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 73static int vfs_domount(struct thread *td, const char *fstype, char *fspath, 74 uint64_t fsflags, struct vfsoptlist **optlist); 75static void free_mntarg(struct mntarg *ma); 76 77static int usermount = 0; 78SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 79 "Unprivileged users may mount and unmount file systems"); 80 81MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 82static uma_zone_t mount_zone; 83 84/* List of mounted filesystems. */ 85struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 86 87/* For any iteration/modification of mountlist */ 88struct mtx mountlist_mtx; 89MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 90 91/* 92 * Global opts, taken by all filesystems 93 */ 94static const char *global_opts[] = { 95 "errmsg", 96 "fstype", 97 "fspath", 98 "ro", 99 "rw", 100 "nosuid", 101 "noexec", 102 NULL 103}; 104 105static int 106mount_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 116static void 117mount_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 126static void 127vfs_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} 133SYSINIT(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. */ 141static void 142vfs_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. */ 153void 154vfs_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 165void 166vfs_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 178static int 179vfs_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 188static int 189vfs_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 */ 200static int 201vfs_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 */ 237static void 238vfs_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 */ 259int 260vfs_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); 330bad: 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 */ 343static void 344vfs_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 */ 366int 367sys_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 428void 429vfs_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 438void 439vfs_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 */ 451struct mount * 452vfs_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 */ 489void 490vfs_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 533int 534vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions) 535{ 536 struct vfsoptlist *optlist; 537 struct vfsopt *opt, *tmp_opt; 538 char *fstype, *fspath, *errmsg; 539 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 540 541 errmsg = fspath = NULL; 542 errmsg_len = fspathlen = 0; 543 errmsg_pos = -1; 544 545 error = vfs_buildopts(fsoptions, &optlist); 546 if (error) 547 return (error); 548 549 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 550 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 551 552 /* 553 * We need these two options before the others, 554 * and they are mandatory for any filesystem. 555 * Ensure they are NUL terminated as well. 556 */ 557 fstypelen = 0; 558 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 559 if (error || fstype[fstypelen - 1] != '\0') { 560 error = EINVAL; 561 if (errmsg != NULL) 562 strncpy(errmsg, "Invalid fstype", errmsg_len); 563 goto bail; 564 } 565 fspathlen = 0; 566 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 567 if (error || fspath[fspathlen - 1] != '\0') { 568 error = EINVAL; 569 if (errmsg != NULL) 570 strncpy(errmsg, "Invalid fspath", errmsg_len); 571 goto bail; 572 } 573 574 /* 575 * We need to see if we have the "update" option 576 * before we call vfs_domount(), since vfs_domount() has special 577 * logic based on MNT_UPDATE. This is very important 578 * when we want to update the root filesystem. 579 */ 580 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 581 if (strcmp(opt->name, "update") == 0) { 582 fsflags |= MNT_UPDATE; 583 vfs_freeopt(optlist, opt); 584 } 585 else if (strcmp(opt->name, "async") == 0) 586 fsflags |= MNT_ASYNC; 587 else if (strcmp(opt->name, "force") == 0) { 588 fsflags |= MNT_FORCE; 589 vfs_freeopt(optlist, opt); 590 } 591 else if (strcmp(opt->name, "reload") == 0) { 592 fsflags |= MNT_RELOAD; 593 vfs_freeopt(optlist, opt); 594 } 595 else if (strcmp(opt->name, "multilabel") == 0) 596 fsflags |= MNT_MULTILABEL; 597 else if (strcmp(opt->name, "noasync") == 0) 598 fsflags &= ~MNT_ASYNC; 599 else if (strcmp(opt->name, "noatime") == 0) 600 fsflags |= MNT_NOATIME; 601 else if (strcmp(opt->name, "atime") == 0) { 602 free(opt->name, M_MOUNT); 603 opt->name = strdup("nonoatime", M_MOUNT); 604 } 605 else if (strcmp(opt->name, "noclusterr") == 0) 606 fsflags |= MNT_NOCLUSTERR; 607 else if (strcmp(opt->name, "clusterr") == 0) { 608 free(opt->name, M_MOUNT); 609 opt->name = strdup("nonoclusterr", M_MOUNT); 610 } 611 else if (strcmp(opt->name, "noclusterw") == 0) 612 fsflags |= MNT_NOCLUSTERW; 613 else if (strcmp(opt->name, "clusterw") == 0) { 614 free(opt->name, M_MOUNT); 615 opt->name = strdup("nonoclusterw", M_MOUNT); 616 } 617 else if (strcmp(opt->name, "noexec") == 0) 618 fsflags |= MNT_NOEXEC; 619 else if (strcmp(opt->name, "exec") == 0) { 620 free(opt->name, M_MOUNT); 621 opt->name = strdup("nonoexec", M_MOUNT); 622 } 623 else if (strcmp(opt->name, "nosuid") == 0) 624 fsflags |= MNT_NOSUID; 625 else if (strcmp(opt->name, "suid") == 0) { 626 free(opt->name, M_MOUNT); 627 opt->name = strdup("nonosuid", M_MOUNT); 628 } 629 else if (strcmp(opt->name, "nosymfollow") == 0) 630 fsflags |= MNT_NOSYMFOLLOW; 631 else if (strcmp(opt->name, "symfollow") == 0) { 632 free(opt->name, M_MOUNT); 633 opt->name = strdup("nonosymfollow", M_MOUNT); 634 } 635 else if (strcmp(opt->name, "noro") == 0) 636 fsflags &= ~MNT_RDONLY; 637 else if (strcmp(opt->name, "rw") == 0) 638 fsflags &= ~MNT_RDONLY; 639 else if (strcmp(opt->name, "ro") == 0) 640 fsflags |= MNT_RDONLY; 641 else if (strcmp(opt->name, "rdonly") == 0) { 642 free(opt->name, M_MOUNT); 643 opt->name = strdup("ro", M_MOUNT); 644 fsflags |= MNT_RDONLY; 645 } 646 else if (strcmp(opt->name, "suiddir") == 0) 647 fsflags |= MNT_SUIDDIR; 648 else if (strcmp(opt->name, "sync") == 0) 649 fsflags |= MNT_SYNCHRONOUS; 650 else if (strcmp(opt->name, "union") == 0) 651 fsflags |= MNT_UNION; 652 else if (strcmp(opt->name, "automounted") == 0) { 653 fsflags |= MNT_AUTOMOUNTED; 654 vfs_freeopt(optlist, opt); 655 } 656 } 657 658 /* 659 * Be ultra-paranoid about making sure the type and fspath 660 * variables will fit in our mp buffers, including the 661 * terminating NUL. 662 */ 663 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) { 664 error = ENAMETOOLONG; 665 goto bail; 666 } 667 668 error = vfs_domount(td, fstype, fspath, fsflags, &optlist); 669bail: 670 /* copyout the errmsg */ 671 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 672 && errmsg_len > 0 && errmsg != NULL) { 673 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 674 bcopy(errmsg, 675 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 676 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 677 } else { 678 copyout(errmsg, 679 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 680 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 681 } 682 } 683 684 if (optlist != NULL) 685 vfs_freeopts(optlist); 686 return (error); 687} 688 689/* 690 * Old mount API. 691 */ 692#ifndef _SYS_SYSPROTO_H_ 693struct mount_args { 694 char *type; 695 char *path; 696 int flags; 697 caddr_t data; 698}; 699#endif 700/* ARGSUSED */ 701int 702sys_mount(td, uap) 703 struct thread *td; 704 struct mount_args /* { 705 char *type; 706 char *path; 707 int flags; 708 caddr_t data; 709 } */ *uap; 710{ 711 char *fstype; 712 struct vfsconf *vfsp = NULL; 713 struct mntarg *ma = NULL; 714 uint64_t flags; 715 int error; 716 717 /* 718 * Mount flags are now 64-bits. On 32-bit architectures only 719 * 32-bits are passed in, but from here on everything handles 720 * 64-bit flags correctly. 721 */ 722 flags = uap->flags; 723 724 AUDIT_ARG_FFLAGS(flags); 725 726 /* 727 * Filter out MNT_ROOTFS. We do not want clients of mount() in 728 * userspace to set this flag, but we must filter it out if we want 729 * MNT_UPDATE on the root file system to work. 730 * MNT_ROOTFS should only be set by the kernel when mounting its 731 * root file system. 732 */ 733 flags &= ~MNT_ROOTFS; 734 735 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 736 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 737 if (error) { 738 free(fstype, M_TEMP); 739 return (error); 740 } 741 742 AUDIT_ARG_TEXT(fstype); 743 vfsp = vfs_byname_kld(fstype, td, &error); 744 free(fstype, M_TEMP); 745 if (vfsp == NULL) 746 return (ENOENT); 747 if (vfsp->vfc_vfsops->vfs_cmount == NULL) 748 return (EOPNOTSUPP); 749 750 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN); 751 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 752 ma = mount_argb(ma, flags & MNT_RDONLY, "noro"); 753 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid"); 754 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec"); 755 756 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags); 757 return (error); 758} 759 760/* 761 * vfs_domount_first(): first file system mount (not update) 762 */ 763static int 764vfs_domount_first( 765 struct thread *td, /* Calling thread. */ 766 struct vfsconf *vfsp, /* File system type. */ 767 char *fspath, /* Mount path. */ 768 struct vnode *vp, /* Vnode to be covered. */ 769 uint64_t fsflags, /* Flags common to all filesystems. */ 770 struct vfsoptlist **optlist /* Options local to the filesystem. */ 771 ) 772{ 773 struct vattr va; 774 struct mount *mp; 775 struct vnode *newdp; 776 int error; 777 778 ASSERT_VOP_ELOCKED(vp, __func__); 779 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 780 781 /* 782 * If the user is not root, ensure that they own the directory 783 * onto which we are attempting to mount. 