Skip to contentSkip to conversation

vfs: do not try to evict inode when super is frozen

PATCH NAK Jaegeuk Kim 2 participants 5 messages Mar 4, 2022 VFS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 vfs: do not try to evict inode when super is frozen

Otherwise, we will get a deadlock.

[freeze test] shrinkder
freeze_super
- pwercpu_down_write(SB_FREEZE_FS)
- super_cache_scan
- down_read(&sb->s_umount)
- prune_icache_sb
- dispose_list
- evict
- f2fs_evict_inode
thaw_super
- down_write(&sb->s_umount);
- __percpu_down_read(SB_FREEZE_FS)

---
 fs/super.c | 6 ++++++
 1 file changed, 6 insertions(+)
fs/super.c +6 -0
index 7af820ba5ad5..f7303d91f874 100644
@@ -79,6 +79,12 @@ static unsigned long super_cache_scan(struct shrinker *shrink,
 	if (!trylock_super(sb))
 		return SHRINK_STOP;
 
+	/* This prevents inode eviction that requires SB_FREEZE_FS. */
+	if (sb->s_writers.frozen == SB_FREEZE_FS) {
+		up_read(&sb->s_umount);
+		return SHRINK_STOP;
+	}
+
 	if (sb->s_op->nr_cached_objects)
 		fs_objects = sb->s_op->nr_cached_objects(sb, sc);
 
Signature · 1 line
2.35.1.616.g0bdcbb4464-goog
#1 lore raw
Quoted 1 line · Jaegeuk Kim

Otherwise, we will get a deadlock.

NACK.

We have to be able to evict clean inodes from memory on frozen inodes because we can still instantiate inodes while the filesytem is frozen. e.g. there's a find running when the filesystem is frozen. What happens if we can't evict clean cached inodes from memory when we run out of memory trying to instantiate new inodes?

Quoted 13 lines

[freeze test] shrinkder
freeze_super
- pwercpu_down_write(SB_FREEZE_FS)
- super_cache_scan
- down_read(&sb->s_umount)
- prune_icache_sb
- dispose_list
- evict
- f2fs_evict_inode
thaw_super
- down_write(&sb->s_umount);
- __percpu_down_read(SB_FREEZE_FS)

That seems like a f2fs bug, not a generic problem.

Filesystems already have to handle stuff like this if an unlinked file is closed while the fs is frozen - we have to handle inode eviction needing to modify the file, and different filesystems handle this differently. Most filesystems simply block in ->evict_inode in this case, but this never occurs from the shrinker context.

IOWs, the shrinker should never be evicting inodes that require the filesystem to immediately block on frozen filesystems. If you have such inodes in cache at the time the filesystem is frozen, then they should be purged from the cache as part of the freeze process so the shrinker won't ever find inodes that it could deadlock on.

Cheers,

Dave.

Signature · 2 lines
Dave Chinner
[email protected]
#2 lore raw
Quoted 10 lines · Dave Chinner
Quoted 1 line · Jaegeuk Kim

Otherwise, we will get a deadlock.

NACK.

We have to be able to evict clean inodes from memory on frozen inodes because we can still instantiate inodes while the filesytem is frozen. e.g. there's a find running when the filesystem is frozen. What happens if we can't evict clean cached inodes from memory when we run out of memory trying to instantiate new inodes?

Ok, that makes sense.

Quoted 24 lines
Quoted 14 lines

[freeze test] shrinkder
freeze_super
- pwercpu_down_write(SB_FREEZE_FS)
- super_cache_scan
- down_read(&sb->s_umount)
- prune_icache_sb
- dispose_list
- evict
- f2fs_evict_inode
thaw_super
- down_write(&sb->s_umount);
- __percpu_down_read(SB_FREEZE_FS)

That seems like a f2fs bug, not a generic problem.

Filesystems already have to handle stuff like this if an unlinked file is closed while the fs is frozen - we have to handle inode eviction needing to modify the file, and different filesystems handle this differently. Most filesystems simply block in ->evict_inode in this case, but this never occurs from the shrinker context.

Quoted 5 lines

IOWs, the shrinker should never be evicting inodes that require the filesystem to immediately block on frozen filesystems. If you have such inodes in cache at the time the filesystem is frozen, then they should be purged from the cache as part of the freeze process so the shrinker won't ever find inodes that it could deadlock on.

If so, is this a bug in drop_caches_sysctl_handler? Or, I shouldn't have used "echo 3 > sysfs/drop_caches" with freezefs in xfstests?

Quoted 7 lines

Cheers,

Dave.

Signature · 2 lines
Dave Chinner
[email protected]
#3 lore raw
Quoted 38 lines · Jaegeuk Kim
Quoted 10 lines · Dave Chinner
Quoted 1 line · Jaegeuk Kim

Otherwise, we will get a deadlock.

NACK.

We have to be able to evict clean inodes from memory on frozen inodes because we can still instantiate inodes while the filesytem is frozen. e.g. there's a find running when the filesystem is frozen. What happens if we can't evict clean cached inodes from memory when we run out of memory trying to instantiate new inodes?

Ok, that makes sense.

Quoted 24 lines
Quoted 14 lines

[freeze test] shrinkder
freeze_super
- pwercpu_down_write(SB_FREEZE_FS)
- super_cache_scan
- down_read(&sb->s_umount)
- prune_icache_sb
- dispose_list
- evict
- f2fs_evict_inode
thaw_super
- down_write(&sb->s_umount);
- __percpu_down_read(SB_FREEZE_FS)

That seems like a f2fs bug, not a generic problem.

Filesystems already have to handle stuff like this if an unlinked file is closed while the fs is frozen - we have to handle inode eviction needing to modify the file, and different filesystems handle this differently. Most filesystems simply block in ->evict_inode in this case, but this never occurs from the shrinker context.

Quoted 8 lines
Quoted 5 lines

IOWs, the shrinker should never be evicting inodes that require the filesystem to immediately block on frozen filesystems. If you have such inodes in cache at the time the filesystem is frozen, then they should be purged from the cache as part of the freeze process so the shrinker won't ever find inodes that it could deadlock on.

If so, is this a bug in drop_caches_sysctl_handler? Or, I shouldn't have used "echo 3 > sysfs/drop_caches" with freezefs in xfstests?

My bad. I totally misunderstood. I'm testing a patch to call evict_inodes() in f2fs_freeze(). Thank you for the comment. :)

Quoted 8 lines
Quoted 8 lines

Cheers,

Dave.

Signature · 2 lines
Dave Chinner
[email protected]
#4 lore raw
In reply to Jaegeuk Kim
Quoted 38 lines · Jaegeuk Kim
Quoted 10 lines · Dave Chinner
Quoted 1 line · Jaegeuk Kim

Otherwise, we will get a deadlock.

NACK.

We have to be able to evict clean inodes from memory on frozen inodes because we can still instantiate inodes while the filesytem is frozen. e.g. there's a find running when the filesystem is frozen. What happens if we can't evict clean cached inodes from memory when we run out of memory trying to instantiate new inodes?

Ok, that makes sense.

Quoted 24 lines
Quoted 14 lines

[freeze test] shrinkder
freeze_super
- pwercpu_down_write(SB_FREEZE_FS)
- super_cache_scan
- down_read(&sb->s_umount)
- prune_icache_sb
- dispose_list
- evict
- f2fs_evict_inode
thaw_super
- down_write(&sb->s_umount);
- __percpu_down_read(SB_FREEZE_FS)

That seems like a f2fs bug, not a generic problem.

Filesystems already have to handle stuff like this if an unlinked file is closed while the fs is frozen - we have to handle inode eviction needing to modify the file, and different filesystems handle this differently. Most filesystems simply block in ->evict_inode in this case, but this never occurs from the shrinker context.

Quoted 7 lines
Quoted 5 lines

IOWs, the shrinker should never be evicting inodes that require the filesystem to immediately block on frozen filesystems. If you have such inodes in cache at the time the filesystem is frozen, then they should be purged from the cache as part of the freeze process so the shrinker won't ever find inodes that it could deadlock on.

If so, is this a bug in drop_caches_sysctl_handler?

IMO, no.

Quoted 2 lines

Or, I shouldn't have
used "echo 3 > sysfs/drop_caches" with freezefs in xfstests?

That should just work.

As I said above - if the filesystem cannot process eviction of certain types of inodes when the filesystem is frozen, it needs to take steps during the freeze process to ensure they can't be evicted when the fs is frozen....

Cheers,

Dave.

Signature · 2 lines
Dave Chinner
[email protected]