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/param.h> 40 #include <sys/conf.h> 41 #include <sys/smp.h> 42 #include <sys/devctl.h> 43 #include <sys/eventhandler.h> 44 #include <sys/fcntl.h> 45 #include <sys/jail.h> 46 #include <sys/kernel.h> 47 #include <sys/ktr.h> 48 #include <sys/libkern.h> 49 #include <sys/limits.h> 50 #include <sys/malloc.h> 51 #include <sys/mount.h> 52 #include <sys/mutex.h> 53 #include <sys/namei.h> 54 #include <sys/priv.h> 55 #include <sys/proc.h> 56 #include <sys/filedesc.h> 57 #include <sys/reboot.h> 58 #include <sys/sbuf.h> 59 #include <sys/stdarg.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 <security/audit/audit.h> 72 #include <security/mac/mac_framework.h> 73 74 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 75 76 static int vfs_domount(struct thread *td, const char *fstype, char *fspath, 77 uint64_t fsflags, bool only_export, bool jail_export, 78 struct vfsoptlist **optlist); 79 static void free_mntarg(struct mntarg *ma); 80 81 static int usermount = 0; 82 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 83 "Unprivileged users may mount and unmount file systems"); 84 85 static bool default_autoro = false; 86 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0, 87 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified"); 88 89 static bool recursive_forced_unmount = false; 90 SYSCTL_BOOL(_vfs, OID_AUTO, recursive_forced_unmount, CTLFLAG_RW, 91 &recursive_forced_unmount, 0, "Recursively unmount stacked upper mounts" 92 " when a file system is forcibly unmounted"); 93 94 static SYSCTL_NODE(_vfs, OID_AUTO, deferred_unmount, 95 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "deferred unmount controls"); 96 97 static unsigned int deferred_unmount_retry_limit = 10; 98 SYSCTL_UINT(_vfs_deferred_unmount, OID_AUTO, retry_limit, CTLFLAG_RW, 99 &deferred_unmount_retry_limit, 0, 100 "Maximum number of retries for deferred unmount failure"); 101 102 static int deferred_unmount_retry_delay_hz; 103 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, retry_delay_hz, CTLFLAG_RW, 104 &deferred_unmount_retry_delay_hz, 0, 105 "Delay in units of [1/kern.hz]s when retrying a failed deferred unmount"); 106 107 static int deferred_unmount_total_retries = 0; 108 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, total_retries, CTLFLAG_RD, 109 &deferred_unmount_total_retries, 0, 110 "Total number of retried deferred unmounts"); 111 112 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 113 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure"); 114 static uma_zone_t mount_zone; 115 116 /* List of mounted filesystems. */ 117 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 118 119 /* For any iteration/modification of mountlist */ 120 struct mtx_padalign __exclusive_cache_line mountlist_mtx; 121 122 EVENTHANDLER_LIST_DEFINE(vfs_mounted); 123 EVENTHANDLER_LIST_DEFINE(vfs_unmounted); 124 125 static void vfs_deferred_unmount(void *arg, int pending); 126 static struct timeout_task deferred_unmount_task; 127 static struct mtx deferred_unmount_lock; 128 MTX_SYSINIT(deferred_unmount, &deferred_unmount_lock, "deferred_unmount", 129 MTX_DEF); 130 static STAILQ_HEAD(, mount) deferred_unmount_list = 131 STAILQ_HEAD_INITIALIZER(deferred_unmount_list); 132 TASKQUEUE_DEFINE_THREAD(deferred_unmount); 133 134 static void mount_devctl_event(const char *type, struct mount *mp, bool donew); 135 136 /* 137 * Global opts, taken by all filesystems 138 */ 139 static const char *global_opts[] = { 140 "errmsg", 141 "fstype", 142 "fspath", 143 "ro", 144 "rw", 145 "nosuid", 146 "noexec", 147 NULL 148 }; 149 150 static int 151 mount_init(void *mem, int size, int flags) 152 { 153 struct mount *mp; 154 155 mp = (struct mount *)mem; 156 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 157 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF); 158 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 159 lockinit(&mp->mnt_renamelock, PVFS, "rename", 0, 0); 160 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO); 161 mp->mnt_ref = 0; 162 mp->mnt_vfs_ops = 1; 163 mp->mnt_rootvnode = NULL; 164 return (0); 165 } 166 167 static void 168 mount_fini(void *mem, int size) 169 { 170 struct mount *mp; 171 172 mp = (struct mount *)mem; 173 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu); 174 lockdestroy(&mp->mnt_renamelock); 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 freeuio(auio); 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 mp->mnt_op = vfsp->vfc_vfsops; 687 mp->mnt_vfc = vfsp; 688 mp->mnt_stat.f_type = vfsp->vfc_typenum; 689 mp->mnt_gen++; 690 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 691 mp->mnt_vnodecovered = vp; 692 mp->mnt_cred = crdup(cred); 693 mp->mnt_stat.f_owner = cred->cr_uid; 694 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 695 mp->mnt_iosize_max = DFLTPHYS; 696 #ifdef MAC 697 mac_mount_init(mp); 698 mac_mount_create(cred, mp); 699 #endif 700 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 701 mp->mnt_upper_pending = 0; 702 TAILQ_INIT(&mp->mnt_uppers); 703 TAILQ_INIT(&mp->mnt_notify); 704 mp->mnt_taskqueue_flags = 0; 705 mp->mnt_unmount_retries = 0; 706 return (mp); 707 } 708 709 /* 710 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 711 */ 712 void 713 vfs_mount_destroy(struct mount *mp) 714 { 715 716 MPPASS(mp->mnt_vfs_ops != 0, mp); 717 718 vfs_assert_mount_counters(mp); 719 720 MNT_ILOCK(mp); 721 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 722 if (mp->mnt_kern_flag & MNTK_MWAIT) { 723 mp->mnt_kern_flag &= ~MNTK_MWAIT; 724 wakeup(mp); 725 } 726 while (mp->mnt_ref) 727 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 728 KASSERT(mp->mnt_ref == 0, 729 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 730 __FILE__, __LINE__)); 731 MPPASS(mp->mnt_writeopcount == 0, mp); 732 MPPASS(mp->mnt_secondary_writes == 0, mp); 733 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 734 struct vnode *vp; 735 736 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 737 vn_printf(vp, "dangling vnode "); 738 panic("unmount: dangling vnode"); 739 } 740 KASSERT(mp->mnt_upper_pending == 0, ("mnt_upper_pending")); 741 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers")); 742 KASSERT(TAILQ_EMPTY(&mp->mnt_notify), ("mnt_notify")); 743 MPPASS(mp->mnt_nvnodelistsize == 0, mp); 744 MPPASS(mp->mnt_lazyvnodelistsize == 0, mp); 745 MPPASS(mp->mnt_lockref == 0, mp); 746 MNT_IUNLOCK(mp); 747 748 MPASSERT(mp->mnt_vfs_ops == 1, mp, 749 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops)); 750 751 MPASSERT(mp->mnt_rootvnode == NULL, mp, 752 ("mount point still has a root vnode %p", mp->mnt_rootvnode)); 753 754 if (mp->mnt_vnodecovered != NULL) 755 vrele(mp->mnt_vnodecovered); 756 #ifdef MAC 757 mac_mount_destroy(mp); 758 #endif 759 if (mp->mnt_opt != NULL) 760 vfs_freeopts(mp->mnt_opt); 761 if (mp->mnt_exjail != NULL) { 762 atomic_subtract_int(&mp->mnt_exjail->cr_prison->pr_exportcnt, 763 1); 764 crfree(mp->mnt_exjail); 765 } 766 if (mp->mnt_export != NULL) { 767 vfs_free_addrlist(mp->mnt_export); 768 free(mp->mnt_export, M_MOUNT); 769 } 770 vfsconf_lock(); 771 mp->mnt_vfc->vfc_refcount--; 772 vfsconf_unlock(); 773 crfree(mp->mnt_cred); 774 uma_zfree(mount_zone, mp); 775 } 776 777 static bool 778 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error) 779 { 780 /* This is an upgrade of an exisiting mount. */ 781 if ((fsflags & MNT_UPDATE) != 0) 782 return (false); 783 /* This is already an R/O mount. */ 784 if ((fsflags & MNT_RDONLY) != 0) 785 return (false); 786 787 switch (error) { 788 case ENODEV: /* generic, geom, ... */ 789 case EACCES: /* cam/scsi, ... */ 790 case EROFS: /* md, mmcsd, ... */ 791 /* 792 * These errors can be returned by the storage layer to signal 793 * that the media is read-only. No harm in the R/O mount 794 * attempt if the error was returned for some other reason. 795 */ 796 return (true); 797 default: 798 return (false); 799 } 800 } 801 802 int 803 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions) 804 { 805 struct vfsoptlist *optlist; 806 struct vfsopt *opt, *tmp_opt; 807 char *fstype, *fspath, *errmsg; 808 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 809 bool autoro, has_nonexport, only_export, jail_export; 810 811 errmsg = fspath = NULL; 812 errmsg_len = fspathlen = 0; 813 errmsg_pos = -1; 814 autoro = default_autoro; 815 816 error = vfs_buildopts(fsoptions, &optlist); 817 if (error) 818 return (error); 819 820 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 821 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 822 823 /* 824 * We need these two options before the others, 825 * and they are mandatory for any filesystem. 826 * Ensure they are NUL terminated as well. 827 */ 828 fstypelen = 0; 829 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 830 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') { 831 error = EINVAL; 832 if (errmsg != NULL) 833 strncpy(errmsg, "Invalid fstype", errmsg_len); 834 goto bail; 835 } 836 fspathlen = 0; 837 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 838 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') { 839 error = EINVAL; 840 if (errmsg != NULL) 841 strncpy(errmsg, "Invalid fspath", errmsg_len); 842 goto bail; 843 } 844 845 /* 846 * Check to see that "export" is only used with the "update", "fstype", 847 * "fspath", "from" and "errmsg" options when in a vnet jail. 848 * These are the ones used to set/update exports by mountd(8). 