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 atomic_add_acq_int(&vfsp->vfc_refcount, 1); 467 mp->mnt_op = vfsp->vfc_vfsops; 468 mp->mnt_vfc = vfsp; 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 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1); 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 else if (strcmp(opt->name, "automounted") == 0) { 653 fsflags |= MNT_AUTOMOUNTED; 654 vfs_freeopt(optlist, opt); 655 } 656 } 657 658 /* 659 * Be ultra-paranoid about making sure the type and fspath 660 * variables will fit in our mp buffers, including the 661 * terminating NUL. 662 */ 663 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) { 664 error = ENAMETOOLONG; 665 goto bail; 666 } 667 668 error = vfs_domount(td, fstype, fspath, fsflags, &optlist); 669 bail: 670 /* copyout the errmsg */ 671 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 672 && errmsg_len > 0 && errmsg != NULL) { 673 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 674 bcopy(errmsg, 675 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 676 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 677 } else { 678 copyout(errmsg, 679 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 680 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 681 } 682 } 683 684 if (optlist != NULL) 685 vfs_freeopts(optlist); 686 return (error); 687 } 688 689 /* 690 * Old mount API. 691 */ 692 #ifndef _SYS_SYSPROTO_H_ 693 struct mount_args { 694 char *type; 695 char *path; 696 int flags; 697 caddr_t data; 698 }; 699 #endif 700 /* ARGSUSED */ 701 int 702 sys_mount(td, uap) 703 struct thread *td; 704 struct mount_args /* { 705 char *type; 706 char *path; 707 int flags; 708 caddr_t data; 709 } */ *uap; 710 { 711 char *fstype; 712 struct vfsconf *vfsp = NULL; 713 struct mntarg *ma = NULL; 714 uint64_t flags; 715 int error; 716 717 /* 718 * Mount flags are now 64-bits. On 32-bit architectures only 719 * 32-bits are passed in, but from here on everything handles 720 * 64-bit flags correctly. 721 */ 722 flags = uap->flags; 723 724 AUDIT_ARG_FFLAGS(flags); 725 726 /* 727 * Filter out MNT_ROOTFS. We do not want clients of mount() in 728 * userspace to set this flag, but we must filter it out if we want 729 * MNT_UPDATE on the root file system to work. 730 * MNT_ROOTFS should only be set by the kernel when mounting its 731 * root file system. 732 */ 733 flags &= ~MNT_ROOTFS; 734 735 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 736 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 737 if (error) { 738 free(fstype, M_TEMP); 739 return (error); 740 } 741 742 AUDIT_ARG_TEXT(fstype); 743 vfsp = vfs_byname_kld(fstype, td, &error); 744 free(fstype, M_TEMP); 745 if (vfsp == NULL) 746 return (ENOENT); 747 if (vfsp->vfc_vfsops->vfs_cmount == NULL) 748 return (EOPNOTSUPP); 749 750 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN); 751 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 752 ma = mount_argb(ma, flags & MNT_RDONLY, "noro"); 753 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid"); 754 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec"); 755 756 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags); 757 return (error); 758 } 759 760 /* 761 * vfs_domount_first(): first file system mount (not update) 762 */ 763 static int 764 vfs_domount_first( 765 struct thread *td, /* Calling thread. */ 766 struct vfsconf *vfsp, /* File system type. */ 767 char *fspath, /* Mount path. */ 768 struct vnode *vp, /* Vnode to be covered. */ 769 uint64_t fsflags, /* Flags common to all filesystems. */ 770 struct vfsoptlist **optlist /* Options local to the filesystem. */ 771 ) 772 { 773 struct vattr va; 774 struct mount *mp; 775 struct vnode *newdp; 776 int error; 777 778 ASSERT_VOP_ELOCKED(vp, __func__); 779 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 780 781 /* 782 * If the user is not root, ensure that they own the directory 783 * onto which we are attempting to mount. 