1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2015 Nexenta Systems, Inc. All rights reserved. 24 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 25 * Copyright (c) 2014, 2021 by Delphix. All rights reserved. 26 * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com> 27 * Copyright 2017 RackTop Systems. 28 * Copyright (c) 2018 Datto Inc. 29 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association. 30 */ 31 32 /* 33 * Routines to manage ZFS mounts. We separate all the nasty routines that have 34 * to deal with the OS. The following functions are the main entry points -- 35 * they are used by mount and unmount and when changing a filesystem's 36 * mountpoint. 37 * 38 * zfs_is_mounted() 39 * zfs_mount() 40 * zfs_mount_at() 41 * zfs_unmount() 42 * zfs_unmountall() 43 * 44 * This file also contains the functions used to manage sharing filesystems via 45 * NFS and iSCSI: 46 * 47 * zfs_is_shared() 48 * zfs_share() 49 * zfs_unshare() 50 * 51 * zfs_is_shared_nfs() 52 * zfs_is_shared_smb() 53 * zfs_share_proto() 54 * zfs_shareall(); 55 * zfs_unshare_nfs() 56 * zfs_unshare_smb() 57 * zfs_unshareall_nfs() 58 * zfs_unshareall_smb() 59 * zfs_unshareall() 60 * zfs_unshareall_bypath() 61 * 62 * The following functions are available for pool consumers, and will 63 * mount/unmount and share/unshare all datasets within pool: 64 * 65 * zpool_enable_datasets() 66 * zpool_disable_datasets() 67 */ 68 69 #include <dirent.h> 70 #include <dlfcn.h> 71 #include <errno.h> 72 #include <fcntl.h> 73 #include <libgen.h> 74 #include <libintl.h> 75 #include <stdio.h> 76 #include <stdlib.h> 77 #include <string.h> 78 #include <unistd.h> 79 #include <zone.h> 80 #include <sys/mntent.h> 81 #include <sys/mount.h> 82 #include <sys/stat.h> 83 #include <sys/vfs.h> 84 #include <sys/dsl_crypt.h> 85 86 #include <libzfs.h> 87 88 #include "libzfs_impl.h" 89 #include <thread_pool.h> 90 91 #include <libshare.h> 92 #include <sys/systeminfo.h> 93 #define MAXISALEN 257 /* based on sysinfo(2) man page */ 94 95 static int mount_tp_nthr = 512; /* tpool threads for multi-threaded mounting */ 96 97 static void zfs_mount_task(void *); 98 static zfs_share_type_t zfs_is_shared_proto(zfs_handle_t *, char **, 99 zfs_share_proto_t); 100 101 /* 102 * The share protocols table must be in the same order as the zfs_share_proto_t 103 * enum in libzfs_impl.h 104 */ 105 static const proto_table_t proto_table[PROTO_END] = { 106 {ZFS_PROP_SHARENFS, "nfs", EZFS_SHARENFSFAILED, EZFS_UNSHARENFSFAILED}, 107 {ZFS_PROP_SHARESMB, "smb", EZFS_SHARESMBFAILED, EZFS_UNSHARESMBFAILED}, 108 }; 109 110 static const zfs_share_proto_t nfs_only[] = { 111 PROTO_NFS, 112 PROTO_END 113 }; 114 115 static const zfs_share_proto_t smb_only[] = { 116 PROTO_SMB, 117 PROTO_END 118 }; 119 static const zfs_share_proto_t share_all_proto[] = { 120 PROTO_NFS, 121 PROTO_SMB, 122 PROTO_END 123 }; 124 125 126 127 static boolean_t 128 dir_is_empty_stat(const char *dirname) 129 { 130 struct stat st; 131 132 /* 133 * We only want to return false if the given path is a non empty 134 * directory, all other errors are handled elsewhere. 135 */ 136 if (stat(dirname, &st) < 0 || !S_ISDIR(st.st_mode)) { 137 return (B_TRUE); 138 } 139 140 /* 141 * An empty directory will still have two entries in it, one 142 * entry for each of "." and "..". 143 */ 144 if (st.st_size > 2) { 145 return (B_FALSE); 146 } 147 148 return (B_TRUE); 149 } 150 151 static boolean_t 152 dir_is_empty_readdir(const char *dirname) 153 { 154 DIR *dirp; 155 struct dirent64 *dp; 156 int dirfd; 157 158 if ((dirfd = openat(AT_FDCWD, dirname, 159 O_RDONLY | O_NDELAY | O_LARGEFILE | O_CLOEXEC, 0)) < 0) { 160 return (B_TRUE); 161 } 162 163 if ((dirp = fdopendir(dirfd)) == NULL) { 164 (void) close(dirfd); 165 return (B_TRUE); 166 } 167 168 while ((dp = readdir64(dirp)) != NULL) { 169 170 if (strcmp(dp->d_name, ".") == 0 || 171 strcmp(dp->d_name, "..") == 0) 172 continue; 173 174 (void) closedir(dirp); 175 return (B_FALSE); 176 } 177 178 (void) closedir(dirp); 179 return (B_TRUE); 180 } 181 182 /* 183 * Returns true if the specified directory is empty. If we can't open the 184 * directory at all, return true so that the mount can fail with a more 185 * informative error message. 186 */ 187 static boolean_t 188 dir_is_empty(const char *dirname) 189 { 190 struct statfs64 st; 191 192 /* 193 * If the statvfs call fails or the filesystem is not a ZFS 194 * filesystem, fall back to the slow path which uses readdir. 195 */ 196 if ((statfs64(dirname, &st) != 0) || 197 (st.f_type != ZFS_SUPER_MAGIC)) { 198 return (dir_is_empty_readdir(dirname)); 199 } 200 201 /* 202 * At this point, we know the provided path is on a ZFS 203 * filesystem, so we can use stat instead of readdir to 204 * determine if the directory is empty or not. We try to avoid 205 * using readdir because that requires opening "dirname"; this 206 * open file descriptor can potentially end up in a child 207 * process if there's a concurrent fork, thus preventing the 208 * zfs_mount() from otherwise succeeding (the open file 209 * descriptor inherited by the child process will cause the 210 * parent's mount to fail with EBUSY). The performance 211 * implications of replacing the open, read, and close with a 212 * single stat is nice; but is not the main motivation for the 213 * added complexity. 214 */ 215 return (dir_is_empty_stat(dirname)); 216 } 217 218 /* 219 * Checks to see if the mount is active. If the filesystem is mounted, we fill 220 * in 'where' with the current mountpoint, and return 1. Otherwise, we return 221 * 0. 