vfs_mount.c revision 179658
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 179658 2008-06-08 20:43:27Z wkoszek $"); 39 40#include <sys/param.h> 41#include <sys/conf.h> 42#include <sys/fcntl.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_mac.h" 72 73#define ROOTNAME "root_device" 74#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 75 76static int vfs_domount(struct thread *td, const char *fstype, 77 char *fspath, int fsflags, void *fsdata); 78static int vfs_mountroot_ask(void); 79static int vfs_mountroot_try(const char *mountfrom); 80static int vfs_donmount(struct thread *td, int fsflags, 81 struct uio *fsoptions); 82static void free_mntarg(struct mntarg *ma); 83static int vfs_getopt_pos(struct vfsoptlist *opts, const char *name); 84 85static int usermount = 0; 86SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 87 "Unprivileged users may mount and unmount file systems"); 88 89MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 90MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 91static uma_zone_t mount_zone; 92 93/* List of mounted filesystems. */ 94struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 95 96/* For any iteration/modification of mountlist */ 97struct mtx mountlist_mtx; 98MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 99 100TAILQ_HEAD(vfsoptlist, vfsopt); 101struct vfsopt { 102 TAILQ_ENTRY(vfsopt) link; 103 char *name; 104 void *value; 105 int len; 106}; 107 108/* 109 * The vnode of the system's root (/ in the filesystem, without chroot 110 * active.) 111 */ 112struct vnode *rootvnode; 113 114/* 115 * The root filesystem is detailed in the kernel environment variable 116 * vfs.root.mountfrom, which is expected to be in the general format 117 * 118 * <vfsname>:[<path>] 119 * vfsname := the name of a VFS known to the kernel and capable 120 * of being mounted as root 121 * path := disk device name or other data used by the filesystem 122 * to locate its physical store 123 */ 124 125/* 126 * Global opts, taken by all filesystems 127 */ 128static const char *global_opts[] = { 129 "errmsg", 130 "fstype", 131 "fspath", 132 "ro", 133 "rw", 134 "nosuid", 135 "noexec", 136 "update", 137 NULL 138}; 139 140/* 141 * The root specifiers we will try if RB_CDROM is specified. 142 */ 143static char *cdrom_rootdevnames[] = { 144 "cd9660:cd0", 145 "cd9660:acd0", 146 NULL 147}; 148 149/* legacy find-root code */ 150char *rootdevnames[2] = {NULL, NULL}; 151#ifndef ROOTDEVNAME 152# define ROOTDEVNAME NULL 153#endif 154static const char *ctrootdevname = ROOTDEVNAME; 155 156/* 157 * --------------------------------------------------------------------- 158 * Functions for building and sanitizing the mount options 159 */ 160 161/* Remove one mount option. */ 162static void 163vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 164{ 165 166 TAILQ_REMOVE(opts, opt, link); 167 free(opt->name, M_MOUNT); 168 if (opt->value != NULL) 169 free(opt->value, M_MOUNT); 170#ifdef INVARIANTS 171 else if (opt->len != 0) 172 panic("%s: mount option with NULL value but length != 0", 173 __func__); 174#endif 175 free(opt, M_MOUNT); 176} 177 178/* Release all resources related to the mount options. */ 179void 180vfs_freeopts(struct vfsoptlist *opts) 181{ 182 struct vfsopt *opt; 183 184 while (!TAILQ_EMPTY(opts)) { 185 opt = TAILQ_FIRST(opts); 186 vfs_freeopt(opts, opt); 187 } 188 free(opts, M_MOUNT); 189} 190 191void 192vfs_deleteopt(struct vfsoptlist *opts, const char *name) 193{ 194 struct vfsopt *opt, *temp; 195 196 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 197 if (strcmp(opt->name, name) == 0) 198 vfs_freeopt(opts, opt); 199 } 200} 201 202/* 203 * Check if options are equal (with or without the "no" prefix). 204 */ 205static int 206vfs_equalopts(const char *opt1, const char *opt2) 207{ 208 209 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 210 if (strcmp(opt1, opt2) == 0) 211 return (1); 212 /* "noopt" vs. "opt" */ 213 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 214 return (1); 215 /* "opt" vs. "noopt" */ 216 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 217 return (1); 218 return (0); 219} 220 221/* 222 * If a mount option is specified several times, 223 * (with or without the "no" prefix) only keep 224 * the last occurence of it. 225 */ 226static void 227vfs_sanitizeopts(struct vfsoptlist *opts) 228{ 229 struct vfsopt *opt, *opt2, *tmp; 230 231 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 232 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 233 while (opt2 != NULL) { 234 if (vfs_equalopts(opt->name, opt2->name)) { 235 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 236 vfs_freeopt(opts, opt2); 237 opt2 = tmp; 238 } else { 239 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 240 } 241 } 242 } 243} 244 245/* 246 * Build a linked list of mount options from a struct uio. 247 */ 248static int 249vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 250{ 251 struct vfsoptlist *opts; 252 struct vfsopt *opt; 253 size_t memused; 254 unsigned int i, iovcnt; 255 int error, namelen, optlen; 256 257 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 258 TAILQ_INIT(opts); 259 memused = 0; 260 iovcnt = auio->uio_iovcnt; 261 for (i = 0; i < iovcnt; i += 2) { 262 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 263 namelen = auio->uio_iov[i].iov_len; 264 optlen = auio->uio_iov[i + 1].iov_len; 265 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 266 opt->value = NULL; 267 opt->len = 0; 268 269 /* 270 * Do this early, so jumps to "bad" will free the current 271 * option. 272 */ 273 TAILQ_INSERT_TAIL(opts, opt, link); 274 memused += sizeof(struct vfsopt) + optlen + namelen; 275 276 /* 277 * Avoid consuming too much memory, and attempts to overflow 278 * memused. 279 */ 280 if (memused > VFS_MOUNTARG_SIZE_MAX || 281 optlen > VFS_MOUNTARG_SIZE_MAX || 282 namelen > VFS_MOUNTARG_SIZE_MAX) { 283 error = EINVAL; 284 goto bad; 285 } 286 287 if (auio->uio_segflg == UIO_SYSSPACE) { 288 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 289 } else { 290 error = copyin(auio->uio_iov[i].iov_base, opt->name, 291 namelen); 292 if (error) 293 goto bad; 294 } 295 /* Ensure names are null-terminated strings. */ 296 if (opt->name[namelen - 1] != '\0') { 297 error = EINVAL; 298 goto bad; 299 } 300 if (optlen != 0) { 301 opt->len = optlen; 302 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 303 if (auio->uio_segflg == UIO_SYSSPACE) { 304 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 305 optlen); 306 } else { 307 error = copyin(auio->uio_iov[i + 1].iov_base, 308 opt->value, optlen); 309 if (error) 310 goto bad; 311 } 312 } 313 } 314 vfs_sanitizeopts(opts); 315 *options = opts; 316 return (0); 317bad: 318 vfs_freeopts(opts); 319 return (error); 320} 321 322/* 323 * Merge the old mount options with the new ones passed 324 * in the MNT_UPDATE case. 325 * 326 * XXX This function will keep a "nofoo" option in the 327 * new options if there is no matching "foo" option 328 * to be cancelled in the old options. This is a bug 329 * if the option's canonical name is "foo". E.g., "noro" 330 * shouldn't end up in the mount point's active options, 331 * but it can. 332 */ 333static void 334vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 335{ 336 struct vfsopt *opt, *opt2, *new; 337 338 TAILQ_FOREACH(opt, opts, link) { 339 /* 340 * Check that this option hasn't been redefined 341 * nor cancelled with a "no" mount option. 