FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_mount.c
1 /*-
2 * Copyright (c) 1999-2004 Poul-Henning Kamp
3 * Copyright (c) 1999 Michael Smith
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/10.4/sys/kern/vfs_mount.c 309208 2016-11-27 09:14:52Z kib $");
39
40 #include <sys/param.h>
41 #include <sys/conf.h>
42 #include <sys/fcntl.h>
43 #include <sys/jail.h>
44 #include <sys/kernel.h>
45 #include <sys/libkern.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/filedesc.h>
53 #include <sys/reboot.h>
54 #include <sys/sbuf.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysproto.h>
57 #include <sys/sx.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysent.h>
60 #include <sys/systm.h>
61 #include <sys/vnode.h>
62 #include <vm/uma.h>
63
64 #include <geom/geom.h>
65
66 #include <machine/stdarg.h>
67
68 #include <security/audit/audit.h>
69 #include <security/mac/mac_framework.h>
70
71 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
72
73 static int vfs_domount(struct thread *td, const char *fstype, char *fspath,
74 uint64_t fsflags, struct vfsoptlist **optlist);
75 static void free_mntarg(struct mntarg *ma);
76
77 static int usermount = 0;
78 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
79 "Unprivileged users may mount and unmount file systems");
80
81 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
82 static uma_zone_t mount_zone;
83
84 /* List of mounted filesystems. */
85 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
86
87 /* For any iteration/modification of mountlist */
88 struct mtx mountlist_mtx;
89 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
90
91 /*
92 * Global opts, taken by all filesystems
93 */
94 static const char *global_opts[] = {
95 "errmsg",
96 "fstype",
97 "fspath",
98 "ro",
99 "rw",
100 "nosuid",
101 "noexec",
102 NULL
103 };
104
105 static int
106 mount_init(void *mem, int size, int flags)
107 {
108 struct mount *mp;
109
110 mp = (struct mount *)mem;
111 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
112 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
113 return (0);
114 }
115
116 static void
117 mount_fini(void *mem, int size)
118 {
119 struct mount *mp;
120
121 mp = (struct mount *)mem;
122 lockdestroy(&mp->mnt_explock);
123 mtx_destroy(&mp->mnt_mtx);
124 }
125
126 static void
127 vfs_mount_init(void *dummy __unused)
128 {
129
130 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
131 NULL, mount_init, mount_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
132 }
133 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
134
135 /*
136 * ---------------------------------------------------------------------
137 * Functions for building and sanitizing the mount options
138 */
139
140 /* Remove one mount option. */
141 static void
142 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
143 {
144
145 TAILQ_REMOVE(opts, opt, link);
146 free(opt->name, M_MOUNT);
147 if (opt->value != NULL)
148 free(opt->value, M_MOUNT);
149 free(opt, M_MOUNT);
150 }
151
152 /* Release all resources related to the mount options. */
153 void
154 vfs_freeopts(struct vfsoptlist *opts)
155 {
156 struct vfsopt *opt;
157
158 while (!TAILQ_EMPTY(opts)) {
159 opt = TAILQ_FIRST(opts);
160 vfs_freeopt(opts, opt);
161 }
162 free(opts, M_MOUNT);
163 }
164
165 void
166 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
167 {
168 struct vfsopt *opt, *temp;
169
170 if (opts == NULL)
171 return;
172 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
173 if (strcmp(opt->name, name) == 0)
174 vfs_freeopt(opts, opt);
175 }
176 }
177
178 static int
179 vfs_isopt_ro(const char *opt)
180 {
181
182 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
183 strcmp(opt, "norw") == 0)
184 return (1);
185 return (0);
186 }
187
188 static int
189 vfs_isopt_rw(const char *opt)
190 {
191
192 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
193 return (1);
194 return (0);
195 }
196
197 /*
198 * Check if options are equal (with or without the "no" prefix).
199 */
200 static int
201 vfs_equalopts(const char *opt1, const char *opt2)
202 {
203 char *p;
204
205 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
206 if (strcmp(opt1, opt2) == 0)
207 return (1);
208 /* "noopt" vs. "opt" */
209 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
210 return (1);
211 /* "opt" vs. "noopt" */
212 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
213 return (1);
214 while ((p = strchr(opt1, '.')) != NULL &&
215 !strncmp(opt1, opt2, ++p - opt1)) {
216 opt2 += p - opt1;
217 opt1 = p;
218 /* "foo.noopt" vs. "foo.opt" */
219 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
220 return (1);
221 /* "foo.opt" vs. "foo.noopt" */
222 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
223 return (1);
224 }
225 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */
226 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
227 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
228 return (1);
229 return (0);
230 }
231
232 /*
233 * If a mount option is specified several times,
234 * (with or without the "no" prefix) only keep
235 * the last occurrence of it.
236 */
237 static void
238 vfs_sanitizeopts(struct vfsoptlist *opts)
239 {
240 struct vfsopt *opt, *opt2, *tmp;
241
242 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
243 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
244 while (opt2 != NULL) {
245 if (vfs_equalopts(opt->name, opt2->name)) {
246 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
247 vfs_freeopt(opts, opt2);
248 opt2 = tmp;
249 } else {
250 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
251 }
252 }
253 }
254 }
255
256 /*
257 * Build a linked list of mount options from a struct uio.
258 */
259 int
260 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
261 {
262 struct vfsoptlist *opts;
263 struct vfsopt *opt;
264 size_t memused, namelen, optlen;
265 unsigned int i, iovcnt;
266 int error;
267
268 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
269 TAILQ_INIT(opts);
270 memused = 0;
271 iovcnt = auio->uio_iovcnt;
272 for (i = 0; i < iovcnt; i += 2) {
273 namelen = auio->uio_iov[i].iov_len;
274 optlen = auio->uio_iov[i + 1].iov_len;
275 memused += sizeof(struct vfsopt) + optlen + namelen;
276 /*
277 * Avoid consuming too much memory, and attempts to overflow
278 * memused.
279 */
280 if (memused > VFS_MOUNTARG_SIZE_MAX ||
281 optlen > VFS_MOUNTARG_SIZE_MAX ||
282 namelen > VFS_MOUNTARG_SIZE_MAX) {
283 error = EINVAL;
284 goto bad;
285 }
286
287 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
288 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
289 opt->value = NULL;
290 opt->len = 0;
291 opt->pos = i / 2;
292 opt->seen = 0;
293
294 /*
295 * Do this early, so jumps to "bad" will free the current
296 * option.
297 */
298 TAILQ_INSERT_TAIL(opts, opt, link);
299
300 if (auio->uio_segflg == UIO_SYSSPACE) {
301 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
302 } else {
303 error = copyin(auio->uio_iov[i].iov_base, opt->name,
304 namelen);
305 if (error)
306 goto bad;
307 }
308 /* Ensure names are null-terminated strings. */
309 if (namelen == 0 || opt->name[namelen - 1] != '\0') {
310 error = EINVAL;
311 goto bad;
312 }
313 if (optlen != 0) {
314 opt->len = optlen;
315 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
316 if (auio->uio_segflg == UIO_SYSSPACE) {
317 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
318 optlen);
319 } else {
320 error = copyin(auio->uio_iov[i + 1].iov_base,
321 opt->value, optlen);
322 if (error)
323 goto bad;
324 }
325 }
326 }
327 vfs_sanitizeopts(opts);
328 *options = opts;
329 return (0);
330 bad:
331 vfs_freeopts(opts);
332 return (error);
333 }
334
335 /*
336 * Merge the old mount options with the new ones passed
337 * in the MNT_UPDATE case.
