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 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); 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>] 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 */ 113 114 /* 115 * Global opts, taken by all filesystems 116 */ 117 static const char *global_opts[] = { 118 "errmsg", 119 "fstype", 120 "fspath", 121 "ro", 122 "rw", 123 "nosuid", 124 "noexec", 125 NULL 126 }; 127 128 /* 129 * The root specifiers we will try if RB_CDROM is specified. 130 */ 131 static char *cdrom_rootdevnames[] = { 132 "cd9660:cd0", 133 "cd9660:acd0", 134 NULL 135 }; 136 137 /* legacy find-root code */ 138 char *rootdevnames[2] = {NULL, NULL}; 139 #ifndef ROOTDEVNAME 140 # define ROOTDEVNAME NULL 141 #endif 142 static const char *ctrootdevname = ROOTDEVNAME; 143 144 /* 145 * --------------------------------------------------------------------- 146 * Functions for building and sanitizing the mount options 147 */ 148 149 /* Remove one mount option. */ 150 static void 151 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 152 { 153 154 TAILQ_REMOVE(opts, opt, link); 155 free(opt->name, M_MOUNT); 156 if (opt->value != NULL) 157 free(opt->value, M_MOUNT); 158 free(opt, M_MOUNT); 159 } 160 161 /* Release all resources related to the mount options. */ 162 void 163 vfs_freeopts(struct vfsoptlist *opts) 164 { 165 struct vfsopt *opt; 166 167 while (!TAILQ_EMPTY(opts)) { 168 opt = TAILQ_FIRST(opts); 169 vfs_freeopt(opts, opt); 170 } 171 free(opts, M_MOUNT); 172 } 173 174 void 175 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 176 { 177 struct vfsopt *opt, *temp; 178 179 if (opts == NULL) 180 return; 181 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 182 if (strcmp(opt->name, name) == 0) 183 vfs_freeopt(opts, opt); 184 } 185 } 186 187 /* 188 * Check if options are equal (with or without the "no" prefix). 189 */ 190 static int 191 vfs_equalopts(const char *opt1, const char *opt2) 192 { 193 char *p; 194 195 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 196 if (strcmp(opt1, opt2) == 0) 197 return (1); 198 /* "noopt" vs. "opt" */ 199 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 200 return (1); 201 /* "opt" vs. "noopt" */ 202 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 203 return (1); 204 while ((p = strchr(opt1, '.')) != NULL && 205 !strncmp(opt1, opt2, ++p - opt1)) { 206 opt2 += p - opt1; 207 opt1 = p; 208 /* "foo.noopt" vs. "foo.opt" */ 209 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 210 return (1); 211 /* "foo.opt" vs. "foo.noopt" */ 212 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 213 return (1); 214 } 215 return (0); 216 } 217 218 /* 219 * If a mount option is specified several times, 220 * (with or without the "no" prefix) only keep 221 * the last occurence of it. 222 */ 223 static void 224 vfs_sanitizeopts(struct vfsoptlist *opts) 225 { 226 struct vfsopt *opt, *opt2, *tmp; 227 228 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 229 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 230 while (opt2 != NULL) { 231 if (vfs_equalopts(opt->name, opt2->name)) { 232 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 233 vfs_freeopt(opts, opt2); 234 opt2 = tmp; 235 } else { 236 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 237 } 238 } 239 } 240 } 241 242 /* 243 * Build a linked list of mount options from a struct uio. 244 */ 245 int 246 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 247 { 248 struct vfsoptlist *opts; 249 struct vfsopt *opt; 250 size_t memused, namelen, optlen; 251 unsigned int i, iovcnt; 252 int error; 253 254 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 255 TAILQ_INIT(opts); 256 memused = 0; 257 iovcnt = auio->uio_iovcnt; 258 for (i = 0; i < iovcnt; i += 2) { 259 namelen = auio->uio_iov[i].iov_len; 260 optlen = auio->uio_iov[i + 1].iov_len; 261 memused += sizeof(struct vfsopt) + optlen + namelen; 262 /* 263 * Avoid consuming too much memory, and attempts to overflow 264 * memused. 265 */ 266 if (memused > VFS_MOUNTARG_SIZE_MAX || 267 optlen > VFS_MOUNTARG_SIZE_MAX || 268 namelen > VFS_MOUNTARG_SIZE_MAX) { 269 error = EINVAL; 270 goto bad; 271 } 272 273 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 274 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 275 opt->value = NULL; 276 opt->len = 0; 277 opt->pos = i / 2; 278 opt->seen = 0; 279 280 /* 281 * Do this early, so jumps to "bad" will free the current 282 * option. 283 */ 284 TAILQ_INSERT_TAIL(opts, opt, link); 285 286 if (auio->uio_segflg == UIO_SYSSPACE) { 287 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 288 } else { 289 error = copyin(auio->uio_iov[i].iov_base, opt->name, 290 namelen); 291 if (error) 292 goto bad; 293 } 294 /* Ensure names are null-terminated strings. */ 295 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 296 error = EINVAL; 297 goto bad; 298 } 299 if (optlen != 0) { 300 opt->len = optlen; 301 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 302 if (auio->uio_segflg == UIO_SYSSPACE) { 303 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 304 optlen); 305 } else { 306 error = copyin(auio->uio_iov[i + 1].iov_base, 307 opt->value, optlen); 308 if (error) 309 goto bad; 310 } 311 } 312 } 313 vfs_sanitizeopts(opts); 314 *options = opts; 315 return (0); 316 bad: 317 vfs_freeopts(opts); 318 return (error); 319 } 320 321 /* 322 * Merge the old mount options with the new ones passed 323 * in the MNT_UPDATE case. 324 * 325 * XXX This function will keep a "nofoo" option in the 326 * new options if there is no matching "foo" option 327 * to be cancelled in the old options. This is a bug 328 * if the option's canonical name is "foo". E.g., "noro" 329 * shouldn't end up in the mount point's active options, 330 * but it can. 331 */ 332 static void 333 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 334 { 335 struct vfsopt *opt, *opt2, *new; 336 337 TAILQ_FOREACH(opt, opts, link) { 338 /* 339 * Check that this option hasn't been redefined 340 * nor cancelled with a "no" mount option. 341 */ 342 opt2 = TAILQ_FIRST(toopts); 343 while (opt2 != NULL) { 344 if (strcmp(opt2->name, opt->name) == 0) 345 goto next; 346 if (strncmp(opt2->name, "no", 2) == 0 && 347 strcmp(opt2->name + 2, opt->name) == 0) { 348 vfs_freeopt(toopts, opt2); 349 goto next; 350 } 351 opt2 = TAILQ_NEXT(opt2, link); 352 } 353 /* We want this option, duplicate it. */ 354 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 355 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 356 strcpy(new->name, opt->name); 357 if (opt->len != 0) { 358 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 359 bcopy(opt->value, new->value, opt->len); 360 } else { 361 new->value = NULL; 362 } 363 new->len = opt->len; 364 new->seen = opt->seen; 365 TAILQ_INSERT_TAIL(toopts, new, link); 366 next: 367 continue; 368 } 369 } 370 371 /* 372 * Mount a filesystem. 373 */ 374 int 375 nmount(td, uap) 376 struct thread *td; 377 struct nmount_args /* { 378 struct iovec *iovp; 379 unsigned int iovcnt; 380 int flags; 381 } */ *uap; 382 { 383 struct uio *auio; 384 int error; 385 u_int iovcnt; 386 387 AUDIT_ARG(fflags, uap->flags); 388 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 389 uap->iovp, uap->iovcnt, uap->flags); 390 391 /* 392 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 393 * userspace to set this flag, but we must filter it out if we want 394 * MNT_UPDATE on the root file system to work. 395 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 396 */ 397 uap->flags &= ~MNT_ROOTFS; 398 399 iovcnt = uap->iovcnt; 400 /* 401 * Check that we have an even number of iovec's 402 * and that we have at least two options. 