vfs_mount.c revision 168355
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 168355 2007-04-04 09:11:34Z rwatson $"); 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 mp->mnt_gen++; 486 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 487 mp->mnt_vnodecovered = vp; 488 mp->mnt_cred = crdup(td->td_ucred); 489 mp->mnt_stat.f_owner = td->td_ucred->cr_uid; 490 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 491 mp->mnt_iosize_max = DFLTPHYS; 492#ifdef MAC 493 mac_init_mount(mp); 494 mac_create_mount(td->td_ucred, mp); 495#endif 496 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 497 return (mp); 498} 499 500/* 501 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 502 */ 503void 504vfs_mount_destroy(struct mount *mp) 505{ 506 int i; 507 508 MNT_ILOCK(mp); 509 for (i = 0; mp->mnt_ref && i < 3; i++) 510 msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz); 511 /* 512 * This will always cause a 3 second delay in rebooting due to 513 * refs on the root mountpoint that never go away. Most of these 514 * are held by init which never exits. 515 */ 516 if (i == 3 && (!rebooting || bootverbose)) 517 printf("Mount point %s had %d dangling refs\n", 518 mp->mnt_stat.f_mntonname, mp->mnt_ref); 519 if (mp->mnt_holdcnt != 0) { 520 printf("Waiting for mount point to be unheld\n"); 521 while (mp->mnt_holdcnt != 0) { 522 mp->mnt_holdcntwaiters++; 523 msleep(&mp->mnt_holdcnt, MNT_MTX(mp), 524 PZERO, "mntdestroy", 0); 525 mp->mnt_holdcntwaiters--; 526 } 527 printf("mount point unheld\n"); 528 } 529 if (mp->mnt_writeopcount > 0) { 530 printf("Waiting for mount point write ops\n"); 531 while (mp->mnt_writeopcount > 0) { 532 mp->mnt_kern_flag |= MNTK_SUSPEND; 533 msleep(&mp->mnt_writeopcount, 534 MNT_MTX(mp), 535 PZERO, "mntdestroy2", 0); 536 } 537 printf("mount point write ops completed\n"); 538 } 539 if (mp->mnt_secondary_writes > 0) { 540 printf("Waiting for mount point secondary write ops\n"); 541 while (mp->mnt_secondary_writes > 0) { 542 mp->mnt_kern_flag |= MNTK_SUSPEND; 543 msleep(&mp->mnt_secondary_writes, 544 MNT_MTX(mp), 545 PZERO, "mntdestroy3", 0); 546 } 547 printf("mount point secondary write ops completed\n"); 548 } 549 MNT_IUNLOCK(mp); 550 mp->mnt_vfc->vfc_refcount--; 551 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 552 struct vnode *vp; 553 554 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 555 vprint("", vp); 556 panic("unmount: dangling vnode"); 557 } 558 MNT_ILOCK(mp); 559 if (mp->mnt_kern_flag & MNTK_MWAIT) 560 wakeup(mp); 561 if (mp->mnt_writeopcount != 0) 562 panic("vfs_mount_destroy: nonzero writeopcount"); 563 if (mp->mnt_secondary_writes != 0) 564 panic("vfs_mount_destroy: nonzero secondary_writes"); 565 if (mp->mnt_nvnodelistsize != 0) 566 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 567 mp->mnt_writeopcount = -1000; 568 mp->mnt_nvnodelistsize = -1000; 569 mp->mnt_secondary_writes = -1000; 570 MNT_IUNLOCK(mp); 571#ifdef MAC 572 mac_destroy_mount(mp); 573#endif 574 if (mp->mnt_opt != NULL) 575 vfs_freeopts(mp->mnt_opt); 576 crfree(mp->mnt_cred); 577 uma_zfree(mount_zone, mp); 578} 579 580static int 581vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 582{ 583 struct vfsoptlist *optlist; 584 struct vfsopt *opt, *noro_opt; 585 char *fstype, *fspath, *errmsg; 586 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 587 int has_rw, has_noro; 588 589 errmsg = NULL; 590 errmsg_len = 0; 591 errmsg_pos = -1; 592 has_rw = 0; 593 has_noro = 0; 594 595 error = vfs_buildopts(fsoptions, &optlist); 596 if (error) 597 return (error); 598 599 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 600 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 601 602 /* 603 * We need these two options before the others, 604 * and they are mandatory for any filesystem. 605 * Ensure they are NUL terminated as well. 606 */ 607 fstypelen = 0; 608 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 609 if (error || fstype[fstypelen - 1] != '\0') { 610 error = EINVAL; 611 if (errmsg != NULL) 612 strncpy(errmsg, "Invalid fstype", errmsg_len); 613 goto bail; 614 } 615 fspathlen = 0; 616 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 617 if (error || fspath[fspathlen - 1] != '\0') { 618 error = EINVAL; 619 if (errmsg != NULL) 620 strncpy(errmsg, "Invalid fspath", errmsg_len); 621 goto bail; 622 } 623 624 /* 625 * We need to see if we have the "update" option 626 * before we call vfs_domount(), since vfs_domount() has special 627 * logic based on MNT_UPDATE. This is very important 628 * when we want to update the root filesystem. 629 */ 630 TAILQ_FOREACH(opt, optlist, link) { 631 if (strcmp(opt->name, "update") == 0) 632 fsflags |= MNT_UPDATE; 633 else if (strcmp(opt->name, "async") == 0) 634 fsflags |= MNT_ASYNC; 635 else if (strcmp(opt->name, "force") == 0) 636 fsflags |= MNT_FORCE; 637 else if (strcmp(opt->name, "multilabel") == 0) 638 fsflags |= MNT_MULTILABEL; 639 else if (strcmp(opt->name, "noasync") == 0) 640 fsflags &= ~MNT_ASYNC; 641 else if (strcmp(opt->name, "noatime") == 0) 642 fsflags |= MNT_NOATIME; 643 else if (strcmp(opt->name, "noclusterr") == 0) 644 fsflags |= MNT_NOCLUSTERR; 645 else if (strcmp(opt->name, "noclusterw") == 0) 646 fsflags |= MNT_NOCLUSTERW; 647 else if (strcmp(opt->name, "noexec") == 0) 648 fsflags |= MNT_NOEXEC; 649 else if (strcmp(opt->name, "nosuid") == 0) 650 fsflags |= MNT_NOSUID; 651 else if (strcmp(opt->name, "nosymfollow") == 0) 652 fsflags |= MNT_NOSYMFOLLOW; 653 else if (strcmp(opt->name, "noro") == 0) { 654 fsflags &= ~MNT_RDONLY; 655 has_noro = 1; 656 } 657 else if (strcmp(opt->name, "rw") == 0) { 658 fsflags &= ~MNT_RDONLY; 659 has_rw = 1; 660 } 661 else if (strcmp(opt->name, "ro") == 0 || 662 strcmp(opt->name, "rdonly") == 0) 663 fsflags |= MNT_RDONLY; 664 else if (strcmp(opt->name, "snapshot") == 0) 665 fsflags |= MNT_SNAPSHOT; 666 else if (strcmp(opt->name, "suiddir") == 0) 667 fsflags |= MNT_SUIDDIR; 668 else if (strcmp(opt->name, "sync") == 0) 669 fsflags |= MNT_SYNCHRONOUS; 670 else if (strcmp(opt->name, "union") == 0) 671 fsflags |= MNT_UNION; 672 } 673 674 /* 675 * If "rw" was specified as a mount option, and we 676 * are trying to update a mount-point from "ro" to "rw", 677 * we need a mount option "noro", since in vfs_mergeopts(), 678 * "noro" will cancel "ro", but "rw" will not do anything. 679 */ 680 if (has_rw && !has_noro) { 681 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 682 noro_opt->name = strdup("noro", M_MOUNT); 683 noro_opt->value = NULL; 684 noro_opt->len = 0; 685 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 686 } 687 688 /* 689 * Be ultra-paranoid about making sure the type and fspath 690 * variables will fit in our mp buffers, including the 691 * terminating NUL. 692 */ 693 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 694 error = ENAMETOOLONG; 695 goto bail; 696 } 697 698 mtx_lock(&Giant); 699 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 700 mtx_unlock(&Giant); 701bail: 702 /* copyout the errmsg */ 703 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 704 && errmsg_len > 0 && errmsg != NULL) { 705 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 706 bcopy(errmsg, 707 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 708 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 709 } else { 710 copyout(errmsg, 711 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 712 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 713 } 714 } 715 716 if (error != 0) 717 vfs_freeopts(optlist); 718 return (error); 719} 720 721/* 722 * Old mount API. 723 */ 724#ifndef _SYS_SYSPROTO_H_ 725struct mount_args { 726 char *type; 727 char *path; 728 int flags; 729 caddr_t data; 730}; 731#endif 732/* ARGSUSED */ 733int 734mount(td, uap) 735 struct thread *td; 736 struct mount_args /* { 737 char *type; 738 char *path; 739 int flags; 740 caddr_t data; 741 } */ *uap; 742{ 743 char *fstype; 744 struct vfsconf *vfsp = NULL; 745 struct mntarg *ma = NULL; 746 int error; 747 748 AUDIT_ARG(fflags, uap->flags); 749 750 /* Kick out MNT_ROOTFS early as it is legal internally */ 751 uap->flags &= ~MNT_ROOTFS; 752 753 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 754 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 755 if (error) { 756 free(fstype, M_TEMP); 757 return (error); 758 } 759 760 AUDIT_ARG(text, fstype); 761 mtx_lock(&Giant); 762 vfsp = vfs_byname_kld(fstype, td, &error); 763 free(fstype, M_TEMP); 764 if (vfsp == NULL) { 765 mtx_unlock(&Giant); 766 return (ENOENT); 767 } 768 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 769 mtx_unlock(&Giant); 770 return (EOPNOTSUPP); 771 } 772 773 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 774 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 775 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 776 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 777 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 778 779 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 780 mtx_unlock(&Giant); 781 return (error); 782} 783 784 785/* 786 * vfs_domount(): actually attempt a filesystem mount. 787 */ 788static int 789vfs_domount( 790 struct thread *td, /* Calling thread. */ 791 const char *fstype, /* Filesystem type. */ 792 char *fspath, /* Mount path. */ 793 int fsflags, /* Flags common to all filesystems. */ 794 void *fsdata /* Options local to the filesystem. */ 795 ) 796{ 797 struct vnode *vp; 798 struct mount *mp; 799 struct vfsconf *vfsp; 800 struct export_args export; 801 int error, flag = 0; 802 struct vattr va; 803 struct nameidata nd; 804 805 mtx_assert(&Giant, MA_OWNED); 806 /* 807 * Be ultra-paranoid about making sure the type and fspath 808 * variables will fit in our mp buffers, including the 809 * terminating NUL. 810 */ 811 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 812 return (ENAMETOOLONG); 813 814 if (jailed(td->td_ucred) || usermount == 0) { 815 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 816 return (error); 817 } 818 819 /* 820 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 821 */ 822 if (fsflags & MNT_EXPORTED) { 823 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 824 if (error) 825 return (error); 826 } 827 if (fsflags & MNT_SUIDDIR) { 828 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 829 if (error) 830 return (error); 831 } 832 /* 833 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 834 */ 835 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 836 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 837 fsflags |= MNT_NOSUID | MNT_USER; 838 } 839 840 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 841 vfsp = NULL; 842 if ((fsflags & MNT_UPDATE) == 0) { 843 /* Don't try to load KLDs if we're mounting the root. */ 844 if (fsflags & MNT_ROOTFS) 845 vfsp = vfs_byname(fstype); 846 else 847 vfsp = vfs_byname_kld(fstype, td, &error); 848 if (vfsp == NULL) 849 return (ENODEV); 850 } 851 /* 852 * Get vnode to be covered 853 */ 854 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 855 fspath, td); 856 if ((error = namei(&nd)) != 0) 857 return (error); 858 NDFREE(&nd, NDF_ONLY_PNBUF); 859 vp = nd.ni_vp; 860 if (fsflags & MNT_UPDATE) { 861 if ((vp->v_vflag & VV_ROOT) == 0) { 862 vput(vp); 863 return (EINVAL); 864 } 865 mp = vp->v_mount; 866 MNT_ILOCK(mp); 867 flag = mp->mnt_flag; 868 /* 869 * We only allow the filesystem to be reloaded if it 870 * is currently mounted read-only. 871 */ 872 if ((fsflags & MNT_RELOAD) && 873 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 874 MNT_IUNLOCK(mp); 875 vput(vp); 876 return (EOPNOTSUPP); /* Needs translation */ 877 } 878 MNT_IUNLOCK(mp); 879 /* 880 * Only privileged root, or (if MNT_USER is set) the user that 881 * did the original mount is permitted to update it. 882 */ 883 error = vfs_suser(mp, td); 884 if (error) { 885 vput(vp); 886 return (error); 887 } 888 if (vfs_busy(mp, LK_NOWAIT, 0, td)) { 889 vput(vp); 890 return (EBUSY); 891 } 892 VI_LOCK(vp); 893 if ((vp->v_iflag & VI_MOUNT) != 0 || 894 vp->v_mountedhere != NULL) { 895 VI_UNLOCK(vp); 896 vfs_unbusy(mp, td); 897 vput(vp); 898 return (EBUSY); 899 } 900 vp->v_iflag |= VI_MOUNT; 901 VI_UNLOCK(vp); 902 MNT_ILOCK(mp); 903 mp->mnt_flag |= fsflags & 904 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 905 MNT_IUNLOCK(mp); 906 VOP_UNLOCK(vp, 0, td); 907 mp->mnt_optnew = fsdata; 908 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 909 } else { 910 /* 911 * If the user is not root, ensure that they own the directory 912 * onto which we are attempting to mount. 913 */ 914 error = VOP_GETATTR(vp, &va, td->td_ucred, td); 915 if (error) { 916 vput(vp); 917 return (error); 918 } 919 if (va.va_uid != td->td_ucred->cr_uid) { 920 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 921 SUSER_ALLOWJAIL); 922 if (error) { 923 vput(vp); 924 return (error); 925 } 926 } 927 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 928 if (error != 0) { 929 vput(vp); 930 return (error); 931 } 932 if (vp->v_type != VDIR) { 933 vput(vp); 934 return (ENOTDIR); 935 } 936 VI_LOCK(vp); 937 if ((vp->v_iflag & VI_MOUNT) != 0 || 938 vp->v_mountedhere != NULL) { 939 VI_UNLOCK(vp); 940 vput(vp); 941 return (EBUSY); 942 } 943 vp->v_iflag |= VI_MOUNT; 944 VI_UNLOCK(vp); 945 946 /* 947 * Allocate and initialize the filesystem. 948 */ 949 mp = vfs_mount_alloc(vp, vfsp, fspath, td); 950 VOP_UNLOCK(vp, 0, td); 951 952 /* XXXMAC: pass to vfs_mount_alloc? */ 953 mp->mnt_optnew = fsdata; 954 } 955 956 /* 957 * Set the mount level flags. 958 */ 959 MNT_ILOCK(mp); 960 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 961 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 962 MNT_RDONLY)); 963 if ((mp->mnt_flag & MNT_ASYNC) == 0) 964 mp->mnt_kern_flag &= ~MNTK_ASYNC; 965 MNT_IUNLOCK(mp); 966 /* 967 * Mount the filesystem. 968 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 969 * get. No freeing of cn_pnbuf. 970 */ 971 error = VFS_MOUNT(mp, td); 972 973 /* 974 * Process the export option only if we are 975 * updating mount options. 976 */ 977 if (!error && (fsflags & MNT_UPDATE)) { 978 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 979 sizeof(export)) == 0) 980 error = vfs_export(mp, &export); 981 } 982 983 if (!error) { 984 if (mp->mnt_opt != NULL) 985 vfs_freeopts(mp->mnt_opt); 986 mp->mnt_opt = mp->mnt_optnew; 987 (void)VFS_STATFS(mp, &mp->mnt_stat, td); 988 } 989 /* 990 * Prevent external consumers of mount options from reading 991 * mnt_optnew. 992 */ 993 mp->mnt_optnew = NULL; 994 if (mp->mnt_flag & MNT_UPDATE) { 995 MNT_ILOCK(mp); 996 if (error) 997 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 998 (flag & ~MNT_QUOTA); 999 else 1000 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1001 MNT_FORCE | MNT_SNAPSHOT); 1002 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1003 mp->mnt_kern_flag |= MNTK_ASYNC; 1004 else 1005 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1006 MNT_IUNLOCK(mp); 1007 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1008 if (mp->mnt_syncer == NULL) 1009 error = vfs_allocate_syncvnode(mp); 1010 } else { 1011 if (mp->mnt_syncer != NULL) 1012 vrele(mp->mnt_syncer); 1013 mp->mnt_syncer = NULL; 1014 } 1015 vfs_unbusy(mp, td); 1016 VI_LOCK(vp); 1017 vp->v_iflag &= ~VI_MOUNT; 1018 VI_UNLOCK(vp); 1019 vrele(vp); 1020 return (error); 1021 } 1022 MNT_ILOCK(mp); 1023 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1024 mp->mnt_kern_flag |= MNTK_ASYNC; 1025 else 1026 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1027 MNT_IUNLOCK(mp); 1028 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1029 /* 1030 * Put the new filesystem on the mount list after root. 1031 */ 1032 cache_purge(vp); 1033 if (!error) { 1034 struct vnode *newdp; 1035 1036 VI_LOCK(vp); 1037 vp->v_iflag &= ~VI_MOUNT; 1038 VI_UNLOCK(vp); 1039 vp->v_mountedhere = mp; 1040 mtx_lock(&mountlist_mtx); 1041 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1042 mtx_unlock(&mountlist_mtx); 1043 vfs_event_signal(NULL, VQ_MOUNT, 0); 1044 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 1045 panic("mount: lost mount"); 1046 mountcheckdirs(vp, newdp); 1047 vput(newdp); 1048 VOP_UNLOCK(vp, 0, td); 1049 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1050 error = vfs_allocate_syncvnode(mp); 1051 vfs_unbusy(mp, td); 1052 if (error) 1053 vrele(vp); 1054 } else { 1055 VI_LOCK(vp); 1056 vp->v_iflag &= ~VI_MOUNT; 1057 VI_UNLOCK(vp); 1058 vfs_unbusy(mp, td); 1059 vfs_mount_destroy(mp); 1060 vput(vp); 1061 } 1062 return (error); 1063} 1064 1065/* 1066 * Unmount a filesystem. 1067 * 1068 * Note: unmount takes a path to the vnode mounted on as argument, not 1069 * special file (as before). 1070 */ 1071#ifndef _SYS_SYSPROTO_H_ 1072struct unmount_args { 1073 char *path; 1074 int flags; 1075}; 1076#endif 1077/* ARGSUSED */ 1078int 1079unmount(td, uap) 1080 struct thread *td; 1081 register struct unmount_args /* { 1082 char *path; 1083 int flags; 1084 } */ *uap; 1085{ 1086 struct mount *mp; 1087 char *pathbuf; 1088 int error, id0, id1; 1089 1090 if (jailed(td->td_ucred) || usermount == 0) { 1091 error = priv_check(td, PRIV_VFS_UNMOUNT); 1092 if (error) 1093 return (error); 1094 } 1095 1096 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1097 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1098 if (error) { 1099 free(pathbuf, M_TEMP); 1100 return (error); 1101 } 1102 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1103 mtx_lock(&Giant); 1104 if (uap->flags & MNT_BYFSID) { 1105 /* Decode the filesystem ID. */ 1106 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1107 mtx_unlock(&Giant); 1108 free(pathbuf, M_TEMP); 1109 return (EINVAL); 1110 } 1111 1112 mtx_lock(&mountlist_mtx); 1113 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1114 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1115 mp->mnt_stat.f_fsid.val[1] == id1) 1116 break; 1117 } 1118 mtx_unlock(&mountlist_mtx); 1119 } else { 1120 mtx_lock(&mountlist_mtx); 1121 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1122 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1123 break; 1124 } 1125 mtx_unlock(&mountlist_mtx); 1126 } 1127 free(pathbuf, M_TEMP); 1128 if (mp == NULL) { 1129 /* 1130 * Previously we returned ENOENT for a nonexistent path and 1131 * EINVAL for a non-mountpoint. We cannot tell these apart 1132 * now, so in the !MNT_BYFSID case return the more likely 1133 * EINVAL for compatibility. 1134 */ 1135 mtx_unlock(&Giant); 1136 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1137 } 1138 1139 /* 1140 * Don't allow unmounting the root filesystem. 1141 */ 1142 if (mp->mnt_flag & MNT_ROOTFS) { 1143 mtx_unlock(&Giant); 1144 return (EINVAL); 1145 } 1146 error = dounmount(mp, uap->flags, td); 1147 mtx_unlock(&Giant); 1148 return (error); 1149} 1150 1151/* 1152 * Do the actual filesystem unmount. 1153 */ 1154int 1155dounmount(mp, flags, td) 1156 struct mount *mp; 1157 int flags; 1158 struct thread *td; 1159{ 1160 struct vnode *coveredvp, *fsrootvp; 1161 int error; 1162 int async_flag; 1163 int mnt_gen_r; 1164 1165 mtx_assert(&Giant, MA_OWNED); 1166 1167 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1168 mnt_gen_r = mp->mnt_gen; 1169 VI_LOCK(coveredvp); 1170 vholdl(coveredvp); 1171 error = vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK, td); 1172 vdrop(coveredvp); 1173 /* 1174 * Check for mp being unmounted while waiting for the 1175 * covered vnode lock. 1176 */ 1177 if (error) 1178 return (error); 1179 if (coveredvp->v_mountedhere != mp || 1180 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1181 VOP_UNLOCK(coveredvp, 0, td); 1182 return (EBUSY); 1183 } 1184 } 1185 /* 1186 * Only privileged root, or (if MNT_USER is set) the user that did the 1187 * original mount is permitted to unmount this filesystem. 1188 */ 1189 error = vfs_suser(mp, td); 1190 if (error) { 1191 if (coveredvp) 1192 VOP_UNLOCK(coveredvp, 0, td); 1193 return (error); 1194 } 1195 1196 MNT_ILOCK(mp); 1197 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1198 MNT_IUNLOCK(mp); 1199 if (coveredvp) 1200 VOP_UNLOCK(coveredvp, 0, td); 1201 return (EBUSY); 1202 } 1203 mp->mnt_kern_flag |= MNTK_UNMOUNT; 1204 /* Allow filesystems to detect that a forced unmount is in progress. */ 1205 if (flags & MNT_FORCE) 1206 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1207 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 1208 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td); 1209 if (error) { 1210 MNT_ILOCK(mp); 1211 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1212 if (mp->mnt_kern_flag & MNTK_MWAIT) 1213 wakeup(mp); 1214 MNT_IUNLOCK(mp); 1215 if (coveredvp) 1216 VOP_UNLOCK(coveredvp, 0, td); 1217 return (error); 1218 } 1219 vn_start_write(NULL, &mp, V_WAIT); 1220 1221 if (mp->mnt_flag & MNT_EXPUBLIC) 1222 vfs_setpublicfs(NULL, NULL, NULL); 1223 1224 vfs_msync(mp, MNT_WAIT); 1225 MNT_ILOCK(mp); 1226 async_flag = mp->mnt_flag & MNT_ASYNC; 1227 mp->mnt_flag &= ~MNT_ASYNC; 1228 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1229 MNT_IUNLOCK(mp); 1230 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1231 if (mp->mnt_syncer != NULL) 1232 vrele(mp->mnt_syncer); 1233 /* 1234 * For forced unmounts, move process cdir/rdir refs on the fs root 1235 * vnode to the covered vnode. For non-forced unmounts we want 1236 * such references to cause an EBUSY error. 1237 */ 1238 if ((flags & MNT_FORCE) && 1239 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1240 if (mp->mnt_vnodecovered != NULL) 1241 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1242 if (fsrootvp == rootvnode) { 1243 vrele(rootvnode); 1244 rootvnode = NULL; 1245 } 1246 vput(fsrootvp); 1247 } 1248 if (((mp->mnt_flag & MNT_RDONLY) || 1249 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1250 (flags & MNT_FORCE)) { 1251 error = VFS_UNMOUNT(mp, flags, td); 1252 } 1253 vn_finished_write(mp); 1254 if (error) { 1255 /* Undo cdir/rdir and rootvnode changes made above. */ 1256 if ((flags & MNT_FORCE) && 1257 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1258 if (mp->mnt_vnodecovered != NULL) 1259 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1260 if (rootvnode == NULL) { 1261 rootvnode = fsrootvp; 1262 vref(rootvnode); 1263 } 1264 vput(fsrootvp); 1265 } 1266 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) 1267 (void) vfs_allocate_syncvnode(mp); 1268 MNT_ILOCK(mp); 1269 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1270 mp->mnt_flag |= async_flag; 1271 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1272 mp->mnt_kern_flag |= MNTK_ASYNC; 1273 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1274 if (mp->mnt_kern_flag & MNTK_MWAIT) 1275 wakeup(mp); 1276 MNT_IUNLOCK(mp); 1277 if (coveredvp) 1278 VOP_UNLOCK(coveredvp, 0, td); 1279 return (error); 1280 } 1281 mtx_lock(&mountlist_mtx); 1282 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1283 mtx_unlock(&mountlist_mtx); 1284 if (coveredvp != NULL) { 1285 coveredvp->v_mountedhere = NULL; 1286 vput(coveredvp); 1287 } 1288 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1289 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1290 vfs_mount_destroy(mp); 1291 return (0); 1292} 1293 1294/* 1295 * --------------------------------------------------------------------- 1296 * Mounting of root filesystem 1297 * 1298 */ 1299 1300struct root_hold_token { 1301 const char *who; 1302 LIST_ENTRY(root_hold_token) list; 1303}; 1304 1305static LIST_HEAD(, root_hold_token) root_holds = 1306 LIST_HEAD_INITIALIZER(&root_holds); 1307 1308static int root_mount_complete = 0; 1309 1310/* 1311 * Hold root mount. 1312 */ 1313struct root_hold_token * 1314root_mount_hold(const char *identifier) 1315{ 1316 struct root_hold_token *h; 1317 1318 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1319 h->who = identifier; 1320 mtx_lock(&mountlist_mtx); 1321 LIST_INSERT_HEAD(&root_holds, h, list); 1322 mtx_unlock(&mountlist_mtx); 1323 return (h); 1324} 1325 1326/* 1327 * Release root mount. 1328 */ 1329void 1330root_mount_rel(struct root_hold_token *h) 1331{ 1332 1333 mtx_lock(&mountlist_mtx); 1334 LIST_REMOVE(h, list); 1335 wakeup(&root_holds); 1336 mtx_unlock(&mountlist_mtx); 1337 free(h, M_DEVBUF); 1338} 1339 1340/* 1341 * Wait for all subsystems to release root mount. 1342 */ 1343static void 1344root_mount_prepare(void) 1345{ 1346 struct root_hold_token *h; 1347 1348 for (;;) { 1349 DROP_GIANT(); 1350 g_waitidle(); 1351 PICKUP_GIANT(); 1352 mtx_lock(&mountlist_mtx); 1353 if (LIST_EMPTY(&root_holds)) { 1354 mtx_unlock(&mountlist_mtx); 1355 break; 1356 } 1357 printf("Root mount waiting for:"); 1358 LIST_FOREACH(h, &root_holds, list) 1359 printf(" %s", h->who); 1360 printf("\n"); 1361 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1362 hz); 1363 } 1364} 1365 1366/* 1367 * Root was mounted, share the good news. 1368 */ 1369static void 1370root_mount_done(void) 1371{ 1372 1373 mtx_lock(&mountlist_mtx); 1374 root_mount_complete = 1; 1375 wakeup(&root_mount_complete); 1376 mtx_unlock(&mountlist_mtx); 1377} 1378 1379/* 1380 * Wait until root is mounted. 1381 */ 1382void 1383root_mount_wait(void) 1384{ 1385 1386 /* 1387 * Panic on an obvious deadlock - the function can't be called from 1388 * a thread which is doing the whole SYSINIT stuff. 1389 */ 1390 KASSERT(curthread->td_proc->p_pid != 0, 1391 ("root_mount_wait: cannot be called from the swapper thread")); 1392 mtx_lock(&mountlist_mtx); 1393 while (!root_mount_complete) { 1394 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1395 hz); 1396 } 1397 mtx_unlock(&mountlist_mtx); 1398} 1399 1400static void 1401set_rootvnode(struct thread *td) 1402{ 1403 struct proc *p; 1404 1405 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1406 panic("Cannot find root vnode"); 1407 1408 p = td->td_proc; 1409 FILEDESC_SLOCK(p->p_fd); 1410 1411 if (p->p_fd->fd_cdir != NULL) 1412 vrele(p->p_fd->fd_cdir); 1413 p->p_fd->fd_cdir = rootvnode; 1414 VREF(rootvnode); 1415 1416 if (p->p_fd->fd_rdir != NULL) 1417 vrele(p->p_fd->fd_rdir); 1418 p->p_fd->fd_rdir = rootvnode; 1419 VREF(rootvnode); 1420 1421 FILEDESC_SUNLOCK(p->p_fd); 1422 1423 VOP_UNLOCK(rootvnode, 0, td); 1424} 1425 1426/* 1427 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1428 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1429 */ 1430 1431static void 1432devfs_first(void) 1433{ 1434 struct thread *td = curthread; 1435 struct vfsoptlist *opts; 1436 struct vfsconf *vfsp; 1437 struct mount *mp = NULL; 1438 int error; 1439 1440 vfsp = vfs_byname("devfs"); 1441 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1442 if (vfsp == NULL) 1443 return; 1444 1445 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td); 1446 1447 error = VFS_MOUNT(mp, td); 1448 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1449 if (error) 1450 return; 1451 1452 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1453 TAILQ_INIT(opts); 1454 mp->mnt_opt = opts; 1455 1456 mtx_lock(&mountlist_mtx); 1457 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1458 mtx_unlock(&mountlist_mtx); 1459 1460 set_rootvnode(td); 1461 1462 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1463 if (error) 1464 printf("kern_symlink /dev -> / returns %d\n", error); 1465} 1466 1467/* 1468 * Surgically move our devfs to be mounted on /dev. 1469 */ 1470 1471static void 1472devfs_fixup(struct thread *td) 1473{ 1474 struct nameidata nd; 1475 int error; 1476 struct vnode *vp, *dvp; 1477 struct mount *mp; 1478 1479 /* Remove our devfs mount from the mountlist and purge the cache */ 1480 mtx_lock(&mountlist_mtx); 1481 mp = TAILQ_FIRST(&mountlist); 1482 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1483 mtx_unlock(&mountlist_mtx); 1484 cache_purgevfs(mp); 1485 1486 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1487 VI_LOCK(dvp); 1488 dvp->v_iflag &= ~VI_MOUNT; 1489 dvp->v_mountedhere = NULL; 1490 VI_UNLOCK(dvp); 1491 1492 /* Set up the real rootvnode, and purge the cache */ 1493 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1494 set_rootvnode(td); 1495 cache_purgevfs(rootvnode->v_mount); 1496 1497 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1498 error = namei(&nd); 1499 if (error) { 1500 printf("Lookup of /dev for devfs, error: %d\n", error); 1501 return; 1502 } 1503 NDFREE(&nd, NDF_ONLY_PNBUF); 1504 vp = nd.ni_vp; 1505 if (vp->v_type != VDIR) { 1506 vput(vp); 1507 } 1508 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 1509 if (error) { 1510 vput(vp); 1511 } 1512 cache_purge(vp); 1513 mp->mnt_vnodecovered = vp; 1514 vp->v_mountedhere = mp; 1515 mtx_lock(&mountlist_mtx); 1516 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1517 mtx_unlock(&mountlist_mtx); 1518 VOP_UNLOCK(vp, 0, td); 1519 vput(dvp); 1520 vfs_unbusy(mp, td); 1521 1522 /* Unlink the no longer needed /dev/dev -> / symlink */ 1523 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1524} 1525 1526/* 1527 * Report errors during filesystem mounting. 