338 *
339 * XXX: This function will keep a "nofoo" option in the new
340 * options. E.g, if the option's canonical name is "foo",
341 * "nofoo" ends up in the mount point's active options.
342 */
343 static void
344 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
345 {
346 struct vfsopt *opt, *new;
347
348 TAILQ_FOREACH(opt, oldopts, link) {
349 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
350 new->name = strdup(opt->name, M_MOUNT);
351 if (opt->len != 0) {
352 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
353 bcopy(opt->value, new->value, opt->len);
354 } else
355 new->value = NULL;
356 new->len = opt->len;
357 new->seen = opt->seen;
358 TAILQ_INSERT_HEAD(toopts, new, link);
359 }
360 vfs_sanitizeopts(toopts);
361 }
362
363 /*
364 * Mount a filesystem.
365 */
366 int
367 sys_nmount(td, uap)
368 struct thread *td;
369 struct nmount_args /* {
370 struct iovec *iovp;
371 unsigned int iovcnt;
372 int flags;
373 } */ *uap;
374 {
375 struct uio *auio;
376 int error;
377 u_int iovcnt;
378 uint64_t flags;
379
380 /*
381 * Mount flags are now 64-bits. On 32-bit archtectures only
382 * 32-bits are passed in, but from here on everything handles
383 * 64-bit flags correctly.
384 */
385 flags = uap->flags;
386
387 AUDIT_ARG_FFLAGS(flags);
388 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
389 uap->iovp, uap->iovcnt, flags);
390
391 /*
392 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
393 * userspace to set this flag, but we must filter it out if we want
394 * MNT_UPDATE on the root file system to work.
395 * MNT_ROOTFS should only be set by the kernel when mounting its
396 * root file system.
397 */
398 flags &= ~MNT_ROOTFS;
399
400 iovcnt = uap->iovcnt;
401 /*
402 * Check that we have an even number of iovec's
403 * and that we have at least two options.
404 */
405 if ((iovcnt & 1) || (iovcnt < 4)) {
406 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
407 uap->iovcnt);
408 return (EINVAL);
409 }
410
411 error = copyinuio(uap->iovp, iovcnt, &auio);
412 if (error) {
413 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
414 __func__, error);
415 return (error);
416 }
417 error = vfs_donmount(td, flags, auio);
418
419 free(auio, M_IOV);
420 return (error);
421 }
422
423 /*
424 * ---------------------------------------------------------------------
425 * Various utility functions
426 */
427
428 void
429 vfs_ref(struct mount *mp)
430 {
431
432 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
433 MNT_ILOCK(mp);
434 MNT_REF(mp);
435 MNT_IUNLOCK(mp);
436 }
437
438 void
439 vfs_rel(struct mount *mp)
440 {
441
442 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
443 MNT_ILOCK(mp);
444 MNT_REL(mp);
445 MNT_IUNLOCK(mp);
446 }
447
448 /*
449 * Allocate and initialize the mount point struct.
450 */
451 struct mount *
452 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
453 struct ucred *cred)
454 {
455 struct mount *mp;
456
457 mp = uma_zalloc(mount_zone, M_WAITOK);
458 bzero(&mp->mnt_startzero,
459 __rangeof(struct mount, mnt_startzero, mnt_endzero));
460 TAILQ_INIT(&mp->mnt_nvnodelist);
461 mp->mnt_nvnodelistsize = 0;
462 TAILQ_INIT(&mp->mnt_activevnodelist);
463 mp->mnt_activevnodelistsize = 0;
464 mp->mnt_ref = 0;
465 (void) vfs_busy(mp, MBF_NOWAIT);
466 atomic_add_acq_int(&vfsp->vfc_refcount, 1);
467 mp->mnt_op = vfsp->vfc_vfsops;
468 mp->mnt_vfc = vfsp;
469 mp->mnt_stat.f_type = vfsp->vfc_typenum;
470 mp->mnt_gen++;
471 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
472 mp->mnt_vnodecovered = vp;
473 mp->mnt_cred = crdup(cred);
474 mp->mnt_stat.f_owner = cred->cr_uid;
475 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
476 mp->mnt_iosize_max = DFLTPHYS;
477 #ifdef MAC
478 mac_mount_init(mp);
479 mac_mount_create(cred, mp);
480 #endif
481 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
482 TAILQ_INIT(&mp->mnt_uppers);
483 return (mp);
484 }
485
486 /*
487 * Destroy the mount struct previously allocated by vfs_mount_alloc().
488 */
489 void
490 vfs_mount_destroy(struct mount *mp)
491 {
492
493 MNT_ILOCK(mp);
494 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
495 if (mp->mnt_kern_flag & MNTK_MWAIT) {
496 mp->mnt_kern_flag &= ~MNTK_MWAIT;
497 wakeup(mp);
498 }
499 while (mp->mnt_ref)
500 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
501 KASSERT(mp->mnt_ref == 0,
502 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
503 __FILE__, __LINE__));
504 if (mp->mnt_writeopcount != 0)
505 panic("vfs_mount_destroy: nonzero writeopcount");
506 if (mp->mnt_secondary_writes != 0)
507 panic("vfs_mount_destroy: nonzero secondary_writes");
508 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
509 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
510 struct vnode *vp;
511
512 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
513 vprint("", vp);
514 panic("unmount: dangling vnode");
515 }
516 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
517 if (mp->mnt_nvnodelistsize != 0)
518 panic("vfs_mount_destroy: nonzero nvnodelistsize");
519 if (mp->mnt_activevnodelistsize != 0)
520 panic("vfs_mount_destroy: nonzero activevnodelistsize");
521 if (mp->mnt_lockref != 0)
522 panic("vfs_mount_destroy: nonzero lock refcount");
523 MNT_IUNLOCK(mp);
524 #ifdef MAC
525 mac_mount_destroy(mp);
526 #endif
527 if (mp->mnt_opt != NULL)
528 vfs_freeopts(mp->mnt_opt);
529 crfree(mp->mnt_cred);
530 uma_zfree(mount_zone, mp);
531 }
532
533 int
534 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
535 {
536 struct vfsoptlist *optlist;
537 struct vfsopt *opt, *tmp_opt;
538 char *fstype, *fspath, *errmsg;
539 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
540
541 errmsg = fspath = NULL;
542 errmsg_len = fspathlen = 0;
543 errmsg_pos = -1;
544
545 error = vfs_buildopts(fsoptions, &optlist);
546 if (error)
547 return (error);
548
549 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
550 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
551
552 /*
553 * We need these two options before the others,
554 * and they are mandatory for any filesystem.
555 * Ensure they are NUL terminated as well.
