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 https://opensource.org/licenses/CDDL-1.0. 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 libblkid to make 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 <ctype.h> 67 #include <errno.h> 68 #include <fcntl.h> 69 #include <libintl.h> 70 #include <libnvpair.h> 71 #include <libzutil.h> 72 #include <limits.h> 73 #include <sys/spa.h> 74 #include <stdio.h> 75 #include <string.h> 76 #include <unistd.h> 77 #include "zpool_util.h" 78 #include <sys/zfs_context.h> 79 #include <sys/stat.h> 80 81 /* 82 * For any given vdev specification, we can have multiple errors. The 83 * vdev_error() function keeps track of whether we have seen an error yet, and 84 * prints out a header if its the first error we've seen. 85 */ 86 boolean_t error_seen; 87 boolean_t is_force; 88 89 void 90 vdev_error(const char *fmt, ...) 91 { 92 va_list ap; 93 94 if (!error_seen) { 95 (void) fprintf(stderr, gettext("invalid vdev specification\n")); 96 if (!is_force) 97 (void) fprintf(stderr, gettext("use '-f' to override " 98 "the following errors:\n")); 99 else 100 (void) fprintf(stderr, gettext("the following errors " 101 "must be manually repaired:\n")); 102 error_seen = B_TRUE; 103 } 104 105 va_start(ap, fmt); 106 (void) vfprintf(stderr, fmt, ap); 107 va_end(ap); 108 } 109 110 /* 111 * Check that a file is valid. All we can do in this case is check that it's 112 * not in use by another pool, and not in use by swap. 113 */ 114 int 115 check_file_generic(const char *file, boolean_t force, boolean_t isspare) 116 { 117 char *name; 118 int fd; 119 int ret = 0; 120 pool_state_t state; 121 boolean_t inuse; 122 123 if ((fd = open(file, O_RDONLY)) < 0) 124 return (0); 125 126 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) { 127 const char *desc; 128 129 switch (state) { 130 case POOL_STATE_ACTIVE: 131 desc = gettext("active"); 132 break; 133 134 case POOL_STATE_EXPORTED: 135 desc = gettext("exported"); 136 break; 137 138 case POOL_STATE_POTENTIALLY_ACTIVE: 139 desc = gettext("potentially active"); 140 break; 141 142 default: 143 desc = gettext("unknown"); 144 break; 145 } 146 147 /* 148 * Allow hot spares to be shared between pools. 149 */ 150 if (state == POOL_STATE_SPARE && isspare) { 151 free(name); 152 (void) close(fd); 153 return (0); 154 } 155 156 if (state == POOL_STATE_ACTIVE || 157 state == POOL_STATE_SPARE || !force) { 158 switch (state) { 159 case POOL_STATE_SPARE: 160 vdev_error(gettext("%s is reserved as a hot " 161 "spare for pool %s\n"), file, name); 162 break; 163 default: 164 vdev_error(gettext("%s is part of %s pool " 165 "'%s'\n"), file, desc, name); 166 break; 167 } 168 ret = -1; 169 } 170 171 free(name); 172 } 173 174 (void) close(fd); 175 return (ret); 176 } 177 178 /* 179 * This may be a shorthand device path or it could be total gibberish. 180 * Check to see if it is a known device available in zfs_vdev_paths. 181 * As part of this check, see if we've been given an entire disk 182 * (minus the slice number). 183 */ 184 static int 185 is_shorthand_path(const char *arg, char *path, size_t path_size, 186 struct stat64 *statbuf, boolean_t *wholedisk) 187 { 188 int error; 189 190 error = zfs_resolve_shortname(arg, path, path_size); 191 if (error == 0) { 192 *wholedisk = zfs_dev_is_whole_disk(path); 193 if (*wholedisk || (stat64(path, statbuf) == 0)) 194 return (0); 195 } 196 197 strlcpy(path, arg, path_size); 198 memset(statbuf, 0, sizeof (*statbuf)); 199 *wholedisk = B_FALSE; 200 201 return (error); 202 } 203 204 /* 205 * Determine if the given path is a hot spare within the given configuration. 206 * If no configuration is given we rely solely on the label. 207 */ 208 static boolean_t 209 is_spare(nvlist_t *config, const char *path) 210 { 211 int fd; 212 pool_state_t state; 213 char *name = NULL; 214 nvlist_t *label; 215 uint64_t guid, spareguid; 216 nvlist_t *nvroot; 217 nvlist_t **spares; 218 uint_t i, nspares; 219 boolean_t inuse; 220 221 if (zpool_is_draid_spare(path)) 222 return (B_TRUE); 223 224 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) 225 return (B_FALSE); 226 227 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 || 228 !inuse || 229 state != POOL_STATE_SPARE || 230 zpool_read_label(fd, &label, NULL) != 0) { 231 free(name); 232 (void) close(fd); 233 return (B_FALSE); 234 } 235 free(name); 236 (void) close(fd); 237 238 if (config == NULL) { 239 nvlist_free(label); 240 return (B_TRUE); 241 } 242 243 verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0); 244 nvlist_free(label); 245 246 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 247 &nvroot) == 0); 248 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 249 &spares, &nspares) == 0) { 250 for (i = 0; i < nspares; i++) { 251 verify(nvlist_lookup_uint64(spares[i], 252 ZPOOL_CONFIG_GUID, &spareguid) == 0); 253 if (spareguid == guid) 254 return (B_TRUE); 255 } 256 } 257 258 return (B_FALSE); 259 } 260 261 /* 262 * Create a leaf vdev. Determine if this is a file or a device. If it's a 263 * device, fill in the device id to make a complete nvlist. Valid forms for a 264 * leaf vdev are: 265 * 266 * /dev/xxx Complete disk path 267 * /xxx Full path to file 268 * xxx Shorthand for <zfs_vdev_paths>/xxx 269 * draid* Virtual dRAID spare 270 */ 271 static nvlist_t * 272 make_leaf_vdev(nvlist_t *props, const char *arg, boolean_t is_primary) 273 { 274 char path[MAXPATHLEN]; 275 struct stat64 statbuf; 276 nvlist_t *vdev = NULL; 277 const char *type = NULL; 278 boolean_t wholedisk = B_FALSE; 279 uint64_t ashift = 0; 280 int err; 281 282 /* 283 * Determine what type of vdev this is, and put the full path into 284 * 'path'. We detect whether this is a device of file afterwards by 285 * checking the st_mode of the file. 286 */ 287 if (arg[0] == '/') { 288 /* 289 * Complete device or file path. Exact type is determined by 290 * examining the file descriptor afterwards. Symbolic links 291 * are resolved to their real paths to determine whole disk 292 * and S_ISBLK/S_ISREG type checks. However, we are careful 293 * to store the given path as ZPOOL_CONFIG_PATH to ensure we 294 * can leverage udev's persistent device labels. 