222 */ 223 boolean_t 224 is_mounted(libzfs_handle_t *zfs_hdl, const char *special, char **where) 225 { 226 struct mnttab entry; 227 228 if (libzfs_mnttab_find(zfs_hdl, special, &entry) != 0) 229 return (B_FALSE); 230 231 if (where != NULL) 232 *where = zfs_strdup(zfs_hdl, entry.mnt_mountp); 233 234 return (B_TRUE); 235 } 236 237 boolean_t 238 zfs_is_mounted(zfs_handle_t *zhp, char **where) 239 { 240 return (is_mounted(zhp->zfs_hdl, zfs_get_name(zhp), where)); 241 } 242 243 /* 244 * Checks any higher order concerns about whether the given dataset is 245 * mountable, false otherwise. zfs_is_mountable_internal specifically assumes 246 * that the caller has verified the sanity of mounting the dataset at 247 * its mountpoint to the extent the caller wants. 248 */ 249 static boolean_t 250 zfs_is_mountable_internal(zfs_handle_t *zhp) 251 { 252 if (zfs_prop_get_int(zhp, ZFS_PROP_ZONED) && 253 getzoneid() == GLOBAL_ZONEID) 254 return (B_FALSE); 255 256 return (B_TRUE); 257 } 258 259 /* 260 * Returns true if the given dataset is mountable, false otherwise. Returns the 261 * mountpoint in 'buf'. 262 */ 263 boolean_t 264 zfs_is_mountable(zfs_handle_t *zhp, char *buf, size_t buflen, 265 zprop_source_t *source, int flags) 266 { 267 char sourceloc[MAXNAMELEN]; 268 zprop_source_t sourcetype; 269 270 if (!zfs_prop_valid_for_type(ZFS_PROP_MOUNTPOINT, zhp->zfs_type, 271 B_FALSE)) 272 return (B_FALSE); 273 274 verify(zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, buf, buflen, 275 &sourcetype, sourceloc, sizeof (sourceloc), B_FALSE) == 0); 276 277 if (strcmp(buf, ZFS_MOUNTPOINT_NONE) == 0 || 278 strcmp(buf, ZFS_MOUNTPOINT_LEGACY) == 0) 279 return (B_FALSE); 280 281 if (zfs_prop_get_int(zhp, ZFS_PROP_CANMOUNT) == ZFS_CANMOUNT_OFF) 282 return (B_FALSE); 283 284 if (!zfs_is_mountable_internal(zhp)) 285 return (B_FALSE); 286 287 if (zfs_prop_get_int(zhp, ZFS_PROP_REDACTED) && !(flags & MS_FORCE)) 288 return (B_FALSE); 289 290 if (source) 291 *source = sourcetype; 292 293 return (B_TRUE); 294 } 295 296 /* 297 * The filesystem is mounted by invoking the system mount utility rather 298 * than by the system call mount(2). This ensures that the /etc/mtab 299 * file is correctly locked for the update. Performing our own locking 300 * and /etc/mtab update requires making an unsafe assumption about how 301 * the mount utility performs its locking. Unfortunately, this also means 302 * in the case of a mount failure we do not have the exact errno. We must 303 * make due with return value from the mount process. 304 * 305 * In the long term a shared library called libmount is under development 306 * which provides a common API to address the locking and errno issues. 307 * Once the standard mount utility has been updated to use this library 308 * we can add an autoconf check to conditionally use it. 309 * 310 * http://www.kernel.org/pub/linux/utils/util-linux/libmount-docs/index.html 311 */ 312 313 static int 314 zfs_add_option(zfs_handle_t *zhp, char *options, int len, 315 zfs_prop_t prop, char *on, char *off) 316 { 317 char *source; 318 uint64_t value; 319 320 /* Skip adding duplicate default options */ 321 if ((strstr(options, on) != NULL) || (strstr(options, off) != NULL)) 322 return (0); 323 324 /* 325 * zfs_prop_get_int() is not used to ensure our mount options 326 * are not influenced by the current /proc/self/mounts contents. 327 */ 328 value = getprop_uint64(zhp, prop, &source); 329 330 (void) strlcat(options, ",", len); 331 (void) strlcat(options, value ? on : off, len); 332 333 return (0); 334 } 335 336 static int 337 zfs_add_options(zfs_handle_t *zhp, char *options, int len) 338 { 339 int error = 0; 340 341 error = zfs_add_option(zhp, options, len, 342 ZFS_PROP_ATIME, MNTOPT_ATIME, MNTOPT_NOATIME); 343 /* 344 * don't add relatime/strictatime when atime=off, otherwise strictatime 345 * will force atime=on 346 */ 347 if (strstr(options, MNTOPT_NOATIME) == NULL) { 348 error = zfs_add_option(zhp, options, len, 349 ZFS_PROP_RELATIME, MNTOPT_RELATIME, MNTOPT_STRICTATIME); 350 } 351 error = error ? error : zfs_add_option(zhp, options, len, 352 ZFS_PROP_DEVICES, MNTOPT_DEVICES, MNTOPT_NODEVICES); 353 error = error ? error : zfs_add_option(zhp, options, len, 354 ZFS_PROP_EXEC, MNTOPT_EXEC, MNTOPT_NOEXEC); 355 error = error ? error : zfs_add_option(zhp, options, len, 356 ZFS_PROP_READONLY, MNTOPT_RO, MNTOPT_RW); 357 error = error ? error : zfs_add_option(zhp, options, len, 358 ZFS_PROP_SETUID, MNTOPT_SETUID, MNTOPT_NOSETUID); 359 error = error ? error : zfs_add_option(zhp, options, len, 360 ZFS_PROP_NBMAND, MNTOPT_NBMAND, MNTOPT_NONBMAND); 361 362 return (error); 363 } 364 365 int 366 zfs_mount(zfs_handle_t *zhp, const char *options, int flags) 367 { 368 char mountpoint[ZFS_MAXPROPLEN]; 369 370 if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), NULL, 371 flags)) 372 return (0); 373 374 return (zfs_mount_at(zhp, options, flags, mountpoint)); 375 } 376 377 /* 378 * Mount the given filesystem. 379 */ 380 int 381 zfs_mount_at(zfs_handle_t *zhp, const char *options, int flags, 382 const char *mountpoint) 383 { 384 struct stat buf; 385 char mntopts[MNT_LINE_MAX]; 386 char overlay[ZFS_MAXPROPLEN]; 387 char prop_encroot[MAXNAMELEN]; 388 boolean_t is_encroot; 389 zfs_handle_t *encroot_hp = zhp; 390 libzfs_handle_t *hdl = zhp->zfs_hdl; 391 uint64_t keystatus; 392 int remount = 0, rc; 393 394 if (options == NULL) { 395 (void) strlcpy(mntopts, MNTOPT_DEFAULTS, sizeof (mntopts)); 396 } else { 397 (void) strlcpy(mntopts, options, sizeof (mntopts)); 398 } 399 400 if (strstr(mntopts, MNTOPT_REMOUNT) != NULL) 401 remount = 1; 402 403 /* Potentially duplicates some checks if invoked by zfs_mount(). */ 404 if (!zfs_is_mountable_internal(zhp)) 405 return (0); 406 407 /* 408 * If the pool is imported read-only then all mounts must be read-only 409 */ 410 if (zpool_get_prop_int(zhp->zpool_hdl, ZPOOL_PROP_READONLY, NULL)) 411 (void) strlcat(mntopts, "," MNTOPT_RO, sizeof (mntopts)); 412 413 /* 414 * Append default mount options which apply to the mount point. 415 * This is done because under Linux (unlike Solaris) multiple mount 416 * points may reference a single super block. This means that just 417 * given a super block there is no back reference to update the per 418 * mount point options. 