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/sbuf.h> 55 #include <sys/syscallsubr.h> 56 #include <sys/sysproto.h> 57 #include <sys/sx.h> 58 #include <sys/sysctl.h> 59 #include <sys/sysent.h> 60 #include <sys/systm.h> 61 #include <sys/vnode.h> 62 #include <vm/uma.h> 63 64 #include <geom/geom.h> 65 66 #include <machine/stdarg.h> 67 68 #include <security/audit/audit.h> 69 #include <security/mac/mac_framework.h> 70 71 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 72 73 static int vfs_domount(struct thread *td, const char *fstype, char *fspath, 74 uint64_t fsflags, struct vfsoptlist **optlist); 75 static void free_mntarg(struct mntarg *ma); 76 77 static int usermount = 0; 78 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 79 "Unprivileged users may mount and unmount file systems"); 80 81 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 82 static uma_zone_t mount_zone; 83 84 /* List of mounted filesystems. */ 85 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 86 87 /* For any iteration/modification of mountlist */ 88 struct mtx mountlist_mtx; 89 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 90 91 /* 92 * Global opts, taken by all filesystems 93 */ 94 static const char *global_opts[] = { 95 "errmsg", 96 "fstype", 97 "fspath", 98 "ro", 99 "rw", 100 "nosuid", 101 "noexec", 102 NULL 103 }; 104 105 static int 106 mount_init(void *mem, int size, int flags) 107 { 108 struct mount *mp; 109 110 mp = (struct mount *)mem; 111 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 112 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 113 return (0); 114 } 115 116 static void 117 mount_fini(void *mem, int size) 118 { 119 struct mount *mp; 120 121 mp = (struct mount *)mem; 122 lockdestroy(&mp->mnt_explock); 123 mtx_destroy(&mp->mnt_mtx); 124 } 125 126 static void 127 vfs_mount_init(void *dummy __unused) 128 { 129 130 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL, 131 NULL, mount_init, mount_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 132 } 133 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL); 134 135 /* 136 * --------------------------------------------------------------------- 137 * Functions for building and sanitizing the mount options 138 */ 139 140 /* Remove one mount option. */ 141 static void 142 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 143 { 144 145 TAILQ_REMOVE(opts, opt, link); 146 free(opt->name, M_MOUNT); 147 if (opt->value != NULL) 148 free(opt->value, M_MOUNT); 149 free(opt, M_MOUNT); 150 } 151 152 /* Release all resources related to the mount options. */ 153 void 154 vfs_freeopts(struct vfsoptlist *opts) 155 { 156 struct vfsopt *opt; 157 158 while (!TAILQ_EMPTY(opts)) { 159 opt = TAILQ_FIRST(opts); 160 vfs_freeopt(opts, opt); 161 } 162 free(opts, M_MOUNT); 163 } 164 165 void 166 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 167 { 168 struct vfsopt *opt, *temp; 169 170 if (opts == NULL) 171 return; 172 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 173 if (strcmp(opt->name, name) == 0) 174 vfs_freeopt(opts, opt); 175 } 176 } 177 178 static int 179 vfs_isopt_ro(const char *opt) 180 { 181 182 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 || 183 strcmp(opt, "norw") == 0) 184 return (1); 185 return (0); 186 } 187 188 static int 189 vfs_isopt_rw(const char *opt) 190 { 191 192 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0) 193 return (1); 194 return (0); 195 } 196 197 /* 198 * Check if options are equal (with or without the "no" prefix). 199 */ 200 static int 201 vfs_equalopts(const char *opt1, const char *opt2) 202 { 203 char *p; 204 205 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 206 if (strcmp(opt1, opt2) == 0) 207 return (1); 208 /* "noopt" vs. "opt" */ 209 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 210 return (1); 211 /* "opt" vs. "noopt" */ 212 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 213 return (1); 214 while ((p = strchr(opt1, '.')) != NULL && 215 !strncmp(opt1, opt2, ++p - opt1)) { 216 opt2 += p - opt1; 217 opt1 = p; 218 /* "foo.noopt" vs. "foo.opt" */ 219 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 220 return (1); 221 /* "foo.opt" vs. "foo.noopt" */ 222 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 223 return (1); 224 } 225 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */ 226 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) && 227 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2))) 228 return (1); 229 return (0); 230 } 231 232 /* 233 * If a mount option is specified several times, 234 * (with or without the "no" prefix) only keep 235 * the last occurrence of it. 236 */ 237 static void 238 vfs_sanitizeopts(struct vfsoptlist *opts) 239 { 240 struct vfsopt *opt, *opt2, *tmp; 241 242 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 243 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 244 while (opt2 != NULL) { 245 if (vfs_equalopts(opt->name, opt2->name)) { 246 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 247 vfs_freeopt(opts, opt2); 248 opt2 = tmp; 249 } else { 250 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 251 } 252 } 253 } 254 } 255 256 /* 257 * Build a linked list of mount options from a struct uio. 258 */ 259 int 260 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 261 { 262 struct vfsoptlist *opts; 263 struct vfsopt *opt; 264 size_t memused, namelen, optlen; 265 unsigned int i, iovcnt; 266 int error; 267 268 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 269 TAILQ_INIT(opts); 270 memused = 0; 271 iovcnt = auio->uio_iovcnt; 272 for (i = 0; i < iovcnt; i += 2) { 273 namelen = auio->uio_iov[i].iov_len; 274 optlen = auio->uio_iov[i + 1].iov_len; 275 memused += sizeof(struct vfsopt) + optlen + namelen; 276 /* 277 * Avoid consuming too much memory, and attempts to overflow 278 * memused. 279 */ 280 if (memused > VFS_MOUNTARG_SIZE_MAX || 281 optlen > VFS_MOUNTARG_SIZE_MAX || 282 namelen > VFS_MOUNTARG_SIZE_MAX) { 283 error = EINVAL; 284 goto bad; 285 } 286 287 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 288 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 289 opt->value = NULL; 290 opt->len = 0; 291 opt->pos = i / 2; 292 opt->seen = 0; 293 294 /* 295 * Do this early, so jumps to "bad" will free the current 296 * option. 