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/param.h>
40 #include <sys/conf.h>
41 #include <sys/smp.h>
42 #include <sys/devctl.h>
43 #include <sys/eventhandler.h>
44 #include <sys/fcntl.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/ktr.h>
48 #include <sys/libkern.h>
49 #include <sys/limits.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/filedesc.h>
57 #include <sys/reboot.h>
58 #include <sys/sbuf.h>
59 #include <sys/stdarg.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysproto.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/systm.h>
65 #include <sys/taskqueue.h>
66 #include <sys/vnode.h>
67 #include <vm/uma.h>
68
69 #include <geom/geom.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, bool jail_export,
78 struct vfsoptlist **optlist);
79 static void free_mntarg(struct mntarg *ma);
80
81 static int usermount = 0;
82 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
83 "Unprivileged users may mount and unmount file systems");
84
85 static bool default_autoro = false;
86 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
87 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
88
89 static bool recursive_forced_unmount = false;
90 SYSCTL_BOOL(_vfs, OID_AUTO, recursive_forced_unmount, CTLFLAG_RW,
91 &recursive_forced_unmount, 0, "Recursively unmount stacked upper mounts"
92 " when a file system is forcibly unmounted");
93
94 static SYSCTL_NODE(_vfs, OID_AUTO, deferred_unmount,
95 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "deferred unmount controls");
96
97 static unsigned int deferred_unmount_retry_limit = 10;
98 SYSCTL_UINT(_vfs_deferred_unmount, OID_AUTO, retry_limit, CTLFLAG_RW,
99 &deferred_unmount_retry_limit, 0,
100 "Maximum number of retries for deferred unmount failure");
101
102 static int deferred_unmount_retry_delay_hz;
103 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, retry_delay_hz, CTLFLAG_RW,
104 &deferred_unmount_retry_delay_hz, 0,
105 "Delay in units of [1/kern.hz]s when retrying a failed deferred unmount");
106
107 static int deferred_unmount_total_retries = 0;
108 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, total_retries, CTLFLAG_RD,
109 &deferred_unmount_total_retries, 0,
110 "Total number of retried deferred unmounts");
111
112 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
113 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
114 static uma_zone_t mount_zone;
115
116 /* List of mounted filesystems. */
117 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
118
119 /* For any iteration/modification of mountlist */
120 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
121
122 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
123 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
124
125 static void vfs_deferred_unmount(void *arg, int pending);
126 static struct timeout_task deferred_unmount_task;
127 static struct mtx deferred_unmount_lock;
128 MTX_SYSINIT(deferred_unmount, &deferred_unmount_lock, "deferred_unmount",
129 MTX_DEF);
130 static STAILQ_HEAD(, mount) deferred_unmount_list =
131 STAILQ_HEAD_INITIALIZER(deferred_unmount_list);
132 TASKQUEUE_DEFINE_THREAD(deferred_unmount);
133
134 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
135
136 /*
137 * Global opts, taken by all filesystems
138 */
139 static const char *global_opts[] = {
140 "errmsg",
141 "fstype",
142 "fspath",
143 "ro",
144 "rw",
145 "nosuid",
146 "noexec",
147 NULL
148 };
149
150 static int
mount_init(void * mem,int size,int flags)151 mount_init(void *mem, int size, int flags)
152 {
153 struct mount *mp;
154
155 mp = (struct mount *)mem;
156 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
157 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
158 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
159 lockinit(&mp->mnt_renamelock, PVFS, "rename", 0, 0);
160 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
161 mp->mnt_ref = 0;
162 mp->mnt_vfs_ops = 1;
163 mp->mnt_rootvnode = NULL;
164 return (0);
165 }
166
167 static void
mount_fini(void * mem,int size)168 mount_fini(void *mem, int size)
169 {
170 struct mount *mp;
171
172 mp = (struct mount *)mem;
173 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
174 lockdestroy(&mp->mnt_renamelock);
175 lockdestroy(&mp->mnt_explock);
176 mtx_destroy(&mp->mnt_listmtx);
177 mtx_destroy(&mp->mnt_mtx);
178 }
179
180 static void
vfs_mount_init(void * dummy __unused)181 vfs_mount_init(void *dummy __unused)
182 {
183 TIMEOUT_TASK_INIT(taskqueue_deferred_unmount, &deferred_unmount_task,
184 0, vfs_deferred_unmount, NULL);
185 deferred_unmount_retry_delay_hz = hz;
186 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
187 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
188 mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
189 }
190 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
191
192 /*
193 * ---------------------------------------------------------------------
194 * Functions for building and sanitizing the mount options
195 */
196
197 /* Remove one mount option. */
198 static void
vfs_freeopt(struct vfsoptlist * opts,struct vfsopt * opt)199 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
200 {
201
202 TAILQ_REMOVE(opts, opt, link);
203 free(opt->name, M_MOUNT);
204 if (opt->value != NULL)
205 free(opt->value, M_MOUNT);
206 free(opt, M_MOUNT);
207 }
208
209 /* Release all resources related to the mount options. */
210 void
vfs_freeopts(struct vfsoptlist * opts)211 vfs_freeopts(struct vfsoptlist *opts)
212 {
213 struct vfsopt *opt;
214
215 while (!TAILQ_EMPTY(opts)) {
216 opt = TAILQ_FIRST(opts);
217 vfs_freeopt(opts, opt);
218 }
219 free(opts, M_MOUNT);
220 }
221
222 void
vfs_deleteopt(struct vfsoptlist * opts,const char * name)223 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
224 {
225 struct vfsopt *opt, *temp;
226
227 if (opts == NULL)
228 return;
229 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
230 if (strcmp(opt->name, name) == 0)
231 vfs_freeopt(opts, opt);
232 }
233 }
234
235 static int
vfs_isopt_ro(const char * opt)236 vfs_isopt_ro(const char *opt)
237 {
238
239 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
240 strcmp(opt, "norw") == 0)
241 return (1);
242 return (0);
243 }
244
245 static int
vfs_isopt_rw(const char * opt)246 vfs_isopt_rw(const char *opt)
247 {
248
249 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
250 return (1);
251 return (0);
252 }
253
254 /*
255 * Check if options are equal (with or without the "no" prefix).
256 */
257 static int
vfs_equalopts(const char * opt1,const char * opt2)258 vfs_equalopts(const char *opt1, const char *opt2)
259 {
260 char *p;
261
262 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
263 if (strcmp(opt1, opt2) == 0)
264 return (1);
265 /* "noopt" vs. "opt" */
266 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
267 return (1);
268 /* "opt" vs. "noopt" */
269 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
270 return (1);
271 while ((p = strchr(opt1, '.')) != NULL &&
272 !strncmp(opt1, opt2, ++p - opt1)) {
273 opt2 += p - opt1;
274 opt1 = p;
275 /* "foo.noopt" vs. "foo.opt" */
276 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
277 return (1);
278 /* "foo.opt" vs. "foo.noopt" */
279 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
280 return (1);
281 }
282 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */
283 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
284 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
285 return (1);
286 return (0);
287 }
288
289 /*
290 * If a mount option is specified several times,
291 * (with or without the "no" prefix) only keep
292 * the last occurrence of it.
293 */
294 static void
vfs_sanitizeopts(struct vfsoptlist * opts)295 vfs_sanitizeopts(struct vfsoptlist *opts)
296 {
297 struct vfsopt *opt, *opt2, *tmp;
298
299 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
300 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
301 while (opt2 != NULL) {
302 if (vfs_equalopts(opt->name, opt2->name)) {
303 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
304 vfs_freeopt(opts, opt2);
305 opt2 = tmp;
306 } else {
307 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
308 }
309 }
310 }
311 }
312
313 /*
314 * Build a linked list of mount options from a struct uio.
315 */
316 int
vfs_buildopts(struct uio * auio,struct vfsoptlist ** options)317 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
318 {
319 struct vfsoptlist *opts;
320 struct vfsopt *opt;
321 size_t memused, namelen, optlen;
322 unsigned int i, iovcnt;
323 int error;
324
325 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
326 TAILQ_INIT(opts);
327 memused = 0;
328 iovcnt = auio->uio_iovcnt;
329 for (i = 0; i < iovcnt; i += 2) {
330 namelen = auio->uio_iov[i].iov_len;
331 optlen = auio->uio_iov[i + 1].iov_len;
332 memused += sizeof(struct vfsopt) + optlen + namelen;
333 /*
334 * Avoid consuming too much memory, and attempts to overflow
335 * memused.
336 */
337 if (memused > VFS_MOUNTARG_SIZE_MAX ||
338 optlen > VFS_MOUNTARG_SIZE_MAX ||
339 namelen > VFS_MOUNTARG_SIZE_MAX) {
340 error = EINVAL;
341 goto bad;
342 }
343
344 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
345 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
346 opt->value = NULL;
347 opt->len = 0;
348 opt->pos = i / 2;
349 opt->seen = 0;
350
351 /*
352 * Do this early, so jumps to "bad" will free the current
353 * option.
354 */
355 TAILQ_INSERT_TAIL(opts, opt, link);
356
357 if (auio->uio_segflg == UIO_SYSSPACE) {
358 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
359 } else {
360 error = copyin(auio->uio_iov[i].iov_base, opt->name,
361 namelen);
362 if (error)
363 goto bad;
364 }
365 /* Ensure names are null-terminated strings. */
366 if (namelen == 0 || opt->name[namelen - 1] != '\0') {
367 error = EINVAL;
368 goto bad;
369 }
370 if (optlen != 0) {
371 opt->len = optlen;
372 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
373 if (auio->uio_segflg == UIO_SYSSPACE) {
374 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
375 optlen);
376 } else {
377 error = copyin(auio->uio_iov[i + 1].iov_base,
378 opt->value, optlen);
379 if (error)
380 goto bad;
381 }
382 }
383 }
384 vfs_sanitizeopts(opts);
385 *options = opts;
386 return (0);
387 bad:
388 vfs_freeopts(opts);
389 return (error);
390 }
391
392 /*
393 * Merge the old mount options with the new ones passed
394 * in the MNT_UPDATE case.
395 *
396 * XXX: This function will keep a "nofoo" option in the new
397 * options. E.g, if the option's canonical name is "foo",
398 * "nofoo" ends up in the mount point's active options.
399 */
400 static void
vfs_mergeopts(struct vfsoptlist * toopts,struct vfsoptlist * oldopts)401 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
402 {
403 struct vfsopt *opt, *new;
404
405 TAILQ_FOREACH(opt, oldopts, link) {
406 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
407 new->name = strdup(opt->name, M_MOUNT);
408 if (opt->len != 0) {
409 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
410 bcopy(opt->value, new->value, opt->len);
411 } else
412 new->value = NULL;
413 new->len = opt->len;
414 new->seen = opt->seen;
415 TAILQ_INSERT_HEAD(toopts, new, link);
416 }
417 vfs_sanitizeopts(toopts);
418 }
419
420 /*
421 * Mount a filesystem.
422 */
423 #ifndef _SYS_SYSPROTO_H_
424 struct nmount_args {
425 struct iovec *iovp;
426 unsigned int iovcnt;
427 int flags;
428 };
429 #endif
430 int
sys_nmount(struct thread * td,struct nmount_args * uap)431 sys_nmount(struct thread *td, struct nmount_args *uap)
432 {
433 struct uio *auio;
434 int error;
435 u_int iovcnt;
436 uint64_t flags;
437
438 /*
439 * Mount flags are now 64-bits. On 32-bit archtectures only
440 * 32-bits are passed in, but from here on everything handles
441 * 64-bit flags correctly.
442 */
443 flags = uap->flags;
444
445 AUDIT_ARG_FFLAGS(flags);
446 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
447 uap->iovp, uap->iovcnt, flags);
448
449 /*
450 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
451 * userspace to set this flag, but we must filter it out if we want
452 * MNT_UPDATE on the root file system to work.
453 * MNT_ROOTFS should only be set by the kernel when mounting its
454 * root file system.
455 */
456 flags &= ~MNT_ROOTFS;
457
458 iovcnt = uap->iovcnt;
459 /*
460 * Check that we have an even number of iovec's
461 * and that we have at least two options.
462 */
463 if ((iovcnt & 1) || (iovcnt < 4)) {
464 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
465 uap->iovcnt);
466 return (EINVAL);
467 }
468
469 error = copyinuio(uap->iovp, iovcnt, &auio);
470 if (error) {
471 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
472 __func__, error);
473 return (error);
474 }
475 error = vfs_donmount(td, flags, auio);
476
477 freeuio(auio);
478 return (error);
479 }
480
481 /*
482 * ---------------------------------------------------------------------
483 * Various utility functions
484 */
485
486 /*
487 * Get a reference on a mount point from a vnode.
488 *
489 * The vnode is allowed to be passed unlocked and race against dooming. Note in
490 * such case there are no guarantees the referenced mount point will still be
491 * associated with it after the function returns.