295 */ 296 if (realpath(arg, path) == NULL) { 297 (void) fprintf(stderr, 298 gettext("cannot resolve path '%s'\n"), arg); 299 return (NULL); 300 } 301 302 wholedisk = zfs_dev_is_whole_disk(path); 303 if (!wholedisk && (stat64(path, &statbuf) != 0)) { 304 (void) fprintf(stderr, 305 gettext("cannot open '%s': %s\n"), 306 path, strerror(errno)); 307 return (NULL); 308 } 309 310 /* After whole disk check restore original passed path */ 311 strlcpy(path, arg, sizeof (path)); 312 } else if (zpool_is_draid_spare(arg)) { 313 if (!is_primary) { 314 (void) fprintf(stderr, 315 gettext("cannot open '%s': dRAID spares can only " 316 "be used to replace primary vdevs\n"), arg); 317 return (NULL); 318 } 319 320 wholedisk = B_TRUE; 321 strlcpy(path, arg, sizeof (path)); 322 type = VDEV_TYPE_DRAID_SPARE; 323 } else { 324 err = is_shorthand_path(arg, path, sizeof (path), 325 &statbuf, &wholedisk); 326 if (err != 0) { 327 /* 328 * If we got ENOENT, then the user gave us 329 * gibberish, so try to direct them with a 330 * reasonable error message. Otherwise, 331 * regurgitate strerror() since it's the best we 332 * can do. 333 */ 334 if (err == ENOENT) { 335 (void) fprintf(stderr, 336 gettext("cannot open '%s': no such " 337 "device in %s\n"), arg, DISK_ROOT); 338 (void) fprintf(stderr, 339 gettext("must be a full path or " 340 "shorthand device name\n")); 341 return (NULL); 342 } else { 343 (void) fprintf(stderr, 344 gettext("cannot open '%s': %s\n"), 345 path, strerror(errno)); 346 return (NULL); 347 } 348 } 349 } 350 351 if (type == NULL) { 352 /* 353 * Determine whether this is a device or a file. 354 */ 355 if (wholedisk || S_ISBLK(statbuf.st_mode)) { 356 type = VDEV_TYPE_DISK; 357 } else if (S_ISREG(statbuf.st_mode)) { 358 type = VDEV_TYPE_FILE; 359 } else { 360 fprintf(stderr, gettext("cannot use '%s': must " 361 "be a block device or regular file\n"), path); 362 return (NULL); 363 } 364 } 365 366 /* 367 * Finally, we have the complete device or file, and we know that it is 368 * acceptable to use. Construct the nvlist to describe this vdev. All 369 * vdevs have a 'path' element, and devices also have a 'devid' element. 370 */ 371 verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0); 372 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0); 373 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0); 374 375 if (strcmp(type, VDEV_TYPE_DISK) == 0) 376 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK, 377 (uint64_t)wholedisk) == 0); 378 379 /* 380 * Override defaults if custom properties are provided. 381 */ 382 if (props != NULL) { 383 const char *value = NULL; 384 385 if (nvlist_lookup_string(props, 386 zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0) { 387 if (zfs_nicestrtonum(NULL, value, &ashift) != 0) { 388 (void) fprintf(stderr, 389 gettext("ashift must be a number.\n")); 390 return (NULL); 391 } 392 if (ashift != 0 && 393 (ashift < ASHIFT_MIN || ashift > ASHIFT_MAX)) { 394 (void) fprintf(stderr, 395 gettext("invalid 'ashift=%" PRIu64 "' " 396 "property: only values between %" PRId32 " " 397 "and %" PRId32 " are allowed.\n"), 398 ashift, ASHIFT_MIN, ASHIFT_MAX); 399 return (NULL); 400 } 401 } 402 } 403 404 /* 405 * If the device is known to incorrectly report its physical sector 406 * size explicitly provide the known correct value. 407 */ 408 if (ashift == 0) { 409 int sector_size; 410 411 if (check_sector_size_database(path, §or_size) == B_TRUE) 412 ashift = highbit64(sector_size) - 1; 413 } 414 415 if (ashift > 0) 416 (void) nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift); 417 418 return (vdev); 419 } 420 421 /* 422 * Go through and verify the replication level of the pool is consistent. 423 * Performs the following checks: 424 * 425 * For the new spec, verifies that devices in mirrors and raidz are the 426 * same size. 427 * 428 * If the current configuration already has inconsistent replication 429 * levels, ignore any other potential problems in the new spec. 430 * 431 * Otherwise, make sure that the current spec (if there is one) and the new 432 * spec have consistent replication levels. 433 * 434 * If there is no current spec (create), make sure new spec has at least 435 * one general purpose vdev. 436 */ 437 typedef struct replication_level { 438 const char *zprl_type; 439 uint64_t zprl_children; 440 uint64_t zprl_parity; 441 } replication_level_t; 442 443 #define ZPOOL_FUZZ (16 * 1024 * 1024) 444 445 /* 446 * N.B. For the purposes of comparing replication levels dRAID can be 447 * considered functionally equivalent to raidz. 448 */ 449 static boolean_t 450 is_raidz_mirror(replication_level_t *a, replication_level_t *b, 451 replication_level_t **raidz, replication_level_t **mirror) 452 { 453 if ((strcmp(a->zprl_type, "raidz") == 0 || 454 strcmp(a->zprl_type, "draid") == 0) && 455 strcmp(b->zprl_type, "mirror") == 0) { 456 *raidz = a; 457 *mirror = b; 458 return (B_TRUE); 459 } 460 return (B_FALSE); 461 } 462 463 /* 464 * Comparison for determining if dRAID and raidz where passed in either order. 465 */ 466 static boolean_t 467 is_raidz_draid(replication_level_t *a, replication_level_t *b) 468 { 469 if ((strcmp(a->zprl_type, "raidz") == 0 || 470 strcmp(a->zprl_type, "draid") == 0) && 471 (strcmp(b->zprl_type, "raidz") == 0 || 472 strcmp(b->zprl_type, "draid") == 0)) { 473 return (B_TRUE); 474 } 475 476 return (B_FALSE); 477 } 478 479 /* 480 * Given a list of toplevel vdevs, return the current replication level. If 481 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then 482 * an error message will be displayed for each self-inconsistent vdev. 483 */ 484 static replication_level_t * 485 get_replication(nvlist_t *nvroot, boolean_t fatal) 486 { 487 nvlist_t **top; 488 uint_t t, toplevels; 489 nvlist_t **child; 490 uint_t c, children; 491 nvlist_t *nv; 492 const char *type; 493 replication_level_t lastrep = {0}; 494 replication_level_t rep; 495 replication_level_t *ret; 496 replication_level_t *raidz, *mirror; 497 boolean_t dontreport; 498 499 ret = safe_malloc(sizeof (replication_level_t)); 500 501 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 502 &top, &toplevels) == 0); 503 504 for (t = 0; t < toplevels; t++) { 505 uint64_t is_log = B_FALSE; 506 507 nv = top[t]; 508 509 /* 510 * For separate logs we ignore the top level vdev replication 511 * constraints. 512 */ 513 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log); 514 if (is_log) 515 continue; 516 517 /* 518 * Ignore holes introduced by removing aux devices, along 519 * with indirect vdevs introduced by previously removed 520 * vdevs. 