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 #include "opt_mac.h" 72 73 #define ROOTNAME "root_device" 74 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 75 76 static void set_rootvnode(void); 77 static int vfs_domount(struct thread *td, const char *fstype, 78 char *fspath, int fsflags, void *fsdata); 79 static int vfs_mountroot_ask(void); 80 static int vfs_mountroot_try(const char *mountfrom, const char *options); 81 static void free_mntarg(struct mntarg *ma); 82 83 static int usermount = 0; 84 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 85 "Unprivileged users may mount and unmount file systems"); 86 87 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 88 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 89 static uma_zone_t mount_zone; 90 91 /* List of mounted filesystems. */ 92 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 93 94 /* For any iteration/modification of mountlist */ 95 struct mtx mountlist_mtx; 96 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 97 98 /* 99 * The vnode of the system's root (/ in the filesystem, without chroot 100 * active.) 101 */ 102 struct vnode *rootvnode; 103 104 /* 105 * The root filesystem is detailed in the kernel environment variable 106 * vfs.root.mountfrom, which is expected to be in the general format 107 * 108 * <vfsname>:[<path>] 109 * vfsname := the name of a VFS known to the kernel and capable 110 * of being mounted as root 111 * path := disk device name or other data used by the filesystem 112 * to locate its physical store 113 * 114 * The environment variable vfs.root.mountfrom options is a comma delimited 115 * set of string mount options. These mount options must be parseable 116 * by nmount() in the kernel. 117 */ 118 119 /* 120 * Global opts, taken by all filesystems 121 */ 122 static const char *global_opts[] = { 123 "errmsg", 124 "fstype", 125 "fspath", 126 "ro", 127 "rw", 128 "nosuid", 129 "noexec", 130 NULL 131 }; 132 133 /* 134 * The root specifiers we will try if RB_CDROM is specified. 135 */ 136 static char *cdrom_rootdevnames[] = { 137 "cd9660:cd0", 138 "cd9660:acd0", 139 NULL 140 }; 141 142 /* legacy find-root code */ 143 char *rootdevnames[2] = {NULL, NULL}; 144 #ifndef ROOTDEVNAME 145 # define ROOTDEVNAME NULL 146 #endif 147 static const char *ctrootdevname = ROOTDEVNAME; 148 149 /* 150 * --------------------------------------------------------------------- 151 * Functions for building and sanitizing the mount options 152 */ 153 154 /* Remove one mount option. */ 155 static void 156 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 157 { 158 159 TAILQ_REMOVE(opts, opt, link); 160 free(opt->name, M_MOUNT); 161 if (opt->value != NULL) 162 free(opt->value, M_MOUNT); 163 free(opt, M_MOUNT); 164 } 165 166 /* Release all resources related to the mount options. */ 167 void 168 vfs_freeopts(struct vfsoptlist *opts) 169 { 170 struct vfsopt *opt; 171 172 while (!TAILQ_EMPTY(opts)) { 173 opt = TAILQ_FIRST(opts); 174 vfs_freeopt(opts, opt); 175 } 176 free(opts, M_MOUNT); 177 } 178 179 void 180 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 181 { 182 struct vfsopt *opt, *temp; 183 184 if (opts == NULL) 185 return; 186 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 187 if (strcmp(opt->name, name) == 0) 188 vfs_freeopt(opts, opt); 189 } 190 } 191 192 /* 193 * Check if options are equal (with or without the "no" prefix). 194 */ 195 static int 196 vfs_equalopts(const char *opt1, const char *opt2) 197 { 198 char *p; 199 200 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 201 if (strcmp(opt1, opt2) == 0) 202 return (1); 203 /* "noopt" vs. "opt" */ 204 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 205 return (1); 206 /* "opt" vs. "noopt" */ 207 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 208 return (1); 209 while ((p = strchr(opt1, '.')) != NULL && 210 !strncmp(opt1, opt2, ++p - opt1)) { 211 opt2 += p - opt1; 212 opt1 = p; 213 /* "foo.noopt" vs. "foo.opt" */ 214 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 215 return (1); 216 /* "foo.opt" vs. "foo.noopt" */ 217 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 218 return (1); 219 } 220 return (0); 221 } 222 223 /* 224 * If a mount option is specified several times, 225 * (with or without the "no" prefix) only keep 226 * the last occurence of it. 227 */ 228 static void 229 vfs_sanitizeopts(struct vfsoptlist *opts) 230 { 231 struct vfsopt *opt, *opt2, *tmp; 232 233 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 234 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 235 while (opt2 != NULL) { 236 if (vfs_equalopts(opt->name, opt2->name)) { 237 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 238 vfs_freeopt(opts, opt2); 239 opt2 = tmp; 240 } else { 241 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 242 } 243 } 244 } 245 } 246 247 /* 248 * Build a linked list of mount options from a struct uio. 249 */ 250 int 251 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 252 { 253 struct vfsoptlist *opts; 254 struct vfsopt *opt; 255 size_t memused, namelen, optlen; 256 unsigned int i, iovcnt; 257 int error; 258 259 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 260 TAILQ_INIT(opts); 261 memused = 0; 262 iovcnt = auio->uio_iovcnt; 263 for (i = 0; i < iovcnt; i += 2) { 264 namelen = auio->uio_iov[i].iov_len; 265 optlen = auio->uio_iov[i + 1].iov_len; 266 memused += sizeof(struct vfsopt) + optlen + namelen; 267 /* 268 * Avoid consuming too much memory, and attempts to overflow 269 * memused. 270 */ 271 if (memused > VFS_MOUNTARG_SIZE_MAX || 272 optlen > VFS_MOUNTARG_SIZE_MAX || 273 namelen > VFS_MOUNTARG_SIZE_MAX) { 274 error = EINVAL; 275 goto bad; 276 } 277 278 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 279 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 280 opt->value = NULL; 281 opt->len = 0; 282 opt->pos = i / 2; 283 opt->seen = 0; 284 285 /* 286 * Do this early, so jumps to "bad" will free the current 287 * option. 288 */ 289 TAILQ_INSERT_TAIL(opts, opt, link); 290 291 if (auio->uio_segflg == UIO_SYSSPACE) { 292 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 293 } else { 294 error = copyin(auio->uio_iov[i].iov_base, opt->name, 295 namelen); 296 if (error) 297 goto bad; 298 } 299 /* Ensure names are null-terminated strings. */ 300 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 301 error = EINVAL; 302 goto bad; 303 } 304 if (optlen != 0) { 305 opt->len = optlen; 306 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 307 if (auio->uio_segflg == UIO_SYSSPACE) { 308 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 309 optlen); 310 } else { 311 error = copyin(auio->uio_iov[i + 1].iov_base, 312 opt->value, optlen); 313 if (error) 314 goto bad; 315 } 316 } 317 } 318 vfs_sanitizeopts(opts); 319 *options = opts; 320 return (0); 321 bad: 322 vfs_freeopts(opts); 323 return (error); 324 } 325 326 /* 327 * Merge the old mount options with the new ones passed 328 * in the MNT_UPDATE case. 329 * 330 * XXX This function will keep a "nofoo" option in the 331 * new options if there is no matching "foo" option 332 * to be cancelled in the old options. This is a bug 333 * if the option's canonical name is "foo". E.g., "noro" 334 * shouldn't end up in the mount point's active options, 335 * but it can. 336 */ 337 static void 338 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 339 { 340 struct vfsopt *opt, *opt2, *new; 341 342 TAILQ_FOREACH(opt, opts, link) { 343 /* 344 * Check that this option hasn't been redefined 345 * nor cancelled with a "no" mount option. 