Currently, we unhash a dying net from netns_ids lists
under rtnl_lock(). It's a leftover from the time when
net::netns_ids was introduced. There was no net::nsid_lock,
and rtnl_lock() was mostly need to order modification
of alive nets nsid idr, i.e. for:
for_each_net(tmp) {
...
id = __peernet2id(tmp, net);
idr_remove(&tmp->netns_ids, id);
...
}
Since we have net::nsid_lock, the modifications are
protected by this local lock, and now we may introduce
better scheme of netns_ids destruction.
Let's look at the functions peernet2id_alloc() and
get_net_ns_by_id(). Previous commits taught these
functions to work well with dying net acquired from
rtnl unlocked lists. And they are the only functions
which can hash a net to netns_ids or obtain from there.
And as easy to check, other netns_ids operating functions
works with id, not with net pointers. So, we do not
need rtnl_lock to synchronize cleanup_net() with all them.
The another property, which is used in the patch,
is that net is unhashed from net_namespace_list
in the only place and by the only process. So,
we avoid excess rcu_read_lock() or rtnl_lock(),
when we'are iterating over the list in unhash_nsid().
All the above makes possible to keep rtnl_lock() locked
only for net->list deletion, and completely avoid it
for netns_ids unhashing and destruction. As these two
doings may take long time (e.g., memory allocation
to send skb), the patch should positively act on
the scalability and signify decrease the time, which
rtnl_lock() is held in cleanup_net().
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
idr_find() is safe under rcu_read_lock() and
maybe_get_net() guarantees that net is alive.
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
peernet2id_alloc() is racy without rtnl_lock() as refcount_read(&peer->count)
under net->nsid_lock does not guarantee, peer is alive:
rcu_read_lock()
peernet2id_alloc() ..
spin_lock_bh(&net->nsid_lock) ..
refcount_read(&peer->count) (!= 0) ..
.. put_net()
.. cleanup_net()
.. for_each_net(tmp)
.. spin_lock_bh(&tmp->nsid_lock)
.. __peernet2id(tmp, net) == -1
.. ..
.. ..
__peernet2id_alloc(alloc == true) ..
.. ..
rcu_read_unlock() ..
.. synchronize_rcu()
.. kmem_cache_free(net)
After the above situation, net::netns_id contains id pointing to freed memory,
and any other dereferencing by the id will operate with this freed memory.
Currently, peernet2id_alloc() is used under rtnl_lock() everywhere except
ovs_vport_cmd_fill_info(), and this race can't occur. But peernet2id_alloc()
is generic interface, and better we fix it before someone really starts
use it in wrong context.
v2: Don't place refcount_read(&net->count) under net->nsid_lock
as suggested by Eric W. Biederman <ebiederm@xmission.com>
v3: Rebase on top of net-next
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since net could be obtained from RCU lists,
and there is a race with net destruction,
the patch converts net::count to refcount_t.
This provides sanity checks for the cases of
incrementing counter of already dead net,
when maybe_get_net() has to used instead
of get_net().
Drivers: allyesconfig and allmodconfig are OK.
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
(I can trivially verify that that idr_remove in cleanup_net happens
after the network namespace count has dropped to zero --EWB)
Function get_net_ns_by_id() does not check for net::count
after it has found a peer in netns_ids idr.
It may dereference a peer, after its count has already been
finaly decremented. This leads to double free and memory
corruption:
put_net(peer) rtnl_lock()
atomic_dec_and_test(&peer->count) [count=0] ...
__put_net(peer) get_net_ns_by_id(net, id)
spin_lock(&cleanup_list_lock)
list_add(&net->cleanup_list, &cleanup_list)
spin_unlock(&cleanup_list_lock)
queue_work() peer = idr_find(&net->netns_ids, id)
| get_net(peer) [count=1]
| ...
| (use after final put)
v ...
cleanup_net() ...
spin_lock(&cleanup_list_lock) ...
list_replace_init(&cleanup_list, ..) ...
spin_unlock(&cleanup_list_lock) ...