1528 */ 1529void 1530vfs_mount_error(struct mount *mp, const char *fmt, ...) 1531{ 1532 struct vfsoptlist *moptlist = mp->mnt_optnew; 1533 va_list ap; 1534 int error, len; 1535 char *errmsg; 1536 1537 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1538 if (error || errmsg == NULL || len <= 0) 1539 return; 1540 1541 va_start(ap, fmt); 1542 vsnprintf(errmsg, (size_t)len, fmt, ap); 1543 va_end(ap); 1544} 1545 1546/* 1547 * Find and mount the root filesystem 1548 */ 1549void 1550vfs_mountroot(void) 1551{ 1552 char *cp; 1553 int error, i, asked = 0; 1554 1555 root_mount_prepare(); 1556 1557 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1558 NULL, NULL, mount_init, mount_fini, 1559 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1560 devfs_first(); 1561 1562 /* 1563 * We are booted with instructions to prompt for the root filesystem. 1564 */ 1565 if (boothowto & RB_ASKNAME) { 1566 if (!vfs_mountroot_ask()) 1567 goto mounted; 1568 asked = 1; 1569 } 1570 1571 /* 1572 * The root filesystem information is compiled in, and we are 1573 * booted with instructions to use it. 1574 */ 1575 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1576 if (!vfs_mountroot_try(ctrootdevname)) 1577 goto mounted; 1578 ctrootdevname = NULL; 1579 } 1580 1581 /* 1582 * We've been given the generic "use CDROM as root" flag. This is 1583 * necessary because one media may be used in many different 1584 * devices, so we need to search for them. 1585 */ 1586 if (boothowto & RB_CDROM) { 1587 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1588 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1589 goto mounted; 1590 } 1591 } 1592 1593 /* 1594 * Try to use the value read by the loader from /etc/fstab, or 1595 * supplied via some other means. This is the preferred 1596 * mechanism. 1597 */ 1598 cp = getenv("vfs.root.mountfrom"); 1599 if (cp != NULL) { 1600 error = vfs_mountroot_try(cp); 1601 freeenv(cp); 1602 if (!error) 1603 goto mounted; 1604 } 1605 1606 /* 1607 * Try values that may have been computed by code during boot 1608 */ 1609 if (!vfs_mountroot_try(rootdevnames[0])) 1610 goto mounted; 1611 if (!vfs_mountroot_try(rootdevnames[1])) 1612 goto mounted; 1613 1614 /* 1615 * If we (still) have a compiled-in default, try it. 1616 */ 1617 if (ctrootdevname != NULL) 1618 if (!vfs_mountroot_try(ctrootdevname)) 1619 goto mounted; 1620 /* 1621 * Everything so far has failed, prompt on the console if we haven't 1622 * already tried that. 1623 */ 1624 if (!asked) 1625 if (!vfs_mountroot_ask()) 1626 goto mounted; 1627 1628 panic("Root mount failed, startup aborted."); 1629 1630mounted: 1631 root_mount_done(); 1632} 1633 1634/* 1635 * Mount (mountfrom) as the root filesystem. 1636 */ 1637static int 1638vfs_mountroot_try(const char *mountfrom) 1639{ 1640 struct mount *mp; 1641 char *vfsname, *path; 1642 time_t timebase; 1643 int error; 1644 char patt[32]; 1645 1646 vfsname = NULL; 1647 path = NULL; 1648 mp = NULL; 1649 error = EINVAL; 1650 1651 if (mountfrom == NULL) 1652 return (error); /* don't complain */ 1653 printf("Trying to mount root from %s\n", mountfrom); 1654 1655 /* parse vfs name and path */ 1656 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1657 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1658 vfsname[0] = path[0] = 0; 1659 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1660 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1661 goto out; 1662 1663 if (path[0] == '\0') 1664 strcpy(path, ROOTNAME); 1665 1666 error = kernel_vmount( 1667 MNT_RDONLY | MNT_ROOTFS, 1668 "fstype", vfsname, 1669 "fspath", "/", 1670 "from", path, 1671 NULL); 1672 if (error == 0) { 1673 /* 1674 * We mount devfs prior to mounting the / FS, so the first 1675 * entry will typically be devfs. 1676 */ 1677 mp = TAILQ_FIRST(&mountlist); 1678 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1679 1680 /* 1681 * Iterate over all currently mounted file systems and use 1682 * the time stamp found to check and/or initialize the RTC. 1683 * Typically devfs has no time stamp and the only other FS 1684 * is the actual / FS. 1685 * Call inittodr() only once and pass it the largest of the 1686 * timestamps we encounter. 1687 */ 1688 timebase = 0; 1689 do { 1690 if (mp->mnt_time > timebase) 1691 timebase = mp->mnt_time; 1692 mp = TAILQ_NEXT(mp, mnt_list); 1693 } while (mp != NULL); 1694 inittodr(timebase); 1695 1696 devfs_fixup(curthread); 1697 } 1698out: 1699 free(path, M_MOUNT); 1700 free(vfsname, M_MOUNT); 1701 return (error); 1702} 1703 1704/* 1705 * --------------------------------------------------------------------- 1706 * Interactive root filesystem selection code. 1707 */ 1708 1709static int 1710vfs_mountroot_ask(void) 1711{ 1712 char name[128]; 1713 1714 for(;;) { 1715 printf("\nManual root filesystem specification:\n"); 1716 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1717#if defined(__amd64__) || defined(__i386__) || defined(__ia64__) 1718 printf(" eg. ufs:da0s1a\n"); 1719#else 1720 printf(" eg. ufs:/dev/da0a\n"); 1721#endif 1722 printf(" ? List valid disk boot devices\n"); 1723 printf(" <empty line> Abort manual input\n"); 1724 printf("\nmountroot> "); 1725 gets(name, sizeof(name), 1); 1726 if (name[0] == '\0') 1727 return (1); 1728 if (name[0] == '?') { 1729 printf("\nList of GEOM managed disk devices:\n "); 1730 g_dev_print(); 1731 continue; 1732 } 1733 if (!vfs_mountroot_try(name)) 1734 return (0); 1735 } 1736} 1737 1738/* 1739 * --------------------------------------------------------------------- 1740 * Functions for querying mount options/arguments from filesystems. 1741 */ 1742 1743/* 1744 * Check that no unknown options are given 1745 */ 1746int 1747vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1748{ 1749 struct vfsopt *opt; 1750 const char **t, *p; 1751 1752 TAILQ_FOREACH(opt, opts, link) { 1753 p = opt->name; 1754 if (p[0] == 'n' && p[1] == 'o') 1755 p += 2; 1756 for(t = global_opts; *t != NULL; t++) 1757 if (!strcmp(*t, p)) 1758 break; 1759 if (*t != NULL) 1760 continue; 1761 for(t = legal; *t != NULL; t++) 1762 if (!strcmp(*t, p)) 1763 break; 1764 if (*t != NULL) 1765 continue; 1766 printf("mount option <%s> is unknown\n", p); 1767 return (EINVAL); 1768 } 1769 return (0); 1770} 1771 1772/* 1773 * Get a mount option by its name. 1774 * 1775 * Return 0 if the option was found, ENOENT otherwise. 1776 * If len is non-NULL it will be filled with the length 1777 * of the option. If buf is non-NULL, it will be filled 1778 * with the address of the option. 1779 */ 1780int 1781vfs_getopt(opts, name, buf, len) 1782 struct vfsoptlist *opts; 1783 const char *name; 1784 void **buf; 1785 int *len; 1786{ 1787 struct vfsopt *opt; 1788 1789 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1790 1791 TAILQ_FOREACH(opt, opts, link) { 1792 if (strcmp(name, opt->name) == 0) { 1793 if (len != NULL) 1794 *len = opt->len; 1795 if (buf != NULL) 1796 *buf = opt->value; 1797 return (0); 1798 } 1799 } 1800 return (ENOENT); 1801} 1802 1803static int 1804vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1805{ 1806 struct vfsopt *opt; 1807 int i; 1808 1809 if (opts == NULL) 1810 return (-1); 1811 1812 i = 0; 1813 TAILQ_FOREACH(opt, opts, link) { 1814 if (strcmp(name, opt->name) == 0) 1815 return (i); 1816 ++i; 1817 } 1818 return (-1); 1819} 1820 1821char * 1822vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1823{ 1824 struct vfsopt *opt; 1825 1826 *error = 0; 1827 TAILQ_FOREACH(opt, opts, link) { 1828 if (strcmp(name, opt->name) != 0) 1829 continue; 1830 if (((char *)opt->value)[opt->len - 1] != '\0') { 1831 *error = EINVAL; 1832 return (NULL); 1833 } 1834 return (opt->value); 1835 } 1836 *error = ENOENT; 1837 return (NULL); 1838} 1839 1840int 1841vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1842{ 1843 struct vfsopt *opt; 1844 1845 TAILQ_FOREACH(opt, opts, link) { 1846 if (strcmp(name, opt->name) == 0) { 1847 if (w != NULL) 1848 *w |= val; 1849 return (1); 1850 } 1851 } 1852 if (w != NULL) 1853 *w &= ~val; 1854 return (0); 1855} 1856 1857int 1858vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1859{ 1860 va_list ap; 1861 struct vfsopt *opt; 1862 int ret; 1863 1864 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1865 1866 TAILQ_FOREACH(opt, opts, link) { 1867 if (strcmp(name, opt->name) != 0) 1868 continue; 1869 if (((char *)opt->value)[opt->len - 1] != '\0') 1870 return (0); 1871 va_start(ap, fmt); 1872 ret = vsscanf(opt->value, fmt, ap); 1873 va_end(ap); 1874 return (ret); 1875 } 1876 return (0); 1877} 1878 1879/* 1880 * Find and copy a mount option. 1881 * 1882 * The size of the buffer has to be specified 1883 * in len, if it is not the same length as the 1884 * mount option, EINVAL is returned. 1885 * Returns ENOENT if the option is not found. 1886 */ 1887int 1888vfs_copyopt(opts, name, dest, len) 1889 struct vfsoptlist *opts; 1890 const char *name; 1891 void *dest; 1892 int len; 1893{ 1894 struct vfsopt *opt; 1895 1896 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1897 1898 TAILQ_FOREACH(opt, opts, link) { 1899 if (strcmp(name, opt->name) == 0) { 1900 if (len != opt->len) 1901 return (EINVAL); 1902 bcopy(opt->value, dest, opt->len); 1903 return (0); 1904 } 1905 } 1906 return (ENOENT); 1907} 1908 1909/* 1910 * This is a helper function for filesystems to traverse their 1911 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1912 */ 1913 1914struct vnode * 1915__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 1916{ 1917 struct vnode *vp; 1918 1919 mtx_assert(MNT_MTX(mp), MA_OWNED); 1920 1921 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1922 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 1923 while (vp != NULL && vp->v_type == VMARKER) 1924 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1925 1926 /* Check if we are done */ 1927 if (vp == NULL) { 1928 __mnt_vnode_markerfree(mvp, mp); 1929 return (NULL); 1930 } 1931 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1932 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1933 return (vp); 1934} 1935 1936struct vnode * 1937__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 1938{ 1939 struct vnode *vp; 1940 1941 mtx_assert(MNT_MTX(mp), MA_OWNED); 1942 1943 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1944 while (vp != NULL && vp->v_type == VMARKER) 1945 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1946 1947 /* Check if we are done */ 1948 if (vp == NULL) { 1949 *mvp = NULL; 1950 return (NULL); 1951 } 1952 mp->mnt_holdcnt++; 1953 MNT_IUNLOCK(mp); 1954 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 1955 M_VNODE_MARKER, 1956 M_WAITOK | M_ZERO); 1957 MNT_ILOCK(mp); 1958 (*mvp)->v_type = VMARKER; 1959 1960 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1961 while (vp != NULL && vp->v_type == VMARKER) 1962 vp = TAILQ_NEXT(vp, v_nmntvnodes); 1963 1964 /* Check if we are done */ 1965 if (vp == NULL) { 1966 MNT_IUNLOCK(mp); 1967 free(*mvp, M_VNODE_MARKER); 1968 MNT_ILOCK(mp); 1969 *mvp = NULL; 1970 mp->mnt_holdcnt--; 1971 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 1972 wakeup(&mp->mnt_holdcnt); 1973 return (NULL); 1974 } 1975 mp->mnt_markercnt++; 1976 (*mvp)->v_mount = mp; 1977 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 1978 return (vp); 1979} 1980 1981 1982void 1983__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 1984{ 1985 1986 if (*mvp == NULL) 1987 return; 1988 1989 mtx_assert(MNT_MTX(mp), MA_OWNED); 1990 1991 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 1992 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 1993 MNT_IUNLOCK(mp); 1994 free(*mvp, M_VNODE_MARKER); 1995 MNT_ILOCK(mp); 1996 *mvp = NULL; 1997 1998 mp->mnt_markercnt--; 1999 mp->mnt_holdcnt--; 2000 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2001 wakeup(&mp->mnt_holdcnt); 2002} 2003 2004 2005int 2006__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 2007{ 2008 int error; 2009 2010 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 2011 if (sbp != &mp->mnt_stat) 2012 *sbp = mp->mnt_stat; 2013 return (error); 2014} 2015 2016void 2017vfs_mountedfrom(struct mount *mp, const char *from) 2018{ 2019 2020 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2021 strlcpy(mp->mnt_stat.f_mntfromname, from, 2022 sizeof mp->mnt_stat.f_mntfromname); 2023} 2024 2025/* 2026 * --------------------------------------------------------------------- 2027 * This is the api for building mount args and mounting filesystems from 2028 * inside the kernel. 2029 * 2030 * The API works by accumulation of individual args. First error is 2031 * latched. 2032 * 2033 * XXX: should be documented in new manpage kernel_mount(9) 2034 */ 2035 2036/* A memory allocation which must be freed when we are done */ 2037struct mntaarg { 2038 SLIST_ENTRY(mntaarg) next; 2039}; 2040 2041/* The header for the mount arguments */ 2042struct mntarg { 2043 struct iovec *v; 2044 int len; 2045 int error; 2046 SLIST_HEAD(, mntaarg) list; 2047}; 2048 2049/* 2050 * Add a boolean argument. 2051 * 2052 * flag is the boolean value. 2053 * name must start with "no". 2054 */ 2055struct mntarg * 2056mount_argb(struct mntarg *ma, int flag, const char *name) 2057{ 2058 2059 KASSERT(name[0] == 'n' && name[1] == 'o', 2060 ("mount_argb(...,%s): name must start with 'no'", name)); 2061 2062 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2063} 2064 2065/* 2066 * Add an argument printf style 2067 */ 2068struct mntarg * 2069mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2070{ 2071 va_list ap; 2072 struct mntaarg *maa; 2073 struct sbuf *sb; 2074 int len; 2075 2076 if (ma == NULL) { 2077 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2078 SLIST_INIT(&ma->list); 2079 } 2080 if (ma->error) 2081 return (ma); 2082 2083 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2084 M_MOUNT, M_WAITOK); 2085 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2086 ma->v[ma->len].iov_len = strlen(name) + 1; 2087 ma->len++; 2088 2089 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND); 2090 va_start(ap, fmt); 2091 sbuf_vprintf(sb, fmt, ap); 2092 va_end(ap); 2093 sbuf_finish(sb); 2094 len = sbuf_len(sb) + 1; 2095 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2096 SLIST_INSERT_HEAD(&ma->list, maa, next); 2097 bcopy(sbuf_data(sb), maa + 1, len); 2098 sbuf_delete(sb); 2099 2100 ma->v[ma->len].iov_base = maa + 1; 2101 ma->v[ma->len].iov_len = len; 2102 ma->len++; 2103 2104 return (ma); 2105} 2106 2107/* 2108 * Add an argument which is a userland string. 2109 */ 2110struct mntarg * 2111mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2112{ 2113 struct mntaarg *maa; 2114 char *tbuf; 2115 2116 if (val == NULL) 2117 return (ma); 2118 if (ma == NULL) { 2119 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2120 SLIST_INIT(&ma->list); 2121 } 2122 if (ma->error) 2123 return (ma); 2124 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2125 SLIST_INSERT_HEAD(&ma->list, maa, next); 2126 tbuf = (void *)(maa + 1); 2127 ma->error = copyinstr(val, tbuf, len, NULL); 2128 return (mount_arg(ma, name, tbuf, -1)); 2129} 2130 2131/* 2132 * Plain argument. 2133 * 2134 * If length is -1, use printf. 2135 */ 2136struct mntarg * 2137mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2138{ 2139 2140 if (ma == NULL) { 2141 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2142 SLIST_INIT(&ma->list); 2143 } 2144 if (ma->error) 2145 return (ma); 2146 2147 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2148 M_MOUNT, M_WAITOK); 2149 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2150 ma->v[ma->len].iov_len = strlen(name) + 1; 2151 ma->len++; 2152 2153 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2154 if (len < 0) 2155 ma->v[ma->len].iov_len = strlen(val) + 1; 2156 else 2157 ma->v[ma->len].iov_len = len; 2158 ma->len++; 2159 return (ma); 2160} 2161 2162/* 2163 * Free a mntarg structure 2164 */ 2165static void 2166free_mntarg(struct mntarg *ma) 2167{ 2168 struct mntaarg *maa; 2169 2170 while (!SLIST_EMPTY(&ma->list)) { 2171 maa = SLIST_FIRST(&ma->list); 2172 SLIST_REMOVE_HEAD(&ma->list, next); 2173 free(maa, M_MOUNT); 2174 } 2175 free(ma->v, M_MOUNT); 2176 free(ma, M_MOUNT); 2177} 2178 2179/* 2180 * Mount a filesystem 2181 */ 2182int 2183kernel_mount(struct mntarg *ma, int flags) 2184{ 2185 struct uio auio; 2186 int error; 2187 2188 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2189 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2190 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2191 2192 auio.uio_iov = ma->v; 2193 auio.uio_iovcnt = ma->len; 2194 auio.uio_segflg = UIO_SYSSPACE; 2195 2196 error = ma->error; 2197 if (!error) 2198 error = vfs_donmount(curthread, flags, &auio); 2199 free_mntarg(ma); 2200 return (error); 2201} 2202 2203/* 2204 * A printflike function to mount a filesystem. 2205 */ 2206int 2207kernel_vmount(int flags, ...) 2208{ 2209 struct mntarg *ma = NULL; 2210 va_list ap; 2211 const char *cp; 2212 const void *vp; 2213 int error; 2214 2215 va_start(ap, flags); 2216 for (;;) { 2217 cp = va_arg(ap, const char *); 2218 if (cp == NULL) 2219 break; 2220 vp = va_arg(ap, const void *); 2221 ma = mount_arg(ma, cp, vp, -1); 2222 } 2223 va_end(ap); 2224 2225 error = kernel_mount(ma, flags); 2226 return (error); 2227} 2228