btrfs: Correctly free extent buffer in case btree_read_extent_buffer_pages fails
PATCH 1/1 btrfs: Correctly free extent buffer in case btree_read_extent_buffer_pages fails
If a an eb fails to be read for whatever reason - it's corrupted on disk and parent transid/key validations fail or IO for eb pages fail then this buffer must be removed from the buffer cache. Currently the code calls free_extent_buffer if an error occurs. Unfortunately this doesn't achieve the desired behavior since btrfs_find_create_tree_block returns with eb->refs == 2. On the other hand free_extent_buffer will only decrement the refs once leavin it added to the buffer cache radix tree. This enables later code to look up the buffer from the cache and utilize it potentially leading to a crash.
The correct way to free the buffer is call free_extent_buffer_stale. This function will correctly call atomic_dec explicitly for the buffer and subsequently call release_extent_buffer which will decrement the final reference thus correctly remove the invalid buffer from buffer cache. This change affects only newly allocated buffers since they have eb->refs == 2.
---
This patch was inspired by Qu's "btrfs: Check the first key and level for cached extent buffer". Though fixing Qu's crash is only a side effect, what I was aiming for is "correct behavior" - in this case immediately remove an eb from the eb cache if it's detected broken. This, however, doesn't eliminate the need for Qu's patch which adds the parent check in read_block_for_search.
I have validated it it using the image from the bugzilla issue and reading /MOUNT/foo/bar/stress/f7. Without my patch (or Qu's) it crashes with either of them I get:
cat: scratch/foo/bar/stress/f7: Input/output error
This also survived a full xfstest run with no regressions.
fs/btrfs/disk-io.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
fs/btrfs/disk-io.c +3 -3
@@ -1043,12 +1043,12 @@ int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr, ret =read_extent_buffer_pages (io_tree, buf,WAIT_PAGE_LOCK , mirror_num);if (ret) { -free_extent_buffer (buf); +free_extent_buffer_stale (buf);return ret; }if (test_bit (EXTENT_BUFFER_CORRUPT , &buf->bflags)) { -free_extent_buffer (buf); +free_extent_buffer_stale (buf);return -EIO ; }else if (extent_buffer_uptodate (buf)) { *eb = buf;
@@ -1102,7 +1102,7 @@ struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, ret =btree_read_extent_buffer_pages (fs_info, buf, parent_transid, level, first_key);if (ret) { -free_extent_buffer (buf); +free_extent_buffer_stale (buf);return ERR_PTR (ret); }return buf;
Signature · 1 line
2.17.1
Quoted 10 lines · Nikolay Borisov
If a an eb fails to be read for whatever reason - it's corrupted on disk and parent transid/key validations fail or IO for eb pages fail then this buffer must be removed from the buffer cache. Currently the code calls
free_extent_bufferif an error occurs. Unfortunately this doesn't achieve the desired behavior sincebtrfs_find_create_tree_blockreturns witheb->refs== 2. On the other handfree_extent_bufferwill only decrement the refs once leavin it added to the buffer cache radix tree. This enables later code to look up the buffer from the cache and utilize it potentially leading to a crash.
Quoted 6 lines
The correct way to free the buffer is call
free_extent_buffer_stale. This function will correctly callatomic_decexplicitly for the buffer and subsequently callrelease_extent_bufferwhich will decrement the final reference thus correctly remove the invalid buffer from buffer cache. This change affects only newly allocated buffers since they haveeb->refs== 2.
Following the same pattern, should every use of btrfs_find_create_tree_block be followed by free_extent_buffer_stale in case of errors?
There's almost the same sequence in readahead_tree_block as you update in reada_tree_block_flagged:
btrfs_find_create_tree_block
read_extent_buffer_pages
free_extent_buffer
so should be the _stale variant used there too?
free_extent_buffer_stale is normally used in other contexts where the 'stale' part means that the eb is known to be disconnected.
