bcachefs: Further reduce iter->trans usage

This is prep work for splitting btree_path out from btree_iter -
btree_path will not have a pointer to btree_trans.

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

11 files changed, 138 insertions, 133 deletions

diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c
index cd0c5009e167..64a731cfc01c 100644
--- a/fs/bcachefs/btree_cache.c
+++ b/fs/bcachefs/btree_cache.c
@@ -632,6 +632,7 @@ err:
 
 /* Slowpath, don't want it inlined into btree_iter_traverse() */
 static noinline struct btree *bch2_btree_node_fill(struct bch_fs *c,
+				struct btree_trans *trans,
 				struct btree_iter *iter,
 				const struct bkey_i *k,
 				enum btree_id btree_id,
@@ -648,8 +649,8 @@ static noinline struct btree *bch2_btree_node_fill(struct bch_fs *c,
 	 * Parent node must be locked, else we could read in a btree node that's
 	 * been freed:
 	 */
-	if (iter && !bch2_btree_node_relock(iter, level + 1)) {
-		btree_trans_restart(iter->trans);
+	if (trans && !bch2_btree_node_relock(trans, iter, level + 1)) {
+		btree_trans_restart(trans);
 		return ERR_PTR(-EINTR);
 	}
 
@@ -680,23 +681,23 @@ static noinline struct btree *bch2_btree_node_fill(struct bch_fs *c,
 	six_unlock_intent(&b->c.lock);
 
 	/* Unlock before doing IO: */
-	if (iter && sync)
-		bch2_trans_unlock(iter->trans);
+	if (trans && sync)
+		bch2_trans_unlock(trans);
 
 	bch2_btree_node_read(c, b, sync);
 
 	if (!sync)
 		return NULL;
 
-	if (iter &&
-	    (!bch2_trans_relock(iter->trans) ||
-	     !bch2_btree_iter_relock_intent(iter))) {
-		BUG_ON(!iter->trans->restarted);
+	if (trans &&
+	    (!bch2_trans_relock(trans) ||
+	     !bch2_btree_iter_relock_intent(trans, iter))) {
+		BUG_ON(!trans->restarted);
 		return ERR_PTR(-EINTR);
 	}
 
 	if (!six_relock_type(&b->c.lock, lock_type, seq)) {
-		btree_trans_restart(iter->trans);
+		btree_trans_restart(trans);
 		return ERR_PTR(-EINTR);
 	}
 
@@ -779,7 +780,7 @@ retry:
 		 * else we could read in a btree node from disk that's been
 		 * freed:
 		 */
-		b = bch2_btree_node_fill(c, iter, k, iter->btree_id,
+		b = bch2_btree_node_fill(c, trans, iter, k, iter->btree_id,
 					 level, lock_type, true);
 
 		/* We raced and found the btree node in the cache */
@@ -821,7 +822,7 @@ lock_node:
 		if (btree_node_read_locked(iter, level + 1))
 			btree_node_unlock(iter, level + 1);
 
-		if (!btree_node_lock(b, k->k.p, level, iter, lock_type,
+		if (!btree_node_lock(trans, iter, b, k->k.p, level, lock_type,
 				     lock_node_check_fn, (void *) k, trace_ip)) {
 			if (!trans->restarted)
 				goto retry;
@@ -832,7 +833,7 @@ lock_node:
 			     b->c.level != level ||
 			     race_fault())) {
 			six_unlock_type(&b->c.lock, lock_type);
-			if (bch2_btree_node_relock(iter, level + 1))
+			if (bch2_btree_node_relock(trans, iter, level + 1))
 				goto retry;
 
 			trace_trans_restart_btree_node_reused(trans->ip,
@@ -856,9 +857,9 @@ lock_node:
 		 * should_be_locked is not set on this iterator yet, so we need
 		 * to relock it specifically:
 		 */
-		if (iter &&
+		if (trans &&
 		    (!bch2_trans_relock(trans) ||
-		     !bch2_btree_iter_relock_intent(iter))) {
+		     !bch2_btree_iter_relock_intent(trans, iter))) {
 			BUG_ON(!trans->restarted);
 			return ERR_PTR(-EINTR);
 		}
@@ -917,7 +918,7 @@ retry:
 		if (nofill)
 			goto out;
 
-		b = bch2_btree_node_fill(c, NULL, k, btree_id,
+		b = bch2_btree_node_fill(c, NULL, NULL, k, btree_id,
 					 level, SIX_LOCK_read, true);
 
 		/* We raced and found the btree node in the cache */
@@ -975,21 +976,24 @@ out:
 	return b;
 }
 
-int bch2_btree_node_prefetch(struct bch_fs *c, struct btree_iter *iter,
+int bch2_btree_node_prefetch(struct bch_fs *c,
+			     struct btree_trans *trans,
+			     struct btree_iter *iter,
 			     const struct bkey_i *k,
 			     enum btree_id btree_id, unsigned level)
 {
 	struct btree_cache *bc = &c->btree_cache;
 	struct btree *b;
 
-	BUG_ON(iter && !btree_node_locked(iter, level + 1));
+	BUG_ON(trans && !btree_node_locked(iter, level + 1));
 	BUG_ON(level >= BTREE_MAX_DEPTH);
 
 	b = btree_cache_find(bc, k);
 	if (b)
 		return 0;
 
