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 static void 940 lines_to_stderr(char *lines[], int lines_cnt) 941 { 942 int i; 943 for (i = 0; i < lines_cnt; i++) { 944 fprintf(stderr, "%s\n", lines[i]); 945 } 946 } 947 948 /* 949 * Go through and find any whole disks in the vdev specification, labelling them 950 * as appropriate. When constructing the vdev spec, we were unable to open this 951 * device in order to provide a devid. Now that we have labelled the disk and 952 * know that slice 0 is valid, we can construct the devid now. 953 * 954 * If the disk was already labeled with an EFI label, we will have gotten the 955 * devid already (because we were able to open the whole disk). Otherwise, we 956 * need to get the devid after we label the disk. 957 */ 958 static int 959 make_disks(zpool_handle_t *zhp, nvlist_t *nv, boolean_t replacing) 960 { 961 nvlist_t **child; 962 uint_t c, children; 963 const char *type, *path; 964 char devpath[MAXPATHLEN]; 965 char udevpath[MAXPATHLEN]; 966 uint64_t wholedisk; 967 struct stat64 statbuf; 968 int is_exclusive = 0; 969 int fd; 970 int ret; 971 972 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0); 973 974 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 975 &child, &children) != 0) { 976 977 if (strcmp(type, VDEV_TYPE_DISK) != 0) 978 return (0); 979 980 /* 981 * We have a disk device. If this is a whole disk write 982 * out the efi partition table, otherwise write zero's to 983 * the first 4k of the partition. This is to ensure that 984 * libblkid will not misidentify the partition due to a 985 * magic value left by the previous filesystem. 986 */ 987 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path)); 988 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 989 &wholedisk)); 990 991 if (!wholedisk) { 992 /* 993 * Update device id string for mpath nodes (Linux only) 994 */ 995 if (is_mpath_whole_disk(path)) 996 update_vdev_config_dev_strs(nv); 997 998 if (!is_spare(NULL, path)) 999 (void) zero_label(path); 1000 return (0); 1001 } 1002 1003 if (realpath(path, devpath) == NULL) { 1004 ret = errno; 1005 (void) fprintf(stderr, 1006 gettext("cannot resolve path '%s'\n"), path); 1007 return (ret); 1008 } 1009 1010 /* 1011 * Remove any previously existing symlink from a udev path to 1012 * the device before labeling the disk. This ensures that 1013 * only newly created links are used. Otherwise there is a 1014 * window between when udev deletes and recreates the link 1015 * during which access attempts will fail with ENOENT. 1016 */ 1017 strlcpy(udevpath, path, MAXPATHLEN); 1018 (void) zfs_append_partition(udevpath, MAXPATHLEN); 1019 1020 fd = open(devpath, O_RDWR|O_EXCL); 1021 if (fd == -1) { 1022 if (errno == EBUSY) 1023 is_exclusive = 1; 1024 #ifdef __FreeBSD__ 1025 if (errno == EPERM) 1026 is_exclusive = 1; 1027 #endif 1028 } else { 1029 (void) close(fd); 1030 } 1031 1032 /* 1033 * If the partition exists, contains a valid spare label, 1034 * and is opened exclusively there is no need to partition 1035 * it. Hot spares have already been partitioned and are 1036 * held open exclusively by the kernel as a safety measure. 1037 * 1038 * If the provided path is for a /dev/disk/ device its 1039 * symbolic link will be removed, partition table created, 1040 * and then block until udev creates the new link. 1041 */ 1042 if (!is_exclusive && !is_spare(NULL, udevpath)) { 1043 char *devnode = strrchr(devpath, '/') + 1; 1044 char **lines = NULL; 1045 int lines_cnt = 0; 1046 1047 ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT)); 1048 if (ret == 0) { 1049 ret = lstat64(udevpath, &statbuf); 1050 if (ret == 0 && S_ISLNK(statbuf.st_mode)) 1051 (void) unlink(udevpath); 1052 } 1053 1054 /* 1055 * When labeling a pool the raw device node name 1056 * is provided as it appears under /dev/. 