xref: /titanic_41/usr/src/uts/common/fs/lofs/lofs_vfsops.c (revision c333dd99c762d509c7eb6cce222221958e23b4c8)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/param.h>
29 #include <sys/errno.h>
30 #include <sys/vfs.h>
31 #include <sys/vfs_opreg.h>
32 #include <sys/vnode.h>
33 #include <sys/uio.h>
34 #include <sys/pathname.h>
35 #include <sys/kmem.h>
36 #include <sys/cred.h>
37 #include <sys/statvfs.h>
38 #include <sys/fs/lofs_info.h>
39 #include <sys/fs/lofs_node.h>
40 #include <sys/mount.h>
41 #include <sys/mntent.h>
42 #include <sys/mkdev.h>
43 #include <sys/priv.h>
44 #include <sys/sysmacros.h>
45 #include <sys/systm.h>
46 #include <sys/cmn_err.h>
47 #include <sys/policy.h>
48 #include <sys/tsol/label.h>
49 #include "fs/fs_subr.h"
50 
51 /*
52  * This is the loadable module wrapper.
53  */
54 #include <sys/modctl.h>
55 
56 static mntopts_t lofs_mntopts;
57 
58 static int lofsinit(int, char *);
59 
60 static vfsdef_t vfw = {
61 	VFSDEF_VERSION,
62 	"lofs",
63 	lofsinit,
64 	VSW_HASPROTO|VSW_STATS,
65 	&lofs_mntopts
66 };
67 
68 /*
69  * LOFS mount options table
70  */
71 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
72 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
73 static char *sub_cancel[] = { MNTOPT_LOFS_NOSUB, NULL };
74 static char *nosub_cancel[] = { MNTOPT_LOFS_SUB, NULL };
75 
76 static mntopt_t mntopts[] = {
77 /*
78  *	option name		cancel option	default arg	flags
79  *		private data
80  */
81 	{ MNTOPT_XATTR,		xattr_cancel,	NULL,		0,
82 		(void *)0 },
83 	{ MNTOPT_NOXATTR,	noxattr_cancel,	NULL,		0,
84 		(void *)0 },
85 	{ MNTOPT_LOFS_SUB,	sub_cancel,	NULL,		0,
86 		(void *)0 },
87 	{ MNTOPT_LOFS_NOSUB,	nosub_cancel,	NULL,		0,
88 		(void *)0 },
89 };
90 
91 static mntopts_t lofs_mntopts = {
92 	sizeof (mntopts) / sizeof (mntopt_t),
93 	mntopts
94 };
95 
96 /*
97  * Module linkage information for the kernel.
98  */
99 
100 static struct modlfs modlfs = {
101 	&mod_fsops, "filesystem for lofs", &vfw
102 };
103 
104 static struct modlinkage modlinkage = {
105 	MODREV_1, (void *)&modlfs, NULL
106 };
107 
108 /*
109  * This is the module initialization routine.
110  */
111 
112 int
113 _init(void)
114 {
115 	int status;
116 
117 	lofs_subrinit();
118 	status = mod_install(&modlinkage);
119 	if (status != 0) {
120 		/*
121 		 * Cleanup previously initialized work.
122 		 */
123 		lofs_subrfini();
124 	}
125 
126 	return (status);
127 }
128 
129 /*
130  * Don't allow the lofs module to be unloaded for now.
131  * There is a memory leak if it gets unloaded.
132  */
133 
134 int
135 _fini(void)
136 {
137 	return (EBUSY);
138 }
139 
140 int
141 _info(struct modinfo *modinfop)
142 {
143 	return (mod_info(&modlinkage, modinfop));
144 }
145 
146 
147 static int lofsfstype;
148 vfsops_t *lo_vfsops;
149 
150 /*
151  * lo mount vfsop
152  * Set up mount info record and attach it to vfs struct.