849 * If only the above options are set in a jail that can run mountd(8), 850 * then the jail_export argument of vfs_domount() will be true. 851 * When jail_export is true, the vfs_suser() check does not cause 852 * failure, but limits the update to exports only. 853 * This allows mountd(8) running within the vnet jail 854 * to export file systems visible within the jail, but 855 * mounted outside of the jail. 856 */ 857 /* 858 * We need to see if we have the "update" option 859 * before we call vfs_domount(), since vfs_domount() has special 860 * logic based on MNT_UPDATE. This is very important 861 * when we want to update the root filesystem. 862 */ 863 has_nonexport = false; 864 only_export = false; 865 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 866 int do_freeopt = 0; 867 868 if (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 only_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 * only_export is set to true only if exports are being 992 * updated and nothing else is being updated. 993 */ 994 if (has_nonexport) 995 only_export = false; 996 /* 997 * If only_export is true and the caller is running within a 998 * vnet prison that can run mountd(8), set jail_export true. 999 */ 1000 jail_export = false; 1001 if (only_export && jailed(td->td_ucred) && 1002 prison_check_nfsd(td->td_ucred)) 1003 jail_export = true; 1004 1005 error = vfs_domount(td, fstype, fspath, fsflags, only_export, 1006 jail_export, &optlist); 1007 if (error == ENODEV) { 1008 error = EINVAL; 1009 if (errmsg != NULL) 1010 strncpy(errmsg, "Invalid fstype", errmsg_len); 1011 goto bail; 1012 } 1013 1014 /* 1015 * See if we can mount in the read-only mode if the error code suggests 1016 * that it could be possible and the mount options allow for that. 1017 * Never try it if "[no]{ro|rw}" has been explicitly requested and not 1018 * overridden by "autoro". 1019 */ 1020 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) { 1021 printf("%s: R/W mount failed, possibly R/O media," 1022 " trying R/O mount\n", __func__); 1023 fsflags |= MNT_RDONLY; 1024 error = vfs_domount(td, fstype, fspath, fsflags, only_export, 1025 jail_export, &optlist); 1026 } 1027 bail: 1028 /* copyout the errmsg */ 1029 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 1030 && errmsg_len > 0 && errmsg != NULL) { 1031 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 1032 bcopy(errmsg, 1033 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 1034 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 1035 } else { 1036 (void)copyout(errmsg, 1037 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 1038 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 1039 } 1040 } 1041 1042 if (optlist != NULL) 1043 vfs_freeopts(optlist); 1044 return (error); 1045 } 1046 1047 /* 1048 * Old mount API. 1049 */ 1050 #ifndef _SYS_SYSPROTO_H_ 1051 struct mount_args { 1052 char *type; 1053 char *path; 1054 int flags; 1055 caddr_t data; 1056 }; 1057 #endif 1058 /* ARGSUSED */ 1059 int 1060 sys_mount(struct thread *td, struct mount_args *uap) 1061 { 1062 char *fstype; 1063 struct vfsconf *vfsp = NULL; 1064 struct mntarg *ma = NULL; 1065 uint64_t flags; 1066 int error; 1067 1068 /* 1069 * Mount flags are now 64-bits. On 32-bit architectures only 1070 * 32-bits are passed in, but from here on everything handles 1071 * 64-bit flags correctly. 1072 */ 1073 flags = uap->flags; 1074 1075 AUDIT_ARG_FFLAGS(flags); 1076 1077 /* 1078 * Filter out MNT_ROOTFS. We do not want clients of mount() in 1079 * userspace to set this flag, but we must filter it out if we want 1080 * MNT_UPDATE on the root file system to work. 1081 * MNT_ROOTFS should only be set by the kernel when mounting its 1082 * root file system. 1083 */ 1084 flags &= ~MNT_ROOTFS; 1085 1086 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 1087 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 1088 if (error) { 1089 free(fstype, M_TEMP); 1090 return (error); 1091 } 1092 1093 AUDIT_ARG_TEXT(fstype); 1094 vfsp = vfs_byname_kld(fstype, td, &error); 1095 free(fstype, M_TEMP); 1096 if (vfsp == NULL) 1097 return (EINVAL); 1098 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 && 1099 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) || 1100 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 && 1101 vfsp->vfc_vfsops->vfs_cmount == NULL)) 1102 return (EOPNOTSUPP); 1103 1104 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN); 1105 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 1106 ma = mount_argb(ma, flags & MNT_RDONLY, "noro"); 1107 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid"); 1108 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec"); 1109 1110 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0) 1111 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags)); 1112 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags)); 1113 } 1114 1115 /* 1116 * vfs_domount_first(): first file system mount (not update) 1117 */ 1118 static int 1119 vfs_domount_first( 1120 struct thread *td, /* Calling thread. */ 1121 struct vfsconf *vfsp, /* File system type. */ 1122 char *fspath, /* Mount path. */ 1123 struct vnode *vp, /* Vnode to be covered. */ 1124 uint64_t fsflags, /* Flags common to all filesystems. */ 1125 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1126 ) 1127 { 1128 struct vattr va; 1129 struct mount *mp; 1130 struct vnode *newdp, *rootvp; 1131 int error, error1; 1132 bool unmounted; 1133 1134 ASSERT_VOP_ELOCKED(vp, __func__); 1135 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 1136 1137 /* 1138 * If the jail of the calling thread lacks permission for this type of 1139 * file system, or is trying to cover its own root, deny immediately. 1140 */ 1141 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred, 1142 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) { 1143 vput(vp); 1144 vfs_unref_vfsconf(vfsp); 1145 return (EPERM); 1146 } 1147 1148 /* 1149 * If the user is not root, ensure that they own the directory 1150 * onto which we are attempting to mount. 1151 */ 1152 error = VOP_GETATTR(vp, &va, td->td_ucred); 1153 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 1154 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN); 1155 if (error == 0) 1156 error = vinvalbuf(vp, V_SAVE, 0, 0); 1157 if (vfsp->vfc_flags & VFCF_FILEMOUNT) { 1158 if (error == 0 && vp->v_type != VDIR && vp->v_type != VREG) 1159 error = EINVAL; 1160 /* 1161 * For file mounts, ensure that there is only one hardlink to the file. 1162 */ 1163 if (error == 0 && vp->v_type == VREG && va.va_nlink != 1) 1164 error = EINVAL; 1165 } else { 1166 if (error == 0 && vp->v_type != VDIR) 1167 error = ENOTDIR; 1168 } 1169 if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0) 1170 error = vn_dir_check_empty(vp); 1171 if (error == 0) { 1172 VI_LOCK(vp); 1173 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 1174 vp->v_iflag |= VI_MOUNT; 1175 else 1176 error = EBUSY; 1177 VI_UNLOCK(vp); 1178 } 1179 if (error != 0) { 1180 vput(vp); 1181 vfs_unref_vfsconf(vfsp); 1182 return (error); 1183 } 1184 vn_seqc_write_begin(vp); 1185 VOP_UNLOCK(vp); 1186 1187 /* Allocate and initialize the filesystem. */ 1188 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 1189 /* XXXMAC: pass to vfs_mount_alloc? */ 1190 mp->mnt_optnew = *optlist; 1191 /* Set the mount level flags. */ 1192 mp->mnt_flag = (fsflags & 1193 (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY | MNT_FORCE)); 1194 1195 /* 1196 * Mount the filesystem. 1197 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1198 * get. No freeing of cn_pnbuf. 1199 */ 1200 error1 = 0; 1201 unmounted = true; 1202 if ((error = VFS_MOUNT(mp)) != 0 || 1203 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 || 1204 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) { 1205 rootvp = NULL; 1206 if (error1 != 0) { 1207 MPASS(error == 0); 1208 rootvp = vfs_cache_root_clear(mp); 1209 if (rootvp != NULL) { 1210 vhold(rootvp); 1211 vrele(rootvp); 1212 } 1213 (void)vn_start_write(NULL, &mp, V_WAIT); 1214 MNT_ILOCK(mp); 1215 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF; 1216 MNT_IUNLOCK(mp); 1217 VFS_PURGE(mp); 1218 error = VFS_UNMOUNT(mp, 0); 1219 vn_finished_write(mp); 1220 if (error != 0) { 1221 printf( 1222 "failed post-mount (%d): rollback unmount returned %d\n", 1223 error1, error); 1224 unmounted = false; 1225 } 1226 error = error1; 1227 } 1228 vfs_unbusy(mp); 1229 mp->mnt_vnodecovered = NULL; 1230 if (unmounted) { 1231 /* XXXKIB wait for mnt_lockref drain? */ 1232 vfs_mount_destroy(mp); 1233 } 1234 VI_LOCK(vp); 1235 vp->v_iflag &= ~VI_MOUNT; 1236 VI_UNLOCK(vp); 1237 if (rootvp != NULL) { 1238 vn_seqc_write_end(rootvp); 1239 vdrop(rootvp); 1240 } 1241 vn_seqc_write_end(vp); 1242 vrele(vp); 1243 return (error); 1244 } 1245 vn_seqc_write_begin(newdp); 1246 VOP_UNLOCK(newdp); 1247 1248 if (mp->mnt_opt != NULL) 1249 vfs_freeopts(mp->mnt_opt); 1250 mp->mnt_opt = mp->mnt_optnew; 1251 *optlist = NULL; 1252 1253 /* 1254 * Prevent external consumers of mount options from reading mnt_optnew. 