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) 2011, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 */
26
27 /*
28 * zhack is a debugging tool that can write changes to ZFS pool using libzpool
29 * for testing purposes. Altering pools with zhack is unsupported and may
30 * result in corrupted pools.
31 */
32
33 #include <zfs_prop.h>
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <ctype.h>
37 #include <sys/stat.h>
38 #include <sys/zfs_context.h>
39 #include <sys/spa.h>
40 #include <sys/spa_impl.h>
41 #include <sys/dmu.h>
42 #include <sys/zap.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/dsl_synctask.h>
45 #include <sys/vdev.h>
46 #include <sys/vdev_impl.h>
47 #include <sys/fs/zfs.h>
48 #include <sys/dmu_objset.h>
49 #include <sys/dsl_pool.h>
50 #include <sys/zio_checksum.h>
51 #include <sys/zio_compress.h>
52 #include <sys/zfeature.h>
53 #include <sys/dmu_tx.h>
54 #include <zfeature_common.h>
55 #include <libzutil.h>
56
57 static importargs_t g_importargs;
58 static char *g_pool;
59 static boolean_t g_readonly;
60
61 typedef enum {
62 ZHACK_REPAIR_OP_UNKNOWN = 0,
63 ZHACK_REPAIR_OP_CKSUM = (1 << 0),
64 ZHACK_REPAIR_OP_UNDETACH = (1 << 1)
65 } zhack_repair_op_t;
66
67 static __attribute__((noreturn)) void
usage(void)68 usage(void)
69 {
70 (void) fprintf(stderr,
71 "Usage: zhack [-c cachefile] [-d dir] <subcommand> <args> ...\n"
72 "where <subcommand> <args> is one of the following:\n"
73 "\n");
74
75 (void) fprintf(stderr,
76 " feature stat <pool>\n"
77 " print information about enabled features\n"
78 " feature enable [-r] [-d desc] <pool> <feature>\n"
79 " add a new enabled feature to the pool\n"
80 " -d <desc> sets the feature's description\n"
81 " -r set read-only compatible flag for feature\n"
82 " feature ref [-md] <pool> <feature>\n"
83 " change the refcount on the given feature\n"
84 " -d decrease instead of increase the refcount\n"
85 " -m add the feature to the label if increasing refcount\n"
86 "\n"
87 " <feature> : should be a feature guid\n"
88 "\n"
89 " label repair <device>\n"
90 " repair labels of a specified device according to options\n"
91 " which may be combined to do their functions in one call\n"
92 " -c repair corrupted label checksums\n"
93 " -u restore the label on a detached device\n"
94 "\n"
95 " <device> : path to vdev\n");
96 exit(1);
97 }
98
99
100 static __attribute__((format(printf, 3, 4))) __attribute__((noreturn)) void
fatal(spa_t * spa,const void * tag,const char * fmt,...)101 fatal(spa_t *spa, const void *tag, const char *fmt, ...)
102 {
103 va_list ap;
104
105 if (spa != NULL) {
106 spa_close(spa, tag);
107 (void) spa_export(g_pool, NULL, B_TRUE, B_FALSE);
108 }
109
110 va_start(ap, fmt);
111 (void) fputs("zhack: ", stderr);
112 (void) vfprintf(stderr, fmt, ap);
113 va_end(ap);
114 (void) fputc('\n', stderr);
115
116 exit(1);
117 }
118
119 static int
space_delta_cb(dmu_object_type_t bonustype,const void * data,zfs_file_info_t * zoi)120 space_delta_cb(dmu_object_type_t bonustype, const void *data,
121 zfs_file_info_t *zoi)
122 {
123 (void) data, (void) zoi;
124
125 /*
126 * Is it a valid type of object to track?
127 */
128 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
129 return (ENOENT);
130 (void) fprintf(stderr, "modifying object that needs user accounting");
131 abort();
132 }
133
134 /*
135 * Target is the dataset whose pool we want to open.
