vfs_mount.c revision 184554
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: head/sys/kern/vfs_mount.c 184554 2008-11-02 10:15:42Z attilio $"); 39 40#include <sys/param.h> 41#include <sys/conf.h> 42#include <sys/fcntl.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_mac.h" 72 73#define ROOTNAME "root_device" 74#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 75 76static int vfs_domount(struct thread *td, const char *fstype, 77 char *fspath, int fsflags, void *fsdata); 78static int vfs_mountroot_ask(void); 79static int vfs_mountroot_try(const char *mountfrom); 80static void free_mntarg(struct mntarg *ma); 81static int vfs_getopt_pos(struct vfsoptlist *opts, const char *name); 82 83static int usermount = 0; 84SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 85 "Unprivileged users may mount and unmount file systems"); 86 87MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 88MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 89static uma_zone_t mount_zone; 90 91/* List of mounted filesystems. */ 92struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 93 94/* For any iteration/modification of mountlist */ 95struct mtx mountlist_mtx; 96MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 97 98TAILQ_HEAD(vfsoptlist, vfsopt); 99struct vfsopt { 100 TAILQ_ENTRY(vfsopt) link; 101 char *name; 102 void *value; 103 int len; 104}; 105 106/* 107 * The vnode of the system's root (/ in the filesystem, without chroot 108 * active.) 109 */ 110struct vnode *rootvnode; 111 112/* 113 * The root filesystem is detailed in the kernel environment variable 114 * vfs.root.mountfrom, which is expected to be in the general format 115 * 116 * <vfsname>:[<path>] 117 * vfsname := the name of a VFS known to the kernel and capable 118 * of being mounted as root 119 * path := disk device name or other data used by the filesystem 120 * to locate its physical store 121 */ 122 123/* 124 * Global opts, taken by all filesystems 125 */ 126static const char *global_opts[] = { 127 "errmsg", 128 "fstype", 129 "fspath", 130 "ro", 131 "rw", 132 "nosuid", 133 "noexec", 134 NULL 135}; 136 137/* 138 * The root specifiers we will try if RB_CDROM is specified. 139 */ 140static char *cdrom_rootdevnames[] = { 141 "cd9660:cd0", 142 "cd9660:acd0", 143 NULL 144}; 145 146/* legacy find-root code */ 147char *rootdevnames[2] = {NULL, NULL}; 148#ifndef ROOTDEVNAME 149# define ROOTDEVNAME NULL 150#endif 151static const char *ctrootdevname = ROOTDEVNAME; 152 153/* 154 * --------------------------------------------------------------------- 155 * Functions for building and sanitizing the mount options 156 */ 157 158/* Remove one mount option. */ 159static void 160vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 161{ 162 163 TAILQ_REMOVE(opts, opt, link); 164 free(opt->name, M_MOUNT); 165 if (opt->value != NULL) 166 free(opt->value, M_MOUNT); 167#ifdef INVARIANTS 168 else if (opt->len != 0) 169 panic("%s: mount option with NULL value but length != 0", 170 __func__); 171#endif 172 free(opt, M_MOUNT); 173} 174 175/* Release all resources related to the mount options. */ 176void 177vfs_freeopts(struct vfsoptlist *opts) 178{ 179 struct vfsopt *opt; 180 181 while (!TAILQ_EMPTY(opts)) { 182 opt = TAILQ_FIRST(opts); 183 vfs_freeopt(opts, opt); 184 } 185 free(opts, M_MOUNT); 186} 187 188void 189vfs_deleteopt(struct vfsoptlist *opts, const char *name) 190{ 191 struct vfsopt *opt, *temp; 192 193 if (opts == NULL) 194 return; 195 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 196 if (strcmp(opt->name, name) == 0) 197 vfs_freeopt(opts, opt); 198 } 199} 200 201/* 202 * Check if options are equal (with or without the "no" prefix). 203 */ 204static int 205vfs_equalopts(const char *opt1, const char *opt2) 206{ 207 208 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 209 if (strcmp(opt1, opt2) == 0) 210 return (1); 211 /* "noopt" vs. "opt" */ 212 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 213 return (1); 214 /* "opt" vs. "noopt" */ 215 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 216 return (1); 217 return (0); 218} 219 220/* 221 * If a mount option is specified several times, 222 * (with or without the "no" prefix) only keep 223 * the last occurence of it. 224 */ 225static void 226vfs_sanitizeopts(struct vfsoptlist *opts) 227{ 228 struct vfsopt *opt, *opt2, *tmp; 229 230 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 231 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 232 while (opt2 != NULL) { 233 if (vfs_equalopts(opt->name, opt2->name)) { 234 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 235 vfs_freeopt(opts, opt2); 236 opt2 = tmp; 237 } else { 238 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 239 } 240 } 241 } 242} 243 244/* 245 * Build a linked list of mount options from a struct uio. 246 */ 247static int 248vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 249{ 250 struct vfsoptlist *opts; 251 struct vfsopt *opt; 252 size_t memused; 253 unsigned int i, iovcnt; 254 int error, namelen, optlen; 255 256 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 257 TAILQ_INIT(opts); 258 memused = 0; 259 iovcnt = auio->uio_iovcnt; 260 for (i = 0; i < iovcnt; i += 2) { 261 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 262 namelen = auio->uio_iov[i].iov_len; 263 optlen = auio->uio_iov[i + 1].iov_len; 264 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 265 opt->value = NULL; 266 opt->len = 0; 267 268 /* 269 * Do this early, so jumps to "bad" will free the current 270 * option. 271 */ 272 TAILQ_INSERT_TAIL(opts, opt, link); 273 memused += sizeof(struct vfsopt) + optlen + namelen; 274 275 /* 276 * Avoid consuming too much memory, and attempts to overflow 277 * memused. 278 */ 279 if (memused > VFS_MOUNTARG_SIZE_MAX || 280 optlen > VFS_MOUNTARG_SIZE_MAX || 281 namelen > VFS_MOUNTARG_SIZE_MAX) { 282 error = EINVAL; 283 goto bad; 284 } 285 286 if (auio->uio_segflg == UIO_SYSSPACE) { 287 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 288 } else { 289 error = copyin(auio->uio_iov[i].iov_base, opt->name, 290 namelen); 291 if (error) 292 goto bad; 293 } 294 /* Ensure names are null-terminated strings. */ 295 if (opt->name[namelen - 1] != '\0') { 296 error = EINVAL; 297 goto bad; 298 } 299 if (optlen != 0) { 300 opt->len = optlen; 301 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 302 if (auio->uio_segflg == UIO_SYSSPACE) { 303 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 304 optlen); 305 } else { 306 error = copyin(auio->uio_iov[i + 1].iov_base, 307 opt->value, optlen); 308 if (error) 309 goto bad; 310 } 311 } 312 } 313 vfs_sanitizeopts(opts); 314 *options = opts; 315 return (0); 316bad: 317 vfs_freeopts(opts); 318 return (error); 319} 320 321/* 322 * Merge the old mount options with the new ones passed 323 * in the MNT_UPDATE case. 324 * 325 * XXX This function will keep a "nofoo" option in the 326 * new options if there is no matching "foo" option 327 * to be cancelled in the old options. This is a bug 328 * if the option's canonical name is "foo". E.g., "noro" 329 * shouldn't end up in the mount point's active options, 330 * but it can. 331 */ 332static void 333vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 334{ 335 struct vfsopt *opt, *opt2, *new; 336 337 TAILQ_FOREACH(opt, opts, link) { 338 /* 339 * Check that this option hasn't been redefined 340 * nor cancelled with a "no" mount option. 