1057 * 1058 * Note that 'zhp' will be NULL when we're creating a 1059 * pool. 1060 */ 1061 if (zpool_prepare_and_label_disk(g_zfs, zhp, devnode, 1062 nv, zhp == NULL ? "create" : 1063 replacing ? "replace" : "add", &lines, 1064 &lines_cnt) != 0) { 1065 (void) fprintf(stderr, 1066 gettext( 1067 "Error preparing/labeling disk.\n")); 1068 if (lines_cnt > 0) { 1069 (void) fprintf(stderr, 1070 gettext("zfs_prepare_disk output:\n")); 1071 lines_to_stderr(lines, lines_cnt); 1072 } 1073 1074 libzfs_free_str_array(lines, lines_cnt); 1075 return (-1); 1076 } 1077 libzfs_free_str_array(lines, lines_cnt); 1078 1079 /* 1080 * Wait for udev to signal the device is available 1081 * by the provided path. 1082 */ 1083 ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT); 1084 if (ret) { 1085 (void) fprintf(stderr, 1086 gettext("missing link: %s was " 1087 "partitioned but %s is missing\n"), 1088 devnode, udevpath); 1089 return (ret); 1090 } 1091 1092 ret = zero_label(udevpath); 1093 if (ret) 1094 return (ret); 1095 } 1096 1097 /* 1098 * Update the path to refer to the partition. The presence of 1099 * the 'whole_disk' field indicates to the CLI that we should 1100 * chop off the partition number when displaying the device in 1101 * future output. 1102 */ 1103 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0); 1104 1105 /* 1106 * Update device id strings for whole disks (Linux only) 1107 */ 1108 update_vdev_config_dev_strs(nv); 1109 1110 return (0); 1111 } 1112 1113 for (c = 0; c < children; c++) 1114 if ((ret = make_disks(zhp, child[c], replacing)) != 0) 1115 return (ret); 1116 1117 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES, 1118 &child, &children) == 0) 1119 for (c = 0; c < children; c++) 1120 if ((ret = make_disks(zhp, child[c], replacing)) != 0) 1121 return (ret); 1122 1123 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE, 1124 &child, &children) == 0) 1125 for (c = 0; c < children; c++) 1126 if ((ret = make_disks(zhp, child[c], replacing)) != 0) 1127 return (ret); 1128 1129 return (0); 1130 } 1131 1132 /* 1133 * Go through and find any devices that are in use. We rely on libdiskmgt for 1134 * the majority of this task. 1135 */ 1136 static boolean_t 1137 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force, 1138 boolean_t replacing, boolean_t isspare) 1139 { 1140 nvlist_t **child; 1141 uint_t c, children; 1142 const char *type, *path; 1143 int ret = 0; 1144 char buf[MAXPATHLEN]; 1145 uint64_t wholedisk = B_FALSE; 1146 boolean_t anyinuse = B_FALSE; 1147 1148 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0); 1149 1150 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1151 &child, &children) != 0) { 1152 1153 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path)); 1154 if (strcmp(type, VDEV_TYPE_DISK) == 0) 1155 verify(!nvlist_lookup_uint64(nv, 1156 ZPOOL_CONFIG_WHOLE_DISK, &wholedisk)); 1157 1158 /* 1159 * As a generic check, we look to see if this is a replace of a 1160 * hot spare within the same pool. If so, we allow it 1161 * regardless of what libblkid or zpool_in_use() says. 1162 */ 1163 if (replacing) { 1164 (void) strlcpy(buf, path, sizeof (buf)); 1165 if (wholedisk) { 1166 ret = zfs_append_partition(buf, sizeof (buf)); 1167 if (ret == -1) 1168 return (-1); 1169 } 1170 1171 if (is_spare(config, buf)) 1172 return (B_FALSE); 1173 } 1174 1175 if (strcmp(type, VDEV_TYPE_DISK) == 