419 */ 420 rc = zfs_add_options(zhp, mntopts, sizeof (mntopts)); 421 if (rc) { 422 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 423 "default options unavailable")); 424 return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED, 425 dgettext(TEXT_DOMAIN, "cannot mount '%s'"), 426 mountpoint)); 427 } 428 429 /* 430 * If the filesystem is encrypted the key must be loaded in order to 431 * mount. If the key isn't loaded, the MS_CRYPT flag decides whether 432 * or not we attempt to load the keys. Note: we must call 433 * zfs_refresh_properties() here since some callers of this function 434 * (most notably zpool_enable_datasets()) may implicitly load our key 435 * by loading the parent's key first. 436 */ 437 if (zfs_prop_get_int(zhp, ZFS_PROP_ENCRYPTION) != ZIO_CRYPT_OFF) { 438 zfs_refresh_properties(zhp); 439 keystatus = zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS); 440 441 /* 442 * If the key is unavailable and MS_CRYPT is set give the 443 * user a chance to enter the key. Otherwise just fail 444 * immediately. 445 */ 446 if (keystatus == ZFS_KEYSTATUS_UNAVAILABLE) { 447 if (flags & MS_CRYPT) { 448 rc = zfs_crypto_get_encryption_root(zhp, 449 &is_encroot, prop_encroot); 450 if (rc) { 451 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 452 "Failed to get encryption root for " 453 "'%s'."), zfs_get_name(zhp)); 454 return (rc); 455 } 456 457 if (!is_encroot) { 458 encroot_hp = zfs_open(hdl, prop_encroot, 459 ZFS_TYPE_DATASET); 460 if (encroot_hp == NULL) 461 return (hdl->libzfs_error); 462 } 463 464 rc = zfs_crypto_load_key(encroot_hp, 465 B_FALSE, NULL); 466 467 if (!is_encroot) 468 zfs_close(encroot_hp); 469 if (rc) 470 return (rc); 471 } else { 472 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 473 "encryption key not loaded")); 474 return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED, 475 dgettext(TEXT_DOMAIN, "cannot mount '%s'"), 476 mountpoint)); 477 } 478 } 479 480 } 481 482 /* 483 * Append zfsutil option so the mount helper allow the mount 484 */ 485 strlcat(mntopts, "," MNTOPT_ZFSUTIL, sizeof (mntopts)); 486 487 /* Create the directory if it doesn't already exist */ 488 if (lstat(mountpoint, &buf) != 0) { 489 if (mkdirp(mountpoint, 0755) != 0) { 490 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 491 "failed to create mountpoint: %s"), 492 strerror(errno)); 493 return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED, 494 dgettext(TEXT_DOMAIN, "cannot mount '%s'"), 495 mountpoint)); 496 } 497 } 498 499 /* 500 * Overlay mounts are enabled by default but may be disabled 501 * via the 'overlay' property. The -O flag remains for compatibility. 502 */ 503 if (!(flags & MS_OVERLAY)) { 504 if (zfs_prop_get(zhp, ZFS_PROP_OVERLAY, overlay, 505 sizeof (overlay), NULL, NULL, 0, B_FALSE) == 0) { 506 if (strcmp(overlay, "on") == 0) { 507 flags |= MS_OVERLAY; 508 } 509 } 510 } 511 512 /* 513 * Determine if the mountpoint is empty. If so, refuse to perform the 514 * mount. We don't perform this check if 'remount' is 515 * specified or if overlay option (-O) is given 516 */ 517 if ((flags & MS_OVERLAY) == 0 && !remount && 518 !dir_is_empty(mountpoint)) { 519 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 520 "directory is not empty")); 521 return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED, 522 dgettext(TEXT_DOMAIN, "cannot mount '%s'"), mountpoint)); 523 } 524 525 /* perform the mount */ 526 rc = do_mount(zhp, mountpoint, mntopts, flags); 527 if (rc) { 528 /* 529 * Generic errors are nasty, but there are just way too many 530 * from mount(), and they're well-understood. We pick a few 531 * common ones to improve upon. 532 */ 533 if (rc == EBUSY) { 534 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 535 "mountpoint or dataset is busy")); 536 } else if (rc == EPERM) { 537 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 538 "Insufficient privileges")); 539 } else if (rc == ENOTSUP) { 540 int spa_version; 541 542 VERIFY(zfs_spa_version(zhp, &spa_version) == 0); 543 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, 544 "Can't mount a version %llu " 545 "file system on a version %d pool. Pool must be" 546 " upgraded to mount this file system."), 547 (u_longlong_t)zfs_prop_get_int(zhp, 548 ZFS_PROP_VERSION), spa_version); 549 } else { 550 zfs_error_aux(hdl, "%s", strerror(rc)); 551 } 552 return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED, 553 dgettext(TEXT_DOMAIN, "cannot mount '%s'"), 554 zhp->zfs_name)); 555 } 556 557 /* remove the mounted entry before re-adding on remount */ 558 if (remount) 559 libzfs_mnttab_remove(hdl, zhp->zfs_name); 560 561 /* add the mounted entry into our cache */ 562 libzfs_mnttab_add(hdl, zfs_get_name(zhp), mountpoint, mntopts); 563 return (0); 564 } 565 566 /* 567 * Unmount a single filesystem. 568 */ 569 static int 570 unmount_one(zfs_handle_t *zhp, const char *mountpoint, int flags) 571 { 572 int error; 573 574 error = do_unmount(zhp, mountpoint, flags); 575 if (error != 0) { 576 int libzfs_err; 577 578 switch (error) { 579 case EBUSY: 580 libzfs_err = EZFS_BUSY; 581 break; 582 case EIO: 583 libzfs_err = EZFS_IO; 584 break; 585 case ENOENT: 586 libzfs_err = EZFS_NOENT; 587 break; 588 case ENOMEM: 589 libzfs_err = EZFS_NOMEM; 590 break; 591 case EPERM: 592 libzfs_err = EZFS_PERM; 593 break; 594 default: 595 libzfs_err = EZFS_UMOUNTFAILED; 596 } 597 if (zhp) { 598 return (zfs_error_fmt(zhp->zfs_hdl, libzfs_err, 599 dgettext(TEXT_DOMAIN, "cannot unmount '%s'"), 600 mountpoint)); 601 } else { 602 return (-1); 603 } 604 } 605 606 return (0); 607 } 608 609 /* 610 * Unmount the given filesystem. 