784 */ 785 error = VOP_GETATTR(vp, &va, td->td_ucred); 786 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 787 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0); 788 if (error == 0) 789 error = vinvalbuf(vp, V_SAVE, 0, 0); 790 if (error == 0 && vp->v_type != VDIR) 791 error = ENOTDIR; 792 if (error == 0) { 793 VI_LOCK(vp); 794 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 795 vp->v_iflag |= VI_MOUNT; 796 else 797 error = EBUSY; 798 VI_UNLOCK(vp); 799 } 800 if (error != 0) { 801 vput(vp); 802 return (error); 803 } 804 VOP_UNLOCK(vp, 0); 805 806 /* Allocate and initialize the filesystem. */ 807 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 808 /* XXXMAC: pass to vfs_mount_alloc? */ 809 mp->mnt_optnew = *optlist; 810 /* Set the mount level flags. */ 811 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 812 813 /* 814 * Mount the filesystem. 815 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 816 * get. No freeing of cn_pnbuf. 817 */ 818 error = VFS_MOUNT(mp); 819 if (error != 0) { 820 vfs_unbusy(mp); 821 vfs_mount_destroy(mp); 822 VI_LOCK(vp); 823 vp->v_iflag &= ~VI_MOUNT; 824 VI_UNLOCK(vp); 825 vrele(vp); 826 return (error); 827 } 828 829 if (mp->mnt_opt != NULL) 830 vfs_freeopts(mp->mnt_opt); 831 mp->mnt_opt = mp->mnt_optnew; 832 *optlist = NULL; 833 (void)VFS_STATFS(mp, &mp->mnt_stat); 834 835 /* 836 * Prevent external consumers of mount options from reading mnt_optnew. 837 */ 838 mp->mnt_optnew = NULL; 839 840 MNT_ILOCK(mp); 841 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 842 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 843 mp->mnt_kern_flag |= MNTK_ASYNC; 844 else 845 mp->mnt_kern_flag &= ~MNTK_ASYNC; 846 MNT_IUNLOCK(mp); 847 848 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 849 cache_purge(vp); 850 VI_LOCK(vp); 851 vp->v_iflag &= ~VI_MOUNT; 852 VI_UNLOCK(vp); 853 vp->v_mountedhere = mp; 854 /* Place the new filesystem at the end of the mount list. */ 855 mtx_lock(&mountlist_mtx); 856 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 857 mtx_unlock(&mountlist_mtx); 858 vfs_event_signal(NULL, VQ_MOUNT, 0); 859 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 860 panic("mount: lost mount"); 861 VOP_UNLOCK(vp, 0); 862 EVENTHANDLER_INVOKE(vfs_mounted, mp, newdp, td); 863 VOP_UNLOCK(newdp, 0); 864 mountcheckdirs(vp, newdp); 865 vrele(newdp); 866 if ((mp->mnt_flag & MNT_RDONLY) == 0) 867 vfs_allocate_syncvnode(mp); 868 vfs_unbusy(mp); 869 return (0); 870} 871 872/* 873 * vfs_domount_update(): update of mounted file system 874 */ 875static int 876vfs_domount_update( 877 struct thread *td, /* Calling thread. */ 878 struct vnode *vp, /* Mount point vnode. */ 879 uint64_t fsflags, /* Flags common to all filesystems. */ 880 struct vfsoptlist **optlist /* Options local to the filesystem. */ 881 ) 882{ 883 struct oexport_args oexport; 884 struct export_args export; 885 struct mount *mp; 886 int error, export_error; 887 uint64_t flag; 888 889 ASSERT_VOP_ELOCKED(vp, __func__); 890 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 891 892 if ((vp->v_vflag & VV_ROOT) == 0) { 893 vput(vp); 894 return (EINVAL); 895 } 896 mp = vp->v_mount; 897 /* 898 * We only allow the filesystem to be reloaded if it 899 * is currently mounted read-only. 900 */ 901 flag = mp->mnt_flag; 902 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 903 vput(vp); 904 return (EOPNOTSUPP); /* Needs translation */ 905 } 906 /* 907 * Only privileged root, or (if MNT_USER is set) the user that 908 * did the original mount is permitted to update it. 909 */ 910 error = vfs_suser(mp, td); 911 if (error != 0) { 912 vput(vp); 913 return (error); 914 } 915 if (vfs_busy(mp, MBF_NOWAIT)) { 916 vput(vp); 917 return (EBUSY); 918 } 919 VI_LOCK(vp); 920 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 921 VI_UNLOCK(vp); 922 vfs_unbusy(mp); 923 vput(vp); 924 return (EBUSY); 925 } 926 vp->v_iflag |= VI_MOUNT; 927 VI_UNLOCK(vp); 928 VOP_UNLOCK(vp, 0); 929 930 MNT_ILOCK(mp); 931 mp->mnt_flag &= ~MNT_UPDATEMASK; 932 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 933 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 934 if ((mp->mnt_flag & MNT_ASYNC) == 0) 935 mp->mnt_kern_flag &= ~MNTK_ASYNC; 936 MNT_IUNLOCK(mp); 937 mp->mnt_optnew = *optlist; 938 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 939 940 /* 941 * Mount the filesystem. 942 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 943 * get. No freeing of cn_pnbuf. 944 */ 945 error = VFS_MOUNT(mp); 946 947 export_error = 0; 948 if (error == 0) { 949 /* Process the export option. */ 950 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 951 sizeof(export)) == 0) { 952 export_error = vfs_export(mp, &export); 953 } else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 954 sizeof(oexport)) == 0) { 955 export.ex_flags = oexport.ex_flags; 956 export.ex_root = oexport.ex_root; 957 export.ex_anon = oexport.ex_anon; 958 export.ex_addr = oexport.ex_addr; 959 export.ex_addrlen = oexport.ex_addrlen; 960 export.ex_mask = oexport.ex_mask; 961 export.ex_masklen = oexport.ex_masklen; 962 export.ex_indexfile = oexport.ex_indexfile; 963 export.ex_numsecflavors = 0; 964 export_error = vfs_export(mp, &export); 965 } 966 } 967 968 MNT_ILOCK(mp); 969 if (error == 0) { 970 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 971 MNT_SNAPSHOT); 972 } else { 973 /* 974 * If we fail, restore old mount flags. MNT_QUOTA is special, 975 * because it is not part of MNT_UPDATEMASK, but it could have 976 * changed in the meantime if quotactl(2) was called. 