vfs_mount.c revision 175294
1241675Suqs/*- 2241675Suqs * Copyright (c) 1999-2004 Poul-Henning Kamp 3241675Suqs * Copyright (c) 1999 Michael Smith 4241675Suqs * Copyright (c) 1989, 1993 5241675Suqs * The Regents of the University of California. All rights reserved. 6241675Suqs * (c) UNIX System Laboratories, Inc. 7241675Suqs * All or some portions of this file are derived from material licensed 8241675Suqs * to the University of California by American Telephone and Telegraph 9241675Suqs * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10241675Suqs * the permission of UNIX System Laboratories, Inc. 11241675Suqs * 12241675Suqs * Redistribution and use in source and binary forms, with or without 13241675Suqs * modification, are permitted provided that the following conditions 14241675Suqs * are met: 15241675Suqs * 1. Redistributions of source code must retain the above copyright 16241675Suqs * notice, this list of conditions and the following disclaimer. 17241675Suqs * 2. Redistributions in binary form must reproduce the above copyright 18241675Suqs * notice, this list of conditions and the following disclaimer in the 19241675Suqs * documentation and/or other materials provided with the distribution. 20241675Suqs * 4. Neither the name of the University nor the names of its contributors 21241675Suqs * may be used to endorse or promote products derived from this software 22241675Suqs * without specific prior written permission. 23241675Suqs * 24241675Suqs * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 25241675Suqs * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26241675Suqs * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27241675Suqs * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 28241675Suqs * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29241675Suqs * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30241675Suqs * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31241675Suqs * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32241675Suqs * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33241675Suqs * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34241675Suqs * SUCH DAMAGE. 35241675Suqs */ 36241675Suqs 37241675Suqs#include <sys/cdefs.h> 38241675Suqs__FBSDID("$FreeBSD: head/sys/kern/vfs_mount.c 175294 2008-01-13 14:44:15Z attilio $"); 39241675Suqs 40241675Suqs#include <sys/param.h> 41241675Suqs#include <sys/conf.h> 42241675Suqs#include <sys/clock.h> 43241675Suqs#include <sys/jail.h> 44241675Suqs#include <sys/kernel.h> 45241675Suqs#include <sys/libkern.h> 46241675Suqs#include <sys/malloc.h> 47241675Suqs#include <sys/mount.h> 48241675Suqs#include <sys/mutex.h> 49241675Suqs#include <sys/namei.h> 50241675Suqs#include <sys/priv.h> 51241675Suqs#include <sys/proc.h> 52241675Suqs#include <sys/filedesc.h> 53241675Suqs#include <sys/reboot.h> 54241675Suqs#include <sys/syscallsubr.h> 55241675Suqs#include <sys/sysproto.h> 56241675Suqs#include <sys/sx.h> 57241675Suqs#include <sys/sysctl.h> 58241675Suqs#include <sys/sysent.h> 59241675Suqs#include <sys/systm.h> 60241675Suqs#include <sys/vnode.h> 61241675Suqs#include <vm/uma.h> 62241675Suqs 63241675Suqs#include <geom/geom.h> 64241675Suqs 65241675Suqs#include <machine/stdarg.h> 66241675Suqs 67241675Suqs#include <security/audit/audit.h> 68241675Suqs#include <security/mac/mac_framework.h> 69241675Suqs 70241675Suqs#include "opt_rootdevname.h" 71241675Suqs#include "opt_ddb.h" 72241675Suqs#include "opt_mac.h" 73241675Suqs 74241675Suqs#ifdef DDB 75241675Suqs#include <ddb/ddb.h> 76241675Suqs#endif 77241675Suqs 78241675Suqs#define ROOTNAME "root_device" 79241675Suqs#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 80241675Suqs 81241675Suqsstatic int vfs_domount(struct thread *td, const char *fstype, 82241675Suqs char *fspath, int fsflags, void *fsdata); 83241675Suqsstatic int vfs_mountroot_ask(void); 84241675Suqsstatic int vfs_mountroot_try(const char *mountfrom); 85241675Suqsstatic int vfs_donmount(struct thread *td, int fsflags, 86241675Suqs struct uio *fsoptions); 87241675Suqsstatic void free_mntarg(struct mntarg *ma); 88241675Suqsstatic int vfs_getopt_pos(struct vfsoptlist *opts, const char *name); 89241675Suqs 90241675Suqsstatic int usermount = 0; 91241675SuqsSYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 92241675Suqs "Unprivileged users may mount and unmount file systems"); 93241675Suqs 94241675SuqsMALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 95241675SuqsMALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 96241675Suqsstatic uma_zone_t mount_zone; 97241675Suqs 98241675Suqs/* List of mounted filesystems. */ 99241675Suqsstruct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 100241675Suqs 101241675Suqs/* For any iteration/modification of mountlist */ 102241675Suqsstruct mtx mountlist_mtx; 103241675SuqsMTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 104241675Suqs 105241675SuqsTAILQ_HEAD(vfsoptlist, vfsopt); 106241675Suqsstruct vfsopt { 107241675Suqs TAILQ_ENTRY(vfsopt) link; 108241675Suqs char *name; 109241675Suqs void *value; 110241675Suqs int len; 111241675Suqs}; 112241675Suqs 113241675Suqs/* 114241675Suqs * The vnode of the system's root (/ in the filesystem, without chroot 115241675Suqs * active.) 116241675Suqs */ 117241675Suqsstruct vnode *rootvnode; 118241675Suqs 119241675Suqs/* 120241675Suqs * The root filesystem is detailed in the kernel environment variable 121241675Suqs * vfs.root.mountfrom, which is expected to be in the general format 122241675Suqs * 123241675Suqs * <vfsname>:[<path>] 124241675Suqs * vfsname := the name of a VFS known to the kernel and capable 125241675Suqs * of being mounted as root 126241675Suqs * path := disk device name or other data used by the filesystem 127241675Suqs * to locate its physical store 128241675Suqs */ 129241675Suqs 130241675Suqs/* 131241675Suqs * Global opts, taken by all filesystems 132241675Suqs */ 133241675Suqsstatic const char *global_opts[] = { 134241675Suqs "errmsg", 135241675Suqs "fstype", 136241675Suqs "fspath", 137241675Suqs "ro", 138241675Suqs "rw", 139241675Suqs "nosuid", 140241675Suqs "noexec", 141241675Suqs "update", 142241675Suqs NULL 143241675Suqs}; 144241675Suqs 145241675Suqs/* 146241675Suqs * The root specifiers we will try if RB_CDROM is specified. 147241675Suqs */ 148241675Suqsstatic char *cdrom_rootdevnames[] = { 149241675Suqs "cd9660:cd0", 150241675Suqs "cd9660:acd0", 151241675Suqs NULL 152241675Suqs}; 153241675Suqs 154241675Suqs/* legacy find-root code */ 155241675Suqschar *rootdevnames[2] = {NULL, NULL}; 156241675Suqs#ifndef ROOTDEVNAME 157241675Suqs# define ROOTDEVNAME NULL 158241675Suqs#endif 159241675Suqsstatic const char *ctrootdevname = ROOTDEVNAME; 160241675Suqs 161241675Suqs/* 162241675Suqs * --------------------------------------------------------------------- 163241675Suqs * Functions for building and sanitizing the mount options 164241675Suqs */ 165241675Suqs 166241675Suqs/* Remove one mount option. */ 167241675Suqsstatic void 168241675Suqsvfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 169241675Suqs{ 170241675Suqs 171241675Suqs TAILQ_REMOVE(opts, opt, link); 172241675Suqs free(opt->name, M_MOUNT); 173241675Suqs if (opt->value != NULL) 174241675Suqs free(opt->value, M_MOUNT); 175241675Suqs#ifdef INVARIANTS 176241675Suqs else if (opt->len != 0) 177241675Suqs panic("%s: mount option with NULL value but length != 0", 178241675Suqs __func__); 179241675Suqs#endif 180241675Suqs free(opt, M_MOUNT); 181241675Suqs} 182241675Suqs 183241675Suqs/* Release all resources related to the mount options. */ 184241675Suqsvoid 185241675Suqsvfs_freeopts(struct vfsoptlist *opts) 186241675Suqs{ 187241675Suqs struct vfsopt *opt; 188241675Suqs 189241675Suqs while (!TAILQ_EMPTY(opts)) { 190241675Suqs opt = TAILQ_FIRST(opts); 191241675Suqs vfs_freeopt(opts, opt); 192241675Suqs } 193241675Suqs free(opts, M_MOUNT); 194241675Suqs} 195241675Suqs 196241675Suqsvoid 197241675Suqsvfs_deleteopt(struct vfsoptlist *opts, const char *name) 198241675Suqs{ 199241675Suqs struct vfsopt *opt, *temp; 200241675Suqs 201241675Suqs TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 202241675Suqs if (strcmp(opt->name, name) == 0) 203241675Suqs vfs_freeopt(opts, opt); 204241675Suqs } 205241675Suqs} 206241675Suqs 207241675Suqs/* 208241675Suqs * Check if options are equal (with or without the "no" prefix). 