Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_mount.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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/malloc.h> 53 #include <sys/mount.h> 54 #include <sys/mutex.h> 55 #include <sys/namei.h> 56 #include <sys/priv.h> 57 #include <sys/proc.h> 58 #include <sys/filedesc.h> 59 #include <sys/reboot.h> 60 #include <sys/sbuf.h> 61 #include <sys/syscallsubr.h> 62 #include <sys/sysproto.h> 63 #include <sys/sx.h> 64 #include <sys/sysctl.h> 65 #include <sys/sysent.h> 66 #include <sys/systm.h> 67 #include <sys/vnode.h> 68 #include <vm/uma.h> 69 70 #include <geom/geom.h> 71 72 #include <machine/stdarg.h> 73 74 #include <security/audit/audit.h> 75 #include <security/mac/mac_framework.h> 76 77 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 78 79 static int vfs_domount(struct thread *td, const char *fstype, char *fspath, 80 uint64_t fsflags, struct vfsoptlist **optlist); 81 static void free_mntarg(struct mntarg *ma); 82 83 static int usermount = 0; 84 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 85 "Unprivileged users may mount and unmount file systems"); 86 87 static bool default_autoro = false; 88 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0, 89 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified"); 90 91 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 92 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure"); 93 static uma_zone_t mount_zone; 94 95 /* List of mounted filesystems. */ 96 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 97 98 /* For any iteration/modification of mountlist */ 99 struct mtx_padalign __exclusive_cache_line mountlist_mtx; 100 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 101 102 EVENTHANDLER_LIST_DEFINE(vfs_mounted); 103 EVENTHANDLER_LIST_DEFINE(vfs_unmounted); 104 105 static void mount_devctl_event(const char *type, struct mount *mp, bool donew); 106 107 /* 108 * Global opts, taken by all filesystems 109 */ 110 static const char *global_opts[] = { 111 "errmsg", 112 "fstype", 113 "fspath", 114 "ro", 115 "rw", 116 "nosuid", 117 "noexec", 118 NULL 119 }; 120 121 static int 122 mount_init(void *mem, int size, int flags) 123 { 124 struct mount *mp; 125 126 mp = (struct mount *)mem; 127 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 128 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF); 129 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 130 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO); 131 mp->mnt_ref = 0; 132 mp->mnt_vfs_ops = 1; 133 mp->mnt_rootvnode = NULL; 134 return (0); 135 } 136 137 static void 138 mount_fini(void *mem, int size) 139 { 140 struct mount *mp; 141 142 mp = (struct mount *)mem; 143 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu); 144 lockdestroy(&mp->mnt_explock); 145 mtx_destroy(&mp->mnt_listmtx); 146 mtx_destroy(&mp->mnt_mtx); 147 } 148 149 static void 150 vfs_mount_init(void *dummy __unused) 151 { 152 153 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL, 154 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE); 155 } 156 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL); 157 158 /* 159 * --------------------------------------------------------------------- 160 * Functions for building and sanitizing the mount options 161 */ 162 163 /* Remove one mount option. */ 164 static void 165 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 166 { 167 168 TAILQ_REMOVE(opts, opt, link); 169 free(opt->name, M_MOUNT); 170 if (opt->value != NULL) 171 free(opt->value, M_MOUNT); 172 free(opt, M_MOUNT); 173 } 174 175 /* Release all resources related to the mount options. */ 176 void 177 vfs_freeopts(struct vfsoptlist *opts) 178 { 179 struct vfsopt *opt; 180 181 while (!TAILQ_EMPTY(opts)) { 182 opt = TAILQ_FIRST(opts); 183 vfs_freeopt(opts, opt); 184 } 185 free(opts, M_MOUNT); 186 } 187 188 void 189 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 190 { 191 struct vfsopt *opt, *temp; 192 193 if (opts == NULL) 194 return; 195 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 196 if (strcmp(opt->name, name) == 0) 197 vfs_freeopt(opts, opt); 198 } 199 } 200 201 static int 202 vfs_isopt_ro(const char *opt) 203 { 204 205 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 || 206 strcmp(opt, "norw") == 0) 207 return (1); 208 return (0); 209 } 210 211 static int 212 vfs_isopt_rw(const char *opt) 213 { 214 215 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0) 216 return (1); 217 return (0); 218 } 219 220 /* 221 * Check if options are equal (with or without the "no" prefix). 222 */ 223 static int 224 vfs_equalopts(const char *opt1, const char *opt2) 225 { 226 char *p; 227 228 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 229 if (strcmp(opt1, opt2) == 0) 230 return (1); 231 /* "noopt" vs. "opt" */ 232 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 233 return (1); 234 /* "opt" vs. "noopt" */ 235 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 236 return (1); 237 while ((p = strchr(opt1, '.')) != NULL && 238 !strncmp(opt1, opt2, ++p - opt1)) { 239 opt2 += p - opt1; 240 opt1 = p; 241 /* "foo.noopt" vs. "foo.opt" */ 242 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 243 return (1); 244 /* "foo.opt" vs. "foo.noopt" */ 245 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 246 return (1); 247 } 248 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */ 249 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) && 250 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2))) 251 return (1); 252 return (0); 253 } 254 255 /* 256 * If a mount option is specified several times, 257 * (with or without the "no" prefix) only keep 258 * the last occurrence of it. 259 */ 260 static void 261 vfs_sanitizeopts(struct vfsoptlist *opts) 262 { 263 struct vfsopt *opt, *opt2, *tmp; 264 265 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 266 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 267 while (opt2 != NULL) { 268 if (vfs_equalopts(opt->name, opt2->name)) { 269 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 270 vfs_freeopt(opts, opt2); 271 opt2 = tmp; 272 } else { 273 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 274 } 275 } 276 } 277 } 278 279 /* 280 * Build a linked list of mount options from a struct uio. 281 */ 282 int 283 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 284 { 285 struct vfsoptlist *opts; 286 struct vfsopt *opt; 287 size_t memused, namelen, optlen; 288 unsigned int i, iovcnt; 289 int error; 290 291 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 292 TAILQ_INIT(opts); 293 memused = 0; 294 iovcnt = auio->uio_iovcnt; 295 for (i = 0; i < iovcnt; i += 2) { 296 namelen = auio->uio_iov[i].iov_len; 297 optlen = auio->uio_iov[i + 1].iov_len; 298 memused += sizeof(struct vfsopt) + optlen + namelen; 299 /* 300 * Avoid consuming too much memory, and attempts to overflow 301 * memused. 302 */ 303 if (memused > VFS_MOUNTARG_SIZE_MAX || 304 optlen > VFS_MOUNTARG_SIZE_MAX || 305 namelen > VFS_MOUNTARG_SIZE_MAX) { 306 error = EINVAL; 307 goto bad; 308 } 309 310 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 311 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 312 opt->value = NULL; 313 opt->len = 0; 314 opt->pos = i / 2; 315 opt->seen = 0; 316 317 /* 318 * Do this early, so jumps to "bad" will free the current 319 * option. 320 */ 321 TAILQ_INSERT_TAIL(opts, opt, link); 322 323 if (auio->uio_segflg == UIO_SYSSPACE) { 324 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 325 } else { 326 error = copyin(auio->uio_iov[i].iov_base, opt->name, 327 namelen); 328 if (error) 329 goto bad; 330 } 331 /* Ensure names are null-terminated strings. */ 332 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 333 error = EINVAL; 334 goto bad; 335 } 336 if (optlen != 0) { 337 opt->len = optlen; 338 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 339 if (auio->uio_segflg == UIO_SYSSPACE) { 340 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 341 optlen); 342 } else { 343 error = copyin(auio->uio_iov[i + 1].iov_base, 344 opt->value, optlen); 345 if (error) 346 goto bad; 347 } 348 } 349 } 350 vfs_sanitizeopts(opts); 351 *options = opts; 352 return (0); 353 bad: 354 vfs_freeopts(opts); 355 return (error); 356 } 357 358 /* 359 * Merge the old mount options with the new ones passed 360 * in the MNT_UPDATE case. 361 * 362 * XXX: This function will keep a "nofoo" option in the new 363 * options. E.g, if the option's canonical name is "foo", 364 * "nofoo" ends up in the mount point's active options. 