-	b = bch2_btree_node_fill(c, iter, k, btree_id, level, SIX_LOCK_read, false);
+	b = bch2_btree_node_fill(c, trans, iter, k, btree_id,
+				 level, SIX_LOCK_read, false);
 	return PTR_ERR_OR_ZERO(b);
 }
 
diff --git a/fs/bcachefs/btree_cache.h b/fs/bcachefs/btree_cache.h
index 5032293e8628..6c1c69f3abcf 100644
--- a/fs/bcachefs/btree_cache.h
+++ b/fs/bcachefs/btree_cache.h
@@ -29,8 +29,9 @@ struct btree *bch2_btree_node_get(struct btree_trans *, struct btree_iter *,
 struct btree *bch2_btree_node_get_noiter(struct bch_fs *, const struct bkey_i *,
 					 enum btree_id, unsigned, bool);
 
-int bch2_btree_node_prefetch(struct bch_fs *, struct btree_iter *,
-			     const struct bkey_i *, enum btree_id, unsigned);
+int bch2_btree_node_prefetch(struct bch_fs *, struct btree_trans *,
+			     struct btree_iter *, const struct bkey_i *,
+			     enum btree_id, unsigned);
 
 void bch2_btree_node_evict(struct bch_fs *, const struct bkey_i *);
 
diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index 5f69a63e1340..a2eca98e4fb5 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -108,17 +108,14 @@ void bch2_btree_node_unlock_write(struct btree_trans *trans,
 	bch2_btree_node_unlock_write_inlined(trans, iter, b);
 }
 
-void __bch2_btree_node_lock_write(struct btree_trans *trans,
-			struct btree_iter *iter, struct btree *b)
+void __bch2_btree_node_lock_write(struct btree_trans *trans, struct btree *b)
 {
-	struct btree_iter *linked;
+	struct btree_iter *iter;
 	unsigned readers = 0;
 
-	EBUG_ON(!btree_node_intent_locked(iter, b->c.level));
-
-	trans_for_each_iter(trans, linked)
-		if (linked->l[b->c.level].b == b &&
-		    btree_node_read_locked(linked, b->c.level))
+	trans_for_each_iter(trans, iter)
+		if (iter->l[b->c.level].b == b &&
+		    btree_node_read_locked(iter, b->c.level))
 			readers++;
 
 	/*
@@ -134,7 +131,8 @@ void __bch2_btree_node_lock_write(struct btree_trans *trans,
 		     &b->c.lock.state.counter);
 }
 
-bool __bch2_btree_node_relock(struct btree_iter *iter, unsigned level)
+bool __bch2_btree_node_relock(struct btree_trans *trans,
+			      struct btree_iter *iter, unsigned level)
 {
 	struct btree *b = btree_iter_node(iter, level);
 	int want = __btree_lock_want(iter, level);
@@ -147,7 +145,7 @@ bool __bch2_btree_node_relock(struct btree_iter *iter, unsigned level)
 
 	if (six_relock_type(&b->c.lock, want, iter->l[level].lock_seq) ||
 	    (btree_node_lock_seq_matches(iter, b, level) &&
-	     btree_node_lock_increment(iter->trans, b, level, want))) {
+	     btree_node_lock_increment(trans, b, level, want))) {
 		mark_btree_node_locked(iter, level, want);
 		return true;
 	} else {
@@ -155,7 +153,8 @@ bool __bch2_btree_node_relock(struct btree_iter *iter, unsigned level)
 	}
 }
 
-static bool bch2_btree_node_upgrade(struct btree_iter *iter, unsigned level)
+static bool bch2_btree_node_upgrade(struct btree_trans *trans,
+				    struct btree_iter *iter, unsigned level)
 {
 	struct btree *b = iter->l[level].b;
 
@@ -176,7 +175,7 @@ static bool bch2_btree_node_upgrade(struct btree_iter *iter, unsigned level)
 		goto success;
 
 	if (btree_node_lock_seq_matches(iter, b, level) &&
-	    btree_node_lock_increment(iter->trans, b, level, BTREE_NODE_INTENT_LOCKED)) {
+	    btree_node_lock_increment(trans, b, level, BTREE_NODE_INTENT_LOCKED)) {
 		btree_node_unlock(iter, level);
 		goto success;
 	}
@@ -199,8 +198,8 @@ static inline bool btree_iter_get_locks(struct btree_trans *trans,
 			break;
 
 		if (!(upgrade
-		      ? bch2_btree_node_upgrade(iter, l)
-		      : bch2_btree_node_relock(iter, l))) {
+		      ? bch2_btree_node_upgrade(trans, iter, l)
+		      : bch2_btree_node_relock(trans, iter, l))) {
 			(upgrade
 			 ? trace_node_upgrade_fail
 			 : trace_node_relock_fail)(trans->ip, trace_ip,
@@ -248,13 +247,13 @@ static struct bpos btree_node_pos(struct btree_bkey_cached_common *_b,
 }
 
