Skip to contentSkip to conversation

Removing a subvolume by an ordinary user

PATCH In review Goffredo Baroncelli 2 participants 4 messages Oct 11, 2010 – Oct 18, 2010 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 Removing a subvolume by an ordinary user

Hi all,

enclosed you can find a patch which permits to remove a volume via the rmdir(2) syscall by an ordinary user.
The rules for a subvolume removal are the same ones of a directory:
- the user must have the write permission on the parent directory
- the subvolume must be empty

The mains differences between calling rmdir(2) on a subvolume and calling the BTRFS_IOC_SNAP_DESTROY ioctl are:

- rmdir(2) requires the subvolume to be empty (the user has to empty the subvolume before removing it, like the rm -rf command does)
- rmdir(2) is a synchronous operation (instead BTRFS_IOC_SNAP_DESTROY works in background)

The previous statements have the following (nice) consequences:
- the CAP_ADMIN capability is not required anymore to remove a subvolume. BTRFS_IOC_SNAP_DESTROY requires CAP_ADMIN because the subvolume removal is performed in background so it would not be possible to return an error if the user has not the privilege of removing a file.
- when the rmdir(2) syscall returns the space is really freed, and there is no necessity to create another API to wait the ends of the BTRFS_IOC_SNAP_DESTROY ioctl.

The only advantage of the BTRFS_IOC_SNAP_DESTROY ioctl is its greater speed in removing a not empty subvolume.

I simplify the code respect my previous post, removing a un-needed call to the function may_destroy_subvol().

You can pull the code from the branch named "rmdir-subvolume" of the following repository:

http://cassiopea.homelinux.net/git/btrfs-unstable.git

As usual, comments are welcome

Regards
G.Baroncelli

fs/btrfs/inode.c +82 -2
index f08427c..0df2fb0 100644
@@ -2944,6 +2944,84 @@ int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+static noinline int btrfs_snap_destroy(struct inode *dir,
+				       struct dentry *dentry)
+{
+
+	struct inode *inode;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct btrfs_root *dest = NULL;
+	struct btrfs_trans_handle *trans;
+	int ret;
+	int err = 0;
+
+	if (IS_ERR(dentry)) {
+		err = PTR_ERR(dentry);
+		goto out;
+	}
+
+	if (!dentry->d_inode) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	inode = dentry->d_inode;
+	if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	dest = BTRFS_I(inode)->root;
+
+	down_write(&root->fs_info->subvol_sem);
+
+	/* remove this check because the directory is empty.
+	 * err = may_destroy_subvol(dest);
+	 * if (err)
+	 *	goto out_up_write;
+	 */
+
+	trans = btrfs_start_transaction(root, 0);
+	if (IS_ERR(trans)) {
+		err = PTR_ERR(trans);
+		goto out_up_write;
+	}
+	trans->block_rsv = &root->fs_info->global_block_rsv;
+
+	ret = btrfs_unlink_subvol(trans, root, dir,
+				dest->root_key.objectid,
+				dentry->d_name.name,
+				dentry->d_name.len);
+	BUG_ON(ret);
+
+	btrfs_record_root_in_trans(trans, dest);
+
+	memset(&dest->root_item.drop_progress, 0,
+		sizeof(dest->root_item.drop_progress));
+	dest->root_item.drop_level = 0;
+	btrfs_set_root_refs(&dest->root_item, 0);
+
+	if (!xchg(&dest->orphan_item_inserted, 1)) {
+		ret = btrfs_insert_orphan_item(trans,
+					root->fs_info->tree_root,
+					dest->root_key.objectid);
+		BUG_ON(ret);
+	}
+
+	ret = btrfs_commit_transaction(trans, root);
+	BUG_ON(ret);
+	inode->i_flags |= S_DEAD;
+out_up_write:
+	up_write(&root->fs_info->subvol_sem);
+	if (!err) {
+		shrink_dcache_sb(root->fs_info->sb);
+		btrfs_invalidate_inodes(dest);
+		/*d_delete(dentry);*/
+	}
+out:
+	return err;
+}
+
 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 {
 	struct inode *inode = dentry->d_inode;
@@ -2952,10 +3030,12 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 	struct btrfs_trans_handle *trans;
 	unsigned long nr = 0;
 
-	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
-	    inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
+	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 		return -ENOTEMPTY;
 
+	if (inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
+		return btrfs_snap_destroy(dir, dentry);
+
 	trans = __unlink_start_trans(dir, dentry);
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
Signature · 2 lines
gpg key@ keyserver.linux.it: Goffredo Baroncelli (ghigo) <kreijackATinwind.it>
Key fingerprint = 4769 7E51 5293 D36C 814E  C054 BF04 F161 3DC5 0512
#1 lore raw
Quoted 88 lines · Goffredo Baroncelli

Hi all,

enclosed you can find a patch which permits to remove a volume via the rmdir(2) syscall by an ordinary user.
The rules for a subvolume removal are the same ones of a directory:
- the user must have the write permission on the parent directory
- the subvolume must be empty

The mains differences between calling rmdir(2) on a subvolume and calling the BTRFS_IOC_SNAP_DESTROY ioctl are:

- rmdir(2) requires the subvolume to be empty (the user has to empty the subvolume before removing it, like the rm -rf command does)
- rmdir(2) is a synchronous operation (instead BTRFS_IOC_SNAP_DESTROY works in background)

The previous statements have the following (nice) consequences:
- the CAP_ADMIN capability is not required anymore to remove a subvolume. BTRFS_IOC_SNAP_DESTROY requires CAP_ADMIN because the subvolume removal is performed in background so it would not be possible to return an error if the user has not the privilege of removing a file.
- when the rmdir(2) syscall returns the space is really freed, and there is no necessity to create another API to wait the ends of the BTRFS_IOC_SNAP_DESTROY ioctl.