346 */ 347 opt2 = TAILQ_FIRST(toopts); 348 while (opt2 != NULL) { 349 if (strcmp(opt2->name, opt->name) == 0) 350 goto next; 351 if (strncmp(opt2->name, "no", 2) == 0 && 352 strcmp(opt2->name + 2, opt->name) == 0) { 353 vfs_freeopt(toopts, opt2); 354 goto next; 355 } 356 opt2 = TAILQ_NEXT(opt2, link); 357 } 358 /* We want this option, duplicate it. */ 359 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 360 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 361 strcpy(new->name, opt->name); 362 if (opt->len != 0) { 363 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 364 bcopy(opt->value, new->value, opt->len); 365 } else { 366 new->value = NULL; 367 } 368 new->len = opt->len; 369 new->seen = opt->seen; 370 TAILQ_INSERT_TAIL(toopts, new, link); 371 next: 372 continue; 373 } 374 } 375 376 /* 377 * Mount a filesystem. 378 */ 379 int 380 nmount(td, uap) 381 struct thread *td; 382 struct nmount_args /* { 383 struct iovec *iovp; 384 unsigned int iovcnt; 385 int flags; 386 } */ *uap; 387 { 388 struct uio *auio; 389 int error; 390 u_int iovcnt; 391 392 AUDIT_ARG(fflags, uap->flags); 393 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 394 uap->iovp, uap->iovcnt, uap->flags); 395 396 /* 397 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 398 * userspace to set this flag, but we must filter it out if we want 399 * MNT_UPDATE on the root file system to work. 400 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 401 */ 402 uap->flags &= ~MNT_ROOTFS; 403 404 iovcnt = uap->iovcnt; 405 /* 406 * Check that we have an even number of iovec's 407 * and that we have at least two options. 408 */ 409 if ((iovcnt & 1) || (iovcnt < 4)) { 410 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 411 uap->iovcnt); 412 return (EINVAL); 413 } 414 415 error = copyinuio(uap->iovp, iovcnt, &auio); 416 if (error) { 417 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 418 __func__, error); 419 return (error); 420 } 421 error = vfs_donmount(td, uap->flags, auio); 422 423 free(auio, M_IOV); 424 return (error); 425 } 426 427 /* 428 * --------------------------------------------------------------------- 429 * Various utility functions 430 */ 431 432 void 433 vfs_ref(struct mount *mp) 434 { 435 436 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 437 MNT_ILOCK(mp); 438 MNT_REF(mp); 439 MNT_IUNLOCK(mp); 440 } 441 442 void 443 vfs_rel(struct mount *mp) 444 { 445 446 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 447 MNT_ILOCK(mp); 448 MNT_REL(mp); 449 MNT_IUNLOCK(mp); 450 } 451 452 static int 453 mount_init(void *mem, int size, int flags) 454 { 455 struct mount *mp; 456 457 mp = (struct mount *)mem; 458 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 459 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 460 return (0); 461 } 462 463 static void 464 mount_fini(void *mem, int size) 465 { 466 struct mount *mp; 467 468 mp = (struct mount *)mem; 469 lockdestroy(&mp->mnt_explock); 470 mtx_destroy(&mp->mnt_mtx); 471 } 472 473 /* 474 * Allocate and initialize the mount point struct. 475 */ 476 struct mount * 477 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 478 struct ucred *cred) 479 { 480 struct mount *mp; 481 482 mp = uma_zalloc(mount_zone, M_WAITOK); 483 bzero(&mp->mnt_startzero, 484 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 485 TAILQ_INIT(&mp->mnt_nvnodelist); 486 mp->mnt_nvnodelistsize = 0; 487 mp->mnt_ref = 0; 488 (void) vfs_busy(mp, MBF_NOWAIT); 489 mp->mnt_op = vfsp->vfc_vfsops; 490 mp->mnt_vfc = vfsp; 491 vfsp->vfc_refcount++; /* XXX Unlocked */ 492 mp->mnt_stat.f_type = vfsp->vfc_typenum; 493 mp->mnt_gen++; 494 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 495 mp->mnt_vnodecovered = vp; 496 mp->mnt_cred = crdup(cred); 497 mp->mnt_stat.f_owner = cred->cr_uid; 498 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 499 mp->mnt_iosize_max = DFLTPHYS; 500 #ifdef MAC 501 mac_mount_init(mp); 502 mac_mount_create(cred, mp); 503 #endif 504 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 505 return (mp); 506 } 507 508 /* 509 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 510 */ 511 void 512 vfs_mount_destroy(struct mount *mp) 513 { 514 515 MNT_ILOCK(mp); 516 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 517 if (mp->mnt_kern_flag & MNTK_MWAIT) { 518 mp->mnt_kern_flag &= ~MNTK_MWAIT; 519 wakeup(mp); 520 } 521 while (mp->mnt_ref) 522 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 523 KASSERT(mp->mnt_ref == 0, 524 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 525 __FILE__, __LINE__)); 526 if (mp->mnt_writeopcount != 0) 527 panic("vfs_mount_destroy: nonzero writeopcount"); 528 if (mp->mnt_secondary_writes != 0) 529 panic("vfs_mount_destroy: nonzero secondary_writes"); 530 mp->mnt_vfc->vfc_refcount--; 531 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 532 struct vnode *vp; 533 534 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 535 vprint("", vp); 536 panic("unmount: dangling vnode"); 537 } 538 if (mp->mnt_nvnodelistsize != 0) 539 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 540 if (mp->mnt_lockref != 0) 541 panic("vfs_mount_destroy: nonzero lock refcount"); 542 MNT_IUNLOCK(mp); 543 #ifdef MAC 544 mac_mount_destroy(mp); 545 #endif 546 if (mp->mnt_opt != NULL) 547 vfs_freeopts(mp->mnt_opt); 548 crfree(mp->mnt_cred); 549 uma_zfree(mount_zone, mp); 550 } 551 552 int 553 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 554 { 555 struct vfsoptlist *optlist; 556 struct vfsopt *opt, *noro_opt, *tmp_opt; 557 char *fstype, *fspath, *errmsg; 558 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 559 int has_rw, has_noro; 560 561 errmsg = fspath = NULL; 562 errmsg_len = has_noro = has_rw = fspathlen = 0; 563 errmsg_pos = -1; 564 565 error = vfs_buildopts(fsoptions, &optlist); 566 if (error) 567 return (error); 568 569 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 570 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 571 572 /* 573 * We need these two options before the others, 574 * and they are mandatory for any filesystem. 575 * Ensure they are NUL terminated as well. 576 */ 577 fstypelen = 0; 578 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 579 if (error || fstype[fstypelen - 1] != '\0') { 580 error = EINVAL; 581 if (errmsg != NULL) 582 strncpy(errmsg, "Invalid fstype", errmsg_len); 583 goto bail; 584 } 585 fspathlen = 0; 586 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 587 if (error || fspath[fspathlen - 1] != '\0') { 588 error = EINVAL; 589 if (errmsg != NULL) 590 strncpy(errmsg, "Invalid fspath", errmsg_len); 591 goto bail; 592 } 593 594 /* 595 * We need to see if we have the "update" option 596 * before we call vfs_domount(), since vfs_domount() has special 597 * logic based on MNT_UPDATE. This is very important 598 * when we want to update the root filesystem. 599 */ 600 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 601 if (strcmp(opt->name, "update") == 0) { 602 fsflags |= MNT_UPDATE; 603 vfs_freeopt(optlist, opt); 604 } 605 else if (strcmp(opt->name, "async") == 0) 606 fsflags |= MNT_ASYNC; 607 else if (strcmp(opt->name, "force") == 0) { 608 fsflags |= MNT_FORCE; 609 vfs_freeopt(optlist, opt); 610 } 611 else if (strcmp(opt->name, "reload") == 0) { 612 fsflags |= MNT_RELOAD; 613 vfs_freeopt(optlist, opt); 614 } 615 else if (strcmp(opt->name, "multilabel") == 0) 616 fsflags |= MNT_MULTILABEL; 617 else if (strcmp(opt->name, "noasync") == 0) 618 fsflags &= ~MNT_ASYNC; 619 else if (strcmp(opt->name, "noatime") == 0) 620 fsflags |= MNT_NOATIME; 621 else if (strcmp(opt->name, "atime") == 0) { 622 free(opt->name, M_MOUNT); 623 opt->name = strdup("nonoatime", M_MOUNT); 624 } 625 else if (strcmp(opt->name, "noclusterr") == 0) 626 fsflags |= MNT_NOCLUSTERR; 627 else if (strcmp(opt->name, "clusterr") == 0) { 628 free(opt->name, M_MOUNT); 629 opt->name = strdup("nonoclusterr", M_MOUNT); 630 } 631 else if (strcmp(opt->name, "noclusterw") == 0) 632 fsflags |= MNT_NOCLUSTERW; 633 else if (strcmp(opt->name, "clusterw") == 0) { 634 free(opt->name, M_MOUNT); 635 opt->name = strdup("nonoclusterw", M_MOUNT); 636 } 637 else if (strcmp(opt->name, "noexec") == 0) 638 fsflags |= MNT_NOEXEC; 639 else if (strcmp(opt->name, "exec") == 0) { 640 