linux_dsm_epyc7002/net/ipv6/ip6_fib.c

2469 lines
56 KiB
C
Raw Normal View History

/*
* Linux INET6 implementation
* Forwarding Information Database
*
* Authors:
* Pedro Roque <roque@di.fc.ul.pt>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*
* Changes:
* Yuji SEKIYA @USAGI: Support default route on router node;
* remove ip6_null_entry from the top of
* routing table.
* Ville Nuorvala: Fixed routing subtrees.
*/
#define pr_fmt(fmt) "IPv6: " fmt
#include <linux/errno.h>
#include <linux/types.h>
#include <linux/net.h>
#include <linux/route.h>
#include <linux/netdevice.h>
#include <linux/in6.h>
#include <linux/init.h>
#include <linux/list.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 15:04:11 +07:00
#include <linux/slab.h>
#include <net/ip.h>
#include <net/ipv6.h>
#include <net/ndisc.h>
#include <net/addrconf.h>
#include <net/lwtunnel.h>
#include <net/fib_notifier.h>
#include <net/ip6_fib.h>
#include <net/ip6_route.h>
static struct kmem_cache *fib6_node_kmem __read_mostly;
struct fib6_cleaner {
struct fib6_walker w;
struct net *net;
int (*func)(struct fib6_info *, void *arg);
int sernum;
void *arg;
bool skip_notify;
};
#ifdef CONFIG_IPV6_SUBTREES
#define FWS_INIT FWS_S
#else
#define FWS_INIT FWS_L
#endif
static struct fib6_info *fib6_find_prefix(struct net *net,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
struct fib6_table *table,
struct fib6_node *fn);
static struct fib6_node *fib6_repair_tree(struct net *net,
struct fib6_table *table,
struct fib6_node *fn);
static int fib6_walk(struct net *net, struct fib6_walker *w);
static int fib6_walk_continue(struct fib6_walker *w);
/*
* A routing update causes an increase of the serial number on the
* affected subtree. This allows for cached routes to be asynchronously
* tested when modifications are made to the destination cache as a
* result of redirects, path MTU changes, etc.
*/
treewide: setup_timer() -> timer_setup() (2 field) This converts all remaining setup_timer() calls that use a nested field to reach a struct timer_list. Coccinelle does not have an easy way to match multiple fields, so a new script is needed to change the matches of "&_E->_timer" into "&_E->_field1._timer" in all the rules. spatch --very-quiet --all-includes --include-headers \ -I ./arch/x86/include -I ./arch/x86/include/generated \ -I ./include -I ./arch/x86/include/uapi \ -I ./arch/x86/include/generated/uapi -I ./include/uapi \ -I ./include/generated/uapi --include ./include/linux/kconfig.h \ --dir . \ --cocci-file ~/src/data/timer_setup-2fields.cocci @fix_address_of depends@ expression e; @@ setup_timer( -&(e) +&e , ...) // Update any raw setup_timer() usages that have a NULL callback, but // would otherwise match change_timer_function_usage, since the latter // will update all function assignments done in the face of a NULL // function initialization in setup_timer(). @change_timer_function_usage_NULL@ expression _E; identifier _field1; identifier _timer; type _cast_data; @@ ( -setup_timer(&_E->_field1._timer, NULL, _E); +timer_setup(&_E->_field1._timer, NULL, 0); | -setup_timer(&_E->_field1._timer, NULL, (_cast_data)_E); +timer_setup(&_E->_field1._timer, NULL, 0); | -setup_timer(&_E._field1._timer, NULL, &_E); +timer_setup(&_E._field1._timer, NULL, 0); | -setup_timer(&_E._field1._timer, NULL, (_cast_data)&_E); +timer_setup(&_E._field1._timer, NULL, 0); ) @change_timer_function_usage@ expression _E; identifier _field1; identifier _timer; struct timer_list _stl; identifier _callback; type _cast_func, _cast_data; @@ ( -setup_timer(&_E->_field1._timer, _callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, &_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, &_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)&_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)&_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, &_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, &_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)&_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)&_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | _E->_field1._timer@_stl.function = _callback; | _E->_field1._timer@_stl.function = &_callback; | _E->_field1._timer@_stl.function = (_cast_func)_callback; | _E->_field1._timer@_stl.function = (_cast_func)&_callback; | _E._field1._timer@_stl.function = _callback; | _E._field1._timer@_stl.function = &_callback; | _E._field1._timer@_stl.function = (_cast_func)_callback; | _E._field1._timer@_stl.function = (_cast_func)&_callback; ) // callback(unsigned long arg) @change_callback_handle_cast depends on change_timer_function_usage@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _origtype; identifier _origarg; type _handletype; identifier _handle; @@ void _callback( -_origtype _origarg +struct timer_list *t ) { ( ... when != _origarg _handletype *_handle = -(_handletype *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle = -(void *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle; ... when != _handle _handle = -(_handletype *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle; ... when != _handle _handle = -(void *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg ) } // callback(unsigned long arg) without existing variable @change_callback_handle_cast_no_arg depends on change_timer_function_usage && !change_callback_handle_cast@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _origtype; identifier _origarg; type _handletype; @@ void _callback( -_origtype _origarg +struct timer_list *t ) { + _handletype *_origarg = from_timer(_origarg, t, _field1._timer); + ... when != _origarg - (_handletype *)_origarg + _origarg ... when != _origarg } // Avoid already converted callbacks. @match_callback_converted depends on change_timer_function_usage && !change_callback_handle_cast && !change_callback_handle_cast_no_arg@ identifier change_timer_function_usage._callback; identifier t; @@ void _callback(struct timer_list *t) { ... } // callback(struct something *handle) @change_callback_handle_arg depends on change_timer_function_usage && !match_callback_converted && !change_callback_handle_cast && !change_callback_handle_cast_no_arg@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _handletype; identifier _handle; @@ void _callback( -_handletype *_handle +struct timer_list *t ) { + _handletype *_handle = from_timer(_handle, t, _field1._timer); ... } // If change_callback_handle_arg ran on an empty function, remove // the added handler. @unchange_callback_handle_arg depends on change_timer_function_usage && change_callback_handle_arg@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _handletype; identifier _handle; identifier t; @@ void _callback(struct timer_list *t) { - _handletype *_handle = from_timer(_handle, t, _field1._timer); } // We only want to refactor the setup_timer() data argument if we've found // the matching callback. This undoes changes in change_timer_function_usage. @unchange_timer_function_usage depends on change_timer_function_usage && !change_callback_handle_cast && !change_callback_handle_cast_no_arg && !change_callback_handle_arg@ expression change_timer_function_usage._E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type change_timer_function_usage._cast_data; @@ ( -timer_setup(&_E->_field1._timer, _callback, 0); +setup_timer(&_E->_field1._timer, _callback, (_cast_data)_E); | -timer_setup(&_E._field1._timer, _callback, 0); +setup_timer(&_E._field1._timer, _callback, (_cast_data)&_E); ) // If we fixed a callback from a .function assignment, fix the // assignment cast now. @change_timer_function_assignment depends on change_timer_function_usage && (change_callback_handle_cast || change_callback_handle_cast_no_arg || change_callback_handle_arg)@ expression change_timer_function_usage._E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type _cast_func; typedef TIMER_FUNC_TYPE; @@ ( _E->_field1._timer.function = -_callback +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -&_callback +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -(_cast_func)_callback; +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -(_cast_func)&_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -&_callback; +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -(_cast_func)_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -(_cast_func)&_callback +(TIMER_FUNC_TYPE)_callback ; ) // Sometimes timer functions are called directly. Replace matched args. @change_timer_function_calls depends on change_timer_function_usage && (change_callback_handle_cast || change_callback_handle_cast_no_arg || change_callback_handle_arg)@ expression _E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type _cast_data; @@ _callback( ( -(_cast_data)_E +&_E->_field1._timer | -(_cast_data)&_E +&_E._field1._timer | -_E +&_E->_field1._timer ) ) // If a timer has been configured without a data argument, it can be // converted without regard to the callback argument, since it is unused. @match_timer_function_unused_data@ expression _E; identifier _field1; identifier _timer; identifier _callback; @@ ( -setup_timer(&_E->_field1._timer, _callback, 0); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, 0L); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, 0UL); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0L); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0UL); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0L); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0UL); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0); +timer_setup(_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0L); +timer_setup(_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0UL); +timer_setup(_field1._timer, _callback, 0); ) @change_callback_unused_data depends on match_timer_function_unused_data@ identifier match_timer_function_unused_data._callback; type _origtype; identifier _origarg; @@ void _callback( -_origtype _origarg +struct timer_list *unused ) { ... when != _origarg } Signed-off-by: Kees Cook <keescook@chromium.org>
2017-10-18 10:21:24 +07:00
static void fib6_gc_timer_cb(struct timer_list *t);
#define FOR_WALKERS(net, w) \
list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
static void fib6_walker_link(struct net *net, struct fib6_walker *w)
{
write_lock_bh(&net->ipv6.fib6_walker_lock);
list_add(&w->lh, &net->ipv6.fib6_walkers);
write_unlock_bh(&net->ipv6.fib6_walker_lock);
}
static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
{
write_lock_bh(&net->ipv6.fib6_walker_lock);
list_del(&w->lh);
write_unlock_bh(&net->ipv6.fib6_walker_lock);
}
static int fib6_new_sernum(struct net *net)
{
int new, old;
do {
old = atomic_read(&net->ipv6.fib6_sernum);
new = old < INT_MAX ? old + 1 : 1;
} while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
old, new) != old);
return new;
}
enum {
FIB6_NO_SERNUM_CHANGE = 0,
};
void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
{
struct fib6_node *fn;
fn = rcu_dereference_protected(f6i->fib6_node,
lockdep_is_held(&f6i->fib6_table->tb6_lock));
if (fn)
fn->fn_sernum = fib6_new_sernum(net);
}
/*
* Auxiliary address test functions for the radix tree.
*
* These assume a 32bit processor (although it will work on
* 64bit processors)
*/
/*
* test bit
*/
#if defined(__LITTLE_ENDIAN)
# define BITOP_BE32_SWIZZLE (0x1F & ~7)
#else
# define BITOP_BE32_SWIZZLE 0
#endif
static __be32 addr_bit_set(const void *token, int fn_bit)
{
const __be32 *addr = token;
/*
* Here,
* 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
* is optimized version of
* htonl(1 << ((~fn_bit)&0x1F))
* See include/asm-generic/bitops/le.h.
*/
return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
addr[fn_bit >> 5];
}
struct fib6_info *fib6_info_alloc(gfp_t gfp_flags)
{
struct fib6_info *f6i;
f6i = kzalloc(sizeof(*f6i), gfp_flags);
if (!f6i)
return NULL;
f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
if (!f6i->rt6i_pcpu) {
kfree(f6i);
return NULL;
}
INIT_LIST_HEAD(&f6i->fib6_siblings);
atomic_inc(&f6i->fib6_ref);
return f6i;
}
net/ipv6: respect rcu grace period before freeing fib6_info syzbot reported use after free that is caused by fib6_info being freed without a proper RCU grace period. CPU: 0 PID: 1407 Comm: udevd Not tainted 4.17.0+ #39 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1b9/0x294 lib/dump_stack.c:113 print_address_description+0x6c/0x20b mm/kasan/report.c:256 kasan_report_error mm/kasan/report.c:354 [inline] kasan_report.cold.7+0x242/0x2fe mm/kasan/report.c:412 __asan_report_load8_noabort+0x14/0x20 mm/kasan/report.c:433 __read_once_size include/linux/compiler.h:188 [inline] find_rr_leaf net/ipv6/route.c:705 [inline] rt6_select net/ipv6/route.c:761 [inline] fib6_table_lookup+0x12b7/0x14d0 net/ipv6/route.c:1823 ip6_pol_route+0x1c2/0x1020 net/ipv6/route.c:1856 ip6_pol_route_output+0x54/0x70 net/ipv6/route.c:2082 fib6_rule_lookup+0x211/0x6d0 net/ipv6/fib6_rules.c:122 ip6_route_output_flags+0x2c5/0x350 net/ipv6/route.c:2110 ip6_route_output include/net/ip6_route.h:82 [inline] icmpv6_xrlim_allow net/ipv6/icmp.c:211 [inline] icmp6_send+0x147c/0x2da0 net/ipv6/icmp.c:535 icmpv6_send+0x17a/0x300 net/ipv6/ip6_icmp.c:43 ip6_link_failure+0xa5/0x790 net/ipv6/route.c:2244 dst_link_failure include/net/dst.h:427 [inline] ndisc_error_report+0xd1/0x1c0 net/ipv6/ndisc.c:695 neigh_invalidate+0x246/0x550 net/core/neighbour.c:892 neigh_timer_handler+0xaf9/0xde0 net/core/neighbour.c:978 call_timer_fn+0x230/0x940 kernel/time/timer.c:1326 expire_timers kernel/time/timer.c:1363 [inline] __run_timers+0x79e/0xc50 kernel/time/timer.c:1666 run_timer_softirq+0x4c/0x70 kernel/time/timer.c:1692 __do_softirq+0x2e0/0xaf5 kernel/softirq.c:284 invoke_softirq kernel/softirq.c:364 [inline] irq_exit+0x1d1/0x200 kernel/softirq.c:404 exiting_irq arch/x86/include/asm/apic.h:527 [inline] smp_apic_timer_interrupt+0x17e/0x710 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xf/0x20 arch/x86/entry/entry_64.S:863 </IRQ> RIP: 0010:strlen+0x5e/0xa0 lib/string.c:482 Code: 24 00 74 3b 48 bb 00 00 00 00 00 fc ff df 4c 89 e0 48 83 c0 01 48 89 c2 48 89 c1 48 c1 ea 03 83 e1 07 0f b6 14 1a 38 ca 7f 04 <84> d2 75 23 80 38 00 75 de 48 83 c4 08 4c 29 e0 5b 41 5c 5d c3 48 RSP: 0018:ffff8801af117850 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff13 RAX: ffff880197f53bd0 RBX: dffffc0000000000 RCX: 0000000000000000 RDX: 0000000000000000 RSI: ffffffff81c5b06c RDI: ffff880197f53bc0 RBP: ffff8801af117868 R08: ffff88019a976540 R09: 0000000000000000 R10: ffff88019a976540 R11: 0000000000000000 R12: ffff880197f53bc0 R13: ffff880197f53bc0 R14: ffffffff899e4e90 R15: ffff8801d91c6a00 strlen include/linux/string.h:267 [inline] getname_kernel+0x24/0x370 fs/namei.c:218 open_exec+0x17/0x70 fs/exec.c:882 load_elf_binary+0x968/0x5610 fs/binfmt_elf.c:780 search_binary_handler+0x17d/0x570 fs/exec.c:1653 exec_binprm fs/exec.c:1695 [inline] __do_execve_file.isra.35+0x16fe/0x2710 fs/exec.c:1819 do_execveat_common fs/exec.c:1866 [inline] do_execve fs/exec.c:1883 [inline] __do_sys_execve fs/exec.c:1964 [inline] __se_sys_execve fs/exec.c:1959 [inline] __x64_sys_execve+0x8f/0xc0 fs/exec.c:1959 do_syscall_64+0x1b1/0x800 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x7f1576a46207 Code: 77 19 f4 48 89 d7 44 89 c0 0f 05 48 3d 00 f0 ff ff 76 e0 f7 d8 64 41 89 01 eb d8 f7 d8 64 41 89 01 eb df b8 3b 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 02 f3 c3 48 8b 15 00 8c 2d 00 f7 d8 64 89 02 RSP: 002b:00007ffff2784568 EFLAGS: 00000202 ORIG_RAX: 000000000000003b RAX: ffffffffffffffda RBX: 00000000ffffffff RCX: 00007f1576a46207 RDX: 0000000001215b10 RSI: 00007ffff2784660 RDI: 00007ffff2785670 RBP: 0000000000625500 R08: 000000000000589c R09: 000000000000589c R10: 0000000000000000 R11: 0000000000000202 R12: 0000000001215b10 R13: 0000000000000007 R14: 0000000001204250 R15: 0000000000000005 Allocated by task 12188: save_stack+0x43/0xd0 mm/kasan/kasan.c:448 set_track mm/kasan/kasan.c:460 [inline] kasan_kmalloc+0xc4/0xe0 mm/kasan/kasan.c:553 kmem_cache_alloc_trace+0x152/0x780 mm/slab.c:3620 kmalloc include/linux/slab.h:513 [inline] kzalloc include/linux/slab.h:706 [inline] fib6_info_alloc+0xbb/0x280 net/ipv6/ip6_fib.c:152 ip6_route_info_create+0x782/0x2b50 net/ipv6/route.c:3013 ip6_route_add+0x23/0xb0 net/ipv6/route.c:3154 ipv6_route_ioctl+0x5a5/0x760 net/ipv6/route.c:3660 inet6_ioctl+0x100/0x1f0 net/ipv6/af_inet6.c:546 sock_do_ioctl+0xe4/0x3e0 net/socket.c:973 sock_ioctl+0x30d/0x680 net/socket.c:1097 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1cf/0x16f0 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b1/0x800 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe Freed by task 1402: save_stack+0x43/0xd0 mm/kasan/kasan.c:448 set_track mm/kasan/kasan.c:460 [inline] __kasan_slab_free+0x11a/0x170 mm/kasan/kasan.c:521 kasan_slab_free+0xe/0x10 mm/kasan/kasan.c:528 __cache_free mm/slab.c:3498 [inline] kfree+0xd9/0x260 mm/slab.c:3813 fib6_info_destroy+0x29b/0x350 net/ipv6/ip6_fib.c:207 fib6_info_release include/net/ip6_fib.h:286 [inline] __ip6_del_rt_siblings net/ipv6/route.c:3235 [inline] ip6_route_del+0x11c4/0x13b0 net/ipv6/route.c:3316 ipv6_route_ioctl+0x616/0x760 net/ipv6/route.c:3663 inet6_ioctl+0x100/0x1f0 net/ipv6/af_inet6.c:546 sock_do_ioctl+0xe4/0x3e0 net/socket.c:973 sock_ioctl+0x30d/0x680 net/socket.c:1097 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1cf/0x16f0 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b1/0x800 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe The buggy address belongs to the object at ffff8801b5df2580 which belongs to the cache kmalloc-256 of size 256 The buggy address is located 8 bytes inside of 256-byte region [ffff8801b5df2580, ffff8801b5df2680) The buggy address belongs to the page: page:ffffea0006d77c80 count:1 mapcount:0 mapping:ffff8801da8007c0 index:0xffff8801b5df2e40 flags: 0x2fffc0000000100(slab) raw: 02fffc0000000100 ffffea0006c5cc48 ffffea0007363308 ffff8801da8007c0 raw: ffff8801b5df2e40 ffff8801b5df2080 0000000100000006 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff8801b5df2480: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff8801b5df2500: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc > ffff8801b5df2580: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ^ ffff8801b5df2600: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff8801b5df2680: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb Fixes: a64efe142f5e ("net/ipv6: introduce fib6_info struct and helpers") Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Ahern <dsahern@gmail.com> Reported-by: syzbot+9e6d75e3edef427ee888@syzkaller.appspotmail.com Acked-by: David Ahern <dsahern@gmail.com> Tested-by: David Ahern <dsahern@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-06-18 19:24:31 +07:00
void fib6_info_destroy_rcu(struct rcu_head *head)
{
net/ipv6: respect rcu grace period before freeing fib6_info syzbot reported use after free that is caused by fib6_info being freed without a proper RCU grace period. CPU: 0 PID: 1407 Comm: udevd Not tainted 4.17.0+ #39 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1b9/0x294 lib/dump_stack.c:113 print_address_description+0x6c/0x20b mm/kasan/report.c:256 kasan_report_error mm/kasan/report.c:354 [inline] kasan_report.cold.7+0x242/0x2fe mm/kasan/report.c:412 __asan_report_load8_noabort+0x14/0x20 mm/kasan/report.c:433 __read_once_size include/linux/compiler.h:188 [inline] find_rr_leaf net/ipv6/route.c:705 [inline] rt6_select net/ipv6/route.c:761 [inline] fib6_table_lookup+0x12b7/0x14d0 net/ipv6/route.c:1823 ip6_pol_route+0x1c2/0x1020 net/ipv6/route.c:1856 ip6_pol_route_output+0x54/0x70 net/ipv6/route.c:2082 fib6_rule_lookup+0x211/0x6d0 net/ipv6/fib6_rules.c:122 ip6_route_output_flags+0x2c5/0x350 net/ipv6/route.c:2110 ip6_route_output include/net/ip6_route.h:82 [inline] icmpv6_xrlim_allow net/ipv6/icmp.c:211 [inline] icmp6_send+0x147c/0x2da0 net/ipv6/icmp.c:535 icmpv6_send+0x17a/0x300 net/ipv6/ip6_icmp.c:43 ip6_link_failure+0xa5/0x790 net/ipv6/route.c:2244 dst_link_failure include/net/dst.h:427 [inline] ndisc_error_report+0xd1/0x1c0 net/ipv6/ndisc.c:695 neigh_invalidate+0x246/0x550 net/core/neighbour.c:892 neigh_timer_handler+0xaf9/0xde0 net/core/neighbour.c:978 call_timer_fn+0x230/0x940 kernel/time/timer.c:1326 expire_timers kernel/time/timer.c:1363 [inline] __run_timers+0x79e/0xc50 kernel/time/timer.c:1666 run_timer_softirq+0x4c/0x70 kernel/time/timer.c:1692 __do_softirq+0x2e0/0xaf5 kernel/softirq.c:284 invoke_softirq kernel/softirq.c:364 [inline] irq_exit+0x1d1/0x200 kernel/softirq.c:404 exiting_irq arch/x86/include/asm/apic.h:527 [inline] smp_apic_timer_interrupt+0x17e/0x710 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xf/0x20 arch/x86/entry/entry_64.S:863 </IRQ> RIP: 0010:strlen+0x5e/0xa0 lib/string.c:482 Code: 24 00 74 3b 48 bb 00 00 00 00 00 fc ff df 4c 89 e0 48 83 c0 01 48 89 c2 48 89 c1 48 c1 ea 03 83 e1 07 0f b6 14 1a 38 ca 7f 04 <84> d2 75 23 80 38 00 75 de 48 83 c4 08 4c 29 e0 5b 41 5c 5d c3 48 RSP: 0018:ffff8801af117850 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff13 RAX: ffff880197f53bd0 RBX: dffffc0000000000 RCX: 0000000000000000 RDX: 0000000000000000 RSI: ffffffff81c5b06c RDI: ffff880197f53bc0 RBP: ffff8801af117868 R08: ffff88019a976540 R09: 0000000000000000 R10: ffff88019a976540 R11: 0000000000000000 R12: ffff880197f53bc0 R13: ffff880197f53bc0 R14: ffffffff899e4e90 R15: ffff8801d91c6a00 strlen include/linux/string.h:267 [inline] getname_kernel+0x24/0x370 fs/namei.c:218 open_exec+0x17/0x70 fs/exec.c:882 load_elf_binary+0x968/0x5610 fs/binfmt_elf.c:780 search_binary_handler+0x17d/0x570 fs/exec.c:1653 exec_binprm fs/exec.c:1695 [inline] __do_execve_file.isra.35+0x16fe/0x2710 fs/exec.c:1819 do_execveat_common fs/exec.c:1866 [inline] do_execve fs/exec.c:1883 [inline] __do_sys_execve fs/exec.c:1964 [inline] __se_sys_execve fs/exec.c:1959 [inline] __x64_sys_execve+0x8f/0xc0 fs/exec.c:1959 do_syscall_64+0x1b1/0x800 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x7f1576a46207 Code: 77 19 f4 48 89 d7 44 89 c0 0f 05 48 3d 00 f0 ff ff 76 e0 f7 d8 64 41 89 01 eb d8 f7 d8 64 41 89 01 eb df b8 3b 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 02 f3 c3 48 8b 15 00 8c 2d 00 f7 d8 64 89 02 RSP: 002b:00007ffff2784568 EFLAGS: 00000202 ORIG_RAX: 000000000000003b RAX: ffffffffffffffda RBX: 00000000ffffffff RCX: 00007f1576a46207 RDX: 0000000001215b10 RSI: 00007ffff2784660 RDI: 00007ffff2785670 RBP: 0000000000625500 R08: 000000000000589c R09: 000000000000589c R10: 0000000000000000 R11: 0000000000000202 R12: 0000000001215b10 R13: 0000000000000007 R14: 0000000001204250 R15: 0000000000000005 Allocated by task 12188: save_stack+0x43/0xd0 mm/kasan/kasan.c:448 set_track mm/kasan/kasan.c:460 [inline] kasan_kmalloc+0xc4/0xe0 mm/kasan/kasan.c:553 kmem_cache_alloc_trace+0x152/0x780 mm/slab.c:3620 kmalloc include/linux/slab.h:513 [inline] kzalloc include/linux/slab.h:706 [inline] fib6_info_alloc+0xbb/0x280 net/ipv6/ip6_fib.c:152 ip6_route_info_create+0x782/0x2b50 net/ipv6/route.c:3013 ip6_route_add+0x23/0xb0 net/ipv6/route.c:3154 ipv6_route_ioctl+0x5a5/0x760 net/ipv6/route.c:3660 inet6_ioctl+0x100/0x1f0 net/ipv6/af_inet6.c:546 sock_do_ioctl+0xe4/0x3e0 net/socket.c:973 sock_ioctl+0x30d/0x680 net/socket.c:1097 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1cf/0x16f0 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b1/0x800 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe Freed by task 1402: save_stack+0x43/0xd0 mm/kasan/kasan.c:448 set_track mm/kasan/kasan.c:460 [inline] __kasan_slab_free+0x11a/0x170 mm/kasan/kasan.c:521 kasan_slab_free+0xe/0x10 mm/kasan/kasan.c:528 __cache_free mm/slab.c:3498 [inline] kfree+0xd9/0x260 mm/slab.c:3813 fib6_info_destroy+0x29b/0x350 net/ipv6/ip6_fib.c:207 fib6_info_release include/net/ip6_fib.h:286 [inline] __ip6_del_rt_siblings net/ipv6/route.c:3235 [inline] ip6_route_del+0x11c4/0x13b0 net/ipv6/route.c:3316 ipv6_route_ioctl+0x616/0x760 net/ipv6/route.c:3663 inet6_ioctl+0x100/0x1f0 net/ipv6/af_inet6.c:546 sock_do_ioctl+0xe4/0x3e0 net/socket.c:973 sock_ioctl+0x30d/0x680 net/socket.c:1097 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1cf/0x16f0 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b1/0x800 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe The buggy address belongs to the object at ffff8801b5df2580 which belongs to the cache kmalloc-256 of size 256 The buggy address is located 8 bytes inside of 256-byte region [ffff8801b5df2580, ffff8801b5df2680) The buggy address belongs to the page: page:ffffea0006d77c80 count:1 mapcount:0 mapping:ffff8801da8007c0 index:0xffff8801b5df2e40 flags: 0x2fffc0000000100(slab) raw: 02fffc0000000100 ffffea0006c5cc48 ffffea0007363308 ffff8801da8007c0 raw: ffff8801b5df2e40 ffff8801b5df2080 0000000100000006 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff8801b5df2480: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff8801b5df2500: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc > ffff8801b5df2580: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ^ ffff8801b5df2600: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff8801b5df2680: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb Fixes: a64efe142f5e ("net/ipv6: introduce fib6_info struct and helpers") Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Ahern <dsahern@gmail.com> Reported-by: syzbot+9e6d75e3edef427ee888@syzkaller.appspotmail.com Acked-by: David Ahern <dsahern@gmail.com> Tested-by: David Ahern <dsahern@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-06-18 19:24:31 +07:00
struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
struct rt6_exception_bucket *bucket;
WARN_ON(f6i->fib6_node);
bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1);
if (bucket) {
f6i->rt6i_exception_bucket = NULL;
kfree(bucket);
}
if (f6i->rt6i_pcpu) {
int cpu;
for_each_possible_cpu(cpu) {
struct rt6_info **ppcpu_rt;
struct rt6_info *pcpu_rt;
ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
pcpu_rt = *ppcpu_rt;
if (pcpu_rt) {
dst_dev_put(&pcpu_rt->dst);
dst_release(&pcpu_rt->dst);
*ppcpu_rt = NULL;
}
}
free_percpu(f6i->rt6i_pcpu);
}
lwtstate_put(f6i->fib6_nh.nh_lwtstate);
if (f6i->fib6_nh.nh_dev)
dev_put(f6i->fib6_nh.nh_dev);
ip_fib_metrics_put(f6i->fib6_metrics);
kfree(f6i);
}
net/ipv6: respect rcu grace period before freeing fib6_info syzbot reported use after free that is caused by fib6_info being freed without a proper RCU grace period. CPU: 0 PID: 1407 Comm: udevd Not tainted 4.17.0+ #39 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1b9/0x294 lib/dump_stack.c:113 print_address_description+0x6c/0x20b mm/kasan/report.c:256 kasan_report_error mm/kasan/report.c:354 [inline] kasan_report.cold.7+0x242/0x2fe mm/kasan/report.c:412 __asan_report_load8_noabort+0x14/0x20 mm/kasan/report.c:433 __read_once_size include/linux/compiler.h:188 [inline] find_rr_leaf net/ipv6/route.c:705 [inline] rt6_select net/ipv6/route.c:761 [inline] fib6_table_lookup+0x12b7/0x14d0 net/ipv6/route.c:1823 ip6_pol_route+0x1c2/0x1020 net/ipv6/route.c:1856 ip6_pol_route_output+0x54/0x70 net/ipv6/route.c:2082 fib6_rule_lookup+0x211/0x6d0 net/ipv6/fib6_rules.c:122 ip6_route_output_flags+0x2c5/0x350 net/ipv6/route.c:2110 ip6_route_output include/net/ip6_route.h:82 [inline] icmpv6_xrlim_allow net/ipv6/icmp.c:211 [inline] icmp6_send+0x147c/0x2da0 net/ipv6/icmp.c:535 icmpv6_send+0x17a/0x300 net/ipv6/ip6_icmp.c:43 ip6_link_failure+0xa5/0x790 net/ipv6/route.c:2244 dst_link_failure include/net/dst.h:427 [inline] ndisc_error_report+0xd1/0x1c0 net/ipv6/ndisc.c:695 neigh_invalidate+0x246/0x550 net/core/neighbour.c:892 neigh_timer_handler+0xaf9/0xde0 net/core/neighbour.c:978 call_timer_fn+0x230/0x940 kernel/time/timer.c:1326 expire_timers kernel/time/timer.c:1363 [inline] __run_timers+0x79e/0xc50 kernel/time/timer.c:1666 run_timer_softirq+0x4c/0x70 kernel/time/timer.c:1692 __do_softirq+0x2e0/0xaf5 kernel/softirq.c:284 invoke_softirq kernel/softirq.c:364 [inline] irq_exit+0x1d1/0x200 kernel/softirq.c:404 exiting_irq arch/x86/include/asm/apic.h:527 [inline] smp_apic_timer_interrupt+0x17e/0x710 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xf/0x20 arch/x86/entry/entry_64.S:863 </IRQ> RIP: 0010:strlen+0x5e/0xa0 lib/string.c:482 Code: 24 00 74 3b 48 bb 00 00 00 00 00 fc ff df 4c 89 e0 48 83 c0 01 48 89 c2 48 89 c1 48 c1 ea 03 83 e1 07 0f b6 14 1a 38 ca 7f 04 <84> d2 75 23 80 38 00 75 de 48 83 c4 08 4c 29 e0 5b 41 5c 5d c3 48 RSP: 0018:ffff8801af117850 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff13 RAX: ffff880197f53bd0 RBX: dffffc0000000000 RCX: 0000000000000000 RDX: 0000000000000000 RSI: ffffffff81c5b06c RDI: ffff880197f53bc0 RBP: ffff8801af117868 R08: ffff88019a976540 R09: 0000000000000000 R10: ffff88019a976540 R11: 0000000000000000 R12: ffff880197f53bc0 R13: ffff880197f53bc0 R14: ffffffff899e4e90 R15: ffff8801d91c6a00 strlen include/linux/string.h:267 [inline] getname_kernel+0x24/0x370 fs/namei.c:218 open_exec+0x17/0x70 fs/exec.c:882 load_elf_binary+0x968/0x5610 fs/binfmt_elf.c:780 search_binary_handler+0x17d/0x570 fs/exec.c:1653 exec_binprm fs/exec.c:1695 [inline] __do_execve_file.isra.35+0x16fe/0x2710 fs/exec.c:1819 do_execveat_common fs/exec.c:1866 [inline] do_execve fs/exec.c:1883 [inline] __do_sys_execve fs/exec.c:1964 [inline] __se_sys_execve fs/exec.c:1959 [inline] __x64_sys_execve+0x8f/0xc0 fs/exec.c:1959 do_syscall_64+0x1b1/0x800 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x7f1576a46207 Code: 77 19 f4 48 89 d7 44 89 c0 0f 05 48 3d 00 f0 ff ff 76 e0 f7 d8 64 41 89 01 eb d8 f7 d8 64 41 89 01 eb df b8 3b 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 02 f3 c3 48 8b 15 00 8c 2d 00 f7 d8 64 89 02 RSP: 002b:00007ffff2784568 EFLAGS: 00000202 ORIG_RAX: 000000000000003b RAX: ffffffffffffffda RBX: 00000000ffffffff RCX: 00007f1576a46207 RDX: 0000000001215b10 RSI: 00007ffff2784660 RDI: 00007ffff2785670 RBP: 0000000000625500 R08: 000000000000589c R09: 000000000000589c R10: 0000000000000000 R11: 0000000000000202 R12: 0000000001215b10 R13: 0000000000000007 R14: 0000000001204250 R15: 0000000000000005 Allocated by task 12188: save_stack+0x43/0xd0 mm/kasan/kasan.c:448 set_track mm/kasan/kasan.c:460 [inline] kasan_kmalloc+0xc4/0xe0 mm/kasan/kasan.c:553 kmem_cache_alloc_trace+0x152/0x780 mm/slab.c:3620 kmalloc include/linux/slab.h:513 [inline] kzalloc include/linux/slab.h:706 [inline] fib6_info_alloc+0xbb/0x280 net/ipv6/ip6_fib.c:152 ip6_route_info_create+0x782/0x2b50 net/ipv6/route.c:3013 ip6_route_add+0x23/0xb0 net/ipv6/route.c:3154 ipv6_route_ioctl+0x5a5/0x760 net/ipv6/route.c:3660 inet6_ioctl+0x100/0x1f0 net/ipv6/af_inet6.c:546 sock_do_ioctl+0xe4/0x3e0 net/socket.c:973 sock_ioctl+0x30d/0x680 net/socket.c:1097 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1cf/0x16f0 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b1/0x800 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe Freed by task 1402: save_stack+0x43/0xd0 mm/kasan/kasan.c:448 set_track mm/kasan/kasan.c:460 [inline] __kasan_slab_free+0x11a/0x170 mm/kasan/kasan.c:521 kasan_slab_free+0xe/0x10 mm/kasan/kasan.c:528 __cache_free mm/slab.c:3498 [inline] kfree+0xd9/0x260 mm/slab.c:3813 fib6_info_destroy+0x29b/0x350 net/ipv6/ip6_fib.c:207 fib6_info_release include/net/ip6_fib.h:286 [inline] __ip6_del_rt_siblings net/ipv6/route.c:3235 [inline] ip6_route_del+0x11c4/0x13b0 net/ipv6/route.c:3316 ipv6_route_ioctl+0x616/0x760 net/ipv6/route.c:3663 inet6_ioctl+0x100/0x1f0 net/ipv6/af_inet6.c:546 sock_do_ioctl+0xe4/0x3e0 net/socket.c:973 sock_ioctl+0x30d/0x680 net/socket.c:1097 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1cf/0x16f0 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b1/0x800 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe The buggy address belongs to the object at ffff8801b5df2580 which belongs to the cache kmalloc-256 of size 256 The buggy address is located 8 bytes inside of 256-byte region [ffff8801b5df2580, ffff8801b5df2680) The buggy address belongs to the page: page:ffffea0006d77c80 count:1 mapcount:0 mapping:ffff8801da8007c0 index:0xffff8801b5df2e40 flags: 0x2fffc0000000100(slab) raw: 02fffc0000000100 ffffea0006c5cc48 ffffea0007363308 ffff8801da8007c0 raw: ffff8801b5df2e40 ffff8801b5df2080 0000000100000006 0000000000000000 page dumped because: kasan: bad access detected Memory state around the buggy address: ffff8801b5df2480: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff8801b5df2500: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc > ffff8801b5df2580: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ^ ffff8801b5df2600: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb ffff8801b5df2680: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb Fixes: a64efe142f5e ("net/ipv6: introduce fib6_info struct and helpers") Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Ahern <dsahern@gmail.com> Reported-by: syzbot+9e6d75e3edef427ee888@syzkaller.appspotmail.com Acked-by: David Ahern <dsahern@gmail.com> Tested-by: David Ahern <dsahern@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-06-18 19:24:31 +07:00
EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
static struct fib6_node *node_alloc(struct net *net)
{
struct fib6_node *fn;
fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
if (fn)
net->ipv6.rt6_stats->fib_nodes++;
return fn;
}
static void node_free_immediate(struct net *net, struct fib6_node *fn)
{
kmem_cache_free(fib6_node_kmem, fn);
net->ipv6.rt6_stats->fib_nodes--;
}
static void node_free_rcu(struct rcu_head *head)
{
struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
kmem_cache_free(fib6_node_kmem, fn);
}
static void node_free(struct net *net, struct fib6_node *fn)
{
call_rcu(&fn->rcu, node_free_rcu);
net->ipv6.rt6_stats->fib_nodes--;
}
static void fib6_free_table(struct fib6_table *table)
{
inetpeer_invalidate_tree(&table->tb6_peers);
kfree(table);
}
static void fib6_link_table(struct net *net, struct fib6_table *tb)
{
unsigned int h;
/*
* Initialize table lock at a single place to give lockdep a key,
* tables aren't visible prior to being linked to the list.
*/
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_lock_init(&tb->tb6_lock);
h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
/*
* No protection necessary, this is the only list mutatation
* operation, tables never disappear once they exist.
*/
hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
}
#ifdef CONFIG_IPV6_MULTIPLE_TABLES
static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
{
struct fib6_table *table;
table = kzalloc(sizeof(*table), GFP_ATOMIC);
if (table) {
table->tb6_id = id;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(table->tb6_root.leaf,
net->ipv6.fib6_null_entry);
table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
inet_peer_base_init(&table->tb6_peers);
}
return table;
}
struct fib6_table *fib6_new_table(struct net *net, u32 id)
{
struct fib6_table *tb;
if (id == 0)
id = RT6_TABLE_MAIN;
tb = fib6_get_table(net, id);
if (tb)
return tb;
tb = fib6_alloc_table(net, id);
if (tb)
fib6_link_table(net, tb);
return tb;
}
EXPORT_SYMBOL_GPL(fib6_new_table);
struct fib6_table *fib6_get_table(struct net *net, u32 id)
{
struct fib6_table *tb;
struct hlist_head *head;
unsigned int h;
if (id == 0)
id = RT6_TABLE_MAIN;
h = id & (FIB6_TABLE_HASHSZ - 1);
rcu_read_lock();
head = &net->ipv6.fib_table_hash[h];
hlist: drop the node parameter from iterators I'm not sure why, but the hlist for each entry iterators were conceived list_for_each_entry(pos, head, member) The hlist ones were greedy and wanted an extra parameter: hlist_for_each_entry(tpos, pos, head, member) Why did they need an extra pos parameter? I'm not quite sure. Not only they don't really need it, it also prevents the iterator from looking exactly like the list iterator, which is unfortunate. Besides the semantic patch, there was some manual work required: - Fix up the actual hlist iterators in linux/list.h - Fix up the declaration of other iterators based on the hlist ones. - A very small amount of places were using the 'node' parameter, this was modified to use 'obj->member' instead. - Coccinelle didn't handle the hlist_for_each_entry_safe iterator properly, so those had to be fixed up manually. The semantic patch which is mostly the work of Peter Senna Tschudin is here: @@ iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host; type T; expression a,c,d,e; identifier b; statement S; @@ -T b; <+... when != b ( hlist_for_each_entry(a, - b, c, d) S | hlist_for_each_entry_continue(a, - b, c) S | hlist_for_each_entry_from(a, - b, c) S | hlist_for_each_entry_rcu(a, - b, c, d) S | hlist_for_each_entry_rcu_bh(a, - b, c, d) S | hlist_for_each_entry_continue_rcu_bh(a, - b, c) S | for_each_busy_worker(a, c, - b, d) S | ax25_uid_for_each(a, - b, c) S | ax25_for_each(a, - b, c) S | inet_bind_bucket_for_each(a, - b, c) S | sctp_for_each_hentry(a, - b, c) S | sk_for_each(a, - b, c) S | sk_for_each_rcu(a, - b, c) S | sk_for_each_from -(a, b) +(a) S + sk_for_each_from(a) S | sk_for_each_safe(a, - b, c, d) S | sk_for_each_bound(a, - b, c) S | hlist_for_each_entry_safe(a, - b, c, d, e) S | hlist_for_each_entry_continue_rcu(a, - b, c) S | nr_neigh_for_each(a, - b, c) S | nr_neigh_for_each_safe(a, - b, c, d) S | nr_node_for_each(a, - b, c) S | nr_node_for_each_safe(a, - b, c, d) S | - for_each_gfn_sp(a, c, d, b) S + for_each_gfn_sp(a, c, d) S | - for_each_gfn_indirect_valid_sp(a, c, d, b) S + for_each_gfn_indirect_valid_sp(a, c, d) S | for_each_host(a, - b, c) S | for_each_host_safe(a, - b, c, d) S | for_each_mesh_entry(a, - b, c, d) S ) ...+> [akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c] [akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c] [akpm@linux-foundation.org: checkpatch fixes] [akpm@linux-foundation.org: fix warnings] [akpm@linux-foudnation.org: redo intrusive kvm changes] Tested-by: Peter Senna Tschudin <peter.senna@gmail.com> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Sasha Levin <sasha.levin@oracle.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Gleb Natapov <gleb@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 08:06:00 +07:00
hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
if (tb->tb6_id == id) {
rcu_read_unlock();
return tb;
}
}
rcu_read_unlock();
return NULL;
}
EXPORT_SYMBOL_GPL(fib6_get_table);
static void __net_init fib6_tables_init(struct net *net)
{
fib6_link_table(net, net->ipv6.fib6_main_tbl);
fib6_link_table(net, net->ipv6.fib6_local_tbl);
}
#else
struct fib6_table *fib6_new_table(struct net *net, u32 id)
{
return fib6_get_table(net, id);
}
struct fib6_table *fib6_get_table(struct net *net, u32 id)
{
return net->ipv6.fib6_main_tbl;
}
struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
const struct sk_buff *skb,
int flags, pol_lookup_t lookup)
{
struct rt6_info *rt;
rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
if (rt->dst.error == -EAGAIN) {
ip6_rt_put(rt);
rt = net->ipv6.ip6_null_entry;
dst_hold(&rt->dst);
}
return &rt->dst;
}
/* called with rcu lock held; no reference taken on fib6_info */
struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
int flags)
{
return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags);
}
static void __net_init fib6_tables_init(struct net *net)
{
fib6_link_table(net, net->ipv6.fib6_main_tbl);
}
#endif
unsigned int fib6_tables_seq_read(struct net *net)
{
unsigned int h, fib_seq = 0;
rcu_read_lock();
for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
struct hlist_head *head = &net->ipv6.fib_table_hash[h];
struct fib6_table *tb;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
hlist_for_each_entry_rcu(tb, head, tb6_hlist)
fib_seq += tb->fib_seq;
}
rcu_read_unlock();
return fib_seq;
}
static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
enum fib_event_type event_type,
struct fib6_info *rt)
{
struct fib6_entry_notifier_info info = {
.rt = rt,
};
return call_fib6_notifier(nb, net, event_type, &info.info);
}
static int call_fib6_entry_notifiers(struct net *net,
enum fib_event_type event_type,
struct fib6_info *rt,
struct netlink_ext_ack *extack)
{
struct fib6_entry_notifier_info info = {
.info.extack = extack,
.rt = rt,
};
rt->fib6_table->fib_seq++;
return call_fib6_notifiers(net, event_type, &info.info);
}
struct fib6_dump_arg {
struct net *net;
struct notifier_block *nb;
};
static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
{
if (rt == arg->net->ipv6.fib6_null_entry)
return;
call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
}
static int fib6_node_dump(struct fib6_walker *w)
{
struct fib6_info *rt;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
for_each_fib6_walker_rt(w)
fib6_rt_dump(rt, w->args);
w->leaf = NULL;
return 0;
}
static void fib6_table_dump(struct net *net, struct fib6_table *tb,
struct fib6_walker *w)
{
w->root = &tb->tb6_root;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_lock_bh(&tb->tb6_lock);
fib6_walk(net, w);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_unlock_bh(&tb->tb6_lock);
}
/* Called with rcu_read_lock() */
int fib6_tables_dump(struct net *net, struct notifier_block *nb)
{
struct fib6_dump_arg arg;
struct fib6_walker *w;
unsigned int h;
w = kzalloc(sizeof(*w), GFP_ATOMIC);
if (!w)
return -ENOMEM;
w->func = fib6_node_dump;
arg.net = net;
arg.nb = nb;
w->args = &arg;
for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
struct hlist_head *head = &net->ipv6.fib_table_hash[h];
struct fib6_table *tb;
hlist_for_each_entry_rcu(tb, head, tb6_hlist)
fib6_table_dump(net, tb, w);
}
kfree(w);
return 0;
}
static int fib6_dump_node(struct fib6_walker *w)
{
int res;
struct fib6_info *rt;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
for_each_fib6_walker_rt(w) {
res = rt6_dump_route(rt, w->args);
if (res < 0) {
/* Frame is full, suspend walking */
w->leaf = rt;
return 1;
}
/* Multipath routes are dumped in one route with the
* RTA_MULTIPATH attribute. Jump 'rt' to point to the
* last sibling of this route (no need to dump the
* sibling routes again)
*/
if (rt->fib6_nsiblings)
rt = list_last_entry(&rt->fib6_siblings,
struct fib6_info,
fib6_siblings);
}
w->leaf = NULL;
return 0;
}
static void fib6_dump_end(struct netlink_callback *cb)
{
struct net *net = sock_net(cb->skb->sk);
struct fib6_walker *w = (void *)cb->args[2];
if (w) {
if (cb->args[4]) {
cb->args[4] = 0;
fib6_walker_unlink(net, w);
}
cb->args[2] = 0;
kfree(w);
}
cb->done = (void *)cb->args[3];
cb->args[1] = 3;
}
static int fib6_dump_done(struct netlink_callback *cb)
{
fib6_dump_end(cb);
return cb->done ? cb->done(cb) : 0;
}
static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
struct netlink_callback *cb)
{
struct net *net = sock_net(skb->sk);
struct fib6_walker *w;
int res;
w = (void *)cb->args[2];
w->root = &table->tb6_root;
if (cb->args[4] == 0) {
ipv6: fib: fix crash when changing large fib while dumping it When the fib size exceeds what can be dumped in a single skb, the dump is suspended and resumed once the last skb has been received by userspace. When the fib is changed while the dump is suspended, the walker might contain stale pointers, causing a crash when the dump is resumed. BUG: unable to handle kernel NULL pointer dereference at 0000000000000018 IP: [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] PGD 5347a067 PUD 65c7067 PMD 0 Oops: 0000 [#1] PREEMPT SMP ... RIP: 0010:[<ffffffffa01bce04>] [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] ... Call Trace: [<ffffffff8104aca3>] ? mutex_spin_on_owner+0x59/0x71 [<ffffffffa01bd105>] inet6_dump_fib+0x11b/0x1b9 [ipv6] [<ffffffff81371af4>] netlink_dump+0x5b/0x19e [<ffffffff8134f288>] ? consume_skb+0x28/0x2a [<ffffffff81373b69>] netlink_recvmsg+0x1ab/0x2c6 [<ffffffff81372781>] ? netlink_unicast+0xfa/0x151 [<ffffffff813483e0>] __sock_recvmsg+0x6d/0x79 [<ffffffff81348a53>] sock_recvmsg+0xca/0xe3 [<ffffffff81066d4b>] ? autoremove_wake_function+0x0/0x38 [<ffffffff811ed1f8>] ? radix_tree_lookup_slot+0xe/0x10 [<ffffffff810b3ed7>] ? find_get_page+0x90/0xa5 [<ffffffff810b5dc5>] ? filemap_fault+0x201/0x34f [<ffffffff810ef152>] ? fget_light+0x2f/0xac [<ffffffff813519e7>] ? verify_iovec+0x4f/0x94 [<ffffffff81349a65>] sys_recvmsg+0x14d/0x223 Store the serial number when beginning to walk the fib and reload pointers when continuing to walk after a change occured. Similar to other dumping functions, this might cause unrelated entries to be missed when entries are deleted. Tested-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-02-08 12:19:03 +07:00
w->count = 0;
w->skip = 0;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_lock_bh(&table->tb6_lock);
res = fib6_walk(net, w);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_unlock_bh(&table->tb6_lock);
ipv6: fib: fix crash when changing large fib while dumping it When the fib size exceeds what can be dumped in a single skb, the dump is suspended and resumed once the last skb has been received by userspace. When the fib is changed while the dump is suspended, the walker might contain stale pointers, causing a crash when the dump is resumed. BUG: unable to handle kernel NULL pointer dereference at 0000000000000018 IP: [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] PGD 5347a067 PUD 65c7067 PMD 0 Oops: 0000 [#1] PREEMPT SMP ... RIP: 0010:[<ffffffffa01bce04>] [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] ... Call Trace: [<ffffffff8104aca3>] ? mutex_spin_on_owner+0x59/0x71 [<ffffffffa01bd105>] inet6_dump_fib+0x11b/0x1b9 [ipv6] [<ffffffff81371af4>] netlink_dump+0x5b/0x19e [<ffffffff8134f288>] ? consume_skb+0x28/0x2a [<ffffffff81373b69>] netlink_recvmsg+0x1ab/0x2c6 [<ffffffff81372781>] ? netlink_unicast+0xfa/0x151 [<ffffffff813483e0>] __sock_recvmsg+0x6d/0x79 [<ffffffff81348a53>] sock_recvmsg+0xca/0xe3 [<ffffffff81066d4b>] ? autoremove_wake_function+0x0/0x38 [<ffffffff811ed1f8>] ? radix_tree_lookup_slot+0xe/0x10 [<ffffffff810b3ed7>] ? find_get_page+0x90/0xa5 [<ffffffff810b5dc5>] ? filemap_fault+0x201/0x34f [<ffffffff810ef152>] ? fget_light+0x2f/0xac [<ffffffff813519e7>] ? verify_iovec+0x4f/0x94 [<ffffffff81349a65>] sys_recvmsg+0x14d/0x223 Store the serial number when beginning to walk the fib and reload pointers when continuing to walk after a change occured. Similar to other dumping functions, this might cause unrelated entries to be missed when entries are deleted. Tested-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-02-08 12:19:03 +07:00
if (res > 0) {
cb->args[4] = 1;
ipv6: fib: fix crash when changing large fib while dumping it When the fib size exceeds what can be dumped in a single skb, the dump is suspended and resumed once the last skb has been received by userspace. When the fib is changed while the dump is suspended, the walker might contain stale pointers, causing a crash when the dump is resumed. BUG: unable to handle kernel NULL pointer dereference at 0000000000000018 IP: [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] PGD 5347a067 PUD 65c7067 PMD 0 Oops: 0000 [#1] PREEMPT SMP ... RIP: 0010:[<ffffffffa01bce04>] [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] ... Call Trace: [<ffffffff8104aca3>] ? mutex_spin_on_owner+0x59/0x71 [<ffffffffa01bd105>] inet6_dump_fib+0x11b/0x1b9 [ipv6] [<ffffffff81371af4>] netlink_dump+0x5b/0x19e [<ffffffff8134f288>] ? consume_skb+0x28/0x2a [<ffffffff81373b69>] netlink_recvmsg+0x1ab/0x2c6 [<ffffffff81372781>] ? netlink_unicast+0xfa/0x151 [<ffffffff813483e0>] __sock_recvmsg+0x6d/0x79 [<ffffffff81348a53>] sock_recvmsg+0xca/0xe3 [<ffffffff81066d4b>] ? autoremove_wake_function+0x0/0x38 [<ffffffff811ed1f8>] ? radix_tree_lookup_slot+0xe/0x10 [<ffffffff810b3ed7>] ? find_get_page+0x90/0xa5 [<ffffffff810b5dc5>] ? filemap_fault+0x201/0x34f [<ffffffff810ef152>] ? fget_light+0x2f/0xac [<ffffffff813519e7>] ? verify_iovec+0x4f/0x94 [<ffffffff81349a65>] sys_recvmsg+0x14d/0x223 Store the serial number when beginning to walk the fib and reload pointers when continuing to walk after a change occured. Similar to other dumping functions, this might cause unrelated entries to be missed when entries are deleted. Tested-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-02-08 12:19:03 +07:00
cb->args[5] = w->root->fn_sernum;
}
} else {
ipv6: fib: fix crash when changing large fib while dumping it When the fib size exceeds what can be dumped in a single skb, the dump is suspended and resumed once the last skb has been received by userspace. When the fib is changed while the dump is suspended, the walker might contain stale pointers, causing a crash when the dump is resumed. BUG: unable to handle kernel NULL pointer dereference at 0000000000000018 IP: [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] PGD 5347a067 PUD 65c7067 PMD 0 Oops: 0000 [#1] PREEMPT SMP ... RIP: 0010:[<ffffffffa01bce04>] [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] ... Call Trace: [<ffffffff8104aca3>] ? mutex_spin_on_owner+0x59/0x71 [<ffffffffa01bd105>] inet6_dump_fib+0x11b/0x1b9 [ipv6] [<ffffffff81371af4>] netlink_dump+0x5b/0x19e [<ffffffff8134f288>] ? consume_skb+0x28/0x2a [<ffffffff81373b69>] netlink_recvmsg+0x1ab/0x2c6 [<ffffffff81372781>] ? netlink_unicast+0xfa/0x151 [<ffffffff813483e0>] __sock_recvmsg+0x6d/0x79 [<ffffffff81348a53>] sock_recvmsg+0xca/0xe3 [<ffffffff81066d4b>] ? autoremove_wake_function+0x0/0x38 [<ffffffff811ed1f8>] ? radix_tree_lookup_slot+0xe/0x10 [<ffffffff810b3ed7>] ? find_get_page+0x90/0xa5 [<ffffffff810b5dc5>] ? filemap_fault+0x201/0x34f [<ffffffff810ef152>] ? fget_light+0x2f/0xac [<ffffffff813519e7>] ? verify_iovec+0x4f/0x94 [<ffffffff81349a65>] sys_recvmsg+0x14d/0x223 Store the serial number when beginning to walk the fib and reload pointers when continuing to walk after a change occured. Similar to other dumping functions, this might cause unrelated entries to be missed when entries are deleted. Tested-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-02-08 12:19:03 +07:00
if (cb->args[5] != w->root->fn_sernum) {
/* Begin at the root if the tree changed */
cb->args[5] = w->root->fn_sernum;
w->state = FWS_INIT;
w->node = w->root;
w->skip = w->count;
} else
w->skip = 0;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_lock_bh(&table->tb6_lock);
res = fib6_walk_continue(w);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_unlock_bh(&table->tb6_lock);
if (res <= 0) {
fib6_walker_unlink(net, w);
cb->args[4] = 0;
}
}
return res;
}
static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
{
const struct nlmsghdr *nlh = cb->nlh;
struct net *net = sock_net(skb->sk);
struct rt6_rtnl_dump_arg arg = {};
unsigned int h, s_h;
unsigned int e = 0, s_e;
struct fib6_walker *w;
struct fib6_table *tb;
struct hlist_head *head;
int res = 0;
if (cb->strict_check) {
int err;
err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
if (err < 0)
return err;
} else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
struct rtmsg *rtm = nlmsg_data(nlh);
arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
}
/* fib entries are never clones */
if (arg.filter.flags & RTM_F_CLONED)
goto out;
w = (void *)cb->args[2];
if (!w) {
/* New dump:
*
* 1. hook callback destructor.
*/
cb->args[3] = (long)cb->done;
cb->done = fib6_dump_done;
/*
* 2. allocate and initialize walker.