0) 1176 ret = check_device(path, force, isspare, wholedisk); 1177 1178 else if (strcmp(type, VDEV_TYPE_FILE) == 0) 1179 ret = check_file(path, force, isspare); 1180 1181 return (ret != 0); 1182 } 1183 1184 for (c = 0; c < children; c++) 1185 if (is_device_in_use(config, child[c], force, replacing, 1186 B_FALSE)) 1187 anyinuse = B_TRUE; 1188 1189 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES, 1190 &child, &children) == 0) 1191 for (c = 0; c < children; c++) 1192 if (is_device_in_use(config, child[c], force, replacing, 1193 B_TRUE)) 1194 anyinuse = B_TRUE; 1195 1196 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE, 1197 &child, &children) == 0) 1198 for (c = 0; c < children; c++) 1199 if (is_device_in_use(config, child[c], force, replacing, 1200 B_FALSE)) 1201 anyinuse = B_TRUE; 1202 1203 return (anyinuse); 1204 } 1205 1206 /* 1207 * Returns the parity level extracted from a raidz or draid type. 1208 * If the parity cannot be determined zero is returned. 1209 */ 1210 static int 1211 get_parity(const char *type) 1212 { 1213 long parity = 0; 1214 const char *p; 1215 1216 if (strncmp(type, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0) { 1217 p = type + strlen(VDEV_TYPE_RAIDZ); 1218 1219 if (*p == '\0') { 1220 /* when unspecified default to single parity */ 1221 return (1); 1222 } else if (*p == '0') { 1223 /* no zero prefixes allowed */ 1224 return (0); 1225 } else { 1226 /* 0-3, no suffixes allowed */ 1227 char *end; 1228 errno = 0; 1229 parity = strtol(p, &end, 10); 1230 if (errno != 0 || *end != '\0' || 1231 parity < 1 || parity > VDEV_RAIDZ_MAXPARITY) { 1232 return (0); 1233 } 1234 } 1235 } else if (strncmp(type, VDEV_TYPE_DRAID, 1236 strlen(VDEV_TYPE_DRAID)) == 0) { 1237 p = type + strlen(VDEV_TYPE_DRAID); 1238 1239 if (*p == '\0' || *p == ':') { 1240 /* when unspecified default to single parity */ 1241 return (1); 1242 } else if (*p == '0') { 1243 /* no zero prefixes allowed */ 1244 return (0); 1245 } else { 1246 /* 0-3, allowed suffixes: '\0' or ':' */ 1247 char *end; 1248 errno = 0; 1249 parity = strtol(p, &end, 10); 1250 if (errno != 0 || 1251 parity < 1 || parity > VDEV_DRAID_MAXPARITY || 1252 (*end != '\0' && *end != ':')) { 1253 return (0); 1254 } 1255 } 1256 } 1257 1258 return ((int)parity); 1259 } 1260 1261 /* 1262 * Assign the minimum and maximum number of devices allowed for 1263 * the specified type. On error NULL is returned, otherwise the 1264 * type prefix is returned (raidz, mirror, etc). 1265 */ 1266 static const char * 1267 is_grouping(const char *type, int *mindev, int *maxdev) 1268 { 1269 int nparity; 1270 1271 if (strncmp(type, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 || 1272 strncmp(type, VDEV_TYPE_DRAID, strlen(VDEV_TYPE_DRAID)) == 0) { 1273 nparity = get_parity(type); 1274 if (nparity == 0) 1275 return (NULL); 1276 if (mindev != NULL) 1277 *mindev = nparity + 1; 1278 if (maxdev != NULL) 1279 *maxdev = 255; 1280 1281 if (strncmp(type, VDEV_TYPE_RAIDZ, 1282 strlen(VDEV_TYPE_RAIDZ)) == 0) { 1283 return (VDEV_TYPE_RAIDZ); 1284 } else { 1285 return (VDEV_TYPE_DRAID); 1286 } 1287 } 1288 1289 if (maxdev != NULL) 1290 *maxdev = INT_MAX; 1291 1292 if (strcmp(type, "mirror") == 0) { 1293 if (mindev != NULL) 1294 *mindev = 2; 1295 return (VDEV_TYPE_MIRROR); 1296 } 1297 1298 if (strcmp(type, "spare") == 0) { 1299 if (mindev != NULL) 1300 *mindev = 1; 1301 return (VDEV_TYPE_SPARE); 1302 } 1303 1304 if (strcmp(type, "log") == 0) { 1305 if (mindev != NULL) 1306 *mindev = 1; 1307 return (VDEV_TYPE_LOG); 1308 } 1309 1310 if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0 || 1311 strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) { 1312 if (mindev != NULL) 1313 *mindev = 1; 1314 return (type); 1315 } 1316 1317 if (strcmp(type, "cache") == 0) { 1318 if (mindev != NULL) 1319 *mindev = 1; 1320 return (VDEV_TYPE_L2CACHE); 1321 } 1322 1323 return (NULL); 1324 } 1325 1326 /* 1327 * Extract the configuration parameters encoded in the dRAID type and 1328 * use them to generate a dRAID configuration. The expected format is: 1329 * 1330 * draid[<parity>][:<data><d|D>][:<children><c|C>][:<spares><s|S>] 1331 * 1332 * The intent is to be able to generate a good configuration when no 1333 * additional information is provided. The only mandatory component 1334 * of the 'type' is the 'draid' prefix. If a value is not provided 1335 * then reasonable defaults are used. The optional components may 1336 * appear in any order but the d/s/c suffix is required. 1337 * 1338 * Valid inputs: 1339 * - data: number of data devices per group (1-255) 1340 * - parity: number of parity blocks per group (1-3) 1341 * - spares: number of distributed spare (0-100) 1342 * - children: total number of devices (1-255) 1343 * 1344 * Examples: 1345 * - zpool create tank draid <devices...> 1346 * - zpool create tank draid2:8d:51c:2s <devices...> 1347 */ 1348 static int 1349 draid_config_by_type(nvlist_t *nv, const char *type, uint64_t children) 1350 { 1351 uint64_t nparity = 1; 1352 uint64_t nspares = 0; 1353 uint64_t ndata = UINT64_MAX; 1354 uint64_t ngroups = 1; 1355 long value; 1356 1357 if (strncmp(type, VDEV_TYPE_DRAID, strlen(VDEV_TYPE_DRAID)) != 0) 1358 return (EINVAL); 1359 1360 nparity = (uint64_t)get_parity(type); 1361 if (nparity == 0 || nparity > VDEV_DRAID_MAXPARITY) { 1362 fprintf(stderr, 1363 gettext("invalid dRAID parity level %llu; must be " 1364 "between 1 and %d\n"), (u_longlong_t)nparity, 1365 VDEV_DRAID_MAXPARITY); 1366 return (EINVAL); 1367 } 1368 1369 char *p = (char *)type; 1370 while ((p = strchr(p, ':')) != NULL) { 1371 char *end; 1372 1373 p = p + 1; 1374 errno = 0; 1375 1376 if (!isdigit(p[0])) { 1377 (void) fprintf(stderr, gettext("invalid dRAID " 1378 "syntax; expected [:<number><c|d|s>] not '%s'\n"), 1379 type); 1380 return (EINVAL); 1381 } 1382 1383 /* Expected non-zero value with c/d/s suffix */ 1384 value = strtol(p, &end, 10); 1385 char suffix = tolower(*end); 1386 if (errno != 0 || 1387 (suffix != 'c' && suffix != 'd' && suffix != 's')) { 1388 (void) fprintf(stderr, gettext("invalid dRAID " 1389 "syntax; expected [:<number><c|d|s>] not '%s'\n"), 1390 type); 1391 return (EINVAL); 1392 } 1393 1394 if (suffix == 'c') { 1395 if ((uint64_t)value != children) { 1396 fprintf(stderr, 1397 gettext("invalid number of dRAID children; " 1398 "%llu required but %llu provided\n"), 1399 (u_longlong_t)value, 1400 (u_longlong_t)children); 1401 return (EINVAL); 1402 } 1403 } else if (suffix == 'd') { 1404 ndata = (uint64_t)value; 1405 } else if (suffix == 's') { 1406 nspares = (uint64_t)value; 1407 } else { 1408 verify(0); /* Unreachable */ 1409 } 1410 } 1411 1412 /* 1413 * When a specific number of data disks is not provided limit a 1414 * redundancy group to 8 data disks. This value was selected to 1415 * provide a reasonable tradeoff between capacity and performance. 