xref: /titanic_50/usr/src/uts/common/fs/lofs/lofs_vfsops.c (revision 03831d35f7499c87d51205817c93e9a8d42c4bae)
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 2006 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/vnode.h>
32 #include <sys/uio.h>
33 #include <sys/pathname.h>
34 #include <sys/kmem.h>
35 #include <sys/cred.h>
36 #include <sys/statvfs.h>
37 #include <sys/fs/lofs_info.h>
38 #include <sys/fs/lofs_node.h>
39 #include <sys/mount.h>
40 #include <sys/mntent.h>
41 #include <sys/mkdev.h>
42 #include <sys/priv.h>
43 #include <sys/sysmacros.h>
44 #include <sys/systm.h>
45 #include <sys/cmn_err.h>
46 #include <sys/policy.h>
47 #include <sys/tsol/label.h>
48 #include "fs/fs_subr.h"
49 
50 /*
51  * This is the loadable module wrapper.
52  */
53 #include <sys/modctl.h>
54 
55 static mntopts_t lofs_mntopts;
56 
57 static int lofsinit(int, char *);
58 
59 static vfsdef_t vfw = {
60 	VFSDEF_VERSION,
61 	"lofs",
62 	lofsinit,
63 	VSW_HASPROTO|VSW_STATS,
64 	&lofs_mntopts
65 };
66 
67 /*
68  * Stuff needed to support "zonedevfs" mode.
69  */
70 static major_t lofs_major;
71 static minor_t lofs_minor;
72 static kmutex_t lofs_minor_lock;
73 
74 /*
75  * LOFS mount options table
76  */
77 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
78 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
79 static char *zonedevfs_cancel[] = { MNTOPT_LOFS_NOZONEDEVFS, NULL };
80 static char *nozonedevfs_cancel[] = { MNTOPT_LOFS_ZONEDEVFS, NULL };
81 static char *sub_cancel[] = { MNTOPT_LOFS_NOSUB, NULL };
82 static char *nosub_cancel[] = { MNTOPT_LOFS_SUB, NULL };
83 
84 static mntopt_t mntopts[] = {
85 /*
86  *	option name		cancel option	default arg	flags
87  *		private data
88  */
89 	{ MNTOPT_XATTR,		xattr_cancel,	NULL,		0,
90 		(void *)0 },
91 	{ MNTOPT_NOXATTR,	noxattr_cancel,	NULL,		0,
92 		(void *)0 },
93 	{ MNTOPT_LOFS_ZONEDEVFS,	zonedevfs_cancel,	NULL,	0,
94 		(void *)0 },
95 	{ MNTOPT_LOFS_NOZONEDEVFS,	nozonedevfs_cancel,	NULL,	0,
96 		(void *)0 },
97 	{ MNTOPT_LOFS_SUB,	sub_cancel,	NULL,		0,
98 		(void *)0 },
99 	{ MNTOPT_LOFS_NOSUB,	nosub_cancel,	NULL,		0,
100 		(void *)0 },
101 };
102 
103 static mntopts_t lofs_mntopts = {
104 	sizeof (mntopts) / sizeof (mntopt_t),
105 	mntopts
106 };
107 
108 /*
109  * Module linkage information for the kernel.
110  */
111 
112 static struct modlfs modlfs = {
113 	&mod_fsops, "filesystem for lofs", &vfw
114 };
115 
116 static struct modlinkage modlinkage = {
117 	MODREV_1, (void *)&modlfs, NULL
118 };
119 
120 /*
121  * This is the module initialization routine.
122  */
123 
124 int
125 _init(void)
126 {
127 	int status;
128 
129 	lofs_subrinit();
130 	status = mod_install(&modlinkage);
131 	if (status != 0) {
132 		/*
133 		 * Cleanup previously initialized work.
134 		 */
135 		lofs_subrfini();
136 	}
137 
138 	return (status);
139 }
140 
141 /*
142  * Don't allow the lofs module to be unloaded for now.
143  * There is a memory leak if it gets unloaded.
144  */
145 
146 int
147 _fini(void)
148 {
149 	return (EBUSY);
150 }
151 
152 int
153 _info(struct modinfo *modinfop)
154 {
155 	return (mod_info(&modlinkage, modinfop));
156 }
157 
158 
159 static int lofsfstype;
160 vfsops_t *lo_vfsops;
161 
162 /*
163  * lo mount vfsop
164  * Set up mount info record and attach it to vfs struct.
