?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Fix gcc uninitialized warning in FreeBSD zio_crypt.c
[zfs.git] / cmd / zdb / zdb.c
blob42b82c08249f47773978d0bca6394e8807a754e6
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2016 Nexenta Systems, Inc.
27 * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC.
28 * Copyright (c) 2015, 2017, Intel Corporation.
29 * Copyright (c) 2020 Datto Inc.
30 * Copyright (c) 2020, The FreeBSD Foundation [1]
31 *
32 * [1] Portions of this software were developed by Allan Jude
33 * under sponsorship from the FreeBSD Foundation.
34 * Copyright (c) 2021 Allan Jude
35 * Copyright (c) 2021 Toomas Soome <tsoome@me.com>
36 * Copyright (c) 2023, 2024, Klara Inc.
37 * Copyright (c) 2023, Rob Norris <robn@despairlabs.com>
38 */
39
40 #include <stdio.h>
41 #include <unistd.h>
42 #include <stdlib.h>
43 #include <ctype.h>
44 #include <getopt.h>
45 #include <openssl/evp.h>
46 #include <sys/zfs_context.h>
47 #include <sys/spa.h>
48 #include <sys/spa_impl.h>
49 #include <sys/dmu.h>
50 #include <sys/zap.h>
51 #include <sys/zap_impl.h>
52 #include <sys/fs/zfs.h>
53 #include <sys/zfs_znode.h>
54 #include <sys/zfs_sa.h>
55 #include <sys/sa.h>
56 #include <sys/sa_impl.h>
57 #include <sys/vdev.h>
58 #include <sys/vdev_impl.h>
59 #include <sys/metaslab_impl.h>
60 #include <sys/dmu_objset.h>
61 #include <sys/dsl_dir.h>
62 #include <sys/dsl_dataset.h>
63 #include <sys/dsl_pool.h>
64 #include <sys/dsl_bookmark.h>
65 #include <sys/dbuf.h>
66 #include <sys/zil.h>
67 #include <sys/zil_impl.h>
68 #include <sys/stat.h>
69 #include <sys/resource.h>
70 #include <sys/dmu_send.h>
71 #include <sys/dmu_traverse.h>
72 #include <sys/zio_checksum.h>
73 #include <sys/zio_compress.h>
74 #include <sys/zfs_fuid.h>
75 #include <sys/arc.h>
76 #include <sys/arc_impl.h>
77 #include <sys/ddt.h>
78 #include <sys/ddt_impl.h>
79 #include <sys/zfeature.h>
80 #include <sys/abd.h>
81 #include <sys/blkptr.h>
82 #include <sys/dsl_crypt.h>
83 #include <sys/dsl_scan.h>
84 #include <sys/btree.h>
85 #include <sys/brt.h>
86 #include <sys/brt_impl.h>
87 #include <zfs_comutil.h>
88 #include <sys/zstd/zstd.h>
89 #include <sys/backtrace.h>
90
91 #include <libnvpair.h>
92 #include <libzutil.h>
93 #include <libzfs_core.h>
94
95 #include <libzdb.h>
96
97 #include "zdb.h"
98
99
100 extern int reference_tracking_enable;
101 extern int zfs_recover;
102 extern uint_t zfs_vdev_async_read_max_active;
103 extern boolean_t spa_load_verify_dryrun;
104 extern boolean_t spa_mode_readable_spacemaps;
105 extern uint_t zfs_reconstruct_indirect_combinations_max;
106 extern uint_t zfs_btree_verify_intensity;
107
108 static const char cmdname[] = "zdb";
109 uint8_t dump_opt[256];
110
111 typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size);
112
113 static uint64_t *zopt_metaslab = NULL;
114 static unsigned zopt_metaslab_args = 0;
115
116
117 static zopt_object_range_t *zopt_object_ranges = NULL;
118 static unsigned zopt_object_args = 0;
119
120 static int flagbits[256];
121
122
123 static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */
124 static int leaked_objects = 0;
125 static range_tree_t *mos_refd_objs;
126 static spa_t *spa;
127 static objset_t *os;
128 static boolean_t kernel_init_done;
129
130 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *,
131 boolean_t);
132 static void mos_obj_refd(uint64_t);
133 static void mos_obj_refd_multiple(uint64_t);
134 static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free,
135 dmu_tx_t *tx);
136
137
138
139 static void zdb_print_blkptr(const blkptr_t *bp, int flags);
140 static void zdb_exit(int reason);
141
142 typedef struct sublivelist_verify_block_refcnt {
143 /* block pointer entry in livelist being verified */
144 blkptr_t svbr_blk;
145
146 /*
147 * Refcount gets incremented to 1 when we encounter the first
148 * FREE entry for the svfbr block pointer and a node for it
149 * is created in our ZDB verification/tracking metadata.
150 *
151 * As we encounter more FREE entries we increment this counter
152 * and similarly decrement it whenever we find the respective
153 * ALLOC entries for this block.
154 *
155 * When the refcount gets to 0 it means that all the FREE and
156 * ALLOC entries of this block have paired up and we no longer
157 * need to track it in our verification logic (e.g. the node
158 * containing this struct in our verification data structure
159 * should be freed).
160 *
161 * [refer to sublivelist_verify_blkptr() for the actual code]
162 */
163 uint32_t svbr_refcnt;
164 } sublivelist_verify_block_refcnt_t;
165
166 static int
167 sublivelist_block_refcnt_compare(const void *larg, const void *rarg)
168 {
169 const sublivelist_verify_block_refcnt_t *l = larg;
170 const sublivelist_verify_block_refcnt_t *r = rarg;
171 return (livelist_compare(&l->svbr_blk, &r->svbr_blk));
172 }
173
174 static int
175 sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free,
176 dmu_tx_t *tx)
177 {
178 ASSERT3P(tx, ==, NULL);
179 struct sublivelist_verify *sv = arg;
180 sublivelist_verify_block_refcnt_t current = {
181 .svbr_blk = *bp,
182
183 /*
184 * Start with 1 in case this is the first free entry.
185 * This field is not used for our B-Tree comparisons
186 * anyway.
187 */
188 .svbr_refcnt = 1,
189 };
190
191 zfs_btree_index_t where;
192 sublivelist_verify_block_refcnt_t *pair =
193 zfs_btree_find(&sv->sv_pair, &current, &where);
194 if (free) {
195 if (pair == NULL) {
196 /* first free entry for this block pointer */
197 zfs_btree_add(&sv->sv_pair, &current);
198 } else {
199 pair->svbr_refcnt++;
200 }
201 } else {
202 if (pair == NULL) {
203 /* block that is currently marked as allocated */
204 for (int i = 0; i < SPA_DVAS_PER_BP; i++) {
205 if (DVA_IS_EMPTY(&bp->blk_dva[i]))
206 break;
207 sublivelist_verify_block_t svb = {
208 .svb_dva = bp->blk_dva[i],
209 .svb_allocated_txg =
210 BP_GET_LOGICAL_BIRTH(bp)
211 };
212
213 if (zfs_btree_find(&sv->sv_leftover, &svb,
214 &where) == NULL) {
215 zfs_btree_add_idx(&sv->sv_leftover,
216 &svb, &where);
217 }
218 }
219 } else {
220 /* alloc matches a free entry */
221 pair->svbr_refcnt--;
222 if (pair->svbr_refcnt == 0) {
223 /* all allocs and frees have been matched */
224 zfs_btree_remove_idx(&sv->sv_pair, &where);
225 }
226 }
227 }
228
229 return (0);
230 }
231
232 static int
233 sublivelist_verify_func(void *args, dsl_deadlist_entry_t *dle)
234 {
235 int err;
236 struct sublivelist_verify *sv = args;
237
238 zfs_btree_create(&sv->sv_pair, sublivelist_block_refcnt_compare, NULL,
239 sizeof (sublivelist_verify_block_refcnt_t));
240
241 err = bpobj_iterate_nofree(&dle->dle_bpobj, sublivelist_verify_blkptr,
242 sv, NULL);
243
244 sublivelist_verify_block_refcnt_t *e;
245 zfs_btree_index_t *cookie = NULL;
246 while ((e = zfs_btree_destroy_nodes(&sv->sv_pair, &cookie)) != NULL) {
247 char blkbuf[BP_SPRINTF_LEN];
248 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
249 &e->svbr_blk, B_TRUE);
250 (void) printf("\tERROR: %d unmatched FREE(s): %s\n",
251 e->svbr_refcnt, blkbuf);
252 }
253 zfs_btree_destroy(&sv->sv_pair);
254
255 return (err);
256 }
257
258 static int
259 livelist_block_compare(const void *larg, const void *rarg)
260 {
261 const sublivelist_verify_block_t *l = larg;
262 const sublivelist_verify_block_t *r = rarg;
263
264 if (DVA_GET_VDEV(&l->svb_dva) < DVA_GET_VDEV(&r->svb_dva))
265 return (-1);
266 else if (DVA_GET_VDEV(&l->svb_dva) > DVA_GET_VDEV(&r->svb_dva))
267 return (+1);
268
269 if (DVA_GET_OFFSET(&l->svb_dva) < DVA_GET_OFFSET(&r->svb_dva))
270 return (-1);
271 else if (DVA_GET_OFFSET(&l->svb_dva) > DVA_GET_OFFSET(&r->svb_dva))
272 return (+1);
273
274 if (DVA_GET_ASIZE(&l->svb_dva) < DVA_GET_ASIZE(&r->svb_dva))
275 return (-1);
276 else if (DVA_GET_ASIZE(&l->svb_dva) > DVA_GET_ASIZE(&r->svb_dva))
277 return (+1);
278
279 return (0);
280 }
281
282 /*
283 * Check for errors in a livelist while tracking all unfreed ALLOCs in the
284 * sublivelist_verify_t: sv->sv_leftover
285 */
286 static void
287 livelist_verify(dsl_deadlist_t *dl, void *arg)
288 {
289 sublivelist_verify_t *sv = arg;
290 dsl_deadlist_iterate(dl, sublivelist_verify_func, sv);
291 }
292
293 /*
294 * Check for errors in the livelist entry and discard the intermediary
295 * data structures
296 */
297 static int
298 sublivelist_verify_lightweight(void *args, dsl_deadlist_entry_t *dle)
299 {
300 (void) args;
301 sublivelist_verify_t sv;
302 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
303 sizeof (sublivelist_verify_block_t));
304 int err = sublivelist_verify_func(&sv, dle);
305 zfs_btree_clear(&sv.sv_leftover);
306 zfs_btree_destroy(&sv.sv_leftover);
307 return (err);
308 }
309
310 typedef struct metaslab_verify {
311 /*
312 * Tree containing all the leftover ALLOCs from the livelists
313 * that are part of this metaslab.