209241675Suqs */ 210241675Suqsstatic int 211241675Suqsvfs_equalopts(const char *opt1, const char *opt2) 212241675Suqs{ 213241675Suqs 214241675Suqs /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 215241675Suqs if (strcmp(opt1, opt2) == 0) 216241675Suqs return (1); 217241675Suqs /* "noopt" vs. "opt" */ 218241675Suqs if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 219241675Suqs return (1); 220241675Suqs /* "opt" vs. "noopt" */ 221241675Suqs if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 222241675Suqs return (1); 223241675Suqs return (0); 224241675Suqs} 225241675Suqs 226241675Suqs/* 227241675Suqs * If a mount option is specified several times, 228241675Suqs * (with or without the "no" prefix) only keep 229241675Suqs * the last occurence of it. 230241675Suqs */ 231241675Suqsstatic void 232241675Suqsvfs_sanitizeopts(struct vfsoptlist *opts) 233241675Suqs{ 234241675Suqs struct vfsopt *opt, *opt2, *tmp; 235241675Suqs 236241675Suqs TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 237241675Suqs opt2 = TAILQ_PREV(opt, vfsoptlist, link); 238241675Suqs while (opt2 != NULL) { 239241675Suqs if (vfs_equalopts(opt->name, opt2->name)) { 240241675Suqs tmp = TAILQ_PREV(opt2, vfsoptlist, link); 241241675Suqs vfs_freeopt(opts, opt2); 242241675Suqs opt2 = tmp; 243241675Suqs } else { 244241675Suqs opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 245241675Suqs } 246241675Suqs } 247241675Suqs } 248241675Suqs} 249241675Suqs 250241675Suqs/* 251241675Suqs * Build a linked list of mount options from a struct uio. 252241675Suqs */ 253241675Suqsstatic int 254241675Suqsvfs_buildopts(struct uio *auio, struct vfsoptlist **options) 255241675Suqs{ 256241675Suqs struct vfsoptlist *opts; 257241675Suqs struct vfsopt *opt; 258241675Suqs size_t memused; 259241675Suqs unsigned int i, iovcnt; 260241675Suqs int error, namelen, optlen; 261241675Suqs 262241675Suqs opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 263241675Suqs TAILQ_INIT(opts); 264241675Suqs memused = 0; 265241675Suqs iovcnt = auio->uio_iovcnt; 266241675Suqs for (i = 0; i < iovcnt; i += 2) { 267241675Suqs opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 268241675Suqs namelen = auio->uio_iov[i].iov_len; 269241675Suqs optlen = auio->uio_iov[i + 1].iov_len; 270241675Suqs opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 271241675Suqs opt->value = NULL; 272241675Suqs opt->len = 0; 273241675Suqs 274241675Suqs /* 275241675Suqs * Do this early, so jumps to "bad" will free the current 276241675Suqs * option. 277241675Suqs */ 278241675Suqs TAILQ_INSERT_TAIL(opts, opt, link); 279241675Suqs memused += sizeof(struct vfsopt) + optlen + namelen; 280241675Suqs 281241675Suqs /* 282241675Suqs * Avoid consuming too much memory, and attempts to overflow 283241675Suqs * memused. 284241675Suqs */ 285241675Suqs if (memused > VFS_MOUNTARG_SIZE_MAX || 286241675Suqs optlen > VFS_MOUNTARG_SIZE_MAX || 287241675Suqs namelen > VFS_MOUNTARG_SIZE_MAX) { 288241675Suqs error = EINVAL; 289241675Suqs goto bad; 290241675Suqs } 291241675Suqs 292241675Suqs if (auio->uio_segflg == UIO_SYSSPACE) { 293241675Suqs bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 294241675Suqs } else { 295241675Suqs error = copyin(auio->uio_iov[i].iov_base, opt->name, 296241675Suqs namelen); 297241675Suqs if (error) 298241675Suqs goto bad; 299241675Suqs } 300241675Suqs /* Ensure names are null-terminated strings. */ 301241675Suqs if (opt->name[namelen - 1] != '\0') { 302241675Suqs error = EINVAL; 303241675Suqs goto bad; 304241675Suqs } 305241675Suqs if (optlen != 0) { 306241675Suqs opt->len = optlen; 307241675Suqs opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 308241675Suqs if (auio->uio_segflg == UIO_SYSSPACE) { 309241675Suqs bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 310241675Suqs optlen); 311241675Suqs } else { 312241675Suqs error = copyin(auio->uio_iov[i + 1].iov_base, 313241675Suqs opt->value, optlen); 314241675Suqs if (error) 315241675Suqs goto bad; 316241675Suqs } 317241675Suqs } 318241675Suqs } 319241675Suqs vfs_sanitizeopts(opts); 320241675Suqs *options = opts; 321241675Suqs return (0); 322241675Suqsbad: 323241675Suqs vfs_freeopts(opts); 324241675Suqs return (error); 325241675Suqs} 326241675Suqs 327241675Suqs/* 328241675Suqs * Merge the old mount options with the new ones passed 329241675Suqs * in the MNT_UPDATE case. 330241675Suqs */ 331241675Suqsstatic void 332241675Suqsvfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 333241675Suqs{ 334241675Suqs struct vfsopt *opt, *opt2, *new; 335241675Suqs 336241675Suqs TAILQ_FOREACH(opt, opts, link) { 337241675Suqs /* 338241675Suqs * Check that this option hasn't been redefined 339241675Suqs * nor cancelled with a "no" mount option. 340241675Suqs */ 341241675Suqs opt2 = TAILQ_FIRST(toopts); 342241675Suqs while (opt2 != NULL) { 343241675Suqs if (strcmp(opt2->name, opt->name) == 0) 344241675Suqs goto next; 345241675Suqs if (strncmp(opt2->name, "no", 2) == 0 && 346241675Suqs strcmp(opt2->name + 2, opt->name) == 0) { 347241675Suqs vfs_freeopt(toopts, opt2); 348241675Suqs goto next; 349241675Suqs } 350241675Suqs opt2 = TAILQ_NEXT(opt2, link); 351241675Suqs } 352241675Suqs /* We want this option, duplicate it. */ 353241675Suqs new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 354241675Suqs new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 355241675Suqs strcpy(new->name, opt->name); 356241675Suqs if (opt->len != 0) { 357241675Suqs new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 358241675Suqs bcopy(opt->value, new->value, opt->len); 359241675Suqs } else { 360241675Suqs new->value = NULL; 361241675Suqs } 362241675Suqs new->len = opt->len; 363241675Suqs TAILQ_INSERT_TAIL(toopts, new, link); 364241675Suqsnext: 365241675Suqs continue; 366241675Suqs } 367241675Suqs} 368241675Suqs 369241675Suqs/* 370241675Suqs * Mount a filesystem. 371241675Suqs */ 372241675Suqsint 373241675Suqsnmount(td, uap) 374241675Suqs struct thread *td; 375241675Suqs struct nmount_args /* { 376241675Suqs struct iovec *iovp; 377241675Suqs unsigned int iovcnt; 378241675Suqs int flags; 379241675Suqs } */ *uap; 380241675Suqs{ 381241675Suqs struct uio *auio; 382241675Suqs struct iovec *iov; 383241675Suqs unsigned int i; 384241675Suqs int error; 385241675Suqs u_int iovcnt; 386241675Suqs 387241675Suqs AUDIT_ARG(fflags, uap->flags); 388241675Suqs 389241675Suqs /* 390241675Suqs * Filter out MNT_ROOTFS. We do not want clients of nmount() in 391241675Suqs * userspace to set this flag, but we must filter it out if we want 392241675Suqs * MNT_UPDATE on the root file system to work. 393241675Suqs * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 394241675Suqs */ 395241675Suqs uap->flags &= ~MNT_ROOTFS; 396241675Suqs 397241675Suqs iovcnt = uap->iovcnt; 398241675Suqs /* 399241675Suqs * Check that we have an even number of iovec's 400241675Suqs * and that we have at least two options. 401241675Suqs */ 402241675Suqs if ((iovcnt & 1) || (iovcnt < 4)) 403241675Suqs return (EINVAL); 404241675Suqs 405241675Suqs error = copyinuio(uap->iovp, iovcnt, &auio); 406241675Suqs if (error) 407241675Suqs return (error); 408241675Suqs iov = auio->uio_iov; 409241675Suqs for (i = 0; i < iovcnt; i++) { 410241675Suqs if (iov->iov_len > MMAXOPTIONLEN) { 411241675Suqs free(auio, M_IOV); 412241675Suqs return (EINVAL); 413241675Suqs } 414241675Suqs iov++; 415241675Suqs } 416241675Suqs error = vfs_donmount(td, uap->flags, auio); 417241675Suqs 418241675Suqs free(auio, M_IOV); 419241675Suqs return (error); 420241675Suqs} 421241675Suqs 422241675Suqs/* 423241675Suqs * --------------------------------------------------------------------- 424241675Suqs * Various utility functions 425241675Suqs */ 426241675Suqs 427241675Suqsvoid 428241675Suqsvfs_ref(struct mount *mp) 429241675Suqs{ 430241675Suqs 431241675Suqs MNT_ILOCK(mp); 432241675Suqs MNT_REF(mp); 433241675Suqs MNT_IUNLOCK(mp); 434241675Suqs} 435241675Suqs 436241675Suqsvoid 437241675Suqsvfs_rel(struct mount *mp) 438241675Suqs{ 439241675Suqs 440241675Suqs MNT_ILOCK(mp); 441241675Suqs MNT_REL(mp); 442241675Suqs MNT_IUNLOCK(mp); 443241675Suqs} 444241675Suqs 445241675Suqsstatic int 446241675Suqsmount_init(void *mem, int size, int flags) 447241675Suqs{ 448241675Suqs struct mount *mp; 449241675Suqs 450241675Suqs mp = (struct mount *)mem; 451241675Suqs mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 452241675Suqs lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 453241675Suqs return (0); 454241675Suqs} 455241675Suqs 456241675Suqsstatic void 457241675Suqsmount_fini(void *mem, int size) 458241675Suqs{ 459241675Suqs struct mount *mp; 460241675Suqs 461241675Suqs mp = (struct mount *)mem; 462241675Suqs lockdestroy(&mp->mnt_lock); 463241675Suqs mtx_destroy(&mp->mnt_mtx); 464241675Suqs} 465241675Suqs 466241675Suqs/* 467241675Suqs * Allocate and initialize the mount point struct. 