297 */ 298 TAILQ_INSERT_TAIL(opts, opt, link); 299 300 if (auio->uio_segflg == UIO_SYSSPACE) { 301 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 302 } else { 303 error = copyin(auio->uio_iov[i].iov_base, opt->name, 304 namelen); 305 if (error) 306 goto bad; 307 } 308 /* Ensure names are null-terminated strings. */ 309 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 310 error = EINVAL; 311 goto bad; 312 } 313 if (optlen != 0) { 314 opt->len = optlen; 315 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 316 if (auio->uio_segflg == UIO_SYSSPACE) { 317 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 318 optlen); 319 } else { 320 error = copyin(auio->uio_iov[i + 1].iov_base, 321 opt->value, optlen); 322 if (error) 323 goto bad; 324 } 325 } 326 } 327 vfs_sanitizeopts(opts); 328 *options = opts; 329 return (0); 330 bad: 331 vfs_freeopts(opts); 332 return (error); 333 } 334 335 /* 336 * Merge the old mount options with the new ones passed 337 * in the MNT_UPDATE case. 338 * 339 * XXX: This function will keep a "nofoo" option in the new 340 * options. E.g, if the option's canonical name is "foo", 341 * "nofoo" ends up in the mount point's active options. 342 */ 343 static void 344 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts) 345 { 346 struct vfsopt *opt, *new; 347 348 TAILQ_FOREACH(opt, oldopts, link) { 349 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 350 new->name = strdup(opt->name, M_MOUNT); 351 if (opt->len != 0) { 352 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 353 bcopy(opt->value, new->value, opt->len); 354 } else 355 new->value = NULL; 356 new->len = opt->len; 357 new->seen = opt->seen; 358 TAILQ_INSERT_HEAD(toopts, new, link); 359 } 360 vfs_sanitizeopts(toopts); 361 } 362 363 /* 364 * Mount a filesystem. 365 */ 366 int 367 sys_nmount(td, uap) 368 struct thread *td; 369 struct nmount_args /* { 370 struct iovec *iovp; 371 unsigned int iovcnt; 372 int flags; 373 } */ *uap; 374 { 375 struct uio *auio; 376 int error; 377 u_int iovcnt; 378 uint64_t flags; 379 380 /* 381 * Mount flags are now 64-bits. On 32-bit archtectures only 382 * 32-bits are passed in, but from here on everything handles 383 * 64-bit flags correctly. 384 */ 385 flags = uap->flags; 386 387 AUDIT_ARG_FFLAGS(flags); 388 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 389 uap->iovp, uap->iovcnt, 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 by the kernel when mounting its 396 * root file system. 397 */ 398 flags &= ~MNT_ROOTFS; 399 400 iovcnt = uap->iovcnt; 401 /* 402 * Check that we have an even number of iovec's 403 * and that we have at least two options. 404 */ 405 if ((iovcnt & 1) || (iovcnt < 4)) { 406 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 407 uap->iovcnt); 408 return (EINVAL); 409 } 410 411 error = copyinuio(uap->iovp, iovcnt, &auio); 412 if (error) { 413 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 414 __func__, error); 415 return (error); 416 } 417 error = vfs_donmount(td, flags, auio); 418 419 free(auio, M_IOV); 420 return (error); 421 } 422 423 /* 424 * --------------------------------------------------------------------- 425 * Various utility functions 426 */ 427 428 void 429 vfs_ref(struct mount *mp) 430 { 431 432 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 433 MNT_ILOCK(mp); 434 MNT_REF(mp); 435 MNT_IUNLOCK(mp); 436 } 437 438 void 439 vfs_rel(struct mount *mp) 440 { 441 442 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 443 MNT_ILOCK(mp); 444 MNT_REL(mp); 445 MNT_IUNLOCK(mp); 446 } 447 448 /* 449 * Allocate and initialize the mount point struct. 450 */ 451 struct mount * 452 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 453 struct ucred *cred) 454 { 455 struct mount *mp; 456 457 mp = uma_zalloc(mount_zone, M_WAITOK); 458 bzero(&mp->mnt_startzero, 459 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 460 TAILQ_INIT(&mp->mnt_nvnodelist); 461 mp->mnt_nvnodelistsize = 0; 462 TAILQ_INIT(&mp->mnt_activevnodelist); 463 mp->mnt_activevnodelistsize = 0; 464 mp->mnt_ref = 0; 465 (void) vfs_busy(mp, MBF_NOWAIT); 466 mp->mnt_op = vfsp->vfc_vfsops; 467 mp->mnt_vfc = vfsp; 468 vfsp->vfc_refcount++; /* XXX Unlocked */ 469 mp->mnt_stat.f_type = vfsp->vfc_typenum; 470 mp->mnt_gen++; 471 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 472 mp->mnt_vnodecovered = vp; 473 mp->mnt_cred = crdup(cred); 474 mp->mnt_stat.f_owner = cred->cr_uid; 475 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 476 mp->mnt_iosize_max = DFLTPHYS; 477 #ifdef MAC 478 mac_mount_init(mp); 479 mac_mount_create(cred, mp); 480 #endif 481 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 482 TAILQ_INIT(&mp->mnt_uppers); 483 return (mp); 484 } 485 486 /* 487 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 488 */ 489 void 490 vfs_mount_destroy(struct mount *mp) 491 { 492 493 MNT_ILOCK(mp); 494 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 495 if (mp->mnt_kern_flag & MNTK_MWAIT) { 496 mp->mnt_kern_flag &= ~MNTK_MWAIT; 497 wakeup(mp); 498 } 499 while (mp->mnt_ref) 500 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 501 KASSERT(mp->mnt_ref == 0, 502 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 503 __FILE__, __LINE__)); 504 if (mp->mnt_writeopcount != 0) 505 panic("vfs_mount_destroy: nonzero writeopcount"); 506 if (mp->mnt_secondary_writes != 0) 507 panic("vfs_mount_destroy: nonzero secondary_writes"); 508 mp->mnt_vfc->vfc_refcount--; 509 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 510 struct vnode *vp; 511 512 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 513 vprint("", vp); 514 panic("unmount: dangling vnode"); 515 } 516 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers")); 517 if (mp->mnt_nvnodelistsize != 0) 518 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 519 if (mp->mnt_activevnodelistsize != 0) 520 panic("vfs_mount_destroy: nonzero activevnodelistsize"); 521 if (mp->mnt_lockref != 0) 522 panic("vfs_mount_destroy: nonzero lock refcount"); 523 MNT_IUNLOCK(mp); 524 #ifdef MAC 525 mac_mount_destroy(mp); 526 #endif 527 if (mp->mnt_opt != NULL) 528 vfs_freeopts(mp->mnt_opt); 529 crfree(mp->mnt_cred); 530 uma_zfree(mount_zone, mp); 531 } 532 533 int 534 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions) 535 { 536 struct vfsoptlist *optlist; 537 struct vfsopt *opt, *tmp_opt; 538 char *fstype, *fspath, *errmsg; 539 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 540 541 errmsg = fspath = NULL; 542 errmsg_len = fspathlen = 0; 543 errmsg_pos = -1; 544 545 error = vfs_buildopts(fsoptions, &optlist); 546 if (error) 547 return (error); 548 549 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 550 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 551 552 /* 553 * We need these two options before the others, 554 * and they are mandatory for any filesystem. 