784 */ 785 error = VOP_GETATTR(vp, &va, td->td_ucred); 786 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 787 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0); 788 if (error == 0) 789 error = vinvalbuf(vp, V_SAVE, 0, 0); 790 if (error == 0 && vp->v_type != VDIR) 791 error = ENOTDIR; 792 if (error == 0) { 793 VI_LOCK(vp); 794 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 795 vp->v_iflag |= VI_MOUNT; 796 else 797 error = EBUSY; 798 VI_UNLOCK(vp); 799 } 800 if (error != 0) { 801 vput(vp); 802 return (error); 803 } 804 VOP_UNLOCK(vp, 0); 805 806 /* Allocate and initialize the filesystem. */ 807 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 808 /* XXXMAC: pass to vfs_mount_alloc? */ 809 mp->mnt_optnew = *optlist; 810 /* Set the mount level flags. */ 811 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 812 813 /* 814 * Mount the filesystem. 815 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 816 * get. No freeing of cn_pnbuf. 817 */ 818 error = VFS_MOUNT(mp); 819 if (error != 0) { 820 vfs_unbusy(mp); 821 vfs_mount_destroy(mp); 822 VI_LOCK(vp); 823 vp->v_iflag &= ~VI_MOUNT; 824 VI_UNLOCK(vp); 825 vrele(vp); 826 return (error); 827 } 828 829 if (mp->mnt_opt != NULL) 830 vfs_freeopts(mp->mnt_opt); 831 mp->mnt_opt = mp->mnt_optnew; 832 *optlist = NULL; 833 (void)VFS_STATFS(mp, &mp->mnt_stat); 834 835 /* 836 * Prevent external consumers of mount options from reading mnt_optnew. 837 */ 838 mp->mnt_optnew = NULL; 839 840 MNT_ILOCK(mp); 841 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 842 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 843 mp->mnt_kern_flag |= MNTK_ASYNC; 844 else 845 mp->mnt_kern_flag &= ~MNTK_ASYNC; 846 MNT_IUNLOCK(mp); 847 848 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 849 cache_purge(vp); 850 VI_LOCK(vp); 851 vp->v_iflag &= ~VI_MOUNT; 852 VI_UNLOCK(vp); 853 vp->v_mountedhere = mp; 854 /* Place the new filesystem at the end of the mount list. */ 855 mtx_lock(&mountlist_mtx); 856 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 857 mtx_unlock(&mountlist_mtx); 858 vfs_event_signal(NULL, VQ_MOUNT, 0); 859 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 860 panic("mount: lost mount"); 861 VOP_UNLOCK(vp, 0); 862 EVENTHANDLER_INVOKE(vfs_mounted, mp, newdp, td); 863 VOP_UNLOCK(newdp, 0); 864 mountcheckdirs(vp, newdp); 865 vrele(newdp); 866 if ((mp->mnt_flag & MNT_RDONLY) == 0) 867 vfs_allocate_syncvnode(mp); 868 vfs_unbusy(mp); 869 return (0); 870 } 871 872 /* 873 * vfs_domount_update(): update of mounted file system 874 */ 875 static int 876 vfs_domount_update( 877 struct thread *td, /* Calling thread. */ 878 struct vnode *vp, /* Mount point vnode. */ 879 uint64_t fsflags, /* Flags common to all filesystems. */ 880 struct vfsoptlist **optlist /* Options local to the filesystem. */ 881 ) 882 { 883 struct oexport_args oexport; 884 struct export_args export; 885 struct mount *mp; 886 int error, export_error; 887 uint64_t flag; 888 889 ASSERT_VOP_ELOCKED(vp, __func__); 890 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 891 mp = vp->v_mount; 892 893 if ((vp->v_vflag & VV_ROOT) == 0) { 894 if (vfs_copyopt(*optlist, "export", &export, sizeof(export)) 895 == 0) 896 error = EXDEV; 897 else 898 error = EINVAL; 899 vput(vp); 900 return (error); 901 } 902 903 /* 904 * We only allow the filesystem to be reloaded if it 905 * is currently mounted read-only. 906 */ 907 flag = mp->mnt_flag; 908 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 909 vput(vp); 910 return (EOPNOTSUPP); /* Needs translation */ 911 } 912 /* 913 * Only privileged root, or (if MNT_USER is set) the user that 914 * did the original mount is permitted to update it. 915 */ 916 error = vfs_suser(mp, td); 917 if (error != 0) { 918 vput(vp); 919 return (error); 920 } 921 if (vfs_busy(mp, MBF_NOWAIT)) { 922 vput(vp); 923 return (EBUSY); 924 } 925 VI_LOCK(vp); 926 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 927 VI_UNLOCK(vp); 928 vfs_unbusy(mp); 929 vput(vp); 930 return (EBUSY); 931 } 932 vp->v_iflag |= VI_MOUNT; 933 VI_UNLOCK(vp); 934 VOP_UNLOCK(vp, 0); 935 936 MNT_ILOCK(mp); 937 mp->mnt_flag &= ~MNT_UPDATEMASK; 938 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 939 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 940 if ((mp->mnt_flag & MNT_ASYNC) == 0) 941 mp->mnt_kern_flag &= ~MNTK_ASYNC; 942 MNT_IUNLOCK(mp); 943 mp->mnt_optnew = *optlist; 944 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 945 946 /* 947 * Mount the filesystem. 948 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 949 * get. No freeing of cn_pnbuf. 