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) 2011, 2019 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Nexenta Systems, Inc.
28 * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC.
29 * Copyright (c) 2015, 2017, Intel Corporation.
30 * Copyright (c) 2020 Datto Inc.
31 * Copyright (c) 2020, The FreeBSD Foundation [1]
32 *
33 * [1] Portions of this software were developed by Allan Jude
34 * under sponsorship from the FreeBSD Foundation.
35 * Copyright (c) 2021 Allan Jude
36 * Copyright (c) 2021 Toomas Soome <tsoome@me.com>
37 * Copyright (c) 2023, 2024, Klara Inc.
38 * Copyright (c) 2023, Rob Norris <robn@despairlabs.com>
39 */
40
41 #include <stdio.h>
42 #include <unistd.h>
43 #include <stdlib.h>
44 #include <ctype.h>
45 #include <getopt.h>
46 #include <openssl/evp.h>
47 #include <sys/zfs_context.h>
48 #include <sys/spa.h>
49 #include <sys/spa_impl.h>
50 #include <sys/dmu.h>
51 #include <sys/zap.h>
52 #include <sys/zap_impl.h>
53 #include <sys/fs/zfs.h>
54 #include <sys/zfs_znode.h>
55 #include <sys/zfs_sa.h>
56 #include <sys/sa.h>
57 #include <sys/sa_impl.h>
58 #include <sys/vdev.h>
59 #include <sys/vdev_impl.h>
60 #include <sys/metaslab_impl.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/dsl_dir.h>
63 #include <sys/dsl_dataset.h>
64 #include <sys/dsl_pool.h>
65 #include <sys/dsl_bookmark.h>
66 #include <sys/dbuf.h>
67 #include <sys/zil.h>
68 #include <sys/zil_impl.h>
69 #include <sys/stat.h>
70 #include <sys/resource.h>
71 #include <sys/dmu_send.h>
72 #include <sys/dmu_traverse.h>
73 #include <sys/zio_checksum.h>
74 #include <sys/zio_compress.h>
75 #include <sys/zfs_fuid.h>
76 #include <sys/arc.h>
77 #include <sys/arc_impl.h>
78 #include <sys/ddt.h>
79 #include <sys/ddt_impl.h>
80 #include <sys/zfeature.h>
81 #include <sys/abd.h>
82 #include <sys/blkptr.h>
83 #include <sys/dsl_crypt.h>
84 #include <sys/dsl_scan.h>
85 #include <sys/btree.h>
86 #include <sys/brt.h>
87 #include <sys/brt_impl.h>
88 #include <zfs_comutil.h>
89 #include <sys/zstd/zstd.h>
90 #include <sys/backtrace.h>
91
92 #include <libnvpair.h>
93 #include <libzutil.h>
94 #include <libzfs_core.h>
95
96 #include <libzdb.h>
97
98 #include "zdb.h"
99
100
101 extern int reference_tracking_enable;
102 extern int zfs_recover;
103 extern uint_t zfs_vdev_async_read_max_active;
104 extern boolean_t spa_load_verify_dryrun;
105 extern boolean_t spa_mode_readable_spacemaps;
106 extern uint_t zfs_reconstruct_indirect_combinations_max;
107 extern uint_t zfs_btree_verify_intensity;
108
109 enum {
110 ARG_ALLOCATED = 256,
111 ARG_BLOCK_BIN_MODE,
112 ARG_BLOCK_CLASSES,
113 };
114
115 static const char cmdname[] = "zdb";
116 uint8_t dump_opt[512];
117
118 typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size);
119
120 static uint64_t *zopt_metaslab = NULL;
121 static unsigned zopt_metaslab_args = 0;
122
123
124 static zopt_object_range_t *zopt_object_ranges = NULL;
125 static unsigned zopt_object_args = 0;
126
127 static int flagbits[256];
128
129
130 static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */
131 static int leaked_objects = 0;
132 static zfs_range_tree_t *mos_refd_objs;
133 static spa_t *spa;
134 static objset_t *os;
135 static boolean_t kernel_init_done;
136 static boolean_t corruption_found = B_FALSE;
137
138 static enum {
139 BIN_AUTO = 0,
140 BIN_PSIZE,
141 BIN_LSIZE,
142 BIN_ASIZE,
143 } block_bin_mode = BIN_AUTO;
144
145 static enum {
146 CLASS_NORMAL = 1 << 1,
147 CLASS_SPECIAL = 1 << 2,
148 CLASS_DEDUP = 1 << 3,
149 CLASS_OTHER = 1 << 4,
150 } block_classes = 0;
151
152 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *,
153 boolean_t);
154 static void mos_obj_refd(uint64_t);
155 static void mos_obj_refd_multiple(uint64_t);
156 static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free,
157 dmu_tx_t *tx);
158
159
160
161 static void zdb_print_blkptr(const blkptr_t *bp, int flags);
162 static void zdb_exit(int reason);
163
164 typedef struct sublivelist_verify_block_refcnt {
165 /* block pointer entry in livelist being verified */
166 blkptr_t svbr_blk;
167
168 /*
169 * Refcount gets incremented to 1 when we encounter the first
170 * FREE entry for the svfbr block pointer and a node for it
171 * is created in our ZDB verification/tracking metadata.
172 *
173 * As we encounter more FREE entries we increment this counter
174 * and similarly decrement it whenever we find the respective
175 * ALLOC entries for this block.
176 *
177 * When the refcount gets to 0 it means that all the FREE and
178 * ALLOC entries of this block have paired up and we no longer
179 * need to track it in our verification logic (e.g. the node
180 * containing this struct in our verification data structure
181 * should be freed).
182 *
183 * [refer to sublivelist_verify_blkptr() for the actual code]
184 */
185 uint32_t svbr_refcnt;
186 } sublivelist_verify_block_refcnt_t;
187
188 static int
sublivelist_block_refcnt_compare(const void * larg,const void * rarg)189 sublivelist_block_refcnt_compare(const void *larg, const void *rarg)
190 {
191 const sublivelist_verify_block_refcnt_t *l = larg;
192 const sublivelist_verify_block_refcnt_t *r = rarg;
193 return (livelist_compare(&l->svbr_blk, &r->svbr_blk));
194 }
195
196 static int
sublivelist_verify_blkptr(void * arg,const blkptr_t * bp,boolean_t free,dmu_tx_t * tx)197 sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free,
198 dmu_tx_t *tx)
199 {
200 ASSERT0P(tx);
201 struct sublivelist_verify *sv = arg;
202 sublivelist_verify_block_refcnt_t current = {
203 .svbr_blk = *bp,
204
205 /*
206 * Start with 1 in case this is the first free entry.
207 * This field is not used for our B-Tree comparisons
208 * anyway.
209 */
210 .svbr_refcnt = 1,
211 };
212
213 zfs_btree_index_t where;
214 sublivelist_verify_block_refcnt_t *pair =
215 zfs_btree_find(&sv->sv_pair, ¤t, &where);
216 if (free) {
217 if (pair == NULL) {
218 /* first free entry for this block pointer */
219 zfs_btree_add(&sv->sv_pair, ¤t);
220 } else {
221 pair->svbr_refcnt++;
222 }
223 } else {
224 if (pair == NULL) {
225 /* block that is currently marked as allocated */
226 for (int i = 0; i < SPA_DVAS_PER_BP; i++) {
227 if (DVA_IS_EMPTY(&bp->blk_dva[i]))
228 break;
229 sublivelist_verify_block_t svb = {
230 .svb_dva = bp->blk_dva[i],
231 .svb_allocated_txg =
232 BP_GET_BIRTH(bp)
233 };
234
235 if (zfs_btree_find(&sv->sv_leftover, &svb,
236 &where) == NULL) {
237 zfs_btree_add_idx(&sv->sv_leftover,
238 &svb, &where);
239 }
240 }
241 } else {
242 /* alloc matches a free entry */
243 pair->svbr_refcnt--;
244 if (pair->svbr_refcnt == 0) {
245 /* all allocs and frees have been matched */
246 zfs_btree_remove_idx(&sv->sv_pair, &where);
247 }
248 }
249 }
250
251 return (0);
252 }
253
254 static int
sublivelist_verify_func(void * args,dsl_deadlist_entry_t * dle)255 sublivelist_verify_func(void *args, dsl_deadlist_entry_t *dle)
256 {
257 int err;
258 struct sublivelist_verify *sv = args;
259
260 zfs_btree_create(&sv->sv_pair, sublivelist_block_refcnt_compare, NULL,
261 sizeof (sublivelist_verify_block_refcnt_t));
262
263 err = bpobj_iterate_nofree(&dle->dle_bpobj, sublivelist_verify_blkptr,
264 sv, NULL);
265
266 sublivelist_verify_block_refcnt_t *e;
267 zfs_btree_index_t *cookie = NULL;
268 while ((e = zfs_btree_destroy_nodes(&sv->sv_pair, &cookie)) != NULL) {
269 char blkbuf[BP_SPRINTF_LEN];
270 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
271 &e->svbr_blk, B_TRUE);
272 (void) printf("\tERROR: %d unmatched FREE(s): %s\n",
273 e->svbr_refcnt, blkbuf);
274 corruption_found = B_TRUE;
275 }
276 zfs_btree_destroy(&sv->sv_pair);
277
278 return (err);
279 }
280
281 static int
livelist_block_compare(const void * larg,const void * rarg)282 livelist_block_compare(const void *larg, const void *rarg)
283 {
284 const sublivelist_verify_block_t *l = larg;
285 const sublivelist_verify_block_t *r = rarg;
286
287 if (DVA_GET_VDEV(&l->svb_dva) < DVA_GET_VDEV(&r->svb_dva))
288 return (-1);
289 else if (DVA_GET_VDEV(&l->svb_dva) > DVA_GET_VDEV(&r->svb_dva))
290 return (+1);
291
292 if (DVA_GET_OFFSET(&l->svb_dva) < DVA_GET_OFFSET(&r->svb_dva))
293 return (-1);
294 else if (DVA_GET_OFFSET(&l->svb_dva) > DVA_GET_OFFSET(&r->svb_dva))
295 return (+1);
296
297 if (DVA_GET_ASIZE(&l->svb_dva) < DVA_GET_ASIZE(&r->svb_dva))
298 return (-1);
299 else if (DVA_GET_ASIZE(&l->svb_dva) > DVA_GET_ASIZE(&r->svb_dva))
300 return (+1);
301
302 return (0);
303 }
304
305 /*
306 * Check for errors in a livelist while tracking all unfreed ALLOCs in the
307 * sublivelist_verify_t: sv->sv_leftover
308 */
309 static void
livelist_verify(dsl_deadlist_t * dl,void * arg)310 livelist_verify(dsl_deadlist_t *dl, void *arg)
311 {
312 sublivelist_verify_t *sv = arg;
313 dsl_deadlist_iterate(dl, sublivelist_verify_func, sv);
314 }
315
316 /*
317 * Check for errors in the livelist entry and discard the intermediary
318 * data structures
319 */
320 static int
sublivelist_verify_lightweight(void * args,dsl_deadlist_entry_t * dle)321 sublivelist_verify_lightweight(void *args, dsl_deadlist_entry_t *dle)
322 {
323 (void) args;
324 sublivelist_verify_t sv;
325 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
326 sizeof (sublivelist_verify_block_t));
327 int err = sublivelist_verify_func(&sv, dle);
328 zfs_btree_clear(&sv.sv_leftover);
329 zfs_btree_destroy(&sv.sv_leftover);
330 return (err);
331 }
332
333 typedef struct metaslab_verify {
334 /*
335 * Tree containing all the leftover ALLOCs from the livelists
336 * that are part of this metaslab.
337 */
338 zfs_btree_t mv_livelist_allocs;
339
340 /*
341 * Metaslab information.
342 */
343 uint64_t mv_vdid;
344 uint64_t mv_msid;
345 uint64_t mv_start;
346 uint64_t mv_end;
347
348 /*
349 * What's currently allocated for this metaslab.
350 */
351 zfs_range_tree_t *mv_allocated;
352 } metaslab_verify_t;
353
354 typedef void ll_iter_t(dsl_deadlist_t *ll, void *arg);
355
356 typedef int (*zdb_log_sm_cb_t)(spa_t *spa, space_map_entry_t *sme, uint64_t txg,
357 void *arg);
358
359 typedef struct unflushed_iter_cb_arg {
360 spa_t *uic_spa;
361 uint64_t uic_txg;
362 void *uic_arg;
363 zdb_log_sm_cb_t uic_cb;
364 } unflushed_iter_cb_arg_t;
365
366 static int
iterate_through_spacemap_logs_cb(space_map_entry_t * sme,void * arg)367 iterate_through_spacemap_logs_cb(space_map_entry_t *sme, void *arg)
368 {
369 unflushed_iter_cb_arg_t *uic = arg;
370 return (uic->uic_cb(uic->uic_spa, sme, uic->uic_txg, uic->uic_arg));
371 }
372
373 static void
iterate_through_spacemap_logs(spa_t * spa,zdb_log_sm_cb_t cb,void * arg)374 iterate_through_spacemap_logs(spa_t *spa, zdb_log_sm_cb_t cb, void *arg)
375 {
376 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
377 return;
378
379 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
380 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
381 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
382 space_map_t *sm = NULL;
383 VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
384 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
385
386 unflushed_iter_cb_arg_t uic = {
387 .uic_spa = spa,
388 .uic_txg = sls->sls_txg,
389 .uic_arg = arg,
390 .uic_cb = cb
391 };
392 VERIFY0(space_map_iterate(sm, space_map_length(sm),
393 iterate_through_spacemap_logs_cb, &uic));
394 space_map_close(sm);
395 }
396 spa_config_exit(spa, SCL_CONFIG, FTAG);
397 }
398
399 static void
verify_livelist_allocs(metaslab_verify_t * mv,uint64_t txg,uint64_t offset,uint64_t size)400 verify_livelist_allocs(metaslab_verify_t *mv, uint64_t txg,
401 uint64_t offset, uint64_t size)
402 {
403 sublivelist_verify_block_t svb = {{{0}}};
404 DVA_SET_VDEV(&svb.svb_dva, mv->mv_vdid);
405 DVA_SET_OFFSET(&svb.svb_dva, offset);
406 DVA_SET_ASIZE(&svb.svb_dva, 0);
407 zfs_btree_index_t where;
408 uint64_t end_offset = offset + size;
409
410 /*
411 * Look for an exact match for spacemap entry in the livelist entries.
412 * Then, look for other livelist entries that fall within the range
413 * of the spacemap entry as it may have been condensed
414 */
415 sublivelist_verify_block_t *found =
416 zfs_btree_find(&mv->mv_livelist_allocs, &svb, &where);
417 if (found == NULL) {
418 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where);
419 }
420 for (; found != NULL && DVA_GET_VDEV(&found->svb_dva) == mv->mv_vdid &&
421 DVA_GET_OFFSET(&found->svb_dva) < end_offset;
422 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
423 if (found->svb_allocated_txg <= txg) {
424 (void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
425 "from TXG %llx FREED at TXG %llx\n",
426 (u_longlong_t)DVA_GET_OFFSET(&found->svb_dva),
427 (u_longlong_t)DVA_GET_ASIZE(&found->svb_dva),
428 (u_longlong_t)found->svb_allocated_txg,
429 (u_longlong_t)txg);
430 corruption_found = B_TRUE;
431 }
432 }
433 }
434
435 static int
metaslab_spacemap_validation_cb(space_map_entry_t * sme,void * arg)436 metaslab_spacemap_validation_cb(space_map_entry_t *sme, void *arg)
437 {
438 metaslab_verify_t *mv = arg;
439 uint64_t offset = sme->sme_offset;
440 uint64_t size = sme->sme_run;
441 uint64_t txg = sme->sme_txg;
442
443 if (sme->sme_type == SM_ALLOC) {
444 if (zfs_range_tree_contains(mv->mv_allocated,
445 offset, size)) {
446 (void) printf("ERROR: DOUBLE ALLOC: "
447 "%llu [%llx:%llx] "
448 "%llu:%llu LOG_SM\n",
449 (u_longlong_t)txg, (u_longlong_t)offset,
450 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
451 (u_longlong_t)mv->mv_msid);
452 corruption_found = B_TRUE;
453 } else {
454 zfs_range_tree_add(mv->mv_allocated,
455 offset, size);
456 }
457 } else {
458 if (!zfs_range_tree_contains(mv->mv_allocated,
459 offset, size)) {
460 (void) printf("ERROR: DOUBLE FREE: "
461 "%llu [%llx:%llx] "
462 "%llu:%llu LOG_SM\n",
463 (u_longlong_t)txg, (u_longlong_t)offset,
464 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
465 (u_longlong_t)mv->mv_msid);
466 corruption_found = B_TRUE;
467 } else {
468 zfs_range_tree_remove(mv->mv_allocated,
469 offset, size);
470 }
471 }
472
473 if (sme->sme_type != SM_ALLOC) {
474 /*
475 * If something is freed in the spacemap, verify that
476 * it is not listed as allocated in the livelist.
477 */
478 verify_livelist_allocs(mv, txg, offset, size);
479 }
480 return (0);
481 }
482
483 static int
spacemap_check_sm_log_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)484 spacemap_check_sm_log_cb(spa_t *spa, space_map_entry_t *sme,
485 uint64_t txg, void *arg)
486 {
487 metaslab_verify_t *mv = arg;
488 uint64_t offset = sme->sme_offset;
489 uint64_t vdev_id = sme->sme_vdev;
490
491 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
492
493 /* skip indirect vdevs */
494 if (!vdev_is_concrete(vd))
495 return (0);
496
497 if (vdev_id != mv->mv_vdid)
498 return (0);
499
500 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
501 if (ms->ms_id != mv->mv_msid)
502 return (0);
503
504 if (txg < metaslab_unflushed_txg(ms))
505 return (0);
506
507
508 ASSERT3U(txg, ==, sme->sme_txg);
509 return (metaslab_spacemap_validation_cb(sme, mv));
510 }
511
512 static void
spacemap_check_sm_log(spa_t * spa,metaslab_verify_t * mv)513 spacemap_check_sm_log(spa_t *spa, metaslab_verify_t *mv)
514 {
515 iterate_through_spacemap_logs(spa, spacemap_check_sm_log_cb, mv);
516 }
517
518 static void
spacemap_check_ms_sm(space_map_t * sm,metaslab_verify_t * mv)519 spacemap_check_ms_sm(space_map_t *sm, metaslab_verify_t *mv)
520 {
521 if (sm == NULL)
522 return;
523
524 VERIFY0(space_map_iterate(sm, space_map_length(sm),
525 metaslab_spacemap_validation_cb, mv));
526 }
527
528 static void iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg);
529
530 /*
531 * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
532 * they are part of that metaslab (mv_msid).
533 */
534 static void
mv_populate_livelist_allocs(metaslab_verify_t * mv,sublivelist_verify_t * sv)535 mv_populate_livelist_allocs(metaslab_verify_t *mv, sublivelist_verify_t *sv)
536 {
537 zfs_btree_index_t where;
538 sublivelist_verify_block_t *svb;
539 ASSERT3U(zfs_btree_numnodes(&mv->mv_livelist_allocs), ==, 0);
540 for (svb = zfs_btree_first(&sv->sv_leftover, &where);
541 svb != NULL;
542 svb = zfs_btree_next(&sv->sv_leftover, &where, &where)) {
543 if (DVA_GET_VDEV(&svb->svb_dva) != mv->mv_vdid)
544 continue;
545
546 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start &&
547 (DVA_GET_OFFSET(&svb->svb_dva) +
548 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_start) {
549 (void) printf("ERROR: Found block that crosses "
550 "metaslab boundary: <%llu:%llx:%llx>\n",
551 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
552 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
553 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
554 corruption_found = B_TRUE;
555 continue;
556 }
557
558 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start)
559 continue;
560
561 if (DVA_GET_OFFSET(&svb->svb_dva) >= mv->mv_end)
562 continue;
563
564 if ((DVA_GET_OFFSET(&svb->svb_dva) +
565 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_end) {
566 (void) printf("ERROR: Found block that crosses "
567 "metaslab boundary: <%llu:%llx:%llx>\n",
568 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
569 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
570 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
571 corruption_found = B_TRUE;
572 continue;
573 }
574
575 zfs_btree_add(&mv->mv_livelist_allocs, svb);
576 }
577
578 for (svb = zfs_btree_first(&mv->mv_livelist_allocs, &where);
579 svb != NULL;
580 svb = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
581 zfs_btree_remove(&sv->sv_leftover, svb);
582 }
583 }
584
585 /*
586 * [Livelist Check]
587 * Iterate through all the sublivelists and:
588 * - report leftover frees (**)
589 * - record leftover ALLOCs together with their TXG [see Cross Check]
590 *
591 * (**) Note: Double ALLOCs are valid in datasets that have dedup
592 * enabled. Similarly double FREEs are allowed as well but
593 * only if they pair up with a corresponding ALLOC entry once
594 * we our done with our sublivelist iteration.
595 *
596 * [Spacemap Check]
597 * for each metaslab:
598 * - iterate over spacemap and then the metaslab's entries in the
599 * spacemap log, then report any double FREEs and ALLOCs (do not
600 * blow up).
601 *
602 * [Cross Check]
603 * After finishing the Livelist Check phase and while being in the
604 * Spacemap Check phase, we find all the recorded leftover ALLOCs
605 * of the livelist check that are part of the metaslab that we are
606 * currently looking at in the Spacemap Check. We report any entries
607 * that are marked as ALLOCs in the livelists but have been actually
608 * freed (and potentially allocated again) after their TXG stamp in
609 * the spacemaps. Also report any ALLOCs from the livelists that
610 * belong to indirect vdevs (e.g. their vdev completed removal).
611 *
612 * Note that this will miss Log Spacemap entries that cancelled each other
613 * out before being flushed to the metaslab, so we are not guaranteed
614 * to match all erroneous ALLOCs.
615 */
616 static void
livelist_metaslab_validate(spa_t * spa)617 livelist_metaslab_validate(spa_t *spa)
618 {
619 (void) printf("Verifying deleted livelist entries\n");
620
621 sublivelist_verify_t sv;
622 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
623 sizeof (sublivelist_verify_block_t));
624 iterate_deleted_livelists(spa, livelist_verify, &sv);
625
626 (void) printf("Verifying metaslab entries\n");
627 vdev_t *rvd = spa->spa_root_vdev;
628 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
629 vdev_t *vd = rvd->vdev_child[c];
630
631 if (!vdev_is_concrete(vd))
632 continue;
633
634 for (uint64_t mid = 0; mid < vd->vdev_ms_count; mid++) {
635 metaslab_t *m = vd->vdev_ms[mid];
636
637 (void) fprintf(stderr,
638 "\rverifying concrete vdev %llu, "
639 "metaslab %llu of %llu ...",
640 (longlong_t)vd->vdev_id,
641 (longlong_t)mid,
642 (longlong_t)vd->vdev_ms_count);
643
644 uint64_t shift, start;
645 zfs_range_seg_type_t type =
646 metaslab_calculate_range_tree_type(vd, m,
647 &start, &shift);
648 metaslab_verify_t mv;
649 mv.mv_allocated = zfs_range_tree_create_flags(
650 NULL, type, NULL, start, shift,
651 0, "livelist_metaslab_validate:mv_allocated");
652 mv.mv_vdid = vd->vdev_id;
653 mv.mv_msid = m->ms_id;
654 mv.mv_start = m->ms_start;
655 mv.mv_end = m->ms_start + m->ms_size;
656 zfs_btree_create(&mv.mv_livelist_allocs,
657 livelist_block_compare, NULL,
658 sizeof (sublivelist_verify_block_t));
659
660 mv_populate_livelist_allocs(&mv, &sv);
661
662 spacemap_check_ms_sm(m->ms_sm, &mv);
663 spacemap_check_sm_log(spa, &mv);
664
665 zfs_range_tree_vacate(mv.mv_allocated, NULL, NULL);
666 zfs_range_tree_destroy(mv.mv_allocated);
667 zfs_btree_clear(&mv.mv_livelist_allocs);
668 zfs_btree_destroy(&mv.mv_livelist_allocs);
669 }
670 }
671 (void) fprintf(stderr, "\n");
672
673 /*
674 * If there are any segments in the leftover tree after we walked
675 * through all the metaslabs in the concrete vdevs then this means
676 * that we have segments in the livelists that belong to indirect
677 * vdevs and are marked as allocated.
678 */
679 if (zfs_btree_numnodes(&sv.sv_leftover) == 0) {
680 zfs_btree_destroy(&sv.sv_leftover);
681 return;
682 }
683 (void) printf("ERROR: Found livelist blocks marked as allocated "
684 "for indirect vdevs:\n");
685 corruption_found = B_TRUE;
686
687 zfs_btree_index_t *where = NULL;
688 sublivelist_verify_block_t *svb;
689 while ((svb = zfs_btree_destroy_nodes(&sv.sv_leftover, &where)) !=
690 NULL) {
691 int vdev_id = DVA_GET_VDEV(&svb->svb_dva);
692 ASSERT3U(vdev_id, <, rvd->vdev_children);
693 vdev_t *vd = rvd->vdev_child[vdev_id];
694 ASSERT(!vdev_is_concrete(vd));
695 (void) printf("<%d:%llx:%llx> TXG %llx\n",
696 vdev_id, (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
697 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva),
698 (u_longlong_t)svb->svb_allocated_txg);
699 }
700 (void) printf("\n");
701 zfs_btree_destroy(&sv.sv_leftover);
702 }
703
704 /*
705 * These libumem hooks provide a reasonable set of defaults for the allocator's
706 * debugging facilities.
707 */
708 const char *
_umem_debug_init(void)709 _umem_debug_init(void)
710 {
711 return ("default,verbose"); /* $UMEM_DEBUG setting */
712 }
713
714 const char *
_umem_logging_init(void)715 _umem_logging_init(void)
716 {
717 return ("fail,contents"); /* $UMEM_LOGGING setting */
718 }
719
720 static void
usage(void)721 usage(void)
722 {
723 (void) fprintf(stderr,
724 "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
725 "[-I <inflight I/Os>]\n"
726 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
727 "\t\t[-K <key>]\n"
728 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
729 "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
730 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
731 "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
732 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
733 "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
734 "\t%s [-v] <bookmark>\n"
735 "\t%s -C [-A] [-U <cache>] [<poolname>]\n"
736 "\t%s -l [-Aqu] <device>\n"
737 "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
738 "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
739 "\t%s -O [-K <key>] <dataset> <path>\n"
740 "\t%s -r [-K <key>] <dataset> <path> <destination>\n"
741 "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
742 "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
743 "\t%s -E [-A] word0:word1:...:word15\n"
744 "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
745 "<poolname>\n\n",
746 cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
747 cmdname, cmdname, cmdname, cmdname, cmdname);
748
749 (void) fprintf(stderr, " Dataset name must include at least one "
750 "separator character '/' or '@'\n");
751 (void) fprintf(stderr, " If dataset name is specified, only that "
752 "dataset is dumped\n");
753 (void) fprintf(stderr, " If object numbers or object number "
754 "ranges are specified, only those\n"
755 " objects or ranges are dumped.\n\n");
756 (void) fprintf(stderr,
757 " Object ranges take the form <start>:<end>[:<flags>]\n"
758 " start Starting object number\n"
759 " end Ending object number, or -1 for no upper bound\n"
760 " flags Optional flags to select object types:\n"
761 " A All objects (this is the default)\n"
762 " d ZFS directories\n"
763 " f ZFS files \n"
764 " m SPA space maps\n"
765 " z ZAPs\n"
766 " - Negate effect of next flag\n\n");
767 (void) fprintf(stderr, " Options to control amount of output:\n");
768 (void) fprintf(stderr, " -b --block-stats "
769 "block statistics\n");
770 (void) fprintf(stderr, " --bin=(lsize|psize|asize) "
771 "bin blocks based on this size in all three columns\n");
772 (void) fprintf(stderr,
773 " --class=(normal|special|dedup|other)[,...]\n"
774 " only consider blocks from "
775 "these allocation classes\n");
776 (void) fprintf(stderr, " -B --backup "
777 "backup stream\n");
778 (void) fprintf(stderr, " -c --checksum "
779 "checksum all metadata (twice for all data) blocks\n");
780 (void) fprintf(stderr, " -C --config "
781 "config (or cachefile if alone)\n");
782 (void) fprintf(stderr, " -d --datasets "
783 "dataset(s)\n");
784 (void) fprintf(stderr, " -D --dedup-stats "
785 "dedup statistics\n");
786 (void) fprintf(stderr, " -E --embedded-block-pointer=INTEGER\n"
787 " decode and display block "
788 "from an embedded block pointer\n");
789 (void) fprintf(stderr, " -h --history "
790 "pool history\n");
791 (void) fprintf(stderr, " -i --intent-logs "
792 "intent logs\n");
793 (void) fprintf(stderr, " -l --label "
794 "read label contents\n");
795 (void) fprintf(stderr, " -k --checkpointed-state "
796 "examine the checkpointed state of the pool\n");
797 (void) fprintf(stderr, " -L --disable-leak-tracking "
798 "disable leak tracking (do not load spacemaps)\n");
799 (void) fprintf(stderr, " -m --metaslabs "
800 "metaslabs\n");
801 (void) fprintf(stderr, " -M --metaslab-groups "
802 "metaslab groups\n");
803 (void) fprintf(stderr, " -O --object-lookups "
804 "perform object lookups by path\n");
805 (void) fprintf(stderr, " -r --copy-object "
806 "copy an object by path to file\n");
807 (void) fprintf(stderr, " -R --read-block "
808 "read and display block from a device\n");
809 (void) fprintf(stderr, " -s --io-stats "
810 "report stats on zdb's I/O\n");
811 (void) fprintf(stderr, " -S --simulate-dedup "
812 "simulate dedup to measure effect\n");
813 (void) fprintf(stderr, " -v --verbose "
814 "verbose (applies to all others)\n");
815 (void) fprintf(stderr, " -y --livelist "
816 "perform livelist and metaslab validation on any livelists being "
817 "deleted\n\n");
818 (void) fprintf(stderr, " Below options are intended for use "
819 "with other options:\n");
820 (void) fprintf(stderr, " -A --ignore-assertions "
821 "ignore assertions (-A), enable panic recovery (-AA) or both "
822 "(-AAA)\n");
823 (void) fprintf(stderr, " -e --exported "
824 "pool is exported/destroyed/has altroot/not in a cachefile\n");
825 (void) fprintf(stderr, " -F --automatic-rewind "
826 "attempt automatic rewind within safe range of transaction "
827 "groups\n");
828 (void) fprintf(stderr, " -G --dump-debug-msg "
829 "dump zfs_dbgmsg buffer before exiting\n");
830 (void) fprintf(stderr, " -I --inflight=INTEGER "
831 "specify the maximum number of checksumming I/Os "
832 "[default is 200]\n");
833 (void) fprintf(stderr, " -K --key=KEY "
834 "decryption key for encrypted dataset\n");
835 (void) fprintf(stderr, " -o --option=\"NAME=VALUE\" "
836 "set the named tunable to the given value\n");
837 (void) fprintf(stderr, " -p --path==PATH "
838 "use one or more with -e to specify path to vdev dir\n");
839 (void) fprintf(stderr, " -P --parseable "
840 "print numbers in parseable form\n");
841 (void) fprintf(stderr, " -q --skip-label "
842 "don't print label contents\n");
843 (void) fprintf(stderr, " -t --txg=INTEGER "
844 "highest txg to use when searching for uberblocks\n");
845 (void) fprintf(stderr, " -T --brt-stats "
846 "BRT statistics\n");
847 (void) fprintf(stderr, " -u --uberblock "
848 "uberblock\n");
849 (void) fprintf(stderr, " -U --cachefile=PATH "
850 "use alternate cachefile\n");
851 (void) fprintf(stderr, " -V --verbatim "
852 "do verbatim import\n");
853 (void) fprintf(stderr, " -x --dump-blocks=PATH "
854 "dump all read blocks into specified directory\n");
855 (void) fprintf(stderr, " -X --extreme-rewind "
856 "attempt extreme rewind (does not work with dataset)\n");
857 (void) fprintf(stderr, " -Y --all-reconstruction "
858 "attempt all reconstruction combinations for split blocks\n");
859 (void) fprintf(stderr, " -Z --zstd-headers "
860 "show ZSTD headers \n");
861 (void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
862 "to make only that option verbose\n");
863 (void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
864 zdb_exit(2);
865 }
866
867 static void
dump_debug_buffer(void)868 dump_debug_buffer(void)
869 {
870 ssize_t ret __attribute__((unused));
871
872 if (!dump_opt['G'])
873 return;
874 /*
875 * We use write() instead of printf() so that this function
876 * is safe to call from a signal handler.
877 */
878 ret = write(STDERR_FILENO, "\n", 1);
879 zfs_dbgmsg_print(STDERR_FILENO, "zdb");
880 }
881
sig_handler(int signo)882 static void sig_handler(int signo)
883 {
884 struct sigaction action;
885
886 libspl_backtrace(STDERR_FILENO);
887 dump_debug_buffer();
888
889 /*
890 * Restore default action and re-raise signal so SIGSEGV and
891 * SIGABRT can trigger a core dump.
892 */
893 action.sa_handler = SIG_DFL;
894 sigemptyset(&action.sa_mask);
895 action.sa_flags = 0;
896 (void) sigaction(signo, &action, NULL);
897 raise(signo);
898 }
899
900 /*
901 * Called for usage errors that are discovered after a call to spa_open(),
902 * dmu_bonus_hold(), or pool_match(). abort() is called for other errors.
903 */
904
905 static void
fatal(const char * fmt,...)906 fatal(const char *fmt, ...)
907 {
908 va_list ap;
909
910 va_start(ap, fmt);
911 (void) fprintf(stderr, "%s: ", cmdname);
912 (void) vfprintf(stderr, fmt, ap);
913 va_end(ap);
914 (void) fprintf(stderr, "\n");
915
916 dump_debug_buffer();
917
918 zdb_exit(1);
919 }
920
921 static void
dump_packed_nvlist(objset_t * os,uint64_t object,void * data,size_t size)922 dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size)
923 {
924 (void) size;
925 nvlist_t *nv;
926 size_t nvsize = *(uint64_t *)data;
927 char *packed = umem_alloc(nvsize, UMEM_NOFAIL);
928
929 VERIFY0(dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH));
930
931 VERIFY0(nvlist_unpack(packed, nvsize, &nv, 0));
932
933 umem_free(packed, nvsize);
934
935 dump_nvlist(nv, 8);
936
937 nvlist_free(nv);
938 }
939
940 static void
dump_history_offsets(objset_t * os,uint64_t object,void * data,size_t size)941 dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size)
942 {
943 (void) os, (void) object, (void) size;
944 spa_history_phys_t *shp = data;
945
946 if (shp == NULL)
947 return;
948
949 (void) printf("\t\tpool_create_len = %llu\n",
950 (u_longlong_t)shp->sh_pool_create_len);
951 (void) printf("\t\tphys_max_off = %llu\n",
952 (u_longlong_t)shp->sh_phys_max_off);
953 (void) printf("\t\tbof = %llu\n",
954 (u_longlong_t)shp->sh_bof);
955 (void) printf("\t\teof = %llu\n",
956 (u_longlong_t)shp->sh_eof);
957 (void) printf("\t\trecords_lost = %llu\n",
958 (u_longlong_t)shp->sh_records_lost);
959 }
960
961 static void
zdb_nicenum(uint64_t num,char * buf,size_t buflen)962 zdb_nicenum(uint64_t num, char *buf, size_t buflen)
963 {
964 if (dump_opt['P'])
965 (void) snprintf(buf, buflen, "%llu", (longlong_t)num);
966 else
967 nicenum(num, buf, buflen);
968 }
969
970 static void
zdb_nicebytes(uint64_t bytes,char * buf,size_t buflen)971 zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen)
972 {
973 if (dump_opt['P'])
974 (void) snprintf(buf, buflen, "%llu", (longlong_t)bytes);
975 else
976 zfs_nicebytes(bytes, buf, buflen);
977 }
978
979 static const char histo_stars[] = "****************************************";
980 static const uint64_t histo_width = sizeof (histo_stars) - 1;
981
982 static void
dump_histogram(const uint64_t * histo,int size,int offset)983 dump_histogram(const uint64_t *histo, int size, int offset)
984 {
985 int i;
986 int minidx = size - 1;
987 int maxidx = 0;
988 uint64_t max = 0;
989
990 for (i = 0; i < size; i++) {
991 if (histo[i] == 0)
992 continue;
993 if (histo[i] > max)
994 max = histo[i];
995 if (i > maxidx)
996 maxidx = i;
997 if (i < minidx)
998 minidx = i;
999 }
1000
1001 if (max < histo_width)
1002 max = histo_width;
1003
1004 for (i = minidx; i <= maxidx; i++) {
1005 (void) printf("\t\t\t%3u: %6llu %s\n",
1006 i + offset, (u_longlong_t)histo[i],
1007 &histo_stars[(max - histo[i]) * histo_width / max]);
1008 }
1009 }
1010
1011 static void
dump_zap_stats(objset_t * os,uint64_t object)1012 dump_zap_stats(objset_t *os, uint64_t object)
1013 {
1014 int error;
1015 zap_stats_t zs;
1016
1017 error = zap_get_stats(os, object, &zs);
1018 if (error)
1019 return;
1020
1021 if (zs.zs_ptrtbl_len == 0) {
1022 ASSERT(zs.zs_num_blocks == 1);
1023 (void) printf("\tmicrozap: %llu bytes, %llu entries\n",
1024 (u_longlong_t)zs.zs_blocksize,
1025 (u_longlong_t)zs.zs_num_entries);
1026 return;
1027 }
1028
1029 (void) printf("\tFat ZAP stats:\n");
1030
1031 (void) printf("\t\tPointer table:\n");
1032 (void) printf("\t\t\t%llu elements\n",
1033 (u_longlong_t)zs.zs_ptrtbl_len);
1034 (void) printf("\t\t\tzt_blk: %llu\n",
1035 (u_longlong_t)zs.zs_ptrtbl_zt_blk);
1036 (void) printf("\t\t\tzt_numblks: %llu\n",
1037 (u_longlong_t)zs.zs_ptrtbl_zt_numblks);
1038 (void) printf("\t\t\tzt_shift: %llu\n",
1039 (u_longlong_t)zs.zs_ptrtbl_zt_shift);
1040 (void) printf("\t\t\tzt_blks_copied: %llu\n",
1041 (u_longlong_t)zs.zs_ptrtbl_blks_copied);
1042 (void) printf("\t\t\tzt_nextblk: %llu\n",
1043 (u_longlong_t)zs.zs_ptrtbl_nextblk);
1044
1045 (void) printf("\t\tZAP entries: %llu\n",
1046 (u_longlong_t)zs.zs_num_entries);
1047 (void) printf("\t\tLeaf blocks: %llu\n",
1048 (u_longlong_t)zs.zs_num_leafs);
1049 (void) printf("\t\tTotal blocks: %llu\n",
1050 (u_longlong_t)zs.zs_num_blocks);
1051 (void) printf("\t\tzap_block_type: 0x%llx\n",
1052 (u_longlong_t)zs.zs_block_type);
1053 (void) printf("\t\tzap_magic: 0x%llx\n",
1054 (u_longlong_t)zs.zs_magic);
1055 (void) printf("\t\tzap_salt: 0x%llx\n",
1056 (u_longlong_t)zs.zs_salt);
1057
1058 (void) printf("\t\tLeafs with 2^n pointers:\n");
1059 dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);
1060
1061 (void) printf("\t\tBlocks with n*5 entries:\n");
1062 dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);
1063
1064 (void) printf("\t\tBlocks n/10 full:\n");
1065 dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);
1066
1067 (void) printf("\t\tEntries with n chunks:\n");
1068 dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);
1069
1070 (void) printf("\t\tBuckets with n entries:\n");
1071 dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
1072 }
1073
1074 static void
dump_none(objset_t * os,uint64_t object,void * data,size_t size)1075 dump_none(objset_t *os, uint64_t object, void *data, size_t size)
1076 {
1077 (void) os, (void) object, (void) data, (void) size;
1078 }
1079
1080 static void
dump_unknown(objset_t * os,uint64_t object,void * data,size_t size)1081 dump_unknown(objset_t *os, uint64_t object, void *data, size_t size)
1082 {
1083 (void) os, (void) object, (void) data, (void) size;
1084 (void) printf("\tUNKNOWN OBJECT TYPE\n");
1085 }
1086
1087 static void
dump_uint8(objset_t * os,uint64_t object,void * data,size_t size)1088 dump_uint8(objset_t *os, uint64_t object, void *data, size_t size)
1089 {
1090 (void) os, (void) object, (void) data, (void) size;
1091 }
1092
1093 static void
dump_uint64(objset_t * os,uint64_t object,void * data,size_t size)1094 dump_uint64(objset_t *os, uint64_t object, void *data, size_t size)
1095 {
1096 uint64_t *arr;
1097 uint64_t oursize;
1098 if (dump_opt['d'] < 6)
1099 return;
1100
1101 if (data == NULL) {
1102 dmu_object_info_t doi;
1103
1104 VERIFY0(dmu_object_info(os, object, &doi));
1105 size = doi.doi_max_offset;
1106 /*
1107 * We cap the size at 1 mebibyte here to prevent
1108 * allocation failures and nigh-infinite printing if the
1109 * object is extremely large.
1110 */
1111 oursize = MIN(size, 1 << 20);
1112 arr = kmem_alloc(oursize, KM_SLEEP);
1113
1114 int err = dmu_read(os, object, 0, oursize, arr, 0);
1115 if (err != 0) {
1116 (void) printf("got error %u from dmu_read\n", err);
1117 kmem_free(arr, oursize);
1118 return;
1119 }
1120 } else {
1121 /*
1122 * Even though the allocation is already done in this code path,
1123 * we still cap the size to prevent excessive printing.