977 * All in all we want current value of MNT_QUOTA, not the old 978 * one. 979 */ 980 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 981 } 982 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 983 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 984 mp->mnt_kern_flag |= MNTK_ASYNC; 985 else 986 mp->mnt_kern_flag &= ~MNTK_ASYNC; 987 MNT_IUNLOCK(mp); 988 989 if (error != 0) 990 goto end; 991 992 if (mp->mnt_opt != NULL) 993 vfs_freeopts(mp->mnt_opt); 994 mp->mnt_opt = mp->mnt_optnew; 995 *optlist = NULL; 996 (void)VFS_STATFS(mp, &mp->mnt_stat); 997 /* 998 * Prevent external consumers of mount options from reading 999 * mnt_optnew. 1000 */ 1001 mp->mnt_optnew = NULL; 1002 1003 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1004 vfs_allocate_syncvnode(mp); 1005 else 1006 vfs_deallocate_syncvnode(mp); 1007end: 1008 vfs_unbusy(mp); 1009 VI_LOCK(vp); 1010 vp->v_iflag &= ~VI_MOUNT; 1011 VI_UNLOCK(vp); 1012 vrele(vp); 1013 return (error != 0 ? error : export_error); 1014} 1015 1016/* 1017 * vfs_domount(): actually attempt a filesystem mount. 1018 */ 1019static int 1020vfs_domount( 1021 struct thread *td, /* Calling thread. */ 1022 const char *fstype, /* Filesystem type. */ 1023 char *fspath, /* Mount path. */ 1024 uint64_t fsflags, /* Flags common to all filesystems. */ 1025 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1026 ) 1027{ 1028 struct vfsconf *vfsp; 1029 struct nameidata nd; 1030 struct vnode *vp; 1031 char *pathbuf; 1032 int error; 1033 1034 /* 1035 * Be ultra-paranoid about making sure the type and fspath 1036 * variables will fit in our mp buffers, including the 1037 * terminating NUL. 1038 */ 1039 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1040 return (ENAMETOOLONG); 1041 1042 if (jailed(td->td_ucred) || usermount == 0) { 1043 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1044 return (error); 1045 } 1046 1047 /* 1048 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1049 */ 1050 if (fsflags & MNT_EXPORTED) { 1051 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1052 if (error) 1053 return (error); 1054 } 1055 if (fsflags & MNT_SUIDDIR) { 1056 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1057 if (error) 1058 return (error); 1059 } 1060 /* 1061 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1062 */ 1063 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1064 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1065 fsflags |= MNT_NOSUID | MNT_USER; 1066 } 1067 1068 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1069 vfsp = NULL; 1070 if ((fsflags & MNT_UPDATE) == 0) { 1071 /* Don't try to load KLDs if we're mounting the root. */ 1072 if (fsflags & MNT_ROOTFS) 1073 vfsp = vfs_byname(fstype); 1074 else 1075 vfsp = vfs_byname_kld(fstype, td, &error); 1076 if (vfsp == NULL) 1077 return (ENODEV); 1078 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 1079 return (EPERM); 1080 } 1081 1082 /* 1083 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1084 */ 1085 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1086 UIO_SYSSPACE, fspath, td); 1087 error = namei(&nd); 1088 if (error != 0) 1089 return (error); 1090 NDFREE(&nd, NDF_ONLY_PNBUF); 1091 vp = nd.ni_vp; 1092 if ((fsflags & MNT_UPDATE) == 0) { 1093 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1094 strcpy(pathbuf, fspath); 1095 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN); 1096 /* debug.disablefullpath == 1 results in ENODEV */ 1097 if (error == 0 || error == ENODEV) { 1098 error = vfs_domount_first(td, vfsp, pathbuf, vp, 1099 fsflags, optlist); 1100 } 1101 free(pathbuf, M_TEMP); 1102 } else 1103 error = vfs_domount_update(td, vp, fsflags, optlist); 1104 1105 ASSERT_VI_UNLOCKED(vp, __func__); 1106 ASSERT_VOP_UNLOCKED(vp, __func__); 1107 1108 return (error); 1109} 1110 1111/* 1112 * Unmount a filesystem. 1113 * 1114 * Note: unmount takes a path to the vnode mounted on as argument, not 1115 * special file (as before). 