bcachefs: Check for bucket journal seq in the future

This fixes an assertion pop in bch2_journal_noflush_seq() - log the
error to the superblock and continue instead.

Reported-by: syzbot+85700120f75fc10d4e18@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c
index 6de0387ede999f884a33cbc70ba32930b4369c0b..e8c246e5803c2756336abb6878af67fd4c252521 100644 (file)
--- a/fs/bcachefs/alloc_background.c
+++ b/fs/bcachefs/alloc_background.c
@@ -926,37 +926,43 @@ int bch2_trigger_alloc(struct btree_trans *trans,
        }
 
        if ((flags & BTREE_TRIGGER_atomic) && (flags & BTREE_TRIGGER_insert)) {
-               u64 journal_seq = trans->journal_res.seq;
-               u64 bucket_journal_seq = new_a->journal_seq;
+               u64 transaction_seq = trans->journal_res.seq;
 
-               if ((flags & BTREE_TRIGGER_insert) &&
-                   data_type_is_empty(old_a->data_type) !=
-                   data_type_is_empty(new_a->data_type) &&
-                   new.k->type == KEY_TYPE_alloc_v4) {
-                       struct bch_alloc_v4 *v = bkey_s_to_alloc_v4(new).v;
+               if (log_fsck_err_on(transaction_seq && new_a->journal_seq > transaction_seq,
+                                   trans, alloc_key_journal_seq_in_future,
+                                   "bucket journal seq in future (currently at %llu)\n%s",
+                                   journal_cur_seq(&c->journal),
+                                   (bch2_bkey_val_to_text(&buf, c, new.s_c), buf.buf)))
+                       new_a->journal_seq = transaction_seq;
 
-                       /*
-                        * If the btree updates referring to a bucket weren't flushed
-                        * before the bucket became empty again, then the we don't have
-                        * to wait on a journal flush before we can reuse the bucket:
-                        */
-                       v->journal_seq = bucket_journal_seq =
-                               data_type_is_empty(new_a->data_type) &&
-                               (journal_seq == v->journal_seq ||
-                                bch2_journal_noflush_seq(&c->journal, v->journal_seq))
-                               ? 0 : journal_seq;
-               }
+               int is_empty_delta = (int) data_type_is_empty(new_a->data_type) -
+                                    (int) data_type_is_empty(old_a->data_type);
 
-               if (!data_type_is_empty(old_a->data_type) &&
-                   data_type_is_empty(new_a->data_type) &&
-                   bucket_journal_seq) {
-                       ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
-                                       c->journal.flushed_seq_ondisk,
-                                       new.k->p.inode, new.k->p.offset,
-                                       bucket_journal_seq);
-                       if (bch2_fs_fatal_err_on(ret, c,
-                                       "setting bucket_needs_journal_commit: %s", bch2_err_str(ret)))
-                               goto err;
+               /* Record journal sequence number of empty -> nonempty transition: */
+               if (is_empty_delta < 0)
+                       new_a->journal_seq = max(new_a->journal_seq, transaction_seq);
+
+               /*
+                * Bucket becomes empty: mark it as waiting for a journal flush,
+                * unless updates since empty -> nonempty transition were never
+                * flushed - we may need to ask the journal not to flush
+                * intermediate sequence numbers:
+                */
+               if (is_empty_delta > 0) {
+                       if (new_a->journal_seq == transaction_seq ||
+                           bch2_journal_noflush_seq(&c->journal, new_a->journal_seq))
+                               new_a->journal_seq = 0;
+                       else {
+                               new_a->journal_seq = transaction_seq;
+
+                               ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
+                                               c->journal.flushed_seq_ondisk,
+                                               new.k->p.inode, new.k->p.offset,
+                                               transaction_seq);
+                               if (bch2_fs_fatal_err_on(ret, c,
+                                               "setting bucket_needs_journal_commit: %s", bch2_err_str(ret)))
+                                       goto err;
+                       }
                }
 
                if (new_a->gen != old_a->gen) {
@@ -1004,6 +1010,7 @@ int bch2_trigger_alloc(struct btree_trans *trans,
                rcu_read_unlock();
        }
 err:
+fsck_err:
        printbuf_exit(&buf);
        bch2_dev_put(ca);
        return ret;

diff --git a/fs/bcachefs/sb-errors_format.h b/fs/bcachefs/sb-errors_format.h
index 382fcafa815a9d1ea27043aa11e083503131fd67..8e3a6c5da10d40d7cfc1c443922f5445718825f7 100644 (file)
--- a/fs/bcachefs/sb-errors_format.h
+++ b/fs/bcachefs/sb-errors_format.h
@@ -122,6 +122,7 @@ enum bch_fsck_flags {
        x(alloc_key_cached_sectors_wrong,                       109,    FSCK_AUTOFIX)   \
        x(alloc_key_stripe_wrong,                               110,    FSCK_AUTOFIX)   \
        x(alloc_key_stripe_redundancy_wrong,                    111,    FSCK_AUTOFIX)   \
+       x(alloc_key_journal_seq_in_future,                      298,    FSCK_AUTOFIX)   \
        x(bucket_sector_count_overflow,                         112,    0)              \
        x(bucket_metadata_type_mismatch,                        113,    0)              \
        x(need_discard_key_wrong,                               114,    FSCK_AUTOFIX)   \
@@ -308,7 +309,7 @@ enum bch_fsck_flags {
        x(logged_op_but_clean,                                  283,    FSCK_AUTOFIX)   \
        x(compression_opt_not_marked_in_sb,                     295,    FSCK_AUTOFIX)   \
        x(compression_type_not_marked_in_sb,                    296,    FSCK_AUTOFIX)   \
-       x(MAX,                                                  298,    0)
+       x(MAX,                                                  299,    0)
 
 enum bch_sb_error_id {
 #define x(t, n, ...) BCH_FSCK_ERR_##t = n,