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("$FreeBSD$"); 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 75 static void set_rootvnode(void); 76 static int vfs_domount(struct thread *td, const char *fstype, 77 char *fspath, int fsflags, void *fsdata); 78 static int vfs_mountroot_ask(void); 79 static int vfs_mountroot_try(const char *mountfrom, const char *options); 80 static void free_mntarg(struct mntarg *ma); 81 82 static int usermount = 0; 83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 84 "Unprivileged users may mount and unmount file systems"); 85 86 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 87 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 88 static uma_zone_t mount_zone; 89 90 /* List of mounted filesystems. */ 91 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 92 93 /* For any iteration/modification of mountlist */ 94 struct mtx mountlist_mtx; 95 MTX_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 */ 101 struct 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 */ 125 static 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 */ 139 static char *cdrom_rootdevnames[] = { 140 "cd9660:cd0", 141 "cd9660:acd0", 142 NULL 143 }; 144 145 /* legacy find-root code */ 146 char *rootdevnames[2] = {NULL, NULL}; 147 #ifndef ROOTDEVNAME 148 # define ROOTDEVNAME NULL 149 #endif 150 static const char *ctrootdevname = ROOTDEVNAME; 151 152 /* 153 * --------------------------------------------------------------------- 154 * Functions for building and sanitizing the mount options 155 */ 156 157 /* Remove one mount option. */ 158 static void 159 vfs_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. */ 170 void 171 vfs_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 } 181 182 void 183 vfs_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 */ 198 static int 199 vfs_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 */ 231 static void 232 vfs_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 */ 253 int 254 vfs_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); 324 bad: 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 */ 340 static void 341 vfs_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); 374 next: 375 continue; 376 } 377 } 378 379 /* 380 * Mount a filesystem. 381 */ 382 int 383 nmount(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 */ 434 435 void 436 vfs_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 } 444 445 void 446 vfs_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 } 454 455 static int 456 mount_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 } 465 466 static void 467 mount_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 */ 479 struct mount * 480 vfs_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 */ 514 void 515 vfs_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 } 554 555 int 556 vfs_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 mtx_lock(&Giant); 708 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 709 mtx_unlock(&Giant); 710 bail: 711 /* copyout the errmsg */ 712 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 713 && errmsg_len > 0 && errmsg != NULL) { 714 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 715 bcopy(errmsg, 716 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 717 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 718 } else { 719 copyout(errmsg, 720 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 721 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 722 } 723 } 724 725 if (error != 0) 726 vfs_freeopts(optlist); 727 return (error); 728 } 729 730 /* 731 * Old mount API. 732 */ 733 #ifndef _SYS_SYSPROTO_H_ 734 struct mount_args { 735 char *type; 736 char *path; 737 int flags; 738 caddr_t data; 739 }; 740 #endif 741 /* ARGSUSED */ 742 int 743 mount(td, uap) 744 struct thread *td; 745 struct mount_args /* { 746 char *type; 747 char *path; 748 int flags; 749 caddr_t data; 750 } */ *uap; 751 { 752 char *fstype; 753 struct vfsconf *vfsp = NULL; 754 struct mntarg *ma = NULL; 755 int error; 756 757 AUDIT_ARG_FFLAGS(uap->flags); 758 759 /* 760 * Filter out MNT_ROOTFS. We do not want clients of mount() in 761 * userspace to set this flag, but we must filter it out if we want 762 * MNT_UPDATE on the root file system to work. 763 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 764 */ 765 uap->flags &= ~MNT_ROOTFS; 766 767 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 768 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 769 if (error) { 770 free(fstype, M_TEMP); 771 return (error); 772 } 773 774 AUDIT_ARG_TEXT(fstype); 775 mtx_lock(&Giant); 776 vfsp = vfs_byname_kld(fstype, td, &error); 777 free(fstype, M_TEMP); 778 if (vfsp == NULL) { 779 mtx_unlock(&Giant); 780 return (ENOENT); 781 } 782 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 783 mtx_unlock(&Giant); 784 return (EOPNOTSUPP); 785 } 786 787 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 788 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 789 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 790 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 791 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 792 793 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags); 794 mtx_unlock(&Giant); 795 return (error); 796 } 797 798 799 /* 800 * vfs_domount(): actually attempt a filesystem mount. 