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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2013, 2018 by Delphix. All rights reserved. 25 * Copyright (c) 2016, 2017 Intel Corporation. 26 * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>. 27 */ 28 29 /* 30 * Functions to convert between a list of vdevs and an nvlist representing the 31 * configuration. Each entry in the list can be one of: 32 * 33 * Device vdevs 34 * disk=(path=..., devid=...) 35 * file=(path=...) 36 * 37 * Group vdevs 38 * raidz[1|2]=(...) 39 * mirror=(...) 40 * 41 * Hot spares 42 * 43 * While the underlying implementation supports it, group vdevs cannot contain 44 * other group vdevs. All userland verification of devices is contained within 45 * this file. If successful, the nvlist returned can be passed directly to the 46 * kernel; we've done as much verification as possible in userland. 47 * 48 * Hot spares are a special case, and passed down as an array of disk vdevs, at 49 * the same level as the root of the vdev tree. 50 * 51 * The only function exported by this file is 'make_root_vdev'. The 52 * function performs several passes: 53 * 54 * 1. Construct the vdev specification. Performs syntax validation and 55 * makes sure each device is valid. 56 * 2. Check for devices in use. Using libdiskmgt, makes sure that no 57 * devices are also in use. Some can be overridden using the 'force' 58 * flag, others cannot. 59 * 3. Check for replication errors if the 'force' flag is not specified. 60 * validates that the replication level is consistent across the 61 * entire pool. 62 * 4. Call libzfs to label any whole disks with an EFI label. 63 */ 64 65 #include <assert.h> 66 #include <devid.h> 67 #include <errno.h> 68 #include <fcntl.h> 69 #include <libdiskmgt.h> 70 #include <libintl.h> 71 #include <libnvpair.h> 72 #include <limits.h> 73 #include <sys/spa.h> 74 #include <stdio.h> 75 #include <string.h> 76 #include <unistd.h> 77 #include <sys/efi_partition.h> 78 #include <sys/stat.h> 79 #include <sys/vtoc.h> 80 #include <sys/mntent.h> 81 82 #include "zpool_util.h" 83 84 #define BACKUP_SLICE "s2" 85 86 /* 87 * For any given vdev specification, we can have multiple errors. The 88 * vdev_error() function keeps track of whether we have seen an error yet, and 89 * prints out a header if its the first error we've seen. 90 */ 91 boolean_t error_seen; 92 boolean_t is_force; 93 94 /*PRINTFLIKE1*/ 95 static void 96 vdev_error(const char *fmt, ...) 97 { 98 va_list ap; 99 100 if (!error_seen) { 101 (void) fprintf(stderr, gettext("invalid vdev specification\n")); 102 if (!is_force) 103 (void) fprintf(stderr, gettext("use '-f' to override " 104 "the following errors:\n")); 105 else 106 (void) fprintf(stderr, gettext("the following errors " 107 "must be manually repaired:\n")); 108 error_seen = B_TRUE; 109 } 110 111 va_start(ap, fmt); 112 (void) vfprintf(stderr, fmt, ap); 113 va_end(ap); 114 } 115 116 static void 117 libdiskmgt_error(int error) 118 { 119 /* 120 * ENXIO/ENODEV is a valid error message if the device doesn't live in 121 * /dev/dsk. Don't bother printing an error message in this case. 122 */ 123 if (error == ENXIO || error == ENODEV) 124 return; 125 126 (void) fprintf(stderr, gettext("warning: device in use checking " 127 "failed: %s\n"), strerror(error)); 128 } 129 130 /* 131 * Validate a device, passing the bulk of the work off to libdiskmgt. 132 */ 133 static int 134 check_slice(const char *path, int force, boolean_t wholedisk, boolean_t isspare) 135 { 136 char *msg; 137 int error = 0; 138 dm_who_type_t who; 139 140 if (force) 141 who = DM_WHO_ZPOOL_FORCE; 142 else if (isspare) 143 who = DM_WHO_ZPOOL_SPARE; 144 else 145 who = DM_WHO_ZPOOL; 146 147 if (dm_inuse((char *)path, &msg, who, &error) || error) { 148 if (error != 0) { 149 libdiskmgt_error(error); 150 return (0); 151 } else { 152 vdev_error("%s", msg); 153 free(msg); 154 return (-1); 155 } 156 } 157 158 /* 159 * If we're given a whole disk, ignore overlapping slices since we're 160 * about to label it anyway. 161 */ 162 error = 0; 163 if (!wholedisk && !force && 164 (dm_isoverlapping((char *)path, &msg, &error) || error)) { 165 if (error == 0) { 166 /* dm_isoverlapping returned -1 */ 167 vdev_error(gettext("%s overlaps with %s\n"), path, msg); 168 free(msg); 169 return (-1); 170 } else if (error != ENODEV) { 171 /* libdiskmgt's devcache only handles physical drives */ 172 libdiskmgt_error(error); 173 return (0); 174 } 175 } 176 177 return (0); 178 } 179 180 181 /* 182 * Validate a whole disk. Iterate over all slices on the disk and make sure 183 * that none is in use by calling check_slice(). 184 */ 185 static int 186 check_disk(const char *name, dm_descriptor_t disk, int force, int isspare) 187 { 188 dm_descriptor_t *drive, *media, *slice; 189 int err = 0; 190 int i; 191 int ret; 192 193 /* 194 * Get the drive associated with this disk. This should never fail, 195 * because we already have an alias handle open for the device. 196 */ 197 if ((drive = dm_get_associated_descriptors(disk, DM_DRIVE, 198 &err)) == NULL || *drive == 0) { 199 if (err) 200 libdiskmgt_error(err); 201 return (0); 202 } 203 204 if ((media = dm_get_associated_descriptors(*drive, DM_MEDIA, 205 &err)) == NULL) { 206 dm_free_descriptors(drive); 207 if (err) 208 libdiskmgt_error(err); 209 return (0); 210 } 211 212 dm_free_descriptors(drive); 213 214 /* 215 * It is possible that the user has specified a removable media drive, 216 * and the media is not present. 