403 */ 404 if ((iovcnt & 1) || (iovcnt < 4)) { 405 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 406 uap->iovcnt); 407 return (EINVAL); 408 } 409 410 error = copyinuio(uap->iovp, iovcnt, &auio); 411 if (error) { 412 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 413 __func__, error); 414 return (error); 415 } 416 error = vfs_donmount(td, uap->flags, auio); 417 418 free(auio, M_IOV); 419 return (error); 420 } 421 422 /* 423 * --------------------------------------------------------------------- 424 * Various utility functions 425 */ 426 427 void 428 vfs_ref(struct mount *mp) 429 { 430 431 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 432 MNT_ILOCK(mp); 433 MNT_REF(mp); 434 MNT_IUNLOCK(mp); 435 } 436 437 void 438 vfs_rel(struct mount *mp) 439 { 440 441 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 442 MNT_ILOCK(mp); 443 MNT_REL(mp); 444 MNT_IUNLOCK(mp); 445 } 446 447 static int 448 mount_init(void *mem, int size, int flags) 449 { 450 struct mount *mp; 451 452 mp = (struct mount *)mem; 453 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 454 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 455 return (0); 456 } 457 458 static void 459 mount_fini(void *mem, int size) 460 { 461 struct mount *mp; 462 463 mp = (struct mount *)mem; 464 lockdestroy(&mp->mnt_explock); 465 mtx_destroy(&mp->mnt_mtx); 466 } 467 468 /* 469 * Allocate and initialize the mount point struct. 470 */ 471 struct mount * 472 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 473 struct ucred *cred) 474 { 475 struct mount *mp; 476 477 mp = uma_zalloc(mount_zone, M_WAITOK); 478 bzero(&mp->mnt_startzero, 479 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 480 TAILQ_INIT(&mp->mnt_nvnodelist); 481 mp->mnt_nvnodelistsize = 0; 482 mp->mnt_ref = 0; 483 (void) vfs_busy(mp, MBF_NOWAIT); 484 mp->mnt_op = vfsp->vfc_vfsops; 485 mp->mnt_vfc = vfsp; 486 vfsp->vfc_refcount++; /* XXX Unlocked */ 487 mp->mnt_stat.f_type = vfsp->vfc_typenum; 488 mp->mnt_gen++; 489 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 490 mp->mnt_vnodecovered = vp; 491 mp->mnt_cred = crdup(cred); 492 mp->mnt_stat.f_owner = cred->cr_uid; 493 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 494 mp->mnt_iosize_max = DFLTPHYS; 495 #ifdef MAC 496 mac_mount_init(mp); 497 mac_mount_create(cred, mp); 498 #endif 499 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 500 return (mp); 501 } 502 503 /* 504 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 505 */ 506 void 507 vfs_mount_destroy(struct mount *mp) 508 { 509 510 MNT_ILOCK(mp); 511 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 512 if (mp->mnt_kern_flag & MNTK_MWAIT) { 513 mp->mnt_kern_flag &= ~MNTK_MWAIT; 514 wakeup(mp); 515 } 516 while (mp->mnt_ref) 517 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 518 KASSERT(mp->mnt_ref == 0, 519 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 520 __FILE__, __LINE__)); 521 if (mp->mnt_writeopcount != 0) 522 panic("vfs_mount_destroy: nonzero writeopcount"); 523 if (mp->mnt_secondary_writes != 0) 524 panic("vfs_mount_destroy: nonzero secondary_writes"); 525 mp->mnt_vfc->vfc_refcount--; 526 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 527 struct vnode *vp; 528 529 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 530 vprint("", vp); 531 panic("unmount: dangling vnode"); 532 } 533 if (mp->mnt_nvnodelistsize != 0) 534 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 535 if (mp->mnt_lockref != 0) 536 panic("vfs_mount_destroy: nonzero lock refcount"); 537 MNT_IUNLOCK(mp); 538 #ifdef MAC 539 mac_mount_destroy(mp); 540 #endif 541 if (mp->mnt_opt != NULL) 542 vfs_freeopts(mp->mnt_opt); 543 crfree(mp->mnt_cred); 544 uma_zfree(mount_zone, mp); 545 } 546 547 int 548 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 549 { 550 struct vfsoptlist *optlist; 551 struct vfsopt *opt, *noro_opt, *tmp_opt; 552 char *fstype, *fspath, *errmsg; 553 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 554 int has_rw, has_noro; 555 556 errmsg = fspath = NULL; 557 errmsg_len = has_noro = has_rw = fspathlen = 0; 558 errmsg_pos = -1; 559 560 error = vfs_buildopts(fsoptions, &optlist); 561 if (error) 562 return (error); 563 564 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 565 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 566 567 /* 568 * We need these two options before the others, 569 * and they are mandatory for any filesystem. 570 * Ensure they are NUL terminated as well. 571 */ 572 fstypelen = 0; 573 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 574 if (error || fstype[fstypelen - 1] != '\0') { 575 error = EINVAL; 576 if (errmsg != NULL) 577 strncpy(errmsg, "Invalid fstype", errmsg_len); 578 goto bail; 579 } 580 fspathlen = 0; 581 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 582 if (error || fspath[fspathlen - 1] != '\0') { 583 error = EINVAL; 584 if (errmsg != NULL) 585 strncpy(errmsg, "Invalid fspath", errmsg_len); 586 goto bail; 587 } 588 589 /* 590 * We need to see if we have the "update" option 591 * before we call vfs_domount(), since vfs_domount() has special 592 * logic based on MNT_UPDATE. This is very important 593 * when we want to update the root filesystem. 594 */ 595 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 596 if (strcmp(opt->name, "update") == 0) { 597 fsflags |= MNT_UPDATE; 598 vfs_freeopt(optlist, opt); 599 } 600 else if (strcmp(opt->name, "async") == 0) 601 fsflags |= MNT_ASYNC; 602 else if (strcmp(opt->name, "force") == 0) { 603 fsflags |= MNT_FORCE; 604 vfs_freeopt(optlist, opt); 605 } 606 else if (strcmp(opt->name, "reload") == 0) { 607 fsflags |= MNT_RELOAD; 608 vfs_freeopt(optlist, opt); 609 } 610 else if (strcmp(opt->name, "multilabel") == 0) 611 fsflags |= MNT_MULTILABEL; 612 else if (strcmp(opt->name, "noasync") == 0) 613 fsflags &= ~MNT_ASYNC; 614 else if (strcmp(opt->name, "noatime") == 0) 615 fsflags |= MNT_NOATIME; 616 else if (strcmp(opt->name, "atime") == 0) { 617 free(opt->name, M_MOUNT); 618 opt->name = strdup("nonoatime", M_MOUNT); 619 } 620 else if (strcmp(opt->name, "noclusterr") == 0) 621 fsflags |= MNT_NOCLUSTERR; 622 else if (strcmp(opt->name, "clusterr") == 0) { 623 free(opt->name, M_MOUNT); 624 opt->name = strdup("nonoclusterr", M_MOUNT); 625 } 626 else if (strcmp(opt->name, "noclusterw") == 0) 627 fsflags |= MNT_NOCLUSTERW; 628 else if (strcmp(opt->name, "clusterw") == 0) { 629 free(opt->name, M_MOUNT); 630 opt->name = strdup("nonoclusterw", M_MOUNT); 631 } 632 else if (strcmp(opt->name, "noexec") == 0) 633 fsflags |= MNT_NOEXEC; 634 else if (strcmp(opt->name, "exec") == 0) { 635 free(opt->name, M_MOUNT); 636 opt->name = strdup("nonoexec", M_MOUNT); 637 } 638 else if (strcmp(opt->name, "nosuid") == 0) 639 fsflags |= MNT_NOSUID; 640 else if (strcmp(opt->name, "suid") == 0) { 641 free(opt->name, M_MOUNT); 642 opt->name = strdup("nonosuid", M_MOUNT); 643 } 644 else if (strcmp(opt->name, "nosymfollow") == 0) 645 fsflags |= MNT_NOSYMFOLLOW; 646 else if (strcmp(opt->name, "symfollow") == 0) { 647 free(opt->name, M_MOUNT); 648 opt->name = strdup("nonosymfollow", M_MOUNT); 649 } 650 else if (strcmp(opt->name, "noro") == 0) { 651 fsflags &= ~MNT_RDONLY; 652 has_noro = 1; 653 } 654 else if (strcmp(opt->name, "rw") == 0) { 655 fsflags &= ~MNT_RDONLY; 656 has_rw = 1; 657 } 658 else if (strcmp(opt->name, "ro") == 0) 659 fsflags |= MNT_RDONLY; 660 else if (strcmp(opt->name, "rdonly") == 0) { 661 free(opt->name, M_MOUNT); 662 opt->name = strdup("ro", M_MOUNT); 663 fsflags |= MNT_RDONLY; 664 } 665 else if (strcmp(opt->name, "suiddir") == 0) 666 fsflags |= MNT_SUIDDIR; 667 else if (strcmp(opt->name, "sync") == 0) 668 fsflags |= MNT_SYNCHRONOUS; 669 else if (strcmp(opt->name, "union") == 0) 670 fsflags |= MNT_UNION; 671 } 672 673 /* 674 * If "rw" was specified as a mount option, and we 675 * are trying to update a mount-point from "ro" to "rw", 676 * we need a mount option "noro", since in vfs_mergeopts(), 677 * "noro" will cancel "ro", but "rw" will not do anything. 