1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1999-2004 Poul-Henning Kamp 5 * Copyright (c) 1999 Michael Smith 6 * Copyright (c) 1989, 1993 7 * The Regents of the University of California. All rights reserved. 8 * (c) UNIX System Laboratories, Inc. 9 * All or some portions of this file are derived from material licensed 10 * to the University of California by American Telephone and Telegraph 11 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 12 * the permission of UNIX System Laboratories, Inc. 13 * 14 * Redistribution and use in source and binary forms, with or without 15 * modification, are permitted provided that the following conditions 16 * are met: 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 3. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 */ 38 39 #include <sys/cdefs.h> 40 #include <sys/param.h> 41 #include <sys/conf.h> 42 #include <sys/smp.h> 43 #include <sys/devctl.h> 44 #include <sys/eventhandler.h> 45 #include <sys/fcntl.h> 46 #include <sys/jail.h> 47 #include <sys/kernel.h> 48 #include <sys/ktr.h> 49 #include <sys/libkern.h> 50 #include <sys/limits.h> 51 #include <sys/malloc.h> 52 #include <sys/mount.h> 53 #include <sys/mutex.h> 54 #include <sys/namei.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/filedesc.h> 58 #include <sys/reboot.h> 59 #include <sys/sbuf.h> 60 #include <sys/syscallsubr.h> 61 #include <sys/sysproto.h> 62 #include <sys/sx.h> 63 #include <sys/sysctl.h> 64 #include <sys/systm.h> 65 #include <sys/taskqueue.h> 66 #include <sys/vnode.h> 67 #include <vm/uma.h> 68 69 #include <geom/geom.h> 70 71 #include <machine/stdarg.h> 72 73 #include <security/audit/audit.h> 74 #include <security/mac/mac_framework.h> 75 76 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 77 78 static int vfs_domount(struct thread *td, const char *fstype, char *fspath, 79 uint64_t fsflags, bool jail_export, 80 struct vfsoptlist **optlist); 81 static void free_mntarg(struct mntarg *ma); 82 83 static int usermount = 0; 84 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 85 "Unprivileged users may mount and unmount file systems"); 86 87 static bool default_autoro = false; 88 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0, 89 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified"); 90 91 static bool recursive_forced_unmount = false; 92 SYSCTL_BOOL(_vfs, OID_AUTO, recursive_forced_unmount, CTLFLAG_RW, 93 &recursive_forced_unmount, 0, "Recursively unmount stacked upper mounts" 94 " when a file system is forcibly unmounted"); 95 96 static SYSCTL_NODE(_vfs, OID_AUTO, deferred_unmount, 97 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "deferred unmount controls"); 98 99 static unsigned int deferred_unmount_retry_limit = 10; 100 SYSCTL_UINT(_vfs_deferred_unmount, OID_AUTO, retry_limit, CTLFLAG_RW, 101 &deferred_unmount_retry_limit, 0, 102 "Maximum number of retries for deferred unmount failure"); 103 104 static int deferred_unmount_retry_delay_hz; 105 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, retry_delay_hz, CTLFLAG_RW, 106 &deferred_unmount_retry_delay_hz, 0, 107 "Delay in units of [1/kern.hz]s when retrying a failed deferred unmount"); 108 109 static int deferred_unmount_total_retries = 0; 110 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, total_retries, CTLFLAG_RD, 111 &deferred_unmount_total_retries, 0, 112 "Total number of retried deferred unmounts"); 113 114 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 115 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure"); 116 static uma_zone_t mount_zone; 117 118 /* List of mounted filesystems. */ 119 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 120 121 /* For any iteration/modification of mountlist */ 122 struct mtx_padalign __exclusive_cache_line mountlist_mtx; 123 124 EVENTHANDLER_LIST_DEFINE(vfs_mounted); 125 EVENTHANDLER_LIST_DEFINE(vfs_unmounted); 126 127 static void vfs_deferred_unmount(void *arg, int pending); 128 static struct timeout_task deferred_unmount_task; 129 static struct mtx deferred_unmount_lock; 130 MTX_SYSINIT(deferred_unmount, &deferred_unmount_lock, "deferred_unmount", 131 MTX_DEF); 132 static STAILQ_HEAD(, mount) deferred_unmount_list = 133 STAILQ_HEAD_INITIALIZER(deferred_unmount_list); 134 TASKQUEUE_DEFINE_THREAD(deferred_unmount); 135 136 static void mount_devctl_event(const char *type, struct mount *mp, bool donew); 137 138 /* 139 * Global opts, taken by all filesystems 140 */ 141 static const char *global_opts[] = { 142 "errmsg", 143 "fstype", 144 "fspath", 145 "ro", 146 "rw", 147 "nosuid", 148 "noexec", 149 NULL 150 }; 151 152 static int 153 mount_init(void *mem, int size, int flags) 154 { 155 struct mount *mp; 156 157 mp = (struct mount *)mem; 158 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 159 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF); 160 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 161 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO); 162 mp->mnt_ref = 0; 163 mp->mnt_vfs_ops = 1; 164 mp->mnt_rootvnode = NULL; 165 return (0); 166 } 167 168 static void 169 mount_fini(void *mem, int size) 170 { 171 struct mount *mp; 172 173 mp = (struct mount *)mem; 174 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu); 175 lockdestroy(&mp->mnt_explock); 176 mtx_destroy(&mp->mnt_listmtx); 177 mtx_destroy(&mp->mnt_mtx); 178 } 179 180 static void 181 vfs_mount_init(void *dummy __unused) 182 { 183 TIMEOUT_TASK_INIT(taskqueue_deferred_unmount, &deferred_unmount_task, 184 0, vfs_deferred_unmount, NULL); 185 deferred_unmount_retry_delay_hz = hz; 186 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL, 187 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE); 188 mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF); 189 } 190 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL); 191 192 /* 193 * --------------------------------------------------------------------- 194 * Functions for building and sanitizing the mount options 195 */ 196 197 /* Remove one mount option. */ 198 static void 199 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 200 { 201 202 TAILQ_REMOVE(opts, opt, link); 203 free(opt->name, M_MOUNT); 204 if (opt->value != NULL) 205 free(opt->value, M_MOUNT); 206 free(opt, M_MOUNT); 207 } 208 209 /* Release all resources related to the mount options. */ 210 void 211 vfs_freeopts(struct vfsoptlist *opts) 212 { 213 struct vfsopt *opt; 214 215 while (!TAILQ_EMPTY(opts)) { 216 opt = TAILQ_FIRST(opts); 217 vfs_freeopt(opts, opt); 218 } 219 free(opts, M_MOUNT); 220 } 221 222 void 223 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 224 { 225 struct vfsopt *opt, *temp; 226 227 if (opts == NULL) 228 return; 229 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 230 if (strcmp(opt->name, name) == 0) 231 vfs_freeopt(opts, opt); 232 } 233 } 234 235 static int 236 vfs_isopt_ro(const char *opt) 237 { 238 239 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 || 240 strcmp(opt, "norw") == 0) 241 return (1); 242 return (0); 243 } 244 245 static int 246 vfs_isopt_rw(const char *opt) 247 { 248 249 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0) 250 return (1); 251 return (0); 252 } 253 254 /* 255 * Check if options are equal (with or without the "no" prefix). 256 */ 257 static int 258 vfs_equalopts(const char *opt1, const char *opt2) 259 { 260 char *p; 261 262 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 263 if (strcmp(opt1, opt2) == 0) 264 return (1); 265 /* "noopt" vs. "opt" */ 266 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 267 return (1); 268 /* "opt" vs. "noopt" */ 269 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 270 return (1); 271 while ((p = strchr(opt1, '.')) != NULL && 272 !strncmp(opt1, opt2, ++p - opt1)) { 273 opt2 += p - opt1; 274 opt1 = p; 275 /* "foo.noopt" vs. "foo.opt" */ 276 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 277 return (1); 278 /* "foo.opt" vs. "foo.noopt" */ 279 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 280 return (1); 281 } 282 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */ 283 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) && 284 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2))) 285 return (1); 286 return (0); 287 } 288 289 /* 290 * If a mount option is specified several times, 291 * (with or without the "no" prefix) only keep 292 * the last occurrence of it. 293 */ 294 static void 295 vfs_sanitizeopts(struct vfsoptlist *opts) 296 { 297 struct vfsopt *opt, *opt2, *tmp; 298 299 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 300 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 301 while (opt2 != NULL) { 302 if (vfs_equalopts(opt->name, opt2->name)) { 303 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 304 vfs_freeopt(opts, opt2); 305 opt2 = tmp; 306 } else { 307 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 308 } 309 } 310 } 311 } 312 313 /* 314 * Build a linked list of mount options from a struct uio. 315 */ 316 int 317 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 318 { 319 struct vfsoptlist *opts; 320 struct vfsopt *opt; 321 size_t memused, namelen, optlen; 322 unsigned int i, iovcnt; 323 int error; 324 325 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 326 TAILQ_INIT(opts); 327 memused = 0; 328 iovcnt = auio->uio_iovcnt; 329 for (i = 0; i < iovcnt; i += 2) { 330 namelen = auio->uio_iov[i].iov_len; 331 optlen = auio->uio_iov[i + 1].iov_len; 332 memused += sizeof(struct vfsopt) + optlen + namelen; 333 /* 334 * Avoid consuming too much memory, and attempts to overflow 335 * memused. 336 */ 337 if (memused > VFS_MOUNTARG_SIZE_MAX || 338 optlen > VFS_MOUNTARG_SIZE_MAX || 339 namelen > VFS_MOUNTARG_SIZE_MAX) { 340 error = EINVAL; 341 goto bad; 342 } 343 344 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 345 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 346 opt->value = NULL; 347 opt->len = 0; 348 opt->pos = i / 2; 349 opt->seen = 0; 350 351 /* 352 * Do this early, so jumps to "bad" will free the current 353 * option. 354 */ 355 TAILQ_INSERT_TAIL(opts, opt, link); 356 357 if (auio->uio_segflg == UIO_SYSSPACE) { 358 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 359 } else { 360 error = copyin(auio->uio_iov[i].iov_base, opt->name, 361 namelen); 362 if (error) 363 goto bad; 364 } 365 /* Ensure names are null-terminated strings. */ 366 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 367 error = EINVAL; 368 goto bad; 369 } 370 if (optlen != 0) { 371 opt->len = optlen; 372 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 373 if (auio->uio_segflg == UIO_SYSSPACE) { 374 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 375 optlen); 376 } else { 377 error = copyin(auio->uio_iov[i + 1].iov_base, 378 opt->value, optlen); 379 if (error) 380 goto bad; 381 } 382 } 383 } 384 vfs_sanitizeopts(opts); 385 *options = opts; 386 return (0); 387 bad: 388 vfs_freeopts(opts); 389 return (error); 390 } 391 392 /* 393 * Merge the old mount options with the new ones passed 394 * in the MNT_UPDATE case. 395 * 396 * XXX: This function will keep a "nofoo" option in the new 397 * options. E.g, if the option's canonical name is "foo", 398 * "nofoo" ends up in the mount point's active options. 