bcachefs: Convert raw uses of bch2_btree_iter_link() to new transactions

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

2 files changed, 130 insertions, 93 deletions

diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c
index 1a28000f20df..e4d2b39e0d82 100644
--- a/fs/bcachefs/fs-io.c
+++ b/fs/bcachefs/fs-io.c
@@ -390,7 +390,8 @@ static int bchfs_write_index_update(struct bch_write_op *wop)
 	struct bchfs_write_op *op = container_of(wop,
 				struct bchfs_write_op, op);
 	struct keylist *keys = &op->op.insert_keys;
-	struct btree_iter extent_iter, inode_iter;
+	struct btree_trans trans;
+	struct btree_iter *extent_iter, *inode_iter = NULL;
 	struct bchfs_extent_trans_hook hook;
 	struct bkey_i *k = bch2_keylist_front(keys);
 	s64 orig_sectors_added = op->sectors_added;
@@ -398,12 +399,13 @@ static int bchfs_write_index_update(struct bch_write_op *wop)
 
 	BUG_ON(k->k.p.inode != op->inode->v.i_ino);
 
-	bch2_btree_iter_init(&extent_iter, wop->c, BTREE_ID_EXTENTS,
-			     bkey_start_pos(&bch2_keylist_front(keys)->k),
-			     BTREE_ITER_INTENT);
-	bch2_btree_iter_init(&inode_iter, wop->c, BTREE_ID_INODES,
-			     POS(extent_iter.pos.inode, 0),
-			     BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
+	bch2_trans_init(&trans, wop->c);
+
+	extent_iter = bch2_trans_get_iter(&trans,
+				BTREE_ID_EXTENTS,
+				bkey_start_pos(&bch2_keylist_front(keys)->k),
+				BTREE_ITER_INTENT);
+	BUG_ON(IS_ERR(extent_iter));
 
 	hook.op			= op;
 	hook.hook.fn		= bchfs_extent_update_hook;
@@ -417,23 +419,28 @@ static int bchfs_write_index_update(struct bch_write_op *wop)
 			hook.need_inode_update = true;
 
 		/* optimization for fewer transaction restarts: */
-		ret = bch2_btree_iter_traverse(&extent_iter);
+		ret = bch2_btree_iter_traverse(extent_iter);
 		if (ret)
 			goto err;
 
 		if (hook.need_inode_update) {
 			struct bkey_s_c inode;
 
-			if (!btree_iter_linked(&inode_iter))
-				bch2_btree_iter_link(&extent_iter, &inode_iter);
+			if (!inode_iter) {
+				inode_iter = bch2_trans_get_iter(&trans,
+					BTREE_ID_INODES,
+					POS(extent_iter->pos.inode, 0),
+					BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
+				BUG_ON(IS_ERR(inode_iter));
+			}
 
-			inode = bch2_btree_iter_peek_slot(&inode_iter);
+			inode = bch2_btree_iter_peek_slot(inode_iter);
 			if ((ret = btree_iter_err(inode)))
 				goto err;
 
 			if (WARN_ONCE(inode.k->type != BCH_INODE_FS,
 				      "inode %llu not found when updating",
-				      extent_iter.pos.inode)) {
+				      extent_iter->pos.inode)) {
 				ret = -ENOENT;
 				break;
 			}
@@ -441,7 +448,7 @@ static int bchfs_write_index_update(struct bch_write_op *wop)
 			if (WARN_ONCE(bkey_bytes(inode.k) >
 				      sizeof(hook.inode_p),
 				      "inode %llu too big (%zu bytes, buf %zu)",
-				      extent_iter.pos.inode,
+				      extent_iter->pos.inode,
 				      bkey_bytes(inode.k),
 				      sizeof(hook.inode_p))) {
 				ret = -ENOENT;
@@ -453,7 +460,7 @@ static int bchfs_write_index_update(struct bch_write_op *wop)
 					       &hook.inode_u);
 			if (WARN_ONCE(ret,
 				      "error %i unpacking inode %llu",
-				      ret, extent_iter.pos.inode)) {
+				      ret, extent_iter->pos.inode)) {
 				ret = -ENOENT;
 				break;
 			}
@@ -463,8 +470,8 @@ static int bchfs_write_index_update(struct bch_write_op *wop)
 					BTREE_INSERT_NOFAIL|
 					BTREE_INSERT_ATOMIC|
 					BTREE_INSERT_USE_RESERVE,
-					BTREE_INSERT_ENTRY(&extent_iter, k),
-					BTREE_INSERT_ENTRY_EXTRA_RES(&inode_iter,
+					BTREE_INSERT_ENTRY(extent_iter, k),
+					BTREE_INSERT_ENTRY_EXTRA_RES(inode_iter,
 							&hook.inode_p.inode.k_i, 2));
 		} else {
 			ret = bch2_btree_insert_at(wop->c, &wop->res,
@@ -472,10 +479,10 @@ static int bchfs_write_index_update(struct bch_write_op *wop)
 					BTREE_INSERT_NOFAIL|
 					BTREE_INSERT_ATOMIC|
 					BTREE_INSERT_USE_RESERVE,
-					BTREE_INSERT_ENTRY(&extent_iter, k));
+					BTREE_INSERT_ENTRY(extent_iter, k));
 		}
 
