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/eventhandler.h> 46 #include <sys/fcntl.h> 47 #include <sys/jail.h> 48 #include <sys/kernel.h> 49 #include <sys/ktr.h> 50 #include <sys/libkern.h> 51 #include <sys/malloc.h> 52 #include <sys/mount.h> 53 #include <sys/mutex.h> 54 #include <sys/namei.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/filedesc.h> 58 #include <sys/reboot.h> 59 #include <sys/sbuf.h> 60 #include <sys/syscallsubr.h> 61 #include <sys/sysproto.h> 62 #include <sys/sx.h> 63 #include <sys/sysctl.h> 64 #include <sys/sysent.h> 65 #include <sys/systm.h> 66 #include <sys/vnode.h> 67 #include <vm/uma.h> 68 69 #include <geom/geom.h> 70 71 #include <machine/stdarg.h> 72 73 #include <security/audit/audit.h> 74 #include <security/mac/mac_framework.h> 75 76 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 77 78 static int vfs_domount(struct thread *td, const char *fstype, char *fspath, 79 uint64_t fsflags, struct vfsoptlist **optlist); 80 static void free_mntarg(struct mntarg *ma); 81 82 static int usermount = 0; 83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 84 "Unprivileged users may mount and unmount file systems"); 85 86 static bool default_autoro = false; 87 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0, 88 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified"); 89 90 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 91 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure"); 92 static uma_zone_t mount_zone; 93 94 /* List of mounted filesystems. */ 95 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 96 97 /* For any iteration/modification of mountlist */ 98 struct mtx mountlist_mtx; 99 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 100 101 EVENTHANDLER_LIST_DEFINE(vfs_mounted); 102 EVENTHANDLER_LIST_DEFINE(vfs_unmounted); 103 104 /* 105 * Global opts, taken by all filesystems 106 */ 107 static const char *global_opts[] = { 108 "errmsg", 109 "fstype", 110 "fspath", 111 "ro", 112 "rw", 113 "nosuid", 114 "noexec", 115 NULL 116 }; 117 118 static int 119 mount_init(void *mem, int size, int flags) 120 { 121 struct mount *mp; 122 123 mp = (struct mount *)mem; 124 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 125 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF); 126 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 127 mp->mnt_thread_in_ops_pcpu = uma_zalloc_pcpu(pcpu_zone_int, 128 M_WAITOK | M_ZERO); 129 mp->mnt_ref_pcpu = uma_zalloc_pcpu(pcpu_zone_int, 130 M_WAITOK | M_ZERO); 131 mp->mnt_lockref_pcpu = uma_zalloc_pcpu(pcpu_zone_int, 132 M_WAITOK | M_ZERO); 133 mp->mnt_writeopcount_pcpu = uma_zalloc_pcpu(pcpu_zone_int, 134 M_WAITOK | M_ZERO); 135 mp->mnt_ref = 0; 136 mp->mnt_vfs_ops = 1; 137 mp->mnt_rootvnode = NULL; 138 return (0); 139 } 140 141 static void 142 mount_fini(void *mem, int size) 143 { 144 struct mount *mp; 145 146 mp = (struct mount *)mem; 147 uma_zfree_pcpu(pcpu_zone_int, mp->mnt_writeopcount_pcpu); 148 uma_zfree_pcpu(pcpu_zone_int, mp->mnt_lockref_pcpu); 149 uma_zfree_pcpu(pcpu_zone_int, mp->mnt_ref_pcpu); 150 uma_zfree_pcpu(pcpu_zone_int, mp->mnt_thread_in_ops_pcpu); 151 lockdestroy(&mp->mnt_explock); 152 mtx_destroy(&mp->mnt_listmtx); 153 mtx_destroy(&mp->mnt_mtx); 154 } 155 156 static void 157 vfs_mount_init(void *dummy __unused) 158 { 159 160 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL, 161 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE); 162 } 163 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL); 164 165 /* 166 * --------------------------------------------------------------------- 167 * Functions for building and sanitizing the mount options 168 */ 169 170 /* Remove one mount option. */ 171 static void 172 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 173 { 174 175 TAILQ_REMOVE(opts, opt, link); 176 free(opt->name, M_MOUNT); 177 if (opt->value != NULL) 178 free(opt->value, M_MOUNT); 179 free(opt, M_MOUNT); 180 } 181 182 /* Release all resources related to the mount options. */ 183 void 184 vfs_freeopts(struct vfsoptlist *opts) 185 { 186 struct vfsopt *opt; 187 188 while (!TAILQ_EMPTY(opts)) { 189 opt = TAILQ_FIRST(opts); 190 vfs_freeopt(opts, opt); 191 } 192 free(opts, M_MOUNT); 193 } 194 195 void 196 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 197 { 198 struct vfsopt *opt, *temp; 199 200 if (opts == NULL) 201 return; 202 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 203 if (strcmp(opt->name, name) == 0) 204 vfs_freeopt(opts, opt); 205 } 206 } 207 208 static int 209 vfs_isopt_ro(const char *opt) 210 { 211 212 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 || 213 strcmp(opt, "norw") == 0) 214 return (1); 215 return (0); 216 } 217 218 static int 219 vfs_isopt_rw(const char *opt) 220 { 221 222 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0) 223 return (1); 224 return (0); 225 } 226 227 /* 228 * Check if options are equal (with or without the "no" prefix). 229 */ 230 static int 231 vfs_equalopts(const char *opt1, const char *opt2) 232 { 233 char *p; 234 235 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 236 if (strcmp(opt1, opt2) == 0) 237 return (1); 238 /* "noopt" vs. "opt" */ 239 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 240 return (1); 241 /* "opt" vs. "noopt" */ 242 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 243 return (1); 244 while ((p = strchr(opt1, '.')) != NULL && 245 !strncmp(opt1, opt2, ++p - opt1)) { 246 opt2 += p - opt1; 247 opt1 = p; 248 /* "foo.noopt" vs. "foo.opt" */ 249 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 250 return (1); 251 /* "foo.opt" vs. "foo.noopt" */ 252 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 253 return (1); 254 } 255 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */ 256 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) && 257 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2))) 258 return (1); 259 return (0); 260 } 261 262 /* 263 * If a mount option is specified several times, 264 * (with or without the "no" prefix) only keep 265 * the last occurrence of it. 266 */ 267 static void 268 vfs_sanitizeopts(struct vfsoptlist *opts) 269 { 270 struct vfsopt *opt, *opt2, *tmp; 271 272 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 273 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 274 while (opt2 != NULL) { 275 if (vfs_equalopts(opt->name, opt2->name)) { 276 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 277 vfs_freeopt(opts, opt2); 278 opt2 = tmp; 279 } else { 280 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 281 } 282 } 283 } 284 } 285 286 /* 287 * Build a linked list of mount options from a struct uio. 288 */ 289 int 290 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 291 { 292 struct vfsoptlist *opts; 293 struct vfsopt *opt; 294 size_t memused, namelen, optlen; 295 unsigned int i, iovcnt; 296 int error; 297 298 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 299 TAILQ_INIT(opts); 300 memused = 0; 301 iovcnt = auio->uio_iovcnt; 302 for (i = 0; i < iovcnt; i += 2) { 303 namelen = auio->uio_iov[i].iov_len; 304 optlen = auio->uio_iov[i + 1].iov_len; 305 memused += sizeof(struct vfsopt) + optlen + namelen; 306 /* 307 * Avoid consuming too much memory, and attempts to overflow 308 * memused. 309 */ 310 if (memused > VFS_MOUNTARG_SIZE_MAX || 311 optlen > VFS_MOUNTARG_SIZE_MAX || 312 namelen > VFS_MOUNTARG_SIZE_MAX) { 313 error = EINVAL; 314 goto bad; 315 } 316 317 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 318 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 319 opt->value = NULL; 320 opt->len = 0; 321 opt->pos = i / 2; 322 opt->seen = 0; 323 324 /* 325 * Do this early, so jumps to "bad" will free the current 326 * option. 