bcachefs: Hook up RENAME_WHITEOUT in rename.

This is needed for overlayfs, which is used by container managers.

Signed-off-by: Sasha Finkelstein <fnkl.kernel@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

4 files changed, 52 insertions, 14 deletions

diff --git a/fs/bcachefs/btree_iter.h b/fs/bcachefs/btree_iter.h
index 6d87e57745da..78e63ad7d380 100644
--- a/fs/bcachefs/btree_iter.h
+++ b/fs/bcachefs/btree_iter.h
@@ -529,6 +529,12 @@ void bch2_set_btree_iter_dontneed(struct btree_iter *);
 
 void *__bch2_trans_kmalloc(struct btree_trans *, size_t);
 
+/**
+ * bch2_trans_kmalloc - allocate memory for use by the current transaction
+ *
+ * Must be called after bch2_trans_begin, which on second and further calls
+ * frees all memory allocated in this transaction
+ */
 static inline void *bch2_trans_kmalloc(struct btree_trans *trans, size_t size)
 {
 	size = roundup(size, 8);
diff --git a/fs/bcachefs/fs-common.c b/fs/bcachefs/fs-common.c
index 508d029ac53d..7e10a9ddcfd9 100644
--- a/fs/bcachefs/fs-common.c
+++ b/fs/bcachefs/fs-common.c
@@ -42,7 +42,8 @@ int bch2_create_trans(struct btree_trans *trans,
 	if (ret)
 		goto err;
 
-	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir, BTREE_ITER_intent);
+	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir,
+			      BTREE_ITER_intent|BTREE_ITER_with_updates);
 	if (ret)
 		goto err;
 
@@ -163,7 +164,7 @@ int bch2_create_trans(struct btree_trans *trans,
 					 name,
 					 dir_target,
 					 &dir_offset,
-					 STR_HASH_must_create);
+					 STR_HASH_must_create|BTREE_ITER_with_updates);
 		if (ret)
 			goto err;
 
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 866ce1f1a6d1..dfadfb82399f 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -736,15 +736,16 @@ static int bch2_rename2(struct mnt_idmap *idmap,
 	struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 	struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 	struct bch_inode_unpacked dst_dir_u, src_dir_u;
-	struct bch_inode_unpacked src_inode_u, dst_inode_u;
+	struct bch_inode_unpacked src_inode_u, dst_inode_u, *whiteout_inode_u;
 	struct btree_trans *trans;
 	enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 		? BCH_RENAME_EXCHANGE
 		: dst_dentry->d_inode
 		? BCH_RENAME_OVERWRITE : BCH_RENAME;
+	bool whiteout = !!(flags & RENAME_WHITEOUT);
 	int ret;
 
-	if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
+	if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE|RENAME_WHITEOUT))
 		return -EINVAL;
 
 	if (mode == BCH_RENAME_OVERWRITE) {
@@ -785,18 +786,48 @@ static int bch2_rename2(struct mnt_idmap *idmap,
 		if (ret)
 			goto err;
 	}
+retry:
+	bch2_trans_begin(trans);
 
-	ret = commit_do(trans, NULL, NULL, 0,
-			bch2_rename_trans(trans,
-					  inode_inum(src_dir), &src_dir_u,
-					  inode_inum(dst_dir), &dst_dir_u,
-					  &src_inode_u,
-					  &dst_inode_u,
-					  &src_dentry->d_name,
-					  &dst_dentry->d_name,
-					  mode));
+	ret = bch2_rename_trans(trans,
+				inode_inum(src_dir), &src_dir_u,
+				inode_inum(dst_dir), &dst_dir_u,
+				&src_inode_u,
+				&dst_inode_u,
+				&src_dentry->d_name,
+				&dst_dentry->d_name,
+				mode);
 	if (unlikely(ret))
+		goto err_tx_restart;
+
+	if (whiteout) {
+		whiteout_inode_u = bch2_trans_kmalloc_nomemzero(trans, sizeof(*whiteout_inode_u));
+		ret = PTR_ERR_OR_ZERO(whiteout_inode_u);
+		if (unlikely(ret))
+			goto err_tx_restart;
+		bch2_inode_init_early(c, whiteout_inode_u);
+
+		ret = bch2_create_trans(trans,
+					inode_inum(src_dir), &src_dir_u,
+					whiteout_inode_u,
+					&src_dentry->d_name,
+					from_kuid(i_user_ns(&src_dir->v), current_fsuid()),
+					from_kgid(i_user_ns(&src_dir->v), current_fsgid()),
+					S_IFCHR|WHITEOUT_MODE, 0,
+					NULL, NULL, (subvol_inum) { 0 }, 0) ?:
+		      bch2_quota_acct(c, bch_qid(whiteout_inode_u), Q_INO, 1,
+				      KEY_TYPE_QUOTA_PREALLOC);
+		if (unlikely(ret))
+			goto err_tx_restart;
+	}
+
+	ret = bch2_trans_commit(trans, NULL, NULL, 0);
+	if (unlikely(ret)) {
+err_tx_restart:
+		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+			goto retry;
 		goto err;
+	}
 
 	BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 	BUG_ON(dst_inode &&
diff --git a/fs/bcachefs/str_hash.h b/fs/bcachefs/str_hash.h
index c8c266cb5797..215eed4cce6d 100644
--- a/fs/bcachefs/str_hash.h
+++ b/fs/bcachefs/str_hash.h
@@ -270,7 +270,7 @@ int bch2_hash_set_in_snapshot(struct btree_trans *trans,
 				desc.hash_bkey(info, bkey_i_to_s_c(insert)),
 				snapshot),
 			   POS(insert->k.p.inode, U64_MAX),
-			   BTREE_ITER_slots|BTREE_ITER_intent, k, ret) {
+			   BTREE_ITER_slots|BTREE_ITER_intent|flags, k, ret) {
 		if (is_visible_key(desc, inum, k)) {
 			if (!desc.cmp_bkey(k, bkey_i_to_s_c(insert)))
 				goto found;