vfs_mount.c revision 213365
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: head/sys/kern/vfs_mount.c 213365 2010-10-02 19:44:13Z marcel $"); 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/syscallsubr.h> 55#include <sys/sysproto.h> 56#include <sys/sx.h> 57#include <sys/sysctl.h> 58#include <sys/sysent.h> 59#include <sys/systm.h> 60#include <sys/vnode.h> 61#include <vm/uma.h> 62 63#include <geom/geom.h> 64 65#include <machine/stdarg.h> 66 67#include <security/audit/audit.h> 68#include <security/mac/mac_framework.h> 69 70#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 71 72static int vfs_domount(struct thread *td, const char *fstype, 73 char *fspath, int fsflags, void *fsdata); 74static void free_mntarg(struct mntarg *ma); 75 76static int usermount = 0; 77SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 78 "Unprivileged users may mount and unmount file systems"); 79 80MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 81MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 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} 177 178/* 179 * Check if options are equal (with or without the "no" prefix). 180 */ 181static int 182vfs_equalopts(const char *opt1, const char *opt2) 183{ 184 char *p; 185 186 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 187 if (strcmp(opt1, opt2) == 0) 188 return (1); 189 /* "noopt" vs. "opt" */ 190 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 191 return (1); 192 /* "opt" vs. "noopt" */ 193 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 194 return (1); 195 while ((p = strchr(opt1, '.')) != NULL && 196 !strncmp(opt1, opt2, ++p - opt1)) { 197 opt2 += p - opt1; 198 opt1 = p; 199 /* "foo.noopt" vs. "foo.opt" */ 200 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 201 return (1); 202 /* "foo.opt" vs. "foo.noopt" */ 203 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 204 return (1); 205 } 206 return (0); 207} 208 209/* 210 * If a mount option is specified several times, 211 * (with or without the "no" prefix) only keep 212 * the last occurence of it. 213 */ 214static void 215vfs_sanitizeopts(struct vfsoptlist *opts) 216{ 217 struct vfsopt *opt, *opt2, *tmp; 218 219 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 220 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 221 while (opt2 != NULL) { 222 if (vfs_equalopts(opt->name, opt2->name)) { 223 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 224 vfs_freeopt(opts, opt2); 225 opt2 = tmp; 226 } else { 227 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 228 } 229 } 230 } 231} 232 233/* 234 * Build a linked list of mount options from a struct uio. 235 */ 236int 237vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 238{ 239 struct vfsoptlist *opts; 240 struct vfsopt *opt; 241 size_t memused, namelen, optlen; 242 unsigned int i, iovcnt; 243 int error; 244 245 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 246 TAILQ_INIT(opts); 247 memused = 0; 248 iovcnt = auio->uio_iovcnt; 249 for (i = 0; i < iovcnt; i += 2) { 250 namelen = auio->uio_iov[i].iov_len; 251 optlen = auio->uio_iov[i + 1].iov_len; 252 memused += sizeof(struct vfsopt) + optlen + namelen; 253 /* 254 * Avoid consuming too much memory, and attempts to overflow 255 * memused. 256 */ 257 if (memused > VFS_MOUNTARG_SIZE_MAX || 258 optlen > VFS_MOUNTARG_SIZE_MAX || 259 namelen > VFS_MOUNTARG_SIZE_MAX) { 260 error = EINVAL; 261 goto bad; 262 } 263 264 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 265 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 266 opt->value = NULL; 267 opt->len = 0; 268 opt->pos = i / 2; 269 opt->seen = 0; 270 271 /* 272 * Do this early, so jumps to "bad" will free the current 273 * option. 274 */ 275 TAILQ_INSERT_TAIL(opts, opt, link); 276 277 if (auio->uio_segflg == UIO_SYSSPACE) { 278 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 279 } else { 280 error = copyin(auio->uio_iov[i].iov_base, opt->name, 281 namelen); 282 if (error) 283 goto bad; 284 } 285 /* Ensure names are null-terminated strings. */ 286 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 287 error = EINVAL; 288 goto bad; 289 } 290 if (optlen != 0) { 291 opt->len = optlen; 292 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 293 if (auio->uio_segflg == UIO_SYSSPACE) { 294 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 295 optlen); 296 } else { 297 error = copyin(auio->uio_iov[i + 1].iov_base, 298 opt->value, optlen); 299 if (error) 300 goto bad; 301 } 302 } 303 } 304 vfs_sanitizeopts(opts); 305 *options = opts; 306 return (0); 307bad: 308 vfs_freeopts(opts); 309 return (error); 310} 311 312/* 313 * Merge the old mount options with the new ones passed 314 * in the MNT_UPDATE case. 315 * 316 * XXX This function will keep a "nofoo" option in the 317 * new options if there is no matching "foo" option 318 * to be cancelled in the old options. This is a bug 319 * if the option's canonical name is "foo". E.g., "noro" 320 * shouldn't end up in the mount point's active options, 321 * but it can. 322 */ 323static void 324vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 325{ 326 struct vfsopt *opt, *opt2, *new; 327 328 TAILQ_FOREACH(opt, opts, link) { 329 /* 330 * Check that this option hasn't been redefined 331 * nor cancelled with a "no" mount option. 332 */ 333 opt2 = TAILQ_FIRST(toopts); 334 while (opt2 != NULL) { 335 if (strcmp(opt2->name, opt->name) == 0) 336 goto next; 337 if (strncmp(opt2->name, "no", 2) == 0 && 338 strcmp(opt2->name + 2, opt->name) == 0) { 339 vfs_freeopt(toopts, opt2); 340 goto next; 341 } 342 opt2 = TAILQ_NEXT(opt2, link); 343 } 344 /* We want this option, duplicate it. */ 345 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 346 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 347 strcpy(new->name, opt->name); 348 if (opt->len != 0) { 349 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 350 bcopy(opt->value, new->value, opt->len); 351 } else { 352 new->value = NULL; 353 } 354 new->len = opt->len; 355 new->seen = opt->seen; 356 TAILQ_INSERT_TAIL(toopts, new, link); 357next: 358 continue; 359 } 360} 361 362/* 363 * Mount a filesystem. 