399 */ 400 static void 401 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts) 402 { 403 struct vfsopt *opt, *new; 404 405 TAILQ_FOREACH(opt, oldopts, link) { 406 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 407 new->name = strdup(opt->name, M_MOUNT); 408 if (opt->len != 0) { 409 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 410 bcopy(opt->value, new->value, opt->len); 411 } else 412 new->value = NULL; 413 new->len = opt->len; 414 new->seen = opt->seen; 415 TAILQ_INSERT_HEAD(toopts, new, link); 416 } 417 vfs_sanitizeopts(toopts); 418 } 419 420 /* 421 * Mount a filesystem. 422 */ 423 #ifndef _SYS_SYSPROTO_H_ 424 struct nmount_args { 425 struct iovec *iovp; 426 unsigned int iovcnt; 427 int flags; 428 }; 429 #endif 430 int 431 sys_nmount(struct thread *td, struct nmount_args *uap) 432 { 433 struct uio *auio; 434 int error; 435 u_int iovcnt; 436 uint64_t flags; 437 438 /* 439 * Mount flags are now 64-bits. On 32-bit archtectures only 440 * 32-bits are passed in, but from here on everything handles 441 * 64-bit flags correctly. 442 */ 443 flags = uap->flags; 444 445 AUDIT_ARG_FFLAGS(flags); 446 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 447 uap->iovp, uap->iovcnt, flags); 448 449 /* 450 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 451 * userspace to set this flag, but we must filter it out if we want 452 * MNT_UPDATE on the root file system to work. 453 * MNT_ROOTFS should only be set by the kernel when mounting its 454 * root file system. 455 */ 456 flags &= ~MNT_ROOTFS; 457 458 iovcnt = uap->iovcnt; 459 /* 460 * Check that we have an even number of iovec's 461 * and that we have at least two options. 462 */ 463 if ((iovcnt & 1) || (iovcnt < 4)) { 464 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 465 uap->iovcnt); 466 return (EINVAL); 467 } 468 469 error = copyinuio(uap->iovp, iovcnt, &auio); 470 if (error) { 471 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 472 __func__, error); 473 return (error); 474 } 475 error = vfs_donmount(td, flags, auio); 476 477 free(auio, M_IOV); 478 return (error); 479 } 480 481 /* 482 * --------------------------------------------------------------------- 483 * Various utility functions 484 */ 485 486 /* 487 * Get a reference on a mount point from a vnode. 488 * 489 * The vnode is allowed to be passed unlocked and race against dooming. Note in 490 * such case there are no guarantees the referenced mount point will still be 491 * associated with it after the function returns. 492 */ 493 struct mount * 494 vfs_ref_from_vp(struct vnode *vp) 495 { 496 struct mount *mp; 497 struct mount_pcpu *mpcpu; 498 499 mp = atomic_load_ptr(&vp->v_mount); 500 if (__predict_false(mp == NULL)) { 501 return (mp); 502 } 503 if (vfs_op_thread_enter(mp, mpcpu)) { 504 if (__predict_true(mp == vp->v_mount)) { 505 vfs_mp_count_add_pcpu(mpcpu, ref, 1); 506 vfs_op_thread_exit(mp, mpcpu); 507 } else { 508 vfs_op_thread_exit(mp, mpcpu); 509 mp = NULL; 510 } 511 } else { 512 MNT_ILOCK(mp); 513 if (mp == vp->v_mount) { 514 MNT_REF(mp); 515 MNT_IUNLOCK(mp); 516 } else { 517 MNT_IUNLOCK(mp); 518 mp = NULL; 519 } 520 } 521 return (mp); 522 } 523 524 void 525 vfs_ref(struct mount *mp) 526 { 527 struct mount_pcpu *mpcpu; 528 529 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 530 if (vfs_op_thread_enter(mp, mpcpu)) { 531 vfs_mp_count_add_pcpu(mpcpu, ref, 1); 532 vfs_op_thread_exit(mp, mpcpu); 533 return; 534 } 535 536 MNT_ILOCK(mp); 537 MNT_REF(mp); 538 MNT_IUNLOCK(mp); 539 } 540 541 /* 542 * Register ump as an upper mount of the mount associated with 543 * vnode vp. This registration will be tracked through 544 * mount_upper_node upper, which should be allocated by the 545 * caller and stored in per-mount data associated with mp. 546 * 547 * If successful, this function will return the mount associated 548 * with vp, and will ensure that it cannot be unmounted until 549 * ump has been unregistered as one of its upper mounts. 550 * 551 * Upon failure this function will return NULL. 552 */ 553 struct mount * 554 vfs_register_upper_from_vp(struct vnode *vp, struct mount *ump, 555 struct mount_upper_node *upper) 556 { 557 struct mount *mp; 558 559 mp = atomic_load_ptr(&vp->v_mount); 560 if (mp == NULL) 561 return (NULL); 562 MNT_ILOCK(mp); 563 if (mp != vp->v_mount || 564 ((mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_RECURSE)) != 0)) { 565 MNT_IUNLOCK(mp); 566 return (NULL); 567 } 568 KASSERT(ump != mp, ("upper and lower mounts are identical")); 569 upper->mp = ump; 570 MNT_REF(mp); 571 TAILQ_INSERT_TAIL(&mp->mnt_uppers, upper, mnt_upper_link); 572 MNT_IUNLOCK(mp); 573 return (mp); 574 } 575 576 /* 577 * Register upper mount ump to receive vnode unlink/reclaim 578 * notifications from lower mount mp. This registration will 579 * be tracked through mount_upper_node upper, which should be 580 * allocated by the caller and stored in per-mount data 581 * associated with mp. 582 * 583 * ump must already be registered as an upper mount of mp 584 * through a call to vfs_register_upper_from_vp(). 585 */ 586 void 587 vfs_register_for_notification(struct mount *mp, struct mount *ump, 588 struct mount_upper_node *upper) 589 { 590 upper->mp = ump; 591 MNT_ILOCK(mp); 592 TAILQ_INSERT_TAIL(&mp->mnt_notify, upper, mnt_upper_link); 593 MNT_IUNLOCK(mp); 594 } 595 596 static void 597 vfs_drain_upper_locked(struct mount *mp) 598 { 599 mtx_assert(MNT_MTX(mp), MA_OWNED); 600 while (mp->mnt_upper_pending != 0) { 601 mp->mnt_kern_flag |= MNTK_UPPER_WAITER; 602 msleep(&mp->mnt_uppers, MNT_MTX(mp), 0, "mntupw", 0); 603 } 604 } 605 606 /* 607 * Undo a previous call to vfs_register_for_notification(). 608 * The mount represented by upper must be currently registered 609 * as an upper mount for mp. 610 */ 611 void 612 vfs_unregister_for_notification(struct mount *mp, 613 struct mount_upper_node *upper) 614 { 615 MNT_ILOCK(mp); 616 vfs_drain_upper_locked(mp); 617 TAILQ_REMOVE(&mp->mnt_notify, upper, mnt_upper_link); 618 MNT_IUNLOCK(mp); 619 } 620 621 /* 622 * Undo a previous call to vfs_register_upper_from_vp(). 623 * This must be done before mp can be unmounted. 624 */ 625 void 626 vfs_unregister_upper(struct mount *mp, struct mount_upper_node *upper) 627 { 628 MNT_ILOCK(mp); 629 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0, 630 ("registered upper with pending unmount")); 631 vfs_drain_upper_locked(mp); 632 TAILQ_REMOVE(&mp->mnt_uppers, upper, mnt_upper_link); 633 if ((mp->mnt_kern_flag & MNTK_TASKQUEUE_WAITER) != 0 && 634 TAILQ_EMPTY(&mp->mnt_uppers)) { 635 mp->mnt_kern_flag &= ~MNTK_TASKQUEUE_WAITER; 636 wakeup(&mp->mnt_taskqueue_link); 637 } 638 MNT_REL(mp); 639 MNT_IUNLOCK(mp); 640 } 641 642 void 643 vfs_rel(struct mount *mp) 644 { 645 struct mount_pcpu *mpcpu; 646 647 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 648 if (vfs_op_thread_enter(mp, mpcpu)) { 649 vfs_mp_count_sub_pcpu(mpcpu, ref, 1); 650 vfs_op_thread_exit(mp, mpcpu); 651 return; 652 } 653 654 MNT_ILOCK(mp); 655 MNT_REL(mp); 656 MNT_IUNLOCK(mp); 657 } 658 659 /* 660 * Allocate and initialize the mount point struct. 661 */ 662 struct mount * 663 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 664 struct ucred *cred) 665 { 666 struct mount *mp; 667 668 mp = uma_zalloc(mount_zone, M_WAITOK); 669 bzero(&mp->mnt_startzero, 670 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 671 mp->mnt_kern_flag = 0; 672 mp->mnt_flag = 0; 673 mp->mnt_rootvnode = NULL; 674 mp->mnt_vnodecovered = NULL; 675 mp->mnt_op = NULL; 676 mp->mnt_vfc = NULL; 677 TAILQ_INIT(&mp->mnt_nvnodelist); 678 mp->mnt_nvnodelistsize = 0; 679 TAILQ_INIT(&mp->mnt_lazyvnodelist); 680 mp->mnt_lazyvnodelistsize = 0; 681 MPPASS(mp->mnt_ref == 0 && mp->mnt_lockref == 0 && 682 mp->mnt_writeopcount == 0, mp); 683 MPASSERT(mp->mnt_vfs_ops == 1, mp, 684 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops)); 685 (void) vfs_busy(mp, MBF_NOWAIT); 686 atomic_add_acq_int(&vfsp->vfc_refcount, 1); 687 mp->mnt_op = vfsp->vfc_vfsops; 688 mp->mnt_vfc = vfsp; 689 mp->mnt_stat.f_type = vfsp->vfc_typenum; 690 mp->mnt_gen++; 691 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 692 mp->mnt_vnodecovered = vp; 693 mp->mnt_cred = crdup(cred); 694 mp->mnt_stat.f_owner = cred->cr_uid; 695 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 696 mp->mnt_iosize_max = DFLTPHYS; 697 #ifdef MAC 698 mac_mount_init(mp); 699 mac_mount_create(cred, mp); 700 #endif 701 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 702 mp->mnt_upper_pending = 0; 703 TAILQ_INIT(&mp->mnt_uppers); 704 TAILQ_INIT(&mp->mnt_notify); 705 mp->mnt_taskqueue_flags = 0; 706 mp->mnt_unmount_retries = 0; 707 return (mp); 708 } 709 710 /* 711 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 712 */ 713 void 714 vfs_mount_destroy(struct mount *mp) 715 { 716 717 MPPASS(mp->mnt_vfs_ops != 0, mp); 718 719 vfs_assert_mount_counters(mp); 720 721 MNT_ILOCK(mp); 722 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 723 if (mp->mnt_kern_flag & MNTK_MWAIT) { 724 mp->mnt_kern_flag &= ~MNTK_MWAIT; 725 wakeup(mp); 726 } 727 while (mp->mnt_ref) 728 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 729 KASSERT(mp->mnt_ref == 0, 730 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 731 __FILE__, __LINE__)); 732 MPPASS(mp->mnt_writeopcount == 0, mp); 733 MPPASS(mp->mnt_secondary_writes == 0, mp); 734 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1); 735 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 736 struct vnode *vp; 737 738 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 739 vn_printf(vp, "dangling vnode "); 740 panic("unmount: dangling vnode"); 741 } 742 KASSERT(mp->mnt_upper_pending == 0, ("mnt_upper_pending")); 743 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers")); 744 KASSERT(TAILQ_EMPTY(&mp->mnt_notify), ("mnt_notify")); 745 MPPASS(mp->mnt_nvnodelistsize == 0, mp); 746 MPPASS(mp->mnt_lazyvnodelistsize == 0, mp); 747 MPPASS(mp->mnt_lockref == 0, mp); 748 MNT_IUNLOCK(mp); 749 750 MPASSERT(mp->mnt_vfs_ops == 1, mp, 751 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops)); 752 753 MPASSERT(mp->mnt_rootvnode == NULL, mp, 754 ("mount point still has a root vnode %p", mp->mnt_rootvnode)); 755 756 if (mp->mnt_vnodecovered != NULL) 757 vrele(mp->mnt_vnodecovered); 758 #ifdef MAC 759 mac_mount_destroy(mp); 760 #endif 761 if (mp->mnt_opt != NULL) 762 vfs_freeopts(mp->mnt_opt); 763 if (mp->mnt_exjail != NULL) { 764 atomic_subtract_int(&mp->mnt_exjail->cr_prison->pr_exportcnt, 765 1); 766 crfree(mp->mnt_exjail); 767 } 768 if (mp->mnt_export != NULL) { 769 vfs_free_addrlist(mp->mnt_export); 770 free(mp->mnt_export, M_MOUNT); 771 } 772 crfree(mp->mnt_cred); 773 uma_zfree(mount_zone, mp); 774 } 775 776 static bool 777 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error) 778 { 779 /* This is an upgrade of an exisiting mount. */ 780 if ((fsflags & MNT_UPDATE) != 0) 781 return (false); 782 /* This is already an R/O mount. */ 783 if ((fsflags & MNT_RDONLY) != 0) 784 return (false); 785 786 switch (error) { 787 case ENODEV: /* generic, geom, ... */ 788 case EACCES: /* cam/scsi, ... */ 789 case EROFS: /* md, mmcsd, ... */ 790 /* 791 * These errors can be returned by the storage layer to signal 792 * that the media is read-only. No harm in the R/O mount 793 * attempt if the error was returned for some other reason. 