The only advantage of the BTRFS_IOC_SNAP_DESTROY ioctl is its greater speed in removing a not empty subvolume.

I simplify the code respect my previous post, removing a un-needed call to the function may_destroy_subvol().

You can pull the code from the branch named "rmdir-subvolume" of the following repository:

http://cassiopea.homelinux.net/git/btrfs-unstable.git

As usual, comments are welcome

Regards
G.Baroncelli

fs/btrfs/inode.c +39 -0
index f08427c..0df2fb0 100644
@@ -2944,6 +2944,84 @@ int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+static noinline int btrfs_snap_destroy(struct inode *dir,
+				       struct dentry *dentry)
+{
+
+	struct inode *inode;
+	struct btrfs_root *root = BTRFS_I(dir)->root;
+	struct btrfs_root *dest = NULL;
+	struct btrfs_trans_handle *trans;
+	int ret;
+	int err = 0;
+
+	if (IS_ERR(dentry)) {
+		err = PTR_ERR(dentry);
+		goto out;
+	}
+
+	if (!dentry->d_inode) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	inode = dentry->d_inode;
+	if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	dest = BTRFS_I(inode)->root;
+
+	down_write(&root->fs_info->subvol_sem);
+
+	/* remove this check because the directory is empty.
+	 * err = may_destroy_subvol(dest);
+	 * if (err)
+	 *	goto out_up_write;
+	 */
+
+	trans = btrfs_start_transaction(root, 0);
+	if (IS_ERR(trans)) {

fs/btrfs/inode.c int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, · line 2995

Quoted 10 lines
fs/btrfs/inode.c +10 -0
+		err = PTR_ERR(trans);
+		goto out_up_write;
+	}
+	trans->block_rsv = &root->fs_info->global_block_rsv;
+
+	ret = btrfs_unlink_subvol(trans, root, dir,
+				dest->root_key.objectid,
+				dentry->d_name.name,
+				dentry->d_name.len);
+	BUG_ON(ret);

Is it really a good idea to add even more dead end BUG_ON() calls instead of handling errors when they happen?

Quoted 49 lines
fs/btrfs/inode.c +33 -2
+
+	btrfs_record_root_in_trans(trans, dest);
+
+	memset(&dest->root_item.drop_progress, 0,
+		sizeof(dest->root_item.drop_progress));
+	dest->root_item.drop_level = 0;
+	btrfs_set_root_refs(&dest->root_item, 0);
+
+	if (!xchg(&dest->orphan_item_inserted, 1)) {
+		ret = btrfs_insert_orphan_item(trans,
+					root->fs_info->tree_root,
+					dest->root_key.objectid);
+		BUG_ON(ret);
+	}
+
+	ret = btrfs_commit_transaction(trans, root);
+	BUG_ON(ret);
+	inode->i_flags |= S_DEAD;
+out_up_write:
+	up_write(&root->fs_info->subvol_sem);
+	if (!err) {
+		shrink_dcache_sb(root->fs_info->sb);
+		btrfs_invalidate_inodes(dest);
+		/*d_delete(dentry);*/
+	}
+out:
+	return err;
+}
+
 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 {
 	struct inode *inode = dentry->d_inode;
@@ -2952,10 +3030,12 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 	struct btrfs_trans_handle *trans;
 	unsigned long nr = 0;
 
-	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
-	    inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
+	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 		return -ENOTEMPTY;
 
+	if (inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
+		return btrfs_snap_destroy(dir, dentry);
+
 	trans = __unlink_start_trans(dir, dentry);
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
#2 lore raw
Quoted 1 line · Ian Kent

On Mon, 2010-10-11 at 20:08 +0200, Goffredo Baroncelli wrote:

[...]

Quoted 9 lines

fs/btrfs/inode.c int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, · line 2995

Quoted 5 lines
fs/btrfs/inode.c +5 -0
+	ret = btrfs_unlink_subvol(trans, root, dir,
+				dest->root_key.objectid,
+				dentry->d_name.name,
+				dentry->d_name.len);
+	BUG_ON(ret);

Is it really a good idea to add even more dead end BUG_ON() calls instead of handling errors when they happen?

I agree with your comment, I copied these lines from the ioctl DESTROY_SUBVOL. Unfortunately I don't have the know-how to handle an error at this level. So I can only leave the code as the original one.

[...]

Signature · 2 lines
gpg key@ keyserver.linux.it: Goffredo Baroncelli (ghigo) <[email protected]>
Key fingerprint = 4769 7E51 5293 D36C 814E  C054 BF04 F161 3DC5 0512
#3 lore raw
Quoted 13 lines · Goffredo Baroncelli
Quoted 1 line · Ian Kent

On Mon, 2010-10-11 at 20:08 +0200, Goffredo Baroncelli wrote:

[...]

Quoted 9 lines

fs/btrfs/inode.c int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, · line 2995

Quoted 5 lines
fs/btrfs/inode.c +5 -0
+	ret = btrfs_unlink_subvol(trans, root, dir,
+				dest->root_key.objectid,
+				dentry->d_name.name,
+				dentry->d_name.len);
+	BUG_ON(ret);

Is it really a good idea to add even more dead end BUG_ON() calls instead of handling errors when they happen?

Quoted 3 lines

I agree with your comment, I copied these lines from the ioctl DESTROY_SUBVOL. Unfortunately I don't have the know-how to handle an error at this level. So I can only leave the code as the original one.

Right, sadly someone is going to have to go through and reconcile all these things at some point.

I picked a function and started analyzing what would be needed (including dependent functions) to handle the error return instead of BUG_ON()ing and it's not trivial at all.

Ian