bcachefs: Centralize marking of replicas in btree update path

6 files changed, 41 insertions, 52 deletions

diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index 44349159dfbe..a7eda1141108 100644
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -439,11 +439,11 @@ enum btree_insert_ret {
 	BTREE_INSERT_OK,
 	/* extent spanned multiple leaf nodes: have to traverse to next node: */
 	BTREE_INSERT_NEED_TRAVERSE,
-	/* write lock held for too long */
 	/* leaf node needs to be split */
 	BTREE_INSERT_BTREE_NODE_FULL,
 	BTREE_INSERT_ENOSPC,
 	BTREE_INSERT_NEED_GC_LOCK,
+	BTREE_INSERT_NEED_MARK_REPLICAS,
 };
 
 enum btree_gc_coalesce_fail_reason {
diff --git a/fs/bcachefs/btree_update_leaf.c b/fs/bcachefs/btree_update_leaf.c
index eab685a4db31..913d74fb900b 100644
--- a/fs/bcachefs/btree_update_leaf.c
+++ b/fs/bcachefs/btree_update_leaf.c
@@ -11,6 +11,7 @@
 #include "journal.h"
 #include "journal_reclaim.h"
 #include "keylist.h"
+#include "replicas.h"
 
 #include <linux/sort.h>
 #include <trace/events/bcachefs.h>
@@ -300,8 +301,8 @@ static inline int btree_trans_cmp(struct btree_insert_entry l,
 
 static enum btree_insert_ret
 btree_key_can_insert(struct btree_insert *trans,
-		      struct btree_insert_entry *insert,
-		      unsigned *u64s)
+		     struct btree_insert_entry *insert,
+		     unsigned *u64s)
 {
 	struct bch_fs *c = trans->c;
 	struct btree *b = insert->iter->l[0].b;
@@ -310,6 +311,11 @@ btree_key_can_insert(struct btree_insert *trans,
 	if (unlikely(btree_node_fake(b)))
 		return BTREE_INSERT_BTREE_NODE_FULL;
 
+	if (!bch2_bkey_replicas_marked(c,
+			insert->iter->btree_id,
+			bkey_i_to_s_c(insert->k)))
+		return BTREE_INSERT_NEED_MARK_REPLICAS;
+
 	ret = !btree_node_is_extents(b)
 		? BTREE_INSERT_OK
 		: bch2_extent_can_insert(trans, insert, u64s);
@@ -326,8 +332,7 @@ btree_key_can_insert(struct btree_insert *trans,
  * Get journal reservation, take write locks, and attempt to do btree update(s):
  */
 static inline int do_btree_insert_at(struct btree_insert *trans,
-				     struct btree_iter **split,
-				     bool *cycle_gc_lock)
+				     struct btree_insert_entry **stopped_at)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_insert_entry *i;
@@ -371,22 +376,10 @@ static inline int do_btree_insert_at(struct btree_insert *trans,
 			u64s = 0;
 
 		u64s += i->k->k.u64s;
-		switch (btree_key_can_insert(trans, i, &u64s)) {
-		case BTREE_INSERT_OK:
-			break;
-		case BTREE_INSERT_BTREE_NODE_FULL:
-			ret = -EINTR;
-			*split = i->iter;
-			goto out;
-		case BTREE_INSERT_ENOSPC:
-			ret = -ENOSPC;
+		ret = btree_key_can_insert(trans, i, &u64s);
+		if (ret) {
+			*stopped_at = i;
 			goto out;
-		case BTREE_INSERT_NEED_GC_LOCK:
-			ret = -EINTR;
-			*cycle_gc_lock = true;
-			goto out;
-		default:
-			BUG();
 		}
 	}
 
@@ -444,8 +437,7 @@ int __bch2_btree_insert_at(struct btree_insert *trans)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_insert_entry *i;
-	struct btree_iter *linked, *split = NULL;
-	bool cycle_gc_lock = false;
+	struct btree_iter *linked;
 	unsigned flags;
 	int ret;
 
@@ -465,9 +457,6 @@ int __bch2_btree_insert_at(struct btree_insert *trans)
 	if (unlikely(!percpu_ref_tryget(&c->writes)))
 		return -EROFS;
 retry:
-	split = NULL;
-	cycle_gc_lock = false;
-
 	trans_for_each_entry(trans, i) {
 		unsigned old_locks_want = i->iter->locks_want;
 		unsigned old_uptodate = i->iter->uptodate;
@@ -485,7 +474,7 @@ retry:
 		}
 	}
 
-	ret = do_btree_insert_at(trans, &split, &cycle_gc_lock);
+	ret = do_btree_insert_at(trans, &i);
 	if (unlikely(ret))
 		goto err;
 
@@ -520,8 +509,9 @@ err:
 	if (!trans->did_work)
 		flags &= ~BTREE_INSERT_NOUNLOCK;
 
-	if (split) {
-		ret = bch2_btree_split_leaf(c, split, flags);
+	switch (ret) {
+	case BTREE_INSERT_BTREE_NODE_FULL:
+		ret = bch2_btree_split_leaf(c, i->iter, flags);
 
