1 /*- 2 * Copyright (c) 1999-2004 Poul-Henning Kamp 3 * Copyright (c) 1999 Michael Smith 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include <sys/param.h> 41 #include <sys/conf.h> 42 #include <sys/fcntl.h> 43 #include <sys/jail.h> 44 #include <sys/kernel.h> 45 #include <sys/libkern.h> 46 #include <sys/malloc.h> 47 #include <sys/mount.h> 48 #include <sys/mutex.h> 49 #include <sys/namei.h> 50 #include <sys/priv.h> 51 #include <sys/proc.h> 52 #include <sys/filedesc.h> 53 #include <sys/reboot.h> 54 #include <sys/syscallsubr.h> 55 #include <sys/sysproto.h> 56 #include <sys/sx.h> 57 #include <sys/sysctl.h> 58 #include <sys/sysent.h> 59 #include <sys/systm.h> 60 #include <sys/vnode.h> 61 #include <vm/uma.h> 62 63 #include <geom/geom.h> 64 65 #include <machine/stdarg.h> 66 67 #include <security/audit/audit.h> 68 #include <security/mac/mac_framework.h> 69 70 #include "opt_rootdevname.h" 71 72 #define ROOTNAME "root_device" 73 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 74 75 static void set_rootvnode(void); 76 static int vfs_domount(struct thread *td, const char *fstype, 77 char *fspath, int fsflags, void *fsdata); 78 static int vfs_mountroot_ask(void); 79 static int vfs_mountroot_try(const char *mountfrom, const char *options); 80 static void free_mntarg(struct mntarg *ma); 81 82 static int usermount = 0; 83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 84 "Unprivileged users may mount and unmount file systems"); 85 86 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 87 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 88 static uma_zone_t mount_zone; 89 90 /* List of mounted filesystems. */ 91 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 92 93 /* For any iteration/modification of mountlist */ 94 struct mtx mountlist_mtx; 95 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 96 97 /* 98 * The vnode of the system's root (/ in the filesystem, without chroot 99 * active.) 100 */ 101 struct vnode *rootvnode; 102 103 /* 104 * The root filesystem is detailed in the kernel environment variable 105 * vfs.root.mountfrom, which is expected to be in the general format 106 * 107 * <vfsname>:[<path>] 108 * vfsname := the name of a VFS known to the kernel and capable 109 * of being mounted as root 110 * path := disk device name or other data used by the filesystem 111 * to locate its physical store 112 * 113 * The environment variable vfs.root.mountfrom options is a comma delimited 114 * set of string mount options. These mount options must be parseable 115 * by nmount() in the kernel. 116 */ 117 118 /* 119 * Global opts, taken by all filesystems 120 */ 121 static const char *global_opts[] = { 122 "errmsg", 123 "fstype", 124 "fspath", 125 "ro", 126 "rw", 127 "nosuid", 128 "noexec", 129 NULL 130 }; 131 132 /* 133 * The root specifiers we will try if RB_CDROM is specified. 134 */ 135 static char *cdrom_rootdevnames[] = { 136 "cd9660:cd0", 137 "cd9660:acd0", 138 NULL 139 }; 140 141 /* legacy find-root code */ 142 char *rootdevnames[2] = {NULL, NULL}; 143 #ifndef ROOTDEVNAME 144 # define ROOTDEVNAME NULL 145 #endif 146 static const char *ctrootdevname = ROOTDEVNAME; 147 148 /* 149 * --------------------------------------------------------------------- 150 * Functions for building and sanitizing the mount options 151 */ 152 153 /* Remove one mount option. */ 154 static void 155 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 156 { 157 158 TAILQ_REMOVE(opts, opt, link); 159 free(opt->name, M_MOUNT); 160 if (opt->value != NULL) 161 free(opt->value, M_MOUNT); 162 free(opt, M_MOUNT); 163 } 164 165 /* Release all resources related to the mount options. */ 166 void 167 vfs_freeopts(struct vfsoptlist *opts) 168 { 169 struct vfsopt *opt; 170 171 while (!TAILQ_EMPTY(opts)) { 172 opt = TAILQ_FIRST(opts); 173 vfs_freeopt(opts, opt); 174 } 175 free(opts, M_MOUNT); 176 } 177 178 void 179 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 180 { 181 struct vfsopt *opt, *temp; 182 183 if (opts == NULL) 184 return; 185 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 186 if (strcmp(opt->name, name) == 0) 187 vfs_freeopt(opts, opt); 188 } 189 } 190 191 /* 192 * Check if options are equal (with or without the "no" prefix). 193 */ 194 static int 195 vfs_equalopts(const char *opt1, const char *opt2) 196 { 197 char *p; 198 199 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 200 if (strcmp(opt1, opt2) == 0) 201 return (1); 202 /* "noopt" vs. "opt" */ 203 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 204 return (1); 205 /* "opt" vs. "noopt" */ 206 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 207 return (1); 208 while ((p = strchr(opt1, '.')) != NULL && 209 !strncmp(opt1, opt2, ++p - opt1)) { 210 opt2 += p - opt1; 211 opt1 = p; 212 /* "foo.noopt" vs. "foo.opt" */ 213 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 214 return (1); 215 /* "foo.opt" vs. "foo.noopt" */ 216 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 217 return (1); 218 } 219 return (0); 220 } 221 222 /* 223 * If a mount option is specified several times, 224 * (with or without the "no" prefix) only keep 225 * the last occurence of it. 226 */ 227 static void 228 vfs_sanitizeopts(struct vfsoptlist *opts) 229 { 230 struct vfsopt *opt, *opt2, *tmp; 231 232 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 233 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 234 while (opt2 != NULL) { 235 if (vfs_equalopts(opt->name, opt2->name)) { 236 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 237 vfs_freeopt(opts, opt2); 238 opt2 = tmp; 239 } else { 240 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 241 } 242 } 243 } 244 } 245 246 /* 247 * Build a linked list of mount options from a struct uio. 248 */ 249 int 250 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 251 { 252 struct vfsoptlist *opts; 253 struct vfsopt *opt; 254 size_t memused, namelen, optlen; 255 unsigned int i, iovcnt; 256 int error; 257 258 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 259 TAILQ_INIT(opts); 260 memused = 0; 261 iovcnt = auio->uio_iovcnt; 262 for (i = 0; i < iovcnt; i += 2) { 263 namelen = auio->uio_iov[i].iov_len; 264 optlen = auio->uio_iov[i + 1].iov_len; 265 memused += sizeof(struct vfsopt) + optlen + namelen; 266 /* 267 * Avoid consuming too much memory, and attempts to overflow 268 * memused. 269 */ 270 if (memused > VFS_MOUNTARG_SIZE_MAX || 271 optlen > VFS_MOUNTARG_SIZE_MAX || 272 namelen > VFS_MOUNTARG_SIZE_MAX) { 273 error = EINVAL; 274 goto bad; 275 } 276 277 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 278 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 279 opt->value = NULL; 280 opt->len = 0; 281 opt->pos = i / 2; 282 opt->seen = 0; 283 284 /* 285 * Do this early, so jumps to "bad" will free the current 286 * option. 287 */ 288 TAILQ_INSERT_TAIL(opts, opt, link); 289 290 if (auio->uio_segflg == UIO_SYSSPACE) { 291 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 292 } else { 293 error = copyin(auio->uio_iov[i].iov_base, opt->name, 294 namelen); 295 if (error) 296 goto bad; 297 } 298 /* Ensure names are null-terminated strings. */ 299 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 300 error = EINVAL; 301 goto bad; 302 } 303 if (optlen != 0) { 304 opt->len = optlen; 305 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 306 if (auio->uio_segflg == UIO_SYSSPACE) { 307 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 308 optlen); 309 } else { 310 error = copyin(auio->uio_iov[i + 1].iov_base, 311 opt->value, optlen); 312 if (error) 313 goto bad; 314 } 315 } 316 } 317 vfs_sanitizeopts(opts); 318 *options = opts; 319 return (0); 320 bad: 321 vfs_freeopts(opts); 322 return (error); 323 } 324 325 /* 326 * Merge the old mount options with the new ones passed 327 * in the MNT_UPDATE case. 328 * 329 * XXX This function will keep a "nofoo" option in the 330 * new options if there is no matching "foo" option 331 * to be cancelled in the old options. This is a bug 332 * if the option's canonical name is "foo". E.g., "noro" 333 * shouldn't end up in the mount point's active options, 334 * but it can. 335 */ 336 static void 337 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 338 { 339 struct vfsopt *opt, *opt2, *new; 340 341 TAILQ_FOREACH(opt, opts, link) { 342 /* 343 * Check that this option hasn't been redefined 344 * nor cancelled with a "no" mount option. 