vfs_mount.c revision 168185
1/*- 2 * Copyright (c) 1999-2004 Poul-Henning Kamp 3 * Copyright (c) 1999 Michael Smith 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: head/sys/kern/vfs_mount.c 168185 2007-03-31 22:44:45Z pjd $"); 39 40#include <sys/param.h> 41#include <sys/conf.h> 42#include <sys/clock.h> 43#include <sys/jail.h> 44#include <sys/kernel.h> 45#include <sys/libkern.h> 46#include <sys/malloc.h> 47#include <sys/mount.h> 48#include <sys/mutex.h> 49#include <sys/namei.h> 50#include <sys/priv.h> 51#include <sys/proc.h> 52#include <sys/filedesc.h> 53#include <sys/reboot.h> 54#include <sys/syscallsubr.h> 55#include <sys/sysproto.h> 56#include <sys/sx.h> 57#include <sys/sysctl.h> 58#include <sys/sysent.h> 59#include <sys/systm.h> 60#include <sys/vnode.h> 61#include <vm/uma.h> 62 63#include <geom/geom.h> 64 65#include <machine/stdarg.h> 66 67#include <security/audit/audit.h> 68#include <security/mac/mac_framework.h> 69 70#include "opt_rootdevname.h" 71#include "opt_ddb.h" 72#include "opt_mac.h" 73 74#ifdef DDB 75#include <ddb/ddb.h> 76#endif 77 78#define ROOTNAME "root_device" 79#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 80 81static int vfs_domount(struct thread *td, const char *fstype, 82 char *fspath, int fsflags, void *fsdata); 83static struct mount *vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp, 84 const char *fspath, struct thread *td); 85static int vfs_mountroot_ask(void); 86static int vfs_mountroot_try(const char *mountfrom); 87static int vfs_donmount(struct thread *td, int fsflags, 88 struct uio *fsoptions); 89static void free_mntarg(struct mntarg *ma); 90static int vfs_getopt_pos(struct vfsoptlist *opts, const char *name); 91 92static int usermount = 0; 93SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 94 "Unprivileged users may mount and unmount file systems"); 95 96MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 97MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 98static uma_zone_t mount_zone; 99 100/* List of mounted filesystems. */ 101struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 102 103/* For any iteration/modification of mountlist */ 104struct mtx mountlist_mtx; 105MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 106 107TAILQ_HEAD(vfsoptlist, vfsopt); 108struct vfsopt { 109 TAILQ_ENTRY(vfsopt) link; 110 char *name; 111 void *value; 112 int len; 113}; 114 115/* 116 * The vnode of the system's root (/ in the filesystem, without chroot 117 * active.) 118 */ 119struct vnode *rootvnode; 120 121/* 122 * The root filesystem is detailed in the kernel environment variable 123 * vfs.root.mountfrom, which is expected to be in the general format 124 * 125 * <vfsname>:[<path>] 126 * vfsname := the name of a VFS known to the kernel and capable 127 * of being mounted as root 128 * path := disk device name or other data used by the filesystem 129 * to locate its physical store 130 */ 131 132/* 133 * Global opts, taken by all filesystems 134 */ 135static const char *global_opts[] = { 136 "errmsg", 137 "fstype", 138 "fspath", 139 "rdonly", 140 "ro", 141 "rw", 142 "suid", 143 "exec", 144 "update", 145 NULL 146}; 147 148/* 149 * The root specifiers we will try if RB_CDROM is specified. 150 */ 151static char *cdrom_rootdevnames[] = { 152 "cd9660:cd0", 153 "cd9660:acd0", 154 NULL 155}; 156 157/* legacy find-root code */ 158char *rootdevnames[2] = {NULL, NULL}; 159#ifndef ROOTDEVNAME 160# define ROOTDEVNAME NULL 161#endif 162static const char *ctrootdevname = ROOTDEVNAME; 163 164/* 165 * --------------------------------------------------------------------- 166 * Functions for building and sanitizing the mount options 167 */ 168 169/* Remove one mount option. */ 170static void 171vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 172{ 173 174 TAILQ_REMOVE(opts, opt, link); 175 free(opt->name, M_MOUNT); 176 if (opt->value != NULL) 177 free(opt->value, M_MOUNT); 178#ifdef INVARIANTS 179 else if (opt->len != 0) 180 panic("%s: mount option with NULL value but length != 0", 181 __func__); 182#endif 183 free(opt, M_MOUNT); 184} 185 186/* Release all resources related to the mount options. */ 187void 188vfs_freeopts(struct vfsoptlist *opts) 189{ 190 struct vfsopt *opt; 191 192 while (!TAILQ_EMPTY(opts)) { 193 opt = TAILQ_FIRST(opts); 194 vfs_freeopt(opts, opt); 195 } 196 free(opts, M_MOUNT); 197} 198 199void 200vfs_deleteopt(struct vfsoptlist *opts, const char *name) 201{ 202 struct vfsopt *opt, *temp; 203 204 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 205 if (strcmp(opt->name, name) == 0) 206 vfs_freeopt(opts, opt); 207 } 208} 209 210/* 211 * Check if options are equal (with or without the "no" prefix). 212 */ 213static int 214vfs_equalopts(const char *opt1, const char *opt2) 215{ 216 217 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 218 if (strcmp(opt1, opt2) == 0) 219 return (1); 220 /* "noopt" vs. "opt" */ 221 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 222 return (1); 223 /* "opt" vs. "noopt" */ 224 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 225 return (1); 226 return (0); 227} 228 229/* 230 * If a mount option is specified several times, 231 * (with or without the "no" prefix) only keep 232 * the last occurence of it. 233 */ 234static void 235vfs_sanitizeopts(struct vfsoptlist *opts) 236{ 237 struct vfsopt *opt, *opt2, *tmp; 238 239 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 240 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 241 while (opt2 != NULL) { 242 if (vfs_equalopts(opt->name, opt2->name)) { 243 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 244 vfs_freeopt(opts, opt2); 245 opt2 = tmp; 246 } else { 247 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 248 } 249 } 250 } 251} 252 253/* 254 * Build a linked list of mount options from a struct uio. 255 */ 256static int 257vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 258{ 259 struct vfsoptlist *opts; 260 struct vfsopt *opt; 261 size_t memused; 262 unsigned int i, iovcnt; 263 int error, namelen, optlen; 264 265 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 266 TAILQ_INIT(opts); 267 memused = 0; 268 iovcnt = auio->uio_iovcnt; 269 for (i = 0; i < iovcnt; i += 2) { 270 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 271 namelen = auio->uio_iov[i].iov_len; 272 optlen = auio->uio_iov[i + 1].iov_len; 273 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 274 opt->value = NULL; 275 opt->len = 0; 276 277 /* 278 * Do this early, so jumps to "bad" will free the current 279 * option. 280 */ 281 TAILQ_INSERT_TAIL(opts, opt, link); 282 memused += sizeof(struct vfsopt) + optlen + namelen; 283 284 /* 285 * Avoid consuming too much memory, and attempts to overflow 286 * memused. 