950 */ 951 error = VFS_MOUNT(mp); 952 953 export_error = 0; 954 if (error == 0) { 955 /* Process the export option. */ 956 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 957 sizeof(export)) == 0) { 958 export_error = vfs_export(mp, &export); 959 } else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 960 sizeof(oexport)) == 0) { 961 export.ex_flags = oexport.ex_flags; 962 export.ex_root = oexport.ex_root; 963 export.ex_anon = oexport.ex_anon; 964 export.ex_addr = oexport.ex_addr; 965 export.ex_addrlen = oexport.ex_addrlen; 966 export.ex_mask = oexport.ex_mask; 967 export.ex_masklen = oexport.ex_masklen; 968 export.ex_indexfile = oexport.ex_indexfile; 969 export.ex_numsecflavors = 0; 970 export_error = vfs_export(mp, &export); 971 } 972 } 973 974 MNT_ILOCK(mp); 975 if (error == 0) { 976 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 977 MNT_SNAPSHOT); 978 } else { 979 /* 980 * If we fail, restore old mount flags. MNT_QUOTA is special, 981 * because it is not part of MNT_UPDATEMASK, but it could have 982 * changed in the meantime if quotactl(2) was called. 983 * All in all we want current value of MNT_QUOTA, not the old 984 * one. 985 */ 986 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 987 } 988 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 989 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 990 mp->mnt_kern_flag |= MNTK_ASYNC; 991 else 992 mp->mnt_kern_flag &= ~MNTK_ASYNC; 993 MNT_IUNLOCK(mp); 994 995 if (error != 0) 996 goto end; 997 998 if (mp->mnt_opt != NULL) 999 vfs_freeopts(mp->mnt_opt); 1000 mp->mnt_opt = mp->mnt_optnew; 1001 *optlist = NULL; 1002 (void)VFS_STATFS(mp, &mp->mnt_stat); 1003 /* 1004 * Prevent external consumers of mount options from reading 1005 * mnt_optnew. 1006 */ 1007 mp->mnt_optnew = NULL; 1008 1009 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1010 vfs_allocate_syncvnode(mp); 1011 else 1012 vfs_deallocate_syncvnode(mp); 1013 end: 1014 vfs_unbusy(mp); 1015 VI_LOCK(vp); 1016 vp->v_iflag &= ~VI_MOUNT; 1017 VI_UNLOCK(vp); 1018 vrele(vp); 1019 return (error != 0 ? error : export_error); 1020 } 1021 1022 /* 1023 * vfs_domount(): actually attempt a filesystem mount. 1024 */ 1025 static int 1026 vfs_domount( 1027 struct thread *td, /* Calling thread. */ 1028 const char *fstype, /* Filesystem type. */ 1029 char *fspath, /* Mount path. */ 1030 uint64_t fsflags, /* Flags common to all filesystems. */ 1031 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1032 ) 1033 { 1034 struct vfsconf *vfsp; 1035 struct nameidata nd; 1036 struct vnode *vp; 1037 char *pathbuf; 1038 int error; 1039 1040 /* 1041 * Be ultra-paranoid about making sure the type and fspath 1042 * variables will fit in our mp buffers, including the 1043 * terminating NUL. 1044 */ 1045 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1046 return (ENAMETOOLONG); 1047 1048 if (jailed(td->td_ucred) || usermount == 0) { 1049 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1050 return (error); 1051 } 1052 1053 /* 1054 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1055 */ 1056 if (fsflags & MNT_EXPORTED) { 1057 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1058 if (error) 1059 return (error); 1060 } 1061 if (fsflags & MNT_SUIDDIR) { 1062 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1063 if (error) 1064 return (error); 1065 } 1066 /* 1067 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1068 */ 1069 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1070 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1071 fsflags |= MNT_NOSUID | MNT_USER; 1072 } 1073 1074 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1075 vfsp = NULL; 1076 if ((fsflags & MNT_UPDATE) == 0) { 1077 /* Don't try to load KLDs if we're mounting the root. */ 1078 if (fsflags & MNT_ROOTFS) 1079 vfsp = vfs_byname(fstype); 1080 else 1081 vfsp = vfs_byname_kld(fstype, td, &error); 1082 if (vfsp == NULL) 1083 return (ENODEV); 1084 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 1085 return (EPERM); 1086 } 1087 1088 /* 1089 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1090 */ 1091 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1092 UIO_SYSSPACE, fspath, td); 1093 error = namei(&nd); 1094 if (error != 0) 1095 return (error); 1096 NDFREE(&nd, NDF_ONLY_PNBUF); 1097 vp = nd.ni_vp; 1098 if ((fsflags & MNT_UPDATE) == 0) { 1099 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1100 strcpy(pathbuf, fspath); 1101 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN); 1102 /* debug.disablefullpath == 1 results in ENODEV */ 1103 if (error == 0 || error == ENODEV) { 1104 error = vfs_domount_first(td, vfsp, pathbuf, vp, 1105 fsflags, optlist); 1106 } 1107 free(pathbuf, M_TEMP); 1108 } else 1109 error = vfs_domount_update(td, vp, fsflags, optlist); 1110 1111 return (error); 1112 } 1113 1114 /* 1115 * Unmount a filesystem. 