 /* Slowpath: */
-bool __bch2_btree_node_lock(struct btree *b, struct bpos pos,
-			    unsigned level, struct btree_iter *iter,
+bool __bch2_btree_node_lock(struct btree_trans *trans,
+			    struct btree_iter *iter,
+			    struct btree *b, struct bpos pos, unsigned level,
 			    enum six_lock_type type,
 			    six_lock_should_sleep_fn should_sleep_fn, void *p,
 			    unsigned long ip)
 {
-	struct btree_trans *trans = iter->trans;
 	struct btree_iter *linked, *deadlock_iter = NULL;
 	u64 start_time = local_clock();
 	unsigned reason = 9;
@@ -361,16 +360,10 @@ bool __bch2_btree_node_lock(struct btree *b, struct bpos pos,
 /* Btree iterator locking: */
 
 #ifdef CONFIG_BCACHEFS_DEBUG
-static void bch2_btree_iter_verify_locks(struct btree_trans *trans,
-					 struct btree_iter *iter)
+static void bch2_btree_iter_verify_locks(struct btree_iter *iter)
 {
 	unsigned l;
 
-	if (!(trans->iters_linked & (1ULL << iter->idx))) {
-		BUG_ON(iter->nodes_locked);
-		return;
-	}
-
 	for (l = 0; btree_iter_node(iter, l); l++) {
 		if (iter->uptodate >= BTREE_ITER_NEED_RELOCK &&
 		    !btree_node_locked(iter, l))
@@ -386,25 +379,24 @@ void bch2_btree_trans_verify_locks(struct btree_trans *trans)
 	struct btree_iter *iter;
 
 	trans_for_each_iter(trans, iter)
-		bch2_btree_iter_verify_locks(trans, iter);
+		bch2_btree_iter_verify_locks(iter);
 }
 #else
-static inline void bch2_btree_iter_verify_locks(struct btree_trans *trans,
-						struct btree_iter *iter) {}
+static inline void bch2_btree_iter_verify_locks(struct btree_iter *iter) {}
 #endif
 
 /*
  * Only for btree_cache.c - only relocks intent locks
  */
-bool bch2_btree_iter_relock_intent(struct btree_iter *iter)
+bool bch2_btree_iter_relock_intent(struct btree_trans *trans,
+				   struct btree_iter *iter)
 {
-	struct btree_trans *trans = iter->trans;
 	unsigned l;
 
 	for (l = iter->level;
 	     l < iter->locks_want && btree_iter_node(iter, l);
 	     l++) {
-		if (!bch2_btree_node_relock(iter, l)) {
+		if (!bch2_btree_node_relock(trans, iter, l)) {
 			trace_node_relock_fail(trans->ip, _RET_IP_,
 					btree_iter_type(iter) == BTREE_ITER_CACHED,
 					iter->btree_id, &iter->real_pos,
@@ -435,10 +427,10 @@ static bool bch2_btree_iter_relock(struct btree_trans *trans,
 	return ret;
 }
 
-bool __bch2_btree_iter_upgrade(struct btree_iter *iter,
+bool __bch2_btree_iter_upgrade(struct btree_trans *trans,
+			       struct btree_iter *iter,
 			       unsigned new_locks_want)
 {
-	struct btree_trans *trans = iter->trans;
 	struct btree_iter *linked;
 
 	EBUG_ON(iter->locks_want >= new_locks_want);
@@ -503,7 +495,7 @@ void __bch2_btree_iter_downgrade(struct btree_iter *iter,
 		}
 	}
 
-	bch2_btree_trans_verify_locks(iter->trans);
+	bch2_btree_iter_verify_locks(iter);
 }
 
 void bch2_trans_downgrade(struct btree_trans *trans)
@@ -554,12 +546,13 @@ void bch2_trans_unlock(struct btree_trans *trans)
 
 #ifdef CONFIG_BCACHEFS_DEBUG
 
-static void bch2_btree_iter_verify_cached(struct btree_iter *iter)
+static void bch2_btree_iter_verify_cached(struct btree_trans *trans,
+					  struct btree_iter *iter)
 {
 	struct bkey_cached *ck;
 	bool locked = btree_node_locked(iter, 0);
 
-	if (!bch2_btree_node_relock(iter, 0))
+	if (!bch2_btree_node_relock(trans, iter, 0))
 		return;
 
 	ck = (void *) iter->l[0].b;
@@ -570,8 +563,8 @@ static void bch2_btree_iter_verify_cached(struct btree_iter *iter)
 		btree_node_unlock(iter, 0);
 }
 
-static void bch2_btree_iter_verify_level(struct btree_iter *iter,
-					 unsigned level)
+static void bch2_btree_iter_verify_level(struct btree_trans *trans,
+				struct btree_iter *iter, unsigned level)
 {
 	struct btree_iter_level *l;
 	struct btree_node_iter tmp;
@@ -589,7 +582,7 @@ static void bch2_btree_iter_verify_level(struct btree_iter *iter,
 
 	if (btree_iter_type(iter) == BTREE_ITER_CACHED) {
 		if (!level)
-			bch2_btree_iter_verify_cached(iter);
+			bch2_btree_iter_verify_cached(trans, iter);
 		return;
 	}
 
@@ -598,7 +591,7 @@ static void bch2_btree_iter_verify_level(struct btree_iter *iter,
 	if (!btree_iter_node(iter, level))
 		return;
 
-	if (!bch2_btree_node_relock(iter, level))
+	if (!bch2_btree_node_relock(trans, iter, level))
 		return;
 