*/
w = kzalloc(sizeof(*w), GFP_ATOMIC);
if (!w)
return -ENOMEM;
w->func = fib6_dump_node;
cb->args[2] = (long)w;
}
arg.skb = skb;
arg.cb = cb;
arg.net = net;
w->args = &arg;
if (arg.filter.table_id) {
tb = fib6_get_table(net, arg.filter.table_id);
if (!tb) {
if (arg.filter.dump_all_families)
goto out;
NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
return -ENOENT;
}
if (!cb->args[0]) {
res = fib6_dump_table(tb, skb, cb);
if (!res)
cb->args[0] = 1;
}
goto out;
}
s_h = cb->args[0];
s_e = cb->args[1];
rcu_read_lock();
for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
e = 0;
head = &net->ipv6.fib_table_hash[h];
hlist: drop the node parameter from iterators I'm not sure why, but the hlist for each entry iterators were conceived list_for_each_entry(pos, head, member) The hlist ones were greedy and wanted an extra parameter: hlist_for_each_entry(tpos, pos, head, member) Why did they need an extra pos parameter? I'm not quite sure. Not only they don't really need it, it also prevents the iterator from looking exactly like the list iterator, which is unfortunate. Besides the semantic patch, there was some manual work required: - Fix up the actual hlist iterators in linux/list.h - Fix up the declaration of other iterators based on the hlist ones. - A very small amount of places were using the 'node' parameter, this was modified to use 'obj->member' instead. - Coccinelle didn't handle the hlist_for_each_entry_safe iterator properly, so those had to be fixed up manually. The semantic patch which is mostly the work of Peter Senna Tschudin is here: @@ iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host; type T; expression a,c,d,e; identifier b; statement S; @@ -T b; <+... when != b ( hlist_for_each_entry(a, - b, c, d) S | hlist_for_each_entry_continue(a, - b, c) S | hlist_for_each_entry_from(a, - b, c) S | hlist_for_each_entry_rcu(a, - b, c, d) S | hlist_for_each_entry_rcu_bh(a, - b, c, d) S | hlist_for_each_entry_continue_rcu_bh(a, - b, c) S | for_each_busy_worker(a, c, - b, d) S | ax25_uid_for_each(a, - b, c) S | ax25_for_each(a, - b, c) S | inet_bind_bucket_for_each(a, - b, c) S | sctp_for_each_hentry(a, - b, c) S | sk_for_each(a, - b, c) S | sk_for_each_rcu(a, - b, c) S | sk_for_each_from -(a, b) +(a) S + sk_for_each_from(a) S | sk_for_each_safe(a, - b, c, d) S | sk_for_each_bound(a, - b, c) S | hlist_for_each_entry_safe(a, - b, c, d, e) S | hlist_for_each_entry_continue_rcu(a, - b, c) S | nr_neigh_for_each(a, - b, c) S | nr_neigh_for_each_safe(a, - b, c, d) S | nr_node_for_each(a, - b, c) S | nr_node_for_each_safe(a, - b, c, d) S | - for_each_gfn_sp(a, c, d, b) S + for_each_gfn_sp(a, c, d) S | - for_each_gfn_indirect_valid_sp(a, c, d, b) S + for_each_gfn_indirect_valid_sp(a, c, d) S | for_each_host(a, - b, c) S | for_each_host_safe(a, - b, c, d) S | for_each_mesh_entry(a, - b, c, d) S ) ...+> [akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c] [akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c] [akpm@linux-foundation.org: checkpatch fixes] [akpm@linux-foundation.org: fix warnings] [akpm@linux-foudnation.org: redo intrusive kvm changes] Tested-by: Peter Senna Tschudin <peter.senna@gmail.com> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Sasha Levin <sasha.levin@oracle.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Gleb Natapov <gleb@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 08:06:00 +07:00
hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
if (e < s_e)
goto next;
res = fib6_dump_table(tb, skb, cb);
if (res != 0)
goto out_unlock;
next:
e++;
}
}
out_unlock:
rcu_read_unlock();
cb->args[1] = e;
cb->args[0] = h;
out:
res = res < 0 ? res : skb->len;
if (res <= 0)
fib6_dump_end(cb);
return res;
}
void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
{
if (!f6i)
return;
if (f6i->fib6_metrics == &dst_default_metrics) {
struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
if (!p)
return;
refcount_set(&p->refcnt, 1);
f6i->fib6_metrics = p;
}
f6i->fib6_metrics->metrics[metric - 1] = val;
}
/*
* Routing Table
*
* return the appropriate node for a routing tree "add" operation
* by either creating and inserting or by returning an existing
* node.
*/
static struct fib6_node *fib6_add_1(struct net *net,
struct fib6_table *table,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
struct fib6_node *root,
struct in6_addr *addr, int plen,
int offset, int allow_create,
int replace_required,
struct netlink_ext_ack *extack)
{
struct fib6_node *fn, *in, *ln;
struct fib6_node *pn = NULL;
struct rt6key *key;
int bit;
__be32 dir = 0;
RT6_TRACE("fib6_add_1\n");
/* insert node in tree */
fn = root;
do {
struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
lockdep_is_held(&table->tb6_lock));
key = (struct rt6key *)((u8 *)leaf + offset);
/*
* Prefix match
*/
if (plen < fn->fn_bit ||
!ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
if (!allow_create) {
if (replace_required) {
NL_SET_ERR_MSG(extack,
"Can not replace route - no match found");
pr_warn("Can't replace route, no match found\n");
return ERR_PTR(-ENOENT);
}
pr_warn("NLM_F_CREATE should be set when creating new route\n");
}
goto insert_above;
}
/*
* Exact match ?
*/
if (plen == fn->fn_bit) {
/* clean up an intermediate node */
if (!(fn->fn_flags & RTN_RTINFO)) {
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RCU_INIT_POINTER(fn->leaf, NULL);
fib6_info_release(leaf);
ipv6: remove null_entry before adding default route In the current code, when creating a new fib6 table, tb6_root.leaf gets initialized to net->ipv6.ip6_null_entry. If a default route is being added with rt->rt6i_metric = 0xffffffff, fib6_add() will add this route after net->ipv6.ip6_null_entry. As null_entry is shared, it could cause problem. In order to fix it, set fn->leaf to NULL before calling fib6_add_rt2node() when trying to add the first default route. And reset fn->leaf to null_entry when adding fails or when deleting the last default route. syzkaller reported the following issue which is fixed by this commit: WARNING: suspicious RCU usage 4.15.0-rc5+ #171 Not tainted ----------------------------- net/ipv6/ip6_fib.c:1702 suspicious rcu_dereference_protected() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 4 locks held by swapper/0/0: #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] lockdep_copy_map include/linux/lockdep.h:178 [inline] #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] call_timer_fn+0x1c6/0x820 kernel/time/timer.c:1310 #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] spin_lock_bh include/linux/spinlock.h:315 [inline] #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] fib6_run_gc+0x9d/0x3c0 net/ipv6/ip6_fib.c:2007 #2: (rcu_read_lock){....}, at: [<0000000091db762d>] __fib6_clean_all+0x0/0x3a0 net/ipv6/ip6_fib.c:1560 #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] spin_lock_bh include/linux/spinlock.h:315 [inline] #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] __fib6_clean_all+0x1d0/0x3a0 net/ipv6/ip6_fib.c:1948 stack backtrace: CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc5+ #171 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 lockdep_rcu_suspicious+0x123/0x170 kernel/locking/lockdep.c:4585 fib6_del+0xcaa/0x11b0 net/ipv6/ip6_fib.c:1701 fib6_clean_node+0x3aa/0x4f0 net/ipv6/ip6_fib.c:1892 fib6_walk_continue+0x46c/0x8a0 net/ipv6/ip6_fib.c:1815 fib6_walk+0x91/0xf0 net/ipv6/ip6_fib.c:1863 fib6_clean_tree+0x1e6/0x340 net/ipv6/ip6_fib.c:1933 __fib6_clean_all+0x1f4/0x3a0 net/ipv6/ip6_fib.c:1949 fib6_clean_all net/ipv6/ip6_fib.c:1960 [inline] fib6_run_gc+0x16b/0x3c0 net/ipv6/ip6_fib.c:2016 fib6_gc_timer_cb+0x20/0x30 net/ipv6/ip6_fib.c:2033 call_timer_fn+0x228/0x820 kernel/time/timer.c:1320 expire_timers kernel/time/timer.c:1357 [inline] __run_timers+0x7ee/0xb70 kernel/time/timer.c:1660 run_timer_softirq+0x4c/0xb0 kernel/time/timer.c:1686 __do_softirq+0x2d7/0xb85 kernel/softirq.c:285 invoke_softirq kernel/softirq.c:365 [inline] irq_exit+0x1cc/0x200 kernel/softirq.c:405 exiting_irq arch/x86/include/asm/apic.h:540 [inline] smp_apic_timer_interrupt+0x16b/0x700 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xa9/0xb0 arch/x86/entry/entry_64.S:904 </IRQ> Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-09 01:34:00 +07:00
/* remove null_entry in the root node */
} else if (fn->fn_flags & RTN_TL_ROOT &&
rcu_access_pointer(fn->leaf) ==
net->ipv6.fib6_null_entry) {
ipv6: remove null_entry before adding default route In the current code, when creating a new fib6 table, tb6_root.leaf gets initialized to net->ipv6.ip6_null_entry. If a default route is being added with rt->rt6i_metric = 0xffffffff, fib6_add() will add this route after net->ipv6.ip6_null_entry. As null_entry is shared, it could cause problem. In order to fix it, set fn->leaf to NULL before calling fib6_add_rt2node() when trying to add the first default route. And reset fn->leaf to null_entry when adding fails or when deleting the last default route. syzkaller reported the following issue which is fixed by this commit: WARNING: suspicious RCU usage 4.15.0-rc5+ #171 Not tainted ----------------------------- net/ipv6/ip6_fib.c:1702 suspicious rcu_dereference_protected() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 4 locks held by swapper/0/0: #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] lockdep_copy_map include/linux/lockdep.h:178 [inline] #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] call_timer_fn+0x1c6/0x820 kernel/time/timer.c:1310 #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] spin_lock_bh include/linux/spinlock.h:315 [inline] #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] fib6_run_gc+0x9d/0x3c0 net/ipv6/ip6_fib.c:2007 #2: (rcu_read_lock){....}, at: [<0000000091db762d>] __fib6_clean_all+0x0/0x3a0 net/ipv6/ip6_fib.c:1560 #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] spin_lock_bh include/linux/spinlock.h:315 [inline] #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] __fib6_clean_all+0x1d0/0x3a0 net/ipv6/ip6_fib.c:1948 stack backtrace: CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc5+ #171 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 lockdep_rcu_suspicious+0x123/0x170 kernel/locking/lockdep.c:4585 fib6_del+0xcaa/0x11b0 net/ipv6/ip6_fib.c:1701 fib6_clean_node+0x3aa/0x4f0 net/ipv6/ip6_fib.c:1892 fib6_walk_continue+0x46c/0x8a0 net/ipv6/ip6_fib.c:1815 fib6_walk+0x91/0xf0 net/ipv6/ip6_fib.c:1863 fib6_clean_tree+0x1e6/0x340 net/ipv6/ip6_fib.c:1933 __fib6_clean_all+0x1f4/0x3a0 net/ipv6/ip6_fib.c:1949 fib6_clean_all net/ipv6/ip6_fib.c:1960 [inline] fib6_run_gc+0x16b/0x3c0 net/ipv6/ip6_fib.c:2016 fib6_gc_timer_cb+0x20/0x30 net/ipv6/ip6_fib.c:2033 call_timer_fn+0x228/0x820 kernel/time/timer.c:1320 expire_timers kernel/time/timer.c:1357 [inline] __run_timers+0x7ee/0xb70 kernel/time/timer.c:1660 run_timer_softirq+0x4c/0xb0 kernel/time/timer.c:1686 __do_softirq+0x2d7/0xb85 kernel/softirq.c:285 invoke_softirq kernel/softirq.c:365 [inline] irq_exit+0x1cc/0x200 kernel/softirq.c:405 exiting_irq arch/x86/include/asm/apic.h:540 [inline] smp_apic_timer_interrupt+0x16b/0x700 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xa9/0xb0 arch/x86/entry/entry_64.S:904 </IRQ> Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-09 01:34:00 +07:00
RCU_INIT_POINTER(fn->leaf, NULL);
}
return fn;
}
/*
* We have more bits to go
*/
/* Try to walk down on tree. */
dir = addr_bit_set(addr, fn->fn_bit);
pn = fn;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
fn = dir ?
rcu_dereference_protected(fn->right,
lockdep_is_held(&table->tb6_lock)) :
rcu_dereference_protected(fn->left,
lockdep_is_held(&table->tb6_lock));
} while (fn);
if (!allow_create) {
/* We should not create new node because
* NLM_F_REPLACE was specified without NLM_F_CREATE
* I assume it is safe to require NLM_F_CREATE when
* REPLACE flag is used! Later we may want to remove the
* check for replace_required, because according
* to netlink specification, NLM_F_CREATE
* MUST be specified if new route is created.
* That would keep IPv6 consistent with IPv4
*/
if (replace_required) {
NL_SET_ERR_MSG(extack,
"Can not replace route - no match found");
pr_warn("Can't replace route, no match found\n");
return ERR_PTR(-ENOENT);
}
pr_warn("NLM_F_CREATE should be set when creating new route\n");
}
/*
* We walked to the bottom of tree.
* Create new leaf node without children.
*/
ln = node_alloc(net);
if (!ln)
return ERR_PTR(-ENOMEM);
ln->fn_bit = plen;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RCU_INIT_POINTER(ln->parent, pn);
if (dir)
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(pn->right, ln);
else
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(pn->left, ln);
return ln;
insert_above:
/*
* split since we don't have a common prefix anymore or
* we have a less significant route.
* we've to insert an intermediate node on the list
* this new node will point to the one we need to create
* and the current
*/
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
pn = rcu_dereference_protected(fn->parent,
lockdep_is_held(&table->tb6_lock));
/* find 1st bit in difference between the 2 addrs.
See comment in __ipv6_addr_diff: bit may be an invalid value,
but if it is >= plen, the value is ignored in any case.
*/
bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
/*
* (intermediate)[in]
* / \
* (new leaf node)[ln] (old node)[fn]
*/
if (plen > bit) {
in = node_alloc(net);
ln = node_alloc(net);
if (!in || !ln) {
if (in)
node_free_immediate(net, in);
if (ln)
node_free_immediate(net, ln);
return ERR_PTR(-ENOMEM);
}
/*
* new intermediate node.
* RTN_RTINFO will
* be off since that an address that chooses one of
* the branches would not match less specific routes
* in the other branch
*/
in->fn_bit = bit;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RCU_INIT_POINTER(in->parent, pn);
in->leaf = fn->leaf;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
atomic_inc(&rcu_dereference_protected(in->leaf,
lockdep_is_held(&table->tb6_lock))->fib6_ref);
/* update parent pointer */
if (dir)
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(pn->right, in);
else
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(pn->left, in);
ln->fn_bit = plen;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RCU_INIT_POINTER(ln->parent, in);
rcu_assign_pointer(fn->parent, in);
if (addr_bit_set(addr, bit)) {
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(in->right, ln);
rcu_assign_pointer(in->left, fn);
} else {
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(in->left, ln);
rcu_assign_pointer(in->right, fn);
}
} else { /* plen <= bit */
/*
* (new leaf node)[ln]
* / \
* (old node)[fn] NULL
*/
ln = node_alloc(net);
if (!ln)
return ERR_PTR(-ENOMEM);
ln->fn_bit = plen;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RCU_INIT_POINTER(ln->parent, pn);
if (addr_bit_set(&key->addr, plen))
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RCU_INIT_POINTER(ln->right, fn);
else
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RCU_INIT_POINTER(ln->left, fn);
rcu_assign_pointer(fn->parent, ln);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (dir)
rcu_assign_pointer(pn->right, ln);
else
rcu_assign_pointer(pn->left, ln);
}
return ln;
}
static void fib6_drop_pcpu_from(struct fib6_info *f6i,
const struct fib6_table *table)
{
int cpu;
/* release the reference to this fib entry from
* all of its cached pcpu routes
*/
for_each_possible_cpu(cpu) {
struct rt6_info **ppcpu_rt;
struct rt6_info *pcpu_rt;
ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
pcpu_rt = *ppcpu_rt;
if (pcpu_rt) {
struct fib6_info *from;
from = rcu_dereference_protected(pcpu_rt->from,
lockdep_is_held(&table->tb6_lock));
rcu_assign_pointer(pcpu_rt->from, NULL);
fib6_info_release(from);
}
ipv6: do not overwrite inetpeer metrics prematurely If an IPv6 host route with metrics exists, an attempt to add a new route for the same target with different metrics fails but rewrites the metrics anyway: 12sp0:~ # ip route add fec0::1 dev eth0 rto_min 1000 12sp0:~ # ip -6 route show fe80::/64 dev eth0 proto kernel metric 256 fec0::1 dev eth0 metric 1024 rto_min lock 1s 12sp0:~ # ip route add fec0::1 dev eth0 rto_min 1500 RTNETLINK answers: File exists 12sp0:~ # ip -6 route show fe80::/64 dev eth0 proto kernel metric 256 fec0::1 dev eth0 metric 1024 rto_min lock 1.5s This is caused by all IPv6 host routes using the metrics in their inetpeer (or the shared default). This also holds for the new route created in ip6_route_add() which shares the metrics with the already existing route and thus ip6_route_add() rewrites the metrics even if the new route ends up not being used at all. Another problem is that old metrics in inetpeer can reappear unexpectedly for a new route, e.g. 12sp0:~ # ip route add fec0::1 dev eth0 rto_min 1000 12sp0:~ # ip route del fec0::1 12sp0:~ # ip route add fec0::1 dev eth0 12sp0:~ # ip route change fec0::1 dev eth0 hoplimit 10 12sp0:~ # ip -6 route show fe80::/64 dev eth0 proto kernel metric 256 fec0::1 dev eth0 metric 1024 hoplimit 10 rto_min lock 1s Resolve the first problem by moving the setting of metrics down into fib6_add_rt2node() to the point we are sure we are inserting the new route into the tree. Second problem is addressed by introducing new flag DST_METRICS_FORCE_OVERWRITE which is set for a new host route in ip6_route_add() and makes ipv6_cow_metrics() always overwrite the metrics in inetpeer (even if they are not "new"); it is reset after that. v5: use a flag in _metrics member rather than one in flags v4: fix a typo making a condition always true (thanks to Hannes Frederic Sowa) v3: rewritten based on David Miller's idea to move setting the metrics (and allocation in non-host case) down to the point we already know the route is to be inserted. Also rebased to net-next as it is quite late in the cycle. Signed-off-by: Michal Kubecek <mkubecek@suse.cz> Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-27 19:04:08 +07:00
}
}
static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
struct net *net)
{
struct fib6_table *table = rt->fib6_table;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (atomic_read(&rt->fib6_ref) != 1) {
/* This route is used as dummy address holder in some split
* nodes. It is not leaked, but it still holds other resources,
* which must be released in time. So, scan ascendant nodes
* and replace dummy references to this route with references
* to still alive ones.
*/
while (fn) {
struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
lockdep_is_held(&table->tb6_lock));
struct fib6_info *new_leaf;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
new_leaf = fib6_find_prefix(net, table, fn);
atomic_inc(&new_leaf->fib6_ref);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(fn->leaf, new_leaf);
fib6_info_release(rt);
}
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
fn = rcu_dereference_protected(fn->parent,
lockdep_is_held(&table->tb6_lock));
}
if (rt->rt6i_pcpu)
fib6_drop_pcpu_from(rt, table);
}
}
/*
* Insert routing information in a node.
*/
static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
struct nl_info *info,
struct netlink_ext_ack *extack)
{
struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
lockdep_is_held(&rt->fib6_table->tb6_lock));
struct fib6_info *iter = NULL;
struct fib6_info __rcu **ins;
struct fib6_info __rcu **fallback_ins = NULL;
int replace = (info->nlh &&
(info->nlh->nlmsg_flags & NLM_F_REPLACE));
int add = (!info->nlh ||
(info->nlh->nlmsg_flags & NLM_F_CREATE));
int found = 0;
bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
u16 nlflags = NLM_F_EXCL;
ipv6: do not overwrite inetpeer metrics prematurely If an IPv6 host route with metrics exists, an attempt to add a new route for the same target with different metrics fails but rewrites the metrics anyway: 12sp0:~ # ip route add fec0::1 dev eth0 rto_min 1000 12sp0:~ # ip -6 route show fe80::/64 dev eth0 proto kernel metric 256 fec0::1 dev eth0 metric 1024 rto_min lock 1s 12sp0:~ # ip route add fec0::1 dev eth0 rto_min 1500 RTNETLINK answers: File exists 12sp0:~ # ip -6 route show fe80::/64 dev eth0 proto kernel metric 256 fec0::1 dev eth0 metric 1024 rto_min lock 1.5s This is caused by all IPv6 host routes using the metrics in their inetpeer (or the shared default). This also holds for the new route created in ip6_route_add() which shares the metrics with the already existing route and thus ip6_route_add() rewrites the metrics even if the new route ends up not being used at all. Another problem is that old metrics in inetpeer can reappear unexpectedly for a new route, e.g. 12sp0:~ # ip route add fec0::1 dev eth0 rto_min 1000 12sp0:~ # ip route del fec0::1 12sp0:~ # ip route add fec0::1 dev eth0 12sp0:~ # ip route change fec0::1 dev eth0 hoplimit 10 12sp0:~ # ip -6 route show fe80::/64 dev eth0 proto kernel metric 256 fec0::1 dev eth0 metric 1024 hoplimit 10 rto_min lock 1s Resolve the first problem by moving the setting of metrics down into fib6_add_rt2node() to the point we are sure we are inserting the new route into the tree. Second problem is addressed by introducing new flag DST_METRICS_FORCE_OVERWRITE which is set for a new host route in ip6_route_add() and makes ipv6_cow_metrics() always overwrite the metrics in inetpeer (even if they are not "new"); it is reset after that. v5: use a flag in _metrics member rather than one in flags v4: fix a typo making a condition always true (thanks to Hannes Frederic Sowa) v3: rewritten based on David Miller's idea to move setting the metrics (and allocation in non-host case) down to the point we already know the route is to be inserted. Also rebased to net-next as it is quite late in the cycle. Signed-off-by: Michal Kubecek <mkubecek@suse.cz> Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-27 19:04:08 +07:00
int err;
if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
nlflags |= NLM_F_APPEND;
ins = &fn->leaf;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
for (iter = leaf; iter;
iter = rcu_dereference_protected(iter->fib6_next,
lockdep_is_held(&rt->fib6_table->tb6_lock))) {
/*
* Search for duplicates
*/
if (iter->fib6_metric == rt->fib6_metric) {
/*
* Same priority level
*/
if (info->nlh &&
(info->nlh->nlmsg_flags & NLM_F_EXCL))
return -EEXIST;
nlflags &= ~NLM_F_EXCL;
if (replace) {
if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
found++;
break;
}
if (rt_can_ecmp)
fallback_ins = fallback_ins ?: ins;
goto next_iter;
}
if (rt6_duplicate_nexthop(iter, rt)) {
if (rt->fib6_nsiblings)
rt->fib6_nsiblings = 0;
if (!(iter->fib6_flags & RTF_EXPIRES))
return -EEXIST;
if (!(rt->fib6_flags & RTF_EXPIRES))
fib6_clean_expires(iter);
else
fib6_set_expires(iter, rt->expires);
if (rt->fib6_pmtu)
fib6_metric_set(iter, RTAX_MTU,
rt->fib6_pmtu);
return -EEXIST;
}
/* If we have the same destination and the same metric,
* but not the same gateway, then the route we try to
* add is sibling to this route, increment our counter
* of siblings, and later we will add our route to the
* list.
* Only static routes (which don't have flag
* RTF_EXPIRES) are used for ECMPv6.
*
* To avoid long list, we only had siblings if the
* route have a gateway.
*/
if (rt_can_ecmp &&
rt6_qualify_for_ecmp(iter))
rt->fib6_nsiblings++;
}
if (iter->fib6_metric > rt->fib6_metric)
break;
next_iter:
ins = &iter->fib6_next;
}
if (fallback_ins && !found) {
/* No ECMP-able route found, replace first non-ECMP one */
ins = fallback_ins;
iter = rcu_dereference_protected(*ins,
lockdep_is_held(&rt->fib6_table->tb6_lock));
found++;
}
[IPV6]: Fix routing round-robin locking. As per RFC2461, section 6.3.6, item #2, when no routers on the matching list are known to be reachable or probably reachable we do round robin on those available routes so that we make sure to probe as many of them as possible to detect when one becomes reachable faster. Each routing table has a rwlock protecting the tree and the linked list of routes at each leaf. The round robin code executes during lookup and thus with the rwlock taken as a reader. A small local spinlock tries to provide protection but this does not work at all for two reasons: 1) The round-robin list manipulation, as coded, goes like this (with read lock held): walk routes finding head and tail spin_lock(); rotate list using head and tail spin_unlock(); While one thread is rotating the list, another thread can end up with stale values of head and tail and then proceed to corrupt the list when it gets the lock. This ends up causing the OOPS in fib6_add() later onthat many people have been hitting. 2) All the other code paths that run with the rwlock held as a reader do not expect the list to change on them, they expect it to remain completely fixed while they hold the lock in that way. So, simply stated, it is impossible to implement this correctly using a manipulation of the list without violating the rwlock locking semantics. Reimplement using a per-fib6_node round-robin pointer. This way we don't need to manipulate the list at all, and since the round-robin pointer can only ever point to real existing entries we don't need to perform any locking on the changing of the round-robin pointer itself. We only need to reset the round-robin pointer to NULL when the entry it is pointing to is removed. The idea is from Thomas Graf and it is very similar to how this was implemented before the advanced router selection code when in. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-03-25 10:36:25 +07:00
/* Reset round-robin state, if necessary */
if (ins == &fn->leaf)
fn->rr_ptr = NULL;
/* Link this route to others same route. */
if (rt->fib6_nsiblings) {
unsigned int fib6_nsiblings;
struct fib6_info *sibling, *temp_sibling;
/* Find the first route that have the same metric */
sibling = leaf;
while (sibling) {
if (sibling->fib6_metric == rt->fib6_metric &&
rt6_qualify_for_ecmp(sibling)) {
list_add_tail(&rt->fib6_siblings,
&sibling->fib6_siblings);
break;
}
sibling = rcu_dereference_protected(sibling->fib6_next,
lockdep_is_held(&rt->fib6_table->tb6_lock));
}
/* For each sibling in the list, increment the counter of
* siblings. BUG() if counters does not match, list of siblings
* is broken!
