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, 74 char *fspath, int 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 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 83 static uma_zone_t mount_zone; 84 85 /* List of mounted filesystems. */ 86 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 87 88 /* For any iteration/modification of mountlist */ 89 struct mtx mountlist_mtx; 90 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 91 92 /* 93 * Global opts, taken by all filesystems 94 */ 95 static const char *global_opts[] = { 96 "errmsg", 97 "fstype", 98 "fspath", 99 "ro", 100 "rw", 101 "nosuid", 102 "noexec", 103 NULL 104 }; 105 106 static int 107 mount_init(void *mem, int size, int flags) 108 { 109 struct mount *mp; 110 111 mp = (struct mount *)mem; 112 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 113 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 114 return (0); 115 } 116 117 static void 118 mount_fini(void *mem, int size) 119 { 120 struct mount *mp; 121 122 mp = (struct mount *)mem; 123 lockdestroy(&mp->mnt_explock); 124 mtx_destroy(&mp->mnt_mtx); 125 } 126 127 static void 128 vfs_mount_init(void *dummy __unused) 129 { 130 131 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL, 132 NULL, mount_init, mount_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 133 } 134 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL); 135 136 /* 137 * --------------------------------------------------------------------- 138 * Functions for building and sanitizing the mount options 139 */ 140 141 /* Remove one mount option. */ 142 static void 143 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 144 { 145 146 TAILQ_REMOVE(opts, opt, link); 147 free(opt->name, M_MOUNT); 148 if (opt->value != NULL) 149 free(opt->value, M_MOUNT); 150 free(opt, M_MOUNT); 151 } 152 153 /* Release all resources related to the mount options. */ 154 void 155 vfs_freeopts(struct vfsoptlist *opts) 156 { 157 struct vfsopt *opt; 158 159 while (!TAILQ_EMPTY(opts)) { 160 opt = TAILQ_FIRST(opts); 161 vfs_freeopt(opts, opt); 162 } 163 free(opts, M_MOUNT); 164 } 165 166 void 167 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 168 { 169 struct vfsopt *opt, *temp; 170 171 if (opts == NULL) 172 return; 173 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 174 if (strcmp(opt->name, name) == 0) 175 vfs_freeopt(opts, opt); 176 } 177 } 178 179 static int 180 vfs_isopt_ro(const char *opt) 181 { 182 183 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 || 184 strcmp(opt, "norw") == 0) 185 return (1); 186 return (0); 187 } 188 189 static int 190 vfs_isopt_rw(const char *opt) 191 { 192 193 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0) 194 return (1); 195 return (0); 196 } 197 198 /* 199 * Check if options are equal (with or without the "no" prefix). 200 */ 201 static int 202 vfs_equalopts(const char *opt1, const char *opt2) 203 { 204 char *p; 205 206 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 207 if (strcmp(opt1, opt2) == 0) 208 return (1); 209 /* "noopt" vs. "opt" */ 210 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 211 return (1); 212 /* "opt" vs. "noopt" */ 213 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 214 return (1); 215 while ((p = strchr(opt1, '.')) != NULL && 216 !strncmp(opt1, opt2, ++p - opt1)) { 217 opt2 += p - opt1; 218 opt1 = p; 219 /* "foo.noopt" vs. "foo.opt" */ 220 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 221 return (1); 222 /* "foo.opt" vs. "foo.noopt" */ 223 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 224 return (1); 225 } 226 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */ 227 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) && 228 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2))) 229 return (1); 230 return (0); 231 } 232 233 /* 234 * If a mount option is specified several times, 235 * (with or without the "no" prefix) only keep 236 * the last occurence of it. 237 */ 238 static void 239 vfs_sanitizeopts(struct vfsoptlist *opts) 240 { 241 struct vfsopt *opt, *opt2, *tmp; 242 243 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 244 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 245 while (opt2 != NULL) { 246 if (vfs_equalopts(opt->name, opt2->name)) { 247 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 248 vfs_freeopt(opts, opt2); 249 opt2 = tmp; 250 } else { 251 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 252 } 253 } 254 } 255 } 256 257 /* 258 * Build a linked list of mount options from a struct uio. 259 */ 260 int 261 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 262 { 263 struct vfsoptlist *opts; 264 struct vfsopt *opt; 265 size_t memused, namelen, optlen; 266 unsigned int i, iovcnt; 267 int error; 268 269 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 270 TAILQ_INIT(opts); 271 memused = 0; 272 iovcnt = auio->uio_iovcnt; 273 for (i = 0; i < iovcnt; i += 2) { 274 namelen = auio->uio_iov[i].iov_len; 275 optlen = auio->uio_iov[i + 1].iov_len; 276 memused += sizeof(struct vfsopt) + optlen + namelen; 277 /* 278 * Avoid consuming too much memory, and attempts to overflow 279 * memused. 280 */ 281 if (memused > VFS_MOUNTARG_SIZE_MAX || 282 optlen > VFS_MOUNTARG_SIZE_MAX || 283 namelen > VFS_MOUNTARG_SIZE_MAX) { 284 error = EINVAL; 285 goto bad; 286 } 287 288 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 289 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 290 opt->value = NULL; 291 opt->len = 0; 292 opt->pos = i / 2; 293 opt->seen = 0; 294 295 /* 296 * Do this early, so jumps to "bad" will free the current 297 * option. 298 */ 299 TAILQ_INSERT_TAIL(opts, opt, link); 300 301 if (auio->uio_segflg == UIO_SYSSPACE) { 302 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 303 } else { 304 error = copyin(auio->uio_iov[i].iov_base, opt->name, 305 namelen); 306 if (error) 307 goto bad; 308 } 309 /* Ensure names are null-terminated strings. */ 310 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 311 error = EINVAL; 312 goto bad; 313 } 314 if (optlen != 0) { 315 opt->len = optlen; 316 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 317 if (auio->uio_segflg == UIO_SYSSPACE) { 318 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 319 optlen); 320 } else { 321 error = copyin(auio->uio_iov[i + 1].iov_base, 322 opt->value, optlen); 323 if (error) 324 goto bad; 325 } 326 } 327 } 328 vfs_sanitizeopts(opts); 329 *options = opts; 330 return (0); 331 bad: 332 vfs_freeopts(opts); 333 return (error); 334 } 335 336 /* 337 * Merge the old mount options with the new ones passed 338 * in the MNT_UPDATE case. 339 * 340 * XXX: This function will keep a "nofoo" option in the new 341 * options. E.g, if the option's canonical name is "foo", 342 * "nofoo" ends up in the mount point's active options. 