556 */
557 fstypelen = 0;
558 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
559 if (error || fstype[fstypelen - 1] != '\0') {
560 error = EINVAL;
561 if (errmsg != NULL)
562 strncpy(errmsg, "Invalid fstype", errmsg_len);
563 goto bail;
564 }
565 fspathlen = 0;
566 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
567 if (error || fspath[fspathlen - 1] != '\0') {
568 error = EINVAL;
569 if (errmsg != NULL)
570 strncpy(errmsg, "Invalid fspath", errmsg_len);
571 goto bail;
572 }
573
574 /*
575 * We need to see if we have the "update" option
576 * before we call vfs_domount(), since vfs_domount() has special
577 * logic based on MNT_UPDATE. This is very important
578 * when we want to update the root filesystem.
579 */
580 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
581 if (strcmp(opt->name, "update") == 0) {
582 fsflags |= MNT_UPDATE;
583 vfs_freeopt(optlist, opt);
584 }
585 else if (strcmp(opt->name, "async") == 0)
586 fsflags |= MNT_ASYNC;
587 else if (strcmp(opt->name, "force") == 0) {
588 fsflags |= MNT_FORCE;
589 vfs_freeopt(optlist, opt);
590 }
591 else if (strcmp(opt->name, "reload") == 0) {
592 fsflags |= MNT_RELOAD;
593 vfs_freeopt(optlist, opt);
594 }
595 else if (strcmp(opt->name, "multilabel") == 0)
596 fsflags |= MNT_MULTILABEL;
597 else if (strcmp(opt->name, "noasync") == 0)
598 fsflags &= ~MNT_ASYNC;
599 else if (strcmp(opt->name, "noatime") == 0)
600 fsflags |= MNT_NOATIME;
601 else if (strcmp(opt->name, "atime") == 0) {
602 free(opt->name, M_MOUNT);
603 opt->name = strdup("nonoatime", M_MOUNT);
604 }
605 else if (strcmp(opt->name, "noclusterr") == 0)
606 fsflags |= MNT_NOCLUSTERR;
607 else if (strcmp(opt->name, "clusterr") == 0) {
608 free(opt->name, M_MOUNT);
609 opt->name = strdup("nonoclusterr", M_MOUNT);
610 }
611 else if (strcmp(opt->name, "noclusterw") == 0)
612 fsflags |= MNT_NOCLUSTERW;
613 else if (strcmp(opt->name, "clusterw") == 0) {
614 free(opt->name, M_MOUNT);
615 opt->name = strdup("nonoclusterw", M_MOUNT);
616 }
617 else if (strcmp(opt->name, "noexec") == 0)
618 fsflags |= MNT_NOEXEC;
619 else if (strcmp(opt->name, "exec") == 0) {
620 free(opt->name, M_MOUNT);
621 opt->name = strdup("nonoexec", M_MOUNT);
622 }
623 else if (strcmp(opt->name, "nosuid") == 0)
624 fsflags |= MNT_NOSUID;
625 else if (strcmp(opt->name, "suid") == 0) {
626 free(opt->name, M_MOUNT);
627 opt->name = strdup("nonosuid", M_MOUNT);
628 }
629 else if (strcmp(opt->name, "nosymfollow") == 0)
630 fsflags |= MNT_NOSYMFOLLOW;
631 else if (strcmp(opt->name, "symfollow") == 0) {
632 free(opt->name, M_MOUNT);
633 opt->name = strdup("nonosymfollow", M_MOUNT);
634 }
635 else if (strcmp(opt->name, "noro") == 0)
636 fsflags &= ~MNT_RDONLY;
637 else if (strcmp(opt->name, "rw") == 0)
638 fsflags &= ~MNT_RDONLY;
639 else if (strcmp(opt->name, "ro") == 0)
640 fsflags |= MNT_RDONLY;
641 else if (strcmp(opt->name, "rdonly") == 0) {
642 free(opt->name, M_MOUNT);
643 opt->name = strdup("ro", M_MOUNT);
644 fsflags |= MNT_RDONLY;
645 }
646 else if (strcmp(opt->name, "suiddir") == 0)
647 fsflags |= MNT_SUIDDIR;
648 else if (strcmp(opt->name, "sync") == 0)
649 fsflags |= MNT_SYNCHRONOUS;
650 else if (strcmp(opt->name, "union") == 0)
651 fsflags |= MNT_UNION;
652 else if (strcmp(opt->name, "automounted") == 0) {
653 fsflags |= MNT_AUTOMOUNTED;
654 vfs_freeopt(optlist, opt);
655 }
656 }
657
658 /*
659 * Be ultra-paranoid about making sure the type and fspath
660 * variables will fit in our mp buffers, including the
661 * terminating NUL.
662 */
663 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
664 error = ENAMETOOLONG;
665 goto bail;
666 }
667
668 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
669 bail:
670 /* copyout the errmsg */
671 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
672 && errmsg_len > 0 && errmsg != NULL) {
673 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
674 bcopy(errmsg,
675 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
676 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
677 } else {
678 copyout(errmsg,
679 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
680 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
681 }
682 }
683
684 if (optlist != NULL)
685 vfs_freeopts(optlist);
686 return (error);
687 }
688
689 /*
690 * Old mount API.
691 */
692 #ifndef _SYS_SYSPROTO_H_
693 struct mount_args {
694 char *type;
695 char *path;
696 int flags;
697 caddr_t data;
698 };
699 #endif
700 /* ARGSUSED */
701 int
702 sys_mount(td, uap)
703 struct thread *td;
704 struct mount_args /* {
705 char *type;
706 char *path;
707 int flags;
708 caddr_t data;
709 } */ *uap;
710 {
711 char *fstype;
712 struct vfsconf *vfsp = NULL;
713 struct mntarg *ma = NULL;
714 uint64_t flags;
715 int error;
716
717 /*
718 * Mount flags are now 64-bits. On 32-bit architectures only
719 * 32-bits are passed in, but from here on everything handles
720 * 64-bit flags correctly.
721 */
722 flags = uap->flags;
723
724 AUDIT_ARG_FFLAGS(flags);
725
726 /*
727 * Filter out MNT_ROOTFS. We do not want clients of mount() in
728 * userspace to set this flag, but we must filter it out if we want
729 * MNT_UPDATE on the root file system to work.
730 * MNT_ROOTFS should only be set by the kernel when mounting its
731 * root file system.
732 */
733 flags &= ~MNT_ROOTFS;
734
735 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
736 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
737 if (error) {
738 free(fstype, M_TEMP);
739 return (error);
740 }
741
742 AUDIT_ARG_TEXT(fstype);
743 vfsp = vfs_byname_kld(fstype, td, &error);
744 free(fstype, M_TEMP);
745 if (vfsp == NULL)
746 return (ENOENT);
747 if (vfsp->vfc_vfsops->vfs_cmount == NULL)
748 return (EOPNOTSUPP);
749
750 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
751 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
752 ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
753 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
754 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
755
756 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags);
757 return (error);
758 }
759
760 /*
761 * vfs_domount_first(): first file system mount (not update)
762 */
763 static int
764 vfs_domount_first(
765 struct thread *td, /* Calling thread. */
766 struct vfsconf *vfsp, /* File system type. */
767 char *fspath, /* Mount path. */
768 struct vnode *vp, /* Vnode to be covered. */
769 uint64_t fsflags, /* Flags common to all filesystems. */
770 struct vfsoptlist **optlist /* Options local to the filesystem. */
771 )
772 {
773 struct vattr va;
774 struct mount *mp;
775 struct vnode *newdp;
776 int error;
777
778 ASSERT_VOP_ELOCKED(vp, __func__);
779 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
780
781 /*
782 * If the user is not root, ensure that they own the directory
783 * onto which we are attempting to mount.
