vfs_mount.c revision 161584
11541Srgrimes/*- 21541Srgrimes * Copyright (c) 1999-2004 Poul-Henning Kamp 31541Srgrimes * Copyright (c) 1999 Michael Smith 41541Srgrimes * Copyright (c) 1989, 1993 51541Srgrimes * The Regents of the University of California. All rights reserved. 61541Srgrimes * (c) UNIX System Laboratories, Inc. 71541Srgrimes * All or some portions of this file are derived from material licensed 81541Srgrimes * to the University of California by American Telephone and Telegraph 91541Srgrimes * Co. or Unix System Laboratories, Inc. and are reproduced herein with 101541Srgrimes * the permission of UNIX System Laboratories, Inc. 111541Srgrimes * 121541Srgrimes * Redistribution and use in source and binary forms, with or without 131541Srgrimes * modification, are permitted provided that the following conditions 141541Srgrimes * are met: 151541Srgrimes * 1. Redistributions of source code must retain the above copyright 161541Srgrimes * notice, this list of conditions and the following disclaimer. 171541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 181541Srgrimes * notice, this list of conditions and the following disclaimer in the 191541Srgrimes * documentation and/or other materials provided with the distribution. 201541Srgrimes * 4. Neither the name of the University nor the names of its contributors 211541Srgrimes * may be used to endorse or promote products derived from this software 221541Srgrimes * without specific prior written permission. 231541Srgrimes * 241541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 251541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 261541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 271541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 281541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 291541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 301541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 311541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 321541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 331541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 341541Srgrimes * SUCH DAMAGE. 351541Srgrimes */ 361541Srgrimes 37116182Sobrien#include <sys/cdefs.h> 38116182Sobrien__FBSDID("$FreeBSD: head/sys/kern/vfs_mount.c 161584 2006-08-24 18:52:28Z marius $"); 39116182Sobrien 40170075Semaste#include <sys/param.h> 41150968Sglebius#include <sys/conf.h> 42147565Speter#include <sys/jail.h> 4344666Sphk#include <sys/kernel.h> 44116874Ssmkelly#include <sys/libkern.h> 4544666Sphk#include <sys/mac.h> 461541Srgrimes#include <sys/malloc.h> 471541Srgrimes#include <sys/mount.h> 481541Srgrimes#include <sys/mutex.h> 49131927Smarcel#include <sys/namei.h> 501541Srgrimes#include <sys/proc.h> 51201879Sattilio#include <sys/filedesc.h> 52201879Sattilio#include <sys/reboot.h> 5374914Sjhb#include <sys/syscallsubr.h> 5467365Sjhb#include <sys/sysproto.h> 551541Srgrimes#include <sys/sx.h> 56111024Sjeff#include <sys/sysctl.h> 571541Srgrimes#include <sys/sysent.h> 58104964Sjeff#include <sys/systm.h> 593308Sphk#include <sys/vnode.h> 60201879Sattilio#include <vm/uma.h> 6176078Sjhb 622320Sdg#include <geom/geom.h> 6312662Sdg 6412662Sdg#include <machine/stdarg.h> 653308Sphk 6667551Sjhb#include <security/audit/audit.h> 6767551Sjhb 68114216Skan#include "opt_rootdevname.h" 69102926Sphk#include "opt_ddb.h" 701541Srgrimes#include "opt_mac.h" 711541Srgrimes 721541Srgrimes#ifdef DDB 731541Srgrimes#include <ddb/ddb.h> 741541Srgrimes#endif 75146799Sjkoshy 76146799Sjkoshy#define ROOTNAME "root_device" 77233628Sfabient#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 78233628Sfabient 79146799Sjkoshystatic int vfs_domount(struct thread *td, const char *fstype, 80146799Sjkoshy char *fspath, int fsflags, void *fsdata); 8187902Sluigistatic struct mount *vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp, 8287902Sluigi const char *fspath, struct thread *td); 8387902Sluigistatic int vfs_mountroot_ask(void); 8431639Sfsmpstatic int vfs_mountroot_try(const char *mountfrom); 8592723Salfredstatic int vfs_donmount(struct thread *td, int fsflags, 86177253Srwatson struct uio *fsoptions); 8710358Sjulianstatic void free_mntarg(struct mntarg *ma); 88169803Sjeffstatic void vfs_mount_destroy(struct mount *); 89170468Sattiliostatic int vfs_getopt_pos(struct vfsoptlist *opts, const char *name); 90169803Sjeff 91147692Speterstatic int usermount = 0; 92147692SpeterSYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 93147692Speter "Unprivileged users may mount and unmount file systems"); 94147692Speter 95174070SpeterMALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 96147703SpsMALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 97147692Speterstatic uma_zone_t mount_zone; 98147692Speter 99174070Speter/* List of mounted filesystems. */ 100157822Sjhbstruct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 101174070Speter 102174070Speter/* For any iteration/modification of mountlist */ 103147703Spsstruct mtx mountlist_mtx; 104147692SpeterMTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 105147692Speter 106147692SpeterTAILQ_HEAD(vfsoptlist, vfsopt); 107147692Speterstruct vfsopt { 108147692Speter TAILQ_ENTRY(vfsopt) link; 109147692Speter char *name; 110147692Speter void *value; 111147692Speter int len; 112147692Speter}; 113147692Speter 114147692Speter/* 115147692Speter * The vnode of the system's root (/ in the filesystem, without chroot 116147692Speter * active.) 117147692Speter */ 118147692Speterstruct vnode *rootvnode; 119192304Sed 120147692Speter/* 121147692Speter * The root filesystem is detailed in the kernel environment variable 122174070Speter * vfs.root.mountfrom, which is expected to be in the general format 123174070Speter * 124174070Speter * <vfsname>:[<path>] 125174070Speter * vfsname := the name of a VFS known to the kernel and capable 126174070Speter * of being mounted as root 127174070Speter * path := disk device name or other data used by the filesystem 128174070Speter * to locate its physical store 129174072Srwatson */ 130174070Speter 131174070Speter/* 132174070Speter * Global opts, taken by all filesystems 133174072Srwatson */ 134174070Speterstatic const char *global_opts[] = { 135174070Speter "errmsg", 136174070Speter "fstype", 137174070Speter "fspath", 138174070Speter "rdonly", 139174070Speter "ro", 140174070Speter "rw", 141174070Speter "suid", 142174070Speter "exec", 143174070Speter "update", 144174070Speter NULL 145174070Speter}; 146174070Speter 147174070Speter/* 148174070Speter * The root specifiers we will try if RB_CDROM is specified. 149174070Speter */ 150174070Speterstatic char *cdrom_rootdevnames[] = { 151174070Speter "cd9660:cd0", 152174070Speter "cd9660:acd0", 153174070Speter NULL 154174070Speter}; 155174070Speter 156174070Speter/* legacy find-root code */ 157174070Speterchar *rootdevnames[2] = {NULL, NULL}; 158174070Speter#ifndef ROOTDEVNAME 159174070Speter# define ROOTDEVNAME NULL 160174070Speter#endif 161174070Speterstatic const char *ctrootdevname = ROOTDEVNAME; 162174070Speter 163192304Sed/* 164174070Speter * --------------------------------------------------------------------- 165174070Speter * Functions for building and sanitizing the mount options 166201879Sattilio */ 167206482Sattilio 168206879Sattilio/* Remove one mount option. */ 169206482Sattiliostatic void 170206482Sattiliovfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 171206482Sattilio{ 172206482Sattilio 173201879Sattilio TAILQ_REMOVE(opts, opt, link); 174201879Sattilio free(opt->name, M_MOUNT); 175201879Sattilio if (opt->value != NULL) 176201879Sattilio free(opt->value, M_MOUNT); 177201879Sattilio#ifdef INVARIANTS 178201879Sattilio else if (opt->len != 0) 179201879Sattilio panic("%s: mount option with NULL value but length != 0", 180201879Sattilio __func__); 181201879Sattilio#endif 182201879Sattilio free(opt, M_MOUNT); 183206482Sattilio} 184201879Sattilio 185201879Sattilio/* Release all resources related to the mount options. */ 186201879Sattiliostatic void 187201879Sattiliovfs_freeopts(struct vfsoptlist *opts) 188201879Sattilio{ 189201879Sattilio struct vfsopt *opt; 190201879Sattilio 191201879Sattilio while (!TAILQ_EMPTY(opts)) { 192201879Sattilio opt = TAILQ_FIRST(opts); 193201879Sattilio vfs_freeopt(opts, opt); 194201879Sattilio } 195201879Sattilio free(opts, M_MOUNT); 196201879Sattilio} 197201879Sattilio 198201879Sattilio/* 199201879Sattilio * Check if options are equal (with or without the "no" prefix). 