Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_mount.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * Copyright (c) 1999-2004 Poul-Henning Kamp 3 * Copyright (c) 1999 Michael Smith 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include <sys/param.h> 41 #include <sys/conf.h> 42 #include <sys/clock.h> 43 #include <sys/jail.h> 44 #include <sys/kernel.h> 45 #include <sys/libkern.h> 46 #include <sys/malloc.h> 47 #include <sys/mount.h> 48 #include <sys/mutex.h> 49 #include <sys/namei.h> 50 #include <sys/priv.h> 51 #include <sys/proc.h> 52 #include <sys/filedesc.h> 53 #include <sys/reboot.h> 54 #include <sys/syscallsubr.h> 55 #include <sys/sysproto.h> 56 #include <sys/sx.h> 57 #include <sys/sysctl.h> 58 #include <sys/sysent.h> 59 #include <sys/systm.h> 60 #include <sys/vnode.h> 61 #include <vm/uma.h> 62 63 #include <geom/geom.h> 64 65 #include <machine/stdarg.h> 66 67 #include <security/audit/audit.h> 68 #include <security/mac/mac_framework.h> 69 70 #include "opt_rootdevname.h" 71 #include "opt_ddb.h" 72 #include "opt_mac.h" 73 74 #ifdef DDB 75 #include <ddb/ddb.h> 76 #endif 77 78 #define ROOTNAME "root_device" 79 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 80 81 static int vfs_domount(struct thread *td, const char *fstype, 82 char *fspath, int fsflags, void *fsdata); 83 static int vfs_mountroot_ask(void); 84 static int vfs_mountroot_try(const char *mountfrom); 85 static void free_mntarg(struct mntarg *ma); 86 static int vfs_getopt_pos(struct vfsoptlist *opts, const char *name); 87 88 static int usermount = 0; 89 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 90 "Unprivileged users may mount and unmount file systems"); 91 92 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 93 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 94 static uma_zone_t mount_zone; 95 96 /* List of mounted filesystems. */ 97 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 98 99 /* For any iteration/modification of mountlist */ 100 struct mtx mountlist_mtx; 101 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 102 103 /* 104 * The vnode of the system's root (/ in the filesystem, without chroot 105 * active.) 106 */ 107 struct vnode *rootvnode; 108 109 /* 110 * The root filesystem is detailed in the kernel environment variable 111 * vfs.root.mountfrom, which is expected to be in the general format 112 * 113 * <vfsname>:[<path>] 114 * vfsname := the name of a VFS known to the kernel and capable 115 * of being mounted as root 116 * path := disk device name or other data used by the filesystem 117 * to locate its physical store 118 */ 119 120 /* 121 * Global opts, taken by all filesystems 122 */ 123 static const char *global_opts[] = { 124 "errmsg", 125 "fstype", 126 "fspath", 127 "ro", 128 "rw", 129 "nosuid", 130 "noexec", 131 NULL 132 }; 133 134 /* 135 * The root specifiers we will try if RB_CDROM is specified. 136 */ 137 static char *cdrom_rootdevnames[] = { 138 "cd9660:cd0", 139 "cd9660:acd0", 140 NULL 141 }; 142 143 /* legacy find-root code */ 144 char *rootdevnames[2] = {NULL, NULL}; 145 #ifndef ROOTDEVNAME 146 # define ROOTDEVNAME NULL 147 #endif 148 static const char *ctrootdevname = ROOTDEVNAME; 149 150 /* 151 * --------------------------------------------------------------------- 152 * Functions for building and sanitizing the mount options 153 */ 154 155 /* Remove one mount option. */ 156 static void 157 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 158 { 159 160 TAILQ_REMOVE(opts, opt, link); 161 free(opt->name, M_MOUNT); 162 if (opt->value != NULL) 163 free(opt->value, M_MOUNT); 164 #ifdef INVARIANTS 165 else if (opt->len != 0) 166 panic("%s: mount option with NULL value but length != 0", 167 __func__); 168 #endif 169 free(opt, M_MOUNT); 170 } 171 172 /* Release all resources related to the mount options. */ 173 void 174 vfs_freeopts(struct vfsoptlist *opts) 175 { 176 struct vfsopt *opt; 177 178 while (!TAILQ_EMPTY(opts)) { 179 opt = TAILQ_FIRST(opts); 180 vfs_freeopt(opts, opt); 181 } 182 free(opts, M_MOUNT); 183 } 184 185 void 186 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 187 { 188 struct vfsopt *opt, *temp; 189 190 if (opts == NULL) 191 return; 192 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 193 if (strcmp(opt->name, name) == 0) 194 vfs_freeopt(opts, opt); 195 } 196 } 197 198 static int 199 vfs_isopt_ro(const char *opt) 200 { 201 202 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 || 203 strcmp(opt, "norw") == 0) 204 return (1); 205 return (0); 206 } 207 208 static int 209 vfs_isopt_rw(const char *opt) 210 { 211 212 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0) 213 return (1); 214 return (0); 215 } 216 217 /* 218 * Check if options are equal (with or without the "no" prefix). 219 */ 220 static int 221 vfs_equalopts(const char *opt1, const char *opt2) 222 { 223 224 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 225 if (strcmp(opt1, opt2) == 0) 226 return (1); 227 /* "noopt" vs. "opt" */ 228 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 229 return (1); 230 /* "opt" vs. "noopt" */ 231 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 232 return (1); 233 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */ 234 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) && 235 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2))) 236 return (1); 237 return (0); 238 } 239 240 /* 241 * If a mount option is specified several times, 242 * (with or without the "no" prefix) only keep 243 * the last occurence of it. 244 */ 245 static void 246 vfs_sanitizeopts(struct vfsoptlist *opts) 247 { 248 struct vfsopt *opt, *opt2, *tmp; 249 250 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 251 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 252 while (opt2 != NULL) { 253 if (vfs_equalopts(opt->name, opt2->name)) { 254 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 255 vfs_freeopt(opts, opt2); 256 opt2 = tmp; 257 } else { 258 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 259 } 260 } 261 } 262 } 263 264 /* 265 * Build a linked list of mount options from a struct uio. 266 */ 267 static int 268 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 269 { 270 struct vfsoptlist *opts; 271 struct vfsopt *opt; 272 size_t memused; 273 unsigned int i, iovcnt; 274 int error, namelen, optlen; 275 276 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 277 TAILQ_INIT(opts); 278 memused = 0; 279 iovcnt = auio->uio_iovcnt; 280 for (i = 0; i < iovcnt; i += 2) { 281 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 282 namelen = auio->uio_iov[i].iov_len; 283 optlen = auio->uio_iov[i + 1].iov_len; 284 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 285 opt->value = NULL; 286 opt->len = 0; 287 288 /* 289 * Do this early, so jumps to "bad" will free the current 290 * option. 291 */ 292 TAILQ_INSERT_TAIL(opts, opt, link); 293 memused += sizeof(struct vfsopt) + optlen + namelen; 294 295 /* 296 * Avoid consuming too much memory, and attempts to overflow 297 * memused. 