327 */ 328 TAILQ_INSERT_TAIL(opts, opt, link); 329 330 if (auio->uio_segflg == UIO_SYSSPACE) { 331 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 332 } else { 333 error = copyin(auio->uio_iov[i].iov_base, opt->name, 334 namelen); 335 if (error) 336 goto bad; 337 } 338 /* Ensure names are null-terminated strings. */ 339 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 340 error = EINVAL; 341 goto bad; 342 } 343 if (optlen != 0) { 344 opt->len = optlen; 345 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 346 if (auio->uio_segflg == UIO_SYSSPACE) { 347 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 348 optlen); 349 } else { 350 error = copyin(auio->uio_iov[i + 1].iov_base, 351 opt->value, optlen); 352 if (error) 353 goto bad; 354 } 355 } 356 } 357 vfs_sanitizeopts(opts); 358 *options = opts; 359 return (0); 360 bad: 361 vfs_freeopts(opts); 362 return (error); 363 } 364 365 /* 366 * Merge the old mount options with the new ones passed 367 * in the MNT_UPDATE case. 368 * 369 * XXX: This function will keep a "nofoo" option in the new 370 * options. E.g, if the option's canonical name is "foo", 371 * "nofoo" ends up in the mount point's active options. 372 */ 373 static void 374 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts) 375 { 376 struct vfsopt *opt, *new; 377 378 TAILQ_FOREACH(opt, oldopts, link) { 379 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 380 new->name = strdup(opt->name, M_MOUNT); 381 if (opt->len != 0) { 382 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 383 bcopy(opt->value, new->value, opt->len); 384 } else 385 new->value = NULL; 386 new->len = opt->len; 387 new->seen = opt->seen; 388 TAILQ_INSERT_HEAD(toopts, new, link); 389 } 390 vfs_sanitizeopts(toopts); 391 } 392 393 /* 394 * Mount a filesystem. 395 */ 396 #ifndef _SYS_SYSPROTO_H_ 397 struct nmount_args { 398 struct iovec *iovp; 399 unsigned int iovcnt; 400 int flags; 401 }; 402 #endif 403 int 404 sys_nmount(struct thread *td, struct nmount_args *uap) 405 { 406 struct uio *auio; 407 int error; 408 u_int iovcnt; 409 uint64_t flags; 410 411 /* 412 * Mount flags are now 64-bits. On 32-bit archtectures only 413 * 32-bits are passed in, but from here on everything handles 414 * 64-bit flags correctly. 415 */ 416 flags = uap->flags; 417 418 AUDIT_ARG_FFLAGS(flags); 419 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 420 uap->iovp, uap->iovcnt, flags); 421 422 /* 423 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 424 * userspace to set this flag, but we must filter it out if we want 425 * MNT_UPDATE on the root file system to work. 426 * MNT_ROOTFS should only be set by the kernel when mounting its 427 * root file system. 428 */ 429 flags &= ~MNT_ROOTFS; 430 431 iovcnt = uap->iovcnt; 432 /* 433 * Check that we have an even number of iovec's 434 * and that we have at least two options. 435 */ 436 if ((iovcnt & 1) || (iovcnt < 4)) { 437 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 438 uap->iovcnt); 439 return (EINVAL); 440 } 441 442 error = copyinuio(uap->iovp, iovcnt, &auio); 443 if (error) { 444 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 445 __func__, error); 446 return (error); 447 } 448 error = vfs_donmount(td, flags, auio); 449 450 free(auio, M_IOV); 451 return (error); 452 } 453 454 /* 455 * --------------------------------------------------------------------- 456 * Various utility functions 457 */ 458 459 void 460 vfs_ref(struct mount *mp) 461 { 462 463 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 464 if (vfs_op_thread_enter(mp)) { 465 vfs_mp_count_add_pcpu(mp, ref, 1); 466 vfs_op_thread_exit(mp); 467 return; 468 } 469 470 MNT_ILOCK(mp); 471 MNT_REF(mp); 472 MNT_IUNLOCK(mp); 473 } 474 475 void 476 vfs_rel(struct mount *mp) 477 { 478 479 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 480 if (vfs_op_thread_enter(mp)) { 481 vfs_mp_count_sub_pcpu(mp, ref, 1); 482 vfs_op_thread_exit(mp); 483 return; 484 } 485 486 MNT_ILOCK(mp); 487 MNT_REL(mp); 488 MNT_IUNLOCK(mp); 489 } 490 491 /* 492 * Allocate and initialize the mount point struct. 493 */ 494 struct mount * 495 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 496 struct ucred *cred) 497 { 498 struct mount *mp; 499 500 mp = uma_zalloc(mount_zone, M_WAITOK); 501 bzero(&mp->mnt_startzero, 502 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 503 TAILQ_INIT(&mp->mnt_nvnodelist); 504 mp->mnt_nvnodelistsize = 0; 505 TAILQ_INIT(&mp->mnt_lazyvnodelist); 506 mp->mnt_lazyvnodelistsize = 0; 507 if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 || 508 mp->mnt_writeopcount != 0) 509 panic("%s: non-zero counters on new mp %p\n", __func__, mp); 510 if (mp->mnt_vfs_ops != 1) 511 panic("%s: vfs_ops should be 1 but %d found\n", __func__, 512 mp->mnt_vfs_ops); 513 (void) vfs_busy(mp, MBF_NOWAIT); 514 atomic_add_acq_int(&vfsp->vfc_refcount, 1); 515 mp->mnt_op = vfsp->vfc_vfsops; 516 mp->mnt_vfc = vfsp; 517 mp->mnt_stat.f_type = vfsp->vfc_typenum; 518 mp->mnt_gen++; 519 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 520 mp->mnt_vnodecovered = vp; 521 mp->mnt_cred = crdup(cred); 522 mp->mnt_stat.f_owner = cred->cr_uid; 523 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 524 mp->mnt_iosize_max = DFLTPHYS; 525 #ifdef MAC 526 mac_mount_init(mp); 527 mac_mount_create(cred, mp); 528 #endif 529 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 530 TAILQ_INIT(&mp->mnt_uppers); 531 return (mp); 532 } 533 534 /* 535 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 536 */ 537 void 538 vfs_mount_destroy(struct mount *mp) 539 { 540 541 if (mp->mnt_vfs_ops == 0) 542 panic("%s: entered with zero vfs_ops\n", __func__); 543 544 vfs_assert_mount_counters(mp); 545 546 MNT_ILOCK(mp); 547 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 548 if (mp->mnt_kern_flag & MNTK_MWAIT) { 549 mp->mnt_kern_flag &= ~MNTK_MWAIT; 550 wakeup(mp); 551 } 552 while (mp->mnt_ref) 553 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 554 KASSERT(mp->mnt_ref == 0, 555 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 556 __FILE__, __LINE__)); 557 if (mp->mnt_writeopcount != 0) 558 panic("vfs_mount_destroy: nonzero writeopcount"); 559 if (mp->mnt_secondary_writes != 0) 560 panic("vfs_mount_destroy: nonzero secondary_writes"); 561 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1); 562 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 563 struct vnode *vp; 564 565 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 566 vn_printf(vp, "dangling vnode "); 567 panic("unmount: dangling vnode"); 568 } 569 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers")); 570 if (mp->mnt_nvnodelistsize != 0) 571 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 572 if (mp->mnt_lazyvnodelistsize != 0) 573 panic("vfs_mount_destroy: nonzero lazyvnodelistsize"); 574 if (mp->mnt_lockref != 0) 575 panic("vfs_mount_destroy: nonzero lock refcount"); 576 MNT_IUNLOCK(mp); 577 578 if (mp->mnt_vfs_ops != 1) 579 panic("%s: vfs_ops should be 1 but %d found\n", __func__, 580 mp->mnt_vfs_ops); 581 582 if (mp->mnt_rootvnode != NULL) 583 panic("%s: mount point still has a root vnode %p\n", __func__, 584 mp->mnt_rootvnode); 585 586 if (mp->mnt_vnodecovered != NULL) 587 vrele(mp->mnt_vnodecovered); 588 #ifdef MAC 589 mac_mount_destroy(mp); 590 #endif 591 if (mp->mnt_opt != NULL) 592 vfs_freeopts(mp->mnt_opt); 593 crfree(mp->mnt_cred); 594 uma_zfree(mount_zone, mp); 595 } 596 597 static bool 598 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error) 599 { 600 /* This is an upgrade of an exisiting mount. */ 601 if ((fsflags & MNT_UPDATE) != 0) 602 return (false); 603 /* This is already an R/O mount. */ 604 if ((fsflags & MNT_RDONLY) != 0) 605 return (false); 606 607 switch (error) { 608 case ENODEV: /* generic, geom, ... */ 609 case EACCES: /* cam/scsi, ... */ 610 case EROFS: /* md, mmcsd, ... */ 611 /* 612 * These errors can be returned by the storage layer to signal 613 * that the media is read-only. No harm in the R/O mount 614 * attempt if the error was returned for some other reason. 