FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_mount.c
1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1999-2004 Poul-Henning Kamp
5 * Copyright (c) 1999 Michael Smith
6 * Copyright (c) 1989, 1993
7 * The Regents of the University of California. All rights reserved.
8 * (c) UNIX System Laboratories, Inc.
9 * All or some portions of this file are derived from material licensed
10 * to the University of California by American Telephone and Telegraph
11 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12 * the permission of UNIX System Laboratories, Inc.
13 *
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
16 * are met:
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in the
21 * documentation and/or other materials provided with the distribution.
22 * 3. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 */
38
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41
42 #include <sys/param.h>
43 #include <sys/conf.h>
44 #include <sys/eventhandler.h>
45 #include <sys/fcntl.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/libkern.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
52 #include <sys/namei.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/filedesc.h>
56 #include <sys/reboot.h>
57 #include <sys/sbuf.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/sysproto.h>
60 #include <sys/sx.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysent.h>
63 #include <sys/systm.h>
64 #include <sys/vnode.h>
65 #include <vm/uma.h>
66
67 #include <geom/geom.h>
68
69 #include <machine/stdarg.h>
70
71 #include <security/audit/audit.h>
72 #include <security/mac/mac_framework.h>
73
74 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
75
76 static int vfs_domount(struct thread *td, const char *fstype, char *fspath,
77 uint64_t fsflags, struct vfsoptlist **optlist);
78 static void free_mntarg(struct mntarg *ma);
79
80 static int usermount = 0;
81 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
82 "Unprivileged users may mount and unmount file systems");
83
84 static bool default_autoro = false;
85 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
86 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
87
88 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
89 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
90 static uma_zone_t mount_zone;
91
92 /* List of mounted filesystems. */
93 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
94
95 /* For any iteration/modification of mountlist */
96 struct mtx mountlist_mtx;
97 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
98
99 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
100 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
101
102 /*
103 * Global opts, taken by all filesystems
104 */
105 static const char *global_opts[] = {
106 "errmsg",
107 "fstype",
108 "fspath",
109 "ro",
110 "rw",
111 "nosuid",
112 "noexec",
113 NULL
114 };
115
116 static int
117 mount_init(void *mem, int size, int flags)
118 {
119 struct mount *mp;
120
121 mp = (struct mount *)mem;
122 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
123 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
124 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
125 return (0);
126 }
127
128 static void
129 mount_fini(void *mem, int size)
130 {
131 struct mount *mp;
132
133 mp = (struct mount *)mem;
134 lockdestroy(&mp->mnt_explock);
135 mtx_destroy(&mp->mnt_listmtx);
136 mtx_destroy(&mp->mnt_mtx);
137 }
138
139 static void
140 vfs_mount_init(void *dummy __unused)
141 {
142
143 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
144 NULL, mount_init, mount_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
145 }
146 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
147
148 /*
149 * ---------------------------------------------------------------------
150 * Functions for building and sanitizing the mount options
151 */
152
153 /* Remove one mount option. */
154 static void
155 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
156 {
157
158 TAILQ_REMOVE(opts, opt, link);
159 free(opt->name, M_MOUNT);
160 if (opt->value != NULL)
161 free(opt->value, M_MOUNT);
162 free(opt, M_MOUNT);
163 }
164
165 /* Release all resources related to the mount options. */
166 void
167 vfs_freeopts(struct vfsoptlist *opts)
168 {
169 struct vfsopt *opt;
170
171 while (!TAILQ_EMPTY(opts)) {
172 opt = TAILQ_FIRST(opts);
173 vfs_freeopt(opts, opt);
174 }
175 free(opts, M_MOUNT);
176 }
177
178 void
179 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
180 {
181 struct vfsopt *opt, *temp;
182
183 if (opts == NULL)
184 return;
185 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
186 if (strcmp(opt->name, name) == 0)
187 vfs_freeopt(opts, opt);
188 }
189 }
190
191 static int
192 vfs_isopt_ro(const char *opt)
193 {
194
195 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
196 strcmp(opt, "norw") == 0)
197 return (1);
198 return (0);
199 }
200
201 static int
202 vfs_isopt_rw(const char *opt)
203 {
204
205 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
206 return (1);
207 return (0);
208 }
209
210 /*
211 * Check if options are equal (with or without the "no" prefix).
212 */
213 static int
214 vfs_equalopts(const char *opt1, const char *opt2)
215 {
216 char *p;
217
218 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
219 if (strcmp(opt1, opt2) == 0)
220 return (1);
221 /* "noopt" vs. "opt" */
222 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
223 return (1);
224 /* "opt" vs. "noopt" */
225 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
226 return (1);
227 while ((p = strchr(opt1, '.')) != NULL &&
228 !strncmp(opt1, opt2, ++p - opt1)) {
229 opt2 += p - opt1;
230 opt1 = p;
231 /* "foo.noopt" vs. "foo.opt" */
232 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
233 return (1);
234 /* "foo.opt" vs. "foo.noopt" */
235 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
236 return (1);
237 }
238 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */
239 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
240 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
241 return (1);
242 return (0);
243 }
244
245 /*
246 * If a mount option is specified several times,
247 * (with or without the "no" prefix) only keep
248 * the last occurrence of it.
249 */
250 static void
251 vfs_sanitizeopts(struct vfsoptlist *opts)
252 {
253 struct vfsopt *opt, *opt2, *tmp;
254
255 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
256 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
257 while (opt2 != NULL) {
258 if (vfs_equalopts(opt->name, opt2->name)) {
259 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
260 vfs_freeopt(opts, opt2);
261 opt2 = tmp;
262 } else {
263 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
264 }
265 }
266 }
267 }
268
269 /*
270 * Build a linked list of mount options from a struct uio.
271 */
272 int
273 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
274 {
275 struct vfsoptlist *opts;
276 struct vfsopt *opt;
277 size_t memused, namelen, optlen;
278 unsigned int i, iovcnt;
279 int error;
280
281 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
282 TAILQ_INIT(opts);
283 memused = 0;
284 iovcnt = auio->uio_iovcnt;
285 for (i = 0; i < iovcnt; i += 2) {
286 namelen = auio->uio_iov[i].iov_len;
287 optlen = auio->uio_iov[i + 1].iov_len;
288 memused += sizeof(struct vfsopt) + optlen + namelen;
289 /*
290 * Avoid consuming too much memory, and attempts to overflow
291 * memused.
292 */
293 if (memused > VFS_MOUNTARG_SIZE_MAX ||
294 optlen > VFS_MOUNTARG_SIZE_MAX ||
295 namelen > VFS_MOUNTARG_SIZE_MAX) {
296 error = EINVAL;
297 goto bad;
298 }
299
300 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
301 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
302 opt->value = NULL;
303 opt->len = 0;
304 opt->pos = i / 2;
305 opt->seen = 0;
306
307 /*
308 * Do this early, so jumps to "bad" will free the current
309 * option.
310 */
311 TAILQ_INSERT_TAIL(opts, opt, link);
312
313 if (auio->uio_segflg == UIO_SYSSPACE) {
314 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
315 } else {
316 error = copyin(auio->uio_iov[i].iov_base, opt->name,
317 namelen);
318 if (error)
319 goto bad;
320 }
321 /* Ensure names are null-terminated strings. */
322 if (namelen == 0 || opt->name[namelen - 1] != '\0') {
323 error = EINVAL;
324 goto bad;
325 }
326 if (optlen != 0) {
327 opt->len = optlen;
328 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
329 if (auio->uio_segflg == UIO_SYSSPACE) {
330 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
331 optlen);
332 } else {
333 error = copyin(auio->uio_iov[i + 1].iov_base,
334 opt->value, optlen);
335 if (error)
336 goto bad;
337 }
338 }
339 }
340 vfs_sanitizeopts(opts);
341 *options = opts;
342 return (0);
343 bad:
344 vfs_freeopts(opts);
345 return (error);
346 }
347
348 /*
349 * Merge the old mount options with the new ones passed
350 * in the MNT_UPDATE case.