678 */ 679 if (has_rw && !has_noro) { 680 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 681 noro_opt->name = strdup("noro", M_MOUNT); 682 noro_opt->value = NULL; 683 noro_opt->len = 0; 684 noro_opt->pos = -1; 685 noro_opt->seen = 1; 686 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 687 } 688 689 /* 690 * Be ultra-paranoid about making sure the type and fspath 691 * variables will fit in our mp buffers, including the 692 * terminating NUL. 693 */ 694 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 695 error = ENAMETOOLONG; 696 goto bail; 697 } 698 699 mtx_lock(&Giant); 700 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 701 mtx_unlock(&Giant); 702 bail: 703 /* copyout the errmsg */ 704 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 705 && errmsg_len > 0 && errmsg != NULL) { 706 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 707 bcopy(errmsg, 708 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 709 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 710 } else { 711 copyout(errmsg, 712 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 713 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 714 } 715 } 716 717 if (error != 0) 718 vfs_freeopts(optlist); 719 return (error); 720 } 721 722 /* 723 * Old mount API. 724 */ 725 #ifndef _SYS_SYSPROTO_H_ 726 struct mount_args { 727 char *type; 728 char *path; 729 int flags; 730 caddr_t data; 731 }; 732 #endif 733 /* ARGSUSED */ 734 int 735 mount(td, uap) 736 struct thread *td; 737 struct mount_args /* { 738 char *type; 739 char *path; 740 int flags; 741 caddr_t data; 742 } */ *uap; 743 { 744 char *fstype; 745 struct vfsconf *vfsp = NULL; 746 struct mntarg *ma = NULL; 747 int error; 748 749 AUDIT_ARG(fflags, uap->flags); 750 751 /* 752 * Filter out MNT_ROOTFS. We do not want clients of mount() in 753 * userspace to set this flag, but we must filter it out if we want 754 * MNT_UPDATE on the root file system to work. 755 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 756 */ 757 uap->flags &= ~MNT_ROOTFS; 758 759 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 760 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 761 if (error) { 762 free(fstype, M_TEMP); 763 return (error); 764 } 765 766 AUDIT_ARG(text, fstype); 767 mtx_lock(&Giant); 768 vfsp = vfs_byname_kld(fstype, td, &error); 769 free(fstype, M_TEMP); 770 if (vfsp == NULL) { 771 mtx_unlock(&Giant); 772 return (ENOENT); 773 } 774 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 775 mtx_unlock(&Giant); 776 return (EOPNOTSUPP); 777 } 778 779 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 780 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 781 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 782 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 783 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 784 785 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 786 mtx_unlock(&Giant); 787 return (error); 788 } 789 790 791 /* 792 * vfs_domount(): actually attempt a filesystem mount. 793 */ 794 static int 795 vfs_domount( 796 struct thread *td, /* Calling thread. */ 797 const char *fstype, /* Filesystem type. */ 798 char *fspath, /* Mount path. */ 799 int fsflags, /* Flags common to all filesystems. */ 800 void *fsdata /* Options local to the filesystem. */ 801 ) 802 { 803 struct vnode *vp; 804 struct mount *mp; 805 struct vfsconf *vfsp; 806 struct oexport_args oexport; 807 struct export_args export; 808 int error, flag = 0; 809 struct vattr va; 810 struct nameidata nd; 811 812 mtx_assert(&Giant, MA_OWNED); 813 /* 814 * Be ultra-paranoid about making sure the type and fspath 815 * variables will fit in our mp buffers, including the 816 * terminating NUL. 817 */ 818 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 819 return (ENAMETOOLONG); 820 821 if (jailed(td->td_ucred) || usermount == 0) { 822 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 823 return (error); 824 } 825 826 /* 827 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 828 */ 829 if (fsflags & MNT_EXPORTED) { 830 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 831 if (error) 832 return (error); 833 } 834 if (fsflags & MNT_SUIDDIR) { 835 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 836 if (error) 837 return (error); 838 } 839 /* 840 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 841 */ 842 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 843 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 844 fsflags |= MNT_NOSUID | MNT_USER; 845 } 846 847 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 848 vfsp = NULL; 849 if ((fsflags & MNT_UPDATE) == 0) { 850 /* Don't try to load KLDs if we're mounting the root. */ 851 if (fsflags & MNT_ROOTFS) 852 vfsp = vfs_byname(fstype); 853 else 854 vfsp = vfs_byname_kld(fstype, td, &error); 855 if (vfsp == NULL) 856 return (ENODEV); 857 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 858 return (EPERM); 859 } 860 /* 861 * Get vnode to be covered 862 */ 863 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 864 fspath, td); 865 if ((error = namei(&nd)) != 0) 866 return (error); 867 NDFREE(&nd, NDF_ONLY_PNBUF); 868 vp = nd.ni_vp; 869 if (fsflags & MNT_UPDATE) { 870 if ((vp->v_vflag & VV_ROOT) == 0) { 871 vput(vp); 872 return (EINVAL); 873 } 874 mp = vp->v_mount; 875 MNT_ILOCK(mp); 876 flag = mp->mnt_flag; 877 /* 878 * We only allow the filesystem to be reloaded if it 879 * is currently mounted read-only. 880 */ 881 if ((fsflags & MNT_RELOAD) && 882 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 883 MNT_IUNLOCK(mp); 884 vput(vp); 885 return (EOPNOTSUPP); /* Needs translation */ 886 } 887 MNT_IUNLOCK(mp); 888 /* 889 * Only privileged root, or (if MNT_USER is set) the user that 890 * did the original mount is permitted to update it. 891 */ 892 error = vfs_suser(mp, td); 893 if (error) { 894 vput(vp); 895 return (error); 896 } 897 if (vfs_busy(mp, MBF_NOWAIT)) { 898 vput(vp); 899 return (EBUSY); 900 } 901 VI_LOCK(vp); 902 if ((vp->v_iflag & VI_MOUNT) != 0 || 903 vp->v_mountedhere != NULL) { 904 VI_UNLOCK(vp); 905 vfs_unbusy(mp); 906 vput(vp); 907 return (EBUSY); 908 } 909 vp->v_iflag |= VI_MOUNT; 910 VI_UNLOCK(vp); 911 MNT_ILOCK(mp); 912 mp->mnt_flag |= fsflags & 913 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 914 MNT_IUNLOCK(mp); 915 VOP_UNLOCK(vp, 0); 916 mp->mnt_optnew = fsdata; 917 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 918 } else { 919 /* 920 * If the user is not root, ensure that they own the directory 921 * onto which we are attempting to mount. 