... ...
... put_net(peer)
... atomic_dec_and_test(&peer->count) [count=0]
... spin_lock(&cleanup_list_lock)
... list_add(&net->cleanup_list, &cleanup_list)
... spin_unlock(&cleanup_list_lock)
... queue_work()
... rtnl_unlock()
rtnl_lock() ...
for_each_net(tmp) { ...
id = __peernet2id(tmp, peer) ...
spin_lock_irq(&tmp->nsid_lock) ...
idr_remove(&tmp->netns_ids, id) ...
... ...
net_drop_ns() ...
net_free(peer) ...
} ...
|
v
cleanup_net()
...
(Second free of peer)
Also, put_net() on the right cpu may reorder with left's cpu
list_replace_init(&cleanup_list, ..), and then cleanup_list
will be corrupted.
Since cleanup_net() is executed in worker thread, while
put_net(peer) can happen everywhere, there should be
enough time for concurrent get_net_ns_by_id() to pick
the peer up, and the race does not seem to be unlikely.
The patch fixes the problem in standard way.
(Also, there is possible problem in peernet2id_alloc(), which requires
check for net::count under nsid_lock and maybe_get_net(peer), but
in current stable kernel it's used under rtnl_lock() and it has to be
safe. Openswitch begun to use peernet2id_alloc(), and possibly it should
be fixed too. While this is not in stable kernel yet, so I'll send
a separate message to netdev@ later).
Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Fixes: 0c7aecd4bd "netns: add rtnl cmd to add and get peer netns ids"
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Both functions take nsid_lock and don't rely on rtnl lock.
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This change allows us to later indicate to rtnetlink core that certain
doit functions should be called without acquiring rtnl_mutex.
This change should have no effect, we simply replace the last (now
unused) calcit argument with the new flag.
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
refcount_t type and corresponding API should be
used instead of atomic_t when the variable is used as
a reference counter. This allows to avoid accidental
refcounter overflows that might lead to use-after-free
situations.
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
Signed-off-by: Hans Liljestrand <ishkamiel@gmail.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David Windsor <dwindsor@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pablo Neira Ayuso says:
====================
Netfilter updates for net-next
The following patchset contains Netfilter updates for your net-next
tree. This batch contains connection tracking updates for the cleanup
iteration path, patches from Florian Westphal:
X) Skip unconfirmed conntracks in nf_ct_iterate_cleanup_net(), just set
dying bit to let the CPU release them.
X) Add nf_ct_iterate_destroy() to be used on module removal, to kill
conntrack from all namespace.
X) Restart iteration on hashtable resizing, since both may occur at
the same time.
X) Use the new nf_ct_iterate_destroy() to remove conntrack with NAT
mapping on module removal.
X) Use nf_ct_iterate_destroy() to remove conntrack entries helper
module removal, from Liping Zhang.
X) Use nf_ct_iterate_cleanup_net() to remove the timeout extension
if user requests this, also from Liping.
X) Add net_ns_barrier() and use it from FTP helper, so make sure
no concurrent namespace removal happens at the same time while
the helper module is being removed.
X) Use NFPROTO_MAX in layer 3 conntrack protocol array, to reduce
module size. Same thing in nf_tables.
Updates for the nf_tables infrastructure:
X) Prepare usage of the extended ACK reporting infrastructure for
nf_tables.
X) Remove unnecessary forward declaration in nf_tables hash set.
X) Skip set size estimation if number of element is not specified.
X) Changes to accomodate a (faster) unresizable hash set implementation,
for anonymous sets and dynamic size fixed sets with no timeouts.
X) Faster lookup function for unresizable hash table for 2 and 4
bytes key.
And, finally, a bunch of asorted small updates and cleanups:
X) Do not hold reference to netdev from ipt_CLUSTER, instead subscribe
to device events and look up for index from the packet path, this
is fixing an issue that is present since the very beginning, patch
from Xin Long.
X) Use nf_register_net_hook() in ipt_CLUSTER, from Florian Westphal.