351 *
352 * XXX: This function will keep a "nofoo" option in the new
353 * options. E.g, if the option's canonical name is "foo",
354 * "nofoo" ends up in the mount point's active options.
355 */
356 static void
357 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
358 {
359 struct vfsopt *opt, *new;
360
361 TAILQ_FOREACH(opt, oldopts, link) {
362 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
363 new->name = strdup(opt->name, M_MOUNT);
364 if (opt->len != 0) {
365 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
366 bcopy(opt->value, new->value, opt->len);
367 } else
368 new->value = NULL;
369 new->len = opt->len;
370 new->seen = opt->seen;
371 TAILQ_INSERT_HEAD(toopts, new, link);
372 }
373 vfs_sanitizeopts(toopts);
374 }
375
376 /*
377 * Mount a filesystem.
378 */
379 #ifndef _SYS_SYSPROTO_H_
380 struct nmount_args {
381 struct iovec *iovp;
382 unsigned int iovcnt;
383 int flags;
384 };
385 #endif
386 int
387 sys_nmount(struct thread *td, struct nmount_args *uap)
388 {
389 struct uio *auio;
390 int error;
391 u_int iovcnt;
392 uint64_t flags;
393
394 /*
395 * Mount flags are now 64-bits. On 32-bit archtectures only
396 * 32-bits are passed in, but from here on everything handles
397 * 64-bit flags correctly.
398 */
399 flags = uap->flags;
400
401 AUDIT_ARG_FFLAGS(flags);
402 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
403 uap->iovp, uap->iovcnt, flags);
404
405 /*
406 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
407 * userspace to set this flag, but we must filter it out if we want
408 * MNT_UPDATE on the root file system to work.
409 * MNT_ROOTFS should only be set by the kernel when mounting its
410 * root file system.
411 */
412 flags &= ~MNT_ROOTFS;
413
414 iovcnt = uap->iovcnt;
415 /*
416 * Check that we have an even number of iovec's
417 * and that we have at least two options.
418 */
419 if ((iovcnt & 1) || (iovcnt < 4)) {
420 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
421 uap->iovcnt);
422 return (EINVAL);
423 }
424
425 error = copyinuio(uap->iovp, iovcnt, &auio);
426 if (error) {
427 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
428 __func__, error);
429 return (error);
430 }
431 error = vfs_donmount(td, flags, auio);
432
433 free(auio, M_IOV);
434 return (error);
435 }
436
437 /*
438 * ---------------------------------------------------------------------
439 * Various utility functions
440 */
441
442 void
443 vfs_ref(struct mount *mp)
444 {
445
446 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
447 MNT_ILOCK(mp);
448 MNT_REF(mp);
449 MNT_IUNLOCK(mp);
450 }
451
452 void
453 vfs_rel(struct mount *mp)
454 {
455
456 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
457 MNT_ILOCK(mp);
458 MNT_REL(mp);
459 MNT_IUNLOCK(mp);
460 }
461
462 /*
463 * Allocate and initialize the mount point struct.
464 */
465 struct mount *
466 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
467 struct ucred *cred)
468 {
469 struct mount *mp;
470
471 mp = uma_zalloc(mount_zone, M_WAITOK);
472 bzero(&mp->mnt_startzero,
473 __rangeof(struct mount, mnt_startzero, mnt_endzero));
474 TAILQ_INIT(&mp->mnt_nvnodelist);
475 mp->mnt_nvnodelistsize = 0;
476 TAILQ_INIT(&mp->mnt_activevnodelist);
477 mp->mnt_activevnodelistsize = 0;
478 TAILQ_INIT(&mp->mnt_tmpfreevnodelist);
479 mp->mnt_tmpfreevnodelistsize = 0;
480 mp->mnt_ref = 0;
481 (void) vfs_busy(mp, MBF_NOWAIT);
482 atomic_add_acq_int(&vfsp->vfc_refcount, 1);
483 mp->mnt_op = vfsp->vfc_vfsops;
484 mp->mnt_vfc = vfsp;
485 mp->mnt_stat.f_type = vfsp->vfc_typenum;
486 mp->mnt_gen++;
487 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
488 mp->mnt_vnodecovered = vp;
489 mp->mnt_cred = crdup(cred);
490 mp->mnt_stat.f_owner = cred->cr_uid;
491 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
492 mp->mnt_iosize_max = DFLTPHYS;
493 #ifdef MAC
494 mac_mount_init(mp);
495 mac_mount_create(cred, mp);
496 #endif
497 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
498 TAILQ_INIT(&mp->mnt_uppers);
499 return (mp);
500 }
501
502 /*
503 * Destroy the mount struct previously allocated by vfs_mount_alloc().
504 */
505 void
506 vfs_mount_destroy(struct mount *mp)
507 {
508
509 MNT_ILOCK(mp);
510 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
511 if (mp->mnt_kern_flag & MNTK_MWAIT) {
512 mp->mnt_kern_flag &= ~MNTK_MWAIT;
513 wakeup(mp);
514 }
515 while (mp->mnt_ref)
516 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
517 KASSERT(mp->mnt_ref == 0,
518 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
519 __FILE__, __LINE__));
520 if (mp->mnt_writeopcount != 0)
521 panic("vfs_mount_destroy: nonzero writeopcount");
522 if (mp->mnt_secondary_writes != 0)
523 panic("vfs_mount_destroy: nonzero secondary_writes");
524 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
525 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
526 struct vnode *vp;
527
528 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
529 vn_printf(vp, "dangling vnode ");
530 panic("unmount: dangling vnode");
531 }
532 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
533 if (mp->mnt_nvnodelistsize != 0)
534 panic("vfs_mount_destroy: nonzero nvnodelistsize");
535 if (mp->mnt_activevnodelistsize != 0)
536 panic("vfs_mount_destroy: nonzero activevnodelistsize");
537 if (mp->mnt_tmpfreevnodelistsize != 0)
538 panic("vfs_mount_destroy: nonzero tmpfreevnodelistsize");
539 if (mp->mnt_lockref != 0)
540 panic("vfs_mount_destroy: nonzero lock refcount");
541 MNT_IUNLOCK(mp);
542 if (mp->mnt_vnodecovered != NULL)
543 vrele(mp->mnt_vnodecovered);
544 #ifdef MAC
545 mac_mount_destroy(mp);
546 #endif
547 if (mp->mnt_opt != NULL)
548 vfs_freeopts(mp->mnt_opt);
549 crfree(mp->mnt_cred);
550 uma_zfree(mount_zone, mp);
551 }
552
553 static bool
554 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
555 {
556 /* This is an upgrade of an exisiting mount. */
557 if ((fsflags & MNT_UPDATE) != 0)
558 return (false);
559 /* This is already an R/O mount. */
560 if ((fsflags & MNT_RDONLY) != 0)
561 return (false);
562
563 switch (error) {
564 case ENODEV: /* generic, geom, ... */
565 case EACCES: /* cam/scsi, ... */
566 case EROFS: /* md, mmcsd, ... */
567 /*
568 * These errors can be returned by the storage layer to signal
569 * that the media is read-only. No harm in the R/O mount
570 * attempt if the error was returned for some other reason.