794 */ 795 return (true); 796 default: 797 return (false); 798 } 799 } 800 801 int 802 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions) 803 { 804 struct vfsoptlist *optlist; 805 struct vfsopt *opt, *tmp_opt; 806 char *fstype, *fspath, *errmsg; 807 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 808 bool autoro, has_nonexport, jail_export; 809 810 errmsg = fspath = NULL; 811 errmsg_len = fspathlen = 0; 812 errmsg_pos = -1; 813 autoro = default_autoro; 814 815 error = vfs_buildopts(fsoptions, &optlist); 816 if (error) 817 return (error); 818 819 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 820 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 821 822 /* 823 * We need these two options before the others, 824 * and they are mandatory for any filesystem. 825 * Ensure they are NUL terminated as well. 826 */ 827 fstypelen = 0; 828 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 829 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') { 830 error = EINVAL; 831 if (errmsg != NULL) 832 strncpy(errmsg, "Invalid fstype", errmsg_len); 833 goto bail; 834 } 835 fspathlen = 0; 836 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 837 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') { 838 error = EINVAL; 839 if (errmsg != NULL) 840 strncpy(errmsg, "Invalid fspath", errmsg_len); 841 goto bail; 842 } 843 844 /* 845 * Check to see that "export" is only used with the "update", "fstype", 846 * "fspath", "from" and "errmsg" options when in a vnet jail. 847 * These are the ones used to set/update exports by mountd(8). 848 * If only the above options are set in a jail that can run mountd(8), 849 * then the jail_export argument of vfs_domount() will be true. 850 * When jail_export is true, the vfs_suser() check does not cause 851 * failure, but limits the update to exports only. 852 * This allows mountd(8) running within the vnet jail 853 * to export file systems visible within the jail, but 854 * mounted outside of the jail. 855 */ 856 /* 857 * We need to see if we have the "update" option 858 * before we call vfs_domount(), since vfs_domount() has special 859 * logic based on MNT_UPDATE. This is very important 860 * when we want to update the root filesystem. 861 */ 862 has_nonexport = false; 863 jail_export = false; 864 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 865 int do_freeopt = 0; 866 867 if (jailed(td->td_ucred) && 868 strcmp(opt->name, "export") != 0 && 869 strcmp(opt->name, "update") != 0 && 870 strcmp(opt->name, "fstype") != 0 && 871 strcmp(opt->name, "fspath") != 0 && 872 strcmp(opt->name, "from") != 0 && 873 strcmp(opt->name, "errmsg") != 0) 874 has_nonexport = true; 875 if (strcmp(opt->name, "update") == 0) { 876 fsflags |= MNT_UPDATE; 877 do_freeopt = 1; 878 } 879 else if (strcmp(opt->name, "async") == 0) 880 fsflags |= MNT_ASYNC; 881 else if (strcmp(opt->name, "force") == 0) { 882 fsflags |= MNT_FORCE; 883 do_freeopt = 1; 884 } 885 else if (strcmp(opt->name, "reload") == 0) { 886 fsflags |= MNT_RELOAD; 887 do_freeopt = 1; 888 } 889 else if (strcmp(opt->name, "multilabel") == 0) 890 fsflags |= MNT_MULTILABEL; 891 else if (strcmp(opt->name, "noasync") == 0) 892 fsflags &= ~MNT_ASYNC; 893 else if (strcmp(opt->name, "noatime") == 0) 894 fsflags |= MNT_NOATIME; 895 else if (strcmp(opt->name, "atime") == 0) { 896 free(opt->name, M_MOUNT); 897 opt->name = strdup("nonoatime", M_MOUNT); 898 } 899 else if (strcmp(opt->name, "noclusterr") == 0) 900 fsflags |= MNT_NOCLUSTERR; 901 else if (strcmp(opt->name, "clusterr") == 0) { 902 free(opt->name, M_MOUNT); 903 opt->name = strdup("nonoclusterr", M_MOUNT); 904 } 905 else if (strcmp(opt->name, "noclusterw") == 0) 906 fsflags |= MNT_NOCLUSTERW; 907 else if (strcmp(opt->name, "clusterw") == 0) { 908 free(opt->name, M_MOUNT); 909 opt->name = strdup("nonoclusterw", M_MOUNT); 910 } 911 else if (strcmp(opt->name, "noexec") == 0) 912 fsflags |= MNT_NOEXEC; 913 else if (strcmp(opt->name, "exec") == 0) { 914 free(opt->name, M_MOUNT); 915 opt->name = strdup("nonoexec", M_MOUNT); 916 } 917 else if (strcmp(opt->name, "nosuid") == 0) 918 fsflags |= MNT_NOSUID; 919 else if (strcmp(opt->name, "suid") == 0) { 920 free(opt->name, M_MOUNT); 921 opt->name = strdup("nonosuid", M_MOUNT); 922 } 923 else if (strcmp(opt->name, "nosymfollow") == 0) 924 fsflags |= MNT_NOSYMFOLLOW; 925 else if (strcmp(opt->name, "symfollow") == 0) { 926 free(opt->name, M_MOUNT); 927 opt->name = strdup("nonosymfollow", M_MOUNT); 928 } 929 else if (strcmp(opt->name, "noro") == 0) { 930 fsflags &= ~MNT_RDONLY; 931 autoro = false; 932 } 933 else if (strcmp(opt->name, "rw") == 0) { 934 fsflags &= ~MNT_RDONLY; 935 autoro = false; 936 } 937 else if (strcmp(opt->name, "ro") == 0) { 938 fsflags |= MNT_RDONLY; 939 autoro = false; 940 } 941 else if (strcmp(opt->name, "rdonly") == 0) { 942 free(opt->name, M_MOUNT); 943 opt->name = strdup("ro", M_MOUNT); 944 fsflags |= MNT_RDONLY; 945 autoro = false; 946 } 947 else if (strcmp(opt->name, "autoro") == 0) { 948 do_freeopt = 1; 949 autoro = true; 950 } 951 else if (strcmp(opt->name, "suiddir") == 0) 952 fsflags |= MNT_SUIDDIR; 953 else if (strcmp(opt->name, "sync") == 0) 954 fsflags |= MNT_SYNCHRONOUS; 955 else if (strcmp(opt->name, "union") == 0) 956 fsflags |= MNT_UNION; 957 else if (strcmp(opt->name, "export") == 0) { 958 fsflags |= MNT_EXPORTED; 959 jail_export = true; 960 } else if (strcmp(opt->name, "automounted") == 0) { 961 fsflags |= MNT_AUTOMOUNTED; 962 do_freeopt = 1; 963 } else if (strcmp(opt->name, "nocover") == 0) { 964 fsflags |= MNT_NOCOVER; 965 do_freeopt = 1; 966 } else if (strcmp(opt->name, "cover") == 0) { 967 fsflags &= ~MNT_NOCOVER; 968 do_freeopt = 1; 969 } else if (strcmp(opt->name, "emptydir") == 0) { 970 fsflags |= MNT_EMPTYDIR; 971 do_freeopt = 1; 972 } else if (strcmp(opt->name, "noemptydir") == 0) { 973 fsflags &= ~MNT_EMPTYDIR; 974 do_freeopt = 1; 975 } 976 if (do_freeopt) 977 vfs_freeopt(optlist, opt); 978 } 979 980 /* 981 * Be ultra-paranoid about making sure the type and fspath 982 * variables will fit in our mp buffers, including the 983 * terminating NUL. 984 */ 985 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) { 986 error = ENAMETOOLONG; 987 goto bail; 988 } 989 990 /* 991 * If has_nonexport is true or the caller is not running within a 992 * vnet prison that can run mountd(8), set jail_export false. 993 */ 994 if (has_nonexport || !jailed(td->td_ucred) || 995 !prison_check_nfsd(td->td_ucred)) 996 jail_export = false; 997 998 error = vfs_domount(td, fstype, fspath, fsflags, jail_export, &optlist); 999 if (error == ENOENT) { 1000 error = EINVAL; 1001 if (errmsg != NULL) 1002 strncpy(errmsg, "Invalid fstype", errmsg_len); 1003 goto bail; 1004 } 1005 1006 /* 1007 * See if we can mount in the read-only mode if the error code suggests 1008 * that it could be possible and the mount options allow for that. 1009 * Never try it if "[no]{ro|rw}" has been explicitly requested and not 1010 * overridden by "autoro". 1011 */ 1012 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) { 1013 printf("%s: R/W mount failed, possibly R/O media," 1014 " trying R/O mount\n", __func__); 1015 fsflags |= MNT_RDONLY; 1016 error = vfs_domount(td, fstype, fspath, fsflags, jail_export, 1017 &optlist); 1018 } 1019 bail: 1020 /* copyout the errmsg */ 1021 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 1022 && errmsg_len > 0 && errmsg != NULL) { 1023 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 1024 bcopy(errmsg, 1025 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 1026 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 1027 } else { 1028 copyout(errmsg, 1029 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 1030 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 1031 } 1032 } 1033 1034 if (optlist != NULL) 1035 vfs_freeopts(optlist); 1036 return (error); 1037 } 1038 1039 /* 1040 * Old mount API. 1041 */ 1042 #ifndef _SYS_SYSPROTO_H_ 1043 struct mount_args { 1044 char *type; 1045 char *path; 1046 int flags; 1047 caddr_t data; 1048 }; 1049 #endif 1050 /* ARGSUSED */ 1051 int 1052 sys_mount(struct thread *td, struct mount_args *uap) 1053 { 1054 char *fstype; 1055 struct vfsconf *vfsp = NULL; 1056 struct mntarg *ma = NULL; 1057 uint64_t flags; 1058 int error; 1059 1060 /* 1061 * Mount flags are now 64-bits. On 32-bit architectures only 1062 * 32-bits are passed in, but from here on everything handles 1063 * 64-bit flags correctly. 1064 */ 1065 flags = uap->flags; 1066 1067 AUDIT_ARG_FFLAGS(flags); 1068 1069 /* 1070 * Filter out MNT_ROOTFS. We do not want clients of mount() in 1071 * userspace to set this flag, but we must filter it out if we want 1072 * MNT_UPDATE on the root file system to work. 1073 * MNT_ROOTFS should only be set by the kernel when mounting its 1074 * root file system. 1075 */ 1076 flags &= ~MNT_ROOTFS; 1077 1078 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 1079 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 1080 if (error) { 1081 free(fstype, M_TEMP); 1082 return (error); 1083 } 1084 1085 AUDIT_ARG_TEXT(fstype); 1086 vfsp = vfs_byname_kld(fstype, td, &error); 1087 free(fstype, M_TEMP); 1088 if (vfsp == NULL) 1089 return (ENOENT); 1090 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 && 1091 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) || 1092 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 && 1093 vfsp->vfc_vfsops->vfs_cmount == NULL)) 1094 return (EOPNOTSUPP); 1095 1096 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN); 1097 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 1098 ma = mount_argb(ma, flags & MNT_RDONLY, "noro"); 1099 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid"); 1100 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec"); 1101 1102 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0) 1103 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags)); 1104 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags)); 1105 } 1106 1107 /* 1108 * vfs_domount_first(): first file system mount (not update) 1109 */ 1110 static int 1111 vfs_domount_first( 1112 struct thread *td, /* Calling thread. */ 1113 struct vfsconf *vfsp, /* File system type. */ 1114 char *fspath, /* Mount path. */ 1115 struct vnode *vp, /* Vnode to be covered. */ 1116 uint64_t fsflags, /* Flags common to all filesystems. */ 1117 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1118 ) 1119 { 1120 struct vattr va; 1121 struct mount *mp; 1122 struct vnode *newdp, *rootvp; 1123 int error, error1; 1124 bool unmounted; 1125 1126 ASSERT_VOP_ELOCKED(vp, __func__); 1127 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 1128 1129 /* 1130 * If the jail of the calling thread lacks permission for this type of 1131 * file system, or is trying to cover its own root, deny immediately. 1132 */ 1133 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred, 1134 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) { 1135 vput(vp); 1136 return (EPERM); 1137 } 1138 1139 /* 1140 * If the user is not root, ensure that they own the directory 1141 * onto which we are attempting to mount. 