298 */ 299 if (memused > VFS_MOUNTARG_SIZE_MAX || 300 optlen > VFS_MOUNTARG_SIZE_MAX || 301 namelen > VFS_MOUNTARG_SIZE_MAX) { 302 error = EINVAL; 303 goto bad; 304 } 305 306 if (auio->uio_segflg == UIO_SYSSPACE) { 307 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 308 } else { 309 error = copyin(auio->uio_iov[i].iov_base, opt->name, 310 namelen); 311 if (error) 312 goto bad; 313 } 314 /* Ensure names are null-terminated strings. */ 315 if (opt->name[namelen - 1] != '\0') { 316 error = EINVAL; 317 goto bad; 318 } 319 if (optlen != 0) { 320 opt->len = optlen; 321 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 322 if (auio->uio_segflg == UIO_SYSSPACE) { 323 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 324 optlen); 325 } else { 326 error = copyin(auio->uio_iov[i + 1].iov_base, 327 opt->value, optlen); 328 if (error) 329 goto bad; 330 } 331 } 332 } 333 vfs_sanitizeopts(opts); 334 *options = opts; 335 return (0); 336 bad: 337 vfs_freeopts(opts); 338 return (error); 339 } 340 341 /* 342 * Merge the old mount options with the new ones passed 343 * in the MNT_UPDATE case. 344 */ 345 static void 346 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 347 { 348 struct vfsopt *opt, *opt2, *new; 349 350 TAILQ_FOREACH(opt, opts, link) { 351 /* 352 * Check that this option hasn't been redefined 353 * nor cancelled with a "no" mount option. 354 */ 355 opt2 = TAILQ_FIRST(toopts); 356 while (opt2 != NULL) { 357 if (strcmp(opt2->name, opt->name) == 0) 358 goto next; 359 if (strncmp(opt2->name, "no", 2) == 0 && 360 strcmp(opt2->name + 2, opt->name) == 0) { 361 vfs_freeopt(toopts, opt2); 362 goto next; 363 } 364 opt2 = TAILQ_NEXT(opt2, link); 365 } 366 /* We want this option, duplicate it. */ 367 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 368 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 369 strcpy(new->name, opt->name); 370 if (opt->len != 0) { 371 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 372 bcopy(opt->value, new->value, opt->len); 373 } else { 374 new->value = NULL; 375 } 376 new->len = opt->len; 377 TAILQ_INSERT_TAIL(toopts, new, link); 378 next: 379 continue; 380 } 381 } 382 383 /* 384 * Mount a filesystem. 385 */ 386 int 387 nmount(td, uap) 388 struct thread *td; 389 struct nmount_args /* { 390 struct iovec *iovp; 391 unsigned int iovcnt; 392 int flags; 393 } */ *uap; 394 { 395 struct uio *auio; 396 struct iovec *iov; 397 unsigned int i; 398 int error; 399 u_int iovcnt; 400 401 AUDIT_ARG(fflags, uap->flags); 402 403 /* 404 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 405 * userspace to set this flag, but we must filter it out if we want 406 * MNT_UPDATE on the root file system to work. 407 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 408 */ 409 uap->flags &= ~MNT_ROOTFS; 410 411 iovcnt = uap->iovcnt; 412 /* 413 * Check that we have an even number of iovec's 414 * and that we have at least two options. 415 */ 416 if ((iovcnt & 1) || (iovcnt < 4)) 417 return (EINVAL); 418 419 error = copyinuio(uap->iovp, iovcnt, &auio); 420 if (error) 421 return (error); 422 iov = auio->uio_iov; 423 for (i = 0; i < iovcnt; i++) { 424 if (iov->iov_len > MMAXOPTIONLEN) { 425 free(auio, M_IOV); 426 return (EINVAL); 427 } 428 iov++; 429 } 430 error = vfs_donmount(td, uap->flags, auio); 431 432 free(auio, M_IOV); 433 return (error); 434 } 435 436 /* 437 * --------------------------------------------------------------------- 438 * Various utility functions 439 */ 440 441 void 442 vfs_ref(struct mount *mp) 443 { 444 445 MNT_ILOCK(mp); 446 MNT_REF(mp); 447 MNT_IUNLOCK(mp); 448 } 449 450 void 451 vfs_rel(struct mount *mp) 452 { 453 454 MNT_ILOCK(mp); 455 MNT_REL(mp); 456 MNT_IUNLOCK(mp); 457 } 458 459 static int 460 mount_init(void *mem, int size, int flags) 461 { 462 struct mount *mp; 463 464 mp = (struct mount *)mem; 465 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 466 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 467 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 468 return (0); 469 } 470 471 static void 472 mount_fini(void *mem, int size) 473 { 474 struct mount *mp; 475 476 mp = (struct mount *)mem; 477 lockdestroy(&mp->mnt_explock); 478 lockdestroy(&mp->mnt_lock); 479 mtx_destroy(&mp->mnt_mtx); 480 } 481 482 /* 483 * Allocate and initialize the mount point struct. 484 */ 485 struct mount * 486 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, 487 const char *fspath, struct thread *td) 488 { 489 struct mount *mp; 490 491 mp = uma_zalloc(mount_zone, M_WAITOK); 492 bzero(&mp->mnt_startzero, 493 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 494 TAILQ_INIT(&mp->mnt_nvnodelist); 495 mp->mnt_nvnodelistsize = 0; 496 mp->mnt_ref = 0; 497 (void) vfs_busy(mp, LK_NOWAIT, 0, td); 498 mp->mnt_op = vfsp->vfc_vfsops; 499 mp->mnt_vfc = vfsp; 500 vfsp->vfc_refcount++; /* XXX Unlocked */ 501 mp->mnt_stat.f_type = vfsp->vfc_typenum; 502 mp->mnt_gen++; 503 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 504 mp->mnt_vnodecovered = vp; 505 mp->mnt_cred = crdup(td->td_ucred); 506 mp->mnt_stat.f_owner = td->td_ucred->cr_uid; 507 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 508 mp->mnt_iosize_max = DFLTPHYS; 509 #ifdef MAC 510 mac_init_mount(mp); 511 mac_create_mount(td->td_ucred, mp); 512 #endif 513 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 514 return (mp); 515 } 516 517 /* 518 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 519 */ 520 void 521 vfs_mount_destroy(struct mount *mp) 522 { 523 int i; 524 525 MNT_ILOCK(mp); 526 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 527 if (mp->mnt_kern_flag & MNTK_MWAIT) { 528 mp->mnt_kern_flag &= ~MNTK_MWAIT; 529 wakeup(mp); 530 } 531 for (i = 0; mp->mnt_ref && i < 3; i++) 532 msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz); 533 /* 534 * This will always cause a 3 second delay in rebooting due to 535 * refs on the root mountpoint that never go away. Most of these 536 * are held by init which never exits. 537 */ 538 if (i == 3 && (!rebooting || bootverbose)) 539 printf("Mount point %s had %d dangling refs\n", 540 mp->mnt_stat.f_mntonname, mp->mnt_ref); 541 if (mp->mnt_holdcnt != 0) { 542 printf("Waiting for mount point to be unheld\n"); 543 while (mp->mnt_holdcnt != 0) { 544 mp->mnt_holdcntwaiters++; 545 msleep(&mp->mnt_holdcnt, MNT_MTX(mp), 546 PZERO, "mntdestroy", 0); 547 mp->mnt_holdcntwaiters--; 548 } 549 printf("mount point unheld\n"); 550 } 551 if (mp->mnt_writeopcount > 0) { 552 printf("Waiting for mount point write ops\n"); 553 while (mp->mnt_writeopcount > 0) { 554 mp->mnt_kern_flag |= MNTK_SUSPEND; 555 msleep(&mp->mnt_writeopcount, 556 MNT_MTX(mp), 557 PZERO, "mntdestroy2", 0); 558 } 559 printf("mount point write ops completed\n"); 560 } 561 if (mp->mnt_secondary_writes > 0) { 562 printf("Waiting for mount point secondary write ops\n"); 563 while (mp->mnt_secondary_writes > 0) { 564 mp->mnt_kern_flag |= MNTK_SUSPEND; 565 msleep(&mp->mnt_secondary_writes, 566 MNT_MTX(mp), 567 PZERO, "mntdestroy3", 0); 568 } 569 printf("mount point secondary write ops completed\n"); 570 } 571 MNT_IUNLOCK(mp); 572 mp->mnt_vfc->vfc_refcount--; 573 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 574 struct vnode *vp; 575 576 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 577 vprint("", vp); 578 panic("unmount: dangling vnode"); 579 } 580 MNT_ILOCK(mp); 581 if (mp->mnt_kern_flag & MNTK_MWAIT) 582 wakeup(mp); 583 if (mp->mnt_writeopcount != 0) 584 panic("vfs_mount_destroy: nonzero writeopcount"); 585 if (mp->mnt_secondary_writes != 0) 586 panic("vfs_mount_destroy: nonzero secondary_writes"); 587 if (mp->mnt_nvnodelistsize != 0) 588 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 589 mp->mnt_writeopcount = -1000; 590 mp->mnt_nvnodelistsize = -1000; 591 mp->mnt_secondary_writes = -1000; 592 MNT_IUNLOCK(mp); 593 #ifdef MAC 594 mac_destroy_mount(mp); 595 #endif 596 if (mp->mnt_opt != NULL) 597 vfs_freeopts(mp->mnt_opt); 598 crfree(mp->mnt_cred); 599 uma_zfree(mount_zone, mp); 600 } 601 602 int 603 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 604 { 605 struct vfsoptlist *optlist; 606 struct vfsopt *opt, *noro_opt, *tmp_opt; 607 char *fstype, *fspath, *errmsg; 608 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 609 int has_rw, has_noro; 610 611 errmsg = NULL; 612 errmsg_len = 0; 613 errmsg_pos = -1; 614 has_rw = 0; 615 has_noro = 0; 616 617 error = vfs_buildopts(fsoptions, &optlist); 618 if (error) 619 return (error); 620 621 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 622 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 623 624 /* 625 * We need these two options before the others, 626 * and they are mandatory for any filesystem. 