217 */ 218 if (*media == 0) { 219 dm_free_descriptors(media); 220 vdev_error(gettext("'%s' has no media in drive\n"), name); 221 return (-1); 222 } 223 224 if ((slice = dm_get_associated_descriptors(*media, DM_SLICE, 225 &err)) == NULL) { 226 dm_free_descriptors(media); 227 if (err) 228 libdiskmgt_error(err); 229 return (0); 230 } 231 232 dm_free_descriptors(media); 233 234 ret = 0; 235 236 /* 237 * Iterate over all slices and report any errors. We don't care about 238 * overlapping slices because we are using the whole disk. 239 */ 240 for (i = 0; slice[i] != 0; i++) { 241 char *name = dm_get_name(slice[i], &err); 242 243 if (check_slice(name, force, B_TRUE, isspare) != 0) 244 ret = -1; 245 246 dm_free_name(name); 247 } 248 249 dm_free_descriptors(slice); 250 return (ret); 251 } 252 253 /* 254 * Validate a device. 255 */ 256 static int 257 check_device(const char *path, boolean_t force, boolean_t isspare) 258 { 259 dm_descriptor_t desc; 260 int err; 261 char *dev; 262 263 /* 264 * For whole disks, libdiskmgt does not include the leading dev path. 265 */ 266 dev = strrchr(path, '/'); 267 assert(dev != NULL); 268 dev++; 269 if ((desc = dm_get_descriptor_by_name(DM_ALIAS, dev, &err)) != 0) { 270 err = check_disk(path, desc, force, isspare); 271 dm_free_descriptor(desc); 272 return (err); 273 } 274 275 return (check_slice(path, force, B_FALSE, isspare)); 276 } 277 278 /* 279 * Check that a file is valid. All we can do in this case is check that it's 280 * not in use by another pool, and not in use by swap. 281 */ 282 static int 283 check_file(const char *file, boolean_t force, boolean_t isspare) 284 { 285 char *name; 286 int fd; 287 int ret = 0; 288 int err; 289 pool_state_t state; 290 boolean_t inuse; 291 292 if (dm_inuse_swap(file, &err)) { 293 if (err) 294 libdiskmgt_error(err); 295 else 296 vdev_error(gettext("%s is currently used by swap. " 297 "Please see swap(1M).\n"), file); 298 return (-1); 299 } 300 301 if ((fd = open(file, O_RDONLY)) < 0) 302 return (0); 303 304 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) { 305 const char *desc; 306 307 switch (state) { 308 case POOL_STATE_ACTIVE: 309 desc = gettext("active"); 310 break; 311 312 case POOL_STATE_EXPORTED: 313 desc = gettext("exported"); 314 break; 315 316 case POOL_STATE_POTENTIALLY_ACTIVE: 317 desc = gettext("potentially active"); 318 break; 319 320 default: 321 desc = gettext("unknown"); 322 break; 323 } 324 325 /* 326 * Allow hot spares to be shared between pools. 327 */ 328 if (state == POOL_STATE_SPARE && isspare) 329 return (0); 330 331 if (state == POOL_STATE_ACTIVE || 332 state == POOL_STATE_SPARE || !force) { 333 switch (state) { 334 case POOL_STATE_SPARE: 335 vdev_error(gettext("%s is reserved as a hot " 336 "spare for pool %s\n"), file, name); 337 break; 338 default: 339 vdev_error(gettext("%s is part of %s pool " 340 "'%s'\n"), file, desc, name); 341 break; 342 } 343 ret = -1; 344 } 345 346 free(name); 347 } 348 349 (void) close(fd); 350 return (ret); 351 } 352 353 354 /* 355 * By "whole disk" we mean an entire physical disk (something we can 356 * label, toggle the write cache on, etc.) as opposed to the full 357 * capacity of a pseudo-device such as lofi or did. We act as if we 358 * are labeling the disk, which should be a pretty good test of whether 359 * it's a viable device or not. Returns B_TRUE if it is and B_FALSE if 360 * it isn't. 361 */ 362 static boolean_t 363 is_whole_disk(const char *arg) 364 { 365 struct dk_gpt *label; 366 int fd; 367 char path[MAXPATHLEN]; 368 369 (void) snprintf(path, sizeof (path), "%s%s%s", 370 ZFS_RDISK_ROOT, strrchr(arg, '/'), BACKUP_SLICE); 371 if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) 372 return (B_FALSE); 373 if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) { 374 (void) close(fd); 375 return (B_FALSE); 376 } 377 efi_free(label); 378 (void) close(fd); 379 return (B_TRUE); 380 } 381 382 /* 383 * Create a leaf vdev. Determine if this is a file or a device. If it's a 384 * device, fill in the device id to make a complete nvlist. Valid forms for a 385 * leaf vdev are: 386 * 387 * /dev/dsk/xxx Complete disk path 388 * /xxx Full path to file 389 * xxx Shorthand for /dev/dsk/xxx 390 */ 391 static nvlist_t * 392 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log) 393 { 394 char path[MAXPATHLEN]; 395 struct stat64 statbuf; 396 nvlist_t *vdev = NULL; 397 char *type = NULL; 398 boolean_t wholedisk = B_FALSE; 399 uint64_t ashift = 0; 400 401 /* 402 * Determine what type of vdev this is, and put the full path into 403 * 'path'. We detect whether this is a device of file afterwards by 404 * checking the st_mode of the file. 405 */ 406 if (arg[0] == '/') { 407 /* 408 * Complete device or file path. Exact type is determined by 409 * examining the file descriptor afterwards. 410 */ 411 wholedisk = is_whole_disk(arg); 412 if (!wholedisk && (stat64(arg, &statbuf) != 0)) { 413 (void) fprintf(stderr, 414 gettext("cannot open '%s': %s\n"), 415 arg, strerror(errno)); 416 return (NULL); 417 } 418 419 (void) strlcpy(path, arg, sizeof (path)); 420 } else { 421 /* 422 * This may be a short path for a device, or it could be total 423 * gibberish. Check to see if it's a known device in 424 * /dev/dsk/. As part of this check, see if we've been given a 425 * an entire disk (minus the slice number). 