165  */
166 /*ARGSUSED*/
167 static int
168 lo_mount(struct vfs *vfsp,
169 	struct vnode *vp,
170 	struct mounta *uap,
171 	struct cred *cr)
172 {
173 	int error;
174 	struct vnode *srootvp = NULL;	/* the server's root */
175 	struct vnode *realrootvp;
176 	struct loinfo *li;
177 	int is_zonedevfs = 0;
178 	int nodev;
179 
180 	nodev = vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL);
181 
182 	if ((error = secpolicy_fs_mount(cr, vp, vfsp)) != 0)
183 		return (EPERM);
184 
185 	/*
186 	 * Loopback devices which get "nodevices" added can be done without
187 	 * "nodevices" set because we cannot import devices into a zone
188 	 * with loopback.  Note that we have all zone privileges when
189 	 * this happens; if not, we'd have gotten "nosuid".
190 	 */
191 	if (!nodev && vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL))
192 		vfs_setmntopt(vfsp, MNTOPT_DEVICES, NULL, VFS_NODISPLAY);
193 
194 	/*
195 	 * We must ensure that only the global zone applies the 'zonedevfs'
196 	 * option; we don't want non-global zones to be able to establish
197 	 * lofs mounts using the special dev_t we use to ensure that the
198 	 * contents of a zone's /dev cannot be victim to link(2) or rename(2).
199 	 * See below, where we set all of this up.
200 	 *
201 	 * Since this is more like a privilege check, we use crgetzoneid(cr)
202 	 * instead of getzoneid().
203 	 */
204 	is_zonedevfs = vfs_optionisset(vfsp, MNTOPT_LOFS_ZONEDEVFS, NULL);
205 	if (crgetzoneid(cr) != GLOBAL_ZONEID && is_zonedevfs)
206 		return (EPERM);
207 
208 	mutex_enter(&vp->v_lock);
209 	if (!(uap->flags & MS_OVERLAY) &&
210 	    (vp->v_count != 1 || (vp->v_flag & VROOT))) {
211 		mutex_exit(&vp->v_lock);
212 		return (EBUSY);
213 	}
214 	mutex_exit(&vp->v_lock);
215 
216 	/*
217 	 * Find real root, and make vfs point to real vfs
218 	 */
219 	if (error = lookupname(uap->spec, (uap->flags & MS_SYSSPACE) ?
220 		UIO_SYSSPACE : UIO_USERSPACE, FOLLOW, NULLVPP,
221 	    &realrootvp))
222 		return (error);
223 
224 	/*
225 	 * Enforce MAC policy if needed.
226 	 *
227 	 * Loopback mounts must not allow writing up. The dominance test
228 	 * is intended to prevent a global zone caller from accidentally
229 	 * creating write-up conditions between two labeled zones.
230 	 * Local zones can't violate MAC on their own without help from
231 	 * the global zone because they can't name a pathname that
232 	 * they don't already have.
233 	 *
234 	 * The special case check for the NET_MAC_AWARE process flag is
235 	 * to support the case of the automounter in the global zone. We
236 	 * permit automounting of local zone directories such as home
237 	 * directories, into the global zone as required by setlabel,
238 	 * zonecopy, and saving of desktop sessions. Such mounts are
239 	 * trusted not to expose the contents of one zone's directories
240 	 * to another by leaking them through the global zone.
241 	 */
242 	if (is_system_labeled() && crgetzoneid(cr) == GLOBAL_ZONEID) {
243 		void *specname;
244 		zone_t *from_zptr;
245 		zone_t *to_zptr;
246 
247 		if (uap->flags & MS_SYSSPACE) {
248 			specname = uap->spec;
249 		} else {
250 			specname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
251 			error = copyinstr(uap->spec, specname, MAXPATHLEN,
252 			    NULL);
253 			if (error) {
254 				kmem_free(specname, MAXPATHLEN);
255 				return (error);
256 			}
257 		}
258 		from_zptr = zone_find_by_path(specname);
259 		if (!(uap->flags & MS_SYSSPACE))
260 			kmem_free(specname, MAXPATHLEN);
261 
262 		to_zptr = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
263 
264 		/*
265 		 * Special case for zone devfs: the zone for /dev will
266 		 * incorrectly appear as the global zone since it's not
267 		 * under the zone rootpath.  So for zone devfs check allow
268 		 * read-write mounts.
269 		 */
270 
271 		if (from_zptr != to_zptr && !is_zonedevfs) {
272 			/*
273 			 * We know at this point that the labels aren't equal
274 			 * because the zone pointers aren't equal, and zones
275 			 * can't share a label.