343 */ 344 static void 345 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts) 346 { 347 struct vfsopt *opt, *new; 348 349 TAILQ_FOREACH(opt, oldopts, link) { 350 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 351 new->name = strdup(opt->name, M_MOUNT); 352 if (opt->len != 0) { 353 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 354 bcopy(opt->value, new->value, opt->len); 355 } else 356 new->value = NULL; 357 new->len = opt->len; 358 new->seen = opt->seen; 359 TAILQ_INSERT_HEAD(toopts, new, link); 360 } 361 vfs_sanitizeopts(toopts); 362 } 363 364 /* 365 * Mount a filesystem. 366 */ 367 int 368 sys_nmount(td, uap) 369 struct thread *td; 370 struct nmount_args /* { 371 struct iovec *iovp; 372 unsigned int iovcnt; 373 int flags; 374 } */ *uap; 375 { 376 struct uio *auio; 377 int error; 378 u_int iovcnt; 379 380 AUDIT_ARG_FFLAGS(uap->flags); 381 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 382 uap->iovp, uap->iovcnt, uap->flags); 383 384 /* 385 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 386 * userspace to set this flag, but we must filter it out if we want 387 * MNT_UPDATE on the root file system to work. 388 * MNT_ROOTFS should only be set by the kernel when mounting its 389 * root file system. 390 */ 391 uap->flags &= ~MNT_ROOTFS; 392 393 iovcnt = uap->iovcnt; 394 /* 395 * Check that we have an even number of iovec's 396 * and that we have at least two options. 397 */ 398 if ((iovcnt & 1) || (iovcnt < 4)) { 399 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 400 uap->iovcnt); 401 return (EINVAL); 402 } 403 404 error = copyinuio(uap->iovp, iovcnt, &auio); 405 if (error) { 406 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 407 __func__, error); 408 return (error); 409 } 410 error = vfs_donmount(td, uap->flags, auio); 411 412 free(auio, M_IOV); 413 return (error); 414 } 415 416 /* 417 * --------------------------------------------------------------------- 418 * Various utility functions 419 */ 420 421 void 422 vfs_ref(struct mount *mp) 423 { 424 425 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 426 MNT_ILOCK(mp); 427 MNT_REF(mp); 428 MNT_IUNLOCK(mp); 429 } 430 431 void 432 vfs_rel(struct mount *mp) 433 { 434 435 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 436 MNT_ILOCK(mp); 437 MNT_REL(mp); 438 MNT_IUNLOCK(mp); 439 } 440 441 /* 442 * Allocate and initialize the mount point struct. 443 */ 444 struct mount * 445 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 446 struct ucred *cred) 447 { 448 struct mount *mp; 449 450 mp = uma_zalloc(mount_zone, M_WAITOK); 451 bzero(&mp->mnt_startzero, 452 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 453 TAILQ_INIT(&mp->mnt_nvnodelist); 454 mp->mnt_nvnodelistsize = 0; 455 mp->mnt_ref = 0; 456 (void) vfs_busy(mp, MBF_NOWAIT); 457 mp->mnt_op = vfsp->vfc_vfsops; 458 mp->mnt_vfc = vfsp; 459 vfsp->vfc_refcount++; /* XXX Unlocked */ 460 mp->mnt_stat.f_type = vfsp->vfc_typenum; 461 mp->mnt_gen++; 462 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 463 mp->mnt_vnodecovered = vp; 464 mp->mnt_cred = crdup(cred); 465 mp->mnt_stat.f_owner = cred->cr_uid; 466 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 467 mp->mnt_iosize_max = DFLTPHYS; 468 #ifdef MAC 469 mac_mount_init(mp); 470 mac_mount_create(cred, mp); 471 #endif 472 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 473 return (mp); 474 } 475 476 /* 477 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 478 */ 479 void 480 vfs_mount_destroy(struct mount *mp) 481 { 482 483 MNT_ILOCK(mp); 484 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 485 if (mp->mnt_kern_flag & MNTK_MWAIT) { 486 mp->mnt_kern_flag &= ~MNTK_MWAIT; 487 wakeup(mp); 488 } 489 while (mp->mnt_ref) 490 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 491 KASSERT(mp->mnt_ref == 0, 492 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 493 __FILE__, __LINE__)); 494 if (mp->mnt_writeopcount != 0) 495 panic("vfs_mount_destroy: nonzero writeopcount"); 496 if (mp->mnt_secondary_writes != 0) 497 panic("vfs_mount_destroy: nonzero secondary_writes"); 498 mp->mnt_vfc->vfc_refcount--; 499 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 500 struct vnode *vp; 501 502 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 503 vprint("", vp); 504 panic("unmount: dangling vnode"); 505 } 506 if (mp->mnt_nvnodelistsize != 0) 507 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 508 if (mp->mnt_lockref != 0) 509 panic("vfs_mount_destroy: nonzero lock refcount"); 510 MNT_IUNLOCK(mp); 511 #ifdef MAC 512 mac_mount_destroy(mp); 513 #endif 514 if (mp->mnt_opt != NULL) 515 vfs_freeopts(mp->mnt_opt); 516 crfree(mp->mnt_cred); 517 uma_zfree(mount_zone, mp); 518 } 519 520 int 521 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 522 { 523 struct vfsoptlist *optlist; 524 struct vfsopt *opt, *tmp_opt; 525 char *fstype, *fspath, *errmsg; 526 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 527 528 errmsg = fspath = NULL; 529 errmsg_len = fspathlen = 0; 530 errmsg_pos = -1; 531 532 error = vfs_buildopts(fsoptions, &optlist); 533 if (error) 534 return (error); 535 536 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 537 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 538 539 /* 540 * We need these two options before the others, 541 * and they are mandatory for any filesystem. 