 	BUG_ON(!btree_iter_pos_in_node(iter, l->b));
@@ -688,10 +681,10 @@ static void bch2_btree_iter_verify(struct btree_iter *iter)
 			break;
 		}
 
-		bch2_btree_iter_verify_level(iter, i);
+		bch2_btree_iter_verify_level(trans, iter, i);
 	}
 
-	bch2_btree_iter_verify_locks(trans, iter);
+	bch2_btree_iter_verify_locks(iter);
 }
 
 static void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter)
@@ -715,13 +708,15 @@ void bch2_btree_trans_verify_iters(struct btree_trans *trans, struct btree *b)
 		return;
 
 	trans_for_each_iter_with_node(trans, b, iter)
-		bch2_btree_iter_verify_level(iter, b->c.level);
+		bch2_btree_iter_verify_level(trans, iter, b->c.level);
 }
 
 #else
 
-static inline void bch2_btree_iter_verify_level(struct btree_iter *iter, unsigned l) {}
-static inline void bch2_btree_iter_verify(struct btree_iter *iter) {}
+static inline void bch2_btree_iter_verify_level(struct btree_trans *trans,
+					struct btree_iter *iter, unsigned l) {}
+static inline void bch2_btree_iter_verify(struct btree_trans *trans,
+					struct btree_iter *iter) {}
 static inline void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter) {}
 
 #endif
@@ -767,7 +762,7 @@ void bch2_btree_iter_fix_key_modified(struct btree_trans *trans,
 
 	trans_for_each_iter_with_node(trans, b, linked) {
 		__bch2_btree_iter_fix_key_modified(linked, b, where);
-		bch2_btree_iter_verify_level(linked, b->c.level);
+		bch2_btree_iter_verify_level(trans, linked, b->c.level);
 	}
 }
 
@@ -892,7 +887,7 @@ void bch2_btree_node_iter_fix(struct btree_trans *trans,
 		__bch2_btree_node_iter_fix(linked, b,
 					   &linked->l[b->c.level].iter, t,
 					   where, clobber_u64s, new_u64s);
-		bch2_btree_iter_verify_level(linked, b->c.level);
+		bch2_btree_iter_verify_level(trans, linked, b->c.level);
 	}
 }
 
@@ -977,7 +972,8 @@ static inline bool btree_iter_advance_to_pos(struct btree_iter *iter,
 /*
  * Verify that iterator for parent node points to child node:
  */
-static void btree_iter_verify_new_node(struct btree_iter *iter, struct btree *b)
+static void btree_iter_verify_new_node(struct btree_trans *trans,
+				       struct btree_iter *iter, struct btree *b)
 {
 	struct btree_iter_level *l;
 	unsigned plevel;
@@ -993,7 +989,7 @@ static void btree_iter_verify_new_node(struct btree_iter *iter, struct btree *b)
 
 	parent_locked = btree_node_locked(iter, plevel);
 
-	if (!bch2_btree_node_relock(iter, plevel))
+	if (!bch2_btree_node_relock(trans, iter, plevel))
 		return;
 
 	l = &iter->l[plevel];
@@ -1007,7 +1003,7 @@ static void btree_iter_verify_new_node(struct btree_iter *iter, struct btree *b)
 		char buf4[100];
 		struct bkey uk = bkey_unpack_key(b, k);
 
-		bch2_dump_btree_node(iter->trans->c, l->b);
+		bch2_dump_btree_node(trans->c, l->b);
 		bch2_bpos_to_text(&PBUF(buf1), iter->real_pos);
 		bch2_bkey_to_text(&PBUF(buf2), &uk);
 		bch2_bpos_to_text(&PBUF(buf3), b->data->min_key);
@@ -1024,8 +1020,8 @@ static void btree_iter_verify_new_node(struct btree_iter *iter, struct btree *b)
 		btree_node_unlock(iter, b->c.level + 1);
 }
 
-static inline void __btree_iter_init(struct btree_iter *iter,
-				     unsigned level)
+static inline void __btree_iter_level_init(struct btree_iter *iter,
+					   unsigned level)
 {
 	struct btree_iter_level *l = &iter->l[level];
 
@@ -1041,19 +1037,20 @@ static inline void __btree_iter_init(struct btree_iter *iter,
 	btree_iter_set_dirty(iter, BTREE_ITER_NEED_PEEK);
 }
 
-static inline void btree_iter_node_set(struct btree_iter *iter,
-				       struct btree *b)
+static inline void btree_iter_level_init(struct btree_trans *trans,
+					 struct btree_iter *iter,
+					 struct btree *b)
 {
 	BUG_ON(btree_iter_type(iter) == BTREE_ITER_CACHED);
 
-	btree_iter_verify_new_node(iter, b);
+	btree_iter_verify_new_node(trans, iter, b);
 
 	EBUG_ON(!btree_iter_pos_in_node(iter, b));
 	EBUG_ON(b->c.lock.state.seq & 1);
 
 	iter->l[b->c.level].lock_seq = b->c.lock.state.seq;
 	iter->l[b->c.level].b = b;
-	__btree_iter_init(iter, b->c.level);
+	__btree_iter_level_init(iter, b->c.level);
 }
 
 /*
@@ -1082,7 +1079,7 @@ void bch2_btree_iter_node_replace(struct btree_trans *trans,
 				mark_btree_node_locked(linked, b->c.level, t);
 			}
 