-		BUG_ON(bkey_cmp(extent_iter.pos, bkey_start_pos(&k->k)));
+		BUG_ON(bkey_cmp(extent_iter->pos, bkey_start_pos(&k->k)));
 
 		if (WARN_ONCE(!ret != !k->k.size,
 			      "ret %i k->size %u", ret, k->k.size))
@@ -486,12 +493,11 @@ err:
 		if (ret)
 			break;
 
-		BUG_ON(bkey_cmp(extent_iter.pos, k->k.p) < 0);
+		BUG_ON(bkey_cmp(extent_iter->pos, k->k.p) < 0);
 		bch2_keylist_pop_front(keys);
 	} while (!bch2_keylist_empty(keys));
 
-	bch2_btree_iter_unlock(&extent_iter);
-	bch2_btree_iter_unlock(&inode_iter);
+	bch2_trans_exit(&trans);
 
 	if (op->is_dio) {
 		struct dio_write *dio = container_of(op, struct dio_write, iop);
@@ -2343,8 +2349,8 @@ static long bch2_fcollapse(struct bch_inode_info *inode,
 {
 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
 	struct address_space *mapping = inode->v.i_mapping;
-	struct btree_iter src;
-	struct btree_iter dst;
+	struct btree_trans trans;
+	struct btree_iter *src, *dst;
 	BKEY_PADDED(k) copy;
 	struct bkey_s_c k;
 	struct i_sectors_hook i_sectors_hook = i_sectors_hook_init(inode, 0);
@@ -2354,13 +2360,17 @@ static long bch2_fcollapse(struct bch_inode_info *inode,
 	if ((offset | len) & (block_bytes(c) - 1))
 		return -EINVAL;
 
-	bch2_btree_iter_init(&dst, c, BTREE_ID_EXTENTS,
+	bch2_trans_init(&trans, c);
+
+	dst = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 			     POS(inode->v.i_ino, offset >> 9),
 			     BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
+	BUG_ON(IS_ERR(dst));
+
 	/* position will be set from dst iter's position: */
-	bch2_btree_iter_init(&src, c, BTREE_ID_EXTENTS, POS_MIN,
+	src = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS, POS_MIN,
 			     BTREE_ITER_SLOTS);
-	bch2_btree_iter_link(&src, &dst);
+	BUG_ON(IS_ERR(src));
 
 	/*
 	 * We need i_mutex to keep the page cache consistent with the extents
@@ -2389,24 +2399,24 @@ static long bch2_fcollapse(struct bch_inode_info *inode,
 	if (ret)
 		goto err;
 
-	while (bkey_cmp(dst.pos,
+	while (bkey_cmp(dst->pos,
 			POS(inode->v.i_ino,
 			    round_up(new_size, PAGE_SIZE) >> 9)) < 0) {
 		struct disk_reservation disk_res;
 
-		bch2_btree_iter_set_pos(&src,
-			POS(dst.pos.inode, dst.pos.offset + (len >> 9)));
+		bch2_btree_iter_set_pos(src,
+			POS(dst->pos.inode, dst->pos.offset + (len >> 9)));
 
-		k = bch2_btree_iter_peek_slot(&src);
+		k = bch2_btree_iter_peek_slot(src);
 		if ((ret = btree_iter_err(k)))
 			goto btree_iter_err;
 
 		bkey_reassemble(&copy.k, k);
 
-		bch2_cut_front(src.pos, &copy.k);
+		bch2_cut_front(src->pos, &copy.k);
 		copy.k.k.p.offset -= len >> 9;
 