801 */ 802 static int 803 vfs_domount( 804 struct thread *td, /* Calling thread. */ 805 const char *fstype, /* Filesystem type. */ 806 char *fspath, /* Mount path. */ 807 int fsflags, /* Flags common to all filesystems. */ 808 void *fsdata /* Options local to the filesystem. */ 809 ) 810 { 811 struct vnode *vp; 812 struct mount *mp; 813 struct vfsconf *vfsp; 814 struct oexport_args oexport; 815 struct export_args export; 816 int error, flag = 0; 817 struct vattr va; 818 struct nameidata nd; 819 820 mtx_assert(&Giant, MA_OWNED); 821 /* 822 * Be ultra-paranoid about making sure the type and fspath 823 * variables will fit in our mp buffers, including the 824 * terminating NUL. 825 */ 826 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 827 return (ENAMETOOLONG); 828 829 if (jailed(td->td_ucred) || usermount == 0) { 830 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 831 return (error); 832 } 833 834 /* 835 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 836 */ 837 if (fsflags & MNT_EXPORTED) { 838 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 839 if (error) 840 return (error); 841 } 842 if (fsflags & MNT_SUIDDIR) { 843 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 844 if (error) 845 return (error); 846 } 847 /* 848 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 849 */ 850 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 851 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 852 fsflags |= MNT_NOSUID | MNT_USER; 853 } 854 855 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 856 vfsp = NULL; 857 if ((fsflags & MNT_UPDATE) == 0) { 858 /* Don't try to load KLDs if we're mounting the root. */ 859 if (fsflags & MNT_ROOTFS) 860 vfsp = vfs_byname(fstype); 861 else 862 vfsp = vfs_byname_kld(fstype, td, &error); 863 if (vfsp == NULL) 864 return (ENODEV); 865 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 866 return (EPERM); 867 } 868 /* 869 * Get vnode to be covered 870 */ 871 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 872 fspath, td); 873 if ((error = namei(&nd)) != 0) 874 return (error); 875 NDFREE(&nd, NDF_ONLY_PNBUF); 876 vp = nd.ni_vp; 877 if (fsflags & MNT_UPDATE) { 878 if ((vp->v_vflag & VV_ROOT) == 0) { 879 vput(vp); 880 return (EINVAL); 881 } 882 mp = vp->v_mount; 883 MNT_ILOCK(mp); 884 flag = mp->mnt_flag; 885 /* 886 * We only allow the filesystem to be reloaded if it 887 * is currently mounted read-only. 888 */ 889 if ((fsflags & MNT_RELOAD) && 890 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 891 MNT_IUNLOCK(mp); 892 vput(vp); 893 return (EOPNOTSUPP); /* Needs translation */ 894 } 895 MNT_IUNLOCK(mp); 896 /* 897 * Only privileged root, or (if MNT_USER is set) the user that 898 * did the original mount is permitted to update it. 899 */ 900 error = vfs_suser(mp, td); 901 if (error) { 902 vput(vp); 903 return (error); 904 } 905 if (vfs_busy(mp, MBF_NOWAIT)) { 906 vput(vp); 907 return (EBUSY); 908 } 909 VI_LOCK(vp); 910 if ((vp->v_iflag & VI_MOUNT) != 0 || 911 vp->v_mountedhere != NULL) { 912 VI_UNLOCK(vp); 913 vfs_unbusy(mp); 914 vput(vp); 915 return (EBUSY); 916 } 917 vp->v_iflag |= VI_MOUNT; 918 VI_UNLOCK(vp); 919 MNT_ILOCK(mp); 920 mp->mnt_flag |= fsflags & 921 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 922 MNT_IUNLOCK(mp); 923 VOP_UNLOCK(vp, 0); 924 mp->mnt_optnew = fsdata; 925 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 926 } else { 927 /* 928 * If the user is not root, ensure that they own the directory 929 * onto which we are attempting to mount. 930 */ 931 error = VOP_GETATTR(vp, &va, td->td_ucred); 932 if (error) { 933 vput(vp); 934 return (error); 935 } 936 if (va.va_uid != td->td_ucred->cr_uid) { 937 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 938 0); 939 if (error) { 940 vput(vp); 941 return (error); 942 } 943 } 944 error = vinvalbuf(vp, V_SAVE, 0, 0); 945 if (error != 0) { 946 vput(vp); 947 return (error); 948 } 949 if (vp->v_type != VDIR) { 950 vput(vp); 951 return (ENOTDIR); 952 } 953 VI_LOCK(vp); 954 if ((vp->v_iflag & VI_MOUNT) != 0 || 955 vp->v_mountedhere != NULL) { 956 VI_UNLOCK(vp); 957 vput(vp); 958 return (EBUSY); 959 } 960 vp->v_iflag |= VI_MOUNT; 961 VI_UNLOCK(vp); 962 VOP_UNLOCK(vp, 0); 963 964 /* 965 * Allocate and initialize the filesystem. 966 */ 967 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 968 969 /* XXXMAC: pass to vfs_mount_alloc? */ 970 mp->mnt_optnew = fsdata; 971 } 972 973 /* 974 * Set the mount level flags. 975 */ 976 MNT_ILOCK(mp); 977 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 978 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 979 MNT_RDONLY)); 980 if ((mp->mnt_flag & MNT_ASYNC) == 0) 981 mp->mnt_kern_flag &= ~MNTK_ASYNC; 982 MNT_IUNLOCK(mp); 983 /* 984 * Mount the filesystem. 985 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 986 * get. No freeing of cn_pnbuf. 987 */ 988 error = VFS_MOUNT(mp); 989 990 /* 991 * Process the export option only if we are 992 * updating mount options. 993 */ 994 if (!error && (fsflags & MNT_UPDATE)) { 995 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 996 sizeof(export)) == 0) 997 error = vfs_export(mp, &export); 998 else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 999 sizeof(oexport)) == 0) { 1000 export.ex_flags = oexport.ex_flags; 1001 export.ex_root = oexport.ex_root; 1002 export.ex_anon = oexport.ex_anon; 1003 export.ex_addr = oexport.ex_addr; 1004 export.ex_addrlen = oexport.ex_addrlen; 1005 export.ex_mask = oexport.ex_mask; 1006 export.ex_masklen = oexport.ex_masklen; 1007 export.ex_indexfile = oexport.ex_indexfile; 1008 export.ex_numsecflavors = 0; 1009 error = vfs_export(mp, &export); 1010 } 1011 } 1012 1013 if (!error) { 1014 if (mp->mnt_opt != NULL) 1015 vfs_freeopts(mp->mnt_opt); 1016 mp->mnt_opt = mp->mnt_optnew; 1017 (void)VFS_STATFS(mp, &mp->mnt_stat); 1018 } 1019 /* 1020 * Prevent external consumers of mount options from reading 1021 * mnt_optnew. 1022 */ 1023 mp->mnt_optnew = NULL; 1024 if (mp->mnt_flag & MNT_UPDATE) { 1025 MNT_ILOCK(mp); 1026 if (error) 1027 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1028 (flag & ~MNT_QUOTA); 1029 else 1030 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1031 MNT_FORCE | MNT_SNAPSHOT); 1032 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1033 mp->mnt_kern_flag |= MNTK_ASYNC; 1034 else 1035 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1036 MNT_IUNLOCK(mp); 1037 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1038 if (mp->mnt_syncer == NULL) 1039 error = vfs_allocate_syncvnode(mp); 1040 } else { 1041 if (mp->mnt_syncer != NULL) 1042 vrele(mp->mnt_syncer); 1043 mp->mnt_syncer = NULL; 1044 } 1045 vfs_unbusy(mp); 1046 VI_LOCK(vp); 1047 vp->v_iflag &= ~VI_MOUNT; 1048 VI_UNLOCK(vp); 1049 vrele(vp); 1050 return (error); 1051 } 1052 MNT_ILOCK(mp); 1053 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1054 mp->mnt_kern_flag |= MNTK_ASYNC; 1055 else 1056 