1142 */ 1143 error = VOP_GETATTR(vp, &va, td->td_ucred); 1144 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 1145 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN); 1146 if (error == 0) 1147 error = vinvalbuf(vp, V_SAVE, 0, 0); 1148 if (vfsp->vfc_flags & VFCF_FILEMOUNT) { 1149 if (error == 0 && vp->v_type != VDIR && vp->v_type != VREG) 1150 error = EINVAL; 1151 /* 1152 * For file mounts, ensure that there is only one hardlink to the file. 1153 */ 1154 if (error == 0 && vp->v_type == VREG && va.va_nlink != 1) 1155 error = EINVAL; 1156 } else { 1157 if (error == 0 && vp->v_type != VDIR) 1158 error = ENOTDIR; 1159 } 1160 if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0) 1161 error = vn_dir_check_empty(vp); 1162 if (error == 0) { 1163 VI_LOCK(vp); 1164 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 1165 vp->v_iflag |= VI_MOUNT; 1166 else 1167 error = EBUSY; 1168 VI_UNLOCK(vp); 1169 } 1170 if (error != 0) { 1171 vput(vp); 1172 return (error); 1173 } 1174 vn_seqc_write_begin(vp); 1175 VOP_UNLOCK(vp); 1176 1177 /* Allocate and initialize the filesystem. */ 1178 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 1179 /* XXXMAC: pass to vfs_mount_alloc? */ 1180 mp->mnt_optnew = *optlist; 1181 /* Set the mount level flags. */ 1182 mp->mnt_flag = (fsflags & 1183 (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY | MNT_FORCE)); 1184 1185 /* 1186 * Mount the filesystem. 1187 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1188 * get. No freeing of cn_pnbuf. 1189 */ 1190 error1 = 0; 1191 unmounted = true; 1192 if ((error = VFS_MOUNT(mp)) != 0 || 1193 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 || 1194 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) { 1195 rootvp = NULL; 1196 if (error1 != 0) { 1197 MPASS(error == 0); 1198 rootvp = vfs_cache_root_clear(mp); 1199 if (rootvp != NULL) { 1200 vhold(rootvp); 1201 vrele(rootvp); 1202 } 1203 (void)vn_start_write(NULL, &mp, V_WAIT); 1204 MNT_ILOCK(mp); 1205 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF; 1206 MNT_IUNLOCK(mp); 1207 VFS_PURGE(mp); 1208 error = VFS_UNMOUNT(mp, 0); 1209 vn_finished_write(mp); 1210 if (error != 0) { 1211 printf( 1212 "failed post-mount (%d): rollback unmount returned %d\n", 1213 error1, error); 1214 unmounted = false; 1215 } 1216 error = error1; 1217 } 1218 vfs_unbusy(mp); 1219 mp->mnt_vnodecovered = NULL; 1220 if (unmounted) { 1221 /* XXXKIB wait for mnt_lockref drain? */ 1222 vfs_mount_destroy(mp); 1223 } 1224 VI_LOCK(vp); 1225 vp->v_iflag &= ~VI_MOUNT; 1226 VI_UNLOCK(vp); 1227 if (rootvp != NULL) { 1228 vn_seqc_write_end(rootvp); 1229 vdrop(rootvp); 1230 } 1231 vn_seqc_write_end(vp); 1232 vrele(vp); 1233 return (error); 1234 } 1235 vn_seqc_write_begin(newdp); 1236 VOP_UNLOCK(newdp); 1237 1238 if (mp->mnt_opt != NULL) 1239 vfs_freeopts(mp->mnt_opt); 1240 mp->mnt_opt = mp->mnt_optnew; 1241 *optlist = NULL; 1242 1243 /* 1244 * Prevent external consumers of mount options from reading mnt_optnew. 1245 */ 1246 mp->mnt_optnew = NULL; 1247 1248 MNT_ILOCK(mp); 1249 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1250 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1251 mp->mnt_kern_flag |= MNTK_ASYNC; 1252 else 1253 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1254 MNT_IUNLOCK(mp); 1255 1256 /* 1257 * VIRF_MOUNTPOINT and v_mountedhere need to be set under the 1258 * vp lock to satisfy vfs_lookup() requirements. 1259 */ 1260 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1261 VI_LOCK(vp); 1262 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT); 1263 vp->v_mountedhere = mp; 1264 VI_UNLOCK(vp); 1265 VOP_UNLOCK(vp); 1266 cache_purge(vp); 1267 1268 /* 1269 * We need to lock both vnodes. 1270 * 1271 * Use vn_lock_pair to avoid establishing an ordering between vnodes 1272 * from different filesystems. 1273 */ 1274 vn_lock_pair(vp, false, LK_EXCLUSIVE, newdp, false, LK_EXCLUSIVE); 1275 1276 VI_LOCK(vp); 1277 vp->v_iflag &= ~VI_MOUNT; 1278 VI_UNLOCK(vp); 1279 /* Place the new filesystem at the end of the mount list. */ 1280 mtx_lock(&mountlist_mtx); 1281 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1282 mtx_unlock(&mountlist_mtx); 1283 vfs_event_signal(NULL, VQ_MOUNT, 0); 1284 VOP_UNLOCK(vp); 1285 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td); 1286 VOP_UNLOCK(newdp); 1287 mount_devctl_event("MOUNT", mp, false); 1288 mountcheckdirs(vp, newdp); 1289 vn_seqc_write_end(vp); 1290 vn_seqc_write_end(newdp); 1291 vrele(newdp); 1292 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1293 vfs_allocate_syncvnode(mp); 1294 vfs_op_exit(mp); 1295 vfs_unbusy(mp); 1296 return (0); 1297 } 1298 1299 /* 1300 * vfs_domount_update(): update of mounted file system 1301 */ 1302 static int 1303 vfs_domount_update( 1304 struct thread *td, /* Calling thread. */ 1305 struct vnode *vp, /* Mount point vnode. */ 1306 uint64_t fsflags, /* Flags common to all filesystems. */ 1307 bool jail_export, /* Got export option in vnet prison. */ 1308 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1309 ) 1310 { 1311 struct export_args export; 1312 struct o2export_args o2export; 1313 struct vnode *rootvp; 1314 void *bufp; 1315 struct mount *mp; 1316 int error, export_error, i, len, fsid_up_len; 1317 uint64_t flag; 1318 gid_t *grps; 1319 fsid_t *fsid_up; 1320 bool vfs_suser_failed; 1321 1322 ASSERT_VOP_ELOCKED(vp, __func__); 1323 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 1324 mp = vp->v_mount; 1325 1326 if ((vp->v_vflag & VV_ROOT) == 0) { 1327 if (vfs_copyopt(*optlist, "export", &export, sizeof(export)) 1328 == 0) 1329 error = EXDEV; 1330 else 1331 error = EINVAL; 1332 vput(vp); 1333 return (error); 1334 } 1335 1336 /* 1337 * We only allow the filesystem to be reloaded if it 1338 * is currently mounted read-only. 1339 */ 1340 flag = mp->mnt_flag; 1341 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 1342 vput(vp); 1343 return (EOPNOTSUPP); /* Needs translation */ 1344 } 1345 /* 1346 * Only privileged root, or (if MNT_USER is set) the user that 1347 * did the original mount is permitted to update it. 1348 */ 1349 /* 1350 * For the case of mountd(8) doing exports in a jail, the vfs_suser() 1351 * call does not cause failure. vfs_domount() has already checked 1352 * that "root" is doing this and vfs_suser() will fail when 1353 * the file system has been mounted outside the jail. 1354 * jail_export set true indicates that "export" is not mixed 1355 * with other options that change mount behaviour. 1356 */ 1357 vfs_suser_failed = false; 1358 error = vfs_suser(mp, td); 1359 if (jail_export && error != 0) { 1360 error = 0; 1361 vfs_suser_failed = true; 1362 } 1363 if (error != 0) { 1364 vput(vp); 1365 return (error); 1366 } 1367 if (vfs_busy(mp, MBF_NOWAIT)) { 1368 vput(vp); 1369 return (EBUSY); 1370 } 1371 VI_LOCK(vp); 1372 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 1373 VI_UNLOCK(vp); 1374 vfs_unbusy(mp); 1375 vput(vp); 1376 return (EBUSY); 1377 } 1378 vp->v_iflag |= VI_MOUNT; 1379 VI_UNLOCK(vp); 1380 VOP_UNLOCK(vp); 1381 1382 rootvp = NULL; 1383 1384 if (vfs_getopt(*optlist, "fsid", (void **)&fsid_up, 1385 &fsid_up_len) == 0) { 1386 if (fsid_up_len != sizeof(*fsid_up)) { 1387 error = EINVAL; 1388 goto end; 1389 } 1390 if (fsidcmp(&fsid_up, &mp->mnt_stat.f_fsid) != 0) { 1391 error = ENOENT; 1392 goto end; 1393 } 1394 vfs_deleteopt(*optlist, "fsid"); 1395 } 1396 1397 vfs_op_enter(mp); 1398 vn_seqc_write_begin(vp); 1399 1400 MNT_ILOCK(mp); 1401 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1402 MNT_IUNLOCK(mp); 1403 error = EBUSY; 1404 goto end; 1405 } 1406 if (vfs_suser_failed) { 1407 KASSERT((fsflags & (MNT_EXPORTED | MNT_UPDATE)) == 1408 (MNT_EXPORTED | MNT_UPDATE), 1409 ("%s: jailed export did not set expected fsflags", 1410 __func__)); 1411 /* 1412 * For this case, only MNT_UPDATE and 1413 * MNT_EXPORTED have been set in fsflags 1414 * by the options. Only set MNT_UPDATE, 1415 * since that is the one that would be set 1416 * when set in fsflags, below. 1417 */ 1418 mp->mnt_flag |= MNT_UPDATE; 1419 } else { 1420 mp->mnt_flag &= ~MNT_UPDATEMASK; 1421 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 1422 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 1423 if ((mp->mnt_flag & MNT_ASYNC) == 0) 1424 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1425 } 1426 rootvp = vfs_cache_root_clear(mp); 1427 MNT_IUNLOCK(mp); 1428 mp->mnt_optnew = *optlist; 1429 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 1430 1431 /* 1432 * Mount the filesystem. 1433 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1434 * get. No freeing of cn_pnbuf. 1435 */ 1436 /* 1437 * For the case of mountd(8) doing exports from within a vnet jail, 1438 * "from" is typically not set correctly such that VFS_MOUNT() will 1439 * return ENOENT. It is not obvious that VFS_MOUNT() ever needs to be 1440 * called when mountd is doing exports, but this check only applies to 1441 * the specific case where it is running inside a vnet jail, to 1442 * avoid any POLA violation. 1443 */ 1444 error = 0; 1445 if (!jail_export) 1446 error = VFS_MOUNT(mp); 1447 1448 export_error = 0; 1449 /* Process the export option. */ 1450 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp, 1451 &len) == 0) { 1452 /* Assume that there is only 1 ABI for each length. */ 1453 switch (len) { 1454 case (sizeof(struct oexport_args)): 1455 bzero(&o2export, sizeof(o2export)); 1456 /* FALLTHROUGH */ 1457 case (sizeof(o2export)): 1458 bcopy(bufp, &o2export, len); 1459 export.ex_flags = (uint64_t)o2export.ex_flags; 1460 export.ex_root = o2export.ex_root; 1461 export.ex_uid = o2export.ex_anon.cr_uid; 1462 export.ex_groups = NULL; 1463 export.ex_ngroups = o2export.ex_anon.cr_ngroups; 1464 if (export.ex_ngroups > 0) { 1465 if (export.ex_ngroups <= XU_NGROUPS) { 1466 export.ex_groups = malloc( 1467 export.ex_ngroups * sizeof(gid_t), 1468 M_TEMP, M_WAITOK); 1469 for (i = 0; i < export.ex_ngroups; i++) 1470 export.ex_groups[i] = 1471 o2export.ex_anon.cr_groups[i]; 1472 } else 1473 export_error = EINVAL; 1474 } else if (export.ex_ngroups < 0) 1475 export_error = EINVAL; 1476 export.ex_addr = o2export.ex_addr; 1477 export.ex_addrlen = o2export.ex_addrlen; 1478 export.ex_mask = o2export.ex_mask; 1479 export.ex_masklen = o2export.ex_masklen; 1480 export.ex_indexfile = o2export.ex_indexfile; 1481 export.ex_numsecflavors = o2export.ex_numsecflavors; 1482 if (export.ex_numsecflavors < MAXSECFLAVORS) { 1483 for (i = 0; i < export.ex_numsecflavors; i++) 1484 export.ex_secflavors[i] = 1485 o2export.ex_secflavors[i]; 1486 } else 1487 export_error = EINVAL; 1488 if (export_error == 0) 1489 export_error = vfs_export(mp, &export, true); 1490 free(export.ex_groups, M_TEMP); 1491 break; 1492 case (sizeof(export)): 1493 bcopy(bufp, &export, len); 1494 grps = NULL; 1495 if (export.ex_ngroups > 0) { 1496 if (export.ex_ngroups <= NGROUPS_MAX) { 1497 grps = malloc(export.ex_ngroups * 1498 sizeof(gid_t), M_TEMP, M_WAITOK); 1499 export_error = copyin(export.ex_groups, 1500 grps, export.ex_ngroups * 1501 sizeof(gid_t)); 1502 if (export_error == 0) 1503 export.ex_groups = grps; 1504 } else 1505 export_error = EINVAL; 1506 } else if (export.ex_ngroups == 0) 1507 export.ex_groups = NULL; 1508 else 1509 export_error = EINVAL; 1510 if (export_error == 0) 1511 export_error = vfs_export(mp, &export, true); 1512 free(grps, M_TEMP); 1513 break; 1514 default: 1515 export_error = EINVAL; 1516 break; 1517 } 1518 } 1519 1520 MNT_ILOCK(mp); 1521 if (error == 0) { 1522 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 1523 MNT_SNAPSHOT); 1524 } else { 1525 /* 1526 * If we fail, restore old mount flags. MNT_QUOTA is special, 1527 * because it is not part of MNT_UPDATEMASK, but it could have 1528 * changed in the meantime if quotactl(2) was called. 