555 * Ensure they are NUL terminated as well. 556 */ 557 fstypelen = 0; 558 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 559 if (error || fstype[fstypelen - 1] != '\0') { 560 error = EINVAL; 561 if (errmsg != NULL) 562 strncpy(errmsg, "Invalid fstype", errmsg_len); 563 goto bail; 564 } 565 fspathlen = 0; 566 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 567 if (error || fspath[fspathlen - 1] != '\0') { 568 error = EINVAL; 569 if (errmsg != NULL) 570 strncpy(errmsg, "Invalid fspath", errmsg_len); 571 goto bail; 572 } 573 574 /* 575 * We need to see if we have the "update" option 576 * before we call vfs_domount(), since vfs_domount() has special 577 * logic based on MNT_UPDATE. This is very important 578 * when we want to update the root filesystem. 579 */ 580 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 581 if (strcmp(opt->name, "update") == 0) { 582 fsflags |= MNT_UPDATE; 583 vfs_freeopt(optlist, opt); 584 } 585 else if (strcmp(opt->name, "async") == 0) 586 fsflags |= MNT_ASYNC; 587 else if (strcmp(opt->name, "force") == 0) { 588 fsflags |= MNT_FORCE; 589 vfs_freeopt(optlist, opt); 590 } 591 else if (strcmp(opt->name, "reload") == 0) { 592 fsflags |= MNT_RELOAD; 593 vfs_freeopt(optlist, opt); 594 } 595 else if (strcmp(opt->name, "multilabel") == 0) 596 fsflags |= MNT_MULTILABEL; 597 else if (strcmp(opt->name, "noasync") == 0) 598 fsflags &= ~MNT_ASYNC; 599 else if (strcmp(opt->name, "noatime") == 0) 600 fsflags |= MNT_NOATIME; 601 else if (strcmp(opt->name, "atime") == 0) { 602 free(opt->name, M_MOUNT); 603 opt->name = strdup("nonoatime", M_MOUNT); 604 } 605 else if (strcmp(opt->name, "noclusterr") == 0) 606 fsflags |= MNT_NOCLUSTERR; 607 else if (strcmp(opt->name, "clusterr") == 0) { 608 free(opt->name, M_MOUNT); 609 opt->name = strdup("nonoclusterr", M_MOUNT); 610 } 611 else if (strcmp(opt->name, "noclusterw") == 0) 612 fsflags |= MNT_NOCLUSTERW; 613 else if (strcmp(opt->name, "clusterw") == 0) { 614 free(opt->name, M_MOUNT); 615 opt->name = strdup("nonoclusterw", M_MOUNT); 616 } 617 else if (strcmp(opt->name, "noexec") == 0) 618 fsflags |= MNT_NOEXEC; 619 else if (strcmp(opt->name, "exec") == 0) { 620 free(opt->name, M_MOUNT); 621 opt->name = strdup("nonoexec", M_MOUNT); 622 } 623 else if (strcmp(opt->name, "nosuid") == 0) 624 fsflags |= MNT_NOSUID; 625 else if (strcmp(opt->name, "suid") == 0) { 626 free(opt->name, M_MOUNT); 627 opt->name = strdup("nonosuid", M_MOUNT); 628 } 629 else if (strcmp(opt->name, "nosymfollow") == 0) 630 fsflags |= MNT_NOSYMFOLLOW; 631 else if (strcmp(opt->name, "symfollow") == 0) { 632 free(opt->name, M_MOUNT); 633 opt->name = strdup("nonosymfollow", M_MOUNT); 634 } 635 else if (strcmp(opt->name, "noro") == 0) 636 fsflags &= ~MNT_RDONLY; 637 else if (strcmp(opt->name, "rw") == 0) 638 fsflags &= ~MNT_RDONLY; 639 else if (strcmp(opt->name, "ro") == 0) 640 fsflags |= MNT_RDONLY; 641 else if (strcmp(opt->name, "rdonly") == 0) { 642 free(opt->name, M_MOUNT); 643 opt->name = strdup("ro", M_MOUNT); 644 fsflags |= MNT_RDONLY; 645 } 646 else if (strcmp(opt->name, "suiddir") == 0) 647 fsflags |= MNT_SUIDDIR; 648 else if (strcmp(opt->name, "sync") == 0) 649 fsflags |= MNT_SYNCHRONOUS; 650 else if (strcmp(opt->name, "union") == 0) 651 fsflags |= MNT_UNION; 652 } 653 654 /* 655 * Be ultra-paranoid about making sure the type and fspath 656 * variables will fit in our mp buffers, including the 657 * terminating NUL. 658 */ 659 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) { 660 error = ENAMETOOLONG; 661 goto bail; 662 } 663 664 error = vfs_domount(td, fstype, fspath, fsflags, &optlist); 665 bail: 666 /* copyout the errmsg */ 667 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 668 && errmsg_len > 0 && errmsg != NULL) { 669 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 670 bcopy(errmsg, 671 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 672 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 673 } else { 674 copyout(errmsg, 675 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 676 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 677 } 678 } 679 680 if (optlist != NULL) 681 vfs_freeopts(optlist); 682 return (error); 683 } 684 685 /* 686 * Old mount API. 687 */ 688 #ifndef _SYS_SYSPROTO_H_ 689 struct mount_args { 690 char *type; 691 char *path; 692 int flags; 693 caddr_t data; 694 }; 695 #endif 696 /* ARGSUSED */ 697 int 698 sys_mount(td, uap) 699 struct thread *td; 700 struct mount_args /* { 701 char *type; 702 char *path; 703 int flags; 704 caddr_t data; 705 } */ *uap; 706 { 707 char *fstype; 708 struct vfsconf *vfsp = NULL; 709 struct mntarg *ma = NULL; 710 uint64_t flags; 711 int error; 712 713 /* 714 * Mount flags are now 64-bits. On 32-bit architectures only 715 * 32-bits are passed in, but from here on everything handles 716 * 64-bit flags correctly. 717 */ 718 flags = uap->flags; 719 720 AUDIT_ARG_FFLAGS(flags); 721 722 /* 723 * Filter out MNT_ROOTFS. We do not want clients of mount() in 724 * userspace to set this flag, but we must filter it out if we want 725 * MNT_UPDATE on the root file system to work. 726 * MNT_ROOTFS should only be set by the kernel when mounting its 727 * root file system. 