1416 */ 1417 if (ndata == UINT64_MAX) { 1418 if (children > nspares + nparity) { 1419 ndata = MIN(children - nspares - nparity, 8); 1420 } else { 1421 fprintf(stderr, gettext("request number of " 1422 "distributed spares %llu and parity level %llu\n" 1423 "leaves no disks available for data\n"), 1424 (u_longlong_t)nspares, (u_longlong_t)nparity); 1425 return (EINVAL); 1426 } 1427 } 1428 1429 /* Verify the maximum allowed group size is never exceeded. */ 1430 if (ndata == 0 || (ndata + nparity > children - nspares)) { 1431 fprintf(stderr, gettext("requested number of dRAID data " 1432 "disks per group %llu is too high,\nat most %llu disks " 1433 "are available for data\n"), (u_longlong_t)ndata, 1434 (u_longlong_t)(children - nspares - nparity)); 1435 return (EINVAL); 1436 } 1437 1438 /* 1439 * Verify the requested number of spares can be satisfied. 1440 * An arbitrary limit of 100 distributed spares is applied. 1441 */ 1442 if (nspares > 100 || nspares > (children - (ndata + nparity))) { 1443 fprintf(stderr, 1444 gettext("invalid number of dRAID spares %llu; additional " 1445 "disks would be required\n"), (u_longlong_t)nspares); 1446 return (EINVAL); 1447 } 1448 1449 /* Verify the requested number children is sufficient. */ 1450 if (children < (ndata + nparity + nspares)) { 1451 fprintf(stderr, gettext("%llu disks were provided, but at " 1452 "least %llu disks are required for this config\n"), 1453 (u_longlong_t)children, 1454 (u_longlong_t)(ndata + nparity + nspares)); 1455 } 1456 1457 if (children > VDEV_DRAID_MAX_CHILDREN) { 1458 fprintf(stderr, gettext("%llu disks were provided, but " 1459 "dRAID only supports up to %u disks"), 1460 (u_longlong_t)children, VDEV_DRAID_MAX_CHILDREN); 1461 } 1462 1463 /* 1464 * Calculate the minimum number of groups required to fill a slice. 1465 * This is the LCM of the stripe width (ndata + nparity) and the 1466 * number of data drives (children - nspares). 1467 */ 1468 while (ngroups * (ndata + nparity) % (children - nspares) != 0) 1469 ngroups++; 1470 1471 /* Store the basic dRAID configuration. */ 1472 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, nparity); 1473 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NDATA, ndata); 1474 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NSPARES, nspares); 1475 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NGROUPS, ngroups); 1476 1477 return (0); 1478 } 1479 1480 /* 1481 * Construct a syntactically valid vdev specification, 1482 * and ensure that all devices and files exist and can be opened. 1483 * Note: we don't bother freeing anything in the error paths 1484 * because the program is just going to exit anyway. 1485 */ 1486 static nvlist_t * 1487 construct_spec(nvlist_t *props, int argc, char **argv) 1488 { 1489 nvlist_t *nvroot, *nv, **top, **spares, **l2cache; 1490 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache; 1491 const char *type, *fulltype; 1492 boolean_t is_log, is_special, is_dedup, is_spare; 1493 boolean_t seen_logs; 1494 1495 top = NULL; 1496 toplevels = 0; 1497 spares = NULL; 1498 l2cache = NULL; 1499 nspares = 0; 1500 nlogs = 0; 1501 nl2cache = 0; 1502 is_log = is_special = is_dedup = is_spare = B_FALSE; 1503 seen_logs = B_FALSE; 1504 nvroot = NULL; 1505 1506 while (argc > 0) { 1507 fulltype = argv[0]; 1508 nv = NULL; 1509 1510 /* 1511 * If it's a mirror, raidz, or draid the subsequent arguments 1512 * are its leaves -- until we encounter the next mirror, 1513 * raidz or draid. 1514 */ 1515 if ((type = is_grouping(fulltype, &mindev, &maxdev)) != NULL) { 1516 nvlist_t **child = NULL; 1517 int c, children = 0; 1518 1519 if (strcmp(type, VDEV_TYPE_SPARE) == 0) { 1520 if (spares != NULL) { 1521 (void) fprintf(stderr, 1522 gettext("invalid vdev " 1523 "specification: 'spare' can be " 1524 "specified only once\n")); 1525 goto spec_out; 1526 } 1527 is_spare = B_TRUE; 1528 is_log = is_special = is_dedup = B_FALSE; 1529 } 1530 1531 if (strcmp(type, VDEV_TYPE_LOG) == 0) { 1532 if (seen_logs) { 1533 (void) fprintf(stderr, 1534 gettext("invalid vdev " 1535 "specification: 'log' can be " 1536 "specified only once\n")); 1537 goto spec_out; 1538 } 1539 seen_logs = B_TRUE; 1540 is_log = B_TRUE; 1541 is_special = is_dedup = is_spare = B_FALSE; 1542 argc--; 1543 argv++; 1544 /* 1545 * A log is not a real grouping device. 