521 */ 522 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0); 523 if (strcmp(type, VDEV_TYPE_HOLE) == 0 || 524 strcmp(type, VDEV_TYPE_INDIRECT) == 0) 525 continue; 526 527 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 528 &child, &children) != 0) { 529 /* 530 * This is a 'file' or 'disk' vdev. 531 */ 532 rep.zprl_type = type; 533 rep.zprl_children = 1; 534 rep.zprl_parity = 0; 535 } else { 536 int64_t vdev_size; 537 538 /* 539 * This is a mirror or RAID-Z vdev. Go through and make 540 * sure the contents are all the same (files vs. disks), 541 * keeping track of the number of elements in the 542 * process. 543 * 544 * We also check that the size of each vdev (if it can 545 * be determined) is the same. 546 */ 547 rep.zprl_type = type; 548 rep.zprl_children = 0; 549 550 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0 || 551 strcmp(type, VDEV_TYPE_DRAID) == 0) { 552 verify(nvlist_lookup_uint64(nv, 553 ZPOOL_CONFIG_NPARITY, 554 &rep.zprl_parity) == 0); 555 assert(rep.zprl_parity != 0); 556 } else { 557 rep.zprl_parity = 0; 558 } 559 560 /* 561 * The 'dontreport' variable indicates that we've 562 * already reported an error for this spec, so don't 563 * bother doing it again. 564 */ 565 type = NULL; 566 dontreport = 0; 567 vdev_size = -1LL; 568 for (c = 0; c < children; c++) { 569 nvlist_t *cnv = child[c]; 570 const char *path; 571 struct stat64 statbuf; 572 int64_t size = -1LL; 573 const char *childtype; 574 int fd, err; 575 576 rep.zprl_children++; 577 578 verify(nvlist_lookup_string(cnv, 579 ZPOOL_CONFIG_TYPE, &childtype) == 0); 580 581 /* 582 * If this is a replacing or spare vdev, then 583 * get the real first child of the vdev: do this 584 * in a loop because replacing and spare vdevs 585 * can be nested. 586 */ 587 while (strcmp(childtype, 588 VDEV_TYPE_REPLACING) == 0 || 589 strcmp(childtype, VDEV_TYPE_SPARE) == 0) { 590 nvlist_t **rchild; 591 uint_t rchildren; 592 593 verify(nvlist_lookup_nvlist_array(cnv, 594 ZPOOL_CONFIG_CHILDREN, &rchild, 595 &rchildren) == 0); 596 assert(rchildren == 2); 597 cnv = rchild[0]; 598 599 verify(nvlist_lookup_string(cnv, 600 ZPOOL_CONFIG_TYPE, 601 &childtype) == 0); 602 } 603 604 verify(nvlist_lookup_string(cnv, 605 ZPOOL_CONFIG_PATH, &path) == 0); 606 607 /* 608 * If we have a raidz/mirror that combines disks 609 * with files, report it as an error. 610 */ 611 if (!dontreport && type != NULL && 612 strcmp(type, childtype) != 0) { 613 if (ret != NULL) 614 free(ret); 615 ret = NULL; 616 if (fatal) 617 vdev_error(gettext( 618 "mismatched replication " 619 "level: %s contains both " 620 "files and devices\n"), 621 rep.zprl_type); 622 else 623 return (NULL); 624 dontreport = B_TRUE; 625 } 626 627 /* 628 * According to stat(2), the value of 'st_size' 629 * is undefined for block devices and character 630 * devices. But there is no effective way to 631 * determine the real size in userland. 632 * 633 * Instead, we'll take advantage of an 634 * implementation detail of spec_size(). If the 635 * device is currently open, then we (should) 636 * return a valid size. 637 * 638 * If we still don't get a valid size (indicated 639 * by a size of 0 or MAXOFFSET_T), then ignore 640 * this device altogether. 641 */ 642 if ((fd = open(path, O_RDONLY)) >= 0) { 643 err = fstat64_blk(fd, &statbuf); 644 (void) close(fd); 645 } else { 646 err = stat64(path, &statbuf); 647 } 648 649 if (err != 0 || 650 statbuf.st_size == 0 || 651 statbuf.st_size == MAXOFFSET_T) 652 continue; 653 654 size = statbuf.st_size; 655 656 /* 657 * Also make sure that devices and 658 * slices have a consistent size. If 659 * they differ by a significant amount 660 * (~16MB) then report an error. 661 */ 662 if (!dontreport && 663 (vdev_size != -1LL && 664 (llabs(size - vdev_size) > 665 ZPOOL_FUZZ))) { 666 if (ret != NULL) 667 free(ret); 668 ret = NULL; 669 if (fatal) 670 vdev_error(gettext( 671 "%s contains devices of " 672 "different sizes\n"), 673 rep.zprl_type); 674 else 675 return (NULL); 676 dontreport = B_TRUE; 677 } 678 679 type = childtype; 680 vdev_size = size; 681 } 682 } 683 684 /* 685 * At this point, we have the replication of the last toplevel 686 * vdev in 'rep'. Compare it to 'lastrep' to see if it is 687 * different. 688 */ 689 if (lastrep.zprl_type != NULL) { 690 if (is_raidz_mirror(&lastrep, &rep, &raidz, &mirror) || 691 is_raidz_mirror(&rep, &lastrep, &raidz, &mirror)) { 692 /* 693 * Accepted raidz and mirror when they can 694 * handle the same number of disk failures. 695 */ 696 if (raidz->zprl_parity != 697 mirror->zprl_children - 1) { 698 if (ret != NULL) 699 free(ret); 700 ret = NULL; 701 if (fatal) 702 vdev_error(gettext( 703 "mismatched replication " 704 "level: " 705 "%s and %s vdevs with " 706 "different redundancy, " 707 "%llu vs. %llu (%llu-way) " 708 "are present\n"), 709 raidz->zprl_type, 710 mirror->zprl_type, 711 (u_longlong_t) 712 raidz->zprl_parity, 713 (u_longlong_t) 714 mirror->zprl_children - 1, 715 (u_longlong_t) 716 mirror->zprl_children); 717 else 718 return (NULL); 719 } 720 } else if (is_raidz_draid(&lastrep, &rep)) { 721 /* 722 * Accepted raidz and draid when they can 723 * handle the same number of disk failures. 724 */ 725 if (lastrep.zprl_parity != rep.zprl_parity) { 726 if (ret != NULL) 727 free(ret); 728 ret = NULL; 729 if (fatal) 730 vdev_error(gettext( 731 "mismatched replication " 732 "level: %s and %s vdevs " 733 "with different " 734 "redundancy, %llu vs. " 735 "%llu are present\n"), 736 lastrep.zprl_type, 737 rep.zprl_type, 738 (u_longlong_t) 739 lastrep.zprl_parity, 740 (u_longlong_t) 741 rep.zprl_parity); 742 else 743 return (NULL); 744 } 745 } else if (strcmp(lastrep.zprl_type, rep.zprl_type) != 746 0) { 747 if (ret != NULL) 748 free(ret); 749 ret = NULL; 750 if (fatal) 751 vdev_error(gettext( 752 "mismatched replication level: " 753 "both %s and %s vdevs are " 754 "present\n"), 755 lastrep.zprl_type, rep.zprl_type); 756 else 757 return (NULL); 758 } else if (lastrep.zprl_parity != rep.zprl_parity) { 759 if (ret) 760 free(ret); 761 ret = NULL; 762 if (fatal) 763 vdev_error(gettext( 764 "mismatched replication level: " 765 "both %llu and %llu device parity " 766 "%s vdevs are present\n"), 767 (u_longlong_t) 768 lastrep.zprl_parity, 769 (u_longlong_t)rep.zprl_parity, 770 rep.zprl_type); 771 else 772 return (NULL); 773 } else if (lastrep.zprl_children != rep.zprl_children) { 774 if (ret) 775 free(ret); 776 ret = NULL; 777 if (fatal) 778 vdev_error(gettext( 779 "mismatched replication level: " 780 "both %llu-way and %llu-way %s " 781 "vdevs are present\n"), 782 (u_longlong_t) 783 lastrep.zprl_children, 784 (u_longlong_t) 785 rep.zprl_children, 786 rep.zprl_type); 787 else 788 return (NULL); 789 } 790 } 791 lastrep = rep; 792 } 793 794 if (ret != NULL) 795 *ret = rep; 796 797 return (ret); 798 } 799 800 /* 801 * Check the replication level of the vdev spec against the current pool. Calls 802 * get_replication() to make sure the new spec is self-consistent. If the pool 803 * has a consistent replication level, then we ignore any errors. Otherwise, 804 * report any difference between the two. 805 */ 806 static int 807 check_replication(nvlist_t *config, nvlist_t *newroot) 808 { 809 nvlist_t **child; 810 uint_t children; 811 replication_level_t *current = NULL, *new; 812 replication_level_t *raidz, *mirror; 813 int ret; 814 815 /* 816 * If we have a current pool configuration, check to see if it's 817 * self-consistent. If not, simply return success. 