-		BUG_ON(bkey_cmp(dst.pos, bkey_start_pos(&copy.k.k)));
+		BUG_ON(bkey_cmp(dst->pos, bkey_start_pos(&copy.k.k)));
 
 		ret = bch2_disk_reservation_get(c, &disk_res, copy.k.k.size,
 				bch2_extent_nr_dirty_ptrs(bkey_i_to_s_c(&copy.k)),
@@ -2417,14 +2427,13 @@ static long bch2_fcollapse(struct bch_inode_info *inode,
 					   &inode->ei_journal_seq,
 					   BTREE_INSERT_ATOMIC|
 					   BTREE_INSERT_NOFAIL,
-					   BTREE_INSERT_ENTRY(&dst, &copy.k));
+					   BTREE_INSERT_ENTRY(dst, &copy.k));
 		bch2_disk_reservation_put(c, &disk_res);
 btree_iter_err:
 		if (ret == -EINTR)
 			ret = 0;
 		if (ret) {
-			bch2_btree_iter_unlock(&src);
-			bch2_btree_iter_unlock(&dst);
+			bch2_trans_exit(&trans);
 			goto err_put_sectors_dirty;
 		}
 		/*
@@ -2432,11 +2441,10 @@ btree_iter_err:
 		 * pointers... which isn't a _super_ serious problem...
 		 */
 
-		bch2_btree_iter_cond_resched(&src);
+		bch2_btree_iter_cond_resched(src);
 	}
 
-	bch2_btree_iter_unlock(&src);
-	bch2_btree_iter_unlock(&dst);
+	bch2_trans_exit(&trans);
 
 	ret = bch2_inode_truncate(c, inode->v.i_ino,
 				 round_up(new_size, block_bytes(c)) >> 9,
diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c
index c36b32531b66..f6035cc7859a 100644
--- a/fs/bcachefs/fsck.c
+++ b/fs/bcachefs/fsck.c
@@ -126,16 +126,22 @@ static int walk_inode(struct bch_fs *c, struct inode_walker *w, u64 inum)
 
 struct hash_check {
 	struct bch_hash_info	info;
-	struct btree_iter	chain;
-	struct btree_iter	iter;
+	struct btree_trans	*trans;
+
+	/* start of current chain of hash collisions: */
+	struct btree_iter	*chain;
+
+	/* next offset in current chain of hash collisions: */
 	u64			next;
 };
 
 static void hash_check_init(const struct bch_hash_desc desc,
-			    struct hash_check *h, struct bch_fs *c)
+			    struct btree_trans *trans,
+			    struct hash_check *h)
 {
-	bch2_btree_iter_init(&h->chain, c, desc.btree_id, POS_MIN, 0);
-	bch2_btree_iter_init(&h->iter, c, desc.btree_id, POS_MIN, 0);
+	h->trans = trans;
+	h->chain = bch2_trans_get_iter(trans, desc.btree_id, POS_MIN, 0);
+	h->next = -1;
 }
 
 static void hash_check_set_inode(struct hash_check *h, struct bch_fs *c,
@@ -206,6 +212,42 @@ err:
 	return ret;
 }
 
+static int hash_check_duplicates(const struct bch_hash_desc desc,
+				 struct hash_check *h, struct bch_fs *c,
+				 struct btree_iter *k_iter, struct bkey_s_c k)
+{
+	struct btree_iter *iter;
+	struct bkey_s_c k2;
+	char buf[200];
+	int ret = 0;
+
+	if (!bkey_cmp(h->chain->pos, k_iter->pos))
+		return 0;
+
+	iter = bch2_trans_copy_iter(h->trans, h->chain);
+	BUG_ON(IS_ERR(iter));
+
+	for_each_btree_key_continue(iter, 0, k2) {
+		if (bkey_cmp(k2.k->p, k.k->p) >= 0)
+			break;
+
+		if (fsck_err_on(k2.k->type == desc.key_type &&
+				!desc.cmp_bkey(k, k2), c,
+				"duplicate hash table keys:\n%s",
+				(bch2_bkey_val_to_text(c, bkey_type(0, desc.btree_id),
+						       buf, sizeof(buf), k), buf))) {
+			ret = fsck_hash_delete_at(desc, &h->info, k_iter);
+			if (ret)
+				return ret;
+			ret = 1;
+			break;
+		}
+	}
+fsck_err:
+	bch2_trans_iter_free(h->trans, iter);
+	return ret;
+}
+
 static int hash_check_key(const struct bch_hash_desc desc,
 			  struct hash_check *h, struct bch_fs *c,
 			  struct btree_iter *k_iter, struct bkey_s_c k)
@@ -218,13 +260,8 @@ static int hash_check_key(const struct bch_hash_desc desc,
 	    k.k->type != desc.key_type)
 		return 0;
 