200214682Sjhb */ 201201879Sattiliostatic int 202201879Sattiliovfs_equalopts(const char *opt1, const char *opt2) 203201879Sattilio{ 204201879Sattilio 205201879Sattilio /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 206220390Sjhb if (strcmp(opt1, opt2) == 0) 207220390Sjhb return (1); 208220390Sjhb /* "noopt" vs. "opt" */ 209220390Sjhb if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 210201879Sattilio return (1); 211209761Sattilio /* "opt" vs. "noopt" */ 212209761Sattilio if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 213209761Sattilio return (1); 214209761Sattilio return (0); 215209761Sattilio} 216209761Sattilio 217201879Sattilio/* 218209761Sattilio * If a mount option is specified several times, 219201879Sattilio * (with or without the "no" prefix) only keep 220201879Sattilio * the last occurence of it. 221201879Sattilio */ 222201879Sattiliostatic void 223201879Sattiliovfs_sanitizeopts(struct vfsoptlist *opts) 224201879Sattilio{ 225201879Sattilio struct vfsopt *opt, *opt2, *tmp; 226206482Sattilio 227201879Sattilio TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 228201879Sattilio opt2 = TAILQ_PREV(opt, vfsoptlist, link); 229201879Sattilio while (opt2 != NULL) { 230201879Sattilio if (vfs_equalopts(opt->name, opt2->name)) { 231201879Sattilio tmp = TAILQ_PREV(opt2, vfsoptlist, link); 232201879Sattilio vfs_freeopt(opts, opt2); 233201879Sattilio opt2 = tmp; 234201879Sattilio } else { 235201879Sattilio opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 236201879Sattilio } 237201879Sattilio } 238201879Sattilio } 239201879Sattilio} 240201879Sattilio 241201879Sattilio/* 242209761Sattilio * Build a linked list of mount options from a struct uio. 243209761Sattilio */ 244209761Sattiliostatic int 245201879Sattiliovfs_buildopts(struct uio *auio, struct vfsoptlist **options) 246201879Sattilio{ 247201879Sattilio struct vfsoptlist *opts; 248201879Sattilio struct vfsopt *opt; 249201879Sattilio size_t memused; 250201879Sattilio unsigned int i, iovcnt; 251201879Sattilio int error, namelen, optlen; 252201879Sattilio 253201879Sattilio opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 254201879Sattilio TAILQ_INIT(opts); 255201879Sattilio memused = 0; 256201879Sattilio iovcnt = auio->uio_iovcnt; 257201879Sattilio for (i = 0; i < iovcnt; i += 2) { 258201879Sattilio opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 259201879Sattilio namelen = auio->uio_iov[i].iov_len; 260201879Sattilio optlen = auio->uio_iov[i + 1].iov_len; 261201879Sattilio opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 262201879Sattilio opt->value = NULL; 263201879Sattilio opt->len = 0; 264201879Sattilio 265201879Sattilio /* 266206482Sattilio * Do this early, so jumps to "bad" will free the current 267206482Sattilio * option. 268206482Sattilio */ 269206482Sattilio TAILQ_INSERT_TAIL(opts, opt, link); 270201879Sattilio memused += sizeof(struct vfsopt) + optlen + namelen; 271206482Sattilio 272206482Sattilio /* 273206482Sattilio * Avoid consuming too much memory, and attempts to overflow 274206482Sattilio * memused. 275206482Sattilio */ 276206482Sattilio if (memused > VFS_MOUNTARG_SIZE_MAX || 277206482Sattilio optlen > VFS_MOUNTARG_SIZE_MAX || 278206482Sattilio namelen > VFS_MOUNTARG_SIZE_MAX) { 279206482Sattilio error = EINVAL; 280206482Sattilio goto bad; 281206482Sattilio } 282206482Sattilio 283206482Sattilio if (auio->uio_segflg == UIO_SYSSPACE) { 284201879Sattilio bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 285201879Sattilio } else { 286201879Sattilio error = copyin(auio->uio_iov[i].iov_base, opt->name, 287201879Sattilio namelen); 288201879Sattilio if (error) 289201879Sattilio goto bad; 290201879Sattilio } 291201879Sattilio /* Ensure names are null-terminated strings. */ 292201879Sattilio if (opt->name[namelen - 1] != '\0') { 293201879Sattilio error = EINVAL; 294201879Sattilio goto bad; 295201879Sattilio } 296201879Sattilio if (optlen != 0) { 297214682Sjhb opt->len = optlen; 298201879Sattilio opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 299201879Sattilio if (auio->uio_segflg == UIO_SYSSPACE) { 300201879Sattilio bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 301201879Sattilio optlen); 302201879Sattilio } else { 303201879Sattilio error = copyin(auio->uio_iov[i + 1].iov_base, 304201879Sattilio opt->value, optlen); 305201879Sattilio if (error) 306201879Sattilio goto bad; 307201879Sattilio } 308201879Sattilio } 309227309Sed } 310227309Sed vfs_sanitizeopts(opts); 311201879Sattilio *options = opts; 312201879Sattilio return (0); 313201879Sattiliobad: 314201879Sattilio vfs_freeopts(opts); 315201879Sattilio return (error); 316201879Sattilio} 317201879Sattilio 318201879Sattilio/* 319201879Sattilio * Merge the old mount options with the new ones passed 320201879Sattilio * in the MNT_UPDATE case. 321174070Speter */ 322174070Speterstatic void 323174070Spetervfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 324174070Speter{ 325174070Speter struct vfsopt *opt, *opt2, *new; 326174070Speter 327174070Speter TAILQ_FOREACH(opt, opts, link) { 328174070Speter /* 329209059Sjhb * Check that this option hasn't been redefined 330174070Speter * nor cancelled with a "no" mount option. 331174070Speter */ 332174070Speter opt2 = TAILQ_FIRST(toopts); 333174070Speter while (opt2 != NULL) { 334174070Speter if (strcmp(opt2->name, opt->name) == 0) 335174070Speter goto next; 336126383Sphk if (strncmp(opt2->name, "no", 2) == 0 && 337126383Sphk strcmp(opt2->name + 2, opt->name) == 0) { 338116874Ssmkelly vfs_freeopt(toopts, opt2); 339126383Sphk goto next; 340116874Ssmkelly } 341126383Sphk opt2 = TAILQ_NEXT(opt2, link); 342126383Sphk } 343126383Sphk /* We want this option, duplicate it. */ 344116874Ssmkelly new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 3451541Srgrimes new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 3461541Srgrimes strcpy(new->name, opt->name); 3471541Srgrimes if (opt->len != 0) { 34834618Sphk new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 34934618Sphk bcopy(opt->value, new->value, opt->len); 35033690Sphk } else { 35134618Sphk new->value = NULL; 35234618Sphk } 35333690Sphk new->len = opt->len; 35434618Sphk TAILQ_INSERT_TAIL(toopts, new, link); 35534618Sphknext: 35634618Sphk continue; 35734618Sphk } 3581541Srgrimes} 3591541Srgrimes 36034618Sphk/* 36134618Sphk * --------------------------------------------------------------------- 36234618Sphk * Mount a filesystem 36334618Sphk */ 36434618Sphkint 3651541Srgrimesnmount(td, uap) 36634618Sphk struct thread *td; 36734618Sphk struct nmount_args /* { 3681541Srgrimes struct iovec *iovp; 36934618Sphk unsigned int iovcnt; 37034618Sphk int flags; 37134618Sphk } */ *uap; 37233690Sphk{ 37333690Sphk struct uio *auio; 37433690Sphk struct iovec *iov; 3751541Srgrimes unsigned int i; 3761541Srgrimes int error; 3771541Srgrimes u_int iovcnt; 3781541Srgrimes 379110296Sjake AUDIT_ARG(fflags, uap->flags); 3801541Srgrimes 381110296Sjake /* Kick out MNT_ROOTFS early as it is legal internally */ 3821541Srgrimes if (uap->flags & MNT_ROOTFS) 383215701Sdim return (EINVAL); 384212601Smav 385208494Smav iovcnt = uap->iovcnt; 3861541Srgrimes /* 3871541Srgrimes * Check that we have an even number of iovec's 3881541Srgrimes * and that we have at least two options. 