468241675Suqs */ 469241675Suqsstruct mount * 470241675Suqsvfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, 471241675Suqs const char *fspath, struct thread *td) 472241675Suqs{ 473241675Suqs struct mount *mp; 474241675Suqs 475241675Suqs mp = uma_zalloc(mount_zone, M_WAITOK); 476241675Suqs bzero(&mp->mnt_startzero, 477241675Suqs __rangeof(struct mount, mnt_startzero, mnt_endzero)); 478241675Suqs TAILQ_INIT(&mp->mnt_nvnodelist); 479241675Suqs mp->mnt_nvnodelistsize = 0; 480241675Suqs mp->mnt_ref = 0; 481241675Suqs (void) vfs_busy(mp, LK_NOWAIT, 0, td); 482241675Suqs mp->mnt_op = vfsp->vfc_vfsops; 483241675Suqs mp->mnt_vfc = vfsp; 484241675Suqs vfsp->vfc_refcount++; /* XXX Unlocked */ 485241675Suqs mp->mnt_stat.f_type = vfsp->vfc_typenum; 486241675Suqs mp->mnt_gen++; 487241675Suqs strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 488241675Suqs mp->mnt_vnodecovered = vp; 489241675Suqs mp->mnt_cred = crdup(td->td_ucred); 490241675Suqs mp->mnt_stat.f_owner = td->td_ucred->cr_uid; 491241675Suqs strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 492241675Suqs mp->mnt_iosize_max = DFLTPHYS; 493241675Suqs#ifdef MAC 494241675Suqs mac_mount_init(mp); 495241675Suqs mac_mount_create(td->td_ucred, mp); 496241675Suqs#endif 497241675Suqs arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 498241675Suqs return (mp); 499241675Suqs} 500241675Suqs 501241675Suqs/* 502241675Suqs * Destroy the mount struct previously allocated by vfs_mount_alloc(). 503241675Suqs */ 504241675Suqsvoid 505241675Suqsvfs_mount_destroy(struct mount *mp) 506241675Suqs{ 507241675Suqs int i; 508241675Suqs 509241675Suqs MNT_ILOCK(mp); 510241675Suqs for (i = 0; mp->mnt_ref && i < 3; i++) 511241675Suqs msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz); 512241675Suqs /* 513241675Suqs * This will always cause a 3 second delay in rebooting due to 514241675Suqs * refs on the root mountpoint that never go away. Most of these 515241675Suqs * are held by init which never exits. 516241675Suqs */ 517241675Suqs if (i == 3 && (!rebooting || bootverbose)) 518241675Suqs printf("Mount point %s had %d dangling refs\n", 519241675Suqs mp->mnt_stat.f_mntonname, mp->mnt_ref); 520241675Suqs if (mp->mnt_holdcnt != 0) { 521241675Suqs printf("Waiting for mount point to be unheld\n"); 522241675Suqs while (mp->mnt_holdcnt != 0) { 523241675Suqs mp->mnt_holdcntwaiters++; 524241675Suqs msleep(&mp->mnt_holdcnt, MNT_MTX(mp), 525241675Suqs PZERO, "mntdestroy", 0); 526241675Suqs mp->mnt_holdcntwaiters--; 527241675Suqs } 528241675Suqs printf("mount point unheld\n"); 529241675Suqs } 530241675Suqs if (mp->mnt_writeopcount > 0) { 531241675Suqs printf("Waiting for mount point write ops\n"); 532241675Suqs while (mp->mnt_writeopcount > 0) { 533241675Suqs mp->mnt_kern_flag |= MNTK_SUSPEND; 534241675Suqs msleep(&mp->mnt_writeopcount, 535241675Suqs MNT_MTX(mp), 536241675Suqs PZERO, "mntdestroy2", 0); 537241675Suqs } 538241675Suqs printf("mount point write ops completed\n"); 539241675Suqs } 540241675Suqs if (mp->mnt_secondary_writes > 0) { 541241675Suqs printf("Waiting for mount point secondary write ops\n"); 542241675Suqs while (mp->mnt_secondary_writes > 0) { 543241675Suqs mp->mnt_kern_flag |= MNTK_SUSPEND; 544241675Suqs msleep(&mp->mnt_secondary_writes, 545241675Suqs MNT_MTX(mp), 546241675Suqs PZERO, "mntdestroy3", 0); 547241675Suqs } 548241675Suqs printf("mount point secondary write ops completed\n"); 549241675Suqs } 550241675Suqs MNT_IUNLOCK(mp); 551241675Suqs mp->mnt_vfc->vfc_refcount--; 552241675Suqs if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 553241675Suqs struct vnode *vp; 554241675Suqs 555241675Suqs TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 556241675Suqs vprint("", vp); 557241675Suqs panic("unmount: dangling vnode"); 558241675Suqs } 559241675Suqs MNT_ILOCK(mp); 560241675Suqs if (mp->mnt_kern_flag & MNTK_MWAIT) 561241675Suqs wakeup(mp); 562241675Suqs if (mp->mnt_writeopcount != 0) 563241675Suqs panic("vfs_mount_destroy: nonzero writeopcount"); 564241675Suqs if (mp->mnt_secondary_writes != 0) 565241675Suqs panic("vfs_mount_destroy: nonzero secondary_writes"); 566241675Suqs if (mp->mnt_nvnodelistsize != 0) 567241675Suqs panic("vfs_mount_destroy: nonzero nvnodelistsize"); 568241675Suqs mp->mnt_writeopcount = -1000; 569241675Suqs mp->mnt_nvnodelistsize = -1000; 570241675Suqs mp->mnt_secondary_writes = -1000; 571241675Suqs MNT_IUNLOCK(mp); 572241675Suqs#ifdef MAC 573241675Suqs mac_mount_destroy(mp); 574241675Suqs#endif 575241675Suqs if (mp->mnt_opt != NULL) 576241675Suqs vfs_freeopts(mp->mnt_opt); 577241675Suqs crfree(mp->mnt_cred); 578241675Suqs uma_zfree(mount_zone, mp); 579241675Suqs} 580241675Suqs 581241675Suqsstatic int 582241675Suqsvfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 583241675Suqs{ 584241675Suqs struct vfsoptlist *optlist; 585241675Suqs struct vfsopt *opt, *noro_opt; 586241675Suqs char *fstype, *fspath, *errmsg; 587241675Suqs int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 588241675Suqs int has_rw, has_noro; 589241675Suqs 590241675Suqs errmsg = NULL; 591241675Suqs errmsg_len = 0; 592241675Suqs errmsg_pos = -1; 593241675Suqs has_rw = 0; 594241675Suqs has_noro = 0; 595241675Suqs 596241675Suqs error = vfs_buildopts(fsoptions, &optlist); 597241675Suqs if (error) 598241675Suqs return (error); 599241675Suqs 600241675Suqs if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 601241675Suqs errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 602241675Suqs 603241675Suqs /* 604241675Suqs * We need these two options before the others, 605241675Suqs * and they are mandatory for any filesystem. 606241675Suqs * Ensure they are NUL terminated as well. 607241675Suqs */ 608241675Suqs fstypelen = 0; 609241675Suqs error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 610241675Suqs if (error || fstype[fstypelen - 1] != '\0') { 611241675Suqs error = EINVAL; 612241675Suqs if (errmsg != NULL) 613241675Suqs strncpy(errmsg, "Invalid fstype", errmsg_len); 614241675Suqs goto bail; 615241675Suqs } 616241675Suqs fspathlen = 0; 617241675Suqs error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 618241675Suqs if (error || fspath[fspathlen - 1] != '\0') { 619241675Suqs error = EINVAL; 620241675Suqs if (errmsg != NULL) 621241675Suqs strncpy(errmsg, "Invalid fspath", errmsg_len); 622241675Suqs goto bail; 623241675Suqs } 624241675Suqs 625241675Suqs /* 626241675Suqs * We need to see if we have the "update" option 627241675Suqs * before we call vfs_domount(), since vfs_domount() has special 628241675Suqs * logic based on MNT_UPDATE. This is very important 629241675Suqs * when we want to update the root filesystem. 630241675Suqs */ 631241675Suqs TAILQ_FOREACH(opt, optlist, link) { 632241675Suqs if (strcmp(opt->name, "update") == 0) 633241675Suqs fsflags |= MNT_UPDATE; 634241675Suqs else if (strcmp(opt->name, "async") == 0) 635241675Suqs fsflags |= MNT_ASYNC; 636241675Suqs else if (strcmp(opt->name, "force") == 0) 637241675Suqs fsflags |= MNT_FORCE; 638241675Suqs else if (strcmp(opt->name, "multilabel") == 0) 639241675Suqs fsflags |= MNT_MULTILABEL; 640241675Suqs else if (strcmp(opt->name, "noasync") == 0) 641241675Suqs fsflags &= ~MNT_ASYNC; 642241675Suqs else if (strcmp(opt->name, "noatime") == 0) 643241675Suqs fsflags |= MNT_NOATIME; 644241675Suqs else if (strcmp(opt->name, "atime") == 0) { 645241675Suqs free(opt->name, M_MOUNT); 646241675Suqs opt->name = strdup("nonoatime", M_MOUNT); 647241675Suqs } 648241675Suqs else if (strcmp(opt->name, "noclusterr") == 0) 649241675Suqs fsflags |= MNT_NOCLUSTERR; 650241675Suqs else if (strcmp(opt->name, "clusterr") == 0) { 651241675Suqs free(opt->name, M_MOUNT); 652241675Suqs opt->name = strdup("nonoclusterr", M_MOUNT); 653241675Suqs } 654241675Suqs else if (strcmp(opt->name, "noclusterw") == 0) 655241675Suqs fsflags |= MNT_NOCLUSTERW; 656241675Suqs else if (strcmp(opt->name, "clusterw") == 0) { 657241675Suqs free(opt->name, M_MOUNT); 658241675Suqs opt->name = strdup("nonoclusterw", M_MOUNT); 659241675Suqs } 660241675Suqs else if (strcmp(opt->name, "noexec") == 0) 661241675Suqs fsflags |= MNT_NOEXEC; 662241675Suqs else if (strcmp(opt->name, "exec") == 0) { 663241675Suqs free(opt->name, M_MOUNT); 664241675Suqs opt->name = strdup("nonoexec", M_MOUNT); 665241675Suqs } 666241675Suqs else if (strcmp(opt->name, "nosuid") == 0) 667241675Suqs fsflags |= MNT_NOSUID; 668241675Suqs else if (strcmp(opt->name, "suid") == 0) { 669241675Suqs free(opt->name, M_MOUNT); 