364 */ 365int 366nmount(td, uap) 367 struct thread *td; 368 struct nmount_args /* { 369 struct iovec *iovp; 370 unsigned int iovcnt; 371 int flags; 372 } */ *uap; 373{ 374 struct uio *auio; 375 int error; 376 u_int iovcnt; 377 378 AUDIT_ARG_FFLAGS(uap->flags); 379 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 380 uap->iovp, uap->iovcnt, uap->flags); 381 382 /* 383 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 384 * userspace to set this flag, but we must filter it out if we want 385 * MNT_UPDATE on the root file system to work. 386 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 387 */ 388 uap->flags &= ~MNT_ROOTFS; 389 390 iovcnt = uap->iovcnt; 391 /* 392 * Check that we have an even number of iovec's 393 * and that we have at least two options. 394 */ 395 if ((iovcnt & 1) || (iovcnt < 4)) { 396 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 397 uap->iovcnt); 398 return (EINVAL); 399 } 400 401 error = copyinuio(uap->iovp, iovcnt, &auio); 402 if (error) { 403 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 404 __func__, error); 405 return (error); 406 } 407 error = vfs_donmount(td, uap->flags, auio); 408 409 free(auio, M_IOV); 410 return (error); 411} 412 413/* 414 * --------------------------------------------------------------------- 415 * Various utility functions 416 */ 417 418void 419vfs_ref(struct mount *mp) 420{ 421 422 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 423 MNT_ILOCK(mp); 424 MNT_REF(mp); 425 MNT_IUNLOCK(mp); 426} 427 428void 429vfs_rel(struct mount *mp) 430{ 431 432 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 433 MNT_ILOCK(mp); 434 MNT_REL(mp); 435 MNT_IUNLOCK(mp); 436} 437 438/* 439 * Allocate and initialize the mount point struct. 440 */ 441struct mount * 442vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 443 struct ucred *cred) 444{ 445 struct mount *mp; 446 447 mp = uma_zalloc(mount_zone, M_WAITOK); 448 bzero(&mp->mnt_startzero, 449 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 450 TAILQ_INIT(&mp->mnt_nvnodelist); 451 mp->mnt_nvnodelistsize = 0; 452 mp->mnt_ref = 0; 453 (void) vfs_busy(mp, MBF_NOWAIT); 454 mp->mnt_op = vfsp->vfc_vfsops; 455 mp->mnt_vfc = vfsp; 456 vfsp->vfc_refcount++; /* XXX Unlocked */ 457 mp->mnt_stat.f_type = vfsp->vfc_typenum; 458 mp->mnt_gen++; 459 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 460 mp->mnt_vnodecovered = vp; 461 mp->mnt_cred = crdup(cred); 462 mp->mnt_stat.f_owner = cred->cr_uid; 463 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 464 mp->mnt_iosize_max = DFLTPHYS; 465#ifdef MAC 466 mac_mount_init(mp); 467 mac_mount_create(cred, mp); 468#endif 469 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 470 return (mp); 471} 472 473/* 474 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 475 */ 476void 477vfs_mount_destroy(struct mount *mp) 478{ 479 480 MNT_ILOCK(mp); 481 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 482 if (mp->mnt_kern_flag & MNTK_MWAIT) { 483 mp->mnt_kern_flag &= ~MNTK_MWAIT; 484 wakeup(mp); 485 } 486 while (mp->mnt_ref) 487 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 488 KASSERT(mp->mnt_ref == 0, 489 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 490 __FILE__, __LINE__)); 491 if (mp->mnt_writeopcount != 0) 492 panic("vfs_mount_destroy: nonzero writeopcount"); 493 if (mp->mnt_secondary_writes != 0) 494 panic("vfs_mount_destroy: nonzero secondary_writes"); 495 mp->mnt_vfc->vfc_refcount--; 496 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 497 struct vnode *vp; 498 499 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 500 vprint("", vp); 501 panic("unmount: dangling vnode"); 502 } 503 if (mp->mnt_nvnodelistsize != 0) 504 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 505 if (mp->mnt_lockref != 0) 506 panic("vfs_mount_destroy: nonzero lock refcount"); 507 MNT_IUNLOCK(mp); 508#ifdef MAC 509 mac_mount_destroy(mp); 510#endif 511 if (mp->mnt_opt != NULL) 512 vfs_freeopts(mp->mnt_opt); 513 crfree(mp->mnt_cred); 514 uma_zfree(mount_zone, mp); 515} 516 517int 518vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 519{ 520 struct vfsoptlist *optlist; 521 struct vfsopt *opt, *noro_opt, *tmp_opt; 522 char *fstype, *fspath, *errmsg; 523 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 524 int has_rw, has_noro; 525 526 errmsg = fspath = NULL; 527 errmsg_len = has_noro = has_rw = fspathlen = 0; 528 errmsg_pos = -1; 529 530 error = vfs_buildopts(fsoptions, &optlist); 531 if (error) 532 return (error); 533 534 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 535 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 536 537 /* 538 * We need these two options before the others, 539 * and they are mandatory for any filesystem. 540 * Ensure they are NUL terminated as well. 541 */ 542 fstypelen = 0; 543 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 544 if (error || fstype[fstypelen - 1] != '\0') { 545 error = EINVAL; 546 if (errmsg != NULL) 547 strncpy(errmsg, "Invalid fstype", errmsg_len); 548 goto bail; 549 } 550 fspathlen = 0; 551 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 552 if (error || fspath[fspathlen - 1] != '\0') { 553 error = EINVAL; 554 if (errmsg != NULL) 555 strncpy(errmsg, "Invalid fspath", errmsg_len); 556 goto bail; 557 } 558 559 /* 560 * We need to see if we have the "update" option 561 * before we call vfs_domount(), since vfs_domount() has special 562 * logic based on MNT_UPDATE. This is very important 563 * when we want to update the root filesystem. 