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1057 MNT_IUNLOCK(mp); 1058 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1059 /* 1060 * Put the new filesystem on the mount list after root. 1061 */ 1062 cache_purge(vp); 1063 VI_LOCK(vp); 1064 vp->v_iflag &= ~VI_MOUNT; 1065 VI_UNLOCK(vp); 1066 if (!error) { 1067 struct vnode *newdp; 1068 1069 vp->v_mountedhere = mp; 1070 mtx_lock(&mountlist_mtx); 1071 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1072 mtx_unlock(&mountlist_mtx); 1073 vfs_event_signal(NULL, VQ_MOUNT, 0); 1074 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 1075 panic("mount: lost mount"); 1076 VOP_UNLOCK(newdp, 0); 1077 VOP_UNLOCK(vp, 0); 1078 mountcheckdirs(vp, newdp); 1079 vrele(newdp); 1080 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1081 error = vfs_allocate_syncvnode(mp); 1082 vfs_unbusy(mp); 1083 if (error) 1084 vrele(vp); 1085 } else { 1086 vfs_unbusy(mp); 1087 vfs_mount_destroy(mp); 1088 vput(vp); 1089 } 1090 return (error); 1091 } 1092 1093 /* 1094 * Unmount a filesystem. 1095 * 1096 * Note: unmount takes a path to the vnode mounted on as argument, not 1097 * special file (as before). 1098 */ 1099 #ifndef _SYS_SYSPROTO_H_ 1100 struct unmount_args { 1101 char *path; 1102 int flags; 1103 }; 1104 #endif 1105 /* ARGSUSED */ 1106 int 1107 unmount(td, uap) 1108 struct thread *td; 1109 register struct unmount_args /* { 1110 char *path; 1111 int flags; 1112 } */ *uap; 1113 { 1114 struct mount *mp; 1115 char *pathbuf; 1116 int error, id0, id1; 1117 1118 AUDIT_ARG_VALUE(uap->flags); 1119 if (jailed(td->td_ucred) || usermount == 0) { 1120 error = priv_check(td, PRIV_VFS_UNMOUNT); 1121 if (error) 1122 return (error); 1123 } 1124 1125 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1126 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1127 if (error) { 1128 free(pathbuf, M_TEMP); 1129 return (error); 1130 } 1131 mtx_lock(&Giant); 1132 if (uap->flags & MNT_BYFSID) { 1133 AUDIT_ARG_TEXT(pathbuf); 1134 /* Decode the filesystem ID. */ 1135 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1136 mtx_unlock(&Giant); 1137 free(pathbuf, M_TEMP); 1138 return (EINVAL); 1139 } 1140 1141 mtx_lock(&mountlist_mtx); 1142 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1143 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1144 mp->mnt_stat.f_fsid.val[1] == id1) 1145 break; 1146 } 1147 mtx_unlock(&mountlist_mtx); 1148 } else { 1149 AUDIT_ARG_UPATH1(td, pathbuf); 1150 mtx_lock(&mountlist_mtx); 1151 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1152 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1153 break; 1154 } 1155 mtx_unlock(&mountlist_mtx); 1156 } 1157 free(pathbuf, M_TEMP); 1158 if (mp == NULL) { 1159 /* 1160 * Previously we returned ENOENT for a nonexistent path and 1161 * EINVAL for a non-mountpoint. We cannot tell these apart 1162 * now, so in the !MNT_BYFSID case return the more likely 1163 * EINVAL for compatibility. 1164 */ 1165 mtx_unlock(&Giant); 1166 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1167 } 1168 1169 /* 1170 * Don't allow unmounting the root filesystem. 1171 */ 1172 if (mp->mnt_flag & MNT_ROOTFS) { 1173 mtx_unlock(&Giant); 1174 return (EINVAL); 1175 } 1176 error = dounmount(mp, uap->flags, td); 1177 mtx_unlock(&Giant); 1178 return (error); 1179 } 1180 1181 /* 1182 * Do the actual filesystem unmount. 1183 */ 1184 int 1185 dounmount(mp, flags, td) 1186 struct mount *mp; 1187 int flags; 1188 struct thread *td; 1189 { 1190 struct vnode *coveredvp, *fsrootvp; 1191 int error; 1192 int async_flag; 1193 int mnt_gen_r; 1194 1195 mtx_assert(&Giant, MA_OWNED); 1196 1197 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1198 mnt_gen_r = mp->mnt_gen; 1199 VI_LOCK(coveredvp); 1200 vholdl(coveredvp); 1201 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1202 vdrop(coveredvp); 1203 /* 1204 * Check for mp being unmounted while waiting for the 1205 * covered vnode lock. 1206 */ 1207 if (coveredvp->v_mountedhere != mp || 1208 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1209 VOP_UNLOCK(coveredvp, 0); 1210 return (EBUSY); 1211 } 1212 } 1213 /* 1214 * Only privileged root, or (if MNT_USER is set) the user that did the 1215 * original mount is permitted to unmount this filesystem. 1216 */ 1217 error = vfs_suser(mp, td); 1218 if (error) { 1219 if (coveredvp) 1220 VOP_UNLOCK(coveredvp, 0); 1221 return (error); 1222 } 1223 1224 MNT_ILOCK(mp); 1225 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1226 MNT_IUNLOCK(mp); 1227 if (coveredvp) 1228 VOP_UNLOCK(coveredvp, 0); 1229 return (EBUSY); 1230 } 1231 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1232 /* Allow filesystems to detect that a forced unmount is in progress. */ 1233 if (flags & MNT_FORCE) 1234 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1235 error = 0; 1236 if (mp->mnt_lockref) { 1237 if ((flags & MNT_FORCE) == 0) { 1238 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1239 MNTK_UNMOUNTF); 1240 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1241 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1242 wakeup(mp); 1243 } 1244 MNT_IUNLOCK(mp); 1245 if (coveredvp) 1246 VOP_UNLOCK(coveredvp, 0); 1247 return (EBUSY); 1248 } 1249 mp->mnt_kern_flag |= MNTK_DRAINING; 1250 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1251 "mount drain", 0); 1252 } 1253 MNT_IUNLOCK(mp); 1254 KASSERT(mp->mnt_lockref == 0, 1255 ("%s: invalid lock refcount in the drain path @ %s:%d", 1256 __func__, __FILE__, __LINE__)); 1257 KASSERT(error == 0, 1258 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1259 __func__, __FILE__, __LINE__)); 1260 vn_start_write(NULL, &mp, V_WAIT); 1261 1262 if (mp->mnt_flag & MNT_EXPUBLIC) 1263 vfs_setpublicfs(NULL, NULL, NULL); 1264 1265 vfs_msync(mp, MNT_WAIT); 1266 MNT_ILOCK(mp); 1267 async_flag = mp->mnt_flag & MNT_ASYNC; 1268 mp->mnt_flag &= ~MNT_ASYNC; 1269 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1270 MNT_IUNLOCK(mp); 1271 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1272 if (mp->mnt_syncer != NULL) 1273 vrele(mp->mnt_syncer); 1274 /* 1275 * For forced unmounts, move process cdir/rdir refs on the fs root 1276 * vnode to the covered vnode. For non-forced unmounts we want 1277 * such references to cause an EBUSY error. 1278 */ 1279 if ((flags & MNT_FORCE) && 1280 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1281 if (mp->mnt_vnodecovered != NULL) 1282 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1283 if (fsrootvp == rootvnode) { 1284 vrele(rootvnode); 1285 rootvnode = NULL; 1286 } 1287 vput(fsrootvp); 1288 } 1289 if (((mp->mnt_flag & MNT_RDONLY) || 1290 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1291 error = VFS_UNMOUNT(mp, flags); 1292 vn_finished_write(mp); 1293 /* 1294 * If we failed to flush the dirty blocks for this mount point, 1295 * undo all the cdir/rdir and rootvnode changes we made above. 1296 * Unless we failed to do so because the device is reporting that 1297 * it doesn't exist anymore. 1298 */ 1299 if (error && error != ENXIO) { 1300 if ((flags & MNT_FORCE) && 1301 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1302 if (mp->mnt_vnodecovered != NULL) 1303 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1304 if (rootvnode == NULL) { 1305 rootvnode = fsrootvp; 1306 vref(rootvnode); 1307 } 1308 vput(fsrootvp); 1309 } 1310 MNT_ILOCK(mp); 1311 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1312 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1313 MNT_IUNLOCK(mp); 1314 (void) vfs_allocate_syncvnode(mp); 1315 MNT_ILOCK(mp); 1316 } 1317 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1318 mp->mnt_flag |= async_flag; 1319 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1320 mp->mnt_kern_flag |= MNTK_ASYNC; 1321 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1322 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1323 wakeup(mp); 1324 } 1325 MNT_IUNLOCK(mp); 1326 if (coveredvp) 1327 VOP_UNLOCK(coveredvp, 0); 1328 return (error); 1329 } 1330 mtx_lock(&mountlist_mtx); 1331 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1332 mtx_unlock(&mountlist_mtx); 1333 if (coveredvp != NULL) { 1334 coveredvp->v_mountedhere = NULL; 1335 vput(coveredvp); 1336 } 1337 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1338 vfs_mount_destroy(mp); 1339 return (0); 1340 } 1341 1342 /* 1343 * --------------------------------------------------------------------- 1344 * Mounting of root filesystem 1345 * 1346 */ 1347 1348 struct root_hold_token { 1349 const char *who; 1350 LIST_ENTRY(root_hold_token) list; 1351 }; 1352 1353 static LIST_HEAD(, root_hold_token) root_holds = 1354 LIST_HEAD_INITIALIZER(root_holds); 1355 1356 static int root_mount_complete; 1357 1358 /* 1359 * Hold root mount. 1360 */ 1361 struct root_hold_token * 1362 root_mount_hold(const char *identifier) 1363 { 1364 struct root_hold_token *h; 1365 1366 if (root_mounted()) 1367 return (NULL); 1368 1369 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1370 h->who = identifier; 1371 mtx_lock(&mountlist_mtx); 1372 LIST_INSERT_HEAD(&root_holds, h, list); 1373 mtx_unlock(&mountlist_mtx); 1374 return (h); 1375 } 1376 1377 /* 1378 * Release root mount. 1379 */ 1380 void 1381 root_mount_rel(struct root_hold_token *h) 1382 { 1383 1384 if (h == NULL) 1385 return; 1386 mtx_lock(&mountlist_mtx); 1387 LIST_REMOVE(h, list); 1388 wakeup(&root_holds); 1389 mtx_unlock(&mountlist_mtx); 1390 free(h, M_DEVBUF); 1391 } 1392 1393 /* 1394 * Wait for all subsystems to release root mount. 1395 */ 1396 static void 1397 root_mount_prepare(void) 1398 { 1399 struct root_hold_token *h; 1400 struct timeval lastfail; 1401 int curfail = 0; 1402 1403 for (;;) { 1404 DROP_GIANT(); 1405 g_waitidle(); 1406 PICKUP_GIANT(); 1407 mtx_lock(&mountlist_mtx); 1408 if (LIST_EMPTY(&root_holds)) { 1409 mtx_unlock(&mountlist_mtx); 1410 break; 1411 } 1412 if (ppsratecheck(&lastfail, &curfail, 1)) { 1413 printf("Root mount waiting for:"); 1414 LIST_FOREACH(h, &root_holds, list) 1415 printf(" %s", h->who); 1416 printf("\n"); 1417 } 1418 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1419 hz); 1420 } 1421 } 1422 1423 /* 1424 * Root was mounted, share the good news. 1425 */ 1426 static void 1427 root_mount_done(void) 1428 { 1429 1430 /* Keep prison0's root in sync with the global rootvnode. */ 1431 mtx_lock(&prison0.pr_mtx); 1432 prison0.pr_root = rootvnode; 1433 vref(prison0.pr_root); 1434 mtx_unlock(&prison0.pr_mtx); 1435 /* 1436 * Use a mutex to prevent the wakeup being missed and waiting for 1437 * an extra 1 second sleep. 1438 */ 1439 mtx_lock(&mountlist_mtx); 1440 root_mount_complete = 1; 1441 wakeup(&root_mount_complete); 1442 mtx_unlock(&mountlist_mtx); 1443 } 1444 1445 /* 1446 * Return true if root is already mounted. 1447 */ 1448 int 1449 root_mounted(void) 1450 { 1451 1452 /* No mutex is acquired here because int stores are atomic. */ 1453 return (root_mount_complete); 1454 } 1455 1456 /* 1457 * Wait until root is mounted. 1458 */ 1459 void 1460 root_mount_wait(void) 1461 { 1462 1463 /* 1464 * Panic on an obvious deadlock - the function can't be called from 1465 * a thread which is doing the whole SYSINIT stuff. 1466 */ 1467 KASSERT(curthread->td_proc->p_pid != 0, 1468 ("root_mount_wait: cannot be called from the swapper thread")); 1469 mtx_lock(&mountlist_mtx); 1470 while (!root_mount_complete) { 1471 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1472 hz); 1473 } 1474 mtx_unlock(&mountlist_mtx); 1475 } 1476 1477 static void 1478 set_rootvnode() 1479 { 1480 struct proc *p; 1481 1482 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode)) 1483 panic("Cannot find root vnode"); 1484 1485 VOP_UNLOCK(rootvnode, 0); 1486 1487 p = curthread->td_proc; 1488 FILEDESC_XLOCK(p->p_fd); 1489 1490 if (p->p_fd->fd_cdir != NULL) 1491 vrele(p->p_fd->fd_cdir); 1492 p->p_fd->fd_cdir = rootvnode; 1493 VREF(rootvnode); 1494 1495 if (p->p_fd->fd_rdir != NULL) 1496 vrele(p->p_fd->fd_rdir); 1497 p->p_fd->fd_rdir = rootvnode; 1498 VREF(rootvnode); 1499 1500 FILEDESC_XUNLOCK(p->p_fd); 1501 1502 EVENTHANDLER_INVOKE(mountroot); 1503 } 1504 1505 /* 1506 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1507 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1508 */ 1509 1510 static void 1511 devfs_first(void) 1512 { 1513 struct thread *td = curthread; 1514 struct vfsoptlist *opts; 1515 struct vfsconf *vfsp; 1516 struct mount *mp = NULL; 1517 int error; 1518 1519 vfsp = vfs_byname("devfs"); 1520 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1521 if (vfsp == NULL) 1522 return; 1523 1524 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 1525 1526 error = VFS_MOUNT(mp); 1527 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1528 if (error) 1529 return; 1530 1531 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1532 TAILQ_INIT(opts); 1533 mp->mnt_opt = opts; 1534 1535 mtx_lock(&mountlist_mtx); 1536 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1537 mtx_unlock(&mountlist_mtx); 1538 1539 set_rootvnode(); 1540 1541 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1542 if (error) 1543 printf("kern_symlink /dev -> / returns %d\n", error); 1544 } 1545 1546 /* 1547 * Surgically move our devfs to be mounted on /dev. 1548 */ 1549 1550 static void 1551 devfs_fixup(struct thread *td) 1552 { 1553 struct nameidata nd; 1554 int error; 1555 struct vnode *vp, *dvp; 1556 struct mount *mp; 1557 1558 /* Remove our devfs mount from the mountlist and purge the cache */ 1559 mtx_lock(&mountlist_mtx); 1560 mp = TAILQ_FIRST(&mountlist); 1561 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1562 mtx_unlock(&mountlist_mtx); 1563 cache_purgevfs(mp); 1564 1565 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp); 1566 VI_LOCK(dvp); 1567 dvp->v_iflag &= ~VI_MOUNT; 1568 VI_UNLOCK(dvp); 1569 dvp->v_mountedhere = NULL; 1570 1571 /* Set up the real rootvnode, and purge the cache */ 1572 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1573 set_rootvnode(); 1574 cache_purgevfs(rootvnode->v_mount); 1575 1576 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1577 error = namei(&nd); 1578 if (error) { 1579 printf("Lookup of /dev for devfs, error: %d\n", error); 1580 return; 1581 } 1582 NDFREE(&nd, NDF_ONLY_PNBUF); 1583 vp = nd.ni_vp; 1584 if (vp->v_type != VDIR) { 1585 vput(vp); 1586 } 1587 error = vinvalbuf(vp, V_SAVE, 0, 0); 1588 if (error) { 1589 vput(vp); 1590 } 1591 cache_purge(vp); 1592 mp->mnt_vnodecovered = vp; 1593 vp->v_mountedhere = mp; 1594 mtx_lock(&mountlist_mtx); 1595 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1596 mtx_unlock(&mountlist_mtx); 1597 VOP_UNLOCK(vp, 0); 1598 vput(dvp); 1599 vfs_unbusy(mp); 1600 1601 /* Unlink the no longer needed /dev/dev -> / symlink */ 1602 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1603 } 1604 1605 /* 1606 * Report errors during filesystem mounting. 1607 */ 1608 void 1609 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1610 { 1611 struct vfsoptlist *moptlist = mp->mnt_optnew; 1612 va_list ap; 1613 int error, len; 1614 char *errmsg; 1615 1616 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1617 if (error || errmsg == NULL || len <= 0) 1618 return; 1619 1620 va_start(ap, fmt); 1621 vsnprintf(errmsg, (size_t)len, fmt, ap); 1622 va_end(ap); 1623 } 1624 1625 void 1626 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1627 { 1628 va_list ap; 1629 int error, len; 1630 char *errmsg; 1631 1632 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1633 if (error || errmsg == NULL || len <= 0) 1634 return; 1635 1636 va_start(ap, fmt); 1637 vsnprintf(errmsg, (size_t)len, fmt, ap); 1638 va_end(ap); 1639 } 1640 1641 /* 1642 * Find and mount the root filesystem 1643 */ 1644 void 1645 vfs_mountroot(void) 1646 { 1647 char *cp, *cpt, *options, *tmpdev; 1648 int error, i, asked = 0; 1649 1650 options = NULL; 1651 1652 root_mount_prepare(); 1653 1654 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1655 NULL, NULL, mount_init, mount_fini, 1656 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1657 devfs_first(); 1658 1659 /* 1660 * We are booted with instructions to prompt for the root filesystem. 1661 */ 1662 if (boothowto & RB_ASKNAME) { 1663 if (!vfs_mountroot_ask()) 1664 goto mounted; 1665 asked = 1; 1666 } 1667 1668 options = getenv("vfs.root.mountfrom.options"); 1669 1670 /* 1671 * The root filesystem information is compiled in, and we are 1672 * booted with instructions to use it. 1673 */ 1674 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1675 if (!vfs_mountroot_try(ctrootdevname, options)) 1676 goto mounted; 1677 ctrootdevname = NULL; 1678 } 1679 1680 /* 1681 * We've been given the generic "use CDROM as root" flag. This is 1682 * necessary because one media may be used in many different 1683 * devices, so we need to search for them. 1684 */ 1685 if (boothowto & RB_CDROM) { 1686 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1687 if (!vfs_mountroot_try(cdrom_rootdevnames[i], options)) 1688 goto mounted; 1689 } 1690 } 1691 1692 /* 1693 * Try to use the value read by the loader from /etc/fstab, or 1694 * supplied via some other means. This is the preferred 1695 * mechanism. 1696 */ 1697 cp = getenv("vfs.root.mountfrom"); 1698 if (cp != NULL) { 1699 cpt = cp; 1700 while ((tmpdev = strsep(&cpt, " \t")) != NULL) { 1701 error = vfs_mountroot_try(tmpdev, options); 1702 if (error == 0) { 1703 freeenv(cp); 1704 goto mounted; 1705 } 1706 } 1707 freeenv(cp); 1708 } 1709 1710 /* 1711 * Try values that may have been computed by code during boot 1712 */ 1713 if (!vfs_mountroot_try(rootdevnames[0], options)) 1714 goto mounted; 1715 if (!vfs_mountroot_try(rootdevnames[1], options)) 1716 goto mounted; 1717 1718 /* 1719 * If we (still) have a compiled-in default, try it. 1720 */ 1721 if (ctrootdevname != NULL) 1722 if (!vfs_mountroot_try(ctrootdevname, options)) 1723 goto mounted; 1724 /* 1725 * Everything so far has failed, prompt on the console if we haven't 1726 * already tried that. 1727 */ 1728 if (!asked) 1729 if (!vfs_mountroot_ask()) 1730 goto mounted; 1731 1732 panic("Root mount failed, startup aborted."); 1733 1734 mounted: 1735 root_mount_done(); 1736 freeenv(options); 1737 } 1738 1739 static struct mntarg * 1740 parse_mountroot_options(struct mntarg *ma, const char *options) 1741 { 1742 char *p; 1743 char *name, *name_arg; 1744 char *val, *val_arg; 1745 char *opts; 1746 1747 if (options == NULL || options[0] == '\0') 1748 return (ma); 1749 1750 p = opts = strdup(options, M_MOUNT); 1751 if (opts == NULL) { 1752 return (ma); 1753 } 1754 1755 while((name = strsep(&p, ",")) != NULL) { 1756 if (name[0] == '\0') 1757 break; 1758 1759 val = strchr(name, '='); 1760 if (val != NULL) { 1761 *val = '\0'; 1762 ++val; 1763 } 1764 if( strcmp(name, "rw") == 0 || 1765 strcmp(name, "noro") == 0) { 1766 /* 1767 * The first time we mount the root file system, 1768 * we need to mount 'ro', so We need to ignore 1769 * 'rw' and 'noro' mount options. 1770 */ 1771 continue; 1772 } 1773 name_arg = strdup(name, M_MOUNT); 1774 val_arg = NULL; 1775 if (val != NULL) 1776 val_arg = strdup(val, M_MOUNT); 1777 1778 ma = mount_arg(ma, name_arg, val_arg, 1779 (val_arg != NULL ? -1 : 0)); 1780 } 1781 free(opts, M_MOUNT); 1782 return (ma); 1783 } 1784 1785 /* 1786 * Mount (mountfrom) as the root filesystem. 1787 */ 1788 static int 1789 vfs_mountroot_try(const char *mountfrom, const char *options) 1790 { 1791 struct mount *mp; 1792 struct mntarg *ma; 1793 char *vfsname, *path; 1794 time_t timebase; 1795 int error; 1796 char patt[32]; 1797 char errmsg[255]; 1798 1799 vfsname = NULL; 1800 path = NULL; 1801 mp = NULL; 1802 ma = NULL; 1803 error = EINVAL; 1804 bzero(errmsg, sizeof(errmsg)); 1805 1806 if (mountfrom == NULL) 1807 return (error); /* don't complain */ 1808 printf("Trying to mount root from %s\n", mountfrom); 1809 1810 /* parse vfs name and path */ 1811 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1812 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1813 vfsname[0] = path[0] = 0; 1814 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1815 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1816 goto out; 1817 1818 if (path[0] == '\0') 1819 strcpy(path, ROOTNAME); 1820 1821 ma = mount_arg(ma, "fstype", vfsname, -1); 1822 ma = mount_arg(ma, "fspath", "/", -1); 1823 ma = mount_arg(ma, "from", path, -1); 1824 ma = mount_arg(ma, "errmsg", errmsg, sizeof(errmsg)); 1825 ma = mount_arg(ma, "ro", NULL, 0); 1826 ma = parse_mountroot_options(ma, options); 1827 error = kernel_mount(ma, MNT_ROOTFS); 1828 1829 if (error == 0) { 1830 /* 1831 * We mount devfs prior to mounting the / FS, so the first 1832 * entry will typically be devfs. 1833 */ 1834 mp = TAILQ_FIRST(&mountlist); 1835 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1836 1837 /* 1838 * Iterate over all currently mounted file systems and use 1839 * the time stamp found to check and/or initialize the RTC. 1840 * Typically devfs has no time stamp and the only other FS 1841 * is the actual / FS. 1842 * Call inittodr() only once and pass it the largest of the 1843 * timestamps we encounter. 1844 */ 1845 timebase = 0; 1846 do { 1847 if (mp->mnt_time > timebase) 1848 timebase = mp->mnt_time; 1849 mp = TAILQ_NEXT(mp, mnt_list); 1850 } while (mp != NULL); 1851 inittodr(timebase); 1852 1853 devfs_fixup(curthread); 1854 } 1855 1856 if (error != 0 ) { 1857 printf("ROOT MOUNT ERROR: %s\n", errmsg); 1858 printf("If you have invalid mount options, reboot, and "); 1859 printf("first try the following from\n"); 1860 printf("the loader prompt:\n\n"); 1861 printf(" set vfs.root.mountfrom.options=rw\n\n"); 1862 printf("and then remove invalid mount options from "); 1863 printf("/etc/fstab.\n\n"); 1864 } 1865 out: 1866 free(path, M_MOUNT); 1867 free(vfsname, M_MOUNT); 1868 return (error); 1869 } 1870 1871 /* 1872 * --------------------------------------------------------------------- 1873 * Interactive root filesystem selection code. 1874 */ 1875 1876 static int 1877 vfs_mountroot_ask(void) 1878 { 1879 char name[128]; 1880 char *mountfrom; 1881 char *options; 1882 1883 for(;;) { 1884 printf("Loader variables:\n"); 1885 printf("vfs.root.mountfrom="); 1886 mountfrom = getenv("vfs.root.mountfrom"); 1887 if (mountfrom != NULL) { 1888 printf("%s", mountfrom); 1889 } 1890 printf("\n"); 1891 printf("vfs.root.mountfrom.options="); 1892 options = getenv("vfs.root.mountfrom.options"); 1893 if (options != NULL) { 1894 printf("%s", options); 1895 } 1896 printf("\n"); 1897 freeenv(mountfrom); 1898 freeenv(options); 1899 printf("\nManual root filesystem specification:\n"); 1900 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1901 printf(" eg. zfs:tank\n"); 1902 printf(" eg. ufs:/dev/da0s1a\n"); 1903 printf(" eg. cd9660:/dev/acd0\n"); 1904 printf(" This is equivalent to: "); 1905 printf("mount -t cd9660 /dev/acd0 /\n"); 1906 printf("\n"); 1907 printf(" ? List valid disk boot devices\n"); 1908 printf(" <empty line> Abort manual input\n"); 1909 printf("\nmountroot> "); 1910 gets(name, sizeof(name), 1); 1911 if (name[0] == '\0') 1912 return (1); 1913 if (name[0] == '?') { 1914 printf("\nList of GEOM managed disk devices:\n "); 1915 g_dev_print(); 1916 continue; 1917 } 1918 if (!vfs_mountroot_try(name, NULL)) 1919 return (0); 1920 } 1921 } 1922 1923 /* 1924 * --------------------------------------------------------------------- 1925 * Functions for querying mount options/arguments from filesystems. 1926 */ 1927 1928 /* 1929 * Check that no unknown options are given 1930 */ 1931 int 1932 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1933 { 1934 struct vfsopt *opt; 1935 char errmsg[255]; 1936 const char **t, *p, *q; 1937 int ret = 0; 1938 1939 TAILQ_FOREACH(opt, opts, link) { 1940 p = opt->name; 1941 q = NULL; 1942 if (p[0] == 'n' && p[1] == 'o') 1943 q = p + 2; 1944 for(t = global_opts; *t != NULL; t++) { 1945 if (strcmp(*t, p) == 0) 1946 break; 1947 if (q != NULL) { 1948 if (strcmp(*t, q) == 0) 1949 break; 1950 } 1951 } 1952 if (*t != NULL) 1953 continue; 1954 for(t = legal; *t != NULL; t++) { 1955 if (strcmp(*t, p) == 0) 1956 break; 1957 if (q != NULL) { 1958 if (strcmp(*t, q) == 0) 1959 break; 1960 } 1961 } 1962 if (*t != NULL) 1963 continue; 1964 snprintf(errmsg, sizeof(errmsg), 1965 "mount option <%s> is unknown", p); 1966 printf("%s\n", errmsg); 1967 ret = EINVAL; 1968 } 1969 if (ret != 0) { 1970 TAILQ_FOREACH(opt, opts, link) { 1971 if (strcmp(opt->name, "errmsg") == 0) { 1972 strncpy((char *)opt->value, errmsg, opt->len); 1973 } 1974 } 1975 } 1976 return (ret); 1977 } 1978 1979 /* 1980 * Get a mount option by its name. 1981 * 1982 * Return 0 if the option was found, ENOENT otherwise. 