670241675Suqs opt->name = strdup("nonosuid", M_MOUNT); 671241675Suqs } 672241675Suqs else if (strcmp(opt->name, "nosymfollow") == 0) 673241675Suqs fsflags |= MNT_NOSYMFOLLOW; 674241675Suqs else if (strcmp(opt->name, "symfollow") == 0) { 675241675Suqs free(opt->name, M_MOUNT); 676241675Suqs opt->name = strdup("nonosymfollow", M_MOUNT); 677241675Suqs } 678241675Suqs else if (strcmp(opt->name, "noro") == 0) { 679241675Suqs fsflags &= ~MNT_RDONLY; 680241675Suqs has_noro = 1; 681241675Suqs } 682241675Suqs else if (strcmp(opt->name, "rw") == 0) { 683241675Suqs fsflags &= ~MNT_RDONLY; 684241675Suqs has_rw = 1; 685241675Suqs } 686241675Suqs else if (strcmp(opt->name, "ro") == 0) 687241675Suqs fsflags |= MNT_RDONLY; 688241675Suqs else if (strcmp(opt->name, "rdonly") == 0) { 689241675Suqs free(opt->name, M_MOUNT); 690241675Suqs opt->name = strdup("ro", M_MOUNT); 691241675Suqs fsflags |= MNT_RDONLY; 692241675Suqs } 693241675Suqs else if (strcmp(opt->name, "snapshot") == 0) 694241675Suqs fsflags |= MNT_SNAPSHOT; 695241675Suqs else if (strcmp(opt->name, "suiddir") == 0) 696241675Suqs fsflags |= MNT_SUIDDIR; 697241675Suqs else if (strcmp(opt->name, "sync") == 0) 698241675Suqs fsflags |= MNT_SYNCHRONOUS; 699241675Suqs else if (strcmp(opt->name, "union") == 0) 700241675Suqs fsflags |= MNT_UNION; 701241675Suqs } 702241675Suqs 703241675Suqs /* 704241675Suqs * If "rw" was specified as a mount option, and we 705241675Suqs * are trying to update a mount-point from "ro" to "rw", 706241675Suqs * we need a mount option "noro", since in vfs_mergeopts(), 707241675Suqs * "noro" will cancel "ro", but "rw" will not do anything. 708241675Suqs */ 709241675Suqs if (has_rw && !has_noro) { 710241675Suqs noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 711241675Suqs noro_opt->name = strdup("noro", M_MOUNT); 712241675Suqs noro_opt->value = NULL; 713241675Suqs noro_opt->len = 0; 714241675Suqs TAILQ_INSERT_TAIL(optlist, noro_opt, link); 715241675Suqs } 716241675Suqs 717241675Suqs /* 718241675Suqs * Be ultra-paranoid about making sure the type and fspath 719241675Suqs * variables will fit in our mp buffers, including the 720241675Suqs * terminating NUL. 721241675Suqs */ 722241675Suqs if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 723241675Suqs error = ENAMETOOLONG; 724241675Suqs goto bail; 725241675Suqs } 726241675Suqs 727241675Suqs mtx_lock(&Giant); 728241675Suqs error = vfs_domount(td, fstype, fspath, fsflags, optlist); 729241675Suqs mtx_unlock(&Giant); 730241675Suqsbail: 731241675Suqs /* copyout the errmsg */ 732241675Suqs if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 733241675Suqs && errmsg_len > 0 && errmsg != NULL) { 734241675Suqs if (fsoptions->uio_segflg == UIO_SYSSPACE) { 735241675Suqs bcopy(errmsg, 736 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 737 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 738 } else { 739 copyout(errmsg, 740 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 741 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 742 } 743 } 744 745 if (error != 0) 746 vfs_freeopts(optlist); 747 return (error); 748} 749 750/* 751 * Old mount API. 752 */ 753#ifndef _SYS_SYSPROTO_H_ 754struct mount_args { 755 char *type; 756 char *path; 757 int flags; 758 caddr_t data; 759}; 760#endif 761/* ARGSUSED */ 762int 763mount(td, uap) 764 struct thread *td; 765 struct mount_args /* { 766 char *type; 767 char *path; 768 int flags; 769 caddr_t data; 770 } */ *uap; 771{ 772 char *fstype; 773 struct vfsconf *vfsp = NULL; 774 struct mntarg *ma = NULL; 775 int error; 776 777 AUDIT_ARG(fflags, uap->flags); 778 779 /* 780 * Filter out MNT_ROOTFS. We do not want clients of mount() in 781 * userspace to set this flag, but we must filter it out if we want 782 * MNT_UPDATE on the root file system to work. 783 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 784 */ 785 uap->flags &= ~MNT_ROOTFS; 786 787 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 788 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 789 if (error) { 790 free(fstype, M_TEMP); 791 return (error); 792 } 793 794 AUDIT_ARG(text, fstype); 795 mtx_lock(&Giant); 796 vfsp = vfs_byname_kld(fstype, td, &error); 797 free(fstype, M_TEMP); 798 if (vfsp == NULL) { 799 mtx_unlock(&Giant); 800 return (ENOENT); 801 } 802 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 803 mtx_unlock(&Giant); 804 return (EOPNOTSUPP); 805 } 806 807 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 808 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 809 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 810 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 811 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 812 813 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 814 mtx_unlock(&Giant); 815 return (error); 816} 817 818 819/* 820 * vfs_domount(): actually attempt a filesystem mount. 821 */ 822static int 823vfs_domount( 824 struct thread *td, /* Calling thread. */ 825 const char *fstype, /* Filesystem type. */ 826 char *fspath, /* Mount path. */ 827 int fsflags, /* Flags common to all filesystems. */ 828 void *fsdata /* Options local to the filesystem. */ 829 ) 830{ 831 struct vnode *vp; 832 struct mount *mp; 833 struct vfsconf *vfsp; 834 struct export_args export; 835 int error, flag = 0; 836 struct vattr va; 837 struct nameidata nd; 838 839 mtx_assert(&Giant, MA_OWNED); 840 /* 841 * Be ultra-paranoid about making sure the type and fspath 842 * variables will fit in our mp buffers, including the 843 * terminating NUL. 844 */ 845 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 846 return (ENAMETOOLONG); 847 848 if (jailed(td->td_ucred) || usermount == 0) { 849 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 850 return (error); 851 } 852 853 /* 854 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 855 */ 856 if (fsflags & MNT_EXPORTED) { 857 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 858 if (error) 859 return (error); 860 } 861 if (fsflags & MNT_SUIDDIR) { 862 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 863 if (error) 864 return (error); 865 } 866 /* 867 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 868 */ 869 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 870 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 871 fsflags |= MNT_NOSUID | MNT_USER; 872 } 873 874 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 875 vfsp = NULL; 876 if ((fsflags & MNT_UPDATE) == 0) { 877 /* Don't try to load KLDs if we're mounting the root. */ 878 if (fsflags & MNT_ROOTFS) 879 vfsp = vfs_byname(fstype); 880 else 881 vfsp = vfs_byname_kld(fstype, td, &error); 882 if (vfsp == NULL) 883 return (ENODEV); 884 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 885 return (EPERM); 886 } 887 /* 888 * Get vnode to be covered 889 */ 890 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 891 fspath, td); 892 if ((error = namei(&nd)) != 0) 893 return (error); 894 NDFREE(&nd, NDF_ONLY_PNBUF); 895 vp = nd.ni_vp; 896 if (fsflags & MNT_UPDATE) { 897 if ((vp->v_vflag & VV_ROOT) == 0) { 898 vput(vp); 899 return (EINVAL); 900 } 901 mp = vp->v_mount; 902 MNT_ILOCK(mp); 903 flag = mp->mnt_flag; 904 /* 905 * We only allow the filesystem to be reloaded if it 906 * is currently mounted read-only. 907 */ 908 if ((fsflags & MNT_RELOAD) && 909 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 910 MNT_IUNLOCK(mp); 911 vput(vp); 912 return (EOPNOTSUPP); /* Needs translation */ 913 } 914 MNT_IUNLOCK(mp); 915 /* 916 * Only privileged root, or (if MNT_USER is set) the user that 917 * did the original mount is permitted to update it. 918 */ 919 error = vfs_suser(mp, td); 920 if (error) { 921 vput(vp); 922 return (error); 923 } 924 if (vfs_busy(mp, LK_NOWAIT, 0, td)) { 925 vput(vp); 926 return (EBUSY); 927 } 928 VI_LOCK(vp); 929 if ((vp->v_iflag & VI_MOUNT) != 0 || 930 vp->v_mountedhere != NULL) { 931 VI_UNLOCK(vp); 932 vfs_unbusy(mp, td); 933 vput(vp); 934 return (EBUSY); 935 } 936 vp->v_iflag |= VI_MOUNT; 937 VI_UNLOCK(vp); 938 MNT_ILOCK(mp); 939 mp->mnt_flag |= fsflags & 940 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 941 MNT_IUNLOCK(mp); 942 VOP_UNLOCK(vp, 0); 943 mp->mnt_optnew = fsdata; 944 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 945 } else { 946 /* 947 * If the user is not root, ensure that they own the directory 948 * onto which we are attempting to mount. 