922 */ 923 error = VOP_GETATTR(vp, &va, td->td_ucred); 924 if (error) { 925 vput(vp); 926 return (error); 927 } 928 if (va.va_uid != td->td_ucred->cr_uid) { 929 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 930 0); 931 if (error) { 932 vput(vp); 933 return (error); 934 } 935 } 936 error = vinvalbuf(vp, V_SAVE, 0, 0); 937 if (error != 0) { 938 vput(vp); 939 return (error); 940 } 941 if (vp->v_type != VDIR) { 942 vput(vp); 943 return (ENOTDIR); 944 } 945 VI_LOCK(vp); 946 if ((vp->v_iflag & VI_MOUNT) != 0 || 947 vp->v_mountedhere != NULL) { 948 VI_UNLOCK(vp); 949 vput(vp); 950 return (EBUSY); 951 } 952 vp->v_iflag |= VI_MOUNT; 953 VI_UNLOCK(vp); 954 955 /* 956 * Allocate and initialize the filesystem. 957 */ 958 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 959 VOP_UNLOCK(vp, 0); 960 961 /* XXXMAC: pass to vfs_mount_alloc? */ 962 mp->mnt_optnew = fsdata; 963 } 964 965 /* 966 * Set the mount level flags. 967 */ 968 MNT_ILOCK(mp); 969 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 970 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 971 MNT_RDONLY)); 972 if ((mp->mnt_flag & MNT_ASYNC) == 0) 973 mp->mnt_kern_flag &= ~MNTK_ASYNC; 974 MNT_IUNLOCK(mp); 975 /* 976 * Mount the filesystem. 977 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 978 * get. No freeing of cn_pnbuf. 979 */ 980 error = VFS_MOUNT(mp, td); 981 982 /* 983 * Process the export option only if we are 984 * updating mount options. 985 */ 986 if (!error && (fsflags & MNT_UPDATE)) { 987 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 988 sizeof(export)) == 0) 989 error = vfs_export(mp, &export); 990 else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 991 sizeof(oexport)) == 0) { 992 export.ex_flags = oexport.ex_flags; 993 export.ex_root = oexport.ex_root; 994 export.ex_anon = oexport.ex_anon; 995 export.ex_addr = oexport.ex_addr; 996 export.ex_addrlen = oexport.ex_addrlen; 997 export.ex_mask = oexport.ex_mask; 998 export.ex_masklen = oexport.ex_masklen; 999 export.ex_indexfile = oexport.ex_indexfile; 1000 export.ex_numsecflavors = 0; 1001 error = vfs_export(mp, &export); 1002 } 1003 } 1004 1005 if (!error) { 1006 if (mp->mnt_opt != NULL) 1007 vfs_freeopts(mp->mnt_opt); 1008 mp->mnt_opt = mp->mnt_optnew; 1009 (void)VFS_STATFS(mp, &mp->mnt_stat, td); 1010 } 1011 /* 1012 * Prevent external consumers of mount options from reading 1013 * mnt_optnew. 1014 */ 1015 mp->mnt_optnew = NULL; 1016 if (mp->mnt_flag & MNT_UPDATE) { 1017 MNT_ILOCK(mp); 1018 if (error) 1019 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1020 (flag & ~MNT_QUOTA); 1021 else 1022 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1023 MNT_FORCE | MNT_SNAPSHOT); 1024 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1025 mp->mnt_kern_flag |= MNTK_ASYNC; 1026 else 1027 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1028 MNT_IUNLOCK(mp); 1029 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1030 if (mp->mnt_syncer == NULL) 1031 error = vfs_allocate_syncvnode(mp); 1032 } else { 1033 if (mp->mnt_syncer != NULL) 1034 vrele(mp->mnt_syncer); 1035 mp->mnt_syncer = NULL; 1036 } 1037 vfs_unbusy(mp); 1038 VI_LOCK(vp); 1039 vp->v_iflag &= ~VI_MOUNT; 1040 VI_UNLOCK(vp); 1041 vrele(vp); 1042 return (error); 1043 } 1044 MNT_ILOCK(mp); 1045 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1046 mp->mnt_kern_flag |= MNTK_ASYNC; 1047 else 1048 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1049 MNT_IUNLOCK(mp); 1050 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1051 /* 1052 * Put the new filesystem on the mount list after root. 1053 */ 1054 cache_purge(vp); 1055 if (!error) { 1056 struct vnode *newdp; 1057 1058 VI_LOCK(vp); 1059 vp->v_iflag &= ~VI_MOUNT; 1060 VI_UNLOCK(vp); 1061 vp->v_mountedhere = mp; 1062 mtx_lock(&mountlist_mtx); 1063 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1064 mtx_unlock(&mountlist_mtx); 1065 vfs_event_signal(NULL, VQ_MOUNT, 0); 1066 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 1067 panic("mount: lost mount"); 1068 mountcheckdirs(vp, newdp); 1069 vput(newdp); 1070 VOP_UNLOCK(vp, 0); 1071 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1072 error = vfs_allocate_syncvnode(mp); 1073 vfs_unbusy(mp); 1074 if (error) 1075 vrele(vp); 1076 } else { 1077 VI_LOCK(vp); 1078 vp->v_iflag &= ~VI_MOUNT; 1079 VI_UNLOCK(vp); 1080 vfs_unbusy(mp); 1081 vfs_mount_destroy(mp); 1082 vput(vp); 1083 } 1084 return (error); 1085 } 1086 1087 /* 1088 * Unmount a filesystem. 1089 * 1090 * Note: unmount takes a path to the vnode mounted on as argument, not 1091 * special file (as before). 1092 */ 1093 #ifndef _SYS_SYSPROTO_H_ 1094 struct unmount_args { 1095 char *path; 1096 int flags; 1097 }; 1098 #endif 1099 /* ARGSUSED */ 1100 int 1101 unmount(td, uap) 1102 struct thread *td; 1103 register struct unmount_args /* { 1104 char *path; 1105 int flags; 1106 } */ *uap; 1107 { 1108 struct mount *mp; 1109 char *pathbuf; 1110 int error, id0, id1; 1111 1112 if (jailed(td->td_ucred) || usermount == 0) { 1113 error = priv_check(td, PRIV_VFS_UNMOUNT); 1114 if (error) 1115 return (error); 1116 } 1117 1118 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1119 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1120 if (error) { 1121 free(pathbuf, M_TEMP); 1122 return (error); 1123 } 1124 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1125 mtx_lock(&Giant); 1126 if (uap->flags & MNT_BYFSID) { 1127 /* Decode the filesystem ID. */ 1128 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1129 mtx_unlock(&Giant); 1130 free(pathbuf, M_TEMP); 1131 return (EINVAL); 1132 } 1133 1134 mtx_lock(&mountlist_mtx); 1135 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1136 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1137 mp->mnt_stat.f_fsid.val[1] == id1) 1138 break; 1139 } 1140 mtx_unlock(&mountlist_mtx); 1141 } else { 1142 mtx_lock(&mountlist_mtx); 1143 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1144 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1145 break; 1146 } 1147 mtx_unlock(&mountlist_mtx); 1148 } 1149 free(pathbuf, M_TEMP); 1150 if (mp == NULL) { 1151 /* 1152 * Previously we returned ENOENT for a nonexistent path and 1153 * EINVAL for a non-mountpoint. We cannot tell these apart 1154 * now, so in the !MNT_BYFSID case return the more likely 1155 * EINVAL for compatibility. 1156 */ 1157 mtx_unlock(&Giant); 1158 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1159 } 1160 1161 /* 1162 * Don't allow unmounting the root filesystem. 1163 */ 1164 if (mp->mnt_flag & MNT_ROOTFS) { 1165 mtx_unlock(&Giant); 1166 return (EINVAL); 1167 } 1168 error = dounmount(mp, uap->flags, td); 1169 mtx_unlock(&Giant); 1170 return (error); 1171 } 1172 1173 /* 1174 * Do the actual filesystem unmount. 1175 */ 1176 int 1177 dounmount(mp, flags, td) 1178 struct mount *mp; 1179 int flags; 1180 struct thread *td; 1181 { 1182 struct vnode *coveredvp, *fsrootvp; 1183 int error; 1184 int async_flag; 1185 int mnt_gen_r; 1186 1187 mtx_assert(&Giant, MA_OWNED); 1188 1189 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1190 mnt_gen_r = mp->mnt_gen; 1191 VI_LOCK(coveredvp); 1192 vholdl(coveredvp); 1193 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1194 vdrop(coveredvp); 1195 /* 1196 * Check for mp being unmounted while waiting for the 1197 * covered vnode lock. 1198 */ 1199 if (coveredvp->v_mountedhere != mp || 1200 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1201 VOP_UNLOCK(coveredvp, 0); 1202 return (EBUSY); 1203 } 1204 } 1205 /* 1206 * Only privileged root, or (if MNT_USER is set) the user that did the 1207 * original mount is permitted to unmount this filesystem. 