345 */ 346 opt2 = TAILQ_FIRST(toopts); 347 while (opt2 != NULL) { 348 if (strcmp(opt2->name, opt->name) == 0) 349 goto next; 350 if (strncmp(opt2->name, "no", 2) == 0 && 351 strcmp(opt2->name + 2, opt->name) == 0) { 352 vfs_freeopt(toopts, opt2); 353 goto next; 354 } 355 opt2 = TAILQ_NEXT(opt2, link); 356 } 357 /* We want this option, duplicate it. */ 358 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 359 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 360 strcpy(new->name, opt->name); 361 if (opt->len != 0) { 362 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 363 bcopy(opt->value, new->value, opt->len); 364 } else { 365 new->value = NULL; 366 } 367 new->len = opt->len; 368 new->seen = opt->seen; 369 TAILQ_INSERT_TAIL(toopts, new, link); 370 next: 371 continue; 372 } 373 } 374 375 /* 376 * Mount a filesystem. 377 */ 378 int 379 nmount(td, uap) 380 struct thread *td; 381 struct nmount_args /* { 382 struct iovec *iovp; 383 unsigned int iovcnt; 384 int flags; 385 } */ *uap; 386 { 387 struct uio *auio; 388 int error; 389 u_int iovcnt; 390 391 AUDIT_ARG(fflags, uap->flags); 392 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 393 uap->iovp, uap->iovcnt, uap->flags); 394 395 /* 396 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 397 * userspace to set this flag, but we must filter it out if we want 398 * MNT_UPDATE on the root file system to work. 399 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 400 */ 401 uap->flags &= ~MNT_ROOTFS; 402 403 iovcnt = uap->iovcnt; 404 /* 405 * Check that we have an even number of iovec's 406 * and that we have at least two options. 407 */ 408 if ((iovcnt & 1) || (iovcnt < 4)) { 409 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 410 uap->iovcnt); 411 return (EINVAL); 412 } 413 414 error = copyinuio(uap->iovp, iovcnt, &auio); 415 if (error) { 416 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 417 __func__, error); 418 return (error); 419 } 420 error = vfs_donmount(td, uap->flags, auio); 421 422 free(auio, M_IOV); 423 return (error); 424 } 425 426 /* 427 * --------------------------------------------------------------------- 428 * Various utility functions 429 */ 430 431 void 432 vfs_ref(struct mount *mp) 433 { 434 435 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 436 MNT_ILOCK(mp); 437 MNT_REF(mp); 438 MNT_IUNLOCK(mp); 439 } 440 441 void 442 vfs_rel(struct mount *mp) 443 { 444 445 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 446 MNT_ILOCK(mp); 447 MNT_REL(mp); 448 MNT_IUNLOCK(mp); 449 } 450 451 static int 452 mount_init(void *mem, int size, int flags) 453 { 454 struct mount *mp; 455 456 mp = (struct mount *)mem; 457 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 458 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 459 return (0); 460 } 461 462 static void 463 mount_fini(void *mem, int size) 464 { 465 struct mount *mp; 466 467 mp = (struct mount *)mem; 468 lockdestroy(&mp->mnt_explock); 469 mtx_destroy(&mp->mnt_mtx); 470 } 471 472 /* 473 * Allocate and initialize the mount point struct. 474 */ 475 struct mount * 476 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 477 struct ucred *cred) 478 { 479 struct mount *mp; 480 481 mp = uma_zalloc(mount_zone, M_WAITOK); 482 bzero(&mp->mnt_startzero, 483 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 484 TAILQ_INIT(&mp->mnt_nvnodelist); 485 mp->mnt_nvnodelistsize = 0; 486 mp->mnt_ref = 0; 487 (void) vfs_busy(mp, MBF_NOWAIT); 488 mp->mnt_op = vfsp->vfc_vfsops; 489 mp->mnt_vfc = vfsp; 490 vfsp->vfc_refcount++; /* XXX Unlocked */ 491 mp->mnt_stat.f_type = vfsp->vfc_typenum; 492 mp->mnt_gen++; 493 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 494 mp->mnt_vnodecovered = vp; 495 mp->mnt_cred = crdup(cred); 496 mp->mnt_stat.f_owner = cred->cr_uid; 497 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 498 mp->mnt_iosize_max = DFLTPHYS; 499 #ifdef MAC 500 mac_mount_init(mp); 501 mac_mount_create(cred, mp); 502 #endif 503 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 504 return (mp); 505 } 506 507 /* 508 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 509 */ 510 void 511 vfs_mount_destroy(struct mount *mp) 512 { 513 514 MNT_ILOCK(mp); 515 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 516 if (mp->mnt_kern_flag & MNTK_MWAIT) { 517 mp->mnt_kern_flag &= ~MNTK_MWAIT; 518 wakeup(mp); 519 } 520 while (mp->mnt_ref) 521 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 522 KASSERT(mp->mnt_ref == 0, 523 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 524 __FILE__, __LINE__)); 525 if (mp->mnt_writeopcount != 0) 526 panic("vfs_mount_destroy: nonzero writeopcount"); 527 if (mp->mnt_secondary_writes != 0) 528 panic("vfs_mount_destroy: nonzero secondary_writes"); 529 mp->mnt_vfc->vfc_refcount--; 530 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 531 struct vnode *vp; 532 533 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 534 vprint("", vp); 535 panic("unmount: dangling vnode"); 536 } 537 if (mp->mnt_nvnodelistsize != 0) 538 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 539 if (mp->mnt_lockref != 0) 540 panic("vfs_mount_destroy: nonzero lock refcount"); 541 MNT_IUNLOCK(mp); 542 #ifdef MAC 543 mac_mount_destroy(mp); 544 #endif 545 if (mp->mnt_opt != NULL) 546 vfs_freeopts(mp->mnt_opt); 547 crfree(mp->mnt_cred); 548 uma_zfree(mount_zone, mp); 549 } 550 551 int 552 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 553 { 554 struct vfsoptlist *optlist; 555 struct vfsopt *opt, *noro_opt, *tmp_opt; 556 char *fstype, *fspath, *errmsg; 557 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 558 int has_rw, has_noro; 559 560 errmsg = fspath = NULL; 561 errmsg_len = has_noro = has_rw = fspathlen = 0; 562 errmsg_pos = -1; 563 564 error = vfs_buildopts(fsoptions, &optlist); 565 if (error) 566 return (error); 567 568 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 569 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 570 571 /* 572 * We need these two options before the others, 573 * and they are mandatory for any filesystem. 574 * Ensure they are NUL terminated as well. 575 */ 576 fstypelen = 0; 577 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 578 if (error || fstype[fstypelen - 1] != '\0') { 579 error = EINVAL; 580 if (errmsg != NULL) 581 strncpy(errmsg, "Invalid fstype", errmsg_len); 582 goto bail; 583 } 584 fspathlen = 0; 585 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 586 if (error || fspath[fspathlen - 1] != '\0') { 587 error = EINVAL; 588 if (errmsg != NULL) 589 strncpy(errmsg, "Invalid fspath", errmsg_len); 590 goto bail; 591 } 592 593 /* 594 * We need to see if we have the "update" option 595 * before we call vfs_domount(), since vfs_domount() has special 596 * logic based on MNT_UPDATE. This is very important 597 * when we want to update the root filesystem. 