1116 */ 1117#ifndef _SYS_SYSPROTO_H_ 1118struct unmount_args { 1119 char *path; 1120 int flags; 1121}; 1122#endif 1123/* ARGSUSED */ 1124int 1125sys_unmount(td, uap) 1126 struct thread *td; 1127 register struct unmount_args /* { 1128 char *path; 1129 int flags; 1130 } */ *uap; 1131{ 1132 struct nameidata nd; 1133 struct mount *mp; 1134 char *pathbuf; 1135 int error, id0, id1; 1136 1137 AUDIT_ARG_VALUE(uap->flags); 1138 if (jailed(td->td_ucred) || usermount == 0) { 1139 error = priv_check(td, PRIV_VFS_UNMOUNT); 1140 if (error) 1141 return (error); 1142 } 1143 1144 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1145 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1146 if (error) { 1147 free(pathbuf, M_TEMP); 1148 return (error); 1149 } 1150 if (uap->flags & MNT_BYFSID) { 1151 AUDIT_ARG_TEXT(pathbuf); 1152 /* Decode the filesystem ID. */ 1153 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1154 free(pathbuf, M_TEMP); 1155 return (EINVAL); 1156 } 1157 1158 mtx_lock(&mountlist_mtx); 1159 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1160 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1161 mp->mnt_stat.f_fsid.val[1] == id1) 1162 break; 1163 } 1164 mtx_unlock(&mountlist_mtx); 1165 } else { 1166 /* 1167 * Try to find global path for path argument. 1168 */ 1169 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1170 UIO_SYSSPACE, pathbuf, td); 1171 if (namei(&nd) == 0) { 1172 NDFREE(&nd, NDF_ONLY_PNBUF); 1173 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf, 1174 MNAMELEN); 1175 if (error == 0 || error == ENODEV) 1176 vput(nd.ni_vp); 1177 } 1178 mtx_lock(&mountlist_mtx); 1179 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1180 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1181 break; 1182 } 1183 mtx_unlock(&mountlist_mtx); 1184 } 1185 free(pathbuf, M_TEMP); 1186 if (mp == NULL) { 1187 /* 1188 * Previously we returned ENOENT for a nonexistent path and 1189 * EINVAL for a non-mountpoint. We cannot tell these apart 1190 * now, so in the !MNT_BYFSID case return the more likely 1191 * EINVAL for compatibility. 1192 */ 1193 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1194 } 1195 1196 /* 1197 * Don't allow unmounting the root filesystem. 1198 */ 1199 if (mp->mnt_flag & MNT_ROOTFS) 1200 return (EINVAL); 1201 error = dounmount(mp, uap->flags, td); 1202 return (error); 1203} 1204 1205/* 1206 * Do the actual filesystem unmount. 1207 */ 1208int 1209dounmount(mp, flags, td) 1210 struct mount *mp; 1211 int flags; 1212 struct thread *td; 1213{ 1214 struct vnode *coveredvp, *fsrootvp; 1215 int error; 1216 uint64_t async_flag; 1217 int mnt_gen_r; 1218 1219 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1220 mnt_gen_r = mp->mnt_gen; 1221 VI_LOCK(coveredvp); 1222 vholdl(coveredvp); 1223 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1224 vdrop(coveredvp); 1225 /* 1226 * Check for mp being unmounted while waiting for the 1227 * covered vnode lock. 1228 */ 1229 if (coveredvp->v_mountedhere != mp || 1230 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1231 VOP_UNLOCK(coveredvp, 0); 1232 return (EBUSY); 1233 } 1234 } 1235 /* 1236 * Only privileged root, or (if MNT_USER is set) the user that did the 1237 * original mount is permitted to unmount this filesystem. 1238 */ 1239 error = vfs_suser(mp, td); 1240 if (error) { 1241 if (coveredvp) 1242 VOP_UNLOCK(coveredvp, 0); 1243 return (error); 1244 } 1245 1246 vn_start_write(NULL, &mp, V_WAIT); 1247 MNT_ILOCK(mp); 1248 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 || 1249 !TAILQ_EMPTY(&mp->mnt_uppers)) { 1250 MNT_IUNLOCK(mp); 1251 if (coveredvp) 1252 VOP_UNLOCK(coveredvp, 0); 1253 vn_finished_write(mp); 1254 return (EBUSY); 1255 } 1256 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1257 /* Allow filesystems to detect that a forced unmount is in progress. */ 1258 if (flags & MNT_FORCE) { 1259 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1260 MNT_IUNLOCK(mp); 1261 /* 1262 * Must be done after setting MNTK_UNMOUNTF and before 1263 * waiting for mnt_lockref to become 0. 1264 */ 1265 VFS_PURGE(mp); 1266 MNT_ILOCK(mp); 1267 } 1268 error = 0; 1269 if (mp->mnt_lockref) { 1270 mp->mnt_kern_flag |= MNTK_DRAINING; 1271 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1272 "mount drain", 0); 1273 } 1274 MNT_IUNLOCK(mp); 1275 KASSERT(mp->mnt_lockref == 0, 1276 ("%s: invalid lock refcount in the drain path @ %s:%d", 1277 __func__, __FILE__, __LINE__)); 1278 KASSERT(error == 0, 1279 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1280 __func__, __FILE__, __LINE__)); 1281 1282 if (mp->mnt_flag & MNT_EXPUBLIC) 1283 vfs_setpublicfs(NULL, NULL, NULL); 1284 1285 vfs_msync(mp, MNT_WAIT); 1286 MNT_ILOCK(mp); 1287 async_flag = mp->mnt_flag & MNT_ASYNC; 1288 mp->mnt_flag &= ~MNT_ASYNC; 1289 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1290 MNT_IUNLOCK(mp); 1291 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1292 vfs_deallocate_syncvnode(mp); 1293 /* 1294 * For forced unmounts, move process cdir/rdir refs on the fs root 1295 * vnode to the covered vnode. For non-forced unmounts we want 1296 * such references to cause an EBUSY error. 1297 */ 1298 if ((flags & MNT_FORCE) && 1299 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1300 if (mp->mnt_vnodecovered != NULL) 1301 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1302 if (fsrootvp == rootvnode) { 1303 vrele(rootvnode); 1304 rootvnode = NULL; 1305 } 1306 vput(fsrootvp); 1307 } 1308 if (((mp->mnt_flag & MNT_RDONLY) || 1309 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1310 error = VFS_UNMOUNT(mp, flags); 1311 vn_finished_write(mp); 1312 /* 1313 * If we failed to flush the dirty blocks for this mount point, 1314 * undo all the cdir/rdir and rootvnode changes we made above. 