1124 */
1125 oursize = MIN(size, 1 << 20);
1126 arr = data;
1127 }
1128
1129 if (size == 0) {
1130 if (data == NULL)
1131 kmem_free(arr, oursize);
1132 (void) printf("\t\t[]\n");
1133 return;
1134 }
1135
1136 (void) printf("\t\t[%0llx", (u_longlong_t)arr[0]);
1137 for (size_t i = 1; i * sizeof (uint64_t) < oursize; i++) {
1138 if (i % 4 != 0)
1139 (void) printf(", %0llx", (u_longlong_t)arr[i]);
1140 else
1141 (void) printf(",\n\t\t%0llx", (u_longlong_t)arr[i]);
1142 }
1143 if (oursize != size)
1144 (void) printf(", ... ");
1145 (void) printf("]\n");
1146
1147 if (data == NULL)
1148 kmem_free(arr, oursize);
1149 }
1150
1151 static void
dump_zap(objset_t * os,uint64_t object,void * data,size_t size)1152 dump_zap(objset_t *os, uint64_t object, void *data, size_t size)
1153 {
1154 (void) data, (void) size;
1155 zap_cursor_t zc;
1156 zap_attribute_t *attrp = zap_attribute_long_alloc();
1157 void *prop;
1158 unsigned i;
1159
1160 dump_zap_stats(os, object);
1161 (void) printf("\n");
1162
1163 for (zap_cursor_init(&zc, os, object);
1164 zap_cursor_retrieve(&zc, attrp) == 0;
1165 zap_cursor_advance(&zc)) {
1166 boolean_t key64 =
1167 !!(zap_getflags(zc.zc_zap) & ZAP_FLAG_UINT64_KEY);
1168
1169 if (key64)
1170 (void) printf("\t\t0x%010" PRIu64 "x = ",
1171 *(uint64_t *)attrp->za_name);
1172 else
1173 (void) printf("\t\t%s = ", attrp->za_name);
1174
1175 if (attrp->za_num_integers == 0) {
1176 (void) printf("\n");
1177 continue;
1178 }
1179 prop = umem_zalloc(attrp->za_num_integers *
1180 attrp->za_integer_length, UMEM_NOFAIL);
1181
1182 if (key64)
1183 (void) zap_lookup_uint64(os, object,
1184 (const uint64_t *)attrp->za_name, 1,
1185 attrp->za_integer_length, attrp->za_num_integers,
1186 prop);
1187 else
1188 (void) zap_lookup(os, object, attrp->za_name,
1189 attrp->za_integer_length, attrp->za_num_integers,
1190 prop);
1191
1192 if (attrp->za_integer_length == 1 && !key64) {
1193 if (strcmp(attrp->za_name,
1194 DSL_CRYPTO_KEY_MASTER_KEY) == 0 ||
1195 strcmp(attrp->za_name,
1196 DSL_CRYPTO_KEY_HMAC_KEY) == 0 ||
1197 strcmp(attrp->za_name, DSL_CRYPTO_KEY_IV) == 0 ||
1198 strcmp(attrp->za_name, DSL_CRYPTO_KEY_MAC) == 0 ||
1199 strcmp(attrp->za_name,
1200 DMU_POOL_CHECKSUM_SALT) == 0) {
1201 uint8_t *u8 = prop;
1202
1203 for (i = 0; i < attrp->za_num_integers; i++) {
1204 (void) printf("%02x", u8[i]);
1205 }
1206 } else {
1207 (void) printf("%s", (char *)prop);
1208 }
1209 } else {
1210 for (i = 0; i < attrp->za_num_integers; i++) {
1211 switch (attrp->za_integer_length) {
1212 case 1:
1213 (void) printf("%u ",
1214 ((uint8_t *)prop)[i]);
1215 break;
1216 case 2:
1217 (void) printf("%u ",
1218 ((uint16_t *)prop)[i]);
1219 break;
1220 case 4:
1221 (void) printf("%u ",
1222 ((uint32_t *)prop)[i]);
1223 break;
1224 case 8:
1225 (void) printf("%lld ",
1226 (u_longlong_t)((int64_t *)prop)[i]);
1227 break;
1228 }
1229 }
1230 }
1231 (void) printf("\n");
1232 umem_free(prop,
1233 attrp->za_num_integers * attrp->za_integer_length);
1234 }
1235 zap_cursor_fini(&zc);
1236 zap_attribute_free(attrp);
1237 }
1238
1239 static void
dump_bpobj(objset_t * os,uint64_t object,void * data,size_t size)1240 dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size)
1241 {
1242 bpobj_phys_t *bpop = data;
1243 uint64_t i;
1244 char bytes[32], comp[32], uncomp[32];
1245
1246 /* make sure the output won't get truncated */
1247 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
1248 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
1249 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
1250
1251 if (bpop == NULL)
1252 return;
1253
1254 zdb_nicenum(bpop->bpo_bytes, bytes, sizeof (bytes));
1255 zdb_nicenum(bpop->bpo_comp, comp, sizeof (comp));
1256 zdb_nicenum(bpop->bpo_uncomp, uncomp, sizeof (uncomp));
1257
1258 (void) printf("\t\tnum_blkptrs = %llu\n",
1259 (u_longlong_t)bpop->bpo_num_blkptrs);
1260 (void) printf("\t\tbytes = %s\n", bytes);
1261 if (size >= BPOBJ_SIZE_V1) {
1262 (void) printf("\t\tcomp = %s\n", comp);
1263 (void) printf("\t\tuncomp = %s\n", uncomp);
1264 }
1265 if (size >= BPOBJ_SIZE_V2) {
1266 (void) printf("\t\tsubobjs = %llu\n",
1267 (u_longlong_t)bpop->bpo_subobjs);
1268 (void) printf("\t\tnum_subobjs = %llu\n",
1269 (u_longlong_t)bpop->bpo_num_subobjs);
1270 }
1271 if (size >= sizeof (*bpop)) {
1272 (void) printf("\t\tnum_freed = %llu\n",
1273 (u_longlong_t)bpop->bpo_num_freed);
1274 }
1275
1276 if (dump_opt['d'] < 5)
1277 return;
1278
1279 for (i = 0; i < bpop->bpo_num_blkptrs; i++) {
1280 char blkbuf[BP_SPRINTF_LEN];
1281 blkptr_t bp;
1282
1283 int err = dmu_read(os, object,
1284 i * sizeof (bp), sizeof (bp), &bp, 0);
1285 if (err != 0) {
1286 (void) printf("got error %u from dmu_read\n", err);
1287 break;
1288 }
1289 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp,
1290 BP_GET_FREE(&bp));
1291 (void) printf("\t%s\n", blkbuf);
1292 }
1293 }
1294
1295 static void
dump_bpobj_subobjs(objset_t * os,uint64_t object,void * data,size_t size)1296 dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size)
1297 {
1298 (void) data, (void) size;
1299 dmu_object_info_t doi;
1300 int64_t i;
1301
1302 VERIFY0(dmu_object_info(os, object, &doi));
1303 uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP);
1304
1305 int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0);
1306 if (err != 0) {
1307 (void) printf("got error %u from dmu_read\n", err);
1308 kmem_free(subobjs, doi.doi_max_offset);
1309 return;
1310 }
1311
1312 int64_t last_nonzero = -1;
1313 for (i = 0; i < doi.doi_max_offset / 8; i++) {
1314 if (subobjs[i] != 0)
1315 last_nonzero = i;
1316 }
1317
1318 for (i = 0; i <= last_nonzero; i++) {
1319 (void) printf("\t%llu\n", (u_longlong_t)subobjs[i]);
1320 }
1321 kmem_free(subobjs, doi.doi_max_offset);
1322 }
1323
1324 static void
dump_ddt_zap(objset_t * os,uint64_t object,void * data,size_t size)1325 dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size)
1326 {
1327 (void) data, (void) size;
1328 dump_zap_stats(os, object);
1329 /* contents are printed elsewhere, properly decoded */
1330 }
1331
1332 static void
dump_sa_attrs(objset_t * os,uint64_t object,void * data,size_t size)1333 dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size)
1334 {
1335 (void) data, (void) size;
1336 zap_cursor_t zc;
1337 zap_attribute_t *attrp = zap_attribute_alloc();
1338
1339 dump_zap_stats(os, object);
1340 (void) printf("\n");
1341
1342 for (zap_cursor_init(&zc, os, object);
1343 zap_cursor_retrieve(&zc, attrp) == 0;
1344 zap_cursor_advance(&zc)) {
1345 (void) printf("\t\t%s = ", attrp->za_name);
1346 if (attrp->za_num_integers == 0) {
1347 (void) printf("\n");
1348 continue;
1349 }
1350 (void) printf(" %llx : [%d:%d:%d]\n",
1351 (u_longlong_t)attrp->za_first_integer,
1352 (int)ATTR_LENGTH(attrp->za_first_integer),
1353 (int)ATTR_BSWAP(attrp->za_first_integer),
1354 (int)ATTR_NUM(attrp->za_first_integer));
1355 }
1356 zap_cursor_fini(&zc);
1357 zap_attribute_free(attrp);
1358 }
1359
1360 static void
dump_sa_layouts(objset_t * os,uint64_t object,void * data,size_t size)1361 dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size)
1362 {
1363 (void) data, (void) size;
1364 zap_cursor_t zc;
1365 zap_attribute_t *attrp = zap_attribute_alloc();
1366 uint16_t *layout_attrs;
1367 unsigned i;
1368
1369 dump_zap_stats(os, object);
1370 (void) printf("\n");
1371
1372 for (zap_cursor_init(&zc, os, object);
1373 zap_cursor_retrieve(&zc, attrp) == 0;
1374 zap_cursor_advance(&zc)) {
1375 (void) printf("\t\t%s = [", attrp->za_name);
1376 if (attrp->za_num_integers == 0) {
1377 (void) printf("\n");
1378 continue;
1379 }
1380
1381 VERIFY(attrp->za_integer_length == 2);
1382 layout_attrs = umem_zalloc(attrp->za_num_integers *
1383 attrp->za_integer_length, UMEM_NOFAIL);
1384
1385 VERIFY(zap_lookup(os, object, attrp->za_name,
1386 attrp->za_integer_length,
1387 attrp->za_num_integers, layout_attrs) == 0);
1388
1389 for (i = 0; i != attrp->za_num_integers; i++)
1390 (void) printf(" %d ", (int)layout_attrs[i]);
1391 (void) printf("]\n");
1392 umem_free(layout_attrs,
1393 attrp->za_num_integers * attrp->za_integer_length);
1394 }
1395 zap_cursor_fini(&zc);
1396 zap_attribute_free(attrp);
1397 }
1398
1399 static void
dump_zpldir(objset_t * os,uint64_t object,void * data,size_t size)1400 dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
1401 {
1402 (void) data, (void) size;
1403 zap_cursor_t zc;
1404 zap_attribute_t *attrp = zap_attribute_long_alloc();
1405 const char *typenames[] = {
1406 /* 0 */ "not specified",
1407 /* 1 */ "FIFO",
1408 /* 2 */ "Character Device",
1409 /* 3 */ "3 (invalid)",
1410 /* 4 */ "Directory",
1411 /* 5 */ "5 (invalid)",
1412 /* 6 */ "Block Device",
1413 /* 7 */ "7 (invalid)",
1414 /* 8 */ "Regular File",
1415 /* 9 */ "9 (invalid)",
1416 /* 10 */ "Symbolic Link",
1417 /* 11 */ "11 (invalid)",
1418 /* 12 */ "Socket",
1419 /* 13 */ "Door",
1420 /* 14 */ "Event Port",
1421 /* 15 */ "15 (invalid)",
1422 };
1423
1424 dump_zap_stats(os, object);
1425 (void) printf("\n");
1426
1427 for (zap_cursor_init(&zc, os, object);
1428 zap_cursor_retrieve(&zc, attrp) == 0;
1429 zap_cursor_advance(&zc)) {
1430 (void) printf("\t\t%s = %lld (type: %s)\n",
1431 attrp->za_name, ZFS_DIRENT_OBJ(attrp->za_first_integer),
1432 typenames[ZFS_DIRENT_TYPE(attrp->za_first_integer)]);
1433 }
1434 zap_cursor_fini(&zc);
1435 zap_attribute_free(attrp);
1436 }
1437
1438 static int
get_dtl_refcount(vdev_t * vd)1439 get_dtl_refcount(vdev_t *vd)
1440 {
1441 int refcount = 0;
1442
1443 if (vd->vdev_ops->vdev_op_leaf) {
1444 space_map_t *sm = vd->vdev_dtl_sm;
1445
1446 if (sm != NULL &&
1447 sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1448 return (1);
1449 return (0);
1450 }
1451
1452 for (unsigned c = 0; c < vd->vdev_children; c++)
1453 refcount += get_dtl_refcount(vd->vdev_child[c]);
1454 return (refcount);
1455 }
1456
1457 static int
get_metaslab_refcount(vdev_t * vd)1458 get_metaslab_refcount(vdev_t *vd)
1459 {
1460 int refcount = 0;
1461
1462 if (vd->vdev_top == vd) {
1463 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
1464 space_map_t *sm = vd->vdev_ms[m]->ms_sm;
1465
1466 if (sm != NULL &&
1467 sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1468 refcount++;
1469 }
1470 }
1471 for (unsigned c = 0; c < vd->vdev_children; c++)
1472 refcount += get_metaslab_refcount(vd->vdev_child[c]);
1473
1474 return (refcount);
1475 }
1476
1477 static int
get_obsolete_refcount(vdev_t * vd)1478 get_obsolete_refcount(vdev_t *vd)
1479 {
1480 uint64_t obsolete_sm_object;
1481 int refcount = 0;
1482
1483 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1484 if (vd->vdev_top == vd && obsolete_sm_object != 0) {
1485 dmu_object_info_t doi;
1486 VERIFY0(dmu_object_info(vd->vdev_spa->spa_meta_objset,
1487 obsolete_sm_object, &doi));
1488 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1489 refcount++;
1490 }
1491 } else {
1492 ASSERT0P(vd->vdev_obsolete_sm);
1493 ASSERT0(obsolete_sm_object);
1494 }
1495 for (unsigned c = 0; c < vd->vdev_children; c++) {
1496 refcount += get_obsolete_refcount(vd->vdev_child[c]);
1497 }
1498
1499 return (refcount);
1500 }
1501
1502 static int
get_prev_obsolete_spacemap_refcount(spa_t * spa)1503 get_prev_obsolete_spacemap_refcount(spa_t *spa)
1504 {
1505 uint64_t prev_obj =
1506 spa->spa_condensing_indirect_phys.scip_prev_obsolete_sm_object;
1507 if (prev_obj != 0) {
1508 dmu_object_info_t doi;
1509 VERIFY0(dmu_object_info(spa->spa_meta_objset, prev_obj, &doi));
1510 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1511 return (1);
1512 }
1513 }
1514 return (0);
1515 }
1516
1517 static int
get_checkpoint_refcount(vdev_t * vd)1518 get_checkpoint_refcount(vdev_t *vd)
1519 {
1520 int refcount = 0;
1521
1522 if (vd->vdev_top == vd && vd->vdev_top_zap != 0 &&
1523 zap_contains(spa_meta_objset(vd->vdev_spa),
1524 vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) == 0)
1525 refcount++;
1526
1527 for (uint64_t c = 0; c < vd->vdev_children; c++)
1528 refcount += get_checkpoint_refcount(vd->vdev_child[c]);
1529
1530 return (refcount);
1531 }
1532
1533 static int
get_log_spacemap_refcount(spa_t * spa)1534 get_log_spacemap_refcount(spa_t *spa)
1535 {
1536 return (avl_numnodes(&spa->spa_sm_logs_by_txg));
1537 }
1538
1539 static int
verify_spacemap_refcounts(spa_t * spa)1540 verify_spacemap_refcounts(spa_t *spa)
1541 {
1542 uint64_t expected_refcount = 0;
1543 uint64_t actual_refcount;
1544
1545 (void) feature_get_refcount(spa,
1546 &spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM],
1547 &expected_refcount);
1548 actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
1549 actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);
1550 actual_refcount += get_obsolete_refcount(spa->spa_root_vdev);
1551 actual_refcount += get_prev_obsolete_spacemap_refcount(spa);
1552 actual_refcount += get_checkpoint_refcount(spa->spa_root_vdev);
1553 actual_refcount += get_log_spacemap_refcount(spa);
1554
1555 if (expected_refcount != actual_refcount) {
1556 (void) printf("space map refcount mismatch: expected %lld != "
1557 "actual %lld\n",
1558 (longlong_t)expected_refcount,
1559 (longlong_t)actual_refcount);
1560 return (2);
1561 }
1562 return (0);
1563 }
1564
1565 static void
dump_spacemap(objset_t * os,space_map_t * sm)1566 dump_spacemap(objset_t *os, space_map_t *sm)
1567 {
1568 const char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1569 "INVALID", "INVALID", "INVALID", "INVALID" };
1570
1571 if (sm == NULL)
1572 return;
1573
1574 (void) printf("space map object %llu:\n",
1575 (longlong_t)sm->sm_object);
1576 (void) printf(" smp_length = 0x%llx\n",
1577 (longlong_t)sm->sm_phys->smp_length);
1578 (void) printf(" smp_alloc = 0x%llx\n",
1579 (longlong_t)sm->sm_phys->smp_alloc);
1580
1581 if (dump_opt['d'] < 6 && dump_opt['m'] < 4)
1582 return;
1583
1584 /*
1585 * Print out the freelist entries in both encoded and decoded form.
1586 */
1587 uint8_t mapshift = sm->sm_shift;
1588 int64_t alloc = 0;
1589 uint64_t word, entry_id = 0;
1590 for (uint64_t offset = 0; offset < space_map_length(sm);
1591 offset += sizeof (word)) {
1592
1593 VERIFY0(dmu_read(os, space_map_object(sm), offset,
1594 sizeof (word), &word, DMU_READ_PREFETCH));
1595
1596 if (sm_entry_is_debug(word)) {
1597 uint64_t de_txg = SM_DEBUG_TXG_DECODE(word);
1598 uint64_t de_sync_pass = SM_DEBUG_SYNCPASS_DECODE(word);
1599 if (de_txg == 0) {
1600 (void) printf(
1601 "\t [%6llu] PADDING\n",
1602 (u_longlong_t)entry_id);
1603 } else {
1604 (void) printf(
1605 "\t [%6llu] %s: txg %llu pass %llu\n",
1606 (u_longlong_t)entry_id,
1607 ddata[SM_DEBUG_ACTION_DECODE(word)],
1608 (u_longlong_t)de_txg,
1609 (u_longlong_t)de_sync_pass);
1610 }
1611 entry_id++;
1612 continue;
1613 }
1614
1615 char entry_type;
1616 uint64_t entry_off, entry_run, entry_vdev;
1617
1618 if (sm_entry_is_single_word(word)) {
1619 entry_type = (SM_TYPE_DECODE(word) == SM_ALLOC) ?
1620 'A' : 'F';
1621 entry_off = (SM_OFFSET_DECODE(word) << mapshift) +
1622 sm->sm_start;
1623 entry_run = SM_RUN_DECODE(word) << mapshift;
1624
1625 (void) printf("\t [%6llu] %c "
1626 "range: %012llx-%012llx size: %08llx\n",
1627 (u_longlong_t)entry_id, entry_type,
1628 (u_longlong_t)entry_off,
1629 (u_longlong_t)(entry_off + entry_run - 1),
1630 (u_longlong_t)entry_run);
1631 } else {
1632 /* it is a two-word entry so we read another word */
1633 ASSERT(sm_entry_is_double_word(word));
1634
1635 uint64_t extra_word;
1636 offset += sizeof (extra_word);
1637 ASSERT3U(offset, <, space_map_length(sm));
1638 VERIFY0(dmu_read(os, space_map_object(sm), offset,
1639 sizeof (extra_word), &extra_word,
1640 DMU_READ_PREFETCH));
1641
1642 entry_run = SM2_RUN_DECODE(word) << mapshift;
1643 entry_vdev = SM2_VDEV_DECODE(word);
1644 entry_type = (SM2_TYPE_DECODE(extra_word) == SM_ALLOC) ?
1645 'A' : 'F';
1646 entry_off = (SM2_OFFSET_DECODE(extra_word) <<
1647 mapshift) + sm->sm_start;
1648
1649 if (zopt_metaslab_args == 0 ||
1650 zopt_metaslab[0] == entry_vdev) {
1651 (void) printf("\t [%6llu] %c "
1652 "range: %012llx-%012llx size: %08llx "
1653 "vdev: %llu\n",
1654 (u_longlong_t)entry_id, entry_type,
1655 (u_longlong_t)entry_off,
1656 (u_longlong_t)(entry_off + entry_run - 1),
1657 (u_longlong_t)entry_run,
1658 (u_longlong_t)entry_vdev);
1659 }
1660 }
1661
1662 if (entry_type == 'A')
1663 alloc += entry_run;
1664 else
1665 alloc -= entry_run;
1666 entry_id++;
1667 }
1668 if (alloc != space_map_allocated(sm)) {
1669 (void) printf("space_map_object alloc (%lld) INCONSISTENT "
1670 "with space map summary (%lld)\n",
1671 (longlong_t)space_map_allocated(sm), (longlong_t)alloc);
1672 }
1673 }
1674
1675 static void
dump_metaslab_stats(metaslab_t * msp)1676 dump_metaslab_stats(metaslab_t *msp)
1677 {
1678 char maxbuf[32];
1679 zfs_range_tree_t *rt = msp->ms_allocatable;
1680 zfs_btree_t *t = &msp->ms_allocatable_by_size;
1681 int free_pct = zfs_range_tree_space(rt) * 100 / msp->ms_size;
1682
1683 /* max sure nicenum has enough space */
1684 _Static_assert(sizeof (maxbuf) >= NN_NUMBUF_SZ, "maxbuf truncated");
1685
1686 zdb_nicenum(metaslab_largest_allocatable(msp), maxbuf, sizeof (maxbuf));
1687
1688 (void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
1689 "segments", zfs_btree_numnodes(t), "maxsize", maxbuf,
1690 "freepct", free_pct);
1691 (void) printf("\tIn-memory histogram:\n");
1692 dump_histogram(rt->rt_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1693 }
1694
1695 static void
dump_allocated(void * arg,uint64_t start,uint64_t size)1696 dump_allocated(void *arg, uint64_t start, uint64_t size)
1697 {
1698 uint64_t *off = arg;
1699 if (*off != start)
1700 (void) printf("ALLOC: %"PRIu64" %"PRIu64"\n", *off,
1701 start - *off);
1702 *off = start + size;
1703 }
1704
1705 static void
dump_metaslab(metaslab_t * msp)1706 dump_metaslab(metaslab_t *msp)
1707 {
1708 vdev_t *vd = msp->ms_group->mg_vd;
1709 spa_t *spa = vd->vdev_spa;
1710 space_map_t *sm = msp->ms_sm;
1711 char freebuf[32];
1712
1713 zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf,
1714 sizeof (freebuf));
1715
1716 (void) printf(
1717 "\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
1718 (u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
1719 (u_longlong_t)space_map_object(sm), freebuf);
1720
1721 if (dump_opt[ARG_ALLOCATED] ||
1722 (dump_opt['m'] > 2 && !dump_opt['L'])) {
1723 mutex_enter(&msp->ms_lock);
1724 VERIFY0(metaslab_load(msp));
1725 }
1726
1727 if (dump_opt['m'] > 2 && !dump_opt['L']) {
1728 zfs_range_tree_stat_verify(msp->ms_allocatable);
1729 dump_metaslab_stats(msp);
1730 }
1731
1732 if (dump_opt[ARG_ALLOCATED]) {
1733 uint64_t off = msp->ms_start;
1734 zfs_range_tree_walk(msp->ms_allocatable, dump_allocated,
1735 &off);
1736 if (off != msp->ms_start + msp->ms_size)
1737 (void) printf("ALLOC: %"PRIu64" %"PRIu64"\n", off,
1738 msp->ms_size - off);
1739 }
1740
1741 if (dump_opt['m'] > 1 && sm != NULL &&
1742 spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) {
1743 /*
1744 * The space map histogram represents free space in chunks
1745 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1746 */
1747 (void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1748 (u_longlong_t)msp->ms_fragmentation);
1749 dump_histogram(sm->sm_phys->smp_histogram,
1750 SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift);
1751 }
1752
1753 if (dump_opt[ARG_ALLOCATED] ||
1754 (dump_opt['m'] > 2 && !dump_opt['L'])) {
1755 metaslab_unload(msp);
1756 mutex_exit(&msp->ms_lock);
1757 }
1758
1759 if (vd->vdev_ops == &vdev_draid_ops)
1760 ASSERT3U(msp->ms_size, <=, 1ULL << vd->vdev_ms_shift);
1761 else
1762 ASSERT3U(msp->ms_size, ==, 1ULL << vd->vdev_ms_shift);
1763
1764 dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
1765
1766 if (spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
1767 (void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1768 (u_longlong_t)metaslab_unflushed_txg(msp));
1769 }
1770 }
1771
1772 static void
print_vdev_metaslab_header(vdev_t * vd)1773 print_vdev_metaslab_header(vdev_t *vd)
1774 {
1775 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
1776 const char *bias_str = "";
1777 if (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) {
1778 bias_str = VDEV_ALLOC_BIAS_LOG;
1779 } else if (alloc_bias == VDEV_BIAS_SPECIAL) {
1780 bias_str = VDEV_ALLOC_BIAS_SPECIAL;
1781 } else if (alloc_bias == VDEV_BIAS_DEDUP) {
1782 bias_str = VDEV_ALLOC_BIAS_DEDUP;
1783 }
1784
1785 uint64_t ms_flush_data_obj = 0;
1786 if (vd->vdev_top_zap != 0) {
1787 int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
1788 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
1789 sizeof (uint64_t), 1, &ms_flush_data_obj);
1790 if (error != ENOENT) {
1791 ASSERT0(error);
1792 }
1793 }
1794
1795 (void) printf("\tvdev %10llu\t%s metaslab shift %4llu",
1796 (u_longlong_t)vd->vdev_id, bias_str,
1797 (u_longlong_t)vd->vdev_ms_shift);
1798
1799 if (ms_flush_data_obj != 0) {
1800 (void) printf(" ms_unflushed_phys object %llu",
1801 (u_longlong_t)ms_flush_data_obj);
1802 }
1803
1804 (void) printf("\n\t%-10s%5llu %-19s %-15s %-12s\n",
1805 "metaslabs", (u_longlong_t)vd->vdev_ms_count,
1806 "offset", "spacemap", "free");
1807 (void) printf("\t%15s %19s %15s %12s\n",
1808 "---------------", "-------------------",
1809 "---------------", "------------");
1810 }
1811
1812 static void
dump_metaslab_groups(spa_t * spa,boolean_t show_special)1813 dump_metaslab_groups(spa_t *spa, boolean_t show_special)
1814 {
1815 vdev_t *rvd = spa->spa_root_vdev;
1816 metaslab_class_t *mc = spa_normal_class(spa);
1817 metaslab_class_t *smc = spa_special_class(spa);
1818 uint64_t fragmentation;
1819
1820 metaslab_class_histogram_verify(mc);
1821
1822 for (unsigned c = 0; c < rvd->vdev_children; c++) {
1823 vdev_t *tvd = rvd->vdev_child[c];
1824 metaslab_group_t *mg = tvd->vdev_mg;
1825
1826 if (mg == NULL || (mg->mg_class != mc &&
1827 (!show_special || mg->mg_class != smc)))
1828 continue;
1829
1830 metaslab_group_histogram_verify(mg);
1831 mg->mg_fragmentation = metaslab_group_fragmentation(mg);
1832
1833 (void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1834 "fragmentation",
1835 (u_longlong_t)tvd->vdev_id,
1836 (u_longlong_t)tvd->vdev_ms_count);
1837 if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
1838 (void) printf("%3s\n", "-");
1839 } else {
1840 (void) printf("%3llu%%\n",
1841 (u_longlong_t)mg->mg_fragmentation);
1842 }
1843 dump_histogram(mg->mg_histogram,
1844 ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1845 }
1846
1847 (void) printf("\tpool %s\tfragmentation", spa_name(spa));
1848 fragmentation = metaslab_class_fragmentation(mc);
1849 if (fragmentation == ZFS_FRAG_INVALID)
1850 (void) printf("\t%3s\n", "-");
1851 else
1852 (void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation);
1853 dump_histogram(mc->mc_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1854 }
1855
1856 static void
print_vdev_indirect(vdev_t * vd)1857 print_vdev_indirect(vdev_t *vd)
1858 {
1859 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
1860 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
1861 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
1862
1863 if (vim == NULL) {
1864 ASSERT0P(vib);
1865 return;
1866 }
1867
1868 ASSERT3U(vdev_indirect_mapping_object(vim), ==,
1869 vic->vic_mapping_object);
1870 ASSERT3U(vdev_indirect_births_object(vib), ==,
1871 vic->vic_births_object);
1872
1873 (void) printf("indirect births obj %llu:\n",
1874 (longlong_t)vic->vic_births_object);
1875 (void) printf(" vib_count = %llu\n",
1876 (longlong_t)vdev_indirect_births_count(vib));
1877 for (uint64_t i = 0; i < vdev_indirect_births_count(vib); i++) {
1878 vdev_indirect_birth_entry_phys_t *cur_vibe =
1879 &vib->vib_entries[i];
1880 (void) printf("\toffset %llx -> txg %llu\n",
1881 (longlong_t)cur_vibe->vibe_offset,
1882 (longlong_t)cur_vibe->vibe_phys_birth_txg);
1883 }
1884 (void) printf("\n");
1885
1886 (void) printf("indirect mapping obj %llu:\n",
1887 (longlong_t)vic->vic_mapping_object);
1888 (void) printf(" vim_max_offset = 0x%llx\n",
1889 (longlong_t)vdev_indirect_mapping_max_offset(vim));
1890 (void) printf(" vim_bytes_mapped = 0x%llx\n",
1891 (longlong_t)vdev_indirect_mapping_bytes_mapped(vim));
1892 (void) printf(" vim_count = %llu\n",
1893 (longlong_t)vdev_indirect_mapping_num_entries(vim));
1894
1895 if (dump_opt['d'] <= 5 && dump_opt['m'] <= 3)
1896 return;
1897
1898 uint32_t *counts = vdev_indirect_mapping_load_obsolete_counts(vim);
1899
1900 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
1901 vdev_indirect_mapping_entry_phys_t *vimep =
1902 &vim->vim_entries[i];
1903 (void) printf("\t<%llx:%llx:%llx> -> "
1904 "<%llx:%llx:%llx> (%x obsolete)\n",
1905 (longlong_t)vd->vdev_id,
1906 (longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
1907 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1908 (longlong_t)DVA_GET_VDEV(&vimep->vimep_dst),
1909 (longlong_t)DVA_GET_OFFSET(&vimep->vimep_dst),
1910 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1911 counts[i]);
1912 }
1913 (void) printf("\n");
1914
1915 uint64_t obsolete_sm_object;
1916 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1917 if (obsolete_sm_object != 0) {
1918 objset_t *mos = vd->vdev_spa->spa_meta_objset;
1919 (void) printf("obsolete space map object %llu:\n",
1920 (u_longlong_t)obsolete_sm_object);
1921 ASSERT(vd->vdev_obsolete_sm != NULL);
1922 ASSERT3U(space_map_object(vd->vdev_obsolete_sm), ==,
1923 obsolete_sm_object);
1924 dump_spacemap(mos, vd->vdev_obsolete_sm);
1925 (void) printf("\n");
1926 }
1927 }
1928
1929 static void
dump_metaslabs(spa_t * spa)1930 dump_metaslabs(spa_t *spa)
1931 {
1932 vdev_t *vd, *rvd = spa->spa_root_vdev;
1933 uint64_t m, c = 0, children = rvd->vdev_children;
1934
1935 (void) printf("\nMetaslabs:\n");
1936
1937 if (zopt_metaslab_args > 0) {
1938 c = zopt_metaslab[0];
1939
1940 if (c >= children)
1941 (void) fatal("bad vdev id: %llu", (u_longlong_t)c);
1942
1943 if (zopt_metaslab_args > 1) {
1944 vd = rvd->vdev_child[c];
1945 print_vdev_metaslab_header(vd);
1946
1947 for (m = 1; m < zopt_metaslab_args; m++) {
1948 if (zopt_metaslab[m] < vd->vdev_ms_count)
1949 dump_metaslab(
1950 vd->vdev_ms[zopt_metaslab[m]]);
1951 else
1952 (void) fprintf(stderr, "bad metaslab "
1953 "number %llu\n",
1954 (u_longlong_t)zopt_metaslab[m]);
1955 }
1956 (void) printf("\n");
1957 return;
1958 }
1959 children = c + 1;
1960 }
1961 for (; c < children; c++) {
1962 vd = rvd->vdev_child[c];
1963 print_vdev_metaslab_header(vd);
1964
1965 print_vdev_indirect(vd);
1966
1967 for (m = 0; m < vd->vdev_ms_count; m++)
1968 dump_metaslab(vd->vdev_ms[m]);
1969 (void) printf("\n");
1970 }
1971 }
1972
1973 static void
dump_log_spacemaps(spa_t * spa)1974 dump_log_spacemaps(spa_t *spa)
1975 {
1976 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1977 return;
1978
1979 (void) printf("\nLog Space Maps in Pool:\n");
1980 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
1981 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1982 space_map_t *sm = NULL;
1983 VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
1984 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
1985
1986 (void) printf("Log Spacemap object %llu txg %llu\n",
1987 (u_longlong_t)sls->sls_sm_obj, (u_longlong_t)sls->sls_txg);
1988 dump_spacemap(spa->spa_meta_objset, sm);
1989 space_map_close(sm);
1990 }
1991 (void) printf("\n");
1992 }
1993
1994 static void
dump_ddt_entry(const ddt_t * ddt,const ddt_lightweight_entry_t * ddlwe,uint64_t index)1995 dump_ddt_entry(const ddt_t *ddt, const ddt_lightweight_entry_t *ddlwe,
1996 uint64_t index)
1997 {
1998 const ddt_key_t *ddk = &ddlwe->ddlwe_key;
1999 char blkbuf[BP_SPRINTF_LEN];
2000 blkptr_t blk;
2001 int p;
2002
2003 for (p = 0; p < DDT_NPHYS(ddt); p++) {
2004 const ddt_univ_phys_t *ddp = &ddlwe->ddlwe_phys;
2005 ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p);
2006
2007 if (ddt_phys_birth(ddp, v) == 0)
2008 continue;
2009 ddt_bp_create(ddt->ddt_checksum, ddk, ddp, v, &blk);
2010 snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
2011 (void) printf("index %llx refcnt %llu phys %d %s\n",
2012 (u_longlong_t)index, (u_longlong_t)ddt_phys_refcnt(ddp, v),
2013 p, blkbuf);
2014 }
2015 }
2016
2017 static void
dump_dedup_ratio(const ddt_stat_t * dds)2018 dump_dedup_ratio(const ddt_stat_t *dds)
2019 {
2020 double rL, rP, rD, D, dedup, compress, copies;
2021
2022 if (dds->dds_blocks == 0)
2023 return;
2024
2025 rL = (double)dds->dds_ref_lsize;
2026 rP = (double)dds->dds_ref_psize;
2027 rD = (double)dds->dds_ref_dsize;
2028 D = (double)dds->dds_dsize;
2029
2030 dedup = rD / D;
2031 compress = rL / rP;
2032 copies = rD / rP;
2033
2034 (void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
2035 "dedup * compress / copies = %.2f\n\n",
2036 dedup, compress, copies, dedup * compress / copies);
2037 }
2038
2039 static void
dump_ddt_log(ddt_t * ddt)2040 dump_ddt_log(ddt_t *ddt)
2041 {
2042 if (ddt->ddt_version != DDT_VERSION_FDT ||
2043 !(ddt->ddt_flags & DDT_FLAG_LOG))
2044 return;
2045
2046 for (int n = 0; n < 2; n++) {
2047 ddt_log_t *ddl = &ddt->ddt_log[n];
2048
2049 char flagstr[64] = {0};
2050 if (ddl->ddl_flags > 0) {
2051 flagstr[0] = ' ';
2052 int c = 1;
2053 if (ddl->ddl_flags & DDL_FLAG_FLUSHING)
2054 c += strlcpy(&flagstr[c], " FLUSHING",
2055 sizeof (flagstr) - c);
2056 if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT)
2057 c += strlcpy(&flagstr[c], " CHECKPOINT",
2058 sizeof (flagstr) - c);
2059 if (ddl->ddl_flags &
2060 ~(DDL_FLAG_FLUSHING|DDL_FLAG_CHECKPOINT))
2061 c += strlcpy(&flagstr[c], " UNKNOWN",
2062 sizeof (flagstr) - c);
2063 flagstr[1] = '[';
2064 flagstr[c] = ']';
2065 }
2066
2067 uint64_t count = avl_numnodes(&ddl->ddl_tree);
2068
2069 printf(DMU_POOL_DDT_LOG ": flags=0x%02x%s; obj=%llu; "
2070 "len=%llu; txg=%llu; entries=%llu\n",
2071 zio_checksum_table[ddt->ddt_checksum].ci_name, n,
2072 ddl->ddl_flags, flagstr,
2073 (u_longlong_t)ddl->ddl_object,
2074 (u_longlong_t)ddl->ddl_length,
2075 (u_longlong_t)ddl->ddl_first_txg, (u_longlong_t)count);
2076
2077 if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT) {
2078 const ddt_key_t *ddk = &ddl->ddl_checkpoint;
2079 printf(" checkpoint: "
2080 "%016llx:%016llx:%016llx:%016llx:%016llx\n",
2081 (u_longlong_t)ddk->ddk_cksum.zc_word[0],
2082 (u_longlong_t)ddk->ddk_cksum.zc_word[1],
2083 (u_longlong_t)ddk->ddk_cksum.zc_word[2],
2084 (u_longlong_t)ddk->ddk_cksum.zc_word[3],
2085 (u_longlong_t)ddk->ddk_prop);
2086 }
2087
2088 if (count == 0 || dump_opt['D'] < 4)
2089 continue;
2090
2091 ddt_lightweight_entry_t ddlwe;
2092 uint64_t index = 0;
2093 for (ddt_log_entry_t *ddle = avl_first(&ddl->ddl_tree);
2094 ddle; ddle = AVL_NEXT(&ddl->ddl_tree, ddle)) {
2095 DDT_LOG_ENTRY_TO_LIGHTWEIGHT(ddt, ddle, &ddlwe);
2096 dump_ddt_entry(ddt, &ddlwe, index++);
2097 }
2098 }
2099 }
2100
2101 static void
dump_ddt_object(ddt_t * ddt,ddt_type_t type,ddt_class_t class)2102 dump_ddt_object(ddt_t *ddt, ddt_type_t type, ddt_class_t class)
2103 {
2104 char name[DDT_NAMELEN];
2105 ddt_lightweight_entry_t ddlwe;
2106 uint64_t walk = 0;
2107 dmu_object_info_t doi;
2108 uint64_t count, dspace, mspace;
2109 int error;
2110
2111 error = ddt_object_info(ddt, type, class, &doi);
2112
2113 if (error == ENOENT)
2114 return;
2115 ASSERT0(error);
2116
2117 error = ddt_object_count(ddt, type, class, &count);
2118 ASSERT0(error);
2119 if (count == 0)
2120 return;
2121
2122 dspace = doi.doi_physical_blocks_512 << 9;
2123 mspace = doi.doi_fill_count * doi.doi_data_block_size;
2124
2125 ddt_object_name(ddt, type, class, name);
2126
2127 (void) printf("%s: dspace=%llu; mspace=%llu; entries=%llu\n", name,
2128 (u_longlong_t)dspace, (u_longlong_t)mspace, (u_longlong_t)count);
2129
2130 if (dump_opt['D'] < 3)
2131 return;
2132
2133 (void) printf("%s: object=%llu\n", name,
2134 (u_longlong_t)ddt->ddt_object[type][class]);
2135 zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class]);
2136
2137 if (dump_opt['D'] < 4)
2138 return;
2139
2140 if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE)
2141 return;
2142
2143 (void) printf("%s contents:\n\n", name);
2144
2145 while ((error = ddt_object_walk(ddt, type, class, &walk, &ddlwe)) == 0)
2146 dump_ddt_entry(ddt, &ddlwe, walk);
2147
2148 ASSERT3U(error, ==, ENOENT);
2149
2150 (void) printf("\n");
2151 }
2152
2153 static void
dump_ddt(ddt_t * ddt)2154 dump_ddt(ddt_t *ddt)
2155 {
2156 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
2157 return;
2158
2159 char flagstr[64] = {0};
2160 if (ddt->ddt_flags > 0) {
2161 flagstr[0] = ' ';
2162 int c = 1;
2163 if (ddt->ddt_flags & DDT_FLAG_FLAT)
2164 c += strlcpy(&flagstr[c], " FLAT",
2165 sizeof (flagstr) - c);
2166 if (ddt->ddt_flags & DDT_FLAG_LOG)
2167 c += strlcpy(&flagstr[c], " LOG",
2168 sizeof (flagstr) - c);
2169 if (ddt->ddt_flags & ~DDT_FLAG_MASK)
2170 c += strlcpy(&flagstr[c], " UNKNOWN",
2171 sizeof (flagstr) - c);
2172 flagstr[1] = '[';
2173 flagstr[c] = ']';
2174 }
2175
2176 printf("DDT-%s: version=%llu [%s]; flags=0x%02llx%s; rootobj=%llu\n",
2177 zio_checksum_table[ddt->ddt_checksum].ci_name,
2178 (u_longlong_t)ddt->ddt_version,
2179 (ddt->ddt_version == 0) ? "LEGACY" :
2180 (ddt->ddt_version == 1) ? "FDT" : "UNKNOWN",
2181 (u_longlong_t)ddt->ddt_flags, flagstr,
2182 (u_longlong_t)ddt->ddt_dir_object);
2183
2184 for (ddt_type_t type = 0; type < DDT_TYPES; type++)
2185 for (ddt_class_t class = 0; class < DDT_CLASSES; class++)
2186 dump_ddt_object(ddt, type, class);
2187
2188 dump_ddt_log(ddt);
2189 }
2190
2191 static void
dump_all_ddts(spa_t * spa)2192 dump_all_ddts(spa_t *spa)
2193 {
2194 ddt_histogram_t ddh_total = {{{0}}};
2195 ddt_stat_t dds_total = {0};
2196
2197 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++)
2198 dump_ddt(spa->spa_ddt[c]);
2199
2200 ddt_get_dedup_stats(spa, &dds_total);
2201
2202 if (dds_total.dds_blocks == 0) {
2203 (void) printf("All DDTs are empty\n");
2204 return;
2205 }
2206
2207 (void) printf("\n");
2208
2209 if (dump_opt['D'] > 1) {
2210 (void) printf("DDT histogram (aggregated over all DDTs):\n");
2211 ddt_get_dedup_histogram(spa, &ddh_total);
2212 zpool_dump_ddt(&dds_total, &ddh_total);
2213 }
2214
2215 dump_dedup_ratio(&dds_total);
2216
2217 /*
2218 * Dump a histogram of unique class entry age
2219 */
2220 if (dump_opt['D'] == 3 && getenv("ZDB_DDT_UNIQUE_AGE_HIST") != NULL) {
2221 ddt_age_histo_t histogram;
2222
2223 (void) printf("DDT walk unique, building age histogram...\n");
2224 ddt_prune_walk(spa, 0, &histogram);
2225
2226 /*
2227 * print out histogram for unique entry class birth
2228 */
2229 if (histogram.dah_entries > 0) {
2230 (void) printf("%5s %9s %4s\n",
2231 "age", "blocks", "amnt");
2232 (void) printf("%5s %9s %4s\n",
2233 "-----", "---------", "----");
2234 for (int i = 0; i < HIST_BINS; i++) {
2235 (void) printf("%5d %9d %4d%%\n", 1 << i,
2236 (int)histogram.dah_age_histo[i],
2237 (int)((histogram.dah_age_histo[i] * 100) /
2238 histogram.dah_entries));
2239 }
2240 }
2241 }
2242 }
2243
2244 static void
dump_brt(spa_t * spa)2245 dump_brt(spa_t *spa)
2246 {
2247 if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) {
2248 printf("BRT: unsupported on this pool\n");
2249 return;
2250 }
2251
2252 if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
2253 printf("BRT: empty\n");
2254 return;
2255 }
2256
2257 char count[32], used[32], saved[32];
2258 zdb_nicebytes(brt_get_used(spa), used, sizeof (used));
2259 zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved));
2260 uint64_t ratio = brt_get_ratio(spa);
2261 printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved,
2262 (u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100));
2263
2264 if (dump_opt['T'] < 2)
2265 return;
2266
2267 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2268 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2269 if (!brtvd->bv_initiated) {
2270 printf("BRT: vdev %" PRIu64 ": empty\n", vdevid);
2271 continue;
2272 }
2273
2274 zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count));
2275 zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used));
2276 zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved));
2277 printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n",
2278 vdevid, count, used, saved);
2279 }
2280
2281 if (dump_opt['T'] < 3)
2282 return;
2283
2284 /* -TTT shows a per-vdev histograms; -TTTT shows all entries */
2285 boolean_t do_histo = dump_opt['T'] == 3;
2286
2287 char dva[64];
2288
2289 if (!do_histo)
2290 printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2291
2292 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2293 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2294 if (!brtvd->bv_initiated)
2295 continue;
2296
2297 uint64_t counts[64] = {};
2298
2299 zap_cursor_t zc;
2300 zap_attribute_t *za = zap_attribute_alloc();
2301 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2302 brtvd->bv_mos_entries);
2303 zap_cursor_retrieve(&zc, za) == 0;
2304 zap_cursor_advance(&zc)) {
2305 uint64_t refcnt;
2306 VERIFY0(zap_lookup_uint64(spa->spa_meta_objset,
2307 brtvd->bv_mos_entries,
2308 (const uint64_t *)za->za_name, 1,
2309 za->za_integer_length, za->za_num_integers,
2310 &refcnt));
2311
2312 if (do_histo)
2313 counts[highbit64(refcnt)]++;
2314 else {
2315 uint64_t offset =
2316 *(const uint64_t *)za->za_name;
2317
2318 snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx",
2319 vdevid, (u_longlong_t)offset);
2320 printf("%-16s %-10llu\n", dva,
2321 (u_longlong_t)refcnt);
2322 }
2323 }
2324 zap_cursor_fini(&zc);
2325 zap_attribute_free(za);
2326
2327 if (do_histo) {
2328 printf("\nBRT: vdev %" PRIu64
2329 ": DVAs with 2^n refcnts:\n", vdevid);
2330 dump_histogram(counts, 64, 0);
2331 }
2332 }
2333 }
2334
2335 static void
dump_dtl_seg(void * arg,uint64_t start,uint64_t size)2336 dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
2337 {
2338 char *prefix = arg;
2339
2340 (void) printf("%s [%llu,%llu) length %llu\n",
2341 prefix,
2342 (u_longlong_t)start,
2343 (u_longlong_t)(start + size),
2344 (u_longlong_t)(size));
2345 }
2346
2347 static void
dump_dtl(vdev_t * vd,int indent)2348 dump_dtl(vdev_t *vd, int indent)
2349 {
2350 spa_t *spa = vd->vdev_spa;
2351 boolean_t required;
2352 const char *name[DTL_TYPES] = { "missing", "partial", "scrub",
2353 "outage" };
2354 char prefix[256];
2355
2356 spa_vdev_state_enter(spa, SCL_NONE);
2357 required = vdev_dtl_required(vd);
2358 (void) spa_vdev_state_exit(spa, NULL, 0);
2359
2360 if (indent == 0)
2361 (void) printf("\nDirty time logs:\n\n");
2362
2363 (void) printf("\t%*s%s [%s]\n", indent, "",
2364 vd->vdev_path ? vd->vdev_path :
2365 vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa),
2366 required ? "DTL-required" : "DTL-expendable");
2367
2368 for (int t = 0; t < DTL_TYPES; t++) {
2369 zfs_range_tree_t *rt = vd->vdev_dtl[t];
2370 if (zfs_range_tree_space(rt) == 0)
2371 continue;
2372 (void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
2373 indent + 2, "", name[t]);
2374 zfs_range_tree_walk(rt, dump_dtl_seg, prefix);
2375 if (dump_opt['d'] > 5 && vd->vdev_children == 0)
2376 dump_spacemap(spa->spa_meta_objset,
2377 vd->vdev_dtl_sm);
2378 }
2379
2380 for (unsigned c = 0; c < vd->vdev_children; c++)
2381 dump_dtl(vd->vdev_child[c], indent + 4);
2382 }
2383
2384 static void
dump_history(spa_t * spa)2385 dump_history(spa_t *spa)
2386 {
2387 nvlist_t **events = NULL;
2388 char *buf;
2389 uint64_t resid, len, off = 0;
2390 uint_t num = 0;
2391 int error;
2392 char tbuf[30];
2393
2394 if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) {
2395 (void) fprintf(stderr, "%s: unable to allocate I/O buffer\n",
2396 __func__);
2397 return;
2398 }
2399
2400 do {
2401 len = SPA_OLD_MAXBLOCKSIZE;
2402
2403 if ((error = spa_history_get(spa, &off, &len, buf)) != 0) {
2404 (void) fprintf(stderr, "Unable to read history: "
2405 "error %d\n", error);
2406 free(buf);
2407 return;
2408 }
2409
2410 if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0)
2411 break;
2412
2413 off -= resid;
2414 } while (len != 0);
2415
2416 (void) printf("\nHistory:\n");
2417 for (unsigned i = 0; i < num; i++) {
2418 boolean_t printed = B_FALSE;
2419
2420 if (nvlist_exists(events[i], ZPOOL_HIST_TIME)) {
2421 time_t tsec;
2422 struct tm t;
2423
2424 tsec = fnvlist_lookup_uint64(events[i],
2425 ZPOOL_HIST_TIME);
2426 (void) localtime_r(&tsec, &t);
2427 (void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t);
2428 } else {
2429 tbuf[0] = '\0';
2430 }
2431
2432 if (nvlist_exists(events[i], ZPOOL_HIST_CMD)) {
2433 (void) printf("%s %s\n", tbuf,
2434 fnvlist_lookup_string(events[i], ZPOOL_HIST_CMD));
2435 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_EVENT)) {
2436 uint64_t ievent;
2437
2438 ievent = fnvlist_lookup_uint64(events[i],
2439 ZPOOL_HIST_INT_EVENT);
2440 if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS)
2441 goto next;
2442
2443 (void) printf(" %s [internal %s txg:%ju] %s\n",
2444 tbuf,
2445 zfs_history_event_names[ievent],
2446 fnvlist_lookup_uint64(events[i],
2447 ZPOOL_HIST_TXG),
2448 fnvlist_lookup_string(events[i],
2449 ZPOOL_HIST_INT_STR));
2450 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_NAME)) {
2451 (void) printf("%s [txg:%ju] %s", tbuf,
2452 fnvlist_lookup_uint64(events[i],
2453 ZPOOL_HIST_TXG),
2454 fnvlist_lookup_string(events[i],
2455 ZPOOL_HIST_INT_NAME));
2456
2457 if (nvlist_exists(events[i], ZPOOL_HIST_DSNAME)) {
2458 (void) printf(" %s (%llu)",
2459 fnvlist_lookup_string(events[i],
2460 ZPOOL_HIST_DSNAME),
2461 (u_longlong_t)fnvlist_lookup_uint64(
2462 events[i],
2463 ZPOOL_HIST_DSID));
2464 }
2465
2466 (void) printf(" %s\n", fnvlist_lookup_string(events[i],
2467 ZPOOL_HIST_INT_STR));
2468 } else if (nvlist_exists(events[i], ZPOOL_HIST_IOCTL)) {
2469 (void) printf("%s ioctl %s\n", tbuf,
2470 fnvlist_lookup_string(events[i],
2471 ZPOOL_HIST_IOCTL));
2472
2473 if (nvlist_exists(events[i], ZPOOL_HIST_INPUT_NVL)) {
2474 (void) printf(" input:\n");
2475 dump_nvlist(fnvlist_lookup_nvlist(events[i],
2476 ZPOOL_HIST_INPUT_NVL), 8);
2477 }
2478 if (nvlist_exists(events[i], ZPOOL_HIST_OUTPUT_NVL)) {
2479 (void) printf(" output:\n");
2480 dump_nvlist(fnvlist_lookup_nvlist(events[i],
2481 ZPOOL_HIST_OUTPUT_NVL), 8);
2482 }
2483 if (nvlist_exists(events[i], ZPOOL_HIST_ERRNO)) {
2484 (void) printf(" errno: %lld\n",
2485 (longlong_t)fnvlist_lookup_int64(events[i],
2486 ZPOOL_HIST_ERRNO));
2487 }
2488 } else {
2489 goto next;
2490 }
2491
2492 printed = B_TRUE;
2493 next:
2494 if (dump_opt['h'] > 1) {
2495 if (!printed)
2496 (void) printf("unrecognized record:\n");
2497 dump_nvlist(events[i], 2);
2498 }
2499 }
2500 free(buf);
2501 }
2502
2503 static void
dump_dnode(objset_t * os,uint64_t object,void * data,size_t size)2504 dump_dnode(objset_t *os, uint64_t object, void *data, size_t size)
2505 {
2506 (void) os, (void) object, (void) data, (void) size;
2507 }
2508
2509 static uint64_t
blkid2offset(const dnode_phys_t * dnp,const blkptr_t * bp,const zbookmark_phys_t * zb)2510 blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
2511 const zbookmark_phys_t *zb)
2512 {
2513 if (dnp == NULL) {
2514 ASSERT(zb->zb_level < 0);
2515 if (zb->zb_object == 0)
2516 return (zb->zb_blkid);
2517 return (zb->zb_blkid * BP_GET_LSIZE(bp));
2518 }
2519
2520 ASSERT(zb->zb_level >= 0);
2521
2522 return ((zb->zb_blkid <<
2523 (zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) *
2524 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
2525 }
2526
2527 static void
snprintf_zstd_header(spa_t * spa,char * blkbuf,size_t buflen,const blkptr_t * bp)2528 snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
2529 const blkptr_t *bp)
2530 {
2531 static abd_t *pabd = NULL;
2532 void *buf;
2533 zio_t *zio;
2534 zfs_zstdhdr_t zstd_hdr;
2535 int error;
2536
2537 if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD)
2538 return;
2539
2540 if (BP_IS_HOLE(bp))
2541 return;
2542
2543 if (BP_IS_EMBEDDED(bp)) {
2544 buf = malloc(SPA_MAXBLOCKSIZE);
2545 if (buf == NULL) {
2546 (void) fprintf(stderr, "out of memory\n");
2547 zdb_exit(1);
2548 }
2549 decode_embedded_bp_compressed(bp, buf);
2550 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2551 free(buf);
2552 zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2553 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2554 (void) snprintf(blkbuf + strlen(blkbuf),
2555 buflen - strlen(blkbuf),
2556 " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2557 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2558 zfs_get_hdrlevel(&zstd_hdr));
2559 return;
2560 }
2561
2562 if (!pabd)
2563 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
2564 zio = zio_root(spa, NULL, NULL, 0);
2565
2566 /* Decrypt but don't decompress so we can read the compression header */
2567 zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL,
