1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or https://opensource.org/licenses/CDDL-1.0.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22
23 /*
24 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 * Copyright (c) 2013, 2018 by Delphix. All rights reserved.
26 * Copyright (c) 2016, 2017 Intel Corporation.
27 * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>.
28 */
29
30 /*
31 * Functions to convert between a list of vdevs and an nvlist representing the
32 * configuration. Each entry in the list can be one of:
33 *
34 * Device vdevs
35 * disk=(path=..., devid=...)
36 * file=(path=...)
37 *
38 * Group vdevs
39 * raidz[1|2]=(...)
40 * mirror=(...)
41 *
42 * Hot spares
43 *
44 * While the underlying implementation supports it, group vdevs cannot contain
45 * other group vdevs. All userland verification of devices is contained within
46 * this file. If successful, the nvlist returned can be passed directly to the
47 * kernel; we've done as much verification as possible in userland.
48 *
49 * Hot spares are a special case, and passed down as an array of disk vdevs, at
50 * the same level as the root of the vdev tree.
51 *
52 * The only function exported by this file is 'make_root_vdev'. The
53 * function performs several passes:
54 *
55 * 1. Construct the vdev specification. Performs syntax validation and
56 * makes sure each device is valid.
57 * 2. Check for devices in use. Using libblkid to make sure that no
58 * devices are also in use. Some can be overridden using the 'force'
59 * flag, others cannot.
60 * 3. Check for replication errors if the 'force' flag is not specified.
61 * validates that the replication level is consistent across the
62 * entire pool.
63 * 4. Call libzfs to label any whole disks with an EFI label.
64 */
65
66 #include <assert.h>
67 #include <ctype.h>
68 #include <errno.h>
69 #include <fcntl.h>
70 #include <libintl.h>
71 #include <libnvpair.h>
72 #include <libzutil.h>
73 #include <limits.h>
74 #include <sys/spa.h>
75 #include <stdio.h>
76 #include <string.h>
77 #include <unistd.h>
78 #include "zpool_util.h"
79 #include <sys/zfs_context.h>
80 #include <sys/stat.h>
81
82 /*
83 * For any given vdev specification, we can have multiple errors. The
84 * vdev_error() function keeps track of whether we have seen an error yet, and
85 * prints out a header if its the first error we've seen.
86 */
87 boolean_t error_seen;
88 boolean_t is_force;
89
90 void
vdev_error(const char * fmt,...)91 vdev_error(const char *fmt, ...)
92 {
93 va_list ap;
94
95 if (!error_seen) {
96 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
97 if (!is_force)
98 (void) fprintf(stderr, gettext("use '-f' to override "
99 "the following errors:\n"));
100 else
101 (void) fprintf(stderr, gettext("the following errors "
102 "must be manually repaired:\n"));
103 error_seen = B_TRUE;
104 }
105
106 va_start(ap, fmt);
107 (void) vfprintf(stderr, fmt, ap);
108 va_end(ap);
109 }
110
111 /*
112 * Check that a file is valid. All we can do in this case is check that it's
113 * not in use by another pool, and not in use by swap.
114 */
115 int
check_file_generic(const char * file,boolean_t force,boolean_t isspare)116 check_file_generic(const char *file, boolean_t force, boolean_t isspare)
117 {
118 char *name;
119 int fd;
120 int ret = 0;
121 pool_state_t state;
122 boolean_t inuse;
123
124 if ((fd = open(file, O_RDONLY)) < 0)
125 return (0);
126
127 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
128 const char *desc;
129
130 switch (state) {
131 case POOL_STATE_ACTIVE:
132 desc = gettext("active");
133 break;
134
135 case POOL_STATE_EXPORTED:
136 desc = gettext("exported");
137 break;
138
139 case POOL_STATE_POTENTIALLY_ACTIVE:
140 desc = gettext("potentially active");
141 break;
142
143 default:
144 desc = gettext("unknown");
145 break;
146 }
147
148 /*
149 * Allow hot spares to be shared between pools.
150 */
151 if (state == POOL_STATE_SPARE && isspare) {
152 free(name);
153 (void) close(fd);
154 return (0);
155 }
156
157 if (state == POOL_STATE_ACTIVE ||
158 state == POOL_STATE_SPARE || !force) {
159 switch (state) {
160 case POOL_STATE_SPARE:
161 vdev_error(gettext("%s is reserved as a hot "
162 "spare for pool %s\n"), file, name);
163 break;
164 default:
165 vdev_error(gettext("%s is part of %s pool "
166 "'%s'\n"), file, desc, name);
167 break;
168 }
169 ret = -1;
170 }
171
172 free(name);
173 }
174
175 (void) close(fd);
176 return (ret);
177 }
178
179 /*
180 * This may be a shorthand device path or it could be total gibberish.
181 * Check to see if it is a known device available in zfs_vdev_paths.
182 * As part of this check, see if we've been given an entire disk
183 * (minus the slice number).
184 */
185 static int
is_shorthand_path(const char * arg,char * path,size_t path_size,struct stat64 * statbuf,boolean_t * wholedisk)186 is_shorthand_path(const char *arg, char *path, size_t path_size,
187 struct stat64 *statbuf, boolean_t *wholedisk)
188 {
189 int error;
190
191 error = zfs_resolve_shortname(arg, path, path_size);
192 if (error == 0) {
193 *wholedisk = zfs_dev_is_whole_disk(path);
194 if (*wholedisk || (stat64(path, statbuf) == 0))
195 return (0);
196 }
197
198 strlcpy(path, arg, path_size);
199 memset(statbuf, 0, sizeof (*statbuf));
200 *wholedisk = B_FALSE;
201
202 return (error);
203 }
204
205 /*
206 * Determine if the given path is a hot spare within the given configuration.
207 * If no configuration is given we rely solely on the label.
208 */
209 static boolean_t
is_spare(nvlist_t * config,const char * path)210 is_spare(nvlist_t *config, const char *path)
211 {
212 int fd;
213 pool_state_t state;
214 char *name = NULL;
215 nvlist_t *label;
216 uint64_t guid, spareguid;
217 nvlist_t *nvroot;
218 nvlist_t **spares;
219 uint_t i, nspares;
220 boolean_t inuse;
221
222 if (zpool_is_draid_spare(path))
223 return (B_TRUE);
224
225 if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
226 return (B_FALSE);
227
228 if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
229 !inuse ||
230 state != POOL_STATE_SPARE ||
231 zpool_read_label(fd, &label, NULL) != 0) {
232 free(name);
233 (void) close(fd);
234 return (B_FALSE);
235 }
236 free(name);
237 (void) close(fd);
238
239 if (config == NULL) {
240 nvlist_free(label);
241 return (B_TRUE);
242 }
243
244 verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
245 nvlist_free(label);
246
247 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
248 &nvroot) == 0);
249 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
250 &spares, &nspares) == 0) {
251 for (i = 0; i < nspares; i++) {
252 verify(nvlist_lookup_uint64(spares[i],
253 ZPOOL_CONFIG_GUID, &spareguid) == 0);
254 if (spareguid == guid)
255 return (B_TRUE);
256 }
257 }
258
259 return (B_FALSE);
260 }
261
262 /*
263 * Create a leaf vdev. Determine if this is a file or a device. If it's a
264 * device, fill in the device id to make a complete nvlist. Valid forms for a
265 * leaf vdev are:
266 *
267 * /dev/xxx Complete disk path
268 * /xxx Full path to file
269 * xxx Shorthand for <zfs_vdev_paths>/xxx
270 * draid* Virtual dRAID spare
271 */
272 static nvlist_t *
make_leaf_vdev(const char * arg,boolean_t is_primary,uint64_t ashift)273 make_leaf_vdev(const char *arg, boolean_t is_primary, uint64_t ashift)
274 {
275 char path[MAXPATHLEN];
276 struct stat64 statbuf;
277 nvlist_t *vdev = NULL;
278 const char *type = NULL;
279 boolean_t wholedisk = B_FALSE;
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 /* Lookup and add the enclosure sysfs path (if exists) */
376 update_vdev_config_dev_sysfs_path(vdev, path,
377 ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH);
378
379 if (strcmp(type, VDEV_TYPE_DISK) == 0)
380 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
381 (uint64_t)wholedisk) == 0);
382
383 /*
384 * If the device is known to incorrectly report its physical sector
385 * size explicitly provide the known correct value.