728 */ 729 flags &= ~MNT_ROOTFS; 730 731 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 732 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 733 if (error) { 734 free(fstype, M_TEMP); 735 return (error); 736 } 737 738 AUDIT_ARG_TEXT(fstype); 739 mtx_lock(&Giant); 740 vfsp = vfs_byname_kld(fstype, td, &error); 741 free(fstype, M_TEMP); 742 if (vfsp == NULL) { 743 mtx_unlock(&Giant); 744 return (ENOENT); 745 } 746 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 747 mtx_unlock(&Giant); 748 return (EOPNOTSUPP); 749 } 750 751 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN); 752 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 753 ma = mount_argb(ma, flags & MNT_RDONLY, "noro"); 754 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid"); 755 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec"); 756 757 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags); 758 mtx_unlock(&Giant); 759 return (error); 760 } 761 762 /* 763 * vfs_domount_first(): first file system mount (not update) 764 */ 765 static int 766 vfs_domount_first( 767 struct thread *td, /* Calling thread. */ 768 struct vfsconf *vfsp, /* File system type. */ 769 char *fspath, /* Mount path. */ 770 struct vnode *vp, /* Vnode to be covered. */ 771 uint64_t fsflags, /* Flags common to all filesystems. */ 772 struct vfsoptlist **optlist /* Options local to the filesystem. */ 773 ) 774 { 775 struct vattr va; 776 struct mount *mp; 777 struct vnode *newdp; 778 int error; 779 780 mtx_assert(&Giant, MA_OWNED); 781 ASSERT_VOP_ELOCKED(vp, __func__); 782 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 783 784 /* 785 * If the user is not root, ensure that they own the directory 786 * onto which we are attempting to mount. 787 */ 788 error = VOP_GETATTR(vp, &va, td->td_ucred); 789 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 790 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0); 791 if (error == 0) 792 error = vinvalbuf(vp, V_SAVE, 0, 0); 793 if (error == 0 && vp->v_type != VDIR) 794 error = ENOTDIR; 795 if (error == 0) { 796 VI_LOCK(vp); 797 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 798 vp->v_iflag |= VI_MOUNT; 799 else 800 error = EBUSY; 801 VI_UNLOCK(vp); 802 } 803 if (error != 0) { 804 vput(vp); 805 return (error); 806 } 807 VOP_UNLOCK(vp, 0); 808 809 /* Allocate and initialize the filesystem. */ 810 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 811 /* XXXMAC: pass to vfs_mount_alloc? */ 812 mp->mnt_optnew = *optlist; 813 /* Set the mount level flags. */ 814 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 815 816 /* 817 * Mount the filesystem. 818 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 819 * get. No freeing of cn_pnbuf. 820 */ 821 error = VFS_MOUNT(mp); 822 if (error != 0) { 823 vfs_unbusy(mp); 824 vfs_mount_destroy(mp); 825 VI_LOCK(vp); 826 vp->v_iflag &= ~VI_MOUNT; 827 VI_UNLOCK(vp); 828 vrele(vp); 829 return (error); 830 } 831 832 if (mp->mnt_opt != NULL) 833 vfs_freeopts(mp->mnt_opt); 834 mp->mnt_opt = mp->mnt_optnew; 835 *optlist = NULL; 836 (void)VFS_STATFS(mp, &mp->mnt_stat); 837 838 /* 839 * Prevent external consumers of mount options from reading mnt_optnew. 840 */ 841 mp->mnt_optnew = NULL; 842 843 MNT_ILOCK(mp); 844 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 845 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 846 mp->mnt_kern_flag |= MNTK_ASYNC; 847 else 848 mp->mnt_kern_flag &= ~MNTK_ASYNC; 849 MNT_IUNLOCK(mp); 850 851 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 852 cache_purge(vp); 853 VI_LOCK(vp); 854 vp->v_iflag &= ~VI_MOUNT; 855 VI_UNLOCK(vp); 856 vp->v_mountedhere = mp; 857 /* Place the new filesystem at the end of the mount list. */ 858 mtx_lock(&mountlist_mtx); 859 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 860 mtx_unlock(&mountlist_mtx); 861 vfs_event_signal(NULL, VQ_MOUNT, 0); 862 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 863 panic("mount: lost mount"); 864 VOP_UNLOCK(vp, 0); 865 EVENTHANDLER_INVOKE(vfs_mounted, mp, newdp, td); 866 VOP_UNLOCK(newdp, 0); 867 mountcheckdirs(vp, newdp); 868 vrele(newdp); 869 if ((mp->mnt_flag & MNT_RDONLY) == 0) 870 vfs_allocate_syncvnode(mp); 871 vfs_unbusy(mp); 872 return (0); 873 } 874 875 /* 876 * vfs_domount_update(): update of mounted file system 877 */ 878 static int 879 vfs_domount_update( 880 struct thread *td, /* Calling thread. */ 881 struct vnode *vp, /* Mount point vnode. */ 882 uint64_t fsflags, /* Flags common to all filesystems. */ 883 struct vfsoptlist **optlist /* Options local to the filesystem. */ 884 ) 885 { 886 struct oexport_args oexport; 887 struct export_args export; 888 struct mount *mp; 889 int error, export_error; 890 uint64_t flag; 891 892 mtx_assert(&Giant, MA_OWNED); 893 ASSERT_VOP_ELOCKED(vp, __func__); 894 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 895 896 if ((vp->v_vflag & VV_ROOT) == 0) { 897 vput(vp); 898 return (EINVAL); 899 } 900 mp = vp->v_mount; 901 /* 902 * We only allow the filesystem to be reloaded if it 903 * is currently mounted read-only. 904 */ 905 flag = mp->mnt_flag; 906 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 907 vput(vp); 908 return (EOPNOTSUPP); /* Needs translation */ 909 } 910 /* 911 * Only privileged root, or (if MNT_USER is set) the user that 912 * did the original mount is permitted to update it. 913 */ 914 error = vfs_suser(mp, td); 915 if (error != 0) { 916 vput(vp); 917 return (error); 918 } 919 if (vfs_busy(mp, MBF_NOWAIT)) { 920 vput(vp); 921 return (EBUSY); 922 } 923 VI_LOCK(vp); 924 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 925 VI_UNLOCK(vp); 926 vfs_unbusy(mp); 927 vput(vp); 928 return (EBUSY); 929 } 930 vp->v_iflag |= VI_MOUNT; 931 VI_UNLOCK(vp); 932 VOP_UNLOCK(vp, 0); 933 934 MNT_ILOCK(mp); 935 mp->mnt_flag &= ~MNT_UPDATEMASK; 936 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 937 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 938 if ((mp->mnt_flag & MNT_ASYNC) == 0) 939 mp->mnt_kern_flag &= ~MNTK_ASYNC; 940 MNT_IUNLOCK(mp); 941 mp->mnt_optnew = *optlist; 942 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 943 944 /* 945 * Mount the filesystem. 946 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 947 * get. No freeing of cn_pnbuf. 948 */ 949 error = VFS_MOUNT(mp); 950 951 export_error = 0; 952 if (error == 0) { 953 /* Process the export option. */ 954 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 955 sizeof(export)) == 0) { 956 export_error = vfs_export(mp, &export); 957 } else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 958 sizeof(oexport)) == 0) { 959 export.ex_flags = oexport.ex_flags; 960 export.ex_root = oexport.ex_root; 961 export.ex_anon = oexport.ex_anon; 962 export.ex_addr = oexport.ex_addr; 963 export.ex_addrlen = oexport.ex_addrlen; 964 export.ex_mask = oexport.ex_mask; 965 export.ex_masklen = oexport.ex_masklen; 966 export.ex_indexfile = oexport.ex_indexfile; 967 export.ex_numsecflavors = 0; 968 export_error = vfs_export(mp, &export); 969 } 970 } 971 972 MNT_ILOCK(mp); 973 if (error == 0) { 974 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 975 MNT_SNAPSHOT); 976 } else { 977 /* 978 * If we fail, restore old mount flags. MNT_QUOTA is special, 979 * because it is not part of MNT_UPDATEMASK, but it could have 980 * changed in the meantime if quotactl(2) was called. 