-			btree_iter_node_set(linked, b);
+			btree_iter_level_init(trans, linked, b);
 		}
 }
 
@@ -1109,7 +1106,7 @@ void bch2_btree_iter_reinit_node(struct btree_trans *trans,
 	struct btree_iter *linked;
 
 	trans_for_each_iter_with_node(trans, b, linked)
-		__btree_iter_init(linked, b->c.level);
+		__btree_iter_level_init(linked, b->c.level);
 }
 
 static int lock_root_check_fn(struct six_lock *lock, void *p)
@@ -1150,8 +1147,8 @@ static inline int btree_iter_lock_root(struct btree_trans *trans,
 		}
 
 		lock_type = __btree_lock_want(iter, iter->level);
-		if (unlikely(!btree_node_lock(b, SPOS_MAX, iter->level,
-					      iter, lock_type,
+		if (unlikely(!btree_node_lock(trans, iter, b, SPOS_MAX,
+					      iter->level, lock_type,
 					      lock_root_check_fn, rootp,
 					      trace_ip))) {
 			if (trans->restarted)
@@ -1169,7 +1166,7 @@ static inline int btree_iter_lock_root(struct btree_trans *trans,
 				iter->l[i].b = NULL;
 
 			mark_btree_node_locked(iter, iter->level, lock_type);
-			btree_iter_node_set(iter, b);
+			btree_iter_level_init(trans, iter, b);
 			return 0;
 		}
 
@@ -1194,7 +1191,7 @@ static int btree_iter_prefetch(struct btree_trans *trans, struct btree_iter *ite
 	bch2_bkey_buf_init(&tmp);
 
 	while (nr && !ret) {
-		if (!bch2_btree_node_relock(iter, iter->level))
+		if (!bch2_btree_node_relock(trans, iter, iter->level))
 			break;
 
 		bch2_btree_node_iter_advance(&node_iter, l->b);
@@ -1203,8 +1200,8 @@ static int btree_iter_prefetch(struct btree_trans *trans, struct btree_iter *ite
 			break;
 
 		bch2_bkey_buf_unpack(&tmp, c, l->b, k);
-		ret = bch2_btree_node_prefetch(c, iter, tmp.k, iter->btree_id,
-					       iter->level - 1);
+		ret = bch2_btree_node_prefetch(c, trans, iter, tmp.k,
+					iter->btree_id, iter->level - 1);
 	}
 
 	if (!was_locked)
@@ -1214,7 +1211,8 @@ static int btree_iter_prefetch(struct btree_trans *trans, struct btree_iter *ite
 	return ret;
 }
 
-static noinline void btree_node_mem_ptr_set(struct btree_iter *iter,
+static noinline void btree_node_mem_ptr_set(struct btree_trans *trans,
+					    struct btree_iter *iter,
 					    unsigned plevel, struct btree *b)
 {
 	struct btree_iter_level *l = &iter->l[plevel];
@@ -1222,7 +1220,7 @@ static noinline void btree_node_mem_ptr_set(struct btree_iter *iter,
 	struct bkey_packed *k;
 	struct bch_btree_ptr_v2 *bp;
 
-	if (!bch2_btree_node_relock(iter, plevel))
+	if (!bch2_btree_node_relock(trans, iter, plevel))
 		return;
 
 	k = bch2_btree_node_iter_peek_all(&l->iter, l->b);
@@ -1259,11 +1257,11 @@ static __always_inline int btree_iter_down(struct btree_trans *trans,
 		goto err;
 
 	mark_btree_node_locked(iter, level, lock_type);
-	btree_iter_node_set(iter, b);
+	btree_iter_level_init(trans, iter, b);
 
 	if (tmp.k->k.type == KEY_TYPE_btree_ptr_v2 &&
 	    unlikely(b != btree_node_mem_ptr(tmp.k)))
-		btree_node_mem_ptr_set(iter, level + 1, b);
+		btree_node_mem_ptr_set(trans, iter, level + 1, b);
 
 	if (iter->flags & BTREE_ITER_PREFETCH)
 		ret = btree_iter_prefetch(trans, iter);
@@ -1272,7 +1270,7 @@ static __always_inline int btree_iter_down(struct btree_trans *trans,
 		btree_node_unlock(iter, level + 1);
 	iter->level = level;
 
-	bch2_btree_iter_verify_locks(trans, iter);
+	bch2_btree_iter_verify_locks(iter);
 err:
 	bch2_bkey_buf_exit(&tmp, c);
 	return ret;
@@ -1306,9 +1304,9 @@ retry_all:
 
 		if (iter1->btree_id == iter2->btree_id &&
 		    iter1->locks_want < iter2->locks_want)
-			__bch2_btree_iter_upgrade(iter1, iter2->locks_want);
+			__bch2_btree_iter_upgrade(trans, iter1, iter2->locks_want);
 		else if (!iter1->locks_want && iter2->locks_want)
-			__bch2_btree_iter_upgrade(iter1, 1);
+			__bch2_btree_iter_upgrade(trans, iter1, 1);
 	}
 
 	bch2_trans_unlock(trans);
@@ -1364,11 +1362,12 @@ static int bch2_btree_iter_traverse_all(struct btree_trans *trans)
 	return __btree_iter_traverse_all(trans, 0, _RET_IP_);
 }
 