1315 * Unless we failed to do so because the device is reporting that 1316 * it doesn't exist anymore. 1317 */ 1318 if (error && error != ENXIO) { 1319 if ((flags & MNT_FORCE) && 1320 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1321 if (mp->mnt_vnodecovered != NULL) 1322 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1323 if (rootvnode == NULL) { 1324 rootvnode = fsrootvp; 1325 vref(rootvnode); 1326 } 1327 vput(fsrootvp); 1328 } 1329 MNT_ILOCK(mp); 1330 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1331 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1332 MNT_IUNLOCK(mp); 1333 vfs_allocate_syncvnode(mp); 1334 MNT_ILOCK(mp); 1335 } 1336 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1337 mp->mnt_flag |= async_flag; 1338 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1339 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1340 mp->mnt_kern_flag |= MNTK_ASYNC; 1341 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1342 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1343 wakeup(mp); 1344 } 1345 MNT_IUNLOCK(mp); 1346 if (coveredvp) 1347 VOP_UNLOCK(coveredvp, 0); 1348 return (error); 1349 } 1350 mtx_lock(&mountlist_mtx); 1351 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1352 mtx_unlock(&mountlist_mtx); 1353 EVENTHANDLER_INVOKE(vfs_unmounted, mp, td); 1354 if (coveredvp != NULL) { 1355 coveredvp->v_mountedhere = NULL; 1356 vput(coveredvp); 1357 } 1358 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1359 vfs_mount_destroy(mp); 1360 return (0); 1361} 1362 1363/* 1364 * Report errors during filesystem mounting. 1365 */ 1366void 1367vfs_mount_error(struct mount *mp, const char *fmt, ...) 1368{ 1369 struct vfsoptlist *moptlist = mp->mnt_optnew; 1370 va_list ap; 1371 int error, len; 1372 char *errmsg; 1373 1374 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1375 if (error || errmsg == NULL || len <= 0) 1376 return; 1377 1378 va_start(ap, fmt); 1379 vsnprintf(errmsg, (size_t)len, fmt, ap); 1380 va_end(ap); 1381} 1382 1383void 1384vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1385{ 1386 va_list ap; 1387 int error, len; 1388 char *errmsg; 1389 1390 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1391 if (error || errmsg == NULL || len <= 0) 1392 return; 1393 1394 va_start(ap, fmt); 1395 vsnprintf(errmsg, (size_t)len, fmt, ap); 1396 va_end(ap); 1397} 1398 1399/* 1400 * --------------------------------------------------------------------- 1401 * Functions for querying mount options/arguments from filesystems. 1402 */ 1403 1404/* 1405 * Check that no unknown options are given 1406 */ 1407int 1408vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1409{ 1410 struct vfsopt *opt; 1411 char errmsg[255]; 1412 const char **t, *p, *q; 1413 int ret = 0; 1414 1415 TAILQ_FOREACH(opt, opts, link) { 1416 p = opt->name; 1417 q = NULL; 1418 if (p[0] == 'n' && p[1] == 'o') 1419 q = p + 2; 1420 for(t = global_opts; *t != NULL; t++) { 1421 if (strcmp(*t, p) == 0) 1422 break; 1423 if (q != NULL) { 1424 if (strcmp(*t, q) == 0) 1425 break; 1426 } 1427 } 1428 if (*t != NULL) 1429 continue; 1430 for(t = legal; *t != NULL; t++) { 1431 if (strcmp(*t, p) == 0) 1432 break; 1433 if (q != NULL) { 1434 if (strcmp(*t, q) == 0) 1435 break; 1436 } 1437 } 1438 if (*t != NULL) 1439 continue; 1440 snprintf(errmsg, sizeof(errmsg), 1441 "mount option <%s> is unknown", p); 1442 ret = EINVAL; 1443 } 1444 if (ret != 0) { 1445 TAILQ_FOREACH(opt, opts, link) { 1446 if (strcmp(opt->name, "errmsg") == 0) { 1447 strncpy((char *)opt->value, errmsg, opt->len); 1448 break; 1449 } 1450 } 1451 if (opt == NULL) 1452 printf("%s\n", errmsg); 1453 } 1454 return (ret); 1455} 1456 1457/* 1458 * Get a mount option by its name. 1459 * 1460 * Return 0 if the option was found, ENOENT otherwise. 1461 * If len is non-NULL it will be filled with the length 1462 * of the option. If buf is non-NULL, it will be filled 1463 * with the address of the option. 1464 */ 1465int 1466vfs_getopt(opts, name, buf, len) 1467 struct vfsoptlist *opts; 1468 const char *name; 1469 void **buf; 1470 int *len; 1471{ 1472 struct vfsopt *opt; 1473 1474 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1475 1476 TAILQ_FOREACH(opt, opts, link) { 1477 if (strcmp(name, opt->name) == 0) { 1478 opt->seen = 1; 1479 if (len != NULL) 1480 *len = opt->len; 1481 if (buf != NULL) 1482 *buf = opt->value; 1483 return (0); 1484 } 1485 } 1486 return (ENOENT); 1487} 1488 1489int 1490vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1491{ 1492 struct vfsopt *opt; 1493 1494 if (opts == NULL) 1495 return (-1); 1496 1497 TAILQ_FOREACH(opt, opts, link) { 1498 if (strcmp(name, opt->name) == 0) { 1499 opt->seen = 1; 1500 return (opt->pos); 1501 } 1502 } 1503 return (-1); 1504} 1505 1506int 1507vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value) 1508{ 1509 char *opt_value, *vtp; 1510 quad_t iv; 1511 int error, opt_len; 1512 1513 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len); 1514 if (error != 0) 1515 return (error); 1516 if (opt_len == 0 || opt_value == NULL) 1517 return (EINVAL); 1518 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0') 1519 return (EINVAL); 1520 iv = strtoq(opt_value, &vtp, 0); 1521 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0')) 1522 return (EINVAL); 1523 if (iv < 0) 1524 return (EINVAL); 1525 switch (vtp[0]) { 1526 case 't': 1527 case 'T': 1528 iv *= 1024; 1529 case 'g': 1530 case 'G': 1531 iv *= 1024; 1532 case 'm': 1533 case 'M': 1534 iv *= 1024; 1535 case 'k': 1536 case 'K': 1537 iv *= 1024; 1538 case '\0': 1539 break; 1540 default: 1541 return (EINVAL); 1542 } 1543 *value = iv; 1544 1545 return (0); 1546} 1547 1548char * 1549vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1550{ 1551 struct vfsopt *opt; 1552 1553 *error = 0; 1554 TAILQ_FOREACH(opt, opts, link) { 1555 if (strcmp(name, opt->name) != 0) 1556 continue; 1557 opt->seen = 1; 1558 if (opt->len == 0 || 1559 ((char *)opt->value)[opt->len - 1] != '\0') { 1560 *error = EINVAL; 1561 return (NULL); 1562 } 1563 return (opt->value); 1564 } 1565 *error = ENOENT; 1566 return (NULL); 1567} 1568 1569int 1570vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, 1571 uint64_t val) 1572{ 1573 struct vfsopt *opt; 1574 1575 TAILQ_FOREACH(opt, opts, link) { 1576 if (strcmp(name, opt->name) == 0) { 1577 opt->seen = 1; 1578 if (w != NULL) 1579 *w |= val; 1580 return (1); 1581 } 1582 } 1583 if (w != NULL) 1584 *w &= ~val; 1585 return (0); 1586} 1587 1588int 1589vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1590{ 1591 va_list ap; 1592 struct vfsopt *opt; 1593 int ret; 1594 1595 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1596 1597 TAILQ_FOREACH(opt, opts, link) { 1598 if (strcmp(name, opt->name) != 0) 1599 continue; 1600 opt->seen = 1; 1601 if (opt->len == 0 || opt->value == NULL) 1602 return (0); 1603 if (((char *)opt->value)[opt->len - 1] != '\0') 1604 return (0); 1605 va_start(ap, fmt); 1606 ret = vsscanf(opt->value, fmt, ap); 1607 va_end(ap); 1608 return (ret); 1609 } 1610 return (0); 1611} 1612 1613int 1614vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 1615{ 1616 struct vfsopt *opt; 1617 1618 TAILQ_FOREACH(opt, opts, link) { 1619 if (strcmp(name, opt->name) != 0) 1620 continue; 1621 opt->seen = 1; 1622 if (opt->value == NULL) 1623 opt->len = len; 1624 else { 1625 if (opt->len != len) 1626 return (EINVAL); 1627 bcopy(value, opt->value, len); 1628 } 1629 return (0); 1630 } 1631 return (ENOENT); 1632} 1633 1634int 1635vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 1636{ 1637 struct vfsopt *opt; 1638 1639 TAILQ_FOREACH(opt, opts, link) { 1640 if (strcmp(name, opt->name) != 0) 1641 continue; 1642 opt->seen = 1; 1643 if (opt->value == NULL) 1644 opt->len = len; 1645 else { 1646 if (opt->len < len) 1647 return (EINVAL); 1648 opt->len = len; 1649 bcopy(value, opt->value, len); 1650 } 1651 return (0); 1652 } 1653 return (ENOENT); 1654} 1655 1656int 1657vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 1658{ 1659 struct vfsopt *opt; 1660 1661 TAILQ_FOREACH(opt, opts, link) { 1662 if (strcmp(name, opt->name) != 0) 1663 continue; 1664 opt->seen = 1; 1665 if (opt->value == NULL) 1666 opt->len = strlen(value) + 1; 1667 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 1668 return (EINVAL); 1669 return (0); 1670 } 1671 return (ENOENT); 1672} 1673 1674/* 1675 * Find and copy a mount option. 1676 * 1677 * The size of the buffer has to be specified 1678 * in len, if it is not the same length as the 1679 * mount option, EINVAL is returned. 1680 * Returns ENOENT if the option is not found. 1681 */ 1682int 1683vfs_copyopt(opts, name, dest, len) 1684 struct vfsoptlist *opts; 1685 const char *name; 1686 void *dest; 1687 int len; 1688{ 1689 struct vfsopt *opt; 1690 1691 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1692 1693 TAILQ_FOREACH(opt, opts, link) { 1694 if (strcmp(name, opt->name) == 0) { 1695 opt->seen = 1; 1696 if (len != opt->len) 1697 return (EINVAL); 1698 bcopy(opt->value, dest, opt->len); 1699 return (0); 1700 } 1701 } 1702 return (ENOENT); 1703} 1704 1705int 1706__vfs_statfs(struct mount *mp, struct statfs *sbp) 1707{ 1708 int error; 1709 1710 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 1711 if (sbp != &mp->mnt_stat) 1712 *sbp = mp->mnt_stat; 1713 return (error); 1714} 1715 1716void 1717vfs_mountedfrom(struct mount *mp, const char *from) 1718{ 1719 1720 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 1721 strlcpy(mp->mnt_stat.f_mntfromname, from, 1722 sizeof mp->mnt_stat.f_mntfromname); 1723} 1724 1725/* 1726 * --------------------------------------------------------------------- 1727 * This is the api for building mount args and mounting filesystems from 1728 * inside the kernel. 