949 */ 950 error = VOP_GETATTR(vp, &va, td->td_ucred, td); 951 if (error) { 952 vput(vp); 953 return (error); 954 } 955 if (va.va_uid != td->td_ucred->cr_uid) { 956 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 957 0); 958 if (error) { 959 vput(vp); 960 return (error); 961 } 962 } 963 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 964 if (error != 0) { 965 vput(vp); 966 return (error); 967 } 968 if (vp->v_type != VDIR) { 969 vput(vp); 970 return (ENOTDIR); 971 } 972 VI_LOCK(vp); 973 if ((vp->v_iflag & VI_MOUNT) != 0 || 974 vp->v_mountedhere != NULL) { 975 VI_UNLOCK(vp); 976 vput(vp); 977 return (EBUSY); 978 } 979 vp->v_iflag |= VI_MOUNT; 980 VI_UNLOCK(vp); 981 982 /* 983 * Allocate and initialize the filesystem. 984 */ 985 mp = vfs_mount_alloc(vp, vfsp, fspath, td); 986 VOP_UNLOCK(vp, 0); 987 988 /* XXXMAC: pass to vfs_mount_alloc? */ 989 mp->mnt_optnew = fsdata; 990 } 991 992 /* 993 * Set the mount level flags. 994 */ 995 MNT_ILOCK(mp); 996 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 997 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 998 MNT_RDONLY)); 999 if ((mp->mnt_flag & MNT_ASYNC) == 0) 1000 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1001 MNT_IUNLOCK(mp); 1002 /* 1003 * Mount the filesystem. 1004 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1005 * get. No freeing of cn_pnbuf. 1006 */ 1007 error = VFS_MOUNT(mp, td); 1008 1009 /* 1010 * Process the export option only if we are 1011 * updating mount options. 1012 */ 1013 if (!error && (fsflags & MNT_UPDATE)) { 1014 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 1015 sizeof(export)) == 0) 1016 error = vfs_export(mp, &export); 1017 } 1018 1019 if (!error) { 1020 if (mp->mnt_opt != NULL) 1021 vfs_freeopts(mp->mnt_opt); 1022 mp->mnt_opt = mp->mnt_optnew; 1023 (void)VFS_STATFS(mp, &mp->mnt_stat, td); 1024 } 1025 /* 1026 * Prevent external consumers of mount options from reading 1027 * mnt_optnew. 1028 */ 1029 mp->mnt_optnew = NULL; 1030 if (mp->mnt_flag & MNT_UPDATE) { 1031 MNT_ILOCK(mp); 1032 if (error) 1033 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1034 (flag & ~MNT_QUOTA); 1035 else 1036 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1037 MNT_FORCE | MNT_SNAPSHOT); 1038 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1039 mp->mnt_kern_flag |= MNTK_ASYNC; 1040 else 1041 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1042 MNT_IUNLOCK(mp); 1043 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1044 if (mp->mnt_syncer == NULL) 1045 error = vfs_allocate_syncvnode(mp); 1046 } else { 1047 if (mp->mnt_syncer != NULL) 1048 vrele(mp->mnt_syncer); 1049 mp->mnt_syncer = NULL; 1050 } 1051 vfs_unbusy(mp, td); 1052 VI_LOCK(vp); 1053 vp->v_iflag &= ~VI_MOUNT; 1054 VI_UNLOCK(vp); 1055 vrele(vp); 1056 return (error); 1057 } 1058 MNT_ILOCK(mp); 1059 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1060 mp->mnt_kern_flag |= MNTK_ASYNC; 1061 else 1062 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1063 MNT_IUNLOCK(mp); 1064 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1065 /* 1066 * Put the new filesystem on the mount list after root. 1067 */ 1068 cache_purge(vp); 1069 if (!error) { 1070 struct vnode *newdp; 1071 1072 VI_LOCK(vp); 1073 vp->v_iflag &= ~VI_MOUNT; 1074 VI_UNLOCK(vp); 1075 vp->v_mountedhere = mp; 1076 mtx_lock(&mountlist_mtx); 1077 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1078 mtx_unlock(&mountlist_mtx); 1079 vfs_event_signal(NULL, VQ_MOUNT, 0); 1080 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 1081 panic("mount: lost mount"); 1082 mountcheckdirs(vp, newdp); 1083 vput(newdp); 1084 VOP_UNLOCK(vp, 0); 1085 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1086 error = vfs_allocate_syncvnode(mp); 1087 vfs_unbusy(mp, td); 1088 if (error) 1089 vrele(vp); 1090 } else { 1091 VI_LOCK(vp); 1092 vp->v_iflag &= ~VI_MOUNT; 1093 VI_UNLOCK(vp); 1094 vfs_unbusy(mp, td); 1095 vfs_mount_destroy(mp); 1096 vput(vp); 1097 } 1098 return (error); 1099} 1100 1101/* 1102 * Unmount a filesystem. 1103 * 1104 * Note: unmount takes a path to the vnode mounted on as argument, not 1105 * special file (as before). 1106 */ 1107#ifndef _SYS_SYSPROTO_H_ 1108struct unmount_args { 1109 char *path; 1110 int flags; 1111}; 1112#endif 1113/* ARGSUSED */ 1114int 1115unmount(td, uap) 1116 struct thread *td; 1117 register struct unmount_args /* { 1118 char *path; 1119 int flags; 1120 } */ *uap; 1121{ 1122 struct mount *mp; 1123 char *pathbuf; 1124 int error, id0, id1; 1125 1126 if (jailed(td->td_ucred) || usermount == 0) { 1127 error = priv_check(td, PRIV_VFS_UNMOUNT); 1128 if (error) 1129 return (error); 1130 } 1131 1132 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1133 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1134 if (error) { 1135 free(pathbuf, M_TEMP); 1136 return (error); 1137 } 1138 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1139 mtx_lock(&Giant); 1140 if (uap->flags & MNT_BYFSID) { 1141 /* Decode the filesystem ID. */ 1142 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1143 mtx_unlock(&Giant); 1144 free(pathbuf, M_TEMP); 1145 return (EINVAL); 1146 } 1147 1148 mtx_lock(&mountlist_mtx); 1149 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1150 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1151 mp->mnt_stat.f_fsid.val[1] == id1) 1152 break; 1153 } 1154 mtx_unlock(&mountlist_mtx); 1155 } else { 1156 mtx_lock(&mountlist_mtx); 1157 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1158 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1159 break; 1160 } 1161 mtx_unlock(&mountlist_mtx); 1162 } 1163 free(pathbuf, M_TEMP); 1164 if (mp == NULL) { 1165 /* 1166 * Previously we returned ENOENT for a nonexistent path and 1167 * EINVAL for a non-mountpoint. We cannot tell these apart 1168 * now, so in the !MNT_BYFSID case return the more likely 1169 * EINVAL for compatibility. 1170 */ 1171 mtx_unlock(&Giant); 1172 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1173 } 1174 1175 /* 1176 * Don't allow unmounting the root filesystem. 1177 */ 1178 if (mp->mnt_flag & MNT_ROOTFS) { 1179 mtx_unlock(&Giant); 1180 return (EINVAL); 1181 } 1182 error = dounmount(mp, uap->flags, td); 1183 mtx_unlock(&Giant); 1184 return (error); 1185} 1186 1187/* 1188 * Do the actual filesystem unmount. 1189 */ 1190int 1191dounmount(mp, flags, td) 1192 struct mount *mp; 1193 int flags; 1194 struct thread *td; 1195{ 1196 struct vnode *coveredvp, *fsrootvp; 1197 int error; 1198 int async_flag; 1199 int mnt_gen_r; 1200 1201 mtx_assert(&Giant, MA_OWNED); 1202 1203 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1204 mnt_gen_r = mp->mnt_gen; 1205 VI_LOCK(coveredvp); 1206 vholdl(coveredvp); 1207 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1208 vdrop(coveredvp); 1209 /* 1210 * Check for mp being unmounted while waiting for the 1211 * covered vnode lock. 1212 */ 1213 if (coveredvp->v_mountedhere != mp || 1214 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1215 VOP_UNLOCK(coveredvp, 0); 1216 return (EBUSY); 1217 } 1218 } 1219 /* 1220 * Only privileged root, or (if MNT_USER is set) the user that did the 1221 * original mount is permitted to unmount this filesystem. 1222 */ 1223 error = vfs_suser(mp, td); 1224 if (error) { 1225 if (coveredvp) 1226 VOP_UNLOCK(coveredvp, 0); 1227 return (error); 1228 } 1229 1230 MNT_ILOCK(mp); 1231 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1232 MNT_IUNLOCK(mp); 1233 if (coveredvp) 1234 VOP_UNLOCK(coveredvp, 0); 1235 return (EBUSY); 1236 } 1237 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1238 /* Allow filesystems to detect that a forced unmount is in progress. */ 1239 if (flags & MNT_FORCE) 1240 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1241 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 1242 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td); 1243 if (error) { 1244 MNT_ILOCK(mp); 1245 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1246 MNTK_UNMOUNTF); 1247 if (mp->mnt_kern_flag & MNTK_MWAIT) 1248 wakeup(mp); 1249 MNT_IUNLOCK(mp); 1250 if (coveredvp) 1251 VOP_UNLOCK(coveredvp, 0); 1252 return (error); 1253 } 1254 vn_start_write(NULL, &mp, V_WAIT); 1255 1256 if (mp->mnt_flag & MNT_EXPUBLIC) 1257 vfs_setpublicfs(NULL, NULL, NULL); 1258 1259 vfs_msync(mp, MNT_WAIT); 1260 MNT_ILOCK(mp); 1261 async_flag = mp->mnt_flag & MNT_ASYNC; 1262 mp->mnt_flag &= ~MNT_ASYNC; 1263 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1264 MNT_IUNLOCK(mp); 1265 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1266 if (mp->mnt_syncer != NULL) 1267 vrele(mp->mnt_syncer); 1268 /* 1269 * For forced unmounts, move process cdir/rdir refs on the fs root 1270 * vnode to the covered vnode. For non-forced unmounts we want 1271 * such references to cause an EBUSY error. 