38910358Sjulian */ 39010358Sjulian if ((iovcnt & 1) || (iovcnt < 4)) 39112569Sbde return (EINVAL); 39212569Sbde 3931541Srgrimes error = copyinuio(uap->iovp, iovcnt, &auio); 3941541Srgrimes if (error) 3951541Srgrimes return (error); 3961541Srgrimes iov = auio->uio_iov; 3971541Srgrimes for (i = 0; i < iovcnt; i++) { 3981541Srgrimes if (iov->iov_len > MMAXOPTIONLEN) { 3991541Srgrimes free(auio, M_IOV); 400209371Smav return (EINVAL); 4011541Srgrimes } 4021541Srgrimes iov++; 4031541Srgrimes } 4041541Srgrimes error = vfs_donmount(td, uap->flags, auio); 4051541Srgrimes 4061541Srgrimes free(auio, M_IOV); 4071541Srgrimes return (error); 4081541Srgrimes} 4091541Srgrimes 410126383Sphk/* 411126383Sphk * --------------------------------------------------------------------- 412126383Sphk * Various utility functions 4131541Srgrimes */ 4141541Srgrimes 4151541Srgrimesvoid 416110296Sjakevfs_ref(struct mount *mp) 417153666Sjhb{ 418110296Sjake 41976078Sjhb MNT_ILOCK(mp); 42076078Sjhb MNT_REF(mp); 421153666Sjhb MNT_IUNLOCK(mp); 42276078Sjhb} 42376078Sjhb 424110296Sjakevoid 42583366Sjulianvfs_rel(struct mount *mp) 426172207Sjeff{ 42776078Sjhb 42876078Sjhb MNT_ILOCK(mp); 42976078Sjhb MNT_REL(mp); 43076078Sjhb MNT_IUNLOCK(mp); 431163709Sjb} 432172207Sjeff 433163709Sjbstatic int 434170297Sjeffmount_init(void *mem, int size, int flags) 435170297Sjeff{ 436172207Sjeff struct mount *mp; 437172207Sjeff 438170297Sjeff mp = (struct mount *)mem; 439163709Sjb mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 440170297Sjeff lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 441170297Sjeff return (0); 442172207Sjeff} 443172207Sjeff 444170297Sjeffstatic void 445170297Sjeffmount_fini(void *mem, int size) 446170297Sjeff{ 447212541Smav struct mount *mp; 448172207Sjeff 449170297Sjeff mp = (struct mount *)mem; 450146799Sjkoshy lockdestroy(&mp->mnt_lock); 451233628Sfabient mtx_destroy(&mp->mnt_mtx); 452146799Sjkoshy} 453146799Sjkoshy 454233628Sfabient/* 455233628Sfabient * Allocate and initialize the mount point struct. 456146799Sjkoshy */ 457177859Sjeffstatic struct mount * 45876078Sjhbvfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, 45976078Sjhb const char *fspath, struct thread *td) 46076078Sjhb{ 4611541Srgrimes struct mount *mp; 4621541Srgrimes 4631541Srgrimes mp = uma_zalloc(mount_zone, M_WAITOK); 464153666Sjhb bzero(&mp->mnt_startzero, 4651541Srgrimes __rangeof(struct mount, mnt_startzero, mnt_endzero)); 4661541Srgrimes TAILQ_INIT(&mp->mnt_nvnodelist); 467177859Sjeff mp->mnt_nvnodelistsize = 0; 468153666Sjhb mp->mnt_ref = 0; 469212603Smav (void) vfs_busy(mp, LK_NOWAIT, 0, td); 470212541Smav mp->mnt_op = vfsp->vfc_vfsops; 4711541Srgrimes mp->mnt_vfc = vfsp; 4721541Srgrimes vfsp->vfc_refcount++; /* XXX Unlocked */ 47376078Sjhb mp->mnt_stat.f_type = vfsp->vfc_typenum; 47476078Sjhb mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 4751541Srgrimes strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 476110296Sjake mp->mnt_vnodecovered = vp; 477153666Sjhb mp->mnt_cred = crdup(td->td_ucred); 478153666Sjhb mp->mnt_stat.f_owner = td->td_ucred->cr_uid; 479110296Sjake strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 48087902Sluigi mp->mnt_iosize_max = DFLTPHYS; 48190550Sluigi#ifdef MAC 48287902Sluigi mac_init_mount(mp); 483126383Sphk mac_create_mount(td->td_ucred, mp); 484126383Sphk#endif 485116874Ssmkelly arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 486126383Sphk return (mp); 4871541Srgrimes} 4881541Srgrimes 489212541Smav/* 490232783Smav * Destroy the mount struct previously allocated by vfs_mount_alloc(). 491212541Smav */ 492212541Smavstatic void 493212541Smavvfs_mount_destroy(struct mount *mp) 494212541Smav{ 495212541Smav int i; 496212601Smav 497212601Smav MNT_ILOCK(mp); 498212601Smav for (i = 0; mp->mnt_ref && i < 3; i++) 499212601Smav msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz); 500212541Smav /* 501212541Smav * This will always cause a 3 second delay in rebooting due to 502212541Smav * refs on the root mountpoint that never go away. Most of these 503212541Smav * are held by init which never exits. 504212541Smav */ 505212541Smav if (i == 3 && (!rebooting || bootverbose)) 506212541Smav printf("Mount point %s had %d dangling refs\n", 507212541Smav mp->mnt_stat.f_mntonname, mp->mnt_ref); 508212541Smav if (mp->mnt_holdcnt != 0) { 509212541Smav printf("Waiting for mount point to be unheld\n"); 510212541Smav while (mp->mnt_holdcnt != 0) { 511212541Smav mp->mnt_holdcntwaiters++; 512212541Smav msleep(&mp->mnt_holdcnt, MNT_MTX(mp), 513212541Smav PZERO, "mntdestroy", 0); 514212541Smav mp->mnt_holdcntwaiters--; 515212541Smav } 516212541Smav printf("mount point unheld\n"); 517212541Smav } 518212541Smav if (mp->mnt_writeopcount > 0) { 519212541Smav printf("Waiting for mount point write ops\n"); 520212541Smav while (mp->mnt_writeopcount > 0) { 521212541Smav mp->mnt_kern_flag |= MNTK_SUSPEND; 522212541Smav msleep(&mp->mnt_writeopcount, 523212541Smav MNT_MTX(mp), 524212541Smav PZERO, "mntdestroy2", 0); 525212541Smav } 526212541Smav printf("mount point write ops completed\n"); 527212541Smav } 528212541Smav if (mp->mnt_secondary_writes > 0) { 529212541Smav printf("Waiting for mount point secondary write ops\n"); 530212541Smav while (mp->mnt_secondary_writes > 0) { 531212541Smav mp->mnt_kern_flag |= MNTK_SUSPEND; 532212541Smav msleep(&mp->mnt_secondary_writes, 533212541Smav MNT_MTX(mp), 534212541Smav PZERO, "mntdestroy3", 0); 535212541Smav } 536212541Smav printf("mount point secondary write ops completed\n"); 537212541Smav } 538212541Smav MNT_IUNLOCK(mp); 539212541Smav mp->mnt_vfc->vfc_refcount--; 540212541Smav if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 541212541Smav struct vnode *vp; 542212541Smav 543212541Smav TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 544233628Sfabient vprint("", vp); 545233628Sfabient panic("unmount: dangling vnode"); 546212541Smav } 547212541Smav MNT_ILOCK(mp); 548212541Smav if (mp->mnt_kern_flag & MNTK_MWAIT) 549212541Smav wakeup(mp); 550212601Smav if (mp->mnt_writeopcount != 0) 551212601Smav panic("vfs_mount_destroy: nonzero writeopcount"); 552212603Smav if (mp->mnt_secondary_writes != 0) 553212541Smav panic("vfs_mount_destroy: nonzero secondary_writes"); 554212601Smav if (mp->mnt_nvnodelistsize != 0) 555212601Smav panic("vfs_mount_destroy: nonzero nvnodelistsize"); 556212541Smav mp->mnt_writeopcount = -1000; 557212601Smav mp->mnt_nvnodelistsize = -1000; 558212601Smav mp->mnt_secondary_writes = -1000; 559212541Smav MNT_IUNLOCK(mp); 560212541Smav#ifdef MAC 561212601Smav mac_destroy_mount(mp); 562212601Smav#endif 563212541Smav if (mp->mnt_opt != NULL) 564212541Smav vfs_freeopts(mp->mnt_opt); 565212541Smav crfree(mp->mnt_cred); 566212541Smav uma_zfree(mount_zone, mp); 567212541Smav} 568212541Smav 569212541Smavstatic int 570212541Smavvfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 571212541Smav{ 572212541Smav struct vfsoptlist *optlist; 573212541Smav struct vfsopt *opt, *noro_opt; 574212541Smav char *fstype, *fspath, *errmsg; 575212541Smav int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 576212541Smav int has_rw, has_noro; 577212541Smav 578212541Smav errmsg_len = 0; 5791541Srgrimes errmsg_pos = -1; 58034961Sphk has_rw = 0; 5811541Srgrimes has_noro = 0; 5821541Srgrimes 58334961Sphk error = vfs_buildopts(fsoptions, &optlist); 5841541Srgrimes if (error) 5851541Srgrimes return (error); 5865081Sbde 5875081Sbde if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 5881541Srgrimes errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 5891541Srgrimes else 5905081Sbde errmsg_len = 0; 5915081Sbde 5925081Sbde /* 5935081Sbde * We need these two options before the others, 5945081Sbde * and they are mandatory for any filesystem. 5955081Sbde * Ensure they are NUL terminated as well. 5961541Srgrimes */ 5975081Sbde fstypelen = 0; 5985081Sbde error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 5995081Sbde if (error || fstype[fstypelen - 1] != '\0') { 6005081Sbde error = EINVAL; 6015081Sbde if (errmsg != NULL) 6025081Sbde strncpy(errmsg, "Invalid fstype", errmsg_len); 6035081Sbde goto bail; 6045081Sbde } 6055081Sbde fspathlen = 0; 6065081Sbde error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 6075081Sbde if (error || fspath[fspathlen - 1] != '\0') { 6081541Srgrimes error = EINVAL; 60934961Sphk if (errmsg != NULL) 61034961Sphk strncpy(errmsg, "Invalid fspath", errmsg_len); 6115081Sbde goto bail; 6125081Sbde } 6135081Sbde 6145081Sbde /* 6155081Sbde * We need to see if we have the "update" option 6165081Sbde * before we call vfs_domount(), since vfs_domount() has special 61734618Sphk * logic based on MNT_UPDATE. This is very important 61833690Sphk * when we want to update the root filesystem. 