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