153  */
154 /*ARGSUSED*/
155 static int
156 lo_mount(struct vfs *vfsp,
157 	struct vnode *vp,
158 	struct mounta *uap,
159 	struct cred *cr)
160 {
161 	int error;
162 	struct vnode *srootvp = NULL;	/* the server's root */
163 	struct vnode *realrootvp;
164 	struct loinfo *li;
165 	int nodev;
166 
167 	nodev = vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL);
168 
169 	if ((error = secpolicy_fs_mount(cr, vp, vfsp)) != 0)
170 		return (EPERM);
171 
172 	/*
173 	 * Loopback devices which get "nodevices" added can be done without
174 	 * "nodevices" set because we cannot import devices into a zone
175 	 * with loopback.  Note that we have all zone privileges when
176 	 * this happens; if not, we'd have gotten "nosuid".
177 	 */
178 	if (!nodev && vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL))
179 		vfs_setmntopt(vfsp, MNTOPT_DEVICES, NULL, VFS_NODISPLAY);
180 
181 	mutex_enter(&vp->v_lock);
182 	if (!(uap->flags & MS_OVERLAY) &&
183 	    (vp->v_count != 1 || (vp->v_flag & VROOT))) {
184 		mutex_exit(&vp->v_lock);
185 		return (EBUSY);
186 	}
187 	mutex_exit(&vp->v_lock);
188 
189 	/*
190 	 * Find real root, and make vfs point to real vfs
191 	 */
192 
193 	if (error = lookupname(uap->spec, (uap->flags & MS_SYSSPACE) ?
194 	    UIO_SYSSPACE : UIO_USERSPACE, FOLLOW, NULLVPP, &realrootvp))
195 		return (error);
196 
197 	/*
198 	 * Enforce MAC policy if needed.
199 	 *
200 	 * Loopback mounts must not allow writing up. The dominance test
201 	 * is intended to prevent a global zone caller from accidentally
202 	 * creating write-up conditions between two labeled zones.
203 	 * Local zones can't violate MAC on their own without help from
204 	 * the global zone because they can't name a pathname that
205 	 * they don't already have.
206 	 *
207 	 * The special case check for the NET_MAC_AWARE process flag is
208 	 * to support the case of the automounter in the global zone. We
209 	 * permit automounting of local zone directories such as home
210 	 * directories, into the global zone as required by setlabel,
211 	 * zonecopy, and saving of desktop sessions. Such mounts are
212 	 * trusted not to expose the contents of one zone's directories
213 	 * to another by leaking them through the global zone.
214 	 */
215 	if (is_system_labeled() && crgetzoneid(cr) == GLOBAL_ZONEID) {
216 		char	specname[MAXPATHLEN];
217 		zone_t	*from_zptr;
218 		zone_t	*to_zptr;
219 
220 		if (vnodetopath(NULL, realrootvp, specname,
221 		    sizeof (specname), CRED()) != 0)
222 			return (EACCES);
223 
224 		from_zptr = zone_find_by_path(specname);
225 		to_zptr = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
226 
227 		/*
228 		 * Special case for zone devfs: the zone for /dev will
229 		 * incorrectly appear as the global zone since it's not
230 		 * under the zone rootpath.  So for zone devfs check allow
231 		 * read-write mounts.
232 		 *
233 		 * Second special case for scratch zones used for Live Upgrade:
234 		 * this is used to mount the zone's root from /root to /a in
235 		 * the scratch zone.  As with the other special case, this
236 		 * appears to be outside of the zone because it's not under
237 		 * the zone rootpath, which is $ZONEPATH/lu in the scratch
238 		 * zone case.
239 		 */
240 
241 		if (from_zptr != to_zptr &&
242 		    !(to_zptr->zone_flags & ZF_IS_SCRATCH)) {
243 			/*
244 			 * We know at this point that the labels aren't equal
245 			 * because the zone pointers aren't equal, and zones
246 			 * can't share a label.
247 			 *
248 			 * If the source is the global zone then making
249 			 * it available to a local zone must be done in
250 			 * read-only mode as the label will become admin_low.
251 			 *
252 			 * If it is a mount between local zones then if
253 			 * the current process is in the global zone and has
254 			 * the NET_MAC_AWARE flag, then regular read-write
255 			 * access is allowed.  If it's in some other zone, but
256 			 * the label on the mount point dominates the original
257 			 * source, then allow the mount as read-only
258 			 * ("read-down").
