On Tue, Jul 23, 2024 at 09:08:00PM GMT, Camila Alvarez wrote:Yes, sorry for the delay. The closure_sync will still wait, but even after waiting there's no space left.
bch2_extent_fallocate should not return -BCH_ERR_transaction_restart_nested when there are no buckets available.
Returning -BCH_ERR_transaction_restart_nested causes the sector allocation to retry even if there's no space available.
It seems like you're saying this fixes a livelock, but I don't see that:
if we return operation_blocked, the closure_sync we do before we return
will still wait.
More explanation?
Reported-by: syzbot+1e515cab343dbe5aa38a@xxxxxxxxxxxxxxxxxxxxxxxxx
Signed-off-by: Camila Alvarez <cam.alvarez.i@xxxxxxxxx>
---
fs/bcachefs/io_misc.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/fs/bcachefs/io_misc.c b/fs/bcachefs/io_misc.c
index 82f9170dab3f..b5b92ef030df 100644
--- a/fs/bcachefs/io_misc.c
+++ b/fs/bcachefs/io_misc.c
@@ -90,8 +90,6 @@ int bch2_extent_fallocate(struct btree_trans *trans,
opts.data_replicas,
opts.data_replicas,
BCH_WATERMARK_normal, 0, &cl, &wp);
- if (bch2_err_matches(ret, BCH_ERR_operation_blocked))
- ret = -BCH_ERR_transaction_restart_nested;
if (ret)
goto err;
--
2.34.1