1255 */ 1256 mp->mnt_optnew = NULL; 1257 1258 MNT_ILOCK(mp); 1259 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1260 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1261 mp->mnt_kern_flag |= MNTK_ASYNC; 1262 else 1263 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1264 MNT_IUNLOCK(mp); 1265 1266 /* 1267 * VIRF_MOUNTPOINT and v_mountedhere need to be set under the 1268 * vp lock to satisfy vfs_lookup() requirements. 1269 */ 1270 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY); 1271 VI_LOCK(vp); 1272 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT); 1273 vp->v_mountedhere = mp; 1274 VI_UNLOCK(vp); 1275 VOP_UNLOCK(vp); 1276 cache_purge(vp); 1277 1278 /* 1279 * We need to lock both vnodes. 1280 * 1281 * Use vn_lock_pair to avoid establishing an ordering between vnodes 1282 * from different filesystems. 1283 */ 1284 vn_lock_pair(vp, false, LK_EXCLUSIVE, newdp, false, LK_EXCLUSIVE); 1285 1286 VI_LOCK(vp); 1287 vp->v_iflag &= ~VI_MOUNT; 1288 VI_UNLOCK(vp); 1289 /* Place the new filesystem at the end of the mount list. */ 1290 mtx_lock(&mountlist_mtx); 1291 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1292 mtx_unlock(&mountlist_mtx); 1293 vfs_event_signal(NULL, VQ_MOUNT, 0); 1294 VOP_UNLOCK(vp); 1295 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td); 1296 VOP_UNLOCK(newdp); 1297 mount_devctl_event("MOUNT", mp, false); 1298 mountcheckdirs(vp, newdp); 1299 vn_seqc_write_end(vp); 1300 vn_seqc_write_end(newdp); 1301 vrele(newdp); 1302 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1303 vfs_allocate_syncvnode(mp); 1304 vfs_op_exit(mp); 1305 vfs_unbusy(mp); 1306 return (0); 1307 } 1308 1309 /* 1310 * vfs_domount_update(): update of mounted file system 1311 */ 1312 static int 1313 vfs_domount_update( 1314 struct thread *td, /* Calling thread. */ 1315 struct vnode *vp, /* Mount point vnode. */ 1316 uint64_t fsflags, /* Flags common to all filesystems. */ 1317 bool only_export, /* Got export option. */ 1318 bool jail_export, /* Got export option in vnet prison. */ 1319 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1320 ) 1321 { 1322 struct export_args export; 1323 struct o2export_args o2export; 1324 struct vnode *rootvp; 1325 void *bufp; 1326 struct mount *mp; 1327 int error, export_error, i, len, fsid_up_len; 1328 uint64_t flag, mnt_union; 1329 gid_t *grps; 1330 fsid_t *fsid_up; 1331 bool vfs_suser_failed; 1332 1333 ASSERT_VOP_ELOCKED(vp, __func__); 1334 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 1335 mp = vp->v_mount; 1336 1337 if ((vp->v_vflag & VV_ROOT) == 0) { 1338 if (vfs_copyopt(*optlist, "export", &export, sizeof(export)) 1339 == 0) 1340 error = EXDEV; 1341 else 1342 error = EINVAL; 1343 vput(vp); 1344 return (error); 1345 } 1346 1347 /* 1348 * We only allow the filesystem to be reloaded if it 1349 * is currently mounted read-only. 1350 */ 1351 flag = mp->mnt_flag; 1352 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 1353 vput(vp); 1354 return (EOPNOTSUPP); /* Needs translation */ 1355 } 1356 /* 1357 * Only privileged root, or (if MNT_USER is set) the user that 1358 * did the original mount is permitted to update it. 1359 */ 1360 /* 1361 * For the case of mountd(8) doing exports in a jail, the vfs_suser() 1362 * call does not cause failure. vfs_domount() has already checked 1363 * that "root" is doing this and vfs_suser() will fail when 1364 * the file system has been mounted outside the jail. 1365 * jail_export set true indicates that "export" is not mixed 1366 * with other options that change mount behaviour. 1367 */ 1368 vfs_suser_failed = false; 1369 error = vfs_suser(mp, td); 1370 if (jail_export && error != 0) { 1371 error = 0; 1372 vfs_suser_failed = true; 1373 } 1374 if (error != 0) { 1375 vput(vp); 1376 return (error); 1377 } 1378 if (vfs_busy(mp, MBF_NOWAIT)) { 1379 vput(vp); 1380 return (EBUSY); 1381 } 1382 VI_LOCK(vp); 1383 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 1384 VI_UNLOCK(vp); 1385 vfs_unbusy(mp); 1386 vput(vp); 1387 return (EBUSY); 1388 } 1389 vp->v_iflag |= VI_MOUNT; 1390 VI_UNLOCK(vp); 1391 VOP_UNLOCK(vp); 1392 1393 rootvp = NULL; 1394 vfs_op_enter(mp); 1395 vn_seqc_write_begin(vp); 1396 1397 if (vfs_getopt(*optlist, "fsid", (void **)&fsid_up, 1398 &fsid_up_len) == 0) { 1399 if (fsid_up_len != sizeof(*fsid_up)) { 1400 error = EINVAL; 1401 goto end; 1402 } 1403 if (fsidcmp(fsid_up, &mp->mnt_stat.f_fsid) != 0) { 1404 error = ENOENT; 1405 goto end; 1406 } 1407 vfs_deleteopt(*optlist, "fsid"); 1408 } 1409 1410 mnt_union = 0; 1411 MNT_ILOCK(mp); 1412 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1413 MNT_IUNLOCK(mp); 1414 error = EBUSY; 1415 goto end; 1416 } 1417 if (vfs_suser_failed) { 1418 KASSERT((fsflags & (MNT_EXPORTED | MNT_UPDATE)) == 1419 (MNT_EXPORTED | MNT_UPDATE), 1420 ("%s: jailed export did not set expected fsflags", 1421 __func__)); 1422 /* 1423 * For this case, only MNT_UPDATE and 1424 * MNT_EXPORTED have been set in fsflags 1425 * by the options. Only set MNT_UPDATE, 1426 * since that is the one that would be set 1427 * when set in fsflags, below. 1428 */ 1429 mp->mnt_flag |= MNT_UPDATE; 1430 } else { 1431 mp->mnt_flag &= ~MNT_UPDATEMASK; 1432 if ((mp->mnt_flag & MNT_UNION) == 0 && 1433 (fsflags & MNT_UNION) != 0) { 1434 fsflags &= ~MNT_UNION; 1435 mnt_union = MNT_UNION; 1436 } 1437 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 1438 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 1439 if ((mp->mnt_flag & MNT_ASYNC) == 0) 1440 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1441 } 1442 rootvp = vfs_cache_root_clear(mp); 1443 MNT_IUNLOCK(mp); 1444 mp->mnt_optnew = *optlist; 1445 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 1446 1447 /* 1448 * Mount the filesystem. 1449 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1450 * get. No freeing of cn_pnbuf. 1451 */ 1452 /* 1453 * When only updating mount exports, VFS_MOUNT() does not need to 1454 * be called, as indicated by only_export being set true. 1455 * For the case of mountd(8) doing exports from within a vnet jail, 1456 * "from" is typically not set correctly such that VFS_MOUNT() will 1457 * return ENOENT. For ZFS, there is a locking bug which can result in 1458 * deadlock if VFS_MOUNT() is called when extended attributes are 1459 * being updated. 1460 */ 1461 error = 0; 1462 if (!only_export) 1463 error = VFS_MOUNT(mp); 1464 1465 export_error = 0; 1466 /* Process the export option. */ 1467 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp, 1468 &len) == 0) { 1469 /* Assume that there is only 1 ABI for each length. */ 1470 switch (len) { 1471 case (sizeof(struct oexport_args)): 1472 bzero(&o2export, sizeof(o2export)); 1473 /* FALLTHROUGH */ 1474 case (sizeof(o2export)): 1475 bcopy(bufp, &o2export, len); 1476 export.ex_flags = (uint64_t)o2export.ex_flags; 1477 export.ex_root = o2export.ex_root; 1478 export.ex_uid = o2export.ex_anon.cr_uid; 1479 export.ex_groups = NULL; 1480 export.ex_ngroups = o2export.ex_anon.cr_ngroups; 1481 if (export.ex_ngroups > 0) { 1482 if (export.ex_ngroups <= XU_NGROUPS) { 1483 export.ex_groups = malloc( 1484 export.ex_ngroups * sizeof(gid_t), 1485 M_TEMP, M_WAITOK); 1486 for (i = 0; i < export.ex_ngroups; i++) 1487 export.ex_groups[i] = 1488 o2export.ex_anon.cr_groups[i]; 1489 } else 1490 export_error = EINVAL; 1491 } else if (export.ex_ngroups < 0) 1492 export_error = EINVAL; 1493 export.ex_addr = o2export.ex_addr; 1494 export.ex_addrlen = o2export.ex_addrlen; 1495 export.ex_mask = o2export.ex_mask; 1496 export.ex_masklen = o2export.ex_masklen; 1497 export.ex_indexfile = o2export.ex_indexfile; 1498 export.ex_numsecflavors = o2export.ex_numsecflavors; 1499 if (export.ex_numsecflavors < MAXSECFLAVORS) { 1500 for (i = 0; i < export.ex_numsecflavors; i++) 1501 export.ex_secflavors[i] = 1502 o2export.ex_secflavors[i]; 1503 } else 1504 export_error = EINVAL; 1505 if (export_error == 0) 1506 export_error = vfs_export(mp, &export, true); 1507 free(export.ex_groups, M_TEMP); 1508 break; 1509 case (sizeof(export)): 1510 bcopy(bufp, &export, len); 1511 grps = NULL; 1512 if (export.ex_ngroups > 0) { 1513 if (export.ex_ngroups <= ngroups_max + 1) { 1514 grps = malloc(export.ex_ngroups * 1515 sizeof(gid_t), M_TEMP, M_WAITOK); 1516 export_error = copyin(export.ex_groups, 1517 grps, export.ex_ngroups * 1518 sizeof(gid_t)); 1519 if (export_error == 0) 1520 export.ex_groups = grps; 1521 } else 1522 export_error = EINVAL; 1523 } else if (export.ex_ngroups == 0) 1524 export.ex_groups = NULL; 1525 else 1526 export_error = EINVAL; 1527 if (export_error == 0) 1528 export_error = vfs_export(mp, &export, true); 1529 free(grps, M_TEMP); 1530 break; 1531 default: 1532 export_error = EINVAL; 1533 break; 1534 } 1535 } 1536 1537 MNT_ILOCK(mp); 1538 if (error == 0) { 1539 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 1540 MNT_SNAPSHOT); 1541 mp->mnt_flag |= mnt_union; 1542 } else { 1543 /* 1544 * If we fail, restore old mount flags. MNT_QUOTA is special, 1545 * because it is not part of MNT_UPDATEMASK, but it could have 1546 * changed in the meantime if quotactl(2) was called. 