1546 * We just set is_log and continue. 1547 */ 1548 continue; 1549 } 1550 1551 if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0) { 1552 is_special = B_TRUE; 1553 is_log = is_dedup = is_spare = B_FALSE; 1554 argc--; 1555 argv++; 1556 continue; 1557 } 1558 1559 if (strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) { 1560 is_dedup = B_TRUE; 1561 is_log = is_special = is_spare = B_FALSE; 1562 argc--; 1563 argv++; 1564 continue; 1565 } 1566 1567 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) { 1568 if (l2cache != NULL) { 1569 (void) fprintf(stderr, 1570 gettext("invalid vdev " 1571 "specification: 'cache' can be " 1572 "specified only once\n")); 1573 goto spec_out; 1574 } 1575 is_log = is_special = B_FALSE; 1576 is_dedup = is_spare = B_FALSE; 1577 } 1578 1579 if (is_log || is_special || is_dedup) { 1580 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) { 1581 (void) fprintf(stderr, 1582 gettext("invalid vdev " 1583 "specification: unsupported '%s' " 1584 "device: %s\n"), is_log ? "log" : 1585 "special", type); 1586 goto spec_out; 1587 } 1588 nlogs++; 1589 } 1590 1591 for (c = 1; c < argc; c++) { 1592 if (is_grouping(argv[c], NULL, NULL) != NULL) 1593 break; 1594 1595 children++; 1596 child = realloc(child, 1597 children * sizeof (nvlist_t *)); 1598 if (child == NULL) 1599 zpool_no_memory(); 1600 if ((nv = make_leaf_vdev(props, argv[c], 1601 !(is_log || is_special || is_dedup || 1602 is_spare))) == NULL) { 1603 for (c = 0; c < children - 1; c++) 1604 nvlist_free(child[c]); 1605 free(child); 1606 goto spec_out; 1607 } 1608 1609 child[children - 1] = nv; 1610 } 1611 1612 if (children < mindev) { 1613 (void) fprintf(stderr, gettext("invalid vdev " 1614 "specification: %s requires at least %d " 1615 "devices\n"), argv[0], mindev); 1616 for (c = 0; c < children; c++) 1617 nvlist_free(child[c]); 1618 free(child); 1619 goto spec_out; 1620 } 1621 1622 if (children > maxdev) { 1623 (void) fprintf(stderr, gettext("invalid vdev " 1624 "specification: %s supports no more than " 1625 "%d devices\n"), argv[0], maxdev); 1626 for (c = 0; c < children; c++) 1627 nvlist_free(child[c]); 1628 free(child); 1629 goto spec_out; 1630 } 1631 1632 argc -= c; 1633 argv += c; 1634 1635 if (strcmp(type, VDEV_TYPE_SPARE) == 0) { 1636 spares = child; 1637 nspares = children; 1638 continue; 1639 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) { 1640 l2cache = child; 1641 nl2cache = children; 1642 continue; 1643 } else { 1644 /* create a top-level vdev with children */ 1645 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME, 1646 0) == 0); 1647 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE, 1648 type) == 0); 1649 verify(nvlist_add_uint64(nv, 1650 ZPOOL_CONFIG_IS_LOG, is_log) == 0); 1651 if (is_log) { 1652 verify(nvlist_add_string(nv, 1653 ZPOOL_CONFIG_ALLOCATION_BIAS, 1654 VDEV_ALLOC_BIAS_LOG) == 0); 1655 } 1656 if (is_special) { 1657 verify(nvlist_add_string(nv, 1658 ZPOOL_CONFIG_ALLOCATION_BIAS, 1659 VDEV_ALLOC_BIAS_SPECIAL) == 0); 1660 } 1661 if (is_dedup) { 1662 verify(nvlist_add_string(nv, 1663 ZPOOL_CONFIG_ALLOCATION_BIAS, 1664 VDEV_ALLOC_BIAS_DEDUP) == 0); 1665 } 1666 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) { 1667 verify(nvlist_add_uint64(nv, 1668 ZPOOL_CONFIG_NPARITY, 1669 mindev - 1) == 0); 1670 } 1671 if (strcmp(type, VDEV_TYPE_DRAID) == 0) { 1672 if (draid_config_by_type(nv, 1673 fulltype, children) != 0) { 1674 for (c = 0; c < children; c++) 1675 nvlist_free(child[c]); 1676 free(child); 1677 goto spec_out; 1678 } 1679 } 1680 verify(nvlist_add_nvlist_array(nv, 1681 ZPOOL_CONFIG_CHILDREN, 1682 (const nvlist_t **)child, children) == 0); 1683 1684 for (c = 0; c < children; c++) 1685 nvlist_free(child[c]); 1686 free(child); 1687 } 1688 } else { 1689 /* 1690 * We have a device. Pass off to make_leaf_vdev() to 1691 * construct the appropriate nvlist describing the vdev. 1692 */ 1693 if ((nv = make_leaf_vdev(props, argv[0], !(is_log || 1694 is_special || is_dedup || is_spare))) == NULL) 1695 goto spec_out; 1696 1697 verify(nvlist_add_uint64(nv, 1698 ZPOOL_CONFIG_IS_LOG, is_log) == 0); 1699 if (is_log) { 1700 verify(nvlist_add_string(nv, 1701 ZPOOL_CONFIG_ALLOCATION_BIAS, 1702 VDEV_ALLOC_BIAS_LOG) == 0); 1703 nlogs++; 1704 } 1705 1706 if (is_special) { 1707 verify(nvlist_add_string(nv, 1708 ZPOOL_CONFIG_ALLOCATION_BIAS, 1709 VDEV_ALLOC_BIAS_SPECIAL) == 0); 1710 } 1711 if (is_dedup) { 1712 verify(nvlist_add_string(nv, 1713 ZPOOL_CONFIG_ALLOCATION_BIAS, 1714 VDEV_ALLOC_BIAS_DEDUP) == 0); 1715 } 1716 argc--; 1717 argv++; 1718 } 1719 1720 toplevels++; 1721 top = realloc(top, toplevels * sizeof (nvlist_t *)); 1722 if (top == NULL) 1723 zpool_no_memory(); 1724 top[toplevels - 1] = nv; 1725 } 1726 1727 if (toplevels == 0 && nspares == 0 && nl2cache == 0) { 1728 (void) fprintf(stderr, gettext("invalid vdev " 1729 "specification: at least one toplevel vdev must be " 1730 "specified\n")); 1731 goto spec_out; 1732 } 1733 1734 if (seen_logs && nlogs == 0) { 1735 (void) fprintf(stderr, gettext("invalid vdev specification: " 1736 "log requires at least 1 device\n")); 1737 goto spec_out; 1738 } 1739 1740 /* 1741 * Finally, create nvroot and add all top-level vdevs to it. 1742 */ 1743 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0); 1744 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 1745 VDEV_TYPE_ROOT) == 0); 1746 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 1747 (const nvlist_t **)top, toplevels) == 0); 1748 if (nspares != 0) 1749 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 1750 (const nvlist_t **)spares, nspares) == 0); 1751 if (nl2cache != 0) 1752 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 1753 (const nvlist_t **)l2cache, nl2cache) == 0); 1754 1755 spec_out: 1756 for (t = 0; t < toplevels; t++) 1757 nvlist_free(top[t]); 1758 for (t = 0; t < nspares; t++) 1759 nvlist_free(spares[t]); 1760 for (t = 0; t < nl2cache; t++) 1761 nvlist_free(l2cache[t]); 1762 1763 free(spares); 1764 free(l2cache); 1765 free(top); 1766 1767 return (nvroot); 1768 } 1769 1770 nvlist_t * 1771 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props, 1772 splitflags_t flags, int argc, char **argv) 1773 { 1774 nvlist_t *newroot = NULL, **child; 1775 uint_t c, children; 1776 1777 if (argc > 0) { 1778 if ((newroot = construct_spec(props, argc, argv)) == NULL) { 1779 (void) fprintf(stderr, gettext("Unable to build a " 1780 "pool from the specified devices\n")); 1781 return (NULL); 1782 } 1783 1784 if (!flags.dryrun && make_disks(zhp, newroot, B_FALSE) != 0) { 1785 nvlist_free(newroot); 1786 return (NULL); 1787 } 1788 1789 /* avoid any tricks in the spec */ 1790 verify(nvlist_lookup_nvlist_array(newroot, 1791 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0); 1792 for (c = 0; c < children; c++) { 1793 const char *path; 1794 const char *type; 1795 int min, max; 1796 1797 verify(nvlist_lookup_string(child[c], 1798 ZPOOL_CONFIG_PATH, &path) == 0); 1799 if ((type = is_grouping(path, &min, &max)) != NULL) { 1800 (void) fprintf(stderr, gettext("Cannot use " 1801 "'%s' as a device for splitting\n"), type); 1802 nvlist_free(newroot); 1803 return (NULL); 1804 } 1805 } 1806 } 1807 1808 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) { 1809 nvlist_free(newroot); 1810 return (NULL); 1811 } 1812 1813 return (newroot); 1814 } 1815 1816 static int 1817 num_normal_vdevs(nvlist_t *nvroot) 1818 { 1819 nvlist_t **top; 1820 uint_t t, toplevels, normal = 0; 1821 1822 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 1823 &top, &toplevels) == 0); 1824 1825 for (t = 0; t < toplevels; t++) { 1826 uint64_t log = B_FALSE; 1827 1828 (void) nvlist_lookup_uint64(top[t], ZPOOL_CONFIG_IS_LOG, &log); 1829 if (log) 1830 continue; 1831 if (nvlist_exists(top[t], ZPOOL_CONFIG_ALLOCATION_BIAS)) 1832 continue; 1833 1834 normal++; 1835 } 1836 1837 return (normal); 1838 } 1839 1840 /* 1841 * Get and validate the contents of the given vdev specification. This ensures 1842 * that the nvlist returned is well-formed, that all the devices exist, and that 1843 * they are not currently in use by any other known consumer. The 'poolconfig' 1844 * parameter is the current configuration of the pool when adding devices 1845 * existing pool, and is used to perform additional checks, such as changing the 1846 * replication level of the pool. It can be 'NULL' to indicate that this is a 1847 * new pool. The 'force' flag controls whether devices should be forcefully 1848 * added, even if they appear in use. 1849 */ 1850 nvlist_t * 1851 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep, 1852 boolean_t replacing, boolean_t dryrun, int argc, char **argv) 1853 { 1854 nvlist_t *newroot; 1855 nvlist_t *poolconfig = NULL; 1856 is_force = force; 1857 1858 /* 1859 * Construct the vdev specification. If this is successful, we know 1860 * that we have a valid specification, and that all devices can be 1861 * opened. 1862 */ 1863 if ((newroot = construct_spec(props, argc, argv)) == NULL) 1864 return (NULL); 1865 1866 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) { 1867 nvlist_free(newroot); 1868 return (NULL); 1869 } 1870 1871 /* 1872 * Validate each device to make sure that it's not shared with another 1873 * subsystem. We do this even if 'force' is set, because there are some 1874 * uses (such as a dedicated dump device) that even '-f' cannot 1875 * override. 1876 */ 1877 if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) { 1878 nvlist_free(newroot); 1879 return (NULL); 1880 } 1881 1882 /* 1883 * Check the replication level of the given vdevs and report any errors 1884 * found. We include the existing pool spec, if any, as we need to 1885 * catch changes against the existing replication level. 1886 */ 1887 if (check_rep && check_replication(poolconfig, newroot) != 0) { 1888 nvlist_free(newroot); 1889 return (NULL); 1890 } 1891 1892 /* 1893 * On pool create the new vdev spec must have one normal vdev. 1894 */ 1895 if (poolconfig == NULL && num_normal_vdevs(newroot) == 0) { 1896 vdev_error(gettext("at least one general top-level vdev must " 1897 "be specified\n")); 1898 nvlist_free(newroot); 1899 return (NULL); 1900 } 1901 1902 /* 1903 * Run through the vdev specification and label any whole disks found. 1904 */ 1905 if (!dryrun && make_disks(zhp, newroot, replacing) != 0) { 1906 nvlist_free(newroot); 1907 return (NULL); 1908 } 1909 1910 return (newroot); 1911 } 1912