Skip to contentSkip to conversation

btrfs: abort transaction in do_remap_reloc_trans() on failure

PATCH Superseded Mark Harmstone 2 participants 3 messages Apr 7 – Apr 8 Btrfs file system

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 btrfs: abort transaction in do_remap_reloc_trans() on failure

If one of the calls made by do_remap_reloc_trans() fails, we can leave the remap tree in an inconsistent state. Abort the transaction if this happens, to prevent the corrupt state from reaching the disk.

---
 fs/btrfs/relocation.c | 1 +
 1 file changed, 1 insertion(+)
fs/btrfs/relocation.c +1 -0
index 1c42c5180bddd5..277a0042967665 100644
@@ -5089,6 +5089,7 @@ static int do_remap_reloc_trans(struct btrfs_fs_info *fs_info,
 	btrfs_free_reserved_extent(fs_info, new_addr, length, 0);
 
 	mutex_unlock(&fs_info->remap_mutex);
+	btrfs_abort_transaction(trans, ret);
 	btrfs_end_transaction(trans);
 
 	return ret;
Signature · 1 line
2.52.0
#1 lore raw
Quoted 14 lines · Mark Harmstone

If one of the calls made by do_remap_reloc_trans() fails, we can leave the remap tree in an inconsistent state. Abort the transaction if this happens, to prevent the corrupt state from reaching the disk.

---
 fs/btrfs/relocation.c | 1 +
 1 file changed, 1 insertion(+)
fs/btrfs/relocation.c +0 -0
index 1c42c5180bddd5..277a0042967665 100644

fs/btrfs/relocation.c static int do_remap_reloc_trans(struct btrfs_fs_info *fs_info, · line 5092

Quoted 4 lines
fs/btrfs/relocation.c +1 -0
@@ -5089,6 +5089,7 @@ static int do_remap_reloc_trans(struct btrfs_fs_info *fs_info,
 	btrfs_free_reserved_extent(fs_info, new_addr, length, 0);
 
 	mutex_unlock(&fs_info->remap_mutex);
+	btrfs_abort_transaction(trans, ret);

Counterintuitively to the usual pattern of sharing code in fail: labels, I think it is quite nice to put a separate abort at each failing code path that requires an abort. That way we get better diagnostic information on what actually went wrong.

e.g. a recent patch from Filipe unfolding it across a function call:
https://lore.kernel.org/linux-btrfs/99743ac0b4b9a69dcaa8f9adcaf301bd1a0f1a89.1769163248.git.fdmanana@suse.com/

Thanks,
Boris

Quoted 6 lines
fs/btrfs/relocation.c +0 -0
 	btrfs_end_transaction(trans);
 
 	return ret;
Signature · 1 line
2.52.0
#2 lore raw
Quoted 25 lines · Boris Burkov
Quoted 14 lines · Mark Harmstone

If one of the calls made by do_remap_reloc_trans() fails, we can leave the remap tree in an inconsistent state. Abort the transaction if this happens, to prevent the corrupt state from reaching the disk.

---
  fs/btrfs/relocation.c | 1 +
  1 file changed, 1 insertion(+)
fs/btrfs/relocation.c +0 -0
index 1c42c5180bddd5..277a0042967665 100644

fs/btrfs/relocation.c static int do_remap_reloc_trans(struct btrfs_fs_info *fs_info, · line 5092

Quoted 4 lines
fs/btrfs/relocation.c +1 -0
@@ -5089,6 +5089,7 @@ static int do_remap_reloc_trans(struct btrfs_fs_info *fs_info,
  	btrfs_free_reserved_extent(fs_info, new_addr, length, 0);
 
  	mutex_unlock(&fs_info->remap_mutex);
+	btrfs_abort_transaction(trans, ret);

Counterintuitively to the usual pattern of sharing code in fail: labels, I think it is quite nice to put a separate abort at each failing code path that requires an abort. That way we get better diagnostic information on what actually went wrong.

Quoted 2 lines

e.g. a recent patch from Filipe unfolding it across a function call:
https://lore.kernel.org/linux-btrfs/99743ac0b4b9a69dcaa8f9adcaf301bd1a0f1a89.1769163248.git.fdmanana@suse.com/

Thanks Boris - if aborting a transaction logs the line number, that's a very good reason to do it in place. This also shows that the call to btrfs_add_to_free_space_tree after add_remap_entry fails is pointless, as the transaction is about to be aborted.

Quoted 9 lines

Thanks,
Boris

Quoted 6 lines
fs/btrfs/relocation.c +0 -0
  	btrfs_end_transaction(trans);
 
  	return ret;
Signature · 1 line
2.52.0