492 */
493 struct mount *
vfs_ref_from_vp(struct vnode * vp)494 vfs_ref_from_vp(struct vnode *vp)
495 {
496 struct mount *mp;
497 struct mount_pcpu *mpcpu;
498
499 mp = atomic_load_ptr(&vp->v_mount);
500 if (__predict_false(mp == NULL)) {
501 return (mp);
502 }
503 if (vfs_op_thread_enter(mp, mpcpu)) {
504 if (__predict_true(mp == vp->v_mount)) {
505 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
506 vfs_op_thread_exit(mp, mpcpu);
507 } else {
508 vfs_op_thread_exit(mp, mpcpu);
509 mp = NULL;
510 }
511 } else {
512 MNT_ILOCK(mp);
513 if (mp == vp->v_mount) {
514 MNT_REF(mp);
515 MNT_IUNLOCK(mp);
516 } else {
517 MNT_IUNLOCK(mp);
518 mp = NULL;
519 }
520 }
521 return (mp);
522 }
523
524 void
vfs_ref(struct mount * mp)525 vfs_ref(struct mount *mp)
526 {
527 struct mount_pcpu *mpcpu;
528
529 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
530 if (vfs_op_thread_enter(mp, mpcpu)) {
531 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
532 vfs_op_thread_exit(mp, mpcpu);
533 return;
534 }
535
536 MNT_ILOCK(mp);
537 MNT_REF(mp);
538 MNT_IUNLOCK(mp);
539 }
540
541 /*
542 * Register ump as an upper mount of the mount associated with
543 * vnode vp. This registration will be tracked through
544 * mount_upper_node upper, which should be allocated by the
545 * caller and stored in per-mount data associated with mp.
546 *
547 * If successful, this function will return the mount associated
548 * with vp, and will ensure that it cannot be unmounted until
549 * ump has been unregistered as one of its upper mounts.
550 *
551 * Upon failure this function will return NULL.
552 */
553 struct mount *
vfs_register_upper_from_vp(struct vnode * vp,struct mount * ump,struct mount_upper_node * upper)554 vfs_register_upper_from_vp(struct vnode *vp, struct mount *ump,
555 struct mount_upper_node *upper)
556 {
557 struct mount *mp;
558
559 mp = atomic_load_ptr(&vp->v_mount);
560 if (mp == NULL)
561 return (NULL);
562 MNT_ILOCK(mp);
563 if (mp != vp->v_mount ||
564 ((mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_RECURSE)) != 0)) {
565 MNT_IUNLOCK(mp);
566 return (NULL);
567 }
568 KASSERT(ump != mp, ("upper and lower mounts are identical"));
569 upper->mp = ump;
570 MNT_REF(mp);
571 TAILQ_INSERT_TAIL(&mp->mnt_uppers, upper, mnt_upper_link);
572 MNT_IUNLOCK(mp);
573 return (mp);
574 }
575
576 /*
577 * Register upper mount ump to receive vnode unlink/reclaim
578 * notifications from lower mount mp. This registration will
579 * be tracked through mount_upper_node upper, which should be
580 * allocated by the caller and stored in per-mount data
581 * associated with mp.
582 *
583 * ump must already be registered as an upper mount of mp
584 * through a call to vfs_register_upper_from_vp().
585 */
586 void
vfs_register_for_notification(struct mount * mp,struct mount * ump,struct mount_upper_node * upper)587 vfs_register_for_notification(struct mount *mp, struct mount *ump,
588 struct mount_upper_node *upper)
589 {
590 upper->mp = ump;
591 MNT_ILOCK(mp);
592 TAILQ_INSERT_TAIL(&mp->mnt_notify, upper, mnt_upper_link);
593 MNT_IUNLOCK(mp);
594 }
595
596 static void
vfs_drain_upper_locked(struct mount * mp)597 vfs_drain_upper_locked(struct mount *mp)
598 {
599 mtx_assert(MNT_MTX(mp), MA_OWNED);
600 while (mp->mnt_upper_pending != 0) {
601 mp->mnt_kern_flag |= MNTK_UPPER_WAITER;
602 msleep(&mp->mnt_uppers, MNT_MTX(mp), 0, "mntupw", 0);
603 }
604 }
605
606 /*
607 * Undo a previous call to vfs_register_for_notification().
608 * The mount represented by upper must be currently registered
609 * as an upper mount for mp.
610 */
611 void
vfs_unregister_for_notification(struct mount * mp,struct mount_upper_node * upper)612 vfs_unregister_for_notification(struct mount *mp,
613 struct mount_upper_node *upper)
614 {
615 MNT_ILOCK(mp);
616 vfs_drain_upper_locked(mp);
617 TAILQ_REMOVE(&mp->mnt_notify, upper, mnt_upper_link);
618 MNT_IUNLOCK(mp);
619 }
620
621 /*
622 * Undo a previous call to vfs_register_upper_from_vp().
623 * This must be done before mp can be unmounted.
624 */
625 void
vfs_unregister_upper(struct mount * mp,struct mount_upper_node * upper)626 vfs_unregister_upper(struct mount *mp, struct mount_upper_node *upper)
627 {
628 MNT_ILOCK(mp);
629 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
630 ("registered upper with pending unmount"));
631 vfs_drain_upper_locked(mp);
632 TAILQ_REMOVE(&mp->mnt_uppers, upper, mnt_upper_link);
633 if ((mp->mnt_kern_flag & MNTK_TASKQUEUE_WAITER) != 0 &&
634 TAILQ_EMPTY(&mp->mnt_uppers)) {
635 mp->mnt_kern_flag &= ~MNTK_TASKQUEUE_WAITER;
636 wakeup(&mp->mnt_taskqueue_link);
637 }
638 MNT_REL(mp);
639 MNT_IUNLOCK(mp);
640 }
641
642 void
vfs_rel(struct mount * mp)643 vfs_rel(struct mount *mp)
644 {
645 struct mount_pcpu *mpcpu;
646
647 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
648 if (vfs_op_thread_enter(mp, mpcpu)) {
649 vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
650 vfs_op_thread_exit(mp, mpcpu);
651 return;
652 }
653
654 MNT_ILOCK(mp);
655 MNT_REL(mp);
656 MNT_IUNLOCK(mp);
657 }
658
659 /*
660 * Allocate and initialize the mount point struct.
661 */
662 struct mount *
vfs_mount_alloc(struct vnode * vp,struct vfsconf * vfsp,const char * fspath,struct ucred * cred)663 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
664 struct ucred *cred)
665 {
666 struct mount *mp;
667
668 mp = uma_zalloc(mount_zone, M_WAITOK);
669 bzero(&mp->mnt_startzero,
670 __rangeof(struct mount, mnt_startzero, mnt_endzero));
671 mp->mnt_kern_flag = 0;
672 mp->mnt_flag = 0;
673 mp->mnt_rootvnode = NULL;
674 mp->mnt_vnodecovered = NULL;
675 mp->mnt_op = NULL;
676 mp->mnt_vfc = NULL;
677 TAILQ_INIT(&mp->mnt_nvnodelist);
678 mp->mnt_nvnodelistsize = 0;
679 TAILQ_INIT(&mp->mnt_lazyvnodelist);
680 mp->mnt_lazyvnodelistsize = 0;
681 MPPASS(mp->mnt_ref == 0 && mp->mnt_lockref == 0 &&
682 mp->mnt_writeopcount == 0, mp);
683 MPASSERT(mp->mnt_vfs_ops == 1, mp,
684 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
685 (void) vfs_busy(mp, MBF_NOWAIT);
686 mp->mnt_op = vfsp->vfc_vfsops;
687 mp->mnt_vfc = vfsp;
688 mp->mnt_stat.f_type = vfsp->vfc_typenum;
689 mp->mnt_gen++;
690 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
691 mp->mnt_vnodecovered = vp;
692 mp->mnt_cred = crdup(cred);
693 mp->mnt_stat.f_owner = cred->cr_uid;
694 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
695 mp->mnt_iosize_max = DFLTPHYS;
696 #ifdef MAC
697 mac_mount_init(mp);
698 mac_mount_create(cred, mp);
699 #endif
700 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
701 mp->mnt_upper_pending = 0;
702 TAILQ_INIT(&mp->mnt_uppers);
703 TAILQ_INIT(&mp->mnt_notify);
704 mp->mnt_taskqueue_flags = 0;
705 mp->mnt_unmount_retries = 0;
706 return (mp);
707 }
708
709 /*
710 * Destroy the mount struct previously allocated by vfs_mount_alloc().
711 */
712 void
vfs_mount_destroy(struct mount * mp)713 vfs_mount_destroy(struct mount *mp)
714 {
715
716 MPPASS(mp->mnt_vfs_ops != 0, mp);
717
718 vfs_assert_mount_counters(mp);
719
720 MNT_ILOCK(mp);
721 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
722 if (mp->mnt_kern_flag & MNTK_MWAIT) {
723 mp->mnt_kern_flag &= ~MNTK_MWAIT;
724 wakeup(mp);
725 }
726 while (mp->mnt_ref)
727 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
728 KASSERT(mp->mnt_ref == 0,
729 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
730 __FILE__, __LINE__));
731 MPPASS(mp->mnt_writeopcount == 0, mp);
732 MPPASS(mp->mnt_secondary_writes == 0, mp);
733 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
734 struct vnode *vp;
735
736 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
737 vn_printf(vp, "dangling vnode ");
738 panic("unmount: dangling vnode");
739 }
740 KASSERT(mp->mnt_upper_pending == 0, ("mnt_upper_pending"));
741 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
742 KASSERT(TAILQ_EMPTY(&mp->mnt_notify), ("mnt_notify"));
743 MPPASS(mp->mnt_nvnodelistsize == 0, mp);
744 MPPASS(mp->mnt_lazyvnodelistsize == 0, mp);
745 MPPASS(mp->mnt_lockref == 0, mp);
746 MNT_IUNLOCK(mp);
747
748 MPASSERT(mp->mnt_vfs_ops == 1, mp,
749 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
750
751 MPASSERT(mp->mnt_rootvnode == NULL, mp,
752 ("mount point still has a root vnode %p", mp->mnt_rootvnode));
753
754 if (mp->mnt_vnodecovered != NULL)
755 vrele(mp->mnt_vnodecovered);
756 #ifdef MAC
757 mac_mount_destroy(mp);
758 #endif
759 if (mp->mnt_opt != NULL)
760 vfs_freeopts(mp->mnt_opt);
761 if (mp->mnt_exjail != NULL) {
762 atomic_subtract_int(&mp->mnt_exjail->cr_prison->pr_exportcnt,
763 1);
764 crfree(mp->mnt_exjail);
765 }
766 if (mp->mnt_export != NULL) {
767 vfs_free_addrlist(mp->mnt_export);
768 free(mp->mnt_export, M_MOUNT);
769 }
770 vfsconf_lock();
771 mp->mnt_vfc->vfc_refcount--;
772 vfsconf_unlock();
773 crfree(mp->mnt_cred);
774 uma_zfree(mount_zone, mp);
775 }
776
777 static bool
vfs_should_downgrade_to_ro_mount(uint64_t fsflags,int error)778 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
779 {
780 /* This is an upgrade of an exisiting mount. */
781 if ((fsflags & MNT_UPDATE) != 0)
782 return (false);
783 /* This is already an R/O mount. */
784 if ((fsflags & MNT_RDONLY) != 0)
785 return (false);
786
787 switch (error) {
788 case ENODEV: /* generic, geom, ... */
789 case EACCES: /* cam/scsi, ... */
790 case EROFS: /* md, mmcsd, ... */
791 /*
792 * These errors can be returned by the storage layer to signal
793 * that the media is read-only. No harm in the R/O mount
794 * attempt if the error was returned for some other reason.
795 */
796 return (true);
797 default:
798 return (false);
799 }
800 }
801
802 int
vfs_donmount(struct thread * td,uint64_t fsflags,struct uio * fsoptions)803 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
804 {
805 struct vfsoptlist *optlist;
806 struct vfsopt *opt, *tmp_opt;
807 char *fstype, *fspath, *errmsg;
808 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
809 bool autoro, has_nonexport, jail_export;
810
811 errmsg = fspath = NULL;
812 errmsg_len = fspathlen = 0;
813 errmsg_pos = -1;
814 autoro = default_autoro;
815
816 error = vfs_buildopts(fsoptions, &optlist);
817 if (error)
818 return (error);
819
820 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
821 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
822
823 /*
824 * We need these two options before the others,
825 * and they are mandatory for any filesystem.
826 * Ensure they are NUL terminated as well.
827 */
828 fstypelen = 0;
829 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
830 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
831 error = EINVAL;
832 if (errmsg != NULL)
833 strncpy(errmsg, "Invalid fstype", errmsg_len);
834 goto bail;
835 }
836 fspathlen = 0;
837 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
838 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
839 error = EINVAL;
840 if (errmsg != NULL)
841 strncpy(errmsg, "Invalid fspath", errmsg_len);
842 goto bail;
843 }
844
845 /*
846 * Check to see that "export" is only used with the "update", "fstype",
847 * "fspath", "from" and "errmsg" options when in a vnet jail.
848 * These are the ones used to set/update exports by mountd(8).
849 * If only the above options are set in a jail that can run mountd(8),
850 * then the jail_export argument of vfs_domount() will be true.
851 * When jail_export is true, the vfs_suser() check does not cause
852 * failure, but limits the update to exports only.
853 * This allows mountd(8) running within the vnet jail
854 * to export file systems visible within the jail, but
855 * mounted outside of the jail.