*/
fib6_nsiblings = 0;
list_for_each_entry_safe(sibling, temp_sibling,
&rt->fib6_siblings, fib6_siblings) {
sibling->fib6_nsiblings++;
BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
fib6_nsiblings++;
}
BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
rt6_multipath_rebalance(temp_sibling);
}
/*
* insert node
*/
if (!replace) {
if (!add)
pr_warn("NLM_F_CREATE should be set when creating new route\n");
add:
nlflags |= NLM_F_CREATE;
err = call_fib6_entry_notifiers(info->nl_net,
FIB_EVENT_ENTRY_ADD,
rt, extack);
if (err)
return err;
rcu_assign_pointer(rt->fib6_next, iter);
atomic_inc(&rt->fib6_ref);
rcu_assign_pointer(rt->fib6_node, fn);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(*ins, rt);
if (!info->skip_notify)
inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
if (!(fn->fn_flags & RTN_RTINFO)) {
info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
fn->fn_flags |= RTN_RTINFO;
}
} else {
int nsiblings;
if (!found) {
if (add)
goto add;
pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
return -ENOENT;
}
err = call_fib6_entry_notifiers(info->nl_net,
FIB_EVENT_ENTRY_REPLACE,
rt, extack);
if (err)
return err;
atomic_inc(&rt->fib6_ref);
rcu_assign_pointer(rt->fib6_node, fn);
rt->fib6_next = iter->fib6_next;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(*ins, rt);
if (!info->skip_notify)
inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
if (!(fn->fn_flags & RTN_RTINFO)) {
info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
fn->fn_flags |= RTN_RTINFO;
}
nsiblings = iter->fib6_nsiblings;
iter->fib6_node = NULL;
fib6_purge_rt(iter, fn, info->nl_net);
if (rcu_access_pointer(fn->rr_ptr) == iter)
fn->rr_ptr = NULL;
fib6_info_release(iter);
if (nsiblings) {
/* Replacing an ECMP route, remove all siblings */
ins = &rt->fib6_next;
iter = rcu_dereference_protected(*ins,
lockdep_is_held(&rt->fib6_table->tb6_lock));
while (iter) {
if (iter->fib6_metric > rt->fib6_metric)
break;
if (rt6_qualify_for_ecmp(iter)) {
*ins = iter->fib6_next;
iter->fib6_node = NULL;
fib6_purge_rt(iter, fn, info->nl_net);
if (rcu_access_pointer(fn->rr_ptr) == iter)
fn->rr_ptr = NULL;
fib6_info_release(iter);
nsiblings--;
info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
} else {
ins = &iter->fib6_next;
}
iter = rcu_dereference_protected(*ins,
lockdep_is_held(&rt->fib6_table->tb6_lock));
}
WARN_ON(nsiblings != 0);
}
}
return 0;
}
static void fib6_start_gc(struct net *net, struct fib6_info *rt)
{
if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
(rt->fib6_flags & RTF_EXPIRES))
mod_timer(&net->ipv6.ip6_fib_timer,
jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
}
void fib6_force_start_gc(struct net *net)
{
if (!timer_pending(&net->ipv6.ip6_fib_timer))
mod_timer(&net->ipv6.ip6_fib_timer,
jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
}
static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
int sernum)
ipv6: update fn_sernum after route is inserted to tree fib6_add() logic currently calls fib6_add_1() to figure out what node should be used for the newly added route and then call fib6_add_rt2node() to insert the route to the node. And during the call of fib6_add_1(), fn_sernum is updated for all nodes that share the same prefix as the new route. This does not have issue in the current code because reader thread will not be able to access the tree while writer thread is inserting new route to it. However, it is not the case once we transition to use RCU. Reader thread could potentially see the new fn_sernum before the new route is inserted. As a result, reader thread's route lookup will return a stale route with the new fn_sernum. In order to solve this issue, we remove all the update of fn_sernum in fib6_add_1(), and instead, introduce a new function that updates fn_sernum for all related nodes and call this functions once the route is successfully inserted to the tree. Also, smp_wmb() is used after a route is successfully inserted into the fib tree and right before the updated of fn->sernum. And smp_rmb() is used right after fn->sernum is accessed in rt6_get_cookie_safe(). This is to guarantee that when the reader thread sees the new fn->sernum, the new route is already inserted in the tree in memory. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:07 +07:00
{
struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
lockdep_is_held(&rt->fib6_table->tb6_lock));
ipv6: update fn_sernum after route is inserted to tree fib6_add() logic currently calls fib6_add_1() to figure out what node should be used for the newly added route and then call fib6_add_rt2node() to insert the route to the node. And during the call of fib6_add_1(), fn_sernum is updated for all nodes that share the same prefix as the new route. This does not have issue in the current code because reader thread will not be able to access the tree while writer thread is inserting new route to it. However, it is not the case once we transition to use RCU. Reader thread could potentially see the new fn_sernum before the new route is inserted. As a result, reader thread's route lookup will return a stale route with the new fn_sernum. In order to solve this issue, we remove all the update of fn_sernum in fib6_add_1(), and instead, introduce a new function that updates fn_sernum for all related nodes and call this functions once the route is successfully inserted to the tree. Also, smp_wmb() is used after a route is successfully inserted into the fib tree and right before the updated of fn->sernum. And smp_rmb() is used right after fn->sernum is accessed in rt6_get_cookie_safe(). This is to guarantee that when the reader thread sees the new fn->sernum, the new route is already inserted in the tree in memory. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:07 +07:00
/* paired with smp_rmb() in rt6_get_cookie_safe() */
smp_wmb();
while (fn) {
fn->fn_sernum = sernum;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
fn = rcu_dereference_protected(fn->parent,
lockdep_is_held(&rt->fib6_table->tb6_lock));
ipv6: update fn_sernum after route is inserted to tree fib6_add() logic currently calls fib6_add_1() to figure out what node should be used for the newly added route and then call fib6_add_rt2node() to insert the route to the node. And during the call of fib6_add_1(), fn_sernum is updated for all nodes that share the same prefix as the new route. This does not have issue in the current code because reader thread will not be able to access the tree while writer thread is inserting new route to it. However, it is not the case once we transition to use RCU. Reader thread could potentially see the new fn_sernum before the new route is inserted. As a result, reader thread's route lookup will return a stale route with the new fn_sernum. In order to solve this issue, we remove all the update of fn_sernum in fib6_add_1(), and instead, introduce a new function that updates fn_sernum for all related nodes and call this functions once the route is successfully inserted to the tree. Also, smp_wmb() is used after a route is successfully inserted into the fib tree and right before the updated of fn->sernum. And smp_rmb() is used right after fn->sernum is accessed in rt6_get_cookie_safe(). This is to guarantee that when the reader thread sees the new fn->sernum, the new route is already inserted in the tree in memory. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:07 +07:00
}
}
void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
{
__fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
}
/*
* Add routing information to the routing tree.
* <destination addr>/<source addr>
* with source addr info in sub-trees
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
* Need to own table->tb6_lock
*/
int fib6_add(struct fib6_node *root, struct fib6_info *rt,
struct nl_info *info, struct netlink_ext_ack *extack)
{
struct fib6_table *table = rt->fib6_table;
struct fib6_node *fn, *pn = NULL;
int err = -ENOMEM;
int allow_create = 1;
int replace_required = 0;
int sernum = fib6_new_sernum(info->nl_net);
if (info->nlh) {
if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
allow_create = 0;
if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
replace_required = 1;
}
if (!allow_create && !replace_required)
pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
fn = fib6_add_1(info->nl_net, table, root,
&rt->fib6_dst.addr, rt->fib6_dst.plen,
offsetof(struct fib6_info, fib6_dst), allow_create,
ipv6: update fn_sernum after route is inserted to tree fib6_add() logic currently calls fib6_add_1() to figure out what node should be used for the newly added route and then call fib6_add_rt2node() to insert the route to the node. And during the call of fib6_add_1(), fn_sernum is updated for all nodes that share the same prefix as the new route. This does not have issue in the current code because reader thread will not be able to access the tree while writer thread is inserting new route to it. However, it is not the case once we transition to use RCU. Reader thread could potentially see the new fn_sernum before the new route is inserted. As a result, reader thread's route lookup will return a stale route with the new fn_sernum. In order to solve this issue, we remove all the update of fn_sernum in fib6_add_1(), and instead, introduce a new function that updates fn_sernum for all related nodes and call this functions once the route is successfully inserted to the tree. Also, smp_wmb() is used after a route is successfully inserted into the fib tree and right before the updated of fn->sernum. And smp_rmb() is used right after fn->sernum is accessed in rt6_get_cookie_safe(). This is to guarantee that when the reader thread sees the new fn->sernum, the new route is already inserted in the tree in memory. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:07 +07:00
replace_required, extack);
if (IS_ERR(fn)) {
err = PTR_ERR(fn);
fn = NULL;
goto out;
}
pn = fn;
#ifdef CONFIG_IPV6_SUBTREES
if (rt->fib6_src.plen) {
struct fib6_node *sn;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (!rcu_access_pointer(fn->subtree)) {
struct fib6_node *sfn;
/*
* Create subtree.
*
* fn[main tree]
* |
* sfn[subtree root]
* \
* sn[new leaf node]
*/
/* Create subtree root node */
sfn = node_alloc(info->nl_net);
if (!sfn)
ipv6: repair fib6 tree in failure case In fib6_add(), it is possible that fib6_add_1() picks an intermediate node and sets the node's fn->leaf to NULL in order to add this new route. However, if fib6_add_rt2node() fails to add the new route for some reason, fn->leaf will be left as NULL and could potentially cause crash when fn->leaf is accessed in fib6_locate(). This patch makes sure fib6_repair_tree() is called to properly repair fn->leaf in the above failure case. Here is the syzkaller reported general protection fault in fib6_locate: kasan: CONFIG_KASAN_INLINE enabled kasan: GPF could be caused by NULL-ptr deref or user memory access general protection fault: 0000 [#1] SMP KASAN Modules linked in: CPU: 0 PID: 40937 Comm: syz-executor3 Not tainted Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 task: ffff8801d7d64100 ti: ffff8801d01a0000 task.ti: ffff8801d01a0000 RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] __ipv6_prefix_equal64_half include/net/ipv6.h:475 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] ipv6_prefix_equal include/net/ipv6.h:492 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate_1 net/ipv6/ip6_fib.c:1210 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate+0x281/0x3c0 net/ipv6/ip6_fib.c:1233 RSP: 0018:ffff8801d01a36a8 EFLAGS: 00010202 RAX: 0000000000000020 RBX: ffff8801bc790e00 RCX: ffffc90002983000 RDX: 0000000000001219 RSI: ffff8801d01a37a0 RDI: 0000000000000100 RBP: ffff8801d01a36f0 R08: 00000000000000ff R09: 0000000000000000 R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000000001 R13: dffffc0000000000 R14: ffff8801d01a37a0 R15: 0000000000000000 FS: 00007f6afd68c700(0000) GS:ffff8801db400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000004c6340 CR3: 00000000ba41f000 CR4: 00000000001426f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Stack: ffff8801d01a37a8 ffff8801d01a3780 ffffed003a0346f5 0000000c82a23ea0 ffff8800b7bd7700 ffff8801d01a3780 ffff8800b6a1c940 ffffffff82a23ea0 ffff8801d01a3920 ffff8801d01a3748 ffffffff82a223d6 ffff8801d7d64988 Call Trace: [<ffffffff82a223d6>] ip6_route_del+0x106/0x570 net/ipv6/route.c:2109 [<ffffffff82a23f9d>] inet6_rtm_delroute+0xfd/0x100 net/ipv6/route.c:3075 [<ffffffff82621359>] rtnetlink_rcv_msg+0x549/0x7a0 net/core/rtnetlink.c:3450 [<ffffffff8274c1d1>] netlink_rcv_skb+0x141/0x370 net/netlink/af_netlink.c:2281 [<ffffffff82613ddf>] rtnetlink_rcv+0x2f/0x40 net/core/rtnetlink.c:3456 [<ffffffff8274ad38>] netlink_unicast_kernel net/netlink/af_netlink.c:1206 [inline] [<ffffffff8274ad38>] netlink_unicast+0x518/0x750 net/netlink/af_netlink.c:1232 [<ffffffff8274b83e>] netlink_sendmsg+0x8ce/0xc30 net/netlink/af_netlink.c:1778 [<ffffffff82564aff>] sock_sendmsg_nosec net/socket.c:609 [inline] [<ffffffff82564aff>] sock_sendmsg+0xcf/0x110 net/socket.c:619 [<ffffffff82564d62>] sock_write_iter+0x222/0x3a0 net/socket.c:834 [<ffffffff8178523d>] new_sync_write+0x1dd/0x2b0 fs/read_write.c:478 [<ffffffff817853f4>] __vfs_write+0xe4/0x110 fs/read_write.c:491 [<ffffffff81786c38>] vfs_write+0x178/0x4b0 fs/read_write.c:538 [<ffffffff817892a9>] SYSC_write fs/read_write.c:585 [inline] [<ffffffff817892a9>] SyS_write+0xd9/0x1b0 fs/read_write.c:577 [<ffffffff82c71e32>] entry_SYSCALL_64_fastpath+0x12/0x17 Note: there is no "Fixes" tag as this seems to be a bug introduced very early. Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-19 07:14:49 +07:00
goto failure;
atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(sfn->leaf,
info->nl_net->ipv6.fib6_null_entry);
sfn->fn_flags = RTN_ROOT;
/* Now add the first leaf node to new subtree */
sn = fib6_add_1(info->nl_net, table, sfn,
&rt->fib6_src.addr, rt->fib6_src.plen,
offsetof(struct fib6_info, fib6_src),
ipv6: update fn_sernum after route is inserted to tree fib6_add() logic currently calls fib6_add_1() to figure out what node should be used for the newly added route and then call fib6_add_rt2node() to insert the route to the node. And during the call of fib6_add_1(), fn_sernum is updated for all nodes that share the same prefix as the new route. This does not have issue in the current code because reader thread will not be able to access the tree while writer thread is inserting new route to it. However, it is not the case once we transition to use RCU. Reader thread could potentially see the new fn_sernum before the new route is inserted. As a result, reader thread's route lookup will return a stale route with the new fn_sernum. In order to solve this issue, we remove all the update of fn_sernum in fib6_add_1(), and instead, introduce a new function that updates fn_sernum for all related nodes and call this functions once the route is successfully inserted to the tree. Also, smp_wmb() is used after a route is successfully inserted into the fib tree and right before the updated of fn->sernum. And smp_rmb() is used right after fn->sernum is accessed in rt6_get_cookie_safe(). This is to guarantee that when the reader thread sees the new fn->sernum, the new route is already inserted in the tree in memory. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:07 +07:00
allow_create, replace_required, extack);
if (IS_ERR(sn)) {
/* If it is failed, discard just allocated
ipv6: repair fib6 tree in failure case In fib6_add(), it is possible that fib6_add_1() picks an intermediate node and sets the node's fn->leaf to NULL in order to add this new route. However, if fib6_add_rt2node() fails to add the new route for some reason, fn->leaf will be left as NULL and could potentially cause crash when fn->leaf is accessed in fib6_locate(). This patch makes sure fib6_repair_tree() is called to properly repair fn->leaf in the above failure case. Here is the syzkaller reported general protection fault in fib6_locate: kasan: CONFIG_KASAN_INLINE enabled kasan: GPF could be caused by NULL-ptr deref or user memory access general protection fault: 0000 [#1] SMP KASAN Modules linked in: CPU: 0 PID: 40937 Comm: syz-executor3 Not tainted Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 task: ffff8801d7d64100 ti: ffff8801d01a0000 task.ti: ffff8801d01a0000 RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] __ipv6_prefix_equal64_half include/net/ipv6.h:475 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] ipv6_prefix_equal include/net/ipv6.h:492 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate_1 net/ipv6/ip6_fib.c:1210 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate+0x281/0x3c0 net/ipv6/ip6_fib.c:1233 RSP: 0018:ffff8801d01a36a8 EFLAGS: 00010202 RAX: 0000000000000020 RBX: ffff8801bc790e00 RCX: ffffc90002983000 RDX: 0000000000001219 RSI: ffff8801d01a37a0 RDI: 0000000000000100 RBP: ffff8801d01a36f0 R08: 00000000000000ff R09: 0000000000000000 R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000000001 R13: dffffc0000000000 R14: ffff8801d01a37a0 R15: 0000000000000000 FS: 00007f6afd68c700(0000) GS:ffff8801db400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000004c6340 CR3: 00000000ba41f000 CR4: 00000000001426f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Stack: ffff8801d01a37a8 ffff8801d01a3780 ffffed003a0346f5 0000000c82a23ea0 ffff8800b7bd7700 ffff8801d01a3780 ffff8800b6a1c940 ffffffff82a23ea0 ffff8801d01a3920 ffff8801d01a3748 ffffffff82a223d6 ffff8801d7d64988 Call Trace: [<ffffffff82a223d6>] ip6_route_del+0x106/0x570 net/ipv6/route.c:2109 [<ffffffff82a23f9d>] inet6_rtm_delroute+0xfd/0x100 net/ipv6/route.c:3075 [<ffffffff82621359>] rtnetlink_rcv_msg+0x549/0x7a0 net/core/rtnetlink.c:3450 [<ffffffff8274c1d1>] netlink_rcv_skb+0x141/0x370 net/netlink/af_netlink.c:2281 [<ffffffff82613ddf>] rtnetlink_rcv+0x2f/0x40 net/core/rtnetlink.c:3456 [<ffffffff8274ad38>] netlink_unicast_kernel net/netlink/af_netlink.c:1206 [inline] [<ffffffff8274ad38>] netlink_unicast+0x518/0x750 net/netlink/af_netlink.c:1232 [<ffffffff8274b83e>] netlink_sendmsg+0x8ce/0xc30 net/netlink/af_netlink.c:1778 [<ffffffff82564aff>] sock_sendmsg_nosec net/socket.c:609 [inline] [<ffffffff82564aff>] sock_sendmsg+0xcf/0x110 net/socket.c:619 [<ffffffff82564d62>] sock_write_iter+0x222/0x3a0 net/socket.c:834 [<ffffffff8178523d>] new_sync_write+0x1dd/0x2b0 fs/read_write.c:478 [<ffffffff817853f4>] __vfs_write+0xe4/0x110 fs/read_write.c:491 [<ffffffff81786c38>] vfs_write+0x178/0x4b0 fs/read_write.c:538 [<ffffffff817892a9>] SYSC_write fs/read_write.c:585 [inline] [<ffffffff817892a9>] SyS_write+0xd9/0x1b0 fs/read_write.c:577 [<ffffffff82c71e32>] entry_SYSCALL_64_fastpath+0x12/0x17 Note: there is no "Fixes" tag as this seems to be a bug introduced very early. Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-19 07:14:49 +07:00
root, and then (in failure) stale node
in main tree.
*/
node_free_immediate(info->nl_net, sfn);
err = PTR_ERR(sn);
ipv6: repair fib6 tree in failure case In fib6_add(), it is possible that fib6_add_1() picks an intermediate node and sets the node's fn->leaf to NULL in order to add this new route. However, if fib6_add_rt2node() fails to add the new route for some reason, fn->leaf will be left as NULL and could potentially cause crash when fn->leaf is accessed in fib6_locate(). This patch makes sure fib6_repair_tree() is called to properly repair fn->leaf in the above failure case. Here is the syzkaller reported general protection fault in fib6_locate: kasan: CONFIG_KASAN_INLINE enabled kasan: GPF could be caused by NULL-ptr deref or user memory access general protection fault: 0000 [#1] SMP KASAN Modules linked in: CPU: 0 PID: 40937 Comm: syz-executor3 Not tainted Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 task: ffff8801d7d64100 ti: ffff8801d01a0000 task.ti: ffff8801d01a0000 RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] __ipv6_prefix_equal64_half include/net/ipv6.h:475 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] ipv6_prefix_equal include/net/ipv6.h:492 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate_1 net/ipv6/ip6_fib.c:1210 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate+0x281/0x3c0 net/ipv6/ip6_fib.c:1233 RSP: 0018:ffff8801d01a36a8 EFLAGS: 00010202 RAX: 0000000000000020 RBX: ffff8801bc790e00 RCX: ffffc90002983000 RDX: 0000000000001219 RSI: ffff8801d01a37a0 RDI: 0000000000000100 RBP: ffff8801d01a36f0 R08: 00000000000000ff R09: 0000000000000000 R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000000001 R13: dffffc0000000000 R14: ffff8801d01a37a0 R15: 0000000000000000 FS: 00007f6afd68c700(0000) GS:ffff8801db400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000004c6340 CR3: 00000000ba41f000 CR4: 00000000001426f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Stack: ffff8801d01a37a8 ffff8801d01a3780 ffffed003a0346f5 0000000c82a23ea0 ffff8800b7bd7700 ffff8801d01a3780 ffff8800b6a1c940 ffffffff82a23ea0 ffff8801d01a3920 ffff8801d01a3748 ffffffff82a223d6 ffff8801d7d64988 Call Trace: [<ffffffff82a223d6>] ip6_route_del+0x106/0x570 net/ipv6/route.c:2109 [<ffffffff82a23f9d>] inet6_rtm_delroute+0xfd/0x100 net/ipv6/route.c:3075 [<ffffffff82621359>] rtnetlink_rcv_msg+0x549/0x7a0 net/core/rtnetlink.c:3450 [<ffffffff8274c1d1>] netlink_rcv_skb+0x141/0x370 net/netlink/af_netlink.c:2281 [<ffffffff82613ddf>] rtnetlink_rcv+0x2f/0x40 net/core/rtnetlink.c:3456 [<ffffffff8274ad38>] netlink_unicast_kernel net/netlink/af_netlink.c:1206 [inline] [<ffffffff8274ad38>] netlink_unicast+0x518/0x750 net/netlink/af_netlink.c:1232 [<ffffffff8274b83e>] netlink_sendmsg+0x8ce/0xc30 net/netlink/af_netlink.c:1778 [<ffffffff82564aff>] sock_sendmsg_nosec net/socket.c:609 [inline] [<ffffffff82564aff>] sock_sendmsg+0xcf/0x110 net/socket.c:619 [<ffffffff82564d62>] sock_write_iter+0x222/0x3a0 net/socket.c:834 [<ffffffff8178523d>] new_sync_write+0x1dd/0x2b0 fs/read_write.c:478 [<ffffffff817853f4>] __vfs_write+0xe4/0x110 fs/read_write.c:491 [<ffffffff81786c38>] vfs_write+0x178/0x4b0 fs/read_write.c:538 [<ffffffff817892a9>] SYSC_write fs/read_write.c:585 [inline] [<ffffffff817892a9>] SyS_write+0xd9/0x1b0 fs/read_write.c:577 [<ffffffff82c71e32>] entry_SYSCALL_64_fastpath+0x12/0x17 Note: there is no "Fixes" tag as this seems to be a bug introduced very early. Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-19 07:14:49 +07:00
goto failure;
}
/* Now link new subtree to main tree */
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(sfn->parent, fn);
rcu_assign_pointer(fn->subtree, sfn);
} else {
sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
&rt->fib6_src.addr, rt->fib6_src.plen,
offsetof(struct fib6_info, fib6_src),
ipv6: update fn_sernum after route is inserted to tree fib6_add() logic currently calls fib6_add_1() to figure out what node should be used for the newly added route and then call fib6_add_rt2node() to insert the route to the node. And during the call of fib6_add_1(), fn_sernum is updated for all nodes that share the same prefix as the new route. This does not have issue in the current code because reader thread will not be able to access the tree while writer thread is inserting new route to it. However, it is not the case once we transition to use RCU. Reader thread could potentially see the new fn_sernum before the new route is inserted. As a result, reader thread's route lookup will return a stale route with the new fn_sernum. In order to solve this issue, we remove all the update of fn_sernum in fib6_add_1(), and instead, introduce a new function that updates fn_sernum for all related nodes and call this functions once the route is successfully inserted to the tree. Also, smp_wmb() is used after a route is successfully inserted into the fib tree and right before the updated of fn->sernum. And smp_rmb() is used right after fn->sernum is accessed in rt6_get_cookie_safe(). This is to guarantee that when the reader thread sees the new fn->sernum, the new route is already inserted in the tree in memory. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:07 +07:00
allow_create, replace_required, extack);
if (IS_ERR(sn)) {
err = PTR_ERR(sn);
ipv6: repair fib6 tree in failure case In fib6_add(), it is possible that fib6_add_1() picks an intermediate node and sets the node's fn->leaf to NULL in order to add this new route. However, if fib6_add_rt2node() fails to add the new route for some reason, fn->leaf will be left as NULL and could potentially cause crash when fn->leaf is accessed in fib6_locate(). This patch makes sure fib6_repair_tree() is called to properly repair fn->leaf in the above failure case. Here is the syzkaller reported general protection fault in fib6_locate: kasan: CONFIG_KASAN_INLINE enabled kasan: GPF could be caused by NULL-ptr deref or user memory access general protection fault: 0000 [#1] SMP KASAN Modules linked in: CPU: 0 PID: 40937 Comm: syz-executor3 Not tainted Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 task: ffff8801d7d64100 ti: ffff8801d01a0000 task.ti: ffff8801d01a0000 RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] __ipv6_prefix_equal64_half include/net/ipv6.h:475 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] ipv6_prefix_equal include/net/ipv6.h:492 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate_1 net/ipv6/ip6_fib.c:1210 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate+0x281/0x3c0 net/ipv6/ip6_fib.c:1233 RSP: 0018:ffff8801d01a36a8 EFLAGS: 00010202 RAX: 0000000000000020 RBX: ffff8801bc790e00 RCX: ffffc90002983000 RDX: 0000000000001219 RSI: ffff8801d01a37a0 RDI: 0000000000000100 RBP: ffff8801d01a36f0 R08: 00000000000000ff R09: 0000000000000000 R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000000001 R13: dffffc0000000000 R14: ffff8801d01a37a0 R15: 0000000000000000 FS: 00007f6afd68c700(0000) GS:ffff8801db400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000004c6340 CR3: 00000000ba41f000 CR4: 00000000001426f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Stack: ffff8801d01a37a8 ffff8801d01a3780 ffffed003a0346f5 0000000c82a23ea0 ffff8800b7bd7700 ffff8801d01a3780 ffff8800b6a1c940 ffffffff82a23ea0 ffff8801d01a3920 ffff8801d01a3748 ffffffff82a223d6 ffff8801d7d64988 Call Trace: [<ffffffff82a223d6>] ip6_route_del+0x106/0x570 net/ipv6/route.c:2109 [<ffffffff82a23f9d>] inet6_rtm_delroute+0xfd/0x100 net/ipv6/route.c:3075 [<ffffffff82621359>] rtnetlink_rcv_msg+0x549/0x7a0 net/core/rtnetlink.c:3450 [<ffffffff8274c1d1>] netlink_rcv_skb+0x141/0x370 net/netlink/af_netlink.c:2281 [<ffffffff82613ddf>] rtnetlink_rcv+0x2f/0x40 net/core/rtnetlink.c:3456 [<ffffffff8274ad38>] netlink_unicast_kernel net/netlink/af_netlink.c:1206 [inline] [<ffffffff8274ad38>] netlink_unicast+0x518/0x750 net/netlink/af_netlink.c:1232 [<ffffffff8274b83e>] netlink_sendmsg+0x8ce/0xc30 net/netlink/af_netlink.c:1778 [<ffffffff82564aff>] sock_sendmsg_nosec net/socket.c:609 [inline] [<ffffffff82564aff>] sock_sendmsg+0xcf/0x110 net/socket.c:619 [<ffffffff82564d62>] sock_write_iter+0x222/0x3a0 net/socket.c:834 [<ffffffff8178523d>] new_sync_write+0x1dd/0x2b0 fs/read_write.c:478 [<ffffffff817853f4>] __vfs_write+0xe4/0x110 fs/read_write.c:491 [<ffffffff81786c38>] vfs_write+0x178/0x4b0 fs/read_write.c:538 [<ffffffff817892a9>] SYSC_write fs/read_write.c:585 [inline] [<ffffffff817892a9>] SyS_write+0xd9/0x1b0 fs/read_write.c:577 [<ffffffff82c71e32>] entry_SYSCALL_64_fastpath+0x12/0x17 Note: there is no "Fixes" tag as this seems to be a bug introduced very early. Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-19 07:14:49 +07:00
goto failure;
}
}
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (!rcu_access_pointer(fn->leaf)) {
ipv6: don't let tb6_root node share routes with other node After commit 4512c43eac7e, if we add a route to the subtree of tb6_root which does not have any route attached to it yet, the current code will let tb6_root and the node in the subtree share the same route. This could cause problem cause tb6_root has RTN_INFO flag marked and the tree repair and clean up code will not work properly. This commit makes sure tb6_root->leaf points back to null_entry instead of sharing route with other node. It fixes the following syzkaller reported issue: BUG: KASAN: use-after-free in ipv6_prefix_equal include/net/ipv6.h:540 [inline] BUG: KASAN: use-after-free in fib6_add_1+0x165f/0x1790 net/ipv6/ip6_fib.c:618 Read of size 8 at addr ffff8801bc043498 by task syz-executor5/19819 CPU: 1 PID: 19819 Comm: syz-executor5 Not tainted 4.15.0-rc7+ #186 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 print_address_description+0x73/0x250 mm/kasan/report.c:252 kasan_report_error mm/kasan/report.c:351 [inline] kasan_report+0x25b/0x340 mm/kasan/report.c:409 __asan_report_load8_noabort+0x14/0x20 mm/kasan/report.c:430 ipv6_prefix_equal include/net/ipv6.h:540 [inline] fib6_add_1+0x165f/0x1790 net/ipv6/ip6_fib.c:618 fib6_add+0x5fa/0x1540 net/ipv6/ip6_fib.c:1214 __ip6_ins_rt+0x6c/0x90 net/ipv6/route.c:1003 ip6_route_add+0x141/0x190 net/ipv6/route.c:2790 ipv6_route_ioctl+0x4db/0x6b0 net/ipv6/route.c:3299 inet6_ioctl+0xef/0x1e0 net/ipv6/af_inet6.c:520 sock_do_ioctl+0x65/0xb0 net/socket.c:958 sock_ioctl+0x2c2/0x440 net/socket.c:1055 vfs_ioctl fs/ioctl.c:46 [inline] do_vfs_ioctl+0x1b1/0x1520 fs/ioctl.c:686 SYSC_ioctl fs/ioctl.c:701 [inline] SyS_ioctl+0x8f/0xc0 fs/ioctl.c:692 entry_SYSCALL_64_fastpath+0x23/0x9a RIP: 0033:0x452ac9 RSP: 002b:00007fd42b321c58 EFLAGS: 00000212 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 000000000071bea0 RCX: 0000000000452ac9 RDX: 0000000020fd7000 RSI: 000000000000890b RDI: 0000000000000013 RBP: 000000000000049e R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000212 R12: 00000000006f4f70 R13: 00000000ffffffff R14: 00007fd42b3226d4 R15: 0000000000000000 Fixes: 4512c43eac7e ("ipv6: remove null_entry before adding default route") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-19 01:40:03 +07:00
if (fn->fn_flags & RTN_TL_ROOT) {
/* put back null_entry for root node */
rcu_assign_pointer(fn->leaf,
info->nl_net->ipv6.fib6_null_entry);
ipv6: don't let tb6_root node share routes with other node After commit 4512c43eac7e, if we add a route to the subtree of tb6_root which does not have any route attached to it yet, the current code will let tb6_root and the node in the subtree share the same route. This could cause problem cause tb6_root has RTN_INFO flag marked and the tree repair and clean up code will not work properly. This commit makes sure tb6_root->leaf points back to null_entry instead of sharing route with other node. It fixes the following syzkaller reported issue: BUG: KASAN: use-after-free in ipv6_prefix_equal include/net/ipv6.h:540 [inline] BUG: KASAN: use-after-free in fib6_add_1+0x165f/0x1790 net/ipv6/ip6_fib.c:618 Read of size 8 at addr ffff8801bc043498 by task syz-executor5/19819 CPU: 1 PID: 19819 Comm: syz-executor5 Not tainted 4.15.0-rc7+ #186 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 print_address_description+0x73/0x250 mm/kasan/report.c:252 kasan_report_error mm/kasan/report.c:351 [inline] kasan_report+0x25b/0x340 mm/kasan/report.c:409 __asan_report_load8_noabort+0x14/0x20 mm/kasan/report.c:430 ipv6_prefix_equal include/net/ipv6.h:540 [inline] fib6_add_1+0x165f/0x1790 net/ipv6/ip6_fib.c:618 fib6_add+0x5fa/0x1540 net/ipv6/ip6_fib.c:1214 __ip6_ins_rt+0x6c/0x90 net/ipv6/route.c:1003 ip6_route_add+0x141/0x190 net/ipv6/route.c:2790 ipv6_route_ioctl+0x4db/0x6b0 net/ipv6/route.c:3299 inet6_ioctl+0xef/0x1e0 net/ipv6/af_inet6.c:520 sock_do_ioctl+0x65/0xb0 net/socket.c:958 sock_ioctl+0x2c2/0x440 net/socket.c:1055 vfs_ioctl fs/ioctl.c:46 [inline] do_vfs_ioctl+0x1b1/0x1520 fs/ioctl.c:686 SYSC_ioctl fs/ioctl.c:701 [inline] SyS_ioctl+0x8f/0xc0 fs/ioctl.c:692 entry_SYSCALL_64_fastpath+0x23/0x9a RIP: 0033:0x452ac9 RSP: 002b:00007fd42b321c58 EFLAGS: 00000212 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 000000000071bea0 RCX: 0000000000452ac9 RDX: 0000000020fd7000 RSI: 000000000000890b RDI: 0000000000000013 RBP: 000000000000049e R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000212 R12: 00000000006f4f70 R13: 00000000ffffffff R14: 00007fd42b3226d4 R15: 0000000000000000 Fixes: 4512c43eac7e ("ipv6: remove null_entry before adding default route") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-19 01:40:03 +07:00
} else {
atomic_inc(&rt->fib6_ref);
ipv6: don't let tb6_root node share routes with other node After commit 4512c43eac7e, if we add a route to the subtree of tb6_root which does not have any route attached to it yet, the current code will let tb6_root and the node in the subtree share the same route. This could cause problem cause tb6_root has RTN_INFO flag marked and the tree repair and clean up code will not work properly. This commit makes sure tb6_root->leaf points back to null_entry instead of sharing route with other node. It fixes the following syzkaller reported issue: BUG: KASAN: use-after-free in ipv6_prefix_equal include/net/ipv6.h:540 [inline] BUG: KASAN: use-after-free in fib6_add_1+0x165f/0x1790 net/ipv6/ip6_fib.c:618 Read of size 8 at addr ffff8801bc043498 by task syz-executor5/19819 CPU: 1 PID: 19819 Comm: syz-executor5 Not tainted 4.15.0-rc7+ #186 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 print_address_description+0x73/0x250 mm/kasan/report.c:252 kasan_report_error mm/kasan/report.c:351 [inline] kasan_report+0x25b/0x340 mm/kasan/report.c:409 __asan_report_load8_noabort+0x14/0x20 mm/kasan/report.c:430 ipv6_prefix_equal include/net/ipv6.h:540 [inline] fib6_add_1+0x165f/0x1790 net/ipv6/ip6_fib.c:618 fib6_add+0x5fa/0x1540 net/ipv6/ip6_fib.c:1214 __ip6_ins_rt+0x6c/0x90 net/ipv6/route.c:1003 ip6_route_add+0x141/0x190 net/ipv6/route.c:2790 ipv6_route_ioctl+0x4db/0x6b0 net/ipv6/route.c:3299 inet6_ioctl+0xef/0x1e0 net/ipv6/af_inet6.c:520 sock_do_ioctl+0x65/0xb0 net/socket.c:958 sock_ioctl+0x2c2/0x440 net/socket.c:1055 vfs_ioctl fs/ioctl.c:46 [inline] do_vfs_ioctl+0x1b1/0x1520 fs/ioctl.c:686 SYSC_ioctl fs/ioctl.c:701 [inline] SyS_ioctl+0x8f/0xc0 fs/ioctl.c:692 entry_SYSCALL_64_fastpath+0x23/0x9a RIP: 0033:0x452ac9 RSP: 002b:00007fd42b321c58 EFLAGS: 00000212 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 000000000071bea0 RCX: 0000000000452ac9 RDX: 0000000020fd7000 RSI: 000000000000890b RDI: 0000000000000013 RBP: 000000000000049e R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000212 R12: 00000000006f4f70 R13: 00000000ffffffff R14: 00007fd42b3226d4 R15: 0000000000000000 Fixes: 4512c43eac7e ("ipv6: remove null_entry before adding default route") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-19 01:40:03 +07:00
rcu_assign_pointer(fn->leaf, rt);
}
}
fn = sn;
}
#endif
err = fib6_add_rt2node(fn, rt, info, extack);
ipv6: update fn_sernum after route is inserted to tree fib6_add() logic currently calls fib6_add_1() to figure out what node should be used for the newly added route and then call fib6_add_rt2node() to insert the route to the node. And during the call of fib6_add_1(), fn_sernum is updated for all nodes that share the same prefix as the new route. This does not have issue in the current code because reader thread will not be able to access the tree while writer thread is inserting new route to it. However, it is not the case once we transition to use RCU. Reader thread could potentially see the new fn_sernum before the new route is inserted. As a result, reader thread's route lookup will return a stale route with the new fn_sernum. In order to solve this issue, we remove all the update of fn_sernum in fib6_add_1(), and instead, introduce a new function that updates fn_sernum for all related nodes and call this functions once the route is successfully inserted to the tree. Also, smp_wmb() is used after a route is successfully inserted into the fib tree and right before the updated of fn->sernum. And smp_rmb() is used right after fn->sernum is accessed in rt6_get_cookie_safe(). This is to guarantee that when the reader thread sees the new fn->sernum, the new route is already inserted in the tree in memory. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:07 +07:00
if (!err) {
__fib6_update_sernum_upto_root(rt, sernum);
fib6_start_gc(info->nl_net, rt);
ipv6: update fn_sernum after route is inserted to tree fib6_add() logic currently calls fib6_add_1() to figure out what node should be used for the newly added route and then call fib6_add_rt2node() to insert the route to the node. And during the call of fib6_add_1(), fn_sernum is updated for all nodes that share the same prefix as the new route. This does not have issue in the current code because reader thread will not be able to access the tree while writer thread is inserting new route to it. However, it is not the case once we transition to use RCU. Reader thread could potentially see the new fn_sernum before the new route is inserted. As a result, reader thread's route lookup will return a stale route with the new fn_sernum. In order to solve this issue, we remove all the update of fn_sernum in fib6_add_1(), and instead, introduce a new function that updates fn_sernum for all related nodes and call this functions once the route is successfully inserted to the tree. Also, smp_wmb() is used after a route is successfully inserted into the fib tree and right before the updated of fn->sernum. And smp_rmb() is used right after fn->sernum is accessed in rt6_get_cookie_safe(). This is to guarantee that when the reader thread sees the new fn->sernum, the new route is already inserted in the tree in memory. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:07 +07:00
}
out:
if (err) {
#ifdef CONFIG_IPV6_SUBTREES
/*
* If fib6_add_1 has cleared the old leaf pointer in the
* super-tree leaf node we have to find a new one for it.