1729 * 1730 * The API works by accumulation of individual args. First error is 1731 * latched. 1732 * 1733 * XXX: should be documented in new manpage kernel_mount(9) 1734 */ 1735 1736/* A memory allocation which must be freed when we are done */ 1737struct mntaarg { 1738 SLIST_ENTRY(mntaarg) next; 1739}; 1740 1741/* The header for the mount arguments */ 1742struct mntarg { 1743 struct iovec *v; 1744 int len; 1745 int error; 1746 SLIST_HEAD(, mntaarg) list; 1747}; 1748 1749/* 1750 * Add a boolean argument. 1751 * 1752 * flag is the boolean value. 1753 * name must start with "no". 1754 */ 1755struct mntarg * 1756mount_argb(struct mntarg *ma, int flag, const char *name) 1757{ 1758 1759 KASSERT(name[0] == 'n' && name[1] == 'o', 1760 ("mount_argb(...,%s): name must start with 'no'", name)); 1761 1762 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 1763} 1764 1765/* 1766 * Add an argument printf style 1767 */ 1768struct mntarg * 1769mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 1770{ 1771 va_list ap; 1772 struct mntaarg *maa; 1773 struct sbuf *sb; 1774 int len; 1775 1776 if (ma == NULL) { 1777 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1778 SLIST_INIT(&ma->list); 1779 } 1780 if (ma->error) 1781 return (ma); 1782 1783 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1784 M_MOUNT, M_WAITOK); 1785 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1786 ma->v[ma->len].iov_len = strlen(name) + 1; 1787 ma->len++; 1788 1789 sb = sbuf_new_auto(); 1790 va_start(ap, fmt); 1791 sbuf_vprintf(sb, fmt, ap); 1792 va_end(ap); 1793 sbuf_finish(sb); 1794 len = sbuf_len(sb) + 1; 1795 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1796 SLIST_INSERT_HEAD(&ma->list, maa, next); 1797 bcopy(sbuf_data(sb), maa + 1, len); 1798 sbuf_delete(sb); 1799 1800 ma->v[ma->len].iov_base = maa + 1; 1801 ma->v[ma->len].iov_len = len; 1802 ma->len++; 1803 1804 return (ma); 1805} 1806 1807/* 1808 * Add an argument which is a userland string. 1809 */ 1810struct mntarg * 1811mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 1812{ 1813 struct mntaarg *maa; 1814 char *tbuf; 1815 1816 if (val == NULL) 1817 return (ma); 1818 if (ma == NULL) { 1819 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1820 SLIST_INIT(&ma->list); 1821 } 1822 if (ma->error) 1823 return (ma); 1824 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1825 SLIST_INSERT_HEAD(&ma->list, maa, next); 1826 tbuf = (void *)(maa + 1); 1827 ma->error = copyinstr(val, tbuf, len, NULL); 1828 return (mount_arg(ma, name, tbuf, -1)); 1829} 1830 1831/* 1832 * Plain argument. 1833 * 1834 * If length is -1, treat value as a C string. 1835 */ 1836struct mntarg * 1837mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 1838{ 1839 1840 if (ma == NULL) { 1841 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1842 SLIST_INIT(&ma->list); 1843 } 1844 if (ma->error) 1845 return (ma); 1846 1847 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1848 M_MOUNT, M_WAITOK); 1849 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1850 ma->v[ma->len].iov_len = strlen(name) + 1; 1851 ma->len++; 1852 1853 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 1854 if (len < 0) 1855 ma->v[ma->len].iov_len = strlen(val) + 1; 1856 else 1857 ma->v[ma->len].iov_len = len; 1858 ma->len++; 1859 return (ma); 1860} 1861 1862/* 1863 * Free a mntarg structure 1864 */ 1865static void 1866free_mntarg(struct mntarg *ma) 1867{ 1868 struct mntaarg *maa; 1869 1870 while (!SLIST_EMPTY(&ma->list)) { 1871 maa = SLIST_FIRST(&ma->list); 1872 SLIST_REMOVE_HEAD(&ma->list, next); 1873 free(maa, M_MOUNT); 1874 } 1875 free(ma->v, M_MOUNT); 1876 free(ma, M_MOUNT); 1877} 1878 1879/* 1880 * Mount a filesystem 1881 */ 1882int 1883kernel_mount(struct mntarg *ma, uint64_t flags) 1884{ 1885 struct uio auio; 1886 int error; 1887 1888 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 1889 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 1890 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 1891 1892 auio.uio_iov = ma->v; 1893 auio.uio_iovcnt = ma->len; 1894 auio.uio_segflg = UIO_SYSSPACE; 1895 1896 error = ma->error; 1897 if (!error) 1898 error = vfs_donmount(curthread, flags, &auio); 1899 free_mntarg(ma); 1900 return (error); 1901} 1902 1903/* 1904 * A printflike function to mount a filesystem. 1905 */ 1906int 1907kernel_vmount(int flags, ...) 1908{ 1909 struct mntarg *ma = NULL; 1910 va_list ap; 1911 const char *cp; 1912 const void *vp; 1913 int error; 1914 1915 va_start(ap, flags); 1916 for (;;) { 1917 cp = va_arg(ap, const char *); 1918 if (cp == NULL) 1919 break; 1920 vp = va_arg(ap, const void *); 1921 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 1922 } 1923 va_end(ap); 1924 1925 error = kernel_mount(ma, flags); 1926 return (error); 1927} 1928 1929void 1930vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp) 1931{ 1932 1933 bcopy(oexp, exp, sizeof(*oexp)); 1934 exp->ex_numsecflavors = 0; 1935} 1936