1116 * 1117 * Note: unmount takes a path to the vnode mounted on as argument, not 1118 * special file (as before). 1119 */ 1120 #ifndef _SYS_SYSPROTO_H_ 1121 struct unmount_args { 1122 char *path; 1123 int flags; 1124 }; 1125 #endif 1126 /* ARGSUSED */ 1127 int 1128 sys_unmount(struct thread *td, struct unmount_args *uap) 1129 { 1130 struct nameidata nd; 1131 struct mount *mp; 1132 char *pathbuf; 1133 int error, id0, id1; 1134 1135 AUDIT_ARG_VALUE(uap->flags); 1136 if (jailed(td->td_ucred) || usermount == 0) { 1137 error = priv_check(td, PRIV_VFS_UNMOUNT); 1138 if (error) 1139 return (error); 1140 } 1141 1142 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1143 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1144 if (error) { 1145 free(pathbuf, M_TEMP); 1146 return (error); 1147 } 1148 if (uap->flags & MNT_BYFSID) { 1149 AUDIT_ARG_TEXT(pathbuf); 1150 /* Decode the filesystem ID. */ 1151 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1152 free(pathbuf, M_TEMP); 1153 return (EINVAL); 1154 } 1155 1156 mtx_lock(&mountlist_mtx); 1157 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1158 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1159 mp->mnt_stat.f_fsid.val[1] == id1) { 1160 vfs_ref(mp); 1161 break; 1162 } 1163 } 1164 mtx_unlock(&mountlist_mtx); 1165 } else { 1166 /* 1167 * Try to find global path for path argument. 1168 */ 1169 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1170 UIO_SYSSPACE, pathbuf, td); 1171 if (namei(&nd) == 0) { 1172 NDFREE(&nd, NDF_ONLY_PNBUF); 1173 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf, 1174 MNAMELEN); 1175 if (error == 0 || error == ENODEV) 1176 vput(nd.ni_vp); 1177 } 1178 mtx_lock(&mountlist_mtx); 1179 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1180 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) { 1181 vfs_ref(mp); 1182 break; 1183 } 1184 } 1185 mtx_unlock(&mountlist_mtx); 1186 } 1187 free(pathbuf, M_TEMP); 1188 if (mp == NULL) { 1189 /* 1190 * Previously we returned ENOENT for a nonexistent path and 1191 * EINVAL for a non-mountpoint. We cannot tell these apart 1192 * now, so in the !MNT_BYFSID case return the more likely 1193 * EINVAL for compatibility. 1194 */ 1195 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1196 } 1197 1198 /* 1199 * Don't allow unmounting the root filesystem. 1200 */ 1201 if (mp->mnt_flag & MNT_ROOTFS) { 1202 vfs_rel(mp); 1203 return (EINVAL); 1204 } 1205 error = dounmount(mp, uap->flags, td); 1206 return (error); 1207 } 1208 1209 /* 1210 * Do the actual filesystem unmount. 1211 */ 1212 int 1213 dounmount(struct mount *mp, int flags, struct thread *td) 1214 { 1215 struct vnode *coveredvp, *fsrootvp; 1216 int error; 1217 uint64_t async_flag; 1218 int mnt_gen_r; 1219 1220 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1221 mnt_gen_r = mp->mnt_gen; 1222 VI_LOCK(coveredvp); 1223 vholdl(coveredvp); 1224 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1225 vdrop(coveredvp); 1226 /* 1227 * Check for mp being unmounted while waiting for the 1228 * covered vnode lock. 1229 */ 1230 if (coveredvp->v_mountedhere != mp || 1231 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1232 VOP_UNLOCK(coveredvp, 0); 1233 vfs_rel(mp); 1234 return (EBUSY); 1235 } 1236 } 1237 /* 1238 * Only privileged root, or (if MNT_USER is set) the user that did the 1239 * original mount is permitted to unmount this filesystem. 