571 */
572 return (true);
573 default:
574 return (false);
575 }
576 }
577
578 int
579 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
580 {
581 struct vfsoptlist *optlist;
582 struct vfsopt *opt, *tmp_opt;
583 char *fstype, *fspath, *errmsg;
584 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
585 bool autoro;
586
587 errmsg = fspath = NULL;
588 errmsg_len = fspathlen = 0;
589 errmsg_pos = -1;
590 autoro = default_autoro;
591
592 error = vfs_buildopts(fsoptions, &optlist);
593 if (error)
594 return (error);
595
596 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
597 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
598
599 /*
600 * We need these two options before the others,
601 * and they are mandatory for any filesystem.
602 * Ensure they are NUL terminated as well.
603 */
604 fstypelen = 0;
605 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
606 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
607 error = EINVAL;
608 if (errmsg != NULL)
609 strncpy(errmsg, "Invalid fstype", errmsg_len);
610 goto bail;
611 }
612 fspathlen = 0;
613 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
614 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
615 error = EINVAL;
616 if (errmsg != NULL)
617 strncpy(errmsg, "Invalid fspath", errmsg_len);
618 goto bail;
619 }
620
621 /*
622 * We need to see if we have the "update" option
623 * before we call vfs_domount(), since vfs_domount() has special
624 * logic based on MNT_UPDATE. This is very important
625 * when we want to update the root filesystem.
626 */
627 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
628 if (strcmp(opt->name, "update") == 0) {
629 fsflags |= MNT_UPDATE;
630 vfs_freeopt(optlist, opt);
631 }
632 else if (strcmp(opt->name, "async") == 0)
633 fsflags |= MNT_ASYNC;
634 else if (strcmp(opt->name, "force") == 0) {
635 fsflags |= MNT_FORCE;
636 vfs_freeopt(optlist, opt);
637 }
638 else if (strcmp(opt->name, "reload") == 0) {
639 fsflags |= MNT_RELOAD;
640 vfs_freeopt(optlist, opt);
641 }
642 else if (strcmp(opt->name, "multilabel") == 0)
643 fsflags |= MNT_MULTILABEL;
644 else if (strcmp(opt->name, "noasync") == 0)
645 fsflags &= ~MNT_ASYNC;
646 else if (strcmp(opt->name, "noatime") == 0)
647 fsflags |= MNT_NOATIME;
648 else if (strcmp(opt->name, "atime") == 0) {
649 free(opt->name, M_MOUNT);
650 opt->name = strdup("nonoatime", M_MOUNT);
651 }
652 else if (strcmp(opt->name, "noclusterr") == 0)
653 fsflags |= MNT_NOCLUSTERR;
654 else if (strcmp(opt->name, "clusterr") == 0) {
655 free(opt->name, M_MOUNT);
656 opt->name = strdup("nonoclusterr", M_MOUNT);
657 }
658 else if (strcmp(opt->name, "noclusterw") == 0)
659 fsflags |= MNT_NOCLUSTERW;
660 else if (strcmp(opt->name, "clusterw") == 0) {
661 free(opt->name, M_MOUNT);
662 opt->name = strdup("nonoclusterw", M_MOUNT);
663 }
664 else if (strcmp(opt->name, "noexec") == 0)
665 fsflags |= MNT_NOEXEC;
666 else if (strcmp(opt->name, "exec") == 0) {
667 free(opt->name, M_MOUNT);
668 opt->name = strdup("nonoexec", M_MOUNT);
669 }
670 else if (strcmp(opt->name, "nosuid") == 0)
671 fsflags |= MNT_NOSUID;
672 else if (strcmp(opt->name, "suid") == 0) {
673 free(opt->name, M_MOUNT);
674 opt->name = strdup("nonosuid", M_MOUNT);
675 }
676 else if (strcmp(opt->name, "nosymfollow") == 0)
677 fsflags |= MNT_NOSYMFOLLOW;
678 else if (strcmp(opt->name, "symfollow") == 0) {
679 free(opt->name, M_MOUNT);
680 opt->name = strdup("nonosymfollow", M_MOUNT);
681 }
682 else if (strcmp(opt->name, "noro") == 0) {
683 fsflags &= ~MNT_RDONLY;
684 autoro = false;
685 }
686 else if (strcmp(opt->name, "rw") == 0) {
687 fsflags &= ~MNT_RDONLY;
688 autoro = false;
689 }
690 else if (strcmp(opt->name, "ro") == 0) {
691 fsflags |= MNT_RDONLY;
692 autoro = false;
693 }
694 else if (strcmp(opt->name, "rdonly") == 0) {
695 free(opt->name, M_MOUNT);
696 opt->name = strdup("ro", M_MOUNT);
697 fsflags |= MNT_RDONLY;
698 autoro = false;
699 }
700 else if (strcmp(opt->name, "autoro") == 0) {
701 vfs_freeopt(optlist, opt);
702 autoro = true;
703 }
704 else if (strcmp(opt->name, "suiddir") == 0)
705 fsflags |= MNT_SUIDDIR;
706 else if (strcmp(opt->name, "sync") == 0)
707 fsflags |= MNT_SYNCHRONOUS;
708 else if (strcmp(opt->name, "union") == 0)
709 fsflags |= MNT_UNION;
710 else if (strcmp(opt->name, "automounted") == 0) {
711 fsflags |= MNT_AUTOMOUNTED;
712 vfs_freeopt(optlist, opt);
713 }
714 }
715
716 /*
717 * Be ultra-paranoid about making sure the type and fspath
718 * variables will fit in our mp buffers, including the
719 * terminating NUL.
720 */
721 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
722 error = ENAMETOOLONG;
723 goto bail;
724 }
725
726 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
727
728 /*
729 * See if we can mount in the read-only mode if the error code suggests
730 * that it could be possible and the mount options allow for that.
731 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
732 * overridden by "autoro".
733 */
734 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
735 printf("%s: R/W mount failed, possibly R/O media,"
736 " trying R/O mount\n", __func__);
737 fsflags |= MNT_RDONLY;
738 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
739 }
740 bail:
741 /* copyout the errmsg */
742 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
743 && errmsg_len > 0 && errmsg != NULL) {
744 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
745 bcopy(errmsg,
746 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
747 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
748 } else {
749 copyout(errmsg,
750 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
751 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
752 }
753 }
754
755 if (optlist != NULL)
756 vfs_freeopts(optlist);
757 return (error);
758 }
759
760 /*
761 * Old mount API.
762 */
763 #ifndef _SYS_SYSPROTO_H_
764 struct mount_args {
765 char *type;
766 char *path;
767 int flags;
768 caddr_t data;
769 };
770 #endif
771 /* ARGSUSED */
772 int
773 sys_mount(struct thread *td, struct mount_args *uap)
774 {
775 char *fstype;
776 struct vfsconf *vfsp = NULL;
777 struct mntarg *ma = NULL;
778 uint64_t flags;
779 int error;
780
781 /*
782 * Mount flags are now 64-bits. On 32-bit architectures only
783 * 32-bits are passed in, but from here on everything handles
784 * 64-bit flags correctly.
785 */
786 flags = uap->flags;
787
788 AUDIT_ARG_FFLAGS(flags);
789
790 /*
791 * Filter out MNT_ROOTFS. We do not want clients of mount() in
792 * userspace to set this flag, but we must filter it out if we want
793 * MNT_UPDATE on the root file system to work.
794 * MNT_ROOTFS should only be set by the kernel when mounting its
795 * root file system.
