[PATCH] btrfs-progs: fix a wrong return value for write_and_map_eb()

[PATCH] btrfs-progs: fix a wrong return value for write_and_map_eb()

[Qu Wenruo]

[BUG]
With commit "btrfs-progs: use write_data_to_disk() to replace
write_extent_to_disk()", "mkfs.btrfs -m raid5" will always fail:

 kernel-shared/transaction.c:155: __commit_transaction: BUG_ON `ret` triggered, value 65536
 ./mkfs.btrfs(+0x57dd2)[0x557e610addd2]
 ./mkfs.btrfs(+0x57e70)[0x557e610ade70]
 ./mkfs.btrfs(__commit_transaction+0x139)[0x557e610ae4d4]
 ./mkfs.btrfs(btrfs_commit_transaction+0x214)[0x557e610ae746]
 ./mkfs.btrfs(main+0x19d9)[0x557e61066d34]
 /usr/lib/libc.so.6(+0x2d310)[0x7f2bf222d310]
 /usr/lib/libc.so.6(__libc_start_main+0x81)[0x7f2bf222d3c1]
 ./mkfs.btrfs(_start+0x25)[0x557e61062925]
 Aborted (core dumped)

[CAUSE]
Commit "btrfs-progs: use write_data_to_disk() to replace
write_extent_to_disk()" refactor write_and_map_eb() very slightly, but
it changed the return value on successful writeback for RAID56.

Previously we return 0, but now we return the stripe length, causing
btrfs_commit_transaction() to freak out.

[FIX]
Fix it by just reseting ret to 0 if nothing wrong happened for RAID56.

Please fold this fix into commit "btrfs-progs: use write_data_to_disk()
to replace write_extent_to_disk()".

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 kernel-shared/disk-io.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/kernel-shared/disk-io.c b/kernel-shared/disk-io.c
index 4aae7a35768c..26b1c9aa192a 100644
--- a/kernel-shared/disk-io.c
+++ b/kernel-shared/disk-io.c
@@ -477,6 +477,8 @@ int write_and_map_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
 			error(
 		"failed to write raid56 stripe for bytenr %llu length %llu: %m",
 				eb->start, length);
+		} else {
+			ret = 0;
 		}
 		goto out;
 	}
-- 
2.35.1

Re: [PATCH] btrfs-progs: fix a wrong return value for write_and_map_eb()

[David Sterba]

On Mon, Apr 18, 2022 at 09:54:24AM +0800, Qu Wenruo wrote:
> [BUG]
> With commit "btrfs-progs: use write_data_to_disk() to replace
> write_extent_to_disk()", "mkfs.btrfs -m raid5" will always fail:
> 
>  kernel-shared/transaction.c:155: __commit_transaction: BUG_ON `ret` triggered, value 65536
>  ./mkfs.btrfs(+0x57dd2)[0x557e610addd2]
>  ./mkfs.btrfs(+0x57e70)[0x557e610ade70]
>  ./mkfs.btrfs(__commit_transaction+0x139)[0x557e610ae4d4]
>  ./mkfs.btrfs(btrfs_commit_transaction+0x214)[0x557e610ae746]
>  ./mkfs.btrfs(main+0x19d9)[0x557e61066d34]
>  /usr/lib/libc.so.6(+0x2d310)[0x7f2bf222d310]
>  /usr/lib/libc.so.6(__libc_start_main+0x81)[0x7f2bf222d3c1]
>  ./mkfs.btrfs(_start+0x25)[0x557e61062925]
>  Aborted (core dumped)
> 
> [CAUSE]
> Commit "btrfs-progs: use write_data_to_disk() to replace
> write_extent_to_disk()" refactor write_and_map_eb() very slightly, but
> it changed the return value on successful writeback for RAID56.
> 
> Previously we return 0, but now we return the stripe length, causing
> btrfs_commit_transaction() to freak out.
> 
> [FIX]
> Fix it by just reseting ret to 0 if nothing wrong happened for RAID56.
> 
> Please fold this fix into commit "btrfs-progs: use write_data_to_disk()
> to replace write_extent_to_disk()".

Now folded and the original series with raid56 repair added to devel
again, thanks.