X) Use ebt_invalid_target() whenever possible in the ebtables tree,
from Gao Feng.
X) Calm down compilation warning in nf_dup infrastructure, patch from
stephen hemminger.
X) Statify functions in nftables rt expression, also from stephen.
X) Update Makefile to use canonical method to specify nf_tables-objs.
From Jike Song.
X) Use nf_conntrack_helpers_register() in amanda and H323.
X) Space cleanup for ctnetlink, from linzhang.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Quoting Joe Stringer:
If a user loads nf_conntrack_ftp, sends FTP traffic through a network
namespace, destroys that namespace then unloads the FTP helper module,
then the kernel will crash.
Events that lead to the crash:
1. conntrack is created with ftp helper in netns x
2. This netns is destroyed
3. netns destruction is scheduled
4. netns destruction wq starts, removes netns from global list
5. ftp helper is unloaded, which resets all helpers of the conntracks
via for_each_net()
but because netns is already gone from list the for_each_net() loop
doesn't include it, therefore all of these conntracks are unaffected.
6. helper module unload finishes
7. netns wq invokes destructor for rmmod'ed helper
CC: "Eric W. Biederman" <ebiederm@xmission.com>
Reported-by: Joe Stringer <joe@ovn.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
When the user tries to assign a specific nsid, idr_alloc() is called with
the range [nsid, nsid+1]. If this nsid is already used, idr_alloc() returns
ENOSPC (No space left on device). In our case, it's better to return
EEXIST to make it clear that the nsid is not available.
CC: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It helps the user to identify errors.
CC: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The default value for somaxconn is set in sysctl_core_net_init(), but this
function is not called when kernel is configured without CONFIG_SYSCTL.
This results in the kernel not being able to accept TCP connections,
because the backlog has zero size. Usually, the user ends up with:
"TCP: request_sock_TCP: Possible SYN flooding on port 7. Dropping request. Check SNMP counters."
If SYN cookies are not enabled the connection is rejected.
Before ef547f2ac1 (tcp: remove max_qlen_log), the effects were less
severe, because the backlog was always at least eight slots long.
Signed-off-by: Roman Kapl <roman.kapl@sysgo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Initialise init_net.count to 1 for its pointer from init_nsproxy lest
someone tries to do a get_net() and a put_net() in a process in which
current->ns_proxy->net_ns points to the initial network namespace.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add netlink_ext_ack arg to rtnl_doit_func. Pass extack arg to nlmsg_parse
for doit functions that call it directly.
This is the first step to using extended error reporting in rtnetlink.
>From here individual subsystems can be updated to set netlink_ext_ack as
needed.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pass the new extended ACK reporting struct to all of the generic
netlink parsing functions. For now, pass NULL in almost all callers
(except for some in the core.)
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
But first update the code that uses these facilities with the
new header.
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull audit updates from Paul Moore:
"After the small number of patches for v4.9, we've got a much bigger
pile for v4.10.
The bulk of these patches involve a rework of the audit backlog queue
to enable us to move the netlink multicasting out of the task/thread
that generates the audit record and into the kernel thread that emits
the record (just like we do for the audit unicast to auditd).
While we were playing with the backlog queue(s) we fixed a number of
other little problems with the code, and from all the testing so far
things look to be in much better shape now. Doing this also allowed us
to re-enable disabling IRQs for some netns operations ("netns: avoid
disabling irq for netns id").
The remaining patches fix some small problems that are well documented
in the commit descriptions, as well as adding session ID filtering
support"
* 'stable-4.10' of git://git.infradead.org/users/pcmoore/audit:
audit: use proper refcount locking on audit_sock
netns: avoid disabling irq for netns id
audit: don't ever sleep on a command record/message
audit: handle a clean auditd shutdown with grace
audit: wake up kauditd_thread after auditd registers
audit: rework audit_log_start()
audit: rework the audit queue handling
audit: rename the queues and kauditd related functions
audit: queue netlink multicast sends just like we do for unicast sends
audit: fixup audit_init()
audit: move kaudit thread start from auditd registration to kaudit init (#2)
audit: add support for session ID user filter
audit: fix formatting of AUDIT_CONFIG_CHANGE events
audit: skip sessionid sentinel value when auto-incrementing
audit: tame initialization warning len_abuf in audit_log_execve_info
audit: less stack usage for /proc/*/loginuid
net_generic() function is both a) inline and b) used ~600 times.