542 * Ensure they are NUL terminated as well. 543 */ 544 fstypelen = 0; 545 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 546 if (error || fstype[fstypelen - 1] != '\0') { 547 error = EINVAL; 548 if (errmsg != NULL) 549 strncpy(errmsg, "Invalid fstype", errmsg_len); 550 goto bail; 551 } 552 fspathlen = 0; 553 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 554 if (error || fspath[fspathlen - 1] != '\0') { 555 error = EINVAL; 556 if (errmsg != NULL) 557 strncpy(errmsg, "Invalid fspath", errmsg_len); 558 goto bail; 559 } 560 561 /* 562 * We need to see if we have the "update" option 563 * before we call vfs_domount(), since vfs_domount() has special 564 * logic based on MNT_UPDATE. This is very important 565 * when we want to update the root filesystem. 566 */ 567 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 568 if (strcmp(opt->name, "update") == 0) { 569 fsflags |= MNT_UPDATE; 570 vfs_freeopt(optlist, opt); 571 } 572 else if (strcmp(opt->name, "async") == 0) 573 fsflags |= MNT_ASYNC; 574 else if (strcmp(opt->name, "force") == 0) { 575 fsflags |= MNT_FORCE; 576 vfs_freeopt(optlist, opt); 577 } 578 else if (strcmp(opt->name, "reload") == 0) { 579 fsflags |= MNT_RELOAD; 580 vfs_freeopt(optlist, opt); 581 } 582 else if (strcmp(opt->name, "multilabel") == 0) 583 fsflags |= MNT_MULTILABEL; 584 else if (strcmp(opt->name, "noasync") == 0) 585 fsflags &= ~MNT_ASYNC; 586 else if (strcmp(opt->name, "noatime") == 0) 587 fsflags |= MNT_NOATIME; 588 else if (strcmp(opt->name, "atime") == 0) { 589 free(opt->name, M_MOUNT); 590 opt->name = strdup("nonoatime", M_MOUNT); 591 } 592 else if (strcmp(opt->name, "noclusterr") == 0) 593 fsflags |= MNT_NOCLUSTERR; 594 else if (strcmp(opt->name, "clusterr") == 0) { 595 free(opt->name, M_MOUNT); 596 opt->name = strdup("nonoclusterr", M_MOUNT); 597 } 598 else if (strcmp(opt->name, "noclusterw") == 0) 599 fsflags |= MNT_NOCLUSTERW; 600 else if (strcmp(opt->name, "clusterw") == 0) { 601 free(opt->name, M_MOUNT); 602 opt->name = strdup("nonoclusterw", M_MOUNT); 603 } 604 else if (strcmp(opt->name, "noexec") == 0) 605 fsflags |= MNT_NOEXEC; 606 else if (strcmp(opt->name, "exec") == 0) { 607 free(opt->name, M_MOUNT); 608 opt->name = strdup("nonoexec", M_MOUNT); 609 } 610 else if (strcmp(opt->name, "nosuid") == 0) 611 fsflags |= MNT_NOSUID; 612 else if (strcmp(opt->name, "suid") == 0) { 613 free(opt->name, M_MOUNT); 614 opt->name = strdup("nonosuid", M_MOUNT); 615 } 616 else if (strcmp(opt->name, "nosymfollow") == 0) 617 fsflags |= MNT_NOSYMFOLLOW; 618 else if (strcmp(opt->name, "symfollow") == 0) { 619 free(opt->name, M_MOUNT); 620 opt->name = strdup("nonosymfollow", M_MOUNT); 621 } 622 else if (strcmp(opt->name, "noro") == 0) 623 fsflags &= ~MNT_RDONLY; 624 else if (strcmp(opt->name, "rw") == 0) 625 fsflags &= ~MNT_RDONLY; 626 else if (strcmp(opt->name, "ro") == 0) 627 fsflags |= MNT_RDONLY; 628 else if (strcmp(opt->name, "rdonly") == 0) { 629 free(opt->name, M_MOUNT); 630 opt->name = strdup("ro", M_MOUNT); 631 fsflags |= MNT_RDONLY; 632 } 633 else if (strcmp(opt->name, "suiddir") == 0) 634 fsflags |= MNT_SUIDDIR; 635 else if (strcmp(opt->name, "sync") == 0) 636 fsflags |= MNT_SYNCHRONOUS; 637 else if (strcmp(opt->name, "union") == 0) 638 fsflags |= MNT_UNION; 639 } 640 641 /* 642 * Be ultra-paranoid about making sure the type and fspath 643 * variables will fit in our mp buffers, including the 644 * terminating NUL. 645 */ 646 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 647 error = ENAMETOOLONG; 648 goto bail; 649 } 650 651 error = vfs_domount(td, fstype, fspath, fsflags, &optlist); 652 bail: 653 /* copyout the errmsg */ 654 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 655 && errmsg_len > 0 && errmsg != NULL) { 656 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 657 bcopy(errmsg, 658 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 659 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 660 } else { 661 copyout(errmsg, 662 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 663 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 664 } 665 } 666 667 if (optlist != NULL) 668 vfs_freeopts(optlist); 669 return (error); 670 } 671 672 /* 673 * Old mount API. 674 */ 675 #ifndef _SYS_SYSPROTO_H_ 676 struct mount_args { 677 char *type; 678 char *path; 679 int flags; 680 caddr_t data; 681 }; 682 #endif 683 /* ARGSUSED */ 684 int 685 sys_mount(td, uap) 686 struct thread *td; 687 struct mount_args /* { 688 char *type; 689 char *path; 690 int flags; 691 caddr_t data; 692 } */ *uap; 693 { 694 char *fstype; 695 struct vfsconf *vfsp = NULL; 696 struct mntarg *ma = NULL; 697 int error; 698 699 AUDIT_ARG_FFLAGS(uap->flags); 700 701 /* 702 * Filter out MNT_ROOTFS. We do not want clients of mount() in 703 * userspace to set this flag, but we must filter it out if we want 704 * MNT_UPDATE on the root file system to work. 705 * MNT_ROOTFS should only be set by the kernel when mounting its 706 * root file system. 707 */ 708 uap->flags &= ~MNT_ROOTFS; 709 710 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 711 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 712 if (error) { 713 free(fstype, M_TEMP); 714 return (error); 715 } 716 717 AUDIT_ARG_TEXT(fstype); 718 mtx_lock(&Giant); 719 vfsp = vfs_byname_kld(fstype, td, &error); 720 free(fstype, M_TEMP); 721 if (vfsp == NULL) { 722 mtx_unlock(&Giant); 723 return (ENOENT); 724 } 725 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 726 mtx_unlock(&Giant); 727 return (EOPNOTSUPP); 728 } 729 730 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 731 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 732 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 733 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 734 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 735 736 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags); 737 mtx_unlock(&Giant); 738 return (error); 739 } 740 741 /* 742 * vfs_domount_first(): first file system mount (not update) 743 */ 744 static int 745 vfs_domount_first( 746 struct thread *td, /* Calling thread. */ 747 struct vfsconf *vfsp, /* File system type. */ 748 char *fspath, /* Mount path. */ 749 struct vnode *vp, /* Vnode to be covered. */ 750 int fsflags, /* Flags common to all filesystems. */ 751 struct vfsoptlist **optlist /* Options local to the filesystem. */ 752 ) 753 { 754 struct vattr va; 755 struct mount *mp; 756 struct vnode *newdp; 757 int error; 758 759 mtx_assert(&Giant, MA_OWNED); 760 ASSERT_VOP_ELOCKED(vp, __func__); 761 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 762 763 /* 764 * If the user is not root, ensure that they own the directory 765 * onto which we are attempting to mount. 