Hello!
On Thu 12-03-20 10:38:17, lishan wrote:
> I have a long-standing problem,
> In the ocfs2 file system, there is a panic problem related to b_committed_data,
> details can be found in the historical mail.
>
> Refer to the old version of the ext3 file system,
> ocfs2 file system uses the undo process to ensure that metadata
> which is not committed to disk is not reused.
>
> I understand that for transactions that have not been committed,
> b_committed_data will not be NULL, so,
> in what scenario do you think b_committed_data will be set to NULL?
I'm sorry for a late reply. I was on vacation last week. Yes, so once you
call jbd2_journal_get_undo_access(), jh->b_committed_data can become NULL
only during transaction commit. So I'm not sure how it could happen that
ocfs2_block_group_clear_bits() sees NULL b_committed data after doing
jbd2_journal_get_undo_access()... That certainly requires more
investigation before papering over the problem. I can see the reporter had
a crashdumping enabled so should be able to show in which state the
journal, handle, and jh is and maybe that will give us a clue.
Honza
> On 2020/3/11 9:15, Joseph Qi wrote:
> >
> >
> > On 2020/3/9 16:26, lishan wrote:
> >> A NULL pointer panic dereference in ocfs2_block_group_clear_bits() happen again.
> >> The information of NULL pointer stack as follows:
> >>
> >> PID: 81866 TASK: ffffa07c3c21ae80 CPU: 66 COMMAND: "fallocate"
> >> #0 [ffff0000b4d6b0b0] machine_kexec at ffff0000800a2954
> >> #1 [ffff0000b4d6b110] __crash_kexec at ffff0000801bab34
> >> #2 [ffff0000b4d6b2a0] panic at ffff0000800f02cc
> >> #3 [ffff0000b4d6b380] die at ffff00008008f6ac
> >> #4 [ffff0000b4d6b3d0] bug_handler at ffff00008008f744
> >> #5 [ffff0000b4d6b400] brk_handler at ffff000080085d1c
> >> #6 [ffff0000b4d6b420] do_debug_exception at ffff000080081194
> >> #7 [ffff0000b4d6b630] el1_dbg at ffff00008008332c
> >> PC: ffff00000190e9c0 [_ocfs2_free_suballoc_bits+1608]
> >> LR: ffff00000190e990 [_ocfs2_free_suballoc_bits+1560]
> >> SP: ffff0000b4d6b640 PSTATE: 60400009
> >> X29: ffff0000b4d6b650 X28: 0000000000000000 X27: 00000000000052f3
> >> X26: ffff807c511a9570 X25: ffff807ca0054000 X24: 00000000000052f2
> >> X23: 0000000000000001 X22: ffff807c7cde7a90 X21: ffff0000811d9000
> >> X20: ffff807c5e7d2000 X19: ffff00000190c768 X18: 0000000000000000
> >> X17: 0000000000000000 X16: ffff000080a032f0 X15: 0000000000000000
> >> X14: ffffffffffffffff X13: fffffffffffffff7 X12: ffffffffffffffff
> >> X11: 0000000000000038 X10: 0101010101010101 X9: ffffffffffffffff
> >> X8: 7f7f7f7f7f7f7f7f X7: 0000000000000000 X6: 0000000000000080
> >> X5: 0000000000000000 X4: 0000000000000002 X3: ffff00000199f390
> >> X2: a603c08321456e00 X1: ffff807c7cde7a90 X0: 0000000000000000
> >> #8 [ffff0000b4d6b650] _ocfs2_free_suballoc_bits at ffff00000190e9bc [ocfs2]
> >> #9 [ffff0000b4d6b710] _ocfs2_free_clusters at ffff0000019110d4 [ocfs2]
> >> #10 [ffff0000b4d6b790] ocfs2_free_clusters at ffff000001913e94 [ocfs2]
> >> #11 [ffff0000b4d6b7d0] __ocfs2_flush_truncate_log at ffff0000018b5294 [ocfs2]
> >> #12 [ffff0000b4d6b8a0] ocfs2_remove_btree_range at ffff0000018bb34c [ocfs2]
> >> #13 [ffff0000b4d6b960] ocfs2_commit_truncate at ffff0000018bc76c [ocfs2]