386 */
387 if (ashift == 0) {
388 int sector_size;
389
390 if (check_sector_size_database(path, §or_size) == B_TRUE)
391 ashift = highbit64(sector_size) - 1;
392 }
393
394 if (ashift > 0)
395 (void) nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
396
397 return (vdev);
398 }
399
400 /*
401 * Go through and verify the replication level of the pool is consistent.
402 * Performs the following checks:
403 *
404 * For the new spec, verifies that devices in mirrors and raidz are the
405 * same size.
406 *
407 * If the current configuration already has inconsistent replication
408 * levels, ignore any other potential problems in the new spec.
409 *
410 * Otherwise, make sure that the current spec (if there is one) and the new
411 * spec have consistent replication levels.
412 *
413 * If there is no current spec (create), make sure new spec has at least
414 * one general purpose vdev.
415 */
416 typedef struct replication_level {
417 const char *zprl_type;
418 uint64_t zprl_children;
419 uint64_t zprl_parity;
420 } replication_level_t;
421
422 #define ZPOOL_FUZZ (16 * 1024 * 1024)
423
424 /*
425 * N.B. For the purposes of comparing replication levels dRAID can be
426 * considered functionally equivalent to raidz.
427 */
428 static boolean_t
is_raidz_mirror(replication_level_t * a,replication_level_t * b,replication_level_t ** raidz,replication_level_t ** mirror)429 is_raidz_mirror(replication_level_t *a, replication_level_t *b,
430 replication_level_t **raidz, replication_level_t **mirror)
431 {
432 if ((strcmp(a->zprl_type, "raidz") == 0 ||
433 strcmp(a->zprl_type, "draid") == 0) &&
434 strcmp(b->zprl_type, "mirror") == 0) {
435 *raidz = a;
436 *mirror = b;
437 return (B_TRUE);
438 }
439 return (B_FALSE);
440 }
441
442 /*
443 * Comparison for determining if dRAID and raidz where passed in either order.
444 */
445 static boolean_t
is_raidz_draid(replication_level_t * a,replication_level_t * b)446 is_raidz_draid(replication_level_t *a, replication_level_t *b)
447 {
448 if ((strcmp(a->zprl_type, "raidz") == 0 ||
449 strcmp(a->zprl_type, "draid") == 0) &&
450 (strcmp(b->zprl_type, "raidz") == 0 ||
451 strcmp(b->zprl_type, "draid") == 0)) {
452 return (B_TRUE);
453 }
454
455 return (B_FALSE);
456 }
457
458 /*
459 * Given a list of toplevel vdevs, return the current replication level. If
460 * the config is inconsistent, then NULL is returned. If 'fatal' is set, then
461 * an error message will be displayed for each self-inconsistent vdev.
462 */
463 static replication_level_t *
get_replication(nvlist_t * nvroot,boolean_t fatal)464 get_replication(nvlist_t *nvroot, boolean_t fatal)
465 {
466 nvlist_t **top;
467 uint_t t, toplevels;
468 nvlist_t **child;
469 uint_t c, children;
470 nvlist_t *nv;
471 const char *type;
472 replication_level_t lastrep = {0};
473 replication_level_t rep;
474 replication_level_t *ret;
475 replication_level_t *raidz, *mirror;
476 boolean_t dontreport;
477
478 ret = safe_malloc(sizeof (replication_level_t));
479
480 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
481 &top, &toplevels) == 0);
482
483 for (t = 0; t < toplevels; t++) {
484 uint64_t is_log = B_FALSE;
485
486 nv = top[t];
487
488 /*
489 * For separate logs we ignore the top level vdev replication
490 * constraints.
491 */
492 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
493 if (is_log)
494 continue;
495
496 /*
497 * Ignore holes introduced by removing aux devices, along
498 * with indirect vdevs introduced by previously removed
499 * vdevs.
500 */
501 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
502 if (strcmp(type, VDEV_TYPE_HOLE) == 0 ||
503 strcmp(type, VDEV_TYPE_INDIRECT) == 0)
504 continue;
505
506 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
507 &child, &children) != 0) {
508 /*
509 * This is a 'file' or 'disk' vdev.
510 */
511 rep.zprl_type = type;
512 rep.zprl_children = 1;
513 rep.zprl_parity = 0;
514 } else {
515 int64_t vdev_size;
516
517 /*
518 * This is a mirror or RAID-Z vdev. Go through and make
519 * sure the contents are all the same (files vs. disks),
520 * keeping track of the number of elements in the
521 * process.
522 *
523 * We also check that the size of each vdev (if it can
524 * be determined) is the same.
525 */
526 rep.zprl_type = type;
527 rep.zprl_children = 0;
528
529 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
530 strcmp(type, VDEV_TYPE_DRAID) == 0) {
531 verify(nvlist_lookup_uint64(nv,
532 ZPOOL_CONFIG_NPARITY,
533 &rep.zprl_parity) == 0);
534 assert(rep.zprl_parity != 0);
535 } else {
536 rep.zprl_parity = 0;
537 }
538
539 /*
540 * The 'dontreport' variable indicates that we've
541 * already reported an error for this spec, so don't
542 * bother doing it again.
543 */
544 type = NULL;
545 dontreport = 0;
546 vdev_size = -1LL;
547 for (c = 0; c < children; c++) {
548 nvlist_t *cnv = child[c];
549 const char *path;
550 struct stat64 statbuf;
551 const char *childtype;
552 int fd, err;
553
554 rep.zprl_children++;
555
556 verify(nvlist_lookup_string(cnv,
557 ZPOOL_CONFIG_TYPE, &childtype) == 0);
558
559 /*
560 * If this is a replacing or spare vdev, then
561 * get the real first child of the vdev: do this
562 * in a loop because replacing and spare vdevs
563 * can be nested.
564 */
565 while (strcmp(childtype,
566 VDEV_TYPE_REPLACING) == 0 ||
567 strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
568 nvlist_t **rchild;
569 uint_t rchildren;
570
571 verify(nvlist_lookup_nvlist_array(cnv,
572 ZPOOL_CONFIG_CHILDREN, &rchild,
573 &rchildren) == 0);
574 assert(rchildren == 2);
575 cnv = rchild[0];
576
577 verify(nvlist_lookup_string(cnv,
578 ZPOOL_CONFIG_TYPE,
579 &childtype) == 0);
580 }
581
582 verify(nvlist_lookup_string(cnv,
583 ZPOOL_CONFIG_PATH, &path) == 0);
584
585 /*
586 * Skip active spares they should never cause
587 * the pool to be evaluated as inconsistent.