1529 * All in all we want current value of MNT_QUOTA, not the old 1530 * one. 1531 */ 1532 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 1533 } 1534 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1535 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1536 mp->mnt_kern_flag |= MNTK_ASYNC; 1537 else 1538 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1539 MNT_IUNLOCK(mp); 1540 1541 if (error != 0) 1542 goto end; 1543 1544 mount_devctl_event("REMOUNT", mp, true); 1545 if (mp->mnt_opt != NULL) 1546 vfs_freeopts(mp->mnt_opt); 1547 mp->mnt_opt = mp->mnt_optnew; 1548 *optlist = NULL; 1549 (void)VFS_STATFS(mp, &mp->mnt_stat); 1550 /* 1551 * Prevent external consumers of mount options from reading 1552 * mnt_optnew. 1553 */ 1554 mp->mnt_optnew = NULL; 1555 1556 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1557 vfs_allocate_syncvnode(mp); 1558 else 1559 vfs_deallocate_syncvnode(mp); 1560 end: 1561 vfs_op_exit(mp); 1562 if (rootvp != NULL) { 1563 vn_seqc_write_end(rootvp); 1564 vrele(rootvp); 1565 } 1566 vn_seqc_write_end(vp); 1567 vfs_unbusy(mp); 1568 VI_LOCK(vp); 1569 vp->v_iflag &= ~VI_MOUNT; 1570 VI_UNLOCK(vp); 1571 vrele(vp); 1572 return (error != 0 ? error : export_error); 1573 } 1574 1575 /* 1576 * vfs_domount(): actually attempt a filesystem mount. 1577 */ 1578 static int 1579 vfs_domount( 1580 struct thread *td, /* Calling thread. */ 1581 const char *fstype, /* Filesystem type. */ 1582 char *fspath, /* Mount path. */ 1583 uint64_t fsflags, /* Flags common to all filesystems. */ 1584 bool jail_export, /* Got export option in vnet prison. */ 1585 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1586 ) 1587 { 1588 struct vfsconf *vfsp; 1589 struct nameidata nd; 1590 struct vnode *vp; 1591 char *pathbuf; 1592 int error; 1593 1594 /* 1595 * Be ultra-paranoid about making sure the type and fspath 1596 * variables will fit in our mp buffers, including the 1597 * terminating NUL. 1598 */ 1599 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1600 return (ENAMETOOLONG); 1601 1602 if (jail_export) { 1603 error = priv_check(td, PRIV_NFS_DAEMON); 1604 if (error) 1605 return (error); 1606 } else if (jailed(td->td_ucred) || usermount == 0) { 1607 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1608 return (error); 1609 } 1610 1611 /* 1612 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1613 */ 1614 if (fsflags & MNT_EXPORTED) { 1615 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1616 if (error) 1617 return (error); 1618 } 1619 if (fsflags & MNT_SUIDDIR) { 1620 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1621 if (error) 1622 return (error); 1623 } 1624 /* 1625 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1626 */ 1627 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1628 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1629 fsflags |= MNT_NOSUID | MNT_USER; 1630 } 1631 1632 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1633 vfsp = NULL; 1634 if ((fsflags & MNT_UPDATE) == 0) { 1635 /* Don't try to load KLDs if we're mounting the root. */ 1636 if (fsflags & MNT_ROOTFS) { 1637 if ((vfsp = vfs_byname(fstype)) == NULL) 1638 return (ENODEV); 1639 } else { 1640 if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL) 1641 return (error); 1642 } 1643 } 1644 1645 /* 1646 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1647 */ 1648 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT, 1649 UIO_SYSSPACE, fspath); 1650 error = namei(&nd); 1651 if (error != 0) 1652 return (error); 1653 vp = nd.ni_vp; 1654 /* 1655 * Don't allow stacking file mounts to work around problems with the way 1656 * that namei sets nd.ni_dvp to vp_crossmp for these. 1657 */ 1658 if (vp->v_type == VREG) 1659 fsflags |= MNT_NOCOVER; 1660 if ((fsflags & MNT_UPDATE) == 0) { 1661 if ((vp->v_vflag & VV_ROOT) != 0 && 1662 (fsflags & MNT_NOCOVER) != 0) { 1663 vput(vp); 1664 error = EBUSY; 1665 goto out; 1666 } 1667 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1668 strcpy(pathbuf, fspath); 1669 /* 1670 * Note: we allow any vnode type here. If the path sanity check 1671 * succeeds, the type will be validated in vfs_domount_first 1672 * above. 1673 */ 1674 if (vp->v_type == VDIR) 1675 error = vn_path_to_global_path(td, vp, pathbuf, 1676 MNAMELEN); 1677 else 1678 error = vn_path_to_global_path_hardlink(td, vp, 1679 nd.ni_dvp, pathbuf, MNAMELEN, 1680 nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen); 1681 if (error == 0) { 1682 error = vfs_domount_first(td, vfsp, pathbuf, vp, 1683 fsflags, optlist); 1684 } 1685 free(pathbuf, M_TEMP); 1686 } else 1687 error = vfs_domount_update(td, vp, fsflags, jail_export, 1688 optlist); 1689 1690 out: 1691 NDFREE_PNBUF(&nd); 1692 vrele(nd.ni_dvp); 1693 1694 return (error); 1695 } 1696 1697 /* 1698 * Unmount a filesystem. 1699 * 1700 * Note: unmount takes a path to the vnode mounted on as argument, not 1701 * special file (as before). 1702 */ 1703 #ifndef _SYS_SYSPROTO_H_ 1704 struct unmount_args { 1705 char *path; 1706 int flags; 1707 }; 1708 #endif 1709 /* ARGSUSED */ 1710 int 1711 sys_unmount(struct thread *td, struct unmount_args *uap) 1712 { 1713 1714 return (kern_unmount(td, uap->path, uap->flags)); 1715 } 1716 1717 int 1718 kern_unmount(struct thread *td, const char *path, int flags) 1719 { 1720 struct nameidata nd; 1721 struct mount *mp; 1722 char *fsidbuf, *pathbuf; 1723 fsid_t fsid; 1724 int error; 1725 1726 AUDIT_ARG_VALUE(flags); 1727 if (jailed(td->td_ucred) || usermount == 0) { 1728 error = priv_check(td, PRIV_VFS_UNMOUNT); 1729 if (error) 1730 return (error); 1731 } 1732 1733 if (flags & MNT_BYFSID) { 1734 fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1735 error = copyinstr(path, fsidbuf, MNAMELEN, NULL); 1736 if (error) { 1737 free(fsidbuf, M_TEMP); 1738 return (error); 1739 } 1740 1741 AUDIT_ARG_TEXT(fsidbuf); 1742 /* Decode the filesystem ID. */ 1743 if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) { 1744 free(fsidbuf, M_TEMP); 1745 return (EINVAL); 1746 } 1747 1748 mp = vfs_getvfs(&fsid); 1749 free(fsidbuf, M_TEMP); 1750 if (mp == NULL) { 1751 return (ENOENT); 1752 } 1753 } else { 1754 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1755 error = copyinstr(path, pathbuf, MNAMELEN, NULL); 1756 if (error) { 1757 free(pathbuf, M_TEMP); 1758 return (error); 1759 } 1760 1761 /* 1762 * Try to find global path for path argument. 1763 */ 1764 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1765 UIO_SYSSPACE, pathbuf); 1766 if (namei(&nd) == 0) { 1767 NDFREE_PNBUF(&nd); 1768 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf, 1769 MNAMELEN); 1770 if (error == 0) 1771 vput(nd.ni_vp); 1772 } 1773 mtx_lock(&mountlist_mtx); 1774 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1775 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) { 1776 vfs_ref(mp); 1777 break; 1778 } 1779 } 1780 mtx_unlock(&mountlist_mtx); 1781 free(pathbuf, M_TEMP); 1782 if (mp == NULL) { 1783 /* 1784 * Previously we returned ENOENT for a nonexistent path and 1785 * EINVAL for a non-mountpoint. We cannot tell these apart 1786 * now, so in the !MNT_BYFSID case return the more likely 1787 * EINVAL for compatibility. 1788 */ 1789 return (EINVAL); 1790 } 1791 } 1792 1793 /* 1794 * Don't allow unmounting the root filesystem. 1795 */ 1796 if (mp->mnt_flag & MNT_ROOTFS) { 1797 vfs_rel(mp); 1798 return (EINVAL); 1799 } 1800 error = dounmount(mp, flags, td); 1801 return (error); 1802 } 1803 1804 /* 1805 * Return error if any of the vnodes, ignoring the root vnode 1806 * and the syncer vnode, have non-zero usecount. 1807 * 1808 * This function is purely advisory - it can return false positives 1809 * and negatives. 1810 */ 1811 static int 1812 vfs_check_usecounts(struct mount *mp) 1813 { 1814 struct vnode *vp, *mvp; 1815 1816 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1817 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON && 1818 vp->v_usecount != 0) { 1819 VI_UNLOCK(vp); 1820 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1821 return (EBUSY); 1822 } 1823 VI_UNLOCK(vp); 1824 } 1825 1826 return (0); 1827 } 1828 1829 static void 1830 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags) 1831 { 1832 1833 mtx_assert(MNT_MTX(mp), MA_OWNED); 1834 mp->mnt_kern_flag &= ~mntkflags; 1835 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) { 1836 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1837 wakeup(mp); 1838 } 1839 vfs_op_exit_locked(mp); 1840 MNT_IUNLOCK(mp); 1841 if (coveredvp != NULL) { 1842 VOP_UNLOCK(coveredvp); 1843 vdrop(coveredvp); 1844 } 1845 vn_finished_write(mp); 1846 vfs_rel(mp); 1847 } 1848 1849 /* 1850 * There are various reference counters associated with the mount point. 1851 * Normally it is permitted to modify them without taking the mnt ilock, 1852 * but this behavior can be temporarily disabled if stable value is needed 1853 * or callers are expected to block (e.g. to not allow new users during 1854 * forced unmount). 1855 */ 1856 void 1857 vfs_op_enter(struct mount *mp) 1858 { 1859 struct mount_pcpu *mpcpu; 1860 int cpu; 1861 1862 MNT_ILOCK(mp); 1863 mp->mnt_vfs_ops++; 1864 if (mp->mnt_vfs_ops > 1) { 1865 MNT_IUNLOCK(mp); 1866 return; 1867 } 1868 vfs_op_barrier_wait(mp); 1869 CPU_FOREACH(cpu) { 1870 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1871 1872 mp->mnt_ref += mpcpu->mntp_ref; 1873 mpcpu->mntp_ref = 0; 1874 1875 mp->mnt_lockref += mpcpu->mntp_lockref; 1876 mpcpu->mntp_lockref = 0; 1877 1878 mp->mnt_writeopcount += mpcpu->mntp_writeopcount; 1879 mpcpu->mntp_writeopcount = 0; 1880 } 1881 MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 && 1882 mp->mnt_writeopcount >= 0, mp, 1883 ("invalid count(s): ref %d lockref %d writeopcount %d", 1884 mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount)); 1885 MNT_IUNLOCK(mp); 1886 vfs_assert_mount_counters(mp); 1887 } 1888 1889 void 1890 vfs_op_exit_locked(struct mount *mp) 1891 { 1892 1893 mtx_assert(MNT_MTX(mp), MA_OWNED); 1894 1895 MPASSERT(mp->mnt_vfs_ops > 0, mp, 1896 ("invalid vfs_ops count %d", mp->mnt_vfs_ops)); 1897 MPASSERT(mp->mnt_vfs_ops > 1 || 1898 (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp, 1899 ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops)); 1900 mp->mnt_vfs_ops--; 1901 } 1902 1903 void 1904 vfs_op_exit(struct mount *mp) 1905 { 1906 1907 MNT_ILOCK(mp); 1908 vfs_op_exit_locked(mp); 1909 MNT_IUNLOCK(mp); 1910 } 1911 1912 struct vfs_op_barrier_ipi { 1913 struct mount *mp; 1914 struct smp_rendezvous_cpus_retry_arg srcra; 1915 }; 1916 1917 static void 1918 vfs_op_action_func(void *arg) 1919 { 1920 struct vfs_op_barrier_ipi *vfsopipi; 1921 struct mount *mp; 1922 1923 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1924 mp = vfsopipi->mp; 1925 1926 