314 */
315 zfs_btree_t mv_livelist_allocs;
316
317 /*
318 * Metaslab information.
319 */
320 uint64_t mv_vdid;
321 uint64_t mv_msid;
322 uint64_t mv_start;
323 uint64_t mv_end;
324
325 /*
326 * What's currently allocated for this metaslab.
327 */
328 range_tree_t *mv_allocated;
329 } metaslab_verify_t;
330
331 typedef void ll_iter_t(dsl_deadlist_t *ll, void *arg);
332
333 typedef int (*zdb_log_sm_cb_t)(spa_t *spa, space_map_entry_t *sme, uint64_t txg,
334 void *arg);
335
336 typedef struct unflushed_iter_cb_arg {
337 spa_t *uic_spa;
338 uint64_t uic_txg;
339 void *uic_arg;
340 zdb_log_sm_cb_t uic_cb;
341 } unflushed_iter_cb_arg_t;
342
343 static int
344 iterate_through_spacemap_logs_cb(space_map_entry_t *sme, void *arg)
345 {
346 unflushed_iter_cb_arg_t *uic = arg;
347 return (uic->uic_cb(uic->uic_spa, sme, uic->uic_txg, uic->uic_arg));
348 }
349
350 static void
351 iterate_through_spacemap_logs(spa_t *spa, zdb_log_sm_cb_t cb, void *arg)
352 {
353 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
354 return;
355
356 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
357 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
358 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
359 space_map_t *sm = NULL;
360 VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
361 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
362
363 unflushed_iter_cb_arg_t uic = {
364 .uic_spa = spa,
365 .uic_txg = sls->sls_txg,
366 .uic_arg = arg,
367 .uic_cb = cb
368 };
369 VERIFY0(space_map_iterate(sm, space_map_length(sm),
370 iterate_through_spacemap_logs_cb, &uic));
371 space_map_close(sm);
372 }
373 spa_config_exit(spa, SCL_CONFIG, FTAG);
374 }
375
376 static void
377 verify_livelist_allocs(metaslab_verify_t *mv, uint64_t txg,
378 uint64_t offset, uint64_t size)
379 {
380 sublivelist_verify_block_t svb = {{{0}}};
381 DVA_SET_VDEV(&svb.svb_dva, mv->mv_vdid);
382 DVA_SET_OFFSET(&svb.svb_dva, offset);
383 DVA_SET_ASIZE(&svb.svb_dva, size);
384 zfs_btree_index_t where;
385 uint64_t end_offset = offset + size;
386
387 /*
388 * Look for an exact match for spacemap entry in the livelist entries.
389 * Then, look for other livelist entries that fall within the range
390 * of the spacemap entry as it may have been condensed
391 */
392 sublivelist_verify_block_t *found =
393 zfs_btree_find(&mv->mv_livelist_allocs, &svb, &where);
394 if (found == NULL) {
395 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where);
396 }
397 for (; found != NULL && DVA_GET_VDEV(&found->svb_dva) == mv->mv_vdid &&
398 DVA_GET_OFFSET(&found->svb_dva) < end_offset;
399 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
400 if (found->svb_allocated_txg <= txg) {
401 (void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
402 "from TXG %llx FREED at TXG %llx\n",
403 (u_longlong_t)DVA_GET_OFFSET(&found->svb_dva),
404 (u_longlong_t)DVA_GET_ASIZE(&found->svb_dva),
405 (u_longlong_t)found->svb_allocated_txg,
406 (u_longlong_t)txg);
407 }
408 }
409 }
410
411 static int
412 metaslab_spacemap_validation_cb(space_map_entry_t *sme, void *arg)
413 {
414 metaslab_verify_t *mv = arg;
415 uint64_t offset = sme->sme_offset;
416 uint64_t size = sme->sme_run;
417 uint64_t txg = sme->sme_txg;
418
419 if (sme->sme_type == SM_ALLOC) {
420 if (range_tree_contains(mv->mv_allocated,
421 offset, size)) {
422 (void) printf("ERROR: DOUBLE ALLOC: "
423 "%llu [%llx:%llx] "
424 "%llu:%llu LOG_SM\n",
425 (u_longlong_t)txg, (u_longlong_t)offset,
426 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
427 (u_longlong_t)mv->mv_msid);
428 } else {
429 range_tree_add(mv->mv_allocated,
430 offset, size);
431 }
432 } else {
433 if (!range_tree_contains(mv->mv_allocated,
434 offset, size)) {
435 (void) printf("ERROR: DOUBLE FREE: "
436 "%llu [%llx:%llx] "
437 "%llu:%llu LOG_SM\n",
438 (u_longlong_t)txg, (u_longlong_t)offset,
439 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
440 (u_longlong_t)mv->mv_msid);
441 } else {
442 range_tree_remove(mv->mv_allocated,
443 offset, size);
444 }
445 }
446
447 if (sme->sme_type != SM_ALLOC) {
448 /*
449 * If something is freed in the spacemap, verify that
450 * it is not listed as allocated in the livelist.
451 */
452 verify_livelist_allocs(mv, txg, offset, size);
453 }
454 return (0);
455 }
456
457 static int
458 spacemap_check_sm_log_cb(spa_t *spa, space_map_entry_t *sme,
459 uint64_t txg, void *arg)
460 {
461 metaslab_verify_t *mv = arg;
462 uint64_t offset = sme->sme_offset;
463 uint64_t vdev_id = sme->sme_vdev;
464
465 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
466
467 /* skip indirect vdevs */
468 if (!vdev_is_concrete(vd))
469 return (0);
470
471 if (vdev_id != mv->mv_vdid)
472 return (0);
473
474 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
475 if (ms->ms_id != mv->mv_msid)
476 return (0);
477
478 if (txg < metaslab_unflushed_txg(ms))
479 return (0);
480
481
482 ASSERT3U(txg, ==, sme->sme_txg);
483 return (metaslab_spacemap_validation_cb(sme, mv));
484 }
485
486 static void
487 spacemap_check_sm_log(spa_t *spa, metaslab_verify_t *mv)
488 {
489 iterate_through_spacemap_logs(spa, spacemap_check_sm_log_cb, mv);
490 }
491
492 static void
493 spacemap_check_ms_sm(space_map_t *sm, metaslab_verify_t *mv)
494 {
495 if (sm == NULL)
496 return;
497
498 VERIFY0(space_map_iterate(sm, space_map_length(sm),
499 metaslab_spacemap_validation_cb, mv));
500 }
501
502 static void iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg);
503
504 /*
505 * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
506 * they are part of that metaslab (mv_msid).
507 */
508 static void
509 mv_populate_livelist_allocs(metaslab_verify_t *mv, sublivelist_verify_t *sv)
510 {
511 zfs_btree_index_t where;
512 sublivelist_verify_block_t *svb;
513 ASSERT3U(zfs_btree_numnodes(&mv->mv_livelist_allocs), ==, 0);
514 for (svb = zfs_btree_first(&sv->sv_leftover, &where);
515 svb != NULL;
516 svb = zfs_btree_next(&sv->sv_leftover, &where, &where)) {
517 if (DVA_GET_VDEV(&svb->svb_dva) != mv->mv_vdid)
518 continue;
519
520 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start &&
521 (DVA_GET_OFFSET(&svb->svb_dva) +
522 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_start) {
523 (void) printf("ERROR: Found block that crosses "
524 "metaslab boundary: <%llu:%llx:%llx>\n",
525 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
526 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
527 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
528 continue;
529 }
530
531 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start)
532 continue;
533
534 if (DVA_GET_OFFSET(&svb->svb_dva) >= mv->mv_end)
535 continue;
536
537 if ((DVA_GET_OFFSET(&svb->svb_dva) +
538 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_end) {
539 (void) printf("ERROR: Found block that crosses "
540 "metaslab boundary: <%llu:%llx:%llx>\n",
541 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
542 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
543 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
544 continue;
545 }
546
547 zfs_btree_add(&mv->mv_livelist_allocs, svb);
548 }
549
550 for (svb = zfs_btree_first(&mv->mv_livelist_allocs, &where);
551 svb != NULL;
552 svb = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
553 zfs_btree_remove(&sv->sv_leftover, svb);
554 }
555 }
556
557 /*
558 * [Livelist Check]
559 * Iterate through all the sublivelists and:
560 * - report leftover frees (**)
561 * - record leftover ALLOCs together with their TXG [see Cross Check]
562 *
563 * (**) Note: Double ALLOCs are valid in datasets that have dedup
564 * enabled. Similarly double FREEs are allowed as well but
565 * only if they pair up with a corresponding ALLOC entry once
566 * we our done with our sublivelist iteration.
567 *
568 * [Spacemap Check]
569 * for each metaslab:
570 * - iterate over spacemap and then the metaslab's entries in the
571 * spacemap log, then report any double FREEs and ALLOCs (do not
572 * blow up).
573 *
574 * [Cross Check]
575 * After finishing the Livelist Check phase and while being in the
576 * Spacemap Check phase, we find all the recorded leftover ALLOCs
577 * of the livelist check that are part of the metaslab that we are
578 * currently looking at in the Spacemap Check. We report any entries
579 * that are marked as ALLOCs in the livelists but have been actually
580 * freed (and potentially allocated again) after their TXG stamp in
581 * the spacemaps. Also report any ALLOCs from the livelists that
582 * belong to indirect vdevs (e.g. their vdev completed removal).
583 *
584 * Note that this will miss Log Spacemap entries that cancelled each other
585 * out before being flushed to the metaslab, so we are not guaranteed
586 * to match all erroneous ALLOCs.
