39
40#include <sys/param.h>
41#include <sys/conf.h>
42#include <sys/jail.h>
43#include <sys/kernel.h>
44#include <sys/libkern.h>
45#include <sys/mac.h>
46#include <sys/malloc.h>
47#include <sys/mount.h>
48#include <sys/mutex.h>
49#include <sys/namei.h>
50#include <sys/proc.h>
51#include <sys/filedesc.h>
52#include <sys/reboot.h>
53#include <sys/syscallsubr.h>
54#include <sys/sysproto.h>
55#include <sys/sx.h>
56#include <sys/sysctl.h>
57#include <sys/sysent.h>
58#include <sys/systm.h>
59#include <sys/vnode.h>
60#include <vm/uma.h>
61
62#include <geom/geom.h>
63
64#include <machine/stdarg.h>
65
66#include <security/audit/audit.h>
67
68#include "opt_rootdevname.h"
69#include "opt_ddb.h"
70#include "opt_mac.h"
71
72#ifdef DDB
73#include <ddb/ddb.h>
74#endif
75
76#define ROOTNAME "root_device"
77#define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
78
79static int vfs_domount(struct thread *td, const char *fstype,
80 char *fspath, int fsflags, void *fsdata);
81static struct mount *vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
82 const char *fspath, struct thread *td);
83static int vfs_mountroot_ask(void);
84static int vfs_mountroot_try(const char *mountfrom);
85static int vfs_donmount(struct thread *td, int fsflags,
86 struct uio *fsoptions);
87static void free_mntarg(struct mntarg *ma);
88static void vfs_mount_destroy(struct mount *);
89static int vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
90
91static int usermount = 0;
92SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
93 "Unprivileged users may mount and unmount file systems");
94
95MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
96MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
97static uma_zone_t mount_zone;
98
99/* List of mounted filesystems. */
100struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
101
102/* For any iteration/modification of mountlist */
103struct mtx mountlist_mtx;
104MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
105
106TAILQ_HEAD(vfsoptlist, vfsopt);
107struct vfsopt {
108 TAILQ_ENTRY(vfsopt) link;
109 char *name;
110 void *value;
111 int len;
112};
113
114/*
115 * The vnode of the system's root (/ in the filesystem, without chroot
116 * active.)
117 */
118struct vnode *rootvnode;
119
120/*
121 * The root filesystem is detailed in the kernel environment variable
122 * vfs.root.mountfrom, which is expected to be in the general format
123 *
124 * <vfsname>:[<path>]
125 * vfsname := the name of a VFS known to the kernel and capable
126 * of being mounted as root
127 * path := disk device name or other data used by the filesystem
128 * to locate its physical store
129 */
130
131/*
132 * Global opts, taken by all filesystems
133 */
134static const char *global_opts[] = {
135 "errmsg",
136 "fstype",
137 "fspath",
138 "rdonly",
139 "ro",
140 "rw",
141 "suid",
142 "exec",
143 "update",
144 NULL
145};
146
147/*
148 * The root specifiers we will try if RB_CDROM is specified.
149 */
150static char *cdrom_rootdevnames[] = {
151 "cd9660:cd0",
152 "cd9660:acd0",
153 NULL
154};
155
156/* legacy find-root code */
157char *rootdevnames[2] = {NULL, NULL};
158#ifndef ROOTDEVNAME
159# define ROOTDEVNAME NULL
160#endif
161static const char *ctrootdevname = ROOTDEVNAME;
162
163/*
164 * ---------------------------------------------------------------------
165 * Functions for building and sanitizing the mount options
166 */
167
168/* Remove one mount option. */
169static void
170vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
171{
172
173 TAILQ_REMOVE(opts, opt, link);
174 free(opt->name, M_MOUNT);
175 if (opt->value != NULL)
176 free(opt->value, M_MOUNT);
177#ifdef INVARIANTS
178 else if (opt->len != 0)
179 panic("%s: mount option with NULL value but length != 0",
180 __func__);
181#endif
182 free(opt, M_MOUNT);
183}
184
185/* Release all resources related to the mount options. */
186static void
187vfs_freeopts(struct vfsoptlist *opts)
188{
189 struct vfsopt *opt;
190
191 while (!TAILQ_EMPTY(opts)) {
192 opt = TAILQ_FIRST(opts);
193 vfs_freeopt(opts, opt);
194 }
195 free(opts, M_MOUNT);
196}
197
198/*
199 * Check if options are equal (with or without the "no" prefix).
200 */
201static int
202vfs_equalopts(const char *opt1, const char *opt2)
203{
204
205 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
206 if (strcmp(opt1, opt2) == 0)
207 return (1);
208 /* "noopt" vs. "opt" */
209 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
210 return (1);
211 /* "opt" vs. "noopt" */
212 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
213 return (1);
214 return (0);
215}
216
217/*
218 * If a mount option is specified several times,
219 * (with or without the "no" prefix) only keep
220 * the last occurence of it.
221 */
222static void
223vfs_sanitizeopts(struct vfsoptlist *opts)
224{
225 struct vfsopt *opt, *opt2, *tmp;
226
227 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
228 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
229 while (opt2 != NULL) {
230 if (vfs_equalopts(opt->name, opt2->name)) {
231 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
232 vfs_freeopt(opts, opt2);
233 opt2 = tmp;
234 } else {
235 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
236 }
237 }
238 }
239}
240
241/*
242 * Build a linked list of mount options from a struct uio.
243 */
244static int
245vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
246{
247 struct vfsoptlist *opts;
248 struct vfsopt *opt;
249 size_t memused;
250 unsigned int i, iovcnt;
251 int error, namelen, optlen;
252
253 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
254 TAILQ_INIT(opts);
255 memused = 0;
256 iovcnt = auio->uio_iovcnt;
257 for (i = 0; i < iovcnt; i += 2) {
258 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
259 namelen = auio->uio_iov[i].iov_len;
260 optlen = auio->uio_iov[i + 1].iov_len;
261 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
262 opt->value = NULL;
263 opt->len = 0;
264
265 /*
266 * Do this early, so jumps to "bad" will free the current
267 * option.
268 */
269 TAILQ_INSERT_TAIL(opts, opt, link);
270 memused += sizeof(struct vfsopt) + optlen + namelen;
271
272 /*
273 * Avoid consuming too much memory, and attempts to overflow
274 * memused.
275 */
276 if (memused > VFS_MOUNTARG_SIZE_MAX ||
277 optlen > VFS_MOUNTARG_SIZE_MAX ||
278 namelen > VFS_MOUNTARG_SIZE_MAX) {
279 error = EINVAL;
280 goto bad;
281 }
282
283 if (auio->uio_segflg == UIO_SYSSPACE) {
284 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
285 } else {
286 error = copyin(auio->uio_iov[i].iov_base, opt->name,
287 namelen);
288 if (error)
289 goto bad;
290 }
291 /* Ensure names are null-terminated strings. */
292 if (opt->name[namelen - 1] != '\0') {
293 error = EINVAL;
294 goto bad;
295 }
296 if (optlen != 0) {
297 opt->len = optlen;
298 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
299 if (auio->uio_segflg == UIO_SYSSPACE) {
300 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
301 optlen);
302 } else {
303 error = copyin(auio->uio_iov[i + 1].iov_base,
304 opt->value, optlen);
305 if (error)
306 goto bad;
307 }
308 }
309 }
310 vfs_sanitizeopts(opts);
311 *options = opts;
312 return (0);
313bad:
314 vfs_freeopts(opts);
315 return (error);
316}
317
318/*
319 * Merge the old mount options with the new ones passed
320 * in the MNT_UPDATE case.
321 */
322static void
323vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
324{
325 struct vfsopt *opt, *opt2, *new;
326
327 TAILQ_FOREACH(opt, opts, link) {
328 /*
329 * Check that this option hasn't been redefined
330 * nor cancelled with a "no" mount option.
331 */
332 opt2 = TAILQ_FIRST(toopts);
333 while (opt2 != NULL) {
334 if (strcmp(opt2->name, opt->name) == 0)
335 goto next;
336 if (strncmp(opt2->name, "no", 2) == 0 &&
337 strcmp(opt2->name + 2, opt->name) == 0) {
338 vfs_freeopt(toopts, opt2);
339 goto next;
340 }
341 opt2 = TAILQ_NEXT(opt2, link);
342 }
343 /* We want this option, duplicate it. */
344 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
345 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
346 strcpy(new->name, opt->name);
347 if (opt->len != 0) {
348 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
349 bcopy(opt->value, new->value, opt->len);
350 } else {
351 new->value = NULL;
352 }
353 new->len = opt->len;
354 TAILQ_INSERT_TAIL(toopts, new, link);
355next:
356 continue;
357 }
358}
359
360/*
361 * ---------------------------------------------------------------------
362 * Mount a filesystem
363 */
364int
365nmount(td, uap)
366 struct thread *td;
367 struct nmount_args /* {
368 struct iovec *iovp;
369 unsigned int iovcnt;
370 int flags;
371 } */ *uap;
372{
373 struct uio *auio;
374 struct iovec *iov;
375 unsigned int i;
376 int error;
377 u_int iovcnt;
378
379 AUDIT_ARG(fflags, uap->flags);
380
381 /* Kick out MNT_ROOTFS early as it is legal internally */
382 if (uap->flags & MNT_ROOTFS)
383 return (EINVAL);
384
385 iovcnt = uap->iovcnt;
386 /*
387 * Check that we have an even number of iovec's
388 * and that we have at least two options.
389 */
390 if ((iovcnt & 1) || (iovcnt < 4))
391 return (EINVAL);
392
393 error = copyinuio(uap->iovp, iovcnt, &auio);
394 if (error)
395 return (error);
396 iov = auio->uio_iov;
397 for (i = 0; i < iovcnt; i++) {
398 if (iov->iov_len > MMAXOPTIONLEN) {
399 free(auio, M_IOV);
400 return (EINVAL);
401 }
402 iov++;
403 }
404 error = vfs_donmount(td, uap->flags, auio);
405
406 free(auio, M_IOV);
407 return (error);
408}
409
410/*
411 * ---------------------------------------------------------------------
412 * Various utility functions
413 */
414
415void
416vfs_ref(struct mount *mp)
417{
418
419 MNT_ILOCK(mp);
420 MNT_REF(mp);
421 MNT_IUNLOCK(mp);
422}
423
424void
425vfs_rel(struct mount *mp)
426{
427
428 MNT_ILOCK(mp);
429 MNT_REL(mp);
430 MNT_IUNLOCK(mp);
431}
432
433static int
434mount_init(void *mem, int size, int flags)
435{
436 struct mount *mp;
437
438 mp = (struct mount *)mem;
439 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
440 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
441 return (0);
442}
443
444static void
445mount_fini(void *mem, int size)
446{
447 struct mount *mp;
448
449 mp = (struct mount *)mem;
450 lockdestroy(&mp->mnt_lock);
451 mtx_destroy(&mp->mnt_mtx);
452}
453
454/*
455 * Allocate and initialize the mount point struct.
456 */
457static struct mount *
458vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
459 const char *fspath, struct thread *td)
460{
461 struct mount *mp;
462
463 mp = uma_zalloc(mount_zone, M_WAITOK);
464 bzero(&mp->mnt_startzero,
465 __rangeof(struct mount, mnt_startzero, mnt_endzero));
466 TAILQ_INIT(&mp->mnt_nvnodelist);
467 mp->mnt_nvnodelistsize = 0;
468 mp->mnt_ref = 0;
469 (void) vfs_busy(mp, LK_NOWAIT, 0, td);
470 mp->mnt_op = vfsp->vfc_vfsops;
471 mp->mnt_vfc = vfsp;
472 vfsp->vfc_refcount++; /* XXX Unlocked */
473 mp->mnt_stat.f_type = vfsp->vfc_typenum;
474 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
475 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
476 mp->mnt_vnodecovered = vp;
477 mp->mnt_cred = crdup(td->td_ucred);
478 mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
479 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
480 mp->mnt_iosize_max = DFLTPHYS;
481#ifdef MAC
482 mac_init_mount(mp);
483 mac_create_mount(td->td_ucred, mp);
484#endif
485 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
486 return (mp);
487}
488
489/*
490 * Destroy the mount struct previously allocated by vfs_mount_alloc().
491 */
492static void
493vfs_mount_destroy(struct mount *mp)
494{
495 int i;
496
497 MNT_ILOCK(mp);
498 for (i = 0; mp->mnt_ref && i < 3; i++)
499 msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
500 /*
501 * This will always cause a 3 second delay in rebooting due to
502 * refs on the root mountpoint that never go away. Most of these
503 * are held by init which never exits.
504 */
505 if (i == 3 && (!rebooting || bootverbose))
506 printf("Mount point %s had %d dangling refs\n",
507 mp->mnt_stat.f_mntonname, mp->mnt_ref);
508 if (mp->mnt_holdcnt != 0) {
509 printf("Waiting for mount point to be unheld\n");
510 while (mp->mnt_holdcnt != 0) {
511 mp->mnt_holdcntwaiters++;
512 msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
513 PZERO, "mntdestroy", 0);
514 mp->mnt_holdcntwaiters--;
515 }
516 printf("mount point unheld\n");
517 }
518 if (mp->mnt_writeopcount > 0) {
519 printf("Waiting for mount point write ops\n");
520 while (mp->mnt_writeopcount > 0) {
521 mp->mnt_kern_flag |= MNTK_SUSPEND;
522 msleep(&mp->mnt_writeopcount,
523 MNT_MTX(mp),
524 PZERO, "mntdestroy2", 0);
525 }
526 printf("mount point write ops completed\n");
527 }
528 if (mp->mnt_secondary_writes > 0) {
529 printf("Waiting for mount point secondary write ops\n");
530 while (mp->mnt_secondary_writes > 0) {
531 mp->mnt_kern_flag |= MNTK_SUSPEND;
532 msleep(&mp->mnt_secondary_writes,
533 MNT_MTX(mp),
534 PZERO, "mntdestroy3", 0);
535 }
536 printf("mount point secondary write ops completed\n");
537 }
538 MNT_IUNLOCK(mp);
539 mp->mnt_vfc->vfc_refcount--;
540 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
541 struct vnode *vp;
542
543 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
544 vprint("", vp);
545 panic("unmount: dangling vnode");
546 }
547 MNT_ILOCK(mp);
548 if (mp->mnt_kern_flag & MNTK_MWAIT)
549 wakeup(mp);
550 if (mp->mnt_writeopcount != 0)
551 panic("vfs_mount_destroy: nonzero writeopcount");
552 if (mp->mnt_secondary_writes != 0)
553 panic("vfs_mount_destroy: nonzero secondary_writes");
554 if (mp->mnt_nvnodelistsize != 0)
555 panic("vfs_mount_destroy: nonzero nvnodelistsize");
556 mp->mnt_writeopcount = -1000;
557 mp->mnt_nvnodelistsize = -1000;
558 mp->mnt_secondary_writes = -1000;
559 MNT_IUNLOCK(mp);
560#ifdef MAC
561 mac_destroy_mount(mp);
562#endif
563 if (mp->mnt_opt != NULL)
564 vfs_freeopts(mp->mnt_opt);
565 crfree(mp->mnt_cred);
566 uma_zfree(mount_zone, mp);
567}
568
569static int
570vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
571{
572 struct vfsoptlist *optlist;
573 struct vfsopt *opt, *noro_opt;
574 char *fstype, *fspath, *errmsg;
575 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
576 int has_rw, has_noro;
577
578 errmsg = NULL;
579 errmsg_len = 0;
580 errmsg_pos = -1;
581 has_rw = 0;
582 has_noro = 0;
583
584 error = vfs_buildopts(fsoptions, &optlist);
585 if (error)
586 return (error);
587
588 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
589 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
590
591 /*
592 * We need these two options before the others,
593 * and they are mandatory for any filesystem.