2568 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS,
2569 NULL));
2570 error = zio_wait(zio);
2571 if (error) {
2572 (void) fprintf(stderr, "read failed: %d\n", error);
2573 return;
2574 }
2575 buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp));
2576 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2577 zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2578 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2579
2580 (void) snprintf(blkbuf + strlen(blkbuf),
2581 buflen - strlen(blkbuf),
2582 " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2583 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2584 zfs_get_hdrlevel(&zstd_hdr));
2585
2586 abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp));
2587 }
2588
2589 static void
snprintf_blkptr_compact(char * blkbuf,size_t buflen,const blkptr_t * bp,boolean_t bp_freed)2590 snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
2591 boolean_t bp_freed)
2592 {
2593 const dva_t *dva = bp->blk_dva;
2594 int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1;
2595 int i;
2596
2597 if (dump_opt['b'] >= 6) {
2598 snprintf_blkptr(blkbuf, buflen, bp);
2599 if (bp_freed) {
2600 (void) snprintf(blkbuf + strlen(blkbuf),
2601 buflen - strlen(blkbuf), " %s", "FREE");
2602 }
2603 return;
2604 }
2605
2606 if (BP_IS_EMBEDDED(bp)) {
2607 (void) sprintf(blkbuf,
2608 "EMBEDDED et=%u %llxL/%llxP B=%llu",
2609 (int)BPE_GET_ETYPE(bp),
2610 (u_longlong_t)BPE_GET_LSIZE(bp),
2611 (u_longlong_t)BPE_GET_PSIZE(bp),
2612 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2613 return;
2614 }
2615
2616 blkbuf[0] = '\0';
2617
2618 for (i = 0; i < ndvas; i++) {
2619 (void) snprintf(blkbuf + strlen(blkbuf),
2620 buflen - strlen(blkbuf), "%llu:%llx:%llx%s ",
2621 (u_longlong_t)DVA_GET_VDEV(&dva[i]),
2622 (u_longlong_t)DVA_GET_OFFSET(&dva[i]),
2623 (u_longlong_t)DVA_GET_ASIZE(&dva[i]),
2624 (DVA_GET_GANG(&dva[i]) ? "G" : ""));
2625 }
2626
2627 if (BP_IS_HOLE(bp)) {
2628 (void) snprintf(blkbuf + strlen(blkbuf),
2629 buflen - strlen(blkbuf),
2630 "%llxL B=%llu",
2631 (u_longlong_t)BP_GET_LSIZE(bp),
2632 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2633 } else {
2634 (void) snprintf(blkbuf + strlen(blkbuf),
2635 buflen - strlen(blkbuf),
2636 "%llxL/%llxP F=%llu B=%llu/%llu",
2637 (u_longlong_t)BP_GET_LSIZE(bp),
2638 (u_longlong_t)BP_GET_PSIZE(bp),
2639 (u_longlong_t)BP_GET_FILL(bp),
2640 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),
2641 (u_longlong_t)BP_GET_PHYSICAL_BIRTH(bp));
2642 if (bp_freed)
2643 (void) snprintf(blkbuf + strlen(blkbuf),
2644 buflen - strlen(blkbuf), " %s", "FREE");
2645 (void) snprintf(blkbuf + strlen(blkbuf),
2646 buflen - strlen(blkbuf),
2647 " cksum=%016llx:%016llx:%016llx:%016llx",
2648 (u_longlong_t)bp->blk_cksum.zc_word[0],
2649 (u_longlong_t)bp->blk_cksum.zc_word[1],
2650 (u_longlong_t)bp->blk_cksum.zc_word[2],
2651 (u_longlong_t)bp->blk_cksum.zc_word[3]);
2652 }
2653 }
2654
2655 static u_longlong_t
print_indirect(spa_t * spa,blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp)2656 print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
2657 const dnode_phys_t *dnp)
2658 {
2659 char blkbuf[BP_SPRINTF_LEN];
2660 u_longlong_t offset;
2661 int l;
2662
2663 offset = (u_longlong_t)blkid2offset(dnp, bp, zb);
2664
2665 (void) printf("%16llx ", offset);
2666
2667 ASSERT(zb->zb_level >= 0);
2668
2669 for (l = dnp->dn_nlevels - 1; l >= -1; l--) {
2670 if (l == zb->zb_level) {
2671 (void) printf("L%llx", (u_longlong_t)zb->zb_level);
2672 } else {
2673 (void) printf(" ");
2674 }
2675 }
2676
2677 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE);
2678 if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD)
2679 snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp);
2680 (void) printf("%s", blkbuf);
2681
2682 if (!BP_IS_EMBEDDED(bp)) {
2683 if (BP_GET_TYPE(bp) != dnp->dn_type) {
2684 (void) printf(" (ERROR: Block pointer type "
2685 "(%llu) does not match dnode type (%hhu))",
2686 BP_GET_TYPE(bp), dnp->dn_type);
2687 corruption_found = B_TRUE;
2688 }
2689 if (BP_GET_LEVEL(bp) != zb->zb_level) {
2690 (void) printf(" (ERROR: Block pointer level "
2691 "(%llu) does not match bookmark level (%lld))",
2692 BP_GET_LEVEL(bp), (longlong_t)zb->zb_level);
2693 corruption_found = B_TRUE;
2694 }
2695 }
2696 (void) printf("\n");
2697
2698 return (offset);
2699 }
2700
2701 static int
visit_indirect(spa_t * spa,const dnode_phys_t * dnp,blkptr_t * bp,const zbookmark_phys_t * zb)2702 visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
2703 blkptr_t *bp, const zbookmark_phys_t *zb)
2704 {
2705 u_longlong_t offset;
2706 int err = 0;
2707
2708 if (BP_GET_BIRTH(bp) == 0)
2709 return (0);
2710
2711 offset = print_indirect(spa, bp, zb, dnp);
2712
2713 if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
2714 arc_flags_t flags = ARC_FLAG_WAIT;
2715 int i;
2716 blkptr_t *cbp;
2717 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
2718 arc_buf_t *buf;
2719 uint64_t fill = 0;
2720 ASSERT(!BP_IS_REDACTED(bp));
2721
2722 err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
2723 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
2724 if (err)
2725 return (err);
2726 ASSERT(buf->b_data);
2727
2728 /* recursively visit blocks below this */
2729 cbp = buf->b_data;
2730 for (i = 0; i < epb; i++, cbp++) {
2731 zbookmark_phys_t czb;
2732
2733 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
2734 zb->zb_level - 1,
2735 zb->zb_blkid * epb + i);
2736 err = visit_indirect(spa, dnp, cbp, &czb);
2737 if (err)
2738 break;
2739 fill += BP_GET_FILL(cbp);
2740 }
2741 if (!err) {
2742 if (fill != BP_GET_FILL(bp)) {
2743 (void) printf("%16llx: Block pointer "
2744 "fill (%llu) does not match calculated "
2745 "value (%llu)\n", offset, BP_GET_FILL(bp),
2746 (u_longlong_t)fill);
2747 corruption_found = B_TRUE;
2748 }
2749 }
2750 arc_buf_destroy(buf, &buf);
2751 }
2752
2753 return (err);
2754 }
2755
2756 static void
dump_indirect(dnode_t * dn)2757 dump_indirect(dnode_t *dn)
2758 {
2759 dnode_phys_t *dnp = dn->dn_phys;
2760 zbookmark_phys_t czb;
2761
2762 (void) printf("Indirect blocks:\n");
2763
2764 SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
2765 dn->dn_object, dnp->dn_nlevels - 1, 0);
2766 for (int j = 0; j < dnp->dn_nblkptr; j++) {
2767 czb.zb_blkid = j;
2768 (void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
2769 &dnp->dn_blkptr[j], &czb);
2770 }
2771
2772 (void) printf("\n");
2773 }
2774
2775 static void
dump_dsl_dir(objset_t * os,uint64_t object,void * data,size_t size)2776 dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size)
2777 {
2778 (void) os, (void) object;
2779 dsl_dir_phys_t *dd = data;
2780 time_t crtime;
2781 char nice[32];
2782
2783 /* make sure nicenum has enough space */
2784 _Static_assert(sizeof (nice) >= NN_NUMBUF_SZ, "nice truncated");
2785
2786 if (dd == NULL)
2787 return;
2788
2789 ASSERT3U(size, >=, sizeof (dsl_dir_phys_t));
2790
2791 crtime = dd->dd_creation_time;
2792 (void) printf("\t\tcreation_time = %s", ctime(&crtime));
2793 (void) printf("\t\thead_dataset_obj = %llu\n",
2794 (u_longlong_t)dd->dd_head_dataset_obj);
2795 (void) printf("\t\tparent_dir_obj = %llu\n",
2796 (u_longlong_t)dd->dd_parent_obj);
2797 (void) printf("\t\torigin_obj = %llu\n",
2798 (u_longlong_t)dd->dd_origin_obj);
2799 (void) printf("\t\tchild_dir_zapobj = %llu\n",
2800 (u_longlong_t)dd->dd_child_dir_zapobj);
2801 zdb_nicenum(dd->dd_used_bytes, nice, sizeof (nice));
2802 (void) printf("\t\tused_bytes = %s\n", nice);
2803 zdb_nicenum(dd->dd_compressed_bytes, nice, sizeof (nice));
2804 (void) printf("\t\tcompressed_bytes = %s\n", nice);
2805 zdb_nicenum(dd->dd_uncompressed_bytes, nice, sizeof (nice));
2806 (void) printf("\t\tuncompressed_bytes = %s\n", nice);
2807 zdb_nicenum(dd->dd_quota, nice, sizeof (nice));
2808 (void) printf("\t\tquota = %s\n", nice);
2809 zdb_nicenum(dd->dd_reserved, nice, sizeof (nice));
2810 (void) printf("\t\treserved = %s\n", nice);
2811 (void) printf("\t\tprops_zapobj = %llu\n",
2812 (u_longlong_t)dd->dd_props_zapobj);
2813 (void) printf("\t\tdeleg_zapobj = %llu\n",
2814 (u_longlong_t)dd->dd_deleg_zapobj);
2815 (void) printf("\t\tflags = %llx\n",
2816 (u_longlong_t)dd->dd_flags);
2817
2818 #define DO(which) \
2819 zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2820 sizeof (nice)); \
2821 (void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2822 DO(HEAD);
2823 DO(SNAP);
2824 DO(CHILD);
2825 DO(CHILD_RSRV);
2826 DO(REFRSRV);
2827 #undef DO
2828 (void) printf("\t\tclones = %llu\n",
2829 (u_longlong_t)dd->dd_clones);
2830 }
2831
2832 static void
dump_dsl_dataset(objset_t * os,uint64_t object,void * data,size_t size)2833 dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
2834 {
2835 (void) os, (void) object;
2836 dsl_dataset_phys_t *ds = data;
2837 time_t crtime;
2838 char used[32], compressed[32], uncompressed[32], unique[32];
2839 char blkbuf[BP_SPRINTF_LEN];
2840
2841 /* make sure nicenum has enough space */
2842 _Static_assert(sizeof (used) >= NN_NUMBUF_SZ, "used truncated");
2843 _Static_assert(sizeof (compressed) >= NN_NUMBUF_SZ,
2844 "compressed truncated");
2845 _Static_assert(sizeof (uncompressed) >= NN_NUMBUF_SZ,
2846 "uncompressed truncated");
2847 _Static_assert(sizeof (unique) >= NN_NUMBUF_SZ, "unique truncated");
2848
2849 if (ds == NULL)
2850 return;
2851
2852 ASSERT(size == sizeof (*ds));
2853 crtime = ds->ds_creation_time;
2854 zdb_nicenum(ds->ds_referenced_bytes, used, sizeof (used));
2855 zdb_nicenum(ds->ds_compressed_bytes, compressed, sizeof (compressed));
2856 zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed,
2857 sizeof (uncompressed));
2858 zdb_nicenum(ds->ds_unique_bytes, unique, sizeof (unique));
2859 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp);
2860
2861 (void) printf("\t\tdir_obj = %llu\n",
2862 (u_longlong_t)ds->ds_dir_obj);
2863 (void) printf("\t\tprev_snap_obj = %llu\n",
2864 (u_longlong_t)ds->ds_prev_snap_obj);
2865 (void) printf("\t\tprev_snap_txg = %llu\n",
2866 (u_longlong_t)ds->ds_prev_snap_txg);
2867 (void) printf("\t\tnext_snap_obj = %llu\n",
2868 (u_longlong_t)ds->ds_next_snap_obj);
2869 (void) printf("\t\tsnapnames_zapobj = %llu\n",
2870 (u_longlong_t)ds->ds_snapnames_zapobj);
2871 (void) printf("\t\tnum_children = %llu\n",
2872 (u_longlong_t)ds->ds_num_children);
2873 (void) printf("\t\tuserrefs_obj = %llu\n",
2874 (u_longlong_t)ds->ds_userrefs_obj);
2875 (void) printf("\t\tcreation_time = %s", ctime(&crtime));
2876 (void) printf("\t\tcreation_txg = %llu\n",
2877 (u_longlong_t)ds->ds_creation_txg);
2878 (void) printf("\t\tdeadlist_obj = %llu\n",
2879 (u_longlong_t)ds->ds_deadlist_obj);
2880 (void) printf("\t\tused_bytes = %s\n", used);
2881 (void) printf("\t\tcompressed_bytes = %s\n", compressed);
2882 (void) printf("\t\tuncompressed_bytes = %s\n", uncompressed);
2883 (void) printf("\t\tunique = %s\n", unique);
2884 (void) printf("\t\tfsid_guid = %llu\n",
2885 (u_longlong_t)ds->ds_fsid_guid);
2886 (void) printf("\t\tguid = %llu\n",
2887 (u_longlong_t)ds->ds_guid);
2888 (void) printf("\t\tflags = %llx\n",
2889 (u_longlong_t)ds->ds_flags);
2890 (void) printf("\t\tnext_clones_obj = %llu\n",
2891 (u_longlong_t)ds->ds_next_clones_obj);
2892 (void) printf("\t\tprops_obj = %llu\n",
2893 (u_longlong_t)ds->ds_props_obj);
2894 (void) printf("\t\tbp = %s\n", blkbuf);
2895 }
2896
2897 static int
dump_bptree_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)2898 dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2899 {
2900 (void) arg, (void) tx;
2901 char blkbuf[BP_SPRINTF_LEN];
2902
2903 if (BP_GET_BIRTH(bp) != 0) {
2904 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
2905 (void) printf("\t%s\n", blkbuf);
2906 }
2907 return (0);
2908 }
2909
2910 static void
dump_bptree(objset_t * os,uint64_t obj,const char * name)2911 dump_bptree(objset_t *os, uint64_t obj, const char *name)
2912 {
2913 char bytes[32];
2914 bptree_phys_t *bt;
2915 dmu_buf_t *db;
2916
2917 /* make sure nicenum has enough space */
2918 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2919
2920 if (dump_opt['d'] < 3)
2921 return;
2922
2923 VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
2924 bt = db->db_data;
2925 zdb_nicenum(bt->bt_bytes, bytes, sizeof (bytes));
2926 (void) printf("\n %s: %llu datasets, %s\n",
2927 name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes);
2928 dmu_buf_rele(db, FTAG);
2929
2930 if (dump_opt['d'] < 5)
2931 return;
2932
2933 (void) printf("\n");
2934
2935 (void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL);
2936 }
2937
2938 static int
dump_bpobj_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)2939 dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
2940 {
2941 (void) arg, (void) tx;
2942 char blkbuf[BP_SPRINTF_LEN];
2943
2944 ASSERT(BP_GET_BIRTH(bp) != 0);
2945 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed);
2946 (void) printf("\t%s\n", blkbuf);
2947 return (0);
2948 }
2949
2950 static void
dump_full_bpobj(bpobj_t * bpo,const char * name,int indent)2951 dump_full_bpobj(bpobj_t *bpo, const char *name, int indent)
2952 {
2953 char bytes[32];
2954 char comp[32];
2955 char uncomp[32];
2956 uint64_t i;
2957
2958 /* make sure nicenum has enough space */
2959 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2960 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
2961 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
2962
2963 if (dump_opt['d'] < 3)
2964 return;
2965
2966 zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes, sizeof (bytes));
2967 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
2968 zdb_nicenum(bpo->bpo_phys->bpo_comp, comp, sizeof (comp));
2969 zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp, sizeof (uncomp));
2970 if (bpo->bpo_havefreed) {
2971 (void) printf(" %*s: object %llu, %llu local "
2972 "blkptrs, %llu freed, %llu subobjs in object %llu, "
2973 "%s (%s/%s comp)\n",
2974 indent * 8, name,
2975 (u_longlong_t)bpo->bpo_object,
2976 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2977 (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2978 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2979 (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2980 bytes, comp, uncomp);
2981 } else {
2982 (void) printf(" %*s: object %llu, %llu local "
2983 "blkptrs, %llu subobjs in object %llu, "
2984 "%s (%s/%s comp)\n",
2985 indent * 8, name,
2986 (u_longlong_t)bpo->bpo_object,
2987 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2988 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2989 (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2990 bytes, comp, uncomp);
2991 }
2992
2993 for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
2994 uint64_t subobj;
2995 bpobj_t subbpo;
2996 int error;
2997 VERIFY0(dmu_read(bpo->bpo_os,
2998 bpo->bpo_phys->bpo_subobjs,
2999 i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3000 error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3001 if (error != 0) {
3002 (void) printf("ERROR %u while trying to open "
3003 "subobj id %llu\n",
3004 error, (u_longlong_t)subobj);
3005 corruption_found = B_TRUE;
3006 continue;
3007 }
3008 dump_full_bpobj(&subbpo, "subobj", indent + 1);
3009 bpobj_close(&subbpo);
3010 }
3011 } else {
3012 if (bpo->bpo_havefreed) {
3013 (void) printf(" %*s: object %llu, %llu blkptrs, "
3014 "%llu freed, %s\n",
3015 indent * 8, name,
3016 (u_longlong_t)bpo->bpo_object,
3017 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
3018 (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
3019 bytes);
3020 } else {
3021 (void) printf(" %*s: object %llu, %llu blkptrs, "
3022 "%s\n",
3023 indent * 8, name,
3024 (u_longlong_t)bpo->bpo_object,
3025 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
3026 bytes);
3027 }
3028 }
3029
3030 if (dump_opt['d'] < 5)
3031 return;
3032
3033
3034 if (indent == 0) {
3035 (void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL);
3036 (void) printf("\n");
3037 }
3038 }
3039
3040 static int
dump_bookmark(dsl_pool_t * dp,char * name,boolean_t print_redact,boolean_t print_list)3041 dump_bookmark(dsl_pool_t *dp, char *name, boolean_t print_redact,
3042 boolean_t print_list)
3043 {
3044 int err = 0;
3045 zfs_bookmark_phys_t prop;
3046 objset_t *mos = dp->dp_spa->spa_meta_objset;
3047 err = dsl_bookmark_lookup(dp, name, NULL, &prop);
3048
3049 if (err != 0) {
3050 return (err);
3051 }
3052
3053 (void) printf("\t#%s: ", strchr(name, '#') + 1);
3054 (void) printf("{guid: %llx creation_txg: %llu creation_time: "
3055 "%llu redaction_obj: %llu}\n", (u_longlong_t)prop.zbm_guid,
3056 (u_longlong_t)prop.zbm_creation_txg,
3057 (u_longlong_t)prop.zbm_creation_time,
3058 (u_longlong_t)prop.zbm_redaction_obj);
3059
3060 IMPLY(print_list, print_redact);
3061 if (!print_redact || prop.zbm_redaction_obj == 0)
3062 return (0);
3063
3064 redaction_list_t *rl;
3065 VERIFY0(dsl_redaction_list_hold_obj(dp,
3066 prop.zbm_redaction_obj, FTAG, &rl));
3067
3068 redaction_list_phys_t *rlp = rl->rl_phys;
3069 (void) printf("\tRedacted:\n\t\tProgress: ");
3070 if (rlp->rlp_last_object != UINT64_MAX ||
3071 rlp->rlp_last_blkid != UINT64_MAX) {
3072 (void) printf("%llu %llu (incomplete)\n",
3073 (u_longlong_t)rlp->rlp_last_object,
3074 (u_longlong_t)rlp->rlp_last_blkid);
3075 } else {
3076 (void) printf("complete\n");
3077 }
3078 (void) printf("\t\tSnapshots: [");
3079 for (unsigned int i = 0; i < rlp->rlp_num_snaps; i++) {
3080 if (i > 0)
3081 (void) printf(", ");
3082 (void) printf("%0llu",
3083 (u_longlong_t)rlp->rlp_snaps[i]);
3084 }
3085 (void) printf("]\n\t\tLength: %llu\n",
3086 (u_longlong_t)rlp->rlp_num_entries);
3087
3088 if (!print_list) {
3089 dsl_redaction_list_rele(rl, FTAG);
3090 return (0);
3091 }
3092
3093 if (rlp->rlp_num_entries == 0) {
3094 dsl_redaction_list_rele(rl, FTAG);
3095 (void) printf("\t\tRedaction List: []\n\n");
3096 return (0);
3097 }
3098
3099 redact_block_phys_t *rbp_buf;
3100 uint64_t size;
3101 dmu_object_info_t doi;
3102
3103 VERIFY0(dmu_object_info(mos, prop.zbm_redaction_obj, &doi));
3104 size = doi.doi_max_offset;
3105 rbp_buf = kmem_alloc(size, KM_SLEEP);
3106
3107 err = dmu_read(mos, prop.zbm_redaction_obj, 0, size,
3108 rbp_buf, 0);
3109 if (err != 0) {
3110 dsl_redaction_list_rele(rl, FTAG);
3111 kmem_free(rbp_buf, size);
3112 return (err);
3113 }
3114
3115 (void) printf("\t\tRedaction List: [{object: %llx, offset: "
3116 "%llx, blksz: %x, count: %llx}",
3117 (u_longlong_t)rbp_buf[0].rbp_object,
3118 (u_longlong_t)rbp_buf[0].rbp_blkid,
3119 (uint_t)(redact_block_get_size(&rbp_buf[0])),
3120 (u_longlong_t)redact_block_get_count(&rbp_buf[0]));
3121
3122 for (size_t i = 1; i < rlp->rlp_num_entries; i++) {
3123 (void) printf(",\n\t\t{object: %llx, offset: %llx, "
3124 "blksz: %x, count: %llx}",
3125 (u_longlong_t)rbp_buf[i].rbp_object,
3126 (u_longlong_t)rbp_buf[i].rbp_blkid,
3127 (uint_t)(redact_block_get_size(&rbp_buf[i])),
3128 (u_longlong_t)redact_block_get_count(&rbp_buf[i]));
3129 }
3130 dsl_redaction_list_rele(rl, FTAG);
3131 kmem_free(rbp_buf, size);
3132 (void) printf("]\n\n");
3133 return (0);
3134 }
3135
3136 static void
dump_bookmarks(objset_t * os,int verbosity)3137 dump_bookmarks(objset_t *os, int verbosity)
3138 {
3139 zap_cursor_t zc;
3140 zap_attribute_t *attrp;
3141 dsl_dataset_t *ds = dmu_objset_ds(os);
3142 dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3143 objset_t *mos = os->os_spa->spa_meta_objset;
3144 if (verbosity < 4)
3145 return;
3146 attrp = zap_attribute_alloc();
3147 dsl_pool_config_enter(dp, FTAG);
3148
3149 for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj);
3150 zap_cursor_retrieve(&zc, attrp) == 0;
3151 zap_cursor_advance(&zc)) {
3152 char osname[ZFS_MAX_DATASET_NAME_LEN];
3153 char buf[ZFS_MAX_DATASET_NAME_LEN];
3154 int len;
3155 dmu_objset_name(os, osname);
3156 len = snprintf(buf, sizeof (buf), "%s#%s", osname,
3157 attrp->za_name);
3158 VERIFY3S(len, <, ZFS_MAX_DATASET_NAME_LEN);
3159 (void) dump_bookmark(dp, buf, verbosity >= 5, verbosity >= 6);
3160 }
3161 zap_cursor_fini(&zc);
3162 dsl_pool_config_exit(dp, FTAG);
3163 zap_attribute_free(attrp);
3164 }
3165
3166 static void
bpobj_count_refd(bpobj_t * bpo)3167 bpobj_count_refd(bpobj_t *bpo)
3168 {
3169 mos_obj_refd(bpo->bpo_object);
3170
3171 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
3172 mos_obj_refd(bpo->bpo_phys->bpo_subobjs);
3173 for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
3174 uint64_t subobj;
3175 bpobj_t subbpo;
3176 int error;
3177 VERIFY0(dmu_read(bpo->bpo_os,
3178 bpo->bpo_phys->bpo_subobjs,
3179 i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3180 error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3181 if (error != 0) {
3182 (void) printf("ERROR %u while trying to open "
3183 "subobj id %llu\n",
3184 error, (u_longlong_t)subobj);
3185 corruption_found = B_TRUE;
3186 continue;
3187 }
3188 bpobj_count_refd(&subbpo);
3189 bpobj_close(&subbpo);
3190 }
3191 }
3192 }
3193
3194 static int
dsl_deadlist_entry_count_refd(void * arg,dsl_deadlist_entry_t * dle)3195 dsl_deadlist_entry_count_refd(void *arg, dsl_deadlist_entry_t *dle)
3196 {
3197 spa_t *spa = arg;
3198 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3199 if (dle->dle_bpobj.bpo_object != empty_bpobj)
3200 bpobj_count_refd(&dle->dle_bpobj);
3201 return (0);
3202 }
3203
3204 static int
dsl_deadlist_entry_dump(void * arg,dsl_deadlist_entry_t * dle)3205 dsl_deadlist_entry_dump(void *arg, dsl_deadlist_entry_t *dle)
3206 {
3207 ASSERT0P(arg);
3208 if (dump_opt['d'] >= 5) {
3209 char buf[128];
3210 (void) snprintf(buf, sizeof (buf),
3211 "mintxg %llu -> obj %llu",
3212 (longlong_t)dle->dle_mintxg,
3213 (longlong_t)dle->dle_bpobj.bpo_object);
3214
3215 dump_full_bpobj(&dle->dle_bpobj, buf, 0);
3216 } else {
3217 (void) printf("mintxg %llu -> obj %llu\n",
3218 (longlong_t)dle->dle_mintxg,
3219 (longlong_t)dle->dle_bpobj.bpo_object);
3220 }
3221 return (0);
3222 }
3223
3224 static void
dump_blkptr_list(dsl_deadlist_t * dl,const char * name)3225 dump_blkptr_list(dsl_deadlist_t *dl, const char *name)
3226 {
3227 char bytes[32];
3228 char comp[32];
3229 char uncomp[32];
3230 char entries[32];
3231 spa_t *spa = dmu_objset_spa(dl->dl_os);
3232 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3233
3234 if (dl->dl_oldfmt) {
3235 if (dl->dl_bpobj.bpo_object != empty_bpobj)
3236 bpobj_count_refd(&dl->dl_bpobj);
3237 } else {
3238 mos_obj_refd(dl->dl_object);
3239 dsl_deadlist_iterate(dl, dsl_deadlist_entry_count_refd, spa);
3240 }
3241
3242 /* make sure nicenum has enough space */
3243 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3244 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
3245 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
3246 _Static_assert(sizeof (entries) >= NN_NUMBUF_SZ, "entries truncated");
3247
3248 if (dump_opt['d'] < 3)
3249 return;
3250
3251 if (dl->dl_oldfmt) {
3252 dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0);
3253 return;
3254 }
3255
3256 zdb_nicenum(dl->dl_phys->dl_used, bytes, sizeof (bytes));
3257 zdb_nicenum(dl->dl_phys->dl_comp, comp, sizeof (comp));
3258 zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp, sizeof (uncomp));
3259 zdb_nicenum(avl_numnodes(&dl->dl_tree), entries, sizeof (entries));
3260 (void) printf("\n %s: %s (%s/%s comp), %s entries\n",
3261 name, bytes, comp, uncomp, entries);
3262
3263 if (dump_opt['d'] < 4)
3264 return;
3265
3266 (void) putchar('\n');
3267
3268 dsl_deadlist_iterate(dl, dsl_deadlist_entry_dump, NULL);
3269 }
3270
3271 static int
verify_dd_livelist(objset_t * os)3272 verify_dd_livelist(objset_t *os)
3273 {
3274 uint64_t ll_used, used, ll_comp, comp, ll_uncomp, uncomp;
3275 dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3276 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
3277
3278 ASSERT(!dmu_objset_is_snapshot(os));
3279 if (!dsl_deadlist_is_open(&dd->dd_livelist))
3280 return (0);
3281
3282 /* Iterate through the livelist to check for duplicates */
3283 dsl_deadlist_iterate(&dd->dd_livelist, sublivelist_verify_lightweight,
3284 NULL);
3285
3286 dsl_pool_config_enter(dp, FTAG);
3287 dsl_deadlist_space(&dd->dd_livelist, &ll_used,
3288 &ll_comp, &ll_uncomp);
3289
3290 dsl_dataset_t *origin_ds;
3291 ASSERT(dsl_pool_config_held(dp));
3292 VERIFY0(dsl_dataset_hold_obj(dp,
3293 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &origin_ds));
3294 VERIFY0(dsl_dataset_space_written(origin_ds, os->os_dsl_dataset,
3295 &used, &comp, &uncomp));
3296 dsl_dataset_rele(origin_ds, FTAG);
3297 dsl_pool_config_exit(dp, FTAG);
3298 /*
3299 * It's possible that the dataset's uncomp space is larger than the
3300 * livelist's because livelists do not track embedded block pointers
3301 */
3302 if (used != ll_used || comp != ll_comp || uncomp < ll_uncomp) {
3303 char nice_used[32], nice_comp[32], nice_uncomp[32];
3304 (void) printf("Discrepancy in space accounting:\n");
3305 zdb_nicenum(used, nice_used, sizeof (nice_used));
3306 zdb_nicenum(comp, nice_comp, sizeof (nice_comp));
3307 zdb_nicenum(uncomp, nice_uncomp, sizeof (nice_uncomp));
3308 (void) printf("dir: used %s, comp %s, uncomp %s\n",
3309 nice_used, nice_comp, nice_uncomp);
3310 zdb_nicenum(ll_used, nice_used, sizeof (nice_used));
3311 zdb_nicenum(ll_comp, nice_comp, sizeof (nice_comp));
3312 zdb_nicenum(ll_uncomp, nice_uncomp, sizeof (nice_uncomp));
3313 (void) printf("livelist: used %s, comp %s, uncomp %s\n",
3314 nice_used, nice_comp, nice_uncomp);
3315 return (1);
3316 }
3317 return (0);
3318 }
3319
3320 static char *key_material = NULL;
3321
3322 static boolean_t
zdb_derive_key(dsl_dir_t * dd,uint8_t * key_out)3323 zdb_derive_key(dsl_dir_t *dd, uint8_t *key_out)
3324 {
3325 uint64_t keyformat, salt, iters;
3326 int i;
3327 unsigned char c;
3328 FILE *f;
3329
3330 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3331 zfs_prop_to_name(ZFS_PROP_KEYFORMAT), sizeof (uint64_t),
3332 1, &keyformat));
3333
3334 switch (keyformat) {
3335 case ZFS_KEYFORMAT_HEX:
3336 for (i = 0; i < WRAPPING_KEY_LEN * 2; i += 2) {
3337 if (!isxdigit(key_material[i]) ||
3338 !isxdigit(key_material[i+1]))
3339 return (B_FALSE);
3340 if (sscanf(&key_material[i], "%02hhx", &c) != 1)
3341 return (B_FALSE);
3342 key_out[i / 2] = c;
3343 }
3344 break;
3345
3346 case ZFS_KEYFORMAT_PASSPHRASE:
3347 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3348 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT),
3349 sizeof (uint64_t), 1, &salt));
3350 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3351 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS),
3352 sizeof (uint64_t), 1, &iters));
3353
3354 if (PKCS5_PBKDF2_HMAC_SHA1(key_material, strlen(key_material),
3355 ((uint8_t *)&salt), sizeof (uint64_t), iters,
3356 WRAPPING_KEY_LEN, key_out) != 1)
3357 return (B_FALSE);
3358
3359 break;
3360
3361 case ZFS_KEYFORMAT_RAW:
3362 if ((f = fopen(key_material, "r")) == NULL)
3363 return (B_FALSE);
3364
3365 if (fread(key_out, 1, WRAPPING_KEY_LEN, f) !=
3366 WRAPPING_KEY_LEN) {
3367 (void) fclose(f);
3368 return (B_FALSE);
3369 }
3370
3371 /* Check the key length */
3372 if (fgetc(f) != EOF) {
3373 (void) fclose(f);
3374 return (B_FALSE);
3375 }
3376
3377 (void) fclose(f);
3378 break;
3379
3380 default:
3381 fatal("no support for key format %u\n",
3382 (unsigned int) keyformat);
3383 }
3384
3385 return (B_TRUE);
3386 }
3387
3388 static char encroot[ZFS_MAX_DATASET_NAME_LEN];
3389 static boolean_t key_loaded = B_FALSE;
3390
3391 static void
zdb_load_key(objset_t * os)3392 zdb_load_key(objset_t *os)
3393 {
3394 dsl_pool_t *dp;
3395 dsl_dir_t *dd, *rdd;
3396 uint8_t key[WRAPPING_KEY_LEN];
3397 uint64_t rddobj;
3398 int err;
3399
3400 dp = spa_get_dsl(os->os_spa);
3401 dd = os->os_dsl_dataset->ds_dir;
3402
3403 dsl_pool_config_enter(dp, FTAG);
3404 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3405 DSL_CRYPTO_KEY_ROOT_DDOBJ, sizeof (uint64_t), 1, &rddobj));
3406 VERIFY0(dsl_dir_hold_obj(dd->dd_pool, rddobj, NULL, FTAG, &rdd));
3407 dsl_dir_name(rdd, encroot);
3408 dsl_dir_rele(rdd, FTAG);
3409
3410 if (!zdb_derive_key(dd, key))
3411 fatal("couldn't derive encryption key");
3412
3413 dsl_pool_config_exit(dp, FTAG);
3414
3415 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_UNAVAILABLE);
3416
3417 dsl_crypto_params_t *dcp;
3418 nvlist_t *crypto_args;
3419
3420 crypto_args = fnvlist_alloc();
3421 fnvlist_add_uint8_array(crypto_args, "wkeydata",
3422 (uint8_t *)key, WRAPPING_KEY_LEN);
3423 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE,
3424 NULL, crypto_args, &dcp));
3425 err = spa_keystore_load_wkey(encroot, dcp, B_FALSE);
3426
3427 dsl_crypto_params_free(dcp, (err != 0));
3428 fnvlist_free(crypto_args);
3429
3430 if (err != 0)
3431 fatal(
3432 "couldn't load encryption key for %s: %s",
3433 encroot, err == ZFS_ERR_CRYPTO_NOTSUP ?
3434 "crypto params not supported" : strerror(err));
3435
3436 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_AVAILABLE);
3437
3438 printf("Unlocked encryption root: %s\n", encroot);
3439 key_loaded = B_TRUE;
3440 }
3441
3442 static void
zdb_unload_key(void)3443 zdb_unload_key(void)
3444 {
3445 if (!key_loaded)
3446 return;
3447
3448 VERIFY0(spa_keystore_unload_wkey(encroot));
3449 key_loaded = B_FALSE;
3450 }
3451
3452 static avl_tree_t idx_tree;
3453 static avl_tree_t domain_tree;
3454 static boolean_t fuid_table_loaded;
3455 static objset_t *sa_os = NULL;
3456 static sa_attr_type_t *sa_attr_table = NULL;
3457
3458 static int
open_objset(const char * path,const void * tag,objset_t ** osp)3459 open_objset(const char *path, const void *tag, objset_t **osp)
3460 {
3461 int err;
3462 uint64_t sa_attrs = 0;
3463 uint64_t version = 0;
3464
3465 VERIFY0P(sa_os);
3466
3467 /*
3468 * We can't own an objset if it's redacted. Therefore, we do this
3469 * dance: hold the objset, then acquire a long hold on its dataset, then
3470 * release the pool (which is held as part of holding the objset).
3471 */
3472
3473 if (dump_opt['K']) {
3474 /* decryption requested, try to load keys */
3475 err = dmu_objset_hold(path, tag, osp);
3476 if (err != 0) {
3477 (void) fprintf(stderr, "failed to hold dataset "
3478 "'%s': %s\n",
3479 path, strerror(err));
3480 return (err);
3481 }
3482 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3483 dsl_pool_rele(dmu_objset_pool(*osp), tag);
3484
3485 /* succeeds or dies */
3486 zdb_load_key(*osp);
3487
3488 /* release it all */
3489 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3490 dsl_dataset_rele(dmu_objset_ds(*osp), tag);
3491 }
3492
3493 int ds_hold_flags = key_loaded ? DS_HOLD_FLAG_DECRYPT : 0;
3494
3495 err = dmu_objset_hold_flags(path, ds_hold_flags, tag, osp);
3496 if (err != 0) {
3497 (void) fprintf(stderr, "failed to hold dataset '%s': %s\n",
3498 path, strerror(err));
3499 return (err);
3500 }
3501 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3502 dsl_pool_rele(dmu_objset_pool(*osp), tag);
3503
3504 if (dmu_objset_type(*osp) == DMU_OST_ZFS &&
3505 (key_loaded || !(*osp)->os_encrypted)) {
3506 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR,
3507 8, 1, &version);
3508 if (version >= ZPL_VERSION_SA) {
3509 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS,
3510 8, 1, &sa_attrs);
3511 }
3512 err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END,
3513 &sa_attr_table);
3514 if (err != 0) {
3515 (void) fprintf(stderr, "sa_setup failed: %s\n",
3516 strerror(err));
3517 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3518 dsl_dataset_rele_flags(dmu_objset_ds(*osp),
3519 ds_hold_flags, tag);
3520 *osp = NULL;
3521 }
3522 }
3523 sa_os = *osp;
3524
3525 return (err);
3526 }
3527
3528 static void
close_objset(objset_t * os,const void * tag)3529 close_objset(objset_t *os, const void *tag)
3530 {
3531 VERIFY3P(os, ==, sa_os);
3532 if (os->os_sa != NULL)
3533 sa_tear_down(os);
3534 dsl_dataset_long_rele(dmu_objset_ds(os), tag);
3535 dsl_dataset_rele_flags(dmu_objset_ds(os),
3536 key_loaded ? DS_HOLD_FLAG_DECRYPT : 0, tag);
3537 sa_attr_table = NULL;
3538 sa_os = NULL;
3539
3540 zdb_unload_key();
3541 }
3542
3543 static void
fuid_table_destroy(void)3544 fuid_table_destroy(void)
3545 {
3546 if (fuid_table_loaded) {
3547 zfs_fuid_table_destroy(&idx_tree, &domain_tree);
3548 fuid_table_loaded = B_FALSE;
3549 }
3550 }
3551
3552 /*
3553 * Clean up DDT internal state. ddt_lookup() adds entries to ddt_tree, which on
3554 * a live pool are normally cleaned up during ddt_sync(). We can't do that (and
3555 * wouldn't want to anyway), but if we don't clean up the presence of stuff on
3556 * ddt_tree will trip asserts in ddt_table_free(). So, we clean up ourselves.
3557 *
3558 * Note that this is not a particularly efficient way to do this, but
3559 * ddt_remove() is the only public method that can do the work we need, and it
3560 * requires the right locks and etc to do the job. This is only ever called
3561 * during zdb shutdown so efficiency is not especially important.
3562 */
3563 static void
zdb_ddt_cleanup(spa_t * spa)3564 zdb_ddt_cleanup(spa_t *spa)
3565 {
3566 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
3567 ddt_t *ddt = spa->spa_ddt[c];
3568 if (!ddt)
3569 continue;
3570
3571 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3572 ddt_enter(ddt);
3573 ddt_entry_t *dde = avl_first(&ddt->ddt_tree), *next;
3574 while (dde) {
3575 next = AVL_NEXT(&ddt->ddt_tree, dde);
3576 dde->dde_io = NULL;
3577 ddt_remove(ddt, dde);
3578 dde = next;
3579 }
3580 ddt_exit(ddt);
3581 spa_config_exit(spa, SCL_CONFIG, FTAG);
3582 }
3583 }
3584
3585 static void
zdb_exit(int reason)3586 zdb_exit(int reason)
3587 {
3588 if (spa != NULL)
3589 zdb_ddt_cleanup(spa);
3590
3591 if (os != NULL) {
3592 close_objset(os, FTAG);
3593 } else if (spa != NULL) {
3594 spa_close(spa, FTAG);
3595 }
3596
3597 fuid_table_destroy();
3598
3599 if (kernel_init_done)
3600 kernel_fini();
3601
3602 exit(reason);
3603 }
3604
3605 /*
3606 * print uid or gid information.
3607 * For normal POSIX id just the id is printed in decimal format.
3608 * For CIFS files with FUID the fuid is printed in hex followed by
3609 * the domain-rid string.
3610 */
3611 static void
print_idstr(uint64_t id,const char * id_type)3612 print_idstr(uint64_t id, const char *id_type)
3613 {
3614 if (FUID_INDEX(id)) {
3615 const char *domain =
3616 zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id));
3617 (void) printf("\t%s %llx [%s-%d]\n", id_type,
3618 (u_longlong_t)id, domain, (int)FUID_RID(id));
3619 } else {
3620 (void) printf("\t%s %llu\n", id_type, (u_longlong_t)id);
3621 }
3622
3623 }
3624
3625 static void
dump_uidgid(objset_t * os,uint64_t uid,uint64_t gid)3626 dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid)
3627 {
3628 uint32_t uid_idx, gid_idx;
3629
3630 uid_idx = FUID_INDEX(uid);
3631 gid_idx = FUID_INDEX(gid);
3632
3633 /* Load domain table, if not already loaded */
3634 if (!fuid_table_loaded && (uid_idx || gid_idx)) {
3635 uint64_t fuid_obj;
3636
3637 /* first find the fuid object. It lives in the master node */
3638 VERIFY0(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES,
3639 8, 1, &fuid_obj));
3640 zfs_fuid_avl_tree_create(&idx_tree, &domain_tree);
3641 (void) zfs_fuid_table_load(os, fuid_obj,
3642 &idx_tree, &domain_tree);
3643 fuid_table_loaded = B_TRUE;
3644 }
3645
3646 print_idstr(uid, "uid");
3647 print_idstr(gid, "gid");
3648 }
3649
3650 static void
dump_znode_sa_xattr(sa_handle_t * hdl)3651 dump_znode_sa_xattr(sa_handle_t *hdl)
3652 {
3653 nvlist_t *sa_xattr;
3654 nvpair_t *elem = NULL;
3655 int sa_xattr_size = 0;
3656 int sa_xattr_entries = 0;
3657 int error;
3658 char *sa_xattr_packed;
3659
3660 error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size);
3661 if (error || sa_xattr_size == 0)
3662 return;
3663
3664 sa_xattr_packed = malloc(sa_xattr_size);
3665 if (sa_xattr_packed == NULL)
3666 return;
3667
3668 error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR],
3669 sa_xattr_packed, sa_xattr_size);
3670 if (error) {
3671 free(sa_xattr_packed);
3672 return;
3673 }
3674
3675 error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0);
3676 if (error) {
3677 free(sa_xattr_packed);
3678 return;
3679 }
3680
3681 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL)
3682 sa_xattr_entries++;
3683
3684 (void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3685 sa_xattr_size, sa_xattr_entries);
3686 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) {
3687 boolean_t can_print = !dump_opt['P'];
3688 uchar_t *value;
3689 uint_t cnt, idx;
3690
3691 (void) printf("\t\t%s = ", nvpair_name(elem));
3692 nvpair_value_byte_array(elem, &value, &cnt);
3693
3694 for (idx = 0; idx < cnt; ++idx) {
3695 if (!isprint(value[idx])) {
3696 can_print = B_FALSE;
3697 break;
3698 }
3699 }
3700
3701 for (idx = 0; idx < cnt; ++idx) {
3702 if (can_print)
3703 (void) putchar(value[idx]);
3704 else
3705 (void) printf("\\%3.3o", value[idx]);
3706 }
3707 (void) putchar('\n');
3708 }
3709
3710 nvlist_free(sa_xattr);
3711 free(sa_xattr_packed);
3712 }
3713
3714 static void
dump_znode_symlink(sa_handle_t * hdl)3715 dump_znode_symlink(sa_handle_t *hdl)
3716 {
3717 int sa_symlink_size = 0;
3718 char linktarget[MAXPATHLEN];
3719 int error;
3720
3721 error = sa_size(hdl, sa_attr_table[ZPL_SYMLINK], &sa_symlink_size);
3722 if (error || sa_symlink_size == 0) {
3723 return;
3724 }
3725 if (sa_symlink_size >= sizeof (linktarget)) {
3726 (void) printf("symlink size %d is too large\n",
3727 sa_symlink_size);
3728 return;
3729 }
3730 linktarget[sa_symlink_size] = '\0';
3731 if (sa_lookup(hdl, sa_attr_table[ZPL_SYMLINK],
3732 &linktarget, sa_symlink_size) == 0)
3733 (void) printf("\ttarget %s\n", linktarget);
3734 }
3735
3736 static void
dump_znode(objset_t * os,uint64_t object,void * data,size_t size)3737 dump_znode(objset_t *os, uint64_t object, void *data, size_t size)
3738 {
3739 (void) data, (void) size;
3740 char path[MAXPATHLEN * 2]; /* allow for xattr and failure prefix */
3741 sa_handle_t *hdl;
3742 uint64_t xattr, rdev, gen;
3743 uint64_t uid, gid, mode, fsize, parent, links;
3744 uint64_t pflags;
3745 uint64_t acctm[2], modtm[2], chgtm[2], crtm[2];
3746 time_t z_crtime, z_atime, z_mtime, z_ctime;
3747 sa_bulk_attr_t bulk[12];
3748 int idx = 0;
3749 int error;
3750
3751 VERIFY3P(os, ==, sa_os);
3752 if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) {
3753 (void) printf("Failed to get handle for SA znode\n");
3754 return;
3755 }
3756
3757 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8);
3758 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8);
3759 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL,
3760 &links, 8);
3761 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8);
3762 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL,
3763 &mode, 8);
3764 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT],
3765 NULL, &parent, 8);
3766 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL,
3767 &fsize, 8);
3768 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL,
3769 acctm, 16);
3770 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL,
3771 modtm, 16);
3772 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL,
3773 crtm, 16);
3774 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL,
3775 chgtm, 16);
3776 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL,
3777 &pflags, 8);
3778
3779 if (sa_bulk_lookup(hdl, bulk, idx)) {
3780 (void) sa_handle_destroy(hdl);
3781 return;
3782 }
3783
3784 z_crtime = (time_t)crtm[0];
3785 z_atime = (time_t)acctm[0];
3786 z_mtime = (time_t)modtm[0];
3787 z_ctime = (time_t)chgtm[0];
3788
3789 if (dump_opt['d'] > 4) {
3790 error = zfs_obj_to_path(os, object, path, sizeof (path));
3791 if (error == ESTALE) {
3792 (void) snprintf(path, sizeof (path), "on delete queue");
3793 } else if (error != 0) {
3794 leaked_objects++;
3795 (void) snprintf(path, sizeof (path),
3796 "path not found, possibly leaked");
3797 }
3798 (void) printf("\tpath %s\n", path);
3799 }
3800
3801 if (S_ISLNK(mode))
3802 dump_znode_symlink(hdl);
3803 dump_uidgid(os, uid, gid);
3804 (void) printf("\tatime %s", ctime(&z_atime));
3805 (void) printf("\tmtime %s", ctime(&z_mtime));
3806 (void) printf("\tctime %s", ctime(&z_ctime));
3807 (void) printf("\tcrtime %s", ctime(&z_crtime));
3808 (void) printf("\tgen %llu\n", (u_longlong_t)gen);
3809 (void) printf("\tmode %llo\n", (u_longlong_t)mode);
3810 (void) printf("\tsize %llu\n", (u_longlong_t)fsize);
3811 (void) printf("\tparent %llu\n", (u_longlong_t)parent);
3812 (void) printf("\tlinks %llu\n", (u_longlong_t)links);
3813 (void) printf("\tpflags %llx\n", (u_longlong_t)pflags);
3814 if (dmu_objset_projectquota_enabled(os) && (pflags & ZFS_PROJID)) {
3815 uint64_t projid;
3816
3817 if (sa_lookup(hdl, sa_attr_table[ZPL_PROJID], &projid,
3818 sizeof (uint64_t)) == 0)
3819 (void) printf("\tprojid %llu\n", (u_longlong_t)projid);
3820 }
3821 if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr,
3822 sizeof (uint64_t)) == 0)
3823 (void) printf("\txattr %llu\n", (u_longlong_t)xattr);
3824 if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev,
3825 sizeof (uint64_t)) == 0)
3826 (void) printf("\trdev 0x%016llx\n", (u_longlong_t)rdev);
3827 dump_znode_sa_xattr(hdl);
3828 sa_handle_destroy(hdl);
3829 }
3830
3831 static void
dump_acl(objset_t * os,uint64_t object,void * data,size_t size)3832 dump_acl(objset_t *os, uint64_t object, void *data, size_t size)
3833 {
3834 (void) os, (void) object, (void) data, (void) size;
3835 }
3836
3837 static void
dump_dmu_objset(objset_t * os,uint64_t object,void * data,size_t size)3838 dump_dmu_objset(objset_t *os, uint64_t object, void *data, size_t size)
3839 {
3840 (void) os, (void) object, (void) data, (void) size;
3841 }
3842
3843 static object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = {
3844 dump_none, /* unallocated */
3845 dump_zap, /* object directory */
3846 dump_uint64, /* object array */
3847 dump_none, /* packed nvlist */
3848 dump_packed_nvlist, /* packed nvlist size */
3849 dump_none, /* bpobj */
3850 dump_bpobj, /* bpobj header */
3851 dump_none, /* SPA space map header */
3852 dump_none, /* SPA space map */
3853 dump_none, /* ZIL intent log */
3854 dump_dnode, /* DMU dnode */
3855 dump_dmu_objset, /* DMU objset */
3856 dump_dsl_dir, /* DSL directory */
3857 dump_zap, /* DSL directory child map */
3858 dump_zap, /* DSL dataset snap map */
3859 dump_zap, /* DSL props */
3860 dump_dsl_dataset, /* DSL dataset */
3861 dump_znode, /* ZFS znode */
3862 dump_acl, /* ZFS V0 ACL */
3863 dump_uint8, /* ZFS plain file */
3864 dump_zpldir, /* ZFS directory */
3865 dump_zap, /* ZFS master node */
3866 dump_zap, /* ZFS delete queue */
3867 dump_uint8, /* zvol object */
3868 dump_zap, /* zvol prop */
3869 dump_uint8, /* other uint8[] */
3870 dump_uint64, /* other uint64[] */
3871 dump_zap, /* other ZAP */
3872 dump_zap, /* persistent error log */
3873 dump_uint8, /* SPA history */
3874 dump_history_offsets, /* SPA history offsets */
3875 dump_zap, /* Pool properties */
3876 dump_zap, /* DSL permissions */
3877 dump_acl, /* ZFS ACL */
3878 dump_uint8, /* ZFS SYSACL */
3879 dump_none, /* FUID nvlist */
3880 dump_packed_nvlist, /* FUID nvlist size */
3881 dump_zap, /* DSL dataset next clones */
3882 dump_zap, /* DSL scrub queue */
3883 dump_zap, /* ZFS user/group/project used */
3884 dump_zap, /* ZFS user/group/project quota */
3885 dump_zap, /* snapshot refcount tags */
3886 dump_ddt_zap, /* DDT ZAP object */
3887 dump_zap, /* DDT statistics */
3888 dump_znode, /* SA object */
3889 dump_zap, /* SA Master Node */
3890 dump_sa_attrs, /* SA attribute registration */
3891 dump_sa_layouts, /* SA attribute layouts */
3892 dump_zap, /* DSL scrub translations */
3893 dump_none, /* fake dedup BP */
3894 dump_zap, /* deadlist */
3895 dump_none, /* deadlist hdr */
3896 dump_zap, /* dsl clones */
3897 dump_bpobj_subobjs, /* bpobj subobjs */
3898 dump_unknown, /* Unknown type, must be last */
3899 };
3900
3901 static boolean_t
match_object_type(dmu_object_type_t obj_type,uint64_t flags)3902 match_object_type(dmu_object_type_t obj_type, uint64_t flags)
3903 {
3904 boolean_t match = B_TRUE;
3905
3906 switch (obj_type) {
3907 case DMU_OT_DIRECTORY_CONTENTS:
3908 if (!(flags & ZOR_FLAG_DIRECTORY))
3909 match = B_FALSE;
3910 break;
3911 case DMU_OT_PLAIN_FILE_CONTENTS:
3912 if (!(flags & ZOR_FLAG_PLAIN_FILE))
3913 match = B_FALSE;
3914 break;
3915 case DMU_OT_SPACE_MAP:
3916 if (!(flags & ZOR_FLAG_SPACE_MAP))
3917 match = B_FALSE;
3918 break;
3919 default:
3920 if (strcmp(zdb_ot_name(obj_type), "zap") == 0) {
3921 if (!(flags & ZOR_FLAG_ZAP))
3922 match = B_FALSE;
3923 break;
3924 }
3925
3926 /*
3927 * If all bits except some of the supported flags are
3928 * set, the user combined the all-types flag (A) with
3929 * a negated flag to exclude some types (e.g. A-f to
3930 * show all object types except plain files).