259 			 */
260 			if (from_zptr->zone_id == GLOBAL_ZONEID) {
261 				/* make the mount read-only */
262 				vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
263 			} else { /* cross-zone mount */
264 				if (to_zptr->zone_id == GLOBAL_ZONEID &&
265 				    /* LINTED: no consequent */
266 				    getpflags(NET_MAC_AWARE, cr) != 0) {
267 					/* Allow the mount as read-write */
268 				} else if (bldominates(
269 				    label2bslabel(to_zptr->zone_slabel),
270 				    label2bslabel(from_zptr->zone_slabel))) {
271 					/* make the mount read-only */
272 					vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
273 				} else {
274 					zone_rele(to_zptr);
275 					zone_rele(from_zptr);
276 					return (EACCES);
277 				}
278 			}
279 		}
280 		zone_rele(to_zptr);
281 		zone_rele(from_zptr);
282 	}
283 
284 	/*
285 	 * realrootvp may be an AUTOFS node, in which case we
286 	 * perform a VOP_ACCESS() to trigger the mount of the
287 	 * intended filesystem, so we loopback mount the intended
288 	 * filesystem instead of the AUTOFS filesystem.
289 	 */
290 	(void) VOP_ACCESS(realrootvp, 0, 0, cr, NULL);
291 
292 	/*
293 	 * We're interested in the top most filesystem.
294 	 * This is specially important when uap->spec is a trigger
295 	 * AUTOFS node, since we're really interested in mounting the
296 	 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
297 	 * call not the AUTOFS node itself.
298 	 */
299 	if (vn_mountedvfs(realrootvp) != NULL) {
300 		if (error = traverse(&realrootvp)) {
301 			VN_RELE(realrootvp);
302 			return (error);
303 		}
304 	}
305 
306 	/*
307 	 * Allocate a vfs info struct and attach it
308 	 */
309 	li = kmem_zalloc(sizeof (struct loinfo), KM_SLEEP);
310 	li->li_realvfs = realrootvp->v_vfsp;
311 	li->li_mountvfs = vfsp;
312 
313 	/*
314 	 * Set mount flags to be inherited by loopback vfs's
315 	 */
316 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) {
317 		li->li_mflag |= VFS_RDONLY;
318 	}
319 	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
320 		li->li_mflag |= (VFS_NOSETUID|VFS_NODEVICES);
321 	}
322 	if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
323 		li->li_mflag |= VFS_NODEVICES;
324 	}
325 	if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
326 		li->li_mflag |= VFS_NOSETUID;
327 	}
328 	/*
329 	 * Permissive flags are added to the "deny" bitmap.
330 	 */
331 	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
332 		li->li_dflag |= VFS_XATTR;
333 	}
334 	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
335 		li->li_dflag |= VFS_NBMAND;
336 	}
337 
338 	/*
339 	 * Propagate inheritable mount flags from the real vfs.
340 	 */
341 	if ((li->li_realvfs->vfs_flag & VFS_RDONLY) &&
342 	    !vfs_optionisset(vfsp, MNTOPT_RO, NULL))
343 		vfs_setmntopt(vfsp, MNTOPT_RO, NULL,
344 		    VFS_NODISPLAY);
345 	if ((li->li_realvfs->vfs_flag & VFS_NOSETUID) &&
346 	    !vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL))
347 		vfs_setmntopt(vfsp, MNTOPT_NOSETUID, NULL,
348 		    VFS_NODISPLAY);
349 	if ((li->li_realvfs->vfs_flag & VFS_NODEVICES) &&
350 	    !vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL))
351 		vfs_setmntopt(vfsp, MNTOPT_NODEVICES, NULL,
352 		    VFS_NODISPLAY);
353 	/*
354 	 * Permissive flags such as VFS_XATTR, as opposed to restrictive flags
355 	 * such as VFS_RDONLY, are handled differently.  An explicit
356 	 * MNTOPT_NOXATTR should override the underlying filesystem's VFS_XATTR.
