1 files changed, 191 insertions, 216 deletions

diff --git a/fs/bcachefs/dirent.c b/fs/bcachefs/dirent.c
index df9913f8967b..1442dacef0de 100644
--- a/fs/bcachefs/dirent.c
+++ b/fs/bcachefs/dirent.c
@@ -1,3 +1,4 @@
+// SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
 #include "bkey_methods.h"
@@ -12,17 +13,10 @@
 
 unsigned bch2_dirent_name_bytes(struct bkey_s_c_dirent d)
 {
-	unsigned len = bkey_val_bytes(d.k) - sizeof(struct bch_dirent);
+	unsigned len = bkey_val_bytes(d.k) -
+		offsetof(struct bch_dirent, d_name);
 
-	while (len && !d.v->d_name[len - 1])
-		--len;
-
-	return len;
-}
-
-static unsigned dirent_val_u64s(unsigned len)
-{
-	return DIV_ROUND_UP(sizeof(struct bch_dirent) + len, sizeof(u64));
+	return strnlen(d.v->d_name, len);
 }
 
 static u64 bch2_dirent_hash(const struct bch_hash_info *info,
@@ -71,8 +65,7 @@ static bool dirent_cmp_bkey(struct bkey_s_c _l, struct bkey_s_c _r)
 
 const struct bch_hash_desc bch2_dirent_hash_desc = {
 	.btree_id	= BTREE_ID_DIRENTS,
-	.key_type	= BCH_DIRENT,
-	.whiteout_type	= BCH_DIRENT_WHITEOUT,
+	.key_type	= KEY_TYPE_dirent,
 	.hash_key	= dirent_hash_key,
 	.hash_bkey	= dirent_hash_bkey,
 	.cmp_key	= dirent_cmp_key,
@@ -81,69 +74,53 @@ const struct bch_hash_desc bch2_dirent_hash_desc = {
 
 const char *bch2_dirent_invalid(const struct bch_fs *c, struct bkey_s_c k)
 {
-	struct bkey_s_c_dirent d;
+	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 	unsigned len;
 
-	switch (k.k->type) {
-	case BCH_DIRENT:
-		if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
-			return "value too small";
+	if (bkey_val_bytes(k.k) < sizeof(struct bch_dirent))
+		return "value too small";
 
-		d = bkey_s_c_to_dirent(k);
-		len = bch2_dirent_name_bytes(d);
+	len = bch2_dirent_name_bytes(d);
+	if (!len)
+		return "empty name";
 
-		if (!len)
-			return "empty name";
-
-		if (bkey_val_u64s(k.k) > dirent_val_u64s(len))
-			return "value too big";
-
-		if (len > NAME_MAX)
-			return "dirent name too big";
-
-		if (memchr(d.v->d_name, '/', len))
-			return "dirent name has invalid characters";
+	/*
+	 * older versions of bcachefs were buggy and creating dirent
+	 * keys that were bigger than necessary:
+	 */
+	if (bkey_val_u64s(k.k) > dirent_val_u64s(len + 7))
+		return "value too big";
 
-		return NULL;
-	case BCH_DIRENT_WHITEOUT:
-		return bkey_val_bytes(k.k) != 0
-			? "value size should be zero"
-			: NULL;
+	if (len > BCH_NAME_MAX)
+		return "dirent name too big";
 
-	default:
-		return "invalid type";
-	}
+	return NULL;
 }
 
-void bch2_dirent_to_text(struct bch_fs *c, char *buf,
-			 size_t size, struct bkey_s_c k)
+void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c,
+			 struct bkey_s_c k)
 {
-	struct bkey_s_c_dirent d;
-	size_t n = 0;
-
-	switch (k.k->type) {
-	case BCH_DIRENT:
-		d = bkey_s_c_to_dirent(k);
-
-		n += bch_scnmemcpy(buf + n, size - n, d.v->d_name,
-				   bch2_dirent_name_bytes(d));
-		n += scnprintf(buf + n, size - n, " -> %llu", d.v->d_inum);
-		break;
-	case BCH_DIRENT_WHITEOUT:
-		scnprintf(buf, size, "whiteout");
-		break;
-	}
+	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
+
+	bch_scnmemcpy(out, d.v->d_name,
+		      bch2_dirent_name_bytes(d));
+	pr_buf(out, " -> %llu", d.v->d_inum);
 }
 