564 */ 565 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 566 if (strcmp(opt->name, "update") == 0) { 567 fsflags |= MNT_UPDATE; 568 vfs_freeopt(optlist, opt); 569 } 570 else if (strcmp(opt->name, "async") == 0) 571 fsflags |= MNT_ASYNC; 572 else if (strcmp(opt->name, "force") == 0) { 573 fsflags |= MNT_FORCE; 574 vfs_freeopt(optlist, opt); 575 } 576 else if (strcmp(opt->name, "reload") == 0) { 577 fsflags |= MNT_RELOAD; 578 vfs_freeopt(optlist, opt); 579 } 580 else if (strcmp(opt->name, "multilabel") == 0) 581 fsflags |= MNT_MULTILABEL; 582 else if (strcmp(opt->name, "noasync") == 0) 583 fsflags &= ~MNT_ASYNC; 584 else if (strcmp(opt->name, "noatime") == 0) 585 fsflags |= MNT_NOATIME; 586 else if (strcmp(opt->name, "atime") == 0) { 587 free(opt->name, M_MOUNT); 588 opt->name = strdup("nonoatime", M_MOUNT); 589 } 590 else if (strcmp(opt->name, "noclusterr") == 0) 591 fsflags |= MNT_NOCLUSTERR; 592 else if (strcmp(opt->name, "clusterr") == 0) { 593 free(opt->name, M_MOUNT); 594 opt->name = strdup("nonoclusterr", M_MOUNT); 595 } 596 else if (strcmp(opt->name, "noclusterw") == 0) 597 fsflags |= MNT_NOCLUSTERW; 598 else if (strcmp(opt->name, "clusterw") == 0) { 599 free(opt->name, M_MOUNT); 600 opt->name = strdup("nonoclusterw", M_MOUNT); 601 } 602 else if (strcmp(opt->name, "noexec") == 0) 603 fsflags |= MNT_NOEXEC; 604 else if (strcmp(opt->name, "exec") == 0) { 605 free(opt->name, M_MOUNT); 606 opt->name = strdup("nonoexec", M_MOUNT); 607 } 608 else if (strcmp(opt->name, "nosuid") == 0) 609 fsflags |= MNT_NOSUID; 610 else if (strcmp(opt->name, "suid") == 0) { 611 free(opt->name, M_MOUNT); 612 opt->name = strdup("nonosuid", M_MOUNT); 613 } 614 else if (strcmp(opt->name, "nosymfollow") == 0) 615 fsflags |= MNT_NOSYMFOLLOW; 616 else if (strcmp(opt->name, "symfollow") == 0) { 617 free(opt->name, M_MOUNT); 618 opt->name = strdup("nonosymfollow", M_MOUNT); 619 } 620 else if (strcmp(opt->name, "noro") == 0) { 621 fsflags &= ~MNT_RDONLY; 622 has_noro = 1; 623 } 624 else if (strcmp(opt->name, "rw") == 0) { 625 fsflags &= ~MNT_RDONLY; 626 has_rw = 1; 627 } 628 else if (strcmp(opt->name, "ro") == 0) 629 fsflags |= MNT_RDONLY; 630 else if (strcmp(opt->name, "rdonly") == 0) { 631 free(opt->name, M_MOUNT); 632 opt->name = strdup("ro", M_MOUNT); 633 fsflags |= MNT_RDONLY; 634 } 635 else if (strcmp(opt->name, "suiddir") == 0) 636 fsflags |= MNT_SUIDDIR; 637 else if (strcmp(opt->name, "sync") == 0) 638 fsflags |= MNT_SYNCHRONOUS; 639 else if (strcmp(opt->name, "union") == 0) 640 fsflags |= MNT_UNION; 641 } 642 643 /* 644 * If "rw" was specified as a mount option, and we 645 * are trying to update a mount-point from "ro" to "rw", 646 * we need a mount option "noro", since in vfs_mergeopts(), 647 * "noro" will cancel "ro", but "rw" will not do anything. 648 */ 649 if (has_rw && !has_noro) { 650 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 651 noro_opt->name = strdup("noro", M_MOUNT); 652 noro_opt->value = NULL; 653 noro_opt->len = 0; 654 noro_opt->pos = -1; 655 noro_opt->seen = 1; 656 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 657 } 658 659 /* 660 * Be ultra-paranoid about making sure the type and fspath 661 * variables will fit in our mp buffers, including the 662 * terminating NUL. 663 */ 664 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 665 error = ENAMETOOLONG; 666 goto bail; 667 } 668 669 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 670bail: 671 /* copyout the errmsg */ 672 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 673 && errmsg_len > 0 && errmsg != NULL) { 674 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 675 bcopy(errmsg, 676 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 677 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 678 } else { 679 copyout(errmsg, 680 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 681 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 682 } 683 } 684 685 if (error != 0) 686 vfs_freeopts(optlist); 687 return (error); 688} 689 690/* 691 * Old mount API. 692 */ 693#ifndef _SYS_SYSPROTO_H_ 694struct mount_args { 695 char *type; 696 char *path; 697 int flags; 698 caddr_t data; 699}; 700#endif 701/* ARGSUSED */ 702int 703mount(td, uap) 704 struct thread *td; 705 struct mount_args /* { 706 char *type; 707 char *path; 708 int flags; 709 caddr_t data; 710 } */ *uap; 711{ 712 char *fstype; 713 struct vfsconf *vfsp = NULL; 714 struct mntarg *ma = NULL; 715 int error; 716 717 AUDIT_ARG_FFLAGS(uap->flags); 718 719 /* 720 * Filter out MNT_ROOTFS. We do not want clients of mount() in 721 * userspace to set this flag, but we must filter it out if we want 722 * MNT_UPDATE on the root file system to work. 723 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 724 */ 725 uap->flags &= ~MNT_ROOTFS; 726 727 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 728 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 729 if (error) { 730 free(fstype, M_TEMP); 731 return (error); 732 } 733 734 AUDIT_ARG_TEXT(fstype); 735 mtx_lock(&Giant); 736 vfsp = vfs_byname_kld(fstype, td, &error); 737 free(fstype, M_TEMP); 738 if (vfsp == NULL) { 739 mtx_unlock(&Giant); 740 return (ENOENT); 741 } 742 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 743 mtx_unlock(&Giant); 744 return (EOPNOTSUPP); 745 } 746 747 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 748 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 749 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 750 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 751 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 752 753 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags); 754 mtx_unlock(&Giant); 755 return (error); 756} 757 758/* 759 * vfs_domount_first(): first file system mount (not update) 760 */ 761static int 762vfs_domount_first( 763 struct thread *td, /* Calling thread. */ 764 struct vfsconf *vfsp, /* File system type. */ 765 char *fspath, /* Mount path. */ 766 struct vnode *vp, /* Vnode to be covered. */ 767 int fsflags, /* Flags common to all filesystems. */ 768 void *fsdata /* Options local to the filesystem. */ 769 ) 770{ 771 struct vattr va; 772 struct mount *mp; 773 struct vnode *newdp; 774 int error; 775 776 mtx_assert(&Giant, MA_OWNED); 777 ASSERT_VOP_ELOCKED(vp, __func__); 778 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 779 780 /* 781 * If the user is not root, ensure that they own the directory 782 * onto which we are attempting to mount. 