Re: [PATCH] Revert "bcachefs: Add asserts to bch2_dev_btree_bitmap_marked_sectors()"

From: Manas
Date: Tue Oct 22 2024 - 09:59:23 EST


On 22.10.2024 09:43, Hongbo Li wrote:
On 2024/10/22 0:48, Manas via B4 Relay wrote:
From: Manas <manas18244@xxxxxxxxxxx>

This reverts commit 60f2b1bcf519416dbffee219132aa949d0c39d0e.

This syzbot bug[1] is triggered due to the BUG_ON assertions added in
__bch2_dev_btree_bitmap_mark. During runtime, m->btree_bitmap_shift is
63 '?'. This triggers both the assertions.

Reverting the commit does not reproduce the said bug.

[1] https://syzkaller.appspot.com/bug?extid=e8eff054face85d7ea41

Signed-off-by: Manas <manas18244@xxxxxxxxxxx>
Reported-by: syzbot+e8eff054face85d7ea41@xxxxxxxxxxxxxxxxxxxxxxxxx
Closes: https://syzkaller.appspot.com/bug?extid=e8eff054face85d7ea41
---
This syzbot bug[1] is triggered due to the BUG_ON assertions added in
__bch2_dev_btree_bitmap_mark. During runtime, m->btree_bitmap_shift is
63 '?'. This triggers both the assertions.

I am unfamiliar with the codebase, and there wasn't a lore discussion
about the assertions in the commit, so I am unsure about the relevance
of these assertions.

Reverting the commit does not reproduce the said bug.

[1] https://syzkaller.appspot.com/bug?extid=e8eff054face85d7ea41
---
fs/bcachefs/sb-members.c | 3 ---
1 file changed, 3 deletions(-)

diff --git a/fs/bcachefs/sb-members.c b/fs/bcachefs/sb-members.c
index fb08dd680dacf82bca414f424024e4a00bf432de..9790fd47338c46d2af30547e1f41a1e578b71aa4 100644
--- a/fs/bcachefs/sb-members.c
+++ b/fs/bcachefs/sb-members.c
@@ -450,9 +450,6 @@ static void __bch2_dev_btree_bitmap_mark(struct bch_sb_field_members_v2 *mi, uns
m->btree_bitmap_shift += resize;
}
- BUG_ON(m->btree_bitmap_shift > 57);
- BUG_ON(end > 64ULL << m->btree_bitmap_shift);
-
May be this is not good way by just removing the BUG_ON. In my humble opinion, the former code have checked m->btree_bitmap_shift in bch2_dev_btree_bitmap_marked_sectors. May be add the similar condition in this helper will be better.

Hi Hongbo, thanks for reviewing this. I was unsure about the fix so I decided to
initiate the conversation by reverting. Yes, that makes sense. I am sending a
fresh patch adding that condition.

--
Manas