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/fcntl.h> 43 #include <sys/jail.h> 44 #include <sys/kernel.h> 45 #include <sys/libkern.h> 46 #include <sys/malloc.h> 47 #include <sys/mount.h> 48 #include <sys/mutex.h> 49 #include <sys/namei.h> 50 #include <sys/priv.h> 51 #include <sys/proc.h> 52 #include <sys/filedesc.h> 53 #include <sys/reboot.h> 54 #include <sys/syscallsubr.h> 55 #include <sys/sysproto.h> 56 #include <sys/sx.h> 57 #include <sys/sysctl.h> 58 #include <sys/sysent.h> 59 #include <sys/systm.h> 60 #include <sys/vnode.h> 61 #include <vm/uma.h> 62 63 #include <geom/geom.h> 64 65 #include <machine/stdarg.h> 66 67 #include <security/audit/audit.h> 68 #include <security/mac/mac_framework.h> 69 70 #include "opt_rootdevname.h" 71 #include "opt_mac.h" 72 73 #define ROOTNAME "root_device" 74 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 75 76 static int vfs_domount(struct thread *td, const char *fstype, 77 char *fspath, int fsflags, void *fsdata); 78 static int vfs_mountroot_ask(void); 79 static int vfs_mountroot_try(const char *mountfrom); 80 static void free_mntarg(struct mntarg *ma); 81 static int vfs_getopt_pos(struct vfsoptlist *opts, const char *name); 82 83 static int usermount = 0; 84 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 85 "Unprivileged users may mount and unmount file systems"); 86 87 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 88 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 89 static uma_zone_t mount_zone; 90 91 /* List of mounted filesystems. */ 92 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 93 94 /* For any iteration/modification of mountlist */ 95 struct mtx mountlist_mtx; 96 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 97 98 TAILQ_HEAD(vfsoptlist, vfsopt); 99 struct vfsopt { 100 TAILQ_ENTRY(vfsopt) link; 101 char *name; 102 void *value; 103 int len; 104 }; 105 106 /* 107 * The vnode of the system's root (/ in the filesystem, without chroot 108 * active.) 109 */ 110 struct vnode *rootvnode; 111 112 /* 113 * The root filesystem is detailed in the kernel environment variable 114 * vfs.root.mountfrom, which is expected to be in the general format 115 * 116 * <vfsname>:[<path>] 117 * vfsname := the name of a VFS known to the kernel and capable 118 * of being mounted as root 119 * path := disk device name or other data used by the filesystem 120 * to locate its physical store 121 */ 122 123 /* 124 * Global opts, taken by all filesystems 125 */ 126 static const char *global_opts[] = { 127 "errmsg", 128 "fstype", 129 "fspath", 130 "ro", 131 "rw", 132 "nosuid", 133 "noexec", 134 NULL 135 }; 136 137 /* 138 * The root specifiers we will try if RB_CDROM is specified. 139 */ 140 static char *cdrom_rootdevnames[] = { 141 "cd9660:cd0", 142 "cd9660:acd0", 143 NULL 144 }; 145 146 /* legacy find-root code */ 147 char *rootdevnames[2] = {NULL, NULL}; 148 #ifndef ROOTDEVNAME 149 # define ROOTDEVNAME NULL 150 #endif 151 static const char *ctrootdevname = ROOTDEVNAME; 152 153 /* 154 * --------------------------------------------------------------------- 155 * Functions for building and sanitizing the mount options 156 */ 157 158 /* Remove one mount option. */ 159 static void 160 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 161 { 162 163 TAILQ_REMOVE(opts, opt, link); 164 free(opt->name, M_MOUNT); 165 if (opt->value != NULL) 166 free(opt->value, M_MOUNT); 167 #ifdef INVARIANTS 168 else if (opt->len != 0) 169 panic("%s: mount option with NULL value but length != 0", 170 __func__); 171 #endif 172 free(opt, M_MOUNT); 173 } 174 175 /* Release all resources related to the mount options. */ 176 void 177 vfs_freeopts(struct vfsoptlist *opts) 178 { 179 struct vfsopt *opt; 180 181 while (!TAILQ_EMPTY(opts)) { 182 opt = TAILQ_FIRST(opts); 183 vfs_freeopt(opts, opt); 184 } 185 free(opts, M_MOUNT); 186 } 187 188 void 189 vfs_deleteopt(struct vfsoptlist *opts, const char *name) 190 { 191 struct vfsopt *opt, *temp; 192 193 if (opts == NULL) 194 return; 195 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 196 if (strcmp(opt->name, name) == 0) 197 vfs_freeopt(opts, opt); 198 } 199 } 200 201 /* 202 * Check if options are equal (with or without the "no" prefix). 203 */ 204 static int 205 vfs_equalopts(const char *opt1, const char *opt2) 206 { 207 208 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 209 if (strcmp(opt1, opt2) == 0) 210 return (1); 211 /* "noopt" vs. "opt" */ 212 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 213 return (1); 214 /* "opt" vs. "noopt" */ 215 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 216 return (1); 217 return (0); 218 } 219 220 /* 221 * If a mount option is specified several times, 222 * (with or without the "no" prefix) only keep 223 * the last occurence of it. 224 */ 225 static void 226 vfs_sanitizeopts(struct vfsoptlist *opts) 227 { 228 struct vfsopt *opt, *opt2, *tmp; 229 230 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 231 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 232 while (opt2 != NULL) { 233 if (vfs_equalopts(opt->name, opt2->name)) { 234 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 235 vfs_freeopt(opts, opt2); 236 opt2 = tmp; 237 } else { 238 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 239 } 240 } 241 } 242 } 243 244 /* 245 * Build a linked list of mount options from a struct uio. 246 */ 247 static int 248 vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 249 { 250 struct vfsoptlist *opts; 251 struct vfsopt *opt; 252 size_t memused; 253 unsigned int i, iovcnt; 254 int error, namelen, optlen; 255 256 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 257 TAILQ_INIT(opts); 258 memused = 0; 259 iovcnt = auio->uio_iovcnt; 260 for (i = 0; i < iovcnt; i += 2) { 261 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 262 namelen = auio->uio_iov[i].iov_len; 263 optlen = auio->uio_iov[i + 1].iov_len; 264 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 265 opt->value = NULL; 266 opt->len = 0; 267 268 /* 269 * Do this early, so jumps to "bad" will free the current 270 * option. 271 */ 272 TAILQ_INSERT_TAIL(opts, opt, link); 273 memused += sizeof(struct vfsopt) + optlen + namelen; 274 275 /* 276 * Avoid consuming too much memory, and attempts to overflow 277 * memused. 278 */ 279 if (memused > VFS_MOUNTARG_SIZE_MAX || 280 optlen > VFS_MOUNTARG_SIZE_MAX || 281 namelen > VFS_MOUNTARG_SIZE_MAX) { 282 error = EINVAL; 283 goto bad; 284 } 285 286 if (auio->uio_segflg == UIO_SYSSPACE) { 287 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 288 } else { 289 error = copyin(auio->uio_iov[i].iov_base, opt->name, 290 namelen); 291 if (error) 292 goto bad; 293 } 294 /* Ensure names are null-terminated strings. */ 295 if (opt->name[namelen - 1] != '\0') { 296 error = EINVAL; 297 goto bad; 298 } 299 if (optlen != 0) { 300 opt->len = optlen; 301 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 302 if (auio->uio_segflg == UIO_SYSSPACE) { 303 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 304 optlen); 305 } else { 306 error = copyin(auio->uio_iov[i + 1].iov_base, 307 opt->value, optlen); 308 if (error) 309 goto bad; 310 } 311 } 312 } 313 vfs_sanitizeopts(opts); 314 *options = opts; 315 return (0); 316 bad: 317 vfs_freeopts(opts); 318 return (error); 319 } 320 321 /* 322 * Merge the old mount options with the new ones passed 323 * in the MNT_UPDATE case. 324 * 325 * XXX This function will keep a "nofoo" option in the 326 * new options if there is no matching "foo" option 327 * to be cancelled in the old options. This is a bug 328 * if the option's canonical name is "foo". E.g., "noro" 329 * shouldn't end up in the mount point's active options, 330 * but it can. 331 */ 332 static void 333 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 334 { 335 struct vfsopt *opt, *opt2, *new; 336 337 TAILQ_FOREACH(opt, opts, link) { 338 /* 339 * Check that this option hasn't been redefined 340 * nor cancelled with a "no" mount option. 