796 */
797 flags &= ~MNT_ROOTFS;
798
799 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
800 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
801 if (error) {
802 free(fstype, M_TEMP);
803 return (error);
804 }
805
806 AUDIT_ARG_TEXT(fstype);
807 vfsp = vfs_byname_kld(fstype, td, &error);
808 free(fstype, M_TEMP);
809 if (vfsp == NULL)
810 return (ENOENT);
811 if (vfsp->vfc_vfsops->vfs_cmount == NULL)
812 return (EOPNOTSUPP);
813
814 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
815 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
816 ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
817 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
818 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
819
820 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags);
821 return (error);
822 }
823
824 /*
825 * vfs_domount_first(): first file system mount (not update)
826 */
827 static int
828 vfs_domount_first(
829 struct thread *td, /* Calling thread. */
830 struct vfsconf *vfsp, /* File system type. */
831 char *fspath, /* Mount path. */
832 struct vnode *vp, /* Vnode to be covered. */
833 uint64_t fsflags, /* Flags common to all filesystems. */
834 struct vfsoptlist **optlist /* Options local to the filesystem. */
835 )
836 {
837 struct vattr va;
838 struct mount *mp;
839 struct vnode *newdp;
840 int error, error1;
841
842 ASSERT_VOP_ELOCKED(vp, __func__);
843 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
844
845 /*
846 * If the jail of the calling thread lacks permission for this type of
847 * file system, or is trying to cover its own root, deny immediately.
848 */
849 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
850 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
851 vput(vp);
852 return (EPERM);
853 }
854
855 /*
856 * If the user is not root, ensure that they own the directory
857 * onto which we are attempting to mount.
858 */
859 error = VOP_GETATTR(vp, &va, td->td_ucred);
860 if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
861 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0);
862 if (error == 0)
863 error = vinvalbuf(vp, V_SAVE, 0, 0);
864 if (error == 0 && vp->v_type != VDIR)
865 error = ENOTDIR;
866 if (error == 0) {
867 VI_LOCK(vp);
868 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
869 vp->v_iflag |= VI_MOUNT;
870 else
871 error = EBUSY;
872 VI_UNLOCK(vp);
873 }
874 if (error != 0) {
875 vput(vp);
876 return (error);
877 }
878 VOP_UNLOCK(vp, 0);
879
880 /* Allocate and initialize the filesystem. */
881 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
882 /* XXXMAC: pass to vfs_mount_alloc? */
883 mp->mnt_optnew = *optlist;
884 /* Set the mount level flags. */
885 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
886
887 /*
888 * Mount the filesystem.
889 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
890 * get. No freeing of cn_pnbuf.
891 */
892 error1 = 0;
893 if ((error = VFS_MOUNT(mp)) != 0 ||
894 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
895 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
896 if (error1 != 0) {
897 error = error1;
898 if ((error1 = VFS_UNMOUNT(mp, 0)) != 0)
899 printf("VFS_UNMOUNT returned %d\n", error1);
900 }
901 vfs_unbusy(mp);
902 mp->mnt_vnodecovered = NULL;
903 vfs_mount_destroy(mp);
904 VI_LOCK(vp);
905 vp->v_iflag &= ~VI_MOUNT;
906 VI_UNLOCK(vp);
907 vrele(vp);
908 return (error);
909 }
910 VOP_UNLOCK(newdp, 0);
911
912 if (mp->mnt_opt != NULL)
913 vfs_freeopts(mp->mnt_opt);
914 mp->mnt_opt = mp->mnt_optnew;
915 *optlist = NULL;
916
917 /*
918 * Prevent external consumers of mount options from reading mnt_optnew.
919 */
920 mp->mnt_optnew = NULL;
921
922 MNT_ILOCK(mp);
923 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
924 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
925 mp->mnt_kern_flag |= MNTK_ASYNC;
926 else
927 mp->mnt_kern_flag &= ~MNTK_ASYNC;
928 MNT_IUNLOCK(mp);
929
930 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
931 cache_purge(vp);
932 VI_LOCK(vp);
933 vp->v_iflag &= ~VI_MOUNT;
934 VI_UNLOCK(vp);
935 vp->v_mountedhere = mp;
936 /* Place the new filesystem at the end of the mount list. */
937 mtx_lock(&mountlist_mtx);
938 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
939 mtx_unlock(&mountlist_mtx);
940 vfs_event_signal(NULL, VQ_MOUNT, 0);
941 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
942 VOP_UNLOCK(vp, 0);
943 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
944 VOP_UNLOCK(newdp, 0);
945 mountcheckdirs(vp, newdp);
946 vrele(newdp);
947 if ((mp->mnt_flag & MNT_RDONLY) == 0)
948 vfs_allocate_syncvnode(mp);
949 vfs_unbusy(mp);
950 return (0);
951 }
952
953 /*
954 * vfs_domount_update(): update of mounted file system
955 */
956 static int
957 vfs_domount_update(
958 struct thread *td, /* Calling thread. */
959 struct vnode *vp, /* Mount point vnode. */
960 uint64_t fsflags, /* Flags common to all filesystems. */
961 struct vfsoptlist **optlist /* Options local to the filesystem. */
962 )
963 {
964 struct export_args export;
965 void *bufp;
966 struct mount *mp;
967 int error, export_error, len;
968 uint64_t flag;
969
970 ASSERT_VOP_ELOCKED(vp, __func__);
971 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
972 mp = vp->v_mount;
973
974 if ((vp->v_vflag & VV_ROOT) == 0) {
975 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
976 == 0)
977 error = EXDEV;
978 else
979 error = EINVAL;
980 vput(vp);
981 return (error);
982 }
983
984 /*
985 * We only allow the filesystem to be reloaded if it
986 * is currently mounted read-only.
987 */
988 flag = mp->mnt_flag;
989 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
990 vput(vp);
991 return (EOPNOTSUPP); /* Needs translation */
992 }
993 /*
994 * Only privileged root, or (if MNT_USER is set) the user that
995 * did the original mount is permitted to update it.
996 */
997 error = vfs_suser(mp, td);
998 if (error != 0) {
999 vput(vp);
1000 return (error);
1001 }
1002 if (vfs_busy(mp, MBF_NOWAIT)) {
1003 vput(vp);
1004 return (EBUSY);
1005 }
1006 VI_LOCK(vp);
1007 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1008 VI_UNLOCK(vp);
1009 vfs_unbusy(mp);
1010 vput(vp);
1011 return (EBUSY);
1012 }
1013 vp->v_iflag |= VI_MOUNT;
1014 VI_UNLOCK(vp);
1015 VOP_UNLOCK(vp, 0);
1016
1017 MNT_ILOCK(mp);
1018 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1019 MNT_IUNLOCK(mp);
1020 error = EBUSY;
1021 goto end;
1022 }
1023 mp->mnt_flag &= ~MNT_UPDATEMASK;
1024 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1025 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1026 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1027 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1028 MNT_IUNLOCK(mp);
1029 mp->mnt_optnew = *optlist;
1030 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1031
1032 /*
1033 * Mount the filesystem.
1034 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1035 * get. No freeing of cn_pnbuf.
1036 */
1037 error = VFS_MOUNT(mp);
1038
1039 export_error = 0;
1040 /* Process the export option. */
1041 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1042 &len) == 0) {
1043 /* Assume that there is only 1 ABI for each length. */
1044 switch (len) {
1045 case (sizeof(struct oexport_args)):
1046 bzero(&export, sizeof(export));
1047 /* FALLTHROUGH */
1048 case (sizeof(export)):
1049 bcopy(bufp, &export, len);
1050 export_error = vfs_export(mp, &export);
1051 break;
1052 default:
1053 export_error = EINVAL;
1054 break;
1055 }
1056 }
1057
1058 MNT_ILOCK(mp);
1059 if (error == 0) {
1060 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1061 MNT_SNAPSHOT);
1062 } else {
1063 /*
1064 * If we fail, restore old mount flags. MNT_QUOTA is special,
1065 * because it is not part of MNT_UPDATEMASK, but it could have
1066 * changed in the meantime if quotactl(2) was called.