1272 */ 1273 if ((flags & MNT_FORCE) && 1274 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1275 if (mp->mnt_vnodecovered != NULL) 1276 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1277 if (fsrootvp == rootvnode) { 1278 vrele(rootvnode); 1279 rootvnode = NULL; 1280 } 1281 vput(fsrootvp); 1282 } 1283 if (((mp->mnt_flag & MNT_RDONLY) || 1284 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1285 (flags & MNT_FORCE)) { 1286 error = VFS_UNMOUNT(mp, flags, td); 1287 } 1288 vn_finished_write(mp); 1289 /* 1290 * If we failed to flush the dirty blocks for this mount point, 1291 * undo all the cdir/rdir and rootvnode changes we made above. 1292 * Unless we failed to do so because the device is reporting that 1293 * it doesn't exist anymore. 1294 */ 1295 if (error && error != ENXIO) { 1296 if ((flags & MNT_FORCE) && 1297 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1298 if (mp->mnt_vnodecovered != NULL) 1299 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1300 if (rootvnode == NULL) { 1301 rootvnode = fsrootvp; 1302 vref(rootvnode); 1303 } 1304 vput(fsrootvp); 1305 } 1306 MNT_ILOCK(mp); 1307 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1308 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1309 MNT_IUNLOCK(mp); 1310 (void) vfs_allocate_syncvnode(mp); 1311 MNT_ILOCK(mp); 1312 } 1313 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1314 mp->mnt_flag |= async_flag; 1315 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1316 mp->mnt_kern_flag |= MNTK_ASYNC; 1317 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1318 if (mp->mnt_kern_flag & MNTK_MWAIT) 1319 wakeup(mp); 1320 MNT_IUNLOCK(mp); 1321 if (coveredvp) 1322 VOP_UNLOCK(coveredvp, 0); 1323 return (error); 1324 } 1325 mtx_lock(&mountlist_mtx); 1326 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1327 mtx_unlock(&mountlist_mtx); 1328 if (coveredvp != NULL) { 1329 coveredvp->v_mountedhere = NULL; 1330 vput(coveredvp); 1331 } 1332 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1333 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1334 vfs_mount_destroy(mp); 1335 return (0); 1336} 1337 1338/* 1339 * --------------------------------------------------------------------- 1340 * Mounting of root filesystem 1341 * 1342 */ 1343 1344struct root_hold_token { 1345 const char *who; 1346 LIST_ENTRY(root_hold_token) list; 1347}; 1348 1349static LIST_HEAD(, root_hold_token) root_holds = 1350 LIST_HEAD_INITIALIZER(&root_holds); 1351 1352static int root_mount_complete; 1353 1354/* 1355 * Hold root mount. 1356 */ 1357struct root_hold_token * 1358root_mount_hold(const char *identifier) 1359{ 1360 struct root_hold_token *h; 1361 1362 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1363 h->who = identifier; 1364 mtx_lock(&mountlist_mtx); 1365 LIST_INSERT_HEAD(&root_holds, h, list); 1366 mtx_unlock(&mountlist_mtx); 1367 return (h); 1368} 1369 1370/* 1371 * Release root mount. 1372 */ 1373void 1374root_mount_rel(struct root_hold_token *h) 1375{ 1376 1377 mtx_lock(&mountlist_mtx); 1378 LIST_REMOVE(h, list); 1379 wakeup(&root_holds); 1380 mtx_unlock(&mountlist_mtx); 1381 free(h, M_DEVBUF); 1382} 1383 1384/* 1385 * Wait for all subsystems to release root mount. 1386 */ 1387static void 1388root_mount_prepare(void) 1389{ 1390 struct root_hold_token *h; 1391 1392 for (;;) { 1393 DROP_GIANT(); 1394 g_waitidle(); 1395 PICKUP_GIANT(); 1396 mtx_lock(&mountlist_mtx); 1397 if (LIST_EMPTY(&root_holds)) { 1398 mtx_unlock(&mountlist_mtx); 1399 break; 1400 } 1401 printf("Root mount waiting for:"); 1402 LIST_FOREACH(h, &root_holds, list) 1403 printf(" %s", h->who); 1404 printf("\n"); 1405 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1406 hz); 1407 } 1408} 1409 1410/* 1411 * Root was mounted, share the good news. 1412 */ 1413static void 1414root_mount_done(void) 1415{ 1416 1417 /* 1418 * Use a mutex to prevent the wakeup being missed and waiting for 1419 * an extra 1 second sleep. 1420 */ 1421 mtx_lock(&mountlist_mtx); 1422 root_mount_complete = 1; 1423 wakeup(&root_mount_complete); 1424 mtx_unlock(&mountlist_mtx); 1425} 1426 1427/* 1428 * Return true if root is already mounted. 1429 */ 1430int 1431root_mounted(void) 1432{ 1433 1434 /* No mutex is acquired here because int stores are atomic. */ 1435 return (root_mount_complete); 1436} 1437 1438/* 1439 * Wait until root is mounted. 1440 */ 1441void 1442root_mount_wait(void) 1443{ 1444 1445 /* 1446 * Panic on an obvious deadlock - the function can't be called from 1447 * a thread which is doing the whole SYSINIT stuff. 1448 */ 1449 KASSERT(curthread->td_proc->p_pid != 0, 1450 ("root_mount_wait: cannot be called from the swapper thread")); 1451 mtx_lock(&mountlist_mtx); 1452 while (!root_mount_complete) { 1453 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1454 hz); 1455 } 1456 mtx_unlock(&mountlist_mtx); 1457} 1458 1459static void 1460set_rootvnode(struct thread *td) 1461{ 1462 struct proc *p; 1463 1464 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1465 panic("Cannot find root vnode"); 1466 1467 p = td->td_proc; 1468 FILEDESC_SLOCK(p->p_fd); 1469 1470 if (p->p_fd->fd_cdir != NULL) 1471 vrele(p->p_fd->fd_cdir); 1472 p->p_fd->fd_cdir = rootvnode; 1473 VREF(rootvnode); 1474 1475 if (p->p_fd->fd_rdir != NULL) 1476 vrele(p->p_fd->fd_rdir); 1477 p->p_fd->fd_rdir = rootvnode; 1478 VREF(rootvnode); 1479 1480 FILEDESC_SUNLOCK(p->p_fd); 1481 1482 VOP_UNLOCK(rootvnode, 0); 1483} 1484 1485/* 1486 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1487 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1488 */ 1489 1490static void 1491devfs_first(void) 1492{ 1493 struct thread *td = curthread; 1494 struct vfsoptlist *opts; 1495 struct vfsconf *vfsp; 1496 struct mount *mp = NULL; 1497 int error; 1498 1499 vfsp = vfs_byname("devfs"); 1500 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1501 if (vfsp == NULL) 1502 return; 1503 1504 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td); 1505 1506 error = VFS_MOUNT(mp, td); 1507 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1508 if (error) 1509 return; 1510 1511 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1512 TAILQ_INIT(opts); 1513 mp->mnt_opt = opts; 1514 1515 mtx_lock(&mountlist_mtx); 1516 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1517 mtx_unlock(&mountlist_mtx); 1518 1519 set_rootvnode(td); 1520 1521 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1522 if (error) 1523 printf("kern_symlink /dev -> / returns %d\n", error); 1524} 1525 1526/* 1527 * Surgically move our devfs to be mounted on /dev. 1528 */ 1529 1530static void 1531devfs_fixup(struct thread *td) 1532{ 1533 struct nameidata nd; 1534 int error; 1535 struct vnode *vp, *dvp; 1536 struct mount *mp; 1537 1538 /* Remove our devfs mount from the mountlist and purge the cache */ 1539 mtx_lock(&mountlist_mtx); 1540 mp = TAILQ_FIRST(&mountlist); 1541 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1542 mtx_unlock(&mountlist_mtx); 1543 cache_purgevfs(mp); 1544 1545 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1546 VI_LOCK(dvp); 1547 dvp->v_iflag &= ~VI_MOUNT; 1548 VI_UNLOCK(dvp); 1549 dvp->v_mountedhere = NULL; 1550 1551 /* Set up the real rootvnode, and purge the cache */ 1552 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1553 set_rootvnode(td); 1554 cache_purgevfs(rootvnode->v_mount); 1555 1556 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1557 error = namei(&nd); 1558 if (error) { 1559 printf("Lookup of /dev for devfs, error: %d\n", error); 1560 return; 1561 } 1562 NDFREE(&nd, NDF_ONLY_PNBUF); 1563 vp = nd.ni_vp; 1564 if (vp->v_type != VDIR) { 1565 vput(vp); 1566 } 1567 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 1568 if (error) { 1569 vput(vp); 1570 } 1571 cache_purge(vp); 1572 mp->mnt_vnodecovered = vp; 1573 vp->v_mountedhere = mp; 1574 mtx_lock(&mountlist_mtx); 1575 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1576 mtx_unlock(&mountlist_mtx); 1577 VOP_UNLOCK(vp, 0); 1578 vput(dvp); 1579 vfs_unbusy(mp, td); 1580 1581 /* Unlink the no longer needed /dev/dev -> / symlink */ 1582 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1583} 1584 1585/* 1586 * Report errors during filesystem mounting. 1587 */ 1588void 1589vfs_mount_error(struct mount *mp, const char *fmt, ...) 