btrfs: fortification for GFP_NOFS allocations
cover btrfs: fortification for GFP_NOFS allocations
Hi,
these two patches were sent as a part of a larger RFC which aims at allowing GFP_NOFS allocations to fail to help sort out memory reclaim issues bound to the current behavior
(http://marc.info/?l=linux-mm&m=143876830616538&w=2).
It is clear that move to the GFP_NOFS behavior change is a long term plan but these patches should be good enough even with that change in place. It also seems that Chris wasn't opposed and would be willing to take them http://marc.info/?l=linux-mm&m=143991792427165&w=2 so here we come. I have rephrased the changeslogs to not refer to the patch which changes the NOFS behavior.
Just to clarify. These two patches allowed my particular testcase (mentioned in the cover referenced above) to survive it doesn't mean that the failing GFP_NOFS are OK now. I have seen some other places where GFP_NOFS allocation is followed by BUG_ON(ALLOC_FAILED). I have not encountered them though.
Let me know if you would prefer other changes.
PATCH 1/2 btrfs: Prevent from early transaction abort
From: Michal Hocko <[email protected]>
Btrfs relies on GFP_NOFS allocation when committing the transaction but this allocation context is rather weak wrt. reclaim capabilities. The page allocator currently tries hard to not fail these allocations if they are small (<=PAGE_ALLOC_COSTLY_ORDER) so this is not a problem currently but there is an attempt to move away from the default no-fail behavior and allow these allocation to fail more eagerly. And this would lead to a pre-mature transaction abort as follows:
[ 55.328093] Call Trace: [ 55.328890] [<ffffffff8154e6f0>] dump_stack+0x4f/0x7b [ 55.330518] [<ffffffff8108fa28>] ? console_unlock+0x334/0x363 [ 55.332738] [<ffffffff8110873e>] __alloc_pages_nodemask+0x81d/0x8d4 [ 55.334910] [<ffffffff81100752>] pagecache_get_page+0x10e/0x20c [ 55.336844] [<ffffffffa007d916>] alloc_extent_buffer+0xd0/0x350 [btrfs] [ 55.338973] [<ffffffffa0059d8c>] btrfs_find_create_tree_block+0x15/0x17 [btrfs] [ 55.341329] [<ffffffffa004f728>] btrfs_alloc_tree_block+0x18c/0x405 [btrfs] [ 55.343566] [<ffffffffa003fa34>] split_leaf+0x1e4/0x6a6 [btrfs] [ 55.345577] [<ffffffffa0040567>] btrfs_search_slot+0x671/0x831 [btrfs] [ 55.347679] [<ffffffff810682d7>] ? get_parent_ip+0xe/0x3e [ 55.349434] [<ffffffffa0041cb2>] btrfs_insert_empty_items+0x5d/0xa8 [btrfs] [ 55.351681] [<ffffffffa004ecfb>] __btrfs_run_delayed_refs+0x7a6/0xf35 [btrfs] [ 55.353979] [<ffffffffa00512ea>] btrfs_run_delayed_refs+0x6e/0x226 [btrfs] [ 55.356212] [<ffffffffa0060e21>] ? start_transaction+0x192/0x534 [btrfs] [ 55.358378] [<ffffffffa0060e21>] ? start_transaction+0x192/0x534 [btrfs] [ 55.360626] [<ffffffffa0060221>] btrfs_commit_transaction+0x4c/0xaba [btrfs] [ 55.362894] [<ffffffffa0060e21>] ? start_transaction+0x192/0x534 [btrfs] [ 55.365221] [<ffffffffa0073428>] btrfs_sync_file+0x29c/0x310 [btrfs] [ 55.367273] [<ffffffff81186808>] vfs_fsync_range+0x8f/0x9e [ 55.369047] [<ffffffff81186833>] vfs_fsync+0x1c/0x1e [ 55.370654] [<ffffffff81186869>] do_fsync+0x34/0x4e [ 55.372246] [<ffffffff81186ab3>] SyS_fsync+0x10/0x14 [ 55.373851] [<ffffffff81554f97>] system_call_fastpath+0x12/0x6f [ 55.381070] BTRFS: error (device hdb1) in btrfs_run_delayed_refs:2821: errno=-12 Out of memory [ 55.382431] BTRFS warning (device hdb1): Skipping commit of aborted transaction. [ 55.382433] BTRFS warning (device hdb1): cleanup_transaction:1692: Aborting unused transaction(IO failure). [ 55.384280] ------------[ cut here ]------------ [ 55.384312] WARNING: CPU: 0 PID: 3010 at fs/btrfs/delayed-ref.c:438 btrfs_select_ref_head+0xd9/0xfe [btrfs]() [...] [ 55.384337] Call Trace: [ 55.384353] [<ffffffff8154e6f0>] dump_stack+0x4f/0x7b [ 55.384357] [<ffffffff8107f717>] ? down_trylock+0x2d/0x37 [ 55.384359] [<ffffffff81046977>] warn_slowpath_common+0xa1/0xbb [ 55.384398] [<ffffffffa00a1d6b>] ? btrfs_select_ref_head+0xd9/0xfe [btrfs] [ 55.384400] [<ffffffff81046a34>] warn_slowpath_null+0x1a/0x1c [ 55.384423] [<ffffffffa00a1d6b>] btrfs_select_ref_head+0xd9/0xfe [btrfs] [ 55.384446] [<ffffffffa004e5f7>] ? __btrfs_run_delayed_refs+0xa2/0xf35 [btrfs] [ 55.384455] [<ffffffffa004e600>] __btrfs_run_delayed_refs+0xab/0xf35 [btrfs] [ 55.384476] [<ffffffffa00512ea>] btrfs_run_delayed_refs+0x6e/0x226 [btrfs] [ 55.384499] [<ffffffffa0060e21>] ? start_transaction+0x192/0x534 [btrfs] [ 55.384521] [<ffffffffa0060e21>] ? start_transaction+0x192/0x534 [btrfs] [ 55.384543] [<ffffffffa0060221>] btrfs_commit_transaction+0x4c/0xaba [btrfs] [ 55.384565] [<ffffffffa0060e21>] ? start_transaction+0x192/0x534 [btrfs] [ 55.384588] [<ffffffffa0073428>] btrfs_sync_file+0x29c/0x310 [btrfs] [ 55.384591] [<ffffffff81186808>] vfs_fsync_range+0x8f/0x9e [ 55.384592] [<ffffffff81186833>] vfs_fsync+0x1c/0x1e [ 55.384593] [<ffffffff81186869>] do_fsync+0x34/0x4e [ 55.384594] [<ffffffff81186ab3>] SyS_fsync+0x10/0x14 [ 55.384595] [<ffffffff81554f97>] system_call_fastpath+0x12/0x6f [...] [ 55.384608] ---[ end trace c29799da1d4dd621 ]--- [ 55.437323] BTRFS info (device hdb1): forced readonly [ 55.438815] BTRFS info (device hdb1): delayed_refs has NO entry
Fix this by being explicit about the no-fail behavior of this allocation path and use __GFP_NOFAIL.
--- fs/btrfs/extent_io.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-)
fs/btrfs/extent_io.c +2 -4
@@ -4607,9 +4607,7 @@ __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, {struct extent_buffer *eb =NULL ; - eb =kmem_cache_zalloc (extent_buffer_cache,GFP_NOFS ); -if (eb ==NULL ) -return NULL ; + eb =kmem_cache_zalloc (extent_buffer_cache,GFP_NOFS |__GFP_NOFAIL); eb->start = start; eb->len = len; eb->fs_info = fs_info;
@@ -4867,7 +4865,7 @@ struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,return NULL ;for (i =0 ; i < num_pages; i++, index++) { - p =find_or_create_page (mapping, index,GFP_NOFS ); + p =find_or_create_page (mapping, index,GFP_NOFS |__GFP_NOFAIL);if (!p) goto free_eb;
Signature · 1 line
2.5.0
PATCH 2/2 btrfs: use __GFP_NOFAIL in alloc_btrfs_bio
In reply to [email protected]From: Michal Hocko <[email protected]>
alloc_btrfs_bio relies on GFP_NOFS allocation when committing the transaction but this allocation context is rather weak wrt. reclaim capabilities. The page allocator currently tries hard to not fail these allocations if they are small (<=PAGE_ALLOC_COSTLY_ORDER) but it can still fail if the _current_ process is the OOM killer victim. Moreover there is an attempt to move away from the default no-fail behavior and allow these allocation to fail more eagerly. This would lead to:
[ 37.928625] kernel BUG at fs/btrfs/extent_io.c:4045
which is clearly undesirable and the nofail behavior should be explicit if the allocation failure cannot be tolerated.
--- fs/btrfs/volumes.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
fs/btrfs/volumes.c +1 -3
@@ -4914,9 +4914,7 @@ static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)* and the stripes */ sizeof (u64) * (total_stripes), - GFP_NOFS); -if (!bbio) -return NULL ; + GFP_NOFS|__GFP_NOFAIL);atomic_set (&bbio->error,0 );atomic_set (&bbio->refs,1 );
Signature · 1 line
2.5.0
Quoted 13 lines · On Wed, Aug 19, 2015 at 02:17:39PM +0200, [email protected] wrote:
Hi,
these two patches were sent as a part of a larger RFC which aims at allowingGFP_NOFSallocations to fail to help sort out memory reclaim issues bound to the current behavior
(http://marc.info/?l=linux-mm&m=143876830616538&w=2).It is clear that move to the
GFP_NOFSbehavior change is a long term plan but these patches should be good enough even with that change in place. It also seems that Chris wasn't opposed and would be willing to take them http://marc.info/?l=linux-mm&m=143991792427165&w=2 so here we come. I have rephrased the changeslogs to not refer to the patch which changes the NOFS behavior.