818 */ 819 if (config != NULL) { 820 nvlist_t *nvroot; 821 822 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 823 &nvroot) == 0); 824 if ((current = get_replication(nvroot, B_FALSE)) == NULL) 825 return (0); 826 } 827 /* 828 * for spares there may be no children, and therefore no 829 * replication level to check 830 */ 831 if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN, 832 &child, &children) != 0) || (children == 0)) { 833 free(current); 834 return (0); 835 } 836 837 /* 838 * If all we have is logs then there's no replication level to check. 839 */ 840 if (num_logs(newroot) == children) { 841 free(current); 842 return (0); 843 } 844 845 /* 846 * Get the replication level of the new vdev spec, reporting any 847 * inconsistencies found. 848 */ 849 if ((new = get_replication(newroot, B_TRUE)) == NULL) { 850 free(current); 851 return (-1); 852 } 853 854 /* 855 * Check to see if the new vdev spec matches the replication level of 856 * the current pool. 857 */ 858 ret = 0; 859 if (current != NULL) { 860 if (is_raidz_mirror(current, new, &raidz, &mirror) || 861 is_raidz_mirror(new, current, &raidz, &mirror)) { 862 if (raidz->zprl_parity != mirror->zprl_children - 1) { 863 vdev_error(gettext( 864 "mismatched replication level: pool and " 865 "new vdev with different redundancy, %s " 866 "and %s vdevs, %llu vs. %llu (%llu-way)\n"), 867 raidz->zprl_type, 868 mirror->zprl_type, 869 (u_longlong_t)raidz->zprl_parity, 870 (u_longlong_t)mirror->zprl_children - 1, 871 (u_longlong_t)mirror->zprl_children); 872 ret = -1; 873 } 874 } else if (strcmp(current->zprl_type, new->zprl_type) != 0) { 875 vdev_error(gettext( 876 "mismatched replication level: pool uses %s " 877 "and new vdev is %s\n"), 878 current->zprl_type, new->zprl_type); 879 ret = -1; 880 } else if (current->zprl_parity != new->zprl_parity) { 881 vdev_error(gettext( 882 "mismatched replication level: pool uses %llu " 883 "device parity and new vdev uses %llu\n"), 884 (u_longlong_t)current->zprl_parity, 885 (u_longlong_t)new->zprl_parity); 886 ret = -1; 887 } else if (current->zprl_children != new->zprl_children) { 888 vdev_error(gettext( 889 "mismatched replication level: pool uses %llu-way " 890 "%s and new vdev uses %llu-way %s\n"), 891 (u_longlong_t)current->zprl_children, 892 current->zprl_type, 893 (u_longlong_t)new->zprl_children, 894 new->zprl_type); 895 ret = -1; 896 } 897 } 898 899 free(new); 900 if (current != NULL) 901 free(current); 902 903 return (ret); 904 } 905 906 static int 907 zero_label(const char *path) 908 { 909 const int size = 4096; 910 char buf[size]; 911 int err, fd; 912 913 if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) { 914 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"), 915 path, strerror(errno)); 916 return (-1); 917 } 918 919 memset(buf, 0, size); 920 err = write(fd, buf, size); 921 (void) fdatasync(fd); 922 (void) close(fd); 923 924 if (err == -1) { 925 (void) fprintf(stderr, gettext("cannot zero first %d bytes " 926 "of '%s': %s\n"), size, path, strerror(errno)); 927 return (-1); 928 } 929 930 if (err != size) { 931 (void) fprintf(stderr, gettext("could only zero %d/%d bytes " 932 "of '%s'\n"), err, size, path); 933 return (-1); 934 } 935 936 return (0); 937 } 938 939 /* 940 * Go through and find any whole disks in the vdev specification, labelling them 941 * as appropriate. When constructing the vdev spec, we were unable to open this 942 * device in order to provide a devid. Now that we have labelled the disk and 943 * know that slice 0 is valid, we can construct the devid now. 944 * 945 * If the disk was already labeled with an EFI label, we will have gotten the 946 * devid already (because we were able to open the whole disk). Otherwise, we 947 * need to get the devid after we label the disk. 948 */ 949 static int 950 make_disks(zpool_handle_t *zhp, nvlist_t *nv) 951 { 952 nvlist_t **child; 953 uint_t c, children; 954 const char *type, *path; 955 char devpath[MAXPATHLEN]; 956 char udevpath[MAXPATHLEN]; 957 uint64_t wholedisk; 958 struct stat64 statbuf; 959 int is_exclusive = 0; 960 int fd; 961 int ret; 962 963 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0); 964 965 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 966 &child, &children) != 0) { 967 968 if (strcmp(type, VDEV_TYPE_DISK) != 0) 969 return (0); 970 971 /* 972 * We have a disk device. If this is a whole disk write 973 * out the efi partition table, otherwise write zero's to 974 * the first 4k of the partition. This is to ensure that 975 * libblkid will not misidentify the partition due to a 976 * magic value left by the previous filesystem. 977 */ 978 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path)); 979 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 980 &wholedisk)); 981 982 if (!wholedisk) { 983 /* 984 * Update device id string for mpath nodes (Linux only) 985 */ 986 if (is_mpath_whole_disk(path)) 987 update_vdev_config_dev_strs(nv); 988 989 if (!is_spare(NULL, path)) 990 (void) zero_label(path); 991 return (0); 992 } 993 994 if (realpath(path, devpath) == NULL) { 995 ret = errno; 996 (void) fprintf(stderr, 997 gettext("cannot resolve path '%s'\n"), path); 998 return (ret); 999 } 1000 1001 /* 1002 * Remove any previously existing symlink from a udev path to 1003 * the device before labeling the disk. This ensures that 1004 * only newly created links are used. Otherwise there is a 1005 * window between when udev deletes and recreates the link 1006 * during which access attempts will fail with ENOENT. 