-static struct bkey_i_dirent *dirent_create_key(u8 type,
-				const struct qstr *name, u64 dst)
+static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans,
+				u8 type, const struct qstr *name, u64 dst)
 {
 	struct bkey_i_dirent *dirent;
 	unsigned u64s = BKEY_U64s + dirent_val_u64s(name->len);
 
-	dirent = kmalloc(u64s * sizeof(u64), GFP_NOFS);
-	if (!dirent)
-		return NULL;
+	if (name->len > BCH_NAME_MAX)
+		return ERR_PTR(-ENAMETOOLONG);
+
+	BUG_ON(u64s > U8_MAX);
+
+	dirent = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
+	if (IS_ERR(dirent))
+		return dirent;
 
 	bkey_dirent_init(&dirent->k_i);
 	dirent->k.u64s = u64s;
@@ -153,30 +130,39 @@ static struct bkey_i_dirent *dirent_create_key(u8 type,
 	memcpy(dirent->v.d_name, name->name, name->len);
 	memset(dirent->v.d_name + name->len, 0,
 	       bkey_val_bytes(&dirent->k) -
-	       (sizeof(struct bch_dirent) + name->len));
+	       offsetof(struct bch_dirent, d_name) -
+	       name->len);
 
 	EBUG_ON(bch2_dirent_name_bytes(dirent_i_to_s_c(dirent)) != name->len);
 
 	return dirent;
 }
 
-int bch2_dirent_create(struct bch_fs *c, u64 dir_inum,
-		       const struct bch_hash_info *hash_info,
-		       u8 type, const struct qstr *name, u64 dst_inum,
-		       u64 *journal_seq, int flags)
+int __bch2_dirent_create(struct btree_trans *trans,
+			 u64 dir_inum, const struct bch_hash_info *hash_info,
+			 u8 type, const struct qstr *name, u64 dst_inum,
+			 int flags)
 {
 	struct bkey_i_dirent *dirent;
 	int ret;
 
-	dirent = dirent_create_key(type, name, dst_inum);
-	if (!dirent)
-		return -ENOMEM;
+	dirent = dirent_create_key(trans, type, name, dst_inum);
+	ret = PTR_ERR_OR_ZERO(dirent);
+	if (ret)
+		return ret;
 
-	ret = bch2_hash_set(bch2_dirent_hash_desc, hash_info, c, dir_inum,
-			   journal_seq, &dirent->k_i, flags);
-	kfree(dirent);
+	return bch2_hash_set(trans, bch2_dirent_hash_desc, hash_info,
+			     dir_inum, &dirent->k_i, flags);
+}
 
-	return ret;
+int bch2_dirent_create(struct bch_fs *c, u64 dir_inum,
+		       const struct bch_hash_info *hash_info,
+		       u8 type, const struct qstr *name, u64 dst_inum,
+		       u64 *journal_seq, int flags)
+{
+	return bch2_trans_do(c, journal_seq, flags,
+		__bch2_dirent_create(&trans, dir_inum, hash_info,
+				     type, name, dst_inum, flags));
 }
 
 static void dirent_copy_target(struct bkey_i_dirent *dst,
@@ -192,147 +178,119 @@ static struct bpos bch2_dirent_pos(struct bch_inode_info *inode,
 	return POS(inode->v.i_ino, bch2_dirent_hash(&inode->ei_str_hash, name));
 }
 