3931 */
3932 if ((flags | ZOR_SUPPORTED_FLAGS) != ZOR_FLAG_ALL_TYPES)
3933 match = B_FALSE;
3934
3935 break;
3936 }
3937
3938 return (match);
3939 }
3940
3941 static void
dump_object(objset_t * os,uint64_t object,int verbosity,boolean_t * print_header,uint64_t * dnode_slots_used,uint64_t flags)3942 dump_object(objset_t *os, uint64_t object, int verbosity,
3943 boolean_t *print_header, uint64_t *dnode_slots_used, uint64_t flags)
3944 {
3945 dmu_buf_t *db = NULL;
3946 dmu_object_info_t doi;
3947 dnode_t *dn;
3948 boolean_t dnode_held = B_FALSE;
3949 void *bonus = NULL;
3950 size_t bsize = 0;
3951 char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
3952 char bonus_size[32];
3953 char aux[50];
3954 int error;
3955
3956 /* make sure nicenum has enough space */
3957 _Static_assert(sizeof (iblk) >= NN_NUMBUF_SZ, "iblk truncated");
3958 _Static_assert(sizeof (dblk) >= NN_NUMBUF_SZ, "dblk truncated");
3959 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, "lsize truncated");
3960 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, "asize truncated");
3961 _Static_assert(sizeof (bonus_size) >= NN_NUMBUF_SZ,
3962 "bonus_size truncated");
3963
3964 if (*print_header) {
3965 (void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n",
3966 "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3967 "lsize", "%full", "type");
3968 *print_header = 0;
3969 }
3970
3971 if (object == 0) {
3972 dn = DMU_META_DNODE(os);
3973 dmu_object_info_from_dnode(dn, &doi);
3974 } else {
3975 /*
3976 * Encrypted datasets will have sensitive bonus buffers
3977 * encrypted. Therefore we cannot hold the bonus buffer and
3978 * must hold the dnode itself instead.
3979 */
3980 error = dmu_object_info(os, object, &doi);
3981 if (error)
3982 fatal("dmu_object_info() failed, errno %u", error);
3983
3984 if (!key_loaded && os->os_encrypted &&
3985 DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) {
3986 error = dnode_hold(os, object, FTAG, &dn);
3987 if (error)
3988 fatal("dnode_hold() failed, errno %u", error);
3989 dnode_held = B_TRUE;
3990 } else {
3991 error = dmu_bonus_hold(os, object, FTAG, &db);
3992 if (error)
3993 fatal("dmu_bonus_hold(%llu) failed, errno %u",
3994 object, error);
3995 bonus = db->db_data;
3996 bsize = db->db_size;
3997 dn = DB_DNODE((dmu_buf_impl_t *)db);
3998 }
3999 }
4000
4001 /*
4002 * Default to showing all object types if no flags were specified.
4003 */
4004 if (flags != 0 && flags != ZOR_FLAG_ALL_TYPES &&
4005 !match_object_type(doi.doi_type, flags))
4006 goto out;
4007
4008 if (dnode_slots_used)
4009 *dnode_slots_used = doi.doi_dnodesize / DNODE_MIN_SIZE;
4010
4011 zdb_nicenum(doi.doi_metadata_block_size, iblk, sizeof (iblk));
4012 zdb_nicenum(doi.doi_data_block_size, dblk, sizeof (dblk));
4013 zdb_nicenum(doi.doi_max_offset, lsize, sizeof (lsize));
4014 zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize, sizeof (asize));
4015 zdb_nicenum(doi.doi_bonus_size, bonus_size, sizeof (bonus_size));
4016 zdb_nicenum(doi.doi_dnodesize, dnsize, sizeof (dnsize));
4017 (void) snprintf(fill, sizeof (fill), "%6.2f", 100.0 *
4018 doi.doi_fill_count * doi.doi_data_block_size / (object == 0 ?
4019 DNODES_PER_BLOCK : 1) / doi.doi_max_offset);
4020
4021 aux[0] = '\0';
4022
4023 if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) {
4024 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
4025 " (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum));
4026 }
4027
4028 if (doi.doi_compress == ZIO_COMPRESS_INHERIT &&
4029 ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) {
4030 const char *compname = NULL;
4031 if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION,
4032 ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel),
4033 &compname) == 0) {
4034 (void) snprintf(aux + strlen(aux),
4035 sizeof (aux) - strlen(aux), " (Z=inherit=%s)",
4036 compname);
4037 } else {
4038 (void) snprintf(aux + strlen(aux),
4039 sizeof (aux) - strlen(aux),
4040 " (Z=inherit=%s-unknown)",
4041 ZDB_COMPRESS_NAME(os->os_compress));
4042 }
4043 } else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) {
4044 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
4045 " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress));
4046 } else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
4047 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
4048 " (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress));
4049 }
4050
4051 (void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n",
4052 (u_longlong_t)object, doi.doi_indirection, iblk, dblk,
4053 asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux);
4054
4055 if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
4056 (void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n",
4057 "", "", "", "", "", "", bonus_size, "bonus",
4058 zdb_ot_name(doi.doi_bonus_type));
4059 }
4060
4061 if (verbosity >= 4) {
4062 (void) printf("\tdnode flags: %s%s%s%s\n",
4063 (dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ?
4064 "USED_BYTES " : "",
4065 (dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ?
4066 "USERUSED_ACCOUNTED " : "",
4067 (dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ?
4068 "USEROBJUSED_ACCOUNTED " : "",
4069 (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ?
4070 "SPILL_BLKPTR" : "");
4071 (void) printf("\tdnode maxblkid: %llu\n",
4072 (longlong_t)dn->dn_phys->dn_maxblkid);
4073
4074 if (!dnode_held) {
4075 object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os,
4076 object, bonus, bsize);
4077 } else {
4078 (void) printf("\t\t(bonus encrypted)\n");
4079 }
4080
4081 if (key_loaded ||
4082 (!os->os_encrypted || !DMU_OT_IS_ENCRYPTED(doi.doi_type))) {
4083 object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object,
4084 NULL, 0);
4085 } else {
4086 (void) printf("\t\t(object encrypted)\n");
4087 }
4088
4089 *print_header = B_TRUE;
4090 }
4091
4092 if (verbosity >= 5) {
4093 if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
4094 char blkbuf[BP_SPRINTF_LEN];
4095 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
4096 DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE);
4097 (void) printf("\nSpill block: %s\n", blkbuf);
4098 }
4099 dump_indirect(dn);
4100 }
4101
4102 if (verbosity >= 5) {
4103 /*
4104 * Report the list of segments that comprise the object.
4105 */
4106 uint64_t start = 0;
4107 uint64_t end;
4108 uint64_t blkfill = 1;
4109 int minlvl = 1;
4110
4111 if (dn->dn_type == DMU_OT_DNODE) {
4112 minlvl = 0;
4113 blkfill = DNODES_PER_BLOCK;
4114 }
4115
4116 for (;;) {
4117 char segsize[32];
4118 /* make sure nicenum has enough space */
4119 _Static_assert(sizeof (segsize) >= NN_NUMBUF_SZ,
4120 "segsize truncated");
4121 error = dnode_next_offset(dn,
4122 0, &start, minlvl, blkfill, 0);
4123 if (error)
4124 break;
4125 end = start;
4126 error = dnode_next_offset(dn,
4127 DNODE_FIND_HOLE, &end, minlvl, blkfill, 0);
4128 zdb_nicenum(end - start, segsize, sizeof (segsize));
4129 (void) printf("\t\tsegment [%016llx, %016llx)"
4130 " size %5s\n", (u_longlong_t)start,
4131 (u_longlong_t)end, segsize);
4132 if (error)
4133 break;
4134 start = end;
4135 }
4136 }
4137
4138 out:
4139 if (db != NULL)
4140 dmu_buf_rele(db, FTAG);
4141 if (dnode_held)
4142 dnode_rele(dn, FTAG);
4143 }
4144
4145 static void
count_dir_mos_objects(dsl_dir_t * dd)4146 count_dir_mos_objects(dsl_dir_t *dd)
4147 {
4148 mos_obj_refd(dd->dd_object);
4149 mos_obj_refd(dsl_dir_phys(dd)->dd_child_dir_zapobj);
4150 mos_obj_refd(dsl_dir_phys(dd)->dd_deleg_zapobj);
4151 mos_obj_refd(dsl_dir_phys(dd)->dd_props_zapobj);
4152 mos_obj_refd(dsl_dir_phys(dd)->dd_clones);
4153
4154 /*
4155 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
4156 * Ignore the references after the first one.
4157 */
4158 mos_obj_refd_multiple(dd->dd_crypto_obj);
4159 }
4160
4161 static void
count_ds_mos_objects(dsl_dataset_t * ds)4162 count_ds_mos_objects(dsl_dataset_t *ds)
4163 {
4164 mos_obj_refd(ds->ds_object);
4165 mos_obj_refd(dsl_dataset_phys(ds)->ds_next_clones_obj);
4166 mos_obj_refd(dsl_dataset_phys(ds)->ds_props_obj);
4167 mos_obj_refd(dsl_dataset_phys(ds)->ds_userrefs_obj);
4168 mos_obj_refd(dsl_dataset_phys(ds)->ds_snapnames_zapobj);
4169 mos_obj_refd(ds->ds_bookmarks_obj);
4170
4171 if (!dsl_dataset_is_snapshot(ds)) {
4172 count_dir_mos_objects(ds->ds_dir);
4173 }
4174 }
4175
4176 static const char *const objset_types[DMU_OST_NUMTYPES] = {
4177 "NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
4178
4179 /*
4180 * Parse a string denoting a range of object IDs of the form
4181 * <start>[:<end>[:flags]], and store the results in zor.
4182 * Return 0 on success. On error, return 1 and update the msg
4183 * pointer to point to a descriptive error message.
4184 */
4185 static int
parse_object_range(char * range,zopt_object_range_t * zor,const char ** msg)4186 parse_object_range(char *range, zopt_object_range_t *zor, const char **msg)
4187 {
4188 uint64_t flags = 0;
4189 char *p, *s, *dup, *flagstr, *tmp = NULL;
4190 size_t len;
4191 int i;
4192 int rc = 0;
4193
4194 if (strchr(range, ':') == NULL) {
4195 zor->zor_obj_start = strtoull(range, &p, 0);
4196 if (*p != '\0') {
4197 *msg = "Invalid characters in object ID";
4198 rc = 1;
4199 }
4200 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4201 zor->zor_obj_end = zor->zor_obj_start;
4202 return (rc);
4203 }
4204
4205 if (strchr(range, ':') == range) {
4206 *msg = "Invalid leading colon";
4207 rc = 1;
4208 return (rc);
4209 }
4210
4211 len = strlen(range);
4212 if (range[len - 1] == ':') {
4213 *msg = "Invalid trailing colon";
4214 rc = 1;
4215 return (rc);
4216 }
4217
4218 dup = strdup(range);
4219 s = strtok_r(dup, ":", &tmp);
4220 zor->zor_obj_start = strtoull(s, &p, 0);
4221
4222 if (*p != '\0') {
4223 *msg = "Invalid characters in start object ID";
4224 rc = 1;
4225 goto out;
4226 }
4227
4228 s = strtok_r(NULL, ":", &tmp);
4229 zor->zor_obj_end = strtoull(s, &p, 0);
4230
4231 if (*p != '\0') {
4232 *msg = "Invalid characters in end object ID";
4233 rc = 1;
4234 goto out;
4235 }
4236
4237 if (zor->zor_obj_start > zor->zor_obj_end) {
4238 *msg = "Start object ID may not exceed end object ID";
4239 rc = 1;
4240 goto out;
4241 }
4242
4243 s = strtok_r(NULL, ":", &tmp);
4244 if (s == NULL) {
4245 zor->zor_flags = ZOR_FLAG_ALL_TYPES;
4246 goto out;
4247 } else if (strtok_r(NULL, ":", &tmp) != NULL) {
4248 *msg = "Invalid colon-delimited field after flags";
4249 rc = 1;
4250 goto out;
4251 }
4252
4253 flagstr = s;
4254 for (i = 0; flagstr[i]; i++) {
4255 int bit;
4256 boolean_t negation = (flagstr[i] == '-');
4257
4258 if (negation) {
4259 i++;
4260 if (flagstr[i] == '\0') {
4261 *msg = "Invalid trailing negation operator";
4262 rc = 1;
4263 goto out;
4264 }
4265 }
4266 bit = flagbits[(uchar_t)flagstr[i]];
4267 if (bit == 0) {
4268 *msg = "Invalid flag";
4269 rc = 1;
4270 goto out;
4271 }
4272 if (negation)
4273 flags &= ~bit;
4274 else
4275 flags |= bit;
4276 }
4277 zor->zor_flags = flags;
4278
4279 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4280 zor->zor_obj_end = ZDB_MAP_OBJECT_ID(zor->zor_obj_end);
4281
4282 out:
4283 free(dup);
4284 return (rc);
4285 }
4286
4287 static void
dump_objset(objset_t * os)4288 dump_objset(objset_t *os)
4289 {
4290 dmu_objset_stats_t dds = { 0 };
4291 uint64_t object, object_count;
4292 uint64_t refdbytes, usedobjs, scratch;
4293 char numbuf[32];
4294 char blkbuf[BP_SPRINTF_LEN + 20];
4295 char osname[ZFS_MAX_DATASET_NAME_LEN];
4296 const char *type = "UNKNOWN";
4297 int verbosity = dump_opt['d'];
4298 boolean_t print_header;
4299 unsigned i;
4300 int error;
4301 uint64_t total_slots_used = 0;
4302 uint64_t max_slot_used = 0;
4303 uint64_t dnode_slots;
4304 uint64_t obj_start;
4305 uint64_t obj_end;
4306 uint64_t flags;
4307
4308 /* make sure nicenum has enough space */
4309 _Static_assert(sizeof (numbuf) >= NN_NUMBUF_SZ, "numbuf truncated");
4310
4311 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
4312 dmu_objset_fast_stat(os, &dds);
4313 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
4314
4315 print_header = B_TRUE;
4316
4317 if (dds.dds_type < DMU_OST_NUMTYPES)
4318 type = objset_types[dds.dds_type];
4319
4320 if (dds.dds_type == DMU_OST_META) {
4321 dds.dds_creation_txg = TXG_INITIAL;
4322 usedobjs = BP_GET_FILL(os->os_rootbp);
4323 refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)->
4324 dd_used_bytes;
4325 } else {
4326 dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
4327 }
4328
4329 ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
4330
4331 zdb_nicenum(refdbytes, numbuf, sizeof (numbuf));
4332
4333 if (verbosity >= 4) {
4334 (void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp ");
4335 (void) snprintf_blkptr(blkbuf + strlen(blkbuf),
4336 sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp);
4337 } else {
4338 blkbuf[0] = '\0';
4339 }
4340
4341 dmu_objset_name(os, osname);
4342
4343 (void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
4344 "%s, %llu objects%s%s\n",
4345 osname, type, (u_longlong_t)dmu_objset_id(os),
4346 (u_longlong_t)dds.dds_creation_txg,
4347 numbuf, (u_longlong_t)usedobjs, blkbuf,
4348 (dds.dds_inconsistent) ? " (inconsistent)" : "");
4349
4350 for (i = 0; i < zopt_object_args; i++) {
4351 obj_start = zopt_object_ranges[i].zor_obj_start;
4352 obj_end = zopt_object_ranges[i].zor_obj_end;
4353 flags = zopt_object_ranges[i].zor_flags;
4354
4355 object = obj_start;
4356 if (object == 0 || obj_start == obj_end)
4357 dump_object(os, object, verbosity, &print_header, NULL,
4358 flags);
4359 else
4360 object--;
4361
4362 while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) &&
4363 object <= obj_end) {
4364 dump_object(os, object, verbosity, &print_header, NULL,
4365 flags);
4366 }
4367 }
4368
4369 if (zopt_object_args > 0) {
4370 (void) printf("\n");
4371 return;
4372 }
4373
4374 if (dump_opt['i'] != 0 || verbosity >= 2)
4375 dump_intent_log(dmu_objset_zil(os));
4376
4377 if (dmu_objset_ds(os) != NULL) {
4378 dsl_dataset_t *ds = dmu_objset_ds(os);
4379 dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
4380 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
4381 !dmu_objset_is_snapshot(os)) {
4382 dump_blkptr_list(&ds->ds_dir->dd_livelist, "Livelist");
4383 if (verify_dd_livelist(os) != 0)
4384 fatal("livelist is incorrect");
4385 }
4386
4387 if (dsl_dataset_remap_deadlist_exists(ds)) {
4388 (void) printf("ds_remap_deadlist:\n");
4389 dump_blkptr_list(&ds->ds_remap_deadlist, "Deadlist");
4390 }
4391 count_ds_mos_objects(ds);
4392 }
4393
4394 if (dmu_objset_ds(os) != NULL)
4395 dump_bookmarks(os, verbosity);
4396
4397 if (verbosity < 2)
4398 return;
4399
4400 if (BP_IS_HOLE(os->os_rootbp))
4401 return;
4402
4403 dump_object(os, 0, verbosity, &print_header, NULL, 0);
4404 object_count = 0;
4405 if (DMU_USERUSED_DNODE(os) != NULL &&
4406 DMU_USERUSED_DNODE(os)->dn_type != 0) {
4407 dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header,
4408 NULL, 0);
4409 dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header,
4410 NULL, 0);
4411 }
4412
4413 if (DMU_PROJECTUSED_DNODE(os) != NULL &&
4414 DMU_PROJECTUSED_DNODE(os)->dn_type != 0)
4415 dump_object(os, DMU_PROJECTUSED_OBJECT, verbosity,
4416 &print_header, NULL, 0);
4417
4418 object = 0;
4419 while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
4420 dump_object(os, object, verbosity, &print_header, &dnode_slots,
4421 0);
4422 object_count++;
4423 total_slots_used += dnode_slots;
4424 max_slot_used = object + dnode_slots - 1;
4425 }
4426
4427 (void) printf("\n");
4428
4429 (void) printf(" Dnode slots:\n");
4430 (void) printf("\tTotal used: %10llu\n",
4431 (u_longlong_t)total_slots_used);
4432 (void) printf("\tMax used: %10llu\n",
4433 (u_longlong_t)max_slot_used);
4434 (void) printf("\tPercent empty: %10lf\n",
4435 (double)(max_slot_used - total_slots_used)*100 /
4436 (double)max_slot_used);
4437 (void) printf("\n");
4438
4439 if (error != ESRCH) {
4440 (void) fprintf(stderr, "dmu_object_next() = %d\n", error);
4441 abort();
4442 }
4443
4444 ASSERT3U(object_count, ==, usedobjs);
4445
4446 if (leaked_objects != 0) {
4447 (void) printf("%d potentially leaked objects detected\n",
4448 leaked_objects);
4449 leaked_objects = 0;
4450 }
4451 }
4452
4453 static void
dump_uberblock(uberblock_t * ub,const char * header,const char * footer)4454 dump_uberblock(uberblock_t *ub, const char *header, const char *footer)
4455 {
4456 time_t timestamp = ub->ub_timestamp;
4457
4458 (void) printf("%s", header ? header : "");
4459 (void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic);
4460 (void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version);
4461 (void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg);
4462 (void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum);
4463 (void) printf("\ttimestamp = %llu UTC = %s",
4464 (u_longlong_t)ub->ub_timestamp, ctime(×tamp));
4465
4466 char blkbuf[BP_SPRINTF_LEN];
4467 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4468 (void) printf("\tbp = %s\n", blkbuf);
4469
4470 (void) printf("\tmmp_magic = %016llx\n",
4471 (u_longlong_t)ub->ub_mmp_magic);
4472 if (MMP_VALID(ub)) {
4473 (void) printf("\tmmp_delay = %0llu\n",
4474 (u_longlong_t)ub->ub_mmp_delay);
4475 if (MMP_SEQ_VALID(ub))
4476 (void) printf("\tmmp_seq = %u\n",
4477 (unsigned int) MMP_SEQ(ub));
4478 if (MMP_FAIL_INT_VALID(ub))
4479 (void) printf("\tmmp_fail = %u\n",
4480 (unsigned int) MMP_FAIL_INT(ub));
4481 if (MMP_INTERVAL_VALID(ub))
4482 (void) printf("\tmmp_write = %u\n",
4483 (unsigned int) MMP_INTERVAL(ub));
4484 /* After MMP_* to make summarize_uberblock_mmp cleaner */
4485 (void) printf("\tmmp_valid = %x\n",
4486 (unsigned int) ub->ub_mmp_config & 0xFF);
4487 }
4488
4489 if (dump_opt['u'] >= 4) {
4490 char blkbuf[BP_SPRINTF_LEN];
4491 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4492 (void) printf("\trootbp = %s\n", blkbuf);
4493 }
4494 (void) printf("\tcheckpoint_txg = %llu\n",
4495 (u_longlong_t)ub->ub_checkpoint_txg);
4496
4497 (void) printf("\traidz_reflow state=%u off=%llu\n",
4498 (int)RRSS_GET_STATE(ub),
4499 (u_longlong_t)RRSS_GET_OFFSET(ub));
4500
4501 (void) printf("%s", footer ? footer : "");
4502 }
4503
4504 static void
dump_config(spa_t * spa)4505 dump_config(spa_t *spa)
4506 {
4507 dmu_buf_t *db;
4508 size_t nvsize = 0;
4509 int error = 0;
4510
4511
4512 error = dmu_bonus_hold(spa->spa_meta_objset,
4513 spa->spa_config_object, FTAG, &db);
4514
4515 if (error == 0) {
4516 nvsize = *(uint64_t *)db->db_data;
4517 dmu_buf_rele(db, FTAG);
4518
4519 (void) printf("\nMOS Configuration:\n");
4520 dump_packed_nvlist(spa->spa_meta_objset,
4521 spa->spa_config_object, (void *)&nvsize, 1);
4522 } else {
4523 (void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d",
4524 (u_longlong_t)spa->spa_config_object, error);
4525 }
4526 }
4527
4528 static void
dump_cachefile(const char * cachefile)4529 dump_cachefile(const char *cachefile)
4530 {
4531 int fd;
4532 struct stat64 statbuf;
4533 char *buf;
4534 nvlist_t *config;
4535
4536 if ((fd = open64(cachefile, O_RDONLY)) < 0) {
4537 (void) printf("cannot open '%s': %s\n", cachefile,
4538 strerror(errno));
4539 zdb_exit(1);
4540 }
4541
4542 if (fstat64(fd, &statbuf) != 0) {
4543 (void) printf("failed to stat '%s': %s\n", cachefile,
4544 strerror(errno));
4545 zdb_exit(1);
4546 }
4547
4548 if ((buf = malloc(statbuf.st_size)) == NULL) {
4549 (void) fprintf(stderr, "failed to allocate %llu bytes\n",
4550 (u_longlong_t)statbuf.st_size);
4551 zdb_exit(1);
4552 }
4553
4554 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
4555 (void) fprintf(stderr, "failed to read %llu bytes\n",
4556 (u_longlong_t)statbuf.st_size);
4557 zdb_exit(1);
4558 }
4559
4560 (void) close(fd);
4561
4562 if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) {
4563 (void) fprintf(stderr, "failed to unpack nvlist\n");
4564 zdb_exit(1);
4565 }
4566
4567 free(buf);
4568
4569 dump_nvlist(config, 0);
4570
4571 nvlist_free(config);
4572 }
4573
4574 /*
4575 * ZFS label nvlist stats
4576 */
4577 typedef struct zdb_nvl_stats {
4578 int zns_list_count;
4579 int zns_leaf_count;
4580 size_t zns_leaf_largest;
4581 size_t zns_leaf_total;
4582 nvlist_t *zns_string;
4583 nvlist_t *zns_uint64;
4584 nvlist_t *zns_boolean;
4585 } zdb_nvl_stats_t;
4586
4587 static void
collect_nvlist_stats(nvlist_t * nvl,zdb_nvl_stats_t * stats)4588 collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats)
4589 {
4590 nvlist_t *list, **array;
4591 nvpair_t *nvp = NULL;
4592 const char *name;
4593 uint_t i, items;
4594
4595 stats->zns_list_count++;
4596
4597 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4598 name = nvpair_name(nvp);
4599
4600 switch (nvpair_type(nvp)) {
4601 case DATA_TYPE_STRING:
4602 fnvlist_add_string(stats->zns_string, name,
4603 fnvpair_value_string(nvp));
4604 break;
4605 case DATA_TYPE_UINT64:
4606 fnvlist_add_uint64(stats->zns_uint64, name,
4607 fnvpair_value_uint64(nvp));
4608 break;
4609 case DATA_TYPE_BOOLEAN:
4610 fnvlist_add_boolean(stats->zns_boolean, name);
4611 break;
4612 case DATA_TYPE_NVLIST:
4613 if (nvpair_value_nvlist(nvp, &list) == 0)
4614 collect_nvlist_stats(list, stats);
4615 break;
4616 case DATA_TYPE_NVLIST_ARRAY:
4617 if (nvpair_value_nvlist_array(nvp, &array, &items) != 0)
4618 break;
4619
4620 for (i = 0; i < items; i++) {
4621 collect_nvlist_stats(array[i], stats);
4622
4623 /* collect stats on leaf vdev */
4624 if (strcmp(name, "children") == 0) {
4625 size_t size;
4626
4627 (void) nvlist_size(array[i], &size,
4628 NV_ENCODE_XDR);
4629 stats->zns_leaf_total += size;
4630 if (size > stats->zns_leaf_largest)
4631 stats->zns_leaf_largest = size;
4632 stats->zns_leaf_count++;
4633 }
4634 }
4635 break;
4636 default:
4637 (void) printf("skip type %d!\n", (int)nvpair_type(nvp));
4638 }
4639 }
4640 }
4641
4642 static void
dump_nvlist_stats(nvlist_t * nvl,size_t cap)4643 dump_nvlist_stats(nvlist_t *nvl, size_t cap)
4644 {
4645 zdb_nvl_stats_t stats = { 0 };
4646 size_t size, sum = 0, total;
4647 size_t noise;
4648
4649 /* requires nvlist with non-unique names for stat collection */
4650 VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0));
4651 VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0));
4652 VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0));
4653 VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR));
4654
4655 (void) printf("\n\nZFS Label NVList Config Stats:\n");
4656
4657 VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR));
4658 (void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4659 (int)total, (int)(cap - total), 100.0 * total / cap);
4660
4661 collect_nvlist_stats(nvl, &stats);
4662
4663 VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR));
4664 size -= noise;
4665 sum += size;
4666 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4667 (int)fnvlist_num_pairs(stats.zns_uint64),
4668 (int)size, 100.0 * size / total);
4669
4670 VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR));
4671 size -= noise;
4672 sum += size;
4673 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4674 (int)fnvlist_num_pairs(stats.zns_string),
4675 (int)size, 100.0 * size / total);
4676
4677 VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR));
4678 size -= noise;
4679 sum += size;
4680 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4681 (int)fnvlist_num_pairs(stats.zns_boolean),
4682 (int)size, 100.0 * size / total);
4683
4684 size = total - sum; /* treat remainder as nvlist overhead */
4685 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4686 stats.zns_list_count, (int)size, 100.0 * size / total);
4687
4688 if (stats.zns_leaf_count > 0) {
4689 size_t average = stats.zns_leaf_total / stats.zns_leaf_count;
4690
4691 (void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4692 stats.zns_leaf_count, (int)average);
4693 (void) printf("%24d bytes largest\n",
4694 (int)stats.zns_leaf_largest);
4695
4696 if (dump_opt['l'] >= 3 && average > 0)
4697 (void) printf(" space for %d additional leaf vdevs\n",
4698 (int)((cap - total) / average));
4699 }
4700 (void) printf("\n");
4701
4702 nvlist_free(stats.zns_string);
4703 nvlist_free(stats.zns_uint64);
4704 nvlist_free(stats.zns_boolean);
4705 }
4706
4707 typedef struct cksum_record {
4708 zio_cksum_t cksum;
4709 boolean_t labels[VDEV_LABELS];
4710 avl_node_t link;
4711 } cksum_record_t;
4712
4713 static int
cksum_record_compare(const void * x1,const void * x2)4714 cksum_record_compare(const void *x1, const void *x2)
4715 {
4716 const cksum_record_t *l = (cksum_record_t *)x1;
4717 const cksum_record_t *r = (cksum_record_t *)x2;
4718 int arraysize = ARRAY_SIZE(l->cksum.zc_word);
4719 int difference = 0;
4720
4721 for (int i = 0; i < arraysize; i++) {
4722 difference = TREE_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]);
4723 if (difference)
4724 break;
4725 }
4726
4727 return (difference);
4728 }
4729
4730 static cksum_record_t *
cksum_record_alloc(zio_cksum_t * cksum,int l)4731 cksum_record_alloc(zio_cksum_t *cksum, int l)
4732 {
4733 cksum_record_t *rec;
4734
4735 rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL);
4736 rec->cksum = *cksum;
4737 rec->labels[l] = B_TRUE;
4738
4739 return (rec);
4740 }
4741
4742 static cksum_record_t *
cksum_record_lookup(avl_tree_t * tree,zio_cksum_t * cksum)4743 cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum)
4744 {
4745 cksum_record_t lookup = { .cksum = *cksum };
4746 avl_index_t where;
4747
4748 return (avl_find(tree, &lookup, &where));
4749 }
4750
4751 static cksum_record_t *
cksum_record_insert(avl_tree_t * tree,zio_cksum_t * cksum,int l)4752 cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l)
4753 {
4754 cksum_record_t *rec;
4755
4756 rec = cksum_record_lookup(tree, cksum);
4757 if (rec) {
4758 rec->labels[l] = B_TRUE;
4759 } else {
4760 rec = cksum_record_alloc(cksum, l);
4761 avl_add(tree, rec);
4762 }
4763
4764 return (rec);
4765 }
4766
4767 static int
first_label(cksum_record_t * rec)4768 first_label(cksum_record_t *rec)
4769 {
4770 for (int i = 0; i < VDEV_LABELS; i++)
4771 if (rec->labels[i])
4772 return (i);
4773
4774 return (-1);
4775 }
4776
4777 static void
print_label_numbers(const char * prefix,const cksum_record_t * rec)4778 print_label_numbers(const char *prefix, const cksum_record_t *rec)
4779 {
4780 fputs(prefix, stdout);
4781 for (int i = 0; i < VDEV_LABELS; i++)
4782 if (rec->labels[i] == B_TRUE)
4783 printf("%d ", i);
4784 putchar('\n');
4785 }
4786
4787 #define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4788
4789 typedef struct zdb_label {
4790 vdev_label_t label;
4791 uint64_t label_offset;
4792 nvlist_t *config_nv;
4793 cksum_record_t *config;
4794 cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT];
4795 boolean_t header_printed;
4796 boolean_t read_failed;
4797 boolean_t cksum_valid;
4798 } zdb_label_t;
4799
4800 static void
print_label_header(zdb_label_t * label,int l)4801 print_label_header(zdb_label_t *label, int l)
4802 {
4803
4804 if (dump_opt['q'])
4805 return;
4806
4807 if (label->header_printed == B_TRUE)
4808 return;
4809
4810 (void) printf("------------------------------------\n");
4811 (void) printf("LABEL %d %s\n", l,
4812 label->cksum_valid ? "" : "(Bad label cksum)");
4813 (void) printf("------------------------------------\n");
4814
4815 label->header_printed = B_TRUE;
4816 }
4817
4818 static void
print_l2arc_header(void)4819 print_l2arc_header(void)
4820 {
4821 (void) printf("------------------------------------\n");
4822 (void) printf("L2ARC device header\n");
4823 (void) printf("------------------------------------\n");
4824 }
4825
4826 static void
print_l2arc_log_blocks(void)4827 print_l2arc_log_blocks(void)
4828 {
4829 (void) printf("------------------------------------\n");
4830 (void) printf("L2ARC device log blocks\n");
4831 (void) printf("------------------------------------\n");
4832 }
4833
4834 static void
dump_l2arc_log_entries(uint64_t log_entries,l2arc_log_ent_phys_t * le,uint64_t i)4835 dump_l2arc_log_entries(uint64_t log_entries,
4836 l2arc_log_ent_phys_t *le, uint64_t i)
4837 {
4838 for (int j = 0; j < log_entries; j++) {
4839 dva_t dva = le[j].le_dva;
4840 (void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4841 "vdev: %llu, offset: %llu\n",
4842 (u_longlong_t)i, j + 1,
4843 (u_longlong_t)DVA_GET_ASIZE(&dva),
4844 (u_longlong_t)DVA_GET_VDEV(&dva),
4845 (u_longlong_t)DVA_GET_OFFSET(&dva));
4846 (void) printf("|\t\t\t\tbirth: %llu\n",
4847 (u_longlong_t)le[j].le_birth);
4848 (void) printf("|\t\t\t\tlsize: %llu\n",
4849 (u_longlong_t)L2BLK_GET_LSIZE((&le[j])->le_prop));
4850 (void) printf("|\t\t\t\tpsize: %llu\n",
4851 (u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop));
4852 (void) printf("|\t\t\t\tcompr: %llu\n",
4853 (u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop));
4854 (void) printf("|\t\t\t\tcomplevel: %llu\n",
4855 (u_longlong_t)(&le[j])->le_complevel);
4856 (void) printf("|\t\t\t\ttype: %llu\n",
4857 (u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop));
4858 (void) printf("|\t\t\t\tprotected: %llu\n",
4859 (u_longlong_t)L2BLK_GET_PROTECTED((&le[j])->le_prop));
4860 (void) printf("|\t\t\t\tprefetch: %llu\n",
4861 (u_longlong_t)L2BLK_GET_PREFETCH((&le[j])->le_prop));
4862 (void) printf("|\t\t\t\taddress: %llu\n",
4863 (u_longlong_t)le[j].le_daddr);
4864 (void) printf("|\t\t\t\tARC state: %llu\n",
4865 (u_longlong_t)L2BLK_GET_STATE((&le[j])->le_prop));
4866 (void) printf("|\n");
4867 }
4868 (void) printf("\n");
4869 }
4870
4871 static void
dump_l2arc_log_blkptr(const l2arc_log_blkptr_t * lbps)4872 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t *lbps)
4873 {
4874 (void) printf("|\t\tdaddr: %llu\n", (u_longlong_t)lbps->lbp_daddr);
4875 (void) printf("|\t\tpayload_asize: %llu\n",
4876 (u_longlong_t)lbps->lbp_payload_asize);
4877 (void) printf("|\t\tpayload_start: %llu\n",
4878 (u_longlong_t)lbps->lbp_payload_start);
4879 (void) printf("|\t\tlsize: %llu\n",
4880 (u_longlong_t)L2BLK_GET_LSIZE(lbps->lbp_prop));
4881 (void) printf("|\t\tasize: %llu\n",
4882 (u_longlong_t)L2BLK_GET_PSIZE(lbps->lbp_prop));
4883 (void) printf("|\t\tcompralgo: %llu\n",
4884 (u_longlong_t)L2BLK_GET_COMPRESS(lbps->lbp_prop));
4885 (void) printf("|\t\tcksumalgo: %llu\n",
4886 (u_longlong_t)L2BLK_GET_CHECKSUM(lbps->lbp_prop));
4887 (void) printf("|\n\n");
4888 }
4889
4890 static void
dump_l2arc_log_blocks(int fd,const l2arc_dev_hdr_phys_t * l2dhdr,l2arc_dev_hdr_phys_t * rebuild)4891 dump_l2arc_log_blocks(int fd, const l2arc_dev_hdr_phys_t *l2dhdr,
4892 l2arc_dev_hdr_phys_t *rebuild)
4893 {
4894 l2arc_log_blk_phys_t this_lb;
4895 uint64_t asize;
4896 l2arc_log_blkptr_t lbps[2];
4897 zio_cksum_t cksum;
4898 int failed = 0;
4899 l2arc_dev_t dev;
4900
4901 if (!dump_opt['q'])
4902 print_l2arc_log_blocks();
4903 memcpy(lbps, l2dhdr->dh_start_lbps, sizeof (lbps));
4904
4905 dev.l2ad_evict = l2dhdr->dh_evict;
4906 dev.l2ad_start = l2dhdr->dh_start;
4907 dev.l2ad_end = l2dhdr->dh_end;
4908
4909 if (l2dhdr->dh_start_lbps[0].lbp_daddr == 0) {
4910 /* no log blocks to read */
4911 if (!dump_opt['q']) {
4912 (void) printf("No log blocks to read\n");
4913 (void) printf("\n");
4914 }
4915 return;
4916 } else {
4917 dev.l2ad_hand = lbps[0].lbp_daddr +
4918 L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4919 }
4920
4921 dev.l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST);
4922
4923 for (;;) {
4924 if (!l2arc_log_blkptr_valid(&dev, &lbps[0]))
4925 break;
4926
4927 /* L2BLK_GET_PSIZE returns aligned size for log blocks */
4928 asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4929 if (pread64(fd, &this_lb, asize, lbps[0].lbp_daddr) != asize) {
4930 if (!dump_opt['q']) {
4931 (void) printf("Error while reading next log "
4932 "block\n\n");
4933 }
4934 break;
4935 }
4936
4937 fletcher_4_native_varsize(&this_lb, asize, &cksum);
4938 if (!ZIO_CHECKSUM_EQUAL(cksum, lbps[0].lbp_cksum)) {
4939 failed++;
4940 if (!dump_opt['q']) {
4941 (void) printf("Invalid cksum\n");
4942 dump_l2arc_log_blkptr(&lbps[0]);
4943 }
4944 break;
4945 }
4946
4947 switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) {
4948 case ZIO_COMPRESS_OFF:
4949 break;
4950 default: {
4951 abd_t *abd = abd_alloc_linear(asize, B_TRUE);
4952 abd_copy_from_buf_off(abd, &this_lb, 0, asize);
4953 abd_t dabd;
4954 abd_get_from_buf_struct(&dabd, &this_lb,
4955 sizeof (this_lb));
4956 int err = zio_decompress_data(L2BLK_GET_COMPRESS(
4957 (&lbps[0])->lbp_prop), abd, &dabd,
4958 asize, sizeof (this_lb), NULL);
4959 abd_free(&dabd);
4960 abd_free(abd);
4961 if (err != 0) {
4962 (void) printf("L2ARC block decompression "
4963 "failed\n");
4964 goto out;
4965 }
4966 break;
4967 }
4968 }
4969
4970 if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
4971 byteswap_uint64_array(&this_lb, sizeof (this_lb));
4972 if (this_lb.lb_magic != L2ARC_LOG_BLK_MAGIC) {
4973 if (!dump_opt['q'])
4974 (void) printf("Invalid log block magic\n\n");
4975 break;
4976 }
4977
4978 rebuild->dh_lb_count++;
4979 rebuild->dh_lb_asize += asize;
4980 if (dump_opt['l'] > 1 && !dump_opt['q']) {
4981 (void) printf("lb[%4llu]\tmagic: %llu\n",
4982 (u_longlong_t)rebuild->dh_lb_count,
4983 (u_longlong_t)this_lb.lb_magic);
4984 dump_l2arc_log_blkptr(&lbps[0]);
4985 }
4986
4987 if (dump_opt['l'] > 2 && !dump_opt['q'])
4988 dump_l2arc_log_entries(l2dhdr->dh_log_entries,
4989 this_lb.lb_entries,
4990 rebuild->dh_lb_count);
4991
4992 if (l2arc_range_check_overlap(lbps[1].lbp_payload_start,
4993 lbps[0].lbp_payload_start, dev.l2ad_evict) &&
4994 !dev.l2ad_first)
4995 break;
4996
4997 lbps[0] = lbps[1];
4998 lbps[1] = this_lb.lb_prev_lbp;
4999 }
5000 out:
5001 if (!dump_opt['q']) {
5002 (void) printf("log_blk_count:\t %llu with valid cksum\n",
5003 (u_longlong_t)rebuild->dh_lb_count);
5004 (void) printf("\t\t %d with invalid cksum\n", failed);
5005 (void) printf("log_blk_asize:\t %llu\n\n",
5006 (u_longlong_t)rebuild->dh_lb_asize);
5007 }
5008 }
5009
5010 static int
dump_l2arc_header(int fd)5011 dump_l2arc_header(int fd)
5012 {
5013 l2arc_dev_hdr_phys_t l2dhdr = {0}, rebuild = {0};
5014 int error = B_FALSE;
5015
5016 if (pread64(fd, &l2dhdr, sizeof (l2dhdr),
5017 VDEV_LABEL_START_SIZE) != sizeof (l2dhdr)) {
5018 error = B_TRUE;
5019 } else {
5020 if (l2dhdr.dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC))
5021 byteswap_uint64_array(&l2dhdr, sizeof (l2dhdr));
5022
5023 if (l2dhdr.dh_magic != L2ARC_DEV_HDR_MAGIC)
5024 error = B_TRUE;
5025 }
5026
5027 if (error) {
5028 (void) printf("L2ARC device header not found\n\n");
5029 /* Do not return an error here for backward compatibility */
5030 return (0);
5031 } else if (!dump_opt['q']) {
5032 print_l2arc_header();
5033
5034 (void) printf(" magic: %llu\n",
5035 (u_longlong_t)l2dhdr.dh_magic);
5036 (void) printf(" version: %llu\n",
5037 (u_longlong_t)l2dhdr.dh_version);
5038 (void) printf(" pool_guid: %llu\n",
5039 (u_longlong_t)l2dhdr.dh_spa_guid);
5040 (void) printf(" flags: %llu\n",
5041 (u_longlong_t)l2dhdr.dh_flags);
5042 (void) printf(" start_lbps[0]: %llu\n",
5043 (u_longlong_t)
5044 l2dhdr.dh_start_lbps[0].lbp_daddr);
5045 (void) printf(" start_lbps[1]: %llu\n",
5046 (u_longlong_t)
5047 l2dhdr.dh_start_lbps[1].lbp_daddr);
5048 (void) printf(" log_blk_ent: %llu\n",
5049 (u_longlong_t)l2dhdr.dh_log_entries);
5050 (void) printf(" start: %llu\n",
5051 (u_longlong_t)l2dhdr.dh_start);
5052 (void) printf(" end: %llu\n",
5053 (u_longlong_t)l2dhdr.dh_end);
5054 (void) printf(" evict: %llu\n",
5055 (u_longlong_t)l2dhdr.dh_evict);
5056 (void) printf(" lb_asize_refcount: %llu\n",
5057 (u_longlong_t)l2dhdr.dh_lb_asize);
5058 (void) printf(" lb_count_refcount: %llu\n",
5059 (u_longlong_t)l2dhdr.dh_lb_count);
5060 (void) printf(" trim_action_time: %llu\n",
5061 (u_longlong_t)l2dhdr.dh_trim_action_time);
5062 (void) printf(" trim_state: %llu\n\n",
5063 (u_longlong_t)l2dhdr.dh_trim_state);
5064 }
5065
5066 dump_l2arc_log_blocks(fd, &l2dhdr, &rebuild);
5067 /*
5068 * The total aligned size of log blocks and the number of log blocks
5069 * reported in the header of the device may be less than what zdb
5070 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
5071 * This happens because dump_l2arc_log_blocks() lacks the memory
5072 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
5073 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
5074 * and dh_lb_count will be lower to begin with than what exists on the
5075 * device. This is normal and zdb should not exit with an error. The
5076 * opposite case should never happen though, the values reported in the
5077 * header should never be higher than what dump_l2arc_log_blocks() and
5078 * l2arc_rebuild() report. If this happens there is a leak in the
5079 * accounting of log blocks.