1007 */ 1008 strlcpy(udevpath, path, MAXPATHLEN); 1009 (void) zfs_append_partition(udevpath, MAXPATHLEN); 1010 1011 fd = open(devpath, O_RDWR|O_EXCL); 1012 if (fd == -1) { 1013 if (errno == EBUSY) 1014 is_exclusive = 1; 1015 #ifdef __FreeBSD__ 1016 if (errno == EPERM) 1017 is_exclusive = 1; 1018 #endif 1019 } else { 1020 (void) close(fd); 1021 } 1022 1023 /* 1024 * If the partition exists, contains a valid spare label, 1025 * and is opened exclusively there is no need to partition 1026 * it. Hot spares have already been partitioned and are 1027 * held open exclusively by the kernel as a safety measure. 1028 * 1029 * If the provided path is for a /dev/disk/ device its 1030 * symbolic link will be removed, partition table created, 1031 * and then block until udev creates the new link. 1032 */ 1033 if (!is_exclusive && !is_spare(NULL, udevpath)) { 1034 char *devnode = strrchr(devpath, '/') + 1; 1035 1036 ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT)); 1037 if (ret == 0) { 1038 ret = lstat64(udevpath, &statbuf); 1039 if (ret == 0 && S_ISLNK(statbuf.st_mode)) 1040 (void) unlink(udevpath); 1041 } 1042 1043 /* 1044 * When labeling a pool the raw device node name 1045 * is provided as it appears under /dev/. 1046 */ 1047 if (zpool_label_disk(g_zfs, zhp, devnode) == -1) 1048 return (-1); 1049 1050 /* 1051 * Wait for udev to signal the device is available 1052 * by the provided path. 1053 */ 1054 ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT); 1055 if (ret) { 1056 (void) fprintf(stderr, 1057 gettext("missing link: %s was " 1058 "partitioned but %s is missing\n"), 1059 devnode, udevpath); 1060 return (ret); 1061 } 1062 1063 ret = zero_label(udevpath); 1064 if (ret) 1065 return (ret); 1066 } 1067 1068 /* 1069 * Update the path to refer to the partition. The presence of 1070 * the 'whole_disk' field indicates to the CLI that we should 1071 * chop off the partition number when displaying the device in 1072 * future output. 1073 */ 1074 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0); 1075 1076 /* 1077 * Update device id strings for whole disks (Linux only) 1078 */ 1079 update_vdev_config_dev_strs(nv); 1080 1081 return (0); 1082 } 1083 1084 for (c = 0; c < children; c++) 1085 if ((ret = make_disks(zhp, child[c])) != 0) 1086 return (ret); 1087 1088 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES, 1089 &child, &children) == 0) 1090 for (c = 0; c < children; c++) 1091 if ((ret = make_disks(zhp, child[c])) != 0) 1092 return (ret); 1093 1094 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE, 1095 &child, &children) == 0) 1096 for (c = 0; c < children; c++) 1097 if ((ret = make_disks(zhp, child[c])) != 0) 1098 return (ret); 1099 1100 return (0); 1101 } 1102 1103 /* 1104 * Go through and find any devices that are in use. We rely on libdiskmgt for 1105 * the majority of this task. 1106 */ 1107 static boolean_t 1108 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force, 1109 boolean_t replacing, boolean_t isspare) 1110 { 1111 nvlist_t **child; 1112 uint_t c, children; 1113 const char *type, *path; 1114 int ret = 0; 1115 char buf[MAXPATHLEN]; 1116 uint64_t wholedisk = B_FALSE; 1117 boolean_t anyinuse = B_FALSE; 1118 1119 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0); 1120 1121 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1122 &child, &children) != 0) { 1123 1124 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path)); 1125 if (strcmp(type, VDEV_TYPE_DISK) == 0) 1126 verify(!nvlist_lookup_uint64(nv, 1127 ZPOOL_CONFIG_WHOLE_DISK, &wholedisk)); 1128 1129 /* 1130 * As a generic check, we look to see if this is a replace of a 1131 * hot spare within the same pool. If so, we allow it 1132 * regardless of what libblkid or zpool_in_use() says. 1133 */ 1134 if (replacing) { 1135 (void) strlcpy(buf, path, sizeof (buf)); 1136 if (wholedisk) { 1137 ret = zfs_append_partition(buf, sizeof (buf)); 1138 if (ret == -1) 1139 return (-1); 1140 } 1141 1142 if (is_spare(config, buf)) 1143 return (B_FALSE); 1144 } 1145 1146 if (strcmp(type, VDEV_TYPE_DISK) == 0) 1147 ret = check_device(path, force, isspare, wholedisk); 1148 1149 else if (strcmp(type, VDEV_TYPE_FILE) == 0) 1150 ret = check_file(path, force, isspare); 1151 1152 return (ret != 0); 1153 } 1154 1155 for (c = 0; c < children; c++) 1156 if (is_device_in_use(config, child[c], force, replacing, 1157 B_FALSE)) 1158 anyinuse = B_TRUE; 1159 1160 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES, 1161 &child, &children) == 0) 1162 for (c = 0; c < children; c++) 1163 if (is_device_in_use(config, child[c], force, replacing, 1164 B_TRUE)) 1165 anyinuse = B_TRUE; 1166 1167 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE, 1168 &child, &children) == 0) 1169 for (c = 0; c < children; c++) 1170 if (is_device_in_use(config, child[c], force, replacing, 1171 B_FALSE)) 1172 anyinuse = B_TRUE; 1173 1174 return (anyinuse); 1175 } 1176 1177 /* 1178 * Returns the parity level extracted from a raidz or draid type. 1179 * If the parity cannot be determined zero is returned. 1180 */ 1181 static int 1182 get_parity(const char *type) 1183 { 1184 long parity = 0; 1185 const char *p; 1186 1187 if (strncmp(type, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0) { 1188 p = type + strlen(VDEV_TYPE_RAIDZ); 1189 1190 if (*p == '\0') { 1191 /* when unspecified default to single parity */ 1192 return (1); 1193 } else if (*p == '0') { 1194 /* no zero prefixes allowed */ 1195 return (0); 1196 } else { 1197 /* 0-3, no suffixes allowed */ 1198 char *end; 1199 errno = 0; 1200 parity = strtol(p, &end, 10); 1201 if (errno != 0 || *end != '\0' || 1202 parity < 1 || parity > VDEV_RAIDZ_MAXPARITY) { 1203 return (0); 1204 } 1205 } 1206 } else if (strncmp(type, VDEV_TYPE_DRAID, 1207 strlen(VDEV_TYPE_DRAID)) == 0) { 1208 p = type + strlen(VDEV_TYPE_DRAID); 1209 1210 if (*p == '\0' || *p == ':') { 1211 /* when unspecified default to single parity */ 1212 return (1); 1213 } else if (*p == '0') { 1214 /* no zero prefixes allowed */ 1215 return (0); 1216 } else { 1217 /* 0-3, allowed suffixes: '\0' or ':' */ 1218 char *end; 1219 errno = 0; 1220 parity = strtol(p, &end, 10); 1221 if (errno != 0 || 1222 parity < 1 || parity > VDEV_DRAID_MAXPARITY || 1223 (*end != '\0' && *end != ':')) { 1224 return (0); 1225 } 1226 } 1227 } 1228 1229 return ((int)parity); 1230 } 1231 1232 /* 1233 * Assign the minimum and maximum number of devices allowed for 1234 * the specified type. On error NULL is returned, otherwise the 1235 * type prefix is returned (raidz, mirror, etc). 