-int bch2_dirent_rename(struct bch_fs *c,
+int bch2_dirent_rename(struct btree_trans *trans,
 		struct bch_inode_info *src_dir, const struct qstr *src_name,
 		struct bch_inode_info *dst_dir, const struct qstr *dst_name,
-		u64 *journal_seq, enum bch_rename_mode mode)
+		enum bch_rename_mode mode)
 {
-	struct btree_iter src_iter, dst_iter, whiteout_iter;
+	struct btree_iter *src_iter, *dst_iter;
 	struct bkey_s_c old_src, old_dst;
-	struct bkey delete;
 	struct bkey_i_dirent *new_src = NULL, *new_dst = NULL;
-	struct bpos src_pos = bch2_dirent_pos(src_dir, src_name);
 	struct bpos dst_pos = bch2_dirent_pos(dst_dir, dst_name);
-	bool need_whiteout;
-	int ret = -ENOMEM;
-
-	bch2_btree_iter_init(&src_iter, c, BTREE_ID_DIRENTS, src_pos,
-			     BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
-	bch2_btree_iter_init(&dst_iter, c, BTREE_ID_DIRENTS, dst_pos,
-			     BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
-	bch2_btree_iter_link(&src_iter, &dst_iter);
-
-	bch2_btree_iter_init(&whiteout_iter, c, BTREE_ID_DIRENTS, src_pos,
-			     BTREE_ITER_SLOTS);
-	bch2_btree_iter_link(&src_iter, &whiteout_iter);
-
-	if (mode == BCH_RENAME_EXCHANGE) {
-		new_src = dirent_create_key(0, src_name, 0);
-		if (!new_src)
-			goto err;
-	} else {
-		new_src = (void *) &delete;
-	}
-
-	new_dst = dirent_create_key(0, dst_name, 0);
-	if (!new_dst)
-		goto err;
-retry:
-	/*
-	 * Note that on -EINTR/dropped locks we're not restarting the lookup
-	 * from the original hashed position (like we do when creating dirents,
-	 * in bch_hash_set) -  we never move existing dirents to different slot:
-	 */
-	old_src = bch2_hash_lookup_at(bch2_dirent_hash_desc,
-				     &src_dir->ei_str_hash,
-				     &src_iter, src_name);
-	if ((ret = btree_iter_err(old_src)))
-		goto err;
-
-	ret = bch2_hash_needs_whiteout(bch2_dirent_hash_desc,
-				&src_dir->ei_str_hash,
-				&whiteout_iter, &src_iter);
-	if (ret < 0)
-		goto err;
-	need_whiteout = ret;
+	int ret;
 
 	/*
+	 * Lookup dst:
+	 *
 	 * Note that in BCH_RENAME mode, we're _not_ checking if
 	 * the target already exists - we're relying on the VFS
 	 * to do that check for us for correctness:
 	 */
-	old_dst = mode == BCH_RENAME
-		? bch2_hash_hole_at(bch2_dirent_hash_desc, &dst_iter)
-		: bch2_hash_lookup_at(bch2_dirent_hash_desc,
-				     &dst_dir->ei_str_hash,
-				     &dst_iter, dst_name);
-	if ((ret = btree_iter_err(old_dst)))
-		goto err;
-
-	switch (mode) {
-	case BCH_RENAME:
+	dst_iter = mode == BCH_RENAME
+		? bch2_hash_hole(trans, bch2_dirent_hash_desc,
+				 &dst_dir->ei_str_hash,
+				 dst_dir->v.i_ino, dst_name)
+		: bch2_hash_lookup(trans, bch2_dirent_hash_desc,
+				   &dst_dir->ei_str_hash,
+				   dst_dir->v.i_ino, dst_name,
+				   BTREE_ITER_INTENT);
+	if (IS_ERR(dst_iter))
+		return PTR_ERR(dst_iter);
+	old_dst = bch2_btree_iter_peek_slot(dst_iter);
+
+	/* Lookup src: */
+	src_iter = bch2_hash_lookup(trans, bch2_dirent_hash_desc,
+				    &src_dir->ei_str_hash,
+				    src_dir->v.i_ino, src_name,
+				    BTREE_ITER_INTENT);
+	if (IS_ERR(src_iter))
+		return PTR_ERR(src_iter);
+	old_src = bch2_btree_iter_peek_slot(src_iter);
+
+	/* Create new dst key: */
+	new_dst = dirent_create_key(trans, 0, dst_name, 0);
+	if (IS_ERR(new_dst))
+		return PTR_ERR(new_dst);
+
+	dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
+	new_dst->k.p = dst_iter->pos;
+
+	/* Create new src key: */
+	if (mode == BCH_RENAME_EXCHANGE) {
+		new_src = dirent_create_key(trans, 0, src_name, 0);
+		if (IS_ERR(new_src))
+			return PTR_ERR(new_src);
+
+		dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
+		new_src->k.p = src_iter->pos;
+	} else {
+		new_src = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
+		if (IS_ERR(new_src))
+			return PTR_ERR(new_src);
 		bkey_init(&new_src->k);
-		dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
+		new_src->k.p = src_iter->pos;
 