1067 * All in all we want current value of MNT_QUOTA, not the old
1068 * one.
1069 */
1070 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1071 }
1072 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1073 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1074 mp->mnt_kern_flag |= MNTK_ASYNC;
1075 else
1076 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1077 MNT_IUNLOCK(mp);
1078
1079 if (error != 0)
1080 goto end;
1081
1082 if (mp->mnt_opt != NULL)
1083 vfs_freeopts(mp->mnt_opt);
1084 mp->mnt_opt = mp->mnt_optnew;
1085 *optlist = NULL;
1086 (void)VFS_STATFS(mp, &mp->mnt_stat);
1087 /*
1088 * Prevent external consumers of mount options from reading
1089 * mnt_optnew.
1090 */
1091 mp->mnt_optnew = NULL;
1092
1093 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1094 vfs_allocate_syncvnode(mp);
1095 else
1096 vfs_deallocate_syncvnode(mp);
1097 end:
1098 vfs_unbusy(mp);
1099 VI_LOCK(vp);
1100 vp->v_iflag &= ~VI_MOUNT;
1101 VI_UNLOCK(vp);
1102 vrele(vp);
1103 return (error != 0 ? error : export_error);
1104 }
1105
1106 /*
1107 * vfs_domount(): actually attempt a filesystem mount.
1108 */
1109 static int
1110 vfs_domount(
1111 struct thread *td, /* Calling thread. */
1112 const char *fstype, /* Filesystem type. */
1113 char *fspath, /* Mount path. */
1114 uint64_t fsflags, /* Flags common to all filesystems. */
1115 struct vfsoptlist **optlist /* Options local to the filesystem. */
1116 )
1117 {
1118 struct vfsconf *vfsp;
1119 struct nameidata nd;
1120 struct vnode *vp;
1121 char *pathbuf;
1122 int error;
1123
1124 /*
1125 * Be ultra-paranoid about making sure the type and fspath
1126 * variables will fit in our mp buffers, including the
1127 * terminating NUL.
1128 */
1129 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1130 return (ENAMETOOLONG);
1131
1132 if (jailed(td->td_ucred) || usermount == 0) {
1133 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1134 return (error);
1135 }
1136
1137 /*
1138 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1139 */
1140 if (fsflags & MNT_EXPORTED) {
1141 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1142 if (error)
1143 return (error);
1144 }
1145 if (fsflags & MNT_SUIDDIR) {
1146 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1147 if (error)
1148 return (error);
1149 }
1150 /*
1151 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1152 */
1153 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1154 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1155 fsflags |= MNT_NOSUID | MNT_USER;
1156 }
1157
1158 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1159 vfsp = NULL;
1160 if ((fsflags & MNT_UPDATE) == 0) {
1161 /* Don't try to load KLDs if we're mounting the root. */
1162 if (fsflags & MNT_ROOTFS)
1163 vfsp = vfs_byname(fstype);
1164 else
1165 vfsp = vfs_byname_kld(fstype, td, &error);
1166 if (vfsp == NULL)
1167 return (ENODEV);
1168 }
1169
1170 /*
1171 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1172 */
1173 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1174 UIO_SYSSPACE, fspath, td);
1175 error = namei(&nd);
1176 if (error != 0)
1177 return (error);
1178 NDFREE(&nd, NDF_ONLY_PNBUF);
1179 vp = nd.ni_vp;
1180 if ((fsflags & MNT_UPDATE) == 0) {
1181 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1182 strcpy(pathbuf, fspath);
1183 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1184 /* debug.disablefullpath == 1 results in ENODEV */
1185 if (error == 0 || error == ENODEV) {
1186 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1187 fsflags, optlist);
1188 }
1189 free(pathbuf, M_TEMP);
1190 } else
1191 error = vfs_domount_update(td, vp, fsflags, optlist);
1192
1193 return (error);
1194 }
1195
1196 /*
1197 * Unmount a filesystem.
1198 *
1199 * Note: unmount takes a path to the vnode mounted on as argument, not
1200 * special file (as before).
1201 */
1202 #ifndef _SYS_SYSPROTO_H_
1203 struct unmount_args {
1204 char *path;
1205 int flags;
1206 };
1207 #endif
1208 /* ARGSUSED */
1209 int
1210 sys_unmount(struct thread *td, struct unmount_args *uap)
1211 {
1212 struct nameidata nd;
1213 struct mount *mp;
1214 char *pathbuf;
1215 int error, id0, id1;
1216
1217 AUDIT_ARG_VALUE(uap->flags);
1218 if (jailed(td->td_ucred) || usermount == 0) {
1219 error = priv_check(td, PRIV_VFS_UNMOUNT);
1220 if (error)
1221 return (error);
1222 }
1223
1224 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1225 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1226 if (error) {
1227 free(pathbuf, M_TEMP);
1228 return (error);
1229 }
1230 if (uap->flags & MNT_BYFSID) {
1231 AUDIT_ARG_TEXT(pathbuf);
1232 /* Decode the filesystem ID. */
1233 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1234 free(pathbuf, M_TEMP);
1235 return (EINVAL);
1236 }
1237
1238 mtx_lock(&mountlist_mtx);
1239 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1240 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1241 mp->mnt_stat.f_fsid.val[1] == id1) {
1242 vfs_ref(mp);
1243 break;
1244 }
1245 }
1246 mtx_unlock(&mountlist_mtx);
1247 } else {
1248 /*
1249 * Try to find global path for path argument.
1250 */
1251 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1252 UIO_SYSSPACE, pathbuf, td);
1253 if (namei(&nd) == 0) {
1254 NDFREE(&nd, NDF_ONLY_PNBUF);
1255 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1256 MNAMELEN);
1257 if (error == 0 || error == ENODEV)
1258 vput(nd.ni_vp);
1259 }
1260 mtx_lock(&mountlist_mtx);
1261 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1262 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1263 vfs_ref(mp);
1264 break;
1265 }
1266 }
1267 mtx_unlock(&mountlist_mtx);
1268 }
1269 free(pathbuf, M_TEMP);
1270 if (mp == NULL) {
1271 /*
1272 * Previously we returned ENOENT for a nonexistent path and
1273 * EINVAL for a non-mountpoint. We cannot tell these apart
1274 * now, so in the !MNT_BYFSID case return the more likely
1275 * EINVAL for compatibility.
1276 */
1277 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1278 }
1279
1280 /*
1281 * Don't allow unmounting the root filesystem.
1282 */
1283 if (mp->mnt_flag & MNT_ROOTFS) {
1284 vfs_rel(mp);
1285 return (EINVAL);
1286 }
1287 error = dounmount(mp, uap->flags, td);
1288 return (error);
1289 }
1290
1291 /*
1292 * Return error if any of the vnodes, ignoring the root vnode
1293 * and the syncer vnode, have non-zero usecount.
1294 *
1295 * This function is purely advisory - it can return false positives
1296 * and negatives.
1297 */
1298 static int
1299 vfs_check_usecounts(struct mount *mp)
1300 {
1301 struct vnode *vp, *mvp;
1302
1303 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1304 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1305 vp->v_usecount != 0) {
1306 VI_UNLOCK(vp);
1307 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1308 return (EBUSY);
1309 }
1310 VI_UNLOCK(vp);
1311 }
1312
1313 return (0);
1314 }
1315
1316 static void
1317 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1318 {
1319
1320 mtx_assert(MNT_MTX(mp), MA_OWNED);
1321 mp->mnt_kern_flag &= ~mntkflags;
1322 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1323 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1324 wakeup(mp);
1325 }
1326 MNT_IUNLOCK(mp);
1327 if (coveredvp != NULL) {
1328 VOP_UNLOCK(coveredvp, 0);
1329 vdrop(coveredvp);
1330 }
1331 vn_finished_write(mp);
1332 }
1333
1334 /*
1335 * Do the actual filesystem unmount.
