[PATCH] bcachefs: Fix infinite loop in backpointers fsck on upgrade

[PATCH] bcachefs: Fix infinite loop in backpointers fsck on upgrade

[Kent Overstreet]

On upgrade from before bcachefs_metadata_version_backpointer_bucket_gen
we'll typically skip the write buffer flush, because we expect many
invalid/missing backpointers and it's faster to just repair them.

When we find backpointers that sum up to more than the bucket sector
counters, that's a special situation - we need to walk those
backpointers to kill the bad ones, and so that case needs a write buffer
flush to avoid an infinite loop.

Reported-by: Jérôme Poulin <jeromepoulin@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/backpointers.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

diff --git a/fs/bcachefs/backpointers.c b/fs/bcachefs/backpointers.c
index 77d93beb3c8f..9d0c3cb634e8 100644
--- a/fs/bcachefs/backpointers.c
+++ b/fs/bcachefs/backpointers.c
@@ -943,20 +943,24 @@ static int check_bucket_backpointer_mismatch(struct btree_trans *trans, struct b
 	if (sectors[ALLOC_dirty]  != a->dirty_sectors ||
 	    sectors[ALLOC_cached] != a->cached_sectors ||
 	    sectors[ALLOC_stripe] != a->stripe_sectors) {
-		if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_backpointer_bucket_gen) {
+		if (sectors[ALLOC_dirty]  > a->dirty_sectors ||
+		    sectors[ALLOC_cached] > a->cached_sectors ||
+		    sectors[ALLOC_stripe] > a->stripe_sectors) {
 			ret = bch2_backpointers_maybe_flush(trans, alloc_k, last_flushed);
 			if (ret)
 				goto err;
-		}
 
-		if (sectors[ALLOC_dirty]  > a->dirty_sectors ||
-		    sectors[ALLOC_cached] > a->cached_sectors ||
-		    sectors[ALLOC_stripe] > a->stripe_sectors) {
 			ret = check_bucket_backpointers_to_extents(trans, ca, alloc_k.k->p) ?:
 				bch_err_throw(c, transaction_restart_nested);
 			goto err;
 		}
 
+		if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_backpointer_bucket_gen) {
+			ret = bch2_backpointers_maybe_flush(trans, alloc_k, last_flushed);
+			if (ret)
+				goto err;
+		}
+
 		bool empty = (sectors[ALLOC_dirty] +
 			      sectors[ALLOC_stripe] +
 			      sectors[ALLOC_cached]) == 0;
-- 
2.50.0