Update bcachefs sources to c68fda3e8f bcachefs: Fix btree node read error path

4 files changed, 75 insertions, 93 deletions

diff --git a/.bcachefs_revision b/.bcachefs_revision
index 5eb25511..0e58a6dd 100644
--- a/.bcachefs_revision
+++ b/.bcachefs_revision
@@ -1 +1 @@
-600598598b7c6d2069a374a14ad4925f39a30faa
+c68fda3e8f4106f1e1881e650f702d5bc2d301cb
diff --git a/libbcachefs/btree_io.c b/libbcachefs/btree_io.c
index 4629617b..a04d48ae 100644
--- a/libbcachefs/btree_io.c
+++ b/libbcachefs/btree_io.c
@@ -1488,23 +1488,32 @@ void bch2_btree_node_read(struct bch_fs *c, struct btree *b,
 	struct btree_read_bio *rb;
 	struct bch_dev *ca;
 	struct bio *bio;
-	struct printbuf buf = PRINTBUF;
 	int ret;
 
-	btree_pos_to_text(&buf, c, b);
 	trace_btree_read(c, b);
 
 	if (bch2_verify_all_btree_replicas &&
 	    !btree_node_read_all_replicas(c, b, sync))
-		goto out;
+		return;
 
 	ret = bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key),
 					 NULL, &pick);
-	if (bch2_fs_fatal_err_on(ret <= 0, c,
-			"btree node read error: no device to read from\n"
-			" at %s", buf.buf)) {
+
+	if (ret <= 0) {
+		struct printbuf buf = PRINTBUF;
+
+		pr_buf(&buf, "btree node read error: no device to read from\n at ");
+		btree_pos_to_text(&buf, c, b);
+		bch_err(c, "%s", buf.buf);
+
+		if (test_bit(BCH_FS_TOPOLOGY_REPAIR_DONE, &c->flags))
+			bch2_fatal_error(c);
+
 		set_btree_node_read_error(b);
-		goto out;
+		clear_btree_node_read_in_flight(b);
+		wake_up_bit(&b->flags, BTREE_NODE_read_in_flight);
+		printbuf_exit(&buf);
+		return;
 	}
 
 	ca = bch_dev_bkey_exists(c, pick.ptr.dev);
@@ -1546,8 +1555,6 @@ void bch2_btree_node_read(struct bch_fs *c, struct btree *b,
 		else
 			queue_work(c->io_complete_wq, &rb->work);
 	}
-out:
-	printbuf_exit(&buf);
 }
 
 int bch2_btree_root_read(struct bch_fs *c, enum btree_id id,
diff --git a/libbcachefs/recovery.c b/libbcachefs/recovery.c
index 04afe245..1f14c0b6 100644
--- a/libbcachefs/recovery.c
+++ b/libbcachefs/recovery.c
@@ -153,21 +153,17 @@ static void journal_iters_fix(struct bch_fs *c)
 {
 	struct journal_keys *keys = &c->journal_keys;
 	/* The key we just inserted is immediately before the gap: */
-	struct journal_key *n = &keys->d[keys->gap - 1];
 	size_t gap_end = keys->gap + (keys->size - keys->nr);
 	struct btree_and_journal_iter *iter;
 
 	/*
-	 * If an iterator points one after the key we just inserted,
-	 * and the key we just inserted compares > the iterator's position,
-	 * decrement the iterator so it points at the key we just inserted:
+	 * If an iterator points one after the key we just inserted, decrement
+	 * the iterator so it points at the key we just inserted - if the
+	 * decrement was unnecessary, bch2_btree_and_journal_iter_peek() will
+	 * handle that:
 	 */
 	list_for_each_entry(iter, &c->journal_iters, journal.list)
-		if (iter->journal.idx == gap_end &&
-		    iter->last &&
-		    iter->b->c.btree_id == n->btree_id &&
-		    iter->b->c.level	== n->level &&
-		    bpos_cmp(n->k->k.p, iter->unpacked.p) > 0)
+		if (iter->journal.idx == gap_end)
 			iter->journal.idx = keys->gap - 1;
 }
 
@@ -297,30 +293,30 @@ void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 		keys->d[idx].overwritten = true;
 }
 
-static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
+static void bch2_journal_iter_advance(struct journal_iter *iter)
+{
+	if (iter->idx < iter->keys->size) {
+		iter->idx++;
+		if (iter->idx == iter->keys->gap)
+			iter->idx += iter->keys->size - iter->keys->nr;
+	}
+}
+
+struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
 {
 	struct journal_key *k = iter->keys->d + iter->idx;
 
-	while (k < iter->keys->d + iter->keys->nr &&
+	while (k < iter->keys->d + iter->keys->size &&
 	       k->btree_id	== iter->btree_id &&
 	       k->level		== iter->level) {
 		if (!k->overwritten)
-			return k->k;
+			return bkey_i_to_s_c(k->k);
 
-		iter->idx++;
+		bch2_journal_iter_advance(iter);
 		k = iter->keys->d + iter->idx;
 	}
 