1236 */ 1237 static const char * 1238 is_grouping(const char *type, int *mindev, int *maxdev) 1239 { 1240 int nparity; 1241 1242 if (strncmp(type, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 || 1243 strncmp(type, VDEV_TYPE_DRAID, strlen(VDEV_TYPE_DRAID)) == 0) { 1244 nparity = get_parity(type); 1245 if (nparity == 0) 1246 return (NULL); 1247 if (mindev != NULL) 1248 *mindev = nparity + 1; 1249 if (maxdev != NULL) 1250 *maxdev = 255; 1251 1252 if (strncmp(type, VDEV_TYPE_RAIDZ, 1253 strlen(VDEV_TYPE_RAIDZ)) == 0) { 1254 return (VDEV_TYPE_RAIDZ); 1255 } else { 1256 return (VDEV_TYPE_DRAID); 1257 } 1258 } 1259 1260 if (maxdev != NULL) 1261 *maxdev = INT_MAX; 1262 1263 if (strcmp(type, "mirror") == 0) { 1264 if (mindev != NULL) 1265 *mindev = 2; 1266 return (VDEV_TYPE_MIRROR); 1267 } 1268 1269 if (strcmp(type, "spare") == 0) { 1270 if (mindev != NULL) 1271 *mindev = 1; 1272 return (VDEV_TYPE_SPARE); 1273 } 1274 1275 if (strcmp(type, "log") == 0) { 1276 if (mindev != NULL) 1277 *mindev = 1; 1278 return (VDEV_TYPE_LOG); 1279 } 1280 1281 if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0 || 1282 strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) { 1283 if (mindev != NULL) 1284 *mindev = 1; 1285 return (type); 1286 } 1287 1288 if (strcmp(type, "cache") == 0) { 1289 if (mindev != NULL) 1290 *mindev = 1; 1291 return (VDEV_TYPE_L2CACHE); 1292 } 1293 1294 return (NULL); 1295 } 1296 1297 /* 1298 * Extract the configuration parameters encoded in the dRAID type and 1299 * use them to generate a dRAID configuration. The expected format is: 1300 * 1301 * draid[<parity>][:<data><d|D>][:<children><c|C>][:<spares><s|S>] 1302 * 1303 * The intent is to be able to generate a good configuration when no 1304 * additional information is provided. The only mandatory component 1305 * of the 'type' is the 'draid' prefix. If a value is not provided 1306 * then reasonable defaults are used. The optional components may 1307 * appear in any order but the d/s/c suffix is required. 1308 * 1309 * Valid inputs: 1310 * - data: number of data devices per group (1-255) 1311 * - parity: number of parity blocks per group (1-3) 1312 * - spares: number of distributed spare (0-100) 1313 * - children: total number of devices (1-255) 1314 * 1315 * Examples: 1316 * - zpool create tank draid <devices...> 1317 * - zpool create tank draid2:8d:51c:2s <devices...> 1318 */ 1319 static int 1320 draid_config_by_type(nvlist_t *nv, const char *type, uint64_t children) 1321 { 1322 uint64_t nparity = 1; 1323 uint64_t nspares = 0; 1324 uint64_t ndata = UINT64_MAX; 1325 uint64_t ngroups = 1; 1326 long value; 1327 1328 if (strncmp(type, VDEV_TYPE_DRAID, strlen(VDEV_TYPE_DRAID)) != 0) 1329 return (EINVAL); 1330 1331 nparity = (uint64_t)get_parity(type); 1332 if (nparity == 0 || nparity > VDEV_DRAID_MAXPARITY) { 1333 fprintf(stderr, 1334 gettext("invalid dRAID parity level %llu; must be " 1335 "between 1 and %d\n"), (u_longlong_t)nparity, 1336 VDEV_DRAID_MAXPARITY); 1337 return (EINVAL); 1338 } 1339 1340 char *p = (char *)type; 1341 while ((p = strchr(p, ':')) != NULL) { 1342 char *end; 1343 1344 p = p + 1; 1345 errno = 0; 1346 1347 if (!isdigit(p[0])) { 1348 (void) fprintf(stderr, gettext("invalid dRAID " 1349 "syntax; expected [:<number><c|d|s>] not '%s'\n"), 1350 type); 1351 return (EINVAL); 1352 } 1353 1354 /* Expected non-zero value with c/d/s suffix */ 1355 value = strtol(p, &end, 10); 1356 char suffix = tolower(*end); 1357 if (errno != 0 || 1358 (suffix != 'c' && suffix != 'd' && suffix != 's')) { 1359 (void) fprintf(stderr, gettext("invalid dRAID " 1360 "syntax; expected [:<number><c|d|s>] not '%s'\n"), 1361 type); 1362 return (EINVAL); 1363 } 1364 1365 if (suffix == 'c') { 1366 if ((uint64_t)value != children) { 1367 fprintf(stderr, 1368 gettext("invalid number of dRAID children; " 1369 "%llu required but %llu provided\n"), 1370 (u_longlong_t)value, 1371 (u_longlong_t)children); 1372 return (EINVAL); 1373 } 1374 } else if (suffix == 'd') { 1375 ndata = (uint64_t)value; 1376 } else if (suffix == 's') { 1377 nspares = (uint64_t)value; 1378 } else { 1379 verify(0); /* Unreachable */ 1380 } 1381 } 1382 1383 /* 1384 * When a specific number of data disks is not provided limit a 1385 * redundancy group to 8 data disks. This value was selected to 1386 * provide a reasonable tradeoff between capacity and performance. 1387 */ 1388 if (ndata == UINT64_MAX) { 1389 if (children > nspares + nparity) { 1390 ndata = MIN(children - nspares - nparity, 8); 1391 } else { 1392 fprintf(stderr, gettext("request number of " 1393 "distributed spares %llu and parity level %llu\n" 1394 "leaves no disks available for data\n"), 1395 (u_longlong_t)nspares, (u_longlong_t)nparity); 1396 return (EINVAL); 1397 } 1398 } 1399 1400 /* Verify the maximum allowed group size is never exceeded. */ 1401 if (ndata == 0 || (ndata + nparity > children - nspares)) { 1402 fprintf(stderr, gettext("requested number of dRAID data " 1403 "disks per group %llu is too high,\nat most %llu disks " 1404 "are available for data\n"), (u_longlong_t)ndata, 1405 (u_longlong_t)(children - nspares - nparity)); 1406 return (EINVAL); 1407 } 1408 1409 /* 1410 * Verify the requested number of spares can be satisfied. 1411 * An arbitrary limit of 100 distributed spares is applied. 1412 */ 1413 if (nspares > 100 || nspares > (children - (ndata + nparity))) { 1414 fprintf(stderr, 1415 gettext("invalid number of dRAID spares %llu; additional " 1416 "disks would be required\n"), (u_longlong_t)nspares); 1417 return (EINVAL); 1418 } 1419 1420 /* Verify the requested number children is sufficient. */ 1421 if (children < (ndata + nparity + nspares)) { 1422 fprintf(stderr, gettext("%llu disks were provided, but at " 1423 "least %llu disks are required for this config\n"), 1424 (u_longlong_t)children, 1425 (u_longlong_t)(ndata + nparity + nspares)); 1426 } 1427 1428 if (children > VDEV_DRAID_MAX_CHILDREN) { 1429 fprintf(stderr, gettext("%llu disks were provided, but " 1430 "dRAID only supports up to %u disks"), 1431 (u_longlong_t)children, VDEV_DRAID_MAX_CHILDREN); 1432 } 1433 1434 /* 1435 * Calculate the minimum number of groups required to fill a slice. 