783 */ 784 error = VOP_GETATTR(vp, &va, td->td_ucred); 785 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 786 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0); 787 if (error == 0) 788 error = vinvalbuf(vp, V_SAVE, 0, 0); 789 if (error == 0 && vp->v_type != VDIR) 790 error = ENOTDIR; 791 if (error == 0) { 792 VI_LOCK(vp); 793 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 794 vp->v_iflag |= VI_MOUNT; 795 else 796 error = EBUSY; 797 VI_UNLOCK(vp); 798 } 799 if (error != 0) { 800 vput(vp); 801 return (error); 802 } 803 VOP_UNLOCK(vp, 0); 804 805 /* Allocate and initialize the filesystem. */ 806 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 807 /* XXXMAC: pass to vfs_mount_alloc? */ 808 mp->mnt_optnew = fsdata; 809 /* Set the mount level flags. */ 810 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 811 812 /* 813 * Mount the filesystem. 814 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 815 * get. No freeing of cn_pnbuf. 816 */ 817 error = VFS_MOUNT(mp); 818 if (error != 0) { 819 vfs_unbusy(mp); 820 vfs_mount_destroy(mp); 821 VI_LOCK(vp); 822 vp->v_iflag &= ~VI_MOUNT; 823 VI_UNLOCK(vp); 824 vrele(vp); 825 return (error); 826 } 827 828 if (mp->mnt_opt != NULL) 829 vfs_freeopts(mp->mnt_opt); 830 mp->mnt_opt = mp->mnt_optnew; 831 (void)VFS_STATFS(mp, &mp->mnt_stat); 832 833 /* 834 * Prevent external consumers of mount options from reading mnt_optnew. 835 */ 836 mp->mnt_optnew = NULL; 837 838 MNT_ILOCK(mp); 839 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 840 mp->mnt_kern_flag |= MNTK_ASYNC; 841 else 842 mp->mnt_kern_flag &= ~MNTK_ASYNC; 843 MNT_IUNLOCK(mp); 844 845 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 846 cache_purge(vp); 847 VI_LOCK(vp); 848 vp->v_iflag &= ~VI_MOUNT; 849 VI_UNLOCK(vp); 850 vp->v_mountedhere = mp; 851 /* Place the new filesystem at the end of the mount list. */ 852 mtx_lock(&mountlist_mtx); 853 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 854 mtx_unlock(&mountlist_mtx); 855 vfs_event_signal(NULL, VQ_MOUNT, 0); 856 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 857 panic("mount: lost mount"); 858 VOP_UNLOCK(newdp, 0); 859 VOP_UNLOCK(vp, 0); 860 mountcheckdirs(vp, newdp); 861 vrele(newdp); 862 if ((mp->mnt_flag & MNT_RDONLY) == 0) 863 vfs_allocate_syncvnode(mp); 864 vfs_unbusy(mp); 865 return (0); 866} 867 868/* 869 * vfs_domount_update(): update of mounted file system 870 */ 871static int 872vfs_domount_update( 873 struct thread *td, /* Calling thread. */ 874 struct vnode *vp, /* Mount point vnode. */ 875 int fsflags, /* Flags common to all filesystems. */ 876 void *fsdata /* Options local to the filesystem. */ 877 ) 878{ 879 struct oexport_args oexport; 880 struct export_args export; 881 struct mount *mp; 882 int error, flag; 883 884 mtx_assert(&Giant, MA_OWNED); 885 ASSERT_VOP_ELOCKED(vp, __func__); 886 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 887 888 if ((vp->v_vflag & VV_ROOT) == 0) { 889 vput(vp); 890 return (EINVAL); 891 } 892 mp = vp->v_mount; 893 /* 894 * We only allow the filesystem to be reloaded if it 895 * is currently mounted read-only. 896 */ 897 flag = mp->mnt_flag; 898 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 899 vput(vp); 900 return (EOPNOTSUPP); /* Needs translation */ 901 } 902 /* 903 * Only privileged root, or (if MNT_USER is set) the user that 904 * did the original mount is permitted to update it. 905 */ 906 error = vfs_suser(mp, td); 907 if (error != 0) { 908 vput(vp); 909 return (error); 910 } 911 if (vfs_busy(mp, MBF_NOWAIT)) { 912 vput(vp); 913 return (EBUSY); 914 } 915 VI_LOCK(vp); 916 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 917 VI_UNLOCK(vp); 918 vfs_unbusy(mp); 919 vput(vp); 920 return (EBUSY); 921 } 922 vp->v_iflag |= VI_MOUNT; 923 VI_UNLOCK(vp); 924 VOP_UNLOCK(vp, 0); 925 926 MNT_ILOCK(mp); 927 mp->mnt_flag &= ~MNT_UPDATEMASK; 928 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 929 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 930 if ((mp->mnt_flag & MNT_ASYNC) == 0) 931 mp->mnt_kern_flag &= ~MNTK_ASYNC; 932 MNT_IUNLOCK(mp); 933 mp->mnt_optnew = fsdata; 934 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 935 936 /* 937 * Mount the filesystem. 938 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 939 * get. No freeing of cn_pnbuf. 940 */ 941 error = VFS_MOUNT(mp); 942 943 if (error == 0) { 944 /* Process the export option. */ 945 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 946 sizeof(export)) == 0) { 947 error = vfs_export(mp, &export); 948 } else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 949 sizeof(oexport)) == 0) { 950 export.ex_flags = oexport.ex_flags; 951 export.ex_root = oexport.ex_root; 952 export.ex_anon = oexport.ex_anon; 953 export.ex_addr = oexport.ex_addr; 954 export.ex_addrlen = oexport.ex_addrlen; 955 export.ex_mask = oexport.ex_mask; 956 export.ex_masklen = oexport.ex_masklen; 957 export.ex_indexfile = oexport.ex_indexfile; 958 export.ex_numsecflavors = 0; 959 error = vfs_export(mp, &export); 960 } 961 } 962 963 MNT_ILOCK(mp); 964 if (error == 0) { 965 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 966 MNT_SNAPSHOT); 967 } else { 968 /* 969 * If we fail, restore old mount flags. MNT_QUOTA is special, 970 * because it is not part of MNT_UPDATEMASK, but it could have 971 * changed in the meantime if quotactl(2) was called. 972 * All in all we want current value of MNT_QUOTA, not the old 973 * one. 974 */ 975 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 976 } 977 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 978 mp->mnt_kern_flag |= MNTK_ASYNC; 979 else 980 mp->mnt_kern_flag &= ~MNTK_ASYNC; 981 MNT_IUNLOCK(mp); 982 983 if (error != 0) 984 goto end; 985 986 if (mp->mnt_opt != NULL) 987 vfs_freeopts(mp->mnt_opt); 988 mp->mnt_opt = mp->mnt_optnew; 989 (void)VFS_STATFS(mp, &mp->mnt_stat); 990 /* 991 * Prevent external consumers of mount options from reading 992 * mnt_optnew. 993 */ 994 mp->mnt_optnew = NULL; 995 996 if ((mp->mnt_flag & MNT_RDONLY) == 0) 997 vfs_allocate_syncvnode(mp); 998 else 999 vfs_deallocate_syncvnode(mp); 1000end: 1001 vfs_unbusy(mp); 1002 VI_LOCK(vp); 1003 vp->v_iflag &= ~VI_MOUNT; 1004 VI_UNLOCK(vp); 1005 vrele(vp); 1006 return (error); 1007} 1008 1009/* 1010 * vfs_domount(): actually attempt a filesystem mount. 