free(opt->name, M_MOUNT); 641 opt->name = strdup("nonoexec", M_MOUNT); 642 } 643 else if (strcmp(opt->name, "nosuid") == 0) 644 fsflags |= MNT_NOSUID; 645 else if (strcmp(opt->name, "suid") == 0) { 646 free(opt->name, M_MOUNT); 647 opt->name = strdup("nonosuid", M_MOUNT); 648 } 649 else if (strcmp(opt->name, "nosymfollow") == 0) 650 fsflags |= MNT_NOSYMFOLLOW; 651 else if (strcmp(opt->name, "symfollow") == 0) { 652 free(opt->name, M_MOUNT); 653 opt->name = strdup("nonosymfollow", M_MOUNT); 654 } 655 else if (strcmp(opt->name, "noro") == 0) { 656 fsflags &= ~MNT_RDONLY; 657 has_noro = 1; 658 } 659 else if (strcmp(opt->name, "rw") == 0) { 660 fsflags &= ~MNT_RDONLY; 661 has_rw = 1; 662 } 663 else if (strcmp(opt->name, "ro") == 0) 664 fsflags |= MNT_RDONLY; 665 else if (strcmp(opt->name, "rdonly") == 0) { 666 free(opt->name, M_MOUNT); 667 opt->name = strdup("ro", M_MOUNT); 668 fsflags |= MNT_RDONLY; 669 } 670 else if (strcmp(opt->name, "suiddir") == 0) 671 fsflags |= MNT_SUIDDIR; 672 else if (strcmp(opt->name, "sync") == 0) 673 fsflags |= MNT_SYNCHRONOUS; 674 else if (strcmp(opt->name, "union") == 0) 675 fsflags |= MNT_UNION; 676 } 677 678 /* 679 * If "rw" was specified as a mount option, and we 680 * are trying to update a mount-point from "ro" to "rw", 681 * we need a mount option "noro", since in vfs_mergeopts(), 682 * "noro" will cancel "ro", but "rw" will not do anything. 683 */ 684 if (has_rw && !has_noro) { 685 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 686 noro_opt->name = strdup("noro", M_MOUNT); 687 noro_opt->value = NULL; 688 noro_opt->len = 0; 689 noro_opt->pos = -1; 690 noro_opt->seen = 1; 691 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 692 } 693 694 /* 695 * Be ultra-paranoid about making sure the type and fspath 696 * variables will fit in our mp buffers, including the 697 * terminating NUL. 698 */ 699 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 700 error = ENAMETOOLONG; 701 goto bail; 702 } 703 704 mtx_lock(&Giant); 705 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 706 mtx_unlock(&Giant); 707 bail: 708 /* copyout the errmsg */ 709 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 710 && errmsg_len > 0 && errmsg != NULL) { 711 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 712 bcopy(errmsg, 713 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 714 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 715 } else { 716 copyout(errmsg, 717 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 718 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 719 } 720 } 721 722 if (error != 0) 723 vfs_freeopts(optlist); 724 return (error); 725 } 726 727 /* 728 * Old mount API. 729 */ 730 #ifndef _SYS_SYSPROTO_H_ 731 struct mount_args { 732 char *type; 733 char *path; 734 int flags; 735 caddr_t data; 736 }; 737 #endif 738 /* ARGSUSED */ 739 int 740 mount(td, uap) 741 struct thread *td; 742 struct mount_args /* { 743 char *type; 744 char *path; 745 int flags; 746 caddr_t data; 747 } */ *uap; 748 { 749 char *fstype; 750 struct vfsconf *vfsp = NULL; 751 struct mntarg *ma = NULL; 752 int error; 753 754 AUDIT_ARG(fflags, uap->flags); 755 756 /* 757 * Filter out MNT_ROOTFS. We do not want clients of mount() in 758 * userspace to set this flag, but we must filter it out if we want 759 * MNT_UPDATE on the root file system to work. 760 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 761 */ 762 uap->flags &= ~MNT_ROOTFS; 763 764 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 765 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 766 if (error) { 767 free(fstype, M_TEMP); 768 return (error); 769 } 770 771 AUDIT_ARG(text, fstype); 772 mtx_lock(&Giant); 773 vfsp = vfs_byname_kld(fstype, td, &error); 774 free(fstype, M_TEMP); 775 if (vfsp == NULL) { 776 mtx_unlock(&Giant); 777 return (ENOENT); 778 } 779 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 780 mtx_unlock(&Giant); 781 return (EOPNOTSUPP); 782 } 783 784 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 785 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 786 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 787 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 788 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 789 790 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags); 791 mtx_unlock(&Giant); 792 return (error); 793 } 794 795 796 /* 797 * vfs_domount(): actually attempt a filesystem mount. 798 */ 799 static int 800 vfs_domount( 801 struct thread *td, /* Calling thread. */ 802 const char *fstype, /* Filesystem type. */ 803 char *fspath, /* Mount path. */ 804 int fsflags, /* Flags common to all filesystems. */ 805 void *fsdata /* Options local to the filesystem. */ 806 ) 807 { 808 struct vnode *vp; 809 struct mount *mp; 810 struct vfsconf *vfsp; 811 struct oexport_args oexport; 812 struct export_args export; 813 int error, flag = 0; 814 struct vattr va; 815 struct nameidata nd; 816 817 mtx_assert(&Giant, MA_OWNED); 818 /* 819 * Be ultra-paranoid about making sure the type and fspath 820 * variables will fit in our mp buffers, including the 821 * terminating NUL. 822 */ 823 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 824 return (ENAMETOOLONG); 825 826 if (jailed(td->td_ucred) || usermount == 0) { 827 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 828 return (error); 829 } 830 831 /* 832 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 833 */ 834 if (fsflags & MNT_EXPORTED) { 835 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 836 if (error) 837 return (error); 838 } 839 if (fsflags & MNT_SUIDDIR) { 840 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 841 if (error) 842 return (error); 843 } 844 /* 845 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 846 */ 847 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 848 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 849 fsflags |= MNT_NOSUID | MNT_USER; 850 } 851 852 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 853 vfsp = NULL; 854 if ((fsflags & MNT_UPDATE) == 0) { 855 /* Don't try to load KLDs if we're mounting the root. */ 856 if (fsflags & MNT_ROOTFS) 857 vfsp = vfs_byname(fstype); 858 else 859 vfsp = vfs_byname_kld(fstype, td, &error); 860 if (vfsp == NULL) 861 return (ENODEV); 862 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 863 return (EPERM); 864 } 865 /* 866 * Get vnode to be covered 867 */ 868 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 869 fspath, td); 870 if ((error = namei(&nd)) != 0) 871 return (error); 872 NDFREE(&nd, NDF_ONLY_PNBUF); 873 vp = nd.ni_vp; 874 if (fsflags & MNT_UPDATE) { 875 if ((vp->v_vflag & VV_ROOT) == 0) { 876 vput(vp); 877 return (EINVAL); 878 } 879 mp = vp->v_mount; 880 MNT_ILOCK(mp); 881 flag = mp->mnt_flag; 882 /* 883 * We only allow the filesystem to be reloaded if it 884 * is currently mounted read-only. 885 */ 886 if ((fsflags & MNT_RELOAD) && 887 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 888 MNT_IUNLOCK(mp); 889 vput(vp); 890 return (EOPNOTSUPP); /* Needs translation */ 891 } 892 MNT_IUNLOCK(mp); 893 /* 894 * Only privileged root, or (if MNT_USER is set) the user that 895 * did the original mount is permitted to update it. 896 */ 897 error = vfs_suser(mp, td); 898 if (error) { 899 vput(vp); 900 return (error); 901 } 902 if (vfs_busy(mp, MBF_NOWAIT)) { 903 vput(vp); 904 return (EBUSY); 905 } 906 VI_LOCK(vp); 907 if ((vp->v_iflag & VI_MOUNT) != 0 || 908 vp->v_mountedhere != NULL) { 909 VI_UNLOCK(vp); 910 vfs_unbusy(mp); 911 vput(vp); 912 return (EBUSY); 913 } 914 vp->v_iflag |= VI_MOUNT; 915 VI_UNLOCK(vp); 916 MNT_ILOCK(mp); 917 mp->mnt_flag |= fsflags & 918 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 919 MNT_IUNLOCK(mp); 920 VOP_UNLOCK(vp, 0); 921 mp->mnt_optnew = fsdata; 922 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 923 } else { 924 /* 925 * If the user is not root, ensure that they own the directory 926 * onto which we are attempting to mount. 