Update bcachefs sources to c9eb15545d bcachefs: Don't call trans_iter_put() on error pointer

.bcachefs_revision

@@ -1 +1 @@
-aae76fba150894dc6c8b21a58b110413531af287
+c9eb15545dce4f5490371afb92033082600473fe

libbcachefs/acl.c

@@ -377,7 +377,7 @@ int bch2_acl_chmod(struct btree_trans *trans,
 	}
 
 	new->k.p = iter->pos;
-	bch2_trans_update(trans, iter, &new->k_i);
+	bch2_trans_update(trans, iter, &new->k_i, 0);
 	*new_acl = acl;
 	acl = NULL;
 err:

libbcachefs/alloc_background.c

@@ -222,8 +222,8 @@ int bch2_alloc_read(struct bch_fs *c, struct journal_keys *journal_keys)
 
 	for_each_btree_key(&trans, iter, BTREE_ID_ALLOC, POS_MIN, 0, k, ret)
 		bch2_mark_key(c, k, 0, 0, NULL, 0,
-			      BCH_BUCKET_MARK_ALLOC_READ|
+			      BTREE_TRIGGER_ALLOC_READ|
-			      BCH_BUCKET_MARK_NOATOMIC);
+			      BTREE_TRIGGER_NOATOMIC);
 
 	ret = bch2_trans_exit(&trans) ?: ret;
 	if (ret) {
@@ -235,8 +235,8 @@ int bch2_alloc_read(struct bch_fs *c, struct journal_keys *journal_keys)
 		if (j->btree_id == BTREE_ID_ALLOC)
 			bch2_mark_key(c, bkey_i_to_s_c(j->k),
 				      0, 0, NULL, 0,
-				      BCH_BUCKET_MARK_ALLOC_READ|
+				      BTREE_TRIGGER_ALLOC_READ|
-				      BCH_BUCKET_MARK_NOATOMIC);
+				      BTREE_TRIGGER_NOATOMIC);
 
 	percpu_down_write(&c->mark_lock);
 	bch2_dev_usage_from_buckets(c);
@@ -314,11 +314,10 @@ retry:
 	a->k.p = iter->pos;
 	bch2_alloc_pack(a, new_u);
 
-	bch2_trans_update(trans, iter, &a->k_i);
+	bch2_trans_update(trans, iter, &a->k_i,
+			  BTREE_TRIGGER_NORUN);
 	ret = bch2_trans_commit(trans, NULL, NULL,
-				BTREE_INSERT_NOFAIL|
+				BTREE_INSERT_NOFAIL|flags);
-				BTREE_INSERT_NOMARK|
-				flags);
 err:
 	if (ret == -EINTR)
 		goto retry;
@@ -383,8 +382,7 @@ int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 	ret = bch2_alloc_write_key(&trans, iter,
 				   BTREE_INSERT_NOFAIL|
 				   BTREE_INSERT_LAZY_RW|
-				   BTREE_INSERT_JOURNAL_REPLAY|
+				   BTREE_INSERT_JOURNAL_REPLAY);
-				   BTREE_INSERT_NOMARK);
 	bch2_trans_exit(&trans);
 	return ret < 0 ? ret : 0;
 }
@@ -901,7 +899,8 @@ retry:
 	a->k.p = iter->pos;
 	bch2_alloc_pack(a, u);
 
-	bch2_trans_update(trans, iter, &a->k_i);
+	bch2_trans_update(trans, iter, &a->k_i,
+			  BTREE_TRIGGER_BUCKET_INVALIDATE);
 
 	/*
 	 * XXX:
@@ -917,7 +916,6 @@ retry:
 				BTREE_INSERT_NOFAIL|
 				BTREE_INSERT_USE_RESERVE|
 				BTREE_INSERT_USE_ALLOC_RESERVE|
-				BTREE_INSERT_BUCKET_INVALIDATE|
 				flags);
 	if (ret == -EINTR)
 		goto retry;

libbcachefs/bkey_methods.c

@@ -156,7 +156,7 @@ void bch2_bkey_debugcheck(struct bch_fs *c, struct btree *b, struct bkey_s_c k)
 		char buf[160];
 
 		bch2_bkey_val_to_text(&PBUF(buf), c, k);
-		bch2_fs_bug(c, "invalid bkey %s: %s", buf, invalid);
+		bch2_fs_inconsistent(c, "invalid bkey %s: %s", buf, invalid);
 		return;
 	}
 

libbcachefs/btree_gc.c

@@ -116,8 +116,8 @@ static int bch2_gc_mark_key(struct bch_fs *c, struct bkey_s_c k,
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const struct bch_extent_ptr *ptr;
 	unsigned flags =
-		BCH_BUCKET_MARK_GC|
+		BTREE_TRIGGER_GC|
-		(initial ? BCH_BUCKET_MARK_NOATOMIC : 0);
+		(initial ? BTREE_TRIGGER_NOATOMIC : 0);
 	int ret = 0;
 
 	if (initial) {
@@ -294,8 +294,8 @@ static int mark_journal_key(struct bch_fs *c, enum btree_id id,
 			   BTREE_ITER_SLOTS, k, ret) {
 		percpu_down_read(&c->mark_lock);
 		ret = bch2_mark_overwrite(&trans, iter, k, insert, NULL,
-					 BCH_BUCKET_MARK_GC|
+					 BTREE_TRIGGER_GC|
-					 BCH_BUCKET_MARK_NOATOMIC);
+					 BTREE_TRIGGER_NOATOMIC);
 		percpu_up_read(&c->mark_lock);
 
 		if (!ret)
@@ -407,7 +407,7 @@ static void bch2_mark_superblocks(struct bch_fs *c)
 	gc_pos_set(c, gc_phase(GC_PHASE_SB));
 
 	for_each_online_member(ca, c, i)
-		bch2_mark_dev_superblock(c, ca, BCH_BUCKET_MARK_GC);
+		bch2_mark_dev_superblock(c, ca, BTREE_TRIGGER_GC);
 	mutex_unlock(&c->sb_lock);
 }
 
@@ -424,7 +424,7 @@ static void bch2_mark_pending_btree_node_frees(struct bch_fs *c)
 		if (d->index_update_done)
 			bch2_mark_key(c, bkey_i_to_s_c(&d->key),
 				      0, 0, NULL, 0,
-				      BCH_BUCKET_MARK_GC);
+				      BTREE_TRIGGER_GC);
 
 	mutex_unlock(&c->btree_interior_update_lock);
 }
@@ -445,7 +445,7 @@ static void bch2_mark_allocator_buckets(struct bch_fs *c)
 		fifo_for_each_entry(i, &ca->free_inc, iter)
 			bch2_mark_alloc_bucket(c, ca, i, true,
 					       gc_pos_alloc(c, NULL),
-					       BCH_BUCKET_MARK_GC);
+					       BTREE_TRIGGER_GC);
 
 
 
@@ -453,7 +453,7 @@ static void bch2_mark_allocator_buckets(struct bch_fs *c)
 			fifo_for_each_entry(i, &ca->free[j], iter)
 				bch2_mark_alloc_bucket(c, ca, i, true,
 						       gc_pos_alloc(c, NULL),
-						       BCH_BUCKET_MARK_GC);
+						       BTREE_TRIGGER_GC);
 	}
 
 	spin_unlock(&c->freelist_lock);
@@ -467,7 +467,7 @@ static void bch2_mark_allocator_buckets(struct bch_fs *c)
 			ca = bch_dev_bkey_exists(c, ob->ptr.dev);
 			bch2_mark_alloc_bucket(c, ca, PTR_BUCKET_NR(ca, &ob->ptr), true,
 					       gc_pos_alloc(c, ob),
-					       BCH_BUCKET_MARK_GC);
+					       BTREE_TRIGGER_GC);
 		}
 		spin_unlock(&ob->lock);
 	}

libbcachefs/btree_iter.c

@@ -1027,10 +1027,7 @@ retry_all:
 	for (i = 0; i < nr_sorted; i++) {
 		iter = &trans->iters[sorted[i]];
 
-		do {
+		ret = btree_iter_traverse_one(iter);
-			ret = btree_iter_traverse_one(iter);
-		} while (ret == -EINTR);
-
 		if (ret)
 			goto retry_all;
 	}
@@ -1793,10 +1790,9 @@ int bch2_trans_iter_free(struct btree_trans *trans,
 static int bch2_trans_realloc_iters(struct btree_trans *trans,
 				    unsigned new_size)
 {
-	void *new_iters, *new_updates, *new_sorted;
+	void *new_iters, *new_updates;
 	size_t iters_bytes;
 	size_t updates_bytes;
-	size_t sorted_bytes;
 
 	new_size = roundup_pow_of_two(new_size);
 
@@ -1810,12 +1806,9 @@ static int bch2_trans_realloc_iters(struct btree_trans *trans,
 	bch2_trans_unlock(trans);
 
 	iters_bytes	= sizeof(struct btree_iter) * new_size;
-	updates_bytes	= sizeof(struct btree_insert_entry) * (new_size + 4);
+	updates_bytes	= sizeof(struct btree_insert_entry) * new_size;
-	sorted_bytes	= sizeof(u8) * (new_size + 4);
 
-	new_iters = kmalloc(iters_bytes +
+	new_iters = kmalloc(iters_bytes + updates_bytes, GFP_NOFS);
-			    updates_bytes +
-			    sorted_bytes, GFP_NOFS);
 	if (new_iters)
 		goto success;
 
@@ -1825,7 +1818,6 @@ static int bch2_trans_realloc_iters(struct btree_trans *trans,
 	trans->used_mempool = true;
 success:
 	new_updates	= new_iters + iters_bytes;
-	new_sorted	= new_updates + updates_bytes;
 
 	memcpy(new_iters, trans->iters,
 	       sizeof(struct btree_iter) * trans->nr_iters);
@@ -1842,7 +1834,6 @@ success:
 
 	trans->iters		= new_iters;
 	trans->updates		= new_updates;
-	trans->updates_sorted	= new_sorted;
 	trans->size		= new_size;
 
 	if (trans->iters_live) {
@@ -1891,6 +1882,7 @@ static struct btree_iter *btree_trans_iter_alloc(struct btree_trans *trans)
 got_slot:
 	BUG_ON(trans->iters_linked & (1ULL << idx));
 	trans->iters_linked |= 1ULL << idx;
+	trans->iters[idx].flags = 0;
 	return &trans->iters[idx];
 }
 
@@ -1906,6 +1898,9 @@ static inline void btree_iter_copy(struct btree_iter *dst,
 		if (btree_node_locked(dst, i))
 			six_lock_increment(&dst->l[i].b->lock,
 					   __btree_lock_want(dst, i));
+
+	dst->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
+	dst->flags &= ~BTREE_ITER_SET_POS_AFTER_COMMIT;
 }
 
 static inline struct bpos bpos_diff(struct bpos l, struct bpos r)
@@ -1956,7 +1951,6 @@ static struct btree_iter *__btree_trans_get_iter(struct btree_trans *trans,
 		iter = best;
 	}
 
-	iter->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
 	iter->flags &= ~(BTREE_ITER_SLOTS|BTREE_ITER_INTENT|BTREE_ITER_PREFETCH);
 	iter->flags |= flags & (BTREE_ITER_SLOTS|BTREE_ITER_INTENT|BTREE_ITER_PREFETCH);
 
@@ -1968,6 +1962,7 @@ static struct btree_iter *__btree_trans_get_iter(struct btree_trans *trans,
 	BUG_ON(iter->btree_id != btree_id);
 	BUG_ON((iter->flags ^ flags) & BTREE_ITER_TYPE);
 	BUG_ON(iter->flags & BTREE_ITER_KEEP_UNTIL_COMMIT);
+	BUG_ON(iter->flags & BTREE_ITER_SET_POS_AFTER_COMMIT);
 	BUG_ON(trans->iters_live & (1ULL << iter->idx));
 
 	trans->iters_live	|= 1ULL << iter->idx;
@@ -2030,7 +2025,6 @@ struct btree_iter *bch2_trans_copy_iter(struct btree_trans *trans,
 	 * it's cheap to copy it again:
 	 */
 	trans->iters_touched &= ~(1ULL << iter->idx);
-	iter->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
 
 	return iter;
 }
@@ -2090,7 +2084,8 @@ void bch2_trans_reset(struct btree_trans *trans, unsigned flags)
 	struct btree_iter *iter;
 
 	trans_for_each_iter(trans, iter)
-		iter->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
+		iter->flags &= ~(BTREE_ITER_KEEP_UNTIL_COMMIT|
+				 BTREE_ITER_SET_POS_AFTER_COMMIT);
 
 	bch2_trans_unlink_iters(trans);
 
@@ -2099,12 +2094,21 @@ void bch2_trans_reset(struct btree_trans *trans, unsigned flags)
 
 	trans->iters_touched &= trans->iters_live;
 
+	trans->need_reset		= 0;
 	trans->nr_updates		= 0;
 
 	if (flags & TRANS_RESET_MEM)
 		trans->mem_top		= 0;
 