588 */
589 if (is_spare(NULL, path))
590 continue;
591
592 /*
593 * If we have a raidz/mirror that combines disks
594 * with files, only report it as an error when
595 * fatal is set to ensure all the replication
596 * checks aren't skipped in check_replication().
597 */
598 if (fatal && !dontreport && type != NULL &&
599 strcmp(type, childtype) != 0) {
600 if (ret != NULL)
601 free(ret);
602 ret = NULL;
603 vdev_error(gettext(
604 "mismatched replication "
605 "level: %s contains both "
606 "files and devices\n"),
607 rep.zprl_type);
608 dontreport = B_TRUE;
609 }
610
611 /*
612 * According to stat(2), the value of 'st_size'
613 * is undefined for block devices and character
614 * devices. But there is no effective way to
615 * determine the real size in userland.
616 *
617 * Instead, we'll take advantage of an
618 * implementation detail of spec_size(). If the
619 * device is currently open, then we (should)
620 * return a valid size.
621 *
622 * If we still don't get a valid size (indicated
623 * by a size of 0 or MAXOFFSET_T), then ignore
624 * this device altogether.
625 */
626 if ((fd = open(path, O_RDONLY)) >= 0) {
627 err = fstat64_blk(fd, &statbuf);
628 (void) close(fd);
629 } else {
630 err = stat64(path, &statbuf);
631 }
632
633 if (err != 0 ||
634 statbuf.st_size == 0 ||
635 statbuf.st_size == MAXOFFSET_T)
636 continue;
637
638 int64_t size = statbuf.st_size;
639
640 /*
641 * Also make sure that devices and
642 * slices have a consistent size. If
643 * they differ by a significant amount
644 * (~16MB) then report an error.
645 */
646 if (!dontreport &&
647 (vdev_size != -1LL &&
648 (llabs(size - vdev_size) >
649 ZPOOL_FUZZ))) {
650 if (ret != NULL)
651 free(ret);
652 ret = NULL;
653 if (fatal)
654 vdev_error(gettext(
655 "%s contains devices of "
656 "different sizes\n"),
657 rep.zprl_type);
658 else
659 return (NULL);
660 dontreport = B_TRUE;
661 }
662
663 type = childtype;
664 vdev_size = size;
665 }
666 }
667
668 /*
669 * At this point, we have the replication of the last toplevel
670 * vdev in 'rep'. Compare it to 'lastrep' to see if it is
671 * different.
672 */
673 if (lastrep.zprl_type != NULL) {
674 if (is_raidz_mirror(&lastrep, &rep, &raidz, &mirror) ||
675 is_raidz_mirror(&rep, &lastrep, &raidz, &mirror)) {
676 /*
677 * Accepted raidz and mirror when they can
678 * handle the same number of disk failures.
679 */
680 if (raidz->zprl_parity !=
681 mirror->zprl_children - 1) {
682 if (ret != NULL)
683 free(ret);
684 ret = NULL;
685 if (fatal)
686 vdev_error(gettext(
687 "mismatched replication "
688 "level: "
689 "%s and %s vdevs with "
690 "different redundancy, "
691 "%llu vs. %llu (%llu-way) "
692 "are present\n"),
693 raidz->zprl_type,
694 mirror->zprl_type,
695 (u_longlong_t)
696 raidz->zprl_parity,
697 (u_longlong_t)
698 mirror->zprl_children - 1,
699 (u_longlong_t)
700 mirror->zprl_children);
701 else
702 return (NULL);
703 }
704 } else if (is_raidz_draid(&lastrep, &rep)) {
705 /*
706 * Accepted raidz and draid when they can
707 * handle the same number of disk failures.
708 */
709 if (lastrep.zprl_parity != rep.zprl_parity) {
710 if (ret != NULL)
711 free(ret);
712 ret = NULL;
713 if (fatal)
714 vdev_error(gettext(
715 "mismatched replication "
716 "level: %s and %s vdevs "
717 "with different "
718 "redundancy, %llu vs. "
719 "%llu are present\n"),
720 lastrep.zprl_type,
721 rep.zprl_type,
722 (u_longlong_t)
723 lastrep.zprl_parity,
724 (u_longlong_t)
725 rep.zprl_parity);
726 else
727 return (NULL);
728 }
729 } else if (strcmp(lastrep.zprl_type, rep.zprl_type) !=
730 0) {
731 if (ret != NULL)
732 free(ret);
733 ret = NULL;
734 if (fatal)
735 vdev_error(gettext(
736 "mismatched replication level: "
737 "both %s and %s vdevs are "
738 "present\n"),
739 lastrep.zprl_type, rep.zprl_type);
740 else
741 return (NULL);
742 } else if (lastrep.zprl_parity != rep.zprl_parity) {
743 if (ret)
744 free(ret);
745 ret = NULL;
746 if (fatal)
747 vdev_error(gettext(
748 "mismatched replication level: "
749 "both %llu and %llu device parity "
750 "%s vdevs are present\n"),
751 (u_longlong_t)
752 lastrep.zprl_parity,
753 (u_longlong_t)rep.zprl_parity,
754 rep.zprl_type);
755 else
756 return (NULL);
757 } else if (lastrep.zprl_children != rep.zprl_children) {
758 if (ret)
759 free(ret);
760 ret = NULL;
761 if (fatal)
762 vdev_error(gettext(
763 "mismatched replication level: "
764 "both %llu-way and %llu-way %s "
765 "vdevs are present\n"),
766 (u_longlong_t)
767 lastrep.zprl_children,
768 (u_longlong_t)
769 rep.zprl_children,
770 rep.zprl_type);
771 else
772 return (NULL);
773 }
774 }
775 lastrep = rep;
776 }
777
778 if (ret != NULL)
779 *ret = rep;
780
781 return (ret);
782 }
783
784 /*
785 * Check the replication level of the vdev spec against the current pool. Calls
786 * get_replication() to make sure the new spec is self-consistent. If the pool
787 * has a consistent replication level, then we ignore any errors. Otherwise,
788 * report any difference between the two.
789 */
790 static int
check_replication(nvlist_t * config,nvlist_t * newroot)791 check_replication(nvlist_t *config, nvlist_t *newroot)
792 {
793 nvlist_t **child;
794 uint_t children;
795 replication_level_t *current = NULL, *new;
796 replication_level_t *raidz, *mirror;
797 int ret;
798
799 /*
800 * If we have a current pool configuration, check to see if it's
801 * self-consistent. If not, simply return success.
802 */
803 if (config != NULL) {
804 nvlist_t *nvroot;
805
806 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
807 &nvroot) == 0);
808 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
809 return (0);
810 }
811 /*
812 * for spares there may be no children, and therefore no
813 * replication level to check
814 */
815 if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
816 &child, &children) != 0) || (children == 0)) {
817 free(current);
818 return (0);
819 }
820
821 /*
822 * If all we have is logs then there's no replication level to check.
823 */
824 if (num_logs(newroot) == children) {
825 free(current);
826 return (0);
827 }
828
829 /*
830 * Get the replication level of the new vdev spec, reporting any
831 * inconsistencies found.
832 */
833 if ((new = get_replication(newroot, B_TRUE)) == NULL) {
834 free(current);
835 return (-1);
836 }
837
838 /*
839 * Check to see if the new vdev spec matches the replication level of
840 * the current pool.