927 */ 928 error = VOP_GETATTR(vp, &va, td->td_ucred); 929 if (error) { 930 vput(vp); 931 return (error); 932 } 933 if (va.va_uid != td->td_ucred->cr_uid) { 934 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 935 0); 936 if (error) { 937 vput(vp); 938 return (error); 939 } 940 } 941 error = vinvalbuf(vp, V_SAVE, 0, 0); 942 if (error != 0) { 943 vput(vp); 944 return (error); 945 } 946 if (vp->v_type != VDIR) { 947 vput(vp); 948 return (ENOTDIR); 949 } 950 VI_LOCK(vp); 951 if ((vp->v_iflag & VI_MOUNT) != 0 || 952 vp->v_mountedhere != NULL) { 953 VI_UNLOCK(vp); 954 vput(vp); 955 return (EBUSY); 956 } 957 vp->v_iflag |= VI_MOUNT; 958 VI_UNLOCK(vp); 959 960 /* 961 * Allocate and initialize the filesystem. 962 */ 963 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 964 VOP_UNLOCK(vp, 0); 965 966 /* XXXMAC: pass to vfs_mount_alloc? */ 967 mp->mnt_optnew = fsdata; 968 } 969 970 /* 971 * Set the mount level flags. 972 */ 973 MNT_ILOCK(mp); 974 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 975 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 976 MNT_RDONLY)); 977 if ((mp->mnt_flag & MNT_ASYNC) == 0) 978 mp->mnt_kern_flag &= ~MNTK_ASYNC; 979 MNT_IUNLOCK(mp); 980 /* 981 * Mount the filesystem. 982 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 983 * get. No freeing of cn_pnbuf. 984 */ 985 error = VFS_MOUNT(mp); 986 987 /* 988 * Process the export option only if we are 989 * updating mount options. 990 */ 991 if (!error && (fsflags & MNT_UPDATE)) { 992 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 993 sizeof(export)) == 0) 994 error = vfs_export(mp, &export); 995 else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 996 sizeof(oexport)) == 0) { 997 export.ex_flags = oexport.ex_flags; 998 export.ex_root = oexport.ex_root; 999 export.ex_anon = oexport.ex_anon; 1000 export.ex_addr = oexport.ex_addr; 1001 export.ex_addrlen = oexport.ex_addrlen; 1002 export.ex_mask = oexport.ex_mask; 1003 export.ex_masklen = oexport.ex_masklen; 1004 export.ex_indexfile = oexport.ex_indexfile; 1005 export.ex_numsecflavors = 0; 1006 error = vfs_export(mp, &export); 1007 } 1008 } 1009 1010 if (!error) { 1011 if (mp->mnt_opt != NULL) 1012 vfs_freeopts(mp->mnt_opt); 1013 mp->mnt_opt = mp->mnt_optnew; 1014 (void)VFS_STATFS(mp, &mp->mnt_stat); 1015 } 1016 /* 1017 * Prevent external consumers of mount options from reading 1018 * mnt_optnew. 1019 */ 1020 mp->mnt_optnew = NULL; 1021 if (mp->mnt_flag & MNT_UPDATE) { 1022 MNT_ILOCK(mp); 1023 if (error) 1024 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1025 (flag & ~MNT_QUOTA); 1026 else 1027 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1028 MNT_FORCE | MNT_SNAPSHOT); 1029 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1030 mp->mnt_kern_flag |= MNTK_ASYNC; 1031 else 1032 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1033 MNT_IUNLOCK(mp); 1034 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1035 if (mp->mnt_syncer == NULL) 1036 error = vfs_allocate_syncvnode(mp); 1037 } else { 1038 if (mp->mnt_syncer != NULL) 1039 vrele(mp->mnt_syncer); 1040 mp->mnt_syncer = NULL; 1041 } 1042 vfs_unbusy(mp); 1043 VI_LOCK(vp); 1044 vp->v_iflag &= ~VI_MOUNT; 1045 VI_UNLOCK(vp); 1046 vrele(vp); 1047 return (error); 1048 } 1049 MNT_ILOCK(mp); 1050 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1051 mp->mnt_kern_flag |= MNTK_ASYNC; 1052 else 1053 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1054 MNT_IUNLOCK(mp); 1055 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1056 /* 1057 * Put the new filesystem on the mount list after root. 1058 */ 1059 cache_purge(vp); 1060 if (!error) { 1061 struct vnode *newdp; 1062 1063 VI_LOCK(vp); 1064 vp->v_iflag &= ~VI_MOUNT; 1065 VI_UNLOCK(vp); 1066 vp->v_mountedhere = mp; 1067 mtx_lock(&mountlist_mtx); 1068 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1069 mtx_unlock(&mountlist_mtx); 1070 vfs_event_signal(NULL, VQ_MOUNT, 0); 1071 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 1072 panic("mount: lost mount"); 1073 mountcheckdirs(vp, newdp); 1074 vput(newdp); 1075 VOP_UNLOCK(vp, 0); 1076 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1077 error = vfs_allocate_syncvnode(mp); 1078 vfs_unbusy(mp); 1079 if (error) 1080 vrele(vp); 1081 } else { 1082 VI_LOCK(vp); 1083 vp->v_iflag &= ~VI_MOUNT; 1084 VI_UNLOCK(vp); 1085 vfs_unbusy(mp); 1086 vfs_mount_destroy(mp); 1087 vput(vp); 1088 } 1089 return (error); 1090 } 1091 1092 /* 1093 * Unmount a filesystem. 1094 * 1095 * Note: unmount takes a path to the vnode mounted on as argument, not 1096 * special file (as before). 1097 */ 1098 #ifndef _SYS_SYSPROTO_H_ 1099 struct unmount_args { 1100 char *path; 1101 int flags; 1102 }; 1103 #endif 1104 /* ARGSUSED */ 1105 int 1106 unmount(td, uap) 1107 struct thread *td; 1108 register struct unmount_args /* { 1109 char *path; 1110 int flags; 1111 } */ *uap; 1112 { 1113 struct mount *mp; 1114 char *pathbuf; 1115 int error, id0, id1; 1116 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 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1130 mtx_lock(&Giant); 1131 if (uap->flags & MNT_BYFSID) { 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 mtx_lock(&mountlist_mtx); 1148 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1149 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1150 break; 1151 } 1152 mtx_unlock(&mountlist_mtx); 1153 } 1154 free(pathbuf, M_TEMP); 1155 if (mp == NULL) { 1156 /* 1157 * Previously we returned ENOENT for a nonexistent path and 1158 * EINVAL for a non-mountpoint. We cannot tell these apart 1159 * now, so in the !MNT_BYFSID case return the more likely 1160 * EINVAL for compatibility. 1161 */ 1162 mtx_unlock(&Giant); 1163 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1164 } 1165 1166 /* 1167 * Don't allow unmounting the root filesystem. 1168 */ 1169 if (mp->mnt_flag & MNT_ROOTFS) { 1170 mtx_unlock(&Giant); 1171 return (EINVAL); 1172 } 1173 error = dounmount(mp, uap->flags, td); 1174 mtx_unlock(&Giant); 1175 return (error); 1176 } 1177 1178 /* 1179 * Do the actual filesystem unmount. 1180 */ 1181 int 1182 dounmount(mp, flags, td) 1183 struct mount *mp; 1184 int flags; 1185 struct thread *td; 1186 { 1187 struct vnode *coveredvp, *fsrootvp; 1188 int error; 1189 int async_flag; 1190 int mnt_gen_r; 1191 1192 mtx_assert(&Giant, MA_OWNED); 1193 1194 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1195 mnt_gen_r = mp->mnt_gen; 1196 VI_LOCK(coveredvp); 1197 vholdl(coveredvp); 1198 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1199 vdrop(coveredvp); 1200 /* 1201 * Check for mp being unmounted while waiting for the 1202 * covered vnode lock. 1203 */ 1204 if (coveredvp->v_mountedhere != mp || 1205 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1206 VOP_UNLOCK(coveredvp, 0); 1207 return (EBUSY); 1208 } 1209 } 1210 /* 1211 * Only privileged root, or (if MNT_USER is set) the user that did the 1212 * original mount is permitted to unmount this filesystem. 