vfs_mount.c revision 212466
1/*- 2 * Copyright (c) 1999-2004 Poul-Henning Kamp 3 * Copyright (c) 1999 Michael Smith 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: head/sys/kern/vfs_mount.c 212466 2010-09-11 13:06:06Z kib $"); 39 40#include <sys/param.h> 41#include <sys/conf.h> 42#include <sys/fcntl.h> 43#include <sys/jail.h> 44#include <sys/kernel.h> 45#include <sys/libkern.h> 46#include <sys/malloc.h> 47#include <sys/mount.h> 48#include <sys/mutex.h> 49#include <sys/namei.h> 50#include <sys/priv.h> 51#include <sys/proc.h> 52#include <sys/filedesc.h> 53#include <sys/reboot.h> 54#include <sys/syscallsubr.h> 55#include <sys/sysproto.h> 56#include <sys/sx.h> 57#include <sys/sysctl.h> 58#include <sys/sysent.h> 59#include <sys/systm.h> 60#include <sys/vnode.h> 61#include <vm/uma.h> 62 63#include <geom/geom.h> 64 65#include <machine/stdarg.h> 66 67#include <security/audit/audit.h> 68#include <security/mac/mac_framework.h> 69 70#include "opt_rootdevname.h" 71 72#define ROOTNAME "root_device" 73#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 74 75static void set_rootvnode(void); 76static int vfs_domount(struct thread *td, const char *fstype, 77 char *fspath, int fsflags, void *fsdata); 78static int vfs_mountroot_ask(void); 79static int vfs_mountroot_try(const char *mountfrom, const char *options); 80static void free_mntarg(struct mntarg *ma); 81 82static int usermount = 0; 83SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 84 "Unprivileged users may mount and unmount file systems"); 85 86MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 87MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 88static uma_zone_t mount_zone; 89 90/* List of mounted filesystems. */ 91struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 92 93/* For any iteration/modification of mountlist */ 94struct mtx mountlist_mtx; 95MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 96 97/* 98 * The vnode of the system's root (/ in the filesystem, without chroot 99 * active.) 100 */ 101struct vnode *rootvnode; 102 103/* 104 * The root filesystem is detailed in the kernel environment variable 105 * vfs.root.mountfrom, which is expected to be in the general format 106 * 107 * <vfsname>:[<path>][ <vfsname>:[<path>] ...] 108 * vfsname := the name of a VFS known to the kernel and capable 109 * of being mounted as root 110 * path := disk device name or other data used by the filesystem 111 * to locate its physical store 112 * 113 * If the environment variable vfs.root.mountfrom is a space separated list, 114 * each list element is tried in turn and the root filesystem will be mounted 115 * from the first one that suceeds. 116 * 117 * The environment variable vfs.root.mountfrom.options is a comma delimited 118 * set of string mount options. These mount options must be parseable 119 * by nmount() in the kernel. 120 */ 121 122/* 123 * Global opts, taken by all filesystems 124 */ 125static const char *global_opts[] = { 126 "errmsg", 127 "fstype", 128 "fspath", 129 "ro", 130 "rw", 131 "nosuid", 132 "noexec", 133 NULL 134}; 135 136/* 137 * The root specifiers we will try if RB_CDROM is specified. 138 */ 139static char *cdrom_rootdevnames[] = { 140 "cd9660:cd0", 141 "cd9660:acd0", 142 NULL 143}; 144 145/* legacy find-root code */ 146char *rootdevnames[2] = {NULL, NULL}; 147#ifndef ROOTDEVNAME 148# define ROOTDEVNAME NULL 149#endif 150static const char *ctrootdevname = ROOTDEVNAME; 151 152/* 153 * --------------------------------------------------------------------- 154 * Functions for building and sanitizing the mount options 155 */ 156 157/* Remove one mount option. */ 158static void 159vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 160{ 161 162 TAILQ_REMOVE(opts, opt, link); 163 free(opt->name, M_MOUNT); 164 if (opt->value != NULL) 165 free(opt->value, M_MOUNT); 166 free(opt, M_MOUNT); 167} 168 169/* Release all resources related to the mount options. */ 170void 171vfs_freeopts(struct vfsoptlist *opts) 172{ 173 struct vfsopt *opt; 174 175 while (!TAILQ_EMPTY(opts)) { 176 opt = TAILQ_FIRST(opts); 177 vfs_freeopt(opts, opt); 178 } 179 free(opts, M_MOUNT); 180} 181 182void 183vfs_deleteopt(struct vfsoptlist *opts, const char *name) 184{ 185 struct vfsopt *opt, *temp; 186 187 if (opts == NULL) 188 return; 189 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 190 if (strcmp(opt->name, name) == 0) 191 vfs_freeopt(opts, opt); 192 } 193} 194 195/* 196 * Check if options are equal (with or without the "no" prefix). 197 */ 198static int 199vfs_equalopts(const char *opt1, const char *opt2) 200{ 201 char *p; 202 203 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 204 if (strcmp(opt1, opt2) == 0) 205 return (1); 206 /* "noopt" vs. "opt" */ 207 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 208 return (1); 209 /* "opt" vs. "noopt" */ 210 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 211 return (1); 212 while ((p = strchr(opt1, '.')) != NULL && 213 !strncmp(opt1, opt2, ++p - opt1)) { 214 opt2 += p - opt1; 215 opt1 = p; 216 /* "foo.noopt" vs. "foo.opt" */ 217 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 218 return (1); 219 /* "foo.opt" vs. "foo.noopt" */ 220 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 221 return (1); 222 } 223 return (0); 224} 225 226/* 227 * If a mount option is specified several times, 228 * (with or without the "no" prefix) only keep 229 * the last occurence of it. 230 */ 231static void 232vfs_sanitizeopts(struct vfsoptlist *opts) 233{ 234 struct vfsopt *opt, *opt2, *tmp; 235 236 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 237 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 238 while (opt2 != NULL) { 239 if (vfs_equalopts(opt->name, opt2->name)) { 240 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 241 vfs_freeopt(opts, opt2); 242 opt2 = tmp; 243 } else { 244 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 245 } 246 } 247 } 248} 249 250/* 251 * Build a linked list of mount options from a struct uio. 252 */ 253int 254vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 255{ 256 struct vfsoptlist *opts; 257 struct vfsopt *opt; 258 size_t memused, namelen, optlen; 259 unsigned int i, iovcnt; 260 int error; 261 262 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 263 TAILQ_INIT(opts); 264 memused = 0; 265 iovcnt = auio->uio_iovcnt; 266 for (i = 0; i < iovcnt; i += 2) { 267 namelen = auio->uio_iov[i].iov_len; 268 optlen = auio->uio_iov[i + 1].iov_len; 269 memused += sizeof(struct vfsopt) + optlen + namelen; 270 /* 271 * Avoid consuming too much memory, and attempts to overflow 272 * memused. 273 */ 274 if (memused > VFS_MOUNTARG_SIZE_MAX || 275 optlen > VFS_MOUNTARG_SIZE_MAX || 276 namelen > VFS_MOUNTARG_SIZE_MAX) { 277 error = EINVAL; 278 goto bad; 279 } 280 281 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 282 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 283 opt->value = NULL; 284 opt->len = 0; 285 opt->pos = i / 2; 286 opt->seen = 0; 287 288 /* 289 * Do this early, so jumps to "bad" will free the current 290 * option. 291 */ 292 TAILQ_INSERT_TAIL(opts, opt, link); 293 294 if (auio->uio_segflg == UIO_SYSSPACE) { 295 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 296 } else { 297 error = copyin(auio->uio_iov[i].iov_base, opt->name, 298 namelen); 299 if (error) 300 goto bad; 301 } 302 /* Ensure names are null-terminated strings. */ 303 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 304 error = EINVAL; 305 goto bad; 306 } 307 if (optlen != 0) { 308 opt->len = optlen; 309 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 310 if (auio->uio_segflg == UIO_SYSSPACE) { 311 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 312 optlen); 313 } else { 314 error = copyin(auio->uio_iov[i + 1].iov_base, 315 opt->value, optlen); 316 if (error) 317 goto bad; 318 } 319 } 320 } 321 vfs_sanitizeopts(opts); 322 *options = opts; 323 return (0); 324bad: 325 vfs_freeopts(opts); 326 return (error); 327} 328 329/* 330 * Merge the old mount options with the new ones passed 331 * in the MNT_UPDATE case. 332 * 333 * XXX This function will keep a "nofoo" option in the 334 * new options if there is no matching "foo" option 335 * to be cancelled in the old options. This is a bug 336 * if the option's canonical name is "foo". E.g., "noro" 337 * shouldn't end up in the mount point's active options, 338 * but it can. 339 */ 340static void 341vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 342{ 343 struct vfsopt *opt, *opt2, *new; 344 345 TAILQ_FOREACH(opt, opts, link) { 346 /* 347 * Check that this option hasn't been redefined 348 * nor cancelled with a "no" mount option. 