 		/*
 		 * if the split succeeded without dropping locks the insert will
@@ -546,9 +536,10 @@ err:
 			trans_restart(" (split)");
 			ret = -EINTR;
 		}
-	}
+		break;
+	case BTREE_INSERT_NEED_GC_LOCK:
+		ret = -EINTR;
 
-	if (cycle_gc_lock) {
 		if (!down_read_trylock(&c->gc_lock)) {
 			if (flags & BTREE_INSERT_NOUNLOCK)
 				goto out;
@@ -557,6 +548,24 @@ err:
 			down_read(&c->gc_lock);
 		}
 		up_read(&c->gc_lock);
+		break;
+	case BTREE_INSERT_ENOSPC:
+		ret = -ENOSPC;
+		break;
+	case BTREE_INSERT_NEED_MARK_REPLICAS:
+		if (flags & BTREE_INSERT_NOUNLOCK) {
+			ret = -EINTR;
+			goto out;
+		}
+
+		bch2_btree_iter_unlock(trans->entries[0].iter);
+		ret = bch2_mark_bkey_replicas(c, i->iter->btree_id,
+					      bkey_i_to_s_c(i->k))
+			?: -EINTR;
+		break;
+	default:
+		BUG_ON(ret >= 0);
+		break;
 	}
 
 	if (ret == -EINTR) {
diff --git a/fs/bcachefs/io.c b/fs/bcachefs/io.c
index eceb48656e60..34cab253c86d 100644
--- a/fs/bcachefs/io.c
+++ b/fs/bcachefs/io.c
@@ -22,7 +22,6 @@
 #include "keylist.h"
 #include "move.h"
 #include "rebalance.h"
-#include "replicas.h"
 #include "super.h"
 #include "super-io.h"
 
@@ -319,13 +318,6 @@ static void __bch2_write_index(struct bch_write_op *op)
 			goto err;
 		}
 
-		if (!(op->flags & BCH_WRITE_NOMARK_REPLICAS)) {
-			ret = bch2_mark_bkey_replicas(c, BKEY_TYPE_EXTENTS,
-						      e.s_c);
-			if (ret)
-				goto err;
-		}
-
 		dst = bkey_next(dst);
 	}
 
diff --git a/fs/bcachefs/io.h b/fs/bcachefs/io.h
index 5bd5f846122e..3ca779745422 100644
--- a/fs/bcachefs/io.h
+++ b/fs/bcachefs/io.h
@@ -30,10 +30,9 @@ enum bch_write_flags {
 	BCH_WRITE_PAGES_OWNED		= (1 << 5),
 	BCH_WRITE_ONLY_SPECIFIED_DEVS	= (1 << 6),
 	BCH_WRITE_NOPUT_RESERVATION	= (1 << 7),
-	BCH_WRITE_NOMARK_REPLICAS	= (1 << 8),
 
 	/* Internal: */
-	BCH_WRITE_JOURNAL_SEQ_PTR	= (1 << 9),
+	BCH_WRITE_JOURNAL_SEQ_PTR	= (1 << 8),
 };
 
 static inline u64 *op_journal_seq(struct bch_write_op *op)
diff --git a/fs/bcachefs/migrate.c b/fs/bcachefs/migrate.c
index c0dfe1c641e2..df4fbae2408a 100644
--- a/fs/bcachefs/migrate.c
+++ b/fs/bcachefs/migrate.c
@@ -71,11 +71,6 @@ static int bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 		 */
 		bch2_extent_normalize(c, e.s);
 
-		ret = bch2_mark_bkey_replicas(c, BKEY_TYPE_EXTENTS,
-					      bkey_i_to_s_c(&tmp.key));
-		if (ret)
-			break;
-
 		iter.pos = bkey_start_pos(&tmp.key.k);
 
 		ret = bch2_btree_insert_at(c, NULL, NULL,
diff --git a/fs/bcachefs/move.c b/fs/bcachefs/move.c
index e93725bf9b39..885792bd4ee3 100644
--- a/fs/bcachefs/move.c
+++ b/fs/bcachefs/move.c
@@ -150,11 +150,6 @@ static int bch2_migrate_index_update(struct bch_write_op *op)
 			goto next;
 		}
 
-		ret = bch2_mark_bkey_replicas(c, BKEY_TYPE_EXTENTS,
-					      extent_i_to_s_c(insert).s_c);
-		if (ret)
-			break;
-
 		ret = bch2_btree_insert_at(c, &op->res,
 				op_journal_seq(op),
 				BTREE_INSERT_ATOMIC|
@@ -239,8 +234,7 @@ int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
 	m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS|
 		BCH_WRITE_PAGES_STABLE|
 		BCH_WRITE_PAGES_OWNED|
-		BCH_WRITE_DATA_ENCODED|
-		BCH_WRITE_NOMARK_REPLICAS;
+		BCH_WRITE_DATA_ENCODED;
 
 	m->op.nr_replicas	= 1;
 	m->op.nr_replicas_required = 1;