136 */
137 static void
zhack_import(char * target,boolean_t readonly)138 zhack_import(char *target, boolean_t readonly)
139 {
140 nvlist_t *config;
141 nvlist_t *props;
142 int error;
143
144 kernel_init(readonly ? SPA_MODE_READ :
145 (SPA_MODE_READ | SPA_MODE_WRITE));
146
147 dmu_objset_register_type(DMU_OST_ZFS, space_delta_cb);
148
149 g_readonly = readonly;
150 g_importargs.can_be_active = readonly;
151 g_pool = strdup(target);
152
153 libpc_handle_t lpch = {
154 .lpc_lib_handle = NULL,
155 .lpc_ops = &libzpool_config_ops,
156 .lpc_printerr = B_TRUE
157 };
158 error = zpool_find_config(&lpch, target, &config, &g_importargs);
159 if (error)
160 fatal(NULL, FTAG, "cannot import '%s'", target);
161
162 props = NULL;
163 if (readonly) {
164 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
165 VERIFY(nvlist_add_uint64(props,
166 zpool_prop_to_name(ZPOOL_PROP_READONLY), 1) == 0);
167 }
168
169 zfeature_checks_disable = B_TRUE;
170 error = spa_import(target, config, props,
171 (readonly ? ZFS_IMPORT_SKIP_MMP : ZFS_IMPORT_NORMAL));
172 fnvlist_free(config);
173 zfeature_checks_disable = B_FALSE;
174 if (error == EEXIST)
175 error = 0;
176
177 if (error)
178 fatal(NULL, FTAG, "can't import '%s': %s", target,
179 strerror(error));
180 }
181
182 static void
zhack_spa_open(char * target,boolean_t readonly,const void * tag,spa_t ** spa)183 zhack_spa_open(char *target, boolean_t readonly, const void *tag, spa_t **spa)
184 {
185 int err;
186
187 zhack_import(target, readonly);
188
189 zfeature_checks_disable = B_TRUE;
190 err = spa_open(target, spa, tag);
191 zfeature_checks_disable = B_FALSE;
192
193 if (err != 0)
194 fatal(*spa, FTAG, "cannot open '%s': %s", target,
195 strerror(err));
196 if (spa_version(*spa) < SPA_VERSION_FEATURES) {
197 fatal(*spa, FTAG, "'%s' has version %d, features not enabled",
198 target, (int)spa_version(*spa));
199 }
200 }
201
202 static void
dump_obj(objset_t * os,uint64_t obj,const char * name)203 dump_obj(objset_t *os, uint64_t obj, const char *name)
204 {
205 zap_cursor_t zc;
206 zap_attribute_t *za = zap_attribute_long_alloc();
207
208 (void) printf("%s_obj:\n", name);
209
210 for (zap_cursor_init(&zc, os, obj);
211 zap_cursor_retrieve(&zc, za) == 0;
212 zap_cursor_advance(&zc)) {
213 if (za->za_integer_length == 8) {
214 ASSERT(za->za_num_integers == 1);
215 (void) printf("\t%s = %llu\n",
216 za->za_name, (u_longlong_t)za->za_first_integer);
217 } else {
218 ASSERT(za->za_integer_length == 1);
219 char val[1024];
220 VERIFY(zap_lookup(os, obj, za->za_name,
221 1, sizeof (val), val) == 0);
222 (void) printf("\t%s = %s\n", za->za_name, val);
223 }
224 }
225 zap_cursor_fini(&zc);
226 zap_attribute_free(za);
227 }
228
229 static void
dump_mos(spa_t * spa)230 dump_mos(spa_t *spa)
231 {
232 nvlist_t *nv = spa->spa_label_features;
233 nvpair_t *pair;
234
235 (void) printf("label config:\n");
236 for (pair = nvlist_next_nvpair(nv, NULL);
237 pair != NULL;
238 pair = nvlist_next_nvpair(nv, pair)) {
239 (void) printf("\t%s\n", nvpair_name(pair));
240 }
241 }
242
243 static void
zhack_do_feature_stat(int argc,char ** argv)244 zhack_do_feature_stat(int argc, char **argv)
245 {
246 spa_t *spa;
247 objset_t *os;
248 char *target;
249
250 argc--;
251 argv++;
252
253 if (argc < 1) {
254 (void) fprintf(stderr, "error: missing pool name\n");
255 usage();
256 }
257 target = argv[0];
258
259 zhack_spa_open(target, B_TRUE, FTAG, &spa);
260 os = spa->spa_meta_objset;
261
262 dump_obj(os, spa->spa_feat_for_read_obj, "for_read");
263 dump_obj(os, spa->spa_feat_for_write_obj, "for_write");
264 dump_obj(os, spa->spa_feat_desc_obj, "descriptions");
265 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
266 dump_obj(os, spa->spa_feat_enabled_txg_obj, "enabled_txg");
267 }
268 dump_mos(spa);
269
270 spa_close(spa, FTAG);
271 }
272
273 static void
zhack_feature_enable_sync(void * arg,dmu_tx_t * tx)274 zhack_feature_enable_sync(void *arg, dmu_tx_t *tx)
275 {
276 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
277 zfeature_info_t *feature = arg;
278
279 feature_enable_sync(spa, feature, tx);
280
281 spa_history_log_internal(spa, "zhack enable feature", tx,
282 "name=%s flags=%u",
283 feature->fi_guid, feature->fi_flags);
284 }
285
286 static void
zhack_do_feature_enable(int argc,char ** argv)287 zhack_do_feature_enable(int argc, char **argv)
288 {
289 int c;
290 char *desc, *target;
291 spa_t *spa;
292 objset_t *mos;
293 zfeature_info_t feature;
294 const spa_feature_t nodeps[] = { SPA_FEATURE_NONE };
295
296 /*
297 * Features are not added to the pool's label until their refcounts
298 * are incremented, so fi_mos can just be left as false for now.