341 */ 342 opt2 = TAILQ_FIRST(toopts); 343 while (opt2 != NULL) { 344 if (strcmp(opt2->name, opt->name) == 0) 345 goto next; 346 if (strncmp(opt2->name, "no", 2) == 0 && 347 strcmp(opt2->name + 2, opt->name) == 0) { 348 vfs_freeopt(toopts, opt2); 349 goto next; 350 } 351 opt2 = TAILQ_NEXT(opt2, link); 352 } 353 /* We want this option, duplicate it. */ 354 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 355 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 356 strcpy(new->name, opt->name); 357 if (opt->len != 0) { 358 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 359 bcopy(opt->value, new->value, opt->len); 360 } else { 361 new->value = NULL; 362 } 363 new->len = opt->len; 364 TAILQ_INSERT_TAIL(toopts, new, link); 365next: 366 continue; 367 } 368} 369 370/* 371 * Mount a filesystem. 372 */ 373int 374nmount(td, uap) 375 struct thread *td; 376 struct nmount_args /* { 377 struct iovec *iovp; 378 unsigned int iovcnt; 379 int flags; 380 } */ *uap; 381{ 382 struct uio *auio; 383 struct iovec *iov; 384 unsigned int i; 385 int error; 386 u_int iovcnt; 387 388 AUDIT_ARG(fflags, uap->flags); 389 390 /* 391 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 392 * userspace to set this flag, but we must filter it out if we want 393 * MNT_UPDATE on the root file system to work. 394 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 395 */ 396 uap->flags &= ~MNT_ROOTFS; 397 398 iovcnt = uap->iovcnt; 399 /* 400 * Check that we have an even number of iovec's 401 * and that we have at least two options. 402 */ 403 if ((iovcnt & 1) || (iovcnt < 4)) 404 return (EINVAL); 405 406 error = copyinuio(uap->iovp, iovcnt, &auio); 407 if (error) 408 return (error); 409 iov = auio->uio_iov; 410 for (i = 0; i < iovcnt; i++) { 411 if (iov->iov_len > MMAXOPTIONLEN) { 412 free(auio, M_IOV); 413 return (EINVAL); 414 } 415 iov++; 416 } 417 error = vfs_donmount(td, uap->flags, auio); 418 419 free(auio, M_IOV); 420 return (error); 421} 422 423/* 424 * --------------------------------------------------------------------- 425 * Various utility functions 426 */ 427 428void 429vfs_ref(struct mount *mp) 430{ 431 432 MNT_ILOCK(mp); 433 MNT_REF(mp); 434 MNT_IUNLOCK(mp); 435} 436 437void 438vfs_rel(struct mount *mp) 439{ 440 441 MNT_ILOCK(mp); 442 MNT_REL(mp); 443 MNT_IUNLOCK(mp); 444} 445 446static int 447mount_init(void *mem, int size, int flags) 448{ 449 struct mount *mp; 450 451 mp = (struct mount *)mem; 452 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 453 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 454 return (0); 455} 456 457static void 458mount_fini(void *mem, int size) 459{ 460 struct mount *mp; 461 462 mp = (struct mount *)mem; 463 lockdestroy(&mp->mnt_explock); 464 mtx_destroy(&mp->mnt_mtx); 465} 466 467/* 468 * Allocate and initialize the mount point struct. 469 */ 470struct mount * 471vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 472 struct ucred *cred) 473{ 474 struct mount *mp; 475 476 mp = uma_zalloc(mount_zone, M_WAITOK); 477 bzero(&mp->mnt_startzero, 478 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 479 TAILQ_INIT(&mp->mnt_nvnodelist); 480 mp->mnt_nvnodelistsize = 0; 481 mp->mnt_ref = 0; 482 (void) vfs_busy(mp, MBF_NOWAIT); 483 mp->mnt_op = vfsp->vfc_vfsops; 484 mp->mnt_vfc = vfsp; 485 vfsp->vfc_refcount++; /* XXX Unlocked */ 486 mp->mnt_stat.f_type = vfsp->vfc_typenum; 487 mp->mnt_gen++; 488 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 489 mp->mnt_vnodecovered = vp; 490 mp->mnt_cred = crdup(cred); 491 mp->mnt_stat.f_owner = cred->cr_uid; 492 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 493 mp->mnt_iosize_max = DFLTPHYS; 494#ifdef MAC 495 mac_mount_init(mp); 496 mac_mount_create(cred, mp); 497#endif 498 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 499 return (mp); 500} 501 502/* 503 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 504 */ 505void 506vfs_mount_destroy(struct mount *mp) 507{ 508 509 MNT_ILOCK(mp); 510 while (mp->mnt_ref) 511 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 512 if (mp->mnt_holdcnt != 0) { 513 printf("Waiting for mount point to be unheld\n"); 514 while (mp->mnt_holdcnt != 0) { 515 mp->mnt_holdcntwaiters++; 516 msleep(&mp->mnt_holdcnt, MNT_MTX(mp), 517 PZERO, "mntdestroy", 0); 518 mp->mnt_holdcntwaiters--; 519 } 520 printf("mount point unheld\n"); 521 } 522 if (mp->mnt_writeopcount > 0) { 523 printf("Waiting for mount point write ops\n"); 524 while (mp->mnt_writeopcount > 0) { 525 mp->mnt_kern_flag |= MNTK_SUSPEND; 526 msleep(&mp->mnt_writeopcount, 527 MNT_MTX(mp), 528 PZERO, "mntdestroy2", 0); 529 } 530 printf("mount point write ops completed\n"); 531 } 532 if (mp->mnt_secondary_writes > 0) { 533 printf("Waiting for mount point secondary write ops\n"); 534 while (mp->mnt_secondary_writes > 0) { 535 mp->mnt_kern_flag |= MNTK_SUSPEND; 536 msleep(&mp->mnt_secondary_writes, 537 MNT_MTX(mp), 538 PZERO, "mntdestroy3", 0); 539 } 540 printf("mount point secondary write ops completed\n"); 541 } 542 MNT_IUNLOCK(mp); 543 mp->mnt_vfc->vfc_refcount--; 544 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 545 struct vnode *vp; 546 547 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 548 vprint("", vp); 549 panic("unmount: dangling vnode"); 550 } 551 MNT_ILOCK(mp); 552 if (mp->mnt_kern_flag & MNTK_MWAIT) 553 wakeup(mp); 554 if (mp->mnt_writeopcount != 0) 555 panic("vfs_mount_destroy: nonzero writeopcount"); 556 if (mp->mnt_secondary_writes != 0) 557 panic("vfs_mount_destroy: nonzero secondary_writes"); 558 if (mp->mnt_nvnodelistsize != 0) 559 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 560 mp->mnt_writeopcount = -1000; 561 mp->mnt_nvnodelistsize = -1000; 562 mp->mnt_secondary_writes = -1000; 563 MNT_IUNLOCK(mp); 564#ifdef MAC 565 mac_mount_destroy(mp); 566#endif 567 if (mp->mnt_opt != NULL) 568 vfs_freeopts(mp->mnt_opt); 569 crfree(mp->mnt_cred); 570 uma_zfree(mount_zone, mp); 571} 572 573int 574vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 575{ 576 struct vfsoptlist *optlist; 577 struct vfsopt *opt, *noro_opt, *tmp_opt; 578 char *fstype, *fspath, *errmsg; 579 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 580 int has_rw, has_noro; 581 582 errmsg = fspath = NULL; 583 errmsg_len = has_noro = has_rw = fspathlen = 0; 584 errmsg_pos = -1; 585 586 error = vfs_buildopts(fsoptions, &optlist); 587 if (error) 588 return (error); 589 590 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 591 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 592 593 /* 594 * We need these two options before the others, 595 * and they are mandatory for any filesystem. 596 * Ensure they are NUL terminated as well. 597 */ 598 fstypelen = 0; 599 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 600 if (error || fstype[fstypelen - 1] != '\0') { 601 error = EINVAL; 602 if (errmsg != NULL) 603 strncpy(errmsg, "Invalid fstype", errmsg_len); 604 goto bail; 605 } 606 fspathlen = 0; 607 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 608 if (error || fspath[fspathlen - 1] != '\0') { 609 error = EINVAL; 610 if (errmsg != NULL) 611 strncpy(errmsg, "Invalid fspath", errmsg_len); 612 goto bail; 613 } 614 615 /* 616 * We need to see if we have the "update" option 617 * before we call vfs_domount(), since vfs_domount() has special 618 * logic based on MNT_UPDATE. This is very important 619 * when we want to update the root filesystem. 