426 */ 427 (void) snprintf(path, sizeof (path), "%s/%s", ZFS_DISK_ROOT, 428 arg); 429 wholedisk = is_whole_disk(path); 430 if (!wholedisk && (stat64(path, &statbuf) != 0)) { 431 /* 432 * If we got ENOENT, then the user gave us 433 * gibberish, so try to direct them with a 434 * reasonable error message. Otherwise, 435 * regurgitate strerror() since it's the best we 436 * can do. 437 */ 438 if (errno == ENOENT) { 439 (void) fprintf(stderr, 440 gettext("cannot open '%s': no such " 441 "device in %s\n"), arg, ZFS_DISK_ROOT); 442 (void) fprintf(stderr, 443 gettext("must be a full path or " 444 "shorthand device name\n")); 445 return (NULL); 446 } else { 447 (void) fprintf(stderr, 448 gettext("cannot open '%s': %s\n"), 449 path, strerror(errno)); 450 return (NULL); 451 } 452 } 453 } 454 455 /* 456 * Determine whether this is a device or a file. 457 */ 458 if (wholedisk || S_ISBLK(statbuf.st_mode)) { 459 type = VDEV_TYPE_DISK; 460 } else if (S_ISREG(statbuf.st_mode)) { 461 type = VDEV_TYPE_FILE; 462 } else { 463 (void) fprintf(stderr, gettext("cannot use '%s': must be a " 464 "block device or regular file\n"), path); 465 return (NULL); 466 } 467 468 /* 469 * Finally, we have the complete device or file, and we know that it is 470 * acceptable to use. Construct the nvlist to describe this vdev. All 471 * vdevs have a 'path' element, and devices also have a 'devid' element. 472 */ 473 verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0); 474 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0); 475 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0); 476 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0); 477 if (is_log) 478 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_ALLOCATION_BIAS, 479 VDEV_ALLOC_BIAS_LOG) == 0); 480 if (strcmp(type, VDEV_TYPE_DISK) == 0) 481 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK, 482 (uint64_t)wholedisk) == 0); 483 484 if (props != NULL) { 485 char *value = NULL; 486 487 if (nvlist_lookup_string(props, 488 zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0) { 489 if (zfs_nicestrtonum(NULL, value, &ashift) != 0) { 490 (void) fprintf(stderr, 491 gettext("ashift must be a number.\n")); 492 return (NULL); 493 } 494 if (ashift != 0 && 495 (ashift < ASHIFT_MIN || ashift > ASHIFT_MAX)) { 496 (void) fprintf(stderr, 497 gettext("invalid 'ashift=%" PRIu64 "' " 498 "property: only values between %" PRId32 " " 499 "and %" PRId32 " are allowed.\n"), 500 ashift, ASHIFT_MIN, ASHIFT_MAX); 501 return (NULL); 502 } 503 } 504 } 505 506 /* 507 * For a whole disk, defer getting its devid until after labeling it. 508 */ 509 if (S_ISBLK(statbuf.st_mode) && !wholedisk) { 510 /* 511 * Get the devid for the device. 512 */ 513 int fd; 514 ddi_devid_t devid; 515 char *minor = NULL, *devid_str = NULL; 516 517 if ((fd = open(path, O_RDONLY)) < 0) { 518 (void) fprintf(stderr, gettext("cannot open '%s': " 519 "%s\n"), path, strerror(errno)); 520 nvlist_free(vdev); 521 return (NULL); 522 } 523 524 if (devid_get(fd, &devid) == 0) { 525 if (devid_get_minor_name(fd, &minor) == 0 && 526 (devid_str = devid_str_encode(devid, minor)) != 527 NULL) { 528 verify(nvlist_add_string(vdev, 529 ZPOOL_CONFIG_DEVID, devid_str) == 0); 530 } 531 if (devid_str != NULL) 532 devid_str_free(devid_str); 533 if (minor != NULL) 534 devid_str_free(minor); 535 devid_free(devid); 536 } 537 538 (void) close(fd); 539 } 540 541 if (ashift > 0) 542 (void) nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift); 543 544 return (vdev); 545 } 546 547 /* 548 * Go through and verify the replication level of the pool is consistent. 549 * Performs the following checks: 550 * 551 * For the new spec, verifies that devices in mirrors and raidz are the 552 * same size. 553 * 554 * If the current configuration already has inconsistent replication 555 * levels, ignore any other potential problems in the new spec. 556 * 557 * Otherwise, make sure that the current spec (if there is one) and the new 558 * spec have consistent replication levels. 559 * 560 * If there is no current spec (create), make sure new spec has at least 561 * one general purpose vdev. 562 */ 563 typedef struct replication_level { 564 char *zprl_type; 565 uint64_t zprl_children; 566 uint64_t zprl_parity; 567 } replication_level_t; 568 569 #define ZPOOL_FUZZ (16 * 1024 * 1024) 570 571 static boolean_t 572 is_raidz_mirror(replication_level_t *a, replication_level_t *b, 573 replication_level_t **raidz, replication_level_t **mirror) 574 { 575 if (strcmp(a->zprl_type, "raidz") == 0 && 576 strcmp(b->zprl_type, "mirror") == 0) { 577 *raidz = a; 578 *mirror = b; 579 return (B_TRUE); 580 } 581 return (B_FALSE); 582 } 583 584 /* 585 * Given a list of toplevel vdevs, return the current replication level. If 586 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then 587 * an error message will be displayed for each self-inconsistent vdev. 588 */ 589 static replication_level_t * 590 get_replication(nvlist_t *nvroot, boolean_t fatal) 591 { 592 nvlist_t **top; 593 uint_t t, toplevels; 594 nvlist_t **child; 595 uint_t c, children; 596 nvlist_t *nv; 597 char *type; 598 replication_level_t lastrep = {0}; 599 replication_level_t rep; 600 replication_level_t *ret; 601 replication_level_t *raidz, *mirror; 602 boolean_t dontreport; 603 604 ret = safe_malloc(sizeof (replication_level_t)); 605 606 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 607 &top, &toplevels) == 0); 608 609 for (t = 0; t < toplevels; t++) { 610 uint64_t is_log = B_FALSE; 611 612 nv = top[t]; 613 614 /* 615 * For separate logs we ignore the top level vdev replication 616 * constraints. 617 */ 618 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log); 619 if (is_log) 620 continue; 621 622 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, 623 &type) == 0); 624 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 625 &child, &children) != 0) { 626 /* 627 * This is a 'file' or 'disk' vdev. 628 */ 629 rep.zprl_type = type; 630 rep.zprl_children = 1; 631 rep.zprl_parity = 0; 632 } else { 633 uint64_t vdev_size; 634 635 /* 636 * This is a mirror or RAID-Z vdev. Go through and make 637 * sure the contents are all the same (files vs. disks), 638 * keeping track of the number of elements in the 639 * process. 