784 */
785 error = VOP_GETATTR(vp, &va, td->td_ucred);
786 if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
787 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0);
788 if (error == 0)
789 error = vinvalbuf(vp, V_SAVE, 0, 0);
790 if (error == 0 && vp->v_type != VDIR)
791 error = ENOTDIR;
792 if (error == 0) {
793 VI_LOCK(vp);
794 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
795 vp->v_iflag |= VI_MOUNT;
796 else
797 error = EBUSY;
798 VI_UNLOCK(vp);
799 }
800 if (error != 0) {
801 vput(vp);
802 return (error);
803 }
804 VOP_UNLOCK(vp, 0);
805
806 /* Allocate and initialize the filesystem. */
807 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
808 /* XXXMAC: pass to vfs_mount_alloc? */
809 mp->mnt_optnew = *optlist;
810 /* Set the mount level flags. */
811 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
812
813 /*
814 * Mount the filesystem.
815 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
816 * get. No freeing of cn_pnbuf.
817 */
818 error = VFS_MOUNT(mp);
819 if (error != 0) {
820 vfs_unbusy(mp);
821 vfs_mount_destroy(mp);
822 VI_LOCK(vp);
823 vp->v_iflag &= ~VI_MOUNT;
824 VI_UNLOCK(vp);
825 vrele(vp);
826 return (error);
827 }
828
829 if (mp->mnt_opt != NULL)
830 vfs_freeopts(mp->mnt_opt);
831 mp->mnt_opt = mp->mnt_optnew;
832 *optlist = NULL;
833 (void)VFS_STATFS(mp, &mp->mnt_stat);
834
835 /*
836 * Prevent external consumers of mount options from reading mnt_optnew.
837 */
838 mp->mnt_optnew = NULL;
839
840 MNT_ILOCK(mp);
841 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
842 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
843 mp->mnt_kern_flag |= MNTK_ASYNC;
844 else
845 mp->mnt_kern_flag &= ~MNTK_ASYNC;
846 MNT_IUNLOCK(mp);
847
848 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
849 cache_purge(vp);
850 VI_LOCK(vp);
851 vp->v_iflag &= ~VI_MOUNT;
852 VI_UNLOCK(vp);
853 vp->v_mountedhere = mp;
854 /* Place the new filesystem at the end of the mount list. */
855 mtx_lock(&mountlist_mtx);
856 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
857 mtx_unlock(&mountlist_mtx);
858 vfs_event_signal(NULL, VQ_MOUNT, 0);
859 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp))
860 panic("mount: lost mount");
861 VOP_UNLOCK(vp, 0);
862 EVENTHANDLER_INVOKE(vfs_mounted, mp, newdp, td);
863 VOP_UNLOCK(newdp, 0);
864 mountcheckdirs(vp, newdp);
865 vrele(newdp);
866 if ((mp->mnt_flag & MNT_RDONLY) == 0)
867 vfs_allocate_syncvnode(mp);
868 vfs_unbusy(mp);
869 return (0);
870 }
871
872 /*
873 * vfs_domount_update(): update of mounted file system
874 */
875 static int
876 vfs_domount_update(
877 struct thread *td, /* Calling thread. */
878 struct vnode *vp, /* Mount point vnode. */
879 uint64_t fsflags, /* Flags common to all filesystems. */
880 struct vfsoptlist **optlist /* Options local to the filesystem. */
881 )
882 {
883 struct oexport_args oexport;
884 struct export_args export;
885 struct mount *mp;
886 int error, export_error;
887 uint64_t flag;
888
889 ASSERT_VOP_ELOCKED(vp, __func__);
890 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
891 mp = vp->v_mount;
892
893 if ((vp->v_vflag & VV_ROOT) == 0) {
894 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
895 == 0)
896 error = EXDEV;
897 else
898 error = EINVAL;
899 vput(vp);
900 return (error);
901 }
902
903 /*
904 * We only allow the filesystem to be reloaded if it
905 * is currently mounted read-only.
906 */
907 flag = mp->mnt_flag;
908 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
909 vput(vp);
910 return (EOPNOTSUPP); /* Needs translation */
911 }
912 /*
913 * Only privileged root, or (if MNT_USER is set) the user that
914 * did the original mount is permitted to update it.
915 */
916 error = vfs_suser(mp, td);
917 if (error != 0) {
918 vput(vp);
919 return (error);
920 }
921 if (vfs_busy(mp, MBF_NOWAIT)) {
922 vput(vp);
923 return (EBUSY);
924 }
925 VI_LOCK(vp);
926 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
927 VI_UNLOCK(vp);
928 vfs_unbusy(mp);
929 vput(vp);
930 return (EBUSY);
931 }
932 vp->v_iflag |= VI_MOUNT;
933 VI_UNLOCK(vp);
934 VOP_UNLOCK(vp, 0);
935
936 MNT_ILOCK(mp);
937 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
938 MNT_IUNLOCK(mp);
939 error = EBUSY;
940 goto end;
941 }
942 mp->mnt_flag &= ~MNT_UPDATEMASK;
943 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
944 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
945 if ((mp->mnt_flag & MNT_ASYNC) == 0)
946 mp->mnt_kern_flag &= ~MNTK_ASYNC;
947 MNT_IUNLOCK(mp);
948 mp->mnt_optnew = *optlist;
949 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
950
951 /*
952 * Mount the filesystem.
953 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
954 * get. No freeing of cn_pnbuf.
955 */
956 error = VFS_MOUNT(mp);
957
958 export_error = 0;
959 if (error == 0) {
960 /* Process the export option. */
961 if (vfs_copyopt(mp->mnt_optnew, "export", &export,
962 sizeof(export)) == 0) {
963 export_error = vfs_export(mp, &export);
964 } else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport,
965 sizeof(oexport)) == 0) {
966 export.ex_flags = oexport.ex_flags;
967 export.ex_root = oexport.ex_root;
968 export.ex_anon = oexport.ex_anon;
969 export.ex_addr = oexport.ex_addr;
970 export.ex_addrlen = oexport.ex_addrlen;
971 export.ex_mask = oexport.ex_mask;
972 export.ex_masklen = oexport.ex_masklen;
973 export.ex_indexfile = oexport.ex_indexfile;
974 export.ex_numsecflavors = 0;
975 export_error = vfs_export(mp, &export);
976 }
977 }
978
979 MNT_ILOCK(mp);
980 if (error == 0) {
981 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
982 MNT_SNAPSHOT);
983 } else {
984 /*
985 * If we fail, restore old mount flags. MNT_QUOTA is special,
986 * because it is not part of MNT_UPDATEMASK, but it could have
987 * changed in the meantime if quotactl(2) was called.
988 * All in all we want current value of MNT_QUOTA, not the old
989 * one.
