bcachefs: Fix ec_stripes_read()

Change it to not mark keys that will be overwritten by keys in the
journal - this fixes a bug where we pop an assertion in
bucket_set_stripe() because of a stale pointer - because the stripe that
has the stale pointer has been deleted.

This code could be factored out and used elsewhere, at some point.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/ec.c

@@ -1267,10 +1267,10 @@ int bch2_stripes_write(struct bch_fs *c, unsigned flags, bool *wrote)
 
 int bch2_stripes_read(struct bch_fs *c, struct journal_keys *journal_keys)
 {
-	struct journal_key *i;
 	struct btree_trans trans;
-	struct btree_iter *iter;
-	struct bkey_s_c k;
+	struct btree_iter *btree_iter;
+	struct journal_iter journal_iter;
+	struct bkey_s_c btree_k, journal_k, k;
 	int ret;
 
 	ret = bch2_fs_ec_start(c);
@@ -1279,10 +1279,41 @@ int bch2_stripes_read(struct bch_fs *c, struct journal_keys *journal_keys)
 
 	bch2_trans_init(&trans, c, 0, 0);
 
-	for_each_btree_key(&trans, iter, BTREE_ID_EC, POS_MIN, 0, k, ret)
+	btree_iter	= bch2_trans_get_iter(&trans, BTREE_ID_EC, POS_MIN, 0);
+	journal_iter	= bch2_journal_iter_init(journal_keys, BTREE_ID_EC);
+
+	btree_k		= bch2_btree_iter_peek(btree_iter);
+	journal_k	= bch2_journal_iter_peek(&journal_iter);
+
+	while (1) {
+		if (btree_k.k && journal_k.k) {
+			int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
+
+			if (cmp < 0) {
+				k = btree_k;
+				btree_k = bch2_btree_iter_next(btree_iter);
+			} else if (cmp == 0) {
+				btree_k = bch2_btree_iter_next(btree_iter);
+				k = journal_k;
+				journal_k = bch2_journal_iter_next(&journal_iter);
+			} else {
+				k = journal_k;
+				journal_k = bch2_journal_iter_next(&journal_iter);
+			}
+		} else if (btree_k.k) {
+			k = btree_k;
+			btree_k = bch2_btree_iter_next(btree_iter);
+		} else if (journal_k.k) {
+			k = journal_k;
+			journal_k = bch2_journal_iter_next(&journal_iter);
+		} else {
+			break;
+		}
+
 		bch2_mark_key(c, k, 0, NULL, 0,
 			      BCH_BUCKET_MARK_ALLOC_READ|
 			      BCH_BUCKET_MARK_NOATOMIC);
+	}
 
 	ret = bch2_trans_exit(&trans) ?: ret;
 	if (ret) {
@@ -1290,13 +1321,6 @@ int bch2_stripes_read(struct bch_fs *c, struct journal_keys *journal_keys)
 		return ret;
 	}
 
-	for_each_journal_key(*journal_keys, i)
-		if (i->btree_id == BTREE_ID_EC)
-			bch2_mark_key(c, bkey_i_to_s_c(i->k),
-				      0, NULL, 0,
-				      BCH_BUCKET_MARK_ALLOC_READ|
-				      BCH_BUCKET_MARK_NOATOMIC);
-
 	return 0;
 }
 

fs/bcachefs/recovery.c

@@ -24,6 +24,42 @@
 
 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
 
+/* iterate over keys read from the journal: */
+
+struct journal_iter bch2_journal_iter_init(struct journal_keys *keys,
+					   enum btree_id id)
+{
+	return (struct journal_iter) {
+		.keys		= keys,
+		.k		= keys->d,
+		.btree_id	= id,
+	};
+}
+
+struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
+{
+	while (1) {
+		if (iter->k == iter->keys->d + iter->keys->nr)
+			return bkey_s_c_null;
+
+		if (iter->k->btree_id == iter->btree_id)
+			return bkey_i_to_s_c(iter->k->k);
+
+		iter->k++;
+	}
+
+	return bkey_s_c_null;
+}
+
+struct bkey_s_c bch2_journal_iter_next(struct journal_iter *iter)
+{
+	if (iter->k == iter->keys->d + iter->keys->nr)
+		return bkey_s_c_null;
+
+	iter->k++;
+	return bch2_journal_iter_peek(iter);
+}
+
 /* sort and dedup all keys in the journal: */
 
 static void journal_entries_free(struct list_head *list)

fs/bcachefs/recovery.h

@@ -18,6 +18,17 @@ struct journal_keys {
 #define for_each_journal_key(keys, i)				\
 	for (i = (keys).d; i < (keys).d + (keys).nr; (i)++)
 
+struct journal_iter {
+	struct journal_keys	*keys;
+	struct journal_key	*k;
+	enum btree_id		btree_id;
+};
+
+struct journal_iter bch2_journal_iter_init(struct journal_keys *,
+					   enum btree_id);
+struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *);
+struct bkey_s_c bch2_journal_iter_next(struct journal_iter *);
+
 int bch2_fs_recovery(struct bch_fs *);
 int bch2_fs_initialize(struct bch_fs *);