856 */
857 /*
858 * We need to see if we have the "update" option
859 * before we call vfs_domount(), since vfs_domount() has special
860 * logic based on MNT_UPDATE. This is very important
861 * when we want to update the root filesystem.
862 */
863 has_nonexport = false;
864 jail_export = false;
865 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
866 int do_freeopt = 0;
867
868 if (jailed(td->td_ucred) &&
869 strcmp(opt->name, "export") != 0 &&
870 strcmp(opt->name, "update") != 0 &&
871 strcmp(opt->name, "fstype") != 0 &&
872 strcmp(opt->name, "fspath") != 0 &&
873 strcmp(opt->name, "from") != 0 &&
874 strcmp(opt->name, "errmsg") != 0)
875 has_nonexport = true;
876 if (strcmp(opt->name, "update") == 0) {
877 fsflags |= MNT_UPDATE;
878 do_freeopt = 1;
879 }
880 else if (strcmp(opt->name, "async") == 0)
881 fsflags |= MNT_ASYNC;
882 else if (strcmp(opt->name, "force") == 0) {
883 fsflags |= MNT_FORCE;
884 do_freeopt = 1;
885 }
886 else if (strcmp(opt->name, "reload") == 0) {
887 fsflags |= MNT_RELOAD;
888 do_freeopt = 1;
889 }
890 else if (strcmp(opt->name, "multilabel") == 0)
891 fsflags |= MNT_MULTILABEL;
892 else if (strcmp(opt->name, "noasync") == 0)
893 fsflags &= ~MNT_ASYNC;
894 else if (strcmp(opt->name, "noatime") == 0)
895 fsflags |= MNT_NOATIME;
896 else if (strcmp(opt->name, "atime") == 0) {
897 free(opt->name, M_MOUNT);
898 opt->name = strdup("nonoatime", M_MOUNT);
899 }
900 else if (strcmp(opt->name, "noclusterr") == 0)
901 fsflags |= MNT_NOCLUSTERR;
902 else if (strcmp(opt->name, "clusterr") == 0) {
903 free(opt->name, M_MOUNT);
904 opt->name = strdup("nonoclusterr", M_MOUNT);
905 }
906 else if (strcmp(opt->name, "noclusterw") == 0)
907 fsflags |= MNT_NOCLUSTERW;
908 else if (strcmp(opt->name, "clusterw") == 0) {
909 free(opt->name, M_MOUNT);
910 opt->name = strdup("nonoclusterw", M_MOUNT);
911 }
912 else if (strcmp(opt->name, "noexec") == 0)
913 fsflags |= MNT_NOEXEC;
914 else if (strcmp(opt->name, "exec") == 0) {
915 free(opt->name, M_MOUNT);
916 opt->name = strdup("nonoexec", M_MOUNT);
917 }
918 else if (strcmp(opt->name, "nosuid") == 0)
919 fsflags |= MNT_NOSUID;
920 else if (strcmp(opt->name, "suid") == 0) {
921 free(opt->name, M_MOUNT);
922 opt->name = strdup("nonosuid", M_MOUNT);
923 }
924 else if (strcmp(opt->name, "nosymfollow") == 0)
925 fsflags |= MNT_NOSYMFOLLOW;
926 else if (strcmp(opt->name, "symfollow") == 0) {
927 free(opt->name, M_MOUNT);
928 opt->name = strdup("nonosymfollow", M_MOUNT);
929 }
930 else if (strcmp(opt->name, "noro") == 0) {
931 fsflags &= ~MNT_RDONLY;
932 autoro = false;
933 }
934 else if (strcmp(opt->name, "rw") == 0) {
935 fsflags &= ~MNT_RDONLY;
936 autoro = false;
937 }
938 else if (strcmp(opt->name, "ro") == 0) {
939 fsflags |= MNT_RDONLY;
940 autoro = false;
941 }
942 else if (strcmp(opt->name, "rdonly") == 0) {
943 free(opt->name, M_MOUNT);
944 opt->name = strdup("ro", M_MOUNT);
945 fsflags |= MNT_RDONLY;
946 autoro = false;
947 }
948 else if (strcmp(opt->name, "autoro") == 0) {
949 do_freeopt = 1;
950 autoro = true;
951 }
952 else if (strcmp(opt->name, "suiddir") == 0)
953 fsflags |= MNT_SUIDDIR;
954 else if (strcmp(opt->name, "sync") == 0)
955 fsflags |= MNT_SYNCHRONOUS;
956 else if (strcmp(opt->name, "union") == 0)
957 fsflags |= MNT_UNION;
958 else if (strcmp(opt->name, "export") == 0) {
959 fsflags |= MNT_EXPORTED;
960 jail_export = true;
961 } else if (strcmp(opt->name, "automounted") == 0) {
962 fsflags |= MNT_AUTOMOUNTED;
963 do_freeopt = 1;
964 } else if (strcmp(opt->name, "nocover") == 0) {
965 fsflags |= MNT_NOCOVER;
966 do_freeopt = 1;
967 } else if (strcmp(opt->name, "cover") == 0) {
968 fsflags &= ~MNT_NOCOVER;
969 do_freeopt = 1;
970 } else if (strcmp(opt->name, "emptydir") == 0) {
971 fsflags |= MNT_EMPTYDIR;
972 do_freeopt = 1;
973 } else if (strcmp(opt->name, "noemptydir") == 0) {
974 fsflags &= ~MNT_EMPTYDIR;
975 do_freeopt = 1;
976 }
977 if (do_freeopt)
978 vfs_freeopt(optlist, opt);
979 }
980
981 /*
982 * Be ultra-paranoid about making sure the type and fspath
983 * variables will fit in our mp buffers, including the
984 * terminating NUL.
985 */
986 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
987 error = ENAMETOOLONG;
988 goto bail;
989 }
990
991 /*
992 * If has_nonexport is true or the caller is not running within a
993 * vnet prison that can run mountd(8), set jail_export false.
994 */
995 if (has_nonexport || !jailed(td->td_ucred) ||
996 !prison_check_nfsd(td->td_ucred))
997 jail_export = false;
998
999 error = vfs_domount(td, fstype, fspath, fsflags, jail_export, &optlist);
1000 if (error == ENODEV) {
1001 error = EINVAL;
1002 if (errmsg != NULL)
1003 strncpy(errmsg, "Invalid fstype", errmsg_len);
1004 goto bail;
1005 }
1006
1007 /*
1008 * See if we can mount in the read-only mode if the error code suggests
1009 * that it could be possible and the mount options allow for that.
1010 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
1011 * overridden by "autoro".
1012 */
1013 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
1014 printf("%s: R/W mount failed, possibly R/O media,"
1015 " trying R/O mount\n", __func__);
1016 fsflags |= MNT_RDONLY;
1017 error = vfs_domount(td, fstype, fspath, fsflags, jail_export,
1018 &optlist);
1019 }
1020 bail:
1021 /* copyout the errmsg */
1022 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
1023 && errmsg_len > 0 && errmsg != NULL) {
1024 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
1025 bcopy(errmsg,
1026 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
1027 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1028 } else {
1029 (void)copyout(errmsg,
1030 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
1031 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1032 }
1033 }
1034
1035 if (optlist != NULL)
1036 vfs_freeopts(optlist);
1037 return (error);
1038 }
1039
1040 /*
1041 * Old mount API.
1042 */
1043 #ifndef _SYS_SYSPROTO_H_
1044 struct mount_args {
1045 char *type;
1046 char *path;
1047 int flags;
1048 caddr_t data;
1049 };
1050 #endif
1051 /* ARGSUSED */
1052 int
sys_mount(struct thread * td,struct mount_args * uap)1053 sys_mount(struct thread *td, struct mount_args *uap)
1054 {
1055 char *fstype;
1056 struct vfsconf *vfsp = NULL;
1057 struct mntarg *ma = NULL;
1058 uint64_t flags;
1059 int error;
1060
1061 /*
1062 * Mount flags are now 64-bits. On 32-bit architectures only
1063 * 32-bits are passed in, but from here on everything handles
1064 * 64-bit flags correctly.
1065 */
1066 flags = uap->flags;
1067
1068 AUDIT_ARG_FFLAGS(flags);
1069
1070 /*
1071 * Filter out MNT_ROOTFS. We do not want clients of mount() in
1072 * userspace to set this flag, but we must filter it out if we want
1073 * MNT_UPDATE on the root file system to work.
1074 * MNT_ROOTFS should only be set by the kernel when mounting its
1075 * root file system.
1076 */
1077 flags &= ~MNT_ROOTFS;
1078
1079 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
1080 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
1081 if (error) {
1082 free(fstype, M_TEMP);
1083 return (error);
1084 }
1085
1086 AUDIT_ARG_TEXT(fstype);
1087 vfsp = vfs_byname_kld(fstype, td, &error);
1088 free(fstype, M_TEMP);
1089 if (vfsp == NULL)
1090 return (EINVAL);
1091 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
1092 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
1093 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
1094 vfsp->vfc_vfsops->vfs_cmount == NULL))
1095 return (EOPNOTSUPP);
1096
1097 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
1098 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
1099 ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
1100 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
1101 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
1102
1103 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
1104 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
1105 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
1106 }
1107
1108 /*
1109 * vfs_domount_first(): first file system mount (not update)
1110 */
1111 static int
vfs_domount_first(struct thread * td,struct vfsconf * vfsp,char * fspath,struct vnode * vp,uint64_t fsflags,struct vfsoptlist ** optlist)1112 vfs_domount_first(
1113 struct thread *td, /* Calling thread. */
1114 struct vfsconf *vfsp, /* File system type. */
1115 char *fspath, /* Mount path. */
1116 struct vnode *vp, /* Vnode to be covered. */
1117 uint64_t fsflags, /* Flags common to all filesystems. */
1118 struct vfsoptlist **optlist /* Options local to the filesystem. */
1119 )
1120 {
1121 struct vattr va;
1122 struct mount *mp;
1123 struct vnode *newdp, *rootvp;
1124 int error, error1;
1125 bool unmounted;
1126
1127 ASSERT_VOP_ELOCKED(vp, __func__);
1128 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
1129
1130 /*
1131 * If the jail of the calling thread lacks permission for this type of
1132 * file system, or is trying to cover its own root, deny immediately.
1133 */
1134 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
1135 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
1136 vput(vp);
1137 vfs_unref_vfsconf(vfsp);
1138 return (EPERM);
1139 }
1140
1141 /*
1142 * If the user is not root, ensure that they own the directory
1143 * onto which we are attempting to mount.
1144 */
1145 error = VOP_GETATTR(vp, &va, td->td_ucred);
1146 if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
1147 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
1148 if (error == 0)
1149 error = vinvalbuf(vp, V_SAVE, 0, 0);
1150 if (vfsp->vfc_flags & VFCF_FILEMOUNT) {
1151 if (error == 0 && vp->v_type != VDIR && vp->v_type != VREG)
1152 error = EINVAL;
1153 /*
1154 * For file mounts, ensure that there is only one hardlink to the file.
1155 */
1156 if (error == 0 && vp->v_type == VREG && va.va_nlink != 1)
1157 error = EINVAL;
1158 } else {
1159 if (error == 0 && vp->v_type != VDIR)
1160 error = ENOTDIR;
1161 }
1162 if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
1163 error = vn_dir_check_empty(vp);
1164 if (error == 0) {
1165 VI_LOCK(vp);
1166 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
1167 vp->v_iflag |= VI_MOUNT;
1168 else
1169 error = EBUSY;
1170 VI_UNLOCK(vp);
1171 }
1172 if (error != 0) {
1173 vput(vp);
1174 vfs_unref_vfsconf(vfsp);
1175 return (error);
1176 }
1177 vn_seqc_write_begin(vp);
1178 VOP_UNLOCK(vp);
1179
1180 /* Allocate and initialize the filesystem. */
1181 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
1182 /* XXXMAC: pass to vfs_mount_alloc? */
1183 mp->mnt_optnew = *optlist;
1184 /* Set the mount level flags. */
1185 mp->mnt_flag = (fsflags &
1186 (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY | MNT_FORCE));
1187
1188 /*
1189 * Mount the filesystem.
1190 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1191 * get. No freeing of cn_pnbuf.
1192 */
1193 error1 = 0;
1194 unmounted = true;
1195 if ((error = VFS_MOUNT(mp)) != 0 ||
1196 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1197 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1198 rootvp = NULL;
1199 if (error1 != 0) {
1200 MPASS(error == 0);
1201 rootvp = vfs_cache_root_clear(mp);
1202 if (rootvp != NULL) {
1203 vhold(rootvp);
1204 vrele(rootvp);
1205 }
1206 (void)vn_start_write(NULL, &mp, V_WAIT);
1207 MNT_ILOCK(mp);
1208 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1209 MNT_IUNLOCK(mp);
1210 VFS_PURGE(mp);
1211 error = VFS_UNMOUNT(mp, 0);
1212 vn_finished_write(mp);
1213 if (error != 0) {
1214 printf(
1215 "failed post-mount (%d): rollback unmount returned %d\n",
1216 error1, error);
1217 unmounted = false;
1218 }
1219 error = error1;
1220 }
1221 vfs_unbusy(mp);
1222 mp->mnt_vnodecovered = NULL;
1223 if (unmounted) {
1224 /* XXXKIB wait for mnt_lockref drain? */
1225 vfs_mount_destroy(mp);
1226 }
1227 VI_LOCK(vp);
1228 vp->v_iflag &= ~VI_MOUNT;
1229 VI_UNLOCK(vp);
1230 if (rootvp != NULL) {
1231 vn_seqc_write_end(rootvp);
1232 vdrop(rootvp);
1233 }
1234 vn_seqc_write_end(vp);
1235 vrele(vp);
1236 return (error);
1237 }
1238 vn_seqc_write_begin(newdp);
1239 VOP_UNLOCK(newdp);
1240
1241 if (mp->mnt_opt != NULL)
1242 vfs_freeopts(mp->mnt_opt);
1243 mp->mnt_opt = mp->mnt_optnew;
1244 *optlist = NULL;
1245
1246 /*
1247 * Prevent external consumers of mount options from reading mnt_optnew.