766 */ 767 error = VOP_GETATTR(vp, &va, td->td_ucred); 768 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 769 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0); 770 if (error == 0) 771 error = vinvalbuf(vp, V_SAVE, 0, 0); 772 if (error == 0 && vp->v_type != VDIR) 773 error = ENOTDIR; 774 if (error == 0) { 775 VI_LOCK(vp); 776 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 777 vp->v_iflag |= VI_MOUNT; 778 else 779 error = EBUSY; 780 VI_UNLOCK(vp); 781 } 782 if (error != 0) { 783 vput(vp); 784 return (error); 785 } 786 VOP_UNLOCK(vp, 0); 787 788 /* Allocate and initialize the filesystem. */ 789 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 790 /* XXXMAC: pass to vfs_mount_alloc? */ 791 mp->mnt_optnew = *optlist; 792 /* Set the mount level flags. */ 793 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 794 795 /* 796 * Mount the filesystem. 797 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 798 * get. No freeing of cn_pnbuf. 799 */ 800 error = VFS_MOUNT(mp); 801 if (error != 0) { 802 vfs_unbusy(mp); 803 vfs_mount_destroy(mp); 804 VI_LOCK(vp); 805 vp->v_iflag &= ~VI_MOUNT; 806 VI_UNLOCK(vp); 807 vrele(vp); 808 return (error); 809 } 810 811 if (mp->mnt_opt != NULL) 812 vfs_freeopts(mp->mnt_opt); 813 mp->mnt_opt = mp->mnt_optnew; 814 *optlist = NULL; 815 (void)VFS_STATFS(mp, &mp->mnt_stat); 816 817 /* 818 * Prevent external consumers of mount options from reading mnt_optnew. 819 */ 820 mp->mnt_optnew = NULL; 821 822 MNT_ILOCK(mp); 823 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 824 mp->mnt_kern_flag |= MNTK_ASYNC; 825 else 826 mp->mnt_kern_flag &= ~MNTK_ASYNC; 827 MNT_IUNLOCK(mp); 828 829 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 830 cache_purge(vp); 831 VI_LOCK(vp); 832 vp->v_iflag &= ~VI_MOUNT; 833 VI_UNLOCK(vp); 834 vp->v_mountedhere = mp; 835 /* Place the new filesystem at the end of the mount list. */ 836 mtx_lock(&mountlist_mtx); 837 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 838 mtx_unlock(&mountlist_mtx); 839 vfs_event_signal(NULL, VQ_MOUNT, 0); 840 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 841 panic("mount: lost mount"); 842 VOP_UNLOCK(newdp, 0); 843 VOP_UNLOCK(vp, 0); 844 mountcheckdirs(vp, newdp); 845 vrele(newdp); 846 if ((mp->mnt_flag & MNT_RDONLY) == 0) 847 vfs_allocate_syncvnode(mp); 848 vfs_unbusy(mp); 849 return (0); 850 } 851 852 /* 853 * vfs_domount_update(): update of mounted file system 854 */ 855 static int 856 vfs_domount_update( 857 struct thread *td, /* Calling thread. */ 858 struct vnode *vp, /* Mount point vnode. */ 859 int fsflags, /* Flags common to all filesystems. */ 860 struct vfsoptlist **optlist /* Options local to the filesystem. */ 861 ) 862 { 863 struct oexport_args oexport; 864 struct export_args export; 865 struct mount *mp; 866 int error, export_error, flag; 867 868 mtx_assert(&Giant, MA_OWNED); 869 ASSERT_VOP_ELOCKED(vp, __func__); 870 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 871 872 if ((vp->v_vflag & VV_ROOT) == 0) { 873 vput(vp); 874 return (EINVAL); 875 } 876 mp = vp->v_mount; 877 /* 878 * We only allow the filesystem to be reloaded if it 879 * is currently mounted read-only. 880 */ 881 flag = mp->mnt_flag; 882 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 883 vput(vp); 884 return (EOPNOTSUPP); /* Needs translation */ 885 } 886 /* 887 * Only privileged root, or (if MNT_USER is set) the user that 888 * did the original mount is permitted to update it. 889 */ 890 error = vfs_suser(mp, td); 891 if (error != 0) { 892 vput(vp); 893 return (error); 894 } 895 if (vfs_busy(mp, MBF_NOWAIT)) { 896 vput(vp); 897 return (EBUSY); 898 } 899 VI_LOCK(vp); 900 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 901 VI_UNLOCK(vp); 902 vfs_unbusy(mp); 903 vput(vp); 904 return (EBUSY); 905 } 906 vp->v_iflag |= VI_MOUNT; 907 VI_UNLOCK(vp); 908 VOP_UNLOCK(vp, 0); 909 910 MNT_ILOCK(mp); 911 mp->mnt_flag &= ~MNT_UPDATEMASK; 912 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 913 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 914 if ((mp->mnt_flag & MNT_ASYNC) == 0) 915 mp->mnt_kern_flag &= ~MNTK_ASYNC; 916 MNT_IUNLOCK(mp); 917 mp->mnt_optnew = *optlist; 918 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 919 920 /* 921 * Mount the filesystem. 922 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 923 * get. No freeing of cn_pnbuf. 924 */ 925 error = VFS_MOUNT(mp); 926 927 export_error = 0; 928 if (error == 0) { 929 /* Process the export option. */ 930 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 931 sizeof(export)) == 0) { 932 export_error = vfs_export(mp, &export); 933 } else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 934 sizeof(oexport)) == 0) { 935 export.ex_flags = oexport.ex_flags; 936 export.ex_root = oexport.ex_root; 937 export.ex_anon = oexport.ex_anon; 938 export.ex_addr = oexport.ex_addr; 939 export.ex_addrlen = oexport.ex_addrlen; 940 export.ex_mask = oexport.ex_mask; 941 export.ex_masklen = oexport.ex_masklen; 942 export.ex_indexfile = oexport.ex_indexfile; 943 export.ex_numsecflavors = 0; 944 export_error = vfs_export(mp, &export); 945 } 946 } 947 948 MNT_ILOCK(mp); 949 if (error == 0) { 950 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 951 MNT_SNAPSHOT); 952 } else { 953 /* 954 * If we fail, restore old mount flags. MNT_QUOTA is special, 955 * because it is not part of MNT_UPDATEMASK, but it could have 956 * changed in the meantime if quotactl(2) was called. 957 * All in all we want current value of MNT_QUOTA, not the old 958 * one. 959 */ 960 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 961 } 962 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 963 mp->mnt_kern_flag |= MNTK_ASYNC; 964 else 965 mp->mnt_kern_flag &= ~MNTK_ASYNC; 966 MNT_IUNLOCK(mp); 967 968 if (error != 0) 969 goto end; 970 971 if (mp->mnt_opt != NULL) 972 vfs_freeopts(mp->mnt_opt); 973 mp->mnt_opt = mp->mnt_optnew; 974 *optlist = NULL; 975 (void)VFS_STATFS(mp, &mp->mnt_stat); 976 /* 977 * Prevent external consumers of mount options from reading 978 * mnt_optnew. 