Home | History | Annotate | Line # | Download | only in unionfs
      1 /*-
      2  * Copyright (c) 1994, 1995 The Regents of the University of California.
      3  * Copyright (c) 1994, 1995 Jan-Simon Pendry.
      4  * Copyright (c) 2005, 2006 Masanori Ozawa <ozawa (at) ongs.co.jp>, ONGS Inc.
      5  * Copyright (c) 2006 Daichi Goto <daichi (at) freebsd.org>
      6  * All rights reserved.
      7  *
      8  * This code is derived from software donated to Berkeley by
      9  * Jan-Simon Pendry.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  * 4. Neither the name of the University nor the names of its contributors
     20  *    may be used to endorse or promote products derived from this software
     21  *    without specific prior written permission.
     22  *
     23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     33  * SUCH DAMAGE.
     34  *
     35  *	@(#)union_vfsops.c	8.20 (Berkeley) 5/20/95
     36  * $FreeBSD: src/sys/fs/unionfs/union_vfsops.c,v 1.89 2008/01/13 14:44:06 attilio Exp $
     37  */
     38 
     39 #include <sys/param.h>
     40 #include <sys/systm.h>
     41 #include <sys/kernel.h>
     42 #include <sys/lock.h>
     43 #include <sys/mount.h>
     44 #include <sys/namei.h>
     45 #include <sys/proc.h>
     46 #include <sys/vnode.h>
     47 #include <sys/stat.h>
     48 #include <sys/sysctl.h>
     49 #include <sys/module.h>
     50 
     51 #include <fs/unionfs/unionfs.h>
     52 
     53 MODULE(MODULE_CLASS_VFS, unionfs, "layerfs");
     54 
     55 MALLOC_DEFINE(M_UNIONFSMNT, "UNIONFS mount", "UNIONFS mount structure");
     56 
     57 struct vfsops unionfs_vfsops;
     58 
     59 VFS_PROTOS(unionfs);
     60 
     61 static struct sysctllog *unionfs_sysctl_log;
     62 
     63 /*
     64  * Mount unionfs layer.
     65  */
     66 int
     67 unionfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
     68 {
     69 	int		error;
     70 	struct vnode   *lowerrootvp;
     71 	struct vnode   *upperrootvp;
     72 	struct unionfs_mount *ump;
     73 	int		below;
     74 	uid_t		uid;
     75 	gid_t		gid;
     76 	u_short		udir;
     77 	u_short		ufile;
     78 	unionfs_copymode copymode;
     79 	unionfs_whitemode whitemode;
     80 	struct pathbuf *pb;
     81 	struct componentname fakecn;
     82 	struct nameidata nd, *ndp;
     83 	struct vattr	va;
     84 	struct union_args *args = data;
     85 	kauth_cred_t	cred;
     86 	size_t		size;
     87 	size_t		len;
     88 	const char	*cp;
     89 	char		*xp;
     90 
     91 	if (args == NULL)
     92 		return EINVAL;
     93 	if (*data_len < sizeof *args)
     94 		return EINVAL;
     95 
     96 	UNIONFSDEBUG("%s(mp = %p)\n", __func__, mp);
     97 
     98 	error = 0;
     99 	below = 0;
    100 	uid = 0;
    101 	gid = 0;
    102 	udir = 0;
    103 	ufile = 0;
    104 	copymode = UNIONFS_TRANSPARENT;	/* default */
    105 	whitemode = UNIONFS_WHITE_ALWAYS;
    106 	ndp = &nd;
    107 
    108 	if (mp->mnt_flag & MNT_ROOTFS) {
    109 		printf("%s: cannot union mount root filesystem\n", __func__);
    110 		return EOPNOTSUPP;
    111 	}
    112 
    113 	if (mp->mnt_flag & MNT_GETARGS) {
    114 		ump = MOUNTTOUNIONFSMOUNT(mp);
    115 		if (ump == NULL)
    116 			return EIO;
    117 		args->target = NULL;
    118 		args->mntflags = ump->um_op;
    119 		*data_len = sizeof *args;
    120 		return 0;
    121 	}
    122 
    123 	/*
    124 	 * Update is a no operation.