61933690Sphk */ 62033690Sphk TAILQ_FOREACH(opt, optlist, link) { 62134961Sphk if (strcmp(opt->name, "update") == 0) 6225081Sbde fsflags |= MNT_UPDATE; 6235081Sbde else if (strcmp(opt->name, "async") == 0) 6245081Sbde fsflags |= MNT_ASYNC; 6255081Sbde else if (strcmp(opt->name, "force") == 0) 6265081Sbde fsflags |= MNT_FORCE; 6275081Sbde else if (strcmp(opt->name, "multilabel") == 0) 6285081Sbde fsflags |= MNT_MULTILABEL; 6295081Sbde else if (strcmp(opt->name, "noasync") == 0) 6305081Sbde fsflags &= ~MNT_ASYNC; 6311541Srgrimes else if (strcmp(opt->name, "noatime") == 0) 6325081Sbde fsflags |= MNT_NOATIME; 6335081Sbde else if (strcmp(opt->name, "noclusterr") == 0) 6345081Sbde fsflags |= MNT_NOCLUSTERR; 63540012Salex else if (strcmp(opt->name, "noclusterw") == 0) 6361541Srgrimes fsflags |= MNT_NOCLUSTERW; 6371541Srgrimes else if (strcmp(opt->name, "noexec") == 0) 6381541Srgrimes fsflags |= MNT_NOEXEC; 6391541Srgrimes else if (strcmp(opt->name, "nosuid") == 0) 6401541Srgrimes fsflags |= MNT_NOSUID; 6411541Srgrimes else if (strcmp(opt->name, "nosymfollow") == 0) 6421541Srgrimes fsflags |= MNT_NOSYMFOLLOW; 6431541Srgrimes else if (strcmp(opt->name, "noro") == 0) { 6441541Srgrimes fsflags &= ~MNT_RDONLY; 6451541Srgrimes has_noro = 1; 6461541Srgrimes } 6471541Srgrimes else if (strcmp(opt->name, "rw") == 0) { 6481541Srgrimes fsflags &= ~MNT_RDONLY; 649113874Sjhb has_rw = 1; 650113874Sjhb } 651110530Sjulian else if (strcmp(opt->name, "ro") == 0 || 652113874Sjhb strcmp(opt->name, "rdonly") == 0) 653113874Sjhb fsflags |= MNT_RDONLY; 654209371Smav else if (strcmp(opt->name, "snapshot") == 0) 655110296Sjake fsflags |= MNT_SNAPSHOT; 656110296Sjake else if (strcmp(opt->name, "suiddir") == 0) 657209371Smav fsflags |= MNT_SUIDDIR; 6581541Srgrimes else if (strcmp(opt->name, "sync") == 0) 6591541Srgrimes fsflags |= MNT_SYNCHRONOUS; 6601541Srgrimes else if (strcmp(opt->name, "union") == 0) 6611541Srgrimes fsflags |= MNT_UNION; 6621541Srgrimes } 6631541Srgrimes 6641541Srgrimes /* 6651541Srgrimes * If "rw" was specified as a mount option, and we 6661541Srgrimes * are trying to update a mount-point from "ro" to "rw", 6671541Srgrimes * we need a mount option "noro", since in vfs_mergeopts(), 6681541Srgrimes * "noro" will cancel "ro", but "rw" will not do anything. 669110530Sjulian */ 670113874Sjhb if (has_rw && !has_noro) { 671113874Sjhb noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 672113874Sjhb noro_opt->name = strdup("noro", M_MOUNT); 673113874Sjhb noro_opt->value = NULL; 674113874Sjhb noro_opt->len = 0; 675123740Speter TAILQ_INSERT_TAIL(optlist, noro_opt, link); 676113874Sjhb } 677110530Sjulian 678128852Scperciva /* 679128852Scperciva * Be ultra-paranoid about making sure the type and fspath 680113874Sjhb * variables will fit in our mp buffers, including the 681209371Smav * terminating NUL. 682110296Sjake */ 683110296Sjake if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 684209371Smav error = ENAMETOOLONG; 6851541Srgrimes goto bail; 6861541Srgrimes } 6871541Srgrimes 6881541Srgrimes mtx_lock(&Giant); 689170174Sjeff error = vfs_domount(td, fstype, fspath, fsflags, optlist); 690170174Sjeff mtx_unlock(&Giant); 691170174Sjeffbail: 692110296Sjake /* copyout the errmsg */ 6931541Srgrimes if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 6941541Srgrimes && errmsg_len > 0 && errmsg != NULL) { 695153666Sjhb if (fsoptions->uio_segflg == UIO_SYSSPACE) { 6961541Srgrimes bcopy(errmsg, 697232783Smav fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 698232783Smav fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 699232783Smav } else { 700232783Smav copyout(errmsg, 701232783Smav fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 702232783Smav fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 703232783Smav } 70417342Sbde } 70517342Sbde 706110296Sjake if (error != 0) 707110296Sjake vfs_freeopts(optlist); 708110296Sjake return (error); 709174070Speter} 7101541Srgrimes 711110296Sjake/* 712110296Sjake * --------------------------------------------------------------------- 713110296Sjake * Old mount API. 714174070Speter */ 715153666Sjhb#ifndef _SYS_SYSPROTO_H_ 71653751Sbdestruct mount_args { 71753751Sbde char *type; 7181541Srgrimes char *path; 719232783Smav int flags; 720130551Sjulian caddr_t data; 721232783Smav}; 7221541Srgrimes#endif 723232783Smav/* ARGSUSED */ 7241541Srgrimesint 7251541Srgrimesmount(td, uap) 7261541Srgrimes struct thread *td; 7271541Srgrimes struct mount_args /* { 7281541Srgrimes char *type; 7291541Srgrimes char *path; 7301541Srgrimes int flags; 7311541Srgrimes caddr_t data; 7321541Srgrimes } */ *uap; 7331541Srgrimes{ 7341541Srgrimes char *fstype; 7351541Srgrimes struct vfsconf *vfsp = NULL; 7361541Srgrimes struct mntarg *ma = NULL; 737151658Sjhb int error; 738151658Sjhb 739232783Smav AUDIT_ARG(fflags, uap->flags); 740232783Smav 74165557Sjasone /* Kick out MNT_ROOTFS early as it is legal internally */ 742232783Smav uap->flags &= ~MNT_ROOTFS; 743232783Smav 744167327Sjulian fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 745232783Smav error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 74665557Sjasone if (error) { 747232783Smav free(fstype, M_TEMP); 74865557Sjasone return (error); 7491541Srgrimes } 7501541Srgrimes 75165782Sjhb AUDIT_ARG(text, fstype); 752131436Sjhb mtx_lock(&Giant); 753131436Sjhb vfsp = vfs_byname_kld(fstype, td, &error); 754170174Sjeff free(fstype, M_TEMP); 755232783Smav if (vfsp == NULL) { 756232783Smav mtx_unlock(&Giant); 757232783Smav return (ENOENT); 758131436Sjhb } 759131436Sjhb if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 760131436Sjhb mtx_unlock(&Giant); 761187357Sjeff return (EOPNOTSUPP); 762187357Sjeff } 763174070Speter 764232783Smav ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 765232783Smav ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 766170297Sjeff ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 767233628Sfabient ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 768233628Sfabient ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 769233628Sfabient 770233628Sfabient error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 77176078Sjhb mtx_unlock(&Giant); 77266716Sjhb return (error); 77376078Sjhb} 774153666Sjhb 77576078Sjhb 776232783Smav/* 777232783Smav * vfs_domount(): actually attempt a filesystem mount. 778232783Smav */ 779232783Smavstatic int 780232783Smavvfs_domount( 781232783Smav struct thread *td, /* Flags common to all filesystems. */ 782232783Smav const char *fstype, /* Filesystem type. */ 783110296Sjake char *fspath, /* Mount path. */ 784110296Sjake int fsflags, /* Flags common to all filesystems. */ 785110296Sjake void *fsdata /* Options local to the filesystem. */ 786153490Sjhb ) 787110296Sjake{ 78876078Sjhb struct vnode *vp; 789111032Sjulian struct mount *mp; 790153666Sjhb struct vfsconf *vfsp; 791110296Sjake struct export_args export; 792110296Sjake int error, flag = 0, kern_flag = 0; 793110296Sjake struct vattr va; 794110530Sjulian struct nameidata nd; 795110530Sjulian 796110296Sjake mtx_assert(&Giant, MA_OWNED); 797113874Sjhb /* 798232783Smav * Be ultra-paranoid about making sure the type and fspath 799110296Sjake * variables will fit in our mp buffers, including the 800110296Sjake * terminating NUL. 801110296Sjake */ 802110296Sjake if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 803110296Sjake return (ENAMETOOLONG); 804110296Sjake 805110296Sjake if (jailed(td->td_ucred)) 806153666Sjhb return (EPERM); 807153666Sjhb if (usermount == 0) { 808110296Sjake if ((error = suser(td)) != 0) 809232783Smav return (error); 810110296Sjake } 811110296Sjake 812110296Sjake /* 813110296Sjake * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 8141541Srgrimes */ 8151541Srgrimes if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) { 8161541Srgrimes if ((error = suser(td)) != 0) 8171541Srgrimes return (error); 8181541Srgrimes } 81912152Sphk /* 82062573Sphk * Silently enforce MNT_NOSUID and MNT_USER for 8211541Srgrimes * unprivileged users. 