587 */
588 static void
589 livelist_metaslab_validate(spa_t *spa)
590 {
591 (void) printf("Verifying deleted livelist entries\n");
592
593 sublivelist_verify_t sv;
594 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
595 sizeof (sublivelist_verify_block_t));
596 iterate_deleted_livelists(spa, livelist_verify, &sv);
597
598 (void) printf("Verifying metaslab entries\n");
599 vdev_t *rvd = spa->spa_root_vdev;
600 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
601 vdev_t *vd = rvd->vdev_child[c];
602
603 if (!vdev_is_concrete(vd))
604 continue;
605
606 for (uint64_t mid = 0; mid < vd->vdev_ms_count; mid++) {
607 metaslab_t *m = vd->vdev_ms[mid];
608
609 (void) fprintf(stderr,
610 "\rverifying concrete vdev %llu, "
611 "metaslab %llu of %llu ...",
612 (longlong_t)vd->vdev_id,
613 (longlong_t)mid,
614 (longlong_t)vd->vdev_ms_count);
615
616 uint64_t shift, start;
617 range_seg_type_t type =
618 metaslab_calculate_range_tree_type(vd, m,
619 &start, &shift);
620 metaslab_verify_t mv;
621 mv.mv_allocated = range_tree_create(NULL,
622 type, NULL, start, shift);
623 mv.mv_vdid = vd->vdev_id;
624 mv.mv_msid = m->ms_id;
625 mv.mv_start = m->ms_start;
626 mv.mv_end = m->ms_start + m->ms_size;
627 zfs_btree_create(&mv.mv_livelist_allocs,
628 livelist_block_compare, NULL,
629 sizeof (sublivelist_verify_block_t));
630
631 mv_populate_livelist_allocs(&mv, &sv);
632
633 spacemap_check_ms_sm(m->ms_sm, &mv);
634 spacemap_check_sm_log(spa, &mv);
635
636 range_tree_vacate(mv.mv_allocated, NULL, NULL);
637 range_tree_destroy(mv.mv_allocated);
638 zfs_btree_clear(&mv.mv_livelist_allocs);
639 zfs_btree_destroy(&mv.mv_livelist_allocs);
640 }
641 }
642 (void) fprintf(stderr, "\n");
643
644 /*
645 * If there are any segments in the leftover tree after we walked
646 * through all the metaslabs in the concrete vdevs then this means
647 * that we have segments in the livelists that belong to indirect
648 * vdevs and are marked as allocated.
649 */
650 if (zfs_btree_numnodes(&sv.sv_leftover) == 0) {
651 zfs_btree_destroy(&sv.sv_leftover);
652 return;
653 }
654 (void) printf("ERROR: Found livelist blocks marked as allocated "
655 "for indirect vdevs:\n");
656
657 zfs_btree_index_t *where = NULL;
658 sublivelist_verify_block_t *svb;
659 while ((svb = zfs_btree_destroy_nodes(&sv.sv_leftover, &where)) !=
660 NULL) {
661 int vdev_id = DVA_GET_VDEV(&svb->svb_dva);
662 ASSERT3U(vdev_id, <, rvd->vdev_children);
663 vdev_t *vd = rvd->vdev_child[vdev_id];
664 ASSERT(!vdev_is_concrete(vd));
665 (void) printf("<%d:%llx:%llx> TXG %llx\n",
666 vdev_id, (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
667 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva),
668 (u_longlong_t)svb->svb_allocated_txg);
669 }
670 (void) printf("\n");
671 zfs_btree_destroy(&sv.sv_leftover);
672 }
673
674 /*
675 * These libumem hooks provide a reasonable set of defaults for the allocator's
676 * debugging facilities.
677 */
678 const char *
679 _umem_debug_init(void)
680 {
681 return ("default,verbose"); /* $UMEM_DEBUG setting */
682 }
683
684 const char *
685 _umem_logging_init(void)
686 {
687 return ("fail,contents"); /* $UMEM_LOGGING setting */
688 }
689
690 static void
691 usage(void)
692 {
693 (void) fprintf(stderr,
694 "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
695 "[-I <inflight I/Os>]\n"
696 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
697 "\t\t[-K <key>]\n"
698 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
699 "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
700 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
701 "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
702 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
703 "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
704 "\t%s [-v] <bookmark>\n"
705 "\t%s -C [-A] [-U <cache>] [<poolname>]\n"
706 "\t%s -l [-Aqu] <device>\n"
707 "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
708 "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
709 "\t%s -O [-K <key>] <dataset> <path>\n"
710 "\t%s -r [-K <key>] <dataset> <path> <destination>\n"
711 "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
712 "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
713 "\t%s -E [-A] word0:word1:...:word15\n"
714 "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
715 "<poolname>\n\n",
716 cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
717 cmdname, cmdname, cmdname, cmdname, cmdname);
718
719 (void) fprintf(stderr, " Dataset name must include at least one "
720 "separator character '/' or '@'\n");
721 (void) fprintf(stderr, " If dataset name is specified, only that "
722 "dataset is dumped\n");
723 (void) fprintf(stderr, " If object numbers or object number "
724 "ranges are specified, only those\n"
725 " objects or ranges are dumped.\n\n");
726 (void) fprintf(stderr,
727 " Object ranges take the form <start>:<end>[:<flags>]\n"
728 " start Starting object number\n"
729 " end Ending object number, or -1 for no upper bound\n"
730 " flags Optional flags to select object types:\n"
731 " A All objects (this is the default)\n"
732 " d ZFS directories\n"
733 " f ZFS files \n"
734 " m SPA space maps\n"
735 " z ZAPs\n"
736 " - Negate effect of next flag\n\n");
737 (void) fprintf(stderr, " Options to control amount of output:\n");
738 (void) fprintf(stderr, " -b --block-stats "
739 "block statistics\n");
740 (void) fprintf(stderr, " -B --backup "
741 "backup stream\n");
742 (void) fprintf(stderr, " -c --checksum "
743 "checksum all metadata (twice for all data) blocks\n");
744 (void) fprintf(stderr, " -C --config "
745 "config (or cachefile if alone)\n");
746 (void) fprintf(stderr, " -d --datasets "
747 "dataset(s)\n");
748 (void) fprintf(stderr, " -D --dedup-stats "
749 "dedup statistics\n");
750 (void) fprintf(stderr, " -E --embedded-block-pointer=INTEGER\n"
751 " decode and display block "
752 "from an embedded block pointer\n");
753 (void) fprintf(stderr, " -h --history "
754 "pool history\n");
755 (void) fprintf(stderr, " -i --intent-logs "
756 "intent logs\n");
757 (void) fprintf(stderr, " -l --label "
758 "read label contents\n");
759 (void) fprintf(stderr, " -k --checkpointed-state "
760 "examine the checkpointed state of the pool\n");
761 (void) fprintf(stderr, " -L --disable-leak-tracking "
762 "disable leak tracking (do not load spacemaps)\n");
763 (void) fprintf(stderr, " -m --metaslabs "
764 "metaslabs\n");
765 (void) fprintf(stderr, " -M --metaslab-groups "
766 "metaslab groups\n");
767 (void) fprintf(stderr, " -O --object-lookups "
768 "perform object lookups by path\n");
769 (void) fprintf(stderr, " -r --copy-object "
770 "copy an object by path to file\n");
771 (void) fprintf(stderr, " -R --read-block "
772 "read and display block from a device\n");
773 (void) fprintf(stderr, " -s --io-stats "
774 "report stats on zdb's I/O\n");
775 (void) fprintf(stderr, " -S --simulate-dedup "
776 "simulate dedup to measure effect\n");
777 (void) fprintf(stderr, " -v --verbose "
778 "verbose (applies to all others)\n");
779 (void) fprintf(stderr, " -y --livelist "
780 "perform livelist and metaslab validation on any livelists being "
781 "deleted\n\n");
782 (void) fprintf(stderr, " Below options are intended for use "
783 "with other options:\n");
784 (void) fprintf(stderr, " -A --ignore-assertions "
785 "ignore assertions (-A), enable panic recovery (-AA) or both "
786 "(-AAA)\n");
787 (void) fprintf(stderr, " -e --exported "
788 "pool is exported/destroyed/has altroot/not in a cachefile\n");
789 (void) fprintf(stderr, " -F --automatic-rewind "
790 "attempt automatic rewind within safe range of transaction "
791 "groups\n");
792 (void) fprintf(stderr, " -G --dump-debug-msg "
793 "dump zfs_dbgmsg buffer before exiting\n");
794 (void) fprintf(stderr, " -I --inflight=INTEGER "
795 "specify the maximum number of checksumming I/Os "
796 "[default is 200]\n");
797 (void) fprintf(stderr, " -K --key=KEY "
798 "decryption key for encrypted dataset\n");
799 (void) fprintf(stderr, " -o --option=\"OPTION=INTEGER\" "
800 "set global variable to an unsigned 32-bit integer\n");
801 (void) fprintf(stderr, " -p --path==PATH "
802 "use one or more with -e to specify path to vdev dir\n");
803 (void) fprintf(stderr, " -P --parseable "
804 "print numbers in parseable form\n");
805 (void) fprintf(stderr, " -q --skip-label "
806 "don't print label contents\n");
807 (void) fprintf(stderr, " -t --txg=INTEGER "
808 "highest txg to use when searching for uberblocks\n");
809 (void) fprintf(stderr, " -T --brt-stats "
810 "BRT statistics\n");
811 (void) fprintf(stderr, " -u --uberblock "
812 "uberblock\n");
813 (void) fprintf(stderr, " -U --cachefile=PATH "
814 "use alternate cachefile\n");
815 (void) fprintf(stderr, " -V --verbatim "
816 "do verbatim import\n");
817 (void) fprintf(stderr, " -x --dump-blocks=PATH "
818 "dump all read blocks into specified directory\n");
819 (void) fprintf(stderr, " -X --extreme-rewind "
820 "attempt extreme rewind (does not work with dataset)\n");
821 (void) fprintf(stderr, " -Y --all-reconstruction "
822 "attempt all reconstruction combinations for split blocks\n");
823 (void) fprintf(stderr, " -Z --zstd-headers "
824 "show ZSTD headers \n");
825 (void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
826 "to make only that option verbose\n");
827 (void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
828 zdb_exit(1);
829 }
830
831 static void
832 dump_debug_buffer(void)
833 {
834 ssize_t ret __attribute__((unused));
835
836 if (!dump_opt['G'])
837 return;
838 /*
839 * We use write() instead of printf() so that this function
840 * is safe to call from a signal handler.
841 */
842 ret = write(STDERR_FILENO, "\n", 1);
843 zfs_dbgmsg_print(STDERR_FILENO, "zdb");
844 }
845
846 static void sig_handler(int signo)
847 {
848 struct sigaction action;
849
850 libspl_backtrace(STDERR_FILENO);
851 dump_debug_buffer();
852
853 /*
854 * Restore default action and re-raise signal so SIGSEGV and
855 * SIGABRT can trigger a core dump.