> >> #14 [ffff0000b4d6ba60] ocfs2_wipe_inode at ffff0000018e57bc [ocfs2]
> >> #15 [ffff0000b4d6bb00] ocfs2_evict_inode at ffff0000018e5db8 [ocfs2]
> >> #16 [ffff0000b4d6bb70] evict at ffff000080365040
> >> #17 [ffff0000b4d6bba0] iput at ffff0000803655d8
> >> #18 [ffff0000b4d6bbe0] ocfs2_dentry_iput at ffff0000018c60a0 [ocfs2]
> >> #19 [ffff0000b4d6bc30] dentry_unlink_inode at ffff00008035ef58
> >> #20 [ffff0000b4d6bc50] __dentry_kill at ffff000080360384
> >> #21 [ffff0000b4d6bc80] dentry_kill at ffff000080360670
> >> #22 [ffff0000b4d6bcb0] dput at ffff00008036093c
> >> #23 [ffff0000b4d6bcf0] __fput at ffff000080343930
> >> #24 [ffff0000b4d6bd40] ____fput at ffff000080343aac
> >> #25 [ffff0000b4d6bd60] task_work_run at ffff0000801172fc
> >>
> >> The direct panic reason is that bh2jh (group_bh)-> b_committed_data is
> >> null. It is presumed that the network was disconnected during the
> >> write process, causing the transaction abort. as follows:
> >> jbd2_journal_abort
> >> .......
> >> jbd2_journal_commit_transaction
> >> jh->b_committed_data = NULL;
> >>
> >> _ocfs2_free_suballoc_bits
> >> ocfs2_block_group_clear_bits
> >> // undo_bg is now set to null
> >> BUG_ON(!undo_bg);
> >>
> >> When applying for free space, if b_committed_data is null,
> >> it will be directly occupied, as follows:
> >> ocfs2_cluster_group_search
> >> ocfs2_block_group_find_clear_bits
> >> ocfs2_test_bg_bit_allocatable:
> >> bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
> >> if (bg)
> >> ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
> >> else
> >> ret = 1;
> >> b_committed_data is an intermediate state backup for bitmap transaction commits,
> >> newly applied space can overwrite previous dirty data,
> >> so, I think, while free clusters, if b_committed_data is null, ignore it.
> >> Host panic directly, too violent.
> >>
> >> Signed-off-by: Shan Li <lishan24@xxxxxxxxxx>
> >> Reviewed-by: Jun Piao <piaojun@xxxxxxxxxx>
> >> ---
> >> fs/ocfs2/suballoc.c | 9 +++++++--
> >> 1 file changed, 7 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/fs/ocfs2/suballoc.c b/fs/ocfs2/suballoc.c
> >> index 939df99d2dec..aaf1b3cbd984 100644
> >> --- a/fs/ocfs2/suballoc.c
> >> +++ b/fs/ocfs2/suballoc.c
> >> @@ -2412,14 +2412,19 @@ static int ocfs2_block_group_clear_bits(handle_t *handle,
> >> if (undo_fn) {
> >> spin_lock(&jh->b_state_lock);
> >> undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
> >> - BUG_ON(!undo_bg);
> >> + if (!undo_bg)
> >> + mlog(ML_NOTICE, "%s: group descriptor # %llu (device %s) journal "
> >> + "b_committed_data had been cleared.\n",
> >> + OCFS2_SB(alloc_inode->i_sb)->uuid_str,
> >> + (unsigned long long)le64_to_cpu(bg->bg_blkno),
> >> + alloc_inode->i_sb->s_id);
> >
> > Seems a kind of workaround.
> > I am worrying about other abnormal cases of NULL b_committed_data, it
> > may lead to a corrupt filesystem.
> > So how about isolating the journal abort case?
> >
> > Thanks,
> > Joseph
> >
> >> }
> >>
> >> tmp = num_bits;
> >> while(tmp--) {
> >> ocfs2_clear_bit((bit_off + tmp),
> >> (unsigned long *) bg->bg_bitmap);
> >> - if (undo_fn)
> >> + if (undo_fn && undo_bg)
> >> undo_fn(bit_off + tmp,
> >> (unsigned long *) undo_bg->bg_bitmap);
> >> }
> >>
> >
> > .
> >
>
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR
