1 files changed, 121 insertions, 109 deletions

diff --git a/fs/bcachefs/btree_update.c b/fs/bcachefs/btree_update.c
index 5d9e02370aff..581ad19b97a6 100644
--- a/fs/bcachefs/btree_update.c
+++ b/fs/bcachefs/btree_update.c
@@ -95,7 +95,6 @@ static noinline int extent_back_merge(struct btree_trans *trans,
 static int need_whiteout_for_snapshot(struct btree_trans *trans,
 				      enum btree_id btree_id, struct bpos pos)
 {
-	struct btree_iter iter;
 	struct bkey_s_c k;
 	u32 snapshot = pos.snapshot;
 	int ret;
@@ -117,7 +116,6 @@ static int need_whiteout_for_snapshot(struct btree_trans *trans,
 			break;
 		}
 	}
-	bch2_trans_iter_exit(trans, &iter);
 
 	return ret;
 }
@@ -131,10 +129,8 @@ int __bch2_insert_snapshot_whiteouts(struct btree_trans *trans,
 	darray_for_each(*s, id) {
 		pos.snapshot = *id;
 
-		struct btree_iter iter;
-		struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, btree, pos,
-						       BTREE_ITER_not_extents|
-						       BTREE_ITER_intent);
+		CLASS(btree_iter, iter)(trans, btree, pos, BTREE_ITER_not_extents|BTREE_ITER_intent);
+		struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter);
 		ret = bkey_err(k);
 		if (ret)
 			break;
@@ -143,7 +139,6 @@ int __bch2_insert_snapshot_whiteouts(struct btree_trans *trans,
 			struct bkey_i *update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 			ret = PTR_ERR_OR_ZERO(update);
 			if (ret) {
-				bch2_trans_iter_exit(trans, &iter);
 				break;
 			}
 
@@ -154,7 +149,6 @@ int __bch2_insert_snapshot_whiteouts(struct btree_trans *trans,
 			ret = bch2_trans_update(trans, &iter, update,
 						BTREE_UPDATE_internal_snapshot_node);
 		}
-		bch2_trans_iter_exit(trans, &iter);
 
 		if (ret)
 			break;
@@ -164,6 +158,19 @@ int __bch2_insert_snapshot_whiteouts(struct btree_trans *trans,
 	return ret;
 }
 
+static inline enum bch_bkey_type extent_whiteout_type(enum btree_id btree, const struct bkey *k)
+{
+	/*
+	 * KEY_TYPE_extent_whiteout indicates that there isn't a real extent
+	 * present at that position: key start positions inclusive of
+	 * KEY_TYPE_extent_whiteout (but not KEY_TYPE_whiteout) are
+	 * monotonically increasing
+	 */
+	return btree_id_is_extents_snapshots(btree) && bkey_deleted(k)
+		? KEY_TYPE_extent_whiteout
+		: KEY_TYPE_whiteout;
+}
+
 int bch2_trans_update_extent_overwrite(struct btree_trans *trans,
 				       struct btree_iter *iter,
 				       enum btree_iter_update_trigger_flags flags,
@@ -221,7 +228,7 @@ int bch2_trans_update_extent_overwrite(struct btree_trans *trans,
 			return ret;
 	}
 
-	if (bkey_le(old.k->p, new.k->p)) {
+	if (!back_split) {
 		update = bch2_trans_kmalloc(trans, sizeof(*update));
 		if ((ret = PTR_ERR_OR_ZERO(update)))
 			return ret;
@@ -230,23 +237,21 @@ int bch2_trans_update_extent_overwrite(struct btree_trans *trans,
 		update->k.p = old.k->p;
 		update->k.p.snapshot = new.k->p.snapshot;
 