1248 */
1249 mp->mnt_optnew = NULL;
1250
1251 MNT_ILOCK(mp);
1252 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1253 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1254 mp->mnt_kern_flag |= MNTK_ASYNC;
1255 else
1256 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1257 MNT_IUNLOCK(mp);
1258
1259 /*
1260 * VIRF_MOUNTPOINT and v_mountedhere need to be set under the
1261 * vp lock to satisfy vfs_lookup() requirements.
1262 */
1263 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1264 VI_LOCK(vp);
1265 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1266 vp->v_mountedhere = mp;
1267 VI_UNLOCK(vp);
1268 VOP_UNLOCK(vp);
1269 cache_purge(vp);
1270
1271 /*
1272 * We need to lock both vnodes.
1273 *
1274 * Use vn_lock_pair to avoid establishing an ordering between vnodes
1275 * from different filesystems.
1276 */
1277 vn_lock_pair(vp, false, LK_EXCLUSIVE, newdp, false, LK_EXCLUSIVE);
1278
1279 VI_LOCK(vp);
1280 vp->v_iflag &= ~VI_MOUNT;
1281 VI_UNLOCK(vp);
1282 /* Place the new filesystem at the end of the mount list. */
1283 mtx_lock(&mountlist_mtx);
1284 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1285 mtx_unlock(&mountlist_mtx);
1286 vfs_event_signal(NULL, VQ_MOUNT, 0);
1287 VOP_UNLOCK(vp);
1288 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1289 VOP_UNLOCK(newdp);
1290 mount_devctl_event("MOUNT", mp, false);
1291 mountcheckdirs(vp, newdp);
1292 vn_seqc_write_end(vp);
1293 vn_seqc_write_end(newdp);
1294 vrele(newdp);
1295 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1296 vfs_allocate_syncvnode(mp);
1297 vfs_op_exit(mp);
1298 vfs_unbusy(mp);
1299 return (0);
1300 }
1301
1302 /*
1303 * vfs_domount_update(): update of mounted file system
1304 */
1305 static int
vfs_domount_update(struct thread * td,struct vnode * vp,uint64_t fsflags,bool jail_export,struct vfsoptlist ** optlist)1306 vfs_domount_update(
1307 struct thread *td, /* Calling thread. */
1308 struct vnode *vp, /* Mount point vnode. */
1309 uint64_t fsflags, /* Flags common to all filesystems. */
1310 bool jail_export, /* Got export option in vnet prison. */
1311 struct vfsoptlist **optlist /* Options local to the filesystem. */
1312 )
1313 {
1314 struct export_args export;
1315 struct o2export_args o2export;
1316 struct vnode *rootvp;
1317 void *bufp;
1318 struct mount *mp;
1319 int error, export_error, i, len, fsid_up_len;
1320 uint64_t flag, mnt_union;
1321 gid_t *grps;
1322 fsid_t *fsid_up;
1323 bool vfs_suser_failed;
1324
1325 ASSERT_VOP_ELOCKED(vp, __func__);
1326 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1327 mp = vp->v_mount;
1328
1329 if ((vp->v_vflag & VV_ROOT) == 0) {
1330 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1331 == 0)
1332 error = EXDEV;
1333 else
1334 error = EINVAL;
1335 vput(vp);
1336 return (error);
1337 }
1338
1339 /*
1340 * We only allow the filesystem to be reloaded if it
1341 * is currently mounted read-only.
1342 */
1343 flag = mp->mnt_flag;
1344 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1345 vput(vp);
1346 return (EOPNOTSUPP); /* Needs translation */
1347 }
1348 /*
1349 * Only privileged root, or (if MNT_USER is set) the user that
1350 * did the original mount is permitted to update it.
1351 */
1352 /*
1353 * For the case of mountd(8) doing exports in a jail, the vfs_suser()
1354 * call does not cause failure. vfs_domount() has already checked
1355 * that "root" is doing this and vfs_suser() will fail when
1356 * the file system has been mounted outside the jail.
1357 * jail_export set true indicates that "export" is not mixed
1358 * with other options that change mount behaviour.
1359 */
1360 vfs_suser_failed = false;
1361 error = vfs_suser(mp, td);
1362 if (jail_export && error != 0) {
1363 error = 0;
1364 vfs_suser_failed = true;
1365 }
1366 if (error != 0) {
1367 vput(vp);
1368 return (error);
1369 }
1370 if (vfs_busy(mp, MBF_NOWAIT)) {
1371 vput(vp);
1372 return (EBUSY);
1373 }
1374 VI_LOCK(vp);
1375 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1376 VI_UNLOCK(vp);
1377 vfs_unbusy(mp);
1378 vput(vp);
1379 return (EBUSY);
1380 }
1381 vp->v_iflag |= VI_MOUNT;
1382 VI_UNLOCK(vp);
1383 VOP_UNLOCK(vp);
1384
1385 rootvp = NULL;
1386 vfs_op_enter(mp);
1387 vn_seqc_write_begin(vp);
1388
1389 if (vfs_getopt(*optlist, "fsid", (void **)&fsid_up,
1390 &fsid_up_len) == 0) {
1391 if (fsid_up_len != sizeof(*fsid_up)) {
1392 error = EINVAL;
1393 goto end;
1394 }
1395 if (fsidcmp(fsid_up, &mp->mnt_stat.f_fsid) != 0) {
1396 error = ENOENT;
1397 goto end;
1398 }
1399 vfs_deleteopt(*optlist, "fsid");
1400 }
1401
1402 mnt_union = 0;
1403 MNT_ILOCK(mp);
1404 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1405 MNT_IUNLOCK(mp);
1406 error = EBUSY;
1407 goto end;
1408 }
1409 if (vfs_suser_failed) {
1410 KASSERT((fsflags & (MNT_EXPORTED | MNT_UPDATE)) ==
1411 (MNT_EXPORTED | MNT_UPDATE),
1412 ("%s: jailed export did not set expected fsflags",
1413 __func__));
1414 /*
1415 * For this case, only MNT_UPDATE and
1416 * MNT_EXPORTED have been set in fsflags
1417 * by the options. Only set MNT_UPDATE,
1418 * since that is the one that would be set
1419 * when set in fsflags, below.
1420 */
1421 mp->mnt_flag |= MNT_UPDATE;
1422 } else {
1423 mp->mnt_flag &= ~MNT_UPDATEMASK;
1424 if ((mp->mnt_flag & MNT_UNION) == 0 &&
1425 (fsflags & MNT_UNION) != 0) {
1426 fsflags &= ~MNT_UNION;
1427 mnt_union = MNT_UNION;
1428 }
1429 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1430 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1431 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1432 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1433 }
1434 rootvp = vfs_cache_root_clear(mp);
1435 MNT_IUNLOCK(mp);
1436 mp->mnt_optnew = *optlist;
1437 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1438
1439 /*
1440 * Mount the filesystem.
1441 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1442 * get. No freeing of cn_pnbuf.
1443 */
1444 /*
1445 * For the case of mountd(8) doing exports from within a vnet jail,
1446 * "from" is typically not set correctly such that VFS_MOUNT() will
1447 * return ENOENT. It is not obvious that VFS_MOUNT() ever needs to be
1448 * called when mountd is doing exports, but this check only applies to
1449 * the specific case where it is running inside a vnet jail, to
1450 * avoid any POLA violation.
1451 */
1452 error = 0;
1453 if (!jail_export)
1454 error = VFS_MOUNT(mp);
1455
1456 export_error = 0;
1457 /* Process the export option. */
1458 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1459 &len) == 0) {
1460 /* Assume that there is only 1 ABI for each length. */
1461 switch (len) {
1462 case (sizeof(struct oexport_args)):
1463 bzero(&o2export, sizeof(o2export));
1464 /* FALLTHROUGH */
1465 case (sizeof(o2export)):
1466 bcopy(bufp, &o2export, len);
1467 export.ex_flags = (uint64_t)o2export.ex_flags;
1468 export.ex_root = o2export.ex_root;
1469 export.ex_uid = o2export.ex_anon.cr_uid;
1470 export.ex_groups = NULL;
1471 export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1472 if (export.ex_ngroups > 0) {
1473 if (export.ex_ngroups <= XU_NGROUPS) {
1474 export.ex_groups = malloc(
1475 export.ex_ngroups * sizeof(gid_t),
1476 M_TEMP, M_WAITOK);
1477 for (i = 0; i < export.ex_ngroups; i++)
1478 export.ex_groups[i] =
1479 o2export.ex_anon.cr_groups[i];
1480 } else
1481 export_error = EINVAL;
1482 } else if (export.ex_ngroups < 0)
1483 export_error = EINVAL;
1484 export.ex_addr = o2export.ex_addr;
1485 export.ex_addrlen = o2export.ex_addrlen;
1486 export.ex_mask = o2export.ex_mask;
1487 export.ex_masklen = o2export.ex_masklen;
1488 export.ex_indexfile = o2export.ex_indexfile;
1489 export.ex_numsecflavors = o2export.ex_numsecflavors;
1490 if (export.ex_numsecflavors < MAXSECFLAVORS) {
1491 for (i = 0; i < export.ex_numsecflavors; i++)
1492 export.ex_secflavors[i] =
1493 o2export.ex_secflavors[i];
1494 } else
1495 export_error = EINVAL;
1496 if (export_error == 0)
1497 export_error = vfs_export(mp, &export, true);
1498 free(export.ex_groups, M_TEMP);
1499 break;
1500 case (sizeof(export)):
1501 bcopy(bufp, &export, len);
1502 grps = NULL;
1503 if (export.ex_ngroups > 0) {
1504 if (export.ex_ngroups <= ngroups_max + 1) {
1505 grps = malloc(export.ex_ngroups *
1506 sizeof(gid_t), M_TEMP, M_WAITOK);
1507 export_error = copyin(export.ex_groups,
1508 grps, export.ex_ngroups *
1509 sizeof(gid_t));
1510 if (export_error == 0)
1511 export.ex_groups = grps;
1512 } else
1513 export_error = EINVAL;
1514 } else if (export.ex_ngroups == 0)
1515 export.ex_groups = NULL;
1516 else
1517 export_error = EINVAL;
1518 if (export_error == 0)
1519 export_error = vfs_export(mp, &export, true);
1520 free(grps, M_TEMP);
1521 break;
1522 default:
1523 export_error = EINVAL;
1524 break;
1525 }
1526 }
1527
1528 MNT_ILOCK(mp);
1529 if (error == 0) {
1530 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1531 MNT_SNAPSHOT);
1532 mp->mnt_flag |= mnt_union;
1533 } else {
1534 /*
1535 * If we fail, restore old mount flags. MNT_QUOTA is special,
1536 * because it is not part of MNT_UPDATEMASK, but it could have
1537 * changed in the meantime if quotactl(2) was called.
1538 * All in all we want current value of MNT_QUOTA, not the old
1539 * one.
1540 */
1541 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1542 }
1543 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1544 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1545 mp->mnt_kern_flag |= MNTK_ASYNC;
1546 else
1547 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1548 MNT_IUNLOCK(mp);
1549
1550 if (error != 0)
1551 goto end;
1552
1553 mount_devctl_event("REMOUNT", mp, true);
1554 if (mp->mnt_opt != NULL)
1555 vfs_freeopts(mp->mnt_opt);
1556 mp->mnt_opt = mp->mnt_optnew;
1557 *optlist = NULL;
1558 (void)VFS_STATFS(mp, &mp->mnt_stat);
1559 /*
1560 * Prevent external consumers of mount options from reading
1561 * mnt_optnew.
1562 */
1563 mp->mnt_optnew = NULL;
1564
1565 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1566 vfs_allocate_syncvnode(mp);
1567 else
1568 vfs_deallocate_syncvnode(mp);
1569 end:
1570 vfs_op_exit(mp);
1571 if (rootvp != NULL) {
1572 vn_seqc_write_end(rootvp);
1573 vrele(rootvp);
1574 }
1575 vn_seqc_write_end(vp);
1576 vfs_unbusy(mp);
1577 VI_LOCK(vp);
1578 vp->v_iflag &= ~VI_MOUNT;
1579 VI_UNLOCK(vp);
1580 vrele(vp);
1581 return (error != 0 ? error : export_error);
1582 }
1583
1584 /*
1585 * vfs_domount(): actually attempt a filesystem mount.