1208 */ 1209 error = vfs_suser(mp, td); 1210 if (error) { 1211 if (coveredvp) 1212 VOP_UNLOCK(coveredvp, 0); 1213 return (error); 1214 } 1215 1216 MNT_ILOCK(mp); 1217 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1218 MNT_IUNLOCK(mp); 1219 if (coveredvp) 1220 VOP_UNLOCK(coveredvp, 0); 1221 return (EBUSY); 1222 } 1223 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1224 /* Allow filesystems to detect that a forced unmount is in progress. */ 1225 if (flags & MNT_FORCE) 1226 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1227 error = 0; 1228 if (mp->mnt_lockref) { 1229 if ((flags & MNT_FORCE) == 0) { 1230 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1231 MNTK_UNMOUNTF); 1232 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1233 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1234 wakeup(mp); 1235 } 1236 MNT_IUNLOCK(mp); 1237 if (coveredvp) 1238 VOP_UNLOCK(coveredvp, 0); 1239 return (EBUSY); 1240 } 1241 mp->mnt_kern_flag |= MNTK_DRAINING; 1242 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1243 "mount drain", 0); 1244 } 1245 MNT_IUNLOCK(mp); 1246 KASSERT(mp->mnt_lockref == 0, 1247 ("%s: invalid lock refcount in the drain path @ %s:%d", 1248 __func__, __FILE__, __LINE__)); 1249 KASSERT(error == 0, 1250 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1251 __func__, __FILE__, __LINE__)); 1252 vn_start_write(NULL, &mp, V_WAIT); 1253 1254 if (mp->mnt_flag & MNT_EXPUBLIC) 1255 vfs_setpublicfs(NULL, NULL, NULL); 1256 1257 vfs_msync(mp, MNT_WAIT); 1258 MNT_ILOCK(mp); 1259 async_flag = mp->mnt_flag & MNT_ASYNC; 1260 mp->mnt_flag &= ~MNT_ASYNC; 1261 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1262 MNT_IUNLOCK(mp); 1263 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1264 if (mp->mnt_syncer != NULL) 1265 vrele(mp->mnt_syncer); 1266 /* 1267 * For forced unmounts, move process cdir/rdir refs on the fs root 1268 * vnode to the covered vnode. For non-forced unmounts we want 1269 * such references to cause an EBUSY error. 1270 */ 1271 if ((flags & MNT_FORCE) && 1272 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1273 if (mp->mnt_vnodecovered != NULL) 1274 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1275 if (fsrootvp == rootvnode) { 1276 vrele(rootvnode); 1277 rootvnode = NULL; 1278 } 1279 vput(fsrootvp); 1280 } 1281 if (((mp->mnt_flag & MNT_RDONLY) || 1282 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1283 (flags & MNT_FORCE)) { 1284 error = VFS_UNMOUNT(mp, flags, td); 1285 } 1286 vn_finished_write(mp); 1287 /* 1288 * If we failed to flush the dirty blocks for this mount point, 1289 * undo all the cdir/rdir and rootvnode changes we made above. 1290 * Unless we failed to do so because the device is reporting that 1291 * it doesn't exist anymore. 1292 */ 1293 if (error && error != ENXIO) { 1294 if ((flags & MNT_FORCE) && 1295 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1296 if (mp->mnt_vnodecovered != NULL) 1297 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1298 if (rootvnode == NULL) { 1299 rootvnode = fsrootvp; 1300 vref(rootvnode); 1301 } 1302 vput(fsrootvp); 1303 } 1304 MNT_ILOCK(mp); 1305 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1306 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1307 MNT_IUNLOCK(mp); 1308 (void) vfs_allocate_syncvnode(mp); 1309 MNT_ILOCK(mp); 1310 } 1311 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1312 mp->mnt_flag |= async_flag; 1313 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1314 mp->mnt_kern_flag |= MNTK_ASYNC; 1315 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1316 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1317 wakeup(mp); 1318 } 1319 MNT_IUNLOCK(mp); 1320 if (coveredvp) 1321 VOP_UNLOCK(coveredvp, 0); 1322 return (error); 1323 } 1324 mtx_lock(&mountlist_mtx); 1325 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1326 mtx_unlock(&mountlist_mtx); 1327 if (coveredvp != NULL) { 1328 coveredvp->v_mountedhere = NULL; 1329 vput(coveredvp); 1330 } 1331 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1332 vfs_mount_destroy(mp); 1333 return (0); 1334 } 1335 1336 /* 1337 * --------------------------------------------------------------------- 1338 * Mounting of root filesystem 1339 * 1340 */ 1341 1342 struct root_hold_token { 1343 const char *who; 1344 LIST_ENTRY(root_hold_token) list; 1345 }; 1346 1347 static LIST_HEAD(, root_hold_token) root_holds = 1348 LIST_HEAD_INITIALIZER(&root_holds); 1349 1350 static int root_mount_complete; 1351 1352 /* 1353 * Hold root mount. 1354 */ 1355 struct root_hold_token * 1356 root_mount_hold(const char *identifier) 1357 { 1358 struct root_hold_token *h; 1359 1360 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1361 h->who = identifier; 1362 mtx_lock(&mountlist_mtx); 1363 LIST_INSERT_HEAD(&root_holds, h, list); 1364 mtx_unlock(&mountlist_mtx); 1365 return (h); 1366 } 1367 1368 /* 1369 * Release root mount. 1370 */ 1371 void 1372 root_mount_rel(struct root_hold_token *h) 1373 { 1374 1375 mtx_lock(&mountlist_mtx); 1376 LIST_REMOVE(h, list); 1377 wakeup(&root_holds); 1378 mtx_unlock(&mountlist_mtx); 1379 free(h, M_DEVBUF); 1380 } 1381 1382 /* 1383 * Wait for all subsystems to release root mount. 1384 */ 1385 static void 1386 root_mount_prepare(void) 1387 { 1388 struct root_hold_token *h; 1389 1390 for (;;) { 1391 DROP_GIANT(); 1392 g_waitidle(); 1393 PICKUP_GIANT(); 1394 mtx_lock(&mountlist_mtx); 1395 if (LIST_EMPTY(&root_holds)) { 1396 mtx_unlock(&mountlist_mtx); 1397 break; 1398 } 1399 printf("Root mount waiting for:"); 1400 LIST_FOREACH(h, &root_holds, list) 1401 printf(" %s", h->who); 1402 printf("\n"); 1403 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1404 hz); 1405 } 1406 } 1407 1408 /* 1409 * Root was mounted, share the good news. 1410 */ 1411 static void 1412 root_mount_done(void) 1413 { 1414 1415 /* 1416 * Use a mutex to prevent the wakeup being missed and waiting for 1417 * an extra 1 second sleep. 1418 */ 1419 mtx_lock(&mountlist_mtx); 1420 root_mount_complete = 1; 1421 wakeup(&root_mount_complete); 1422 mtx_unlock(&mountlist_mtx); 1423 } 1424 1425 /* 1426 * Return true if root is already mounted. 1427 */ 1428 int 1429 root_mounted(void) 1430 { 1431 1432 /* No mutex is acquired here because int stores are atomic. */ 1433 return (root_mount_complete); 1434 } 1435 1436 /* 1437 * Wait until root is mounted. 1438 */ 1439 void 1440 root_mount_wait(void) 1441 { 1442 1443 /* 1444 * Panic on an obvious deadlock - the function can't be called from 1445 * a thread which is doing the whole SYSINIT stuff. 