*/
ipv6: fix general protection fault in fib6_add() In fib6_add(), pn could be NULL if fib6_add_1() failed to return a fib6 node. Checking pn != fn before accessing pn->leaf makes sure pn is not NULL. This fixes the following GPF reported by syzkaller: general protection fault: 0000 [#1] SMP KASAN Dumping ftrace buffer: (ftrace buffer empty) Modules linked in: CPU: 0 PID: 3201 Comm: syzkaller001778 Not tainted 4.15.0-rc5+ #151 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 RIP: 0010:fib6_add+0x736/0x15a0 net/ipv6/ip6_fib.c:1244 RSP: 0018:ffff8801c7626a70 EFLAGS: 00010202 RAX: dffffc0000000000 RBX: 0000000000000020 RCX: ffffffff84794465 RDX: 0000000000000004 RSI: ffff8801d38935f0 RDI: 0000000000000282 RBP: ffff8801c7626da0 R08: 1ffff10038ec4c35 R09: 0000000000000000 R10: ffff8801c7626c68 R11: 0000000000000000 R12: 00000000fffffffe R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000009 FS: 0000000000000000(0000) GS:ffff8801db200000(0063) knlGS:0000000009b70840 CS: 0010 DS: 002b ES: 002b CR0: 0000000080050033 CR2: 0000000020be1000 CR3: 00000001d585a006 CR4: 00000000001606f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: __ip6_ins_rt+0x6c/0x90 net/ipv6/route.c:1006 ip6_route_multipath_add+0xd14/0x16c0 net/ipv6/route.c:3833 inet6_rtm_newroute+0xdc/0x160 net/ipv6/route.c:3957 rtnetlink_rcv_msg+0x733/0x1020 net/core/rtnetlink.c:4411 netlink_rcv_skb+0x21e/0x460 net/netlink/af_netlink.c:2408 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4423 netlink_unicast_kernel net/netlink/af_netlink.c:1275 [inline] netlink_unicast+0x4e8/0x6f0 net/netlink/af_netlink.c:1301 netlink_sendmsg+0xa4a/0xe60 net/netlink/af_netlink.c:1864 sock_sendmsg_nosec net/socket.c:636 [inline] sock_sendmsg+0xca/0x110 net/socket.c:646 sock_write_iter+0x31a/0x5d0 net/socket.c:915 call_write_iter include/linux/fs.h:1772 [inline] do_iter_readv_writev+0x525/0x7f0 fs/read_write.c:653 do_iter_write+0x154/0x540 fs/read_write.c:932 compat_writev+0x225/0x420 fs/read_write.c:1246 do_compat_writev+0x115/0x220 fs/read_write.c:1267 C_SYSC_writev fs/read_write.c:1278 [inline] compat_SyS_writev+0x26/0x30 fs/read_write.c:1274 do_syscall_32_irqs_on arch/x86/entry/common.c:327 [inline] do_fast_syscall_32+0x3ee/0xf9d arch/x86/entry/common.c:389 entry_SYSENTER_compat+0x54/0x63 arch/x86/entry/entry_64_compat.S:125 Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-04 05:11:59 +07:00
if (pn != fn) {
struct fib6_info *pn_leaf =
ipv6: fix general protection fault in fib6_add() In fib6_add(), pn could be NULL if fib6_add_1() failed to return a fib6 node. Checking pn != fn before accessing pn->leaf makes sure pn is not NULL. This fixes the following GPF reported by syzkaller: general protection fault: 0000 [#1] SMP KASAN Dumping ftrace buffer: (ftrace buffer empty) Modules linked in: CPU: 0 PID: 3201 Comm: syzkaller001778 Not tainted 4.15.0-rc5+ #151 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 RIP: 0010:fib6_add+0x736/0x15a0 net/ipv6/ip6_fib.c:1244 RSP: 0018:ffff8801c7626a70 EFLAGS: 00010202 RAX: dffffc0000000000 RBX: 0000000000000020 RCX: ffffffff84794465 RDX: 0000000000000004 RSI: ffff8801d38935f0 RDI: 0000000000000282 RBP: ffff8801c7626da0 R08: 1ffff10038ec4c35 R09: 0000000000000000 R10: ffff8801c7626c68 R11: 0000000000000000 R12: 00000000fffffffe R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000009 FS: 0000000000000000(0000) GS:ffff8801db200000(0063) knlGS:0000000009b70840 CS: 0010 DS: 002b ES: 002b CR0: 0000000080050033 CR2: 0000000020be1000 CR3: 00000001d585a006 CR4: 00000000001606f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: __ip6_ins_rt+0x6c/0x90 net/ipv6/route.c:1006 ip6_route_multipath_add+0xd14/0x16c0 net/ipv6/route.c:3833 inet6_rtm_newroute+0xdc/0x160 net/ipv6/route.c:3957 rtnetlink_rcv_msg+0x733/0x1020 net/core/rtnetlink.c:4411 netlink_rcv_skb+0x21e/0x460 net/netlink/af_netlink.c:2408 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4423 netlink_unicast_kernel net/netlink/af_netlink.c:1275 [inline] netlink_unicast+0x4e8/0x6f0 net/netlink/af_netlink.c:1301 netlink_sendmsg+0xa4a/0xe60 net/netlink/af_netlink.c:1864 sock_sendmsg_nosec net/socket.c:636 [inline] sock_sendmsg+0xca/0x110 net/socket.c:646 sock_write_iter+0x31a/0x5d0 net/socket.c:915 call_write_iter include/linux/fs.h:1772 [inline] do_iter_readv_writev+0x525/0x7f0 fs/read_write.c:653 do_iter_write+0x154/0x540 fs/read_write.c:932 compat_writev+0x225/0x420 fs/read_write.c:1246 do_compat_writev+0x115/0x220 fs/read_write.c:1267 C_SYSC_writev fs/read_write.c:1278 [inline] compat_SyS_writev+0x26/0x30 fs/read_write.c:1274 do_syscall_32_irqs_on arch/x86/entry/common.c:327 [inline] do_fast_syscall_32+0x3ee/0xf9d arch/x86/entry/common.c:389 entry_SYSENTER_compat+0x54/0x63 arch/x86/entry/entry_64_compat.S:125 Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-04 05:11:59 +07:00
rcu_dereference_protected(pn->leaf,
lockdep_is_held(&table->tb6_lock));
if (pn_leaf == rt) {
pn_leaf = NULL;
RCU_INIT_POINTER(pn->leaf, NULL);
fib6_info_release(rt);
}
ipv6: fix general protection fault in fib6_add() In fib6_add(), pn could be NULL if fib6_add_1() failed to return a fib6 node. Checking pn != fn before accessing pn->leaf makes sure pn is not NULL. This fixes the following GPF reported by syzkaller: general protection fault: 0000 [#1] SMP KASAN Dumping ftrace buffer: (ftrace buffer empty) Modules linked in: CPU: 0 PID: 3201 Comm: syzkaller001778 Not tainted 4.15.0-rc5+ #151 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 RIP: 0010:fib6_add+0x736/0x15a0 net/ipv6/ip6_fib.c:1244 RSP: 0018:ffff8801c7626a70 EFLAGS: 00010202 RAX: dffffc0000000000 RBX: 0000000000000020 RCX: ffffffff84794465 RDX: 0000000000000004 RSI: ffff8801d38935f0 RDI: 0000000000000282 RBP: ffff8801c7626da0 R08: 1ffff10038ec4c35 R09: 0000000000000000 R10: ffff8801c7626c68 R11: 0000000000000000 R12: 00000000fffffffe R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000009 FS: 0000000000000000(0000) GS:ffff8801db200000(0063) knlGS:0000000009b70840 CS: 0010 DS: 002b ES: 002b CR0: 0000000080050033 CR2: 0000000020be1000 CR3: 00000001d585a006 CR4: 00000000001606f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: __ip6_ins_rt+0x6c/0x90 net/ipv6/route.c:1006 ip6_route_multipath_add+0xd14/0x16c0 net/ipv6/route.c:3833 inet6_rtm_newroute+0xdc/0x160 net/ipv6/route.c:3957 rtnetlink_rcv_msg+0x733/0x1020 net/core/rtnetlink.c:4411 netlink_rcv_skb+0x21e/0x460 net/netlink/af_netlink.c:2408 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4423 netlink_unicast_kernel net/netlink/af_netlink.c:1275 [inline] netlink_unicast+0x4e8/0x6f0 net/netlink/af_netlink.c:1301 netlink_sendmsg+0xa4a/0xe60 net/netlink/af_netlink.c:1864 sock_sendmsg_nosec net/socket.c:636 [inline] sock_sendmsg+0xca/0x110 net/socket.c:646 sock_write_iter+0x31a/0x5d0 net/socket.c:915 call_write_iter include/linux/fs.h:1772 [inline] do_iter_readv_writev+0x525/0x7f0 fs/read_write.c:653 do_iter_write+0x154/0x540 fs/read_write.c:932 compat_writev+0x225/0x420 fs/read_write.c:1246 do_compat_writev+0x115/0x220 fs/read_write.c:1267 C_SYSC_writev fs/read_write.c:1278 [inline] compat_SyS_writev+0x26/0x30 fs/read_write.c:1274 do_syscall_32_irqs_on arch/x86/entry/common.c:327 [inline] do_fast_syscall_32+0x3ee/0xf9d arch/x86/entry/common.c:389 entry_SYSENTER_compat+0x54/0x63 arch/x86/entry/entry_64_compat.S:125 Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-04 05:11:59 +07:00
if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
pn_leaf = fib6_find_prefix(info->nl_net, table,
pn);
#if RT6_DEBUG >= 2
if (!pn_leaf) {
WARN_ON(!pn_leaf);
pn_leaf =
info->nl_net->ipv6.fib6_null_entry;
ipv6: fix general protection fault in fib6_add() In fib6_add(), pn could be NULL if fib6_add_1() failed to return a fib6 node. Checking pn != fn before accessing pn->leaf makes sure pn is not NULL. This fixes the following GPF reported by syzkaller: general protection fault: 0000 [#1] SMP KASAN Dumping ftrace buffer: (ftrace buffer empty) Modules linked in: CPU: 0 PID: 3201 Comm: syzkaller001778 Not tainted 4.15.0-rc5+ #151 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 RIP: 0010:fib6_add+0x736/0x15a0 net/ipv6/ip6_fib.c:1244 RSP: 0018:ffff8801c7626a70 EFLAGS: 00010202 RAX: dffffc0000000000 RBX: 0000000000000020 RCX: ffffffff84794465 RDX: 0000000000000004 RSI: ffff8801d38935f0 RDI: 0000000000000282 RBP: ffff8801c7626da0 R08: 1ffff10038ec4c35 R09: 0000000000000000 R10: ffff8801c7626c68 R11: 0000000000000000 R12: 00000000fffffffe R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000009 FS: 0000000000000000(0000) GS:ffff8801db200000(0063) knlGS:0000000009b70840 CS: 0010 DS: 002b ES: 002b CR0: 0000000080050033 CR2: 0000000020be1000 CR3: 00000001d585a006 CR4: 00000000001606f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: __ip6_ins_rt+0x6c/0x90 net/ipv6/route.c:1006 ip6_route_multipath_add+0xd14/0x16c0 net/ipv6/route.c:3833 inet6_rtm_newroute+0xdc/0x160 net/ipv6/route.c:3957 rtnetlink_rcv_msg+0x733/0x1020 net/core/rtnetlink.c:4411 netlink_rcv_skb+0x21e/0x460 net/netlink/af_netlink.c:2408 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4423 netlink_unicast_kernel net/netlink/af_netlink.c:1275 [inline] netlink_unicast+0x4e8/0x6f0 net/netlink/af_netlink.c:1301 netlink_sendmsg+0xa4a/0xe60 net/netlink/af_netlink.c:1864 sock_sendmsg_nosec net/socket.c:636 [inline] sock_sendmsg+0xca/0x110 net/socket.c:646 sock_write_iter+0x31a/0x5d0 net/socket.c:915 call_write_iter include/linux/fs.h:1772 [inline] do_iter_readv_writev+0x525/0x7f0 fs/read_write.c:653 do_iter_write+0x154/0x540 fs/read_write.c:932 compat_writev+0x225/0x420 fs/read_write.c:1246 do_compat_writev+0x115/0x220 fs/read_write.c:1267 C_SYSC_writev fs/read_write.c:1278 [inline] compat_SyS_writev+0x26/0x30 fs/read_write.c:1274 do_syscall_32_irqs_on arch/x86/entry/common.c:327 [inline] do_fast_syscall_32+0x3ee/0xf9d arch/x86/entry/common.c:389 entry_SYSENTER_compat+0x54/0x63 arch/x86/entry/entry_64_compat.S:125 Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-04 05:11:59 +07:00
}
#endif
fib6_info_hold(pn_leaf);
ipv6: fix general protection fault in fib6_add() In fib6_add(), pn could be NULL if fib6_add_1() failed to return a fib6 node. Checking pn != fn before accessing pn->leaf makes sure pn is not NULL. This fixes the following GPF reported by syzkaller: general protection fault: 0000 [#1] SMP KASAN Dumping ftrace buffer: (ftrace buffer empty) Modules linked in: CPU: 0 PID: 3201 Comm: syzkaller001778 Not tainted 4.15.0-rc5+ #151 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 RIP: 0010:fib6_add+0x736/0x15a0 net/ipv6/ip6_fib.c:1244 RSP: 0018:ffff8801c7626a70 EFLAGS: 00010202 RAX: dffffc0000000000 RBX: 0000000000000020 RCX: ffffffff84794465 RDX: 0000000000000004 RSI: ffff8801d38935f0 RDI: 0000000000000282 RBP: ffff8801c7626da0 R08: 1ffff10038ec4c35 R09: 0000000000000000 R10: ffff8801c7626c68 R11: 0000000000000000 R12: 00000000fffffffe R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000009 FS: 0000000000000000(0000) GS:ffff8801db200000(0063) knlGS:0000000009b70840 CS: 0010 DS: 002b ES: 002b CR0: 0000000080050033 CR2: 0000000020be1000 CR3: 00000001d585a006 CR4: 00000000001606f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: __ip6_ins_rt+0x6c/0x90 net/ipv6/route.c:1006 ip6_route_multipath_add+0xd14/0x16c0 net/ipv6/route.c:3833 inet6_rtm_newroute+0xdc/0x160 net/ipv6/route.c:3957 rtnetlink_rcv_msg+0x733/0x1020 net/core/rtnetlink.c:4411 netlink_rcv_skb+0x21e/0x460 net/netlink/af_netlink.c:2408 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4423 netlink_unicast_kernel net/netlink/af_netlink.c:1275 [inline] netlink_unicast+0x4e8/0x6f0 net/netlink/af_netlink.c:1301 netlink_sendmsg+0xa4a/0xe60 net/netlink/af_netlink.c:1864 sock_sendmsg_nosec net/socket.c:636 [inline] sock_sendmsg+0xca/0x110 net/socket.c:646 sock_write_iter+0x31a/0x5d0 net/socket.c:915 call_write_iter include/linux/fs.h:1772 [inline] do_iter_readv_writev+0x525/0x7f0 fs/read_write.c:653 do_iter_write+0x154/0x540 fs/read_write.c:932 compat_writev+0x225/0x420 fs/read_write.c:1246 do_compat_writev+0x115/0x220 fs/read_write.c:1267 C_SYSC_writev fs/read_write.c:1278 [inline] compat_SyS_writev+0x26/0x30 fs/read_write.c:1274 do_syscall_32_irqs_on arch/x86/entry/common.c:327 [inline] do_fast_syscall_32+0x3ee/0xf9d arch/x86/entry/common.c:389 entry_SYSENTER_compat+0x54/0x63 arch/x86/entry/entry_64_compat.S:125 Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-04 05:11:59 +07:00
rcu_assign_pointer(pn->leaf, pn_leaf);
}
}
#endif
ipv6: repair fib6 tree in failure case In fib6_add(), it is possible that fib6_add_1() picks an intermediate node and sets the node's fn->leaf to NULL in order to add this new route. However, if fib6_add_rt2node() fails to add the new route for some reason, fn->leaf will be left as NULL and could potentially cause crash when fn->leaf is accessed in fib6_locate(). This patch makes sure fib6_repair_tree() is called to properly repair fn->leaf in the above failure case. Here is the syzkaller reported general protection fault in fib6_locate: kasan: CONFIG_KASAN_INLINE enabled kasan: GPF could be caused by NULL-ptr deref or user memory access general protection fault: 0000 [#1] SMP KASAN Modules linked in: CPU: 0 PID: 40937 Comm: syz-executor3 Not tainted Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 task: ffff8801d7d64100 ti: ffff8801d01a0000 task.ti: ffff8801d01a0000 RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] __ipv6_prefix_equal64_half include/net/ipv6.h:475 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] ipv6_prefix_equal include/net/ipv6.h:492 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate_1 net/ipv6/ip6_fib.c:1210 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate+0x281/0x3c0 net/ipv6/ip6_fib.c:1233 RSP: 0018:ffff8801d01a36a8 EFLAGS: 00010202 RAX: 0000000000000020 RBX: ffff8801bc790e00 RCX: ffffc90002983000 RDX: 0000000000001219 RSI: ffff8801d01a37a0 RDI: 0000000000000100 RBP: ffff8801d01a36f0 R08: 00000000000000ff R09: 0000000000000000 R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000000001 R13: dffffc0000000000 R14: ffff8801d01a37a0 R15: 0000000000000000 FS: 00007f6afd68c700(0000) GS:ffff8801db400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000004c6340 CR3: 00000000ba41f000 CR4: 00000000001426f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Stack: ffff8801d01a37a8 ffff8801d01a3780 ffffed003a0346f5 0000000c82a23ea0 ffff8800b7bd7700 ffff8801d01a3780 ffff8800b6a1c940 ffffffff82a23ea0 ffff8801d01a3920 ffff8801d01a3748 ffffffff82a223d6 ffff8801d7d64988 Call Trace: [<ffffffff82a223d6>] ip6_route_del+0x106/0x570 net/ipv6/route.c:2109 [<ffffffff82a23f9d>] inet6_rtm_delroute+0xfd/0x100 net/ipv6/route.c:3075 [<ffffffff82621359>] rtnetlink_rcv_msg+0x549/0x7a0 net/core/rtnetlink.c:3450 [<ffffffff8274c1d1>] netlink_rcv_skb+0x141/0x370 net/netlink/af_netlink.c:2281 [<ffffffff82613ddf>] rtnetlink_rcv+0x2f/0x40 net/core/rtnetlink.c:3456 [<ffffffff8274ad38>] netlink_unicast_kernel net/netlink/af_netlink.c:1206 [inline] [<ffffffff8274ad38>] netlink_unicast+0x518/0x750 net/netlink/af_netlink.c:1232 [<ffffffff8274b83e>] netlink_sendmsg+0x8ce/0xc30 net/netlink/af_netlink.c:1778 [<ffffffff82564aff>] sock_sendmsg_nosec net/socket.c:609 [inline] [<ffffffff82564aff>] sock_sendmsg+0xcf/0x110 net/socket.c:619 [<ffffffff82564d62>] sock_write_iter+0x222/0x3a0 net/socket.c:834 [<ffffffff8178523d>] new_sync_write+0x1dd/0x2b0 fs/read_write.c:478 [<ffffffff817853f4>] __vfs_write+0xe4/0x110 fs/read_write.c:491 [<ffffffff81786c38>] vfs_write+0x178/0x4b0 fs/read_write.c:538 [<ffffffff817892a9>] SYSC_write fs/read_write.c:585 [inline] [<ffffffff817892a9>] SyS_write+0xd9/0x1b0 fs/read_write.c:577 [<ffffffff82c71e32>] entry_SYSCALL_64_fastpath+0x12/0x17 Note: there is no "Fixes" tag as this seems to be a bug introduced very early. Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-19 07:14:49 +07:00
goto failure;
}
return err;
ipv6: repair fib6 tree in failure case In fib6_add(), it is possible that fib6_add_1() picks an intermediate node and sets the node's fn->leaf to NULL in order to add this new route. However, if fib6_add_rt2node() fails to add the new route for some reason, fn->leaf will be left as NULL and could potentially cause crash when fn->leaf is accessed in fib6_locate(). This patch makes sure fib6_repair_tree() is called to properly repair fn->leaf in the above failure case. Here is the syzkaller reported general protection fault in fib6_locate: kasan: CONFIG_KASAN_INLINE enabled kasan: GPF could be caused by NULL-ptr deref or user memory access general protection fault: 0000 [#1] SMP KASAN Modules linked in: CPU: 0 PID: 40937 Comm: syz-executor3 Not tainted Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 task: ffff8801d7d64100 ti: ffff8801d01a0000 task.ti: ffff8801d01a0000 RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] __ipv6_prefix_equal64_half include/net/ipv6.h:475 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] ipv6_prefix_equal include/net/ipv6.h:492 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate_1 net/ipv6/ip6_fib.c:1210 [inline] RIP: 0010:[<ffffffff82a3e0e1>] [<ffffffff82a3e0e1>] fib6_locate+0x281/0x3c0 net/ipv6/ip6_fib.c:1233 RSP: 0018:ffff8801d01a36a8 EFLAGS: 00010202 RAX: 0000000000000020 RBX: ffff8801bc790e00 RCX: ffffc90002983000 RDX: 0000000000001219 RSI: ffff8801d01a37a0 RDI: 0000000000000100 RBP: ffff8801d01a36f0 R08: 00000000000000ff R09: 0000000000000000 R10: 0000000000000003 R11: 0000000000000000 R12: 0000000000000001 R13: dffffc0000000000 R14: ffff8801d01a37a0 R15: 0000000000000000 FS: 00007f6afd68c700(0000) GS:ffff8801db400000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00000000004c6340 CR3: 00000000ba41f000 CR4: 00000000001426f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Stack: ffff8801d01a37a8 ffff8801d01a3780 ffffed003a0346f5 0000000c82a23ea0 ffff8800b7bd7700 ffff8801d01a3780 ffff8800b6a1c940 ffffffff82a23ea0 ffff8801d01a3920 ffff8801d01a3748 ffffffff82a223d6 ffff8801d7d64988 Call Trace: [<ffffffff82a223d6>] ip6_route_del+0x106/0x570 net/ipv6/route.c:2109 [<ffffffff82a23f9d>] inet6_rtm_delroute+0xfd/0x100 net/ipv6/route.c:3075 [<ffffffff82621359>] rtnetlink_rcv_msg+0x549/0x7a0 net/core/rtnetlink.c:3450 [<ffffffff8274c1d1>] netlink_rcv_skb+0x141/0x370 net/netlink/af_netlink.c:2281 [<ffffffff82613ddf>] rtnetlink_rcv+0x2f/0x40 net/core/rtnetlink.c:3456 [<ffffffff8274ad38>] netlink_unicast_kernel net/netlink/af_netlink.c:1206 [inline] [<ffffffff8274ad38>] netlink_unicast+0x518/0x750 net/netlink/af_netlink.c:1232 [<ffffffff8274b83e>] netlink_sendmsg+0x8ce/0xc30 net/netlink/af_netlink.c:1778 [<ffffffff82564aff>] sock_sendmsg_nosec net/socket.c:609 [inline] [<ffffffff82564aff>] sock_sendmsg+0xcf/0x110 net/socket.c:619 [<ffffffff82564d62>] sock_write_iter+0x222/0x3a0 net/socket.c:834 [<ffffffff8178523d>] new_sync_write+0x1dd/0x2b0 fs/read_write.c:478 [<ffffffff817853f4>] __vfs_write+0xe4/0x110 fs/read_write.c:491 [<ffffffff81786c38>] vfs_write+0x178/0x4b0 fs/read_write.c:538 [<ffffffff817892a9>] SYSC_write fs/read_write.c:585 [inline] [<ffffffff817892a9>] SyS_write+0xd9/0x1b0 fs/read_write.c:577 [<ffffffff82c71e32>] entry_SYSCALL_64_fastpath+0x12/0x17 Note: there is no "Fixes" tag as this seems to be a bug introduced very early. Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-19 07:14:49 +07:00
failure:
ipv6: remove null_entry before adding default route In the current code, when creating a new fib6 table, tb6_root.leaf gets initialized to net->ipv6.ip6_null_entry. If a default route is being added with rt->rt6i_metric = 0xffffffff, fib6_add() will add this route after net->ipv6.ip6_null_entry. As null_entry is shared, it could cause problem. In order to fix it, set fn->leaf to NULL before calling fib6_add_rt2node() when trying to add the first default route. And reset fn->leaf to null_entry when adding fails or when deleting the last default route. syzkaller reported the following issue which is fixed by this commit: WARNING: suspicious RCU usage 4.15.0-rc5+ #171 Not tainted ----------------------------- net/ipv6/ip6_fib.c:1702 suspicious rcu_dereference_protected() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 4 locks held by swapper/0/0: #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] lockdep_copy_map include/linux/lockdep.h:178 [inline] #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] call_timer_fn+0x1c6/0x820 kernel/time/timer.c:1310 #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] spin_lock_bh include/linux/spinlock.h:315 [inline] #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] fib6_run_gc+0x9d/0x3c0 net/ipv6/ip6_fib.c:2007 #2: (rcu_read_lock){....}, at: [<0000000091db762d>] __fib6_clean_all+0x0/0x3a0 net/ipv6/ip6_fib.c:1560 #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] spin_lock_bh include/linux/spinlock.h:315 [inline] #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] __fib6_clean_all+0x1d0/0x3a0 net/ipv6/ip6_fib.c:1948 stack backtrace: CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc5+ #171 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 lockdep_rcu_suspicious+0x123/0x170 kernel/locking/lockdep.c:4585 fib6_del+0xcaa/0x11b0 net/ipv6/ip6_fib.c:1701 fib6_clean_node+0x3aa/0x4f0 net/ipv6/ip6_fib.c:1892 fib6_walk_continue+0x46c/0x8a0 net/ipv6/ip6_fib.c:1815 fib6_walk+0x91/0xf0 net/ipv6/ip6_fib.c:1863 fib6_clean_tree+0x1e6/0x340 net/ipv6/ip6_fib.c:1933 __fib6_clean_all+0x1f4/0x3a0 net/ipv6/ip6_fib.c:1949 fib6_clean_all net/ipv6/ip6_fib.c:1960 [inline] fib6_run_gc+0x16b/0x3c0 net/ipv6/ip6_fib.c:2016 fib6_gc_timer_cb+0x20/0x30 net/ipv6/ip6_fib.c:2033 call_timer_fn+0x228/0x820 kernel/time/timer.c:1320 expire_timers kernel/time/timer.c:1357 [inline] __run_timers+0x7ee/0xb70 kernel/time/timer.c:1660 run_timer_softirq+0x4c/0xb0 kernel/time/timer.c:1686 __do_softirq+0x2d7/0xb85 kernel/softirq.c:285 invoke_softirq kernel/softirq.c:365 [inline] irq_exit+0x1cc/0x200 kernel/softirq.c:405 exiting_irq arch/x86/include/asm/apic.h:540 [inline] smp_apic_timer_interrupt+0x16b/0x700 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xa9/0xb0 arch/x86/entry/entry_64.S:904 </IRQ> Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-09 01:34:00 +07:00
/* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
* 1. fn is an intermediate node and we failed to add the new
* route to it in both subtree creation failure and fib6_add_rt2node()
* failure case.