if (!vfs_op_thread_entered(mp)) 1927 smp_rendezvous_cpus_done(arg); 1928 } 1929 1930 static void 1931 vfs_op_wait_func(void *arg, int cpu) 1932 { 1933 struct vfs_op_barrier_ipi *vfsopipi; 1934 struct mount *mp; 1935 struct mount_pcpu *mpcpu; 1936 1937 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1938 mp = vfsopipi->mp; 1939 1940 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1941 while (atomic_load_int(&mpcpu->mntp_thread_in_ops)) 1942 cpu_spinwait(); 1943 } 1944 1945 void 1946 vfs_op_barrier_wait(struct mount *mp) 1947 { 1948 struct vfs_op_barrier_ipi vfsopipi; 1949 1950 vfsopipi.mp = mp; 1951 1952 smp_rendezvous_cpus_retry(all_cpus, 1953 smp_no_rendezvous_barrier, 1954 vfs_op_action_func, 1955 smp_no_rendezvous_barrier, 1956 vfs_op_wait_func, 1957 &vfsopipi.srcra); 1958 } 1959 1960 #ifdef DIAGNOSTIC 1961 void 1962 vfs_assert_mount_counters(struct mount *mp) 1963 { 1964 struct mount_pcpu *mpcpu; 1965 int cpu; 1966 1967 if (mp->mnt_vfs_ops == 0) 1968 return; 1969 1970 CPU_FOREACH(cpu) { 1971 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1972 if (mpcpu->mntp_ref != 0 || 1973 mpcpu->mntp_lockref != 0 || 1974 mpcpu->mntp_writeopcount != 0) 1975 vfs_dump_mount_counters(mp); 1976 } 1977 } 1978 1979 void 1980 vfs_dump_mount_counters(struct mount *mp) 1981 { 1982 struct mount_pcpu *mpcpu; 1983 int ref, lockref, writeopcount; 1984 int cpu; 1985 1986 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops); 1987 1988 printf(" ref : "); 1989 ref = mp->mnt_ref; 1990 CPU_FOREACH(cpu) { 1991 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1992 printf("%d ", mpcpu->mntp_ref); 1993 ref += mpcpu->mntp_ref; 1994 } 1995 printf("\n"); 1996 printf(" lockref : "); 1997 lockref = mp->mnt_lockref; 1998 CPU_FOREACH(cpu) { 1999 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 2000 printf("%d ", mpcpu->mntp_lockref); 2001 lockref += mpcpu->mntp_lockref; 2002 } 2003 printf("\n"); 2004 printf("writeopcount: "); 2005 writeopcount = mp->mnt_writeopcount; 2006 CPU_FOREACH(cpu) { 2007 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 2008 printf("%d ", mpcpu->mntp_writeopcount); 2009 writeopcount += mpcpu->mntp_writeopcount; 2010 } 2011 printf("\n"); 2012 2013 printf("counter struct total\n"); 2014 printf("ref %-5d %-5d\n", mp->mnt_ref, ref); 2015 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref); 2016 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount); 2017 2018 panic("invalid counts on struct mount"); 2019 } 2020 #endif 2021 2022 int 2023 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which) 2024 { 2025 struct mount_pcpu *mpcpu; 2026 int cpu, sum; 2027 2028 switch (which) { 2029 case MNT_COUNT_REF: 2030 sum = mp->mnt_ref; 2031 break; 2032 case MNT_COUNT_LOCKREF: 2033 sum = mp->mnt_lockref; 2034 break; 2035 case MNT_COUNT_WRITEOPCOUNT: 2036 sum = mp->mnt_writeopcount; 2037 break; 2038 } 2039 2040 CPU_FOREACH(cpu) { 2041 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 2042 switch (which) { 2043 case MNT_COUNT_REF: 2044 sum += mpcpu->mntp_ref; 2045 break; 2046 case MNT_COUNT_LOCKREF: 2047 sum += mpcpu->mntp_lockref; 2048 break; 2049 case MNT_COUNT_WRITEOPCOUNT: 2050 sum += mpcpu->mntp_writeopcount; 2051 break; 2052 } 2053 } 2054 return (sum); 2055 } 2056 2057 static bool 2058 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue, 2059 int timeout_ticks) 2060 { 2061 bool enqueued; 2062 2063 enqueued = false; 2064 mtx_lock(&deferred_unmount_lock); 2065 if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) { 2066 mp->mnt_taskqueue_flags = flags | MNT_DEFERRED; 2067 STAILQ_INSERT_TAIL(&deferred_unmount_list, mp, 2068 mnt_taskqueue_link); 2069 enqueued = true; 2070 } 2071 mtx_unlock(&deferred_unmount_lock); 2072 2073 if (enqueued) { 2074 taskqueue_enqueue_timeout(taskqueue_deferred_unmount, 2075 &deferred_unmount_task, timeout_ticks); 2076 } 2077 2078 return (enqueued); 2079 } 2080 2081 /* 2082 * Taskqueue handler for processing async/recursive unmounts 2083 */ 2084 static void 2085 vfs_deferred_unmount(void *argi __unused, int pending __unused) 2086 { 2087 STAILQ_HEAD(, mount) local_unmounts; 2088 uint64_t flags; 2089 struct mount *mp, *tmp; 2090 int error; 2091 unsigned int retries; 2092 bool unmounted; 2093 2094 STAILQ_INIT(&local_unmounts); 2095 mtx_lock(&deferred_unmount_lock); 2096 STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list); 2097 mtx_unlock(&deferred_unmount_lock); 2098 2099 STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) { 2100 flags = mp->mnt_taskqueue_flags; 2101 KASSERT((flags & MNT_DEFERRED) != 0, 2102 ("taskqueue unmount without MNT_DEFERRED")); 2103 error = dounmount(mp, flags, curthread); 2104 if (error != 0) { 2105 MNT_ILOCK(mp); 2106 unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0); 2107 MNT_IUNLOCK(mp); 2108 2109 /* 2110 * The deferred unmount thread is the only thread that 2111 * modifies the retry counts, so locking/atomics aren't 2112 * needed here. 2113 */ 2114 retries = (mp->mnt_unmount_retries)++; 2115 deferred_unmount_total_retries++; 2116 if (!unmounted && retries < deferred_unmount_retry_limit) { 2117 deferred_unmount_enqueue(mp, flags, true, 2118 -deferred_unmount_retry_delay_hz); 2119 } else { 2120 if (retries >= deferred_unmount_retry_limit) { 2121 printf("giving up on deferred unmount " 2122 "of %s after %d retries, error %d\n", 2123 mp->mnt_stat.f_mntonname, retries, error); 2124 } 2125 vfs_rel(mp); 2126 } 2127 } 2128 } 2129 } 2130 2131 /* 2132 * Do the actual filesystem unmount. 2133 */ 2134 int 2135 dounmount(struct mount *mp, uint64_t flags, struct thread *td) 2136 { 2137 struct mount_upper_node *upper; 2138 struct vnode *coveredvp, *rootvp; 2139 int error; 2140 uint64_t async_flag; 2141 int mnt_gen_r; 2142 unsigned int retries; 2143 2144 KASSERT((flags & MNT_DEFERRED) == 0 || 2145 (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE), 2146 ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE")); 2147 2148 /* 2149 * If the caller has explicitly requested the unmount to be handled by 2150 * the taskqueue and we're not already in taskqueue context, queue 2151 * up the unmount request and exit. This is done prior to any 2152 * credential checks; MNT_DEFERRED should be used only for kernel- 2153 * initiated unmounts and will therefore be processed with the 2154 * (kernel) credentials of the taskqueue thread. Still, callers 2155 * should be sure this is the behavior they want. 2156 */ 2157 if ((flags & MNT_DEFERRED) != 0 && 2158 taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) { 2159 if (!deferred_unmount_enqueue(mp, flags, false, 0)) 2160 vfs_rel(mp); 2161 return (EINPROGRESS); 2162 } 2163 2164 /* 2165 * Only privileged root, or (if MNT_USER is set) the user that did the 2166 * original mount is permitted to unmount this filesystem. 2167 * This check should be made prior to queueing up any recursive 2168 * unmounts of upper filesystems. Those unmounts will be executed 2169 * with kernel thread credentials and are expected to succeed, so 2170 * we must at least ensure the originating context has sufficient 2171 * privilege to unmount the base filesystem before proceeding with 2172 * the uppers. 2173 */ 2174 error = vfs_suser(mp, td); 2175 if (error != 0) { 2176 KASSERT((flags & MNT_DEFERRED) == 0, 2177 ("taskqueue unmount with insufficient privilege")); 2178 vfs_rel(mp); 2179 return (error); 2180 } 2181 2182 if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0)) 2183 flags |= MNT_RECURSE; 2184 2185 if ((flags & MNT_RECURSE) != 0) { 2186 KASSERT((flags & MNT_FORCE) != 0, 2187 ("MNT_RECURSE requires MNT_FORCE")); 2188 2189 MNT_ILOCK(mp); 2190 /* 2191 * Set MNTK_RECURSE to prevent new upper mounts from being 2192 * added, and note that an operation on the uppers list is in 2193 * progress. This will ensure that unregistration from the 2194 * uppers list, and therefore any pending unmount of the upper 2195 * FS, can't complete until after we finish walking the list. 2196 */ 2197 mp->mnt_kern_flag |= MNTK_RECURSE; 2198 mp->mnt_upper_pending++; 2199 TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) { 2200 retries = upper->mp->mnt_unmount_retries; 2201 if (retries > deferred_unmount_retry_limit) { 2202 error = EBUSY; 2203 continue; 2204 } 2205 MNT_IUNLOCK(mp); 2206 2207 vfs_ref(upper->mp); 2208 if (!deferred_unmount_enqueue(upper->mp, flags, 2209 false, 0)) 2210 vfs_rel(upper->mp); 2211 MNT_ILOCK(mp); 2212 } 2213 mp->mnt_upper_pending--; 2214 if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 && 2215 mp->mnt_upper_pending == 0) { 2216 mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER; 2217 wakeup(&mp->mnt_uppers); 2218 } 2219 2220 /* 2221 * If we're not on the taskqueue, wait until the uppers list 2222 * is drained before proceeding with unmount. Otherwise, if 2223 * we are on the taskqueue and there are still pending uppers, 2224 * just re-enqueue on the end of the taskqueue. 2225 */ 2226 if ((flags & MNT_DEFERRED) == 0) { 2227 while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) { 2228 mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER; 2229 error = msleep(&mp->mnt_taskqueue_link, 2230 MNT_MTX(mp), PCATCH, "umntqw", 0); 2231 } 2232 if (error != 0) { 2233 MNT_REL(mp); 2234 MNT_IUNLOCK(mp); 2235 return (error); 2236 } 2237 } else if (!TAILQ_EMPTY(&mp->mnt_uppers)) { 2238 MNT_IUNLOCK(mp); 2239 if (error == 0) 2240 deferred_unmount_enqueue(mp, flags, true, 0); 2241 return (error); 2242 } 2243 MNT_IUNLOCK(mp); 2244 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty")); 2245 } 2246 2247 /* Allow the taskqueue to safely re-enqueue on failure */ 2248 if ((flags & MNT_DEFERRED) != 0) 2249 vfs_ref(mp); 2250 2251 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 2252 mnt_gen_r = mp->mnt_gen; 2253 VI_LOCK(coveredvp); 2254 vholdl(coveredvp); 2255 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 2256 /* 2257 * Check for mp being unmounted while waiting for the 2258 * covered vnode lock. 2259 */ 2260 if (coveredvp->v_mountedhere != mp || 2261 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 2262 VOP_UNLOCK(coveredvp); 2263 vdrop(coveredvp); 2264 vfs_rel(mp); 2265 return (EBUSY); 2266 } 2267 } 2268 2269 vfs_op_enter(mp); 2270 2271 vn_start_write(NULL, &mp, V_WAIT); 2272 MNT_ILOCK(mp); 2273 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 || 2274 (mp->mnt_flag & MNT_UPDATE) != 0 || 2275 !TAILQ_EMPTY(&mp->mnt_uppers)) { 2276 dounmount_cleanup(mp, coveredvp, 0); 2277 return (EBUSY); 2278 } 2279 mp->mnt_kern_flag |= MNTK_UNMOUNT; 2280 rootvp = vfs_cache_root_clear(mp); 2281 if (coveredvp != NULL) 2282 vn_seqc_write_begin(coveredvp); 2283 if (flags & MNT_NONBUSY) { 2284 MNT_IUNLOCK(mp); 2285 error = vfs_check_usecounts(mp); 2286 MNT_ILOCK(mp); 2287 if (error != 0) { 2288 vn_seqc_write_end(coveredvp); 2289 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT); 2290 if (rootvp != NULL) { 2291 vn_seqc_write_end(rootvp); 2292 vrele(rootvp); 2293 } 2294 return (error); 2295 } 2296 } 2297 /* Allow filesystems to detect that a forced unmount is in progress. */ 2298 if (flags & MNT_FORCE) { 2299 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 2300 MNT_IUNLOCK(mp); 2301 /* 2302 * Must be done after setting MNTK_UNMOUNTF and before 2303 * waiting for mnt_lockref to become 0. 