1336 */
1337 int
1338 dounmount(struct mount *mp, int flags, struct thread *td)
1339 {
1340 struct vnode *coveredvp;
1341 int error;
1342 uint64_t async_flag;
1343 int mnt_gen_r;
1344
1345 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1346 mnt_gen_r = mp->mnt_gen;
1347 VI_LOCK(coveredvp);
1348 vholdl(coveredvp);
1349 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1350 /*
1351 * Check for mp being unmounted while waiting for the
1352 * covered vnode lock.
1353 */
1354 if (coveredvp->v_mountedhere != mp ||
1355 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1356 VOP_UNLOCK(coveredvp, 0);
1357 vdrop(coveredvp);
1358 vfs_rel(mp);
1359 return (EBUSY);
1360 }
1361 }
1362
1363 /*
1364 * Only privileged root, or (if MNT_USER is set) the user that did the
1365 * original mount is permitted to unmount this filesystem.
1366 */
1367 error = vfs_suser(mp, td);
1368 if (error != 0) {
1369 if (coveredvp != NULL) {
1370 VOP_UNLOCK(coveredvp, 0);
1371 vdrop(coveredvp);
1372 }
1373 vfs_rel(mp);
1374 return (error);
1375 }
1376
1377 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1378 MNT_ILOCK(mp);
1379 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1380 (mp->mnt_flag & MNT_UPDATE) != 0 ||
1381 !TAILQ_EMPTY(&mp->mnt_uppers)) {
1382 dounmount_cleanup(mp, coveredvp, 0);
1383 return (EBUSY);
1384 }
1385 mp->mnt_kern_flag |= MNTK_UNMOUNT;
1386 if (flags & MNT_NONBUSY) {
1387 MNT_IUNLOCK(mp);
1388 error = vfs_check_usecounts(mp);
1389 MNT_ILOCK(mp);
1390 if (error != 0) {
1391 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1392 return (error);
1393 }
1394 }
1395 /* Allow filesystems to detect that a forced unmount is in progress. */
1396 if (flags & MNT_FORCE) {
1397 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1398 MNT_IUNLOCK(mp);
1399 /*
1400 * Must be done after setting MNTK_UNMOUNTF and before
1401 * waiting for mnt_lockref to become 0.
1402 */
1403 VFS_PURGE(mp);
1404 MNT_ILOCK(mp);
1405 }
1406 error = 0;
1407 if (mp->mnt_lockref) {
1408 mp->mnt_kern_flag |= MNTK_DRAINING;
1409 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1410 "mount drain", 0);
1411 }
1412 MNT_IUNLOCK(mp);
1413 KASSERT(mp->mnt_lockref == 0,
1414 ("%s: invalid lock refcount in the drain path @ %s:%d",
1415 __func__, __FILE__, __LINE__));
1416 KASSERT(error == 0,
1417 ("%s: invalid return value for msleep in the drain path @ %s:%d",
1418 __func__, __FILE__, __LINE__));
1419
1420 if (mp->mnt_flag & MNT_EXPUBLIC)
1421 vfs_setpublicfs(NULL, NULL, NULL);
1422
1423 /*
1424 * From now, we can claim that the use reference on the
1425 * coveredvp is ours, and the ref can be released only by
1426 * successfull unmount by us, or left for later unmount
1427 * attempt. The previously acquired hold reference is no
1428 * longer needed to protect the vnode from reuse.
1429 */
1430 if (coveredvp != NULL)
1431 vdrop(coveredvp);
1432
1433 vfs_msync(mp, MNT_WAIT);
1434 MNT_ILOCK(mp);
1435 async_flag = mp->mnt_flag & MNT_ASYNC;
1436 mp->mnt_flag &= ~MNT_ASYNC;
1437 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1438 MNT_IUNLOCK(mp);
1439 cache_purgevfs(mp, false); /* remove cache entries for this file sys */
1440 vfs_deallocate_syncvnode(mp);
1441 error = VFS_UNMOUNT(mp, flags);
1442 vn_finished_write(mp);
1443 /*
1444 * If we failed to flush the dirty blocks for this mount point,
1445 * undo all the cdir/rdir and rootvnode changes we made above.
1446 * Unless we failed to do so because the device is reporting that
1447 * it doesn't exist anymore.
1448 */
1449 if (error && error != ENXIO) {
1450 MNT_ILOCK(mp);
1451 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1452 MNT_IUNLOCK(mp);
1453 vfs_allocate_syncvnode(mp);
1454 MNT_ILOCK(mp);
1455 }
1456 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1457 mp->mnt_flag |= async_flag;
1458 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1459 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1460 mp->mnt_kern_flag |= MNTK_ASYNC;
1461 if (mp->mnt_kern_flag & MNTK_MWAIT) {
1462 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1463 wakeup(mp);
1464 }
1465 MNT_IUNLOCK(mp);
1466 if (coveredvp)
1467 VOP_UNLOCK(coveredvp, 0);
1468 return (error);
1469 }
1470 mtx_lock(&mountlist_mtx);
1471 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1472 mtx_unlock(&mountlist_mtx);
1473 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1474 if (coveredvp != NULL) {
1475 coveredvp->v_mountedhere = NULL;
1476 VOP_UNLOCK(coveredvp, 0);
1477 }
1478 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1479 if (rootvnode != NULL && mp == rootvnode->v_mount) {
1480 vrele(rootvnode);
1481 rootvnode = NULL;
1482 }
1483 if (mp == rootdevmp)
1484 rootdevmp = NULL;
1485 vfs_mount_destroy(mp);
1486 return (0);
1487 }
1488
1489 /*
1490 * Report errors during filesystem mounting.
1491 */
1492 void
1493 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1494 {
1495 struct vfsoptlist *moptlist = mp->mnt_optnew;
1496 va_list ap;
1497 int error, len;
1498 char *errmsg;
1499
1500 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1501 if (error || errmsg == NULL || len <= 0)
1502 return;
1503
1504 va_start(ap, fmt);
1505 vsnprintf(errmsg, (size_t)len, fmt, ap);
1506 va_end(ap);
1507 }
1508
1509 void
1510 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1511 {
1512 va_list ap;
1513 int error, len;
1514 char *errmsg;
1515
1516 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1517 if (error || errmsg == NULL || len <= 0)
1518 return;
1519
1520 va_start(ap, fmt);
1521 vsnprintf(errmsg, (size_t)len, fmt, ap);
1522 va_end(ap);
1523 }
1524
1525 /*
1526 * ---------------------------------------------------------------------
1527 * Functions for querying mount options/arguments from filesystems.
1528 */
1529
1530 /*
1531 * Check that no unknown options are given
1532 */
1533 int
1534 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1535 {
1536 struct vfsopt *opt;
1537 char errmsg[255];
1538 const char **t, *p, *q;
1539 int ret = 0;
1540
1541 TAILQ_FOREACH(opt, opts, link) {
1542 p = opt->name;
1543 q = NULL;
1544 if (p[0] == 'n' && p[1] == 'o')
1545 q = p + 2;
1546 for(t = global_opts; *t != NULL; t++) {
1547 if (strcmp(*t, p) == 0)
1548 break;
1549 if (q != NULL) {
1550 if (strcmp(*t, q) == 0)
1551 break;
1552 }
1553 }
1554 if (*t != NULL)
1555 continue;
1556 for(t = legal; *t != NULL; t++) {
1557 if (strcmp(*t, p) == 0)
1558 break;
1559 if (q != NULL) {
1560 if (strcmp(*t, q) == 0)
1561 break;
1562 }
1563 }
1564 if (*t != NULL)
1565 continue;
1566 snprintf(errmsg, sizeof(errmsg),
1567 "mount option <%s> is unknown", p);
1568 ret = EINVAL;
1569 }
1570 if (ret != 0) {
1571 TAILQ_FOREACH(opt, opts, link) {
1572 if (strcmp(opt->name, "errmsg") == 0) {
1573 strncpy((char *)opt->value, errmsg, opt->len);
1574 break;
1575 }
1576 }
1577 if (opt == NULL)
1578 printf("%s\n", errmsg);
1579 }
1580 return (ret);
1581 }
1582
1583 /*
1584 * Get a mount option by its name.