1446 */ 1447 KASSERT(curthread->td_proc->p_pid != 0, 1448 ("root_mount_wait: cannot be called from the swapper thread")); 1449 mtx_lock(&mountlist_mtx); 1450 while (!root_mount_complete) { 1451 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1452 hz); 1453 } 1454 mtx_unlock(&mountlist_mtx); 1455 } 1456 1457 static void 1458 set_rootvnode(struct thread *td) 1459 { 1460 struct proc *p; 1461 1462 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1463 panic("Cannot find root vnode"); 1464 1465 p = td->td_proc; 1466 FILEDESC_XLOCK(p->p_fd); 1467 1468 if (p->p_fd->fd_cdir != NULL) 1469 vrele(p->p_fd->fd_cdir); 1470 p->p_fd->fd_cdir = rootvnode; 1471 VREF(rootvnode); 1472 1473 if (p->p_fd->fd_rdir != NULL) 1474 vrele(p->p_fd->fd_rdir); 1475 p->p_fd->fd_rdir = rootvnode; 1476 VREF(rootvnode); 1477 1478 FILEDESC_XUNLOCK(p->p_fd); 1479 1480 VOP_UNLOCK(rootvnode, 0); 1481 1482 EVENTHANDLER_INVOKE(mountroot); 1483 } 1484 1485 /* 1486 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1487 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1488 */ 1489 1490 static void 1491 devfs_first(void) 1492 { 1493 struct thread *td = curthread; 1494 struct vfsoptlist *opts; 1495 struct vfsconf *vfsp; 1496 struct mount *mp = NULL; 1497 int error; 1498 1499 vfsp = vfs_byname("devfs"); 1500 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1501 if (vfsp == NULL) 1502 return; 1503 1504 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 1505 1506 error = VFS_MOUNT(mp, td); 1507 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1508 if (error) 1509 return; 1510 1511 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1512 TAILQ_INIT(opts); 1513 mp->mnt_opt = opts; 1514 1515 mtx_lock(&mountlist_mtx); 1516 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1517 mtx_unlock(&mountlist_mtx); 1518 1519 set_rootvnode(td); 1520 1521 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1522 if (error) 1523 printf("kern_symlink /dev -> / returns %d\n", error); 1524 } 1525 1526 /* 1527 * Surgically move our devfs to be mounted on /dev. 1528 */ 1529 1530 static void 1531 devfs_fixup(struct thread *td) 1532 { 1533 struct nameidata nd; 1534 int error; 1535 struct vnode *vp, *dvp; 1536 struct mount *mp; 1537 1538 /* Remove our devfs mount from the mountlist and purge the cache */ 1539 mtx_lock(&mountlist_mtx); 1540 mp = TAILQ_FIRST(&mountlist); 1541 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1542 mtx_unlock(&mountlist_mtx); 1543 cache_purgevfs(mp); 1544 1545 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1546 VI_LOCK(dvp); 1547 dvp->v_iflag &= ~VI_MOUNT; 1548 VI_UNLOCK(dvp); 1549 dvp->v_mountedhere = NULL; 1550 1551 /* Set up the real rootvnode, and purge the cache */ 1552 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1553 set_rootvnode(td); 1554 cache_purgevfs(rootvnode->v_mount); 1555 1556 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1557 error = namei(&nd); 1558 if (error) { 1559 printf("Lookup of /dev for devfs, error: %d\n", error); 1560 return; 1561 } 1562 NDFREE(&nd, NDF_ONLY_PNBUF); 1563 vp = nd.ni_vp; 1564 if (vp->v_type != VDIR) { 1565 vput(vp); 1566 } 1567 error = vinvalbuf(vp, V_SAVE, 0, 0); 1568 if (error) { 1569 vput(vp); 1570 } 1571 cache_purge(vp); 1572 mp->mnt_vnodecovered = vp; 1573 vp->v_mountedhere = mp; 1574 mtx_lock(&mountlist_mtx); 1575 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1576 mtx_unlock(&mountlist_mtx); 1577 VOP_UNLOCK(vp, 0); 1578 vput(dvp); 1579 vfs_unbusy(mp); 1580 1581 /* Unlink the no longer needed /dev/dev -> / symlink */ 1582 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1583 } 1584 1585 /* 1586 * Report errors during filesystem mounting. 1587 */ 1588 void 1589 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1590 { 1591 struct vfsoptlist *moptlist = mp->mnt_optnew; 1592 va_list ap; 1593 int error, len; 1594 char *errmsg; 1595 1596 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1597 if (error || errmsg == NULL || len <= 0) 1598 return; 1599 1600 va_start(ap, fmt); 1601 vsnprintf(errmsg, (size_t)len, fmt, ap); 1602 va_end(ap); 1603 } 1604 1605 void 1606 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1607 { 1608 va_list ap; 1609 int error, len; 1610 char *errmsg; 1611 1612 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1613 if (error || errmsg == NULL || len <= 0) 1614 return; 1615 1616 va_start(ap, fmt); 1617 vsnprintf(errmsg, (size_t)len, fmt, ap); 1618 va_end(ap); 1619 } 1620 1621 /* 1622 * Find and mount the root filesystem 1623 */ 1624 void 1625 vfs_mountroot(void) 1626 { 1627 char *cp; 1628 int error, i, asked = 0; 1629 1630 root_mount_prepare(); 1631 1632 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1633 NULL, NULL, mount_init, mount_fini, 1634 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1635 devfs_first(); 1636 1637 /* 1638 * We are booted with instructions to prompt for the root filesystem. 1639 */ 1640 if (boothowto & RB_ASKNAME) { 1641 if (!vfs_mountroot_ask()) 1642 goto mounted; 1643 asked = 1; 1644 } 1645 1646 /* 1647 * The root filesystem information is compiled in, and we are 1648 * booted with instructions to use it. 1649 */ 1650 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1651 if (!vfs_mountroot_try(ctrootdevname)) 1652 goto mounted; 1653 ctrootdevname = NULL; 1654 } 1655 1656 /* 1657 * We've been given the generic "use CDROM as root" flag. This is 1658 * necessary because one media may be used in many different 1659 * devices, so we need to search for them. 1660 */ 1661 if (boothowto & RB_CDROM) { 1662 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1663 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1664 goto mounted; 1665 } 1666 } 1667 1668 /* 1669 * Try to use the value read by the loader from /etc/fstab, or 1670 * supplied via some other means. This is the preferred 1671 * mechanism. 1672 */ 1673 cp = getenv("vfs.root.mountfrom"); 1674 if (cp != NULL) { 1675 error = vfs_mountroot_try(cp); 1676 freeenv(cp); 1677 if (!error) 1678 goto mounted; 1679 } 1680 1681 /* 1682 * Try values that may have been computed by code during boot 1683 */ 1684 if (!vfs_mountroot_try(rootdevnames[0])) 1685 goto mounted; 1686 if (!vfs_mountroot_try(rootdevnames[1])) 1687 goto mounted; 1688 1689 /* 1690 * If we (still) have a compiled-in default, try it. 1691 */ 1692 if (ctrootdevname != NULL) 1693 if (!vfs_mountroot_try(ctrootdevname)) 1694 goto mounted; 1695 /* 1696 * Everything so far has failed, prompt on the console if we haven't 1697 * already tried that. 1698 */ 1699 if (!asked) 1700 if (!vfs_mountroot_ask()) 1701 goto mounted; 1702 1703 panic("Root mount failed, startup aborted."); 1704 1705 mounted: 1706 root_mount_done(); 1707 } 1708 1709 /* 1710 * Mount (mountfrom) as the root filesystem. 1711 */ 1712 static int 1713 vfs_mountroot_try(const char *mountfrom) 1714 { 1715 struct mount *mp; 1716 char *vfsname, *path; 1717 time_t timebase; 1718 int error; 1719 char patt[32]; 1720 1721 vfsname = NULL; 1722 path = NULL; 1723 mp = NULL; 1724 error = EINVAL; 1725 1726 if (mountfrom == NULL) 1727 return (error); /* don't complain */ 1728 printf("Trying to mount root from %s\n", mountfrom); 1729 1730 /* parse vfs name and path */ 1731 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1732 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1733 vfsname[0] = path[0] = 0; 1734 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1735 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1736 goto out; 1737 1738 if (path[0] == '\0') 1739 strcpy(path, ROOTNAME); 1740 1741 error = kernel_vmount( 1742 MNT_RDONLY | MNT_ROOTFS, 1743 "fstype", vfsname, 1744 "fspath", "/", 1745 "from", path, 1746 NULL); 1747 if (error == 0) { 1748 /* 1749 * We mount devfs prior to mounting the / FS, so the first 1750 * entry will typically be devfs. 1751 */ 1752 mp = TAILQ_FIRST(&mountlist); 1753 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1754 1755 /* 1756 * Iterate over all currently mounted file systems and use 1757 * the time stamp found to check and/or initialize the RTC. 1758 * Typically devfs has no time stamp and the only other FS 1759 * is the actual / FS. 1760 * Call inittodr() only once and pass it the largest of the 1761 * timestamps we encounter. 