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