990 */
991 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
992 }
993 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
994 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
995 mp->mnt_kern_flag |= MNTK_ASYNC;
996 else
997 mp->mnt_kern_flag &= ~MNTK_ASYNC;
998 MNT_IUNLOCK(mp);
999
1000 if (error != 0)
1001 goto end;
1002
1003 if (mp->mnt_opt != NULL)
1004 vfs_freeopts(mp->mnt_opt);
1005 mp->mnt_opt = mp->mnt_optnew;
1006 *optlist = NULL;
1007 (void)VFS_STATFS(mp, &mp->mnt_stat);
1008 /*
1009 * Prevent external consumers of mount options from reading
1010 * mnt_optnew.
1011 */
1012 mp->mnt_optnew = NULL;
1013
1014 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1015 vfs_allocate_syncvnode(mp);
1016 else
1017 vfs_deallocate_syncvnode(mp);
1018 end:
1019 vfs_unbusy(mp);
1020 VI_LOCK(vp);
1021 vp->v_iflag &= ~VI_MOUNT;
1022 VI_UNLOCK(vp);
1023 vrele(vp);
1024 return (error != 0 ? error : export_error);
1025 }
1026
1027 /*
1028 * vfs_domount(): actually attempt a filesystem mount.
1029 */
1030 static int
1031 vfs_domount(
1032 struct thread *td, /* Calling thread. */
1033 const char *fstype, /* Filesystem type. */
1034 char *fspath, /* Mount path. */
1035 uint64_t fsflags, /* Flags common to all filesystems. */
1036 struct vfsoptlist **optlist /* Options local to the filesystem. */
1037 )
1038 {
1039 struct vfsconf *vfsp;
1040 struct nameidata nd;
1041 struct vnode *vp;
1042 char *pathbuf;
1043 int error;
1044
1045 /*
1046 * Be ultra-paranoid about making sure the type and fspath
1047 * variables will fit in our mp buffers, including the
1048 * terminating NUL.
1049 */
1050 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1051 return (ENAMETOOLONG);
1052
1053 if (jailed(td->td_ucred) || usermount == 0) {
1054 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1055 return (error);
1056 }
1057
1058 /*
1059 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1060 */
1061 if (fsflags & MNT_EXPORTED) {
1062 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1063 if (error)
1064 return (error);
1065 }
1066 if (fsflags & MNT_SUIDDIR) {
1067 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1068 if (error)
1069 return (error);
1070 }
1071 /*
1072 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1073 */
1074 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1075 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1076 fsflags |= MNT_NOSUID | MNT_USER;
1077 }
1078
1079 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1080 vfsp = NULL;
1081 if ((fsflags & MNT_UPDATE) == 0) {
1082 /* Don't try to load KLDs if we're mounting the root. */
1083 if (fsflags & MNT_ROOTFS)
1084 vfsp = vfs_byname(fstype);
1085 else
1086 vfsp = vfs_byname_kld(fstype, td, &error);
1087 if (vfsp == NULL)
1088 return (ENODEV);
1089 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
1090 return (EPERM);
1091 }
1092
1093 /*
1094 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1095 */
1096 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1097 UIO_SYSSPACE, fspath, td);
1098 error = namei(&nd);
1099 if (error != 0)
1100 return (error);
1101 NDFREE(&nd, NDF_ONLY_PNBUF);
1102 vp = nd.ni_vp;
1103 if ((fsflags & MNT_UPDATE) == 0) {
1104 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1105 strcpy(pathbuf, fspath);
1106 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1107 /* debug.disablefullpath == 1 results in ENODEV */
1108 if (error == 0 || error == ENODEV) {
1109 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1110 fsflags, optlist);
1111 }
1112 free(pathbuf, M_TEMP);
1113 } else
1114 error = vfs_domount_update(td, vp, fsflags, optlist);
1115
1116 return (error);
1117 }
1118
1119 /*
1120 * Unmount a filesystem.
1121 *
1122 * Note: unmount takes a path to the vnode mounted on as argument, not
1123 * special file (as before).
1124 */
1125 #ifndef _SYS_SYSPROTO_H_
1126 struct unmount_args {
1127 char *path;
1128 int flags;
1129 };
1130 #endif
1131 /* ARGSUSED */
1132 int
1133 sys_unmount(struct thread *td, struct unmount_args *uap)
1134 {
1135 struct nameidata nd;
1136 struct mount *mp;
1137 char *pathbuf;
1138 int error, id0, id1;
1139
1140 AUDIT_ARG_VALUE(uap->flags);
1141 if (jailed(td->td_ucred) || usermount == 0) {
1142 error = priv_check(td, PRIV_VFS_UNMOUNT);
1143 if (error)
1144 return (error);
1145 }
1146
1147 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1148 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1149 if (error) {
1150 free(pathbuf, M_TEMP);
1151 return (error);
1152 }
1153 if (uap->flags & MNT_BYFSID) {
1154 AUDIT_ARG_TEXT(pathbuf);
1155 /* Decode the filesystem ID. */
1156 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1157 free(pathbuf, M_TEMP);
1158 return (EINVAL);
1159 }
1160
1161 mtx_lock(&mountlist_mtx);
1162 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1163 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1164 mp->mnt_stat.f_fsid.val[1] == id1) {
1165 vfs_ref(mp);
1166 break;
1167 }
1168 }
1169 mtx_unlock(&mountlist_mtx);
1170 } else {
1171 /*
1172 * Try to find global path for path argument.
1173 */
1174 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1175 UIO_SYSSPACE, pathbuf, td);
1176 if (namei(&nd) == 0) {
1177 NDFREE(&nd, NDF_ONLY_PNBUF);
1178 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1179 MNAMELEN);
1180 if (error == 0 || error == ENODEV)
1181 vput(nd.ni_vp);
1182 }
1183 mtx_lock(&mountlist_mtx);
1184 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1185 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1186 vfs_ref(mp);
1187 break;
1188 }
1189 }
1190 mtx_unlock(&mountlist_mtx);
1191 }
1192 free(pathbuf, M_TEMP);
1193 if (mp == NULL) {
1194 /*
1195 * Previously we returned ENOENT for a nonexistent path and
1196 * EINVAL for a non-mountpoint. We cannot tell these apart
1197 * now, so in the !MNT_BYFSID case return the more likely
1198 * EINVAL for compatibility.
1199 */
1200 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1201 }
1202
1203 /*
1204 * Don't allow unmounting the root filesystem.
1205 */
1206 if (mp->mnt_flag & MNT_ROOTFS) {
1207 vfs_rel(mp);
1208 return (EINVAL);
1209 }
1210 error = dounmount(mp, uap->flags, td);
1211 return (error);
1212 }
1213
1214 /*
1215 * Do the actual filesystem unmount.
1216 */
1217 int
1218 dounmount(struct mount *mp, int flags, struct thread *td)
1219 {
1220 struct vnode *coveredvp, *fsrootvp;
1221 int error;
1222 uint64_t async_flag;
1223 int mnt_gen_r;
1224
1225 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1226 mnt_gen_r = mp->mnt_gen;
1227 VI_LOCK(coveredvp);
1228 vholdl(coveredvp);
1229 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1230 /*
1231 * Check for mp being unmounted while waiting for the
1232 * covered vnode lock.