-static inline bool btree_iter_good_node(struct btree_iter *iter,
+static inline bool btree_iter_good_node(struct btree_trans *trans,
+					struct btree_iter *iter,
 					unsigned l, int check_pos)
 {
 	if (!is_btree_node(iter, l) ||
-	    !bch2_btree_node_relock(iter, l))
+	    !bch2_btree_node_relock(trans, iter, l))
 		return false;
 
 	if (check_pos < 0 && btree_iter_pos_before_node(iter, iter->l[l].b))
@@ -1378,13 +1377,14 @@ static inline bool btree_iter_good_node(struct btree_iter *iter,
 	return true;
 }
 
-static inline unsigned btree_iter_up_until_good_node(struct btree_iter *iter,
+static inline unsigned btree_iter_up_until_good_node(struct btree_trans *trans,
+						     struct btree_iter *iter,
 						     int check_pos)
 {
 	unsigned l = iter->level;
 
 	while (btree_iter_node(iter, l) &&
-	       !btree_iter_good_node(iter, l, check_pos)) {
+	       !btree_iter_good_node(trans, iter, l, check_pos)) {
 		btree_node_unlock(iter, l);
 		iter->l[l].b = BTREE_ITER_NO_NODE_UP;
 		l++;
@@ -1419,20 +1419,20 @@ static int btree_iter_traverse_one(struct btree_trans *trans,
 	}
 
 	if (btree_iter_type(iter) == BTREE_ITER_CACHED) {
-		ret = bch2_btree_iter_traverse_cached(iter);
+		ret = bch2_btree_iter_traverse_cached(trans, iter);
 		goto out;
 	}
 
 	if (unlikely(iter->level >= BTREE_MAX_DEPTH))
 		goto out;
 
-	iter->level = btree_iter_up_until_good_node(iter, 0);
+	iter->level = btree_iter_up_until_good_node(trans, iter, 0);
 
 	/* If we need intent locks, take them too: */
 	for (l = iter->level + 1;
 	     l < iter->locks_want && btree_iter_node(iter, l);
 	     l++)
-		if (!bch2_btree_node_relock(iter, l))
+		if (!bch2_btree_node_relock(trans, iter, l))
 			while (iter->level <= l) {
 				btree_node_unlock(iter, iter->level);
 				iter->l[iter->level].b = BTREE_ITER_NO_NODE_UP;
@@ -1643,7 +1643,7 @@ static void btree_iter_set_search_pos(struct btree_iter *iter, struct bpos new_p
 		return;
 	}
 
-	l = btree_iter_up_until_good_node(iter, cmp);
+	l = btree_iter_up_until_good_node(trans, iter, cmp);
 
 	if (btree_iter_node(iter, l)) {
 		/*
@@ -1654,7 +1654,7 @@ static void btree_iter_set_search_pos(struct btree_iter *iter, struct bpos new_p
 		 */
 		if (cmp < 0 ||
 		    !btree_iter_advance_to_pos(iter, &iter->l[l], 8))
-			__btree_iter_init(iter, l);
+			__btree_iter_level_init(iter, l);
 
 		/* Don't leave it locked if we're not supposed to: */
 		if (btree_lock_want(iter, l) == BTREE_NODE_UNLOCKED)
diff --git a/fs/bcachefs/btree_iter.h b/fs/bcachefs/btree_iter.h
index 6dad6f1a2d9b..b6354782a2d0 100644
--- a/fs/bcachefs/btree_iter.h
+++ b/fs/bcachefs/btree_iter.h
@@ -135,7 +135,7 @@ void bch2_btree_node_iter_fix(struct btree_trans *trans, struct btree_iter *,
 			      struct btree *, struct btree_node_iter *,
 			      struct bkey_packed *, unsigned, unsigned);
 
-bool bch2_btree_iter_relock_intent(struct btree_iter *);
+bool bch2_btree_iter_relock_intent(struct btree_trans *, struct btree_iter *);
 
 bool bch2_trans_relock(struct btree_trans *);
 void bch2_trans_unlock(struct btree_trans *);
@@ -148,15 +148,17 @@ static inline int btree_trans_restart(struct btree_trans *trans)
 	return -EINTR;
 }
 
-bool __bch2_btree_iter_upgrade(struct btree_iter *, unsigned);
+bool __bch2_btree_iter_upgrade(struct btree_trans *,
+			       struct btree_iter *, unsigned);
 
-static inline bool bch2_btree_iter_upgrade(struct btree_iter *iter,
+static inline bool bch2_btree_iter_upgrade(struct btree_trans *trans,
+					   struct btree_iter *iter,
 					   unsigned new_locks_want)
 {
 	new_locks_want = min(new_locks_want, BTREE_MAX_DEPTH);
 
 	return iter->locks_want < new_locks_want
-		? __bch2_btree_iter_upgrade(iter, new_locks_want)
+		? __bch2_btree_iter_upgrade(trans, iter, new_locks_want)
 		: iter->uptodate <= BTREE_ITER_NEED_PEEK;
 }
 
diff --git a/fs/bcachefs/btree_key_cache.c b/fs/bcachefs/btree_key_cache.c
index 6bc20813d00d..568c1f2704c2 100644
--- a/fs/bcachefs/btree_key_cache.c
+++ b/fs/bcachefs/btree_key_cache.c
@@ -212,7 +212,7 @@ static int btree_key_cache_fill(struct btree_trans *trans,
 	if (ret)
 		goto err;
 