287 */ 288 if (memused > VFS_MOUNTARG_SIZE_MAX || 289 optlen > VFS_MOUNTARG_SIZE_MAX || 290 namelen > VFS_MOUNTARG_SIZE_MAX) { 291 error = EINVAL; 292 goto bad; 293 } 294 295 if (auio->uio_segflg == UIO_SYSSPACE) { 296 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 297 } else { 298 error = copyin(auio->uio_iov[i].iov_base, opt->name, 299 namelen); 300 if (error) 301 goto bad; 302 } 303 /* Ensure names are null-terminated strings. */ 304 if (opt->name[namelen - 1] != '\0') { 305 error = EINVAL; 306 goto bad; 307 } 308 if (optlen != 0) { 309 opt->len = optlen; 310 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 311 if (auio->uio_segflg == UIO_SYSSPACE) { 312 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 313 optlen); 314 } else { 315 error = copyin(auio->uio_iov[i + 1].iov_base, 316 opt->value, optlen); 317 if (error) 318 goto bad; 319 } 320 } 321 } 322 vfs_sanitizeopts(opts); 323 *options = opts; 324 return (0); 325bad: 326 vfs_freeopts(opts); 327 return (error); 328} 329 330/* 331 * Merge the old mount options with the new ones passed 332 * in the MNT_UPDATE case. 333 */ 334static void 335vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 336{ 337 struct vfsopt *opt, *opt2, *new; 338 339 TAILQ_FOREACH(opt, opts, link) { 340 /* 341 * Check that this option hasn't been redefined 342 * nor cancelled with a "no" mount option. 343 */ 344 opt2 = TAILQ_FIRST(toopts); 345 while (opt2 != NULL) { 346 if (strcmp(opt2->name, opt->name) == 0) 347 goto next; 348 if (strncmp(opt2->name, "no", 2) == 0 && 349 strcmp(opt2->name + 2, opt->name) == 0) { 350 vfs_freeopt(toopts, opt2); 351 goto next; 352 } 353 opt2 = TAILQ_NEXT(opt2, link); 354 } 355 /* We want this option, duplicate it. */ 356 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 357 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 358 strcpy(new->name, opt->name); 359 if (opt->len != 0) { 360 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 361 bcopy(opt->value, new->value, opt->len); 362 } else { 363 new->value = NULL; 364 } 365 new->len = opt->len; 366 TAILQ_INSERT_TAIL(toopts, new, link); 367next: 368 continue; 369 } 370} 371 372/* 373 * Mount a filesystem. 374 */ 375int 376nmount(td, uap) 377 struct thread *td; 378 struct nmount_args /* { 379 struct iovec *iovp; 380 unsigned int iovcnt; 381 int flags; 382 } */ *uap; 383{ 384 struct uio *auio; 385 struct iovec *iov; 386 unsigned int i; 387 int error; 388 u_int iovcnt; 389 390 AUDIT_ARG(fflags, uap->flags); 391 392 /* Kick out MNT_ROOTFS early as it is legal internally */ 393 if (uap->flags & MNT_ROOTFS) 394 return (EINVAL); 395 396 iovcnt = uap->iovcnt; 397 /* 398 * Check that we have an even number of iovec's 399 * and that we have at least two options. 400 */ 401 if ((iovcnt & 1) || (iovcnt < 4)) 402 return (EINVAL); 403 404 error = copyinuio(uap->iovp, iovcnt, &auio); 405 if (error) 406 return (error); 407 iov = auio->uio_iov; 408 for (i = 0; i < iovcnt; i++) { 409 if (iov->iov_len > MMAXOPTIONLEN) { 410 free(auio, M_IOV); 411 return (EINVAL); 412 } 413 iov++; 414 } 415 error = vfs_donmount(td, uap->flags, auio); 416 417 free(auio, M_IOV); 418 return (error); 419} 420 421/* 422 * --------------------------------------------------------------------- 423 * Various utility functions 424 */ 425 426void 427vfs_ref(struct mount *mp) 428{ 429 430 MNT_ILOCK(mp); 431 MNT_REF(mp); 432 MNT_IUNLOCK(mp); 433} 434 435void 436vfs_rel(struct mount *mp) 437{ 438 439 MNT_ILOCK(mp); 440 MNT_REL(mp); 441 MNT_IUNLOCK(mp); 442} 443 444static int 445mount_init(void *mem, int size, int flags) 446{ 447 struct mount *mp; 448 449 mp = (struct mount *)mem; 450 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 451 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 452 return (0); 453} 454 455static void 456mount_fini(void *mem, int size) 457{ 458 struct mount *mp; 459 460 mp = (struct mount *)mem; 461 lockdestroy(&mp->mnt_lock); 462 mtx_destroy(&mp->mnt_mtx); 463} 464 465/* 466 * Allocate and initialize the mount point struct. 467 */ 468static struct mount * 469vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, 470 const char *fspath, struct thread *td) 471{ 472 struct mount *mp; 473 474 mp = uma_zalloc(mount_zone, M_WAITOK); 475 bzero(&mp->mnt_startzero, 476 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 477 TAILQ_INIT(&mp->mnt_nvnodelist); 478 mp->mnt_nvnodelistsize = 0; 479 mp->mnt_ref = 0; 480 (void) vfs_busy(mp, LK_NOWAIT, 0, td); 481 mp->mnt_op = vfsp->vfc_vfsops; 482 mp->mnt_vfc = vfsp; 483 vfsp->vfc_refcount++; /* XXX Unlocked */ 484 mp->mnt_stat.f_type = vfsp->vfc_typenum; 485 MNT_ILOCK(mp); 486 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 487 MNT_IUNLOCK(mp); 488 mp->mnt_gen++; 489 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 490 mp->mnt_vnodecovered = vp; 491 mp->mnt_cred = crdup(td->td_ucred); 492 mp->mnt_stat.f_owner = td->td_ucred->cr_uid; 493 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 494 mp->mnt_iosize_max = DFLTPHYS; 495#ifdef MAC 496 mac_init_mount(mp); 497 mac_create_mount(td->td_ucred, mp); 498#endif 499 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 500 return (mp); 501} 502 503/* 504 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 505 */ 506void 507vfs_mount_destroy(struct mount *mp) 508{ 509 int i; 510 511 MNT_ILOCK(mp); 512 for (i = 0; mp->mnt_ref && i < 3; i++) 513 msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz); 514 /* 515 * This will always cause a 3 second delay in rebooting due to 516 * refs on the root mountpoint that never go away. Most of these 517 * are held by init which never exits. 518 */ 519 if (i == 3 && (!rebooting || bootverbose)) 520 printf("Mount point %s had %d dangling refs\n", 521 mp->mnt_stat.f_mntonname, mp->mnt_ref); 522 if (mp->mnt_holdcnt != 0) { 523 printf("Waiting for mount point to be unheld\n"); 524 while (mp->mnt_holdcnt != 0) { 525 mp->mnt_holdcntwaiters++; 526 msleep(&mp->mnt_holdcnt, MNT_MTX(mp), 527 PZERO, "mntdestroy", 0); 528 mp->mnt_holdcntwaiters--; 529 } 530 printf("mount point unheld\n"); 531 } 532 if (mp->mnt_writeopcount > 0) { 533 printf("Waiting for mount point write ops\n"); 534 while (mp->mnt_writeopcount > 0) { 535 mp->mnt_kern_flag |= MNTK_SUSPEND; 536 msleep(&mp->mnt_writeopcount, 537 MNT_MTX(mp), 538 PZERO, "mntdestroy2", 0); 539 } 540 printf("mount point write ops completed\n"); 541 } 542 if (mp->mnt_secondary_writes > 0) { 543 printf("Waiting for mount point secondary write ops\n"); 544 while (mp->mnt_secondary_writes > 0) { 545 mp->mnt_kern_flag |= MNTK_SUSPEND; 546 msleep(&mp->mnt_secondary_writes, 547 MNT_MTX(mp), 548 PZERO, "mntdestroy3", 0); 549 } 550 printf("mount point secondary write ops completed\n"); 551 } 552 MNT_IUNLOCK(mp); 553 mp->mnt_vfc->vfc_refcount--; 554 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 555 struct vnode *vp; 556 557 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 558 vprint("", vp); 559 panic("unmount: dangling vnode"); 560 } 561 MNT_ILOCK(mp); 562 if (mp->mnt_kern_flag & MNTK_MWAIT) 563 wakeup(mp); 564 if (mp->mnt_writeopcount != 0) 565 panic("vfs_mount_destroy: nonzero writeopcount"); 566 if (mp->mnt_secondary_writes != 0) 567 panic("vfs_mount_destroy: nonzero secondary_writes"); 568 if (mp->mnt_nvnodelistsize != 0) 569 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 570 mp->mnt_writeopcount = -1000; 571 mp->mnt_nvnodelistsize = -1000; 572 mp->mnt_secondary_writes = -1000; 573 MNT_IUNLOCK(mp); 574#ifdef MAC 575 mac_destroy_mount(mp); 576#endif 577 if (mp->mnt_opt != NULL) 578 vfs_freeopts(mp->mnt_opt); 579 crfree(mp->mnt_cred); 580 uma_zfree(mount_zone, mp); 581} 582 583static int 584vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 585{ 586 struct vfsoptlist *optlist; 587 struct vfsopt *opt, *noro_opt; 588 char *fstype, *fspath, *errmsg; 589 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 590 int has_rw, has_noro; 591 592 errmsg = NULL; 593 errmsg_len = 0; 594 errmsg_pos = -1; 595 has_rw = 0; 596 has_noro = 0; 597 598 error = vfs_buildopts(fsoptions, &optlist); 599 if (error) 600 return (error); 601 602 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 603 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 604 605 /* 606 * We need these two options before the others, 607 * and they are mandatory for any filesystem. 