856 */
857 action.sa_handler = SIG_DFL;
858 sigemptyset(&action.sa_mask);
859 action.sa_flags = 0;
860 (void) sigaction(signo, &action, NULL);
861 raise(signo);
862 }
863
864 /*
865 * Called for usage errors that are discovered after a call to spa_open(),
866 * dmu_bonus_hold(), or pool_match(). abort() is called for other errors.
867 */
868
869 static void
870 fatal(const char *fmt, ...)
871 {
872 va_list ap;
873
874 va_start(ap, fmt);
875 (void) fprintf(stderr, "%s: ", cmdname);
876 (void) vfprintf(stderr, fmt, ap);
877 va_end(ap);
878 (void) fprintf(stderr, "\n");
879
880 dump_debug_buffer();
881
882 zdb_exit(1);
883 }
884
885 static void
886 dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size)
887 {
888 (void) size;
889 nvlist_t *nv;
890 size_t nvsize = *(uint64_t *)data;
891 char *packed = umem_alloc(nvsize, UMEM_NOFAIL);
892
893 VERIFY(0 == dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH));
894
895 VERIFY(nvlist_unpack(packed, nvsize, &nv, 0) == 0);
896
897 umem_free(packed, nvsize);
898
899 dump_nvlist(nv, 8);
900
901 nvlist_free(nv);
902 }
903
904 static void
905 dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size)
906 {
907 (void) os, (void) object, (void) size;
908 spa_history_phys_t *shp = data;
909
910 if (shp == NULL)
911 return;
912
913 (void) printf("\t\tpool_create_len = %llu\n",
914 (u_longlong_t)shp->sh_pool_create_len);
915 (void) printf("\t\tphys_max_off = %llu\n",
916 (u_longlong_t)shp->sh_phys_max_off);
917 (void) printf("\t\tbof = %llu\n",
918 (u_longlong_t)shp->sh_bof);
919 (void) printf("\t\teof = %llu\n",
920 (u_longlong_t)shp->sh_eof);
921 (void) printf("\t\trecords_lost = %llu\n",
922 (u_longlong_t)shp->sh_records_lost);
923 }
924
925 static void
926 zdb_nicenum(uint64_t num, char *buf, size_t buflen)
927 {
928 if (dump_opt['P'])
929 (void) snprintf(buf, buflen, "%llu", (longlong_t)num);
930 else
931 nicenum(num, buf, buflen);
932 }
933
934 static void
935 zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen)
936 {
937 if (dump_opt['P'])
938 (void) snprintf(buf, buflen, "%llu", (longlong_t)bytes);
939 else
940 zfs_nicebytes(bytes, buf, buflen);
941 }
942
943 static const char histo_stars[] = "****************************************";
944 static const uint64_t histo_width = sizeof (histo_stars) - 1;
945
946 static void
947 dump_histogram(const uint64_t *histo, int size, int offset)
948 {
949 int i;
950 int minidx = size - 1;
951 int maxidx = 0;
952 uint64_t max = 0;
953
954 for (i = 0; i < size; i++) {
955 if (histo[i] == 0)
956 continue;
957 if (histo[i] > max)
958 max = histo[i];
959 if (i > maxidx)
960 maxidx = i;
961 if (i < minidx)
962 minidx = i;
963 }
964
965 if (max < histo_width)
966 max = histo_width;
967
968 for (i = minidx; i <= maxidx; i++) {
969 (void) printf("\t\t\t%3u: %6llu %s\n",
970 i + offset, (u_longlong_t)histo[i],
971 &histo_stars[(max - histo[i]) * histo_width / max]);
972 }
973 }
974
975 static void
976 dump_zap_stats(objset_t *os, uint64_t object)
977 {
978 int error;
979 zap_stats_t zs;
980
981 error = zap_get_stats(os, object, &zs);
982 if (error)
983 return;
984
985 if (zs.zs_ptrtbl_len == 0) {
986 ASSERT(zs.zs_num_blocks == 1);
987 (void) printf("\tmicrozap: %llu bytes, %llu entries\n",
988 (u_longlong_t)zs.zs_blocksize,
989 (u_longlong_t)zs.zs_num_entries);
990 return;
991 }
992
993 (void) printf("\tFat ZAP stats:\n");
994
995 (void) printf("\t\tPointer table:\n");
996 (void) printf("\t\t\t%llu elements\n",
997 (u_longlong_t)zs.zs_ptrtbl_len);
998 (void) printf("\t\t\tzt_blk: %llu\n",
999 (u_longlong_t)zs.zs_ptrtbl_zt_blk);
1000 (void) printf("\t\t\tzt_numblks: %llu\n",
1001 (u_longlong_t)zs.zs_ptrtbl_zt_numblks);
1002 (void) printf("\t\t\tzt_shift: %llu\n",
1003 (u_longlong_t)zs.zs_ptrtbl_zt_shift);
1004 (void) printf("\t\t\tzt_blks_copied: %llu\n",
1005 (u_longlong_t)zs.zs_ptrtbl_blks_copied);
1006 (void) printf("\t\t\tzt_nextblk: %llu\n",
1007 (u_longlong_t)zs.zs_ptrtbl_nextblk);
1008
1009 (void) printf("\t\tZAP entries: %llu\n",
1010 (u_longlong_t)zs.zs_num_entries);
1011 (void) printf("\t\tLeaf blocks: %llu\n",
1012 (u_longlong_t)zs.zs_num_leafs);
1013 (void) printf("\t\tTotal blocks: %llu\n",
1014 (u_longlong_t)zs.zs_num_blocks);
1015 (void) printf("\t\tzap_block_type: 0x%llx\n",
1016 (u_longlong_t)zs.zs_block_type);
1017 (void) printf("\t\tzap_magic: 0x%llx\n",
1018 (u_longlong_t)zs.zs_magic);
1019 (void) printf("\t\tzap_salt: 0x%llx\n",
1020 (u_longlong_t)zs.zs_salt);
1021
1022 (void) printf("\t\tLeafs with 2^n pointers:\n");
1023 dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);
1024
1025 (void) printf("\t\tBlocks with n*5 entries:\n");
1026 dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);
1027
1028 (void) printf("\t\tBlocks n/10 full:\n");
1029 dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);
1030
1031 (void) printf("\t\tEntries with n chunks:\n");
1032 dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);
1033
1034 (void) printf("\t\tBuckets with n entries:\n");
1035 dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
1036 }
1037
1038 static void
1039 dump_none(objset_t *os, uint64_t object, void *data, size_t size)
1040 {
1041 (void) os, (void) object, (void) data, (void) size;
1042 }
1043
1044 static void
1045 dump_unknown(objset_t *os, uint64_t object, void *data, size_t size)
1046 {
1047 (void) os, (void) object, (void) data, (void) size;
1048 (void) printf("\tUNKNOWN OBJECT TYPE\n");
1049 }
1050
1051 static void
1052 dump_uint8(objset_t *os, uint64_t object, void *data, size_t size)
1053 {
1054 (void) os, (void) object, (void) data, (void) size;
1055 }
1056
1057 static void
1058 dump_uint64(objset_t *os, uint64_t object, void *data, size_t size)
1059 {
1060 uint64_t *arr;
1061 uint64_t oursize;
1062 if (dump_opt['d'] < 6)
1063 return;
1064
1065 if (data == NULL) {
1066 dmu_object_info_t doi;
1067
1068 VERIFY0(dmu_object_info(os, object, &doi));
1069 size = doi.doi_max_offset;
1070 /*
1071 * We cap the size at 1 mebibyte here to prevent
1072 * allocation failures and nigh-infinite printing if the
1073 * object is extremely large.
1074 */
1075 oursize = MIN(size, 1 << 20);
1076 arr = kmem_alloc(oursize, KM_SLEEP);
1077
1078 int err = dmu_read(os, object, 0, oursize, arr, 0);
1079 if (err != 0) {
1080 (void) printf("got error %u from dmu_read\n", err);
1081 kmem_free(arr, oursize);
1082 return;
1083 }
1084 } else {
1085 /*
1086 * Even though the allocation is already done in this code path,
1087 * we still cap the size to prevent excessive printing.