981 * All in all we want current value of MNT_QUOTA, not the old 982 * one. 983 */ 984 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 985 } 986 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 987 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 988 mp->mnt_kern_flag |= MNTK_ASYNC; 989 else 990 mp->mnt_kern_flag &= ~MNTK_ASYNC; 991 MNT_IUNLOCK(mp); 992 993 if (error != 0) 994 goto end; 995 996 if (mp->mnt_opt != NULL) 997 vfs_freeopts(mp->mnt_opt); 998 mp->mnt_opt = mp->mnt_optnew; 999 *optlist = NULL; 1000 (void)VFS_STATFS(mp, &mp->mnt_stat); 1001 /* 1002 * Prevent external consumers of mount options from reading 1003 * mnt_optnew. 1004 */ 1005 mp->mnt_optnew = NULL; 1006 1007 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1008 vfs_allocate_syncvnode(mp); 1009 else 1010 vfs_deallocate_syncvnode(mp); 1011 end: 1012 vfs_unbusy(mp); 1013 VI_LOCK(vp); 1014 vp->v_iflag &= ~VI_MOUNT; 1015 VI_UNLOCK(vp); 1016 vrele(vp); 1017 return (error != 0 ? error : export_error); 1018 } 1019 1020 /* 1021 * vfs_domount(): actually attempt a filesystem mount. 1022 */ 1023 static int 1024 vfs_domount( 1025 struct thread *td, /* Calling thread. */ 1026 const char *fstype, /* Filesystem type. */ 1027 char *fspath, /* Mount path. */ 1028 uint64_t fsflags, /* Flags common to all filesystems. */ 1029 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1030 ) 1031 { 1032 struct vfsconf *vfsp; 1033 struct nameidata nd; 1034 struct vnode *vp; 1035 char *pathbuf; 1036 int error; 1037 1038 /* 1039 * Be ultra-paranoid about making sure the type and fspath 1040 * variables will fit in our mp buffers, including the 1041 * terminating NUL. 1042 */ 1043 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1044 return (ENAMETOOLONG); 1045 1046 if (jailed(td->td_ucred) || usermount == 0) { 1047 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1048 return (error); 1049 } 1050 1051 /* 1052 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1053 */ 1054 if (fsflags & MNT_EXPORTED) { 1055 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1056 if (error) 1057 return (error); 1058 } 1059 if (fsflags & MNT_SUIDDIR) { 1060 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1061 if (error) 1062 return (error); 1063 } 1064 /* 1065 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1066 */ 1067 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1068 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1069 fsflags |= MNT_NOSUID | MNT_USER; 1070 } 1071 1072 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1073 vfsp = NULL; 1074 if ((fsflags & MNT_UPDATE) == 0) { 1075 /* Don't try to load KLDs if we're mounting the root. */ 1076 if (fsflags & MNT_ROOTFS) 1077 vfsp = vfs_byname(fstype); 1078 else 1079 vfsp = vfs_byname_kld(fstype, td, &error); 1080 if (vfsp == NULL) 1081 return (ENODEV); 1082 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 1083 return (EPERM); 1084 } 1085 1086 /* 1087 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1088 */ 1089 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1090 UIO_SYSSPACE, fspath, td); 1091 error = namei(&nd); 1092 if (error != 0) 1093 return (error); 1094 mtx_lock(&Giant); 1095 NDFREE(&nd, NDF_ONLY_PNBUF); 1096 vp = nd.ni_vp; 1097 if ((fsflags & MNT_UPDATE) == 0) { 1098 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1099 strcpy(pathbuf, fspath); 1100 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN); 1101 /* debug.disablefullpath == 1 results in ENODEV */ 1102 if (error == 0 || error == ENODEV) { 1103 error = vfs_domount_first(td, vfsp, pathbuf, vp, 1104 fsflags, optlist); 1105 } 1106 free(pathbuf, M_TEMP); 1107 } else 1108 error = vfs_domount_update(td, vp, fsflags, optlist); 1109 mtx_unlock(&Giant); 1110 1111 ASSERT_VI_UNLOCKED(vp, __func__); 1112 ASSERT_VOP_UNLOCKED(vp, __func__); 1113 1114 return (error); 1115 } 1116 1117 /* 1118 * Unmount a filesystem. 1119 * 1120 * Note: unmount takes a path to the vnode mounted on as argument, not 1121 * special file (as before). 1122 */ 1123 #ifndef _SYS_SYSPROTO_H_ 1124 struct unmount_args { 1125 char *path; 1126 int flags; 1127 }; 1128 #endif 1129 /* ARGSUSED */ 1130 int 1131 sys_unmount(td, uap) 1132 struct thread *td; 1133 register struct unmount_args /* { 1134 char *path; 1135 int flags; 1136 } */ *uap; 1137 { 1138 struct nameidata nd; 1139 struct mount *mp; 1140 char *pathbuf; 1141 int error, id0, id1; 1142 1143 AUDIT_ARG_VALUE(uap->flags); 1144 if (jailed(td->td_ucred) || usermount == 0) { 1145 error = priv_check(td, PRIV_VFS_UNMOUNT); 1146 if (error) 1147 return (error); 1148 } 1149 1150 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1151 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1152 if (error) { 1153 free(pathbuf, M_TEMP); 1154 return (error); 1155 } 1156 mtx_lock(&Giant); 1157 if (uap->flags & MNT_BYFSID) { 1158 AUDIT_ARG_TEXT(pathbuf); 1159 /* Decode the filesystem ID. */ 1160 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1161 mtx_unlock(&Giant); 1162 free(pathbuf, M_TEMP); 1163 return (EINVAL); 1164 } 1165 1166 mtx_lock(&mountlist_mtx); 1167 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1168 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1169 mp->mnt_stat.f_fsid.val[1] == id1) 1170 break; 1171 } 1172 mtx_unlock(&mountlist_mtx); 1173 } else { 1174 /* 1175 * Try to find global path for path argument. 1176 */ 1177 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1178 UIO_SYSSPACE, pathbuf, td); 1179 if (namei(&nd) == 0) { 1180 NDFREE(&nd, NDF_ONLY_PNBUF); 1181 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf, 1182 MNAMELEN); 1183 if (error == 0 || error == ENODEV) 1184 vput(nd.ni_vp); 1185 } 1186 mtx_lock(&mountlist_mtx); 1187 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1188 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1189 break; 1190 } 1191 mtx_unlock(&mountlist_mtx); 1192 } 1193 free(pathbuf, M_TEMP); 1194 if (mp == NULL) { 1195 /* 1196 * Previously we returned ENOENT for a nonexistent path and 1197 * EINVAL for a non-mountpoint. We cannot tell these apart 1198 * now, so in the !MNT_BYFSID case return the more likely 1199 * EINVAL for compatibility. 