1586 */
1587 static int
vfs_domount(struct thread * td,const char * fstype,char * fspath,uint64_t fsflags,bool jail_export,struct vfsoptlist ** optlist)1588 vfs_domount(
1589 struct thread *td, /* Calling thread. */
1590 const char *fstype, /* Filesystem type. */
1591 char *fspath, /* Mount path. */
1592 uint64_t fsflags, /* Flags common to all filesystems. */
1593 bool jail_export, /* Got export option in vnet prison. */
1594 struct vfsoptlist **optlist /* Options local to the filesystem. */
1595 )
1596 {
1597 struct vfsconf *vfsp;
1598 struct nameidata nd;
1599 struct vnode *vp;
1600 char *pathbuf;
1601 int error;
1602
1603 /*
1604 * Be ultra-paranoid about making sure the type and fspath
1605 * variables will fit in our mp buffers, including the
1606 * terminating NUL.
1607 */
1608 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1609 return (ENAMETOOLONG);
1610
1611 if (jail_export) {
1612 error = priv_check(td, PRIV_NFS_DAEMON);
1613 if (error)
1614 return (error);
1615 } else if (jailed(td->td_ucred) || usermount == 0) {
1616 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1617 return (error);
1618 }
1619
1620 /*
1621 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1622 */
1623 if (fsflags & MNT_EXPORTED) {
1624 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1625 if (error)
1626 return (error);
1627 }
1628 if (fsflags & MNT_SUIDDIR) {
1629 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1630 if (error)
1631 return (error);
1632 }
1633 /*
1634 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1635 */
1636 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1637 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1638 fsflags |= MNT_NOSUID | MNT_USER;
1639 }
1640
1641 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1642 vfsp = NULL;
1643 if ((fsflags & MNT_UPDATE) == 0) {
1644 /* Don't try to load KLDs if we're mounting the root. */
1645 if (fsflags & MNT_ROOTFS) {
1646 if ((vfsp = vfs_byname(fstype)) == NULL)
1647 return (ENODEV);
1648 } else {
1649 if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1650 return (error);
1651 }
1652 }
1653
1654 /*
1655 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1656 */
1657 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT,
1658 UIO_SYSSPACE, fspath);
1659 error = namei(&nd);
1660 if (error != 0)
1661 return (error);
1662 vp = nd.ni_vp;
1663 /*
1664 * Don't allow stacking file mounts to work around problems with the way
1665 * that namei sets nd.ni_dvp to vp_crossmp for these.
1666 */
1667 if (vp->v_type == VREG)
1668 fsflags |= MNT_NOCOVER;
1669 if ((fsflags & MNT_UPDATE) == 0) {
1670 if ((vp->v_vflag & VV_ROOT) != 0 &&
1671 (fsflags & MNT_NOCOVER) != 0) {
1672 vput(vp);
1673 error = EBUSY;
1674 goto out;
1675 }
1676 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1677 strcpy(pathbuf, fspath);
1678 /*
1679 * Note: we allow any vnode type here. If the path sanity check
1680 * succeeds, the type will be validated in vfs_domount_first
1681 * above.
1682 */
1683 if (vp->v_type == VDIR)
1684 error = vn_path_to_global_path(td, vp, pathbuf,
1685 MNAMELEN);
1686 else
1687 error = vn_path_to_global_path_hardlink(td, vp,
1688 nd.ni_dvp, pathbuf, MNAMELEN,
1689 nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
1690 if (error == 0) {
1691 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1692 fsflags, optlist);
1693 }
1694 free(pathbuf, M_TEMP);
1695 } else
1696 error = vfs_domount_update(td, vp, fsflags, jail_export,
1697 optlist);
1698
1699 out:
1700 NDFREE_PNBUF(&nd);
1701 vrele(nd.ni_dvp);
1702
1703 return (error);
1704 }
1705
1706 /*
1707 * Unmount a filesystem.
1708 *
1709 * Note: unmount takes a path to the vnode mounted on as argument, not
1710 * special file (as before).
1711 */
1712 #ifndef _SYS_SYSPROTO_H_
1713 struct unmount_args {
1714 char *path;
1715 int flags;
1716 };
1717 #endif
1718 /* ARGSUSED */
1719 int
sys_unmount(struct thread * td,struct unmount_args * uap)1720 sys_unmount(struct thread *td, struct unmount_args *uap)
1721 {
1722
1723 return (kern_unmount(td, uap->path, uap->flags));
1724 }
1725
1726 int
kern_unmount(struct thread * td,const char * path,int flags)1727 kern_unmount(struct thread *td, const char *path, int flags)
1728 {
1729 struct nameidata nd;
1730 struct mount *mp;
1731 char *fsidbuf, *pathbuf;
1732 fsid_t fsid;
1733 int error;
1734
1735 AUDIT_ARG_VALUE(flags);
1736 if (jailed(td->td_ucred) || usermount == 0) {
1737 error = priv_check(td, PRIV_VFS_UNMOUNT);
1738 if (error)
1739 return (error);
1740 }
1741
1742 if (flags & MNT_BYFSID) {
1743 fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1744 error = copyinstr(path, fsidbuf, MNAMELEN, NULL);
1745 if (error) {
1746 free(fsidbuf, M_TEMP);
1747 return (error);
1748 }
1749
1750 AUDIT_ARG_TEXT(fsidbuf);
1751 /* Decode the filesystem ID. */
1752 if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) {
1753 free(fsidbuf, M_TEMP);
1754 return (EINVAL);
1755 }
1756
1757 mp = vfs_getvfs(&fsid);
1758 free(fsidbuf, M_TEMP);
1759 if (mp == NULL) {
1760 return (ENOENT);
1761 }
1762 } else {
1763 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1764 error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1765 if (error) {
1766 free(pathbuf, M_TEMP);
1767 return (error);
1768 }
1769
1770 /*
1771 * Try to find global path for path argument.
1772 */
1773 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1774 UIO_SYSSPACE, pathbuf);
1775 if (namei(&nd) == 0) {
1776 NDFREE_PNBUF(&nd);
1777 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1778 MNAMELEN);
1779 if (error == 0)
1780 vput(nd.ni_vp);
1781 }
1782 mtx_lock(&mountlist_mtx);
1783 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1784 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1785 vfs_ref(mp);
1786 break;
1787 }
1788 }
1789 mtx_unlock(&mountlist_mtx);
1790 free(pathbuf, M_TEMP);
1791 if (mp == NULL) {
1792 /*
1793 * Previously we returned ENOENT for a nonexistent path and
1794 * EINVAL for a non-mountpoint. We cannot tell these apart
1795 * now, so in the !MNT_BYFSID case return the more likely
1796 * EINVAL for compatibility.
1797 */
1798 return (EINVAL);
1799 }
1800 }
1801
1802 /*
1803 * Don't allow unmounting the root filesystem.
1804 */
1805 if (mp->mnt_flag & MNT_ROOTFS) {
1806 vfs_rel(mp);
1807 return (EINVAL);
1808 }
1809 error = dounmount(mp, flags, td);
1810 return (error);
1811 }
1812
1813 /*
1814 * Return error if any of the vnodes, ignoring the root vnode
1815 * and the syncer vnode, have non-zero usecount.
1816 *
1817 * This function is purely advisory - it can return false positives
1818 * and negatives.
1819 */
1820 static int
vfs_check_usecounts(struct mount * mp)1821 vfs_check_usecounts(struct mount *mp)
1822 {
1823 struct vnode *vp, *mvp;
1824
1825 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1826 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1827 vp->v_usecount != 0) {
1828 VI_UNLOCK(vp);
1829 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1830 return (EBUSY);
1831 }
1832 VI_UNLOCK(vp);
1833 }
1834
1835 return (0);
1836 }
1837
1838 static void
dounmount_cleanup(struct mount * mp,struct vnode * coveredvp,int mntkflags)1839 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1840 {
1841
1842 mtx_assert(MNT_MTX(mp), MA_OWNED);
1843 mp->mnt_kern_flag &= ~mntkflags;
1844 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1845 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1846 wakeup(mp);
1847 }
1848 vfs_op_exit_locked(mp);
1849 MNT_IUNLOCK(mp);
1850 if (coveredvp != NULL) {
1851 VOP_UNLOCK(coveredvp);
1852 vdrop(coveredvp);
1853 }
1854 vn_finished_write(mp);
1855 vfs_rel(mp);
1856 }
1857
1858 /*
1859 * There are various reference counters associated with the mount point.
1860 * Normally it is permitted to modify them without taking the mnt ilock,
1861 * but this behavior can be temporarily disabled if stable value is needed
1862 * or callers are expected to block (e.g. to not allow new users during
1863 * forced unmount).
1864 */
1865 void
vfs_op_enter(struct mount * mp)1866 vfs_op_enter(struct mount *mp)
1867 {
1868 struct mount_pcpu *mpcpu;
1869 int cpu;
1870
1871 MNT_ILOCK(mp);
1872 mp->mnt_vfs_ops++;
1873 if (mp->mnt_vfs_ops > 1) {
1874 MNT_IUNLOCK(mp);
1875 return;
1876 }
1877 vfs_op_barrier_wait(mp);
1878 CPU_FOREACH(cpu) {
1879 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1880
1881 mp->mnt_ref += mpcpu->mntp_ref;
1882 mpcpu->mntp_ref = 0;
1883
1884 mp->mnt_lockref += mpcpu->mntp_lockref;
1885 mpcpu->mntp_lockref = 0;
1886
1887 mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1888 mpcpu->mntp_writeopcount = 0;
1889 }
1890 MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1891 mp->mnt_writeopcount >= 0, mp,
1892 ("invalid count(s): ref %d lockref %d writeopcount %d",
1893 mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1894 MNT_IUNLOCK(mp);
1895 vfs_assert_mount_counters(mp);
1896 }
1897
1898 void
vfs_op_exit_locked(struct mount * mp)1899 vfs_op_exit_locked(struct mount *mp)
1900 {
1901
1902 mtx_assert(MNT_MTX(mp), MA_OWNED);
1903
1904 MPASSERT(mp->mnt_vfs_ops > 0, mp,
1905 ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1906 MPASSERT(mp->mnt_vfs_ops > 1 ||
1907 (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1908 ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1909 mp->mnt_vfs_ops--;
1910 }
1911
1912 void
vfs_op_exit(struct mount * mp)1913 vfs_op_exit(struct mount *mp)
1914 {
1915
1916 MNT_ILOCK(mp);
1917 vfs_op_exit_locked(mp);
1918 MNT_IUNLOCK(mp);
1919 }
1920
1921 struct vfs_op_barrier_ipi {
1922 struct mount *mp;
1923 struct smp_rendezvous_cpus_retry_arg srcra;
1924 };
1925
1926 static void
vfs_op_action_func(void * arg)1927 vfs_op_action_func(void *arg)
1928 {
1929 struct vfs_op_barrier_ipi *vfsopipi;
1930 struct mount *mp;
1931
1932 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1933 mp = vfsopipi->mp;
1934
1935 if (!vfs_op_thread_entered(mp))
1936 smp_rendezvous_cpus_done(arg);
1937 }
1938
1939 static void
vfs_op_wait_func(void * arg,int cpu)1940 vfs_op_wait_func(void *arg, int cpu)
1941 {
1942 struct vfs_op_barrier_ipi *vfsopipi;
1943 struct mount *mp;
1944 struct mount_pcpu *mpcpu;
1945
1946 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1947 mp = vfsopipi->mp;
1948
1949 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1950 while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1951 cpu_spinwait();
1952 }
1953
1954 void
vfs_op_barrier_wait(struct mount * mp)1955 vfs_op_barrier_wait(struct mount *mp)
1956 {
1957 struct vfs_op_barrier_ipi vfsopipi;
1958
1959 vfsopipi.mp = mp;
1960
1961 smp_rendezvous_cpus_retry(all_cpus,
1962 smp_no_rendezvous_barrier,
1963 vfs_op_action_func,
1964 smp_no_rendezvous_barrier,
1965 vfs_op_wait_func,
1966 &vfsopipi.srcra);
1967 }
1968
1969 #ifdef DIAGNOSTIC
1970 void
vfs_assert_mount_counters(struct mount * mp)1971 vfs_assert_mount_counters(struct mount *mp)
1972 {
1973 struct mount_pcpu *mpcpu;
1974 int cpu;
1975
1976 if (mp->mnt_vfs_ops == 0)
1977 return;
1978
1979 CPU_FOREACH(cpu) {
1980 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1981 if (mpcpu->mntp_ref != 0 ||
1982 mpcpu->mntp_lockref != 0 ||
1983 mpcpu->mntp_writeopcount != 0)
1984 vfs_dump_mount_counters(mp);
1985 }
1986 }
1987
1988 void
vfs_dump_mount_counters(struct mount * mp)1989 vfs_dump_mount_counters(struct mount *mp)
1990 {
1991 struct mount_pcpu *mpcpu;
1992 int ref, lockref, writeopcount;
1993 int cpu;
1994
1995 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1996
1997 printf(" ref : ");