349 */ 350 opt2 = TAILQ_FIRST(toopts); 351 while (opt2 != NULL) { 352 if (strcmp(opt2->name, opt->name) == 0) 353 goto next; 354 if (strncmp(opt2->name, "no", 2) == 0 && 355 strcmp(opt2->name + 2, opt->name) == 0) { 356 vfs_freeopt(toopts, opt2); 357 goto next; 358 } 359 opt2 = TAILQ_NEXT(opt2, link); 360 } 361 /* We want this option, duplicate it. */ 362 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 363 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 364 strcpy(new->name, opt->name); 365 if (opt->len != 0) { 366 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 367 bcopy(opt->value, new->value, opt->len); 368 } else { 369 new->value = NULL; 370 } 371 new->len = opt->len; 372 new->seen = opt->seen; 373 TAILQ_INSERT_TAIL(toopts, new, link); 374next: 375 continue; 376 } 377} 378 379/* 380 * Mount a filesystem. 381 */ 382int 383nmount(td, uap) 384 struct thread *td; 385 struct nmount_args /* { 386 struct iovec *iovp; 387 unsigned int iovcnt; 388 int flags; 389 } */ *uap; 390{ 391 struct uio *auio; 392 int error; 393 u_int iovcnt; 394 395 AUDIT_ARG_FFLAGS(uap->flags); 396 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 397 uap->iovp, uap->iovcnt, uap->flags); 398 399 /* 400 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 401 * userspace to set this flag, but we must filter it out if we want 402 * MNT_UPDATE on the root file system to work. 403 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 404 */ 405 uap->flags &= ~MNT_ROOTFS; 406 407 iovcnt = uap->iovcnt; 408 /* 409 * Check that we have an even number of iovec's 410 * and that we have at least two options. 411 */ 412 if ((iovcnt & 1) || (iovcnt < 4)) { 413 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 414 uap->iovcnt); 415 return (EINVAL); 416 } 417 418 error = copyinuio(uap->iovp, iovcnt, &auio); 419 if (error) { 420 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 421 __func__, error); 422 return (error); 423 } 424 error = vfs_donmount(td, uap->flags, auio); 425 426 free(auio, M_IOV); 427 return (error); 428} 429 430/* 431 * --------------------------------------------------------------------- 432 * Various utility functions 433 */ 434 435void 436vfs_ref(struct mount *mp) 437{ 438 439 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 440 MNT_ILOCK(mp); 441 MNT_REF(mp); 442 MNT_IUNLOCK(mp); 443} 444 445void 446vfs_rel(struct mount *mp) 447{ 448 449 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 450 MNT_ILOCK(mp); 451 MNT_REL(mp); 452 MNT_IUNLOCK(mp); 453} 454 455static int 456mount_init(void *mem, int size, int flags) 457{ 458 struct mount *mp; 459 460 mp = (struct mount *)mem; 461 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 462 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 463 return (0); 464} 465 466static void 467mount_fini(void *mem, int size) 468{ 469 struct mount *mp; 470 471 mp = (struct mount *)mem; 472 lockdestroy(&mp->mnt_explock); 473 mtx_destroy(&mp->mnt_mtx); 474} 475 476/* 477 * Allocate and initialize the mount point struct. 478 */ 479struct mount * 480vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 481 struct ucred *cred) 482{ 483 struct mount *mp; 484 485 mp = uma_zalloc(mount_zone, M_WAITOK); 486 bzero(&mp->mnt_startzero, 487 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 488 TAILQ_INIT(&mp->mnt_nvnodelist); 489 mp->mnt_nvnodelistsize = 0; 490 mp->mnt_ref = 0; 491 (void) vfs_busy(mp, MBF_NOWAIT); 492 mp->mnt_op = vfsp->vfc_vfsops; 493 mp->mnt_vfc = vfsp; 494 vfsp->vfc_refcount++; /* XXX Unlocked */ 495 mp->mnt_stat.f_type = vfsp->vfc_typenum; 496 mp->mnt_gen++; 497 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 498 mp->mnt_vnodecovered = vp; 499 mp->mnt_cred = crdup(cred); 500 mp->mnt_stat.f_owner = cred->cr_uid; 501 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 502 mp->mnt_iosize_max = DFLTPHYS; 503#ifdef MAC 504 mac_mount_init(mp); 505 mac_mount_create(cred, mp); 506#endif 507 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 508 return (mp); 509} 510 511/* 512 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 513 */ 514void 515vfs_mount_destroy(struct mount *mp) 516{ 517 518 MNT_ILOCK(mp); 519 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 520 if (mp->mnt_kern_flag & MNTK_MWAIT) { 521 mp->mnt_kern_flag &= ~MNTK_MWAIT; 522 wakeup(mp); 523 } 524 while (mp->mnt_ref) 525 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 526 KASSERT(mp->mnt_ref == 0, 527 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 528 __FILE__, __LINE__)); 529 if (mp->mnt_writeopcount != 0) 530 panic("vfs_mount_destroy: nonzero writeopcount"); 531 if (mp->mnt_secondary_writes != 0) 532 panic("vfs_mount_destroy: nonzero secondary_writes"); 533 mp->mnt_vfc->vfc_refcount--; 534 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 535 struct vnode *vp; 536 537 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 538 vprint("", vp); 539 panic("unmount: dangling vnode"); 540 } 541 if (mp->mnt_nvnodelistsize != 0) 542 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 543 if (mp->mnt_lockref != 0) 544 panic("vfs_mount_destroy: nonzero lock refcount"); 545 MNT_IUNLOCK(mp); 546#ifdef MAC 547 mac_mount_destroy(mp); 548#endif 549 if (mp->mnt_opt != NULL) 550 vfs_freeopts(mp->mnt_opt); 551 crfree(mp->mnt_cred); 552 uma_zfree(mount_zone, mp); 553} 554 555int 556vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 557{ 558 struct vfsoptlist *optlist; 559 struct vfsopt *opt, *noro_opt, *tmp_opt; 560 char *fstype, *fspath, *errmsg; 561 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 562 int has_rw, has_noro; 563 564 errmsg = fspath = NULL; 565 errmsg_len = has_noro = has_rw = fspathlen = 0; 566 errmsg_pos = -1; 567 568 error = vfs_buildopts(fsoptions, &optlist); 569 if (error) 570 return (error); 571 572 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 573 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 574 575 /* 576 * We need these two options before the others, 577 * and they are mandatory for any filesystem. 578 * Ensure they are NUL terminated as well. 579 */ 580 fstypelen = 0; 581 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 582 if (error || fstype[fstypelen - 1] != '\0') { 583 error = EINVAL; 584 if (errmsg != NULL) 585 strncpy(errmsg, "Invalid fstype", errmsg_len); 586 goto bail; 587 } 588 fspathlen = 0; 589 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 590 if (error || fspath[fspathlen - 1] != '\0') { 591 error = EINVAL; 592 if (errmsg != NULL) 593 strncpy(errmsg, "Invalid fspath", errmsg_len); 594 goto bail; 595 } 596 597 /* 598 * We need to see if we have the "update" option 599 * before we call vfs_domount(), since vfs_domount() has special 600 * logic based on MNT_UPDATE. This is very important 601 * when we want to update the root filesystem. 602 */ 603 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 604 if (strcmp(opt->name, "update") == 0) { 605 fsflags |= MNT_UPDATE; 606 vfs_freeopt(optlist, opt); 607 } 608 else if (strcmp(opt->name, "async") == 0) 609 fsflags |= MNT_ASYNC; 610 else if (strcmp(opt->name, "force") == 0) { 611 fsflags |= MNT_FORCE; 612 vfs_freeopt(optlist, opt); 613 } 614 else if (strcmp(opt->name, "reload") == 0) { 615 fsflags |= MNT_RELOAD; 616 vfs_freeopt(optlist, opt); 617 } 618 else if (strcmp(opt->name, "multilabel") == 0) 619 fsflags |= MNT_MULTILABEL; 620 else if (strcmp(opt->name, "noasync") == 0) 621 fsflags &= ~MNT_ASYNC; 622 else if (strcmp(opt->name, "noatime") == 0) 623 fsflags |= MNT_NOATIME; 624 else if (strcmp(opt->name, "atime") == 0) { 625 free(opt->name, M_MOUNT); 626 opt->name = strdup("nonoatime", M_MOUNT); 627 } 628 else if (strcmp(opt->name, "noclusterr") == 0) 629 fsflags |= MNT_NOCLUSTERR; 630 else if (strcmp(opt->name, "clusterr") == 0) { 631 free(opt->name, M_MOUNT); 632 opt->name = strdup("nonoclusterr", M_MOUNT); 633 } 634 else if (strcmp(opt->name, "noclusterw") == 0) 635 fsflags |= MNT_NOCLUSTERW; 636 else if (strcmp(opt->name, "clusterw") == 0) { 637 free(opt->name, M_MOUNT); 638 opt->name = strdup("nonoclusterw", M_MOUNT); 639 } 640 else if (strcmp(opt->name, "noexec") == 0) 641 fsflags |= MNT_NOEXEC; 642 else if (strcmp(opt->name, "exec") == 0) { 643 free(opt->name, M_MOUNT); 644 opt->name = strdup("nonoexec", M_MOUNT); 645 } 646 else if (strcmp(opt->name, "nosuid") == 0) 647 fsflags |= MNT_NOSUID; 648 else if (strcmp(opt->name, "suid") == 0) { 649 free(opt->name, M_MOUNT); 650 opt->name = strdup("nonosuid", M_MOUNT); 651 } 652 else if (strcmp(opt->name, "nosymfollow") == 0) 653 fsflags |= MNT_NOSYMFOLLOW; 654 else if (strcmp(opt->name, "symfollow") == 0) { 655 free(opt->name, M_MOUNT); 656 opt->name = strdup("nonosymfollow", M_MOUNT); 657 } 658 else if (strcmp(opt->name, "noro") == 0) { 659 fsflags &= ~MNT_RDONLY; 660 has_noro = 1; 661 } 662 else if (strcmp(opt->name, "rw") == 0) { 663 fsflags &= ~MNT_RDONLY; 664 has_rw = 1; 665 } 666 else if (strcmp(opt->name, "ro") == 0) 667 fsflags |= MNT_RDONLY; 668 else if (strcmp(opt->name, "rdonly") == 0) { 669 free(opt->name, M_MOUNT); 670 opt->name = strdup("ro", M_MOUNT); 671 fsflags |= MNT_RDONLY; 672 } 673 else if (strcmp(opt->name, "suiddir") == 0) 674 fsflags |= MNT_SUIDDIR; 675 else if (strcmp(opt->name, "sync") == 0) 676 fsflags |= MNT_SYNCHRONOUS; 677 else if (strcmp(opt->name, "union") == 0) 678 fsflags |= MNT_UNION; 679 } 680 681 /* 682 * If "rw" was specified as a mount option, and we 683 * are trying to update a mount-point from "ro" to "rw", 684 * we need a mount option "noro", since in vfs_mergeopts(), 685 * "noro" will cancel "ro", but "rw" will not do anything. 686 */ 687 if (has_rw && !has_noro) { 688 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 689 noro_opt->name = strdup("noro", M_MOUNT); 690 noro_opt->value = NULL; 691 noro_opt->len = 0; 692 noro_opt->pos = -1; 693 noro_opt->seen = 1; 694 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 695 } 696 697 /* 698 * Be ultra-paranoid about making sure the type and fspath 699 * variables will fit in our mp buffers, including the 700 * terminating NUL. 701 */ 702 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 703 error = ENAMETOOLONG; 704 goto bail; 705 } 706 707 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 708bail: 709 /* copyout the errmsg */ 710 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 711 && errmsg_len > 0 && errmsg != NULL) { 712 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 713 bcopy(errmsg, 714 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 715 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 716 } else { 717 copyout(errmsg, 718 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 719 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 720 } 721 } 722 723 if (error != 0) 724 vfs_freeopts(optlist); 725 return (error); 726} 727 728/* 729 * Old mount API. 730 */ 731#ifndef _SYS_SYSPROTO_H_ 732struct mount_args { 733 char *type; 734 char *path; 735 int flags; 736 caddr_t data; 737}; 738#endif 739/* ARGSUSED */ 740int 741mount(td, uap) 742 struct thread *td; 743 struct mount_args /* { 744 char *type; 745 char *path; 746 int flags; 747 caddr_t data; 748 } */ *uap; 749{ 750 char *fstype; 751 struct vfsconf *vfsp = NULL; 752 struct mntarg *ma = NULL; 753 int error; 754 755 AUDIT_ARG_FFLAGS(uap->flags); 756 757 /* 758 * Filter out MNT_ROOTFS. We do not want clients of mount() in 759 * userspace to set this flag, but we must filter it out if we want 760 * MNT_UPDATE on the root file system to work. 761 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 762 */ 763 uap->flags &= ~MNT_ROOTFS; 764 765 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 766 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 767 if (error) { 768 free(fstype, M_TEMP); 769 return (error); 770 } 771 772 AUDIT_ARG_TEXT(fstype); 773 mtx_lock(&Giant); 774 vfsp = vfs_byname_kld(fstype, td, &error); 775 free(fstype, M_TEMP); 776 if (vfsp == NULL) { 777 mtx_unlock(&Giant); 778 return (ENOENT); 779 } 780 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 781 mtx_unlock(&Giant); 782 return (EOPNOTSUPP); 783 } 784 785 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 786 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 787 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 788 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 789 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 790 791 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags); 792 mtx_unlock(&Giant); 793 return (error); 794} 795 796/* 797 * vfs_domount_first(): first file system mount (not update) 798 */ 799static int 800vfs_domount_first( 801 struct thread *td, /* Calling thread. */ 802 struct vfsconf *vfsp, /* File system type. */ 803 char *fspath, /* Mount path. */ 804 struct vnode *vp, /* Vnode to be covered. */ 805 int fsflags, /* Flags common to all filesystems. */ 806 void *fsdata /* Options local to the filesystem. */ 807 ) 808{ 809 struct vattr va; 810 struct mount *mp; 811 struct vnode *newdp; 812 int error; 813 814 mtx_assert(&Giant, MA_OWNED); 815 ASSERT_VOP_ELOCKED(vp, __func__); 816 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 817 818 /* 819 * If the user is not root, ensure that they own the directory 820 * onto which we are attempting to mount. 821 */ 822 error = VOP_GETATTR(vp, &va, td->td_ucred); 823 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 824 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0); 825 if (error == 0) 826 error = vinvalbuf(vp, V_SAVE, 0, 0); 827 if (error == 0 && vp->v_type != VDIR) 828 error = ENOTDIR; 829 if (error == 0) { 830 VI_LOCK(vp); 831 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 832 vp->v_iflag |= VI_MOUNT; 833 else 834 error = EBUSY; 835 VI_UNLOCK(vp); 836 } 837 if (error != 0) { 838 vput(vp); 839 return (error); 840 } 841 VOP_UNLOCK(vp, 0); 842 843 /* Allocate and initialize the filesystem. */ 844 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 845 /* XXXMAC: pass to vfs_mount_alloc? */ 846 mp->mnt_optnew = fsdata; 847 /* Set the mount level flags. */ 848 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 849 850 /* 851 * Mount the filesystem. 852 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 853 * get. No freeing of cn_pnbuf. 854 */ 855 error = VFS_MOUNT(mp); 856 if (error != 0) { 857 vfs_unbusy(mp); 858 vfs_mount_destroy(mp); 859 VI_LOCK(vp); 860 vp->v_iflag &= ~VI_MOUNT; 861 VI_UNLOCK(vp); 862 vrele(vp); 863 return (error); 864 } 865 866 if (mp->mnt_opt != NULL) 867 vfs_freeopts(mp->mnt_opt); 868 mp->mnt_opt = mp->mnt_optnew; 869 (void)VFS_STATFS(mp, &mp->mnt_stat); 870 871 /* 872 * Prevent external consumers of mount options from reading mnt_optnew. 873 */ 874 mp->mnt_optnew = NULL; 875 876 MNT_ILOCK(mp); 877 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 878 mp->mnt_kern_flag |= MNTK_ASYNC; 879 else 880 mp->mnt_kern_flag &= ~MNTK_ASYNC; 881 MNT_IUNLOCK(mp); 882 883 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 884 cache_purge(vp); 885 VI_LOCK(vp); 886 vp->v_iflag &= ~VI_MOUNT; 887 VI_UNLOCK(vp); 888 vp->v_mountedhere = mp; 889 /* Place the new filesystem at the end of the mount list. */ 890 mtx_lock(&mountlist_mtx); 891 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 892 mtx_unlock(&mountlist_mtx); 893 vfs_event_signal(NULL, VQ_MOUNT, 0); 894 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 895 panic("mount: lost mount"); 896 VOP_UNLOCK(newdp, 0); 897 VOP_UNLOCK(vp, 0); 898 mountcheckdirs(vp, newdp); 899 vrele(newdp); 900 if ((mp->mnt_flag & MNT_RDONLY) == 0) 901 vfs_allocate_syncvnode(mp); 902 vfs_unbusy(mp); 903 return (0); 904} 905 906/* 907 * vfs_domount_update(): update of mounted file system 908 */ 909static int 910vfs_domount_update( 911 struct thread *td, /* Calling thread. */ 912 struct vnode *vp, /* Mount point vnode. */ 913 int fsflags, /* Flags common to all filesystems. */ 914 void *fsdata /* Options local to the filesystem. */ 915 ) 916{ 917 struct oexport_args oexport; 918 struct export_args export; 919 struct mount *mp; 920 int error, flag; 921 922 mtx_assert(&Giant, MA_OWNED); 923 ASSERT_VOP_ELOCKED(vp, __func__); 924 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 925 926 if ((vp->v_vflag & VV_ROOT) == 0) { 927 vput(vp); 928 return (EINVAL); 929 } 930 mp = vp->v_mount; 931 /* 932 * We only allow the filesystem to be reloaded if it 933 * is currently mounted read-only. 934 */ 935 flag = mp->mnt_flag; 936 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 937 vput(vp); 938 return (EOPNOTSUPP); /* Needs translation */ 939 } 940 /* 941 * Only privileged root, or (if MNT_USER is set) the user that 942 * did the original mount is permitted to update it. 943 */ 944 error = vfs_suser(mp, td); 945 if (error != 0) { 946 vput(vp); 947 return (error); 948 } 949 if (vfs_busy(mp, MBF_NOWAIT)) { 950 vput(vp); 951 return (EBUSY); 952 } 953 VI_LOCK(vp); 954 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 955 VI_UNLOCK(vp); 956 vfs_unbusy(mp); 957 vput(vp); 958 return (EBUSY); 959 } 960 vp->v_iflag |= VI_MOUNT; 961 VI_UNLOCK(vp); 962 VOP_UNLOCK(vp, 0); 963 964 MNT_ILOCK(mp); 965 mp->mnt_flag &= ~MNT_UPDATEMASK; 966 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 967 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 968 if ((mp->mnt_flag & MNT_ASYNC) == 0) 969 mp->mnt_kern_flag &= ~MNTK_ASYNC; 970 MNT_IUNLOCK(mp); 971 mp->mnt_optnew = fsdata; 972 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 973 974 /* 975 * Mount the filesystem. 976 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 977 * get. No freeing of cn_pnbuf. 978 */ 979 error = VFS_MOUNT(mp); 980 981 if (error == 0) { 982 /* Process the export option. */ 983 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 984 sizeof(export)) == 0) { 985 error = vfs_export(mp, &export); 986 } else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 987 sizeof(oexport)) == 0) { 988 export.ex_flags = oexport.ex_flags; 989 export.ex_root = oexport.ex_root; 990 export.ex_anon = oexport.ex_anon; 991 export.ex_addr = oexport.ex_addr; 992 export.ex_addrlen = oexport.ex_addrlen; 993 export.ex_mask = oexport.ex_mask; 994 export.ex_masklen = oexport.ex_masklen; 995 export.ex_indexfile = oexport.ex_indexfile; 996 export.ex_numsecflavors = 0; 997 error = vfs_export(mp, &export); 998 } 999 } 1000 1001 MNT_ILOCK(mp); 1002 if (error == 0) { 1003 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 1004 MNT_SNAPSHOT); 1005 } else { 1006 /* 1007 * If we fail, restore old mount flags. MNT_QUOTA is special, 1008 * because it is not part of MNT_UPDATEMASK, but it could have 1009 * changed in the meantime if quotactl(2) was called. 1010 * All in all we want current value of MNT_QUOTA, not the old 1011 * one. 1012 */ 1013 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 1014 } 1015 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1016 mp->mnt_kern_flag |= MNTK_ASYNC; 1017 else 1018 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1019 MNT_IUNLOCK(mp); 1020 1021 if (error != 0) 1022 goto end; 1023 1024 if (mp->mnt_opt != NULL) 1025 vfs_freeopts(mp->mnt_opt); 1026 mp->mnt_opt = mp->mnt_optnew; 1027 (void)VFS_STATFS(mp, &mp->mnt_stat); 1028 /* 1029 * Prevent external consumers of mount options from reading 1030 * mnt_optnew. 1031 */ 1032 mp->mnt_optnew = NULL; 1033 1034 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1035 vfs_allocate_syncvnode(mp); 1036 else 1037 vfs_deallocate_syncvnode(mp); 1038end: 1039 vfs_unbusy(mp); 1040 VI_LOCK(vp); 1041 vp->v_iflag &= ~VI_MOUNT; 1042 VI_UNLOCK(vp); 1043 vrele(vp); 1044 return (error); 1045} 1046 1047/* 1048 * vfs_domount(): actually attempt a filesystem mount. 1049 */ 1050static int 1051vfs_domount( 1052 struct thread *td, /* Calling thread. */ 1053 const char *fstype, /* Filesystem type. */ 1054 char *fspath, /* Mount path. */ 1055 int fsflags, /* Flags common to all filesystems. */ 1056 void *fsdata /* Options local to the filesystem. */ 1057 ) 1058{ 1059 struct vfsconf *vfsp; 1060 struct nameidata nd; 1061 struct vnode *vp; 1062 int error; 1063 1064 /* 1065 * Be ultra-paranoid about making sure the type and fspath 1066 * variables will fit in our mp buffers, including the 1067 * terminating NUL. 1068 */ 1069 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1070 return (ENAMETOOLONG); 1071 1072 if (jailed(td->td_ucred) || usermount == 0) { 1073 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1074 return (error); 1075 } 1076 1077 /* 1078 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1079 */ 1080 if (fsflags & MNT_EXPORTED) { 1081 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1082 if (error) 1083 return (error); 1084 } 1085 if (fsflags & MNT_SUIDDIR) { 1086 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1087 if (error) 1088 return (error); 1089 } 1090 /* 1091 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1092 */ 1093 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1094 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1095 fsflags |= MNT_NOSUID | MNT_USER; 1096 } 1097 1098 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1099 vfsp = NULL; 1100 if ((fsflags & MNT_UPDATE) == 0) { 1101 /* Don't try to load KLDs if we're mounting the root. */ 1102 if (fsflags & MNT_ROOTFS) 1103 vfsp = vfs_byname(fstype); 1104 else 1105 vfsp = vfs_byname_kld(fstype, td, &error); 1106 if (vfsp == NULL) 1107 return (ENODEV); 1108 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 1109 return (EPERM); 1110 } 1111 1112 /* 1113 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1114 */ 1115 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | MPSAFE | AUDITVNODE1, 1116 UIO_SYSSPACE, fspath, td); 1117 error = namei(&nd); 1118 if (error != 0) 1119 return (error); 1120 if (!NDHASGIANT(&nd)) 1121 mtx_lock(&Giant); 1122 NDFREE(&nd, NDF_ONLY_PNBUF); 1123 vp = nd.ni_vp; 1124 if ((fsflags & MNT_UPDATE) == 0) { 1125 error = vfs_domount_first(td, vfsp, fspath, vp, fsflags, 1126 fsdata); 1127 } else { 1128 error = vfs_domount_update(td, vp, fsflags, fsdata); 1129 } 1130 mtx_unlock(&Giant); 1131 1132 ASSERT_VI_UNLOCKED(vp, __func__); 1133 ASSERT_VOP_UNLOCKED(vp, __func__); 1134 1135 return (error); 1136} 1137 1138/* 1139 * Unmount a filesystem. 1140 * 1141 * Note: unmount takes a path to the vnode mounted on as argument, not 1142 * special file (as before). 1143 */ 1144#ifndef _SYS_SYSPROTO_H_ 1145struct unmount_args { 1146 char *path; 1147 int flags; 1148}; 1149#endif 1150/* ARGSUSED */ 1151int 1152unmount(td, uap) 1153 struct thread *td; 1154 register struct unmount_args /* { 1155 char *path; 1156 int flags; 1157 } */ *uap; 1158{ 1159 struct mount *mp; 1160 char *pathbuf; 1161 int error, id0, id1; 1162 1163 AUDIT_ARG_VALUE(uap->flags); 1164 if (jailed(td->td_ucred) || usermount == 0) { 1165 error = priv_check(td, PRIV_VFS_UNMOUNT); 1166 if (error) 1167 return (error); 1168 } 1169 1170 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1171 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1172 if (error) { 1173 free(pathbuf, M_TEMP); 1174 return (error); 1175 } 1176 mtx_lock(&Giant); 1177 if (uap->flags & MNT_BYFSID) { 1178 AUDIT_ARG_TEXT(pathbuf); 1179 /* Decode the filesystem ID. */ 1180 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1181 mtx_unlock(&Giant); 1182 free(pathbuf, M_TEMP); 1183 return (EINVAL); 1184 } 1185 1186 mtx_lock(&mountlist_mtx); 1187 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1188 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1189 mp->mnt_stat.f_fsid.val[1] == id1) 1190 break; 1191 } 1192 mtx_unlock(&mountlist_mtx); 1193 } else { 1194 AUDIT_ARG_UPATH1(td, pathbuf); 1195 mtx_lock(&mountlist_mtx); 1196 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1197 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1198 break; 1199 } 1200 mtx_unlock(&mountlist_mtx); 1201 } 1202 free(pathbuf, M_TEMP); 1203 if (mp == NULL) { 1204 /* 1205 * Previously we returned ENOENT for a nonexistent path and 1206 * EINVAL for a non-mountpoint. We cannot tell these apart 1207 * now, so in the !MNT_BYFSID case return the more likely 1208 * EINVAL for compatibility. 1209 */ 1210 mtx_unlock(&Giant); 1211 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1212 } 1213 1214 /* 1215 * Don't allow unmounting the root filesystem. 1216 */ 1217 if (mp->mnt_flag & MNT_ROOTFS) { 1218 mtx_unlock(&Giant); 1219 return (EINVAL); 1220 } 1221 error = dounmount(mp, uap->flags, td); 1222 mtx_unlock(&Giant); 1223 return (error); 1224} 1225 1226/* 1227 * Do the actual filesystem unmount. 1228 */ 1229int 1230dounmount(mp, flags, td) 1231 struct mount *mp; 1232 int flags; 1233 struct thread *td; 1234{ 1235 struct vnode *coveredvp, *fsrootvp; 1236 int error; 1237 int async_flag; 1238 int mnt_gen_r; 1239 1240 mtx_assert(&Giant, MA_OWNED); 1241 1242 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1243 mnt_gen_r = mp->mnt_gen; 1244 VI_LOCK(coveredvp); 1245 vholdl(coveredvp); 1246 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1247 vdrop(coveredvp); 1248 /* 1249 * Check for mp being unmounted while waiting for the 1250 * covered vnode lock. 1251 */ 1252 if (coveredvp->v_mountedhere != mp || 1253 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1254 VOP_UNLOCK(coveredvp, 0); 1255 return (EBUSY); 1256 } 1257 } 1258 /* 1259 * Only privileged root, or (if MNT_USER is set) the user that did the 1260 * original mount is permitted to unmount this filesystem. 1261 */ 1262 error = vfs_suser(mp, td); 1263 if (error) { 1264 if (coveredvp) 1265 VOP_UNLOCK(coveredvp, 0); 1266 return (error); 1267 } 1268 1269 MNT_ILOCK(mp); 1270 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1271 MNT_IUNLOCK(mp); 1272 if (coveredvp) 1273 VOP_UNLOCK(coveredvp, 0); 1274 return (EBUSY); 1275 } 1276 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1277 /* Allow filesystems to detect that a forced unmount is in progress. */ 1278 if (flags & MNT_FORCE) 1279 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1280 error = 0; 1281 if (mp->mnt_lockref) { 1282 if ((flags & MNT_FORCE) == 0) { 1283 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1284 MNTK_UNMOUNTF); 1285 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1286 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1287 wakeup(mp); 1288 } 1289 MNT_IUNLOCK(mp); 1290 if (coveredvp) 1291 VOP_UNLOCK(coveredvp, 0); 1292 return (EBUSY); 1293 } 1294 mp->mnt_kern_flag |= MNTK_DRAINING; 1295 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1296 "mount drain", 0); 1297 } 1298 MNT_IUNLOCK(mp); 1299 KASSERT(mp->mnt_lockref == 0, 1300 ("%s: invalid lock refcount in the drain path @ %s:%d", 1301 __func__, __FILE__, __LINE__)); 1302 KASSERT(error == 0, 1303 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1304 __func__, __FILE__, __LINE__)); 1305 vn_start_write(NULL, &mp, V_WAIT); 1306 1307 if (mp->mnt_flag & MNT_EXPUBLIC) 1308 vfs_setpublicfs(NULL, NULL, NULL); 1309 1310 vfs_msync(mp, MNT_WAIT); 1311 MNT_ILOCK(mp); 1312 async_flag = mp->mnt_flag & MNT_ASYNC; 1313 mp->mnt_flag &= ~MNT_ASYNC; 1314 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1315 MNT_IUNLOCK(mp); 1316 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1317 vfs_deallocate_syncvnode(mp); 1318 /* 1319 * For forced unmounts, move process cdir/rdir refs on the fs root 1320 * vnode to the covered vnode. For non-forced unmounts we want 1321 * such references to cause an EBUSY error. 1322 */ 1323 if ((flags & MNT_FORCE) && 1324 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1325 if (mp->mnt_vnodecovered != NULL) 1326 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1327 if (fsrootvp == rootvnode) { 1328 vrele(rootvnode); 1329 rootvnode = NULL; 1330 } 1331 vput(fsrootvp); 1332 } 1333 if (((mp->mnt_flag & MNT_RDONLY) || 1334 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1335 error = VFS_UNMOUNT(mp, flags); 1336 vn_finished_write(mp); 1337 /* 1338 * If we failed to flush the dirty blocks for this mount point, 1339 * undo all the cdir/rdir and rootvnode changes we made above. 1340 * Unless we failed to do so because the device is reporting that 1341 * it doesn't exist anymore. 1342 */ 1343 if (error && error != ENXIO) { 1344 if ((flags & MNT_FORCE) && 1345 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1346 if (mp->mnt_vnodecovered != NULL) 1347 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1348 if (rootvnode == NULL) { 1349 rootvnode = fsrootvp; 1350 vref(rootvnode); 1351 } 1352 vput(fsrootvp); 1353 } 1354 MNT_ILOCK(mp); 1355 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1356 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1357 MNT_IUNLOCK(mp); 1358 vfs_allocate_syncvnode(mp); 1359 MNT_ILOCK(mp); 1360 } 1361 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1362 mp->mnt_flag |= async_flag; 1363 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1364 mp->mnt_kern_flag |= MNTK_ASYNC; 1365 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1366 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1367 wakeup(mp); 1368 } 1369 MNT_IUNLOCK(mp); 1370 if (coveredvp) 1371 VOP_UNLOCK(coveredvp, 0); 1372 return (error); 1373 } 1374 mtx_lock(&mountlist_mtx); 1375 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1376 mtx_unlock(&mountlist_mtx); 1377 if (coveredvp != NULL) { 1378 coveredvp->v_mountedhere = NULL; 1379 vput(coveredvp); 1380 } 1381 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1382 vfs_mount_destroy(mp); 1383 return (0); 1384} 1385 1386/* 1387 * --------------------------------------------------------------------- 1388 * Mounting of root filesystem 1389 * 1390 */ 1391 1392struct root_hold_token { 1393 const char *who; 1394 LIST_ENTRY(root_hold_token) list; 1395}; 1396 1397static LIST_HEAD(, root_hold_token) root_holds = 1398 LIST_HEAD_INITIALIZER(root_holds); 1399 1400static int root_mount_complete; 1401 1402/* 1403 * Hold root mount. 1404 */ 1405struct root_hold_token * 1406root_mount_hold(const char *identifier) 1407{ 1408 struct root_hold_token *h; 1409 1410 if (root_mounted()) 1411 return (NULL); 1412 1413 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1414 h->who = identifier; 1415 mtx_lock(&mountlist_mtx); 1416 LIST_INSERT_HEAD(&root_holds, h, list); 1417 mtx_unlock(&mountlist_mtx); 1418 return (h); 1419} 1420 1421/* 1422 * Release root mount. 1423 */ 1424void 1425root_mount_rel(struct root_hold_token *h) 1426{ 1427 1428 if (h == NULL) 1429 return; 1430 mtx_lock(&mountlist_mtx); 1431 LIST_REMOVE(h, list); 1432 wakeup(&root_holds); 1433 mtx_unlock(&mountlist_mtx); 1434 free(h, M_DEVBUF); 1435} 1436 1437/* 1438 * Wait for all subsystems to release root mount. 1439 */ 1440static void 1441root_mount_prepare(void) 1442{ 1443 struct root_hold_token *h; 1444 struct timeval lastfail; 1445 int curfail = 0; 1446 1447 for (;;) { 1448 DROP_GIANT(); 1449 g_waitidle(); 1450 PICKUP_GIANT(); 1451 mtx_lock(&mountlist_mtx); 1452 if (LIST_EMPTY(&root_holds)) { 1453 mtx_unlock(&mountlist_mtx); 1454 break; 1455 } 1456 if (ppsratecheck(&lastfail, &curfail, 1)) { 1457 printf("Root mount waiting for:"); 1458 LIST_FOREACH(h, &root_holds, list) 1459 printf(" %s", h->who); 1460 printf("\n"); 1461 } 1462 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1463 hz); 1464 } 1465} 1466 1467/* 1468 * Root was mounted, share the good news. 1469 */ 1470static void 1471root_mount_done(void) 1472{ 1473 1474 /* Keep prison0's root in sync with the global rootvnode. */ 1475 mtx_lock(&prison0.pr_mtx); 1476 prison0.pr_root = rootvnode; 1477 vref(prison0.pr_root); 1478 mtx_unlock(&prison0.pr_mtx); 1479 /* 1480 * Use a mutex to prevent the wakeup being missed and waiting for 1481 * an extra 1 second sleep. 1482 */ 1483 mtx_lock(&mountlist_mtx); 1484 root_mount_complete = 1; 1485 wakeup(&root_mount_complete); 1486 mtx_unlock(&mountlist_mtx); 1487} 1488 1489/* 1490 * Return true if root is already mounted. 1491 */ 1492int 1493root_mounted(void) 1494{ 1495 1496 /* No mutex is acquired here because int stores are atomic. */ 1497 return (root_mount_complete); 1498} 1499 1500/* 1501 * Wait until root is mounted. 1502 */ 1503void 1504root_mount_wait(void) 1505{ 1506 1507 /* 1508 * Panic on an obvious deadlock - the function can't be called from 1509 * a thread which is doing the whole SYSINIT stuff. 1510 */ 1511 KASSERT(curthread->td_proc->p_pid != 0, 1512 ("root_mount_wait: cannot be called from the swapper thread")); 1513 mtx_lock(&mountlist_mtx); 1514 while (!root_mount_complete) { 1515 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1516 hz); 1517 } 1518 mtx_unlock(&mountlist_mtx); 1519} 1520 1521static void 1522set_rootvnode() 1523{ 1524 struct proc *p; 1525 1526 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode)) 1527 panic("Cannot find root vnode"); 1528 1529 VOP_UNLOCK(rootvnode, 0); 1530 1531 p = curthread->td_proc; 1532 FILEDESC_XLOCK(p->p_fd); 1533 1534 if (p->p_fd->fd_cdir != NULL) 1535 vrele(p->p_fd->fd_cdir); 1536 p->p_fd->fd_cdir = rootvnode; 1537 VREF(rootvnode); 1538 1539 if (p->p_fd->fd_rdir != NULL) 1540 vrele(p->p_fd->fd_rdir); 1541 p->p_fd->fd_rdir = rootvnode; 1542 VREF(rootvnode); 1543 1544 FILEDESC_XUNLOCK(p->p_fd); 1545 1546 EVENTHANDLER_INVOKE(mountroot); 1547} 1548 1549/* 1550 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1551 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1552 */ 1553 1554static void 1555devfs_first(void) 1556{ 1557 struct thread *td = curthread; 1558 struct vfsoptlist *opts; 1559 struct vfsconf *vfsp; 1560 struct mount *mp = NULL; 1561 int error; 1562 1563 vfsp = vfs_byname("devfs"); 1564 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1565 if (vfsp == NULL) 1566 return; 1567 1568 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 1569 1570 error = VFS_MOUNT(mp); 1571 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1572 if (error) 1573 return; 1574 1575 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1576 TAILQ_INIT(opts); 1577 mp->mnt_opt = opts; 1578 1579 mtx_lock(&mountlist_mtx); 1580 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1581 mtx_unlock(&mountlist_mtx); 1582 1583 set_rootvnode(); 1584 1585 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1586 if (error) 1587 printf("kern_symlink /dev -> / returns %d\n", error); 1588} 1589 1590/* 1591 * Surgically move our devfs to be mounted on /dev. 1592 */ 1593 1594static void 1595devfs_fixup(struct thread *td) 1596{ 1597 struct nameidata nd; 1598 struct vnode *vp, *dvp; 1599 struct mount *mp; 1600 int error; 1601 1602 /* Remove our devfs mount from the mountlist and purge the cache */ 1603 mtx_lock(&mountlist_mtx); 1604 mp = TAILQ_FIRST(&mountlist); 1605 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1606 mtx_unlock(&mountlist_mtx); 1607 cache_purgevfs(mp); 1608 1609 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp); 1610 VI_LOCK(dvp); 1611 dvp->v_iflag &= ~VI_MOUNT; 1612 VI_UNLOCK(dvp); 1613 dvp->v_mountedhere = NULL; 1614 1615 /* Set up the real rootvnode, and purge the cache */ 1616 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1617 set_rootvnode(); 1618 cache_purgevfs(rootvnode->v_mount); 1619 1620 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1621 error = namei(&nd); 1622 if (error) { 1623 printf("Lookup of /dev for devfs, error: %d\n", error); 1624 vput(dvp); 1625 vfs_unbusy(mp); 1626 return; 1627 } 1628 NDFREE(&nd, NDF_ONLY_PNBUF); 1629 vp = nd.ni_vp; 1630 if (vp->v_type != VDIR) { 1631 printf("/dev is not a directory\n"); 1632 vput(dvp); 1633 vput(vp); 1634 vfs_unbusy(mp); 1635 return; 1636 } 1637 error = vinvalbuf(vp, V_SAVE, 0, 0); 1638 if (error) { 1639 printf("vinvalbuf() of /dev failed, error: %d\n", error); 1640 vput(dvp); 1641 vput(vp); 1642 vfs_unbusy(mp); 1643 return; 1644 } 1645 cache_purge(vp); 1646 mp->mnt_vnodecovered = vp; 1647 vp->v_mountedhere = mp; 1648 mtx_lock(&mountlist_mtx); 1649 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1650 mtx_unlock(&mountlist_mtx); 1651 VOP_UNLOCK(vp, 0); 1652 vput(dvp); 1653 vfs_unbusy(mp); 1654 1655 /* Unlink the no longer needed /dev/dev -> / symlink */ 1656 error = kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1657 if (error) 1658 printf("kern_unlink of /dev/dev failed, error: %d\n", error); 1659} 1660 1661/* 1662 * Report errors during filesystem mounting. 1663 */ 1664void 1665vfs_mount_error(struct mount *mp, const char *fmt, ...) 1666{ 1667 struct vfsoptlist *moptlist = mp->mnt_optnew; 1668 va_list ap; 1669 int error, len; 1670 char *errmsg; 1671 1672 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1673 if (error || errmsg == NULL || len <= 0) 1674 return; 1675 1676 va_start(ap, fmt); 1677 vsnprintf(errmsg, (size_t)len, fmt, ap); 1678 va_end(ap); 1679} 1680 1681void 1682vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1683{ 1684 va_list ap; 1685 int error, len; 1686 char *errmsg; 1687 1688 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1689 if (error || errmsg == NULL || len <= 0) 1690 return; 1691 1692 va_start(ap, fmt); 1693 vsnprintf(errmsg, (size_t)len, fmt, ap); 1694 va_end(ap); 1695} 1696 1697/* 1698 * Find and mount the root filesystem 1699 */ 1700void 1701vfs_mountroot(void) 1702{ 1703 char *cp, *cpt, *options, *tmpdev; 1704 int error, i, asked = 0; 1705 1706 options = NULL; 1707 1708 root_mount_prepare(); 1709 1710 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1711 NULL, NULL, mount_init, mount_fini, 1712 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1713 devfs_first(); 1714 1715 /* 1716 * We are booted with instructions to prompt for the root filesystem. 1717 */ 1718 if (boothowto & RB_ASKNAME) { 1719 if (!vfs_mountroot_ask()) 1720 goto mounted; 1721 asked = 1; 1722 } 1723 1724 options = getenv("vfs.root.mountfrom.options"); 1725 1726 /* 1727 * The root filesystem information is compiled in, and we are 1728 * booted with instructions to use it. 1729 */ 1730 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1731 if (!vfs_mountroot_try(ctrootdevname, options)) 1732 goto mounted; 1733 ctrootdevname = NULL; 1734 } 1735 1736 /* 1737 * We've been given the generic "use CDROM as root" flag. This is 1738 * necessary because one media may be used in many different 1739 * devices, so we need to search for them. 1740 */ 1741 if (boothowto & RB_CDROM) { 1742 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1743 if (!vfs_mountroot_try(cdrom_rootdevnames[i], options)) 1744 goto mounted; 1745 } 1746 } 1747 1748 /* 1749 * Try to use the value read by the loader from /etc/fstab, or 1750 * supplied via some other means. This is the preferred 1751 * mechanism. 1752 */ 1753 cp = getenv("vfs.root.mountfrom"); 1754 if (cp != NULL) { 1755 cpt = cp; 1756 while ((tmpdev = strsep(&cpt, " \t")) != NULL) { 1757 error = vfs_mountroot_try(tmpdev, options); 1758 if (error == 0) { 1759 freeenv(cp); 1760 goto mounted; 1761 } 1762 } 1763 freeenv(cp); 1764 } 1765 1766 /* 1767 * Try values that may have been computed by code during boot 1768 */ 1769 if (!vfs_mountroot_try(rootdevnames[0], options)) 1770 goto mounted; 1771 if (!vfs_mountroot_try(rootdevnames[1], options)) 1772 goto mounted; 1773 1774 /* 1775 * If we (still) have a compiled-in default, try it. 1776 */ 1777 if (ctrootdevname != NULL) 1778 if (!vfs_mountroot_try(ctrootdevname, options)) 1779 goto mounted; 1780 /* 1781 * Everything so far has failed, prompt on the console if we haven't 1782 * already tried that. 1783 */ 1784 if (!asked) 1785 if (!vfs_mountroot_ask()) 1786 goto mounted; 1787 1788 panic("Root mount failed, startup aborted."); 1789 1790mounted: 1791 root_mount_done(); 1792 freeenv(options); 1793} 1794 1795static struct mntarg * 1796parse_mountroot_options(struct mntarg *ma, const char *options) 1797{ 1798 char *p; 1799 char *name, *name_arg; 1800 char *val, *val_arg; 1801 char *opts; 1802 1803 if (options == NULL || options[0] == '\0') 1804 return (ma); 1805 1806 p = opts = strdup(options, M_MOUNT); 1807 if (opts == NULL) { 1808 return (ma); 1809 } 1810 1811 while((name = strsep(&p, ",")) != NULL) { 1812 if (name[0] == '\0') 1813 break; 1814 1815 val = strchr(name, '='); 1816 if (val != NULL) { 1817 *val = '\0'; 1818 ++val; 1819 } 1820 if( strcmp(name, "rw") == 0 || 1821 strcmp(name, "noro") == 0) { 1822 /* 1823 * The first time we mount the root file system, 1824 * we need to mount 'ro', so We need to ignore 1825 * 'rw' and 'noro' mount options. 1826 */ 1827 continue; 1828 } 1829 name_arg = strdup(name, M_MOUNT); 1830 val_arg = NULL; 1831 if (val != NULL) 1832 val_arg = strdup(val, M_MOUNT); 1833 1834 ma = mount_arg(ma, name_arg, val_arg, 1835 (val_arg != NULL ? -1 : 0)); 1836 } 1837 free(opts, M_MOUNT); 1838 return (ma); 1839} 1840 1841/* 1842 * Mount (mountfrom) as the root filesystem. 1843 */ 1844static int 1845vfs_mountroot_try(const char *mountfrom, const char *options) 1846{ 1847 struct mount *mp; 1848 struct mntarg *ma; 1849 char *vfsname, *path; 1850 time_t timebase; 1851 int error; 1852 char patt[32]; 1853 char errmsg[255]; 1854 1855 vfsname = NULL; 1856 path = NULL; 1857 mp = NULL; 1858 ma = NULL; 1859 error = EINVAL; 1860 bzero(errmsg, sizeof(errmsg)); 1861 1862 if (mountfrom == NULL) 1863 return (error); /* don't complain */ 1864 printf("Trying to mount root from %s\n", mountfrom); 1865 1866 /* parse vfs name and path */ 1867 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1868 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1869 vfsname[0] = path[0] = 0; 1870 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1871 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1872 goto out; 1873 1874 if (path[0] == '\0') 1875 strcpy(path, ROOTNAME); 1876 1877 ma = mount_arg(ma, "fstype", vfsname, -1); 1878 ma = mount_arg(ma, "fspath", "/", -1); 1879 ma = mount_arg(ma, "from", path, -1); 1880 ma = mount_arg(ma, "errmsg", errmsg, sizeof(errmsg)); 1881 ma = mount_arg(ma, "ro", NULL, 0); 1882 ma = parse_mountroot_options(ma, options); 1883 error = kernel_mount(ma, MNT_ROOTFS); 1884 1885 if (error == 0) { 1886 /* 1887 * We mount devfs prior to mounting the / FS, so the first 1888 * entry will typically be devfs. 1889 */ 1890 mp = TAILQ_FIRST(&mountlist); 1891 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1892 1893 /* 1894 * Iterate over all currently mounted file systems and use 1895 * the time stamp found to check and/or initialize the RTC. 1896 * Typically devfs has no time stamp and the only other FS 1897 * is the actual / FS. 1898 * Call inittodr() only once and pass it the largest of the 1899 * timestamps we encounter. 1900 */ 1901 timebase = 0; 1902 do { 1903 if (mp->mnt_time > timebase) 1904 timebase = mp->mnt_time; 1905 mp = TAILQ_NEXT(mp, mnt_list); 1906 } while (mp != NULL); 1907 inittodr(timebase); 1908 1909 devfs_fixup(curthread); 1910 } 1911 1912 if (error != 0 ) { 1913 printf("ROOT MOUNT ERROR: %s\n", errmsg); 1914 printf("If you have invalid mount options, reboot, and "); 1915 printf("first try the following from\n"); 1916 printf("the loader prompt:\n\n"); 1917 printf(" set vfs.root.mountfrom.options=rw\n\n"); 1918 printf("and then remove invalid mount options from "); 1919 printf("/etc/fstab.\n\n"); 1920 } 1921out: 1922 free(path, M_MOUNT); 1923 free(vfsname, M_MOUNT); 1924 return (error); 1925} 1926 1927/* 1928 * --------------------------------------------------------------------- 1929 * Interactive root filesystem selection code. 1930 */ 1931 1932static int 1933vfs_mountroot_ask(void) 1934{ 1935 char name[128]; 1936 char *mountfrom; 1937 char *options; 1938 1939 for(;;) { 1940 printf("Loader variables:\n"); 1941 printf("vfs.root.mountfrom="); 1942 mountfrom = getenv("vfs.root.mountfrom"); 1943 if (mountfrom != NULL) { 1944 printf("%s", mountfrom); 1945 } 1946 printf("\n"); 1947 printf("vfs.root.mountfrom.options="); 1948 options = getenv("vfs.root.mountfrom.options"); 1949 if (options != NULL) { 1950 printf("%s", options); 1951 } 1952 printf("\n"); 1953 freeenv(mountfrom); 1954 freeenv(options); 1955 printf("\nManual root filesystem specification:\n"); 1956 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1957 printf(" eg. zfs:tank\n"); 1958 printf(" eg. ufs:/dev/da0s1a\n"); 1959 printf(" eg. cd9660:/dev/acd0\n"); 1960 printf(" This is equivalent to: "); 1961 printf("mount -t cd9660 /dev/acd0 /\n"); 1962 printf("\n"); 1963 printf(" ? List valid disk boot devices\n"); 1964 printf(" <empty line> Abort manual input\n"); 1965 printf("\nmountroot> "); 1966 gets(name, sizeof(name), 1); 1967 if (name[0] == '\0') 1968 return (1); 1969 if (name[0] == '?') { 1970 printf("\nList of GEOM managed disk devices:\n "); 1971 g_dev_print(); 1972 continue; 1973 } 1974 if (!vfs_mountroot_try(name, NULL)) 1975 return (0); 1976 } 1977} 1978 1979/* 1980 * --------------------------------------------------------------------- 1981 * Functions for querying mount options/arguments from filesystems. 1982 */ 1983 1984/* 1985 * Check that no unknown options are given 1986 */ 1987int 1988vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1989{ 1990 struct vfsopt *opt; 1991 char errmsg[255]; 1992 const char **t, *p, *q; 1993 int ret = 0; 1994 1995 TAILQ_FOREACH(opt, opts, link) { 1996 p = opt->name; 1997 q = NULL; 1998 if (p[0] == 'n' && p[1] == 'o') 1999 q = p + 2; 2000 for(t = global_opts; *t != NULL; t++) { 2001 if (strcmp(*t, p) == 0) 2002 break; 2003 if (q != NULL) { 2004 if (strcmp(*t, q) == 0) 2005 break; 2006 } 2007 } 2008 if (*t != NULL) 2009 continue; 2010 for(t = legal; *t != NULL; t++) { 2011 if (strcmp(*t, p) == 0) 2012 break; 2013 if (q != NULL) { 2014 if (strcmp(*t, q) == 0) 2015 break; 2016 } 2017 } 2018 if (*t != NULL) 2019 continue; 2020 snprintf(errmsg, sizeof(errmsg), 2021 "mount option <%s> is unknown", p); 2022 printf("%s\n", errmsg); 2023 ret = EINVAL; 2024 } 2025 if (ret != 0) { 2026 TAILQ_FOREACH(opt, opts, link) { 2027 if (strcmp(opt->name, "errmsg") == 0) { 2028 strncpy((char *)opt->value, errmsg, opt->len); 2029 } 2030 } 2031 } 2032 return (ret); 2033} 2034 2035/* 2036 * Get a mount option by its name. 2037 * 2038 * Return 0 if the option was found, ENOENT otherwise. 2039 * If len is non-NULL it will be filled with the length 2040 * of the option. If buf is non-NULL, it will be filled 2041 * with the address of the option. 2042 */ 2043int 2044vfs_getopt(opts, name, buf, len) 2045 struct vfsoptlist *opts; 2046 const char *name; 2047 void **buf; 2048 int *len; 2049{ 2050 struct vfsopt *opt; 2051 2052 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2053 2054 TAILQ_FOREACH(opt, opts, link) { 2055 if (strcmp(name, opt->name) == 0) { 2056 opt->seen = 1; 2057 if (len != NULL) 2058 *len = opt->len; 2059 if (buf != NULL) 2060 *buf = opt->value; 2061 return (0); 2062 } 2063 } 2064 return (ENOENT); 2065} 2066 2067int 2068vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2069{ 2070 struct vfsopt *opt; 2071 2072 if (opts == NULL) 2073 return (-1); 2074 2075 TAILQ_FOREACH(opt, opts, link) { 2076 if (strcmp(name, opt->name) == 0) { 2077 opt->seen = 1; 2078 return (opt->pos); 2079 } 2080 } 2081 return (-1); 2082} 2083 2084char * 2085vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2086{ 2087 struct vfsopt *opt; 2088 2089 *error = 0; 2090 TAILQ_FOREACH(opt, opts, link) { 2091 if (strcmp(name, opt->name) != 0) 2092 continue; 2093 opt->seen = 1; 2094 if (opt->len == 0 || 2095 ((char *)opt->value)[opt->len - 1] != '\0') { 2096 *error = EINVAL; 2097 return (NULL); 2098 } 2099 return (opt->value); 2100 } 2101 *error = ENOENT; 2102 return (NULL); 2103} 2104 2105int 2106vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 2107{ 2108 struct vfsopt *opt; 2109 2110 TAILQ_FOREACH(opt, opts, link) { 2111 if (strcmp(name, opt->name) == 0) { 2112 opt->seen = 1; 2113 if (w != NULL) 2114 *w |= val; 2115 return (1); 2116 } 2117 } 2118 if (w != NULL) 2119 *w &= ~val; 2120 return (0); 2121} 2122 2123int 2124vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2125{ 2126 va_list ap; 2127 struct vfsopt *opt; 2128 int ret; 2129 2130 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2131 2132 TAILQ_FOREACH(opt, opts, link) { 2133 if (strcmp(name, opt->name) != 0) 2134 continue; 2135 opt->seen = 1; 2136 if (opt->len == 0 || opt->value == NULL) 2137 return (0); 2138 if (((char *)opt->value)[opt->len - 1] != '\0') 2139 return (0); 2140 va_start(ap, fmt); 2141 ret = vsscanf(opt->value, fmt, ap); 2142 va_end(ap); 2143 return (ret); 2144 } 2145 return (0); 2146} 2147 2148int 2149vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2150{ 2151 struct vfsopt *opt; 2152 2153 TAILQ_FOREACH(opt, opts, link) { 2154 if (strcmp(name, opt->name) != 0) 2155 continue; 2156 opt->seen = 1; 2157 if (opt->value == NULL) 2158 opt->len = len; 2159 else { 2160 if (opt->len != len) 2161 return (EINVAL); 2162 bcopy(value, opt->value, len); 2163 } 2164 return (0); 2165 } 2166 return (ENOENT); 2167} 2168 2169int 2170vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2171{ 2172 struct vfsopt *opt; 2173 2174 TAILQ_FOREACH(opt, opts, link) { 2175 if (strcmp(name, opt->name) != 0) 2176 continue; 2177 opt->seen = 1; 2178 if (opt->value == NULL) 2179 opt->len = len; 2180 else { 2181 if (opt->len < len) 2182 return (EINVAL); 2183 opt->len = len; 2184 bcopy(value, opt->value, len); 2185 } 2186 return (0); 2187 } 2188 return (ENOENT); 2189} 2190 2191int 2192vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2193{ 2194 struct vfsopt *opt; 2195 2196 TAILQ_FOREACH(opt, opts, link) { 2197 if (strcmp(name, opt->name) != 0) 2198 continue; 2199 opt->seen = 1; 2200 if (opt->value == NULL) 2201 opt->len = strlen(value) + 1; 2202 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2203 return (EINVAL); 2204 return (0); 2205 } 2206 return (ENOENT); 2207} 2208 2209/* 2210 * Find and copy a mount option. 2211 * 2212 * The size of the buffer has to be specified 2213 * in len, if it is not the same length as the 2214 * mount option, EINVAL is returned. 2215 * Returns ENOENT if the option is not found. 2216 */ 2217int 2218vfs_copyopt(opts, name, dest, len) 2219 struct vfsoptlist *opts; 2220 const char *name; 2221 void *dest; 2222 int len; 2223{ 2224 struct vfsopt *opt; 2225 2226 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2227 2228 TAILQ_FOREACH(opt, opts, link) { 2229 if (strcmp(name, opt->name) == 0) { 2230 opt->seen = 1; 2231 if (len != opt->len) 2232 return (EINVAL); 2233 bcopy(opt->value, dest, opt->len); 2234 return (0); 2235 } 2236 } 2237 return (ENOENT); 2238} 2239 2240/* 2241 * This is a helper function for filesystems to traverse their 2242 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2243 */ 2244 2245struct vnode * 2246__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2247{ 2248 struct vnode *vp; 2249 2250 mtx_assert(MNT_MTX(mp), MA_OWNED); 2251 2252 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2253 if ((*mvp)->v_yield++ == 500) { 2254 MNT_IUNLOCK(mp); 2255 (*mvp)->v_yield = 0; 2256 uio_yield(); 2257 MNT_ILOCK(mp); 2258 } 2259 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2260 while (vp != NULL && vp->v_type == VMARKER) 2261 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2262 2263 /* Check if we are done */ 2264 if (vp == NULL) { 2265 __mnt_vnode_markerfree(mvp, mp); 2266 return (NULL); 2267 } 2268 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2269 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2270 return (vp); 2271} 2272 2273struct vnode * 2274__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2275{ 2276 struct vnode *vp; 2277 2278 mtx_assert(MNT_MTX(mp), MA_OWNED); 2279 2280 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2281 while (vp != NULL && vp->v_type == VMARKER) 2282 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2283 2284 /* Check if we are done */ 2285 if (vp == NULL) { 2286 *mvp = NULL; 2287 return (NULL); 2288 } 2289 MNT_REF(mp); 2290 MNT_IUNLOCK(mp); 2291 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2292 M_VNODE_MARKER, 2293 M_WAITOK | M_ZERO); 2294 MNT_ILOCK(mp); 2295 (*mvp)->v_type = VMARKER; 2296 2297 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2298 while (vp != NULL && vp->v_type == VMARKER) 2299 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2300 2301 /* Check if we are done */ 2302 if (vp == NULL) { 2303 MNT_IUNLOCK(mp); 2304 free(*mvp, M_VNODE_MARKER); 2305 MNT_ILOCK(mp); 2306 *mvp = NULL; 2307 MNT_REL(mp); 2308 return (NULL); 2309 } 2310 (*mvp)->v_mount = mp; 2311 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2312 return (vp); 2313} 2314 2315 2316void 2317__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2318{ 2319 2320 if (*mvp == NULL) 2321 return; 2322 2323 mtx_assert(MNT_MTX(mp), MA_OWNED); 2324 2325 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2326 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2327 MNT_IUNLOCK(mp); 2328 free(*mvp, M_VNODE_MARKER); 2329 MNT_ILOCK(mp); 2330 *mvp = NULL; 2331 MNT_REL(mp); 2332} 2333 2334 2335int 2336__vfs_statfs(struct mount *mp, struct statfs *sbp) 2337{ 2338 int error; 2339 2340 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 2341 if (sbp != &mp->mnt_stat) 2342 *sbp = mp->mnt_stat; 2343 return (error); 2344} 2345 2346void 2347vfs_mountedfrom(struct mount *mp, const char *from) 2348{ 2349 2350 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2351 strlcpy(mp->mnt_stat.f_mntfromname, from, 2352 sizeof mp->mnt_stat.f_mntfromname); 2353} 2354 2355/* 2356 * --------------------------------------------------------------------- 2357 * This is the api for building mount args and mounting filesystems from 2358 * inside the kernel. 2359 * 2360 * The API works by accumulation of individual args. First error is 2361 * latched. 2362 * 2363 * XXX: should be documented in new manpage kernel_mount(9) 2364 */ 2365 2366/* A memory allocation which must be freed when we are done */ 2367struct mntaarg { 2368 SLIST_ENTRY(mntaarg) next; 2369}; 2370 2371/* The header for the mount arguments */ 2372struct mntarg { 2373 struct iovec *v; 2374 int len; 2375 int error; 2376 SLIST_HEAD(, mntaarg) list; 2377}; 2378 2379/* 2380 * Add a boolean argument. 2381 * 2382 * flag is the boolean value. 2383 * name must start with "no". 2384 */ 2385struct mntarg * 2386mount_argb(struct mntarg *ma, int flag, const char *name) 2387{ 2388 2389 KASSERT(name[0] == 'n' && name[1] == 'o', 2390 ("mount_argb(...,%s): name must start with 'no'", name)); 2391 2392 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2393} 2394 2395/* 2396 * Add an argument printf style 2397 */ 2398struct mntarg * 2399mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2400{ 2401 va_list ap; 2402 struct mntaarg *maa; 2403 struct sbuf *sb; 2404 int len; 2405 2406 if (ma == NULL) { 2407 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2408 SLIST_INIT(&ma->list); 2409 } 2410 if (ma->error) 2411 return (ma); 2412 2413 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2414 M_MOUNT, M_WAITOK); 2415 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2416 ma->v[ma->len].iov_len = strlen(name) + 1; 2417 ma->len++; 2418 2419 sb = sbuf_new_auto(); 2420 va_start(ap, fmt); 2421 sbuf_vprintf(sb, fmt, ap); 2422 va_end(ap); 2423 sbuf_finish(sb); 2424 len = sbuf_len(sb) + 1; 2425 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2426 SLIST_INSERT_HEAD(&ma->list, maa, next); 2427 bcopy(sbuf_data(sb), maa + 1, len); 2428 sbuf_delete(sb); 2429 2430 ma->v[ma->len].iov_base = maa + 1; 2431 ma->v[ma->len].iov_len = len; 2432 ma->len++; 2433 2434 return (ma); 2435} 2436 2437/* 2438 * Add an argument which is a userland string. 2439 */ 2440struct mntarg * 2441mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2442{ 2443 struct mntaarg *maa; 2444 char *tbuf; 2445 2446 if (val == NULL) 2447 return (ma); 2448 if (ma == NULL) { 2449 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2450 SLIST_INIT(&ma->list); 2451 } 2452 if (ma->error) 2453 return (ma); 2454 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2455 SLIST_INSERT_HEAD(&ma->list, maa, next); 2456 tbuf = (void *)(maa + 1); 2457 ma->error = copyinstr(val, tbuf, len, NULL); 2458 return (mount_arg(ma, name, tbuf, -1)); 2459} 2460 2461/* 2462 * Plain argument. 2463 * 2464 * If length is -1, treat value as a C string. 2465 */ 2466struct mntarg * 2467mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2468{ 2469 2470 if (ma == NULL) { 2471 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2472 SLIST_INIT(&ma->list); 2473 } 2474 if (ma->error) 2475 return (ma); 2476 2477 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2478 M_MOUNT, M_WAITOK); 2479 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2480 ma->v[ma->len].iov_len = strlen(name) + 1; 2481 ma->len++; 2482 2483 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2484 if (len < 0) 2485 ma->v[ma->len].iov_len = strlen(val) + 1; 2486 else 2487 ma->v[ma->len].iov_len = len; 2488 ma->len++; 2489 return (ma); 2490} 2491 2492/* 2493 * Free a mntarg structure 2494 */ 2495static void 2496free_mntarg(struct mntarg *ma) 2497{ 2498 struct mntaarg *maa; 2499 2500 while (!SLIST_EMPTY(&ma->list)) { 2501 maa = SLIST_FIRST(&ma->list); 2502 SLIST_REMOVE_HEAD(&ma->list, next); 2503 free(maa, M_MOUNT); 2504 } 2505 free(ma->v, M_MOUNT); 2506 free(ma, M_MOUNT); 2507} 2508 2509/* 2510 * Mount a filesystem 2511 */ 2512int 2513kernel_mount(struct mntarg *ma, int flags) 2514{ 2515 struct uio auio; 2516 int error; 2517 2518 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2519 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2520 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2521 2522 auio.uio_iov = ma->v; 2523 auio.uio_iovcnt = ma->len; 2524 auio.uio_segflg = UIO_SYSSPACE; 2525 2526 error = ma->error; 2527 if (!error) 2528 error = vfs_donmount(curthread, flags, &auio); 2529 free_mntarg(ma); 2530 return (error); 2531} 2532 2533/* 2534 * A printflike function to mount a filesystem. 2535 */ 2536int 2537kernel_vmount(int flags, ...) 2538{ 2539 struct mntarg *ma = NULL; 2540 va_list ap; 2541 const char *cp; 2542 const void *vp; 2543 int error; 2544 2545 va_start(ap, flags); 2546 for (;;) { 2547 cp = va_arg(ap, const char *); 2548 if (cp == NULL) 2549 break; 2550 vp = va_arg(ap, const void *); 2551 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2552 } 2553 va_end(ap); 2554 2555 error = kernel_mount(ma, flags); 2556 return (error); 2557} 2558