611 */ 612 int 613 zfs_unmount(zfs_handle_t *zhp, const char *mountpoint, int flags) 614 { 615 libzfs_handle_t *hdl = zhp->zfs_hdl; 616 struct mnttab entry; 617 char *mntpt = NULL; 618 boolean_t encroot, unmounted = B_FALSE; 619 620 /* check to see if we need to unmount the filesystem */ 621 if (mountpoint != NULL || ((zfs_get_type(zhp) == ZFS_TYPE_FILESYSTEM) && 622 libzfs_mnttab_find(hdl, zhp->zfs_name, &entry) == 0)) { 623 /* 624 * mountpoint may have come from a call to 625 * getmnt/getmntany if it isn't NULL. If it is NULL, 626 * we know it comes from libzfs_mnttab_find which can 627 * then get freed later. We strdup it to play it safe. 628 */ 629 if (mountpoint == NULL) 630 mntpt = zfs_strdup(hdl, entry.mnt_mountp); 631 else 632 mntpt = zfs_strdup(hdl, mountpoint); 633 634 /* 635 * Unshare and unmount the filesystem 636 */ 637 if (zfs_unshare_proto(zhp, mntpt, share_all_proto) != 0) { 638 free(mntpt); 639 return (-1); 640 } 641 zfs_commit_all_shares(); 642 643 if (unmount_one(zhp, mntpt, flags) != 0) { 644 free(mntpt); 645 (void) zfs_shareall(zhp); 646 zfs_commit_all_shares(); 647 return (-1); 648 } 649 650 libzfs_mnttab_remove(hdl, zhp->zfs_name); 651 free(mntpt); 652 unmounted = B_TRUE; 653 } 654 655 /* 656 * If the MS_CRYPT flag is provided we must ensure we attempt to 657 * unload the dataset's key regardless of whether we did any work 658 * to unmount it. We only do this for encryption roots. 659 */ 660 if ((flags & MS_CRYPT) != 0 && 661 zfs_prop_get_int(zhp, ZFS_PROP_ENCRYPTION) != ZIO_CRYPT_OFF) { 662 zfs_refresh_properties(zhp); 663 664 if (zfs_crypto_get_encryption_root(zhp, &encroot, NULL) != 0 && 665 unmounted) { 666 (void) zfs_mount(zhp, NULL, 0); 667 return (-1); 668 } 669 670 if (encroot && zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) == 671 ZFS_KEYSTATUS_AVAILABLE && 672 zfs_crypto_unload_key(zhp) != 0) { 673 (void) zfs_mount(zhp, NULL, 0); 674 return (-1); 675 } 676 } 677 678 zpool_disable_volume_os(zhp->zfs_name); 679 680 return (0); 681 } 682 683 /* 684 * Unmount this filesystem and any children inheriting the mountpoint property. 685 * To do this, just act like we're changing the mountpoint property, but don't 686 * remount the filesystems afterwards. 687 */ 688 int 689 zfs_unmountall(zfs_handle_t *zhp, int flags) 690 { 691 prop_changelist_t *clp; 692 int ret; 693 694 clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT, 695 CL_GATHER_ITER_MOUNTED, flags); 696 if (clp == NULL) 697 return (-1); 698 699 ret = changelist_prefix(clp); 700 changelist_free(clp); 701 702 return (ret); 703 } 704 705 boolean_t 706 zfs_is_shared(zfs_handle_t *zhp) 707 { 708 zfs_share_type_t rc = 0; 709 const zfs_share_proto_t *curr_proto; 710 711 if (ZFS_IS_VOLUME(zhp)) 712 return (B_FALSE); 713 714 for (curr_proto = share_all_proto; *curr_proto != PROTO_END; 715 curr_proto++) 716 rc |= zfs_is_shared_proto(zhp, NULL, *curr_proto); 717 718 return (rc ? B_TRUE : B_FALSE); 719 } 720 721 /* 722 * Unshare a filesystem by mountpoint. 723 */ 724 int 725 unshare_one(libzfs_handle_t *hdl, const char *name, const char *mountpoint, 726 zfs_share_proto_t proto) 727 { 728 int err; 729 730 err = sa_disable_share(mountpoint, proto_table[proto].p_name); 731 if (err != SA_OK) { 732 return (zfs_error_fmt(hdl, proto_table[proto].p_unshare_err, 733 dgettext(TEXT_DOMAIN, "cannot unshare '%s': %s"), 734 name, sa_errorstr(err))); 735 } 736 return (0); 737 } 738 739 /* 740 * Query libshare for the given mountpoint and protocol, returning 741 * a zfs_share_type_t value. 742 */ 743 zfs_share_type_t 744 is_shared(const char *mountpoint, zfs_share_proto_t proto) 745 { 746 if (sa_is_shared(mountpoint, proto_table[proto].p_name)) { 747 switch (proto) { 748 case PROTO_NFS: 749 return (SHARED_NFS); 750 case PROTO_SMB: 751 return (SHARED_SMB); 752 default: 753 return (SHARED_NOT_SHARED); 754 } 755 } 756 return (SHARED_NOT_SHARED); 757 } 758 759 /* 760 * Share the given filesystem according to the options in the specified 761 * protocol specific properties (sharenfs, sharesmb). We rely 762 * on "libshare" to do the dirty work for us. 763 */ 764 int 765 zfs_share_proto(zfs_handle_t *zhp, const zfs_share_proto_t *proto) 766 { 767 char mountpoint[ZFS_MAXPROPLEN]; 768 char shareopts[ZFS_MAXPROPLEN]; 769 char sourcestr[ZFS_MAXPROPLEN]; 770 const zfs_share_proto_t *curr_proto; 771 zprop_source_t sourcetype; 772 int err = 0; 773 774 if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), NULL, 0)) 775 return (0); 776 777 for (curr_proto = proto; *curr_proto != PROTO_END; curr_proto++) { 778 /* 779 * Return success if there are no share options. 780 */ 781 if (zfs_prop_get(zhp, proto_table[*curr_proto].p_prop, 782 shareopts, sizeof (shareopts), &sourcetype, sourcestr, 783 ZFS_MAXPROPLEN, B_FALSE) != 0 || 784 strcmp(shareopts, "off") == 0) 785 continue; 786 787 /* 788 * If the 'zoned' property is set, then zfs_is_mountable() 789 * will have already bailed out if we are in the global zone. 790 * But local zones cannot be NFS servers, so we ignore it for 791 * local zones as well. 792 */ 793 if (zfs_prop_get_int(zhp, ZFS_PROP_ZONED)) 794 continue; 795 796 err = sa_enable_share(zfs_get_name(zhp), mountpoint, shareopts, 797 proto_table[*curr_proto].p_name); 798 if (err != SA_OK) { 799 return (zfs_error_fmt(zhp->zfs_hdl, 800 proto_table[*curr_proto].p_share_err, 801 dgettext(TEXT_DOMAIN, "cannot share '%s: %s'"), 802 zfs_get_name(zhp), sa_errorstr(err))); 803 } 804 805 } 806 return (0); 807 } 808 809 int 810 zfs_share(zfs_handle_t *zhp) 811 { 812 assert(!ZFS_IS_VOLUME(zhp)); 813 return (zfs_share_proto(zhp, share_all_proto)); 814 } 815 816 int 817 zfs_unshare(zfs_handle_t *zhp) 818 { 819 assert(!ZFS_IS_VOLUME(zhp)); 820 return (zfs_unshareall(zhp)); 821 } 822 823 /* 824 * Check to see if the filesystem is currently shared. 825 */ 826 static zfs_share_type_t 827 zfs_is_shared_proto(zfs_handle_t *zhp, char **where, zfs_share_proto_t proto) 828 { 829 char *mountpoint; 830 zfs_share_type_t rc; 831 832 if (!zfs_is_mounted(zhp, &mountpoint)) 833 return (SHARED_NOT_SHARED); 834 835 if ((rc = is_shared(mountpoint, proto)) 836 != SHARED_NOT_SHARED) { 837 if (where != NULL) 838 *where = mountpoint; 839 else 840 free(mountpoint); 841 return (rc); 842 } else { 843 free(mountpoint); 844 return (SHARED_NOT_SHARED); 845 } 846 } 847 848 boolean_t 849 zfs_is_shared_nfs(zfs_handle_t *zhp, char **where) 850 { 851 return (zfs_is_shared_proto(zhp, where, 852 PROTO_NFS) != SHARED_NOT_SHARED); 853 } 854 855 boolean_t 856 zfs_is_shared_smb(zfs_handle_t *zhp, char **where) 857 { 858 return (zfs_is_shared_proto(zhp, where, 859 PROTO_SMB) != SHARED_NOT_SHARED); 860 } 861 862 /* 863 * zfs_parse_options(options, proto) 864 * 865 * Call the legacy parse interface to get the protocol specific 866 * options using the NULL arg to indicate that this is a "parse" only. 867 */ 868 int 869 zfs_parse_options(char *options, zfs_share_proto_t proto) 870 { 871 return (sa_validate_shareopts(options, proto_table[proto].p_name)); 872 } 873 874 void 875 zfs_commit_proto(const zfs_share_proto_t *proto) 876 { 877 const zfs_share_proto_t *curr_proto; 878 for (curr_proto = proto; *curr_proto != PROTO_END; curr_proto++) 879 sa_commit_shares(proto_table[*curr_proto].p_name); 880 } 881 882 void 883 zfs_commit_nfs_shares(void) 884 { 885 zfs_commit_proto(nfs_only); 886 } 887 888 void 889 zfs_commit_smb_shares(void) 890 { 891 zfs_commit_proto(smb_only); 892 } 893 894 void 895 zfs_commit_all_shares(void) 896 { 897 zfs_commit_proto(share_all_proto); 898 } 899 900 void 901 zfs_commit_shares(const char *proto) 902 { 903 if (proto == NULL) 904 zfs_commit_proto(share_all_proto); 905 else if (strcmp(proto, "nfs") == 0) 906 zfs_commit_proto(nfs_only); 907 else if (strcmp(proto, "smb") == 0) 908 zfs_commit_proto(smb_only); 909 } 910 911 int 912 zfs_share_nfs(zfs_handle_t *zhp) 913 { 914 return (zfs_share_proto(zhp, nfs_only)); 915 } 916 917 int 918 zfs_share_smb(zfs_handle_t *zhp) 919 { 920 return (zfs_share_proto(zhp, smb_only)); 921 } 922 923 int 924 zfs_shareall(zfs_handle_t *zhp) 925 { 926 return (zfs_share_proto(zhp, share_all_proto)); 927 } 928 929 /* 930 * Unshare the given filesystem. 931 */ 932 int 933 zfs_unshare_proto(zfs_handle_t *zhp, const char *mountpoint, 934 const zfs_share_proto_t *proto) 935 { 936 libzfs_handle_t *hdl = zhp->zfs_hdl; 937 struct mnttab entry; 938 char *mntpt = NULL; 939 940 /* check to see if need to unmount the filesystem */ 941 if (mountpoint != NULL) 942 mntpt = zfs_strdup(hdl, mountpoint); 943 944 if (mountpoint != NULL || ((zfs_get_type(zhp) == ZFS_TYPE_FILESYSTEM) && 945 libzfs_mnttab_find(hdl, zfs_get_name(zhp), &entry) == 0)) { 946 const zfs_share_proto_t *curr_proto; 947 948 if (mountpoint == NULL) 949 mntpt = zfs_strdup(zhp->zfs_hdl, entry.mnt_mountp); 950 951 for (curr_proto = proto; *curr_proto != PROTO_END; 952 curr_proto++) { 953 954 if (is_shared(mntpt, *curr_proto)) { 955 if (unshare_one(hdl, zhp->zfs_name, 956 mntpt, *curr_proto) != 0) { 957 if (mntpt != NULL) 958 free(mntpt); 959 return (-1); 960 } 961 } 962 } 963 } 964 if (mntpt != NULL) 965 free(mntpt); 966 967 return (0); 968 } 969 970 int 971 zfs_unshare_nfs(zfs_handle_t *zhp, const char *mountpoint) 972 { 973 return (zfs_unshare_proto(zhp, mountpoint, nfs_only)); 974 } 975 976 int 977 zfs_unshare_smb(zfs_handle_t *zhp, const char *mountpoint) 978 { 979 return (zfs_unshare_proto(zhp, mountpoint, smb_only)); 980 } 981 982 /* 983 * Same as zfs_unmountall(), but for NFS and SMB unshares. 984 */ 985 static int 986 zfs_unshareall_proto(zfs_handle_t *zhp, const zfs_share_proto_t *proto) 987 { 988 prop_changelist_t *clp; 989 int ret; 990 991 clp = changelist_gather(zhp, ZFS_PROP_SHARENFS, 0, 0); 992 if (clp == NULL) 993 return (-1); 994 995 ret = changelist_unshare(clp, proto); 996 changelist_free(clp); 997 998 return (ret); 999 } 1000 1001 int 1002 zfs_unshareall_nfs(zfs_handle_t *zhp) 1003 { 1004 return (zfs_unshareall_proto(zhp, nfs_only)); 1005 } 1006 1007 int 1008 zfs_unshareall_smb(zfs_handle_t *zhp) 1009 { 1010 return (zfs_unshareall_proto(zhp, smb_only)); 1011 } 1012 1013 int 1014 zfs_unshareall(zfs_handle_t *zhp) 1015 { 1016 return (zfs_unshareall_proto(zhp, share_all_proto)); 1017 } 1018 1019 int 1020 zfs_unshareall_bypath(zfs_handle_t *zhp, const char *mountpoint) 1021 { 1022 return (zfs_unshare_proto(zhp, mountpoint, share_all_proto)); 1023 } 1024 1025 int 1026 zfs_unshareall_bytype(zfs_handle_t *zhp, const char *mountpoint, 1027 const char *proto) 1028 { 1029 if (proto == NULL) 1030 return (zfs_unshare_proto(zhp, mountpoint, share_all_proto)); 1031 if (strcmp(proto, "nfs") == 0) 1032 return (zfs_unshare_proto(zhp, mountpoint, nfs_only)); 1033 else if (strcmp(proto, "smb") == 0) 1034 return (zfs_unshare_proto(zhp, mountpoint, smb_only)); 1035 else 1036 return (1); 1037 } 1038 1039 /* 1040 * Remove the mountpoint associated with the current dataset, if necessary. 1041 * We only remove the underlying directory if: 1042 * 1043 * - The mountpoint is not 'none' or 'legacy' 1044 * - The mountpoint is non-empty 1045 * - The mountpoint is the default or inherited 1046 * - The 'zoned' property is set, or we're in a local zone 1047 * 1048 * Any other directories we leave alone. 1049 */ 1050 void 1051 remove_mountpoint(zfs_handle_t *zhp) 1052 { 1053 char mountpoint[ZFS_MAXPROPLEN]; 1054 zprop_source_t source; 1055 1056 if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), 1057 &source, 0)) 1058 return; 1059 1060 if (source == ZPROP_SRC_DEFAULT || 1061 source == ZPROP_SRC_INHERITED) { 1062 /* 1063 * Try to remove the directory, silently ignoring any errors. 1064 * The filesystem may have since been removed or moved around, 1065 * and this error isn't really useful to the administrator in 1066 * any way. 1067 */ 1068 (void) rmdir(mountpoint); 1069 } 1070 } 1071 1072 /* 1073 * Add the given zfs handle to the cb_handles array, dynamically reallocating 1074 * the array if it is out of space. 1075 */ 1076 void 1077 libzfs_add_handle(get_all_cb_t *cbp, zfs_handle_t *zhp) 1078 { 1079 if (cbp->cb_alloc == cbp->cb_used) { 1080 size_t newsz; 1081 zfs_handle_t **newhandles; 1082 1083 newsz = cbp->cb_alloc != 0 ? cbp->cb_alloc * 2 : 64; 1084 newhandles = zfs_realloc(zhp->zfs_hdl, 1085 cbp->cb_handles, cbp->cb_alloc * sizeof (zfs_handle_t *), 1086 newsz * sizeof (zfs_handle_t *)); 1087 cbp->cb_handles = newhandles; 1088 cbp->cb_alloc = newsz; 1089 } 1090 cbp->cb_handles[cbp->cb_used++] = zhp; 1091 } 1092 1093 /* 1094 * Recursive helper function used during file system enumeration 1095 */ 1096 static int 1097 zfs_iter_cb(zfs_handle_t *zhp, void *data) 1098 { 1099 get_all_cb_t *cbp = data; 1100 1101 if (!(zfs_get_type(zhp) & ZFS_TYPE_FILESYSTEM)) { 1102 zfs_close(zhp); 1103 return (0); 1104 } 1105 1106 if (zfs_prop_get_int(zhp, ZFS_PROP_CANMOUNT) == ZFS_CANMOUNT_NOAUTO) { 1107 zfs_close(zhp); 1108 return (0); 1109 } 1110 1111 if (zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) == 1112 ZFS_KEYSTATUS_UNAVAILABLE) { 1113 zfs_close(zhp); 1114 return (0); 1115 } 1116 1117 /* 1118 * If this filesystem is inconsistent and has a receive resume 1119 * token, we can not mount it. 1120 */ 1121 if (zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT) && 1122 zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN, 1123 NULL, 0, NULL, NULL, 0, B_TRUE) == 0) { 1124 zfs_close(zhp); 1125 return (0); 1126 } 1127 1128 libzfs_add_handle(cbp, zhp); 1129 if (zfs_iter_filesystems(zhp, zfs_iter_cb, cbp) != 0) { 1130 zfs_close(zhp); 1131 return (-1); 1132 } 1133 return (0); 1134 } 1135 1136 /* 1137 * Sort comparator that compares two mountpoint paths. We sort these paths so 1138 * that subdirectories immediately follow their parents. This means that we 1139 * effectively treat the '/' character as the lowest value non-nul char. 1140 * Since filesystems from non-global zones can have the same mountpoint 1141 * as other filesystems, the comparator sorts global zone filesystems to 1142 * the top of the list. This means that the global zone will traverse the 1143 * filesystem list in the correct order and can stop when it sees the 1144 * first zoned filesystem. In a non-global zone, only the delegated 1145 * filesystems are seen. 1146 * 1147 * An example sorted list using this comparator would look like: 1148 * 1149 * /foo 1150 * /foo/bar 1151 * /foo/bar/baz 1152 * /foo/baz 1153 * /foo.bar 1154 * /foo (NGZ1) 1155 * /foo (NGZ2) 1156 * 1157 * The mounting code depends on this ordering to deterministically iterate 1158 * over filesystems in order to spawn parallel mount tasks. 1159 */ 1160 static int 1161 mountpoint_cmp(const void *arga, const void *argb) 1162 { 1163 zfs_handle_t *const *zap = arga; 1164 zfs_handle_t *za = *zap; 1165 zfs_handle_t *const *zbp = argb; 1166 zfs_handle_t *zb = *zbp; 1167 char mounta[MAXPATHLEN]; 1168 char mountb[MAXPATHLEN]; 1169 const char *a = mounta; 1170 const char *b = mountb; 1171 boolean_t gota, gotb; 1172 uint64_t zoneda, zonedb; 1173 1174 zoneda = zfs_prop_get_int(za, ZFS_PROP_ZONED); 1175 zonedb = zfs_prop_get_int(zb, ZFS_PROP_ZONED); 1176 if (zoneda && !zonedb) 1177 return (1); 1178 if (!zoneda && zonedb) 1179 return (-1); 1180 1181 gota = (zfs_get_type(za) == ZFS_TYPE_FILESYSTEM); 1182 if (gota) { 1183 verify(zfs_prop_get(za, ZFS_PROP_MOUNTPOINT, mounta, 1184 sizeof (mounta), NULL, NULL, 0, B_FALSE) == 0); 1185 } 1186 gotb = (zfs_get_type(zb) == ZFS_TYPE_FILESYSTEM); 1187 if (gotb) { 1188 verify(zfs_prop_get(zb, ZFS_PROP_MOUNTPOINT, mountb, 1189 sizeof (mountb), NULL, NULL, 0, B_FALSE) == 0); 1190 } 1191 1192 if (gota && gotb) { 1193 while (*a != '\0' && (*a == *b)) { 1194 a++; 1195 b++; 1196 } 1197 if (*a == *b) 1198 return (0); 1199 if (*a == '\0') 1200 return (-1); 1201 if (*b == '\0') 1202 return (1); 1203 if (*a == '/') 1204 return (-1); 1205 if (*b == '/') 1206 return (1); 1207 return (*a < *b ? -1 : *a > *b); 1208 } 1209 1210 if (gota) 1211 return (-1); 1212 if (gotb) 1213 return (1); 1214 1215 /* 1216 * If neither filesystem has a mountpoint, revert to sorting by 1217 * dataset name. 1218 */ 1219 return (strcmp(zfs_get_name(za), zfs_get_name(zb))); 1220 } 1221 1222 /* 1223 * Return true if path2 is a child of path1 or path2 equals path1 or 1224 * path1 is "/" (path2 is always a child of "/"). 1225 */ 1226 static boolean_t 1227 libzfs_path_contains(const char *path1, const char *path2) 1228 { 1229 return (strcmp(path1, path2) == 0 || strcmp(path1, "/") == 0 || 1230 (strstr(path2, path1) == path2 && path2[strlen(path1)] == '/')); 1231 } 1232 1233 /* 1234 * Given a mountpoint specified by idx in the handles array, find the first 1235 * non-descendent of that mountpoint and return its index. Descendant paths 1236 * start with the parent's path. This function relies on the ordering 1237 * enforced by mountpoint_cmp(). 1238 */ 1239 static int 1240 non_descendant_idx(zfs_handle_t **handles, size_t num_handles, int idx) 1241 { 1242 char parent[ZFS_MAXPROPLEN]; 1243 char child[ZFS_MAXPROPLEN]; 1244 int i; 1245 1246 verify(zfs_prop_get(handles[idx], ZFS_PROP_MOUNTPOINT, parent, 1247 sizeof (parent), NULL, NULL, 0, B_FALSE) == 0); 1248 1249 for (i = idx + 1; i < num_handles; i++) { 1250 verify(zfs_prop_get(handles[i], ZFS_PROP_MOUNTPOINT, child, 1251 sizeof (child), NULL, NULL, 0, B_FALSE) == 0); 1252 if (!libzfs_path_contains(parent, child)) 1253 break; 1254 } 1255 return (i); 1256 } 1257 1258 typedef struct mnt_param { 1259 libzfs_handle_t *mnt_hdl; 1260 tpool_t *mnt_tp; 1261 zfs_handle_t **mnt_zhps; /* filesystems to mount */ 1262 size_t mnt_num_handles; 1263 int mnt_idx; /* Index of selected entry to mount */ 1264 zfs_iter_f mnt_func; 1265 void *mnt_data; 1266 } mnt_param_t; 1267 1268 /* 1269 * Allocate and populate the parameter struct for mount function, and 1270 * schedule mounting of the entry selected by idx. 1271 */ 1272 static void 1273 zfs_dispatch_mount(libzfs_handle_t *hdl, zfs_handle_t **handles, 1274 size_t num_handles, int idx, zfs_iter_f func, void *data, tpool_t *tp) 1275 { 1276 mnt_param_t *mnt_param = zfs_alloc(hdl, sizeof (mnt_param_t)); 1277 1278 mnt_param->mnt_hdl = hdl; 1279 mnt_param->mnt_tp = tp; 1280 mnt_param->mnt_zhps = handles; 1281 mnt_param->mnt_num_handles = num_handles; 1282 mnt_param->mnt_idx = idx; 1283 mnt_param->mnt_func = func; 1284 mnt_param->mnt_data = data; 1285 1286 (void) tpool_dispatch(tp, zfs_mount_task, (void*)mnt_param); 1287 } 1288 1289 /* 1290 * This is the structure used to keep state of mounting or sharing operations 1291 * during a call to zpool_enable_datasets(). 