1200 */ 1201 mtx_unlock(&Giant); 1202 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1203 } 1204 1205 /* 1206 * Don't allow unmounting the root filesystem. 1207 */ 1208 if (mp->mnt_flag & MNT_ROOTFS) { 1209 mtx_unlock(&Giant); 1210 return (EINVAL); 1211 } 1212 error = dounmount(mp, uap->flags, td); 1213 mtx_unlock(&Giant); 1214 return (error); 1215 } 1216 1217 /* 1218 * Do the actual filesystem unmount. 1219 */ 1220 int 1221 dounmount(mp, flags, td) 1222 struct mount *mp; 1223 int flags; 1224 struct thread *td; 1225 { 1226 struct vnode *coveredvp, *fsrootvp; 1227 int error; 1228 uint64_t async_flag; 1229 int mnt_gen_r; 1230 1231 mtx_assert(&Giant, MA_OWNED); 1232 1233 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1234 mnt_gen_r = mp->mnt_gen; 1235 VI_LOCK(coveredvp); 1236 vholdl(coveredvp); 1237 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1238 vdrop(coveredvp); 1239 /* 1240 * Check for mp being unmounted while waiting for the 1241 * covered vnode lock. 1242 */ 1243 if (coveredvp->v_mountedhere != mp || 1244 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1245 VOP_UNLOCK(coveredvp, 0); 1246 return (EBUSY); 1247 } 1248 } 1249 /* 1250 * Only privileged root, or (if MNT_USER is set) the user that did the 1251 * original mount is permitted to unmount this filesystem. 1252 */ 1253 error = vfs_suser(mp, td); 1254 if (error) { 1255 if (coveredvp) 1256 VOP_UNLOCK(coveredvp, 0); 1257 return (error); 1258 } 1259 1260 vn_start_write(NULL, &mp, V_WAIT); 1261 MNT_ILOCK(mp); 1262 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 || 1263 !TAILQ_EMPTY(&mp->mnt_uppers)) { 1264 MNT_IUNLOCK(mp); 1265 if (coveredvp) 1266 VOP_UNLOCK(coveredvp, 0); 1267 vn_finished_write(mp); 1268 return (EBUSY); 1269 } 1270 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1271 /* Allow filesystems to detect that a forced unmount is in progress. */ 1272 if (flags & MNT_FORCE) { 1273 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1274 MNT_IUNLOCK(mp); 1275 /* 1276 * Must be done after setting MNTK_UNMOUNTF and before 1277 * waiting for mnt_lockref to become 0. 1278 */ 1279 VFS_PURGE(mp); 1280 MNT_ILOCK(mp); 1281 } 1282 error = 0; 1283 if (mp->mnt_lockref) { 1284 mp->mnt_kern_flag |= MNTK_DRAINING; 1285 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1286 "mount drain", 0); 1287 } 1288 MNT_IUNLOCK(mp); 1289 KASSERT(mp->mnt_lockref == 0, 1290 ("%s: invalid lock refcount in the drain path @ %s:%d", 1291 __func__, __FILE__, __LINE__)); 1292 KASSERT(error == 0, 1293 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1294 __func__, __FILE__, __LINE__)); 1295 1296 if (mp->mnt_flag & MNT_EXPUBLIC) 1297 vfs_setpublicfs(NULL, NULL, NULL); 1298 1299 vfs_msync(mp, MNT_WAIT); 1300 MNT_ILOCK(mp); 1301 async_flag = mp->mnt_flag & MNT_ASYNC; 1302 mp->mnt_flag &= ~MNT_ASYNC; 1303 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1304 MNT_IUNLOCK(mp); 1305 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1306 vfs_deallocate_syncvnode(mp); 1307 /* 1308 * For forced unmounts, move process cdir/rdir refs on the fs root 1309 * vnode to the covered vnode. For non-forced unmounts we want 1310 * such references to cause an EBUSY error. 1311 */ 1312 if ((flags & MNT_FORCE) && 1313 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1314 if (mp->mnt_vnodecovered != NULL) 1315 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1316 if (fsrootvp == rootvnode) { 1317 vrele(rootvnode); 1318 rootvnode = NULL; 1319 } 1320 vput(fsrootvp); 1321 } 1322 if (((mp->mnt_flag & MNT_RDONLY) || 1323 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1324 error = VFS_UNMOUNT(mp, flags); 1325 vn_finished_write(mp); 1326 /* 1327 * If we failed to flush the dirty blocks for this mount point, 1328 * undo all the cdir/rdir and rootvnode changes we made above. 1329 * Unless we failed to do so because the device is reporting that 1330 * it doesn't exist anymore. 1331 */ 1332 if (error && error != ENXIO) { 1333 if ((flags & MNT_FORCE) && 1334 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1335 if (mp->mnt_vnodecovered != NULL) 1336 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1337 if (rootvnode == NULL) { 1338 rootvnode = fsrootvp; 1339 vref(rootvnode); 1340 } 1341 vput(fsrootvp); 1342 } 1343 MNT_ILOCK(mp); 1344 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1345 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1346 MNT_IUNLOCK(mp); 1347 vfs_allocate_syncvnode(mp); 1348 MNT_ILOCK(mp); 1349 } 1350 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1351 mp->mnt_flag |= async_flag; 1352 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1353 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1354 mp->mnt_kern_flag |= MNTK_ASYNC; 1355 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1356 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1357 wakeup(mp); 1358 } 1359 MNT_IUNLOCK(mp); 1360 if (coveredvp) 1361 VOP_UNLOCK(coveredvp, 0); 1362 return (error); 1363 } 1364 mtx_lock(&mountlist_mtx); 1365 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1366 mtx_unlock(&mountlist_mtx); 1367 EVENTHANDLER_INVOKE(vfs_unmounted, mp, td); 1368 if (coveredvp != NULL) { 1369 coveredvp->v_mountedhere = NULL; 1370 vput(coveredvp); 1371 } 1372 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1373 vfs_mount_destroy(mp); 1374 return (0); 1375 } 1376 1377 /* 1378 * Report errors during filesystem mounting. 1379 */ 1380 void 1381 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1382 { 1383 struct vfsoptlist *moptlist = mp->mnt_optnew; 1384 va_list ap; 1385 int error, len; 1386 char *errmsg; 1387 1388 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1389 if (error || errmsg == NULL || len <= 0) 1390 return; 1391 1392 va_start(ap, fmt); 1393 vsnprintf(errmsg, (size_t)len, fmt, ap); 1394 va_end(ap); 1395 } 1396 1397 void 1398 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1399 { 1400 va_list ap; 1401 int error, len; 1402 char *errmsg; 1403 1404 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1405 if (error || errmsg == NULL || len <= 0) 1406 return; 1407 1408 va_start(ap, fmt); 1409 vsnprintf(errmsg, (size_t)len, fmt, ap); 1410 va_end(ap); 1411 } 1412 1413 /* 1414 * --------------------------------------------------------------------- 1415 * Functions for querying mount options/arguments from filesystems. 