-	bch2_btree_iter_traverse_all(trans);
+	if (trans->fs_usage_deltas) {
+		trans->fs_usage_deltas->used = 0;
+		memset(&trans->fs_usage_deltas->memset_start, 0,
+		       (void *) &trans->fs_usage_deltas->memset_end -
+		       (void *) &trans->fs_usage_deltas->memset_start);
+	}
+
+	if (!(flags & TRANS_RESET_NOTRAVERSE))
+		bch2_btree_iter_traverse_all(trans);
 }
 
 void bch2_trans_init(struct btree_trans *trans, struct bch_fs *c,
@@ -2118,7 +2122,6 @@ void bch2_trans_init(struct btree_trans *trans, struct bch_fs *c,
 	trans->size		= ARRAY_SIZE(trans->iters_onstack);
 	trans->iters		= trans->iters_onstack;
 	trans->updates		= trans->updates_onstack;
-	trans->updates_sorted	= trans->updates_sorted_onstack;
 	trans->fs_usage_deltas	= NULL;
 
 	if (expected_nr_iters > trans->size)
@@ -2155,6 +2158,6 @@ int bch2_fs_btree_iter_init(struct bch_fs *c)
 
 	return mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
 			sizeof(struct btree_iter) * nr +
-			sizeof(struct btree_insert_entry) * (nr + 4) +
+			sizeof(struct btree_insert_entry) * nr +
-			sizeof(u8) * (nr + 4));
+			sizeof(u8) * nr);
 }

libbcachefs/btree_iter.h

@@ -291,6 +291,7 @@ struct btree_iter *bch2_trans_get_node_iter(struct btree_trans *,
 
 #define TRANS_RESET_ITERS		(1 << 0)
 #define TRANS_RESET_MEM			(1 << 1)
+#define TRANS_RESET_NOTRAVERSE		(1 << 2)
 
 void bch2_trans_reset(struct btree_trans *, unsigned);
 

libbcachefs/btree_types.h

@@ -194,6 +194,7 @@ enum btree_iter_type {
  */
 #define BTREE_ITER_IS_EXTENTS		(1 << 6)
 #define BTREE_ITER_ERROR		(1 << 7)
+#define BTREE_ITER_SET_POS_AFTER_COMMIT	(1 << 8)
 
 enum btree_iter_uptodate {
 	BTREE_ITER_UPTODATE		= 0,
@@ -210,12 +211,13 @@ enum btree_iter_uptodate {
  * @nodes_intent_locked	- bitmask indicating which locks are intent locks
  */
 struct btree_iter {
-	u8			idx;
-
 	struct btree_trans	*trans;
 	struct bpos		pos;
+	struct bpos		pos_after_commit;
+
+	u16			flags;
+	u8			idx;
 
-	u8			flags;
 	enum btree_iter_uptodate uptodate:4;
 	enum btree_id		btree_id:4;
 	unsigned		level:4,
@@ -242,6 +244,8 @@ static inline enum btree_iter_type btree_iter_type(struct btree_iter *iter)
 }
 
 struct btree_insert_entry {
+	unsigned		trigger_flags;
+	unsigned		trans_triggers_run:1;
 	struct bkey_i		*k;
 	struct btree_iter	*iter;
 };
@@ -262,6 +266,7 @@ struct btree_trans {
 	unsigned		used_mempool:1;
 	unsigned		error:1;
 	unsigned		nounlock:1;
+	unsigned		need_reset:1;
 
 	unsigned		mem_top;
 	unsigned		mem_bytes;
@@ -269,7 +274,6 @@ struct btree_trans {
 
 	struct btree_iter	*iters;
 	struct btree_insert_entry *updates;
-	u8			*updates_sorted;
 
 	/* update path: */
 	struct journal_res	journal_res;
@@ -282,8 +286,7 @@ struct btree_trans {
 	struct replicas_delta_list *fs_usage_deltas;
 
 	struct btree_iter	iters_onstack[2];
-	struct btree_insert_entry updates_onstack[6];
+	struct btree_insert_entry updates_onstack[2];
-	u8			updates_sorted_onstack[6];
 };
 
 #define BTREE_FLAG(flag)						\
@@ -481,6 +484,32 @@ static inline bool btree_node_is_extents(struct btree *b)
 	 (1U << BKEY_TYPE_INODES)|			\
 	 (1U << BKEY_TYPE_REFLINK))
 
+enum btree_trigger_flags {
+	__BTREE_TRIGGER_NORUN,		/* Don't run triggers at all */
+	__BTREE_TRIGGER_NOOVERWRITES,	/* Don't run triggers on overwrites */
+
+	__BTREE_TRIGGER_INSERT,
+	__BTREE_TRIGGER_OVERWRITE,
+	__BTREE_TRIGGER_OVERWRITE_SPLIT,
+
+	__BTREE_TRIGGER_GC,
+	__BTREE_TRIGGER_BUCKET_INVALIDATE,
+	__BTREE_TRIGGER_ALLOC_READ,
+	__BTREE_TRIGGER_NOATOMIC,
+};
+
+#define BTREE_TRIGGER_NORUN		(1U << __BTREE_TRIGGER_NORUN)
+#define BTREE_TRIGGER_NOOVERWRITES	(1U << __BTREE_TRIGGER_NOOVERWRITES)
+
+#define BTREE_TRIGGER_INSERT		(1U << __BTREE_TRIGGER_INSERT)
+#define BTREE_TRIGGER_OVERWRITE		(1U << __BTREE_TRIGGER_OVERWRITE)
+#define BTREE_TRIGGER_OVERWRITE_SPLIT	(1U << __BTREE_TRIGGER_OVERWRITE_SPLIT)
+
+#define BTREE_TRIGGER_GC		(1U << __BTREE_TRIGGER_GC)
+#define BTREE_TRIGGER_BUCKET_INVALIDATE	(1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
+#define BTREE_TRIGGER_ALLOC_READ	(1U << __BTREE_TRIGGER_ALLOC_READ)
+#define BTREE_TRIGGER_NOATOMIC		(1U << __BTREE_TRIGGER_NOATOMIC)
+
 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
 {
 	return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);

libbcachefs/btree_update.h

@@ -15,7 +15,7 @@ bool bch2_btree_bset_insert_key(struct btree_iter *, struct btree *,
 void bch2_btree_journal_key(struct btree_trans *, struct btree_iter *,
 			    struct bkey_i *);
 
-enum {
+enum btree_insert_flags {
 	__BTREE_INSERT_NOUNLOCK,
 	__BTREE_INSERT_NOFAIL,
 	__BTREE_INSERT_NOCHECK_RW,
@@ -24,9 +24,6 @@ enum {
 	__BTREE_INSERT_USE_ALLOC_RESERVE,
 	__BTREE_INSERT_JOURNAL_REPLAY,
 	__BTREE_INSERT_JOURNAL_RESERVED,
-	__BTREE_INSERT_NOMARK_OVERWRITES,
-	__BTREE_INSERT_NOMARK,
-	__BTREE_INSERT_BUCKET_INVALIDATE,
 	__BTREE_INSERT_NOWAIT,
 	__BTREE_INSERT_GC_LOCK_HELD,
 	__BCH_HASH_SET_MUST_CREATE,
@@ -53,14 +50,6 @@ enum {
 
 #define BTREE_INSERT_JOURNAL_RESERVED	(1 << __BTREE_INSERT_JOURNAL_RESERVED)
 
-/* Don't mark overwrites, just new key: */
-#define BTREE_INSERT_NOMARK_OVERWRITES	(1 << __BTREE_INSERT_NOMARK_OVERWRITES)
-
-/* Don't call mark new key at all: */
-#define BTREE_INSERT_NOMARK		(1 << __BTREE_INSERT_NOMARK)
-
-#define BTREE_INSERT_BUCKET_INVALIDATE	(1 << __BTREE_INSERT_BUCKET_INVALIDATE)
-
 /* Don't block on allocation failure (for new btree nodes: */
 #define BTREE_INSERT_NOWAIT		(1 << __BTREE_INSERT_NOWAIT)
 #define BTREE_INSERT_GC_LOCK_HELD	(1 << __BTREE_INSERT_GC_LOCK_HELD)
@@ -83,6 +72,8 @@ int bch2_btree_node_rewrite(struct bch_fs *c, struct btree_iter *,
 int bch2_btree_node_update_key(struct bch_fs *, struct btree_iter *,
 			       struct btree *, struct bkey_i_btree_ptr *);
 
+int bch2_trans_update(struct btree_trans *, struct btree_iter *,
+		      struct bkey_i *, enum btree_trigger_flags);
 int __bch2_trans_commit(struct btree_trans *);
 
 /**
@@ -107,19 +98,6 @@ static inline int bch2_trans_commit(struct btree_trans *trans,
 	return __bch2_trans_commit(trans);
 }
 
-static inline void bch2_trans_update(struct btree_trans *trans,
-				     struct btree_iter *iter,
-				     struct bkey_i *k)
-{
-	EBUG_ON(trans->nr_updates >= trans->nr_iters + 4);
-
-	iter->flags |= BTREE_ITER_KEEP_UNTIL_COMMIT;
-
-	trans->updates[trans->nr_updates++] = (struct btree_insert_entry) {
-		.iter = iter, .k = k
-	};
-}
-
 #define __bch2_trans_do(_trans, _disk_res, _journal_seq,		\
 			_flags,	_reset_flags, _do)			\
 ({									\

libbcachefs/btree_update_interior.c

@@ -193,8 +193,8 @@ found:
 	    gc_pos_cmp(c->gc_pos, gc_phase(GC_PHASE_PENDING_DELETE)) < 0)
 		bch2_mark_key_locked(c, bkey_i_to_s_c(&d->key),
 			      0, 0, NULL, 0,
-			      BCH_BUCKET_MARK_OVERWRITE|
+			      BTREE_TRIGGER_OVERWRITE|
-			      BCH_BUCKET_MARK_GC);
+			      BTREE_TRIGGER_GC);
 }
 
 static void __btree_node_free(struct bch_fs *c, struct btree *b)
@@ -265,13 +265,13 @@ static void bch2_btree_node_free_ondisk(struct bch_fs *c,
 	BUG_ON(!pending->index_update_done);
 
 	bch2_mark_key(c, bkey_i_to_s_c(&pending->key),
-		      0, 0, NULL, 0, BCH_BUCKET_MARK_OVERWRITE);
+		      0, 0, NULL, 0, BTREE_TRIGGER_OVERWRITE);
 
 	if (gc_visited(c, gc_phase(GC_PHASE_PENDING_DELETE)))
 		bch2_mark_key(c, bkey_i_to_s_c(&pending->key),
 			      0, 0, NULL, 0,
-			      BCH_BUCKET_MARK_OVERWRITE|
+			      BTREE_TRIGGER_OVERWRITE|
-			      BCH_BUCKET_MARK_GC);
+			      BTREE_TRIGGER_GC);
 }
 
 static struct btree *__bch2_btree_node_alloc(struct bch_fs *c,
@@ -1084,12 +1084,12 @@ static void bch2_btree_set_root_inmem(struct btree_update *as, struct btree *b)
 
 	bch2_mark_key_locked(c, bkey_i_to_s_c(&b->key),
 		      0, 0, fs_usage, 0,
-		      BCH_BUCKET_MARK_INSERT);
+		      BTREE_TRIGGER_INSERT);
 	if (gc_visited(c, gc_pos_btree_root(b->btree_id)))
 		bch2_mark_key_locked(c, bkey_i_to_s_c(&b->key),
 				     0, 0, NULL, 0,
-				     BCH_BUCKET_MARK_INSERT|
+				     BTREE_TRIGGER_INSERT|
-				     BCH_BUCKET_MARK_GC);
+				     BTREE_TRIGGER_GC);
 
 	if (old && !btree_node_fake(old))
 		bch2_btree_node_free_index(as, NULL,
@@ -1182,13 +1182,13 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as, struct btree *b
 
 	bch2_mark_key_locked(c, bkey_i_to_s_c(insert),
 			     0, 0, fs_usage, 0,
-			     BCH_BUCKET_MARK_INSERT);
+			     BTREE_TRIGGER_INSERT);
 
 	if (gc_visited(c, gc_pos_btree_node(b)))
 		bch2_mark_key_locked(c, bkey_i_to_s_c(insert),
 				     0, 0, NULL, 0,
-				     BCH_BUCKET_MARK_INSERT|
+				     BTREE_TRIGGER_INSERT|
-				     BCH_BUCKET_MARK_GC);
+				     BTREE_TRIGGER_GC);
 
 	while ((k = bch2_btree_node_iter_peek_all(node_iter, b)) &&
 	       bkey_iter_pos_cmp(b, &insert->k.p, k) > 0)
@@ -2031,12 +2031,12 @@ static void __bch2_btree_node_update_key(struct bch_fs *c,
 
 		bch2_mark_key_locked(c, bkey_i_to_s_c(&new_key->k_i),
 			      0, 0, fs_usage, 0,
-			      BCH_BUCKET_MARK_INSERT);
+			      BTREE_TRIGGER_INSERT);
 		if (gc_visited(c, gc_pos_btree_root(b->btree_id)))
 			bch2_mark_key_locked(c, bkey_i_to_s_c(&new_key->k_i),
 					     0, 0, NULL, 0,
-					     BCH_BUCKET_MARK_INSERT||
+					     BTREE_TRIGGER_INSERT||
-					     BCH_BUCKET_MARK_GC);
+					     BTREE_TRIGGER_GC);
 
 		bch2_btree_node_free_index(as, NULL,
 					   bkey_i_to_s_c(&b->key),

libbcachefs/btree_update_leaf.c

@@ -21,18 +21,12 @@
 #include <trace/events/bcachefs.h>
 
 static inline bool same_leaf_as_prev(struct btree_trans *trans,
-				     unsigned idx)
+				     struct btree_insert_entry *i)
 {
-	return idx &&
+	return i != trans->updates &&
-		trans->updates[trans->updates_sorted[idx]].iter->l[0].b ==
+		i[0].iter->l[0].b == i[-1].iter->l[0].b;
-		trans->updates[trans->updates_sorted[idx - 1]].iter->l[0].b;
 }
 
-#define trans_for_each_update_sorted(_trans, _i, _iter)			\
-	for (_iter = 0;							\
-	     _iter < _trans->nr_updates &&				\
-	     (_i = _trans->updates + _trans->updates_sorted[_iter], 1);	\
-	     _iter++)
 
 inline void bch2_btree_node_lock_for_insert(struct bch_fs *c, struct btree *b,
 					    struct btree_iter *iter)
@@ -51,28 +45,6 @@ inline void bch2_btree_node_lock_for_insert(struct bch_fs *c, struct btree *b,
 		bch2_btree_init_next(c, b, iter);
 }
 
-static inline void btree_trans_sort_updates(struct btree_trans *trans)
-{
-	struct btree_insert_entry *l, *r;
-	unsigned nr = 0, pos;
-
-	trans_for_each_update(trans, l) {
-		for (pos = 0; pos < nr; pos++) {
-			r = trans->updates + trans->updates_sorted[pos];
-
-			if (btree_iter_cmp(l->iter, r->iter) <= 0)
-				break;
-		}
-
-		memmove(&trans->updates_sorted[pos + 1],
-			&trans->updates_sorted[pos],
-			(nr - pos) * sizeof(trans->updates_sorted[0]));
-
-		trans->updates_sorted[pos] = l - trans->updates;
-		nr++;
-	}
-}
-
 /* Inserting into a given leaf node (last stage of insert): */
 
 /* Handle overwrites and do insert, for non extents: */
@@ -239,40 +211,39 @@ void bch2_btree_journal_key(struct btree_trans *trans,
 }
 
 static void bch2_insert_fixup_key(struct btree_trans *trans,
-				  struct btree_insert_entry *insert)
+				  struct btree_iter *iter,
+				  struct bkey_i *insert)
 {
-	struct btree_iter *iter = insert->iter;
 	struct btree_iter_level *l = &iter->l[0];
 
 	EBUG_ON(iter->level);
-	EBUG_ON(insert->k->k.u64s >
+	EBUG_ON(insert->k.u64s >
 		bch_btree_keys_u64s_remaining(trans->c, l->b));
 
-	if (likely(bch2_btree_bset_insert_key(iter, l->b, &l->iter,
+	if (likely(bch2_btree_bset_insert_key(iter, l->b, &l->iter, insert)))
-					      insert->k)))
+		bch2_btree_journal_key(trans, iter, insert);
-		bch2_btree_journal_key(trans, iter, insert->k);
 }
 
 /**
  * btree_insert_key - insert a key one key into a leaf node
  */
 static void btree_insert_key_leaf(struct btree_trans *trans,
-				  struct btree_insert_entry *insert)
+				  struct btree_iter *iter,
+				  struct bkey_i *insert)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *iter = insert->iter;
 	struct btree *b = iter->l[0].b;
 	struct bset_tree *t = bset_tree_last(b);
 	int old_u64s = bset_u64s(t);
 	int old_live_u64s = b->nr.live_u64s;
 	int live_u64s_added, u64s_added;
 
-	insert->k->k.needs_whiteout = false;
+	insert->k.needs_whiteout = false;
 
 	if (!btree_node_is_extents(b))
-		bch2_insert_fixup_key(trans, insert);
+		bch2_insert_fixup_key(trans, iter, insert);
 	else
-		bch2_insert_fixup_extent(trans, insert);
+		bch2_insert_fixup_extent(trans, iter, insert);
 
 	live_u64s_added = (int) b->nr.live_u64s - old_live_u64s;
 	u64s_added = (int) bset_u64s(t) - old_u64s;
@@ -286,24 +257,25 @@ static void btree_insert_key_leaf(struct btree_trans *trans,
 	    bch2_maybe_compact_whiteouts(c, b))
 		bch2_btree_iter_reinit_node(iter, b);
 
-	trace_btree_insert_key(c, b, insert->k);
+	trace_btree_insert_key(c, b, insert);
 }
 
 /* Normal update interface: */
 
 static inline void btree_insert_entry_checks(struct btree_trans *trans,
-					     struct btree_insert_entry *i)
+					     struct btree_iter *iter,
+					     struct bkey_i *insert)
 {
 	struct bch_fs *c = trans->c;
 
-	BUG_ON(i->iter->level);
+	BUG_ON(iter->level);
-	BUG_ON(bkey_cmp(bkey_start_pos(&i->k->k), i->iter->pos));
+	BUG_ON(bkey_cmp(bkey_start_pos(&insert->k), iter->pos));
-	EBUG_ON((i->iter->flags & BTREE_ITER_IS_EXTENTS) &&
+	EBUG_ON((iter->flags & BTREE_ITER_IS_EXTENTS) &&
-		bkey_cmp(i->k->k.p, i->iter->l[0].b->key.k.p) > 0);
+		bkey_cmp(insert->k.p, iter->l[0].b->key.k.p) > 0);
 
 	BUG_ON(debug_check_bkeys(c) &&
-	       !bkey_deleted(&i->k->k) &&
+	       !bkey_deleted(&insert->k) &&
-	       bch2_bkey_invalid(c, bkey_i_to_s_c(i->k), i->iter->btree_id));
+	       bch2_bkey_invalid(c, bkey_i_to_s_c(insert), iter->btree_id));
 }
 
 static noinline int
@@ -344,11 +316,12 @@ static inline int bch2_trans_journal_res_get(struct btree_trans *trans,
 
 static enum btree_insert_ret
 btree_key_can_insert(struct btree_trans *trans,
-		     struct btree_insert_entry *insert,
+		     struct btree_iter *iter,
+		     struct bkey_i *insert,
 		     unsigned *u64s)
 {
 	struct bch_fs *c = trans->c;
-	struct btree *b = insert->iter->l[0].b;
+	struct btree *b = iter->l[0].b;
 	static enum btree_insert_ret ret;
 
 	if (unlikely(btree_node_fake(b)))
@@ -356,7 +329,7 @@ btree_key_can_insert(struct btree_trans *trans,
 
 	ret = !btree_node_is_extents(b)
 		? BTREE_INSERT_OK
-		: bch2_extent_can_insert(trans, insert, u64s);
+		: bch2_extent_can_insert(trans, iter, insert, u64s);
 	if (ret)
 		return ret;
 
@@ -367,21 +340,22 @@ btree_key_can_insert(struct btree_trans *trans,
 }
 
 static inline void do_btree_insert_one(struct btree_trans *trans,
-				       struct btree_insert_entry *insert)
+				       struct btree_iter *iter,
+				       struct bkey_i *insert)
 {
-	btree_insert_key_leaf(trans, insert);
+	btree_insert_key_leaf(trans, iter, insert);
 }
 
-static inline bool update_has_trans_triggers(struct btree_insert_entry *i)
+static inline bool iter_has_trans_triggers(struct btree_iter *iter)
 {
-	return BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->iter->btree_id);
+	return BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << iter->btree_id);
 }
 
-static inline bool update_has_nontrans_triggers(struct btree_insert_entry *i)
+static inline bool iter_has_nontrans_triggers(struct btree_iter *iter)
 {
 	return (BTREE_NODE_TYPE_HAS_TRIGGERS &
 		~BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS) &
-		(1U << i->iter->btree_id);
+		(1U << iter->btree_id);
 }
 
 static noinline void bch2_btree_iter_unlock_noinline(struct btree_iter *iter)
@@ -393,17 +367,11 @@ static noinline void bch2_trans_mark_gc(struct btree_trans *trans)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_insert_entry *i;
-	unsigned mark_flags = trans->flags & BTREE_INSERT_BUCKET_INVALIDATE
-		? BCH_BUCKET_MARK_BUCKET_INVALIDATE
-		: 0;
-
-	if (unlikely(trans->flags & BTREE_INSERT_NOMARK))
-		return;
 
 	trans_for_each_update(trans, i)
 		if (gc_visited(c, gc_pos_btree_node(i->iter->l[0].b)))
-			bch2_mark_update(trans, i, NULL,
+			bch2_mark_update(trans, i->iter, i->k, NULL,
-					 mark_flags|BCH_BUCKET_MARK_GC);
+					 i->trigger_flags|BTREE_TRIGGER_GC);
 }
 
 static inline int
@@ -413,10 +381,7 @@ bch2_trans_commit_write_locked(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct bch_fs_usage *fs_usage = NULL;
 	struct btree_insert_entry *i;
-	unsigned mark_flags = trans->flags & BTREE_INSERT_BUCKET_INVALIDATE
+	unsigned u64s = 0;
-		? BCH_BUCKET_MARK_BUCKET_INVALIDATE
-		: 0;
-	unsigned iter, u64s = 0;
 	bool marking = false;
 	int ret;
 
@@ -433,13 +398,13 @@ bch2_trans_commit_write_locked(struct btree_trans *trans,
 
 	prefetch(&trans->c->journal.flags);
 
-	trans_for_each_update_sorted(trans, i, iter) {
+	trans_for_each_update(trans, i) {
 		/* Multiple inserts might go to same leaf: */
-		if (!same_leaf_as_prev(trans, iter))
+		if (!same_leaf_as_prev(trans, i))
 			u64s = 0;
 
 		u64s += i->k->k.u64s;
-		ret = btree_key_can_insert(trans, i, &u64s);
+		ret = btree_key_can_insert(trans, i->iter, i->k, &u64s);
 		if (ret) {
 			*stopped_at = i;
 			return ret;
@@ -488,9 +453,9 @@ bch2_trans_commit_write_locked(struct btree_trans *trans,
 	}
 
 	trans_for_each_update(trans, i)
-		if (likely(!(trans->flags & BTREE_INSERT_NOMARK)) &&
+		if (iter_has_nontrans_triggers(i->iter))
-		    update_has_nontrans_triggers(i))
+			bch2_mark_update(trans, i->iter, i->k,
-			bch2_mark_update(trans, i, fs_usage, mark_flags);
+					 fs_usage, i->trigger_flags);
 
 	if (marking)
 		bch2_trans_fs_usage_apply(trans, fs_usage);
@@ -499,7 +464,7 @@ bch2_trans_commit_write_locked(struct btree_trans *trans,
 		bch2_trans_mark_gc(trans);
 
 	trans_for_each_update(trans, i)
-		do_btree_insert_one(trans, i);
+		do_btree_insert_one(trans, i->iter, i->k);
 err:
 	if (marking) {
 		bch2_fs_usage_scratch_put(c, fs_usage);
@@ -517,7 +482,6 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans,
 {
 	struct btree_insert_entry *i;
 	struct btree_iter *iter;
-	unsigned idx;
 	int ret;
 
 	trans_for_each_update(trans, i)
@@ -549,24 +513,18 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans,
 
 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG))
 		trans_for_each_update(trans, i)
-			btree_insert_entry_checks(trans, i);
+			btree_insert_entry_checks(trans, i->iter, i->k);
 	bch2_btree_trans_verify_locks(trans);
 
-	/*
+	trans_for_each_update(trans, i)
-	 * No more updates can be added - sort updates so we can take write
+		if (!same_leaf_as_prev(trans, i))
-	 * locks in the correct order:
-	 */
-	btree_trans_sort_updates(trans);
-
-	trans_for_each_update_sorted(trans, i, idx)
-		if (!same_leaf_as_prev(trans, idx))
 			bch2_btree_node_lock_for_insert(trans->c,
 						i->iter->l[0].b, i->iter);
 
 	ret = bch2_trans_commit_write_locked(trans, stopped_at);
 
-	trans_for_each_update_sorted(trans, i, idx)
+	trans_for_each_update(trans, i)
-		if (!same_leaf_as_prev(trans, idx))
+		if (!same_leaf_as_prev(trans, i))
 			bch2_btree_node_unlock_write_inlined(i->iter->l[0].b,
 							     i->iter);
 