-		if (bkey_cmp(dst_pos, src_iter.pos) <= 0 &&
-		    bkey_cmp(src_iter.pos, dst_iter.pos) < 0) {
+		if (bkey_cmp(dst_pos, src_iter->pos) <= 0 &&
+		    bkey_cmp(src_iter->pos, dst_iter->pos) < 0) {
 			/*
-			 * If we couldn't insert new_dst at its hashed
-			 * position (dst_pos) due to a hash collision,
-			 * and we're going to be deleting in
-			 * between the hashed position and first empty
-			 * slot we found - just overwrite the pos we
-			 * were going to delete:
-			 *
-			 * Note: this is a correctness issue, in this
-			 * situation bch2_hash_needs_whiteout() could
-			 * return false when the whiteout would have
-			 * been needed if we inserted at the pos
-			 * __dirent_find_hole() found
+			 * We have a hash collision for the new dst key,
+			 * and new_src - the key we're deleting - is between
+			 * new_dst's hashed slot and the slot we're going to be
+			 * inserting it into - oops.  This will break the hash
+			 * table if we don't deal with it:
 			 */
-			new_dst->k.p = src_iter.pos;
-			ret = bch2_btree_insert_at(c, NULL, NULL,
-					journal_seq,
-					BTREE_INSERT_ATOMIC,
-					BTREE_INSERT_ENTRY(&src_iter,
+			if (mode == BCH_RENAME) {
+				/*
+				 * If we're not overwriting, we can just insert
+				 * new_dst at the src position:
+				 */
+				new_dst->k.p = src_iter->pos;
+				bch2_trans_update(trans,
+					BTREE_INSERT_ENTRY(src_iter,
 							   &new_dst->k_i));
-			goto err;
+				return 0;
+			} else {
+				/* If we're overwriting, we can't insert new_dst
+				 * at a different slot because it has to
+				 * overwrite old_dst - just make sure to use a
+				 * whiteout when deleting src:
+				 */
+				new_src->k.type = KEY_TYPE_whiteout;
+			}
+		} else {
+			/* Check if we need a whiteout to delete src: */
+			ret = bch2_hash_needs_whiteout(trans, bch2_dirent_hash_desc,
+						       &src_dir->ei_str_hash,
+						       src_iter);
+			if (ret < 0)
+				return ret;
+
+			if (ret)
+				new_src->k.type = KEY_TYPE_whiteout;
 		}
-
-		if (need_whiteout)
-			new_src->k.type = BCH_DIRENT_WHITEOUT;
-		break;
-	case BCH_RENAME_OVERWRITE:
-		bkey_init(&new_src->k);
-		dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
-
-		if (bkey_cmp(dst_pos, src_iter.pos) <= 0 &&
-		    bkey_cmp(src_iter.pos, dst_iter.pos) < 0) {
-			/*
-			 * Same case described above -
-			 * bch_hash_needs_whiteout could spuriously
-			 * return false, but we have to insert at
-			 * dst_iter.pos because we're overwriting
-			 * another dirent:
-			 */
-			new_src->k.type = BCH_DIRENT_WHITEOUT;
-		} else if (need_whiteout)
-			new_src->k.type = BCH_DIRENT_WHITEOUT;
-		break;
-	case BCH_RENAME_EXCHANGE:
-		dirent_copy_target(new_src, bkey_s_c_to_dirent(old_dst));
-		dirent_copy_target(new_dst, bkey_s_c_to_dirent(old_src));
-		break;
 	}
 
-	new_src->k.p = src_iter.pos;
-	new_dst->k.p = dst_iter.pos;
-	ret = bch2_btree_insert_at(c, NULL, NULL, journal_seq,
-			BTREE_INSERT_ATOMIC,
-			BTREE_INSERT_ENTRY(&src_iter, &new_src->k_i),
-			BTREE_INSERT_ENTRY(&dst_iter, &new_dst->k_i));
-err:
-	if (ret == -EINTR)
-		goto retry;
-
-	bch2_btree_iter_unlock(&whiteout_iter);
-	bch2_btree_iter_unlock(&dst_iter);
-	bch2_btree_iter_unlock(&src_iter);
-
-	if (new_src != (void *) &delete)
-		kfree(new_src);
-	kfree(new_dst);
-	return ret;
+	bch2_trans_update(trans, BTREE_INSERT_ENTRY(src_iter, &new_src->k_i));
+	bch2_trans_update(trans, BTREE_INSERT_ENTRY(dst_iter, &new_dst->k_i));
+	return 0;
+}
+
+int __bch2_dirent_delete(struct btree_trans *trans, u64 dir_inum,
+			 const struct bch_hash_info *hash_info,
+			 const struct qstr *name)
+{
+	return bch2_hash_delete(trans, bch2_dirent_hash_desc, hash_info,
+				dir_inum, name);
 }
 
