refcounts have a generic implementation and an asm optimized one. The
generic version has extra debugging to make sure that once a refcount
goes to zero, refcount_inc won't increase it.
The btrfs delayed inode code wasn't expecting this, and we're tripping
over the warnings when the generic refcounts are used. We ended up with
this race:
Process A Process B
btrfs_get_delayed_node()
spin_lock(root->inode_lock)
radix_tree_lookup()
__btrfs_release_delayed_node()
refcount_dec_and_test(&delayed_node->refs)
our refcount is now zero
refcount_add(2) <---
warning here, refcount
unchanged
spin_lock(root->inode_lock)
radix_tree_delete()
With the generic refcounts, we actually warn again when process B above
tries to release his refcount because refcount_add() turned into a
no-op.
We saw this in production on older kernels without the asm optimized
refcounts.
The fix used here is to use refcount_inc_not_zero() to detect when the
object is in the middle of being freed and return NULL. This is almost
always the right answer anyway, since we usually end up pitching the
delayed_node if it didn't have fresh data in it.
This also changes __btrfs_release_delayed_node() to remove the extra
check for zero refcounts before radix tree deletion.
btrfs_get_delayed_node() was the only path that was allowing refcounts
to go from zero to one.
Fixes: 6de5f18e7b ("btrfs: fix refcount_t usage when deleting btrfs_delayed_node")
CC: <stable@vger.kernel.org> # 4.12+
Signed-off-by: Chris Mason <clm@fb.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The way we handle delalloc metadata reservations has gotten
progressively more complicated over the years. There is so much cruft
and weirdness around keeping the reserved count and outstanding counters
consistent and handling the error cases that it's impossible to
understand.
Fix this by making the delalloc block rsv per-inode. This way we can
calculate the actual size of the outstanding metadata reservations every
time we make a change, and then reserve the delta based on that amount.
This greatly simplifies the code everywhere, and makes the error
handling in btrfs_delalloc_reserve_metadata far less terrifying.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Our dir_context->pos is supposed to hold the next position we're
supposed to look. If we successfully insert a delayed dir index we
could end up with a duplicate entry because we don't increase ctx->pos
after doing the dir_emit.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
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 Sterba <dsterba@suse.com>
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 Sterba <dsterba@suse.com>
This function is internal to btrfs and doesn't really deal with any
VFS members, as such it needn't take a struct inode refrence but
btrfs_inode.
Signed-off-by: Nikolay Borisov <n.borisov.lkml@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently btrfs_ino takes a struct inode and this causes a lot of
internal btrfs functions which consume this ino to take a VFS inode,
rather than btrfs' own struct btrfs_inode. In order to fix this "leak"
of VFS structs into the internals of btrfs first it's necessary to
eliminate all uses of struct inode for the purpose of inode. This patch
does that by using BTRFS_I to convert an inode to btrfs_inode. With
this problem eliminated subsequent patches will start eliminating the
passing of struct inode altogether, eventually resulting in a lot cleaner
code.
Signed-off-by: Nikolay Borisov <n.borisov.lkml@gmail.com>
[ fix btrfs_get_extent tracepoint prototype ]
Signed-off-by: David Sterba <dsterba@suse.com>
This replaces ACCESS_ONCE macro with the corresponding
READ|WRITE macros
Signed-off-by: Seraphime Kirkovski <kirkseraph@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Problem statement: unprivileged user who has read-write access to more than
one btrfs subvolume may easily consume all kernel memory (eventually
triggering oom-killer).