2304 */ 2305 VFS_PURGE(mp); 2306 MNT_ILOCK(mp); 2307 } 2308 error = 0; 2309 if (mp->mnt_lockref) { 2310 mp->mnt_kern_flag |= MNTK_DRAINING; 2311 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 2312 "mount drain", 0); 2313 } 2314 MNT_IUNLOCK(mp); 2315 KASSERT(mp->mnt_lockref == 0, 2316 ("%s: invalid lock refcount in the drain path @ %s:%d", 2317 __func__, __FILE__, __LINE__)); 2318 KASSERT(error == 0, 2319 ("%s: invalid return value for msleep in the drain path @ %s:%d", 2320 __func__, __FILE__, __LINE__)); 2321 2322 /* 2323 * We want to keep the vnode around so that we can vn_seqc_write_end 2324 * after we are done with unmount. Downgrade our reference to a mere 2325 * hold count so that we don't interefere with anything. 2326 */ 2327 if (rootvp != NULL) { 2328 vhold(rootvp); 2329 vrele(rootvp); 2330 } 2331 2332 if (mp->mnt_flag & MNT_EXPUBLIC) 2333 vfs_setpublicfs(NULL, NULL, NULL); 2334 2335 vfs_periodic(mp, MNT_WAIT); 2336 MNT_ILOCK(mp); 2337 async_flag = mp->mnt_flag & MNT_ASYNC; 2338 mp->mnt_flag &= ~MNT_ASYNC; 2339 mp->mnt_kern_flag &= ~MNTK_ASYNC; 2340 MNT_IUNLOCK(mp); 2341 vfs_deallocate_syncvnode(mp); 2342 error = VFS_UNMOUNT(mp, flags); 2343 vn_finished_write(mp); 2344 vfs_rel(mp); 2345 /* 2346 * If we failed to flush the dirty blocks for this mount point, 2347 * undo all the cdir/rdir and rootvnode changes we made above. 2348 * Unless we failed to do so because the device is reporting that 2349 * it doesn't exist anymore. 2350 */ 2351 if (error && error != ENXIO) { 2352 MNT_ILOCK(mp); 2353 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 2354 MNT_IUNLOCK(mp); 2355 vfs_allocate_syncvnode(mp); 2356 MNT_ILOCK(mp); 2357 } 2358 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 2359 mp->mnt_flag |= async_flag; 2360 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 2361 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 2362 mp->mnt_kern_flag |= MNTK_ASYNC; 2363 if (mp->mnt_kern_flag & MNTK_MWAIT) { 2364 mp->mnt_kern_flag &= ~MNTK_MWAIT; 2365 wakeup(mp); 2366 } 2367 vfs_op_exit_locked(mp); 2368 MNT_IUNLOCK(mp); 2369 if (coveredvp) { 2370 vn_seqc_write_end(coveredvp); 2371 VOP_UNLOCK(coveredvp); 2372 vdrop(coveredvp); 2373 } 2374 if (rootvp != NULL) { 2375 vn_seqc_write_end(rootvp); 2376 vdrop(rootvp); 2377 } 2378 return (error); 2379 } 2380 2381 mtx_lock(&mountlist_mtx); 2382 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2383 mtx_unlock(&mountlist_mtx); 2384 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td); 2385 if (coveredvp != NULL) { 2386 VI_LOCK(coveredvp); 2387 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT); 2388 coveredvp->v_mountedhere = NULL; 2389 vn_seqc_write_end_locked(coveredvp); 2390 VI_UNLOCK(coveredvp); 2391 VOP_UNLOCK(coveredvp); 2392 vdrop(coveredvp); 2393 } 2394 mount_devctl_event("UNMOUNT", mp, false); 2395 if (rootvp != NULL) { 2396 vn_seqc_write_end(rootvp); 2397 vdrop(rootvp); 2398 } 2399 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 2400 if (rootvnode != NULL && mp == rootvnode->v_mount) { 2401 vrele(rootvnode); 2402 rootvnode = NULL; 2403 } 2404 if (mp == rootdevmp) 2405 rootdevmp = NULL; 2406 if ((flags & MNT_DEFERRED) != 0) 2407 vfs_rel(mp); 2408 vfs_mount_destroy(mp); 2409 return (0); 2410 } 2411 2412 /* 2413 * Report errors during filesystem mounting. 2414 */ 2415 void 2416 vfs_mount_error(struct mount *mp, const char *fmt, ...) 2417 { 2418 struct vfsoptlist *moptlist = mp->mnt_optnew; 2419 va_list ap; 2420 int error, len; 2421 char *errmsg; 2422 2423 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 2424 if (error || errmsg == NULL || len <= 0) 2425 return; 2426 2427 va_start(ap, fmt); 2428 vsnprintf(errmsg, (size_t)len, fmt, ap); 2429 va_end(ap); 2430 } 2431 2432 void 2433 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 2434 { 2435 va_list ap; 2436 int error, len; 2437 char *errmsg; 2438 2439 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 2440 if (error || errmsg == NULL || len <= 0) 2441 return; 2442 2443 va_start(ap, fmt); 2444 vsnprintf(errmsg, (size_t)len, fmt, ap); 2445 va_end(ap); 2446 } 2447 2448 /* 2449 * --------------------------------------------------------------------- 2450 * Functions for querying mount options/arguments from filesystems. 2451 */ 2452 2453 /* 2454 * Check that no unknown options are given 2455 */ 2456 int 2457 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 2458 { 2459 struct vfsopt *opt; 2460 char errmsg[255]; 2461 const char **t, *p, *q; 2462 int ret = 0; 2463 2464 TAILQ_FOREACH(opt, opts, link) { 2465 p = opt->name; 2466 q = NULL; 2467 if (p[0] == 'n' && p[1] == 'o') 2468 q = p + 2; 2469 for(t = global_opts; *t != NULL; t++) { 2470 if (strcmp(*t, p) == 0) 2471 break; 2472 if (q != NULL) { 2473 if (strcmp(*t, q) == 0) 2474 break; 2475 } 2476 } 2477 if (*t != NULL) 2478 continue; 2479 for(t = legal; *t != NULL; t++) { 2480 if (strcmp(*t, p) == 0) 2481 break; 2482 if (q != NULL) { 2483 if (strcmp(*t, q) == 0) 2484 break; 2485 } 2486 } 2487 if (*t != NULL) 2488 continue; 2489 snprintf(errmsg, sizeof(errmsg), 2490 "mount option <%s> is unknown", p); 2491 ret = EINVAL; 2492 } 2493 if (ret != 0) { 2494 TAILQ_FOREACH(opt, opts, link) { 2495 if (strcmp(opt->name, "errmsg") == 0) { 2496 strncpy((char *)opt->value, errmsg, opt->len); 2497 break; 2498 } 2499 } 2500 if (opt == NULL) 2501 printf("%s\n", errmsg); 2502 } 2503 return (ret); 2504 } 2505 2506 /* 2507 * Get a mount option by its name. 2508 * 2509 * Return 0 if the option was found, ENOENT otherwise. 2510 * If len is non-NULL it will be filled with the length 2511 * of the option. If buf is non-NULL, it will be filled 2512 * with the address of the option. 2513 */ 2514 int 2515 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len) 2516 { 2517 struct vfsopt *opt; 2518 2519 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2520 2521 TAILQ_FOREACH(opt, opts, link) { 2522 if (strcmp(name, opt->name) == 0) { 2523 opt->seen = 1; 2524 if (len != NULL) 2525 *len = opt->len; 2526 if (buf != NULL) 2527 *buf = opt->value; 2528 return (0); 2529 } 2530 } 2531 return (ENOENT); 2532 } 2533 2534 int 2535 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2536 { 2537 struct vfsopt *opt; 2538 2539 if (opts == NULL) 2540 return (-1); 2541 2542 TAILQ_FOREACH(opt, opts, link) { 2543 if (strcmp(name, opt->name) == 0) { 2544 opt->seen = 1; 2545 return (opt->pos); 2546 } 2547 } 2548 return (-1); 2549 } 2550 2551 int 2552 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value) 2553 { 2554 char *opt_value, *vtp; 2555 quad_t iv; 2556 int error, opt_len; 2557 2558 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len); 2559 if (error != 0) 2560 return (error); 2561 if (opt_len == 0 || opt_value == NULL) 2562 return (EINVAL); 2563 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0') 2564 return (EINVAL); 2565 iv = strtoq(opt_value, &vtp, 0); 2566 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0')) 2567 return (EINVAL); 2568 if (iv < 0) 2569 return (EINVAL); 2570 switch (vtp[0]) { 2571 case 't': case 'T': 2572 iv *= 1024; 2573 /* FALLTHROUGH */ 2574 case 'g': case 'G': 2575 iv *= 1024; 2576 /* FALLTHROUGH */ 2577 case 'm': case 'M': 2578 iv *= 1024; 2579 /* FALLTHROUGH */ 2580 case 'k': case 'K': 2581 iv *= 1024; 2582 case '\0': 2583 break; 2584 default: 2585 return (EINVAL); 2586 } 2587 *value = iv; 2588 2589 return (0); 2590 } 2591 2592 char * 2593 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2594 { 2595 struct vfsopt *opt; 2596 2597 *error = 0; 2598 TAILQ_FOREACH(opt, opts, link) { 2599 if (strcmp(name, opt->name) != 0) 2600 continue; 2601 opt->seen = 1; 2602 if (opt->len == 0 || 2603 ((char *)opt->value)[opt->len - 1] != '\0') { 2604 *error = EINVAL; 2605 return (NULL); 2606 } 2607 return (opt->value); 2608 } 2609 *error = ENOENT; 2610 return (NULL); 2611 } 2612 2613 int 2614 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, 2615 uint64_t val) 2616 { 2617 struct vfsopt *opt; 2618 2619 TAILQ_FOREACH(opt, opts, link) { 2620 if (strcmp(name, opt->name) == 0) { 2621 opt->seen = 1; 2622 if (w != NULL) 2623 *w |= val; 2624 return (1); 2625 } 2626 } 2627 if (w != NULL) 2628 *w &= ~val; 2629 return (0); 2630 } 2631 2632 int 2633 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2634 { 2635 va_list ap; 2636 struct vfsopt *opt; 2637 int ret; 2638 2639 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2640 2641 TAILQ_FOREACH(opt, opts, link) { 2642 if (strcmp(name, opt->name) != 0) 2643 continue; 2644 opt->seen = 1; 2645 if (opt->len == 0 || opt->value == NULL) 2646 return (0); 2647 if (((char *)opt->value)[opt->len - 1] != '\0') 2648 return (0); 2649 va_start(ap, fmt); 2650 ret = vsscanf(opt->value, fmt, ap); 2651 va_end(ap); 2652 return (ret); 2653 } 2654 return (0); 2655 } 2656 2657 int 2658 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2659 { 2660 struct vfsopt *opt; 2661 2662 TAILQ_FOREACH(opt, opts, link) { 2663 if (strcmp(name, opt->name) != 0) 2664 continue; 2665 opt->seen = 1; 2666 if (opt->value == NULL) 2667 opt->len = len; 2668 else { 2669 if (opt->len != len) 2670 return (EINVAL); 2671 bcopy(value, opt->value, len); 2672 } 2673 return (0); 2674 } 2675 return (ENOENT); 2676 } 2677 2678 int 2679 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2680 { 2681 struct vfsopt *opt; 2682 2683 TAILQ_FOREACH(opt, opts, link) { 2684 if (strcmp(name, opt->name) != 0) 2685 continue; 2686 opt->seen = 1; 2687 if (opt->value == NULL) 2688 opt->len = len; 2689 else { 2690 if (opt->len < len) 2691 return (EINVAL); 2692 opt->len = len; 2693 bcopy(value, opt->value, len); 2694 } 2695 return (0); 2696 } 2697 return (ENOENT); 2698 } 2699 2700 int 2701 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2702 { 2703 struct vfsopt *opt; 2704 2705 TAILQ_FOREACH(opt, opts, link) { 2706 if (strcmp(name, opt->name) != 0) 2707 continue; 2708 opt->seen = 1; 2709 if (opt->value == NULL) 2710 opt->len = strlen(value) + 1; 2711 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2712 return (EINVAL); 2713 return (0); 2714 } 2715 return (ENOENT); 2716 } 2717 2718 /* 2719 * Find and copy a mount option. 2720 * 2721 * The size of the buffer has to be specified 2722 * in len, if it is not the same length as the 2723 * mount option, EINVAL is returned. 2724 * Returns ENOENT if the option is not found. 2725 */ 2726 int 2727 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len) 2728 { 2729 struct vfsopt *opt; 2730 2731 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2732 2733 TAILQ_FOREACH(opt, opts, link) { 2734 if (strcmp(name, opt->name) == 0) { 2735 opt->seen = 1; 2736 if (len != opt->len) 2737 return (EINVAL); 2738 bcopy(opt->value, dest, opt->len); 2739 return (0); 2740 } 2741 } 2742 return (ENOENT); 2743 } 2744 2745 int 2746 __vfs_statfs(struct mount *mp, struct statfs *sbp) 2747 { 2748 /* 2749 * Filesystems only fill in part of the structure for updates, we 2750 * have to read the entirety first to get all content. 