299 */
300 desc = NULL;
301 feature.fi_uname = "zhack";
302 feature.fi_flags = 0;
303 feature.fi_depends = nodeps;
304 feature.fi_feature = SPA_FEATURE_NONE;
305
306 optind = 1;
307 while ((c = getopt(argc, argv, "+rd:")) != -1) {
308 switch (c) {
309 case 'r':
310 feature.fi_flags |= ZFEATURE_FLAG_READONLY_COMPAT;
311 break;
312 case 'd':
313 if (desc != NULL)
314 free(desc);
315 desc = strdup(optarg);
316 break;
317 default:
318 usage();
319 break;
320 }
321 }
322
323 if (desc == NULL)
324 desc = strdup("zhack injected");
325 feature.fi_desc = desc;
326
327 argc -= optind;
328 argv += optind;
329
330 if (argc < 2) {
331 (void) fprintf(stderr, "error: missing feature or pool name\n");
332 usage();
333 }
334 target = argv[0];
335 feature.fi_guid = argv[1];
336
337 if (!zfeature_is_valid_guid(feature.fi_guid))
338 fatal(NULL, FTAG, "invalid feature guid: %s", feature.fi_guid);
339
340 zhack_spa_open(target, B_FALSE, FTAG, &spa);
341 mos = spa->spa_meta_objset;
342
343 if (zfeature_is_supported(feature.fi_guid))
344 fatal(spa, FTAG, "'%s' is a real feature, will not enable",
345 feature.fi_guid);
346 if (0 == zap_contains(mos, spa->spa_feat_desc_obj, feature.fi_guid))
347 fatal(spa, FTAG, "feature already enabled: %s",
348 feature.fi_guid);
349
350 VERIFY0(dsl_sync_task(spa_name(spa), NULL,
351 zhack_feature_enable_sync, &feature, 5, ZFS_SPACE_CHECK_NORMAL));
352
353 spa_close(spa, FTAG);
354
355 free(desc);
356 }
357
358 static void
feature_incr_sync(void * arg,dmu_tx_t * tx)359 feature_incr_sync(void *arg, dmu_tx_t *tx)
360 {
361 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
362 zfeature_info_t *feature = arg;
363 uint64_t refcount;
364
365 VERIFY0(feature_get_refcount_from_disk(spa, feature, &refcount));
366 feature_sync(spa, feature, refcount + 1, tx);
367 spa_history_log_internal(spa, "zhack feature incr", tx,
368 "name=%s", feature->fi_guid);
369 }
370
371 static void
feature_decr_sync(void * arg,dmu_tx_t * tx)372 feature_decr_sync(void *arg, dmu_tx_t *tx)
373 {
374 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
375 zfeature_info_t *feature = arg;
376 uint64_t refcount;
377
378 VERIFY0(feature_get_refcount_from_disk(spa, feature, &refcount));
379 feature_sync(spa, feature, refcount - 1, tx);
380 spa_history_log_internal(spa, "zhack feature decr", tx,
381 "name=%s", feature->fi_guid);
382 }
383
384 static void
zhack_do_feature_ref(int argc,char ** argv)385 zhack_do_feature_ref(int argc, char **argv)
386 {
387 int c;
388 char *target;
389 boolean_t decr = B_FALSE;
390 spa_t *spa;
391 objset_t *mos;
392 zfeature_info_t feature;
393 const spa_feature_t nodeps[] = { SPA_FEATURE_NONE };
394
395 /*
396 * fi_desc does not matter here because it was written to disk
397 * when the feature was enabled, but we need to properly set the
398 * feature for read or write based on the information we read off
399 * disk later.