5080 */
5081 if (l2dhdr.dh_lb_asize > rebuild.dh_lb_asize ||
5082 l2dhdr.dh_lb_count > rebuild.dh_lb_count)
5083 return (1);
5084
5085 return (0);
5086 }
5087
5088 static void
dump_config_from_label(zdb_label_t * label,size_t buflen,int l)5089 dump_config_from_label(zdb_label_t *label, size_t buflen, int l)
5090 {
5091 if (dump_opt['q'])
5092 return;
5093
5094 if ((dump_opt['l'] < 3) && (first_label(label->config) != l))
5095 return;
5096
5097 print_label_header(label, l);
5098 dump_nvlist(label->config_nv, 4);
5099 print_label_numbers(" labels = ", label->config);
5100
5101 if (dump_opt['l'] >= 2)
5102 dump_nvlist_stats(label->config_nv, buflen);
5103 }
5104
5105 #define ZDB_MAX_UB_HEADER_SIZE 32
5106
5107 static void
dump_label_uberblocks(zdb_label_t * label,uint64_t ashift,int label_num)5108 dump_label_uberblocks(zdb_label_t *label, uint64_t ashift, int label_num)
5109 {
5110
5111 vdev_t vd;
5112 char header[ZDB_MAX_UB_HEADER_SIZE];
5113
5114 vd.vdev_ashift = ashift;
5115 vd.vdev_top = &vd;
5116
5117 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5118 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5119 uberblock_t *ub = (void *)((char *)&label->label + uoff);
5120 cksum_record_t *rec = label->uberblocks[i];
5121
5122 if (rec == NULL) {
5123 if (dump_opt['u'] >= 2) {
5124 print_label_header(label, label_num);
5125 (void) printf(" Uberblock[%d] invalid\n", i);
5126 }
5127 continue;
5128 }
5129
5130 if ((dump_opt['u'] < 3) && (first_label(rec) != label_num))
5131 continue;
5132
5133 if ((dump_opt['u'] < 4) &&
5134 (ub->ub_mmp_magic == MMP_MAGIC) && ub->ub_mmp_delay &&
5135 (i >= VDEV_UBERBLOCK_COUNT(&vd) - MMP_BLOCKS_PER_LABEL))
5136 continue;
5137
5138 print_label_header(label, label_num);
5139 (void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE,
5140 " Uberblock[%d]\n", i);
5141 dump_uberblock(ub, header, "");
5142 print_label_numbers(" labels = ", rec);
5143 }
5144 }
5145
5146 static char curpath[PATH_MAX];
5147
5148 /*
5149 * Iterate through the path components, recursively passing
5150 * current one's obj and remaining path until we find the obj
5151 * for the last one.
5152 */
5153 static int
dump_path_impl(objset_t * os,uint64_t obj,char * name,uint64_t * retobj)5154 dump_path_impl(objset_t *os, uint64_t obj, char *name, uint64_t *retobj)
5155 {
5156 int err;
5157 boolean_t header = B_TRUE;
5158 uint64_t child_obj;
5159 char *s;
5160 dmu_buf_t *db;
5161 dmu_object_info_t doi;
5162
5163 if ((s = strchr(name, '/')) != NULL)
5164 *s = '\0';
5165 err = zap_lookup(os, obj, name, 8, 1, &child_obj);
5166
5167 (void) strlcat(curpath, name, sizeof (curpath));
5168
5169 if (err != 0) {
5170 (void) fprintf(stderr, "failed to lookup %s: %s\n",
5171 curpath, strerror(err));
5172 return (err);
5173 }
5174
5175 child_obj = ZFS_DIRENT_OBJ(child_obj);
5176 err = sa_buf_hold(os, child_obj, FTAG, &db);
5177 if (err != 0) {
5178 (void) fprintf(stderr,
5179 "failed to get SA dbuf for obj %llu: %s\n",
5180 (u_longlong_t)child_obj, strerror(err));
5181 return (EINVAL);
5182 }
5183 dmu_object_info_from_db(db, &doi);
5184 sa_buf_rele(db, FTAG);
5185
5186 if (doi.doi_bonus_type != DMU_OT_SA &&
5187 doi.doi_bonus_type != DMU_OT_ZNODE) {
5188 (void) fprintf(stderr, "invalid bonus type %d for obj %llu\n",
5189 doi.doi_bonus_type, (u_longlong_t)child_obj);
5190 return (EINVAL);
5191 }
5192
5193 if (dump_opt['v'] > 6) {
5194 (void) printf("obj=%llu %s type=%d bonustype=%d\n",
5195 (u_longlong_t)child_obj, curpath, doi.doi_type,
5196 doi.doi_bonus_type);
5197 }
5198
5199 (void) strlcat(curpath, "/", sizeof (curpath));
5200
5201 switch (doi.doi_type) {
5202 case DMU_OT_DIRECTORY_CONTENTS:
5203 if (s != NULL && *(s + 1) != '\0')
5204 return (dump_path_impl(os, child_obj, s + 1, retobj));
5205 zfs_fallthrough;
5206 case DMU_OT_PLAIN_FILE_CONTENTS:
5207 if (retobj != NULL) {
5208 *retobj = child_obj;
5209 } else {
5210 dump_object(os, child_obj, dump_opt['v'], &header,
5211 NULL, 0);
5212 }
5213 return (0);
5214 default:
5215 (void) fprintf(stderr, "object %llu has non-file/directory "
5216 "type %d\n", (u_longlong_t)obj, doi.doi_type);
5217 break;
5218 }
5219
5220 return (EINVAL);
5221 }
5222
5223 /*
5224 * Dump the blocks for the object specified by path inside the dataset.
5225 */
5226 static int
dump_path(char * ds,char * path,uint64_t * retobj)5227 dump_path(char *ds, char *path, uint64_t *retobj)
5228 {
5229 int err;
5230 objset_t *os;
5231 uint64_t root_obj;
5232
5233 err = open_objset(ds, FTAG, &os);
5234 if (err != 0)
5235 return (err);
5236
5237 err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj);
5238 if (err != 0) {
5239 (void) fprintf(stderr, "can't lookup root znode: %s\n",
5240 strerror(err));
5241 close_objset(os, FTAG);
5242 return (EINVAL);
5243 }
5244
5245 (void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds);
5246
5247 err = dump_path_impl(os, root_obj, path, retobj);
5248
5249 close_objset(os, FTAG);
5250 return (err);
5251 }
5252
5253 static int
dump_backup_bytes(objset_t * os,void * buf,int len,void * arg)5254 dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
5255 {
5256 const char *p = (const char *)buf;
5257 ssize_t nwritten;
5258
5259 (void) os;
5260 (void) arg;
5261
5262 /* Write the data out, handling short writes and signals. */
5263 while ((nwritten = write(STDOUT_FILENO, p, len)) < len) {
5264 if (nwritten < 0) {
5265 if (errno == EINTR)
5266 continue;
5267 return (errno);
5268 }
5269 p += nwritten;
5270 len -= nwritten;
5271 }
5272
5273 return (0);
5274 }
5275
5276 static void
dump_backup(const char * pool,uint64_t objset_id,const char * flagstr)5277 dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
5278 {
5279 boolean_t embed = B_FALSE;
5280 boolean_t large_block = B_FALSE;
5281 boolean_t compress = B_FALSE;
5282 boolean_t raw = B_FALSE;
5283
5284 const char *c;
5285 for (c = flagstr; c != NULL && *c != '\0'; c++) {
5286 switch (*c) {
5287 case 'e':
5288 embed = B_TRUE;
5289 break;
5290 case 'L':
5291 large_block = B_TRUE;
5292 break;
5293 case 'c':
5294 compress = B_TRUE;
5295 break;
5296 case 'w':
5297 raw = B_TRUE;
5298 break;
5299 default:
5300 fprintf(stderr, "dump_backup: invalid flag "
5301 "'%c'\n", *c);
5302 return;
5303 }
5304 }
5305
5306 if (isatty(STDOUT_FILENO)) {
5307 fprintf(stderr, "dump_backup: stream cannot be written "
5308 "to a terminal\n");
5309 return;
5310 }
5311
5312 offset_t off = 0;
5313 dmu_send_outparams_t out = {
5314 .dso_outfunc = dump_backup_bytes,
5315 .dso_dryrun = B_FALSE,
5316 };
5317
5318 int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed,
5319 large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO,
5320 &off, &out);
5321 if (err != 0) {
5322 fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",
5323 strerror(err));
5324 return;
5325 }
5326 }
5327
5328 static int
zdb_copy_object(objset_t * os,uint64_t srcobj,char * destfile)5329 zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile)
5330 {
5331 int err = 0;
5332 uint64_t size, readsize, oursize, offset;
5333 ssize_t writesize;
5334 sa_handle_t *hdl;
5335
5336 (void) printf("Copying object %" PRIu64 " to file %s\n", srcobj,
5337 destfile);
5338
5339 VERIFY3P(os, ==, sa_os);
5340 if ((err = sa_handle_get(os, srcobj, NULL, SA_HDL_PRIVATE, &hdl))) {
5341 (void) printf("Failed to get handle for SA znode\n");
5342 return (err);
5343 }
5344 if ((err = sa_lookup(hdl, sa_attr_table[ZPL_SIZE], &size, 8))) {
5345 (void) sa_handle_destroy(hdl);
5346 return (err);
5347 }
5348 (void) sa_handle_destroy(hdl);
5349
5350 (void) printf("Object %" PRIu64 " is %" PRIu64 " bytes\n", srcobj,
5351 size);
5352 if (size == 0) {
5353 return (EINVAL);
5354 }
5355
5356 int fd = open(destfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
5357 if (fd == -1)
5358 return (errno);
5359 /*
5360 * We cap the size at 1 mebibyte here to prevent
5361 * allocation failures and nigh-infinite printing if the
5362 * object is extremely large.
5363 */
5364 oursize = MIN(size, 1 << 20);
5365 offset = 0;
5366 char *buf = kmem_alloc(oursize, KM_NOSLEEP);
5367 if (buf == NULL) {
5368 (void) close(fd);
5369 return (ENOMEM);
5370 }
5371
5372 while (offset < size) {
5373 readsize = MIN(size - offset, 1 << 20);
5374 err = dmu_read(os, srcobj, offset, readsize, buf, 0);
5375 if (err != 0) {
5376 (void) printf("got error %u from dmu_read\n", err);
5377 kmem_free(buf, oursize);
5378 (void) close(fd);
5379 return (err);
5380 }
5381 if (dump_opt['v'] > 3) {
5382 (void) printf("Read offset=%" PRIu64 " size=%" PRIu64
5383 " error=%d\n", offset, readsize, err);
5384 }
5385
5386 writesize = write(fd, buf, readsize);
5387 if (writesize < 0) {
5388 err = errno;
5389 break;
5390 } else if (writesize != readsize) {
5391 /* Incomplete write */
5392 (void) fprintf(stderr, "Short write, only wrote %llu of"
5393 " %" PRIu64 " bytes, exiting...\n",
5394 (u_longlong_t)writesize, readsize);
5395 break;
5396 }
5397
5398 offset += readsize;
5399 }
5400
5401 (void) close(fd);
5402
5403 if (buf != NULL)
5404 kmem_free(buf, oursize);
5405
5406 return (err);
5407 }
5408
5409 static boolean_t
label_cksum_valid(vdev_label_t * label,uint64_t offset)5410 label_cksum_valid(vdev_label_t *label, uint64_t offset)
5411 {
5412 zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
5413 zio_cksum_t expected_cksum;
5414 zio_cksum_t actual_cksum;
5415 zio_cksum_t verifier;
5416 zio_eck_t *eck;
5417 int byteswap;
5418
5419 void *data = (char *)label + offsetof(vdev_label_t, vl_vdev_phys);
5420 eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1;
5421
5422 offset += offsetof(vdev_label_t, vl_vdev_phys);
5423 ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
5424
5425 byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
5426 if (byteswap)
5427 byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
5428
5429 expected_cksum = eck->zec_cksum;
5430 eck->zec_cksum = verifier;
5431
5432 abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE);
5433 ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum);
5434 abd_free(abd);
5435
5436 if (byteswap)
5437 byteswap_uint64_array(&expected_cksum, sizeof (zio_cksum_t));
5438
5439 if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
5440 return (B_TRUE);
5441
5442 return (B_FALSE);
5443 }
5444
5445 static int
dump_label(const char * dev)5446 dump_label(const char *dev)
5447 {
5448 char path[MAXPATHLEN];
5449 zdb_label_t labels[VDEV_LABELS] = {{{{0}}}};
5450 uint64_t psize, ashift, l2cache;
5451 struct stat64 statbuf;
5452 boolean_t config_found = B_FALSE;
5453 boolean_t error = B_FALSE;
5454 boolean_t read_l2arc_header = B_FALSE;
5455 avl_tree_t config_tree;
5456 avl_tree_t uberblock_tree;
5457 void *node, *cookie;
5458 int fd;
5459
5460 /*
5461 * Check if we were given absolute path and use it as is.
5462 * Otherwise if the provided vdev name doesn't point to a file,
5463 * try prepending expected disk paths and partition numbers.
5464 */
5465 (void) strlcpy(path, dev, sizeof (path));
5466 if (dev[0] != '/' && stat64(path, &statbuf) != 0) {
5467 int error;
5468
5469 error = zfs_resolve_shortname(dev, path, MAXPATHLEN);
5470 if (error == 0 && zfs_dev_is_whole_disk(path)) {
5471 if (zfs_append_partition(path, MAXPATHLEN) == -1)
5472 error = ENOENT;
5473 }
5474
5475 if (error || (stat64(path, &statbuf) != 0)) {
5476 (void) printf("failed to find device %s, try "
5477 "specifying absolute path instead\n", dev);
5478 return (1);
5479 }
5480 }
5481
5482 if ((fd = open64(path, O_RDONLY)) < 0) {
5483 (void) printf("cannot open '%s': %s\n", path, strerror(errno));
5484 zdb_exit(1);
5485 }
5486
5487 if (fstat64_blk(fd, &statbuf) != 0) {
5488 (void) printf("failed to stat '%s': %s\n", path,
5489 strerror(errno));
5490 (void) close(fd);
5491 zdb_exit(1);
5492 }
5493
5494 if (S_ISBLK(statbuf.st_mode) && zfs_dev_flush(fd) != 0)
5495 (void) printf("failed to invalidate cache '%s' : %s\n", path,
5496 strerror(errno));
5497
5498 avl_create(&config_tree, cksum_record_compare,
5499 sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5500 avl_create(&uberblock_tree, cksum_record_compare,
5501 sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5502
5503 psize = statbuf.st_size;
5504 psize = P2ALIGN_TYPED(psize, sizeof (vdev_label_t), uint64_t);
5505 ashift = SPA_MINBLOCKSHIFT;
5506
5507 /*
5508 * 1. Read the label from disk
5509 * 2. Verify label cksum
5510 * 3. Unpack the configuration and insert in config tree.
5511 * 4. Traverse all uberblocks and insert in uberblock tree.
5512 */
5513 for (int l = 0; l < VDEV_LABELS; l++) {
5514 zdb_label_t *label = &labels[l];
5515 char *buf = label->label.vl_vdev_phys.vp_nvlist;
5516 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5517 nvlist_t *config;
5518 cksum_record_t *rec;
5519 zio_cksum_t cksum;
5520 vdev_t vd;
5521
5522 label->label_offset = vdev_label_offset(psize, l, 0);
5523
5524 if (pread64(fd, &label->label, sizeof (label->label),
5525 label->label_offset) != sizeof (label->label)) {
5526 if (!dump_opt['q'])
5527 (void) printf("failed to read label %d\n", l);
5528 label->read_failed = B_TRUE;
5529 error = B_TRUE;
5530 continue;
5531 }
5532
5533 label->read_failed = B_FALSE;
5534 label->cksum_valid = label_cksum_valid(&label->label,
5535 label->label_offset);
5536
5537 if (nvlist_unpack(buf, buflen, &config, 0) == 0) {
5538 nvlist_t *vdev_tree = NULL;
5539 size_t size;
5540
5541 if ((nvlist_lookup_nvlist(config,
5542 ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) ||
5543 (nvlist_lookup_uint64(vdev_tree,
5544 ZPOOL_CONFIG_ASHIFT, &ashift) != 0))
5545 ashift = SPA_MINBLOCKSHIFT;
5546
5547 if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0)
5548 size = buflen;
5549
5550 /* If the device is a cache device read the header. */
5551 if (!read_l2arc_header) {
5552 if (nvlist_lookup_uint64(config,
5553 ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 &&
5554 l2cache == POOL_STATE_L2CACHE) {
5555 read_l2arc_header = B_TRUE;
5556 }
5557 }
5558
5559 fletcher_4_native_varsize(buf, size, &cksum);
5560 rec = cksum_record_insert(&config_tree, &cksum, l);
5561
5562 label->config = rec;
5563 label->config_nv = config;
5564 config_found = B_TRUE;
5565 } else {
5566 error = B_TRUE;
5567 }
5568
5569 vd.vdev_ashift = ashift;
5570 vd.vdev_top = &vd;
5571
5572 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5573 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5574 uberblock_t *ub = (void *)((char *)label + uoff);
5575
5576 if (uberblock_verify(ub))
5577 continue;
5578
5579 fletcher_4_native_varsize(ub, sizeof (*ub), &cksum);
5580 rec = cksum_record_insert(&uberblock_tree, &cksum, l);
5581
5582 label->uberblocks[i] = rec;
5583 }
5584 }
5585
5586 /*
5587 * Dump the label and uberblocks.
5588 */
5589 for (int l = 0; l < VDEV_LABELS; l++) {
5590 zdb_label_t *label = &labels[l];
5591 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5592
5593 if (label->read_failed == B_TRUE)
5594 continue;
5595
5596 if (label->config_nv) {
5597 dump_config_from_label(label, buflen, l);
5598 } else {
5599 if (!dump_opt['q'])
5600 (void) printf("failed to unpack label %d\n", l);
5601 }
5602
5603 if (dump_opt['u'])
5604 dump_label_uberblocks(label, ashift, l);
5605
5606 nvlist_free(label->config_nv);
5607 }
5608
5609 /*
5610 * Dump the L2ARC header, if existent.
5611 */
5612 if (read_l2arc_header)
5613 error |= dump_l2arc_header(fd);
5614
5615 cookie = NULL;
5616 while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL)
5617 umem_free(node, sizeof (cksum_record_t));
5618
5619 cookie = NULL;
5620 while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL)
5621 umem_free(node, sizeof (cksum_record_t));
5622
5623 avl_destroy(&config_tree);
5624 avl_destroy(&uberblock_tree);
5625
5626 (void) close(fd);
5627
5628 return (config_found == B_FALSE ? 2 :
5629 (error == B_TRUE ? 1 : 0));
5630 }
5631
5632 static uint64_t dataset_feature_count[SPA_FEATURES];
5633 static uint64_t global_feature_count[SPA_FEATURES];
5634 static uint64_t remap_deadlist_count = 0;
5635
5636 static int
dump_one_objset(const char * dsname,void * arg)5637 dump_one_objset(const char *dsname, void *arg)
5638 {
5639 (void) arg;
5640 int error;
5641 objset_t *os;
5642 spa_feature_t f;
5643
5644 error = open_objset(dsname, FTAG, &os);
5645 if (error != 0)
5646 return (0);
5647
5648 for (f = 0; f < SPA_FEATURES; f++) {
5649 if (!dsl_dataset_feature_is_active(dmu_objset_ds(os), f))
5650 continue;
5651 ASSERT(spa_feature_table[f].fi_flags &
5652 ZFEATURE_FLAG_PER_DATASET);
5653 dataset_feature_count[f]++;
5654 }
5655
5656 if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os))) {
5657 remap_deadlist_count++;
5658 }
5659
5660 for (dsl_bookmark_node_t *dbn =
5661 avl_first(&dmu_objset_ds(os)->ds_bookmarks); dbn != NULL;
5662 dbn = AVL_NEXT(&dmu_objset_ds(os)->ds_bookmarks, dbn)) {
5663 mos_obj_refd(dbn->dbn_phys.zbm_redaction_obj);
5664 if (dbn->dbn_phys.zbm_redaction_obj != 0) {
5665 global_feature_count[
5666 SPA_FEATURE_REDACTION_BOOKMARKS]++;
5667 objset_t *mos = os->os_spa->spa_meta_objset;
5668 dnode_t *rl;
5669 VERIFY0(dnode_hold(mos,
5670 dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl));
5671 if (rl->dn_have_spill) {
5672 global_feature_count[
5673 SPA_FEATURE_REDACTION_LIST_SPILL]++;
5674 }
5675 }
5676 if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)
5677 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN]++;
5678 }
5679
5680 if (dsl_deadlist_is_open(&dmu_objset_ds(os)->ds_dir->dd_livelist) &&
5681 !dmu_objset_is_snapshot(os)) {
5682 global_feature_count[SPA_FEATURE_LIVELIST]++;
5683 }
5684
5685 dump_objset(os);
5686 close_objset(os, FTAG);
5687 fuid_table_destroy();
5688 return (0);
5689 }
5690
5691 /*
5692 * Block statistics.
5693 */
5694 #define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5695 typedef struct zdb_blkstats {
5696 uint64_t zb_asize;
5697 uint64_t zb_lsize;
5698 uint64_t zb_psize;
5699 uint64_t zb_count;
5700 uint64_t zb_gangs;
5701 uint64_t zb_ditto_samevdev;
5702 uint64_t zb_ditto_same_ms;
5703 uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
5704 } zdb_blkstats_t;
5705
5706 /*
5707 * Extended object types to report deferred frees and dedup auto-ditto blocks.
5708 */
5709 #define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0)
5710 #define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1)
5711 #define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2)
5712 #define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3)
5713
5714 static const char *zdb_ot_extname[] = {
5715 "deferred free",
5716 "dedup ditto",
5717 "other",
5718 "Total",
5719 };
5720
5721 #define ZB_TOTAL DN_MAX_LEVELS
5722 #define SPA_MAX_FOR_16M (SPA_MAXBLOCKSHIFT+1)
5723
5724 typedef struct zdb_brt_entry {
5725 dva_t zbre_dva;
5726 uint64_t zbre_refcount;
5727 avl_node_t zbre_node;
5728 } zdb_brt_entry_t;
5729
5730 typedef struct zdb_cb {
5731 zdb_blkstats_t zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
5732 uint64_t zcb_removing_size;
5733 uint64_t zcb_checkpoint_size;
5734 uint64_t zcb_dedup_asize;
5735 uint64_t zcb_dedup_blocks;
5736 uint64_t zcb_clone_asize;
5737 uint64_t zcb_clone_blocks;
5738 uint64_t zcb_psize_count[SPA_MAX_FOR_16M];
5739 uint64_t zcb_lsize_count[SPA_MAX_FOR_16M];
5740 uint64_t zcb_asize_count[SPA_MAX_FOR_16M];
5741 uint64_t zcb_psize_len[SPA_MAX_FOR_16M];
5742 uint64_t zcb_lsize_len[SPA_MAX_FOR_16M];
5743 uint64_t zcb_asize_len[SPA_MAX_FOR_16M];
5744 uint64_t zcb_psize_total;
5745 uint64_t zcb_lsize_total;
5746 uint64_t zcb_asize_total;
5747 uint64_t zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
5748 uint64_t zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
5749 [BPE_PAYLOAD_SIZE + 1];
5750 uint64_t zcb_start;
5751 hrtime_t zcb_lastprint;
5752 uint64_t zcb_totalasize;
5753 uint64_t zcb_errors[256];
5754 int zcb_readfails;
5755 int zcb_haderrors;
5756 spa_t *zcb_spa;
5757 uint32_t **zcb_vd_obsolete_counts;
5758 avl_tree_t zcb_brt;
5759 boolean_t zcb_brt_is_active;
5760 } zdb_cb_t;
5761
5762 /* test if two DVA offsets from same vdev are within the same metaslab */
5763 static boolean_t
same_metaslab(spa_t * spa,uint64_t vdev,uint64_t off1,uint64_t off2)5764 same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2)
5765 {
5766 vdev_t *vd = vdev_lookup_top(spa, vdev);
5767 uint64_t ms_shift = vd->vdev_ms_shift;
5768
5769 return ((off1 >> ms_shift) == (off2 >> ms_shift));
5770 }
5771
5772 /*
5773 * Used to simplify reporting of the histogram data.
5774 */
5775 typedef struct one_histo {
5776 const char *name;
5777 uint64_t *count;
5778 uint64_t *len;
5779 uint64_t cumulative;
5780 } one_histo_t;
5781
5782 /*
5783 * The number of separate histograms processed for psize, lsize and asize.
5784 */
5785 #define NUM_HISTO 3
5786
5787 /*
5788 * This routine will create a fixed column size output of three different
5789 * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5790 * the count, length and cumulative length of the psize, lsize and
5791 * asize blocks.
5792 *
5793 * All three types of blocks are listed on a single line
5794 *
5795 * By default the table is printed in nicenumber format (e.g. 123K) but
5796 * if the '-P' parameter is specified then the full raw number (parseable)
5797 * is printed out.
5798 */
5799 static void
dump_size_histograms(zdb_cb_t * zcb)5800 dump_size_histograms(zdb_cb_t *zcb)
5801 {
5802 /*
5803 * A temporary buffer that allows us to convert a number into
5804 * a string using zdb_nicenumber to allow either raw or human
5805 * readable numbers to be output.
5806 */
5807 char numbuf[32];
5808
5809 /*
5810 * Define titles which are used in the headers of the tables
5811 * printed by this routine.
5812 */
5813 const char blocksize_title1[] = "block";
5814 const char blocksize_title2[] = "size";
5815 const char count_title[] = "Count";
5816 const char length_title[] = "Size";
5817 const char cumulative_title[] = "Cum.";
5818
5819 /*
5820 * Setup the histogram arrays (psize, lsize, and asize).
5821 */
5822 one_histo_t parm_histo[NUM_HISTO];
5823
5824 parm_histo[0].name = "psize";
5825 parm_histo[0].count = zcb->zcb_psize_count;
5826 parm_histo[0].len = zcb->zcb_psize_len;
5827 parm_histo[0].cumulative = 0;
5828
5829 parm_histo[1].name = "lsize";
5830 parm_histo[1].count = zcb->zcb_lsize_count;
5831 parm_histo[1].len = zcb->zcb_lsize_len;
5832 parm_histo[1].cumulative = 0;
5833
5834 parm_histo[2].name = "asize";
5835 parm_histo[2].count = zcb->zcb_asize_count;
5836 parm_histo[2].len = zcb->zcb_asize_len;
5837 parm_histo[2].cumulative = 0;
5838
5839
5840 (void) printf("\nBlock Size Histogram\n");
5841 switch (block_bin_mode) {
5842 case BIN_PSIZE:
5843 printf("(note: all categories are binned by %s)\n", "psize");
5844 break;
5845 case BIN_LSIZE:
5846 printf("(note: all categories are binned by %s)\n", "lsize");
5847 break;
5848 case BIN_ASIZE:
5849 printf("(note: all categories are binned by %s)\n", "asize");
5850 break;
5851 default:
5852 printf("(note: all categories are binned separately)\n");
5853 break;
5854 }
5855 if (block_classes != 0) {
5856 char buf[256] = "";
5857 if (block_classes & CLASS_NORMAL)
5858 strlcat(buf, "\"normal\", ", sizeof (buf));
5859 if (block_classes & CLASS_SPECIAL)
5860 strlcat(buf, "\"special\", ", sizeof (buf));
5861 if (block_classes & CLASS_DEDUP)
5862 strlcat(buf, "\"dedup\", ", sizeof (buf));
5863 if (block_classes & CLASS_OTHER)
5864 strlcat(buf, "\"other\", ", sizeof (buf));
5865 buf[strlen(buf)-2] = '\0';
5866 printf("(note: only blocks in these classes are counted: %s)\n",
5867 buf);
5868 }
5869 /*
5870 * Print the first line titles
5871 */
5872 if (dump_opt['P'])
5873 (void) printf("\n%s\t", blocksize_title1);
5874 else
5875 (void) printf("\n%7s ", blocksize_title1);
5876
5877 for (int j = 0; j < NUM_HISTO; j++) {
5878 if (dump_opt['P']) {
5879 if (j < NUM_HISTO - 1) {
5880 (void) printf("%s\t\t\t", parm_histo[j].name);
5881 } else {
5882 /* Don't print trailing spaces */
5883 (void) printf(" %s", parm_histo[j].name);
5884 }
5885 } else {
5886 if (j < NUM_HISTO - 1) {
5887 /* Left aligned strings in the output */
5888 (void) printf("%-7s ",
5889 parm_histo[j].name);
5890 } else {
5891 /* Don't print trailing spaces */
5892 (void) printf("%s", parm_histo[j].name);
5893 }
5894 }
5895 }
5896 (void) printf("\n");
5897
5898 /*
5899 * Print the second line titles
5900 */
5901 if (dump_opt['P']) {
5902 (void) printf("%s\t", blocksize_title2);
5903 } else {
5904 (void) printf("%7s ", blocksize_title2);
5905 }
5906
5907 for (int i = 0; i < NUM_HISTO; i++) {
5908 if (dump_opt['P']) {
5909 (void) printf("%s\t%s\t%s\t",
5910 count_title, length_title, cumulative_title);
5911 } else {
5912 (void) printf("%7s%7s%7s",
5913 count_title, length_title, cumulative_title);
5914 }
5915 }
5916 (void) printf("\n");
5917
5918 /*
5919 * Print the rows
5920 */
5921 for (int i = SPA_MINBLOCKSHIFT; i < SPA_MAX_FOR_16M; i++) {
5922
5923 /*
5924 * Print the first column showing the blocksize
5925 */
5926 zdb_nicenum((1ULL << i), numbuf, sizeof (numbuf));
5927
5928 if (dump_opt['P']) {
5929 printf("%s", numbuf);
5930 } else {
5931 printf("%7s:", numbuf);
5932 }
5933
5934 /*
5935 * Print the remaining set of 3 columns per size:
5936 * for psize, lsize and asize
5937 */
5938 for (int j = 0; j < NUM_HISTO; j++) {
5939 parm_histo[j].cumulative += parm_histo[j].len[i];
5940
5941 zdb_nicenum(parm_histo[j].count[i],
5942 numbuf, sizeof (numbuf));
5943 if (dump_opt['P'])
5944 (void) printf("\t%s", numbuf);
5945 else
5946 (void) printf("%7s", numbuf);
5947
5948 zdb_nicenum(parm_histo[j].len[i],
5949 numbuf, sizeof (numbuf));
5950 if (dump_opt['P'])
5951 (void) printf("\t%s", numbuf);
5952 else
5953 (void) printf("%7s", numbuf);
5954
5955 zdb_nicenum(parm_histo[j].cumulative,
5956 numbuf, sizeof (numbuf));
5957 if (dump_opt['P'])
5958 (void) printf("\t%s", numbuf);
5959 else
5960 (void) printf("%7s", numbuf);
5961 }
5962 (void) printf("\n");
5963 }
5964 }
5965
5966 static void
zdb_count_block(zdb_cb_t * zcb,zilog_t * zilog,const blkptr_t * bp,dmu_object_type_t type)5967 zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
5968 dmu_object_type_t type)
5969 {
5970 int i;
5971
5972 ASSERT(type < ZDB_OT_TOTAL);
5973
5974 if (zilog && zil_bp_tree_add(zilog, bp) != 0)
5975 return;
5976
5977 /*
5978 * This flag controls if we will issue a claim for the block while
5979 * counting it, to ensure that all blocks are referenced in space maps.
5980 * We don't issue claims if we're not doing leak tracking, because it's
5981 * expensive if the user isn't interested. We also don't claim the
5982 * second or later occurences of cloned or dedup'd blocks, because we
5983 * already claimed them the first time.
5984 */
5985 boolean_t do_claim = !dump_opt['L'];
5986
5987 spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
5988
5989 blkptr_t tempbp;
5990 if (BP_GET_DEDUP(bp)) {
5991 /*
5992 * Dedup'd blocks are special. We need to count them, so we can
5993 * later uncount them when reporting leaked space, and we must
5994 * only claim them once.
5995 *
5996 * We use the existing dedup system to track what we've seen.
5997 * The first time we see a block, we do a ddt_lookup() to see
5998 * if it exists in the DDT. If we're doing leak tracking, we
5999 * claim the block at this time.
6000 *
6001 * Each time we see a block, we reduce the refcount in the
6002 * entry by one, and add to the size and count of dedup'd
6003 * blocks to report at the end.
6004 */
6005
6006 ddt_t *ddt = ddt_select(zcb->zcb_spa, bp);
6007
6008 ddt_enter(ddt);
6009
6010 /*
6011 * Find the block. This will create the entry in memory, but
6012 * we'll know if that happened by its refcount.
6013 */
6014 ddt_entry_t *dde = ddt_lookup(ddt, bp, B_TRUE);
6015
6016 /*
6017 * ddt_lookup() can return NULL if this block didn't exist
6018 * in the DDT and creating it would take the DDT over its
6019 * quota. Since we got the block from disk, it must exist in
6020 * the DDT, so this can't happen. However, when unique entries
6021 * are pruned, the dedup bit can be set with no corresponding
6022 * entry in the DDT.
6023 */
6024 if (dde == NULL) {
6025 ddt_exit(ddt);
6026 goto skipped;
6027 }
6028
6029 /* Get the phys for this variant */
6030 ddt_phys_variant_t v = ddt_phys_select(ddt, dde, bp);
6031
6032 /*
6033 * This entry may have multiple sets of DVAs. We must claim
6034 * each set the first time we see them in a real block on disk,
6035 * or count them on subsequent occurences. We don't have a
6036 * convenient way to track the first time we see each variant,
6037 * so we repurpose dde_io as a set of "seen" flag bits. We can
6038 * do this safely in zdb because it never writes, so it will
6039 * never have a writing zio for this block in that pointer.
6040 */
6041 boolean_t seen = !!(((uintptr_t)dde->dde_io) & (1 << v));
6042 if (!seen)
6043 dde->dde_io =
6044 (void *)(((uintptr_t)dde->dde_io) | (1 << v));
6045
6046 /* Consume a reference for this block. */
6047 if (ddt_phys_total_refcnt(ddt, dde->dde_phys) > 0)
6048 ddt_phys_decref(dde->dde_phys, v);
6049
6050 /*
6051 * If this entry has a single flat phys, it may have been
6052 * extended with additional DVAs at some time in its life.
6053 * This block might be from before it was fully extended, and
6054 * so have fewer DVAs.
6055 *
6056 * If this is the first time we've seen this block, and we
6057 * claimed it as-is, then we would miss the claim on some
6058 * number of DVAs, which would then be seen as leaked.
6059 *
6060 * In all cases, if we've had fewer DVAs, then the asize would
6061 * be too small, and would lead to the pool apparently using
6062 * more space than allocated.
6063 *
6064 * To handle this, we copy the canonical set of DVAs from the
6065 * entry back to the block pointer before we claim it.
6066 */
6067 if (v == DDT_PHYS_FLAT) {
6068 ASSERT3U(BP_GET_PHYSICAL_BIRTH(bp), ==,
6069 ddt_phys_birth(dde->dde_phys, v));
6070 tempbp = *bp;
6071 ddt_bp_fill(dde->dde_phys, v, &tempbp,
6072 BP_GET_PHYSICAL_BIRTH(bp));
6073 bp = &tempbp;
6074 }
6075
6076 if (seen) {
6077 /*
6078 * The second or later time we see this block,
6079 * it's a duplicate and we count it.
6080 */
6081 zcb->zcb_dedup_asize += BP_GET_ASIZE(bp);
6082 zcb->zcb_dedup_blocks++;
6083
6084 /* Already claimed, don't do it again. */
6085 do_claim = B_FALSE;
6086 }
6087
6088 ddt_exit(ddt);
6089 } else if (zcb->zcb_brt_is_active &&
6090 brt_maybe_exists(zcb->zcb_spa, bp)) {
6091 /*
6092 * Cloned blocks are special. We need to count them, so we can
6093 * later uncount them when reporting leaked space, and we must
6094 * only claim them once.
6095 *
6096 * To do this, we keep our own in-memory BRT. For each block
6097 * we haven't seen before, we look it up in the real BRT and
6098 * if its there, we note it and its refcount then proceed as
6099 * normal. If we see the block again, we count it as a clone
6100 * and then give it no further consideration.
6101 */
6102 zdb_brt_entry_t zbre_search, *zbre;
6103 avl_index_t where;
6104
6105 zbre_search.zbre_dva = bp->blk_dva[0];
6106 zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where);
6107 if (zbre == NULL) {
6108 /* Not seen before; track it */
6109 uint64_t refcnt =
6110 brt_entry_get_refcount(zcb->zcb_spa, bp);
6111 if (refcnt > 0) {
6112 zbre = umem_zalloc(sizeof (zdb_brt_entry_t),
6113 UMEM_NOFAIL);
6114 zbre->zbre_dva = bp->blk_dva[0];
6115 zbre->zbre_refcount = refcnt;
6116 avl_insert(&zcb->zcb_brt, zbre, where);
6117 }
6118 } else {
6119 /*
6120 * Second or later occurrence, count it and take a
6121 * refcount.
6122 */
6123 zcb->zcb_clone_asize += BP_GET_ASIZE(bp);
6124 zcb->zcb_clone_blocks++;
6125
6126 zbre->zbre_refcount--;
6127 if (zbre->zbre_refcount == 0) {
6128 avl_remove(&zcb->zcb_brt, zbre);
6129 umem_free(zbre, sizeof (zdb_brt_entry_t));
6130 }
6131
6132 /* Already claimed, don't do it again. */
6133 do_claim = B_FALSE;
6134 }
6135 }
6136
6137 skipped:
6138 for (i = 0; i < 4; i++) {
6139 int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
6140 int t = (i & 1) ? type : ZDB_OT_TOTAL;
6141 int equal;
6142 zdb_blkstats_t *zb = &zcb->zcb_type[l][t];
6143
6144 zb->zb_asize += BP_GET_ASIZE(bp);
6145 zb->zb_lsize += BP_GET_LSIZE(bp);
6146 zb->zb_psize += BP_GET_PSIZE(bp);
6147 zb->zb_count++;
6148
6149 /*
6150 * The histogram is only big enough to record blocks up to
6151 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
6152 * "other", bucket.
6153 */
6154 unsigned idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT;
6155 idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1);
6156 zb->zb_psize_histogram[idx]++;
6157
6158 zb->zb_gangs += BP_COUNT_GANG(bp);
6159
6160 switch (BP_GET_NDVAS(bp)) {
6161 case 2:
6162 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6163 DVA_GET_VDEV(&bp->blk_dva[1])) {
6164 zb->zb_ditto_samevdev++;
6165
6166 if (same_metaslab(zcb->zcb_spa,
6167 DVA_GET_VDEV(&bp->blk_dva[0]),
6168 DVA_GET_OFFSET(&bp->blk_dva[0]),
6169 DVA_GET_OFFSET(&bp->blk_dva[1])))
6170 zb->zb_ditto_same_ms++;
6171 }
6172 break;
6173 case 3:
6174 equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6175 DVA_GET_VDEV(&bp->blk_dva[1])) +
6176 (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6177 DVA_GET_VDEV(&bp->blk_dva[2])) +
6178 (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6179 DVA_GET_VDEV(&bp->blk_dva[2]));
6180 if (equal != 0) {
6181 zb->zb_ditto_samevdev++;
6182
6183 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6184 DVA_GET_VDEV(&bp->blk_dva[1]) &&
6185 same_metaslab(zcb->zcb_spa,
6186 DVA_GET_VDEV(&bp->blk_dva[0]),
6187 DVA_GET_OFFSET(&bp->blk_dva[0]),
6188 DVA_GET_OFFSET(&bp->blk_dva[1])))
6189 zb->zb_ditto_same_ms++;
6190 else if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6191 DVA_GET_VDEV(&bp->blk_dva[2]) &&
6192 same_metaslab(zcb->zcb_spa,
6193 DVA_GET_VDEV(&bp->blk_dva[0]),
6194 DVA_GET_OFFSET(&bp->blk_dva[0]),
6195 DVA_GET_OFFSET(&bp->blk_dva[2])))
6196 zb->zb_ditto_same_ms++;
6197 else if (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6198 DVA_GET_VDEV(&bp->blk_dva[2]) &&
6199 same_metaslab(zcb->zcb_spa,
6200 DVA_GET_VDEV(&bp->blk_dva[1]),
6201 DVA_GET_OFFSET(&bp->blk_dva[1]),
6202 DVA_GET_OFFSET(&bp->blk_dva[2])))
6203 zb->zb_ditto_same_ms++;
6204 }
6205 break;
6206 }
6207 }
6208
6209 spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
6210
6211 if (BP_IS_EMBEDDED(bp)) {
6212 zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
6213 zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
6214 [BPE_GET_PSIZE(bp)]++;
6215 return;
6216 }
6217
6218 if (block_classes != 0) {
6219 spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
6220
6221 uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[0]);
6222 uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[0]);
6223 vdev_t *vd = vdev_lookup_top(zcb->zcb_spa, vdev);
6224 ASSERT(vd != NULL);
6225 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6226 ASSERT(ms != NULL);
6227 metaslab_group_t *mg = ms->ms_group;
6228 ASSERT(mg != NULL);
6229 metaslab_class_t *mc = mg->mg_class;
6230 ASSERT(mc != NULL);
6231
6232 spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
6233
6234 int class;
6235 if (mc == spa_normal_class(zcb->zcb_spa)) {
6236 class = CLASS_NORMAL;
6237 } else if (mc == spa_special_class(zcb->zcb_spa)) {
6238 class = CLASS_SPECIAL;
6239 } else if (mc == spa_dedup_class(zcb->zcb_spa)) {
6240 class = CLASS_DEDUP;
6241 } else {
6242 class = CLASS_OTHER;
6243 }
6244
6245 if (!(block_classes & class)) {
6246 goto hist_skipped;
6247 }
6248 }
6249
6250 /*
6251 * The binning histogram bins by powers of two up to
6252 * SPA_MAXBLOCKSIZE rather than creating bins for
6253 * every possible blocksize found in the pool.
6254 */
6255 int bin;
6256
6257 /*
6258 * Binning strategy: each bin includes blocks up to and including
6259 * the given size (excluding blocks that fit into the previous bin).
6260 * This way, the "4K" bin includes blocks within the (2K; 4K] range.
6261 */
6262 #define BIN(size) (highbit64((size) - 1))
6263
6264 switch (block_bin_mode) {
6265 case BIN_PSIZE: bin = BIN(BP_GET_PSIZE(bp)); break;
6266 case BIN_LSIZE: bin = BIN(BP_GET_LSIZE(bp)); break;
6267 case BIN_ASIZE: bin = BIN(BP_GET_ASIZE(bp)); break;
6268 case BIN_AUTO: break;
6269 default: PANIC("bad block_bin_mode"); abort();
6270 }
6271
6272 if (block_bin_mode == BIN_AUTO)
6273 bin = BIN(BP_GET_PSIZE(bp));
6274
6275 zcb->zcb_psize_count[bin]++;
6276 zcb->zcb_psize_len[bin] += BP_GET_PSIZE(bp);
6277 zcb->zcb_psize_total += BP_GET_PSIZE(bp);
6278
6279 if (block_bin_mode == BIN_AUTO)
6280 bin = BIN(BP_GET_LSIZE(bp));
6281
6282 zcb->zcb_lsize_count[bin]++;
6283 zcb->zcb_lsize_len[bin] += BP_GET_LSIZE(bp);
6284 zcb->zcb_lsize_total += BP_GET_LSIZE(bp);
6285
6286 if (block_bin_mode == BIN_AUTO)
6287 bin = BIN(BP_GET_ASIZE(bp));
6288
6289 zcb->zcb_asize_count[bin]++;
6290 zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp);
6291 zcb->zcb_asize_total += BP_GET_ASIZE(bp);
6292
6293 #undef BIN
6294
6295 hist_skipped:
6296 if (!do_claim)
6297 return;
6298
6299 VERIFY0(zio_wait(zio_claim(NULL, zcb->zcb_spa,
6300 spa_min_claim_txg(zcb->zcb_spa), bp, NULL, NULL,
6301 ZIO_FLAG_CANFAIL)));
6302 }
6303
6304 static void
zdb_blkptr_done(zio_t * zio)6305 zdb_blkptr_done(zio_t *zio)
6306 {
6307 spa_t *spa = zio->io_spa;
6308 blkptr_t *bp = zio->io_bp;
6309 int ioerr = zio->io_error;
6310 zdb_cb_t *zcb = zio->io_private;
6311 zbookmark_phys_t *zb = &zio->io_bookmark;
6312
6313 mutex_enter(&spa->spa_scrub_lock);
6314 spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
6315 cv_broadcast(&spa->spa_scrub_io_cv);
6316
6317 if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
6318 char blkbuf[BP_SPRINTF_LEN];
6319
6320 zcb->zcb_haderrors = 1;
6321 zcb->zcb_errors[ioerr]++;
6322
6323 if (dump_opt['b'] >= 2)
6324 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6325 else
6326 blkbuf[0] = '\0';
6327
6328 (void) printf("zdb_blkptr_cb: "
6329 "Got error %d reading "
6330 "<%llu, %llu, %lld, %llx> %s -- skipping\n",
6331 ioerr,
6332 (u_longlong_t)zb->zb_objset,
6333 (u_longlong_t)zb->zb_object,
6334 (u_longlong_t)zb->zb_level,
6335 (u_longlong_t)zb->zb_blkid,
6336 blkbuf);
6337 }
6338 mutex_exit(&spa->spa_scrub_lock);
6339
6340 abd_free(zio->io_abd);
6341 }
6342
6343 static int
zdb_blkptr_cb(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)6344 zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
6345 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
6346 {
6347 zdb_cb_t *zcb = arg;
6348 dmu_object_type_t type;
6349 boolean_t is_metadata;
6350
6351 if (zb->zb_level == ZB_DNODE_LEVEL)
6352 return (0);
6353
6354 if (dump_opt['b'] >= 5 && BP_GET_BIRTH(bp) > 0) {
6355 char blkbuf[BP_SPRINTF_LEN];
6356 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6357 (void) printf("objset %llu object %llu "
6358 "level %lld offset 0x%llx %s\n",
6359 (u_longlong_t)zb->zb_objset,
6360 (u_longlong_t)zb->zb_object,
6361 (longlong_t)zb->zb_level,
6362 (u_longlong_t)blkid2offset(dnp, bp, zb),
6363 blkbuf);
6364 }
6365
6366 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
6367 return (0);
6368
6369 type = BP_GET_TYPE(bp);
6370
6371 zdb_count_block(zcb, zilog, bp,
6372 (type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type);
6373
6374 is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
6375
6376 if (!BP_IS_EMBEDDED(bp) &&
6377 (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
6378 size_t size = BP_GET_PSIZE(bp);
6379 abd_t *abd = abd_alloc(size, B_FALSE);
6380 int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
6381
6382 /* If it's an intent log block, failure is expected. */
6383 if (zb->zb_level == ZB_ZIL_LEVEL)
6384 flags |= ZIO_FLAG_SPECULATIVE;
6385
6386 mutex_enter(&spa->spa_scrub_lock);
6387 while (spa->spa_load_verify_bytes > max_inflight_bytes)
6388 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
6389 spa->spa_load_verify_bytes += size;
6390 mutex_exit(&spa->spa_scrub_lock);
6391
6392 zio_nowait(zio_read(NULL, spa, bp, abd, size,
6393 zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
6394 }
6395
6396 zcb->zcb_readfails = 0;
6397
6398 /* only call gethrtime() every 100 blocks */
6399 static int iters;
6400 if (++iters > 100)
6401 iters = 0;
6402 else
6403 return (0);
6404
6405 if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
6406 uint64_t now = gethrtime();
6407 char buf[10];
6408 uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
6409 uint64_t kb_per_sec =
6410 1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000));
6411 uint64_t sec_remaining =
6412 (zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec;
6413
6414 /* make sure nicenum has enough space */
6415 _Static_assert(sizeof (buf) >= NN_NUMBUF_SZ, "buf truncated");
6416
6417 zfs_nicebytes(bytes, buf, sizeof (buf));
6418 (void) fprintf(stderr,
6419 "\r%5s completed (%4"PRIu64"MB/s) "
6420 "estimated time remaining: "
6421 "%"PRIu64"hr %02"PRIu64"min %02"PRIu64"sec ",
6422 buf, kb_per_sec / 1024,
6423 sec_remaining / 60 / 60,
6424 sec_remaining / 60 % 60,
6425 sec_remaining % 60);
6426
6427 zcb->zcb_lastprint = now;
6428 }
6429
6430 return (0);
6431 }
6432
6433 static void
zdb_leak(void * arg,uint64_t start,uint64_t size)6434 zdb_leak(void *arg, uint64_t start, uint64_t size)
6435 {
6436 vdev_t *vd = arg;
6437
6438 (void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
6439 (u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
6440 }
6441
6442 static metaslab_ops_t zdb_metaslab_ops = {
6443 NULL /* alloc */
6444 };
6445
6446 static int
load_unflushed_svr_segs_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6447 load_unflushed_svr_segs_cb(spa_t *spa, space_map_entry_t *sme,
6448 uint64_t txg, void *arg)
6449 {
6450 spa_vdev_removal_t *svr = arg;
6451
6452 uint64_t offset = sme->sme_offset;
6453 uint64_t size = sme->sme_run;
6454
6455 /* skip vdevs we don't care about */
6456 if (sme->sme_vdev != svr->svr_vdev_id)
6457 return (0);
6458
6459 vdev_t *vd = vdev_lookup_top(spa, sme->sme_vdev);
6460 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6461 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6462
6463 if (txg < metaslab_unflushed_txg(ms))
6464 return (0);
6465
6466 if (sme->sme_type == SM_ALLOC)
6467 zfs_range_tree_add(svr->svr_allocd_segs, offset, size);
6468 else
6469 zfs_range_tree_remove(svr->svr_allocd_segs, offset, size);
6470
6471 return (0);
6472 }
6473
6474 static void
claim_segment_impl_cb(uint64_t inner_offset,vdev_t * vd,uint64_t offset,uint64_t size,void * arg)6475 claim_segment_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
6476 uint64_t size, void *arg)
6477 {
6478 (void) inner_offset, (void) arg;
6479
6480 /*
6481 * This callback was called through a remap from
6482 * a device being removed. Therefore, the vdev that
6483 * this callback is applied to is a concrete
6484 * vdev.
6485 */
6486 ASSERT(vdev_is_concrete(vd));
6487
6488 VERIFY0(metaslab_claim_impl(vd, offset, size,
6489 spa_min_claim_txg(vd->vdev_spa)));
6490 }
6491
6492 static void
claim_segment_cb(void * arg,uint64_t offset,uint64_t size)6493 claim_segment_cb(void *arg, uint64_t offset, uint64_t size)
6494 {
6495 vdev_t *vd = arg;
6496
6497 vdev_indirect_ops.vdev_op_remap(vd, offset, size,
6498 claim_segment_impl_cb, NULL);
6499 }
6500
6501 /*
6502 * After accounting for all allocated blocks that are directly referenced,
6503 * we might have missed a reference to a block from a partially complete
6504 * (and thus unused) indirect mapping object. We perform a secondary pass
6505 * through the metaslabs we have already mapped and claim the destination
6506 * blocks.
6507 */
6508 static void
zdb_claim_removing(spa_t * spa,zdb_cb_t * zcb)6509 zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb)
6510 {
6511 if (dump_opt['L'])
6512 return;
6513
6514 if (spa->spa_vdev_removal == NULL)
6515 return;
6516
6517 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6518
6519 spa_vdev_removal_t *svr = spa->spa_vdev_removal;
6520 vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id);
6521 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6522
6523 ASSERT0(zfs_range_tree_space(svr->svr_allocd_segs));
6524
6525 zfs_range_tree_t *allocs = zfs_range_tree_create_flags(
6526 NULL, ZFS_RANGE_SEG64, NULL, 0, 0,
6527 0, "zdb_claim_removing:allocs");
6528 for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) {
6529 metaslab_t *msp = vd->vdev_ms[msi];
6530
6531 ASSERT0(zfs_range_tree_space(allocs));
6532 if (msp->ms_sm != NULL)
6533 VERIFY0(space_map_load(msp->ms_sm, allocs, SM_ALLOC));
6534 zfs_range_tree_vacate(allocs, zfs_range_tree_add,
6535 svr->svr_allocd_segs);
6536 }
6537 zfs_range_tree_destroy(allocs);
6538
6539 iterate_through_spacemap_logs(spa, load_unflushed_svr_segs_cb, svr);
6540
6541 /*
6542 * Clear everything past what has been synced,
6543 * because we have not allocated mappings for
6544 * it yet.
6545 */
6546 zfs_range_tree_clear(svr->svr_allocd_segs,
6547 vdev_indirect_mapping_max_offset(vim),
6548 vd->vdev_asize - vdev_indirect_mapping_max_offset(vim));
6549
6550 zcb->zcb_removing_size += zfs_range_tree_space(svr->svr_allocd_segs);
6551 zfs_range_tree_vacate(svr->svr_allocd_segs, claim_segment_cb, vd);
6552
6553 spa_config_exit(spa, SCL_CONFIG, FTAG);
6554 }
6555
6556 static int
increment_indirect_mapping_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6557 increment_indirect_mapping_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6558 dmu_tx_t *tx)
6559 {
6560 (void) tx;
6561 zdb_cb_t *zcb = arg;
6562 spa_t *spa = zcb->zcb_spa;
6563 vdev_t *vd;
6564 const dva_t *dva = &bp->blk_dva[0];
6565
6566 ASSERT(!bp_freed);
6567 ASSERT(!dump_opt['L']);
6568 ASSERT3U(BP_GET_NDVAS(bp), ==, 1);
6569
6570 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6571 vd = vdev_lookup_top(zcb->zcb_spa, DVA_GET_VDEV(dva));
6572 ASSERT3P(vd, !=, NULL);
6573 spa_config_exit(spa, SCL_VDEV, FTAG);
6574
6575 ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0);
6576 ASSERT3P(zcb->zcb_vd_obsolete_counts[vd->vdev_id], !=, NULL);
6577
6578 vdev_indirect_mapping_increment_obsolete_count(
6579 vd->vdev_indirect_mapping,
6580 DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva),
6581 zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6582
6583 return (0);
6584 }
6585
6586 static uint32_t *
zdb_load_obsolete_counts(vdev_t * vd)6587 zdb_load_obsolete_counts(vdev_t *vd)
6588 {
6589 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6590 spa_t *spa = vd->vdev_spa;
6591 spa_condensing_indirect_phys_t *scip =
6592 &spa->spa_condensing_indirect_phys;
6593 uint64_t obsolete_sm_object;
6594 uint32_t *counts;
6595
6596 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
6597 EQUIV(obsolete_sm_object != 0, vd->vdev_obsolete_sm != NULL);
6598 counts = vdev_indirect_mapping_load_obsolete_counts(vim);
6599 if (vd->vdev_obsolete_sm != NULL) {
6600 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6601 vd->vdev_obsolete_sm);
6602 }
6603 if (scip->scip_vdev == vd->vdev_id &&
6604 scip->scip_prev_obsolete_sm_object != 0) {
6605 space_map_t *prev_obsolete_sm = NULL;
6606 VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset,
6607 scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0));
6608 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6609 prev_obsolete_sm);
6610 space_map_close(prev_obsolete_sm);
6611 }
6612 return (counts);
6613 }
6614
6615 typedef struct checkpoint_sm_exclude_entry_arg {
6616 vdev_t *cseea_vd;
6617 uint64_t cseea_checkpoint_size;
6618 } checkpoint_sm_exclude_entry_arg_t;
6619
6620 static int
checkpoint_sm_exclude_entry_cb(space_map_entry_t * sme,void * arg)6621 checkpoint_sm_exclude_entry_cb(space_map_entry_t *sme, void *arg)
6622 {
6623 checkpoint_sm_exclude_entry_arg_t *cseea = arg;
6624 vdev_t *vd = cseea->cseea_vd;
6625 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
6626 uint64_t end = sme->sme_offset + sme->sme_run;
6627
6628 ASSERT(sme->sme_type == SM_FREE);
6629
6630 /*
6631 * Since the vdev_checkpoint_sm exists in the vdev level
6632 * and the ms_sm space maps exist in the metaslab level,
6633 * an entry in the checkpoint space map could theoretically
6634 * cross the boundaries of the metaslab that it belongs.
6635 *
6636 * In reality, because of the way that we populate and
6637 * manipulate the checkpoint's space maps currently,
6638 * there shouldn't be any entries that cross metaslabs.
6639 * Hence the assertion below.
6640 *
6641 * That said, there is no fundamental requirement that
6642 * the checkpoint's space map entries should not cross
6643 * metaslab boundaries. So if needed we could add code
6644 * that handles metaslab-crossing segments in the future.
6645 */
6646 VERIFY3U(sme->sme_offset, >=, ms->ms_start);
6647 VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
6648
6649 /*
6650 * By removing the entry from the allocated segments we
6651 * also verify that the entry is there to begin with.
6652 */
6653 mutex_enter(&ms->ms_lock);
6654 zfs_range_tree_remove(ms->ms_allocatable, sme->sme_offset,
6655 sme->sme_run);
6656 mutex_exit(&ms->ms_lock);
6657
6658 cseea->cseea_checkpoint_size += sme->sme_run;
6659 return (0);
6660 }
6661
6662 static void
zdb_leak_init_vdev_exclude_checkpoint(vdev_t * vd,zdb_cb_t * zcb)6663 zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_cb_t *zcb)
6664 {
6665 spa_t *spa = vd->vdev_spa;
6666 space_map_t *checkpoint_sm = NULL;
6667 uint64_t checkpoint_sm_obj;
6668
6669 /*
6670 * If there is no vdev_top_zap, we are in a pool whose
6671 * version predates the pool checkpoint feature.
6672 */
6673 if (vd->vdev_top_zap == 0)
6674 return;
6675
6676 /*
6677 * If there is no reference of the vdev_checkpoint_sm in
6678 * the vdev_top_zap, then one of the following scenarios
6679 * is true:
6680 *
6681 * 1] There is no checkpoint
6682 * 2] There is a checkpoint, but no checkpointed blocks
6683 * have been freed yet
6684 * 3] The current vdev is indirect
6685 *
6686 * In these cases we return immediately.
6687 */
6688 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
6689 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
6690 return;
6691
6692 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
6693 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1,
6694 &checkpoint_sm_obj));
6695
6696 checkpoint_sm_exclude_entry_arg_t cseea;
6697 cseea.cseea_vd = vd;
6698 cseea.cseea_checkpoint_size = 0;
6699
6700 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
6701 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
6702
6703 VERIFY0(space_map_iterate(checkpoint_sm,
6704 space_map_length(checkpoint_sm),
6705 checkpoint_sm_exclude_entry_cb, &cseea));
6706 space_map_close(checkpoint_sm);
6707
6708 zcb->zcb_checkpoint_size += cseea.cseea_checkpoint_size;
6709 }
6710
6711 static void
zdb_leak_init_exclude_checkpoint(spa_t * spa,zdb_cb_t * zcb)6712 zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb)
6713 {
6714 ASSERT(!dump_opt['L']);
6715
6716 vdev_t *rvd = spa->spa_root_vdev;
6717 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6718 ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id);
6719 zdb_leak_init_vdev_exclude_checkpoint(rvd->vdev_child[c], zcb);
6720 }
6721 }
6722
6723 static int
count_unflushed_space_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6724 count_unflushed_space_cb(spa_t *spa, space_map_entry_t *sme,
6725 uint64_t txg, void *arg)
6726 {
6727 int64_t *ualloc_space = arg;
6728
6729 uint64_t offset = sme->sme_offset;
6730 uint64_t vdev_id = sme->sme_vdev;
6731
6732 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6733 if (!vdev_is_concrete(vd))
6734 return (0);
6735
6736 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6737 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6738
6739 if (txg < metaslab_unflushed_txg(ms))
6740 return (0);
6741
6742 if (sme->sme_type == SM_ALLOC)
6743 *ualloc_space += sme->sme_run;
6744 else
6745 *ualloc_space -= sme->sme_run;
6746
6747 return (0);
6748 }
6749
6750 static int64_t
get_unflushed_alloc_space(spa_t * spa)6751 get_unflushed_alloc_space(spa_t *spa)
6752 {
6753 if (dump_opt['L'])
6754 return (0);
6755
6756 int64_t ualloc_space = 0;
6757 iterate_through_spacemap_logs(spa, count_unflushed_space_cb,
6758 &ualloc_space);
6759 return (ualloc_space);
6760 }
6761
6762 static int
load_unflushed_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6763 load_unflushed_cb(spa_t *spa, space_map_entry_t *sme, uint64_t txg, void *arg)
6764 {
6765 maptype_t *uic_maptype = arg;
6766
6767 uint64_t offset = sme->sme_offset;
6768 uint64_t size = sme->sme_run;
6769 uint64_t vdev_id = sme->sme_vdev;
6770
6771 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6772
6773 /* skip indirect vdevs */
6774 if (!vdev_is_concrete(vd))
6775 return (0);
6776
6777 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6778
6779 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6780 ASSERT(*uic_maptype == SM_ALLOC || *uic_maptype == SM_FREE);
6781
6782 if (txg < metaslab_unflushed_txg(ms))
6783 return (0);
6784
6785 if (*uic_maptype == sme->sme_type)
6786 zfs_range_tree_add(ms->ms_allocatable, offset, size);
6787 else
6788 zfs_range_tree_remove(ms->ms_allocatable, offset, size);
6789
6790 return (0);
6791 }
6792
6793 static void
load_unflushed_to_ms_allocatables(spa_t * spa,maptype_t maptype)6794 load_unflushed_to_ms_allocatables(spa_t *spa, maptype_t maptype)
6795 {
6796 iterate_through_spacemap_logs(spa, load_unflushed_cb, &maptype);
6797 }
6798
6799 static void
load_concrete_ms_allocatable_trees(spa_t * spa,maptype_t maptype)6800 load_concrete_ms_allocatable_trees(spa_t *spa, maptype_t maptype)
6801 {
6802 vdev_t *rvd = spa->spa_root_vdev;
6803 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
6804 vdev_t *vd = rvd->vdev_child[i];
6805
6806 ASSERT3U(i, ==, vd->vdev_id);
6807
6808 if (vd->vdev_ops == &vdev_indirect_ops)
6809 continue;
6810
6811 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6812 metaslab_t *msp = vd->vdev_ms[m];
6813
6814 (void) fprintf(stderr,
6815 "\rloading concrete vdev %llu, "
6816 "metaslab %llu of %llu ...",
6817 (longlong_t)vd->vdev_id,
6818 (longlong_t)msp->ms_id,
6819 (longlong_t)vd->vdev_ms_count);
6820
6821 mutex_enter(&msp->ms_lock);
6822 zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6823
6824 /*
6825 * We don't want to spend the CPU manipulating the
6826 * size-ordered tree, so clear the range_tree ops.
6827 */
6828 msp->ms_allocatable->rt_ops = NULL;
6829
6830 if (msp->ms_sm != NULL) {
6831 VERIFY0(space_map_load(msp->ms_sm,
6832 msp->ms_allocatable, maptype));
6833 }
6834 if (!msp->ms_loaded)
6835 msp->ms_loaded = B_TRUE;
6836 mutex_exit(&msp->ms_lock);
6837 }
6838 }
6839
6840 load_unflushed_to_ms_allocatables(spa, maptype);
6841 }
6842
6843 /*
6844 * vm_idxp is an in-out parameter which (for indirect vdevs) is the
6845 * index in vim_entries that has the first entry in this metaslab.
6846 * On return, it will be set to the first entry after this metaslab.
6847 */
6848 static void
load_indirect_ms_allocatable_tree(vdev_t * vd,metaslab_t * msp,uint64_t * vim_idxp)6849 load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab_t *msp,
6850 uint64_t *vim_idxp)
6851 {
6852 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6853
6854 mutex_enter(&msp->ms_lock);
6855 zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6856
6857 /*
6858 * We don't want to spend the CPU manipulating the
6859 * size-ordered tree, so clear the range_tree ops.
6860 */
6861 msp->ms_allocatable->rt_ops = NULL;
6862
6863 for (; *vim_idxp < vdev_indirect_mapping_num_entries(vim);
6864 (*vim_idxp)++) {
6865 vdev_indirect_mapping_entry_phys_t *vimep =
6866 &vim->vim_entries[*vim_idxp];
6867 uint64_t ent_offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6868 uint64_t ent_len = DVA_GET_ASIZE(&vimep->vimep_dst);
6869 ASSERT3U(ent_offset, >=, msp->ms_start);
6870 if (ent_offset >= msp->ms_start + msp->ms_size)
6871 break;
6872
6873 /*
6874 * Mappings do not cross metaslab boundaries,
6875 * because we create them by walking the metaslabs.
6876 */
6877 ASSERT3U(ent_offset + ent_len, <=,
6878 msp->ms_start + msp->ms_size);
6879 zfs_range_tree_add(msp->ms_allocatable, ent_offset, ent_len);
6880 }
6881
6882 if (!msp->ms_loaded)
6883 msp->ms_loaded = B_TRUE;
6884 mutex_exit(&msp->ms_lock);
6885 }
6886
6887 static void
zdb_leak_init_prepare_indirect_vdevs(spa_t * spa,zdb_cb_t * zcb)6888 zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb)
6889 {
6890 ASSERT(!dump_opt['L']);
6891
6892 vdev_t *rvd = spa->spa_root_vdev;
6893 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6894 vdev_t *vd = rvd->vdev_child[c];
6895
6896 ASSERT3U(c, ==, vd->vdev_id);
6897
6898 if (vd->vdev_ops != &vdev_indirect_ops)
6899 continue;
6900
6901 /*
6902 * Note: we don't check for mapping leaks on
6903 * removing vdevs because their ms_allocatable's
6904 * are used to look for leaks in allocated space.
6905 */
6906 zcb->zcb_vd_obsolete_counts[c] = zdb_load_obsolete_counts(vd);
6907
6908 /*
6909 * Normally, indirect vdevs don't have any
6910 * metaslabs. We want to set them up for
6911 * zio_claim().
6912 */
6913 vdev_metaslab_group_create(vd);
6914 VERIFY0(vdev_metaslab_init(vd, 0));
6915
6916 vdev_indirect_mapping_t *vim __maybe_unused =
6917 vd->vdev_indirect_mapping;
6918 uint64_t vim_idx = 0;
6919 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6920
6921 (void) fprintf(stderr,
6922 "\rloading indirect vdev %llu, "
6923 "metaslab %llu of %llu ...",
6924 (longlong_t)vd->vdev_id,
6925 (longlong_t)vd->vdev_ms[m]->ms_id,
6926 (longlong_t)vd->vdev_ms_count);
6927
6928 load_indirect_ms_allocatable_tree(vd, vd->vdev_ms[m],
6929 &vim_idx);
6930 }
6931 ASSERT3U(vim_idx, ==, vdev_indirect_mapping_num_entries(vim));
6932 }
6933 }
6934
6935 static void
zdb_leak_init(spa_t * spa,zdb_cb_t * zcb)6936 zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
6937 {
6938 zcb->zcb_spa = spa;
6939
6940 if (dump_opt['L'])
6941 return;
6942
6943 dsl_pool_t *dp = spa->spa_dsl_pool;
6944 vdev_t *rvd = spa->spa_root_vdev;
6945
6946 /*
6947 * We are going to be changing the meaning of the metaslab's
6948 * ms_allocatable. Ensure that the allocator doesn't try to
6949 * use the tree.
6950 */
6951 spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
6952 spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
6953 spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6954 spa->spa_special_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6955
6956 zcb->zcb_vd_obsolete_counts =
6957 umem_zalloc(rvd->vdev_children * sizeof (uint32_t *),
6958 UMEM_NOFAIL);
6959
6960 /*
6961 * For leak detection, we overload the ms_allocatable trees
6962 * to contain allocated segments instead of free segments.
6963 * As a result, we can't use the normal metaslab_load/unload
6964 * interfaces.
6965 */
6966 zdb_leak_init_prepare_indirect_vdevs(spa, zcb);
6967 load_concrete_ms_allocatable_trees(spa, SM_ALLOC);
6968
6969 /*
6970 * On load_concrete_ms_allocatable_trees() we loaded all the
6971 * allocated entries from the ms_sm to the ms_allocatable for
6972 * each metaslab. If the pool has a checkpoint or is in the
6973 * middle of discarding a checkpoint, some of these blocks
6974 * may have been freed but their ms_sm may not have been
6975 * updated because they are referenced by the checkpoint. In
6976 * order to avoid false-positives during leak-detection, we
6977 * go through the vdev's checkpoint space map and exclude all
6978 * its entries from their relevant ms_allocatable.
6979 *
6980 * We also aggregate the space held by the checkpoint and add
6981 * it to zcb_checkpoint_size.
6982 *
6983 * Note that at this point we are also verifying that all the
6984 * entries on the checkpoint_sm are marked as allocated in
6985 * the ms_sm of their relevant metaslab.
6986 * [see comment in checkpoint_sm_exclude_entry_cb()]
6987 */
6988 zdb_leak_init_exclude_checkpoint(spa, zcb);
6989 ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa));
6990
6991 /* for cleaner progress output */
6992 (void) fprintf(stderr, "\n");
6993
6994 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
6995 ASSERT(spa_feature_is_enabled(spa,
6996 SPA_FEATURE_DEVICE_REMOVAL));
6997 (void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj,
6998 increment_indirect_mapping_cb, zcb, NULL);
6999 }
7000 }
7001
7002 static boolean_t
zdb_check_for_obsolete_leaks(vdev_t * vd,zdb_cb_t * zcb)7003 zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb)
7004 {
7005 boolean_t leaks = B_FALSE;
7006 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7007 uint64_t total_leaked = 0;
7008 boolean_t are_precise = B_FALSE;
7009
7010 ASSERT(vim != NULL);
7011
7012 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
7013 vdev_indirect_mapping_entry_phys_t *vimep =
7014 &vim->vim_entries[i];
7015 uint64_t obsolete_bytes = 0;
7016 uint64_t offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
7017 metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
7018
7019 /*
7020 * This is not very efficient but it's easy to
7021 * verify correctness.
7022 */
7023 for (uint64_t inner_offset = 0;
7024 inner_offset < DVA_GET_ASIZE(&vimep->vimep_dst);
7025 inner_offset += 1ULL << vd->vdev_ashift) {
7026 if (zfs_range_tree_contains(msp->ms_allocatable,
7027 offset + inner_offset, 1ULL << vd->vdev_ashift)) {
7028 obsolete_bytes += 1ULL << vd->vdev_ashift;
7029 }
7030 }
7031
7032 int64_t bytes_leaked = obsolete_bytes -
7033 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i];
7034 ASSERT3U(DVA_GET_ASIZE(&vimep->vimep_dst), >=,
7035 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]);
7036
7037 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7038 if (bytes_leaked != 0 && (are_precise || dump_opt['d'] >= 5)) {
7039 (void) printf("obsolete indirect mapping count "
7040 "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
7041 (u_longlong_t)vd->vdev_id,
7042 (u_longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
7043 (u_longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
7044 (u_longlong_t)bytes_leaked);
7045 }
7046 total_leaked += ABS(bytes_leaked);
7047 }
7048
7049 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7050 if (!are_precise && total_leaked > 0) {
7051 int pct_leaked = total_leaked * 100 /
7052 vdev_indirect_mapping_bytes_mapped(vim);
7053 (void) printf("cannot verify obsolete indirect mapping "
7054 "counts of vdev %llu because precise feature was not "
7055 "enabled when it was removed: %d%% (%llx bytes) of mapping"
7056 "unreferenced\n",
7057 (u_longlong_t)vd->vdev_id, pct_leaked,
7058 (u_longlong_t)total_leaked);
7059 } else if (total_leaked > 0) {
7060 (void) printf("obsolete indirect mapping count mismatch "
7061 "for vdev %llu -- %llx total bytes mismatched\n",
7062 (u_longlong_t)vd->vdev_id,
7063 (u_longlong_t)total_leaked);
7064 leaks |= B_TRUE;
7065 }
7066
7067 vdev_indirect_mapping_free_obsolete_counts(vim,
7068 zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
7069 zcb->zcb_vd_obsolete_counts[vd->vdev_id] = NULL;
7070
7071 return (leaks);
7072 }
7073
7074 static boolean_t
zdb_leak_fini(spa_t * spa,zdb_cb_t * zcb)7075 zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb)
7076 {
7077 if (dump_opt['L'])
7078 return (B_FALSE);
7079
7080 boolean_t leaks = B_FALSE;
7081 vdev_t *rvd = spa->spa_root_vdev;
7082 for (unsigned c = 0; c < rvd->vdev_children; c++) {
7083 vdev_t *vd = rvd->vdev_child[c];
7084
7085 if (zcb->zcb_vd_obsolete_counts[c] != NULL) {
7086 leaks |= zdb_check_for_obsolete_leaks(vd, zcb);
7087 }
7088
7089 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
7090 metaslab_t *msp = vd->vdev_ms[m];
7091 ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class ==
7092 spa_embedded_log_class(spa) ||
7093 msp->ms_group->mg_class ==
7094 spa_special_embedded_log_class(spa)) ?
7095 vd->vdev_log_mg : vd->vdev_mg);
7096
7097 /*
7098 * ms_allocatable has been overloaded
7099 * to contain allocated segments. Now that
7100 * we finished traversing all blocks, any
7101 * block that remains in the ms_allocatable
7102 * represents an allocated block that we
7103 * did not claim during the traversal.
7104 * Claimed blocks would have been removed
7105 * from the ms_allocatable. For indirect
7106 * vdevs, space remaining in the tree
7107 * represents parts of the mapping that are
7108 * not referenced, which is not a bug.
7109 */
7110 if (vd->vdev_ops == &vdev_indirect_ops) {
7111 zfs_range_tree_vacate(msp->ms_allocatable,
7112 NULL, NULL);
7113 } else {
7114 zfs_range_tree_vacate(msp->ms_allocatable,
7115 zdb_leak, vd);
7116 }
7117 if (msp->ms_loaded) {
7118 msp->ms_loaded = B_FALSE;
7119 }
7120 }
7121 }
7122
7123 umem_free(zcb->zcb_vd_obsolete_counts,
7124 rvd->vdev_children * sizeof (uint32_t *));
7125 zcb->zcb_vd_obsolete_counts = NULL;
7126
7127 return (leaks);
7128 }
7129
7130 static int
count_block_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)7131 count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7132 {
7133 (void) tx;
7134 zdb_cb_t *zcb = arg;
7135
7136 if (dump_opt['b'] >= 5) {
7137 char blkbuf[BP_SPRINTF_LEN];
7138 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
7139 (void) printf("[%s] %s\n",
7140 "deferred free", blkbuf);
7141 }
7142 zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED);
7143 return (0);
7144 }
7145
7146 /*
7147 * Iterate over livelists which have been destroyed by the user but
7148 * are still present in the MOS, waiting to be freed
7149 */
7150 static void
iterate_deleted_livelists(spa_t * spa,ll_iter_t func,void * arg)7151 iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg)
7152 {
7153 objset_t *mos = spa->spa_meta_objset;
7154 uint64_t zap_obj;
7155 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
7156 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
7157 if (err == ENOENT)
7158 return;
7159 ASSERT0(err);
7160
7161 zap_cursor_t zc;
7162 zap_attribute_t *attrp = zap_attribute_alloc();
7163 dsl_deadlist_t ll;
7164 /* NULL out os prior to dsl_deadlist_open in case it's garbage */
7165 ll.dl_os = NULL;
7166 for (zap_cursor_init(&zc, mos, zap_obj);
7167 zap_cursor_retrieve(&zc, attrp) == 0;
7168 (void) zap_cursor_advance(&zc)) {
7169 VERIFY0(dsl_deadlist_open(&ll, mos, attrp->za_first_integer));
7170 func(&ll, arg);
7171 dsl_deadlist_close(&ll);
7172 }
7173 zap_cursor_fini(&zc);
7174 zap_attribute_free(attrp);
7175 }
7176
7177 static int
bpobj_count_block_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)7178 bpobj_count_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
7179 dmu_tx_t *tx)
7180 {
7181 ASSERT(!bp_freed);
7182 return (count_block_cb(arg, bp, tx));
7183 }
7184
7185 static int
livelist_entry_count_blocks_cb(void * args,dsl_deadlist_entry_t * dle)7186 livelist_entry_count_blocks_cb(void *args, dsl_deadlist_entry_t *dle)
7187 {
7188 zdb_cb_t *zbc = args;
7189 bplist_t blks;
7190 bplist_create(&blks);
7191 /* determine which blocks have been alloc'd but not freed */
7192 VERIFY0(dsl_process_sub_livelist(&dle->dle_bpobj, &blks, NULL, NULL));
7193 /* count those blocks */
7194 (void) bplist_iterate(&blks, count_block_cb, zbc, NULL);
7195 bplist_destroy(&blks);
7196 return (0);
7197 }
7198
7199 static void
livelist_count_blocks(dsl_deadlist_t * ll,void * arg)7200 livelist_count_blocks(dsl_deadlist_t *ll, void *arg)
7201 {
7202 dsl_deadlist_iterate(ll, livelist_entry_count_blocks_cb, arg);
7203 }
7204
7205 /*
7206 * Count the blocks in the livelists that have been destroyed by the user
7207 * but haven't yet been freed.
7208 */
7209 static void
deleted_livelists_count_blocks(spa_t * spa,zdb_cb_t * zbc)7210 deleted_livelists_count_blocks(spa_t *spa, zdb_cb_t *zbc)
7211 {
7212 iterate_deleted_livelists(spa, livelist_count_blocks, zbc);
7213 }
7214
7215 static void
dump_livelist_cb(dsl_deadlist_t * ll,void * arg)7216 dump_livelist_cb(dsl_deadlist_t *ll, void *arg)
7217 {
7218 ASSERT0P(arg);
7219 global_feature_count[SPA_FEATURE_LIVELIST]++;
7220 dump_blkptr_list(ll, "Deleted Livelist");
7221 dsl_deadlist_iterate(ll, sublivelist_verify_lightweight, NULL);
7222 }
7223
7224 /*
7225 * Print out, register object references to, and increment feature counts for
7226 * livelists that have been destroyed by the user but haven't yet been freed.
7227 */
7228 static void
deleted_livelists_dump_mos(spa_t * spa)7229 deleted_livelists_dump_mos(spa_t *spa)
7230 {
7231 uint64_t zap_obj;
7232 objset_t *mos = spa->spa_meta_objset;
7233 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
7234 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
7235 if (err == ENOENT)
7236 return;
7237 mos_obj_refd(zap_obj);
7238 iterate_deleted_livelists(spa, dump_livelist_cb, NULL);
7239 }
7240
7241 static int
zdb_brt_entry_compare(const void * zcn1,const void * zcn2)7242 zdb_brt_entry_compare(const void *zcn1, const void *zcn2)
7243 {
7244 const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva;
7245 const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva;
7246 int cmp;
7247
7248 cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
7249 if (cmp == 0)
7250 cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2));
7251
7252 return (cmp);
7253 }
7254
7255 static int
dump_block_stats(spa_t * spa)7256 dump_block_stats(spa_t *spa)
7257 {
7258 zdb_cb_t *zcb;
7259 zdb_blkstats_t *zb, *tzb;
7260 uint64_t norm_alloc, norm_space, total_alloc, total_found;
7261 int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7262 TRAVERSE_NO_DECRYPT | TRAVERSE_HARD;
7263 boolean_t leaks = B_FALSE;
7264 int e, c, err;
7265 bp_embedded_type_t i;
7266
7267 ddt_prefetch_all(spa);
7268
7269 zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL);
7270
7271 if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
7272 avl_create(&zcb->zcb_brt, zdb_brt_entry_compare,
7273 sizeof (zdb_brt_entry_t),
7274 offsetof(zdb_brt_entry_t, zbre_node));
7275 zcb->zcb_brt_is_active = B_TRUE;
7276 }
7277
7278 (void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
7279 (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
7280 (dump_opt['c'] == 1) ? "metadata " : "",
7281 dump_opt['c'] ? "checksums " : "",
7282 (dump_opt['c'] && !dump_opt['L']) ? "and verify " : "",
7283 !dump_opt['L'] ? "nothing leaked " : "");
7284
7285 /*
7286 * When leak detection is enabled we load all space maps as SM_ALLOC
7287 * maps, then traverse the pool claiming each block we discover. If
7288 * the pool is perfectly consistent, the segment trees will be empty
7289 * when we're done. Anything left over is a leak; any block we can't
7290 * claim (because it's not part of any space map) is a double
7291 * allocation, reference to a freed block, or an unclaimed log block.
7292 *
7293 * When leak detection is disabled (-L option) we still traverse the
7294 * pool claiming each block we discover, but we skip opening any space
7295 * maps.
7296 */
7297 zdb_leak_init(spa, zcb);
7298
7299 /*
7300 * If there's a deferred-free bplist, process that first.
7301 */
7302 (void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
7303 bpobj_count_block_cb, zcb, NULL);
7304
7305 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
7306 (void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
7307 bpobj_count_block_cb, zcb, NULL);
7308 }
7309
7310 zdb_claim_removing(spa, zcb);
7311
7312 if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
7313 VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,
7314 spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb,
7315 zcb, NULL));
7316 }
7317
7318 deleted_livelists_count_blocks(spa, zcb);
7319
7320 if (dump_opt['c'] > 1)
7321 flags |= TRAVERSE_PREFETCH_DATA;
7322
7323 zcb->zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
7324 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa));
7325 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
7326 zcb->zcb_totalasize +=
7327 metaslab_class_get_alloc(spa_embedded_log_class(spa));
7328 zcb->zcb_totalasize +=
7329 metaslab_class_get_alloc(spa_special_embedded_log_class(spa));
7330 zcb->zcb_start = zcb->zcb_lastprint = gethrtime();
7331 err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb);
7332
7333 /*
7334 * If we've traversed the data blocks then we need to wait for those
7335 * I/Os to complete. We leverage "The Godfather" zio to wait on
7336 * all async I/Os to complete.
7337 */
7338 if (dump_opt['c']) {
7339 for (c = 0; c < max_ncpus; c++) {
7340 (void) zio_wait(spa->spa_async_zio_root[c]);
7341 spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL,
7342 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
7343 ZIO_FLAG_GODFATHER);
7344 }
7345 }
7346 ASSERT0(spa->spa_load_verify_bytes);
7347
7348 /*
7349 * Done after zio_wait() since zcb_haderrors is modified in
7350 * zdb_blkptr_done()
7351 */
7352 zcb->zcb_haderrors |= err;
7353
7354 if (zcb->zcb_haderrors) {
7355 (void) printf("\nError counts:\n\n");
7356 (void) printf("\t%5s %s\n", "errno", "count");
7357 for (e = 0; e < 256; e++) {
7358 if (zcb->zcb_errors[e] != 0) {
7359 (void) printf("\t%5d %llu\n",
7360 e, (u_longlong_t)zcb->zcb_errors[e]);
7361 }
7362 }
7363 }
7364
7365 /*
7366 * Report any leaked segments.
7367 */
7368 leaks |= zdb_leak_fini(spa, zcb);
7369
7370 tzb = &zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL];
7371
7372 norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7373 norm_space = metaslab_class_get_space(spa_normal_class(spa));
7374
7375 total_alloc = norm_alloc +
7376 metaslab_class_get_alloc(spa_log_class(spa)) +
7377 metaslab_class_get_alloc(spa_embedded_log_class(spa)) +
7378 metaslab_class_get_alloc(spa_special_embedded_log_class(spa)) +
7379 metaslab_class_get_alloc(spa_special_class(spa)) +
7380 metaslab_class_get_alloc(spa_dedup_class(spa)) +
7381 get_unflushed_alloc_space(spa);
7382 total_found =
7383 tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize +
7384 zcb->zcb_removing_size + zcb->zcb_checkpoint_size;
7385
7386 if (total_found == total_alloc && !dump_opt['L']) {
7387 (void) printf("\n\tNo leaks (block sum matches space"
7388 " maps exactly)\n");
7389 } else if (!dump_opt['L']) {
7390 (void) printf("block traversal size %llu != alloc %llu "
7391 "(%s %lld)\n",
7392 (u_longlong_t)total_found,
7393 (u_longlong_t)total_alloc,
7394 (dump_opt['L']) ? "unreachable" : "leaked",
7395 (longlong_t)(total_alloc - total_found));
7396 }
7397
7398 if (tzb->zb_count == 0) {
7399 umem_free(zcb, sizeof (zdb_cb_t));
7400 return (2);
7401 }
7402
7403 (void) printf("\n");
7404 (void) printf("\t%-16s %14llu\n", "bp count:",
7405 (u_longlong_t)tzb->zb_count);
7406 (void) printf("\t%-16s %14llu\n", "ganged count:",
7407 (longlong_t)tzb->zb_gangs);
7408 (void) printf("\t%-16s %14llu avg: %6llu\n", "bp logical:",
7409 (u_longlong_t)tzb->zb_lsize,
7410 (u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
7411 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7412 "bp physical:", (u_longlong_t)tzb->zb_psize,
7413 (u_longlong_t)(tzb->zb_psize / tzb->zb_count),
7414 (double)tzb->zb_lsize / tzb->zb_psize);
7415 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7416 "bp allocated:", (u_longlong_t)tzb->zb_asize,
7417 (u_longlong_t)(tzb->zb_asize / tzb->zb_count),
7418 (double)tzb->zb_lsize / tzb->zb_asize);
7419 (void) printf("\t%-16s %14llu ref>1: %6llu deduplication: %6.2f\n",
7420 "bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize,
7421 (u_longlong_t)zcb->zcb_dedup_blocks,
7422 (double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0);
7423 (void) printf("\t%-16s %14llu count: %6llu\n",
7424 "bp cloned:", (u_longlong_t)zcb->zcb_clone_asize,
7425 (u_longlong_t)zcb->zcb_clone_blocks);
7426 (void) printf("\t%-16s %14llu used: %5.2f%%\n", "Normal class:",
7427 (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
7428
7429 if (spa_special_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7430 uint64_t alloc = metaslab_class_get_alloc(
7431 spa_special_class(spa));
7432 uint64_t space = metaslab_class_get_space(
7433 spa_special_class(spa));
7434
7435 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7436 "Special class", (u_longlong_t)alloc,
7437 100.0 * alloc / space);
7438 }
7439
7440 if (spa_dedup_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7441 uint64_t alloc = metaslab_class_get_alloc(
7442 spa_dedup_class(spa));
7443 uint64_t space = metaslab_class_get_space(
7444 spa_dedup_class(spa));
7445
7446 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7447 "Dedup class", (u_longlong_t)alloc,
7448 100.0 * alloc / space);
7449 }
7450
7451 if (spa_embedded_log_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7452 uint64_t alloc = metaslab_class_get_alloc(
7453 spa_embedded_log_class(spa));
7454 uint64_t space = metaslab_class_get_space(
7455 spa_embedded_log_class(spa));
7456
7457 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7458 "Embedded log class", (u_longlong_t)alloc,
7459 100.0 * alloc / space);
7460 }
7461
7462 if (spa_special_embedded_log_class(spa)->mc_allocator[0].mca_rotor
7463 != NULL) {
7464 uint64_t alloc = metaslab_class_get_alloc(
7465 spa_special_embedded_log_class(spa));
7466 uint64_t space = metaslab_class_get_space(
7467 spa_special_embedded_log_class(spa));
7468
7469 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7470 "Special embedded log", (u_longlong_t)alloc,
7471 100.0 * alloc / space);
7472 }
7473
7474 for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
7475 if (zcb->zcb_embedded_blocks[i] == 0)
7476 continue;
7477 (void) printf("\n");
7478 (void) printf("\tadditional, non-pointer bps of type %u: "
7479 "%10llu\n",
7480 i, (u_longlong_t)zcb->zcb_embedded_blocks[i]);
7481
7482 if (dump_opt['b'] >= 3) {
7483 (void) printf("\t number of (compressed) bytes: "
7484 "number of bps\n");
7485 dump_histogram(zcb->zcb_embedded_histogram[i],
7486 sizeof (zcb->zcb_embedded_histogram[i]) /
7487 sizeof (zcb->zcb_embedded_histogram[i][0]), 0);
7488 }
7489 }
7490
7491 if (tzb->zb_ditto_samevdev != 0) {
7492 (void) printf("\tDittoed blocks on same vdev: %llu\n",
7493 (longlong_t)tzb->zb_ditto_samevdev);
7494 }
7495 if (tzb->zb_ditto_same_ms != 0) {
7496 (void) printf("\tDittoed blocks in same metaslab: %llu\n",
7497 (longlong_t)tzb->zb_ditto_same_ms);
7498 }
7499
7500 for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) {
7501 vdev_t *vd = spa->spa_root_vdev->vdev_child[v];
7502 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7503
7504 if (vim == NULL) {
7505 continue;
7506 }
7507
7508 char mem[32];
7509 zdb_nicenum(vdev_indirect_mapping_num_entries(vim),
7510 mem, vdev_indirect_mapping_size(vim));
7511
7512 (void) printf("\tindirect vdev id %llu has %llu segments "
7513 "(%s in memory)\n",
7514 (longlong_t)vd->vdev_id,
7515 (longlong_t)vdev_indirect_mapping_num_entries(vim), mem);
7516 }
7517
7518 if (dump_opt['b'] >= 2) {
7519 int l, t, level;
7520 char csize[32], lsize[32], psize[32], asize[32];
7521 char avg[32], gang[32];
7522 (void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
7523 "\t avg\t comp\t%%Total\tType\n");
7524
7525 zfs_blkstat_t *mdstats = umem_zalloc(sizeof (zfs_blkstat_t),
7526 UMEM_NOFAIL);
7527
7528 for (t = 0; t <= ZDB_OT_TOTAL; t++) {
7529 const char *typename;
7530
7531 /* make sure nicenum has enough space */
7532 _Static_assert(sizeof (csize) >= NN_NUMBUF_SZ,
7533 "csize truncated");
7534 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ,
7535 "lsize truncated");
7536 _Static_assert(sizeof (psize) >= NN_NUMBUF_SZ,
7537 "psize truncated");
7538 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ,
7539 "asize truncated");
7540 _Static_assert(sizeof (avg) >= NN_NUMBUF_SZ,
7541 "avg truncated");
7542 _Static_assert(sizeof (gang) >= NN_NUMBUF_SZ,
7543 "gang truncated");
7544
7545 if (t < DMU_OT_NUMTYPES)
7546 typename = dmu_ot[t].ot_name;
7547 else
7548 typename = zdb_ot_extname[t - DMU_OT_NUMTYPES];
7549
7550 if (zcb->zcb_type[ZB_TOTAL][t].zb_asize == 0) {
7551 (void) printf("%6s\t%5s\t%5s\t%5s"
7552 "\t%5s\t%5s\t%6s\t%s\n",
7553 "-",
7554 "-",
7555 "-",
7556 "-",
7557 "-",
7558 "-",
7559 "-",
7560 typename);
7561 continue;
7562 }
7563
7564 for (l = ZB_TOTAL - 1; l >= -1; l--) {
7565 level = (l == -1 ? ZB_TOTAL : l);
7566 zb = &zcb->zcb_type[level][t];
7567
7568 if (zb->zb_asize == 0)
7569 continue;
7570
7571 if (level != ZB_TOTAL && t < DMU_OT_NUMTYPES &&
7572 (level > 0 || DMU_OT_IS_METADATA(t))) {
7573 mdstats->zb_count += zb->zb_count;
7574 mdstats->zb_lsize += zb->zb_lsize;
7575 mdstats->zb_psize += zb->zb_psize;
7576 mdstats->zb_asize += zb->zb_asize;
7577 mdstats->zb_gangs += zb->zb_gangs;
7578 }
7579
7580 if (dump_opt['b'] < 3 && level != ZB_TOTAL)
7581 continue;
7582
7583 if (level == 0 && zb->zb_asize ==
7584 zcb->zcb_type[ZB_TOTAL][t].zb_asize)
7585 continue;
7586
7587 zdb_nicenum(zb->zb_count, csize,
7588 sizeof (csize));
7589 zdb_nicenum(zb->zb_lsize, lsize,
7590 sizeof (lsize));
7591 zdb_nicenum(zb->zb_psize, psize,
7592 sizeof (psize));
7593 zdb_nicenum(zb->zb_asize, asize,
7594 sizeof (asize));
7595 zdb_nicenum(zb->zb_asize / zb->zb_count, avg,
7596 sizeof (avg));
7597 zdb_nicenum(zb->zb_gangs, gang, sizeof (gang));
7598
7599 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7600 "\t%5.2f\t%6.2f\t",
7601 csize, lsize, psize, asize, avg,
7602 (double)zb->zb_lsize / zb->zb_psize,
7603 100.0 * zb->zb_asize / tzb->zb_asize);
7604
7605 if (level == ZB_TOTAL)
7606 (void) printf("%s\n", typename);
7607 else
7608 (void) printf(" L%d %s\n",
7609 level, typename);
7610
7611 if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) {
7612 (void) printf("\t number of ganged "
7613 "blocks: %s\n", gang);
7614 }
7615
7616 if (dump_opt['b'] >= 4) {
7617 (void) printf("psize "
7618 "(in 512-byte sectors): "
7619 "number of blocks\n");
7620 dump_histogram(zb->zb_psize_histogram,
7621 PSIZE_HISTO_SIZE, 0);
7622 }
7623 }
7624 }
7625 zdb_nicenum(mdstats->zb_count, csize,
7626 sizeof (csize));
7627 zdb_nicenum(mdstats->zb_lsize, lsize,
7628 sizeof (lsize));
7629 zdb_nicenum(mdstats->zb_psize, psize,
7630 sizeof (psize));
7631 zdb_nicenum(mdstats->zb_asize, asize,
7632 sizeof (asize));
7633 zdb_nicenum(mdstats->zb_asize / mdstats->zb_count, avg,
7634 sizeof (avg));
7635 zdb_nicenum(mdstats->zb_gangs, gang, sizeof (gang));
7636
7637 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7638 "\t%5.2f\t%6.2f\t",
7639 csize, lsize, psize, asize, avg,
7640 (double)mdstats->zb_lsize / mdstats->zb_psize,
7641 100.0 * mdstats->zb_asize / tzb->zb_asize);
7642 (void) printf("%s\n", "Metadata Total");
7643
7644 /* Output a table summarizing block sizes in the pool */
7645 if (dump_opt['b'] >= 2) {
7646 dump_size_histograms(zcb);
7647 }
7648
7649 umem_free(mdstats, sizeof (zfs_blkstat_t));
7650 }
7651
7652 (void) printf("\n");
7653
7654 if (leaks) {
7655 umem_free(zcb, sizeof (zdb_cb_t));
7656 return (2);
7657 }
7658
7659 if (zcb->zcb_haderrors) {
7660 umem_free(zcb, sizeof (zdb_cb_t));
7661 return (3);
7662 }
7663
7664 umem_free(zcb, sizeof (zdb_cb_t));
7665 return (0);
7666 }
7667
7668 typedef struct zdb_ddt_entry {
7669 /* key must be first for ddt_key_compare */
7670 ddt_key_t zdde_key;
7671 uint64_t zdde_ref_blocks;
7672 uint64_t zdde_ref_lsize;
7673 uint64_t zdde_ref_psize;
7674 uint64_t zdde_ref_dsize;
7675 avl_node_t zdde_node;
7676 } zdb_ddt_entry_t;
7677
7678 static int
zdb_ddt_add_cb(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)7679 zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
7680 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
7681 {
7682 (void) zilog, (void) dnp;
7683 avl_tree_t *t = arg;
7684 avl_index_t where;
7685 zdb_ddt_entry_t *zdde, zdde_search;
7686
7687 if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
7688 BP_IS_EMBEDDED(bp))
7689 return (0);
7690
7691 if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
7692 (void) printf("traversing objset %llu, %llu objects, "
7693 "%lu blocks so far\n",
7694 (u_longlong_t)zb->zb_objset,
7695 (u_longlong_t)BP_GET_FILL(bp),
7696 avl_numnodes(t));
7697 }
7698
7699 if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF ||
7700 BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
7701 return (0);
7702
7703 ddt_key_fill(&zdde_search.zdde_key, bp);
7704
7705 zdde = avl_find(t, &zdde_search, &where);
7706
7707 if (zdde == NULL) {
7708 zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL);
7709 zdde->zdde_key = zdde_search.zdde_key;
7710 avl_insert(t, zdde, where);
7711 }
7712
7713 zdde->zdde_ref_blocks += 1;
7714 zdde->zdde_ref_lsize += BP_GET_LSIZE(bp);
7715 zdde->zdde_ref_psize += BP_GET_PSIZE(bp);
7716 zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp);
7717
7718 return (0);
7719 }
7720
7721 static void
dump_simulated_ddt(spa_t * spa)7722 dump_simulated_ddt(spa_t *spa)
7723 {
7724 avl_tree_t t;
7725 void *cookie = NULL;
7726 zdb_ddt_entry_t *zdde;
7727 ddt_histogram_t ddh_total = {{{0}}};
7728 ddt_stat_t dds_total = {0};
7729
7730 avl_create(&t, ddt_key_compare,
7731 sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
7732
7733 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7734
7735 (void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7736 TRAVERSE_NO_DECRYPT, zdb_ddt_add_cb, &t);
7737
7738 spa_config_exit(spa, SCL_CONFIG, FTAG);
7739
7740 while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) {
7741 uint64_t refcnt = zdde->zdde_ref_blocks;
7742 ASSERT(refcnt != 0);
7743
7744 ddt_stat_t *dds = &ddh_total.ddh_stat[highbit64(refcnt) - 1];
7745
7746 dds->dds_blocks += zdde->zdde_ref_blocks / refcnt;
7747 dds->dds_lsize += zdde->zdde_ref_lsize / refcnt;
7748 dds->dds_psize += zdde->zdde_ref_psize / refcnt;
7749 dds->dds_dsize += zdde->zdde_ref_dsize / refcnt;
7750
7751 dds->dds_ref_blocks += zdde->zdde_ref_blocks;
7752 dds->dds_ref_lsize += zdde->zdde_ref_lsize;
7753 dds->dds_ref_psize += zdde->zdde_ref_psize;
7754 dds->dds_ref_dsize += zdde->zdde_ref_dsize;
7755
7756 umem_free(zdde, sizeof (*zdde));
7757 }
7758
7759 avl_destroy(&t);
7760
7761 ddt_histogram_total(&dds_total, &ddh_total);
7762
7763 (void) printf("Simulated DDT histogram:\n");
7764
7765 zpool_dump_ddt(&dds_total, &ddh_total);
7766
7767 dump_dedup_ratio(&dds_total);
7768 }
7769
7770 static int
verify_device_removal_feature_counts(spa_t * spa)7771 verify_device_removal_feature_counts(spa_t *spa)
7772 {
7773 uint64_t dr_feature_refcount = 0;
7774 uint64_t oc_feature_refcount = 0;
7775 uint64_t indirect_vdev_count = 0;
7776 uint64_t precise_vdev_count = 0;
7777 uint64_t obsolete_counts_object_count = 0;
7778 uint64_t obsolete_sm_count = 0;
7779 uint64_t obsolete_counts_count = 0;
7780 uint64_t scip_count = 0;
7781 uint64_t obsolete_bpobj_count = 0;
7782 int ret = 0;
7783
7784 spa_condensing_indirect_phys_t *scip =
7785 &spa->spa_condensing_indirect_phys;
7786 if (scip->scip_next_mapping_object != 0) {
7787 vdev_t *vd = spa->spa_root_vdev->vdev_child[scip->scip_vdev];
7788 ASSERT(scip->scip_prev_obsolete_sm_object != 0);
7789 ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
7790
7791 (void) printf("Condensing indirect vdev %llu: new mapping "
7792 "object %llu, prev obsolete sm %llu\n",
7793 (u_longlong_t)scip->scip_vdev,
7794 (u_longlong_t)scip->scip_next_mapping_object,
7795 (u_longlong_t)scip->scip_prev_obsolete_sm_object);
7796 if (scip->scip_prev_obsolete_sm_object != 0) {
7797 space_map_t *prev_obsolete_sm = NULL;
7798 VERIFY0(space_map_open(&prev_obsolete_sm,
7799 spa->spa_meta_objset,
7800 scip->scip_prev_obsolete_sm_object,
7801 0, vd->vdev_asize, 0));
7802 dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm);
7803 (void) printf("\n");
7804 space_map_close(prev_obsolete_sm);
7805 }
7806
7807 scip_count += 2;
7808 }
7809
7810 for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
7811 vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
7812 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
7813
7814 if (vic->vic_mapping_object != 0) {
7815 ASSERT(vd->vdev_ops == &vdev_indirect_ops ||
7816 vd->vdev_removing);
7817 indirect_vdev_count++;
7818
7819 if (vd->vdev_indirect_mapping->vim_havecounts) {
7820 obsolete_counts_count++;
7821 }
7822 }
7823
7824 boolean_t are_precise;
7825 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7826 if (are_precise) {
7827 ASSERT(vic->vic_mapping_object != 0);
7828 precise_vdev_count++;
7829 }
7830
7831 uint64_t obsolete_sm_object;
7832 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
7833 if (obsolete_sm_object != 0) {
7834 ASSERT(vic->vic_mapping_object != 0);
7835 obsolete_sm_count++;
7836 }
7837 }
7838
7839 (void) feature_get_refcount(spa,
7840 &spa_feature_table[SPA_FEATURE_DEVICE_REMOVAL],
7841 &dr_feature_refcount);
7842 (void) feature_get_refcount(spa,
7843 &spa_feature_table[SPA_FEATURE_OBSOLETE_COUNTS],
7844 &oc_feature_refcount);
7845
7846 if (dr_feature_refcount != indirect_vdev_count) {
7847 ret = 1;
7848 (void) printf("Number of indirect vdevs (%llu) " \
7849 "does not match feature count (%llu)\n",
7850 (u_longlong_t)indirect_vdev_count,
7851 (u_longlong_t)dr_feature_refcount);
7852 } else {
7853 (void) printf("Verified device_removal feature refcount " \
7854 "of %llu is correct\n",
7855 (u_longlong_t)dr_feature_refcount);
7856 }
7857
7858 if (zap_contains(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT,
7859 DMU_POOL_OBSOLETE_BPOBJ) == 0) {
7860 obsolete_bpobj_count++;
7861 }
7862
7863
7864 obsolete_counts_object_count = precise_vdev_count;
7865 obsolete_counts_object_count += obsolete_sm_count;
7866 obsolete_counts_object_count += obsolete_counts_count;
7867 obsolete_counts_object_count += scip_count;
7868 obsolete_counts_object_count += obsolete_bpobj_count;
7869 obsolete_counts_object_count += remap_deadlist_count;
7870
7871 if (oc_feature_refcount != obsolete_counts_object_count) {
7872 ret = 1;
7873 (void) printf("Number of obsolete counts objects (%llu) " \
7874 "does not match feature count (%llu)\n",
7875 (u_longlong_t)obsolete_counts_object_count,
7876 (u_longlong_t)oc_feature_refcount);
7877 (void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
7878 "ob:%llu rd:%llu\n",
7879 (u_longlong_t)precise_vdev_count,
7880 (u_longlong_t)obsolete_sm_count,
7881 (u_longlong_t)obsolete_counts_count,
7882 (u_longlong_t)scip_count,
7883 (u_longlong_t)obsolete_bpobj_count,
7884 (u_longlong_t)remap_deadlist_count);
7885 } else {
7886 (void) printf("Verified indirect_refcount feature refcount " \
7887 "of %llu is correct\n",
7888 (u_longlong_t)oc_feature_refcount);
7889 }
7890 return (ret);
7891 }
7892
7893 static void
zdb_set_skip_mmp(char * target)7894 zdb_set_skip_mmp(char *target)
7895 {
7896 spa_t *spa;
7897
7898 /*
7899 * Disable the activity check to allow examination of
7900 * active pools.
7901 */
7902 mutex_enter(&spa_namespace_lock);
7903 if ((spa = spa_lookup(target)) != NULL) {
7904 spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
7905 }
7906 mutex_exit(&spa_namespace_lock);
7907 }
7908
7909 #define BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7910 /*
7911 * Import the checkpointed state of the pool specified by the target
7912 * parameter as readonly. The function also accepts a pool config
7913 * as an optional parameter, else it attempts to infer the config by
7914 * the name of the target pool.
7915 *
7916 * Note that the checkpointed state's pool name will be the name of
7917 * the original pool with the above suffix appended to it. In addition,
7918 * if the target is not a pool name (e.g. a path to a dataset) then
7919 * the new_path parameter is populated with the updated path to
7920 * reflect the fact that we are looking into the checkpointed state.
7921 *
7922 * The function returns a newly-allocated copy of the name of the
7923 * pool containing the checkpointed state. When this copy is no
7924 * longer needed it should be freed with free(3C). Same thing
7925 * applies to the new_path parameter if allocated.
7926 */
7927 static char *
import_checkpointed_state(char * target,nvlist_t * cfg,boolean_t target_is_spa,char ** new_path)7928 import_checkpointed_state(char *target, nvlist_t *cfg, boolean_t target_is_spa,
7929 char **new_path)
7930 {
7931 int error = 0;
7932 char *poolname, *bogus_name = NULL;
7933 boolean_t freecfg = B_FALSE;
7934
7935 /* If the target is not a pool, the extract the pool name */
7936 char *path_start = strchr(target, '/');
7937 if (target_is_spa || path_start == NULL) {
7938 poolname = target;
7939 } else {
7940 size_t poolname_len = path_start - target;
7941 poolname = strndup(target, poolname_len);
7942 }
7943
7944 if (cfg == NULL) {
7945 zdb_set_skip_mmp(poolname);
7946 error = spa_get_stats(poolname, &cfg, NULL, 0);
7947 if (error != 0) {
7948 fatal("Tried to read config of pool \"%s\" but "
7949 "spa_get_stats() failed with error %d\n",
7950 poolname, error);
7951 }
7952 freecfg = B_TRUE;
7953 }
7954
7955 if (asprintf(&bogus_name, "%s%s", poolname, BOGUS_SUFFIX) == -1) {
7956 if (target != poolname)
7957 free(poolname);
7958 return (NULL);
7959 }
7960 fnvlist_add_string(cfg, ZPOOL_CONFIG_POOL_NAME, bogus_name);
7961
7962 error = spa_import(bogus_name, cfg, NULL,
7963 ZFS_IMPORT_MISSING_LOG | ZFS_IMPORT_CHECKPOINT |
7964 ZFS_IMPORT_SKIP_MMP);
7965 if (freecfg)
7966 nvlist_free(cfg);
7967 if (error != 0) {
7968 fatal("Tried to import pool \"%s\" but spa_import() failed "
7969 "with error %d\n", bogus_name, error);
7970 }
7971
7972 if (new_path != NULL && !target_is_spa) {
7973 if (asprintf(new_path, "%s%s", bogus_name,
7974 path_start != NULL ? path_start : "") == -1) {
7975 free(bogus_name);
7976 if (!target_is_spa && path_start != NULL)
7977 free(poolname);
7978 return (NULL);
7979 }
7980 }
7981
7982 if (target != poolname)
7983 free(poolname);
7984
7985 return (bogus_name);
7986 }
7987
7988 typedef struct verify_checkpoint_sm_entry_cb_arg {
7989 vdev_t *vcsec_vd;
7990
7991 /* the following fields are only used for printing progress */
7992 uint64_t vcsec_entryid;
7993 uint64_t vcsec_num_entries;
7994 } verify_checkpoint_sm_entry_cb_arg_t;
7995
7996 #define ENTRIES_PER_PROGRESS_UPDATE 10000
7997
7998 static int
verify_checkpoint_sm_entry_cb(space_map_entry_t * sme,void * arg)7999 verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, void *arg)
8000 {
8001 verify_checkpoint_sm_entry_cb_arg_t *vcsec = arg;
8002 vdev_t *vd = vcsec->vcsec_vd;
8003 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
8004 uint64_t end = sme->sme_offset + sme->sme_run;
8005
8006 ASSERT(sme->sme_type == SM_FREE);
8007
8008 if ((vcsec->vcsec_entryid % ENTRIES_PER_PROGRESS_UPDATE) == 0) {
8009 (void) fprintf(stderr,
8010 "\rverifying vdev %llu, space map entry %llu of %llu ...",
8011 (longlong_t)vd->vdev_id,
8012 (longlong_t)vcsec->vcsec_entryid,
8013 (longlong_t)vcsec->vcsec_num_entries);
8014 }
8015 vcsec->vcsec_entryid++;
8016
8017 /*
8018 * See comment in checkpoint_sm_exclude_entry_cb()
8019 */
8020 VERIFY3U(sme->sme_offset, >=, ms->ms_start);
8021 VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
8022
8023 /*
8024 * The entries in the vdev_checkpoint_sm should be marked as
8025 * allocated in the checkpointed state of the pool, therefore
8026 * their respective ms_allocateable trees should not contain them.
8027 */
8028 mutex_enter(&ms->ms_lock);
8029 zfs_range_tree_verify_not_present(ms->ms_allocatable,
8030 sme->sme_offset, sme->sme_run);
8031 mutex_exit(&ms->ms_lock);
8032
8033 return (0);
8034 }
8035
8036 /*
8037 * Verify that all segments in the vdev_checkpoint_sm are allocated
8038 * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
8039 * ms_allocatable).
8040 *
8041 * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
8042 * each vdev in the current state of the pool to the metaslab space maps
8043 * (ms_sm) of the checkpointed state of the pool.
8044 *
8045 * Note that the function changes the state of the ms_allocatable
8046 * trees of the current spa_t. The entries of these ms_allocatable
8047 * trees are cleared out and then repopulated from with the free
8048 * entries of their respective ms_sm space maps.
8049 */
8050 static void
verify_checkpoint_vdev_spacemaps(spa_t * checkpoint,spa_t * current)8051 verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, spa_t *current)
8052 {
8053 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
8054 vdev_t *current_rvd = current->spa_root_vdev;
8055
8056 load_concrete_ms_allocatable_trees(checkpoint, SM_FREE);
8057
8058 for (uint64_t c = 0; c < ckpoint_rvd->vdev_children; c++) {
8059 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[c];
8060 vdev_t *current_vd = current_rvd->vdev_child[c];
8061
8062 space_map_t *checkpoint_sm = NULL;
8063 uint64_t checkpoint_sm_obj;
8064
8065 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
8066 /*
8067 * Since we don't allow device removal in a pool
8068 * that has a checkpoint, we expect that all removed
8069 * vdevs were removed from the pool before the
8070 * checkpoint.
8071 */
8072 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
8073 continue;
8074 }
8075
8076 /*
8077 * If the checkpoint space map doesn't exist, then nothing
8078 * here is checkpointed so there's nothing to verify.
8079 */
8080 if (current_vd->vdev_top_zap == 0 ||
8081 zap_contains(spa_meta_objset(current),
8082 current_vd->vdev_top_zap,
8083 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
8084 continue;
8085
8086 VERIFY0(zap_lookup(spa_meta_objset(current),
8087 current_vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
8088 sizeof (uint64_t), 1, &checkpoint_sm_obj));
8089
8090 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current),
8091 checkpoint_sm_obj, 0, current_vd->vdev_asize,
8092 current_vd->vdev_ashift));
8093
8094 verify_checkpoint_sm_entry_cb_arg_t vcsec;
8095 vcsec.vcsec_vd = ckpoint_vd;
8096 vcsec.vcsec_entryid = 0;
8097 vcsec.vcsec_num_entries =
8098 space_map_length(checkpoint_sm) / sizeof (uint64_t);
8099 VERIFY0(space_map_iterate(checkpoint_sm,
8100 space_map_length(checkpoint_sm),
8101 verify_checkpoint_sm_entry_cb, &vcsec));
8102 if (dump_opt['m'] > 3)
8103 dump_spacemap(current->spa_meta_objset, checkpoint_sm);
8104 space_map_close(checkpoint_sm);
8105 }
8106
8107 /*
8108 * If we've added vdevs since we took the checkpoint, ensure
8109 * that their checkpoint space maps are empty.
8110 */
8111 if (ckpoint_rvd->vdev_children < current_rvd->vdev_children) {
8112 for (uint64_t c = ckpoint_rvd->vdev_children;
8113 c < current_rvd->vdev_children; c++) {
8114 vdev_t *current_vd = current_rvd->vdev_child[c];
8115 VERIFY0P(current_vd->vdev_checkpoint_sm);
8116 }
8117 }
8118
8119 /* for cleaner progress output */
8120 (void) fprintf(stderr, "\n");
8121 }
8122
8123 /*
8124 * Verifies that all space that's allocated in the checkpoint is
8125 * still allocated in the current version, by checking that everything
8126 * in checkpoint's ms_allocatable (which is actually allocated, not
8127 * allocatable/free) is not present in current's ms_allocatable.
8128 *
8129 * Note that the function changes the state of the ms_allocatable
8130 * trees of both spas when called. The entries of all ms_allocatable
8131 * trees are cleared out and then repopulated from their respective
8132 * ms_sm space maps. In the checkpointed state we load the allocated
8133 * entries, and in the current state we load the free entries.
8134 */
8135 static void
verify_checkpoint_ms_spacemaps(spa_t * checkpoint,spa_t * current)8136 verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_t *current)
8137 {
8138 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
8139 vdev_t *current_rvd = current->spa_root_vdev;
8140
8141 load_concrete_ms_allocatable_trees(checkpoint, SM_ALLOC);
8142 load_concrete_ms_allocatable_trees(current, SM_FREE);
8143
8144 for (uint64_t i = 0; i < ckpoint_rvd->vdev_children; i++) {
8145 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[i];
8146 vdev_t *current_vd = current_rvd->vdev_child[i];
8147
8148 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
8149 /*
8150 * See comment in verify_checkpoint_vdev_spacemaps()
8151 */
8152 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
8153 continue;
8154 }
8155
8156 for (uint64_t m = 0; m < ckpoint_vd->vdev_ms_count; m++) {
8157 metaslab_t *ckpoint_msp = ckpoint_vd->vdev_ms[m];
8158 metaslab_t *current_msp = current_vd->vdev_ms[m];
8159
8160 (void) fprintf(stderr,
8161 "\rverifying vdev %llu of %llu, "
8162 "metaslab %llu of %llu ...",
8163 (longlong_t)current_vd->vdev_id,
8164 (longlong_t)current_rvd->vdev_children,
8165 (longlong_t)current_vd->vdev_ms[m]->ms_id,
8166 (longlong_t)current_vd->vdev_ms_count);
8167
8168 /*
8169 * We walk through the ms_allocatable trees that
8170 * are loaded with the allocated blocks from the
8171 * ms_sm spacemaps of the checkpoint. For each
8172 * one of these ranges we ensure that none of them
8173 * exists in the ms_allocatable trees of the
8174 * current state which are loaded with the ranges
8175 * that are currently free.
8176 *
8177 * This way we ensure that none of the blocks that
8178 * are part of the checkpoint were freed by mistake.
8179 */
8180 zfs_range_tree_walk(ckpoint_msp->ms_allocatable,
8181 (zfs_range_tree_func_t *)
8182 zfs_range_tree_verify_not_present,
8183 current_msp->ms_allocatable);
8184 }
8185 }
8186
8187 /* for cleaner progress output */
8188 (void) fprintf(stderr, "\n");
8189 }
8190
8191 static void
verify_checkpoint_blocks(spa_t * spa)8192 verify_checkpoint_blocks(spa_t *spa)
8193 {
8194 ASSERT(!dump_opt['L']);
8195
8196 spa_t *checkpoint_spa;
8197 char *checkpoint_pool;
8198 int error = 0;
8199
8200 /*
8201 * We import the checkpointed state of the pool (under a different
8202 * name) so we can do verification on it against the current state
8203 * of the pool.
8204 */
8205 checkpoint_pool = import_checkpointed_state(spa->spa_name, NULL, B_TRUE,
8206 NULL);
8207 ASSERT(strcmp(spa->spa_name, checkpoint_pool) != 0);
8208
8209 error = spa_open(checkpoint_pool, &checkpoint_spa, FTAG);
8210 if (error != 0) {
8211 fatal("Tried to open pool \"%s\" but spa_open() failed with "
8212 "error %d\n", checkpoint_pool, error);
8213 }
8214
8215 /*
8216 * Ensure that ranges in the checkpoint space maps of each vdev
8217 * are allocated according to the checkpointed state's metaslab
8218 * space maps.
8219 */
8220 verify_checkpoint_vdev_spacemaps(checkpoint_spa, spa);
8221
8222 /*
8223 * Ensure that allocated ranges in the checkpoint's metaslab
8224 * space maps remain allocated in the metaslab space maps of
8225 * the current state.
8226 */
8227 verify_checkpoint_ms_spacemaps(checkpoint_spa, spa);
8228
8229 /*
8230 * Once we are done, we get rid of the checkpointed state.
8231 */
8232 spa_close(checkpoint_spa, FTAG);
8233 free(checkpoint_pool);
8234 }
8235
8236 static void
dump_leftover_checkpoint_blocks(spa_t * spa)8237 dump_leftover_checkpoint_blocks(spa_t *spa)
8238 {
8239 vdev_t *rvd = spa->spa_root_vdev;
8240
8241 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
8242 vdev_t *vd = rvd->vdev_child[i];
8243
8244 space_map_t *checkpoint_sm = NULL;
8245 uint64_t checkpoint_sm_obj;
8246
8247 if (vd->vdev_top_zap == 0)
8248 continue;
8249
8250 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
8251 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
8252 continue;
8253
8254 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
8255 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
8256 sizeof (uint64_t), 1, &checkpoint_sm_obj));
8257
8258 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
8259 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
8260 dump_spacemap(spa->spa_meta_objset, checkpoint_sm);
8261 space_map_close(checkpoint_sm);
8262 }
8263 }
8264
8265 static int
verify_checkpoint(spa_t * spa)8266 verify_checkpoint(spa_t *spa)
8267 {
8268 uberblock_t checkpoint;
8269 int error;
8270
8271 if (!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT))
8272 return (0);
8273
8274 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8275 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
8276 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
8277
8278 if (error == ENOENT && !dump_opt['L']) {
8279 /*
8280 * If the feature is active but the uberblock is missing
8281 * then we must be in the middle of discarding the
8282 * checkpoint.
8283 */
8284 (void) printf("\nPartially discarded checkpoint "
8285 "state found:\n");
8286 if (dump_opt['m'] > 3)
8287 dump_leftover_checkpoint_blocks(spa);
8288 return (0);
8289 } else if (error != 0) {
8290 (void) printf("lookup error %d when looking for "
8291 "checkpointed uberblock in MOS\n", error);
8292 return (error);
8293 }
8294 dump_uberblock(&checkpoint, "\nCheckpointed uberblock found:\n", "\n");
8295
8296 if (checkpoint.ub_checkpoint_txg == 0) {
8297 (void) printf("\nub_checkpoint_txg not set in checkpointed "
8298 "uberblock\n");
8299 error = 3;
8300 }
8301
8302 if (error == 0 && !dump_opt['L'])
8303 verify_checkpoint_blocks(spa);
8304
8305 return (error);
8306 }
8307
8308 static void
mos_leaks_cb(void * arg,uint64_t start,uint64_t size)8309 mos_leaks_cb(void *arg, uint64_t start, uint64_t size)
8310 {
8311 (void) arg;
8312 for (uint64_t i = start; i < size; i++) {
8313 (void) printf("MOS object %llu referenced but not allocated\n",
8314 (u_longlong_t)i);
8315 }
8316 }
8317
8318 static void
mos_obj_refd(uint64_t obj)8319 mos_obj_refd(uint64_t obj)
8320 {
8321 if (obj != 0 && mos_refd_objs != NULL)
8322 zfs_range_tree_add(mos_refd_objs, obj, 1);
8323 }
8324
8325 /*
8326 * Call on a MOS object that may already have been referenced.
8327 */
8328 static void
mos_obj_refd_multiple(uint64_t obj)8329 mos_obj_refd_multiple(uint64_t obj)
8330 {
8331 if (obj != 0 && mos_refd_objs != NULL &&
8332 !zfs_range_tree_contains(mos_refd_objs, obj, 1))
8333 zfs_range_tree_add(mos_refd_objs, obj, 1);
8334 }
8335
8336 static void
mos_leak_vdev_top_zap(vdev_t * vd)8337 mos_leak_vdev_top_zap(vdev_t *vd)
8338 {
8339 uint64_t ms_flush_data_obj;
8340 int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
8341 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
8342 sizeof (ms_flush_data_obj), 1, &ms_flush_data_obj);
8343 if (error == ENOENT)
8344 return;
8345 ASSERT0(error);
8346
8347 mos_obj_refd(ms_flush_data_obj);
8348 }
8349
8350 static void
mos_leak_vdev(vdev_t * vd)8351 mos_leak_vdev(vdev_t *vd)
8352 {
8353 mos_obj_refd(vd->vdev_dtl_object);
8354 mos_obj_refd(vd->vdev_ms_array);
8355 mos_obj_refd(vd->vdev_indirect_config.vic_births_object);
8356 mos_obj_refd(vd->vdev_indirect_config.vic_mapping_object);
8357 mos_obj_refd(vd->vdev_leaf_zap);
8358 if (vd->vdev_checkpoint_sm != NULL)
8359 mos_obj_refd(vd->vdev_checkpoint_sm->sm_object);
8360 if (vd->vdev_indirect_mapping != NULL) {
8361 mos_obj_refd(vd->vdev_indirect_mapping->
8362 vim_phys->vimp_counts_object);
8363 }
8364 if (vd->vdev_obsolete_sm != NULL)
8365 mos_obj_refd(vd->vdev_obsolete_sm->sm_object);
8366
8367 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
8368 metaslab_t *ms = vd->vdev_ms[m];
8369 mos_obj_refd(space_map_object(ms->ms_sm));
8370 }
8371
8372 if (vd->vdev_root_zap != 0)
8373 mos_obj_refd(vd->vdev_root_zap);
8374
8375 if (vd->vdev_top_zap != 0) {
8376 mos_obj_refd(vd->vdev_top_zap);
8377 mos_leak_vdev_top_zap(vd);
8378 }
8379
8380 for (uint64_t c = 0; c < vd->vdev_children; c++) {
8381 mos_leak_vdev(vd->vdev_child[c]);
8382 }
8383 }
8384
8385 static void
mos_leak_log_spacemaps(spa_t * spa)8386 mos_leak_log_spacemaps(spa_t *spa)
8387 {
8388 uint64_t spacemap_zap;
8389 int error = zap_lookup(spa_meta_objset(spa),
8390 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LOG_SPACEMAP_ZAP,
8391 sizeof (spacemap_zap), 1, &spacemap_zap);
8392 if (error == ENOENT)
8393 return;
8394 ASSERT0(error);
8395
8396 mos_obj_refd(spacemap_zap);
8397 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
8398 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls))
8399 mos_obj_refd(sls->sls_sm_obj);
8400 }
8401
8402 static void
errorlog_count_refd(objset_t * mos,uint64_t errlog)8403 errorlog_count_refd(objset_t *mos, uint64_t errlog)
8404 {
8405 zap_cursor_t zc;
8406 zap_attribute_t *za = zap_attribute_alloc();
8407 for (zap_cursor_init(&zc, mos, errlog);
8408 zap_cursor_retrieve(&zc, za) == 0;
8409 zap_cursor_advance(&zc)) {
8410 mos_obj_refd(za->za_first_integer);
8411 }
8412 zap_cursor_fini(&zc);
8413 zap_attribute_free(za);
8414 }
8415
8416 static int
dump_mos_leaks(spa_t * spa)8417 dump_mos_leaks(spa_t *spa)
8418 {
8419 int rv = 0;
8420 objset_t *mos = spa->spa_meta_objset;
8421 dsl_pool_t *dp = spa->spa_dsl_pool;
8422
8423 /* Visit and mark all referenced objects in the MOS */
8424
8425 mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT);
8426 mos_obj_refd(spa->spa_pool_props_object);
8427 mos_obj_refd(spa->spa_config_object);
8428 mos_obj_refd(spa->spa_ddt_stat_object);
8429 mos_obj_refd(spa->spa_feat_desc_obj);
8430 mos_obj_refd(spa->spa_feat_enabled_txg_obj);
8431 mos_obj_refd(spa->spa_feat_for_read_obj);
8432 mos_obj_refd(spa->spa_feat_for_write_obj);
8433 mos_obj_refd(spa->spa_history);
8434 mos_obj_refd(spa->spa_errlog_last);
8435 mos_obj_refd(spa->spa_errlog_scrub);
8436
8437 if (spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) {
8438 errorlog_count_refd(mos, spa->spa_errlog_last);
8439 errorlog_count_refd(mos, spa->spa_errlog_scrub);
8440 }
8441
8442 mos_obj_refd(spa->spa_all_vdev_zaps);
8443 mos_obj_refd(spa->spa_dsl_pool->dp_bptree_obj);
8444 mos_obj_refd(spa->spa_dsl_pool->dp_tmp_userrefs_obj);
8445 mos_obj_refd(spa->spa_dsl_pool->dp_scan->scn_phys.scn_queue_obj);
8446 bpobj_count_refd(&spa->spa_deferred_bpobj);
8447 mos_obj_refd(dp->dp_empty_bpobj);
8448 bpobj_count_refd(&dp->dp_obsolete_bpobj);
8449 bpobj_count_refd(&dp->dp_free_bpobj);
8450 mos_obj_refd(spa->spa_l2cache.sav_object);
8451 mos_obj_refd(spa->spa_spares.sav_object);
8452
8453 if (spa->spa_syncing_log_sm != NULL)
8454 mos_obj_refd(spa->spa_syncing_log_sm->sm_object);
8455 mos_leak_log_spacemaps(spa);
8456
8457 mos_obj_refd(spa->spa_condensing_indirect_phys.
8458 scip_next_mapping_object);
8459 mos_obj_refd(spa->spa_condensing_indirect_phys.
8460 scip_prev_obsolete_sm_object);
8461 if (spa->spa_condensing_indirect_phys.scip_next_mapping_object != 0) {
8462 vdev_indirect_mapping_t *vim =
8463 vdev_indirect_mapping_open(mos,
8464 spa->spa_condensing_indirect_phys.scip_next_mapping_object);
8465 mos_obj_refd(vim->vim_phys->vimp_counts_object);
8466 vdev_indirect_mapping_close(vim);
8467 }
8468 deleted_livelists_dump_mos(spa);
8469
8470 if (dp->dp_origin_snap != NULL) {
8471 dsl_dataset_t *ds;
8472
8473 dsl_pool_config_enter(dp, FTAG);
8474 VERIFY0(dsl_dataset_hold_obj(dp,
8475 dsl_dataset_phys(dp->dp_origin_snap)->ds_next_snap_obj,
8476 FTAG, &ds));
8477 count_ds_mos_objects(ds);
8478 dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
8479 dsl_dataset_rele(ds, FTAG);
8480 dsl_pool_config_exit(dp, FTAG);
8481
8482 count_ds_mos_objects(dp->dp_origin_snap);
8483 dump_blkptr_list(&dp->dp_origin_snap->ds_deadlist, "Deadlist");
8484 }
8485 count_dir_mos_objects(dp->dp_mos_dir);
8486 if (dp->dp_free_dir != NULL)
8487 count_dir_mos_objects(dp->dp_free_dir);
8488 if (dp->dp_leak_dir != NULL)
8489 count_dir_mos_objects(dp->dp_leak_dir);
8490
8491 mos_leak_vdev(spa->spa_root_vdev);
8492
8493 for (uint64_t c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
8494 ddt_t *ddt = spa->spa_ddt[c];
8495 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
8496 continue;
8497
8498 /* DDT store objects */
8499 for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
8500 for (ddt_class_t class = 0; class < DDT_CLASSES;
8501 class++) {
8502 mos_obj_refd(ddt->ddt_object[type][class]);
8503 }
8504 }
8505
8506 /* FDT container */
8507 if (ddt->ddt_version == DDT_VERSION_FDT)
8508 mos_obj_refd(ddt->ddt_dir_object);
8509
8510 /* FDT log objects */
8511 if (ddt->ddt_flags & DDT_FLAG_LOG) {
8512 mos_obj_refd(ddt->ddt_log[0].ddl_object);
8513 mos_obj_refd(ddt->ddt_log[1].ddl_object);
8514 }
8515 }
8516
8517 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
8518 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
8519 if (brtvd->bv_initiated) {
8520 mos_obj_refd(brtvd->bv_mos_brtvdev);
8521 mos_obj_refd(brtvd->bv_mos_entries);
8522 }
8523 }
8524
8525 /*
8526 * Visit all allocated objects and make sure they are referenced.
8527 */
8528 uint64_t object = 0;
8529 while (dmu_object_next(mos, &object, B_FALSE, 0) == 0) {
8530 if (zfs_range_tree_contains(mos_refd_objs, object, 1)) {
8531 zfs_range_tree_remove(mos_refd_objs, object, 1);
8532 } else {
8533 dmu_object_info_t doi;
8534 const char *name;
8535 VERIFY0(dmu_object_info(mos, object, &doi));
8536 if (doi.doi_type & DMU_OT_NEWTYPE) {
8537 dmu_object_byteswap_t bswap =
8538 DMU_OT_BYTESWAP(doi.doi_type);
8539 name = dmu_ot_byteswap[bswap].ob_name;
8540 } else {
8541 name = dmu_ot[doi.doi_type].ot_name;
8542 }
8543
8544 (void) printf("MOS object %llu (%s) leaked\n",
8545 (u_longlong_t)object, name);
8546 rv = 2;
8547 }
8548 }
8549 (void) zfs_range_tree_walk(mos_refd_objs, mos_leaks_cb, NULL);
8550 if (!zfs_range_tree_is_empty(mos_refd_objs))
8551 rv = 2;
8552 zfs_range_tree_vacate(mos_refd_objs, NULL, NULL);
8553 zfs_range_tree_destroy(mos_refd_objs);
8554 return (rv);
8555 }
8556
8557 typedef struct log_sm_obsolete_stats_arg {
8558 uint64_t lsos_current_txg;
8559
8560 uint64_t lsos_total_entries;
8561 uint64_t lsos_valid_entries;
8562
8563 uint64_t lsos_sm_entries;
8564 uint64_t lsos_valid_sm_entries;
8565 } log_sm_obsolete_stats_arg_t;
8566
8567 static int
log_spacemap_obsolete_stats_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)8568 log_spacemap_obsolete_stats_cb(spa_t *spa, space_map_entry_t *sme,
8569 uint64_t txg, void *arg)
8570 {
8571 log_sm_obsolete_stats_arg_t *lsos = arg;
8572
8573 uint64_t offset = sme->sme_offset;
8574 uint64_t vdev_id = sme->sme_vdev;
8575
8576 if (lsos->lsos_current_txg == 0) {
8577 /* this is the first log */
8578 lsos->lsos_current_txg = txg;
8579 } else if (lsos->lsos_current_txg < txg) {
8580 /* we just changed log - print stats and reset */
8581 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8582 (u_longlong_t)lsos->lsos_valid_sm_entries,
8583 (u_longlong_t)lsos->lsos_sm_entries,
8584 (u_longlong_t)lsos->lsos_current_txg);
8585 lsos->lsos_valid_sm_entries = 0;
8586 lsos->lsos_sm_entries = 0;
8587 lsos->lsos_current_txg = txg;
8588 }
8589 ASSERT3U(lsos->lsos_current_txg, ==, txg);
8590
8591 lsos->lsos_sm_entries++;
8592 lsos->lsos_total_entries++;
8593
8594 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
8595 if (!vdev_is_concrete(vd))
8596 return (0);
8597
8598 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
8599 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
8600
8601 if (txg < metaslab_unflushed_txg(ms))
8602 return (0);
8603 lsos->lsos_valid_sm_entries++;
8604 lsos->lsos_valid_entries++;
8605 return (0);
8606 }
8607
8608 static void
dump_log_spacemap_obsolete_stats(spa_t * spa)8609 dump_log_spacemap_obsolete_stats(spa_t *spa)
8610 {
8611 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
8612 return;
8613
8614 log_sm_obsolete_stats_arg_t lsos = {0};
8615
8616 (void) printf("Log Space Map Obsolete Entry Statistics:\n");
8617
8618 iterate_through_spacemap_logs(spa,
8619 log_spacemap_obsolete_stats_cb, &lsos);
8620
8621 /* print stats for latest log */
8622 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8623 (u_longlong_t)lsos.lsos_valid_sm_entries,
8624 (u_longlong_t)lsos.lsos_sm_entries,
8625 (u_longlong_t)lsos.lsos_current_txg);
8626
8627 (void) printf("%-8llu valid entries out of %-8llu - total\n\n",
8628 (u_longlong_t)lsos.lsos_valid_entries,
8629 (u_longlong_t)lsos.lsos_total_entries);
8630 }
8631
8632 static void
dump_zpool(spa_t * spa)8633 dump_zpool(spa_t *spa)
8634 {
8635 dsl_pool_t *dp = spa_get_dsl(spa);
8636 int rc = 0;
8637
8638 if (dump_opt['y']) {
8639 livelist_metaslab_validate(spa);
8640 }
8641
8642 if (dump_opt['S']) {
8643 dump_simulated_ddt(spa);
8644 return;
8645 }
8646
8647 if (!dump_opt['e'] && dump_opt['C'] > 1) {
8648 (void) printf("\nCached configuration:\n");
8649 dump_nvlist(spa->spa_config, 8);
8650 }
8651
8652 if (dump_opt['C'])
8653 dump_config(spa);
8654
8655 if (dump_opt['u'])
8656 dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n");
8657
8658 if (dump_opt['D'])
8659 dump_all_ddts(spa);
8660
8661 if (dump_opt['T'])
8662 dump_brt(spa);
8663
8664 if (dump_opt['d'] > 2 || dump_opt['m'])
8665 dump_metaslabs(spa);
8666 if (dump_opt['M'])
8667 dump_metaslab_groups(spa, dump_opt['M'] > 1);
8668 if (dump_opt['d'] > 2 || dump_opt['m']) {
8669 dump_log_spacemaps(spa);
8670 dump_log_spacemap_obsolete_stats(spa);
8671 }
8672
8673 if (dump_opt['d'] || dump_opt['i']) {
8674 spa_feature_t f;
8675 mos_refd_objs = zfs_range_tree_create_flags(
8676 NULL, ZFS_RANGE_SEG64, NULL, 0, 0,
8677 0, "dump_zpool:mos_refd_objs");
8678 dump_objset(dp->dp_meta_objset);
8679
8680 if (dump_opt['d'] >= 3) {
8681 dsl_pool_t *dp = spa->spa_dsl_pool;
8682 dump_full_bpobj(&spa->spa_deferred_bpobj,
8683 "Deferred frees", 0);
8684 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
8685 dump_full_bpobj(&dp->dp_free_bpobj,
8686 "Pool snapshot frees", 0);
8687 }
8688 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
8689 ASSERT(spa_feature_is_enabled(spa,
8690 SPA_FEATURE_DEVICE_REMOVAL));
8691 dump_full_bpobj(&dp->dp_obsolete_bpobj,
8692 "Pool obsolete blocks", 0);
8693 }
8694
8695 if (spa_feature_is_active(spa,
8696 SPA_FEATURE_ASYNC_DESTROY)) {
8697 dump_bptree(spa->spa_meta_objset,
8698 dp->dp_bptree_obj,
8699 "Pool dataset frees");
8700 }
8701 dump_dtl(spa->spa_root_vdev, 0);
8702 }
8703
8704 for (spa_feature_t f = 0; f < SPA_FEATURES; f++)
8705 global_feature_count[f] = UINT64_MAX;
8706 global_feature_count[SPA_FEATURE_REDACTION_BOOKMARKS] = 0;
8707 global_feature_count[SPA_FEATURE_REDACTION_LIST_SPILL] = 0;
8708 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN] = 0;
8709 global_feature_count[SPA_FEATURE_LIVELIST] = 0;
8710
8711 (void) dmu_objset_find(spa_name(spa), dump_one_objset,
8712 NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
8713
8714 if (rc == 0 && !dump_opt['L'])
8715 rc = dump_mos_leaks(spa);
8716
8717 for (f = 0; f < SPA_FEATURES; f++) {
8718 uint64_t refcount;
8719
8720 uint64_t *arr;
8721 if (!(spa_feature_table[f].fi_flags &
8722 ZFEATURE_FLAG_PER_DATASET)) {
8723 if (global_feature_count[f] == UINT64_MAX)
8724 continue;
8725 if (!spa_feature_is_enabled(spa, f)) {
8726 ASSERT0(global_feature_count[f]);
8727 continue;
8728 }
8729 arr = global_feature_count;
8730 } else {
8731 if (!spa_feature_is_enabled(spa, f)) {
8732 ASSERT0(dataset_feature_count[f]);
8733 continue;
8734 }
8735 arr = dataset_feature_count;
8736 }
8737 if (feature_get_refcount(spa, &spa_feature_table[f],
8738 &refcount) == ENOTSUP)
8739 continue;
8740 if (arr[f] != refcount) {
8741 (void) printf("%s feature refcount mismatch: "
8742 "%lld consumers != %lld refcount\n",
8743 spa_feature_table[f].fi_uname,
8744 (longlong_t)arr[f], (longlong_t)refcount);
8745 rc = 2;
8746 } else {
8747 (void) printf("Verified %s feature refcount "
8748 "of %llu is correct\n",
8749 spa_feature_table[f].fi_uname,
8750 (longlong_t)refcount);
8751 }
8752 }
8753
8754 if (rc == 0)
8755 rc = verify_device_removal_feature_counts(spa);
8756 }
8757
8758 if (rc == 0 && (dump_opt['b'] || dump_opt['c']))
8759 rc = dump_block_stats(spa);
8760
8761 if (rc == 0)
8762 rc = verify_spacemap_refcounts(spa);
8763
8764 if (dump_opt['s'])
8765 show_pool_stats(spa);
8766
8767 if (dump_opt['h'])
8768 dump_history(spa);
8769
8770 if (rc == 0)
8771 rc = verify_checkpoint(spa);
8772
8773 if (rc != 0) {
8774 dump_debug_buffer();
8775 zdb_exit(rc);
8776 }
8777 }
8778
8779 #define ZDB_FLAG_CHECKSUM 0x0001
8780 #define ZDB_FLAG_DECOMPRESS 0x0002
8781 #define ZDB_FLAG_BSWAP 0x0004
8782 #define ZDB_FLAG_GBH 0x0008
8783 #define ZDB_FLAG_INDIRECT 0x0010
8784 #define ZDB_FLAG_RAW 0x0020
8785 #define ZDB_FLAG_PRINT_BLKPTR 0x0040
8786 #define ZDB_FLAG_VERBOSE 0x0080
8787
8788 static int flagbits[256];
8789 static char flagbitstr[16];
8790
8791 static void
zdb_print_blkptr(const blkptr_t * bp,int flags)8792 zdb_print_blkptr(const blkptr_t *bp, int flags)
8793 {
8794 char blkbuf[BP_SPRINTF_LEN];
8795
8796 if (flags & ZDB_FLAG_BSWAP)
8797 byteswap_uint64_array((void *)bp, sizeof (blkptr_t));
8798
8799 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
8800 (void) printf("%s\n", blkbuf);
8801 }
8802
8803 static void
zdb_dump_indirect(blkptr_t * bp,int nbps,int flags)8804 zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
8805 {
8806 int i;
8807
8808 for (i = 0; i < nbps; i++)
8809 zdb_print_blkptr(&bp[i], flags);
8810 }
8811
8812 static void
zdb_dump_gbh(void * buf,uint64_t size,int flags)8813 zdb_dump_gbh(void *buf, uint64_t size, int flags)
8814 {
8815 zdb_dump_indirect((blkptr_t *)buf, gbh_nblkptrs(size), flags);
8816 }
8817
8818 static void
zdb_dump_block_raw(void * buf,uint64_t size,int flags)8819 zdb_dump_block_raw(void *buf, uint64_t size, int flags)
8820 {
8821 if (flags & ZDB_FLAG_BSWAP)
8822 byteswap_uint64_array(buf, size);
8823 VERIFY(write(fileno(stdout), buf, size) == size);
8824 }
8825
8826 static void
zdb_dump_block(char * label,void * buf,uint64_t size,int flags)8827 zdb_dump_block(char *label, void *buf, uint64_t size, int flags)
8828 {
8829 uint64_t *d = (uint64_t *)buf;
8830 unsigned nwords = size / sizeof (uint64_t);
8831 int do_bswap = !!(flags & ZDB_FLAG_BSWAP);
8832 unsigned i, j;
8833 const char *hdr;
8834 char *c;
8835
8836
8837 if (do_bswap)
8838 hdr = " 7 6 5 4 3 2 1 0 f e d c b a 9 8";
8839 else
8840 hdr = " 0 1 2 3 4 5 6 7 8 9 a b c d e f";
8841
8842 (void) printf("\n%s\n%6s %s 0123456789abcdef\n", label, "", hdr);
8843
8844 #ifdef _ZFS_LITTLE_ENDIAN
8845 /* correct the endianness */
8846 do_bswap = !do_bswap;
8847 #endif
8848 for (i = 0; i < nwords; i += 2) {
8849 (void) printf("%06llx: %016llx %016llx ",
8850 (u_longlong_t)(i * sizeof (uint64_t)),
8851 (u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]),
8852 (u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1]));
8853
8854 c = (char *)&d[i];
8855 for (j = 0; j < 2 * sizeof (uint64_t); j++)
8856 (void) printf("%c", isprint(c[j]) ? c[j] : '.');
8857 (void) printf("\n");
8858 }
8859 }
8860
8861 /*
8862 * There are two acceptable formats:
8863 * leaf_name - For example: c1t0d0 or /tmp/ztest.0a
8864 * child[.child]* - For example: 0.1.1
8865 *
8866 * The second form can be used to specify arbitrary vdevs anywhere
8867 * in the hierarchy. For example, in a pool with a mirror of
8868 * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
8869 */
8870 static vdev_t *
zdb_vdev_lookup(vdev_t * vdev,const char * path)8871 zdb_vdev_lookup(vdev_t *vdev, const char *path)
8872 {
8873 char *s, *p, *q;
8874 unsigned i;
8875
8876 if (vdev == NULL)
8877 return (NULL);
8878
8879 /* First, assume the x.x.x.x format */
8880 i = strtoul(path, &s, 10);
8881 if (s == path || (s && *s != '.' && *s != '\0'))
8882 goto name;
8883 if (i >= vdev->vdev_children)
8884 return (NULL);
8885
8886 vdev = vdev->vdev_child[i];
8887 if (s && *s == '\0')
8888 return (vdev);
8889 return (zdb_vdev_lookup(vdev, s+1));
8890
8891 name:
8892 for (i = 0; i < vdev->vdev_children; i++) {
8893 vdev_t *vc = vdev->vdev_child[i];
8894
8895 if (vc->vdev_path == NULL) {
8896 vc = zdb_vdev_lookup(vc, path);
8897 if (vc == NULL)
8898 continue;
8899 else
8900 return (vc);
8901 }
8902
8903 p = strrchr(vc->vdev_path, '/');
8904 p = p ? p + 1 : vc->vdev_path;
8905 q = &vc->vdev_path[strlen(vc->vdev_path) - 2];
8906
8907 if (strcmp(vc->vdev_path, path) == 0)
8908 return (vc);
8909 if (strcmp(p, path) == 0)
8910 return (vc);
8911 if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0)
8912 return (vc);
8913 }
8914
8915 return (NULL);
8916 }
8917
8918 static int
name_from_objset_id(spa_t * spa,uint64_t objset_id,char * outstr)8919 name_from_objset_id(spa_t *spa, uint64_t objset_id, char *outstr)
8920 {
8921 dsl_dataset_t *ds;
8922
8923 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
8924 int error = dsl_dataset_hold_obj(spa->spa_dsl_pool, objset_id,
8925 NULL, &ds);
8926 if (error != 0) {
8927 (void) fprintf(stderr, "failed to hold objset %llu: %s\n",
8928 (u_longlong_t)objset_id, strerror(error));
8929 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8930 return (error);
8931 }
8932 dsl_dataset_name(ds, outstr);
8933 dsl_dataset_rele(ds, NULL);
8934 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8935 return (0);
8936 }
8937
8938 static boolean_t
zdb_parse_block_sizes(char * sizes,uint64_t * lsize,uint64_t * psize)8939 zdb_parse_block_sizes(char *sizes, uint64_t *lsize, uint64_t *psize)
8940 {
8941 char *s0, *s1, *tmp = NULL;
8942
8943 if (sizes == NULL)
8944 return (B_FALSE);
8945
8946 s0 = strtok_r(sizes, "/", &tmp);
8947 if (s0 == NULL)
8948 return (B_FALSE);
8949 s1 = strtok_r(NULL, "/", &tmp);
8950 *lsize = strtoull(s0, NULL, 16);
8951 *psize = s1 ? strtoull(s1, NULL, 16) : *lsize;
8952 return (*lsize >= *psize && *psize > 0);
8953 }
8954
8955 #define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg))
8956
8957 static boolean_t
try_decompress_block(abd_t * pabd,uint64_t lsize,uint64_t psize,int flags,int cfunc,void * lbuf,void * lbuf2)8958 try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize,
8959 int flags, int cfunc, void *lbuf, void *lbuf2)
8960 {
8961 if (flags & ZDB_FLAG_VERBOSE) {
8962 (void) fprintf(stderr,
8963 "Trying %05llx -> %05llx (%s)\n",
8964 (u_longlong_t)psize,
8965 (u_longlong_t)lsize,
8966 zio_compress_table[cfunc].ci_name);
8967 }
8968
8969 /*
8970 * We set lbuf to all zeros and lbuf2 to all
8971 * ones, then decompress to both buffers and
8972 * compare their contents. This way we can
8973 * know if decompression filled exactly to
8974 * lsize or if it left some bytes unwritten.
8975 */
8976
8977 memset(lbuf, 0x00, lsize);
8978 memset(lbuf2, 0xff, lsize);
8979
8980 abd_t labd, labd2;
8981 abd_get_from_buf_struct(&labd, lbuf, lsize);
8982 abd_get_from_buf_struct(&labd2, lbuf2, lsize);
8983
8984 boolean_t ret = B_FALSE;
8985 if (zio_decompress_data(cfunc, pabd,
8986 &labd, psize, lsize, NULL) == 0 &&
8987 zio_decompress_data(cfunc, pabd,
8988 &labd2, psize, lsize, NULL) == 0 &&
8989 memcmp(lbuf, lbuf2, lsize) == 0)
8990 ret = B_TRUE;
8991
8992 abd_free(&labd2);
8993 abd_free(&labd);
8994
8995 return (ret);
8996 }
8997
8998 static uint64_t
zdb_decompress_block(abd_t * pabd,void * buf,void * lbuf,uint64_t lsize,uint64_t psize,int flags)8999 zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
9000 uint64_t psize, int flags)
9001 {
9002 (void) buf;
9003 uint64_t orig_lsize = lsize;
9004 boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL));
9005 /*
9006 * We don't know how the data was compressed, so just try
9007 * every decompress function at every inflated blocksize.
9008 */
9009 void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
9010 int cfuncs[ZIO_COMPRESS_FUNCTIONS] = { 0 };
9011 int *cfuncp = cfuncs;
9012 uint64_t maxlsize = SPA_MAXBLOCKSIZE;
9013 uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) |
9014 ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) |
9015 ZIO_COMPRESS_MASK(ZLE);
9016 *cfuncp++ = ZIO_COMPRESS_LZ4;
9017 *cfuncp++ = ZIO_COMPRESS_LZJB;
9018 mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB);
9019 /*
9020 * Every gzip level has the same decompressor, no need to
9021 * run it 9 times per bruteforce attempt.
9022 */
9023 mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3);
9024 mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5);
9025 mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7);
9026 mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9);
9027 for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++)
9028 if (((1ULL << c) & mask) == 0)
9029 *cfuncp++ = c;
9030
9031 /*
9032 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
9033 * could take a while and we should let the user know
9034 * we are not stuck. On the other hand, printing progress
9035 * info gets old after a while. User can specify 'v' flag
9036 * to see the progression.
9037 */
9038 if (lsize == psize)
9039 lsize += SPA_MINBLOCKSIZE;
9040 else
9041 maxlsize = lsize;
9042
9043 for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) {
9044 for (cfuncp = cfuncs; *cfuncp; cfuncp++) {
9045 if (try_decompress_block(pabd, lsize, psize, flags,
9046 *cfuncp, lbuf, lbuf2)) {
9047 tryzle = B_FALSE;
9048 break;
9049 }
9050 }
9051 if (*cfuncp != 0)
9052 break;
9053 }
9054 if (tryzle) {
9055 for (lsize = orig_lsize; lsize <= maxlsize;
9056 lsize += SPA_MINBLOCKSIZE) {
9057 if (try_decompress_block(pabd, lsize, psize, flags,
9058 ZIO_COMPRESS_ZLE, lbuf, lbuf2)) {
9059 *cfuncp = ZIO_COMPRESS_ZLE;
9060 break;
9061 }
9062 }
9063 }
9064 umem_free(lbuf2, SPA_MAXBLOCKSIZE);
9065
9066 if (*cfuncp == ZIO_COMPRESS_ZLE) {
9067 printf("\nZLE decompression was selected. If you "
9068 "suspect the results are wrong,\ntry avoiding ZLE "
9069 "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
9070 }
9071
9072 return (lsize > maxlsize ? -1 : lsize);
9073 }
9074
9075 /*
9076 * Read a block from a pool and print it out. The syntax of the
9077 * block descriptor is:
9078 *
9079 * pool:vdev_specifier:offset:[lsize/]psize[:flags]
9080 *
9081 * pool - The name of the pool you wish to read from
9082 * vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
9083 * offset - offset, in hex, in bytes
9084 * size - Amount of data to read, in hex, in bytes
9085 * flags - A string of characters specifying options
9086 * b: Decode a blkptr at given offset within block
9087 * c: Calculate and display checksums
9088 * d: Decompress data before dumping
9089 * e: Byteswap data before dumping
9090 * g: Display data as a gang block header
9091 * i: Display as an indirect block
9092 * r: Dump raw data to stdout
9093 * v: Verbose
9094 *
9095 */
9096 static void
zdb_read_block(char * thing,spa_t * spa)9097 zdb_read_block(char *thing, spa_t *spa)
9098 {
9099 blkptr_t blk, *bp = &blk;
9100 dva_t *dva = bp->blk_dva;
9101 int flags = 0;
9102 uint64_t offset = 0, psize = 0, lsize = 0, blkptr_offset = 0;
9103 zio_t *zio;
9104 vdev_t *vd;
9105 abd_t *pabd;
9106 void *lbuf, *buf;
9107 char *s, *p, *dup, *flagstr, *sizes, *tmp = NULL;
9108 const char *vdev, *errmsg = NULL;
9109 int i, len, error;
9110 boolean_t borrowed = B_FALSE, found = B_FALSE;
9111
9112 dup = strdup(thing);
9113 s = strtok_r(dup, ":", &tmp);
9114 vdev = s ?: "";
9115 s = strtok_r(NULL, ":", &tmp);
9116 offset = strtoull(s ? s : "", NULL, 16);
9117 sizes = strtok_r(NULL, ":", &tmp);
9118 s = strtok_r(NULL, ":", &tmp);
9119 flagstr = strdup(s ?: "");
9120
9121 if (!zdb_parse_block_sizes(sizes, &lsize, &psize))
9122 errmsg = "invalid size(s)";
9123 if (!IS_P2ALIGNED(psize, DEV_BSIZE) || !IS_P2ALIGNED(lsize, DEV_BSIZE))
9124 errmsg = "size must be a multiple of sector size";
9125 if (!IS_P2ALIGNED(offset, DEV_BSIZE))
9126 errmsg = "offset must be a multiple of sector size";
9127 if (errmsg) {
9128 (void) printf("Invalid block specifier: %s - %s\n",
9129 thing, errmsg);
9130 goto done;
9131 }
9132
9133 tmp = NULL;
9134 for (s = strtok_r(flagstr, ":", &tmp);
9135 s != NULL;
9136 s = strtok_r(NULL, ":", &tmp)) {
9137 len = strlen(flagstr);
9138 for (i = 0; i < len; i++) {
9139 int bit = flagbits[(uchar_t)flagstr[i]];
9140
9141 if (bit == 0) {
9142 (void) printf("***Ignoring flag: %c\n",
9143 (uchar_t)flagstr[i]);
9144 continue;
9145 }
9146 found = B_TRUE;
9147 flags |= bit;
9148
9149 p = &flagstr[i + 1];
9150 if (*p != ':' && *p != '\0') {
9151 int j = 0, nextbit = flagbits[(uchar_t)*p];
9152 char *end, offstr[8] = { 0 };
9153 if ((bit == ZDB_FLAG_PRINT_BLKPTR) &&
9154 (nextbit == 0)) {
9155 /* look ahead to isolate the offset */
9156 while (nextbit == 0 &&
9157 strchr(flagbitstr, *p) == NULL) {
9158 offstr[j] = *p;
9159 j++;
9160 if (i + j > strlen(flagstr))
9161 break;
9162 p++;
9163 nextbit = flagbits[(uchar_t)*p];
9164 }
9165 blkptr_offset = strtoull(offstr, &end,
9166 16);
9167 i += j;
9168 } else if (nextbit == 0) {
9169 (void) printf("***Ignoring flag arg:"
9170 " '%c'\n", (uchar_t)*p);
9171 }
9172 }
9173 }
9174 }
9175 if (blkptr_offset % sizeof (blkptr_t)) {
9176 printf("Block pointer offset 0x%llx "
9177 "must be divisible by 0x%x\n",
9178 (longlong_t)blkptr_offset, (int)sizeof (blkptr_t));
9179 goto done;
9180 }
9181 if (found == B_FALSE && strlen(flagstr) > 0) {
9182 printf("Invalid flag arg: '%s'\n", flagstr);
9183 goto done;
9184 }
9185
9186 vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev);
9187 if (vd == NULL) {
9188 (void) printf("***Invalid vdev: %s\n", vdev);
9189 goto done;
9190 } else {
9191 if (vd->vdev_path)
9192 (void) fprintf(stderr, "Found vdev: %s\n",
9193 vd->vdev_path);
9194 else
9195 (void) fprintf(stderr, "Found vdev type: %s\n",
9196 vd->vdev_ops->vdev_op_type);
9197 }
9198
9199 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
9200 lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
9201
9202 BP_ZERO(bp);
9203
9204 DVA_SET_VDEV(&dva[0], vd->vdev_id);
9205 DVA_SET_OFFSET(&dva[0], offset);
9206 DVA_SET_GANG(&dva[0], 0);
9207 DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
9208
9209 BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
9210
9211 BP_SET_LSIZE(bp, lsize);
9212 BP_SET_PSIZE(bp, psize);
9213 BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF);
9214 BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF);
9215 BP_SET_TYPE(bp, DMU_OT_NONE);
9216 BP_SET_LEVEL(bp, 0);
9217 BP_SET_DEDUP(bp, 0);
9218 BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
9219
9220 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9221 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9222
9223 if (vd == vd->vdev_top) {
9224 /*
9225 * Treat this as a normal block read.
9226 */
9227 zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL,
9228 ZIO_PRIORITY_SYNC_READ,
9229 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL));
9230 } else {
9231 /*
9232 * Treat this as a vdev child I/O.
9233 */
9234 zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd,
9235 psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ,
9236 ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
9237 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | ZIO_FLAG_OPTIONAL,
9238 NULL, NULL));
9239 }
9240
9241 error = zio_wait(zio);
9242 spa_config_exit(spa, SCL_STATE, FTAG);
9243
9244 if (error) {
9245 (void) printf("Read of %s failed, error: %d\n", thing, error);
9246 goto out;
9247 }
9248
9249 uint64_t orig_lsize = lsize;
9250 buf = lbuf;
9251 if (flags & ZDB_FLAG_DECOMPRESS) {
9252 lsize = zdb_decompress_block(pabd, buf, lbuf,
9253 lsize, psize, flags);
9254 if (lsize == -1) {
9255 (void) printf("Decompress of %s failed\n", thing);
9256 goto out;
9257 }
9258 } else {
9259 buf = abd_borrow_buf_copy(pabd, lsize);
9260 borrowed = B_TRUE;
9261 }
9262 /*
9263 * Try to detect invalid block pointer. If invalid, try
9264 * decompressing.
9265 */
9266 if ((flags & ZDB_FLAG_PRINT_BLKPTR || flags & ZDB_FLAG_INDIRECT) &&
9267 !(flags & ZDB_FLAG_DECOMPRESS)) {
9268 const blkptr_t *b = (const blkptr_t *)(void *)
9269 ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9270 if (zfs_blkptr_verify(spa, b,
9271 BLK_CONFIG_NEEDED, BLK_VERIFY_ONLY)) {
9272 abd_return_buf_copy(pabd, buf, lsize);
9273 borrowed = B_FALSE;
9274 buf = lbuf;
9275 lsize = zdb_decompress_block(pabd, buf,
9276 lbuf, lsize, psize, flags);
9277 b = (const blkptr_t *)(void *)
9278 ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9279 if (lsize == -1 || zfs_blkptr_verify(spa, b,
9280 BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) {
9281 printf("invalid block pointer at this DVA\n");
9282 goto out;
9283 }
9284 }
9285 }
9286
9287 if (flags & ZDB_FLAG_PRINT_BLKPTR)
9288 zdb_print_blkptr((blkptr_t *)(void *)
9289 ((uintptr_t)buf + (uintptr_t)blkptr_offset), flags);
9290 else if (flags & ZDB_FLAG_RAW)
9291 zdb_dump_block_raw(buf, lsize, flags);
9292 else if (flags & ZDB_FLAG_INDIRECT)
9293 zdb_dump_indirect((blkptr_t *)buf,
9294 orig_lsize / sizeof (blkptr_t), flags);
9295 else if (flags & ZDB_FLAG_GBH)
9296 zdb_dump_gbh(buf, lsize, flags);
9297 else
9298 zdb_dump_block(thing, buf, lsize, flags);
9299
9300 /*
9301 * If :c was specified, iterate through the checksum table to
9302 * calculate and display each checksum for our specified
9303 * DVA and length.
9304 */
9305 if ((flags & ZDB_FLAG_CHECKSUM) && !(flags & ZDB_FLAG_RAW) &&
9306 !(flags & ZDB_FLAG_GBH)) {
9307 zio_t *czio;
9308 (void) printf("\n");
9309 for (enum zio_checksum ck = ZIO_CHECKSUM_LABEL;
9310 ck < ZIO_CHECKSUM_FUNCTIONS; ck++) {
9311
9312 if ((zio_checksum_table[ck].ci_flags &
9313 ZCHECKSUM_FLAG_EMBEDDED) ||
9314 ck == ZIO_CHECKSUM_NOPARITY) {
9315 continue;
9316 }
9317 BP_SET_CHECKSUM(bp, ck);
9318 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9319 czio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9320 if (vd == vd->vdev_top) {
9321 zio_nowait(zio_read(czio, spa, bp, pabd, psize,
9322 NULL, NULL,
9323 ZIO_PRIORITY_SYNC_READ,
9324 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9325 ZIO_FLAG_DONT_RETRY, NULL));
9326 } else {
9327 zio_nowait(zio_vdev_child_io(czio, bp, vd,
9328 offset, pabd, psize, ZIO_TYPE_READ,
9329 ZIO_PRIORITY_SYNC_READ,
9330 ZIO_FLAG_DONT_PROPAGATE |
9331 ZIO_FLAG_DONT_RETRY |
9332 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9333 ZIO_FLAG_SPECULATIVE |
9334 ZIO_FLAG_OPTIONAL, NULL, NULL));
9335 }
9336 error = zio_wait(czio);
9337 if (error == 0 || error == ECKSUM) {
9338 zio_t *ck_zio = zio_null(NULL, spa, NULL,
9339 NULL, NULL, 0);
9340 ck_zio->io_offset =
9341 DVA_GET_OFFSET(&bp->blk_dva[0]);
9342 ck_zio->io_bp = bp;
9343 zio_checksum_compute(ck_zio, ck, pabd, psize);
9344 printf(
9345 "%12s\t"
9346 "cksum=%016llx:%016llx:%016llx:%016llx\n",
9347 zio_checksum_table[ck].ci_name,
9348 (u_longlong_t)bp->blk_cksum.zc_word[0],
9349 (u_longlong_t)bp->blk_cksum.zc_word[1],
9350 (u_longlong_t)bp->blk_cksum.zc_word[2],
9351 (u_longlong_t)bp->blk_cksum.zc_word[3]);
9352 zio_wait(ck_zio);
9353 } else {
9354 printf("error %d reading block\n", error);
9355 }
9356 spa_config_exit(spa, SCL_STATE, FTAG);
9357 }
9358 }
9359
9360 if (borrowed)
9361 abd_return_buf_copy(pabd, buf, lsize);
9362
9363 out:
9364 abd_free(pabd);
9365 umem_free(lbuf, SPA_MAXBLOCKSIZE);
9366 done:
9367 free(flagstr);
9368 free(dup);
9369 }
9370
9371 static void
zdb_embedded_block(char * thing)9372 zdb_embedded_block(char *thing)
9373 {
9374 blkptr_t bp = {{{{0}}}};
9375 unsigned long long *words = (void *)&bp;
9376 char *buf;
9377 int err;
9378
9379 err = sscanf(thing, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
9380 "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
9381 words + 0, words + 1, words + 2, words + 3,
9382 words + 4, words + 5, words + 6, words + 7,
9383 words + 8, words + 9, words + 10, words + 11,
9384 words + 12, words + 13, words + 14, words + 15);
9385 if (err != 16) {
9386 (void) fprintf(stderr, "invalid input format\n");
9387 zdb_exit(1);
9388 }
9389 ASSERT3U(BPE_GET_LSIZE(&bp), <=, SPA_MAXBLOCKSIZE);
9390 buf = malloc(SPA_MAXBLOCKSIZE);
9391 if (buf == NULL) {
9392 (void) fprintf(stderr, "out of memory\n");
9393 zdb_exit(1);
9394 }
9395 err = decode_embedded_bp(&bp, buf, BPE_GET_LSIZE(&bp));
9396 if (err != 0) {
9397 (void) fprintf(stderr, "decode failed: %u\n", err);
9398 zdb_exit(1);
9399 }
9400 zdb_dump_block_raw(buf, BPE_GET_LSIZE(&bp), 0);
9401 free(buf);
9402 }
9403
9404 /* check for valid hex or decimal numeric string */
9405 static boolean_t
zdb_numeric(char * str)9406 zdb_numeric(char *str)
9407 {
9408 int i = 0, len;
9409
9410 len = strlen(str);
9411 if (len == 0)
9412 return (B_FALSE);
9413 if (strncmp(str, "0x", 2) == 0 || strncmp(str, "0X", 2) == 0)
9414 i = 2;
9415 for (; i < len; i++) {
9416 if (!isxdigit(str[i]))
9417 return (B_FALSE);
9418 }
9419 return (B_TRUE);
9420 }
9421
9422 static int
dummy_get_file_info(dmu_object_type_t bonustype,const void * data,zfs_file_info_t * zoi)9423 dummy_get_file_info(dmu_object_type_t bonustype, const void *data,
9424 zfs_file_info_t *zoi)
9425 {
9426 (void) data, (void) zoi;
9427
9428 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
9429 return (ENOENT);
9430
9431 (void) fprintf(stderr, "dummy_get_file_info: not implemented");
9432 abort();
9433 }
9434
9435 int
main(int argc,char ** argv)9436 main(int argc, char **argv)
9437 {
9438 int c;
9439 int dump_all = 1;
9440 int verbose = 0;
9441 int error = 0;
9442 char **searchdirs = NULL;
9443 int nsearch = 0;
9444 char *target, *target_pool, dsname[ZFS_MAX_DATASET_NAME_LEN];
9445 nvlist_t *policy = NULL;
9446 uint64_t max_txg = UINT64_MAX;
9447 int64_t objset_id = -1;
9448 uint64_t object;
9449 int flags = ZFS_IMPORT_MISSING_LOG;
9450 int rewind = ZPOOL_NEVER_REWIND;
9451 char *spa_config_path_env, *objset_str;
9452 boolean_t target_is_spa = B_TRUE, dataset_lookup = B_FALSE;
9453 nvlist_t *cfg = NULL;
9454 struct sigaction action;
9455 boolean_t force_import = B_FALSE;
9456 boolean_t config_path_console = B_FALSE;
9457 char pbuf[MAXPATHLEN];
9458
9459 dprintf_setup(&argc, argv);
9460
9461 /*
9462 * Set up signal handlers, so if we crash due to bad on-disk data we
9463 * can get more info. Unlike ztest, we don't bail out if we can't set
9464 * up signal handlers, because zdb is very useful without them.
9465 */
9466 action.sa_handler = sig_handler;
9467 sigemptyset(&action.sa_mask);
9468 action.sa_flags = 0;
9469 if (sigaction(SIGSEGV, &action, NULL) < 0) {
9470 (void) fprintf(stderr, "zdb: cannot catch SIGSEGV: %s\n",
9471 strerror(errno));
9472 }
9473 if (sigaction(SIGABRT, &action, NULL) < 0) {
9474 (void) fprintf(stderr, "zdb: cannot catch SIGABRT: %s\n",
9475 strerror(errno));
9476 }
9477
9478 /*
9479 * If there is an environment variable SPA_CONFIG_PATH it overrides
9480 * default spa_config_path setting. If -U flag is specified it will
9481 * override this environment variable settings once again.
9482 */
9483 spa_config_path_env = getenv("SPA_CONFIG_PATH");
9484 if (spa_config_path_env != NULL)
9485 spa_config_path = spa_config_path_env;
9486
9487 /*
9488 * For performance reasons, we set this tunable down. We do so before
9489 * the arg parsing section so that the user can override this value if
9490 * they choose.
9491 */
9492 zfs_btree_verify_intensity = 3;
9493
9494 struct option long_options[] = {
9495 {"ignore-assertions", no_argument, NULL, 'A'},
9496 {"block-stats", no_argument, NULL, 'b'},
9497 {"backup", no_argument, NULL, 'B'},
9498 {"checksum", no_argument, NULL, 'c'},
9499 {"config", no_argument, NULL, 'C'},
9500 {"datasets", no_argument, NULL, 'd'},
9501 {"dedup-stats", no_argument, NULL, 'D'},
9502 {"exported", no_argument, NULL, 'e'},
9503 {"embedded-block-pointer", no_argument, NULL, 'E'},
9504 {"automatic-rewind", no_argument, NULL, 'F'},
9505 {"dump-debug-msg", no_argument, NULL, 'G'},
9506 {"history", no_argument, NULL, 'h'},
9507 {"intent-logs", no_argument, NULL, 'i'},
9508 {"inflight", required_argument, NULL, 'I'},
9509 {"checkpointed-state", no_argument, NULL, 'k'},
9510 {"key", required_argument, NULL, 'K'},
9511 {"label", no_argument, NULL, 'l'},
9512 {"disable-leak-tracking", no_argument, NULL, 'L'},
9513 {"metaslabs", no_argument, NULL, 'm'},
9514 {"metaslab-groups", no_argument, NULL, 'M'},
9515 {"numeric", no_argument, NULL, 'N'},
9516 {"option", required_argument, NULL, 'o'},
9517 {"object-lookups", no_argument, NULL, 'O'},
9518 {"path", required_argument, NULL, 'p'},
9519 {"parseable", no_argument, NULL, 'P'},
9520 {"skip-label", no_argument, NULL, 'q'},
9521 {"copy-object", no_argument, NULL, 'r'},
9522 {"read-block", no_argument, NULL, 'R'},
9523 {"io-stats", no_argument, NULL, 's'},
9524 {"simulate-dedup", no_argument, NULL, 'S'},
9525 {"txg", required_argument, NULL, 't'},
9526 {"brt-stats", no_argument, NULL, 'T'},
9527 {"uberblock", no_argument, NULL, 'u'},
9528 {"cachefile", required_argument, NULL, 'U'},
9529 {"verbose", no_argument, NULL, 'v'},
9530 {"verbatim", no_argument, NULL, 'V'},
9531 {"dump-blocks", required_argument, NULL, 'x'},
9532 {"extreme-rewind", no_argument, NULL, 'X'},
9533 {"all-reconstruction", no_argument, NULL, 'Y'},
9534 {"livelist", no_argument, NULL, 'y'},
9535 {"zstd-headers", no_argument, NULL, 'Z'},
9536 {"allocated-map", no_argument, NULL,
9537 ARG_ALLOCATED},
9538 {"bin", required_argument, NULL,
9539 ARG_BLOCK_BIN_MODE},
9540 {"class", required_argument, NULL,
9541 ARG_BLOCK_CLASSES},
9542 {0, 0, 0, 0}
9543 };
9544
9545 while ((c = getopt_long(argc, argv,
9546 "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
9547 long_options, NULL)) != -1) {
9548 switch (c) {
9549 case 'b':
9550 case 'B':
9551 case 'c':
9552 case 'C':
9553 case 'd':
9554 case 'D':
9555 case 'E':
9556 case 'G':
9557 case 'h':
9558 case 'i':
9559 case 'l':
9560 case 'm':
9561 case 'M':
9562 case 'N':
9563 case 'O':
9564 case 'r':
9565 case 'R':
9566 case 's':
9567 case 'S':
9568 case 'T':
9569 case 'u':
9570 case 'y':
9571 case 'Z':
9572 case ARG_ALLOCATED:
9573 dump_opt[c]++;
9574 dump_all = 0;
9575 break;
9576 case 'A':
9577 case 'e':
9578 case 'F':
9579 case 'k':
9580 case 'L':
9581 case 'P':
9582 case 'q':
9583 case 'X':
9584 dump_opt[c]++;
9585 break;
9586 case 'Y':
9587 zfs_reconstruct_indirect_combinations_max = INT_MAX;
9588 zfs_deadman_enabled = 0;
9589 break;
9590 /* NB: Sort single match options below. */
9591 case 'I':
9592 max_inflight_bytes = strtoull(optarg, NULL, 0);
9593 if (max_inflight_bytes == 0) {
9594 (void) fprintf(stderr, "maximum number "
9595 "of inflight bytes must be greater "
9596 "than 0\n");
9597 usage();
9598 }
9599 break;
9600 case 'K':
9601 dump_opt[c]++;
9602 key_material = strdup(optarg);
9603 /* redact key material in process table */
9604 while (*optarg != '\0') { *optarg++ = '*'; }
9605 break;
9606 case 'o':
9607 dump_opt[c]++;
9608 dump_all = 0;
9609 error = handle_tunable_option(optarg, B_FALSE);
9610 if (error != 0)
9611 zdb_exit(1);
9612 break;
9613 case 'p':
9614 if (searchdirs == NULL) {
9615 searchdirs = umem_alloc(sizeof (char *),
9616 UMEM_NOFAIL);
9617 } else {
9618 char **tmp = umem_alloc((nsearch + 1) *
9619 sizeof (char *), UMEM_NOFAIL);
9620 memcpy(tmp, searchdirs, nsearch *
9621 sizeof (char *));
9622 umem_free(searchdirs,
9623 nsearch * sizeof (char *));
9624 searchdirs = tmp;
9625 }
9626 searchdirs[nsearch++] = optarg;
9627 break;
9628 case 't':
9629 max_txg = strtoull(optarg, NULL, 0);
9630 if (max_txg < TXG_INITIAL) {
9631 (void) fprintf(stderr, "incorrect txg "
9632 "specified: %s\n", optarg);
9633 usage();
9634 }
9635 break;
9636 case 'U':
9637 config_path_console = B_TRUE;
9638 spa_config_path = optarg;
9639 if (spa_config_path[0] != '/') {
9640 (void) fprintf(stderr,
9641 "cachefile must be an absolute path "
9642 "(i.e. start with a slash)\n");
9643 usage();
9644 }
9645 break;
9646 case 'v':
9647 verbose++;
9648 break;
9649 case 'V':
9650 flags = ZFS_IMPORT_VERBATIM;
9651 break;
9652 case 'x':
9653 vn_dumpdir = optarg;
9654 break;
9655 case ARG_BLOCK_BIN_MODE:
9656 if (strcmp(optarg, "lsize") == 0) {
9657 block_bin_mode = BIN_LSIZE;
9658 } else if (strcmp(optarg, "psize") == 0) {
9659 block_bin_mode = BIN_PSIZE;
9660 } else if (strcmp(optarg, "asize") == 0) {
9661 block_bin_mode = BIN_ASIZE;
9662 } else {
9663 (void) fprintf(stderr,
9664 "--bin=\"%s\" must be one of \"lsize\", "
9665 "\"psize\" or \"asize\"\n", optarg);
9666 usage();
9667 }
9668 break;
9669
9670 case ARG_BLOCK_CLASSES: {
9671 char *buf = strdup(optarg), *tok = buf, *next,
9672 *save = NULL;
9673
9674 while ((next = strtok_r(tok, ",", &save)) != NULL) {
9675 tok = NULL;
9676
9677 if (strcmp(next, "normal") == 0) {
9678 block_classes |= CLASS_NORMAL;
9679 } else if (strcmp(next, "special") == 0) {
9680 block_classes |= CLASS_SPECIAL;
9681 } else if (strcmp(next, "dedup") == 0) {
9682 block_classes |= CLASS_DEDUP;
9683 } else if (strcmp(next, "other") == 0) {
9684 block_classes |= CLASS_OTHER;
9685 } else {
9686 (void) fprintf(stderr,
9687 "--class=\"%s\" must be a "
9688 "comma-separated list of either "
9689 "\"normal\", \"special\", "
9690 "\"asize\" or \"other\"; "
9691 "got \"%s\"\n",
9692 optarg, next);
9693 usage();
9694 }
9695 }
9696
9697 if (block_classes == 0) {
9698 (void) fprintf(stderr,
9699 "--class= must be a comma-separated "
9700 "list of either \"normal\", \"special\", "
9701 "\"asize\" or \"other\"; got empty\n");
9702 usage();
9703 }
9704
9705 free(buf);
9706 break;
9707 }
9708 default:
9709 usage();
9710 break;
9711 }
9712 }
9713
9714 if (!dump_opt['e'] && searchdirs != NULL) {
9715 (void) fprintf(stderr, "-p option requires use of -e\n");
9716 usage();
9717 }
9718 #if defined(_LP64)
9719 /*
9720 * ZDB does not typically re-read blocks; therefore limit the ARC
9721 * to 256 MB, which can be used entirely for metadata.
9722 */
9723 zfs_arc_min = 2ULL << SPA_MAXBLOCKSHIFT;
9724 zfs_arc_max = 256 * 1024 * 1024;
9725 #endif
9726
9727 /*
9728 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
9729 * "zdb -b" uses traversal prefetch which uses async reads.
9730 * For good performance, let several of them be active at once.
9731 */
9732 zfs_vdev_async_read_max_active = 10;
9733
9734 /*
9735 * Disable reference tracking for better performance.
9736 */
9737 reference_tracking_enable = B_FALSE;
9738
9739 /*
9740 * Do not fail spa_load when spa_load_verify fails. This is needed
9741 * to load non-idle pools.
9742 */
9743 spa_load_verify_dryrun = B_TRUE;
9744
9745 /*
9746 * ZDB should have ability to read spacemaps.
9747 */
9748 spa_mode_readable_spacemaps = B_TRUE;
9749
9750 if (dump_all)
9751 verbose = MAX(verbose, 1);
9752
9753 for (c = 0; c < 256; c++) {
9754 if (dump_all && strchr("ABeEFkKlLNOPrRSXy", c) == NULL)
9755 dump_opt[c] = 1;
9756 if (dump_opt[c])
9757 dump_opt[c] += verbose;
9758 }
9759
9760 libspl_set_assert_ok((dump_opt['A'] == 1) || (dump_opt['A'] > 2));
9761 zfs_recover = (dump_opt['A'] > 1);
9762
9763 argc -= optind;
9764 argv += optind;
9765 if (argc < 2 && dump_opt['R'])
9766 usage();
9767
9768 target = argv[0];
9769
9770 /*
9771 * Automate cachefile
9772 */
9773 if (!spa_config_path_env && !config_path_console && target &&
9774 libzfs_core_init() == 0) {
9775 char *pname = strdup(target);
9776 const char *value;
9777 nvlist_t *pnvl = NULL;
9778 nvlist_t *vnvl = NULL;
9779
9780 if (strpbrk(pname, "/@") != NULL)
9781 *strpbrk(pname, "/@") = '\0';
9782
9783 if (pname && lzc_get_props(pname, &pnvl) == 0) {
9784 if (nvlist_lookup_nvlist(pnvl, "cachefile",
9785 &vnvl) == 0) {
9786 value = fnvlist_lookup_string(vnvl,
9787 ZPROP_VALUE);
9788 } else {
9789 value = "-";
9790 }
9791 strlcpy(pbuf, value, sizeof (pbuf));
9792 if (pbuf[0] != '\0') {
9793 if (pbuf[0] == '/') {
9794 if (access(pbuf, F_OK) == 0)
9795 spa_config_path = pbuf;
9796 else
9797 force_import = B_TRUE;
9798 } else if ((strcmp(pbuf, "-") == 0 &&
9799 access(ZPOOL_CACHE, F_OK) != 0) ||
9800 strcmp(pbuf, "none") == 0) {
9801 force_import = B_TRUE;
9802 }
9803 }
9804 nvlist_free(vnvl);
9805 }
9806
9807 free(pname);
9808 nvlist_free(pnvl);
9809 libzfs_core_fini();
9810 }
9811
9812 dmu_objset_register_type(DMU_OST_ZFS, dummy_get_file_info);
9813 kernel_init(SPA_MODE_READ);
9814 kernel_init_done = B_TRUE;
9815
9816 if (dump_opt['E']) {
9817 if (argc != 1)
9818 usage();
9819 zdb_embedded_block(argv[0]);
9820 error = 0;
9821 goto fini;
9822 }
9823
9824 if (argc < 1) {
9825 if (!dump_opt['e'] && dump_opt['C']) {
9826 dump_cachefile(spa_config_path);
9827 error = 0;
9828 goto fini;
9829 }
9830 if (dump_opt['o'])
9831 /*
9832 * Avoid blasting tunable options off the top of the
9833 * screen.
9834 */
9835 zdb_exit(1);
9836 usage();
9837 }
9838
9839 if (dump_opt['l']) {
9840 error = dump_label(argv[0]);
9841 goto fini;
9842 }
9843
9844 if (dump_opt['X'] || dump_opt['F'])
9845 rewind = ZPOOL_DO_REWIND |
9846 (dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
9847
9848 /* -N implies -d */
9849 if (dump_opt['N'] && dump_opt['d'] == 0)
9850 dump_opt['d'] = dump_opt['N'];
9851
9852 if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 ||
9853 nvlist_add_uint64(policy, ZPOOL_LOAD_REQUEST_TXG, max_txg) != 0 ||
9854 nvlist_add_uint32(policy, ZPOOL_LOAD_REWIND_POLICY, rewind) != 0)
9855 fatal("internal error: %s", strerror(ENOMEM));
9856
9857 error = 0;
9858
9859 if (strpbrk(target, "/@") != NULL) {
9860 size_t targetlen;
9861
9862 target_pool = strdup(target);
9863 *strpbrk(target_pool, "/@") = '\0';
9864
9865 target_is_spa = B_FALSE;
9866 targetlen = strlen(target);
9867 if (targetlen && target[targetlen - 1] == '/')
9868 target[targetlen - 1] = '\0';
9869
9870 /*
9871 * See if an objset ID was supplied (-d <pool>/<objset ID>).
9872 * To disambiguate tank/100, consider the 100 as objsetID
9873 * if -N was given, otherwise 100 is an objsetID iff
9874 * tank/100 as a named dataset fails on lookup.
9875 */
9876 objset_str = strchr(target, '/');
9877 if (objset_str && strlen(objset_str) > 1 &&
9878 zdb_numeric(objset_str + 1)) {
9879 char *endptr;
9880 errno = 0;
9881 objset_str++;
9882 objset_id = strtoull(objset_str, &endptr, 0);
9883 /* dataset 0 is the same as opening the pool */
9884 if (errno == 0 && endptr != objset_str &&
9885 objset_id != 0) {
9886 if (dump_opt['N'])
9887 dataset_lookup = B_TRUE;
9888 }
9889 /* normal dataset name not an objset ID */
9890 if (endptr == objset_str) {
9891 objset_id = -1;
9892 }
9893 } else if (objset_str && !zdb_numeric(objset_str + 1) &&
9894 dump_opt['N']) {
9895 printf("Supply a numeric objset ID with -N\n");
9896 error = 2;
9897 goto fini;
9898 }
9899 } else {
9900 target_pool = target;
9901 }
9902
9903 if (dump_opt['e'] || force_import) {
9904 importargs_t args = { 0 };
9905
9906 /*
9907 * If path is not provided, search in /dev
9908 */
9909 if (searchdirs == NULL) {
9910 searchdirs = umem_alloc(sizeof (char *), UMEM_NOFAIL);
9911 searchdirs[nsearch++] = (char *)ZFS_DEVDIR;
9912 }
9913
9914 args.paths = nsearch;
9915 args.path = searchdirs;
9916 args.can_be_active = B_TRUE;
9917
9918 libpc_handle_t lpch = {
9919 .lpc_lib_handle = NULL,
9920 .lpc_ops = &libzpool_config_ops,
9921 .lpc_printerr = B_TRUE
9922 };
9923 error = zpool_find_config(&lpch, target_pool, &cfg, &args);
9924
9925 if (error == 0) {
9926
9927 if (nvlist_add_nvlist(cfg,
9928 ZPOOL_LOAD_POLICY, policy) != 0) {
9929 fatal("can't open '%s': %s",
9930 target, strerror(ENOMEM));
9931 }
9932
9933 if (dump_opt['C'] > 1) {
9934 (void) printf("\nConfiguration for import:\n");
9935 dump_nvlist(cfg, 8);
9936 }
9937
9938 /*
9939 * Disable the activity check to allow examination of
9940 * active pools.
9941 */
9942 error = spa_import(target_pool, cfg, NULL,
9943 flags | ZFS_IMPORT_SKIP_MMP);
9944 }
9945 }
9946
9947 if (searchdirs != NULL) {
9948 umem_free(searchdirs, nsearch * sizeof (char *));
9949 searchdirs = NULL;
9950 }
9951
9952 /*
9953 * We need to make sure to process -O option or call
9954 * dump_path after the -e option has been processed,
9955 * which imports the pool to the namespace if it's
9956 * not in the cachefile.
9957 */
9958 if (dump_opt['O']) {
9959 if (argc != 2)
9960 usage();
9961 dump_opt['v'] = verbose + 3;
9962 error = dump_path(argv[0], argv[1], NULL);
9963 goto fini;
9964 }
9965
9966 if (dump_opt['r']) {
9967 target_is_spa = B_FALSE;
9968 if (argc != 3)
9969 usage();
9970 dump_opt['v'] = verbose;
9971 error = dump_path(argv[0], argv[1], &object);
9972 if (error != 0)
9973 fatal("internal error: %s", strerror(error));
9974 }
9975
9976 /*
9977 * import_checkpointed_state makes the assumption that the
9978 * target pool that we pass it is already part of the spa
9979 * namespace. Because of that we need to make sure to call
9980 * it always after the -e option has been processed, which
9981 * imports the pool to the namespace if it's not in the
9982 * cachefile.
9983 */
9984 char *checkpoint_pool = NULL;
9985 char *checkpoint_target = NULL;
9986 if (dump_opt['k']) {
9987 checkpoint_pool = import_checkpointed_state(target, cfg,
9988 target_is_spa, &checkpoint_target);
9989
9990 if (checkpoint_target != NULL)
9991 target = checkpoint_target;
9992 }
9993
9994 if (cfg != NULL) {
9995 nvlist_free(cfg);
9996 cfg = NULL;
9997 }
9998
9999 if (target_pool != target)
10000 free(target_pool);
10001
10002 if (error == 0) {
10003 if (dump_opt['k'] && (target_is_spa || dump_opt['R'])) {
10004 ASSERT(checkpoint_pool != NULL);
10005 ASSERT0P(checkpoint_target);
10006
10007 error = spa_open(checkpoint_pool, &spa, FTAG);
10008 if (error != 0) {
10009 fatal("Tried to open pool \"%s\" but "
10010 "spa_open() failed with error %d\n",
10011 checkpoint_pool, error);
10012 }
10013
10014 } else if (target_is_spa || dump_opt['R'] || dump_opt['B'] ||
10015 objset_id == 0) {
10016 zdb_set_skip_mmp(target);
10017 error = spa_open_rewind(target, &spa, FTAG, policy,
10018 NULL);
10019 if (error) {
10020 /*
10021 * If we're missing the log device then
10022 * try opening the pool after clearing the
10023 * log state.
10024 */
10025 mutex_enter(&spa_namespace_lock);
10026 if ((spa = spa_lookup(target)) != NULL &&
10027 spa->spa_log_state == SPA_LOG_MISSING) {
10028 spa->spa_log_state = SPA_LOG_CLEAR;
10029 error = 0;
10030 }
10031 mutex_exit(&spa_namespace_lock);
10032
10033 if (!error) {
10034 error = spa_open_rewind(target, &spa,
10035 FTAG, policy, NULL);
10036 }
10037 }
10038 } else if (strpbrk(target, "#") != NULL) {
10039 dsl_pool_t *dp;
10040 error = dsl_pool_hold(target, FTAG, &dp);
10041 if (error != 0) {
10042 fatal("can't dump '%s': %s", target,
10043 strerror(error));
10044 }
10045 error = dump_bookmark(dp, target, B_TRUE, verbose > 1);
10046 dsl_pool_rele(dp, FTAG);
10047 if (error != 0) {
10048 fatal("can't dump '%s': %s", target,
10049 strerror(error));
10050 }
10051 goto fini;
10052 } else {
10053 target_pool = strdup(target);
10054 if (strpbrk(target, "/@") != NULL)
10055 *strpbrk(target_pool, "/@") = '\0';
10056
10057 zdb_set_skip_mmp(target);
10058 /*
10059 * If -N was supplied, the user has indicated that
10060 * zdb -d <pool>/<objsetID> is in effect. Otherwise
10061 * we first assume that the dataset string is the
10062 * dataset name. If dmu_objset_hold fails with the
10063 * dataset string, and we have an objset_id, retry the
10064 * lookup with the objsetID.
10065 */
10066 boolean_t retry = B_TRUE;
10067 retry_lookup:
10068 if (dataset_lookup == B_TRUE) {
10069 /*
10070 * Use the supplied id to get the name
10071 * for open_objset.
10072 */
10073 error = spa_open(target_pool, &spa, FTAG);
10074 if (error == 0) {
10075 error = name_from_objset_id(spa,
10076 objset_id, dsname);
10077 spa_close(spa, FTAG);
10078 if (error == 0)
10079 target = dsname;
10080 }
10081 }
10082 if (error == 0) {
10083 if (objset_id > 0 && retry) {
10084 int err = dmu_objset_hold(target, FTAG,
10085 &os);
10086 if (err) {
10087 dataset_lookup = B_TRUE;
10088 retry = B_FALSE;
10089 goto retry_lookup;
10090 } else {
10091 dmu_objset_rele(os, FTAG);
10092 }
10093 }
10094 error = open_objset(target, FTAG, &os);
10095 }
10096 if (error == 0)
10097 spa = dmu_objset_spa(os);
10098 free(target_pool);
10099 }
10100 }
10101 nvlist_free(policy);
10102
10103 if (error)
10104 fatal("can't open '%s': %s", target, strerror(error));
10105
10106 /*
10107 * Set the pool failure mode to panic in order to prevent the pool
10108 * from suspending. A suspended I/O will have no way to resume and
10109 * can prevent the zdb(8) command from terminating as expected.
10110 */
10111 if (spa != NULL)
10112 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
10113
10114 argv++;
10115 argc--;
10116 if (dump_opt['r']) {
10117 error = zdb_copy_object(os, object, argv[1]);
10118 } else if (!dump_opt['R']) {
10119 flagbits['d'] = ZOR_FLAG_DIRECTORY;
10120 flagbits['f'] = ZOR_FLAG_PLAIN_FILE;
10121 flagbits['m'] = ZOR_FLAG_SPACE_MAP;
10122 flagbits['z'] = ZOR_FLAG_ZAP;
10123 flagbits['A'] = ZOR_FLAG_ALL_TYPES;
10124
10125 if (argc > 0 && dump_opt['d']) {
10126 zopt_object_args = argc;
10127 zopt_object_ranges = calloc(zopt_object_args,
10128 sizeof (zopt_object_range_t));
10129 for (unsigned i = 0; i < zopt_object_args; i++) {
10130 int err;
10131 const char *msg = NULL;
10132
10133 err = parse_object_range(argv[i],
10134 &zopt_object_ranges[i], &msg);
10135 if (err != 0)
10136 fatal("Bad object or range: '%s': %s\n",
10137 argv[i], msg ?: "");
10138 }
10139 } else if (argc > 0 && dump_opt['m']) {
10140 zopt_metaslab_args = argc;
10141 zopt_metaslab = calloc(zopt_metaslab_args,
10142 sizeof (uint64_t));
10143 for (unsigned i = 0; i < zopt_metaslab_args; i++) {
10144 errno = 0;
10145 zopt_metaslab[i] = strtoull(argv[i], NULL, 0);
10146 if (zopt_metaslab[i] == 0 && errno != 0)
10147 fatal("bad number %s: %s", argv[i],
10148 strerror(errno));
10149 }
10150 }
10151 if (dump_opt['B']) {
10152 dump_backup(target, objset_id,
10153 argc > 0 ? argv[0] : NULL);
10154 } else if (os != NULL) {
10155 dump_objset(os);
10156 } else if (zopt_object_args > 0 && !dump_opt['m']) {
10157 dump_objset(spa->spa_meta_objset);
10158 } else {
10159 dump_zpool(spa);
10160 }
10161 } else {
10162 flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR;
10163 flagbits['c'] = ZDB_FLAG_CHECKSUM;
10164 flagbits['d'] = ZDB_FLAG_DECOMPRESS;
10165 flagbits['e'] = ZDB_FLAG_BSWAP;
10166 flagbits['g'] = ZDB_FLAG_GBH;
10167 flagbits['i'] = ZDB_FLAG_INDIRECT;
10168 flagbits['r'] = ZDB_FLAG_RAW;
10169 flagbits['v'] = ZDB_FLAG_VERBOSE;
10170
10171 for (int i = 0; i < argc; i++)
10172 zdb_read_block(argv[i], spa);
10173 }
10174
10175 if (dump_opt['k']) {
10176 free(checkpoint_pool);
10177 if (!target_is_spa)
10178 free(checkpoint_target);
10179 }
10180
10181 fini:
10182 if (spa != NULL)
10183 zdb_ddt_cleanup(spa);
10184
10185 if (os != NULL) {
10186 close_objset(os, FTAG);
10187 } else if (spa != NULL) {
10188 spa_close(spa, FTAG);
10189 }
10190
10191 fuid_table_destroy();
10192
10193 dump_debug_buffer();
10194
10195 if (kernel_init_done)
10196 kernel_fini();
10197
10198 if (corruption_found && error == 0)
10199 error = 3;
10200
10201 return (error);
10202 }
10203