1416 */ 1417 1418 /* 1419 * Check that no unknown options are given 1420 */ 1421 int 1422 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1423 { 1424 struct vfsopt *opt; 1425 char errmsg[255]; 1426 const char **t, *p, *q; 1427 int ret = 0; 1428 1429 TAILQ_FOREACH(opt, opts, link) { 1430 p = opt->name; 1431 q = NULL; 1432 if (p[0] == 'n' && p[1] == 'o') 1433 q = p + 2; 1434 for(t = global_opts; *t != NULL; t++) { 1435 if (strcmp(*t, p) == 0) 1436 break; 1437 if (q != NULL) { 1438 if (strcmp(*t, q) == 0) 1439 break; 1440 } 1441 } 1442 if (*t != NULL) 1443 continue; 1444 for(t = legal; *t != NULL; t++) { 1445 if (strcmp(*t, p) == 0) 1446 break; 1447 if (q != NULL) { 1448 if (strcmp(*t, q) == 0) 1449 break; 1450 } 1451 } 1452 if (*t != NULL) 1453 continue; 1454 snprintf(errmsg, sizeof(errmsg), 1455 "mount option <%s> is unknown", p); 1456 ret = EINVAL; 1457 } 1458 if (ret != 0) { 1459 TAILQ_FOREACH(opt, opts, link) { 1460 if (strcmp(opt->name, "errmsg") == 0) { 1461 strncpy((char *)opt->value, errmsg, opt->len); 1462 break; 1463 } 1464 } 1465 if (opt == NULL) 1466 printf("%s\n", errmsg); 1467 } 1468 return (ret); 1469 } 1470 1471 /* 1472 * Get a mount option by its name. 1473 * 1474 * Return 0 if the option was found, ENOENT otherwise. 1475 * If len is non-NULL it will be filled with the length 1476 * of the option. If buf is non-NULL, it will be filled 1477 * with the address of the option. 1478 */ 1479 int 1480 vfs_getopt(opts, name, buf, len) 1481 struct vfsoptlist *opts; 1482 const char *name; 1483 void **buf; 1484 int *len; 1485 { 1486 struct vfsopt *opt; 1487 1488 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1489 1490 TAILQ_FOREACH(opt, opts, link) { 1491 if (strcmp(name, opt->name) == 0) { 1492 opt->seen = 1; 1493 if (len != NULL) 1494 *len = opt->len; 1495 if (buf != NULL) 1496 *buf = opt->value; 1497 return (0); 1498 } 1499 } 1500 return (ENOENT); 1501 } 1502 1503 int 1504 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1505 { 1506 struct vfsopt *opt; 1507 1508 if (opts == NULL) 1509 return (-1); 1510 1511 TAILQ_FOREACH(opt, opts, link) { 1512 if (strcmp(name, opt->name) == 0) { 1513 opt->seen = 1; 1514 return (opt->pos); 1515 } 1516 } 1517 return (-1); 1518 } 1519 1520 int 1521 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value) 1522 { 1523 char *opt_value, *vtp; 1524 quad_t iv; 1525 int error, opt_len; 1526 1527 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len); 1528 if (error != 0) 1529 return (error); 1530 if (opt_len == 0 || opt_value == NULL) 1531 return (EINVAL); 1532 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0') 1533 return (EINVAL); 1534 iv = strtoq(opt_value, &vtp, 0); 1535 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0')) 1536 return (EINVAL); 1537 if (iv < 0) 1538 return (EINVAL); 1539 switch (vtp[0]) { 1540 case 't': 1541 case 'T': 1542 iv *= 1024; 1543 case 'g': 1544 case 'G': 1545 iv *= 1024; 1546 case 'm': 1547 case 'M': 1548 iv *= 1024; 1549 case 'k': 1550 case 'K': 1551 iv *= 1024; 1552 case '\0': 1553 break; 1554 default: 1555 return (EINVAL); 1556 } 1557 *value = iv; 1558 1559 return (0); 1560 } 1561 1562 char * 1563 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1564 { 1565 struct vfsopt *opt; 1566 1567 *error = 0; 1568 TAILQ_FOREACH(opt, opts, link) { 1569 if (strcmp(name, opt->name) != 0) 1570 continue; 1571 opt->seen = 1; 1572 if (opt->len == 0 || 1573 ((char *)opt->value)[opt->len - 1] != '\0') { 1574 *error = EINVAL; 1575 return (NULL); 1576 } 1577 return (opt->value); 1578 } 1579 *error = ENOENT; 1580 return (NULL); 1581 } 1582 1583 int 1584 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, 1585 uint64_t val) 1586 { 1587 struct vfsopt *opt; 1588 1589 TAILQ_FOREACH(opt, opts, link) { 1590 if (strcmp(name, opt->name) == 0) { 1591 opt->seen = 1; 1592 if (w != NULL) 1593 *w |= val; 1594 return (1); 1595 } 1596 } 1597 if (w != NULL) 1598 *w &= ~val; 1599 return (0); 1600 } 1601 1602 int 1603 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1604 { 1605 va_list ap; 1606 struct vfsopt *opt; 1607 int ret; 1608 1609 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1610 1611 TAILQ_FOREACH(opt, opts, link) { 1612 if (strcmp(name, opt->name) != 0) 1613 continue; 1614 opt->seen = 1; 1615 if (opt->len == 0 || opt->value == NULL) 1616 return (0); 1617 if (((char *)opt->value)[opt->len - 1] != '\0') 1618 return (0); 1619 va_start(ap, fmt); 1620 ret = vsscanf(opt->value, fmt, ap); 1621 va_end(ap); 1622 return (ret); 1623 } 1624 return (0); 1625 } 1626 1627 int 1628 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 1629 { 1630 struct vfsopt *opt; 1631 1632 TAILQ_FOREACH(opt, opts, link) { 1633 if (strcmp(name, opt->name) != 0) 1634 continue; 1635 opt->seen = 1; 1636 if (opt->value == NULL) 1637 opt->len = len; 1638 else { 1639 if (opt->len != len) 1640 return (EINVAL); 1641 bcopy(value, opt->value, len); 1642 } 1643 return (0); 1644 } 1645 return (ENOENT); 1646 } 1647 1648 int 1649 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 1650 { 1651 struct vfsopt *opt; 1652 1653 TAILQ_FOREACH(opt, opts, link) { 1654 if (strcmp(name, opt->name) != 0) 1655 continue; 1656 opt->seen = 1; 1657 if (opt->value == NULL) 1658 opt->len = len; 1659 else { 1660 if (opt->len < len) 1661 return (EINVAL); 1662 opt->len = len; 1663 bcopy(value, opt->value, len); 1664 } 1665 return (0); 1666 } 1667 return (ENOENT); 1668 } 1669 1670 int 1671 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 1672 { 1673 struct vfsopt *opt; 1674 1675 TAILQ_FOREACH(opt, opts, link) { 1676 if (strcmp(name, opt->name) != 0) 1677 continue; 1678 opt->seen = 1; 1679 if (opt->value == NULL) 1680 opt->len = strlen(value) + 1; 1681 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 1682 return (EINVAL); 1683 return (0); 1684 } 1685 return (ENOENT); 1686 } 1687 1688 /* 1689 * Find and copy a mount option. 