1998 ref = mp->mnt_ref;
1999 CPU_FOREACH(cpu) {
2000 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2001 printf("%d ", mpcpu->mntp_ref);
2002 ref += mpcpu->mntp_ref;
2003 }
2004 printf("\n");
2005 printf(" lockref : ");
2006 lockref = mp->mnt_lockref;
2007 CPU_FOREACH(cpu) {
2008 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2009 printf("%d ", mpcpu->mntp_lockref);
2010 lockref += mpcpu->mntp_lockref;
2011 }
2012 printf("\n");
2013 printf("writeopcount: ");
2014 writeopcount = mp->mnt_writeopcount;
2015 CPU_FOREACH(cpu) {
2016 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2017 printf("%d ", mpcpu->mntp_writeopcount);
2018 writeopcount += mpcpu->mntp_writeopcount;
2019 }
2020 printf("\n");
2021
2022 printf("counter struct total\n");
2023 printf("ref %-5d %-5d\n", mp->mnt_ref, ref);
2024 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref);
2025 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount);
2026
2027 panic("invalid counts on struct mount");
2028 }
2029 #endif
2030
2031 int
vfs_mount_fetch_counter(struct mount * mp,enum mount_counter which)2032 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
2033 {
2034 struct mount_pcpu *mpcpu;
2035 int cpu, sum;
2036
2037 switch (which) {
2038 case MNT_COUNT_REF:
2039 sum = mp->mnt_ref;
2040 break;
2041 case MNT_COUNT_LOCKREF:
2042 sum = mp->mnt_lockref;
2043 break;
2044 case MNT_COUNT_WRITEOPCOUNT:
2045 sum = mp->mnt_writeopcount;
2046 break;
2047 }
2048
2049 CPU_FOREACH(cpu) {
2050 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2051 switch (which) {
2052 case MNT_COUNT_REF:
2053 sum += mpcpu->mntp_ref;
2054 break;
2055 case MNT_COUNT_LOCKREF:
2056 sum += mpcpu->mntp_lockref;
2057 break;
2058 case MNT_COUNT_WRITEOPCOUNT:
2059 sum += mpcpu->mntp_writeopcount;
2060 break;
2061 }
2062 }
2063 return (sum);
2064 }
2065
2066 static bool
deferred_unmount_enqueue(struct mount * mp,uint64_t flags,bool requeue,int timeout_ticks)2067 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue,
2068 int timeout_ticks)
2069 {
2070 bool enqueued;
2071
2072 enqueued = false;
2073 mtx_lock(&deferred_unmount_lock);
2074 if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) {
2075 mp->mnt_taskqueue_flags = flags | MNT_DEFERRED;
2076 STAILQ_INSERT_TAIL(&deferred_unmount_list, mp,
2077 mnt_taskqueue_link);
2078 enqueued = true;
2079 }
2080 mtx_unlock(&deferred_unmount_lock);
2081
2082 if (enqueued) {
2083 taskqueue_enqueue_timeout(taskqueue_deferred_unmount,
2084 &deferred_unmount_task, timeout_ticks);
2085 }
2086
2087 return (enqueued);
2088 }
2089
2090 /*
2091 * Taskqueue handler for processing async/recursive unmounts
2092 */
2093 static void
vfs_deferred_unmount(void * argi __unused,int pending __unused)2094 vfs_deferred_unmount(void *argi __unused, int pending __unused)
2095 {
2096 STAILQ_HEAD(, mount) local_unmounts;
2097 uint64_t flags;
2098 struct mount *mp, *tmp;
2099 int error;
2100 unsigned int retries;
2101 bool unmounted;
2102
2103 STAILQ_INIT(&local_unmounts);
2104 mtx_lock(&deferred_unmount_lock);
2105 STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list);
2106 mtx_unlock(&deferred_unmount_lock);
2107
2108 STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) {
2109 flags = mp->mnt_taskqueue_flags;
2110 KASSERT((flags & MNT_DEFERRED) != 0,
2111 ("taskqueue unmount without MNT_DEFERRED"));
2112 error = dounmount(mp, flags, curthread);
2113 if (error != 0) {
2114 MNT_ILOCK(mp);
2115 unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0);
2116 MNT_IUNLOCK(mp);
2117
2118 /*
2119 * The deferred unmount thread is the only thread that
2120 * modifies the retry counts, so locking/atomics aren't
2121 * needed here.
2122 */
2123 retries = (mp->mnt_unmount_retries)++;
2124 deferred_unmount_total_retries++;
2125 if (!unmounted && retries < deferred_unmount_retry_limit) {
2126 deferred_unmount_enqueue(mp, flags, true,
2127 -deferred_unmount_retry_delay_hz);
2128 } else {
2129 if (retries >= deferred_unmount_retry_limit) {
2130 printf("giving up on deferred unmount "
2131 "of %s after %d retries, error %d\n",
2132 mp->mnt_stat.f_mntonname, retries, error);
2133 }
2134 vfs_rel(mp);
2135 }
2136 }
2137 }
2138 }
2139
2140 /*
2141 * Do the actual filesystem unmount.
2142 */
2143 int
dounmount(struct mount * mp,uint64_t flags,struct thread * td)2144 dounmount(struct mount *mp, uint64_t flags, struct thread *td)
2145 {
2146 struct mount_upper_node *upper;
2147 struct vnode *coveredvp, *rootvp;
2148 int error;
2149 uint64_t async_flag;
2150 int mnt_gen_r;
2151 unsigned int retries;
2152
2153 KASSERT((flags & MNT_DEFERRED) == 0 ||
2154 (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE),
2155 ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE"));
2156
2157 /*
2158 * If the caller has explicitly requested the unmount to be handled by
2159 * the taskqueue and we're not already in taskqueue context, queue
2160 * up the unmount request and exit. This is done prior to any
2161 * credential checks; MNT_DEFERRED should be used only for kernel-
2162 * initiated unmounts and will therefore be processed with the
2163 * (kernel) credentials of the taskqueue thread. Still, callers
2164 * should be sure this is the behavior they want.
2165 */
2166 if ((flags & MNT_DEFERRED) != 0 &&
2167 taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) {
2168 if (!deferred_unmount_enqueue(mp, flags, false, 0))
2169 vfs_rel(mp);
2170 return (EINPROGRESS);
2171 }
2172
2173 /*
2174 * Only privileged root, or (if MNT_USER is set) the user that did the
2175 * original mount is permitted to unmount this filesystem.
2176 * This check should be made prior to queueing up any recursive
2177 * unmounts of upper filesystems. Those unmounts will be executed
2178 * with kernel thread credentials and are expected to succeed, so
2179 * we must at least ensure the originating context has sufficient
2180 * privilege to unmount the base filesystem before proceeding with
2181 * the uppers.
2182 */
2183 error = vfs_suser(mp, td);
2184 if (error != 0) {
2185 KASSERT((flags & MNT_DEFERRED) == 0,
2186 ("taskqueue unmount with insufficient privilege"));
2187 vfs_rel(mp);
2188 return (error);
2189 }
2190
2191 if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0))
2192 flags |= MNT_RECURSE;
2193
2194 if ((flags & MNT_RECURSE) != 0) {
2195 KASSERT((flags & MNT_FORCE) != 0,
2196 ("MNT_RECURSE requires MNT_FORCE"));
2197
2198 MNT_ILOCK(mp);
2199 /*
2200 * Set MNTK_RECURSE to prevent new upper mounts from being
2201 * added, and note that an operation on the uppers list is in
2202 * progress. This will ensure that unregistration from the
2203 * uppers list, and therefore any pending unmount of the upper
2204 * FS, can't complete until after we finish walking the list.
2205 */
2206 mp->mnt_kern_flag |= MNTK_RECURSE;
2207 mp->mnt_upper_pending++;
2208 TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) {
2209 retries = upper->mp->mnt_unmount_retries;
2210 if (retries > deferred_unmount_retry_limit) {
2211 error = EBUSY;
2212 continue;
2213 }
2214 MNT_IUNLOCK(mp);
2215
2216 vfs_ref(upper->mp);
2217 if (!deferred_unmount_enqueue(upper->mp, flags,
2218 false, 0))
2219 vfs_rel(upper->mp);
2220 MNT_ILOCK(mp);
2221 }
2222 mp->mnt_upper_pending--;
2223 if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 &&
2224 mp->mnt_upper_pending == 0) {
2225 mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER;
2226 wakeup(&mp->mnt_uppers);
2227 }
2228
2229 /*
2230 * If we're not on the taskqueue, wait until the uppers list
2231 * is drained before proceeding with unmount. Otherwise, if
2232 * we are on the taskqueue and there are still pending uppers,
2233 * just re-enqueue on the end of the taskqueue.
2234 */
2235 if ((flags & MNT_DEFERRED) == 0) {
2236 while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) {
2237 mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER;
2238 error = msleep(&mp->mnt_taskqueue_link,
2239 MNT_MTX(mp), PCATCH, "umntqw", 0);
2240 }
2241 if (error != 0) {
2242 MNT_REL(mp);
2243 MNT_IUNLOCK(mp);
2244 return (error);
2245 }
2246 } else if (!TAILQ_EMPTY(&mp->mnt_uppers)) {
2247 MNT_IUNLOCK(mp);
2248 if (error == 0)
2249 deferred_unmount_enqueue(mp, flags, true, 0);
2250 return (error);
2251 }
2252 MNT_IUNLOCK(mp);
2253 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty"));
2254 }
2255
2256 /* Allow the taskqueue to safely re-enqueue on failure */
2257 if ((flags & MNT_DEFERRED) != 0)
2258 vfs_ref(mp);
2259
2260 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
2261 mnt_gen_r = mp->mnt_gen;
2262 VI_LOCK(coveredvp);
2263 vholdl(coveredvp);
2264 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
2265 /*
2266 * Check for mp being unmounted while waiting for the
2267 * covered vnode lock.
2268 */
2269 if (coveredvp->v_mountedhere != mp ||
2270 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
2271 VOP_UNLOCK(coveredvp);
2272 vdrop(coveredvp);
2273 vfs_rel(mp);
2274 return (EBUSY);
2275 }
2276 }
2277
2278 vfs_op_enter(mp);
2279
2280 vn_start_write(NULL, &mp, V_WAIT);
2281 MNT_ILOCK(mp);
2282 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
2283 (mp->mnt_flag & MNT_UPDATE) != 0 ||
2284 !TAILQ_EMPTY(&mp->mnt_uppers)) {
2285 dounmount_cleanup(mp, coveredvp, 0);
2286 return (EBUSY);
2287 }
2288 mp->mnt_kern_flag |= MNTK_UNMOUNT;
2289 rootvp = vfs_cache_root_clear(mp);
2290 if (coveredvp != NULL)
2291 vn_seqc_write_begin(coveredvp);
2292 if (flags & MNT_NONBUSY) {
2293 MNT_IUNLOCK(mp);
2294 error = vfs_check_usecounts(mp);
2295 MNT_ILOCK(mp);
2296 if (error != 0) {
2297 vn_seqc_write_end(coveredvp);
2298 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
2299 if (rootvp != NULL) {
2300 vn_seqc_write_end(rootvp);
2301 vrele(rootvp);
2302 }
2303 return (error);
2304 }
2305 }
2306 /* Allow filesystems to detect that a forced unmount is in progress. */
2307 if (flags & MNT_FORCE) {
2308 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
2309 MNT_IUNLOCK(mp);
2310 /*
2311 * Must be done after setting MNTK_UNMOUNTF and before
2312 * waiting for mnt_lockref to become 0.
2313 */
2314 VFS_PURGE(mp);
2315 MNT_ILOCK(mp);
2316 }
2317 error = 0;
2318 if (mp->mnt_lockref) {
2319 mp->mnt_kern_flag |= MNTK_DRAINING;
2320 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
2321 "mount drain", 0);
2322 }
2323 MNT_IUNLOCK(mp);
2324 KASSERT(mp->mnt_lockref == 0,
2325 ("%s: invalid lock refcount in the drain path @ %s:%d",
2326 __func__, __FILE__, __LINE__));
2327 KASSERT(error == 0,
2328 ("%s: invalid return value for msleep in the drain path @ %s:%d",
2329 __func__, __FILE__, __LINE__));
2330
2331 /*
2332 * We want to keep the vnode around so that we can vn_seqc_write_end
2333 * after we are done with unmount. Downgrade our reference to a mere
2334 * hold count so that we don't interefere with anything.
2335 */
2336 if (rootvp != NULL) {
2337 vhold(rootvp);
2338 vrele(rootvp);
2339 }
2340
2341 if (mp->mnt_flag & MNT_EXPUBLIC)
2342 vfs_setpublicfs(NULL, NULL, NULL);
2343
2344 vfs_periodic(mp, MNT_WAIT);
2345 MNT_ILOCK(mp);
2346 async_flag = mp->mnt_flag & MNT_ASYNC;
2347 mp->mnt_flag &= ~MNT_ASYNC;
2348 mp->mnt_kern_flag &= ~MNTK_ASYNC;
2349 MNT_IUNLOCK(mp);
2350 vfs_deallocate_syncvnode(mp);
2351 error = VFS_UNMOUNT(mp, flags);
2352 vn_finished_write(mp);
2353 vfs_rel(mp);
2354 /*
2355 * If we failed to flush the dirty blocks for this mount point,
2356 * undo all the cdir/rdir and rootvnode changes we made above.
2357 * Unless we failed to do so because the device is reporting that
2358 * it doesn't exist anymore.