1762 */ 1763 timebase = 0; 1764 do { 1765 if (mp->mnt_time > timebase) 1766 timebase = mp->mnt_time; 1767 mp = TAILQ_NEXT(mp, mnt_list); 1768 } while (mp != NULL); 1769 inittodr(timebase); 1770 1771 devfs_fixup(curthread); 1772 } 1773 out: 1774 free(path, M_MOUNT); 1775 free(vfsname, M_MOUNT); 1776 return (error); 1777 } 1778 1779 /* 1780 * --------------------------------------------------------------------- 1781 * Interactive root filesystem selection code. 1782 */ 1783 1784 static int 1785 vfs_mountroot_ask(void) 1786 { 1787 char name[128]; 1788 1789 for(;;) { 1790 printf("\nManual root filesystem specification:\n"); 1791 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1792 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__) 1793 printf(" eg. ufs:da0s1a\n"); 1794 #else 1795 printf(" eg. ufs:/dev/da0a\n"); 1796 #endif 1797 printf(" ? List valid disk boot devices\n"); 1798 printf(" <empty line> Abort manual input\n"); 1799 printf("\nmountroot> "); 1800 gets(name, sizeof(name), 1); 1801 if (name[0] == '\0') 1802 return (1); 1803 if (name[0] == '?') { 1804 printf("\nList of GEOM managed disk devices:\n "); 1805 g_dev_print(); 1806 continue; 1807 } 1808 if (!vfs_mountroot_try(name)) 1809 return (0); 1810 } 1811 } 1812 1813 /* 1814 * --------------------------------------------------------------------- 1815 * Functions for querying mount options/arguments from filesystems. 1816 */ 1817 1818 /* 1819 * Check that no unknown options are given 1820 */ 1821 int 1822 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1823 { 1824 struct vfsopt *opt; 1825 char errmsg[255]; 1826 const char **t, *p, *q; 1827 int ret = 0; 1828 1829 TAILQ_FOREACH(opt, opts, link) { 1830 p = opt->name; 1831 q = NULL; 1832 if (p[0] == 'n' && p[1] == 'o') 1833 q = p + 2; 1834 for(t = global_opts; *t != NULL; t++) { 1835 if (strcmp(*t, p) == 0) 1836 break; 1837 if (q != NULL) { 1838 if (strcmp(*t, q) == 0) 1839 break; 1840 } 1841 } 1842 if (*t != NULL) 1843 continue; 1844 for(t = legal; *t != NULL; t++) { 1845 if (strcmp(*t, p) == 0) 1846 break; 1847 if (q != NULL) { 1848 if (strcmp(*t, q) == 0) 1849 break; 1850 } 1851 } 1852 if (*t != NULL) 1853 continue; 1854 snprintf(errmsg, sizeof(errmsg), 1855 "mount option <%s> is unknown", p); 1856 printf("%s\n", errmsg); 1857 ret = EINVAL; 1858 } 1859 if (ret != 0) { 1860 TAILQ_FOREACH(opt, opts, link) { 1861 if (strcmp(opt->name, "errmsg") == 0) { 1862 strncpy((char *)opt->value, errmsg, opt->len); 1863 } 1864 } 1865 } 1866 return (ret); 1867 } 1868 1869 /* 1870 * Get a mount option by its name. 1871 * 1872 * Return 0 if the option was found, ENOENT otherwise. 1873 * If len is non-NULL it will be filled with the length 1874 * of the option. If buf is non-NULL, it will be filled 1875 * with the address of the option. 1876 */ 1877 int 1878 vfs_getopt(opts, name, buf, len) 1879 struct vfsoptlist *opts; 1880 const char *name; 1881 void **buf; 1882 int *len; 1883 { 1884 struct vfsopt *opt; 1885 1886 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1887 1888 TAILQ_FOREACH(opt, opts, link) { 1889 if (strcmp(name, opt->name) == 0) { 1890 opt->seen = 1; 1891 if (len != NULL) 1892 *len = opt->len; 1893 if (buf != NULL) 1894 *buf = opt->value; 1895 return (0); 1896 } 1897 } 1898 return (ENOENT); 1899 } 1900 1901 int 1902 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1903 { 1904 struct vfsopt *opt; 1905 1906 if (opts == NULL) 1907 return (-1); 1908 1909 TAILQ_FOREACH(opt, opts, link) { 1910 if (strcmp(name, opt->name) == 0) { 1911 opt->seen = 1; 1912 return (opt->pos); 1913 } 1914 } 1915 return (-1); 1916 } 1917 1918 char * 1919 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1920 { 1921 struct vfsopt *opt; 1922 1923 *error = 0; 1924 TAILQ_FOREACH(opt, opts, link) { 1925 if (strcmp(name, opt->name) != 0) 1926 continue; 1927 opt->seen = 1; 1928 if (opt->len == 0 || 1929 ((char *)opt->value)[opt->len - 1] != '\0') { 1930 *error = EINVAL; 1931 return (NULL); 1932 } 1933 return (opt->value); 1934 } 1935 *error = ENOENT; 1936 return (NULL); 1937 } 1938 1939 int 1940 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1941 { 1942 struct vfsopt *opt; 1943 1944 TAILQ_FOREACH(opt, opts, link) { 1945 if (strcmp(name, opt->name) == 0) { 1946 opt->seen = 1; 1947 if (w != NULL) 1948 *w |= val; 1949 return (1); 1950 } 1951 } 1952 if (w != NULL) 1953 *w &= ~val; 1954 return (0); 1955 } 1956 1957 int 1958 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1959 { 1960 va_list ap; 1961 struct vfsopt *opt; 1962 int ret; 1963 1964 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1965 1966 TAILQ_FOREACH(opt, opts, link) { 1967 if (strcmp(name, opt->name) != 0) 1968 continue; 1969 opt->seen = 1; 1970 if (opt->len == 0 || opt->value == NULL) 1971 return (0); 1972 if (((char *)opt->value)[opt->len - 1] != '\0') 1973 return (0); 1974 va_start(ap, fmt); 1975 ret = vsscanf(opt->value, fmt, ap); 1976 va_end(ap); 1977 return (ret); 1978 } 1979 return (0); 1980 } 1981 1982 int 1983 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 1984 { 1985 struct vfsopt *opt; 1986 1987 TAILQ_FOREACH(opt, opts, link) { 1988 if (strcmp(name, opt->name) != 0) 1989 continue; 1990 opt->seen = 1; 1991 if (opt->value == NULL) 1992 opt->len = len; 1993 else { 1994 if (opt->len != len) 1995 return (EINVAL); 1996 bcopy(value, opt->value, len); 1997 } 1998 return (0); 1999 } 2000 return (ENOENT); 2001 } 2002 2003 int 2004 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2005 { 2006 struct vfsopt *opt; 2007 2008 TAILQ_FOREACH(opt, opts, link) { 2009 if (strcmp(name, opt->name) != 0) 2010 continue; 2011 opt->seen = 1; 2012 if (opt->value == NULL) 2013 opt->len = len; 2014 else { 2015 if (opt->len < len) 2016 return (EINVAL); 2017 opt->len = len; 2018 bcopy(value, opt->value, len); 2019 } 2020 return (0); 2021 } 2022 return (ENOENT); 2023 } 2024 2025 int 2026 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2027 { 2028 struct vfsopt *opt; 2029 2030 TAILQ_FOREACH(opt, opts, link) { 2031 if (strcmp(name, opt->name) != 0) 2032 continue; 2033 opt->seen = 1; 2034 if (opt->value == NULL) 2035 opt->len = strlen(value) + 1; 2036 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2037 return (EINVAL); 2038 return (0); 2039 } 2040 return (ENOENT); 2041 } 2042 2043 /* 2044 * Find and copy a mount option. 2045 * 2046 * The size of the buffer has to be specified 2047 * in len, if it is not the same length as the 2048 * mount option, EINVAL is returned. 2049 * Returns ENOENT if the option is not found. 