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