I think that btrfs_find_create_tree_block needs a dedicated freeing function that will make sure all invariants (refs == 2, no other structures connected) are met and then do effectively the same what _stale does now.
Quoted 22 lines · David Sterba
Quoted 10 lines · Nikolay Borisov
If a an eb fails to be read for whatever reason - it's corrupted on disk and parent transid/key validations fail or IO for eb pages fail then this buffer must be removed from the buffer cache. Currently the code calls
free_extent_bufferif an error occurs. Unfortunately this doesn't achieve the desired behavior sincebtrfs_find_create_tree_blockreturns witheb->refs== 2. On the other handfree_extent_bufferwill only decrement the refs once leavin it added to the buffer cache radix tree. This enables later code to look up the buffer from the cache and utilize it potentially leading to a crash.Quoted 6 lines
The correct way to free the buffer is call
free_extent_buffer_stale. This function will correctly callatomic_decexplicitly for the buffer and subsequently callrelease_extent_bufferwhich will decrement the final reference thus correctly remove the invalid buffer from buffer cache. This change affects only newly allocated buffers since they haveeb->refs== 2.Following the same pattern, should every use of
btrfs_find_create_tree_blockbe followed byfree_extent_buffer_stalein case of errors?
Quoted 8 lines
There's almost the same sequence in
readahead_tree_blockas you update inreada_tree_block_flagged:
btrfs_find_create_tree_block
read_extent_buffer_pages
free_extent_bufferso should be the
_stalevariant used there too?
You are right it should also be using the _stale variant. I will fix it.
Despite my cleanup of the extent buffer freeing code I'm still not entirely happy with the special cases we have with the stale or unmapped logic. That's why we need all these special cases ;\
Quoted 8 lines
free_extent_buffer_staleis normally used in other contexts where the 'stale' part means that the eb is known to be disconnected.I think that
btrfs_find_create_tree_blockneeds a dedicated freeing function that will make sure all invariants (refs == 2, no other structures connected) are met and then do effectively the same what_staledoes now.
Sounds good though I'm a bit reluctant to add yet another free_extent_buffer variant. IMO the logic of the extent buffer radix tree need to be revised which could lead to simplifications in the whole ref counting code.
Quoted 1 line
Quoted 30 lines · Nikolay Borisov
Quoted 11 lines
Quoted 6 lines
The correct way to free the buffer is call
free_extent_buffer_stale. This function will correctly callatomic_decexplicitly for the buffer and subsequently callrelease_extent_bufferwhich will decrement the final reference thus correctly remove the invalid buffer from buffer cache. This change affects only newly allocated buffers since they haveeb->refs== 2.Following the same pattern, should every use of
btrfs_find_create_tree_blockbe followed byfree_extent_buffer_stalein case of errors?Quoted 8 lines
There's almost the same sequence in
readahead_tree_blockas you update inreada_tree_block_flagged:
btrfs_find_create_tree_block
read_extent_buffer_pages
free_extent_bufferso should be the
_stalevariant used there too?You are right it should also be using the
_stalevariant. I will fix it.Despite my cleanup of the extent buffer freeing code I'm still not entirely happy with the special cases we have with the stale or unmapped logic. That's why we need all these special cases ;\
Quoted 4 lines
free_extent_buffer_staleis normally used in other contexts where the 'stale' part means that the eb is known to be disconnected.
Quoted 8 lines
Quoted 4 lines
I think that
btrfs_find_create_tree_blockneeds a dedicated freeing function that will make sure all invariants (refs == 2, no other structures connected) are met and then do effectively the same what_staledoes now.Sounds good though I'm a bit reluctant to add yet another
free_extent_buffervariant. IMO the logic of the extent buffer radix tree need to be revised which could lead to simplifications in the whole ref counting code.
Revised maybe, but this is the core of the MM <-> FS interaction and must be correct, even after the simplification. There's RCU in the mix, non-standard refcounting, page tracking. This won't be easy. Adding a admitedly why-yet-another special freeing helper would at least stop abusing the API and annotate the context. That's an intermediate step.