It has the following code inside
...
ptr = ng->ptr[id - 1];
...
"id" is never compile time constant so compiler is forced to subtract 1.
And those decrements or LEA [r32 - 1] instructions add up.
We also start id'ing from 1 to catch bugs where pernet sybsystem id
is not initialized and 0. This is quite pointless idea (nothing will
work or immediate interference with first registered subsystem) in
general but it hints what needs to be done for code size reduction.
Namely, overlaying allocation of pointer array and fixed part of
structure in the beginning and using usual base-0 addressing.
Ids are just cookies, their exact values do not matter, so lets start
with 3 on x86_64.
Code size savings (oh boy): -4.2 KB
As usual, ignore the initial compiler stupidity part of the table.
add/remove: 0/0 grow/shrink: 12/670 up/down: 89/-4297 (-4208)
function old new delta
tipc_nametbl_insert_publ 1250 1270 +20
nlmclnt_lookup_host 686 703 +17
nfsd4_encode_fattr 5930 5941 +11
nfs_get_client 1050 1061 +11
register_pernet_operations 333 342 +9
tcf_mirred_init 843 849 +6
tcf_bpf_init 1143 1149 +6
gss_setup_upcall 990 994 +4
idmap_name_to_id 432 434 +2
ops_init 274 275 +1
nfsd_inject_forget_client 259 260 +1
nfs4_alloc_client 612 613 +1
tunnel_key_walker 164 163 -1
...
tipc_bcbase_select_primary 392 360 -32
mac80211_hwsim_new_radio 2808 2767 -41
ipip6_tunnel_ioctl 2228 2186 -42
tipc_bcast_rcv 715 672 -43
tipc_link_build_proto_msg 1140 1089 -51
nfsd4_lock 3851 3796 -55
tipc_mon_rcv 1012 956 -56
Total: Before=156643951, After=156639743, chg -0.00%
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is precursor to fixing "[id - 1]" bloat inside net_generic().
Name "s" is chosen to complement name "u" often used for dummy unions.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Publishing net_generic pointer is done with silly mistake: new array is
published BEFORE setting freshly acquired pernet subsystem pointer.
memcpy
rcu_assign_pointer
kfree_rcu
ng->ptr[id - 1] = data;
This bug was introduced with commit dec827d174
("[NETNS]: The generic per-net pointers.") in the glorious days of
chopping networking stack into containers proper 8.5 years ago (whee...)
How it didn't trigger for so long?
Well, you need quite specific set of conditions:
*) race window opens once per pernet subsystem addition
(read: modprobe or boot)
*) not every pernet subsystem is eligible (need ->id and ->size)
*) not every pernet subsystem is vulnerable (need incorrect or absense
of ordering of register_pernet_sybsys() and actually using net_generic())
*) to hide the bug even more, default is to preallocate 13 pointers which
is actually quite a lot. You need IPv6, netfilter, bridging etc together
loaded to trigger reallocation in the first place. Trimmed down
config are OK.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
All conflicts were simple overlapping changes except perhaps
for the Thunder driver.
That driver has a change_mtu method explicitly for sending
a message to the hardware. If that fails it returns an
error.
Normally a driver doesn't need an ndo_change_mtu method becuase those
are usually just range changes, which are now handled generically.
But since this extra operation is needed in the Thunder driver, it has
to stay.
However, if the message send fails we have to restore the original
MTU before the change because the entire call chain expects that if
an error is thrown by ndo_change_mtu then the MTU did not change.
Therefore code is added to nicvf_change_mtu to remember the original
MTU, and to restore it upon nicvf_update_hw_max_frs() failue.
Signed-off-by: David S. Miller <davem@davemloft.net>
Make struct pernet_operations::id unsigned.
There are 2 reasons to do so:
1)
This field is really an index into an zero based array and
thus is unsigned entity. Using negative value is out-of-bound
access by definition.
2)
On x86_64 unsigned 32-bit data which are mixed with pointers
via array indexing or offsets added or subtracted to pointers
are preffered to signed 32-bit data.
"int" being used as an array index needs to be sign-extended
to 64-bit before being used.
void f(long *p, int i)
{
g(p[i]);
}
roughly translates to
movsx rsi, esi
mov rdi, [rsi+...]
call g
MOVSX is 3 byte instruction which isn't necessary if the variable is
unsigned because x86_64 is zero extending by default.
Now, there is net_generic() function which, you guessed it right, uses
"int" as an array index:
static inline void *net_generic(const struct net *net, int id)
{
...
ptr = ng->ptr[id - 1];
...
}
And this function is used a lot, so those sign extensions add up.
Patch snipes ~1730 bytes on allyesconfig kernel (without all junk
messing with code generation):
add/remove: 0/0 grow/shrink: 70/598 up/down: 396/-2126 (-1730)
Unfortunately some functions actually grow bigger.
This is a semmingly random artefact of code generation with register
allocator being used differently. gcc decides that some variable
needs to live in new r8+ registers and every access now requires REX
prefix. Or it is shifted into r12, so [r12+0] addressing mode has to be
used which is longer than [r8]
However, overall balance is in negative direction:
add/remove: 0/0 grow/shrink: 70/598 up/down: 396/-2126 (-1730)
function old new delta
nfsd4_lock 3886 3959 +73
tipc_link_build_proto_msg 1096 1140 +44
mac80211_hwsim_new_radio 2776 2808 +32
tipc_mon_rcv 1032 1058 +26
svcauth_gss_legacy_init 1413 1429 +16
tipc_bcbase_select_primary 379 392 +13
nfsd4_exchange_id 1247 1260 +13
nfsd4_setclientid_confirm 782 793 +11
...
put_client_renew_locked 494 480 -14
ip_set_sockfn_get 730 716 -14
geneve_sock_add 829 813 -16
nfsd4_sequence_done 721 703 -18
nlmclnt_lookup_host 708 686 -22
nfsd4_lockt 1085 1063 -22
nfs_get_client 1077 1050 -27
tcf_bpf_init 1106 1076 -30
nfsd4_encode_fattr 5997 5930 -67
Total: Before=154856051, After=154854321, chg -0.00%
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Andrei reports we still allocate netns ID from idr after we destroy
it in cleanup_net().
cleanup_net():
...
idr_destroy(&net->netns_ids);
...
list_for_each_entry_reverse(ops, &pernet_list, list)
ops_exit_list(ops, &net_exit_list);
-> rollback_registered_many()
-> rtmsg_ifinfo_build_skb()
-> rtnl_fill_ifinfo()
-> peernet2id_alloc()
After that point we should not even access net->netns_ids, we
should check the death of the current netns as early as we can in
peernet2id_alloc().
For net-next we can consider to avoid sending rtmsg totally,
it is a good optimization for netns teardown path.
Fixes: 0c7aecd4bd ("netns: add rtnl cmd to add and get peer netns ids")
Reported-by: Andrei Vagin <avagin@gmail.com>
Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Andrei Vagin <avagin@openvz.org>
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Mostly simple overlapping changes.
For example, David Ahern's adjacency list revamp in 'net-next'
conflicted with an adjacency list traversal bug fix in 'net'.
Signed-off-by: David S. Miller <davem@davemloft.net>
net_mutex can be locked for a long time. It may be because many
namespaces are being destroyed or many processes decide to create
a network namespace.
Both these operations are heavy, so it is better to have an ability to
kill a process which is waiting net_mutex.