640 * 641 * We also check that the size of each vdev (if it can 642 * be determined) is the same. 643 */ 644 rep.zprl_type = type; 645 rep.zprl_children = 0; 646 647 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) { 648 verify(nvlist_lookup_uint64(nv, 649 ZPOOL_CONFIG_NPARITY, 650 &rep.zprl_parity) == 0); 651 assert(rep.zprl_parity != 0); 652 } else { 653 rep.zprl_parity = 0; 654 } 655 656 /* 657 * The 'dontreport' variable indicates that we've 658 * already reported an error for this spec, so don't 659 * bother doing it again. 660 */ 661 type = NULL; 662 dontreport = 0; 663 vdev_size = -1ULL; 664 for (c = 0; c < children; c++) { 665 nvlist_t *cnv = child[c]; 666 char *path; 667 struct stat64 statbuf; 668 uint64_t size = -1ULL; 669 char *childtype; 670 int fd, err; 671 672 rep.zprl_children++; 673 674 verify(nvlist_lookup_string(cnv, 675 ZPOOL_CONFIG_TYPE, &childtype) == 0); 676 677 /* 678 * If this is a replacing or spare vdev, then 679 * get the real first child of the vdev: do this 680 * in a loop because replacing and spare vdevs 681 * can be nested. 682 */ 683 while (strcmp(childtype, 684 VDEV_TYPE_REPLACING) == 0 || 685 strcmp(childtype, VDEV_TYPE_SPARE) == 0) { 686 nvlist_t **rchild; 687 uint_t rchildren; 688 689 verify(nvlist_lookup_nvlist_array(cnv, 690 ZPOOL_CONFIG_CHILDREN, &rchild, 691 &rchildren) == 0); 692 assert(rchildren == 2); 693 cnv = rchild[0]; 694 695 verify(nvlist_lookup_string(cnv, 696 ZPOOL_CONFIG_TYPE, 697 &childtype) == 0); 698 } 699 700 verify(nvlist_lookup_string(cnv, 701 ZPOOL_CONFIG_PATH, &path) == 0); 702 703 /* 704 * If we have a raidz/mirror that combines disks 705 * with files, report it as an error. 706 */ 707 if (!dontreport && type != NULL && 708 strcmp(type, childtype) != 0) { 709 if (ret != NULL) 710 free(ret); 711 ret = NULL; 712 if (fatal) 713 vdev_error(gettext( 714 "mismatched replication " 715 "level: %s contains both " 716 "files and devices\n"), 717 rep.zprl_type); 718 else 719 return (NULL); 720 dontreport = B_TRUE; 721 } 722 723 /* 724 * According to stat(2), the value of 'st_size' 725 * is undefined for block devices and character 726 * devices. But there is no effective way to 727 * determine the real size in userland. 728 * 729 * Instead, we'll take advantage of an 730 * implementation detail of spec_size(). If the 731 * device is currently open, then we (should) 732 * return a valid size. 733 * 734 * If we still don't get a valid size (indicated 735 * by a size of 0 or MAXOFFSET_T), then ignore 736 * this device altogether. 737 */ 738 if ((fd = open(path, O_RDONLY)) >= 0) { 739 err = fstat64(fd, &statbuf); 740 (void) close(fd); 741 } else { 742 err = stat64(path, &statbuf); 743 } 744 745 if (err != 0 || 746 statbuf.st_size == 0 || 747 statbuf.st_size == MAXOFFSET_T) 748 continue; 749 750 size = statbuf.st_size; 751 752 /* 753 * Also make sure that devices and 754 * slices have a consistent size. If 755 * they differ by a significant amount 756 * (~16MB) then report an error. 757 */ 758 if (!dontreport && 759 (vdev_size != -1ULL && 760 (labs(size - vdev_size) > 761 ZPOOL_FUZZ))) { 762 if (ret != NULL) 763 free(ret); 764 ret = NULL; 765 if (fatal) 766 vdev_error(gettext( 767 "%s contains devices of " 768 "different sizes\n"), 769 rep.zprl_type); 770 else 771 return (NULL); 772 dontreport = B_TRUE; 773 } 774 775 type = childtype; 776 vdev_size = size; 777 } 778 } 779 780 /* 781 * At this point, we have the replication of the last toplevel 782 * vdev in 'rep'. Compare it to 'lastrep' to see if it is 783 * different. 784 */ 785 if (lastrep.zprl_type != NULL) { 786 if (is_raidz_mirror(&lastrep, &rep, &raidz, &mirror) || 787 is_raidz_mirror(&rep, &lastrep, &raidz, &mirror)) { 788 /* 789 * Accepted raidz and mirror when they can 790 * handle the same number of disk failures. 791 */ 792 if (raidz->zprl_parity != 793 mirror->zprl_children - 1) { 794 if (ret != NULL) 795 free(ret); 796 ret = NULL; 797 if (fatal) 798 vdev_error(gettext( 799 "mismatched replication " 800 "level: " 801 "%s and %s vdevs with " 802 "different redundancy, " 803 "%llu vs. %llu (%llu-way) " 804 "are present\n"), 805 raidz->zprl_type, 806 mirror->zprl_type, 807 raidz->zprl_parity, 808 mirror->zprl_children - 1, 809 mirror->zprl_children); 810 else 811 return (NULL); 812 } 813 } else if (strcmp(lastrep.zprl_type, rep.zprl_type) != 814 0) { 815 if (ret != NULL) 816 free(ret); 817 ret = NULL; 818 if (fatal) 819 vdev_error(gettext( 820 "mismatched replication level: " 821 "both %s and %s vdevs are " 822 "present\n"), 823 lastrep.zprl_type, rep.zprl_type); 824 else 825 return (NULL); 826 } else if (lastrep.zprl_parity != rep.zprl_parity) { 827 if (ret) 828 free(ret); 829 ret = NULL; 830 if (fatal) 831 vdev_error(gettext( 832 "mismatched replication level: " 833 "both %llu and %llu device parity " 834 "%s vdevs are present\n"), 835 lastrep.zprl_parity, 836 rep.zprl_parity, 837 rep.zprl_type); 838 else 839 return (NULL); 840 } else if (lastrep.zprl_children != rep.zprl_children) { 841 if (ret) 842 free(ret); 843 ret = NULL; 844 if (fatal) 845 vdev_error(gettext( 846 "mismatched replication level: " 847 "both %llu-way and %llu-way %s " 848 "vdevs are present\n"), 849 lastrep.zprl_children, 850 rep.zprl_children, 851 rep.zprl_type); 852 else 853 return (NULL); 854 } 855 } 856 lastrep = rep; 857 } 858 859 if (ret != NULL) 860 *ret = rep; 861 862 return (ret); 863 } 864 865 /* 866 * Check the replication level of the vdev spec against the current pool. Calls 867 * get_replication() to make sure the new spec is self-consistent. If the pool 868 * has a consistent replication level, then we ignore any errors. Otherwise, 869 * report any difference between the two. 870 */ 871 static int 872 check_replication(nvlist_t *config, nvlist_t *newroot) 873 { 874 nvlist_t **child; 875 uint_t children; 876 replication_level_t *current = NULL, *new; 877 replication_level_t *raidz, *mirror; 878 int ret; 879 880 /* 881 * If we have a current pool configuration, check to see if it's 882 * self-consistent. If not, simply return success. 