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