    125 	 */
    126 	if (mp->mnt_flag & MNT_UPDATE) {
    127 		printf("%s: cannot update union mount\n", __func__);
    128 		return EOPNOTSUPP;
    129 	}
    130 
    131 	cred = kauth_cred_get();
    132 	vn_lock(mp->mnt_vnodecovered, LK_EXCLUSIVE | LK_RETRY);
    133 	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, cred);
    134 	if (!error) {
    135 		if (udir == 0)
    136 			udir = va.va_mode;
    137 		if (ufile == 0)
    138 			ufile = va.va_mode;
    139 		uid = va.va_uid;
    140 		gid = va.va_gid;
    141 	}
    142 	VOP_UNLOCK(mp->mnt_vnodecovered);
    143 	if (error)
    144 		return error;
    145 
    146 	switch (args->mntflags & UNMNT_OPMASK) {
    147 	case UNMNT_ABOVE:
    148 		below = 0;
    149 		break;
    150 
    151 	case UNMNT_BELOW:
    152 		below = 1;
    153 		break;
    154 
    155 	case UNMNT_REPLACE:
    156 	default:
    157 		return EINVAL;
    158 	}
    159 
    160 	/* If copymode is UNIONFS_TRADITIONAL, uid/gid is mounted user. */
    161 	if (copymode == UNIONFS_TRADITIONAL) {
    162 		uid = kauth_cred_getuid(cred);
    163 		gid = kauth_cred_getgid(cred);
    164 	}
    165 
    166 	UNIONFSDEBUG("%s: uid=%d, gid=%d\n", __func__, uid, gid);
    167 	UNIONFSDEBUG("%s: udir=%#03o, ufile=%#03o\n", __func__, udir, ufile);
    168 	UNIONFSDEBUG("%s: copymode=%d\n", __func__, copymode);
    169 
    170 	/*
    171 	 * Find upper node
    172 	 */
    173 	error = pathbuf_copyin(args->target, &pb);
    174 	if (error)
    175 		return error;
    176 	NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, pb);
    177 	if ((error = namei(ndp))) {
    178 		pathbuf_destroy(pb);
    179 		return error;
    180 	}
    181 
    182 	/* get root vnodes */
    183 	lowerrootvp = mp->mnt_vnodecovered;
    184 	upperrootvp = ndp->ni_vp;
    185 
    186 	vrele(ndp->ni_dvp);
    187 	ndp->ni_dvp = NULLVP;
    188 	pathbuf_destroy(pb);
    189 
    190 	/* create unionfs_mount */
    191 	ump = kmem_zalloc(sizeof(*ump), KM_SLEEP);
    192 
    193 	/*
    194 	 * Save reference
    195 	 */
    196 	if (below) {
    197 		VOP_UNLOCK(upperrootvp);
    198 		vn_lock(lowerrootvp, LK_EXCLUSIVE | LK_RETRY);
    199 		ump->um_lowervp = upperrootvp;
    200 		ump->um_uppervp = lowerrootvp;
    201 	} else {
    202 		ump->um_lowervp = lowerrootvp;
    203 		ump->um_uppervp = upperrootvp;
    204 	}
    205 	ump->um_rootvp = NULLVP;
    206 	ump->um_uid = uid;
    207 	ump->um_gid = gid;
    208 	ump->um_udir = udir;
    209 	ump->um_ufile = ufile;
    210 	ump->um_copymode = copymode;
    211 	ump->um_whitemode = whitemode;
    212 
    213 	if ((lowerrootvp->v_mount->mnt_iflag & IMNT_MPSAFE) &&
    214 	    (upperrootvp->v_mount->mnt_flag & IMNT_MPSAFE))
    215 		mp->mnt_iflag |= IMNT_MPSAFE;
    216 	mp->mnt_data = ump;
    217 
    218 	/*
    219 	 * Copy upper layer's RDONLY flag.
