1 files changed, 47 insertions, 45 deletions

diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index 39c2db68123b..239fcc3c7c99 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -99,7 +99,7 @@ static void btree_node_interior_verify(struct bch_fs *c, struct btree *b)
 
 /* Calculate ideal packed bkey format for new btree nodes: */
 
-void __bch2_btree_calc_format(struct bkey_format_state *s, struct btree *b)
+static void __bch2_btree_calc_format(struct bkey_format_state *s, struct btree *b)
 {
 	struct bkey_packed *k;
 	struct bset_tree *t;
@@ -125,21 +125,20 @@ static struct bkey_format bch2_btree_calc_format(struct btree *b)
 	return bch2_bkey_format_done(&s);
 }
 
-static size_t btree_node_u64s_with_format(struct btree *b,
+static size_t btree_node_u64s_with_format(struct btree_nr_keys nr,
+					  struct bkey_format *old_f,
 					  struct bkey_format *new_f)
 {
-	struct bkey_format *old_f = &b->format;
-
 	/* stupid integer promotion rules */
 	ssize_t delta =
 	    (((int) new_f->key_u64s - old_f->key_u64s) *
-	     (int) b->nr.packed_keys) +
+	     (int) nr.packed_keys) +
 	    (((int) new_f->key_u64s - BKEY_U64s) *
-	     (int) b->nr.unpacked_keys);
+	     (int) nr.unpacked_keys);
 
-	BUG_ON(delta + b->nr.live_u64s < 0);
+	BUG_ON(delta + nr.live_u64s < 0);
 
-	return b->nr.live_u64s + delta;
+	return nr.live_u64s + delta;
 }
 
 /**
@@ -147,16 +146,18 @@ static size_t btree_node_u64s_with_format(struct btree *b,
  *
  * @c:		filesystem handle
  * @b:		btree node to rewrite
+ * @nr:		number of keys for new node (i.e. b->nr)
  * @new_f:	bkey format to translate keys to
  *
  * Returns: true if all re-packed keys will be able to fit in a new node.
  *
  * Assumes all keys will successfully pack with the new format.
  */
-bool bch2_btree_node_format_fits(struct bch_fs *c, struct btree *b,
+static bool bch2_btree_node_format_fits(struct bch_fs *c, struct btree *b,
+				 struct btree_nr_keys nr,
 				 struct bkey_format *new_f)
 {
-	size_t u64s = btree_node_u64s_with_format(b, new_f);
+	size_t u64s = btree_node_u64s_with_format(nr, &b->format, new_f);
 
 	return __vstruct_bytes(struct btree_node, u64s) < btree_bytes(c);
 }
@@ -391,7 +392,7 @@ static struct btree *bch2_btree_node_alloc_replacement(struct btree_update *as,
 	 * The keys might expand with the new format - if they wouldn't fit in
 	 * the btree node anymore, use the old format for now:
 	 */
-	if (!bch2_btree_node_format_fits(as->c, b, &format))
+	if (!bch2_btree_node_format_fits(as->c, b, b->nr, &format))
 		format = b->format;
 
 	SET_BTREE_NODE_SEQ(n->data, BTREE_NODE_SEQ(b->data) + 1);
@@ -513,8 +514,6 @@ static void bch2_btree_update_free(struct btree_update *as, struct btree_trans *
 		up_read(&c->gc_lock);
 	as->took_gc_lock = false;
 
-	bch2_journal_preres_put(&c->journal, &as->journal_preres);
-
 	bch2_journal_pin_drop(&c->journal, &as->journal);
 	bch2_journal_pin_flush(&c->journal, &as->journal);
 	bch2_disk_reservation_put(c, &as->disk_res);
@@ -734,8 +733,6 @@ err:
 
 	bch2_journal_pin_drop(&c->journal, &as->journal);
 
-	bch2_journal_preres_put(&c->journal, &as->journal_preres);
-
 	mutex_lock(&c->btree_interior_update_lock);
 	for (i = 0; i < as->nr_new_nodes; i++) {
 		b = as->new_nodes[i];
@@ -782,9 +779,9 @@ static void btree_interior_update_work(struct work_struct *work)
 	}
 }
 
-static void btree_update_set_nodes_written(struct closure *cl)
+static CLOSURE_CALLBACK(btree_update_set_nodes_written)
 {
-	struct btree_update *as = container_of(cl, struct btree_update, cl);
+	closure_type(as, struct btree_update, cl);
 	struct bch_fs *c = as->c;
 
 	mutex_lock(&c->btree_interior_update_lock);
@@ -1047,7 +1044,6 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	unsigned nr_nodes[2] = { 0, 0 };
 	unsigned update_level = level;
 	enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
-	unsigned journal_flags = 0;
 	int ret = 0;
 	u32 restart_count = trans->restart_count;
 
@@ -1061,9 +1057,16 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	flags &= ~BCH_WATERMARK_MASK;
 	flags |= watermark;
 