1213 */ 1214 error = vfs_suser(mp, td); 1215 if (error) { 1216 if (coveredvp) 1217 VOP_UNLOCK(coveredvp, 0); 1218 return (error); 1219 } 1220 1221 MNT_ILOCK(mp); 1222 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1223 MNT_IUNLOCK(mp); 1224 if (coveredvp) 1225 VOP_UNLOCK(coveredvp, 0); 1226 return (EBUSY); 1227 } 1228 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1229 /* Allow filesystems to detect that a forced unmount is in progress. */ 1230 if (flags & MNT_FORCE) 1231 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1232 error = 0; 1233 if (mp->mnt_lockref) { 1234 if ((flags & MNT_FORCE) == 0) { 1235 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1236 MNTK_UNMOUNTF); 1237 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1238 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1239 wakeup(mp); 1240 } 1241 MNT_IUNLOCK(mp); 1242 if (coveredvp) 1243 VOP_UNLOCK(coveredvp, 0); 1244 return (EBUSY); 1245 } 1246 mp->mnt_kern_flag |= MNTK_DRAINING; 1247 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1248 "mount drain", 0); 1249 } 1250 MNT_IUNLOCK(mp); 1251 KASSERT(mp->mnt_lockref == 0, 1252 ("%s: invalid lock refcount in the drain path @ %s:%d", 1253 __func__, __FILE__, __LINE__)); 1254 KASSERT(error == 0, 1255 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1256 __func__, __FILE__, __LINE__)); 1257 vn_start_write(NULL, &mp, V_WAIT); 1258 1259 if (mp->mnt_flag & MNT_EXPUBLIC) 1260 vfs_setpublicfs(NULL, NULL, NULL); 1261 1262 vfs_msync(mp, MNT_WAIT); 1263 MNT_ILOCK(mp); 1264 async_flag = mp->mnt_flag & MNT_ASYNC; 1265 mp->mnt_flag &= ~MNT_ASYNC; 1266 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1267 MNT_IUNLOCK(mp); 1268 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1269 if (mp->mnt_syncer != NULL) 1270 vrele(mp->mnt_syncer); 1271 /* 1272 * For forced unmounts, move process cdir/rdir refs on the fs root 1273 * vnode to the covered vnode. For non-forced unmounts we want 1274 * such references to cause an EBUSY error. 1275 */ 1276 if ((flags & MNT_FORCE) && 1277 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1278 if (mp->mnt_vnodecovered != NULL) 1279 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1280 if (fsrootvp == rootvnode) { 1281 vrele(rootvnode); 1282 rootvnode = NULL; 1283 } 1284 vput(fsrootvp); 1285 } 1286 if (((mp->mnt_flag & MNT_RDONLY) || 1287 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1288 error = VFS_UNMOUNT(mp, flags); 1289 vn_finished_write(mp); 1290 /* 1291 * If we failed to flush the dirty blocks for this mount point, 1292 * undo all the cdir/rdir and rootvnode changes we made above. 1293 * Unless we failed to do so because the device is reporting that 1294 * it doesn't exist anymore. 1295 */ 1296 if (error && error != ENXIO) { 1297 if ((flags & MNT_FORCE) && 1298 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1299 if (mp->mnt_vnodecovered != NULL) 1300 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1301 if (rootvnode == NULL) { 1302 rootvnode = fsrootvp; 1303 vref(rootvnode); 1304 } 1305 vput(fsrootvp); 1306 } 1307 MNT_ILOCK(mp); 1308 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1309 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1310 MNT_IUNLOCK(mp); 1311 (void) vfs_allocate_syncvnode(mp); 1312 MNT_ILOCK(mp); 1313 } 1314 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1315 mp->mnt_flag |= async_flag; 1316 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1317 mp->mnt_kern_flag |= MNTK_ASYNC; 1318 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1319 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1320 wakeup(mp); 1321 } 1322 MNT_IUNLOCK(mp); 1323 if (coveredvp) 1324 VOP_UNLOCK(coveredvp, 0); 1325 return (error); 1326 } 1327 mtx_lock(&mountlist_mtx); 1328 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1329 mtx_unlock(&mountlist_mtx); 1330 if (coveredvp != NULL) { 1331 coveredvp->v_mountedhere = NULL; 1332 vput(coveredvp); 1333 } 1334 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1335 vfs_mount_destroy(mp); 1336 return (0); 1337 } 1338 1339 /* 1340 * --------------------------------------------------------------------- 1341 * Mounting of root filesystem 1342 * 1343 */ 1344 1345 struct root_hold_token { 1346 const char *who; 1347 LIST_ENTRY(root_hold_token) list; 1348 }; 1349 1350 static LIST_HEAD(, root_hold_token) root_holds = 1351 LIST_HEAD_INITIALIZER(&root_holds); 1352 1353 static int root_mount_complete; 1354 1355 /* 1356 * Hold root mount. 1357 */ 1358 struct root_hold_token * 1359 root_mount_hold(const char *identifier) 1360 { 1361 struct root_hold_token *h; 1362 1363 if (root_mounted()) 1364 return (NULL); 1365 1366 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1367 h->who = identifier; 1368 mtx_lock(&mountlist_mtx); 1369 LIST_INSERT_HEAD(&root_holds, h, list); 1370 mtx_unlock(&mountlist_mtx); 1371 return (h); 1372 } 1373 1374 /* 1375 * Release root mount. 1376 */ 1377 void 1378 root_mount_rel(struct root_hold_token *h) 1379 { 1380 1381 if (h == NULL) 1382 return; 1383 mtx_lock(&mountlist_mtx); 1384 LIST_REMOVE(h, list); 1385 wakeup(&root_holds); 1386 mtx_unlock(&mountlist_mtx); 1387 free(h, M_DEVBUF); 1388 } 1389 1390 /* 1391 * Wait for all subsystems to release root mount. 1392 */ 1393 static void 1394 root_mount_prepare(void) 1395 { 1396 struct root_hold_token *h; 1397 struct timeval lastfail; 1398 int curfail = 0; 1399 1400 for (;;) { 1401 DROP_GIANT(); 1402 g_waitidle(); 1403 PICKUP_GIANT(); 1404 mtx_lock(&mountlist_mtx); 1405 if (LIST_EMPTY(&root_holds)) { 1406 mtx_unlock(&mountlist_mtx); 1407 break; 1408 } 1409 if (ppsratecheck(&lastfail, &curfail, 1)) { 1410 printf("Root mount waiting for:"); 1411 LIST_FOREACH(h, &root_holds, list) 1412 printf(" %s", h->who); 1413 printf("\n"); 1414 } 1415 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1416 hz); 1417 } 1418 } 1419 1420 /* 1421 * Root was mounted, share the good news. 1422 */ 1423 static void 1424 root_mount_done(void) 1425 { 1426 1427 /* Keep prison0's root in sync with the global rootvnode. */ 1428 mtx_lock(&prison0.pr_mtx); 1429 prison0.pr_root = rootvnode; 1430 vref(prison0.pr_root); 1431 mtx_unlock(&prison0.pr_mtx); 1432 /* 1433 * Use a mutex to prevent the wakeup being missed and waiting for 1434 * an extra 1 second sleep. 1435 */ 1436 mtx_lock(&mountlist_mtx); 1437 root_mount_complete = 1; 1438 wakeup(&root_mount_complete); 1439 mtx_unlock(&mountlist_mtx); 1440 } 1441 1442 /* 1443 * Return true if root is already mounted. 1444 */ 1445 int 1446 root_mounted(void) 1447 { 1448 1449 /* No mutex is acquired here because int stores are atomic. */ 1450 return (root_mount_complete); 1451 } 1452 1453 /* 1454 * Wait until root is mounted. 1455 */ 1456 void 1457 root_mount_wait(void) 1458 { 1459 1460 /* 1461 * Panic on an obvious deadlock - the function can't be called from 1462 * a thread which is doing the whole SYSINIT stuff. 1463 */ 1464 KASSERT(curthread->td_proc->p_pid != 0, 1465 ("root_mount_wait: cannot be called from the swapper thread")); 1466 mtx_lock(&mountlist_mtx); 1467 while (!root_mount_complete) { 1468 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1469 hz); 1470 } 1471 mtx_unlock(&mountlist_mtx); 1472 } 1473 1474 static void 1475 set_rootvnode() 1476 { 1477 struct proc *p; 1478 1479 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode)) 1480 panic("Cannot find root vnode"); 1481 1482 p = curthread->td_proc; 1483 FILEDESC_XLOCK(p->p_fd); 1484 1485 if (p->p_fd->fd_cdir != NULL) 1486 vrele(p->p_fd->fd_cdir); 1487 p->p_fd->fd_cdir = rootvnode; 1488 VREF(rootvnode); 1489 1490 if (p->p_fd->fd_rdir != NULL) 1491 vrele(p->p_fd->fd_rdir); 1492 p->p_fd->fd_rdir = rootvnode; 1493 VREF(rootvnode); 1494 1495 FILEDESC_XUNLOCK(p->p_fd); 1496 1497 VOP_UNLOCK(rootvnode, 0); 1498 1499 EVENTHANDLER_INVOKE(mountroot); 1500 } 1501 1502 /* 1503 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1504 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1505 */ 1506 1507 static void 1508 devfs_first(void) 1509 { 1510 struct thread *td = curthread; 1511 struct vfsoptlist *opts; 1512 struct vfsconf *vfsp; 1513 struct mount *mp = NULL; 1514 int error; 1515 1516 vfsp = vfs_byname("devfs"); 1517 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1518 if (vfsp == NULL) 1519 return; 1520 1521 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 1522 1523 error = VFS_MOUNT(mp); 1524 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1525 if (error) 1526 return; 1527 1528 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1529 TAILQ_INIT(opts); 1530 mp->mnt_opt = opts; 1531 1532 mtx_lock(&mountlist_mtx); 1533 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1534 mtx_unlock(&mountlist_mtx); 1535 1536 set_rootvnode(); 1537 1538 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1539 if (error) 1540 printf("kern_symlink /dev -> / returns %d\n", error); 1541 } 1542 1543 /* 1544 * Surgically move our devfs to be mounted on /dev. 1545 */ 1546 1547 static void 1548 devfs_fixup(struct thread *td) 1549 { 1550 struct nameidata nd; 1551 int error; 1552 struct vnode *vp, *dvp; 1553 struct mount *mp; 1554 1555 /* Remove our devfs mount from the mountlist and purge the cache */ 1556 mtx_lock(&mountlist_mtx); 1557 mp = TAILQ_FIRST(&mountlist); 1558 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1559 mtx_unlock(&mountlist_mtx); 1560 cache_purgevfs(mp); 1561 1562 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp); 1563 VI_LOCK(dvp); 1564 dvp->v_iflag &= ~VI_MOUNT; 1565 VI_UNLOCK(dvp); 1566 dvp->v_mountedhere = NULL; 1567 1568 /* Set up the real rootvnode, and purge the cache */ 1569 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1570 set_rootvnode(); 1571 cache_purgevfs(rootvnode->v_mount); 1572 1573 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1574 error = namei(&nd); 1575 if (error) { 1576 printf("Lookup of /dev for devfs, error: %d\n", error); 1577 return; 1578 } 1579 NDFREE(&nd, NDF_ONLY_PNBUF); 1580 vp = nd.ni_vp; 1581 if (vp->v_type != VDIR) { 1582 vput(vp); 1583 } 1584 error = vinvalbuf(vp, V_SAVE, 0, 0); 1585 if (error) { 1586 vput(vp); 1587 } 1588 cache_purge(vp); 1589 mp->mnt_vnodecovered = vp; 1590 vp->v_mountedhere = mp; 1591 mtx_lock(&mountlist_mtx); 1592 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1593 mtx_unlock(&mountlist_mtx); 1594 VOP_UNLOCK(vp, 0); 1595 vput(dvp); 1596 vfs_unbusy(mp); 1597 1598 /* Unlink the no longer needed /dev/dev -> / symlink */ 1599 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1600 } 1601 1602 /* 1603 * Report errors during filesystem mounting. 1604 */ 1605 void 1606 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1607 { 1608 struct vfsoptlist *moptlist = mp->mnt_optnew; 1609 va_list ap; 1610 int error, len; 1611 char *errmsg; 1612 1613 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1614 if (error || errmsg == NULL || len <= 0) 1615 return; 1616 1617 va_start(ap, fmt); 1618 vsnprintf(errmsg, (size_t)len, fmt, ap); 1619 va_end(ap); 1620 } 1621 1622 void 1623 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1624 { 1625 va_list ap; 1626 int error, len; 1627 char *errmsg; 1628 1629 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1630 if (error || errmsg == NULL || len <= 0) 1631 return; 1632 1633 va_start(ap, fmt); 1634 vsnprintf(errmsg, (size_t)len, fmt, ap); 1635 va_end(ap); 1636 } 1637 1638 /* 1639 * Find and mount the root filesystem 1640 */ 1641 void 1642 vfs_mountroot(void) 1643 { 1644 char *cp, *options; 1645 int error, i, asked = 0; 1646 1647 options = NULL; 1648 1649 root_mount_prepare(); 1650 1651 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1652 NULL, NULL, mount_init, mount_fini, 1653 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1654 devfs_first(); 1655 1656 /* 1657 * We are booted with instructions to prompt for the root filesystem. 1658 */ 1659 if (boothowto & RB_ASKNAME) { 1660 if (!vfs_mountroot_ask()) 1661 goto mounted; 1662 asked = 1; 1663 } 1664 1665 options = getenv("vfs.root.mountfrom.options"); 1666 1667 /* 1668 * The root filesystem information is compiled in, and we are 1669 * booted with instructions to use it. 1670 */ 1671 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1672 if (!vfs_mountroot_try(ctrootdevname, options)) 1673 goto mounted; 1674 ctrootdevname = NULL; 1675 } 1676 1677 /* 1678 * We've been given the generic "use CDROM as root" flag. This is 1679 * necessary because one media may be used in many different 1680 * devices, so we need to search for them. 1681 */ 1682 if (boothowto & RB_CDROM) { 1683 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1684 if (!vfs_mountroot_try(cdrom_rootdevnames[i], options)) 1685 goto mounted; 1686 } 1687 } 1688 1689 /* 1690 * Try to use the value read by the loader from /etc/fstab, or 1691 * supplied via some other means. This is the preferred 1692 * mechanism. 1693 */ 1694 cp = getenv("vfs.root.mountfrom"); 1695 if (cp != NULL) { 1696 error = vfs_mountroot_try(cp, options); 1697 freeenv(cp); 1698 if (!error) 1699 goto mounted; 1700 } 1701 1702 /* 1703 * Try values that may have been computed by code during boot 1704 */ 1705 if (!vfs_mountroot_try(rootdevnames[0], options)) 1706 goto mounted; 1707 if (!vfs_mountroot_try(rootdevnames[1], options)) 1708 goto mounted; 1709 1710 /* 1711 * If we (still) have a compiled-in default, try it. 1712 */ 1713 if (ctrootdevname != NULL) 1714 if (!vfs_mountroot_try(ctrootdevname, options)) 1715 goto mounted; 1716 /* 1717 * Everything so far has failed, prompt on the console if we haven't 1718 * already tried that. 1719 */ 1720 if (!asked) 1721 if (!vfs_mountroot_ask()) 1722 goto mounted; 1723 1724 panic("Root mount failed, startup aborted."); 1725 1726 mounted: 1727 root_mount_done(); 1728 freeenv(options); 1729 } 1730 1731 static struct mntarg * 1732 parse_mountroot_options(struct mntarg *ma, const char *options) 1733 { 1734 char *p; 1735 char *name, *name_arg; 1736 char *val, *val_arg; 1737 char *opts; 1738 1739 if (options == NULL || options[0] == '\0') 1740 return (ma); 1741 1742 p = opts = strdup(options, M_MOUNT); 1743 if (opts == NULL) { 1744 return (ma); 1745 } 1746 1747 while((name = strsep(&p, ",")) != NULL) { 1748 if (name[0] == '\0') 1749 break; 1750 1751 val = strchr(name, '='); 1752 if (val != NULL) { 1753 *val = '\0'; 1754 ++val; 1755 } 1756 if( strcmp(name, "rw") == 0 || 1757 strcmp(name, "noro") == 0) { 1758 /* 1759 * The first time we mount the root file system, 1760 * we need to mount 'ro', so We need to ignore 1761 * 'rw' and 'noro' mount options. 1762 */ 1763 continue; 1764 } 1765 name_arg = strdup(name, M_MOUNT); 1766 val_arg = NULL; 1767 if (val != NULL) 1768 val_arg = strdup(val, M_MOUNT); 1769 1770 ma = mount_arg(ma, name_arg, val_arg, 1771 (val_arg != NULL ? -1 : 0)); 1772 } 1773 free(opts, M_MOUNT); 1774 return (ma); 1775 } 1776 1777 /* 1778 * Mount (mountfrom) as the root filesystem. 1779 */ 1780 static int 1781 vfs_mountroot_try(const char *mountfrom, const char *options) 1782 { 1783 struct mount *mp; 1784 struct mntarg *ma; 1785 char *vfsname, *path; 1786 time_t timebase; 1787 int error; 1788 char patt[32]; 1789 char errmsg[255]; 1790 1791 vfsname = NULL; 1792 path = NULL; 1793 mp = NULL; 1794 ma = NULL; 1795 error = EINVAL; 1796 bzero(errmsg, sizeof(errmsg)); 1797 1798 if (mountfrom == NULL) 1799 return (error); /* don't complain */ 1800 printf("Trying to mount root from %s\n", mountfrom); 1801 1802 /* parse vfs name and path */ 1803 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1804 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1805 vfsname[0] = path[0] = 0; 1806 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1807 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1808 goto out; 1809 1810 if (path[0] == '\0') 1811 strcpy(path, ROOTNAME); 1812 1813 ma = mount_arg(ma, "fstype", vfsname, -1); 1814 ma = mount_arg(ma, "fspath", "/", -1); 1815 ma = mount_arg(ma, "from", path, -1); 1816 ma = mount_arg(ma, "errmsg", errmsg, sizeof(errmsg)); 1817 ma = mount_arg(ma, "ro", NULL, 0); 1818 ma = parse_mountroot_options(ma, options); 1819 error = kernel_mount(ma, MNT_ROOTFS); 1820 1821 if (error == 0) { 1822 /* 1823 * We mount devfs prior to mounting the / FS, so the first 1824 * entry will typically be devfs. 1825 */ 1826 mp = TAILQ_FIRST(&mountlist); 1827 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1828 1829 /* 1830 * Iterate over all currently mounted file systems and use 1831 * the time stamp found to check and/or initialize the RTC. 1832 * Typically devfs has no time stamp and the only other FS 1833 * is the actual / FS. 1834 * Call inittodr() only once and pass it the largest of the 1835 * timestamps we encounter. 