* 2. fn is the root node in the table and we fail to add the first
* default route to it.
*/
ipv6: remove null_entry before adding default route In the current code, when creating a new fib6 table, tb6_root.leaf gets initialized to net->ipv6.ip6_null_entry. If a default route is being added with rt->rt6i_metric = 0xffffffff, fib6_add() will add this route after net->ipv6.ip6_null_entry. As null_entry is shared, it could cause problem. In order to fix it, set fn->leaf to NULL before calling fib6_add_rt2node() when trying to add the first default route. And reset fn->leaf to null_entry when adding fails or when deleting the last default route. syzkaller reported the following issue which is fixed by this commit: WARNING: suspicious RCU usage 4.15.0-rc5+ #171 Not tainted ----------------------------- net/ipv6/ip6_fib.c:1702 suspicious rcu_dereference_protected() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 4 locks held by swapper/0/0: #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] lockdep_copy_map include/linux/lockdep.h:178 [inline] #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] call_timer_fn+0x1c6/0x820 kernel/time/timer.c:1310 #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] spin_lock_bh include/linux/spinlock.h:315 [inline] #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] fib6_run_gc+0x9d/0x3c0 net/ipv6/ip6_fib.c:2007 #2: (rcu_read_lock){....}, at: [<0000000091db762d>] __fib6_clean_all+0x0/0x3a0 net/ipv6/ip6_fib.c:1560 #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] spin_lock_bh include/linux/spinlock.h:315 [inline] #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] __fib6_clean_all+0x1d0/0x3a0 net/ipv6/ip6_fib.c:1948 stack backtrace: CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc5+ #171 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 lockdep_rcu_suspicious+0x123/0x170 kernel/locking/lockdep.c:4585 fib6_del+0xcaa/0x11b0 net/ipv6/ip6_fib.c:1701 fib6_clean_node+0x3aa/0x4f0 net/ipv6/ip6_fib.c:1892 fib6_walk_continue+0x46c/0x8a0 net/ipv6/ip6_fib.c:1815 fib6_walk+0x91/0xf0 net/ipv6/ip6_fib.c:1863 fib6_clean_tree+0x1e6/0x340 net/ipv6/ip6_fib.c:1933 __fib6_clean_all+0x1f4/0x3a0 net/ipv6/ip6_fib.c:1949 fib6_clean_all net/ipv6/ip6_fib.c:1960 [inline] fib6_run_gc+0x16b/0x3c0 net/ipv6/ip6_fib.c:2016 fib6_gc_timer_cb+0x20/0x30 net/ipv6/ip6_fib.c:2033 call_timer_fn+0x228/0x820 kernel/time/timer.c:1320 expire_timers kernel/time/timer.c:1357 [inline] __run_timers+0x7ee/0xb70 kernel/time/timer.c:1660 run_timer_softirq+0x4c/0xb0 kernel/time/timer.c:1686 __do_softirq+0x2d7/0xb85 kernel/softirq.c:285 invoke_softirq kernel/softirq.c:365 [inline] irq_exit+0x1cc/0x200 kernel/softirq.c:405 exiting_irq arch/x86/include/asm/apic.h:540 [inline] smp_apic_timer_interrupt+0x16b/0x700 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xa9/0xb0 arch/x86/entry/entry_64.S:904 </IRQ> Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-09 01:34:00 +07:00
if (fn &&
(!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
(fn->fn_flags & RTN_TL_ROOT &&
!rcu_access_pointer(fn->leaf))))
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
fib6_repair_tree(info->nl_net, table, fn);
return err;
}
/*
* Routing tree lookup
*
*/
struct lookup_args {
int offset; /* key offset on fib6_info */
const struct in6_addr *addr; /* search key */
};
static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
struct lookup_args *args)
{
struct fib6_node *fn;
__be32 dir;
if (unlikely(args->offset == 0))
return NULL;
/*
* Descend on a tree
*/
fn = root;
for (;;) {
struct fib6_node *next;
dir = addr_bit_set(args->addr, fn->fn_bit);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
next = dir ? rcu_dereference(fn->right) :
rcu_dereference(fn->left);
if (next) {
fn = next;
continue;
}
break;
}
while (fn) {
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
struct fib6_node *subtree = FIB6_SUBTREE(fn);
if (subtree || fn->fn_flags & RTN_RTINFO) {
struct fib6_info *leaf = rcu_dereference(fn->leaf);
struct rt6key *key;
if (!leaf)
goto backtrack;
key = (struct rt6key *) ((u8 *)leaf + args->offset);
if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
#ifdef CONFIG_IPV6_SUBTREES
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (subtree) {
struct fib6_node *sfn;
sfn = fib6_node_lookup_1(subtree,
args + 1);
if (!sfn)
goto backtrack;
fn = sfn;
}
#endif
if (fn->fn_flags & RTN_RTINFO)
return fn;
}
}
backtrack:
if (fn->fn_flags & RTN_ROOT)
break;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
fn = rcu_dereference(fn->parent);
}
return NULL;
}
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
/* called with rcu_read_lock() held
*/
struct fib6_node *fib6_node_lookup(struct fib6_node *root,
const struct in6_addr *daddr,
const struct in6_addr *saddr)
{
struct fib6_node *fn;
struct lookup_args args[] = {
{
.offset = offsetof(struct fib6_info, fib6_dst),
.addr = daddr,
},
#ifdef CONFIG_IPV6_SUBTREES
{
.offset = offsetof(struct fib6_info, fib6_src),
.addr = saddr,
},
#endif
{
.offset = 0, /* sentinel */
}
};
fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
if (!fn || fn->fn_flags & RTN_TL_ROOT)
fn = root;
return fn;
}
/*
* Get node with specified destination prefix (and source prefix,
* if subtrees are used)
* exact_match == true means we try to find fn with exact match of
* the passed in prefix addr
* exact_match == false means we try to find fn with longest prefix
* match of the passed in prefix addr. This is useful for finding fn
* for cached route as it will be stored in the exception table under
* the node with longest prefix length.
*/
static struct fib6_node *fib6_locate_1(struct fib6_node *root,
const struct in6_addr *addr,
int plen, int offset,
bool exact_match)
{
struct fib6_node *fn, *prev = NULL;
for (fn = root; fn ; ) {
struct fib6_info *leaf = rcu_dereference(fn->leaf);
struct rt6key *key;
/* This node is being deleted */
if (!leaf) {
if (plen <= fn->fn_bit)
goto out;
else
goto next;
}
key = (struct rt6key *)((u8 *)leaf + offset);
/*
* Prefix match
*/
if (plen < fn->fn_bit ||
!ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
goto out;
if (plen == fn->fn_bit)
return fn;
prev = fn;
next:
/*
* We have more bits to go
*/
if (addr_bit_set(addr, fn->fn_bit))
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
fn = rcu_dereference(fn->right);
else
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
fn = rcu_dereference(fn->left);
}
out:
if (exact_match)
return NULL;
else
return prev;
}
struct fib6_node *fib6_locate(struct fib6_node *root,
const struct in6_addr *daddr, int dst_len,
const struct in6_addr *saddr, int src_len,
bool exact_match)
{
struct fib6_node *fn;
fn = fib6_locate_1(root, daddr, dst_len,
offsetof(struct fib6_info, fib6_dst),
exact_match);
#ifdef CONFIG_IPV6_SUBTREES
if (src_len) {
WARN_ON(saddr == NULL);
if (fn) {
struct fib6_node *subtree = FIB6_SUBTREE(fn);
if (subtree) {
fn = fib6_locate_1(subtree, saddr, src_len,
offsetof(struct fib6_info, fib6_src),
exact_match);
}
}
}
#endif
if (fn && fn->fn_flags & RTN_RTINFO)
return fn;
return NULL;
}
/*
* Deletion
*
*/
static struct fib6_info *fib6_find_prefix(struct net *net,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
struct fib6_table *table,
struct fib6_node *fn)
{
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
struct fib6_node *child_left, *child_right;
if (fn->fn_flags & RTN_ROOT)
return net->ipv6.fib6_null_entry;
while (fn) {
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
child_left = rcu_dereference_protected(fn->left,
lockdep_is_held(&table->tb6_lock));
child_right = rcu_dereference_protected(fn->right,
lockdep_is_held(&table->tb6_lock));
if (child_left)
return rcu_dereference_protected(child_left->leaf,
lockdep_is_held(&table->tb6_lock));
if (child_right)
return rcu_dereference_protected(child_right->leaf,
lockdep_is_held(&table->tb6_lock));
fn = FIB6_SUBTREE(fn);
}
return NULL;
}
/*
* Called to trim the tree of intermediate nodes when possible. "fn"
* is the node we want to try and remove.
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
* Need to own table->tb6_lock
*/
static struct fib6_node *fib6_repair_tree(struct net *net,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
struct fib6_table *table,
struct fib6_node *fn)
{
int children;
int nstate;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
struct fib6_node *child;
struct fib6_walker *w;
int iter = 0;
ipv6: remove null_entry before adding default route In the current code, when creating a new fib6 table, tb6_root.leaf gets initialized to net->ipv6.ip6_null_entry. If a default route is being added with rt->rt6i_metric = 0xffffffff, fib6_add() will add this route after net->ipv6.ip6_null_entry. As null_entry is shared, it could cause problem. In order to fix it, set fn->leaf to NULL before calling fib6_add_rt2node() when trying to add the first default route. And reset fn->leaf to null_entry when adding fails or when deleting the last default route. syzkaller reported the following issue which is fixed by this commit: WARNING: suspicious RCU usage 4.15.0-rc5+ #171 Not tainted ----------------------------- net/ipv6/ip6_fib.c:1702 suspicious rcu_dereference_protected() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 4 locks held by swapper/0/0: #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] lockdep_copy_map include/linux/lockdep.h:178 [inline] #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] call_timer_fn+0x1c6/0x820 kernel/time/timer.c:1310 #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] spin_lock_bh include/linux/spinlock.h:315 [inline] #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] fib6_run_gc+0x9d/0x3c0 net/ipv6/ip6_fib.c:2007 #2: (rcu_read_lock){....}, at: [<0000000091db762d>] __fib6_clean_all+0x0/0x3a0 net/ipv6/ip6_fib.c:1560 #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] spin_lock_bh include/linux/spinlock.h:315 [inline] #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] __fib6_clean_all+0x1d0/0x3a0 net/ipv6/ip6_fib.c:1948 stack backtrace: CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc5+ #171 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 lockdep_rcu_suspicious+0x123/0x170 kernel/locking/lockdep.c:4585 fib6_del+0xcaa/0x11b0 net/ipv6/ip6_fib.c:1701 fib6_clean_node+0x3aa/0x4f0 net/ipv6/ip6_fib.c:1892 fib6_walk_continue+0x46c/0x8a0 net/ipv6/ip6_fib.c:1815 fib6_walk+0x91/0xf0 net/ipv6/ip6_fib.c:1863 fib6_clean_tree+0x1e6/0x340 net/ipv6/ip6_fib.c:1933 __fib6_clean_all+0x1f4/0x3a0 net/ipv6/ip6_fib.c:1949 fib6_clean_all net/ipv6/ip6_fib.c:1960 [inline] fib6_run_gc+0x16b/0x3c0 net/ipv6/ip6_fib.c:2016 fib6_gc_timer_cb+0x20/0x30 net/ipv6/ip6_fib.c:2033 call_timer_fn+0x228/0x820 kernel/time/timer.c:1320 expire_timers kernel/time/timer.c:1357 [inline] __run_timers+0x7ee/0xb70 kernel/time/timer.c:1660 run_timer_softirq+0x4c/0xb0 kernel/time/timer.c:1686 __do_softirq+0x2d7/0xb85 kernel/softirq.c:285 invoke_softirq kernel/softirq.c:365 [inline] irq_exit+0x1cc/0x200 kernel/softirq.c:405 exiting_irq arch/x86/include/asm/apic.h:540 [inline] smp_apic_timer_interrupt+0x16b/0x700 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xa9/0xb0 arch/x86/entry/entry_64.S:904 </IRQ> Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-09 01:34:00 +07:00
/* Set fn->leaf to null_entry for root node. */
if (fn->fn_flags & RTN_TL_ROOT) {
rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
ipv6: remove null_entry before adding default route In the current code, when creating a new fib6 table, tb6_root.leaf gets initialized to net->ipv6.ip6_null_entry. If a default route is being added with rt->rt6i_metric = 0xffffffff, fib6_add() will add this route after net->ipv6.ip6_null_entry. As null_entry is shared, it could cause problem. In order to fix it, set fn->leaf to NULL before calling fib6_add_rt2node() when trying to add the first default route. And reset fn->leaf to null_entry when adding fails or when deleting the last default route. syzkaller reported the following issue which is fixed by this commit: WARNING: suspicious RCU usage 4.15.0-rc5+ #171 Not tainted ----------------------------- net/ipv6/ip6_fib.c:1702 suspicious rcu_dereference_protected() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 4 locks held by swapper/0/0: #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] lockdep_copy_map include/linux/lockdep.h:178 [inline] #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] call_timer_fn+0x1c6/0x820 kernel/time/timer.c:1310 #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] spin_lock_bh include/linux/spinlock.h:315 [inline] #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] fib6_run_gc+0x9d/0x3c0 net/ipv6/ip6_fib.c:2007 #2: (rcu_read_lock){....}, at: [<0000000091db762d>] __fib6_clean_all+0x0/0x3a0 net/ipv6/ip6_fib.c:1560 #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] spin_lock_bh include/linux/spinlock.h:315 [inline] #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] __fib6_clean_all+0x1d0/0x3a0 net/ipv6/ip6_fib.c:1948 stack backtrace: CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc5+ #171 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 lockdep_rcu_suspicious+0x123/0x170 kernel/locking/lockdep.c:4585 fib6_del+0xcaa/0x11b0 net/ipv6/ip6_fib.c:1701 fib6_clean_node+0x3aa/0x4f0 net/ipv6/ip6_fib.c:1892 fib6_walk_continue+0x46c/0x8a0 net/ipv6/ip6_fib.c:1815 fib6_walk+0x91/0xf0 net/ipv6/ip6_fib.c:1863 fib6_clean_tree+0x1e6/0x340 net/ipv6/ip6_fib.c:1933 __fib6_clean_all+0x1f4/0x3a0 net/ipv6/ip6_fib.c:1949 fib6_clean_all net/ipv6/ip6_fib.c:1960 [inline] fib6_run_gc+0x16b/0x3c0 net/ipv6/ip6_fib.c:2016 fib6_gc_timer_cb+0x20/0x30 net/ipv6/ip6_fib.c:2033 call_timer_fn+0x228/0x820 kernel/time/timer.c:1320 expire_timers kernel/time/timer.c:1357 [inline] __run_timers+0x7ee/0xb70 kernel/time/timer.c:1660 run_timer_softirq+0x4c/0xb0 kernel/time/timer.c:1686 __do_softirq+0x2d7/0xb85 kernel/softirq.c:285 invoke_softirq kernel/softirq.c:365 [inline] irq_exit+0x1cc/0x200 kernel/softirq.c:405 exiting_irq arch/x86/include/asm/apic.h:540 [inline] smp_apic_timer_interrupt+0x16b/0x700 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xa9/0xb0 arch/x86/entry/entry_64.S:904 </IRQ> Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-09 01:34:00 +07:00
return fn;
}
for (;;) {
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
lockdep_is_held(&table->tb6_lock));
struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
lockdep_is_held(&table->tb6_lock));
struct fib6_node *pn = rcu_dereference_protected(fn->parent,
lockdep_is_held(&table->tb6_lock));
struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
lockdep_is_held(&table->tb6_lock));
struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
lockdep_is_held(&table->tb6_lock));
struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
lockdep_is_held(&table->tb6_lock));
struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
lockdep_is_held(&table->tb6_lock));
struct fib6_info *new_fn_leaf;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
iter++;
WARN_ON(fn->fn_flags & RTN_RTINFO);
WARN_ON(fn->fn_flags & RTN_TL_ROOT);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
WARN_ON(fn_leaf);
children = 0;
child = NULL;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (fn_r)
child = fn_r, children |= 1;
if (fn_l)
child = fn_l, children |= 2;
if (children == 3 || FIB6_SUBTREE(fn)
#ifdef CONFIG_IPV6_SUBTREES
/* Subtree root (i.e. fn) may have one child */
|| (children && fn->fn_flags & RTN_ROOT)
#endif
) {
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
new_fn_leaf = fib6_find_prefix(net, table, fn);
#if RT6_DEBUG >= 2
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (!new_fn_leaf) {
WARN_ON(!new_fn_leaf);
new_fn_leaf = net->ipv6.fib6_null_entry;
}
#endif
fib6_info_hold(new_fn_leaf);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(fn->leaf, new_fn_leaf);
return pn;
}
#ifdef CONFIG_IPV6_SUBTREES
if (FIB6_SUBTREE(pn) == fn) {
WARN_ON(!(fn->fn_flags & RTN_ROOT));
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RCU_INIT_POINTER(pn->subtree, NULL);
nstate = FWS_L;
} else {
WARN_ON(fn->fn_flags & RTN_ROOT);
#endif
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (pn_r == fn)
rcu_assign_pointer(pn->right, child);
else if (pn_l == fn)
rcu_assign_pointer(pn->left, child);
#if RT6_DEBUG >= 2
else
WARN_ON(1);
#endif
if (child)
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(child->parent, pn);
nstate = FWS_R;
#ifdef CONFIG_IPV6_SUBTREES
}
#endif
read_lock(&net->ipv6.fib6_walker_lock);
FOR_WALKERS(net, w) {
if (!child) {
if (w->node == fn) {
RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
w->node = pn;
w->state = nstate;
}
} else {
if (w->node == fn) {
w->node = child;
if (children&2) {
RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
} else {
RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
}
}
}
}
read_unlock(&net->ipv6.fib6_walker_lock);
node_free(net, fn);
if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
return pn;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
RCU_INIT_POINTER(pn->leaf, NULL);
fib6_info_release(pn_leaf);
fn = pn;
}
}
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
struct fib6_info __rcu **rtp, struct nl_info *info)
{
struct fib6_walker *w;
struct fib6_info *rt = rcu_dereference_protected(*rtp,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
lockdep_is_held(&table->tb6_lock));
struct net *net = info->nl_net;
RT6_TRACE("fib6_del_route\n");
/* Unlink it */
*rtp = rt->fib6_next;
rt->fib6_node = NULL;
net->ipv6.rt6_stats->fib_rt_entries--;
net->ipv6.rt6_stats->fib_discarded_routes++;
/* Flush all cached dst in exception table */
rt6_flush_exceptions(rt);
[IPV6]: Fix routing round-robin locking. As per RFC2461, section 6.3.6, item #2, when no routers on the matching list are known to be reachable or probably reachable we do round robin on those available routes so that we make sure to probe as many of them as possible to detect when one becomes reachable faster. Each routing table has a rwlock protecting the tree and the linked list of routes at each leaf. The round robin code executes during lookup and thus with the rwlock taken as a reader. A small local spinlock tries to provide protection but this does not work at all for two reasons: 1) The round-robin list manipulation, as coded, goes like this (with read lock held): walk routes finding head and tail spin_lock(); rotate list using head and tail spin_unlock(); While one thread is rotating the list, another thread can end up with stale values of head and tail and then proceed to corrupt the list when it gets the lock. This ends up causing the OOPS in fib6_add() later onthat many people have been hitting. 2) All the other code paths that run with the rwlock held as a reader do not expect the list to change on them, they expect it to remain completely fixed while they hold the lock in that way. So, simply stated, it is impossible to implement this correctly using a manipulation of the list without violating the rwlock locking semantics. Reimplement using a per-fib6_node round-robin pointer. This way we don't need to manipulate the list at all, and since the round-robin pointer can only ever point to real existing entries we don't need to perform any locking on the changing of the round-robin pointer itself. We only need to reset the round-robin pointer to NULL when the entry it is pointing to is removed. The idea is from Thomas Graf and it is very similar to how this was implemented before the advanced router selection code when in. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-03-25 10:36:25 +07:00
/* Reset round-robin state, if necessary */
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (rcu_access_pointer(fn->rr_ptr) == rt)
[IPV6]: Fix routing round-robin locking. As per RFC2461, section 6.3.6, item #2, when no routers on the matching list are known to be reachable or probably reachable we do round robin on those available routes so that we make sure to probe as many of them as possible to detect when one becomes reachable faster. Each routing table has a rwlock protecting the tree and the linked list of routes at each leaf. The round robin code executes during lookup and thus with the rwlock taken as a reader. A small local spinlock tries to provide protection but this does not work at all for two reasons: 1) The round-robin list manipulation, as coded, goes like this (with read lock held): walk routes finding head and tail spin_lock(); rotate list using head and tail spin_unlock(); While one thread is rotating the list, another thread can end up with stale values of head and tail and then proceed to corrupt the list when it gets the lock. This ends up causing the OOPS in fib6_add() later onthat many people have been hitting. 2) All the other code paths that run with the rwlock held as a reader do not expect the list to change on them, they expect it to remain completely fixed while they hold the lock in that way. So, simply stated, it is impossible to implement this correctly using a manipulation of the list without violating the rwlock locking semantics. Reimplement using a per-fib6_node round-robin pointer. This way we don't need to manipulate the list at all, and since the round-robin pointer can only ever point to real existing entries we don't need to perform any locking on the changing of the round-robin pointer itself. We only need to reset the round-robin pointer to NULL when the entry it is pointing to is removed. The idea is from Thomas Graf and it is very similar to how this was implemented before the advanced router selection code when in. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-03-25 10:36:25 +07:00
fn->rr_ptr = NULL;
/* Remove this entry from other siblings */
if (rt->fib6_nsiblings) {
struct fib6_info *sibling, *next_sibling;
list_for_each_entry_safe(sibling, next_sibling,
&rt->fib6_siblings, fib6_siblings)
sibling->fib6_nsiblings--;
rt->fib6_nsiblings = 0;
list_del_init(&rt->fib6_siblings);
rt6_multipath_rebalance(next_sibling);
}
/* Adjust walkers */
read_lock(&net->ipv6.fib6_walker_lock);
FOR_WALKERS(net, w) {
if (w->state == FWS_C && w->leaf == rt) {
RT6_TRACE("walker %p adjusted by delroute\n", w);
w->leaf = rcu_dereference_protected(rt->fib6_next,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
lockdep_is_held(&table->tb6_lock));
if (!w->leaf)
w->state = FWS_U;
}
}
read_unlock(&net->ipv6.fib6_walker_lock);
ipv6: remove null_entry before adding default route In the current code, when creating a new fib6 table, tb6_root.leaf gets initialized to net->ipv6.ip6_null_entry. If a default route is being added with rt->rt6i_metric = 0xffffffff, fib6_add() will add this route after net->ipv6.ip6_null_entry. As null_entry is shared, it could cause problem. In order to fix it, set fn->leaf to NULL before calling fib6_add_rt2node() when trying to add the first default route. And reset fn->leaf to null_entry when adding fails or when deleting the last default route. syzkaller reported the following issue which is fixed by this commit: WARNING: suspicious RCU usage 4.15.0-rc5+ #171 Not tainted ----------------------------- net/ipv6/ip6_fib.c:1702 suspicious rcu_dereference_protected() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 4 locks held by swapper/0/0: #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] lockdep_copy_map include/linux/lockdep.h:178 [inline] #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] call_timer_fn+0x1c6/0x820 kernel/time/timer.c:1310 #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] spin_lock_bh include/linux/spinlock.h:315 [inline] #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] fib6_run_gc+0x9d/0x3c0 net/ipv6/ip6_fib.c:2007 #2: (rcu_read_lock){....}, at: [<0000000091db762d>] __fib6_clean_all+0x0/0x3a0 net/ipv6/ip6_fib.c:1560 #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] spin_lock_bh include/linux/spinlock.h:315 [inline] #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] __fib6_clean_all+0x1d0/0x3a0 net/ipv6/ip6_fib.c:1948 stack backtrace: CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc5+ #171 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 lockdep_rcu_suspicious+0x123/0x170 kernel/locking/lockdep.c:4585 fib6_del+0xcaa/0x11b0 net/ipv6/ip6_fib.c:1701 fib6_clean_node+0x3aa/0x4f0 net/ipv6/ip6_fib.c:1892 fib6_walk_continue+0x46c/0x8a0 net/ipv6/ip6_fib.c:1815 fib6_walk+0x91/0xf0 net/ipv6/ip6_fib.c:1863 fib6_clean_tree+0x1e6/0x340 net/ipv6/ip6_fib.c:1933 __fib6_clean_all+0x1f4/0x3a0 net/ipv6/ip6_fib.c:1949 fib6_clean_all net/ipv6/ip6_fib.c:1960 [inline] fib6_run_gc+0x16b/0x3c0 net/ipv6/ip6_fib.c:2016 fib6_gc_timer_cb+0x20/0x30 net/ipv6/ip6_fib.c:2033 call_timer_fn+0x228/0x820 kernel/time/timer.c:1320 expire_timers kernel/time/timer.c:1357 [inline] __run_timers+0x7ee/0xb70 kernel/time/timer.c:1660 run_timer_softirq+0x4c/0xb0 kernel/time/timer.c:1686 __do_softirq+0x2d7/0xb85 kernel/softirq.c:285 invoke_softirq kernel/softirq.c:365 [inline] irq_exit+0x1cc/0x200 kernel/softirq.c:405 exiting_irq arch/x86/include/asm/apic.h:540 [inline] smp_apic_timer_interrupt+0x16b/0x700 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xa9/0xb0 arch/x86/entry/entry_64.S:904 </IRQ> Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-09 01:34:00 +07:00
/* If it was last route, call fib6_repair_tree() to:
* 1. For root node, put back null_entry as how the table was created.