@@ -582,8 +540,8 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans,
 	if (trans->flags & BTREE_INSERT_NOUNLOCK)
 		trans->nounlock = true;
 
-	trans_for_each_update_sorted(trans, i, idx)
+	trans_for_each_update(trans, i)
-		if (!same_leaf_as_prev(trans, idx))
+		if (!same_leaf_as_prev(trans, i))
 			bch2_foreground_maybe_merge(trans->c, i->iter,
 						    0, trans->flags);
 
@@ -716,9 +674,12 @@ int __bch2_trans_commit(struct btree_trans *trans)
 {
 	struct btree_insert_entry *i = NULL;
 	struct btree_iter *iter;
+	bool trans_trigger_run;
 	unsigned u64s;
 	int ret = 0;
 
+	BUG_ON(trans->need_reset);
+
 	if (!trans->nr_updates)
 		goto out_noupdates;
 
@@ -738,29 +699,39 @@ int __bch2_trans_commit(struct btree_trans *trans)
 	}
 
 	/*
-	 * note: running triggers will append more updates to the list of
+	 * Running triggers will append more updates to the list of updates as
-	 * updates as we're walking it:
+	 * we're walking it:
 	 */
-	trans_for_each_update(trans, i) {
+	do {
-		/* we know trans->nounlock won't be set here: */
+		trans_trigger_run = false;
-		if (unlikely(!(i->iter->locks_want < 1
-			       ? __bch2_btree_iter_upgrade(i->iter, 1)
-			       : i->iter->uptodate <= BTREE_ITER_NEED_PEEK))) {
-			trace_trans_restart_upgrade(trans->ip);
-			ret = -EINTR;
-			goto out;
-		}
 
-		if (likely(!(trans->flags & BTREE_INSERT_NOMARK)) &&
+		trans_for_each_update(trans, i) {
-		    update_has_trans_triggers(i)) {
+			/* we know trans->nounlock won't be set here: */
-			ret = bch2_trans_mark_update(trans, i->iter, i->k);
+			if (unlikely(!(i->iter->locks_want < 1
-			if (unlikely(ret)) {
+				       ? __bch2_btree_iter_upgrade(i->iter, 1)
-				if (ret == -EINTR)
+				       : i->iter->uptodate <= BTREE_ITER_NEED_PEEK))) {
-					trace_trans_restart_mark(trans->ip);
+				trace_trans_restart_upgrade(trans->ip);
+				ret = -EINTR;
 				goto out;
 			}
-		}
 
+			if (iter_has_trans_triggers(i->iter) &&
+			    !i->trans_triggers_run) {
+				i->trans_triggers_run = true;
+				trans_trigger_run = true;
+
+				ret = bch2_trans_mark_update(trans, i->iter, i->k,
+							     i->trigger_flags);
+				if (unlikely(ret)) {
+					if (ret == -EINTR)
+						trace_trans_restart_mark(trans->ip);
+					goto out;
+				}
+			}
+		}
+	} while (trans_trigger_run);
+
+	trans_for_each_update(trans, i) {
 		u64s = jset_u64s(i->k->k.u64s);
 		if (0)
 			trans->journal_preres_u64s += u64s;
@@ -776,28 +747,22 @@ retry:
 
 	if (ret)
 		goto err;
+
+	trans_for_each_iter(trans, iter)
+		if ((trans->iters_live & (1ULL << iter->idx)) &&
+		    (iter->flags & BTREE_ITER_SET_POS_AFTER_COMMIT)) {
+			if (trans->flags & BTREE_INSERT_NOUNLOCK)
+				bch2_btree_iter_set_pos_same_leaf(iter, iter->pos_after_commit);
+			else
+				bch2_btree_iter_set_pos(iter, iter->pos_after_commit);
+		}
 out:
 	bch2_journal_preres_put(&trans->c->journal, &trans->journal_preres);
 
 	if (likely(!(trans->flags & BTREE_INSERT_NOCHECK_RW)))
 		percpu_ref_put(&trans->c->writes);
 out_noupdates:
-	trans_for_each_iter_all(trans, iter)
+	bch2_trans_reset(trans, TRANS_RESET_MEM|TRANS_RESET_NOTRAVERSE);
-		iter->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
-
-	if (!ret) {
-		bch2_trans_unlink_iters(trans);
-		trans->iters_touched = 0;
-	}
-	trans->nr_updates	= 0;
-	trans->mem_top		= 0;
-
-	if (trans->fs_usage_deltas) {
-		trans->fs_usage_deltas->used = 0;
-		memset(&trans->fs_usage_deltas->memset_start, 0,
-		       (void *) &trans->fs_usage_deltas->memset_end -
-		       (void *) &trans->fs_usage_deltas->memset_start);
-	}
 
 	return ret;
 err:
@@ -808,6 +773,76 @@ err:
 	goto retry;
 }
 
+int bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
+		      struct bkey_i *k, enum btree_trigger_flags flags)
+{
+	struct btree_insert_entry *i, n = (struct btree_insert_entry) {
+		.trigger_flags = flags, .iter = iter, .k = k
+	};
+
+	EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&k->k)));
+
+	iter->flags |= BTREE_ITER_KEEP_UNTIL_COMMIT;
+
+	if (iter->flags & BTREE_ITER_IS_EXTENTS) {
+		iter->pos_after_commit = k->k.p;
+		iter->flags |= BTREE_ITER_SET_POS_AFTER_COMMIT;
+	}
+
+	/*
+	 * Pending updates are kept sorted: first, find position of new update:
+	 */
+	trans_for_each_update(trans, i)
+		if (btree_iter_cmp(iter, i->iter) <= 0)
+			break;
+
+	/*
+	 * Now delete/trim any updates the new update overwrites:
+	 */
+	if (i > trans->updates &&
+	    i[-1].iter->btree_id == iter->btree_id &&
+	    bkey_cmp(iter->pos, i[-1].k->k.p) < 0)
+		bch2_cut_back(n.iter->pos, i[-1].k);
+
+	while (i < trans->updates + trans->nr_updates &&
+	       iter->btree_id == i->iter->btree_id &&
+	       bkey_cmp(n.k->k.p, i->k->k.p) >= 0)
+		array_remove_item(trans->updates, trans->nr_updates,
+				  i - trans->updates);
+
+	if (i < trans->updates + trans->nr_updates &&
+	    iter->btree_id == i->iter->btree_id &&
+	    bkey_cmp(n.k->k.p, i->iter->pos) > 0) {
+		/*
+		 * When we have an extent that overwrites the start of another
+		 * update, trimming that extent will mean the iterator's
+		 * position has to change since the iterator position has to
+		 * match the extent's start pos - but we don't want to change
+		 * the iterator pos if some other code is using it, so we may
+		 * need to clone it:
+		 */
+		if (trans->iters_live & (1ULL << i->iter->idx)) {
+			i->iter = bch2_trans_copy_iter(trans, i->iter);
+			if (IS_ERR(i->iter)) {
+				trans->need_reset = true;
+				return PTR_ERR(i->iter);
+			}
+
+			i->iter->flags |= BTREE_ITER_KEEP_UNTIL_COMMIT;
+			bch2_trans_iter_put(trans, i->iter);
+		}
+
+		bch2_cut_front(n.k->k.p, i->k);
+		bch2_btree_iter_set_pos(i->iter, n.k->k.p);
+	}
+
+	EBUG_ON(trans->nr_updates >= trans->nr_iters);
+
+	array_insert_item(trans->updates, trans->nr_updates,
+			  i - trans->updates, n);
+	return 0;
+}
+
 static int __bch2_btree_insert(struct btree_trans *trans,
 			       enum btree_id id, struct bkey_i *k)
 {
@@ -818,7 +853,7 @@ static int __bch2_btree_insert(struct btree_trans *trans,
 	if (IS_ERR(iter))
 		return PTR_ERR(iter);
 
-	bch2_trans_update(trans, iter, k);
+	bch2_trans_update(trans, iter, k, 0);
 	return 0;
 }
 
@@ -880,7 +915,7 @@ retry:
 				break;
 		}
 
-		bch2_trans_update(trans, iter, &delete);
+		bch2_trans_update(trans, iter, &delete, 0);
 		ret = bch2_trans_commit(trans, NULL, journal_seq,
 					BTREE_INSERT_NOFAIL);
 		if (ret)
@@ -906,7 +941,7 @@ int bch2_btree_delete_at(struct btree_trans *trans,
 	bkey_init(&k.k);
 	k.k.p = iter->pos;
 
-	bch2_trans_update(trans, iter, &k);
+	bch2_trans_update(trans, iter, &k, 0);
 	return bch2_trans_commit(trans, NULL, NULL,
 				 BTREE_INSERT_NOFAIL|
 				 BTREE_INSERT_USE_RESERVE|flags);

libbcachefs/buckets.c

@@ -628,7 +628,7 @@ unwind:
 	percpu_rwsem_assert_held(&c->mark_lock);			\
 									\
 	for (gc = 0; gc < 2 && !ret; gc++)				\
-		if (!gc == !(flags & BCH_BUCKET_MARK_GC) ||		\
+		if (!gc == !(flags & BTREE_TRIGGER_GC) ||		\
 		    (gc && gc_visited(c, pos)))				\
 			ret = fn(c, __VA_ARGS__, gc);			\
 	ret;								\
@@ -710,7 +710,7 @@ static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k,
 			   struct bch_fs_usage *fs_usage,
 			   u64 journal_seq, unsigned flags)
 {
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool gc = flags & BTREE_TRIGGER_GC;
 	struct bkey_alloc_unpacked u;
 	struct bch_dev *ca;
 	struct bucket *g;
@@ -719,8 +719,8 @@ static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k,
 	/*
 	 * alloc btree is read in by bch2_alloc_read, not gc:
 	 */
-	if ((flags & BCH_BUCKET_MARK_GC) &&
+	if ((flags & BTREE_TRIGGER_GC) &&
-	    !(flags & BCH_BUCKET_MARK_BUCKET_INVALIDATE))
+	    !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
 		return 0;
 
 	ca = bch_dev_bkey_exists(c, k.k->p.inode);
@@ -743,7 +743,7 @@ static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k,
 		}
 	}));
 
-	if (!(flags & BCH_BUCKET_MARK_ALLOC_READ))
+	if (!(flags & BTREE_TRIGGER_ALLOC_READ))
 		bch2_dev_usage_update(c, ca, fs_usage, old, m, gc);
 
 	g->io_time[READ]	= u.read_time;
@@ -756,7 +756,7 @@ static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k,
 	 * not:
 	 */
 
-	if ((flags & BCH_BUCKET_MARK_BUCKET_INVALIDATE) &&
+	if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
 	    old.cached_sectors) {
 		update_cached_sectors(c, fs_usage, ca->dev_idx,
 				      -old.cached_sectors);
@@ -842,13 +842,13 @@ static s64 __ptr_disk_sectors_delta(unsigned old_size,
 {
 	BUG_ON(!n || !d);
 
-	if (flags & BCH_BUCKET_MARK_OVERWRITE_SPLIT) {
+	if (flags & BTREE_TRIGGER_OVERWRITE_SPLIT) {
 		BUG_ON(offset + -delta > old_size);
 
 		return -disk_sectors_scaled(n, d, old_size) +
 			disk_sectors_scaled(n, d, offset) +
 			disk_sectors_scaled(n, d, old_size - offset + delta);
-	} else if (flags & BCH_BUCKET_MARK_OVERWRITE) {
+	} else if (flags & BTREE_TRIGGER_OVERWRITE) {
 		BUG_ON(offset + -delta > old_size);
 
 		return -disk_sectors_scaled(n, d, old_size) +
@@ -874,8 +874,8 @@ static void bucket_set_stripe(struct bch_fs *c,
 			      u64 journal_seq,
 			      unsigned flags)
 {
-	bool enabled = !(flags & BCH_BUCKET_MARK_OVERWRITE);
+	bool enabled = !(flags & BTREE_TRIGGER_OVERWRITE);
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool gc = flags & BTREE_TRIGGER_GC;
 	unsigned i;
 
 	for (i = 0; i < v->nr_blocks; i++) {
@@ -922,7 +922,7 @@ static bool bch2_mark_pointer(struct bch_fs *c,
 			      struct bch_fs_usage *fs_usage,
 			      u64 journal_seq, unsigned flags)
 {
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool gc = flags & BTREE_TRIGGER_GC;
 	struct bucket_mark old, new;
 	struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
 	struct bucket *g = PTR_BUCKET(ca, &p.ptr, gc);
@@ -970,7 +970,7 @@ static bool bch2_mark_pointer(struct bch_fs *c,
 			new.data_type = data_type;
 		}
 