1547 * All in all we want current value of MNT_QUOTA, not the old 1548 * one. 1549 */ 1550 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 1551 } 1552 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1553 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1554 mp->mnt_kern_flag |= MNTK_ASYNC; 1555 else 1556 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1557 MNT_IUNLOCK(mp); 1558 1559 if (error != 0) 1560 goto end; 1561 1562 mount_devctl_event("REMOUNT", mp, true); 1563 if (mp->mnt_opt != NULL) 1564 vfs_freeopts(mp->mnt_opt); 1565 mp->mnt_opt = mp->mnt_optnew; 1566 *optlist = NULL; 1567 (void)VFS_STATFS(mp, &mp->mnt_stat); 1568 /* 1569 * Prevent external consumers of mount options from reading 1570 * mnt_optnew. 1571 */ 1572 mp->mnt_optnew = NULL; 1573 1574 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1575 vfs_allocate_syncvnode(mp); 1576 else 1577 vfs_deallocate_syncvnode(mp); 1578 end: 1579 vfs_op_exit(mp); 1580 if (rootvp != NULL) { 1581 vn_seqc_write_end(rootvp); 1582 vrele(rootvp); 1583 } 1584 vn_seqc_write_end(vp); 1585 vfs_unbusy(mp); 1586 VI_LOCK(vp); 1587 vp->v_iflag &= ~VI_MOUNT; 1588 VI_UNLOCK(vp); 1589 vrele(vp); 1590 return (error != 0 ? error : export_error); 1591 } 1592 1593 /* 1594 * vfs_domount(): actually attempt a filesystem mount. 1595 */ 1596 static int 1597 vfs_domount( 1598 struct thread *td, /* Calling thread. */ 1599 const char *fstype, /* Filesystem type. */ 1600 char *fspath, /* Mount path. */ 1601 uint64_t fsflags, /* Flags common to all filesystems. */ 1602 bool only_export, /* Got export option. */ 1603 bool jail_export, /* Got export option in vnet prison. */ 1604 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1605 ) 1606 { 1607 struct vfsconf *vfsp; 1608 struct nameidata nd; 1609 struct vnode *vp; 1610 char *pathbuf; 1611 int error; 1612 1613 /* 1614 * Be ultra-paranoid about making sure the type and fspath 1615 * variables will fit in our mp buffers, including the 1616 * terminating NUL. 1617 */ 1618 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1619 return (ENAMETOOLONG); 1620 1621 if (jail_export) { 1622 error = priv_check(td, PRIV_NFS_DAEMON); 1623 if (error) 1624 return (error); 1625 } else if (jailed(td->td_ucred) || usermount == 0) { 1626 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1627 return (error); 1628 } 1629 1630 /* 1631 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1632 */ 1633 if (fsflags & MNT_EXPORTED) { 1634 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1635 if (error) 1636 return (error); 1637 } 1638 if (fsflags & MNT_SUIDDIR) { 1639 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1640 if (error) 1641 return (error); 1642 } 1643 /* 1644 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1645 */ 1646 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1647 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1648 fsflags |= MNT_NOSUID | MNT_USER; 1649 } 1650 1651 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1652 vfsp = NULL; 1653 if ((fsflags & MNT_UPDATE) == 0) { 1654 /* Don't try to load KLDs if we're mounting the root. */ 1655 if (fsflags & MNT_ROOTFS) { 1656 if ((vfsp = vfs_byname(fstype)) == NULL) 1657 return (ENODEV); 1658 } else { 1659 if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL) 1660 return (error); 1661 } 1662 } 1663 1664 /* 1665 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1666 */ 1667 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT, 1668 UIO_SYSSPACE, fspath); 1669 error = namei(&nd); 1670 if (error != 0) 1671 return (error); 1672 vp = nd.ni_vp; 1673 /* 1674 * Don't allow stacking file mounts to work around problems with the way 1675 * that namei sets nd.ni_dvp to vp_crossmp for these. 1676 */ 1677 if (vp->v_type == VREG) 1678 fsflags |= MNT_NOCOVER; 1679 if ((fsflags & MNT_UPDATE) == 0) { 1680 if ((vp->v_vflag & VV_ROOT) != 0 && 1681 (fsflags & MNT_NOCOVER) != 0) { 1682 vput(vp); 1683 error = EBUSY; 1684 goto out; 1685 } 1686 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1687 strcpy(pathbuf, fspath); 1688 /* 1689 * Note: we allow any vnode type here. If the path sanity check 1690 * succeeds, the type will be validated in vfs_domount_first 1691 * above. 1692 */ 1693 if (vp->v_type == VDIR) 1694 error = vn_path_to_global_path(td, vp, pathbuf, 1695 MNAMELEN); 1696 else 1697 error = vn_path_to_global_path_hardlink(td, vp, 1698 nd.ni_dvp, pathbuf, MNAMELEN, 1699 nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen); 1700 if (error == 0) { 1701 error = vfs_domount_first(td, vfsp, pathbuf, vp, 1702 fsflags, optlist); 1703 } 1704 free(pathbuf, M_TEMP); 1705 } else 1706 error = vfs_domount_update(td, vp, fsflags, only_export, 1707 jail_export, optlist); 1708 1709 out: 1710 NDFREE_PNBUF(&nd); 1711 vrele(nd.ni_dvp); 1712 1713 return (error); 1714 } 1715 1716 /* 1717 * Unmount a filesystem. 1718 * 1719 * Note: unmount takes a path to the vnode mounted on as argument, not 1720 * special file (as before). 1721 */ 1722 #ifndef _SYS_SYSPROTO_H_ 1723 struct unmount_args { 1724 char *path; 1725 int flags; 1726 }; 1727 #endif 1728 /* ARGSUSED */ 1729 int 1730 sys_unmount(struct thread *td, struct unmount_args *uap) 1731 { 1732 1733 return (kern_unmount(td, uap->path, (unsigned)uap->flags)); 1734 } 1735 1736 int 1737 kern_unmount(struct thread *td, const char *path, uint64_t flags) 1738 { 1739 struct nameidata nd; 1740 struct mount *mp; 1741 char *fsidbuf, *pathbuf; 1742 fsid_t fsid; 1743 int error; 1744 1745 AUDIT_ARG_VALUE(flags); 1746 if ((flags & (MNT_DEFERRED | MNT_RECURSE)) != 0) 1747 return (EINVAL); 1748 if (jailed(td->td_ucred) || usermount == 0) { 1749 error = priv_check(td, PRIV_VFS_UNMOUNT); 1750 if (error) 1751 return (error); 1752 } 1753 1754 if (flags & MNT_BYFSID) { 1755 fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1756 error = copyinstr(path, fsidbuf, MNAMELEN, NULL); 1757 if (error) { 1758 free(fsidbuf, M_TEMP); 1759 return (error); 1760 } 1761 1762 AUDIT_ARG_TEXT(fsidbuf); 1763 /* Decode the filesystem ID. */ 1764 if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) { 1765 free(fsidbuf, M_TEMP); 1766 return (EINVAL); 1767 } 1768 1769 mp = vfs_getvfs(&fsid); 1770 free(fsidbuf, M_TEMP); 1771 if (mp == NULL) { 1772 return (ENOENT); 1773 } 1774 } else { 1775 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1776 error = copyinstr(path, pathbuf, MNAMELEN, NULL); 1777 if (error) { 1778 free(pathbuf, M_TEMP); 1779 return (error); 1780 } 1781 1782 /* 1783 * Try to find global path for path argument. 1784 */ 1785 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1786 UIO_SYSSPACE, pathbuf); 1787 if (namei(&nd) == 0) { 1788 NDFREE_PNBUF(&nd); 1789 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf, 1790 MNAMELEN); 1791 if (error == 0) 1792 vput(nd.ni_vp); 1793 } 1794 mtx_lock(&mountlist_mtx); 1795 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1796 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) { 1797 vfs_ref(mp); 1798 break; 1799 } 1800 } 1801 mtx_unlock(&mountlist_mtx); 1802 free(pathbuf, M_TEMP); 1803 if (mp == NULL) { 1804 /* 1805 * Previously we returned ENOENT for a nonexistent path and 1806 * EINVAL for a non-mountpoint. We cannot tell these apart 1807 * now, so in the !MNT_BYFSID case return the more likely 1808 * EINVAL for compatibility. 1809 */ 1810 return (EINVAL); 1811 } 1812 } 1813 1814 /* 1815 * Don't allow unmounting the root filesystem. 1816 */ 1817 if (mp->mnt_flag & MNT_ROOTFS) { 1818 vfs_rel(mp); 1819 return (EINVAL); 1820 } 1821 error = dounmount(mp, flags, td); 1822 return (error); 1823 } 1824 1825 /* 1826 * Return error if any of the vnodes, ignoring the root vnode 1827 * and the syncer vnode, have non-zero usecount. 1828 * 1829 * This function is purely advisory - it can return false positives 1830 * and negatives. 1831 */ 1832 static int 1833 vfs_check_usecounts(struct mount *mp) 1834 { 1835 struct vnode *vp, *mvp; 1836 1837 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1838 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON && 1839 vp->v_usecount != 0) { 1840 VI_UNLOCK(vp); 1841 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1842 return (EBUSY); 1843 } 1844 VI_UNLOCK(vp); 1845 } 1846 1847 return (0); 1848 } 1849 1850 static void 1851 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags) 1852 { 1853 1854 mtx_assert(MNT_MTX(mp), MA_OWNED); 1855 mp->mnt_kern_flag &= ~mntkflags; 1856 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) { 1857 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1858 wakeup(mp); 1859 } 1860 vfs_op_exit_locked(mp); 1861 MNT_IUNLOCK(mp); 1862 if (coveredvp != NULL) { 1863 VOP_UNLOCK(coveredvp); 1864 vdrop(coveredvp); 1865 } 1866 vn_finished_write(mp); 1867 vfs_rel(mp); 1868 } 1869 1870 /* 1871 * There are various reference counters associated with the mount point. 