276 			 *
277 			 * If the source is the global zone then making
278 			 * it available to a local zone must be done in
279 			 * read-only mode as the label will become admin_low.
280 			 *
281 			 * If it is a mount between local zones then if
282 			 * the current process is in the global zone and has
283 			 * the NET_MAC_AWARE flag, then regular read-write
284 			 * access is allowed.  If it's in some other zone, but
285 			 * the label on the mount point dominates the original
286 			 * source, then allow the mount as read-only
287 			 * ("read-down").
288 			 */
289 			if (from_zptr->zone_id == GLOBAL_ZONEID) {
290 				/* make the mount read-only */
291 				vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
292 			} else { /* cross-zone mount */
293 				if (to_zptr->zone_id == GLOBAL_ZONEID &&
294 				    /* LINTED: no consequent */
295 				    getpflags(NET_MAC_AWARE, cr) != 0) {
296 					/* Allow the mount as read-write */
297 				} else if (bldominates(
298 				    label2bslabel(to_zptr->zone_slabel),
299 				    label2bslabel(from_zptr->zone_slabel))) {
300 					/* make the mount read-only */
301 					vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
302 				} else {
303 					zone_rele(to_zptr);
304 					zone_rele(from_zptr);
305 					return (EACCES);
306 				}
307 			}
308 		}
309 		zone_rele(to_zptr);
310 		zone_rele(from_zptr);
311 	}
312 
313 	/*
314 	 * realrootvp may be an AUTOFS node, in which case we
315 	 * perform a VOP_ACCESS() to trigger the mount of the
316 	 * intended filesystem, so we loopback mount the intended
317 	 * filesystem instead of the AUTOFS filesystem.
318 	 */
319 	(void) VOP_ACCESS(realrootvp, 0, 0, cr);
320 
321 	/*
322 	 * We're interested in the top most filesystem.
323 	 * This is specially important when uap->spec is a trigger
324 	 * AUTOFS node, since we're really interested in mounting the
325 	 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
326 	 * call not the AUTOFS node itself.
327 	 */
328 	if (vn_mountedvfs(realrootvp) != NULL) {
329 		if (error = traverse(&realrootvp)) {
330 			VN_RELE(realrootvp);
331 			return (error);
332 		}
333 	}
334 
335 	/*
336 	 * Allocate a vfs info struct and attach it
337 	 */
338 	li = kmem_zalloc(sizeof (struct loinfo), KM_SLEEP);
339 	li->li_realvfs = realrootvp->v_vfsp;
340 	li->li_mountvfs = vfsp;
341 
342 	/*
343 	 * Set mount flags to be inherited by loopback vfs's
344 	 */
345 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) {
346 		li->li_mflag |= VFS_RDONLY;
347 	}
348 	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
349 		li->li_mflag |= (VFS_NOSETUID|VFS_NODEVICES);
350 	}
351 	if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
352 		li->li_mflag |= VFS_NODEVICES;
353 	}
354 	if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
355 		li->li_mflag |= VFS_NOSETUID;
356 	}
357 	/*
358 	 * Permissive flags are added to the "deny" bitmap.
359 	 */
360 	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
361 		li->li_dflag |= VFS_XATTR;
362 	}
363 	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
364 		li->li_dflag |= VFS_NBMAND;
365 	}
366 
367 	/*
368 	 * Propagate inheritable mount flags from the real vfs.
369 	 */
370 	if ((li->li_realvfs->vfs_flag & VFS_RDONLY) &&
371 	    !vfs_optionisset(vfsp, MNTOPT_RO, NULL))
372 		vfs_setmntopt(vfsp, MNTOPT_RO, NULL,
373 		    VFS_NODISPLAY);
374 	if ((li->li_realvfs->vfs_flag & VFS_NOSETUID) &&
375 	    !vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL))
376 		vfs_setmntopt(vfsp, MNTOPT_NOSETUID, NULL,
377 		    VFS_NODISPLAY);
378 	if ((li->li_realvfs->vfs_flag & VFS_NODEVICES) &&
379 	    !vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL))
380 		vfs_setmntopt(vfsp, MNTOPT_NODEVICES, NULL,
381 		    VFS_NODISPLAY);
382 	/*
383 	 * Permissive flags such as VFS_XATTR, as opposed to restrictive flags
384 	 * such as VFS_RDONLY, are handled differently.  An explicit
385 	 * MNTOPT_NOXATTR should override the underlying filesystem's VFS_XATTR.