627 * Ensure they are NUL terminated as well. 628 */ 629 fstypelen = 0; 630 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 631 if (error || fstype[fstypelen - 1] != '\0') { 632 error = EINVAL; 633 if (errmsg != NULL) 634 strncpy(errmsg, "Invalid fstype", errmsg_len); 635 goto bail; 636 } 637 fspathlen = 0; 638 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 639 if (error || fspath[fspathlen - 1] != '\0') { 640 error = EINVAL; 641 if (errmsg != NULL) 642 strncpy(errmsg, "Invalid fspath", errmsg_len); 643 goto bail; 644 } 645 646 /* 647 * We need to see if we have the "update" option 648 * before we call vfs_domount(), since vfs_domount() has special 649 * logic based on MNT_UPDATE. This is very important 650 * when we want to update the root filesystem. 651 */ 652 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 653 if (strcmp(opt->name, "update") == 0) { 654 fsflags |= MNT_UPDATE; 655 vfs_freeopt(optlist, opt); 656 } 657 else if (strcmp(opt->name, "async") == 0) 658 fsflags |= MNT_ASYNC; 659 else if (strcmp(opt->name, "force") == 0) { 660 fsflags |= MNT_FORCE; 661 vfs_freeopt(optlist, opt); 662 } 663 else if (strcmp(opt->name, "reload") == 0) { 664 fsflags |= MNT_RELOAD; 665 vfs_freeopt(optlist, opt); 666 } 667 else if (strcmp(opt->name, "multilabel") == 0) 668 fsflags |= MNT_MULTILABEL; 669 else if (strcmp(opt->name, "noasync") == 0) 670 fsflags &= ~MNT_ASYNC; 671 else if (strcmp(opt->name, "noatime") == 0) 672 fsflags |= MNT_NOATIME; 673 else if (strcmp(opt->name, "atime") == 0) { 674 free(opt->name, M_MOUNT); 675 opt->name = strdup("nonoatime", M_MOUNT); 676 } 677 else if (strcmp(opt->name, "noclusterr") == 0) 678 fsflags |= MNT_NOCLUSTERR; 679 else if (strcmp(opt->name, "clusterr") == 0) { 680 free(opt->name, M_MOUNT); 681 opt->name = strdup("nonoclusterr", M_MOUNT); 682 } 683 else if (strcmp(opt->name, "noclusterw") == 0) 684 fsflags |= MNT_NOCLUSTERW; 685 else if (strcmp(opt->name, "clusterw") == 0) { 686 free(opt->name, M_MOUNT); 687 opt->name = strdup("nonoclusterw", M_MOUNT); 688 } 689 else if (strcmp(opt->name, "noexec") == 0) 690 fsflags |= MNT_NOEXEC; 691 else if (strcmp(opt->name, "exec") == 0) { 692 free(opt->name, M_MOUNT); 693 opt->name = strdup("nonoexec", M_MOUNT); 694 } 695 else if (strcmp(opt->name, "nosuid") == 0) 696 fsflags |= MNT_NOSUID; 697 else if (strcmp(opt->name, "suid") == 0) { 698 free(opt->name, M_MOUNT); 699 opt->name = strdup("nonosuid", M_MOUNT); 700 } 701 else if (strcmp(opt->name, "nosymfollow") == 0) 702 fsflags |= MNT_NOSYMFOLLOW; 703 else if (strcmp(opt->name, "symfollow") == 0) { 704 free(opt->name, M_MOUNT); 705 opt->name = strdup("nonosymfollow", M_MOUNT); 706 } 707 else if (strcmp(opt->name, "noro") == 0) { 708 fsflags &= ~MNT_RDONLY; 709 has_noro = 1; 710 } 711 else if (strcmp(opt->name, "rw") == 0) { 712 fsflags &= ~MNT_RDONLY; 713 has_rw = 1; 714 } 715 else if (strcmp(opt->name, "ro") == 0) 716 fsflags |= MNT_RDONLY; 717 else if (strcmp(opt->name, "rdonly") == 0) { 718 free(opt->name, M_MOUNT); 719 opt->name = strdup("ro", M_MOUNT); 720 fsflags |= MNT_RDONLY; 721 } 722 else if (strcmp(opt->name, "suiddir") == 0) 723 fsflags |= MNT_SUIDDIR; 724 else if (strcmp(opt->name, "sync") == 0) 725 fsflags |= MNT_SYNCHRONOUS; 726 else if (strcmp(opt->name, "union") == 0) 727 fsflags |= MNT_UNION; 728 } 729 730 /* 731 * If "rw" was specified as a mount option, and we 732 * are trying to update a mount-point from "ro" to "rw", 733 * we need a mount option "noro", since in vfs_mergeopts(), 734 * "noro" will cancel "ro", but "rw" will not do anything. 735 */ 736 if (has_rw && !has_noro) { 737 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 738 noro_opt->name = strdup("noro", M_MOUNT); 739 noro_opt->value = NULL; 740 noro_opt->len = 0; 741 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 742 } 743 744 /* 745 * Be ultra-paranoid about making sure the type and fspath 746 * variables will fit in our mp buffers, including the 747 * terminating NUL. 748 */ 749 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 750 error = ENAMETOOLONG; 751 goto bail; 752 } 753 754 mtx_lock(&Giant); 755 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 756 mtx_unlock(&Giant); 757 bail: 758 /* copyout the errmsg */ 759 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 760 && errmsg_len > 0 && errmsg != NULL) { 761 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 762 bcopy(errmsg, 763 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 764 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 765 } else { 766 copyout(errmsg, 767 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 768 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 769 } 770 } 771 772 if (error != 0) 773 vfs_freeopts(optlist); 774 return (error); 775 } 776 777 /* 778 * Old mount API. 779 */ 780 #ifndef _SYS_SYSPROTO_H_ 781 struct mount_args { 782 char *type; 783 char *path; 784 int flags; 785 caddr_t data; 786 }; 787 #endif 788 /* ARGSUSED */ 789 int 790 mount(td, uap) 791 struct thread *td; 792 struct mount_args /* { 793 char *type; 794 char *path; 795 int flags; 796 caddr_t data; 797 } */ *uap; 798 { 799 char *fstype; 800 struct vfsconf *vfsp = NULL; 801 struct mntarg *ma = NULL; 802 int error; 803 804 AUDIT_ARG(fflags, uap->flags); 805 806 /* 807 * Filter out MNT_ROOTFS. We do not want clients of mount() in 808 * userspace to set this flag, but we must filter it out if we want 809 * MNT_UPDATE on the root file system to work. 810 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 811 */ 812 uap->flags &= ~MNT_ROOTFS; 813 814 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 815 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 816 if (error) { 817 free(fstype, M_TEMP); 818 return (error); 819 } 820 821 AUDIT_ARG(text, fstype); 822 mtx_lock(&Giant); 823 vfsp = vfs_byname_kld(fstype, td, &error); 824 free(fstype, M_TEMP); 825 if (vfsp == NULL) { 826 mtx_unlock(&Giant); 827 return (ENOENT); 828 } 829 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 830 mtx_unlock(&Giant); 831 return (EOPNOTSUPP); 832 } 833 834 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 835 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 836 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 837 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 838 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 839 840 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 841 mtx_unlock(&Giant); 842 return (error); 843 } 844 845 846 /* 847 * vfs_domount(): actually attempt a filesystem mount. 