Quoted 7 lines
Just to clarify. These two patches allowed my particular testcase (mentioned in the cover referenced above) to survive it doesn't mean that the failing
GFP_NOFSare OK now. I have seen some other places whereGFP_NOFSallocation is followed byBUG_ON(ALLOC_FAILED). I have not encountered them though.Let me know if you would prefer other changes.
My plan is to start with these two and take more as required.
-chris
Quoted 20 lines · Chris Mason
Quoted 19 lines · On Wed, Aug 19, 2015 at 02:17:39PM +0200, [email protected] wrote:
Hi,
these two patches were sent as a part of a larger RFC which aims at allowingGFP_NOFSallocations to fail to help sort out memory reclaim issues bound to the current behavior
(http://marc.info/?l=linux-mm&m=143876830616538&w=2).It is clear that move to the
GFP_NOFSbehavior change is a long term plan but these patches should be good enough even with that change in place. It also seems that Chris wasn't opposed and would be willing to take them http://marc.info/?l=linux-mm&m=143991792427165&w=2 so here we come. I have rephrased the changeslogs to not refer to the patch which changes the NOFS behavior.Just to clarify. These two patches allowed my particular testcase (mentioned in the cover referenced above) to survive it doesn't mean that the failing
GFP_NOFSare OK now. I have seen some other places whereGFP_NOFSallocation is followed byBUG_ON(ALLOC_FAILED). I have not encountered them though.
Quoted 3 lines
Quoted 1 line
Let me know if you would prefer other changes.
My plan is to start with these two and take more as required.
I've previously noticed in __set_extent_bit() things like:
if (!prealloc && (mask & __GFP_WAIT)) {
prealloc = alloc_extent_state(mask);
BUG_ON(!prealloc);
}and later:
prealloc = alloc_extent_state_atomic(prealloc);
BUG_ON(!prealloc);which internally does:
if (!prealloc)
prealloc = alloc_extent_state(GFP_ATOMIC);The first one could be fixable by adding __GFP_NOFAIL. In fact we've got an internal bug report for that one already. Even without GFP_NOFS being allowed to fail, allocation can already fail when the thread is marked for oom kill, which is likely what happened in that case.
The second case is problematic though, because GFP_ATOMIC | __GFP_NOFAIL is not allowed. GFP_ATOMIC will give you access to memory reserves, which reduces the chance of hitting the BUG_ON(), but it's not a bulletproof solution.
Quoted 7 lines
-chris
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Quoted 32 lines · Vlastimil Babka
Quoted 20 lines · Chris Mason
Quoted 19 lines · On Wed, Aug 19, 2015 at 02:17:39PM +0200, [email protected] wrote:
Hi,
these two patches were sent as a part of a larger RFC which aims at allowingGFP_NOFSallocations to fail to help sort out memory reclaim issues bound to the current behavior
(http://marc.info/?l=linux-mm&m=143876830616538&w=2).It is clear that move to the
GFP_NOFSbehavior change is a long term plan but these patches should be good enough even with that change in place. It also seems that Chris wasn't opposed and would be willing to take them http://marc.info/?l=linux-mm&m=143991792427165&w=2 so here we come. I have rephrased the changeslogs to not refer to the patch which changes the NOFS behavior.Just to clarify. These two patches allowed my particular testcase (mentioned in the cover referenced above) to survive it doesn't mean that the failing
GFP_NOFSare OK now. I have seen some other places whereGFP_NOFSallocation is followed byBUG_ON(ALLOC_FAILED). I have not encountered them though.Quoted 3 lines
Quoted 1 line
Let me know if you would prefer other changes.
My plan is to start with these two and take more as required.
I've previously noticed in
__set_extent_bit()things like:if (!prealloc && (mask & __GFP_WAIT)) { prealloc = alloc_extent_state(mask); BUG_ON(!prealloc); }
Quoted 4 lines
and later:
prealloc = alloc_extent_state_atomic(prealloc); BUG_ON(!prealloc);
Yes. I have noticed also many other places:
$ git grep "BUG_ON.*ENOMEM" -- fs/btrfs/ | wc -l 47
I have talked to David Sterba and he said this is on his todo list. So this will likely take some more time but it is definitely good to sort out.
Signature · 2 lines
Michal Hocko SUSE Labs