342 */ 343 opt2 = TAILQ_FIRST(toopts); 344 while (opt2 != NULL) { 345 if (strcmp(opt2->name, opt->name) == 0) 346 goto next; 347 if (strncmp(opt2->name, "no", 2) == 0 && 348 strcmp(opt2->name + 2, opt->name) == 0) { 349 vfs_freeopt(toopts, opt2); 350 goto next; 351 } 352 opt2 = TAILQ_NEXT(opt2, link); 353 } 354 /* We want this option, duplicate it. */ 355 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 356 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 357 strcpy(new->name, opt->name); 358 if (opt->len != 0) { 359 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 360 bcopy(opt->value, new->value, opt->len); 361 } else { 362 new->value = NULL; 363 } 364 new->len = opt->len; 365 TAILQ_INSERT_TAIL(toopts, new, link); 366next: 367 continue; 368 } 369} 370 371/* 372 * Mount a filesystem. 373 */ 374int 375nmount(td, uap) 376 struct thread *td; 377 struct nmount_args /* { 378 struct iovec *iovp; 379 unsigned int iovcnt; 380 int flags; 381 } */ *uap; 382{ 383 struct uio *auio; 384 struct iovec *iov; 385 unsigned int i; 386 int error; 387 u_int iovcnt; 388 389 AUDIT_ARG(fflags, uap->flags); 390 391 /* 392 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 393 * userspace to set this flag, but we must filter it out if we want 394 * MNT_UPDATE on the root file system to work. 395 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 396 */ 397 uap->flags &= ~MNT_ROOTFS; 398 399 iovcnt = uap->iovcnt; 400 /* 401 * Check that we have an even number of iovec's 402 * and that we have at least two options. 403 */ 404 if ((iovcnt & 1) || (iovcnt < 4)) 405 return (EINVAL); 406 407 error = copyinuio(uap->iovp, iovcnt, &auio); 408 if (error) 409 return (error); 410 iov = auio->uio_iov; 411 for (i = 0; i < iovcnt; i++) { 412 if (iov->iov_len > MMAXOPTIONLEN) { 413 free(auio, M_IOV); 414 return (EINVAL); 415 } 416 iov++; 417 } 418 error = vfs_donmount(td, uap->flags, auio); 419 420 free(auio, M_IOV); 421 return (error); 422} 423 424/* 425 * --------------------------------------------------------------------- 426 * Various utility functions 427 */ 428 429void 430vfs_ref(struct mount *mp) 431{ 432 433 MNT_ILOCK(mp); 434 MNT_REF(mp); 435 MNT_IUNLOCK(mp); 436} 437 438void 439vfs_rel(struct mount *mp) 440{ 441 442 MNT_ILOCK(mp); 443 MNT_REL(mp); 444 MNT_IUNLOCK(mp); 445} 446 447static int 448mount_init(void *mem, int size, int flags) 449{ 450 struct mount *mp; 451 452 mp = (struct mount *)mem; 453 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 454 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 455 return (0); 456} 457 458static void 459mount_fini(void *mem, int size) 460{ 461 struct mount *mp; 462 463 mp = (struct mount *)mem; 464 lockdestroy(&mp->mnt_lock); 465 mtx_destroy(&mp->mnt_mtx); 466} 467 468/* 469 * Allocate and initialize the mount point struct. 470 */ 471struct mount * 472vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, 473 const char *fspath, struct thread *td) 474{ 475 struct mount *mp; 476 477 mp = uma_zalloc(mount_zone, M_WAITOK); 478 bzero(&mp->mnt_startzero, 479 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 480 TAILQ_INIT(&mp->mnt_nvnodelist); 481 mp->mnt_nvnodelistsize = 0; 482 mp->mnt_ref = 0; 483 (void) vfs_busy(mp, LK_NOWAIT, 0, td); 484 mp->mnt_op = vfsp->vfc_vfsops; 485 mp->mnt_vfc = vfsp; 486 vfsp->vfc_refcount++; /* XXX Unlocked */ 487 mp->mnt_stat.f_type = vfsp->vfc_typenum; 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_mount_init(mp); 497 mac_mount_create(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_mount_destroy(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, "atime") == 0) { 647 free(opt->name, M_MOUNT); 648 opt->name = strdup("nonoatime", M_MOUNT); 649 } 650 else if (strcmp(opt->name, "noclusterr") == 0) 651 fsflags |= MNT_NOCLUSTERR; 652 else if (strcmp(opt->name, "clusterr") == 0) { 653 free(opt->name, M_MOUNT); 654 opt->name = strdup("nonoclusterr", M_MOUNT); 655 } 656 else if (strcmp(opt->name, "noclusterw") == 0) 657 fsflags |= MNT_NOCLUSTERW; 658 else if (strcmp(opt->name, "clusterw") == 0) { 659 free(opt->name, M_MOUNT); 660 opt->name = strdup("nonoclusterw", M_MOUNT); 661 } 662 else if (strcmp(opt->name, "noexec") == 0) 663 fsflags |= MNT_NOEXEC; 664 else if (strcmp(opt->name, "exec") == 0) { 665 free(opt->name, M_MOUNT); 666 opt->name = strdup("nonoexec", M_MOUNT); 667 } 668 else if (strcmp(opt->name, "nosuid") == 0) 669 fsflags |= MNT_NOSUID; 670 else if (strcmp(opt->name, "suid") == 0) { 671 free(opt->name, M_MOUNT); 672 opt->name = strdup("nonosuid", M_MOUNT); 673 } 674 else if (strcmp(opt->name, "nosymfollow") == 0) 675 fsflags |= MNT_NOSYMFOLLOW; 676 else if (strcmp(opt->name, "symfollow") == 0) { 677 free(opt->name, M_MOUNT); 678 opt->name = strdup("nonosymfollow", M_MOUNT); 679 } 680 else if (strcmp(opt->name, "noro") == 0) { 681 fsflags &= ~MNT_RDONLY; 682 has_noro = 1; 683 } 684 else if (strcmp(opt->name, "rw") == 0) { 685 fsflags &= ~MNT_RDONLY; 686 has_rw = 1; 687 } 688 else if (strcmp(opt->name, "ro") == 0) 689 fsflags |= MNT_RDONLY; 690 else if (strcmp(opt->name, "rdonly") == 0) { 691 free(opt->name, M_MOUNT); 692 opt->name = strdup("ro", M_MOUNT); 693 fsflags |= MNT_RDONLY; 694 } 695 else if (strcmp(opt->name, "suiddir") == 0) 696 fsflags |= MNT_SUIDDIR; 697 else if (strcmp(opt->name, "sync") == 0) 698 fsflags |= MNT_SYNCHRONOUS; 699 else if (strcmp(opt->name, "union") == 0) 700 fsflags |= MNT_UNION; 701 } 702 703 /* 704 * If "rw" was specified as a mount option, and we 705 * are trying to update a mount-point from "ro" to "rw", 706 * we need a mount option "noro", since in vfs_mergeopts(), 707 * "noro" will cancel "ro", but "rw" will not do anything. 