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