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 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 126 127 EVENTHANDLER_LIST_DEFINE(vfs_mounted); 128 EVENTHANDLER_LIST_DEFINE(vfs_unmounted); 129 130 static void vfs_deferred_unmount(void *arg, int pending); 131 static struct timeout_task deferred_unmount_task; 132 static struct mtx deferred_unmount_lock; 133 MTX_SYSINIT(deferred_unmount, &deferred_unmount_lock, "deferred_unmount", 134 MTX_DEF); 135 static STAILQ_HEAD(, mount) deferred_unmount_list = 136 STAILQ_HEAD_INITIALIZER(deferred_unmount_list); 137 TASKQUEUE_DEFINE_THREAD(deferred_unmount); 138 139 static void mount_devctl_event(const char *type, struct mount *mp, bool donew); 140 141 /* 142 * Global opts, taken by all filesystems 143 */ 144 static const char *global_opts[] = { 145 "errmsg", 146 "fstype", 147 "fspath", 148 "ro", 149 "rw", 150 "nosuid", 151 "noexec", 152 NULL 153 }; 154 155 static int 156 mount_init(void *mem, int size, int flags) 157 { 158 struct mount *mp; 159 160 mp = (struct mount *)mem; 161 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 162 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF); 163 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 164 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO); 165 mp->mnt_ref = 0; 166 mp->mnt_vfs_ops = 1; 167 mp->mnt_rootvnode = NULL; 168 return (0); 169 } 170 171 static void 172 mount_fini(void *mem, int size) 173 { 174 struct mount *mp; 175 176 mp = (struct mount *)mem; 177 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu); 178 lockdestroy(&mp->mnt_explock); 179 mtx_destroy(&mp->mnt_listmtx); 180 mtx_destroy(&mp->mnt_mtx); 181 } 182 183 static void 184 vfs_mount_init(void *dummy __unused) 185 { 186 TIMEOUT_TASK_INIT(taskqueue_deferred_unmount, &deferred_unmount_task, 187 0, vfs_deferred_unmount, NULL); 188 deferred_unmount_retry_delay_hz = hz; 189 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL, 190 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE); 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); 969 970 /* 971 * See if we can mount in the read-only mode if the error code suggests 972 * that it could be possible and the mount options allow for that. 973 * Never try it if "[no]{ro|rw}" has been explicitly requested and not 974 * overridden by "autoro". 975 */ 976 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) { 977 printf("%s: R/W mount failed, possibly R/O media," 978 " trying R/O mount\n", __func__); 979 fsflags |= MNT_RDONLY; 980 error = vfs_domount(td, fstype, fspath, fsflags, &optlist); 981 } 982 bail: 983 /* copyout the errmsg */ 984 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 985 && errmsg_len > 0 && errmsg != NULL) { 986 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 987 bcopy(errmsg, 988 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 989 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 990 } else { 991 copyout(errmsg, 992 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 993 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 994 } 995 } 996 997 if (optlist != NULL) 998 vfs_freeopts(optlist); 999 return (error); 1000 } 1001 1002 /* 1003 * Old mount API. 1004 */ 1005 #ifndef _SYS_SYSPROTO_H_ 1006 struct mount_args { 1007 char *type; 1008 char *path; 1009 int flags; 1010 caddr_t data; 1011 }; 1012 #endif 1013 /* ARGSUSED */ 1014 int 1015 sys_mount(struct thread *td, struct mount_args *uap) 1016 { 1017 char *fstype; 1018 struct vfsconf *vfsp = NULL; 1019 struct mntarg *ma = NULL; 1020 uint64_t flags; 1021 int error; 1022 1023 /* 1024 * Mount flags are now 64-bits. On 32-bit architectures only 1025 * 32-bits are passed in, but from here on everything handles 1026 * 64-bit flags correctly. 1027 */ 1028 flags = uap->flags; 1029 1030 AUDIT_ARG_FFLAGS(flags); 1031 1032 /* 1033 * Filter out MNT_ROOTFS. We do not want clients of mount() in 1034 * userspace to set this flag, but we must filter it out if we want 1035 * MNT_UPDATE on the root file system to work. 1036 * MNT_ROOTFS should only be set by the kernel when mounting its 1037 * root file system. 1038 */ 1039 flags &= ~MNT_ROOTFS; 1040 1041 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 1042 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 1043 if (error) { 1044 free(fstype, M_TEMP); 1045 return (error); 1046 } 1047 1048 AUDIT_ARG_TEXT(fstype); 1049 vfsp = vfs_byname_kld(fstype, td, &error); 1050 free(fstype, M_TEMP); 1051 if (vfsp == NULL) 1052 return (ENOENT); 1053 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 && 1054 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) || 1055 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 && 1056 vfsp->vfc_vfsops->vfs_cmount == NULL)) 1057 return (EOPNOTSUPP); 1058 1059 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN); 1060 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 1061 ma = mount_argb(ma, flags & MNT_RDONLY, "noro"); 1062 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid"); 1063 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec"); 1064 1065 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0) 1066 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags)); 1067 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags)); 1068 } 1069 1070 /* 1071 * vfs_domount_first(): first file system mount (not update) 1072 */ 1073 static int 1074 vfs_domount_first( 1075 struct thread *td, /* Calling thread. */ 1076 struct vfsconf *vfsp, /* File system type. */ 1077 char *fspath, /* Mount path. */ 1078 struct vnode *vp, /* Vnode to be covered. */ 1079 uint64_t fsflags, /* Flags common to all filesystems. */ 1080 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1081 ) 1082 { 1083 struct vattr va; 1084 struct mount *mp; 1085 struct vnode *newdp, *rootvp; 1086 int error, error1; 1087 bool unmounted; 1088 1089 ASSERT_VOP_ELOCKED(vp, __func__); 1090 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 1091 1092 if ((fsflags & MNT_EMPTYDIR) != 0) { 1093 error = vfs_emptydir(vp); 1094 if (error != 0) { 1095 vput(vp); 1096 return (error); 1097 } 1098 } 1099 1100 /* 1101 * If the jail of the calling thread lacks permission for this type of 1102 * file system, or is trying to cover its own root, deny immediately. 1103 */ 1104 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred, 1105 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) { 1106 vput(vp); 1107 return (EPERM); 1108 } 1109 1110 /* 1111 * If the user is not root, ensure that they own the directory 1112 * onto which we are attempting to mount. 1113 */ 1114 error = VOP_GETATTR(vp, &va, td->td_ucred); 1115 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 1116 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN); 1117 if (error == 0) 1118 error = vinvalbuf(vp, V_SAVE, 0, 0); 1119 if (error == 0 && vp->v_type != VDIR) 1120 error = ENOTDIR; 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 & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 1142 1143 /* 1144 * Mount the filesystem. 1145 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1146 * get. No freeing of cn_pnbuf. 1147 */ 1148 error1 = 0; 1149 unmounted = true; 1150 if ((error = VFS_MOUNT(mp)) != 0 || 1151 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 || 1152 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) { 1153 rootvp = NULL; 1154 if (error1 != 0) { 1155 MPASS(error == 0); 1156 rootvp = vfs_cache_root_clear(mp); 1157 if (rootvp != NULL) { 1158 vhold(rootvp); 1159 vrele(rootvp); 1160 } 1161 (void)vn_start_write(NULL, &mp, V_WAIT); 1162 MNT_ILOCK(mp); 1163 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF; 1164 MNT_IUNLOCK(mp); 1165 VFS_PURGE(mp); 1166 error = VFS_UNMOUNT(mp, 0); 1167 vn_finished_write(mp); 1168 if (error != 0) { 1169 printf( 1170 "failed post-mount (%d): rollback unmount returned %d\n", 1171 error1, error); 1172 unmounted = false; 1173 } 1174 error = error1; 1175 } 1176 vfs_unbusy(mp); 1177 mp->mnt_vnodecovered = NULL; 1178 if (unmounted) { 1179 /* XXXKIB wait for mnt_lockref drain? */ 1180 vfs_mount_destroy(mp); 1181 } 1182 VI_LOCK(vp); 1183 vp->v_iflag &= ~VI_MOUNT; 1184 VI_UNLOCK(vp); 1185 if (rootvp != NULL) { 1186 vn_seqc_write_end(rootvp); 1187 vdrop(rootvp); 1188 } 1189 vn_seqc_write_end(vp); 1190 vrele(vp); 1191 return (error); 1192 } 1193 vn_seqc_write_begin(newdp); 1194 VOP_UNLOCK(newdp); 1195 1196 if (mp->mnt_opt != NULL) 1197 vfs_freeopts(mp->mnt_opt); 1198 mp->mnt_opt = mp->mnt_optnew; 1199 *optlist = NULL; 1200 1201 /* 1202 * Prevent external consumers of mount options from reading mnt_optnew. 1203 */ 1204 mp->mnt_optnew = NULL; 1205 1206 MNT_ILOCK(mp); 1207 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1208 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1209 mp->mnt_kern_flag |= MNTK_ASYNC; 1210 else 1211 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1212 MNT_IUNLOCK(mp); 1213 1214 VI_LOCK(vp); 1215 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT); 1216 vp->v_mountedhere = mp; 1217 VI_UNLOCK(vp); 1218 cache_purge(vp); 1219 1220 /* 1221 * We need to lock both vnodes. 1222 * 1223 * Use vn_lock_pair to avoid establishing an ordering between vnodes 1224 * from different filesystems. 1225 */ 1226 vn_lock_pair(vp, false, newdp, false); 1227 1228 VI_LOCK(vp); 1229 vp->v_iflag &= ~VI_MOUNT; 1230 VI_UNLOCK(vp); 1231 /* Place the new filesystem at the end of the mount list. */ 1232 mtx_lock(&mountlist_mtx); 1233 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1234 mtx_unlock(&mountlist_mtx); 1235 vfs_event_signal(NULL, VQ_MOUNT, 0); 1236 VOP_UNLOCK(vp); 1237 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td); 1238 VOP_UNLOCK(newdp); 1239 mount_devctl_event("MOUNT", mp, false); 1240 mountcheckdirs(vp, newdp); 1241 vn_seqc_write_end(vp); 1242 vn_seqc_write_end(newdp); 1243 vrele(newdp); 1244 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1245 vfs_allocate_syncvnode(mp); 1246 vfs_op_exit(mp); 1247 vfs_unbusy(mp); 1248 return (0); 1249 } 1250 1251 /* 1252 * vfs_domount_update(): update of mounted file system 1253 */ 1254 static int 1255 vfs_domount_update( 1256 struct thread *td, /* Calling thread. */ 1257 struct vnode *vp, /* Mount point vnode. */ 1258 uint64_t fsflags, /* Flags common to all filesystems. */ 1259 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1260 ) 1261 { 1262 struct export_args export; 1263 struct o2export_args o2export; 1264 struct vnode *rootvp; 1265 void *bufp; 1266 struct mount *mp; 1267 int error, export_error, i, len; 1268 uint64_t flag; 1269 gid_t *grps; 1270 1271 ASSERT_VOP_ELOCKED(vp, __func__); 1272 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 1273 mp = vp->v_mount; 1274 1275 if ((vp->v_vflag & VV_ROOT) == 0) { 1276 if (vfs_copyopt(*optlist, "export", &export, sizeof(export)) 1277 == 0) 1278 error = EXDEV; 1279 else 1280 error = EINVAL; 1281 vput(vp); 1282 return (error); 1283 } 1284 1285 /* 1286 * We only allow the filesystem to be reloaded if it 1287 * is currently mounted read-only. 1288 */ 1289 flag = mp->mnt_flag; 1290 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 1291 vput(vp); 1292 return (EOPNOTSUPP); /* Needs translation */ 1293 } 1294 /* 1295 * Only privileged root, or (if MNT_USER is set) the user that 1296 * did the original mount is permitted to update it. 1297 */ 1298 error = vfs_suser(mp, td); 1299 if (error != 0) { 1300 vput(vp); 1301 return (error); 1302 } 1303 if (vfs_busy(mp, MBF_NOWAIT)) { 1304 vput(vp); 1305 return (EBUSY); 1306 } 1307 VI_LOCK(vp); 1308 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 1309 VI_UNLOCK(vp); 1310 vfs_unbusy(mp); 1311 vput(vp); 1312 return (EBUSY); 1313 } 1314 vp->v_iflag |= VI_MOUNT; 1315 VI_UNLOCK(vp); 1316 VOP_UNLOCK(vp); 1317 1318 vfs_op_enter(mp); 1319 vn_seqc_write_begin(vp); 1320 1321 rootvp = NULL; 1322 MNT_ILOCK(mp); 1323 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1324 MNT_IUNLOCK(mp); 1325 error = EBUSY; 1326 goto end; 1327 } 1328 mp->mnt_flag &= ~MNT_UPDATEMASK; 1329 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 1330 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 1331 if ((mp->mnt_flag & MNT_ASYNC) == 0) 1332 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1333 rootvp = vfs_cache_root_clear(mp); 1334 MNT_IUNLOCK(mp); 1335 mp->mnt_optnew = *optlist; 1336 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 1337 1338 /* 1339 * Mount the filesystem. 1340 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1341 * get. No freeing of cn_pnbuf. 1342 */ 1343 error = VFS_MOUNT(mp); 1344 1345 export_error = 0; 1346 /* Process the export option. */ 1347 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp, 1348 &len) == 0) { 1349 /* Assume that there is only 1 ABI for each length. */ 1350 switch (len) { 1351 case (sizeof(struct oexport_args)): 1352 bzero(&o2export, sizeof(o2export)); 1353 /* FALLTHROUGH */ 1354 case (sizeof(o2export)): 1355 bcopy(bufp, &o2export, len); 1356 export.ex_flags = (uint64_t)o2export.ex_flags; 1357 export.ex_root = o2export.ex_root; 1358 export.ex_uid = o2export.ex_anon.cr_uid; 1359 export.ex_groups = NULL; 1360 export.ex_ngroups = o2export.ex_anon.cr_ngroups; 1361 if (export.ex_ngroups > 0) { 1362 if (export.ex_ngroups <= XU_NGROUPS) { 1363 export.ex_groups = malloc( 1364 export.ex_ngroups * sizeof(gid_t), 1365 M_TEMP, M_WAITOK); 1366 for (i = 0; i < export.ex_ngroups; i++) 1367 export.ex_groups[i] = 1368 o2export.ex_anon.cr_groups[i]; 1369 } else 1370 export_error = EINVAL; 1371 } else if (export.ex_ngroups < 0) 1372 export_error = EINVAL; 1373 export.ex_addr = o2export.ex_addr; 1374 export.ex_addrlen = o2export.ex_addrlen; 1375 export.ex_mask = o2export.ex_mask; 1376 export.ex_masklen = o2export.ex_masklen; 1377 export.ex_indexfile = o2export.ex_indexfile; 1378 export.ex_numsecflavors = o2export.ex_numsecflavors; 1379 if (export.ex_numsecflavors < MAXSECFLAVORS) { 1380 for (i = 0; i < export.ex_numsecflavors; i++) 1381 export.ex_secflavors[i] = 1382 o2export.ex_secflavors[i]; 1383 } else 1384 export_error = EINVAL; 1385 if (export_error == 0) 1386 export_error = vfs_export(mp, &export); 1387 free(export.ex_groups, M_TEMP); 1388 break; 1389 case (sizeof(export)): 1390 bcopy(bufp, &export, len); 1391 grps = NULL; 1392 if (export.ex_ngroups > 0) { 1393 if (export.ex_ngroups <= NGROUPS_MAX) { 1394 grps = malloc(export.ex_ngroups * 1395 sizeof(gid_t), M_TEMP, M_WAITOK); 1396 export_error = copyin(export.ex_groups, 1397 grps, export.ex_ngroups * 1398 sizeof(gid_t)); 1399 if (export_error == 0) 1400 export.ex_groups = grps; 1401 } else 1402 export_error = EINVAL; 1403 } else if (export.ex_ngroups == 0) 1404 export.ex_groups = NULL; 1405 else 1406 export_error = EINVAL; 1407 if (export_error == 0) 1408 export_error = vfs_export(mp, &export); 1409 free(grps, M_TEMP); 1410 break; 1411 default: 1412 export_error = EINVAL; 1413 break; 1414 } 1415 } 1416 1417 MNT_ILOCK(mp); 1418 if (error == 0) { 1419 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 1420 MNT_SNAPSHOT); 1421 } else { 1422 /* 1423 * If we fail, restore old mount flags. MNT_QUOTA is special, 1424 * because it is not part of MNT_UPDATEMASK, but it could have 1425 * changed in the meantime if quotactl(2) was called. 1426 * All in all we want current value of MNT_QUOTA, not the old 1427 * one. 1428 */ 1429 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 1430 } 1431 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1432 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1433 mp->mnt_kern_flag |= MNTK_ASYNC; 1434 else 1435 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1436 MNT_IUNLOCK(mp); 1437 1438 if (error != 0) 1439 goto end; 1440 1441 mount_devctl_event("REMOUNT", mp, true); 1442 if (mp->mnt_opt != NULL) 1443 vfs_freeopts(mp->mnt_opt); 1444 mp->mnt_opt = mp->mnt_optnew; 1445 *optlist = NULL; 1446 (void)VFS_STATFS(mp, &mp->mnt_stat); 1447 /* 1448 * Prevent external consumers of mount options from reading 1449 * mnt_optnew. 1450 */ 1451 mp->mnt_optnew = NULL; 1452 1453 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1454 vfs_allocate_syncvnode(mp); 1455 else 1456 vfs_deallocate_syncvnode(mp); 1457 end: 1458 vfs_op_exit(mp); 1459 if (rootvp != NULL) { 1460 vn_seqc_write_end(rootvp); 1461 vrele(rootvp); 1462 } 1463 vn_seqc_write_end(vp); 1464 vfs_unbusy(mp); 1465 VI_LOCK(vp); 1466 vp->v_iflag &= ~VI_MOUNT; 1467 VI_UNLOCK(vp); 1468 vrele(vp); 1469 return (error != 0 ? error : export_error); 1470 } 1471 1472 /* 1473 * vfs_domount(): actually attempt a filesystem mount. 1474 */ 1475 static int 1476 vfs_domount( 1477 struct thread *td, /* Calling thread. */ 1478 const char *fstype, /* Filesystem type. */ 1479 char *fspath, /* Mount path. */ 1480 uint64_t fsflags, /* Flags common to all filesystems. */ 1481 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1482 ) 1483 { 1484 struct vfsconf *vfsp; 1485 struct nameidata nd; 1486 struct vnode *vp; 1487 char *pathbuf; 1488 int error; 1489 1490 /* 1491 * Be ultra-paranoid about making sure the type and fspath 1492 * variables will fit in our mp buffers, including the 1493 * terminating NUL. 1494 */ 1495 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1496 return (ENAMETOOLONG); 1497 1498 if (jailed(td->td_ucred) || usermount == 0) { 1499 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1500 return (error); 1501 } 1502 1503 /* 1504 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1505 */ 1506 if (fsflags & MNT_EXPORTED) { 1507 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1508 if (error) 1509 return (error); 1510 } 1511 if (fsflags & MNT_SUIDDIR) { 1512 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1513 if (error) 1514 return (error); 1515 } 1516 /* 1517 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1518 */ 1519 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1520 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1521 fsflags |= MNT_NOSUID | MNT_USER; 1522 } 1523 1524 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1525 vfsp = NULL; 1526 if ((fsflags & MNT_UPDATE) == 0) { 1527 /* Don't try to load KLDs if we're mounting the root. */ 1528 if (fsflags & MNT_ROOTFS) 1529 vfsp = vfs_byname(fstype); 1530 else 1531 vfsp = vfs_byname_kld(fstype, td, &error); 1532 if (vfsp == NULL) 1533 return (ENODEV); 1534 } 1535 1536 /* 1537 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1538 */ 1539 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1540 UIO_SYSSPACE, fspath, td); 1541 error = namei(&nd); 1542 if (error != 0) 1543 return (error); 1544 NDFREE(&nd, NDF_ONLY_PNBUF); 1545 vp = nd.ni_vp; 1546 if ((fsflags & MNT_UPDATE) == 0) { 1547 if ((vp->v_vflag & VV_ROOT) != 0 && 1548 (fsflags & MNT_NOCOVER) != 0) { 1549 vput(vp); 1550 return (EBUSY); 1551 } 1552 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1553 strcpy(pathbuf, fspath); 1554 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN); 1555 if (error == 0) { 1556 error = vfs_domount_first(td, vfsp, pathbuf, vp, 1557 fsflags, optlist); 1558 } 1559 free(pathbuf, M_TEMP); 1560 } else 1561 error = vfs_domount_update(td, vp, fsflags, optlist); 1562 1563 return (error); 1564 } 1565 1566 /* 1567 * Unmount a filesystem. 1568 * 1569 * Note: unmount takes a path to the vnode mounted on as argument, not 1570 * special file (as before). 1571 */ 1572 #ifndef _SYS_SYSPROTO_H_ 1573 struct unmount_args { 1574 char *path; 1575 int flags; 1576 }; 1577 #endif 1578 /* ARGSUSED */ 1579 int 1580 sys_unmount(struct thread *td, struct unmount_args *uap) 1581 { 1582 1583 return (kern_unmount(td, uap->path, uap->flags)); 1584 } 1585 1586 int 1587 kern_unmount(struct thread *td, const char *path, int flags) 1588 { 1589 struct nameidata nd; 1590 struct mount *mp; 1591 char *pathbuf; 1592 int error, id0, id1; 1593 1594 AUDIT_ARG_VALUE(flags); 1595 if (jailed(td->td_ucred) || usermount == 0) { 1596 error = priv_check(td, PRIV_VFS_UNMOUNT); 1597 if (error) 1598 return (error); 1599 } 1600 1601 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1602 error = copyinstr(path, pathbuf, MNAMELEN, NULL); 1603 if (error) { 1604 free(pathbuf, M_TEMP); 1605 return (error); 1606 } 1607 if (flags & MNT_BYFSID) { 1608 AUDIT_ARG_TEXT(pathbuf); 1609 /* Decode the filesystem ID. */ 1610 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1611 free(pathbuf, M_TEMP); 1612 return (EINVAL); 1613 } 1614 1615 mtx_lock(&mountlist_mtx); 1616 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1617 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1618 mp->mnt_stat.f_fsid.val[1] == id1) { 1619 vfs_ref(mp); 1620 break; 1621 } 1622 } 1623 mtx_unlock(&mountlist_mtx); 1624 } else { 1625 /* 1626 * Try to find global path for path argument. 1627 */ 1628 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1629 UIO_SYSSPACE, pathbuf, td); 1630 if (namei(&nd) == 0) { 1631 NDFREE(&nd, NDF_ONLY_PNBUF); 1632 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf, 1633 MNAMELEN); 1634 if (error == 0) 1635 vput(nd.ni_vp); 1636 } 1637 mtx_lock(&mountlist_mtx); 1638 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1639 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) { 1640 vfs_ref(mp); 1641 break; 1642 } 1643 } 1644 mtx_unlock(&mountlist_mtx); 1645 } 1646 free(pathbuf, M_TEMP); 1647 if (mp == NULL) { 1648 /* 1649 * Previously we returned ENOENT for a nonexistent path and 1650 * EINVAL for a non-mountpoint. We cannot tell these apart 1651 * now, so in the !MNT_BYFSID case return the more likely 1652 * EINVAL for compatibility. 1653 */ 1654 return ((flags & MNT_BYFSID) ? ENOENT : EINVAL); 1655 } 1656 1657 /* 1658 * Don't allow unmounting the root filesystem. 1659 */ 1660 if (mp->mnt_flag & MNT_ROOTFS) { 1661 vfs_rel(mp); 1662 return (EINVAL); 1663 } 1664 error = dounmount(mp, flags, td); 1665 return (error); 1666 } 1667 1668 /* 1669 * Return error if any of the vnodes, ignoring the root vnode 1670 * and the syncer vnode, have non-zero usecount. 1671 * 1672 * This function is purely advisory - it can return false positives 1673 * and negatives. 1674 */ 1675 static int 1676 vfs_check_usecounts(struct mount *mp) 1677 { 1678 struct vnode *vp, *mvp; 1679 1680 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1681 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON && 1682 vp->v_usecount != 0) { 1683 VI_UNLOCK(vp); 1684 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1685 return (EBUSY); 1686 } 1687 VI_UNLOCK(vp); 1688 } 1689 1690 return (0); 1691 } 1692 1693 static void 1694 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags) 1695 { 1696 1697 mtx_assert(MNT_MTX(mp), MA_OWNED); 1698 mp->mnt_kern_flag &= ~mntkflags; 1699 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) { 1700 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1701 wakeup(mp); 1702 } 1703 vfs_op_exit_locked(mp); 1704 MNT_IUNLOCK(mp); 1705 if (coveredvp != NULL) { 1706 VOP_UNLOCK(coveredvp); 1707 vdrop(coveredvp); 1708 } 1709 vn_finished_write(mp); 1710 } 1711 1712 /* 1713 * There are various reference counters associated with the mount point. 1714 * Normally it is permitted to modify them without taking the mnt ilock, 1715 * but this behavior can be temporarily disabled if stable value is needed 1716 * or callers are expected to block (e.g. to not allow new users during 1717 * forced unmount). 1718 */ 1719 void 1720 vfs_op_enter(struct mount *mp) 1721 { 1722 struct mount_pcpu *mpcpu; 1723 int cpu; 1724 1725 MNT_ILOCK(mp); 1726 mp->mnt_vfs_ops++; 1727 if (mp->mnt_vfs_ops > 1) { 1728 MNT_IUNLOCK(mp); 1729 return; 1730 } 1731 vfs_op_barrier_wait(mp); 1732 CPU_FOREACH(cpu) { 1733 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1734 1735 mp->mnt_ref += mpcpu->mntp_ref; 1736 mpcpu->mntp_ref = 0; 1737 1738 mp->mnt_lockref += mpcpu->mntp_lockref; 1739 mpcpu->mntp_lockref = 0; 1740 1741 mp->mnt_writeopcount += mpcpu->mntp_writeopcount; 1742 mpcpu->mntp_writeopcount = 0; 1743 } 1744 if (mp->mnt_ref <= 0 || mp->mnt_lockref < 0 || mp->mnt_writeopcount < 0) 1745 panic("%s: invalid count(s) on mp %p: ref %d lockref %d writeopcount %d\n", 1746 __func__, mp, mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount); 1747 MNT_IUNLOCK(mp); 1748 vfs_assert_mount_counters(mp); 1749 } 1750 1751 void 1752 vfs_op_exit_locked(struct mount *mp) 1753 { 1754 1755 mtx_assert(MNT_MTX(mp), MA_OWNED); 1756 1757 if (mp->mnt_vfs_ops <= 0) 1758 panic("%s: invalid vfs_ops count %d for mp %p\n", 1759 __func__, mp->mnt_vfs_ops, mp); 1760 mp->mnt_vfs_ops--; 1761 } 1762 1763 void 1764 vfs_op_exit(struct mount *mp) 1765 { 1766 1767 MNT_ILOCK(mp); 1768 vfs_op_exit_locked(mp); 1769 MNT_IUNLOCK(mp); 1770 } 1771 1772 struct vfs_op_barrier_ipi { 1773 struct mount *mp; 1774 struct smp_rendezvous_cpus_retry_arg srcra; 1775 }; 1776 1777 static void 1778 vfs_op_action_func(void *arg) 1779 { 1780 struct vfs_op_barrier_ipi *vfsopipi; 1781 struct mount *mp; 1782 1783 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1784 mp = vfsopipi->mp; 1785 1786 if (!vfs_op_thread_entered(mp)) 1787 smp_rendezvous_cpus_done(arg); 1788 } 1789 1790 static void 1791 vfs_op_wait_func(void *arg, int cpu) 1792 { 1793 struct vfs_op_barrier_ipi *vfsopipi; 1794 struct mount *mp; 1795 struct mount_pcpu *mpcpu; 1796 1797 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1798 mp = vfsopipi->mp; 1799 1800 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1801 while (atomic_load_int(&mpcpu->mntp_thread_in_ops)) 1802 cpu_spinwait(); 1803 } 1804 1805 void 1806 vfs_op_barrier_wait(struct mount *mp) 1807 { 1808 struct vfs_op_barrier_ipi vfsopipi; 1809 1810 vfsopipi.mp = mp; 1811 1812 smp_rendezvous_cpus_retry(all_cpus, 1813 smp_no_rendezvous_barrier, 1814 vfs_op_action_func, 1815 smp_no_rendezvous_barrier, 1816 vfs_op_wait_func, 1817 &vfsopipi.srcra); 1818 } 1819 1820 #ifdef DIAGNOSTIC 1821 void 1822 vfs_assert_mount_counters(struct mount *mp) 1823 { 1824 struct mount_pcpu *mpcpu; 1825 int cpu; 1826 1827 if (mp->mnt_vfs_ops == 0) 1828 return; 1829 1830 CPU_FOREACH(cpu) { 1831 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1832 if (mpcpu->mntp_ref != 0 || 1833 mpcpu->mntp_lockref != 0 || 1834 mpcpu->mntp_writeopcount != 0) 1835 vfs_dump_mount_counters(mp); 1836 } 1837 } 1838 1839 void 1840 vfs_dump_mount_counters(struct mount *mp) 1841 { 1842 struct mount_pcpu *mpcpu; 1843 int ref, lockref, writeopcount; 1844 int cpu; 1845 1846 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops); 1847 1848 printf(" ref : "); 1849 ref = mp->mnt_ref; 1850 CPU_FOREACH(cpu) { 1851 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1852 printf("%d ", mpcpu->mntp_ref); 1853 ref += mpcpu->mntp_ref; 1854 } 1855 printf("\n"); 1856 printf(" lockref : "); 1857 lockref = mp->mnt_lockref; 1858 CPU_FOREACH(cpu) { 1859 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1860 printf("%d ", mpcpu->mntp_lockref); 1861 lockref += mpcpu->mntp_lockref; 1862 } 1863 printf("\n"); 1864 printf("writeopcount: "); 1865 writeopcount = mp->mnt_writeopcount; 1866 CPU_FOREACH(cpu) { 1867 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1868 printf("%d ", mpcpu->mntp_writeopcount); 1869 writeopcount += mpcpu->mntp_writeopcount; 1870 } 1871 printf("\n"); 1872 1873 printf("counter struct total\n"); 1874 printf("ref %-5d %-5d\n", mp->mnt_ref, ref); 1875 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref); 1876 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount); 1877 1878 panic("invalid counts on struct mount"); 1879 } 1880 #endif 1881 1882 int 1883 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which) 1884 { 1885 struct mount_pcpu *mpcpu; 1886 int cpu, sum; 1887 1888 switch (which) { 1889 case MNT_COUNT_REF: 1890 sum = mp->mnt_ref; 1891 break; 1892 case MNT_COUNT_LOCKREF: 1893 sum = mp->mnt_lockref; 1894 break; 1895 case MNT_COUNT_WRITEOPCOUNT: 1896 sum = mp->mnt_writeopcount; 1897 break; 1898 } 1899 1900 CPU_FOREACH(cpu) { 1901 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1902 switch (which) { 1903 case MNT_COUNT_REF: 1904 sum += mpcpu->mntp_ref; 1905 break; 1906 case MNT_COUNT_LOCKREF: 1907 sum += mpcpu->mntp_lockref; 1908 break; 1909 case MNT_COUNT_WRITEOPCOUNT: 1910 sum += mpcpu->mntp_writeopcount; 1911 break; 1912 } 1913 } 1914 return (sum); 1915 } 1916 1917 static bool 1918 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue, 1919 int timeout_ticks) 1920 { 1921 bool enqueued; 1922 1923 enqueued = false; 1924 mtx_lock(&deferred_unmount_lock); 1925 if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) { 1926 mp->mnt_taskqueue_flags = flags | MNT_DEFERRED; 1927 STAILQ_INSERT_TAIL(&deferred_unmount_list, mp, 1928 mnt_taskqueue_link); 1929 enqueued = true; 1930 } 1931 mtx_unlock(&deferred_unmount_lock); 1932 1933 if (enqueued) { 1934 taskqueue_enqueue_timeout(taskqueue_deferred_unmount, 1935 &deferred_unmount_task, timeout_ticks); 1936 } 1937 1938 return (enqueued); 1939 } 1940 1941 /* 1942 * Taskqueue handler for processing async/recursive unmounts 1943 */ 1944 static void 1945 vfs_deferred_unmount(void *argi __unused, int pending __unused) 1946 { 1947 STAILQ_HEAD(, mount) local_unmounts; 1948 uint64_t flags; 1949 struct mount *mp, *tmp; 1950 int error; 1951 unsigned int retries; 1952 bool unmounted; 1953 1954 STAILQ_INIT(&local_unmounts); 1955 mtx_lock(&deferred_unmount_lock); 1956 STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list); 1957 mtx_unlock(&deferred_unmount_lock); 1958 1959 STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) { 1960 flags = mp->mnt_taskqueue_flags; 1961 KASSERT((flags & MNT_DEFERRED) != 0, 1962 ("taskqueue unmount without MNT_DEFERRED")); 1963 error = dounmount(mp, flags, curthread); 1964 if (error != 0) { 1965 MNT_ILOCK(mp); 1966 unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0); 1967 MNT_IUNLOCK(mp); 1968 1969 /* 1970 * The deferred unmount thread is the only thread that 1971 * modifies the retry counts, so locking/atomics aren't 1972 * needed here. 1973 */ 1974 retries = (mp->mnt_unmount_retries)++; 1975 deferred_unmount_total_retries++; 1976 if (!unmounted && retries < deferred_unmount_retry_limit) { 1977 deferred_unmount_enqueue(mp, flags, true, 1978 -deferred_unmount_retry_delay_hz); 1979 } else { 1980 if (retries >= deferred_unmount_retry_limit) { 1981 printf("giving up on deferred unmount " 1982 "of %s after %d retries, error %d\n", 1983 mp->mnt_stat.f_mntonname, retries, error); 1984 } 1985 vfs_rel(mp); 1986 } 1987 } 1988 } 1989 } 1990 1991 /* 1992 * Do the actual filesystem unmount. 1993 */ 1994 int 1995 dounmount(struct mount *mp, uint64_t flags, struct thread *td) 1996 { 1997 struct mount_upper_node *upper; 1998 struct vnode *coveredvp, *rootvp; 1999 int error; 2000 uint64_t async_flag; 2001 int mnt_gen_r; 2002 unsigned int retries; 2003 2004 KASSERT((flags & MNT_DEFERRED) == 0 || 2005 (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE), 2006 ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE")); 2007 2008 /* 2009 * If the caller has explicitly requested the unmount to be handled by 2010 * the taskqueue and we're not already in taskqueue context, queue 2011 * up the unmount request and exit. This is done prior to any 2012 * credential checks; MNT_DEFERRED should be used only for kernel- 2013 * initiated unmounts and will therefore be processed with the 2014 * (kernel) credentials of the taskqueue thread. Still, callers 2015 * should be sure this is the behavior they want. 2016 */ 2017 if ((flags & MNT_DEFERRED) != 0 && 2018 taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) { 2019 if (!deferred_unmount_enqueue(mp, flags, false, 0)) 2020 vfs_rel(mp); 2021 return (EINPROGRESS); 2022 } 2023 2024 /* 2025 * Only privileged root, or (if MNT_USER is set) the user that did the 2026 * original mount is permitted to unmount this filesystem. 2027 * This check should be made prior to queueing up any recursive 2028 * unmounts of upper filesystems. Those unmounts will be executed 2029 * with kernel thread credentials and are expected to succeed, so 2030 * we must at least ensure the originating context has sufficient 2031 * privilege to unmount the base filesystem before proceeding with 2032 * the uppers. 2033 */ 2034 error = vfs_suser(mp, td); 2035 if (error != 0) { 2036 KASSERT((flags & MNT_DEFERRED) == 0, 2037 ("taskqueue unmount with insufficient privilege")); 2038 vfs_rel(mp); 2039 return (error); 2040 } 2041 2042 if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0)) 2043 flags |= MNT_RECURSE; 2044 2045 if ((flags & MNT_RECURSE) != 0) { 2046 KASSERT((flags & MNT_FORCE) != 0, 2047 ("MNT_RECURSE requires MNT_FORCE")); 2048 2049 MNT_ILOCK(mp); 2050 /* 2051 * Set MNTK_RECURSE to prevent new upper mounts from being 2052 * added, and note that an operation on the uppers list is in 2053 * progress. This will ensure that unregistration from the 2054 * uppers list, and therefore any pending unmount of the upper 2055 * FS, can't complete until after we finish walking the list. 2056 */ 2057 mp->mnt_kern_flag |= MNTK_RECURSE; 2058 mp->mnt_upper_pending++; 2059 TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) { 2060 retries = upper->mp->mnt_unmount_retries; 2061 if (retries > deferred_unmount_retry_limit) { 2062 error = EBUSY; 2063 continue; 2064 } 2065 MNT_IUNLOCK(mp); 2066 2067 vfs_ref(upper->mp); 2068 if (!deferred_unmount_enqueue(upper->mp, flags, 2069 false, 0)) 2070 vfs_rel(upper->mp); 2071 MNT_ILOCK(mp); 2072 } 2073 mp->mnt_upper_pending--; 2074 if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 && 2075 mp->mnt_upper_pending == 0) { 2076 mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER; 2077 wakeup(&mp->mnt_uppers); 2078 } 2079 2080 /* 2081 * If we're not on the taskqueue, wait until the uppers list 2082 * is drained before proceeding with unmount. Otherwise, if 2083 * we are on the taskqueue and there are still pending uppers, 2084 * just re-enqueue on the end of the taskqueue. 2085 */ 2086 if ((flags & MNT_DEFERRED) == 0) { 2087 while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) { 2088 mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER; 2089 error = msleep(&mp->mnt_taskqueue_link, 2090 MNT_MTX(mp), PCATCH, "umntqw", 0); 2091 } 2092 if (error != 0) { 2093 MNT_REL(mp); 2094 MNT_IUNLOCK(mp); 2095 return (error); 2096 } 2097 } else if (!TAILQ_EMPTY(&mp->mnt_uppers)) { 2098 MNT_IUNLOCK(mp); 2099 if (error == 0) 2100 deferred_unmount_enqueue(mp, flags, true, 0); 2101 return (error); 2102 } 2103 MNT_IUNLOCK(mp); 2104 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty")); 2105 } 2106 2107 /* Allow the taskqueue to safely re-enqueue on failure */ 2108 if ((flags & MNT_DEFERRED) != 0) 2109 vfs_ref(mp); 2110 2111 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 2112 mnt_gen_r = mp->mnt_gen; 2113 VI_LOCK(coveredvp); 2114 vholdl(coveredvp); 2115 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 2116 /* 2117 * Check for mp being unmounted while waiting for the 2118 * covered vnode lock. 2119 */ 2120 if (coveredvp->v_mountedhere != mp || 2121 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 2122 VOP_UNLOCK(coveredvp); 2123 vdrop(coveredvp); 2124 vfs_rel(mp); 2125 return (EBUSY); 2126 } 2127 } 2128 2129 vfs_op_enter(mp); 2130 2131 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF); 2132 MNT_ILOCK(mp); 2133 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 || 2134 (mp->mnt_flag & MNT_UPDATE) != 0 || 2135 !TAILQ_EMPTY(&mp->mnt_uppers)) { 2136 dounmount_cleanup(mp, coveredvp, 0); 2137 return (EBUSY); 2138 } 2139 mp->mnt_kern_flag |= MNTK_UNMOUNT; 2140 rootvp = vfs_cache_root_clear(mp); 2141 if (coveredvp != NULL) 2142 vn_seqc_write_begin(coveredvp); 2143 if (flags & MNT_NONBUSY) { 2144 MNT_IUNLOCK(mp); 2145 error = vfs_check_usecounts(mp); 2146 MNT_ILOCK(mp); 2147 if (error != 0) { 2148 vn_seqc_write_end(coveredvp); 2149 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT); 2150 if (rootvp != NULL) { 2151 vn_seqc_write_end(rootvp); 2152 vrele(rootvp); 2153 } 2154 return (error); 2155 } 2156 } 2157 /* Allow filesystems to detect that a forced unmount is in progress. */ 2158 if (flags & MNT_FORCE) { 2159 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 2160 MNT_IUNLOCK(mp); 2161 /* 2162 * Must be done after setting MNTK_UNMOUNTF and before 2163 * waiting for mnt_lockref to become 0. 2164 */ 2165 VFS_PURGE(mp); 2166 MNT_ILOCK(mp); 2167 } 2168 error = 0; 2169 if (mp->mnt_lockref) { 2170 mp->mnt_kern_flag |= MNTK_DRAINING; 2171 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 2172 "mount drain", 0); 2173 } 2174 MNT_IUNLOCK(mp); 2175 KASSERT(mp->mnt_lockref == 0, 2176 ("%s: invalid lock refcount in the drain path @ %s:%d", 2177 __func__, __FILE__, __LINE__)); 2178 KASSERT(error == 0, 2179 ("%s: invalid return value for msleep in the drain path @ %s:%d", 2180 __func__, __FILE__, __LINE__)); 2181 2182 /* 2183 * We want to keep the vnode around so that we can vn_seqc_write_end 2184 * after we are done with unmount. Downgrade our reference to a mere 2185 * hold count so that we don't interefere with anything. 2186 */ 2187 if (rootvp != NULL) { 2188 vhold(rootvp); 2189 vrele(rootvp); 2190 } 2191 2192 if (mp->mnt_flag & MNT_EXPUBLIC) 2193 vfs_setpublicfs(NULL, NULL, NULL); 2194 2195 vfs_periodic(mp, MNT_WAIT); 2196 MNT_ILOCK(mp); 2197 async_flag = mp->mnt_flag & MNT_ASYNC; 2198 mp->mnt_flag &= ~MNT_ASYNC; 2199 mp->mnt_kern_flag &= ~MNTK_ASYNC; 2200 MNT_IUNLOCK(mp); 2201 vfs_deallocate_syncvnode(mp); 2202 error = VFS_UNMOUNT(mp, flags); 2203 vn_finished_write(mp); 2204 /* 2205 * If we failed to flush the dirty blocks for this mount point, 2206 * undo all the cdir/rdir and rootvnode changes we made above. 2207 * Unless we failed to do so because the device is reporting that 2208 * it doesn't exist anymore. 2209 */ 2210 if (error && error != ENXIO) { 2211 MNT_ILOCK(mp); 2212 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 2213 MNT_IUNLOCK(mp); 2214 vfs_allocate_syncvnode(mp); 2215 MNT_ILOCK(mp); 2216 } 2217 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 2218 mp->mnt_flag |= async_flag; 2219 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 2220 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 2221 mp->mnt_kern_flag |= MNTK_ASYNC; 2222 if (mp->mnt_kern_flag & MNTK_MWAIT) { 2223 mp->mnt_kern_flag &= ~MNTK_MWAIT; 2224 wakeup(mp); 2225 } 2226 vfs_op_exit_locked(mp); 2227 MNT_IUNLOCK(mp); 2228 if (coveredvp) { 2229 vn_seqc_write_end(coveredvp); 2230 VOP_UNLOCK(coveredvp); 2231 vdrop(coveredvp); 2232 } 2233 if (rootvp != NULL) { 2234 vn_seqc_write_end(rootvp); 2235 vdrop(rootvp); 2236 } 2237 return (error); 2238 } 2239 2240 mtx_lock(&mountlist_mtx); 2241 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2242 mtx_unlock(&mountlist_mtx); 2243 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td); 2244 if (coveredvp != NULL) { 2245 VI_LOCK(coveredvp); 2246 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT); 2247 coveredvp->v_mountedhere = NULL; 2248 vn_seqc_write_end_locked(coveredvp); 2249 VI_UNLOCK(coveredvp); 2250 VOP_UNLOCK(coveredvp); 2251 vdrop(coveredvp); 2252 } 2253 mount_devctl_event("UNMOUNT", mp, false); 2254 if (rootvp != NULL) { 2255 vn_seqc_write_end(rootvp); 2256 vdrop(rootvp); 2257 } 2258 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 2259 if (rootvnode != NULL && mp == rootvnode->v_mount) { 2260 vrele(rootvnode); 2261 rootvnode = NULL; 2262 } 2263 if (mp == rootdevmp) 2264 rootdevmp = NULL; 2265 if ((flags & MNT_DEFERRED) != 0) 2266 vfs_rel(mp); 2267 vfs_mount_destroy(mp); 2268 return (0); 2269 } 2270 2271 /* 2272 * Report errors during filesystem mounting. 2273 */ 2274 void 2275 vfs_mount_error(struct mount *mp, const char *fmt, ...) 2276 { 2277 struct vfsoptlist *moptlist = mp->mnt_optnew; 2278 va_list ap; 2279 int error, len; 2280 char *errmsg; 2281 2282 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 2283 if (error || errmsg == NULL || len <= 0) 2284 return; 2285 2286 va_start(ap, fmt); 2287 vsnprintf(errmsg, (size_t)len, fmt, ap); 2288 va_end(ap); 2289 } 2290 2291 void 2292 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 2293 { 2294 va_list ap; 2295 int error, len; 2296 char *errmsg; 2297 2298 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 2299 if (error || errmsg == NULL || len <= 0) 2300 return; 2301 2302 va_start(ap, fmt); 2303 vsnprintf(errmsg, (size_t)len, fmt, ap); 2304 va_end(ap); 2305 } 2306 2307 /* 2308 * --------------------------------------------------------------------- 2309 * Functions for querying mount options/arguments from filesystems. 2310 */ 2311 2312 /* 2313 * Check that no unknown options are given 2314 */ 2315 int 2316 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 2317 { 2318 struct vfsopt *opt; 2319 char errmsg[255]; 2320 const char **t, *p, *q; 2321 int ret = 0; 2322 2323 TAILQ_FOREACH(opt, opts, link) { 2324 p = opt->name; 2325 q = NULL; 2326 if (p[0] == 'n' && p[1] == 'o') 2327 q = p + 2; 2328 for(t = global_opts; *t != NULL; t++) { 2329 if (strcmp(*t, p) == 0) 2330 break; 2331 if (q != NULL) { 2332 if (strcmp(*t, q) == 0) 2333 break; 2334 } 2335 } 2336 if (*t != NULL) 2337 continue; 2338 for(t = legal; *t != NULL; t++) { 2339 if (strcmp(*t, p) == 0) 2340 break; 2341 if (q != NULL) { 2342 if (strcmp(*t, q) == 0) 2343 break; 2344 } 2345 } 2346 if (*t != NULL) 2347 continue; 2348 snprintf(errmsg, sizeof(errmsg), 2349 "mount option <%s> is unknown", p); 2350 ret = EINVAL; 2351 } 2352 if (ret != 0) { 2353 TAILQ_FOREACH(opt, opts, link) { 2354 if (strcmp(opt->name, "errmsg") == 0) { 2355 strncpy((char *)opt->value, errmsg, opt->len); 2356 break; 2357 } 2358 } 2359 if (opt == NULL) 2360 printf("%s\n", errmsg); 2361 } 2362 return (ret); 2363 } 2364 2365 /* 2366 * Get a mount option by its name. 2367 * 2368 * Return 0 if the option was found, ENOENT otherwise. 2369 * If len is non-NULL it will be filled with the length 2370 * of the option. If buf is non-NULL, it will be filled 2371 * with the address of the option. 2372 */ 2373 int 2374 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len) 2375 { 2376 struct vfsopt *opt; 2377 2378 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2379 2380 TAILQ_FOREACH(opt, opts, link) { 2381 if (strcmp(name, opt->name) == 0) { 2382 opt->seen = 1; 2383 if (len != NULL) 2384 *len = opt->len; 2385 if (buf != NULL) 2386 *buf = opt->value; 2387 return (0); 2388 } 2389 } 2390 return (ENOENT); 2391 } 2392 2393 int 2394 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2395 { 2396 struct vfsopt *opt; 2397 2398 if (opts == NULL) 2399 return (-1); 2400 2401 TAILQ_FOREACH(opt, opts, link) { 2402 if (strcmp(name, opt->name) == 0) { 2403 opt->seen = 1; 2404 return (opt->pos); 2405 } 2406 } 2407 return (-1); 2408 } 2409 2410 int 2411 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value) 2412 { 2413 char *opt_value, *vtp; 2414 quad_t iv; 2415 int error, opt_len; 2416 2417 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len); 2418 if (error != 0) 2419 return (error); 2420 if (opt_len == 0 || opt_value == NULL) 2421 return (EINVAL); 2422 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0') 2423 return (EINVAL); 2424 iv = strtoq(opt_value, &vtp, 0); 2425 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0')) 2426 return (EINVAL); 2427 if (iv < 0) 2428 return (EINVAL); 2429 switch (vtp[0]) { 2430 case 't': case 'T': 2431 iv *= 1024; 2432 /* FALLTHROUGH */ 2433 case 'g': case 'G': 2434 iv *= 1024; 2435 /* FALLTHROUGH */ 2436 case 'm': case 'M': 2437 iv *= 1024; 2438 /* FALLTHROUGH */ 2439 case 'k': case 'K': 2440 iv *= 1024; 2441 case '\0': 2442 break; 2443 default: 2444 return (EINVAL); 2445 } 2446 *value = iv; 2447 2448 return (0); 2449 } 2450 2451 char * 2452 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2453 { 2454 struct vfsopt *opt; 2455 2456 *error = 0; 2457 TAILQ_FOREACH(opt, opts, link) { 2458 if (strcmp(name, opt->name) != 0) 2459 continue; 2460 opt->seen = 1; 2461 if (opt->len == 0 || 2462 ((char *)opt->value)[opt->len - 1] != '\0') { 2463 *error = EINVAL; 2464 return (NULL); 2465 } 2466 return (opt->value); 2467 } 2468 *error = ENOENT; 2469 return (NULL); 2470 } 2471 2472 int 2473 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, 2474 uint64_t val) 2475 { 2476 struct vfsopt *opt; 2477 2478 TAILQ_FOREACH(opt, opts, link) { 2479 if (strcmp(name, opt->name) == 0) { 2480 opt->seen = 1; 2481 if (w != NULL) 2482 *w |= val; 2483 return (1); 2484 } 2485 } 2486 if (w != NULL) 2487 *w &= ~val; 2488 return (0); 2489 } 2490 2491 int 2492 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2493 { 2494 va_list ap; 2495 struct vfsopt *opt; 2496 int ret; 2497 2498 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2499 2500 TAILQ_FOREACH(opt, opts, link) { 2501 if (strcmp(name, opt->name) != 0) 2502 continue; 2503 opt->seen = 1; 2504 if (opt->len == 0 || opt->value == NULL) 2505 return (0); 2506 if (((char *)opt->value)[opt->len - 1] != '\0') 2507 return (0); 2508 va_start(ap, fmt); 2509 ret = vsscanf(opt->value, fmt, ap); 2510 va_end(ap); 2511 return (ret); 2512 } 2513 return (0); 2514 } 2515 2516 int 2517 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2518 { 2519 struct vfsopt *opt; 2520 2521 TAILQ_FOREACH(opt, opts, link) { 2522 if (strcmp(name, opt->name) != 0) 2523 continue; 2524 opt->seen = 1; 2525 if (opt->value == NULL) 2526 opt->len = len; 2527 else { 2528 if (opt->len != len) 2529 return (EINVAL); 2530 bcopy(value, opt->value, len); 2531 } 2532 return (0); 2533 } 2534 return (ENOENT); 2535 } 2536 2537 int 2538 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2539 { 2540 struct vfsopt *opt; 2541 2542 TAILQ_FOREACH(opt, opts, link) { 2543 if (strcmp(name, opt->name) != 0) 2544 continue; 2545 opt->seen = 1; 2546 if (opt->value == NULL) 2547 opt->len = len; 2548 else { 2549 if (opt->len < len) 2550 return (EINVAL); 2551 opt->len = len; 2552 bcopy(value, opt->value, len); 2553 } 2554 return (0); 2555 } 2556 return (ENOENT); 2557 } 2558 2559 int 2560 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2561 { 2562 struct vfsopt *opt; 2563 2564 TAILQ_FOREACH(opt, opts, link) { 2565 if (strcmp(name, opt->name) != 0) 2566 continue; 2567 opt->seen = 1; 2568 if (opt->value == NULL) 2569 opt->len = strlen(value) + 1; 2570 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2571 return (EINVAL); 2572 return (0); 2573 } 2574 return (ENOENT); 2575 } 2576 2577 /* 2578 * Find and copy a mount option. 2579 * 2580 * The size of the buffer has to be specified 2581 * in len, if it is not the same length as the 2582 * mount option, EINVAL is returned. 2583 * Returns ENOENT if the option is not found. 2584 */ 2585 int 2586 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len) 2587 { 2588 struct vfsopt *opt; 2589 2590 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2591 2592 TAILQ_FOREACH(opt, opts, link) { 2593 if (strcmp(name, opt->name) == 0) { 2594 opt->seen = 1; 2595 if (len != opt->len) 2596 return (EINVAL); 2597 bcopy(opt->value, dest, opt->len); 2598 return (0); 2599 } 2600 } 2601 return (ENOENT); 2602 } 2603 2604 int 2605 __vfs_statfs(struct mount *mp, struct statfs *sbp) 2606 { 2607 2608 /* 2609 * Filesystems only fill in part of the structure for updates, we 2610 * have to read the entirety first to get all content. 2611 */ 2612 if (sbp != &mp->mnt_stat) 2613 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp)); 2614 2615 /* 2616 * Set these in case the underlying filesystem fails to do so. 2617 */ 2618 sbp->f_version = STATFS_VERSION; 2619 sbp->f_namemax = NAME_MAX; 2620 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 2621 2622 return (mp->mnt_op->vfs_statfs(mp, sbp)); 2623 } 2624 2625 void 2626 vfs_mountedfrom(struct mount *mp, const char *from) 2627 { 2628 2629 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2630 strlcpy(mp->mnt_stat.f_mntfromname, from, 2631 sizeof mp->mnt_stat.f_mntfromname); 2632 } 2633 2634 /* 2635 * --------------------------------------------------------------------- 2636 * This is the api for building mount args and mounting filesystems from 2637 * inside the kernel. 2638 * 2639 * The API works by accumulation of individual args. First error is 2640 * latched. 2641 * 2642 * XXX: should be documented in new manpage kernel_mount(9) 2643 */ 2644 2645 /* A memory allocation which must be freed when we are done */ 2646 struct mntaarg { 2647 SLIST_ENTRY(mntaarg) next; 2648 }; 2649 2650 /* The header for the mount arguments */ 2651 struct mntarg { 2652 struct iovec *v; 2653 int len; 2654 int error; 2655 SLIST_HEAD(, mntaarg) list; 2656 }; 2657 2658 /* 2659 * Add a boolean argument. 2660 * 2661 * flag is the boolean value. 2662 * name must start with "no". 2663 */ 2664 struct mntarg * 2665 mount_argb(struct mntarg *ma, int flag, const char *name) 2666 { 2667 2668 KASSERT(name[0] == 'n' && name[1] == 'o', 2669 ("mount_argb(...,%s): name must start with 'no'", name)); 2670 2671 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2672 } 2673 2674 /* 2675 * Add an argument printf style 2676 */ 2677 struct mntarg * 2678 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2679 { 2680 va_list ap; 2681 struct mntaarg *maa; 2682 struct sbuf *sb; 2683 int len; 2684 2685 if (ma == NULL) { 2686 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2687 SLIST_INIT(&ma->list); 2688 } 2689 if (ma->error) 2690 return (ma); 2691 2692 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2693 M_MOUNT, M_WAITOK); 2694 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2695 ma->v[ma->len].iov_len = strlen(name) + 1; 2696 ma->len++; 2697 2698 sb = sbuf_new_auto(); 2699 va_start(ap, fmt); 2700 sbuf_vprintf(sb, fmt, ap); 2701 va_end(ap); 2702 sbuf_finish(sb); 2703 len = sbuf_len(sb) + 1; 2704 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2705 SLIST_INSERT_HEAD(&ma->list, maa, next); 2706 bcopy(sbuf_data(sb), maa + 1, len); 2707 sbuf_delete(sb); 2708 2709 ma->v[ma->len].iov_base = maa + 1; 2710 ma->v[ma->len].iov_len = len; 2711 ma->len++; 2712 2713 return (ma); 2714 } 2715 2716 /* 2717 * Add an argument which is a userland string. 2718 */ 2719 struct mntarg * 2720 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2721 { 2722 struct mntaarg *maa; 2723 char *tbuf; 2724 2725 if (val == NULL) 2726 return (ma); 2727 if (ma == NULL) { 2728 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2729 SLIST_INIT(&ma->list); 2730 } 2731 if (ma->error) 2732 return (ma); 2733 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2734 SLIST_INSERT_HEAD(&ma->list, maa, next); 2735 tbuf = (void *)(maa + 1); 2736 ma->error = copyinstr(val, tbuf, len, NULL); 2737 return (mount_arg(ma, name, tbuf, -1)); 2738 } 2739 2740 /* 2741 * Plain argument. 2742 * 2743 * If length is -1, treat value as a C string. 2744 */ 2745 struct mntarg * 2746 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2747 { 2748 2749 if (ma == NULL) { 2750 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2751 SLIST_INIT(&ma->list); 2752 } 2753 if (ma->error) 2754 return (ma); 2755 2756 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2757 M_MOUNT, M_WAITOK); 2758 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2759 ma->v[ma->len].iov_len = strlen(name) + 1; 2760 ma->len++; 2761 2762 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2763 if (len < 0) 2764 ma->v[ma->len].iov_len = strlen(val) + 1; 2765 else 2766 ma->v[ma->len].iov_len = len; 2767 ma->len++; 2768 return (ma); 2769 } 2770 2771 /* 2772 * Free a mntarg structure 2773 */ 2774 static void 2775 free_mntarg(struct mntarg *ma) 2776 { 2777 struct mntaarg *maa; 2778 2779 while (!SLIST_EMPTY(&ma->list)) { 2780 maa = SLIST_FIRST(&ma->list); 2781 SLIST_REMOVE_HEAD(&ma->list, next); 2782 free(maa, M_MOUNT); 2783 } 2784 free(ma->v, M_MOUNT); 2785 free(ma, M_MOUNT); 2786 } 2787 2788 /* 2789 * Mount a filesystem 2790 */ 2791 int 2792 kernel_mount(struct mntarg *ma, uint64_t flags) 2793 { 2794 struct uio auio; 2795 int error; 2796 2797 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2798 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2799 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2800 2801 auio.uio_iov = ma->v; 2802 auio.uio_iovcnt = ma->len; 2803 auio.uio_segflg = UIO_SYSSPACE; 2804 2805 error = ma->error; 2806 if (!error) 2807 error = vfs_donmount(curthread, flags, &auio); 2808 free_mntarg(ma); 2809 return (error); 2810 } 2811 2812 /* 2813 * A printflike function to mount a filesystem. 2814 */ 2815 int 2816 kernel_vmount(int flags, ...) 2817 { 2818 struct mntarg *ma = NULL; 2819 va_list ap; 2820 const char *cp; 2821 const void *vp; 2822 int error; 2823 2824 va_start(ap, flags); 2825 for (;;) { 2826 cp = va_arg(ap, const char *); 2827 if (cp == NULL) 2828 break; 2829 vp = va_arg(ap, const void *); 2830 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2831 } 2832 va_end(ap); 2833 2834 error = kernel_mount(ma, flags); 2835 return (error); 2836 } 2837 2838 /* Map from mount options to printable formats. */ 2839 static struct mntoptnames optnames[] = { 2840 MNTOPT_NAMES 2841 }; 2842 2843 #define DEVCTL_LEN 1024 2844 static void 2845 mount_devctl_event(const char *type, struct mount *mp, bool donew) 2846 { 2847 const uint8_t *cp; 2848 struct mntoptnames *fp; 2849 struct sbuf sb; 2850 struct statfs *sfp = &mp->mnt_stat; 2851 char *buf; 2852 2853 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT); 2854 if (buf == NULL) 2855 return; 2856 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN); 2857 sbuf_cpy(&sb, "mount-point=\""); 2858 devctl_safe_quote_sb(&sb, sfp->f_mntonname); 2859 sbuf_cat(&sb, "\" mount-dev=\""); 2860 devctl_safe_quote_sb(&sb, sfp->f_mntfromname); 2861 sbuf_cat(&sb, "\" mount-type=\""); 2862 devctl_safe_quote_sb(&sb, sfp->f_fstypename); 2863 sbuf_cat(&sb, "\" fsid=0x"); 2864 cp = (const uint8_t *)&sfp->f_fsid.val[0]; 2865 for (int i = 0; i < sizeof(sfp->f_fsid); i++) 2866 sbuf_printf(&sb, "%02x", cp[i]); 2867 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner); 2868 for (fp = optnames; fp->o_opt != 0; fp++) { 2869 if ((mp->mnt_flag & fp->o_opt) != 0) { 2870 sbuf_cat(&sb, fp->o_name); 2871 sbuf_putc(&sb, ';'); 2872 } 2873 } 2874 sbuf_putc(&sb, '"'); 2875 sbuf_finish(&sb); 2876 2877 /* 2878 * Options are not published because the form of the options depends on 2879 * the file system and may include binary data. In addition, they don't 2880 * necessarily provide enough useful information to be actionable when 2881 * devd processes them. 2882 */ 2883 2884 if (sbuf_error(&sb) == 0) 2885 devctl_notify("VFS", "FS", type, sbuf_data(&sb)); 2886 sbuf_delete(&sb); 2887 free(buf, M_MOUNT); 2888 } 2889 2890 /* 2891 * Suspend write operations on all local writeable filesystems. Does 2892 * full sync of them in the process. 2893 * 2894 * Iterate over the mount points in reverse order, suspending most 2895 * recently mounted filesystems first. It handles a case where a 2896 * filesystem mounted from a md(4) vnode-backed device should be 2897 * suspended before the filesystem that owns the vnode. 2898 */ 2899 void 2900 suspend_all_fs(void) 2901 { 2902 struct mount *mp; 2903 int error; 2904 2905 mtx_lock(&mountlist_mtx); 2906 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 2907 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT); 2908 if (error != 0) 2909 continue; 2910 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL || 2911 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 2912 mtx_lock(&mountlist_mtx); 2913 vfs_unbusy(mp); 2914 continue; 2915 } 2916 error = vfs_write_suspend(mp, 0); 2917 if (error == 0) { 2918 MNT_ILOCK(mp); 2919 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0); 2920 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL; 2921 MNT_IUNLOCK(mp); 2922 mtx_lock(&mountlist_mtx); 2923 } else { 2924 printf("suspend of %s failed, error %d\n", 2925 mp->mnt_stat.f_mntonname, error); 2926 mtx_lock(&mountlist_mtx); 2927 vfs_unbusy(mp); 2928 } 2929 } 2930 mtx_unlock(&mountlist_mtx); 2931 } 2932 2933 void 2934 resume_all_fs(void) 2935 { 2936 struct mount *mp; 2937 2938 mtx_lock(&mountlist_mtx); 2939 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2940 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0) 2941 continue; 2942 mtx_unlock(&mountlist_mtx); 2943 MNT_ILOCK(mp); 2944 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0); 2945 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL; 2946 MNT_IUNLOCK(mp); 2947 vfs_write_resume(mp, 0); 2948 mtx_lock(&mountlist_mtx); 2949 vfs_unbusy(mp); 2950 } 2951 mtx_unlock(&mountlist_mtx); 2952 } 2953