1240 */ 1241 error = vfs_suser(mp, td); 1242 if (error != 0) { 1243 if (coveredvp) 1244 VOP_UNLOCK(coveredvp, 0); 1245 vfs_rel(mp); 1246 return (error); 1247 } 1248 1249 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF); 1250 MNT_ILOCK(mp); 1251 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 || 1252 !TAILQ_EMPTY(&mp->mnt_uppers)) { 1253 MNT_IUNLOCK(mp); 1254 if (coveredvp) 1255 VOP_UNLOCK(coveredvp, 0); 1256 vn_finished_write(mp); 1257 return (EBUSY); 1258 } 1259 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1260 /* Allow filesystems to detect that a forced unmount is in progress. */ 1261 if (flags & MNT_FORCE) { 1262 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1263 MNT_IUNLOCK(mp); 1264 /* 1265 * Must be done after setting MNTK_UNMOUNTF and before 1266 * waiting for mnt_lockref to become 0. 1267 */ 1268 VFS_PURGE(mp); 1269 MNT_ILOCK(mp); 1270 } 1271 error = 0; 1272 if (mp->mnt_lockref) { 1273 mp->mnt_kern_flag |= MNTK_DRAINING; 1274 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1275 "mount drain", 0); 1276 } 1277 MNT_IUNLOCK(mp); 1278 KASSERT(mp->mnt_lockref == 0, 1279 ("%s: invalid lock refcount in the drain path @ %s:%d", 1280 __func__, __FILE__, __LINE__)); 1281 KASSERT(error == 0, 1282 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1283 __func__, __FILE__, __LINE__)); 1284 1285 if (mp->mnt_flag & MNT_EXPUBLIC) 1286 vfs_setpublicfs(NULL, NULL, NULL); 1287 1288 vfs_msync(mp, MNT_WAIT); 1289 MNT_ILOCK(mp); 1290 async_flag = mp->mnt_flag & MNT_ASYNC; 1291 mp->mnt_flag &= ~MNT_ASYNC; 1292 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1293 MNT_IUNLOCK(mp); 1294 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1295 vfs_deallocate_syncvnode(mp); 1296 /* 1297 * For forced unmounts, move process cdir/rdir refs on the fs root 1298 * vnode to the covered vnode. For non-forced unmounts we want 1299 * such references to cause an EBUSY error. 1300 */ 1301 if ((flags & MNT_FORCE) && 1302 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1303 if (mp->mnt_vnodecovered != NULL) 1304 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1305 if (fsrootvp == rootvnode) { 1306 vrele(rootvnode); 1307 rootvnode = NULL; 1308 } 1309 vput(fsrootvp); 1310 } 1311 if ((mp->mnt_flag & MNT_RDONLY) != 0 || (flags & MNT_FORCE) != 0 || 1312 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) 1313 error = VFS_UNMOUNT(mp, flags); 1314 vn_finished_write(mp); 1315 /* 1316 * If we failed to flush the dirty blocks for this mount point, 1317 * undo all the cdir/rdir and rootvnode changes we made above. 1318 * Unless we failed to do so because the device is reporting that 1319 * it doesn't exist anymore. 1320 */ 1321 if (error && error != ENXIO) { 1322 if ((flags & MNT_FORCE) && 1323 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1324 if (mp->mnt_vnodecovered != NULL) 1325 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1326 if (rootvnode == NULL) { 1327 rootvnode = fsrootvp; 1328 vref(rootvnode); 1329 } 1330 vput(fsrootvp); 1331 } 1332 MNT_ILOCK(mp); 1333 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1334 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1335 MNT_IUNLOCK(mp); 1336 vfs_allocate_syncvnode(mp); 1337 MNT_ILOCK(mp); 1338 } 1339 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1340 mp->mnt_flag |= async_flag; 1341 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1342 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1343 mp->mnt_kern_flag |= MNTK_ASYNC; 1344 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1345 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1346 wakeup(mp); 1347 } 1348 MNT_IUNLOCK(mp); 1349 if (coveredvp) 1350 VOP_UNLOCK(coveredvp, 0); 1351 return (error); 1352 } 1353 mtx_lock(&mountlist_mtx); 1354 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1355 mtx_unlock(&mountlist_mtx); 1356 EVENTHANDLER_INVOKE(vfs_unmounted, mp, td); 1357 if (coveredvp != NULL) { 1358 coveredvp->v_mountedhere = NULL; 1359 vput(coveredvp); 1360 } 1361 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1362 if (mp == rootdevmp) 1363 rootdevmp = NULL; 1364 vfs_mount_destroy(mp); 1365 return (0); 1366 } 1367 1368 /* 1369 * Report errors during filesystem mounting. 