608 * Ensure they are NUL terminated as well. 609 */ 610 fstypelen = 0; 611 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 612 if (error || fstype[fstypelen - 1] != '\0') { 613 error = EINVAL; 614 if (errmsg != NULL) 615 strncpy(errmsg, "Invalid fstype", errmsg_len); 616 goto bail; 617 } 618 fspathlen = 0; 619 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 620 if (error || fspath[fspathlen - 1] != '\0') { 621 error = EINVAL; 622 if (errmsg != NULL) 623 strncpy(errmsg, "Invalid fspath", errmsg_len); 624 goto bail; 625 } 626 627 /* 628 * We need to see if we have the "update" option 629 * before we call vfs_domount(), since vfs_domount() has special 630 * logic based on MNT_UPDATE. This is very important 631 * when we want to update the root filesystem. 632 */ 633 TAILQ_FOREACH(opt, optlist, link) { 634 if (strcmp(opt->name, "update") == 0) 635 fsflags |= MNT_UPDATE; 636 else if (strcmp(opt->name, "async") == 0) 637 fsflags |= MNT_ASYNC; 638 else if (strcmp(opt->name, "force") == 0) 639 fsflags |= MNT_FORCE; 640 else if (strcmp(opt->name, "multilabel") == 0) 641 fsflags |= MNT_MULTILABEL; 642 else if (strcmp(opt->name, "noasync") == 0) 643 fsflags &= ~MNT_ASYNC; 644 else if (strcmp(opt->name, "noatime") == 0) 645 fsflags |= MNT_NOATIME; 646 else if (strcmp(opt->name, "noclusterr") == 0) 647 fsflags |= MNT_NOCLUSTERR; 648 else if (strcmp(opt->name, "noclusterw") == 0) 649 fsflags |= MNT_NOCLUSTERW; 650 else if (strcmp(opt->name, "noexec") == 0) 651 fsflags |= MNT_NOEXEC; 652 else if (strcmp(opt->name, "nosuid") == 0) 653 fsflags |= MNT_NOSUID; 654 else if (strcmp(opt->name, "nosymfollow") == 0) 655 fsflags |= MNT_NOSYMFOLLOW; 656 else if (strcmp(opt->name, "noro") == 0) { 657 fsflags &= ~MNT_RDONLY; 658 has_noro = 1; 659 } 660 else if (strcmp(opt->name, "rw") == 0) { 661 fsflags &= ~MNT_RDONLY; 662 has_rw = 1; 663 } 664 else if (strcmp(opt->name, "ro") == 0 || 665 strcmp(opt->name, "rdonly") == 0) 666 fsflags |= MNT_RDONLY; 667 else if (strcmp(opt->name, "snapshot") == 0) 668 fsflags |= MNT_SNAPSHOT; 669 else if (strcmp(opt->name, "suiddir") == 0) 670 fsflags |= MNT_SUIDDIR; 671 else if (strcmp(opt->name, "sync") == 0) 672 fsflags |= MNT_SYNCHRONOUS; 673 else if (strcmp(opt->name, "union") == 0) 674 fsflags |= MNT_UNION; 675 } 676 677 /* 678 * If "rw" was specified as a mount option, and we 679 * are trying to update a mount-point from "ro" to "rw", 680 * we need a mount option "noro", since in vfs_mergeopts(), 681 * "noro" will cancel "ro", but "rw" will not do anything. 682 */ 683 if (has_rw && !has_noro) { 684 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 685 noro_opt->name = strdup("noro", M_MOUNT); 686 noro_opt->value = NULL; 687 noro_opt->len = 0; 688 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 689 } 690 691 /* 692 * Be ultra-paranoid about making sure the type and fspath 693 * variables will fit in our mp buffers, including the 694 * terminating NUL. 695 */ 696 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 697 error = ENAMETOOLONG; 698 goto bail; 699 } 700 701 mtx_lock(&Giant); 702 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 703 mtx_unlock(&Giant); 704bail: 705 /* copyout the errmsg */ 706 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 707 && errmsg_len > 0 && errmsg != NULL) { 708 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 709 bcopy(errmsg, 710 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 711 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 712 } else { 713 copyout(errmsg, 714 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 715 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 716 } 717 } 718 719 if (error != 0) 720 vfs_freeopts(optlist); 721 return (error); 722} 723 724/* 725 * Old mount API. 726 */ 727#ifndef _SYS_SYSPROTO_H_ 728struct mount_args { 729 char *type; 730 char *path; 731 int flags; 732 caddr_t data; 733}; 734#endif 735/* ARGSUSED */ 736int 737mount(td, uap) 738 struct thread *td; 739 struct mount_args /* { 740 char *type; 741 char *path; 742 int flags; 743 caddr_t data; 744 } */ *uap; 745{ 746 char *fstype; 747 struct vfsconf *vfsp = NULL; 748 struct mntarg *ma = NULL; 749 int error; 750 751 AUDIT_ARG(fflags, uap->flags); 752 753 /* Kick out MNT_ROOTFS early as it is legal internally */ 754 uap->flags &= ~MNT_ROOTFS; 755 756 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 757 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 758 if (error) { 759 free(fstype, M_TEMP); 760 return (error); 761 } 762 763 AUDIT_ARG(text, fstype); 764 mtx_lock(&Giant); 765 vfsp = vfs_byname_kld(fstype, td, &error); 766 free(fstype, M_TEMP); 767 if (vfsp == NULL) { 768 mtx_unlock(&Giant); 769 return (ENOENT); 770 } 771 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 772 mtx_unlock(&Giant); 773 return (EOPNOTSUPP); 774 } 775 776 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 777 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 778 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 779 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 780 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 781 782 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 783 mtx_unlock(&Giant); 784 return (error); 785} 786 787 788/* 789 * vfs_domount(): actually attempt a filesystem mount. 