386 	 */
387 	if ((li->li_realvfs->vfs_flag & VFS_XATTR) &&
388 	    !vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL) &&
389 	    !vfs_optionisset(vfsp, MNTOPT_XATTR, NULL))
390 		vfs_setmntopt(vfsp, MNTOPT_XATTR, NULL,
391 		    VFS_NODISPLAY);
392 	if ((li->li_realvfs->vfs_flag & VFS_NBMAND) &&
393 	    !vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL) &&
394 	    !vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL))
395 		vfs_setmntopt(vfsp, MNTOPT_NBMAND, NULL,
396 		    VFS_NODISPLAY);
397 
398 	li->li_refct = 0;
399 	vfsp->vfs_data = (caddr_t)li;
400 	vfsp->vfs_bcount = 0;
401 	vfsp->vfs_fstype = lofsfstype;
402 	vfsp->vfs_bsize = li->li_realvfs->vfs_bsize;
403 
404 	/*
405 	 * Test to see if we need to be in "zone /dev" mode.  In zonedevfs
406 	 * mode, we pull a nasty trick; we make sure that the lofs dev_t does
407 	 * *not* reflect the underlying device, so that no renames or links
408 	 * can occur to or from the /dev hierarchy.
409 	 */
410 	if (is_zonedevfs) {
411 		dev_t dev;
412 
413 		mutex_enter(&lofs_minor_lock);
414 		do {
415 			lofs_minor = (lofs_minor + 1) & MAXMIN32;
416 			dev = makedevice(lofs_major, lofs_minor);
417 		} while (vfs_devismounted(dev));
418 		mutex_exit(&lofs_minor_lock);
419 
420 		vfsp->vfs_dev = dev;
421 		vfs_make_fsid(&vfsp->vfs_fsid, dev, lofsfstype);
422 
423 		li->li_flag |= LO_ZONEDEVFS;
424 	} else {
425 		vfsp->vfs_dev = li->li_realvfs->vfs_dev;
426 		vfsp->vfs_fsid.val[0] = li->li_realvfs->vfs_fsid.val[0];
427 		vfsp->vfs_fsid.val[1] = li->li_realvfs->vfs_fsid.val[1];
428 	}
429 
430 	if (vfs_optionisset(vfsp, MNTOPT_LOFS_NOSUB, NULL)) {
431 		li->li_flag |= LO_NOSUB;
432 	}
433 
434 	/*
435 	 * Setup the hashtable. If the root of this mount isn't a directory,
436 	 * there's no point in allocating a large hashtable. A table with one
437 	 * bucket is sufficient.
438 	 */
439 	if (realrootvp->v_type != VDIR)
440 		lsetup(li, 1);
441 	else
442 		lsetup(li, 0);
443 
444 	/*
445 	 * Make the root vnode
446 	 */
447 	srootvp = makelonode(realrootvp, li, 0);
448 	srootvp->v_flag |= VROOT;
449 	li->li_rootvp = srootvp;
450 
451 #ifdef LODEBUG
452 	lo_dprint(4, "lo_mount: vfs %p realvfs %p root %p realroot %p li %p\n",
453 	    vfsp, li->li_realvfs, srootvp, realrootvp, li);
454 #endif
455 	return (0);
456 }
457 
458 /*
459  * Undo loopback mount
460  */
461 static int
462 lo_unmount(struct vfs *vfsp, int flag, struct cred *cr)
463 {
464 	struct loinfo *li;
465 
466 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
467 		return (EPERM);
468 
469 	/*
470 	 * Forced unmount is not supported by this file system
471 	 * and thus, ENOTSUP, is being returned.
472 	 */
473 	if (flag & MS_FORCE)
474 		return (ENOTSUP);
475 
476 	li = vtoli(vfsp);
477 #ifdef LODEBUG
478 	lo_dprint(4, "lo_unmount(%p) li %p\n", vfsp, li);
479 #endif
480 	if (li->li_refct != 1 || li->li_rootvp->v_count != 1) {
481 #ifdef LODEBUG
482 		lo_dprint(4, "refct %d v_ct %d\n", li->li_refct,
483 		    li->li_rootvp->v_count);
484 #endif
485 		return (EBUSY);
486 	}
487 	VN_RELE(li->li_rootvp);
488 	return (0);
489 }
490 
491 /*
492  * Find root of lofs mount.
493  */
494 static int
495 lo_root(struct vfs *vfsp, struct vnode **vpp)
496 {
497 	*vpp = vtoli(vfsp)->li_rootvp;
498 #ifdef LODEBUG
499 	lo_dprint(4, "lo_root(0x%p) = %p\n", vfsp, *vpp);
500 #endif
501 	/*
502 	 * If the root of the filesystem is a special file, return the specvp
503 	 * version of the vnode. We don't save the specvp vnode in our
504 	 * hashtable since that's exclusively for lnodes.