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Andrei Vagin <avagin@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit bc51dddf98 ("netns: avoid disabling irq for
netns id") as it was found to cause problems with systems running
SELinux/audit, see the mailing list thread below:
* http://marc.info/?t=147694653900002&r=1&w=2
Eventually we should be able to reintroduce this code once we have
rewritten the audit multicast code to queue messages much the same
way we do for unicast messages. A tracking issue for this can be
found below:
* https://github.com/linux-audit/audit-kernel/issues/23
Reported-by: Stephen Smalley <sds@tycho.nsa.gov>
Reported-by: Elad Raz <e@eladraz.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Paul Moore <paul@paul-moore.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull namespace updates from Eric Biederman:
"This set of changes is a number of smaller things that have been
overlooked in other development cycles focused on more fundamental
change. The devpts changes are small things that were a distraction
until we managed to kill off DEVPTS_MULTPLE_INSTANCES. There is an
trivial regression fix to autofs for the unprivileged mount changes
that went in last cycle. A pair of ioctls has been added by Andrey
Vagin making it is possible to discover the relationships between
namespaces when referring to them through file descriptors.
The big user visible change is starting to add simple resource limits
to catch programs that misbehave. With namespaces in general and user
namespaces in particular allowing users to use more kinds of
resources, it has become important to have something to limit errant
programs. Because the purpose of these limits is to catch errant
programs the code needs to be inexpensive to use as it always on, and
the default limits need to be high enough that well behaved programs
on well behaved systems don't encounter them.
To this end, after some review I have implemented per user per user
namespace limits, and use them to limit the number of namespaces. The
limits being per user mean that one user can not exhause the limits of
another user. The limits being per user namespace allow contexts where
the limit is 0 and security conscious folks can remove from their
threat anlysis the code used to manage namespaces (as they have
historically done as it root only). At the same time the limits being
per user namespace allow other parts of the system to use namespaces.
Namespaces are increasingly being used in application sand boxing
scenarios so an all or nothing disable for the entire system for the
security conscious folks makes increasing use of these sandboxes
impossible.
There is also added a limit on the maximum number of mounts present in
a single mount namespace. It is nontrivial to guess what a reasonable
system wide limit on the number of mount structure in the kernel would
be, especially as it various based on how a system is using
containers. A limit on the number of mounts in a mount namespace
however is much easier to understand and set. In most cases in
practice only about 1000 mounts are used. Given that some autofs
scenarious have the potential to be 30,000 to 50,000 mounts I have set
the default limit for the number of mounts at 100,000 which is well
above every known set of users but low enough that the mount hash
tables don't degrade unreaonsably.
These limits are a start. I expect this estabilishes a pattern that
other limits for resources that namespaces use will follow. There has
been interest in making inotify event limits per user per user
namespace as well as interest expressed in making details about what
is going on in the kernel more visible"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace: (28 commits)
autofs: Fix automounts by using current_real_cred()->uid
mnt: Add a per mount namespace limit on the number of mounts
netns: move {inc,dec}_net_namespaces into #ifdef
nsfs: Simplify __ns_get_path
tools/testing: add a test to check nsfs ioctl-s
nsfs: add ioctl to get a parent namespace
nsfs: add ioctl to get an owning user namespace for ns file descriptor
kernel: add a helper to get an owning user namespace for a namespace
devpts: Change the owner of /dev/pts/ptmx to the mounter of /dev/pts
devpts: Remove sync_filesystems
devpts: Make devpts_kill_sb safe if fsi is NULL
devpts: Simplify devpts_mount by using mount_nodev
devpts: Move the creation of /dev/pts/ptmx into fill_super
devpts: Move parse_mount_options into fill_super
userns: When the per user per user namespace limit is reached return ENOSPC
userns; Document per user per user namespace limits.
mntns: Add a limit on the number of mount namespaces.
netns: Add a limit on the number of net namespaces
cgroupns: Add a limit on the number of cgroup namespaces
ipcns: Add a limit on the number of ipc namespaces
...