615 */ 616 return (true); 617 default: 618 return (false); 619 } 620 } 621 622 int 623 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions) 624 { 625 struct vfsoptlist *optlist; 626 struct vfsopt *opt, *tmp_opt; 627 char *fstype, *fspath, *errmsg; 628 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 629 bool autoro; 630 631 errmsg = fspath = NULL; 632 errmsg_len = fspathlen = 0; 633 errmsg_pos = -1; 634 autoro = default_autoro; 635 636 error = vfs_buildopts(fsoptions, &optlist); 637 if (error) 638 return (error); 639 640 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 641 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 642 643 /* 644 * We need these two options before the others, 645 * and they are mandatory for any filesystem. 646 * Ensure they are NUL terminated as well. 647 */ 648 fstypelen = 0; 649 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 650 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') { 651 error = EINVAL; 652 if (errmsg != NULL) 653 strncpy(errmsg, "Invalid fstype", errmsg_len); 654 goto bail; 655 } 656 fspathlen = 0; 657 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 658 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') { 659 error = EINVAL; 660 if (errmsg != NULL) 661 strncpy(errmsg, "Invalid fspath", errmsg_len); 662 goto bail; 663 } 664 665 /* 666 * We need to see if we have the "update" option 667 * before we call vfs_domount(), since vfs_domount() has special 668 * logic based on MNT_UPDATE. This is very important 669 * when we want to update the root filesystem. 670 */ 671 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 672 int do_freeopt = 0; 673 674 if (strcmp(opt->name, "update") == 0) { 675 fsflags |= MNT_UPDATE; 676 do_freeopt = 1; 677 } 678 else if (strcmp(opt->name, "async") == 0) 679 fsflags |= MNT_ASYNC; 680 else if (strcmp(opt->name, "force") == 0) { 681 fsflags |= MNT_FORCE; 682 do_freeopt = 1; 683 } 684 else if (strcmp(opt->name, "reload") == 0) { 685 fsflags |= MNT_RELOAD; 686 do_freeopt = 1; 687 } 688 else if (strcmp(opt->name, "multilabel") == 0) 689 fsflags |= MNT_MULTILABEL; 690 else if (strcmp(opt->name, "noasync") == 0) 691 fsflags &= ~MNT_ASYNC; 692 else if (strcmp(opt->name, "noatime") == 0) 693 fsflags |= MNT_NOATIME; 694 else if (strcmp(opt->name, "atime") == 0) { 695 free(opt->name, M_MOUNT); 696 opt->name = strdup("nonoatime", M_MOUNT); 697 } 698 else if (strcmp(opt->name, "noclusterr") == 0) 699 fsflags |= MNT_NOCLUSTERR; 700 else if (strcmp(opt->name, "clusterr") == 0) { 701 free(opt->name, M_MOUNT); 702 opt->name = strdup("nonoclusterr", M_MOUNT); 703 } 704 else if (strcmp(opt->name, "noclusterw") == 0) 705 fsflags |= MNT_NOCLUSTERW; 706 else if (strcmp(opt->name, "clusterw") == 0) { 707 free(opt->name, M_MOUNT); 708 opt->name = strdup("nonoclusterw", M_MOUNT); 709 } 710 else if (strcmp(opt->name, "noexec") == 0) 711 fsflags |= MNT_NOEXEC; 712 else if (strcmp(opt->name, "exec") == 0) { 713 free(opt->name, M_MOUNT); 714 opt->name = strdup("nonoexec", M_MOUNT); 715 } 716 else if (strcmp(opt->name, "nosuid") == 0) 717 fsflags |= MNT_NOSUID; 718 else if (strcmp(opt->name, "suid") == 0) { 719 free(opt->name, M_MOUNT); 720 opt->name = strdup("nonosuid", M_MOUNT); 721 } 722 else if (strcmp(opt->name, "nosymfollow") == 0) 723 fsflags |= MNT_NOSYMFOLLOW; 724 else if (strcmp(opt->name, "symfollow") == 0) { 725 free(opt->name, M_MOUNT); 726 opt->name = strdup("nonosymfollow", M_MOUNT); 727 } 728 else if (strcmp(opt->name, "noro") == 0) { 729 fsflags &= ~MNT_RDONLY; 730 autoro = false; 731 } 732 else if (strcmp(opt->name, "rw") == 0) { 733 fsflags &= ~MNT_RDONLY; 734 autoro = false; 735 } 736 else if (strcmp(opt->name, "ro") == 0) { 737 fsflags |= MNT_RDONLY; 738 autoro = false; 739 } 740 else if (strcmp(opt->name, "rdonly") == 0) { 741 free(opt->name, M_MOUNT); 742 opt->name = strdup("ro", M_MOUNT); 743 fsflags |= MNT_RDONLY; 744 autoro = false; 745 } 746 else if (strcmp(opt->name, "autoro") == 0) { 747 do_freeopt = 1; 748 autoro = true; 749 } 750 else if (strcmp(opt->name, "suiddir") == 0) 751 fsflags |= MNT_SUIDDIR; 752 else if (strcmp(opt->name, "sync") == 0) 753 fsflags |= MNT_SYNCHRONOUS; 754 else if (strcmp(opt->name, "union") == 0) 755 fsflags |= MNT_UNION; 756 else if (strcmp(opt->name, "automounted") == 0) { 757 fsflags |= MNT_AUTOMOUNTED; 758 do_freeopt = 1; 759 } else if (strcmp(opt->name, "nocover") == 0) { 760 fsflags |= MNT_NOCOVER; 761 do_freeopt = 1; 762 } else if (strcmp(opt->name, "cover") == 0) { 763 fsflags &= ~MNT_NOCOVER; 764 do_freeopt = 1; 765 } else if (strcmp(opt->name, "emptydir") == 0) { 766 fsflags |= MNT_EMPTYDIR; 767 do_freeopt = 1; 768 } else if (strcmp(opt->name, "noemptydir") == 0) { 769 fsflags &= ~MNT_EMPTYDIR; 770 do_freeopt = 1; 771 } 772 if (do_freeopt) 773 vfs_freeopt(optlist, opt); 774 } 775 776 /* 777 * Be ultra-paranoid about making sure the type and fspath 778 * variables will fit in our mp buffers, including the 779 * terminating NUL. 780 */ 781 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) { 782 error = ENAMETOOLONG; 783 goto bail; 784 } 785 786 error = vfs_domount(td, fstype, fspath, fsflags, &optlist); 787 788 /* 789 * See if we can mount in the read-only mode if the error code suggests 790 * that it could be possible and the mount options allow for that. 791 * Never try it if "[no]{ro|rw}" has been explicitly requested and not 792 * overridden by "autoro". 793 */ 794 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) { 795 printf("%s: R/W mount failed, possibly R/O media," 796 " trying R/O mount\n", __func__); 797 fsflags |= MNT_RDONLY; 798 error = vfs_domount(td, fstype, fspath, fsflags, &optlist); 799 } 800 bail: 801 /* copyout the errmsg */ 802 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 803 && errmsg_len > 0 && errmsg != NULL) { 804 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 805 bcopy(errmsg, 806 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 807 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 808 } else { 809 copyout(errmsg, 810 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 811 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 812 } 813 } 814 815 if (optlist != NULL) 816 vfs_freeopts(optlist); 817 return (error); 818 } 819 820 /* 821 * Old mount API. 822 */ 823 #ifndef _SYS_SYSPROTO_H_ 824 struct mount_args { 825 char *type; 826 char *path; 827 int flags; 828 caddr_t data; 829 }; 830 #endif 831 /* ARGSUSED */ 832 int 833 sys_mount(struct thread *td, struct mount_args *uap) 834 { 835 char *fstype; 836 struct vfsconf *vfsp = NULL; 837 struct mntarg *ma = NULL; 838 uint64_t flags; 839 int error; 840 841 /* 842 * Mount flags are now 64-bits. On 32-bit architectures only 843 * 32-bits are passed in, but from here on everything handles 844 * 64-bit flags correctly. 845 */ 846 flags = uap->flags; 847 848 AUDIT_ARG_FFLAGS(flags); 849 850 /* 851 * Filter out MNT_ROOTFS. We do not want clients of mount() in 852 * userspace to set this flag, but we must filter it out if we want 853 * MNT_UPDATE on the root file system to work. 854 * MNT_ROOTFS should only be set by the kernel when mounting its 855 * root file system. 856 */ 857 flags &= ~MNT_ROOTFS; 858 859 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 860 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 861 if (error) { 862 free(fstype, M_TEMP); 863 return (error); 864 } 865 866 AUDIT_ARG_TEXT(fstype); 867 vfsp = vfs_byname_kld(fstype, td, &error); 868 free(fstype, M_TEMP); 869 if (vfsp == NULL) 870 return (ENOENT); 871 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 && 872 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) || 873 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 && 874 vfsp->vfc_vfsops->vfs_cmount == NULL)) 875 return (EOPNOTSUPP); 876 877 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN); 878 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 879 ma = mount_argb(ma, flags & MNT_RDONLY, "noro"); 880 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid"); 881 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec"); 882 883 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0) 884 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags)); 885 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags)); 886 } 887 888 /* 889 * vfs_domount_first(): first file system mount (not update) 890 */ 891 static int 892 vfs_domount_first( 893 struct thread *td, /* Calling thread. */ 894 struct vfsconf *vfsp, /* File system type. */ 895 char *fspath, /* Mount path. */ 896 struct vnode *vp, /* Vnode to be covered. */ 897 uint64_t fsflags, /* Flags common to all filesystems. */ 898 struct vfsoptlist **optlist /* Options local to the filesystem. */ 899 ) 900 { 901 struct vattr va; 902 struct mount *mp; 903 struct vnode *newdp, *rootvp; 904 int error, error1; 905 906 ASSERT_VOP_ELOCKED(vp, __func__); 907 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 908 909 if ((fsflags & MNT_EMPTYDIR) != 0) { 910 error = vfs_emptydir(vp); 911 if (error != 0) { 912 vput(vp); 913 return (error); 914 } 915 } 916 917 /* 918 * If the jail of the calling thread lacks permission for this type of 919 * file system, deny immediately. 