365 */ 366 static void 367 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts) 368 { 369 struct vfsopt *opt, *new; 370 371 TAILQ_FOREACH(opt, oldopts, link) { 372 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 373 new->name = strdup(opt->name, M_MOUNT); 374 if (opt->len != 0) { 375 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 376 bcopy(opt->value, new->value, opt->len); 377 } else 378 new->value = NULL; 379 new->len = opt->len; 380 new->seen = opt->seen; 381 TAILQ_INSERT_HEAD(toopts, new, link); 382 } 383 vfs_sanitizeopts(toopts); 384 } 385 386 /* 387 * Mount a filesystem. 388 */ 389 #ifndef _SYS_SYSPROTO_H_ 390 struct nmount_args { 391 struct iovec *iovp; 392 unsigned int iovcnt; 393 int flags; 394 }; 395 #endif 396 int 397 sys_nmount(struct thread *td, struct nmount_args *uap) 398 { 399 struct uio *auio; 400 int error; 401 u_int iovcnt; 402 uint64_t flags; 403 404 /* 405 * Mount flags are now 64-bits. On 32-bit archtectures only 406 * 32-bits are passed in, but from here on everything handles 407 * 64-bit flags correctly. 408 */ 409 flags = uap->flags; 410 411 AUDIT_ARG_FFLAGS(flags); 412 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 413 uap->iovp, uap->iovcnt, flags); 414 415 /* 416 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 417 * userspace to set this flag, but we must filter it out if we want 418 * MNT_UPDATE on the root file system to work. 419 * MNT_ROOTFS should only be set by the kernel when mounting its 420 * root file system. 421 */ 422 flags &= ~MNT_ROOTFS; 423 424 iovcnt = uap->iovcnt; 425 /* 426 * Check that we have an even number of iovec's 427 * and that we have at least two options. 428 */ 429 if ((iovcnt & 1) || (iovcnt < 4)) { 430 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 431 uap->iovcnt); 432 return (EINVAL); 433 } 434 435 error = copyinuio(uap->iovp, iovcnt, &auio); 436 if (error) { 437 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 438 __func__, error); 439 return (error); 440 } 441 error = vfs_donmount(td, flags, auio); 442 443 free(auio, M_IOV); 444 return (error); 445 } 446 447 /* 448 * --------------------------------------------------------------------- 449 * Various utility functions 450 */ 451 452 void 453 vfs_ref(struct mount *mp) 454 { 455 struct mount_pcpu *mpcpu; 456 457 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 458 if (vfs_op_thread_enter(mp, mpcpu)) { 459 vfs_mp_count_add_pcpu(mpcpu, ref, 1); 460 vfs_op_thread_exit(mp, mpcpu); 461 return; 462 } 463 464 MNT_ILOCK(mp); 465 MNT_REF(mp); 466 MNT_IUNLOCK(mp); 467 } 468 469 void 470 vfs_rel(struct mount *mp) 471 { 472 struct mount_pcpu *mpcpu; 473 474 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 475 if (vfs_op_thread_enter(mp, mpcpu)) { 476 vfs_mp_count_sub_pcpu(mpcpu, ref, 1); 477 vfs_op_thread_exit(mp, mpcpu); 478 return; 479 } 480 481 MNT_ILOCK(mp); 482 MNT_REL(mp); 483 MNT_IUNLOCK(mp); 484 } 485 486 /* 487 * Allocate and initialize the mount point struct. 488 */ 489 struct mount * 490 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 491 struct ucred *cred) 492 { 493 struct mount *mp; 494 495 mp = uma_zalloc(mount_zone, M_WAITOK); 496 bzero(&mp->mnt_startzero, 497 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 498 mp->mnt_kern_flag = 0; 499 mp->mnt_flag = 0; 500 mp->mnt_rootvnode = NULL; 501 mp->mnt_vnodecovered = NULL; 502 mp->mnt_op = NULL; 503 mp->mnt_vfc = NULL; 504 TAILQ_INIT(&mp->mnt_nvnodelist); 505 mp->mnt_nvnodelistsize = 0; 506 TAILQ_INIT(&mp->mnt_lazyvnodelist); 507 mp->mnt_lazyvnodelistsize = 0; 508 if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 || 509 mp->mnt_writeopcount != 0) 510 panic("%s: non-zero counters on new mp %p\n", __func__, mp); 511 if (mp->mnt_vfs_ops != 1) 512 panic("%s: vfs_ops should be 1 but %d found\n", __func__, 513 mp->mnt_vfs_ops); 514 (void) vfs_busy(mp, MBF_NOWAIT); 515 atomic_add_acq_int(&vfsp->vfc_refcount, 1); 516 mp->mnt_op = vfsp->vfc_vfsops; 517 mp->mnt_vfc = vfsp; 518 mp->mnt_stat.f_type = vfsp->vfc_typenum; 519 mp->mnt_gen++; 520 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 521 mp->mnt_vnodecovered = vp; 522 mp->mnt_cred = crdup(cred); 523 mp->mnt_stat.f_owner = cred->cr_uid; 524 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 525 mp->mnt_iosize_max = DFLTPHYS; 526 #ifdef MAC 527 mac_mount_init(mp); 528 mac_mount_create(cred, mp); 529 #endif 530 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 531 TAILQ_INIT(&mp->mnt_uppers); 532 return (mp); 533 } 534 535 /* 536 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 537 */ 538 void 539 vfs_mount_destroy(struct mount *mp) 540 { 541 542 if (mp->mnt_vfs_ops == 0) 543 panic("%s: entered with zero vfs_ops\n", __func__); 544 545 vfs_assert_mount_counters(mp); 546 547 MNT_ILOCK(mp); 548 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 549 if (mp->mnt_kern_flag & MNTK_MWAIT) { 550 mp->mnt_kern_flag &= ~MNTK_MWAIT; 551 wakeup(mp); 552 } 553 while (mp->mnt_ref) 554 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 555 KASSERT(mp->mnt_ref == 0, 556 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 557 __FILE__, __LINE__)); 558 if (mp->mnt_writeopcount != 0) 559 panic("vfs_mount_destroy: nonzero writeopcount"); 560 if (mp->mnt_secondary_writes != 0) 561 panic("vfs_mount_destroy: nonzero secondary_writes"); 562 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1); 563 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 564 struct vnode *vp; 565 566 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 567 vn_printf(vp, "dangling vnode "); 568 panic("unmount: dangling vnode"); 569 } 570 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers")); 571 if (mp->mnt_nvnodelistsize != 0) 572 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 573 if (mp->mnt_lazyvnodelistsize != 0) 574 panic("vfs_mount_destroy: nonzero lazyvnodelistsize"); 575 if (mp->mnt_lockref != 0) 576 panic("vfs_mount_destroy: nonzero lock refcount"); 577 MNT_IUNLOCK(mp); 578 579 if (mp->mnt_vfs_ops != 1) 580 panic("%s: vfs_ops should be 1 but %d found\n", __func__, 581 mp->mnt_vfs_ops); 582 583 if (mp->mnt_rootvnode != NULL) 584 panic("%s: mount point still has a root vnode %p\n", __func__, 585 mp->mnt_rootvnode); 586 587 if (mp->mnt_vnodecovered != NULL) 588 vrele(mp->mnt_vnodecovered); 589 #ifdef MAC 590 mac_mount_destroy(mp); 591 #endif 592 if (mp->mnt_opt != NULL) 593 vfs_freeopts(mp->mnt_opt); 594 crfree(mp->mnt_cred); 595 uma_zfree(mount_zone, mp); 596 } 597 598 static bool 599 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error) 600 { 601 /* This is an upgrade of an exisiting mount. */ 602 if ((fsflags & MNT_UPDATE) != 0) 603 return (false); 604 /* This is already an R/O mount. */ 605 if ((fsflags & MNT_RDONLY) != 0) 606 return (false); 607 608 switch (error) { 609 case ENODEV: /* generic, geom, ... */ 610 case EACCES: /* cam/scsi, ... */ 611 case EROFS: /* md, mmcsd, ... */ 612 /* 613 * These errors can be returned by the storage layer to signal 614 * that the media is read-only. No harm in the R/O mount 615 * attempt if the error was returned for some other reason. 616 */ 617 return (true); 618 default: 619 return (false); 620 } 621 } 622 623 int 624 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions) 625 { 626 struct vfsoptlist *optlist; 627 struct vfsopt *opt, *tmp_opt; 628 char *fstype, *fspath, *errmsg; 629 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 630 bool autoro; 631 632 errmsg = fspath = NULL; 633 errmsg_len = fspathlen = 0; 634 errmsg_pos = -1; 635 autoro = default_autoro; 636 637 error = vfs_buildopts(fsoptions, &optlist); 638 if (error) 639 return (error); 640 641 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 642 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 643 644 /* 645 * We need these two options before the others, 646 * and they are mandatory for any filesystem. 647 * Ensure they are NUL terminated as well. 648 */ 649 fstypelen = 0; 650 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 651 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') { 652 error = EINVAL; 653 if (errmsg != NULL) 654 strncpy(errmsg, "Invalid fstype", errmsg_len); 655 goto bail; 656 } 657 fspathlen = 0; 658 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 659 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') { 660 error = EINVAL; 661 if (errmsg != NULL) 662 strncpy(errmsg, "Invalid fspath", errmsg_len); 663 goto bail; 664 } 665 666 /* 667 * We need to see if we have the "update" option 668 * before we call vfs_domount(), since vfs_domount() has special 669 * logic based on MNT_UPDATE. This is very important 670 * when we want to update the root filesystem. 671 */ 672 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 673 int do_freeopt = 0; 674 675 if (strcmp(opt->name, "update") == 0) { 676 fsflags |= MNT_UPDATE; 677 do_freeopt = 1; 678 } 679 else if (strcmp(opt->name, "async") == 0) 680 fsflags |= MNT_ASYNC; 681 else if (strcmp(opt->name, "force") == 0) { 682 fsflags |= MNT_FORCE; 683 do_freeopt = 1; 684 } 685 else if (strcmp(opt->name, "reload") == 0) { 686 fsflags |= MNT_RELOAD; 687 do_freeopt = 1; 688 } 689 else if (strcmp(opt->name, "multilabel") == 0) 690 fsflags |= MNT_MULTILABEL; 691 else if (strcmp(opt->name, "noasync") == 0) 692 fsflags &= ~MNT_ASYNC; 693 else if (strcmp(opt->name, "noatime") == 0) 694 fsflags |= MNT_NOATIME; 695 else if (strcmp(opt->name, "atime") == 0) { 696 free(opt->name, M_MOUNT); 697 opt->name = strdup("nonoatime", M_MOUNT); 698 } 699 else if (strcmp(opt->name, "noclusterr") == 0) 700 fsflags |= MNT_NOCLUSTERR; 701 else if (strcmp(opt->name, "clusterr") == 0) { 702 free(opt->name, M_MOUNT); 703 opt->name = strdup("nonoclusterr", M_MOUNT); 704 } 705 else if (strcmp(opt->name, "noclusterw") == 0) 706 fsflags |= MNT_NOCLUSTERW; 707 else if (strcmp(opt->name, "clusterw") == 0) { 708 free(opt->name, M_MOUNT); 709 opt->name = strdup("nonoclusterw", M_MOUNT); 710 } 711 else if (strcmp(opt->name, "noexec") == 0) 712 fsflags |= MNT_NOEXEC; 713 else if (strcmp(opt->name, "exec") == 0) { 714 free(opt->name, M_MOUNT); 715 opt->name = strdup("nonoexec", M_MOUNT); 716 } 717 else if (strcmp(opt->name, "nosuid") == 0) 718 fsflags |= MNT_NOSUID; 719 else if (strcmp(opt->name, "suid") == 0) { 720 free(opt->name, M_MOUNT); 721 opt->name = strdup("nonosuid", M_MOUNT); 722 } 723 else if (strcmp(opt->name, "nosymfollow") == 0) 724 fsflags |= MNT_NOSYMFOLLOW; 725 else if (strcmp(opt->name, "symfollow") == 0) { 726 free(opt->name, M_MOUNT); 727 opt->name = strdup("nonosymfollow", M_MOUNT); 728 } 729 else if (strcmp(opt->name, "noro") == 0) { 730 fsflags &= ~MNT_RDONLY; 731 autoro = false; 732 } 733 else if (strcmp(opt->name, "rw") == 0) { 734 fsflags &= ~MNT_RDONLY; 735 autoro = false; 736 } 737 else if (strcmp(opt->name, "ro") == 0) { 738 fsflags |= MNT_RDONLY; 739 autoro = false; 740 } 741 else if (strcmp(opt->name, "rdonly") == 0) { 742 free(opt->name, M_MOUNT); 743 opt->name = strdup("ro", M_MOUNT); 744 fsflags |= MNT_RDONLY; 745 autoro = false; 746 } 747 else if (strcmp(opt->name, "autoro") == 0) { 748 do_freeopt = 1; 749 autoro = true; 750 } 751 else if (strcmp(opt->name, "suiddir") == 0) 752 fsflags |= MNT_SUIDDIR; 753 else if (strcmp(opt->name, "sync") == 0) 754 fsflags |= MNT_SYNCHRONOUS; 755 else if (strcmp(opt->name, "union") == 0) 756 fsflags |= MNT_UNION; 757 else if (strcmp(opt->name, "automounted") == 0) { 758 fsflags |= MNT_AUTOMOUNTED; 759 do_freeopt = 1; 760 } else if (strcmp(opt->name, "nocover") == 0) { 761 fsflags |= MNT_NOCOVER; 762 do_freeopt = 1; 763 } else if (strcmp(opt->name, "cover") == 0) { 764 fsflags &= ~MNT_NOCOVER; 765 do_freeopt = 1; 766 } else if (strcmp(opt->name, "emptydir") == 0) { 767 fsflags |= MNT_EMPTYDIR; 768 do_freeopt = 1; 769 } else if (strcmp(opt->name, "noemptydir") == 0) { 770 fsflags &= ~MNT_EMPTYDIR; 771 do_freeopt = 1; 772 } 773 if (do_freeopt) 774 vfs_freeopt(optlist, opt); 775 } 776 777 /* 778 * Be ultra-paranoid about making sure the type and fspath 779 * variables will fit in our mp buffers, including the 780 * terminating NUL. 781 */ 782 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) { 783 error = ENAMETOOLONG; 784 goto bail; 785 } 786 787 error = vfs_domount(td, fstype, fspath, fsflags, &optlist); 788 789 /* 790 * See if we can mount in the read-only mode if the error code suggests 791 * that it could be possible and the mount options allow for that. 792 * Never try it if "[no]{ro|rw}" has been explicitly requested and not 793 * overridden by "autoro". 794 */ 795 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) { 796 printf("%s: R/W mount failed, possibly R/O media," 797 " trying R/O mount\n", __func__); 798 fsflags |= MNT_RDONLY; 799 error = vfs_domount(td, fstype, fspath, fsflags, &optlist); 800 } 801 bail: 802 /* copyout the errmsg */ 803 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 804 && errmsg_len > 0 && errmsg != NULL) { 805 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 806 bcopy(errmsg, 807 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 808 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 809 } else { 810 copyout(errmsg, 811 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 812 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 813 } 814 } 815 816 if (optlist != NULL) 817 vfs_freeopts(optlist); 818 return (error); 819 } 820 821 /* 822 * Old mount API. 823 */ 824 #ifndef _SYS_SYSPROTO_H_ 825 struct mount_args { 826 char *type; 827 char *path; 828 int flags; 829 caddr_t data; 830 }; 831 #endif 832 /* ARGSUSED */ 833 int 834 sys_mount(struct thread *td, struct mount_args *uap) 835 { 836 char *fstype; 837 struct vfsconf *vfsp = NULL; 838 struct mntarg *ma = NULL; 839 uint64_t flags; 840 int error; 841 842 /* 843 * Mount flags are now 64-bits. On 32-bit architectures only 844 * 32-bits are passed in, but from here on everything handles 845 * 64-bit flags correctly. 846 */ 847 flags = uap->flags; 848 849 AUDIT_ARG_FFLAGS(flags); 850 851 /* 852 * Filter out MNT_ROOTFS. We do not want clients of mount() in 853 * userspace to set this flag, but we must filter it out if we want 854 * MNT_UPDATE on the root file system to work. 855 * MNT_ROOTFS should only be set by the kernel when mounting its 856 * root file system. 857 */ 858 flags &= ~MNT_ROOTFS; 859 860 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 861 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 862 if (error) { 863 free(fstype, M_TEMP); 864 return (error); 865 } 866 867 AUDIT_ARG_TEXT(fstype); 868 vfsp = vfs_byname_kld(fstype, td, &error); 869 free(fstype, M_TEMP); 870 if (vfsp == NULL) 871 return (ENOENT); 872 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 && 873 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) || 874 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 && 875 vfsp->vfc_vfsops->vfs_cmount == NULL)) 876 return (EOPNOTSUPP); 877 878 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN); 879 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 880 ma = mount_argb(ma, flags & MNT_RDONLY, "noro"); 881 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid"); 882 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec"); 883 884 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0) 885 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags)); 886 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags)); 887 } 888 889 /* 890 * vfs_domount_first(): first file system mount (not update) 891 */ 892 static int 893 vfs_domount_first( 894 struct thread *td, /* Calling thread. */ 895 struct vfsconf *vfsp, /* File system type. */ 896 char *fspath, /* Mount path. */ 897 struct vnode *vp, /* Vnode to be covered. */ 898 uint64_t fsflags, /* Flags common to all filesystems. */ 899 struct vfsoptlist **optlist /* Options local to the filesystem. */ 900 ) 901 { 902 struct vattr va; 903 struct mount *mp; 904 struct vnode *newdp, *rootvp; 905 int error, error1; 906 bool unmounted; 907 908 ASSERT_VOP_ELOCKED(vp, __func__); 909 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 910 911 if ((fsflags & MNT_EMPTYDIR) != 0) { 912 error = vfs_emptydir(vp); 913 if (error != 0) { 914 vput(vp); 915 return (error); 916 } 917 } 918 919 /* 920 * If the jail of the calling thread lacks permission for this type of 921 * file system, or is trying to cover its own root, deny immediately. 922 */ 923 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred, 924 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) { 925 vput(vp); 926 return (EPERM); 927 } 928 929 /* 930 * If the user is not root, ensure that they own the directory 931 * onto which we are attempting to mount. 932 */ 933 error = VOP_GETATTR(vp, &va, td->td_ucred); 934 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 935 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN); 936 if (error == 0) 937 error = vinvalbuf(vp, V_SAVE, 0, 0); 938 if (error == 0 && vp->v_type != VDIR) 939 error = ENOTDIR; 940 if (error == 0) { 941 VI_LOCK(vp); 942 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 943 vp->v_iflag |= VI_MOUNT; 944 else 945 error = EBUSY; 946 VI_UNLOCK(vp); 947 } 948 if (error != 0) { 949 vput(vp); 950 return (error); 951 } 952 vn_seqc_write_begin(vp); 953 VOP_UNLOCK(vp); 954 955 /* Allocate and initialize the filesystem. */ 956 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 957 /* XXXMAC: pass to vfs_mount_alloc? */ 958 mp->mnt_optnew = *optlist; 959 /* Set the mount level flags. */ 960 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 961 962 /* 963 * Mount the filesystem. 964 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 965 * get. No freeing of cn_pnbuf. 966 */ 967 error1 = 0; 968 unmounted = true; 969 if ((error = VFS_MOUNT(mp)) != 0 || 970 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 || 971 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) { 972 rootvp = NULL; 973 if (error1 != 0) { 974 MPASS(error == 0); 975 rootvp = vfs_cache_root_clear(mp); 976 if (rootvp != NULL) { 977 vhold(rootvp); 978 vrele(rootvp); 979 } 980 (void)vn_start_write(NULL, &mp, V_WAIT); 981 MNT_ILOCK(mp); 982 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF; 983 MNT_IUNLOCK(mp); 984 VFS_PURGE(mp); 985 error = VFS_UNMOUNT(mp, 0); 986 vn_finished_write(mp); 987 if (error != 0) { 988 printf( 989 "failed post-mount (%d): rollback unmount returned %d\n", 990 error1, error); 991 unmounted = false; 992 } 993 error = error1; 994 } 995 vfs_unbusy(mp); 996 mp->mnt_vnodecovered = NULL; 997 if (unmounted) { 998 /* XXXKIB wait for mnt_lockref drain? */ 999 vfs_mount_destroy(mp); 1000 } 1001 VI_LOCK(vp); 1002 vp->v_iflag &= ~VI_MOUNT; 1003 VI_UNLOCK(vp); 1004 if (rootvp != NULL) { 1005 vn_seqc_write_end(rootvp); 1006 vdrop(rootvp); 1007 } 1008 vn_seqc_write_end(vp); 1009 vrele(vp); 1010 return (error); 1011 } 1012 vn_seqc_write_begin(newdp); 1013 VOP_UNLOCK(newdp); 1014 1015 if (mp->mnt_opt != NULL) 1016 vfs_freeopts(mp->mnt_opt); 1017 mp->mnt_opt = mp->mnt_optnew; 1018 *optlist = NULL; 1019 1020 /* 1021 * Prevent external consumers of mount options from reading mnt_optnew. 1022 */ 1023 mp->mnt_optnew = NULL; 1024 1025 MNT_ILOCK(mp); 1026 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1027 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1028 mp->mnt_kern_flag |= MNTK_ASYNC; 1029 else 1030 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1031 MNT_IUNLOCK(mp); 1032 1033 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1034 cache_purge(vp); 1035 VI_LOCK(vp); 1036 vp->v_iflag &= ~VI_MOUNT; 1037 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT); 1038 vp->v_mountedhere = mp; 1039 VI_UNLOCK(vp); 1040 /* Place the new filesystem at the end of the mount list. */ 1041 mtx_lock(&mountlist_mtx); 1042 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1043 mtx_unlock(&mountlist_mtx); 1044 vfs_event_signal(NULL, VQ_MOUNT, 0); 1045 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY); 1046 VOP_UNLOCK(vp); 1047 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td); 1048 VOP_UNLOCK(newdp); 1049 mount_devctl_event("MOUNT", mp, false); 1050 mountcheckdirs(vp, newdp); 1051 vn_seqc_write_end(vp); 1052 vn_seqc_write_end(newdp); 1053 vrele(newdp); 1054 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1055 vfs_allocate_syncvnode(mp); 1056 vfs_op_exit(mp); 1057 vfs_unbusy(mp); 1058 return (0); 1059 } 1060 1061 /* 1062 * vfs_domount_update(): update of mounted file system 1063 */ 1064 static int 1065 vfs_domount_update( 1066 struct thread *td, /* Calling thread. */ 1067 struct vnode *vp, /* Mount point vnode. */ 1068 uint64_t fsflags, /* Flags common to all filesystems. */ 1069 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1070 ) 1071 { 1072 struct export_args export; 1073 struct o2export_args o2export; 1074 struct vnode *rootvp; 1075 void *bufp; 1076 struct mount *mp; 1077 int error, export_error, i, len; 1078 uint64_t flag; 1079 gid_t *grps; 1080 1081 ASSERT_VOP_ELOCKED(vp, __func__); 1082 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 1083 mp = vp->v_mount; 1084 1085 if ((vp->v_vflag & VV_ROOT) == 0) { 1086 if (vfs_copyopt(*optlist, "export", &export, sizeof(export)) 1087 == 0) 1088 error = EXDEV; 1089 else 1090 error = EINVAL; 1091 vput(vp); 1092 return (error); 1093 } 1094 1095 /* 1096 * We only allow the filesystem to be reloaded if it 1097 * is currently mounted read-only. 1098 */ 1099 flag = mp->mnt_flag; 1100 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 1101 vput(vp); 1102 return (EOPNOTSUPP); /* Needs translation */ 1103 } 1104 /* 1105 * Only privileged root, or (if MNT_USER is set) the user that 1106 * did the original mount is permitted to update it. 1107 */ 1108 error = vfs_suser(mp, td); 1109 if (error != 0) { 1110 vput(vp); 1111 return (error); 1112 } 1113 if (vfs_busy(mp, MBF_NOWAIT)) { 1114 vput(vp); 1115 return (EBUSY); 1116 } 1117 VI_LOCK(vp); 1118 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 1119 VI_UNLOCK(vp); 1120 vfs_unbusy(mp); 1121 vput(vp); 1122 return (EBUSY); 1123 } 1124 vp->v_iflag |= VI_MOUNT; 1125 VI_UNLOCK(vp); 1126 VOP_UNLOCK(vp); 1127 1128 vfs_op_enter(mp); 1129 vn_seqc_write_begin(vp); 1130 1131 rootvp = NULL; 1132 MNT_ILOCK(mp); 1133 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1134 MNT_IUNLOCK(mp); 1135 error = EBUSY; 1136 goto end; 1137 } 1138 mp->mnt_flag &= ~MNT_UPDATEMASK; 1139 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 1140 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 1141 if ((mp->mnt_flag & MNT_ASYNC) == 0) 1142 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1143 rootvp = vfs_cache_root_clear(mp); 1144 MNT_IUNLOCK(mp); 1145 mp->mnt_optnew = *optlist; 1146 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 1147 1148 /* 1149 * Mount the filesystem. 1150 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1151 * get. No freeing of cn_pnbuf. 1152 */ 1153 error = VFS_MOUNT(mp); 1154 1155 export_error = 0; 1156 /* Process the export option. */ 1157 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp, 1158 &len) == 0) { 1159 /* Assume that there is only 1 ABI for each length. */ 1160 switch (len) { 1161 case (sizeof(struct oexport_args)): 1162 bzero(&o2export, sizeof(o2export)); 1163 /* FALLTHROUGH */ 1164 case (sizeof(o2export)): 1165 bcopy(bufp, &o2export, len); 1166 export.ex_flags = (uint64_t)o2export.ex_flags; 1167 export.ex_root = o2export.ex_root; 1168 export.ex_uid = o2export.ex_anon.cr_uid; 1169 export.ex_groups = NULL; 1170 export.ex_ngroups = o2export.ex_anon.cr_ngroups; 1171 if (export.ex_ngroups > 0) { 1172 if (export.ex_ngroups <= XU_NGROUPS) { 1173 export.ex_groups = malloc( 1174 export.ex_ngroups * sizeof(gid_t), 1175 M_TEMP, M_WAITOK); 1176 for (i = 0; i < export.ex_ngroups; i++) 1177 export.ex_groups[i] = 1178 o2export.ex_anon.cr_groups[i]; 1179 } else 1180 export_error = EINVAL; 1181 } else if (export.ex_ngroups < 0) 1182 export_error = EINVAL; 1183 export.ex_addr = o2export.ex_addr; 1184 export.ex_addrlen = o2export.ex_addrlen; 1185 export.ex_mask = o2export.ex_mask; 1186 export.ex_masklen = o2export.ex_masklen; 1187 export.ex_indexfile = o2export.ex_indexfile; 1188 export.ex_numsecflavors = o2export.ex_numsecflavors; 1189 if (export.ex_numsecflavors < MAXSECFLAVORS) { 1190 for (i = 0; i < export.ex_numsecflavors; i++) 1191 export.ex_secflavors[i] = 1192 o2export.ex_secflavors[i]; 1193 } else 1194 export_error = EINVAL; 1195 if (export_error == 0) 1196 export_error = vfs_export(mp, &export); 1197 free(export.ex_groups, M_TEMP); 1198 break; 1199 case (sizeof(export)): 1200 bcopy(bufp, &export, len); 1201 grps = NULL; 1202 if (export.ex_ngroups > 0) { 1203 if (export.ex_ngroups <= NGROUPS_MAX) { 1204 grps = malloc(export.ex_ngroups * 1205 sizeof(gid_t), M_TEMP, M_WAITOK); 1206 export_error = copyin(export.ex_groups, 1207 grps, export.ex_ngroups * 1208 sizeof(gid_t)); 1209 if (export_error == 0) 1210 export.ex_groups = grps; 1211 } else 1212 export_error = EINVAL; 1213 } else if (export.ex_ngroups == 0) 1214 export.ex_groups = NULL; 1215 else 1216 export_error = EINVAL; 1217 if (export_error == 0) 1218 export_error = vfs_export(mp, &export); 1219 free(grps, M_TEMP); 1220 break; 1221 default: 1222 export_error = EINVAL; 1223 break; 1224 } 1225 } 1226 1227 MNT_ILOCK(mp); 1228 if (error == 0) { 1229 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 1230 MNT_SNAPSHOT); 1231 } else { 1232 /* 1233 * If we fail, restore old mount flags. MNT_QUOTA is special, 1234 * because it is not part of MNT_UPDATEMASK, but it could have 1235 * changed in the meantime if quotactl(2) was called. 1236 * All in all we want current value of MNT_QUOTA, not the old 1237 * one. 1238 */ 1239 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 1240 } 1241 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1242 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1243 mp->mnt_kern_flag |= MNTK_ASYNC; 1244 else 1245 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1246 MNT_IUNLOCK(mp); 1247 1248 if (error != 0) 1249 goto end; 1250 1251 mount_devctl_event("REMOUNT", mp, true); 1252 if (mp->mnt_opt != NULL) 1253 vfs_freeopts(mp->mnt_opt); 1254 mp->mnt_opt = mp->mnt_optnew; 1255 *optlist = NULL; 1256 (void)VFS_STATFS(mp, &mp->mnt_stat); 1257 /* 1258 * Prevent external consumers of mount options from reading 1259 * mnt_optnew. 1260 */ 1261 mp->mnt_optnew = NULL; 1262 1263 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1264 vfs_allocate_syncvnode(mp); 1265 else 1266 vfs_deallocate_syncvnode(mp); 1267 end: 1268 vfs_op_exit(mp); 1269 if (rootvp != NULL) { 1270 vn_seqc_write_end(rootvp); 1271 vrele(rootvp); 1272 } 1273 vn_seqc_write_end(vp); 1274 vfs_unbusy(mp); 1275 VI_LOCK(vp); 1276 vp->v_iflag &= ~VI_MOUNT; 1277 VI_UNLOCK(vp); 1278 vrele(vp); 1279 return (error != 0 ? error : export_error); 1280 } 1281 1282 /* 1283 * vfs_domount(): actually attempt a filesystem mount. 1284 */ 1285 static int 1286 vfs_domount( 1287 struct thread *td, /* Calling thread. */ 1288 const char *fstype, /* Filesystem type. */ 1289 char *fspath, /* Mount path. */ 1290 uint64_t fsflags, /* Flags common to all filesystems. */ 1291 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1292 ) 1293 { 1294 struct vfsconf *vfsp; 1295 struct nameidata nd; 1296 struct vnode *vp; 1297 char *pathbuf; 1298 int error; 1299 1300 /* 1301 * Be ultra-paranoid about making sure the type and fspath 1302 * variables will fit in our mp buffers, including the 1303 * terminating NUL. 1304 */ 1305 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1306 return (ENAMETOOLONG); 1307 1308 if (jailed(td->td_ucred) || usermount == 0) { 1309 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1310 return (error); 1311 } 1312 1313 /* 1314 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1315 */ 1316 if (fsflags & MNT_EXPORTED) { 1317 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1318 if (error) 1319 return (error); 1320 } 1321 if (fsflags & MNT_SUIDDIR) { 1322 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1323 if (error) 1324 return (error); 1325 } 1326 /* 1327 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1328 */ 1329 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1330 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1331 fsflags |= MNT_NOSUID | MNT_USER; 1332 } 1333 1334 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1335 vfsp = NULL; 1336 if ((fsflags & MNT_UPDATE) == 0) { 1337 /* Don't try to load KLDs if we're mounting the root. */ 1338 if (fsflags & MNT_ROOTFS) 1339 vfsp = vfs_byname(fstype); 1340 else 1341 vfsp = vfs_byname_kld(fstype, td, &error); 1342 if (vfsp == NULL) 1343 return (ENODEV); 1344 } 1345 1346 /* 1347 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1348 */ 1349 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1350 UIO_SYSSPACE, fspath, td); 1351 error = namei(&nd); 1352 if (error != 0) 1353 return (error); 1354 NDFREE(&nd, NDF_ONLY_PNBUF); 1355 vp = nd.ni_vp; 1356 if ((fsflags & MNT_UPDATE) == 0) { 1357 if ((vp->v_vflag & VV_ROOT) != 0 && 1358 (fsflags & MNT_NOCOVER) != 0) { 1359 vput(vp); 1360 return (EBUSY); 1361 } 1362 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1363 strcpy(pathbuf, fspath); 1364 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN); 1365 if (error == 0) { 1366 error = vfs_domount_first(td, vfsp, pathbuf, vp, 1367 fsflags, optlist); 1368 } 1369 free(pathbuf, M_TEMP); 1370 } else 1371 error = vfs_domount_update(td, vp, fsflags, optlist); 1372 1373 return (error); 1374 } 1375 1376 /* 1377 * Unmount a filesystem. 1378 * 1379 * Note: unmount takes a path to the vnode mounted on as argument, not 1380 * special file (as before). 1381 */ 1382 #ifndef _SYS_SYSPROTO_H_ 1383 struct unmount_args { 1384 char *path; 1385 int flags; 1386 }; 1387 #endif 1388 /* ARGSUSED */ 1389 int 1390 sys_unmount(struct thread *td, struct unmount_args *uap) 1391 { 1392 1393 return (kern_unmount(td, uap->path, uap->flags)); 1394 } 1395 1396 int 1397 kern_unmount(struct thread *td, const char *path, int flags) 1398 { 1399 struct nameidata nd; 1400 struct mount *mp; 1401 char *pathbuf; 1402 int error, id0, id1; 1403 1404 AUDIT_ARG_VALUE(flags); 1405 if (jailed(td->td_ucred) || usermount == 0) { 1406 error = priv_check(td, PRIV_VFS_UNMOUNT); 1407 if (error) 1408 return (error); 1409 } 1410 1411 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1412 error = copyinstr(path, pathbuf, MNAMELEN, NULL); 1413 if (error) { 1414 free(pathbuf, M_TEMP); 1415 return (error); 1416 } 1417 if (flags & MNT_BYFSID) { 1418 AUDIT_ARG_TEXT(pathbuf); 1419 /* Decode the filesystem ID. */ 1420 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1421 free(pathbuf, M_TEMP); 1422 return (EINVAL); 1423 } 1424 1425 mtx_lock(&mountlist_mtx); 1426 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1427 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1428 mp->mnt_stat.f_fsid.val[1] == id1) { 1429 vfs_ref(mp); 1430 break; 1431 } 1432 } 1433 mtx_unlock(&mountlist_mtx); 1434 } else { 1435 /* 1436 * Try to find global path for path argument. 1437 */ 1438 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1439 UIO_SYSSPACE, pathbuf, td); 1440 if (namei(&nd) == 0) { 1441 NDFREE(&nd, NDF_ONLY_PNBUF); 1442 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf, 1443 MNAMELEN); 1444 if (error == 0) 1445 vput(nd.ni_vp); 1446 } 1447 mtx_lock(&mountlist_mtx); 1448 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1449 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) { 1450 vfs_ref(mp); 1451 break; 1452 } 1453 } 1454 mtx_unlock(&mountlist_mtx); 1455 } 1456 free(pathbuf, M_TEMP); 1457 if (mp == NULL) { 1458 /* 1459 * Previously we returned ENOENT for a nonexistent path and 1460 * EINVAL for a non-mountpoint. We cannot tell these apart 1461 * now, so in the !MNT_BYFSID case return the more likely 1462 * EINVAL for compatibility. 1463 */ 1464 return ((flags & MNT_BYFSID) ? ENOENT : EINVAL); 1465 } 1466 1467 /* 1468 * Don't allow unmounting the root filesystem. 1469 */ 1470 if (mp->mnt_flag & MNT_ROOTFS) { 1471 vfs_rel(mp); 1472 return (EINVAL); 1473 } 1474 error = dounmount(mp, flags, td); 1475 return (error); 1476 } 1477 1478 /* 1479 * Return error if any of the vnodes, ignoring the root vnode 1480 * and the syncer vnode, have non-zero usecount. 1481 * 1482 * This function is purely advisory - it can return false positives 1483 * and negatives. 1484 */ 1485 static int 1486 vfs_check_usecounts(struct mount *mp) 1487 { 1488 struct vnode *vp, *mvp; 1489 1490 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1491 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON && 1492 vp->v_usecount != 0) { 1493 VI_UNLOCK(vp); 1494 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1495 return (EBUSY); 1496 } 1497 VI_UNLOCK(vp); 1498 } 1499 1500 return (0); 1501 } 1502 1503 static void 1504 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags) 1505 { 1506 1507 mtx_assert(MNT_MTX(mp), MA_OWNED); 1508 mp->mnt_kern_flag &= ~mntkflags; 1509 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) { 1510 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1511 wakeup(mp); 1512 } 1513 vfs_op_exit_locked(mp); 1514 MNT_IUNLOCK(mp); 1515 if (coveredvp != NULL) { 1516 VOP_UNLOCK(coveredvp); 1517 vdrop(coveredvp); 1518 } 1519 vn_finished_write(mp); 1520 } 1521 1522 /* 1523 * There are various reference counters associated with the mount point. 1524 * Normally it is permitted to modify them without taking the mnt ilock, 1525 * but this behavior can be temporarily disabled if stable value is needed 1526 * or callers are expected to block (e.g. to not allow new users during 1527 * forced unmount). 1528 */ 1529 void 1530 vfs_op_enter(struct mount *mp) 1531 { 1532 struct mount_pcpu *mpcpu; 1533 int cpu; 1534 1535 MNT_ILOCK(mp); 1536 mp->mnt_vfs_ops++; 1537 if (mp->mnt_vfs_ops > 1) { 1538 MNT_IUNLOCK(mp); 1539 return; 1540 } 1541 vfs_op_barrier_wait(mp); 1542 CPU_FOREACH(cpu) { 1543 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1544 1545 mp->mnt_ref += mpcpu->mntp_ref; 1546 mpcpu->mntp_ref = 0; 1547 1548 mp->mnt_lockref += mpcpu->mntp_lockref; 1549 mpcpu->mntp_lockref = 0; 1550 1551 mp->mnt_writeopcount += mpcpu->mntp_writeopcount; 1552 mpcpu->mntp_writeopcount = 0; 1553 } 1554 if (mp->mnt_ref <= 0 || mp->mnt_lockref < 0 || mp->mnt_writeopcount < 0) 1555 panic("%s: invalid count(s) on mp %p: ref %d lockref %d writeopcount %d\n", 1556 __func__, mp, mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount); 1557 MNT_IUNLOCK(mp); 1558 vfs_assert_mount_counters(mp); 1559 } 1560 1561 void 1562 vfs_op_exit_locked(struct mount *mp) 1563 { 1564 1565 mtx_assert(MNT_MTX(mp), MA_OWNED); 1566 1567 if (mp->mnt_vfs_ops <= 0) 1568 panic("%s: invalid vfs_ops count %d for mp %p\n", 1569 __func__, mp->mnt_vfs_ops, mp); 1570 mp->mnt_vfs_ops--; 1571 } 1572 1573 void 1574 vfs_op_exit(struct mount *mp) 1575 { 1576 1577 MNT_ILOCK(mp); 1578 vfs_op_exit_locked(mp); 1579 MNT_IUNLOCK(mp); 1580 } 1581 1582 struct vfs_op_barrier_ipi { 1583 struct mount *mp; 1584 struct smp_rendezvous_cpus_retry_arg srcra; 1585 }; 1586 1587 static void 1588 vfs_op_action_func(void *arg) 1589 { 1590 struct vfs_op_barrier_ipi *vfsopipi; 1591 struct mount *mp; 1592 1593 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1594 mp = vfsopipi->mp; 1595 1596 if (!vfs_op_thread_entered(mp)) 1597 smp_rendezvous_cpus_done(arg); 1598 } 1599 1600 static void 1601 vfs_op_wait_func(void *arg, int cpu) 1602 { 1603 struct vfs_op_barrier_ipi *vfsopipi; 1604 struct mount *mp; 1605 struct mount_pcpu *mpcpu; 1606 1607 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1608 mp = vfsopipi->mp; 1609 1610 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1611 while (atomic_load_int(&mpcpu->mntp_thread_in_ops)) 1612 cpu_spinwait(); 1613 } 1614 1615 void 1616 vfs_op_barrier_wait(struct mount *mp) 1617 { 1618 struct vfs_op_barrier_ipi vfsopipi; 1619 1620 vfsopipi.mp = mp; 1621 1622 smp_rendezvous_cpus_retry(all_cpus, 1623 smp_no_rendezvous_barrier, 1624 vfs_op_action_func, 1625 smp_no_rendezvous_barrier, 1626 vfs_op_wait_func, 1627 &vfsopipi.srcra); 1628 } 1629 1630 #ifdef DIAGNOSTIC 1631 void 1632 vfs_assert_mount_counters(struct mount *mp) 1633 { 1634 struct mount_pcpu *mpcpu; 1635 int cpu; 1636 1637 if (mp->mnt_vfs_ops == 0) 1638 return; 1639 1640 CPU_FOREACH(cpu) { 1641 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1642 if (mpcpu->mntp_ref != 0 || 1643 mpcpu->mntp_lockref != 0 || 1644 mpcpu->mntp_writeopcount != 0) 1645 vfs_dump_mount_counters(mp); 1646 } 1647 } 1648 1649 void 1650 vfs_dump_mount_counters(struct mount *mp) 1651 { 1652 struct mount_pcpu *mpcpu; 1653 int ref, lockref, writeopcount; 1654 int cpu; 1655 1656 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops); 1657 1658 printf(" ref : "); 1659 ref = mp->mnt_ref; 1660 CPU_FOREACH(cpu) { 1661 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1662 printf("%d ", mpcpu->mntp_ref); 1663 ref += mpcpu->mntp_ref; 1664 } 1665 printf("\n"); 1666 printf(" lockref : "); 1667 lockref = mp->mnt_lockref; 1668 CPU_FOREACH(cpu) { 1669 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1670 printf("%d ", mpcpu->mntp_lockref); 1671 lockref += mpcpu->mntp_lockref; 1672 } 1673 printf("\n"); 1674 printf("writeopcount: "); 1675 writeopcount = mp->mnt_writeopcount; 1676 CPU_FOREACH(cpu) { 1677 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1678 printf("%d ", mpcpu->mntp_writeopcount); 1679 writeopcount += mpcpu->mntp_writeopcount; 1680 } 1681 printf("\n"); 1682 1683 printf("counter struct total\n"); 1684 printf("ref %-5d %-5d\n", mp->mnt_ref, ref); 1685 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref); 1686 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount); 1687 1688 panic("invalid counts on struct mount"); 1689 } 1690 #endif 1691 1692 int 1693 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which) 1694 { 1695 struct mount_pcpu *mpcpu; 1696 int cpu, sum; 1697 1698 switch (which) { 1699 case MNT_COUNT_REF: 1700 sum = mp->mnt_ref; 1701 break; 1702 case MNT_COUNT_LOCKREF: 1703 sum = mp->mnt_lockref; 1704 break; 1705 case MNT_COUNT_WRITEOPCOUNT: 1706 sum = mp->mnt_writeopcount; 1707 break; 1708 } 1709 1710 CPU_FOREACH(cpu) { 1711 mpcpu = vfs_mount_pcpu_remote(mp, cpu); 1712 switch (which) { 1713 case MNT_COUNT_REF: 1714 sum += mpcpu->mntp_ref; 1715 break; 1716 case MNT_COUNT_LOCKREF: 1717 sum += mpcpu->mntp_lockref; 1718 break; 1719 case MNT_COUNT_WRITEOPCOUNT: 1720 sum += mpcpu->mntp_writeopcount; 1721 break; 1722 } 1723 } 1724 return (sum); 1725 } 1726 1727 /* 1728 * Do the actual filesystem unmount. 1729 */ 1730 int 1731 dounmount(struct mount *mp, int flags, struct thread *td) 1732 { 1733 struct vnode *coveredvp, *rootvp; 1734 int error; 1735 uint64_t async_flag; 1736 int mnt_gen_r; 1737 1738 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1739 mnt_gen_r = mp->mnt_gen; 1740 VI_LOCK(coveredvp); 1741 vholdl(coveredvp); 1742 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1743 /* 1744 * Check for mp being unmounted while waiting for the 1745 * covered vnode lock. 1746 */ 1747 if (coveredvp->v_mountedhere != mp || 1748 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1749 VOP_UNLOCK(coveredvp); 1750 vdrop(coveredvp); 1751 vfs_rel(mp); 1752 return (EBUSY); 1753 } 1754 } 1755 1756 /* 1757 * Only privileged root, or (if MNT_USER is set) the user that did the 1758 * original mount is permitted to unmount this filesystem. 1759 */ 1760 error = vfs_suser(mp, td); 1761 if (error != 0) { 1762 if (coveredvp != NULL) { 1763 VOP_UNLOCK(coveredvp); 1764 vdrop(coveredvp); 1765 } 1766 vfs_rel(mp); 1767 return (error); 1768 } 1769 1770 vfs_op_enter(mp); 1771 1772 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF); 1773 MNT_ILOCK(mp); 1774 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 || 1775 (mp->mnt_flag & MNT_UPDATE) != 0 || 1776 !TAILQ_EMPTY(&mp->mnt_uppers)) { 1777 dounmount_cleanup(mp, coveredvp, 0); 1778 return (EBUSY); 1779 } 1780 mp->mnt_kern_flag |= MNTK_UNMOUNT; 1781 rootvp = vfs_cache_root_clear(mp); 1782 if (coveredvp != NULL) 1783 vn_seqc_write_begin(coveredvp); 1784 if (flags & MNT_NONBUSY) { 1785 MNT_IUNLOCK(mp); 1786 error = vfs_check_usecounts(mp); 1787 MNT_ILOCK(mp); 1788 if (error != 0) { 1789 vn_seqc_write_end(coveredvp); 1790 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT); 1791 if (rootvp != NULL) { 1792 vn_seqc_write_end(rootvp); 1793 vrele(rootvp); 1794 } 1795 return (error); 1796 } 1797 } 1798 /* Allow filesystems to detect that a forced unmount is in progress. */ 1799 if (flags & MNT_FORCE) { 1800 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1801 MNT_IUNLOCK(mp); 1802 /* 1803 * Must be done after setting MNTK_UNMOUNTF and before 1804 * waiting for mnt_lockref to become 0. 1805 */ 1806 VFS_PURGE(mp); 1807 MNT_ILOCK(mp); 1808 } 1809 error = 0; 1810 if (mp->mnt_lockref) { 1811 mp->mnt_kern_flag |= MNTK_DRAINING; 1812 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1813 "mount drain", 0); 1814 } 1815 MNT_IUNLOCK(mp); 1816 KASSERT(mp->mnt_lockref == 0, 1817 ("%s: invalid lock refcount in the drain path @ %s:%d", 1818 __func__, __FILE__, __LINE__)); 1819 KASSERT(error == 0, 1820 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1821 __func__, __FILE__, __LINE__)); 1822 1823 /* 1824 * We want to keep the vnode around so that we can vn_seqc_write_end 1825 * after we are done with unmount. Downgrade our reference to a mere 1826 * hold count so that we don't interefere with anything. 1827 */ 1828 if (rootvp != NULL) { 1829 vhold(rootvp); 1830 vrele(rootvp); 1831 } 1832 1833 if (mp->mnt_flag & MNT_EXPUBLIC) 1834 vfs_setpublicfs(NULL, NULL, NULL); 1835 1836 vfs_periodic(mp, MNT_WAIT); 1837 MNT_ILOCK(mp); 1838 async_flag = mp->mnt_flag & MNT_ASYNC; 1839 mp->mnt_flag &= ~MNT_ASYNC; 1840 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1841 MNT_IUNLOCK(mp); 1842 vfs_deallocate_syncvnode(mp); 1843 error = VFS_UNMOUNT(mp, flags); 1844 vn_finished_write(mp); 1845 /* 1846 * If we failed to flush the dirty blocks for this mount point, 1847 * undo all the cdir/rdir and rootvnode changes we made above. 1848 * Unless we failed to do so because the device is reporting that 1849 * it doesn't exist anymore. 1850 */ 1851 if (error && error != ENXIO) { 1852 MNT_ILOCK(mp); 1853 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1854 MNT_IUNLOCK(mp); 1855 vfs_allocate_syncvnode(mp); 1856 MNT_ILOCK(mp); 1857 } 1858 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1859 mp->mnt_flag |= async_flag; 1860 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1861 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1862 mp->mnt_kern_flag |= MNTK_ASYNC; 1863 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1864 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1865 wakeup(mp); 1866 } 1867 vfs_op_exit_locked(mp); 1868 MNT_IUNLOCK(mp); 1869 if (coveredvp) { 1870 vn_seqc_write_end(coveredvp); 1871 VOP_UNLOCK(coveredvp); 1872 vdrop(coveredvp); 1873 } 1874 if (rootvp != NULL) { 1875 vn_seqc_write_end(rootvp); 1876 vdrop(rootvp); 1877 } 1878 return (error); 1879 } 1880 mtx_lock(&mountlist_mtx); 1881 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1882 mtx_unlock(&mountlist_mtx); 1883 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td); 1884 if (coveredvp != NULL) { 1885 VI_LOCK(coveredvp); 1886 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT); 1887 coveredvp->v_mountedhere = NULL; 1888 vn_seqc_write_end_locked(coveredvp); 1889 VI_UNLOCK(coveredvp); 1890 VOP_UNLOCK(coveredvp); 1891 vdrop(coveredvp); 1892 } 1893 mount_devctl_event("UNMOUNT", mp, false); 1894 if (rootvp != NULL) { 1895 vn_seqc_write_end(rootvp); 1896 vdrop(rootvp); 1897 } 1898 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1899 if (rootvnode != NULL && mp == rootvnode->v_mount) { 1900 vrele(rootvnode); 1901 rootvnode = NULL; 1902 } 1903 if (mp == rootdevmp) 1904 rootdevmp = NULL; 1905 vfs_mount_destroy(mp); 1906 return (0); 1907 } 1908 1909 /* 1910 * Report errors during filesystem mounting. 1911 */ 1912 void 1913 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1914 { 1915 struct vfsoptlist *moptlist = mp->mnt_optnew; 1916 va_list ap; 1917 int error, len; 1918 char *errmsg; 1919 1920 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1921 if (error || errmsg == NULL || len <= 0) 1922 return; 1923 1924 va_start(ap, fmt); 1925 vsnprintf(errmsg, (size_t)len, fmt, ap); 1926 va_end(ap); 1927 } 1928 1929 void 1930 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1931 { 1932 va_list ap; 1933 int error, len; 1934 char *errmsg; 1935 1936 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1937 if (error || errmsg == NULL || len <= 0) 1938 return; 1939 1940 va_start(ap, fmt); 1941 vsnprintf(errmsg, (size_t)len, fmt, ap); 1942 va_end(ap); 1943 } 1944 1945 /* 1946 * --------------------------------------------------------------------- 1947 * Functions for querying mount options/arguments from filesystems. 1948 */ 1949 1950 /* 1951 * Check that no unknown options are given 1952 */ 1953 int 1954 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1955 { 1956 struct vfsopt *opt; 1957 char errmsg[255]; 1958 const char **t, *p, *q; 1959 int ret = 0; 1960 1961 TAILQ_FOREACH(opt, opts, link) { 1962 p = opt->name; 1963 q = NULL; 1964 if (p[0] == 'n' && p[1] == 'o') 1965 q = p + 2; 1966 for(t = global_opts; *t != NULL; t++) { 1967 if (strcmp(*t, p) == 0) 1968 break; 1969 if (q != NULL) { 1970 if (strcmp(*t, q) == 0) 1971 break; 1972 } 1973 } 1974 if (*t != NULL) 1975 continue; 1976 for(t = legal; *t != NULL; t++) { 1977 if (strcmp(*t, p) == 0) 1978 break; 1979 if (q != NULL) { 1980 if (strcmp(*t, q) == 0) 1981 break; 1982 } 1983 } 1984 if (*t != NULL) 1985 continue; 1986 snprintf(errmsg, sizeof(errmsg), 1987 "mount option <%s> is unknown", p); 1988 ret = EINVAL; 1989 } 1990 if (ret != 0) { 1991 TAILQ_FOREACH(opt, opts, link) { 1992 if (strcmp(opt->name, "errmsg") == 0) { 1993 strncpy((char *)opt->value, errmsg, opt->len); 1994 break; 1995 } 1996 } 1997 if (opt == NULL) 1998 printf("%s\n", errmsg); 1999 } 2000 return (ret); 2001 } 2002 2003 /* 2004 * Get a mount option by its name. 2005 * 2006 * Return 0 if the option was found, ENOENT otherwise. 2007 * If len is non-NULL it will be filled with the length 2008 * of the option. If buf is non-NULL, it will be filled 2009 * with the address of the option. 2010 */ 2011 int 2012 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len) 2013 { 2014 struct vfsopt *opt; 2015 2016 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2017 2018 TAILQ_FOREACH(opt, opts, link) { 2019 if (strcmp(name, opt->name) == 0) { 2020 opt->seen = 1; 2021 if (len != NULL) 2022 *len = opt->len; 2023 if (buf != NULL) 2024 *buf = opt->value; 2025 return (0); 2026 } 2027 } 2028 return (ENOENT); 2029 } 2030 2031 int 2032 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2033 { 2034 struct vfsopt *opt; 2035 2036 if (opts == NULL) 2037 return (-1); 2038 2039 TAILQ_FOREACH(opt, opts, link) { 2040 if (strcmp(name, opt->name) == 0) { 2041 opt->seen = 1; 2042 return (opt->pos); 2043 } 2044 } 2045 return (-1); 2046 } 2047 2048 int 2049 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value) 2050 { 2051 char *opt_value, *vtp; 2052 quad_t iv; 2053 int error, opt_len; 2054 2055 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len); 2056 if (error != 0) 2057 return (error); 2058 if (opt_len == 0 || opt_value == NULL) 2059 return (EINVAL); 2060 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0') 2061 return (EINVAL); 2062 iv = strtoq(opt_value, &vtp, 0); 2063 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0')) 2064 return (EINVAL); 2065 if (iv < 0) 2066 return (EINVAL); 2067 switch (vtp[0]) { 2068 case 't': case 'T': 2069 iv *= 1024; 2070 /* FALLTHROUGH */ 2071 case 'g': case 'G': 2072 iv *= 1024; 2073 /* FALLTHROUGH */ 2074 case 'm': case 'M': 2075 iv *= 1024; 2076 /* FALLTHROUGH */ 2077 case 'k': case 'K': 2078 iv *= 1024; 2079 case '\0': 2080 break; 2081 default: 2082 return (EINVAL); 2083 } 2084 *value = iv; 2085 2086 return (0); 2087 } 2088 2089 char * 2090 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2091 { 2092 struct vfsopt *opt; 2093 2094 *error = 0; 2095 TAILQ_FOREACH(opt, opts, link) { 2096 if (strcmp(name, opt->name) != 0) 2097 continue; 2098 opt->seen = 1; 2099 if (opt->len == 0 || 2100 ((char *)opt->value)[opt->len - 1] != '\0') { 2101 *error = EINVAL; 2102 return (NULL); 2103 } 2104 return (opt->value); 2105 } 2106 *error = ENOENT; 2107 return (NULL); 2108 } 2109 2110 int 2111 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, 2112 uint64_t val) 2113 { 2114 struct vfsopt *opt; 2115 2116 TAILQ_FOREACH(opt, opts, link) { 2117 if (strcmp(name, opt->name) == 0) { 2118 opt->seen = 1; 2119 if (w != NULL) 2120 *w |= val; 2121 return (1); 2122 } 2123 } 2124 if (w != NULL) 2125 *w &= ~val; 2126 return (0); 2127 } 2128 2129 int 2130 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2131 { 2132 va_list ap; 2133 struct vfsopt *opt; 2134 int ret; 2135 2136 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2137 2138 TAILQ_FOREACH(opt, opts, link) { 2139 if (strcmp(name, opt->name) != 0) 2140 continue; 2141 opt->seen = 1; 2142 if (opt->len == 0 || opt->value == NULL) 2143 return (0); 2144 if (((char *)opt->value)[opt->len - 1] != '\0') 2145 return (0); 2146 va_start(ap, fmt); 2147 ret = vsscanf(opt->value, fmt, ap); 2148 va_end(ap); 2149 return (ret); 2150 } 2151 return (0); 2152 } 2153 2154 int 2155 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2156 { 2157 struct vfsopt *opt; 2158 2159 TAILQ_FOREACH(opt, opts, link) { 2160 if (strcmp(name, opt->name) != 0) 2161 continue; 2162 opt->seen = 1; 2163 if (opt->value == NULL) 2164 opt->len = len; 2165 else { 2166 if (opt->len != len) 2167 return (EINVAL); 2168 bcopy(value, opt->value, len); 2169 } 2170 return (0); 2171 } 2172 return (ENOENT); 2173 } 2174 2175 int 2176 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2177 { 2178 struct vfsopt *opt; 2179 2180 TAILQ_FOREACH(opt, opts, link) { 2181 if (strcmp(name, opt->name) != 0) 2182 continue; 2183 opt->seen = 1; 2184 if (opt->value == NULL) 2185 opt->len = len; 2186 else { 2187 if (opt->len < len) 2188 return (EINVAL); 2189 opt->len = len; 2190 bcopy(value, opt->value, len); 2191 } 2192 return (0); 2193 } 2194 return (ENOENT); 2195 } 2196 2197 int 2198 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2199 { 2200 struct vfsopt *opt; 2201 2202 TAILQ_FOREACH(opt, opts, link) { 2203 if (strcmp(name, opt->name) != 0) 2204 continue; 2205 opt->seen = 1; 2206 if (opt->value == NULL) 2207 opt->len = strlen(value) + 1; 2208 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2209 return (EINVAL); 2210 return (0); 2211 } 2212 return (ENOENT); 2213 } 2214 2215 /* 2216 * Find and copy a mount option. 2217 * 2218 * The size of the buffer has to be specified 2219 * in len, if it is not the same length as the 2220 * mount option, EINVAL is returned. 2221 * Returns ENOENT if the option is not found. 2222 */ 2223 int 2224 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len) 2225 { 2226 struct vfsopt *opt; 2227 2228 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2229 2230 TAILQ_FOREACH(opt, opts, link) { 2231 if (strcmp(name, opt->name) == 0) { 2232 opt->seen = 1; 2233 if (len != opt->len) 2234 return (EINVAL); 2235 bcopy(opt->value, dest, opt->len); 2236 return (0); 2237 } 2238 } 2239 return (ENOENT); 2240 } 2241 2242 int 2243 __vfs_statfs(struct mount *mp, struct statfs *sbp) 2244 { 2245 2246 /* 2247 * Filesystems only fill in part of the structure for updates, we 2248 * have to read the entirety first to get all content. 2249 */ 2250 if (sbp != &mp->mnt_stat) 2251 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp)); 2252 2253 /* 2254 * Set these in case the underlying filesystem fails to do so. 2255 */ 2256 sbp->f_version = STATFS_VERSION; 2257 sbp->f_namemax = NAME_MAX; 2258 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 2259 2260 return (mp->mnt_op->vfs_statfs(mp, sbp)); 2261 } 2262 2263 void 2264 vfs_mountedfrom(struct mount *mp, const char *from) 2265 { 2266 2267 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2268 strlcpy(mp->mnt_stat.f_mntfromname, from, 2269 sizeof mp->mnt_stat.f_mntfromname); 2270 } 2271 2272 /* 2273 * --------------------------------------------------------------------- 2274 * This is the api for building mount args and mounting filesystems from 2275 * inside the kernel. 2276 * 2277 * The API works by accumulation of individual args. First error is 2278 * latched. 2279 * 2280 * XXX: should be documented in new manpage kernel_mount(9) 2281 */ 2282 2283 /* A memory allocation which must be freed when we are done */ 2284 struct mntaarg { 2285 SLIST_ENTRY(mntaarg) next; 2286 }; 2287 2288 /* The header for the mount arguments */ 2289 struct mntarg { 2290 struct iovec *v; 2291 int len; 2292 int error; 2293 SLIST_HEAD(, mntaarg) list; 2294 }; 2295 2296 /* 2297 * Add a boolean argument. 2298 * 2299 * flag is the boolean value. 2300 * name must start with "no". 2301 */ 2302 struct mntarg * 2303 mount_argb(struct mntarg *ma, int flag, const char *name) 2304 { 2305 2306 KASSERT(name[0] == 'n' && name[1] == 'o', 2307 ("mount_argb(...,%s): name must start with 'no'", name)); 2308 2309 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2310 } 2311 2312 /* 2313 * Add an argument printf style 2314 */ 2315 struct mntarg * 2316 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2317 { 2318 va_list ap; 2319 struct mntaarg *maa; 2320 struct sbuf *sb; 2321 int len; 2322 2323 if (ma == NULL) { 2324 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2325 SLIST_INIT(&ma->list); 2326 } 2327 if (ma->error) 2328 return (ma); 2329 2330 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2331 M_MOUNT, M_WAITOK); 2332 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2333 ma->v[ma->len].iov_len = strlen(name) + 1; 2334 ma->len++; 2335 2336 sb = sbuf_new_auto(); 2337 va_start(ap, fmt); 2338 sbuf_vprintf(sb, fmt, ap); 2339 va_end(ap); 2340 sbuf_finish(sb); 2341 len = sbuf_len(sb) + 1; 2342 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2343 SLIST_INSERT_HEAD(&ma->list, maa, next); 2344 bcopy(sbuf_data(sb), maa + 1, len); 2345 sbuf_delete(sb); 2346 2347 ma->v[ma->len].iov_base = maa + 1; 2348 ma->v[ma->len].iov_len = len; 2349 ma->len++; 2350 2351 return (ma); 2352 } 2353 2354 /* 2355 * Add an argument which is a userland string. 2356 */ 2357 struct mntarg * 2358 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2359 { 2360 struct mntaarg *maa; 2361 char *tbuf; 2362 2363 if (val == NULL) 2364 return (ma); 2365 if (ma == NULL) { 2366 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2367 SLIST_INIT(&ma->list); 2368 } 2369 if (ma->error) 2370 return (ma); 2371 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2372 SLIST_INSERT_HEAD(&ma->list, maa, next); 2373 tbuf = (void *)(maa + 1); 2374 ma->error = copyinstr(val, tbuf, len, NULL); 2375 return (mount_arg(ma, name, tbuf, -1)); 2376 } 2377 2378 /* 2379 * Plain argument. 2380 * 2381 * If length is -1, treat value as a C string. 2382 */ 2383 struct mntarg * 2384 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2385 { 2386 2387 if (ma == NULL) { 2388 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2389 SLIST_INIT(&ma->list); 2390 } 2391 if (ma->error) 2392 return (ma); 2393 2394 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2395 M_MOUNT, M_WAITOK); 2396 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2397 ma->v[ma->len].iov_len = strlen(name) + 1; 2398 ma->len++; 2399 2400 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2401 if (len < 0) 2402 ma->v[ma->len].iov_len = strlen(val) + 1; 2403 else 2404 ma->v[ma->len].iov_len = len; 2405 ma->len++; 2406 return (ma); 2407 } 2408 2409 /* 2410 * Free a mntarg structure 2411 */ 2412 static void 2413 free_mntarg(struct mntarg *ma) 2414 { 2415 struct mntaarg *maa; 2416 2417 while (!SLIST_EMPTY(&ma->list)) { 2418 maa = SLIST_FIRST(&ma->list); 2419 SLIST_REMOVE_HEAD(&ma->list, next); 2420 free(maa, M_MOUNT); 2421 } 2422 free(ma->v, M_MOUNT); 2423 free(ma, M_MOUNT); 2424 } 2425 2426 /* 2427 * Mount a filesystem 2428 */ 2429 int 2430 kernel_mount(struct mntarg *ma, uint64_t flags) 2431 { 2432 struct uio auio; 2433 int error; 2434 2435 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2436 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2437 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2438 2439 auio.uio_iov = ma->v; 2440 auio.uio_iovcnt = ma->len; 2441 auio.uio_segflg = UIO_SYSSPACE; 2442 2443 error = ma->error; 2444 if (!error) 2445 error = vfs_donmount(curthread, flags, &auio); 2446 free_mntarg(ma); 2447 return (error); 2448 } 2449 2450 /* 2451 * A printflike function to mount a filesystem. 2452 */ 2453 int 2454 kernel_vmount(int flags, ...) 2455 { 2456 struct mntarg *ma = NULL; 2457 va_list ap; 2458 const char *cp; 2459 const void *vp; 2460 int error; 2461 2462 va_start(ap, flags); 2463 for (;;) { 2464 cp = va_arg(ap, const char *); 2465 if (cp == NULL) 2466 break; 2467 vp = va_arg(ap, const void *); 2468 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2469 } 2470 va_end(ap); 2471 2472 error = kernel_mount(ma, flags); 2473 return (error); 2474 } 2475 2476 /* Map from mount options to printable formats. */ 2477 static struct mntoptnames optnames[] = { 2478 MNTOPT_NAMES 2479 }; 2480 2481 static void 2482 mount_devctl_event_mntopt(struct sbuf *sb, const char *what, struct vfsoptlist *opts) 2483 { 2484 struct vfsopt *opt; 2485 2486 if (opts == NULL || TAILQ_EMPTY(opts)) 2487 return; 2488 sbuf_printf(sb, " %s=\"", what); 2489 TAILQ_FOREACH(opt, opts, link) { 2490 if (opt->name[0] == '\0' || (opt->len > 0 && *(char *)opt->value == '\0')) 2491 continue; 2492 devctl_safe_quote_sb(sb, opt->name); 2493 if (opt->len > 0) { 2494 sbuf_putc(sb, '='); 2495 devctl_safe_quote_sb(sb, opt->value); 2496 } 2497 sbuf_putc(sb, ';'); 2498 } 2499 sbuf_putc(sb, '"'); 2500 } 2501 2502 #define DEVCTL_LEN 1024 2503 static void 2504 mount_devctl_event(const char *type, struct mount *mp, bool donew) 2505 { 2506 const uint8_t *cp; 2507 struct mntoptnames *fp; 2508 struct sbuf sb; 2509 struct statfs *sfp = &mp->mnt_stat; 2510 char *buf; 2511 2512 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT); 2513 if (buf == NULL) 2514 return; 2515 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN); 2516 sbuf_cpy(&sb, "mount-point=\""); 2517 devctl_safe_quote_sb(&sb, sfp->f_mntonname); 2518 sbuf_cat(&sb, "\" mount-dev=\""); 2519 devctl_safe_quote_sb(&sb, sfp->f_mntfromname); 2520 sbuf_cat(&sb, "\" mount-type=\""); 2521 devctl_safe_quote_sb(&sb, sfp->f_fstypename); 2522 sbuf_cat(&sb, "\" fsid=0x"); 2523 cp = (const uint8_t *)&sfp->f_fsid.val[0]; 2524 for (int i = 0; i < sizeof(sfp->f_fsid); i++) 2525 sbuf_printf(&sb, "%02x", cp[i]); 2526 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner); 2527 for (fp = optnames; fp->o_opt != 0; fp++) { 2528 if ((mp->mnt_flag & fp->o_opt) != 0) { 2529 sbuf_cat(&sb, fp->o_name); 2530 sbuf_putc(&sb, ';'); 2531 } 2532 } 2533 sbuf_putc(&sb, '"'); 2534 mount_devctl_event_mntopt(&sb, "opt", mp->mnt_opt); 2535 if (donew) 2536 mount_devctl_event_mntopt(&sb, "optnew", mp->mnt_optnew); 2537 sbuf_finish(&sb); 2538 2539 if (sbuf_error(&sb) == 0) 2540 devctl_notify("VFS", "FS", type, sbuf_data(&sb)); 2541 sbuf_delete(&sb); 2542 free(buf, M_MOUNT); 2543 } 2544 2545 /* 2546 * Suspend write operations on all local writeable filesystems. Does 2547 * full sync of them in the process. 2548 * 2549 * Iterate over the mount points in reverse order, suspending most 2550 * recently mounted filesystems first. It handles a case where a 2551 * filesystem mounted from a md(4) vnode-backed device should be 2552 * suspended before the filesystem that owns the vnode. 2553 */ 2554 void 2555 suspend_all_fs(void) 2556 { 2557 struct mount *mp; 2558 int error; 2559 2560 mtx_lock(&mountlist_mtx); 2561 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 2562 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT); 2563 if (error != 0) 2564 continue; 2565 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL || 2566 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { 2567 mtx_lock(&mountlist_mtx); 2568 vfs_unbusy(mp); 2569 continue; 2570 } 2571 error = vfs_write_suspend(mp, 0); 2572 if (error == 0) { 2573 MNT_ILOCK(mp); 2574 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0); 2575 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL; 2576 MNT_IUNLOCK(mp); 2577 mtx_lock(&mountlist_mtx); 2578 } else { 2579 printf("suspend of %s failed, error %d\n", 2580 mp->mnt_stat.f_mntonname, error); 2581 mtx_lock(&mountlist_mtx); 2582 vfs_unbusy(mp); 2583 } 2584 } 2585 mtx_unlock(&mountlist_mtx); 2586 } 2587 2588 void 2589 resume_all_fs(void) 2590 { 2591 struct mount *mp; 2592 2593 mtx_lock(&mountlist_mtx); 2594 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2595 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0) 2596 continue; 2597 mtx_unlock(&mountlist_mtx); 2598 MNT_ILOCK(mp); 2599 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0); 2600 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL; 2601 MNT_IUNLOCK(mp); 2602 vfs_write_resume(mp, 0); 2603 mtx_lock(&mountlist_mtx); 2604 vfs_unbusy(mp); 2605 } 2606 mtx_unlock(&mountlist_mtx); 2607 }
Cache object: 172d89a6b2d3558b282b41423bcb4fa1
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]