Reproducer (./mkrmdir below essentially loops over mkdir/rmdir):
[root@kteam1 ~]# cat prep.sh
DEV=/dev/sdb
mkfs.btrfs -f $DEV
mount $DEV /mnt
for i in `seq 1 16`
do
mkdir /mnt/$i
btrfs subvolume create /mnt/SV_$i
ID=`btrfs subvolume list /mnt |grep "SV_$i$" |cut -d ' ' -f 2`
mount -t btrfs -o subvolid=$ID $DEV /mnt/$i
chmod a+rwx /mnt/$i
done
[root@kteam1 ~]# sh prep.sh
[maxim@kteam1 ~]$ for i in `seq 1 16`; do ./mkrmdir /mnt/$i 2000 2000 & done
[root@kteam1 ~]# for i in `seq 1 4`; do grep "kmalloc-128" /proc/slabinfo | grep -v dma; sleep 60; done
kmalloc-128 10144 10144 128 32 1 : tunables 0 0 0 : slabdata 317 317 0
kmalloc-128 9992352 9992352 128 32 1 : tunables 0 0 0 : slabdata 312261 312261 0
kmalloc-128 24226752 24226752 128 32 1 : tunables 0 0 0 : slabdata 757086 757086 0
kmalloc-128 42754240 42754240 128 32 1 : tunables 0 0 0 : slabdata 1336070 1336070 0
The huge numbers above come from insane number of async_work-s allocated
and queued by btrfs_wq_run_delayed_node.
The problem is caused by btrfs_wq_run_delayed_node() queuing more and more
works if the number of delayed items is above BTRFS_DELAYED_BACKGROUND. The
worker func (btrfs_async_run_delayed_root) processes at least
BTRFS_DELAYED_BATCH items (if they are present in the list). So, the machinery
works as expected while the list is almost empty. As soon as it is getting
bigger, worker func starts to process more than one item at a time, it takes
longer, and the chances to have async_works queued more than needed is getting
higher.
The problem above is worsened by another flaw of delayed-inode implementation:
if async_work was queued in a throttling branch (number of items >=
BTRFS_DELAYED_WRITEBACK), corresponding worker func won't quit until
the number of items < BTRFS_DELAYED_BACKGROUND / 2. So, it is possible that
the func occupies CPU infinitely (up to 30sec in my experiments): while the
func is trying to drain the list, the user activity may add more and more
items to the list.
The patch fixes both problems in straightforward way: refuse queuing too
many works in btrfs_wq_run_delayed_node and bail out of worker func if
at least BTRFS_DELAYED_WRITEBACK items are processed.
Changed in v2: remove support of thresh == NO_THRESHOLD.
Signed-off-by: Maxim Patlasov <mpatlasov@virtuozzo.com>
Signed-off-by: Chris Mason <clm@fb.com>
Cc: stable@vger.kernel.org # v3.15+
Now we only use the root parameter to print the root objectid in
a tracepoint. We can use the root parameter from the transaction
handle for that. It's also used to join the transaction with
async commits, so we remove the comment that it's just for checking.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There are loads of functions in btrfs that accept a root parameter
but only use it to obtain an fs_info pointer. Let's convert those to
just accept an fs_info pointer directly.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This results in btrfs_assert_delayed_root_empty and
btrfs_destroy_delayed_inode taking an fs_info instead of a root.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In routines where someptr->fs_info is referenced multiple times, we
introduce a convenience variable. This makes the code considerably
more readable.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
If we process the last item in the leaf and hit an I/O error while
reading the next leaf, we return -EIO without having adjusted the
position. Since we have emitted dirents, getdents() will return
the byte count to the user instead of the error. Subsequent callers
will emit the last successful dirent again, and return -EIO again,
with the same result. Callers loop forever.
Instead, if we always increment ctx->pos after emitting or skipping
the dirent, we'll be sure that we won't hit the same one again. When
we go to process the next leaf, we won't have emitted any dirents
and the -EIO will be returned to the user properly. We also don't
need to track if we've emitted a dirent already or if we've changed
the position yet.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
CodingStyle chapter 2:
"[...] never break user-visible strings such as printk messages,
because that breaks the ability to grep for them."
This patch unsplits user-visible strings.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We have a lot of random ints in btrfs_fs_info that can be put into flags. This
is mostly equivalent with the exception of how we deal with quota going on or
off, now instead we set a flag when we are turning it on or off and deal with
that appropriately, rather than just having a pending state that the current
quota_enabled gets set to. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
__btrfs_abort_transaction doesn't use its root parameter except to
obtain an fs_info pointer. We can obtain that from trans->root->fs_info
for now and from trans->fs_info in a later patch.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
BTRFS is using a variety of slab caches to satisfy internal needs.
Those slab caches are always allocated with the SLAB_RECLAIM_ACCOUNT,
meaning allocations from the caches are going to be accounted as
SReclaimable. At the same time btrfs is not registering any shrinkers
whatsoever, thus preventing memory from the slabs to be shrunk. This
means those caches are not in fact reclaimable.