 int bch2_dirent_delete(struct bch_fs *c, u64 dir_inum,
@@ -340,66 +298,83 @@ int bch2_dirent_delete(struct bch_fs *c, u64 dir_inum,
 		       const struct qstr *name,
 		       u64 *journal_seq)
 {
-	return bch2_hash_delete(bch2_dirent_hash_desc, hash_info,
-			       c, dir_inum, journal_seq, name);
+	return bch2_trans_do(c, journal_seq,
+			     BTREE_INSERT_ATOMIC|
+			     BTREE_INSERT_NOFAIL,
+		__bch2_dirent_delete(&trans, dir_inum, hash_info, name));
 }
 
 u64 bch2_dirent_lookup(struct bch_fs *c, u64 dir_inum,
 		       const struct bch_hash_info *hash_info,
 		       const struct qstr *name)
 {
-	struct btree_iter iter;
+	struct btree_trans trans;
+	struct btree_iter *iter;
 	struct bkey_s_c k;
-	u64 inum;
+	u64 inum = 0;
 
-	k = bch2_hash_lookup(bch2_dirent_hash_desc, hash_info, c,
-			    dir_inum, &iter, name);
-	if (IS_ERR(k.k)) {
-		bch2_btree_iter_unlock(&iter);
-		return 0;
+	bch2_trans_init(&trans, c, 0, 0);
+
+	iter = bch2_hash_lookup(&trans, bch2_dirent_hash_desc,
+				hash_info, dir_inum, name, 0);
+	if (IS_ERR(iter)) {
+		BUG_ON(PTR_ERR(iter) == -EINTR);
+		goto out;
 	}
 
+	k = bch2_btree_iter_peek_slot(iter);
 	inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
-	bch2_btree_iter_unlock(&iter);
-
+out:
+	bch2_trans_exit(&trans);
 	return inum;
 }
 
-int bch2_empty_dir(struct bch_fs *c, u64 dir_inum)
+int bch2_empty_dir_trans(struct btree_trans *trans, u64 dir_inum)
 {
-	struct btree_iter iter;
+	struct btree_iter *iter;
 	struct bkey_s_c k;
-	int ret = 0;
+	int ret;
 
-	for_each_btree_key(&iter, c, BTREE_ID_DIRENTS, POS(dir_inum, 0), 0, k) {
+	for_each_btree_key(trans, iter, BTREE_ID_DIRENTS,
+			   POS(dir_inum, 0), 0, k, ret) {
 		if (k.k->p.inode > dir_inum)
 			break;
 
-		if (k.k->type == BCH_DIRENT) {
+		if (k.k->type == KEY_TYPE_dirent) {
 			ret = -ENOTEMPTY;
 			break;
 		}
 	}
-	bch2_btree_iter_unlock(&iter);
+	bch2_trans_iter_put(trans, iter);
 
 	return ret;
 }
 
+int bch2_empty_dir(struct bch_fs *c, u64 dir_inum)
+{
+	return bch2_trans_do(c, NULL, 0,
+		bch2_empty_dir_trans(&trans, dir_inum));
+}
+
 int bch2_readdir(struct bch_fs *c, struct file *file,
 		 struct dir_context *ctx)
 {
 	struct bch_inode_info *inode = file_bch_inode(file);
-	struct btree_iter iter;
+	struct btree_trans trans;
+	struct btree_iter *iter;
 	struct bkey_s_c k;
 	struct bkey_s_c_dirent dirent;
 	unsigned len;
+	int ret;
 
 	if (!dir_emit_dots(file, ctx))
 		return 0;
 
-	for_each_btree_key(&iter, c, BTREE_ID_DIRENTS,
-			   POS(inode->v.i_ino, ctx->pos), 0, k) {
-		if (k.k->type != BCH_DIRENT)
+	bch2_trans_init(&trans, c, 0, 0);
+
+	for_each_btree_key(&trans, iter, BTREE_ID_DIRENTS,
+			   POS(inode->v.i_ino, ctx->pos), 0, k, ret) {
+		if (k.k->type != KEY_TYPE_dirent)
 			continue;
 
 		dirent = bkey_s_c_to_dirent(k);
@@ -423,7 +398,7 @@ int bch2_readdir(struct bch_fs *c, struct file *file,
 
 		ctx->pos = k.k->p.offset + 1;
 	}
-	bch2_btree_iter_unlock(&iter);
+	ret = bch2_trans_exit(&trans) ?: ret;
 
-	return 0;
+	return ret;
 }