Re: [PATCH] bcachefs: WARNING in bch2_fs_journal_stop

[Camila Alvarez Inostroza]

On Thu, 18 Jul 2024, Kent Overstreet wrote:

On Wed, Jul 17, 2024 at 06:02:39PM GMT, Camila Alvarez wrote:

last_seq and last_empty_seq suffered from an off by one error when the
journal has no entries.

The indexes were fixed and an assertion is added to check that the
last_empty_seq is always kept under the next valid seq number.

oh nice. I'm going to need to stare at this some more, I still feel like
this code needs to be clearer and less fiddly.

Sure! Please let me know if there's anything I can do to help.

Reported-by: syzbot+4093905737cf289b6...@syzkaller.appspotmail.com
Signed-off-by: Camila Alvarez <cam.alvare...@gmail.com>
---
 fs/bcachefs/journal.c | 23 +++++++++++++----------
 1 file changed, 13 insertions(+), 10 deletions(-)

diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c
index 10b19791ec98..7bbbf4b149e9 100644
--- a/fs/bcachefs/journal.c
+++ b/fs/bcachefs/journal.c
@@ -1201,7 +1201,7 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
        struct journal_replay *i, **_i;
        struct genradix_iter iter;
        bool had_entries = false;
-       u64 last_seq = cur_seq, nr, seq;
+        u64 last_written_seq = cur_seq - 1, last_seq = cur_seq - 1, nr, seq;

        genradix_for_each_reverse(&c->journal_entries, iter, _i) {
                i = *_i;
@@ -1209,11 +1209,11 @@ int bch2_fs_journal_start(struct journal *j, u64 
cur_seq)
                if (journal_replay_ignore(i))
                        continue;

-               last_seq = le64_to_cpu(i->j.last_seq);
+               last_written_seq = le64_to_cpu(i->j.last_seq);
                break;
        }

-       nr = cur_seq - last_seq;
+       nr = cur_seq - last_written_seq;

        if (nr + 1 > j->pin.size) {
                free_fifo(&j->pin);
@@ -1224,14 +1224,14 @@ int bch2_fs_journal_start(struct journal *j, u64 
cur_seq)
                }
        }

-       j->replay_journal_seq        = last_seq;
+       j->replay_journal_seq        = last_written_seq;
        j->replay_journal_seq_end = cur_seq;
-       j->last_seq_ondisk   = last_seq;
-       j->flushed_seq_ondisk        = cur_seq - 1;
-       j->seq_ondisk                = cur_seq - 1;
-       j->pin.front         = last_seq;
+       j->last_seq_ondisk   = last_written_seq;
+       j->flushed_seq_ondisk        = last_seq;
+       j->seq_ondisk                = last_seq;
+       j->pin.front         = last_written_seq;
        j->pin.back          = cur_seq;
-       atomic64_set(&j->seq, cur_seq - 1);
+       atomic64_set(&j->seq, last_seq);

        fifo_for_each_entry_ptr(p, &j->pin, seq)
                journal_pin_list_init(p, 1);
@@ -1261,7 +1261,10 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
        }

        if (!had_entries)
-               j->last_empty_seq = cur_seq;
+               j->last_empty_seq = last_seq;
+
+       WARN(j->last_empty_seq >= cur_seq, "journal startup error: last empty seq 
%llu is higher or equal than the next seq number to be used (%llu)",
+             j->last_empty_seq, cur_seq);

        spin_lock(&j->lock);

--
2.34.1