With the newly enforced limit on the number of namespaces,
we get a build warning if CONFIG_NETNS is disabled:
net/core/net_namespace.c:273:13: error: 'dec_net_namespaces' defined but not used [-Werror=unused-function]
net/core/net_namespace.c:268:24: error: 'inc_net_namespaces' defined but not used [-Werror=unused-function]
This moves the two added functions inside the #ifdef that guards
their callers.
Fixes: 703286608a ("netns: Add a limit on the number of net namespaces")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
From: Andrey Vagin <avagin@openvz.org>
Each namespace has an owning user namespace and now there is not way
to discover these relationships.
Pid and user namepaces are hierarchical. There is no way to discover
parent-child relationships too.
Why we may want to know relationships between namespaces?
One use would be visualization, in order to understand the running
system. Another would be to answer the question: what capability does
process X have to perform operations on a resource governed by namespace
Y?
One more use-case (which usually called abnormal) is checkpoint/restart.
In CRIU we are going to dump and restore nested namespaces.
There [1] was a discussion about which interface to choose to determing
relationships between namespaces.
Eric suggested to add two ioctl-s [2]:
> Grumble, Grumble. I think this may actually a case for creating ioctls
> for these two cases. Now that random nsfs file descriptors are bind
> mountable the original reason for using proc files is not as pressing.
>
> One ioctl for the user namespace that owns a file descriptor.
> One ioctl for the parent namespace of a namespace file descriptor.
Here is an implementaions of these ioctl-s.
$ man man7/namespaces.7
...
Since Linux 4.X, the following ioctl(2) calls are supported for
namespace file descriptors. The correct syntax is:
fd = ioctl(ns_fd, ioctl_type);
where ioctl_type is one of the following:
NS_GET_USERNS
Returns a file descriptor that refers to an owning user names‐
pace.
NS_GET_PARENT
Returns a file descriptor that refers to a parent namespace.
This ioctl(2) can be used for pid and user namespaces. For
user namespaces, NS_GET_PARENT and NS_GET_USERNS have the same
meaning.
In addition to generic ioctl(2) errors, the following specific ones
can occur:
EINVAL NS_GET_PARENT was called for a nonhierarchical namespace.
EPERM The requested namespace is outside of the current namespace
scope.
[1] https://lkml.org/lkml/2016/7/6/158
[2] https://lkml.org/lkml/2016/7/9/101
Changes for v2:
* don't return ENOENT for init_user_ns and init_pid_ns. There is nothing
outside of the init namespace, so we can return EPERM in this case too.
> The fewer special cases the easier the code is to get
> correct, and the easier it is to read. // Eric
Changes for v3:
* rename ns->get_owner() to ns->owner(). get_* usually means that it
grabs a reference.
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: "Michael Kerrisk (man-pages)" <mtk.manpages@gmail.com>
Cc: "W. Trevor King" <wking@tremily.us>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Serge Hallyn <serge.hallyn@canonical.com>
Return -EPERM if an owning user namespace is outside of a process
current user namespace.
v2: In a first version ns_get_owner returned ENOENT for init_user_ns.
This special cases was removed from this version. There is nothing
outside of init_user_ns, so we can return EPERM.
v3: rename ns->get_owner() to ns->owner(). get_* usually means that it
grabs a reference.
Acked-by: Serge Hallyn <serge@hallyn.com>
Signed-off-by: Andrei Vagin <avagin@openvz.org>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
The current error codes returned when a the per user per user
namespace limit are hit (EINVAL, EUSERS, and ENFILE) are wrong. I
asked for advice on linux-api and it we made clear that those were
the wrong error code, but a correct effor code was not suggested.
The best general error code I have found for hitting a resource limit
is ENOSPC. It is not perfect but as it is unambiguous it will serve
until someone comes up with a better error code.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
We never read or change netns id in hardirq context,
the only place we read netns id in softirq context
is in vxlan_xmit(). So, it should be enough to just
disable BH.
Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
netns id should be already allocated each time we change
netns, that is, in dev_change_net_namespace() (more precisely
in rtnl_fill_ifinfo()). It is safe to just call peernet2id() here.
Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When CONFIG_NET_NS is disabled, registering pernet operations causes
init() to be called immediately with init_net as an argument. Unfortunately
this leads to some pernet ops, such as proc_net_ns_init() to be called too
early, when init_net namespace has not been fully initialized. This causes
issues when we want to change pernet ops to use more data from the net
namespace in question, for example reference user namespace that owns our
network namespace.
To fix this we could either play game of musical chairs and rearrange init
order, or we could do the same as when CONFIG_NET_NS is enabled, and
postpone calling pernet ops->init() until namespace is set up properly.
Note that we can not simply undo commit ed160e839d ("[NET]: Cleanup
pernet operation without CONFIG_NET_NS") and use the same implementations
for __register_pernet_operations() and __unregister_pernet_operations(),
because many pernet ops are marked as __net_initdata and will be discarded,
which wreaks havoc on our ops lists. Here we rely on the fact that we only
use lists until init_net is fully initialized, which happens much earlier
than discarding __net_initdata sections.
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The spinlock is used to protect netns_ids which is per net,
so there is no need to use a global spinlock.
Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Unlock was missing on error path.
Fixes: 95f38411df ("netns: use a spin_lock to protect nsid management")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Four minor merge conflicts:
1) qca_spi.c renamed the local variable used for the SPI device
from spi_device to spi, meanwhile the spi_set_drvdata() call
got moved further up in the probe function.
2) Two changes were both adding new members to codel params
structure, and thus we had overlapping changes to the
initializer function.
3) 'net' was making a fix to sk_release_kernel() which is
completely removed in 'net-next'.
4) In net_namespace.c, the rtnl_net_fill() call for GET operations
had the command value fixed, meanwhile 'net-next' adjusted the
argument signature a bit.
This also matches example merge resolutions posted by Stephen
Rothwell over the past two days.
Signed-off-by: David S. Miller <davem@davemloft.net>
Usually, RTM_NEWxxx is returned on a get (same as a dump).
Fixes: 0c7aecd4bd ("netns: add rtnl cmd to add and get peer netns ids")
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
More accurately, listen all netns that have a nsid assigned into the netns
where the netlink socket is opened.
For this purpose, a netlink socket option is added:
NETLINK_LISTEN_ALL_NSID. When this option is set on a netlink socket, this
socket will receive netlink notifications from all netns that have a nsid
assigned into the netns where the socket has been opened. The nsid is sent
to userland via an anscillary data.
With this patch, a daemon needs only one socket to listen many netns. This
is useful when the number of netns is high.
Because 0 is a valid value for a nsid, the field nsid_is_set indicates if
the field nsid is valid or not. skb->cb is initialized to 0 on skb
allocation, thus we are sure that we will never send a nsid 0 by error to
the userland.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Before this patch, nsid were protected by the rtnl lock. The goal of this
patch is to be able to find a nsid without needing to hold the rtnl lock.
The next patch will introduce a netlink socket option to listen to all
netns that have a nsid assigned into the netns where the socket is opened.
Thus, it's important to call rtnl_net_notifyid() outside the spinlock, to
avoid a recursive lock (nsid are notified via rtnl). This was the main
reason of the previous patch.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no functional change with this patch. It will ease the refactoring
of the locking system that protects nsids and the support of the netlink
socket option NETLINK_LISTEN_ALL_NSID.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
In a following commit, a new function will be introduced to only lookup for
a nsid (no allocation if the nsid doesn't exist). To avoid confusion, the
existing function is renamed.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
The goal of this commit is to prepare the rework of the locking of nsnid
protection.
After this patch, rtnl_net_notifyid() will not call anymore __peernet2id(),
ie no idr_* operation into this function.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
All callers of this function expect a nsid, not an error.
Thus, returns NETNSA_NSID_NOT_ASSIGNED in case of error so that callers
don't have to convert the error to NETNSA_NSID_NOT_ASSIGNED.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove duplicated include.
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
