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