920 */ 921 if (jailed(td->td_ucred) && !prison_allow(td->td_ucred, 922 vfsp->vfc_prison_flag)) { 923 vput(vp); 924 return (EPERM); 925 } 926 927 /* 928 * If the user is not root, ensure that they own the directory 929 * onto which we are attempting to mount. 930 */ 931 error = VOP_GETATTR(vp, &va, td->td_ucred); 932 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 933 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN); 934 if (error == 0) 935 error = vinvalbuf(vp, V_SAVE, 0, 0); 936 if (error == 0 && vp->v_type != VDIR) 937 error = ENOTDIR; 938 if (error == 0) { 939 VI_LOCK(vp); 940 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 941 vp->v_iflag |= VI_MOUNT; 942 else 943 error = EBUSY; 944 VI_UNLOCK(vp); 945 } 946 if (error != 0) { 947 vput(vp); 948 return (error); 949 } 950 VOP_UNLOCK(vp); 951 952 /* Allocate and initialize the filesystem. */ 953 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 954 /* XXXMAC: pass to vfs_mount_alloc? */ 955 mp->mnt_optnew = *optlist; 956 /* Set the mount level flags. */ 957 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 958 959 /* 960 * Mount the filesystem. 961 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 962 * get. No freeing of cn_pnbuf. 963 */ 964 error1 = 0; 965 if ((error = VFS_MOUNT(mp)) != 0 || 966 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 || 967 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) { 968 if (error1 != 0) { 969 error = error1; 970 rootvp = vfs_cache_root_clear(mp); 971 if (rootvp != NULL) 972 vrele(rootvp); 973 if ((error1 = VFS_UNMOUNT(mp, 0)) != 0) 974 printf("VFS_UNMOUNT returned %d\n", error1); 975 } 976 vfs_unbusy(mp); 977 mp->mnt_vnodecovered = NULL; 978 vfs_mount_destroy(mp); 979 VI_LOCK(vp); 980 vp->v_iflag &= ~VI_MOUNT; 981 VI_UNLOCK(vp); 982 vrele(vp); 983 return (error); 984 } 985 VOP_UNLOCK(newdp); 986 987 if (mp->mnt_opt != NULL) 988 vfs_freeopts(mp->mnt_opt); 989 mp->mnt_opt = mp->mnt_optnew; 990 *optlist = NULL; 991 992 /* 993 * Prevent external consumers of mount options from reading mnt_optnew. 994 */ 995 mp->mnt_optnew = NULL; 996 997 MNT_ILOCK(mp); 998 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 999 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1000 mp->mnt_kern_flag |= MNTK_ASYNC; 1001 else 1002 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1003 MNT_IUNLOCK(mp); 1004 1005 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1006 cache_purge(vp); 1007 VI_LOCK(vp); 1008 vp->v_iflag &= ~VI_MOUNT; 1009 VI_UNLOCK(vp); 1010 vp->v_mountedhere = mp; 1011 /* Place the new filesystem at the end of the mount list. */ 1012 mtx_lock(&mountlist_mtx); 1013 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1014 mtx_unlock(&mountlist_mtx); 1015 vfs_event_signal(NULL, VQ_MOUNT, 0); 1016 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY); 1017 VOP_UNLOCK(vp); 1018 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td); 1019 VOP_UNLOCK(newdp); 1020 mountcheckdirs(vp, newdp); 1021 vrele(newdp); 1022 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1023 vfs_allocate_syncvnode(mp); 1024 vfs_op_exit(mp); 1025 vfs_unbusy(mp); 1026 return (0); 1027 } 1028 1029 /* 1030 * vfs_domount_update(): update of mounted file system 1031 */ 1032 static int 1033 vfs_domount_update( 1034 struct thread *td, /* Calling thread. */ 1035 struct vnode *vp, /* Mount point vnode. */ 1036 uint64_t fsflags, /* Flags common to all filesystems. */ 1037 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1038 ) 1039 { 1040 struct export_args export; 1041 struct o2export_args o2export; 1042 struct vnode *rootvp; 1043 void *bufp; 1044 struct mount *mp; 1045 int error, export_error, i, len; 1046 uint64_t flag; 1047 gid_t *grps; 1048 1049 ASSERT_VOP_ELOCKED(vp, __func__); 1050 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 1051 mp = vp->v_mount; 1052 1053 if ((vp->v_vflag & VV_ROOT) == 0) { 1054 if (vfs_copyopt(*optlist, "export", &export, sizeof(export)) 1055 == 0) 1056 error = EXDEV; 1057 else 1058 error = EINVAL; 1059 vput(vp); 1060 return (error); 1061 } 1062 1063 /* 1064 * We only allow the filesystem to be reloaded if it 1065 * is currently mounted read-only. 1066 */ 1067 flag = mp->mnt_flag; 1068 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 1069 vput(vp); 1070 return (EOPNOTSUPP); /* Needs translation */ 1071 } 1072 /* 1073 * Only privileged root, or (if MNT_USER is set) the user that 1074 * did the original mount is permitted to update it. 1075 */ 1076 error = vfs_suser(mp, td); 1077 if (error != 0) { 1078 vput(vp); 1079 return (error); 1080 } 1081 if (vfs_busy(mp, MBF_NOWAIT)) { 1082 vput(vp); 1083 return (EBUSY); 1084 } 1085 VI_LOCK(vp); 1086 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 1087 VI_UNLOCK(vp); 1088 vfs_unbusy(mp); 1089 vput(vp); 1090 return (EBUSY); 1091 } 1092 vp->v_iflag |= VI_MOUNT; 1093 VI_UNLOCK(vp); 1094 VOP_UNLOCK(vp); 1095 1096 vfs_op_enter(mp); 1097 1098 MNT_ILOCK(mp); 1099 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 1100 MNT_IUNLOCK(mp); 1101 error = EBUSY; 1102 goto end; 1103 } 1104 mp->mnt_flag &= ~MNT_UPDATEMASK; 1105 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 1106 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 1107 if ((mp->mnt_flag & MNT_ASYNC) == 0) 1108 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1109 rootvp = vfs_cache_root_clear(mp); 1110 MNT_IUNLOCK(mp); 1111 if (rootvp != NULL) 1112 vrele(rootvp); 1113 mp->mnt_optnew = *optlist; 1114 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 1115 1116 /* 1117 * Mount the filesystem. 1118 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1119 * get. No freeing of cn_pnbuf. 1120 */ 1121 error = VFS_MOUNT(mp); 1122 1123 export_error = 0; 1124 /* Process the export option. */ 1125 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp, 1126 &len) == 0) { 1127 /* Assume that there is only 1 ABI for each length. */ 1128 switch (len) { 1129 case (sizeof(struct oexport_args)): 1130 bzero(&o2export, sizeof(o2export)); 1131 /* FALLTHROUGH */ 1132 case (sizeof(o2export)): 1133 bcopy(bufp, &o2export, len); 1134 export.ex_flags = (uint64_t)o2export.ex_flags; 1135 export.ex_root = o2export.ex_root; 1136 export.ex_uid = o2export.ex_anon.cr_uid; 1137 export.ex_groups = NULL; 1138 export.ex_ngroups = o2export.ex_anon.cr_ngroups; 1139 if (export.ex_ngroups > 0) { 1140 if (export.ex_ngroups <= XU_NGROUPS) { 1141 export.ex_groups = malloc( 1142 export.ex_ngroups * sizeof(gid_t), 1143 M_TEMP, M_WAITOK); 1144 for (i = 0; i < export.ex_ngroups; i++) 1145 export.ex_groups[i] = 1146 o2export.ex_anon.cr_groups[i]; 1147 } else 1148 export_error = EINVAL; 1149 } else if (export.ex_ngroups < 0) 1150 export_error = EINVAL; 1151 export.ex_addr = o2export.ex_addr; 1152 export.ex_addrlen = o2export.ex_addrlen; 1153 export.ex_mask = o2export.ex_mask; 1154 export.ex_masklen = o2export.ex_masklen; 1155 export.ex_indexfile = o2export.ex_indexfile; 1156 export.ex_numsecflavors = o2export.ex_numsecflavors; 1157 if (export.ex_numsecflavors < MAXSECFLAVORS) { 1158 for (i = 0; i < export.ex_numsecflavors; i++) 1159 export.ex_secflavors[i] = 1160 o2export.ex_secflavors[i]; 1161 } else 1162 export_error = EINVAL; 1163 if (export_error == 0) 1164 export_error = vfs_export(mp, &export); 1165 free(export.ex_groups, M_TEMP); 1166 break; 1167 case (sizeof(export)): 1168 bcopy(bufp, &export, len); 1169 grps = NULL; 1170 if (export.ex_ngroups > 0) { 1171 if (export.ex_ngroups <= NGROUPS_MAX) { 1172 grps = malloc(export.ex_ngroups * 1173 sizeof(gid_t), M_TEMP, M_WAITOK); 1174 export_error = copyin(export.ex_groups, 1175 grps, export.ex_ngroups * 1176 sizeof(gid_t)); 1177 if (export_error == 0) 1178 export.ex_groups = grps; 1179 } else 1180 export_error = EINVAL; 1181 } else if (export.ex_ngroups == 0) 1182 export.ex_groups = NULL; 1183 else 1184 export_error = EINVAL; 1185 if (export_error == 0) 1186 export_error = vfs_export(mp, &export); 1187 free(grps, M_TEMP); 1188 break; 1189 default: 1190 export_error = EINVAL; 1191 break; 1192 } 1193 } 1194 1195 MNT_ILOCK(mp); 1196 if (error == 0) { 1197 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 1198 MNT_SNAPSHOT); 1199 } else { 1200 /* 1201 * If we fail, restore old mount flags. MNT_QUOTA is special, 1202 * because it is not part of MNT_UPDATEMASK, but it could have 1203 * changed in the meantime if quotactl(2) was called. 1204 * All in all we want current value of MNT_QUOTA, not the old 1205 * one. 1206 */ 1207 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 1208 } 1209 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1210 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1211 mp->mnt_kern_flag |= MNTK_ASYNC; 1212 else 1213 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1214 MNT_IUNLOCK(mp); 1215 1216 if (error != 0) 1217 goto end; 1218 1219 if (mp->mnt_opt != NULL) 1220 vfs_freeopts(mp->mnt_opt); 1221 mp->mnt_opt = mp->mnt_optnew; 1222 *optlist = NULL; 1223 (void)VFS_STATFS(mp, &mp->mnt_stat); 1224 /* 1225 * Prevent external consumers of mount options from reading 1226 * mnt_optnew. 1227 */ 1228 mp->mnt_optnew = NULL; 1229 1230 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1231 vfs_allocate_syncvnode(mp); 1232 else 1233 vfs_deallocate_syncvnode(mp); 1234 end: 1235 vfs_op_exit(mp); 1236 vfs_unbusy(mp); 1237 VI_LOCK(vp); 1238 vp->v_iflag &= ~VI_MOUNT; 1239 VI_UNLOCK(vp); 1240 vrele(vp); 1241 return (error != 0 ? error : export_error); 1242 } 1243 1244 /* 1245 * vfs_domount(): actually attempt a filesystem mount. 1246 */ 1247 static int 1248 vfs_domount( 1249 struct thread *td, /* Calling thread. */ 1250 const char *fstype, /* Filesystem type. */ 1251 char *fspath, /* Mount path. */ 1252 uint64_t fsflags, /* Flags common to all filesystems. */ 1253 struct vfsoptlist **optlist /* Options local to the filesystem. */ 1254 ) 1255 { 1256 struct vfsconf *vfsp; 1257 struct nameidata nd; 1258 struct vnode *vp; 1259 char *pathbuf; 1260 int error; 1261 1262 /* 1263 * Be ultra-paranoid about making sure the type and fspath 1264 * variables will fit in our mp buffers, including the 1265 * terminating NUL. 1266 */ 1267 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1268 return (ENAMETOOLONG); 1269 1270 if (jailed(td->td_ucred) || usermount == 0) { 1271 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1272 return (error); 1273 } 1274 1275 /* 1276 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1277 */ 1278 if (fsflags & MNT_EXPORTED) { 1279 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1280 if (error) 1281 return (error); 1282 } 1283 if (fsflags & MNT_SUIDDIR) { 1284 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1285 if (error) 1286 return (error); 1287 } 1288 /* 1289 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1290 */ 1291 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1292 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1293 fsflags |= MNT_NOSUID | MNT_USER; 1294 } 1295 1296 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1297 vfsp = NULL; 1298 if ((fsflags & MNT_UPDATE) == 0) { 1299 /* Don't try to load KLDs if we're mounting the root. */ 1300 if (fsflags & MNT_ROOTFS) 1301 vfsp = vfs_byname(fstype); 1302 else 1303 vfsp = vfs_byname_kld(fstype, td, &error); 1304 if (vfsp == NULL) 1305 return (ENODEV); 1306 } 1307 1308 /* 1309 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1310 */ 1311 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1312 UIO_SYSSPACE, fspath, td); 1313 error = namei(&nd); 1314 if (error != 0) 1315 return (error); 1316 NDFREE(&nd, NDF_ONLY_PNBUF); 1317 vp = nd.ni_vp; 1318 if ((fsflags & MNT_UPDATE) == 0) { 1319 if ((vp->v_vflag & VV_ROOT) != 0 && 1320 (fsflags & MNT_NOCOVER) != 0) { 1321 vput(vp); 1322 return (EBUSY); 1323 } 1324 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1325 strcpy(pathbuf, fspath); 1326 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN); 1327 if (error == 0) { 1328 error = vfs_domount_first(td, vfsp, pathbuf, vp, 1329 fsflags, optlist); 1330 } 1331 free(pathbuf, M_TEMP); 1332 } else 1333 error = vfs_domount_update(td, vp, fsflags, optlist); 1334 1335 return (error); 1336 } 1337 1338 /* 1339 * Unmount a filesystem. 1340 * 1341 * Note: unmount takes a path to the vnode mounted on as argument, not 1342 * special file (as before). 1343 */ 1344 #ifndef _SYS_SYSPROTO_H_ 1345 struct unmount_args { 1346 char *path; 1347 int flags; 1348 }; 1349 #endif 1350 /* ARGSUSED */ 1351 int 1352 sys_unmount(struct thread *td, struct unmount_args *uap) 1353 { 1354 1355 return (kern_unmount(td, uap->path, uap->flags)); 1356 } 1357 1358 int 1359 kern_unmount(struct thread *td, const char *path, int flags) 1360 { 1361 struct nameidata nd; 1362 struct mount *mp; 1363 char *pathbuf; 1364 int error, id0, id1; 1365 1366 AUDIT_ARG_VALUE(flags); 1367 if (jailed(td->td_ucred) || usermount == 0) { 1368 error = priv_check(td, PRIV_VFS_UNMOUNT); 1369 if (error) 1370 return (error); 1371 } 1372 1373 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1374 error = copyinstr(path, pathbuf, MNAMELEN, NULL); 1375 if (error) { 1376 free(pathbuf, M_TEMP); 1377 return (error); 1378 } 1379 if (flags & MNT_BYFSID) { 1380 AUDIT_ARG_TEXT(pathbuf); 1381 /* Decode the filesystem ID. */ 1382 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1383 free(pathbuf, M_TEMP); 1384 return (EINVAL); 1385 } 1386 1387 mtx_lock(&mountlist_mtx); 1388 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1389 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1390 mp->mnt_stat.f_fsid.val[1] == id1) { 1391 vfs_ref(mp); 1392 break; 1393 } 1394 } 1395 mtx_unlock(&mountlist_mtx); 1396 } else { 1397 /* 1398 * Try to find global path for path argument. 1399 */ 1400 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, 1401 UIO_SYSSPACE, pathbuf, td); 1402 if (namei(&nd) == 0) { 1403 NDFREE(&nd, NDF_ONLY_PNBUF); 1404 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf, 1405 MNAMELEN); 1406 if (error == 0) 1407 vput(nd.ni_vp); 1408 } 1409 mtx_lock(&mountlist_mtx); 1410 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1411 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) { 1412 vfs_ref(mp); 1413 break; 1414 } 1415 } 1416 mtx_unlock(&mountlist_mtx); 1417 } 1418 free(pathbuf, M_TEMP); 1419 if (mp == NULL) { 1420 /* 1421 * Previously we returned ENOENT for a nonexistent path and 1422 * EINVAL for a non-mountpoint. We cannot tell these apart 1423 * now, so in the !MNT_BYFSID case return the more likely 1424 * EINVAL for compatibility. 1425 */ 1426 return ((flags & MNT_BYFSID) ? ENOENT : EINVAL); 1427 } 1428 1429 /* 1430 * Don't allow unmounting the root filesystem. 1431 */ 1432 if (mp->mnt_flag & MNT_ROOTFS) { 1433 vfs_rel(mp); 1434 return (EINVAL); 1435 } 1436 error = dounmount(mp, flags, td); 1437 return (error); 1438 } 1439 1440 /* 1441 * Return error if any of the vnodes, ignoring the root vnode 1442 * and the syncer vnode, have non-zero usecount. 1443 * 1444 * This function is purely advisory - it can return false positives 1445 * and negatives. 1446 */ 1447 static int 1448 vfs_check_usecounts(struct mount *mp) 1449 { 1450 struct vnode *vp, *mvp; 1451 1452 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 1453 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON && 1454 vp->v_usecount != 0) { 1455 VI_UNLOCK(vp); 1456 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 1457 return (EBUSY); 1458 } 1459 VI_UNLOCK(vp); 1460 } 1461 1462 return (0); 1463 } 1464 1465 static void 1466 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags) 1467 { 1468 1469 mtx_assert(MNT_MTX(mp), MA_OWNED); 1470 mp->mnt_kern_flag &= ~mntkflags; 1471 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) { 1472 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1473 wakeup(mp); 1474 } 1475 vfs_op_exit_locked(mp); 1476 MNT_IUNLOCK(mp); 1477 if (coveredvp != NULL) { 1478 VOP_UNLOCK(coveredvp); 1479 vdrop(coveredvp); 1480 } 1481 vn_finished_write(mp); 1482 } 1483 1484 /* 1485 * There are various reference counters associated with the mount point. 1486 * Normally it is permitted to modify them without taking the mnt ilock, 1487 * but this behavior can be temporarily disabled if stable value is needed 1488 * or callers are expected to block (e.g. to not allow new users during 1489 * forced unmount). 1490 */ 1491 void 1492 vfs_op_enter(struct mount *mp) 1493 { 1494 int cpu; 1495 1496 MNT_ILOCK(mp); 1497 mp->mnt_vfs_ops++; 1498 if (mp->mnt_vfs_ops > 1) { 1499 MNT_IUNLOCK(mp); 1500 return; 1501 } 1502 vfs_op_barrier_wait(mp); 1503 CPU_FOREACH(cpu) { 1504 mp->mnt_ref += 1505 zpcpu_replace_cpu(mp->mnt_ref_pcpu, 0, cpu); 1506 mp->mnt_lockref += 1507 zpcpu_replace_cpu(mp->mnt_lockref_pcpu, 0, cpu); 1508 mp->mnt_writeopcount += 1509 zpcpu_replace_cpu(mp->mnt_writeopcount_pcpu, 0, cpu); 1510 } 1511 MNT_IUNLOCK(mp); 1512 vfs_assert_mount_counters(mp); 1513 } 1514 1515 void 1516 vfs_op_exit_locked(struct mount *mp) 1517 { 1518 1519 mtx_assert(MNT_MTX(mp), MA_OWNED); 1520 1521 if (mp->mnt_vfs_ops <= 0) 1522 panic("%s: invalid vfs_ops count %d for mp %p\n", 1523 __func__, mp->mnt_vfs_ops, mp); 1524 mp->mnt_vfs_ops--; 1525 } 1526 1527 void 1528 vfs_op_exit(struct mount *mp) 1529 { 1530 1531 MNT_ILOCK(mp); 1532 vfs_op_exit_locked(mp); 1533 MNT_IUNLOCK(mp); 1534 } 1535 1536 struct vfs_op_barrier_ipi { 1537 struct mount *mp; 1538 struct smp_rendezvous_cpus_retry_arg srcra; 1539 }; 1540 1541 static void 1542 vfs_op_action_func(void *arg) 1543 { 1544 struct vfs_op_barrier_ipi *vfsopipi; 1545 struct mount *mp; 1546 1547 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1548 mp = vfsopipi->mp; 1549 1550 if (!vfs_op_thread_entered(mp)) 1551 smp_rendezvous_cpus_done(arg); 1552 } 1553 1554 static void 1555 vfs_op_wait_func(void *arg, int cpu) 1556 { 1557 struct vfs_op_barrier_ipi *vfsopipi; 1558 struct mount *mp; 1559 int *in_op; 1560 1561 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra); 1562 mp = vfsopipi->mp; 1563 1564 in_op = zpcpu_get_cpu(mp->mnt_thread_in_ops_pcpu, cpu); 1565 while (atomic_load_int(in_op)) 1566 cpu_spinwait(); 1567 } 1568 1569 void 1570 vfs_op_barrier_wait(struct mount *mp) 1571 { 1572 struct vfs_op_barrier_ipi vfsopipi; 1573 1574 vfsopipi.mp = mp; 1575 1576 smp_rendezvous_cpus_retry(all_cpus, 1577 smp_no_rendezvous_barrier, 1578 vfs_op_action_func, 1579 smp_no_rendezvous_barrier, 1580 vfs_op_wait_func, 1581 &vfsopipi.srcra); 1582 } 1583 1584 #ifdef DIAGNOSTIC 1585 void 1586 vfs_assert_mount_counters(struct mount *mp) 1587 { 1588 int cpu; 1589 1590 if (mp->mnt_vfs_ops == 0) 1591 return; 1592 1593 CPU_FOREACH(cpu) { 1594 if (*zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu) != 0 || 1595 *zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu) != 0 || 1596 *zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu) != 0) 1597 vfs_dump_mount_counters(mp); 1598 } 1599 } 1600 1601 void 1602 vfs_dump_mount_counters(struct mount *mp) 1603 { 1604 int cpu, *count; 1605 int ref, lockref, writeopcount; 1606 1607 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops); 1608 1609 printf(" ref : "); 1610 ref = mp->mnt_ref; 1611 CPU_FOREACH(cpu) { 1612 count = zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu); 1613 printf("%d ", *count); 1614 ref += *count; 1615 } 1616 printf("\n"); 1617 printf(" lockref : "); 1618 lockref = mp->mnt_lockref; 1619 CPU_FOREACH(cpu) { 1620 count = zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu); 1621 printf("%d ", *count); 1622 lockref += *count; 1623 } 1624 printf("\n"); 1625 printf("writeopcount: "); 1626 writeopcount = mp->mnt_writeopcount; 1627 CPU_FOREACH(cpu) { 1628 count = zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu); 1629 printf("%d ", *count); 1630 writeopcount += *count; 1631 } 1632 printf("\n"); 1633 1634 printf("counter struct total\n"); 1635 printf("ref %-5d %-5d\n", mp->mnt_ref, ref); 1636 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref); 1637 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount); 1638 1639 panic("invalid counts on struct mount"); 1640 } 1641 #endif 1642 1643 int 1644 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which) 1645 { 1646 int *base, *pcpu; 1647 int cpu, sum; 1648 1649 switch (which) { 1650 case MNT_COUNT_REF: 1651 base = &mp->mnt_ref; 1652 pcpu = mp->mnt_ref_pcpu; 1653 break; 1654 case MNT_COUNT_LOCKREF: 1655 base = &mp->mnt_lockref; 1656 pcpu = mp->mnt_lockref_pcpu; 1657 break; 1658 case MNT_COUNT_WRITEOPCOUNT: 1659 base = &mp->mnt_writeopcount; 1660 pcpu = mp->mnt_writeopcount_pcpu; 1661 break; 1662 } 1663 1664 sum = *base; 1665 CPU_FOREACH(cpu) { 1666 sum += *zpcpu_get_cpu(pcpu, cpu); 1667 } 1668 return (sum); 1669 } 1670 1671 /* 1672 * Do the actual filesystem unmount. 1673 */ 1674 int 1675 dounmount(struct mount *mp, int flags, struct thread *td) 1676 { 1677 struct vnode *coveredvp, *rootvp; 1678 int error; 1679 uint64_t async_flag; 1680 int mnt_gen_r; 1681 1682 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1683 mnt_gen_r = mp->mnt_gen; 1684 VI_LOCK(coveredvp); 1685 vholdl(coveredvp); 1686 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1687 /* 1688 * Check for mp being unmounted while waiting for the 1689 * covered vnode lock. 1690 */ 1691 if (coveredvp->v_mountedhere != mp || 1692 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1693 VOP_UNLOCK(coveredvp); 1694 vdrop(coveredvp); 1695 vfs_rel(mp); 1696 return (EBUSY); 1697 } 1698 } 1699 1700 /* 1701 * Only privileged root, or (if MNT_USER is set) the user that did the 1702 * original mount is permitted to unmount this filesystem. 1703 */ 1704 error = vfs_suser(mp, td); 1705 if (error != 0) { 1706 if (coveredvp != NULL) { 1707 VOP_UNLOCK(coveredvp); 1708 vdrop(coveredvp); 1709 } 1710 vfs_rel(mp); 1711 return (error); 1712 } 1713 1714 vfs_op_enter(mp); 1715 1716 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF); 1717 MNT_ILOCK(mp); 1718 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 || 1719 (mp->mnt_flag & MNT_UPDATE) != 0 || 1720 !TAILQ_EMPTY(&mp->mnt_uppers)) { 1721 dounmount_cleanup(mp, coveredvp, 0); 1722 return (EBUSY); 1723 } 1724 mp->mnt_kern_flag |= MNTK_UNMOUNT; 1725 rootvp = vfs_cache_root_clear(mp); 1726 if (flags & MNT_NONBUSY) { 1727 MNT_IUNLOCK(mp); 1728 error = vfs_check_usecounts(mp); 1729 MNT_ILOCK(mp); 1730 if (error != 0) { 1731 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT); 1732 if (rootvp != NULL) 1733 vrele(rootvp); 1734 return (error); 1735 } 1736 } 1737 /* Allow filesystems to detect that a forced unmount is in progress. */ 1738 if (flags & MNT_FORCE) { 1739 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1740 MNT_IUNLOCK(mp); 1741 /* 1742 * Must be done after setting MNTK_UNMOUNTF and before 1743 * waiting for mnt_lockref to become 0. 1744 */ 1745 VFS_PURGE(mp); 1746 MNT_ILOCK(mp); 1747 } 1748 error = 0; 1749 if (mp->mnt_lockref) { 1750 mp->mnt_kern_flag |= MNTK_DRAINING; 1751 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1752 "mount drain", 0); 1753 } 1754 MNT_IUNLOCK(mp); 1755 KASSERT(mp->mnt_lockref == 0, 1756 ("%s: invalid lock refcount in the drain path @ %s:%d", 1757 __func__, __FILE__, __LINE__)); 1758 KASSERT(error == 0, 1759 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1760 __func__, __FILE__, __LINE__)); 1761 1762 if (rootvp != NULL) 1763 vrele(rootvp); 1764 1765 if (mp->mnt_flag & MNT_EXPUBLIC) 1766 vfs_setpublicfs(NULL, NULL, NULL); 1767 1768 /* 1769 * From now, we can claim that the use reference on the 1770 * coveredvp is ours, and the ref can be released only by 1771 * successfull unmount by us, or left for later unmount 1772 * attempt. The previously acquired hold reference is no 1773 * longer needed to protect the vnode from reuse. 1774 */ 1775 if (coveredvp != NULL) 1776 vdrop(coveredvp); 1777 1778 vfs_periodic(mp, MNT_WAIT); 1779 MNT_ILOCK(mp); 1780 async_flag = mp->mnt_flag & MNT_ASYNC; 1781 mp->mnt_flag &= ~MNT_ASYNC; 1782 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1783 MNT_IUNLOCK(mp); 1784 cache_purgevfs(mp, false); /* remove cache entries for this file sys */ 1785 vfs_deallocate_syncvnode(mp); 1786 error = VFS_UNMOUNT(mp, flags); 1787 vn_finished_write(mp); 1788 /* 1789 * If we failed to flush the dirty blocks for this mount point, 1790 * undo all the cdir/rdir and rootvnode changes we made above. 1791 * Unless we failed to do so because the device is reporting that 1792 * it doesn't exist anymore. 1793 */ 1794 if (error && error != ENXIO) { 1795 MNT_ILOCK(mp); 1796 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1797 MNT_IUNLOCK(mp); 1798 vfs_allocate_syncvnode(mp); 1799 MNT_ILOCK(mp); 1800 } 1801 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1802 mp->mnt_flag |= async_flag; 1803 if ((mp->mnt_flag & MNT_ASYNC) != 0 && 1804 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0) 1805 mp->mnt_kern_flag |= MNTK_ASYNC; 1806 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1807 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1808 wakeup(mp); 1809 } 1810 vfs_op_exit_locked(mp); 1811 MNT_IUNLOCK(mp); 1812 if (coveredvp) 1813 VOP_UNLOCK(coveredvp); 1814 return (error); 1815 } 1816 mtx_lock(&mountlist_mtx); 1817 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1818 mtx_unlock(&mountlist_mtx); 1819 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td); 1820 if (coveredvp != NULL) { 1821 coveredvp->v_mountedhere = NULL; 1822 VOP_UNLOCK(coveredvp); 1823 } 1824 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1825 if (rootvnode != NULL && mp == rootvnode->v_mount) { 1826 vrele(rootvnode); 1827 rootvnode = NULL; 1828 } 1829 if (mp == rootdevmp) 1830 rootdevmp = NULL; 1831 vfs_mount_destroy(mp); 1832 return (0); 1833 } 1834 1835 /* 1836 * Report errors during filesystem mounting. 1837 */ 1838 void 1839 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1840 { 1841 struct vfsoptlist *moptlist = mp->mnt_optnew; 1842 va_list ap; 1843 int error, len; 1844 char *errmsg; 1845 1846 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1847 if (error || errmsg == NULL || len <= 0) 1848 return; 1849 1850 va_start(ap, fmt); 1851 vsnprintf(errmsg, (size_t)len, fmt, ap); 1852 va_end(ap); 1853 } 1854 1855 void 1856 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1857 { 1858 va_list ap; 1859 int error, len; 1860 char *errmsg; 1861 1862 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1863 if (error || errmsg == NULL || len <= 0) 1864 return; 1865 1866 va_start(ap, fmt); 1867 vsnprintf(errmsg, (size_t)len, fmt, ap); 1868 va_end(ap); 1869 } 1870 1871 /* 1872 * --------------------------------------------------------------------- 1873 * Functions for querying mount options/arguments from filesystems. 1874 */ 1875 1876 /* 1877 * Check that no unknown options are given 1878 */ 1879 int 1880 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1881 { 1882 struct vfsopt *opt; 1883 char errmsg[255]; 1884 const char **t, *p, *q; 1885 int ret = 0; 1886 1887 TAILQ_FOREACH(opt, opts, link) { 1888 p = opt->name; 1889 q = NULL; 1890 if (p[0] == 'n' && p[1] == 'o') 1891 q = p + 2; 1892 for(t = global_opts; *t != NULL; t++) { 1893 if (strcmp(*t, p) == 0) 1894 break; 1895 if (q != NULL) { 1896 if (strcmp(*t, q) == 0) 1897 break; 1898 } 1899 } 1900 if (*t != NULL) 1901 continue; 1902 for(t = legal; *t != NULL; t++) { 1903 if (strcmp(*t, p) == 0) 1904 break; 1905 if (q != NULL) { 1906 if (strcmp(*t, q) == 0) 1907 break; 1908 } 1909 } 1910 if (*t != NULL) 1911 continue; 1912 snprintf(errmsg, sizeof(errmsg), 1913 "mount option <%s> is unknown", p); 1914 ret = EINVAL; 1915 } 1916 if (ret != 0) { 1917 TAILQ_FOREACH(opt, opts, link) { 1918 if (strcmp(opt->name, "errmsg") == 0) { 1919 strncpy((char *)opt->value, errmsg, opt->len); 1920 break; 1921 } 1922 } 1923 if (opt == NULL) 1924 printf("%s\n", errmsg); 1925 } 1926 return (ret); 1927 } 1928 1929 /* 1930 * Get a mount option by its name. 1931 * 1932 * Return 0 if the option was found, ENOENT otherwise. 1933 * If len is non-NULL it will be filled with the length 1934 * of the option. If buf is non-NULL, it will be filled 1935 * with the address of the option. 1936 */ 1937 int 1938 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len) 1939 { 1940 struct vfsopt *opt; 1941 1942 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1943 1944 TAILQ_FOREACH(opt, opts, link) { 1945 if (strcmp(name, opt->name) == 0) { 1946 opt->seen = 1; 1947 if (len != NULL) 1948 *len = opt->len; 1949 if (buf != NULL) 1950 *buf = opt->value; 1951 return (0); 1952 } 1953 } 1954 return (ENOENT); 1955 } 1956 1957 int 1958 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1959 { 1960 struct vfsopt *opt; 1961 1962 if (opts == NULL) 1963 return (-1); 1964 1965 TAILQ_FOREACH(opt, opts, link) { 1966 if (strcmp(name, opt->name) == 0) { 1967 opt->seen = 1; 1968 return (opt->pos); 1969 } 1970 } 1971 return (-1); 1972 } 1973 1974 int 1975 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value) 1976 { 1977 char *opt_value, *vtp; 1978 quad_t iv; 1979 int error, opt_len; 1980 1981 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len); 1982 if (error != 0) 1983 return (error); 1984 if (opt_len == 0 || opt_value == NULL) 1985 return (EINVAL); 1986 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0') 1987 return (EINVAL); 1988 iv = strtoq(opt_value, &vtp, 0); 1989 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0')) 1990 return (EINVAL); 1991 if (iv < 0) 1992 return (EINVAL); 1993 switch (vtp[0]) { 1994 case 't': case 'T': 1995 iv *= 1024; 1996 /* FALLTHROUGH */ 1997 case 'g': case 'G': 1998 iv *= 1024; 1999 /* FALLTHROUGH */ 2000 case 'm': case 'M': 2001 iv *= 1024; 2002 /* FALLTHROUGH */ 2003 case 'k': case 'K': 2004 iv *= 1024; 2005 case '\0': 2006 break; 2007 default: 2008 return (EINVAL); 2009 } 2010 *value = iv; 2011 2012 return (0); 2013 } 2014 2015 char * 2016 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2017 { 2018 struct vfsopt *opt; 2019 2020 *error = 0; 2021 TAILQ_FOREACH(opt, opts, link) { 2022 if (strcmp(name, opt->name) != 0) 2023 continue; 2024 opt->seen = 1; 2025 if (opt->len == 0 || 2026 ((char *)opt->value)[opt->len - 1] != '\0') { 2027 *error = EINVAL; 2028 return (NULL); 2029 } 2030 return (opt->value); 2031 } 2032 *error = ENOENT; 2033 return (NULL); 2034 } 2035 2036 int 2037 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, 2038 uint64_t val) 2039 { 2040 struct vfsopt *opt; 2041 2042 TAILQ_FOREACH(opt, opts, link) { 2043 if (strcmp(name, opt->name) == 0) { 2044 opt->seen = 1; 2045 if (w != NULL) 2046 *w |= val; 2047 return (1); 2048 } 2049 } 2050 if (w != NULL) 2051 *w &= ~val; 2052 return (0); 2053 } 2054 2055 int 2056 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2057 { 2058 va_list ap; 2059 struct vfsopt *opt; 2060 int ret; 2061 2062 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2063 2064 TAILQ_FOREACH(opt, opts, link) { 2065 if (strcmp(name, opt->name) != 0) 2066 continue; 2067 opt->seen = 1; 2068 if (opt->len == 0 || opt->value == NULL) 2069 return (0); 2070 if (((char *)opt->value)[opt->len - 1] != '\0') 2071 return (0); 2072 va_start(ap, fmt); 2073 ret = vsscanf(opt->value, fmt, ap); 2074 va_end(ap); 2075 return (ret); 2076 } 2077 return (0); 2078 } 2079 2080 int 2081 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2082 { 2083 struct vfsopt *opt; 2084 2085 TAILQ_FOREACH(opt, opts, link) { 2086 if (strcmp(name, opt->name) != 0) 2087 continue; 2088 opt->seen = 1; 2089 if (opt->value == NULL) 2090 opt->len = len; 2091 else { 2092 if (opt->len != len) 2093 return (EINVAL); 2094 bcopy(value, opt->value, len); 2095 } 2096 return (0); 2097 } 2098 return (ENOENT); 2099 } 2100 2101 int 2102 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2103 { 2104 struct vfsopt *opt; 2105 2106 TAILQ_FOREACH(opt, opts, link) { 2107 if (strcmp(name, opt->name) != 0) 2108 continue; 2109 opt->seen = 1; 2110 if (opt->value == NULL) 2111 opt->len = len; 2112 else { 2113 if (opt->len < len) 2114 return (EINVAL); 2115 opt->len = len; 2116 bcopy(value, opt->value, len); 2117 } 2118 return (0); 2119 } 2120 return (ENOENT); 2121 } 2122 2123 int 2124 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2125 { 2126 struct vfsopt *opt; 2127 2128 TAILQ_FOREACH(opt, opts, link) { 2129 if (strcmp(name, opt->name) != 0) 2130 continue; 2131 opt->seen = 1; 2132 if (opt->value == NULL) 2133 opt->len = strlen(value) + 1; 2134 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2135 return (EINVAL); 2136 return (0); 2137 } 2138 return (ENOENT); 2139 } 2140 2141 /* 2142 * Find and copy a mount option. 2143 * 2144 * The size of the buffer has to be specified 2145 * in len, if it is not the same length as the 2146 * mount option, EINVAL is returned. 2147 * Returns ENOENT if the option is not found. 2148 */ 2149 int 2150 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len) 2151 { 2152 struct vfsopt *opt; 2153 2154 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2155 2156 TAILQ_FOREACH(opt, opts, link) { 2157 if (strcmp(name, opt->name) == 0) { 2158 opt->seen = 1; 2159 if (len != opt->len) 2160 return (EINVAL); 2161 bcopy(opt->value, dest, opt->len); 2162 return (0); 2163 } 2164 } 2165 return (ENOENT); 2166 } 2167 2168 int 2169 __vfs_statfs(struct mount *mp, struct statfs *sbp) 2170 { 2171 2172 /* 2173 * Filesystems only fill in part of the structure for updates, we 2174 * have to read the entirety first to get all content. 2175 */ 2176 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp)); 2177 2178 /* 2179 * Set these in case the underlying filesystem fails to do so. 2180 */ 2181 sbp->f_version = STATFS_VERSION; 2182 sbp->f_namemax = NAME_MAX; 2183 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 2184 2185 return (mp->mnt_op->vfs_statfs(mp, sbp)); 2186 } 2187 2188 void 2189 vfs_mountedfrom(struct mount *mp, const char *from) 2190 { 2191 2192 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2193 strlcpy(mp->mnt_stat.f_mntfromname, from, 2194 sizeof mp->mnt_stat.f_mntfromname); 2195 } 2196 2197 /* 2198 * --------------------------------------------------------------------- 2199 * This is the api for building mount args and mounting filesystems from 2200 * inside the kernel. 2201 * 2202 * The API works by accumulation of individual args. First error is 2203 * latched. 2204 * 2205 * XXX: should be documented in new manpage kernel_mount(9) 2206 */ 2207 2208 /* A memory allocation which must be freed when we are done */ 2209 struct mntaarg { 2210 SLIST_ENTRY(mntaarg) next; 2211 }; 2212 2213 /* The header for the mount arguments */ 2214 struct mntarg { 2215 struct iovec *v; 2216 int len; 2217 int error; 2218 SLIST_HEAD(, mntaarg) list; 2219 }; 2220 2221 /* 2222 * Add a boolean argument. 2223 * 2224 * flag is the boolean value. 2225 * name must start with "no". 2226 */ 2227 struct mntarg * 2228 mount_argb(struct mntarg *ma, int flag, const char *name) 2229 { 2230 2231 KASSERT(name[0] == 'n' && name[1] == 'o', 2232 ("mount_argb(...,%s): name must start with 'no'", name)); 2233 2234 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2235 } 2236 2237 /* 2238 * Add an argument printf style 2239 */ 2240 struct mntarg * 2241 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2242 { 2243 va_list ap; 2244 struct mntaarg *maa; 2245 struct sbuf *sb; 2246 int len; 2247 2248 if (ma == NULL) { 2249 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2250 SLIST_INIT(&ma->list); 2251 } 2252 if (ma->error) 2253 return (ma); 2254 2255 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2256 M_MOUNT, M_WAITOK); 2257 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2258 ma->v[ma->len].iov_len = strlen(name) + 1; 2259 ma->len++; 2260 2261 sb = sbuf_new_auto(); 2262 va_start(ap, fmt); 2263 sbuf_vprintf(sb, fmt, ap); 2264 va_end(ap); 2265 sbuf_finish(sb); 2266 len = sbuf_len(sb) + 1; 2267 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2268 SLIST_INSERT_HEAD(&ma->list, maa, next); 2269 bcopy(sbuf_data(sb), maa + 1, len); 2270 sbuf_delete(sb); 2271 2272 ma->v[ma->len].iov_base = maa + 1; 2273 ma->v[ma->len].iov_len = len; 2274 ma->len++; 2275 2276 return (ma); 2277 } 2278 2279 /* 2280 * Add an argument which is a userland string. 2281 */ 2282 struct mntarg * 2283 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2284 { 2285 struct mntaarg *maa; 2286 char *tbuf; 2287 2288 if (val == NULL) 2289 return (ma); 2290 if (ma == NULL) { 2291 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2292 SLIST_INIT(&ma->list); 2293 } 2294 if (ma->error) 2295 return (ma); 2296 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2297 SLIST_INSERT_HEAD(&ma->list, maa, next); 2298 tbuf = (void *)(maa + 1); 2299 ma->error = copyinstr(val, tbuf, len, NULL); 2300 return (mount_arg(ma, name, tbuf, -1)); 2301 } 2302 2303 /* 2304 * Plain argument. 2305 * 2306 * If length is -1, treat value as a C string. 2307 */ 2308 struct mntarg * 2309 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2310 { 2311 2312 if (ma == NULL) { 2313 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2314 SLIST_INIT(&ma->list); 2315 } 2316 if (ma->error) 2317 return (ma); 2318 2319 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2320 M_MOUNT, M_WAITOK); 2321 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2322 ma->v[ma->len].iov_len = strlen(name) + 1; 2323 ma->len++; 2324 2325 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2326 if (len < 0) 2327 ma->v[ma->len].iov_len = strlen(val) + 1; 2328 else 2329 ma->v[ma->len].iov_len = len; 2330 ma->len++; 2331 return (ma); 2332 } 2333 2334 /* 2335 * Free a mntarg structure 2336 */ 2337 static void 2338 free_mntarg(struct mntarg *ma) 2339 { 2340 struct mntaarg *maa; 2341 2342 while (!SLIST_EMPTY(&ma->list)) { 2343 maa = SLIST_FIRST(&ma->list); 2344 SLIST_REMOVE_HEAD(&ma->list, next); 2345 free(maa, M_MOUNT); 2346 } 2347 free(ma->v, M_MOUNT); 2348 free(ma, M_MOUNT); 2349 } 2350 2351 /* 2352 * Mount a filesystem 2353 */ 2354 int 2355 kernel_mount(struct mntarg *ma, uint64_t flags) 2356 { 2357 struct uio auio; 2358 int error; 2359 2360 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2361 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2362 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2363 2364 auio.uio_iov = ma->v; 2365 auio.uio_iovcnt = ma->len; 2366 auio.uio_segflg = UIO_SYSSPACE; 2367 2368 error = ma->error; 2369 if (!error) 2370 error = vfs_donmount(curthread, flags, &auio); 2371 free_mntarg(ma); 2372 return (error); 2373 } 2374 2375 /* 2376 * A printflike function to mount a filesystem. 2377 */ 2378 int 2379 kernel_vmount(int flags, ...) 2380 { 2381 struct mntarg *ma = NULL; 2382 va_list ap; 2383 const char *cp; 2384 const void *vp; 2385 int error; 2386 2387 va_start(ap, flags); 2388 for (;;) { 2389 cp = va_arg(ap, const char *); 2390 if (cp == NULL) 2391 break; 2392 vp = va_arg(ap, const void *); 2393 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2394 } 2395 va_end(ap); 2396 2397 error = kernel_mount(ma, flags); 2398 return (error); 2399 } 2400