1872 * Normally it is permitted to modify them without taking the mnt ilock, 1873 * but this behavior can be temporarily disabled if stable value is needed 1874 * or callers are expected to block (e.g. to not allow new users during 1875 * forced unmount). 1876 */ 1877 void 1878 vfs_op_enter(struct mount *mp) 1879 { 1880 struct mount_pcpu *mpcpu; 1881 int cpu; 1882 1883 MNT_ILOCK(mp); 1884 mp->mnt_vfs_ops++; 1885 if (mp->mnt_vfs_ops > 1) { 1886 MNT_IUNLOCK(mp); 1887 return; 1888 } 1889 vfs_op_barrier_wait(mp); 1890 CPU_FOREACH(cpu) { 1891 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1892 1893 mp->mnt_ref += mpcpu->mntp_ref; 1894 mpcpu->mntp_ref = 0; 1895 1896 mp->mnt_lockref += mpcpu->mntp_lockref; 1897 mpcpu->mntp_lockref = 0; 1898 1899 mp->mnt_writeopcount += mpcpu->mntp_writeopcount; 1900 mpcpu->mntp_writeopcount = 0; 1901 } 1902 MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 && 1903 mp->mnt_writeopcount >= 0, mp, 1904 ("invalid count(s): ref %d lockref %d writeopcount %d", 1905 mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount)); 1906 MNT_IUNLOCK(mp); 1907 vfs_assert_mount_counters(mp); 1908 } 1909 1910 void 1911 vfs_op_exit_locked(struct mount *mp) 1912 { 1913 1914 mtx_assert(MNT_MTX(mp), MA_OWNED); 1915 1916 MPASSERT(mp->mnt_vfs_ops > 0, mp, 1917 ("invalid vfs_ops count %d", mp->mnt_vfs_ops)); 1918 MPASSERT(mp->mnt_vfs_ops > 1 || 1919 (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp, 1920 ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops)); 1921 mp->mnt_vfs_ops--; 1922 } 1923 1924 void 1925 vfs_op_exit(struct mount *mp) 1926 { 1927 1928 MNT_ILOCK(mp); 1929 vfs_op_exit_locked(mp); 1930 MNT_IUNLOCK(mp); 1931 } 1932 1933 struct vfs_op_barrier_ipi { 1934 struct mount *mp; 1935 struct smp_rendezvous_cpus_retry_arg srcra; 1936 }; 1937 1938 static void 1939 vfs_op_action_func(void *arg) 1940 { 1941 struct vfs_op_barrier_ipi *vfsopipi; 1942 struct mount *mp; 1943 1944 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1945 mp = vfsopipi->mp; 1946 1947 if (!vfs_op_thread_entered(mp)) 1948 smp_rendezvous_cpus_done(arg); 1949 } 1950 1951 static void 1952 vfs_op_wait_func(void *arg, int cpu) 1953 { 1954 struct vfs_op_barrier_ipi *vfsopipi; 1955 struct mount *mp; 1956 struct mount_pcpu *mpcpu; 1957 1958 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1959 mp = vfsopipi->mp; 1960 1961 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1962 while (atomic_load_int(&mpcpu->mntp_thread_in_ops)) 1963 cpu_spinwait(); 1964 } 1965 1966 void 1967 vfs_op_barrier_wait(struct mount *mp) 1968 { 1969 struct vfs_op_barrier_ipi vfsopipi; 1970 1971 vfsopipi.mp = mp; 1972 1973 smp_rendezvous_cpus_retry(all_cpus, 1974 smp_no_rendezvous_barrier, 1975 vfs_op_action_func, 1976 smp_no_rendezvous_barrier, 1977 vfs_op_wait_func, 1978 &vfsopipi.srcra); 1979 } 1980 1981 #ifdef DIAGNOSTIC 1982 void 1983 vfs_assert_mount_counters(struct mount *mp) 1984 { 1985 struct mount_pcpu *mpcpu; 1986 int cpu; 1987 1988 if (mp->mnt_vfs_ops == 0) 1989 return; 1990 1991 CPU_FOREACH(cpu) { 1992 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1993 if (mpcpu->mntp_ref != 0 || 1994 mpcpu->mntp_lockref != 0 || 1995 mpcpu->mntp_writeopcount != 0) 1996 vfs_dump_mount_counters(mp); 1997 } 1998 } 1999 2000 void 2001 vfs_dump_mount_counters(struct mount *mp) 2002 { 2003 struct mount_pcpu *mpcpu; 2004 int ref, lockref, writeopcount; 2005 int cpu; 2006 2007 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops); 2008 2009 printf(" ref : "); 2010 ref = mp->mnt_ref; 2011 CPU_FOREACH(cpu) { 2012 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 2013 printf("%d ", mpcpu->mntp_ref); 2014 ref += mpcpu->mntp_ref; 2015 } 2016 printf("\n"); 2017 printf(" lockref : "); 2018 lockref = mp->mnt_lockref; 2019 CPU_FOREACH(cpu) { 2020 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 2021 printf("%d ", mpcpu->mntp_lockref); 2022 lockref += mpcpu->mntp_lockref; 2023 } 2024 printf("\n"); 2025 printf("writeopcount: "); 2026 writeopcount = mp->mnt_writeopcount; 2027 CPU_FOREACH(cpu) { 2028 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 2029 printf("%d ", mpcpu->mntp_writeopcount); 2030 writeopcount += mpcpu->mntp_writeopcount; 2031 } 2032 printf("\n"); 2033 2034 printf("counter struct total\n"); 2035 printf("ref %-5d %-5d\n", mp->mnt_ref, ref); 2036 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref); 2037 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount); 2038 2039 panic("invalid counts on struct mount"); 2040 } 2041 #endif 2042 2043 int 2044 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which) 2045 { 2046 struct mount_pcpu *mpcpu; 2047 int cpu, sum; 2048 2049 switch (which) { 2050 case MNT_COUNT_REF: 2051 sum = mp->mnt_ref; 2052 break; 2053 case MNT_COUNT_LOCKREF: 2054 sum = mp->mnt_lockref; 2055 break; 2056 case MNT_COUNT_WRITEOPCOUNT: 2057 sum = mp->mnt_writeopcount; 2058 break; 2059 } 2060 2061 CPU_FOREACH(cpu) { 2062 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 2063 switch (which) { 2064 case MNT_COUNT_REF: 2065 sum += mpcpu->mntp_ref; 2066 break; 2067 case MNT_COUNT_LOCKREF: 2068 sum += mpcpu->mntp_lockref; 2069 break; 2070 case MNT_COUNT_WRITEOPCOUNT: 2071 sum += mpcpu->mntp_writeopcount; 2072 break; 2073 } 2074 } 2075 return (sum); 2076 } 2077 2078 static bool 2079 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue, 2080 int timeout_ticks) 2081 { 2082 bool enqueued; 2083 2084 enqueued = false; 2085 mtx_lock(&deferred_unmount_lock); 2086 if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) { 2087 mp->mnt_taskqueue_flags = flags | MNT_DEFERRED; 2088 STAILQ_INSERT_TAIL(&deferred_unmount_list, mp, 2089 mnt_taskqueue_link); 2090 enqueued = true; 2091 } 2092 mtx_unlock(&deferred_unmount_lock); 2093 2094 if (enqueued) { 2095 taskqueue_enqueue_timeout(taskqueue_deferred_unmount, 2096 &deferred_unmount_task, timeout_ticks); 2097 } 2098 2099 return (enqueued); 2100 } 2101 2102 /* 2103 * Taskqueue handler for processing async/recursive unmounts 2104 */ 2105 static void 2106 vfs_deferred_unmount(void *argi __unused, int pending __unused) 2107 { 2108 STAILQ_HEAD(, mount) local_unmounts; 2109 uint64_t flags; 2110 struct mount *mp, *tmp; 2111 int error; 2112 unsigned int retries; 2113 bool unmounted; 2114 2115 STAILQ_INIT(&local_unmounts); 2116 mtx_lock(&deferred_unmount_lock); 2117 STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list); 2118 mtx_unlock(&deferred_unmount_lock); 2119 2120 STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) { 2121 flags = mp->mnt_taskqueue_flags; 2122 KASSERT((flags & MNT_DEFERRED) != 0, 2123 ("taskqueue unmount without MNT_DEFERRED")); 2124 error = dounmount(mp, flags, curthread); 2125 if (error != 0) { 2126 MNT_ILOCK(mp); 2127 unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0); 2128 MNT_IUNLOCK(mp); 2129 2130 /* 2131 * The deferred unmount thread is the only thread that 2132 * modifies the retry counts, so locking/atomics aren't 2133 * needed here. 2134 */ 2135 retries = (mp->mnt_unmount_retries)++; 2136 deferred_unmount_total_retries++; 2137 if (!unmounted && retries < deferred_unmount_retry_limit) { 2138 deferred_unmount_enqueue(mp, flags, true, 2139 -deferred_unmount_retry_delay_hz); 2140 } else { 2141 if (retries >= deferred_unmount_retry_limit) { 2142 printf("giving up on deferred unmount " 2143 "of %s after %d retries, error %d\n", 2144 mp->mnt_stat.f_mntonname, retries, error); 2145 } 2146 vfs_rel(mp); 2147 } 2148 } 2149 } 2150 } 2151 2152 /* 2153 * Do the actual filesystem unmount. 2154 */ 2155 int 2156 dounmount(struct mount *mp, uint64_t flags, struct thread *td) 2157 { 2158 struct mount_upper_node *upper; 2159 struct vnode *coveredvp, *rootvp; 2160 int error; 2161 uint64_t async_flag; 2162 int mnt_gen_r; 2163 unsigned int retries; 2164 2165 KASSERT((flags & MNT_DEFERRED) == 0 || 2166 (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE), 2167 ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE")); 2168 2169 /* 2170 * If the caller has explicitly requested the unmount to be handled by 2171 * the taskqueue and we're not already in taskqueue context, queue 2172 * up the unmount request and exit. This is done prior to any 2173 * credential checks; MNT_DEFERRED should be used only for kernel- 2174 * initiated unmounts and will therefore be processed with the 2175 * (kernel) credentials of the taskqueue thread. Still, callers 2176 * should be sure this is the behavior they want. 2177 */ 2178 if ((flags & MNT_DEFERRED) != 0 && 2179 taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) { 2180 if (!deferred_unmount_enqueue(mp, flags, false, 0)) 2181 vfs_rel(mp); 2182 return (EINPROGRESS); 2183 } 2184 2185 /* 2186 * Only privileged root, or (if MNT_USER is set) the user that did the 2187 * original mount is permitted to unmount this filesystem. 