841 */
842 ret = 0;
843 if (current != NULL) {
844 if (is_raidz_mirror(current, new, &raidz, &mirror) ||
845 is_raidz_mirror(new, current, &raidz, &mirror)) {
846 if (raidz->zprl_parity != mirror->zprl_children - 1) {
847 vdev_error(gettext(
848 "mismatched replication level: pool and "
849 "new vdev with different redundancy, %s "
850 "and %s vdevs, %llu vs. %llu (%llu-way)\n"),
851 raidz->zprl_type,
852 mirror->zprl_type,
853 (u_longlong_t)raidz->zprl_parity,
854 (u_longlong_t)mirror->zprl_children - 1,
855 (u_longlong_t)mirror->zprl_children);
856 ret = -1;
857 }
858 } else if (is_raidz_draid(current, new)) {
859 if (current->zprl_parity != new->zprl_parity) {
860 vdev_error(gettext(
861 "mismatched replication level: pool and "
862 "new vdev with different redundancy, %s "
863 "and %s vdevs, %llu vs. %llu\n"),
864 current->zprl_type,
865 new->zprl_type,
866 (u_longlong_t)current->zprl_parity,
867 (u_longlong_t)new->zprl_parity);
868 ret = -1;
869 }
870 } else if (strcmp(current->zprl_type, new->zprl_type) != 0) {
871 vdev_error(gettext(
872 "mismatched replication level: pool uses %s "
873 "and new vdev is %s\n"),
874 current->zprl_type, new->zprl_type);
875 ret = -1;
876 } else if (current->zprl_parity != new->zprl_parity) {
877 vdev_error(gettext(
878 "mismatched replication level: pool uses %llu "
879 "device parity and new vdev uses %llu\n"),
880 (u_longlong_t)current->zprl_parity,
881 (u_longlong_t)new->zprl_parity);
882 ret = -1;
883 } else if (current->zprl_children != new->zprl_children) {
884 vdev_error(gettext(
885 "mismatched replication level: pool uses %llu-way "
886 "%s and new vdev uses %llu-way %s\n"),
887 (u_longlong_t)current->zprl_children,
888 current->zprl_type,
889 (u_longlong_t)new->zprl_children,
890 new->zprl_type);
891 ret = -1;
892 }
893 }
894
895 free(new);
896 if (current != NULL)
897 free(current);
898
899 return (ret);
900 }
901
902 static int
zero_label(const char * path)903 zero_label(const char *path)
904 {
905 const int size = 4096;
906 char buf[size];
907 int err, fd;
908
909 if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
910 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
911 path, strerror(errno));
912 return (-1);
913 }
914
915 memset(buf, 0, size);
916 err = write(fd, buf, size);
917 (void) fdatasync(fd);
918 (void) close(fd);
919
920 if (err == -1) {
921 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
922 "of '%s': %s\n"), size, path, strerror(errno));
923 return (-1);
924 }
925
926 if (err != size) {
927 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
928 "of '%s'\n"), err, size, path);
929 return (-1);
930 }
931
932 return (0);
933 }
934
935 static void
lines_to_stderr(char * lines[],int lines_cnt)936 lines_to_stderr(char *lines[], int lines_cnt)
937 {
938 int i;
939 for (i = 0; i < lines_cnt; i++) {
940 fprintf(stderr, "%s\n", lines[i]);
941 }
942 }
943
944 /*
945 * Go through and find any whole disks in the vdev specification, labelling them
946 * as appropriate. When constructing the vdev spec, we were unable to open this
947 * device in order to provide a devid. Now that we have labelled the disk and
948 * know that slice 0 is valid, we can construct the devid now.
949 *
950 * If the disk was already labeled with an EFI label, we will have gotten the
951 * devid already (because we were able to open the whole disk). Otherwise, we
952 * need to get the devid after we label the disk.
953 */
954 static int
make_disks(zpool_handle_t * zhp,nvlist_t * nv,boolean_t replacing)955 make_disks(zpool_handle_t *zhp, nvlist_t *nv, boolean_t replacing)
956 {
957 nvlist_t **child;
958 uint_t c, children;
959 const char *type, *path;
960 char devpath[MAXPATHLEN];
961 char udevpath[MAXPATHLEN];
962 uint64_t wholedisk;
963 struct stat64 statbuf;
964 int is_exclusive = 0;
965 int fd;
966 int ret;
967
968 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
969
970 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
971 &child, &children) != 0) {
972
973 if (strcmp(type, VDEV_TYPE_DISK) != 0)
974 return (0);
975
976 /*
977 * We have a disk device. If this is a whole disk write
978 * out the efi partition table, otherwise write zero's to
979 * the first 4k of the partition. This is to ensure that
980 * libblkid will not misidentify the partition due to a
981 * magic value left by the previous filesystem.
982 */
983 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
984 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
985 &wholedisk));
986
987 if (!wholedisk) {
988 /*
989 * Update device id string for mpath nodes (Linux only)
990 */
991 if (is_mpath_whole_disk(path))
992 update_vdev_config_dev_strs(nv);
993
994 if (!is_spare(NULL, path))
995 (void) zero_label(path);
996 return (0);
997 }
998
999 if (realpath(path, devpath) == NULL) {
1000 ret = errno;
1001 (void) fprintf(stderr,
1002 gettext("cannot resolve path '%s'\n"), path);
1003 return (ret);
1004 }
1005
1006 /*
1007 * Remove any previously existing symlink from a udev path to
1008 * the device before labeling the disk. This ensures that
1009 * only newly created links are used. Otherwise there is a
1010 * window between when udev deletes and recreates the link
1011 * during which access attempts will fail with ENOENT.
1012 */
1013 strlcpy(udevpath, path, MAXPATHLEN);
1014 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1015
1016 fd = open(devpath, O_RDWR|O_EXCL);
1017 if (fd == -1) {
1018 if (errno == EBUSY)
1019 is_exclusive = 1;
1020 #ifdef __FreeBSD__
1021 if (errno == EPERM)
1022 is_exclusive = 1;
1023 #endif
1024 } else {
1025 (void) close(fd);
1026 }
1027
1028 /*
1029 * If the partition exists, contains a valid spare label,
1030 * and is opened exclusively there is no need to partition
1031 * it. Hot spares have already been partitioned and are
1032 * held open exclusively by the kernel as a safety measure.
1033 *
1034 * If the provided path is for a /dev/disk/ device its
1035 * symbolic link will be removed, partition table created,
1036 * and then block until udev creates the new link.
1037 */
1038 if (!is_exclusive && !is_spare(NULL, udevpath)) {
1039 char *devnode = strrchr(devpath, '/') + 1;
1040 char **lines = NULL;
1041 int lines_cnt = 0;
1042
1043 ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1044 if (ret == 0) {
1045 ret = lstat64(udevpath, &statbuf);
1046 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1047 (void) unlink(udevpath);
1048 }
1049
1050 /*
1051 * When labeling a pool the raw device node name
1052 * is provided as it appears under /dev/.
1053 *
1054 * Note that 'zhp' will be NULL when we're creating a
1055 * pool.