8221541Srgrimes */ 8231541Srgrimes if (suser(td) != 0) 8241541Srgrimes fsflags |= MNT_NOSUID | MNT_USER; 8251541Srgrimes 82696052Sbde /* Load KLDs before we lock the covered vnode to avoid reversals. */ 8271541Srgrimes vfsp = NULL; 8281541Srgrimes if ((fsflags & MNT_UPDATE) == 0) { 8291541Srgrimes /* Don't try to load KLDs if we're mounting the root. */ 8301541Srgrimes if (fsflags & MNT_ROOTFS) 83112243Sphk vfsp = vfs_byname(fstype); 8321541Srgrimes else 8332858Swollman vfsp = vfs_byname_kld(fstype, td, &error); 834192304Sed if (vfsp == NULL) 835192304Sed return (ENODEV); 83688019Sluigi } 83788019Sluigi /* 838116874Ssmkelly * Get vnode to be covered 839126383Sphk */ 840126383Sphk NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 841126383Sphk fspath, td); 842165260Sn_hibma if ((error = namei(&nd)) != 0) 843116874Ssmkelly return (error); 844126383Sphk NDFREE(&nd, NDF_ONLY_PNBUF); 845116874Ssmkelly vp = nd.ni_vp; 846126386Sphk if (fsflags & MNT_UPDATE) { 847165260Sn_hibma if ((vp->v_vflag & VV_ROOT) == 0) { 848126383Sphk vput(vp); 849126383Sphk return (EINVAL); 850165260Sn_hibma } 851126383Sphk mp = vp->v_mount; 852126383Sphk flag = mp->mnt_flag; 853126383Sphk kern_flag = mp->mnt_kern_flag; 854116874Ssmkelly /* 855116874Ssmkelly * We only allow the filesystem to be reloaded if it 856116874Ssmkelly * is currently mounted read-only. 857116874Ssmkelly */ 858170075Semaste if ((fsflags & MNT_RELOAD) && 859116874Ssmkelly ((mp->mnt_flag & MNT_RDONLY) == 0)) { 860116874Ssmkelly vput(vp); 861116874Ssmkelly return (EOPNOTSUPP); /* Needs translation */ 862116874Ssmkelly } 863116874Ssmkelly /* 864209390Sed * Only privileged root, or (if MNT_USER is set) the user that 865116874Ssmkelly * did the original mount is permitted to update it. 866116874Ssmkelly */ 867116874Ssmkelly error = vfs_suser(mp, td); 868116874Ssmkelly if (error) { 869116874Ssmkelly vput(vp); 870116874Ssmkelly return (error); 871225788Smav } 872157822Sjhb if (vfs_busy(mp, LK_NOWAIT, 0, td)) { 873116874Ssmkelly vput(vp); 874116874Ssmkelly return (EBUSY); 875116874Ssmkelly } 876116874Ssmkelly VI_LOCK(vp); 877116874Ssmkelly if ((vp->v_iflag & VI_MOUNT) != 0 || 878116874Ssmkelly vp->v_mountedhere != NULL) { 879116874Ssmkelly VI_UNLOCK(vp); 880116908Ssmkelly vfs_unbusy(mp, td); 881170075Semaste vput(vp); 882170075Semaste return (EBUSY); 883170075Semaste } 884174898Srwatson vp->v_iflag |= VI_MOUNT; 885170075Semaste VI_UNLOCK(vp); 886116874Ssmkelly mp->mnt_flag |= fsflags & 887170075Semaste (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 888116874Ssmkelly VOP_UNLOCK(vp, 0, td); 889116874Ssmkelly mp->mnt_optnew = fsdata; 890126383Sphk vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 891 } else { 892 /* 893 * If the user is not root, ensure that they own the directory 894 * onto which we are attempting to mount. 895 */ 896 error = VOP_GETATTR(vp, &va, td->td_ucred, td); 897 if (error) { 898 vput(vp); 899 return (error); 900 } 901 if (va.va_uid != td->td_ucred->cr_uid) { 902 if ((error = suser(td)) != 0) { 903 vput(vp); 904 return (error); 905 } 906 } 907 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 908 if (error != 0) { 909 vput(vp); 910 return (error); 911 } 912 if (vp->v_type != VDIR) { 913 vput(vp); 914 return (ENOTDIR); 915 } 916 VI_LOCK(vp); 917 if ((vp->v_iflag & VI_MOUNT) != 0 || 918 vp->v_mountedhere != NULL) { 919 VI_UNLOCK(vp); 920 vput(vp); 921 return (EBUSY); 922 } 923 vp->v_iflag |= VI_MOUNT; 924 VI_UNLOCK(vp); 925 926 /* 927 * Allocate and initialize the filesystem. 928 */ 929 mp = vfs_mount_alloc(vp, vfsp, fspath, td); 930 VOP_UNLOCK(vp, 0, td); 931 932 /* XXXMAC: pass to vfs_mount_alloc? */ 933 mp->mnt_optnew = fsdata; 934 } 935 936 /* 937 * Set the mount level flags. 938 */ 939 if (fsflags & MNT_RDONLY) 940 mp->mnt_flag |= MNT_RDONLY; 941 mp->mnt_flag &=~ MNT_UPDATEMASK; 942 mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS); 943 /* 944 * Mount the filesystem. 945 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 946 * get. No freeing of cn_pnbuf. 947 */ 948 error = VFS_MOUNT(mp, td); 949 950 /* 951 * Process the export option only if we are 952 * updating mount options. 953 */ 954 if (!error && (fsflags & MNT_UPDATE)) { 955 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 956 sizeof(export)) == 0) 957 error = vfs_export(mp, &export); 958 } 959 960 if (!error) { 961 if (mp->mnt_opt != NULL) 962 vfs_freeopts(mp->mnt_opt); 963 mp->mnt_opt = mp->mnt_optnew; 964 (void)VFS_STATFS(mp, &mp->mnt_stat, td); 965 } 966 /* 967 * Prevent external consumers of mount options from reading 968 * mnt_optnew. 969 */ 970 mp->mnt_optnew = NULL; 971 if (mp->mnt_flag & MNT_UPDATE) { 972 mp->mnt_flag &= 973 ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT); 974 if (error) { 975 mp->mnt_flag = flag; 976 mp->mnt_kern_flag = kern_flag; 977 } 978 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 979 if (mp->mnt_syncer == NULL) 980 error = vfs_allocate_syncvnode(mp); 981 } else { 982 if (mp->mnt_syncer != NULL) 983 vrele(mp->mnt_syncer); 984 mp->mnt_syncer = NULL; 985 } 986 vfs_unbusy(mp, td); 987 VI_LOCK(vp); 988 vp->v_iflag &= ~VI_MOUNT; 989 VI_UNLOCK(vp); 990 vrele(vp); 991 return (error); 992 } 993 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 994 /* 995 * Put the new filesystem on the mount list after root. 996 */ 997 cache_purge(vp); 998 if (!error) { 999 struct vnode *newdp; 1000 1001 VI_LOCK(vp); 1002 vp->v_iflag &= ~VI_MOUNT; 1003 VI_UNLOCK(vp); 1004 vp->v_mountedhere = mp; 1005 mtx_lock(&mountlist_mtx); 1006 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1007 mtx_unlock(&mountlist_mtx); 1008 vfs_event_signal(NULL, VQ_MOUNT, 0); 1009 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 1010 panic("mount: lost mount"); 1011 mountcheckdirs(vp, newdp); 1012 vput(newdp); 1013 VOP_UNLOCK(vp, 0, td); 1014 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1015 error = vfs_allocate_syncvnode(mp); 1016 vfs_unbusy(mp, td); 1017 if (error) 1018 vrele(vp); 1019 } else { 1020 VI_LOCK(vp); 1021 vp->v_iflag &= ~VI_MOUNT; 1022 VI_UNLOCK(vp); 1023 vfs_unbusy(mp, td); 1024 vfs_mount_destroy(mp); 1025 vput(vp); 1026 } 1027 return (error); 1028} 1029 1030/* 1031 * --------------------------------------------------------------------- 1032 * Unmount a filesystem. 1033 * 1034 * Note: unmount takes a path to the vnode mounted on as argument, 1035 * not special file (as before). 1036 */ 1037#ifndef _SYS_SYSPROTO_H_ 1038struct unmount_args { 1039 char *path; 1040 int flags; 1041}; 1042#endif 1043/* ARGSUSED */ 1044int 1045unmount(td, uap) 1046 struct thread *td; 1047 register struct unmount_args /* { 1048 char *path; 1049 int flags; 1050 } */ *uap; 1051{ 1052 struct mount *mp; 1053 char *pathbuf; 1054 int error, id0, id1; 1055 1056 if (jailed(td->td_ucred)) 1057 return (EPERM); 1058 if (usermount == 0) { 1059 if ((error = suser(td)) != 0) 1060 return (error); 1061 } 1062 1063 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1064 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1065 if (error) { 1066 free(pathbuf, M_TEMP); 1067 return (error); 1068 } 1069 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1070 mtx_lock(&Giant); 1071 if (uap->flags & MNT_BYFSID) { 1072 /* Decode the filesystem ID. */ 1073 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1074 mtx_unlock(&Giant); 1075 free(pathbuf, M_TEMP); 1076 return (EINVAL); 1077 } 1078 1079 mtx_lock(&mountlist_mtx); 1080 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1081 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1082 mp->mnt_stat.f_fsid.val[1] == id1) 1083 break; 1084 } 1085 mtx_unlock(&mountlist_mtx); 1086 } else { 1087 mtx_lock(&mountlist_mtx); 1088 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1089 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1090 break; 1091 } 1092 mtx_unlock(&mountlist_mtx); 1093 } 1094 free(pathbuf, M_TEMP); 1095 if (mp == NULL) { 1096 /* 1097 * Previously we returned ENOENT for a nonexistent path and 1098 * EINVAL for a non-mountpoint. We cannot tell these apart 1099 * now, so in the !MNT_BYFSID case return the more likely 1100 * EINVAL for compatibility. 1101 */ 1102 mtx_unlock(&Giant); 1103 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1104 } 1105 1106 /* 1107 * Only privileged root, or (if MNT_USER is set) the user that did the 1108 * original mount is permitted to unmount this filesystem. 1109 */ 1110 error = vfs_suser(mp, td); 1111 if (error) { 1112 mtx_unlock(&Giant); 1113 return (error); 1114 } 1115 1116 /* 1117 * Don't allow unmounting the root filesystem. 1118 */ 1119 if (mp->mnt_flag & MNT_ROOTFS) { 1120 mtx_unlock(&Giant); 1121 return (EINVAL); 1122 } 1123 error = dounmount(mp, uap->flags, td); 1124 mtx_unlock(&Giant); 1125 return (error); 1126} 1127 1128/* 1129 * Do the actual filesystem unmount. 