Skip to contentSkip to conversation

btrfs: fix bytes_may_use underflow after a failed tree block allocation

PATCH In review Vlad Poenaru 2 participants 2 messages Nov 18

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 btrfs: fix bytes_may_use underflow after a failed tree block allocation

When btrfs_alloc_tree_block() fails after btrfs_reserve_extent() has succeeded, its error path gives the block back to the block reserve, although the block has already left bytes_may_use. Each such failure leaves bytes_may_use a block short, and it underflows and warns in space-info.h when the global block reserve is released at unmount, or earlier.

btrfs_alloc_tree_block()

  |- btrfs_use_block_rsv()
  |- btrfs_reserve_extent()
  |  \- btrfs_add_reserved_bytes()  <- bytes_may_use -= blocksize
  |- btrfs_init_new_buffer()
  |  \- alloc_extent_buffer()       <- fails with -ENOMEM
  |- btrfs_free_reserved_extent()   <- bytes_reserved -= blocksize

\- btrfs_unuse_block_rsv() <- block rsv += blocksize again

Since commit 19d7f65f032f ("btrfs: convert the buffer_radix to an xarray") in 6.16, alloc_extent_buffer() can fail this way under memory cgroup pressure, because fs_info->buffer_tree is memcg-accounted.

Return without giving the block back once btrfs_reserve_extent() has succeeded.

Fixes 67b7859e9bfa ("btrfs: handle ENOMEM in btrfs_alloc_tree_block")

---
v2:
- Shorter changelog, built around a call trace (Qu Wenruo).
- Dropped the code comment (Boris Burkov).

The code is v1's apart from the removed comment, so v1's testing stands:
its reproducer leaves bytes_may_use below the global reserve without the fix and not with it, and fstests btrfs/002, btrfs/220, generic/475 and generic/476 pass.

v1, with the reproducer:
https://lore.kernel.org/all/[email protected]/

 fs/btrfs/extent-tree.c | 1 +
 1 file changed, 1 insertion(+)
fs/btrfs/extent-tree.c +1 -0
index 9b0a5c6b96f3c..8d4c66bf850bc 100644
@@ -5433,6 +5433,7 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
 	free_extent_buffer(buf);
 out_free_reserved:
 	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, false);
+	return ERR_PTR(ret);
 out_unuse:
 	btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
 	return ERR_PTR(ret);
Reviewed-by on v2 1/1
#1 lore raw
Quoted 20 lines · Vlad Poenaru

When btrfs_alloc_tree_block() fails after btrfs_reserve_extent() has succeeded, its error path gives the block back to the block reserve, although the block has already left bytes_may_use. Each such failure leaves bytes_may_use a block short, and it underflows and warns in space-info.h when the global block reserve is released at unmount, or earlier.

btrfs_alloc_tree_block()

   |- btrfs_use_block_rsv()
   |- btrfs_reserve_extent()
   |  \- btrfs_add_reserved_bytes()  <- bytes_may_use -= blocksize
   |- btrfs_init_new_buffer()
   |  \- alloc_extent_buffer()       <- fails with -ENOMEM
   |- btrfs_free_reserved_extent()   <- bytes_reserved -= blocksize

\- btrfs_unuse_block_rsv() <- block rsv += blocksize again

Since commit 19d7f65f032f ("btrfs: convert the buffer_radix to an xarray") in 6.16, alloc_extent_buffer() can fail this way under memory cgroup pressure, because fs_info->buffer_tree is memcg-accounted.

Quoted 7 lines

Return without giving the block back once btrfs_reserve_extent() has succeeded.

Fixes 67b7859e9bfa ("btrfs: handle ENOMEM in btrfs_alloc_tree_block")

Thanks,
Qu

Quoted 27 lines

---
v2:
- Shorter changelog, built around a call trace (Qu Wenruo).
- Dropped the code comment (Boris Burkov).

The code is v1's apart from the removed comment, so v1's testing stands:
its reproducer leaves bytes_may_use below the global reserve without the fix and not with it, and fstests btrfs/002, btrfs/220, generic/475 and generic/476 pass.

v1, with the reproducer:
https://lore.kernel.org/all/[email protected]/

  fs/btrfs/extent-tree.c | 1 +
  1 file changed, 1 insertion(+)
fs/btrfs/extent-tree.c +1 -0
index 9b0a5c6b96f3c..8d4c66bf850bc 100644
@@ -5433,6 +5433,7 @@ struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
  	free_extent_buffer(buf);
  out_free_reserved:
  	btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, false);
+	return ERR_PTR(ret);
  out_unuse:
  	btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
  	return ERR_PTR(ret);