594 * Ensure they are NUL terminated as well.
595 */
596 fstypelen = 0;
597 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
598 if (error || fstype[fstypelen - 1] != '\0') {
599 error = EINVAL;
600 if (errmsg != NULL)
601 strncpy(errmsg, "Invalid fstype", errmsg_len);
602 goto bail;
603 }
604 fspathlen = 0;
605 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
606 if (error || fspath[fspathlen - 1] != '\0') {
607 error = EINVAL;
608 if (errmsg != NULL)
609 strncpy(errmsg, "Invalid fspath", errmsg_len);
610 goto bail;
611 }
612
613 /*
614 * We need to see if we have the "update" option
615 * before we call vfs_domount(), since vfs_domount() has special
616 * logic based on MNT_UPDATE. This is very important
617 * when we want to update the root filesystem.
618 */
619 TAILQ_FOREACH(opt, optlist, link) {
620 if (strcmp(opt->name, "update") == 0)
621 fsflags |= MNT_UPDATE;
622 else if (strcmp(opt->name, "async") == 0)
623 fsflags |= MNT_ASYNC;
624 else if (strcmp(opt->name, "force") == 0)
625 fsflags |= MNT_FORCE;
626 else if (strcmp(opt->name, "multilabel") == 0)
627 fsflags |= MNT_MULTILABEL;
628 else if (strcmp(opt->name, "noasync") == 0)
629 fsflags &= ~MNT_ASYNC;
630 else if (strcmp(opt->name, "noatime") == 0)
631 fsflags |= MNT_NOATIME;
632 else if (strcmp(opt->name, "noclusterr") == 0)
633 fsflags |= MNT_NOCLUSTERR;
634 else if (strcmp(opt->name, "noclusterw") == 0)
635 fsflags |= MNT_NOCLUSTERW;
636 else if (strcmp(opt->name, "noexec") == 0)
637 fsflags |= MNT_NOEXEC;
638 else if (strcmp(opt->name, "nosuid") == 0)
639 fsflags |= MNT_NOSUID;
640 else if (strcmp(opt->name, "nosymfollow") == 0)
641 fsflags |= MNT_NOSYMFOLLOW;
642 else if (strcmp(opt->name, "noro") == 0) {
643 fsflags &= ~MNT_RDONLY;
644 has_noro = 1;
645 }
646 else if (strcmp(opt->name, "rw") == 0) {
647 fsflags &= ~MNT_RDONLY;
648 has_rw = 1;
649 }
650 else if (strcmp(opt->name, "ro") == 0 ||
651 strcmp(opt->name, "rdonly") == 0)
652 fsflags |= MNT_RDONLY;
653 else if (strcmp(opt->name, "snapshot") == 0)
654 fsflags |= MNT_SNAPSHOT;
655 else if (strcmp(opt->name, "suiddir") == 0)
656 fsflags |= MNT_SUIDDIR;
657 else if (strcmp(opt->name, "sync") == 0)
658 fsflags |= MNT_SYNCHRONOUS;
659 else if (strcmp(opt->name, "union") == 0)
660 fsflags |= MNT_UNION;
661 }
662
663 /*
664 * If "rw" was specified as a mount option, and we
665 * are trying to update a mount-point from "ro" to "rw",
666 * we need a mount option "noro", since in vfs_mergeopts(),
667 * "noro" will cancel "ro", but "rw" will not do anything.
668 */
669 if (has_rw && !has_noro) {
670 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
671 noro_opt->name = strdup("noro", M_MOUNT);
672 noro_opt->value = NULL;
673 noro_opt->len = 0;
674 TAILQ_INSERT_TAIL(optlist, noro_opt, link);
675 }
676
677 /*
678 * Be ultra-paranoid about making sure the type and fspath
679 * variables will fit in our mp buffers, including the
680 * terminating NUL.
681 */
682 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
683 error = ENAMETOOLONG;
684 goto bail;
685 }
686
687 mtx_lock(&Giant);
688 error = vfs_domount(td, fstype, fspath, fsflags, optlist);
689 mtx_unlock(&Giant);
690bail:
691 /* copyout the errmsg */
692 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
693 && errmsg_len > 0 && errmsg != NULL) {
694 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
695 bcopy(errmsg,
696 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
697 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
698 } else {
699 copyout(errmsg,
700 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
701 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
702 }
703 }
704
705 if (error != 0)
706 vfs_freeopts(optlist);
707 return (error);
708}
709
710/*
711 * ---------------------------------------------------------------------
712 * Old mount API.
713 */
714#ifndef _SYS_SYSPROTO_H_
715struct mount_args {
716 char *type;
717 char *path;
718 int flags;
719 caddr_t data;
720};
721#endif
722/* ARGSUSED */
723int
724mount(td, uap)
725 struct thread *td;
726 struct mount_args /* {
727 char *type;
728 char *path;
729 int flags;
730 caddr_t data;
731 } */ *uap;
732{
733 char *fstype;
734 struct vfsconf *vfsp = NULL;
735 struct mntarg *ma = NULL;
736 int error;
737
738 AUDIT_ARG(fflags, uap->flags);
739
740 /* Kick out MNT_ROOTFS early as it is legal internally */
741 uap->flags &= ~MNT_ROOTFS;
742
743 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
744 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
745 if (error) {
746 free(fstype, M_TEMP);
747 return (error);
748 }
749
750 AUDIT_ARG(text, fstype);
751 mtx_lock(&Giant);
752 vfsp = vfs_byname_kld(fstype, td, &error);
753 free(fstype, M_TEMP);
754 if (vfsp == NULL) {
755 mtx_unlock(&Giant);
756 return (ENOENT);
757 }
758 if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
759 mtx_unlock(&Giant);
760 return (EOPNOTSUPP);
761 }
762
763 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
764 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
765 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
766 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
767 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
768
769 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
770 mtx_unlock(&Giant);
771 return (error);
772}
773
774
775/*
776 * vfs_domount(): actually attempt a filesystem mount.
777 */
778static int
779vfs_domount(
780 struct thread *td, /* Calling thread. */
781 const char *fstype, /* Filesystem type. */
782 char *fspath, /* Mount path. */
783 int fsflags, /* Flags common to all filesystems. */
784 void *fsdata /* Options local to the filesystem. */
785 )
786{
787 struct vnode *vp;
788 struct mount *mp;
789 struct vfsconf *vfsp;
790 struct export_args export;
791 int error, flag = 0, kern_flag = 0;
792 struct vattr va;
793 struct nameidata nd;
794
795 mtx_assert(&Giant, MA_OWNED);
796 /*
797 * Be ultra-paranoid about making sure the type and fspath
798 * variables will fit in our mp buffers, including the
799 * terminating NUL.
800 */
801 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
802 return (ENAMETOOLONG);
803
804 if (jailed(td->td_ucred))
805 return (EPERM);
806 if (usermount == 0) {
807 if ((error = suser(td)) != 0)
808 return (error);
809 }
810
811 /*
812 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
813 */
814 if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
815 if ((error = suser(td)) != 0)
816 return (error);
817 }
818 /*
819 * Silently enforce MNT_NOSUID and MNT_USER for
820 * unprivileged users.
821 */
822 if (suser(td) != 0)
823 fsflags |= MNT_NOSUID | MNT_USER;
824
825 /* Load KLDs before we lock the covered vnode to avoid reversals. */
826 vfsp = NULL;
827 if ((fsflags & MNT_UPDATE) == 0) {
828 /* Don't try to load KLDs if we're mounting the root. */
829 if (fsflags & MNT_ROOTFS)
830 vfsp = vfs_byname(fstype);
831 else
832 vfsp = vfs_byname_kld(fstype, td, &error);
833 if (vfsp == NULL)
834 return (ENODEV);
835 }
836 /*
837 * Get vnode to be covered
838 */
839 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
840 fspath, td);
841 if ((error = namei(&nd)) != 0)
842 return (error);
843 NDFREE(&nd, NDF_ONLY_PNBUF);
844 vp = nd.ni_vp;
845 if (fsflags & MNT_UPDATE) {
846 if ((vp->v_vflag & VV_ROOT) == 0) {
847 vput(vp);
848 return (EINVAL);
849 }
850 mp = vp->v_mount;
851 flag = mp->mnt_flag;
852 kern_flag = mp->mnt_kern_flag;
853 /*
854 * We only allow the filesystem to be reloaded if it
855 * is currently mounted read-only.
856 */
857 if ((fsflags & MNT_RELOAD) &&
858 ((mp->mnt_flag & MNT_RDONLY) == 0)) {
859 vput(vp);
860 return (EOPNOTSUPP); /* Needs translation */
861 }
862 /*
863 * Only privileged root, or (if MNT_USER is set) the user that
864 * did the original mount is permitted to update it.
865 */
866 error = vfs_suser(mp, td);
867 if (error) {
868 vput(vp);
869 return (error);
870 }
871 if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
872 vput(vp);
873 return (EBUSY);
874 }
875 VI_LOCK(vp);
876 if ((vp->v_iflag & VI_MOUNT) != 0 ||
877 vp->v_mountedhere != NULL) {
878 VI_UNLOCK(vp);
879 vfs_unbusy(mp, td);
880 vput(vp);
881 return (EBUSY);
882 }
883 vp->v_iflag |= VI_MOUNT;
884 VI_UNLOCK(vp);
885 mp->mnt_flag |= fsflags &
886 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
887 VOP_UNLOCK(vp, 0, td);
888 mp->mnt_optnew = fsdata;
889 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
890 } else {
891 /*
892 * If the user is not root, ensure that they own the directory
893 * onto which we are attempting to mount.
894 */
895 error = VOP_GETATTR(vp, &va, td->td_ucred, td);
896 if (error) {
897 vput(vp);
898 return (error);
899 }
900 if (va.va_uid != td->td_ucred->cr_uid) {
901 if ((error = suser(td)) != 0) {
902 vput(vp);
903 return (error);
904 }
905 }
906 error = vinvalbuf(vp, V_SAVE, td, 0, 0);
907 if (error != 0) {
908 vput(vp);
909 return (error);
910 }
911 if (vp->v_type != VDIR) {
912 vput(vp);
913 return (ENOTDIR);
914 }
915 VI_LOCK(vp);
916 if ((vp->v_iflag & VI_MOUNT) != 0 ||
917 vp->v_mountedhere != NULL) {
918 VI_UNLOCK(vp);
919 vput(vp);
920 return (EBUSY);
921 }
922 vp->v_iflag |= VI_MOUNT;
923 VI_UNLOCK(vp);
924
925 /*
926 * Allocate and initialize the filesystem.
927 */
928 mp = vfs_mount_alloc(vp, vfsp, fspath, td);
929 VOP_UNLOCK(vp, 0, td);
930
931 /* XXXMAC: pass to vfs_mount_alloc? */
932 mp->mnt_optnew = fsdata;
933 }
934
935 /*
936 * Set the mount level flags.
937 */
938 if (fsflags & MNT_RDONLY)
939 mp->mnt_flag |= MNT_RDONLY;
940 mp->mnt_flag &=~ MNT_UPDATEMASK;
941 mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS);
942 /*
943 * Mount the filesystem.
944 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
945 * get. No freeing of cn_pnbuf.
946 */
947 error = VFS_MOUNT(mp, td);
948
949 /*
950 * Process the export option only if we are
951 * updating mount options.
952 */
953 if (!error && (fsflags & MNT_UPDATE)) {
954 if (vfs_copyopt(mp->mnt_optnew, "export", &export,
955 sizeof(export)) == 0)
956 error = vfs_export(mp, &export);
957 }
958
959 if (!error) {
960 if (mp->mnt_opt != NULL)
961 vfs_freeopts(mp->mnt_opt);
962 mp->mnt_opt = mp->mnt_optnew;
963 (void)VFS_STATFS(mp, &mp->mnt_stat, td);
964 }
965 /*
966 * Prevent external consumers of mount options from reading
967 * mnt_optnew.
968 */
969 mp->mnt_optnew = NULL;
970 if (mp->mnt_flag & MNT_UPDATE) {
971 mp->mnt_flag &=
972 ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
973 if (error) {
974 mp->mnt_flag = flag;
975 mp->mnt_kern_flag = kern_flag;
976 }
977 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
978 if (mp->mnt_syncer == NULL)
979 error = vfs_allocate_syncvnode(mp);
980 } else {
981 if (mp->mnt_syncer != NULL)
982 vrele(mp->mnt_syncer);
983 mp->mnt_syncer = NULL;
984 }
985 vfs_unbusy(mp, td);
986 VI_LOCK(vp);
987 vp->v_iflag &= ~VI_MOUNT;
988 VI_UNLOCK(vp);
989 vrele(vp);
990 return (error);
991 }
992 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
993 /*
994 * Put the new filesystem on the mount list after root.
995 */
996 cache_purge(vp);
997 if (!error) {
998 struct vnode *newdp;
999
1000 VI_LOCK(vp);
1001 vp->v_iflag &= ~VI_MOUNT;
1002 VI_UNLOCK(vp);
1003 vp->v_mountedhere = mp;
1004 mtx_lock(&mountlist_mtx);
1005 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1006 mtx_unlock(&mountlist_mtx);
1007 vfs_event_signal(NULL, VQ_MOUNT, 0);
1008 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1009 panic("mount: lost mount");
1010 mountcheckdirs(vp, newdp);
1011 vput(newdp);
1012 VOP_UNLOCK(vp, 0, td);
1013 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1014 error = vfs_allocate_syncvnode(mp);
1015 vfs_unbusy(mp, td);
1016 if (error)
1017 vrele(vp);
1018 } else {
1019 VI_LOCK(vp);
1020 vp->v_iflag &= ~VI_MOUNT;
1021 VI_UNLOCK(vp);
1022 vfs_unbusy(mp, td);
1023 vfs_mount_destroy(mp);
1024 vput(vp);
1025 }
1026 return (error);
1027}
1028
1029/*
1030 * ---------------------------------------------------------------------
1031 * Unmount a filesystem.
1032 *
1033 * Note: unmount takes a path to the vnode mounted on as argument,
1034 * not special file (as before).
1035 */
1036#ifndef _SYS_SYSPROTO_H_
1037struct unmount_args {
1038 char *path;
1039 int flags;
1040};
1041#endif
1042/* ARGSUSED */
1043int
1044unmount(td, uap)
1045 struct thread *td;
1046 register struct unmount_args /* {
1047 char *path;
1048 int flags;
1049 } */ *uap;
1050{
1051 struct mount *mp;
1052 char *pathbuf;
1053 int error, id0, id1;
1054
1055 if (jailed(td->td_ucred))
1056 return (EPERM);
1057 if (usermount == 0) {
1058 if ((error = suser(td)) != 0)
1059 return (error);
1060 }
1061
1062 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1063 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1064 if (error) {
1065 free(pathbuf, M_TEMP);
1066 return (error);
1067 }
1068 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1069 mtx_lock(&Giant);
1070 if (uap->flags & MNT_BYFSID) {
1071 /* Decode the filesystem ID. */
1072 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1073 mtx_unlock(&Giant);
1074 free(pathbuf, M_TEMP);
1075 return (EINVAL);
1076 }
1077
1078 mtx_lock(&mountlist_mtx);
1079 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1080 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1081 mp->mnt_stat.f_fsid.val[1] == id1)
1082 break;
1083 }
1084 mtx_unlock(&mountlist_mtx);
1085 } else {
1086 mtx_lock(&mountlist_mtx);
1087 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1088 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1089 break;
1090 }
1091 mtx_unlock(&mountlist_mtx);
1092 }
1093 free(pathbuf, M_TEMP);
1094 if (mp == NULL) {
1095 /*
1096 * Previously we returned ENOENT for a nonexistent path and
1097 * EINVAL for a non-mountpoint. We cannot tell these apart
1098 * now, so in the !MNT_BYFSID case return the more likely
1099 * EINVAL for compatibility.
1100 */
1101 mtx_unlock(&Giant);
1102 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1103 }
1104
1105 /*
1106 * Don't allow unmounting the root filesystem.
1107 */
1108 if (mp->mnt_flag & MNT_ROOTFS) {
1109 mtx_unlock(&Giant);
1110 return (EINVAL);
1111 }
1112 error = dounmount(mp, uap->flags, td);
1113 mtx_unlock(&Giant);
1114 return (error);
1115}
1116
1117/*
1118 * Do the actual filesystem unmount.
1119 */
1120int
1121dounmount(mp, flags, td)
1122 struct mount *mp;
1123 int flags;
1124 struct thread *td;
1125{
1126 struct vnode *coveredvp, *fsrootvp;
1127 int error;
1128 int async_flag;
1129
1130 mtx_assert(&Giant, MA_OWNED);
1131
1132 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1133 VI_LOCK(coveredvp);
1134 vholdl(coveredvp);
1135 error = vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK, td);
1136 vdrop(coveredvp);
1137 /*
1138 * Check for mp being unmounted while waiting for the
1139 * covered vnode lock.
1140 */
1141 if (error)
1142 return (error);
1143 if (coveredvp->v_mountedhere != mp) {
1144 VOP_UNLOCK(coveredvp, 0, td);
1145 return (EBUSY);
1146 }
1147 }
1148 /*
1149 * Only privileged root, or (if MNT_USER is set) the user that did the
1150 * original mount is permitted to unmount this filesystem.
1151 */
1152 error = vfs_suser(mp, td);
1153 if (error) {
| 39
40#include <sys/param.h>
41#include <sys/conf.h>
42#include <sys/jail.h>
43#include <sys/kernel.h>
44#include <sys/libkern.h>
45#include <sys/mac.h>
46#include <sys/malloc.h>
47#include <sys/mount.h>
48#include <sys/mutex.h>
49#include <sys/namei.h>
50#include <sys/proc.h>
51#include <sys/filedesc.h>
52#include <sys/reboot.h>
53#include <sys/syscallsubr.h>
54#include <sys/sysproto.h>
55#include <sys/sx.h>
56#include <sys/sysctl.h>
57#include <sys/sysent.h>
58#include <sys/systm.h>
59#include <sys/vnode.h>
60#include <vm/uma.h>
61
62#include <geom/geom.h>
63
64#include <machine/stdarg.h>
65
66#include <security/audit/audit.h>
67
68#include "opt_rootdevname.h"
69#include "opt_ddb.h"
70#include "opt_mac.h"
71
72#ifdef DDB
73#include <ddb/ddb.h>
74#endif
75
76#define ROOTNAME "root_device"
77#define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
78
79static int vfs_domount(struct thread *td, const char *fstype,
80 char *fspath, int fsflags, void *fsdata);
81static struct mount *vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
82 const char *fspath, struct thread *td);
83static int vfs_mountroot_ask(void);
84static int vfs_mountroot_try(const char *mountfrom);
85static int vfs_donmount(struct thread *td, int fsflags,
86 struct uio *fsoptions);
87static void free_mntarg(struct mntarg *ma);
88static void vfs_mount_destroy(struct mount *);
89static int vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
90
91static int usermount = 0;
92SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
93 "Unprivileged users may mount and unmount file systems");
94
95MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
96MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
97static uma_zone_t mount_zone;
98
99/* List of mounted filesystems. */
100struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
101
102/* For any iteration/modification of mountlist */
103struct mtx mountlist_mtx;
104MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
105
106TAILQ_HEAD(vfsoptlist, vfsopt);
107struct vfsopt {
108 TAILQ_ENTRY(vfsopt) link;
109 char *name;
110 void *value;
111 int len;
112};
113
114/*
115 * The vnode of the system's root (/ in the filesystem, without chroot
116 * active.)
117 */
118struct vnode *rootvnode;
119
120/*
121 * The root filesystem is detailed in the kernel environment variable
122 * vfs.root.mountfrom, which is expected to be in the general format
123 *
124 * <vfsname>:[<path>]
125 * vfsname := the name of a VFS known to the kernel and capable
126 * of being mounted as root
127 * path := disk device name or other data used by the filesystem
128 * to locate its physical store
129 */
130
131/*
132 * Global opts, taken by all filesystems
133 */
134static const char *global_opts[] = {
135 "errmsg",
136 "fstype",
137 "fspath",
138 "rdonly",
139 "ro",
140 "rw",
141 "suid",
142 "exec",
143 "update",
144 NULL
145};
146
147/*
148 * The root specifiers we will try if RB_CDROM is specified.
149 */
150static char *cdrom_rootdevnames[] = {
151 "cd9660:cd0",
152 "cd9660:acd0",
153 NULL
154};
155
156/* legacy find-root code */
157char *rootdevnames[2] = {NULL, NULL};
158#ifndef ROOTDEVNAME
159# define ROOTDEVNAME NULL
160#endif
161static const char *ctrootdevname = ROOTDEVNAME;
162
163/*
164 * ---------------------------------------------------------------------
165 * Functions for building and sanitizing the mount options
166 */
167
168/* Remove one mount option. */
169static void
170vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
171{
172
173 TAILQ_REMOVE(opts, opt, link);
174 free(opt->name, M_MOUNT);
175 if (opt->value != NULL)
176 free(opt->value, M_MOUNT);
177#ifdef INVARIANTS
178 else if (opt->len != 0)
179 panic("%s: mount option with NULL value but length != 0",
180 __func__);
181#endif
182 free(opt, M_MOUNT);
183}
184
185/* Release all resources related to the mount options. */
186static void
187vfs_freeopts(struct vfsoptlist *opts)
188{
189 struct vfsopt *opt;
190
191 while (!TAILQ_EMPTY(opts)) {
192 opt = TAILQ_FIRST(opts);
193 vfs_freeopt(opts, opt);
194 }
195 free(opts, M_MOUNT);
196}
197
198/*
199 * Check if options are equal (with or without the "no" prefix).
200 */
201static int
202vfs_equalopts(const char *opt1, const char *opt2)
203{
204
205 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
206 if (strcmp(opt1, opt2) == 0)
207 return (1);
208 /* "noopt" vs. "opt" */
209 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
210 return (1);
211 /* "opt" vs. "noopt" */
212 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
213 return (1);
214 return (0);
215}
216
217/*
218 * If a mount option is specified several times,
219 * (with or without the "no" prefix) only keep
220 * the last occurence of it.
221 */
222static void
223vfs_sanitizeopts(struct vfsoptlist *opts)
224{
225 struct vfsopt *opt, *opt2, *tmp;
226
227 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
228 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
229 while (opt2 != NULL) {
230 if (vfs_equalopts(opt->name, opt2->name)) {
231 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
232 vfs_freeopt(opts, opt2);
233 opt2 = tmp;
234 } else {
235 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
236 }
237 }
238 }
239}
240
241/*
242 * Build a linked list of mount options from a struct uio.
243 */
244static int
245vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
246{
247 struct vfsoptlist *opts;
248 struct vfsopt *opt;
249 size_t memused;
250 unsigned int i, iovcnt;
251 int error, namelen, optlen;
252
253 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
254 TAILQ_INIT(opts);
255 memused = 0;
256 iovcnt = auio->uio_iovcnt;
257 for (i = 0; i < iovcnt; i += 2) {
258 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
259 namelen = auio->uio_iov[i].iov_len;
260 optlen = auio->uio_iov[i + 1].iov_len;
261 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
262 opt->value = NULL;
263 opt->len = 0;
264
265 /*
266 * Do this early, so jumps to "bad" will free the current
267 * option.
268 */
269 TAILQ_INSERT_TAIL(opts, opt, link);
270 memused += sizeof(struct vfsopt) + optlen + namelen;
271
272 /*
273 * Avoid consuming too much memory, and attempts to overflow
274 * memused.
275 */
276 if (memused > VFS_MOUNTARG_SIZE_MAX ||
277 optlen > VFS_MOUNTARG_SIZE_MAX ||
278 namelen > VFS_MOUNTARG_SIZE_MAX) {
279 error = EINVAL;
280 goto bad;
281 }
282
283 if (auio->uio_segflg == UIO_SYSSPACE) {
284 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
285 } else {
286 error = copyin(auio->uio_iov[i].iov_base, opt->name,
287 namelen);
288 if (error)
289 goto bad;
290 }
291 /* Ensure names are null-terminated strings. */
292 if (opt->name[namelen - 1] != '\0') {
293 error = EINVAL;
294 goto bad;
295 }
296 if (optlen != 0) {
297 opt->len = optlen;
298 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
299 if (auio->uio_segflg == UIO_SYSSPACE) {
300 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
301 optlen);
302 } else {
303 error = copyin(auio->uio_iov[i + 1].iov_base,
304 opt->value, optlen);
305 if (error)
306 goto bad;
307 }
308 }
309 }
310 vfs_sanitizeopts(opts);
311 *options = opts;
312 return (0);
313bad:
314 vfs_freeopts(opts);
315 return (error);
316}
317
318/*
319 * Merge the old mount options with the new ones passed
320 * in the MNT_UPDATE case.
321 */
322static void
323vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
324{
325 struct vfsopt *opt, *opt2, *new;
326
327 TAILQ_FOREACH(opt, opts, link) {
328 /*
329 * Check that this option hasn't been redefined
330 * nor cancelled with a "no" mount option.
331 */
332 opt2 = TAILQ_FIRST(toopts);
333 while (opt2 != NULL) {
334 if (strcmp(opt2->name, opt->name) == 0)
335 goto next;
336 if (strncmp(opt2->name, "no", 2) == 0 &&
337 strcmp(opt2->name + 2, opt->name) == 0) {
338 vfs_freeopt(toopts, opt2);
339 goto next;
340 }
341 opt2 = TAILQ_NEXT(opt2, link);
342 }
343 /* We want this option, duplicate it. */
344 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
345 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
346 strcpy(new->name, opt->name);
347 if (opt->len != 0) {
348 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
349 bcopy(opt->value, new->value, opt->len);
350 } else {
351 new->value = NULL;
352 }
353 new->len = opt->len;
354 TAILQ_INSERT_TAIL(toopts, new, link);
355next:
356 continue;
357 }
358}
359
360/*
361 * ---------------------------------------------------------------------
362 * Mount a filesystem
363 */
364int
365nmount(td, uap)
366 struct thread *td;
367 struct nmount_args /* {
368 struct iovec *iovp;
369 unsigned int iovcnt;
370 int flags;
371 } */ *uap;
372{
373 struct uio *auio;
374 struct iovec *iov;
375 unsigned int i;
376 int error;
377 u_int iovcnt;
378
379 AUDIT_ARG(fflags, uap->flags);
380
381 /* Kick out MNT_ROOTFS early as it is legal internally */
382 if (uap->flags & MNT_ROOTFS)
383 return (EINVAL);
384
385 iovcnt = uap->iovcnt;
386 /*
387 * Check that we have an even number of iovec's
388 * and that we have at least two options.
389 */
390 if ((iovcnt & 1) || (iovcnt < 4))
391 return (EINVAL);
392
393 error = copyinuio(uap->iovp, iovcnt, &auio);
394 if (error)
395 return (error);
396 iov = auio->uio_iov;
397 for (i = 0; i < iovcnt; i++) {
398 if (iov->iov_len > MMAXOPTIONLEN) {
399 free(auio, M_IOV);
400 return (EINVAL);
401 }
402 iov++;
403 }
404 error = vfs_donmount(td, uap->flags, auio);
405
406 free(auio, M_IOV);
407 return (error);
408}
409
410/*
411 * ---------------------------------------------------------------------
412 * Various utility functions
413 */
414
415void
416vfs_ref(struct mount *mp)
417{
418
419 MNT_ILOCK(mp);
420 MNT_REF(mp);
421 MNT_IUNLOCK(mp);
422}
423
424void
425vfs_rel(struct mount *mp)
426{
427
428 MNT_ILOCK(mp);
429 MNT_REL(mp);
430 MNT_IUNLOCK(mp);
431}
432
433static int
434mount_init(void *mem, int size, int flags)
435{
436 struct mount *mp;
437
438 mp = (struct mount *)mem;
439 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
440 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
441 return (0);
442}
443
444static void
445mount_fini(void *mem, int size)
446{
447 struct mount *mp;
448
449 mp = (struct mount *)mem;
450 lockdestroy(&mp->mnt_lock);
451 mtx_destroy(&mp->mnt_mtx);
452}
453
454/*
455 * Allocate and initialize the mount point struct.
456 */
457static struct mount *
458vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
459 const char *fspath, struct thread *td)
460{
461 struct mount *mp;
462
463 mp = uma_zalloc(mount_zone, M_WAITOK);
464 bzero(&mp->mnt_startzero,
465 __rangeof(struct mount, mnt_startzero, mnt_endzero));
466 TAILQ_INIT(&mp->mnt_nvnodelist);
467 mp->mnt_nvnodelistsize = 0;
468 mp->mnt_ref = 0;
469 (void) vfs_busy(mp, LK_NOWAIT, 0, td);
470 mp->mnt_op = vfsp->vfc_vfsops;
471 mp->mnt_vfc = vfsp;
472 vfsp->vfc_refcount++; /* XXX Unlocked */
473 mp->mnt_stat.f_type = vfsp->vfc_typenum;
474 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
475 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
476 mp->mnt_vnodecovered = vp;
477 mp->mnt_cred = crdup(td->td_ucred);
478 mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
479 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
480 mp->mnt_iosize_max = DFLTPHYS;
481#ifdef MAC
482 mac_init_mount(mp);
483 mac_create_mount(td->td_ucred, mp);
484#endif
485 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
486 return (mp);
487}
488
489/*
490 * Destroy the mount struct previously allocated by vfs_mount_alloc().
491 */
492static void
493vfs_mount_destroy(struct mount *mp)
494{
495 int i;
496
497 MNT_ILOCK(mp);
498 for (i = 0; mp->mnt_ref && i < 3; i++)
499 msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
500 /*
501 * This will always cause a 3 second delay in rebooting due to
502 * refs on the root mountpoint that never go away. Most of these
503 * are held by init which never exits.
504 */
505 if (i == 3 && (!rebooting || bootverbose))
506 printf("Mount point %s had %d dangling refs\n",
507 mp->mnt_stat.f_mntonname, mp->mnt_ref);
508 if (mp->mnt_holdcnt != 0) {
509 printf("Waiting for mount point to be unheld\n");
510 while (mp->mnt_holdcnt != 0) {
511 mp->mnt_holdcntwaiters++;
512 msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
513 PZERO, "mntdestroy", 0);
514 mp->mnt_holdcntwaiters--;
515 }
516 printf("mount point unheld\n");
517 }
518 if (mp->mnt_writeopcount > 0) {
519 printf("Waiting for mount point write ops\n");
520 while (mp->mnt_writeopcount > 0) {
521 mp->mnt_kern_flag |= MNTK_SUSPEND;
522 msleep(&mp->mnt_writeopcount,
523 MNT_MTX(mp),
524 PZERO, "mntdestroy2", 0);
525 }
526 printf("mount point write ops completed\n");
527 }
528 if (mp->mnt_secondary_writes > 0) {
529 printf("Waiting for mount point secondary write ops\n");
530 while (mp->mnt_secondary_writes > 0) {
531 mp->mnt_kern_flag |= MNTK_SUSPEND;
532 msleep(&mp->mnt_secondary_writes,
533 MNT_MTX(mp),
534 PZERO, "mntdestroy3", 0);
535 }
536 printf("mount point secondary write ops completed\n");
537 }
538 MNT_IUNLOCK(mp);
539 mp->mnt_vfc->vfc_refcount--;
540 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
541 struct vnode *vp;
542
543 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
544 vprint("", vp);
545 panic("unmount: dangling vnode");
546 }
547 MNT_ILOCK(mp);
548 if (mp->mnt_kern_flag & MNTK_MWAIT)
549 wakeup(mp);
550 if (mp->mnt_writeopcount != 0)
551 panic("vfs_mount_destroy: nonzero writeopcount");
552 if (mp->mnt_secondary_writes != 0)
553 panic("vfs_mount_destroy: nonzero secondary_writes");
554 if (mp->mnt_nvnodelistsize != 0)
555 panic("vfs_mount_destroy: nonzero nvnodelistsize");
556 mp->mnt_writeopcount = -1000;
557 mp->mnt_nvnodelistsize = -1000;
558 mp->mnt_secondary_writes = -1000;
559 MNT_IUNLOCK(mp);
560#ifdef MAC
561 mac_destroy_mount(mp);
562#endif
563 if (mp->mnt_opt != NULL)
564 vfs_freeopts(mp->mnt_opt);
565 crfree(mp->mnt_cred);
566 uma_zfree(mount_zone, mp);
567}
568
569static int
570vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
571{
572 struct vfsoptlist *optlist;
573 struct vfsopt *opt, *noro_opt;
574 char *fstype, *fspath, *errmsg;
575 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
576 int has_rw, has_noro;
577
578 errmsg = NULL;
579 errmsg_len = 0;
580 errmsg_pos = -1;
581 has_rw = 0;
582 has_noro = 0;
583
584 error = vfs_buildopts(fsoptions, &optlist);
585 if (error)
586 return (error);
587
588 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
589 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
590
591 /*
592 * We need these two options before the others,
593 * and they are mandatory for any filesystem.
594 * Ensure they are NUL terminated as well.
595 */
596 fstypelen = 0;
597 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
598 if (error || fstype[fstypelen - 1] != '\0') {
599 error = EINVAL;
600 if (errmsg != NULL)
601 strncpy(errmsg, "Invalid fstype", errmsg_len);
602 goto bail;
603 }
604 fspathlen = 0;
605 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
606 if (error || fspath[fspathlen - 1] != '\0') {
607 error = EINVAL;
608 if (errmsg != NULL)
609 strncpy(errmsg, "Invalid fspath", errmsg_len);
610 goto bail;
611 }
612
613 /*
614 * We need to see if we have the "update" option
615 * before we call vfs_domount(), since vfs_domount() has special
616 * logic based on MNT_UPDATE. This is very important
617 * when we want to update the root filesystem.
618 */
619 TAILQ_FOREACH(opt, optlist, link) {
620 if (strcmp(opt->name, "update") == 0)
621 fsflags |= MNT_UPDATE;
622 else if (strcmp(opt->name, "async") == 0)
623 fsflags |= MNT_ASYNC;
624 else if (strcmp(opt->name, "force") == 0)
625 fsflags |= MNT_FORCE;
626 else if (strcmp(opt->name, "multilabel") == 0)
627 fsflags |= MNT_MULTILABEL;
628 else if (strcmp(opt->name, "noasync") == 0)
629 fsflags &= ~MNT_ASYNC;
630 else if (strcmp(opt->name, "noatime") == 0)
631 fsflags |= MNT_NOATIME;
632 else if (strcmp(opt->name, "noclusterr") == 0)
633 fsflags |= MNT_NOCLUSTERR;
634 else if (strcmp(opt->name, "noclusterw") == 0)
635 fsflags |= MNT_NOCLUSTERW;
636 else if (strcmp(opt->name, "noexec") == 0)
637 fsflags |= MNT_NOEXEC;
638 else if (strcmp(opt->name, "nosuid") == 0)
639 fsflags |= MNT_NOSUID;
640 else if (strcmp(opt->name, "nosymfollow") == 0)
641 fsflags |= MNT_NOSYMFOLLOW;
642 else if (strcmp(opt->name, "noro") == 0) {
643 fsflags &= ~MNT_RDONLY;
644 has_noro = 1;
645 }
646 else if (strcmp(opt->name, "rw") == 0) {
647 fsflags &= ~MNT_RDONLY;
648 has_rw = 1;
649 }
650 else if (strcmp(opt->name, "ro") == 0 ||
651 strcmp(opt->name, "rdonly") == 0)
652 fsflags |= MNT_RDONLY;
653 else if (strcmp(opt->name, "snapshot") == 0)
654 fsflags |= MNT_SNAPSHOT;
655 else if (strcmp(opt->name, "suiddir") == 0)
656 fsflags |= MNT_SUIDDIR;
657 else if (strcmp(opt->name, "sync") == 0)
658 fsflags |= MNT_SYNCHRONOUS;
659 else if (strcmp(opt->name, "union") == 0)
660 fsflags |= MNT_UNION;
661 }
662
663 /*
664 * If "rw" was specified as a mount option, and we
665 * are trying to update a mount-point from "ro" to "rw",
666 * we need a mount option "noro", since in vfs_mergeopts(),
667 * "noro" will cancel "ro", but "rw" will not do anything.
668 */
669 if (has_rw && !has_noro) {
670 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
671 noro_opt->name = strdup("noro", M_MOUNT);
672 noro_opt->value = NULL;
673 noro_opt->len = 0;
674 TAILQ_INSERT_TAIL(optlist, noro_opt, link);
675 }
676
677 /*
678 * Be ultra-paranoid about making sure the type and fspath
679 * variables will fit in our mp buffers, including the
680 * terminating NUL.
681 */
682 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
683 error = ENAMETOOLONG;
684 goto bail;
685 }
686
687 mtx_lock(&Giant);
688 error = vfs_domount(td, fstype, fspath, fsflags, optlist);
689 mtx_unlock(&Giant);
690bail:
691 /* copyout the errmsg */
692 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
693 && errmsg_len > 0 && errmsg != NULL) {
694 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
695 bcopy(errmsg,
696 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
697 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
698 } else {
699 copyout(errmsg,
700 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
701 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
702 }
703 }
704
705 if (error != 0)
706 vfs_freeopts(optlist);
707 return (error);
708}
709
710/*
711 * ---------------------------------------------------------------------
712 * Old mount API.
713 */
714#ifndef _SYS_SYSPROTO_H_
715struct mount_args {
716 char *type;
717 char *path;
718 int flags;
719 caddr_t data;
720};
721#endif
722/* ARGSUSED */
723int
724mount(td, uap)
725 struct thread *td;
726 struct mount_args /* {
727 char *type;
728 char *path;
729 int flags;
730 caddr_t data;
731 } */ *uap;
732{
733 char *fstype;
734 struct vfsconf *vfsp = NULL;
735 struct mntarg *ma = NULL;
736 int error;
737
738 AUDIT_ARG(fflags, uap->flags);
739
740 /* Kick out MNT_ROOTFS early as it is legal internally */
741 uap->flags &= ~MNT_ROOTFS;
742
743 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
744 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
745 if (error) {
746 free(fstype, M_TEMP);
747 return (error);
748 }
749
750 AUDIT_ARG(text, fstype);
751 mtx_lock(&Giant);
752 vfsp = vfs_byname_kld(fstype, td, &error);
753 free(fstype, M_TEMP);
754 if (vfsp == NULL) {
755 mtx_unlock(&Giant);
756 return (ENOENT);
757 }
758 if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
759 mtx_unlock(&Giant);
760 return (EOPNOTSUPP);
761 }
762
763 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
764 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
765 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
766 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
767 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
768
769 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
770 mtx_unlock(&Giant);
771 return (error);
772}
773
774
775/*
776 * vfs_domount(): actually attempt a filesystem mount.
777 */
778static int
779vfs_domount(
780 struct thread *td, /* Calling thread. */
781 const char *fstype, /* Filesystem type. */
782 char *fspath, /* Mount path. */
783 int fsflags, /* Flags common to all filesystems. */
784 void *fsdata /* Options local to the filesystem. */
785 )
786{
787 struct vnode *vp;
788 struct mount *mp;
789 struct vfsconf *vfsp;
790 struct export_args export;
791 int error, flag = 0, kern_flag = 0;
792 struct vattr va;
793 struct nameidata nd;
794
795 mtx_assert(&Giant, MA_OWNED);
796 /*
797 * Be ultra-paranoid about making sure the type and fspath
798 * variables will fit in our mp buffers, including the
799 * terminating NUL.
800 */
801 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
802 return (ENAMETOOLONG);
803
804 if (jailed(td->td_ucred))
805 return (EPERM);
806 if (usermount == 0) {
807 if ((error = suser(td)) != 0)
808 return (error);
809 }
810
811 /*
812 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
813 */
814 if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
815 if ((error = suser(td)) != 0)
816 return (error);
817 }
818 /*
819 * Silently enforce MNT_NOSUID and MNT_USER for
820 * unprivileged users.
821 */
822 if (suser(td) != 0)
823 fsflags |= MNT_NOSUID | MNT_USER;
824
825 /* Load KLDs before we lock the covered vnode to avoid reversals. */
826 vfsp = NULL;
827 if ((fsflags & MNT_UPDATE) == 0) {
828 /* Don't try to load KLDs if we're mounting the root. */
829 if (fsflags & MNT_ROOTFS)
830 vfsp = vfs_byname(fstype);
831 else
832 vfsp = vfs_byname_kld(fstype, td, &error);
833 if (vfsp == NULL)
834 return (ENODEV);
835 }
836 /*
837 * Get vnode to be covered
838 */
839 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
840 fspath, td);
841 if ((error = namei(&nd)) != 0)
842 return (error);
843 NDFREE(&nd, NDF_ONLY_PNBUF);
844 vp = nd.ni_vp;
845 if (fsflags & MNT_UPDATE) {
846 if ((vp->v_vflag & VV_ROOT) == 0) {
847 vput(vp);
848 return (EINVAL);
849 }
850 mp = vp->v_mount;
851 flag = mp->mnt_flag;
852 kern_flag = mp->mnt_kern_flag;
853 /*
854 * We only allow the filesystem to be reloaded if it
855 * is currently mounted read-only.
856 */
857 if ((fsflags & MNT_RELOAD) &&
858 ((mp->mnt_flag & MNT_RDONLY) == 0)) {
859 vput(vp);
860 return (EOPNOTSUPP); /* Needs translation */
861 }
862 /*
863 * Only privileged root, or (if MNT_USER is set) the user that
864 * did the original mount is permitted to update it.
865 */
866 error = vfs_suser(mp, td);
867 if (error) {
868 vput(vp);
869 return (error);
870 }
871 if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
872 vput(vp);
873 return (EBUSY);
874 }
875 VI_LOCK(vp);
876 if ((vp->v_iflag & VI_MOUNT) != 0 ||
877 vp->v_mountedhere != NULL) {
878 VI_UNLOCK(vp);
879 vfs_unbusy(mp, td);
880 vput(vp);
881 return (EBUSY);
882 }
883 vp->v_iflag |= VI_MOUNT;
884 VI_UNLOCK(vp);
885 mp->mnt_flag |= fsflags &
886 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
887 VOP_UNLOCK(vp, 0, td);
888 mp->mnt_optnew = fsdata;
889 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
890 } else {
891 /*
892 * If the user is not root, ensure that they own the directory
893 * onto which we are attempting to mount.
894 */
895 error = VOP_GETATTR(vp, &va, td->td_ucred, td);
896 if (error) {
897 vput(vp);
898 return (error);
899 }
900 if (va.va_uid != td->td_ucred->cr_uid) {
901 if ((error = suser(td)) != 0) {
902 vput(vp);
903 return (error);
904 }
905 }
906 error = vinvalbuf(vp, V_SAVE, td, 0, 0);
907 if (error != 0) {
908 vput(vp);
909 return (error);
910 }
911 if (vp->v_type != VDIR) {
912 vput(vp);
913 return (ENOTDIR);
914 }
915 VI_LOCK(vp);
916 if ((vp->v_iflag & VI_MOUNT) != 0 ||
917 vp->v_mountedhere != NULL) {
918 VI_UNLOCK(vp);
919 vput(vp);
920 return (EBUSY);
921 }
922 vp->v_iflag |= VI_MOUNT;
923 VI_UNLOCK(vp);
924
925 /*
926 * Allocate and initialize the filesystem.
927 */
928 mp = vfs_mount_alloc(vp, vfsp, fspath, td);
929 VOP_UNLOCK(vp, 0, td);
930
931 /* XXXMAC: pass to vfs_mount_alloc? */
932 mp->mnt_optnew = fsdata;
933 }
934
935 /*
936 * Set the mount level flags.
937 */
938 if (fsflags & MNT_RDONLY)
939 mp->mnt_flag |= MNT_RDONLY;
940 mp->mnt_flag &=~ MNT_UPDATEMASK;
941 mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS);
942 /*
943 * Mount the filesystem.
944 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
945 * get. No freeing of cn_pnbuf.
946 */
947 error = VFS_MOUNT(mp, td);
948
949 /*
950 * Process the export option only if we are
951 * updating mount options.
952 */
953 if (!error && (fsflags & MNT_UPDATE)) {
954 if (vfs_copyopt(mp->mnt_optnew, "export", &export,
955 sizeof(export)) == 0)
956 error = vfs_export(mp, &export);
957 }
958
959 if (!error) {
960 if (mp->mnt_opt != NULL)
961 vfs_freeopts(mp->mnt_opt);
962 mp->mnt_opt = mp->mnt_optnew;
963 (void)VFS_STATFS(mp, &mp->mnt_stat, td);
964 }
965 /*
966 * Prevent external consumers of mount options from reading
967 * mnt_optnew.
968 */
969 mp->mnt_optnew = NULL;
970 if (mp->mnt_flag & MNT_UPDATE) {
971 mp->mnt_flag &=
972 ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
973 if (error) {
974 mp->mnt_flag = flag;
975 mp->mnt_kern_flag = kern_flag;
976 }
977 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
978 if (mp->mnt_syncer == NULL)
979 error = vfs_allocate_syncvnode(mp);
980 } else {
981 if (mp->mnt_syncer != NULL)
982 vrele(mp->mnt_syncer);
983 mp->mnt_syncer = NULL;
984 }
985 vfs_unbusy(mp, td);
986 VI_LOCK(vp);
987 vp->v_iflag &= ~VI_MOUNT;
988 VI_UNLOCK(vp);
989 vrele(vp);
990 return (error);
991 }
992 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
993 /*
994 * Put the new filesystem on the mount list after root.
995 */
996 cache_purge(vp);
997 if (!error) {
998 struct vnode *newdp;
999
1000 VI_LOCK(vp);
1001 vp->v_iflag &= ~VI_MOUNT;
1002 VI_UNLOCK(vp);
1003 vp->v_mountedhere = mp;
1004 mtx_lock(&mountlist_mtx);
1005 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1006 mtx_unlock(&mountlist_mtx);
1007 vfs_event_signal(NULL, VQ_MOUNT, 0);
1008 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1009 panic("mount: lost mount");
1010 mountcheckdirs(vp, newdp);
1011 vput(newdp);
1012 VOP_UNLOCK(vp, 0, td);
1013 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1014 error = vfs_allocate_syncvnode(mp);
1015 vfs_unbusy(mp, td);
1016 if (error)
1017 vrele(vp);
1018 } else {
1019 VI_LOCK(vp);
1020 vp->v_iflag &= ~VI_MOUNT;
1021 VI_UNLOCK(vp);
1022 vfs_unbusy(mp, td);
1023 vfs_mount_destroy(mp);
1024 vput(vp);
1025 }
1026 return (error);
1027}
1028
1029/*
1030 * ---------------------------------------------------------------------
1031 * Unmount a filesystem.
1032 *
1033 * Note: unmount takes a path to the vnode mounted on as argument,
1034 * not special file (as before).
1035 */
1036#ifndef _SYS_SYSPROTO_H_
1037struct unmount_args {
1038 char *path;
1039 int flags;
1040};
1041#endif
1042/* ARGSUSED */
1043int
1044unmount(td, uap)
1045 struct thread *td;
1046 register struct unmount_args /* {
1047 char *path;
1048 int flags;
1049 } */ *uap;
1050{
1051 struct mount *mp;
1052 char *pathbuf;
1053 int error, id0, id1;
1054
1055 if (jailed(td->td_ucred))
1056 return (EPERM);
1057 if (usermount == 0) {
1058 if ((error = suser(td)) != 0)
1059 return (error);
1060 }
1061
1062 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1063 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1064 if (error) {
1065 free(pathbuf, M_TEMP);
1066 return (error);
1067 }
1068 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1069 mtx_lock(&Giant);
1070 if (uap->flags & MNT_BYFSID) {
1071 /* Decode the filesystem ID. */
1072 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1073 mtx_unlock(&Giant);
1074 free(pathbuf, M_TEMP);
1075 return (EINVAL);
1076 }
1077
1078 mtx_lock(&mountlist_mtx);
1079 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1080 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1081 mp->mnt_stat.f_fsid.val[1] == id1)
1082 break;
1083 }
1084 mtx_unlock(&mountlist_mtx);
1085 } else {
1086 mtx_lock(&mountlist_mtx);
1087 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1088 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1089 break;
1090 }
1091 mtx_unlock(&mountlist_mtx);
1092 }
1093 free(pathbuf, M_TEMP);
1094 if (mp == NULL) {
1095 /*
1096 * Previously we returned ENOENT for a nonexistent path and
1097 * EINVAL for a non-mountpoint. We cannot tell these apart
1098 * now, so in the !MNT_BYFSID case return the more likely
1099 * EINVAL for compatibility.
1100 */
1101 mtx_unlock(&Giant);
1102 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1103 }
1104
1105 /*
1106 * Don't allow unmounting the root filesystem.
1107 */
1108 if (mp->mnt_flag & MNT_ROOTFS) {
1109 mtx_unlock(&Giant);
1110 return (EINVAL);
1111 }
1112 error = dounmount(mp, uap->flags, td);
1113 mtx_unlock(&Giant);
1114 return (error);
1115}
1116
1117/*
1118 * Do the actual filesystem unmount.
1119 */
1120int
1121dounmount(mp, flags, td)
1122 struct mount *mp;
1123 int flags;
1124 struct thread *td;
1125{
1126 struct vnode *coveredvp, *fsrootvp;
1127 int error;
1128 int async_flag;
1129
1130 mtx_assert(&Giant, MA_OWNED);
1131
1132 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1133 VI_LOCK(coveredvp);
1134 vholdl(coveredvp);
1135 error = vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK, td);
1136 vdrop(coveredvp);
1137 /*
1138 * Check for mp being unmounted while waiting for the
1139 * covered vnode lock.
1140 */
1141 if (error)
1142 return (error);
1143 if (coveredvp->v_mountedhere != mp) {
1144 VOP_UNLOCK(coveredvp, 0, td);
1145 return (EBUSY);
1146 }
1147 }
1148 /*
1149 * Only privileged root, or (if MNT_USER is set) the user that did the
1150 * original mount is permitted to unmount this filesystem.
1151 */
1152 error = vfs_suser(mp, td);
1153 if (error) {
|
1155 return (error);
1156 }
1157
1158 MNT_ILOCK(mp);
1159 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1160 MNT_IUNLOCK(mp);
1161 if (coveredvp)
1162 VOP_UNLOCK(coveredvp, 0, td);
1163 return (EBUSY);
1164 }
1165 mp->mnt_kern_flag |= MNTK_UNMOUNT;
1166 /* Allow filesystems to detect that a forced unmount is in progress. */
1167 if (flags & MNT_FORCE)
1168 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1169 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1170 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td);
1171 if (error) {
1172 MNT_ILOCK(mp);
1173 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1174 if (mp->mnt_kern_flag & MNTK_MWAIT)
1175 wakeup(mp);
1176 MNT_IUNLOCK(mp);
1177 if (coveredvp)
1178 VOP_UNLOCK(coveredvp, 0, td);
1179 return (error);
1180 }
1181 vn_start_write(NULL, &mp, V_WAIT);
1182
1183 if (mp->mnt_flag & MNT_EXPUBLIC)
1184 vfs_setpublicfs(NULL, NULL, NULL);
1185
1186 vfs_msync(mp, MNT_WAIT);
1187 async_flag = mp->mnt_flag & MNT_ASYNC;
1188 mp->mnt_flag &= ~MNT_ASYNC;
1189 cache_purgevfs(mp); /* remove cache entries for this file sys */
1190 if (mp->mnt_syncer != NULL)
1191 vrele(mp->mnt_syncer);
1192 /*
1193 * For forced unmounts, move process cdir/rdir refs on the fs root
1194 * vnode to the covered vnode. For non-forced unmounts we want
1195 * such references to cause an EBUSY error.
1196 */
1197 if ((flags & MNT_FORCE) &&
1198 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1199 if (mp->mnt_vnodecovered != NULL)
1200 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1201 if (fsrootvp == rootvnode) {
1202 vrele(rootvnode);
1203 rootvnode = NULL;
1204 }
1205 vput(fsrootvp);
1206 }
1207 if (((mp->mnt_flag & MNT_RDONLY) ||
1208 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1209 (flags & MNT_FORCE)) {
1210 error = VFS_UNMOUNT(mp, flags, td);
1211 }
1212 vn_finished_write(mp);
1213 if (error) {
1214 /* Undo cdir/rdir and rootvnode changes made above. */
1215 if ((flags & MNT_FORCE) &&
1216 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1217 if (mp->mnt_vnodecovered != NULL)
1218 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1219 if (rootvnode == NULL) {
1220 rootvnode = fsrootvp;
1221 vref(rootvnode);
1222 }
1223 vput(fsrootvp);
1224 }
1225 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1226 (void) vfs_allocate_syncvnode(mp);
1227 MNT_ILOCK(mp);
1228 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1229 mp->mnt_flag |= async_flag;
1230 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1231 if (mp->mnt_kern_flag & MNTK_MWAIT)
1232 wakeup(mp);
1233 MNT_IUNLOCK(mp);
1234 if (coveredvp)
1235 VOP_UNLOCK(coveredvp, 0, td);
1236 return (error);
1237 }
1238 mtx_lock(&mountlist_mtx);
1239 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1240 mtx_unlock(&mountlist_mtx);
1241 if (coveredvp != NULL) {
1242 coveredvp->v_mountedhere = NULL;
1243 vput(coveredvp);
1244 }
1245 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1246 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1247 vfs_mount_destroy(mp);
1248 return (0);
1249}
1250
1251/*
1252 * ---------------------------------------------------------------------
1253 * Mounting of root filesystem
1254 *
1255 */
1256
1257struct root_hold_token {
1258 const char *who;
1259 LIST_ENTRY(root_hold_token) list;
1260};
1261
1262static LIST_HEAD(, root_hold_token) root_holds =
1263 LIST_HEAD_INITIALIZER(&root_holds);
1264
1265struct root_hold_token *
1266root_mount_hold(const char *identifier)
1267{
1268 struct root_hold_token *h;
1269
1270 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1271 h->who = identifier;
1272 mtx_lock(&mountlist_mtx);
1273 LIST_INSERT_HEAD(&root_holds, h, list);
1274 mtx_unlock(&mountlist_mtx);
1275 return (h);
1276}
1277
1278void
1279root_mount_rel(struct root_hold_token *h)
1280{
1281
1282 mtx_lock(&mountlist_mtx);
1283 LIST_REMOVE(h, list);
1284 wakeup(&root_holds);
1285 mtx_unlock(&mountlist_mtx);
1286 free(h, M_DEVBUF);
1287}
1288
1289static void
1290root_mount_wait(void)
1291{
1292 struct root_hold_token *h;
1293
1294 for (;;) {
1295 DROP_GIANT();
1296 g_waitidle();
1297 PICKUP_GIANT();
1298 mtx_lock(&mountlist_mtx);
1299 if (LIST_EMPTY(&root_holds)) {
1300 mtx_unlock(&mountlist_mtx);
1301 break;
1302 }
1303 printf("Root mount waiting for:");
1304 LIST_FOREACH(h, &root_holds, list)
1305 printf(" %s", h->who);
1306 printf("\n");
1307 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1308 hz);
1309 }
1310}
1311
1312static void
1313set_rootvnode(struct thread *td)
1314{
1315 struct proc *p;
1316
1317 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1318 panic("Cannot find root vnode");
1319
1320 p = td->td_proc;
1321 FILEDESC_LOCK(p->p_fd);
1322
1323 if (p->p_fd->fd_cdir != NULL)
1324 vrele(p->p_fd->fd_cdir);
1325 p->p_fd->fd_cdir = rootvnode;
1326 VREF(rootvnode);
1327
1328 if (p->p_fd->fd_rdir != NULL)
1329 vrele(p->p_fd->fd_rdir);
1330 p->p_fd->fd_rdir = rootvnode;
1331 VREF(rootvnode);
1332
1333 FILEDESC_UNLOCK(p->p_fd);
1334
1335 VOP_UNLOCK(rootvnode, 0, td);
1336}
1337
1338/*
1339 * Mount /devfs as our root filesystem, but do not put it on the mountlist
1340 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup.
1341 */
1342
1343static void
1344devfs_first(void)
1345{
1346 struct thread *td = curthread;
1347 struct vfsoptlist *opts;
1348 struct vfsconf *vfsp;
1349 struct mount *mp = NULL;
1350 int error;
1351
1352 vfsp = vfs_byname("devfs");
1353 KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1354 if (vfsp == NULL)
1355 return;
1356
1357 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1358
1359 error = VFS_MOUNT(mp, td);
1360 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1361 if (error)
1362 return;
1363
1364 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1365 TAILQ_INIT(opts);
1366 mp->mnt_opt = opts;
1367
1368 mtx_lock(&mountlist_mtx);
1369 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1370 mtx_unlock(&mountlist_mtx);
1371
1372 set_rootvnode(td);
1373
1374 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1375 if (error)
1376 printf("kern_symlink /dev -> / returns %d\n", error);
1377}
1378
1379/*
1380 * Surgically move our devfs to be mounted on /dev.
1381 */
1382
1383static void
1384devfs_fixup(struct thread *td)
1385{
1386 struct nameidata nd;
1387 int error;
1388 struct vnode *vp, *dvp;
1389 struct mount *mp;
1390
1391 /* Remove our devfs mount from the mountlist and purge the cache */
1392 mtx_lock(&mountlist_mtx);
1393 mp = TAILQ_FIRST(&mountlist);
1394 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1395 mtx_unlock(&mountlist_mtx);
1396 cache_purgevfs(mp);
1397
1398 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1399 VI_LOCK(dvp);
1400 dvp->v_iflag &= ~VI_MOUNT;
1401 dvp->v_mountedhere = NULL;
1402 VI_UNLOCK(dvp);
1403
1404 /* Set up the real rootvnode, and purge the cache */
1405 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1406 set_rootvnode(td);
1407 cache_purgevfs(rootvnode->v_mount);
1408
1409 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1410 error = namei(&nd);
1411 if (error) {
1412 printf("Lookup of /dev for devfs, error: %d\n", error);
1413 return;
1414 }
1415 NDFREE(&nd, NDF_ONLY_PNBUF);
1416 vp = nd.ni_vp;
1417 if (vp->v_type != VDIR) {
1418 vput(vp);
1419 }
1420 error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1421 if (error) {
1422 vput(vp);
1423 }
1424 cache_purge(vp);
1425 mp->mnt_vnodecovered = vp;
1426 vp->v_mountedhere = mp;
1427 mtx_lock(&mountlist_mtx);
1428 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1429 mtx_unlock(&mountlist_mtx);
1430 VOP_UNLOCK(vp, 0, td);
1431 vput(dvp);
1432 vfs_unbusy(mp, td);
1433
1434 /* Unlink the no longer needed /dev/dev -> / symlink */
1435 kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1436}
1437
1438/*
1439 * Report errors during filesystem mounting.
1440 */
1441void
1442vfs_mount_error(struct mount *mp, const char *fmt, ...)
1443{
1444 struct vfsoptlist *moptlist = mp->mnt_optnew;
1445 va_list ap;
1446 int error, len;
1447 char *errmsg;
1448
1449 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1450 if (error || errmsg == NULL || len <= 0)
1451 return;
1452
1453 va_start(ap, fmt);
1454 vsnprintf(errmsg, (size_t)len, fmt, ap);
1455 va_end(ap);
1456}
1457
1458/*
1459 * Find and mount the root filesystem
1460 */
1461void
1462vfs_mountroot(void)
1463{
1464 char *cp;
1465 int error, i, asked = 0;
1466
1467 root_mount_wait();
1468
1469 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1470 NULL, NULL, mount_init, mount_fini,
1471 UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1472 devfs_first();
1473
1474 /*
1475 * We are booted with instructions to prompt for the root filesystem.
1476 */
1477 if (boothowto & RB_ASKNAME) {
1478 if (!vfs_mountroot_ask())
1479 return;
1480 asked = 1;
1481 }
1482
1483 /*
1484 * The root filesystem information is compiled in, and we are
1485 * booted with instructions to use it.
1486 */
1487 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1488 if (!vfs_mountroot_try(ctrootdevname))
1489 return;
1490 ctrootdevname = NULL;
1491 }
1492
1493 /*
1494 * We've been given the generic "use CDROM as root" flag. This is
1495 * necessary because one media may be used in many different
1496 * devices, so we need to search for them.
1497 */
1498 if (boothowto & RB_CDROM) {
1499 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1500 if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1501 return;
1502 }
1503 }
1504
1505 /*
1506 * Try to use the value read by the loader from /etc/fstab, or
1507 * supplied via some other means. This is the preferred
1508 * mechanism.
1509 */
1510 cp = getenv("vfs.root.mountfrom");
1511 if (cp != NULL) {
1512 error = vfs_mountroot_try(cp);
1513 freeenv(cp);
1514 if (!error)
1515 return;
1516 }
1517
1518 /*
1519 * Try values that may have been computed by code during boot
1520 */
1521 if (!vfs_mountroot_try(rootdevnames[0]))
1522 return;
1523 if (!vfs_mountroot_try(rootdevnames[1]))
1524 return;
1525
1526 /*
1527 * If we (still) have a compiled-in default, try it.
1528 */
1529 if (ctrootdevname != NULL)
1530 if (!vfs_mountroot_try(ctrootdevname))
1531 return;
1532 /*
1533 * Everything so far has failed, prompt on the console if we haven't
1534 * already tried that.
1535 */
1536 if (!asked)
1537 if (!vfs_mountroot_ask())
1538 return;
1539
1540 panic("Root mount failed, startup aborted.");
1541}
1542
1543/*
1544 * Mount (mountfrom) as the root filesystem.
1545 */
1546static int
1547vfs_mountroot_try(const char *mountfrom)
1548{
1549 struct mount *mp;
1550 char *vfsname, *path;
1551 time_t timebase;
1552 int error;
1553 char patt[32];
1554
1555 vfsname = NULL;
1556 path = NULL;
1557 mp = NULL;
1558 error = EINVAL;
1559
1560 if (mountfrom == NULL)
1561 return (error); /* don't complain */
1562 printf("Trying to mount root from %s\n", mountfrom);
1563
1564 /* parse vfs name and path */
1565 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1566 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1567 vfsname[0] = path[0] = 0;
1568 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1569 if (sscanf(mountfrom, patt, vfsname, path) < 1)
1570 goto out;
1571
1572 if (path[0] == '\0')
1573 strcpy(path, ROOTNAME);
1574
1575 error = kernel_vmount(
1576 MNT_RDONLY | MNT_ROOTFS,
1577 "fstype", vfsname,
1578 "fspath", "/",
1579 "from", path,
1580 NULL);
1581 if (error == 0) {
1582 /*
1583 * We mount devfs prior to mounting the / FS, so the first
1584 * entry will typically be devfs.
1585 */
1586 mp = TAILQ_FIRST(&mountlist);
1587 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1588
1589 /*
1590 * Iterate over all currently mounted file systems and use
1591 * the time stamp found to check and/or initialize the RTC.
1592 * Typically devfs has no time stamp and the only other FS
1593 * is the actual / FS.
1594 * Call inittodr() only once and pass it the largest of the
1595 * timestamps we encounter.
1596 */
1597 timebase = 0;
1598 do {
1599 if (mp->mnt_time > timebase)
1600 timebase = mp->mnt_time;
1601 mp = TAILQ_NEXT(mp, mnt_list);
1602 } while (mp != NULL);
1603 inittodr(timebase);
1604
1605 devfs_fixup(curthread);
1606 }
1607out:
1608 free(path, M_MOUNT);
1609 free(vfsname, M_MOUNT);
1610 return (error);
1611}
1612
1613/*
1614 * ---------------------------------------------------------------------
1615 * Interactive root filesystem selection code.
1616 */
1617
1618static int
1619vfs_mountroot_ask(void)
1620{
1621 char name[128];
1622
1623 for(;;) {
1624 printf("\nManual root filesystem specification:\n");
1625 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n");
1626#if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1627 printf(" eg. ufs:da0s1a\n");
1628#else
1629 printf(" eg. ufs:/dev/da0a\n");
1630#endif
1631 printf(" ? List valid disk boot devices\n");
1632 printf(" <empty line> Abort manual input\n");
1633 printf("\nmountroot> ");
1634 gets(name, sizeof(name), 1);
1635 if (name[0] == '\0')
1636 return (1);
1637 if (name[0] == '?') {
1638 printf("\nList of GEOM managed disk devices:\n ");
1639 g_dev_print();
1640 continue;
1641 }
1642 if (!vfs_mountroot_try(name))
1643 return (0);
1644 }
1645}
1646
1647/*
1648 * ---------------------------------------------------------------------
1649 * Functions for querying mount options/arguments from filesystems.
1650 */
1651
1652/*
1653 * Check that no unknown options are given
1654 */
1655int
1656vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1657{
1658 struct vfsopt *opt;
1659 const char **t, *p;
1660
1661
1662 TAILQ_FOREACH(opt, opts, link) {
1663 p = opt->name;
1664 if (p[0] == 'n' && p[1] == 'o')
1665 p += 2;
1666 for(t = global_opts; *t != NULL; t++)
1667 if (!strcmp(*t, p))
1668 break;
1669 if (*t != NULL)
1670 continue;
1671 for(t = legal; *t != NULL; t++)
1672 if (!strcmp(*t, p))
1673 break;
1674 if (*t != NULL)
1675 continue;
1676 printf("mount option <%s> is unknown\n", p);
1677 return (EINVAL);
1678 }
1679 return (0);
1680}
1681
1682/*
1683 * Get a mount option by its name.
1684 *
1685 * Return 0 if the option was found, ENOENT otherwise.
1686 * If len is non-NULL it will be filled with the length
1687 * of the option. If buf is non-NULL, it will be filled
1688 * with the address of the option.
1689 */
1690int
1691vfs_getopt(opts, name, buf, len)
1692 struct vfsoptlist *opts;
1693 const char *name;
1694 void **buf;
1695 int *len;
1696{
1697 struct vfsopt *opt;
1698
1699 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1700
1701 TAILQ_FOREACH(opt, opts, link) {
1702 if (strcmp(name, opt->name) == 0) {
1703 if (len != NULL)
1704 *len = opt->len;
1705 if (buf != NULL)
1706 *buf = opt->value;
1707 return (0);
1708 }
1709 }
1710 return (ENOENT);
1711}
1712
1713static int
1714vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1715{
1716 struct vfsopt *opt;
1717 int i;
1718
1719 if (opts == NULL)
1720 return (-1);
1721
1722 i = 0;
1723 TAILQ_FOREACH(opt, opts, link) {
1724 if (strcmp(name, opt->name) == 0)
1725 return (i);
1726 ++i;
1727 }
1728 return (-1);
1729}
1730
1731char *
1732vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1733{
1734 struct vfsopt *opt;
1735
1736 *error = 0;
1737 TAILQ_FOREACH(opt, opts, link) {
1738 if (strcmp(name, opt->name) != 0)
1739 continue;
1740 if (((char *)opt->value)[opt->len - 1] != '\0') {
1741 *error = EINVAL;
1742 return (NULL);
1743 }
1744 return (opt->value);
1745 }
1746 return (NULL);
1747}
1748
1749int
1750vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1751{
1752 struct vfsopt *opt;
1753
1754 TAILQ_FOREACH(opt, opts, link) {
1755 if (strcmp(name, opt->name) == 0) {
1756 if (w != NULL)
1757 *w |= val;
1758 return (1);
1759 }
1760 }
1761 if (w != NULL)
1762 *w &= ~val;
1763 return (0);
1764}
1765
1766int
1767vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1768{
1769 va_list ap;
1770 struct vfsopt *opt;
1771 int ret;
1772
1773 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1774
1775 TAILQ_FOREACH(opt, opts, link) {
1776 if (strcmp(name, opt->name) != 0)
1777 continue;
1778 if (((char *)opt->value)[opt->len - 1] != '\0')
1779 return (0);
1780 va_start(ap, fmt);
1781 ret = vsscanf(opt->value, fmt, ap);
1782 va_end(ap);
1783 return (ret);
1784 }
1785 return (0);
1786}
1787
1788/*
1789 * Find and copy a mount option.
1790 *
1791 * The size of the buffer has to be specified
1792 * in len, if it is not the same length as the
1793 * mount option, EINVAL is returned.
1794 * Returns ENOENT if the option is not found.
1795 */
1796int
1797vfs_copyopt(opts, name, dest, len)
1798 struct vfsoptlist *opts;
1799 const char *name;
1800 void *dest;
1801 int len;
1802{
1803 struct vfsopt *opt;
1804
1805 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1806
1807 TAILQ_FOREACH(opt, opts, link) {
1808 if (strcmp(name, opt->name) == 0) {
1809 if (len != opt->len)
1810 return (EINVAL);
1811 bcopy(opt->value, dest, opt->len);
1812 return (0);
1813 }
1814 }
1815 return (ENOENT);
1816}
1817
1818/*
1819 * This is a helper function for filesystems to traverse their
1820 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h
1821 */
1822
1823struct vnode *
1824__mnt_vnode_next(struct vnode **mvp, struct mount *mp)
1825{
1826 struct vnode *vp;
1827
1828 mtx_assert(MNT_MTX(mp), MA_OWNED);
1829
1830 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1831 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
1832 while (vp != NULL && vp->v_type == VMARKER)
1833 vp = TAILQ_NEXT(vp, v_nmntvnodes);
1834
1835 /* Check if we are done */
1836 if (vp == NULL) {
1837 __mnt_vnode_markerfree(mvp, mp);
1838 return (NULL);
1839 }
1840 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1841 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1842 return (vp);
1843}
1844
1845struct vnode *
1846__mnt_vnode_first(struct vnode **mvp, struct mount *mp)
1847{
1848 struct vnode *vp;
1849
1850 mtx_assert(MNT_MTX(mp), MA_OWNED);
1851
1852 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1853 while (vp != NULL && vp->v_type == VMARKER)
1854 vp = TAILQ_NEXT(vp, v_nmntvnodes);
1855
1856 /* Check if we are done */
1857 if (vp == NULL) {
1858 *mvp = NULL;
1859 return (NULL);
1860 }
1861 mp->mnt_holdcnt++;
1862 MNT_IUNLOCK(mp);
1863 *mvp = (struct vnode *) malloc(sizeof(struct vnode),
1864 M_VNODE_MARKER,
1865 M_WAITOK | M_ZERO);
1866 MNT_ILOCK(mp);
1867 (*mvp)->v_type = VMARKER;
1868
1869 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1870 while (vp != NULL && vp->v_type == VMARKER)
1871 vp = TAILQ_NEXT(vp, v_nmntvnodes);
1872
1873 /* Check if we are done */
1874 if (vp == NULL) {
1875 MNT_IUNLOCK(mp);
1876 free(*mvp, M_VNODE_MARKER);
1877 MNT_ILOCK(mp);
1878 *mvp = NULL;
1879 mp->mnt_holdcnt--;
1880 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1881 wakeup(&mp->mnt_holdcnt);
1882 return (NULL);
1883 }
1884 mp->mnt_markercnt++;
1885 (*mvp)->v_mount = mp;
1886 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1887 return (vp);
1888}
1889
1890
1891void
1892__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
1893{
1894
1895 if (*mvp == NULL)
1896 return;
1897
1898 mtx_assert(MNT_MTX(mp), MA_OWNED);
1899
1900 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1901 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1902 MNT_IUNLOCK(mp);
1903 free(*mvp, M_VNODE_MARKER);
1904 MNT_ILOCK(mp);
1905 *mvp = NULL;
1906
1907 mp->mnt_markercnt--;
1908 mp->mnt_holdcnt--;
1909 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1910 wakeup(&mp->mnt_holdcnt);
1911}
1912
1913
1914int
1915__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
1916{
1917 int error;
1918
1919 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
1920 if (sbp != &mp->mnt_stat)
1921 *sbp = mp->mnt_stat;
1922 return (error);
1923}
1924
1925void
1926vfs_mountedfrom(struct mount *mp, const char *from)
1927{
1928
1929 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1930 strlcpy(mp->mnt_stat.f_mntfromname, from,
1931 sizeof mp->mnt_stat.f_mntfromname);
1932}
1933
1934/*
1935 * ---------------------------------------------------------------------
1936 * This is the api for building mount args and mounting filesystems from
1937 * inside the kernel.
1938 *
1939 * The API works by accumulation of individual args. First error is
1940 * latched.
1941 *
1942 * XXX: should be documented in new manpage kernel_mount(9)
1943 */
1944
1945/* A memory allocation which must be freed when we are done */
1946struct mntaarg {
1947 SLIST_ENTRY(mntaarg) next;
1948};
1949
1950/* The header for the mount arguments */
1951struct mntarg {
1952 struct iovec *v;
1953 int len;
1954 int error;
1955 SLIST_HEAD(, mntaarg) list;
1956};
1957
1958/*
1959 * Add a boolean argument.
1960 *
1961 * flag is the boolean value.
1962 * name must start with "no".
1963 */
1964struct mntarg *
1965mount_argb(struct mntarg *ma, int flag, const char *name)
1966{
1967
1968 KASSERT(name[0] == 'n' && name[1] == 'o',
1969 ("mount_argb(...,%s): name must start with 'no'", name));
1970
1971 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1972}
1973
1974/*
1975 * Add an argument printf style
1976 */
1977struct mntarg *
1978mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1979{
1980 va_list ap;
1981 struct mntaarg *maa;
1982 struct sbuf *sb;
1983 int len;
1984
1985 if (ma == NULL) {
1986 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1987 SLIST_INIT(&ma->list);
1988 }
1989 if (ma->error)
1990 return (ma);
1991
1992 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1993 M_MOUNT, M_WAITOK);
1994 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1995 ma->v[ma->len].iov_len = strlen(name) + 1;
1996 ma->len++;
1997
1998 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
1999 va_start(ap, fmt);
2000 sbuf_vprintf(sb, fmt, ap);
2001 va_end(ap);
2002 sbuf_finish(sb);
2003 len = sbuf_len(sb) + 1;
2004 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2005 SLIST_INSERT_HEAD(&ma->list, maa, next);
2006 bcopy(sbuf_data(sb), maa + 1, len);
2007 sbuf_delete(sb);
2008
2009 ma->v[ma->len].iov_base = maa + 1;
2010 ma->v[ma->len].iov_len = len;
2011 ma->len++;
2012
2013 return (ma);
2014}
2015
2016/*
2017 * Add an argument which is a userland string.
2018 */
2019struct mntarg *
2020mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2021{
2022 struct mntaarg *maa;
2023 char *tbuf;
2024
2025 if (val == NULL)
2026 return (ma);
2027 if (ma == NULL) {
2028 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2029 SLIST_INIT(&ma->list);
2030 }
2031 if (ma->error)
2032 return (ma);
2033 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2034 SLIST_INSERT_HEAD(&ma->list, maa, next);
2035 tbuf = (void *)(maa + 1);
2036 ma->error = copyinstr(val, tbuf, len, NULL);
2037 return (mount_arg(ma, name, tbuf, -1));
2038}
2039
2040/*
2041 * Plain argument.
2042 *
2043 * If length is -1, use printf.
2044 */
2045struct mntarg *
2046mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2047{
2048
2049 if (ma == NULL) {
2050 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2051 SLIST_INIT(&ma->list);
2052 }
2053 if (ma->error)
2054 return (ma);
2055
2056 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2057 M_MOUNT, M_WAITOK);
2058 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2059 ma->v[ma->len].iov_len = strlen(name) + 1;
2060 ma->len++;
2061
2062 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2063 if (len < 0)
2064 ma->v[ma->len].iov_len = strlen(val) + 1;
2065 else
2066 ma->v[ma->len].iov_len = len;
2067 ma->len++;
2068 return (ma);
2069}
2070
2071/*
2072 * Free a mntarg structure
2073 */
2074static void
2075free_mntarg(struct mntarg *ma)
2076{
2077 struct mntaarg *maa;
2078
2079 while (!SLIST_EMPTY(&ma->list)) {
2080 maa = SLIST_FIRST(&ma->list);
2081 SLIST_REMOVE_HEAD(&ma->list, next);
2082 free(maa, M_MOUNT);
2083 }
2084 free(ma->v, M_MOUNT);
2085 free(ma, M_MOUNT);
2086}
2087
2088/*
2089 * Mount a filesystem
2090 */
2091int
2092kernel_mount(struct mntarg *ma, int flags)
2093{
2094 struct uio auio;
2095 int error;
2096
2097 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2098 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2099 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2100
2101 auio.uio_iov = ma->v;
2102 auio.uio_iovcnt = ma->len;
2103 auio.uio_segflg = UIO_SYSSPACE;
2104
2105 error = ma->error;
2106 if (!error)
2107 error = vfs_donmount(curthread, flags, &auio);
2108 free_mntarg(ma);
2109 return (error);
2110}
2111
2112/*
2113 * A printflike function to mount a filesystem.
2114 */
2115int
2116kernel_vmount(int flags, ...)
2117{
2118 struct mntarg *ma = NULL;
2119 va_list ap;
2120 const char *cp;
2121 const void *vp;
2122 int error;
2123
2124 va_start(ap, flags);
2125 for (;;) {
2126 cp = va_arg(ap, const char *);
2127 if (cp == NULL)
2128 break;
2129 vp = va_arg(ap, const void *);
2130 ma = mount_arg(ma, cp, vp, -1);
2131 }
2132 va_end(ap);
2133
2134 error = kernel_mount(ma, flags);
2135 return (error);
2136}
| 1156 return (error);
1157 }
1158
1159 MNT_ILOCK(mp);
1160 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1161 MNT_IUNLOCK(mp);
1162 if (coveredvp)
1163 VOP_UNLOCK(coveredvp, 0, td);
1164 return (EBUSY);
1165 }
1166 mp->mnt_kern_flag |= MNTK_UNMOUNT;
1167 /* Allow filesystems to detect that a forced unmount is in progress. */
1168 if (flags & MNT_FORCE)
1169 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1170 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1171 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td);
1172 if (error) {
1173 MNT_ILOCK(mp);
1174 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1175 if (mp->mnt_kern_flag & MNTK_MWAIT)
1176 wakeup(mp);
1177 MNT_IUNLOCK(mp);
1178 if (coveredvp)
1179 VOP_UNLOCK(coveredvp, 0, td);
1180 return (error);
1181 }
1182 vn_start_write(NULL, &mp, V_WAIT);
1183
1184 if (mp->mnt_flag & MNT_EXPUBLIC)
1185 vfs_setpublicfs(NULL, NULL, NULL);
1186
1187 vfs_msync(mp, MNT_WAIT);
1188 async_flag = mp->mnt_flag & MNT_ASYNC;
1189 mp->mnt_flag &= ~MNT_ASYNC;
1190 cache_purgevfs(mp); /* remove cache entries for this file sys */
1191 if (mp->mnt_syncer != NULL)
1192 vrele(mp->mnt_syncer);
1193 /*
1194 * For forced unmounts, move process cdir/rdir refs on the fs root
1195 * vnode to the covered vnode. For non-forced unmounts we want
1196 * such references to cause an EBUSY error.
1197 */
1198 if ((flags & MNT_FORCE) &&
1199 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1200 if (mp->mnt_vnodecovered != NULL)
1201 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1202 if (fsrootvp == rootvnode) {
1203 vrele(rootvnode);
1204 rootvnode = NULL;
1205 }
1206 vput(fsrootvp);
1207 }
1208 if (((mp->mnt_flag & MNT_RDONLY) ||
1209 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1210 (flags & MNT_FORCE)) {
1211 error = VFS_UNMOUNT(mp, flags, td);
1212 }
1213 vn_finished_write(mp);
1214 if (error) {
1215 /* Undo cdir/rdir and rootvnode changes made above. */
1216 if ((flags & MNT_FORCE) &&
1217 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1218 if (mp->mnt_vnodecovered != NULL)
1219 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1220 if (rootvnode == NULL) {
1221 rootvnode = fsrootvp;
1222 vref(rootvnode);
1223 }
1224 vput(fsrootvp);
1225 }
1226 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1227 (void) vfs_allocate_syncvnode(mp);
1228 MNT_ILOCK(mp);
1229 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1230 mp->mnt_flag |= async_flag;
1231 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1232 if (mp->mnt_kern_flag & MNTK_MWAIT)
1233 wakeup(mp);
1234 MNT_IUNLOCK(mp);
1235 if (coveredvp)
1236 VOP_UNLOCK(coveredvp, 0, td);
1237 return (error);
1238 }
1239 mtx_lock(&mountlist_mtx);
1240 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1241 mtx_unlock(&mountlist_mtx);
1242 if (coveredvp != NULL) {
1243 coveredvp->v_mountedhere = NULL;
1244 vput(coveredvp);
1245 }
1246 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1247 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1248 vfs_mount_destroy(mp);
1249 return (0);
1250}
1251
1252/*
1253 * ---------------------------------------------------------------------
1254 * Mounting of root filesystem
1255 *
1256 */
1257
1258struct root_hold_token {
1259 const char *who;
1260 LIST_ENTRY(root_hold_token) list;
1261};
1262
1263static LIST_HEAD(, root_hold_token) root_holds =
1264 LIST_HEAD_INITIALIZER(&root_holds);
1265
1266struct root_hold_token *
1267root_mount_hold(const char *identifier)
1268{
1269 struct root_hold_token *h;
1270
1271 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1272 h->who = identifier;
1273 mtx_lock(&mountlist_mtx);
1274 LIST_INSERT_HEAD(&root_holds, h, list);
1275 mtx_unlock(&mountlist_mtx);
1276 return (h);
1277}
1278
1279void
1280root_mount_rel(struct root_hold_token *h)
1281{
1282
1283 mtx_lock(&mountlist_mtx);
1284 LIST_REMOVE(h, list);
1285 wakeup(&root_holds);
1286 mtx_unlock(&mountlist_mtx);
1287 free(h, M_DEVBUF);
1288}
1289
1290static void
1291root_mount_wait(void)
1292{
1293 struct root_hold_token *h;
1294
1295 for (;;) {
1296 DROP_GIANT();
1297 g_waitidle();
1298 PICKUP_GIANT();
1299 mtx_lock(&mountlist_mtx);
1300 if (LIST_EMPTY(&root_holds)) {
1301 mtx_unlock(&mountlist_mtx);
1302 break;
1303 }
1304 printf("Root mount waiting for:");
1305 LIST_FOREACH(h, &root_holds, list)
1306 printf(" %s", h->who);
1307 printf("\n");
1308 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1309 hz);
1310 }
1311}
1312
1313static void
1314set_rootvnode(struct thread *td)
1315{
1316 struct proc *p;
1317
1318 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1319 panic("Cannot find root vnode");
1320
1321 p = td->td_proc;
1322 FILEDESC_LOCK(p->p_fd);
1323
1324 if (p->p_fd->fd_cdir != NULL)
1325 vrele(p->p_fd->fd_cdir);
1326 p->p_fd->fd_cdir = rootvnode;
1327 VREF(rootvnode);
1328
1329 if (p->p_fd->fd_rdir != NULL)
1330 vrele(p->p_fd->fd_rdir);
1331 p->p_fd->fd_rdir = rootvnode;
1332 VREF(rootvnode);
1333
1334 FILEDESC_UNLOCK(p->p_fd);
1335
1336 VOP_UNLOCK(rootvnode, 0, td);
1337}
1338
1339/*
1340 * Mount /devfs as our root filesystem, but do not put it on the mountlist
1341 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup.
1342 */
1343
1344static void
1345devfs_first(void)
1346{
1347 struct thread *td = curthread;
1348 struct vfsoptlist *opts;
1349 struct vfsconf *vfsp;
1350 struct mount *mp = NULL;
1351 int error;
1352
1353 vfsp = vfs_byname("devfs");
1354 KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1355 if (vfsp == NULL)
1356 return;
1357
1358 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1359
1360 error = VFS_MOUNT(mp, td);
1361 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1362 if (error)
1363 return;
1364
1365 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1366 TAILQ_INIT(opts);
1367 mp->mnt_opt = opts;
1368
1369 mtx_lock(&mountlist_mtx);
1370 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1371 mtx_unlock(&mountlist_mtx);
1372
1373 set_rootvnode(td);
1374
1375 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1376 if (error)
1377 printf("kern_symlink /dev -> / returns %d\n", error);
1378}
1379
1380/*
1381 * Surgically move our devfs to be mounted on /dev.
1382 */
1383
1384static void
1385devfs_fixup(struct thread *td)
1386{
1387 struct nameidata nd;
1388 int error;
1389 struct vnode *vp, *dvp;
1390 struct mount *mp;
1391
1392 /* Remove our devfs mount from the mountlist and purge the cache */
1393 mtx_lock(&mountlist_mtx);
1394 mp = TAILQ_FIRST(&mountlist);
1395 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1396 mtx_unlock(&mountlist_mtx);
1397 cache_purgevfs(mp);
1398
1399 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1400 VI_LOCK(dvp);
1401 dvp->v_iflag &= ~VI_MOUNT;
1402 dvp->v_mountedhere = NULL;
1403 VI_UNLOCK(dvp);
1404
1405 /* Set up the real rootvnode, and purge the cache */
1406 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1407 set_rootvnode(td);
1408 cache_purgevfs(rootvnode->v_mount);
1409
1410 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1411 error = namei(&nd);
1412 if (error) {
1413 printf("Lookup of /dev for devfs, error: %d\n", error);
1414 return;
1415 }
1416 NDFREE(&nd, NDF_ONLY_PNBUF);
1417 vp = nd.ni_vp;
1418 if (vp->v_type != VDIR) {
1419 vput(vp);
1420 }
1421 error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1422 if (error) {
1423 vput(vp);
1424 }
1425 cache_purge(vp);
1426 mp->mnt_vnodecovered = vp;
1427 vp->v_mountedhere = mp;
1428 mtx_lock(&mountlist_mtx);
1429 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1430 mtx_unlock(&mountlist_mtx);
1431 VOP_UNLOCK(vp, 0, td);
1432 vput(dvp);
1433 vfs_unbusy(mp, td);
1434
1435 /* Unlink the no longer needed /dev/dev -> / symlink */
1436 kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1437}
1438
1439/*
1440 * Report errors during filesystem mounting.
1441 */
1442void
1443vfs_mount_error(struct mount *mp, const char *fmt, ...)
1444{
1445 struct vfsoptlist *moptlist = mp->mnt_optnew;
1446 va_list ap;
1447 int error, len;
1448 char *errmsg;
1449
1450 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1451 if (error || errmsg == NULL || len <= 0)
1452 return;
1453
1454 va_start(ap, fmt);
1455 vsnprintf(errmsg, (size_t)len, fmt, ap);
1456 va_end(ap);
1457}
1458
1459/*
1460 * Find and mount the root filesystem
1461 */
1462void
1463vfs_mountroot(void)
1464{
1465 char *cp;
1466 int error, i, asked = 0;
1467
1468 root_mount_wait();
1469
1470 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1471 NULL, NULL, mount_init, mount_fini,
1472 UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1473 devfs_first();
1474
1475 /*
1476 * We are booted with instructions to prompt for the root filesystem.
1477 */
1478 if (boothowto & RB_ASKNAME) {
1479 if (!vfs_mountroot_ask())
1480 return;
1481 asked = 1;
1482 }
1483
1484 /*
1485 * The root filesystem information is compiled in, and we are
1486 * booted with instructions to use it.
1487 */
1488 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1489 if (!vfs_mountroot_try(ctrootdevname))
1490 return;
1491 ctrootdevname = NULL;
1492 }
1493
1494 /*
1495 * We've been given the generic "use CDROM as root" flag. This is
1496 * necessary because one media may be used in many different
1497 * devices, so we need to search for them.
1498 */
1499 if (boothowto & RB_CDROM) {
1500 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1501 if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1502 return;
1503 }
1504 }
1505
1506 /*
1507 * Try to use the value read by the loader from /etc/fstab, or
1508 * supplied via some other means. This is the preferred
1509 * mechanism.
1510 */
1511 cp = getenv("vfs.root.mountfrom");
1512 if (cp != NULL) {
1513 error = vfs_mountroot_try(cp);
1514 freeenv(cp);
1515 if (!error)
1516 return;
1517 }
1518
1519 /*
1520 * Try values that may have been computed by code during boot
1521 */
1522 if (!vfs_mountroot_try(rootdevnames[0]))
1523 return;
1524 if (!vfs_mountroot_try(rootdevnames[1]))
1525 return;
1526
1527 /*
1528 * If we (still) have a compiled-in default, try it.
1529 */
1530 if (ctrootdevname != NULL)
1531 if (!vfs_mountroot_try(ctrootdevname))
1532 return;
1533 /*
1534 * Everything so far has failed, prompt on the console if we haven't
1535 * already tried that.
1536 */
1537 if (!asked)
1538 if (!vfs_mountroot_ask())
1539 return;
1540
1541 panic("Root mount failed, startup aborted.");
1542}
1543
1544/*
1545 * Mount (mountfrom) as the root filesystem.
1546 */
1547static int
1548vfs_mountroot_try(const char *mountfrom)
1549{
1550 struct mount *mp;
1551 char *vfsname, *path;
1552 time_t timebase;
1553 int error;
1554 char patt[32];
1555
1556 vfsname = NULL;
1557 path = NULL;
1558 mp = NULL;
1559 error = EINVAL;
1560
1561 if (mountfrom == NULL)
1562 return (error); /* don't complain */
1563 printf("Trying to mount root from %s\n", mountfrom);
1564
1565 /* parse vfs name and path */
1566 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1567 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1568 vfsname[0] = path[0] = 0;
1569 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1570 if (sscanf(mountfrom, patt, vfsname, path) < 1)
1571 goto out;
1572
1573 if (path[0] == '\0')
1574 strcpy(path, ROOTNAME);
1575
1576 error = kernel_vmount(
1577 MNT_RDONLY | MNT_ROOTFS,
1578 "fstype", vfsname,
1579 "fspath", "/",
1580 "from", path,
1581 NULL);
1582 if (error == 0) {
1583 /*
1584 * We mount devfs prior to mounting the / FS, so the first
1585 * entry will typically be devfs.
1586 */
1587 mp = TAILQ_FIRST(&mountlist);
1588 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1589
1590 /*
1591 * Iterate over all currently mounted file systems and use
1592 * the time stamp found to check and/or initialize the RTC.
1593 * Typically devfs has no time stamp and the only other FS
1594 * is the actual / FS.
1595 * Call inittodr() only once and pass it the largest of the
1596 * timestamps we encounter.
1597 */
1598 timebase = 0;
1599 do {
1600 if (mp->mnt_time > timebase)
1601 timebase = mp->mnt_time;
1602 mp = TAILQ_NEXT(mp, mnt_list);
1603 } while (mp != NULL);
1604 inittodr(timebase);
1605
1606 devfs_fixup(curthread);
1607 }
1608out:
1609 free(path, M_MOUNT);
1610 free(vfsname, M_MOUNT);
1611 return (error);
1612}
1613
1614/*
1615 * ---------------------------------------------------------------------
1616 * Interactive root filesystem selection code.
1617 */
1618
1619static int
1620vfs_mountroot_ask(void)
1621{
1622 char name[128];
1623
1624 for(;;) {
1625 printf("\nManual root filesystem specification:\n");
1626 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n");
1627#if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1628 printf(" eg. ufs:da0s1a\n");
1629#else
1630 printf(" eg. ufs:/dev/da0a\n");
1631#endif
1632 printf(" ? List valid disk boot devices\n");
1633 printf(" <empty line> Abort manual input\n");
1634 printf("\nmountroot> ");
1635 gets(name, sizeof(name), 1);
1636 if (name[0] == '\0')
1637 return (1);
1638 if (name[0] == '?') {
1639 printf("\nList of GEOM managed disk devices:\n ");
1640 g_dev_print();
1641 continue;
1642 }
1643 if (!vfs_mountroot_try(name))
1644 return (0);
1645 }
1646}
1647
1648/*
1649 * ---------------------------------------------------------------------
1650 * Functions for querying mount options/arguments from filesystems.
1651 */
1652
1653/*
1654 * Check that no unknown options are given
1655 */
1656int
1657vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1658{
1659 struct vfsopt *opt;
1660 const char **t, *p;
1661
1662
1663 TAILQ_FOREACH(opt, opts, link) {
1664 p = opt->name;
1665 if (p[0] == 'n' && p[1] == 'o')
1666 p += 2;
1667 for(t = global_opts; *t != NULL; t++)
1668 if (!strcmp(*t, p))
1669 break;
1670 if (*t != NULL)
1671 continue;
1672 for(t = legal; *t != NULL; t++)
1673 if (!strcmp(*t, p))
1674 break;
1675 if (*t != NULL)
1676 continue;
1677 printf("mount option <%s> is unknown\n", p);
1678 return (EINVAL);
1679 }
1680 return (0);
1681}
1682
1683/*
1684 * Get a mount option by its name.
1685 *
1686 * Return 0 if the option was found, ENOENT otherwise.
1687 * If len is non-NULL it will be filled with the length
1688 * of the option. If buf is non-NULL, it will be filled
1689 * with the address of the option.
1690 */
1691int
1692vfs_getopt(opts, name, buf, len)
1693 struct vfsoptlist *opts;
1694 const char *name;
1695 void **buf;
1696 int *len;
1697{
1698 struct vfsopt *opt;
1699
1700 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1701
1702 TAILQ_FOREACH(opt, opts, link) {
1703 if (strcmp(name, opt->name) == 0) {
1704 if (len != NULL)
1705 *len = opt->len;
1706 if (buf != NULL)
1707 *buf = opt->value;
1708 return (0);
1709 }
1710 }
1711 return (ENOENT);
1712}
1713
1714static int
1715vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1716{
1717 struct vfsopt *opt;
1718 int i;
1719
1720 if (opts == NULL)
1721 return (-1);
1722
1723 i = 0;
1724 TAILQ_FOREACH(opt, opts, link) {
1725 if (strcmp(name, opt->name) == 0)
1726 return (i);
1727 ++i;
1728 }
1729 return (-1);
1730}
1731
1732char *
1733vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1734{
1735 struct vfsopt *opt;
1736
1737 *error = 0;
1738 TAILQ_FOREACH(opt, opts, link) {
1739 if (strcmp(name, opt->name) != 0)
1740 continue;
1741 if (((char *)opt->value)[opt->len - 1] != '\0') {
1742 *error = EINVAL;
1743 return (NULL);
1744 }
1745 return (opt->value);
1746 }
1747 return (NULL);
1748}
1749
1750int
1751vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1752{
1753 struct vfsopt *opt;
1754
1755 TAILQ_FOREACH(opt, opts, link) {
1756 if (strcmp(name, opt->name) == 0) {
1757 if (w != NULL)
1758 *w |= val;
1759 return (1);
1760 }
1761 }
1762 if (w != NULL)
1763 *w &= ~val;
1764 return (0);
1765}
1766
1767int
1768vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1769{
1770 va_list ap;
1771 struct vfsopt *opt;
1772 int ret;
1773
1774 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1775
1776 TAILQ_FOREACH(opt, opts, link) {
1777 if (strcmp(name, opt->name) != 0)
1778 continue;
1779 if (((char *)opt->value)[opt->len - 1] != '\0')
1780 return (0);
1781 va_start(ap, fmt);
1782 ret = vsscanf(opt->value, fmt, ap);
1783 va_end(ap);
1784 return (ret);
1785 }
1786 return (0);
1787}
1788
1789/*
1790 * Find and copy a mount option.
1791 *
1792 * The size of the buffer has to be specified
1793 * in len, if it is not the same length as the
1794 * mount option, EINVAL is returned.
1795 * Returns ENOENT if the option is not found.
1796 */
1797int
1798vfs_copyopt(opts, name, dest, len)
1799 struct vfsoptlist *opts;
1800 const char *name;
1801 void *dest;
1802 int len;
1803{
1804 struct vfsopt *opt;
1805
1806 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1807
1808 TAILQ_FOREACH(opt, opts, link) {
1809 if (strcmp(name, opt->name) == 0) {
1810 if (len != opt->len)
1811 return (EINVAL);
1812 bcopy(opt->value, dest, opt->len);
1813 return (0);
1814 }
1815 }
1816 return (ENOENT);
1817}
1818
1819/*
1820 * This is a helper function for filesystems to traverse their
1821 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h
1822 */
1823
1824struct vnode *
1825__mnt_vnode_next(struct vnode **mvp, struct mount *mp)
1826{
1827 struct vnode *vp;
1828
1829 mtx_assert(MNT_MTX(mp), MA_OWNED);
1830
1831 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1832 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
1833 while (vp != NULL && vp->v_type == VMARKER)
1834 vp = TAILQ_NEXT(vp, v_nmntvnodes);
1835
1836 /* Check if we are done */
1837 if (vp == NULL) {
1838 __mnt_vnode_markerfree(mvp, mp);
1839 return (NULL);
1840 }
1841 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1842 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1843 return (vp);
1844}
1845
1846struct vnode *
1847__mnt_vnode_first(struct vnode **mvp, struct mount *mp)
1848{
1849 struct vnode *vp;
1850
1851 mtx_assert(MNT_MTX(mp), MA_OWNED);
1852
1853 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1854 while (vp != NULL && vp->v_type == VMARKER)
1855 vp = TAILQ_NEXT(vp, v_nmntvnodes);
1856
1857 /* Check if we are done */
1858 if (vp == NULL) {
1859 *mvp = NULL;
1860 return (NULL);
1861 }
1862 mp->mnt_holdcnt++;
1863 MNT_IUNLOCK(mp);
1864 *mvp = (struct vnode *) malloc(sizeof(struct vnode),
1865 M_VNODE_MARKER,
1866 M_WAITOK | M_ZERO);
1867 MNT_ILOCK(mp);
1868 (*mvp)->v_type = VMARKER;
1869
1870 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1871 while (vp != NULL && vp->v_type == VMARKER)
1872 vp = TAILQ_NEXT(vp, v_nmntvnodes);
1873
1874 /* Check if we are done */
1875 if (vp == NULL) {
1876 MNT_IUNLOCK(mp);
1877 free(*mvp, M_VNODE_MARKER);
1878 MNT_ILOCK(mp);
1879 *mvp = NULL;
1880 mp->mnt_holdcnt--;
1881 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1882 wakeup(&mp->mnt_holdcnt);
1883 return (NULL);
1884 }
1885 mp->mnt_markercnt++;
1886 (*mvp)->v_mount = mp;
1887 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1888 return (vp);
1889}
1890
1891
1892void
1893__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
1894{
1895
1896 if (*mvp == NULL)
1897 return;
1898
1899 mtx_assert(MNT_MTX(mp), MA_OWNED);
1900
1901 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1902 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1903 MNT_IUNLOCK(mp);
1904 free(*mvp, M_VNODE_MARKER);
1905 MNT_ILOCK(mp);
1906 *mvp = NULL;
1907
1908 mp->mnt_markercnt--;
1909 mp->mnt_holdcnt--;
1910 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1911 wakeup(&mp->mnt_holdcnt);
1912}
1913
1914
1915int
1916__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
1917{
1918 int error;
1919
1920 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
1921 if (sbp != &mp->mnt_stat)
1922 *sbp = mp->mnt_stat;
1923 return (error);
1924}
1925
1926void
1927vfs_mountedfrom(struct mount *mp, const char *from)
1928{
1929
1930 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1931 strlcpy(mp->mnt_stat.f_mntfromname, from,
1932 sizeof mp->mnt_stat.f_mntfromname);
1933}
1934
1935/*
1936 * ---------------------------------------------------------------------
1937 * This is the api for building mount args and mounting filesystems from
1938 * inside the kernel.
1939 *
1940 * The API works by accumulation of individual args. First error is
1941 * latched.
1942 *
1943 * XXX: should be documented in new manpage kernel_mount(9)
1944 */
1945
1946/* A memory allocation which must be freed when we are done */
1947struct mntaarg {
1948 SLIST_ENTRY(mntaarg) next;
1949};
1950
1951/* The header for the mount arguments */
1952struct mntarg {
1953 struct iovec *v;
1954 int len;
1955 int error;
1956 SLIST_HEAD(, mntaarg) list;
1957};
1958
1959/*
1960 * Add a boolean argument.
1961 *
1962 * flag is the boolean value.
1963 * name must start with "no".
1964 */
1965struct mntarg *
1966mount_argb(struct mntarg *ma, int flag, const char *name)
1967{
1968
1969 KASSERT(name[0] == 'n' && name[1] == 'o',
1970 ("mount_argb(...,%s): name must start with 'no'", name));
1971
1972 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1973}
1974
1975/*
1976 * Add an argument printf style
1977 */
1978struct mntarg *
1979mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1980{
1981 va_list ap;
1982 struct mntaarg *maa;
1983 struct sbuf *sb;
1984 int len;
1985
1986 if (ma == NULL) {
1987 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1988 SLIST_INIT(&ma->list);
1989 }
1990 if (ma->error)
1991 return (ma);
1992
1993 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1994 M_MOUNT, M_WAITOK);
1995 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1996 ma->v[ma->len].iov_len = strlen(name) + 1;
1997 ma->len++;
1998
1999 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
2000 va_start(ap, fmt);
2001 sbuf_vprintf(sb, fmt, ap);
2002 va_end(ap);
2003 sbuf_finish(sb);
2004 len = sbuf_len(sb) + 1;
2005 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2006 SLIST_INSERT_HEAD(&ma->list, maa, next);
2007 bcopy(sbuf_data(sb), maa + 1, len);
2008 sbuf_delete(sb);
2009
2010 ma->v[ma->len].iov_base = maa + 1;
2011 ma->v[ma->len].iov_len = len;
2012 ma->len++;
2013
2014 return (ma);
2015}
2016
2017/*
2018 * Add an argument which is a userland string.
2019 */
2020struct mntarg *
2021mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2022{
2023 struct mntaarg *maa;
2024 char *tbuf;
2025
2026 if (val == NULL)
2027 return (ma);
2028 if (ma == NULL) {
2029 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2030 SLIST_INIT(&ma->list);
2031 }
2032 if (ma->error)
2033 return (ma);
2034 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2035 SLIST_INSERT_HEAD(&ma->list, maa, next);
2036 tbuf = (void *)(maa + 1);
2037 ma->error = copyinstr(val, tbuf, len, NULL);
2038 return (mount_arg(ma, name, tbuf, -1));
2039}
2040
2041/*
2042 * Plain argument.
2043 *
2044 * If length is -1, use printf.
2045 */
2046struct mntarg *
2047mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2048{
2049
2050 if (ma == NULL) {
2051 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2052 SLIST_INIT(&ma->list);
2053 }
2054 if (ma->error)
2055 return (ma);
2056
2057 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2058 M_MOUNT, M_WAITOK);
2059 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2060 ma->v[ma->len].iov_len = strlen(name) + 1;
2061 ma->len++;
2062
2063 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2064 if (len < 0)
2065 ma->v[ma->len].iov_len = strlen(val) + 1;
2066 else
2067 ma->v[ma->len].iov_len = len;
2068 ma->len++;
2069 return (ma);
2070}
2071
2072/*
2073 * Free a mntarg structure
2074 */
2075static void
2076free_mntarg(struct mntarg *ma)
2077{
2078 struct mntaarg *maa;
2079
2080 while (!SLIST_EMPTY(&ma->list)) {
2081 maa = SLIST_FIRST(&ma->list);
2082 SLIST_REMOVE_HEAD(&ma->list, next);
2083 free(maa, M_MOUNT);
2084 }
2085 free(ma->v, M_MOUNT);
2086 free(ma, M_MOUNT);
2087}
2088
2089/*
2090 * Mount a filesystem
2091 */
2092int
2093kernel_mount(struct mntarg *ma, int flags)
2094{
2095 struct uio auio;
2096 int error;
2097
2098 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2099 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2100 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2101
2102 auio.uio_iov = ma->v;
2103 auio.uio_iovcnt = ma->len;
2104 auio.uio_segflg = UIO_SYSSPACE;
2105
2106 error = ma->error;
2107 if (!error)
2108 error = vfs_donmount(curthread, flags, &auio);
2109 free_mntarg(ma);
2110 return (error);
2111}
2112
2113/*
2114 * A printflike function to mount a filesystem.
2115 */
2116int
2117kernel_vmount(int flags, ...)
2118{
2119 struct mntarg *ma = NULL;
2120 va_list ap;
2121 const char *cp;
2122 const void *vp;
2123 int error;
2124
2125 va_start(ap, flags);
2126 for (;;) {
2127 cp = va_arg(ap, const char *);
2128 if (cp == NULL)
2129 break;
2130 vp = va_arg(ap, const void *);
2131 ma = mount_arg(ma, cp, vp, -1);
2132 }
2133 va_end(ap);
2134
2135 error = kernel_mount(ma, flags);
2136 return (error);
2137}
|