1130 */ 1131int 1132dounmount(mp, flags, td) 1133 struct mount *mp; 1134 int flags; 1135 struct thread *td; 1136{ 1137 struct vnode *coveredvp, *fsrootvp; 1138 int error; 1139 int async_flag; 1140 1141 mtx_assert(&Giant, MA_OWNED); 1142 1143 if ((coveredvp = mp->mnt_vnodecovered) != NULL) 1144 vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY, td); 1145 MNT_ILOCK(mp); 1146 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1147 MNT_IUNLOCK(mp); 1148 if (coveredvp) 1149 VOP_UNLOCK(coveredvp, 0, td); 1150 return (EBUSY); 1151 } 1152 mp->mnt_kern_flag |= MNTK_UNMOUNT; 1153 /* Allow filesystems to detect that a forced unmount is in progress. */ 1154 if (flags & MNT_FORCE) 1155 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1156 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 1157 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td); 1158 if (error) { 1159 MNT_ILOCK(mp); 1160 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1161 if (mp->mnt_kern_flag & MNTK_MWAIT) 1162 wakeup(mp); 1163 MNT_IUNLOCK(mp); 1164 if (coveredvp) 1165 VOP_UNLOCK(coveredvp, 0, td); 1166 return (error); 1167 } 1168 vn_start_write(NULL, &mp, V_WAIT); 1169 1170 if (mp->mnt_flag & MNT_EXPUBLIC) 1171 vfs_setpublicfs(NULL, NULL, NULL); 1172 1173 vfs_msync(mp, MNT_WAIT); 1174 async_flag = mp->mnt_flag & MNT_ASYNC; 1175 mp->mnt_flag &= ~MNT_ASYNC; 1176 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1177 if (mp->mnt_syncer != NULL) 1178 vrele(mp->mnt_syncer); 1179 /* 1180 * For forced unmounts, move process cdir/rdir refs on the fs root 1181 * vnode to the covered vnode. For non-forced unmounts we want 1182 * such references to cause an EBUSY error. 1183 */ 1184 if ((flags & MNT_FORCE) && 1185 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1186 if (mp->mnt_vnodecovered != NULL) 1187 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1188 if (fsrootvp == rootvnode) { 1189 vrele(rootvnode); 1190 rootvnode = NULL; 1191 } 1192 vput(fsrootvp); 1193 } 1194 if (((mp->mnt_flag & MNT_RDONLY) || 1195 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1196 (flags & MNT_FORCE)) { 1197 error = VFS_UNMOUNT(mp, flags, td); 1198 } 1199 vn_finished_write(mp); 1200 if (error) { 1201 /* Undo cdir/rdir and rootvnode changes made above. */ 1202 if ((flags & MNT_FORCE) && 1203 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1204 if (mp->mnt_vnodecovered != NULL) 1205 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1206 if (rootvnode == NULL) { 1207 rootvnode = fsrootvp; 1208 vref(rootvnode); 1209 } 1210 vput(fsrootvp); 1211 } 1212 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) 1213 (void) vfs_allocate_syncvnode(mp); 1214 MNT_ILOCK(mp); 1215 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1216 mp->mnt_flag |= async_flag; 1217 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1218 if (mp->mnt_kern_flag & MNTK_MWAIT) 1219 wakeup(mp); 1220 MNT_IUNLOCK(mp); 1221 if (coveredvp) 1222 VOP_UNLOCK(coveredvp, 0, td); 1223 return (error); 1224 } 1225 mtx_lock(&mountlist_mtx); 1226 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1227 mtx_unlock(&mountlist_mtx); 1228 if (coveredvp != NULL) { 1229 coveredvp->v_mountedhere = NULL; 1230 vput(coveredvp); 1231 } 1232 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1233 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1234 vfs_mount_destroy(mp); 1235 return (0); 1236} 1237 1238/* 1239 * --------------------------------------------------------------------- 1240 * Mounting of root filesystem 1241 * 1242 */ 1243 1244struct root_hold_token { 1245 const char *who; 1246 LIST_ENTRY(root_hold_token) list; 1247}; 1248 1249static LIST_HEAD(, root_hold_token) root_holds = 1250 LIST_HEAD_INITIALIZER(&root_holds); 1251 1252struct root_hold_token * 1253root_mount_hold(const char *identifier) 1254{ 1255 struct root_hold_token *h; 1256 1257 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1258 h->who = identifier; 1259 mtx_lock(&mountlist_mtx); 1260 LIST_INSERT_HEAD(&root_holds, h, list); 1261 mtx_unlock(&mountlist_mtx); 1262 return (h); 1263} 1264 1265void 1266root_mount_rel(struct root_hold_token *h) 1267{ 1268 1269 mtx_lock(&mountlist_mtx); 1270 LIST_REMOVE(h, list); 1271 wakeup(&root_holds); 1272 mtx_unlock(&mountlist_mtx); 1273 free(h, M_DEVBUF); 1274} 1275 1276static void 1277root_mount_wait(void) 1278{ 1279 struct root_hold_token *h; 1280 1281 for (;;) { 1282 DROP_GIANT(); 1283 g_waitidle(); 1284 PICKUP_GIANT(); 1285 mtx_lock(&mountlist_mtx); 1286 if (LIST_EMPTY(&root_holds)) { 1287 mtx_unlock(&mountlist_mtx); 1288 break; 1289 } 1290 printf("Root mount waiting for:"); 1291 LIST_FOREACH(h, &root_holds, list) 1292 printf(" %s", h->who); 1293 printf("\n"); 1294 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1295 hz); 1296 } 1297} 1298 1299static void 1300set_rootvnode(struct thread *td) 1301{ 1302 struct proc *p; 1303 1304 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1305 panic("Cannot find root vnode"); 1306 1307 p = td->td_proc; 1308 FILEDESC_LOCK(p->p_fd); 1309 1310 if (p->p_fd->fd_cdir != NULL) 1311 vrele(p->p_fd->fd_cdir); 1312 p->p_fd->fd_cdir = rootvnode; 1313 VREF(rootvnode); 1314 1315 if (p->p_fd->fd_rdir != NULL) 1316 vrele(p->p_fd->fd_rdir); 1317 p->p_fd->fd_rdir = rootvnode; 1318 VREF(rootvnode); 1319 1320 FILEDESC_UNLOCK(p->p_fd); 1321 1322 VOP_UNLOCK(rootvnode, 0, td); 1323} 1324 1325/* 1326 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1327 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1328 */ 1329 1330static void 1331devfs_first(void) 1332{ 1333 struct thread *td = curthread; 1334 struct vfsoptlist *opts; 1335 struct vfsconf *vfsp; 1336 struct mount *mp = NULL; 1337 int error; 1338 1339 vfsp = vfs_byname("devfs"); 1340 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1341 if (vfsp == NULL) 1342 return; 1343 1344 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td); 1345 1346 error = VFS_MOUNT(mp, td); 1347 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1348 if (error) 1349 return; 1350 1351 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1352 TAILQ_INIT(opts); 1353 mp->mnt_opt = opts; 1354 1355 mtx_lock(&mountlist_mtx); 1356 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1357 mtx_unlock(&mountlist_mtx); 1358 1359 set_rootvnode(td); 1360 1361 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1362 if (error) 1363 printf("kern_symlink /dev -> / returns %d\n", error); 1364} 1365 1366/* 1367 * Surgically move our devfs to be mounted on /dev. 1368 */ 1369 1370static void 1371devfs_fixup(struct thread *td) 1372{ 1373 struct nameidata nd; 1374 int error; 1375 struct vnode *vp, *dvp; 1376 struct mount *mp; 1377 1378 /* Remove our devfs mount from the mountlist and purge the cache */ 1379 mtx_lock(&mountlist_mtx); 1380 mp = TAILQ_FIRST(&mountlist); 1381 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1382 mtx_unlock(&mountlist_mtx); 1383 cache_purgevfs(mp); 1384 1385 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1386 VI_LOCK(dvp); 1387 dvp->v_iflag &= ~VI_MOUNT; 1388 dvp->v_mountedhere = NULL; 1389 VI_UNLOCK(dvp); 1390 1391 /* Set up the real rootvnode, and purge the cache */ 1392 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1393 set_rootvnode(td); 1394 cache_purgevfs(rootvnode->v_mount); 1395 1396 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1397 error = namei(&nd); 1398 if (error) { 1399 printf("Lookup of /dev for devfs, error: %d\n", error); 1400 return; 1401 } 1402 NDFREE(&nd, NDF_ONLY_PNBUF); 1403 vp = nd.ni_vp; 1404 if (vp->v_type != VDIR) { 1405 vput(vp); 1406 } 1407 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 1408 if (error) { 1409 vput(vp); 1410 } 1411 cache_purge(vp); 1412 mp->mnt_vnodecovered = vp; 1413 vp->v_mountedhere = mp; 1414 mtx_lock(&mountlist_mtx); 1415 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1416 mtx_unlock(&mountlist_mtx); 1417 VOP_UNLOCK(vp, 0, td); 1418 vput(dvp); 1419 vfs_unbusy(mp, td); 1420 1421 /* Unlink the no longer needed /dev/dev -> / symlink */ 1422 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1423} 1424 1425/* 1426 * Report errors during filesystem mounting. 1427 */ 1428void 1429vfs_mount_error(struct mount *mp, const char *fmt, ...) 1430{ 1431 struct vfsoptlist *moptlist = mp->mnt_optnew; 1432 va_list ap; 1433 int error, len; 1434 char *errmsg; 1435 1436 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1437 if (error || errmsg == NULL || len <= 0) 1438 return; 1439 1440 va_start(ap, fmt); 1441 vsnprintf(errmsg, (size_t)len, fmt, ap); 1442 va_end(ap); 1443} 1444 1445/* 1446 * Find and mount the root filesystem 1447 */ 1448void 1449vfs_mountroot(void) 1450{ 1451 char *cp; 1452 int error, i, asked = 0; 1453 1454 root_mount_wait(); 1455 1456 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1457 NULL, NULL, mount_init, mount_fini, 1458 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1459 devfs_first(); 1460 1461 /* 1462 * We are booted with instructions to prompt for the root filesystem. 