1836 */ 1837 timebase = 0; 1838 do { 1839 if (mp->mnt_time > timebase) 1840 timebase = mp->mnt_time; 1841 mp = TAILQ_NEXT(mp, mnt_list); 1842 } while (mp != NULL); 1843 inittodr(timebase); 1844 1845 devfs_fixup(curthread); 1846 } 1847 1848 if (error != 0 ) { 1849 printf("ROOT MOUNT ERROR: %s\n", errmsg); 1850 printf("If you have invalid mount options, reboot, and "); 1851 printf("first try the following from\n"); 1852 printf("the loader prompt:\n\n"); 1853 printf(" set vfs.root.mountfrom.options=rw\n\n"); 1854 printf("and then remove invalid mount options from "); 1855 printf("/etc/fstab.\n\n"); 1856 } 1857 out: 1858 free(path, M_MOUNT); 1859 free(vfsname, M_MOUNT); 1860 return (error); 1861 } 1862 1863 /* 1864 * --------------------------------------------------------------------- 1865 * Interactive root filesystem selection code. 1866 */ 1867 1868 static int 1869 vfs_mountroot_ask(void) 1870 { 1871 char name[128]; 1872 char *mountfrom; 1873 char *options; 1874 1875 for(;;) { 1876 printf("Loader variables:\n"); 1877 printf("vfs.root.mountfrom="); 1878 mountfrom = getenv("vfs.root.mountfrom"); 1879 if (mountfrom != NULL) { 1880 printf("%s", mountfrom); 1881 } 1882 printf("\n"); 1883 printf("vfs.root.mountfrom.options="); 1884 options = getenv("vfs.root.mountfrom.options"); 1885 if (options != NULL) { 1886 printf("%s", options); 1887 } 1888 printf("\n"); 1889 freeenv(mountfrom); 1890 freeenv(options); 1891 printf("\nManual root filesystem specification:\n"); 1892 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1893 printf(" eg. ufs:/dev/da0s1a\n"); 1894 printf(" eg. cd9660:/dev/acd0\n"); 1895 printf(" This is equivalent to: "); 1896 printf("mount -t cd9660 /dev/acd0 /\n"); 1897 printf("\n"); 1898 printf(" ? List valid disk boot devices\n"); 1899 printf(" <empty line> Abort manual input\n"); 1900 printf("\nmountroot> "); 1901 gets(name, sizeof(name), 1); 1902 if (name[0] == '\0') 1903 return (1); 1904 if (name[0] == '?') { 1905 printf("\nList of GEOM managed disk devices:\n "); 1906 g_dev_print(); 1907 continue; 1908 } 1909 if (!vfs_mountroot_try(name, NULL)) 1910 return (0); 1911 } 1912 } 1913 1914 /* 1915 * --------------------------------------------------------------------- 1916 * Functions for querying mount options/arguments from filesystems. 1917 */ 1918 1919 /* 1920 * Check that no unknown options are given 1921 */ 1922 int 1923 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1924 { 1925 struct vfsopt *opt; 1926 char errmsg[255]; 1927 const char **t, *p, *q; 1928 int ret = 0; 1929 1930 TAILQ_FOREACH(opt, opts, link) { 1931 p = opt->name; 1932 q = NULL; 1933 if (p[0] == 'n' && p[1] == 'o') 1934 q = p + 2; 1935 for(t = global_opts; *t != NULL; t++) { 1936 if (strcmp(*t, p) == 0) 1937 break; 1938 if (q != NULL) { 1939 if (strcmp(*t, q) == 0) 1940 break; 1941 } 1942 } 1943 if (*t != NULL) 1944 continue; 1945 for(t = legal; *t != NULL; t++) { 1946 if (strcmp(*t, p) == 0) 1947 break; 1948 if (q != NULL) { 1949 if (strcmp(*t, q) == 0) 1950 break; 1951 } 1952 } 1953 if (*t != NULL) 1954 continue; 1955 snprintf(errmsg, sizeof(errmsg), 1956 "mount option <%s> is unknown", p); 1957 printf("%s\n", errmsg); 1958 ret = EINVAL; 1959 } 1960 if (ret != 0) { 1961 TAILQ_FOREACH(opt, opts, link) { 1962 if (strcmp(opt->name, "errmsg") == 0) { 1963 strncpy((char *)opt->value, errmsg, opt->len); 1964 } 1965 } 1966 } 1967 return (ret); 1968 } 1969 1970 /* 1971 * Get a mount option by its name. 1972 * 1973 * Return 0 if the option was found, ENOENT otherwise. 1974 * If len is non-NULL it will be filled with the length 1975 * of the option. If buf is non-NULL, it will be filled 1976 * with the address of the option. 1977 */ 1978 int 1979 vfs_getopt(opts, name, buf, len) 1980 struct vfsoptlist *opts; 1981 const char *name; 1982 void **buf; 1983 int *len; 1984 { 1985 struct vfsopt *opt; 1986 1987 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1988 1989 TAILQ_FOREACH(opt, opts, link) { 1990 if (strcmp(name, opt->name) == 0) { 1991 opt->seen = 1; 1992 if (len != NULL) 1993 *len = opt->len; 1994 if (buf != NULL) 1995 *buf = opt->value; 1996 return (0); 1997 } 1998 } 1999 return (ENOENT); 2000 } 2001 2002 int 2003 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2004 { 2005 struct vfsopt *opt; 2006 2007 if (opts == NULL) 2008 return (-1); 2009 2010 TAILQ_FOREACH(opt, opts, link) { 2011 if (strcmp(name, opt->name) == 0) { 2012 opt->seen = 1; 2013 return (opt->pos); 2014 } 2015 } 2016 return (-1); 2017 } 2018 2019 char * 2020 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2021 { 2022 struct vfsopt *opt; 2023 2024 *error = 0; 2025 TAILQ_FOREACH(opt, opts, link) { 2026 if (strcmp(name, opt->name) != 0) 2027 continue; 2028 opt->seen = 1; 2029 if (opt->len == 0 || 2030 ((char *)opt->value)[opt->len - 1] != '\0') { 2031 *error = EINVAL; 2032 return (NULL); 2033 } 2034 return (opt->value); 2035 } 2036 *error = ENOENT; 2037 return (NULL); 2038 } 2039 2040 int 2041 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 2042 { 2043 struct vfsopt *opt; 2044 2045 TAILQ_FOREACH(opt, opts, link) { 2046 if (strcmp(name, opt->name) == 0) { 2047 opt->seen = 1; 2048 if (w != NULL) 2049 *w |= val; 2050 return (1); 2051 } 2052 } 2053 if (w != NULL) 2054 *w &= ~val; 2055 return (0); 2056 } 2057 2058 int 2059 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2060 { 2061 va_list ap; 2062 struct vfsopt *opt; 2063 int ret; 2064 2065 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2066 2067 TAILQ_FOREACH(opt, opts, link) { 2068 if (strcmp(name, opt->name) != 0) 2069 continue; 2070 opt->seen = 1; 2071 if (opt->len == 0 || opt->value == NULL) 2072 return (0); 2073 if (((char *)opt->value)[opt->len - 1] != '\0') 2074 return (0); 2075 va_start(ap, fmt); 2076 ret = vsscanf(opt->value, fmt, ap); 2077 va_end(ap); 2078 return (ret); 2079 } 2080 return (0); 2081 } 2082 2083 int 2084 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2085 { 2086 struct vfsopt *opt; 2087 2088 TAILQ_FOREACH(opt, opts, link) { 2089 if (strcmp(name, opt->name) != 0) 2090 continue; 2091 opt->seen = 1; 2092 if (opt->value == NULL) 2093 opt->len = len; 2094 else { 2095 if (opt->len != len) 2096 return (EINVAL); 2097 bcopy(value, opt->value, len); 2098 } 2099 return (0); 2100 } 2101 return (ENOENT); 2102 } 2103 2104 int 2105 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2106 { 2107 struct vfsopt *opt; 2108 2109 TAILQ_FOREACH(opt, opts, link) { 2110 if (strcmp(name, opt->name) != 0) 2111 continue; 2112 opt->seen = 1; 2113 if (opt->value == NULL) 2114 opt->len = len; 2115 else { 2116 if (opt->len < len) 2117 return (EINVAL); 2118 opt->len = len; 2119 bcopy(value, opt->value, len); 2120 } 2121 return (0); 2122 } 2123 return (ENOENT); 2124 } 2125 2126 int 2127 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2128 { 2129 struct vfsopt *opt; 2130 2131 TAILQ_FOREACH(opt, opts, link) { 2132 if (strcmp(name, opt->name) != 0) 2133 continue; 2134 opt->seen = 1; 2135 if (opt->value == NULL) 2136 opt->len = strlen(value) + 1; 2137 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2138 return (EINVAL); 2139 return (0); 2140 } 2141 return (ENOENT); 2142 } 2143 2144 /* 2145 * Find and copy a mount option. 2146 * 2147 * The size of the buffer has to be specified 2148 * in len, if it is not the same length as the 2149 * mount option, EINVAL is returned. 2150 * Returns ENOENT if the option is not found. 