To fix this remove the SLAB_RECLAIM_ACCOUNT on all caches apart from the
inode cache, since this one is being freed by the generic VFS super_block
shrinker. Also set the transaction related caches as SLAB_TEMPORARY,
to better document the lifetime of the objects (it just translates
to SLAB_RECLAIM_ACCOUNT).
Signed-off-by: Nikolay Borisov <n.borisov.lkml@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We reserve space for the inode update when we first reserve space for writing to
a file. However there are lots of ways that we can use this reservation and not
have it for subsequent ordered extents. Previously we'd fall through and try to
reserve metadata bytes for this, then we'd just steal the full reservation from
the delalloc_block_rsv, and if that didn't have enough space we'd steal the full
reservation from the global reserve. The problem with this is we can easily
just return ENOSPC and fallback to updating the inode item directly. In the
worst case (assuming 4k nodesize) we'd steal 64kib from the global reserve if we
fall all the way through, however if we just fallback and update the inode
directly we'd only steal 4k * BTRFS_PATH_MAX in the worst case which is 32kib.
We would have also just added the extent item for the inode so we likely will
have already cow'ed down most of the way to the leaf containing the inode item,
so we are more often than not only need one or two nodesize's worth of
reservations. Given the reservation for the extent itself is also a worst case
we will likely already have space to cover the inode update.
This change will make us behave better in the theoretical worst case, and much
better in the case that we don't have our reservation and cannot reserve more
metadata. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
So btrfs_block_rsv_migrate just unconditionally calls block_rsv_migrate_bytes.
Not only this but it unconditionally changes the size of the block_rsv. This
isn't a bug strictly speaking, but it makes truncate block rsv's look funny
because every time we migrate bytes over its size grows, even though we only
want it to be a specific size. So collapse this into one function that takes an
update_size argument and make truncate and evict not update the size for
consistency sake. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Commit fe742fd4f9 ("Revert "btrfs: switch to ->iterate_shared()"")
backed out the conversion to ->iterate_shared() for Btrfs because the
delayed inode handling in btrfs_real_readdir() is racy. However, we can
still do readdir in parallel if there are no delayed nodes.
This is a temporary fix which upgrades the shared inode lock to an
exclusive lock only when we have delayed items until we come up with a
more complete solution. While we're here, rename the
btrfs_{get,put}_delayed_items functions to make it very clear that
they're just for readdir.
Tested with xfstests and by doing a parallel kernel build:
while make tinyconfig && make -j4 && git clean dqfx; do
:
done
along with a bunch of parallel finds in another shell:
while true; do
for ((i=0; i<4; i++)); do
find . >/dev/null &
done
wait
done
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
Dont print warning for ENOSPC error unless ENOSPC_DEBUG is enabled. Use
btrfs_debug if it is enabled.
Signed-off-by: Ashish Samant <ashish.samant@oracle.com>
[ preserve the WARN_ON ]
Signed-off-by: David Sterba <dsterba@suse.com>
Cleanup.
kmem_cache_destroy has support NULL argument checking,
so drop the double null testing before calling it.
Signed-off-by: Kinglong Mee <kinglongmee@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The value of ctx->pos in the last readdir call is supposed to be set to
INT_MAX due to 32bit compatibility, unless 'pos' is intentially set to a
larger value, then it's LLONG_MAX.
There's a report from PaX SIZE_OVERFLOW plugin that "ctx->pos++"
overflows (https://forums.grsecurity.net/viewtopic.php?f=1&t=4284), on a
64bit arch, where the value is 0x7fffffffffffffff ie. LLONG_MAX before
the increment.
We can get to that situation like that:
* emit all regular readdir entries
* still in the same call to readdir, bump the last pos to INT_MAX
* next call to readdir will not emit any entries, but will reach the
bump code again, finds pos to be INT_MAX and sets it to LLONG_MAX
Normally this is not a problem, but if we call readdir again, we'll find
'pos' set to LLONG_MAX and the unconditional increment will overflow.