1011 */ 1012static int 1013vfs_domount( 1014 struct thread *td, /* Calling thread. */ 1015 const char *fstype, /* Filesystem type. */ 1016 char *fspath, /* Mount path. */ 1017 int fsflags, /* Flags common to all filesystems. */ 1018 void *fsdata /* Options local to the filesystem. */ 1019 ) 1020{ 1021 struct vfsconf *vfsp; 1022 struct nameidata nd; 1023 struct vnode *vp; 1024 int error; 1025 1026 /* 1027 * Be ultra-paranoid about making sure the type and fspath 1028 * variables will fit in our mp buffers, including the 1029 * terminating NUL. 1030 */ 1031 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1032 return (ENAMETOOLONG); 1033 1034 if (jailed(td->td_ucred) || usermount == 0) { 1035 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1036 return (error); 1037 } 1038 1039 /* 1040 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1041 */ 1042 if (fsflags & MNT_EXPORTED) { 1043 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1044 if (error) 1045 return (error); 1046 } 1047 if (fsflags & MNT_SUIDDIR) { 1048 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1049 if (error) 1050 return (error); 1051 } 1052 /* 1053 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1054 */ 1055 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1056 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1057 fsflags |= MNT_NOSUID | MNT_USER; 1058 } 1059 1060 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1061 vfsp = NULL; 1062 if ((fsflags & MNT_UPDATE) == 0) { 1063 /* Don't try to load KLDs if we're mounting the root. */ 1064 if (fsflags & MNT_ROOTFS) 1065 vfsp = vfs_byname(fstype); 1066 else 1067 vfsp = vfs_byname_kld(fstype, td, &error); 1068 if (vfsp == NULL) 1069 return (ENODEV); 1070 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 1071 return (EPERM); 1072 } 1073 1074 /* 1075 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1076 */ 1077 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | MPSAFE | AUDITVNODE1, 1078 UIO_SYSSPACE, fspath, td); 1079 error = namei(&nd); 1080 if (error != 0) 1081 return (error); 1082 if (!NDHASGIANT(&nd)) 1083 mtx_lock(&Giant); 1084 NDFREE(&nd, NDF_ONLY_PNBUF); 1085 vp = nd.ni_vp; 1086 if ((fsflags & MNT_UPDATE) == 0) { 1087 error = vfs_domount_first(td, vfsp, fspath, vp, fsflags, 1088 fsdata); 1089 } else { 1090 error = vfs_domount_update(td, vp, fsflags, fsdata); 1091 } 1092 mtx_unlock(&Giant); 1093 1094 ASSERT_VI_UNLOCKED(vp, __func__); 1095 ASSERT_VOP_UNLOCKED(vp, __func__); 1096 1097 return (error); 1098} 1099 1100/* 1101 * Unmount a filesystem. 1102 * 1103 * Note: unmount takes a path to the vnode mounted on as argument, not 1104 * special file (as before). 1105 */ 1106#ifndef _SYS_SYSPROTO_H_ 1107struct unmount_args { 1108 char *path; 1109 int flags; 1110}; 1111#endif 1112/* ARGSUSED */ 1113int 1114unmount(td, uap) 1115 struct thread *td; 1116 register struct unmount_args /* { 1117 char *path; 1118 int flags; 1119 } */ *uap; 1120{ 1121 struct mount *mp; 1122 char *pathbuf; 1123 int error, id0, id1; 1124 1125 AUDIT_ARG_VALUE(uap->flags); 1126 if (jailed(td->td_ucred) || usermount == 0) { 1127 error = priv_check(td, PRIV_VFS_UNMOUNT); 1128 if (error) 1129 return (error); 1130 } 1131 1132 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1133 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1134 if (error) { 1135 free(pathbuf, M_TEMP); 1136 return (error); 1137 } 1138 mtx_lock(&Giant); 1139 if (uap->flags & MNT_BYFSID) { 1140 AUDIT_ARG_TEXT(pathbuf); 1141 /* Decode the filesystem ID. */ 1142 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1143 mtx_unlock(&Giant); 1144 free(pathbuf, M_TEMP); 1145 return (EINVAL); 1146 } 1147 1148 mtx_lock(&mountlist_mtx); 1149 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1150 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1151 mp->mnt_stat.f_fsid.val[1] == id1) 1152 break; 1153 } 1154 mtx_unlock(&mountlist_mtx); 1155 } else { 1156 AUDIT_ARG_UPATH1(td, pathbuf); 1157 mtx_lock(&mountlist_mtx); 1158 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1159 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1160 break; 1161 } 1162 mtx_unlock(&mountlist_mtx); 1163 } 1164 free(pathbuf, M_TEMP); 1165 if (mp == NULL) { 1166 /* 1167 * Previously we returned ENOENT for a nonexistent path and 1168 * EINVAL for a non-mountpoint. We cannot tell these apart 1169 * now, so in the !MNT_BYFSID case return the more likely 1170 * EINVAL for compatibility. 1171 */ 1172 mtx_unlock(&Giant); 1173 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1174 } 1175 1176 /* 1177 * Don't allow unmounting the root filesystem. 1178 */ 1179 if (mp->mnt_flag & MNT_ROOTFS) { 1180 mtx_unlock(&Giant); 1181 return (EINVAL); 1182 } 1183 error = dounmount(mp, uap->flags, td); 1184 mtx_unlock(&Giant); 1185 return (error); 1186} 1187 1188/* 1189 * Do the actual filesystem unmount. 1190 */ 1191int 1192dounmount(mp, flags, td) 1193 struct mount *mp; 1194 int flags; 1195 struct thread *td; 1196{ 1197 struct vnode *coveredvp, *fsrootvp; 1198 int error; 1199 int async_flag; 1200 int mnt_gen_r; 1201 1202 mtx_assert(&Giant, MA_OWNED); 1203 1204 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1205 mnt_gen_r = mp->mnt_gen; 1206 VI_LOCK(coveredvp); 1207 vholdl(coveredvp); 1208 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1209 vdrop(coveredvp); 1210 /* 1211 * Check for mp being unmounted while waiting for the 1212 * covered vnode lock. 1213 */ 1214 if (coveredvp->v_mountedhere != mp || 1215 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1216 VOP_UNLOCK(coveredvp, 0); 1217 return (EBUSY); 1218 } 1219 } 1220 /* 1221 * Only privileged root, or (if MNT_USER is set) the user that did the 1222 * original mount is permitted to unmount this filesystem. 