2188 * This check should be made prior to queueing up any recursive 2189 * unmounts of upper filesystems. Those unmounts will be executed 2190 * with kernel thread credentials and are expected to succeed, so 2191 * we must at least ensure the originating context has sufficient 2192 * privilege to unmount the base filesystem before proceeding with 2193 * the uppers. 2194 */ 2195 error = vfs_suser(mp, td); 2196 if (error != 0) { 2197 KASSERT((flags & MNT_DEFERRED) == 0, 2198 ("taskqueue unmount with insufficient privilege")); 2199 vfs_rel(mp); 2200 return (error); 2201 } 2202 2203 if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0)) 2204 flags |= MNT_RECURSE; 2205 2206 if ((flags & MNT_RECURSE) != 0) { 2207 KASSERT((flags & MNT_FORCE) != 0, 2208 ("MNT_RECURSE requires MNT_FORCE")); 2209 2210 MNT_ILOCK(mp); 2211 /* 2212 * Set MNTK_RECURSE to prevent new upper mounts from being 2213 * added, and note that an operation on the uppers list is in 2214 * progress. This will ensure that unregistration from the 2215 * uppers list, and therefore any pending unmount of the upper 2216 * FS, can't complete until after we finish walking the list. 2217 */ 2218 mp->mnt_kern_flag |= MNTK_RECURSE; 2219 mp->mnt_upper_pending++; 2220 TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) { 2221 retries = upper->mp->mnt_unmount_retries; 2222 if (retries > deferred_unmount_retry_limit) { 2223 error = EBUSY; 2224 continue; 2225 } 2226 MNT_IUNLOCK(mp); 2227 2228 vfs_ref(upper->mp); 2229 if (!deferred_unmount_enqueue(upper->mp, flags, 2230 false, 0)) 2231 vfs_rel(upper->mp); 2232 MNT_ILOCK(mp); 2233 } 2234 mp->mnt_upper_pending--; 2235 if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 && 2236 mp->mnt_upper_pending == 0) { 2237 mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER; 2238 wakeup(&mp->mnt_uppers); 2239 } 2240 2241 /* 2242 * If we're not on the taskqueue, wait until the uppers list 2243 * is drained before proceeding with unmount. Otherwise, if 2244 * we are on the taskqueue and there are still pending uppers, 2245 * just re-enqueue on the end of the taskqueue. 2246 */ 2247 if ((flags & MNT_DEFERRED) == 0) { 2248 while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) { 2249 mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER; 2250 error = msleep(&mp->mnt_taskqueue_link, 2251 MNT_MTX(mp), PCATCH, "umntqw", 0); 2252 } 2253 if (error != 0) { 2254 MNT_REL(mp); 2255 MNT_IUNLOCK(mp); 2256 return (error); 2257 } 2258 } else if (!TAILQ_EMPTY(&mp->mnt_uppers)) { 2259 MNT_IUNLOCK(mp); 2260 if (error == 0) 2261 deferred_unmount_enqueue(mp, flags, true, 0); 2262 return (error); 2263 } 2264 MNT_IUNLOCK(mp); 2265 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty")); 2266 } 2267 2268 /* Allow the taskqueue to safely re-enqueue on failure */ 2269 if ((flags & MNT_DEFERRED) != 0) 2270 vfs_ref(mp); 2271 2272 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 2273 mnt_gen_r = mp->mnt_gen; 2274 VI_LOCK(coveredvp); 2275 vholdl(coveredvp); 2276 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 2277 /* 2278 * Check for mp being unmounted while waiting for the 2279 * covered vnode lock. 2280 */ 2281 if (coveredvp->v_mountedhere != mp || 2282 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 2283 VOP_UNLOCK(coveredvp); 2284 vdrop(coveredvp); 2285 vfs_rel(mp); 2286 return (EBUSY); 2287 } 2288 } 2289 2290 vfs_op_enter(mp); 2291 2292 vn_start_write(NULL, &mp, V_WAIT); 2293 MNT_ILOCK(mp); 2294 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 || 2295 (mp->mnt_flag & MNT_UPDATE) != 0 || 2296 !TAILQ_EMPTY(&mp->mnt_uppers)) { 2297 dounmount_cleanup(mp, coveredvp, 0); 2298 return (EBUSY); 2299 } 2300 mp->mnt_kern_flag |= MNTK_UNMOUNT; 2301 rootvp = vfs_cache_root_clear(mp); 2302 if (coveredvp != NULL) 2303 vn_seqc_write_begin(coveredvp); 2304 if (flags & MNT_NONBUSY) { 2305 MNT_IUNLOCK(mp); 2306 error = vfs_check_usecounts(mp); 2307 MNT_ILOCK(mp); 2308 if (error != 0) { 2309 vn_seqc_write_end(coveredvp); 2310 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT); 2311 if (rootvp != NULL) { 2312 vn_seqc_write_end(rootvp); 2313 vrele(rootvp); 2314 } 2315 return (error); 2316 } 2317 } 2318 /* Allow filesystems to detect that a forced unmount is in progress. */ 2319 if (flags & MNT_FORCE) { 2320 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 2321 MNT_IUNLOCK(mp); 2322 /* 2323 * Must be done after setting MNTK_UNMOUNTF and before 2324 * waiting for mnt_lockref to become 0. 2325 */ 2326 VFS_PURGE(mp); 2327 MNT_ILOCK(mp); 2328 } 2329 error = 0; 2330 if (mp->mnt_lockref) { 2331 mp->mnt_kern_flag |= MNTK_DRAINING; 2332 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 2333 "mount drain", 0); 2334 } 2335 MNT_IUNLOCK(mp); 2336 KASSERT(mp->mnt_lockref == 0, 2337 ("%s: invalid lock refcount in the drain path @ %s:%d", 2338 __func__, __FILE__, __LINE__)); 2339 KASSERT(error == 0, 2340 ("%s: invalid return value for msleep in the drain path @ %s:%d", 2341 __func__, __FILE__, __LINE__)); 2342 2343 /* 2344 * We want to keep the vnode around so that we can vn_seqc_write_end 2345 * after we are done with unmount. Downgrade our reference to a mere 2346 * hold count so that we don't interefere with anything. 2347 */ 2348 if (rootvp != NULL) { 2349 vhold(rootvp); 2350 vrele(rootvp); 2351 } 2352 2353 if (mp->mnt_flag & MNT_EXPUBLIC) 2354 vfs_setpublicfs(NULL, NULL, NULL); 2355 2356 vfs_periodic(mp, MNT_WAIT); 2357 MNT_ILOCK(mp); 2358 async_flag = mp->mnt_flag & MNT_ASYNC; 2359 mp->mnt_flag &= ~MNT_ASYNC; 2360 mp->mnt_kern_flag &= ~MNTK_ASYNC; 2361 MNT_IUNLOCK(mp); 2362 vfs_deallocate_syncvnode(mp); 2363 error = VFS_UNMOUNT(mp, flags); 2364 vn_finished_write(mp); 2365 vfs_rel(mp); 2366 /* 2367 * If we failed to flush the dirty blocks for this mount point, 2368 * undo all the cdir/rdir and rootvnode changes we made above. 2369 * Unless we failed to do so because the device is reporting that 2370 * it doesn't exist anymore. 2371 */ 2372 if (error && error != ENXIO) { 2373 MNT_ILOCK(mp); 2374 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 2375 MNT_IUNLOCK(mp); 2376 vfs_allocate_syncvnode(mp); 2377 MNT_ILOCK(mp); 2378 } 2379 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 2380 mp->mnt_flag |= async_flag; 2381 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 2382 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 2383 mp->mnt_kern_flag |= MNTK_ASYNC; 2384 if (mp->mnt_kern_flag & MNTK_MWAIT) { 2385 mp->mnt_kern_flag &= ~MNTK_MWAIT; 2386 wakeup(mp); 2387 } 2388 vfs_op_exit_locked(mp); 2389 MNT_IUNLOCK(mp); 2390 if (coveredvp) { 2391 vn_seqc_write_end(coveredvp); 2392 VOP_UNLOCK(coveredvp); 2393 vdrop(coveredvp); 2394 } 2395 if (rootvp != NULL) { 2396 vn_seqc_write_end(rootvp); 2397 vdrop(rootvp); 2398 } 2399 return (error); 2400 } 2401 2402 mtx_lock(&mountlist_mtx); 2403 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2404 mtx_unlock(&mountlist_mtx); 2405 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td); 2406 if (coveredvp != NULL) { 2407 VI_LOCK(coveredvp); 2408 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT); 2409 coveredvp->v_mountedhere = NULL; 2410 vn_seqc_write_end_locked(coveredvp); 2411 VI_UNLOCK(coveredvp); 2412 VOP_UNLOCK(coveredvp); 2413 vdrop(coveredvp); 2414 } 2415 mount_devctl_event("UNMOUNT", mp, false); 2416 if (rootvp != NULL) { 2417 vn_seqc_write_end(rootvp); 2418 vdrop(rootvp); 2419 } 2420 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 2421 if (rootvnode != NULL && mp == rootvnode->v_mount) { 2422 vrele(rootvnode); 2423 rootvnode = NULL; 2424 } 2425 if (mp == rootdevmp) 2426 rootdevmp = NULL; 2427 if ((flags & MNT_DEFERRED) != 0) 2428 vfs_rel(mp); 2429 vfs_mount_destroy(mp); 2430 return (0); 2431 } 2432 2433 /* 2434 * Report errors during filesystem mounting. 2435 */ 2436 void 2437 vfs_mount_error(struct mount *mp, const char *fmt, ...) 2438 { 2439 struct vfsoptlist *moptlist = mp->mnt_optnew; 2440 va_list ap; 2441 int error, len; 2442 char *errmsg; 2443 2444 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 2445 if (error || errmsg == NULL || len <= 0) 2446 return; 2447 2448 va_start(ap, fmt); 2449 vsnprintf(errmsg, (size_t)len, fmt, ap); 2450 va_end(ap); 2451 } 2452 2453 void 2454 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 2455 { 2456 va_list ap; 2457 int error, len; 2458 char *errmsg; 2459 2460 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 2461 if (error || errmsg == NULL || len <= 0) 2462 return; 2463 2464 va_start(ap, fmt); 2465 vsnprintf(errmsg, (size_t)len, fmt, ap); 2466 va_end(ap); 2467 } 2468 2469 /* 2470 * --------------------------------------------------------------------- 2471 * Functions for querying mount options/arguments from filesystems. 