1463 */ 1464 if (boothowto & RB_ASKNAME) { 1465 if (!vfs_mountroot_ask()) 1466 return; 1467 asked = 1; 1468 } 1469 1470 /* 1471 * The root filesystem information is compiled in, and we are 1472 * booted with instructions to use it. 1473 */ 1474 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1475 if (!vfs_mountroot_try(ctrootdevname)) 1476 return; 1477 ctrootdevname = NULL; 1478 } 1479 1480 /* 1481 * We've been given the generic "use CDROM as root" flag. This is 1482 * necessary because one media may be used in many different 1483 * devices, so we need to search for them. 1484 */ 1485 if (boothowto & RB_CDROM) { 1486 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1487 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1488 return; 1489 } 1490 } 1491 1492 /* 1493 * Try to use the value read by the loader from /etc/fstab, or 1494 * supplied via some other means. This is the preferred 1495 * mechanism. 1496 */ 1497 cp = getenv("vfs.root.mountfrom"); 1498 if (cp != NULL) { 1499 error = vfs_mountroot_try(cp); 1500 freeenv(cp); 1501 if (!error) 1502 return; 1503 } 1504 1505 /* 1506 * Try values that may have been computed by code during boot 1507 */ 1508 if (!vfs_mountroot_try(rootdevnames[0])) 1509 return; 1510 if (!vfs_mountroot_try(rootdevnames[1])) 1511 return; 1512 1513 /* 1514 * If we (still) have a compiled-in default, try it. 1515 */ 1516 if (ctrootdevname != NULL) 1517 if (!vfs_mountroot_try(ctrootdevname)) 1518 return; 1519 /* 1520 * Everything so far has failed, prompt on the console if we haven't 1521 * already tried that. 1522 */ 1523 if (!asked) 1524 if (!vfs_mountroot_ask()) 1525 return; 1526 1527 panic("Root mount failed, startup aborted."); 1528} 1529 1530/* 1531 * Mount (mountfrom) as the root filesystem. 1532 */ 1533static int 1534vfs_mountroot_try(const char *mountfrom) 1535{ 1536 struct mount *mp; 1537 char *vfsname, *path; 1538 time_t timebase; 1539 int error; 1540 char patt[32]; 1541 1542 vfsname = NULL; 1543 path = NULL; 1544 mp = NULL; 1545 error = EINVAL; 1546 1547 if (mountfrom == NULL) 1548 return (error); /* don't complain */ 1549 printf("Trying to mount root from %s\n", mountfrom); 1550 1551 /* parse vfs name and path */ 1552 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1553 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1554 vfsname[0] = path[0] = 0; 1555 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1556 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1557 goto out; 1558 1559 if (path[0] == '\0') 1560 strcpy(path, ROOTNAME); 1561 1562 error = kernel_vmount( 1563 MNT_RDONLY | MNT_ROOTFS, 1564 "fstype", vfsname, 1565 "fspath", "/", 1566 "from", path, 1567 NULL); 1568 if (error == 0) { 1569 /* 1570 * We mount devfs prior to mounting the / FS, so the first 1571 * entry will typically be devfs. 1572 */ 1573 mp = TAILQ_FIRST(&mountlist); 1574 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1575 1576 /* 1577 * Iterate over all currently mounted file systems and use 1578 * the time stamp found to check and/or initialize the RTC. 1579 * Typically devfs has no time stamp and the only other FS 1580 * is the actual / FS. 1581 * Call inittodr() only once and pass it the largest of the 1582 * timestamps we encounter. 1583 */ 1584 timebase = 0; 1585 do { 1586 if (mp->mnt_time > timebase) 1587 timebase = mp->mnt_time; 1588 mp = TAILQ_NEXT(mp, mnt_list); 1589 } while (mp != NULL); 1590 inittodr(timebase); 1591 1592 devfs_fixup(curthread); 1593 } 1594out: 1595 free(path, M_MOUNT); 1596 free(vfsname, M_MOUNT); 1597 return (error); 1598} 1599 1600/* 1601 * --------------------------------------------------------------------- 1602 * Interactive root filesystem selection code. 1603 */ 1604 1605static int 1606vfs_mountroot_ask(void) 1607{ 1608 char name[128]; 1609 1610 for(;;) { 1611 printf("\nManual root filesystem specification:\n"); 1612 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1613#if defined(__amd64__) || defined(__i386__) || defined(__ia64__) 1614 printf(" eg. ufs:da0s1a\n"); 1615#else 1616 printf(" eg. ufs:/dev/da0a\n"); 1617#endif 1618 printf(" ? List valid disk boot devices\n"); 1619 printf(" <empty line> Abort manual input\n"); 1620 printf("\nmountroot> "); 1621 gets(name, sizeof(name), 1); 1622 if (name[0] == '\0') 1623 return (1); 1624 if (name[0] == '?') { 1625 printf("\nList of GEOM managed disk devices:\n "); 1626 g_dev_print(); 1627 continue; 1628 } 1629 if (!vfs_mountroot_try(name)) 1630 return (0); 1631 } 1632} 1633 1634/* 1635 * --------------------------------------------------------------------- 1636 * Functions for querying mount options/arguments from filesystems. 1637 */ 1638 1639/* 1640 * Check that no unknown options are given 1641 */ 1642int 1643vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1644{ 1645 struct vfsopt *opt; 1646 const char **t, *p; 1647 1648 1649 TAILQ_FOREACH(opt, opts, link) { 1650 p = opt->name; 1651 if (p[0] == 'n' && p[1] == 'o') 1652 p += 2; 1653 for(t = global_opts; *t != NULL; t++) 1654 if (!strcmp(*t, p)) 1655 break; 1656 if (*t != NULL) 1657 continue; 1658 for(t = legal; *t != NULL; t++) 1659 if (!strcmp(*t, p)) 1660 break; 1661 if (*t != NULL) 1662 continue; 1663 printf("mount option <%s> is unknown\n", p); 1664 return (EINVAL); 1665 } 1666 return (0); 1667} 1668 1669/* 1670 * Get a mount option by its name. 1671 * 1672 * Return 0 if the option was found, ENOENT otherwise. 1673 * If len is non-NULL it will be filled with the length 1674 * of the option. If buf is non-NULL, it will be filled 1675 * with the address of the option. 1676 */ 1677int 1678vfs_getopt(opts, name, buf, len) 1679 struct vfsoptlist *opts; 1680 const char *name; 1681 void **buf; 1682 int *len; 1683{ 1684 struct vfsopt *opt; 1685 1686 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1687 1688 TAILQ_FOREACH(opt, opts, link) { 1689 if (strcmp(name, opt->name) == 0) { 1690 if (len != NULL) 1691 *len = opt->len; 1692 if (buf != NULL) 1693 *buf = opt->value; 1694 return (0); 1695 } 1696 } 1697 return (ENOENT); 1698} 1699 1700static int 1701vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1702{ 1703 struct vfsopt *opt; 1704 int i; 1705 1706 if (opts == NULL) 1707 return (-1); 1708 1709 i = 0; 1710 TAILQ_FOREACH(opt, opts, link) { 1711 if (strcmp(name, opt->name) == 0) 1712 return (i); 1713 ++i; 1714 } 1715 return (-1); 1716} 1717 1718char * 1719vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1720{ 1721 struct vfsopt *opt; 1722 1723 *error = 0; 1724 TAILQ_FOREACH(opt, opts, link) { 1725 if (strcmp(name, opt->name) != 0) 1726 continue; 1727 if (((char *)opt->value)[opt->len - 1] != '\0') { 1728 *error = EINVAL; 1729 return (NULL); 1730 } 1731 return (opt->value); 1732 } 1733 return (NULL); 1734} 1735 1736int 1737vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1738{ 1739 struct vfsopt *opt; 1740 1741 TAILQ_FOREACH(opt, opts, link) { 1742 if (strcmp(name, opt->name) == 0) { 1743 if (w != NULL) 1744 *w |= val; 1745 return (1); 1746 } 1747 } 1748 if (w != NULL) 1749 *w &= ~val; 1750 return (0); 1751} 1752 1753int 1754vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1755{ 1756 va_list ap; 1757 struct vfsopt *opt; 1758 int ret; 1759 1760 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1761 1762 TAILQ_FOREACH(opt, opts, link) { 1763 if (strcmp(name, opt->name) != 0) 1764 continue; 1765 if (((char *)opt->value)[opt->len - 1] != '\0') 1766 return (0); 1767 va_start(ap, fmt); 1768 ret = vsscanf(opt->value, fmt, ap); 1769 va_end(ap); 1770 return (ret); 1771 } 1772 return (0); 1773} 1774 1775/* 1776 * Find and copy a mount option. 1777 * 1778 * The size of the buffer has to be specified 1779 * in len, if it is not the same length as the 1780 * mount option, EINVAL is returned. 1781 * Returns ENOENT if the option is not found. 