620 */ 621 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 622 if (strcmp(opt->name, "update") == 0) { 623 fsflags |= MNT_UPDATE; 624 vfs_freeopt(optlist, opt); 625 } 626 else if (strcmp(opt->name, "async") == 0) 627 fsflags |= MNT_ASYNC; 628 else if (strcmp(opt->name, "force") == 0) { 629 fsflags |= MNT_FORCE; 630 vfs_freeopt(optlist, opt); 631 } 632 else if (strcmp(opt->name, "reload") == 0) { 633 fsflags |= MNT_RELOAD; 634 vfs_freeopt(optlist, opt); 635 } 636 else if (strcmp(opt->name, "multilabel") == 0) 637 fsflags |= MNT_MULTILABEL; 638 else if (strcmp(opt->name, "noasync") == 0) 639 fsflags &= ~MNT_ASYNC; 640 else if (strcmp(opt->name, "noatime") == 0) 641 fsflags |= MNT_NOATIME; 642 else if (strcmp(opt->name, "atime") == 0) { 643 free(opt->name, M_MOUNT); 644 opt->name = strdup("nonoatime", M_MOUNT); 645 } 646 else if (strcmp(opt->name, "noclusterr") == 0) 647 fsflags |= MNT_NOCLUSTERR; 648 else if (strcmp(opt->name, "clusterr") == 0) { 649 free(opt->name, M_MOUNT); 650 opt->name = strdup("nonoclusterr", M_MOUNT); 651 } 652 else if (strcmp(opt->name, "noclusterw") == 0) 653 fsflags |= MNT_NOCLUSTERW; 654 else if (strcmp(opt->name, "clusterw") == 0) { 655 free(opt->name, M_MOUNT); 656 opt->name = strdup("nonoclusterw", M_MOUNT); 657 } 658 else if (strcmp(opt->name, "noexec") == 0) 659 fsflags |= MNT_NOEXEC; 660 else if (strcmp(opt->name, "exec") == 0) { 661 free(opt->name, M_MOUNT); 662 opt->name = strdup("nonoexec", M_MOUNT); 663 } 664 else if (strcmp(opt->name, "nosuid") == 0) 665 fsflags |= MNT_NOSUID; 666 else if (strcmp(opt->name, "suid") == 0) { 667 free(opt->name, M_MOUNT); 668 opt->name = strdup("nonosuid", M_MOUNT); 669 } 670 else if (strcmp(opt->name, "nosymfollow") == 0) 671 fsflags |= MNT_NOSYMFOLLOW; 672 else if (strcmp(opt->name, "symfollow") == 0) { 673 free(opt->name, M_MOUNT); 674 opt->name = strdup("nonosymfollow", M_MOUNT); 675 } 676 else if (strcmp(opt->name, "noro") == 0) { 677 fsflags &= ~MNT_RDONLY; 678 has_noro = 1; 679 } 680 else if (strcmp(opt->name, "rw") == 0) { 681 fsflags &= ~MNT_RDONLY; 682 has_rw = 1; 683 } 684 else if (strcmp(opt->name, "ro") == 0) 685 fsflags |= MNT_RDONLY; 686 else if (strcmp(opt->name, "rdonly") == 0) { 687 free(opt->name, M_MOUNT); 688 opt->name = strdup("ro", M_MOUNT); 689 fsflags |= MNT_RDONLY; 690 } 691 else if (strcmp(opt->name, "suiddir") == 0) 692 fsflags |= MNT_SUIDDIR; 693 else if (strcmp(opt->name, "sync") == 0) 694 fsflags |= MNT_SYNCHRONOUS; 695 else if (strcmp(opt->name, "union") == 0) 696 fsflags |= MNT_UNION; 697 } 698 699 /* 700 * If "rw" was specified as a mount option, and we 701 * are trying to update a mount-point from "ro" to "rw", 702 * we need a mount option "noro", since in vfs_mergeopts(), 703 * "noro" will cancel "ro", but "rw" will not do anything. 704 */ 705 if (has_rw && !has_noro) { 706 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 707 noro_opt->name = strdup("noro", M_MOUNT); 708 noro_opt->value = NULL; 709 noro_opt->len = 0; 710 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 711 } 712 713 /* 714 * Be ultra-paranoid about making sure the type and fspath 715 * variables will fit in our mp buffers, including the 716 * terminating NUL. 717 */ 718 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 719 error = ENAMETOOLONG; 720 goto bail; 721 } 722 723 mtx_lock(&Giant); 724 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 725 mtx_unlock(&Giant); 726bail: 727 /* copyout the errmsg */ 728 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 729 && errmsg_len > 0 && errmsg != NULL) { 730 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 731 bcopy(errmsg, 732 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 733 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 734 } else { 735 copyout(errmsg, 736 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 737 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 738 } 739 } 740 741 if (error != 0) 742 vfs_freeopts(optlist); 743 return (error); 744} 745 746/* 747 * Old mount API. 748 */ 749#ifndef _SYS_SYSPROTO_H_ 750struct mount_args { 751 char *type; 752 char *path; 753 int flags; 754 caddr_t data; 755}; 756#endif 757/* ARGSUSED */ 758int 759mount(td, uap) 760 struct thread *td; 761 struct mount_args /* { 762 char *type; 763 char *path; 764 int flags; 765 caddr_t data; 766 } */ *uap; 767{ 768 char *fstype; 769 struct vfsconf *vfsp = NULL; 770 struct mntarg *ma = NULL; 771 int error; 772 773 AUDIT_ARG(fflags, uap->flags); 774 775 /* 776 * Filter out MNT_ROOTFS. We do not want clients of mount() in 777 * userspace to set this flag, but we must filter it out if we want 778 * MNT_UPDATE on the root file system to work. 779 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 780 */ 781 uap->flags &= ~MNT_ROOTFS; 782 783 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 784 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 785 if (error) { 786 free(fstype, M_TEMP); 787 return (error); 788 } 789 790 AUDIT_ARG(text, fstype); 791 mtx_lock(&Giant); 792 vfsp = vfs_byname_kld(fstype, td, &error); 793 free(fstype, M_TEMP); 794 if (vfsp == NULL) { 795 mtx_unlock(&Giant); 796 return (ENOENT); 797 } 798 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 799 mtx_unlock(&Giant); 800 return (EOPNOTSUPP); 801 } 802 803 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 804 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 805 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 806 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 807 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 808 809 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 810 mtx_unlock(&Giant); 811 return (error); 812} 813 814 815/* 816 * vfs_domount(): actually attempt a filesystem mount. 817 */ 818static int 819vfs_domount( 820 struct thread *td, /* Calling thread. */ 821 const char *fstype, /* Filesystem type. */ 822 char *fspath, /* Mount path. */ 823 int fsflags, /* Flags common to all filesystems. */ 824 void *fsdata /* Options local to the filesystem. */ 825 ) 826{ 827 struct vnode *vp; 828 struct mount *mp; 829 struct vfsconf *vfsp; 830 struct export_args export; 831 int error, flag = 0; 832 struct vattr va; 833 struct nameidata nd; 834 835 mtx_assert(&Giant, MA_OWNED); 836 /* 837 * Be ultra-paranoid about making sure the type and fspath 838 * variables will fit in our mp buffers, including the 839 * terminating NUL. 840 */ 841 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 842 return (ENAMETOOLONG); 843 844 if (jailed(td->td_ucred) || usermount == 0) { 845 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 846 return (error); 847 } 848 849 /* 850 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 851 */ 852 if (fsflags & MNT_EXPORTED) { 853 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 854 if (error) 855 return (error); 856 } 857 if (fsflags & MNT_SUIDDIR) { 858 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 859 if (error) 860 return (error); 861 } 862 /* 863 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 864 */ 865 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 866 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 867 fsflags |= MNT_NOSUID | MNT_USER; 868 } 869 870 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 871 vfsp = NULL; 872 if ((fsflags & MNT_UPDATE) == 0) { 873 /* Don't try to load