On Wed 15-01-25 18:53:41, Jan Kara wrote: > On Wed 15-01-25 13:00:23, Heming Zhao wrote: > > Hello Jan, > > > > On 1/15/25 09:32, Liebes Wang wrote: > > > The bisection log shows the first cause commit is a09decff5c32060639a685581c380f51b14e1fc2: > > > a09decff5c32 jbd2: clear JBD2_ABORT flag before journal_reset to update log tail info when load journal > > > > > > The full bisection log is attached. Hope this helps. > > > > This bisearch commit a09decff5c32 appears to be the root cause > > of this issue. It fixed one issue but introduced another. > > > > Syzbot tested the patch with calling jbd2_journal_wipe() with 'write=1'. > > The Syzbot test result [1] shows that the same WARN_ON() is triggered > > in a subsequent routine – the classic whack-a-mole! > > > > Back to commit a09decff5c32, it opened a door to allow jbd2 to update > > sb regardless of whether the value of sb items are correct. > > > > To fix a09decff5c32, it seems that jbd2 needs to add more sanity check > > codes in a sub-routine of jbd2_journal_load(). > > > > btw, in my view, this is a jbd2 issue not ocfs2/ext4 issue. > > > > [1]: https://lore.kernel.org/ocfs2-devel/04a9ad29-51de-4b50-a5bb-56f91817639d@xxxxxxxx/T/#m86d01f83d808868bb5e6548d30f79b4f9f889b13 > > Thanks for debugging this! So I'm not 100% convinced this is only jbd2 bug > because jbd2_journal_recover() was never intended to be called after > jbd2_journal_skip_recovery() (called from jbd2_journal_wipe()). You're > supposed to call either jbd2_journal_wipe() or jbd2_journal_recover() but > not both. So IMO this needs fixing in OCFS2 code. That being said you've > also pointed at one bug in jbd2 code - the WARN_ON(!sb->s_sequence) in > jbd2_journal_update_sb_log_tail() is indeed wrong. We were inconsistent > inside jbd2 whether TID 0 is considered valid or not and relatively > recently we've decided to accept TID 0 as valid but this place was left > out. I'll send a fix for that. OK, after checking again OCFS2 is indeed fine here. I'm sorry for the confusion. I'll send appropriate jbd2 fixes shortly. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