The report from Victor at
(http://thread.gmane.org/gmane.comp.file-systems.btrfs/49500) with debugging
print shows that pattern:
Overflow: e
Overflow: 7fffffff
Overflow: 7fffffffffffffff
PAX: size overflow detected in function btrfs_real_readdir
fs/btrfs/inode.c:5760 cicus.935_282 max, count: 9, decl: pos; num: 0;
context: dir_context;
CPU: 0 PID: 2630 Comm: polkitd Not tainted 4.2.3-grsec #1
Hardware name: Gigabyte Technology Co., Ltd. H81ND2H/H81ND2H, BIOS F3 08/11/2015
ffffffff81901608 0000000000000000 ffffffff819015e6 ffffc90004973d48
ffffffff81742f0f 0000000000000007 ffffffff81901608 ffffc90004973d78
ffffffff811cb706 0000000000000000 ffff8800d47359e0 ffffc90004973ed8
Call Trace:
[<ffffffff81742f0f>] dump_stack+0x4c/0x7f
[<ffffffff811cb706>] report_size_overflow+0x36/0x40
[<ffffffff812ef0bc>] btrfs_real_readdir+0x69c/0x6d0
[<ffffffff811dafc8>] iterate_dir+0xa8/0x150
[<ffffffff811e6d8d>] ? __fget_light+0x2d/0x70
[<ffffffff811dba3a>] SyS_getdents+0xba/0x1c0
Overflow: 1a
[<ffffffff811db070>] ? iterate_dir+0x150/0x150
[<ffffffff81749b69>] entry_SYSCALL_64_fastpath+0x12/0x83
The jump from 7fffffff to 7fffffffffffffff happens when new dir entries
are not yet synced and are processed from the delayed list. Then the code
could go to the bump section again even though it might not emit any new
dir entries from the delayed list.
The fix avoids entering the "bump" section again once we've finished
emitting the entries, both for synced and delayed entries.
References: https://forums.grsecurity.net/viewtopic.php?f=1&t=4284
Reported-by: Victor <services@swwu.com>
CC: stable@vger.kernel.org
Signed-off-by: David Sterba <dsterba@suse.com>
Tested-by: Holger Hoffstätte <holger.hoffstaette@googlemail.com>
Signed-off-by: Chris Mason <clm@fb.com>
It's slightly cleaner to zero-out the delayed node upon allocation
than to do it by hand in btrfs_init_delayed_node() for a few members
Signed-off-by: Alexandru Moise <00moses.alexander00@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We need to fill inode when we found a node for it in delayed_nodes_tree.
But we did not fill the ->last_trans currently, it will cause the test
of xfstest/generic/311 fail. Scenario of the 311 is shown as below:
Problem:
(1). test_fd = open(fname, O_RDWR|O_DIRECT)
(2). pwrite(test_fd, buf, 4096, 0)
(3). close(test_fd)
(4). drop_all_caches() <-------- "echo 3 > /proc/sys/vm/drop_caches"
(5). test_fd = open(fname, O_RDWR|O_DIRECT)
(6). fsync(test_fd);
<-------- we did not get the correct log entry for the file
Reason:
When we re-open this file in (5), we would find a node
in delayed_nodes_tree and fill the inode we are lookup with the
information. But the ->last_trans is not filled, then the fsync()
will check the ->last_trans and found it's 0 then say this inode
is already in our tree which is commited, not recording the extents
for it.
Fix:
This patch fill the ->last_trans properly and set the
runtime_flags if needed in this situation. Then we can get the
log entries we expected after (6) and generic/311 passed.
Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com>
Reviewed-by: Miao Xie <miaoxie@huawei.com>
Signed-off-by: Chris Mason <clm@fb.com>
This is the second independent patch of a larger project to cleanup
btrfs's internal usage of btrfs_root. Many functions take btrfs_root
only to grab the fs_info struct.
By requiring a root these functions cause programmer overhead. That
these functions can accept any valid root is not obvious until
inspection.
This patch reduces the specificity of such functions to accept the
fs_info directly.
These patches can be applied independently and thus are not being
submitted as a patch series. There should be about 26 patches by the
project's completion. Each patch will cleanup between 1 and 34 functions
apiece. Each patch covers a single file's functions.