1088 */
1089 oursize = MIN(size, 1 << 20);
1090 arr = data;
1091 }
1092
1093 if (size == 0) {
1094 if (data == NULL)
1095 kmem_free(arr, oursize);
1096 (void) printf("\t\t[]\n");
1097 return;
1098 }
1099
1100 (void) printf("\t\t[%0llx", (u_longlong_t)arr[0]);
1101 for (size_t i = 1; i * sizeof (uint64_t) < oursize; i++) {
1102 if (i % 4 != 0)
1103 (void) printf(", %0llx", (u_longlong_t)arr[i]);
1104 else
1105 (void) printf(",\n\t\t%0llx", (u_longlong_t)arr[i]);
1106 }
1107 if (oursize != size)
1108 (void) printf(", ... ");
1109 (void) printf("]\n");
1110
1111 if (data == NULL)
1112 kmem_free(arr, oursize);
1113 }
1114
1115 static void
1116 dump_zap(objset_t *os, uint64_t object, void *data, size_t size)
1117 {
1118 (void) data, (void) size;
1119 zap_cursor_t zc;
1120 zap_attribute_t *attrp = zap_attribute_long_alloc();
1121 void *prop;
1122 unsigned i;
1123
1124 dump_zap_stats(os, object);
1125 (void) printf("\n");
1126
1127 for (zap_cursor_init(&zc, os, object);
1128 zap_cursor_retrieve(&zc, attrp) == 0;
1129 zap_cursor_advance(&zc)) {
1130 boolean_t key64 =
1131 !!(zap_getflags(zc.zc_zap) & ZAP_FLAG_UINT64_KEY);
1132
1133 if (key64)
1134 (void) printf("\t\t0x%010" PRIu64 "x = ",
1135 *(uint64_t *)attrp->za_name);
1136 else
1137 (void) printf("\t\t%s = ", attrp->za_name);
1138
1139 if (attrp->za_num_integers == 0) {
1140 (void) printf("\n");
1141 continue;
1142 }
1143 prop = umem_zalloc(attrp->za_num_integers *
1144 attrp->za_integer_length, UMEM_NOFAIL);
1145
1146 if (key64)
1147 (void) zap_lookup_uint64(os, object,
1148 (const uint64_t *)attrp->za_name, 1,
1149 attrp->za_integer_length, attrp->za_num_integers,
1150 prop);
1151 else
1152 (void) zap_lookup(os, object, attrp->za_name,
1153 attrp->za_integer_length, attrp->za_num_integers,
1154 prop);
1155
1156 if (attrp->za_integer_length == 1 && !key64) {
1157 if (strcmp(attrp->za_name,
1158 DSL_CRYPTO_KEY_MASTER_KEY) == 0 ||
1159 strcmp(attrp->za_name,
1160 DSL_CRYPTO_KEY_HMAC_KEY) == 0 ||
1161 strcmp(attrp->za_name, DSL_CRYPTO_KEY_IV) == 0 ||
1162 strcmp(attrp->za_name, DSL_CRYPTO_KEY_MAC) == 0 ||
1163 strcmp(attrp->za_name,
1164 DMU_POOL_CHECKSUM_SALT) == 0) {
1165 uint8_t *u8 = prop;
1166
1167 for (i = 0; i < attrp->za_num_integers; i++) {
1168 (void) printf("%02x", u8[i]);
1169 }
1170 } else {
1171 (void) printf("%s", (char *)prop);
1172 }
1173 } else {
1174 for (i = 0; i < attrp->za_num_integers; i++) {
1175 switch (attrp->za_integer_length) {
1176 case 1:
1177 (void) printf("%u ",
1178 ((uint8_t *)prop)[i]);
1179 break;
1180 case 2:
1181 (void) printf("%u ",
1182 ((uint16_t *)prop)[i]);
1183 break;
1184 case 4:
1185 (void) printf("%u ",
1186 ((uint32_t *)prop)[i]);
1187 break;
1188 case 8:
1189 (void) printf("%lld ",
1190 (u_longlong_t)((int64_t *)prop)[i]);
1191 break;
1192 }
1193 }
1194 }
1195 (void) printf("\n");
1196 umem_free(prop,
1197 attrp->za_num_integers * attrp->za_integer_length);
1198 }
1199 zap_cursor_fini(&zc);
1200 zap_attribute_free(attrp);
1201 }
1202
1203 static void
1204 dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size)
1205 {
1206 bpobj_phys_t *bpop = data;
1207 uint64_t i;
1208 char bytes[32], comp[32], uncomp[32];
1209
1210 /* make sure the output won't get truncated */
1211 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
1212 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
1213 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
1214
1215 if (bpop == NULL)
1216 return;
1217
1218 zdb_nicenum(bpop->bpo_bytes, bytes, sizeof (bytes));
1219 zdb_nicenum(bpop->bpo_comp, comp, sizeof (comp));
1220 zdb_nicenum(bpop->bpo_uncomp, uncomp, sizeof (uncomp));
1221
1222 (void) printf("\t\tnum_blkptrs = %llu\n",
1223 (u_longlong_t)bpop->bpo_num_blkptrs);
1224 (void) printf("\t\tbytes = %s\n", bytes);
1225 if (size >= BPOBJ_SIZE_V1) {
1226 (void) printf("\t\tcomp = %s\n", comp);
1227 (void) printf("\t\tuncomp = %s\n", uncomp);
1228 }
1229 if (size >= BPOBJ_SIZE_V2) {
1230 (void) printf("\t\tsubobjs = %llu\n",
1231 (u_longlong_t)bpop->bpo_subobjs);
1232 (void) printf("\t\tnum_subobjs = %llu\n",
1233 (u_longlong_t)bpop->bpo_num_subobjs);
1234 }
1235 if (size >= sizeof (*bpop)) {
1236 (void) printf("\t\tnum_freed = %llu\n",
1237 (u_longlong_t)bpop->bpo_num_freed);
1238 }
1239
1240 if (dump_opt['d'] < 5)
1241 return;
1242
1243 for (i = 0; i < bpop->bpo_num_blkptrs; i++) {
1244 char blkbuf[BP_SPRINTF_LEN];
1245 blkptr_t bp;
1246
1247 int err = dmu_read(os, object,
1248 i * sizeof (bp), sizeof (bp), &bp, 0);
1249 if (err != 0) {
1250 (void) printf("got error %u from dmu_read\n", err);
1251 break;
1252 }
1253 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp,
1254 BP_GET_FREE(&bp));
1255 (void) printf("\t%s\n", blkbuf);
1256 }
1257 }
1258
1259 static void
1260 dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size)
1261 {
1262 (void) data, (void) size;
1263 dmu_object_info_t doi;
1264 int64_t i;
1265
1266 VERIFY0(dmu_object_info(os, object, &doi));
1267 uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP);
1268
1269 int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0);
1270 if (err != 0) {
1271 (void) printf("got error %u from dmu_read\n", err);
1272 kmem_free(subobjs, doi.doi_max_offset);
1273 return;
1274 }
1275
1276 int64_t last_nonzero = -1;
1277 for (i = 0; i < doi.doi_max_offset / 8; i++) {
1278 if (subobjs[i] != 0)
1279 last_nonzero = i;
1280 }
1281
1282 for (i = 0; i <= last_nonzero; i++) {
1283 (void) printf("\t%llu\n", (u_longlong_t)subobjs[i]);
1284 }
1285 kmem_free(subobjs, doi.doi_max_offset);
1286 }
1287
1288 static void
1289 dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size)
1290 {
1291 (void) data, (void) size;
1292 dump_zap_stats(os, object);
1293 /* contents are printed elsewhere, properly decoded */
1294 }
1295
1296 static void
1297 dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size)
1298 {
1299 (void) data, (void) size;
1300 zap_cursor_t zc;
1301 zap_attribute_t *attrp = zap_attribute_alloc();
1302
1303 dump_zap_stats(os, object);
1304 (void) printf("\n");
1305
1306 for (zap_cursor_init(&zc, os, object);
1307 zap_cursor_retrieve(&zc, attrp) == 0;
1308 zap_cursor_advance(&zc)) {
1309 (void) printf("\t\t%s = ", attrp->za_name);
1310 if (attrp->za_num_integers == 0) {
1311 (void) printf("\n");
1312 continue;
1313 }
1314 (void) printf(" %llx : [%d:%d:%d]\n",
1315 (u_longlong_t)attrp->za_first_integer,
1316 (int)ATTR_LENGTH(attrp->za_first_integer),
1317 (int)ATTR_BSWAP(attrp->za_first_integer),
1318 (int)ATTR_NUM(attrp->za_first_integer));
1319 }
1320 zap_cursor_fini(&zc);
1321 zap_attribute_free(attrp);
1322 }
1323
1324 static void
1325 dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size)
1326 {
1327 (void) data, (void) size;
1328 zap_cursor_t zc;
1329 zap_attribute_t *attrp = zap_attribute_alloc();
1330 uint16_t *layout_attrs;
1331 unsigned i;
1332
1333 dump_zap_stats(os, object);
1334 (void) printf("\n");
1335
1336 for (zap_cursor_init(&zc, os, object);
1337 zap_cursor_retrieve(&zc, attrp) == 0;
1338 zap_cursor_advance(&zc)) {
1339 (void) printf("\t\t%s = [", attrp->za_name);
1340 if (attrp->za_num_integers == 0) {
1341 (void) printf("\n");
1342 continue;
1343 }
1344
1345 VERIFY(attrp->za_integer_length == 2);
1346 layout_attrs = umem_zalloc(attrp->za_num_integers *
1347 attrp->za_integer_length, UMEM_NOFAIL);
1348
1349 VERIFY(zap_lookup(os, object, attrp->za_name,
1350 attrp->za_integer_length,
1351 attrp->za_num_integers, layout_attrs) == 0);
1352
1353 for (i = 0; i != attrp->za_num_integers; i++)
1354 (void) printf(" %d ", (int)layout_attrs[i]);
1355 (void) printf("]\n");
1356 umem_free(layout_attrs,
1357 attrp->za_num_integers * attrp->za_integer_length);
1358 }
1359 zap_cursor_fini(&zc);
1360 zap_attribute_free(attrp);
1361 }
1362
1363 static void
1364 dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
1365 {
1366 (void) data, (void) size;
1367 zap_cursor_t zc;
1368 zap_attribute_t *attrp = zap_attribute_long_alloc();
1369 const char *typenames[] = {
1370 /* 0 */ "not specified",
1371 /* 1 */ "FIFO",
1372 /* 2 */ "Character Device",
1373 /* 3 */ "3 (invalid)",
1374 /* 4 */ "Directory",
1375 /* 5 */ "5 (invalid)",
1376 /* 6 */ "Block Device",
1377 /* 7 */ "7 (invalid)",
1378 /* 8 */ "Regular File",
1379 /* 9 */ "9 (invalid)",
1380 /* 10 */ "Symbolic Link",
1381 /* 11 */ "11 (invalid)",
1382 /* 12 */ "Socket",
1383 /* 13 */ "Door",
1384 /* 14 */ "Event Port",
1385 /* 15 */ "15 (invalid)",
1386 };
1387
1388 dump_zap_stats(os, object);
1389 (void) printf("\n");
1390
1391 for (zap_cursor_init(&zc, os, object);
1392 zap_cursor_retrieve(&zc, attrp) == 0;
1393 zap_cursor_advance(&zc)) {
1394 (void) printf("\t\t%s = %lld (type: %s)\n",
1395 attrp->za_name, ZFS_DIRENT_OBJ(attrp->za_first_integer),
1396 typenames[ZFS_DIRENT_TYPE(attrp->za_first_integer)]);
1397 }
1398 zap_cursor_fini(&zc);
1399 zap_attribute_free(attrp);