1983 * If len is non-NULL it will be filled with the length 1984 * of the option. If buf is non-NULL, it will be filled 1985 * with the address of the option. 1986 */ 1987 int 1988 vfs_getopt(opts, name, buf, len) 1989 struct vfsoptlist *opts; 1990 const char *name; 1991 void **buf; 1992 int *len; 1993 { 1994 struct vfsopt *opt; 1995 1996 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1997 1998 TAILQ_FOREACH(opt, opts, link) { 1999 if (strcmp(name, opt->name) == 0) { 2000 opt->seen = 1; 2001 if (len != NULL) 2002 *len = opt->len; 2003 if (buf != NULL) 2004 *buf = opt->value; 2005 return (0); 2006 } 2007 } 2008 return (ENOENT); 2009 } 2010 2011 int 2012 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2013 { 2014 struct vfsopt *opt; 2015 2016 if (opts == NULL) 2017 return (-1); 2018 2019 TAILQ_FOREACH(opt, opts, link) { 2020 if (strcmp(name, opt->name) == 0) { 2021 opt->seen = 1; 2022 return (opt->pos); 2023 } 2024 } 2025 return (-1); 2026 } 2027 2028 char * 2029 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2030 { 2031 struct vfsopt *opt; 2032 2033 *error = 0; 2034 TAILQ_FOREACH(opt, opts, link) { 2035 if (strcmp(name, opt->name) != 0) 2036 continue; 2037 opt->seen = 1; 2038 if (opt->len == 0 || 2039 ((char *)opt->value)[opt->len - 1] != '\0') { 2040 *error = EINVAL; 2041 return (NULL); 2042 } 2043 return (opt->value); 2044 } 2045 *error = ENOENT; 2046 return (NULL); 2047 } 2048 2049 int 2050 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 2051 { 2052 struct vfsopt *opt; 2053 2054 TAILQ_FOREACH(opt, opts, link) { 2055 if (strcmp(name, opt->name) == 0) { 2056 opt->seen = 1; 2057 if (w != NULL) 2058 *w |= val; 2059 return (1); 2060 } 2061 } 2062 if (w != NULL) 2063 *w &= ~val; 2064 return (0); 2065 } 2066 2067 int 2068 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2069 { 2070 va_list ap; 2071 struct vfsopt *opt; 2072 int ret; 2073 2074 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2075 2076 TAILQ_FOREACH(opt, opts, link) { 2077 if (strcmp(name, opt->name) != 0) 2078 continue; 2079 opt->seen = 1; 2080 if (opt->len == 0 || opt->value == NULL) 2081 return (0); 2082 if (((char *)opt->value)[opt->len - 1] != '\0') 2083 return (0); 2084 va_start(ap, fmt); 2085 ret = vsscanf(opt->value, fmt, ap); 2086 va_end(ap); 2087 return (ret); 2088 } 2089 return (0); 2090 } 2091 2092 int 2093 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2094 { 2095 struct vfsopt *opt; 2096 2097 TAILQ_FOREACH(opt, opts, link) { 2098 if (strcmp(name, opt->name) != 0) 2099 continue; 2100 opt->seen = 1; 2101 if (opt->value == NULL) 2102 opt->len = len; 2103 else { 2104 if (opt->len != len) 2105 return (EINVAL); 2106 bcopy(value, opt->value, len); 2107 } 2108 return (0); 2109 } 2110 return (ENOENT); 2111 } 2112 2113 int 2114 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2115 { 2116 struct vfsopt *opt; 2117 2118 TAILQ_FOREACH(opt, opts, link) { 2119 if (strcmp(name, opt->name) != 0) 2120 continue; 2121 opt->seen = 1; 2122 if (opt->value == NULL) 2123 opt->len = len; 2124 else { 2125 if (opt->len < len) 2126 return (EINVAL); 2127 opt->len = len; 2128 bcopy(value, opt->value, len); 2129 } 2130 return (0); 2131 } 2132 return (ENOENT); 2133 } 2134 2135 int 2136 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2137 { 2138 struct vfsopt *opt; 2139 2140 TAILQ_FOREACH(opt, opts, link) { 2141 if (strcmp(name, opt->name) != 0) 2142 continue; 2143 opt->seen = 1; 2144 if (opt->value == NULL) 2145 opt->len = strlen(value) + 1; 2146 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2147 return (EINVAL); 2148 return (0); 2149 } 2150 return (ENOENT); 2151 } 2152 2153 /* 2154 * Find and copy a mount option. 2155 * 2156 * The size of the buffer has to be specified 2157 * in len, if it is not the same length as the 2158 * mount option, EINVAL is returned. 2159 * Returns ENOENT if the option is not found. 2160 */ 2161 int 2162 vfs_copyopt(opts, name, dest, len) 2163 struct vfsoptlist *opts; 2164 const char *name; 2165 void *dest; 2166 int len; 2167 { 2168 struct vfsopt *opt; 2169 2170 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2171 2172 TAILQ_FOREACH(opt, opts, link) { 2173 if (strcmp(name, opt->name) == 0) { 2174 opt->seen = 1; 2175 if (len != opt->len) 2176 return (EINVAL); 2177 bcopy(opt->value, dest, opt->len); 2178 return (0); 2179 } 2180 } 2181 return (ENOENT); 2182 } 2183 2184 /* 2185 * This is a helper function for filesystems to traverse their 2186 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2187 */ 2188 2189 struct vnode * 2190 __mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2191 { 2192 struct vnode *vp; 2193 2194 mtx_assert(MNT_MTX(mp), MA_OWNED); 2195 2196 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2197 if ((*mvp)->v_yield++ == 500) { 2198 MNT_IUNLOCK(mp); 2199 (*mvp)->v_yield = 0; 2200 uio_yield(); 2201 MNT_ILOCK(mp); 2202 } 2203 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2204 while (vp != NULL && vp->v_type == VMARKER) 2205 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2206 2207 /* Check if we are done */ 2208 if (vp == NULL) { 2209 __mnt_vnode_markerfree(mvp, mp); 2210 return (NULL); 2211 } 2212 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2213 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2214 return (vp); 2215 } 2216 2217 struct vnode * 2218 __mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2219 { 2220 struct vnode *vp; 2221 2222 mtx_assert(MNT_MTX(mp), MA_OWNED); 2223 2224 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2225 while (vp != NULL && vp->v_type == VMARKER) 2226 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2227 2228 /* Check if we are done */ 2229 if (vp == NULL) { 2230 *mvp = NULL; 2231 return (NULL); 2232 } 2233 MNT_REF(mp); 2234 MNT_IUNLOCK(mp); 2235 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2236 M_VNODE_MARKER, 2237 M_WAITOK | M_ZERO); 2238 MNT_ILOCK(mp); 2239 (*mvp)->v_type = VMARKER; 2240 2241 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2242 while (vp != NULL && vp->v_type == VMARKER) 2243 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2244 2245 /* Check if we are done */ 2246 if (vp == NULL) { 2247 MNT_IUNLOCK(mp); 2248 free(*mvp, M_VNODE_MARKER); 2249 MNT_ILOCK(mp); 2250 *mvp = NULL; 2251 MNT_REL(mp); 2252 return (NULL); 2253 } 2254 (*mvp)->v_mount = mp; 2255 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2256 return (vp); 2257 } 2258 2259 2260 void 2261 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2262 { 2263 2264 if (*mvp == NULL) 2265 return; 2266 2267 mtx_assert(MNT_MTX(mp), MA_OWNED); 2268 2269 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2270 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2271 MNT_IUNLOCK(mp); 2272 free(*mvp, M_VNODE_MARKER); 2273 MNT_ILOCK(mp); 2274 *mvp = NULL; 2275 MNT_REL(mp); 2276 } 2277 2278 2279 int 2280 __vfs_statfs(struct mount *mp, struct statfs *sbp) 2281 { 2282 int error; 2283 2284 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 2285 if (sbp != &mp->mnt_stat) 2286 *sbp = mp->mnt_stat; 2287 return (error); 2288 } 2289 2290 void 2291 vfs_mountedfrom(struct mount *mp, const char *from) 2292 { 2293 2294 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2295 strlcpy(mp->mnt_stat.f_mntfromname, from, 2296 sizeof mp->mnt_stat.f_mntfromname); 2297 } 2298 2299 /* 2300 * --------------------------------------------------------------------- 2301 * This is the api for building mount args and mounting filesystems from 2302 * inside the kernel. 2303 * 2304 * The API works by accumulation of individual args. First error is 2305 * latched. 2306 * 2307 * XXX: should be documented in new manpage kernel_mount(9) 2308 */ 2309 2310 /* A memory allocation which must be freed when we are done */ 2311 struct mntaarg { 2312 SLIST_ENTRY(mntaarg) next; 2313 }; 2314 2315 /* The header for the mount arguments */ 2316 struct mntarg { 2317 struct iovec *v; 2318 int len; 2319 int error; 2320 SLIST_HEAD(, mntaarg) list; 2321 }; 2322 2323 /* 2324 * Add a boolean argument. 2325 * 2326 * flag is the boolean value. 2327 * name must start with "no". 2328 */ 2329 struct mntarg * 2330 mount_argb(struct mntarg *ma, int flag, const char *name) 2331 { 2332 2333 KASSERT(name[0] == 'n' && name[1] == 'o', 2334 ("mount_argb(...,%s): name must start with 'no'", name)); 2335 2336 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2337 } 2338 2339 /* 2340 * Add an argument printf style 2341 */ 2342 struct mntarg * 2343 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2344 { 2345 va_list ap; 2346 struct mntaarg *maa; 2347 struct sbuf *sb; 2348 int len; 2349 2350 if (ma == NULL) { 2351 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2352 SLIST_INIT(&ma->list); 2353 } 2354 if (ma->error) 2355 return (ma); 2356 2357 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2358 M_MOUNT, M_WAITOK); 2359 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2360 ma->v[ma->len].iov_len = strlen(name) + 1; 2361 ma->len++; 2362 2363 sb = sbuf_new_auto(); 2364 va_start(ap, fmt); 2365 sbuf_vprintf(sb, fmt, ap); 2366 va_end(ap); 2367 sbuf_finish(sb); 2368 len = sbuf_len(sb) + 1; 2369 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2370 SLIST_INSERT_HEAD(&ma->list, maa, next); 2371 bcopy(sbuf_data(sb), maa + 1, len); 2372 sbuf_delete(sb); 2373 2374 ma->v[ma->len].iov_base = maa + 1; 2375 ma->v[ma->len].iov_len = len; 2376 ma->len++; 2377 2378 return (ma); 2379 } 2380 2381 /* 2382 * Add an argument which is a userland string. 2383 */ 2384 struct mntarg * 2385 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2386 { 2387 struct mntaarg *maa; 2388 char *tbuf; 2389 2390 if (val == NULL) 2391 return (ma); 2392 if (ma == NULL) { 2393 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2394 SLIST_INIT(&ma->list); 2395 } 2396 if (ma->error) 2397 return (ma); 2398 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2399 SLIST_INSERT_HEAD(&ma->list, maa, next); 2400 tbuf = (void *)(maa + 1); 2401 ma->error = copyinstr(val, tbuf, len, NULL); 2402 return (mount_arg(ma, name, tbuf, -1)); 2403 } 2404 2405 /* 2406 * Plain argument. 2407 * 2408 * If length is -1, treat value as a C string. 2409 */ 2410 struct mntarg * 2411 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2412 { 2413 2414 if (ma == NULL) { 2415 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2416 SLIST_INIT(&ma->list); 2417 } 2418 if (ma->error) 2419 return (ma); 2420 2421 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2422 M_MOUNT, M_WAITOK); 2423 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2424 ma->v[ma->len].iov_len = strlen(name) + 1; 2425 ma->len++; 2426 2427 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2428 if (len < 0) 2429 ma->v[ma->len].iov_len = strlen(val) + 1; 2430 else 2431 ma->v[ma->len].iov_len = len; 2432 ma->len++; 2433 return (ma); 2434 } 2435 2436 /* 2437 * Free a mntarg structure 2438 */ 2439 static void 2440 free_mntarg(struct mntarg *ma) 2441 { 2442 struct mntaarg *maa; 2443 2444 while (!SLIST_EMPTY(&ma->list)) { 2445 maa = SLIST_FIRST(&ma->list); 2446 SLIST_REMOVE_HEAD(&ma->list, next); 2447 free(maa, M_MOUNT); 2448 } 2449 free(ma->v, M_MOUNT); 2450 free(ma, M_MOUNT); 2451 } 2452 2453 /* 2454 * Mount a filesystem 2455 */ 2456 int 2457 kernel_mount(struct mntarg *ma, int flags) 2458 { 2459 struct uio auio; 2460 int error; 2461 2462 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2463 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2464 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2465 2466 auio.uio_iov = ma->v; 2467 auio.uio_iovcnt = ma->len; 2468 auio.uio_segflg = UIO_SYSSPACE; 2469 2470 error = ma->error; 2471 if (!error) 2472 error = vfs_donmount(curthread, flags, &auio); 2473 free_mntarg(ma); 2474 return (error); 2475 } 2476 2477 /* 2478 * A printflike function to mount a filesystem. 2479 */ 2480 int 2481 kernel_vmount(int flags, ...) 2482 { 2483 struct mntarg *ma = NULL; 2484 va_list ap; 2485 const char *cp; 2486 const void *vp; 2487 int error; 2488 2489 va_start(ap, flags); 2490 for (;;) { 2491 cp = va_arg(ap, const char *); 2492 if (cp == NULL) 2493 break; 2494 vp = va_arg(ap, const void *); 2495 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2496 } 2497 va_end(ap); 2498 2499 error = kernel_mount(ma, flags); 2500 return (error); 2501 } 2502