    220 	 */
    221 	mp->mnt_flag |= ump->um_uppervp->v_mount->mnt_flag & MNT_RDONLY;
    222 
    223 	/*
    224 	 * Check whiteout
    225 	 */
    226 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
    227 		memset(&fakecn, 0, sizeof(fakecn));
    228 		fakecn.cn_nameiop = LOOKUP;
    229 		error = VOP_WHITEOUT(ump->um_uppervp, &fakecn, LOOKUP);
    230 		if (error) {
    231 			if (below) {
    232 				VOP_UNLOCK(ump->um_uppervp);
    233 				vrele(upperrootvp);
    234 			} else
    235 				vput(ump->um_uppervp);
    236 			goto out;
    237 		}
    238 	}
    239 
    240 	/*
    241 	 * Unlock the node
    242 	 */
    243 	VOP_UNLOCK(ump->um_uppervp);
    244 
    245 	ump->um_op = args->mntflags & UNMNT_OPMASK;
    246 
    247 	/*
    248 	 * Get the unionfs root vnode.
    249 	 */
    250 	error = unionfs_nodeget(mp, ump->um_uppervp, ump->um_lowervp,
    251 	    NULLVP, &(ump->um_rootvp), NULL);
    252 	vrele(upperrootvp);
    253 	if (error) {
    254 		goto out;
    255 	}
    256 
    257 	/*
    258 	 * Check mnt_flag
    259 	 */
    260 	if ((ump->um_lowervp->v_mount->mnt_flag & MNT_LOCAL) &&
    261 	    (ump->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
    262 		mp->mnt_flag |= MNT_LOCAL;
    263 
    264 	/*
    265 	 * Get new fsid
    266 	 */
    267 	vfs_getnewfsid(mp);
    268 
    269 	error = set_statvfs_info(path, UIO_USERSPACE, NULL, UIO_USERSPACE,
    270 	    mp->mnt_op->vfs_name, mp, curlwp);
    271 	if (error) {
    272 		unionfs_noderem(ump->um_rootvp);
    273 		goto out;
    274 	}
    275 
    276 	switch (ump->um_op) {
    277 	case UNMNT_ABOVE:
    278 		cp = "<above>:";
    279 		break;
    280 	case UNMNT_BELOW:
    281 		cp = "<below>:";
    282 		break;
    283 	default:
    284 #ifdef DIAGNOSTIC
    285 		panic("%s: bad um_op", __func__);
    286 #endif
    287 		break;
    288 	}
    289 	len = strlen(cp);
    290 	memcpy(mp->mnt_stat.f_mntfromname, cp, len);
    291 	xp = mp->mnt_stat.f_mntfromname + len;
    292 	len = MNAMELEN - len;
    293 	(void) copyinstr(args->target, xp, len - 1, &size);
    294 	memset(xp + size, 0, len - size);
    295 
    296 	UNIONFSDEBUG("%s: from %s, on %s\n", __func__,
    297 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
    298 
    299 	return 0;
    300 out:
    301 	kmem_free(ump, sizeof(*ump));
    302 	mp->mnt_data = NULL;
    303 	return error;
    304 }
    305 
    306 /*
    307  * Free reference to unionfs layer
    308  */
    309 int
    310 unionfs_unmount(struct mount *mp, int mntflags)
    311 {
    312 	struct unionfs_mount *ump;
    313 	int		error;
    314 	int		freeing;
    315 	int		flags;
    316 
    317 	UNIONFSDEBUG("%s: mp = %p\n", __func__, mp);
    318 
    319 	ump = MOUNTTOUNIONFSMOUNT(mp);
    320 	flags = 0;
    321 
    322 	if (mntflags & MNT_FORCE)
    323 		flags |= FORCECLOSE;
    324 
    325 	/* vflush (no need to call vrele) */
    326 	for (freeing = 0; (error = vflush(mp, NULL, flags)) != 0;) {
    327 		struct vnode_iterator *marker;
    328 		struct vnode *vp;
    329 		int n;
    330 
    331 		/* count #vnodes held on mount list */
    332 		vfs_vnode_iterator_init(mp, &marker);
    333 		n = 0;
    334 		while ((vp = vfs_vnode_iterator_next(marker, NULL, NULL)))
    335 			n++;
    336 		vfs_vnode_iterator_destroy(marker);
    337 
    338 		/* if this is unchanged then stop */
    339 		if (n == freeing)
    340 			break;
    341 
    342 		/* otherwise try once more time */
    343 		freeing = n;
    344 	}
    345 
    346 	if (error)
    347 		return error;
    348 
    349 	kmem_free(ump, sizeof(*ump));
    350 	mp->mnt_data = NULL;
    351 
    352 	return (0);
    353 }
    354 
    355 int
    356 unionfs_root(struct mount *mp, int lktype, struct vnode **vpp)
    357 {
    358 	struct unionfs_mount *ump;
    359 	struct vnode   *vp;
    360 
    361 	ump = MOUNTTOUNIONFSMOUNT(mp);
    362 	vp = ump->um_rootvp;
    363 
    364 	UNIONFSDEBUG("%s: rootvp=%p locked=%#x\n", __func__,
    365 	    vp, VOP_ISLOCKED(vp));
    366 
    367 	vref(vp);
    368 	vn_lock(vp, lktype);
    369 
    370 	*vpp = vp;
    371 
    372 	return 0;
    373 }
    374 
    375 int
    376 unionfs_quotactl(struct mount *mp, struct quotactl_args *args)
    377 {
    378 	struct unionfs_mount *ump;
    379 
    380 	ump = MOUNTTOUNIONFSMOUNT(mp);
    381 
    382 	/*
    383 	 * Writing is always performed to upper vnode.