1233 */
1234 if (coveredvp->v_mountedhere != mp ||
1235 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1236 VOP_UNLOCK(coveredvp, 0);
1237 vdrop(coveredvp);
1238 vfs_rel(mp);
1239 return (EBUSY);
1240 }
1241 }
1242
1243 /*
1244 * Only privileged root, or (if MNT_USER is set) the user that did the
1245 * original mount is permitted to unmount this filesystem.
1246 */
1247 error = vfs_suser(mp, td);
1248 if (error != 0) {
1249 if (coveredvp != NULL) {
1250 VOP_UNLOCK(coveredvp, 0);
1251 vdrop(coveredvp);
1252 }
1253 vfs_rel(mp);
1254 return (error);
1255 }
1256
1257 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1258 MNT_ILOCK(mp);
1259 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1260 (mp->mnt_flag & MNT_UPDATE) != 0 ||
1261 !TAILQ_EMPTY(&mp->mnt_uppers)) {
1262 MNT_IUNLOCK(mp);
1263 if (coveredvp != NULL) {
1264 VOP_UNLOCK(coveredvp, 0);
1265 vdrop(coveredvp);
1266 }
1267 vn_finished_write(mp);
1268 return (EBUSY);
1269 }
1270 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1271 /* Allow filesystems to detect that a forced unmount is in progress. */
1272 if (flags & MNT_FORCE) {
1273 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1274 MNT_IUNLOCK(mp);
1275 /*
1276 * Must be done after setting MNTK_UNMOUNTF and before
1277 * waiting for mnt_lockref to become 0.
1278 */
1279 VFS_PURGE(mp);
1280 MNT_ILOCK(mp);
1281 }
1282 error = 0;
1283 if (mp->mnt_lockref) {
1284 mp->mnt_kern_flag |= MNTK_DRAINING;
1285 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1286 "mount drain", 0);
1287 }
1288 MNT_IUNLOCK(mp);
1289 KASSERT(mp->mnt_lockref == 0,
1290 ("%s: invalid lock refcount in the drain path @ %s:%d",
1291 __func__, __FILE__, __LINE__));
1292 KASSERT(error == 0,
1293 ("%s: invalid return value for msleep in the drain path @ %s:%d",
1294 __func__, __FILE__, __LINE__));
1295
1296 if (mp->mnt_flag & MNT_EXPUBLIC)
1297 vfs_setpublicfs(NULL, NULL, NULL);
1298
1299 /*
1300 * From now, we can claim that the use reference on the
1301 * coveredvp is ours, and the ref can be released only by
1302 * successfull unmount by us, or left for later unmount
1303 * attempt. The previously acquired hold reference is no
1304 * longer needed to protect the vnode from reuse.
1305 */
1306 if (coveredvp != NULL)
1307 vdrop(coveredvp);
1308
1309 vfs_msync(mp, MNT_WAIT);
1310 MNT_ILOCK(mp);
1311 async_flag = mp->mnt_flag & MNT_ASYNC;
1312 mp->mnt_flag &= ~MNT_ASYNC;
1313 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1314 MNT_IUNLOCK(mp);
1315 cache_purgevfs(mp); /* remove cache entries for this file sys */
1316 vfs_deallocate_syncvnode(mp);
1317 /*
1318 * For forced unmounts, move process cdir/rdir refs on the fs root
1319 * vnode to the covered vnode. For non-forced unmounts we want
1320 * such references to cause an EBUSY error.
1321 */
1322 if ((flags & MNT_FORCE) &&
1323 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1324 if (mp->mnt_vnodecovered != NULL &&
1325 (mp->mnt_flag & MNT_IGNORE) == 0)
1326 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1327 if (fsrootvp == rootvnode) {
1328 vrele(rootvnode);
1329 rootvnode = NULL;
1330 }
1331 vput(fsrootvp);
1332 }
1333 if ((mp->mnt_flag & MNT_RDONLY) != 0 || (flags & MNT_FORCE) != 0 ||
1334 (error = VFS_SYNC(mp, MNT_WAIT)) == 0)
1335 error = VFS_UNMOUNT(mp, flags);
1336 vn_finished_write(mp);
1337 /*
1338 * If we failed to flush the dirty blocks for this mount point,
1339 * undo all the cdir/rdir and rootvnode changes we made above.
1340 * Unless we failed to do so because the device is reporting that
1341 * it doesn't exist anymore.
1342 */
1343 if (error && error != ENXIO) {
1344 if ((flags & MNT_FORCE) &&
1345 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1346 if (mp->mnt_vnodecovered != NULL &&
1347 (mp->mnt_flag & MNT_IGNORE) == 0)
1348 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1349 if (rootvnode == NULL) {
1350 rootvnode = fsrootvp;
1351 vref(rootvnode);
1352 }
1353 vput(fsrootvp);
1354 }
1355 MNT_ILOCK(mp);
1356 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1357 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1358 MNT_IUNLOCK(mp);
1359 vfs_allocate_syncvnode(mp);
1360 MNT_ILOCK(mp);
1361 }
1362 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1363 mp->mnt_flag |= async_flag;
1364 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1365 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1366 mp->mnt_kern_flag |= MNTK_ASYNC;
1367 if (mp->mnt_kern_flag & MNTK_MWAIT) {
1368 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1369 wakeup(mp);
1370 }
1371 MNT_IUNLOCK(mp);
1372 if (coveredvp)
1373 VOP_UNLOCK(coveredvp, 0);
1374 return (error);
1375 }
1376 mtx_lock(&mountlist_mtx);
1377 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1378 mtx_unlock(&mountlist_mtx);
1379 EVENTHANDLER_INVOKE(vfs_unmounted, mp, td);
1380 if (coveredvp != NULL) {
1381 coveredvp->v_mountedhere = NULL;
1382 vput(coveredvp);
1383 }
1384 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1385 if (mp == rootdevmp)
1386 rootdevmp = NULL;
1387 vfs_mount_destroy(mp);
1388 return (0);
1389 }
1390
1391 /*
1392 * Report errors during filesystem mounting.
1393 */
1394 void
1395 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1396 {
1397 struct vfsoptlist *moptlist = mp->mnt_optnew;
1398 va_list ap;
1399 int error, len;
1400 char *errmsg;
1401
1402 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1403 if (error || errmsg == NULL || len <= 0)
1404 return;
1405
1406 va_start(ap, fmt);
1407 vsnprintf(errmsg, (size_t)len, fmt, ap);
1408 va_end(ap);
1409 }
1410
1411 void
1412 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1413 {
1414 va_list ap;
1415 int error, len;
1416 char *errmsg;
1417
1418 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1419 if (error || errmsg == NULL || len <= 0)
1420 return;
1421
1422 va_start(ap, fmt);
1423 vsnprintf(errmsg, (size_t)len, fmt, ap);
1424 va_end(ap);
1425 }
1426
1427 /*
1428 * ---------------------------------------------------------------------
1429 * Functions for querying mount options/arguments from filesystems.