1782 */ 1783int 1784vfs_copyopt(opts, name, dest, len) 1785 struct vfsoptlist *opts; 1786 const char *name; 1787 void *dest; 1788 int len; 1789{ 1790 struct vfsopt *opt; 1791 1792 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1793 1794 TAILQ_FOREACH(opt, opts, link) { 1795 if (strcmp(name, opt->name) == 0) { 1796 if (len != opt->len) 1797 return (EINVAL); 1798 bcopy(opt->value, dest, opt->len); 1799 return (0); 1800 } 1801 } 1802 return (ENOENT); 1803} 1804 1805/* 1806 * This is a helper function for filesystems to traverse their 1807 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1808 */ 1809 1810struct vnode * 1811__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 1812{ 1813 struct vnode *vp; 1814 1815 mtx_assert(MNT_MTX(mp), MA_OWNED); 1816 1817 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1818 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 1819 while (vp != NULL && vp->v_type == VMARKER) 1820 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1821 1822 /* Check if we are done */ 1823 if (vp == NULL) { 1824 __mnt_vnode_markerfree(mvp, mp); 1825 return (NULL); 1826 } 1827 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1828 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1829 return (vp); 1830} 1831 1832struct vnode * 1833__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 1834{ 1835 struct vnode *vp; 1836 1837 mtx_assert(MNT_MTX(mp), MA_OWNED); 1838 1839 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1840 while (vp != NULL && vp->v_type == VMARKER) 1841 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1842 1843 /* Check if we are done */ 1844 if (vp == NULL) { 1845 *mvp = NULL; 1846 return (NULL); 1847 } 1848 mp->mnt_holdcnt++; 1849 MNT_IUNLOCK(mp); 1850 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 1851 M_VNODE_MARKER, 1852 M_WAITOK | M_ZERO); 1853 MNT_ILOCK(mp); 1854 (*mvp)->v_type = VMARKER; 1855 1856 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1857 while (vp != NULL && vp->v_type == VMARKER) 1858 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1859 1860 /* Check if we are done */ 1861 if (vp == NULL) { 1862 MNT_IUNLOCK(mp); 1863 free(*mvp, M_VNODE_MARKER); 1864 MNT_ILOCK(mp); 1865 *mvp = NULL; 1866 mp->mnt_holdcnt--; 1867 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 1868 wakeup(&mp->mnt_holdcnt); 1869 return (NULL); 1870 } 1871 mp->mnt_markercnt++; 1872 (*mvp)->v_mount = mp; 1873 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1874 return (vp); 1875} 1876 1877 1878void 1879__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 1880{ 1881 1882 if (*mvp == NULL) 1883 return; 1884 1885 mtx_assert(MNT_MTX(mp), MA_OWNED); 1886 1887 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1888 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1889 MNT_IUNLOCK(mp); 1890 free(*mvp, M_VNODE_MARKER); 1891 MNT_ILOCK(mp); 1892 *mvp = NULL; 1893 1894 mp->mnt_markercnt--; 1895 mp->mnt_holdcnt--; 1896 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 1897 wakeup(&mp->mnt_holdcnt); 1898} 1899 1900 1901int 1902__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 1903{ 1904 int error; 1905 1906 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 1907 if (sbp != &mp->mnt_stat) 1908 *sbp = mp->mnt_stat; 1909 return (error); 1910} 1911 1912void 1913vfs_mountedfrom(struct mount *mp, const char *from) 1914{ 1915 1916 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 1917 strlcpy(mp->mnt_stat.f_mntfromname, from, 1918 sizeof mp->mnt_stat.f_mntfromname); 1919} 1920 1921/* 1922 * --------------------------------------------------------------------- 1923 * This is the api for building mount args and mounting filesystems from 1924 * inside the kernel. 1925 * 1926 * The API works by accumulation of individual args. First error is 1927 * latched. 1928 * 1929 * XXX: should be documented in new manpage kernel_mount(9) 1930 */ 1931 1932/* A memory allocation which must be freed when we are done */ 1933struct mntaarg { 1934 SLIST_ENTRY(mntaarg) next; 1935}; 1936 1937/* The header for the mount arguments */ 1938struct mntarg { 1939 struct iovec *v; 1940 int len; 1941 int error; 1942 SLIST_HEAD(, mntaarg) list; 1943}; 1944 1945/* 1946 * Add a boolean argument. 1947 * 1948 * flag is the boolean value. 1949 * name must start with "no". 1950 */ 1951struct mntarg * 1952mount_argb(struct mntarg *ma, int flag, const char *name) 1953{ 1954 1955 KASSERT(name[0] == 'n' && name[1] == 'o', 1956 ("mount_argb(...,%s): name must start with 'no'", name)); 1957 1958 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 1959} 1960 1961/* 1962 * Add an argument printf style 1963 */ 1964struct mntarg * 1965mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 1966{ 1967 va_list ap; 1968 struct mntaarg *maa; 1969 struct sbuf *sb; 1970 int len; 1971 1972 if (ma == NULL) { 1973 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1974 SLIST_INIT(&ma->list); 1975 } 1976 if (ma->error) 1977 return (ma); 1978 1979 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1980 M_MOUNT, M_WAITOK); 1981 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1982 ma->v[ma->len].iov_len = strlen(name) + 1; 1983 ma->len++; 1984 1985 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND); 1986 va_start(ap, fmt); 1987 sbuf_vprintf(sb, fmt, ap); 1988 va_end(ap); 1989 sbuf_finish(sb); 1990 len = sbuf_len(sb) + 1; 1991 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1992 SLIST_INSERT_HEAD(&ma->list, maa, next); 1993 bcopy(sbuf_data(sb), maa + 1, len); 1994 sbuf_delete(sb); 1995 1996 ma->v[ma->len].iov_base = maa + 1; 1997 ma->v[ma->len].iov_len = len; 1998 ma->len++; 1999 2000 return (ma); 2001} 2002 2003/* 2004 * Add an argument which is a userland string. 2005 */ 2006struct mntarg * 2007mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2008{ 2009 struct mntaarg *maa; 2010 char *tbuf; 2011 2012 if (val == NULL) 2013 return (ma); 2014 if (ma == NULL) { 2015 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2016 SLIST_INIT(&ma->list); 2017 } 2018 if (ma->error) 2019 return (ma); 2020 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2021 SLIST_INSERT_HEAD(&ma->list, maa, next); 2022 tbuf = (void *)(maa + 1); 2023 ma->error = copyinstr(val, tbuf, len, NULL); 2024 return (mount_arg(ma, name, tbuf, -1)); 2025} 2026 2027/* 2028 * Plain argument. 2029 * 2030 * If length is -1, use printf. 2031 */ 2032struct mntarg * 2033mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2034{ 2035 2036 if (ma == NULL) { 2037 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2038 SLIST_INIT(&ma->list); 2039 } 2040 if (ma->error) 2041 return (ma); 2042 2043 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2044 M_MOUNT, M_WAITOK); 2045 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2046 ma->v[ma->len].iov_len = strlen(name) + 1; 2047 ma->len++; 2048 2049 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2050 if (len < 0) 2051 ma->v[ma->len].iov_len = strlen(val) + 1; 2052 else 2053 ma->v[ma->len].iov_len = len; 2054 ma->len++; 2055 return (ma); 2056} 2057 2058/* 2059 * Free a mntarg structure 2060 */ 2061static void 2062free_mntarg(struct mntarg *ma) 2063{ 2064 struct mntaarg *maa; 2065 2066 while (!SLIST_EMPTY(&ma->list)) { 2067 maa = SLIST_FIRST(&ma->list); 2068 SLIST_REMOVE_HEAD(&ma->list, next); 2069 free(maa, M_MOUNT); 2070 } 2071 free(ma->v, M_MOUNT); 2072 free(ma, M_MOUNT); 2073} 2074 2075/* 2076 * Mount a filesystem 2077 */ 2078int 2079kernel_mount(struct mntarg *ma, int flags) 2080{ 2081 struct uio auio; 2082 int error; 2083 2084 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2085 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2086 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2087 2088 auio.uio_iov = ma->v; 2089 auio.uio_iovcnt = ma->len; 2090 auio.uio_segflg = UIO_SYSSPACE; 2091 2092 error = ma->error; 2093 if (!error) 2094 error = vfs_donmount(curthread, flags, &auio); 2095 free_mntarg(ma); 2096 return (error); 2097} 2098 2099/* 2100 * A printflike function to mount a filesystem. 2101 */ 2102int 2103kernel_vmount(int flags, ...) 2104{ 2105 struct mntarg *ma = NULL; 2106 va_list ap; 2107 const char *cp; 2108 const void *vp; 2109 int error; 2110 2111 va_start(ap, flags); 2112 for (;;) { 2113 cp = va_arg(ap, const char *); 2114 if (cp == NULL) 2115 break; 2116 vp = va_arg(ap, const void *); 2117 ma = mount_arg(ma, cp, vp, -1); 2118 } 2119 va_end(ap); 2120 2121 error = kernel_mount(ma, flags); 2122 return (error); 2123} 2124