1436 * This is the LCM of the stripe width (ndata + nparity) and the 1437 * number of data drives (children - nspares). 1438 */ 1439 while (ngroups * (ndata + nparity) % (children - nspares) != 0) 1440 ngroups++; 1441 1442 /* Store the basic dRAID configuration. */ 1443 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, nparity); 1444 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NDATA, ndata); 1445 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NSPARES, nspares); 1446 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NGROUPS, ngroups); 1447 1448 return (0); 1449 } 1450 1451 /* 1452 * Construct a syntactically valid vdev specification, 1453 * and ensure that all devices and files exist and can be opened. 1454 * Note: we don't bother freeing anything in the error paths 1455 * because the program is just going to exit anyway. 1456 */ 1457 static nvlist_t * 1458 construct_spec(nvlist_t *props, int argc, char **argv) 1459 { 1460 nvlist_t *nvroot, *nv, **top, **spares, **l2cache; 1461 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache; 1462 const char *type, *fulltype; 1463 boolean_t is_log, is_special, is_dedup, is_spare; 1464 boolean_t seen_logs; 1465 1466 top = NULL; 1467 toplevels = 0; 1468 spares = NULL; 1469 l2cache = NULL; 1470 nspares = 0; 1471 nlogs = 0; 1472 nl2cache = 0; 1473 is_log = is_special = is_dedup = is_spare = B_FALSE; 1474 seen_logs = B_FALSE; 1475 nvroot = NULL; 1476 1477 while (argc > 0) { 1478 fulltype = argv[0]; 1479 nv = NULL; 1480 1481 /* 1482 * If it's a mirror, raidz, or draid the subsequent arguments 1483 * are its leaves -- until we encounter the next mirror, 1484 * raidz or draid. 1485 */ 1486 if ((type = is_grouping(fulltype, &mindev, &maxdev)) != NULL) { 1487 nvlist_t **child = NULL; 1488 int c, children = 0; 1489 1490 if (strcmp(type, VDEV_TYPE_SPARE) == 0) { 1491 if (spares != NULL) { 1492 (void) fprintf(stderr, 1493 gettext("invalid vdev " 1494 "specification: 'spare' can be " 1495 "specified only once\n")); 1496 goto spec_out; 1497 } 1498 is_spare = B_TRUE; 1499 is_log = is_special = is_dedup = B_FALSE; 1500 } 1501 1502 if (strcmp(type, VDEV_TYPE_LOG) == 0) { 1503 if (seen_logs) { 1504 (void) fprintf(stderr, 1505 gettext("invalid vdev " 1506 "specification: 'log' can be " 1507 "specified only once\n")); 1508 goto spec_out; 1509 } 1510 seen_logs = B_TRUE; 1511 is_log = B_TRUE; 1512 is_special = is_dedup = is_spare = B_FALSE; 1513 argc--; 1514 argv++; 1515 /* 1516 * A log is not a real grouping device. 1517 * We just set is_log and continue. 1518 */ 1519 continue; 1520 } 1521 1522 if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0) { 1523 is_special = B_TRUE; 1524 is_log = is_dedup = is_spare = B_FALSE; 1525 argc--; 1526 argv++; 1527 continue; 1528 } 1529 1530 if (strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) { 1531 is_dedup = B_TRUE; 1532 is_log = is_special = is_spare = B_FALSE; 1533 argc--; 1534 argv++; 1535 continue; 1536 } 1537 1538 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) { 1539 if (l2cache != NULL) { 1540 (void) fprintf(stderr, 1541 gettext("invalid vdev " 1542 "specification: 'cache' can be " 1543 "specified only once\n")); 1544 goto spec_out; 1545 } 1546 is_log = is_special = B_FALSE; 1547 is_dedup = is_spare = B_FALSE; 1548 } 1549 1550 if (is_log || is_special || is_dedup) { 1551 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) { 1552 (void) fprintf(stderr, 1553 gettext("invalid vdev " 1554 "specification: unsupported '%s' " 1555 "device: %s\n"), is_log ? "log" : 1556 "special", type); 1557 goto spec_out; 1558 } 1559 nlogs++; 1560 } 1561 1562 for (c = 1; c < argc; c++) { 1563 if (is_grouping(argv[c], NULL, NULL) != NULL) 1564 break; 1565 1566 children++; 1567 child = realloc(child, 1568 children * sizeof (nvlist_t *)); 1569 if (child == NULL) 1570 zpool_no_memory(); 1571 if ((nv = make_leaf_vdev(props, argv[c], 1572 !(is_log || is_special || is_dedup || 1573 is_spare))) == NULL) { 1574 for (c = 0; c < children - 1; c++) 1575 nvlist_free(child[c]); 1576 free(child); 1577 goto spec_out; 1578 } 1579 1580 child[children - 1] = nv; 1581 } 1582 1583 if (children < mindev) { 1584 (void) fprintf(stderr, gettext("invalid vdev " 1585 "specification: %s requires at least %d " 1586 "devices\n"), argv[0], mindev); 1587 for (c = 0; c < children; c++) 1588 nvlist_free(child[c]); 1589 free(child); 1590 goto spec_out; 1591 } 1592 1593 if (children > maxdev) { 1594 (void) fprintf(stderr, gettext("invalid vdev " 1595 "specification: %s supports no more than " 1596 "%d devices\n"), argv[0], maxdev); 1597 for (c = 0; c < children; c++) 1598 nvlist_free(child[c]); 1599 free(child); 1600 goto spec_out; 1601 } 1602 1603 argc -= c; 1604 argv += c; 1605 1606 if (strcmp(type, VDEV_TYPE_SPARE) == 0) { 1607 spares = child; 1608 nspares = children; 1609 continue; 1610 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) { 1611 l2cache = child; 1612 nl2cache = children; 1613 continue; 1614 } else { 1615 /* create a top-level vdev with children */ 1616 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME, 1617 0) == 0); 1618 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 1619 type) == 0); 1620 verify(nvlist_add_uint64(nv, 1621 ZPOOL_CONFIG_IS_LOG, is_log) == 0); 1622 if (is_log) { 1623 verify(nvlist_add_string(nv, 1624 ZPOOL_CONFIG_ALLOCATION_BIAS, 1625 VDEV_ALLOC_BIAS_LOG) == 0); 1626 } 1627 if (is_special) { 1628 verify(nvlist_add_string(nv, 1629 ZPOOL_CONFIG_ALLOCATION_BIAS, 1630 VDEV_ALLOC_BIAS_SPECIAL) == 0); 1631 } 1632 if (is_dedup) { 1633 verify(nvlist_add_string(nv, 1634 ZPOOL_CONFIG_ALLOCATION_BIAS, 1635 VDEV_ALLOC_BIAS_DEDUP) == 0); 1636 } 1637 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) { 1638 verify(nvlist_add_uint64(nv, 1639 ZPOOL_CONFIG_NPARITY, 1640 mindev - 1) == 0); 1641 } 1642 if (strcmp(type, VDEV_TYPE_DRAID) == 0) { 1643 if (draid_config_by_type(nv, 1644 fulltype, children) != 0) { 1645 for (c = 0; c < children; c++) 1646 nvlist_free(child[c]); 1647 free(child); 1648 goto spec_out; 1649 } 1650 } 1651 verify(nvlist_add_nvlist_array(nv, 1652 ZPOOL_CONFIG_CHILDREN, 1653 (const nvlist_t **)child, children) == 0); 1654 1655 for (c = 0; c < children; c++) 1656 nvlist_free(child[c]); 1657 free(child); 1658 } 1659 } else { 1660 /* 1661 * We have a device. Pass off to make_leaf_vdev() to 1662 * construct the appropriate nvlist describing the vdev. 1663 */ 1664 if ((nv = make_leaf_vdev(props, argv[0], !(is_log || 1665 is_special || is_dedup || is_spare))) == NULL) 1666 goto spec_out; 1667 1668 verify(nvlist_add_uint64(nv, 1669 ZPOOL_CONFIG_IS_LOG, is_log) == 0); 1670 if (is_log) { 1671 verify(nvlist_add_string(nv, 1672 ZPOOL_CONFIG_ALLOCATION_BIAS, 1673 VDEV_ALLOC_BIAS_LOG) == 0); 1674 nlogs++; 1675 } 1676 1677 if (is_special) { 1678 verify(nvlist_add_string(nv, 1679 ZPOOL_CONFIG_ALLOCATION_BIAS, 1680 VDEV_ALLOC_BIAS_SPECIAL) == 0); 1681 } 1682 if (is_dedup) { 1683 verify(nvlist_add_string(nv, 1684 ZPOOL_CONFIG_ALLOCATION_BIAS, 1685 VDEV_ALLOC_BIAS_DEDUP) == 0); 1686 } 1687 argc--; 1688 argv++; 1689 } 1690 1691 toplevels++; 1692 top = realloc(top, toplevels * sizeof (nvlist_t *)); 1693 if (top == NULL) 1694 zpool_no_memory(); 1695 top[toplevels - 1] = nv; 1696 } 1697 1698 if (toplevels == 0 && nspares == 0 && nl2cache == 0) { 1699 (void) fprintf(stderr, gettext("invalid vdev " 1700 "specification: at least one toplevel vdev must be " 1701 "specified\n")); 1702 goto spec_out; 1703 } 1704 1705 if (seen_logs && nlogs == 0) { 1706 (void) fprintf(stderr, gettext("invalid vdev specification: " 1707 "log requires at least 1 device\n")); 1708 goto spec_out; 1709 } 1710 1711 /* 1712 * Finally, create nvroot and add all top-level vdevs to it. 1713 */ 1714 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0); 1715 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 1716 VDEV_TYPE_ROOT) == 0); 1717 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 1718 (const nvlist_t **)top, toplevels) == 0); 1719 if (nspares != 0) 1720 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 1721 (const nvlist_t **)spares, nspares) == 0); 1722 if (nl2cache != 0) 1723 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 1724 (const nvlist_t **)l2cache, nl2cache) == 0); 1725 1726 spec_out: 1727 for (t = 0; t < toplevels; t++) 1728 nvlist_free(top[t]); 1729 for (t = 0; t < nspares; t++) 1730 nvlist_free(spares[t]); 1731 for (t = 0; t < nl2cache; t++) 1732 nvlist_free(l2cache[t]); 1733 1734 free(spares); 1735 free(l2cache); 1736 free(top); 1737 1738 return (nvroot); 1739 } 1740 1741 nvlist_t * 1742 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props, 1743 splitflags_t flags, int argc, char **argv) 1744 { 1745 nvlist_t *newroot = NULL, **child; 1746 uint_t c, children; 1747 1748 if (argc > 0) { 1749 if ((newroot = construct_spec(props, argc, argv)) == NULL) { 1750 (void) fprintf(stderr, gettext("Unable to build a " 1751 "pool from the specified devices\n")); 1752 return (NULL); 1753 } 1754 1755 if (!flags.dryrun && make_disks(zhp, newroot) != 0) { 1756 nvlist_free(newroot); 1757 return (NULL); 1758 } 1759 1760 /* avoid any tricks in the spec */ 1761 verify(nvlist_lookup_nvlist_array(newroot, 1762 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0); 1763 for (c = 0; c < children; c++) { 1764 const char *path; 1765 const char *type; 1766 int min, max; 1767 1768 verify(nvlist_lookup_string(child[c], 1769 ZPOOL_CONFIG_PATH, &path) == 0); 1770 if ((type = is_grouping(path, &min, &max)) != NULL) { 1771 (void) fprintf(stderr, gettext("Cannot use " 1772 "'%s' as a device for splitting\n"), type); 1773 nvlist_free(newroot); 1774 return (NULL); 1775 } 1776 } 1777 } 1778 1779 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) { 1780 nvlist_free(newroot); 1781 return (NULL); 1782 } 1783 1784 return (newroot); 1785 } 1786 1787 static int 1788 num_normal_vdevs(nvlist_t *nvroot) 1789 { 1790 nvlist_t **top; 1791 uint_t t, toplevels, normal = 0; 1792 1793 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 1794 &top, &toplevels) == 0); 1795 1796 for (t = 0; t < toplevels; t++) { 1797 uint64_t log = B_FALSE; 1798 1799 (void) nvlist_lookup_uint64(top[t], ZPOOL_CONFIG_IS_LOG, &log); 1800 if (log) 1801 continue; 1802 if (nvlist_exists(top[t], ZPOOL_CONFIG_ALLOCATION_BIAS)) 1803 continue; 1804 1805 normal++; 1806 } 1807 1808 return (normal); 1809 } 1810 1811 /* 1812 * Get and validate the contents of the given vdev specification. This ensures 1813 * that the nvlist returned is well-formed, that all the devices exist, and that 1814 * they are not currently in use by any other known consumer. The 'poolconfig' 1815 * parameter is the current configuration of the pool when adding devices 1816 * existing pool, and is used to perform additional checks, such as changing the 1817 * replication level of the pool. It can be 'NULL' to indicate that this is a 1818 * new pool. The 'force' flag controls whether devices should be forcefully 1819 * added, even if they appear in use. 1820 */ 1821 nvlist_t * 1822 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep, 1823 boolean_t replacing, boolean_t dryrun, int argc, char **argv) 1824 { 1825 nvlist_t *newroot; 1826 nvlist_t *poolconfig = NULL; 1827 is_force = force; 1828 1829 /* 1830 * Construct the vdev specification. If this is successful, we know 1831 * that we have a valid specification, and that all devices can be 1832 * opened. 1833 */ 1834 if ((newroot = construct_spec(props, argc, argv)) == NULL) 1835 return (NULL); 1836 1837 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) { 1838 nvlist_free(newroot); 1839 return (NULL); 1840 } 1841 1842 /* 1843 * Validate each device to make sure that it's not shared with another 1844 * subsystem. We do this even if 'force' is set, because there are some 1845 * uses (such as a dedicated dump device) that even '-f' cannot 1846 * override. 1847 */ 1848 if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) { 1849 nvlist_free(newroot); 1850 return (NULL); 1851 } 1852 1853 /* 1854 * Check the replication level of the given vdevs and report any errors 1855 * found. We include the existing pool spec, if any, as we need to 1856 * catch changes against the existing replication level. 1857 */ 1858 if (check_rep && check_replication(poolconfig, newroot) != 0) { 1859 nvlist_free(newroot); 1860 return (NULL); 1861 } 1862 1863 /* 1864 * On pool create the new vdev spec must have one normal vdev. 1865 */ 1866 if (poolconfig == NULL && num_normal_vdevs(newroot) == 0) { 1867 vdev_error(gettext("at least one general top-level vdev must " 1868 "be specified\n")); 1869 nvlist_free(newroot); 1870 return (NULL); 1871 } 1872 1873 /* 1874 * Run through the vdev specification and label any whole disks found. 1875 */ 1876 if (!dryrun && make_disks(zhp, newroot) != 0) { 1877 nvlist_free(newroot); 1878 return (NULL); 1879 } 1880 1881 return (newroot); 1882 } 1883