1430 */
1431
1432 /*
1433 * Check that no unknown options are given
1434 */
1435 int
1436 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1437 {
1438 struct vfsopt *opt;
1439 char errmsg[255];
1440 const char **t, *p, *q;
1441 int ret = 0;
1442
1443 TAILQ_FOREACH(opt, opts, link) {
1444 p = opt->name;
1445 q = NULL;
1446 if (p[0] == 'n' && p[1] == 'o')
1447 q = p + 2;
1448 for(t = global_opts; *t != NULL; t++) {
1449 if (strcmp(*t, p) == 0)
1450 break;
1451 if (q != NULL) {
1452 if (strcmp(*t, q) == 0)
1453 break;
1454 }
1455 }
1456 if (*t != NULL)
1457 continue;
1458 for(t = legal; *t != NULL; t++) {
1459 if (strcmp(*t, p) == 0)
1460 break;
1461 if (q != NULL) {
1462 if (strcmp(*t, q) == 0)
1463 break;
1464 }
1465 }
1466 if (*t != NULL)
1467 continue;
1468 snprintf(errmsg, sizeof(errmsg),
1469 "mount option <%s> is unknown", p);
1470 ret = EINVAL;
1471 }
1472 if (ret != 0) {
1473 TAILQ_FOREACH(opt, opts, link) {
1474 if (strcmp(opt->name, "errmsg") == 0) {
1475 strncpy((char *)opt->value, errmsg, opt->len);
1476 break;
1477 }
1478 }
1479 if (opt == NULL)
1480 printf("%s\n", errmsg);
1481 }
1482 return (ret);
1483 }
1484
1485 /*
1486 * Get a mount option by its name.
1487 *
1488 * Return 0 if the option was found, ENOENT otherwise.
1489 * If len is non-NULL it will be filled with the length
1490 * of the option. If buf is non-NULL, it will be filled
1491 * with the address of the option.
1492 */
1493 int
1494 vfs_getopt(opts, name, buf, len)
1495 struct vfsoptlist *opts;
1496 const char *name;
1497 void **buf;
1498 int *len;
1499 {
1500 struct vfsopt *opt;
1501
1502 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1503
1504 TAILQ_FOREACH(opt, opts, link) {
1505 if (strcmp(name, opt->name) == 0) {
1506 opt->seen = 1;
1507 if (len != NULL)
1508 *len = opt->len;
1509 if (buf != NULL)
1510 *buf = opt->value;
1511 return (0);
1512 }
1513 }
1514 return (ENOENT);
1515 }
1516
1517 int
1518 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1519 {
1520 struct vfsopt *opt;
1521
1522 if (opts == NULL)
1523 return (-1);
1524
1525 TAILQ_FOREACH(opt, opts, link) {
1526 if (strcmp(name, opt->name) == 0) {
1527 opt->seen = 1;
1528 return (opt->pos);
1529 }
1530 }
1531 return (-1);
1532 }
1533
1534 int
1535 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
1536 {
1537 char *opt_value, *vtp;
1538 quad_t iv;
1539 int error, opt_len;
1540
1541 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
1542 if (error != 0)
1543 return (error);
1544 if (opt_len == 0 || opt_value == NULL)
1545 return (EINVAL);
1546 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
1547 return (EINVAL);
1548 iv = strtoq(opt_value, &vtp, 0);
1549 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
1550 return (EINVAL);
1551 if (iv < 0)
1552 return (EINVAL);
1553 switch (vtp[0]) {
1554 case 't':
1555 case 'T':
1556 iv *= 1024;
1557 case 'g':
1558 case 'G':
1559 iv *= 1024;
1560 case 'm':
1561 case 'M':
1562 iv *= 1024;
1563 case 'k':
1564 case 'K':
1565 iv *= 1024;
1566 case '\0':
1567 break;
1568 default:
1569 return (EINVAL);
1570 }
1571 *value = iv;
1572
1573 return (0);
1574 }
1575
1576 char *
1577 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1578 {
1579 struct vfsopt *opt;
1580
1581 *error = 0;
1582 TAILQ_FOREACH(opt, opts, link) {
1583 if (strcmp(name, opt->name) != 0)
1584 continue;
1585 opt->seen = 1;
1586 if (opt->len == 0 ||
1587 ((char *)opt->value)[opt->len - 1] != '\0') {
1588 *error = EINVAL;
1589 return (NULL);
1590 }
1591 return (opt->value);
1592 }
1593 *error = ENOENT;
1594 return (NULL);
1595 }
1596
1597 int
1598 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
1599 uint64_t val)
1600 {
1601 struct vfsopt *opt;
1602
1603 TAILQ_FOREACH(opt, opts, link) {
1604 if (strcmp(name, opt->name) == 0) {
1605 opt->seen = 1;
1606 if (w != NULL)
1607 *w |= val;
1608 return (1);
1609 }
1610 }
1611 if (w != NULL)
1612 *w &= ~val;
1613 return (0);
1614 }
1615
1616 int
1617 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1618 {
1619 va_list ap;
1620 struct vfsopt *opt;
1621 int ret;
1622
1623 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1624
1625 TAILQ_FOREACH(opt, opts, link) {
1626 if (strcmp(name, opt->name) != 0)
1627 continue;
1628 opt->seen = 1;
1629 if (opt->len == 0 || opt->value == NULL)
1630 return (0);
1631 if (((char *)opt->value)[opt->len - 1] != '\0')
1632 return (0);
1633 va_start(ap, fmt);
1634 ret = vsscanf(opt->value, fmt, ap);
1635 va_end(ap);
1636 return (ret);
1637 }
1638 return (0);
1639 }
1640
1641 int
1642 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
1643 {
1644 struct vfsopt *opt;
1645
1646 TAILQ_FOREACH(opt, opts, link) {
1647 if (strcmp(name, opt->name) != 0)
1648 continue;
1649 opt->seen = 1;
1650 if (opt->value == NULL)
1651 opt->len = len;
1652 else {
1653 if (opt->len != len)
1654 return (EINVAL);
1655 bcopy(value, opt->value, len);
1656 }
1657 return (0);
1658 }
1659 return (ENOENT);
1660 }
1661
1662 int
1663 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
1664 {
1665 struct vfsopt *opt;
1666
1667 TAILQ_FOREACH(opt, opts, link) {
1668 if (strcmp(name, opt->name) != 0)
1669 continue;
1670 opt->seen = 1;
1671 if (opt->value == NULL)
1672 opt->len = len;
1673 else {
1674 if (opt->len < len)
1675 return (EINVAL);
1676 opt->len = len;
1677 bcopy(value, opt->value, len);
1678 }
1679 return (0);
1680 }
1681 return (ENOENT);
1682 }
1683
1684 int
1685 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
1686 {
1687 struct vfsopt *opt;
1688
1689 TAILQ_FOREACH(opt, opts, link) {
1690 if (strcmp(name, opt->name) != 0)
1691 continue;
1692 opt->seen = 1;
1693 if (opt->value == NULL)
1694 opt->len = strlen(value) + 1;
1695 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
1696 return (EINVAL);
1697 return (0);
1698 }
1699 return (ENOENT);
1700 }
1701
1702 /*
1703 * Find and copy a mount option.