-		if (new.k->p.snapshot != old.k->p.snapshot) {
-			update->k.type = KEY_TYPE_whiteout;
-		} else if (btree_type_has_snapshots(btree_id)) {
-			ret = need_whiteout_for_snapshot(trans, btree_id, update->k.p);
+		if (btree_type_has_snapshots(btree_id)) {
+			ret =   new.k->p.snapshot != old.k->p.snapshot
+				? 1
+				: need_whiteout_for_snapshot(trans, btree_id, update->k.p);
 			if (ret < 0)
 				return ret;
 			if (ret)
-				update->k.type = KEY_TYPE_whiteout;
+				update->k.type = extent_whiteout_type(iter->btree_id, new.k);
 		}
 
 		ret = bch2_btree_insert_nonextent(trans, btree_id, update,
 					  BTREE_UPDATE_internal_snapshot_node|flags);
 		if (ret)
 			return ret;
-	}
-
-	if (back_split) {
+	} else {
 		update = bch2_bkey_make_mut_noupdate(trans, old);
 		if ((ret = PTR_ERR_OR_ZERO(update)))
 			return ret;
@@ -268,18 +273,17 @@ static int bch2_trans_update_extent(struct btree_trans *trans,
 				    struct bkey_i *insert,
 				    enum btree_iter_update_trigger_flags flags)
 {
-	struct btree_iter iter;
-	struct bkey_s_c k;
 	enum btree_id btree_id = orig_iter->btree_id;
-	int ret = 0;
 
-	bch2_trans_iter_init(trans, &iter, btree_id, bkey_start_pos(&insert->k),
-			     BTREE_ITER_intent|
-			     BTREE_ITER_with_updates|
-			     BTREE_ITER_not_extents);
-	k = bch2_btree_iter_peek_max(trans, &iter, POS(insert->k.p.inode, U64_MAX));
-	if ((ret = bkey_err(k)))
-		goto err;
+	CLASS(btree_iter, iter)(trans, btree_id, bkey_start_pos(&insert->k),
+				BTREE_ITER_intent|
+				BTREE_ITER_with_updates|
+				BTREE_ITER_not_extents|
+				BTREE_ITER_nofilter_whiteouts);
+	struct bkey_s_c k = bch2_btree_iter_peek_max(&iter, POS(insert->k.p.inode, U64_MAX));
+	int ret = bkey_err(k);
+	if (ret)
+		return ret;
 	if (!k.k)
 		goto out;
 
@@ -287,26 +291,54 @@ static int bch2_trans_update_extent(struct btree_trans *trans,
 		if (bch2_bkey_maybe_mergable(k.k, &insert->k)) {
 			ret = extent_front_merge(trans, &iter, k, &insert, flags);
 			if (ret)
-				goto err;
+				return ret;
 		}
 
 		goto next;
 	}
 
-	while (bkey_gt(insert->k.p, bkey_start_pos(k.k))) {
-		bool done = bkey_lt(insert->k.p, k.k->p);
+	while (true) {
+		BUG_ON(bkey_le(k.k->p, bkey_start_pos(&insert->k)));
 
-		ret = bch2_trans_update_extent_overwrite(trans, &iter, flags, k, bkey_i_to_s_c(insert));
-		if (ret)
-			goto err;
+		/*
+		 * When KEY_TYPE_whiteout is included, bkey_start_pos is not
+		 * monotonically increasing
+		 */
+		if (k.k->type != KEY_TYPE_whiteout && bkey_le(insert->k.p, bkey_start_pos(k.k)))
+			break;
+
+		bool done = k.k->type != KEY_TYPE_whiteout && bkey_lt(insert->k.p, k.k->p);
+
+		if (bkey_extent_whiteout(k.k)) {
+			enum bch_bkey_type whiteout_type = extent_whiteout_type(btree_id, &insert->k);
+
+			if (bkey_le(k.k->p, insert->k.p) &&
+			    k.k->type != whiteout_type) {
+				struct bkey_i *update = bch2_bkey_make_mut_noupdate(trans, k);
+				ret = PTR_ERR_OR_ZERO(update);
+				if (ret)
+					return ret;
+
+				update->k.p.snapshot = iter.snapshot;
+				update->k.type = whiteout_type;
+
+				ret = bch2_trans_update(trans, &iter, update, 0);
+				if (ret)
+					return ret;
+			}
+		} else {
+			ret = bch2_trans_update_extent_overwrite(trans, &iter, flags, k, bkey_i_to_s_c(insert));
+			if (ret)
+				return ret;
+		}
 
 		if (done)
 			goto out;
 next:
-		bch2_btree_iter_advance(trans, &iter);
-		k = bch2_btree_iter_peek_max(trans, &iter, POS(insert->k.p.inode, U64_MAX));
+		bch2_btree_iter_advance(&iter);
+		k = bch2_btree_iter_peek_max(&iter, POS(insert->k.p.inode, U64_MAX));
 		if ((ret = bkey_err(k)))
-			goto err;
+			return ret;
 		if (!k.k)
 			goto out;
 	}
@@ -314,15 +346,12 @@ next:
 	if (bch2_bkey_maybe_mergable(&insert->k, k.k)) {
 		ret = extent_back_merge(trans, &iter, insert, k);
 		if (ret)
-			goto err;
+			return ret;
 	}
 out:
-	if (!bkey_deleted(&insert->k))
-		ret = bch2_btree_insert_nonextent(trans, btree_id, insert, flags);
-err:
-	bch2_trans_iter_exit(trans, &iter);
-
-	return ret;
+	return !bkey_deleted(&insert->k)
+		? bch2_btree_insert_nonextent(trans, btree_id, insert, flags)
+		: 0;
 }
 
 static inline struct btree_insert_entry *
@@ -594,13 +623,13 @@ int bch2_bkey_get_empty_slot(struct btree_trans *trans, struct btree_iter *iter,
 			     enum btree_id btree, struct bpos end)
 {
 	bch2_trans_iter_init(trans, iter, btree, end, BTREE_ITER_intent);
-	struct bkey_s_c k = bch2_btree_iter_peek_prev(trans, iter);
+	struct bkey_s_c k = bch2_btree_iter_peek_prev(iter);
 	int ret = bkey_err(k);
 	if (ret)
 		goto err;
 
-	bch2_btree_iter_advance(trans, iter);
-	k = bch2_btree_iter_peek_slot(trans, iter);
+	bch2_btree_iter_advance(iter);
+	k = bch2_btree_iter_peek_slot(iter);
 	ret = bkey_err(k);
 	if (ret)
 		goto err;
@@ -614,7 +643,7 @@ int bch2_bkey_get_empty_slot(struct btree_trans *trans, struct btree_iter *iter,
 
 	return 0;
 err:
-	bch2_trans_iter_exit(trans, iter);
+	bch2_trans_iter_exit(iter);
 	return ret;
 }
 
@@ -629,29 +658,21 @@ int bch2_btree_insert_nonextent(struct btree_trans *trans,
 				enum btree_id btree, struct bkey_i *k,
 				enum btree_iter_update_trigger_flags flags)
 {
-	struct btree_iter iter;
-	int ret;
-
-	bch2_trans_iter_init(trans, &iter, btree, k->k.p,
-			     BTREE_ITER_cached|
-			     BTREE_ITER_not_extents|
-			     BTREE_ITER_intent);
-	ret   = bch2_btree_iter_traverse(trans, &iter) ?:
+	CLASS(btree_iter, iter)(trans, btree, k->k.p,
+				BTREE_ITER_cached|
+				BTREE_ITER_not_extents|
+				BTREE_ITER_intent);
+	return  bch2_btree_iter_traverse(&iter) ?:
 		bch2_trans_update(trans, &iter, k, flags);
-	bch2_trans_iter_exit(trans, &iter);
-	return ret;
 }
 
-int bch2_btree_insert_trans(struct btree_trans *trans, enum btree_id id,
+int bch2_btree_insert_trans(struct btree_trans *trans, enum btree_id btree,
 			    struct bkey_i *k, enum btree_iter_update_trigger_flags flags)
 {
-	struct btree_iter iter;
-	bch2_trans_iter_init(trans, &iter, id, bkey_start_pos(&k->k),
-			     BTREE_ITER_intent|flags);
-	int ret = bch2_btree_iter_traverse(trans, &iter) ?:
-		  bch2_trans_update(trans, &iter, k, flags);
-	bch2_trans_iter_exit(trans, &iter);
-	return ret;
+	CLASS(btree_iter, iter)(trans, btree, bkey_start_pos(&k->k),
+				BTREE_ITER_intent|flags);
+	return  bch2_btree_iter_traverse(&iter) ?:
+		bch2_trans_update(trans, &iter, k, flags);
 }
 