979 */ 980 mp->mnt_optnew = NULL; 981 982 if ((mp->mnt_flag & MNT_RDONLY) == 0) 983 vfs_allocate_syncvnode(mp); 984 else 985 vfs_deallocate_syncvnode(mp); 986 end: 987 vfs_unbusy(mp); 988 VI_LOCK(vp); 989 vp->v_iflag &= ~VI_MOUNT; 990 VI_UNLOCK(vp); 991 vrele(vp); 992 return (error != 0 ? error : export_error); 993 } 994 995 /* 996 * vfs_domount(): actually attempt a filesystem mount. 997 */ 998 static int 999 vfs_domount( 1000 struct thread *td, /* Calling thread. */ 1001 const char *fstype, /* Filesystem type. */ 1002 char *fspath, /* Mount path. */ 1003 int fsflags, /* Flags common to all filesystems. */ 1004 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1005 ) 1006 { 1007 struct vfsconf *vfsp; 1008 struct nameidata nd; 1009 struct vnode *vp; 1010 int error; 1011 1012 /* 1013 * Be ultra-paranoid about making sure the type and fspath 1014 * variables will fit in our mp buffers, including the 1015 * terminating NUL. 1016 */ 1017 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1018 return (ENAMETOOLONG); 1019 1020 if (jailed(td->td_ucred) || usermount == 0) { 1021 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1022 return (error); 1023 } 1024 1025 /* 1026 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1027 */ 1028 if (fsflags & MNT_EXPORTED) { 1029 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1030 if (error) 1031 return (error); 1032 } 1033 if (fsflags & MNT_SUIDDIR) { 1034 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1035 if (error) 1036 return (error); 1037 } 1038 /* 1039 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1040 */ 1041 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1042 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1043 fsflags |= MNT_NOSUID | MNT_USER; 1044 } 1045 1046 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1047 vfsp = NULL; 1048 if ((fsflags & MNT_UPDATE) == 0) { 1049 /* Don't try to load KLDs if we're mounting the root. */ 1050 if (fsflags & MNT_ROOTFS) 1051 vfsp = vfs_byname(fstype); 1052 else 1053 vfsp = vfs_byname_kld(fstype, td, &error); 1054 if (vfsp == NULL) 1055 return (ENODEV); 1056 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 1057 return (EPERM); 1058 } 1059 1060 /* 1061 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1062 */ 1063 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | MPSAFE | AUDITVNODE1, 1064 UIO_SYSSPACE, fspath, td); 1065 error = namei(&nd); 1066 if (error != 0) 1067 return (error); 1068 if (!NDHASGIANT(&nd)) 1069 mtx_lock(&Giant); 1070 NDFREE(&nd, NDF_ONLY_PNBUF); 1071 vp = nd.ni_vp; 1072 if ((fsflags & MNT_UPDATE) == 0) { 1073 error = vfs_domount_first(td, vfsp, fspath, vp, fsflags, 1074 optlist); 1075 } else { 1076 error = vfs_domount_update(td, vp, fsflags, optlist); 1077 } 1078 mtx_unlock(&Giant); 1079 1080 ASSERT_VI_UNLOCKED(vp, __func__); 1081 ASSERT_VOP_UNLOCKED(vp, __func__); 1082 1083 return (error); 1084 } 1085 1086 /* 1087 * Unmount a filesystem. 1088 * 1089 * Note: unmount takes a path to the vnode mounted on as argument, not 1090 * special file (as before). 1091 */ 1092 #ifndef _SYS_SYSPROTO_H_ 1093 struct unmount_args { 1094 char *path; 1095 int flags; 1096 }; 1097 #endif 1098 /* ARGSUSED */ 1099 int 1100 sys_unmount(td, uap) 1101 struct thread *td; 1102 register struct unmount_args /* { 1103 char *path; 1104 int flags; 1105 } */ *uap; 1106 { 1107 struct mount *mp; 1108 char *pathbuf; 1109 int error, id0, id1; 1110 1111 AUDIT_ARG_VALUE(uap->flags); 1112 if (jailed(td->td_ucred) || usermount == 0) { 1113 error = priv_check(td, PRIV_VFS_UNMOUNT); 1114 if (error) 1115 return (error); 1116 } 1117 1118 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1119 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1120 if (error) { 1121 free(pathbuf, M_TEMP); 1122 return (error); 1123 } 1124 mtx_lock(&Giant); 1125 if (uap->flags & MNT_BYFSID) { 1126 AUDIT_ARG_TEXT(pathbuf); 1127 /* Decode the filesystem ID. */ 1128 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1129 mtx_unlock(&Giant); 1130 free(pathbuf, M_TEMP); 1131 return (EINVAL); 1132 } 1133 1134 mtx_lock(&mountlist_mtx); 1135 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1136 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1137 mp->mnt_stat.f_fsid.val[1] == id1) 1138 break; 1139 } 1140 mtx_unlock(&mountlist_mtx); 1141 } else { 1142 AUDIT_ARG_UPATH1(td, pathbuf); 1143 mtx_lock(&mountlist_mtx); 1144 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1145 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1146 break; 1147 } 1148 mtx_unlock(&mountlist_mtx); 1149 } 1150 free(pathbuf, M_TEMP); 1151 if (mp == NULL) { 1152 /* 1153 * Previously we returned ENOENT for a nonexistent path and 1154 * EINVAL for a non-mountpoint. We cannot tell these apart 1155 * now, so in the !MNT_BYFSID case return the more likely 1156 * EINVAL for compatibility. 1157 */ 1158 mtx_unlock(&Giant); 1159 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1160 } 1161 1162 /* 1163 * Don't allow unmounting the root filesystem. 1164 */ 1165 if (mp->mnt_flag & MNT_ROOTFS) { 1166 mtx_unlock(&Giant); 1167 return (EINVAL); 1168 } 1169 error = dounmount(mp, uap->flags, td); 1170 mtx_unlock(&Giant); 1171 return (error); 1172 } 1173 1174 /* 1175 * Do the actual filesystem unmount. 1176 */ 1177 int 1178 dounmount(mp, flags, td) 1179 struct mount *mp; 1180 int flags; 1181 struct thread *td; 1182 { 1183 struct vnode *coveredvp, *fsrootvp; 1184 int error; 1185 int async_flag; 1186 int mnt_gen_r; 1187 1188 mtx_assert(&Giant, MA_OWNED); 1189 1190 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1191 mnt_gen_r = mp->mnt_gen; 1192 VI_LOCK(coveredvp); 1193 vholdl(coveredvp); 1194 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1195 vdrop(coveredvp); 1196 /* 1197 * Check for mp being unmounted while waiting for the 1198 * covered vnode lock. 1199 */ 1200 if (coveredvp->v_mountedhere != mp || 1201 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1202 VOP_UNLOCK(coveredvp, 0); 1203 return (EBUSY); 1204 } 1205 } 1206 /* 1207 * Only privileged root, or (if MNT_USER is set) the user that did the 1208 * original mount is permitted to unmount this filesystem. 