This patch affects the following function(s):
1) btrfs_wq_run_delayed_node
Signed-off-by: Daniel Dressler <danieru.dressler@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.cz>
This patch adds a new member to the 'struct btrfs_inode' structure to hold
the file creation time.
Signed-off-by: chandan <chandanrmail@gmail.com>
[refreshed, removed btrfs_inode_otime]
Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
Commit 1d52c78afb (Btrfs: try not to ENOSPC on log replay) added a
check to skip delayed inode updates during log replay because it
confuses the enospc code. But the delayed processing will end up
ignoring delayed refs from log replay because the inode itself wasn't
put through the delayed code.
This can end up triggering a warning at commit time:
WARNING: CPU: 2 PID: 778 at fs/btrfs/delayed-inode.c:1410 btrfs_assert_delayed_root_empty+0x32/0x34()
Which is repeated for each commit because we never process the delayed
inode ref update.
The fix used here is to change btrfs_delayed_delete_inode_ref to return
an error if we're currently in log replay. The caller will do the ref
deletion immediately and everything will work properly.
Signed-off-by: Chris Mason <clm@fb.com>
cc: stable@vger.kernel.org # v3.18 and any stable series that picked 1d52c78afb
btrfs_set_key_type and btrfs_key_type are used inconsistently along with
open coded variants. Other members of btrfs_key are accessed directly
without any helpers anyway.
Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
This has been reported and discussed for a long time, and this hang occurs in
both 3.15 and 3.16.
Btrfs now migrates to use kernel workqueue, but it introduces this hang problem.
Btrfs has a kind of work queued as an ordered way, which means that its
ordered_func() must be processed in the way of FIFO, so it usually looks like --
normal_work_helper(arg)
work = container_of(arg, struct btrfs_work, normal_work);
work->func() <---- (we name it work X)
for ordered_work in wq->ordered_list
ordered_work->ordered_func()
ordered_work->ordered_free()
The hang is a rare case, first when we find free space, we get an uncached block
group, then we go to read its free space cache inode for free space information,
so it will
file a readahead request
btrfs_readpages()
for page that is not in page cache
__do_readpage()
submit_extent_page()
btrfs_submit_bio_hook()
btrfs_bio_wq_end_io()
submit_bio()
end_workqueue_bio() <--(ret by the 1st endio)
queue a work(named work Y) for the 2nd
also the real endio()
So the hang occurs when work Y's work_struct and work X's work_struct happens
to share the same address.
A bit more explanation,
A,B,C -- struct btrfs_work
arg -- struct work_struct
kthread:
worker_thread()
pick up a work_struct from @worklist
process_one_work(arg)
worker->current_work = arg; <-- arg is A->normal_work
worker->current_func(arg)
normal_work_helper(arg)
A = container_of(arg, struct btrfs_work, normal_work);
A->func()
A->ordered_func()
A->ordered_free() <-- A gets freed
B->ordered_func()
submit_compressed_extents()
find_free_extent()
load_free_space_inode()
... <-- (the above readhead stack)
end_workqueue_bio()
btrfs_queue_work(work C)
B->ordered_free()
As if work A has a high priority in wq->ordered_list and there are more ordered
works queued after it, such as B->ordered_func(), its memory could have been
freed before normal_work_helper() returns, which means that kernel workqueue
code worker_thread() still has worker->current_work pointer to be work
A->normal_work's, ie. arg's address.
Meanwhile, work C is allocated after work A is freed, work C->normal_work
and work A->normal_work are likely to share the same address(I confirmed this
with ftrace output, so I'm not just guessing, it's rare though).
When another kthread picks up work C->normal_work to process, and finds our
kthread is processing it(see find_worker_executing_work()), it'll think
work C as a collision and skip then, which ends up nobody processing work C.
So the situation is that our kthread is waiting forever on work C.
Besides, there're other cases that can lead to deadlock, but the real problem
is that all btrfs workqueue shares one work->func, -- normal_work_helper,
so this makes each workqueue to have its own helper function, but only a
wraper pf normal_work_helper.
With this patch, I no long hit the above hang.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
On heavy workloads, we're seeing soft lockup warnings on
root->inode_lock in __btrfs_release_delayed_node. The low hanging fruit
is to reduce the size of the critical section.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