1370 */ 1371 void 1372 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1373 { 1374 struct vfsoptlist *moptlist = mp->mnt_optnew; 1375 va_list ap; 1376 int error, len; 1377 char *errmsg; 1378 1379 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1380 if (error || errmsg == NULL || len <= 0) 1381 return; 1382 1383 va_start(ap, fmt); 1384 vsnprintf(errmsg, (size_t)len, fmt, ap); 1385 va_end(ap); 1386 } 1387 1388 void 1389 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1390 { 1391 va_list ap; 1392 int error, len; 1393 char *errmsg; 1394 1395 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1396 if (error || errmsg == NULL || len <= 0) 1397 return; 1398 1399 va_start(ap, fmt); 1400 vsnprintf(errmsg, (size_t)len, fmt, ap); 1401 va_end(ap); 1402 } 1403 1404 /* 1405 * --------------------------------------------------------------------- 1406 * Functions for querying mount options/arguments from filesystems. 1407 */ 1408 1409 /* 1410 * Check that no unknown options are given 1411 */ 1412 int 1413 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1414 { 1415 struct vfsopt *opt; 1416 char errmsg[255]; 1417 const char **t, *p, *q; 1418 int ret = 0; 1419 1420 TAILQ_FOREACH(opt, opts, link) { 1421 p = opt->name; 1422 q = NULL; 1423 if (p[0] == 'n' && p[1] == 'o') 1424 q = p + 2; 1425 for(t = global_opts; *t != NULL; t++) { 1426 if (strcmp(*t, p) == 0) 1427 break; 1428 if (q != NULL) { 1429 if (strcmp(*t, q) == 0) 1430 break; 1431 } 1432 } 1433 if (*t != NULL) 1434 continue; 1435 for(t = legal; *t != NULL; t++) { 1436 if (strcmp(*t, p) == 0) 1437 break; 1438 if (q != NULL) { 1439 if (strcmp(*t, q) == 0) 1440 break; 1441 } 1442 } 1443 if (*t != NULL) 1444 continue; 1445 snprintf(errmsg, sizeof(errmsg), 1446 "mount option <%s> is unknown", p); 1447 ret = EINVAL; 1448 } 1449 if (ret != 0) { 1450 TAILQ_FOREACH(opt, opts, link) { 1451 if (strcmp(opt->name, "errmsg") == 0) { 1452 strncpy((char *)opt->value, errmsg, opt->len); 1453 break; 1454 } 1455 } 1456 if (opt == NULL) 1457 printf("%s\n", errmsg); 1458 } 1459 return (ret); 1460 } 1461 1462 /* 1463 * Get a mount option by its name. 1464 * 1465 * Return 0 if the option was found, ENOENT otherwise. 1466 * If len is non-NULL it will be filled with the length 1467 * of the option. If buf is non-NULL, it will be filled 1468 * with the address of the option. 1469 */ 1470 int 1471 vfs_getopt(opts, name, buf, len) 1472 struct vfsoptlist *opts; 1473 const char *name; 1474 void **buf; 1475 int *len; 1476 { 1477 struct vfsopt *opt; 1478 1479 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1480 1481 TAILQ_FOREACH(opt, opts, link) { 1482 if (strcmp(name, opt->name) == 0) { 1483 opt->seen = 1; 1484 if (len != NULL) 1485 *len = opt->len; 1486 if (buf != NULL) 1487 *buf = opt->value; 1488 return (0); 1489 } 1490 } 1491 return (ENOENT); 1492 } 1493 1494 int 1495 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1496 { 1497 struct vfsopt *opt; 1498 1499 if (opts == NULL) 1500 return (-1); 1501 1502 TAILQ_FOREACH(opt, opts, link) { 1503 if (strcmp(name, opt->name) == 0) { 1504 opt->seen = 1; 1505 return (opt->pos); 1506 } 1507 } 1508 return (-1); 1509 } 1510 1511 int 1512 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value) 1513 { 1514 char *opt_value, *vtp; 1515 quad_t iv; 1516 int error, opt_len; 1517 1518 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len); 1519 if (error != 0) 1520 return (error); 1521 if (opt_len == 0 || opt_value == NULL) 1522 return (EINVAL); 1523 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0') 1524 return (EINVAL); 1525 iv = strtoq(opt_value, &vtp, 0); 1526 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0')) 1527 return (EINVAL); 1528 if (iv < 0) 1529 return (EINVAL); 1530 switch (vtp[0]) { 1531 case 't': 1532 case 'T': 1533 iv *= 1024; 1534 case 'g': 1535 case 'G': 1536 iv *= 1024; 1537 case 'm': 1538 case 'M': 1539 iv *= 1024; 1540 case 'k': 1541 case 'K': 1542 iv *= 1024; 1543 case '\0': 1544 break; 1545 default: 1546 return (EINVAL); 1547 } 1548 *value = iv; 1549 1550 return (0); 1551 } 1552 1553 char * 1554 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1555 { 1556 struct vfsopt *opt; 1557 1558 *error = 0; 1559 TAILQ_FOREACH(opt, opts, link) { 1560 if (strcmp(name, opt->name) != 0) 1561 continue; 1562 opt->seen = 1; 1563 if (opt->len == 0 || 1564 ((char *)opt->value)[opt->len - 1] != '\0') { 1565 *error = EINVAL; 1566 return (NULL); 1567 } 1568 return (opt->value); 1569 } 1570 *error = ENOENT; 1571 return (NULL); 1572 } 1573 1574 int 1575 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, 1576 uint64_t val) 