* 2. For other nodes, expunge its radix tree node.
*/
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (!rcu_access_pointer(fn->leaf)) {
ipv6: remove null_entry before adding default route In the current code, when creating a new fib6 table, tb6_root.leaf gets initialized to net->ipv6.ip6_null_entry. If a default route is being added with rt->rt6i_metric = 0xffffffff, fib6_add() will add this route after net->ipv6.ip6_null_entry. As null_entry is shared, it could cause problem. In order to fix it, set fn->leaf to NULL before calling fib6_add_rt2node() when trying to add the first default route. And reset fn->leaf to null_entry when adding fails or when deleting the last default route. syzkaller reported the following issue which is fixed by this commit: WARNING: suspicious RCU usage 4.15.0-rc5+ #171 Not tainted ----------------------------- net/ipv6/ip6_fib.c:1702 suspicious rcu_dereference_protected() usage! other info that might help us debug this: rcu_scheduler_active = 2, debug_locks = 1 4 locks held by swapper/0/0: #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] lockdep_copy_map include/linux/lockdep.h:178 [inline] #0: ((&net->ipv6.ip6_fib_timer)){+.-.}, at: [<00000000d43f631b>] call_timer_fn+0x1c6/0x820 kernel/time/timer.c:1310 #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] spin_lock_bh include/linux/spinlock.h:315 [inline] #1: (&(&net->ipv6.fib6_gc_lock)->rlock){+.-.}, at: [<000000002ff9d65c>] fib6_run_gc+0x9d/0x3c0 net/ipv6/ip6_fib.c:2007 #2: (rcu_read_lock){....}, at: [<0000000091db762d>] __fib6_clean_all+0x0/0x3a0 net/ipv6/ip6_fib.c:1560 #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] spin_lock_bh include/linux/spinlock.h:315 [inline] #3: (&(&tb->tb6_lock)->rlock){+.-.}, at: [<000000009e503581>] __fib6_clean_all+0x1d0/0x3a0 net/ipv6/ip6_fib.c:1948 stack backtrace: CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.15.0-rc5+ #171 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <IRQ> __dump_stack lib/dump_stack.c:17 [inline] dump_stack+0x194/0x257 lib/dump_stack.c:53 lockdep_rcu_suspicious+0x123/0x170 kernel/locking/lockdep.c:4585 fib6_del+0xcaa/0x11b0 net/ipv6/ip6_fib.c:1701 fib6_clean_node+0x3aa/0x4f0 net/ipv6/ip6_fib.c:1892 fib6_walk_continue+0x46c/0x8a0 net/ipv6/ip6_fib.c:1815 fib6_walk+0x91/0xf0 net/ipv6/ip6_fib.c:1863 fib6_clean_tree+0x1e6/0x340 net/ipv6/ip6_fib.c:1933 __fib6_clean_all+0x1f4/0x3a0 net/ipv6/ip6_fib.c:1949 fib6_clean_all net/ipv6/ip6_fib.c:1960 [inline] fib6_run_gc+0x16b/0x3c0 net/ipv6/ip6_fib.c:2016 fib6_gc_timer_cb+0x20/0x30 net/ipv6/ip6_fib.c:2033 call_timer_fn+0x228/0x820 kernel/time/timer.c:1320 expire_timers kernel/time/timer.c:1357 [inline] __run_timers+0x7ee/0xb70 kernel/time/timer.c:1660 run_timer_softirq+0x4c/0xb0 kernel/time/timer.c:1686 __do_softirq+0x2d7/0xb85 kernel/softirq.c:285 invoke_softirq kernel/softirq.c:365 [inline] irq_exit+0x1cc/0x200 kernel/softirq.c:405 exiting_irq arch/x86/include/asm/apic.h:540 [inline] smp_apic_timer_interrupt+0x16b/0x700 arch/x86/kernel/apic/apic.c:1052 apic_timer_interrupt+0xa9/0xb0 arch/x86/entry/entry_64.S:904 </IRQ> Reported-by: syzbot <syzkaller@googlegroups.com> Fixes: 66f5d6ce53e6 ("ipv6: replace rwlock with rcu and spinlock in fib6_table") Signed-off-by: Wei Wang <weiwan@google.com> Acked-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-01-09 01:34:00 +07:00
if (!(fn->fn_flags & RTN_TL_ROOT)) {
fn->fn_flags &= ~RTN_RTINFO;
net->ipv6.rt6_stats->fib_route_nodes--;
}
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
fn = fib6_repair_tree(net, table, fn);
}
fib6_purge_rt(rt, fn, net);
call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
if (!info->skip_notify)
inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
fib6_info_release(rt);
}
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
/* Need to own table->tb6_lock */
int fib6_del(struct fib6_info *rt, struct nl_info *info)
{
struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
lockdep_is_held(&rt->fib6_table->tb6_lock));
struct fib6_table *table = rt->fib6_table;
struct net *net = info->nl_net;
struct fib6_info __rcu **rtp;
struct fib6_info __rcu **rtp_next;
if (!fn || rt == net->ipv6.fib6_null_entry)
return -ENOENT;
WARN_ON(!(fn->fn_flags & RTN_RTINFO));
/*
* Walk the leaf entries looking for ourself
*/
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
struct fib6_info *cur = rcu_dereference_protected(*rtp,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
lockdep_is_held(&table->tb6_lock));
if (rt == cur) {
fib6_del_route(table, fn, rtp, info);
return 0;
}
rtp_next = &cur->fib6_next;
}
return -ENOENT;
}
/*
* Tree traversal function.
*
* Certainly, it is not interrupt safe.
* However, it is internally reenterable wrt itself and fib6_add/fib6_del.
* It means, that we can modify tree during walking
* and use this function for garbage collection, clone pruning,
* cleaning tree when a device goes down etc. etc.
*
* It guarantees that every node will be traversed,
* and that it will be traversed only once.
*
* Callback function w->func may return:
* 0 -> continue walking.
* positive value -> walking is suspended (used by tree dumps,
* and probably by gc, if it will be split to several slices)
* negative value -> terminate walking.
*
* The function itself returns:
* 0 -> walk is complete.
* >0 -> walk is incomplete (i.e. suspended)
* <0 -> walk is terminated by an error.
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
*
* This function is called with tb6_lock held.
*/
static int fib6_walk_continue(struct fib6_walker *w)
{
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
struct fib6_node *fn, *pn, *left, *right;
/* w->root should always be table->tb6_root */
WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
for (;;) {
fn = w->node;
if (!fn)
return 0;
switch (w->state) {
#ifdef CONFIG_IPV6_SUBTREES
case FWS_S:
if (FIB6_SUBTREE(fn)) {
w->node = FIB6_SUBTREE(fn);
continue;
}
w->state = FWS_L;
#endif
/* fall through */
case FWS_L:
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
left = rcu_dereference_protected(fn->left, 1);
if (left) {
w->node = left;
w->state = FWS_INIT;
continue;
}
w->state = FWS_R;
/* fall through */
case FWS_R:
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
right = rcu_dereference_protected(fn->right, 1);
if (right) {
w->node = right;
w->state = FWS_INIT;
continue;
}
w->state = FWS_C;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
w->leaf = rcu_dereference_protected(fn->leaf, 1);
/* fall through */
case FWS_C:
if (w->leaf && fn->fn_flags & RTN_RTINFO) {
ipv6: fib: fix crash when changing large fib while dumping it When the fib size exceeds what can be dumped in a single skb, the dump is suspended and resumed once the last skb has been received by userspace. When the fib is changed while the dump is suspended, the walker might contain stale pointers, causing a crash when the dump is resumed. BUG: unable to handle kernel NULL pointer dereference at 0000000000000018 IP: [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] PGD 5347a067 PUD 65c7067 PMD 0 Oops: 0000 [#1] PREEMPT SMP ... RIP: 0010:[<ffffffffa01bce04>] [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] ... Call Trace: [<ffffffff8104aca3>] ? mutex_spin_on_owner+0x59/0x71 [<ffffffffa01bd105>] inet6_dump_fib+0x11b/0x1b9 [ipv6] [<ffffffff81371af4>] netlink_dump+0x5b/0x19e [<ffffffff8134f288>] ? consume_skb+0x28/0x2a [<ffffffff81373b69>] netlink_recvmsg+0x1ab/0x2c6 [<ffffffff81372781>] ? netlink_unicast+0xfa/0x151 [<ffffffff813483e0>] __sock_recvmsg+0x6d/0x79 [<ffffffff81348a53>] sock_recvmsg+0xca/0xe3 [<ffffffff81066d4b>] ? autoremove_wake_function+0x0/0x38 [<ffffffff811ed1f8>] ? radix_tree_lookup_slot+0xe/0x10 [<ffffffff810b3ed7>] ? find_get_page+0x90/0xa5 [<ffffffff810b5dc5>] ? filemap_fault+0x201/0x34f [<ffffffff810ef152>] ? fget_light+0x2f/0xac [<ffffffff813519e7>] ? verify_iovec+0x4f/0x94 [<ffffffff81349a65>] sys_recvmsg+0x14d/0x223 Store the serial number when beginning to walk the fib and reload pointers when continuing to walk after a change occured. Similar to other dumping functions, this might cause unrelated entries to be missed when entries are deleted. Tested-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-02-08 12:19:03 +07:00
int err;
if (w->skip) {
w->skip--;
goto skip;
ipv6: fib: fix crash when changing large fib while dumping it When the fib size exceeds what can be dumped in a single skb, the dump is suspended and resumed once the last skb has been received by userspace. When the fib is changed while the dump is suspended, the walker might contain stale pointers, causing a crash when the dump is resumed. BUG: unable to handle kernel NULL pointer dereference at 0000000000000018 IP: [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] PGD 5347a067 PUD 65c7067 PMD 0 Oops: 0000 [#1] PREEMPT SMP ... RIP: 0010:[<ffffffffa01bce04>] [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] ... Call Trace: [<ffffffff8104aca3>] ? mutex_spin_on_owner+0x59/0x71 [<ffffffffa01bd105>] inet6_dump_fib+0x11b/0x1b9 [ipv6] [<ffffffff81371af4>] netlink_dump+0x5b/0x19e [<ffffffff8134f288>] ? consume_skb+0x28/0x2a [<ffffffff81373b69>] netlink_recvmsg+0x1ab/0x2c6 [<ffffffff81372781>] ? netlink_unicast+0xfa/0x151 [<ffffffff813483e0>] __sock_recvmsg+0x6d/0x79 [<ffffffff81348a53>] sock_recvmsg+0xca/0xe3 [<ffffffff81066d4b>] ? autoremove_wake_function+0x0/0x38 [<ffffffff811ed1f8>] ? radix_tree_lookup_slot+0xe/0x10 [<ffffffff810b3ed7>] ? find_get_page+0x90/0xa5 [<ffffffff810b5dc5>] ? filemap_fault+0x201/0x34f [<ffffffff810ef152>] ? fget_light+0x2f/0xac [<ffffffff813519e7>] ? verify_iovec+0x4f/0x94 [<ffffffff81349a65>] sys_recvmsg+0x14d/0x223 Store the serial number when beginning to walk the fib and reload pointers when continuing to walk after a change occured. Similar to other dumping functions, this might cause unrelated entries to be missed when entries are deleted. Tested-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-02-08 12:19:03 +07:00
}
err = w->func(w);
if (err)
return err;
ipv6: fib: fix crash when changing large fib while dumping it When the fib size exceeds what can be dumped in a single skb, the dump is suspended and resumed once the last skb has been received by userspace. When the fib is changed while the dump is suspended, the walker might contain stale pointers, causing a crash when the dump is resumed. BUG: unable to handle kernel NULL pointer dereference at 0000000000000018 IP: [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] PGD 5347a067 PUD 65c7067 PMD 0 Oops: 0000 [#1] PREEMPT SMP ... RIP: 0010:[<ffffffffa01bce04>] [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] ... Call Trace: [<ffffffff8104aca3>] ? mutex_spin_on_owner+0x59/0x71 [<ffffffffa01bd105>] inet6_dump_fib+0x11b/0x1b9 [ipv6] [<ffffffff81371af4>] netlink_dump+0x5b/0x19e [<ffffffff8134f288>] ? consume_skb+0x28/0x2a [<ffffffff81373b69>] netlink_recvmsg+0x1ab/0x2c6 [<ffffffff81372781>] ? netlink_unicast+0xfa/0x151 [<ffffffff813483e0>] __sock_recvmsg+0x6d/0x79 [<ffffffff81348a53>] sock_recvmsg+0xca/0xe3 [<ffffffff81066d4b>] ? autoremove_wake_function+0x0/0x38 [<ffffffff811ed1f8>] ? radix_tree_lookup_slot+0xe/0x10 [<ffffffff810b3ed7>] ? find_get_page+0x90/0xa5 [<ffffffff810b5dc5>] ? filemap_fault+0x201/0x34f [<ffffffff810ef152>] ? fget_light+0x2f/0xac [<ffffffff813519e7>] ? verify_iovec+0x4f/0x94 [<ffffffff81349a65>] sys_recvmsg+0x14d/0x223 Store the serial number when beginning to walk the fib and reload pointers when continuing to walk after a change occured. Similar to other dumping functions, this might cause unrelated entries to be missed when entries are deleted. Tested-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-02-08 12:19:03 +07:00
w->count++;
continue;
}
skip:
w->state = FWS_U;
/* fall through */
case FWS_U:
if (fn == w->root)
return 0;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
pn = rcu_dereference_protected(fn->parent, 1);
left = rcu_dereference_protected(pn->left, 1);
right = rcu_dereference_protected(pn->right, 1);
w->node = pn;
#ifdef CONFIG_IPV6_SUBTREES
if (FIB6_SUBTREE(pn) == fn) {
WARN_ON(!(fn->fn_flags & RTN_ROOT));
w->state = FWS_L;
continue;
}
#endif
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (left == fn) {
w->state = FWS_R;
continue;
}
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
if (right == fn) {
w->state = FWS_C;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
w->leaf = rcu_dereference_protected(w->node->leaf, 1);
continue;
}
#if RT6_DEBUG >= 2
WARN_ON(1);
#endif
}
}
}
static int fib6_walk(struct net *net, struct fib6_walker *w)
{
int res;
w->state = FWS_INIT;
w->node = w->root;
fib6_walker_link(net, w);
res = fib6_walk_continue(w);
if (res <= 0)
fib6_walker_unlink(net, w);
return res;
}
static int fib6_clean_node(struct fib6_walker *w)
{
int res;
struct fib6_info *rt;
struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
struct nl_info info = {
.nl_net = c->net,
.skip_notify = c->skip_notify,
};
if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
w->node->fn_sernum != c->sernum)
w->node->fn_sernum = c->sernum;
if (!c->func) {
WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
w->leaf = NULL;
return 0;
}
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
for_each_fib6_walker_rt(w) {
res = c->func(rt, c->arg);
if (res == -1) {
w->leaf = rt;
res = fib6_del(rt, &info);
if (res) {
#if RT6_DEBUG >= 2
pr_debug("%s: del failed: rt=%p@%p err=%d\n",
__func__, rt,
rcu_access_pointer(rt->fib6_node),
res);
#endif
continue;
}
return 0;
} else if (res == -2) {
if (WARN_ON(!rt->fib6_nsiblings))
continue;
rt = list_last_entry(&rt->fib6_siblings,
struct fib6_info, fib6_siblings);
continue;
}
WARN_ON(res != 0);
}
w->leaf = rt;
return 0;
}
/*
* Convenient frontend to tree walker.
*
* func is called on each route.
* It may return -2 -> skip multipath route.
* -1 -> delete this route.
* 0 -> continue walking
*/
static void fib6_clean_tree(struct net *net, struct fib6_node *root,
int (*func)(struct fib6_info *, void *arg),
int sernum, void *arg, bool skip_notify)
{
struct fib6_cleaner c;
c.w.root = root;
c.w.func = fib6_clean_node;
ipv6: fib: fix crash when changing large fib while dumping it When the fib size exceeds what can be dumped in a single skb, the dump is suspended and resumed once the last skb has been received by userspace. When the fib is changed while the dump is suspended, the walker might contain stale pointers, causing a crash when the dump is resumed. BUG: unable to handle kernel NULL pointer dereference at 0000000000000018 IP: [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] PGD 5347a067 PUD 65c7067 PMD 0 Oops: 0000 [#1] PREEMPT SMP ... RIP: 0010:[<ffffffffa01bce04>] [<ffffffffa01bce04>] fib6_walk_continue+0xbb/0x124 [ipv6] ... Call Trace: [<ffffffff8104aca3>] ? mutex_spin_on_owner+0x59/0x71 [<ffffffffa01bd105>] inet6_dump_fib+0x11b/0x1b9 [ipv6] [<ffffffff81371af4>] netlink_dump+0x5b/0x19e [<ffffffff8134f288>] ? consume_skb+0x28/0x2a [<ffffffff81373b69>] netlink_recvmsg+0x1ab/0x2c6 [<ffffffff81372781>] ? netlink_unicast+0xfa/0x151 [<ffffffff813483e0>] __sock_recvmsg+0x6d/0x79 [<ffffffff81348a53>] sock_recvmsg+0xca/0xe3 [<ffffffff81066d4b>] ? autoremove_wake_function+0x0/0x38 [<ffffffff811ed1f8>] ? radix_tree_lookup_slot+0xe/0x10 [<ffffffff810b3ed7>] ? find_get_page+0x90/0xa5 [<ffffffff810b5dc5>] ? filemap_fault+0x201/0x34f [<ffffffff810ef152>] ? fget_light+0x2f/0xac [<ffffffff813519e7>] ? verify_iovec+0x4f/0x94 [<ffffffff81349a65>] sys_recvmsg+0x14d/0x223 Store the serial number when beginning to walk the fib and reload pointers when continuing to walk after a change occured. Similar to other dumping functions, this might cause unrelated entries to be missed when entries are deleted. Tested-by: Ben Greear <greearb@candelatech.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-02-08 12:19:03 +07:00
c.w.count = 0;
c.w.skip = 0;
c.func = func;
c.sernum = sernum;
c.arg = arg;
c.net = net;
c.skip_notify = skip_notify;
fib6_walk(net, &c.w);
}
static void __fib6_clean_all(struct net *net,
int (*func)(struct fib6_info *, void *),
int sernum, void *arg, bool skip_notify)
{
struct fib6_table *table;
struct hlist_head *head;
unsigned int h;
rcu_read_lock();
for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
head = &net->ipv6.fib_table_hash[h];
hlist: drop the node parameter from iterators I'm not sure why, but the hlist for each entry iterators were conceived list_for_each_entry(pos, head, member) The hlist ones were greedy and wanted an extra parameter: hlist_for_each_entry(tpos, pos, head, member) Why did they need an extra pos parameter? I'm not quite sure. Not only they don't really need it, it also prevents the iterator from looking exactly like the list iterator, which is unfortunate. Besides the semantic patch, there was some manual work required: - Fix up the actual hlist iterators in linux/list.h - Fix up the declaration of other iterators based on the hlist ones. - A very small amount of places were using the 'node' parameter, this was modified to use 'obj->member' instead. - Coccinelle didn't handle the hlist_for_each_entry_safe iterator properly, so those had to be fixed up manually. The semantic patch which is mostly the work of Peter Senna Tschudin is here: @@ iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host; type T; expression a,c,d,e; identifier b; statement S; @@ -T b; <+... when != b ( hlist_for_each_entry(a, - b, c, d) S | hlist_for_each_entry_continue(a, - b, c) S | hlist_for_each_entry_from(a, - b, c) S | hlist_for_each_entry_rcu(a, - b, c, d) S | hlist_for_each_entry_rcu_bh(a, - b, c, d) S | hlist_for_each_entry_continue_rcu_bh(a, - b, c) S | for_each_busy_worker(a, c, - b, d) S | ax25_uid_for_each(a, - b, c) S | ax25_for_each(a, - b, c) S | inet_bind_bucket_for_each(a, - b, c) S | sctp_for_each_hentry(a, - b, c) S | sk_for_each(a, - b, c) S | sk_for_each_rcu(a, - b, c) S | sk_for_each_from -(a, b) +(a) S + sk_for_each_from(a) S | sk_for_each_safe(a, - b, c, d) S | sk_for_each_bound(a, - b, c) S | hlist_for_each_entry_safe(a, - b, c, d, e) S | hlist_for_each_entry_continue_rcu(a, - b, c) S | nr_neigh_for_each(a, - b, c) S | nr_neigh_for_each_safe(a, - b, c, d) S | nr_node_for_each(a, - b, c) S | nr_node_for_each_safe(a, - b, c, d) S | - for_each_gfn_sp(a, c, d, b) S + for_each_gfn_sp(a, c, d) S | - for_each_gfn_indirect_valid_sp(a, c, d, b) S + for_each_gfn_indirect_valid_sp(a, c, d) S | for_each_host(a, - b, c) S | for_each_host_safe(a, - b, c, d) S | for_each_mesh_entry(a, - b, c, d) S ) ...+> [akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c] [akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c] [akpm@linux-foundation.org: checkpatch fixes] [akpm@linux-foundation.org: fix warnings] [akpm@linux-foudnation.org: redo intrusive kvm changes] Tested-by: Peter Senna Tschudin <peter.senna@gmail.com> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Sasha Levin <sasha.levin@oracle.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Gleb Natapov <gleb@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-28 08:06:00 +07:00
hlist_for_each_entry_rcu(table, head, tb6_hlist) {
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_lock_bh(&table->tb6_lock);
fib6_clean_tree(net, &table->tb6_root,
func, sernum, arg, skip_notify);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_unlock_bh(&table->tb6_lock);
}
}
rcu_read_unlock();
}
void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
void *arg)
{
__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
}
void fib6_clean_all_skip_notify(struct net *net,
int (*func)(struct fib6_info *, void *),
void *arg)
{
__fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
}
static void fib6_flush_trees(struct net *net)
{
int new_sernum = fib6_new_sernum(net);
__fib6_clean_all(net, NULL, new_sernum, NULL, false);
}
/*
* Garbage collection
*/
static int fib6_age(struct fib6_info *rt, void *arg)
{
struct fib6_gc_args *gc_args = arg;
unsigned long now = jiffies;
/*
* check addrconf expiration here.
* Routes are expired even if they are in use.
*/
if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
if (time_after(now, rt->expires)) {
RT6_TRACE("expiring %p\n", rt);
return -1;
}
gc_args->more++;
}
/* Also age clones in the exception table.
* Note, that clones are aged out
* only if they are not in use now.