1223 */ 1224 error = vfs_suser(mp, td); 1225 if (error) { 1226 if (coveredvp) 1227 VOP_UNLOCK(coveredvp, 0); 1228 return (error); 1229 } 1230 1231 MNT_ILOCK(mp); 1232 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1233 MNT_IUNLOCK(mp); 1234 if (coveredvp) 1235 VOP_UNLOCK(coveredvp, 0); 1236 return (EBUSY); 1237 } 1238 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1239 /* Allow filesystems to detect that a forced unmount is in progress. */ 1240 if (flags & MNT_FORCE) 1241 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1242 error = 0; 1243 if (mp->mnt_lockref) { 1244 if ((flags & MNT_FORCE) == 0) { 1245 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1246 MNTK_UNMOUNTF); 1247 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1248 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1249 wakeup(mp); 1250 } 1251 MNT_IUNLOCK(mp); 1252 if (coveredvp) 1253 VOP_UNLOCK(coveredvp, 0); 1254 return (EBUSY); 1255 } 1256 mp->mnt_kern_flag |= MNTK_DRAINING; 1257 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1258 "mount drain", 0); 1259 } 1260 MNT_IUNLOCK(mp); 1261 KASSERT(mp->mnt_lockref == 0, 1262 ("%s: invalid lock refcount in the drain path @ %s:%d", 1263 __func__, __FILE__, __LINE__)); 1264 KASSERT(error == 0, 1265 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1266 __func__, __FILE__, __LINE__)); 1267 vn_start_write(NULL, &mp, V_WAIT); 1268 1269 if (mp->mnt_flag & MNT_EXPUBLIC) 1270 vfs_setpublicfs(NULL, NULL, NULL); 1271 1272 vfs_msync(mp, MNT_WAIT); 1273 MNT_ILOCK(mp); 1274 async_flag = mp->mnt_flag & MNT_ASYNC; 1275 mp->mnt_flag &= ~MNT_ASYNC; 1276 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1277 MNT_IUNLOCK(mp); 1278 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1279 vfs_deallocate_syncvnode(mp); 1280 /* 1281 * For forced unmounts, move process cdir/rdir refs on the fs root 1282 * vnode to the covered vnode. For non-forced unmounts we want 1283 * such references to cause an EBUSY error. 1284 */ 1285 if ((flags & MNT_FORCE) && 1286 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1287 if (mp->mnt_vnodecovered != NULL) 1288 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1289 if (fsrootvp == rootvnode) { 1290 vrele(rootvnode); 1291 rootvnode = NULL; 1292 } 1293 vput(fsrootvp); 1294 } 1295 if (((mp->mnt_flag & MNT_RDONLY) || 1296 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1297 error = VFS_UNMOUNT(mp, flags); 1298 vn_finished_write(mp); 1299 /* 1300 * If we failed to flush the dirty blocks for this mount point, 1301 * undo all the cdir/rdir and rootvnode changes we made above. 1302 * Unless we failed to do so because the device is reporting that 1303 * it doesn't exist anymore. 1304 */ 1305 if (error && error != ENXIO) { 1306 if ((flags & MNT_FORCE) && 1307 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1308 if (mp->mnt_vnodecovered != NULL) 1309 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1310 if (rootvnode == NULL) { 1311 rootvnode = fsrootvp; 1312 vref(rootvnode); 1313 } 1314 vput(fsrootvp); 1315 } 1316 MNT_ILOCK(mp); 1317 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1318 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1319 MNT_IUNLOCK(mp); 1320 vfs_allocate_syncvnode(mp); 1321 MNT_ILOCK(mp); 1322 } 1323 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1324 mp->mnt_flag |= async_flag; 1325 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1326 mp->mnt_kern_flag |= MNTK_ASYNC; 1327 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1328 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1329 wakeup(mp); 1330 } 1331 MNT_IUNLOCK(mp); 1332 if (coveredvp) 1333 VOP_UNLOCK(coveredvp, 0); 1334 return (error); 1335 } 1336 mtx_lock(&mountlist_mtx); 1337 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1338 mtx_unlock(&mountlist_mtx); 1339 if (coveredvp != NULL) { 1340 coveredvp->v_mountedhere = NULL; 1341 vput(coveredvp); 1342 } 1343 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1344 vfs_mount_destroy(mp); 1345 return (0); 1346} 1347 1348/* 1349 * Report errors during filesystem mounting. 1350 */ 1351void 1352vfs_mount_error(struct mount *mp, const char *fmt, ...) 1353{ 1354 struct vfsoptlist *moptlist = mp->mnt_optnew; 1355 va_list ap; 1356 int error, len; 1357 char *errmsg; 1358 1359 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1360 if (error || errmsg == NULL || len <= 0) 1361 return; 1362 1363 va_start(ap, fmt); 1364 vsnprintf(errmsg, (size_t)len, fmt, ap); 1365 va_end(ap); 1366} 1367 1368void 1369vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1370{ 1371 va_list ap; 1372 int error, len; 1373 char *errmsg; 1374 1375 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1376 if (error || errmsg == NULL || len <= 0) 1377 return; 1378 1379 va_start(ap, fmt); 1380 vsnprintf(errmsg, (size_t)len, fmt, ap); 1381 va_end(ap); 1382} 1383 1384/* 1385 * --------------------------------------------------------------------- 1386 * Functions for querying mount options/arguments from filesystems. 1387 */ 1388 1389/* 1390 * Check that no unknown options are given 1391 */ 1392int 1393vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1394{ 1395 struct vfsopt *opt; 1396 char errmsg[255]; 1397 const char **t, *p, *q; 1398 int ret = 0; 1399 1400 TAILQ_FOREACH(opt, opts, link) { 1401 p = opt->name; 1402 q = NULL; 1403 if (p[0] == 'n' && p[1] == 'o') 1404 q = p + 2; 1405 for(t = global_opts; *t != NULL; t++) { 1406 if (strcmp(*t, p) == 0) 1407 break; 1408 if (q != NULL) { 1409 if (strcmp(*t, q) == 0) 1410 break; 1411 } 1412 } 1413 if (*t != NULL) 1414 continue; 1415 for(t = legal; *t != NULL; t++) { 1416 if (strcmp(*t, p) == 0) 1417 break; 1418 if (q != NULL) { 1419 if (strcmp(*t, q) == 0) 1420 break; 1421 } 1422 } 1423 if (*t != NULL) 1424 continue; 1425 snprintf(errmsg, sizeof(errmsg), 1426 "mount option <%s> is unknown", p); 1427 printf("%s\n", errmsg); 1428 ret = EINVAL; 1429 } 1430 if (ret != 0) { 1431 TAILQ_FOREACH(opt, opts, link) { 1432 if (strcmp(opt->name, "errmsg") == 0) { 1433 strncpy((char *)opt->value, errmsg, opt->len); 1434 } 1435 } 1436 } 1437 return (ret); 1438} 1439 1440/* 1441 * Get a mount option by its name. 1442 * 1443 * Return 0 if the option was found, ENOENT otherwise. 1444 * If len is non-NULL it will be filled with the length 1445 * of the option. If buf is non-NULL, it will be filled 1446 * with the address of the option. 