598 */ 599 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 600 if (strcmp(opt->name, "update") == 0) { 601 fsflags |= MNT_UPDATE; 602 vfs_freeopt(optlist, opt); 603 } 604 else if (strcmp(opt->name, "async") == 0) 605 fsflags |= MNT_ASYNC; 606 else if (strcmp(opt->name, "force") == 0) { 607 fsflags |= MNT_FORCE; 608 vfs_freeopt(optlist, opt); 609 } 610 else if (strcmp(opt->name, "reload") == 0) { 611 fsflags |= MNT_RELOAD; 612 vfs_freeopt(optlist, opt); 613 } 614 else if (strcmp(opt->name, "multilabel") == 0) 615 fsflags |= MNT_MULTILABEL; 616 else if (strcmp(opt->name, "noasync") == 0) 617 fsflags &= ~MNT_ASYNC; 618 else if (strcmp(opt->name, "noatime") == 0) 619 fsflags |= MNT_NOATIME; 620 else if (strcmp(opt->name, "atime") == 0) { 621 free(opt->name, M_MOUNT); 622 opt->name = strdup("nonoatime", M_MOUNT); 623 } 624 else if (strcmp(opt->name, "noclusterr") == 0) 625 fsflags |= MNT_NOCLUSTERR; 626 else if (strcmp(opt->name, "clusterr") == 0) { 627 free(opt->name, M_MOUNT); 628 opt->name = strdup("nonoclusterr", M_MOUNT); 629 } 630 else if (strcmp(opt->name, "noclusterw") == 0) 631 fsflags |= MNT_NOCLUSTERW; 632 else if (strcmp(opt->name, "clusterw") == 0) { 633 free(opt->name, M_MOUNT); 634 opt->name = strdup("nonoclusterw", M_MOUNT); 635 } 636 else if (strcmp(opt->name, "noexec") == 0) 637 fsflags |= MNT_NOEXEC; 638 else if (strcmp(opt->name, "exec") == 0) { 639 free(opt->name, M_MOUNT); 640 opt->name = strdup("nonoexec", M_MOUNT); 641 } 642 else if (strcmp(opt->name, "nosuid") == 0) 643 fsflags |= MNT_NOSUID; 644 else if (strcmp(opt->name, "suid") == 0) { 645 free(opt->name, M_MOUNT); 646 opt->name = strdup("nonosuid", M_MOUNT); 647 } 648 else if (strcmp(opt->name, "nosymfollow") == 0) 649 fsflags |= MNT_NOSYMFOLLOW; 650 else if (strcmp(opt->name, "symfollow") == 0) { 651 free(opt->name, M_MOUNT); 652 opt->name = strdup("nonosymfollow", M_MOUNT); 653 } 654 else if (strcmp(opt->name, "noro") == 0) { 655 fsflags &= ~MNT_RDONLY; 656 has_noro = 1; 657 } 658 else if (strcmp(opt->name, "rw") == 0) { 659 fsflags &= ~MNT_RDONLY; 660 has_rw = 1; 661 } 662 else if (strcmp(opt->name, "ro") == 0) 663 fsflags |= MNT_RDONLY; 664 else if (strcmp(opt->name, "rdonly") == 0) { 665 free(opt->name, M_MOUNT); 666 opt->name = strdup("ro", M_MOUNT); 667 fsflags |= MNT_RDONLY; 668 } 669 else if (strcmp(opt->name, "suiddir") == 0) 670 fsflags |= MNT_SUIDDIR; 671 else if (strcmp(opt->name, "sync") == 0) 672 fsflags |= MNT_SYNCHRONOUS; 673 else if (strcmp(opt->name, "union") == 0) 674 fsflags |= MNT_UNION; 675 } 676 677 /* 678 * If "rw" was specified as a mount option, and we 679 * are trying to update a mount-point from "ro" to "rw", 680 * we need a mount option "noro", since in vfs_mergeopts(), 681 * "noro" will cancel "ro", but "rw" will not do anything. 682 */ 683 if (has_rw && !has_noro) { 684 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 685 noro_opt->name = strdup("noro", M_MOUNT); 686 noro_opt->value = NULL; 687 noro_opt->len = 0; 688 noro_opt->pos = -1; 689 noro_opt->seen = 1; 690 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 691 } 692 693 /* 694 * Be ultra-paranoid about making sure the type and fspath 695 * variables will fit in our mp buffers, including the 696 * terminating NUL. 697 */ 698 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 699 error = ENAMETOOLONG; 700 goto bail; 701 } 702 703 mtx_lock(&Giant); 704 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 705 mtx_unlock(&Giant); 706 bail: 707 /* copyout the errmsg */ 708 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 709 && errmsg_len > 0 && errmsg != NULL) { 710 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 711 bcopy(errmsg, 712 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 713 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 714 } else { 715 copyout(errmsg, 716 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 717 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 718 } 719 } 720 721 if (error != 0) 722 vfs_freeopts(optlist); 723 return (error); 724 } 725 726 /* 727 * Old mount API. 728 */ 729 #ifndef _SYS_SYSPROTO_H_ 730 struct mount_args { 731 char *type; 732 char *path; 733 int flags; 734 caddr_t data; 735 }; 736 #endif 737 /* ARGSUSED */ 738 int 739 mount(td, uap) 740 struct thread *td; 741 struct mount_args /* { 742 char *type; 743 char *path; 744 int flags; 745 caddr_t data; 746 } */ *uap; 747 { 748 char *fstype; 749 struct vfsconf *vfsp = NULL; 750 struct mntarg *ma = NULL; 751 int error; 752 753 AUDIT_ARG(fflags, uap->flags); 754 755 /* 756 * Filter out MNT_ROOTFS. We do not want clients of mount() in 757 * userspace to set this flag, but we must filter it out if we want 758 * MNT_UPDATE on the root file system to work. 759 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 760 */ 761 uap->flags &= ~MNT_ROOTFS; 762 763 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 764 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 765 if (error) { 766 free(fstype, M_TEMP); 767 return (error); 768 } 769 770 AUDIT_ARG(text, fstype); 771 mtx_lock(&Giant); 772 vfsp = vfs_byname_kld(fstype, td, &error); 773 free(fstype, M_TEMP); 774 if (vfsp == NULL) { 775 mtx_unlock(&Giant); 776 return (ENOENT); 777 } 778 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 779 mtx_unlock(&Giant); 780 return (EOPNOTSUPP); 781 } 782 783 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 784 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 785 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 786 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 787 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 788 789 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags); 790 mtx_unlock(&Giant); 791 return (error); 792 } 793 794 795 /* 796 * vfs_domount(): actually attempt a filesystem mount. 797 */ 798 static int 799 vfs_domount( 800 struct thread *td, /* Calling thread. */ 801 const char *fstype, /* Filesystem type. */ 802 char *fspath, /* Mount path. */ 803 int fsflags, /* Flags common to all filesystems. */ 804 void *fsdata /* Options local to the filesystem. */ 805 ) 806 { 807 struct vnode *vp; 808 struct mount *mp; 809 struct vfsconf *vfsp; 810 struct oexport_args oexport; 811 struct export_args export; 812 int error, flag = 0; 813 struct vattr va; 814 struct nameidata nd; 815 816 mtx_assert(&Giant, MA_OWNED); 817 /* 818 * Be ultra-paranoid about making sure the type and fspath 819 * variables will fit in our mp buffers, including the 820 * terminating NUL. 821 */ 822 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 823 return (ENAMETOOLONG); 824 825 if (jailed(td->td_ucred) || usermount == 0) { 826 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 827 return (error); 828 } 829 830 /* 831 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 832 */ 833 if (fsflags & MNT_EXPORTED) { 834 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 835 if (error) 836 return (error); 837 } 838 if (fsflags & MNT_SUIDDIR) { 839 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 840 if (error) 841 return (error); 842 } 843 /* 844 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 845 */ 846 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 847 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 848 fsflags |= MNT_NOSUID | MNT_USER; 849 } 850 851 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 852 vfsp = NULL; 853 if ((fsflags & MNT_UPDATE) == 0) { 854 /* Don't try to load KLDs if we're mounting the root. */ 855 if (fsflags & MNT_ROOTFS) 856 vfsp = vfs_byname(fstype); 857 else 858 vfsp = vfs_byname_kld(fstype, td, &error); 859 if (vfsp == NULL) 860 return (ENODEV); 861 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 862 return (EPERM); 863 } 864 /* 865 * Get vnode to be covered 866 */ 867 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 868 fspath, td); 869 if ((error = namei(&nd)) != 0) 870 return (error); 871 NDFREE(&nd, NDF_ONLY_PNBUF); 872 vp = nd.ni_vp; 873 if (fsflags & MNT_UPDATE) { 874 if ((vp->v_vflag & VV_ROOT) == 0) { 875 vput(vp); 876 return (EINVAL); 877 } 878 mp = vp->v_mount; 879 MNT_ILOCK(mp); 880 flag = mp->mnt_flag; 881 /* 882 * We only allow the filesystem to be reloaded if it 883 * is currently mounted read-only. 884 */ 885 if ((fsflags & MNT_RELOAD) && 886 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 887 MNT_IUNLOCK(mp); 888 vput(vp); 889 return (EOPNOTSUPP); /* Needs translation */ 890 } 891 MNT_IUNLOCK(mp); 892 /* 893 * Only privileged root, or (if MNT_USER is set) the user that 894 * did the original mount is permitted to update it. 895 */ 896 error = vfs_suser(mp, td); 897 if (error) { 898 vput(vp); 899 return (error); 900 } 901 if (vfs_busy(mp, MBF_NOWAIT)) { 902 vput(vp); 903 return (EBUSY); 904 } 905 VI_LOCK(vp); 906 if ((vp->v_iflag & VI_MOUNT) != 0 || 907 vp->v_mountedhere != NULL) { 908 VI_UNLOCK(vp); 909 vfs_unbusy(mp); 910 vput(vp); 911 return (EBUSY); 912 } 913 vp->v_iflag |= VI_MOUNT; 914 VI_UNLOCK(vp); 915 MNT_ILOCK(mp); 916 mp->mnt_flag |= fsflags & 917 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 918 MNT_IUNLOCK(mp); 919 VOP_UNLOCK(vp, 0); 920 mp->mnt_optnew = fsdata; 921 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 922 } else { 923 /* 924 * If the user is not root, ensure that they own the directory 925 * onto which we are attempting to mount. 