357 	 */
358 	if ((li->li_realvfs->vfs_flag & VFS_XATTR) &&
359 	    !vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL) &&
360 	    !vfs_optionisset(vfsp, MNTOPT_XATTR, NULL))
361 		vfs_setmntopt(vfsp, MNTOPT_XATTR, NULL,
362 		    VFS_NODISPLAY);
363 	if ((li->li_realvfs->vfs_flag & VFS_NBMAND) &&
364 	    !vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL) &&
365 	    !vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL))
366 		vfs_setmntopt(vfsp, MNTOPT_NBMAND, NULL,
367 		    VFS_NODISPLAY);
368 
369 	li->li_refct = 0;
370 	vfsp->vfs_data = (caddr_t)li;
371 	vfsp->vfs_bcount = 0;
372 	vfsp->vfs_fstype = lofsfstype;
373 	vfsp->vfs_bsize = li->li_realvfs->vfs_bsize;
374 
375 	vfsp->vfs_dev = li->li_realvfs->vfs_dev;
376 	vfsp->vfs_fsid.val[0] = li->li_realvfs->vfs_fsid.val[0];
377 	vfsp->vfs_fsid.val[1] = li->li_realvfs->vfs_fsid.val[1];
378 
379 	if (vfs_optionisset(vfsp, MNTOPT_LOFS_NOSUB, NULL)) {
380 		li->li_flag |= LO_NOSUB;
381 	}
382 
383 	/*
384 	 * Propagate any VFS features
385 	 */
386 
387 	vfs_propagate_features(li->li_realvfs, vfsp);
388 
389 	/*
390 	 * Setup the hashtable. If the root of this mount isn't a directory,
391 	 * there's no point in allocating a large hashtable. A table with one
392 	 * bucket is sufficient.
393 	 */
394 	if (realrootvp->v_type != VDIR)
395 		lsetup(li, 1);
396 	else
397 		lsetup(li, 0);
398 
399 	/*
400 	 * Make the root vnode
401 	 */
402 	srootvp = makelonode(realrootvp, li, 0);
403 	srootvp->v_flag |= VROOT;
404 	li->li_rootvp = srootvp;
405 
406 #ifdef LODEBUG
407 	lo_dprint(4, "lo_mount: vfs %p realvfs %p root %p realroot %p li %p\n",
408 	    vfsp, li->li_realvfs, srootvp, realrootvp, li);
409 #endif
410 	return (0);
411 }
412 
413 /*
414  * Undo loopback mount
415  */
416 static int
417 lo_unmount(struct vfs *vfsp, int flag, struct cred *cr)
418 {
419 	struct loinfo *li;
420 
421 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
422 		return (EPERM);
423 
424 	/*
425 	 * Forced unmount is not supported by this file system
426 	 * and thus, ENOTSUP, is being returned.
427 	 */
428 	if (flag & MS_FORCE)
429 		return (ENOTSUP);
430 
431 	li = vtoli(vfsp);
432 #ifdef LODEBUG
433 	lo_dprint(4, "lo_unmount(%p) li %p\n", vfsp, li);
434 #endif
435 	if (li->li_refct != 1 || li->li_rootvp->v_count != 1) {
436 #ifdef LODEBUG
437 		lo_dprint(4, "refct %d v_ct %d\n", li->li_refct,
438 		    li->li_rootvp->v_count);
439 #endif
440 		return (EBUSY);
441 	}
442 	VN_RELE(li->li_rootvp);
443 	return (0);
444 }
445 
446 /*
447  * Find root of lofs mount.
448  */
449 static int
450 lo_root(struct vfs *vfsp, struct vnode **vpp)
451 {
452 	*vpp = vtoli(vfsp)->li_rootvp;
453 #ifdef LODEBUG
454 	lo_dprint(4, "lo_root(0x%p) = %p\n", vfsp, *vpp);
455 #endif
456 	/*
457 	 * If the root of the filesystem is a special file, return the specvp
458 	 * version of the vnode. We don't save the specvp vnode in our
459 	 * hashtable since that's exclusively for lnodes.