-		if (flags & BCH_BUCKET_MARK_NOATOMIC) {
+		if (flags & BTREE_TRIGGER_NOATOMIC) {
 			g->_mark = new;
 			break;
 		}
@@ -1008,7 +1008,7 @@ static int bch2_mark_stripe_ptr(struct bch_fs *c,
 				unsigned *nr_data,
 				unsigned *nr_parity)
 {
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool gc = flags & BTREE_TRIGGER_GC;
 	struct stripe *m;
 	unsigned old, new;
 	int blocks_nonempty_delta;
@@ -1121,7 +1121,7 @@ static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
 			    struct bch_fs_usage *fs_usage,
 			    u64 journal_seq, unsigned flags)
 {
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool gc = flags & BTREE_TRIGGER_GC;
 	struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
 	size_t idx = s.k->p.offset;
 	struct stripe *m = genradix_ptr(&c->stripes[gc], idx);
@@ -1129,14 +1129,14 @@ static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
 
 	spin_lock(&c->ec_stripes_heap_lock);
 
-	if (!m || ((flags & BCH_BUCKET_MARK_OVERWRITE) && !m->alive)) {
+	if (!m || ((flags & BTREE_TRIGGER_OVERWRITE) && !m->alive)) {
 		spin_unlock(&c->ec_stripes_heap_lock);
 		bch_err_ratelimited(c, "error marking nonexistent stripe %zu",
 				    idx);
 		return -1;
 	}
 
-	if (!(flags & BCH_BUCKET_MARK_OVERWRITE)) {
+	if (!(flags & BTREE_TRIGGER_OVERWRITE)) {
 		m->sectors	= le16_to_cpu(s.v->sectors);
 		m->algorithm	= s.v->algorithm;
 		m->nr_blocks	= s.v->nr_blocks;
@@ -1152,7 +1152,7 @@ static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
 #endif
 
 		/* gc recalculates these fields: */
-		if (!(flags & BCH_BUCKET_MARK_GC)) {
+		if (!(flags & BTREE_TRIGGER_GC)) {
 			for (i = 0; i < s.v->nr_blocks; i++) {
 				m->block_sectors[i] =
 					stripe_blockcount_get(s.v, i);
@@ -1185,16 +1185,16 @@ int bch2_mark_key_locked(struct bch_fs *c,
 
 	preempt_disable();
 
-	if (!fs_usage || (flags & BCH_BUCKET_MARK_GC))
+	if (!fs_usage || (flags & BTREE_TRIGGER_GC))
 		fs_usage = fs_usage_ptr(c, journal_seq,
-					flags & BCH_BUCKET_MARK_GC);
+					flags & BTREE_TRIGGER_GC);
 
 	switch (k.k->type) {
 	case KEY_TYPE_alloc:
 		ret = bch2_mark_alloc(c, k, fs_usage, journal_seq, flags);
 		break;
 	case KEY_TYPE_btree_ptr:
-		sectors = !(flags & BCH_BUCKET_MARK_OVERWRITE)
+		sectors = !(flags & BTREE_TRIGGER_OVERWRITE)
 			?  c->opts.btree_node_size
 			: -c->opts.btree_node_size;
 
@@ -1210,7 +1210,7 @@ int bch2_mark_key_locked(struct bch_fs *c,
 		ret = bch2_mark_stripe(c, k, fs_usage, journal_seq, flags);
 		break;
 	case KEY_TYPE_inode:
-		if (!(flags & BCH_BUCKET_MARK_OVERWRITE))
+		if (!(flags & BTREE_TRIGGER_OVERWRITE))
 			fs_usage->nr_inodes++;
 		else
 			fs_usage->nr_inodes--;
@@ -1260,7 +1260,7 @@ inline int bch2_mark_overwrite(struct btree_trans *trans,
 	unsigned		offset = 0;
 	s64			sectors = 0;
 
-	flags |= BCH_BUCKET_MARK_OVERWRITE;
+	flags |= BTREE_TRIGGER_OVERWRITE;
 
 	if (btree_node_is_extents(b)
 	    ? bkey_cmp(new->k.p, bkey_start_pos(old.k)) <= 0
@@ -1288,7 +1288,7 @@ inline int bch2_mark_overwrite(struct btree_trans *trans,
 			offset = bkey_start_offset(&new->k) -
 				bkey_start_offset(old.k);
 			sectors = -((s64) new->k.size);
-			flags |= BCH_BUCKET_MARK_OVERWRITE_SPLIT;
+			flags |= BTREE_TRIGGER_OVERWRITE_SPLIT;
 			break;
 		}
 
@@ -1300,26 +1300,29 @@ inline int bch2_mark_overwrite(struct btree_trans *trans,
 }
 
 int bch2_mark_update(struct btree_trans *trans,
-		     struct btree_insert_entry *insert,
+		     struct btree_iter *iter,
+		     struct bkey_i *insert,
 		     struct bch_fs_usage *fs_usage,
 		     unsigned flags)
 {
 	struct bch_fs		*c = trans->c;
-	struct btree_iter	*iter = insert->iter;
 	struct btree		*b = iter->l[0].b;
 	struct btree_node_iter	node_iter = iter->l[0].iter;
 	struct bkey_packed	*_k;
 	int ret = 0;
 
+	if (unlikely(flags & BTREE_TRIGGER_NORUN))
+		return 0;
+
 	if (!btree_node_type_needs_gc(iter->btree_id))
 		return 0;
 
-	bch2_mark_key_locked(c, bkey_i_to_s_c(insert->k),
+	bch2_mark_key_locked(c, bkey_i_to_s_c(insert),
-		0, insert->k->k.size,
+		0, insert->k.size,
 		fs_usage, trans->journal_res.seq,
-		BCH_BUCKET_MARK_INSERT|flags);
+		BTREE_TRIGGER_INSERT|flags);
 
-	if (unlikely(trans->flags & BTREE_INSERT_NOMARK_OVERWRITES))
+	if (unlikely(flags & BTREE_TRIGGER_NOOVERWRITES))
 		return 0;
 
 	/*
@@ -1328,7 +1331,7 @@ int bch2_mark_update(struct btree_trans *trans,
 	 */
 	if ((iter->btree_id == BTREE_ID_ALLOC ||
 	     iter->btree_id == BTREE_ID_EC) &&
-	    !bkey_deleted(&insert->k->k))
+	    !bkey_deleted(&insert->k))
 		return 0;
 
 	while ((_k = bch2_btree_node_iter_peek_filter(&node_iter, b,
@@ -1336,7 +1339,7 @@ int bch2_mark_update(struct btree_trans *trans,
 		struct bkey		unpacked;
 		struct bkey_s_c		k = bkey_disassemble(b, _k, &unpacked);
 
-		ret = bch2_mark_overwrite(trans, iter, k, insert->k,
+		ret = bch2_mark_overwrite(trans, iter, k, insert,
 					  fs_usage, flags);
 		if (ret <= 0)
 			break;
@@ -1430,30 +1433,6 @@ static int trans_get_key(struct btree_trans *trans,
 	return ret;
 }
 
-static void *trans_update_key(struct btree_trans *trans,
-			      struct btree_iter *iter,
-			      unsigned u64s)
-{
-	struct btree_insert_entry *i;
-	struct bkey_i *new_k;
-
-	new_k = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
-	if (IS_ERR(new_k))
-		return new_k;
-
-	bkey_init(&new_k->k);
-	new_k->k.p = iter->pos;
-
-	trans_for_each_update(trans, i)
-		if (i->iter == iter) {
-			i->k = new_k;
-			return new_k;
-		}
-
-	bch2_trans_update(trans, iter, new_k);
-	return new_k;
-}
-
 static int bch2_trans_mark_pointer(struct btree_trans *trans,
 			struct extent_ptr_decoded p,
 			s64 sectors, enum bch_data_type data_type)
@@ -1537,7 +1516,7 @@ static int bch2_trans_mark_pointer(struct btree_trans *trans,
 	u.data_type = u.dirty_sectors || u.cached_sectors
 		? data_type : 0;
 
-	a = trans_update_key(trans, iter, BKEY_ALLOC_U64s_MAX);
+	a = bch2_trans_kmalloc(trans, BKEY_ALLOC_U64s_MAX * 8);
 	ret = PTR_ERR_OR_ZERO(a);
 	if (ret)
 		goto out;
@@ -1545,6 +1524,7 @@ static int bch2_trans_mark_pointer(struct btree_trans *trans,
 	bkey_alloc_init(&a->k_i);
 	a->k.p = iter->pos;
 	bch2_alloc_pack(a, u);
+	bch2_trans_update(trans, iter, &a->k_i, 0);
 out:
 	bch2_trans_iter_put(trans, iter);
 	return ret;
@@ -1559,9 +1539,8 @@ static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter *iter;
-	struct bkey_i *new_k;
 	struct bkey_s_c k;
-	struct bkey_s_stripe s;
+	struct bkey_i_stripe *s;
 	int ret = 0;
 
 	ret = trans_get_key(trans, BTREE_ID_EC, POS(0, p.idx), &iter, &k);
@@ -1576,21 +1555,21 @@ static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
 		goto out;
 	}
 
-	new_k = trans_update_key(trans, iter, k.k->u64s);
+	s = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
-	ret = PTR_ERR_OR_ZERO(new_k);
+	ret = PTR_ERR_OR_ZERO(s);
 	if (ret)
 		goto out;
 
-	bkey_reassemble(new_k, k);
+	bkey_reassemble(&s->k_i, k);
-	s = bkey_i_to_s_stripe(new_k);
 
-	stripe_blockcount_set(s.v, p.block,
+	stripe_blockcount_set(&s->v, p.block,
-		stripe_blockcount_get(s.v, p.block) +
+		stripe_blockcount_get(&s->v, p.block) +
 		sectors);
 
-	*nr_data	= s.v->nr_blocks - s.v->nr_redundant;
+	*nr_data	= s->v.nr_blocks - s->v.nr_redundant;
-	*nr_parity	= s.v->nr_redundant;
+	*nr_parity	= s->v.nr_redundant;
-	bch2_bkey_to_replicas(&r->e, s.s_c);
+	bch2_bkey_to_replicas(&r->e, bkey_i_to_s_c(&s->k_i));
+	bch2_trans_update(trans, iter, &s->k_i, 0);
 out:
 	bch2_trans_iter_put(trans, iter);
 	return ret;
@@ -1671,7 +1650,6 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter *iter;
-	struct bkey_i *new_k;
 	struct bkey_s_c k;
 	struct bkey_i_reflink_v *r_v;
 	s64 ret;
@@ -1689,7 +1667,7 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 		goto err;
 	}
 
-	if ((flags & BCH_BUCKET_MARK_OVERWRITE) &&
+	if ((flags & BTREE_TRIGGER_OVERWRITE) &&
 	    (bkey_start_offset(k.k) < idx ||
 	     k.k->p.offset > idx + sectors))
 		goto out;
@@ -1697,21 +1675,22 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 	bch2_btree_iter_set_pos(iter, bkey_start_pos(k.k));
 	BUG_ON(iter->uptodate > BTREE_ITER_NEED_PEEK);
 
-	new_k = trans_update_key(trans, iter, k.k->u64s);
+	r_v = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
-	ret = PTR_ERR_OR_ZERO(new_k);
+	ret = PTR_ERR_OR_ZERO(r_v);
 	if (ret)
 		goto err;
 
-	bkey_reassemble(new_k, k);
+	bkey_reassemble(&r_v->k_i, k);
-	r_v = bkey_i_to_reflink_v(new_k);
 
 	le64_add_cpu(&r_v->v.refcount,
-		     !(flags & BCH_BUCKET_MARK_OVERWRITE) ? 1 : -1);
+		     !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1);
 
 	if (!r_v->v.refcount) {
 		r_v->k.type = KEY_TYPE_deleted;
 		set_bkey_val_u64s(&r_v->k, 0);
 	}
+
+	bch2_trans_update(trans, iter, &r_v->k_i, 0);
 out:
 	ret = k.k->p.offset - idx;
 err:
@@ -1750,7 +1729,7 @@ int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c k,
 
 	switch (k.k->type) {
 	case KEY_TYPE_btree_ptr:
-		sectors = !(flags & BCH_BUCKET_MARK_OVERWRITE)
+		sectors = !(flags & BTREE_TRIGGER_OVERWRITE)
 			?  c->opts.btree_node_size
 			: -c->opts.btree_node_size;
 
@@ -1763,7 +1742,7 @@ int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c k,
 	case KEY_TYPE_inode:
 		d = replicas_deltas_realloc(trans, 0);
 
-		if (!(flags & BCH_BUCKET_MARK_OVERWRITE))
+		if (!(flags & BTREE_TRIGGER_OVERWRITE))
 			d->nr_inodes++;
 		else
 			d->nr_inodes--;
@@ -1791,22 +1770,26 @@ int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c k,
 
 int bch2_trans_mark_update(struct btree_trans *trans,
 			   struct btree_iter *iter,
-			   struct bkey_i *insert)
+			   struct bkey_i *insert,
+			   unsigned flags)
 {
 	struct btree		*b = iter->l[0].b;
 	struct btree_node_iter	node_iter = iter->l[0].iter;
 	struct bkey_packed	*_k;
 	int ret;
 