790 */ 791static int 792vfs_domount( 793 struct thread *td, /* Calling thread. */ 794 const char *fstype, /* Filesystem type. */ 795 char *fspath, /* Mount path. */ 796 int fsflags, /* Flags common to all filesystems. */ 797 void *fsdata /* Options local to the filesystem. */ 798 ) 799{ 800 struct vnode *vp; 801 struct mount *mp; 802 struct vfsconf *vfsp; 803 struct export_args export; 804 int error, flag = 0; 805 struct vattr va; 806 struct nameidata nd; 807 808 mtx_assert(&Giant, MA_OWNED); 809 /* 810 * Be ultra-paranoid about making sure the type and fspath 811 * variables will fit in our mp buffers, including the 812 * terminating NUL. 813 */ 814 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 815 return (ENAMETOOLONG); 816 817 if (jailed(td->td_ucred) || usermount == 0) { 818 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 819 return (error); 820 } 821 822 /* 823 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 824 */ 825 if (fsflags & MNT_EXPORTED) { 826 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 827 if (error) 828 return (error); 829 } 830 if (fsflags & MNT_SUIDDIR) { 831 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 832 if (error) 833 return (error); 834 } 835 /* 836 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 837 */ 838 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 839 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 840 fsflags |= MNT_NOSUID | MNT_USER; 841 } 842 843 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 844 vfsp = NULL; 845 if ((fsflags & MNT_UPDATE) == 0) { 846 /* Don't try to load KLDs if we're mounting the root. */ 847 if (fsflags & MNT_ROOTFS) 848 vfsp = vfs_byname(fstype); 849 else 850 vfsp = vfs_byname_kld(fstype, td, &error); 851 if (vfsp == NULL) 852 return (ENODEV); 853 } 854 /* 855 * Get vnode to be covered 856 */ 857 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 858 fspath, td); 859 if ((error = namei(&nd)) != 0) 860 return (error); 861 NDFREE(&nd, NDF_ONLY_PNBUF); 862 vp = nd.ni_vp; 863 if (fsflags & MNT_UPDATE) { 864 if ((vp->v_vflag & VV_ROOT) == 0) { 865 vput(vp); 866 return (EINVAL); 867 } 868 mp = vp->v_mount; 869 MNT_ILOCK(mp); 870 flag = mp->mnt_flag; 871 /* 872 * We only allow the filesystem to be reloaded if it 873 * is currently mounted read-only. 874 */ 875 if ((fsflags & MNT_RELOAD) && 876 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 877 MNT_IUNLOCK(mp); 878 vput(vp); 879 return (EOPNOTSUPP); /* Needs translation */ 880 } 881 MNT_IUNLOCK(mp); 882 /* 883 * Only privileged root, or (if MNT_USER is set) the user that 884 * did the original mount is permitted to update it. 885 */ 886 error = vfs_suser(mp, td); 887 if (error) { 888 vput(vp); 889 return (error); 890 } 891 if (vfs_busy(mp, LK_NOWAIT, 0, td)) { 892 vput(vp); 893 return (EBUSY); 894 } 895 VI_LOCK(vp); 896 if ((vp->v_iflag & VI_MOUNT) != 0 || 897 vp->v_mountedhere != NULL) { 898 VI_UNLOCK(vp); 899 vfs_unbusy(mp, td); 900 vput(vp); 901 return (EBUSY); 902 } 903 vp->v_iflag |= VI_MOUNT; 904 VI_UNLOCK(vp); 905 MNT_ILOCK(mp); 906 mp->mnt_flag |= fsflags & 907 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 908 MNT_IUNLOCK(mp); 909 VOP_UNLOCK(vp, 0, td); 910 mp->mnt_optnew = fsdata; 911 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 912 } else { 913 /* 914 * If the user is not root, ensure that they own the directory 915 * onto which we are attempting to mount. 916 */ 917 error = VOP_GETATTR(vp, &va, td->td_ucred, td); 918 if (error) { 919 vput(vp); 920 return (error); 921 } 922 if (va.va_uid != td->td_ucred->cr_uid) { 923 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 924 SUSER_ALLOWJAIL); 925 if (error) { 926 vput(vp); 927 return (error); 928 } 929 } 930 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 931 if (error != 0) { 932 vput(vp); 933 return (error); 934 } 935 if (vp->v_type != VDIR) { 936 vput(vp); 937 return (ENOTDIR); 938 } 939 VI_LOCK(vp); 940 if ((vp->v_iflag & VI_MOUNT) != 0 || 941 vp->v_mountedhere != NULL) { 942 VI_UNLOCK(vp); 943 vput(vp); 944 return (EBUSY); 945 } 946 vp->v_iflag |= VI_MOUNT; 947 VI_UNLOCK(vp); 948 949 /* 950 * Allocate and initialize the filesystem. 951 */ 952 mp = vfs_mount_alloc(vp, vfsp, fspath, td); 953 VOP_UNLOCK(vp, 0, td); 954 955 /* XXXMAC: pass to vfs_mount_alloc? */ 956 mp->mnt_optnew = fsdata; 957 } 958 959 /* 960 * Set the mount level flags. 961 */ 962 MNT_ILOCK(mp); 963 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 964 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 965 MNT_RDONLY)); 966 if ((mp->mnt_flag & MNT_ASYNC) == 0) 967 mp->mnt_kern_flag &= ~MNTK_ASYNC; 968 MNT_IUNLOCK(mp); 969 /* 970 * Mount the filesystem. 971 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 972 * get. No freeing of cn_pnbuf. 973 */ 974 error = VFS_MOUNT(mp, td); 975 976 /* 977 * Process the export option only if we are 978 * updating mount options. 979 */ 980 if (!error && (fsflags & MNT_UPDATE)) { 981 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 982 sizeof(export)) == 0) 983 error = vfs_export(mp, &export); 984 } 985 986 if (!error) { 987 if (mp->mnt_opt != NULL) 988 vfs_freeopts(mp->mnt_opt); 989 mp->mnt_opt = mp->mnt_optnew; 990 (void)VFS_STATFS(mp, &mp->mnt_stat, td); 991 } 992 /* 993 * Prevent external consumers of mount options from reading 994 * mnt_optnew. 995 */ 996 mp->mnt_optnew = NULL; 997 if (mp->mnt_flag & MNT_UPDATE) { 998 MNT_ILOCK(mp); 999 if (error) 1000 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1001 (flag & ~MNT_QUOTA); 1002 else 1003 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1004 MNT_FORCE | MNT_SNAPSHOT); 1005 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1006 mp->mnt_kern_flag |= MNTK_ASYNC; 1007 else 1008 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1009 MNT_IUNLOCK(mp); 1010 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1011 if (mp->mnt_syncer == NULL) 1012 error = vfs_allocate_syncvnode(mp); 1013 } else { 1014 if (mp->mnt_syncer != NULL) 1015 vrele(mp->mnt_syncer); 1016 mp->mnt_syncer = NULL; 1017 } 1018 vfs_unbusy(mp, td); 1019 VI_LOCK(vp); 1020 vp->v_iflag &= ~VI_MOUNT; 1021 VI_UNLOCK(vp); 1022 vrele(vp); 1023 return (error); 1024 } 1025 MNT_ILOCK(mp); 1026 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1027 mp->mnt_kern_flag |= MNTK_ASYNC; 1028 else 1029 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1030 MNT_IUNLOCK(mp); 1031 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1032 /* 1033 * Put the new filesystem on the mount list after root. 1034 */ 1035 cache_purge(vp); 1036 if (!error) { 1037 struct vnode *newdp; 1038 1039 VI_LOCK(vp); 1040 vp->v_iflag &= ~VI_MOUNT; 1041 VI_UNLOCK(vp); 1042 vp->v_mountedhere = mp; 1043 mtx_lock(&mountlist_mtx); 1044 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1045 mtx_unlock(&mountlist_mtx); 1046 vfs_event_signal(NULL, VQ_MOUNT, 0); 1047 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 1048 panic("mount: lost mount"); 1049 mountcheckdirs(vp, newdp); 1050 vput(newdp); 1051 VOP_UNLOCK(vp, 0, td); 1052 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1053 error = vfs_allocate_syncvnode(mp); 1054 vfs_unbusy(mp, td); 1055 if (error) 1056 vrele(vp); 1057 } else { 1058 VI_LOCK(vp); 1059 vp->v_iflag &= ~VI_MOUNT; 1060 VI_UNLOCK(vp); 1061 vfs_unbusy(mp, td); 1062 vfs_mount_destroy(mp); 1063 vput(vp); 1064 } 1065 return (error); 1066} 1067 1068/* 1069 * Unmount a filesystem. 1070 * 1071 * Note: unmount takes a path to the vnode mounted on as argument, not 1072 * special file (as before). 1073 */ 1074#ifndef _SYS_SYSPROTO_H_ 1075struct unmount_args { 1076 char *path; 1077 int flags; 1078}; 1079#endif 1080/* ARGSUSED */ 1081int 1082unmount(td, uap) 1083 struct thread *td; 1084 register struct unmount_args /* { 1085 char *path; 1086 int flags; 1087 } */ *uap; 1088{ 1089 struct mount *mp; 1090 char *pathbuf; 1091 int error, id0, id1; 1092 1093 if (jailed(td->td_ucred) || usermount == 0) { 1094 error = priv_check(td, PRIV_VFS_UNMOUNT); 1095 if (error) 1096 return (error); 1097 } 1098 1099 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1100 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1101 if (error) { 1102 free(pathbuf, M_TEMP); 1103 return (error); 1104 } 1105 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1106 mtx_lock(&Giant); 1107 if (uap->flags & MNT_BYFSID) { 1108 /* Decode the filesystem ID. */ 1109 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1110 mtx_unlock(&Giant); 1111 free(pathbuf, M_TEMP); 1112 return (EINVAL); 1113 } 1114 1115 mtx_lock(&mountlist_mtx); 1116 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1117 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1118 mp->mnt_stat.f_fsid.val[1] == id1) 1119 break; 1120 } 1121 mtx_unlock(&mountlist_mtx); 1122 } else { 1123 mtx_lock(&mountlist_mtx); 1124 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1125 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1126 break; 1127 } 1128 mtx_unlock(&mountlist_mtx); 1129 } 1130 free(pathbuf, M_TEMP); 1131 if (mp == NULL) { 1132 /* 1133 * Previously we returned ENOENT for a nonexistent path and 1134 * EINVAL for a non-mountpoint. We cannot tell these apart 1135 * now, so in the !MNT_BYFSID case return the more likely 1136 * EINVAL for compatibility. 1137 */ 1138 mtx_unlock(&Giant); 1139 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1140 } 1141 1142 /* 1143 * Don't allow unmounting the root filesystem. 1144 */ 1145 if (mp->mnt_flag & MNT_ROOTFS) { 1146 mtx_unlock(&Giant); 1147 return (EINVAL); 1148 } 1149 error = dounmount(mp, uap->flags, td); 1150 mtx_unlock(&Giant); 1151 return (error); 1152} 1153 1154/* 1155 * Do the actual filesystem unmount. 1156 */ 1157int 1158dounmount(mp, flags, td) 1159 struct mount *mp; 1160 int flags; 1161 struct thread *td; 1162{ 1163 struct vnode *coveredvp, *fsrootvp; 1164 int error; 1165 int async_flag; 1166 int mnt_gen_r; 1167 1168 mtx_assert(&Giant, MA_OWNED); 1169 1170 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1171 mnt_gen_r = mp->mnt_gen; 1172 VI_LOCK(coveredvp); 1173 vholdl(coveredvp); 1174 error = vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK, td); 1175 vdrop(coveredvp); 1176 /* 1177 * Check for mp being unmounted while waiting for the 1178 * covered vnode lock. 1179 */ 1180 if (error) 1181 return (error); 1182 if (coveredvp->v_mountedhere != mp || 1183 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1184 VOP_UNLOCK(coveredvp, 0, td); 1185 return (EBUSY); 1186 } 1187 } 1188 /* 1189 * Only privileged root, or (if MNT_USER is set) the user that did the 1190 * original mount is permitted to unmount this filesystem. 1191 */ 1192 error = vfs_suser(mp, td); 1193 if (error) { 1194 if (coveredvp) 1195 VOP_UNLOCK(coveredvp, 0, td); 1196 return (error); 1197 } 1198 1199 MNT_ILOCK(mp); 1200 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1201 MNT_IUNLOCK(mp); 1202 if (coveredvp) 1203 VOP_UNLOCK(coveredvp, 0, td); 1204 return (EBUSY); 1205 } 1206 mp->mnt_kern_flag |= MNTK_UNMOUNT; 1207 /* Allow filesystems to detect that a forced unmount is in progress. */ 1208 if (flags & MNT_FORCE) 1209 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1210 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 1211 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td); 1212 if (error) { 1213 MNT_ILOCK(mp); 1214 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1215 if (mp->mnt_kern_flag & MNTK_MWAIT) 1216 wakeup(mp); 1217 MNT_IUNLOCK(mp); 1218 if (coveredvp) 1219 VOP_UNLOCK(coveredvp, 0, td); 1220 return (error); 1221 } 1222 vn_start_write(NULL, &mp, V_WAIT); 1223 1224 if (mp->mnt_flag & MNT_EXPUBLIC) 1225 vfs_setpublicfs(NULL, NULL, NULL); 1226 1227 vfs_msync(mp, MNT_WAIT); 1228 MNT_ILOCK(mp); 1229 async_flag = mp->mnt_flag & MNT_ASYNC; 1230 mp->mnt_flag &= ~MNT_ASYNC; 1231 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1232 MNT_IUNLOCK(mp); 1233 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1234 if (mp->mnt_syncer != NULL) 1235 vrele(mp->mnt_syncer); 1236 /* 1237 * For forced unmounts, move process cdir/rdir refs on the fs root 1238 * vnode to the covered vnode. For non-forced unmounts we want 1239 * such references to cause an EBUSY error. 1240 */ 1241 if ((flags & MNT_FORCE) && 1242 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1243 if (mp->mnt_vnodecovered != NULL) 1244 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1245 if (fsrootvp == rootvnode) { 1246 vrele(rootvnode); 1247 rootvnode = NULL; 1248 } 1249 vput(fsrootvp); 1250 } 1251 if (((mp->mnt_flag & MNT_RDONLY) || 1252 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1253 (flags & MNT_FORCE)) { 1254 error = VFS_UNMOUNT(mp, flags, td); 1255 } 1256 vn_finished_write(mp); 1257 if (error) { 1258 /* Undo cdir/rdir and rootvnode changes made above. */ 1259 if ((flags & MNT_FORCE) && 1260 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1261 if (mp->mnt_vnodecovered != NULL) 1262 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1263 if (rootvnode == NULL) { 1264 rootvnode = fsrootvp; 1265 vref(rootvnode); 1266 } 1267 vput(fsrootvp); 1268 } 1269 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) 1270 (void) vfs_allocate_syncvnode(mp); 1271 MNT_ILOCK(mp); 1272 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1273 mp->mnt_flag |= async_flag; 1274 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1275 mp->mnt_kern_flag |= MNTK_ASYNC; 1276 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1277 if (mp->mnt_kern_flag & MNTK_MWAIT) 1278 wakeup(mp); 1279 MNT_IUNLOCK(mp); 1280 if (coveredvp) 1281 VOP_UNLOCK(coveredvp, 0, td); 1282 return (error); 1283 } 1284 mtx_lock(&mountlist_mtx); 1285 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1286 mtx_unlock(&mountlist_mtx); 1287 if (coveredvp != NULL) { 1288 coveredvp->v_mountedhere = NULL; 1289 vput(coveredvp); 1290 } 1291 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1292 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1293 vfs_mount_destroy(mp); 1294 return (0); 1295} 1296 1297/* 1298 * --------------------------------------------------------------------- 1299 * Mounting of root filesystem 1300 * 1301 */ 1302 1303struct root_hold_token { 1304 const char *who; 1305 LIST_ENTRY(root_hold_token) list; 1306}; 1307 1308static LIST_HEAD(, root_hold_token) root_holds = 1309 LIST_HEAD_INITIALIZER(&root_holds); 1310 1311struct root_hold_token * 1312root_mount_hold(const char *identifier) 1313{ 1314 struct root_hold_token *h; 1315 1316 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1317 h->who = identifier; 1318 mtx_lock(&mountlist_mtx); 1319 LIST_INSERT_HEAD(&root_holds, h, list); 1320 mtx_unlock(&mountlist_mtx); 1321 return (h); 1322} 1323 1324void 1325root_mount_rel(struct root_hold_token *h) 1326{ 1327 1328 mtx_lock(&mountlist_mtx); 1329 LIST_REMOVE(h, list); 