1704 *
1705 * The size of the buffer has to be specified
1706 * in len, if it is not the same length as the
1707 * mount option, EINVAL is returned.
1708 * Returns ENOENT if the option is not found.
1709 */
1710 int
1711 vfs_copyopt(opts, name, dest, len)
1712 struct vfsoptlist *opts;
1713 const char *name;
1714 void *dest;
1715 int len;
1716 {
1717 struct vfsopt *opt;
1718
1719 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1720
1721 TAILQ_FOREACH(opt, opts, link) {
1722 if (strcmp(name, opt->name) == 0) {
1723 opt->seen = 1;
1724 if (len != opt->len)
1725 return (EINVAL);
1726 bcopy(opt->value, dest, opt->len);
1727 return (0);
1728 }
1729 }
1730 return (ENOENT);
1731 }
1732
1733 int
1734 __vfs_statfs(struct mount *mp, struct statfs *sbp)
1735 {
1736 int error;
1737
1738 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat);
1739 if (sbp != &mp->mnt_stat)
1740 *sbp = mp->mnt_stat;
1741 return (error);
1742 }
1743
1744 void
1745 vfs_mountedfrom(struct mount *mp, const char *from)
1746 {
1747
1748 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1749 strlcpy(mp->mnt_stat.f_mntfromname, from,
1750 sizeof mp->mnt_stat.f_mntfromname);
1751 }
1752
1753 /*
1754 * ---------------------------------------------------------------------
1755 * This is the api for building mount args and mounting filesystems from
1756 * inside the kernel.
1757 *
1758 * The API works by accumulation of individual args. First error is
1759 * latched.
1760 *
1761 * XXX: should be documented in new manpage kernel_mount(9)
1762 */
1763
1764 /* A memory allocation which must be freed when we are done */
1765 struct mntaarg {
1766 SLIST_ENTRY(mntaarg) next;
1767 };
1768
1769 /* The header for the mount arguments */
1770 struct mntarg {
1771 struct iovec *v;
1772 int len;
1773 int error;
1774 SLIST_HEAD(, mntaarg) list;
1775 };
1776
1777 /*
1778 * Add a boolean argument.
1779 *
1780 * flag is the boolean value.
1781 * name must start with "no".
1782 */
1783 struct mntarg *
1784 mount_argb(struct mntarg *ma, int flag, const char *name)
1785 {
1786
1787 KASSERT(name[0] == 'n' && name[1] == 'o',
1788 ("mount_argb(...,%s): name must start with 'no'", name));
1789
1790 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1791 }
1792
1793 /*
1794 * Add an argument printf style
1795 */
1796 struct mntarg *
1797 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1798 {
1799 va_list ap;
1800 struct mntaarg *maa;
1801 struct sbuf *sb;
1802 int len;
1803
1804 if (ma == NULL) {
1805 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1806 SLIST_INIT(&ma->list);
1807 }
1808 if (ma->error)
1809 return (ma);
1810
1811 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1812 M_MOUNT, M_WAITOK);
1813 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1814 ma->v[ma->len].iov_len = strlen(name) + 1;
1815 ma->len++;
1816
1817 sb = sbuf_new_auto();
1818 va_start(ap, fmt);
1819 sbuf_vprintf(sb, fmt, ap);
1820 va_end(ap);
1821 sbuf_finish(sb);
1822 len = sbuf_len(sb) + 1;
1823 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1824 SLIST_INSERT_HEAD(&ma->list, maa, next);
1825 bcopy(sbuf_data(sb), maa + 1, len);
1826 sbuf_delete(sb);
1827
1828 ma->v[ma->len].iov_base = maa + 1;
1829 ma->v[ma->len].iov_len = len;
1830 ma->len++;
1831
1832 return (ma);
1833 }
1834
1835 /*
1836 * Add an argument which is a userland string.
1837 */
1838 struct mntarg *
1839 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
1840 {
1841 struct mntaarg *maa;
1842 char *tbuf;
1843
1844 if (val == NULL)
1845 return (ma);
1846 if (ma == NULL) {
1847 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1848 SLIST_INIT(&ma->list);
1849 }
1850 if (ma->error)
1851 return (ma);
1852 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1853 SLIST_INSERT_HEAD(&ma->list, maa, next);
1854 tbuf = (void *)(maa + 1);
1855 ma->error = copyinstr(val, tbuf, len, NULL);
1856 return (mount_arg(ma, name, tbuf, -1));
1857 }
1858
1859 /*
1860 * Plain argument.
1861 *
1862 * If length is -1, treat value as a C string.
1863 */
1864 struct mntarg *
1865 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
1866 {
1867
1868 if (ma == NULL) {
1869 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1870 SLIST_INIT(&ma->list);
1871 }
1872 if (ma->error)
1873 return (ma);
1874
1875 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1876 M_MOUNT, M_WAITOK);
1877 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1878 ma->v[ma->len].iov_len = strlen(name) + 1;
1879 ma->len++;
1880
1881 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
1882 if (len < 0)
1883 ma->v[ma->len].iov_len = strlen(val) + 1;
1884 else
1885 ma->v[ma->len].iov_len = len;
1886 ma->len++;
1887 return (ma);
1888 }
1889
1890 /*
1891 * Free a mntarg structure
1892 */
1893 static void
1894 free_mntarg(struct mntarg *ma)
1895 {
1896 struct mntaarg *maa;
1897
1898 while (!SLIST_EMPTY(&ma->list)) {
1899 maa = SLIST_FIRST(&ma->list);
1900 SLIST_REMOVE_HEAD(&ma->list, next);
1901 free(maa, M_MOUNT);
1902 }
1903 free(ma->v, M_MOUNT);
1904 free(ma, M_MOUNT);
1905 }
1906
1907 /*
1908 * Mount a filesystem
1909 */
1910 int
1911 kernel_mount(struct mntarg *ma, uint64_t flags)
1912 {
1913 struct uio auio;
1914 int error;
1915
1916 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
1917 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
1918 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
1919
1920 auio.uio_iov = ma->v;
1921 auio.uio_iovcnt = ma->len;
1922 auio.uio_segflg = UIO_SYSSPACE;
1923
1924 error = ma->error;
1925 if (!error)
1926 error = vfs_donmount(curthread, flags, &auio);
1927 free_mntarg(ma);
1928 return (error);
1929 }
1930
1931 /*
1932 * A printflike function to mount a filesystem.
1933 */
1934 int
1935 kernel_vmount(int flags, ...)
1936 {
1937 struct mntarg *ma = NULL;
1938 va_list ap;
1939 const char *cp;
1940 const void *vp;
1941 int error;
1942
1943 va_start(ap, flags);
1944 for (;;) {
1945 cp = va_arg(ap, const char *);
1946 if (cp == NULL)
1947 break;
1948 vp = va_arg(ap, const void *);
1949 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
1950 }
1951 va_end(ap);
1952
1953 error = kernel_mount(ma, flags);
1954 return (error);
1955 }
1956
1957 void
1958 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp)
1959 {
1960
1961 bcopy(oexp, exp, sizeof(*oexp));
1962 exp->ex_numsecflavors = 0;
1963 }
Cache object: dcbbaca643c336b1233d16829f822b06
|