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