1585 *
1586 * Return 0 if the option was found, ENOENT otherwise.
1587 * If len is non-NULL it will be filled with the length
1588 * of the option. If buf is non-NULL, it will be filled
1589 * with the address of the option.
1590 */
1591 int
1592 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
1593 {
1594 struct vfsopt *opt;
1595
1596 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1597
1598 TAILQ_FOREACH(opt, opts, link) {
1599 if (strcmp(name, opt->name) == 0) {
1600 opt->seen = 1;
1601 if (len != NULL)
1602 *len = opt->len;
1603 if (buf != NULL)
1604 *buf = opt->value;
1605 return (0);
1606 }
1607 }
1608 return (ENOENT);
1609 }
1610
1611 int
1612 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1613 {
1614 struct vfsopt *opt;
1615
1616 if (opts == NULL)
1617 return (-1);
1618
1619 TAILQ_FOREACH(opt, opts, link) {
1620 if (strcmp(name, opt->name) == 0) {
1621 opt->seen = 1;
1622 return (opt->pos);
1623 }
1624 }
1625 return (-1);
1626 }
1627
1628 int
1629 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
1630 {
1631 char *opt_value, *vtp;
1632 quad_t iv;
1633 int error, opt_len;
1634
1635 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
1636 if (error != 0)
1637 return (error);
1638 if (opt_len == 0 || opt_value == NULL)
1639 return (EINVAL);
1640 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
1641 return (EINVAL);
1642 iv = strtoq(opt_value, &vtp, 0);
1643 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
1644 return (EINVAL);
1645 if (iv < 0)
1646 return (EINVAL);
1647 switch (vtp[0]) {
1648 case 't': case 'T':
1649 iv *= 1024;
1650 /* FALLTHROUGH */
1651 case 'g': case 'G':
1652 iv *= 1024;
1653 /* FALLTHROUGH */
1654 case 'm': case 'M':
1655 iv *= 1024;
1656 /* FALLTHROUGH */
1657 case 'k': case 'K':
1658 iv *= 1024;
1659 case '\0':
1660 break;
1661 default:
1662 return (EINVAL);
1663 }
1664 *value = iv;
1665
1666 return (0);
1667 }
1668
1669 char *
1670 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1671 {
1672 struct vfsopt *opt;
1673
1674 *error = 0;
1675 TAILQ_FOREACH(opt, opts, link) {
1676 if (strcmp(name, opt->name) != 0)
1677 continue;
1678 opt->seen = 1;
1679 if (opt->len == 0 ||
1680 ((char *)opt->value)[opt->len - 1] != '\0') {
1681 *error = EINVAL;
1682 return (NULL);
1683 }
1684 return (opt->value);
1685 }
1686 *error = ENOENT;
1687 return (NULL);
1688 }
1689
1690 int
1691 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
1692 uint64_t val)
1693 {
1694 struct vfsopt *opt;
1695
1696 TAILQ_FOREACH(opt, opts, link) {
1697 if (strcmp(name, opt->name) == 0) {
1698 opt->seen = 1;
1699 if (w != NULL)
1700 *w |= val;
1701 return (1);
1702 }
1703 }
1704 if (w != NULL)
1705 *w &= ~val;
1706 return (0);
1707 }
1708
1709 int
1710 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1711 {
1712 va_list ap;
1713 struct vfsopt *opt;
1714 int ret;
1715
1716 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1717
1718 TAILQ_FOREACH(opt, opts, link) {
1719 if (strcmp(name, opt->name) != 0)
1720 continue;
1721 opt->seen = 1;
1722 if (opt->len == 0 || opt->value == NULL)
1723 return (0);
1724 if (((char *)opt->value)[opt->len - 1] != '\0')
1725 return (0);
1726 va_start(ap, fmt);
1727 ret = vsscanf(opt->value, fmt, ap);
1728 va_end(ap);
1729 return (ret);
1730 }
1731 return (0);
1732 }
1733
1734 int
1735 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
1736 {
1737 struct vfsopt *opt;
1738
1739 TAILQ_FOREACH(opt, opts, link) {
1740 if (strcmp(name, opt->name) != 0)
1741 continue;
1742 opt->seen = 1;
1743 if (opt->value == NULL)
1744 opt->len = len;
1745 else {
1746 if (opt->len != len)
1747 return (EINVAL);
1748 bcopy(value, opt->value, len);
1749 }
1750 return (0);
1751 }
1752 return (ENOENT);
1753 }
1754
1755 int
1756 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
1757 {
1758 struct vfsopt *opt;
1759
1760 TAILQ_FOREACH(opt, opts, link) {
1761 if (strcmp(name, opt->name) != 0)
1762 continue;
1763 opt->seen = 1;
1764 if (opt->value == NULL)
1765 opt->len = len;
1766 else {
1767 if (opt->len < len)
1768 return (EINVAL);
1769 opt->len = len;
1770 bcopy(value, opt->value, len);
1771 }
1772 return (0);
1773 }
1774 return (ENOENT);
1775 }
1776
1777 int
1778 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
1779 {
1780 struct vfsopt *opt;
1781
1782 TAILQ_FOREACH(opt, opts, link) {
1783 if (strcmp(name, opt->name) != 0)
1784 continue;
1785 opt->seen = 1;
1786 if (opt->value == NULL)
1787 opt->len = strlen(value) + 1;
1788 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
1789 return (EINVAL);
1790 return (0);
1791 }
1792 return (ENOENT);
1793 }
1794
1795 /*
1796 * Find and copy a mount option.
1797 *
1798 * The size of the buffer has to be specified
1799 * in len, if it is not the same length as the
1800 * mount option, EINVAL is returned.
1801 * Returns ENOENT if the option is not found.
1802 */
1803 int
1804 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
1805 {
1806 struct vfsopt *opt;
1807
1808 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1809
1810 TAILQ_FOREACH(opt, opts, link) {
1811 if (strcmp(name, opt->name) == 0) {
1812 opt->seen = 1;
1813 if (len != opt->len)
1814 return (EINVAL);
1815 bcopy(opt->value, dest, opt->len);
1816 return (0);
1817 }
1818 }
1819 return (ENOENT);
1820 }
1821
1822 int
1823 __vfs_statfs(struct mount *mp, struct statfs *sbp)
1824 {
1825 int error;
1826
1827 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat);
1828 if (sbp != &mp->mnt_stat)
1829 *sbp = mp->mnt_stat;
1830 return (error);
1831 }
1832
1833 void
1834 vfs_mountedfrom(struct mount *mp, const char *from)
1835 {
1836
1837 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1838 strlcpy(mp->mnt_stat.f_mntfromname, from,
1839 sizeof mp->mnt_stat.f_mntfromname);
1840 }
1841
1842 /*
1843 * ---------------------------------------------------------------------
1844 * This is the api for building mount args and mounting filesystems from
1845 * inside the kernel.
1846 *
1847 * The API works by accumulation of individual args. First error is
1848 * latched.
