vfs_mount.c revision 138509
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 138509 2004-12-07 08:15:41Z phk $"); 39 40#include <sys/param.h> 41#include <sys/conf.h> 42#include <sys/cons.h> 43#include <sys/jail.h> 44#include <sys/kernel.h> 45#include <sys/mac.h> 46#include <sys/malloc.h> 47#include <sys/mount.h> 48#include <sys/mutex.h> 49#include <sys/namei.h> 50#include <sys/proc.h> 51#include <sys/filedesc.h> 52#include <sys/reboot.h> 53#include <sys/syscallsubr.h> 54#include <sys/sysproto.h> 55#include <sys/sx.h> 56#include <sys/sysctl.h> 57#include <sys/sysent.h> 58#include <sys/systm.h> 59#include <sys/vnode.h> 60 61#include <geom/geom.h> 62 63#include <machine/stdarg.h> 64 65#include "opt_rootdevname.h" 66#include "opt_ddb.h" 67#include "opt_mac.h" 68 69#ifdef DDB 70#include <ddb/ddb.h> 71#endif 72 73#define ROOTNAME "root_device" 74#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 75 76static void checkdirs(struct vnode *olddp, struct vnode *newdp); 77static void gets(char *cp); 78static int vfs_domount(struct thread *td, const char *fstype, 79 char *fspath, int fsflags, void *fsdata); 80static int vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp, 81 const char *fspath, struct thread *td, struct mount **mpp); 82static int vfs_mountroot_ask(void); 83static int vfs_mountroot_try(const char *mountfrom); 84static int vfs_donmount(struct thread *td, int fsflags, 85 struct uio *fsoptions); 86 87static int usermount = 0; 88SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 89 "Unprivileged users may mount and unmount file systems"); 90 91MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 92 93/* List of mounted filesystems. */ 94struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 95 96/* For any iteration/modification of mountlist */ 97struct mtx mountlist_mtx; 98 99TAILQ_HEAD(vfsoptlist, vfsopt); 100struct vfsopt { 101 TAILQ_ENTRY(vfsopt) link; 102 char *name; 103 void *value; 104 int len; 105}; 106 107/* 108 * The vnode of the system's root (/ in the filesystem, without chroot 109 * active.) 110 */ 111struct vnode *rootvnode; 112 113/* 114 * The root filesystem is detailed in the kernel environment variable 115 * vfs.root.mountfrom, which is expected to be in the general format 116 * 117 * <vfsname>:[<path>] 118 * vfsname := the name of a VFS known to the kernel and capable 119 * of being mounted as root 120 * path := disk device name or other data used by the filesystem 121 * to locate its physical store 122 */ 123 124/* 125 * Global opts, taken by all filesystems 126 */ 127static const char *global_opts[] = { 128 "fstype", 129 "fspath", 130 "ro", 131 "suid", 132 "exec", 133 NULL 134}; 135 136/* 137 * The root specifiers we will try if RB_CDROM is specified. 138 */ 139static char *cdrom_rootdevnames[] = { 140 "cd9660:cd0", 141 "cd9660:acd0", 142 NULL 143}; 144 145/* legacy find-root code */ 146char *rootdevnames[2] = {NULL, NULL}; 147#ifndef ROOTDEVNAME 148# define ROOTDEVNAME NULL 149#endif 150const char *ctrootdevname = ROOTDEVNAME; 151 152/* 153 * --------------------------------------------------------------------- 154 * Functions for building and sanitizing the mount options 155 */ 156 157/* Remove one mount option. */ 158static void 159vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 160{ 161 162 TAILQ_REMOVE(opts, opt, link); 163 free(opt->name, M_MOUNT); 164 if (opt->value != NULL) 165 free(opt->value, M_MOUNT); 166#ifdef INVARIANTS 167 else if (opt->len != 0) 168 panic("%s: mount option with NULL value but length != 0", 169 __func__); 170#endif 171 free(opt, M_MOUNT); 172} 173 174/* Release all resources related to the mount options. */ 175static void 176vfs_freeopts(struct vfsoptlist *opts) 177{ 178 struct vfsopt *opt; 179 180 while (!TAILQ_EMPTY(opts)) { 181 opt = TAILQ_FIRST(opts); 182 vfs_freeopt(opts, opt); 183 } 184 free(opts, M_MOUNT); 185} 186 187/* 188 * Check if options are equal (with or without the "no" prefix). 189 */ 190static int 191vfs_equalopts(const char *opt1, const char *opt2) 192{ 193 194 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 195 if (strcmp(opt1, opt2) == 0) 196 return (1); 197 /* "noopt" vs. "opt" */ 198 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 199 return (1); 200 /* "opt" vs. "noopt" */ 201 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 202 return (1); 203 return (0); 204} 205 206/* 207 * If a mount option is specified several times, 208 * (with or without the "no" prefix) only keep 209 * the last occurence of it. 210 */ 211static void 212vfs_sanitizeopts(struct vfsoptlist *opts) 213{ 214 struct vfsopt *opt, *opt2, *tmp; 215 216 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 217 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 218 while (opt2 != NULL) { 219 if (vfs_equalopts(opt->name, opt2->name)) { 220 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 221 vfs_freeopt(opts, opt2); 222 opt2 = tmp; 223 } else { 224 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 225 } 226 } 227 } 228} 229 230/* 231 * Build a linked list of mount options from a struct uio. 232 */ 233static int 234vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 235{ 236 struct vfsoptlist *opts; 237 struct vfsopt *opt; 238 size_t memused; 239 unsigned int i, iovcnt; 240 int error, namelen, optlen; 241 242 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 243 TAILQ_INIT(opts); 244 memused = 0; 245 iovcnt = auio->uio_iovcnt; 246 for (i = 0; i < iovcnt; i += 2) { 247 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 248 namelen = auio->uio_iov[i].iov_len; 249 optlen = auio->uio_iov[i + 1].iov_len; 250 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 251 opt->value = NULL; 252 opt->len = 0; 253 254 /* 255 * Do this early, so jumps to "bad" will free the current 256 * option. 257 */ 258 TAILQ_INSERT_TAIL(opts, opt, link); 259 memused += sizeof(struct vfsopt) + optlen + namelen; 260 261 /* 262 * Avoid consuming too much memory, and attempts to overflow 263 * memused. 264 */ 265 if (memused > VFS_MOUNTARG_SIZE_MAX || 266 optlen > VFS_MOUNTARG_SIZE_MAX || 267 namelen > VFS_MOUNTARG_SIZE_MAX) { 268 error = EINVAL; 269 goto bad; 270 } 271 272 if (auio->uio_segflg == UIO_SYSSPACE) { 273 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 274 } else { 275 error = copyin(auio->uio_iov[i].iov_base, opt->name, 276 namelen); 277 if (error) 278 goto bad; 279 } 280 /* Ensure names are null-terminated strings. */ 281 if (opt->name[namelen - 1] != '\0') { 282 error = EINVAL; 283 goto bad; 284 } 285 if (optlen != 0) { 286 opt->len = optlen; 287 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 288 if (auio->uio_segflg == UIO_SYSSPACE) { 289 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 290 optlen); 291 } else { 292 error = copyin(auio->uio_iov[i + 1].iov_base, 293 opt->value, optlen); 294 if (error) 295 goto bad; 296 } 297 } 298 } 299 vfs_sanitizeopts(opts); 300 *options = opts; 301 return (0); 302bad: 303 vfs_freeopts(opts); 304 return (error); 305} 306 307/* 308 * Merge the old mount options with the new ones passed 309 * in the MNT_UPDATE case. 310 */ 311static void 312vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 313{ 314 struct vfsopt *opt, *opt2, *new; 315 316 TAILQ_FOREACH(opt, opts, link) { 317 /* 318 * Check that this option hasn't been redefined 319 * nor cancelled with a "no" mount option. 320 */ 321 opt2 = TAILQ_FIRST(toopts); 322 while (opt2 != NULL) { 323 if (strcmp(opt2->name, opt->name) == 0) 324 goto next; 325 if (strncmp(opt2->name, "no", 2) == 0 && 326 strcmp(opt2->name + 2, opt->name) == 0) { 327 vfs_freeopt(toopts, opt2); 328 goto next; 329 } 330 opt2 = TAILQ_NEXT(opt2, link); 331 } 332 /* We want this option, duplicate it. */ 333 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 334 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 335 strcpy(new->name, opt->name); 336 if (opt->len != 0) { 337 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 338 bcopy(opt->value, new->value, opt->len); 339 } else { 340 new->value = NULL; 341 } 342 new->len = opt->len; 343 TAILQ_INSERT_TAIL(toopts, new, link); 344next: 345 continue; 346 } 347} 348 349/* 350 * --------------------------------------------------------------------- 351 * Mount a filesystem 352 */ 353int 354nmount(td, uap) 355 struct thread *td; 356 struct nmount_args /* { 357 struct iovec *iovp; 358 unsigned int iovcnt; 359 int flags; 360 } */ *uap; 361{ 362 struct uio *auio; 363 struct iovec *iov; 364 unsigned int i; 365 int error; 366 u_int iovcnt; 367 368 /* Kick out MNT_ROOTFS early as it is legal internally */ 369 if (uap->flags & MNT_ROOTFS) 370 return (EINVAL); 371 372 iovcnt = uap->iovcnt; 373 /* 374 * Check that we have an even number of iovec's 375 * and that we have at least two options. 376 */ 377 if ((iovcnt & 1) || (iovcnt < 4)) 378 return (EINVAL); 379 380 error = copyinuio(uap->iovp, iovcnt, &auio); 381 if (error) 382 return (error); 383 iov = auio->uio_iov; 384 for (i = 0; i < iovcnt; i++) { 385 if (iov->iov_len > MMAXOPTIONLEN) { 386 free(auio, M_IOV); 387 return (EINVAL); 388 } 389 iov++; 390 } 391 error = vfs_donmount(td, uap->flags, auio); 392 free(auio, M_IOV); 393 return (error); 394} 395 396/* 397 * --------------------------------------------------------------------- 398 * Various utility functions 399 */ 400 401/* 402 * Allocate and initialize the mount point struct. 403 */ 404static int 405vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, 406 const char *fspath, struct thread *td, struct mount **mpp) 407{ 408 struct mount *mp; 409 410 mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 411 TAILQ_INIT(&mp->mnt_nvnodelist); 412 mp->mnt_nvnodelistsize = 0; 413 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 414 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); 415 vfs_busy(mp, LK_NOWAIT, 0, td); 416 mp->mnt_op = vfsp->vfc_vfsops; 417 mp->mnt_vfc = vfsp; 418 vfsp->vfc_refcount++; 419 mp->mnt_stat.f_type = vfsp->vfc_typenum; 420 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 421 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 422 mp->mnt_vnodecovered = vp; 423 mp->mnt_cred = crdup(td->td_ucred); 424 mp->mnt_stat.f_owner = td->td_ucred->cr_uid; 425 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 426 mp->mnt_iosize_max = DFLTPHYS; 427#ifdef MAC 428 mac_init_mount(mp); 429 mac_create_mount(td->td_ucred, mp); 430#endif 431 *mpp = mp; 432 return (0); 433} 434 435/* 436 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 437 */ 438void 439vfs_mount_destroy(struct mount *mp, struct thread *td) 440{ 441 442 mp->mnt_vfc->vfc_refcount--; 443 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) 444 panic("unmount: dangling vnode"); 445 vfs_unbusy(mp,td); 446 lockdestroy(&mp->mnt_lock); 447 mtx_destroy(&mp->mnt_mtx); 448 if (mp->mnt_kern_flag & MNTK_MWAIT) 449 wakeup(mp); 450#ifdef MAC 451 mac_destroy_mount(mp); 452#endif 453 if (mp->mnt_opt != NULL) 454 vfs_freeopts(mp->mnt_opt); 455 crfree(mp->mnt_cred); 456 free(mp, M_MOUNT); 457} 458 459static int 460vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 461{ 462 struct vfsoptlist *optlist; 463 char *fstype, *fspath; 464 int error, fstypelen, fspathlen; 465 466 error = vfs_buildopts(fsoptions, &optlist); 467 if (error) 468 return (error); 469 470 /* 471 * We need these two options before the others, 472 * and they are mandatory for any filesystem. 473 * Ensure they are NUL terminated as well. 474 */ 475 fstypelen = 0; 476 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 477 if (error || fstype[fstypelen - 1] != '\0') { 478 error = EINVAL; 479 goto bail; 480 } 481 fspathlen = 0; 482 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 483 if (error || fspath[fspathlen - 1] != '\0') { 484 error = EINVAL; 485 goto bail; 486 } 487 488 /* 489 * Be ultra-paranoid about making sure the type and fspath 490 * variables will fit in our mp buffers, including the 491 * terminating NUL. 492 */ 493 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 494 error = ENAMETOOLONG; 495 goto bail; 496 } 497 498 mtx_lock(&Giant); 499 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 500 mtx_unlock(&Giant); 501bail: 502 if (error) 503 vfs_freeopts(optlist); 504 return (error); 505} 506 507/* 508 * --------------------------------------------------------------------- 509 * Old mount API. 510 */ 511#ifndef _SYS_SYSPROTO_H_ 512struct mount_args { 513 char *type; 514 char *path; 515 int flags; 516 caddr_t data; 517}; 518#endif 519/* ARGSUSED */ 520int 521mount(td, uap) 522 struct thread *td; 523 struct mount_args /* { 524 char *type; 525 char *path; 526 int flags; 527 caddr_t data; 528 } */ *uap; 529{ 530 char *fstype; 531 struct vfsconf *vfsp = NULL; 532 struct mntarg *ma = NULL; 533 int error; 534 535 /* Kick out MNT_ROOTFS early as it is legal internally */ 536 uap->flags &= ~MNT_ROOTFS; 537 538 if (uap->data == NULL) 539 return (EINVAL); 540 541 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 542 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 543 if (!error) { 544 mtx_lock(&Giant); /* XXX ? */ 545 vfsp = vfs_byname_kld(fstype, td, &error); 546 mtx_unlock(&Giant); 547 } 548 free(fstype, M_TEMP); 549 if (error) 550 return (error); 551 if (vfsp == NULL) 552 return (ENOENT); 553 if (vfsp->vfc_vfsops->vfs_cmount == NULL) 554 return (EOPNOTSUPP); 555 556 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 557 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 558 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 559 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 560 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 561 562 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 563 return (error); 564} 565 566 567/* 568 * vfs_domount(): actually attempt a filesystem mount. 569 */ 570static int 571vfs_domount( 572 struct thread *td, /* Flags common to all filesystems. */ 573 const char *fstype, /* Filesystem type. */ 574 char *fspath, /* Mount path. */ 575 int fsflags, /* Flags common to all filesystems. */ 576 void *fsdata /* Options local to the filesystem. */ 577 ) 578{ 579 struct vnode *vp; 580 struct mount *mp; 581 struct vfsconf *vfsp; 582 int error, flag = 0, kern_flag = 0; 583 struct vattr va; 584 struct nameidata nd; 585 586 mtx_assert(&Giant, MA_OWNED); 587 588 /* 589 * Be ultra-paranoid about making sure the type and fspath 590 * variables will fit in our mp buffers, including the 591 * terminating NUL. 592 */ 593 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 594 return (ENAMETOOLONG); 595 596 if (jailed(td->td_ucred)) 597 return (EPERM); 598 if (usermount == 0) { 599 if ((error = suser(td)) != 0) 600 return (error); 601 } 602 603 /* 604 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 605 */ 606 if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) { 607 if ((error = suser(td)) != 0) 608 return (error); 609 } 610 /* 611 * Silently enforce MNT_NOSUID and MNT_USER for 612 * unprivileged users. 613 */ 614 if (suser(td) != 0) 615 fsflags |= MNT_NOSUID | MNT_USER; 616 /* 617 * Get vnode to be covered 618 */ 619 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td); 620 if ((error = namei(&nd)) != 0) 621 return (error); 622 NDFREE(&nd, NDF_ONLY_PNBUF); 623 vp = nd.ni_vp; 624 if (fsflags & MNT_UPDATE) { 625 if ((vp->v_vflag & VV_ROOT) == 0) { 626 vput(vp); 627 return (EINVAL); 628 } 629 mp = vp->v_mount; 630 flag = mp->mnt_flag; 631 kern_flag = mp->mnt_kern_flag; 632 /* 633 * We only allow the filesystem to be reloaded if it 634 * is currently mounted read-only. 635 */ 636 if ((fsflags & MNT_RELOAD) && 637 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 638 vput(vp); 639 return (EOPNOTSUPP); /* Needs translation */ 640 } 641 /* 642 * Only privileged root, or (if MNT_USER is set) the user that 643 * did the original mount is permitted to update it. 644 */ 645 error = vfs_suser(mp, td); 646 if (error) { 647 vput(vp); 648 return (error); 649 } 650 if (vfs_busy(mp, LK_NOWAIT, 0, td)) { 651 vput(vp); 652 return (EBUSY); 653 } 654 VI_LOCK(vp); 655 if ((vp->v_iflag & VI_MOUNT) != 0 || 656 vp->v_mountedhere != NULL) { 657 VI_UNLOCK(vp); 658 vfs_unbusy(mp, td); 659 vput(vp); 660 return (EBUSY); 661 } 662 vp->v_iflag |= VI_MOUNT; 663 VI_UNLOCK(vp); 664 mp->mnt_flag |= fsflags & 665 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 666 VOP_UNLOCK(vp, 0, td); 667 mp->mnt_optnew = fsdata; 668 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 669 } else { 670 /* 671 * If the user is not root, ensure that they own the directory 672 * onto which we are attempting to mount. 673 */ 674 error = VOP_GETATTR(vp, &va, td->td_ucred, td); 675 if (error) { 676 vput(vp); 677 return (error); 678 } 679 if (va.va_uid != td->td_ucred->cr_uid) { 680 if ((error = suser(td)) != 0) { 681 vput(vp); 682 return (error); 683 } 684 } 685 error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0); 686 if (error != 0) { 687 vput(vp); 688 return (error); 689 } 690 if (vp->v_type != VDIR) { 691 vput(vp); 692 return (ENOTDIR); 693 } 694 vfsp = vfs_byname_kld(fstype, td, &error); 695 if (vfsp == NULL) { 696 vput(vp); 697 return (error); 698 } 699 VI_LOCK(vp); 700 if ((vp->v_iflag & VI_MOUNT) != 0 || 701 vp->v_mountedhere != NULL) { 702 VI_UNLOCK(vp); 703 vput(vp); 704 return (EBUSY); 705 } 706 vp->v_iflag |= VI_MOUNT; 707 VI_UNLOCK(vp); 708 709 /* 710 * Allocate and initialize the filesystem. 711 */ 712 error = vfs_mount_alloc(vp, vfsp, fspath, td, &mp); 713 if (error) { 714 vput(vp); 715 return (error); 716 } 717 VOP_UNLOCK(vp, 0, td); 718 719 /* XXXMAC: pass to vfs_mount_alloc? */ 720 mp->mnt_optnew = fsdata; 721 } 722 723 /* 724 * Set the mount level flags. 725 */ 726 if (fsflags & MNT_RDONLY) 727 mp->mnt_flag |= MNT_RDONLY; 728 mp->mnt_flag &=~ MNT_UPDATEMASK; 729 mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS); 730 /* 731 * Mount the filesystem. 732 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 733 * get. No freeing of cn_pnbuf. 734 */ 735 error = VFS_MOUNT(mp, td); 736 if (!error) { 737 if (mp->mnt_opt != NULL) 738 vfs_freeopts(mp->mnt_opt); 739 mp->mnt_opt = mp->mnt_optnew; 740 VFS_STATFS(mp, &mp->mnt_stat, td); 741 } 742 /* 743 * Prevent external consumers of mount options from reading 744 * mnt_optnew. 745 */ 746 mp->mnt_optnew = NULL; 747 if (mp->mnt_flag & MNT_UPDATE) { 748 mp->mnt_flag &= 749 ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT); 750 if (error) { 751 mp->mnt_flag = flag; 752 mp->mnt_kern_flag = kern_flag; 753 } 754 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 755 if (mp->mnt_syncer == NULL) 756 error = vfs_allocate_syncvnode(mp); 757 } else { 758 if (mp->mnt_syncer != NULL) 759 vrele(mp->mnt_syncer); 760 mp->mnt_syncer = NULL; 761 } 762 vfs_unbusy(mp, td); 763 VI_LOCK(vp); 764 vp->v_iflag &= ~VI_MOUNT; 765 VI_UNLOCK(vp); 766 vrele(vp); 767 return (error); 768 } 769 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 770 /* 771 * Put the new filesystem on the mount list after root. 772 */ 773 cache_purge(vp); 774 if (!error) { 775 struct vnode *newdp; 776 777 VI_LOCK(vp); 778 vp->v_iflag &= ~VI_MOUNT; 779 VI_UNLOCK(vp); 780 vp->v_mountedhere = mp; 781 mtx_lock(&mountlist_mtx); 782 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 783 mtx_unlock(&mountlist_mtx); 784 vfs_event_signal(NULL, VQ_MOUNT, 0); 785 if (VFS_ROOT(mp, &newdp, td)) 786 panic("mount: lost mount"); 787 checkdirs(vp, newdp); 788 vput(newdp); 789 VOP_UNLOCK(vp, 0, td); 790 if ((mp->mnt_flag & MNT_RDONLY) == 0) 791 error = vfs_allocate_syncvnode(mp); 792 vfs_unbusy(mp, td); 793 if (error || (error = VFS_START(mp, 0, td)) != 0) 794 vrele(vp); 795 } else { 796 VI_LOCK(vp); 797 vp->v_iflag &= ~VI_MOUNT; 798 VI_UNLOCK(vp); 799 vfs_mount_destroy(mp, td); 800 vput(vp); 801 } 802 return (error); 803} 804 805/* 806 * Scan all active processes to see if any of them have a current 807 * or root directory of `olddp'. If so, replace them with the new 808 * mount point. 809 */ 810static void 811checkdirs(olddp, newdp) 812 struct vnode *olddp, *newdp; 813{ 814 struct filedesc *fdp; 815 struct proc *p; 816 int nrele; 817 818 if (vrefcnt(olddp) == 1) 819 return; 820 sx_slock(&allproc_lock); 821 LIST_FOREACH(p, &allproc, p_list) { 822 mtx_lock(&fdesc_mtx); 823 fdp = p->p_fd; 824 if (fdp == NULL) { 825 mtx_unlock(&fdesc_mtx); 826 continue; 827 } 828 nrele = 0; 829 FILEDESC_LOCK_FAST(fdp); 830 if (fdp->fd_cdir == olddp) { 831 vref(newdp); 832 fdp->fd_cdir = newdp; 833 nrele++; 834 } 835 if (fdp->fd_rdir == olddp) { 836 vref(newdp); 837 fdp->fd_rdir = newdp; 838 nrele++; 839 } 840 FILEDESC_UNLOCK_FAST(fdp); 841 mtx_unlock(&fdesc_mtx); 842 while (nrele--) 843 vrele(olddp); 844 } 845 sx_sunlock(&allproc_lock); 846 if (rootvnode == olddp) { 847 vrele(rootvnode); 848 vref(newdp); 849 rootvnode = newdp; 850 } 851} 852 853/* 854 * --------------------------------------------------------------------- 855 * Unmount a filesystem. 856 * 857 * Note: unmount takes a path to the vnode mounted on as argument, 858 * not special file (as before). 859 */ 860#ifndef _SYS_SYSPROTO_H_ 861struct unmount_args { 862 char *path; 863 int flags; 864}; 865#endif 866/* ARGSUSED */ 867int 868unmount(td, uap) 869 struct thread *td; 870 register struct unmount_args /* { 871 char *path; 872 int flags; 873 } */ *uap; 874{ 875 struct mount *mp; 876 char *pathbuf; 877 int error, id0, id1; 878 879 if (jailed(td->td_ucred)) 880 return (EPERM); 881 if (usermount == 0) { 882 if ((error = suser(td)) != 0) 883 return (error); 884 } 885 886 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 887 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 888 if (error) { 889 free(pathbuf, M_TEMP); 890 return (error); 891 } 892 if (uap->flags & MNT_BYFSID) { 893 /* Decode the filesystem ID. */ 894 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 895 free(pathbuf, M_TEMP); 896 return (EINVAL); 897 } 898 899 mtx_lock(&mountlist_mtx); 900 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 901 if (mp->mnt_stat.f_fsid.val[0] == id0 && 902 mp->mnt_stat.f_fsid.val[1] == id1) 903 break; 904 } 905 mtx_unlock(&mountlist_mtx); 906 } else { 907 mtx_lock(&mountlist_mtx); 908 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 909 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 910 break; 911 } 912 mtx_unlock(&mountlist_mtx); 913 } 914 free(pathbuf, M_TEMP); 915 if (mp == NULL) { 916 /* 917 * Previously we returned ENOENT for a nonexistent path and 918 * EINVAL for a non-mountpoint. We cannot tell these apart 919 * now, so in the !MNT_BYFSID case return the more likely 920 * EINVAL for compatibility. 921 */ 922 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 923 } 924 925 /* 926 * Only privileged root, or (if MNT_USER is set) the user that did the 927 * original mount is permitted to unmount this filesystem. 928 */ 929 error = vfs_suser(mp, td); 930 if (error) 931 return (error); 932 933 /* 934 * Don't allow unmounting the root filesystem. 935 */ 936 if (mp->mnt_flag & MNT_ROOTFS) 937 return (EINVAL); 938 mtx_lock(&Giant); 939 error = dounmount(mp, uap->flags, td); 940 mtx_unlock(&Giant); 941 return (error); 942} 943 944/* 945 * Do the actual filesystem unmount. 946 */ 947int 948dounmount(mp, flags, td) 949 struct mount *mp; 950 int flags; 951 struct thread *td; 952{ 953 struct vnode *coveredvp, *fsrootvp; 954 int error; 955 int async_flag; 956 957 mtx_assert(&Giant, MA_OWNED); 958 959 mtx_lock(&mountlist_mtx); 960 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 961 mtx_unlock(&mountlist_mtx); 962 return (EBUSY); 963 } 964 mp->mnt_kern_flag |= MNTK_UNMOUNT; 965 /* Allow filesystems to detect that a forced unmount is in progress. */ 966 if (flags & MNT_FORCE) 967 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 968 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 969 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), &mountlist_mtx, td); 970 if (error) { 971 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 972 if (mp->mnt_kern_flag & MNTK_MWAIT) 973 wakeup(mp); 974 return (error); 975 } 976 vn_start_write(NULL, &mp, V_WAIT); 977 978 if (mp->mnt_flag & MNT_EXPUBLIC) 979 vfs_setpublicfs(NULL, NULL, NULL); 980 981 vfs_msync(mp, MNT_WAIT); 982 async_flag = mp->mnt_flag & MNT_ASYNC; 983 mp->mnt_flag &= ~MNT_ASYNC; 984 cache_purgevfs(mp); /* remove cache entries for this file sys */ 985 if (mp->mnt_syncer != NULL) 986 vrele(mp->mnt_syncer); 987 /* 988 * For forced unmounts, move process cdir/rdir refs on the fs root 989 * vnode to the covered vnode. For non-forced unmounts we want 990 * such references to cause an EBUSY error. 991 */ 992 if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) { 993 if (mp->mnt_vnodecovered != NULL) 994 checkdirs(fsrootvp, mp->mnt_vnodecovered); 995 if (fsrootvp == rootvnode) { 996 vrele(rootvnode); 997 rootvnode = NULL; 998 } 999 vput(fsrootvp); 1000 } 1001 if (((mp->mnt_flag & MNT_RDONLY) || 1002 (error = VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td)) == 0) || 1003 (flags & MNT_FORCE)) { 1004 error = VFS_UNMOUNT(mp, flags, td); 1005 } 1006 vn_finished_write(mp); 1007 if (error) { 1008 /* Undo cdir/rdir and rootvnode changes made above. */ 1009 if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) { 1010 if (mp->mnt_vnodecovered != NULL) 1011 checkdirs(mp->mnt_vnodecovered, fsrootvp); 1012 if (rootvnode == NULL) { 1013 rootvnode = fsrootvp; 1014 vref(rootvnode); 1015 } 1016 vput(fsrootvp); 1017 } 1018 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) 1019 (void) vfs_allocate_syncvnode(mp); 1020 mtx_lock(&mountlist_mtx); 1021 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1022 mp->mnt_flag |= async_flag; 1023 lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK, 1024 &mountlist_mtx, td); 1025 if (mp->mnt_kern_flag & MNTK_MWAIT) 1026 wakeup(mp); 1027 return (error); 1028 } 1029 mtx_lock(&mountlist_mtx); 1030 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1031 if ((coveredvp = mp->mnt_vnodecovered) != NULL) 1032 coveredvp->v_mountedhere = NULL; 1033 mtx_unlock(&mountlist_mtx); 1034 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1035 vfs_mount_destroy(mp, td); 1036 if (coveredvp != NULL) 1037 vrele(coveredvp); 1038 return (0); 1039} 1040 1041/* 1042 * --------------------------------------------------------------------- 1043 * Mounting of root filesystem 1044 * 1045 */ 1046 1047static void 1048set_rootvnode(struct thread *td) 1049{ 1050 struct proc *p; 1051 1052 if (VFS_ROOT(TAILQ_FIRST(&mountlist), &rootvnode, td)) 1053 panic("Cannot find root vnode"); 1054 1055 p = td->td_proc; 1056 FILEDESC_LOCK(p->p_fd); 1057 1058 if (p->p_fd->fd_cdir != NULL) 1059 vrele(p->p_fd->fd_cdir); 1060 p->p_fd->fd_cdir = rootvnode; 1061 VREF(rootvnode); 1062 1063 if (p->p_fd->fd_rdir != NULL) 1064 vrele(p->p_fd->fd_rdir); 1065 p->p_fd->fd_rdir = rootvnode; 1066 VREF(rootvnode); 1067 1068 FILEDESC_UNLOCK(p->p_fd); 1069 1070 VOP_UNLOCK(rootvnode, 0, td); 1071} 1072 1073/* 1074 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1075 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1076 */ 1077 1078static struct mount * 1079devfs_first(void) 1080{ 1081 struct thread *td = curthread; 1082 struct vfsconf *vfsp; 1083 struct mount *mp = NULL; 1084 int error; 1085 1086 vfsp = vfs_byname("devfs"); 1087 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1088 if (vfsp == NULL) 1089 return(NULL); 1090 1091 error = vfs_mount_alloc(NULLVP, vfsp, "/dev", td, &mp); 1092 KASSERT(error == 0, ("vfs_mount_alloc failed %d", error)); 1093 if (error) 1094 return (NULL); 1095 1096 error = VFS_MOUNT(mp, curthread); 1097 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1098 if (error) 1099 return (NULL); 1100 1101 VFS_START(mp, 0, td); 1102 1103 mtx_lock(&mountlist_mtx); 1104 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1105 mtx_unlock(&mountlist_mtx); 1106 1107 set_rootvnode(td); 1108 1109 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1110 printf("kern_symlink = %d\n", error); 1111 1112 return (mp); 1113} 1114 1115/* 1116 * Surgically move our devfs to be mounted on /dev. 1117 */ 1118 1119static void 1120devfs_fixup(struct thread *td) 1121{ 1122 struct nameidata nd; 1123 int error; 1124 struct vnode *vp, *dvp; 1125 struct mount *mp; 1126 1127 /* Remove our devfs mount from the mountlist and purge the cache */ 1128 mtx_lock(&mountlist_mtx); 1129 mp = TAILQ_FIRST(&mountlist); 1130 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1131 mtx_unlock(&mountlist_mtx); 1132 cache_purgevfs(mp); 1133 1134 VFS_ROOT(mp, &dvp, td); 1135 VI_LOCK(dvp); 1136 dvp->v_iflag &= ~VI_MOUNT; 1137 dvp->v_mountedhere = NULL; 1138 VI_UNLOCK(dvp); 1139 1140 /* Set up the real rootvnode, and purge the cache */ 1141 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1142 set_rootvnode(td); 1143 cache_purgevfs(rootvnode->v_mount); 1144 1145 1146#if 0 1147 /* We may have a chance... */ 1148 error = kern_mkdir(td, "/dev", UIO_SYSSPACE, 0700); 1149 printf("kern_mkdir = %d\n", error); 1150#endif 1151 1152 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1153 error = namei(&nd); 1154 if (error) { 1155 printf("Lookup /dev -> %d\n", error); 1156 return; 1157 } 1158 NDFREE(&nd, NDF_ONLY_PNBUF); 1159 vp = nd.ni_vp; 1160 if (vp->v_type != VDIR) { 1161 vput(vp); 1162 } 1163 error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0); 1164 if (error) { 1165 vput(vp); 1166 } 1167 cache_purge(vp); 1168 mp->mnt_vnodecovered = vp; 1169 vp->v_mountedhere = mp; 1170 mtx_lock(&mountlist_mtx); 1171 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1172 mtx_unlock(&mountlist_mtx); 1173 VOP_UNLOCK(vp, 0, td); 1174 vfs_unbusy(mp, td); 1175 VREF(vp); 1176 vput(vp); 1177 vput(dvp); 1178 1179} 1180 1181/* 1182 * Find and mount the root filesystem 1183 */ 1184void 1185vfs_mountroot(void) 1186{ 1187 char *cp; 1188 int error, i, asked = 0; 1189 struct mount *mp; 1190 1191 /* 1192 * Wait for GEOM to settle down 1193 */ 1194 DROP_GIANT(); 1195 g_waitidle(); 1196 PICKUP_GIANT(); 1197 1198 mp = devfs_first(); 1199 1200 /* 1201 * We are booted with instructions to prompt for the root filesystem. 1202 */ 1203 if (boothowto & RB_ASKNAME) { 1204 if (!vfs_mountroot_ask()) 1205 return; 1206 asked = 1; 1207 } 1208 1209 /* 1210 * The root filesystem information is compiled in, and we are 1211 * booted with instructions to use it. 1212 */ 1213 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1214 if (!vfs_mountroot_try(ctrootdevname)) 1215 return; 1216 ctrootdevname = NULL; 1217 } 1218 1219 /* 1220 * We've been given the generic "use CDROM as root" flag. This is 1221 * necessary because one media may be used in many different 1222 * devices, so we need to search for them. 1223 */ 1224 if (boothowto & RB_CDROM) { 1225 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1226 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1227 return; 1228 } 1229 } 1230 1231 /* 1232 * Try to use the value read by the loader from /etc/fstab, or 1233 * supplied via some other means. This is the preferred 1234 * mechanism. 1235 */ 1236 cp = getenv("vfs.root.mountfrom"); 1237 if (cp != NULL) { 1238 error = vfs_mountroot_try(cp); 1239 freeenv(cp); 1240 if (!error) 1241 return; 1242 } 1243 1244 /* 1245 * Try values that may have been computed by code during boot 1246 */ 1247 if (!vfs_mountroot_try(rootdevnames[0])) 1248 return; 1249 if (!vfs_mountroot_try(rootdevnames[1])) 1250 return; 1251 1252 /* 1253 * If we (still) have a compiled-in default, try it. 1254 */ 1255 if (ctrootdevname != NULL) 1256 if (!vfs_mountroot_try(ctrootdevname)) 1257 return; 1258 /* 1259 * Everything so far has failed, prompt on the console if we haven't 1260 * already tried that. 1261 */ 1262 if (!asked) 1263 if (!vfs_mountroot_ask()) 1264 return; 1265 1266 panic("Root mount failed, startup aborted."); 1267} 1268 1269/* 1270 * Mount (mountfrom) as the root filesystem. 1271 */ 1272static int 1273vfs_mountroot_try(const char *mountfrom) 1274{ 1275 struct mount *mp; 1276 char *vfsname, *path; 1277 int error; 1278 char patt[32]; 1279 int s; 1280 1281 vfsname = NULL; 1282 path = NULL; 1283 mp = NULL; 1284 error = EINVAL; 1285 1286 if (mountfrom == NULL) 1287 return (error); /* don't complain */ 1288 1289 s = splcam(); /* Overkill, but annoying without it */ 1290 printf("Trying to mount root from %s\n", mountfrom); 1291 splx(s); 1292 1293 /* parse vfs name and path */ 1294 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1295 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1296 vfsname[0] = path[0] = 0; 1297 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1298 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1299 return (error); 1300 1301 if (path[0] == '\0') 1302 strcpy(path, ROOTNAME); 1303 1304 error = kernel_vmount( 1305 MNT_RDONLY | MNT_ROOTFS, 1306 "fstype", vfsname, 1307 "fspath", "/", 1308 "from", path, 1309 NULL); 1310 printf("kernel_vmount = %d\n", error); 1311 if (error == 0) { 1312 mp = TAILQ_FIRST(&mountlist); 1313 1314 /* sanity check system clock against root fs timestamp */ 1315 inittodr(mp->mnt_time); 1316 vfs_unbusy(mp, curthread); 1317 error = VFS_START(mp, 0, curthread); 1318 1319 devfs_fixup(curthread); 1320 } 1321 return (error); 1322} 1323 1324/* 1325 * --------------------------------------------------------------------- 1326 * Interactive root filesystem selection code. 1327 */ 1328 1329static int 1330vfs_mountroot_ask(void) 1331{ 1332 char name[128]; 1333 1334 for(;;) { 1335 printf("\nManual root filesystem specification:\n"); 1336 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1337#if defined(__i386__) || defined(__ia64__) 1338 printf(" eg. ufs:da0s1a\n"); 1339#else 1340 printf(" eg. ufs:/dev/da0a\n"); 1341#endif 1342 printf(" ? List valid disk boot devices\n"); 1343 printf(" <empty line> Abort manual input\n"); 1344 printf("\nmountroot> "); 1345 gets(name); 1346 if (name[0] == '\0') 1347 return (1); 1348 if (name[0] == '?') { 1349 printf("\nList of GEOM managed disk devices:\n "); 1350 g_dev_print(); 1351 continue; 1352 } 1353 if (!vfs_mountroot_try(name)) 1354 return (0); 1355 } 1356} 1357 1358/* 1359 * Local helper function for vfs_mountroot_ask. 1360 */ 1361static void 1362gets(char *cp) 1363{ 1364 char *lp; 1365 int c; 1366 1367 lp = cp; 1368 for (;;) { 1369 printf("%c", c = cngetc() & 0177); 1370 switch (c) { 1371 case -1: 1372 case '\n': 1373 case '\r': 1374 *lp++ = '\0'; 1375 return; 1376 case '\b': 1377 case '\177': 1378 if (lp > cp) { 1379 printf(" \b"); 1380 lp--; 1381 } 1382 continue; 1383 case '#': 1384 lp--; 1385 if (lp < cp) 1386 lp = cp; 1387 continue; 1388 case '@': 1389 case 'u' & 037: 1390 lp = cp; 1391 printf("%c", '\n'); 1392 continue; 1393 default: 1394 *lp++ = c; 1395 } 1396 } 1397} 1398 1399/* 1400 * --------------------------------------------------------------------- 1401 * Functions for querying mount options/arguments from filesystems. 1402 */ 1403 1404/* 1405 * Check that no unknown options are given 1406 */ 1407int 1408vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1409{ 1410 struct vfsopt *opt; 1411 const char **t, *p; 1412 1413 1414 TAILQ_FOREACH(opt, opts, link) { 1415 p = opt->name; 1416 if (p[0] == 'n' && p[1] == 'o') 1417 p += 2; 1418 for(t = global_opts; *t != NULL; t++) 1419 if (!strcmp(*t, p)) 1420 break; 1421 if (*t != NULL) 1422 continue; 1423 for(t = legal; *t != NULL; t++) 1424 if (!strcmp(*t, p)) 1425 break; 1426 if (*t != NULL) 1427 continue; 1428 printf("mount option <%s> is unknown\n", p); 1429 return (EINVAL); 1430 } 1431 return (0); 1432} 1433 1434/* 1435 * Get a mount option by its name. 1436 * 1437 * Return 0 if the option was found, ENOENT otherwise. 1438 * If len is non-NULL it will be filled with the length 1439 * of the option. If buf is non-NULL, it will be filled 1440 * with the address of the option. 1441 */ 1442int 1443vfs_getopt(opts, name, buf, len) 1444 struct vfsoptlist *opts; 1445 const char *name; 1446 void **buf; 1447 int *len; 1448{ 1449 struct vfsopt *opt; 1450 1451 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1452 1453 TAILQ_FOREACH(opt, opts, link) { 1454 if (strcmp(name, opt->name) == 0) { 1455 if (len != NULL) 1456 *len = opt->len; 1457 if (buf != NULL) 1458 *buf = opt->value; 1459 return (0); 1460 } 1461 } 1462 return (ENOENT); 1463} 1464 1465char * 1466vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1467{ 1468 struct vfsopt *opt; 1469 1470 *error = 0; 1471 TAILQ_FOREACH(opt, opts, link) { 1472 if (strcmp(name, opt->name) != 0) 1473 continue; 1474 if (((char *)opt->value)[opt->len - 1] != '\0') { 1475 *error = EINVAL; 1476 return (NULL); 1477 } 1478 return (opt->value); 1479 } 1480 return (NULL); 1481} 1482 1483int 1484vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1485{ 1486 struct vfsopt *opt; 1487 1488 TAILQ_FOREACH(opt, opts, link) { 1489 if (strcmp(name, opt->name) == 0) { 1490 if (w != NULL) 1491 *w |= val; 1492 return (1); 1493 } 1494 } 1495 if (w != NULL) 1496 *w &= ~val; 1497 return (0); 1498} 1499 1500int 1501vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1502{ 1503 va_list ap; 1504 struct vfsopt *opt; 1505 int ret; 1506 1507 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1508 1509 TAILQ_FOREACH(opt, opts, link) { 1510 if (strcmp(name, opt->name) != 0) 1511 continue; 1512 if (((char *)opt->value)[opt->len - 1] != '\0') 1513 return (0); 1514 va_start(ap, fmt); 1515 ret = vsscanf(opt->value, fmt, ap); 1516 va_end(ap); 1517 return (ret); 1518 } 1519 return (0); 1520} 1521 1522/* 1523 * Find and copy a mount option. 1524 * 1525 * The size of the buffer has to be specified 1526 * in len, if it is not the same length as the 1527 * mount option, EINVAL is returned. 1528 * Returns ENOENT if the option is not found. 1529 */ 1530int 1531vfs_copyopt(opts, name, dest, len) 1532 struct vfsoptlist *opts; 1533 const char *name; 1534 void *dest; 1535 int len; 1536{ 1537 struct vfsopt *opt; 1538 1539 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1540 1541 TAILQ_FOREACH(opt, opts, link) { 1542 if (strcmp(name, opt->name) == 0) { 1543 if (len != opt->len) 1544 return (EINVAL); 1545 bcopy(opt->value, dest, opt->len); 1546 return (0); 1547 } 1548 } 1549 return (ENOENT); 1550} 1551 1552/* 1553 * This is a helper function for filesystems to traverse their 1554 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1555 */ 1556 1557struct vnode * 1558__mnt_vnode_next(struct vnode **nvp, struct mount *mp) 1559{ 1560 struct vnode *vp; 1561 1562 mtx_assert(&mp->mnt_mtx, MA_OWNED); 1563 1564 vp = *nvp; 1565 /* Check if we are done */ 1566 if (vp == NULL) 1567 return (NULL); 1568 /* If our next vnode is no longer ours, start over */ 1569 if (vp->v_mount != mp) 1570 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1571 /* Save pointer to next vnode in list */ 1572 if (vp != NULL) 1573 *nvp = TAILQ_NEXT(vp, v_nmntvnodes); 1574 else 1575 *nvp = NULL; 1576 return (vp); 1577} 1578 1579int 1580__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 1581{ 1582 int error; 1583 1584 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 1585 if (sbp != &mp->mnt_stat) 1586 memcpy(sbp, &mp->mnt_stat, sizeof sbp); 1587 return (error); 1588} 1589 1590void 1591vfs_mountedfrom(struct mount *mp, const char *from) 1592{ 1593 1594 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 1595 strlcpy(mp->mnt_stat.f_mntfromname, from, 1596 sizeof mp->mnt_stat.f_mntfromname); 1597} 1598 1599/* 1600 * --------------------------------------------------------------------- 1601 * This is the api for building mount args and mounting filesystems from 1602 * inside the kernel. 1603 * 1604 * The API works by accumulation of individual args. First error is 1605 * latched. 1606 * 1607 * XXX: should be documented in new manpage kernel_mount(9) 1608 */ 1609 1610/* A memory allocation which must be freed when we are done */ 1611struct mntaarg { 1612 SLIST_ENTRY(mntaarg) next; 1613}; 1614 1615/* The header for the mount arguments */ 1616struct mntarg { 1617 struct iovec *v; 1618 int len; 1619 int error; 1620 SLIST_HEAD(, mntaarg) list; 1621}; 1622 1623/* 1624 * Add a boolean argument. 1625 * 1626 * flag is the boolean value. 1627 * name must start with "no". 1628 */ 1629struct mntarg * 1630mount_argb(struct mntarg *ma, int flag, const char *name) 1631{ 1632 1633 KASSERT(name[0] == 'n' && name[1] == 'o', 1634 ("mount_argb(...,%s): name must start with 'no'", name)); 1635 1636 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 1637} 1638 1639/* 1640 * Add an argument printf style 1641 */ 1642struct mntarg * 1643mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 1644{ 1645 va_list ap; 1646 struct mntaarg *maa; 1647 struct sbuf *sb; 1648 int len; 1649 1650 if (ma == NULL) { 1651 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1652 SLIST_INIT(&ma->list); 1653 } 1654 if (ma->error) 1655 return (ma); 1656 1657 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1658 M_MOUNT, M_WAITOK); 1659 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1660 ma->v[ma->len].iov_len = strlen(name) + 1; 1661 ma->len++; 1662 1663 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND); 1664 va_start(ap, fmt); 1665 sbuf_vprintf(sb, fmt, ap); 1666 va_end(ap); 1667 sbuf_finish(sb); 1668 len = sbuf_len(sb) + 1; 1669 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1670 SLIST_INSERT_HEAD(&ma->list, maa, next); 1671 bcopy(sbuf_data(sb), maa + 1, len); 1672 sbuf_delete(sb); 1673 1674 ma->v[ma->len].iov_base = maa + 1; 1675 ma->v[ma->len].iov_len = len; 1676 ma->len++; 1677 1678 return (ma); 1679} 1680 1681/* 1682 * Add an argument which is a userland string. 1683 */ 1684struct mntarg * 1685mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 1686{ 1687 struct mntaarg *maa; 1688 char *tbuf; 1689 1690 if (val == NULL) 1691 return (ma); 1692 if (ma == NULL) { 1693 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1694 SLIST_INIT(&ma->list); 1695 } 1696 if (ma->error) 1697 return (ma); 1698 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1699 SLIST_INSERT_HEAD(&ma->list, maa, next); 1700 tbuf = (void *)(maa + 1); 1701 ma->error = copyinstr(val, tbuf, len, NULL); 1702 return (mount_arg(ma, name, tbuf, -1)); 1703} 1704 1705/* 1706 * Plain argument. 1707 * 1708 * If length is -1, use printf. 1709 */ 1710struct mntarg * 1711mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 1712{ 1713 1714 if (ma == NULL) { 1715 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1716 SLIST_INIT(&ma->list); 1717 } 1718 if (ma->error) 1719 return (ma); 1720 1721 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1722 M_MOUNT, M_WAITOK); 1723 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1724 ma->v[ma->len].iov_len = strlen(name) + 1; 1725 ma->len++; 1726 1727 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 1728 if (len < 0) 1729 ma->v[ma->len].iov_len = strlen(val) + 1; 1730 else 1731 ma->v[ma->len].iov_len = len; 1732 ma->len++; 1733 return (ma); 1734} 1735 1736/* 1737 * Free a mntarg structure 1738 */ 1739void 1740free_mntarg(struct mntarg *ma) 1741{ 1742 struct mntaarg *maa; 1743 1744 while (!SLIST_EMPTY(&ma->list)) { 1745 maa = SLIST_FIRST(&ma->list); 1746 SLIST_REMOVE_HEAD(&ma->list, next); 1747 free(maa, M_MOUNT); 1748 } 1749 free(ma->v, M_MOUNT); 1750 free(ma, M_MOUNT); 1751} 1752 1753/* 1754 * Mount a filesystem 1755 */ 1756int 1757kernel_mount(struct mntarg *ma, int flags) 1758{ 1759 struct uio auio; 1760 int error; 1761 1762 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 1763 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 1764 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 1765 1766 auio.uio_iov = ma->v; 1767 auio.uio_iovcnt = ma->len; 1768 auio.uio_segflg = UIO_SYSSPACE; 1769 1770 error = ma->error; 1771 if (!error) 1772 error = vfs_donmount(curthread, flags, &auio); 1773 free_mntarg(ma); 1774 return (error); 1775} 1776 1777/* 1778 * A printflike function to mount a filesystem. 1779 */ 1780int 1781kernel_vmount(int flags, ...) 1782{ 1783 struct mntarg *ma = NULL; 1784 va_list ap; 1785 const char *cp; 1786 const void *vp; 1787 int error; 1788 1789 va_start(ap, flags); 1790 for (;;) { 1791 cp = va_arg(ap, const char *); 1792 if (cp == NULL) 1793 break; 1794 vp = va_arg(ap, const void *); 1795 ma = mount_arg(ma, cp, vp, -1); 1796 } 1797 va_end(ap); 1798 1799 error = kernel_mount(ma, flags); 1800 return (error); 1801} 1802