1330 wakeup(&root_holds); 1331 mtx_unlock(&mountlist_mtx); 1332 free(h, M_DEVBUF); 1333} 1334 1335static void 1336root_mount_wait(void) 1337{ 1338 struct root_hold_token *h; 1339 1340 for (;;) { 1341 DROP_GIANT(); 1342 g_waitidle(); 1343 PICKUP_GIANT(); 1344 mtx_lock(&mountlist_mtx); 1345 if (LIST_EMPTY(&root_holds)) { 1346 mtx_unlock(&mountlist_mtx); 1347 break; 1348 } 1349 printf("Root mount waiting for:"); 1350 LIST_FOREACH(h, &root_holds, list) 1351 printf(" %s", h->who); 1352 printf("\n"); 1353 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1354 hz); 1355 } 1356} 1357 1358static void 1359set_rootvnode(struct thread *td) 1360{ 1361 struct proc *p; 1362 1363 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1364 panic("Cannot find root vnode"); 1365 1366 p = td->td_proc; 1367 FILEDESC_LOCK(p->p_fd); 1368 1369 if (p->p_fd->fd_cdir != NULL) 1370 vrele(p->p_fd->fd_cdir); 1371 p->p_fd->fd_cdir = rootvnode; 1372 VREF(rootvnode); 1373 1374 if (p->p_fd->fd_rdir != NULL) 1375 vrele(p->p_fd->fd_rdir); 1376 p->p_fd->fd_rdir = rootvnode; 1377 VREF(rootvnode); 1378 1379 FILEDESC_UNLOCK(p->p_fd); 1380 1381 VOP_UNLOCK(rootvnode, 0, td); 1382} 1383 1384/* 1385 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1386 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1387 */ 1388 1389static void 1390devfs_first(void) 1391{ 1392 struct thread *td = curthread; 1393 struct vfsoptlist *opts; 1394 struct vfsconf *vfsp; 1395 struct mount *mp = NULL; 1396 int error; 1397 1398 vfsp = vfs_byname("devfs"); 1399 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1400 if (vfsp == NULL) 1401 return; 1402 1403 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td); 1404 1405 error = VFS_MOUNT(mp, td); 1406 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1407 if (error) 1408 return; 1409 1410 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1411 TAILQ_INIT(opts); 1412 mp->mnt_opt = opts; 1413 1414 mtx_lock(&mountlist_mtx); 1415 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1416 mtx_unlock(&mountlist_mtx); 1417 1418 set_rootvnode(td); 1419 1420 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1421 if (error) 1422 printf("kern_symlink /dev -> / returns %d\n", error); 1423} 1424 1425/* 1426 * Surgically move our devfs to be mounted on /dev. 1427 */ 1428 1429static void 1430devfs_fixup(struct thread *td) 1431{ 1432 struct nameidata nd; 1433 int error; 1434 struct vnode *vp, *dvp; 1435 struct mount *mp; 1436 1437 /* Remove our devfs mount from the mountlist and purge the cache */ 1438 mtx_lock(&mountlist_mtx); 1439 mp = TAILQ_FIRST(&mountlist); 1440 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1441 mtx_unlock(&mountlist_mtx); 1442 cache_purgevfs(mp); 1443 1444 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1445 VI_LOCK(dvp); 1446 dvp->v_iflag &= ~VI_MOUNT; 1447 dvp->v_mountedhere = NULL; 1448 VI_UNLOCK(dvp); 1449 1450 /* Set up the real rootvnode, and purge the cache */ 1451 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1452 set_rootvnode(td); 1453 cache_purgevfs(rootvnode->v_mount); 1454 1455 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1456 error = namei(&nd); 1457 if (error) { 1458 printf("Lookup of /dev for devfs, error: %d\n", error); 1459 return; 1460 } 1461 NDFREE(&nd, NDF_ONLY_PNBUF); 1462 vp = nd.ni_vp; 1463 if (vp->v_type != VDIR) { 1464 vput(vp); 1465 } 1466 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 1467 if (error) { 1468 vput(vp); 1469 } 1470 cache_purge(vp); 1471 mp->mnt_vnodecovered = vp; 1472 vp->v_mountedhere = mp; 1473 mtx_lock(&mountlist_mtx); 1474 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1475 mtx_unlock(&mountlist_mtx); 1476 VOP_UNLOCK(vp, 0, td); 1477 vput(dvp); 1478 vfs_unbusy(mp, td); 1479 1480 /* Unlink the no longer needed /dev/dev -> / symlink */ 1481 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1482} 1483 1484/* 1485 * Report errors during filesystem mounting. 1486 */ 1487void 1488vfs_mount_error(struct mount *mp, const char *fmt, ...) 1489{ 1490 struct vfsoptlist *moptlist = mp->mnt_optnew; 1491 va_list ap; 1492 int error, len; 1493 char *errmsg; 1494 1495 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1496 if (error || errmsg == NULL || len <= 0) 1497 return; 1498 1499 va_start(ap, fmt); 1500 vsnprintf(errmsg, (size_t)len, fmt, ap); 1501 va_end(ap); 1502} 1503 1504/* 1505 * Find and mount the root filesystem 1506 */ 1507void 1508vfs_mountroot(void) 1509{ 1510 char *cp; 1511 int error, i, asked = 0; 1512 1513 root_mount_wait(); 1514 1515 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1516 NULL, NULL, mount_init, mount_fini, 1517 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1518 devfs_first(); 1519 1520 /* 1521 * We are booted with instructions to prompt for the root filesystem. 1522 */ 1523 if (boothowto & RB_ASKNAME) { 1524 if (!vfs_mountroot_ask()) 1525 return; 1526 asked = 1; 1527 } 1528 1529 /* 1530 * The root filesystem information is compiled in, and we are 1531 * booted with instructions to use it. 1532 */ 1533 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1534 if (!vfs_mountroot_try(ctrootdevname)) 1535 return; 1536 ctrootdevname = NULL; 1537 } 1538 1539 /* 1540 * We've been given the generic "use CDROM as root" flag. This is 1541 * necessary because one media may be used in many different 1542 * devices, so we need to search for them. 1543 */ 1544 if (boothowto & RB_CDROM) { 1545 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1546 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1547 return; 1548 } 1549 } 1550 1551 /* 1552 * Try to use the value read by the loader from /etc/fstab, or 1553 * supplied via some other means. This is the preferred 1554 * mechanism. 1555 */ 1556 cp = getenv("vfs.root.mountfrom"); 1557 if (cp != NULL) { 1558 error = vfs_mountroot_try(cp); 1559 freeenv(cp); 1560 if (!error) 1561 return; 1562 } 1563 1564 /* 1565 * Try values that may have been computed by code during boot 1566 */ 1567 if (!vfs_mountroot_try(rootdevnames[0])) 1568 return; 1569 if (!vfs_mountroot_try(rootdevnames[1])) 1570 return; 1571 1572 /* 1573 * If we (still) have a compiled-in default, try it. 1574 */ 1575 if (ctrootdevname != NULL) 1576 if (!vfs_mountroot_try(ctrootdevname)) 1577 return; 1578 /* 1579 * Everything so far has failed, prompt on the console if we haven't 1580 * already tried that. 1581 */ 1582 if (!asked) 1583 if (!vfs_mountroot_ask()) 1584 return; 1585 1586 panic("Root mount failed, startup aborted."); 1587} 1588 1589/* 1590 * Mount (mountfrom) as the root filesystem. 1591 */ 1592static int 1593vfs_mountroot_try(const char *mountfrom) 1594{ 1595 struct mount *mp; 1596 char *vfsname, *path; 1597 time_t timebase; 1598 int error; 1599 char patt[32]; 1600 1601 vfsname = NULL; 1602 path = NULL; 1603 mp = NULL; 1604 error = EINVAL; 1605 1606 if (mountfrom == NULL) 1607 return (error); /* don't complain */ 1608 printf("Trying to mount root from %s\n", mountfrom); 1609 1610 /* parse vfs name and path */ 1611 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1612 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1613 vfsname[0] = path[0] = 0; 1614 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1615 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1616 goto out; 1617 1618 if (path[0] == '\0') 1619 strcpy(path, ROOTNAME); 1620 1621 error = kernel_vmount( 1622 MNT_RDONLY | MNT_ROOTFS, 1623 "fstype", vfsname, 1624 "fspath", "/", 1625 "from", path, 1626 NULL); 1627 if (error == 0) { 1628 /* 1629 * We mount devfs prior to mounting the / FS, so the first 1630 * entry will typically be devfs. 1631 */ 1632 mp = TAILQ_FIRST(&mountlist); 1633 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1634 1635 /* 1636 * Iterate over all currently mounted file systems and use 1637 * the time stamp found to check and/or initialize the RTC. 1638 * Typically devfs has no time stamp and the only other FS 1639 * is the actual / FS. 1640 * Call inittodr() only once and pass it the largest of the 1641 * timestamps we encounter. 1642 */ 1643 timebase = 0; 1644 do { 1645 if (mp->mnt_time > timebase) 1646 timebase = mp->mnt_time; 1647 mp = TAILQ_NEXT(mp, mnt_list); 1648 } while (mp != NULL); 1649 inittodr(timebase); 1650 1651 devfs_fixup(curthread); 1652 } 1653out: 1654 free(path, M_MOUNT); 1655 free(vfsname, M_MOUNT); 1656 return (error); 1657} 1658 1659/* 1660 * --------------------------------------------------------------------- 1661 * Interactive root filesystem selection code. 1662 */ 1663 1664static int 1665vfs_mountroot_ask(void) 1666{ 1667 char name[128]; 1668 1669 for(;;) { 1670 printf("\nManual root filesystem specification:\n"); 1671 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1672#if defined(__amd64__) || defined(__i386__) || defined(__ia64__) 1673 printf(" eg. ufs:da0s1a\n"); 1674#else 1675 printf(" eg. ufs:/dev/da0a\n"); 1676#endif 1677 printf(" ? List valid disk boot devices\n"); 1678 printf(" <empty line> Abort manual input\n"); 1679 printf("\nmountroot> "); 1680 gets(name, sizeof(name), 1); 1681 if (name[0] == '\0') 1682 return (1); 1683 if (name[0] == '?') { 1684 printf("\nList of GEOM managed disk devices:\n "); 1685 g_dev_print(); 1686 continue; 1687 } 1688 if (!vfs_mountroot_try(name)) 1689 return (0); 1690 } 1691} 1692 1693/* 1694 * --------------------------------------------------------------------- 1695 * Functions for querying mount options/arguments from filesystems. 1696 */ 1697 1698/* 1699 * Check that no unknown options are given 1700 */ 1701int 1702vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1703{ 1704 struct vfsopt *opt; 1705 const char **t, *p; 1706 1707 TAILQ_FOREACH(opt, opts, link) { 1708 p = opt->name; 1709 if (p[0] == 'n' && p[1] == 'o') 1710 p += 2; 1711 for(t = global_opts; *t != NULL; t++) 1712 if (!strcmp(*t, p)) 1713 break; 1714 if (*t != NULL) 1715 continue; 1716 for(t = legal; *t != NULL; t++) 1717 if (!strcmp(*t, p)) 1718 break; 1719 if (*t != NULL) 1720 continue; 1721 printf("mount option <%s> is unknown\n", p); 1722 return (EINVAL); 1723 } 1724 return (0); 1725} 1726 1727/* 1728 * Get a mount option by its name. 1729 * 1730 * Return 0 if the option was found, ENOENT otherwise. 1731 * If len is non-NULL it will be filled with the length 1732 * of the option. If buf is non-NULL, it will be filled 1733 * with the address of the option. 1734 */ 1735int 1736vfs_getopt(opts, name, buf, len) 1737 struct vfsoptlist *opts; 1738 const char *name; 1739 void **buf; 1740 int *len; 1741{ 1742 struct vfsopt *opt; 1743 1744 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1745 1746 TAILQ_FOREACH(opt, opts, link) { 1747 if (strcmp(name, opt->name) == 0) { 1748 if (len != NULL) 1749 *len = opt->len; 1750 if (buf != NULL) 1751 *buf = opt->value; 1752 return (0); 1753 } 1754 } 1755 return (ENOENT); 1756} 1757 1758static int 1759vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1760{ 1761 struct vfsopt *opt; 1762 int i; 1763 1764 if (opts == NULL) 1765 return (-1); 1766 1767 i = 0; 1768 TAILQ_FOREACH(opt, opts, link) { 1769 if (strcmp(name, opt->name) == 0) 1770 return (i); 1771 ++i; 1772 } 1773 return (-1); 1774} 1775 1776char * 1777vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1778{ 1779 struct vfsopt *opt; 1780 1781 *error = 0; 1782 TAILQ_FOREACH(opt, opts, link) { 1783 if (strcmp(name, opt->name) != 0) 1784 continue; 1785 if (((char *)opt->value)[opt->len - 1] != '\0') { 1786 *error = EINVAL; 1787 return (NULL); 1788 } 1789 return (opt->value); 1790 } 1791 *error = ENOENT; 1792 return (NULL); 1793} 1794 1795int 1796vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1797{ 1798 struct vfsopt *opt; 1799 1800 TAILQ_FOREACH(opt, opts, link) { 1801 if (strcmp(name, opt->name) == 0) { 1802 if (w != NULL) 1803 *w |= val; 1804 return (1); 1805 } 1806 } 1807 if (w != NULL) 1808 *w &= ~val; 1809 return (0); 1810} 1811 1812int 1813vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1814{ 1815 va_list ap; 1816 struct vfsopt *opt; 1817 int ret; 1818 1819 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1820 1821 TAILQ_FOREACH(opt, opts, link) { 1822 if (strcmp(name, opt->name) != 0) 1823 continue; 1824 if (((char *)opt->value)[opt->len - 1] != '\0') 1825 return (0); 1826 va_start(ap, fmt); 1827 ret = vsscanf(opt->value, fmt, ap); 1828 va_end(ap); 1829 return (ret); 1830 } 1831 return (0); 1832} 1833 1834/* 1835 * Find and copy a mount option. 1836 * 1837 * The size of the buffer has to be specified 1838 * in len, if it is not the same length as the 1839 * mount option, EINVAL is returned. 1840 * Returns ENOENT if the option is not found. 1841 */ 1842int 1843vfs_copyopt(opts, name, dest, len) 1844 struct vfsoptlist *opts; 1845 const char *name; 1846 void *dest; 1847 int len; 1848{ 1849 struct vfsopt *opt; 1850 1851 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1852 1853 TAILQ_FOREACH(opt, opts, link) { 1854 if (strcmp(name, opt->name) == 0) { 1855 if (len != opt->len) 1856 return (EINVAL); 1857 bcopy(opt->value, dest, opt->len); 1858 return (0); 1859 } 1860 } 1861 return (ENOENT); 1862} 1863 1864/* 1865 * This is a helper function for filesystems to traverse their 1866 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1867 */ 1868 1869struct vnode * 1870__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 1871{ 1872 struct vnode *vp; 1873 1874 mtx_assert(MNT_MTX(mp), MA_OWNED); 1875 1876 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1877 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 1878 while (vp != NULL && vp->v_type == VMARKER) 1879 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1880 1881 /* Check if we are done */ 1882 if (vp == NULL) { 1883 __mnt_vnode_markerfree(mvp, mp); 1884 return (NULL); 1885 } 1886 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1887 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1888 return (vp); 1889} 1890 1891struct vnode * 1892__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 1893{ 1894 struct vnode *vp; 1895 1896 mtx_assert(MNT_MTX(mp), MA_OWNED); 1897 1898 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1899 while (vp != NULL && vp->v_type == VMARKER) 1900 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1901 1902 /* Check if we are done */ 1903 if (vp == NULL) { 1904 *mvp = NULL; 1905 return (NULL); 1906 } 1907 mp->mnt_holdcnt++; 1908 MNT_IUNLOCK(mp); 1909 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 1910 M_VNODE_MARKER, 1911 M_WAITOK | M_ZERO); 1912 MNT_ILOCK(mp); 1913 (*mvp)->v_type = VMARKER; 1914 1915 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1916 while (vp != NULL && vp->v_type == VMARKER) 1917 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1918 1919 /* Check if we are done */ 1920 if (vp == NULL) { 1921 MNT_IUNLOCK(mp); 1922 free(*mvp, M_VNODE_MARKER); 1923 MNT_ILOCK(mp); 1924 *mvp = NULL; 1925 mp->mnt_holdcnt--; 1926 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 1927 wakeup(&mp->mnt_holdcnt); 1928 return (NULL); 1929 } 1930 mp->mnt_markercnt++; 1931 (*mvp)->v_mount = mp; 1932 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1933 return (vp); 1934} 1935 1936 1937void 1938__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 1939{ 1940 1941 if (*mvp == NULL) 1942 return; 1943 1944 mtx_assert(MNT_MTX(mp), MA_OWNED); 1945 1946 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1947 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1948 MNT_IUNLOCK(mp); 1949 free(*mvp, M_VNODE_MARKER); 1950 MNT_ILOCK(mp); 1951 *mvp = NULL; 1952 1953 mp->mnt_markercnt--; 1954 mp->mnt_holdcnt--; 1955 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 1956 wakeup(&mp->mnt_holdcnt); 1957} 1958 1959 1960int 1961__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 1962{ 1963 int error; 1964 1965 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 1966 if (sbp != &mp->mnt_stat) 1967 *sbp = mp->mnt_stat; 1968 return (error); 1969} 1970 1971void 1972vfs_mountedfrom(struct mount *mp, const char *from) 1973{ 1974 1975 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 1976 strlcpy(mp->mnt_stat.f_mntfromname, from, 1977 sizeof mp->mnt_stat.f_mntfromname); 1978} 1979 1980/* 1981 * --------------------------------------------------------------------- 1982 * This is the api for building mount args and mounting filesystems from 1983 * inside the kernel. 1984 * 1985 * The API works by accumulation of individual args. First error is 1986 * latched. 1987 * 1988 * XXX: should be documented in new manpage kernel_mount(9) 1989 */ 1990 1991/* A memory allocation which must be freed when we are done */ 1992struct mntaarg { 1993 SLIST_ENTRY(mntaarg) next; 1994}; 1995 1996/* The header for the mount arguments */ 1997struct mntarg { 1998 struct iovec *v; 1999 int len; 2000 int error; 2001 SLIST_HEAD(, mntaarg) list; 2002}; 2003 2004/* 2005 * Add a boolean argument. 2006 * 2007 * flag is the boolean value. 2008 * name must start with "no". 2009 */ 2010struct mntarg * 2011mount_argb(struct mntarg *ma, int flag, const char *name) 2012{ 2013 2014 KASSERT(name[0] == 'n' && name[1] == 'o', 2015 ("mount_argb(...,%s): name must start with 'no'", name)); 2016 2017 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2018} 2019 2020/* 2021 * Add an argument printf style 2022 */ 2023struct mntarg * 2024mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2025{ 2026 va_list ap; 2027 struct mntaarg *maa; 2028 struct sbuf *sb; 2029 int len; 2030 2031 if (ma == NULL) { 2032 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2033 SLIST_INIT(&ma->list); 2034 } 2035 if (ma->error) 2036 return (ma); 2037 2038 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2039 M_MOUNT, M_WAITOK); 2040 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2041 ma->v[ma->len].iov_len = strlen(name) + 1; 2042 ma->len++; 2043 2044 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND); 2045 va_start(ap, fmt); 2046 sbuf_vprintf(sb, fmt, ap); 2047 va_end(ap); 2048 sbuf_finish(sb); 2049 len = sbuf_len(sb) + 1; 2050 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2051 SLIST_INSERT_HEAD(&ma->list, maa, next); 2052 bcopy(sbuf_data(sb), maa + 1, len); 2053 sbuf_delete(sb); 2054 2055 ma->v[ma->len].iov_base = maa + 1; 2056 ma->v[ma->len].iov_len = len; 2057 ma->len++; 2058 2059 return (ma); 2060} 2061 2062/* 2063 * Add an argument which is a userland string. 2064 */ 2065struct mntarg * 2066mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2067{ 2068 struct mntaarg *maa; 2069 char *tbuf; 2070 2071 if (val == NULL) 2072 return (ma); 2073 if (ma == NULL) { 2074 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2075 SLIST_INIT(&ma->list); 2076 } 2077 if (ma->error) 2078 return (ma); 2079 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2080 SLIST_INSERT_HEAD(&ma->list, maa, next); 2081 tbuf = (void *)(maa + 1); 2082 ma->error = copyinstr(val, tbuf, len, NULL); 2083 return (mount_arg(ma, name, tbuf, -1)); 2084} 2085 2086/* 2087 * Plain argument. 2088 * 2089 * If length is -1, use printf. 2090 */ 2091struct mntarg * 2092mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2093{ 2094 2095 if (ma == NULL) { 2096 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2097 SLIST_INIT(&ma->list); 2098 } 2099 if (ma->error) 2100 return (ma); 2101 2102 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2103 M_MOUNT, M_WAITOK); 2104 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2105 ma->v[ma->len].iov_len = strlen(name) + 1; 2106 ma->len++; 2107 2108 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2109 if (len < 0) 2110 ma->v[ma->len].iov_len = strlen(val) + 1; 2111 else 2112 ma->v[ma->len].iov_len = len; 2113 ma->len++; 2114 return (ma); 2115} 2116 2117/* 2118 * Free a mntarg structure 2119 */ 2120static void 2121free_mntarg(struct mntarg *ma) 2122{ 2123 struct mntaarg *maa; 2124 2125 while (!SLIST_EMPTY(&ma->list)) { 2126 maa = SLIST_FIRST(&ma->list); 2127 SLIST_REMOVE_HEAD(&ma->list, next); 2128 free(maa, M_MOUNT); 2129 } 2130 free(ma->v, M_MOUNT); 2131 free(ma, M_MOUNT); 2132} 2133 2134/* 2135 * Mount a filesystem 2136 */ 2137int 2138kernel_mount(struct mntarg *ma, int flags) 2139{ 2140 struct uio auio; 2141 int error; 2142 2143 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2144 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2145 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2146 2147 auio.uio_iov = ma->v; 2148 auio.uio_iovcnt = ma->len; 2149 auio.uio_segflg = UIO_SYSSPACE; 2150 2151 error = ma->error; 2152 if (!error) 2153 error = vfs_donmount(curthread, flags, &auio); 2154 free_mntarg(ma); 2155 return (error); 2156} 2157 2158/* 2159 * A printflike function to mount a filesystem. 2160 */ 2161int 2162kernel_vmount(int flags, ...) 2163{ 2164 struct mntarg *ma = NULL; 2165 va_list ap; 2166 const char *cp; 2167 const void *vp; 2168 int error; 2169 2170 va_start(ap, flags); 2171 for (;;) { 2172 cp = va_arg(ap, const char *); 2173 if (cp == NULL) 2174 break; 2175 vp = va_arg(ap, const void *); 2176 ma = mount_arg(ma, cp, vp, -1); 2177 } 2178 va_end(ap); 2179 2180 error = kernel_mount(ma, flags); 2181 return (error); 2182} 2183