1056 */
1057 if (zpool_prepare_and_label_disk(g_zfs, zhp, devnode,
1058 nv, zhp == NULL ? "create" :
1059 replacing ? "replace" : "add", &lines,
1060 &lines_cnt) != 0) {
1061 (void) fprintf(stderr,
1062 gettext(
1063 "Error preparing/labeling disk.\n"));
1064 if (lines_cnt > 0) {
1065 (void) fprintf(stderr,
1066 gettext("zfs_prepare_disk output:\n"));
1067 lines_to_stderr(lines, lines_cnt);
1068 }
1069
1070 libzfs_free_str_array(lines, lines_cnt);
1071 return (-1);
1072 }
1073 libzfs_free_str_array(lines, lines_cnt);
1074
1075 /*
1076 * Wait for udev to signal the device is available
1077 * by the provided path.
1078 */
1079 ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1080 if (ret) {
1081 (void) fprintf(stderr,
1082 gettext("missing link: %s was "
1083 "partitioned but %s is missing\n"),
1084 devnode, udevpath);
1085 return (ret);
1086 }
1087
1088 ret = zero_label(udevpath);
1089 if (ret)
1090 return (ret);
1091 }
1092
1093 /*
1094 * Update the path to refer to the partition. The presence of
1095 * the 'whole_disk' field indicates to the CLI that we should
1096 * chop off the partition number when displaying the device in
1097 * future output.
1098 */
1099 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1100
1101 /*
1102 * Update device id strings for whole disks (Linux only)
1103 */
1104 update_vdev_config_dev_strs(nv);
1105
1106 return (0);
1107 }
1108
1109 for (c = 0; c < children; c++)
1110 if ((ret = make_disks(zhp, child[c], replacing)) != 0)
1111 return (ret);
1112
1113 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1114 &child, &children) == 0)
1115 for (c = 0; c < children; c++)
1116 if ((ret = make_disks(zhp, child[c], replacing)) != 0)
1117 return (ret);
1118
1119 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1120 &child, &children) == 0)
1121 for (c = 0; c < children; c++)
1122 if ((ret = make_disks(zhp, child[c], replacing)) != 0)
1123 return (ret);
1124
1125 return (0);
1126 }
1127
1128 /*
1129 * Go through and find any devices that are in use. We rely on libdiskmgt for
1130 * the majority of this task.
1131 */
1132 static boolean_t
is_device_in_use(nvlist_t * config,nvlist_t * nv,boolean_t force,boolean_t replacing,boolean_t isspare)1133 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1134 boolean_t replacing, boolean_t isspare)
1135 {
1136 nvlist_t **child;
1137 uint_t c, children;
1138 const char *type, *path;
1139 int ret = 0;
1140 char buf[MAXPATHLEN];
1141 uint64_t wholedisk = B_FALSE;
1142 boolean_t anyinuse = B_FALSE;
1143
1144 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1145
1146 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1147 &child, &children) != 0) {
1148
1149 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1150 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1151 verify(!nvlist_lookup_uint64(nv,
1152 ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1153
1154 /*
1155 * As a generic check, we look to see if this is a replace of a
1156 * hot spare within the same pool. If so, we allow it
1157 * regardless of what libblkid or zpool_in_use() says.
1158 */
1159 if (replacing) {
1160 (void) strlcpy(buf, path, sizeof (buf));
1161 if (wholedisk) {
1162 ret = zfs_append_partition(buf, sizeof (buf));
1163 if (ret == -1)
1164 return (-1);
1165 }
1166
1167 if (is_spare(config, buf))
1168 return (B_FALSE);
1169 }
1170
1171 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1172 ret = check_device(path, force, isspare, wholedisk);
1173
1174 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1175 ret = check_file(path, force, isspare);
1176
1177 return (ret != 0);
1178 }
1179
1180 for (c = 0; c < children; c++)
1181 if (is_device_in_use(config, child[c], force, replacing,
1182 B_FALSE))
1183 anyinuse = B_TRUE;
1184
1185 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1186 &child, &children) == 0)
1187 for (c = 0; c < children; c++)
1188 if (is_device_in_use(config, child[c], force, replacing,
1189 B_TRUE))
1190 anyinuse = B_TRUE;
1191
1192 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1193 &child, &children) == 0)
1194 for (c = 0; c < children; c++)
1195 if (is_device_in_use(config, child[c], force, replacing,
1196 B_FALSE))
1197 anyinuse = B_TRUE;
1198
1199 return (anyinuse);
1200 }
1201
1202 /*
1203 * Returns the parity level extracted from a raidz or draid type.
1204 * If the parity cannot be determined zero is returned.
1205 */
1206 static int
get_parity(const char * type)1207 get_parity(const char *type)
1208 {
1209 long parity = 0;
1210 const char *p;
1211
1212 if (strncmp(type, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0) {
1213 p = type + strlen(VDEV_TYPE_RAIDZ);
1214
1215 if (*p == '\0') {
1216 /* when unspecified default to single parity */
1217 return (1);
1218 } else if (*p == '0') {
1219 /* no zero prefixes allowed */
1220 return (0);
1221 } else {
1222 /* 0-3, no suffixes allowed */
1223 char *end;
1224 errno = 0;
1225 parity = strtol(p, &end, 10);
1226 if (errno != 0 || *end != '\0' ||
1227 parity < 1 || parity > VDEV_RAIDZ_MAXPARITY) {
1228 return (0);
1229 }
1230 }
1231 } else if (strncmp(type, VDEV_TYPE_DRAID,
1232 strlen(VDEV_TYPE_DRAID)) == 0) {
1233 p = type + strlen(VDEV_TYPE_DRAID);
1234
1235 if (*p == '\0' || *p == ':') {
1236 /* when unspecified default to single parity */
1237 return (1);
1238 } else if (*p == '0') {
1239 /* no zero prefixes allowed */
1240 return (0);
1241 } else {
1242 /* 0-3, allowed suffixes: '\0' or ':' */
1243 char *end;
1244 errno = 0;
1245 parity = strtol(p, &end, 10);
1246 if (errno != 0 ||
1247 parity < 1 || parity > VDEV_DRAID_MAXPARITY ||
1248 (*end != '\0' && *end != ':')) {
1249 return (0);
1250 }
1251 }
1252 }
1253
1254 return ((int)parity);
1255 }
1256
1257 /*
1258 * Assign the minimum and maximum number of devices allowed for
1259 * the specified type. On error NULL is returned, otherwise the
1260 * type prefix is returned (raidz, mirror, etc).
1261 */
1262 static const char *
is_grouping(const char * type,int * mindev,int * maxdev)1263 is_grouping(const char *type, int *mindev, int *maxdev)
1264 {
1265 int nparity;
1266
1267 if (strncmp(type, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1268 strncmp(type, VDEV_TYPE_DRAID, strlen(VDEV_TYPE_DRAID)) == 0) {
1269 nparity = get_parity(type);
1270 if (nparity == 0)
1271 return (NULL);
1272 if (mindev != NULL)
1273 *mindev = nparity + 1;
1274 if (maxdev != NULL)
1275 *maxdev = 255;
1276
1277 if (strncmp(type, VDEV_TYPE_RAIDZ,
1278 strlen(VDEV_TYPE_RAIDZ)) == 0) {
1279 return (VDEV_TYPE_RAIDZ);
1280 } else {
1281 return (VDEV_TYPE_DRAID);
1282 }
1283 }
1284
1285 if (maxdev != NULL)
1286 *maxdev = INT_MAX;
1287
1288 if (strcmp(type, "mirror") == 0) {
1289 if (mindev != NULL)
1290 *mindev = 2;
1291 return (VDEV_TYPE_MIRROR);
1292 }
1293
1294 if (strcmp(type, "spare") == 0) {
1295 if (mindev != NULL)
1296 *mindev = 1;
1297 return (VDEV_TYPE_SPARE);
1298 }
1299
1300 if (strcmp(type, "log") == 0) {
1301 if (mindev != NULL)
1302 *mindev = 1;
1303 return (VDEV_TYPE_LOG);
1304 }
1305
1306 if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0 ||
1307 strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) {
1308 if (mindev != NULL)
1309 *mindev = 1;
1310 return (type);
1311 }
1312
1313 if (strcmp(type, "cache") == 0) {
1314 if (mindev != NULL)
1315 *mindev = 1;
1316 return (VDEV_TYPE_L2CACHE);
1317 }
1318
1319 return (NULL);
1320 }
1321
1322 /*
1323 * Extract the configuration parameters encoded in the dRAID type and
1324 * use them to generate a dRAID configuration. The expected format is:
1325 *
1326 * draid[<parity>][:<data><d|D>][:<children><c|C>][:<spares><s|S>]
1327 *
1328 * The intent is to be able to generate a good configuration when no
1329 * additional information is provided. The only mandatory component
1330 * of the 'type' is the 'draid' prefix. If a value is not provided
1331 * then reasonable defaults are used. The optional components may
1332 * appear in any order but the d/s/c suffix is required.
1333 *
1334 * Valid inputs:
1335 * - data: number of data devices per group (1-255)
1336 * - parity: number of parity blocks per group (1-3)
1337 * - spares: number of distributed spare (0-100)
1338 * - children: total number of devices (1-255)
1339 *
1340 * Examples:
1341 * - zpool create tank draid <devices...>
1342 * - zpool create tank draid2:8d:51c:2s <devices...>
1343 */
1344 static int
draid_config_by_type(nvlist_t * nv,const char * type,uint64_t children)1345 draid_config_by_type(nvlist_t *nv, const char *type, uint64_t children)
1346 {
1347 uint64_t nparity;
1348 uint64_t nspares = 0;
1349 uint64_t ndata = UINT64_MAX;
1350 uint64_t ngroups = 1;
1351 long value;
1352
1353 if (strncmp(type, VDEV_TYPE_DRAID, strlen(VDEV_TYPE_DRAID)) != 0)
1354 return (EINVAL);
1355
1356 nparity = (uint64_t)get_parity(type);
1357 if (nparity == 0 || nparity > VDEV_DRAID_MAXPARITY) {
1358 fprintf(stderr,
1359 gettext("invalid dRAID parity level %llu; must be "
1360 "between 1 and %d\n"), (u_longlong_t)nparity,
1361 VDEV_DRAID_MAXPARITY);
1362 return (EINVAL);
1363 }
1364
1365 char *p = (char *)type;
1366 while ((p = strchr(p, ':')) != NULL) {
1367 char *end;
1368
1369 p = p + 1;
1370 errno = 0;
1371
1372 if (!isdigit(p[0])) {
1373 (void) fprintf(stderr, gettext("invalid dRAID "
1374 "syntax; expected [:<number><c|d|s>] not '%s'\n"),
1375 type);
1376 return (EINVAL);
1377 }
1378
1379 /* Expected non-zero value with c/d/s suffix */
1380 value = strtol(p, &end, 10);
1381 char suffix = tolower(*end);
1382 if (errno != 0 ||
1383 (suffix != 'c' && suffix != 'd' && suffix != 's')) {
1384 (void) fprintf(stderr, gettext("invalid dRAID "
1385 "syntax; expected [:<number><c|d|s>] not '%s'\n"),
1386 type);
1387 return (EINVAL);
1388 }
1389
1390 if (suffix == 'c') {
1391 if ((uint64_t)value != children) {
1392 fprintf(stderr,
1393 gettext("invalid number of dRAID children; "
1394 "%llu required but %llu provided\n"),
1395 (u_longlong_t)value,
1396 (u_longlong_t)children);
1397 return (EINVAL);
1398 }
1399 } else if (suffix == 'd') {
1400 ndata = (uint64_t)value;
1401 } else if (suffix == 's') {
1402 nspares = (uint64_t)value;
1403 } else {
1404 verify(0); /* Unreachable */
1405 }
1406 }
1407
1408 /*
1409 * When a specific number of data disks is not provided limit a
1410 * redundancy group to 8 data disks. This value was selected to
1411 * provide a reasonable tradeoff between capacity and performance.
1412 */
1413 if (ndata == UINT64_MAX) {
1414 if (children > nspares + nparity) {
1415 ndata = MIN(children - nspares - nparity, 8);
1416 } else {
1417 fprintf(stderr, gettext("request number of "
1418 "distributed spares %llu and parity level %llu\n"
1419 "leaves no disks available for data\n"),
1420 (u_longlong_t)nspares, (u_longlong_t)nparity);
1421 return (EINVAL);
1422 }
1423 }
1424
1425 /* Verify the maximum allowed group size is never exceeded. */
1426 if (ndata == 0 || (ndata + nparity > children - nspares)) {
1427 fprintf(stderr, gettext("requested number of dRAID data "
1428 "disks per group %llu is too high,\nat most %llu disks "
1429 "are available for data\n"), (u_longlong_t)ndata,
1430 (u_longlong_t)(children - nspares - nparity));
1431 return (EINVAL);
1432 }
1433
1434 /*
1435 * Verify the requested number of spares can be satisfied.
1436 * An arbitrary limit of 100 distributed spares is applied.
1437 */
1438 if (nspares > 100 || nspares > (children - (ndata + nparity))) {
1439 fprintf(stderr,
1440 gettext("invalid number of dRAID spares %llu; additional "
1441 "disks would be required\n"), (u_longlong_t)nspares);
1442 return (EINVAL);
1443 }
1444
1445 /* Verify the requested number children is sufficient. */
1446 if (children < (ndata + nparity + nspares)) {
1447 fprintf(stderr, gettext("%llu disks were provided, but at "
1448 "least %llu disks are required for this config\n"),
1449 (u_longlong_t)children,
1450 (u_longlong_t)(ndata + nparity + nspares));
1451 }
1452
1453 if (children > VDEV_DRAID_MAX_CHILDREN) {
1454 fprintf(stderr, gettext("%llu disks were provided, but "
1455 "dRAID only supports up to %u disks"),
1456 (u_longlong_t)children, VDEV_DRAID_MAX_CHILDREN);
1457 }
1458
1459 /*
1460 * Calculate the minimum number of groups required to fill a slice.
1461 * This is the LCM of the stripe width (ndata + nparity) and the
1462 * number of data drives (children - nspares).
1463 */
1464 while (ngroups * (ndata + nparity) % (children - nspares) != 0)
1465 ngroups++;
1466
1467 /* Store the basic dRAID configuration. */
1468 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, nparity);
1469 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NDATA, ndata);
1470 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NSPARES, nspares);
1471 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DRAID_NGROUPS, ngroups);
1472
1473 return (0);
1474 }
1475
1476 /*
1477 * Construct a syntactically valid vdev specification,
1478 * and ensure that all devices and files exist and can be opened.
1479 * Note: we don't bother freeing anything in the error paths
1480 * because the program is just going to exit anyway.
1481 */
1482 static nvlist_t *
construct_spec(nvlist_t * props,int argc,char ** argv)1483 construct_spec(nvlist_t *props, int argc, char **argv)
1484 {
1485 nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1486 int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1487 const char *type, *fulltype;
1488 boolean_t is_log, is_special, is_dedup, is_spare;
1489 boolean_t seen_logs;
1490 uint64_t ashift = 0;
1491
1492 if (props != NULL) {
1493 const char *value = NULL;
1494
1495 if (nvlist_lookup_string(props,
1496 zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0) {
1497 if (zfs_nicestrtonum(NULL, value, &ashift) != 0) {
1498 (void) fprintf(stderr,
1499 gettext("ashift must be a number.\n"));
1500 return (NULL);
1501 }
1502 if (ashift != 0 &&
1503 (ashift < ASHIFT_MIN || ashift > ASHIFT_MAX)) {
1504 (void) fprintf(stderr,
1505 gettext("invalid 'ashift=%" PRIu64 "' "
1506 "property: only values between %" PRId32 " "
1507 "and %" PRId32 " are allowed.\n"),
1508 ashift, ASHIFT_MIN, ASHIFT_MAX);
1509 return (NULL);
1510 }
1511 }
1512 }
1513
1514 top = NULL;
1515 toplevels = 0;
1516 spares = NULL;
1517 l2cache = NULL;
1518 nspares = 0;
1519 nlogs = 0;
1520 nl2cache = 0;
1521 is_log = is_special = is_dedup = is_spare = B_FALSE;
1522 seen_logs = B_FALSE;
1523 nvroot = NULL;
1524
1525 while (argc > 0) {
1526 fulltype = argv[0];
1527 nv = NULL;
1528
1529 /*
1530 * If it's a mirror, raidz, or draid the subsequent arguments
1531 * are its leaves -- until we encounter the next mirror,
1532 * raidz or draid.
1533 */
1534 if ((type = is_grouping(fulltype, &mindev, &maxdev)) != NULL) {
1535 nvlist_t **child = NULL;
1536 int c, children = 0;
1537
1538 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1539 if (spares != NULL) {
1540 (void) fprintf(stderr,
1541 gettext("invalid vdev "
1542 "specification: 'spare' can be "
1543 "specified only once\n"));
1544 goto spec_out;
1545 }
1546 is_spare = B_TRUE;
1547 is_log = is_special = is_dedup = B_FALSE;
1548 }
1549
1550 if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1551 if (seen_logs) {
1552 (void) fprintf(stderr,
1553 gettext("invalid vdev "
1554 "specification: 'log' can be "
1555 "specified only once\n"));
1556 goto spec_out;
1557 }
1558 seen_logs = B_TRUE;
1559 is_log = B_TRUE;
1560 is_special = is_dedup = is_spare = B_FALSE;
1561 argc--;
1562 argv++;
1563 /*
1564 * A log is not a real grouping device.
1565 * We just set is_log and continue.
1566 */
1567 continue;
1568 }
1569
1570 if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0) {
1571 is_special = B_TRUE;
1572 is_log = is_dedup = is_spare = B_FALSE;
1573 argc--;
1574 argv++;
1575 continue;
1576 }
1577
1578 if (strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) {
1579 is_dedup = B_TRUE;
1580 is_log = is_special = is_spare = B_FALSE;
1581 argc--;
1582 argv++;
1583 continue;
1584 }
1585
1586 if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1587 if (l2cache != NULL) {
1588 (void) fprintf(stderr,
1589 gettext("invalid vdev "
1590 "specification: 'cache' can be "
1591 "specified only once\n"));
1592 goto spec_out;
1593 }
1594 is_log = is_special = B_FALSE;
1595 is_dedup = is_spare = B_FALSE;
1596 }
1597
1598 if (is_log) {
1599 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1600 (void) fprintf(stderr,
1601 gettext("invalid vdev "
1602 "specification: unsupported 'log' "
1603 "device: %s\n"), type);
1604 goto spec_out;
1605 }
1606 nlogs++;
1607 }
1608
1609 for (c = 1; c < argc; c++) {
1610 if (is_grouping(argv[c], NULL, NULL) != NULL)
1611 break;
1612
1613 children++;
1614 child = realloc(child,
1615 children * sizeof (nvlist_t *));
1616 if (child == NULL)
1617 zpool_no_memory();
1618 if ((nv = make_leaf_vdev(argv[c],
1619 !(is_log || is_special || is_dedup ||
1620 is_spare), ashift)) == NULL) {
1621 for (c = 0; c < children - 1; c++)
1622 nvlist_free(child[c]);
1623 free(child);
1624 goto spec_out;
1625 }
1626
1627 child[children - 1] = nv;
1628 }
1629
1630 if (children < mindev) {
1631 (void) fprintf(stderr, gettext("invalid vdev "
1632 "specification: %s requires at least %d "
1633 "devices\n"), argv[0], mindev);
1634 for (c = 0; c < children; c++)
1635 nvlist_free(child[c]);
1636 free(child);
1637 goto spec_out;
1638 }
1639
1640 if (children > maxdev) {
1641 (void) fprintf(stderr, gettext("invalid vdev "
1642 "specification: %s supports no more than "
1643 "%d devices\n"), argv[0], maxdev);
1644 for (c = 0; c < children; c++)
1645 nvlist_free(child[c]);
1646 free(child);
1647 goto spec_out;
1648 }
1649
1650 argc -= c;
1651 argv += c;
1652
1653 if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1654 spares = child;
1655 nspares = children;
1656 continue;
1657 } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1658 l2cache = child;
1659 nl2cache = children;
1660 continue;
1661 } else {
1662 /* create a top-level vdev with children */
1663 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1664 0) == 0);
1665 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1666 type) == 0);
1667 verify(nvlist_add_uint64(nv,
1668 ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1669 if (is_log) {
1670 verify(nvlist_add_string(nv,
1671 ZPOOL_CONFIG_ALLOCATION_BIAS,
1672 VDEV_ALLOC_BIAS_LOG) == 0);
1673 }
1674 if (is_special) {
1675 verify(nvlist_add_string(nv,
1676 ZPOOL_CONFIG_ALLOCATION_BIAS,
1677 VDEV_ALLOC_BIAS_SPECIAL) == 0);
1678 }
1679 if (is_dedup) {
1680 verify(nvlist_add_string(nv,
1681 ZPOOL_CONFIG_ALLOCATION_BIAS,
1682 VDEV_ALLOC_BIAS_DEDUP) == 0);
1683 }
1684 if (ashift > 0) {
1685 fnvlist_add_uint64(nv,
1686 ZPOOL_CONFIG_ASHIFT, ashift);
1687 }
1688 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1689 verify(nvlist_add_uint64(nv,
1690 ZPOOL_CONFIG_NPARITY,
1691 mindev - 1) == 0);
1692 }
1693 if (strcmp(type, VDEV_TYPE_DRAID) == 0) {
1694 if (draid_config_by_type(nv,
1695 fulltype, children) != 0) {
1696 for (c = 0; c < children; c++)
1697 nvlist_free(child[c]);
1698 free(child);
1699 goto spec_out;
1700 }
1701 }
1702 verify(nvlist_add_nvlist_array(nv,
1703 ZPOOL_CONFIG_CHILDREN,
1704 (const nvlist_t **)child, children) == 0);
1705
1706 for (c = 0; c < children; c++)
1707 nvlist_free(child[c]);
1708 free(child);
1709 }
1710 } else {
1711 /*
1712 * We have a device. Pass off to make_leaf_vdev() to
1713 * construct the appropriate nvlist describing the vdev.
1714 */
1715 if ((nv = make_leaf_vdev(argv[0], !(is_log ||
1716 is_special || is_dedup || is_spare),
1717 ashift)) == NULL)
1718 goto spec_out;
1719
1720 verify(nvlist_add_uint64(nv,
1721 ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1722 if (is_log) {
1723 verify(nvlist_add_string(nv,
1724 ZPOOL_CONFIG_ALLOCATION_BIAS,
1725 VDEV_ALLOC_BIAS_LOG) == 0);
1726 nlogs++;
1727 }
1728
1729 if (is_special) {
1730 verify(nvlist_add_string(nv,
1731 ZPOOL_CONFIG_ALLOCATION_BIAS,
1732 VDEV_ALLOC_BIAS_SPECIAL) == 0);
1733 }
1734 if (is_dedup) {
1735 verify(nvlist_add_string(nv,
1736 ZPOOL_CONFIG_ALLOCATION_BIAS,
1737 VDEV_ALLOC_BIAS_DEDUP) == 0);
1738 }
1739 argc--;
1740 argv++;
1741 }
1742
1743 toplevels++;
1744 top = realloc(top, toplevels * sizeof (nvlist_t *));
1745 if (top == NULL)
1746 zpool_no_memory();
1747 top[toplevels - 1] = nv;
1748 }
1749
1750 if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1751 (void) fprintf(stderr, gettext("invalid vdev "
1752 "specification: at least one toplevel vdev must be "
1753 "specified\n"));
1754 goto spec_out;
1755 }
1756
1757 if (seen_logs && nlogs == 0) {
1758 (void) fprintf(stderr, gettext("invalid vdev specification: "
1759 "log requires at least 1 device\n"));
1760 goto spec_out;
1761 }
1762
1763 /*
1764 * Finally, create nvroot and add all top-level vdevs to it.
1765 */
1766 verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1767 verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1768 VDEV_TYPE_ROOT) == 0);
1769 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1770 (const nvlist_t **)top, toplevels) == 0);
1771 if (nspares != 0)
1772 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1773 (const nvlist_t **)spares, nspares) == 0);
1774 if (nl2cache != 0)
1775 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1776 (const nvlist_t **)l2cache, nl2cache) == 0);
1777
1778 spec_out:
1779 for (t = 0; t < toplevels; t++)
1780 nvlist_free(top[t]);
1781 for (t = 0; t < nspares; t++)
1782 nvlist_free(spares[t]);
1783 for (t = 0; t < nl2cache; t++)
1784 nvlist_free(l2cache[t]);
1785
1786 free(spares);
1787 free(l2cache);
1788 free(top);
1789
1790 return (nvroot);
1791 }
1792
1793 nvlist_t *
split_mirror_vdev(zpool_handle_t * zhp,char * newname,nvlist_t * props,splitflags_t flags,int argc,char ** argv)1794 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1795 splitflags_t flags, int argc, char **argv)
1796 {
1797 nvlist_t *newroot = NULL, **child;
1798 uint_t c, children;
1799
1800 if (argc > 0) {
1801 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1802 (void) fprintf(stderr, gettext("Unable to build a "
1803 "pool from the specified devices\n"));
1804 return (NULL);
1805 }
1806
1807 if (!flags.dryrun && make_disks(zhp, newroot, B_FALSE) != 0) {
1808 nvlist_free(newroot);
1809 return (NULL);
1810 }
1811
1812 /* avoid any tricks in the spec */
1813 verify(nvlist_lookup_nvlist_array(newroot,
1814 ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1815 for (c = 0; c < children; c++) {
1816 const char *path;
1817 const char *type;
1818 int min, max;
1819
1820 verify(nvlist_lookup_string(child[c],
1821 ZPOOL_CONFIG_PATH, &path) == 0);
1822 if ((type = is_grouping(path, &min, &max)) != NULL) {
1823 (void) fprintf(stderr, gettext("Cannot use "
1824 "'%s' as a device for splitting\n"), type);
1825 nvlist_free(newroot);
1826 return (NULL);
1827 }
1828 }
1829 }
1830
1831 if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1832 nvlist_free(newroot);
1833 return (NULL);
1834 }
1835
1836 return (newroot);
1837 }
1838
1839 static int
num_normal_vdevs(nvlist_t * nvroot)1840 num_normal_vdevs(nvlist_t *nvroot)
1841 {
1842 nvlist_t **top;
1843 uint_t t, toplevels, normal = 0;
1844
1845 verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1846 &top, &toplevels) == 0);
1847
1848 for (t = 0; t < toplevels; t++) {
1849 uint64_t log = B_FALSE;
1850
1851 (void) nvlist_lookup_uint64(top[t], ZPOOL_CONFIG_IS_LOG, &log);
1852 if (log)
1853 continue;
1854 if (nvlist_exists(top[t], ZPOOL_CONFIG_ALLOCATION_BIAS))
1855 continue;
1856
1857 normal++;
1858 }
1859
1860 return (normal);
1861 }
1862
1863 /*
1864 * Get and validate the contents of the given vdev specification. This ensures
1865 * that the nvlist returned is well-formed, that all the devices exist, and that
1866 * they are not currently in use by any other known consumer. The 'poolconfig'
1867 * parameter is the current configuration of the pool when adding devices
1868 * existing pool, and is used to perform additional checks, such as changing the
1869 * replication level of the pool. It can be 'NULL' to indicate that this is a
1870 * new pool. The 'force' flag controls whether devices should be forcefully
1871 * added, even if they appear in use.
1872 */
1873 nvlist_t *
make_root_vdev(zpool_handle_t * zhp,nvlist_t * props,int force,int check_rep,boolean_t replacing,boolean_t dryrun,int argc,char ** argv)1874 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1875 boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1876 {
1877 nvlist_t *newroot;
1878 nvlist_t *poolconfig = NULL;
1879 is_force = force;
1880
1881 /*
1882 * Construct the vdev specification. If this is successful, we know
1883 * that we have a valid specification, and that all devices can be
1884 * opened.
1885 */
1886 if ((newroot = construct_spec(props, argc, argv)) == NULL)
1887 return (NULL);
1888
1889 if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1890 nvlist_free(newroot);
1891 return (NULL);
1892 }
1893
1894 /*
1895 * Validate each device to make sure that it's not shared with another
1896 * subsystem. We do this even if 'force' is set, because there are some
1897 * uses (such as a dedicated dump device) that even '-f' cannot
1898 * override.
1899 */
1900 if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1901 nvlist_free(newroot);
1902 return (NULL);
1903 }
1904
1905 /*
1906 * Check the replication level of the given vdevs and report any errors
1907 * found. We include the existing pool spec, if any, as we need to
1908 * catch changes against the existing replication level.
1909 */
1910 if (check_rep && check_replication(poolconfig, newroot) != 0) {
1911 nvlist_free(newroot);
1912 return (NULL);
1913 }
1914
1915 /*
1916 * On pool create the new vdev spec must have one normal vdev.
1917 */
1918 if (poolconfig == NULL && num_normal_vdevs(newroot) == 0) {
1919 vdev_error(gettext("at least one general top-level vdev must "
1920 "be specified\n"));
1921 nvlist_free(newroot);
1922 return (NULL);
1923 }
1924
1925 /*
1926 * Run through the vdev specification and label any whole disks found.
1927 */
1928 if (!dryrun && make_disks(zhp, newroot, replacing) != 0) {
1929 nvlist_free(newroot);
1930 return (NULL);
1931 }
1932
1933 return (newroot);
1934 }
1935