+	if (unlikely(flags & BTREE_TRIGGER_NORUN))
+		return 0;
+
 	if (!btree_node_type_needs_gc(iter->btree_id))
 		return 0;
 
 	ret = bch2_trans_mark_key(trans, bkey_i_to_s_c(insert),
-			0, insert->k.size, BCH_BUCKET_MARK_INSERT);
+			0, insert->k.size, BTREE_TRIGGER_INSERT);
 	if (ret)
 		return ret;
 
-	if (unlikely(trans->flags & BTREE_INSERT_NOMARK_OVERWRITES))
+	if (unlikely(flags & BTREE_TRIGGER_NOOVERWRITES))
 		return 0;
 
 	while ((_k = bch2_btree_node_iter_peek_filter(&node_iter, b,
@@ -1815,7 +1798,7 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 		struct bkey_s_c		k;
 		unsigned		offset = 0;
 		s64			sectors = 0;
-		unsigned		flags = BCH_BUCKET_MARK_OVERWRITE;
+		unsigned		flags = BTREE_TRIGGER_OVERWRITE;
 
 		k = bkey_disassemble(b, _k, &unpacked);
 
@@ -1845,7 +1828,7 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 				offset = bkey_start_offset(&insert->k) -
 					bkey_start_offset(k.k);
 				sectors = -((s64) insert->k.size);
-				flags |= BCH_BUCKET_MARK_OVERWRITE_SPLIT;
+				flags |= BTREE_TRIGGER_OVERWRITE_SPLIT;
 				break;
 			}
 

libbcachefs/buckets.h

@@ -258,14 +258,6 @@ void bch2_mark_metadata_bucket(struct bch_fs *, struct bch_dev *,
 			       size_t, enum bch_data_type, unsigned,
 			       struct gc_pos, unsigned);
 
-#define BCH_BUCKET_MARK_INSERT			(1 << 0)
-#define BCH_BUCKET_MARK_OVERWRITE		(1 << 1)
-#define BCH_BUCKET_MARK_OVERWRITE_SPLIT		(1 << 2)
-#define BCH_BUCKET_MARK_BUCKET_INVALIDATE	(1 << 3)
-#define BCH_BUCKET_MARK_GC			(1 << 4)
-#define BCH_BUCKET_MARK_ALLOC_READ		(1 << 5)
-#define BCH_BUCKET_MARK_NOATOMIC		(1 << 6)
-
 int bch2_mark_key_locked(struct bch_fs *, struct bkey_s_c, unsigned, s64,
 			 struct bch_fs_usage *, u64, unsigned);
 int bch2_mark_key(struct bch_fs *, struct bkey_s_c, unsigned, s64,
@@ -276,17 +268,16 @@ int bch2_fs_usage_apply(struct bch_fs *, struct bch_fs_usage *,
 int bch2_mark_overwrite(struct btree_trans *, struct btree_iter *,
 			struct bkey_s_c, struct bkey_i *,
 			struct bch_fs_usage *, unsigned);
-int bch2_mark_update(struct btree_trans *, struct btree_insert_entry *,
+int bch2_mark_update(struct btree_trans *, struct btree_iter *,
-		     struct bch_fs_usage *, unsigned);
+		     struct bkey_i *, struct bch_fs_usage *, unsigned);
 
 int bch2_replicas_delta_list_apply(struct bch_fs *,
 				   struct bch_fs_usage *,
 				   struct replicas_delta_list *);
 int bch2_trans_mark_key(struct btree_trans *, struct bkey_s_c,
 			unsigned, s64, unsigned);
-int bch2_trans_mark_update(struct btree_trans *,
+int bch2_trans_mark_update(struct btree_trans *, struct btree_iter *iter,
-			   struct btree_iter *iter,
+			   struct bkey_i *insert, unsigned);
-			   struct bkey_i *insert);
 void bch2_trans_fs_usage_apply(struct btree_trans *, struct bch_fs_usage *);
 
 /* disk reservations: */

libbcachefs/dirent.c

@@ -246,7 +246,7 @@ int bch2_dirent_rename(struct btree_trans *trans,
 				 */
 				new_dst->k.p = src_iter->pos;
 				bch2_trans_update(trans, src_iter,
-						  &new_dst->k_i);
+						  &new_dst->k_i, 0);
 				return 0;
 			} else {
 				/* If we're overwriting, we can't insert new_dst
@@ -268,8 +268,8 @@ int bch2_dirent_rename(struct btree_trans *trans,
 		}
 	}
 
-	bch2_trans_update(trans, src_iter, &new_src->k_i);
+	bch2_trans_update(trans, src_iter, &new_src->k_i, 0);
-	bch2_trans_update(trans, dst_iter, &new_dst->k_i);
+	bch2_trans_update(trans, dst_iter, &new_dst->k_i, 0);
 	return 0;
 }
 
@@ -331,7 +331,9 @@ int bch2_empty_dir_trans(struct btree_trans *trans, u64 dir_inum)
 			break;
 		}
 	}
-	bch2_trans_iter_put(trans, iter);
+
+	if (!IS_ERR(iter))
+		bch2_trans_iter_put(trans, iter);
 
 	return ret;
 }

libbcachefs/ec.c

@@ -736,7 +736,7 @@ found_slot:
 
 	stripe->k.p = iter->pos;
 
-	bch2_trans_update(&trans, iter, &stripe->k_i);
+	bch2_trans_update(&trans, iter, &stripe->k_i, 0);
 
 	ret = bch2_trans_commit(&trans, NULL, NULL,
 				BTREE_INSERT_NOFAIL);
@@ -818,7 +818,7 @@ static int ec_stripe_update_ptrs(struct bch_fs *c,
 
 		extent_stripe_ptr_add(e, s, ec_ptr, idx);
 
-		bch2_trans_update(&trans, iter, sk.k);
+		bch2_trans_update(&trans, iter, sk.k, 0);
 
 		ret = bch2_trans_commit(&trans, NULL, NULL,
 					BTREE_INSERT_NOFAIL|
@@ -1230,7 +1230,7 @@ static int __bch2_stripe_write_key(struct btree_trans *trans,
 
 	spin_unlock(&c->ec_stripes_heap_lock);
 
-	bch2_trans_update(trans, iter, &new_key->k_i);
+	bch2_trans_update(trans, iter, &new_key->k_i, 0);
 
 	return bch2_trans_commit(trans, NULL, NULL,
 				 BTREE_INSERT_NOFAIL|flags);
@@ -1316,8 +1316,8 @@ int bch2_stripes_read(struct bch_fs *c, struct journal_keys *journal_keys)
 
 		bch2_mark_key(c, btree ? btree_k : journal_k,
 			      0, 0, NULL, 0,
-			      BCH_BUCKET_MARK_ALLOC_READ|
+			      BTREE_TRIGGER_ALLOC_READ|
-			      BCH_BUCKET_MARK_NOATOMIC);
+			      BTREE_TRIGGER_NOATOMIC);
 
 		if (btree)
 			btree_k = bch2_btree_iter_next(btree_iter);

libbcachefs/error.h

@@ -16,26 +16,6 @@ struct work_struct;
 
 /* Error messages: */
 
-/*
- * Very fatal logic/inconsistency errors: these indicate that we've majorly
- * screwed up at runtime, i.e. it's not likely that it was just caused by the
- * data on disk being inconsistent. These BUG():
- *
- * XXX: audit and convert to inconsistent() checks
- */
-
-#define bch2_fs_bug(c, ...)						\
-do {									\
-	bch_err(c, __VA_ARGS__);					\
-	BUG();								\
-} while (0)
-
-#define bch2_fs_bug_on(cond, c, ...)					\
-do {									\
-	if (cond)							\
-		bch2_fs_bug(c, __VA_ARGS__);				\
-} while (0)
-
 /*
  * Inconsistency errors: The on disk data is inconsistent. If these occur during
  * initial recovery, they don't indicate a bug in the running code - we walk all

libbcachefs/extent_update.c

@@ -166,10 +166,11 @@ int bch2_extent_is_atomic(struct bkey_i *k, struct btree_iter *iter)
 
 enum btree_insert_ret
 bch2_extent_can_insert(struct btree_trans *trans,
-		       struct btree_insert_entry *insert,
+		       struct btree_iter *iter,
+		       struct bkey_i *insert,
 		       unsigned *u64s)
 {
-	struct btree_iter_level *l = &insert->iter->l[0];
+	struct btree_iter_level *l = &iter->l[0];
 	struct btree_node_iter node_iter = l->iter;
 	struct bkey_packed *_k;
 	struct bkey unpacked;
@@ -179,12 +180,12 @@ bch2_extent_can_insert(struct btree_trans *trans,
 						      KEY_TYPE_discard))) {
 		struct bkey_s_c k = bkey_disassemble(l->b, _k, &unpacked);
 		enum bch_extent_overlap overlap =
-			bch2_extent_overlap(&insert->k->k, k.k);
+			bch2_extent_overlap(&insert->k, k.k);
 
-		if (bkey_cmp(bkey_start_pos(k.k), insert->k->k.p) >= 0)
+		if (bkey_cmp(bkey_start_pos(k.k), insert->k.p) >= 0)
 			break;
 
-		overlap = bch2_extent_overlap(&insert->k->k, k.k);
+		overlap = bch2_extent_overlap(&insert->k, k.k);
 
 		/*
 		 * If we're overwriting an existing extent, we may need to emit
@@ -192,8 +193,8 @@ bch2_extent_can_insert(struct btree_trans *trans,
 		 * position:
 		 */
 		if (k.k->needs_whiteout &&
-		    (!bkey_whiteout(&insert->k->k) ||
+		    (!bkey_whiteout(&insert->k) ||
-		     bkey_cmp(k.k->p, insert->k->k.p)))
+		     bkey_cmp(k.k->p, insert->k.p)))
 			*u64s += BKEY_U64s;
 
 		/*
@@ -507,11 +508,10 @@ extent_squash(struct bch_fs *c, struct btree_iter *iter,
  * key insertion needs to continue/be retried.
  */
 void bch2_insert_fixup_extent(struct btree_trans *trans,
-			      struct btree_insert_entry *insert_entry)
+			      struct btree_iter *iter,
+			      struct bkey_i *insert)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *iter	= insert_entry->iter;
-	struct bkey_i *insert	= insert_entry->k;
 	struct btree_iter_level *l = &iter->l[0];
 	struct btree_node_iter node_iter = l->iter;
 	bool do_update		= !bkey_whiteout(&insert->k);

libbcachefs/extent_update.h

@@ -10,9 +10,10 @@ int bch2_extent_trim_atomic(struct bkey_i *, struct btree_iter *);
 int bch2_extent_is_atomic(struct bkey_i *, struct btree_iter *);
 
 enum btree_insert_ret
-bch2_extent_can_insert(struct btree_trans *, struct btree_insert_entry *,
+bch2_extent_can_insert(struct btree_trans *, struct btree_iter *,
-		       unsigned *);
+		       struct bkey_i *, unsigned *);
 void bch2_insert_fixup_extent(struct btree_trans *,
-			      struct btree_insert_entry *);
+			      struct btree_iter *,
+			      struct bkey_i *);
 
 #endif /* _BCACHEFS_EXTENT_UPDATE_H */

libbcachefs/extents.c

@@ -172,14 +172,17 @@ void bch2_btree_ptr_debugcheck(struct bch_fs *c, struct bkey_s_c k)
 	struct bucket_mark mark;
 	struct bch_dev *ca;
 
-	bch2_fs_bug_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
-		       !bch2_bkey_replicas_marked(c, k, false), c,
-		       "btree key bad (replicas not marked in superblock):\n%s",
-		       (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
-
 	if (!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
 		return;
 
+	if (!percpu_down_read_trylock(&c->mark_lock))
+		return;
+
+	bch2_fs_inconsistent_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
+		!bch2_bkey_replicas_marked(c, k, false), c,
+		"btree key bad (replicas not marked in superblock):\n%s",
+		(bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
+
 	bkey_for_each_ptr(ptrs, ptr) {
 		ca = bch_dev_bkey_exists(c, ptr->dev);
 
@@ -194,13 +197,15 @@ void bch2_btree_ptr_debugcheck(struct bch_fs *c, struct bkey_s_c k)
 		    mark.dirty_sectors < c->opts.btree_node_size)
 			goto err;
 	}
-
+out:
+	percpu_up_read(&c->mark_lock);
 	return;
 err:
-	bch2_bkey_val_to_text(&PBUF(buf), c, k);
+	bch2_fs_inconsistent(c, "%s btree pointer %s: bucket %zi gen %i mark %08x",
-	bch2_fs_bug(c, "%s btree pointer %s: bucket %zi gen %i mark %08x",
+		err, (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf),
-		    err, buf, PTR_BUCKET_NR(ca, ptr),
+		PTR_BUCKET_NR(ca, ptr),
-		    mark.gen, (unsigned) mark.v.counter);
+		mark.gen, (unsigned) mark.v.counter);
+	goto out;
 }
 