926 */ 927 error = VOP_GETATTR(vp, &va, td->td_ucred); 928 if (error) { 929 vput(vp); 930 return (error); 931 } 932 if (va.va_uid != td->td_ucred->cr_uid) { 933 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 934 0); 935 if (error) { 936 vput(vp); 937 return (error); 938 } 939 } 940 error = vinvalbuf(vp, V_SAVE, 0, 0); 941 if (error != 0) { 942 vput(vp); 943 return (error); 944 } 945 if (vp->v_type != VDIR) { 946 vput(vp); 947 return (ENOTDIR); 948 } 949 VI_LOCK(vp); 950 if ((vp->v_iflag & VI_MOUNT) != 0 || 951 vp->v_mountedhere != NULL) { 952 VI_UNLOCK(vp); 953 vput(vp); 954 return (EBUSY); 955 } 956 vp->v_iflag |= VI_MOUNT; 957 VI_UNLOCK(vp); 958 959 /* 960 * Allocate and initialize the filesystem. 961 */ 962 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 963 VOP_UNLOCK(vp, 0); 964 965 /* XXXMAC: pass to vfs_mount_alloc? */ 966 mp->mnt_optnew = fsdata; 967 } 968 969 /* 970 * Set the mount level flags. 971 */ 972 MNT_ILOCK(mp); 973 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 974 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 975 MNT_RDONLY)); 976 if ((mp->mnt_flag & MNT_ASYNC) == 0) 977 mp->mnt_kern_flag &= ~MNTK_ASYNC; 978 MNT_IUNLOCK(mp); 979 /* 980 * Mount the filesystem. 981 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 982 * get. No freeing of cn_pnbuf. 983 */ 984 error = VFS_MOUNT(mp); 985 986 /* 987 * Process the export option only if we are 988 * updating mount options. 989 */ 990 if (!error && (fsflags & MNT_UPDATE)) { 991 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 992 sizeof(export)) == 0) 993 error = vfs_export(mp, &export); 994 else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 995 sizeof(oexport)) == 0) { 996 export.ex_flags = oexport.ex_flags; 997 export.ex_root = oexport.ex_root; 998 export.ex_anon = oexport.ex_anon; 999 export.ex_addr = oexport.ex_addr; 1000 export.ex_addrlen = oexport.ex_addrlen; 1001 export.ex_mask = oexport.ex_mask; 1002 export.ex_masklen = oexport.ex_masklen; 1003 export.ex_indexfile = oexport.ex_indexfile; 1004 export.ex_numsecflavors = 0; 1005 error = vfs_export(mp, &export); 1006 } 1007 } 1008 1009 if (!error) { 1010 if (mp->mnt_opt != NULL) 1011 vfs_freeopts(mp->mnt_opt); 1012 mp->mnt_opt = mp->mnt_optnew; 1013 (void)VFS_STATFS(mp, &mp->mnt_stat); 1014 } 1015 /* 1016 * Prevent external consumers of mount options from reading 1017 * mnt_optnew. 1018 */ 1019 mp->mnt_optnew = NULL; 1020 if (mp->mnt_flag & MNT_UPDATE) { 1021 MNT_ILOCK(mp); 1022 if (error) 1023 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1024 (flag & ~MNT_QUOTA); 1025 else 1026 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1027 MNT_FORCE | MNT_SNAPSHOT); 1028 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1029 mp->mnt_kern_flag |= MNTK_ASYNC; 1030 else 1031 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1032 MNT_IUNLOCK(mp); 1033 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1034 if (mp->mnt_syncer == NULL) 1035 error = vfs_allocate_syncvnode(mp); 1036 } else { 1037 if (mp->mnt_syncer != NULL) 1038 vrele(mp->mnt_syncer); 1039 mp->mnt_syncer = NULL; 1040 } 1041 vfs_unbusy(mp); 1042 VI_LOCK(vp); 1043 vp->v_iflag &= ~VI_MOUNT; 1044 VI_UNLOCK(vp); 1045 vrele(vp); 1046 return (error); 1047 } 1048 MNT_ILOCK(mp); 1049 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1050 mp->mnt_kern_flag |= MNTK_ASYNC; 1051 else 1052 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1053 MNT_IUNLOCK(mp); 1054 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1055 /* 1056 * Put the new filesystem on the mount list after root. 1057 */ 1058 cache_purge(vp); 1059 if (!error) { 1060 struct vnode *newdp; 1061 1062 VI_LOCK(vp); 1063 vp->v_iflag &= ~VI_MOUNT; 1064 VI_UNLOCK(vp); 1065 vp->v_mountedhere = mp; 1066 mtx_lock(&mountlist_mtx); 1067 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1068 mtx_unlock(&mountlist_mtx); 1069 vfs_event_signal(NULL, VQ_MOUNT, 0); 1070 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 1071 panic("mount: lost mount"); 1072 mountcheckdirs(vp, newdp); 1073 vput(newdp); 1074 VOP_UNLOCK(vp, 0); 1075 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1076 error = vfs_allocate_syncvnode(mp); 1077 vfs_unbusy(mp); 1078 if (error) 1079 vrele(vp); 1080 } else { 1081 VI_LOCK(vp); 1082 vp->v_iflag &= ~VI_MOUNT; 1083 VI_UNLOCK(vp); 1084 vfs_unbusy(mp); 1085 vfs_mount_destroy(mp); 1086 vput(vp); 1087 } 1088 return (error); 1089 } 1090 1091 /* 1092 * Unmount a filesystem. 1093 * 1094 * Note: unmount takes a path to the vnode mounted on as argument, not 1095 * special file (as before). 1096 */ 1097 #ifndef _SYS_SYSPROTO_H_ 1098 struct unmount_args { 1099 char *path; 1100 int flags; 1101 }; 1102 #endif 1103 /* ARGSUSED */ 1104 int 1105 unmount(td, uap) 1106 struct thread *td; 1107 register struct unmount_args /* { 1108 char *path; 1109 int flags; 1110 } */ *uap; 1111 { 1112 struct mount *mp; 1113 char *pathbuf; 1114 int error, id0, id1; 1115 1116 if (jailed(td->td_ucred) || usermount == 0) { 1117 error = priv_check(td, PRIV_VFS_UNMOUNT); 1118 if (error) 1119 return (error); 1120 } 1121 1122 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1123 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1124 if (error) { 1125 free(pathbuf, M_TEMP); 1126 return (error); 1127 } 1128 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1129 mtx_lock(&Giant); 1130 if (uap->flags & MNT_BYFSID) { 1131 /* Decode the filesystem ID. */ 1132 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1133 mtx_unlock(&Giant); 1134 free(pathbuf, M_TEMP); 1135 return (EINVAL); 1136 } 1137 1138 mtx_lock(&mountlist_mtx); 1139 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1140 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1141 mp->mnt_stat.f_fsid.val[1] == id1) 1142 break; 1143 } 1144 mtx_unlock(&mountlist_mtx); 1145 } else { 1146 mtx_lock(&mountlist_mtx); 1147 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1148 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1149 break; 1150 } 1151 mtx_unlock(&mountlist_mtx); 1152 } 1153 free(pathbuf, M_TEMP); 1154 if (mp == NULL) { 1155 /* 1156 * Previously we returned ENOENT for a nonexistent path and 1157 * EINVAL for a non-mountpoint. We cannot tell these apart 1158 * now, so in the !MNT_BYFSID case return the more likely 1159 * EINVAL for compatibility. 1160 */ 1161 mtx_unlock(&Giant); 1162 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1163 } 1164 1165 /* 1166 * Don't allow unmounting the root filesystem. 1167 */ 1168 if (mp->mnt_flag & MNT_ROOTFS) { 1169 mtx_unlock(&Giant); 1170 return (EINVAL); 1171 } 1172 error = dounmount(mp, uap->flags, td); 1173 mtx_unlock(&Giant); 1174 return (error); 1175 } 1176 1177 /* 1178 * Do the actual filesystem unmount. 1179 */ 1180 int 1181 dounmount(mp, flags, td) 1182 struct mount *mp; 1183 int flags; 1184 struct thread *td; 1185 { 1186 struct vnode *coveredvp, *fsrootvp; 1187 int error; 1188 int async_flag; 1189 int mnt_gen_r; 1190 1191 mtx_assert(&Giant, MA_OWNED); 1192 1193 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1194 mnt_gen_r = mp->mnt_gen; 1195 VI_LOCK(coveredvp); 1196 vholdl(coveredvp); 1197 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1198 vdrop(coveredvp); 1199 /* 1200 * Check for mp being unmounted while waiting for the 1201 * covered vnode lock. 1202 */ 1203 if (coveredvp->v_mountedhere != mp || 1204 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1205 VOP_UNLOCK(coveredvp, 0); 1206 return (EBUSY); 1207 } 1208 } 1209 /* 1210 * Only privileged root, or (if MNT_USER is set) the user that did the 1211 * original mount is permitted to unmount this filesystem. 