-	if (k.k->p.offset != h->next) {
-		if (!btree_iter_linked(&h->chain)) {
-			bch2_btree_iter_link(k_iter, &h->chain);
-			bch2_btree_iter_link(k_iter, &h->iter);
-		}
-		bch2_btree_iter_copy(&h->chain, k_iter);
-	}
+	if (k.k->p.offset != h->next)
+		bch2_btree_iter_copy(h->chain, k_iter);
 	h->next = k.k->p.offset + 1;
 
 	if (k.k->type != desc.key_type)
@@ -232,11 +269,11 @@ static int hash_check_key(const struct bch_hash_desc desc,
 
 	hashed = desc.hash_bkey(&h->info, k);
 
-	if (fsck_err_on(hashed < h->chain.pos.offset ||
+	if (fsck_err_on(hashed < h->chain->pos.offset ||
 			hashed > k.k->p.offset, c,
 			"hash table key at wrong offset: %llu, "
 			"hashed to %llu chain starts at %llu\n%s",
-			k.k->p.offset, hashed, h->chain.pos.offset,
+			k.k->p.offset, hashed, h->chain->pos.offset,
 			(bch2_bkey_val_to_text(c, bkey_type(0, desc.btree_id),
 					       buf, sizeof(buf), k), buf))) {
 		ret = hash_redo_key(desc, h, c, k_iter, k, hashed);
@@ -247,25 +284,7 @@ static int hash_check_key(const struct bch_hash_desc desc,
 		return 1;
 	}
 
-	if (!bkey_cmp(h->chain.pos, k_iter->pos))
-		return 0;
-
-	bch2_btree_iter_copy(&h->iter, &h->chain);
-	while (bkey_cmp(h->iter.pos, k_iter->pos) < 0) {
-		struct bkey_s_c k2 = bch2_btree_iter_peek(&h->iter);
-
-		if (fsck_err_on(k2.k->type == desc.key_type &&
-				!desc.cmp_bkey(k, k2), c,
-				"duplicate hash table keys:\n%s",
-				(bch2_bkey_val_to_text(c, bkey_type(0, desc.btree_id),
-						       buf, sizeof(buf), k), buf))) {
-			ret = fsck_hash_delete_at(desc, &h->info, &h->iter);
-			if (ret)
-				return ret;
-			return 1;
-		}
-		bch2_btree_iter_next(&h->iter);
-	}
+	ret = hash_check_duplicates(desc, h, c, k_iter, k);
 fsck_err:
 	return ret;
 }
@@ -367,7 +386,8 @@ static int check_dirents(struct bch_fs *c)
 {
 	struct inode_walker w = inode_walker_init();
 	struct hash_check h;
-	struct btree_iter iter;
+	struct btree_trans trans;
+	struct btree_iter *iter;
 	struct bkey_s_c k;
 	unsigned name_len;
 	char buf[200];
@@ -375,10 +395,16 @@ static int check_dirents(struct bch_fs *c)
 
 	bch_verbose(c, "checking dirents");
 
-	hash_check_init(bch2_dirent_hash_desc, &h, c);
+	bch2_trans_init(&trans, c);
 
-	for_each_btree_key(&iter, c, BTREE_ID_DIRENTS,
-			   POS(BCACHEFS_ROOT_INO, 0), 0, k) {
+	BUG_ON(bch2_trans_preload_iters(&trans));
+
+	iter = bch2_trans_get_iter(&trans, BTREE_ID_DIRENTS,
+				   POS(BCACHEFS_ROOT_INO, 0), 0);
+
+	hash_check_init(bch2_dirent_hash_desc, &trans, &h);
+
+	for_each_btree_key_continue(iter, 0, k) {
 		struct bkey_s_c_dirent d;
 		struct bch_inode_unpacked target;
 		bool have_target;
@@ -397,7 +423,7 @@ static int check_dirents(struct bch_fs *c)
 				mode_to_type(w.inode.bi_mode),
 				(bch2_bkey_val_to_text(c, BTREE_ID_DIRENTS,
 						       buf, sizeof(buf), k), buf))) {
-			ret = bch2_btree_delete_at(&iter, 0);
+			ret = bch2_btree_delete_at(iter, 0);
 			if (ret)
 				goto err;
 			continue;
@@ -406,7 +432,7 @@ static int check_dirents(struct bch_fs *c)
 		if (w.first_this_inode && w.have_inode)
 			hash_check_set_inode(&h, c, &w.inode);
 