*/
rt6_age_exceptions(rt, gc_args, now);
return 0;
}
void fib6_run_gc(unsigned long expires, struct net *net, bool force)
{
struct fib6_gc_args gc_args;
unsigned long now;
if (force) {
spin_lock_bh(&net->ipv6.fib6_gc_lock);
} else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
return;
}
gc_args.timeout = expires ? (int)expires :
net->ipv6.sysctl.ip6_rt_gc_interval;
gc_args.more = 0;
fib6_clean_all(net, fib6_age, &gc_args);
now = jiffies;
net->ipv6.ip6_rt_last_gc = now;
if (gc_args.more)
mod_timer(&net->ipv6.ip6_fib_timer,
round_jiffies(now
+ net->ipv6.sysctl.ip6_rt_gc_interval));
else
del_timer(&net->ipv6.ip6_fib_timer);
spin_unlock_bh(&net->ipv6.fib6_gc_lock);
}
treewide: setup_timer() -> timer_setup() (2 field) This converts all remaining setup_timer() calls that use a nested field to reach a struct timer_list. Coccinelle does not have an easy way to match multiple fields, so a new script is needed to change the matches of "&_E->_timer" into "&_E->_field1._timer" in all the rules. spatch --very-quiet --all-includes --include-headers \ -I ./arch/x86/include -I ./arch/x86/include/generated \ -I ./include -I ./arch/x86/include/uapi \ -I ./arch/x86/include/generated/uapi -I ./include/uapi \ -I ./include/generated/uapi --include ./include/linux/kconfig.h \ --dir . \ --cocci-file ~/src/data/timer_setup-2fields.cocci @fix_address_of depends@ expression e; @@ setup_timer( -&(e) +&e , ...) // Update any raw setup_timer() usages that have a NULL callback, but // would otherwise match change_timer_function_usage, since the latter // will update all function assignments done in the face of a NULL // function initialization in setup_timer(). @change_timer_function_usage_NULL@ expression _E; identifier _field1; identifier _timer; type _cast_data; @@ ( -setup_timer(&_E->_field1._timer, NULL, _E); +timer_setup(&_E->_field1._timer, NULL, 0); | -setup_timer(&_E->_field1._timer, NULL, (_cast_data)_E); +timer_setup(&_E->_field1._timer, NULL, 0); | -setup_timer(&_E._field1._timer, NULL, &_E); +timer_setup(&_E._field1._timer, NULL, 0); | -setup_timer(&_E._field1._timer, NULL, (_cast_data)&_E); +timer_setup(&_E._field1._timer, NULL, 0); ) @change_timer_function_usage@ expression _E; identifier _field1; identifier _timer; struct timer_list _stl; identifier _callback; type _cast_func, _cast_data; @@ ( -setup_timer(&_E->_field1._timer, _callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, &_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, &_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)&_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)&_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, &_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, &_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)&_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)&_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | _E->_field1._timer@_stl.function = _callback; | _E->_field1._timer@_stl.function = &_callback; | _E->_field1._timer@_stl.function = (_cast_func)_callback; | _E->_field1._timer@_stl.function = (_cast_func)&_callback; | _E._field1._timer@_stl.function = _callback; | _E._field1._timer@_stl.function = &_callback; | _E._field1._timer@_stl.function = (_cast_func)_callback; | _E._field1._timer@_stl.function = (_cast_func)&_callback; ) // callback(unsigned long arg) @change_callback_handle_cast depends on change_timer_function_usage@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _origtype; identifier _origarg; type _handletype; identifier _handle; @@ void _callback( -_origtype _origarg +struct timer_list *t ) { ( ... when != _origarg _handletype *_handle = -(_handletype *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle = -(void *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle; ... when != _handle _handle = -(_handletype *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle; ... when != _handle _handle = -(void *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg ) } // callback(unsigned long arg) without existing variable @change_callback_handle_cast_no_arg depends on change_timer_function_usage && !change_callback_handle_cast@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _origtype; identifier _origarg; type _handletype; @@ void _callback( -_origtype _origarg +struct timer_list *t ) { + _handletype *_origarg = from_timer(_origarg, t, _field1._timer); + ... when != _origarg - (_handletype *)_origarg + _origarg ... when != _origarg } // Avoid already converted callbacks. @match_callback_converted depends on change_timer_function_usage && !change_callback_handle_cast && !change_callback_handle_cast_no_arg@ identifier change_timer_function_usage._callback; identifier t; @@ void _callback(struct timer_list *t) { ... } // callback(struct something *handle) @change_callback_handle_arg depends on change_timer_function_usage && !match_callback_converted && !change_callback_handle_cast && !change_callback_handle_cast_no_arg@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _handletype; identifier _handle; @@ void _callback( -_handletype *_handle +struct timer_list *t ) { + _handletype *_handle = from_timer(_handle, t, _field1._timer); ... } // If change_callback_handle_arg ran on an empty function, remove // the added handler. @unchange_callback_handle_arg depends on change_timer_function_usage && change_callback_handle_arg@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _handletype; identifier _handle; identifier t; @@ void _callback(struct timer_list *t) { - _handletype *_handle = from_timer(_handle, t, _field1._timer); } // We only want to refactor the setup_timer() data argument if we've found // the matching callback. This undoes changes in change_timer_function_usage. @unchange_timer_function_usage depends on change_timer_function_usage && !change_callback_handle_cast && !change_callback_handle_cast_no_arg && !change_callback_handle_arg@ expression change_timer_function_usage._E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type change_timer_function_usage._cast_data; @@ ( -timer_setup(&_E->_field1._timer, _callback, 0); +setup_timer(&_E->_field1._timer, _callback, (_cast_data)_E); | -timer_setup(&_E._field1._timer, _callback, 0); +setup_timer(&_E._field1._timer, _callback, (_cast_data)&_E); ) // If we fixed a callback from a .function assignment, fix the // assignment cast now. @change_timer_function_assignment depends on change_timer_function_usage && (change_callback_handle_cast || change_callback_handle_cast_no_arg || change_callback_handle_arg)@ expression change_timer_function_usage._E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type _cast_func; typedef TIMER_FUNC_TYPE; @@ ( _E->_field1._timer.function = -_callback +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -&_callback +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -(_cast_func)_callback; +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -(_cast_func)&_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -&_callback; +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -(_cast_func)_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -(_cast_func)&_callback +(TIMER_FUNC_TYPE)_callback ; ) // Sometimes timer functions are called directly. Replace matched args. @change_timer_function_calls depends on change_timer_function_usage && (change_callback_handle_cast || change_callback_handle_cast_no_arg || change_callback_handle_arg)@ expression _E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type _cast_data; @@ _callback( ( -(_cast_data)_E +&_E->_field1._timer | -(_cast_data)&_E +&_E._field1._timer | -_E +&_E->_field1._timer ) ) // If a timer has been configured without a data argument, it can be // converted without regard to the callback argument, since it is unused. @match_timer_function_unused_data@ expression _E; identifier _field1; identifier _timer; identifier _callback; @@ ( -setup_timer(&_E->_field1._timer, _callback, 0); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, 0L); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, 0UL); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0L); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0UL); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0L); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0UL); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0); +timer_setup(_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0L); +timer_setup(_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0UL); +timer_setup(_field1._timer, _callback, 0); ) @change_callback_unused_data depends on match_timer_function_unused_data@ identifier match_timer_function_unused_data._callback; type _origtype; identifier _origarg; @@ void _callback( -_origtype _origarg +struct timer_list *unused ) { ... when != _origarg } Signed-off-by: Kees Cook <keescook@chromium.org>
2017-10-18 10:21:24 +07:00
static void fib6_gc_timer_cb(struct timer_list *t)
{
treewide: setup_timer() -> timer_setup() (2 field) This converts all remaining setup_timer() calls that use a nested field to reach a struct timer_list. Coccinelle does not have an easy way to match multiple fields, so a new script is needed to change the matches of "&_E->_timer" into "&_E->_field1._timer" in all the rules. spatch --very-quiet --all-includes --include-headers \ -I ./arch/x86/include -I ./arch/x86/include/generated \ -I ./include -I ./arch/x86/include/uapi \ -I ./arch/x86/include/generated/uapi -I ./include/uapi \ -I ./include/generated/uapi --include ./include/linux/kconfig.h \ --dir . \ --cocci-file ~/src/data/timer_setup-2fields.cocci @fix_address_of depends@ expression e; @@ setup_timer( -&(e) +&e , ...) // Update any raw setup_timer() usages that have a NULL callback, but // would otherwise match change_timer_function_usage, since the latter // will update all function assignments done in the face of a NULL // function initialization in setup_timer(). @change_timer_function_usage_NULL@ expression _E; identifier _field1; identifier _timer; type _cast_data; @@ ( -setup_timer(&_E->_field1._timer, NULL, _E); +timer_setup(&_E->_field1._timer, NULL, 0); | -setup_timer(&_E->_field1._timer, NULL, (_cast_data)_E); +timer_setup(&_E->_field1._timer, NULL, 0); | -setup_timer(&_E._field1._timer, NULL, &_E); +timer_setup(&_E._field1._timer, NULL, 0); | -setup_timer(&_E._field1._timer, NULL, (_cast_data)&_E); +timer_setup(&_E._field1._timer, NULL, 0); ) @change_timer_function_usage@ expression _E; identifier _field1; identifier _timer; struct timer_list _stl; identifier _callback; type _cast_func, _cast_data; @@ ( -setup_timer(&_E->_field1._timer, _callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, &_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, &_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)&_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)&_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, &_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, &_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)&_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)&_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | _E->_field1._timer@_stl.function = _callback; | _E->_field1._timer@_stl.function = &_callback; | _E->_field1._timer@_stl.function = (_cast_func)_callback; | _E->_field1._timer@_stl.function = (_cast_func)&_callback; | _E._field1._timer@_stl.function = _callback; | _E._field1._timer@_stl.function = &_callback; | _E._field1._timer@_stl.function = (_cast_func)_callback; | _E._field1._timer@_stl.function = (_cast_func)&_callback; ) // callback(unsigned long arg) @change_callback_handle_cast depends on change_timer_function_usage@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _origtype; identifier _origarg; type _handletype; identifier _handle; @@ void _callback( -_origtype _origarg +struct timer_list *t ) { ( ... when != _origarg _handletype *_handle = -(_handletype *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle = -(void *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle; ... when != _handle _handle = -(_handletype *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle; ... when != _handle _handle = -(void *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg ) } // callback(unsigned long arg) without existing variable @change_callback_handle_cast_no_arg depends on change_timer_function_usage && !change_callback_handle_cast@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _origtype; identifier _origarg; type _handletype; @@ void _callback( -_origtype _origarg +struct timer_list *t ) { + _handletype *_origarg = from_timer(_origarg, t, _field1._timer); + ... when != _origarg - (_handletype *)_origarg + _origarg ... when != _origarg } // Avoid already converted callbacks. @match_callback_converted depends on change_timer_function_usage && !change_callback_handle_cast && !change_callback_handle_cast_no_arg@ identifier change_timer_function_usage._callback; identifier t; @@ void _callback(struct timer_list *t) { ... } // callback(struct something *handle) @change_callback_handle_arg depends on change_timer_function_usage && !match_callback_converted && !change_callback_handle_cast && !change_callback_handle_cast_no_arg@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _handletype; identifier _handle; @@ void _callback( -_handletype *_handle +struct timer_list *t ) { + _handletype *_handle = from_timer(_handle, t, _field1._timer); ... } // If change_callback_handle_arg ran on an empty function, remove // the added handler. @unchange_callback_handle_arg depends on change_timer_function_usage && change_callback_handle_arg@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _handletype; identifier _handle; identifier t; @@ void _callback(struct timer_list *t) { - _handletype *_handle = from_timer(_handle, t, _field1._timer); } // We only want to refactor the setup_timer() data argument if we've found // the matching callback. This undoes changes in change_timer_function_usage. @unchange_timer_function_usage depends on change_timer_function_usage && !change_callback_handle_cast && !change_callback_handle_cast_no_arg && !change_callback_handle_arg@ expression change_timer_function_usage._E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type change_timer_function_usage._cast_data; @@ ( -timer_setup(&_E->_field1._timer, _callback, 0); +setup_timer(&_E->_field1._timer, _callback, (_cast_data)_E); | -timer_setup(&_E._field1._timer, _callback, 0); +setup_timer(&_E._field1._timer, _callback, (_cast_data)&_E); ) // If we fixed a callback from a .function assignment, fix the // assignment cast now. @change_timer_function_assignment depends on change_timer_function_usage && (change_callback_handle_cast || change_callback_handle_cast_no_arg || change_callback_handle_arg)@ expression change_timer_function_usage._E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type _cast_func; typedef TIMER_FUNC_TYPE; @@ ( _E->_field1._timer.function = -_callback +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -&_callback +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -(_cast_func)_callback; +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -(_cast_func)&_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -&_callback; +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -(_cast_func)_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -(_cast_func)&_callback +(TIMER_FUNC_TYPE)_callback ; ) // Sometimes timer functions are called directly. Replace matched args. @change_timer_function_calls depends on change_timer_function_usage && (change_callback_handle_cast || change_callback_handle_cast_no_arg || change_callback_handle_arg)@ expression _E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type _cast_data; @@ _callback( ( -(_cast_data)_E +&_E->_field1._timer | -(_cast_data)&_E +&_E._field1._timer | -_E +&_E->_field1._timer ) ) // If a timer has been configured without a data argument, it can be // converted without regard to the callback argument, since it is unused. @match_timer_function_unused_data@ expression _E; identifier _field1; identifier _timer; identifier _callback; @@ ( -setup_timer(&_E->_field1._timer, _callback, 0); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, 0L); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, 0UL); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0L); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0UL); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0L); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0UL); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0); +timer_setup(_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0L); +timer_setup(_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0UL); +timer_setup(_field1._timer, _callback, 0); ) @change_callback_unused_data depends on match_timer_function_unused_data@ identifier match_timer_function_unused_data._callback; type _origtype; identifier _origarg; @@ void _callback( -_origtype _origarg +struct timer_list *unused ) { ... when != _origarg } Signed-off-by: Kees Cook <keescook@chromium.org>
2017-10-18 10:21:24 +07:00
struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
fib6_run_gc(0, arg, true);
}
static int __net_init fib6_net_init(struct net *net)
{
size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
int err;
err = fib6_notifier_init(net);
if (err)
return err;
spin_lock_init(&net->ipv6.fib6_gc_lock);
rwlock_init(&net->ipv6.fib6_walker_lock);
INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
treewide: setup_timer() -> timer_setup() (2 field) This converts all remaining setup_timer() calls that use a nested field to reach a struct timer_list. Coccinelle does not have an easy way to match multiple fields, so a new script is needed to change the matches of "&_E->_timer" into "&_E->_field1._timer" in all the rules. spatch --very-quiet --all-includes --include-headers \ -I ./arch/x86/include -I ./arch/x86/include/generated \ -I ./include -I ./arch/x86/include/uapi \ -I ./arch/x86/include/generated/uapi -I ./include/uapi \ -I ./include/generated/uapi --include ./include/linux/kconfig.h \ --dir . \ --cocci-file ~/src/data/timer_setup-2fields.cocci @fix_address_of depends@ expression e; @@ setup_timer( -&(e) +&e , ...) // Update any raw setup_timer() usages that have a NULL callback, but // would otherwise match change_timer_function_usage, since the latter // will update all function assignments done in the face of a NULL // function initialization in setup_timer(). @change_timer_function_usage_NULL@ expression _E; identifier _field1; identifier _timer; type _cast_data; @@ ( -setup_timer(&_E->_field1._timer, NULL, _E); +timer_setup(&_E->_field1._timer, NULL, 0); | -setup_timer(&_E->_field1._timer, NULL, (_cast_data)_E); +timer_setup(&_E->_field1._timer, NULL, 0); | -setup_timer(&_E._field1._timer, NULL, &_E); +timer_setup(&_E._field1._timer, NULL, 0); | -setup_timer(&_E._field1._timer, NULL, (_cast_data)&_E); +timer_setup(&_E._field1._timer, NULL, 0); ) @change_timer_function_usage@ expression _E; identifier _field1; identifier _timer; struct timer_list _stl; identifier _callback; type _cast_func, _cast_data; @@ ( -setup_timer(&_E->_field1._timer, _callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, &_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, &_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)&_callback, _E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, (_cast_func)&_callback, (_cast_data)_E); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, &_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, &_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)&_callback, (_cast_data)_E); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, (_cast_func)&_callback, (_cast_data)&_E); +timer_setup(&_E._field1._timer, _callback, 0); | _E->_field1._timer@_stl.function = _callback; | _E->_field1._timer@_stl.function = &_callback; | _E->_field1._timer@_stl.function = (_cast_func)_callback; | _E->_field1._timer@_stl.function = (_cast_func)&_callback; | _E._field1._timer@_stl.function = _callback; | _E._field1._timer@_stl.function = &_callback; | _E._field1._timer@_stl.function = (_cast_func)_callback; | _E._field1._timer@_stl.function = (_cast_func)&_callback; ) // callback(unsigned long arg) @change_callback_handle_cast depends on change_timer_function_usage@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _origtype; identifier _origarg; type _handletype; identifier _handle; @@ void _callback( -_origtype _origarg +struct timer_list *t ) { ( ... when != _origarg _handletype *_handle = -(_handletype *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle = -(void *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle; ... when != _handle _handle = -(_handletype *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg | ... when != _origarg _handletype *_handle; ... when != _handle _handle = -(void *)_origarg; +from_timer(_handle, t, _field1._timer); ... when != _origarg ) } // callback(unsigned long arg) without existing variable @change_callback_handle_cast_no_arg depends on change_timer_function_usage && !change_callback_handle_cast@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _origtype; identifier _origarg; type _handletype; @@ void _callback( -_origtype _origarg +struct timer_list *t ) { + _handletype *_origarg = from_timer(_origarg, t, _field1._timer); + ... when != _origarg - (_handletype *)_origarg + _origarg ... when != _origarg } // Avoid already converted callbacks. @match_callback_converted depends on change_timer_function_usage && !change_callback_handle_cast && !change_callback_handle_cast_no_arg@ identifier change_timer_function_usage._callback; identifier t; @@ void _callback(struct timer_list *t) { ... } // callback(struct something *handle) @change_callback_handle_arg depends on change_timer_function_usage && !match_callback_converted && !change_callback_handle_cast && !change_callback_handle_cast_no_arg@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _handletype; identifier _handle; @@ void _callback( -_handletype *_handle +struct timer_list *t ) { + _handletype *_handle = from_timer(_handle, t, _field1._timer); ... } // If change_callback_handle_arg ran on an empty function, remove // the added handler. @unchange_callback_handle_arg depends on change_timer_function_usage && change_callback_handle_arg@ identifier change_timer_function_usage._callback; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; type _handletype; identifier _handle; identifier t; @@ void _callback(struct timer_list *t) { - _handletype *_handle = from_timer(_handle, t, _field1._timer); } // We only want to refactor the setup_timer() data argument if we've found // the matching callback. This undoes changes in change_timer_function_usage. @unchange_timer_function_usage depends on change_timer_function_usage && !change_callback_handle_cast && !change_callback_handle_cast_no_arg && !change_callback_handle_arg@ expression change_timer_function_usage._E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type change_timer_function_usage._cast_data; @@ ( -timer_setup(&_E->_field1._timer, _callback, 0); +setup_timer(&_E->_field1._timer, _callback, (_cast_data)_E); | -timer_setup(&_E._field1._timer, _callback, 0); +setup_timer(&_E._field1._timer, _callback, (_cast_data)&_E); ) // If we fixed a callback from a .function assignment, fix the // assignment cast now. @change_timer_function_assignment depends on change_timer_function_usage && (change_callback_handle_cast || change_callback_handle_cast_no_arg || change_callback_handle_arg)@ expression change_timer_function_usage._E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type _cast_func; typedef TIMER_FUNC_TYPE; @@ ( _E->_field1._timer.function = -_callback +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -&_callback +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -(_cast_func)_callback; +(TIMER_FUNC_TYPE)_callback ; | _E->_field1._timer.function = -(_cast_func)&_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -&_callback; +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -(_cast_func)_callback +(TIMER_FUNC_TYPE)_callback ; | _E._field1._timer.function = -(_cast_func)&_callback +(TIMER_FUNC_TYPE)_callback ; ) // Sometimes timer functions are called directly. Replace matched args. @change_timer_function_calls depends on change_timer_function_usage && (change_callback_handle_cast || change_callback_handle_cast_no_arg || change_callback_handle_arg)@ expression _E; identifier change_timer_function_usage._field1; identifier change_timer_function_usage._timer; identifier change_timer_function_usage._callback; type _cast_data; @@ _callback( ( -(_cast_data)_E +&_E->_field1._timer | -(_cast_data)&_E +&_E._field1._timer | -_E +&_E->_field1._timer ) ) // If a timer has been configured without a data argument, it can be // converted without regard to the callback argument, since it is unused. @match_timer_function_unused_data@ expression _E; identifier _field1; identifier _timer; identifier _callback; @@ ( -setup_timer(&_E->_field1._timer, _callback, 0); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, 0L); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E->_field1._timer, _callback, 0UL); +timer_setup(&_E->_field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0L); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_E._field1._timer, _callback, 0UL); +timer_setup(&_E._field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0L); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(&_field1._timer, _callback, 0UL); +timer_setup(&_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0); +timer_setup(_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0L); +timer_setup(_field1._timer, _callback, 0); | -setup_timer(_field1._timer, _callback, 0UL); +timer_setup(_field1._timer, _callback, 0); ) @change_callback_unused_data depends on match_timer_function_unused_data@ identifier match_timer_function_unused_data._callback; type _origtype; identifier _origarg; @@ void _callback( -_origtype _origarg +struct timer_list *unused ) { ... when != _origarg } Signed-off-by: Kees Cook <keescook@chromium.org>
2017-10-18 10:21:24 +07:00
timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
if (!net->ipv6.rt6_stats)
goto out_timer;
/* Avoid false sharing : Use at least a full cache line */
size = max_t(size_t, size, L1_CACHE_BYTES);
net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
if (!net->ipv6.fib_table_hash)
goto out_rt6_stats;
net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
GFP_KERNEL);
if (!net->ipv6.fib6_main_tbl)
goto out_fib_table_hash;
net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
net->ipv6.fib6_null_entry);
net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
#ifdef CONFIG_IPV6_MULTIPLE_TABLES
net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
GFP_KERNEL);
if (!net->ipv6.fib6_local_tbl)
goto out_fib6_main_tbl;
net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
net->ipv6.fib6_null_entry);
net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
#endif
fib6_tables_init(net);
return 0;
#ifdef CONFIG_IPV6_MULTIPLE_TABLES
out_fib6_main_tbl:
kfree(net->ipv6.fib6_main_tbl);
#endif
out_fib_table_hash:
kfree(net->ipv6.fib_table_hash);
out_rt6_stats:
kfree(net->ipv6.rt6_stats);
out_timer:
fib6_notifier_exit(net);
return -ENOMEM;
}
static void fib6_net_exit(struct net *net)
{
unsigned int i;
del_timer_sync(&net->ipv6.ip6_fib_timer);
for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
struct hlist_head *head = &net->ipv6.fib_table_hash[i];
struct hlist_node *tmp;
struct fib6_table *tb;
hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
hlist_del(&tb->tb6_hlist);
fib6_free_table(tb);
}
}
kfree(net->ipv6.fib_table_hash);
kfree(net->ipv6.rt6_stats);
fib6_notifier_exit(net);
}
static struct pernet_operations fib6_net_ops = {
.init = fib6_net_init,
.exit = fib6_net_exit,
};
int __init fib6_init(void)
{
int ret = -ENOMEM;
fib6_node_kmem = kmem_cache_create("fib6_nodes",
sizeof(struct fib6_node),
0, SLAB_HWCACHE_ALIGN,
NULL);
if (!fib6_node_kmem)
goto out;
ret = register_pernet_subsys(&fib6_net_ops);
if (ret)
goto out_kmem_cache_create;
ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
inet6_dump_fib, 0);
if (ret)
goto out_unregister_subsys;
__fib6_flush_trees = fib6_flush_trees;
out:
return ret;
out_unregister_subsys:
unregister_pernet_subsys(&fib6_net_ops);
out_kmem_cache_create:
kmem_cache_destroy(fib6_node_kmem);
goto out;
}
void fib6_gc_cleanup(void)
{
unregister_pernet_subsys(&fib6_net_ops);
kmem_cache_destroy(fib6_node_kmem);
}
#ifdef CONFIG_PROC_FS
static int ipv6_route_seq_show(struct seq_file *seq, void *v)
{
struct fib6_info *rt = v;
struct ipv6_route_iter *iter = seq->private;
const struct net_device *dev;
seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
#ifdef CONFIG_IPV6_SUBTREES
seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
#else
seq_puts(seq, "00000000000000000000000000000000 00 ");
#endif
if (rt->fib6_flags & RTF_GATEWAY)
seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
else
seq_puts(seq, "00000000000000000000000000000000");
dev = rt->fib6_nh.nh_dev;
seq_printf(seq, " %08x %08x %08x %08x %8s\n",
rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
rt->fib6_flags, dev ? dev->name : "");
iter->w.leaf = NULL;
return 0;
}
static int ipv6_route_yield(struct fib6_walker *w)
{
struct ipv6_route_iter *iter = w->args;
if (!iter->skip)
return 1;
do {
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
iter->w.leaf = rcu_dereference_protected(
iter->w.leaf->fib6_next,
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
lockdep_is_held(&iter->tbl->tb6_lock));
iter->skip--;
if (!iter->skip && iter->w.leaf)
return 1;
} while (iter->w.leaf);
return 0;
}
static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
struct net *net)
{
memset(&iter->w, 0, sizeof(iter->w));
iter->w.func = ipv6_route_yield;
iter->w.root = &iter->tbl->tb6_root;
iter->w.state = FWS_INIT;
iter->w.node = iter->w.root;
iter->w.args = iter;
iter->sernum = iter->w.root->fn_sernum;
INIT_LIST_HEAD(&iter->w.lh);
fib6_walker_link(net, &iter->w);
}
static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
struct net *net)
{
unsigned int h;
struct hlist_node *node;
if (tbl) {
h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
} else {
h = 0;
node = NULL;
}
while (!node && h < FIB6_TABLE_HASHSZ) {
node = rcu_dereference_bh(
hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
}
return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
}
static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
{
if (iter->sernum != iter->w.root->fn_sernum) {
iter->sernum = iter->w.root->fn_sernum;
iter->w.state = FWS_INIT;
iter->w.node = iter->w.root;
WARN_ON(iter->w.skip);
iter->w.skip = iter->w.count;
}
}
static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
int r;
struct fib6_info *n;
struct net *net = seq_file_net(seq);
struct ipv6_route_iter *iter = seq->private;
if (!v)
goto iter_table;
n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
if (n) {
++*pos;
return n;
}
iter_table:
ipv6_route_check_sernum(iter);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_lock_bh(&iter->tbl->tb6_lock);
r = fib6_walk_continue(&iter->w);
ipv6: replace rwlock with rcu and spinlock in fib6_table With all the preparation work before, we are now ready to replace rwlock with rcu and spinlock in fib6_table. That means now all fib6_node in fib6_table are protected by rcu. And when freeing fib6_node, call_rcu() is used to wait for the rcu grace period before releasing the memory. When accessing fib6_node, corresponding rcu APIs need to be used. And all previous sessions protected by the write lock will now be protected by the spin lock per table. All previous sessions protected by read lock will now be protected by rcu_read_lock(). A couple of things to note here: 1. As part of the work of replacing rwlock with rcu, the linked list of fn->leaf now has to be rcu protected as well. So both fn->leaf and rt->dst.rt6_next are now __rcu tagged and corresponding rcu APIs are used when manipulating them. 2. For fn->rr_ptr, first of all, it also needs to be rcu protected now and is tagged with __rcu and rcu APIs are used in corresponding places. Secondly, fn->rr_ptr is changed in rt6_select() which is a reader thread. This makes the issue a bit complicated. We think a valid solution for it is to let rt6_select() grab the tb6_lock if it decides to change it. As it is not in the normal operation and only happens when there is no valid neighbor cache for the route, we think the performance impact should be low. 3. fib6_walk_continue() has to be called with tb6_lock held even in the route dumping related functions, e.g. inet6_dump_fib(), fib6_tables_dump() and ipv6_route_seq_ops. It is because fib6_walk_continue() makes modifications to the walker structure, and so are fib6_repair_tree() and fib6_del_route(). In order to do proper syncing between them, we need to let fib6_walk_continue() hold the lock. We may be able to do further improvement on the way we do the tree walk to get rid of the need for holding the spin lock. But not for now. 4. When fib6_del_route() removes a route from the tree, we no longer mark rt->dst.rt6_next to NULL to make simultaneous reader be able to further traverse the list with rcu. However, rt->dst.rt6_next is only valid within this same rcu period. No one should access it later. 5. All the operation of atomic_inc(rt->rt6i_ref) is changed to be performed before we publish this route (either by linking it to fn->leaf or insert it in the list pointed by fn->leaf) just to be safe because as soon as we publish the route, some read thread will be able to access it. Signed-off-by: Wei Wang <weiwan@google.com> Signed-off-by: Martin KaFai Lau <kafai@fb.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-10-07 02:06:10 +07:00
spin_unlock_bh(&iter->tbl->tb6_lock);
if (r > 0) {
if (v)
++*pos;
return iter->w.leaf;
} else if (r < 0) {
fib6_walker_unlink(net, &iter->w);
return NULL;
}
fib6_walker_unlink(net, &iter->w);
iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
if (!iter->tbl)
return NULL;
ipv6_route_seq_setup_walk(iter, net);
goto iter_table;
}
static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
__acquires(RCU_BH)
{
struct net *net = seq_file_net(seq);
struct ipv6_route_iter *iter = seq->private;
rcu_read_lock_bh();
iter->tbl = ipv6_route_seq_next_table(NULL, net);
iter->skip = *pos;
if (iter->tbl) {
ipv6_route_seq_setup_walk(iter, net);
return ipv6_route_seq_next(seq, NULL, pos);
} else {
return NULL;
}
}
static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
{
struct fib6_walker *w = &iter->w;
return w->node && !(w->state == FWS_U && w->node == w->root);
}
static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
__releases(RCU_BH)
{
struct net *net = seq_file_net(seq);
struct ipv6_route_iter *iter = seq->private;
if (ipv6_route_iter_active(iter))
fib6_walker_unlink(net, &iter->w);
rcu_read_unlock_bh();
}
const struct seq_operations ipv6_route_seq_ops = {
.start = ipv6_route_seq_start,
.next = ipv6_route_seq_next,
.stop = ipv6_route_seq_stop,
.show = ipv6_route_seq_show
};
#endif /* CONFIG_PROC_FS */