KLDs if we're mounting the root. */ 874 if (fsflags & MNT_ROOTFS) 875 vfsp = vfs_byname(fstype); 876 else 877 vfsp = vfs_byname_kld(fstype, td, &error); 878 if (vfsp == NULL) 879 return (ENODEV); 880 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 881 return (EPERM); 882 } 883 /* 884 * Get vnode to be covered 885 */ 886 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 887 fspath, td); 888 if ((error = namei(&nd)) != 0) 889 return (error); 890 NDFREE(&nd, NDF_ONLY_PNBUF); 891 vp = nd.ni_vp; 892 if (fsflags & MNT_UPDATE) { 893 if ((vp->v_vflag & VV_ROOT) == 0) { 894 vput(vp); 895 return (EINVAL); 896 } 897 mp = vp->v_mount; 898 MNT_ILOCK(mp); 899 flag = mp->mnt_flag; 900 /* 901 * We only allow the filesystem to be reloaded if it 902 * is currently mounted read-only. 903 */ 904 if ((fsflags & MNT_RELOAD) && 905 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 906 MNT_IUNLOCK(mp); 907 vput(vp); 908 return (EOPNOTSUPP); /* Needs translation */ 909 } 910 MNT_IUNLOCK(mp); 911 /* 912 * Only privileged root, or (if MNT_USER is set) the user that 913 * did the original mount is permitted to update it. 914 */ 915 error = vfs_suser(mp, td); 916 if (error) { 917 vput(vp); 918 return (error); 919 } 920 if (vfs_busy(mp, MBF_NOWAIT)) { 921 vput(vp); 922 return (EBUSY); 923 } 924 VI_LOCK(vp); 925 if ((vp->v_iflag & VI_MOUNT) != 0 || 926 vp->v_mountedhere != NULL) { 927 VI_UNLOCK(vp); 928 vfs_unbusy(mp); 929 vput(vp); 930 return (EBUSY); 931 } 932 vp->v_iflag |= VI_MOUNT; 933 VI_UNLOCK(vp); 934 MNT_ILOCK(mp); 935 mp->mnt_flag |= fsflags & 936 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 937 MNT_IUNLOCK(mp); 938 VOP_UNLOCK(vp, 0); 939 mp->mnt_optnew = fsdata; 940 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 941 } else { 942 /* 943 * If the user is not root, ensure that they own the directory 944 * onto which we are attempting to mount. 945 */ 946 error = VOP_GETATTR(vp, &va, td->td_ucred); 947 if (error) { 948 vput(vp); 949 return (error); 950 } 951 if (va.va_uid != td->td_ucred->cr_uid) { 952 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 953 0); 954 if (error) { 955 vput(vp); 956 return (error); 957 } 958 } 959 error = vinvalbuf(vp, V_SAVE, 0, 0); 960 if (error != 0) { 961 vput(vp); 962 return (error); 963 } 964 if (vp->v_type != VDIR) { 965 vput(vp); 966 return (ENOTDIR); 967 } 968 VI_LOCK(vp); 969 if ((vp->v_iflag & VI_MOUNT) != 0 || 970 vp->v_mountedhere != NULL) { 971 VI_UNLOCK(vp); 972 vput(vp); 973 return (EBUSY); 974 } 975 vp->v_iflag |= VI_MOUNT; 976 VI_UNLOCK(vp); 977 978 /* 979 * Allocate and initialize the filesystem. 980 */ 981 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 982 VOP_UNLOCK(vp, 0); 983 984 /* XXXMAC: pass to vfs_mount_alloc? */ 985 mp->mnt_optnew = fsdata; 986 } 987 988 /* 989 * Set the mount level flags. 990 */ 991 MNT_ILOCK(mp); 992 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 993 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 994 MNT_RDONLY)); 995 if ((mp->mnt_flag & MNT_ASYNC) == 0) 996 mp->mnt_kern_flag &= ~MNTK_ASYNC; 997 MNT_IUNLOCK(mp); 998 /* 999 * Mount the filesystem. 1000 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1001 * get. No freeing of cn_pnbuf. 1002 */ 1003 error = VFS_MOUNT(mp, td); 1004 1005 /* 1006 * Process the export option only if we are 1007 * updating mount options. 1008 */ 1009 if (!error && (fsflags & MNT_UPDATE)) { 1010 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 1011 sizeof(export)) == 0) 1012 error = vfs_export(mp, &export); 1013 } 1014 1015 if (!error) { 1016 if (mp->mnt_opt != NULL) 1017 vfs_freeopts(mp->mnt_opt); 1018 mp->mnt_opt = mp->mnt_optnew; 1019 (void)VFS_STATFS(mp, &mp->mnt_stat, td); 1020 } 1021 /* 1022 * Prevent external consumers of mount options from reading 1023 * mnt_optnew. 1024 */ 1025 mp->mnt_optnew = NULL; 1026 if (mp->mnt_flag & MNT_UPDATE) { 1027 MNT_ILOCK(mp); 1028 if (error) 1029 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1030 (flag & ~MNT_QUOTA); 1031 else 1032 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1033 MNT_FORCE | MNT_SNAPSHOT); 1034 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1035 mp->mnt_kern_flag |= MNTK_ASYNC; 1036 else 1037 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1038 MNT_IUNLOCK(mp); 1039 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1040 if (mp->mnt_syncer == NULL) 1041 error = vfs_allocate_syncvnode(mp); 1042 } else { 1043 if (mp->mnt_syncer != NULL) 1044 vrele(mp->mnt_syncer); 1045 mp->mnt_syncer = NULL; 1046 } 1047 vfs_unbusy(mp); 1048 VI_LOCK(vp); 1049 vp->v_iflag &= ~VI_MOUNT; 1050 VI_UNLOCK(vp); 1051 vrele(vp); 1052 return (error); 1053 } 1054 MNT_ILOCK(mp); 1055 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1056 mp->mnt_kern_flag |= MNTK_ASYNC; 1057 else 1058 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1059 MNT_IUNLOCK(mp); 1060 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1061 /* 1062 * Put the new filesystem on the mount list after root. 1063 */ 1064 cache_purge(vp); 1065 if (!error) { 1066 struct vnode *newdp; 1067 1068 VI_LOCK(vp); 1069 vp->v_iflag &= ~VI_MOUNT; 1070 VI_UNLOCK(vp); 1071 vp->v_mountedhere = mp; 1072 mtx_lock(&mountlist_mtx); 1073 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1074 mtx_unlock(&mountlist_mtx); 1075 vfs_event_signal(NULL, VQ_MOUNT, 0); 1076 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 1077 panic("mount: lost mount"); 1078 mountcheckdirs(vp, newdp); 1079 vput(newdp); 1080 VOP_UNLOCK(vp, 0); 1081 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1082 error = vfs_allocate_syncvnode(mp); 1083 vfs_unbusy(mp); 1084 if (error) 1085 vrele(vp); 1086 } else { 1087 VI_LOCK(vp); 1088 vp->v_iflag &= ~VI_MOUNT; 1089 VI_UNLOCK(vp); 1090 vfs_unbusy(mp); 1091 vfs_mount_destroy(mp); 1092 vput(vp); 1093 } 1094 return (error); 1095} 1096 1097/* 1098 * Unmount a filesystem. 1099 * 1100 * Note: unmount takes a path to the vnode mounted on as argument, not 1101 * special file (as before). 1102 */ 1103#ifndef _SYS_SYSPROTO_H_ 1104struct unmount_args { 1105 char *path; 1106 int flags; 1107}; 1108#endif 1109/* ARGSUSED */ 1110int 1111unmount(td, uap) 1112 struct thread *td; 1113 register struct unmount_args /* { 1114 char *path; 1115 int flags; 1116 } */ *uap; 1117{ 1118 struct mount *mp; 1119 char *pathbuf; 1120 int error, id0, id1; 1121 1122 if (jailed(td->td_ucred) || usermount == 0) { 1123 error = priv_check(td, PRIV_VFS_UNMOUNT); 1124 if (error) 1125 return (error); 1126 } 1127 1128 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1129 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1130 if (error) { 1131 free(pathbuf, M_TEMP); 1132 return (error); 1133 } 1134 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1135 mtx_lock(&Giant); 1136 if (uap->flags & MNT_BYFSID) { 1137 /* Decode the filesystem ID. */ 1138 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1139 mtx_unlock(&Giant); 1140 free(pathbuf, M_TEMP); 1141 return (EINVAL); 1142 } 1143 1144 mtx_lock(&mountlist_mtx); 1145 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1146 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1147 mp->mnt_stat.f_fsid.val[1] == id1) 1148 break; 1149 } 1150 mtx_unlock(&mountlist_mtx); 1151 } else { 1152 mtx_lock(&mountlist_mtx); 1153 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1154 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1155 break; 1156 } 1157 mtx_unlock(&mountlist_mtx); 1158 } 1159 free(pathbuf, M_TEMP); 1160 if (mp == NULL) { 1161 /* 1162 * Previously we returned ENOENT for a nonexistent path and 1163 * EINVAL for a non-mountpoint. We cannot tell these apart 1164 * now, so in the !MNT_BYFSID case return the more likely 1165 * EINVAL for compatibility. 