1209 */ 1210 error = vfs_suser(mp, td); 1211 if (error) { 1212 if (coveredvp) 1213 VOP_UNLOCK(coveredvp, 0); 1214 return (error); 1215 } 1216 1217 MNT_ILOCK(mp); 1218 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1219 MNT_IUNLOCK(mp); 1220 if (coveredvp) 1221 VOP_UNLOCK(coveredvp, 0); 1222 return (EBUSY); 1223 } 1224 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1225 /* Allow filesystems to detect that a forced unmount is in progress. */ 1226 if (flags & MNT_FORCE) 1227 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1228 error = 0; 1229 if (mp->mnt_lockref) { 1230 if ((flags & MNT_FORCE) == 0) { 1231 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1232 MNTK_UNMOUNTF); 1233 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1234 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1235 wakeup(mp); 1236 } 1237 MNT_IUNLOCK(mp); 1238 if (coveredvp) 1239 VOP_UNLOCK(coveredvp, 0); 1240 return (EBUSY); 1241 } 1242 mp->mnt_kern_flag |= MNTK_DRAINING; 1243 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1244 "mount drain", 0); 1245 } 1246 MNT_IUNLOCK(mp); 1247 KASSERT(mp->mnt_lockref == 0, 1248 ("%s: invalid lock refcount in the drain path @ %s:%d", 1249 __func__, __FILE__, __LINE__)); 1250 KASSERT(error == 0, 1251 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1252 __func__, __FILE__, __LINE__)); 1253 vn_start_write(NULL, &mp, V_WAIT); 1254 1255 if (mp->mnt_flag & MNT_EXPUBLIC) 1256 vfs_setpublicfs(NULL, NULL, NULL); 1257 1258 vfs_msync(mp, MNT_WAIT); 1259 MNT_ILOCK(mp); 1260 async_flag = mp->mnt_flag & MNT_ASYNC; 1261 mp->mnt_flag &= ~MNT_ASYNC; 1262 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1263 MNT_IUNLOCK(mp); 1264 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1265 vfs_deallocate_syncvnode(mp); 1266 /* 1267 * For forced unmounts, move process cdir/rdir refs on the fs root 1268 * vnode to the covered vnode. For non-forced unmounts we want 1269 * such references to cause an EBUSY error. 1270 */ 1271 if ((flags & MNT_FORCE) && 1272 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1273 if (mp->mnt_vnodecovered != NULL) 1274 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1275 if (fsrootvp == rootvnode) { 1276 vrele(rootvnode); 1277 rootvnode = NULL; 1278 } 1279 vput(fsrootvp); 1280 } 1281 if (((mp->mnt_flag & MNT_RDONLY) || 1282 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1283 error = VFS_UNMOUNT(mp, flags); 1284 vn_finished_write(mp); 1285 /* 1286 * If we failed to flush the dirty blocks for this mount point, 1287 * undo all the cdir/rdir and rootvnode changes we made above. 1288 * Unless we failed to do so because the device is reporting that 1289 * it doesn't exist anymore. 1290 */ 1291 if (error && error != ENXIO) { 1292 if ((flags & MNT_FORCE) && 1293 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1294 if (mp->mnt_vnodecovered != NULL) 1295 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1296 if (rootvnode == NULL) { 1297 rootvnode = fsrootvp; 1298 vref(rootvnode); 1299 } 1300 vput(fsrootvp); 1301 } 1302 MNT_ILOCK(mp); 1303 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1304 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1305 MNT_IUNLOCK(mp); 1306 vfs_allocate_syncvnode(mp); 1307 MNT_ILOCK(mp); 1308 } 1309 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1310 mp->mnt_flag |= async_flag; 1311 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1312 mp->mnt_kern_flag |= MNTK_ASYNC; 1313 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1314 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1315 wakeup(mp); 1316 } 1317 MNT_IUNLOCK(mp); 1318 if (coveredvp) 1319 VOP_UNLOCK(coveredvp, 0); 1320 return (error); 1321 } 1322 mtx_lock(&mountlist_mtx); 1323 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1324 mtx_unlock(&mountlist_mtx); 1325 if (coveredvp != NULL) { 1326 coveredvp->v_mountedhere = NULL; 1327 vput(coveredvp); 1328 } 1329 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1330 vfs_mount_destroy(mp); 1331 return (0); 1332 } 1333 1334 /* 1335 * Report errors during filesystem mounting. 1336 */ 1337 void 1338 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1339 { 1340 struct vfsoptlist *moptlist = mp->mnt_optnew; 1341 va_list ap; 1342 int error, len; 1343 char *errmsg; 1344 1345 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1346 if (error || errmsg == NULL || len <= 0) 1347 return; 1348 1349 va_start(ap, fmt); 1350 vsnprintf(errmsg, (size_t)len, fmt, ap); 1351 va_end(ap); 1352 } 1353 1354 void 1355 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1356 { 1357 va_list ap; 1358 int error, len; 1359 char *errmsg; 1360 1361 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1362 if (error || errmsg == NULL || len <= 0) 1363 return; 1364 1365 va_start(ap, fmt); 1366 vsnprintf(errmsg, (size_t)len, fmt, ap); 1367 va_end(ap); 1368 } 1369 1370 /* 1371 * --------------------------------------------------------------------- 1372 * Functions for querying mount options/arguments from filesystems. 1373 */ 1374 1375 /* 1376 * Check that no unknown options are given 1377 */ 1378 int 1379 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1380 { 1381 struct vfsopt *opt; 1382 char errmsg[255]; 1383 const char **t, *p, *q; 1384 int ret = 0; 1385 1386 TAILQ_FOREACH(opt, opts, link) { 1387 p = opt->name; 1388 q = NULL; 1389 if (p[0] == 'n' && p[1] == 'o') 1390 q = p + 2; 1391 for(t = global_opts; *t != NULL; t++) { 1392 if (strcmp(*t, p) == 0) 1393 break; 1394 if (q != NULL) { 1395 if (strcmp(*t, q) == 0) 1396 break; 1397 } 1398 } 1399 if (*t != NULL) 1400 continue; 1401 for(t = legal; *t != NULL; t++) { 1402 if (strcmp(*t, p) == 0) 1403 break; 1404 if (q != NULL) { 1405 if (strcmp(*t, q) == 0) 1406 break; 1407 } 1408 } 1409 if (*t != NULL) 1410 continue; 1411 snprintf(errmsg, sizeof(errmsg), 1412 "mount option <%s> is unknown", p); 1413 ret = EINVAL; 1414 } 1415 if (ret != 0) { 1416 TAILQ_FOREACH(opt, opts, link) { 1417 if (strcmp(opt->name, "errmsg") == 0) { 1418 strncpy((char *)opt->value, errmsg, opt->len); 1419 break; 1420 } 1421 } 1422 if (opt == NULL) 1423 printf("%s\n", errmsg); 1424 } 1425 return (ret); 1426 } 1427 1428 /* 1429 * Get a mount option by its name. 1430 * 1431 * Return 0 if the option was found, ENOENT otherwise. 