    384 	 */
    385 	return VFS_QUOTACTL(ump->um_uppervp->v_mount, args);
    386 }
    387 
    388 int
    389 unionfs_statvfs(struct mount *mp, struct statvfs *sbp)
    390 {
    391 	struct unionfs_mount *ump;
    392 	int		error;
    393 	uint64_t	lbsize;
    394 	struct statvfs *sbuf = kmem_zalloc(sizeof(*sbuf), KM_SLEEP);
    395 
    396 	ump = MOUNTTOUNIONFSMOUNT(mp);
    397 
    398 	UNIONFSDEBUG("%s(mp = %p, lvp = %p, uvp = %p)\n",
    399 	    __func__, mp, ump->um_lowervp, ump->um_uppervp);
    400 
    401 	error = VFS_STATVFS(ump->um_lowervp->v_mount, sbuf);
    402 	if (error)
    403 		goto done;
    404 
    405 	/* now copy across the "interesting" information and fake the rest */
    406 	sbp->f_blocks = sbuf->f_blocks;
    407 	sbp->f_files = sbuf->f_files;
    408 
    409 	lbsize = sbuf->f_bsize;
    410 
    411 	error = VFS_STATVFS(ump->um_uppervp->v_mount, sbuf);
    412 	if (error)
    413 		goto done;
    414 
    415 	/*
    416 	 * The FS type etc is copy from upper vfs.
    417 	 * (write able vfs have priority)
    418 	 */
    419 	sbp->f_flag = sbuf->f_flag;
    420 	sbp->f_bsize = sbuf->f_bsize;
    421 	sbp->f_iosize = sbuf->f_iosize;
    422 
    423 	if (sbuf->f_bsize != lbsize)
    424 		sbp->f_blocks = ((off_t)sbp->f_blocks * lbsize) / sbuf->f_bsize;
    425 
    426 	sbp->f_blocks += sbuf->f_blocks;
    427 	sbp->f_bfree = sbuf->f_bfree;
    428 	sbp->f_bavail = sbuf->f_bavail;
    429 	sbp->f_files += sbuf->f_files;
    430 	sbp->f_ffree = sbuf->f_ffree;
    431 
    432  done:
    433 	kmem_free(sbuf, sizeof(*sbuf));
    434 	return error;
    435 }
    436 
    437 int
    438 unionfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
    439 {
    440 	/* nothing to do */
    441 	return 0;
    442 }
    443 
    444 int
    445 unionfs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
    446     int namespace, const char *attrname)
    447 {
    448 	struct unionfs_mount *ump;
    449 	struct unionfs_node *unp;
    450 
    451 	ump = MOUNTTOUNIONFSMOUNT(mp);
    452 	unp = VTOUNIONFS(filename_vp);
    453 
    454 	if (unp->un_uppervp != NULLVP) {
    455 		return VFS_EXTATTRCTL(ump->um_uppervp->v_mount, cmd,
    456 		    unp->un_uppervp, namespace, attrname);
    457 	} else {
    458 		return VFS_EXTATTRCTL(ump->um_lowervp->v_mount, cmd,
    459 		    unp->un_lowervp, namespace, attrname);
    460 	}
    461 }
    462 
    463 /*
    464  * Initialize
    465  */
    466 void
    467 unionfs_init(void)
    468 {
    469 	UNIONFSDEBUG("%s\n", __func__);	/* printed during system boot */
    470 }
    471 
    472 static int
    473 unionfs_renamelock_enter(struct mount *mp)
    474 {
    475 	struct unionfs_mount *um = MOUNTTOUNIONFSMOUNT(mp);
    476 
    477 	/* Lock just the upper fs, where the action happens. */