1212 */ 1213 error = vfs_suser(mp, td); 1214 if (error) { 1215 if (coveredvp) 1216 VOP_UNLOCK(coveredvp, 0); 1217 return (error); 1218 } 1219 1220 MNT_ILOCK(mp); 1221 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1222 MNT_IUNLOCK(mp); 1223 if (coveredvp) 1224 VOP_UNLOCK(coveredvp, 0); 1225 return (EBUSY); 1226 } 1227 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1228 /* Allow filesystems to detect that a forced unmount is in progress. */ 1229 if (flags & MNT_FORCE) 1230 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1231 error = 0; 1232 if (mp->mnt_lockref) { 1233 if ((flags & MNT_FORCE) == 0) { 1234 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1235 MNTK_UNMOUNTF); 1236 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1237 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1238 wakeup(mp); 1239 } 1240 MNT_IUNLOCK(mp); 1241 if (coveredvp) 1242 VOP_UNLOCK(coveredvp, 0); 1243 return (EBUSY); 1244 } 1245 mp->mnt_kern_flag |= MNTK_DRAINING; 1246 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1247 "mount drain", 0); 1248 } 1249 MNT_IUNLOCK(mp); 1250 KASSERT(mp->mnt_lockref == 0, 1251 ("%s: invalid lock refcount in the drain path @ %s:%d", 1252 __func__, __FILE__, __LINE__)); 1253 KASSERT(error == 0, 1254 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1255 __func__, __FILE__, __LINE__)); 1256 vn_start_write(NULL, &mp, V_WAIT); 1257 1258 if (mp->mnt_flag & MNT_EXPUBLIC) 1259 vfs_setpublicfs(NULL, NULL, NULL); 1260 1261 vfs_msync(mp, MNT_WAIT); 1262 MNT_ILOCK(mp); 1263 async_flag = mp->mnt_flag & MNT_ASYNC; 1264 mp->mnt_flag &= ~MNT_ASYNC; 1265 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1266 MNT_IUNLOCK(mp); 1267 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1268 if (mp->mnt_syncer != NULL) 1269 vrele(mp->mnt_syncer); 1270 /* 1271 * For forced unmounts, move process cdir/rdir refs on the fs root 1272 * vnode to the covered vnode. For non-forced unmounts we want 1273 * such references to cause an EBUSY error. 1274 */ 1275 if ((flags & MNT_FORCE) && 1276 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1277 if (mp->mnt_vnodecovered != NULL) 1278 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1279 if (fsrootvp == rootvnode) { 1280 vrele(rootvnode); 1281 rootvnode = NULL; 1282 } 1283 vput(fsrootvp); 1284 } 1285 if (((mp->mnt_flag & MNT_RDONLY) || 1286 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1287 error = VFS_UNMOUNT(mp, flags); 1288 vn_finished_write(mp); 1289 /* 1290 * If we failed to flush the dirty blocks for this mount point, 1291 * undo all the cdir/rdir and rootvnode changes we made above. 1292 * Unless we failed to do so because the device is reporting that 1293 * it doesn't exist anymore. 1294 */ 1295 if (error && error != ENXIO) { 1296 if ((flags & MNT_FORCE) && 1297 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1298 if (mp->mnt_vnodecovered != NULL) 1299 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1300 if (rootvnode == NULL) { 1301 rootvnode = fsrootvp; 1302 vref(rootvnode); 1303 } 1304 vput(fsrootvp); 1305 } 1306 MNT_ILOCK(mp); 1307 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1308 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1309 MNT_IUNLOCK(mp); 1310 (void) vfs_allocate_syncvnode(mp); 1311 MNT_ILOCK(mp); 1312 } 1313 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1314 mp->mnt_flag |= async_flag; 1315 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1316 mp->mnt_kern_flag |= MNTK_ASYNC; 1317 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1318 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1319 wakeup(mp); 1320 } 1321 MNT_IUNLOCK(mp); 1322 if (coveredvp) 1323 VOP_UNLOCK(coveredvp, 0); 1324 return (error); 1325 } 1326 mtx_lock(&mountlist_mtx); 1327 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1328 mtx_unlock(&mountlist_mtx); 1329 if (coveredvp != NULL) { 1330 coveredvp->v_mountedhere = NULL; 1331 vput(coveredvp); 1332 } 1333 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1334 vfs_mount_destroy(mp); 1335 return (0); 1336 } 1337 1338 /* 1339 * --------------------------------------------------------------------- 1340 * Mounting of root filesystem 1341 * 1342 */ 1343 1344 struct root_hold_token { 1345 const char *who; 1346 LIST_ENTRY(root_hold_token) list; 1347 }; 1348 1349 static LIST_HEAD(, root_hold_token) root_holds = 1350 LIST_HEAD_INITIALIZER(&root_holds); 1351 1352 static int root_mount_complete; 1353 1354 /* 1355 * Hold root mount. 1356 */ 1357 struct root_hold_token * 1358 root_mount_hold(const char *identifier) 1359 { 1360 struct root_hold_token *h; 1361 1362 if (root_mounted()) 1363 return (NULL); 1364 1365 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1366 h->who = identifier; 1367 mtx_lock(&mountlist_mtx); 1368 LIST_INSERT_HEAD(&root_holds, h, list); 1369 mtx_unlock(&mountlist_mtx); 1370 return (h); 1371 } 1372 1373 /* 1374 * Release root mount. 1375 */ 1376 void 1377 root_mount_rel(struct root_hold_token *h) 1378 { 1379 1380 if (h == NULL) 1381 return; 1382 mtx_lock(&mountlist_mtx); 1383 LIST_REMOVE(h, list); 1384 wakeup(&root_holds); 1385 mtx_unlock(&mountlist_mtx); 1386 free(h, M_DEVBUF); 1387 } 1388 1389 /* 1390 * Wait for all subsystems to release root mount. 1391 */ 1392 static void 1393 root_mount_prepare(void) 1394 { 1395 struct root_hold_token *h; 1396 struct timeval lastfail; 1397 int curfail = 0; 1398 1399 for (;;) { 1400 DROP_GIANT(); 1401 g_waitidle(); 1402 PICKUP_GIANT(); 1403 mtx_lock(&mountlist_mtx); 1404 if (LIST_EMPTY(&root_holds)) { 1405 mtx_unlock(&mountlist_mtx); 1406 break; 1407 } 1408 if (ppsratecheck(&lastfail, &curfail, 1)) { 1409 printf("Root mount waiting for:"); 1410 LIST_FOREACH(h, &root_holds, list) 1411 printf(" %s", h->who); 1412 printf("\n"); 1413 } 1414 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1415 hz); 1416 } 1417 } 1418 1419 /* 1420 * Root was mounted, share the good news. 1421 */ 1422 static void 1423 root_mount_done(void) 1424 { 1425 1426 /* Keep prison0's root in sync with the global rootvnode. */ 1427 mtx_lock(&prison0.pr_mtx); 1428 prison0.pr_root = rootvnode; 1429 vref(prison0.pr_root); 1430 mtx_unlock(&prison0.pr_mtx); 1431 /* 1432 * Use a mutex to prevent the wakeup being missed and waiting for 1433 * an extra 1 second sleep. 1434 */ 1435 mtx_lock(&mountlist_mtx); 1436 root_mount_complete = 1; 1437 wakeup(&root_mount_complete); 1438 mtx_unlock(&mountlist_mtx); 1439 } 1440 1441 /* 1442 * Return true if root is already mounted. 1443 */ 1444 int 1445 root_mounted(void) 1446 { 1447 1448 /* No mutex is acquired here because int stores are atomic. */ 1449 return (root_mount_complete); 1450 } 1451 1452 /* 1453 * Wait until root is mounted. 1454 */ 1455 void 1456 root_mount_wait(void) 1457 { 1458 1459 /* 1460 * Panic on an obvious deadlock - the function can't be called from 1461 * a thread which is doing the whole SYSINIT stuff. 1462 */ 1463 KASSERT(curthread->td_proc->p_pid != 0, 1464 ("root_mount_wait: cannot be called from the swapper thread")); 1465 mtx_lock(&mountlist_mtx); 1466 while (!root_mount_complete) { 1467 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1468 hz); 1469 } 1470 mtx_unlock(&mountlist_mtx); 1471 } 1472 1473 static void 1474 set_rootvnode() 1475 { 1476 struct proc *p; 1477 1478 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode)) 1479 panic("Cannot find root vnode"); 1480 1481 p = curthread->td_proc; 1482 FILEDESC_XLOCK(p->p_fd); 1483 1484 if (p->p_fd->fd_cdir != NULL) 1485 vrele(p->p_fd->fd_cdir); 1486 p->p_fd->fd_cdir = rootvnode; 1487 VREF(rootvnode); 1488 1489 if (p->p_fd->fd_rdir != NULL) 1490 vrele(p->p_fd->fd_rdir); 1491 p->p_fd->fd_rdir = rootvnode; 1492 VREF(rootvnode); 1493 1494 FILEDESC_XUNLOCK(p->p_fd); 1495 1496 VOP_UNLOCK(rootvnode, 0); 1497 1498 EVENTHANDLER_INVOKE(mountroot); 1499 } 1500 1501 /* 1502 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1503 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1504 */ 1505 1506 static void 1507 devfs_first(void) 1508 { 1509 struct thread *td = curthread; 1510 struct vfsoptlist *opts; 1511 struct vfsconf *vfsp; 1512 struct mount *mp = NULL; 1513 int error; 1514 