505 	 */
506 	if (IS_DEVVP(*vpp)) {
507 		struct vnode *svp;
508 
509 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, kcred);
510 		if (svp == NULL)
511 			return (ENOSYS);
512 		*vpp = svp;
513 	} else {
514 		VN_HOLD(*vpp);
515 	}
516 
517 	return (0);
518 }
519 
520 /*
521  * Get file system statistics.
522  */
523 static int
524 lo_statvfs(register struct vfs *vfsp, struct statvfs64 *sbp)
525 {
526 	vnode_t *realrootvp;
527 
528 #ifdef LODEBUG
529 	lo_dprint(4, "lostatvfs %p\n", vfsp);
530 #endif
531 	/*
532 	 * Using realrootvp->v_vfsp (instead of the realvfsp that was
533 	 * cached) is necessary to make lofs work woth forced UFS unmounts.
534 	 * In the case of a forced unmount, UFS stores a set of dummy vfsops
535 	 * in all the (i)vnodes in the filesystem. The dummy ops simply
536 	 * returns back EIO.
537 	 */
538 	(void) lo_realvfs(vfsp, &realrootvp);
539 	if (realrootvp != NULL)
540 		return (VFS_STATVFS(realrootvp->v_vfsp, sbp));
541 	else
542 		return (EIO);
543 }
544 
545 /*
546  * LOFS doesn't have any data or metadata to flush, pending I/O on the
547  * underlying filesystem will be flushed when such filesystem is synched.
548  */
549 /* ARGSUSED */
550 static int
551 lo_sync(struct vfs *vfsp,
552 	short flag,
553 	struct cred *cr)
554 {
555 #ifdef LODEBUG
556 	lo_dprint(4, "lo_sync: %p\n", vfsp);
557 #endif
558 	return (0);
559 }
560 
561 /*
562  * Obtain the vnode from the underlying filesystem.
563  */
564 static int
565 lo_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp)
566 {
567 	vnode_t *realrootvp;
568 
569 #ifdef LODEBUG
570 	lo_dprint(4, "lo_vget: %p\n", vfsp);
571 #endif
572 	(void) lo_realvfs(vfsp, &realrootvp);
573 	if (realrootvp != NULL)
574 		return (VFS_VGET(realrootvp->v_vfsp, vpp, fidp));
575 	else
576 		return (EIO);
577 }
578 
579 /*
580  * Free mount-specific data.
581  */
582 static void
583 lo_freevfs(struct vfs *vfsp)
584 {
585 	struct loinfo *li = vtoli(vfsp);
586 
587 	ldestroy(li);
588 	kmem_free(li, sizeof (struct loinfo));
589 }
590 
591 static int
592 lofsinit(int fstyp, char *name)
593 {
594 	static const fs_operation_def_t lo_vfsops_template[] = {
595 		VFSNAME_MOUNT, lo_mount,
596 		VFSNAME_UNMOUNT, lo_unmount,
597 		VFSNAME_ROOT, lo_root,
598 		VFSNAME_STATVFS, lo_statvfs,
599 		VFSNAME_SYNC, (fs_generic_func_p) lo_sync,
600 		VFSNAME_VGET, lo_vget,
601 		VFSNAME_FREEVFS, (fs_generic_func_p) lo_freevfs,
602 		NULL, NULL
603 	};
604 	int error;
605 
606 	error = vfs_setfsops(fstyp, lo_vfsops_template, &lo_vfsops);
607 	if (error != 0) {
608 		cmn_err(CE_WARN, "lofsinit: bad vfs ops template");
609 		return (error);
610 	}
611 
612 	error = vn_make_ops(name, lo_vnodeops_template, &lo_vnodeops);
613 	if (error != 0) {
614 		(void) vfs_freevfsops_by_type(fstyp);
615 		cmn_err(CE_WARN, "lofsinit: bad vnode ops template");
616 		return (error);
617 	}
618 
619 	lofsfstype = fstyp;
620 
621 	if ((lofs_major = getudev()) == (major_t)-1) {
622 		(void) vfs_freevfsops_by_type(fstyp);
623 		cmn_err(CE_WARN, "lofsinit: Can't get unique device number.");
624 		return (ENXIO);
625 	}
626 
627 	lofs_minor = 0;
628 	mutex_init(&lofs_minor_lock, NULL, MUTEX_DEFAULT, NULL);
629 
630 	return (0);
631 }
632