    478 	return VFS_RENAMELOCK_ENTER(um->um_uppervp->v_mount);
    479 }
    480 
    481 static void
    482 unionfs_renamelock_exit(struct mount *mp)
    483 {
    484 	struct unionfs_mount *um = MOUNTTOUNIONFSMOUNT(mp);
    485 
    486 	VFS_RENAMELOCK_EXIT(um->um_uppervp->v_mount);
    487 }
    488 
    489 int
    490 unionfs_start(struct mount *mp, int flags)
    491 {
    492 
    493 	return 0;
    494 }
    495 
    496 void
    497 unionfs_done(void)
    498 {
    499 
    500 	/* Make sure to unset the readdir hook. */
    501 	vn_union_readdir_hook = NULL;
    502 }
    503 
    504 extern const struct vnodeopv_desc unionfs_vnodeop_opv_desc;
    505 
    506 const struct vnodeopv_desc * const unionfs_vnodeopv_descs[] = {
    507 	&unionfs_vnodeop_opv_desc,
    508 	NULL,
    509 };
    510 
    511 struct vfsops unionfs_vfsops = {
    512 	.vfs_name = MOUNT_UNION,
    513 	.vfs_min_mount_data = sizeof (struct unionfs_args),
    514 	.vfs_mount = unionfs_mount,
    515 	.vfs_start = unionfs_start,
    516 	.vfs_unmount = unionfs_unmount,
    517 	.vfs_root = unionfs_root,
    518 	.vfs_quotactl = (void *)eopnotsupp,
    519 	.vfs_statvfs = unionfs_statvfs,
    520 	.vfs_sync = unionfs_sync,
    521 	.vfs_vget = (void *)eopnotsupp,
    522 	.vfs_fhtovp = (void *)eopnotsupp,
    523 	.vfs_vptofh = (void *)eopnotsupp,
    524 	.vfs_init = unionfs_init,
    525 	.vfs_done = unionfs_done,
    526 	.vfs_snapshot = (void *)eopnotsupp,
    527 	.vfs_extattrctl = vfs_stdextattrctl,
    528 	.vfs_suspendctl = (void *)eopnotsupp,
    529 	.vfs_renamelock_enter = unionfs_renamelock_enter,
    530 	.vfs_renamelock_exit = unionfs_renamelock_exit,
    531 	.vfs_fsync = (void *)eopnotsupp,
    532 	.vfs_opv_descs = unionfs_vnodeopv_descs
    533 };
    534 
    535 static int
    536 unionfs_modcmd(modcmd_t cmd, void *arg)
    537 {
    538 	int error;
    539 
    540 	switch (cmd) {
    541 	case MODULE_CMD_INIT:
    542 		error = vfs_attach(&unionfs_vfsops);
    543 		if (error != 0)
    544 			break;
    545 		sysctl_createv(&unionfs_sysctl_log, 0, NULL, NULL,
    546 		    CTLFLAG_PERMANENT,
    547 		    CTLTYPE_NODE, "union",
    548 		    SYSCTL_DESCR("Union file system"),
    549 		    NULL, 0, NULL, 0,
    550 		    CTL_VFS, 15, CTL_EOL);
    551 		/*
    552 		 * XXX the "15" above could be dynamic, thereby eliminating
    553 		 * one more instance of the "number to vfs" mapping problem,
    554 		 * but "15" is the order as taken from sys/mount.h
    555 		 */
    556 		break;
    557 	case MODULE_CMD_FINI:
    558 		error = vfs_detach(&unionfs_vfsops);
    559 		if (error != 0)
    560 			break;
    561 		sysctl_teardown(&unionfs_sysctl_log);
    562 		break;
    563 	default:
    564 		error = ENOTTY;
    565 		break;
    566 	}
    567 
    568 	return error;
    569 }
    570