848 */ 849 static int 850 vfs_domount( 851 struct thread *td, /* Calling thread. */ 852 const char *fstype, /* Filesystem type. */ 853 char *fspath, /* Mount path. */ 854 int fsflags, /* Flags common to all filesystems. */ 855 void *fsdata /* Options local to the filesystem. */ 856 ) 857 { 858 struct vnode *vp; 859 struct mount *mp; 860 struct vfsconf *vfsp; 861 struct export_args export; 862 int error, flag = 0; 863 struct vattr va; 864 struct nameidata nd; 865 866 mtx_assert(&Giant, MA_OWNED); 867 /* 868 * Be ultra-paranoid about making sure the type and fspath 869 * variables will fit in our mp buffers, including the 870 * terminating NUL. 871 */ 872 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 873 return (ENAMETOOLONG); 874 875 if (jailed(td->td_ucred) || usermount == 0) { 876 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 877 return (error); 878 } 879 880 /* 881 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 882 */ 883 if (fsflags & MNT_EXPORTED) { 884 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 885 if (error) 886 return (error); 887 } 888 if (fsflags & MNT_SUIDDIR) { 889 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 890 if (error) 891 return (error); 892 } 893 /* 894 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 895 */ 896 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 897 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 898 fsflags |= MNT_NOSUID | MNT_USER; 899 } 900 901 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 902 vfsp = NULL; 903 if ((fsflags & MNT_UPDATE) == 0) { 904 /* Don't try to load KLDs if we're mounting the root. */ 905 if (fsflags & MNT_ROOTFS) 906 vfsp = vfs_byname(fstype); 907 else 908 vfsp = vfs_byname_kld(fstype, td, &error); 909 if (vfsp == NULL) 910 return (ENODEV); 911 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 912 return (EPERM); 913 } 914 /* 915 * Get vnode to be covered 916 */ 917 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 918 fspath, td); 919 if ((error = namei(&nd)) != 0) 920 return (error); 921 NDFREE(&nd, NDF_ONLY_PNBUF); 922 vp = nd.ni_vp; 923 if (fsflags & MNT_UPDATE) { 924 if ((vp->v_vflag & VV_ROOT) == 0) { 925 vput(vp); 926 return (EINVAL); 927 } 928 mp = vp->v_mount; 929 MNT_ILOCK(mp); 930 flag = mp->mnt_flag; 931 /* 932 * We only allow the filesystem to be reloaded if it 933 * is currently mounted read-only. 934 */ 935 if ((fsflags & MNT_RELOAD) && 936 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 937 MNT_IUNLOCK(mp); 938 vput(vp); 939 return (EOPNOTSUPP); /* Needs translation */ 940 } 941 MNT_IUNLOCK(mp); 942 /* 943 * Only privileged root, or (if MNT_USER is set) the user that 944 * did the original mount is permitted to update it. 945 */ 946 error = vfs_suser(mp, td); 947 if (error) { 948 vput(vp); 949 return (error); 950 } 951 if (vfs_busy(mp, LK_NOWAIT, 0, td)) { 952 vput(vp); 953 return (EBUSY); 954 } 955 VI_LOCK(vp); 956 if ((vp->v_iflag & VI_MOUNT) != 0 || 957 vp->v_mountedhere != NULL) { 958 VI_UNLOCK(vp); 959 vfs_unbusy(mp, td); 960 vput(vp); 961 return (EBUSY); 962 } 963 vp->v_iflag |= VI_MOUNT; 964 VI_UNLOCK(vp); 965 MNT_ILOCK(mp); 966 mp->mnt_flag |= fsflags & 967 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 968 MNT_IUNLOCK(mp); 969 VOP_UNLOCK(vp, 0, td); 970 mp->mnt_optnew = fsdata; 971 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 972 } else { 973 /* 974 * If the user is not root, ensure that they own the directory 975 * onto which we are attempting to mount. 976 */ 977 error = VOP_GETATTR(vp, &va, td->td_ucred, td); 978 if (error) { 979 vput(vp); 980 return (error); 981 } 982 if (va.va_uid != td->td_ucred->cr_uid) { 983 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 984 0); 985 if (error) { 986 vput(vp); 987 return (error); 988 } 989 } 990 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 991 if (error != 0) { 992 vput(vp); 993 return (error); 994 } 995 if (vp->v_type != VDIR) { 996 vput(vp); 997 return (ENOTDIR); 998 } 999 VI_LOCK(vp); 1000 if ((vp->v_iflag & VI_MOUNT) != 0 || 1001 vp->v_mountedhere != NULL) { 1002 VI_UNLOCK(vp); 1003 vput(vp); 1004 return (EBUSY); 1005 } 1006 vp->v_iflag |= VI_MOUNT; 1007 VI_UNLOCK(vp); 1008 1009 /* 1010 * Allocate and initialize the filesystem. 1011 */ 1012 mp = vfs_mount_alloc(vp, vfsp, fspath, td); 1013 VOP_UNLOCK(vp, 0, td); 1014 1015 /* XXXMAC: pass to vfs_mount_alloc? */ 1016 mp->mnt_optnew = fsdata; 1017 } 1018 1019 /* 1020 * Set the mount level flags. 1021 */ 1022 MNT_ILOCK(mp); 1023 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 1024 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 1025 MNT_RDONLY)); 1026 if ((mp->mnt_flag & MNT_ASYNC) == 0) 1027 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1028 MNT_IUNLOCK(mp); 1029 /* 1030 * Mount the filesystem. 1031 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1032 * get. No freeing of cn_pnbuf. 1033 */ 1034 error = VFS_MOUNT(mp, td); 1035 1036 /* 1037 * Process the export option only if we are 1038 * updating mount options. 1039 */ 1040 if (!error && (fsflags & MNT_UPDATE)) { 1041 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 1042 sizeof(export)) == 0) 1043 error = vfs_export(mp, &export); 1044 } 1045 1046 if (!error) { 1047 if (mp->mnt_opt != NULL) 1048 vfs_freeopts(mp->mnt_opt); 1049 mp->mnt_opt = mp->mnt_optnew; 1050 (void)VFS_STATFS(mp, &mp->mnt_stat, td); 1051 } 1052 /* 1053 * Prevent external consumers of mount options from reading 1054 * mnt_optnew. 1055 */ 1056 mp->mnt_optnew = NULL; 1057 if (mp->mnt_flag & MNT_UPDATE) { 1058 MNT_ILOCK(mp); 1059 if (error) 1060 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1061 (flag & ~MNT_QUOTA); 1062 else 1063 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1064 MNT_FORCE | MNT_SNAPSHOT); 1065 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1066 mp->mnt_kern_flag |= MNTK_ASYNC; 1067 else 1068 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1069 MNT_IUNLOCK(mp); 1070 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1071 if (mp->mnt_syncer == NULL) 1072 error = vfs_allocate_syncvnode(mp); 1073 } else { 1074 if (mp->mnt_syncer != NULL) 1075 vrele(mp->mnt_syncer); 1076 mp->mnt_syncer = NULL; 1077 } 1078 vfs_unbusy(mp, td); 1079 VI_LOCK(vp); 1080 vp->v_iflag &= ~VI_MOUNT; 1081 VI_UNLOCK(vp); 1082 vrele(vp); 1083 return (error); 1084 } 1085 MNT_ILOCK(mp); 1086 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1087 mp->mnt_kern_flag |= MNTK_ASYNC; 1088 else 1089 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1090 MNT_IUNLOCK(mp); 1091 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1092 /* 1093 * Put the new filesystem on the mount list after root. 1094 */ 1095 cache_purge(vp); 1096 if (!error) { 1097 struct vnode *newdp; 1098 1099 VI_LOCK(vp); 1100 vp->v_iflag &= ~VI_MOUNT; 1101 VI_UNLOCK(vp); 1102 vp->v_mountedhere = mp; 1103 mtx_lock(&mountlist_mtx); 1104 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1105 mtx_unlock(&mountlist_mtx); 1106 vfs_event_signal(NULL, VQ_MOUNT, 0); 1107 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 1108 panic("mount: lost mount"); 1109 VOP_UNLOCK(newdp, 0, td); 1110 VOP_UNLOCK(vp, 0, td); 1111 mountcheckdirs(vp, newdp); 1112 vrele(newdp); 1113 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1114 error = vfs_allocate_syncvnode(mp); 1115 vfs_unbusy(mp, td); 1116 if (error) 1117 vrele(vp); 1118 } else { 1119 VI_LOCK(vp); 1120 vp->v_iflag &= ~VI_MOUNT; 1121 VI_UNLOCK(vp); 1122 vfs_unbusy(mp, td); 1123 vfs_mount_destroy(mp); 1124 vput(vp); 1125 } 1126 return (error); 1127 } 1128 1129 /* 1130 * Unmount a filesystem. 1131 * 1132 * Note: unmount takes a path to the vnode mounted on as argument, not 1133 * special file (as before). 1134 */ 1135 #ifndef _SYS_SYSPROTO_H_ 1136 struct unmount_args { 1137 char *path; 1138 int flags; 1139 }; 1140 #endif 1141 /* ARGSUSED */ 1142 int 1143 unmount(td, uap) 1144 struct thread *td; 1145 register struct unmount_args /* { 1146 char *path; 1147 int flags; 1148 } */ *uap; 1149 { 1150 struct mount *mp; 1151 char *pathbuf; 1152 int error, id0, id1; 1153 1154 if (jailed(td->td_ucred) || usermount == 0) { 1155 error = priv_check(td, PRIV_VFS_UNMOUNT); 1156 if (error) 1157 return (error); 1158 } 1159 1160 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1161 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1162 if (error) { 1163 free(pathbuf, M_TEMP); 1164 return (error); 1165 } 1166 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1167 mtx_lock(&Giant); 1168 if (uap->flags & MNT_BYFSID) { 1169 /* Decode the filesystem ID. */ 1170 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1171 mtx_unlock(&Giant); 1172 free(pathbuf, M_TEMP); 1173 return (EINVAL); 1174 } 1175 1176 mtx_lock(&mountlist_mtx); 1177 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1178 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1179 mp->mnt_stat.f_fsid.val[1] == id1) 1180 break; 1181 } 1182 mtx_unlock(&mountlist_mtx); 1183 } else { 1184 mtx_lock(&mountlist_mtx); 1185 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1186 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1187 break; 1188 } 1189 mtx_unlock(&mountlist_mtx); 1190 } 1191 free(pathbuf, M_TEMP); 1192 if (mp == NULL) { 1193 /* 1194 * Previously we returned ENOENT for a nonexistent path and 1195 * EINVAL for a non-mountpoint. We cannot tell these apart 1196 * now, so in the !MNT_BYFSID case return the more likely 1197 * EINVAL for compatibility. 1198 */ 1199 mtx_unlock(&Giant); 1200 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1201 } 1202 1203 /* 1204 * Don't allow unmounting the root filesystem. 1205 */ 1206 if (mp->mnt_flag & MNT_ROOTFS) { 1207 mtx_unlock(&Giant); 1208 return (EINVAL); 1209 } 1210 error = dounmount(mp, uap->flags, td); 1211 mtx_unlock(&Giant); 1212 return (error); 1213 } 1214 1215 /* 1216 * Do the actual filesystem unmount. 1217 */ 1218 int 1219 dounmount(mp, flags, td) 1220 struct mount *mp; 1221 int flags; 1222 struct thread *td; 1223 { 1224 struct vnode *coveredvp, *fsrootvp; 1225 int error; 1226 int async_flag; 1227 int mnt_gen_r; 1228 1229 mtx_assert(&Giant, MA_OWNED); 1230 1231 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1232 mnt_gen_r = mp->mnt_gen; 1233 VI_LOCK(coveredvp); 1234 vholdl(coveredvp); 1235 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY, td); 1236 vdrop(coveredvp); 1237 /* 1238 * Check for mp being unmounted while waiting for the 1239 * covered vnode lock. 1240 */ 1241 if (coveredvp->v_mountedhere != mp || 1242 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1243 VOP_UNLOCK(coveredvp, 0, td); 1244 return (EBUSY); 1245 } 1246 } 1247 /* 1248 * Only privileged root, or (if MNT_USER is set) the user that did the 1249 * original mount is permitted to unmount this filesystem. 1250 */ 1251 error = vfs_suser(mp, td); 1252 if (error) { 1253 if (coveredvp) 1254 VOP_UNLOCK(coveredvp, 0, td); 1255 return (error); 1256 } 1257 1258 MNT_ILOCK(mp); 1259 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1260 MNT_IUNLOCK(mp); 1261 if (coveredvp) 1262 VOP_UNLOCK(coveredvp, 0, td); 1263 return (EBUSY); 1264 } 1265 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1266 /* Allow filesystems to detect that a forced unmount is in progress. */ 1267 if (flags & MNT_FORCE) 1268 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1269 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 1270 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td); 1271 if (error) { 1272 MNT_ILOCK(mp); 1273 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1274 MNTK_UNMOUNTF); 1275 if (mp->mnt_kern_flag & MNTK_MWAIT) 1276 wakeup(mp); 1277 MNT_IUNLOCK(mp); 1278 if (coveredvp) 1279 VOP_UNLOCK(coveredvp, 0, td); 1280 return (error); 1281 } 1282 vn_start_write(NULL, &mp, V_WAIT); 1283 1284 if (mp->mnt_flag & MNT_EXPUBLIC) 1285 vfs_setpublicfs(NULL, NULL, NULL); 1286 1287 vfs_msync(mp, MNT_WAIT); 1288 MNT_ILOCK(mp); 1289 async_flag = mp->mnt_flag & MNT_ASYNC; 1290 mp->mnt_flag &= ~MNT_ASYNC; 1291 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1292 MNT_IUNLOCK(mp); 1293 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1294 if (mp->mnt_syncer != NULL) 1295 vrele(mp->mnt_syncer); 1296 /* 1297 * For forced unmounts, move process cdir/rdir refs on the fs root 1298 * vnode to the covered vnode. For non-forced unmounts we want 1299 * such references to cause an EBUSY error. 1300 */ 1301 if ((flags & MNT_FORCE) && 1302 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1303 if (mp->mnt_vnodecovered != NULL) 1304 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1305 if (fsrootvp == rootvnode) { 1306 vrele(rootvnode); 1307 rootvnode = NULL; 1308 } 1309 vput(fsrootvp); 1310 } 1311 if (((mp->mnt_flag & MNT_RDONLY) || 1312 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1313 (flags & MNT_FORCE)) { 1314 error = VFS_UNMOUNT(mp, flags, td); 1315 } 1316 vn_finished_write(mp); 1317 /* 1318 * If we failed to flush the dirty blocks for this mount point, 1319 * undo all the cdir/rdir and rootvnode changes we made above. 1320 * Unless we failed to do so because the device is reporting that 1321 * it doesn't exist anymore. 1322 */ 1323 if (error && error != ENXIO) { 1324 if ((flags & MNT_FORCE) && 1325 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1326 if (mp->mnt_vnodecovered != NULL) 1327 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1328 if (rootvnode == NULL) { 1329 rootvnode = fsrootvp; 1330 vref(rootvnode); 1331 } 1332 vput(fsrootvp); 1333 } 1334 MNT_ILOCK(mp); 1335 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1336 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1337 MNT_IUNLOCK(mp); 1338 (void) vfs_allocate_syncvnode(mp); 1339 MNT_ILOCK(mp); 1340 } 1341 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1342 mp->mnt_flag |= async_flag; 1343 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1344 mp->mnt_kern_flag |= MNTK_ASYNC; 1345 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1346 if (mp->mnt_kern_flag & MNTK_MWAIT) 1347 wakeup(mp); 1348 MNT_IUNLOCK(mp); 1349 if (coveredvp) 1350 VOP_UNLOCK(coveredvp, 0, td); 1351 return (error); 1352 } 1353 mtx_lock(&mountlist_mtx); 1354 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1355 mtx_unlock(&mountlist_mtx); 1356 if (coveredvp != NULL) { 1357 coveredvp->v_mountedhere = NULL; 1358 vput(coveredvp); 1359 } 1360 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1361 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1362 vfs_mount_destroy(mp); 1363 return (0); 1364 } 1365 1366 /* 1367 * --------------------------------------------------------------------- 1368 * Mounting of root filesystem 1369 * 1370 */ 1371 1372 struct root_hold_token { 1373 const char *who; 1374 LIST_ENTRY(root_hold_token) list; 1375 }; 1376 1377 static LIST_HEAD(, root_hold_token) root_holds = 1378 LIST_HEAD_INITIALIZER(&root_holds); 1379 1380 static int root_mount_complete; 1381 1382 /* 1383 * Hold root mount. 1384 */ 1385 struct root_hold_token * 1386 root_mount_hold(const char *identifier) 1387 { 1388 struct root_hold_token *h; 1389 1390 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1391 h->who = identifier; 1392 mtx_lock(&mountlist_mtx); 1393 LIST_INSERT_HEAD(&root_holds, h, list); 1394 mtx_unlock(&mountlist_mtx); 1395 return (h); 1396 } 1397 1398 /* 1399 * Release root mount. 1400 */ 1401 void 1402 root_mount_rel(struct root_hold_token *h) 1403 { 1404 1405 mtx_lock(&mountlist_mtx); 1406 LIST_REMOVE(h, list); 1407 wakeup(&root_holds); 1408 mtx_unlock(&mountlist_mtx); 1409 free(h, M_DEVBUF); 1410 } 1411 1412 /* 1413 * Wait for all subsystems to release root mount. 1414 */ 1415 static void 1416 root_mount_prepare(void) 1417 { 1418 struct root_hold_token *h; 1419 1420 for (;;) { 1421 DROP_GIANT(); 1422 g_waitidle(); 1423 PICKUP_GIANT(); 1424 mtx_lock(&mountlist_mtx); 1425 if (LIST_EMPTY(&root_holds)) { 1426 mtx_unlock(&mountlist_mtx); 1427 break; 1428 } 1429 printf("Root mount waiting for:"); 1430 LIST_FOREACH(h, &root_holds, list) 1431 printf(" %s", h->who); 1432 printf("\n"); 1433 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1434 hz); 1435 } 1436 } 1437 1438 /* 1439 * Root was mounted, share the good news. 1440 */ 1441 static void 1442 root_mount_done(void) 1443 { 1444 1445 /* 1446 * Use a mutex to prevent the wakeup being missed and waiting for 1447 * an extra 1 second sleep. 1448 */ 1449 mtx_lock(&mountlist_mtx); 1450 root_mount_complete = 1; 1451 wakeup(&root_mount_complete); 1452 mtx_unlock(&mountlist_mtx); 1453 } 1454 1455 /* 1456 * Return true if root is already mounted. 1457 */ 1458 int 1459 root_mounted(void) 1460 { 1461 1462 /* No mutex is acquired here because int stores are atomic. */ 1463 return (root_mount_complete); 1464 } 1465 1466 /* 1467 * Wait until root is mounted. 1468 */ 1469 void 1470 root_mount_wait(void) 1471 { 1472 1473 /* 1474 * Panic on an obvious deadlock - the function can't be called from 1475 * a thread which is doing the whole SYSINIT stuff. 1476 */ 1477 KASSERT(curthread->td_proc->p_pid != 0, 1478 ("root_mount_wait: cannot be called from the swapper thread")); 1479 mtx_lock(&mountlist_mtx); 1480 while (!root_mount_complete) { 1481 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1482 hz); 1483 } 1484 mtx_unlock(&mountlist_mtx); 1485 } 1486 1487 static void 1488 set_rootvnode(struct thread *td) 1489 { 1490 struct proc *p; 1491 1492 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1493 panic("Cannot find root vnode"); 1494 1495 VOP_UNLOCK(rootvnode, 0, td); 1496 1497 p = td->td_proc; 1498 FILEDESC_XLOCK(p->p_fd); 1499 1500 if (p->p_fd->fd_cdir != NULL) 1501 vrele(p->p_fd->fd_cdir); 1502 p->p_fd->fd_cdir = rootvnode; 1503 VREF(rootvnode); 1504 1505 if (p->p_fd->fd_rdir != NULL) 1506 vrele(p->p_fd->fd_rdir); 1507 p->p_fd->fd_rdir = rootvnode; 1508 VREF(rootvnode); 1509 1510 FILEDESC_XUNLOCK(p->p_fd); 1511 1512 EVENTHANDLER_INVOKE(mountroot); 1513 } 1514 1515 /* 1516 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1517 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1518 */ 1519 1520 static void 1521 devfs_first(void) 1522 { 1523 struct thread *td = curthread; 1524 struct vfsoptlist *opts; 1525 struct vfsconf *vfsp; 1526 struct mount *mp = NULL; 1527 int error; 1528 1529 vfsp = vfs_byname("devfs"); 1530 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1531 if (vfsp == NULL) 1532 return; 1533 1534 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td); 1535 1536 error = VFS_MOUNT(mp, td); 1537 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1538 if (error) 1539 return; 1540 1541 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1542 TAILQ_INIT(opts); 1543 mp->mnt_opt = opts; 1544 1545 mtx_lock(&mountlist_mtx); 1546 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1547 mtx_unlock(&mountlist_mtx); 1548 1549 set_rootvnode(td); 1550 1551 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1552 if (error) 1553 printf("kern_symlink /dev -> / returns %d\n", error); 1554 } 1555 1556 /* 1557 * Surgically move our devfs to be mounted on /dev. 1558 */ 1559 1560 static void 1561 devfs_fixup(struct thread *td) 1562 { 1563 struct nameidata nd; 1564 int error; 1565 struct vnode *vp, *dvp; 1566 struct mount *mp; 1567 1568 /* Remove our devfs mount from the mountlist and purge the cache */ 1569 mtx_lock(&mountlist_mtx); 1570 mp = TAILQ_FIRST(&mountlist); 1571 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1572 mtx_unlock(&mountlist_mtx); 1573 cache_purgevfs(mp); 1574 1575 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1576 VI_LOCK(dvp); 1577 dvp->v_iflag &= ~VI_MOUNT; 1578 VI_UNLOCK(dvp); 1579 dvp->v_mountedhere = NULL; 1580 1581 /* Set up the real rootvnode, and purge the cache */ 1582 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1583 set_rootvnode(td); 1584 cache_purgevfs(rootvnode->v_mount); 1585 1586 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1587 error = namei(&nd); 1588 if (error) { 1589 printf("Lookup of /dev for devfs, error: %d\n", error); 1590 return; 1591 } 1592 NDFREE(&nd, NDF_ONLY_PNBUF); 1593 vp = nd.ni_vp; 1594 if (vp->v_type != VDIR) { 1595 vput(vp); 1596 } 1597 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 1598 if (error) { 1599 vput(vp); 1600 } 1601 cache_purge(vp); 1602 mp->mnt_vnodecovered = vp; 1603 vp->v_mountedhere = mp; 1604 mtx_lock(&mountlist_mtx); 1605 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1606 mtx_unlock(&mountlist_mtx); 1607 VOP_UNLOCK(vp, 0, td); 1608 vput(dvp); 1609 vfs_unbusy(mp, td); 1610 1611 /* Unlink the no longer needed /dev/dev -> / symlink */ 1612 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1613 } 1614 1615 /* 1616 * Report errors during filesystem mounting. 1617 */ 1618 void 1619 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1620 { 1621 struct vfsoptlist *moptlist = mp->mnt_optnew; 1622 va_list ap; 1623 int error, len; 1624 char *errmsg; 1625 1626 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1627 if (error || errmsg == NULL || len <= 0) 1628 return; 1629 1630 va_start(ap, fmt); 1631 vsnprintf(errmsg, (size_t)len, fmt, ap); 1632 va_end(ap); 1633 } 1634 1635 /* 1636 * Find and mount the root filesystem 1637 */ 1638 void 1639 vfs_mountroot(void) 1640 { 1641 char *cp; 1642 int error, i, asked = 0; 1643 1644 root_mount_prepare(); 1645 1646 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1647 NULL, NULL, mount_init, mount_fini, 1648 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1649 devfs_first(); 1650 1651 /* 1652 * We are booted with instructions to prompt for the root filesystem. 1653 */ 1654 if (boothowto & RB_ASKNAME) { 1655 if (!vfs_mountroot_ask()) 1656 goto mounted; 1657 asked = 1; 1658 } 1659 1660 /* 1661 * The root filesystem information is compiled in, and we are 1662 * booted with instructions to use it. 1663 */ 1664 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1665 if (!vfs_mountroot_try(ctrootdevname)) 1666 goto mounted; 1667 ctrootdevname = NULL; 1668 } 1669 1670 /* 1671 * We've been given the generic "use CDROM as root" flag. This is 1672 * necessary because one media may be used in many different 1673 * devices, so we need to search for them. 1674 */ 1675 if (boothowto & RB_CDROM) { 1676 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1677 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1678 goto mounted; 1679 } 1680 } 1681 1682 /* 1683 * Try to use the value read by the loader from /etc/fstab, or 1684 * supplied via some other means. This is the preferred 1685 * mechanism. 1686 */ 1687 cp = getenv("vfs.root.mountfrom"); 1688 if (cp != NULL) { 1689 error = vfs_mountroot_try(cp); 1690 freeenv(cp); 1691 if (!error) 1692 goto mounted; 1693 } 1694 1695 /* 1696 * Try values that may have been computed by code during boot 1697 */ 1698 if (!vfs_mountroot_try(rootdevnames[0])) 1699 goto mounted; 1700 if (!vfs_mountroot_try(rootdevnames[1])) 1701 goto mounted; 1702 1703 /* 1704 * If we (still) have a compiled-in default, try it. 1705 */ 1706 if (ctrootdevname != NULL) 1707 if (!vfs_mountroot_try(ctrootdevname)) 1708 goto mounted; 1709 /* 1710 * Everything so far has failed, prompt on the console if we haven't 1711 * already tried that. 1712 */ 1713 if (!asked) 1714 if (!vfs_mountroot_ask()) 1715 goto mounted; 1716 1717 panic("Root mount failed, startup aborted."); 1718 1719 mounted: 1720 root_mount_done(); 1721 } 1722 1723 /* 1724 * Mount (mountfrom) as the root filesystem. 1725 */ 1726 static int 1727 vfs_mountroot_try(const char *mountfrom) 1728 { 1729 struct mount *mp; 1730 char *vfsname, *path; 1731 time_t timebase; 1732 int error; 1733 char patt[32]; 1734 1735 vfsname = NULL; 1736 path = NULL; 1737 mp = NULL; 1738 error = EINVAL; 1739 1740 if (mountfrom == NULL) 1741 return (error); /* don't complain */ 1742 printf("Trying to mount root from %s\n", mountfrom); 1743 1744 /* parse vfs name and path */ 1745 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1746 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1747 vfsname[0] = path[0] = 0; 1748 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1749 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1750 goto out; 1751 1752 if (path[0] == '\0') 1753 strcpy(path, ROOTNAME); 1754 1755 error = kernel_vmount( 1756 MNT_RDONLY | MNT_ROOTFS, 1757 "fstype", vfsname, 1758 "fspath", "/", 1759 "from", path, 1760 NULL); 1761 if (error == 0) { 1762 /* 1763 * We mount devfs prior to mounting the / FS, so the first 1764 * entry will typically be devfs. 1765 */ 1766 mp = TAILQ_FIRST(&mountlist); 1767 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1768 1769 /* 1770 * Iterate over all currently mounted file systems and use 1771 * the time stamp found to check and/or initialize the RTC. 1772 * Typically devfs has no time stamp and the only other FS 1773 * is the actual / FS. 1774 * Call inittodr() only once and pass it the largest of the 1775 * timestamps we encounter. 1776 */ 1777 timebase = 0; 1778 do { 1779 if (mp->mnt_time > timebase) 1780 timebase = mp->mnt_time; 1781 mp = TAILQ_NEXT(mp, mnt_list); 1782 } while (mp != NULL); 1783 inittodr(timebase); 1784 1785 devfs_fixup(curthread); 1786 } 1787 out: 1788 free(path, M_MOUNT); 1789 free(vfsname, M_MOUNT); 1790 return (error); 1791 } 1792 1793 /* 1794 * --------------------------------------------------------------------- 1795 * Interactive root filesystem selection code. 1796 */ 1797 1798 static int 1799 vfs_mountroot_ask(void) 1800 { 1801 char name[128]; 1802 1803 for(;;) { 1804 printf("\nManual root filesystem specification:\n"); 1805 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1806 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__) 1807 printf(" eg. ufs:da0s1a\n"); 1808 #else 1809 printf(" eg. ufs:/dev/da0a\n"); 1810 #endif 1811 printf(" ? List valid disk boot devices\n"); 1812 printf(" <empty line> Abort manual input\n"); 1813 printf("\nmountroot> "); 1814 gets(name, sizeof(name), 1); 1815 if (name[0] == '\0') 1816 return (1); 1817 if (name[0] == '?') { 1818 printf("\nList of GEOM managed disk devices:\n "); 1819 g_dev_print(); 1820 continue; 1821 } 1822 if (!vfs_mountroot_try(name)) 1823 return (0); 1824 } 1825 } 1826 1827 /* 1828 * --------------------------------------------------------------------- 1829 * Functions for querying mount options/arguments from filesystems. 1830 */ 1831 1832 /* 1833 * Check that no unknown options are given 1834 */ 1835 int 1836 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1837 { 1838 struct vfsopt *opt; 1839 char errmsg[255]; 1840 const char **t, *p, *q; 1841 int ret = 0; 1842 1843 TAILQ_FOREACH(opt, opts, link) { 1844 p = opt->name; 1845 q = NULL; 1846 if (p[0] == 'n' && p[1] == 'o') 1847 q = p + 2; 1848 for(t = global_opts; *t != NULL; t++) { 1849 if (strcmp(*t, p) == 0) 1850 break; 1851 if (q != NULL) { 1852 if (strcmp(*t, q) == 0) 1853 break; 1854 } 1855 } 1856 if (*t != NULL) 1857 continue; 1858 for(t = legal; *t != NULL; t++) { 1859 if (strcmp(*t, p) == 0) 1860 break; 1861 if (q != NULL) { 1862 if (strcmp(*t, q) == 0) 1863 break; 1864 } 1865 } 1866 if (*t != NULL) 1867 continue; 1868 snprintf(errmsg, sizeof(errmsg), 1869 "mount option <%s> is unknown", p); 1870 printf("%s\n", errmsg); 1871 ret = EINVAL; 1872 } 1873 if (ret != 0) { 1874 TAILQ_FOREACH(opt, opts, link) { 1875 if (strcmp(opt->name, "errmsg") == 0) { 1876 strncpy((char *)opt->value, errmsg, opt->len); 1877 } 1878 } 1879 } 1880 return (ret); 1881 } 1882 1883 /* 1884 * Get a mount option by its name. 1885 * 1886 * Return 0 if the option was found, ENOENT otherwise. 1887 * If len is non-NULL it will be filled with the length 1888 * of the option. If buf is non-NULL, it will be filled 1889 * with the address of the option. 1890 */ 1891 int 1892 vfs_getopt(opts, name, buf, len) 1893 struct vfsoptlist *opts; 1894 const char *name; 1895 void **buf; 1896 int *len; 1897 { 1898 struct vfsopt *opt; 1899 1900 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1901 1902 TAILQ_FOREACH(opt, opts, link) { 1903 if (strcmp(name, opt->name) == 0) { 1904 if (len != NULL) 1905 *len = opt->len; 1906 if (buf != NULL) 1907 *buf = opt->value; 1908 return (0); 1909 } 1910 } 1911 return (ENOENT); 1912 } 1913 1914 static int 1915 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1916 { 1917 struct vfsopt *opt; 1918 int i; 1919 1920 if (opts == NULL) 1921 return (-1); 1922 1923 i = 0; 1924 TAILQ_FOREACH(opt, opts, link) { 1925 if (strcmp(name, opt->name) == 0) 1926 return (i); 1927 ++i; 1928 } 1929 return (-1); 1930 } 1931 1932 char * 1933 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1934 { 1935 struct vfsopt *opt; 1936 1937 *error = 0; 1938 TAILQ_FOREACH(opt, opts, link) { 1939 if (strcmp(name, opt->name) != 0) 1940 continue; 1941 if (((char *)opt->value)[opt->len - 1] != '\0') { 1942 *error = EINVAL; 1943 return (NULL); 1944 } 1945 return (opt->value); 1946 } 1947 *error = ENOENT; 1948 return (NULL); 1949 } 1950 1951 int 1952 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1953 { 1954 struct vfsopt *opt; 1955 1956 TAILQ_FOREACH(opt, opts, link) { 1957 if (strcmp(name, opt->name) == 0) { 1958 if (w != NULL) 1959 *w |= val; 1960 return (1); 1961 } 1962 } 1963 if (w != NULL) 1964 *w &= ~val; 1965 return (0); 1966 } 1967 1968 int 1969 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1970 { 1971 va_list ap; 1972 struct vfsopt *opt; 1973 int ret; 1974 1975 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1976 1977 TAILQ_FOREACH(opt, opts, link) { 1978 if (strcmp(name, opt->name) != 0) 1979 continue; 1980 if (opt->len == 0 || opt->value == NULL) 1981 return (0); 1982 if (((char *)opt->value)[opt->len - 1] != '\0') 1983 return (0); 1984 va_start(ap, fmt); 1985 ret = vsscanf(opt->value, fmt, ap); 1986 va_end(ap); 1987 return (ret); 1988 } 1989 return (0); 1990 } 1991 1992 /* 1993 * Find and copy a mount option. 1994 * 1995 * The size of the buffer has to be specified 1996 * in len, if it is not the same length as the 1997 * mount option, EINVAL is returned. 1998 * Returns ENOENT if the option is not found. 1999 */ 2000 int 2001 vfs_copyopt(opts, name, dest, len) 2002 struct vfsoptlist *opts; 2003 const char *name; 2004 void *dest; 2005 int len; 2006 { 2007 struct vfsopt *opt; 2008 2009 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2010 2011 TAILQ_FOREACH(opt, opts, link) { 2012 if (strcmp(name, opt->name) == 0) { 2013 if (len != opt->len) 2014 return (EINVAL); 2015 bcopy(opt->value, dest, opt->len); 2016 return (0); 2017 } 2018 } 2019 return (ENOENT); 2020 } 2021 2022 /* 2023 * This is a helper function for filesystems to traverse their 2024 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2025 */ 2026 2027 struct vnode * 2028 __mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2029 { 2030 struct vnode *vp; 2031 2032 mtx_assert(MNT_MTX(mp), MA_OWNED); 2033 2034 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2035 if ((*mvp)->v_yield++ == 500) { 2036 MNT_IUNLOCK(mp); 2037 (*mvp)->v_yield = 0; 2038 uio_yield(); 2039 MNT_ILOCK(mp); 2040 } 2041 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2042 while (vp != NULL && vp->v_type == VMARKER) 2043 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2044 2045 /* Check if we are done */ 2046 if (vp == NULL) { 2047 __mnt_vnode_markerfree(mvp, mp); 2048 return (NULL); 2049 } 2050 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2051 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2052 return (vp); 2053 } 2054 2055 struct vnode * 2056 __mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2057 { 2058 struct vnode *vp; 2059 2060 mtx_assert(MNT_MTX(mp), MA_OWNED); 2061 2062 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2063 while (vp != NULL && vp->v_type == VMARKER) 2064 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2065 2066 /* Check if we are done */ 2067 if (vp == NULL) { 2068 *mvp = NULL; 2069 return (NULL); 2070 } 2071 mp->mnt_holdcnt++; 2072 MNT_IUNLOCK(mp); 2073 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2074 M_VNODE_MARKER, 2075 M_WAITOK | M_ZERO); 2076 MNT_ILOCK(mp); 2077 (*mvp)->v_type = VMARKER; 2078 2079 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2080 while (vp != NULL && vp->v_type == VMARKER) 2081 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2082 2083 /* Check if we are done */ 2084 if (vp == NULL) { 2085 MNT_IUNLOCK(mp); 2086 free(*mvp, M_VNODE_MARKER); 2087 MNT_ILOCK(mp); 2088 *mvp = NULL; 2089 mp->mnt_holdcnt--; 2090 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2091 wakeup(&mp->mnt_holdcnt); 2092 return (NULL); 2093 } 2094 mp->mnt_markercnt++; 2095 (*mvp)->v_mount = mp; 2096 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2097 return (vp); 2098 } 2099 2100 2101 void 2102 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2103 { 2104 2105 if (*mvp == NULL) 2106 return; 2107 2108 mtx_assert(MNT_MTX(mp), MA_OWNED); 2109 2110 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2111 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2112 MNT_IUNLOCK(mp); 2113 free(*mvp, M_VNODE_MARKER); 2114 MNT_ILOCK(mp); 2115 *mvp = NULL; 2116 2117 mp->mnt_markercnt--; 2118 mp->mnt_holdcnt--; 2119 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2120 wakeup(&mp->mnt_holdcnt); 2121 } 2122 2123 2124 int 2125 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 2126 { 2127 int error; 2128 2129 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 2130 if (sbp != &mp->mnt_stat) 2131 *sbp = mp->mnt_stat; 2132 return (error); 2133 } 2134 2135 void 2136 vfs_mountedfrom(struct mount *mp, const char *from) 2137 { 2138 2139 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2140 strlcpy(mp->mnt_stat.f_mntfromname, from, 2141 sizeof mp->mnt_stat.f_mntfromname); 2142 } 2143 2144 /* 2145 * --------------------------------------------------------------------- 2146 * This is the api for building mount args and mounting filesystems from 2147 * inside the kernel. 2148 * 2149 * The API works by accumulation of individual args. First error is 2150 * latched. 2151 * 2152 * XXX: should be documented in new manpage kernel_mount(9) 2153 */ 2154 2155 /* A memory allocation which must be freed when we are done */ 2156 struct mntaarg { 2157 SLIST_ENTRY(mntaarg) next; 2158 }; 2159 2160 /* The header for the mount arguments */ 2161 struct mntarg { 2162 struct iovec *v; 2163 int len; 2164 int error; 2165 SLIST_HEAD(, mntaarg) list; 2166 }; 2167 2168 /* 2169 * Add a boolean argument. 2170 * 2171 * flag is the boolean value. 2172 * name must start with "no". 2173 */ 2174 struct mntarg * 2175 mount_argb(struct mntarg *ma, int flag, const char *name) 2176 { 2177 2178 KASSERT(name[0] == 'n' && name[1] == 'o', 2179 ("mount_argb(...,%s): name must start with 'no'", name)); 2180 2181 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2182 } 2183 2184 /* 2185 * Add an argument printf style 2186 */ 2187 struct mntarg * 2188 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2189 { 2190 va_list ap; 2191 struct mntaarg *maa; 2192 struct sbuf *sb; 2193 int len; 2194 2195 if (ma == NULL) { 2196 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2197 SLIST_INIT(&ma->list); 2198 } 2199 if (ma->error) 2200 return (ma); 2201 2202 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2203 M_MOUNT, M_WAITOK); 2204 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2205 ma->v[ma->len].iov_len = strlen(name) + 1; 2206 ma->len++; 2207 2208 sb = sbuf_new_auto(); 2209 va_start(ap, fmt); 2210 sbuf_vprintf(sb, fmt, ap); 2211 va_end(ap); 2212 sbuf_finish(sb); 2213 len = sbuf_len(sb) + 1; 2214 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2215 SLIST_INSERT_HEAD(&ma->list, maa, next); 2216 bcopy(sbuf_data(sb), maa + 1, len); 2217 sbuf_delete(sb); 2218 2219 ma->v[ma->len].iov_base = maa + 1; 2220 ma->v[ma->len].iov_len = len; 2221 ma->len++; 2222 2223 return (ma); 2224 } 2225 2226 /* 2227 * Add an argument which is a userland string. 2228 */ 2229 struct mntarg * 2230 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2231 { 2232 struct mntaarg *maa; 2233 char *tbuf; 2234 2235 if (val == NULL) 2236 return (ma); 2237 if (ma == NULL) { 2238 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2239 SLIST_INIT(&ma->list); 2240 } 2241 if (ma->error) 2242 return (ma); 2243 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2244 SLIST_INSERT_HEAD(&ma->list, maa, next); 2245 tbuf = (void *)(maa + 1); 2246 ma->error = copyinstr(val, tbuf, len, NULL); 2247 return (mount_arg(ma, name, tbuf, -1)); 2248 } 2249 2250 /* 2251 * Plain argument. 2252 * 2253 * If length is -1, use printf. 2254 */ 2255 struct mntarg * 2256 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2257 { 2258 2259 if (ma == NULL) { 2260 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2261 SLIST_INIT(&ma->list); 2262 } 2263 if (ma->error) 2264 return (ma); 2265 2266 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2267 M_MOUNT, M_WAITOK); 2268 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2269 ma->v[ma->len].iov_len = strlen(name) + 1; 2270 ma->len++; 2271 2272 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2273 if (len < 0) 2274 ma->v[ma->len].iov_len = strlen(val) + 1; 2275 else 2276 ma->v[ma->len].iov_len = len; 2277 ma->len++; 2278 return (ma); 2279 } 2280 2281 /* 2282 * Free a mntarg structure 2283 */ 2284 static void 2285 free_mntarg(struct mntarg *ma) 2286 { 2287 struct mntaarg *maa; 2288 2289 while (!SLIST_EMPTY(&ma->list)) { 2290 maa = SLIST_FIRST(&ma->list); 2291 SLIST_REMOVE_HEAD(&ma->list, next); 2292 free(maa, M_MOUNT); 2293 } 2294 free(ma->v, M_MOUNT); 2295 free(ma, M_MOUNT); 2296 } 2297 2298 /* 2299 * Mount a filesystem 2300 */ 2301 int 2302 kernel_mount(struct mntarg *ma, int flags) 2303 { 2304 struct uio auio; 2305 int error; 2306 2307 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2308 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2309 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2310 2311 auio.uio_iov = ma->v; 2312 auio.uio_iovcnt = ma->len; 2313 auio.uio_segflg = UIO_SYSSPACE; 2314 2315 error = ma->error; 2316 if (!error) 2317 error = vfs_donmount(curthread, flags, &auio); 2318 free_mntarg(ma); 2319 return (error); 2320 } 2321 2322 /* 2323 * A printflike function to mount a filesystem. 2324 */ 2325 int 2326 kernel_vmount(int flags, ...) 2327 { 2328 struct mntarg *ma = NULL; 2329 va_list ap; 2330 const char *cp; 2331 const void *vp; 2332 int error; 2333 2334 va_start(ap, flags); 2335 for (;;) { 2336 cp = va_arg(ap, const char *); 2337 if (cp == NULL) 2338 break; 2339 vp = va_arg(ap, const void *); 2340 ma = mount_arg(ma, cp, vp, -1); 2341 } 2342 va_end(ap); 2343 2344 error = kernel_mount(ma, flags); 2345 return (error); 2346 }
Cache object: 9ae70ff5dd77eb97c6977c9bdb645707
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]