 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
@@ -223,29 +228,17 @@ void bch2_extent_debugcheck(struct bch_fs *c, struct bkey_s_c k)
 	struct extent_ptr_decoded p;
 	char buf[160];
 
-	/*
+	if (!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
-	 * XXX: we should be doing most/all of these checks at startup time,
-	 * where we check bch2_bkey_invalid() in btree_node_read_done()
-	 *
-	 * But note that we can't check for stale pointers or incorrect gc marks
-	 * until after journal replay is done (it might be an extent that's
-	 * going to get overwritten during replay)
-	 */
-
-	if (percpu_down_read_trylock(&c->mark_lock)) {
-		bch2_fs_bug_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
-			       !bch2_bkey_replicas_marked_locked(c, e.s_c, false), c,
-			       "extent key bad (replicas not marked in superblock):\n%s",
-			       (bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf));
-		percpu_up_read(&c->mark_lock);
-	}
-	/*
-	 * If journal replay hasn't finished, we might be seeing keys
-	 * that will be overwritten by the time journal replay is done:
-	 */
-	if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
 		return;
 
+	if (!percpu_down_read_trylock(&c->mark_lock))
+		return;
+
+	bch2_fs_inconsistent_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
+		!bch2_bkey_replicas_marked_locked(c, e.s_c, false), c,
+		"extent key bad (replicas not marked in superblock):\n%s",
+		(bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf));
+
 	extent_for_each_ptr_decode(e, p, entry) {
 		struct bch_dev *ca	= bch_dev_bkey_exists(c, p.ptr.dev);
 		struct bucket_mark mark = ptr_bucket_mark(ca, &p.ptr);
@@ -255,21 +248,24 @@ void bch2_extent_debugcheck(struct bch_fs *c, struct bkey_s_c k)
 			? mark.cached_sectors
 			: mark.dirty_sectors;
 
-		bch2_fs_bug_on(stale && !p.ptr.cached, c,
+		bch2_fs_inconsistent_on(stale && !p.ptr.cached, c,
-			       "stale dirty pointer (ptr gen %u bucket %u",
+			"stale dirty pointer (ptr gen %u bucket %u",
-			       p.ptr.gen, mark.gen);
+			p.ptr.gen, mark.gen);
 
-		bch2_fs_bug_on(stale > 96, c, "key too stale: %i", stale);
+		bch2_fs_inconsistent_on(stale > 96, c,
+			"key too stale: %i", stale);
 
-		bch2_fs_bug_on(!stale &&
+		bch2_fs_inconsistent_on(!stale &&
-			       (mark.data_type != BCH_DATA_USER ||
+			(mark.data_type != BCH_DATA_USER ||
-				mark_sectors < disk_sectors), c,
+			 mark_sectors < disk_sectors), c,
-			       "extent pointer not marked: %s:\n"
+			"extent pointer not marked: %s:\n"
-			       "type %u sectors %u < %u",
+			"type %u sectors %u < %u",
-			       (bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf),
+			(bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf),
-			       mark.data_type,
+			mark.data_type,
-			       mark_sectors, disk_sectors);
+			mark_sectors, disk_sectors);
 	}
+
+	percpu_up_read(&c->mark_lock);
 }
 
 void bch2_extent_to_text(struct printbuf *out, struct bch_fs *c,

libbcachefs/extents.h

@@ -8,7 +8,6 @@
 
 struct bch_fs;
 struct btree_trans;
-struct btree_insert_entry;
 
 /* extent entries: */
 

libbcachefs/fs-io.c

@@ -2411,7 +2411,7 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 	struct address_space *mapping = inode->v.i_mapping;
 	struct bkey_on_stack copy;
 	struct btree_trans trans;
-	struct btree_iter *src, *dst, *del = NULL;
+	struct btree_iter *src, *dst;
 	loff_t shift, new_size;
 	u64 src_start;
 	int ret;
@@ -2493,7 +2493,7 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 		struct bpos next_pos;
 		struct bpos move_pos = POS(inode->v.i_ino, offset >> 9);
 		struct bpos atomic_end;
-		unsigned commit_flags = 0;
+		unsigned trigger_flags = 0;
 
 		k = insert
 			? bch2_btree_iter_peek_prev(src)
@@ -2541,38 +2541,12 @@ reassemble:
 
 		next_pos = insert ? bkey_start_pos(&delete.k) : delete.k.p;
 
-		/*
-		 * If the new and old keys overlap (because we're moving an
-		 * extent that's bigger than the amount we're collapsing by),
-		 * we need to trim the delete key here so they don't overlap
-		 * because overlaps on insertions aren't handled before
-		 * triggers are run, so the overwrite will get double counted
-		 * by the triggers machinery:
-		 */
-		if (insert &&
-		    bkey_cmp(bkey_start_pos(&copy.k->k), delete.k.p) < 0) {
-			bch2_cut_back(bkey_start_pos(&copy.k->k), &delete);
-		} else if (!insert &&
-			   bkey_cmp(copy.k->k.p,
-				    bkey_start_pos(&delete.k)) > 0) {
-			bch2_cut_front(copy.k->k.p, &delete);
-
-			del = bch2_trans_copy_iter(&trans, src);
-			BUG_ON(IS_ERR_OR_NULL(del));
-
-			bch2_btree_iter_set_pos(del,
-				bkey_start_pos(&delete.k));
-		}
-
-		bch2_trans_update(&trans, dst, copy.k);
-		bch2_trans_update(&trans, del ?: src, &delete);
-
 		if (copy.k->k.size == k.k->size) {
 			/*
 			 * If we're moving the entire extent, we can skip
 			 * running triggers:
 			 */
-			commit_flags |= BTREE_INSERT_NOMARK;
+			trigger_flags |= BTREE_TRIGGER_NORUN;
 		} else {
 			/* We might end up splitting compressed extents: */
 			unsigned nr_ptrs =
@@ -2584,16 +2558,13 @@ reassemble:
 			BUG_ON(ret);
 		}
 
-		ret = bch2_trans_commit(&trans, &disk_res,
+		ret =   bch2_trans_update(&trans, src, &delete, trigger_flags) ?:
-					&inode->ei_journal_seq,
+			bch2_trans_update(&trans, dst, copy.k, trigger_flags) ?:
-					BTREE_INSERT_NOFAIL|
+			bch2_trans_commit(&trans, &disk_res,
-					commit_flags);
+					  &inode->ei_journal_seq,
+					  BTREE_INSERT_NOFAIL);
 		bch2_disk_reservation_put(c, &disk_res);
 bkey_err:
-		if (del)
-			bch2_trans_iter_put(&trans, del);
-		del = NULL;
-
 		if (!ret)
 			bch2_btree_iter_set_pos(src, next_pos);
 

libbcachefs/fsck.c

@@ -192,7 +192,7 @@ static int hash_redo_key(const struct bch_hash_desc desc,
 
 	bkey_init(&delete.k);
 	delete.k.p = k_iter->pos;
-	bch2_trans_update(trans, k_iter, &delete);
+	bch2_trans_update(trans, k_iter, &delete, 0);
 
 	return  bch2_hash_set(trans, desc, &h->info, k_iter->pos.inode,
 			      tmp, BCH_HASH_SET_MUST_CREATE) ?:
@@ -388,7 +388,7 @@ static int check_dirent_hash(struct btree_trans *trans, struct hash_check *h,
 				      BTREE_INSERT_NOFAIL|
 				      BTREE_INSERT_LAZY_RW,
 				      TRANS_RESET_MEM,
-			(bch2_trans_update(trans, iter, &d->k_i), 0));
+			(bch2_trans_update(trans, iter, &d->k_i, 0), 0));
 		if (ret)
 			goto err;
 
@@ -661,7 +661,7 @@ retry:
 					      BTREE_INSERT_NOFAIL|
 					      BTREE_INSERT_LAZY_RW,
 					      TRANS_RESET_MEM,
-				(bch2_trans_update(&trans, iter, &n->k_i), 0));
+				(bch2_trans_update(&trans, iter, &n->k_i, 0), 0));
 			kfree(n);
 			if (ret)
 				goto err;
@@ -1276,7 +1276,7 @@ static int check_inode(struct btree_trans *trans,
 				      BTREE_INSERT_NOFAIL|
 				      BTREE_INSERT_LAZY_RW,
 				      TRANS_RESET_MEM,
-			(bch2_trans_update(trans, iter, &p.inode.k_i), 0));
+			(bch2_trans_update(trans, iter, &p.inode.k_i, 0), 0));
 		if (ret)
 			bch_err(c, "error in fsck: error %i "
 				"updating inode", ret);

libbcachefs/inode.c

@@ -223,7 +223,7 @@ int bch2_inode_write(struct btree_trans *trans,
 		return PTR_ERR(inode_p);
 
 	bch2_inode_pack(inode_p, inode);
-	bch2_trans_update(trans, iter, &inode_p->inode.k_i);
+	bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 	return 0;
 }
 
@@ -411,7 +411,7 @@ again:
 			inode_u->bi_generation	= bkey_generation(k);
 
 			bch2_inode_pack(inode_p, inode_u);
-			bch2_trans_update(trans, iter, &inode_p->inode.k_i);
+			bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 			return 0;
 		}
 	}
@@ -493,7 +493,7 @@ int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
 			delete.v.bi_generation = cpu_to_le32(bi_generation);
 		}
 
-		bch2_trans_update(&trans, iter, &delete.k_i);
+		bch2_trans_update(&trans, iter, &delete.k_i, 0);
 
 		ret = bch2_trans_commit(&trans, NULL, NULL,
 					BTREE_INSERT_NOFAIL);

libbcachefs/io.c

@@ -292,13 +292,13 @@ int bch2_extent_update(struct btree_trans *trans,
 		if (delta || new_i_size) {
 			bch2_inode_pack(&inode_p, &inode_u);
 			bch2_trans_update(trans, inode_iter,
-					  &inode_p.inode.k_i);
+					  &inode_p.inode.k_i, 0);
 		}
 
 		bch2_trans_iter_put(trans, inode_iter);
 	}
 
-	bch2_trans_update(trans, iter, k);
+	bch2_trans_update(trans, iter, k, 0);
 
 	ret = bch2_trans_commit(trans, disk_res, journal_seq,
 				BTREE_INSERT_NOCHECK_RW|
@@ -1738,7 +1738,7 @@ retry:
 	if (!bch2_bkey_narrow_crcs(new.k, new_crc))
 		goto out;
 
-	bch2_trans_update(&trans, iter, new.k);
+	bch2_trans_update(&trans, iter, new.k, 0);
 	ret = bch2_trans_commit(&trans, NULL, NULL,
 				BTREE_INSERT_NOFAIL|
 				BTREE_INSERT_NOWAIT);
@@ -1979,7 +1979,7 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig,
 			goto hole;
 
 		iter.bi_size	= pick.crc.compressed_size << 9;
-		goto noclone;
+		goto get_bio;
 	}
 
 	if (!(flags & BCH_READ_LAST_FRAGMENT) ||
@@ -2026,7 +2026,7 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig,
 		pick.crc.live_size		= bvec_iter_sectors(iter);
 		offset_into_extent		= 0;
 	}
-
+get_bio:
 	if (rbio) {
 		/*
 		 * promote already allocated bounce rbio:
@@ -2064,7 +2064,6 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig,
 		rbio->bio.bi_iter = iter;
 		rbio->split	= true;
 	} else {
-noclone:
 		rbio = orig;
 		rbio->bio.bi_iter = iter;
 		EBUG_ON(bio_flagged(&rbio->bio, BIO_CHAIN));

libbcachefs/migrate.c

@@ -53,9 +53,6 @@ static int __bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags
 	while ((k = bch2_btree_iter_peek(iter)).k &&
 	       !(ret = bkey_err(k))) {
 		if (!bch2_bkey_has_device(k, dev_idx)) {
-			ret = bch2_mark_bkey_replicas(c, k);
-			if (ret)
-				break;
 			bch2_btree_iter_next(iter);
 			continue;
 		}
@@ -76,7 +73,7 @@ static int __bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags
 
 		bch2_btree_iter_set_pos(iter, bkey_start_pos(&sk.k->k));
 
-		bch2_trans_update(&trans, iter, sk.k);
+		bch2_trans_update(&trans, iter, sk.k, 0);
 
 		ret = bch2_trans_commit(&trans, NULL, NULL,
 					BTREE_INSERT_NOFAIL);
@@ -129,34 +126,27 @@ static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 			struct bkey_i_btree_ptr *new_key;
 retry:
 			if (!bch2_bkey_has_device(bkey_i_to_s_c(&b->key),
-						  dev_idx)) {
+						  dev_idx))
-				/*
+				continue;
-				 * we might have found a btree node key we
-				 * needed to update, and then tried to update it
-				 * but got -EINTR after upgrading the iter, but
-				 * then raced and the node is now gone:
-				 */
-				bch2_btree_iter_downgrade(iter);
 