400 */
401 feature.fi_uname = "zhack";
402 feature.fi_flags = 0;
403 feature.fi_desc = NULL;
404 feature.fi_depends = nodeps;
405 feature.fi_feature = SPA_FEATURE_NONE;
406
407 optind = 1;
408 while ((c = getopt(argc, argv, "+md")) != -1) {
409 switch (c) {
410 case 'm':
411 feature.fi_flags |= ZFEATURE_FLAG_MOS;
412 break;
413 case 'd':
414 decr = B_TRUE;
415 break;
416 default:
417 usage();
418 break;
419 }
420 }
421 argc -= optind;
422 argv += optind;
423
424 if (argc < 2) {
425 (void) fprintf(stderr, "error: missing feature or pool name\n");
426 usage();
427 }
428 target = argv[0];
429 feature.fi_guid = argv[1];
430
431 if (!zfeature_is_valid_guid(feature.fi_guid))
432 fatal(NULL, FTAG, "invalid feature guid: %s", feature.fi_guid);
433
434 zhack_spa_open(target, B_FALSE, FTAG, &spa);
435 mos = spa->spa_meta_objset;
436
437 if (zfeature_is_supported(feature.fi_guid)) {
438 fatal(spa, FTAG,
439 "'%s' is a real feature, will not change refcount",
440 feature.fi_guid);
441 }
442
443 if (0 == zap_contains(mos, spa->spa_feat_for_read_obj,
444 feature.fi_guid)) {
445 feature.fi_flags &= ~ZFEATURE_FLAG_READONLY_COMPAT;
446 } else if (0 == zap_contains(mos, spa->spa_feat_for_write_obj,
447 feature.fi_guid)) {
448 feature.fi_flags |= ZFEATURE_FLAG_READONLY_COMPAT;
449 } else {
450 fatal(spa, FTAG, "feature is not enabled: %s", feature.fi_guid);
451 }
452
453 if (decr) {
454 uint64_t count;
455 if (feature_get_refcount_from_disk(spa, &feature,
456 &count) == 0 && count == 0) {
457 fatal(spa, FTAG, "feature refcount already 0: %s",
458 feature.fi_guid);
459 }
460 }
461
462 VERIFY0(dsl_sync_task(spa_name(spa), NULL,
463 decr ? feature_decr_sync : feature_incr_sync, &feature,
464 5, ZFS_SPACE_CHECK_NORMAL));
465
466 spa_close(spa, FTAG);
467 }
468
469 static int
zhack_do_feature(int argc,char ** argv)470 zhack_do_feature(int argc, char **argv)
471 {
472 char *subcommand;
473
474 argc--;
475 argv++;
476 if (argc == 0) {
477 (void) fprintf(stderr,
478 "error: no feature operation specified\n");
479 usage();
480 }
481
482 subcommand = argv[0];
483 if (strcmp(subcommand, "stat") == 0) {
484 zhack_do_feature_stat(argc, argv);
485 } else if (strcmp(subcommand, "enable") == 0) {
486 zhack_do_feature_enable(argc, argv);
487 } else if (strcmp(subcommand, "ref") == 0) {
488 zhack_do_feature_ref(argc, argv);
489 } else {
490 (void) fprintf(stderr, "error: unknown subcommand: %s\n",
491 subcommand);
492 usage();
493 }
494
495 return (0);
496 }
497
498 #define ASHIFT_UBERBLOCK_SHIFT(ashift) \
499 MIN(MAX(ashift, UBERBLOCK_SHIFT), \
500 MAX_UBERBLOCK_SHIFT)
501 #define ASHIFT_UBERBLOCK_SIZE(ashift) \
502 (1ULL << ASHIFT_UBERBLOCK_SHIFT(ashift))
503
504 #define REPAIR_LABEL_STATUS_CKSUM (1 << 0)
505 #define REPAIR_LABEL_STATUS_UB (1 << 1)
506
507 static int
zhack_repair_read_label(const int fd,vdev_label_t * vl,const uint64_t label_offset,const int l)508 zhack_repair_read_label(const int fd, vdev_label_t *vl,
509 const uint64_t label_offset, const int l)
510 {
511 const int err = pread64(fd, vl, sizeof (vdev_label_t), label_offset);
512
513 if (err == -1) {
514 (void) fprintf(stderr,
515 "error: cannot read label %d: %s\n",
516 l, strerror(errno));
517 return (err);
518 } else if (err != sizeof (vdev_label_t)) {
519 (void) fprintf(stderr,
520 "error: bad label %d read size\n", l);
521 return (err);
522 }