-	if (!bch2_btree_node_relock(ck_iter, 0)) {
+	if (!bch2_btree_node_relock(trans, ck_iter, 0)) {
 		trace_transaction_restart_ip(trans->ip, _THIS_IP_);
 		ret = btree_trans_restart(trans);
 		goto err;
@@ -265,9 +265,8 @@ static int bkey_cached_check_fn(struct six_lock *lock, void *p)
 }
 
 __flatten
-int bch2_btree_iter_traverse_cached(struct btree_iter *iter)
+int bch2_btree_iter_traverse_cached(struct btree_trans *trans, struct btree_iter *iter)
 {
-	struct btree_trans *trans = iter->trans;
 	struct bch_fs *c = trans->c;
 	struct bkey_cached *ck;
 	int ret = 0;
@@ -276,7 +275,7 @@ int bch2_btree_iter_traverse_cached(struct btree_iter *iter)
 
 	iter->l[1].b = NULL;
 
-	if (bch2_btree_node_relock(iter, 0)) {
+	if (bch2_btree_node_relock(trans, iter, 0)) {
 		ck = (void *) iter->l[0].b;
 		goto fill;
 	}
@@ -301,7 +300,7 @@ retry:
 	} else {
 		enum six_lock_type lock_want = __btree_lock_want(iter, 0);
 
-		if (!btree_node_lock((void *) ck, iter->pos, 0, iter, lock_want,
+		if (!btree_node_lock(trans, iter, (void *) ck, iter->pos, 0, lock_want,
 				     bkey_cached_check_fn, iter, _THIS_IP_)) {
 			if (!trans->restarted)
 				goto retry;
@@ -325,7 +324,7 @@ retry:
 fill:
 	if (!ck->valid && !(iter->flags & BTREE_ITER_CACHED_NOFILL)) {
 		if (!iter->locks_want &&
-		    !!__bch2_btree_iter_upgrade(iter, 1)) {
+		    !!__bch2_btree_iter_upgrade(trans, iter, 1)) {
 			trace_transaction_restart_ip(trans->ip, _THIS_IP_);
 			BUG_ON(!trans->restarted);
 			ret = -EINTR;
@@ -343,7 +342,7 @@ fill:
 	iter->uptodate = BTREE_ITER_NEED_PEEK;
 
 	if ((iter->flags & BTREE_ITER_INTENT) &&
-	    !bch2_btree_iter_upgrade(iter, 1)) {
+	    !bch2_btree_iter_upgrade(trans, iter, 1)) {
 		BUG_ON(!trans->restarted);
 		ret = -EINTR;
 	}
diff --git a/fs/bcachefs/btree_key_cache.h b/fs/bcachefs/btree_key_cache.h
index 7e2b0a08f745..d890632e4425 100644
--- a/fs/bcachefs/btree_key_cache.h
+++ b/fs/bcachefs/btree_key_cache.h
@@ -26,7 +26,7 @@ int bch2_btree_key_cache_journal_flush(struct journal *,
 struct bkey_cached *
 bch2_btree_key_cache_find(struct bch_fs *, enum btree_id, struct bpos);
 
-int bch2_btree_iter_traverse_cached(struct btree_iter *);
+int bch2_btree_iter_traverse_cached(struct btree_trans *, struct btree_iter *);
 
 bool bch2_btree_insert_key_cached(struct btree_trans *,
 			struct btree_iter *, struct bkey_i *);
diff --git a/fs/bcachefs/btree_locking.h b/fs/bcachefs/btree_locking.h
index 869c498e3f78..f8b358f8f2c1 100644
--- a/fs/bcachefs/btree_locking.h
+++ b/fs/bcachefs/btree_locking.h
@@ -167,40 +167,38 @@ static inline bool btree_node_lock_increment(struct btree_trans *trans,
 	return false;
 }
 
-bool __bch2_btree_node_lock(struct btree *, struct bpos, unsigned,
-			    struct btree_iter *, enum six_lock_type,
-			    six_lock_should_sleep_fn, void *,
-			    unsigned long);
+bool __bch2_btree_node_lock(struct btree_trans *, struct btree_iter *,
+			    struct btree *, struct bpos, unsigned,
+			    enum six_lock_type, six_lock_should_sleep_fn,
+			    void *, unsigned long);
 
-static inline bool btree_node_lock(struct btree *b,
-			struct bpos pos, unsigned level,
+static inline bool btree_node_lock(struct btree_trans *trans,
 			struct btree_iter *iter,
+			struct btree *b, struct bpos pos, unsigned level,
 			enum six_lock_type type,
 			six_lock_should_sleep_fn should_sleep_fn, void *p,
 			unsigned long ip)
 {
-	struct btree_trans *trans = iter->trans;
-
 	EBUG_ON(level >= BTREE_MAX_DEPTH);
 	EBUG_ON(!(trans->iters_linked & (1ULL << iter->idx)));
 
 	return likely(six_trylock_type(&b->c.lock, type)) ||
 		btree_node_lock_increment(trans, b, level, type) ||
-		__bch2_btree_node_lock(b, pos, level, iter, type,
+		__bch2_btree_node_lock(trans, iter, b, pos, level, type,
 				       should_sleep_fn, p, ip);
 }
 