1577 { 1578 struct vfsopt *opt; 1579 1580 TAILQ_FOREACH(opt, opts, link) { 1581 if (strcmp(name, opt->name) == 0) { 1582 opt->seen = 1; 1583 if (w != NULL) 1584 *w |= val; 1585 return (1); 1586 } 1587 } 1588 if (w != NULL) 1589 *w &= ~val; 1590 return (0); 1591 } 1592 1593 int 1594 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1595 { 1596 va_list ap; 1597 struct vfsopt *opt; 1598 int ret; 1599 1600 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1601 1602 TAILQ_FOREACH(opt, opts, link) { 1603 if (strcmp(name, opt->name) != 0) 1604 continue; 1605 opt->seen = 1; 1606 if (opt->len == 0 || opt->value == NULL) 1607 return (0); 1608 if (((char *)opt->value)[opt->len - 1] != '\0') 1609 return (0); 1610 va_start(ap, fmt); 1611 ret = vsscanf(opt->value, fmt, ap); 1612 va_end(ap); 1613 return (ret); 1614 } 1615 return (0); 1616 } 1617 1618 int 1619 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 1620 { 1621 struct vfsopt *opt; 1622 1623 TAILQ_FOREACH(opt, opts, link) { 1624 if (strcmp(name, opt->name) != 0) 1625 continue; 1626 opt->seen = 1; 1627 if (opt->value == NULL) 1628 opt->len = len; 1629 else { 1630 if (opt->len != len) 1631 return (EINVAL); 1632 bcopy(value, opt->value, len); 1633 } 1634 return (0); 1635 } 1636 return (ENOENT); 1637 } 1638 1639 int 1640 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 1641 { 1642 struct vfsopt *opt; 1643 1644 TAILQ_FOREACH(opt, opts, link) { 1645 if (strcmp(name, opt->name) != 0) 1646 continue; 1647 opt->seen = 1; 1648 if (opt->value == NULL) 1649 opt->len = len; 1650 else { 1651 if (opt->len < len) 1652 return (EINVAL); 1653 opt->len = len; 1654 bcopy(value, opt->value, len); 1655 } 1656 return (0); 1657 } 1658 return (ENOENT); 1659 } 1660 1661 int 1662 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 1663 { 1664 struct vfsopt *opt; 1665 1666 TAILQ_FOREACH(opt, opts, link) { 1667 if (strcmp(name, opt->name) != 0) 1668 continue; 1669 opt->seen = 1; 1670 if (opt->value == NULL) 1671 opt->len = strlen(value) + 1; 1672 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 1673 return (EINVAL); 1674 return (0); 1675 } 1676 return (ENOENT); 1677 } 1678 1679 /* 1680 * Find and copy a mount option. 1681 * 1682 * The size of the buffer has to be specified 1683 * in len, if it is not the same length as the 1684 * mount option, EINVAL is returned. 1685 * Returns ENOENT if the option is not found. 1686 */ 1687 int 1688 vfs_copyopt(opts, name, dest, len) 1689 struct vfsoptlist *opts; 1690 const char *name; 1691 void *dest; 1692 int len; 1693 { 1694 struct vfsopt *opt; 1695 1696 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1697 1698 TAILQ_FOREACH(opt, opts, link) { 1699 if (strcmp(name, opt->name) == 0) { 1700 opt->seen = 1; 1701 if (len != opt->len) 1702 return (EINVAL); 1703 bcopy(opt->value, dest, opt->len); 1704 return (0); 1705 } 1706 } 1707 return (ENOENT); 1708 } 1709 1710 int 1711 __vfs_statfs(struct mount *mp, struct statfs *sbp) 1712 { 1713 int error; 1714 1715 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 1716 if (sbp != &mp->mnt_stat) 1717 *sbp = mp->mnt_stat; 1718 return (error); 1719 } 1720 1721 void 1722 vfs_mountedfrom(struct mount *mp, const char *from) 1723 { 1724 1725 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 1726 strlcpy(mp->mnt_stat.f_mntfromname, from, 1727 sizeof mp->mnt_stat.f_mntfromname); 1728 } 1729 1730 /* 1731 * --------------------------------------------------------------------- 1732 * This is the api for building mount args and mounting filesystems from 1733 * inside the kernel. 1734 * 1735 * The API works by accumulation of individual args. First error is 1736 * latched. 1737 * 1738 * XXX: should be documented in new manpage kernel_mount(9) 1739 */ 1740 1741 /* A memory allocation which must be freed when we are done */ 1742 struct mntaarg { 1743 SLIST_ENTRY(mntaarg) next; 1744 }; 1745 1746 /* The header for the mount arguments */ 1747 struct mntarg { 1748 struct iovec *v; 1749 int len; 1750 int error; 1751 SLIST_HEAD(, mntaarg) list; 1752 }; 1753 1754 /* 1755 * Add a boolean argument. 1756 * 1757 * flag is the boolean value. 1758 * name must start with "no". 