1432 * If len is non-NULL it will be filled with the length 1433 * of the option. If buf is non-NULL, it will be filled 1434 * with the address of the option. 1435 */ 1436 int 1437 vfs_getopt(opts, name, buf, len) 1438 struct vfsoptlist *opts; 1439 const char *name; 1440 void **buf; 1441 int *len; 1442 { 1443 struct vfsopt *opt; 1444 1445 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1446 1447 TAILQ_FOREACH(opt, opts, link) { 1448 if (strcmp(name, opt->name) == 0) { 1449 opt->seen = 1; 1450 if (len != NULL) 1451 *len = opt->len; 1452 if (buf != NULL) 1453 *buf = opt->value; 1454 return (0); 1455 } 1456 } 1457 return (ENOENT); 1458 } 1459 1460 int 1461 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1462 { 1463 struct vfsopt *opt; 1464 1465 if (opts == NULL) 1466 return (-1); 1467 1468 TAILQ_FOREACH(opt, opts, link) { 1469 if (strcmp(name, opt->name) == 0) { 1470 opt->seen = 1; 1471 return (opt->pos); 1472 } 1473 } 1474 return (-1); 1475 } 1476 1477 char * 1478 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1479 { 1480 struct vfsopt *opt; 1481 1482 *error = 0; 1483 TAILQ_FOREACH(opt, opts, link) { 1484 if (strcmp(name, opt->name) != 0) 1485 continue; 1486 opt->seen = 1; 1487 if (opt->len == 0 || 1488 ((char *)opt->value)[opt->len - 1] != '\0') { 1489 *error = EINVAL; 1490 return (NULL); 1491 } 1492 return (opt->value); 1493 } 1494 *error = ENOENT; 1495 return (NULL); 1496 } 1497 1498 int 1499 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, 1500 uint64_t val) 1501 { 1502 struct vfsopt *opt; 1503 1504 TAILQ_FOREACH(opt, opts, link) { 1505 if (strcmp(name, opt->name) == 0) { 1506 opt->seen = 1; 1507 if (w != NULL) 1508 *w |= val; 1509 return (1); 1510 } 1511 } 1512 if (w != NULL) 1513 *w &= ~val; 1514 return (0); 1515 } 1516 1517 int 1518 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1519 { 1520 va_list ap; 1521 struct vfsopt *opt; 1522 int ret; 1523 1524 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1525 1526 TAILQ_FOREACH(opt, opts, link) { 1527 if (strcmp(name, opt->name) != 0) 1528 continue; 1529 opt->seen = 1; 1530 if (opt->len == 0 || opt->value == NULL) 1531 return (0); 1532 if (((char *)opt->value)[opt->len - 1] != '\0') 1533 return (0); 1534 va_start(ap, fmt); 1535 ret = vsscanf(opt->value, fmt, ap); 1536 va_end(ap); 1537 return (ret); 1538 } 1539 return (0); 1540 } 1541 1542 int 1543 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 1544 { 1545 struct vfsopt *opt; 1546 1547 TAILQ_FOREACH(opt, opts, link) { 1548 if (strcmp(name, opt->name) != 0) 1549 continue; 1550 opt->seen = 1; 1551 if (opt->value == NULL) 1552 opt->len = len; 1553 else { 1554 if (opt->len != len) 1555 return (EINVAL); 1556 bcopy(value, opt->value, len); 1557 } 1558 return (0); 1559 } 1560 return (ENOENT); 1561 } 1562 1563 int 1564 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 1565 { 1566 struct vfsopt *opt; 1567 1568 TAILQ_FOREACH(opt, opts, link) { 1569 if (strcmp(name, opt->name) != 0) 1570 continue; 1571 opt->seen = 1; 1572 if (opt->value == NULL) 1573 opt->len = len; 1574 else { 1575 if (opt->len < len) 1576 return (EINVAL); 1577 opt->len = len; 1578 bcopy(value, opt->value, len); 1579 } 1580 return (0); 1581 } 1582 return (ENOENT); 1583 } 1584 1585 int 1586 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 1587 { 1588 struct vfsopt *opt; 1589 1590 TAILQ_FOREACH(opt, opts, link) { 1591 if (strcmp(name, opt->name) != 0) 1592 continue; 1593 opt->seen = 1; 1594 if (opt->value == NULL) 1595 opt->len = strlen(value) + 1; 1596 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 1597 return (EINVAL); 1598 return (0); 1599 } 1600 return (ENOENT); 1601 } 1602 1603 /* 1604 * Find and copy a mount option. 1605 * 1606 * The size of the buffer has to be specified 1607 * in len, if it is not the same length as the 1608 * mount option, EINVAL is returned. 1609 * Returns ENOENT if the option is not found. 1610 */ 1611 int 1612 vfs_copyopt(opts, name, dest, len) 1613 struct vfsoptlist *opts; 1614 const char *name; 1615 void *dest; 1616 int len; 1617 { 1618 struct vfsopt *opt; 1619 1620 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1621 1622 TAILQ_FOREACH(opt, opts, link) { 1623 if (strcmp(name, opt->name) == 0) { 1624 opt->seen = 1; 1625 if (len != opt->len) 1626 return (EINVAL); 1627 bcopy(opt->value, dest, opt->len); 1628 return (0); 1629 } 1630 } 1631 return (ENOENT); 1632 } 1633 1634 /* 1635 * This is a helper function for filesystems to traverse their 1636 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1637 */ 1638 1639 struct vnode * 1640 __mnt_vnode_next(struct vnode **mvp, struct mount *mp) 1641 { 1642 struct vnode *vp; 1643 1644 mtx_assert(MNT_MTX(mp), MA_OWNED); 1645 1646 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1647 if (should_yield()) { 1648 MNT_IUNLOCK(mp); 1649 kern_yield(PRI_UNCHANGED); 1650 MNT_ILOCK(mp); 1651 } 1652 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 1653 while (vp != NULL && vp->v_type == VMARKER) 1654 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1655 1656 /* Check if we are done */ 1657 if (vp == NULL) { 1658 __mnt_vnode_markerfree(mvp, mp); 1659 return (NULL); 1660 } 1661 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1662 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1663 return (vp); 1664 } 1665 1666 struct vnode * 1667 __mnt_vnode_first(struct vnode **mvp, struct mount *mp) 1668 { 1669 struct vnode *vp; 1670 1671 mtx_assert(MNT_MTX(mp), MA_OWNED); 1672 1673 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1674 while (vp != NULL && vp->v_type == VMARKER) 1675 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1676 1677 /* Check if we are done */ 1678 if (vp == NULL) { 1679 *mvp = NULL; 1680 return (NULL); 1681 } 1682 MNT_REF(mp); 1683 MNT_IUNLOCK(mp); 1684 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 1685 M_VNODE_MARKER, 1686 M_WAITOK | M_ZERO); 1687 MNT_ILOCK(mp); 1688 (*mvp)->v_type = VMARKER; 1689 1690 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1691 while (vp != NULL && vp->v_type == VMARKER) 1692 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1693 1694 /* Check if we are done */ 1695 if (vp == NULL) { 1696 MNT_IUNLOCK(mp); 1697 free(*mvp, M_VNODE_MARKER); 1698 MNT_ILOCK(mp); 1699 *mvp = NULL; 