1849 *
1850 * XXX: should be documented in new manpage kernel_mount(9)
1851 */
1852
1853 /* A memory allocation which must be freed when we are done */
1854 struct mntaarg {
1855 SLIST_ENTRY(mntaarg) next;
1856 };
1857
1858 /* The header for the mount arguments */
1859 struct mntarg {
1860 struct iovec *v;
1861 int len;
1862 int error;
1863 SLIST_HEAD(, mntaarg) list;
1864 };
1865
1866 /*
1867 * Add a boolean argument.
1868 *
1869 * flag is the boolean value.
1870 * name must start with "no".
1871 */
1872 struct mntarg *
1873 mount_argb(struct mntarg *ma, int flag, const char *name)
1874 {
1875
1876 KASSERT(name[0] == 'n' && name[1] == 'o',
1877 ("mount_argb(...,%s): name must start with 'no'", name));
1878
1879 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1880 }
1881
1882 /*
1883 * Add an argument printf style
1884 */
1885 struct mntarg *
1886 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1887 {
1888 va_list ap;
1889 struct mntaarg *maa;
1890 struct sbuf *sb;
1891 int len;
1892
1893 if (ma == NULL) {
1894 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1895 SLIST_INIT(&ma->list);
1896 }
1897 if (ma->error)
1898 return (ma);
1899
1900 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1901 M_MOUNT, M_WAITOK);
1902 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1903 ma->v[ma->len].iov_len = strlen(name) + 1;
1904 ma->len++;
1905
1906 sb = sbuf_new_auto();
1907 va_start(ap, fmt);
1908 sbuf_vprintf(sb, fmt, ap);
1909 va_end(ap);
1910 sbuf_finish(sb);
1911 len = sbuf_len(sb) + 1;
1912 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1913 SLIST_INSERT_HEAD(&ma->list, maa, next);
1914 bcopy(sbuf_data(sb), maa + 1, len);
1915 sbuf_delete(sb);
1916
1917 ma->v[ma->len].iov_base = maa + 1;
1918 ma->v[ma->len].iov_len = len;
1919 ma->len++;
1920
1921 return (ma);
1922 }
1923
1924 /*
1925 * Add an argument which is a userland string.
1926 */
1927 struct mntarg *
1928 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
1929 {
1930 struct mntaarg *maa;
1931 char *tbuf;
1932
1933 if (val == NULL)
1934 return (ma);
1935 if (ma == NULL) {
1936 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1937 SLIST_INIT(&ma->list);
1938 }
1939 if (ma->error)
1940 return (ma);
1941 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1942 SLIST_INSERT_HEAD(&ma->list, maa, next);
1943 tbuf = (void *)(maa + 1);
1944 ma->error = copyinstr(val, tbuf, len, NULL);
1945 return (mount_arg(ma, name, tbuf, -1));
1946 }
1947
1948 /*
1949 * Plain argument.
1950 *
1951 * If length is -1, treat value as a C string.
1952 */
1953 struct mntarg *
1954 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
1955 {
1956
1957 if (ma == NULL) {
1958 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1959 SLIST_INIT(&ma->list);
1960 }
1961 if (ma->error)
1962 return (ma);
1963
1964 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1965 M_MOUNT, M_WAITOK);
1966 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1967 ma->v[ma->len].iov_len = strlen(name) + 1;
1968 ma->len++;
1969
1970 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
1971 if (len < 0)
1972 ma->v[ma->len].iov_len = strlen(val) + 1;
1973 else
1974 ma->v[ma->len].iov_len = len;
1975 ma->len++;
1976 return (ma);
1977 }
1978
1979 /*
1980 * Free a mntarg structure
1981 */
1982 static void
1983 free_mntarg(struct mntarg *ma)
1984 {
1985 struct mntaarg *maa;
1986
1987 while (!SLIST_EMPTY(&ma->list)) {
1988 maa = SLIST_FIRST(&ma->list);
1989 SLIST_REMOVE_HEAD(&ma->list, next);
1990 free(maa, M_MOUNT);
1991 }
1992 free(ma->v, M_MOUNT);
1993 free(ma, M_MOUNT);
1994 }
1995
1996 /*
1997 * Mount a filesystem
1998 */
1999 int
2000 kernel_mount(struct mntarg *ma, uint64_t flags)
2001 {
2002 struct uio auio;
2003 int error;
2004
2005 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2006 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2007 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2008
2009 auio.uio_iov = ma->v;
2010 auio.uio_iovcnt = ma->len;
2011 auio.uio_segflg = UIO_SYSSPACE;
2012
2013 error = ma->error;
2014 if (!error)
2015 error = vfs_donmount(curthread, flags, &auio);
2016 free_mntarg(ma);
2017 return (error);
2018 }
2019
2020 /*
2021 * A printflike function to mount a filesystem.
2022 */
2023 int
2024 kernel_vmount(int flags, ...)
2025 {
2026 struct mntarg *ma = NULL;
2027 va_list ap;
2028 const char *cp;
2029 const void *vp;
2030 int error;
2031
2032 va_start(ap, flags);
2033 for (;;) {
2034 cp = va_arg(ap, const char *);
2035 if (cp == NULL)
2036 break;
2037 vp = va_arg(ap, const void *);
2038 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2039 }
2040 va_end(ap);
2041
2042 error = kernel_mount(ma, flags);
2043 return (error);
2044 }
2045
2046 /*
2047 * Convert the old export args format into new export args.
2048 *
2049 * The old export args struct does not have security flavors. Otherwise, the
2050 * structs are identical. The default security flavor 'sys' is applied by
2051 * vfs_export when .ex_numsecflavors is 0.
2052 */
2053 void
2054 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp)
2055 {
2056
2057 bcopy(oexp, exp, sizeof(*oexp));
2058 exp->ex_numsecflavors = 0;
2059 }
2060
2061 /*
2062 * Suspend write operations on all local writeable filesystems. Does
2063 * full sync of them in the process.
2064 *
2065 * Iterate over the mount points in reverse order, suspending most
2066 * recently mounted filesystems first. It handles a case where a
2067 * filesystem mounted from a md(4) vnode-backed device should be
2068 * suspended before the filesystem that owns the vnode.
2069 */
2070 void
2071 suspend_all_fs(void)
2072 {
2073 struct mount *mp;
2074 int error;
2075
2076 mtx_lock(&mountlist_mtx);
2077 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2078 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2079 if (error != 0)
2080 continue;
2081 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2082 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2083 mtx_lock(&mountlist_mtx);
2084 vfs_unbusy(mp);
2085 continue;
2086 }
2087 error = vfs_write_suspend(mp, 0);
2088 if (error == 0) {
2089 MNT_ILOCK(mp);
2090 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2091 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2092 MNT_IUNLOCK(mp);
2093 mtx_lock(&mountlist_mtx);
2094 } else {
2095 printf("suspend of %s failed, error %d\n",
2096 mp->mnt_stat.f_mntonname, error);
2097 mtx_lock(&mountlist_mtx);
2098 vfs_unbusy(mp);
2099 }
2100 }
2101 mtx_unlock(&mountlist_mtx);
2102 }
2103
2104 void
2105 resume_all_fs(void)
2106 {
2107 struct mount *mp;
2108
2109 mtx_lock(&mountlist_mtx);
2110 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2111 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
2112 continue;
2113 mtx_unlock(&mountlist_mtx);
2114 MNT_ILOCK(mp);
2115 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
2116 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
2117 MNT_IUNLOCK(mp);
2118 vfs_write_resume(mp, 0);
2119 mtx_lock(&mountlist_mtx);
2120 vfs_unbusy(mp);
2121 }
2122 mtx_unlock(&mountlist_mtx);
2123 }
Cache object: 5d2dd328feb9b115e60cca542501a5ac
|