Revert "bcachefs: Delete some obsolete journal_seq_blacklist code"

This reverts commit f95b61228efd04c9c158123da5827c96e9773b29.

It turns out, we're seeing filesystems in the wild end up with
blacklisted btree node bsets - this should not be happening, and until
we understand why and fix it we need to keep this code around.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

fs/bcachefs/bcachefs.h

@@ -755,6 +755,7 @@ struct bch_fs {
 	/* JOURNAL SEQ BLACKLIST */
 	struct journal_seq_blacklist_table *
 				journal_seq_blacklist_table;
+	struct work_struct	journal_seq_blacklist_gc_work;
 
 	/* ALLOCATOR */
 	spinlock_t		freelist_lock;

fs/bcachefs/journal_seq_blacklist.c

@@ -241,3 +241,81 @@ const struct bch_sb_field_ops bch_sb_field_ops_journal_seq_blacklist = {
 	.validate	= bch2_sb_journal_seq_blacklist_validate,
 	.to_text	= bch2_sb_journal_seq_blacklist_to_text
 };
+
+void bch2_blacklist_entries_gc(struct work_struct *work)
+{
+	struct bch_fs *c = container_of(work, struct bch_fs,
+					journal_seq_blacklist_gc_work);
+	struct journal_seq_blacklist_table *t;
+	struct bch_sb_field_journal_seq_blacklist *bl;
+	struct journal_seq_blacklist_entry *src, *dst;
+	struct btree_trans trans;
+	unsigned i, nr, new_nr;
+	int ret;
+
+	bch2_trans_init(&trans, c, 0, 0);
+
+	for (i = 0; i < BTREE_ID_NR; i++) {
+		struct btree_iter iter;
+		struct btree *b;
+
+		bch2_trans_node_iter_init(&trans, &iter, i, POS_MIN,
+					  0, 0, BTREE_ITER_PREFETCH);
+retry:
+		bch2_trans_begin(&trans);
+
+		b = bch2_btree_iter_peek_node(&iter);
+
+		while (!(ret = PTR_ERR_OR_ZERO(b)) &&
+		       b &&
+		       !test_bit(BCH_FS_STOPPING, &c->flags))
+			b = bch2_btree_iter_next_node(&iter);
+
+		if (ret == -EINTR)
+			goto retry;
+
+		bch2_trans_iter_exit(&trans, &iter);
+	}
+
+	bch2_trans_exit(&trans);
+	if (ret)
+		return;
+
+	mutex_lock(&c->sb_lock);
+	bl = bch2_sb_get_journal_seq_blacklist(c->disk_sb.sb);
+	if (!bl)
+		goto out;
+
+	nr = blacklist_nr_entries(bl);
+	dst = bl->start;
+
+	t = c->journal_seq_blacklist_table;
+	BUG_ON(nr != t->nr);
+
+	for (src = bl->start, i = eytzinger0_first(t->nr);
+	     src < bl->start + nr;
+	     src++, i = eytzinger0_next(i, nr)) {
+		BUG_ON(t->entries[i].start	!= le64_to_cpu(src->start));
+		BUG_ON(t->entries[i].end	!= le64_to_cpu(src->end));
+
+		if (t->entries[i].dirty)
+			*dst++ = *src;
+	}
+
+	new_nr = dst - bl->start;
+
+	bch_info(c, "nr blacklist entries was %u, now %u", nr, new_nr);
+
+	if (new_nr != nr) {
+		bl = bch2_sb_resize_journal_seq_blacklist(&c->disk_sb,
+				new_nr ? sb_blacklist_u64s(new_nr) : 0);
+		BUG_ON(new_nr && !bl);
+
+		if (!new_nr)
+			c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_journal_seq_blacklist_v3));
+
+		bch2_write_super(c);
+	}
+out:
+	mutex_unlock(&c->sb_lock);
+}

fs/bcachefs/journal_seq_blacklist.h

@@ -17,4 +17,6 @@ int bch2_blacklist_table_initialize(struct bch_fs *);
 
 extern const struct bch_sb_field_ops bch_sb_field_ops_journal_seq_blacklist;
 
+void bch2_blacklist_entries_gc(struct work_struct *);
+
 #endif /* _BCACHEFS_JOURNAL_SEQ_BLACKLIST_H */

fs/bcachefs/recovery.c

@@ -1065,6 +1065,16 @@ int bch2_fs_recovery(struct bch_fs *c)
 	if (ret)
 		goto err;
 
+	/*
+	 * After an unclean shutdown, skip then next few journal sequence
+	 * numbers as they may have been referenced by btree writes that
+	 * happened before their corresponding journal writes - those btree
+	 * writes need to be ignored, by skipping and blacklisting the next few
+	 * journal sequence numbers:
+	 */
+	if (!c->sb.clean)
+		journal_seq += 8;
+
 	if (blacklist_seq != journal_seq) {
 		ret = bch2_journal_seq_blacklist_add(c,
 					blacklist_seq, journal_seq);
@@ -1210,14 +1220,6 @@ int bch2_fs_recovery(struct bch_fs *c)
 	}
 
 	mutex_lock(&c->sb_lock);
-	/*
-	 * With journal replay done, we can clear the journal seq blacklist
-	 * table:
-	 */
-	BUG_ON(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags));
-	if (le16_to_cpu(c->sb.version_min) >= bcachefs_metadata_version_btree_ptr_sectors_written)
-		bch2_sb_resize_journal_seq_blacklist(&c->disk_sb, 0);
-
 	if (c->opts.version_upgrade) {
 		c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
 		c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
@@ -1259,6 +1261,10 @@ int bch2_fs_recovery(struct bch_fs *c)
 		bch_info(c, "scanning for old btree nodes done");
 	}
 
+	if (c->journal_seq_blacklist_table &&
+	    c->journal_seq_blacklist_table->nr > 128)
+		queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
+
 	ret = 0;
 out:
 	set_bit(BCH_FS_FSCK_DONE, &c->flags);

fs/bcachefs/super.c

@@ -535,6 +535,8 @@ void __bch2_fs_stop(struct bch_fs *c)
 
 	set_bit(BCH_FS_STOPPING, &c->flags);
 
+	cancel_work_sync(&c->journal_seq_blacklist_gc_work);
+
 	down_write(&c->state_lock);
 	bch2_fs_read_only(c);
 	up_write(&c->state_lock);
@@ -698,6 +700,9 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 
 	spin_lock_init(&c->btree_write_error_lock);
 
+	INIT_WORK(&c->journal_seq_blacklist_gc_work,
+		  bch2_blacklist_entries_gc);
+
 	INIT_LIST_HEAD(&c->journal_entries);
 	INIT_LIST_HEAD(&c->journal_iters);