1400 }
1401
1402 static int
1403 get_dtl_refcount(vdev_t *vd)
1404 {
1405 int refcount = 0;
1406
1407 if (vd->vdev_ops->vdev_op_leaf) {
1408 space_map_t *sm = vd->vdev_dtl_sm;
1409
1410 if (sm != NULL &&
1411 sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1412 return (1);
1413 return (0);
1414 }
1415
1416 for (unsigned c = 0; c < vd->vdev_children; c++)
1417 refcount += get_dtl_refcount(vd->vdev_child[c]);
1418 return (refcount);
1419 }
1420
1421 static int
1422 get_metaslab_refcount(vdev_t *vd)
1423 {
1424 int refcount = 0;
1425
1426 if (vd->vdev_top == vd) {
1427 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
1428 space_map_t *sm = vd->vdev_ms[m]->ms_sm;
1429
1430 if (sm != NULL &&
1431 sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1432 refcount++;
1433 }
1434 }
1435 for (unsigned c = 0; c < vd->vdev_children; c++)
1436 refcount += get_metaslab_refcount(vd->vdev_child[c]);
1437
1438 return (refcount);
1439 }
1440
1441 static int
1442 get_obsolete_refcount(vdev_t *vd)
1443 {
1444 uint64_t obsolete_sm_object;
1445 int refcount = 0;
1446
1447 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1448 if (vd->vdev_top == vd && obsolete_sm_object != 0) {
1449 dmu_object_info_t doi;
1450 VERIFY0(dmu_object_info(vd->vdev_spa->spa_meta_objset,
1451 obsolete_sm_object, &doi));
1452 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1453 refcount++;
1454 }
1455 } else {
1456 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
1457 ASSERT3U(obsolete_sm_object, ==, 0);
1458 }
1459 for (unsigned c = 0; c < vd->vdev_children; c++) {
1460 refcount += get_obsolete_refcount(vd->vdev_child[c]);
1461 }
1462
1463 return (refcount);
1464 }
1465
1466 static int
1467 get_prev_obsolete_spacemap_refcount(spa_t *spa)
1468 {
1469 uint64_t prev_obj =
1470 spa->spa_condensing_indirect_phys.scip_prev_obsolete_sm_object;
1471 if (prev_obj != 0) {
1472 dmu_object_info_t doi;
1473 VERIFY0(dmu_object_info(spa->spa_meta_objset, prev_obj, &doi));
1474 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1475 return (1);
1476 }
1477 }
1478 return (0);
1479 }
1480
1481 static int
1482 get_checkpoint_refcount(vdev_t *vd)
1483 {
1484 int refcount = 0;
1485
1486 if (vd->vdev_top == vd && vd->vdev_top_zap != 0 &&
1487 zap_contains(spa_meta_objset(vd->vdev_spa),
1488 vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) == 0)
1489 refcount++;
1490
1491 for (uint64_t c = 0; c < vd->vdev_children; c++)
1492 refcount += get_checkpoint_refcount(vd->vdev_child[c]);
1493
1494 return (refcount);
1495 }
1496
1497 static int
1498 get_log_spacemap_refcount(spa_t *spa)
1499 {
1500 return (avl_numnodes(&spa->spa_sm_logs_by_txg));
1501 }
1502
1503 static int
1504 verify_spacemap_refcounts(spa_t *spa)
1505 {
1506 uint64_t expected_refcount = 0;
1507 uint64_t actual_refcount;
1508
1509 (void) feature_get_refcount(spa,
1510 &spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM],
1511 &expected_refcount);
1512 actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
1513 actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);
1514 actual_refcount += get_obsolete_refcount(spa->spa_root_vdev);
1515 actual_refcount += get_prev_obsolete_spacemap_refcount(spa);
1516 actual_refcount += get_checkpoint_refcount(spa->spa_root_vdev);
1517 actual_refcount += get_log_spacemap_refcount(spa);
1518
1519 if (expected_refcount != actual_refcount) {
1520 (void) printf("space map refcount mismatch: expected %lld != "
1521 "actual %lld\n",
1522 (longlong_t)expected_refcount,
1523 (longlong_t)actual_refcount);
1524 return (2);
1525 }
1526 return (0);
1527 }
1528
1529 static void
1530 dump_spacemap(objset_t *os, space_map_t *sm)
1531 {
1532 const char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1533 "INVALID", "INVALID", "INVALID", "INVALID" };
1534
1535 if (sm == NULL)
1536 return;
1537
1538 (void) printf("space map object %llu:\n",
1539 (longlong_t)sm->sm_object);
1540 (void) printf(" smp_length = 0x%llx\n",
1541 (longlong_t)sm->sm_phys->smp_length);
1542 (void) printf(" smp_alloc = 0x%llx\n",
1543 (longlong_t)sm->sm_phys->smp_alloc);
1544
1545 if (dump_opt['d'] < 6 && dump_opt['m'] < 4)
1546 return;
1547
1548 /*
1549 * Print out the freelist entries in both encoded and decoded form.
1550 */
1551 uint8_t mapshift = sm->sm_shift;
1552 int64_t alloc = 0;
1553 uint64_t word, entry_id = 0;
1554 for (uint64_t offset = 0; offset < space_map_length(sm);
1555 offset += sizeof (word)) {
1556
1557 VERIFY0(dmu_read(os, space_map_object(sm), offset,
1558 sizeof (word), &word, DMU_READ_PREFETCH));
1559
1560 if (sm_entry_is_debug(word)) {
1561 uint64_t de_txg = SM_DEBUG_TXG_DECODE(word);
1562 uint64_t de_sync_pass = SM_DEBUG_SYNCPASS_DECODE(word);
1563 if (de_txg == 0) {
1564 (void) printf(
1565 "\t [%6llu] PADDING\n",
1566 (u_longlong_t)entry_id);
1567 } else {
1568 (void) printf(
1569 "\t [%6llu] %s: txg %llu pass %llu\n",
1570 (u_longlong_t)entry_id,
1571 ddata[SM_DEBUG_ACTION_DECODE(word)],
1572 (u_longlong_t)de_txg,
1573 (u_longlong_t)de_sync_pass);
1574 }
1575 entry_id++;
1576 continue;
1577 }
1578
1579 uint8_t words;
1580 char entry_type;
1581 uint64_t entry_off, entry_run, entry_vdev = SM_NO_VDEVID;
1582
1583 if (sm_entry_is_single_word(word)) {
1584 entry_type = (SM_TYPE_DECODE(word) == SM_ALLOC) ?
1585 'A' : 'F';
1586 entry_off = (SM_OFFSET_DECODE(word) << mapshift) +
1587 sm->sm_start;
1588 entry_run = SM_RUN_DECODE(word) << mapshift;
1589 words = 1;
1590 } else {
1591 /* it is a two-word entry so we read another word */
1592 ASSERT(sm_entry_is_double_word(word));
1593
1594 uint64_t extra_word;
1595 offset += sizeof (extra_word);
1596 VERIFY0(dmu_read(os, space_map_object(sm), offset,
1597 sizeof (extra_word), &extra_word,
1598 DMU_READ_PREFETCH));
1599
1600 ASSERT3U(offset, <=, space_map_length(sm));
1601
1602 entry_run = SM2_RUN_DECODE(word) << mapshift;
1603 entry_vdev = SM2_VDEV_DECODE(word);
1604 entry_type = (SM2_TYPE_DECODE(extra_word) == SM_ALLOC) ?
1605 'A' : 'F';
1606 entry_off = (SM2_OFFSET_DECODE(extra_word) <<
1607 mapshift) + sm->sm_start;
1608 words = 2;
1609 }
1610
1611 (void) printf("\t [%6llu] %c range:"
1612 " %010llx-%010llx size: %06llx vdev: %06llu words: %u\n",
1613 (u_longlong_t)entry_id,
1614 entry_type, (u_longlong_t)entry_off,
1615 (u_longlong_t)(entry_off + entry_run),
1616 (u_longlong_t)entry_run,
1617 (u_longlong_t)entry_vdev, words);
1618
1619 if (entry_type == 'A')
1620 alloc += entry_run;
1621 else
1622 alloc -= entry_run;
1623 entry_id++;
1624 }
1625 if (alloc != space_map_allocated(sm)) {
1626 (void) printf("space_map_object alloc (%lld) INCONSISTENT "
1627 "with space map summary (%lld)\n",
1628 (longlong_t)space_map_allocated(sm), (longlong_t)alloc);
1629 }
1630 }
1631
1632 static void
1633 dump_metaslab_stats(metaslab_t *msp)
1634 {
1635 char maxbuf[32];
1636 range_tree_t *rt = msp->ms_allocatable;
1637 zfs_btree_t *t = &msp->ms_allocatable_by_size;
1638 int free_pct = range_tree_space(rt) * 100 / msp->ms_size;
1639
1640 /* max sure nicenum has enough space */
1641 _Static_assert(sizeof (maxbuf) >= NN_NUMBUF_SZ, "maxbuf truncated");
1642
1643 zdb_nicenum(metaslab_largest_allocatable(msp), maxbuf, sizeof (maxbuf));
1644
1645 (void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
1646 "segments", zfs_btree_numnodes(t), "maxsize", maxbuf,
1647 "freepct", free_pct);
1648 (void) printf("\tIn-memory histogram:\n");
1649 dump_histogram(rt->rt_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1650 }
1651
1652 static void
1653 dump_metaslab(metaslab_t *msp)
1654 {
1655 vdev_t *vd = msp->ms_group->mg_vd;
1656 spa_t *spa = vd->vdev_spa;
1657 space_map_t *sm = msp->ms_sm;
1658 char freebuf[32];
1659
1660 zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf,
1661 sizeof (freebuf));
1662
1663 (void) printf(
1664 "\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
1665 (u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
1666 (u_longlong_t)space_map_object(sm), freebuf);
1667
1668 if (dump_opt['m'] > 2 && !dump_opt['L']) {
1669 mutex_enter(&msp->ms_lock);
1670 VERIFY0(metaslab_load(msp));
1671 range_tree_stat_verify(msp->ms_allocatable);
1672 dump_metaslab_stats(msp);
1673 metaslab_unload(msp);
1674 mutex_exit(&msp->ms_lock);
1675 }
1676
1677 if (dump_opt['m'] > 1 && sm != NULL &&
1678 spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) {
1679 /*
1680 * The space map histogram represents free space in chunks
1681 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1682 */
1683 (void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1684 (u_longlong_t)msp->ms_fragmentation);
1685 dump_histogram(sm->sm_phys->smp_histogram,
1686 SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift);
1687 }
1688
1689 if (vd->vdev_ops == &vdev_draid_ops)
1690 ASSERT3U(msp->ms_size, <=, 1ULL << vd->vdev_ms_shift);
1691 else