1447 */ 1448int 1449vfs_getopt(opts, name, buf, len) 1450 struct vfsoptlist *opts; 1451 const char *name; 1452 void **buf; 1453 int *len; 1454{ 1455 struct vfsopt *opt; 1456 1457 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1458 1459 TAILQ_FOREACH(opt, opts, link) { 1460 if (strcmp(name, opt->name) == 0) { 1461 opt->seen = 1; 1462 if (len != NULL) 1463 *len = opt->len; 1464 if (buf != NULL) 1465 *buf = opt->value; 1466 return (0); 1467 } 1468 } 1469 return (ENOENT); 1470} 1471 1472int 1473vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1474{ 1475 struct vfsopt *opt; 1476 1477 if (opts == NULL) 1478 return (-1); 1479 1480 TAILQ_FOREACH(opt, opts, link) { 1481 if (strcmp(name, opt->name) == 0) { 1482 opt->seen = 1; 1483 return (opt->pos); 1484 } 1485 } 1486 return (-1); 1487} 1488 1489char * 1490vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1491{ 1492 struct vfsopt *opt; 1493 1494 *error = 0; 1495 TAILQ_FOREACH(opt, opts, link) { 1496 if (strcmp(name, opt->name) != 0) 1497 continue; 1498 opt->seen = 1; 1499 if (opt->len == 0 || 1500 ((char *)opt->value)[opt->len - 1] != '\0') { 1501 *error = EINVAL; 1502 return (NULL); 1503 } 1504 return (opt->value); 1505 } 1506 *error = ENOENT; 1507 return (NULL); 1508} 1509 1510int 1511vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1512{ 1513 struct vfsopt *opt; 1514 1515 TAILQ_FOREACH(opt, opts, link) { 1516 if (strcmp(name, opt->name) == 0) { 1517 opt->seen = 1; 1518 if (w != NULL) 1519 *w |= val; 1520 return (1); 1521 } 1522 } 1523 if (w != NULL) 1524 *w &= ~val; 1525 return (0); 1526} 1527 1528int 1529vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1530{ 1531 va_list ap; 1532 struct vfsopt *opt; 1533 int ret; 1534 1535 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1536 1537 TAILQ_FOREACH(opt, opts, link) { 1538 if (strcmp(name, opt->name) != 0) 1539 continue; 1540 opt->seen = 1; 1541 if (opt->len == 0 || opt->value == NULL) 1542 return (0); 1543 if (((char *)opt->value)[opt->len - 1] != '\0') 1544 return (0); 1545 va_start(ap, fmt); 1546 ret = vsscanf(opt->value, fmt, ap); 1547 va_end(ap); 1548 return (ret); 1549 } 1550 return (0); 1551} 1552 1553int 1554vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 1555{ 1556 struct vfsopt *opt; 1557 1558 TAILQ_FOREACH(opt, opts, link) { 1559 if (strcmp(name, opt->name) != 0) 1560 continue; 1561 opt->seen = 1; 1562 if (opt->value == NULL) 1563 opt->len = len; 1564 else { 1565 if (opt->len != len) 1566 return (EINVAL); 1567 bcopy(value, opt->value, len); 1568 } 1569 return (0); 1570 } 1571 return (ENOENT); 1572} 1573 1574int 1575vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 1576{ 1577 struct vfsopt *opt; 1578 1579 TAILQ_FOREACH(opt, opts, link) { 1580 if (strcmp(name, opt->name) != 0) 1581 continue; 1582 opt->seen = 1; 1583 if (opt->value == NULL) 1584 opt->len = len; 1585 else { 1586 if (opt->len < len) 1587 return (EINVAL); 1588 opt->len = len; 1589 bcopy(value, opt->value, len); 1590 } 1591 return (0); 1592 } 1593 return (ENOENT); 1594} 1595 1596int 1597vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 1598{ 1599 struct vfsopt *opt; 1600 1601 TAILQ_FOREACH(opt, opts, link) { 1602 if (strcmp(name, opt->name) != 0) 1603 continue; 1604 opt->seen = 1; 1605 if (opt->value == NULL) 1606 opt->len = strlen(value) + 1; 1607 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 1608 return (EINVAL); 1609 return (0); 1610 } 1611 return (ENOENT); 1612} 1613 1614/* 1615 * Find and copy a mount option. 1616 * 1617 * The size of the buffer has to be specified 1618 * in len, if it is not the same length as the 1619 * mount option, EINVAL is returned. 1620 * Returns ENOENT if the option is not found. 1621 */ 1622int 1623vfs_copyopt(opts, name, dest, len) 1624 struct vfsoptlist *opts; 1625 const char *name; 1626 void *dest; 1627 int len; 1628{ 1629 struct vfsopt *opt; 1630 1631 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1632 1633 TAILQ_FOREACH(opt, opts, link) { 1634 if (strcmp(name, opt->name) == 0) { 1635 opt->seen = 1; 1636 if (len != opt->len) 1637 return (EINVAL); 1638 bcopy(opt->value, dest, opt->len); 1639 return (0); 1640 } 1641 } 1642 return (ENOENT); 1643} 1644 1645/* 1646 * This is a helper function for filesystems to traverse their 1647 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1648 */ 1649 1650struct vnode * 1651__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 1652{ 1653 struct vnode *vp; 1654 1655 mtx_assert(MNT_MTX(mp), MA_OWNED); 1656 1657 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1658 if ((*mvp)->v_yield++ == 500) { 1659 MNT_IUNLOCK(mp); 1660 (*mvp)->v_yield = 0; 1661 uio_yield(); 1662 MNT_ILOCK(mp); 1663 } 1664 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 1665 while (vp != NULL && vp->v_type == VMARKER) 1666 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1667 1668 /* Check if we are done */ 1669 if (vp == NULL) { 1670 __mnt_vnode_markerfree(mvp, mp); 1671 return (NULL); 1672 } 1673 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1674 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1675 return (vp); 1676} 1677 1678struct vnode * 1679__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 1680{ 1681 struct vnode *vp; 1682 1683 mtx_assert(MNT_MTX(mp), MA_OWNED); 1684 1685 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1686 while (vp != NULL && vp->v_type == VMARKER) 1687 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1688 1689 /* Check if we are done */ 1690 if (vp == NULL) { 1691 *mvp = NULL; 1692 return (NULL); 1693 } 1694 MNT_REF(mp); 1695 MNT_IUNLOCK(mp); 1696 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 1697 M_VNODE_MARKER, 1698 M_WAITOK | M_ZERO); 1699 MNT_ILOCK(mp); 1700 (*mvp)->v_type = VMARKER; 1701 1702 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1703 while (vp != NULL && vp->v_type == VMARKER) 1704 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1705 1706 /* Check if we are done */ 1707 if (vp == NULL) { 1708 MNT_IUNLOCK(mp); 