1515 vfsp = vfs_byname("devfs"); 1516 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1517 if (vfsp == NULL) 1518 return; 1519 1520 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 1521 1522 error = VFS_MOUNT(mp); 1523 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1524 if (error) 1525 return; 1526 1527 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1528 TAILQ_INIT(opts); 1529 mp->mnt_opt = opts; 1530 1531 mtx_lock(&mountlist_mtx); 1532 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1533 mtx_unlock(&mountlist_mtx); 1534 1535 set_rootvnode(); 1536 1537 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1538 if (error) 1539 printf("kern_symlink /dev -> / returns %d\n", error); 1540 } 1541 1542 /* 1543 * Surgically move our devfs to be mounted on /dev. 1544 */ 1545 1546 static void 1547 devfs_fixup(struct thread *td) 1548 { 1549 struct nameidata nd; 1550 int error; 1551 struct vnode *vp, *dvp; 1552 struct mount *mp; 1553 1554 /* Remove our devfs mount from the mountlist and purge the cache */ 1555 mtx_lock(&mountlist_mtx); 1556 mp = TAILQ_FIRST(&mountlist); 1557 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1558 mtx_unlock(&mountlist_mtx); 1559 cache_purgevfs(mp); 1560 1561 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp); 1562 VI_LOCK(dvp); 1563 dvp->v_iflag &= ~VI_MOUNT; 1564 VI_UNLOCK(dvp); 1565 dvp->v_mountedhere = NULL; 1566 1567 /* Set up the real rootvnode, and purge the cache */ 1568 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1569 set_rootvnode(); 1570 cache_purgevfs(rootvnode->v_mount); 1571 1572 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1573 error = namei(&nd); 1574 if (error) { 1575 printf("Lookup of /dev for devfs, error: %d\n", error); 1576 return; 1577 } 1578 NDFREE(&nd, NDF_ONLY_PNBUF); 1579 vp = nd.ni_vp; 1580 if (vp->v_type != VDIR) { 1581 vput(vp); 1582 } 1583 error = vinvalbuf(vp, V_SAVE, 0, 0); 1584 if (error) { 1585 vput(vp); 1586 } 1587 cache_purge(vp); 1588 mp->mnt_vnodecovered = vp; 1589 vp->v_mountedhere = mp; 1590 mtx_lock(&mountlist_mtx); 1591 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1592 mtx_unlock(&mountlist_mtx); 1593 VOP_UNLOCK(vp, 0); 1594 vput(dvp); 1595 vfs_unbusy(mp); 1596 1597 /* Unlink the no longer needed /dev/dev -> / symlink */ 1598 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1599 } 1600 1601 /* 1602 * Report errors during filesystem mounting. 1603 */ 1604 void 1605 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1606 { 1607 struct vfsoptlist *moptlist = mp->mnt_optnew; 1608 va_list ap; 1609 int error, len; 1610 char *errmsg; 1611 1612 error = vfs_getopt(moptlist, "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 void 1622 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1623 { 1624 va_list ap; 1625 int error, len; 1626 char *errmsg; 1627 1628 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1629 if (error || errmsg == NULL || len <= 0) 1630 return; 1631 1632 va_start(ap, fmt); 1633 vsnprintf(errmsg, (size_t)len, fmt, ap); 1634 va_end(ap); 1635 } 1636 1637 /* 1638 * Find and mount the root filesystem 1639 */ 1640 void 1641 vfs_mountroot(void) 1642 { 1643 char *cp, *options; 1644 int error, i, asked = 0; 1645 1646 options = NULL; 1647 1648 root_mount_prepare(); 1649 1650 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1651 NULL, NULL, mount_init, mount_fini, 1652 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1653 devfs_first(); 1654 1655 /* 1656 * We are booted with instructions to prompt for the root filesystem. 1657 */ 1658 if (boothowto & RB_ASKNAME) { 1659 if (!vfs_mountroot_ask()) 1660 goto mounted; 1661 asked = 1; 1662 } 1663 1664 options = getenv("vfs.root.mountfrom.options"); 1665 1666 /* 1667 * The root filesystem information is compiled in, and we are 1668 * booted with instructions to use it. 1669 */ 1670 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1671 if (!vfs_mountroot_try(ctrootdevname, options)) 1672 goto mounted; 1673 ctrootdevname = NULL; 1674 } 1675 1676 /* 1677 * We've been given the generic "use CDROM as root" flag. This is 1678 * necessary because one media may be used in many different 1679 * devices, so we need to search for them. 1680 */ 1681 if (boothowto & RB_CDROM) { 1682 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1683 if (!vfs_mountroot_try(cdrom_rootdevnames[i], options)) 1684 goto mounted; 1685 } 1686 } 1687 1688 /* 1689 * Try to use the value read by the loader from /etc/fstab, or 1690 * supplied via some other means. This is the preferred 1691 * mechanism. 1692 */ 1693 cp = getenv("vfs.root.mountfrom"); 1694 if (cp != NULL) { 1695 error = vfs_mountroot_try(cp, options); 1696 freeenv(cp); 1697 if (!error) 1698 goto mounted; 1699 } 1700 1701 /* 1702 * Try values that may have been computed by code during boot 1703 */ 1704 if (!vfs_mountroot_try(rootdevnames[0], options)) 1705 goto mounted; 1706 if (!vfs_mountroot_try(rootdevnames[1], options)) 1707 goto mounted; 1708 1709 /* 1710 * If we (still) have a compiled-in default, try it. 1711 */ 1712 if (ctrootdevname != NULL) 1713 if (!vfs_mountroot_try(ctrootdevname, options)) 1714 goto mounted; 1715 /* 1716 * Everything so far has failed, prompt on the console if we haven't 1717 * already tried that. 1718 */ 1719 if (!asked) 1720 if (!vfs_mountroot_ask()) 1721 goto mounted; 1722 1723 panic("Root mount failed, startup aborted."); 1724 1725 mounted: 1726 root_mount_done(); 1727 freeenv(options); 1728 } 1729 1730 static struct mntarg * 1731 parse_mountroot_options(struct mntarg *ma, const char *options) 1732 { 1733 char *p; 1734 char *name, *name_arg; 1735 char *val, *val_arg; 1736 char *opts; 1737 1738 if (options == NULL || options[0] == '\0') 1739 return (ma); 1740 1741 p = opts = strdup(options, M_MOUNT); 1742 if (opts == NULL) { 1743 return (ma); 1744 } 1745 1746 while((name = strsep(&p, ",")) != NULL) { 1747 if (name[0] == '\0') 1748 break; 1749 1750 val = strchr(name, '='); 1751 if (val != NULL) { 1752 *val = '\0'; 1753 ++val; 1754 } 1755 if( strcmp(name, "rw") == 0 || 1756 strcmp(name, "noro") == 0) { 1757 /* 1758 * The first time we mount the root file system, 1759 * we need to mount 'ro', so We need to ignore 1760 * 'rw' and 'noro' mount options. 1761 */ 1762 continue; 1763 } 1764 name_arg = strdup(name, M_MOUNT); 1765 val_arg = NULL; 1766 if (val != NULL) 1767 val_arg = strdup(val, M_MOUNT); 1768 1769 ma = mount_arg(ma, name_arg, val_arg, 1770 (val_arg != NULL ? -1 : 0)); 1771 } 1772 free(opts, M_MOUNT); 1773 return (ma); 1774 } 1775 1776 /* 1777 * Mount (mountfrom) as the root filesystem. 1778 */ 1779 static int 1780 vfs_mountroot_try(const char *mountfrom, const char *options) 1781 { 1782 struct mount *mp; 1783 struct mntarg *ma; 1784 char *vfsname, *path; 1785 time_t timebase; 1786 int error; 1787 char patt[32]; 1788 char errmsg[255]; 1789 1790 vfsname = NULL; 1791 path = NULL; 1792 mp = NULL; 1793 ma = NULL; 1794 error = EINVAL; 1795 bzero(errmsg, sizeof(errmsg)); 1796 1797 if (mountfrom == NULL) 1798 return (error); /* don't complain */ 1799 printf("Trying to mount root from %s\n", mountfrom); 1800 1801 /* parse vfs name and path */ 1802 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1803 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1804 vfsname[0] = path[0] = 0; 1805 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1806 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1807 goto out; 1808 1809 if (path[0] == '\0') 1810 strcpy(path, ROOTNAME); 1811 1812 ma = mount_arg(ma, "fstype", vfsname, -1); 1813 ma = mount_arg(ma, "fspath", "/", -1); 1814 ma = mount_arg(ma, "from", path, -1); 1815 ma = mount_arg(ma, "errmsg", errmsg, sizeof(errmsg)); 1816 ma = mount_arg(ma, "ro", NULL, 0); 1817 ma = parse_mountroot_options(ma, options); 1818 error = kernel_mount(ma, MNT_ROOTFS); 1819 1820 if (error == 0) { 1821 /* 1822 * We mount devfs prior to mounting the / FS, so the first 1823 * entry will typically be devfs. 