1692 ASSERT3U(msp->ms_size, ==, 1ULL << vd->vdev_ms_shift);
1693
1694 dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
1695
1696 if (spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
1697 (void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1698 (u_longlong_t)metaslab_unflushed_txg(msp));
1699 }
1700 }
1701
1702 static void
1703 print_vdev_metaslab_header(vdev_t *vd)
1704 {
1705 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
1706 const char *bias_str = "";
1707 if (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) {
1708 bias_str = VDEV_ALLOC_BIAS_LOG;
1709 } else if (alloc_bias == VDEV_BIAS_SPECIAL) {
1710 bias_str = VDEV_ALLOC_BIAS_SPECIAL;
1711 } else if (alloc_bias == VDEV_BIAS_DEDUP) {
1712 bias_str = VDEV_ALLOC_BIAS_DEDUP;
1713 }
1714
1715 uint64_t ms_flush_data_obj = 0;
1716 if (vd->vdev_top_zap != 0) {
1717 int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
1718 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
1719 sizeof (uint64_t), 1, &ms_flush_data_obj);
1720 if (error != ENOENT) {
1721 ASSERT0(error);
1722 }
1723 }
1724
1725 (void) printf("\tvdev %10llu %s",
1726 (u_longlong_t)vd->vdev_id, bias_str);
1727
1728 if (ms_flush_data_obj != 0) {
1729 (void) printf(" ms_unflushed_phys object %llu",
1730 (u_longlong_t)ms_flush_data_obj);
1731 }
1732
1733 (void) printf("\n\t%-10s%5llu %-19s %-15s %-12s\n",
1734 "metaslabs", (u_longlong_t)vd->vdev_ms_count,
1735 "offset", "spacemap", "free");
1736 (void) printf("\t%15s %19s %15s %12s\n",
1737 "---------------", "-------------------",
1738 "---------------", "------------");
1739 }
1740
1741 static void
1742 dump_metaslab_groups(spa_t *spa, boolean_t show_special)
1743 {
1744 vdev_t *rvd = spa->spa_root_vdev;
1745 metaslab_class_t *mc = spa_normal_class(spa);
1746 metaslab_class_t *smc = spa_special_class(spa);
1747 uint64_t fragmentation;
1748
1749 metaslab_class_histogram_verify(mc);
1750
1751 for (unsigned c = 0; c < rvd->vdev_children; c++) {
1752 vdev_t *tvd = rvd->vdev_child[c];
1753 metaslab_group_t *mg = tvd->vdev_mg;
1754
1755 if (mg == NULL || (mg->mg_class != mc &&
1756 (!show_special || mg->mg_class != smc)))
1757 continue;
1758
1759 metaslab_group_histogram_verify(mg);
1760 mg->mg_fragmentation = metaslab_group_fragmentation(mg);
1761
1762 (void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1763 "fragmentation",
1764 (u_longlong_t)tvd->vdev_id,
1765 (u_longlong_t)tvd->vdev_ms_count);
1766 if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
1767 (void) printf("%3s\n", "-");
1768 } else {
1769 (void) printf("%3llu%%\n",
1770 (u_longlong_t)mg->mg_fragmentation);
1771 }
1772 dump_histogram(mg->mg_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1773 }
1774
1775 (void) printf("\tpool %s\tfragmentation", spa_name(spa));
1776 fragmentation = metaslab_class_fragmentation(mc);
1777 if (fragmentation == ZFS_FRAG_INVALID)
1778 (void) printf("\t%3s\n", "-");
1779 else
1780 (void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation);
1781 dump_histogram(mc->mc_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1782 }
1783
1784 static void
1785 print_vdev_indirect(vdev_t *vd)
1786 {
1787 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
1788 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
1789 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
1790
1791 if (vim == NULL) {
1792 ASSERT3P(vib, ==, NULL);
1793 return;
1794 }
1795
1796 ASSERT3U(vdev_indirect_mapping_object(vim), ==,
1797 vic->vic_mapping_object);
1798 ASSERT3U(vdev_indirect_births_object(vib), ==,
1799 vic->vic_births_object);
1800
1801 (void) printf("indirect births obj %llu:\n",
1802 (longlong_t)vic->vic_births_object);
1803 (void) printf(" vib_count = %llu\n",
1804 (longlong_t)vdev_indirect_births_count(vib));
1805 for (uint64_t i = 0; i < vdev_indirect_births_count(vib); i++) {
1806 vdev_indirect_birth_entry_phys_t *cur_vibe =
1807 &vib->vib_entries[i];
1808 (void) printf("\toffset %llx -> txg %llu\n",
1809 (longlong_t)cur_vibe->vibe_offset,
1810 (longlong_t)cur_vibe->vibe_phys_birth_txg);
1811 }
1812 (void) printf("\n");
1813
1814 (void) printf("indirect mapping obj %llu:\n",
1815 (longlong_t)vic->vic_mapping_object);
1816 (void) printf(" vim_max_offset = 0x%llx\n",
1817 (longlong_t)vdev_indirect_mapping_max_offset(vim));
1818 (void) printf(" vim_bytes_mapped = 0x%llx\n",
1819 (longlong_t)vdev_indirect_mapping_bytes_mapped(vim));
1820 (void) printf(" vim_count = %llu\n",
1821 (longlong_t)vdev_indirect_mapping_num_entries(vim));
1822
1823 if (dump_opt['d'] <= 5 && dump_opt['m'] <= 3)
1824 return;
1825
1826 uint32_t *counts = vdev_indirect_mapping_load_obsolete_counts(vim);
1827
1828 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
1829 vdev_indirect_mapping_entry_phys_t *vimep =
1830 &vim->vim_entries[i];
1831 (void) printf("\t<%llx:%llx:%llx> -> "
1832 "<%llx:%llx:%llx> (%x obsolete)\n",
1833 (longlong_t)vd->vdev_id,
1834 (longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
1835 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1836 (longlong_t)DVA_GET_VDEV(&vimep->vimep_dst),
1837 (longlong_t)DVA_GET_OFFSET(&vimep->vimep_dst),
1838 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1839 counts[i]);
1840 }
1841 (void) printf("\n");
1842
1843 uint64_t obsolete_sm_object;
1844 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1845 if (obsolete_sm_object != 0) {
1846 objset_t *mos = vd->vdev_spa->spa_meta_objset;
1847 (void) printf("obsolete space map object %llu:\n",
1848 (u_longlong_t)obsolete_sm_object);
1849 ASSERT(vd->vdev_obsolete_sm != NULL);
1850 ASSERT3U(space_map_object(vd->vdev_obsolete_sm), ==,
1851 obsolete_sm_object);
1852 dump_spacemap(mos, vd->vdev_obsolete_sm);
1853 (void) printf("\n");
1854 }
1855 }
1856
1857 static void
1858 dump_metaslabs(spa_t *spa)
1859 {
1860 vdev_t *vd, *rvd = spa->spa_root_vdev;
1861 uint64_t m, c = 0, children = rvd->vdev_children;
1862
1863 (void) printf("\nMetaslabs:\n");
1864
1865 if (!dump_opt['d'] && zopt_metaslab_args > 0) {
1866 c = zopt_metaslab[0];
1867
1868 if (c >= children)
1869 (void) fatal("bad vdev id: %llu", (u_longlong_t)c);
1870
1871 if (zopt_metaslab_args > 1) {
1872 vd = rvd->vdev_child[c];
1873 print_vdev_metaslab_header(vd);
1874
1875 for (m = 1; m < zopt_metaslab_args; m++) {
1876 if (zopt_metaslab[m] < vd->vdev_ms_count)
1877 dump_metaslab(
1878 vd->vdev_ms[zopt_metaslab[m]]);
1879 else
1880 (void) fprintf(stderr, "bad metaslab "
1881 "number %llu\n",
1882 (u_longlong_t)zopt_metaslab[m]);
1883 }
1884 (void) printf("\n");
1885 return;
1886 }
1887 children = c + 1;
1888 }
1889 for (; c < children; c++) {
1890 vd = rvd->vdev_child[c];
1891 print_vdev_metaslab_header(vd);
1892
1893 print_vdev_indirect(vd);
1894
1895 for (m = 0; m < vd->vdev_ms_count; m++)
1896 dump_metaslab(vd->vdev_ms[m]);
1897 (void) printf("\n");
1898 }
1899 }
1900
1901 static void
1902 dump_log_spacemaps(spa_t *spa)
1903 {
1904 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1905 return;
1906
1907 (void) printf("\nLog Space Maps in Pool:\n");
1908 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
1909 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1910 space_map_t *sm = NULL;
1911 VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
1912 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
1913
1914 (void) printf("Log Spacemap object %llu txg %llu\n",
1915 (u_longlong_t)sls->sls_sm_obj, (u_longlong_t)sls->sls_txg);
1916 dump_spacemap(spa->spa_meta_objset, sm);
1917 space_map_close(sm);
1918 }
1919 (void) printf("\n");
1920 }
1921
1922 static void
1923 dump_ddt_entry(const ddt_t *ddt, const ddt_lightweight_entry_t *ddlwe,
1924 uint64_t index)
1925 {
1926 const ddt_key_t *ddk = &ddlwe->ddlwe_key;
1927 char blkbuf[BP_SPRINTF_LEN];
1928 blkptr_t blk;
1929 int p;
1930
1931 for (p = 0; p < DDT_NPHYS(ddt); p++) {
1932 const ddt_univ_phys_t *ddp = &ddlwe->ddlwe_phys;
1933 ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p);
1934
1935 if (ddt_phys_birth(ddp, v) == 0)
1936 continue;
1937 ddt_bp_create(ddt->ddt_checksum, ddk, ddp, v, &blk);
1938 snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
1939 (void) printf("index %llx refcnt %llu phys %d %s\n",
1940 (u_longlong_t)index, (u_longlong_t)ddt_phys_refcnt(ddp, v),
1941 p, blkbuf);
1942 }
1943 }
1944
1945 static void
1946 dump_dedup_ratio(const ddt_stat_t *dds)
1947 {
1948 double rL, rP, rD, D, dedup, compress, copies;
1949
1950 if (dds->dds_blocks == 0)
1951 return;
1952
1953 rL = (double)dds->dds_ref_lsize;
1954 rP = (double)dds->dds_ref_psize;
1955 rD = (double)dds->dds_ref_dsize;
1956 D = (double)dds->dds_dsize;
1957
1958 dedup = rD / D;
1959 compress = rL / rP;
1960 copies = rD / rP;
1961
1962 (void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
1963 "dedup * compress / copies = %.2f\n\n",
1964 dedup, compress, copies, dedup * compress / copies);
1965 }
1966