1759 */ 1760 struct mntarg * 1761 mount_argb(struct mntarg *ma, int flag, const char *name) 1762 { 1763 1764 KASSERT(name[0] == 'n' && name[1] == 'o', 1765 ("mount_argb(...,%s): name must start with 'no'", name)); 1766 1767 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 1768 } 1769 1770 /* 1771 * Add an argument printf style 1772 */ 1773 struct mntarg * 1774 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 1775 { 1776 va_list ap; 1777 struct mntaarg *maa; 1778 struct sbuf *sb; 1779 int len; 1780 1781 if (ma == NULL) { 1782 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1783 SLIST_INIT(&ma->list); 1784 } 1785 if (ma->error) 1786 return (ma); 1787 1788 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1789 M_MOUNT, M_WAITOK); 1790 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1791 ma->v[ma->len].iov_len = strlen(name) + 1; 1792 ma->len++; 1793 1794 sb = sbuf_new_auto(); 1795 va_start(ap, fmt); 1796 sbuf_vprintf(sb, fmt, ap); 1797 va_end(ap); 1798 sbuf_finish(sb); 1799 len = sbuf_len(sb) + 1; 1800 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1801 SLIST_INSERT_HEAD(&ma->list, maa, next); 1802 bcopy(sbuf_data(sb), maa + 1, len); 1803 sbuf_delete(sb); 1804 1805 ma->v[ma->len].iov_base = maa + 1; 1806 ma->v[ma->len].iov_len = len; 1807 ma->len++; 1808 1809 return (ma); 1810 } 1811 1812 /* 1813 * Add an argument which is a userland string. 1814 */ 1815 struct mntarg * 1816 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 1817 { 1818 struct mntaarg *maa; 1819 char *tbuf; 1820 1821 if (val == NULL) 1822 return (ma); 1823 if (ma == NULL) { 1824 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1825 SLIST_INIT(&ma->list); 1826 } 1827 if (ma->error) 1828 return (ma); 1829 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1830 SLIST_INSERT_HEAD(&ma->list, maa, next); 1831 tbuf = (void *)(maa + 1); 1832 ma->error = copyinstr(val, tbuf, len, NULL); 1833 return (mount_arg(ma, name, tbuf, -1)); 1834 } 1835 1836 /* 1837 * Plain argument. 1838 * 1839 * If length is -1, treat value as a C string. 1840 */ 1841 struct mntarg * 1842 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 1843 { 1844 1845 if (ma == NULL) { 1846 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1847 SLIST_INIT(&ma->list); 1848 } 1849 if (ma->error) 1850 return (ma); 1851 1852 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1853 M_MOUNT, M_WAITOK); 1854 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1855 ma->v[ma->len].iov_len = strlen(name) + 1; 1856 ma->len++; 1857 1858 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 1859 if (len < 0) 1860 ma->v[ma->len].iov_len = strlen(val) + 1; 1861 else 1862 ma->v[ma->len].iov_len = len; 1863 ma->len++; 1864 return (ma); 1865 } 1866 1867 /* 1868 * Free a mntarg structure 1869 */ 1870 static void 1871 free_mntarg(struct mntarg *ma) 1872 { 1873 struct mntaarg *maa; 1874 1875 while (!SLIST_EMPTY(&ma->list)) { 1876 maa = SLIST_FIRST(&ma->list); 1877 SLIST_REMOVE_HEAD(&ma->list, next); 1878 free(maa, M_MOUNT); 1879 } 1880 free(ma->v, M_MOUNT); 1881 free(ma, M_MOUNT); 1882 } 1883 1884 /* 1885 * Mount a filesystem 1886 */ 1887 int 1888 kernel_mount(struct mntarg *ma, uint64_t flags) 1889 { 1890 struct uio auio; 1891 int error; 1892 1893 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 1894 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 1895 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 1896 1897 auio.uio_iov = ma->v; 1898 auio.uio_iovcnt = ma->len; 1899 auio.uio_segflg = UIO_SYSSPACE; 1900 1901 error = ma->error; 1902 if (!error) 1903 error = vfs_donmount(curthread, flags, &auio); 1904 free_mntarg(ma); 1905 return (error); 1906 } 1907 1908 /* 1909 * A printflike function to mount a filesystem. 1910 */ 1911 int 1912 kernel_vmount(int flags, ...) 1913 { 1914 struct mntarg *ma = NULL; 1915 va_list ap; 1916 const char *cp; 1917 const void *vp; 1918 int error; 1919 1920 va_start(ap, flags); 1921 for (;;) { 1922 cp = va_arg(ap, const char *); 1923 if (cp == NULL) 1924 break; 1925 vp = va_arg(ap, const void *); 1926 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 1927 } 1928 va_end(ap); 1929 1930 error = kernel_mount(ma, flags); 1931 return (error); 1932 } 1933 1934 void 1935 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp) 1936 { 1937 1938 bcopy(oexp, exp, sizeof(*oexp)); 1939 exp->ex_numsecflavors = 0; 1940 } 1941