2050 */ 2051 int 2052 vfs_copyopt(opts, name, dest, len) 2053 struct vfsoptlist *opts; 2054 const char *name; 2055 void *dest; 2056 int len; 2057 { 2058 struct vfsopt *opt; 2059 2060 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2061 2062 TAILQ_FOREACH(opt, opts, link) { 2063 if (strcmp(name, opt->name) == 0) { 2064 opt->seen = 1; 2065 if (len != opt->len) 2066 return (EINVAL); 2067 bcopy(opt->value, dest, opt->len); 2068 return (0); 2069 } 2070 } 2071 return (ENOENT); 2072 } 2073 2074 /* 2075 * This is a helper function for filesystems to traverse their 2076 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2077 */ 2078 2079 struct vnode * 2080 __mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2081 { 2082 struct vnode *vp; 2083 2084 mtx_assert(MNT_MTX(mp), MA_OWNED); 2085 2086 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2087 if ((*mvp)->v_yield++ == 500) { 2088 MNT_IUNLOCK(mp); 2089 (*mvp)->v_yield = 0; 2090 uio_yield(); 2091 MNT_ILOCK(mp); 2092 } 2093 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2094 while (vp != NULL && vp->v_type == VMARKER) 2095 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2096 2097 /* Check if we are done */ 2098 if (vp == NULL) { 2099 __mnt_vnode_markerfree(mvp, mp); 2100 return (NULL); 2101 } 2102 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2103 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2104 return (vp); 2105 } 2106 2107 struct vnode * 2108 __mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2109 { 2110 struct vnode *vp; 2111 2112 mtx_assert(MNT_MTX(mp), MA_OWNED); 2113 2114 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2115 while (vp != NULL && vp->v_type == VMARKER) 2116 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2117 2118 /* Check if we are done */ 2119 if (vp == NULL) { 2120 *mvp = NULL; 2121 return (NULL); 2122 } 2123 MNT_REF(mp); 2124 MNT_IUNLOCK(mp); 2125 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2126 M_VNODE_MARKER, 2127 M_WAITOK | M_ZERO); 2128 MNT_ILOCK(mp); 2129 (*mvp)->v_type = VMARKER; 2130 2131 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2132 while (vp != NULL && vp->v_type == VMARKER) 2133 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2134 2135 /* Check if we are done */ 2136 if (vp == NULL) { 2137 MNT_IUNLOCK(mp); 2138 free(*mvp, M_VNODE_MARKER); 2139 MNT_ILOCK(mp); 2140 *mvp = NULL; 2141 MNT_REL(mp); 2142 return (NULL); 2143 } 2144 (*mvp)->v_mount = mp; 2145 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2146 return (vp); 2147 } 2148 2149 2150 void 2151 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2152 { 2153 2154 if (*mvp == NULL) 2155 return; 2156 2157 mtx_assert(MNT_MTX(mp), MA_OWNED); 2158 2159 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2160 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2161 MNT_IUNLOCK(mp); 2162 free(*mvp, M_VNODE_MARKER); 2163 MNT_ILOCK(mp); 2164 *mvp = NULL; 2165 MNT_REL(mp); 2166 } 2167 2168 2169 int 2170 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 2171 { 2172 int error; 2173 2174 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 2175 if (sbp != &mp->mnt_stat) 2176 *sbp = mp->mnt_stat; 2177 return (error); 2178 } 2179 2180 void 2181 vfs_mountedfrom(struct mount *mp, const char *from) 2182 { 2183 2184 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2185 strlcpy(mp->mnt_stat.f_mntfromname, from, 2186 sizeof mp->mnt_stat.f_mntfromname); 2187 } 2188 2189 /* 2190 * --------------------------------------------------------------------- 2191 * This is the api for building mount args and mounting filesystems from 2192 * inside the kernel. 2193 * 2194 * The API works by accumulation of individual args. First error is 2195 * latched. 2196 * 2197 * XXX: should be documented in new manpage kernel_mount(9) 2198 */ 2199 2200 /* A memory allocation which must be freed when we are done */ 2201 struct mntaarg { 2202 SLIST_ENTRY(mntaarg) next; 2203 }; 2204 2205 /* The header for the mount arguments */ 2206 struct mntarg { 2207 struct iovec *v; 2208 int len; 2209 int error; 2210 SLIST_HEAD(, mntaarg) list; 2211 }; 2212 2213 /* 2214 * Add a boolean argument. 2215 * 2216 * flag is the boolean value. 2217 * name must start with "no". 2218 */ 2219 struct mntarg * 2220 mount_argb(struct mntarg *ma, int flag, const char *name) 2221 { 2222 2223 KASSERT(name[0] == 'n' && name[1] == 'o', 2224 ("mount_argb(...,%s): name must start with 'no'", name)); 2225 2226 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2227 } 2228 2229 /* 2230 * Add an argument printf style 2231 */ 2232 struct mntarg * 2233 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2234 { 2235 va_list ap; 2236 struct mntaarg *maa; 2237 struct sbuf *sb; 2238 int len; 2239 2240 if (ma == NULL) { 2241 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2242 SLIST_INIT(&ma->list); 2243 } 2244 if (ma->error) 2245 return (ma); 2246 2247 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2248 M_MOUNT, M_WAITOK); 2249 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2250 ma->v[ma->len].iov_len = strlen(name) + 1; 2251 ma->len++; 2252 2253 sb = sbuf_new_auto(); 2254 va_start(ap, fmt); 2255 sbuf_vprintf(sb, fmt, ap); 2256 va_end(ap); 2257 sbuf_finish(sb); 2258 len = sbuf_len(sb) + 1; 2259 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2260 SLIST_INSERT_HEAD(&ma->list, maa, next); 2261 bcopy(sbuf_data(sb), maa + 1, len); 2262 sbuf_delete(sb); 2263 2264 ma->v[ma->len].iov_base = maa + 1; 2265 ma->v[ma->len].iov_len = len; 2266 ma->len++; 2267 2268 return (ma); 2269 } 2270 2271 /* 2272 * Add an argument which is a userland string. 2273 */ 2274 struct mntarg * 2275 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2276 { 2277 struct mntaarg *maa; 2278 char *tbuf; 2279 2280 if (val == NULL) 2281 return (ma); 2282 if (ma == NULL) { 2283 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2284 SLIST_INIT(&ma->list); 2285 } 2286 if (ma->error) 2287 return (ma); 2288 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2289 SLIST_INSERT_HEAD(&ma->list, maa, next); 2290 tbuf = (void *)(maa + 1); 2291 ma->error = copyinstr(val, tbuf, len, NULL); 2292 return (mount_arg(ma, name, tbuf, -1)); 2293 } 2294 2295 /* 2296 * Plain argument. 2297 * 2298 * If length is -1, treat value as a C string. 2299 */ 2300 struct mntarg * 2301 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2302 { 2303 2304 if (ma == NULL) { 2305 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2306 SLIST_INIT(&ma->list); 2307 } 2308 if (ma->error) 2309 return (ma); 2310 2311 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2312 M_MOUNT, M_WAITOK); 2313 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2314 ma->v[ma->len].iov_len = strlen(name) + 1; 2315 ma->len++; 2316 2317 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2318 if (len < 0) 2319 ma->v[ma->len].iov_len = strlen(val) + 1; 2320 else 2321 ma->v[ma->len].iov_len = len; 2322 ma->len++; 2323 return (ma); 2324 } 2325 2326 /* 2327 * Free a mntarg structure 2328 */ 2329 static void 2330 free_mntarg(struct mntarg *ma) 2331 { 2332 struct mntaarg *maa; 2333 2334 while (!SLIST_EMPTY(&ma->list)) { 2335 maa = SLIST_FIRST(&ma->list); 2336 SLIST_REMOVE_HEAD(&ma->list, next); 2337 free(maa, M_MOUNT); 2338 } 2339 free(ma->v, M_MOUNT); 2340 free(ma, M_MOUNT); 2341 } 2342 2343 /* 2344 * Mount a filesystem 2345 */ 2346 int 2347 kernel_mount(struct mntarg *ma, int flags) 2348 { 2349 struct uio auio; 2350 int error; 2351 2352 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2353 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2354 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2355 2356 auio.uio_iov = ma->v; 2357 auio.uio_iovcnt = ma->len; 2358 auio.uio_segflg = UIO_SYSSPACE; 2359 2360 error = ma->error; 2361 if (!error) 2362 error = vfs_donmount(curthread, flags, &auio); 2363 free_mntarg(ma); 2364 return (error); 2365 } 2366 2367 /* 2368 * A printflike function to mount a filesystem. 2369 */ 2370 int 2371 kernel_vmount(int flags, ...) 2372 { 2373 struct mntarg *ma = NULL; 2374 va_list ap; 2375 const char *cp; 2376 const void *vp; 2377 int error; 2378 2379 va_start(ap, flags); 2380 for (;;) { 2381 cp = va_arg(ap, const char *); 2382 if (cp == NULL) 2383 break; 2384 vp = va_arg(ap, const void *); 2385 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2386 } 2387 va_end(ap); 2388 2389 error = kernel_mount(ma, flags); 2390 return (error); 2391 } 2392