-	if (flags & BTREE_INSERT_JOURNAL_RECLAIM)
-		journal_flags |= JOURNAL_RES_GET_NONBLOCK;
-	journal_flags |= watermark;
+	if (!(flags & BTREE_INSERT_JOURNAL_RECLAIM) &&
+	    watermark < c->journal.watermark) {
+		struct journal_res res = { 0 };
+
+		ret = drop_locks_do(trans,
+			bch2_journal_res_get(&c->journal, &res, 1,
+					     watermark|JOURNAL_RES_GET_CHECK));
+		if (ret)
+			return ERR_PTR(ret);
+	}
 
 	while (1) {
 		nr_nodes[!!update_level] += 1 + split;
@@ -1080,8 +1083,12 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 			break;
 		}
 
+		/*
+		 * Always check for space for two keys, even if we won't have to
+		 * split at prior level - it might have been a merge instead:
+		 */
 		if (bch2_btree_node_insert_fits(c, path->l[update_level].b,
-					BKEY_BTREE_PTR_U64s_MAX * (1 + split)))
+						BKEY_BTREE_PTR_U64s_MAX * 2))
 			break;
 
 		split = path->l[update_level].b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c);
@@ -1129,27 +1136,6 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
 	if (ret)
 		goto err;
 
-	ret = bch2_journal_preres_get(&c->journal, &as->journal_preres,
-				      BTREE_UPDATE_JOURNAL_RES,
-				      journal_flags|JOURNAL_RES_GET_NONBLOCK);
-	if (ret) {
-		if (flags & BTREE_INSERT_JOURNAL_RECLAIM) {
-			ret = -BCH_ERR_journal_reclaim_would_deadlock;
-			goto err;
-		}
-
-		ret = drop_locks_do(trans,
-			bch2_journal_preres_get(&c->journal, &as->journal_preres,
-					      BTREE_UPDATE_JOURNAL_RES,
-					      journal_flags));
-		if (ret == -BCH_ERR_journal_preres_get_blocked) {
-			trace_and_count(c, trans_restart_journal_preres_get, trans, _RET_IP_, journal_flags);
-			ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_journal_preres_get);
-		}
-		if (ret)
-			goto err;
-	}
-
 	ret = bch2_disk_reservation_get(c, &as->disk_res,
 			(nr_nodes[0] + nr_nodes[1]) * btree_sectors(c),
 			c->opts.metadata_replicas,
@@ -1360,8 +1346,11 @@ static void __btree_split_node(struct btree_update *as,
 	struct bkey_packed *out[2];
 	struct bkey uk;
 	unsigned u64s, n1_u64s = (b->nr.live_u64s * 3) / 5;
+	struct { unsigned nr_keys, val_u64s; } nr_keys[2];
 	int i;
 
+	memset(&nr_keys, 0, sizeof(nr_keys));
+
 	for (i = 0; i < 2; i++) {
 		BUG_ON(n[i]->nsets != 1);
 
@@ -1383,6 +1372,9 @@ static void __btree_split_node(struct btree_update *as,
 		if (!i)
 			n1_pos = uk.p;
 		bch2_bkey_format_add_key(&format[i], &uk);
+
+		nr_keys[i].nr_keys++;
+		nr_keys[i].val_u64s += bkeyp_val_u64s(&b->format, k);
 	}
 
 	btree_set_min(n[0], b->data->min_key);
@@ -1395,6 +1387,12 @@ static void __btree_split_node(struct btree_update *as,
 		bch2_bkey_format_add_pos(&format[i], n[i]->data->max_key);
 
 		n[i]->data->format = bch2_bkey_format_done(&format[i]);
+
+		unsigned u64s = nr_keys[i].nr_keys * n[i]->data->format.key_u64s +
+			nr_keys[i].val_u64s;
+		if (__vstruct_bytes(struct btree_node, u64s) > btree_bytes(as->c))
+			n[i]->data->format = b->format;
+
 		btree_node_set_format(n[i], n[i]->data->format);
 	}
 
@@ -1837,8 +1835,8 @@ int __bch2_foreground_maybe_merge(struct btree_trans *trans,
 	bch2_bkey_format_add_pos(&new_s, next->data->max_key);
 	new_f = bch2_bkey_format_done(&new_s);
 
-	sib_u64s = btree_node_u64s_with_format(b, &new_f) +
-		btree_node_u64s_with_format(m, &new_f);
+	sib_u64s = btree_node_u64s_with_format(b->nr, &b->format, &new_f) +
+		btree_node_u64s_with_format(m->nr, &m->format, &new_f);
 
 	if (sib_u64s > BTREE_FOREGROUND_MERGE_HYSTERESIS(c)) {
 		sib_u64s -= BTREE_FOREGROUND_MERGE_HYSTERESIS(c);
@@ -2296,6 +2294,10 @@ int bch2_btree_node_update_key_get_iter(struct btree_trans *trans,
 
 	BUG_ON(!btree_node_hashed(b));
 
+	struct bch_extent_ptr *ptr;
+	bch2_bkey_drop_ptrs(bkey_i_to_s(new_key), ptr,
+			    !bch2_bkey_has_device(bkey_i_to_s(&b->key), ptr->dev));
+
 	ret = bch2_btree_node_update_key(trans, &iter, b, new_key,
 					 commit_flags, skip_triggers);
 out: