vfs_mount.c revision 195939
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 195939 2009-07-29 07:44:43Z rwatson $"); 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>] 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 * The environment variable vfs.root.mountfrom options is a comma delimited 114 * set of string mount options. These mount options must be parseable 115 * by nmount() in the kernel. 116 */ 117 118/* 119 * Global opts, taken by all filesystems 120 */ 121static const char *global_opts[] = { 122 "errmsg", 123 "fstype", 124 "fspath", 125 "ro", 126 "rw", 127 "nosuid", 128 "noexec", 129 NULL 130}; 131 132/* 133 * The root specifiers we will try if RB_CDROM is specified. 134 */ 135static char *cdrom_rootdevnames[] = { 136 "cd9660:cd0", 137 "cd9660:acd0", 138 NULL 139}; 140 141/* legacy find-root code */ 142char *rootdevnames[2] = {NULL, NULL}; 143#ifndef ROOTDEVNAME 144# define ROOTDEVNAME NULL 145#endif 146static const char *ctrootdevname = ROOTDEVNAME; 147 148/* 149 * --------------------------------------------------------------------- 150 * Functions for building and sanitizing the mount options 151 */ 152 153/* Remove one mount option. */ 154static void 155vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 156{ 157 158 TAILQ_REMOVE(opts, opt, link); 159 free(opt->name, M_MOUNT); 160 if (opt->value != NULL) 161 free(opt->value, M_MOUNT); 162 free(opt, M_MOUNT); 163} 164 165/* Release all resources related to the mount options. */ 166void 167vfs_freeopts(struct vfsoptlist *opts) 168{ 169 struct vfsopt *opt; 170 171 while (!TAILQ_EMPTY(opts)) { 172 opt = TAILQ_FIRST(opts); 173 vfs_freeopt(opts, opt); 174 } 175 free(opts, M_MOUNT); 176} 177 178void 179vfs_deleteopt(struct vfsoptlist *opts, const char *name) 180{ 181 struct vfsopt *opt, *temp; 182 183 if (opts == NULL) 184 return; 185 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 186 if (strcmp(opt->name, name) == 0) 187 vfs_freeopt(opts, opt); 188 } 189} 190 191/* 192 * Check if options are equal (with or without the "no" prefix). 193 */ 194static int 195vfs_equalopts(const char *opt1, const char *opt2) 196{ 197 char *p; 198 199 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 200 if (strcmp(opt1, opt2) == 0) 201 return (1); 202 /* "noopt" vs. "opt" */ 203 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 204 return (1); 205 /* "opt" vs. "noopt" */ 206 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 207 return (1); 208 while ((p = strchr(opt1, '.')) != NULL && 209 !strncmp(opt1, opt2, ++p - opt1)) { 210 opt2 += p - opt1; 211 opt1 = p; 212 /* "foo.noopt" vs. "foo.opt" */ 213 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 214 return (1); 215 /* "foo.opt" vs. "foo.noopt" */ 216 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 217 return (1); 218 } 219 return (0); 220} 221 222/* 223 * If a mount option is specified several times, 224 * (with or without the "no" prefix) only keep 225 * the last occurence of it. 226 */ 227static void 228vfs_sanitizeopts(struct vfsoptlist *opts) 229{ 230 struct vfsopt *opt, *opt2, *tmp; 231 232 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 233 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 234 while (opt2 != NULL) { 235 if (vfs_equalopts(opt->name, opt2->name)) { 236 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 237 vfs_freeopt(opts, opt2); 238 opt2 = tmp; 239 } else { 240 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 241 } 242 } 243 } 244} 245 246/* 247 * Build a linked list of mount options from a struct uio. 248 */ 249int 250vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 251{ 252 struct vfsoptlist *opts; 253 struct vfsopt *opt; 254 size_t memused, namelen, optlen; 255 unsigned int i, iovcnt; 256 int error; 257 258 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 259 TAILQ_INIT(opts); 260 memused = 0; 261 iovcnt = auio->uio_iovcnt; 262 for (i = 0; i < iovcnt; i += 2) { 263 namelen = auio->uio_iov[i].iov_len; 264 optlen = auio->uio_iov[i + 1].iov_len; 265 memused += sizeof(struct vfsopt) + optlen + namelen; 266 /* 267 * Avoid consuming too much memory, and attempts to overflow 268 * memused. 269 */ 270 if (memused > VFS_MOUNTARG_SIZE_MAX || 271 optlen > VFS_MOUNTARG_SIZE_MAX || 272 namelen > VFS_MOUNTARG_SIZE_MAX) { 273 error = EINVAL; 274 goto bad; 275 } 276 277 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 278 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 279 opt->value = NULL; 280 opt->len = 0; 281 opt->pos = i / 2; 282 opt->seen = 0; 283 284 /* 285 * Do this early, so jumps to "bad" will free the current 286 * option. 287 */ 288 TAILQ_INSERT_TAIL(opts, opt, link); 289 290 if (auio->uio_segflg == UIO_SYSSPACE) { 291 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 292 } else { 293 error = copyin(auio->uio_iov[i].iov_base, opt->name, 294 namelen); 295 if (error) 296 goto bad; 297 } 298 /* Ensure names are null-terminated strings. */ 299 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 300 error = EINVAL; 301 goto bad; 302 } 303 if (optlen != 0) { 304 opt->len = optlen; 305 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 306 if (auio->uio_segflg == UIO_SYSSPACE) { 307 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 308 optlen); 309 } else { 310 error = copyin(auio->uio_iov[i + 1].iov_base, 311 opt->value, optlen); 312 if (error) 313 goto bad; 314 } 315 } 316 } 317 vfs_sanitizeopts(opts); 318 *options = opts; 319 return (0); 320bad: 321 vfs_freeopts(opts); 322 return (error); 323} 324 325/* 326 * Merge the old mount options with the new ones passed 327 * in the MNT_UPDATE case. 328 * 329 * XXX This function will keep a "nofoo" option in the 330 * new options if there is no matching "foo" option 331 * to be cancelled in the old options. This is a bug 332 * if the option's canonical name is "foo". E.g., "noro" 333 * shouldn't end up in the mount point's active options, 334 * but it can. 335 */ 336static void 337vfs_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 new->seen = opt->seen; 369 TAILQ_INSERT_TAIL(toopts, new, link); 370next: 371 continue; 372 } 373} 374 375/* 376 * Mount a filesystem. 377 */ 378int 379nmount(td, uap) 380 struct thread *td; 381 struct nmount_args /* { 382 struct iovec *iovp; 383 unsigned int iovcnt; 384 int flags; 385 } */ *uap; 386{ 387 struct uio *auio; 388 int error; 389 u_int iovcnt; 390 391 AUDIT_ARG_FFLAGS(uap->flags); 392 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 393 uap->iovp, uap->iovcnt, uap->flags); 394 395 /* 396 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 397 * userspace to set this flag, but we must filter it out if we want 398 * MNT_UPDATE on the root file system to work. 399 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 400 */ 401 uap->flags &= ~MNT_ROOTFS; 402 403 iovcnt = uap->iovcnt; 404 /* 405 * Check that we have an even number of iovec's 406 * and that we have at least two options. 407 */ 408 if ((iovcnt & 1) || (iovcnt < 4)) { 409 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 410 uap->iovcnt); 411 return (EINVAL); 412 } 413 414 error = copyinuio(uap->iovp, iovcnt, &auio); 415 if (error) { 416 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 417 __func__, error); 418 return (error); 419 } 420 error = vfs_donmount(td, uap->flags, auio); 421 422 free(auio, M_IOV); 423 return (error); 424} 425 426/* 427 * --------------------------------------------------------------------- 428 * Various utility functions 429 */ 430 431void 432vfs_ref(struct mount *mp) 433{ 434 435 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 436 MNT_ILOCK(mp); 437 MNT_REF(mp); 438 MNT_IUNLOCK(mp); 439} 440 441void 442vfs_rel(struct mount *mp) 443{ 444 445 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 446 MNT_ILOCK(mp); 447 MNT_REL(mp); 448 MNT_IUNLOCK(mp); 449} 450 451static int 452mount_init(void *mem, int size, int flags) 453{ 454 struct mount *mp; 455 456 mp = (struct mount *)mem; 457 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 458 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 459 return (0); 460} 461 462static void 463mount_fini(void *mem, int size) 464{ 465 struct mount *mp; 466 467 mp = (struct mount *)mem; 468 lockdestroy(&mp->mnt_explock); 469 mtx_destroy(&mp->mnt_mtx); 470} 471 472/* 473 * Allocate and initialize the mount point struct. 474 */ 475struct mount * 476vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 477 struct ucred *cred) 478{ 479 struct mount *mp; 480 481 mp = uma_zalloc(mount_zone, M_WAITOK); 482 bzero(&mp->mnt_startzero, 483 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 484 TAILQ_INIT(&mp->mnt_nvnodelist); 485 mp->mnt_nvnodelistsize = 0; 486 mp->mnt_ref = 0; 487 (void) vfs_busy(mp, MBF_NOWAIT); 488 mp->mnt_op = vfsp->vfc_vfsops; 489 mp->mnt_vfc = vfsp; 490 vfsp->vfc_refcount++; /* XXX Unlocked */ 491 mp->mnt_stat.f_type = vfsp->vfc_typenum; 492 mp->mnt_gen++; 493 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 494 mp->mnt_vnodecovered = vp; 495 mp->mnt_cred = crdup(cred); 496 mp->mnt_stat.f_owner = cred->cr_uid; 497 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 498 mp->mnt_iosize_max = DFLTPHYS; 499#ifdef MAC 500 mac_mount_init(mp); 501 mac_mount_create(cred, mp); 502#endif 503 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 504 return (mp); 505} 506 507/* 508 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 509 */ 510void 511vfs_mount_destroy(struct mount *mp) 512{ 513 514 MNT_ILOCK(mp); 515 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 516 if (mp->mnt_kern_flag & MNTK_MWAIT) { 517 mp->mnt_kern_flag &= ~MNTK_MWAIT; 518 wakeup(mp); 519 } 520 while (mp->mnt_ref) 521 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 522 KASSERT(mp->mnt_ref == 0, 523 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 524 __FILE__, __LINE__)); 525 if (mp->mnt_writeopcount != 0) 526 panic("vfs_mount_destroy: nonzero writeopcount"); 527 if (mp->mnt_secondary_writes != 0) 528 panic("vfs_mount_destroy: nonzero secondary_writes"); 529 mp->mnt_vfc->vfc_refcount--; 530 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 531 struct vnode *vp; 532 533 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 534 vprint("", vp); 535 panic("unmount: dangling vnode"); 536 } 537 if (mp->mnt_nvnodelistsize != 0) 538 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 539 if (mp->mnt_lockref != 0) 540 panic("vfs_mount_destroy: nonzero lock refcount"); 541 MNT_IUNLOCK(mp); 542#ifdef MAC 543 mac_mount_destroy(mp); 544#endif 545 if (mp->mnt_opt != NULL) 546 vfs_freeopts(mp->mnt_opt); 547 crfree(mp->mnt_cred); 548 uma_zfree(mount_zone, mp); 549} 550 551int 552vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 553{ 554 struct vfsoptlist *optlist; 555 struct vfsopt *opt, *noro_opt, *tmp_opt; 556 char *fstype, *fspath, *errmsg; 557 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 558 int has_rw, has_noro; 559 560 errmsg = fspath = NULL; 561 errmsg_len = has_noro = has_rw = fspathlen = 0; 562 errmsg_pos = -1; 563 564 error = vfs_buildopts(fsoptions, &optlist); 565 if (error) 566 return (error); 567 568 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 569 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 570 571 /* 572 * We need these two options before the others, 573 * and they are mandatory for any filesystem. 574 * Ensure they are NUL terminated as well. 575 */ 576 fstypelen = 0; 577 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 578 if (error || fstype[fstypelen - 1] != '\0') { 579 error = EINVAL; 580 if (errmsg != NULL) 581 strncpy(errmsg, "Invalid fstype", errmsg_len); 582 goto bail; 583 } 584 fspathlen = 0; 585 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 586 if (error || fspath[fspathlen - 1] != '\0') { 587 error = EINVAL; 588 if (errmsg != NULL) 589 strncpy(errmsg, "Invalid fspath", errmsg_len); 590 goto bail; 591 } 592 593 /* 594 * We need to see if we have the "update" option 595 * before we call vfs_domount(), since vfs_domount() has special 596 * logic based on MNT_UPDATE. This is very important 597 * when we want to update the root filesystem. 598 */ 599 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 600 if (strcmp(opt->name, "update") == 0) { 601 fsflags |= MNT_UPDATE; 602 vfs_freeopt(optlist, opt); 603 } 604 else if (strcmp(opt->name, "async") == 0) 605 fsflags |= MNT_ASYNC; 606 else if (strcmp(opt->name, "force") == 0) { 607 fsflags |= MNT_FORCE; 608 vfs_freeopt(optlist, opt); 609 } 610 else if (strcmp(opt->name, "reload") == 0) { 611 fsflags |= MNT_RELOAD; 612 vfs_freeopt(optlist, opt); 613 } 614 else if (strcmp(opt->name, "multilabel") == 0) 615 fsflags |= MNT_MULTILABEL; 616 else if (strcmp(opt->name, "noasync") == 0) 617 fsflags &= ~MNT_ASYNC; 618 else if (strcmp(opt->name, "noatime") == 0) 619 fsflags |= MNT_NOATIME; 620 else if (strcmp(opt->name, "atime") == 0) { 621 free(opt->name, M_MOUNT); 622 opt->name = strdup("nonoatime", M_MOUNT); 623 } 624 else if (strcmp(opt->name, "noclusterr") == 0) 625 fsflags |= MNT_NOCLUSTERR; 626 else if (strcmp(opt->name, "clusterr") == 0) { 627 free(opt->name, M_MOUNT); 628 opt->name = strdup("nonoclusterr", M_MOUNT); 629 } 630 else if (strcmp(opt->name, "noclusterw") == 0) 631 fsflags |= MNT_NOCLUSTERW; 632 else if (strcmp(opt->name, "clusterw") == 0) { 633 free(opt->name, M_MOUNT); 634 opt->name = strdup("nonoclusterw", M_MOUNT); 635 } 636 else if (strcmp(opt->name, "noexec") == 0) 637 fsflags |= MNT_NOEXEC; 638 else if (strcmp(opt->name, "exec") == 0) { 639 free(opt->name, M_MOUNT); 640 opt->name = strdup("nonoexec", M_MOUNT); 641 } 642 else if (strcmp(opt->name, "nosuid") == 0) 643 fsflags |= MNT_NOSUID; 644 else if (strcmp(opt->name, "suid") == 0) { 645 free(opt->name, M_MOUNT); 646 opt->name = strdup("nonosuid", M_MOUNT); 647 } 648 else if (strcmp(opt->name, "nosymfollow") == 0) 649 fsflags |= MNT_NOSYMFOLLOW; 650 else if (strcmp(opt->name, "symfollow") == 0) { 651 free(opt->name, M_MOUNT); 652 opt->name = strdup("nonosymfollow", M_MOUNT); 653 } 654 else if (strcmp(opt->name, "noro") == 0) { 655 fsflags &= ~MNT_RDONLY; 656 has_noro = 1; 657 } 658 else if (strcmp(opt->name, "rw") == 0) { 659 fsflags &= ~MNT_RDONLY; 660 has_rw = 1; 661 } 662 else if (strcmp(opt->name, "ro") == 0) 663 fsflags |= MNT_RDONLY; 664 else if (strcmp(opt->name, "rdonly") == 0) { 665 free(opt->name, M_MOUNT); 666 opt->name = strdup("ro", M_MOUNT); 667 fsflags |= MNT_RDONLY; 668 } 669 else if (strcmp(opt->name, "suiddir") == 0) 670 fsflags |= MNT_SUIDDIR; 671 else if (strcmp(opt->name, "sync") == 0) 672 fsflags |= MNT_SYNCHRONOUS; 673 else if (strcmp(opt->name, "union") == 0) 674 fsflags |= MNT_UNION; 675 } 676 677 /* 678 * If "rw" was specified as a mount option, and we 679 * are trying to update a mount-point from "ro" to "rw", 680 * we need a mount option "noro", since in vfs_mergeopts(), 681 * "noro" will cancel "ro", but "rw" will not do anything. 682 */ 683 if (has_rw && !has_noro) { 684 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 685 noro_opt->name = strdup("noro", M_MOUNT); 686 noro_opt->value = NULL; 687 noro_opt->len = 0; 688 noro_opt->pos = -1; 689 noro_opt->seen = 1; 690 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 691 } 692 693 /* 694 * Be ultra-paranoid about making sure the type and fspath 695 * variables will fit in our mp buffers, including the 696 * terminating NUL. 697 */ 698 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 699 error = ENAMETOOLONG; 700 goto bail; 701 } 702 703 mtx_lock(&Giant); 704 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 705 mtx_unlock(&Giant); 706bail: 707 /* copyout the errmsg */ 708 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 709 && errmsg_len > 0 && errmsg != NULL) { 710 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 711 bcopy(errmsg, 712 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 713 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 714 } else { 715 copyout(errmsg, 716 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 717 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 718 } 719 } 720 721 if (error != 0) 722 vfs_freeopts(optlist); 723 return (error); 724} 725 726/* 727 * Old mount API. 728 */ 729#ifndef _SYS_SYSPROTO_H_ 730struct mount_args { 731 char *type; 732 char *path; 733 int flags; 734 caddr_t data; 735}; 736#endif 737/* ARGSUSED */ 738int 739mount(td, uap) 740 struct thread *td; 741 struct mount_args /* { 742 char *type; 743 char *path; 744 int flags; 745 caddr_t data; 746 } */ *uap; 747{ 748 char *fstype; 749 struct vfsconf *vfsp = NULL; 750 struct mntarg *ma = NULL; 751 int error; 752 753 AUDIT_ARG_FFLAGS(uap->flags); 754 755 /* 756 * Filter out MNT_ROOTFS. We do not want clients of mount() in 757 * userspace to set this flag, but we must filter it out if we want 758 * MNT_UPDATE on the root file system to work. 759 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 760 */ 761 uap->flags &= ~MNT_ROOTFS; 762 763 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 764 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 765 if (error) { 766 free(fstype, M_TEMP); 767 return (error); 768 } 769 770 AUDIT_ARG_TEXT(fstype); 771 mtx_lock(&Giant); 772 vfsp = vfs_byname_kld(fstype, td, &error); 773 free(fstype, M_TEMP); 774 if (vfsp == NULL) { 775 mtx_unlock(&Giant); 776 return (ENOENT); 777 } 778 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 779 mtx_unlock(&Giant); 780 return (EOPNOTSUPP); 781 } 782 783 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 784 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 785 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 786 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 787 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 788 789 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags); 790 mtx_unlock(&Giant); 791 return (error); 792} 793 794 795/* 796 * vfs_domount(): actually attempt a filesystem mount. 797 */ 798static int 799vfs_domount( 800 struct thread *td, /* Calling thread. */ 801 const char *fstype, /* Filesystem type. */ 802 char *fspath, /* Mount path. */ 803 int fsflags, /* Flags common to all filesystems. */ 804 void *fsdata /* Options local to the filesystem. */ 805 ) 806{ 807 struct vnode *vp; 808 struct mount *mp; 809 struct vfsconf *vfsp; 810 struct oexport_args oexport; 811 struct export_args export; 812 int error, flag = 0; 813 struct vattr va; 814 struct nameidata nd; 815 816 mtx_assert(&Giant, MA_OWNED); 817 /* 818 * Be ultra-paranoid about making sure the type and fspath 819 * variables will fit in our mp buffers, including the 820 * terminating NUL. 821 */ 822 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 823 return (ENAMETOOLONG); 824 825 if (jailed(td->td_ucred) || usermount == 0) { 826 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 827 return (error); 828 } 829 830 /* 831 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 832 */ 833 if (fsflags & MNT_EXPORTED) { 834 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 835 if (error) 836 return (error); 837 } 838 if (fsflags & MNT_SUIDDIR) { 839 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 840 if (error) 841 return (error); 842 } 843 /* 844 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 845 */ 846 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 847 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 848 fsflags |= MNT_NOSUID | MNT_USER; 849 } 850 851 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 852 vfsp = NULL; 853 if ((fsflags & MNT_UPDATE) == 0) { 854 /* Don't try to load KLDs if we're mounting the root. */ 855 if (fsflags & MNT_ROOTFS) 856 vfsp = vfs_byname(fstype); 857 else 858 vfsp = vfs_byname_kld(fstype, td, &error); 859 if (vfsp == NULL) 860 return (ENODEV); 861 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 862 return (EPERM); 863 } 864 /* 865 * Get vnode to be covered 866 */ 867 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 868 fspath, td); 869 if ((error = namei(&nd)) != 0) 870 return (error); 871 NDFREE(&nd, NDF_ONLY_PNBUF); 872 vp = nd.ni_vp; 873 if (fsflags & MNT_UPDATE) { 874 if ((vp->v_vflag & VV_ROOT) == 0) { 875 vput(vp); 876 return (EINVAL); 877 } 878 mp = vp->v_mount; 879 MNT_ILOCK(mp); 880 flag = mp->mnt_flag; 881 /* 882 * We only allow the filesystem to be reloaded if it 883 * is currently mounted read-only. 884 */ 885 if ((fsflags & MNT_RELOAD) && 886 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 887 MNT_IUNLOCK(mp); 888 vput(vp); 889 return (EOPNOTSUPP); /* Needs translation */ 890 } 891 MNT_IUNLOCK(mp); 892 /* 893 * Only privileged root, or (if MNT_USER is set) the user that 894 * did the original mount is permitted to update it. 895 */ 896 error = vfs_suser(mp, td); 897 if (error) { 898 vput(vp); 899 return (error); 900 } 901 if (vfs_busy(mp, MBF_NOWAIT)) { 902 vput(vp); 903 return (EBUSY); 904 } 905 VI_LOCK(vp); 906 if ((vp->v_iflag & VI_MOUNT) != 0 || 907 vp->v_mountedhere != NULL) { 908 VI_UNLOCK(vp); 909 vfs_unbusy(mp); 910 vput(vp); 911 return (EBUSY); 912 } 913 vp->v_iflag |= VI_MOUNT; 914 VI_UNLOCK(vp); 915 MNT_ILOCK(mp); 916 mp->mnt_flag |= fsflags & 917 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 918 MNT_IUNLOCK(mp); 919 VOP_UNLOCK(vp, 0); 920 mp->mnt_optnew = fsdata; 921 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 922 } else { 923 /* 924 * If the user is not root, ensure that they own the directory 925 * onto which we are attempting to mount. 926 */ 927 error = VOP_GETATTR(vp, &va, td->td_ucred); 928 if (error) { 929 vput(vp); 930 return (error); 931 } 932 if (va.va_uid != td->td_ucred->cr_uid) { 933 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 934 0); 935 if (error) { 936 vput(vp); 937 return (error); 938 } 939 } 940 error = vinvalbuf(vp, V_SAVE, 0, 0); 941 if (error != 0) { 942 vput(vp); 943 return (error); 944 } 945 if (vp->v_type != VDIR) { 946 vput(vp); 947 return (ENOTDIR); 948 } 949 VI_LOCK(vp); 950 if ((vp->v_iflag & VI_MOUNT) != 0 || 951 vp->v_mountedhere != NULL) { 952 VI_UNLOCK(vp); 953 vput(vp); 954 return (EBUSY); 955 } 956 vp->v_iflag |= VI_MOUNT; 957 VI_UNLOCK(vp); 958 959 /* 960 * Allocate and initialize the filesystem. 961 */ 962 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 963 VOP_UNLOCK(vp, 0); 964 965 /* XXXMAC: pass to vfs_mount_alloc? */ 966 mp->mnt_optnew = fsdata; 967 } 968 969 /* 970 * Set the mount level flags. 971 */ 972 MNT_ILOCK(mp); 973 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 974 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 975 MNT_RDONLY)); 976 if ((mp->mnt_flag & MNT_ASYNC) == 0) 977 mp->mnt_kern_flag &= ~MNTK_ASYNC; 978 MNT_IUNLOCK(mp); 979 /* 980 * Mount the filesystem. 981 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 982 * get. No freeing of cn_pnbuf. 983 */ 984 error = VFS_MOUNT(mp); 985 986 /* 987 * Process the export option only if we are 988 * updating mount options. 989 */ 990 if (!error && (fsflags & MNT_UPDATE)) { 991 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 992 sizeof(export)) == 0) 993 error = vfs_export(mp, &export); 994 else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 995 sizeof(oexport)) == 0) { 996 export.ex_flags = oexport.ex_flags; 997 export.ex_root = oexport.ex_root; 998 export.ex_anon = oexport.ex_anon; 999 export.ex_addr = oexport.ex_addr; 1000 export.ex_addrlen = oexport.ex_addrlen; 1001 export.ex_mask = oexport.ex_mask; 1002 export.ex_masklen = oexport.ex_masklen; 1003 export.ex_indexfile = oexport.ex_indexfile; 1004 export.ex_numsecflavors = 0; 1005 error = vfs_export(mp, &export); 1006 } 1007 } 1008 1009 if (!error) { 1010 if (mp->mnt_opt != NULL) 1011 vfs_freeopts(mp->mnt_opt); 1012 mp->mnt_opt = mp->mnt_optnew; 1013 (void)VFS_STATFS(mp, &mp->mnt_stat); 1014 } 1015 /* 1016 * Prevent external consumers of mount options from reading 1017 * mnt_optnew. 1018 */ 1019 mp->mnt_optnew = NULL; 1020 if (mp->mnt_flag & MNT_UPDATE) { 1021 MNT_ILOCK(mp); 1022 if (error) 1023 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1024 (flag & ~MNT_QUOTA); 1025 else 1026 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1027 MNT_FORCE | MNT_SNAPSHOT); 1028 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1029 mp->mnt_kern_flag |= MNTK_ASYNC; 1030 else 1031 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1032 MNT_IUNLOCK(mp); 1033 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1034 if (mp->mnt_syncer == NULL) 1035 error = vfs_allocate_syncvnode(mp); 1036 } else { 1037 if (mp->mnt_syncer != NULL) 1038 vrele(mp->mnt_syncer); 1039 mp->mnt_syncer = NULL; 1040 } 1041 vfs_unbusy(mp); 1042 VI_LOCK(vp); 1043 vp->v_iflag &= ~VI_MOUNT; 1044 VI_UNLOCK(vp); 1045 vrele(vp); 1046 return (error); 1047 } 1048 MNT_ILOCK(mp); 1049 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1050 mp->mnt_kern_flag |= MNTK_ASYNC; 1051 else 1052 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1053 MNT_IUNLOCK(mp); 1054 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1055 /* 1056 * Put the new filesystem on the mount list after root. 1057 */ 1058 cache_purge(vp); 1059 if (!error) { 1060 struct vnode *newdp; 1061 1062 VI_LOCK(vp); 1063 vp->v_iflag &= ~VI_MOUNT; 1064 VI_UNLOCK(vp); 1065 vp->v_mountedhere = mp; 1066 mtx_lock(&mountlist_mtx); 1067 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1068 mtx_unlock(&mountlist_mtx); 1069 vfs_event_signal(NULL, VQ_MOUNT, 0); 1070 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 1071 panic("mount: lost mount"); 1072 mountcheckdirs(vp, newdp); 1073 vput(newdp); 1074 VOP_UNLOCK(vp, 0); 1075 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1076 error = vfs_allocate_syncvnode(mp); 1077 vfs_unbusy(mp); 1078 if (error) 1079 vrele(vp); 1080 } else { 1081 VI_LOCK(vp); 1082 vp->v_iflag &= ~VI_MOUNT; 1083 VI_UNLOCK(vp); 1084 vfs_unbusy(mp); 1085 vfs_mount_destroy(mp); 1086 vput(vp); 1087 } 1088 return (error); 1089} 1090 1091/* 1092 * Unmount a filesystem. 1093 * 1094 * Note: unmount takes a path to the vnode mounted on as argument, not 1095 * special file (as before). 1096 */ 1097#ifndef _SYS_SYSPROTO_H_ 1098struct unmount_args { 1099 char *path; 1100 int flags; 1101}; 1102#endif 1103/* ARGSUSED */ 1104int 1105unmount(td, uap) 1106 struct thread *td; 1107 register struct unmount_args /* { 1108 char *path; 1109 int flags; 1110 } */ *uap; 1111{ 1112 struct mount *mp; 1113 char *pathbuf; 1114 int error, id0, id1; 1115 1116 AUDIT_ARG_VALUE(uap->flags); 1117 if (jailed(td->td_ucred) || usermount == 0) { 1118 error = priv_check(td, PRIV_VFS_UNMOUNT); 1119 if (error) 1120 return (error); 1121 } 1122 1123 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1124 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1125 if (error) { 1126 free(pathbuf, M_TEMP); 1127 return (error); 1128 } 1129 mtx_lock(&Giant); 1130 if (uap->flags & MNT_BYFSID) { 1131 AUDIT_ARG_TEXT(pathbuf); 1132 /* Decode the filesystem ID. */ 1133 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1134 mtx_unlock(&Giant); 1135 free(pathbuf, M_TEMP); 1136 return (EINVAL); 1137 } 1138 1139 mtx_lock(&mountlist_mtx); 1140 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1141 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1142 mp->mnt_stat.f_fsid.val[1] == id1) 1143 break; 1144 } 1145 mtx_unlock(&mountlist_mtx); 1146 } else { 1147 AUDIT_ARG_UPATH1(td, pathbuf); 1148 mtx_lock(&mountlist_mtx); 1149 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1150 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1151 break; 1152 } 1153 mtx_unlock(&mountlist_mtx); 1154 } 1155 free(pathbuf, M_TEMP); 1156 if (mp == NULL) { 1157 /* 1158 * Previously we returned ENOENT for a nonexistent path and 1159 * EINVAL for a non-mountpoint. We cannot tell these apart 1160 * now, so in the !MNT_BYFSID case return the more likely 1161 * EINVAL for compatibility. 1162 */ 1163 mtx_unlock(&Giant); 1164 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1165 } 1166 1167 /* 1168 * Don't allow unmounting the root filesystem. 1169 */ 1170 if (mp->mnt_flag & MNT_ROOTFS) { 1171 mtx_unlock(&Giant); 1172 return (EINVAL); 1173 } 1174 error = dounmount(mp, uap->flags, td); 1175 mtx_unlock(&Giant); 1176 return (error); 1177} 1178 1179/* 1180 * Do the actual filesystem unmount. 1181 */ 1182int 1183dounmount(mp, flags, td) 1184 struct mount *mp; 1185 int flags; 1186 struct thread *td; 1187{ 1188 struct vnode *coveredvp, *fsrootvp; 1189 int error; 1190 int async_flag; 1191 int mnt_gen_r; 1192 1193 mtx_assert(&Giant, MA_OWNED); 1194 1195 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1196 mnt_gen_r = mp->mnt_gen; 1197 VI_LOCK(coveredvp); 1198 vholdl(coveredvp); 1199 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1200 vdrop(coveredvp); 1201 /* 1202 * Check for mp being unmounted while waiting for the 1203 * covered vnode lock. 1204 */ 1205 if (coveredvp->v_mountedhere != mp || 1206 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1207 VOP_UNLOCK(coveredvp, 0); 1208 return (EBUSY); 1209 } 1210 } 1211 /* 1212 * Only privileged root, or (if MNT_USER is set) the user that did the 1213 * original mount is permitted to unmount this filesystem. 1214 */ 1215 error = vfs_suser(mp, td); 1216 if (error) { 1217 if (coveredvp) 1218 VOP_UNLOCK(coveredvp, 0); 1219 return (error); 1220 } 1221 1222 MNT_ILOCK(mp); 1223 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1224 MNT_IUNLOCK(mp); 1225 if (coveredvp) 1226 VOP_UNLOCK(coveredvp, 0); 1227 return (EBUSY); 1228 } 1229 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1230 /* Allow filesystems to detect that a forced unmount is in progress. */ 1231 if (flags & MNT_FORCE) 1232 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1233 error = 0; 1234 if (mp->mnt_lockref) { 1235 if ((flags & MNT_FORCE) == 0) { 1236 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1237 MNTK_UNMOUNTF); 1238 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1239 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1240 wakeup(mp); 1241 } 1242 MNT_IUNLOCK(mp); 1243 if (coveredvp) 1244 VOP_UNLOCK(coveredvp, 0); 1245 return (EBUSY); 1246 } 1247 mp->mnt_kern_flag |= MNTK_DRAINING; 1248 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1249 "mount drain", 0); 1250 } 1251 MNT_IUNLOCK(mp); 1252 KASSERT(mp->mnt_lockref == 0, 1253 ("%s: invalid lock refcount in the drain path @ %s:%d", 1254 __func__, __FILE__, __LINE__)); 1255 KASSERT(error == 0, 1256 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1257 __func__, __FILE__, __LINE__)); 1258 vn_start_write(NULL, &mp, V_WAIT); 1259 1260 if (mp->mnt_flag & MNT_EXPUBLIC) 1261 vfs_setpublicfs(NULL, NULL, NULL); 1262 1263 vfs_msync(mp, MNT_WAIT); 1264 MNT_ILOCK(mp); 1265 async_flag = mp->mnt_flag & MNT_ASYNC; 1266 mp->mnt_flag &= ~MNT_ASYNC; 1267 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1268 MNT_IUNLOCK(mp); 1269 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1270 if (mp->mnt_syncer != NULL) 1271 vrele(mp->mnt_syncer); 1272 /* 1273 * For forced unmounts, move process cdir/rdir refs on the fs root 1274 * vnode to the covered vnode. For non-forced unmounts we want 1275 * such references to cause an EBUSY error. 1276 */ 1277 if ((flags & MNT_FORCE) && 1278 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1279 if (mp->mnt_vnodecovered != NULL) 1280 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1281 if (fsrootvp == rootvnode) { 1282 vrele(rootvnode); 1283 rootvnode = NULL; 1284 } 1285 vput(fsrootvp); 1286 } 1287 if (((mp->mnt_flag & MNT_RDONLY) || 1288 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1289 error = VFS_UNMOUNT(mp, flags); 1290 vn_finished_write(mp); 1291 /* 1292 * If we failed to flush the dirty blocks for this mount point, 1293 * undo all the cdir/rdir and rootvnode changes we made above. 1294 * Unless we failed to do so because the device is reporting that 1295 * it doesn't exist anymore. 1296 */ 1297 if (error && error != ENXIO) { 1298 if ((flags & MNT_FORCE) && 1299 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1300 if (mp->mnt_vnodecovered != NULL) 1301 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1302 if (rootvnode == NULL) { 1303 rootvnode = fsrootvp; 1304 vref(rootvnode); 1305 } 1306 vput(fsrootvp); 1307 } 1308 MNT_ILOCK(mp); 1309 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1310 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1311 MNT_IUNLOCK(mp); 1312 (void) vfs_allocate_syncvnode(mp); 1313 MNT_ILOCK(mp); 1314 } 1315 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1316 mp->mnt_flag |= async_flag; 1317 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1318 mp->mnt_kern_flag |= MNTK_ASYNC; 1319 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1320 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1321 wakeup(mp); 1322 } 1323 MNT_IUNLOCK(mp); 1324 if (coveredvp) 1325 VOP_UNLOCK(coveredvp, 0); 1326 return (error); 1327 } 1328 mtx_lock(&mountlist_mtx); 1329 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1330 mtx_unlock(&mountlist_mtx); 1331 if (coveredvp != NULL) { 1332 coveredvp->v_mountedhere = NULL; 1333 vput(coveredvp); 1334 } 1335 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1336 vfs_mount_destroy(mp); 1337 return (0); 1338} 1339 1340/* 1341 * --------------------------------------------------------------------- 1342 * Mounting of root filesystem 1343 * 1344 */ 1345 1346struct root_hold_token { 1347 const char *who; 1348 LIST_ENTRY(root_hold_token) list; 1349}; 1350 1351static LIST_HEAD(, root_hold_token) root_holds = 1352 LIST_HEAD_INITIALIZER(&root_holds); 1353 1354static int root_mount_complete; 1355 1356/* 1357 * Hold root mount. 1358 */ 1359struct root_hold_token * 1360root_mount_hold(const char *identifier) 1361{ 1362 struct root_hold_token *h; 1363 1364 if (root_mounted()) 1365 return (NULL); 1366 1367 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1368 h->who = identifier; 1369 mtx_lock(&mountlist_mtx); 1370 LIST_INSERT_HEAD(&root_holds, h, list); 1371 mtx_unlock(&mountlist_mtx); 1372 return (h); 1373} 1374 1375/* 1376 * Release root mount. 1377 */ 1378void 1379root_mount_rel(struct root_hold_token *h) 1380{ 1381 1382 if (h == NULL) 1383 return; 1384 mtx_lock(&mountlist_mtx); 1385 LIST_REMOVE(h, list); 1386 wakeup(&root_holds); 1387 mtx_unlock(&mountlist_mtx); 1388 free(h, M_DEVBUF); 1389} 1390 1391/* 1392 * Wait for all subsystems to release root mount. 1393 */ 1394static void 1395root_mount_prepare(void) 1396{ 1397 struct root_hold_token *h; 1398 struct timeval lastfail; 1399 int curfail = 0; 1400 1401 for (;;) { 1402 DROP_GIANT(); 1403 g_waitidle(); 1404 PICKUP_GIANT(); 1405 mtx_lock(&mountlist_mtx); 1406 if (LIST_EMPTY(&root_holds)) { 1407 mtx_unlock(&mountlist_mtx); 1408 break; 1409 } 1410 if (ppsratecheck(&lastfail, &curfail, 1)) { 1411 printf("Root mount waiting for:"); 1412 LIST_FOREACH(h, &root_holds, list) 1413 printf(" %s", h->who); 1414 printf("\n"); 1415 } 1416 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1417 hz); 1418 } 1419} 1420 1421/* 1422 * Root was mounted, share the good news. 1423 */ 1424static void 1425root_mount_done(void) 1426{ 1427 1428 /* Keep prison0's root in sync with the global rootvnode. */ 1429 mtx_lock(&prison0.pr_mtx); 1430 prison0.pr_root = rootvnode; 1431 vref(prison0.pr_root); 1432 mtx_unlock(&prison0.pr_mtx); 1433 /* 1434 * Use a mutex to prevent the wakeup being missed and waiting for 1435 * an extra 1 second sleep. 1436 */ 1437 mtx_lock(&mountlist_mtx); 1438 root_mount_complete = 1; 1439 wakeup(&root_mount_complete); 1440 mtx_unlock(&mountlist_mtx); 1441} 1442 1443/* 1444 * Return true if root is already mounted. 1445 */ 1446int 1447root_mounted(void) 1448{ 1449 1450 /* No mutex is acquired here because int stores are atomic. */ 1451 return (root_mount_complete); 1452} 1453 1454/* 1455 * Wait until root is mounted. 1456 */ 1457void 1458root_mount_wait(void) 1459{ 1460 1461 /* 1462 * Panic on an obvious deadlock - the function can't be called from 1463 * a thread which is doing the whole SYSINIT stuff. 1464 */ 1465 KASSERT(curthread->td_proc->p_pid != 0, 1466 ("root_mount_wait: cannot be called from the swapper thread")); 1467 mtx_lock(&mountlist_mtx); 1468 while (!root_mount_complete) { 1469 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1470 hz); 1471 } 1472 mtx_unlock(&mountlist_mtx); 1473} 1474 1475static void 1476set_rootvnode() 1477{ 1478 struct proc *p; 1479 1480 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode)) 1481 panic("Cannot find root vnode"); 1482 1483 p = curthread->td_proc; 1484 FILEDESC_XLOCK(p->p_fd); 1485 1486 if (p->p_fd->fd_cdir != NULL) 1487 vrele(p->p_fd->fd_cdir); 1488 p->p_fd->fd_cdir = rootvnode; 1489 VREF(rootvnode); 1490 1491 if (p->p_fd->fd_rdir != NULL) 1492 vrele(p->p_fd->fd_rdir); 1493 p->p_fd->fd_rdir = rootvnode; 1494 VREF(rootvnode); 1495 1496 FILEDESC_XUNLOCK(p->p_fd); 1497 1498 VOP_UNLOCK(rootvnode, 0); 1499 1500 EVENTHANDLER_INVOKE(mountroot); 1501} 1502 1503/* 1504 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1505 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1506 */ 1507 1508static void 1509devfs_first(void) 1510{ 1511 struct thread *td = curthread; 1512 struct vfsoptlist *opts; 1513 struct vfsconf *vfsp; 1514 struct mount *mp = NULL; 1515 int error; 1516 1517 vfsp = vfs_byname("devfs"); 1518 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1519 if (vfsp == NULL) 1520 return; 1521 1522 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 1523 1524 error = VFS_MOUNT(mp); 1525 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1526 if (error) 1527 return; 1528 1529 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1530 TAILQ_INIT(opts); 1531 mp->mnt_opt = opts; 1532 1533 mtx_lock(&mountlist_mtx); 1534 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1535 mtx_unlock(&mountlist_mtx); 1536 1537 set_rootvnode(); 1538 1539 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1540 if (error) 1541 printf("kern_symlink /dev -> / returns %d\n", error); 1542} 1543 1544/* 1545 * Surgically move our devfs to be mounted on /dev. 1546 */ 1547 1548static void 1549devfs_fixup(struct thread *td) 1550{ 1551 struct nameidata nd; 1552 int error; 1553 struct vnode *vp, *dvp; 1554 struct mount *mp; 1555 1556 /* Remove our devfs mount from the mountlist and purge the cache */ 1557 mtx_lock(&mountlist_mtx); 1558 mp = TAILQ_FIRST(&mountlist); 1559 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1560 mtx_unlock(&mountlist_mtx); 1561 cache_purgevfs(mp); 1562 1563 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp); 1564 VI_LOCK(dvp); 1565 dvp->v_iflag &= ~VI_MOUNT; 1566 VI_UNLOCK(dvp); 1567 dvp->v_mountedhere = NULL; 1568 1569 /* Set up the real rootvnode, and purge the cache */ 1570 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1571 set_rootvnode(); 1572 cache_purgevfs(rootvnode->v_mount); 1573 1574 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1575 error = namei(&nd); 1576 if (error) { 1577 printf("Lookup of /dev for devfs, error: %d\n", error); 1578 return; 1579 } 1580 NDFREE(&nd, NDF_ONLY_PNBUF); 1581 vp = nd.ni_vp; 1582 if (vp->v_type != VDIR) { 1583 vput(vp); 1584 } 1585 error = vinvalbuf(vp, V_SAVE, 0, 0); 1586 if (error) { 1587 vput(vp); 1588 } 1589 cache_purge(vp); 1590 mp->mnt_vnodecovered = vp; 1591 vp->v_mountedhere = mp; 1592 mtx_lock(&mountlist_mtx); 1593 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1594 mtx_unlock(&mountlist_mtx); 1595 VOP_UNLOCK(vp, 0); 1596 vput(dvp); 1597 vfs_unbusy(mp); 1598 1599 /* Unlink the no longer needed /dev/dev -> / symlink */ 1600 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1601} 1602 1603/* 1604 * Report errors during filesystem mounting. 1605 */ 1606void 1607vfs_mount_error(struct mount *mp, const char *fmt, ...) 1608{ 1609 struct vfsoptlist *moptlist = mp->mnt_optnew; 1610 va_list ap; 1611 int error, len; 1612 char *errmsg; 1613 1614 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1615 if (error || errmsg == NULL || len <= 0) 1616 return; 1617 1618 va_start(ap, fmt); 1619 vsnprintf(errmsg, (size_t)len, fmt, ap); 1620 va_end(ap); 1621} 1622 1623void 1624vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1625{ 1626 va_list ap; 1627 int error, len; 1628 char *errmsg; 1629 1630 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1631 if (error || errmsg == NULL || len <= 0) 1632 return; 1633 1634 va_start(ap, fmt); 1635 vsnprintf(errmsg, (size_t)len, fmt, ap); 1636 va_end(ap); 1637} 1638 1639/* 1640 * Find and mount the root filesystem 1641 */ 1642void 1643vfs_mountroot(void) 1644{ 1645 char *cp, *options; 1646 int error, i, asked = 0; 1647 1648 options = NULL; 1649 1650 root_mount_prepare(); 1651 1652 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1653 NULL, NULL, mount_init, mount_fini, 1654 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1655 devfs_first(); 1656 1657 /* 1658 * We are booted with instructions to prompt for the root filesystem. 1659 */ 1660 if (boothowto & RB_ASKNAME) { 1661 if (!vfs_mountroot_ask()) 1662 goto mounted; 1663 asked = 1; 1664 } 1665 1666 options = getenv("vfs.root.mountfrom.options"); 1667 1668 /* 1669 * The root filesystem information is compiled in, and we are 1670 * booted with instructions to use it. 1671 */ 1672 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1673 if (!vfs_mountroot_try(ctrootdevname, options)) 1674 goto mounted; 1675 ctrootdevname = NULL; 1676 } 1677 1678 /* 1679 * We've been given the generic "use CDROM as root" flag. This is 1680 * necessary because one media may be used in many different 1681 * devices, so we need to search for them. 1682 */ 1683 if (boothowto & RB_CDROM) { 1684 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1685 if (!vfs_mountroot_try(cdrom_rootdevnames[i], options)) 1686 goto mounted; 1687 } 1688 } 1689 1690 /* 1691 * Try to use the value read by the loader from /etc/fstab, or 1692 * supplied via some other means. This is the preferred 1693 * mechanism. 1694 */ 1695 cp = getenv("vfs.root.mountfrom"); 1696 if (cp != NULL) { 1697 error = vfs_mountroot_try(cp, options); 1698 freeenv(cp); 1699 if (!error) 1700 goto mounted; 1701 } 1702 1703 /* 1704 * Try values that may have been computed by code during boot 1705 */ 1706 if (!vfs_mountroot_try(rootdevnames[0], options)) 1707 goto mounted; 1708 if (!vfs_mountroot_try(rootdevnames[1], options)) 1709 goto mounted; 1710 1711 /* 1712 * If we (still) have a compiled-in default, try it. 1713 */ 1714 if (ctrootdevname != NULL) 1715 if (!vfs_mountroot_try(ctrootdevname, options)) 1716 goto mounted; 1717 /* 1718 * Everything so far has failed, prompt on the console if we haven't 1719 * already tried that. 1720 */ 1721 if (!asked) 1722 if (!vfs_mountroot_ask()) 1723 goto mounted; 1724 1725 panic("Root mount failed, startup aborted."); 1726 1727mounted: 1728 root_mount_done(); 1729 freeenv(options); 1730} 1731 1732static struct mntarg * 1733parse_mountroot_options(struct mntarg *ma, const char *options) 1734{ 1735 char *p; 1736 char *name, *name_arg; 1737 char *val, *val_arg; 1738 char *opts; 1739 1740 if (options == NULL || options[0] == '\0') 1741 return (ma); 1742 1743 p = opts = strdup(options, M_MOUNT); 1744 if (opts == NULL) { 1745 return (ma); 1746 } 1747 1748 while((name = strsep(&p, ",")) != NULL) { 1749 if (name[0] == '\0') 1750 break; 1751 1752 val = strchr(name, '='); 1753 if (val != NULL) { 1754 *val = '\0'; 1755 ++val; 1756 } 1757 if( strcmp(name, "rw") == 0 || 1758 strcmp(name, "noro") == 0) { 1759 /* 1760 * The first time we mount the root file system, 1761 * we need to mount 'ro', so We need to ignore 1762 * 'rw' and 'noro' mount options. 1763 */ 1764 continue; 1765 } 1766 name_arg = strdup(name, M_MOUNT); 1767 val_arg = NULL; 1768 if (val != NULL) 1769 val_arg = strdup(val, M_MOUNT); 1770 1771 ma = mount_arg(ma, name_arg, val_arg, 1772 (val_arg != NULL ? -1 : 0)); 1773 } 1774 free(opts, M_MOUNT); 1775 return (ma); 1776} 1777 1778/* 1779 * Mount (mountfrom) as the root filesystem. 1780 */ 1781static int 1782vfs_mountroot_try(const char *mountfrom, const char *options) 1783{ 1784 struct mount *mp; 1785 struct mntarg *ma; 1786 char *vfsname, *path; 1787 time_t timebase; 1788 int error; 1789 char patt[32]; 1790 char errmsg[255]; 1791 1792 vfsname = NULL; 1793 path = NULL; 1794 mp = NULL; 1795 ma = NULL; 1796 error = EINVAL; 1797 bzero(errmsg, sizeof(errmsg)); 1798 1799 if (mountfrom == NULL) 1800 return (error); /* don't complain */ 1801 printf("Trying to mount root from %s\n", mountfrom); 1802 1803 /* parse vfs name and path */ 1804 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1805 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1806 vfsname[0] = path[0] = 0; 1807 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1808 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1809 goto out; 1810 1811 if (path[0] == '\0') 1812 strcpy(path, ROOTNAME); 1813 1814 ma = mount_arg(ma, "fstype", vfsname, -1); 1815 ma = mount_arg(ma, "fspath", "/", -1); 1816 ma = mount_arg(ma, "from", path, -1); 1817 ma = mount_arg(ma, "errmsg", errmsg, sizeof(errmsg)); 1818 ma = mount_arg(ma, "ro", NULL, 0); 1819 ma = parse_mountroot_options(ma, options); 1820 error = kernel_mount(ma, MNT_ROOTFS); 1821 1822 if (error == 0) { 1823 /* 1824 * We mount devfs prior to mounting the / FS, so the first 1825 * entry will typically be devfs. 1826 */ 1827 mp = TAILQ_FIRST(&mountlist); 1828 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1829 1830 /* 1831 * Iterate over all currently mounted file systems and use 1832 * the time stamp found to check and/or initialize the RTC. 1833 * Typically devfs has no time stamp and the only other FS 1834 * is the actual / FS. 1835 * Call inittodr() only once and pass it the largest of the 1836 * timestamps we encounter. 1837 */ 1838 timebase = 0; 1839 do { 1840 if (mp->mnt_time > timebase) 1841 timebase = mp->mnt_time; 1842 mp = TAILQ_NEXT(mp, mnt_list); 1843 } while (mp != NULL); 1844 inittodr(timebase); 1845 1846 devfs_fixup(curthread); 1847 } 1848 1849 if (error != 0 ) { 1850 printf("ROOT MOUNT ERROR: %s\n", errmsg); 1851 printf("If you have invalid mount options, reboot, and "); 1852 printf("first try the following from\n"); 1853 printf("the loader prompt:\n\n"); 1854 printf(" set vfs.root.mountfrom.options=rw\n\n"); 1855 printf("and then remove invalid mount options from "); 1856 printf("/etc/fstab.\n\n"); 1857 } 1858out: 1859 free(path, M_MOUNT); 1860 free(vfsname, M_MOUNT); 1861 return (error); 1862} 1863 1864/* 1865 * --------------------------------------------------------------------- 1866 * Interactive root filesystem selection code. 1867 */ 1868 1869static int 1870vfs_mountroot_ask(void) 1871{ 1872 char name[128]; 1873 char *mountfrom; 1874 char *options; 1875 1876 for(;;) { 1877 printf("Loader variables:\n"); 1878 printf("vfs.root.mountfrom="); 1879 mountfrom = getenv("vfs.root.mountfrom"); 1880 if (mountfrom != NULL) { 1881 printf("%s", mountfrom); 1882 } 1883 printf("\n"); 1884 printf("vfs.root.mountfrom.options="); 1885 options = getenv("vfs.root.mountfrom.options"); 1886 if (options != NULL) { 1887 printf("%s", options); 1888 } 1889 printf("\n"); 1890 freeenv(mountfrom); 1891 freeenv(options); 1892 printf("\nManual root filesystem specification:\n"); 1893 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1894 printf(" eg. ufs:/dev/da0s1a\n"); 1895 printf(" eg. cd9660:/dev/acd0\n"); 1896 printf(" This is equivalent to: "); 1897 printf("mount -t cd9660 /dev/acd0 /\n"); 1898 printf("\n"); 1899 printf(" ? List valid disk boot devices\n"); 1900 printf(" <empty line> Abort manual input\n"); 1901 printf("\nmountroot> "); 1902 gets(name, sizeof(name), 1); 1903 if (name[0] == '\0') 1904 return (1); 1905 if (name[0] == '?') { 1906 printf("\nList of GEOM managed disk devices:\n "); 1907 g_dev_print(); 1908 continue; 1909 } 1910 if (!vfs_mountroot_try(name, NULL)) 1911 return (0); 1912 } 1913} 1914 1915/* 1916 * --------------------------------------------------------------------- 1917 * Functions for querying mount options/arguments from filesystems. 1918 */ 1919 1920/* 1921 * Check that no unknown options are given 1922 */ 1923int 1924vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1925{ 1926 struct vfsopt *opt; 1927 char errmsg[255]; 1928 const char **t, *p, *q; 1929 int ret = 0; 1930 1931 TAILQ_FOREACH(opt, opts, link) { 1932 p = opt->name; 1933 q = NULL; 1934 if (p[0] == 'n' && p[1] == 'o') 1935 q = p + 2; 1936 for(t = global_opts; *t != NULL; t++) { 1937 if (strcmp(*t, p) == 0) 1938 break; 1939 if (q != NULL) { 1940 if (strcmp(*t, q) == 0) 1941 break; 1942 } 1943 } 1944 if (*t != NULL) 1945 continue; 1946 for(t = legal; *t != NULL; t++) { 1947 if (strcmp(*t, p) == 0) 1948 break; 1949 if (q != NULL) { 1950 if (strcmp(*t, q) == 0) 1951 break; 1952 } 1953 } 1954 if (*t != NULL) 1955 continue; 1956 snprintf(errmsg, sizeof(errmsg), 1957 "mount option <%s> is unknown", p); 1958 printf("%s\n", errmsg); 1959 ret = EINVAL; 1960 } 1961 if (ret != 0) { 1962 TAILQ_FOREACH(opt, opts, link) { 1963 if (strcmp(opt->name, "errmsg") == 0) { 1964 strncpy((char *)opt->value, errmsg, opt->len); 1965 } 1966 } 1967 } 1968 return (ret); 1969} 1970 1971/* 1972 * Get a mount option by its name. 1973 * 1974 * Return 0 if the option was found, ENOENT otherwise. 1975 * If len is non-NULL it will be filled with the length 1976 * of the option. If buf is non-NULL, it will be filled 1977 * with the address of the option. 1978 */ 1979int 1980vfs_getopt(opts, name, buf, len) 1981 struct vfsoptlist *opts; 1982 const char *name; 1983 void **buf; 1984 int *len; 1985{ 1986 struct vfsopt *opt; 1987 1988 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1989 1990 TAILQ_FOREACH(opt, opts, link) { 1991 if (strcmp(name, opt->name) == 0) { 1992 opt->seen = 1; 1993 if (len != NULL) 1994 *len = opt->len; 1995 if (buf != NULL) 1996 *buf = opt->value; 1997 return (0); 1998 } 1999 } 2000 return (ENOENT); 2001} 2002 2003int 2004vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2005{ 2006 struct vfsopt *opt; 2007 2008 if (opts == NULL) 2009 return (-1); 2010 2011 TAILQ_FOREACH(opt, opts, link) { 2012 if (strcmp(name, opt->name) == 0) { 2013 opt->seen = 1; 2014 return (opt->pos); 2015 } 2016 } 2017 return (-1); 2018} 2019 2020char * 2021vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2022{ 2023 struct vfsopt *opt; 2024 2025 *error = 0; 2026 TAILQ_FOREACH(opt, opts, link) { 2027 if (strcmp(name, opt->name) != 0) 2028 continue; 2029 opt->seen = 1; 2030 if (opt->len == 0 || 2031 ((char *)opt->value)[opt->len - 1] != '\0') { 2032 *error = EINVAL; 2033 return (NULL); 2034 } 2035 return (opt->value); 2036 } 2037 *error = ENOENT; 2038 return (NULL); 2039} 2040 2041int 2042vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 2043{ 2044 struct vfsopt *opt; 2045 2046 TAILQ_FOREACH(opt, opts, link) { 2047 if (strcmp(name, opt->name) == 0) { 2048 opt->seen = 1; 2049 if (w != NULL) 2050 *w |= val; 2051 return (1); 2052 } 2053 } 2054 if (w != NULL) 2055 *w &= ~val; 2056 return (0); 2057} 2058 2059int 2060vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2061{ 2062 va_list ap; 2063 struct vfsopt *opt; 2064 int ret; 2065 2066 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2067 2068 TAILQ_FOREACH(opt, opts, link) { 2069 if (strcmp(name, opt->name) != 0) 2070 continue; 2071 opt->seen = 1; 2072 if (opt->len == 0 || opt->value == NULL) 2073 return (0); 2074 if (((char *)opt->value)[opt->len - 1] != '\0') 2075 return (0); 2076 va_start(ap, fmt); 2077 ret = vsscanf(opt->value, fmt, ap); 2078 va_end(ap); 2079 return (ret); 2080 } 2081 return (0); 2082} 2083 2084int 2085vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2086{ 2087 struct vfsopt *opt; 2088 2089 TAILQ_FOREACH(opt, opts, link) { 2090 if (strcmp(name, opt->name) != 0) 2091 continue; 2092 opt->seen = 1; 2093 if (opt->value == NULL) 2094 opt->len = len; 2095 else { 2096 if (opt->len != len) 2097 return (EINVAL); 2098 bcopy(value, opt->value, len); 2099 } 2100 return (0); 2101 } 2102 return (ENOENT); 2103} 2104 2105int 2106vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2107{ 2108 struct vfsopt *opt; 2109 2110 TAILQ_FOREACH(opt, opts, link) { 2111 if (strcmp(name, opt->name) != 0) 2112 continue; 2113 opt->seen = 1; 2114 if (opt->value == NULL) 2115 opt->len = len; 2116 else { 2117 if (opt->len < len) 2118 return (EINVAL); 2119 opt->len = len; 2120 bcopy(value, opt->value, len); 2121 } 2122 return (0); 2123 } 2124 return (ENOENT); 2125} 2126 2127int 2128vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2129{ 2130 struct vfsopt *opt; 2131 2132 TAILQ_FOREACH(opt, opts, link) { 2133 if (strcmp(name, opt->name) != 0) 2134 continue; 2135 opt->seen = 1; 2136 if (opt->value == NULL) 2137 opt->len = strlen(value) + 1; 2138 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2139 return (EINVAL); 2140 return (0); 2141 } 2142 return (ENOENT); 2143} 2144 2145/* 2146 * Find and copy a mount option. 2147 * 2148 * The size of the buffer has to be specified 2149 * in len, if it is not the same length as the 2150 * mount option, EINVAL is returned. 2151 * Returns ENOENT if the option is not found. 2152 */ 2153int 2154vfs_copyopt(opts, name, dest, len) 2155 struct vfsoptlist *opts; 2156 const char *name; 2157 void *dest; 2158 int len; 2159{ 2160 struct vfsopt *opt; 2161 2162 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2163 2164 TAILQ_FOREACH(opt, opts, link) { 2165 if (strcmp(name, opt->name) == 0) { 2166 opt->seen = 1; 2167 if (len != opt->len) 2168 return (EINVAL); 2169 bcopy(opt->value, dest, opt->len); 2170 return (0); 2171 } 2172 } 2173 return (ENOENT); 2174} 2175 2176/* 2177 * This is a helper function for filesystems to traverse their 2178 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2179 */ 2180 2181struct vnode * 2182__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2183{ 2184 struct vnode *vp; 2185 2186 mtx_assert(MNT_MTX(mp), MA_OWNED); 2187 2188 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2189 if ((*mvp)->v_yield++ == 500) { 2190 MNT_IUNLOCK(mp); 2191 (*mvp)->v_yield = 0; 2192 uio_yield(); 2193 MNT_ILOCK(mp); 2194 } 2195 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2196 while (vp != NULL && vp->v_type == VMARKER) 2197 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2198 2199 /* Check if we are done */ 2200 if (vp == NULL) { 2201 __mnt_vnode_markerfree(mvp, mp); 2202 return (NULL); 2203 } 2204 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2205 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2206 return (vp); 2207} 2208 2209struct vnode * 2210__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2211{ 2212 struct vnode *vp; 2213 2214 mtx_assert(MNT_MTX(mp), MA_OWNED); 2215 2216 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2217 while (vp != NULL && vp->v_type == VMARKER) 2218 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2219 2220 /* Check if we are done */ 2221 if (vp == NULL) { 2222 *mvp = NULL; 2223 return (NULL); 2224 } 2225 MNT_REF(mp); 2226 MNT_IUNLOCK(mp); 2227 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2228 M_VNODE_MARKER, 2229 M_WAITOK | M_ZERO); 2230 MNT_ILOCK(mp); 2231 (*mvp)->v_type = VMARKER; 2232 2233 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2234 while (vp != NULL && vp->v_type == VMARKER) 2235 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2236 2237 /* Check if we are done */ 2238 if (vp == NULL) { 2239 MNT_IUNLOCK(mp); 2240 free(*mvp, M_VNODE_MARKER); 2241 MNT_ILOCK(mp); 2242 *mvp = NULL; 2243 MNT_REL(mp); 2244 return (NULL); 2245 } 2246 (*mvp)->v_mount = mp; 2247 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2248 return (vp); 2249} 2250 2251 2252void 2253__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2254{ 2255 2256 if (*mvp == NULL) 2257 return; 2258 2259 mtx_assert(MNT_MTX(mp), MA_OWNED); 2260 2261 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2262 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2263 MNT_IUNLOCK(mp); 2264 free(*mvp, M_VNODE_MARKER); 2265 MNT_ILOCK(mp); 2266 *mvp = NULL; 2267 MNT_REL(mp); 2268} 2269 2270 2271int 2272__vfs_statfs(struct mount *mp, struct statfs *sbp) 2273{ 2274 int error; 2275 2276 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 2277 if (sbp != &mp->mnt_stat) 2278 *sbp = mp->mnt_stat; 2279 return (error); 2280} 2281 2282void 2283vfs_mountedfrom(struct mount *mp, const char *from) 2284{ 2285 2286 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2287 strlcpy(mp->mnt_stat.f_mntfromname, from, 2288 sizeof mp->mnt_stat.f_mntfromname); 2289} 2290 2291/* 2292 * --------------------------------------------------------------------- 2293 * This is the api for building mount args and mounting filesystems from 2294 * inside the kernel. 2295 * 2296 * The API works by accumulation of individual args. First error is 2297 * latched. 2298 * 2299 * XXX: should be documented in new manpage kernel_mount(9) 2300 */ 2301 2302/* A memory allocation which must be freed when we are done */ 2303struct mntaarg { 2304 SLIST_ENTRY(mntaarg) next; 2305}; 2306 2307/* The header for the mount arguments */ 2308struct mntarg { 2309 struct iovec *v; 2310 int len; 2311 int error; 2312 SLIST_HEAD(, mntaarg) list; 2313}; 2314 2315/* 2316 * Add a boolean argument. 2317 * 2318 * flag is the boolean value. 2319 * name must start with "no". 2320 */ 2321struct mntarg * 2322mount_argb(struct mntarg *ma, int flag, const char *name) 2323{ 2324 2325 KASSERT(name[0] == 'n' && name[1] == 'o', 2326 ("mount_argb(...,%s): name must start with 'no'", name)); 2327 2328 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2329} 2330 2331/* 2332 * Add an argument printf style 2333 */ 2334struct mntarg * 2335mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2336{ 2337 va_list ap; 2338 struct mntaarg *maa; 2339 struct sbuf *sb; 2340 int len; 2341 2342 if (ma == NULL) { 2343 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2344 SLIST_INIT(&ma->list); 2345 } 2346 if (ma->error) 2347 return (ma); 2348 2349 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2350 M_MOUNT, M_WAITOK); 2351 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2352 ma->v[ma->len].iov_len = strlen(name) + 1; 2353 ma->len++; 2354 2355 sb = sbuf_new_auto(); 2356 va_start(ap, fmt); 2357 sbuf_vprintf(sb, fmt, ap); 2358 va_end(ap); 2359 sbuf_finish(sb); 2360 len = sbuf_len(sb) + 1; 2361 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2362 SLIST_INSERT_HEAD(&ma->list, maa, next); 2363 bcopy(sbuf_data(sb), maa + 1, len); 2364 sbuf_delete(sb); 2365 2366 ma->v[ma->len].iov_base = maa + 1; 2367 ma->v[ma->len].iov_len = len; 2368 ma->len++; 2369 2370 return (ma); 2371} 2372 2373/* 2374 * Add an argument which is a userland string. 2375 */ 2376struct mntarg * 2377mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2378{ 2379 struct mntaarg *maa; 2380 char *tbuf; 2381 2382 if (val == NULL) 2383 return (ma); 2384 if (ma == NULL) { 2385 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2386 SLIST_INIT(&ma->list); 2387 } 2388 if (ma->error) 2389 return (ma); 2390 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2391 SLIST_INSERT_HEAD(&ma->list, maa, next); 2392 tbuf = (void *)(maa + 1); 2393 ma->error = copyinstr(val, tbuf, len, NULL); 2394 return (mount_arg(ma, name, tbuf, -1)); 2395} 2396 2397/* 2398 * Plain argument. 2399 * 2400 * If length is -1, treat value as a C string. 2401 */ 2402struct mntarg * 2403mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2404{ 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 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2420 if (len < 0) 2421 ma->v[ma->len].iov_len = strlen(val) + 1; 2422 else 2423 ma->v[ma->len].iov_len = len; 2424 ma->len++; 2425 return (ma); 2426} 2427 2428/* 2429 * Free a mntarg structure 2430 */ 2431static void 2432free_mntarg(struct mntarg *ma) 2433{ 2434 struct mntaarg *maa; 2435 2436 while (!SLIST_EMPTY(&ma->list)) { 2437 maa = SLIST_FIRST(&ma->list); 2438 SLIST_REMOVE_HEAD(&ma->list, next); 2439 free(maa, M_MOUNT); 2440 } 2441 free(ma->v, M_MOUNT); 2442 free(ma, M_MOUNT); 2443} 2444 2445/* 2446 * Mount a filesystem 2447 */ 2448int 2449kernel_mount(struct mntarg *ma, int flags) 2450{ 2451 struct uio auio; 2452 int error; 2453 2454 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2455 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2456 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2457 2458 auio.uio_iov = ma->v; 2459 auio.uio_iovcnt = ma->len; 2460 auio.uio_segflg = UIO_SYSSPACE; 2461 2462 error = ma->error; 2463 if (!error) 2464 error = vfs_donmount(curthread, flags, &auio); 2465 free_mntarg(ma); 2466 return (error); 2467} 2468 2469/* 2470 * A printflike function to mount a filesystem. 2471 */ 2472int 2473kernel_vmount(int flags, ...) 2474{ 2475 struct mntarg *ma = NULL; 2476 va_list ap; 2477 const char *cp; 2478 const void *vp; 2479 int error; 2480 2481 va_start(ap, flags); 2482 for (;;) { 2483 cp = va_arg(ap, const char *); 2484 if (cp == NULL) 2485 break; 2486 vp = va_arg(ap, const void *); 2487 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2488 } 2489 va_end(ap); 2490 2491 error = kernel_mount(ma, flags); 2492 return (error); 2493} 2494