523
524 return (0);
525 }
526
527 static void
zhack_repair_calc_cksum(const int byteswap,void * data,const uint64_t offset,const uint64_t abdsize,zio_eck_t * eck,zio_cksum_t * cksum)528 zhack_repair_calc_cksum(const int byteswap, void *data, const uint64_t offset,
529 const uint64_t abdsize, zio_eck_t *eck, zio_cksum_t *cksum)
530 {
531 zio_cksum_t verifier;
532 zio_cksum_t current_cksum;
533 zio_checksum_info_t *ci;
534 abd_t *abd;
535
536 ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
537
538 if (byteswap)
539 byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
540
541 current_cksum = eck->zec_cksum;
542 eck->zec_cksum = verifier;
543
544 ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
545 abd = abd_get_from_buf(data, abdsize);
546 ci->ci_func[byteswap](abd, abdsize, NULL, cksum);
547 abd_free(abd);
548
549 eck->zec_cksum = current_cksum;
550 }
551
552 static int
zhack_repair_check_label(uberblock_t * ub,const int l,const char ** cfg_keys,const size_t cfg_keys_len,nvlist_t * cfg,nvlist_t * vdev_tree_cfg,uint64_t * ashift)553 zhack_repair_check_label(uberblock_t *ub, const int l, const char **cfg_keys,
554 const size_t cfg_keys_len, nvlist_t *cfg, nvlist_t *vdev_tree_cfg,
555 uint64_t *ashift)
556 {
557 int err;
558
559 if (ub->ub_txg != 0) {
560 (void) fprintf(stderr,
561 "error: label %d: UB TXG of 0 expected, but got %"
562 PRIu64 "\n",
563 l, ub->ub_txg);
564 (void) fprintf(stderr, "It would appear the device was not "
565 "properly removed.\n");
566 return (1);
567 }
568
569 for (int i = 0; i < cfg_keys_len; i++) {
570 uint64_t val;
571 err = nvlist_lookup_uint64(cfg, cfg_keys[i], &val);
572 if (err) {
573 (void) fprintf(stderr,
574 "error: label %d, %d: "
575 "cannot find nvlist key %s\n",
576 l, i, cfg_keys[i]);
577 return (err);
578 }
579 }
580
581 err = nvlist_lookup_nvlist(cfg,
582 ZPOOL_CONFIG_VDEV_TREE, &vdev_tree_cfg);
583 if (err) {
584 (void) fprintf(stderr,
585 "error: label %d: cannot find nvlist key %s\n",
586 l, ZPOOL_CONFIG_VDEV_TREE);
587 return (err);
588 }
589
590 err = nvlist_lookup_uint64(vdev_tree_cfg,
591 ZPOOL_CONFIG_ASHIFT, ashift);
592 if (err) {
593 (void) fprintf(stderr,
594 "error: label %d: cannot find nvlist key %s\n",
595 l, ZPOOL_CONFIG_ASHIFT);
596 return (err);
597 }
598
599 if (*ashift == 0) {
600 (void) fprintf(stderr,
601 "error: label %d: nvlist key %s is zero\n",
602 l, ZPOOL_CONFIG_ASHIFT);
603 return (err);
604 }
605
606 return (0);
607 }
608
609 static int
zhack_repair_undetach(uberblock_t * ub,nvlist_t * cfg,const int l)610 zhack_repair_undetach(uberblock_t *ub, nvlist_t *cfg, const int l)
611 {
612 /*
613 * Uberblock root block pointer has valid birth TXG.
614 * Copying it to the label NVlist
615 */
616 if (BP_GET_LOGICAL_BIRTH(&ub->ub_rootbp) != 0) {
617 const uint64_t txg = BP_GET_LOGICAL_BIRTH(&ub->ub_rootbp);
618 ub->ub_txg = txg;
619
620 if (nvlist_remove_all(cfg, ZPOOL_CONFIG_CREATE_TXG) != 0) {
621 (void) fprintf(stderr,
622 "error: label %d: "
623 "Failed to remove pool creation TXG\n",
624 l);
625 return (1);
626 }
627
628 if (nvlist_remove_all(cfg, ZPOOL_CONFIG_POOL_TXG) != 0) {
629 (void) fprintf(stderr,
630 "error: label %d: Failed to remove pool TXG to "
631 "be replaced.\n",
632 l);
633 return (1);
634 }
635
636 if (nvlist_add_uint64(cfg, ZPOOL_CONFIG_POOL_TXG, txg) != 0) {
637 (void) fprintf(stderr,