1700 MNT_REL(mp); 1701 return (NULL); 1702 } 1703 (*mvp)->v_mount = mp; 1704 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1705 return (vp); 1706 } 1707 1708 1709 void 1710 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 1711 { 1712 1713 if (*mvp == NULL) 1714 return; 1715 1716 mtx_assert(MNT_MTX(mp), MA_OWNED); 1717 1718 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1719 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1720 MNT_IUNLOCK(mp); 1721 free(*mvp, M_VNODE_MARKER); 1722 MNT_ILOCK(mp); 1723 *mvp = NULL; 1724 MNT_REL(mp); 1725 } 1726 1727 1728 int 1729 __vfs_statfs(struct mount *mp, struct statfs *sbp) 1730 { 1731 int error; 1732 1733 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 1734 if (sbp != &mp->mnt_stat) 1735 *sbp = mp->mnt_stat; 1736 return (error); 1737 } 1738 1739 void 1740 vfs_mountedfrom(struct mount *mp, const char *from) 1741 { 1742 1743 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 1744 strlcpy(mp->mnt_stat.f_mntfromname, from, 1745 sizeof mp->mnt_stat.f_mntfromname); 1746 } 1747 1748 /* 1749 * --------------------------------------------------------------------- 1750 * This is the api for building mount args and mounting filesystems from 1751 * inside the kernel. 1752 * 1753 * The API works by accumulation of individual args. First error is 1754 * latched. 1755 * 1756 * XXX: should be documented in new manpage kernel_mount(9) 1757 */ 1758 1759 /* A memory allocation which must be freed when we are done */ 1760 struct mntaarg { 1761 SLIST_ENTRY(mntaarg) next; 1762 }; 1763 1764 /* The header for the mount arguments */ 1765 struct mntarg { 1766 struct iovec *v; 1767 int len; 1768 int error; 1769 SLIST_HEAD(, mntaarg) list; 1770 }; 1771 1772 /* 1773 * Add a boolean argument. 1774 * 1775 * flag is the boolean value. 1776 * name must start with "no". 1777 */ 1778 struct mntarg * 1779 mount_argb(struct mntarg *ma, int flag, const char *name) 1780 { 1781 1782 KASSERT(name[0] == 'n' && name[1] == 'o', 1783 ("mount_argb(...,%s): name must start with 'no'", name)); 1784 1785 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 1786 } 1787 1788 /* 1789 * Add an argument printf style 1790 */ 1791 struct mntarg * 1792 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 1793 { 1794 va_list ap; 1795 struct mntaarg *maa; 1796 struct sbuf *sb; 1797 int len; 1798 1799 if (ma == NULL) { 1800 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1801 SLIST_INIT(&ma->list); 1802 } 1803 if (ma->error) 1804 return (ma); 1805 1806 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1807 M_MOUNT, M_WAITOK); 1808 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1809 ma->v[ma->len].iov_len = strlen(name) + 1; 1810 ma->len++; 1811 1812 sb = sbuf_new_auto(); 1813 va_start(ap, fmt); 1814 sbuf_vprintf(sb, fmt, ap); 1815 va_end(ap); 1816 sbuf_finish(sb); 1817 len = sbuf_len(sb) + 1; 1818 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1819 SLIST_INSERT_HEAD(&ma->list, maa, next); 1820 bcopy(sbuf_data(sb), maa + 1, len); 1821 sbuf_delete(sb); 1822 1823 ma->v[ma->len].iov_base = maa + 1; 1824 ma->v[ma->len].iov_len = len; 1825 ma->len++; 1826 1827 return (ma); 1828 } 1829 1830 /* 1831 * Add an argument which is a userland string. 1832 */ 1833 struct mntarg * 1834 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 1835 { 1836 struct mntaarg *maa; 1837 char *tbuf; 1838 1839 if (val == NULL) 1840 return (ma); 1841 if (ma == NULL) { 1842 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1843 SLIST_INIT(&ma->list); 1844 } 1845 if (ma->error) 1846 return (ma); 1847 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1848 SLIST_INSERT_HEAD(&ma->list, maa, next); 1849 tbuf = (void *)(maa + 1); 1850 ma->error = copyinstr(val, tbuf, len, NULL); 1851 return (mount_arg(ma, name, tbuf, -1)); 1852 } 1853 1854 /* 1855 * Plain argument. 1856 * 1857 * If length is -1, treat value as a C string. 1858 */ 1859 struct mntarg * 1860 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 1861 { 1862 1863 if (ma == NULL) { 1864 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1865 SLIST_INIT(&ma->list); 1866 } 1867 if (ma->error) 1868 return (ma); 1869 1870 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1871 M_MOUNT, M_WAITOK); 1872 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1873 ma->v[ma->len].iov_len = strlen(name) + 1; 1874 ma->len++; 1875 1876 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 1877 if (len < 0) 1878 ma->v[ma->len].iov_len = strlen(val) + 1; 1879 else 1880 ma->v[ma->len].iov_len = len; 1881 ma->len++; 1882 return (ma); 1883 } 1884 1885 /* 1886 * Free a mntarg structure 1887 */ 1888 static void 1889 free_mntarg(struct mntarg *ma) 1890 { 1891 struct mntaarg *maa; 1892 1893 while (!SLIST_EMPTY(&ma->list)) { 1894 maa = SLIST_FIRST(&ma->list); 1895 SLIST_REMOVE_HEAD(&ma->list, next); 1896 free(maa, M_MOUNT); 1897 } 1898 free(ma->v, M_MOUNT); 1899 free(ma, M_MOUNT); 1900 } 1901 1902 /* 1903 * Mount a filesystem 1904 */ 1905 int 1906 kernel_mount(struct mntarg *ma, int flags) 1907 { 1908 struct uio auio; 1909 int error; 1910 1911 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 1912 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 1913 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 1914 1915 auio.uio_iov = ma->v; 1916 auio.uio_iovcnt = ma->len; 1917 auio.uio_segflg = UIO_SYSSPACE; 1918 1919 error = ma->error; 1920 if (!error) 1921 error = vfs_donmount(curthread, flags, &auio); 1922 free_mntarg(ma); 1923 return (error); 1924 } 1925 1926 /* 1927 * A printflike function to mount a filesystem. 1928 */ 1929 int 1930 kernel_vmount(int flags, ...) 1931 { 1932 struct mntarg *ma = NULL; 1933 va_list ap; 1934 const char *cp; 1935 const void *vp; 1936 int error; 1937 1938 va_start(ap, flags); 1939 for (;;) { 1940 cp = va_arg(ap, const char *); 1941 if (cp == NULL) 1942 break; 1943 vp = va_arg(ap, const void *); 1944 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 1945 } 1946 va_end(ap); 1947 1948 error = kernel_mount(ma, flags); 1949 return (error); 1950 } 1951 1952 void 1953 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp) 1954 { 1955 1956 bcopy(oexp, exp, sizeof(*oexp)); 1957 exp->ex_numsecflavors = 0; 1958 } 1959