 /**
@@ -661,21 +682,23 @@ int bch2_btree_insert_trans(struct btree_trans *trans, enum btree_id id,
  * @k:			key to insert
  * @disk_res:		must be non-NULL whenever inserting or potentially
  *			splitting data extents
- * @flags:		transaction commit flags
+ * @commit_flags:	transaction commit flags
  * @iter_flags:		btree iter update trigger flags
  *
  * Returns:		0 on success, error code on failure
  */
 int bch2_btree_insert(struct bch_fs *c, enum btree_id id, struct bkey_i *k,
-		      struct disk_reservation *disk_res, int flags,
+		      struct disk_reservation *disk_res,
+		      enum bch_trans_commit_flags commit_flags,
 		      enum btree_iter_update_trigger_flags iter_flags)
 {
-	return bch2_trans_commit_do(c, disk_res, NULL, flags,
-			     bch2_btree_insert_trans(trans, id, k, iter_flags));
+	CLASS(btree_trans, trans)(c);
+	return commit_do(trans, disk_res, NULL, commit_flags,
+			 bch2_btree_insert_trans(trans, id, k, iter_flags));
 }
 
-int bch2_btree_delete_at(struct btree_trans *trans,
-			 struct btree_iter *iter, unsigned update_flags)
+int bch2_btree_delete_at(struct btree_trans *trans, struct btree_iter *iter,
+			 enum btree_iter_update_trigger_flags flags)
 {
 	struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k));
 	int ret = PTR_ERR_OR_ZERO(k);
@@ -684,38 +707,32 @@ int bch2_btree_delete_at(struct btree_trans *trans,
 
 	bkey_init(&k->k);
 	k->k.p = iter->pos;
-	return bch2_trans_update(trans, iter, k, update_flags);
+	return bch2_trans_update(trans, iter, k, flags);
 }
 
 int bch2_btree_delete(struct btree_trans *trans,
 		      enum btree_id btree, struct bpos pos,
-		      unsigned update_flags)
+		      enum btree_iter_update_trigger_flags flags)
 {
-	struct btree_iter iter;
-	int ret;
-
-	bch2_trans_iter_init(trans, &iter, btree, pos,
-			     BTREE_ITER_cached|
-			     BTREE_ITER_intent);
-	ret   = bch2_btree_iter_traverse(trans, &iter) ?:
-		bch2_btree_delete_at(trans, &iter, update_flags);
-	bch2_trans_iter_exit(trans, &iter);
-
-	return ret;
+	CLASS(btree_iter, iter)(trans, btree, pos,
+				BTREE_ITER_cached|
+				BTREE_ITER_intent);
+	return  bch2_btree_iter_traverse(&iter) ?:
+		bch2_btree_delete_at(trans, &iter, flags);
 }
 
-int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
+int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id btree,
 				  struct bpos start, struct bpos end,
-				  unsigned update_flags,
+				  enum btree_iter_update_trigger_flags flags,
 				  u64 *journal_seq)
 {
 	u32 restart_count = trans->restart_count;
-	struct btree_iter iter;
 	struct bkey_s_c k;
 	int ret = 0;
 
-	bch2_trans_iter_init(trans, &iter, id, start, BTREE_ITER_intent);
-	while ((k = bch2_btree_iter_peek_max(trans, &iter, end)).k) {
+	CLASS(btree_iter, iter)(trans, btree, start, BTREE_ITER_intent|flags);
+
+	while ((k = bch2_btree_iter_peek_max(&iter, end)).k) {
 		struct disk_reservation disk_res =
 			bch2_disk_reservation_init(trans->c, 0);
 		struct bkey_i delete;
@@ -747,7 +764,7 @@ int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
 					bpos_min(end, k.k->p).offset -
 					iter.pos.offset);
 