2151 */ 2152 int 2153 vfs_copyopt(opts, name, dest, len) 2154 struct vfsoptlist *opts; 2155 const char *name; 2156 void *dest; 2157 int len; 2158 { 2159 struct vfsopt *opt; 2160 2161 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2162 2163 TAILQ_FOREACH(opt, opts, link) { 2164 if (strcmp(name, opt->name) == 0) { 2165 opt->seen = 1; 2166 if (len != opt->len) 2167 return (EINVAL); 2168 bcopy(opt->value, dest, opt->len); 2169 return (0); 2170 } 2171 } 2172 return (ENOENT); 2173 } 2174 2175 /* 2176 * This is a helper function for filesystems to traverse their 2177 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2178 */ 2179 2180 struct vnode * 2181 __mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2182 { 2183 struct vnode *vp; 2184 2185 mtx_assert(MNT_MTX(mp), MA_OWNED); 2186 2187 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2188 if ((*mvp)->v_yield++ == 500) { 2189 MNT_IUNLOCK(mp); 2190 (*mvp)->v_yield = 0; 2191 uio_yield(); 2192 MNT_ILOCK(mp); 2193 } 2194 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2195 while (vp != NULL && vp->v_type == VMARKER) 2196 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2197 2198 /* Check if we are done */ 2199 if (vp == NULL) { 2200 __mnt_vnode_markerfree(mvp, mp); 2201 return (NULL); 2202 } 2203 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2204 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2205 return (vp); 2206 } 2207 2208 struct vnode * 2209 __mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2210 { 2211 struct vnode *vp; 2212 2213 mtx_assert(MNT_MTX(mp), MA_OWNED); 2214 2215 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2216 while (vp != NULL && vp->v_type == VMARKER) 2217 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2218 2219 /* Check if we are done */ 2220 if (vp == NULL) { 2221 *mvp = NULL; 2222 return (NULL); 2223 } 2224 MNT_REF(mp); 2225 MNT_IUNLOCK(mp); 2226 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2227 M_VNODE_MARKER, 2228 M_WAITOK | M_ZERO); 2229 MNT_ILOCK(mp); 2230 (*mvp)->v_type = VMARKER; 2231 2232 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2233 while (vp != NULL && vp->v_type == VMARKER) 2234 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2235 2236 /* Check if we are done */ 2237 if (vp == NULL) { 2238 MNT_IUNLOCK(mp); 2239 free(*mvp, M_VNODE_MARKER); 2240 MNT_ILOCK(mp); 2241 *mvp = NULL; 2242 MNT_REL(mp); 2243 return (NULL); 2244 } 2245 (*mvp)->v_mount = mp; 2246 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2247 return (vp); 2248 } 2249 2250 2251 void 2252 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2253 { 2254 2255 if (*mvp == NULL) 2256 return; 2257 2258 mtx_assert(MNT_MTX(mp), MA_OWNED); 2259 2260 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2261 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2262 MNT_IUNLOCK(mp); 2263 free(*mvp, M_VNODE_MARKER); 2264 MNT_ILOCK(mp); 2265 *mvp = NULL; 2266 MNT_REL(mp); 2267 } 2268 2269 2270 int 2271 __vfs_statfs(struct mount *mp, struct statfs *sbp) 2272 { 2273 int error; 2274 2275 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 2276 if (sbp != &mp->mnt_stat) 2277 *sbp = mp->mnt_stat; 2278 return (error); 2279 } 2280 2281 void 2282 vfs_mountedfrom(struct mount *mp, const char *from) 2283 { 2284 2285 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2286 strlcpy(mp->mnt_stat.f_mntfromname, from, 2287 sizeof mp->mnt_stat.f_mntfromname); 2288 } 2289 2290 /* 2291 * --------------------------------------------------------------------- 2292 * This is the api for building mount args and mounting filesystems from 2293 * inside the kernel. 2294 * 2295 * The API works by accumulation of individual args. First error is 2296 * latched. 2297 * 2298 * XXX: should be documented in new manpage kernel_mount(9) 2299 */ 2300 2301 /* A memory allocation which must be freed when we are done */ 2302 struct mntaarg { 2303 SLIST_ENTRY(mntaarg) next; 2304 }; 2305 2306 /* The header for the mount arguments */ 2307 struct mntarg { 2308 struct iovec *v; 2309 int len; 2310 int error; 2311 SLIST_HEAD(, mntaarg) list; 2312 }; 2313 2314 /* 2315 * Add a boolean argument. 2316 * 2317 * flag is the boolean value. 2318 * name must start with "no". 2319 */ 2320 struct mntarg * 2321 mount_argb(struct mntarg *ma, int flag, const char *name) 2322 { 2323 2324 KASSERT(name[0] == 'n' && name[1] == 'o', 2325 ("mount_argb(...,%s): name must start with 'no'", name)); 2326 2327 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2328 } 2329 2330 /* 2331 * Add an argument printf style 2332 */ 2333 struct mntarg * 2334 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2335 { 2336 va_list ap; 2337 struct mntaarg *maa; 2338 struct sbuf *sb; 2339 int len; 2340 2341 if (ma == NULL) { 2342 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2343 SLIST_INIT(&ma->list); 2344 } 2345 if (ma->error) 2346 return (ma); 2347 2348 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2349 M_MOUNT, M_WAITOK); 2350 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2351 ma->v[ma->len].iov_len = strlen(name) + 1; 2352 ma->len++; 2353 2354 sb = sbuf_new_auto(); 2355 va_start(ap, fmt); 2356 sbuf_vprintf(sb, fmt, ap); 2357 va_end(ap); 2358 sbuf_finish(sb); 2359 len = sbuf_len(sb) + 1; 2360 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2361 SLIST_INSERT_HEAD(&ma->list, maa, next); 2362 bcopy(sbuf_data(sb), maa + 1, len); 2363 sbuf_delete(sb); 2364 2365 ma->v[ma->len].iov_base = maa + 1; 2366 ma->v[ma->len].iov_len = len; 2367 ma->len++; 2368 2369 return (ma); 2370 } 2371 2372 /* 2373 * Add an argument which is a userland string. 2374 */ 2375 struct mntarg * 2376 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2377 { 2378 struct mntaarg *maa; 2379 char *tbuf; 2380 2381 if (val == NULL) 2382 return (ma); 2383 if (ma == NULL) { 2384 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2385 SLIST_INIT(&ma->list); 2386 } 2387 if (ma->error) 2388 return (ma); 2389 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2390 SLIST_INSERT_HEAD(&ma->list, maa, next); 2391 tbuf = (void *)(maa + 1); 2392 ma->error = copyinstr(val, tbuf, len, NULL); 2393 return (mount_arg(ma, name, tbuf, -1)); 2394 } 2395 2396 /* 2397 * Plain argument. 2398 * 2399 * If length is -1, treat value as a C string. 2400 */ 2401 struct mntarg * 2402 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2403 { 2404 2405 if (ma == NULL) { 2406 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2407 SLIST_INIT(&ma->list); 2408 } 2409 if (ma->error) 2410 return (ma); 2411 2412 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2413 M_MOUNT, M_WAITOK); 2414 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2415 ma->v[ma->len].iov_len = strlen(name) + 1; 2416 ma->len++; 2417 2418 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2419 if (len < 0) 2420 ma->v[ma->len].iov_len = strlen(val) + 1; 2421 else 2422 ma->v[ma->len].iov_len = len; 2423 ma->len++; 2424 return (ma); 2425 } 2426 2427 /* 2428 * Free a mntarg structure 2429 */ 2430 static void 2431 free_mntarg(struct mntarg *ma) 2432 { 2433 struct mntaarg *maa; 2434 2435 while (!SLIST_EMPTY(&ma->list)) { 2436 maa = SLIST_FIRST(&ma->list); 2437 SLIST_REMOVE_HEAD(&ma->list, next); 2438 free(maa, M_MOUNT); 2439 } 2440 free(ma->v, M_MOUNT); 2441 free(ma, M_MOUNT); 2442 } 2443 2444 /* 2445 * Mount a filesystem 2446 */ 2447 int 2448 kernel_mount(struct mntarg *ma, int flags) 2449 { 2450 struct uio auio; 2451 int error; 2452 2453 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2454 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2455 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2456 2457 auio.uio_iov = ma->v; 2458 auio.uio_iovcnt = ma->len; 2459 auio.uio_segflg = UIO_SYSSPACE; 2460 2461 error = ma->error; 2462 if (!error) 2463 error = vfs_donmount(curthread, flags, &auio); 2464 free_mntarg(ma); 2465 return (error); 2466 } 2467 2468 /* 2469 * A printflike function to mount a filesystem. 2470 */ 2471 int 2472 kernel_vmount(int flags, ...) 2473 { 2474 struct mntarg *ma = NULL; 2475 va_list ap; 2476 const char *cp; 2477 const void *vp; 2478 int error; 2479 2480 va_start(ap, flags); 2481 for (;;) { 2482 cp = va_arg(ap, const char *); 2483 if (cp == NULL) 2484 break; 2485 vp = va_arg(ap, const void *); 2486 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2487 } 2488 va_end(ap); 2489 2490 error = kernel_mount(ma, flags); 2491 return (error); 2492 } 2493