2359 */
2360 if (error && error != ENXIO) {
2361 MNT_ILOCK(mp);
2362 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
2363 MNT_IUNLOCK(mp);
2364 vfs_allocate_syncvnode(mp);
2365 MNT_ILOCK(mp);
2366 }
2367 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
2368 mp->mnt_flag |= async_flag;
2369 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
2370 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
2371 mp->mnt_kern_flag |= MNTK_ASYNC;
2372 if (mp->mnt_kern_flag & MNTK_MWAIT) {
2373 mp->mnt_kern_flag &= ~MNTK_MWAIT;
2374 wakeup(mp);
2375 }
2376 vfs_op_exit_locked(mp);
2377 MNT_IUNLOCK(mp);
2378 if (coveredvp) {
2379 vn_seqc_write_end(coveredvp);
2380 VOP_UNLOCK(coveredvp);
2381 vdrop(coveredvp);
2382 }
2383 if (rootvp != NULL) {
2384 vn_seqc_write_end(rootvp);
2385 vdrop(rootvp);
2386 }
2387 return (error);
2388 }
2389
2390 mtx_lock(&mountlist_mtx);
2391 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2392 mtx_unlock(&mountlist_mtx);
2393 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
2394 if (coveredvp != NULL) {
2395 VI_LOCK(coveredvp);
2396 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
2397 coveredvp->v_mountedhere = NULL;
2398 vn_seqc_write_end_locked(coveredvp);
2399 VI_UNLOCK(coveredvp);
2400 VOP_UNLOCK(coveredvp);
2401 vdrop(coveredvp);
2402 }
2403 mount_devctl_event("UNMOUNT", mp, false);
2404 if (rootvp != NULL) {
2405 vn_seqc_write_end(rootvp);
2406 vdrop(rootvp);
2407 }
2408 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
2409 if (rootvnode != NULL && mp == rootvnode->v_mount) {
2410 vrele(rootvnode);
2411 rootvnode = NULL;
2412 }
2413 if (mp == rootdevmp)
2414 rootdevmp = NULL;
2415 if ((flags & MNT_DEFERRED) != 0)
2416 vfs_rel(mp);
2417 vfs_mount_destroy(mp);
2418 return (0);
2419 }
2420
2421 /*
2422 * Report errors during filesystem mounting.
2423 */
2424 void
vfs_mount_error(struct mount * mp,const char * fmt,...)2425 vfs_mount_error(struct mount *mp, const char *fmt, ...)
2426 {
2427 struct vfsoptlist *moptlist = mp->mnt_optnew;
2428 va_list ap;
2429 int error, len;
2430 char *errmsg;
2431
2432 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
2433 if (error || errmsg == NULL || len <= 0)
2434 return;
2435
2436 va_start(ap, fmt);
2437 vsnprintf(errmsg, (size_t)len, fmt, ap);
2438 va_end(ap);
2439 }
2440
2441 void
vfs_opterror(struct vfsoptlist * opts,const char * fmt,...)2442 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2443 {
2444 va_list ap;
2445 int error, len;
2446 char *errmsg;
2447
2448 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2449 if (error || errmsg == NULL || len <= 0)
2450 return;
2451
2452 va_start(ap, fmt);
2453 vsnprintf(errmsg, (size_t)len, fmt, ap);
2454 va_end(ap);
2455 }
2456
2457 /*
2458 * ---------------------------------------------------------------------
2459 * Functions for querying mount options/arguments from filesystems.
2460 */
2461
2462 /*
2463 * Check that no unknown options are given
2464 */
2465 int
vfs_filteropt(struct vfsoptlist * opts,const char ** legal)2466 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2467 {
2468 struct vfsopt *opt;
2469 char errmsg[255];
2470 const char **t, *p, *q;
2471 int ret = 0;
2472
2473 TAILQ_FOREACH(opt, opts, link) {
2474 p = opt->name;
2475 q = NULL;
2476 if (p[0] == 'n' && p[1] == 'o')
2477 q = p + 2;
2478 for(t = global_opts; *t != NULL; t++) {
2479 if (strcmp(*t, p) == 0)
2480 break;
2481 if (q != NULL) {
2482 if (strcmp(*t, q) == 0)
2483 break;
2484 }
2485 }
2486 if (*t != NULL)
2487 continue;
2488 for(t = legal; *t != NULL; t++) {
2489 if (strcmp(*t, p) == 0)
2490 break;
2491 if (q != NULL) {
2492 if (strcmp(*t, q) == 0)
2493 break;
2494 }
2495 }
2496 if (*t != NULL)
2497 continue;
2498 snprintf(errmsg, sizeof(errmsg),
2499 "mount option <%s> is unknown", p);
2500 ret = EINVAL;
2501 }
2502 if (ret != 0) {
2503 TAILQ_FOREACH(opt, opts, link) {
2504 if (strcmp(opt->name, "errmsg") == 0) {
2505 strncpy((char *)opt->value, errmsg, opt->len);
2506 break;
2507 }
2508 }
2509 if (opt == NULL)
2510 printf("%s\n", errmsg);
2511 }
2512 return (ret);
2513 }
2514
2515 /*
2516 * Get a mount option by its name.
2517 *
2518 * Return 0 if the option was found, ENOENT otherwise.
2519 * If len is non-NULL it will be filled with the length
2520 * of the option. If buf is non-NULL, it will be filled
2521 * with the address of the option.
2522 */
2523 int
vfs_getopt(struct vfsoptlist * opts,const char * name,void ** buf,int * len)2524 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2525 {
2526 struct vfsopt *opt;
2527
2528 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2529
2530 TAILQ_FOREACH(opt, opts, link) {
2531 if (strcmp(name, opt->name) == 0) {
2532 opt->seen = 1;
2533 if (len != NULL)
2534 *len = opt->len;
2535 if (buf != NULL)
2536 *buf = opt->value;
2537 return (0);
2538 }
2539 }
2540 return (ENOENT);
2541 }
2542
2543 int
vfs_getopt_pos(struct vfsoptlist * opts,const char * name)2544 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2545 {
2546 struct vfsopt *opt;
2547
2548 if (opts == NULL)
2549 return (-1);
2550
2551 TAILQ_FOREACH(opt, opts, link) {
2552 if (strcmp(name, opt->name) == 0) {
2553 opt->seen = 1;
2554 return (opt->pos);
2555 }
2556 }
2557 return (-1);
2558 }
2559
2560 int
vfs_getopt_size(struct vfsoptlist * opts,const char * name,off_t * value)2561 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2562 {
2563 char *opt_value, *vtp;
2564 quad_t iv;
2565 int error, opt_len;
2566
2567 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2568 if (error != 0)
2569 return (error);
2570 if (opt_len == 0 || opt_value == NULL)
2571 return (EINVAL);
2572 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2573 return (EINVAL);
2574 iv = strtoq(opt_value, &vtp, 0);
2575 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2576 return (EINVAL);
2577 if (iv < 0)
2578 return (EINVAL);
2579 switch (vtp[0]) {
2580 case 't': case 'T':
2581 iv *= 1024;
2582 /* FALLTHROUGH */
2583 case 'g': case 'G':
2584 iv *= 1024;
2585 /* FALLTHROUGH */
2586 case 'm': case 'M':
2587 iv *= 1024;
2588 /* FALLTHROUGH */
2589 case 'k': case 'K':
2590 iv *= 1024;
2591 case '\0':
2592 break;
2593 default:
2594 return (EINVAL);
2595 }
2596 *value = iv;
2597
2598 return (0);
2599 }
2600
2601 char *
vfs_getopts(struct vfsoptlist * opts,const char * name,int * error)2602 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2603 {
2604 struct vfsopt *opt;
2605
2606 *error = 0;
2607 TAILQ_FOREACH(opt, opts, link) {
2608 if (strcmp(name, opt->name) != 0)
2609 continue;
2610 opt->seen = 1;
2611 if (opt->len == 0 ||
2612 ((char *)opt->value)[opt->len - 1] != '\0') {
2613 *error = EINVAL;
2614 return (NULL);
2615 }
2616 return (opt->value);
2617 }
2618 *error = ENOENT;
2619 return (NULL);
2620 }
2621
2622 int
vfs_flagopt(struct vfsoptlist * opts,const char * name,uint64_t * w,uint64_t val)2623 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2624 uint64_t val)
2625 {
2626 struct vfsopt *opt;
2627
2628 TAILQ_FOREACH(opt, opts, link) {
2629 if (strcmp(name, opt->name) == 0) {
2630 opt->seen = 1;
2631 if (w != NULL)
2632 *w |= val;
2633 return (1);
2634 }
2635 }
2636 if (w != NULL)
2637 *w &= ~val;
2638 return (0);
2639 }
2640
2641 int
vfs_scanopt(struct vfsoptlist * opts,const char * name,const char * fmt,...)2642 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2643 {
2644 va_list ap;
2645 struct vfsopt *opt;
2646 int ret;
2647
2648 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2649
2650 TAILQ_FOREACH(opt, opts, link) {
2651 if (strcmp(name, opt->name) != 0)
2652 continue;
2653 opt->seen = 1;
2654 if (opt->len == 0 || opt->value == NULL)
2655 return (0);
2656 if (((char *)opt->value)[opt->len - 1] != '\0')
2657 return (0);
2658 va_start(ap, fmt);
2659 ret = vsscanf(opt->value, fmt, ap);
2660 va_end(ap);
2661 return (ret);
2662 }
2663 return (0);
2664 }
2665
2666 int
vfs_setopt(struct vfsoptlist * opts,const char * name,void * value,int len)2667 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2668 {
2669 struct vfsopt *opt;
2670
2671 TAILQ_FOREACH(opt, opts, link) {
2672 if (strcmp(name, opt->name) != 0)
2673 continue;
2674 opt->seen = 1;
2675 if (opt->value == NULL)
2676 opt->len = len;
2677 else {
2678 if (opt->len != len)
2679 return (EINVAL);
2680 bcopy(value, opt->value, len);
2681 }
2682 return (0);
2683 }
2684 return (ENOENT);
2685 }
2686
2687 int
vfs_setopt_part(struct vfsoptlist * opts,const char * name,void * value,int len)2688 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2689 {
2690 struct vfsopt *opt;
2691
2692 TAILQ_FOREACH(opt, opts, link) {
2693 if (strcmp(name, opt->name) != 0)
2694 continue;
2695 opt->seen = 1;
2696 if (opt->value == NULL)
2697 opt->len = len;
2698 else {
2699 if (opt->len < len)
2700 return (EINVAL);
2701 opt->len = len;
2702 bcopy(value, opt->value, len);
2703 }
2704 return (0);
2705 }
2706 return (ENOENT);
2707 }
2708
2709 int
vfs_setopts(struct vfsoptlist * opts,const char * name,const char * value)2710 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2711 {
2712 struct vfsopt *opt;
2713
2714 TAILQ_FOREACH(opt, opts, link) {
2715 if (strcmp(name, opt->name) != 0)
2716 continue;
2717 opt->seen = 1;
2718 if (opt->value == NULL)
2719 opt->len = strlen(value) + 1;
2720 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2721 return (EINVAL);
2722 return (0);
2723 }
2724 return (ENOENT);
2725 }
2726
2727 /*
2728 * Find and copy a mount option.
2729 *
2730 * The size of the buffer has to be specified
2731 * in len, if it is not the same length as the
2732 * mount option, EINVAL is returned.
2733 * Returns ENOENT if the option is not found.
2734 */
2735 int
vfs_copyopt(struct vfsoptlist * opts,const char * name,void * dest,int len)2736 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2737 {
2738 struct vfsopt *opt;
2739
2740 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2741
2742 TAILQ_FOREACH(opt, opts, link) {
2743 if (strcmp(name, opt->name) == 0) {
2744 opt->seen = 1;
2745 if (len != opt->len)
2746 return (EINVAL);
2747 bcopy(opt->value, dest, opt->len);
2748 return (0);
2749 }
2750 }
2751 return (ENOENT);
2752 }
2753
2754 int
__vfs_statfs(struct mount * mp,struct statfs * sbp)2755 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2756 {
2757 /*
2758 * Filesystems only fill in part of the structure for updates, we
2759 * have to read the entirety first to get all content.
2760 */
2761 if (sbp != &mp->mnt_stat)
2762 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2763
2764 /*
2765 * Set these in case the underlying filesystem fails to do so.
2766 */
2767 sbp->f_version = STATFS_VERSION;
2768 sbp->f_namemax = NAME_MAX;
2769 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2770 sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize;
2771
2772 return (mp->mnt_op->vfs_statfs(mp, sbp));
2773 }
2774
2775 void
vfs_mountedfrom(struct mount * mp,const char * from)2776 vfs_mountedfrom(struct mount *mp, const char *from)
2777 {
2778
2779 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2780 strlcpy(mp->mnt_stat.f_mntfromname, from,
2781 sizeof mp->mnt_stat.f_mntfromname);
2782 }
2783
2784 /*
2785 * ---------------------------------------------------------------------
2786 * This is the api for building mount args and mounting filesystems from
2787 * inside the kernel.
2788 *
2789 * The API works by accumulation of individual args. First error is
2790 * latched.