638 "error: label %d: "
639 "Failed to add pool TXG of %" PRIu64 "\n",
640 l, txg);
641 return (1);
642 }
643 }
644
645 return (0);
646 }
647
648 static boolean_t
zhack_repair_write_label(const int l,const int fd,const int byteswap,void * data,zio_eck_t * eck,const uint64_t offset,const uint64_t abdsize)649 zhack_repair_write_label(const int l, const int fd, const int byteswap,
650 void *data, zio_eck_t *eck, const uint64_t offset, const uint64_t abdsize)
651 {
652 zio_cksum_t actual_cksum;
653 zhack_repair_calc_cksum(byteswap, data, offset, abdsize, eck,
654 &actual_cksum);
655 zio_cksum_t expected_cksum = eck->zec_cksum;
656 ssize_t err;
657
658 if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
659 return (B_FALSE);
660
661 eck->zec_cksum = actual_cksum;
662
663 err = pwrite64(fd, data, abdsize, offset);
664 if (err == -1) {
665 (void) fprintf(stderr, "error: cannot write label %d: %s\n",
666 l, strerror(errno));
667 return (B_FALSE);
668 } else if (err != abdsize) {
669 (void) fprintf(stderr, "error: bad write size label %d\n", l);
670 return (B_FALSE);
671 } else {
672 (void) fprintf(stderr,
673 "label %d: wrote %" PRIu64 " bytes at offset %" PRIu64 "\n",
674 l, abdsize, offset);
675 }
676
677 return (B_TRUE);
678 }
679
680 static void
zhack_repair_write_uberblock(vdev_label_t * vl,const int l,const uint64_t ashift,const int fd,const int byteswap,const uint64_t label_offset,uint32_t * labels_repaired)681 zhack_repair_write_uberblock(vdev_label_t *vl, const int l,
682 const uint64_t ashift, const int fd, const int byteswap,
683 const uint64_t label_offset, uint32_t *labels_repaired)
684 {
685 void *ub_data =
686 (char *)vl + offsetof(vdev_label_t, vl_uberblock);
687 zio_eck_t *ub_eck =
688 (zio_eck_t *)
689 ((char *)(ub_data) + (ASHIFT_UBERBLOCK_SIZE(ashift))) - 1;
690
691 if (ub_eck->zec_magic != 0) {
692 (void) fprintf(stderr,
693 "error: label %d: "
694 "Expected Uberblock checksum magic number to "
695 "be 0, but got %" PRIu64 "\n",
696 l, ub_eck->zec_magic);
697 (void) fprintf(stderr, "It would appear there's already "
698 "a checksum for the uberblock.\n");
699 return;
700 }
701
702
703 ub_eck->zec_magic = byteswap ? BSWAP_64(ZEC_MAGIC) : ZEC_MAGIC;
704
705 if (zhack_repair_write_label(l, fd, byteswap,
706 ub_data, ub_eck,
707 label_offset + offsetof(vdev_label_t, vl_uberblock),
708 ASHIFT_UBERBLOCK_SIZE(ashift)))
709 labels_repaired[l] |= REPAIR_LABEL_STATUS_UB;
710 }
711
712 static void
zhack_repair_print_cksum(FILE * stream,const zio_cksum_t * cksum)713 zhack_repair_print_cksum(FILE *stream, const zio_cksum_t *cksum)
714 {
715 (void) fprintf(stream,
716 "%016llx:%016llx:%016llx:%016llx",
717 (u_longlong_t)cksum->zc_word[0],
718 (u_longlong_t)cksum->zc_word[1],
719 (u_longlong_t)cksum->zc_word[2],
720 (u_longlong_t)cksum->zc_word[3]);
721 }
722
723 static int
zhack_repair_test_cksum(const int byteswap,void * vdev_data,zio_eck_t * vdev_eck,const uint64_t vdev_phys_offset,const int l)724 zhack_repair_test_cksum(const int byteswap, void *vdev_data,
725 zio_eck_t *vdev_eck, const uint64_t vdev_phys_offset, const int l)
726 {
727 const zio_cksum_t expected_cksum = vdev_eck->zec_cksum;
728 zio_cksum_t actual_cksum;
729 zhack_repair_calc_cksum(byteswap, vdev_data, vdev_phys_offset,
730 VDEV_PHYS_SIZE, vdev_eck, &actual_cksum);
731 const uint64_t expected_magic = byteswap ?
732 BSWAP_64(ZEC_MAGIC) : ZEC_MAGIC;
733 const uint64_t actual_magic = vdev_eck->zec_magic;
734 int err = 0;
735 if (actual_magic != expected_magic) {
736 (void) fprintf(stderr, "error: label %d: "
737 "Expected "
738 "the nvlist checksum magic number to not be %"
739 PRIu64 " not %" PRIu64 "\n",
740 l, expected_magic, actual_magic);
741 err = ECKSUM;
742 }
743 if (!ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum)) {
744 (void) fprintf(stderr, "error: label %d: "
745 "Expected the nvlist checksum to be ", l);
746 (void) zhack_repair_print_cksum(stderr,
747 &expected_cksum);
748 (void) fprintf(stderr, " not ");
749 zhack_repair_print_cksum(stderr, &actual_cksum);
750 (void) fprintf(stderr, "\n");
751 err = ECKSUM;
752 }
753 return (err);
754 }
755
756 static void
zhack_repair_one_label(const zhack_repair_op_t op,const int fd,vdev_label_t * vl,const uint64_t label_offset,const int l,uint32_t * labels_repaired)757 zhack_repair_one_label(const zhack_repair_op_t op, const int fd,
758 vdev_label_t *vl, const uint64_t label_offset, const int l,
759 uint32_t *labels_repaired)
760 {
761 ssize_t err;
762 uberblock_t *ub = (uberblock_t *)vl->vl_uberblock;
763 void *vdev_data =
764 (char *)vl + offsetof(vdev_label_t, vl_vdev_phys);
765 zio_eck_t *vdev_eck =
766 (zio_eck_t *)((char *)(vdev_data) + VDEV_PHYS_SIZE) - 1;
767 const uint64_t vdev_phys_offset =
768 label_offset + offsetof(vdev_label_t, vl_vdev_phys);
769 const char *cfg_keys[] = { ZPOOL_CONFIG_VERSION,
770 ZPOOL_CONFIG_POOL_STATE, ZPOOL_CONFIG_GUID };
771 nvlist_t *cfg;
772 nvlist_t *vdev_tree_cfg = NULL;
773 uint64_t ashift;
774 int byteswap;
775
776 err = zhack_repair_read_label(fd, vl, label_offset, l);
777 if (err)
778 return;
779
780 if (vdev_eck->zec_magic == 0) {
781 (void) fprintf(stderr, "error: label %d: "
782 "Expected the nvlist checksum magic number to not be zero"
783 "\n",
784 l);
785 (void) fprintf(stderr, "There should already be a checksum "
786 "for the label.\n");
787 return;
788 }
789
790 byteswap =
791 (vdev_eck->zec_magic == BSWAP_64((uint64_t)ZEC_MAGIC));
792
793 if (byteswap) {
794 byteswap_uint64_array(&vdev_eck->zec_cksum,
795 sizeof (zio_cksum_t));
796 vdev_eck->zec_magic = BSWAP_64(vdev_eck->zec_magic);
797 }
798
799 if ((op & ZHACK_REPAIR_OP_CKSUM) == 0 &&
800 zhack_repair_test_cksum(byteswap, vdev_data, vdev_eck,
801 vdev_phys_offset, l) != 0) {
802 (void) fprintf(stderr, "It would appear checksums are "
803 "corrupted. Try zhack repair label -c <device>\n");
804 return;
805 }
806
807 err = nvlist_unpack(vl->vl_vdev_phys.vp_nvlist,
808 VDEV_PHYS_SIZE - sizeof (zio_eck_t), &cfg, 0);
809 if (err) {
810 (void) fprintf(stderr,
811 "error: cannot unpack nvlist label %d\n", l);
812 return;
813 }
814
815 err = zhack_repair_check_label(ub,
816 l, cfg_keys, ARRAY_SIZE(cfg_keys), cfg, vdev_tree_cfg, &ashift);
817 if (err)
818 return;
819
820 if ((op & ZHACK_REPAIR_OP_UNDETACH) != 0) {
821 char *buf;
822 size_t buflen;
823
824 err = zhack_repair_undetach(ub, cfg, l);
825 if (err)
826 return;
827
828 buf = vl->vl_vdev_phys.vp_nvlist;
829 buflen = VDEV_PHYS_SIZE - sizeof (zio_eck_t);
830 if (nvlist_pack(cfg, &buf, &buflen, NV_ENCODE_XDR, 0) != 0) {
831 (void) fprintf(stderr,
832 "error: label %d: Failed to pack nvlist\n", l);
833 return;
834 }
835
836 zhack_repair_write_uberblock(vl,
837 l, ashift, fd, byteswap, label_offset, labels_repaired);
838 }
839
840 if (zhack_repair_write_label(l, fd, byteswap, vdev_data, vdev_eck,
841 vdev_phys_offset, VDEV_PHYS_SIZE))
842 labels_repaired[l] |= REPAIR_LABEL_STATUS_CKSUM;
843
844 fsync(fd);
845 }
846
847 static const char *
zhack_repair_label_status(const uint32_t label_status,const uint32_t to_check)848 zhack_repair_label_status(const uint32_t label_status,
849 const uint32_t to_check)
850 {
851 return ((label_status & to_check) != 0 ? "repaired" : "skipped");
852 }
853
854 static int
zhack_label_repair(const zhack_repair_op_t op,const int argc,char ** argv)855 zhack_label_repair(const zhack_repair_op_t op, const int argc, char **argv)
856 {
857 uint32_t labels_repaired[VDEV_LABELS] = {0};
858 vdev_label_t labels[VDEV_LABELS] = {{{0}}};
859 struct stat64 st;
860 int fd;
861 off_t filesize;
862 uint32_t repaired = 0;
863
864 abd_init();
865
866 if (argc < 1) {
867 (void) fprintf(stderr, "error: missing device\n");
868 usage();
869 }
870
871 if ((fd = open(argv[0], O_RDWR)) == -1)
872 fatal(NULL, FTAG, "cannot open '%s': %s", argv[0],
873 strerror(errno));
874
875 if (fstat64_blk(fd, &st) != 0)
876 fatal(NULL, FTAG, "cannot stat '%s': %s", argv[0],
877 strerror(errno));
878
879 filesize = st.st_size;
880 (void) fprintf(stderr, "Calculated filesize to be %jd\n",
881 (intmax_t)filesize);
882
883 if (filesize % sizeof (vdev_label_t) != 0)
884 filesize =
885 (filesize / sizeof (vdev_label_t)) * sizeof (vdev_label_t);
886
887 for (int l = 0; l < VDEV_LABELS; l++) {
888 zhack_repair_one_label(op, fd, &labels[l],
889 vdev_label_offset(filesize, l, 0), l, labels_repaired);
890 }
891
892 close(fd);
893
894 abd_fini();
895
896 for (int l = 0; l < VDEV_LABELS; l++) {
897 const uint32_t lr = labels_repaired[l];
898 (void) printf("label %d: ", l);
899 (void) printf("uberblock: %s ",
900 zhack_repair_label_status(lr, REPAIR_LABEL_STATUS_UB));
901 (void) printf("checksum: %s\n",
902 zhack_repair_label_status(lr, REPAIR_LABEL_STATUS_CKSUM));
903 repaired |= lr;
904 }
905
906 if (repaired > 0)
907 return (0);
908
909 return (1);
910 }
911
912 static int
zhack_do_label_repair(int argc,char ** argv)913 zhack_do_label_repair(int argc, char **argv)
914 {
915 zhack_repair_op_t op = ZHACK_REPAIR_OP_UNKNOWN;
916 int c;
917
918 optind = 1;
919 while ((c = getopt(argc, argv, "+cu")) != -1) {
920 switch (c) {
921 case 'c':
922 op |= ZHACK_REPAIR_OP_CKSUM;
923 break;
924 case 'u':
925 op |= ZHACK_REPAIR_OP_UNDETACH;
926 break;
927 default:
928 usage();
929 break;
930 }
931 }
932
933 argc -= optind;
934 argv += optind;
935
936 if (op == ZHACK_REPAIR_OP_UNKNOWN)
937 op = ZHACK_REPAIR_OP_CKSUM;
938
939 return (zhack_label_repair(op, argc, argv));
940 }
941
942 static int
zhack_do_label(int argc,char ** argv)943 zhack_do_label(int argc, char **argv)
944 {
945 char *subcommand;
946 int err;
947
948 argc--;
949 argv++;
950 if (argc == 0) {
951 (void) fprintf(stderr,
952 "error: no label operation specified\n");
953 usage();
954 }
955
956 subcommand = argv[0];
957 if (strcmp(subcommand, "repair") == 0) {
958 err = zhack_do_label_repair(argc, argv);
959 } else {
960 (void) fprintf(stderr, "error: unknown subcommand: %s\n",
961 subcommand);
962 usage();
963 }
964
965 return (err);
966 }
967
968 #define MAX_NUM_PATHS 1024
969
970 int
main(int argc,char ** argv)971 main(int argc, char **argv)
972 {
973 char *path[MAX_NUM_PATHS];
974 const char *subcommand;
975 int rv = 0;
976 int c;
977
978 g_importargs.path = path;
979
980 dprintf_setup(&argc, argv);
981 zfs_prop_init();
982
983 while ((c = getopt(argc, argv, "+c:d:")) != -1) {
984 switch (c) {
985 case 'c':
986 g_importargs.cachefile = optarg;
987 break;
988 case 'd':
989 assert(g_importargs.paths < MAX_NUM_PATHS);
990 g_importargs.path[g_importargs.paths++] = optarg;
991 break;
992 default:
993 usage();
994 break;
995 }
996 }
997
998 argc -= optind;
999 argv += optind;
1000 optind = 1;
1001
1002 if (argc == 0) {
1003 (void) fprintf(stderr, "error: no command specified\n");
1004 usage();
1005 }
1006
1007 subcommand = argv[0];
1008
1009 if (strcmp(subcommand, "feature") == 0) {
1010 rv = zhack_do_feature(argc, argv);
1011 } else if (strcmp(subcommand, "label") == 0) {
1012 return (zhack_do_label(argc, argv));
1013 } else {
1014 (void) fprintf(stderr, "error: unknown subcommand: %s\n",
1015 subcommand);
1016 usage();
1017 }
1018
1019 if (!g_readonly && spa_export(g_pool, NULL, B_TRUE, B_FALSE) != 0) {
1020 fatal(NULL, FTAG, "pool export failed; "
1021 "changes may not be committed to disk\n");
1022 }
1023
1024 kernel_fini();
1025
1026 return (rv);
1027 }
1028