1709 free(*mvp, M_VNODE_MARKER); 1710 MNT_ILOCK(mp); 1711 *mvp = NULL; 1712 MNT_REL(mp); 1713 return (NULL); 1714 } 1715 (*mvp)->v_mount = mp; 1716 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1717 return (vp); 1718} 1719 1720 1721void 1722__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 1723{ 1724 1725 if (*mvp == NULL) 1726 return; 1727 1728 mtx_assert(MNT_MTX(mp), MA_OWNED); 1729 1730 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1731 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1732 MNT_IUNLOCK(mp); 1733 free(*mvp, M_VNODE_MARKER); 1734 MNT_ILOCK(mp); 1735 *mvp = NULL; 1736 MNT_REL(mp); 1737} 1738 1739 1740int 1741__vfs_statfs(struct mount *mp, struct statfs *sbp) 1742{ 1743 int error; 1744 1745 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 1746 if (sbp != &mp->mnt_stat) 1747 *sbp = mp->mnt_stat; 1748 return (error); 1749} 1750 1751void 1752vfs_mountedfrom(struct mount *mp, const char *from) 1753{ 1754 1755 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 1756 strlcpy(mp->mnt_stat.f_mntfromname, from, 1757 sizeof mp->mnt_stat.f_mntfromname); 1758} 1759 1760/* 1761 * --------------------------------------------------------------------- 1762 * This is the api for building mount args and mounting filesystems from 1763 * inside the kernel. 1764 * 1765 * The API works by accumulation of individual args. First error is 1766 * latched. 1767 * 1768 * XXX: should be documented in new manpage kernel_mount(9) 1769 */ 1770 1771/* A memory allocation which must be freed when we are done */ 1772struct mntaarg { 1773 SLIST_ENTRY(mntaarg) next; 1774}; 1775 1776/* The header for the mount arguments */ 1777struct mntarg { 1778 struct iovec *v; 1779 int len; 1780 int error; 1781 SLIST_HEAD(, mntaarg) list; 1782}; 1783 1784/* 1785 * Add a boolean argument. 1786 * 1787 * flag is the boolean value. 1788 * name must start with "no". 1789 */ 1790struct mntarg * 1791mount_argb(struct mntarg *ma, int flag, const char *name) 1792{ 1793 1794 KASSERT(name[0] == 'n' && name[1] == 'o', 1795 ("mount_argb(...,%s): name must start with 'no'", name)); 1796 1797 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 1798} 1799 1800/* 1801 * Add an argument printf style 1802 */ 1803struct mntarg * 1804mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 1805{ 1806 va_list ap; 1807 struct mntaarg *maa; 1808 struct sbuf *sb; 1809 int len; 1810 1811 if (ma == NULL) { 1812 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1813 SLIST_INIT(&ma->list); 1814 } 1815 if (ma->error) 1816 return (ma); 1817 1818 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1819 M_MOUNT, M_WAITOK); 1820 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1821 ma->v[ma->len].iov_len = strlen(name) + 1; 1822 ma->len++; 1823 1824 sb = sbuf_new_auto(); 1825 va_start(ap, fmt); 1826 sbuf_vprintf(sb, fmt, ap); 1827 va_end(ap); 1828 sbuf_finish(sb); 1829 len = sbuf_len(sb) + 1; 1830 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1831 SLIST_INSERT_HEAD(&ma->list, maa, next); 1832 bcopy(sbuf_data(sb), maa + 1, len); 1833 sbuf_delete(sb); 1834 1835 ma->v[ma->len].iov_base = maa + 1; 1836 ma->v[ma->len].iov_len = len; 1837 ma->len++; 1838 1839 return (ma); 1840} 1841 1842/* 1843 * Add an argument which is a userland string. 1844 */ 1845struct mntarg * 1846mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 1847{ 1848 struct mntaarg *maa; 1849 char *tbuf; 1850 1851 if (val == NULL) 1852 return (ma); 1853 if (ma == NULL) { 1854 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1855 SLIST_INIT(&ma->list); 1856 } 1857 if (ma->error) 1858 return (ma); 1859 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1860 SLIST_INSERT_HEAD(&ma->list, maa, next); 1861 tbuf = (void *)(maa + 1); 1862 ma->error = copyinstr(val, tbuf, len, NULL); 1863 return (mount_arg(ma, name, tbuf, -1)); 1864} 1865 1866/* 1867 * Plain argument. 1868 * 1869 * If length is -1, treat value as a C string. 1870 */ 1871struct mntarg * 1872mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 1873{ 1874 1875 if (ma == NULL) { 1876 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1877 SLIST_INIT(&ma->list); 1878 } 1879 if (ma->error) 1880 return (ma); 1881 1882 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1883 M_MOUNT, M_WAITOK); 1884 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1885 ma->v[ma->len].iov_len = strlen(name) + 1; 1886 ma->len++; 1887 1888 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 1889 if (len < 0) 1890 ma->v[ma->len].iov_len = strlen(val) + 1; 1891 else 1892 ma->v[ma->len].iov_len = len; 1893 ma->len++; 1894 return (ma); 1895} 1896 1897/* 1898 * Free a mntarg structure 1899 */ 1900static void 1901free_mntarg(struct mntarg *ma) 1902{ 1903 struct mntaarg *maa; 1904 1905 while (!SLIST_EMPTY(&ma->list)) { 1906 maa = SLIST_FIRST(&ma->list); 1907 SLIST_REMOVE_HEAD(&ma->list, next); 1908 free(maa, M_MOUNT); 1909 } 1910 free(ma->v, M_MOUNT); 1911 free(ma, M_MOUNT); 1912} 1913 1914/* 1915 * Mount a filesystem 1916 */ 1917int 1918kernel_mount(struct mntarg *ma, int flags) 1919{ 1920 struct uio auio; 1921 int error; 1922 1923 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 1924 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 1925 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 1926 1927 auio.uio_iov = ma->v; 1928 auio.uio_iovcnt = ma->len; 1929 auio.uio_segflg = UIO_SYSSPACE; 1930 1931 error = ma->error; 1932 if (!error) 1933 error = vfs_donmount(curthread, flags, &auio); 1934 free_mntarg(ma); 1935 return (error); 1936} 1937 1938/* 1939 * A printflike function to mount a filesystem. 1940 */ 1941int 1942kernel_vmount(int flags, ...) 1943{ 1944 struct mntarg *ma = NULL; 1945 va_list ap; 1946 const char *cp; 1947 const void *vp; 1948 int error; 1949 1950 va_start(ap, flags); 1951 for (;;) { 1952 cp = va_arg(ap, const char *); 1953 if (cp == NULL) 1954 break; 1955 vp = va_arg(ap, const void *); 1956 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 1957 } 1958 va_end(ap); 1959 1960 error = kernel_mount(ma, flags); 1961 return (error); 1962} 1963