708 */ 709 if (has_rw && !has_noro) { 710 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 711 noro_opt->name = strdup("noro", M_MOUNT); 712 noro_opt->value = NULL; 713 noro_opt->len = 0; 714 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 715 } 716 717 /* 718 * Be ultra-paranoid about making sure the type and fspath 719 * variables will fit in our mp buffers, including the 720 * terminating NUL. 721 */ 722 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 723 error = ENAMETOOLONG; 724 goto bail; 725 } 726 727 mtx_lock(&Giant); 728 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 729 mtx_unlock(&Giant); 730bail: 731 /* copyout the errmsg */ 732 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 733 && errmsg_len > 0 && errmsg != NULL) { 734 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 735 bcopy(errmsg, 736 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 737 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 738 } else { 739 copyout(errmsg, 740 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 741 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 742 } 743 } 744 745 if (error != 0) 746 vfs_freeopts(optlist); 747 return (error); 748} 749 750/* 751 * Old mount API. 752 */ 753#ifndef _SYS_SYSPROTO_H_ 754struct mount_args { 755 char *type; 756 char *path; 757 int flags; 758 caddr_t data; 759}; 760#endif 761/* ARGSUSED */ 762int 763mount(td, uap) 764 struct thread *td; 765 struct mount_args /* { 766 char *type; 767 char *path; 768 int flags; 769 caddr_t data; 770 } */ *uap; 771{ 772 char *fstype; 773 struct vfsconf *vfsp = NULL; 774 struct mntarg *ma = NULL; 775 int error; 776 777 AUDIT_ARG(fflags, uap->flags); 778 779 /* 780 * Filter out MNT_ROOTFS. We do not want clients of mount() in 781 * userspace to set this flag, but we must filter it out if we want 782 * MNT_UPDATE on the root file system to work. 783 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 784 */ 785 uap->flags &= ~MNT_ROOTFS; 786 787 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 788 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 789 if (error) { 790 free(fstype, M_TEMP); 791 return (error); 792 } 793 794 AUDIT_ARG(text, fstype); 795 mtx_lock(&Giant); 796 vfsp = vfs_byname_kld(fstype, td, &error); 797 free(fstype, M_TEMP); 798 if (vfsp == NULL) { 799 mtx_unlock(&Giant); 800 return (ENOENT); 801 } 802 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 803 mtx_unlock(&Giant); 804 return (EOPNOTSUPP); 805 } 806 807 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 808 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 809 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 810 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 811 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 812 813 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 814 mtx_unlock(&Giant); 815 return (error); 816} 817 818 819/* 820 * vfs_domount(): actually attempt a filesystem mount. 821 */ 822static int 823vfs_domount( 824 struct thread *td, /* Calling thread. */ 825 const char *fstype, /* Filesystem type. */ 826 char *fspath, /* Mount path. */ 827 int fsflags, /* Flags common to all filesystems. */ 828 void *fsdata /* Options local to the filesystem. */ 829 ) 830{ 831 struct vnode *vp; 832 struct mount *mp; 833 struct vfsconf *vfsp; 834 struct export_args export; 835 int error, flag = 0; 836 struct vattr va; 837 struct nameidata nd; 838 839 mtx_assert(&Giant, MA_OWNED); 840 /* 841 * Be ultra-paranoid about making sure the type and fspath 842 * variables will fit in our mp buffers, including the 843 * terminating NUL. 844 */ 845 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 846 return (ENAMETOOLONG); 847 848 if (jailed(td->td_ucred) || usermount == 0) { 849 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 850 return (error); 851 } 852 853 /* 854 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 855 */ 856 if (fsflags & MNT_EXPORTED) { 857 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 858 if (error) 859 return (error); 860 } 861 if (fsflags & MNT_SUIDDIR) { 862 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 863 if (error) 864 return (error); 865 } 866 /* 867 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 868 */ 869 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 870 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 871 fsflags |= MNT_NOSUID | MNT_USER; 872 } 873 874 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 875 vfsp = NULL; 876 if ((fsflags & MNT_UPDATE) == 0) { 877 /* Don't try to load KLDs if we're mounting the root. */ 878 if (fsflags & MNT_ROOTFS) 879 vfsp = vfs_byname(fstype); 880 else 881 vfsp = vfs_byname_kld(fstype, td, &error); 882 if (vfsp == NULL) 883 return (ENODEV); 884 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 885 return (EPERM); 886 } 887 /* 888 * Get vnode to be covered 889 */ 890 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 891 fspath, td); 892 if ((error = namei(&nd)) != 0) 893 return (error); 894 NDFREE(&nd, NDF_ONLY_PNBUF); 895 vp = nd.ni_vp; 896 if (fsflags & MNT_UPDATE) { 897 if ((vp->v_vflag & VV_ROOT) == 0) { 898 vput(vp); 899 return (EINVAL); 900 } 901 mp = vp->v_mount; 902 MNT_ILOCK(mp); 903 flag = mp->mnt_flag; 904 /* 905 * We only allow the filesystem to be reloaded if it 906 * is currently mounted read-only. 907 */ 908 if ((fsflags & MNT_RELOAD) && 909 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 910 MNT_IUNLOCK(mp); 911 vput(vp); 912 return (EOPNOTSUPP); /* Needs translation */ 913 } 914 MNT_IUNLOCK(mp); 915 /* 916 * Only privileged root, or (if MNT_USER is set) the user that 917 * did the original mount is permitted to update it. 918 */ 919 error = vfs_suser(mp, td); 920 if (error) { 921 vput(vp); 922 return (error); 923 } 924 if (vfs_busy(mp, LK_NOWAIT, 0, td)) { 925 vput(vp); 926 return (EBUSY); 927 } 928 VI_LOCK(vp); 929 if ((vp->v_iflag & VI_MOUNT) != 0 || 930 vp->v_mountedhere != NULL) { 931 VI_UNLOCK(vp); 932 vfs_unbusy(mp, td); 933 vput(vp); 934 return (EBUSY); 935 } 936 vp->v_iflag |= VI_MOUNT; 937 VI_UNLOCK(vp); 938 MNT_ILOCK(mp); 939 mp->mnt_flag |= fsflags & 940 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 941 MNT_IUNLOCK(mp); 942 VOP_UNLOCK(vp, 0); 943 mp->mnt_optnew = fsdata; 944 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 945 } else { 946 /* 947 * If the user is not root, ensure that they own the directory 948 * onto which we are attempting to mount. 949 */ 950 error = VOP_GETATTR(vp, &va, td->td_ucred, td); 951 if (error) { 952 vput(vp); 953 return (error); 954 } 955 if (va.va_uid != td->td_ucred->cr_uid) { 956 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 957 0); 958 if (error) { 959 vput(vp); 960 return (error); 961 } 962 } 963 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 964 if (error != 0) { 965 vput(vp); 966 return (error); 967 } 968 if (vp->v_type != VDIR) { 969 vput(vp); 970 return (ENOTDIR); 971 } 972 VI_LOCK(vp); 973 if ((vp->v_iflag & VI_MOUNT) != 0 || 974 vp->v_mountedhere != NULL) { 975 VI_UNLOCK(vp); 976 vput(vp); 977 return (EBUSY); 978 } 979 vp->v_iflag |= VI_MOUNT; 980 VI_UNLOCK(vp); 981 982 /* 983 * Allocate and initialize the filesystem. 984 */ 985 mp = vfs_mount_alloc(vp, vfsp, fspath, td); 986 VOP_UNLOCK(vp, 0); 987 988 /* XXXMAC: pass to vfs_mount_alloc? */ 989 mp->mnt_optnew = fsdata; 990 } 991 992 /* 993 * Set the mount level flags. 