341 */ 342 opt2 = TAILQ_FIRST(toopts); 343 while (opt2 != NULL) { 344 if (strcmp(opt2->name, opt->name) == 0) 345 goto next; 346 if (strncmp(opt2->name, "no", 2) == 0 && 347 strcmp(opt2->name + 2, opt->name) == 0) { 348 vfs_freeopt(toopts, opt2); 349 goto next; 350 } 351 opt2 = TAILQ_NEXT(opt2, link); 352 } 353 /* We want this option, duplicate it. */ 354 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 355 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 356 strcpy(new->name, opt->name); 357 if (opt->len != 0) { 358 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 359 bcopy(opt->value, new->value, opt->len); 360 } else { 361 new->value = NULL; 362 } 363 new->len = opt->len; 364 TAILQ_INSERT_TAIL(toopts, new, link); 365 next: 366 continue; 367 } 368 } 369 370 /* 371 * Mount a filesystem. 372 */ 373 int 374 nmount(td, uap) 375 struct thread *td; 376 struct nmount_args /* { 377 struct iovec *iovp; 378 unsigned int iovcnt; 379 int flags; 380 } */ *uap; 381 { 382 struct uio *auio; 383 struct iovec *iov; 384 unsigned int i; 385 int error; 386 u_int iovcnt; 387 388 AUDIT_ARG(fflags, uap->flags); 389 390 /* 391 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 392 * userspace to set this flag, but we must filter it out if we want 393 * MNT_UPDATE on the root file system to work. 394 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 395 */ 396 uap->flags &= ~MNT_ROOTFS; 397 398 iovcnt = uap->iovcnt; 399 /* 400 * Check that we have an even number of iovec's 401 * and that we have at least two options. 402 */ 403 if ((iovcnt & 1) || (iovcnt < 4)) 404 return (EINVAL); 405 406 error = copyinuio(uap->iovp, iovcnt, &auio); 407 if (error) 408 return (error); 409 iov = auio->uio_iov; 410 for (i = 0; i < iovcnt; i++) { 411 if (iov->iov_len > MMAXOPTIONLEN) { 412 free(auio, M_IOV); 413 return (EINVAL); 414 } 415 iov++; 416 } 417 error = vfs_donmount(td, uap->flags, auio); 418 419 free(auio, M_IOV); 420 return (error); 421 } 422 423 /* 424 * --------------------------------------------------------------------- 425 * Various utility functions 426 */ 427 428 void 429 vfs_ref(struct mount *mp) 430 { 431 432 MNT_ILOCK(mp); 433 MNT_REF(mp); 434 MNT_IUNLOCK(mp); 435 } 436 437 void 438 vfs_rel(struct mount *mp) 439 { 440 441 MNT_ILOCK(mp); 442 MNT_REL(mp); 443 MNT_IUNLOCK(mp); 444 } 445 446 static int 447 mount_init(void *mem, int size, int flags) 448 { 449 struct mount *mp; 450 451 mp = (struct mount *)mem; 452 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 453 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 454 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 455 return (0); 456 } 457 458 static void 459 mount_fini(void *mem, int size) 460 { 461 struct mount *mp; 462 463 mp = (struct mount *)mem; 464 lockdestroy(&mp->mnt_explock); 465 lockdestroy(&mp->mnt_lock); 466 mtx_destroy(&mp->mnt_mtx); 467 } 468 469 /* 470 * Allocate and initialize the mount point struct. 471 */ 472 struct mount * 473 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 474 struct ucred *cred) 475 { 476 struct mount *mp; 477 478 mp = uma_zalloc(mount_zone, M_WAITOK); 479 bzero(&mp->mnt_startzero, 480 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 481 TAILQ_INIT(&mp->mnt_nvnodelist); 482 mp->mnt_nvnodelistsize = 0; 483 mp->mnt_ref = 0; 484 (void) vfs_busy(mp, LK_NOWAIT, 0); 485 mp->mnt_op = vfsp->vfc_vfsops; 486 mp->mnt_vfc = vfsp; 487 vfsp->vfc_refcount++; /* XXX Unlocked */ 488 mp->mnt_stat.f_type = vfsp->vfc_typenum; 489 mp->mnt_gen++; 490 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 491 mp->mnt_vnodecovered = vp; 492 mp->mnt_cred = crdup(cred); 493 mp->mnt_stat.f_owner = cred->cr_uid; 494 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 495 mp->mnt_iosize_max = DFLTPHYS; 496 #ifdef MAC 497 mac_mount_init(mp); 498 mac_mount_create(cred, mp); 499 #endif 500 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 501 return (mp); 502 } 503 504 /* 505 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 506 */ 507 void 508 vfs_mount_destroy(struct mount *mp) 509 { 510 int i; 511 512 MNT_ILOCK(mp); 513 for (i = 0; mp->mnt_ref && i < 3; i++) 514 msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz); 515 /* 516 * This will always cause a 3 second delay in rebooting due to 517 * refs on the root mountpoint that never go away. Most of these 518 * are held by init which never exits. 519 */ 520 if (i == 3 && (!rebooting || bootverbose)) 521 printf("Mount point %s had %d dangling refs\n", 522 mp->mnt_stat.f_mntonname, mp->mnt_ref); 523 if (mp->mnt_holdcnt != 0) { 524 printf("Waiting for mount point to be unheld\n"); 525 while (mp->mnt_holdcnt != 0) { 526 mp->mnt_holdcntwaiters++; 527 msleep(&mp->mnt_holdcnt, MNT_MTX(mp), 528 PZERO, "mntdestroy", 0); 529 mp->mnt_holdcntwaiters--; 530 } 531 printf("mount point unheld\n"); 532 } 533 if (mp->mnt_writeopcount > 0) { 534 printf("Waiting for mount point write ops\n"); 535 while (mp->mnt_writeopcount > 0) { 536 mp->mnt_kern_flag |= MNTK_SUSPEND; 537 msleep(&mp->mnt_writeopcount, 538 MNT_MTX(mp), 539 PZERO, "mntdestroy2", 0); 540 } 541 printf("mount point write ops completed\n"); 542 } 543 if (mp->mnt_secondary_writes > 0) { 544 printf("Waiting for mount point secondary write ops\n"); 545 while (mp->mnt_secondary_writes > 0) { 546 mp->mnt_kern_flag |= MNTK_SUSPEND; 547 msleep(&mp->mnt_secondary_writes, 548 MNT_MTX(mp), 549 PZERO, "mntdestroy3", 0); 550 } 551 printf("mount point secondary write ops completed\n"); 552 } 553 MNT_IUNLOCK(mp); 554 mp->mnt_vfc->vfc_refcount--; 555 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 556 struct vnode *vp; 557 558 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 559 vprint("", vp); 560 panic("unmount: dangling vnode"); 561 } 562 MNT_ILOCK(mp); 563 if (mp->mnt_kern_flag & MNTK_MWAIT) 564 wakeup(mp); 565 if (mp->mnt_writeopcount != 0) 566 panic("vfs_mount_destroy: nonzero writeopcount"); 567 if (mp->mnt_secondary_writes != 0) 568 panic("vfs_mount_destroy: nonzero secondary_writes"); 569 if (mp->mnt_nvnodelistsize != 0) 570 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 571 mp->mnt_writeopcount = -1000; 572 mp->mnt_nvnodelistsize = -1000; 573 mp->mnt_secondary_writes = -1000; 574 MNT_IUNLOCK(mp); 575 #ifdef MAC 576 mac_mount_destroy(mp); 577 #endif 578 if (mp->mnt_opt != NULL) 579 vfs_freeopts(mp->mnt_opt); 580 crfree(mp->mnt_cred); 581 uma_zfree(mount_zone, mp); 582 } 583 584 int 585 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 586 { 587 struct vfsoptlist *optlist; 588 struct vfsopt *opt, *noro_opt, *tmp_opt; 589 char *fstype, *fspath, *errmsg; 590 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 591 int has_rw, has_noro; 592 593 errmsg = fspath = NULL; 594 errmsg_len = has_noro = has_rw = fspathlen = 0; 595 errmsg_pos = -1; 596 597 error = vfs_buildopts(fsoptions, &optlist); 598 if (error) 599 return (error); 600 601 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 602 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 603 604 /* 605 * We need these two options before the others, 606 * and they are mandatory for any filesystem. 