-				ret = bch2_mark_bkey_replicas(c, bkey_i_to_s_c(&b->key));
+			bkey_copy(&tmp.k, &b->key);
-				if (ret)
+			new_key = bkey_i_to_btree_ptr(&tmp.k);
-					goto err;
-			} else {
-				bkey_copy(&tmp.k, &b->key);
-				new_key = bkey_i_to_btree_ptr(&tmp.k);
 
-				ret = drop_dev_ptrs(c, bkey_i_to_s(&new_key->k_i),
+			ret = drop_dev_ptrs(c, bkey_i_to_s(&new_key->k_i),
-						    dev_idx, flags, true);
+					    dev_idx, flags, true);
-				if (ret)
+			if (ret) {
-					goto err;
+				bch_err(c, "Cannot drop device without losing data");
+				goto err;
+			}
 
-				ret = bch2_btree_node_update_key(c, iter, b, new_key);
+			ret = bch2_btree_node_update_key(c, iter, b, new_key);
-				if (ret == -EINTR) {
+			if (ret == -EINTR) {
-					b = bch2_btree_iter_peek_node(iter);
+				b = bch2_btree_iter_peek_node(iter);
-					goto retry;
+				goto retry;
-				}
+			}
-				if (ret)
+			if (ret) {
-					goto err;
+				bch_err(c, "Error updating btree node key: %i", ret);
+				goto err;
 			}
 		}
 		bch2_trans_iter_free(&trans, iter);
@@ -167,9 +157,10 @@ retry:
 		closure_wait_event(&c->btree_interior_update_wait,
 				   !bch2_btree_interior_updates_nr_pending(c) ||
 				   c->btree_roots_dirty);
+		if (c->btree_roots_dirty)
+			bch2_journal_meta(&c->journal);
 		if (!bch2_btree_interior_updates_nr_pending(c))
 			break;
-		bch2_journal_meta(&c->journal);
 	}
 
 	ret = 0;
@@ -184,6 +175,5 @@ err:
 int bch2_dev_data_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 {
 	return bch2_dev_usrdata_drop(c, dev_idx, flags) ?:
-		bch2_dev_metadata_drop(c, dev_idx, flags) ?:
+		bch2_dev_metadata_drop(c, dev_idx, flags);
-		bch2_replicas_gc2(c);
 }

libbcachefs/move.c

@@ -150,7 +150,7 @@ static int bch2_migrate_index_update(struct bch_write_op *op)
 			goto next;
 		}
 
-		bch2_trans_update(&trans, iter, insert);
+		bch2_trans_update(&trans, iter, insert, 0);
 
 		ret = bch2_trans_commit(&trans, &op->res,
 				op_journal_seq(op),

libbcachefs/quota.c

@@ -752,7 +752,7 @@ static int bch2_set_quota(struct super_block *sb, struct kqid qid,
 	if (qdq->d_fieldmask & QC_INO_HARD)
 		new_quota.v.c[Q_INO].hardlimit = cpu_to_le64(qdq->d_ino_hardlimit);
 
-	bch2_trans_update(&trans, iter, &new_quota.k_i);
+	bch2_trans_update(&trans, iter, &new_quota.k_i, 0);
 
 	ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 

libbcachefs/recovery.c

@@ -300,28 +300,24 @@ retry:
 		bch2_cut_front(split_iter->pos, split);
 		bch2_cut_back(atomic_end, split);
 
-		bch2_trans_update(&trans, split_iter, split);
+		bch2_trans_update(&trans, split_iter, split, !remark
+				  ? BTREE_TRIGGER_NORUN
+				  : BTREE_TRIGGER_NOOVERWRITES);
 		bch2_btree_iter_set_pos(iter, split->k.p);
 	} while (bkey_cmp(iter->pos, k->k.p) < 0);
 
 	if (remark) {
 		ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 					  0, -((s64) k->k.size),
-					  BCH_BUCKET_MARK_OVERWRITE) ?:
+					  BTREE_TRIGGER_OVERWRITE);
-		      bch2_trans_commit(&trans, &disk_res, NULL,
+		if (ret)
-					BTREE_INSERT_NOFAIL|
+			goto err;
-					BTREE_INSERT_LAZY_RW|
-					BTREE_INSERT_NOMARK_OVERWRITES);
-	} else {
-		ret = bch2_trans_commit(&trans, &disk_res, NULL,
-					BTREE_INSERT_NOFAIL|
-					BTREE_INSERT_LAZY_RW|
-					BTREE_INSERT_JOURNAL_REPLAY|
-					BTREE_INSERT_NOMARK);
 	}
 
-	if (ret)
+	ret = bch2_trans_commit(&trans, &disk_res, NULL,
-		goto err;
+				BTREE_INSERT_NOFAIL|
+				BTREE_INSERT_LAZY_RW|
+				BTREE_INSERT_JOURNAL_REPLAY);
 err:
 	if (ret == -EINTR)
 		goto retry;
@@ -331,6 +327,30 @@ err:
 	return bch2_trans_exit(&trans) ?: ret;
 }
 
+static int __bch2_journal_replay_key(struct btree_trans *trans,
+				     enum btree_id id, struct bkey_i *k)
+{
+	struct btree_iter *iter;
+
+	iter = bch2_trans_get_iter(trans, id, bkey_start_pos(&k->k),
+				   BTREE_ITER_INTENT);
+	if (IS_ERR(iter))
+		return PTR_ERR(iter);
+
+	bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
+	return 0;
+}
+
+static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
+				   struct bkey_i *k)
+{
+	return bch2_trans_do(c, NULL, NULL,
+			     BTREE_INSERT_NOFAIL|
+			     BTREE_INSERT_LAZY_RW|
+			     BTREE_INSERT_JOURNAL_REPLAY,
+			     __bch2_journal_replay_key(&trans, id, k));
+}
+
 static int bch2_journal_replay(struct bch_fs *c,
 			       struct journal_keys keys)
 {
@@ -348,12 +368,7 @@ static int bch2_journal_replay(struct bch_fs *c,
 		else if (btree_node_type_is_extents(i->btree_id))
 			ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 		else
-			ret = bch2_btree_insert(c, i->btree_id, i->k,
+			ret = bch2_journal_replay_key(c, i->btree_id, i->k);
-						NULL, NULL,
-						BTREE_INSERT_NOFAIL|
-						BTREE_INSERT_LAZY_RW|
-						BTREE_INSERT_JOURNAL_REPLAY|
-						BTREE_INSERT_NOMARK);
 
 		if (ret) {
 			bch_err(c, "journal replay: error %d while replaying key",

libbcachefs/reflink.c

@@ -115,7 +115,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	r_v->v.refcount	= 0;
 	memcpy(r_v->v.start, e->v.start, bkey_val_bytes(&e->k));
 
-	bch2_trans_update(trans, reflink_iter, &r_v->k_i);
+	bch2_trans_update(trans, reflink_iter, &r_v->k_i, 0);
 
 	r_p = bch2_trans_kmalloc(trans, sizeof(*r_p));
 	if (IS_ERR(r_p))
@@ -126,7 +126,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	set_bkey_val_bytes(&r_p->k, sizeof(r_p->v));
 	r_p->v.idx = cpu_to_le64(bkey_start_offset(&r_v->k));
 
-	bch2_trans_update(trans, extent_iter, &r_p->k_i);
+	bch2_trans_update(trans, extent_iter, &r_p->k_i, 0);
 err:
 	if (!IS_ERR(reflink_iter)) {
 		c->reflink_hint = reflink_iter->pos.offset;

libbcachefs/replicas.c

@@ -84,10 +84,10 @@ static void extent_to_replicas(struct bkey_s_c k,
 		if (p.ptr.cached)
 			continue;
 
-		if (p.has_ec)
+		if (!p.has_ec)
+			r->devs[r->nr_devs++] = p.ptr.dev;
+		else
 			r->nr_required = 0;
-
-		r->devs[r->nr_devs++] = p.ptr.dev;
 	}
 }
 

libbcachefs/str_hash.h

@@ -281,7 +281,7 @@ not_found:
 			swap(iter, slot);
 
 		insert->k.p = iter->pos;
-		bch2_trans_update(trans, iter, insert);
+		bch2_trans_update(trans, iter, insert, 0);
 	}
 
 	goto out;
@@ -308,7 +308,7 @@ int bch2_hash_delete_at(struct btree_trans *trans,
 	delete->k.p = iter->pos;
 	delete->k.type = ret ? KEY_TYPE_whiteout : KEY_TYPE_deleted;
 
-	bch2_trans_update(trans, iter, delete);
+	bch2_trans_update(trans, iter, delete, 0);
 	return 0;
 }
 

libbcachefs/super.c

@@ -1416,7 +1416,11 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 
 	mutex_lock(&c->state_lock);
 
-	percpu_ref_put(&ca->ref); /* XXX */
+	/*
+	 * We consume a reference to ca->ref, regardless of whether we succeed
+	 * or fail:
+	 */
+	percpu_ref_put(&ca->ref);
 
 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
 		bch_err(ca, "Cannot remove without losing data");
@@ -1425,11 +1429,6 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 
 	__bch2_dev_read_only(c, ca);
 
-	/*
-	 * XXX: verify that dev_idx is really not in use anymore, anywhere
-	 *
-	 * flag_data_bad() does not check btree pointers
-	 */
 	ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
 	if (ret) {
 		bch_err(ca, "Remove failed: error %i dropping data", ret);
@@ -1442,17 +1441,6 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 		goto err;
 	}
 
-	data = bch2_dev_has_data(c, ca);
-	if (data) {
-		char data_has_str[100];
-
-		bch2_flags_to_text(&PBUF(data_has_str),
-				   bch2_data_types, data);
-		bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
-		ret = -EBUSY;
-		goto err;
-	}
-
 	ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
 				      POS(ca->dev_idx, 0),
 				      POS(ca->dev_idx + 1, 0),
@@ -1467,12 +1455,33 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 	 * (overwritten) keys that point to the device we're removing:
 	 */
 	bch2_journal_flush_all_pins(&c->journal);
+	/*
+	 * hack to ensure bch2_replicas_gc2() clears out entries to this device
+	 */
+	bch2_journal_meta(&c->journal);
 	ret = bch2_journal_error(&c->journal);
 	if (ret) {
 		bch_err(ca, "Remove failed, journal error");
 		goto err;
 	}
 
+	ret = bch2_replicas_gc2(c);
+	if (ret) {
+		bch_err(ca, "Remove failed: error %i from replicas gc", ret);
+		goto err;
+	}
+
+	data = bch2_dev_has_data(c, ca);
+	if (data) {
+		char data_has_str[100];
+
+		bch2_flags_to_text(&PBUF(data_has_str),
+				   bch2_data_types, data);
+		bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
+		ret = -EBUSY;
+		goto err;
+	}
+
 	__bch2_dev_offline(c, ca);
 
 	mutex_lock(&c->sb_lock);

libbcachefs/sysfs.c

@@ -916,8 +916,6 @@ SHOW(bch2_dev)
 			bch2_disk_path_to_text(&out, &c->disk_sb,
 					       ca->mi.group - 1);
 			mutex_unlock(&c->sb_lock);
-		} else {
-			pr_buf(&out, "none");
 		}
 
 		pr_buf(&out, "\n");

libbcachefs/tests.c

@@ -43,7 +43,7 @@ static void test_delete(struct bch_fs *c, u64 nr)
 	ret = bch2_btree_iter_traverse(iter);
 	BUG_ON(ret);
 
-	bch2_trans_update(&trans, iter, &k.k_i);
+	bch2_trans_update(&trans, iter, &k.k_i, 0);
 	ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 	BUG_ON(ret);
 
@@ -75,7 +75,7 @@ static void test_delete_written(struct bch_fs *c, u64 nr)
 	ret = bch2_btree_iter_traverse(iter);
 	BUG_ON(ret);
 
-	bch2_trans_update(&trans, iter, &k.k_i);
+	bch2_trans_update(&trans, iter, &k.k_i, 0);
 	ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 	BUG_ON(ret);
 
@@ -465,7 +465,7 @@ static void rand_mixed(struct bch_fs *c, u64 nr)
 			bkey_cookie_init(&k.k_i);
 			k.k.p = iter->pos;
 
-			bch2_trans_update(&trans, iter, &k.k_i);
+			bch2_trans_update(&trans, iter, &k.k_i, 0);
 			ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 			BUG_ON(ret);
 		}
@@ -509,7 +509,7 @@ static void seq_insert(struct bch_fs *c, u64 nr)
 			   BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
 		insert.k.p = iter->pos;
 
-		bch2_trans_update(&trans, iter, &insert.k_i);
+		bch2_trans_update(&trans, iter, &insert.k_i, 0);
 		ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 		BUG_ON(ret);
 
@@ -548,7 +548,7 @@ static void seq_overwrite(struct bch_fs *c, u64 nr)
 
 		bkey_reassemble(&u.k_i, k);
 
-		bch2_trans_update(&trans, iter, &u.k_i);
+		bch2_trans_update(&trans, iter, &u.k_i, 0);
 		ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 		BUG_ON(ret);
 	}