From: Goffredo Baroncelli Subject: [PATCH V3] Removing a subvolume with a simple "rm -rf" Date: Sun, 17 Oct 2010 18:09:50 +0200 Message-ID: <201010171809.58152.kreijack@libero.it> Reply-To: kreijack@libero.it Mime-Version: 1.0 Content-Type: multipart/signed; boundary="nextPart1417182.XgMMnMG3a8"; protocol="application/pgp-signature"; micalg=pgp-sha1 Cc: Chris Mason To: linux-btrfs@vger.kernel.org Return-path: List-ID: --nextPart1417182.XgMMnMG3a8 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable Hi all, enclosed you can find a new version of the patch which permits to remove a= =20 volume via the rmdir(2) syscall by a non-root user. The goal of this patch= =20 is to permits to remove a subvolume with a simple "rm -rf" command. The rules for a subvolume removal are the same ones of a directory: =2D the user must have the write permission on the parent directory =2D the subvolume must be empty The mains differences between calling rmdir(2) on a subvolume and calling t= he=20 BTRFS_IOC_SNAP_DESTROY ioctl are: =2D rmdir(2) requires the subvolume to be empty (the user has to empty the= =20 subvolume before removing it, like the rm -rf command does) =2D rmdir(2) is a synchronous operation (instead BTRFS_IOC_SNAP_DESTROY wor= ks=20 in background) The previous statements have the following (nice) consequences: =2D the CAP_ADMIN capability is not mandatory anymore to remove a subvolume= =2E=20 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=20 the user has not the privilege of removing a file. This explain why this io= ctl may be executed only by root. =2D When the rmdir(2) syscall returns, the space is really freed. 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(). Moreover I added a call to the d_invalidate() function, which was missed in my previous version. You can pull the code from the branch named "rmdir-subvolume" of the follow= ing repository: http://cassiopea.homelinux.net/git/btrfs-unstable.git =20 As usual, comments are welcome Regards G.Baroncelli diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index f08427c..f732ddf 100644 =2D-- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -2944,6 +2944,88 @@ int btrfs_unlink_subvol(struct btrfs_trans_handle *t= rans, return 0; } =20 +static noinline int btrfs_snap_destroy(struct inode *dir, + struct dentry *dentry) +{ + + struct inode *inode; + struct btrfs_root *root =3D BTRFS_I(dir)->root; + struct btrfs_root *dest =3D NULL; + struct btrfs_trans_handle *trans; + int ret; + int err =3D 0; + + if (IS_ERR(dentry)) { + err =3D PTR_ERR(dentry); + goto out; + } + + if (!dentry->d_inode) { + err =3D -ENOENT; + goto out; + } + + inode =3D dentry->d_inode; + if (inode->i_ino !=3D BTRFS_FIRST_FREE_OBJECTID) { + err =3D -EINVAL; + goto out; + } + + dest =3D BTRFS_I(inode)->root; +=09 + err =3D d_invalidate(dentry); + if (err) + goto out; + + down_write(&root->fs_info->subvol_sem); + + /* remove this check because the directory is empty. + * err =3D may_destroy_subvol(dest); + * if (err) + * goto out_up_write; + */ + + trans =3D btrfs_start_transaction(root, 0); + if (IS_ERR(trans)) { + err =3D PTR_ERR(trans); + goto out_up_write; + } + trans->block_rsv =3D &root->fs_info->global_block_rsv; + + ret =3D 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 =3D 0; + btrfs_set_root_refs(&dest->root_item, 0); + + if (!xchg(&dest->orphan_item_inserted, 1)) { + ret =3D btrfs_insert_orphan_item(trans, + root->fs_info->tree_root, + dest->root_key.objectid); + BUG_ON(ret); + } + + ret =3D btrfs_commit_transaction(trans, root); + BUG_ON(ret); + inode->i_flags |=3D 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 =3D dentry->d_inode; @@ -2952,10 +3034,12 @@ static int btrfs_rmdir(struct inode *dir, struct de= ntry *dentry) struct btrfs_trans_handle *trans; unsigned long nr =3D 0; =20 =2D if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || =2D inode->i_ino =3D=3D BTRFS_FIRST_FREE_OBJECTID) + if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) return -ENOTEMPTY; =20 + if (inode->i_ino =3D=3D BTRFS_FIRST_FREE_OBJECTID) + return btrfs_snap_destroy(dir, dentry); + trans =3D __unlink_start_trans(dir, dentry); if (IS_ERR(trans)) return PTR_ERR(trans); =2D-=20 gpg key@ keyserver.linux.it: Goffredo Baroncelli (ghigo) Key fingerprint =3D 4769 7E51 5293 D36C 814E C054 BF04 F161 3DC5 0512 --nextPart1417182.XgMMnMG3a8 Content-Type: application/pgp-signature; name=signature.asc Content-Description: This is a digitally signed message part. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.10 (GNU/Linux) iD8DBQBMux/OvwTxYT3FBRIRAqYaAJ40yZ6xq8Y7LcQRTsHLSjY+F/WCsQCfaQ4Y FjwodBVsjJSdTYVH/ioXCw4= =CrQH -----END PGP SIGNATURE----- --nextPart1417182.XgMMnMG3a8--