2791 *
2792 * XXX: should be documented in new manpage kernel_mount(9)
2793 */
2794
2795 /* A memory allocation which must be freed when we are done */
2796 struct mntaarg {
2797 SLIST_ENTRY(mntaarg) next;
2798 };
2799
2800 /* The header for the mount arguments */
2801 struct mntarg {
2802 struct iovec *v;
2803 int len;
2804 int error;
2805 SLIST_HEAD(, mntaarg) list;
2806 };
2807
2808 /*
2809 * Add a boolean argument.
2810 *
2811 * flag is the boolean value.
2812 * name must start with "no".
2813 */
2814 struct mntarg *
mount_argb(struct mntarg * ma,int flag,const char * name)2815 mount_argb(struct mntarg *ma, int flag, const char *name)
2816 {
2817
2818 KASSERT(name[0] == 'n' && name[1] == 'o',
2819 ("mount_argb(...,%s): name must start with 'no'", name));
2820
2821 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2822 }
2823
2824 /*
2825 * Add an argument printf style
2826 */
2827 struct mntarg *
mount_argf(struct mntarg * ma,const char * name,const char * fmt,...)2828 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2829 {
2830 va_list ap;
2831 struct mntaarg *maa;
2832 struct sbuf *sb;
2833 int len;
2834
2835 if (ma == NULL) {
2836 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2837 SLIST_INIT(&ma->list);
2838 }
2839 if (ma->error)
2840 return (ma);
2841
2842 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2843 M_MOUNT, M_WAITOK);
2844 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2845 ma->v[ma->len].iov_len = strlen(name) + 1;
2846 ma->len++;
2847
2848 sb = sbuf_new_auto();
2849 va_start(ap, fmt);
2850 sbuf_vprintf(sb, fmt, ap);
2851 va_end(ap);
2852 sbuf_finish(sb);
2853 len = sbuf_len(sb) + 1;
2854 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2855 SLIST_INSERT_HEAD(&ma->list, maa, next);
2856 bcopy(sbuf_data(sb), maa + 1, len);
2857 sbuf_delete(sb);
2858
2859 ma->v[ma->len].iov_base = maa + 1;
2860 ma->v[ma->len].iov_len = len;
2861 ma->len++;
2862
2863 return (ma);
2864 }
2865
2866 /*
2867 * Add an argument which is a userland string.
2868 */
2869 struct mntarg *
mount_argsu(struct mntarg * ma,const char * name,const void * val,int len)2870 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2871 {
2872 struct mntaarg *maa;
2873 char *tbuf;
2874
2875 if (val == NULL)
2876 return (ma);
2877 if (ma == NULL) {
2878 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2879 SLIST_INIT(&ma->list);
2880 }
2881 if (ma->error)
2882 return (ma);
2883 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2884 SLIST_INSERT_HEAD(&ma->list, maa, next);
2885 tbuf = (void *)(maa + 1);
2886 ma->error = copyinstr(val, tbuf, len, NULL);
2887 return (mount_arg(ma, name, tbuf, -1));
2888 }
2889
2890 /*
2891 * Plain argument.
2892 *
2893 * If length is -1, treat value as a C string.
2894 */
2895 struct mntarg *
mount_arg(struct mntarg * ma,const char * name,const void * val,int len)2896 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2897 {
2898
2899 if (ma == NULL) {
2900 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2901 SLIST_INIT(&ma->list);
2902 }
2903 if (ma->error)
2904 return (ma);
2905
2906 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2907 M_MOUNT, M_WAITOK);
2908 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2909 ma->v[ma->len].iov_len = strlen(name) + 1;
2910 ma->len++;
2911
2912 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2913 if (len < 0)
2914 ma->v[ma->len].iov_len = strlen(val) + 1;
2915 else
2916 ma->v[ma->len].iov_len = len;
2917 ma->len++;
2918 return (ma);
2919 }
2920
2921 /*
2922 * Free a mntarg structure
2923 */
2924 static void
free_mntarg(struct mntarg * ma)2925 free_mntarg(struct mntarg *ma)
2926 {
2927 struct mntaarg *maa;
2928
2929 while (!SLIST_EMPTY(&ma->list)) {
2930 maa = SLIST_FIRST(&ma->list);
2931 SLIST_REMOVE_HEAD(&ma->list, next);
2932 free(maa, M_MOUNT);
2933 }
2934 free(ma->v, M_MOUNT);
2935 free(ma, M_MOUNT);
2936 }
2937
2938 /*
2939 * Mount a filesystem
2940 */
2941 int
kernel_mount(struct mntarg * ma,uint64_t flags)2942 kernel_mount(struct mntarg *ma, uint64_t flags)
2943 {
2944 struct uio auio;
2945 int error;
2946
2947 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2948 KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2949 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2950
2951 error = ma->error;
2952 if (error == 0) {
2953 auio.uio_iov = ma->v;
2954 auio.uio_iovcnt = ma->len;
2955 auio.uio_segflg = UIO_SYSSPACE;
2956 error = vfs_donmount(curthread, flags, &auio);
2957 }
2958 free_mntarg(ma);
2959 return (error);
2960 }
2961
2962 /* Map from mount options to printable formats. */
2963 static struct mntoptnames optnames[] = {
2964 MNTOPT_NAMES
2965 };
2966
2967 #define DEVCTL_LEN 1024
2968 static void
mount_devctl_event(const char * type,struct mount * mp,bool donew)2969 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2970 {
2971 const uint8_t *cp;
2972 struct mntoptnames *fp;
2973 struct sbuf sb;
2974 struct statfs *sfp = &mp->mnt_stat;
2975 char *buf;
2976
2977 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2978 if (buf == NULL)
2979 return;
2980 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2981 sbuf_cpy(&sb, "mount-point=\"");
2982 devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2983 sbuf_cat(&sb, "\" mount-dev=\"");
2984 devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2985 sbuf_cat(&sb, "\" mount-type=\"");
2986 devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2987 sbuf_cat(&sb, "\" fsid=0x");
2988 cp = (const uint8_t *)&sfp->f_fsid.val[0];
2989 for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2990 sbuf_printf(&sb, "%02x", cp[i]);
2991 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2992 for (fp = optnames; fp->o_opt != 0; fp++) {
2993 if ((mp->mnt_flag & fp->o_opt) != 0) {
2994 sbuf_cat(&sb, fp->o_name);
2995 sbuf_putc(&sb, ';');
2996 }
2997 }
2998 sbuf_putc(&sb, '"');
2999 sbuf_finish(&sb);
3000
3001 /*
3002 * Options are not published because the form of the options depends on
3003 * the file system and may include binary data. In addition, they don't
3004 * necessarily provide enough useful information to be actionable when
3005 * devd processes them.
3006 */
3007
3008 if (sbuf_error(&sb) == 0)
3009 devctl_notify("VFS", "FS", type, sbuf_data(&sb));
3010 sbuf_delete(&sb);
3011 free(buf, M_MOUNT);
3012 }
3013
3014 /*
3015 * Force remount specified mount point to read-only. The argument
3016 * must be busied to avoid parallel unmount attempts.
3017 *
3018 * Intended use is to prevent further writes if some metadata
3019 * inconsistency is detected. Note that the function still flushes
3020 * all cached metadata and data for the mount point, which might be
3021 * not always suitable.
3022 */
3023 int
vfs_remount_ro(struct mount * mp)3024 vfs_remount_ro(struct mount *mp)
3025 {
3026 struct vfsoptlist *opts;
3027 struct vfsopt *opt;
3028 struct vnode *vp_covered, *rootvp;
3029 int error;
3030
3031 vfs_op_enter(mp);
3032 KASSERT(mp->mnt_lockref > 0,
3033 ("vfs_remount_ro: mp %p is not busied", mp));
3034 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
3035 ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
3036
3037 rootvp = NULL;
3038 vp_covered = mp->mnt_vnodecovered;
3039 error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
3040 if (error != 0) {
3041 vfs_op_exit(mp);
3042 return (error);
3043 }
3044 VI_LOCK(vp_covered);
3045 if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
3046 VI_UNLOCK(vp_covered);
3047 vput(vp_covered);
3048 vfs_op_exit(mp);
3049 return (EBUSY);
3050 }
3051 vp_covered->v_iflag |= VI_MOUNT;
3052 VI_UNLOCK(vp_covered);
3053 vn_seqc_write_begin(vp_covered);
3054
3055 MNT_ILOCK(mp);
3056 if ((mp->mnt_flag & MNT_RDONLY) != 0) {
3057 MNT_IUNLOCK(mp);
3058 error = EBUSY;
3059 goto out;
3060 }
3061 mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
3062 rootvp = vfs_cache_root_clear(mp);
3063 MNT_IUNLOCK(mp);
3064
3065 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
3066 TAILQ_INIT(opts);
3067 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
3068 opt->name = strdup("ro", M_MOUNT);
3069 opt->value = NULL;
3070 TAILQ_INSERT_TAIL(opts, opt, link);
3071 vfs_mergeopts(opts, mp->mnt_opt);
3072 mp->mnt_optnew = opts;
3073
3074 error = VFS_MOUNT(mp);
3075
3076 if (error == 0) {
3077 MNT_ILOCK(mp);
3078 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
3079 MNT_IUNLOCK(mp);
3080 vfs_deallocate_syncvnode(mp);
3081 if (mp->mnt_opt != NULL)
3082 vfs_freeopts(mp->mnt_opt);
3083 mp->mnt_opt = mp->mnt_optnew;
3084 } else {
3085 MNT_ILOCK(mp);
3086 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
3087 MNT_IUNLOCK(mp);
3088 vfs_freeopts(mp->mnt_optnew);
3089 }
3090 mp->mnt_optnew = NULL;
3091
3092 out:
3093 vfs_op_exit(mp);
3094 VI_LOCK(vp_covered);
3095 vp_covered->v_iflag &= ~VI_MOUNT;
3096 VI_UNLOCK(vp_covered);
3097 vput(vp_covered);
3098 vn_seqc_write_end(vp_covered);
3099 if (rootvp != NULL) {
3100 vn_seqc_write_end(rootvp);
3101 vrele(rootvp);
3102 }
3103 return (error);
3104 }
3105
3106 /*
3107 * Suspend write operations on all local writeable filesystems. Does
3108 * full sync of them in the process.
3109 *
3110 * Iterate over the mount points in reverse order, suspending most
3111 * recently mounted filesystems first. It handles a case where a
3112 * filesystem mounted from a md(4) vnode-backed device should be
3113 * suspended before the filesystem that owns the vnode.
3114 */
3115 void
suspend_all_fs(void)3116 suspend_all_fs(void)
3117 {
3118 struct mount *mp;
3119 int error;
3120
3121 mtx_lock(&mountlist_mtx);
3122 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
3123 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
3124 if (error != 0)
3125 continue;
3126 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
3127 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
3128 mtx_lock(&mountlist_mtx);
3129 vfs_unbusy(mp);
3130 continue;
3131 }
3132 error = vfs_write_suspend(mp, 0);
3133 if (error == 0) {
3134 MNT_ILOCK(mp);
3135 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
3136 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
3137 MNT_IUNLOCK(mp);
3138 mtx_lock(&mountlist_mtx);
3139 } else {
3140 printf("suspend of %s failed, error %d\n",
3141 mp->mnt_stat.f_mntonname, error);
3142 mtx_lock(&mountlist_mtx);
3143 vfs_unbusy(mp);
3144 }
3145 }
3146 mtx_unlock(&mountlist_mtx);
3147 }
3148
3149 /*
3150 * Clone the mnt_exjail field to a new mount point.
3151 */
3152 void
vfs_exjail_clone(struct mount * inmp,struct mount * outmp)3153 vfs_exjail_clone(struct mount *inmp, struct mount *outmp)
3154 {
3155 struct ucred *cr;
3156 struct prison *pr;
3157
3158 MNT_ILOCK(inmp);
3159 cr = inmp->mnt_exjail;
3160 if (cr != NULL) {
3161 crhold(cr);
3162 MNT_IUNLOCK(inmp);
3163 pr = cr->cr_prison;
3164 sx_slock(&allprison_lock);
3165 if (!prison_isalive(pr)) {
3166 sx_sunlock(&allprison_lock);
3167 crfree(cr);
3168 return;
3169 }
3170 MNT_ILOCK(outmp);
3171 if (outmp->mnt_exjail == NULL) {
3172 outmp->mnt_exjail = cr;
3173 atomic_add_int(&pr->pr_exportcnt, 1);
3174 cr = NULL;
3175 }
3176 MNT_IUNLOCK(outmp);
3177 sx_sunlock(&allprison_lock);
3178 if (cr != NULL)
3179 crfree(cr);
3180 } else
3181 MNT_IUNLOCK(inmp);
3182 }
3183
3184 void
resume_all_fs(void)3185 resume_all_fs(void)
3186 {
3187 struct mount *mp;
3188
3189 mtx_lock(&mountlist_mtx);
3190 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3191 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
3192 continue;
3193 mtx_unlock(&mountlist_mtx);
3194 MNT_ILOCK(mp);
3195 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
3196 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
3197 MNT_IUNLOCK(mp);
3198 vfs_write_resume(mp, 0);
3199 mtx_lock(&mountlist_mtx);
3200 vfs_unbusy(mp);
3201 }
3202 mtx_unlock(&mountlist_mtx);
3203 }
3204