994 */ 995 MNT_ILOCK(mp); 996 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 997 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 998 MNT_RDONLY)); 999 if ((mp->mnt_flag & MNT_ASYNC) == 0) 1000 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1001 MNT_IUNLOCK(mp); 1002 /* 1003 * Mount the filesystem. 1004 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1005 * get. No freeing of cn_pnbuf. 1006 */ 1007 error = VFS_MOUNT(mp, td); 1008 1009 /* 1010 * Process the export option only if we are 1011 * updating mount options. 1012 */ 1013 if (!error && (fsflags & MNT_UPDATE)) { 1014 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 1015 sizeof(export)) == 0) 1016 error = vfs_export(mp, &export); 1017 } 1018 1019 if (!error) { 1020 if (mp->mnt_opt != NULL) 1021 vfs_freeopts(mp->mnt_opt); 1022 mp->mnt_opt = mp->mnt_optnew; 1023 (void)VFS_STATFS(mp, &mp->mnt_stat, td); 1024 } 1025 /* 1026 * Prevent external consumers of mount options from reading 1027 * mnt_optnew. 1028 */ 1029 mp->mnt_optnew = NULL; 1030 if (mp->mnt_flag & MNT_UPDATE) { 1031 MNT_ILOCK(mp); 1032 if (error) 1033 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1034 (flag & ~MNT_QUOTA); 1035 else 1036 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1037 MNT_FORCE | MNT_SNAPSHOT); 1038 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1039 mp->mnt_kern_flag |= MNTK_ASYNC; 1040 else 1041 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1042 MNT_IUNLOCK(mp); 1043 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1044 if (mp->mnt_syncer == NULL) 1045 error = vfs_allocate_syncvnode(mp); 1046 } else { 1047 if (mp->mnt_syncer != NULL) 1048 vrele(mp->mnt_syncer); 1049 mp->mnt_syncer = NULL; 1050 } 1051 vfs_unbusy(mp, td); 1052 VI_LOCK(vp); 1053 vp->v_iflag &= ~VI_MOUNT; 1054 VI_UNLOCK(vp); 1055 vrele(vp); 1056 return (error); 1057 } 1058 MNT_ILOCK(mp); 1059 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1060 mp->mnt_kern_flag |= MNTK_ASYNC; 1061 else 1062 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1063 MNT_IUNLOCK(mp); 1064 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1065 /* 1066 * Put the new filesystem on the mount list after root. 1067 */ 1068 cache_purge(vp); 1069 if (!error) { 1070 struct vnode *newdp; 1071 1072 VI_LOCK(vp); 1073 vp->v_iflag &= ~VI_MOUNT; 1074 VI_UNLOCK(vp); 1075 vp->v_mountedhere = mp; 1076 mtx_lock(&mountlist_mtx); 1077 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1078 mtx_unlock(&mountlist_mtx); 1079 vfs_event_signal(NULL, VQ_MOUNT, 0); 1080 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 1081 panic("mount: lost mount"); 1082 mountcheckdirs(vp, newdp); 1083 vput(newdp); 1084 VOP_UNLOCK(vp, 0); 1085 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1086 error = vfs_allocate_syncvnode(mp); 1087 vfs_unbusy(mp, td); 1088 if (error) 1089 vrele(vp); 1090 } else { 1091 VI_LOCK(vp); 1092 vp->v_iflag &= ~VI_MOUNT; 1093 VI_UNLOCK(vp); 1094 vfs_unbusy(mp, td); 1095 vfs_mount_destroy(mp); 1096 vput(vp); 1097 } 1098 return (error); 1099} 1100 1101/* 1102 * Unmount a filesystem. 1103 * 1104 * Note: unmount takes a path to the vnode mounted on as argument, not 1105 * special file (as before). 1106 */ 1107#ifndef _SYS_SYSPROTO_H_ 1108struct unmount_args { 1109 char *path; 1110 int flags; 1111}; 1112#endif 1113/* ARGSUSED */ 1114int 1115unmount(td, uap) 1116 struct thread *td; 1117 register struct unmount_args /* { 1118 char *path; 1119 int flags; 1120 } */ *uap; 1121{ 1122 struct mount *mp; 1123 char *pathbuf; 1124 int error, id0, id1; 1125 1126 if (jailed(td->td_ucred) || usermount == 0) { 1127 error = priv_check(td, PRIV_VFS_UNMOUNT); 1128 if (error) 1129 return (error); 1130 } 1131 1132 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1133 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1134 if (error) { 1135 free(pathbuf, M_TEMP); 1136 return (error); 1137 } 1138 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1139 mtx_lock(&Giant); 1140 if (uap->flags & MNT_BYFSID) { 1141 /* Decode the filesystem ID. */ 1142 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1143 mtx_unlock(&Giant); 1144 free(pathbuf, M_TEMP); 1145 return (EINVAL); 1146 } 1147 1148 mtx_lock(&mountlist_mtx); 1149 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1150 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1151 mp->mnt_stat.f_fsid.val[1] == id1) 1152 break; 1153 } 1154 mtx_unlock(&mountlist_mtx); 1155 } else { 1156 mtx_lock(&mountlist_mtx); 1157 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1158 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1159 break; 1160 } 1161 mtx_unlock(&mountlist_mtx); 1162 } 1163 free(pathbuf, M_TEMP); 1164 if (mp == NULL) { 1165 /* 1166 * Previously we returned ENOENT for a nonexistent path and 1167 * EINVAL for a non-mountpoint. We cannot tell these apart 1168 * now, so in the !MNT_BYFSID case return the more likely 1169 * EINVAL for compatibility. 1170 */ 1171 mtx_unlock(&Giant); 1172 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1173 } 1174 1175 /* 1176 * Don't allow unmounting the root filesystem. 1177 */ 1178 if (mp->mnt_flag & MNT_ROOTFS) { 1179 mtx_unlock(&Giant); 1180 return (EINVAL); 1181 } 1182 error = dounmount(mp, uap->flags, td); 1183 mtx_unlock(&Giant); 1184 return (error); 1185} 1186 1187/* 1188 * Do the actual filesystem unmount. 1189 */ 1190int 1191dounmount(mp, flags, td) 1192 struct mount *mp; 1193 int flags; 1194 struct thread *td; 1195{ 1196 struct vnode *coveredvp, *fsrootvp; 1197 int error; 1198 int async_flag; 1199 int mnt_gen_r; 1200 1201 mtx_assert(&Giant, MA_OWNED); 1202 1203 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1204 mnt_gen_r = mp->mnt_gen; 1205 VI_LOCK(coveredvp); 1206 vholdl(coveredvp); 1207 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1208 vdrop(coveredvp); 1209 /* 1210 * Check for mp being unmounted while waiting for the 1211 * covered vnode lock. 1212 */ 1213 if (coveredvp->v_mountedhere != mp || 1214 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1215 VOP_UNLOCK(coveredvp, 0); 1216 return (EBUSY); 1217 } 1218 } 1219 /* 1220 * Only privileged root, or (if MNT_USER is set) the user that did the 1221 * original mount is permitted to unmount this filesystem. 1222 */ 1223 error = vfs_suser(mp, td); 1224 if (error) { 1225 if (coveredvp) 1226 VOP_UNLOCK(coveredvp, 0); 1227 return (error); 1228 } 1229 1230 MNT_ILOCK(mp); 1231 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1232 MNT_IUNLOCK(mp); 1233 if (coveredvp) 1234 VOP_UNLOCK(coveredvp, 0); 1235 return (EBUSY); 1236 } 1237 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1238 /* Allow filesystems to detect that a forced unmount is in progress. */ 1239 if (flags & MNT_FORCE) 1240 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1241 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 1242 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp)); 1243 if (error) { 1244 MNT_ILOCK(mp); 1245 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1246 MNTK_UNMOUNTF); 1247 if (mp->mnt_kern_flag & MNTK_MWAIT) 1248 wakeup(mp); 1249 MNT_IUNLOCK(mp); 1250 if (coveredvp) 1251 VOP_UNLOCK(coveredvp, 0); 1252 return (error); 1253 } 1254 vn_start_write(NULL, &mp, V_WAIT); 1255 1256 if (mp->mnt_flag & MNT_EXPUBLIC) 1257 vfs_setpublicfs(NULL, NULL, NULL); 1258 1259 vfs_msync(mp, MNT_WAIT); 1260 MNT_ILOCK(mp); 1261 async_flag = mp->mnt_flag & MNT_ASYNC; 1262 mp->mnt_flag &= ~MNT_ASYNC; 1263 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1264 MNT_IUNLOCK(mp); 1265 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1266 if (mp->mnt_syncer != NULL) 1267 vrele(mp->mnt_syncer); 1268 /* 1269 * For forced unmounts, move process cdir/rdir refs on the fs root 1270 * vnode to the covered vnode. For non-forced unmounts we want 1271 * such references to cause an EBUSY error. 1272 */ 1273 if ((flags & MNT_FORCE) && 1274 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1275 if (mp->mnt_vnodecovered != NULL) 1276 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1277 if (fsrootvp == rootvnode) { 1278 vrele(rootvnode); 1279 rootvnode = NULL; 1280 } 1281 vput(fsrootvp); 1282 } 1283 if (((mp->mnt_flag & MNT_RDONLY) || 1284 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1285 (flags & MNT_FORCE)) { 1286 error = VFS_UNMOUNT(mp, flags, td); 1287 } 1288 vn_finished_write(mp); 1289 /* 1290 * If we failed to flush the dirty blocks for this mount point, 1291 * undo all the cdir/rdir and rootvnode changes we made above. 1292 * Unless we failed to do so because the device is reporting that 1293 * it doesn't exist anymore. 1294 */ 1295 if (error && error != ENXIO) { 1296 if ((flags & MNT_FORCE) && 1297 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1298 if (mp->mnt_vnodecovered != NULL) 1299 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1300 if (rootvnode == NULL) { 1301 rootvnode = fsrootvp; 1302 vref(rootvnode); 1303 } 1304 vput(fsrootvp); 1305 } 1306 MNT_ILOCK(mp); 1307 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1308 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1309 MNT_IUNLOCK(mp); 1310 (void) vfs_allocate_syncvnode(mp); 1311 MNT_ILOCK(mp); 1312 } 1313 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1314 mp->mnt_flag |= async_flag; 1315 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1316 mp->mnt_kern_flag |= MNTK_ASYNC; 1317 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL); 1318 if (mp->mnt_kern_flag & MNTK_MWAIT) 1319 wakeup(mp); 1320 MNT_IUNLOCK(mp); 1321 if (coveredvp) 1322 VOP_UNLOCK(coveredvp, 0); 1323 return (error); 1324 } 1325 mtx_lock(&mountlist_mtx); 1326 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1327 mtx_unlock(&mountlist_mtx); 1328 if (coveredvp != NULL) { 1329 coveredvp->v_mountedhere = NULL; 1330 vput(coveredvp); 1331 } 1332 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1333 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL); 1334 vfs_mount_destroy(mp); 1335 return (0); 1336} 1337 1338/* 1339 * --------------------------------------------------------------------- 1340 * Mounting of root filesystem 1341 * 1342 */ 1343 1344struct root_hold_token { 1345 const char *who; 1346 LIST_ENTRY(root_hold_token) list; 1347}; 1348 1349static LIST_HEAD(, root_hold_token) root_holds = 1350 LIST_HEAD_INITIALIZER(&root_holds); 1351 1352static int root_mount_complete; 1353 1354/* 1355 * Hold root mount. 1356 */ 1357struct root_hold_token * 1358root_mount_hold(const char *identifier) 1359{ 1360 struct root_hold_token *h; 1361 1362 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1363 h->who = identifier; 1364 mtx_lock(&mountlist_mtx); 1365 LIST_INSERT_HEAD(&root_holds, h, list); 1366 mtx_unlock(&mountlist_mtx); 1367 return (h); 1368} 1369 1370/* 1371 * Release root mount. 1372 */ 1373void 1374root_mount_rel(struct root_hold_token *h) 1375{ 1376 1377 mtx_lock(&mountlist_mtx); 1378 LIST_REMOVE(h, list); 1379 wakeup(&root_holds); 1380 mtx_unlock(&mountlist_mtx); 1381 free(h, M_DEVBUF); 1382} 1383 1384/* 1385 * Wait for all subsystems to release root mount. 1386 */ 1387static void 1388root_mount_prepare(void) 1389{ 1390 struct root_hold_token *h; 1391 1392 for (;;) { 1393 DROP_GIANT(); 1394 g_waitidle(); 1395 PICKUP_GIANT(); 1396 mtx_lock(&mountlist_mtx); 1397 if (LIST_EMPTY(&root_holds)) { 1398 mtx_unlock(&mountlist_mtx); 1399 break; 1400 } 1401 printf("Root mount waiting for:"); 1402 LIST_FOREACH(h, &root_holds, list) 1403 printf(" %s", h->who); 1404 printf("\n"); 1405 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1406 hz); 1407 } 1408} 1409 1410/* 1411 * Root was mounted, share the good news. 1412 */ 1413static void 1414root_mount_done(void) 1415{ 1416 1417 /* 1418 * Use a mutex to prevent the wakeup being missed and waiting for 1419 * an extra 1 second sleep. 1420 */ 1421 mtx_lock(&mountlist_mtx); 1422 root_mount_complete = 1; 1423 wakeup(&root_mount_complete); 1424 mtx_unlock(&mountlist_mtx); 1425} 1426 1427/* 1428 * Return true if root is already mounted. 1429 */ 1430int 1431root_mounted(void) 1432{ 1433 1434 /* No mutex is acquired here because int stores are atomic. */ 1435 return (root_mount_complete); 1436} 1437 1438/* 1439 * Wait until root is mounted. 1440 */ 1441void 1442root_mount_wait(void) 1443{ 1444 1445 /* 1446 * Panic on an obvious deadlock - the function can't be called from 1447 * a thread which is doing the whole SYSINIT stuff. 1448 */ 1449 KASSERT(curthread->td_proc->p_pid != 0, 1450 ("root_mount_wait: cannot be called from the swapper thread")); 1451 mtx_lock(&mountlist_mtx); 1452 while (!root_mount_complete) { 1453 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1454 hz); 1455 } 1456 mtx_unlock(&mountlist_mtx); 1457} 1458 1459static void 1460set_rootvnode(struct thread *td) 1461{ 1462 struct proc *p; 1463 1464 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1465 panic("Cannot find root vnode"); 1466 1467 p = td->td_proc; 1468 FILEDESC_XLOCK(p->p_fd); 1469 1470 if (p->p_fd->fd_cdir != NULL) 1471 vrele(p->p_fd->fd_cdir); 1472 p->p_fd->fd_cdir = rootvnode; 1473 VREF(rootvnode); 1474 1475 if (p->p_fd->fd_rdir != NULL) 1476 vrele(p->p_fd->fd_rdir); 1477 p->p_fd->fd_rdir = rootvnode; 1478 VREF(rootvnode); 1479 1480 FILEDESC_XUNLOCK(p->p_fd); 1481 1482 VOP_UNLOCK(rootvnode, 0); 1483 1484 EVENTHANDLER_INVOKE(mountroot); 1485} 1486 1487/* 1488 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1489 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1490 */ 1491 1492static void 1493devfs_first(void) 1494{ 1495 struct thread *td = curthread; 1496 struct vfsoptlist *opts; 1497 struct vfsconf *vfsp; 1498 struct mount *mp = NULL; 1499 int error; 1500 1501 vfsp = vfs_byname("devfs"); 1502 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1503 if (vfsp == NULL) 1504 return; 1505 1506 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td); 1507 1508 error = VFS_MOUNT(mp, td); 1509 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1510 if (error) 1511 return; 1512 1513 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1514 TAILQ_INIT(opts); 1515 mp->mnt_opt = opts; 1516 1517 mtx_lock(&mountlist_mtx); 1518 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1519 mtx_unlock(&mountlist_mtx); 1520 1521 set_rootvnode(td); 1522 1523 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1524 if (error) 1525 printf("kern_symlink /dev -> / returns %d\n", error); 1526} 1527 1528/* 1529 * Surgically move our devfs to be mounted on /dev. 1530 */ 1531 1532static void 1533devfs_fixup(struct thread *td) 1534{ 1535 struct nameidata nd; 1536 int error; 1537 struct vnode *vp, *dvp; 1538 struct mount *mp; 1539 1540 /* Remove our devfs mount from the mountlist and purge the cache */ 1541 mtx_lock(&mountlist_mtx); 1542 mp = TAILQ_FIRST(&mountlist); 1543 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1544 mtx_unlock(&mountlist_mtx); 1545 cache_purgevfs(mp); 1546 1547 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1548 VI_LOCK(dvp); 1549 dvp->v_iflag &= ~VI_MOUNT; 1550 VI_UNLOCK(dvp); 1551 dvp->v_mountedhere = NULL; 1552 1553 /* Set up the real rootvnode, and purge the cache */ 1554 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1555 set_rootvnode(td); 1556 cache_purgevfs(rootvnode->v_mount); 1557 1558 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1559 error = namei(&nd); 1560 if (error) { 1561 printf("Lookup of /dev for devfs, error: %d\n", error); 1562 return; 1563 } 1564 NDFREE(&nd, NDF_ONLY_PNBUF); 1565 vp = nd.ni_vp; 1566 if (vp->v_type != VDIR) { 1567 vput(vp); 1568 } 1569 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 1570 if (error) { 1571 vput(vp); 1572 } 1573 cache_purge(vp); 1574 mp->mnt_vnodecovered = vp; 1575 vp->v_mountedhere = mp; 1576 mtx_lock(&mountlist_mtx); 1577 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1578 mtx_unlock(&mountlist_mtx); 1579 VOP_UNLOCK(vp, 0); 1580 vput(dvp); 1581 vfs_unbusy(mp, td); 1582 1583 /* Unlink the no longer needed /dev/dev -> / symlink */ 1584 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1585} 1586 1587/* 1588 * Report errors during filesystem mounting. 1589 */ 1590void 1591vfs_mount_error(struct mount *mp, const char *fmt, ...) 1592{ 1593 struct vfsoptlist *moptlist = mp->mnt_optnew; 1594 va_list ap; 1595 int error, len; 1596 char *errmsg; 1597 1598 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1599 if (error || errmsg == NULL || len <= 0) 1600 return; 1601 1602 va_start(ap, fmt); 1603 vsnprintf(errmsg, (size_t)len, fmt, ap); 1604 va_end(ap); 1605} 1606 1607/* 1608 * Find and mount the root filesystem 1609 */ 1610void 1611vfs_mountroot(void) 1612{ 1613 char *cp; 1614 int error, i, asked = 0; 1615 1616 root_mount_prepare(); 1617 1618 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1619 NULL, NULL, mount_init, mount_fini, 1620 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1621 devfs_first(); 1622 1623 /* 1624 * We are booted with instructions to prompt for the root filesystem. 1625 */ 1626 if (boothowto & RB_ASKNAME) { 1627 if (!vfs_mountroot_ask()) 1628 goto mounted; 1629 asked = 1; 1630 } 1631 1632 /* 1633 * The root filesystem information is compiled in, and we are 1634 * booted with instructions to use it. 1635 */ 1636 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1637 if (!vfs_mountroot_try(ctrootdevname)) 1638 goto mounted; 1639 ctrootdevname = NULL; 1640 } 1641 1642 /* 1643 * We've been given the generic "use CDROM as root" flag. This is 1644 * necessary because one media may be used in many different 1645 * devices, so we need to search for them. 1646 */ 1647 if (boothowto & RB_CDROM) { 1648 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1649 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1650 goto mounted; 1651 } 1652 } 1653 1654 /* 1655 * Try to use the value read by the loader from /etc/fstab, or 1656 * supplied via some other means. This is the preferred 1657 * mechanism. 1658 */ 1659 cp = getenv("vfs.root.mountfrom"); 1660 if (cp != NULL) { 1661 error = vfs_mountroot_try(cp); 1662 freeenv(cp); 1663 if (!error) 1664 goto mounted; 1665 } 1666 1667 /* 1668 * Try values that may have been computed by code during boot 1669 */ 1670 if (!vfs_mountroot_try(rootdevnames[0])) 1671 goto mounted; 1672 if (!vfs_mountroot_try(rootdevnames[1])) 1673 goto mounted; 1674 1675 /* 1676 * If we (still) have a compiled-in default, try it. 1677 */ 1678 if (ctrootdevname != NULL) 1679 if (!vfs_mountroot_try(ctrootdevname)) 1680 goto mounted; 1681 /* 1682 * Everything so far has failed, prompt on the console if we haven't 1683 * already tried that. 1684 */ 1685 if (!asked) 1686 if (!vfs_mountroot_ask()) 1687 goto mounted; 1688 1689 panic("Root mount failed, startup aborted."); 1690 1691mounted: 1692 root_mount_done(); 1693} 1694 1695/* 1696 * Mount (mountfrom) as the root filesystem. 1697 */ 1698static int 1699vfs_mountroot_try(const char *mountfrom) 1700{ 1701 struct mount *mp; 1702 char *vfsname, *path; 1703 time_t timebase; 1704 int error; 1705 char patt[32]; 1706 1707 vfsname = NULL; 1708 path = NULL; 1709 mp = NULL; 1710 error = EINVAL; 1711 1712 if (mountfrom == NULL) 1713 return (error); /* don't complain */ 1714 printf("Trying to mount root from %s\n", mountfrom); 1715 1716 /* parse vfs name and path */ 1717 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1718 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1719 vfsname[0] = path[0] = 0; 1720 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1721 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1722 goto out; 1723 1724 if (path[0] == '\0') 1725 strcpy(path, ROOTNAME); 1726 1727 error = kernel_vmount( 1728 MNT_RDONLY | MNT_ROOTFS, 1729 "fstype", vfsname, 1730 "fspath", "/", 1731 "from", path, 1732 NULL); 1733 if (error == 0) { 1734 /* 1735 * We mount devfs prior to mounting the / FS, so the first 1736 * entry will typically be devfs. 1737 */ 1738 mp = TAILQ_FIRST(&mountlist); 1739 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1740 1741 /* 1742 * Iterate over all currently mounted file systems and use 1743 * the time stamp found to check and/or initialize the RTC. 1744 * Typically devfs has no time stamp and the only other FS 1745 * is the actual / FS. 1746 * Call inittodr() only once and pass it the largest of the 1747 * timestamps we encounter. 