1690 * 1691 * The size of the buffer has to be specified 1692 * in len, if it is not the same length as the 1693 * mount option, EINVAL is returned. 1694 * Returns ENOENT if the option is not found. 1695 */ 1696 int 1697 vfs_copyopt(opts, name, dest, len) 1698 struct vfsoptlist *opts; 1699 const char *name; 1700 void *dest; 1701 int len; 1702 { 1703 struct vfsopt *opt; 1704 1705 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1706 1707 TAILQ_FOREACH(opt, opts, link) { 1708 if (strcmp(name, opt->name) == 0) { 1709 opt->seen = 1; 1710 if (len != opt->len) 1711 return (EINVAL); 1712 bcopy(opt->value, dest, opt->len); 1713 return (0); 1714 } 1715 } 1716 return (ENOENT); 1717 } 1718 1719 int 1720 __vfs_statfs(struct mount *mp, struct statfs *sbp) 1721 { 1722 int error; 1723 1724 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 1725 if (sbp != &mp->mnt_stat) 1726 *sbp = mp->mnt_stat; 1727 return (error); 1728 } 1729 1730 void 1731 vfs_mountedfrom(struct mount *mp, const char *from) 1732 { 1733 1734 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 1735 strlcpy(mp->mnt_stat.f_mntfromname, from, 1736 sizeof mp->mnt_stat.f_mntfromname); 1737 } 1738 1739 /* 1740 * --------------------------------------------------------------------- 1741 * This is the api for building mount args and mounting filesystems from 1742 * inside the kernel. 1743 * 1744 * The API works by accumulation of individual args. First error is 1745 * latched. 1746 * 1747 * XXX: should be documented in new manpage kernel_mount(9) 1748 */ 1749 1750 /* A memory allocation which must be freed when we are done */ 1751 struct mntaarg { 1752 SLIST_ENTRY(mntaarg) next; 1753 }; 1754 1755 /* The header for the mount arguments */ 1756 struct mntarg { 1757 struct iovec *v; 1758 int len; 1759 int error; 1760 SLIST_HEAD(, mntaarg) list; 1761 }; 1762 1763 /* 1764 * Add a boolean argument. 1765 * 1766 * flag is the boolean value. 1767 * name must start with "no". 1768 */ 1769 struct mntarg * 1770 mount_argb(struct mntarg *ma, int flag, const char *name) 1771 { 1772 1773 KASSERT(name[0] == 'n' && name[1] == 'o', 1774 ("mount_argb(...,%s): name must start with 'no'", name)); 1775 1776 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 1777 } 1778 1779 /* 1780 * Add an argument printf style 1781 */ 1782 struct mntarg * 1783 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 1784 { 1785 va_list ap; 1786 struct mntaarg *maa; 1787 struct sbuf *sb; 1788 int len; 1789 1790 if (ma == NULL) { 1791 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1792 SLIST_INIT(&ma->list); 1793 } 1794 if (ma->error) 1795 return (ma); 1796 1797 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1798 M_MOUNT, M_WAITOK); 1799 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1800 ma->v[ma->len].iov_len = strlen(name) + 1; 1801 ma->len++; 1802 1803 sb = sbuf_new_auto(); 1804 va_start(ap, fmt); 1805 sbuf_vprintf(sb, fmt, ap); 1806 va_end(ap); 1807 sbuf_finish(sb); 1808 len = sbuf_len(sb) + 1; 1809 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1810 SLIST_INSERT_HEAD(&ma->list, maa, next); 1811 bcopy(sbuf_data(sb), maa + 1, len); 1812 sbuf_delete(sb); 1813 1814 ma->v[ma->len].iov_base = maa + 1; 1815 ma->v[ma->len].iov_len = len; 1816 ma->len++; 1817 1818 return (ma); 1819 } 1820 1821 /* 1822 * Add an argument which is a userland string. 1823 */ 1824 struct mntarg * 1825 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 1826 { 1827 struct mntaarg *maa; 1828 char *tbuf; 1829 1830 if (val == NULL) 1831 return (ma); 1832 if (ma == NULL) { 1833 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1834 SLIST_INIT(&ma->list); 1835 } 1836 if (ma->error) 1837 return (ma); 1838 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1839 SLIST_INSERT_HEAD(&ma->list, maa, next); 1840 tbuf = (void *)(maa + 1); 1841 ma->error = copyinstr(val, tbuf, len, NULL); 1842 return (mount_arg(ma, name, tbuf, -1)); 1843 } 1844 1845 /* 1846 * Plain argument. 1847 * 1848 * If length is -1, treat value as a C string. 1849 */ 1850 struct mntarg * 1851 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 1852 { 1853 1854 if (ma == NULL) { 1855 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1856 SLIST_INIT(&ma->list); 1857 } 1858 if (ma->error) 1859 return (ma); 1860 1861 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1862 M_MOUNT, M_WAITOK); 1863 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1864 ma->v[ma->len].iov_len = strlen(name) + 1; 1865 ma->len++; 1866 1867 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 1868 if (len < 0) 1869 ma->v[ma->len].iov_len = strlen(val) + 1; 1870 else 1871 ma->v[ma->len].iov_len = len; 1872 ma->len++; 1873 return (ma); 1874 } 1875 1876 /* 1877 * Free a mntarg structure 1878 */ 1879 static void 1880 free_mntarg(struct mntarg *ma) 1881 { 1882 struct mntaarg *maa; 1883 1884 while (!SLIST_EMPTY(&ma->list)) { 1885 maa = SLIST_FIRST(&ma->list); 1886 SLIST_REMOVE_HEAD(&ma->list, next); 1887 free(maa, M_MOUNT); 1888 } 1889 free(ma->v, M_MOUNT); 1890 free(ma, M_MOUNT); 1891 } 1892 1893 /* 1894 * Mount a filesystem 1895 */ 1896 int 1897 kernel_mount(struct mntarg *ma, uint64_t flags) 1898 { 1899 struct uio auio; 1900 int error; 1901 1902 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 1903 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 1904 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 1905 1906 auio.uio_iov = ma->v; 1907 auio.uio_iovcnt = ma->len; 1908 auio.uio_segflg = UIO_SYSSPACE; 1909 1910 error = ma->error; 1911 if (!error) 1912 error = vfs_donmount(curthread, flags, &auio); 1913 free_mntarg(ma); 1914 return (error); 1915 } 1916 1917 /* 1918 * A printflike function to mount a filesystem. 1919 */ 1920 int 1921 kernel_vmount(int flags, ...) 1922 { 1923 struct mntarg *ma = NULL; 1924 va_list ap; 1925 const char *cp; 1926 const void *vp; 1927 int error; 1928 1929 va_start(ap, flags); 1930 for (;;) { 1931 cp = va_arg(ap, const char *); 1932 if (cp == NULL) 1933 break; 1934 vp = va_arg(ap, const void *); 1935 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 1936 } 1937 va_end(ap); 1938 1939 error = kernel_mount(ma, flags); 1940 return (error); 1941 } 1942 1943 void 1944 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp) 1945 { 1946 1947 bcopy(oexp, exp, sizeof(*oexp)); 1948 exp->ex_numsecflavors = 0; 1949 } 1950