2472 */ 2473 2474 /* 2475 * Check that no unknown options are given 2476 */ 2477 int 2478 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 2479 { 2480 struct vfsopt *opt; 2481 char errmsg[255]; 2482 const char **t, *p, *q; 2483 int ret = 0; 2484 2485 TAILQ_FOREACH(opt, opts, link) { 2486 p = opt->name; 2487 q = NULL; 2488 if (p[0] == 'n' && p[1] == 'o') 2489 q = p + 2; 2490 for(t = global_opts; *t != NULL; t++) { 2491 if (strcmp(*t, p) == 0) 2492 break; 2493 if (q != NULL) { 2494 if (strcmp(*t, q) == 0) 2495 break; 2496 } 2497 } 2498 if (*t != NULL) 2499 continue; 2500 for(t = legal; *t != NULL; t++) { 2501 if (strcmp(*t, p) == 0) 2502 break; 2503 if (q != NULL) { 2504 if (strcmp(*t, q) == 0) 2505 break; 2506 } 2507 } 2508 if (*t != NULL) 2509 continue; 2510 snprintf(errmsg, sizeof(errmsg), 2511 "mount option <%s> is unknown", p); 2512 ret = EINVAL; 2513 } 2514 if (ret != 0) { 2515 TAILQ_FOREACH(opt, opts, link) { 2516 if (strcmp(opt->name, "errmsg") == 0) { 2517 strncpy((char *)opt->value, errmsg, opt->len); 2518 break; 2519 } 2520 } 2521 if (opt == NULL) 2522 printf("%s\n", errmsg); 2523 } 2524 return (ret); 2525 } 2526 2527 /* 2528 * Get a mount option by its name. 2529 * 2530 * Return 0 if the option was found, ENOENT otherwise. 2531 * If len is non-NULL it will be filled with the length 2532 * of the option. If buf is non-NULL, it will be filled 2533 * with the address of the option. 2534 */ 2535 int 2536 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len) 2537 { 2538 struct vfsopt *opt; 2539 2540 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2541 2542 TAILQ_FOREACH(opt, opts, link) { 2543 if (strcmp(name, opt->name) == 0) { 2544 opt->seen = 1; 2545 if (len != NULL) 2546 *len = opt->len; 2547 if (buf != NULL) 2548 *buf = opt->value; 2549 return (0); 2550 } 2551 } 2552 return (ENOENT); 2553 } 2554 2555 int 2556 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2557 { 2558 struct vfsopt *opt; 2559 2560 if (opts == NULL) 2561 return (-1); 2562 2563 TAILQ_FOREACH(opt, opts, link) { 2564 if (strcmp(name, opt->name) == 0) { 2565 opt->seen = 1; 2566 return (opt->pos); 2567 } 2568 } 2569 return (-1); 2570 } 2571 2572 int 2573 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value) 2574 { 2575 char *opt_value, *vtp; 2576 quad_t iv; 2577 int error, opt_len; 2578 2579 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len); 2580 if (error != 0) 2581 return (error); 2582 if (opt_len == 0 || opt_value == NULL) 2583 return (EINVAL); 2584 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0') 2585 return (EINVAL); 2586 iv = strtoq(opt_value, &vtp, 0); 2587 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0')) 2588 return (EINVAL); 2589 if (iv < 0) 2590 return (EINVAL); 2591 switch (vtp[0]) { 2592 case 't': case 'T': 2593 iv *= 1024; 2594 /* FALLTHROUGH */ 2595 case 'g': case 'G': 2596 iv *= 1024; 2597 /* FALLTHROUGH */ 2598 case 'm': case 'M': 2599 iv *= 1024; 2600 /* FALLTHROUGH */ 2601 case 'k': case 'K': 2602 iv *= 1024; 2603 case '\0': 2604 break; 2605 default: 2606 return (EINVAL); 2607 } 2608 *value = iv; 2609 2610 return (0); 2611 } 2612 2613 char * 2614 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2615 { 2616 struct vfsopt *opt; 2617 2618 *error = 0; 2619 TAILQ_FOREACH(opt, opts, link) { 2620 if (strcmp(name, opt->name) != 0) 2621 continue; 2622 opt->seen = 1; 2623 if (opt->len == 0 || 2624 ((char *)opt->value)[opt->len - 1] != '\0') { 2625 *error = EINVAL; 2626 return (NULL); 2627 } 2628 return (opt->value); 2629 } 2630 *error = ENOENT; 2631 return (NULL); 2632 } 2633 2634 int 2635 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, 2636 uint64_t val) 2637 { 2638 struct vfsopt *opt; 2639 2640 TAILQ_FOREACH(opt, opts, link) { 2641 if (strcmp(name, opt->name) == 0) { 2642 opt->seen = 1; 2643 if (w != NULL) 2644 *w |= val; 2645 return (1); 2646 } 2647 } 2648 if (w != NULL) 2649 *w &= ~val; 2650 return (0); 2651 } 2652 2653 int 2654 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2655 { 2656 va_list ap; 2657 struct vfsopt *opt; 2658 int ret; 2659 2660 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2661 2662 TAILQ_FOREACH(opt, opts, link) { 2663 if (strcmp(name, opt->name) != 0) 2664 continue; 2665 opt->seen = 1; 2666 if (opt->len == 0 || opt->value == NULL) 2667 return (0); 2668 if (((char *)opt->value)[opt->len - 1] != '\0') 2669 return (0); 2670 va_start(ap, fmt); 2671 ret = vsscanf(opt->value, fmt, ap); 2672 va_end(ap); 2673 return (ret); 2674 } 2675 return (0); 2676 } 2677 2678 int 2679 vfs_setopt(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 bcopy(value, opt->value, len); 2693 } 2694 return (0); 2695 } 2696 return (ENOENT); 2697 } 2698 2699 int 2700 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2701 { 2702 struct vfsopt *opt; 2703 2704 TAILQ_FOREACH(opt, opts, link) { 2705 if (strcmp(name, opt->name) != 0) 2706 continue; 2707 opt->seen = 1; 2708 if (opt->value == NULL) 2709 opt->len = len; 2710 else { 2711 if (opt->len < len) 2712 return (EINVAL); 2713 opt->len = len; 2714 bcopy(value, opt->value, len); 2715 } 2716 return (0); 2717 } 2718 return (ENOENT); 2719 } 2720 2721 int 2722 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2723 { 2724 struct vfsopt *opt; 2725 2726 TAILQ_FOREACH(opt, opts, link) { 2727 if (strcmp(name, opt->name) != 0) 2728 continue; 2729 opt->seen = 1; 2730 if (opt->value == NULL) 2731 opt->len = strlen(value) + 1; 2732 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2733 return (EINVAL); 2734 return (0); 2735 } 2736 return (ENOENT); 2737 } 2738 2739 /* 2740 * Find and copy a mount option. 2741 * 2742 * The size of the buffer has to be specified 2743 * in len, if it is not the same length as the 2744 * mount option, EINVAL is returned. 2745 * Returns ENOENT if the option is not found. 2746 */ 2747 int 2748 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len) 2749 { 2750 struct vfsopt *opt; 2751 2752 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2753 2754 TAILQ_FOREACH(opt, opts, link) { 2755 if (strcmp(name, opt->name) == 0) { 2756 opt->seen = 1; 2757 if (len != opt->len) 2758 return (EINVAL); 2759 bcopy(opt->value, dest, opt->len); 2760 return (0); 2761 } 2762 } 2763 return (ENOENT); 2764 } 2765 2766 int 2767 __vfs_statfs(struct mount *mp, struct statfs *sbp) 2768 { 2769 /* 2770 * Filesystems only fill in part of the structure for updates, we 2771 * have to read the entirety first to get all content. 2772 */ 2773 if (sbp != &mp->mnt_stat) 2774 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp)); 2775 2776 /* 2777 * Set these in case the underlying filesystem fails to do so. 2778 */ 2779 sbp->f_version = STATFS_VERSION; 2780 sbp->f_namemax = NAME_MAX; 2781 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 2782 sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize; 2783 2784 return (mp->mnt_op->vfs_statfs(mp, sbp)); 2785 } 2786 2787 void 2788 vfs_mountedfrom(struct mount *mp, const char *from) 2789 { 2790 2791 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2792 strlcpy(mp->mnt_stat.f_mntfromname, from, 2793 sizeof mp->mnt_stat.f_mntfromname); 2794 } 2795 2796 /* 2797 * --------------------------------------------------------------------- 2798 * This is the api for building mount args and mounting filesystems from 2799 * inside the kernel. 2800 * 2801 * The API works by accumulation of individual args. First error is 2802 * latched. 2803 * 2804 * XXX: should be documented in new manpage kernel_mount(9) 2805 */ 2806 2807 /* A memory allocation which must be freed when we are done */ 2808 struct mntaarg { 2809 SLIST_ENTRY(mntaarg) next; 2810 }; 2811 2812 /* The header for the mount arguments */ 2813 struct mntarg { 2814 struct iovec *v; 2815 int len; 2816 int error; 2817 SLIST_HEAD(, mntaarg) list; 2818 }; 2819 2820 /* 2821 * Add a boolean argument. 2822 * 2823 * flag is the boolean value. 2824 * name must start with "no". 2825 */ 2826 struct mntarg * 2827 mount_argb(struct mntarg *ma, int flag, const char *name) 2828 { 2829 2830 KASSERT(name[0] == 'n' && name[1] == 'o', 2831 ("mount_argb(...,%s): name must start with 'no'", name)); 2832 2833 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2834 } 2835 2836 /* 2837 * Add an argument printf style 2838 */ 2839 struct mntarg * 2840 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2841 { 2842 va_list ap; 2843 struct mntaarg *maa; 2844 struct sbuf *sb; 2845 int len; 2846 2847 if (ma == NULL) { 2848 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2849 SLIST_INIT(&ma->list); 2850 } 2851 if (ma->error) 2852 return (ma); 2853 2854 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2855 M_MOUNT, M_WAITOK); 2856 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2857 ma->v[ma->len].iov_len = strlen(name) + 1; 2858 ma->len++; 2859 2860 sb = sbuf_new_auto(); 2861 va_start(ap, fmt); 2862 sbuf_vprintf(sb, fmt, ap); 2863 va_end(ap); 2864 sbuf_finish(sb); 2865 len = sbuf_len(sb) + 1; 2866 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2867 SLIST_INSERT_HEAD(&ma->list, maa, next); 2868 bcopy(sbuf_data(sb), maa + 1, len); 2869 sbuf_delete(sb); 2870 2871 ma->v[ma->len].iov_base = maa + 1; 2872 ma->v[ma->len].iov_len = len; 2873 ma->len++; 2874 2875 return (ma); 2876 } 2877 2878 /* 2879 * Add an argument which is a userland string. 