xref: /freebsd/sys/kern/vfs_mount.c (revision 7562eaabc01a48e6b11d5b558c41e3b92dae5c2d)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * Copyright (c) 1999 Michael Smith
35  * All rights reserved.
36  * Copyright (c) 1999 Poul-Henning Kamp
37  * All rights reserved.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  */
60 
61 #include <sys/cdefs.h>
62 __FBSDID("$FreeBSD$");
63 
64 #include <sys/param.h>
65 #include <sys/conf.h>
66 #include <sys/cons.h>
67 #include <sys/jail.h>
68 #include <sys/kernel.h>
69 #include <sys/linker.h>
70 #include <sys/mac.h>
71 #include <sys/malloc.h>
72 #include <sys/mount.h>
73 #include <sys/mutex.h>
74 #include <sys/namei.h>
75 #include <sys/proc.h>
76 #include <sys/filedesc.h>
77 #include <sys/reboot.h>
78 #include <sys/sysproto.h>
79 #include <sys/sx.h>
80 #include <sys/sysctl.h>
81 #include <sys/sysent.h>
82 #include <sys/systm.h>
83 #include <sys/vnode.h>
84 
85 #include <geom/geom.h>
86 
87 #include <machine/stdarg.h>
88 
89 #include "opt_rootdevname.h"
90 #include "opt_ddb.h"
91 #include "opt_mac.h"
92 
93 #ifdef DDB
94 #include <ddb/ddb.h>
95 #endif
96 
97 #define	ROOTNAME		"root_device"
98 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
99 
100 static void	checkdirs(struct vnode *olddp, struct vnode *newdp);
101 static void	gets(char *cp);
102 static int	vfs_domount(struct thread *td, const char *fstype,
103 		    char *fspath, int fsflags, void *fsdata, int compat);
104 static int	vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
105 		    const char *fspath, struct thread *td, struct mount **mpp);
106 static int	vfs_mountroot_ask(void);
107 static int	vfs_mountroot_try(const char *mountfrom);
108 static int	vfs_donmount(struct thread *td, int fsflags,
109 		    struct uio *fsoptions);
110 
111 static int	usermount = 0;
112 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
113     "Unprivileged users may mount and unmount file systems");
114 
115 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
116 
117 /* List of mounted filesystems. */
118 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
119 
120 /* For any iteration/modification of mountlist */
121 struct mtx mountlist_mtx;
122 
123 /*
124  * The vnode of the system's root (/ in the filesystem, without chroot
125  * active.)
126  */
127 struct vnode	*rootvnode;
128 
129 /*
130  * The root filesystem is detailed in the kernel environment variable
131  * vfs.root.mountfrom, which is expected to be in the general format
132  *
133  * <vfsname>:[<path>]
134  * vfsname   := the name of a VFS known to the kernel and capable
135  *              of being mounted as root
136  * path      := disk device name or other data used by the filesystem
137  *              to locate its physical store
138  */
139 
140 /*
141  * The root specifiers we will try if RB_CDROM is specified.
142  */
143 static char *cdrom_rootdevnames[] = {
144 	"cd9660:cd0",
145 	"cd9660:acd0",
146 	NULL
147 };
148 
149 /* legacy find-root code */
150 char		*rootdevnames[2] = {NULL, NULL};
151 struct cdev *rootdev = NULL;
152 #ifdef ROOTDEVNAME
153 const char	*ctrootdevname = ROOTDEVNAME;
154 #else
155 const char	*ctrootdevname = NULL;
156 #endif
157 
158 /*
159  * Has to be dynamic as the value of rootdev can change; however, it can't
160  * change after the root is mounted, so a user process can't access this
161  * sysctl until after the value is unchangeable.
162  */
163 static int
164 sysctl_rootdev(SYSCTL_HANDLER_ARGS)
165 {
166 	int error;
167 
168 	/* _RD prevents this from happening. */
169 	KASSERT(req->newptr == NULL, ("Attempt to change root device name"));
170 
171 	if (rootdev != NULL)
172 		error = sysctl_handle_string(oidp, rootdev->si_name, 0, req);
173 	else
174 		error = sysctl_handle_string(oidp, "", 0, req);
175 
176 	return (error);
177 }
178 
179 SYSCTL_PROC(_kern, OID_AUTO, rootdev, CTLTYPE_STRING | CTLFLAG_RD,
180     0, 0, sysctl_rootdev, "A", "Root file system device");
181 
182 /* Remove one mount option. */
183 static void
184 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
185 {
186 
187 	TAILQ_REMOVE(opts, opt, link);
188 	free(opt->name, M_MOUNT);
189 	if (opt->value != NULL)
190 		free(opt->value, M_MOUNT);
191 #ifdef INVARIANTS
192 	else if (opt->len != 0)
193 		panic("%s: mount option with NULL value but length != 0",
194 		    __func__);
195 #endif
196 	free(opt, M_MOUNT);
197 }
198 
199 /* Release all resources related to the mount options. */
200 static void
201 vfs_freeopts(struct vfsoptlist *opts)
202 {
203 	struct vfsopt *opt;
204 
205 	while (!TAILQ_EMPTY(opts)) {
206 		opt = TAILQ_FIRST(opts);
207 		vfs_freeopt(opts, opt);
208 	}
209 	free(opts, M_MOUNT);
210 }
211 
212 /*
213  * Check if options are equal (with or without the "no" prefix).
214  */
215 static int
216 vfs_equalopts(const char *opt1, const char *opt2)
217 {
218 
219 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
220 	if (strcmp(opt1, opt2) == 0)
221 		return (1);
222 	/* "noopt" vs. "opt" */
223 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
224 		return (1);
225 	/* "opt" vs. "noopt" */
226 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
227 		return (1);
228 	return (0);
229 }
230 
231 /*
232  * If a mount option is specified several times,
233  * (with or without the "no" prefix) only keep
234  * the last occurence of it.
235  */
236 static void
237 vfs_sanitizeopts(struct vfsoptlist *opts)
238 {
239 	struct vfsopt *opt, *opt2, *tmp;
240 
241 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
242 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
243 		while (opt2 != NULL) {
244 			if (vfs_equalopts(opt->name, opt2->name)) {
245 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
246 				vfs_freeopt(opts, opt2);
247 				opt2 = tmp;
248 			} else {
249 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
250 			}
251 		}
252 	}
253 }
254 
255 /*
256  * Build a linked list of mount options from a struct uio.
257  */
258 static int
259 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
260 {
261 	struct vfsoptlist *opts;
262 	struct vfsopt *opt;
263 	size_t memused;
264 	unsigned int i, iovcnt;
265 	int error, namelen, optlen;
266 
267 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
268 	TAILQ_INIT(opts);
269 	memused = 0;
270 	iovcnt = auio->uio_iovcnt;
271 	for (i = 0; i < iovcnt; i += 2) {
272 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
273 		namelen = auio->uio_iov[i].iov_len;
274 		optlen = auio->uio_iov[i + 1].iov_len;
275 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
276 		opt->value = NULL;
277 		opt->len = 0;
278 
279 		/*
280 		 * Do this early, so jumps to "bad" will free the current
281 		 * option.
282 		 */
283 		TAILQ_INSERT_TAIL(opts, opt, link);
284 		memused += sizeof(struct vfsopt) + optlen + namelen;
285 
286 		/*
287 		 * Avoid consuming too much memory, and attempts to overflow
288 		 * memused.
289 		 */
290 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
291 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
292 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
293 			error = EINVAL;
294 			goto bad;
295 		}
296 
297 		if (auio->uio_segflg == UIO_SYSSPACE) {
298 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
299 		} else {
300 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
301 			    namelen);
302 			if (error)
303 				goto bad;
304 		}
305 		/* Ensure names are null-terminated strings. */
306 		if (opt->name[namelen - 1] != '\0') {
307 			error = EINVAL;
308 			goto bad;
309 		}
310 		if (optlen != 0) {
311 			opt->len = optlen;
312 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
313 			if (auio->uio_segflg == UIO_SYSSPACE) {
314 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
315 				    optlen);
316 			} else {
317 				error = copyin(auio->uio_iov[i + 1].iov_base,
318 				    opt->value, optlen);
319 				if (error)
320 					goto bad;
321 			}
322 		}
323 	}
324 	vfs_sanitizeopts(opts);
325 	*options = opts;
326 	return (0);
327 bad:
328 	vfs_freeopts(opts);
329 	return (error);
330 }
331 
332 /*
333  * Merge the old mount options with the new ones passed
334  * in the MNT_UPDATE case.
335  */
336 static void
337 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
338 {
339 	struct vfsopt *opt, *opt2, *new;
340 
341 	TAILQ_FOREACH(opt, opts, link) {
342 		/*
343 		 * Check that this option hasn't been redefined
344 		 * nor cancelled with a "no" mount option.
345 		 */
346 		opt2 = TAILQ_FIRST(toopts);
347 		while (opt2 != NULL) {
348 			if (strcmp(opt2->name, opt->name) == 0)
349 				goto next;
350 			if (strncmp(opt2->name, "no", 2) == 0 &&
351 			    strcmp(opt2->name + 2, opt->name) == 0) {
352 				vfs_freeopt(toopts, opt2);
353 				goto next;
354 			}
355 			opt2 = TAILQ_NEXT(opt2, link);
356 		}
357 		/* We want this option, duplicate it. */
358 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
359 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
360 		strcpy(new->name, opt->name);
361 		if (opt->len != 0) {
362 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
363 			bcopy(opt->value, new->value, opt->len);
364 		} else {
365 			new->value = NULL;
366 		}
367 		new->len = opt->len;
368 		TAILQ_INSERT_TAIL(toopts, new, link);
369 next:
370 		continue;
371 	}
372 }
373 
374 /*
375  * New mount API.
376  */
377 int
378 nmount(td, uap)
379 	struct thread *td;
380 	struct nmount_args /* {
381 		struct iovec *iovp;
382 		unsigned int iovcnt;
383 		int flags;
384 	} */ *uap;
385 {
386 	struct uio *auio;
387 	struct iovec *iov;
388 	unsigned int i;
389 	int error;
390 	u_int iovcnt;
391 
392 	iovcnt = uap->iovcnt;
393 	/*
394 	 * Check that we have an even number of iovec's
395 	 * and that we have at least two options.
396 	 */
397 	if ((iovcnt & 1) || (iovcnt < 4))
398 		return (EINVAL);
399 	error = copyinuio(uap->iovp, iovcnt, &auio);
400 	if (error)
401 		return (error);
402 	iov = auio->uio_iov;
403 	for (i = 0; i < iovcnt; i++) {
404 		if (iov->iov_len > MMAXOPTIONLEN) {
405 			free(auio, M_IOV);
406 			return (EINVAL);
407 		}
408 		iov++;
409 	}
410 	error = vfs_donmount(td, uap->flags, auio);
411 	free(auio, M_IOV);
412 	return (error);
413 }
414 
415 int
416 kernel_mount(struct iovec *iovp, u_int iovcnt, int flags)
417 {
418 	struct uio auio;
419 	int error;
420 
421 	/*
422 	 * Check that we have an even number of iovec's
423 	 * and that we have at least two options.
424 	 */
425 	if ((iovcnt & 1) || (iovcnt < 4))
426 		return (EINVAL);
427 
428 	auio.uio_iov = iovp;
429 	auio.uio_iovcnt = iovcnt;
430 	auio.uio_segflg = UIO_SYSSPACE;
431 
432 	error = vfs_donmount(curthread, flags, &auio);
433 	return (error);
434 }
435 
436 int
437 kernel_vmount(int flags, ...)
438 {
439 	struct iovec *iovp;
440 	struct uio auio;
441 	va_list ap;
442 	u_int iovcnt, iovlen, len;
443 	const char *cp;
444 	char *buf, *pos;
445 	size_t n;
446 	int error, i;
447 
448 	len = 0;
449 	va_start(ap, flags);
450 	for (iovcnt = 0; (cp = va_arg(ap, const char *)) != NULL; iovcnt++)
451 		len += strlen(cp) + 1;
452 	va_end(ap);
453 
454 	if (iovcnt < 4 || iovcnt & 1)
455 		return (EINVAL);
456 
457 	iovlen = iovcnt * sizeof (struct iovec);
458 	MALLOC(iovp, struct iovec *, iovlen, M_MOUNT, M_WAITOK);
459 	MALLOC(buf, char *, len, M_MOUNT, M_WAITOK);
460 	pos = buf;
461 	va_start(ap, flags);
462 	for (i = 0; i < iovcnt; i++) {
463 		cp = va_arg(ap, const char *);
464 		copystr(cp, pos, len - (pos - buf), &n);
465 		iovp[i].iov_base = pos;
466 		iovp[i].iov_len = n;
467 		pos += n;
468 	}
469 	va_end(ap);
470 
471 	auio.uio_iov = iovp;
472 	auio.uio_iovcnt = iovcnt;
473 	auio.uio_segflg = UIO_SYSSPACE;
474 
475 	error = vfs_donmount(curthread, flags, &auio);
476 	FREE(iovp, M_MOUNT);
477 	FREE(buf, M_MOUNT);
478 	return (error);
479 }
480 
481 /*
482  * Allocate and initialize the mount point struct.
483  */
484 static int
485 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
486     const char *fspath, struct thread *td, struct mount **mpp)
487 {
488 	struct mount *mp;
489 
490 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
491 	TAILQ_INIT(&mp->mnt_nvnodelist);
492 	mp->mnt_nvnodelistsize = 0;
493 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
494 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
495 	vfs_busy(mp, LK_NOWAIT, 0, td);
496 	mp->mnt_op = vfsp->vfc_vfsops;
497 	mp->mnt_vfc = vfsp;
498 	vfsp->vfc_refcount++;
499 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
500 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
501 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
502 	mp->mnt_vnodecovered = vp;
503 	mp->mnt_cred = crdup(td->td_ucred);
504 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
505 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
506 	mp->mnt_iosize_max = DFLTPHYS;
507 #ifdef MAC
508 	mac_init_mount(mp);
509 	mac_create_mount(td->td_ucred, mp);
510 #endif
511 	*mpp = mp;
512 	return (0);
513 }
514 
515 /*
516  * Destroy the mount struct previously allocated by vfs_mount_alloc().
517  */
518 void
519 vfs_mount_destroy(struct mount *mp, struct thread *td)
520 {
521 
522 	mp->mnt_vfc->vfc_refcount--;
523 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
524 		panic("unmount: dangling vnode");
525 	vfs_unbusy(mp,td);
526 	lockdestroy(&mp->mnt_lock);
527 	mtx_destroy(&mp->mnt_mtx);
528 	if (mp->mnt_kern_flag & MNTK_MWAIT)
529 		wakeup(mp);
530 #ifdef MAC
531 	mac_destroy_mount(mp);
532 #endif
533 	if (mp->mnt_opt != NULL)
534 		vfs_freeopts(mp->mnt_opt);
535 	crfree(mp->mnt_cred);
536 	free(mp, M_MOUNT);
537 }
538 
539 static int
540 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
541 {
542 	struct vfsoptlist *optlist;
543 	char *fstype, *fspath;
544 	int error, fstypelen, fspathlen;
545 
546 	error = vfs_buildopts(fsoptions, &optlist);
547 	if (error)
548 		return (error);
549 
550 	/*
551 	 * We need these two options before the others,
552 	 * and they are mandatory for any filesystem.
553 	 * Ensure they are NUL terminated as well.
554 	 */
555 	fstypelen = 0;
556 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
557 	if (error || fstype[fstypelen - 1] != '\0') {
558 		error = EINVAL;
559 		goto bail;
560 	}
561 	fspathlen = 0;
562 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
563 	if (error || fspath[fspathlen - 1] != '\0') {
564 		error = EINVAL;
565 		goto bail;
566 	}
567 
568 	/*
569 	 * Be ultra-paranoid about making sure the type and fspath
570 	 * variables will fit in our mp buffers, including the
571 	 * terminating NUL.
572 	 */
573 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
574 		error = ENAMETOOLONG;
575 		goto bail;
576 	}
577 
578 	error = vfs_domount(td, fstype, fspath, fsflags, optlist, 0);
579 bail:
580 	if (error)
581 		vfs_freeopts(optlist);
582 	return (error);
583 }
584 
585 /*
586  * Old mount API.
587  */
588 #ifndef _SYS_SYSPROTO_H_
589 struct mount_args {
590 	char	*type;
591 	char	*path;
592 	int	flags;
593 	caddr_t	data;
594 };
595 #endif
596 /* ARGSUSED */
597 int
598 mount(td, uap)
599 	struct thread *td;
600 	struct mount_args /* {
601 		char *type;
602 		char *path;
603 		int flags;
604 		caddr_t data;
605 	} */ *uap;
606 {
607 	char *fstype;
608 	char *fspath;
609 	int error;
610 
611 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
612 	fspath = malloc(MNAMELEN, M_TEMP, M_WAITOK);
613 
614 	/*
615 	 * vfs_mount() actually takes a kernel string for `type' and
616 	 * `path' now, so extract them.
617 	 */
618 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
619 	if (error == 0)
620 		error = copyinstr(uap->path, fspath, MNAMELEN, NULL);
621 	if (error == 0)
622 		error = vfs_domount(td, fstype, fspath, uap->flags,
623 		    uap->data, 1);
624 	free(fstype, M_TEMP);
625 	free(fspath, M_TEMP);
626 	return (error);
627 }
628 
629 /*
630  * vfs_mount(): actually attempt a filesystem mount.
631  *
632  * This routine is designed to be a "generic" entry point for routines
633  * that wish to mount a filesystem. All parameters except `fsdata' are
634  * pointers into kernel space. `fsdata' is currently still a pointer
635  * into userspace.
636  */
637 int
638 vfs_mount(td, fstype, fspath, fsflags, fsdata)
639 	struct thread *td;
640 	const char *fstype;
641 	char *fspath;
642 	int fsflags;
643 	void *fsdata;
644 {
645 
646 	return (vfs_domount(td, fstype, fspath, fsflags, fsdata, 1));
647 }
648 
649 /*
650  * vfs_domount(): actually attempt a filesystem mount.
651  */
652 static int
653 vfs_domount(
654 	struct thread *td,	/* Flags common to all filesystems. */
655 	const char *fstype,	/* Filesystem type. */
656 	char *fspath,		/* Mount path. */
657 	int fsflags,		/* Flags common to all filesystems. */
658 	void *fsdata,		/* Options local to the filesystem. */
659 	int compat		/* Invocation from compat syscall. */
660 	)
661 {
662 	linker_file_t lf;
663 	struct vnode *vp;
664 	struct mount *mp;
665 	struct vfsconf *vfsp;
666 	int error, flag = 0, kern_flag = 0;
667 	struct vattr va;
668 	struct nameidata nd;
669 
670 	/*
671 	 * Be ultra-paranoid about making sure the type and fspath
672 	 * variables will fit in our mp buffers, including the
673 	 * terminating NUL.
674 	 */
675 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
676 		return (ENAMETOOLONG);
677 
678 	if (jailed(td->td_ucred))
679 		return (EPERM);
680 	if (usermount == 0) {
681 		if ((error = suser(td)) != 0)
682 			return (error);
683 	}
684 
685 	/*
686 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
687 	 */
688 	if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
689 		if ((error = suser(td)) != 0)
690 			return (error);
691 	}
692 	/*
693 	 * Silently enforce MNT_NODEV, MNT_NOSUID and MNT_USER for
694 	 * unprivileged users.
695 	 */
696 	if (suser(td) != 0)
697 		fsflags |= MNT_NODEV | MNT_NOSUID | MNT_USER;
698 	/*
699 	 * Get vnode to be covered
700 	 */
701 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td);
702 	if ((error = namei(&nd)) != 0)
703 		return (error);
704 	NDFREE(&nd, NDF_ONLY_PNBUF);
705 	vp = nd.ni_vp;
706 	if (fsflags & MNT_UPDATE) {
707 		if ((vp->v_vflag & VV_ROOT) == 0) {
708 			vput(vp);
709 			return (EINVAL);
710 		}
711 		mp = vp->v_mount;
712 		flag = mp->mnt_flag;
713 		kern_flag = mp->mnt_kern_flag;
714 		/*
715 		 * We only allow the filesystem to be reloaded if it
716 		 * is currently mounted read-only.
717 		 */
718 		if ((fsflags & MNT_RELOAD) &&
719 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
720 			vput(vp);
721 			return (EOPNOTSUPP);	/* Needs translation */
722 		}
723 		/*
724 		 * Only privileged root, or (if MNT_USER is set) the user that
725 		 * did the original mount is permitted to update it.
726 		 */
727 		error = vfs_suser(mp, td);
728 		if (error) {
729 			vput(vp);
730 			return (error);
731 		}
732 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
733 			vput(vp);
734 			return (EBUSY);
735 		}
736 		VI_LOCK(vp);
737 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
738 		    vp->v_mountedhere != NULL) {
739 			VI_UNLOCK(vp);
740 			vfs_unbusy(mp, td);
741 			vput(vp);
742 			return (EBUSY);
743 		}
744 		vp->v_iflag |= VI_MOUNT;
745 		VI_UNLOCK(vp);
746 		mp->mnt_flag |= fsflags &
747 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT);
748 		VOP_UNLOCK(vp, 0, td);
749 		if (compat == 0) {
750 			mp->mnt_optnew = fsdata;
751 			vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
752 		}
753 	} else {
754 		/*
755 		 * If the user is not root, ensure that they own the directory
756 		 * onto which we are attempting to mount.
757 		 */
758 		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
759 		if (error) {
760 			vput(vp);
761 			return (error);
762 		}
763 		if (va.va_uid != td->td_ucred->cr_uid) {
764 			if ((error = suser(td)) != 0) {
765 				vput(vp);
766 				return (error);
767 			}
768 		}
769 		if ((error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0)) != 0) {
770 			vput(vp);
771 			return (error);
772 		}
773 		if (vp->v_type != VDIR) {
774 			vput(vp);
775 			return (ENOTDIR);
776 		}
777 		vfsp = vfs_byname(fstype);
778 		if (vfsp == NULL) {
779 			/* Only load modules for root (very important!). */
780 			if ((error = suser(td)) != 0) {
781 				vput(vp);
782 				return (error);
783 			}
784 			error = securelevel_gt(td->td_ucred, 0);
785 			if (error) {
786 				vput(vp);
787 				return (error);
788 			}
789 			error = linker_load_module(NULL, fstype, NULL, NULL, &lf);
790 			if (error || lf == NULL) {
791 				vput(vp);
792 				if (lf == NULL)
793 					error = ENODEV;
794 				return (error);
795 			}
796 			lf->userrefs++;
797 			/* Look up again to see if the VFS was loaded. */
798 			vfsp = vfs_byname(fstype);
799 			if (vfsp == NULL) {
800 				lf->userrefs--;
801 				linker_file_unload(lf, LINKER_UNLOAD_FORCE);
802 				vput(vp);
803 				return (ENODEV);
804 			}
805 		}
806 		VI_LOCK(vp);
807 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
808 		    vp->v_mountedhere != NULL) {
809 			VI_UNLOCK(vp);
810 			vput(vp);
811 			return (EBUSY);
812 		}
813 		vp->v_iflag |= VI_MOUNT;
814 		VI_UNLOCK(vp);
815 
816 		/*
817 		 * Allocate and initialize the filesystem.
818 		 */
819 		error = vfs_mount_alloc(vp, vfsp, fspath, td, &mp);
820 		if (error) {
821 			vput(vp);
822 			return (error);
823 		}
824 		VOP_UNLOCK(vp, 0, td);
825 
826 		/* XXXMAC: pass to vfs_mount_alloc? */
827 		if (compat == 0)
828 			mp->mnt_optnew = fsdata;
829 	}
830 	/*
831 	 * Check if the fs implements the type VFS_[O]MOUNT()
832 	 * function we are looking for.
833 	 */
834 	if ((compat == 0) == (mp->mnt_op->vfs_omount != NULL)) {
835 		printf("%s doesn't support the %s mount syscall\n",
836 		    mp->mnt_vfc->vfc_name, compat ? "old" : "new");
837 		VI_LOCK(vp);
838 		vp->v_iflag &= ~VI_MOUNT;
839 		VI_UNLOCK(vp);
840 		if (mp->mnt_flag & MNT_UPDATE)
841 			vfs_unbusy(mp, td);
842 		else
843 			vfs_mount_destroy(mp, td);
844 		vrele(vp);
845 		return (EOPNOTSUPP);
846 	}
847 
848 	/*
849 	 * Set the mount level flags.
850 	 */
851 	if (fsflags & MNT_RDONLY)
852 		mp->mnt_flag |= MNT_RDONLY;
853 	else if (mp->mnt_flag & MNT_RDONLY)
854 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
855 	mp->mnt_flag &=~ MNT_UPDATEMASK;
856 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE);
857 	/*
858 	 * Mount the filesystem.
859 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
860 	 * get.  No freeing of cn_pnbuf.
861 	 */
862 	if (compat)
863 	    error = VFS_OMOUNT(mp, fspath, fsdata, td);
864 	else
865 	    error = VFS_MOUNT(mp, td);
866 	if (!error) {
867 		if (mp->mnt_opt != NULL)
868 			vfs_freeopts(mp->mnt_opt);
869 		mp->mnt_opt = mp->mnt_optnew;
870 	}
871 	/*
872 	 * Prevent external consumers of mount options from reading
873 	 * mnt_optnew.
874 	*/
875 	mp->mnt_optnew = NULL;
876 	if (mp->mnt_flag & MNT_UPDATE) {
877 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
878 			mp->mnt_flag &= ~MNT_RDONLY;
879 		mp->mnt_flag &=
880 		    ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
881 		mp->mnt_kern_flag &= ~MNTK_WANTRDWR;
882 		if (error) {
883 			mp->mnt_flag = flag;
884 			mp->mnt_kern_flag = kern_flag;
885 		}
886 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
887 			if (mp->mnt_syncer == NULL)
888 				error = vfs_allocate_syncvnode(mp);
889 		} else {
890 			if (mp->mnt_syncer != NULL)
891 				vrele(mp->mnt_syncer);
892 			mp->mnt_syncer = NULL;
893 		}
894 		vfs_unbusy(mp, td);
895 		VI_LOCK(vp);
896 		vp->v_iflag &= ~VI_MOUNT;
897 		VI_UNLOCK(vp);
898 		vrele(vp);
899 		return (error);
900 	}
901 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
902 	/*
903 	 * Put the new filesystem on the mount list after root.
904 	 */
905 	cache_purge(vp);
906 	if (!error) {
907 		struct vnode *newdp;
908 
909 		VI_LOCK(vp);
910 		vp->v_iflag &= ~VI_MOUNT;
911 		VI_UNLOCK(vp);
912 		vp->v_mountedhere = mp;
913 		mtx_lock(&mountlist_mtx);
914 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
915 		mtx_unlock(&mountlist_mtx);
916 		vfs_event_signal(NULL, VQ_MOUNT, 0);
917 		if (VFS_ROOT(mp, &newdp, td))
918 			panic("mount: lost mount");
919 		checkdirs(vp, newdp);
920 		vput(newdp);
921 		VOP_UNLOCK(vp, 0, td);
922 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
923 			error = vfs_allocate_syncvnode(mp);
924 		vfs_unbusy(mp, td);
925 		if (error || (error = VFS_START(mp, 0, td)) != 0)
926 			vrele(vp);
927 	} else {
928 		VI_LOCK(vp);
929 		vp->v_iflag &= ~VI_MOUNT;
930 		VI_UNLOCK(vp);
931 		vfs_mount_destroy(mp, td);
932 		vput(vp);
933 	}
934 	return (error);
935 }
936 
937 /*
938  * Scan all active processes to see if any of them have a current
939  * or root directory of `olddp'. If so, replace them with the new
940  * mount point.
941  */
942 static void
943 checkdirs(olddp, newdp)
944 	struct vnode *olddp, *newdp;
945 {
946 	struct filedesc *fdp;
947 	struct proc *p;
948 	int nrele;
949 
950 	if (vrefcnt(olddp) == 1)
951 		return;
952 	sx_slock(&allproc_lock);
953 	LIST_FOREACH(p, &allproc, p_list) {
954 		mtx_lock(&fdesc_mtx);
955 		fdp = p->p_fd;
956 		if (fdp == NULL) {
957 			mtx_unlock(&fdesc_mtx);
958 			continue;
959 		}
960 		nrele = 0;
961 		FILEDESC_LOCK(fdp);
962 		if (fdp->fd_cdir == olddp) {
963 			VREF(newdp);
964 			fdp->fd_cdir = newdp;
965 			nrele++;
966 		}
967 		if (fdp->fd_rdir == olddp) {
968 			VREF(newdp);
969 			fdp->fd_rdir = newdp;
970 			nrele++;
971 		}
972 		FILEDESC_UNLOCK(fdp);
973 		mtx_unlock(&fdesc_mtx);
974 		while (nrele--)
975 			vrele(olddp);
976 	}
977 	sx_sunlock(&allproc_lock);
978 	if (rootvnode == olddp) {
979 		vrele(rootvnode);
980 		VREF(newdp);
981 		rootvnode = newdp;
982 	}
983 }
984 
985 /*
986  * Unmount a filesystem.
987  *
988  * Note: unmount takes a path to the vnode mounted on as argument,
989  * not special file (as before).
990  */
991 #ifndef _SYS_SYSPROTO_H_
992 struct unmount_args {
993 	char	*path;
994 	int	flags;
995 };
996 #endif
997 /* ARGSUSED */
998 int
999 unmount(td, uap)
1000 	struct thread *td;
1001 	register struct unmount_args /* {
1002 		char *path;
1003 		int flags;
1004 	} */ *uap;
1005 {
1006 	struct mount *mp;
1007 	char *pathbuf;
1008 	int error, id0, id1;
1009 
1010 	if (jailed(td->td_ucred))
1011 		return (EPERM);
1012 	if (usermount == 0) {
1013 		if ((error = suser(td)) != 0)
1014 			return (error);
1015 	}
1016 
1017 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1018 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1019 	if (error) {
1020 		free(pathbuf, M_TEMP);
1021 		return (error);
1022 	}
1023 	if (uap->flags & MNT_BYFSID) {
1024 		/* Decode the filesystem ID. */
1025 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1026 			free(pathbuf, M_TEMP);
1027 			return (EINVAL);
1028 		}
1029 
1030 		mtx_lock(&mountlist_mtx);
1031 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1032 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1033 			    mp->mnt_stat.f_fsid.val[1] == id1)
1034 				break;
1035 		}
1036 		mtx_unlock(&mountlist_mtx);
1037 	} else {
1038 		mtx_lock(&mountlist_mtx);
1039 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1040 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1041 				break;
1042 		}
1043 		mtx_unlock(&mountlist_mtx);
1044 	}
1045 	free(pathbuf, M_TEMP);
1046 	if (mp == NULL) {
1047 		/*
1048 		 * Previously we returned ENOENT for a nonexistent path and
1049 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1050 		 * now, so in the !MNT_BYFSID case return the more likely
1051 		 * EINVAL for compatibility.
1052 		 */
1053 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1054 	}
1055 
1056 	/*
1057 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1058 	 * original mount is permitted to unmount this filesystem.
1059 	 */
1060 	error = vfs_suser(mp, td);
1061 	if (error)
1062 		return (error);
1063 
1064 	/*
1065 	 * Don't allow unmounting the root filesystem.
1066 	 */
1067 	if (mp->mnt_flag & MNT_ROOTFS)
1068 		return (EINVAL);
1069 	return (dounmount(mp, uap->flags, td));
1070 }
1071 
1072 /*
1073  * Do the actual filesystem unmount.
1074  */
1075 int
1076 dounmount(mp, flags, td)
1077 	struct mount *mp;
1078 	int flags;
1079 	struct thread *td;
1080 {
1081 	struct vnode *coveredvp, *fsrootvp;
1082 	int error;
1083 	int async_flag;
1084 
1085 	mtx_lock(&mountlist_mtx);
1086 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1087 		mtx_unlock(&mountlist_mtx);
1088 		return (EBUSY);
1089 	}
1090 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1091 	/* Allow filesystems to detect that a forced unmount is in progress. */
1092 	if (flags & MNT_FORCE)
1093 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1094 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1095 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), &mountlist_mtx, td);
1096 	if (error) {
1097 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1098 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1099 			wakeup(mp);
1100 		return (error);
1101 	}
1102 	vn_start_write(NULL, &mp, V_WAIT);
1103 
1104 	if (mp->mnt_flag & MNT_EXPUBLIC)
1105 		vfs_setpublicfs(NULL, NULL, NULL);
1106 
1107 	vfs_msync(mp, MNT_WAIT);
1108 	async_flag = mp->mnt_flag & MNT_ASYNC;
1109 	mp->mnt_flag &= ~MNT_ASYNC;
1110 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1111 	if (mp->mnt_syncer != NULL)
1112 		vrele(mp->mnt_syncer);
1113 	/*
1114 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1115 	 * vnode to the covered vnode.  For non-forced unmounts we want
1116 	 * such references to cause an EBUSY error.
1117 	 */
1118 	if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) {
1119 		if (mp->mnt_vnodecovered != NULL)
1120 			checkdirs(fsrootvp, mp->mnt_vnodecovered);
1121 		if (fsrootvp == rootvnode) {
1122 			vrele(rootvnode);
1123 			rootvnode = NULL;
1124 		}
1125 		vput(fsrootvp);
1126 	}
1127 	if (((mp->mnt_flag & MNT_RDONLY) ||
1128 	     (error = VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td)) == 0) ||
1129 	    (flags & MNT_FORCE)) {
1130 		error = VFS_UNMOUNT(mp, flags, td);
1131 	}
1132 	vn_finished_write(mp);
1133 	if (error) {
1134 		/* Undo cdir/rdir and rootvnode changes made above. */
1135 		if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) {
1136 			if (mp->mnt_vnodecovered != NULL)
1137 				checkdirs(mp->mnt_vnodecovered, fsrootvp);
1138 			if (rootvnode == NULL) {
1139 				rootvnode = fsrootvp;
1140 				vref(rootvnode);
1141 			}
1142 			vput(fsrootvp);
1143 		}
1144 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1145 			(void) vfs_allocate_syncvnode(mp);
1146 		mtx_lock(&mountlist_mtx);
1147 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1148 		mp->mnt_flag |= async_flag;
1149 		lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK,
1150 		    &mountlist_mtx, td);
1151 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1152 			wakeup(mp);
1153 		return (error);
1154 	}
1155 	mtx_lock(&mountlist_mtx);
1156 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1157 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1158 		coveredvp->v_mountedhere = NULL;
1159 	mtx_unlock(&mountlist_mtx);
1160 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1161 	vfs_mount_destroy(mp, td);
1162 	if (coveredvp != NULL)
1163 		vrele(coveredvp);
1164 	return (0);
1165 }
1166 
1167 /*
1168  * Lookup a filesystem type, and if found allocate and initialize
1169  * a mount structure for it.
1170  *
1171  * Devname is usually updated by mount(8) after booting.
1172  */
1173 int
1174 vfs_rootmountalloc(fstypename, devname, mpp)
1175 	char *fstypename;
1176 	char *devname;
1177 	struct mount **mpp;
1178 {
1179 	struct thread *td = curthread;	/* XXX */
1180 	struct vfsconf *vfsp;
1181 	struct mount *mp;
1182 	int error;
1183 
1184 	if (fstypename == NULL)
1185 		return (ENODEV);
1186 	vfsp = vfs_byname(fstypename);
1187 	if (vfsp == NULL)
1188 		return (ENODEV);
1189 	error = vfs_mount_alloc(NULLVP, vfsp, "/", td, &mp);
1190 	if (error)
1191 		return (error);
1192 	mp->mnt_flag |= MNT_RDONLY | MNT_ROOTFS;
1193 	strlcpy(mp->mnt_stat.f_mntfromname, devname, MNAMELEN);
1194 	*mpp = mp;
1195 	return (0);
1196 }
1197 
1198 /*
1199  * Find and mount the root filesystem
1200  */
1201 void
1202 vfs_mountroot(void)
1203 {
1204 	char *cp;
1205 	int error, i, asked = 0;
1206 
1207 
1208 	/*
1209 	 * Wait for GEOM to settle down
1210 	 */
1211 	DROP_GIANT();
1212 	g_waitidle();
1213 	PICKUP_GIANT();
1214 
1215 	/*
1216 	 * We are booted with instructions to prompt for the root filesystem.
1217 	 */
1218 	if (boothowto & RB_ASKNAME) {
1219 		if (!vfs_mountroot_ask())
1220 			return;
1221 		asked = 1;
1222 	}
1223 
1224 	/*
1225 	 * The root filesystem information is compiled in, and we are
1226 	 * booted with instructions to use it.
1227 	 */
1228 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1229 		if (!vfs_mountroot_try(ctrootdevname))
1230 			return;
1231 		ctrootdevname = NULL;
1232 	}
1233 
1234 	/*
1235 	 * We've been given the generic "use CDROM as root" flag.  This is
1236 	 * necessary because one media may be used in many different
1237 	 * devices, so we need to search for them.
1238 	 */
1239 	if (boothowto & RB_CDROM) {
1240 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1241 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1242 				return;
1243 		}
1244 	}
1245 
1246 	/*
1247 	 * Try to use the value read by the loader from /etc/fstab, or
1248 	 * supplied via some other means.  This is the preferred
1249 	 * mechanism.
1250 	 */
1251 	cp = getenv("vfs.root.mountfrom");
1252 	if (cp != NULL) {
1253 		error = vfs_mountroot_try(cp);
1254 		freeenv(cp);
1255 		if (!error)
1256 			return;
1257 	}
1258 
1259 	/*
1260 	 * Try values that may have been computed by code during boot
1261 	 */
1262 	if (!vfs_mountroot_try(rootdevnames[0]))
1263 		return;
1264 	if (!vfs_mountroot_try(rootdevnames[1]))
1265 		return;
1266 
1267 	/*
1268 	 * If we (still) have a compiled-in default, try it.
1269 	 */
1270 	if (ctrootdevname != NULL)
1271 		if (!vfs_mountroot_try(ctrootdevname))
1272 			return;
1273 
1274 	/*
1275 	 * Everything so far has failed, prompt on the console if we haven't
1276 	 * already tried that.
1277 	 */
1278 	if (!asked)
1279 		if (!vfs_mountroot_ask())
1280 			return;
1281 	panic("Root mount failed, startup aborted.");
1282 }
1283 
1284 /*
1285  * Mount (mountfrom) as the root filesystem.
1286  */
1287 static int
1288 vfs_mountroot_try(const char *mountfrom)
1289 {
1290         struct mount	*mp;
1291 	char		*vfsname, *path;
1292 	const char	*devname;
1293 	int		error;
1294 	char		patt[32];
1295 	int		s;
1296 
1297 	vfsname = NULL;
1298 	path    = NULL;
1299 	mp      = NULL;
1300 	error   = EINVAL;
1301 
1302 	if (mountfrom == NULL)
1303 		return (error);		/* don't complain */
1304 
1305 	s = splcam();			/* Overkill, but annoying without it */
1306 	printf("Mounting root from %s\n", mountfrom);
1307 	splx(s);
1308 
1309 	/* parse vfs name and path */
1310 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1311 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1312 	vfsname[0] = path[0] = 0;
1313 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1314 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1315 		goto done;
1316 
1317 	/* allocate a root mount */
1318 	error = vfs_rootmountalloc(vfsname, path[0] != 0 ? path : ROOTNAME,
1319 	    &mp);
1320 	if (error != 0) {
1321 		printf("Can't allocate root mount for filesystem '%s': %d\n",
1322 		       vfsname, error);
1323 		goto done;
1324 	}
1325 
1326 	/*
1327 	 * do our best to set rootdev
1328 	 * XXX: This does not belong here!
1329 	 */
1330 	if (path[0] != '\0') {
1331 		struct cdev *diskdev;
1332 		diskdev = getdiskbyname(path);
1333 		if (diskdev != NULL)
1334 			rootdev = diskdev;
1335 		else
1336 			printf("setrootbyname failed\n");
1337 	}
1338 
1339 	/* If the root device is a type "memory disk", mount RW */
1340 	if (rootdev != NULL && rootdev->si_devsw != NULL) {
1341 		devname = devtoname(rootdev);
1342 		if (devname[0] == 'm' && devname[1] == 'd')
1343 			mp->mnt_flag &= ~MNT_RDONLY;
1344 	}
1345 
1346 	error = VFS_OMOUNT(mp, NULL, NULL, curthread);
1347 
1348 done:
1349 	if (vfsname != NULL)
1350 		free(vfsname, M_MOUNT);
1351 	if (path != NULL)
1352 		free(path, M_MOUNT);
1353 	if (error != 0) {
1354 		if (mp != NULL)
1355 			vfs_mount_destroy(mp, curthread);
1356 		printf("Root mount failed: %d\n", error);
1357 	} else {
1358 
1359 		/* register with list of mounted filesystems */
1360 		mtx_lock(&mountlist_mtx);
1361 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1362 		mtx_unlock(&mountlist_mtx);
1363 
1364 		/* sanity check system clock against root fs timestamp */
1365 		inittodr(mp->mnt_time);
1366 		vfs_unbusy(mp, curthread);
1367 		error = VFS_START(mp, 0, curthread);
1368 	}
1369 	return (error);
1370 }
1371 
1372 /*
1373  * Spin prompting on the console for a suitable root filesystem
1374  */
1375 static int
1376 vfs_mountroot_ask(void)
1377 {
1378 	char name[128];
1379 
1380 	for(;;) {
1381 		printf("\nManual root filesystem specification:\n");
1382 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1383 #if defined(__i386__) || defined(__ia64__)
1384 		printf("                       eg. ufs:da0s1a\n");
1385 #else
1386 		printf("                       eg. ufs:/dev/da0a\n");
1387 #endif
1388 		printf("  ?                  List valid disk boot devices\n");
1389 		printf("  <empty line>       Abort manual input\n");
1390 		printf("\nmountroot> ");
1391 		gets(name);
1392 		if (name[0] == '\0')
1393 			return (1);
1394 		if (name[0] == '?') {
1395 			printf("\nList of GEOM managed disk devices:\n  ");
1396 			g_dev_print();
1397 			continue;
1398 		}
1399 		if (!vfs_mountroot_try(name))
1400 			return (0);
1401 	}
1402 }
1403 
1404 /*
1405  * Local helper function for vfs_mountroot_ask.
1406  */
1407 static void
1408 gets(char *cp)
1409 {
1410 	char *lp;
1411 	int c;
1412 
1413 	lp = cp;
1414 	for (;;) {
1415 		printf("%c", c = cngetc() & 0177);
1416 		switch (c) {
1417 		case -1:
1418 		case '\n':
1419 		case '\r':
1420 			*lp++ = '\0';
1421 			return;
1422 		case '\b':
1423 		case '\177':
1424 			if (lp > cp) {
1425 				printf(" \b");
1426 				lp--;
1427 			}
1428 			continue;
1429 		case '#':
1430 			lp--;
1431 			if (lp < cp)
1432 				lp = cp;
1433 			continue;
1434 		case '@':
1435 		case 'u' & 037:
1436 			lp = cp;
1437 			printf("%c", '\n');
1438 			continue;
1439 		default:
1440 			*lp++ = c;
1441 		}
1442 	}
1443 }
1444 
1445 /*
1446  * Convert a given name to the cdev pointer of the device, which is probably
1447  * but not by definition, a disk.  Mount a DEVFS (on nothing), look the name
1448  * up, extract the cdev from the vnode and unmount it again.  Unfortunately
1449  * we cannot use the vnode directly (because we unmount the DEVFS again)
1450  * so the filesystems still have to do the bdevvp() stunt.
1451  */
1452 struct cdev *
1453 getdiskbyname(char *name)
1454 {
1455 	char *cp = name;
1456 	struct cdev *dev = NULL;
1457 	struct thread *td = curthread;
1458 	struct vfsconf *vfsp;
1459 	struct mount *mp = NULL;
1460 	struct vnode *vroot = NULL;
1461 	struct nameidata nid;
1462 	int error;
1463 
1464 	if (!bcmp(cp, "/dev/", 5))
1465 		cp += 5;
1466 
1467 	do {
1468 		vfsp = vfs_byname("devfs");
1469 		if (vfsp == NULL)
1470 			break;
1471 		error = vfs_mount_alloc(NULLVP, vfsp, "/dev", td, &mp);
1472 		if (error)
1473 			break;
1474 		mp->mnt_flag |= MNT_RDONLY;
1475 
1476 		error = VFS_MOUNT(mp, curthread);
1477 		if (error)
1478 			break;
1479 		VFS_START(mp, 0, td);
1480 		VFS_ROOT(mp, &vroot, td);
1481 		VOP_UNLOCK(vroot, 0, td);
1482 
1483 		NDINIT(&nid, LOOKUP, NOCACHE|FOLLOW,
1484 		    UIO_SYSSPACE, cp, curthread);
1485 		nid.ni_startdir = vroot;
1486 		nid.ni_pathlen = strlen(cp);
1487 		nid.ni_cnd.cn_cred = curthread->td_ucred;
1488 		nid.ni_cnd.cn_nameptr = cp;
1489 
1490 		error = lookup(&nid);
1491 		if (error)
1492 			break;
1493 		dev = vn_todev (nid.ni_vp);
1494 		NDFREE(&nid, 0);
1495 	} while (0);
1496 
1497 	if (vroot != NULL)
1498 		VFS_UNMOUNT(mp, 0, td);
1499 	if (mp != NULL)
1500 		vfs_mount_destroy(mp, td);
1501   	return (dev);
1502 }
1503 
1504 /* Show the struct cdev *for a disk specified by name */
1505 #ifdef DDB
1506 DB_SHOW_COMMAND(disk, db_getdiskbyname)
1507 {
1508 	struct cdev *dev;
1509 
1510 	if (modif[0] == '\0') {
1511 		db_error("usage: show disk/devicename");
1512 		return;
1513 	}
1514 	dev = getdiskbyname(modif);
1515 	if (dev != NULL)
1516 		db_printf("struct cdev *= %p\n", dev);
1517 	else
1518 		db_printf("No disk device matched.\n");
1519 }
1520 #endif
1521 
1522 /*
1523  * Get a mount option by its name.
1524  *
1525  * Return 0 if the option was found, ENOENT otherwise.
1526  * If len is non-NULL it will be filled with the length
1527  * of the option. If buf is non-NULL, it will be filled
1528  * with the address of the option.
1529  */
1530 int
1531 vfs_getopt(opts, name, buf, len)
1532 	struct vfsoptlist *opts;
1533 	const char *name;
1534 	void **buf;
1535 	int *len;
1536 {
1537 	struct vfsopt *opt;
1538 
1539 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1540 
1541 	TAILQ_FOREACH(opt, opts, link) {
1542 		if (strcmp(name, opt->name) == 0) {
1543 			if (len != NULL)
1544 				*len = opt->len;
1545 			if (buf != NULL)
1546 				*buf = opt->value;
1547 			return (0);
1548 		}
1549 	}
1550 	return (ENOENT);
1551 }
1552 
1553 /*
1554  * Find and copy a mount option.
1555  *
1556  * The size of the buffer has to be specified
1557  * in len, if it is not the same length as the
1558  * mount option, EINVAL is returned.
1559  * Returns ENOENT if the option is not found.
1560  */
1561 int
1562 vfs_copyopt(opts, name, dest, len)
1563 	struct vfsoptlist *opts;
1564 	const char *name;
1565 	void *dest;
1566 	int len;
1567 {
1568 	struct vfsopt *opt;
1569 
1570 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1571 
1572 	TAILQ_FOREACH(opt, opts, link) {
1573 		if (strcmp(name, opt->name) == 0) {
1574 			if (len != opt->len)
1575 				return (EINVAL);
1576 			bcopy(opt->value, dest, opt->len);
1577 			return (0);
1578 		}
1579 	}
1580 	return (ENOENT);
1581 }
1582 
1583 
1584 /*
1585  * This is a helper function for filesystems to traverse their
1586  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1587  */
1588 
1589 struct vnode *
1590 __mnt_vnode_next(struct vnode **nvp, struct mount *mp)
1591 {
1592 	struct vnode *vp;
1593 
1594 	mtx_assert(&mp->mnt_mtx, MA_OWNED);
1595 	vp = *nvp;
1596 	/* Check if we are done */
1597 	if (vp == NULL)
1598 		return (NULL);
1599 	/* If our next vnode is no longer ours, start over */
1600 	if (vp->v_mount != mp)
1601 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1602 	/* Save pointer to next vnode in list */
1603 	if (vp != NULL)
1604 		*nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1605 	else
1606 		*nvp = NULL;
1607 	return (vp);
1608 }
1609