1824 */ 1825 mp = TAILQ_FIRST(&mountlist); 1826 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1827 1828 /* 1829 * Iterate over all currently mounted file systems and use 1830 * the time stamp found to check and/or initialize the RTC. 1831 * Typically devfs has no time stamp and the only other FS 1832 * is the actual / FS. 1833 * Call inittodr() only once and pass it the largest of the 1834 * timestamps we encounter. 1835 */ 1836 timebase = 0; 1837 do { 1838 if (mp->mnt_time > timebase) 1839 timebase = mp->mnt_time; 1840 mp = TAILQ_NEXT(mp, mnt_list); 1841 } while (mp != NULL); 1842 inittodr(timebase); 1843 1844 devfs_fixup(curthread); 1845 } 1846 1847 if (error != 0 ) { 1848 printf("ROOT MOUNT ERROR: %s\n", errmsg); 1849 printf("If you have invalid mount options, reboot, and "); 1850 printf("first try the following from\n"); 1851 printf("the loader prompt:\n\n"); 1852 printf(" set vfs.root.mountfrom.options=rw\n\n"); 1853 printf("and then remove invalid mount options from "); 1854 printf("/etc/fstab.\n\n"); 1855 } 1856 out: 1857 free(path, M_MOUNT); 1858 free(vfsname, M_MOUNT); 1859 return (error); 1860 } 1861 1862 /* 1863 * --------------------------------------------------------------------- 1864 * Interactive root filesystem selection code. 1865 */ 1866 1867 static int 1868 vfs_mountroot_ask(void) 1869 { 1870 char name[128]; 1871 char *mountfrom; 1872 char *options; 1873 1874 for(;;) { 1875 printf("Loader variables:\n"); 1876 printf("vfs.root.mountfrom="); 1877 mountfrom = getenv("vfs.root.mountfrom"); 1878 if (mountfrom != NULL) { 1879 printf("%s", mountfrom); 1880 } 1881 printf("\n"); 1882 printf("vfs.root.mountfrom.options="); 1883 options = getenv("vfs.root.mountfrom.options"); 1884 if (options != NULL) { 1885 printf("%s", options); 1886 } 1887 printf("\n"); 1888 freeenv(mountfrom); 1889 freeenv(options); 1890 printf("\nManual root filesystem specification:\n"); 1891 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1892 printf(" eg. ufs:/dev/da0s1a\n"); 1893 printf(" eg. cd9660:/dev/acd0\n"); 1894 printf(" This is equivalent to: "); 1895 printf("mount -t cd9660 /dev/acd0 /\n"); 1896 printf("\n"); 1897 printf(" ? List valid disk boot devices\n"); 1898 printf(" <empty line> Abort manual input\n"); 1899 printf("\nmountroot> "); 1900 gets(name, sizeof(name), 1); 1901 if (name[0] == '\0') 1902 return (1); 1903 if (name[0] == '?') { 1904 printf("\nList of GEOM managed disk devices:\n "); 1905 g_dev_print(); 1906 continue; 1907 } 1908 if (!vfs_mountroot_try(name, NULL)) 1909 return (0); 1910 } 1911 } 1912 1913 /* 1914 * --------------------------------------------------------------------- 1915 * Functions for querying mount options/arguments from filesystems. 1916 */ 1917 1918 /* 1919 * Check that no unknown options are given 1920 */ 1921 int 1922 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1923 { 1924 struct vfsopt *opt; 1925 char errmsg[255]; 1926 const char **t, *p, *q; 1927 int ret = 0; 1928 1929 TAILQ_FOREACH(opt, opts, link) { 1930 p = opt->name; 1931 q = NULL; 1932 if (p[0] == 'n' && p[1] == 'o') 1933 q = p + 2; 1934 for(t = global_opts; *t != NULL; t++) { 1935 if (strcmp(*t, p) == 0) 1936 break; 1937 if (q != NULL) { 1938 if (strcmp(*t, q) == 0) 1939 break; 1940 } 1941 } 1942 if (*t != NULL) 1943 continue; 1944 for(t = legal; *t != NULL; t++) { 1945 if (strcmp(*t, p) == 0) 1946 break; 1947 if (q != NULL) { 1948 if (strcmp(*t, q) == 0) 1949 break; 1950 } 1951 } 1952 if (*t != NULL) 1953 continue; 1954 snprintf(errmsg, sizeof(errmsg), 1955 "mount option <%s> is unknown", p); 1956 printf("%s\n", errmsg); 1957 ret = EINVAL; 1958 } 1959 if (ret != 0) { 1960 TAILQ_FOREACH(opt, opts, link) { 1961 if (strcmp(opt->name, "errmsg") == 0) { 1962 strncpy((char *)opt->value, errmsg, opt->len); 1963 } 1964 } 1965 } 1966 return (ret); 1967 } 1968 1969 /* 1970 * Get a mount option by its name. 1971 * 1972 * Return 0 if the option was found, ENOENT otherwise. 1973 * If len is non-NULL it will be filled with the length 1974 * of the option. If buf is non-NULL, it will be filled 1975 * with the address of the option. 1976 */ 1977 int 1978 vfs_getopt(opts, name, buf, len) 1979 struct vfsoptlist *opts; 1980 const char *name; 1981 void **buf; 1982 int *len; 1983 { 1984 struct vfsopt *opt; 1985 1986 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1987 1988 TAILQ_FOREACH(opt, opts, link) { 1989 if (strcmp(name, opt->name) == 0) { 1990 opt->seen = 1; 1991 if (len != NULL) 1992 *len = opt->len; 1993 if (buf != NULL) 1994 *buf = opt->value; 1995 return (0); 1996 } 1997 } 1998 return (ENOENT); 1999 } 2000 2001 int 2002 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2003 { 2004 struct vfsopt *opt; 2005 2006 if (opts == NULL) 2007 return (-1); 2008 2009 TAILQ_FOREACH(opt, opts, link) { 2010 if (strcmp(name, opt->name) == 0) { 2011 opt->seen = 1; 2012 return (opt->pos); 2013 } 2014 } 2015 return (-1); 2016 } 2017 2018 char * 2019 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2020 { 2021 struct vfsopt *opt; 2022 2023 *error = 0; 2024 TAILQ_FOREACH(opt, opts, link) { 2025 if (strcmp(name, opt->name) != 0) 2026 continue; 2027 opt->seen = 1; 2028 if (opt->len == 0 || 2029 ((char *)opt->value)[opt->len - 1] != '\0') { 2030 *error = EINVAL; 2031 return (NULL); 2032 } 2033 return (opt->value); 2034 } 2035 *error = ENOENT; 2036 return (NULL); 2037 } 2038 2039 int 2040 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 2041 { 2042 struct vfsopt *opt; 2043 2044 TAILQ_FOREACH(opt, opts, link) { 2045 if (strcmp(name, opt->name) == 0) { 2046 opt->seen = 1; 2047 if (w != NULL) 2048 *w |= val; 2049 return (1); 2050 } 2051 } 2052 if (w != NULL) 2053 *w &= ~val; 2054 return (0); 2055 } 2056 2057 int 2058 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2059 { 2060 va_list ap; 2061 struct vfsopt *opt; 2062 int ret; 2063 2064 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2065 2066 TAILQ_FOREACH(opt, opts, link) { 2067 if (strcmp(name, opt->name) != 0) 2068 continue; 2069 opt->seen = 1; 2070 if (opt->len == 0 || opt->value == NULL) 2071 return (0); 2072 if (((char *)opt->value)[opt->len - 1] != '\0') 2073 return (0); 2074 va_start(ap, fmt); 2075 ret = vsscanf(opt->value, fmt, ap); 2076 va_end(ap); 2077 return (ret); 2078 } 2079 return (0); 2080 } 2081 2082 int 2083 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2084 { 2085 struct vfsopt *opt; 2086 2087 TAILQ_FOREACH(opt, opts, link) { 2088 if (strcmp(name, opt->name) != 0) 2089 continue; 2090 opt->seen = 1; 2091 if (opt->value == NULL) 2092 opt->len = len; 2093 else { 2094 if (opt->len != len) 2095 return (EINVAL); 2096 bcopy(value, opt->value, len); 2097 } 2098 return (0); 2099 } 2100 return (ENOENT); 2101 } 2102 2103 int 2104 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2105 { 2106 struct vfsopt *opt; 2107 2108 TAILQ_FOREACH(opt, opts, link) { 2109 if (strcmp(name, opt->name) != 0) 2110 continue; 2111 opt->seen = 1; 2112 if (opt->value == NULL) 2113 opt->len = len; 2114 else { 2115 if (opt->len < len) 2116 return (EINVAL); 2117 opt->len = len; 2118 bcopy(value, opt->value, len); 2119 } 2120 return (0); 2121 } 2122 return (ENOENT); 2123 } 2124 2125 int 2126 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2127 { 2128 struct vfsopt *opt; 2129 2130 TAILQ_FOREACH(opt, opts, link) { 2131 if (strcmp(name, opt->name) != 0) 2132 continue; 2133 opt->seen = 1; 2134 if (opt->value == NULL) 2135 opt->len = strlen(value) + 1; 2136 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2137 return (EINVAL); 2138 return (0); 2139 } 2140 return (ENOENT); 2141 } 2142 2143 /* 2144 * Find and copy a mount option. 2145 * 2146 * The size of the buffer has to be specified 2147 * in len, if it is not the same length as the 2148 * mount option, EINVAL is returned. 2149 * Returns ENOENT if the option is not found. 