1590{ 1591 struct vfsoptlist *moptlist = mp->mnt_optnew; 1592 va_list ap; 1593 int error, len; 1594 char *errmsg; 1595 1596 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1597 if (error || errmsg == NULL || len <= 0) 1598 return; 1599 1600 va_start(ap, fmt); 1601 vsnprintf(errmsg, (size_t)len, fmt, ap); 1602 va_end(ap); 1603} 1604 1605/* 1606 * Find and mount the root filesystem 1607 */ 1608void 1609vfs_mountroot(void) 1610{ 1611 char *cp; 1612 int error, i, asked = 0; 1613 1614 root_mount_prepare(); 1615 1616 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1617 NULL, NULL, mount_init, mount_fini, 1618 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1619 devfs_first(); 1620 1621 /* 1622 * We are booted with instructions to prompt for the root filesystem. 1623 */ 1624 if (boothowto & RB_ASKNAME) { 1625 if (!vfs_mountroot_ask()) 1626 goto mounted; 1627 asked = 1; 1628 } 1629 1630 /* 1631 * The root filesystem information is compiled in, and we are 1632 * booted with instructions to use it. 1633 */ 1634 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1635 if (!vfs_mountroot_try(ctrootdevname)) 1636 goto mounted; 1637 ctrootdevname = NULL; 1638 } 1639 1640 /* 1641 * We've been given the generic "use CDROM as root" flag. This is 1642 * necessary because one media may be used in many different 1643 * devices, so we need to search for them. 1644 */ 1645 if (boothowto & RB_CDROM) { 1646 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1647 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1648 goto mounted; 1649 } 1650 } 1651 1652 /* 1653 * Try to use the value read by the loader from /etc/fstab, or 1654 * supplied via some other means. This is the preferred 1655 * mechanism. 1656 */ 1657 cp = getenv("vfs.root.mountfrom"); 1658 if (cp != NULL) { 1659 error = vfs_mountroot_try(cp); 1660 freeenv(cp); 1661 if (!error) 1662 goto mounted; 1663 } 1664 1665 /* 1666 * Try values that may have been computed by code during boot 1667 */ 1668 if (!vfs_mountroot_try(rootdevnames[0])) 1669 goto mounted; 1670 if (!vfs_mountroot_try(rootdevnames[1])) 1671 goto mounted; 1672 1673 /* 1674 * If we (still) have a compiled-in default, try it. 1675 */ 1676 if (ctrootdevname != NULL) 1677 if (!vfs_mountroot_try(ctrootdevname)) 1678 goto mounted; 1679 /* 1680 * Everything so far has failed, prompt on the console if we haven't 1681 * already tried that. 1682 */ 1683 if (!asked) 1684 if (!vfs_mountroot_ask()) 1685 goto mounted; 1686 1687 panic("Root mount failed, startup aborted."); 1688 1689mounted: 1690 root_mount_done(); 1691} 1692 1693/* 1694 * Mount (mountfrom) as the root filesystem. 1695 */ 1696static int 1697vfs_mountroot_try(const char *mountfrom) 1698{ 1699 struct mount *mp; 1700 char *vfsname, *path; 1701 time_t timebase; 1702 int error; 1703 char patt[32]; 1704 1705 vfsname = NULL; 1706 path = NULL; 1707 mp = NULL; 1708 error = EINVAL; 1709 1710 if (mountfrom == NULL) 1711 return (error); /* don't complain */ 1712 printf("Trying to mount root from %s\n", mountfrom); 1713 1714 /* parse vfs name and path */ 1715 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1716 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1717 vfsname[0] = path[0] = 0; 1718 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1719 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1720 goto out; 1721 1722 if (path[0] == '\0') 1723 strcpy(path, ROOTNAME); 1724 1725 error = kernel_vmount( 1726 MNT_RDONLY | MNT_ROOTFS, 1727 "fstype", vfsname, 1728 "fspath", "/", 1729 "from", path, 1730 NULL); 1731 if (error == 0) { 1732 /* 1733 * We mount devfs prior to mounting the / FS, so the first 1734 * entry will typically be devfs. 1735 */ 1736 mp = TAILQ_FIRST(&mountlist); 1737 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1738 1739 /* 1740 * Iterate over all currently mounted file systems and use 1741 * the time stamp found to check and/or initialize the RTC. 1742 * Typically devfs has no time stamp and the only other FS 1743 * is the actual / FS. 1744 * Call inittodr() only once and pass it the largest of the 1745 * timestamps we encounter. 1746 */ 1747 timebase = 0; 1748 do { 1749 if (mp->mnt_time > timebase) 1750 timebase = mp->mnt_time; 1751 mp = TAILQ_NEXT(mp, mnt_list); 1752 } while (mp != NULL); 1753 inittodr(timebase); 1754 1755 devfs_fixup(curthread); 1756 } 1757out: 1758 free(path, M_MOUNT); 1759 free(vfsname, M_MOUNT); 1760 return (error); 1761} 1762 1763/* 1764 * --------------------------------------------------------------------- 1765 * Interactive root filesystem selection code. 1766 */ 1767 1768static int 1769vfs_mountroot_ask(void) 1770{ 1771 char name[128]; 1772 1773 for(;;) { 1774 printf("\nManual root filesystem specification:\n"); 1775 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1776#if defined(__amd64__) || defined(__i386__) || defined(__ia64__) 1777 printf(" eg. ufs:da0s1a\n"); 1778#else 1779 printf(" eg. ufs:/dev/da0a\n"); 1780#endif 1781 printf(" ? List valid disk boot devices\n"); 1782 printf(" <empty line> Abort manual input\n"); 1783 printf("\nmountroot> "); 1784 gets(name, sizeof(name), 1); 1785 if (name[0] == '\0') 1786 return (1); 1787 if (name[0] == '?') { 1788 printf("\nList of GEOM managed disk devices:\n "); 1789 g_dev_print(); 1790 continue; 1791 } 1792 if (!vfs_mountroot_try(name)) 1793 return (0); 1794 } 1795} 1796 1797/* 1798 * --------------------------------------------------------------------- 1799 * Functions for querying mount options/arguments from filesystems. 1800 */ 1801 1802/* 1803 * Check that no unknown options are given 1804 */ 1805int 1806vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1807{ 1808 struct vfsopt *opt; 1809 char errmsg[255]; 1810 const char **t, *p, *q; 1811 int ret = 0; 1812 1813 TAILQ_FOREACH(opt, opts, link) { 1814 p = opt->name; 1815 q = NULL; 1816 if (p[0] == 'n' && p[1] == 'o') 1817 q = p + 2; 1818 for(t = global_opts; *t != NULL; t++) { 1819 if (strcmp(*t, p) == 0) 1820 break; 1821 if (q != NULL) { 1822 if (strcmp(*t, q) == 0) 1823 break; 1824 } 1825 } 1826 if (*t != NULL) 1827 continue; 1828 for(t = legal; *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 sprintf(errmsg, "mount option <%s> is unknown", p); 1839 printf("%s\n", errmsg); 1840 ret = EINVAL; 1841 } 1842 if (ret != 0) { 1843 TAILQ_FOREACH(opt, opts, link) { 1844 if (strcmp(opt->name, "errmsg") == 0) { 1845 strncpy((char *)opt->value, errmsg, opt->len); 1846 } 1847 } 1848 } 1849 return (ret); 1850} 1851 1852/* 1853 * Get a mount option by its name. 1854 * 1855 * Return 0 if the option was found, ENOENT otherwise. 1856 * If len is non-NULL it will be filled with the length 1857 * of the option. If buf is non-NULL, it will be filled 1858 * with the address of the option. 1859 */ 1860int 1861vfs_getopt(opts, name, buf, len) 1862 struct vfsoptlist *opts; 1863 const char *name; 1864 void **buf; 1865 int *len; 1866{ 1867 struct vfsopt *opt; 1868 1869 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1870 1871 TAILQ_FOREACH(opt, opts, link) { 1872 if (strcmp(name, opt->name) == 0) { 1873 if (len != NULL) 1874 *len = opt->len; 1875 if (buf != NULL) 1876 *buf = opt->value; 1877 return (0); 1878 } 1879 } 1880 return (ENOENT); 1881} 1882 1883static int 1884vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1885{ 1886 struct vfsopt *opt; 1887 int i; 1888 1889 if (opts == NULL) 1890 return (-1); 1891 1892 i = 0; 1893 TAILQ_FOREACH(opt, opts, link) { 1894 if (strcmp(name, opt->name) == 0) 1895 return (i); 1896 ++i; 1897 } 1898 return (-1); 1899} 1900 1901char * 1902vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1903{ 1904 struct vfsopt *opt; 1905 1906 *error = 0; 1907 TAILQ_FOREACH(opt, opts, link) { 1908 if (strcmp(name, opt->name) != 0) 1909 continue; 1910 if (((char *)opt->value)[opt->len - 1] != '\0') { 1911 *error = EINVAL; 1912 return (NULL); 1913 } 1914 return (opt->value); 1915 } 1916 *error = ENOENT; 1917 return (NULL); 1918} 1919 1920int 1921vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1922{ 1923 struct vfsopt *opt; 1924 1925 TAILQ_FOREACH(opt, opts, link) { 1926 if (strcmp(name, opt->name) == 0) { 1927 if (w != NULL) 1928 *w |= val; 1929 return (1); 1930 } 1931 } 1932 if (w != NULL) 1933 *w &= ~val; 1934 return (0); 1935} 1936 1937int 1938vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1939{ 1940 va_list ap; 1941 struct vfsopt *opt; 1942 int ret; 1943 1944 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1945 1946 TAILQ_FOREACH(opt, opts, link) { 1947 if (strcmp(name, opt->name) != 0) 1948 continue; 1949 if (opt->len == 0 || opt->value == NULL) 1950 return (0); 1951 if (((char *)opt->value)[opt->len - 1] != '\0') 1952 return (0); 1953 va_start(ap, fmt); 1954 ret = vsscanf(opt->value, fmt, ap); 1955 va_end(ap); 1956 return (ret); 1957 } 1958 return (0); 1959} 1960 1961/* 1962 * Find and copy a mount option. 