607 * Ensure they are NUL terminated as well. 608 */ 609 fstypelen = 0; 610 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 611 if (error || fstype[fstypelen - 1] != '\0') { 612 error = EINVAL; 613 if (errmsg != NULL) 614 strncpy(errmsg, "Invalid fstype", errmsg_len); 615 goto bail; 616 } 617 fspathlen = 0; 618 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 619 if (error || fspath[fspathlen - 1] != '\0') { 620 error = EINVAL; 621 if (errmsg != NULL) 622 strncpy(errmsg, "Invalid fspath", errmsg_len); 623 goto bail; 624 } 625 626 /* 627 * We need to see if we have the "update" option 628 * before we call vfs_domount(), since vfs_domount() has special 629 * logic based on MNT_UPDATE. This is very important 630 * when we want to update the root filesystem. 631 */ 632 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 633 if (strcmp(opt->name, "update") == 0) { 634 fsflags |= MNT_UPDATE; 635 vfs_freeopt(optlist, opt); 636 } 637 else if (strcmp(opt->name, "async") == 0) 638 fsflags |= MNT_ASYNC; 639 else if (strcmp(opt->name, "force") == 0) { 640 fsflags |= MNT_FORCE; 641 vfs_freeopt(optlist, opt); 642 } 643 else if (strcmp(opt->name, "reload") == 0) { 644 fsflags |= MNT_RELOAD; 645 vfs_freeopt(optlist, opt); 646 } 647 else if (strcmp(opt->name, "multilabel") == 0) 648 fsflags |= MNT_MULTILABEL; 649 else if (strcmp(opt->name, "noasync") == 0) 650 fsflags &= ~MNT_ASYNC; 651 else if (strcmp(opt->name, "noatime") == 0) 652 fsflags |= MNT_NOATIME; 653 else if (strcmp(opt->name, "atime") == 0) { 654 free(opt->name, M_MOUNT); 655 opt->name = strdup("nonoatime", M_MOUNT); 656 } 657 else if (strcmp(opt->name, "noclusterr") == 0) 658 fsflags |= MNT_NOCLUSTERR; 659 else if (strcmp(opt->name, "clusterr") == 0) { 660 free(opt->name, M_MOUNT); 661 opt->name = strdup("nonoclusterr", M_MOUNT); 662 } 663 else if (strcmp(opt->name, "noclusterw") == 0) 664 fsflags |= MNT_NOCLUSTERW; 665 else if (strcmp(opt->name, "clusterw") == 0) { 666 free(opt->name, M_MOUNT); 667 opt->name = strdup("nonoclusterw", M_MOUNT); 668 } 669 else if (strcmp(opt->name, "noexec") == 0) 670 fsflags |= MNT_NOEXEC; 671 else if (strcmp(opt->name, "exec") == 0) { 672 free(opt->name, M_MOUNT); 673 opt->name = strdup("nonoexec", M_MOUNT); 674 } 675 else if (strcmp(opt->name, "nosuid") == 0) 676 fsflags |= MNT_NOSUID; 677 else if (strcmp(opt->name, "suid") == 0) { 678 free(opt->name, M_MOUNT); 679 opt->name = strdup("nonosuid", M_MOUNT); 680 } 681 else if (strcmp(opt->name, "nosymfollow") == 0) 682 fsflags |= MNT_NOSYMFOLLOW; 683 else if (strcmp(opt->name, "symfollow") == 0) { 684 free(opt->name, M_MOUNT); 685 opt->name = strdup("nonosymfollow", M_MOUNT); 686 } 687 else if (strcmp(opt->name, "noro") == 0) { 688 fsflags &= ~MNT_RDONLY; 689 has_noro = 1; 690 } 691 else if (strcmp(opt->name, "rw") == 0) { 692 fsflags &= ~MNT_RDONLY; 693 has_rw = 1; 694 } 695 else if (strcmp(opt->name, "ro") == 0) 696 fsflags |= MNT_RDONLY; 697 else if (strcmp(opt->name, "rdonly") == 0) { 698 free(opt->name, M_MOUNT); 699 opt->name = strdup("ro", M_MOUNT); 700 fsflags |= MNT_RDONLY; 701 } 702 else if (strcmp(opt->name, "suiddir") == 0) 703 fsflags |= MNT_SUIDDIR; 704 else if (strcmp(opt->name, "sync") == 0) 705 fsflags |= MNT_SYNCHRONOUS; 706 else if (strcmp(opt->name, "union") == 0) 707 fsflags |= MNT_UNION; 708 } 709 710 /* 711 * If "rw" was specified as a mount option, and we 712 * are trying to update a mount-point from "ro" to "rw", 713 * we need a mount option "noro", since in vfs_mergeopts(), 714 * "noro" will cancel "ro", but "rw" will not do anything. 715 */ 716 if (has_rw && !has_noro) { 717 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 718 noro_opt->name = strdup("noro", M_MOUNT); 719 noro_opt->value = NULL; 720 noro_opt->len = 0; 721 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 722 } 723 724 /* 725 * Be ultra-paranoid about making sure the type and fspath 726 * variables will fit in our mp buffers, including the 727 * terminating NUL. 728 */ 729 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 730 error = ENAMETOOLONG; 731 goto bail; 732 } 733 734 mtx_lock(&Giant); 735 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 736 mtx_unlock(&Giant); 737 bail: 738 /* copyout the errmsg */ 739 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 740 && errmsg_len > 0 && errmsg != NULL) { 741 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 742 bcopy(errmsg, 743 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 744 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 745 } else { 746 copyout(errmsg, 747 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 748 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 749 } 750 } 751 752 if (error != 0) 753 vfs_freeopts(optlist); 754 return (error); 755 } 756 757 /* 758 * Old mount API. 759 */ 760 #ifndef _SYS_SYSPROTO_H_ 761 struct mount_args { 762 char *type; 763 char *path; 764 int flags; 765 caddr_t data; 766 }; 767 #endif 768 /* ARGSUSED */ 769 int 770 mount(td, uap) 771 struct thread *td; 772 struct mount_args /* { 773 char *type; 774 char *path; 775 int flags; 776 caddr_t data; 777 } */ *uap; 778 { 779 char *fstype; 780 struct vfsconf *vfsp = NULL; 781 struct mntarg *ma = NULL; 782 int error; 783 784 AUDIT_ARG(fflags, uap->flags); 785 786 /* 787 * Filter out MNT_ROOTFS. We do not want clients of mount() in 788 * userspace to set this flag, but we must filter it out if we want 789 * MNT_UPDATE on the root file system to work. 790 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 791 */ 792 uap->flags &= ~MNT_ROOTFS; 793 794 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 795 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 796 if (error) { 797 free(fstype, M_TEMP); 798 return (error); 799 } 800 801 AUDIT_ARG(text, fstype); 802 mtx_lock(&Giant); 803 vfsp = vfs_byname_kld(fstype, td, &error); 804 free(fstype, M_TEMP); 805 if (vfsp == NULL) { 806 mtx_unlock(&Giant); 807 return (ENOENT); 808 } 809 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 810 mtx_unlock(&Giant); 811 return (EOPNOTSUPP); 812 } 813 814 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 815 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 816 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 817 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 818 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 819 820 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 821 mtx_unlock(&Giant); 822 return (error); 823 } 824 825 826 /* 827 * vfs_domount(): actually attempt a filesystem mount. 828 */ 829 static int 830 vfs_domount( 831 struct thread *td, /* Calling thread. */ 832 const char *fstype, /* Filesystem type. */ 833 char *fspath, /* Mount path. */ 834 int fsflags, /* Flags common to all filesystems. */ 835 void *fsdata /* Options local to the filesystem. */ 836 ) 837 { 838 struct vnode *vp; 839 struct mount *mp; 840 struct vfsconf *vfsp; 841 struct export_args export; 842 int error, flag = 0; 843 struct vattr va; 844 struct nameidata nd; 845 846 mtx_assert(&Giant, MA_OWNED); 847 /* 848 * Be ultra-paranoid about making sure the type and fspath 849 * variables will fit in our mp buffers, including the 850 * terminating NUL. 851 */ 852 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 853 return (ENAMETOOLONG); 854 855 if (jailed(td->td_ucred) || usermount == 0) { 856 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 857 return (error); 858 } 859 860 /* 861 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 862 */ 863 if (fsflags & MNT_EXPORTED) { 864 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 865 if (error) 866 return (error); 867 } 868 if (fsflags & MNT_SUIDDIR) { 869 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 870 if (error) 871 return (error); 872 } 873 /* 874 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 875 */ 876 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 877 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 878 fsflags |= MNT_NOSUID | MNT_USER; 879 } 880 881 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 882 vfsp = NULL; 883 if ((fsflags & MNT_UPDATE) == 0) { 884 /* Don't try to load KLDs if we're mounting the root. */ 885 if (fsflags & MNT_ROOTFS) 886 vfsp = vfs_byname(fstype); 887 else 888 vfsp = vfs_byname_kld(fstype, td, &error); 889 if (vfsp == NULL) 890 return (ENODEV); 891 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 892 return (EPERM); 893 } 894 /* 895 * Get vnode to be covered 896 */ 897 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, 898 fspath, td); 899 if ((error = namei(&nd)) != 0) 900 return (error); 901 NDFREE(&nd, NDF_ONLY_PNBUF); 902 vp = nd.ni_vp; 903 if (fsflags & MNT_UPDATE) { 904 if ((vp->v_vflag & VV_ROOT) == 0) { 905 vput(vp); 906 return (EINVAL); 907 } 908 mp = vp->v_mount; 909 MNT_ILOCK(mp); 910 flag = mp->mnt_flag; 911 /* 912 * We only allow the filesystem to be reloaded if it 913 * is currently mounted read-only. 914 */ 915 if ((fsflags & MNT_RELOAD) && 916 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 917 MNT_IUNLOCK(mp); 918 vput(vp); 919 return (EOPNOTSUPP); /* Needs translation */ 920 } 921 MNT_IUNLOCK(mp); 922 /* 923 * Only privileged root, or (if MNT_USER is set) the user that 924 * did the original mount is permitted to update it. 925 */ 926 error = vfs_suser(mp, td); 927 if (error) { 928 vput(vp); 929 return (error); 930 } 931 if (vfs_busy(mp, LK_NOWAIT, 0)) { 932 vput(vp); 933 return (EBUSY); 934 } 935 VI_LOCK(vp); 936 if ((vp->v_iflag & VI_MOUNT) != 0 || 937 vp->v_mountedhere != NULL) { 938 VI_UNLOCK(vp); 939 vfs_unbusy(mp); 940 vput(vp); 941 return (EBUSY); 942 } 943 vp->v_iflag |= VI_MOUNT; 944 VI_UNLOCK(vp); 945 MNT_ILOCK(mp); 946 mp->mnt_flag |= fsflags & 947 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 948 MNT_IUNLOCK(mp); 949 VOP_UNLOCK(vp, 0); 950 mp->mnt_optnew = fsdata; 951 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 952 } else { 953 /* 954 * If the user is not root, ensure that they own the directory 955 * onto which we are attempting to mount. 956 */ 957 error = VOP_GETATTR(vp, &va, td->td_ucred); 958 if (error) { 959 vput(vp); 960 return (error); 961 } 962 if (va.va_uid != td->td_ucred->cr_uid) { 963 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 964 0); 965 if (error) { 966 vput(vp); 967 return (error); 968 } 969 } 970 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 971 if (error != 0) { 972 vput(vp); 973 return (error); 974 } 975 if (vp->v_type != VDIR) { 976 vput(vp); 977 return (ENOTDIR); 978 } 979 VI_LOCK(vp); 980 if ((vp->v_iflag & VI_MOUNT) != 0 || 981 vp->v_mountedhere != NULL) { 982 VI_UNLOCK(vp); 983 vput(vp); 984 return (EBUSY); 985 } 986 vp->v_iflag |= VI_MOUNT; 987 VI_UNLOCK(vp); 988 989 /* 990 * Allocate and initialize the filesystem. 991 */ 992 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 993 VOP_UNLOCK(vp, 0); 994 995 /* XXXMAC: pass to vfs_mount_alloc? */ 996 mp->mnt_optnew = fsdata; 997 } 998 999 /* 1000 * Set the mount level flags. 1001 */ 1002 MNT_ILOCK(mp); 1003 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) | 1004 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS | 1005 MNT_RDONLY)); 1006 if ((mp->mnt_flag & MNT_ASYNC) == 0) 1007 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1008 MNT_IUNLOCK(mp); 1009 /* 1010 * Mount the filesystem. 1011 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 1012 * get. No freeing of cn_pnbuf. 1013 */ 1014 error = VFS_MOUNT(mp, td); 1015 1016 /* 1017 * Process the export option only if we are 1018 * updating mount options. 1019 */ 1020 if (!error && (fsflags & MNT_UPDATE)) { 1021 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 1022 sizeof(export)) == 0) 1023 error = vfs_export(mp, &export); 1024 } 1025 1026 if (!error) { 1027 if (mp->mnt_opt != NULL) 1028 vfs_freeopts(mp->mnt_opt); 1029 mp->mnt_opt = mp->mnt_optnew; 1030 (void)VFS_STATFS(mp, &mp->mnt_stat, td); 1031 } 1032 /* 1033 * Prevent external consumers of mount options from reading 1034 * mnt_optnew. 1035 */ 1036 mp->mnt_optnew = NULL; 1037 if (mp->mnt_flag & MNT_UPDATE) { 1038 MNT_ILOCK(mp); 1039 if (error) 1040 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | 1041 (flag & ~MNT_QUOTA); 1042 else 1043 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | 1044 MNT_FORCE | MNT_SNAPSHOT); 1045 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1046 mp->mnt_kern_flag |= MNTK_ASYNC; 1047 else 1048 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1049 MNT_IUNLOCK(mp); 1050 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1051 if (mp->mnt_syncer == NULL) 1052 error = vfs_allocate_syncvnode(mp); 1053 } else { 1054 if (mp->mnt_syncer != NULL) 1055 vrele(mp->mnt_syncer); 1056 mp->mnt_syncer = NULL; 1057 } 1058 vfs_unbusy(mp); 1059 VI_LOCK(vp); 1060 vp->v_iflag &= ~VI_MOUNT; 1061 VI_UNLOCK(vp); 1062 vrele(vp); 1063 return (error); 1064 } 1065 MNT_ILOCK(mp); 1066 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1067 mp->mnt_kern_flag |= MNTK_ASYNC; 1068 else 1069 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1070 MNT_IUNLOCK(mp); 1071 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1072 /* 1073 * Put the new filesystem on the mount list after root. 1074 */ 1075 cache_purge(vp); 1076 if (!error) { 1077 struct vnode *newdp; 1078 1079 VI_LOCK(vp); 1080 vp->v_iflag &= ~VI_MOUNT; 1081 VI_UNLOCK(vp); 1082 vp->v_mountedhere = mp; 1083 mtx_lock(&mountlist_mtx); 1084 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1085 mtx_unlock(&mountlist_mtx); 1086 vfs_event_signal(NULL, VQ_MOUNT, 0); 1087 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 1088 panic("mount: lost mount"); 1089 mountcheckdirs(vp, newdp); 1090 vput(newdp); 1091 VOP_UNLOCK(vp, 0); 1092 if ((mp->mnt_flag & MNT_RDONLY) == 0) 1093 error = vfs_allocate_syncvnode(mp); 1094 vfs_unbusy(mp); 1095 if (error) 1096 vrele(vp); 1097 } else { 1098 VI_LOCK(vp); 1099 vp->v_iflag &= ~VI_MOUNT; 1100 VI_UNLOCK(vp); 1101 vfs_unbusy(mp); 1102 vfs_mount_destroy(mp); 1103 vput(vp); 1104 } 1105 return (error); 1106 } 1107 1108 /* 1109 * Unmount a filesystem. 1110 * 1111 * Note: unmount takes a path to the vnode mounted on as argument, not 1112 * special file (as before). 1113 */ 1114 #ifndef _SYS_SYSPROTO_H_ 1115 struct unmount_args { 1116 char *path; 1117 int flags; 1118 }; 1119 #endif 1120 /* ARGSUSED */ 1121 int 1122 unmount(td, uap) 1123 struct thread *td; 1124 register struct unmount_args /* { 1125 char *path; 1126 int flags; 1127 } */ *uap; 1128 { 1129 struct mount *mp; 1130 char *pathbuf; 1131 int error, id0, id1; 1132 1133 if (jailed(td->td_ucred) || usermount == 0) { 1134 error = priv_check(td, PRIV_VFS_UNMOUNT); 1135 if (error) 1136 return (error); 1137 } 1138 1139 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1140 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1141 if (error) { 1142 free(pathbuf, M_TEMP); 1143 return (error); 1144 } 1145 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1); 1146 mtx_lock(&Giant); 1147 if (uap->flags & MNT_BYFSID) { 1148 /* Decode the filesystem ID. */ 1149 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1150 mtx_unlock(&Giant); 1151 free(pathbuf, M_TEMP); 1152 return (EINVAL); 1153 } 1154 1155 mtx_lock(&mountlist_mtx); 1156 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1157 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1158 mp->mnt_stat.f_fsid.val[1] == id1) 1159 break; 1160 } 1161 mtx_unlock(&mountlist_mtx); 1162 } else { 1163 mtx_lock(&mountlist_mtx); 1164 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1165 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1166 break; 1167 } 1168 mtx_unlock(&mountlist_mtx); 1169 } 1170 free(pathbuf, M_TEMP); 1171 if (mp == NULL) { 1172 /* 1173 * Previously we returned ENOENT for a nonexistent path and 1174 * EINVAL for a non-mountpoint. We cannot tell these apart 1175 * now, so in the !MNT_BYFSID case return the more likely 1176 * EINVAL for compatibility. 1177 */ 1178 mtx_unlock(&Giant); 1179 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1180 } 1181 1182 /* 1183 * Don't allow unmounting the root filesystem. 1184 */ 1185 if (mp->mnt_flag & MNT_ROOTFS) { 1186 mtx_unlock(&Giant); 1187 return (EINVAL); 1188 } 1189 error = dounmount(mp, uap->flags, td); 1190 mtx_unlock(&Giant); 1191 return (error); 1192 } 1193 1194 /* 1195 * Do the actual filesystem unmount. 1196 */ 1197 int 1198 dounmount(mp, flags, td) 1199 struct mount *mp; 1200 int flags; 1201 struct thread *td; 1202 { 1203 struct vnode *coveredvp, *fsrootvp; 1204 int error; 1205 int async_flag; 1206 int mnt_gen_r; 1207 1208 mtx_assert(&Giant, MA_OWNED); 1209 1210 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1211 mnt_gen_r = mp->mnt_gen; 1212 VI_LOCK(coveredvp); 1213 vholdl(coveredvp); 1214 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1215 vdrop(coveredvp); 1216 /* 1217 * Check for mp being unmounted while waiting for the 1218 * covered vnode lock. 1219 */ 1220 if (coveredvp->v_mountedhere != mp || 1221 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1222 VOP_UNLOCK(coveredvp, 0); 1223 return (EBUSY); 1224 } 1225 } 1226 /* 1227 * Only privileged root, or (if MNT_USER is set) the user that did the 1228 * original mount is permitted to unmount this filesystem. 1229 */ 1230 error = vfs_suser(mp, td); 1231 if (error) { 1232 if (coveredvp) 1233 VOP_UNLOCK(coveredvp, 0); 1234 return (error); 1235 } 1236 1237 MNT_ILOCK(mp); 1238 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1239 MNT_IUNLOCK(mp); 1240 if (coveredvp) 1241 VOP_UNLOCK(coveredvp, 0); 1242 return (EBUSY); 1243 } 1244 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1245 /* Allow filesystems to detect that a forced unmount is in progress. */ 1246 if (flags & MNT_FORCE) 1247 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1248 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 1249 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp)); 1250 if (error) { 1251 MNT_ILOCK(mp); 1252 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1253 MNTK_UNMOUNTF); 1254 if (mp->mnt_kern_flag & MNTK_MWAIT) 1255 wakeup(mp); 1256 MNT_IUNLOCK(mp); 1257 if (coveredvp) 1258 VOP_UNLOCK(coveredvp, 0); 1259 return (error); 1260 } 1261 vn_start_write(NULL, &mp, V_WAIT); 1262 1263 if (mp->mnt_flag & MNT_EXPUBLIC) 1264 vfs_setpublicfs(NULL, NULL, NULL); 1265 1266 vfs_msync(mp, MNT_WAIT); 1267 MNT_ILOCK(mp); 1268 async_flag = mp->mnt_flag & MNT_ASYNC; 1269 mp->mnt_flag &= ~MNT_ASYNC; 1270 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1271 MNT_IUNLOCK(mp); 1272 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1273 if (mp->mnt_syncer != NULL) 1274 vrele(mp->mnt_syncer); 1275 /* 1276 * For forced unmounts, move process cdir/rdir refs on the fs root 1277 * vnode to the covered vnode. For non-forced unmounts we want 1278 * such references to cause an EBUSY error. 1279 */ 1280 if ((flags & MNT_FORCE) && 1281 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1282 if (mp->mnt_vnodecovered != NULL) 1283 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1284 if (fsrootvp == rootvnode) { 1285 vrele(rootvnode); 1286 rootvnode = NULL; 1287 } 1288 vput(fsrootvp); 1289 } 1290 if (((mp->mnt_flag & MNT_RDONLY) || 1291 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1292 (flags & MNT_FORCE)) { 1293 error = VFS_UNMOUNT(mp, flags, td); 1294 } 1295 vn_finished_write(mp); 1296 /* 1297 * If we failed to flush the dirty blocks for this mount point, 1298 * undo all the cdir/rdir and rootvnode changes we made above. 1299 * Unless we failed to do so because the device is reporting that 1300 * it doesn't exist anymore. 1301 */ 1302 if (error && error != ENXIO) { 1303 if ((flags & MNT_FORCE) && 1304 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1305 if (mp->mnt_vnodecovered != NULL) 1306 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1307 if (rootvnode == NULL) { 1308 rootvnode = fsrootvp; 1309 vref(rootvnode); 1310 } 1311 vput(fsrootvp); 1312 } 1313 MNT_ILOCK(mp); 1314 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1315 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1316 MNT_IUNLOCK(mp); 1317 (void) vfs_allocate_syncvnode(mp); 1318 MNT_ILOCK(mp); 1319 } 1320 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1321 mp->mnt_flag |= async_flag; 1322 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1323 mp->mnt_kern_flag |= MNTK_ASYNC; 1324 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL); 1325 if (mp->mnt_kern_flag & MNTK_MWAIT) 1326 wakeup(mp); 1327 MNT_IUNLOCK(mp); 1328 if (coveredvp) 1329 VOP_UNLOCK(coveredvp, 0); 1330 return (error); 1331 } 1332 mtx_lock(&mountlist_mtx); 1333 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1334 mtx_unlock(&mountlist_mtx); 1335 if (coveredvp != NULL) { 1336 coveredvp->v_mountedhere = NULL; 1337 vput(coveredvp); 1338 } 1339 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1340 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL); 1341 vfs_mount_destroy(mp); 1342 return (0); 1343 } 1344 1345 /* 1346 * --------------------------------------------------------------------- 1347 * Mounting of root filesystem 1348 * 1349 */ 1350 1351 struct root_hold_token { 1352 const char *who; 1353 LIST_ENTRY(root_hold_token) list; 1354 }; 1355 1356 static LIST_HEAD(, root_hold_token) root_holds = 1357 LIST_HEAD_INITIALIZER(&root_holds); 1358 1359 static int root_mount_complete; 1360 1361 /* 1362 * Hold root mount. 1363 */ 1364 struct root_hold_token * 1365 root_mount_hold(const char *identifier) 1366 { 1367 struct root_hold_token *h; 1368 1369 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1370 h->who = identifier; 1371 mtx_lock(&mountlist_mtx); 1372 LIST_INSERT_HEAD(&root_holds, h, list); 1373 mtx_unlock(&mountlist_mtx); 1374 return (h); 1375 } 1376 1377 /* 1378 * Release root mount. 1379 */ 1380 void 1381 root_mount_rel(struct root_hold_token *h) 1382 { 1383 1384 mtx_lock(&mountlist_mtx); 1385 LIST_REMOVE(h, list); 1386 wakeup(&root_holds); 1387 mtx_unlock(&mountlist_mtx); 1388 free(h, M_DEVBUF); 1389 } 1390 1391 /* 1392 * Wait for all subsystems to release root mount. 1393 */ 1394 static void 1395 root_mount_prepare(void) 1396 { 1397 struct root_hold_token *h; 1398 1399 for (;;) { 1400 DROP_GIANT(); 1401 g_waitidle(); 1402 PICKUP_GIANT(); 1403 mtx_lock(&mountlist_mtx); 1404 if (LIST_EMPTY(&root_holds)) { 1405 mtx_unlock(&mountlist_mtx); 1406 break; 1407 } 1408 printf("Root mount waiting for:"); 1409 LIST_FOREACH(h, &root_holds, list) 1410 printf(" %s", h->who); 1411 printf("\n"); 1412 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1413 hz); 1414 } 1415 } 1416 1417 /* 1418 * Root was mounted, share the good news. 1419 */ 1420 static void 1421 root_mount_done(void) 1422 { 1423 1424 /* 1425 * Use a mutex to prevent the wakeup being missed and waiting for 1426 * an extra 1 second sleep. 1427 */ 1428 mtx_lock(&mountlist_mtx); 1429 root_mount_complete = 1; 1430 wakeup(&root_mount_complete); 1431 mtx_unlock(&mountlist_mtx); 1432 } 1433 1434 /* 1435 * Return true if root is already mounted. 1436 */ 1437 int 1438 root_mounted(void) 1439 { 1440 1441 /* No mutex is acquired here because int stores are atomic. */ 1442 return (root_mount_complete); 1443 } 1444 1445 /* 1446 * Wait until root is mounted. 1447 */ 1448 void 1449 root_mount_wait(void) 1450 { 1451 1452 /* 1453 * Panic on an obvious deadlock - the function can't be called from 1454 * a thread which is doing the whole SYSINIT stuff. 1455 */ 1456 KASSERT(curthread->td_proc->p_pid != 0, 1457 ("root_mount_wait: cannot be called from the swapper thread")); 1458 mtx_lock(&mountlist_mtx); 1459 while (!root_mount_complete) { 1460 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1461 hz); 1462 } 1463 mtx_unlock(&mountlist_mtx); 1464 } 1465 1466 static void 1467 set_rootvnode(struct thread *td) 1468 { 1469 struct proc *p; 1470 1471 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1472 panic("Cannot find root vnode"); 1473 1474 p = td->td_proc; 1475 FILEDESC_XLOCK(p->p_fd); 1476 1477 if (p->p_fd->fd_cdir != NULL) 1478 vrele(p->p_fd->fd_cdir); 1479 p->p_fd->fd_cdir = rootvnode; 1480 VREF(rootvnode); 1481 1482 if (p->p_fd->fd_rdir != NULL) 1483 vrele(p->p_fd->fd_rdir); 1484 p->p_fd->fd_rdir = rootvnode; 1485 VREF(rootvnode); 1486 1487 FILEDESC_XUNLOCK(p->p_fd); 1488 1489 VOP_UNLOCK(rootvnode, 0); 1490 1491 EVENTHANDLER_INVOKE(mountroot); 1492 } 1493 1494 /* 1495 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1496 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1497 */ 1498 1499 static void 1500 devfs_first(void) 1501 { 1502 struct thread *td = curthread; 1503 struct vfsoptlist *opts; 1504 struct vfsconf *vfsp; 1505 struct mount *mp = NULL; 1506 int error; 1507 1508 vfsp = vfs_byname("devfs"); 1509 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1510 if (vfsp == NULL) 1511 return; 1512 1513 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 1514 1515 error = VFS_MOUNT(mp, td); 1516 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1517 if (error) 1518 return; 1519 1520 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1521 TAILQ_INIT(opts); 1522 mp->mnt_opt = opts; 1523 1524 mtx_lock(&mountlist_mtx); 1525 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1526 mtx_unlock(&mountlist_mtx); 1527 1528 set_rootvnode(td); 1529 1530 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1531 if (error) 1532 printf("kern_symlink /dev -> / returns %d\n", error); 1533 } 1534 1535 /* 1536 * Surgically move our devfs to be mounted on /dev. 1537 */ 1538 1539 static void 1540 devfs_fixup(struct thread *td) 1541 { 1542 struct nameidata nd; 1543 int error; 1544 struct vnode *vp, *dvp; 1545 struct mount *mp; 1546 1547 /* Remove our devfs mount from the mountlist and purge the cache */ 1548 mtx_lock(&mountlist_mtx); 1549 mp = TAILQ_FIRST(&mountlist); 1550 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1551 mtx_unlock(&mountlist_mtx); 1552 cache_purgevfs(mp); 1553 1554 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1555 VI_LOCK(dvp); 1556 dvp->v_iflag &= ~VI_MOUNT; 1557 VI_UNLOCK(dvp); 1558 dvp->v_mountedhere = NULL; 1559 1560 /* Set up the real rootvnode, and purge the cache */ 1561 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1562 set_rootvnode(td); 1563 cache_purgevfs(rootvnode->v_mount); 1564 1565 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1566 error = namei(&nd); 1567 if (error) { 1568 printf("Lookup of /dev for devfs, error: %d\n", error); 1569 return; 1570 } 1571 NDFREE(&nd, NDF_ONLY_PNBUF); 1572 vp = nd.ni_vp; 1573 if (vp->v_type != VDIR) { 1574 vput(vp); 1575 } 1576 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 1577 if (error) { 1578 vput(vp); 1579 } 1580 cache_purge(vp); 1581 mp->mnt_vnodecovered = vp; 1582 vp->v_mountedhere = mp; 1583 mtx_lock(&mountlist_mtx); 1584 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1585 mtx_unlock(&mountlist_mtx); 1586 VOP_UNLOCK(vp, 0); 1587 vput(dvp); 1588 vfs_unbusy(mp); 1589 1590 /* Unlink the no longer needed /dev/dev -> / symlink */ 1591 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1592 } 1593 1594 /* 1595 * Report errors during filesystem mounting. 1596 */ 1597 void 1598 vfs_mount_error(struct mount *mp, const char *fmt, ...) 1599 { 1600 struct vfsoptlist *moptlist = mp->mnt_optnew; 1601 va_list ap; 1602 int error, len; 1603 char *errmsg; 1604 1605 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1606 if (error || errmsg == NULL || len <= 0) 1607 return; 1608 1609 va_start(ap, fmt); 1610 vsnprintf(errmsg, (size_t)len, fmt, ap); 1611 va_end(ap); 1612 } 1613 1614 /* 1615 * Find and mount the root filesystem 1616 */ 1617 void 1618 vfs_mountroot(void) 1619 { 1620 char *cp; 1621 int error, i, asked = 0; 1622 1623 root_mount_prepare(); 1624 1625 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1626 NULL, NULL, mount_init, mount_fini, 1627 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1628 devfs_first(); 1629 1630 /* 1631 * We are booted with instructions to prompt for the root filesystem. 1632 */ 1633 if (boothowto & RB_ASKNAME) { 1634 if (!vfs_mountroot_ask()) 1635 goto mounted; 1636 asked = 1; 1637 } 1638 1639 /* 1640 * The root filesystem information is compiled in, and we are 1641 * booted with instructions to use it. 1642 */ 1643 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1644 if (!vfs_mountroot_try(ctrootdevname)) 1645 goto mounted; 1646 ctrootdevname = NULL; 1647 } 1648 1649 /* 1650 * We've been given the generic "use CDROM as root" flag. This is 1651 * necessary because one media may be used in many different 1652 * devices, so we need to search for them. 1653 */ 1654 if (boothowto & RB_CDROM) { 1655 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1656 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1657 goto mounted; 1658 } 1659 } 1660 1661 /* 1662 * Try to use the value read by the loader from /etc/fstab, or 1663 * supplied via some other means. This is the preferred 1664 * mechanism. 1665 */ 1666 cp = getenv("vfs.root.mountfrom"); 1667 if (cp != NULL) { 1668 error = vfs_mountroot_try(cp); 1669 freeenv(cp); 1670 if (!error) 1671 goto mounted; 1672 } 1673 1674 /* 1675 * Try values that may have been computed by code during boot 1676 */ 1677 if (!vfs_mountroot_try(rootdevnames[0])) 1678 goto mounted; 1679 if (!vfs_mountroot_try(rootdevnames[1])) 1680 goto mounted; 1681 1682 /* 1683 * If we (still) have a compiled-in default, try it. 1684 */ 1685 if (ctrootdevname != NULL) 1686 if (!vfs_mountroot_try(ctrootdevname)) 1687 goto mounted; 1688 /* 1689 * Everything so far has failed, prompt on the console if we haven't 1690 * already tried that. 1691 */ 1692 if (!asked) 1693 if (!vfs_mountroot_ask()) 1694 goto mounted; 1695 1696 panic("Root mount failed, startup aborted."); 1697 1698 mounted: 1699 root_mount_done(); 1700 } 1701 1702 /* 1703 * Mount (mountfrom) as the root filesystem. 1704 */ 1705 static int 1706 vfs_mountroot_try(const char *mountfrom) 1707 { 1708 struct mount *mp; 1709 char *vfsname, *path; 1710 time_t timebase; 1711 int error; 1712 char patt[32]; 1713 1714 vfsname = NULL; 1715 path = NULL; 1716 mp = NULL; 1717 error = EINVAL; 1718 1719 if (mountfrom == NULL) 1720 return (error); /* don't complain */ 1721 printf("Trying to mount root from %s\n", mountfrom); 1722 1723 /* parse vfs name and path */ 1724 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1725 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1726 vfsname[0] = path[0] = 0; 1727 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1728 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1729 goto out; 1730 1731 if (path[0] == '\0') 1732 strcpy(path, ROOTNAME); 1733 1734 error = kernel_vmount( 1735 MNT_RDONLY | MNT_ROOTFS, 1736 "fstype", vfsname, 1737 "fspath", "/", 1738 "from", path, 1739 NULL); 1740 if (error == 0) { 1741 /* 1742 * We mount devfs prior to mounting the / FS, so the first 1743 * entry will typically be devfs. 1744 */ 1745 mp = TAILQ_FIRST(&mountlist); 1746 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1747 1748 /* 1749 * Iterate over all currently mounted file systems and use 1750 * the time stamp found to check and/or initialize the RTC. 1751 * Typically devfs has no time stamp and the only other FS 1752 * is the actual / FS. 1753 * Call inittodr() only once and pass it the largest of the 1754 * timestamps we encounter. 1755 */ 1756 timebase = 0; 1757 do { 1758 if (mp->mnt_time > timebase) 1759 timebase = mp->mnt_time; 1760 mp = TAILQ_NEXT(mp, mnt_list); 1761 } while (mp != NULL); 1762 inittodr(timebase); 1763 1764 devfs_fixup(curthread); 1765 } 1766 out: 1767 free(path, M_MOUNT); 1768 free(vfsname, M_MOUNT); 1769 return (error); 1770 } 1771 1772 /* 1773 * --------------------------------------------------------------------- 1774 * Interactive root filesystem selection code. 1775 */ 1776 1777 static int 1778 vfs_mountroot_ask(void) 1779 { 1780 char name[128]; 1781 1782 for(;;) { 1783 printf("\nManual root filesystem specification:\n"); 1784 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1785 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__) 1786 printf(" eg. ufs:da0s1a\n"); 1787 #else 1788 printf(" eg. ufs:/dev/da0a\n"); 1789 #endif 1790 printf(" ? List valid disk boot devices\n"); 1791 printf(" <empty line> Abort manual input\n"); 1792 printf("\nmountroot> "); 1793 gets(name, sizeof(name), 1); 1794 if (name[0] == '\0') 1795 return (1); 1796 if (name[0] == '?') { 1797 printf("\nList of GEOM managed disk devices:\n "); 1798 g_dev_print(); 1799 continue; 1800 } 1801 if (!vfs_mountroot_try(name)) 1802 return (0); 1803 } 1804 } 1805 1806 /* 1807 * --------------------------------------------------------------------- 1808 * Functions for querying mount options/arguments from filesystems. 1809 */ 1810 1811 /* 1812 * Check that no unknown options are given 1813 */ 1814 int 1815 vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1816 { 1817 struct vfsopt *opt; 1818 char errmsg[255]; 1819 const char **t, *p, *q; 1820 int ret = 0; 1821 1822 TAILQ_FOREACH(opt, opts, link) { 1823 p = opt->name; 1824 q = NULL; 1825 if (p[0] == 'n' && p[1] == 'o') 1826 q = p + 2; 1827 for(t = global_opts; *t != NULL; t++) { 1828 if (strcmp(*t, p) == 0) 1829 break; 1830 if (q != NULL) { 1831 if (strcmp(*t, q) == 0) 1832 break; 1833 } 1834 } 1835 if (*t != NULL) 1836 continue; 1837 for(t = legal; *t != NULL; t++) { 1838 if (strcmp(*t, p) == 0) 1839 break; 1840 if (q != NULL) { 1841 if (strcmp(*t, q) == 0) 1842 break; 1843 } 1844 } 1845 if (*t != NULL) 1846 continue; 1847 snprintf(errmsg, sizeof(errmsg), 1848 "mount option <%s> is unknown", p); 1849 printf("%s\n", errmsg); 1850 ret = EINVAL; 1851 } 1852 if (ret != 0) { 1853 TAILQ_FOREACH(opt, opts, link) { 1854 if (strcmp(opt->name, "errmsg") == 0) { 1855 strncpy((char *)opt->value, errmsg, opt->len); 1856 } 1857 } 1858 } 1859 return (ret); 1860 } 1861 1862 /* 1863 * Get a mount option by its name. 1864 * 1865 * Return 0 if the option was found, ENOENT otherwise. 1866 * If len is non-NULL it will be filled with the length 1867 * of the option. If buf is non-NULL, it will be filled 1868 * with the address of the option. 1869 */ 1870 int 1871 vfs_getopt(opts, name, buf, len) 1872 struct vfsoptlist *opts; 1873 const char *name; 1874 void **buf; 1875 int *len; 1876 { 1877 struct vfsopt *opt; 1878 1879 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1880 1881 TAILQ_FOREACH(opt, opts, link) { 1882 if (strcmp(name, opt->name) == 0) { 1883 if (len != NULL) 1884 *len = opt->len; 1885 if (buf != NULL) 1886 *buf = opt->value; 1887 return (0); 1888 } 1889 } 1890 return (ENOENT); 1891 } 1892 1893 static int 1894 vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 1895 { 1896 struct vfsopt *opt; 1897 int i; 1898 1899 if (opts == NULL) 1900 return (-1); 1901 1902 i = 0; 1903 TAILQ_FOREACH(opt, opts, link) { 1904 if (strcmp(name, opt->name) == 0) 1905 return (i); 1906 ++i; 1907 } 1908 return (-1); 1909 } 1910 1911 char * 1912 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1913 { 1914 struct vfsopt *opt; 1915 1916 *error = 0; 1917 TAILQ_FOREACH(opt, opts, link) { 1918 if (strcmp(name, opt->name) != 0) 1919 continue; 1920 if (((char *)opt->value)[opt->len - 1] != '\0') { 1921 *error = EINVAL; 1922 return (NULL); 1923 } 1924 return (opt->value); 1925 } 1926 *error = ENOENT; 1927 return (NULL); 1928 } 1929 1930 int 1931 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1932 { 1933 struct vfsopt *opt; 1934 1935 TAILQ_FOREACH(opt, opts, link) { 1936 if (strcmp(name, opt->name) == 0) { 1937 if (w != NULL) 1938 *w |= val; 1939 return (1); 1940 } 1941 } 1942 if (w != NULL) 1943 *w &= ~val; 1944 return (0); 1945 } 1946 1947 int 1948 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1949 { 1950 va_list ap; 1951 struct vfsopt *opt; 1952 int ret; 1953 1954 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1955 1956 TAILQ_FOREACH(opt, opts, link) { 1957 if (strcmp(name, opt->name) != 0) 1958 continue; 1959 if (opt->len == 0 || opt->value == NULL) 1960 return (0); 1961 if (((char *)opt->value)[opt->len - 1] != '\0') 1962 return (0); 1963 va_start(ap, fmt); 1964 ret = vsscanf(opt->value, fmt, ap); 1965 va_end(ap); 1966 return (ret); 1967 } 1968 return (0); 1969 } 1970 1971 /* 1972 * Find and copy a mount option. 1973 * 1974 * The size of the buffer has to be specified 1975 * in len, if it is not the same length as the 1976 * mount option, EINVAL is returned. 1977 * Returns ENOENT if the option is not found. 1978 */ 1979 int 1980 vfs_copyopt(opts, name, dest, len) 1981 struct vfsoptlist *opts; 1982 const char *name; 1983 void *dest; 1984 int len; 1985 { 1986 struct vfsopt *opt; 1987 1988 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1989 1990 TAILQ_FOREACH(opt, opts, link) { 1991 if (strcmp(name, opt->name) == 0) { 1992 if (len != opt->len) 1993 return (EINVAL); 1994 bcopy(opt->value, dest, opt->len); 1995 return (0); 1996 } 1997 } 1998 return (ENOENT); 1999 } 2000 2001 /* 2002 * This is a helper function for filesystems to traverse their 2003 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2004 */ 2005 2006 struct vnode * 2007 __mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2008 { 2009 struct vnode *vp; 2010 2011 mtx_assert(MNT_MTX(mp), MA_OWNED); 2012 2013 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2014 if ((*mvp)->v_yield++ == 500) { 2015 MNT_IUNLOCK(mp); 2016 (*mvp)->v_yield = 0; 2017 uio_yield(); 2018 MNT_ILOCK(mp); 2019 } 2020 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2021 while (vp != NULL && vp->v_type == VMARKER) 2022 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2023 2024 /* Check if we are done */ 2025 if (vp == NULL) { 2026 __mnt_vnode_markerfree(mvp, mp); 2027 return (NULL); 2028 } 2029 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2030 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2031 return (vp); 2032 } 2033 2034 struct vnode * 2035 __mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2036 { 2037 struct vnode *vp; 2038 2039 mtx_assert(MNT_MTX(mp), MA_OWNED); 2040 2041 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2042 while (vp != NULL && vp->v_type == VMARKER) 2043 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2044 2045 /* Check if we are done */ 2046 if (vp == NULL) { 2047 *mvp = NULL; 2048 return (NULL); 2049 } 2050 mp->mnt_holdcnt++; 2051 MNT_IUNLOCK(mp); 2052 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2053 M_VNODE_MARKER, 2054 M_WAITOK | M_ZERO); 2055 MNT_ILOCK(mp); 2056 (*mvp)->v_type = VMARKER; 2057 2058 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2059 while (vp != NULL && vp->v_type == VMARKER) 2060 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2061 2062 /* Check if we are done */ 2063 if (vp == NULL) { 2064 MNT_IUNLOCK(mp); 2065 free(*mvp, M_VNODE_MARKER); 2066 MNT_ILOCK(mp); 2067 *mvp = NULL; 2068 mp->mnt_holdcnt--; 2069 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2070 wakeup(&mp->mnt_holdcnt); 2071 return (NULL); 2072 } 2073 mp->mnt_markercnt++; 2074 (*mvp)->v_mount = mp; 2075 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2076 return (vp); 2077 } 2078 2079 2080 void 2081 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2082 { 2083 2084 if (*mvp == NULL) 2085 return; 2086 2087 mtx_assert(MNT_MTX(mp), MA_OWNED); 2088 2089 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2090 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2091 MNT_IUNLOCK(mp); 2092 free(*mvp, M_VNODE_MARKER); 2093 MNT_ILOCK(mp); 2094 *mvp = NULL; 2095 2096 mp->mnt_markercnt--; 2097 mp->mnt_holdcnt--; 2098 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0) 2099 wakeup(&mp->mnt_holdcnt); 2100 } 2101 2102 2103 int 2104 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 2105 { 2106 int error; 2107 2108 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 2109 if (sbp != &mp->mnt_stat) 2110 *sbp = mp->mnt_stat; 2111 return (error); 2112 } 2113 2114 void 2115 vfs_mountedfrom(struct mount *mp, const char *from) 2116 { 2117 2118 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2119 strlcpy(mp->mnt_stat.f_mntfromname, from, 2120 sizeof mp->mnt_stat.f_mntfromname); 2121 } 2122 2123 /* 2124 * --------------------------------------------------------------------- 2125 * This is the api for building mount args and mounting filesystems from 2126 * inside the kernel. 2127 * 2128 * The API works by accumulation of individual args. First error is 2129 * latched. 2130 * 2131 * XXX: should be documented in new manpage kernel_mount(9) 2132 */ 2133 2134 /* A memory allocation which must be freed when we are done */ 2135 struct mntaarg { 2136 SLIST_ENTRY(mntaarg) next; 2137 }; 2138 2139 /* The header for the mount arguments */ 2140 struct mntarg { 2141 struct iovec *v; 2142 int len; 2143 int error; 2144 SLIST_HEAD(, mntaarg) list; 2145 }; 2146 2147 /* 2148 * Add a boolean argument. 2149 * 2150 * flag is the boolean value. 2151 * name must start with "no". 2152 */ 2153 struct mntarg * 2154 mount_argb(struct mntarg *ma, int flag, const char *name) 2155 { 2156 2157 KASSERT(name[0] == 'n' && name[1] == 'o', 2158 ("mount_argb(...,%s): name must start with 'no'", name)); 2159 2160 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2161 } 2162 2163 /* 2164 * Add an argument printf style 2165 */ 2166 struct mntarg * 2167 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2168 { 2169 va_list ap; 2170 struct mntaarg *maa; 2171 struct sbuf *sb; 2172 int len; 2173 2174 if (ma == NULL) { 2175 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2176 SLIST_INIT(&ma->list); 2177 } 2178 if (ma->error) 2179 return (ma); 2180 2181 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2182 M_MOUNT, M_WAITOK); 2183 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2184 ma->v[ma->len].iov_len = strlen(name) + 1; 2185 ma->len++; 2186 2187 sb = sbuf_new_auto(); 2188 va_start(ap, fmt); 2189 sbuf_vprintf(sb, fmt, ap); 2190 va_end(ap); 2191 sbuf_finish(sb); 2192 len = sbuf_len(sb) + 1; 2193 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2194 SLIST_INSERT_HEAD(&ma->list, maa, next); 2195 bcopy(sbuf_data(sb), maa + 1, len); 2196 sbuf_delete(sb); 2197 2198 ma->v[ma->len].iov_base = maa + 1; 2199 ma->v[ma->len].iov_len = len; 2200 ma->len++; 2201 2202 return (ma); 2203 } 2204 2205 /* 2206 * Add an argument which is a userland string. 2207 */ 2208 struct mntarg * 2209 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2210 { 2211 struct mntaarg *maa; 2212 char *tbuf; 2213 2214 if (val == NULL) 2215 return (ma); 2216 if (ma == NULL) { 2217 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2218 SLIST_INIT(&ma->list); 2219 } 2220 if (ma->error) 2221 return (ma); 2222 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2223 SLIST_INSERT_HEAD(&ma->list, maa, next); 2224 tbuf = (void *)(maa + 1); 2225 ma->error = copyinstr(val, tbuf, len, NULL); 2226 return (mount_arg(ma, name, tbuf, -1)); 2227 } 2228 2229 /* 2230 * Plain argument. 2231 * 2232 * If length is -1, treat value as a C string. 2233 */ 2234 struct mntarg * 2235 mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2236 { 2237 2238 if (ma == NULL) { 2239 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2240 SLIST_INIT(&ma->list); 2241 } 2242 if (ma->error) 2243 return (ma); 2244 2245 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2246 M_MOUNT, M_WAITOK); 2247 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2248 ma->v[ma->len].iov_len = strlen(name) + 1; 2249 ma->len++; 2250 2251 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2252 if (len < 0) 2253 ma->v[ma->len].iov_len = strlen(val) + 1; 2254 else 2255 ma->v[ma->len].iov_len = len; 2256 ma->len++; 2257 return (ma); 2258 } 2259 2260 /* 2261 * Free a mntarg structure 2262 */ 2263 static void 2264 free_mntarg(struct mntarg *ma) 2265 { 2266 struct mntaarg *maa; 2267 2268 while (!SLIST_EMPTY(&ma->list)) { 2269 maa = SLIST_FIRST(&ma->list); 2270 SLIST_REMOVE_HEAD(&ma->list, next); 2271 free(maa, M_MOUNT); 2272 } 2273 free(ma->v, M_MOUNT); 2274 free(ma, M_MOUNT); 2275 } 2276 2277 /* 2278 * Mount a filesystem 2279 */ 2280 int 2281 kernel_mount(struct mntarg *ma, int flags) 2282 { 2283 struct uio auio; 2284 int error; 2285 2286 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2287 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2288 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2289 2290 auio.uio_iov = ma->v; 2291 auio.uio_iovcnt = ma->len; 2292 auio.uio_segflg = UIO_SYSSPACE; 2293 2294 error = ma->error; 2295 if (!error) 2296 error = vfs_donmount(curthread, flags, &auio); 2297 free_mntarg(ma); 2298 return (error); 2299 } 2300 2301 /* 2302 * A printflike function to mount a filesystem. 2303 */ 2304 int 2305 kernel_vmount(int flags, ...) 2306 { 2307 struct mntarg *ma = NULL; 2308 va_list ap; 2309 const char *cp; 2310 const void *vp; 2311 int error; 2312 2313 va_start(ap, flags); 2314 for (;;) { 2315 cp = va_arg(ap, const char *); 2316 if (cp == NULL) 2317 break; 2318 vp = va_arg(ap, const void *); 2319 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2320 } 2321 va_end(ap); 2322 2323 error = kernel_mount(ma, flags); 2324 return (error); 2325 } 2326