2880 */ 2881 struct mntarg * 2882 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2883 { 2884 struct mntaarg *maa; 2885 char *tbuf; 2886 2887 if (val == NULL) 2888 return (ma); 2889 if (ma == NULL) { 2890 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2891 SLIST_INIT(&ma->list); 2892 } 2893 if (ma->error) 2894 return (ma); 2895 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2896 SLIST_INSERT_HEAD(&ma->list, maa, next); 2897 tbuf = (void *)(maa + 1); 2898 ma->error = copyinstr(val, tbuf, len, NULL); 2899 return (mount_arg(ma, name, tbuf, -1)); 2900 } 2901 2902 /* 2903 * Plain argument. 2904 * 2905 * If length is -1, treat value as a C string. 2906 */ 2907 struct mntarg * 2908 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2909 { 2910 2911 if (ma == NULL) { 2912 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2913 SLIST_INIT(&ma->list); 2914 } 2915 if (ma->error) 2916 return (ma); 2917 2918 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2919 M_MOUNT, M_WAITOK); 2920 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2921 ma->v[ma->len].iov_len = strlen(name) + 1; 2922 ma->len++; 2923 2924 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2925 if (len < 0) 2926 ma->v[ma->len].iov_len = strlen(val) + 1; 2927 else 2928 ma->v[ma->len].iov_len = len; 2929 ma->len++; 2930 return (ma); 2931 } 2932 2933 /* 2934 * Free a mntarg structure 2935 */ 2936 static void 2937 free_mntarg(struct mntarg *ma) 2938 { 2939 struct mntaarg *maa; 2940 2941 while (!SLIST_EMPTY(&ma->list)) { 2942 maa = SLIST_FIRST(&ma->list); 2943 SLIST_REMOVE_HEAD(&ma->list, next); 2944 free(maa, M_MOUNT); 2945 } 2946 free(ma->v, M_MOUNT); 2947 free(ma, M_MOUNT); 2948 } 2949 2950 /* 2951 * Mount a filesystem 2952 */ 2953 int 2954 kernel_mount(struct mntarg *ma, uint64_t flags) 2955 { 2956 struct uio auio; 2957 int error; 2958 2959 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2960 KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v")); 2961 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2962 2963 error = ma->error; 2964 if (error == 0) { 2965 auio.uio_iov = ma->v; 2966 auio.uio_iovcnt = ma->len; 2967 auio.uio_segflg = UIO_SYSSPACE; 2968 error = vfs_donmount(curthread, flags, &auio); 2969 } 2970 free_mntarg(ma); 2971 return (error); 2972 } 2973 2974 /* Map from mount options to printable formats. */ 2975 static struct mntoptnames optnames[] = { 2976 MNTOPT_NAMES 2977 }; 2978 2979 #define DEVCTL_LEN 1024 2980 static void 2981 mount_devctl_event(const char *type, struct mount *mp, bool donew) 2982 { 2983 const uint8_t *cp; 2984 struct mntoptnames *fp; 2985 struct sbuf sb; 2986 struct statfs *sfp = &mp->mnt_stat; 2987 char *buf; 2988 2989 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT); 2990 if (buf == NULL) 2991 return; 2992 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN); 2993 sbuf_cpy(&sb, "mount-point=\""); 2994 devctl_safe_quote_sb(&sb, sfp->f_mntonname); 2995 sbuf_cat(&sb, "\" mount-dev=\""); 2996 devctl_safe_quote_sb(&sb, sfp->f_mntfromname); 2997 sbuf_cat(&sb, "\" mount-type=\""); 2998 devctl_safe_quote_sb(&sb, sfp->f_fstypename); 2999 sbuf_cat(&sb, "\" fsid=0x"); 3000 cp = (const uint8_t *)&sfp->f_fsid.val[0]; 3001 for (int i = 0; i < sizeof(sfp->f_fsid); i++) 3002 sbuf_printf(&sb, "%02x", cp[i]); 3003 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner); 3004 for (fp = optnames; fp->o_opt != 0; fp++) { 3005 if ((mp->mnt_flag & fp->o_opt) != 0) { 3006 sbuf_cat(&sb, fp->o_name); 3007 sbuf_putc(&sb, ';'); 3008 } 3009 } 3010 sbuf_putc(&sb, '"'); 3011 sbuf_finish(&sb); 3012 3013 /* 3014 * Options are not published because the form of the options depends on 3015 * the file system and may include binary data. In addition, they don't 3016 * necessarily provide enough useful information to be actionable when 3017 * devd processes them. 3018 */ 3019 3020 if (sbuf_error(&sb) == 0) 3021 devctl_notify("VFS", "FS", type, sbuf_data(&sb)); 3022 sbuf_delete(&sb); 3023 free(buf, M_MOUNT); 3024 } 3025 3026 /* 3027 * Force remount specified mount point to read-only. The argument 3028 * must be busied to avoid parallel unmount attempts. 3029 * 3030 * Intended use is to prevent further writes if some metadata 3031 * inconsistency is detected. Note that the function still flushes 3032 * all cached metadata and data for the mount point, which might be 3033 * not always suitable. 3034 */ 3035 int 3036 vfs_remount_ro(struct mount *mp) 3037 { 3038 struct vfsoptlist *opts; 3039 struct vfsopt *opt; 3040 struct vnode *vp_covered, *rootvp; 3041 int error; 3042 3043 vfs_op_enter(mp); 3044 KASSERT(mp->mnt_lockref > 0, 3045 ("vfs_remount_ro: mp %p is not busied", mp)); 3046 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0, 3047 ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp)); 3048 3049 rootvp = NULL; 3050 vp_covered = mp->mnt_vnodecovered; 3051 error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT); 3052 if (error != 0) { 3053 vfs_op_exit(mp); 3054 return (error); 3055 } 3056 VI_LOCK(vp_covered); 3057 if ((vp_covered->v_iflag & VI_MOUNT) != 0) { 3058 VI_UNLOCK(vp_covered); 3059 vput(vp_covered); 3060 vfs_op_exit(mp); 3061 return (EBUSY); 3062 } 3063 vp_covered->v_iflag |= VI_MOUNT; 3064 VI_UNLOCK(vp_covered); 3065 vn_seqc_write_begin(vp_covered); 3066 3067 MNT_ILOCK(mp); 3068 if ((mp->mnt_flag & MNT_RDONLY) != 0) { 3069 MNT_IUNLOCK(mp); 3070 error = EBUSY; 3071 goto out; 3072 } 3073 mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY; 3074 rootvp = vfs_cache_root_clear(mp); 3075 MNT_IUNLOCK(mp); 3076 3077 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO); 3078 TAILQ_INIT(opts); 3079 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO); 3080 opt->name = strdup("ro", M_MOUNT); 3081 opt->value = NULL; 3082 TAILQ_INSERT_TAIL(opts, opt, link); 3083 vfs_mergeopts(opts, mp->mnt_opt); 3084 mp->mnt_optnew = opts; 3085 3086 error = VFS_MOUNT(mp); 3087 3088 if (error == 0) { 3089 MNT_ILOCK(mp); 3090 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE); 3091 MNT_IUNLOCK(mp); 3092 vfs_deallocate_syncvnode(mp); 3093 if (mp->mnt_opt != NULL) 3094 vfs_freeopts(mp->mnt_opt); 3095 mp->mnt_opt = mp->mnt_optnew; 3096 } else { 3097 MNT_ILOCK(mp); 3098 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY); 3099 MNT_IUNLOCK(mp); 3100 vfs_freeopts(mp->mnt_optnew); 3101 } 3102 mp->mnt_optnew = NULL; 3103 3104 out: 3105 vfs_op_exit(mp); 3106 VI_LOCK(vp_covered); 3107 vp_covered->v_iflag &= ~VI_MOUNT; 3108 VI_UNLOCK(vp_covered); 3109 vput(vp_covered); 3110 vn_seqc_write_end(vp_covered); 3111 if (rootvp != NULL) { 3112 vn_seqc_write_end(rootvp); 3113 vrele(rootvp); 3114 } 3115 return (error); 3116 } 3117 3118 /* 3119 * Suspend write operations on all local writeable filesystems. Does 3120 * full sync of them in the process. 3121 * 3122 * Iterate over the mount points in reverse order, suspending most 3123 * recently mounted filesystems first. It handles a case where a 3124 * filesystem mounted from a md(4) vnode-backed device should be 3125 * suspended before the filesystem that owns the vnode. 3126 */ 3127 void 3128 suspend_all_fs(void) 3129 { 3130 struct mount *mp; 3131 int error; 3132 3133 mtx_lock(&mountlist_mtx); 3134 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 3135 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT); 3136 if (error != 0) 3137 continue; 3138 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL || 3139 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 3140 mtx_lock(&mountlist_mtx); 3141 vfs_unbusy(mp); 3142 continue; 3143 } 3144 error = vfs_write_suspend(mp, 0); 3145 if (error == 0) { 3146 MNT_ILOCK(mp); 3147 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0); 3148 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL; 3149 MNT_IUNLOCK(mp); 3150 mtx_lock(&mountlist_mtx); 3151 } else { 3152 printf("suspend of %s failed, error %d\n", 3153 mp->mnt_stat.f_mntonname, error); 3154 mtx_lock(&mountlist_mtx); 3155 vfs_unbusy(mp); 3156 } 3157 } 3158 mtx_unlock(&mountlist_mtx); 3159 } 3160 3161 /* 3162 * Clone the mnt_exjail field to a new mount point. 3163 */ 3164 void 3165 vfs_exjail_clone(struct mount *inmp, struct mount *outmp) 3166 { 3167 struct ucred *cr; 3168 struct prison *pr; 3169 3170 MNT_ILOCK(inmp); 3171 cr = inmp->mnt_exjail; 3172 if (cr != NULL) { 3173 crhold(cr); 3174 MNT_IUNLOCK(inmp); 3175 pr = cr->cr_prison; 3176 sx_slock(&allprison_lock); 3177 if (!prison_isalive(pr)) { 3178 sx_sunlock(&allprison_lock); 3179 crfree(cr); 3180 return; 3181 } 3182 MNT_ILOCK(outmp); 3183 if (outmp->mnt_exjail == NULL) { 3184 outmp->mnt_exjail = cr; 3185 atomic_add_int(&pr->pr_exportcnt, 1); 3186 cr = NULL; 3187 } 3188 MNT_IUNLOCK(outmp); 3189 sx_sunlock(&allprison_lock); 3190 if (cr != NULL) 3191 crfree(cr); 3192 } else 3193 MNT_IUNLOCK(inmp); 3194 } 3195 3196 void 3197 resume_all_fs(void) 3198 { 3199 struct mount *mp; 3200 3201 mtx_lock(&mountlist_mtx); 3202 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3203 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0) 3204 continue; 3205 mtx_unlock(&mountlist_mtx); 3206 MNT_ILOCK(mp); 3207 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0); 3208 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL; 3209 MNT_IUNLOCK(mp); 3210 vfs_write_resume(mp, 0); 3211 mtx_lock(&mountlist_mtx); 3212 vfs_unbusy(mp); 3213 } 3214 mtx_unlock(&mountlist_mtx); 3215 } 3216