1748 */ 1749 timebase = 0; 1750 do { 1751 if (mp->mnt_time > timebase) 1752 timebase = mp->mnt_time; 1753 mp = TAILQ_NEXT(mp, mnt_list); 1754 } while (mp != NULL); 1755 inittodr(timebase); 1756 1757 devfs_fixup(curthread); 1758 } 1759out: 1760 free(path, M_MOUNT); 1761 free(vfsname, M_MOUNT); 1762 return (error); 1763} 1764 1765/* 1766 * --------------------------------------------------------------------- 1767 * Interactive root filesystem selection code. 1768 */ 1769 1770static int 1771vfs_mountroot_ask(void) 1772{ 1773 char name[128]; 1774 1775 for(;;) { 1776 printf("\nManual root filesystem specification:\n"); 1777 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1778#if defined(__amd64__) || defined(__i386__) || defined(__ia64__) 1779 printf(" eg. ufs:da0s1a\n"); 1780#else 1781 printf(" eg. ufs:/dev/da0a\n"); 1782#endif 1783 printf(" ? List valid disk boot devices\n"); 1784 printf(" <empty line> Abort manual input\n"); 1785 printf("\nmountroot> "); 1786 gets(name, sizeof(name), 1); 1787 if (name[0] == '\0') 1788 return (1); 1789 if (name[0] == '?') { 1790 printf("\nList of GEOM managed disk devices:\n "); 1791 g_dev_print(); 1792 continue; 1793 } 1794 if (!vfs_mountroot_try(name)) 1795 return (0); 1796 } 1797} 1798 1799/* 1800 * --------------------------------------------------------------------- 1801 * Functions for querying mount options/arguments from filesystems. 1802 */ 1803 1804/* 1805 * Check that no unknown options are given 1806 */ 1807int 1808vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1809{ 1810 struct vfsopt *opt; 1811 char errmsg[255]; 1812 const char **t, *p, *q; 1813 int ret = 0; 1814 1815 TAILQ_FOREACH(opt, opts, link) { 1816 p = opt->name; 1817 q = NULL; 1818 if (p[0] == 'n' && p[1] == 'o') 1819 q = p + 2; 1820 for(t = global_opts; *t != NULL; t++) { 1821 if (strcmp(*t, p) == 0) 1822 break; 1823 if (q != NULL) { 1824 if (strcmp(*t, q) == 0) 1825 break; 1826 } 1827 } 1828 if (*t != NULL) 1829 continue; 1830 for(t = legal; *t != NULL; t++) { 1831 if (strcmp(*t, p) == 0) 1832 break; 1833 if (q != NULL) { 1834 if (strcmp(*t, q) == 0) 1835 break; 1836 } 1837 } 1838 if (*t != NULL) 1839 continue; 1840 sprintf(errmsg, "mount option <%s> is unknown", p); 1841 printf("%s\n", errmsg); 1842 ret = EINVAL; 1843 } 1844 if (ret != 0) { 1845 TAILQ_FOREACH(opt, opts, link) { 1846 if (strcmp(opt->name, "errmsg") == 0) { 1847 strncpy((char *)opt->value, errmsg, opt->len); 1848 } 1849 } 1850 } 1851 return (ret); 1852} 1853 1854/* 1855 * Get a mount option by its name. 1856 * 1857 * Return 0 if the option was found, ENOENT otherwise. 1858 * If len is non-NULL it will be filled with the length 1859 * of the option. If buf is non-NULL, it will be filled 1860 * with the address of the option. 1861 */ 1862int 1863vfs_getopt(opts, name, buf, len) 1864 struct vfsoptlist *opts; 1865 const char *name; 1866 void **buf; 1867 int *len; 1868{ 1869 struct vfsopt *opt; 1870 1871 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1872 1873 TAILQ_FOREACH(opt, opts, link) { 1874 if (strcmp(name, opt->name) == 0) { 1875 if (len != NULL) 1876 *len = opt->len; 1877 if (buf != NULL) 1878 *buf = opt->value; 1879 return (0); 1880 } 1881 } 1882 return (ENOENT); 1883} 1884 1885static int 1886vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1887{ 1888 struct vfsopt *opt; 1889 int i; 1890 1891 if (opts == NULL) 1892 return (-1); 1893 1894 i = 0; 1895 TAILQ_FOREACH(opt, opts, link) { 1896 if (strcmp(name, opt->name) == 0) 1897 return (i); 1898 ++i; 1899 } 1900 return (-1); 1901} 1902 1903char * 1904vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1905{ 1906 struct vfsopt *opt; 1907 1908 *error = 0; 1909 TAILQ_FOREACH(opt, opts, link) { 1910 if (strcmp(name, opt->name) != 0) 1911 continue; 1912 if (((char *)opt->value)[opt->len - 1] != '\0') { 1913 *error = EINVAL; 1914 return (NULL); 1915 } 1916 return (opt->value); 1917 } 1918 *error = ENOENT; 1919 return (NULL); 1920} 1921 1922int 1923vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1924{ 1925 struct vfsopt *opt; 1926 1927 TAILQ_FOREACH(opt, opts, link) { 1928 if (strcmp(name, opt->name) == 0) { 1929 if (w != NULL) 1930 *w |= val; 1931 return (1); 1932 } 1933 } 1934 if (w != NULL) 1935 *w &= ~val; 1936 return (0); 1937} 1938 1939int 1940vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1941{ 1942 va_list ap; 1943 struct vfsopt *opt; 1944 int ret; 1945 1946 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1947 1948 TAILQ_FOREACH(opt, opts, link) { 1949 if (strcmp(name, opt->name) != 0) 1950 continue; 1951 if (opt->len == 0 || opt->value == NULL) 1952 return (0); 1953 if (((char *)opt->value)[opt->len - 1] != '\0') 1954 return (0); 1955 va_start(ap, fmt); 1956 ret = vsscanf(opt->value, fmt, ap); 1957 va_end(ap); 1958 return (ret); 1959 } 1960 return (0); 1961} 1962 1963/* 1964 * Find and copy a mount option. 1965 * 1966 * The size of the buffer has to be specified 1967 * in len, if it is not the same length as the 1968 * mount option, EINVAL is returned. 1969 * Returns ENOENT if the option is not found. 1970 */ 1971int 1972vfs_copyopt(opts, name, dest, len) 1973 struct vfsoptlist *opts; 1974 const char *name; 1975 void *dest; 1976 int len; 1977{ 1978 struct vfsopt *opt; 1979 1980 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1981 1982 TAILQ_FOREACH(opt, opts, link) { 1983 if (strcmp(name, opt->name) == 0) { 1984 if (len != opt->len) 1985 return (EINVAL); 1986 bcopy(opt->value, dest, opt->len); 1987 return (0); 1988 } 1989 } 1990 return (ENOENT); 1991} 1992 1993/* 1994 * This is a helper function for filesystems to traverse their 1995 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1996 */ 1997 1998struct vnode * 1999__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2000{ 2001 struct vnode *vp; 2002 2003 mtx_assert(MNT_MTX(mp), MA_OWNED); 2004 2005 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2006 if ((*mvp)->v_yield++ == 500) { 2007 MNT_IUNLOCK(mp); 2008 (*mvp)->v_yield = 0; 2009 uio_yield(); 2010 MNT_ILOCK(mp); 2011 } 2012 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2013 while (vp != NULL && vp->v_type == VMARKER) 2014 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2015 2016 /* Check if we are done */ 2017 if (vp == NULL) { 2018 __mnt_vnode_markerfree(mvp, mp); 2019 return (NULL); 2020 } 2021 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2022 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2023 return (vp); 2024} 2025 2026struct vnode * 2027__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2028{ 2029 struct vnode *vp; 2030 2031 mtx_assert(MNT_MTX(mp), MA_OWNED); 2032 2033 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2034 while (vp != NULL && vp->v_type == VMARKER) 2035 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2036 2037 /* Check if we are done */ 2038 if (vp == NULL) { 2039 *mvp = NULL; 2040 return (NULL); 2041 } 2042 mp->mnt_holdcnt++; 2043 MNT_IUNLOCK(mp); 2044 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2045 M_VNODE_MARKER, 2046 M_WAITOK | M_ZERO); 2047 MNT_ILOCK(mp); 2048 (*mvp)->v_type = VMARKER; 2049 2050 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2051 while (vp != NULL && vp->v_type == VMARKER) 2052 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2053 2054 /* Check if we are done */ 2055 if (vp == NULL) { 2056 MNT_IUNLOCK(mp); 2057 free(*mvp, M_VNODE_MARKER); 2058 MNT_ILOCK(mp); 2059 *mvp = NULL; 2060 mp->mnt_holdcnt--; 2061 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2062 wakeup(&mp->mnt_holdcnt); 2063 return (NULL); 2064 } 2065 mp->mnt_markercnt++; 2066 (*mvp)->v_mount = mp; 2067 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2068 return (vp); 2069} 2070 2071 2072void 2073__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2074{ 2075 2076 if (*mvp == NULL) 2077 return; 2078 2079 mtx_assert(MNT_MTX(mp), MA_OWNED); 2080 2081 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2082 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2083 MNT_IUNLOCK(mp); 2084 free(*mvp, M_VNODE_MARKER); 2085 MNT_ILOCK(mp); 2086 *mvp = NULL; 2087 2088 mp->mnt_markercnt--; 2089 mp->mnt_holdcnt--; 2090 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2091 wakeup(&mp->mnt_holdcnt); 2092} 2093 2094 2095int 2096__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 2097{ 2098 int error; 2099 2100 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 2101 if (sbp != &mp->mnt_stat) 2102 *sbp = mp->mnt_stat; 2103 return (error); 2104} 2105 2106void 2107vfs_mountedfrom(struct mount *mp, const char *from) 2108{ 2109 2110 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2111 strlcpy(mp->mnt_stat.f_mntfromname, from, 2112 sizeof mp->mnt_stat.f_mntfromname); 2113} 2114 2115/* 2116 * --------------------------------------------------------------------- 2117 * This is the api for building mount args and mounting filesystems from 2118 * inside the kernel. 2119 * 2120 * The API works by accumulation of individual args. First error is 2121 * latched. 2122 * 2123 * XXX: should be documented in new manpage kernel_mount(9) 2124 */ 2125 2126/* A memory allocation which must be freed when we are done */ 2127struct mntaarg { 2128 SLIST_ENTRY(mntaarg) next; 2129}; 2130 2131/* The header for the mount arguments */ 2132struct mntarg { 2133 struct iovec *v; 2134 int len; 2135 int error; 2136 SLIST_HEAD(, mntaarg) list; 2137}; 2138 2139/* 2140 * Add a boolean argument. 2141 * 2142 * flag is the boolean value. 2143 * name must start with "no". 2144 */ 2145struct mntarg * 2146mount_argb(struct mntarg *ma, int flag, const char *name) 2147{ 2148 2149 KASSERT(name[0] == 'n' && name[1] == 'o', 2150 ("mount_argb(...,%s): name must start with 'no'", name)); 2151 2152 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2153} 2154 2155/* 2156 * Add an argument printf style 2157 */ 2158struct mntarg * 2159mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2160{ 2161 va_list ap; 2162 struct mntaarg *maa; 2163 struct sbuf *sb; 2164 int len; 2165 2166 if (ma == NULL) { 2167 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2168 SLIST_INIT(&ma->list); 2169 } 2170 if (ma->error) 2171 return (ma); 2172 2173 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2174 M_MOUNT, M_WAITOK); 2175 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2176 ma->v[ma->len].iov_len = strlen(name) + 1; 2177 ma->len++; 2178 2179 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND); 2180 va_start(ap, fmt); 2181 sbuf_vprintf(sb, fmt, ap); 2182 va_end(ap); 2183 sbuf_finish(sb); 2184 len = sbuf_len(sb) + 1; 2185 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2186 SLIST_INSERT_HEAD(&ma->list, maa, next); 2187 bcopy(sbuf_data(sb), maa + 1, len); 2188 sbuf_delete(sb); 2189 2190 ma->v[ma->len].iov_base = maa + 1; 2191 ma->v[ma->len].iov_len = len; 2192 ma->len++; 2193 2194 return (ma); 2195} 2196 2197/* 2198 * Add an argument which is a userland string. 2199 */ 2200struct mntarg * 2201mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2202{ 2203 struct mntaarg *maa; 2204 char *tbuf; 2205 2206 if (val == NULL) 2207 return (ma); 2208 if (ma == NULL) { 2209 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2210 SLIST_INIT(&ma->list); 2211 } 2212 if (ma->error) 2213 return (ma); 2214 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2215 SLIST_INSERT_HEAD(&ma->list, maa, next); 2216 tbuf = (void *)(maa + 1); 2217 ma->error = copyinstr(val, tbuf, len, NULL); 2218 return (mount_arg(ma, name, tbuf, -1)); 2219} 2220 2221/* 2222 * Plain argument. 2223 * 2224 * If length is -1, treat value as a C string. 2225 */ 2226struct mntarg * 2227mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2228{ 2229 2230 if (ma == NULL) { 2231 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2232 SLIST_INIT(&ma->list); 2233 } 2234 if (ma->error) 2235 return (ma); 2236 2237 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2238 M_MOUNT, M_WAITOK); 2239 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2240 ma->v[ma->len].iov_len = strlen(name) + 1; 2241 ma->len++; 2242 2243 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2244 if (len < 0) 2245 ma->v[ma->len].iov_len = strlen(val) + 1; 2246 else 2247 ma->v[ma->len].iov_len = len; 2248 ma->len++; 2249 return (ma); 2250} 2251 2252/* 2253 * Free a mntarg structure 2254 */ 2255static void 2256free_mntarg(struct mntarg *ma) 2257{ 2258 struct mntaarg *maa; 2259 2260 while (!SLIST_EMPTY(&ma->list)) { 2261 maa = SLIST_FIRST(&ma->list); 2262 SLIST_REMOVE_HEAD(&ma->list, next); 2263 free(maa, M_MOUNT); 2264 } 2265 free(ma->v, M_MOUNT); 2266 free(ma, M_MOUNT); 2267} 2268 2269/* 2270 * Mount a filesystem 2271 */ 2272int 2273kernel_mount(struct mntarg *ma, int flags) 2274{ 2275 struct uio auio; 2276 int error; 2277 2278 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2279 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2280 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2281 2282 auio.uio_iov = ma->v; 2283 auio.uio_iovcnt = ma->len; 2284 auio.uio_segflg = UIO_SYSSPACE; 2285 2286 error = ma->error; 2287 if (!error) 2288 error = vfs_donmount(curthread, flags, &auio); 2289 free_mntarg(ma); 2290 return (error); 2291} 2292 2293/* 2294 * A printflike function to mount a filesystem. 2295 */ 2296int 2297kernel_vmount(int flags, ...) 2298{ 2299 struct mntarg *ma = NULL; 2300 va_list ap; 2301 const char *cp; 2302 const void *vp; 2303 int error; 2304 2305 va_start(ap, flags); 2306 for (;;) { 2307 cp = va_arg(ap, const char *); 2308 if (cp == NULL) 2309 break; 2310 vp = va_arg(ap, const void *); 2311 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2312 } 2313 va_end(ap); 2314 2315 error = kernel_mount(ma, flags); 2316 return (error); 2317} 2318