-		ret = hash_check_key(bch2_dirent_hash_desc, &h, c, &iter, k);
+		ret = hash_check_key(bch2_dirent_hash_desc, &h, c, iter, k);
 		if (ret > 0) {
 			ret = 0;
 			continue;
@@ -430,7 +456,7 @@ static int check_dirents(struct bch_fs *c)
 		    fsck_err_on(name_len == 2 &&
 				!memcmp(d.v->d_name, "..", 2), c,
 				".. dirent")) {
-			ret = remove_dirent(c, &iter, d);
+			ret = remove_dirent(c, iter, d);
 			if (ret)
 				goto err;
 			continue;
@@ -440,7 +466,7 @@ static int check_dirents(struct bch_fs *c)
 				"dirent points to own directory:\n%s",
 				(bch2_bkey_val_to_text(c, BTREE_ID_DIRENTS,
 						       buf, sizeof(buf), k), buf))) {
-			ret = remove_dirent(c, &iter, d);
+			ret = remove_dirent(c, iter, d);
 			if (ret)
 				goto err;
 			continue;
@@ -457,7 +483,7 @@ static int check_dirents(struct bch_fs *c)
 				"dirent points to missing inode:\n%s",
 				(bch2_bkey_val_to_text(c, BTREE_ID_DIRENTS,
 						       buf, sizeof(buf), k), buf))) {
-			ret = remove_dirent(c, &iter, d);
+			ret = remove_dirent(c, iter, d);
 			if (ret)
 				goto err;
 			continue;
@@ -483,7 +509,7 @@ static int check_dirents(struct bch_fs *c)
 
 			ret = bch2_btree_insert_at(c, NULL, NULL, NULL,
 					BTREE_INSERT_NOFAIL,
-					BTREE_INSERT_ENTRY(&iter, &n->k_i));
+					BTREE_INSERT_ENTRY(iter, &n->k_i));
 			kfree(n);
 			if (ret)
 				goto err;
@@ -492,9 +518,7 @@ static int check_dirents(struct bch_fs *c)
 	}
 err:
 fsck_err:
-	bch2_btree_iter_unlock(&h.chain);
-	bch2_btree_iter_unlock(&h.iter);
-	return bch2_btree_iter_unlock(&iter) ?: ret;
+	return bch2_trans_exit(&trans) ?: ret;
 }
 
 /*
@@ -505,16 +529,23 @@ static int check_xattrs(struct bch_fs *c)
 {
 	struct inode_walker w = inode_walker_init();
 	struct hash_check h;
-	struct btree_iter iter;
+	struct btree_trans trans;
+	struct btree_iter *iter;
 	struct bkey_s_c k;
 	int ret = 0;
 
 	bch_verbose(c, "checking xattrs");
 
-	hash_check_init(bch2_xattr_hash_desc, &h, c);
+	bch2_trans_init(&trans, c);
 
-	for_each_btree_key(&iter, c, BTREE_ID_XATTRS,
-			   POS(BCACHEFS_ROOT_INO, 0), 0, k) {
+	BUG_ON(bch2_trans_preload_iters(&trans));
+
+	iter = bch2_trans_get_iter(&trans, BTREE_ID_XATTRS,
+				   POS(BCACHEFS_ROOT_INO, 0), 0);
+
+	hash_check_init(bch2_xattr_hash_desc, &trans, &h);
+
+	for_each_btree_key_continue(iter, 0, k) {
 		ret = walk_inode(c, &w, k.k->p.inode);
 		if (ret)
 			break;
@@ -522,7 +553,7 @@ static int check_xattrs(struct bch_fs *c)
 		if (fsck_err_on(!w.have_inode, c,
 				"xattr for missing inode %llu",
 				k.k->p.inode)) {
-			ret = bch2_btree_delete_at(&iter, 0);
+			ret = bch2_btree_delete_at(iter, 0);
 			if (ret)
 				goto err;
 			continue;
@@ -531,15 +562,13 @@ static int check_xattrs(struct bch_fs *c)
 		if (w.first_this_inode && w.have_inode)
 			hash_check_set_inode(&h, c, &w.inode);
 
-		ret = hash_check_key(bch2_xattr_hash_desc, &h, c, &iter, k);
+		ret = hash_check_key(bch2_xattr_hash_desc, &h, c, iter, k);
 		if (ret)
 			goto fsck_err;
 	}
 err:
 fsck_err:
-	bch2_btree_iter_unlock(&h.chain);
-	bch2_btree_iter_unlock(&h.iter);
-	return bch2_btree_iter_unlock(&iter) ?: ret;
+	return bch2_trans_exit(&trans) ?: ret;
 }
 
 /* Get root directory, create if it doesn't exist: */