1963 * 1964 * The size of the buffer has to be specified 1965 * in len, if it is not the same length as the 1966 * mount option, EINVAL is returned. 1967 * Returns ENOENT if the option is not found. 1968 */ 1969int 1970vfs_copyopt(opts, name, dest, len) 1971 struct vfsoptlist *opts; 1972 const char *name; 1973 void *dest; 1974 int len; 1975{ 1976 struct vfsopt *opt; 1977 1978 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1979 1980 TAILQ_FOREACH(opt, opts, link) { 1981 if (strcmp(name, opt->name) == 0) { 1982 if (len != opt->len) 1983 return (EINVAL); 1984 bcopy(opt->value, dest, opt->len); 1985 return (0); 1986 } 1987 } 1988 return (ENOENT); 1989} 1990 1991/* 1992 * This is a helper function for filesystems to traverse their 1993 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1994 */ 1995 1996struct vnode * 1997__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 1998{ 1999 struct vnode *vp; 2000 2001 mtx_assert(MNT_MTX(mp), MA_OWNED); 2002 2003 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2004 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2005 while (vp != NULL && vp->v_type == VMARKER) 2006 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2007 2008 /* Check if we are done */ 2009 if (vp == NULL) { 2010 __mnt_vnode_markerfree(mvp, mp); 2011 return (NULL); 2012 } 2013 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2014 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2015 return (vp); 2016} 2017 2018struct vnode * 2019__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2020{ 2021 struct vnode *vp; 2022 2023 mtx_assert(MNT_MTX(mp), MA_OWNED); 2024 2025 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2026 while (vp != NULL && vp->v_type == VMARKER) 2027 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2028 2029 /* Check if we are done */ 2030 if (vp == NULL) { 2031 *mvp = NULL; 2032 return (NULL); 2033 } 2034 mp->mnt_holdcnt++; 2035 MNT_IUNLOCK(mp); 2036 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2037 M_VNODE_MARKER, 2038 M_WAITOK | M_ZERO); 2039 MNT_ILOCK(mp); 2040 (*mvp)->v_type = VMARKER; 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 MNT_IUNLOCK(mp); 2049 free(*mvp, M_VNODE_MARKER); 2050 MNT_ILOCK(mp); 2051 *mvp = NULL; 2052 mp->mnt_holdcnt--; 2053 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2054 wakeup(&mp->mnt_holdcnt); 2055 return (NULL); 2056 } 2057 mp->mnt_markercnt++; 2058 (*mvp)->v_mount = mp; 2059 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2060 return (vp); 2061} 2062 2063 2064void 2065__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2066{ 2067 2068 if (*mvp == NULL) 2069 return; 2070 2071 mtx_assert(MNT_MTX(mp), MA_OWNED); 2072 2073 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2074 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2075 MNT_IUNLOCK(mp); 2076 free(*mvp, M_VNODE_MARKER); 2077 MNT_ILOCK(mp); 2078 *mvp = NULL; 2079 2080 mp->mnt_markercnt--; 2081 mp->mnt_holdcnt--; 2082 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2083 wakeup(&mp->mnt_holdcnt); 2084} 2085 2086 2087int 2088__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 2089{ 2090 int error; 2091 2092 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 2093 if (sbp != &mp->mnt_stat) 2094 *sbp = mp->mnt_stat; 2095 return (error); 2096} 2097 2098void 2099vfs_mountedfrom(struct mount *mp, const char *from) 2100{ 2101 2102 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2103 strlcpy(mp->mnt_stat.f_mntfromname, from, 2104 sizeof mp->mnt_stat.f_mntfromname); 2105} 2106 2107/* 2108 * --------------------------------------------------------------------- 2109 * This is the api for building mount args and mounting filesystems from 2110 * inside the kernel. 2111 * 2112 * The API works by accumulation of individual args. First error is 2113 * latched. 2114 * 2115 * XXX: should be documented in new manpage kernel_mount(9) 2116 */ 2117 2118/* A memory allocation which must be freed when we are done */ 2119struct mntaarg { 2120 SLIST_ENTRY(mntaarg) next; 2121}; 2122 2123/* The header for the mount arguments */ 2124struct mntarg { 2125 struct iovec *v; 2126 int len; 2127 int error; 2128 SLIST_HEAD(, mntaarg) list; 2129}; 2130 2131/* 2132 * Add a boolean argument. 2133 * 2134 * flag is the boolean value. 2135 * name must start with "no". 2136 */ 2137struct mntarg * 2138mount_argb(struct mntarg *ma, int flag, const char *name) 2139{ 2140 2141 KASSERT(name[0] == 'n' && name[1] == 'o', 2142 ("mount_argb(...,%s): name must start with 'no'", name)); 2143 2144 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2145} 2146 2147/* 2148 * Add an argument printf style 2149 */ 2150struct mntarg * 2151mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2152{ 2153 va_list ap; 2154 struct mntaarg *maa; 2155 struct sbuf *sb; 2156 int len; 2157 2158 if (ma == NULL) { 2159 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2160 SLIST_INIT(&ma->list); 2161 } 2162 if (ma->error) 2163 return (ma); 2164 2165 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2166 M_MOUNT, M_WAITOK); 2167 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2168 ma->v[ma->len].iov_len = strlen(name) + 1; 2169 ma->len++; 2170 2171 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND); 2172 va_start(ap, fmt); 2173 sbuf_vprintf(sb, fmt, ap); 2174 va_end(ap); 2175 sbuf_finish(sb); 2176 len = sbuf_len(sb) + 1; 2177 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2178 SLIST_INSERT_HEAD(&ma->list, maa, next); 2179 bcopy(sbuf_data(sb), maa + 1, len); 2180 sbuf_delete(sb); 2181 2182 ma->v[ma->len].iov_base = maa + 1; 2183 ma->v[ma->len].iov_len = len; 2184 ma->len++; 2185 2186 return (ma); 2187} 2188 2189/* 2190 * Add an argument which is a userland string. 2191 */ 2192struct mntarg * 2193mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2194{ 2195 struct mntaarg *maa; 2196 char *tbuf; 2197 2198 if (val == NULL) 2199 return (ma); 2200 if (ma == NULL) { 2201 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2202 SLIST_INIT(&ma->list); 2203 } 2204 if (ma->error) 2205 return (ma); 2206 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2207 SLIST_INSERT_HEAD(&ma->list, maa, next); 2208 tbuf = (void *)(maa + 1); 2209 ma->error = copyinstr(val, tbuf, len, NULL); 2210 return (mount_arg(ma, name, tbuf, -1)); 2211} 2212 2213/* 2214 * Plain argument. 2215 * 2216 * If length is -1, use printf. 2217 */ 2218struct mntarg * 2219mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2220{ 2221 2222 if (ma == NULL) { 2223 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2224 SLIST_INIT(&ma->list); 2225 } 2226 if (ma->error) 2227 return (ma); 2228 2229 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2230 M_MOUNT, M_WAITOK); 2231 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2232 ma->v[ma->len].iov_len = strlen(name) + 1; 2233 ma->len++; 2234 2235 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2236 if (len < 0) 2237 ma->v[ma->len].iov_len = strlen(val) + 1; 2238 else 2239 ma->v[ma->len].iov_len = len; 2240 ma->len++; 2241 return (ma); 2242} 2243 2244/* 2245 * Free a mntarg structure 2246 */ 2247static void 2248free_mntarg(struct mntarg *ma) 2249{ 2250 struct mntaarg *maa; 2251 2252 while (!SLIST_EMPTY(&ma->list)) { 2253 maa = SLIST_FIRST(&ma->list); 2254 SLIST_REMOVE_HEAD(&ma->list, next); 2255 free(maa, M_MOUNT); 2256 } 2257 free(ma->v, M_MOUNT); 2258 free(ma, M_MOUNT); 2259} 2260 2261/* 2262 * Mount a filesystem 2263 */ 2264int 2265kernel_mount(struct mntarg *ma, int flags) 2266{ 2267 struct uio auio; 2268 int error; 2269 2270 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2271 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2272 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2273 2274 auio.uio_iov = ma->v; 2275 auio.uio_iovcnt = ma->len; 2276 auio.uio_segflg = UIO_SYSSPACE; 2277 2278 error = ma->error; 2279 if (!error) 2280 error = vfs_donmount(curthread, flags, &auio); 2281 free_mntarg(ma); 2282 return (error); 2283} 2284 2285/* 2286 * A printflike function to mount a filesystem. 2287 */ 2288int 2289kernel_vmount(int flags, ...) 2290{ 2291 struct mntarg *ma = NULL; 2292 va_list ap; 2293 const char *cp; 2294 const void *vp; 2295 int error; 2296 2297 va_start(ap, flags); 2298 for (;;) { 2299 cp = va_arg(ap, const char *); 2300 if (cp == NULL) 2301 break; 2302 vp = va_arg(ap, const void *); 2303 ma = mount_arg(ma, cp, vp, -1); 2304 } 2305 va_end(ap); 2306 2307 error = kernel_mount(ma, flags); 2308 return (error); 2309} 2310