2150 */ 2151 int 2152 vfs_copyopt(opts, name, dest, len) 2153 struct vfsoptlist *opts; 2154 const char *name; 2155 void *dest; 2156 int len; 2157 { 2158 struct vfsopt *opt; 2159 2160 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2161 2162 TAILQ_FOREACH(opt, opts, link) { 2163 if (strcmp(name, opt->name) == 0) { 2164 opt->seen = 1; 2165 if (len != opt->len) 2166 return (EINVAL); 2167 bcopy(opt->value, dest, opt->len); 2168 return (0); 2169 } 2170 } 2171 return (ENOENT); 2172 } 2173 2174 /* 2175 * This is a helper function for filesystems to traverse their 2176 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2177 */ 2178 2179 struct vnode * 2180 __mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2181 { 2182 struct vnode *vp; 2183 2184 mtx_assert(MNT_MTX(mp), MA_OWNED); 2185 2186 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2187 if ((*mvp)->v_yield++ == 500) { 2188 MNT_IUNLOCK(mp); 2189 (*mvp)->v_yield = 0; 2190 uio_yield(); 2191 MNT_ILOCK(mp); 2192 } 2193 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2194 while (vp != NULL && vp->v_type == VMARKER) 2195 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2196 2197 /* Check if we are done */ 2198 if (vp == NULL) { 2199 __mnt_vnode_markerfree(mvp, mp); 2200 return (NULL); 2201 } 2202 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2203 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2204 return (vp); 2205 } 2206 2207 struct vnode * 2208 __mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2209 { 2210 struct vnode *vp; 2211 2212 mtx_assert(MNT_MTX(mp), MA_OWNED); 2213 2214 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2215 while (vp != NULL && vp->v_type == VMARKER) 2216 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2217 2218 /* Check if we are done */ 2219 if (vp == NULL) { 2220 *mvp = NULL; 2221 return (NULL); 2222 } 2223 MNT_REF(mp); 2224 MNT_IUNLOCK(mp); 2225 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2226 M_VNODE_MARKER, 2227 M_WAITOK | M_ZERO); 2228 MNT_ILOCK(mp); 2229 (*mvp)->v_type = VMARKER; 2230 2231 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2232 while (vp != NULL && vp->v_type == VMARKER) 2233 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2234 2235 /* Check if we are done */ 2236 if (vp == NULL) { 2237 MNT_IUNLOCK(mp); 2238 free(*mvp, M_VNODE_MARKER); 2239 MNT_ILOCK(mp); 2240 *mvp = NULL; 2241 MNT_REL(mp); 2242 return (NULL); 2243 } 2244 (*mvp)->v_mount = mp; 2245 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2246 return (vp); 2247 } 2248 2249 2250 void 2251 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2252 { 2253 2254 if (*mvp == NULL) 2255 return; 2256 2257 mtx_assert(MNT_MTX(mp), MA_OWNED); 2258 2259 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2260 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2261 MNT_IUNLOCK(mp); 2262 free(*mvp, M_VNODE_MARKER); 2263 MNT_ILOCK(mp); 2264 *mvp = NULL; 2265 MNT_REL(mp); 2266 } 2267 2268 2269 int 2270 __vfs_statfs(struct mount *mp, struct statfs *sbp) 2271 { 2272 int error; 2273 2274 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 2275 if (sbp != &mp->mnt_stat) 2276 *sbp = mp->mnt_stat; 2277 return (error); 2278 } 2279 2280 void 2281 vfs_mountedfrom(struct mount *mp, const char *from) 2282 { 2283 2284 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2285 strlcpy(mp->mnt_stat.f_mntfromname, from, 2286 sizeof mp->mnt_stat.f_mntfromname); 2287 } 2288 2289 /* 2290 * --------------------------------------------------------------------- 2291 * This is the api for building mount args and mounting filesystems from 2292 * inside the kernel. 2293 * 2294 * The API works by accumulation of individual args. First error is 2295 * latched. 2296 * 2297 * XXX: should be documented in new manpage kernel_mount(9) 2298 */ 2299 2300 /* A memory allocation which must be freed when we are done */ 2301 struct mntaarg { 2302 SLIST_ENTRY(mntaarg) next; 2303 }; 2304 2305 /* The header for the mount arguments */ 2306 struct mntarg { 2307 struct iovec *v; 2308 int len; 2309 int error; 2310 SLIST_HEAD(, mntaarg) list; 2311 }; 2312 2313 /* 2314 * Add a boolean argument. 2315 * 2316 * flag is the boolean value. 2317 * name must start with "no". 2318 */ 2319 struct mntarg * 2320 mount_argb(struct mntarg *ma, int flag, const char *name) 2321 { 2322 2323 KASSERT(name[0] == 'n' && name[1] == 'o', 2324 ("mount_argb(...,%s): name must start with 'no'", name)); 2325 2326 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2327 } 2328 2329 /* 2330 * Add an argument printf style 2331 */ 2332 struct mntarg * 2333 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2334 { 2335 va_list ap; 2336 struct mntaarg *maa; 2337 struct sbuf *sb; 2338 int len; 2339 2340 if (ma == NULL) { 2341 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2342 SLIST_INIT(&ma->list); 2343 } 2344 if (ma->error) 2345 return (ma); 2346 2347 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2348 M_MOUNT, M_WAITOK); 2349 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2350 ma->v[ma->len].iov_len = strlen(name) + 1; 2351 ma->len++; 2352 2353 sb = sbuf_new_auto(); 2354 va_start(ap, fmt); 2355 sbuf_vprintf(sb, fmt, ap); 2356 va_end(ap); 2357 sbuf_finish(sb); 2358 len = sbuf_len(sb) + 1; 2359 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2360 SLIST_INSERT_HEAD(&ma->list, maa, next); 2361 bcopy(sbuf_data(sb), maa + 1, len); 2362 sbuf_delete(sb); 2363 2364 ma->v[ma->len].iov_base = maa + 1; 2365 ma->v[ma->len].iov_len = len; 2366 ma->len++; 2367 2368 return (ma); 2369 } 2370 2371 /* 2372 * Add an argument which is a userland string. 2373 */ 2374 struct mntarg * 2375 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2376 { 2377 struct mntaarg *maa; 2378 char *tbuf; 2379 2380 if (val == NULL) 2381 return (ma); 2382 if (ma == NULL) { 2383 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2384 SLIST_INIT(&ma->list); 2385 } 2386 if (ma->error) 2387 return (ma); 2388 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2389 SLIST_INSERT_HEAD(&ma->list, maa, next); 2390 tbuf = (void *)(maa + 1); 2391 ma->error = copyinstr(val, tbuf, len, NULL); 2392 return (mount_arg(ma, name, tbuf, -1)); 2393 } 2394 2395 /* 2396 * Plain argument. 2397 * 2398 * If length is -1, treat value as a C string. 2399 */ 2400 struct mntarg * 2401 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2402 { 2403 2404 if (ma == NULL) { 2405 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2406 SLIST_INIT(&ma->list); 2407 } 2408 if (ma->error) 2409 return (ma); 2410 2411 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2412 M_MOUNT, M_WAITOK); 2413 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2414 ma->v[ma->len].iov_len = strlen(name) + 1; 2415 ma->len++; 2416 2417 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2418 if (len < 0) 2419 ma->v[ma->len].iov_len = strlen(val) + 1; 2420 else 2421 ma->v[ma->len].iov_len = len; 2422 ma->len++; 2423 return (ma); 2424 } 2425 2426 /* 2427 * Free a mntarg structure 2428 */ 2429 static void 2430 free_mntarg(struct mntarg *ma) 2431 { 2432 struct mntaarg *maa; 2433 2434 while (!SLIST_EMPTY(&ma->list)) { 2435 maa = SLIST_FIRST(&ma->list); 2436 SLIST_REMOVE_HEAD(&ma->list, next); 2437 free(maa, M_MOUNT); 2438 } 2439 free(ma->v, M_MOUNT); 2440 free(ma, M_MOUNT); 2441 } 2442 2443 /* 2444 * Mount a filesystem 2445 */ 2446 int 2447 kernel_mount(struct mntarg *ma, int flags) 2448 { 2449 struct uio auio; 2450 int error; 2451 2452 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2453 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2454 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2455 2456 auio.uio_iov = ma->v; 2457 auio.uio_iovcnt = ma->len; 2458 auio.uio_segflg = UIO_SYSSPACE; 2459 2460 error = ma->error; 2461 if (!error) 2462 error = vfs_donmount(curthread, flags, &auio); 2463 free_mntarg(ma); 2464 return (error); 2465 } 2466 2467 /* 2468 * A printflike function to mount a filesystem. 2469 */ 2470 int 2471 kernel_vmount(int flags, ...) 2472 { 2473 struct mntarg *ma = NULL; 2474 va_list ap; 2475 const char *cp; 2476 const void *vp; 2477 int error; 2478 2479 va_start(ap, flags); 2480 for (;;) { 2481 cp = va_arg(ap, const char *); 2482 if (cp == NULL) 2483 break; 2484 vp = va_arg(ap, const void *); 2485 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2486 } 2487 va_end(ap); 2488 2489 error = kernel_mount(ma, flags); 2490 return (error); 2491 } 2492