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