-		ret   = bch2_trans_update(trans, &iter, &delete, update_flags) ?:
+		ret   = bch2_trans_update(trans, &iter, &delete, flags) ?:
 			bch2_trans_commit(trans, &disk_res, journal_seq,
 					  BCH_TRANS_COMMIT_no_enospc);
 		bch2_disk_reservation_put(trans->c, &disk_res);
@@ -765,7 +782,6 @@ err:
 		if (ret)
 			break;
 	}
-	bch2_trans_iter_exit(trans, &iter);
 
 	return ret ?: trans_was_restarted(trans, restart_count);
 }
@@ -777,12 +793,11 @@ err:
  */
 int bch2_btree_delete_range(struct bch_fs *c, enum btree_id id,
 			    struct bpos start, struct bpos end,
-			    unsigned update_flags,
+			    enum btree_iter_update_trigger_flags flags,
 			    u64 *journal_seq)
 {
-	int ret = bch2_trans_run(c,
-			bch2_btree_delete_range_trans(trans, id, start, end,
-						      update_flags, journal_seq));
+	CLASS(btree_trans, trans)(c);
+	int ret = bch2_btree_delete_range_trans(trans, id, start, end, flags, journal_seq);
 	if (ret == -BCH_ERR_transaction_restart_nested)
 		ret = 0;
 	return ret;
@@ -807,13 +822,10 @@ int bch2_btree_bit_mod_iter(struct btree_trans *trans, struct btree_iter *iter,
 int bch2_btree_bit_mod(struct btree_trans *trans, enum btree_id btree,
 		       struct bpos pos, bool set)
 {
-	struct btree_iter iter;
-	bch2_trans_iter_init(trans, &iter, btree, pos, BTREE_ITER_intent);
+	CLASS(btree_iter, iter)(trans, btree, pos, BTREE_ITER_intent);
 
-	int ret = bch2_btree_iter_traverse(trans, &iter) ?:
-		  bch2_btree_bit_mod_iter(trans, &iter, set);
-	bch2_trans_iter_exit(trans, &iter);
-	return ret;
+	return  bch2_btree_iter_traverse(&iter) ?:
+		bch2_btree_bit_mod_iter(trans, &iter, set);
 }
 
 int bch2_btree_bit_mod_buffered(struct btree_trans *trans, enum btree_id btree,
@@ -876,31 +888,31 @@ static int
 __bch2_fs_log_msg(struct bch_fs *c, unsigned commit_flags, const char *fmt,
 		  va_list args)
 {
-	struct printbuf buf = PRINTBUF;
+	CLASS(printbuf, buf)();
 	prt_vprintf(&buf, fmt, args);
 
 	unsigned u64s = DIV_ROUND_UP(buf.pos, sizeof(u64));
 
 	int ret = buf.allocation_failure ? -BCH_ERR_ENOMEM_trans_log_msg : 0;
 	if (ret)
-		goto err;
+		return ret;
 
 	if (!test_bit(JOURNAL_running, &c->journal.flags)) {
 		ret = darray_make_room(&c->journal.early_journal_entries, jset_u64s(u64s));
 		if (ret)
-			goto err;
+			return ret;
 
 		struct jset_entry_log *l = (void *) &darray_top(c->journal.early_journal_entries);
 		journal_entry_init(&l->entry, BCH_JSET_ENTRY_log, 0, 1, u64s);
 		memcpy_and_pad(l->d, u64s * sizeof(u64), buf.buf, buf.pos, 0);
 		c->journal.early_journal_entries.nr += jset_u64s(u64s);
 	} else {
-		ret = bch2_trans_commit_do(c, NULL, NULL, commit_flags,
-			bch2_trans_log_msg(trans, &buf));
+		CLASS(btree_trans, trans)(c);
+		ret = commit_do(trans, NULL, NULL, commit_flags,
+				bch2_trans_log_msg(trans, &buf));
 	}
-err:
-	printbuf_exit(&buf);
-	return ret;
+
+	return 0;
 }
 
 __printf(2, 3)