460 	 */
461 	if (IS_DEVVP(*vpp)) {
462 		struct vnode *svp;
463 
464 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, kcred);
465 		if (svp == NULL)
466 			return (ENOSYS);
467 		*vpp = svp;
468 	} else {
469 		VN_HOLD(*vpp);
470 	}
471 
472 	return (0);
473 }
474 
475 /*
476  * Get file system statistics.
477  */
478 static int
479 lo_statvfs(register struct vfs *vfsp, struct statvfs64 *sbp)
480 {
481 	vnode_t *realrootvp;
482 
483 #ifdef LODEBUG
484 	lo_dprint(4, "lostatvfs %p\n", vfsp);
485 #endif
486 	/*
487 	 * Using realrootvp->v_vfsp (instead of the realvfsp that was
488 	 * cached) is necessary to make lofs work woth forced UFS unmounts.
489 	 * In the case of a forced unmount, UFS stores a set of dummy vfsops
490 	 * in all the (i)vnodes in the filesystem. The dummy ops simply
491 	 * returns back EIO.
492 	 */
493 	(void) lo_realvfs(vfsp, &realrootvp);
494 	if (realrootvp != NULL)
495 		return (VFS_STATVFS(realrootvp->v_vfsp, sbp));
496 	else
497 		return (EIO);
498 }
499 
500 /*
501  * LOFS doesn't have any data or metadata to flush, pending I/O on the
502  * underlying filesystem will be flushed when such filesystem is synched.
503  */
504 /* ARGSUSED */
505 static int
506 lo_sync(struct vfs *vfsp,
507 	short flag,
508 	struct cred *cr)
509 {
510 #ifdef LODEBUG
511 	lo_dprint(4, "lo_sync: %p\n", vfsp);
512 #endif
513 	return (0);
514 }
515 
516 /*
517  * Obtain the vnode from the underlying filesystem.
518  */
519 static int
520 lo_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp)
521 {
522 	vnode_t *realrootvp;
523 
524 #ifdef LODEBUG
525 	lo_dprint(4, "lo_vget: %p\n", vfsp);
526 #endif
527 	(void) lo_realvfs(vfsp, &realrootvp);
528 	if (realrootvp != NULL)
529 		return (VFS_VGET(realrootvp->v_vfsp, vpp, fidp));
530 	else
531 		return (EIO);
532 }
533 
534 /*
535  * Free mount-specific data.
536  */
537 static void
538 lo_freevfs(struct vfs *vfsp)
539 {
540 	struct loinfo *li = vtoli(vfsp);
541 
542 	ldestroy(li);
543 	kmem_free(li, sizeof (struct loinfo));
544 }
545 
546 static int
547 lofsinit(int fstyp, char *name)
548 {
549 	static const fs_operation_def_t lo_vfsops_template[] = {
550 		VFSNAME_MOUNT,		{ .vfs_mount = lo_mount },
551 		VFSNAME_UNMOUNT,	{ .vfs_unmount = lo_unmount },
552 		VFSNAME_ROOT,		{ .vfs_root = lo_root },
553 		VFSNAME_STATVFS,	{ .vfs_statvfs = lo_statvfs },
554 		VFSNAME_SYNC,		{ .vfs_sync = lo_sync },
555 		VFSNAME_VGET,		{ .vfs_vget = lo_vget },
556 		VFSNAME_FREEVFS,	{ .vfs_freevfs = lo_freevfs },
557 		NULL,			NULL
558 	};
559 	int error;
560 
561 	error = vfs_setfsops(fstyp, lo_vfsops_template, &lo_vfsops);
562 	if (error != 0) {
563 		cmn_err(CE_WARN, "lofsinit: bad vfs ops template");
564 		return (error);
565 	}
566 
567 	error = vn_make_ops(name, lo_vnodeops_template, &lo_vnodeops);
568 	if (error != 0) {
569 		(void) vfs_freevfsops_by_type(fstyp);
570 		cmn_err(CE_WARN, "lofsinit: bad vnode ops template");
571 		return (error);
572 	}
573 
574 	lofsfstype = fstyp;
575 
576 	return (0);
577 }
578