-	return NULL;
-}
-
-static void bch2_journal_iter_advance(struct journal_iter *iter)
-{
-	if (iter->idx < iter->keys->size) {
-		iter->idx++;
-		if (iter->idx == iter->keys->gap)
-			iter->idx += iter->keys->size - iter->keys->nr;
-	}
+	return bkey_s_c_null;
 }
 
 static void bch2_journal_iter_exit(struct journal_iter *iter)
@@ -352,71 +348,49 @@ static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 
 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 {
-	switch (iter->last) {
-	case none:
-		break;
-	case btree:
-		bch2_journal_iter_advance_btree(iter);
-		break;
-	case journal:
-		bch2_journal_iter_advance(&iter->journal);
-		break;
-	}
-
-	iter->last = none;
+	if (!bpos_cmp(iter->pos, SPOS_MAX))
+		iter->at_end = true;
+	else
+		iter->pos = bpos_successor(iter->pos);
 }
 
 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 {
-	struct bkey_s_c ret;
+	struct bkey_s_c btree_k, journal_k, ret;
+again:
+	if (iter->at_end)
+		return bkey_s_c_null;
 
-	while (1) {
-		struct bkey_s_c btree_k		=
-			bch2_journal_iter_peek_btree(iter);
-		struct bkey_s_c journal_k	=
-			bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
-
-		if (btree_k.k && journal_k.k) {
-			int cmp = bpos_cmp(btree_k.k->p, journal_k.k->p);
+	while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
+	       bpos_cmp(btree_k.k->p, iter->pos) < 0)
+		bch2_journal_iter_advance_btree(iter);
 
-			if (!cmp)
-				bch2_journal_iter_advance_btree(iter);
+	while ((journal_k = bch2_journal_iter_peek(&iter->journal)).k &&
+	       bpos_cmp(journal_k.k->p, iter->pos) < 0)
+		bch2_journal_iter_advance(&iter->journal);
 
-			iter->last = cmp < 0 ? btree : journal;
-		} else if (btree_k.k) {
-			iter->last = btree;
-		} else if (journal_k.k) {
-			iter->last = journal;
-		} else {
-			iter->last = none;
-			return bkey_s_c_null;
-		}
+	ret = journal_k.k &&
+		(!btree_k.k || bpos_cmp(journal_k.k->p, btree_k.k->p) <= 0)
+		? journal_k
+		: btree_k;
 
-		ret = iter->last == journal ? journal_k : btree_k;
+	if (ret.k && iter->b && bpos_cmp(ret.k->p, iter->b->data->max_key) > 0)
+		ret = bkey_s_c_null;
 
-		if (iter->b &&
-		    bpos_cmp(ret.k->p, iter->b->data->max_key) > 0) {
-			iter->journal.idx = iter->journal.keys->nr;
-			iter->last = none;
-			return bkey_s_c_null;
+	if (ret.k) {
+		iter->pos = ret.k->p;
+		if (bkey_deleted(ret.k)) {
+			bch2_btree_and_journal_iter_advance(iter);
+			goto again;
 		}
-
-		if (!bkey_deleted(ret.k))
-			break;
-
-		bch2_btree_and_journal_iter_advance(iter);
+	} else {
+		iter->pos = SPOS_MAX;
+		iter->at_end = true;
 	}
 
 	return ret;
 }
 
-struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
-{
-	bch2_btree_and_journal_iter_advance(iter);
-
-	return bch2_btree_and_journal_iter_peek(iter);
-}
-
 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 {
 	bch2_journal_iter_exit(&iter->journal);
@@ -434,6 +408,8 @@ void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter
 	iter->node_iter = node_iter;
 	bch2_journal_iter_init(c, &iter->journal, b->c.btree_id, b->c.level, pos);
 	INIT_LIST_HEAD(&iter->journal.list);
+	iter->pos = b->data->min_key;
+	iter->at_end = false;
 }
 
 /*
@@ -1169,9 +1145,6 @@ use_clean:
 		blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 	}
 
-	if (c->opts.read_journal_only)
-		goto out;
-
 	if (c->opts.reconstruct_alloc) {
 		c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 		drop_alloc_keys(&c->journal_keys);
@@ -1202,6 +1175,13 @@ use_clean:
 		}
 	}
 
+	/*
+	 * note: cmd_list_journal needs the blacklist table fully up to date so
+	 * it can asterisk ignored journal entries:
+	 */
+	if (c->opts.read_journal_only)
+		goto out;
+
 	ret = bch2_fs_journal_start(&c->journal, journal_seq);
 	if (ret)
 		goto err;
diff --git a/libbcachefs/recovery.h b/libbcachefs/recovery.h
index 52db06b2..8c0348e8 100644
--- a/libbcachefs/recovery.h
+++ b/libbcachefs/recovery.h
@@ -20,12 +20,8 @@ struct btree_and_journal_iter {
 	struct bkey		unpacked;
 
 	struct journal_iter	journal;
-
-	enum last_key_returned {
-		none,
-		btree,
-		journal,
-	}			last;
+	struct bpos		pos;
+	bool			at_end;
 };
 
 struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *, enum btree_id,
@@ -44,7 +40,6 @@ void bch2_journal_key_overwritten(struct bch_fs *, enum btree_id,
 
 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *);
 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *);
-struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *);
 
 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *);
 void __bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *,