vfs_mount.c revision 138461
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 138461 2004-12-06 16:39:05Z 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/sysproto.h> 54#include <sys/sx.h> 55#include <sys/sysctl.h> 56#include <sys/sysent.h> 57#include <sys/systm.h> 58#include <sys/vnode.h> 59 60#include <geom/geom.h> 61 62#include <machine/stdarg.h> 63 64#include "opt_rootdevname.h" 65#include "opt_ddb.h" 66#include "opt_mac.h" 67 68#ifdef DDB 69#include <ddb/ddb.h> 70#endif 71 72#define ROOTNAME "root_device" 73#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 74 75static void checkdirs(struct vnode *olddp, struct vnode *newdp); 76static struct cdev *getdiskbyname(char *_name); 77static void gets(char *cp); 78static int vfs_domount(struct thread *td, const char *fstype, 79 char *fspath, int fsflags, void *fsdata, int compat); 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 * The root specifiers we will try if RB_CDROM is specified. 126 */ 127static char *cdrom_rootdevnames[] = { 128 "cd9660:cd0", 129 "cd9660:acd0", 130 NULL 131}; 132 133/* legacy find-root code */ 134char *rootdevnames[2] = {NULL, NULL}; 135struct cdev *rootdev = NULL; 136#ifdef ROOTDEVNAME 137const char *ctrootdevname = ROOTDEVNAME; 138#else 139const char *ctrootdevname = NULL; 140#endif 141 142/* Remove one mount option. */ 143static void 144vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 145{ 146 147 TAILQ_REMOVE(opts, opt, link); 148 free(opt->name, M_MOUNT); 149 if (opt->value != NULL) 150 free(opt->value, M_MOUNT); 151#ifdef INVARIANTS 152 else if (opt->len != 0) 153 panic("%s: mount option with NULL value but length != 0", 154 __func__); 155#endif 156 free(opt, M_MOUNT); 157} 158 159/* Release all resources related to the mount options. */ 160static void 161vfs_freeopts(struct vfsoptlist *opts) 162{ 163 struct vfsopt *opt; 164 165 while (!TAILQ_EMPTY(opts)) { 166 opt = TAILQ_FIRST(opts); 167 vfs_freeopt(opts, opt); 168 } 169 free(opts, M_MOUNT); 170} 171 172/* 173 * Check if options are equal (with or without the "no" prefix). 174 */ 175static int 176vfs_equalopts(const char *opt1, const char *opt2) 177{ 178 179 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 180 if (strcmp(opt1, opt2) == 0) 181 return (1); 182 /* "noopt" vs. "opt" */ 183 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 184 return (1); 185 /* "opt" vs. "noopt" */ 186 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 187 return (1); 188 return (0); 189} 190 191/* 192 * If a mount option is specified several times, 193 * (with or without the "no" prefix) only keep 194 * the last occurence of it. 195 */ 196static void 197vfs_sanitizeopts(struct vfsoptlist *opts) 198{ 199 struct vfsopt *opt, *opt2, *tmp; 200 201 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 202 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 203 while (opt2 != NULL) { 204 if (vfs_equalopts(opt->name, opt2->name)) { 205 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 206 vfs_freeopt(opts, opt2); 207 opt2 = tmp; 208 } else { 209 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 210 } 211 } 212 } 213} 214 215/* 216 * Build a linked list of mount options from a struct uio. 217 */ 218static int 219vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 220{ 221 struct vfsoptlist *opts; 222 struct vfsopt *opt; 223 size_t memused; 224 unsigned int i, iovcnt; 225 int error, namelen, optlen; 226 227 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 228 TAILQ_INIT(opts); 229 memused = 0; 230 iovcnt = auio->uio_iovcnt; 231 for (i = 0; i < iovcnt; i += 2) { 232 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 233 namelen = auio->uio_iov[i].iov_len; 234 optlen = auio->uio_iov[i + 1].iov_len; 235 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 236 opt->value = NULL; 237 opt->len = 0; 238 239 /* 240 * Do this early, so jumps to "bad" will free the current 241 * option. 242 */ 243 TAILQ_INSERT_TAIL(opts, opt, link); 244 memused += sizeof(struct vfsopt) + optlen + namelen; 245 246 /* 247 * Avoid consuming too much memory, and attempts to overflow 248 * memused. 249 */ 250 if (memused > VFS_MOUNTARG_SIZE_MAX || 251 optlen > VFS_MOUNTARG_SIZE_MAX || 252 namelen > VFS_MOUNTARG_SIZE_MAX) { 253 error = EINVAL; 254 goto bad; 255 } 256 257 if (auio->uio_segflg == UIO_SYSSPACE) { 258 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 259 } else { 260 error = copyin(auio->uio_iov[i].iov_base, opt->name, 261 namelen); 262 if (error) 263 goto bad; 264 } 265 /* Ensure names are null-terminated strings. */ 266 if (opt->name[namelen - 1] != '\0') { 267 error = EINVAL; 268 goto bad; 269 } 270 if (optlen != 0) { 271 opt->len = optlen; 272 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 273 if (auio->uio_segflg == UIO_SYSSPACE) { 274 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 275 optlen); 276 } else { 277 error = copyin(auio->uio_iov[i + 1].iov_base, 278 opt->value, optlen); 279 if (error) 280 goto bad; 281 } 282 } 283 } 284 vfs_sanitizeopts(opts); 285 *options = opts; 286 return (0); 287bad: 288 vfs_freeopts(opts); 289 return (error); 290} 291 292/* 293 * Merge the old mount options with the new ones passed 294 * in the MNT_UPDATE case. 295 */ 296static void 297vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 298{ 299 struct vfsopt *opt, *opt2, *new; 300 301 TAILQ_FOREACH(opt, opts, link) { 302 /* 303 * Check that this option hasn't been redefined 304 * nor cancelled with a "no" mount option. 305 */ 306 opt2 = TAILQ_FIRST(toopts); 307 while (opt2 != NULL) { 308 if (strcmp(opt2->name, opt->name) == 0) 309 goto next; 310 if (strncmp(opt2->name, "no", 2) == 0 && 311 strcmp(opt2->name + 2, opt->name) == 0) { 312 vfs_freeopt(toopts, opt2); 313 goto next; 314 } 315 opt2 = TAILQ_NEXT(opt2, link); 316 } 317 /* We want this option, duplicate it. */ 318 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 319 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 320 strcpy(new->name, opt->name); 321 if (opt->len != 0) { 322 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 323 bcopy(opt->value, new->value, opt->len); 324 } else { 325 new->value = NULL; 326 } 327 new->len = opt->len; 328 TAILQ_INSERT_TAIL(toopts, new, link); 329next: 330 continue; 331 } 332} 333 334/* 335 * New mount API. 336 */ 337int 338nmount(td, uap) 339 struct thread *td; 340 struct nmount_args /* { 341 struct iovec *iovp; 342 unsigned int iovcnt; 343 int flags; 344 } */ *uap; 345{ 346 struct uio *auio; 347 struct iovec *iov; 348 unsigned int i; 349 int error; 350 u_int iovcnt; 351 352 /* Kick out MNT_ROOTFS early as it is legal internally */ 353 if (uap->flags & MNT_ROOTFS) 354 return (EINVAL); 355 356 iovcnt = uap->iovcnt; 357 /* 358 * Check that we have an even number of iovec's 359 * and that we have at least two options. 360 */ 361 if ((iovcnt & 1) || (iovcnt < 4)) 362 return (EINVAL); 363 364 error = copyinuio(uap->iovp, iovcnt, &auio); 365 if (error) 366 return (error); 367 iov = auio->uio_iov; 368 for (i = 0; i < iovcnt; i++) { 369 if (iov->iov_len > MMAXOPTIONLEN) { 370 free(auio, M_IOV); 371 return (EINVAL); 372 } 373 iov++; 374 } 375 error = vfs_donmount(td, uap->flags, auio); 376 free(auio, M_IOV); 377 return (error); 378} 379 380/* 381 * Allocate and initialize the mount point struct. 382 */ 383static int 384vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, 385 const char *fspath, struct thread *td, struct mount **mpp) 386{ 387 struct mount *mp; 388 389 mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 390 TAILQ_INIT(&mp->mnt_nvnodelist); 391 mp->mnt_nvnodelistsize = 0; 392 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 393 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); 394 vfs_busy(mp, LK_NOWAIT, 0, td); 395 mp->mnt_op = vfsp->vfc_vfsops; 396 mp->mnt_vfc = vfsp; 397 vfsp->vfc_refcount++; 398 mp->mnt_stat.f_type = vfsp->vfc_typenum; 399 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 400 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 401 mp->mnt_vnodecovered = vp; 402 mp->mnt_cred = crdup(td->td_ucred); 403 mp->mnt_stat.f_owner = td->td_ucred->cr_uid; 404 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 405 mp->mnt_iosize_max = DFLTPHYS; 406#ifdef MAC 407 mac_init_mount(mp); 408 mac_create_mount(td->td_ucred, mp); 409#endif 410 *mpp = mp; 411 return (0); 412} 413 414/* 415 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 416 */ 417void 418vfs_mount_destroy(struct mount *mp, struct thread *td) 419{ 420 421 mp->mnt_vfc->vfc_refcount--; 422 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) 423 panic("unmount: dangling vnode"); 424 vfs_unbusy(mp,td); 425 lockdestroy(&mp->mnt_lock); 426 mtx_destroy(&mp->mnt_mtx); 427 if (mp->mnt_kern_flag & MNTK_MWAIT) 428 wakeup(mp); 429#ifdef MAC 430 mac_destroy_mount(mp); 431#endif 432 if (mp->mnt_opt != NULL) 433 vfs_freeopts(mp->mnt_opt); 434 crfree(mp->mnt_cred); 435 free(mp, M_MOUNT); 436} 437 438static int 439vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 440{ 441 struct vfsoptlist *optlist; 442 char *fstype, *fspath; 443 int error, fstypelen, fspathlen; 444 445 error = vfs_buildopts(fsoptions, &optlist); 446 if (error) 447 return (error); 448 449 /* 450 * We need these two options before the others, 451 * and they are mandatory for any filesystem. 452 * Ensure they are NUL terminated as well. 453 */ 454 fstypelen = 0; 455 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 456 if (error || fstype[fstypelen - 1] != '\0') { 457 error = EINVAL; 458 goto bail; 459 } 460 fspathlen = 0; 461 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 462 if (error || fspath[fspathlen - 1] != '\0') { 463 error = EINVAL; 464 goto bail; 465 } 466 467 /* 468 * Be ultra-paranoid about making sure the type and fspath 469 * variables will fit in our mp buffers, including the 470 * terminating NUL. 471 */ 472 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 473 error = ENAMETOOLONG; 474 goto bail; 475 } 476 477 mtx_lock(&Giant); 478 error = vfs_domount(td, fstype, fspath, fsflags, optlist, 0); 479 mtx_unlock(&Giant); 480bail: 481 if (error) 482 vfs_freeopts(optlist); 483 return (error); 484} 485 486/* 487 * Old mount API. 488 */ 489#ifndef _SYS_SYSPROTO_H_ 490struct mount_args { 491 char *type; 492 char *path; 493 int flags; 494 caddr_t data; 495}; 496#endif 497/* ARGSUSED */ 498int 499mount(td, uap) 500 struct thread *td; 501 struct mount_args /* { 502 char *type; 503 char *path; 504 int flags; 505 caddr_t data; 506 } */ *uap; 507{ 508 char *fstype; 509 char *fspath; 510 struct vfsconf *vfsp; 511 struct mntarg *ma = NULL; 512 int error; 513 514 /* Kick out MNT_ROOTFS early as it is legal internally */ 515 if (uap->flags & MNT_ROOTFS) 516 return (EINVAL); 517 518 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 519 520 /* 521 * vfs_mount() actually takes a kernel string for `type' and 522 * `path' now, so extract them. 523 */ 524 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 525 mtx_lock(&Giant); /* XXX ? */ 526 vfsp = vfs_byname_kld(fstype, td, &error); 527 mtx_unlock(&Giant); /* XXX ? */ 528 if (vfsp == NULL) { 529 free(fstype, M_TEMP); 530 return (ENOENT); 531 } 532 fspath = malloc(MNAMELEN, M_TEMP, M_WAITOK); 533 error = copyinstr(uap->path, fspath, MNAMELEN, NULL); 534 if (error == 0 && vfsp->vfc_vfsops->vfs_cmount != NULL) { 535 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 536 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 537 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 538 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 539 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 540 error = vfsp->vfc_vfsops->vfs_cmount( 541 ma, uap->data, uap->flags, td); 542 } else if (error == 0) { 543 mtx_lock(&Giant); 544 error = vfs_domount(td, fstype, fspath, 545 uap->flags, uap->data, 1); 546 mtx_unlock(&Giant); 547 } 548 free(fstype, M_TEMP); 549 free(fspath, M_TEMP); 550 return (error); 551} 552 553/* 554 * vfs_domount(): actually attempt a filesystem mount. 555 */ 556static int 557vfs_domount( 558 struct thread *td, /* Flags common to all filesystems. */ 559 const char *fstype, /* Filesystem type. */ 560 char *fspath, /* Mount path. */ 561 int fsflags, /* Flags common to all filesystems. */ 562 void *fsdata, /* Options local to the filesystem. */ 563 int compat /* Invocation from compat syscall. */ 564 ) 565{ 566 struct vnode *vp; 567 struct mount *mp; 568 struct vfsconf *vfsp; 569 int error, flag = 0, kern_flag = 0; 570 struct vattr va; 571 struct nameidata nd; 572 573 mtx_assert(&Giant, MA_OWNED); 574 575 /* 576 * Be ultra-paranoid about making sure the type and fspath 577 * variables will fit in our mp buffers, including the 578 * terminating NUL. 579 */ 580 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 581 return (ENAMETOOLONG); 582 583 if (jailed(td->td_ucred)) 584 return (EPERM); 585 if (usermount == 0) { 586 if ((error = suser(td)) != 0) 587 return (error); 588 } 589 590 /* 591 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 592 */ 593 if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) { 594 if ((error = suser(td)) != 0) 595 return (error); 596 } 597 /* 598 * Silently enforce MNT_NOSUID and MNT_USER for 599 * unprivileged users. 600 */ 601 if (suser(td) != 0) 602 fsflags |= MNT_NOSUID | MNT_USER; 603 /* 604 * Get vnode to be covered 605 */ 606 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td); 607 if ((error = namei(&nd)) != 0) 608 return (error); 609 NDFREE(&nd, NDF_ONLY_PNBUF); 610 vp = nd.ni_vp; 611 if (fsflags & MNT_UPDATE) { 612 if ((vp->v_vflag & VV_ROOT) == 0) { 613 vput(vp); 614 return (EINVAL); 615 } 616 mp = vp->v_mount; 617 flag = mp->mnt_flag; 618 kern_flag = mp->mnt_kern_flag; 619 /* 620 * We only allow the filesystem to be reloaded if it 621 * is currently mounted read-only. 622 */ 623 if ((fsflags & MNT_RELOAD) && 624 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 625 vput(vp); 626 return (EOPNOTSUPP); /* Needs translation */ 627 } 628 /* 629 * Only privileged root, or (if MNT_USER is set) the user that 630 * did the original mount is permitted to update it. 631 */ 632 error = vfs_suser(mp, td); 633 if (error) { 634 vput(vp); 635 return (error); 636 } 637 if (vfs_busy(mp, LK_NOWAIT, 0, td)) { 638 vput(vp); 639 return (EBUSY); 640 } 641 VI_LOCK(vp); 642 if ((vp->v_iflag & VI_MOUNT) != 0 || 643 vp->v_mountedhere != NULL) { 644 VI_UNLOCK(vp); 645 vfs_unbusy(mp, td); 646 vput(vp); 647 return (EBUSY); 648 } 649 vp->v_iflag |= VI_MOUNT; 650 VI_UNLOCK(vp); 651 mp->mnt_flag |= fsflags & 652 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT); 653 VOP_UNLOCK(vp, 0, td); 654 if (compat == 0) { 655 mp->mnt_optnew = fsdata; 656 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 657 } 658 } else { 659 /* 660 * If the user is not root, ensure that they own the directory 661 * onto which we are attempting to mount. 662 */ 663 error = VOP_GETATTR(vp, &va, td->td_ucred, td); 664 if (error) { 665 vput(vp); 666 return (error); 667 } 668 if (va.va_uid != td->td_ucred->cr_uid) { 669 if ((error = suser(td)) != 0) { 670 vput(vp); 671 return (error); 672 } 673 } 674 if ((error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0)) != 0) { 675 vput(vp); 676 return (error); 677 } 678 if (vp->v_type != VDIR) { 679 vput(vp); 680 return (ENOTDIR); 681 } 682 vfsp = vfs_byname_kld(fstype, td, &error); 683 if (vfsp == NULL) { 684 vput(vp); 685 return (error); 686 } 687 VI_LOCK(vp); 688 if ((vp->v_iflag & VI_MOUNT) != 0 || 689 vp->v_mountedhere != NULL) { 690 VI_UNLOCK(vp); 691 vput(vp); 692 return (EBUSY); 693 } 694 vp->v_iflag |= VI_MOUNT; 695 VI_UNLOCK(vp); 696 697 /* 698 * Allocate and initialize the filesystem. 699 */ 700 error = vfs_mount_alloc(vp, vfsp, fspath, td, &mp); 701 if (error) { 702 vput(vp); 703 return (error); 704 } 705 VOP_UNLOCK(vp, 0, td); 706 707 /* XXXMAC: pass to vfs_mount_alloc? */ 708 if (compat == 0) 709 mp->mnt_optnew = fsdata; 710 } 711 /* 712 * Check if the fs implements the type VFS_[O]MOUNT() 713 * function we are looking for. 714 */ 715 if ((compat && (mp->mnt_op->vfs_omount == NULL)) || 716 (!compat && (mp->mnt_op->vfs_mount == NULL))) { 717 printf("%s doesn't support the %s mount syscall\n", 718 mp->mnt_vfc->vfc_name, compat ? "old" : "new"); 719 VI_LOCK(vp); 720 vp->v_iflag &= ~VI_MOUNT; 721 VI_UNLOCK(vp); 722 if (mp->mnt_flag & MNT_UPDATE) 723 vfs_unbusy(mp, td); 724 else 725 vfs_mount_destroy(mp, td); 726 vrele(vp); 727 return (EOPNOTSUPP); 728 } 729 730 /* 731 * Set the mount level flags. 732 */ 733 if (fsflags & MNT_RDONLY) 734 mp->mnt_flag |= MNT_RDONLY; 735 else if (mp->mnt_flag & MNT_RDONLY) 736 mp->mnt_kern_flag |= MNTK_WANTRDWR; 737 mp->mnt_flag &=~ MNT_UPDATEMASK; 738 mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE); 739 /* 740 * Mount the filesystem. 741 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 742 * get. No freeing of cn_pnbuf. 743 */ 744 if (compat) 745 error = VFS_OMOUNT(mp, fspath, fsdata, td); 746 else 747 error = VFS_MOUNT(mp, td); 748 if (!error) { 749 if (mp->mnt_opt != NULL) 750 vfs_freeopts(mp->mnt_opt); 751 mp->mnt_opt = mp->mnt_optnew; 752 } 753 /* 754 * Prevent external consumers of mount options from reading 755 * mnt_optnew. 756 */ 757 mp->mnt_optnew = NULL; 758 if (mp->mnt_flag & MNT_UPDATE) { 759 if (mp->mnt_kern_flag & MNTK_WANTRDWR) 760 mp->mnt_flag &= ~MNT_RDONLY; 761 mp->mnt_flag &= 762 ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT); 763 mp->mnt_kern_flag &= ~MNTK_WANTRDWR; 764 if (error) { 765 mp->mnt_flag = flag; 766 mp->mnt_kern_flag = kern_flag; 767 } 768 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 769 if (mp->mnt_syncer == NULL) 770 error = vfs_allocate_syncvnode(mp); 771 } else { 772 if (mp->mnt_syncer != NULL) 773 vrele(mp->mnt_syncer); 774 mp->mnt_syncer = NULL; 775 } 776 vfs_unbusy(mp, td); 777 VI_LOCK(vp); 778 vp->v_iflag &= ~VI_MOUNT; 779 VI_UNLOCK(vp); 780 vrele(vp); 781 return (error); 782 } 783 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 784 /* 785 * Put the new filesystem on the mount list after root. 786 */ 787 cache_purge(vp); 788 if (!error) { 789 struct vnode *newdp; 790 791 VI_LOCK(vp); 792 vp->v_iflag &= ~VI_MOUNT; 793 VI_UNLOCK(vp); 794 vp->v_mountedhere = mp; 795 mtx_lock(&mountlist_mtx); 796 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 797 mtx_unlock(&mountlist_mtx); 798 vfs_event_signal(NULL, VQ_MOUNT, 0); 799 if (VFS_ROOT(mp, &newdp, td)) 800 panic("mount: lost mount"); 801 checkdirs(vp, newdp); 802 vput(newdp); 803 VOP_UNLOCK(vp, 0, td); 804 if ((mp->mnt_flag & MNT_RDONLY) == 0) 805 error = vfs_allocate_syncvnode(mp); 806 vfs_unbusy(mp, td); 807 if (error || (error = VFS_START(mp, 0, td)) != 0) 808 vrele(vp); 809 } else { 810 VI_LOCK(vp); 811 vp->v_iflag &= ~VI_MOUNT; 812 VI_UNLOCK(vp); 813 vfs_mount_destroy(mp, td); 814 vput(vp); 815 } 816 return (error); 817} 818 819/* 820 * Scan all active processes to see if any of them have a current 821 * or root directory of `olddp'. If so, replace them with the new 822 * mount point. 823 */ 824static void 825checkdirs(olddp, newdp) 826 struct vnode *olddp, *newdp; 827{ 828 struct filedesc *fdp; 829 struct proc *p; 830 int nrele; 831 832 if (vrefcnt(olddp) == 1) 833 return; 834 sx_slock(&allproc_lock); 835 LIST_FOREACH(p, &allproc, p_list) { 836 mtx_lock(&fdesc_mtx); 837 fdp = p->p_fd; 838 if (fdp == NULL) { 839 mtx_unlock(&fdesc_mtx); 840 continue; 841 } 842 nrele = 0; 843 FILEDESC_LOCK_FAST(fdp); 844 if (fdp->fd_cdir == olddp) { 845 vref(newdp); 846 fdp->fd_cdir = newdp; 847 nrele++; 848 } 849 if (fdp->fd_rdir == olddp) { 850 vref(newdp); 851 fdp->fd_rdir = newdp; 852 nrele++; 853 } 854 FILEDESC_UNLOCK_FAST(fdp); 855 mtx_unlock(&fdesc_mtx); 856 while (nrele--) 857 vrele(olddp); 858 } 859 sx_sunlock(&allproc_lock); 860 if (rootvnode == olddp) { 861 vrele(rootvnode); 862 vref(newdp); 863 rootvnode = newdp; 864 } 865} 866 867/* 868 * Unmount a filesystem. 869 * 870 * Note: unmount takes a path to the vnode mounted on as argument, 871 * not special file (as before). 872 */ 873#ifndef _SYS_SYSPROTO_H_ 874struct unmount_args { 875 char *path; 876 int flags; 877}; 878#endif 879/* ARGSUSED */ 880int 881unmount(td, uap) 882 struct thread *td; 883 register struct unmount_args /* { 884 char *path; 885 int flags; 886 } */ *uap; 887{ 888 struct mount *mp; 889 char *pathbuf; 890 int error, id0, id1; 891 892 if (jailed(td->td_ucred)) 893 return (EPERM); 894 if (usermount == 0) { 895 if ((error = suser(td)) != 0) 896 return (error); 897 } 898 899 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 900 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 901 if (error) { 902 free(pathbuf, M_TEMP); 903 return (error); 904 } 905 if (uap->flags & MNT_BYFSID) { 906 /* Decode the filesystem ID. */ 907 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 908 free(pathbuf, M_TEMP); 909 return (EINVAL); 910 } 911 912 mtx_lock(&mountlist_mtx); 913 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 914 if (mp->mnt_stat.f_fsid.val[0] == id0 && 915 mp->mnt_stat.f_fsid.val[1] == id1) 916 break; 917 } 918 mtx_unlock(&mountlist_mtx); 919 } else { 920 mtx_lock(&mountlist_mtx); 921 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 922 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 923 break; 924 } 925 mtx_unlock(&mountlist_mtx); 926 } 927 free(pathbuf, M_TEMP); 928 if (mp == NULL) { 929 /* 930 * Previously we returned ENOENT for a nonexistent path and 931 * EINVAL for a non-mountpoint. We cannot tell these apart 932 * now, so in the !MNT_BYFSID case return the more likely 933 * EINVAL for compatibility. 934 */ 935 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 936 } 937 938 /* 939 * Only privileged root, or (if MNT_USER is set) the user that did the 940 * original mount is permitted to unmount this filesystem. 941 */ 942 error = vfs_suser(mp, td); 943 if (error) 944 return (error); 945 946 /* 947 * Don't allow unmounting the root filesystem. 948 */ 949 if (mp->mnt_flag & MNT_ROOTFS) 950 return (EINVAL); 951 mtx_lock(&Giant); 952 error = dounmount(mp, uap->flags, td); 953 mtx_unlock(&Giant); 954 return (error); 955} 956 957/* 958 * Do the actual filesystem unmount. 959 */ 960int 961dounmount(mp, flags, td) 962 struct mount *mp; 963 int flags; 964 struct thread *td; 965{ 966 struct vnode *coveredvp, *fsrootvp; 967 int error; 968 int async_flag; 969 970 mtx_assert(&Giant, MA_OWNED); 971 972 mtx_lock(&mountlist_mtx); 973 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 974 mtx_unlock(&mountlist_mtx); 975 return (EBUSY); 976 } 977 mp->mnt_kern_flag |= MNTK_UNMOUNT; 978 /* Allow filesystems to detect that a forced unmount is in progress. */ 979 if (flags & MNT_FORCE) 980 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 981 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 982 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), &mountlist_mtx, td); 983 if (error) { 984 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 985 if (mp->mnt_kern_flag & MNTK_MWAIT) 986 wakeup(mp); 987 return (error); 988 } 989 vn_start_write(NULL, &mp, V_WAIT); 990 991 if (mp->mnt_flag & MNT_EXPUBLIC) 992 vfs_setpublicfs(NULL, NULL, NULL); 993 994 vfs_msync(mp, MNT_WAIT); 995 async_flag = mp->mnt_flag & MNT_ASYNC; 996 mp->mnt_flag &= ~MNT_ASYNC; 997 cache_purgevfs(mp); /* remove cache entries for this file sys */ 998 if (mp->mnt_syncer != NULL) 999 vrele(mp->mnt_syncer); 1000 /* 1001 * For forced unmounts, move process cdir/rdir refs on the fs root 1002 * vnode to the covered vnode. For non-forced unmounts we want 1003 * such references to cause an EBUSY error. 1004 */ 1005 if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) { 1006 if (mp->mnt_vnodecovered != NULL) 1007 checkdirs(fsrootvp, mp->mnt_vnodecovered); 1008 if (fsrootvp == rootvnode) { 1009 vrele(rootvnode); 1010 rootvnode = NULL; 1011 } 1012 vput(fsrootvp); 1013 } 1014 if (((mp->mnt_flag & MNT_RDONLY) || 1015 (error = VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td)) == 0) || 1016 (flags & MNT_FORCE)) { 1017 error = VFS_UNMOUNT(mp, flags, td); 1018 } 1019 vn_finished_write(mp); 1020 if (error) { 1021 /* Undo cdir/rdir and rootvnode changes made above. */ 1022 if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) { 1023 if (mp->mnt_vnodecovered != NULL) 1024 checkdirs(mp->mnt_vnodecovered, fsrootvp); 1025 if (rootvnode == NULL) { 1026 rootvnode = fsrootvp; 1027 vref(rootvnode); 1028 } 1029 vput(fsrootvp); 1030 } 1031 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) 1032 (void) vfs_allocate_syncvnode(mp); 1033 mtx_lock(&mountlist_mtx); 1034 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1035 mp->mnt_flag |= async_flag; 1036 lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK, 1037 &mountlist_mtx, td); 1038 if (mp->mnt_kern_flag & MNTK_MWAIT) 1039 wakeup(mp); 1040 return (error); 1041 } 1042 mtx_lock(&mountlist_mtx); 1043 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1044 if ((coveredvp = mp->mnt_vnodecovered) != NULL) 1045 coveredvp->v_mountedhere = NULL; 1046 mtx_unlock(&mountlist_mtx); 1047 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1048 vfs_mount_destroy(mp, td); 1049 if (coveredvp != NULL) 1050 vrele(coveredvp); 1051 return (0); 1052} 1053 1054/* 1055 * Find and mount the root filesystem 1056 */ 1057void 1058vfs_mountroot(void) 1059{ 1060 char *cp; 1061 int error, i, asked = 0; 1062 1063 1064 /* 1065 * Wait for GEOM to settle down 1066 */ 1067 DROP_GIANT(); 1068 g_waitidle(); 1069 PICKUP_GIANT(); 1070 1071 /* 1072 * We are booted with instructions to prompt for the root filesystem. 1073 */ 1074 if (boothowto & RB_ASKNAME) { 1075 if (!vfs_mountroot_ask()) 1076 return; 1077 asked = 1; 1078 } 1079 1080 /* 1081 * The root filesystem information is compiled in, and we are 1082 * booted with instructions to use it. 1083 */ 1084 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1085 if (!vfs_mountroot_try(ctrootdevname)) 1086 return; 1087 ctrootdevname = NULL; 1088 } 1089 1090 /* 1091 * We've been given the generic "use CDROM as root" flag. This is 1092 * necessary because one media may be used in many different 1093 * devices, so we need to search for them. 1094 */ 1095 if (boothowto & RB_CDROM) { 1096 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1097 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1098 return; 1099 } 1100 } 1101 1102 /* 1103 * Try to use the value read by the loader from /etc/fstab, or 1104 * supplied via some other means. This is the preferred 1105 * mechanism. 1106 */ 1107 cp = getenv("vfs.root.mountfrom"); 1108 if (cp != NULL) { 1109 error = vfs_mountroot_try(cp); 1110 freeenv(cp); 1111 if (!error) 1112 return; 1113 } 1114 1115 /* 1116 * Try values that may have been computed by code during boot 1117 */ 1118 if (!vfs_mountroot_try(rootdevnames[0])) 1119 return; 1120 if (!vfs_mountroot_try(rootdevnames[1])) 1121 return; 1122 1123 /* 1124 * If we (still) have a compiled-in default, try it. 1125 */ 1126 if (ctrootdevname != NULL) 1127 if (!vfs_mountroot_try(ctrootdevname)) 1128 return; 1129 1130 /* 1131 * Everything so far has failed, prompt on the console if we haven't 1132 * already tried that. 1133 */ 1134 if (!asked) 1135 if (!vfs_mountroot_ask()) 1136 return; 1137 panic("Root mount failed, startup aborted."); 1138} 1139 1140/* 1141 * Mount (mountfrom) as the root filesystem. 1142 */ 1143static int 1144vfs_mountroot_try(const char *mountfrom) 1145{ 1146 struct mount *mp; 1147 struct thread *td = curthread; 1148 struct vfsconf *vfsp; 1149 char *vfsname, *path; 1150 int error; 1151 char patt[32]; 1152 int s; 1153 1154 vfsname = NULL; 1155 path = NULL; 1156 mp = NULL; 1157 error = EINVAL; 1158 1159 if (mountfrom == NULL) 1160 return (error); /* don't complain */ 1161 1162 s = splcam(); /* Overkill, but annoying without it */ 1163 printf("Trying to mount root from %s\n", mountfrom); 1164 splx(s); 1165 1166 /* parse vfs name and path */ 1167 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1168 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1169 vfsname[0] = path[0] = 0; 1170 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1171 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1172 goto done; 1173 1174 if (path[0] == '\0') 1175 strcpy(path, ROOTNAME); 1176 1177 vfsp = vfs_byname(vfsname); 1178 if (vfsp == NULL) { 1179 printf("Can't find filesystem \"%s\"\n", vfsname); 1180 goto done; 1181 } 1182 error = vfs_mount_alloc(NULLVP, vfsp, "/", td, &mp); 1183 if (error) { 1184 printf("Could not alloc mountpoint\n"); 1185 goto done; 1186 } 1187 1188 mp->mnt_flag |= MNT_RDONLY | MNT_ROOTFS; 1189 1190 strlcpy(mp->mnt_stat.f_mntfromname, path, MNAMELEN); 1191 1192 /* 1193 * do our best to set rootdev 1194 * XXX: This does not belong here! 1195 */ 1196 if (path[0] != '\0') { 1197 struct cdev *diskdev; 1198 diskdev = getdiskbyname(path); 1199 if (diskdev != NULL) 1200 rootdev = diskdev; 1201 else 1202 printf("setrootbyname failed\n"); 1203 } 1204 1205 error = VFS_OMOUNT(mp, path, NULL, curthread); 1206 1207done: 1208 if (vfsname != NULL) 1209 free(vfsname, M_MOUNT); 1210 if (path != NULL) 1211 free(path, M_MOUNT); 1212 if (error != 0) { 1213 if (mp != NULL) 1214 vfs_mount_destroy(mp, curthread); 1215 printf("Root mount failed: %d\n", error); 1216 } else { 1217 1218 /* register with list of mounted filesystems */ 1219 mtx_lock(&mountlist_mtx); 1220 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1221 mtx_unlock(&mountlist_mtx); 1222 1223 /* sanity check system clock against root fs timestamp */ 1224 inittodr(mp->mnt_time); 1225 vfs_unbusy(mp, curthread); 1226 error = VFS_START(mp, 0, curthread); 1227 } 1228 return (error); 1229} 1230 1231/* 1232 * Spin prompting on the console for a suitable root filesystem 1233 */ 1234static int 1235vfs_mountroot_ask(void) 1236{ 1237 char name[128]; 1238 1239 for(;;) { 1240 printf("\nManual root filesystem specification:\n"); 1241 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1242#if defined(__i386__) || defined(__ia64__) 1243 printf(" eg. ufs:da0s1a\n"); 1244#else 1245 printf(" eg. ufs:/dev/da0a\n"); 1246#endif 1247 printf(" ? List valid disk boot devices\n"); 1248 printf(" <empty line> Abort manual input\n"); 1249 printf("\nmountroot> "); 1250 gets(name); 1251 if (name[0] == '\0') 1252 return (1); 1253 if (name[0] == '?') { 1254 printf("\nList of GEOM managed disk devices:\n "); 1255 g_dev_print(); 1256 continue; 1257 } 1258 if (!vfs_mountroot_try(name)) 1259 return (0); 1260 } 1261} 1262 1263/* 1264 * Local helper function for vfs_mountroot_ask. 1265 */ 1266static void 1267gets(char *cp) 1268{ 1269 char *lp; 1270 int c; 1271 1272 lp = cp; 1273 for (;;) { 1274 printf("%c", c = cngetc() & 0177); 1275 switch (c) { 1276 case -1: 1277 case '\n': 1278 case '\r': 1279 *lp++ = '\0'; 1280 return; 1281 case '\b': 1282 case '\177': 1283 if (lp > cp) { 1284 printf(" \b"); 1285 lp--; 1286 } 1287 continue; 1288 case '#': 1289 lp--; 1290 if (lp < cp) 1291 lp = cp; 1292 continue; 1293 case '@': 1294 case 'u' & 037: 1295 lp = cp; 1296 printf("%c", '\n'); 1297 continue; 1298 default: 1299 *lp++ = c; 1300 } 1301 } 1302} 1303 1304/* 1305 * Convert a given name to the cdev pointer of the device, which is probably 1306 * but not by definition, a disk. Mount a DEVFS (on nothing), look the name 1307 * up, extract the cdev from the vnode and unmount it again. Unfortunately 1308 * we cannot use the vnode directly (because we unmount the DEVFS again) 1309 * so the filesystems still have to do the bdevvp() stunt. 1310 */ 1311static struct cdev * 1312getdiskbyname(char *name) 1313{ 1314 char *cp = name; 1315 struct cdev *dev = NULL; 1316 struct thread *td = curthread; 1317 struct vfsconf *vfsp; 1318 struct mount *mp = NULL; 1319 struct vnode *vroot = NULL; 1320 struct nameidata nid; 1321 int error; 1322 1323 if (!bcmp(cp, "/dev/", 5)) 1324 cp += 5; 1325 1326 do { 1327 vfsp = vfs_byname("devfs"); 1328 if (vfsp == NULL) 1329 break; 1330 error = vfs_mount_alloc(NULLVP, vfsp, "/dev", td, &mp); 1331 if (error) 1332 break; 1333 mp->mnt_flag |= MNT_RDONLY; 1334 1335 error = VFS_MOUNT(mp, curthread); 1336 if (error) 1337 break; 1338 VFS_START(mp, 0, td); 1339 VFS_ROOT(mp, &vroot, td); 1340 VOP_UNLOCK(vroot, 0, td); 1341 1342 NDINIT(&nid, LOOKUP, NOCACHE|FOLLOW, 1343 UIO_SYSSPACE, cp, curthread); 1344 nid.ni_startdir = vroot; 1345 nid.ni_pathlen = strlen(cp); 1346 nid.ni_cnd.cn_cred = curthread->td_ucred; 1347 nid.ni_cnd.cn_nameptr = cp; 1348 1349 error = lookup(&nid); 1350 if (error) 1351 break; 1352 if (nid.ni_vp->v_type != VCHR) 1353 dev = NULL; 1354 else 1355 dev = nid.ni_vp->v_rdev; 1356 NDFREE(&nid, 0); 1357 } while (0); 1358 1359 if (vroot != NULL) 1360 VFS_UNMOUNT(mp, 0, td); 1361 if (mp != NULL) 1362 vfs_mount_destroy(mp, td); 1363 return (dev); 1364} 1365 1366/* Show the struct cdev *for a disk specified by name */ 1367#ifdef DDB 1368DB_SHOW_COMMAND(disk, db_getdiskbyname) 1369{ 1370 struct cdev *dev; 1371 1372 if (modif[0] == '\0') { 1373 db_error("usage: show disk/devicename"); 1374 return; 1375 } 1376 dev = getdiskbyname(modif); 1377 if (dev != NULL) 1378 db_printf("struct cdev *= %p\n", dev); 1379 else 1380 db_printf("No disk device matched.\n"); 1381} 1382#endif 1383 1384/* 1385 * Get a mount option by its name. 1386 * 1387 * Return 0 if the option was found, ENOENT otherwise. 1388 * If len is non-NULL it will be filled with the length 1389 * of the option. If buf is non-NULL, it will be filled 1390 * with the address of the option. 1391 */ 1392int 1393vfs_getopt(opts, name, buf, len) 1394 struct vfsoptlist *opts; 1395 const char *name; 1396 void **buf; 1397 int *len; 1398{ 1399 struct vfsopt *opt; 1400 1401 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1402 1403 TAILQ_FOREACH(opt, opts, link) { 1404 if (strcmp(name, opt->name) == 0) { 1405 if (len != NULL) 1406 *len = opt->len; 1407 if (buf != NULL) 1408 *buf = opt->value; 1409 return (0); 1410 } 1411 } 1412 return (ENOENT); 1413} 1414 1415/* 1416 * Find and copy a mount option. 1417 * 1418 * The size of the buffer has to be specified 1419 * in len, if it is not the same length as the 1420 * mount option, EINVAL is returned. 1421 * Returns ENOENT if the option is not found. 1422 */ 1423int 1424vfs_copyopt(opts, name, dest, len) 1425 struct vfsoptlist *opts; 1426 const char *name; 1427 void *dest; 1428 int len; 1429{ 1430 struct vfsopt *opt; 1431 1432 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1433 1434 TAILQ_FOREACH(opt, opts, link) { 1435 if (strcmp(name, opt->name) == 0) { 1436 if (len != opt->len) 1437 return (EINVAL); 1438 bcopy(opt->value, dest, opt->len); 1439 return (0); 1440 } 1441 } 1442 return (ENOENT); 1443} 1444 1445 1446/* 1447 * This is a helper function for filesystems to traverse their 1448 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1449 */ 1450 1451struct vnode * 1452__mnt_vnode_next(struct vnode **nvp, struct mount *mp) 1453{ 1454 struct vnode *vp; 1455 1456 mtx_assert(&mp->mnt_mtx, MA_OWNED); 1457 1458 vp = *nvp; 1459 /* Check if we are done */ 1460 if (vp == NULL) 1461 return (NULL); 1462 /* If our next vnode is no longer ours, start over */ 1463 if (vp->v_mount != mp) 1464 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1465 /* Save pointer to next vnode in list */ 1466 if (vp != NULL) 1467 *nvp = TAILQ_NEXT(vp, v_nmntvnodes); 1468 else 1469 *nvp = NULL; 1470 return (vp); 1471} 1472 1473int 1474__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 1475{ 1476 int error; 1477 1478 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 1479 if (sbp != &mp->mnt_stat) 1480 memcpy(sbp, &mp->mnt_stat, sizeof sbp); 1481 return (error); 1482} 1483 1484/* 1485 * --------------------------------------------------------------------- 1486 * This is the api for building mount args and mounting filesystems from 1487 * inside the kernel. 1488 * 1489 * The API works by accumulation of individual args. First error is 1490 * latched. 1491 * 1492 * XXX: should be documented in new manpage kernel_mount(9) 1493 */ 1494 1495/* A memory allocation which must be freed when we are done */ 1496struct mntaarg { 1497 SLIST_ENTRY(mntaarg) next; 1498}; 1499 1500/* The header for the mount arguments */ 1501struct mntarg { 1502 struct iovec *v; 1503 int len; 1504 int error; 1505 SLIST_HEAD(, mntaarg) list; 1506}; 1507 1508/* 1509 * Add a boolean argument. 1510 * 1511 * flag is the boolean value. 1512 * name must start with "no". 1513 */ 1514struct mntarg * 1515mount_argb(struct mntarg *ma, int flag, const char *name) 1516{ 1517 1518 KASSERT(name[0] == 'n' && name[1] == 'o', 1519 ("mount_argb(...,%s): name must start with 'no'", name)); 1520 1521 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 1522} 1523 1524/* 1525 * Add an argument printf style 1526 */ 1527struct mntarg * 1528mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 1529{ 1530 va_list ap; 1531 struct mntaarg *maa; 1532 struct sbuf *sb; 1533 int len; 1534 1535 if (ma == NULL) { 1536 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1537 SLIST_INIT(&ma->list); 1538 } 1539 if (ma->error) 1540 return (ma); 1541 1542 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1543 M_MOUNT, M_WAITOK); 1544 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1545 ma->v[ma->len].iov_len = strlen(name) + 1; 1546 ma->len++; 1547 1548 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND); 1549 va_start(ap, fmt); 1550 sbuf_vprintf(sb, fmt, ap); 1551 va_end(ap); 1552 sbuf_finish(sb); 1553 len = sbuf_len(sb) + 1; 1554 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1555 SLIST_INSERT_HEAD(&ma->list, maa, next); 1556 bcopy(sbuf_data(sb), maa + 1, len); 1557 sbuf_delete(sb); 1558 1559 ma->v[ma->len].iov_base = maa + 1; 1560 ma->v[ma->len].iov_len = len; 1561 ma->len++; 1562 1563 return (ma); 1564} 1565 1566/* 1567 * Add an argument which is a userland string. 1568 */ 1569struct mntarg * 1570mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 1571{ 1572 struct mntaarg *maa; 1573 char *tbuf; 1574 1575 if (val == NULL) 1576 return (ma); 1577 if (ma == NULL) { 1578 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1579 SLIST_INIT(&ma->list); 1580 } 1581 if (ma->error) 1582 return (ma); 1583 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1584 SLIST_INSERT_HEAD(&ma->list, maa, next); 1585 tbuf = (void *)(maa + 1); 1586 ma->error = copyinstr(val, tbuf, len, NULL); 1587 return (mount_arg(ma, name, tbuf, -1)); 1588} 1589 1590/* 1591 * Plain argument. 1592 * 1593 * If length is -1, use printf. 1594 */ 1595struct mntarg * 1596mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 1597{ 1598 1599 if (ma == NULL) { 1600 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1601 SLIST_INIT(&ma->list); 1602 } 1603 if (ma->error) 1604 return (ma); 1605 1606 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1607 M_MOUNT, M_WAITOK); 1608 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1609 ma->v[ma->len].iov_len = strlen(name) + 1; 1610 ma->len++; 1611 1612 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 1613 if (len < 0) 1614 ma->v[ma->len].iov_len = strlen(val) + 1; 1615 else 1616 ma->v[ma->len].iov_len = len; 1617 ma->len++; 1618 return (ma); 1619} 1620 1621/* 1622 * Free a mntarg structure 1623 */ 1624void 1625free_mntarg(struct mntarg *ma) 1626{ 1627 struct mntaarg *maa; 1628 1629 while (!SLIST_EMPTY(&ma->list)) { 1630 maa = SLIST_FIRST(&ma->list); 1631 SLIST_REMOVE_HEAD(&ma->list, next); 1632 free(maa, M_MOUNT); 1633 } 1634 free(ma->v, M_MOUNT); 1635 free(ma, M_MOUNT); 1636} 1637 1638/* 1639 * Mount a filesystem 1640 */ 1641int 1642kernel_mount(struct mntarg *ma, int flags) 1643{ 1644 struct uio auio; 1645 int error; 1646 1647 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 1648 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 1649 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 1650 1651 auio.uio_iov = ma->v; 1652 auio.uio_iovcnt = ma->len; 1653 auio.uio_segflg = UIO_SYSSPACE; 1654 1655 error = ma->error; 1656 if (!error) 1657 error = vfs_donmount(curthread, flags, &auio); 1658 free_mntarg(ma); 1659 return (error); 1660} 1661 1662/* 1663 * A printflike function to mount a filesystem. 1664 */ 1665int 1666kernel_vmount(int flags, ...) 1667{ 1668 struct mntarg *ma = NULL; 1669 va_list ap; 1670 const char *cp; 1671 const void *vp; 1672 int error; 1673 1674 va_start(ap, flags); 1675 for (;;) { 1676 cp = va_arg(ap, const char *); 1677 if (cp == NULL) 1678 break; 1679 vp = va_arg(ap, const void *); 1680 ma = mount_arg(ma, cp, vp, -1); 1681 } 1682 va_end(ap); 1683 1684 error = kernel_mount(ma, flags); 1685 return (error); 1686} 1687