2751 */ 2752 if (sbp != &mp->mnt_stat) 2753 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp)); 2754 2755 /* 2756 * Set these in case the underlying filesystem fails to do so. 2757 */ 2758 sbp->f_version = STATFS_VERSION; 2759 sbp->f_namemax = NAME_MAX; 2760 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 2761 sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize; 2762 2763 return (mp->mnt_op->vfs_statfs(mp, sbp)); 2764 } 2765 2766 void 2767 vfs_mountedfrom(struct mount *mp, const char *from) 2768 { 2769 2770 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2771 strlcpy(mp->mnt_stat.f_mntfromname, from, 2772 sizeof mp->mnt_stat.f_mntfromname); 2773 } 2774 2775 /* 2776 * --------------------------------------------------------------------- 2777 * This is the api for building mount args and mounting filesystems from 2778 * inside the kernel. 2779 * 2780 * The API works by accumulation of individual args. First error is 2781 * latched. 2782 * 2783 * XXX: should be documented in new manpage kernel_mount(9) 2784 */ 2785 2786 /* A memory allocation which must be freed when we are done */ 2787 struct mntaarg { 2788 SLIST_ENTRY(mntaarg) next; 2789 }; 2790 2791 /* The header for the mount arguments */ 2792 struct mntarg { 2793 struct iovec *v; 2794 int len; 2795 int error; 2796 SLIST_HEAD(, mntaarg) list; 2797 }; 2798 2799 /* 2800 * Add a boolean argument. 2801 * 2802 * flag is the boolean value. 2803 * name must start with "no". 2804 */ 2805 struct mntarg * 2806 mount_argb(struct mntarg *ma, int flag, const char *name) 2807 { 2808 2809 KASSERT(name[0] == 'n' && name[1] == 'o', 2810 ("mount_argb(...,%s): name must start with 'no'", name)); 2811 2812 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2813 } 2814 2815 /* 2816 * Add an argument printf style 2817 */ 2818 struct mntarg * 2819 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2820 { 2821 va_list ap; 2822 struct mntaarg *maa; 2823 struct sbuf *sb; 2824 int len; 2825 2826 if (ma == NULL) { 2827 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2828 SLIST_INIT(&ma->list); 2829 } 2830 if (ma->error) 2831 return (ma); 2832 2833 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2834 M_MOUNT, M_WAITOK); 2835 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2836 ma->v[ma->len].iov_len = strlen(name) + 1; 2837 ma->len++; 2838 2839 sb = sbuf_new_auto(); 2840 va_start(ap, fmt); 2841 sbuf_vprintf(sb, fmt, ap); 2842 va_end(ap); 2843 sbuf_finish(sb); 2844 len = sbuf_len(sb) + 1; 2845 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2846 SLIST_INSERT_HEAD(&ma->list, maa, next); 2847 bcopy(sbuf_data(sb), maa + 1, len); 2848 sbuf_delete(sb); 2849 2850 ma->v[ma->len].iov_base = maa + 1; 2851 ma->v[ma->len].iov_len = len; 2852 ma->len++; 2853 2854 return (ma); 2855 } 2856 2857 /* 2858 * Add an argument which is a userland string. 2859 */ 2860 struct mntarg * 2861 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2862 { 2863 struct mntaarg *maa; 2864 char *tbuf; 2865 2866 if (val == NULL) 2867 return (ma); 2868 if (ma == NULL) { 2869 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2870 SLIST_INIT(&ma->list); 2871 } 2872 if (ma->error) 2873 return (ma); 2874 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2875 SLIST_INSERT_HEAD(&ma->list, maa, next); 2876 tbuf = (void *)(maa + 1); 2877 ma->error = copyinstr(val, tbuf, len, NULL); 2878 return (mount_arg(ma, name, tbuf, -1)); 2879 } 2880 2881 /* 2882 * Plain argument. 2883 * 2884 * If length is -1, treat value as a C string. 2885 */ 2886 struct mntarg * 2887 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2888 { 2889 2890 if (ma == NULL) { 2891 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2892 SLIST_INIT(&ma->list); 2893 } 2894 if (ma->error) 2895 return (ma); 2896 2897 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2898 M_MOUNT, M_WAITOK); 2899 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2900 ma->v[ma->len].iov_len = strlen(name) + 1; 2901 ma->len++; 2902 2903 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2904 if (len < 0) 2905 ma->v[ma->len].iov_len = strlen(val) + 1; 2906 else 2907 ma->v[ma->len].iov_len = len; 2908 ma->len++; 2909 return (ma); 2910 } 2911 2912 /* 2913 * Free a mntarg structure 2914 */ 2915 static void 2916 free_mntarg(struct mntarg *ma) 2917 { 2918 struct mntaarg *maa; 2919 2920 while (!SLIST_EMPTY(&ma->list)) { 2921 maa = SLIST_FIRST(&ma->list); 2922 SLIST_REMOVE_HEAD(&ma->list, next); 2923 free(maa, M_MOUNT); 2924 } 2925 free(ma->v, M_MOUNT); 2926 free(ma, M_MOUNT); 2927 } 2928 2929 /* 2930 * Mount a filesystem 2931 */ 2932 int 2933 kernel_mount(struct mntarg *ma, uint64_t flags) 2934 { 2935 struct uio auio; 2936 int error; 2937 2938 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2939 KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v")); 2940 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2941 2942 error = ma->error; 2943 if (error == 0) { 2944 auio.uio_iov = ma->v; 2945 auio.uio_iovcnt = ma->len; 2946 auio.uio_segflg = UIO_SYSSPACE; 2947 error = vfs_donmount(curthread, flags, &auio); 2948 } 2949 free_mntarg(ma); 2950 return (error); 2951 } 2952 2953 /* Map from mount options to printable formats. */ 2954 static struct mntoptnames optnames[] = { 2955 MNTOPT_NAMES 2956 }; 2957 2958 #define DEVCTL_LEN 1024 2959 static void 2960 mount_devctl_event(const char *type, struct mount *mp, bool donew) 2961 { 2962 const uint8_t *cp; 2963 struct mntoptnames *fp; 2964 struct sbuf sb; 2965 struct statfs *sfp = &mp->mnt_stat; 2966 char *buf; 2967 2968 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT); 2969 if (buf == NULL) 2970 return; 2971 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN); 2972 sbuf_cpy(&sb, "mount-point=\""); 2973 devctl_safe_quote_sb(&sb, sfp->f_mntonname); 2974 sbuf_cat(&sb, "\" mount-dev=\""); 2975 devctl_safe_quote_sb(&sb, sfp->f_mntfromname); 2976 sbuf_cat(&sb, "\" mount-type=\""); 2977 devctl_safe_quote_sb(&sb, sfp->f_fstypename); 2978 sbuf_cat(&sb, "\" fsid=0x"); 2979 cp = (const uint8_t *)&sfp->f_fsid.val[0]; 2980 for (int i = 0; i < sizeof(sfp->f_fsid); i++) 2981 sbuf_printf(&sb, "%02x", cp[i]); 2982 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner); 2983 for (fp = optnames; fp->o_opt != 0; fp++) { 2984 if ((mp->mnt_flag & fp->o_opt) != 0) { 2985 sbuf_cat(&sb, fp->o_name); 2986 sbuf_putc(&sb, ';'); 2987 } 2988 } 2989 sbuf_putc(&sb, '"'); 2990 sbuf_finish(&sb); 2991 2992 /* 2993 * Options are not published because the form of the options depends on 2994 * the file system and may include binary data. In addition, they don't 2995 * necessarily provide enough useful information to be actionable when 2996 * devd processes them. 2997 */ 2998 2999 if (sbuf_error(&sb) == 0) 3000 devctl_notify("VFS", "FS", type, sbuf_data(&sb)); 3001 sbuf_delete(&sb); 3002 free(buf, M_MOUNT); 3003 } 3004 3005 /* 3006 * Force remount specified mount point to read-only. The argument 3007 * must be busied to avoid parallel unmount attempts. 3008 * 3009 * Intended use is to prevent further writes if some metadata 3010 * inconsistency is detected. Note that the function still flushes 3011 * all cached metadata and data for the mount point, which might be 3012 * not always suitable. 3013 */ 3014 int 3015 vfs_remount_ro(struct mount *mp) 3016 { 3017 struct vfsoptlist *opts; 3018 struct vfsopt *opt; 3019 struct vnode *vp_covered, *rootvp; 3020 int error; 3021 3022 vfs_op_enter(mp); 3023 KASSERT(mp->mnt_lockref > 0, 3024 ("vfs_remount_ro: mp %p is not busied", mp)); 3025 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0, 3026 ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp)); 3027 3028 rootvp = NULL; 3029 vp_covered = mp->mnt_vnodecovered; 3030 error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT); 3031 if (error != 0) { 3032 vfs_op_exit(mp); 3033 return (error); 3034 } 3035 VI_LOCK(vp_covered); 3036 if ((vp_covered->v_iflag & VI_MOUNT) != 0) { 3037 VI_UNLOCK(vp_covered); 3038 vput(vp_covered); 3039 vfs_op_exit(mp); 3040 return (EBUSY); 3041 } 3042 vp_covered->v_iflag |= VI_MOUNT; 3043 VI_UNLOCK(vp_covered); 3044 vn_seqc_write_begin(vp_covered); 3045 3046 MNT_ILOCK(mp); 3047 if ((mp->mnt_flag & MNT_RDONLY) != 0) { 3048 MNT_IUNLOCK(mp); 3049 error = EBUSY; 3050 goto out; 3051 } 3052 mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY; 3053 rootvp = vfs_cache_root_clear(mp); 3054 MNT_IUNLOCK(mp); 3055 3056 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO); 3057 TAILQ_INIT(opts); 3058 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO); 3059 opt->name = strdup("ro", M_MOUNT); 3060 opt->value = NULL; 3061 TAILQ_INSERT_TAIL(opts, opt, link); 3062 vfs_mergeopts(opts, mp->mnt_opt); 3063 mp->mnt_optnew = opts; 3064 3065 error = VFS_MOUNT(mp); 3066 3067 if (error == 0) { 3068 MNT_ILOCK(mp); 3069 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE); 3070 MNT_IUNLOCK(mp); 3071 vfs_deallocate_syncvnode(mp); 3072 if (mp->mnt_opt != NULL) 3073 vfs_freeopts(mp->mnt_opt); 3074 mp->mnt_opt = mp->mnt_optnew; 3075 } else { 3076 MNT_ILOCK(mp); 3077 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY); 3078 MNT_IUNLOCK(mp); 3079 vfs_freeopts(mp->mnt_optnew); 3080 } 3081 mp->mnt_optnew = NULL; 3082 3083 out: 3084 vfs_op_exit(mp); 3085 VI_LOCK(vp_covered); 3086 vp_covered->v_iflag &= ~VI_MOUNT; 3087 VI_UNLOCK(vp_covered); 3088 vput(vp_covered); 3089 vn_seqc_write_end(vp_covered); 3090 if (rootvp != NULL) { 3091 vn_seqc_write_end(rootvp); 3092 vrele(rootvp); 3093 } 3094 return (error); 3095 } 3096 3097 /* 3098 * Suspend write operations on all local writeable filesystems. Does 3099 * full sync of them in the process. 3100 * 3101 * Iterate over the mount points in reverse order, suspending most 3102 * recently mounted filesystems first. It handles a case where a 3103 * filesystem mounted from a md(4) vnode-backed device should be 3104 * suspended before the filesystem that owns the vnode. 3105 */ 3106 void 3107 suspend_all_fs(void) 3108 { 3109 struct mount *mp; 3110 int error; 3111 3112 mtx_lock(&mountlist_mtx); 3113 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 3114 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT); 3115 if (error != 0) 3116 continue; 3117 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL || 3118 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 3119 mtx_lock(&mountlist_mtx); 3120 vfs_unbusy(mp); 3121 continue; 3122 } 3123 error = vfs_write_suspend(mp, 0); 3124 if (error == 0) { 3125 MNT_ILOCK(mp); 3126 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0); 3127 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL; 3128 MNT_IUNLOCK(mp); 3129 mtx_lock(&mountlist_mtx); 3130 } else { 3131 printf("suspend of %s failed, error %d\n", 3132 mp->mnt_stat.f_mntonname, error); 3133 mtx_lock(&mountlist_mtx); 3134 vfs_unbusy(mp); 3135 } 3136 } 3137 mtx_unlock(&mountlist_mtx); 3138 } 3139 3140 void 3141 resume_all_fs(void) 3142 { 3143 struct mount *mp; 3144 3145 mtx_lock(&mountlist_mtx); 3146 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3147 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0) 3148 continue; 3149 mtx_unlock(&mountlist_mtx); 3150 MNT_ILOCK(mp); 3151 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0); 3152 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL; 3153 MNT_IUNLOCK(mp); 3154 vfs_write_resume(mp, 0); 3155 mtx_lock(&mountlist_mtx); 3156 vfs_unbusy(mp); 3157 } 3158 mtx_unlock(&mountlist_mtx); 3159 } 3160