1292 */ 1293 typedef struct mount_state { 1294 /* 1295 * ms_mntstatus is set to -1 if any mount fails. While multiple threads 1296 * could update this variable concurrently, no synchronization is 1297 * needed as it's only ever set to -1. 1298 */ 1299 int ms_mntstatus; 1300 int ms_mntflags; 1301 const char *ms_mntopts; 1302 } mount_state_t; 1303 1304 static int 1305 zfs_mount_one(zfs_handle_t *zhp, void *arg) 1306 { 1307 mount_state_t *ms = arg; 1308 int ret = 0; 1309 1310 /* 1311 * don't attempt to mount encrypted datasets with 1312 * unloaded keys 1313 */ 1314 if (zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) == 1315 ZFS_KEYSTATUS_UNAVAILABLE) 1316 return (0); 1317 1318 if (zfs_mount(zhp, ms->ms_mntopts, ms->ms_mntflags) != 0) 1319 ret = ms->ms_mntstatus = -1; 1320 return (ret); 1321 } 1322 1323 static int 1324 zfs_share_one(zfs_handle_t *zhp, void *arg) 1325 { 1326 mount_state_t *ms = arg; 1327 int ret = 0; 1328 1329 if (zfs_share(zhp) != 0) 1330 ret = ms->ms_mntstatus = -1; 1331 return (ret); 1332 } 1333 1334 /* 1335 * Thread pool function to mount one file system. On completion, it finds and 1336 * schedules its children to be mounted. This depends on the sorting done in 1337 * zfs_foreach_mountpoint(). Note that the degenerate case (chain of entries 1338 * each descending from the previous) will have no parallelism since we always 1339 * have to wait for the parent to finish mounting before we can schedule 1340 * its children. 1341 */ 1342 static void 1343 zfs_mount_task(void *arg) 1344 { 1345 mnt_param_t *mp = arg; 1346 int idx = mp->mnt_idx; 1347 zfs_handle_t **handles = mp->mnt_zhps; 1348 size_t num_handles = mp->mnt_num_handles; 1349 char mountpoint[ZFS_MAXPROPLEN]; 1350 1351 verify(zfs_prop_get(handles[idx], ZFS_PROP_MOUNTPOINT, mountpoint, 1352 sizeof (mountpoint), NULL, NULL, 0, B_FALSE) == 0); 1353 1354 if (mp->mnt_func(handles[idx], mp->mnt_data) != 0) 1355 goto out; 1356 1357 /* 1358 * We dispatch tasks to mount filesystems with mountpoints underneath 1359 * this one. We do this by dispatching the next filesystem with a 1360 * descendant mountpoint of the one we just mounted, then skip all of 1361 * its descendants, dispatch the next descendant mountpoint, and so on. 1362 * The non_descendant_idx() function skips over filesystems that are 1363 * descendants of the filesystem we just dispatched. 1364 */ 1365 for (int i = idx + 1; i < num_handles; 1366 i = non_descendant_idx(handles, num_handles, i)) { 1367 char child[ZFS_MAXPROPLEN]; 1368 verify(zfs_prop_get(handles[i], ZFS_PROP_MOUNTPOINT, 1369 child, sizeof (child), NULL, NULL, 0, B_FALSE) == 0); 1370 1371 if (!libzfs_path_contains(mountpoint, child)) 1372 break; /* not a descendant, return */ 1373 zfs_dispatch_mount(mp->mnt_hdl, handles, num_handles, i, 1374 mp->mnt_func, mp->mnt_data, mp->mnt_tp); 1375 } 1376 1377 out: 1378 free(mp); 1379 } 1380 1381 /* 1382 * Issue the func callback for each ZFS handle contained in the handles 1383 * array. This function is used to mount all datasets, and so this function 1384 * guarantees that filesystems for parent mountpoints are called before their 1385 * children. As such, before issuing any callbacks, we first sort the array 1386 * of handles by mountpoint. 1387 * 1388 * Callbacks are issued in one of two ways: 1389 * 1390 * 1. Sequentially: If the parallel argument is B_FALSE or the ZFS_SERIAL_MOUNT 1391 * environment variable is set, then we issue callbacks sequentially. 1392 * 1393 * 2. In parallel: If the parallel argument is B_TRUE and the ZFS_SERIAL_MOUNT 1394 * environment variable is not set, then we use a tpool to dispatch threads 1395 * to mount filesystems in parallel. This function dispatches tasks to mount 1396 * the filesystems at the top-level mountpoints, and these tasks in turn 1397 * are responsible for recursively mounting filesystems in their children 1398 * mountpoints. 1399 */ 1400 void 1401 zfs_foreach_mountpoint(libzfs_handle_t *hdl, zfs_handle_t **handles, 1402 size_t num_handles, zfs_iter_f func, void *data, boolean_t parallel) 1403 { 1404 zoneid_t zoneid = getzoneid(); 1405 1406 /* 1407 * The ZFS_SERIAL_MOUNT environment variable is an undocumented 1408 * variable that can be used as a convenience to do a/b comparison 1409 * of serial vs. parallel mounting. 1410 */ 1411 boolean_t serial_mount = !parallel || 1412 (getenv("ZFS_SERIAL_MOUNT") != NULL); 1413 1414 /* 1415 * Sort the datasets by mountpoint. See mountpoint_cmp for details 1416 * of how these are sorted. 1417 */ 1418 qsort(handles, num_handles, sizeof (zfs_handle_t *), mountpoint_cmp); 1419 1420 if (serial_mount) { 1421 for (int i = 0; i < num_handles; i++) { 1422 func(handles[i], data); 1423 } 1424 return; 1425 } 1426 1427 /* 1428 * Issue the callback function for each dataset using a parallel 1429 * algorithm that uses a thread pool to manage threads. 1430 */ 1431 tpool_t *tp = tpool_create(1, mount_tp_nthr, 0, NULL); 1432 1433 /* 1434 * There may be multiple "top level" mountpoints outside of the pool's 1435 * root mountpoint, e.g.: /foo /bar. Dispatch a mount task for each of 1436 * these. 1437 */ 1438 for (int i = 0; i < num_handles; 1439 i = non_descendant_idx(handles, num_handles, i)) { 1440 /* 1441 * Since the mountpoints have been sorted so that the zoned 1442 * filesystems are at the end, a zoned filesystem seen from 1443 * the global zone means that we're done. 1444 */ 1445 if (zoneid == GLOBAL_ZONEID && 1446 zfs_prop_get_int(handles[i], ZFS_PROP_ZONED)) 1447 break; 1448 zfs_dispatch_mount(hdl, handles, num_handles, i, func, data, 1449 tp); 1450 } 1451 1452 tpool_wait(tp); /* wait for all scheduled mounts to complete */ 1453 tpool_destroy(tp); 1454 } 1455 1456 /* 1457 * Mount and share all datasets within the given pool. This assumes that no 1458 * datasets within the pool are currently mounted. 1459 */ 1460 int 1461 zpool_enable_datasets(zpool_handle_t *zhp, const char *mntopts, int flags) 1462 { 1463 get_all_cb_t cb = { 0 }; 1464 mount_state_t ms = { 0 }; 1465 zfs_handle_t *zfsp; 1466 int ret = 0; 1467 1468 if ((zfsp = zfs_open(zhp->zpool_hdl, zhp->zpool_name, 1469 ZFS_TYPE_DATASET)) == NULL) 1470 goto out; 1471 1472 /* 1473 * Gather all non-snapshot datasets within the pool. Start by adding 1474 * the root filesystem for this pool to the list, and then iterate 1475 * over all child filesystems. 1476 */ 1477 libzfs_add_handle(&cb, zfsp); 1478 if (zfs_iter_filesystems(zfsp, zfs_iter_cb, &cb) != 0) 1479 goto out; 1480 1481 /* 1482 * Mount all filesystems 1483 */ 1484 ms.ms_mntopts = mntopts; 1485 ms.ms_mntflags = flags; 1486 zfs_foreach_mountpoint(zhp->zpool_hdl, cb.cb_handles, cb.cb_used, 1487 zfs_mount_one, &ms, B_TRUE); 1488 if (ms.ms_mntstatus != 0) 1489 ret = ms.ms_mntstatus; 1490 1491 /* 1492 * Share all filesystems that need to be shared. This needs to be 1493 * a separate pass because libshare is not mt-safe, and so we need 1494 * to share serially. 1495 */ 1496 ms.ms_mntstatus = 0; 1497 zfs_foreach_mountpoint(zhp->zpool_hdl, cb.cb_handles, cb.cb_used, 1498 zfs_share_one, &ms, B_FALSE); 1499 if (ms.ms_mntstatus != 0) 1500 ret = ms.ms_mntstatus; 1501 else 1502 zfs_commit_all_shares(); 1503 1504 out: 1505 for (int i = 0; i < cb.cb_used; i++) 1506 zfs_close(cb.cb_handles[i]); 1507 free(cb.cb_handles); 1508 1509 return (ret); 1510 } 1511 1512 struct sets_s { 1513 char *mountpoint; 1514 zfs_handle_t *dataset; 1515 }; 1516 1517 static int 1518 mountpoint_compare(const void *a, const void *b) 1519 { 1520 const struct sets_s *mounta = (struct sets_s *)a; 1521 const struct sets_s *mountb = (struct sets_s *)b; 1522 1523 return (strcmp(mountb->mountpoint, mounta->mountpoint)); 1524 } 1525 1526 /* 1527 * Unshare and unmount all datasets within the given pool. We don't want to 1528 * rely on traversing the DSL to discover the filesystems within the pool, 1529 * because this may be expensive (if not all of them are mounted), and can fail 1530 * arbitrarily (on I/O error, for example). Instead, we walk /proc/self/mounts 1531 * and gather all the filesystems that are currently mounted. 1532 */ 1533 int 1534 zpool_disable_datasets(zpool_handle_t *zhp, boolean_t force) 1535 { 1536 int used, alloc; 1537 FILE *mnttab; 1538 struct mnttab entry; 1539 size_t namelen; 1540 struct sets_s *sets = NULL; 1541 libzfs_handle_t *hdl = zhp->zpool_hdl; 1542 int i; 1543 int ret = -1; 1544 int flags = (force ? MS_FORCE : 0); 1545 1546 namelen = strlen(zhp->zpool_name); 1547 1548 if ((mnttab = fopen(MNTTAB, "re")) == NULL) 1549 return (ENOENT); 1550 1551 used = alloc = 0; 1552 while (getmntent(mnttab, &entry) == 0) { 1553 /* 1554 * Ignore non-ZFS entries. 1555 */ 1556 if (entry.mnt_fstype == NULL || 1557 strcmp(entry.mnt_fstype, MNTTYPE_ZFS) != 0) 1558 continue; 1559 1560 /* 1561 * Ignore filesystems not within this pool. 1562 */ 1563 if (entry.mnt_mountp == NULL || 1564 strncmp(entry.mnt_special, zhp->zpool_name, namelen) != 0 || 1565 (entry.mnt_special[namelen] != '/' && 1566 entry.mnt_special[namelen] != '\0')) 1567 continue; 1568 1569 /* 1570 * At this point we've found a filesystem within our pool. Add 1571 * it to our growing list. 1572 */ 1573 if (used == alloc) { 1574 if (alloc == 0) { 1575 1576 if ((sets = zfs_alloc(hdl, 1577 8 * sizeof (struct sets_s))) == NULL) 1578 goto out; 1579 1580 alloc = 8; 1581 } else { 1582 void *ptr; 1583 1584 if ((ptr = zfs_realloc(hdl, sets, 1585 alloc * sizeof (struct sets_s), 1586 alloc * 2 * sizeof (struct sets_s))) 1587 == NULL) 1588 goto out; 1589 sets = ptr; 1590 1591 alloc *= 2; 1592 } 1593 } 1594 1595 if ((sets[used].mountpoint = zfs_strdup(hdl, 1596 entry.mnt_mountp)) == NULL) 1597 goto out; 1598 1599 /* 1600 * This is allowed to fail, in case there is some I/O error. It 1601 * is only used to determine if we need to remove the underlying 1602 * mountpoint, so failure is not fatal. 1603 */ 1604 sets[used].dataset = make_dataset_handle(hdl, 1605 entry.mnt_special); 1606 1607 used++; 1608 } 1609 1610 /* 1611 * At this point, we have the entire list of filesystems, so sort it by 1612 * mountpoint. 1613 */ 1614 if (used != 0) 1615 qsort(sets, used, sizeof (struct sets_s), mountpoint_compare); 1616 1617 /* 1618 * Walk through and first unshare everything. 1619 */ 1620 for (i = 0; i < used; i++) { 1621 const zfs_share_proto_t *curr_proto; 1622 for (curr_proto = share_all_proto; *curr_proto != PROTO_END; 1623 curr_proto++) { 1624 if (is_shared(sets[i].mountpoint, *curr_proto) && 1625 unshare_one(hdl, sets[i].mountpoint, 1626 sets[i].mountpoint, *curr_proto) != 0) 1627 goto out; 1628 } 1629 } 1630 zfs_commit_all_shares(); 1631 1632 /* 1633 * Now unmount everything, removing the underlying directories as 1634 * appropriate. 1635 */ 1636 for (i = 0; i < used; i++) { 1637 if (unmount_one(sets[i].dataset, sets[i].mountpoint, 1638 flags) != 0) 1639 goto out; 1640 } 1641 1642 for (i = 0; i < used; i++) { 1643 if (sets[i].dataset) 1644 remove_mountpoint(sets[i].dataset); 1645 } 1646 1647 zpool_disable_datasets_os(zhp, force); 1648 1649 ret = 0; 1650 out: 1651 (void) fclose(mnttab); 1652 for (i = 0; i < used; i++) { 1653 if (sets[i].dataset) 1654 zfs_close(sets[i].dataset); 1655 free(sets[i].mountpoint); 1656 } 1657 free(sets); 1658 1659 return (ret); 1660 } 1661