-bool __bch2_btree_node_relock(struct btree_iter *, unsigned);
+bool __bch2_btree_node_relock(struct btree_trans *, struct btree_iter *, unsigned);
 
-static inline bool bch2_btree_node_relock(struct btree_iter *iter,
-					  unsigned level)
+static inline bool bch2_btree_node_relock(struct btree_trans *trans,
+					  struct btree_iter *iter, unsigned level)
 {
 	EBUG_ON(btree_node_locked(iter, level) &&
 		btree_node_locked_type(iter, level) !=
 		__btree_lock_want(iter, level));
 
 	return likely(btree_node_locked(iter, level)) ||
-		__bch2_btree_node_relock(iter, level);
+		__bch2_btree_node_relock(trans, iter, level);
 }
 
 /*
@@ -225,8 +223,7 @@ bch2_btree_node_unlock_write_inlined(struct btree_trans *trans, struct btree_ite
 void bch2_btree_node_unlock_write(struct btree_trans *,
 			struct btree_iter *, struct btree *);
 
-void __bch2_btree_node_lock_write(struct btree_trans *,
-			struct btree_iter *, struct btree *);
+void __bch2_btree_node_lock_write(struct btree_trans *, struct btree *);
 
 static inline void bch2_btree_node_lock_write(struct btree_trans *trans,
 					      struct btree_iter *iter,
@@ -234,9 +231,10 @@ static inline void bch2_btree_node_lock_write(struct btree_trans *trans,
 {
 	EBUG_ON(iter->l[b->c.level].b != b);
 	EBUG_ON(iter->l[b->c.level].lock_seq != b->c.lock.state.seq);
+	EBUG_ON(!btree_node_intent_locked(iter, b->c.level));
 
 	if (unlikely(!six_trylock_write(&b->c.lock)))
-		__bch2_btree_node_lock_write(trans, iter, b);
+		__bch2_btree_node_lock_write(trans, b);
 }
 
 #endif /* _BCACHEFS_BTREE_LOCKING_H */
diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index 3d7c47712b74..d18d539bcc8e 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -937,7 +937,7 @@ retry:
 	 * XXX: figure out how far we might need to split,
 	 * instead of locking/reserving all the way to the root:
 	 */
-	if (!bch2_btree_iter_upgrade(iter, U8_MAX)) {
+	if (!bch2_btree_iter_upgrade(trans, iter, U8_MAX)) {
 		trace_trans_restart_iter_upgrade(trans->ip, _RET_IP_,
 						 iter->btree_id,
 						 &iter->real_pos);
diff --git a/fs/bcachefs/btree_update_interior.h b/fs/bcachefs/btree_update_interior.h
index 07046dab614b..13b3a1bf0f4f 100644
--- a/fs/bcachefs/btree_update_interior.h
+++ b/fs/bcachefs/btree_update_interior.h
@@ -132,7 +132,7 @@ static inline int bch2_foreground_maybe_merge_sibling(struct btree_trans *trans,
 	if (iter->uptodate >= BTREE_ITER_NEED_TRAVERSE)
 		return 0;
 
-	if (!bch2_btree_node_relock(iter, level))
+	if (!bch2_btree_node_relock(trans, iter, level))
 		return 0;
 
 	b = iter->l[level].b;
diff --git a/fs/bcachefs/btree_update_leaf.c b/fs/bcachefs/btree_update_leaf.c
index 21fd6b2a7532..aabf408a5114 100644
--- a/fs/bcachefs/btree_update_leaf.c
+++ b/fs/bcachefs/btree_update_leaf.c
@@ -561,7 +561,7 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans,
 	 */
 	trans_for_each_iter(trans, iter)
 		if (iter->nodes_locked != iter->nodes_intent_locked &&
-		    !bch2_btree_iter_upgrade(iter, 1)) {
+		    !bch2_btree_iter_upgrade(trans, iter, 1)) {
 			trace_trans_restart_upgrade(trans->ip, trace_ip,
 						    iter->btree_id,
 						    &iter->real_pos);
@@ -783,7 +783,8 @@ int __bch2_trans_commit(struct btree_trans *trans)
 	trans_for_each_update(trans, i) {
 		BUG_ON(!i->iter->should_be_locked);
 
-		if (unlikely(!bch2_btree_iter_upgrade(i->iter, i->level + 1))) {
+		if (unlikely(!bch2_btree_iter_upgrade(trans, i->iter,
+						      i->level + 1))) {
 			trace_trans_restart_upgrade(trans->ip, _RET_IP_,
 						    i->iter->btree_id,
 						    &i->iter->pos);
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index b02af94f4037..71b0f14f41f3 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -326,8 +326,8 @@ static void btree_and_journal_iter_prefetch(struct bch_fs *c, struct btree *b,
 	       (k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 		bch2_bkey_buf_reassemble(&tmp, c, k);
 
-		bch2_btree_node_prefetch(c, NULL, tmp.k,
-					b->c.btree_id, b->c.level - 1);
+		bch2_btree_node_prefetch(c, NULL, NULL, tmp.k,
+					 b->c.btree_id, b->c.level - 1);
 
 		bch2_btree_and_journal_iter_advance(&iter);
 		i++;