883 */ 884 if (config != NULL) { 885 nvlist_t *nvroot; 886 887 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 888 &nvroot) == 0); 889 if ((current = get_replication(nvroot, B_FALSE)) == NULL) 890 return (0); 891 } 892 /* 893 * for spares there may be no children, and therefore no 894 * replication level to check 895 */ 896 if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN, 897 &child, &children) != 0) || (children == 0)) { 898 free(current); 899 return (0); 900 } 901 902 /* 903 * If all we have is logs then there's no replication level to check. 904 */ 905 if (num_logs(newroot) == children) { 906 free(current); 907 return (0); 908 } 909 910 /* 911 * Get the replication level of the new vdev spec, reporting any 912 * inconsistencies found. 913 */ 914 if ((new = get_replication(newroot, B_TRUE)) == NULL) { 915 free(current); 916 return (-1); 917 } 918 919 /* 920 * Check to see if the new vdev spec matches the replication level of 921 * the current pool. 922 */ 923 ret = 0; 924 if (current != NULL) { 925 if (is_raidz_mirror(current, new, &raidz, &mirror) || 926 is_raidz_mirror(new, current, &raidz, &mirror)) { 927 if (raidz->zprl_parity != mirror->zprl_children - 1) { 928 vdev_error(gettext( 929 "mismatched replication level: pool and " 930 "new vdev with different redundancy, %s " 931 "and %s vdevs, %llu vs. %llu (%llu-way)\n"), 932 raidz->zprl_type, 933 mirror->zprl_type, 934 raidz->zprl_parity, 935 mirror->zprl_children - 1, 936 mirror->zprl_children); 937 ret = -1; 938 } 939 } else if (strcmp(current->zprl_type, new->zprl_type) != 0) { 940 vdev_error(gettext( 941 "mismatched replication level: pool uses %s " 942 "and new vdev is %s\n"), 943 current->zprl_type, new->zprl_type); 944 ret = -1; 945 } else if (current->zprl_parity != new->zprl_parity) { 946 vdev_error(gettext( 947 "mismatched replication level: pool uses %llu " 948 "device parity and new vdev uses %llu\n"), 949 current->zprl_parity, new->zprl_parity); 950 ret = -1; 951 } else if (current->zprl_children != new->zprl_children) { 952 vdev_error(gettext( 953 "mismatched replication level: pool uses %llu-way " 954 "%s and new vdev uses %llu-way %s\n"), 955 current->zprl_children, current->zprl_type, 956 new->zprl_children, new->zprl_type); 957 ret = -1; 958 } 959 } 960 961 free(new); 962 if (current != NULL) 963 free(current); 964 965 return (ret); 966 } 967 968 /* 969 * Go through and find any whole disks in the vdev specification, labelling them 970 * as appropriate. When constructing the vdev spec, we were unable to open this 971 * device in order to provide a devid. Now that we have labelled the disk and 972 * know the pool slice is valid, we can construct the devid now. 973 * 974 * If the disk was already labeled with an EFI label, we will have gotten the 975 * devid already (because we were able to open the whole disk). Otherwise, we 976 * need to get the devid after we label the disk. 977 */ 978 static int 979 make_disks(zpool_handle_t *zhp, nvlist_t *nv, zpool_boot_label_t boot_type, 980 uint64_t boot_size) 981 { 982 nvlist_t **child; 983 uint_t c, children; 984 char *type, *path, *diskname; 985 char buf[MAXPATHLEN]; 986 uint64_t wholedisk; 987 int fd; 988 int ret; 989 int slice; 990 ddi_devid_t devid; 991 char *minor = NULL, *devid_str = NULL; 992 993 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0); 994 995 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 996 &child, &children) != 0) { 997 998 if (strcmp(type, VDEV_TYPE_DISK) != 0) 999 return (0); 1000 1001 /* 1002 * We have a disk device. Get the path to the device 1003 * and see if it's a whole disk by appending the backup 1004 * slice and stat()ing the device. 1005 */ 1006 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0); 1007 1008 diskname = strrchr(path, '/'); 1009 assert(diskname != NULL); 1010 diskname++; 1011 1012 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 1013 &wholedisk) != 0 || !wholedisk) { 1014 /* 1015 * This is not whole disk, return error if 1016 * boot partition creation was requested 1017 */ 1018 if (boot_type == ZPOOL_CREATE_BOOT_LABEL) { 1019 (void) fprintf(stderr, 1020 gettext("creating boot partition is only " 1021 "supported on whole disk vdevs: %s\n"), 1022 diskname); 1023 return (-1); 1024 } 1025 return (0); 1026 } 1027 1028 ret = zpool_label_disk(g_zfs, zhp, diskname, boot_type, 1029 boot_size, &slice); 1030 if (ret == -1) 1031 return (ret); 1032 1033 /* 1034 * Fill in the devid, now that we've labeled the disk. 1035 */ 1036 (void) snprintf(buf, sizeof (buf), "%ss%d", path, slice); 1037 if ((fd = open(buf, O_RDONLY)) < 0) { 1038 (void) fprintf(stderr, 1039 gettext("cannot open '%s': %s\n"), 1040 buf, strerror(errno)); 1041 return (-1); 1042 } 1043 1044 if (devid_get(fd, &devid) == 0) { 1045 if (devid_get_minor_name(fd, &minor) == 0 && 1046 (devid_str = devid_str_encode(devid, minor)) != 1047 NULL) { 1048 verify(nvlist_add_string(nv, 1049 ZPOOL_CONFIG_DEVID, devid_str) == 0); 1050 } 1051 if (devid_str != NULL) 1052 devid_str_free(devid_str); 1053 if (minor != NULL) 1054 devid_str_free(minor); 1055 devid_free(devid); 1056 } 1057 1058 /* 1059 * Update the path to refer to the pool slice. The presence of 1060 * the 'whole_disk' field indicates to the CLI that we should 1061 * chop off the slice number when displaying the device in 1062 * future output. 1063 */ 1064 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0); 1065 1066 (void) close(fd); 1067 1068 return (0); 1069 } 1070 1071 /* illumos kernel does not support booting from multi-vdev pools. */ 1072 if ((boot_type == ZPOOL_CREATE_BOOT_LABEL)) { 1073 if ((strcmp(type, VDEV_TYPE_ROOT) == 0) && children > 1) { 1074 (void) fprintf(stderr, gettext("boot pool " 1075 "can not have more than one vdev\n")); 1076 return (-1); 1077 } 1078 } 1079 1080 for (c = 0; c < children; c++) { 1081 ret = make_disks(zhp, child[c], boot_type, boot_size); 1082 if (ret != 0) 1083 return (ret); 1084 } 1085 1086 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES, 1087 &child, &children) == 0) 1088 for (c = 0; c < children; c++) { 1089 ret = make_disks(zhp, child[c], boot_type, boot_size); 1090 if (ret != 0) 1091 return (ret); 1092 } 1093 1094 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE, 1095 &child, &children) == 0) 1096 for (c = 0; c < children; c++) { 1097 ret = make_disks(zhp, child[c], boot_type, boot_size); 1098 if (ret != 0) 1099 return (ret); 1100 } 1101 1102 return (0); 1103 } 1104 1105 /* 1106 * Determine if the given path is a hot spare within the given configuration. 1107 */ 1108 static boolean_t 1109 is_spare(nvlist_t *config, const char *path) 1110 { 1111 int fd; 1112 pool_state_t state; 1113 char *name = NULL; 1114 nvlist_t *label; 1115 uint64_t guid, spareguid; 1116 nvlist_t *nvroot; 1117 nvlist_t **spares; 1118 uint_t i, nspares; 1119 boolean_t inuse; 1120 1121 if ((fd = open(path, O_RDONLY)) < 0) 1122 return (B_FALSE); 1123 1124 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 || 1125 !inuse || 1126 state != POOL_STATE_SPARE || 1127 zpool_read_label(fd, &label) != 0) { 1128 free(name); 1129 (void) close(fd); 1130 return (B_FALSE); 1131 } 1132 free(name); 1133 (void) close(fd); 1134 1135 verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0); 1136 nvlist_free(label); 1137 1138 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 1139 &nvroot) == 0); 1140 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 1141 &spares, &nspares) == 0) { 1142 for (i = 0; i < nspares; i++) { 1143 verify(nvlist_lookup_uint64(spares[i], 1144 ZPOOL_CONFIG_GUID, &spareguid) == 0); 1145 if (spareguid == guid) 1146 return (B_TRUE); 1147 } 1148 } 1149 1150 return (B_FALSE); 1151 } 1152 1153 /* 1154 * Go through and find any devices that are in use. We rely on libdiskmgt for 1155 * the majority of this task. 1156 */ 1157 static boolean_t 1158 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force, 1159 boolean_t replacing, boolean_t isspare) 1160 { 1161 nvlist_t **child; 1162 uint_t c, children; 1163 char *type, *path; 1164 int ret = 0; 1165 char buf[MAXPATHLEN]; 1166 uint64_t wholedisk; 1167 boolean_t anyinuse = B_FALSE; 1168 1169 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0); 1170 1171 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1172 &child, &children) != 0) { 1173 1174 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0); 1175 1176 /* 1177 * As a generic check, we look to see if this is a replace of a 1178 * hot spare within the same pool. If so, we allow it 1179 * regardless of what libdiskmgt or zpool_in_use() says. 1180 */ 1181 if (replacing) { 1182 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 1183 &wholedisk) == 0 && wholedisk) 1184 (void) snprintf(buf, sizeof (buf), "%ss0", 1185 path); 1186 else 1187 (void) strlcpy(buf, path, sizeof (buf)); 1188 1189 if (is_spare(config, buf)) 1190 return (B_FALSE); 1191 } 1192 1193 if (strcmp(type, VDEV_TYPE_DISK) == 0) 1194 ret = check_device(path, force, isspare); 1195 else if (strcmp(type, VDEV_TYPE_FILE) == 0) 1196 ret = check_file(path, force, isspare); 1197 1198 return (ret != 0); 1199 } 1200 1201 for (c = 0; c < children; c++) 1202 if (is_device_in_use(config, child[c], force, replacing, 1203 B_FALSE)) 1204 anyinuse = B_TRUE; 1205 1206 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES, 1207 &child, &children) == 0) 1208 for (c = 0; c < children; c++) 1209 if (is_device_in_use(config, child[c], force, replacing, 1210 B_TRUE)) 1211 anyinuse = B_TRUE; 1212 1213 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE, 1214 &child, &children) == 0) 1215 for (c = 0; c < children; c++) 1216 if (is_device_in_use(config, child[c], force, replacing, 1217 B_FALSE)) 1218 anyinuse = B_TRUE; 1219 1220 return (anyinuse); 1221 } 1222 1223 static const char * 1224 is_grouping(const char *type, int *mindev, int *maxdev) 1225 { 1226 if (strncmp(type, "raidz", 5) == 0) { 1227 const char *p = type + 5; 1228 char *end; 1229 long nparity; 1230 1231 if (*p == '\0') { 1232 nparity = 1; 1233 } else if (*p == '0') { 1234 return (NULL); /* no zero prefixes allowed */ 1235 } else { 1236 errno = 0; 1237 nparity = strtol(p, &end, 10); 1238 if (errno != 0 || nparity < 1 || nparity >= 255 || 1239 *end != '\0') 1240 return (NULL); 1241 } 1242 1243 if (mindev != NULL) 1244 *mindev = nparity + 1; 1245 if (maxdev != NULL) 1246 *maxdev = 255; 1247 return (VDEV_TYPE_RAIDZ); 1248 } 1249 1250 if (maxdev != NULL) 1251 *maxdev = INT_MAX; 1252 1253 if (strcmp(type, "mirror") == 0) { 1254 if (mindev != NULL) 1255 *mindev = 2; 1256 return (VDEV_TYPE_MIRROR); 1257 } 1258 1259 if (strcmp(type, "spare") == 0) { 1260 if (mindev != NULL) 1261 *mindev = 1; 1262 return (VDEV_TYPE_SPARE); 1263 } 1264 1265 if (strcmp(type, "log") == 0) { 1266 if (mindev != NULL) 1267 *mindev = 1; 1268 return (VDEV_TYPE_LOG); 1269 } 1270 1271 if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0 || 1272 strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) { 1273 if (mindev != NULL) 1274 *mindev = 1; 1275 return (type); 1276 } 1277 1278 if (strcmp(type, "cache") == 0) { 1279 if (mindev != NULL) 1280 *mindev = 1; 1281 return (VDEV_TYPE_L2CACHE); 1282 } 1283 1284 return (NULL); 1285 } 1286 1287 /* 1288 * Construct a syntactically valid vdev specification, 1289 * and ensure that all devices and files exist and can be opened. 1290 * Note: we don't bother freeing anything in the error paths 1291 * because the program is just going to exit anyway. 1292 */ 1293 nvlist_t * 1294 construct_spec(nvlist_t *props, int argc, char **argv) 1295 { 1296 nvlist_t *nvroot, *nv, **top, **spares, **l2cache; 1297 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache; 1298 const char *type; 1299 uint64_t is_log, is_special, is_dedup; 1300 boolean_t seen_logs; 1301 1302 top = NULL; 1303 toplevels = 0; 1304 spares = NULL; 1305 l2cache = NULL; 1306 nspares = 0; 1307 nlogs = 0; 1308 nl2cache = 0; 1309 is_log = is_special = is_dedup = B_FALSE; 1310 seen_logs = B_FALSE; 1311 1312 while (argc > 0) { 1313 nv = NULL; 1314 1315 /* 1316 * If it's a mirror or raidz, the subsequent arguments are 1317 * its leaves -- until we encounter the next mirror or raidz. 1318 */ 1319 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) { 1320 nvlist_t **child = NULL; 1321 int c, children = 0; 1322 1323 if (strcmp(type, VDEV_TYPE_SPARE) == 0) { 1324 if (spares != NULL) { 1325 (void) fprintf(stderr, 1326 gettext("invalid vdev " 1327 "specification: 'spare' can be " 1328 "specified only once\n")); 1329 return (NULL); 1330 } 1331 is_log = is_special = is_dedup = B_FALSE; 1332 } 1333 1334 if (strcmp(type, VDEV_TYPE_LOG) == 0) { 1335 if (seen_logs) { 1336 (void) fprintf(stderr, 1337 gettext("invalid vdev " 1338 "specification: 'log' can be " 1339 "specified only once\n")); 1340 return (NULL); 1341 } 1342 seen_logs = B_TRUE; 1343 is_log = B_TRUE; 1344 is_special = B_FALSE; 1345 is_dedup = B_FALSE; 1346 argc--; 1347 argv++; 1348 /* 1349 * A log is not a real grouping device. 1350 * We just set is_log and continue. 1351 */ 1352 continue; 1353 } 1354 1355 if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0) { 1356 is_special = B_TRUE; 1357 is_log = B_FALSE; 1358 is_dedup = B_FALSE; 1359 argc--; 1360 argv++; 1361 continue; 1362 } 1363 1364 if (strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) { 1365 is_dedup = B_TRUE; 1366 is_log = B_FALSE; 1367 is_special = B_FALSE; 1368 argc--; 1369 argv++; 1370 continue; 1371 } 1372 1373 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) { 1374 if (l2cache != NULL) { 1375 (void) fprintf(stderr, 1376 gettext("invalid vdev " 1377 "specification: 'cache' can be " 1378 "specified only once\n")); 1379 return (NULL); 1380 } 1381 is_log = is_special = is_dedup = B_FALSE; 1382 } 1383 1384 if (is_log || is_special || is_dedup) { 1385 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) { 1386 (void) fprintf(stderr, 1387 gettext("invalid vdev " 1388 "specification: unsupported '%s' " 1389 "device: %s\n"), is_log ? "log" : 1390 "special", type); 1391 return (NULL); 1392 } 1393 nlogs++; 1394 } 1395 1396 for (c = 1; c < argc; c++) { 1397 if (is_grouping(argv[c], NULL, NULL) != NULL) 1398 break; 1399 children++; 1400 child = realloc(child, 1401 children * sizeof (nvlist_t *)); 1402 if (child == NULL) 1403 zpool_no_memory(); 1404 if ((nv = make_leaf_vdev(props, argv[c], 1405 B_FALSE)) == NULL) 1406 return (NULL); 1407 child[children - 1] = nv; 1408 } 1409 1410 if (children < mindev) { 1411 (void) fprintf(stderr, gettext("invalid vdev " 1412 "specification: %s requires at least %d " 1413 "devices\n"), argv[0], mindev); 1414 return (NULL); 1415 } 1416 1417 if (children > maxdev) { 1418 (void) fprintf(stderr, gettext("invalid vdev " 1419 "specification: %s supports no more than " 1420 "%d devices\n"), argv[0], maxdev); 1421 return (NULL); 1422 } 1423 1424 argc -= c; 1425 argv += c; 1426 1427 if (strcmp(type, VDEV_TYPE_SPARE) == 0) { 1428 spares = child; 1429 nspares = children; 1430 continue; 1431 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) { 1432 l2cache = child; 1433 nl2cache = children; 1434 continue; 1435 } else { 1436 /* create a top-level vdev with children */ 1437 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME, 1438 0) == 0); 1439 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 1440 type) == 0); 1441 verify(nvlist_add_uint64(nv, 1442 ZPOOL_CONFIG_IS_LOG, is_log) == 0); 1443 if (is_log) 1444 verify(nvlist_add_string(nv, 1445 ZPOOL_CONFIG_ALLOCATION_BIAS, 1446 VDEV_ALLOC_BIAS_LOG) == 0); 1447 if (is_special) { 1448 verify(nvlist_add_string(nv, 1449 ZPOOL_CONFIG_ALLOCATION_BIAS, 1450 VDEV_ALLOC_BIAS_SPECIAL) == 0); 1451 } 1452 if (is_dedup) { 1453 verify(nvlist_add_string(nv, 1454 ZPOOL_CONFIG_ALLOCATION_BIAS, 1455 VDEV_ALLOC_BIAS_DEDUP) == 0); 1456 } 1457 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) { 1458 verify(nvlist_add_uint64(nv, 1459 ZPOOL_CONFIG_NPARITY, 1460 mindev - 1) == 0); 1461 } 1462 verify(nvlist_add_nvlist_array(nv, 1463 ZPOOL_CONFIG_CHILDREN, child, 1464 children) == 0); 1465 1466 for (c = 0; c < children; c++) 1467 nvlist_free(child[c]); 1468 free(child); 1469 } 1470 } else { 1471 /* 1472 * We have a device. Pass off to make_leaf_vdev() to 1473 * construct the appropriate nvlist describing the vdev. 1474 */ 1475 if ((nv = make_leaf_vdev(props, argv[0], is_log)) 1476 == NULL) 1477 return (NULL); 1478 if (is_log) 1479 nlogs++; 1480 if (is_special) { 1481 verify(nvlist_add_string(nv, 1482 ZPOOL_CONFIG_ALLOCATION_BIAS, 1483 VDEV_ALLOC_BIAS_SPECIAL) == 0); 1484 } 1485 if (is_dedup) { 1486 verify(nvlist_add_string(nv, 1487 ZPOOL_CONFIG_ALLOCATION_BIAS, 1488 VDEV_ALLOC_BIAS_DEDUP) == 0); 1489 } 1490 argc--; 1491 argv++; 1492 } 1493 1494 toplevels++; 1495 top = realloc(top, toplevels * sizeof (nvlist_t *)); 1496 if (top == NULL) 1497 zpool_no_memory(); 1498 top[toplevels - 1] = nv; 1499 } 1500 1501 if (toplevels == 0 && nspares == 0 && nl2cache == 0) { 1502 (void) fprintf(stderr, gettext("invalid vdev " 1503 "specification: at least one toplevel vdev must be " 1504 "specified\n")); 1505 return (NULL); 1506 } 1507 1508 if (seen_logs && nlogs == 0) { 1509 (void) fprintf(stderr, gettext("invalid vdev specification: " 1510 "log requires at least 1 device\n")); 1511 return (NULL); 1512 } 1513 1514 /* 1515 * Finally, create nvroot and add all top-level vdevs to it. 1516 */ 1517 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0); 1518 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 1519 VDEV_TYPE_ROOT) == 0); 1520 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 1521 top, toplevels) == 0); 1522 if (nspares != 0) 1523 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 1524 spares, nspares) == 0); 1525 if (nl2cache != 0) 1526 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 1527 l2cache, nl2cache) == 0); 1528 1529 for (t = 0; t < toplevels; t++) 1530 nvlist_free(top[t]); 1531 for (t = 0; t < nspares; t++) 1532 nvlist_free(spares[t]); 1533 for (t = 0; t < nl2cache; t++) 1534 nvlist_free(l2cache[t]); 1535 if (spares) 1536 free(spares); 1537 if (l2cache) 1538 free(l2cache); 1539 free(top); 1540 1541 return (nvroot); 1542 } 1543 1544 nvlist_t * 1545 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props, 1546 splitflags_t flags, int argc, char **argv) 1547 { 1548 nvlist_t *newroot = NULL, **child; 1549 uint_t c, children; 1550 zpool_boot_label_t boot_type; 1551 1552 if (argc > 0) { 1553 if ((newroot = construct_spec(props, argc, argv)) == NULL) { 1554 (void) fprintf(stderr, gettext("Unable to build a " 1555 "pool from the specified devices\n")); 1556 return (NULL); 1557 } 1558 1559 if (zpool_is_bootable(zhp)) 1560 boot_type = ZPOOL_COPY_BOOT_LABEL; 1561 else 1562 boot_type = ZPOOL_NO_BOOT_LABEL; 1563 1564 if (!flags.dryrun && 1565 make_disks(zhp, newroot, boot_type, 0) != 0) { 1566 nvlist_free(newroot); 1567 return (NULL); 1568 } 1569 1570 /* avoid any tricks in the spec */ 1571 verify(nvlist_lookup_nvlist_array(newroot, 1572 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0); 1573 for (c = 0; c < children; c++) { 1574 char *path; 1575 const char *type; 1576 int min, max; 1577 1578 verify(nvlist_lookup_string(child[c], 1579 ZPOOL_CONFIG_PATH, &path) == 0); 1580 if ((type = is_grouping(path, &min, &max)) != NULL) { 1581 (void) fprintf(stderr, gettext("Cannot use " 1582 "'%s' as a device for splitting\n"), type); 1583 nvlist_free(newroot); 1584 return (NULL); 1585 } 1586 } 1587 } 1588 1589 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) { 1590 nvlist_free(newroot); 1591 return (NULL); 1592 } 1593 1594 return (newroot); 1595 } 1596 1597 static int 1598 num_normal_vdevs(nvlist_t *nvroot) 1599 { 1600 nvlist_t **top; 1601 uint_t t, toplevels, normal = 0; 1602 1603 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 1604 &top, &toplevels) == 0); 1605 1606 for (t = 0; t < toplevels; t++) { 1607 uint64_t log = B_FALSE; 1608 1609 (void) nvlist_lookup_uint64(top[t], ZPOOL_CONFIG_IS_LOG, &log); 1610 if (log) 1611 continue; 1612 if (nvlist_exists(top[t], ZPOOL_CONFIG_ALLOCATION_BIAS)) 1613 continue; 1614 1615 normal++; 1616 } 1617 1618 return (normal); 1619 } 1620 1621 /* 1622 * Get and validate the contents of the given vdev specification. This ensures 1623 * that the nvlist returned is well-formed, that all the devices exist, and that 1624 * they are not currently in use by any other known consumer. The 'poolconfig' 1625 * parameter is the current configuration of the pool when adding devices 1626 * existing pool, and is used to perform additional checks, such as changing the 1627 * replication level of the pool. It can be 'NULL' to indicate that this is a 1628 * new pool. The 'force' flag controls whether devices should be forcefully 1629 * added, even if they appear in use. 1630 */ 1631 nvlist_t * 1632 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep, 1633 boolean_t replacing, boolean_t dryrun, zpool_boot_label_t boot_type, 1634 uint64_t boot_size, int argc, char **argv) 1635 { 1636 nvlist_t *newroot; 1637 nvlist_t *poolconfig = NULL; 1638 is_force = force; 1639 1640 /* 1641 * Construct the vdev specification. If this is successful, we know 1642 * that we have a valid specification, and that all devices can be 1643 * opened. 1644 */ 1645 if ((newroot = construct_spec(props, argc, argv)) == NULL) 1646 return (NULL); 1647 1648 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) 1649 return (NULL); 1650 1651 /* 1652 * Validate each device to make sure that its not shared with another 1653 * subsystem. We do this even if 'force' is set, because there are some 1654 * uses (such as a dedicated dump device) that even '-f' cannot 1655 * override. 1656 */ 1657 if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) { 1658 nvlist_free(newroot); 1659 return (NULL); 1660 } 1661 1662 /* 1663 * Check the replication level of the given vdevs and report any errors 1664 * found. We include the existing pool spec, if any, as we need to 1665 * catch changes against the existing replication level. 1666 */ 1667 if (check_rep && check_replication(poolconfig, newroot) != 0) { 1668 nvlist_free(newroot); 1669 return (NULL); 1670 } 1671 1672 /* 1673 * On pool create the new vdev spec must have one normal vdev. 1674 */ 1675 if (poolconfig == NULL && num_normal_vdevs(newroot) == 0) { 1676 vdev_error(gettext("at least one general top-level vdev must " 1677 "be specified\n")); 1678 nvlist_free(newroot); 1679 return (NULL); 1680 } 1681 1682 /* 1683 * Run through the vdev specification and label any whole disks found. 1684 */ 1685 if (!dryrun && make_disks(zhp, newroot, boot_type, boot_size) != 0) { 1686 nvlist_free(newroot); 1687 return (NULL); 1688 } 1689 1690 return (newroot); 1691 } 1692