1166 */ 1167 mtx_unlock(&Giant); 1168 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1169 } 1170 1171 /* 1172 * Don't allow unmounting the root filesystem. 1173 */ 1174 if (mp->mnt_flag & MNT_ROOTFS) { 1175 mtx_unlock(&Giant); 1176 return (EINVAL); 1177 } 1178 error = dounmount(mp, uap->flags, td); 1179 mtx_unlock(&Giant); 1180 return (error); 1181} 1182 1183/* 1184 * Do the actual filesystem unmount. 1185 */ 1186int 1187dounmount(mp, flags, td) 1188 struct mount *mp; 1189 int flags; 1190 struct thread *td; 1191{ 1192 struct vnode *coveredvp, *fsrootvp; 1193 int error; 1194 int async_flag; 1195 int mnt_gen_r; 1196 1197 mtx_assert(&Giant, MA_OWNED); 1198 1199 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1200 mnt_gen_r = mp->mnt_gen; 1201 VI_LOCK(coveredvp); 1202 vholdl(coveredvp); 1203 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1204 vdrop(coveredvp); 1205 /* 1206 * Check for mp being unmounted while waiting for the 1207 * covered vnode lock. 1208 */ 1209 if (coveredvp->v_mountedhere != mp || 1210 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1211 VOP_UNLOCK(coveredvp, 0); 1212 return (EBUSY); 1213 } 1214 } 1215 /* 1216 * Only privileged root, or (if MNT_USER is set) the user that did the 1217 * original mount is permitted to unmount this filesystem. 1218 */ 1219 error = vfs_suser(mp, td); 1220 if (error) { 1221 if (coveredvp) 1222 VOP_UNLOCK(coveredvp, 0); 1223 return (error); 1224 } 1225 1226 MNT_ILOCK(mp); 1227 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1228 MNT_IUNLOCK(mp); 1229 if (coveredvp) 1230 VOP_UNLOCK(coveredvp, 0); 1231 return (EBUSY); 1232 } 1233 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1234 /* Allow filesystems to detect that a forced unmount is in progress. */ 1235 if (flags & MNT_FORCE) 1236 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1237 error = 0; 1238 if (mp->mnt_lockref) { 1239 if (flags & MNT_FORCE) { 1240 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1241 MNTK_UNMOUNTF); 1242 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1243 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1244 wakeup(mp); 1245 } 1246 MNT_IUNLOCK(mp); 1247 if (coveredvp) 1248 VOP_UNLOCK(coveredvp, 0); 1249 return (EBUSY); 1250 } 1251 mp->mnt_kern_flag |= MNTK_DRAINING; 1252 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1253 "mount drain", 0); 1254 } 1255 MNT_IUNLOCK(mp); 1256 KASSERT(mp->mnt_lockref == 0, 1257 ("%s: invalid lock refcount in the drain path @ %s:%d", 1258 __func__, __FILE__, __LINE__)); 1259 KASSERT(error == 0, 1260 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1261 __func__, __FILE__, __LINE__)); 1262 vn_start_write(NULL, &mp, V_WAIT); 1263 1264 if (mp->mnt_flag & MNT_EXPUBLIC) 1265 vfs_setpublicfs(NULL, NULL, NULL); 1266 1267 vfs_msync(mp, MNT_WAIT); 1268 MNT_ILOCK(mp); 1269 async_flag = mp->mnt_flag & MNT_ASYNC; 1270 mp->mnt_flag &= ~MNT_ASYNC; 1271 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1272 MNT_IUNLOCK(mp); 1273 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1274 if (mp->mnt_syncer != NULL) 1275 vrele(mp->mnt_syncer); 1276 /* 1277 * For forced unmounts, move process cdir/rdir refs on the fs root 1278 * vnode to the covered vnode. For non-forced unmounts we want 1279 * such references to cause an EBUSY error. 1280 */ 1281 if ((flags & MNT_FORCE) && 1282 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1283 if (mp->mnt_vnodecovered != NULL) 1284 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1285 if (fsrootvp == rootvnode) { 1286 vrele(rootvnode); 1287 rootvnode = NULL; 1288 } 1289 vput(fsrootvp); 1290 } 1291 if (((mp->mnt_flag & MNT_RDONLY) || 1292 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1293 (flags & MNT_FORCE)) { 1294 error = VFS_UNMOUNT(mp, flags, td); 1295 } 1296 vn_finished_write(mp); 1297 /* 1298 * If we failed to flush the dirty blocks for this mount point, 1299 * undo all the cdir/rdir and rootvnode changes we made above. 1300 * Unless we failed to do so because the device is reporting that 1301 * it doesn't exist anymore. 1302 */ 1303 if (error && error != ENXIO) { 1304 if ((flags & MNT_FORCE) && 1305 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1306 if (mp->mnt_vnodecovered != NULL) 1307 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1308 if (rootvnode == NULL) { 1309 rootvnode = fsrootvp; 1310 vref(rootvnode); 1311 } 1312 vput(fsrootvp); 1313 } 1314 MNT_ILOCK(mp); 1315 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1316 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1317 MNT_IUNLOCK(mp); 1318 (void) vfs_allocate_syncvnode(mp); 1319 MNT_ILOCK(mp); 1320 } 1321 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1322 mp->mnt_flag |= async_flag; 1323 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1324 mp->mnt_kern_flag |= MNTK_ASYNC; 1325 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1326 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1327 wakeup(mp); 1328 } 1329 MNT_IUNLOCK(mp); 1330 if (coveredvp) 1331 VOP_UNLOCK(coveredvp, 0); 1332 return (error); 1333 } 1334 mtx_lock(&mountlist_mtx); 1335 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1336 mtx_unlock(&mountlist_mtx); 1337 if (coveredvp != NULL) { 1338 coveredvp->v_mountedhere = NULL; 1339 vput(coveredvp); 1340 } 1341 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1342 vfs_mount_destroy(mp); 1343 return (0); 1344} 1345 1346/* 1347 * --------------------------------------------------------------------- 1348 * Mounting of root filesystem 1349 * 1350 */ 1351 1352struct root_hold_token { 1353 const char *who; 1354 LIST_ENTRY(root_hold_token) list; 1355}; 1356 1357static LIST_HEAD(, root_hold_token) root_holds = 1358 LIST_HEAD_INITIALIZER(&root_holds); 1359 1360static int root_mount_complete; 1361 1362/* 1363 * Hold root mount. 1364 */ 1365struct root_hold_token * 1366root_mount_hold(const char *identifier) 1367{ 1368 struct root_hold_token *h; 1369 1370 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1371 h->who = identifier; 1372 mtx_lock(&mountlist_mtx); 1373 LIST_INSERT_HEAD(&root_holds, h, list); 1374 mtx_unlock(&mountlist_mtx); 1375 return (h); 1376} 1377 1378/* 1379 * Release root mount. 1380 */ 1381void 1382root_mount_rel(struct root_hold_token *h) 1383{ 1384 1385 mtx_lock(&mountlist_mtx); 1386 LIST_REMOVE(h, list); 1387 wakeup(&root_holds); 1388 mtx_unlock(&mountlist_mtx); 1389 free(h, M_DEVBUF); 1390} 1391 1392/* 1393 * Wait for all subsystems to release root mount. 1394 */ 1395static void 1396root_mount_prepare(void) 1397{ 1398 struct root_hold_token *h; 1399 1400 for (;;) { 1401 DROP_GIANT(); 1402 g_waitidle(); 1403 PICKUP_GIANT(); 1404 mtx_lock(&mountlist_mtx); 1405 if (LIST_EMPTY(&root_holds)) { 1406 mtx_unlock(&mountlist_mtx); 1407 break; 1408 } 1409 printf("Root mount waiting for:"); 1410 LIST_FOREACH(h, &root_holds, list) 1411 printf(" %s", h->who); 1412 printf("\n"); 1413 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1414 hz); 1415 } 1416} 1417 1418/* 1419 * Root was mounted, share the good news. 1420 */ 1421static void 1422root_mount_done(void) 1423{ 1424 1425 /* 1426 * Use a mutex to prevent the wakeup being missed and waiting for 1427 * an extra 1 second sleep. 1428 */ 1429 mtx_lock(&mountlist_mtx); 1430 root_mount_complete = 1; 1431 wakeup(&root_mount_complete); 1432 mtx_unlock(&mountlist_mtx); 1433} 1434 1435/* 1436 * Return true if root is already mounted. 1437 */ 1438int 1439root_mounted(void) 1440{ 1441 1442 /* No mutex is acquired here because int stores are atomic. */ 1443 return (root_mount_complete); 1444} 1445 1446/* 1447 * Wait until root is mounted. 1448 */ 1449void 1450root_mount_wait(void) 1451{ 1452 1453 /* 1454 * Panic on an obvious deadlock - the function can't be called from 1455 * a thread which is doing the whole SYSINIT stuff. 1456 */ 1457 KASSERT(curthread->td_proc->p_pid != 0, 1458 ("root_mount_wait: cannot be called from the swapper thread")); 1459 mtx_lock(&mountlist_mtx); 1460 while (!root_mount_complete) { 1461 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1462 hz); 1463 } 1464 mtx_unlock(&mountlist_mtx); 1465} 1466 1467static void 1468set_rootvnode(struct thread *td) 1469{ 1470 struct proc *p; 1471 1472 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1473 panic("Cannot find root vnode"); 1474 1475 p = td->td_proc; 1476 FILEDESC_XLOCK(p->p_fd); 1477 1478 if (p->p_fd->fd_cdir != NULL) 1479 vrele(p->p_fd->fd_cdir); 1480 p->p_fd->fd_cdir = rootvnode; 1481 VREF(rootvnode); 1482 1483 if (p->p_fd->fd_rdir != NULL) 1484 vrele(p->p_fd->fd_rdir); 1485 p->p_fd->fd_rdir = rootvnode; 1486 VREF(rootvnode); 1487 1488 FILEDESC_XUNLOCK(p->p_fd); 1489 1490 VOP_UNLOCK(rootvnode, 0); 1491 1492 EVENTHANDLER_INVOKE(mountroot); 1493} 1494 1495/* 1496 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1497 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1498 */ 1499 1500static void 1501devfs_first(void) 1502{ 1503 struct thread *td = curthread; 1504 struct vfsoptlist *opts; 1505 struct vfsconf *vfsp; 1506 struct mount *mp = NULL; 1507 int error; 1508 1509 vfsp = vfs_byname("devfs"); 1510 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1511 if (vfsp == NULL) 1512 return; 1513 1514 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 1515 1516 error = VFS_MOUNT(mp, td); 1517 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1518 if (error) 1519 return; 1520 1521 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1522 TAILQ_INIT(opts); 1523 mp->mnt_opt = opts; 1524 1525 mtx_lock(&mountlist_mtx); 1526 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1527 mtx_unlock(&mountlist_mtx); 1528 1529 set_rootvnode(td); 1530 1531 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1532 if (error) 1533 printf("kern_symlink /dev -> / returns %d\n", error); 1534} 1535 1536/* 1537 * Surgically move our devfs to be mounted on /dev. 1538 */ 1539 1540static void 1541devfs_fixup(struct thread *td) 1542{ 1543 struct nameidata nd; 1544 int error; 1545 struct vnode *vp, *dvp; 1546 struct mount *mp; 1547 1548 /* Remove our devfs mount from the mountlist and purge the cache */ 1549 mtx_lock(&mountlist_mtx); 1550 mp = TAILQ_FIRST(&mountlist); 1551 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1552 mtx_unlock(&mountlist_mtx); 1553 cache_purgevfs(mp); 1554 1555 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1556 VI_LOCK(dvp); 1557 dvp->v_iflag &= ~VI_MOUNT; 1558 VI_UNLOCK(dvp); 1559 dvp->v_mountedhere = NULL; 1560 1561 /* Set up the real rootvnode, and purge the cache */ 1562 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1563 set_rootvnode(td); 1564 cache_purgevfs(rootvnode->v_mount); 1565 1566 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1567 error = namei(&nd); 1568 if (error) { 1569 printf("Lookup of /dev for devfs, error: %d\n", error); 1570 return; 1571 } 1572 NDFREE(&nd, NDF_ONLY_PNBUF); 1573 vp = nd.ni_vp; 1574 if (vp->v_type != VDIR) { 1575 vput(vp); 1576 } 1577 error = vinvalbuf(vp, V_SAVE, 0, 0); 1578 if (error) { 1579 vput(vp); 1580 } 1581 cache_purge(vp); 1582 mp->mnt_vnodecovered = vp; 1583 vp->v_mountedhere = mp; 1584 mtx_lock(&mountlist_mtx); 1585 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1586 mtx_unlock(&mountlist_mtx); 1587 VOP_UNLOCK(vp, 0); 1588 vput(dvp); 1589 vfs_unbusy(mp); 1590 1591 /* Unlink the no longer needed /dev/dev -> / symlink */ 1592 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1593} 1594 1595/* 1596 * Report errors during filesystem mounting. 1597 */ 1598void 1599vfs_mount_error(struct mount *mp, const char *fmt, ...) 1600{ 1601 struct vfsoptlist *moptlist = mp->mnt_optnew; 1602 va_list ap; 1603 int error, len; 1604 char *errmsg; 1605 1606 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1607 if (error || errmsg == NULL || len <= 0) 1608 return; 1609 1610 va_start(ap, fmt); 1611 vsnprintf(errmsg, (size_t)len, fmt, ap); 1612 va_end(ap); 1613} 1614 1615/* 1616 * Find and mount the root filesystem 1617 */ 1618void 1619vfs_mountroot(void) 1620{ 1621 char *cp; 1622 int error, i, asked = 0; 1623 1624 root_mount_prepare(); 1625 1626 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1627 NULL, NULL, mount_init, mount_fini, 1628 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1629 devfs_first(); 1630 1631 /* 1632 * We are booted with instructions to prompt for the root filesystem. 1633 */ 1634 if (boothowto & RB_ASKNAME) { 1635 if (!vfs_mountroot_ask()) 1636 goto mounted; 1637 asked = 1; 1638 } 1639 1640 /* 1641 * The root filesystem information is compiled in, and we are 1642 * booted with instructions to use it. 1643 */ 1644 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1645 if (!vfs_mountroot_try(ctrootdevname)) 1646 goto mounted; 1647 ctrootdevname = NULL; 1648 } 1649 1650 /* 1651 * We've been given the generic "use CDROM as root" flag. This is 1652 * necessary because one media may be used in many different 1653 * devices, so we need to search for them. 1654 */ 1655 if (boothowto & RB_CDROM) { 1656 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1657 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1658 goto mounted; 1659 } 1660 } 1661 1662 /* 1663 * Try to use the value read by the loader from /etc/fstab, or 1664 * supplied via some other means. This is the preferred 1665 * mechanism. 1666 */ 1667 cp = getenv("vfs.root.mountfrom"); 1668 if (cp != NULL) { 1669 error = vfs_mountroot_try(cp); 1670 freeenv(cp); 1671 if (!error) 1672 goto mounted; 1673 } 1674 1675 /* 1676 * Try values that may have been computed by code during boot 1677 */ 1678 if (!vfs_mountroot_try(rootdevnames[0])) 1679 goto mounted; 1680 if (!vfs_mountroot_try(rootdevnames[1])) 1681 goto mounted; 1682 1683 /* 1684 * If we (still) have a compiled-in default, try it. 1685 */ 1686 if (ctrootdevname != NULL) 1687 if (!vfs_mountroot_try(ctrootdevname)) 1688 goto mounted; 1689 /* 1690 * Everything so far has failed, prompt on the console if we haven't 1691 * already tried that. 1692 */ 1693 if (!asked) 1694 if (!vfs_mountroot_ask()) 1695 goto mounted; 1696 1697 panic("Root mount failed, startup aborted."); 1698 1699mounted: 1700 root_mount_done(); 1701} 1702 1703/* 1704 * Mount (mountfrom) as the root filesystem. 1705 */ 1706static int 1707vfs_mountroot_try(const char *mountfrom) 1708{ 1709 struct mount *mp; 1710 char *vfsname, *path; 1711 time_t timebase; 1712 int error; 1713 char patt[32]; 1714 1715 vfsname = NULL; 1716 path = NULL; 1717 mp = NULL; 1718 error = EINVAL; 1719 1720 if (mountfrom == NULL) 1721 return (error); /* don't complain */ 1722 printf("Trying to mount root from %s\n", mountfrom); 1723 1724 /* parse vfs name and path */ 1725 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1726 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1727 vfsname[0] = path[0] = 0; 1728 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1729 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1730 goto out; 1731 1732 if (path[0] == '\0') 1733 strcpy(path, ROOTNAME); 1734 1735 error = kernel_vmount( 1736 MNT_RDONLY | MNT_ROOTFS, 1737 "fstype", vfsname, 1738 "fspath", "/", 1739 "from", path, 1740 NULL); 1741 if (error == 0) { 1742 /* 1743 * We mount devfs prior to mounting the / FS, so the first 1744 * entry will typically be devfs. 1745 */ 1746 mp = TAILQ_FIRST(&mountlist); 1747 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1748 1749 /* 1750 * Iterate over all currently mounted file systems and use 1751 * the time stamp found to check and/or initialize the RTC. 1752 * Typically devfs has no time stamp and the only other FS 1753 * is the actual / FS. 1754 * Call inittodr() only once and pass it the largest of the 1755 * timestamps we encounter. 1756 */ 1757 timebase = 0; 1758 do { 1759 if (mp->mnt_time > timebase) 1760 timebase = mp->mnt_time; 1761 mp = TAILQ_NEXT(mp, mnt_list); 1762 } while (mp != NULL); 1763 inittodr(timebase); 1764 1765 devfs_fixup(curthread); 1766 } 1767out: 1768 free(path, M_MOUNT); 1769 free(vfsname, M_MOUNT); 1770 return (error); 1771} 1772 1773/* 1774 * --------------------------------------------------------------------- 1775 * Interactive root filesystem selection code. 1776 */ 1777 1778static int 1779vfs_mountroot_ask(void) 1780{ 1781 char name[128]; 1782 1783 for(;;) { 1784 printf("\nManual root filesystem specification:\n"); 1785 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1786#if defined(__amd64__) || defined(__i386__) || defined(__ia64__) 1787 printf(" eg. ufs:da0s1a\n"); 1788#else 1789 printf(" eg. ufs:/dev/da0a\n"); 1790#endif 1791 printf(" ? List valid disk boot devices\n"); 1792 printf(" <empty line> Abort manual input\n"); 1793 printf("\nmountroot> "); 1794 gets(name, sizeof(name), 1); 1795 if (name[0] == '\0') 1796 return (1); 1797 if (name[0] == '?') { 1798 printf("\nList of GEOM managed disk devices:\n "); 1799 g_dev_print(); 1800 continue; 1801 } 1802 if (!vfs_mountroot_try(name)) 1803 return (0); 1804 } 1805} 1806 1807/* 1808 * --------------------------------------------------------------------- 1809 * Functions for querying mount options/arguments from filesystems. 1810 */ 1811 1812/* 1813 * Check that no unknown options are given 1814 */ 1815int 1816vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1817{ 1818 struct vfsopt *opt; 1819 char errmsg[255]; 1820 const char **t, *p, *q; 1821 int ret = 0; 1822 1823 TAILQ_FOREACH(opt, opts, link) { 1824 p = opt->name; 1825 q = NULL; 1826 if (p[0] == 'n' && p[1] == 'o') 1827 q = p + 2; 1828 for(t = global_opts; *t != NULL; t++) { 1829 if (strcmp(*t, p) == 0) 1830 break; 1831 if (q != NULL) { 1832 if (strcmp(*t, q) == 0) 1833 break; 1834 } 1835 } 1836 if (*t != NULL) 1837 continue; 1838 for(t = legal; *t != NULL; t++) { 1839 if (strcmp(*t, p) == 0) 1840 break; 1841 if (q != NULL) { 1842 if (strcmp(*t, q) == 0) 1843 break; 1844 } 1845 } 1846 if (*t != NULL) 1847 continue; 1848 snprintf(errmsg, sizeof(errmsg), 1849 "mount option <%s> is unknown", p); 1850 printf("%s\n", errmsg); 1851 ret = EINVAL; 1852 } 1853 if (ret != 0) { 1854 TAILQ_FOREACH(opt, opts, link) { 1855 if (strcmp(opt->name, "errmsg") == 0) { 1856 strncpy((char *)opt->value, errmsg, opt->len); 1857 } 1858 } 1859 } 1860 return (ret); 1861} 1862 1863/* 1864 * Get a mount option by its name. 1865 * 1866 * Return 0 if the option was found, ENOENT otherwise. 1867 * If len is non-NULL it will be filled with the length 1868 * of the option. If buf is non-NULL, it will be filled 1869 * with the address of the option. 1870 */ 1871int 1872vfs_getopt(opts, name, buf, len) 1873 struct vfsoptlist *opts; 1874 const char *name; 1875 void **buf; 1876 int *len; 1877{ 1878 struct vfsopt *opt; 1879 1880 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1881 1882 TAILQ_FOREACH(opt, opts, link) { 1883 if (strcmp(name, opt->name) == 0) { 1884 if (len != NULL) 1885 *len = opt->len; 1886 if (buf != NULL) 1887 *buf = opt->value; 1888 return (0); 1889 } 1890 } 1891 return (ENOENT); 1892} 1893 1894static int 1895vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1896{ 1897 struct vfsopt *opt; 1898 int i; 1899 1900 if (opts == NULL) 1901 return (-1); 1902 1903 i = 0; 1904 TAILQ_FOREACH(opt, opts, link) { 1905 if (strcmp(name, opt->name) == 0) 1906 return (i); 1907 ++i; 1908 } 1909 return (-1); 1910} 1911 1912char * 1913vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1914{ 1915 struct vfsopt *opt; 1916 1917 *error = 0; 1918 TAILQ_FOREACH(opt, opts, link) { 1919 if (strcmp(name, opt->name) != 0) 1920 continue; 1921 if (((char *)opt->value)[opt->len - 1] != '\0') { 1922 *error = EINVAL; 1923 return (NULL); 1924 } 1925 return (opt->value); 1926 } 1927 *error = ENOENT; 1928 return (NULL); 1929} 1930 1931int 1932vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1933{ 1934 struct vfsopt *opt; 1935 1936 TAILQ_FOREACH(opt, opts, link) { 1937 if (strcmp(name, opt->name) == 0) { 1938 if (w != NULL) 1939 *w |= val; 1940 return (1); 1941 } 1942 } 1943 if (w != NULL) 1944 *w &= ~val; 1945 return (0); 1946} 1947 1948int 1949vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1950{ 1951 va_list ap; 1952 struct vfsopt *opt; 1953 int ret; 1954 1955 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1956 1957 TAILQ_FOREACH(opt, opts, link) { 1958 if (strcmp(name, opt->name) != 0) 1959 continue; 1960 if (opt->len == 0 || opt->value == NULL) 1961 return (0); 1962 if (((char *)opt->value)[opt->len - 1] != '\0') 1963 return (0); 1964 va_start(ap, fmt); 1965 ret = vsscanf(opt->value, fmt, ap); 1966 va_end(ap); 1967 return (ret); 1968 } 1969 return (0); 1970} 1971 1972/* 1973 * Find and copy a mount option. 1974 * 1975 * The size of the buffer has to be specified 1976 * in len, if it is not the same length as the 1977 * mount option, EINVAL is returned. 1978 * Returns ENOENT if the option is not found. 1979 */ 1980int 1981vfs_copyopt(opts, name, dest, len) 1982 struct vfsoptlist *opts; 1983 const char *name; 1984 void *dest; 1985 int len; 1986{ 1987 struct vfsopt *opt; 1988 1989 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1990 1991 TAILQ_FOREACH(opt, opts, link) { 1992 if (strcmp(name, opt->name) == 0) { 1993 if (len != opt->len) 1994 return (EINVAL); 1995 bcopy(opt->value, dest, opt->len); 1996 return (0); 1997 } 1998 } 1999 return (ENOENT); 2000} 2001 2002/* 2003 * This is a helper function for filesystems to traverse their 2004 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2005 */ 2006 2007struct vnode * 2008__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2009{ 2010 struct vnode *vp; 2011 2012 mtx_assert(MNT_MTX(mp), MA_OWNED); 2013 2014 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2015 if ((*mvp)->v_yield++ == 500) { 2016 MNT_IUNLOCK(mp); 2017 (*mvp)->v_yield = 0; 2018 uio_yield(); 2019 MNT_ILOCK(mp); 2020 } 2021 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2022 while (vp != NULL && vp->v_type == VMARKER) 2023 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2024 2025 /* Check if we are done */ 2026 if (vp == NULL) { 2027 __mnt_vnode_markerfree(mvp, mp); 2028 return (NULL); 2029 } 2030 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2031 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2032 return (vp); 2033} 2034 2035struct vnode * 2036__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2037{ 2038 struct vnode *vp; 2039 2040 mtx_assert(MNT_MTX(mp), MA_OWNED); 2041 2042 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2043 while (vp != NULL && vp->v_type == VMARKER) 2044 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2045 2046 /* Check if we are done */ 2047 if (vp == NULL) { 2048 *mvp = NULL; 2049 return (NULL); 2050 } 2051 mp->mnt_holdcnt++; 2052 MNT_IUNLOCK(mp); 2053 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2054 M_VNODE_MARKER, 2055 M_WAITOK | M_ZERO); 2056 MNT_ILOCK(mp); 2057 (*mvp)->v_type = VMARKER; 2058 2059 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2060 while (vp != NULL && vp->v_type == VMARKER) 2061 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2062 2063 /* Check if we are done */ 2064 if (vp == NULL) { 2065 MNT_IUNLOCK(mp); 2066 free(*mvp, M_VNODE_MARKER); 2067 MNT_ILOCK(mp); 2068 *mvp = NULL; 2069 mp->mnt_holdcnt--; 2070 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2071 wakeup(&mp->mnt_holdcnt); 2072 return (NULL); 2073 } 2074 (*mvp)->v_mount = mp; 2075 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2076 return (vp); 2077} 2078 2079 2080void 2081__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2082{ 2083 2084 if (*mvp == NULL) 2085 return; 2086 2087 mtx_assert(MNT_MTX(mp), MA_OWNED); 2088 2089 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2090 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2091 MNT_IUNLOCK(mp); 2092 free(*mvp, M_VNODE_MARKER); 2093 MNT_ILOCK(mp); 2094 *mvp = NULL; 2095 2096 mp->mnt_holdcnt--; 2097 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2098 wakeup(&mp->mnt_holdcnt); 2099} 2100 2101 2102int 2103__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 2104{ 2105 int error; 2106 2107 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 2108 if (sbp != &mp->mnt_stat) 2109 *sbp = mp->mnt_stat; 2110 return (error); 2111} 2112 2113void 2114vfs_mountedfrom(struct mount *mp, const char *from) 2115{ 2116 2117 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2118 strlcpy(mp->mnt_stat.f_mntfromname, from, 2119 sizeof mp->mnt_stat.f_mntfromname); 2120} 2121 2122/* 2123 * --------------------------------------------------------------------- 2124 * This is the api for building mount args and mounting filesystems from 2125 * inside the kernel. 2126 * 2127 * The API works by accumulation of individual args. First error is 2128 * latched. 2129 * 2130 * XXX: should be documented in new manpage kernel_mount(9) 2131 */ 2132 2133/* A memory allocation which must be freed when we are done */ 2134struct mntaarg { 2135 SLIST_ENTRY(mntaarg) next; 2136}; 2137 2138/* The header for the mount arguments */ 2139struct mntarg { 2140 struct iovec *v; 2141 int len; 2142 int error; 2143 SLIST_HEAD(, mntaarg) list; 2144}; 2145 2146/* 2147 * Add a boolean argument. 2148 * 2149 * flag is the boolean value. 2150 * name must start with "no". 2151 */ 2152struct mntarg * 2153mount_argb(struct mntarg *ma, int flag, const char *name) 2154{ 2155 2156 KASSERT(name[0] == 'n' && name[1] == 'o', 2157 ("mount_argb(...,%s): name must start with 'no'", name)); 2158 2159 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2160} 2161 2162/* 2163 * Add an argument printf style 2164 */ 2165struct mntarg * 2166mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2167{ 2168 va_list ap; 2169 struct mntaarg *maa; 2170 struct sbuf *sb; 2171 int len; 2172 2173 if (ma == NULL) { 2174 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2175 SLIST_INIT(&ma->list); 2176 } 2177 if (ma->error) 2178 return (ma); 2179 2180 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2181 M_MOUNT, M_WAITOK); 2182 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2183 ma->v[ma->len].iov_len = strlen(name) + 1; 2184 ma->len++; 2185 2186 sb = sbuf_new_auto(); 2187 va_start(ap, fmt); 2188 sbuf_vprintf(sb, fmt, ap); 2189 va_end(ap); 2190 sbuf_finish(sb); 2191 len = sbuf_len(sb) + 1; 2192 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2193 SLIST_INSERT_HEAD(&ma->list, maa, next); 2194 bcopy(sbuf_data(sb), maa + 1, len); 2195 sbuf_delete(sb); 2196 2197 ma->v[ma->len].iov_base = maa + 1; 2198 ma->v[ma->len].iov_len = len; 2199 ma->len++; 2200 2201 return (ma); 2202} 2203 2204/* 2205 * Add an argument which is a userland string. 2206 */ 2207struct mntarg * 2208mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2209{ 2210 struct mntaarg *maa; 2211 char *tbuf; 2212 2213 if (val == NULL) 2214 return (ma); 2215 if (ma == NULL) { 2216 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2217 SLIST_INIT(&ma->list); 2218 } 2219 if (ma->error) 2220 return (ma); 2221 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2222 SLIST_INSERT_HEAD(&ma->list, maa, next); 2223 tbuf = (void *)(maa + 1); 2224 ma->error = copyinstr(val, tbuf, len, NULL); 2225 return (mount_arg(ma, name, tbuf, -1)); 2226} 2227 2228/* 2229 * Plain argument. 2230 * 2231 * If length is -1, treat value as a C string. 2232 */ 2233struct mntarg * 2234mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2235{ 2236 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 2244 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2245 M_MOUNT, M_WAITOK); 2246 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2247 ma->v[ma->len].iov_len = strlen(name) + 1; 2248 ma->len++; 2249 2250 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2251 if (len < 0) 2252 ma->v[ma->len].iov_len = strlen(val) + 1; 2253 else 2254 ma->v[ma->len].iov_len = len; 2255 ma->len++; 2256 return (ma); 2257} 2258 2259/* 2260 * Free a mntarg structure 2261 */ 2262static void 2263free_mntarg(struct mntarg *ma) 2264{ 2265 struct mntaarg *maa; 2266 2267 while (!SLIST_EMPTY(&ma->list)) { 2268 maa = SLIST_FIRST(&ma->list); 2269 SLIST_REMOVE_HEAD(&ma->list, next); 2270 free(maa, M_MOUNT); 2271 } 2272 free(ma->v, M_MOUNT); 2273 free(ma, M_MOUNT); 2274} 2275 2276/* 2277 * Mount a filesystem 2278 */ 2279int 2280kernel_mount(struct mntarg *ma, int flags) 2281{ 2282 struct uio auio; 2283 int error; 2284 2285 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2286 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2287 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2288 2289 auio.uio_iov = ma->v; 2290 auio.uio_iovcnt = ma->len; 2291 auio.uio_segflg = UIO_SYSSPACE; 2292 2293 error = ma->error; 2294 if (!error) 2295 error = vfs_donmount(curthread, flags, &auio); 2296 free_mntarg(ma); 2297 return (error); 2298} 2299 2300/* 2301 * A printflike function to mount a filesystem. 2302 */ 2303int 2304kernel_vmount(int flags, ...) 2305{ 2306 struct mntarg *ma = NULL; 2307 va_list ap; 2308 const char *cp; 2309 const void *vp; 2310 int error; 2311 2312 va_start(ap, flags); 2313 for (;;) { 2314 cp = va_arg(ap, const char *); 2315 if (cp == NULL) 2316 break; 2317 vp = va_arg(ap, const void *); 2318 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2319 } 2320 va_end(ap); 2321 2322 error = kernel_mount(ma, flags); 2323 return (error); 2324} 2325