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