1967 static void
1968 dump_ddt_log(ddt_t *ddt)
1969 {
1970 for (int n = 0; n < 2; n++) {
1971 ddt_log_t *ddl = &ddt->ddt_log[n];
1972
1973 uint64_t count = avl_numnodes(&ddl->ddl_tree);
1974 if (count == 0)
1975 continue;
1976
1977 printf(DMU_POOL_DDT_LOG ": %lu log entries\n",
1978 zio_checksum_table[ddt->ddt_checksum].ci_name, n, count);
1979
1980 if (dump_opt['D'] < 4)
1981 continue;
1982
1983 ddt_lightweight_entry_t ddlwe;
1984 uint64_t index = 0;
1985 for (ddt_log_entry_t *ddle = avl_first(&ddl->ddl_tree);
1986 ddle; ddle = AVL_NEXT(&ddl->ddl_tree, ddle)) {
1987 DDT_LOG_ENTRY_TO_LIGHTWEIGHT(ddt, ddle, &ddlwe);
1988 dump_ddt_entry(ddt, &ddlwe, index++);
1989 }
1990 }
1991 }
1992
1993 static void
1994 dump_ddt(ddt_t *ddt, ddt_type_t type, ddt_class_t class)
1995 {
1996 char name[DDT_NAMELEN];
1997 ddt_lightweight_entry_t ddlwe;
1998 uint64_t walk = 0;
1999 dmu_object_info_t doi;
2000 uint64_t count, dspace, mspace;
2001 int error;
2002
2003 error = ddt_object_info(ddt, type, class, &doi);
2004
2005 if (error == ENOENT)
2006 return;
2007 ASSERT(error == 0);
2008
2009 error = ddt_object_count(ddt, type, class, &count);
2010 ASSERT(error == 0);
2011 if (count == 0)
2012 return;
2013
2014 dspace = doi.doi_physical_blocks_512 << 9;
2015 mspace = doi.doi_fill_count * doi.doi_data_block_size;
2016
2017 ddt_object_name(ddt, type, class, name);
2018
2019 (void) printf("%s: %llu entries, size %llu on disk, %llu in core\n",
2020 name,
2021 (u_longlong_t)count,
2022 (u_longlong_t)dspace,
2023 (u_longlong_t)mspace);
2024
2025 if (dump_opt['D'] < 3)
2026 return;
2027
2028 zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class]);
2029
2030 if (dump_opt['D'] < 4)
2031 return;
2032
2033 if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE)
2034 return;
2035
2036 (void) printf("%s contents:\n\n", name);
2037
2038 while ((error = ddt_object_walk(ddt, type, class, &walk, &ddlwe)) == 0)
2039 dump_ddt_entry(ddt, &ddlwe, walk);
2040
2041 ASSERT3U(error, ==, ENOENT);
2042
2043 (void) printf("\n");
2044 }
2045
2046 static void
2047 dump_all_ddts(spa_t *spa)
2048 {
2049 ddt_histogram_t ddh_total = {{{0}}};
2050 ddt_stat_t dds_total = {0};
2051
2052 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
2053 ddt_t *ddt = spa->spa_ddt[c];
2054 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
2055 continue;
2056 for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
2057 for (ddt_class_t class = 0; class < DDT_CLASSES;
2058 class++) {
2059 dump_ddt(ddt, type, class);
2060 }
2061 }
2062 dump_ddt_log(ddt);
2063 }
2064
2065 ddt_get_dedup_stats(spa, &dds_total);
2066
2067 if (dds_total.dds_blocks == 0) {
2068 (void) printf("All DDTs are empty\n");
2069 return;
2070 }
2071
2072 (void) printf("\n");
2073
2074 if (dump_opt['D'] > 1) {
2075 (void) printf("DDT histogram (aggregated over all DDTs):\n");
2076 ddt_get_dedup_histogram(spa, &ddh_total);
2077 zpool_dump_ddt(&dds_total, &ddh_total);
2078 }
2079
2080 dump_dedup_ratio(&dds_total);
2081
2082 /*
2083 * Dump a histogram of unique class entry age
2084 */
2085 if (dump_opt['D'] == 3 && getenv("ZDB_DDT_UNIQUE_AGE_HIST") != NULL) {
2086 ddt_age_histo_t histogram;
2087
2088 (void) printf("DDT walk unique, building age histogram...\n");
2089 ddt_prune_walk(spa, 0, &histogram);
2090
2091 /*
2092 * print out histogram for unique entry class birth
2093 */
2094 if (histogram.dah_entries > 0) {
2095 (void) printf("%5s %9s %4s\n",
2096 "age", "blocks", "amnt");
2097 (void) printf("%5s %9s %4s\n",
2098 "-----", "---------", "----");
2099 for (int i = 0; i < HIST_BINS; i++) {
2100 (void) printf("%5d %9d %4d%%\n", 1 << i,
2101 (int)histogram.dah_age_histo[i],
2102 (int)((histogram.dah_age_histo[i] * 100) /
2103 histogram.dah_entries));
2104 }
2105 }
2106 }
2107 }
2108
2109 static void
2110 dump_brt(spa_t *spa)
2111 {
2112 if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) {
2113 printf("BRT: unsupported on this pool\n");
2114 return;
2115 }
2116
2117 if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
2118 printf("BRT: empty\n");
2119 return;
2120 }
2121
2122 brt_t *brt = spa->spa_brt;
2123 VERIFY(brt);
2124
2125 char count[32], used[32], saved[32];
2126 zdb_nicebytes(brt_get_used(spa), used, sizeof (used));
2127 zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved));
2128 uint64_t ratio = brt_get_ratio(spa);
2129 printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved,
2130 (u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100));
2131
2132 if (dump_opt['T'] < 2)
2133 return;
2134
2135 for (uint64_t vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) {
2136 brt_vdev_t *brtvd = &brt->brt_vdevs[vdevid];
2137 if (brtvd == NULL)
2138 continue;
2139
2140 if (!brtvd->bv_initiated) {
2141 printf("BRT: vdev %" PRIu64 ": empty\n", vdevid);
2142 continue;
2143 }
2144
2145 zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count));
2146 zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used));
2147 zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved));
2148 printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n",
2149 vdevid, count, used, saved);
2150 }
2151
2152 if (dump_opt['T'] < 3)
2153 return;
2154
2155 char dva[64];
2156 printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2157
2158 for (uint64_t vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) {
2159 brt_vdev_t *brtvd = &brt->brt_vdevs[vdevid];
2160 if (brtvd == NULL || !brtvd->bv_initiated)
2161 continue;
2162
2163 zap_cursor_t zc;
2164 zap_attribute_t *za = zap_attribute_alloc();
2165 for (zap_cursor_init(&zc, brt->brt_mos, brtvd->bv_mos_entries);
2166 zap_cursor_retrieve(&zc, za) == 0;
2167 zap_cursor_advance(&zc)) {
2168 uint64_t refcnt;
2169 VERIFY0(zap_lookup_uint64(brt->brt_mos,
2170 brtvd->bv_mos_entries,
2171 (const uint64_t *)za->za_name, 1,
2172 za->za_integer_length, za->za_num_integers,
2173 &refcnt));
2174
2175 uint64_t offset = *(const uint64_t *)za->za_name;
2176
2177 snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx", vdevid,
2178 (u_longlong_t)offset);
2179 printf("%-16s %-10llu\n", dva, (u_longlong_t)refcnt);
2180 }
2181 zap_cursor_fini(&zc);
2182 zap_attribute_free(za);
2183 }
2184 }
2185
2186 static void
2187 dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
2188 {
2189 char *prefix = arg;
2190
2191 (void) printf("%s [%llu,%llu) length %llu\n",
2192 prefix,
2193 (u_longlong_t)start,
2194 (u_longlong_t)(start + size),
2195 (u_longlong_t)(size));
2196 }
2197
2198 static void
2199 dump_dtl(vdev_t *vd, int indent)
2200 {
2201 spa_t *spa = vd->vdev_spa;
2202 boolean_t required;
2203 const char *name[DTL_TYPES] = { "missing", "partial", "scrub",
2204 "outage" };
2205 char prefix[256];
2206
2207 spa_vdev_state_enter(spa, SCL_NONE);
2208 required = vdev_dtl_required(vd);
2209 (void) spa_vdev_state_exit(spa, NULL, 0);
2210
2211 if (indent == 0)
2212 (void) printf("\nDirty time logs:\n\n");
2213
2214 (void) printf("\t%*s%s [%s]\n", indent, "",
2215 vd->vdev_path ? vd->vdev_path :
2216 vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa),
2217 required ? "DTL-required" : "DTL-expendable");
2218
2219 for (int t = 0; t < DTL_TYPES; t++) {
2220 range_tree_t *rt = vd->vdev_dtl[t];
2221 if (range_tree_space(rt) == 0)
2222 continue;
2223 (void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
2224 indent + 2, "", name[t]);
2225 range_tree_walk(rt, dump_dtl_seg, prefix);
2226 if (dump_opt['d'] > 5 && vd->vdev_children == 0)
2227 dump_spacemap(spa->spa_meta_objset,
2228 vd->vdev_dtl_sm);
2229 }
2230
2231 for (unsigned c = 0; c < vd->vdev_children; c++)
2232 dump_dtl(vd->vdev_child[c], indent + 4);
2233 }
2234
2235 static void
2236 dump_history(spa_t *spa)
2237 {
2238 nvlist_t **events = NULL;
2239 char *buf;
2240 uint64_t resid, len, off = 0;
2241 uint_t num = 0;
2242 int error;
2243 char tbuf[30];
2244
2245 if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) {
2246 (void) fprintf(stderr, "%s: unable to allocate I/O buffer\n",
2247 __func__);
2248 return;
2249 }
2250
2251 do {
2252 len = SPA_OLD_MAXBLOCKSIZE;
2253
2254 if ((error = spa_history_get(spa, &off, &len, buf)) != 0) {
2255 (void) fprintf(stderr, "Unable to read history: "
2256 "error %d\n", error);
2257 free(buf);
2258 return;
2259 }
2260
2261 if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0)
2262 break;
2263
2264 off -= resid;
2265 } while (len != 0);
2266
2267 (void) printf("\nHistory:\n");
2268 for (unsigned i = 0; i < num; i++) {
2269 boolean_t printed = B_FALSE;
2270
2271 if (nvlist_exists(events[i], ZPOOL_HIST_TIME)) {
2272 time_t tsec;
2273 struct tm t;
2274
2275 tsec = fnvlist_lookup_uint64(events[i],
2276 ZPOOL_HIST_TIME);
2277 (void) localtime_r(&tsec, &t);
2278 (void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t);
2279 } else {
2280 tbuf[0] = '\0';
2281 }
2282
2283 if (nvlist_exists(events[i], ZPOOL_HIST_CMD)) {
2284 (void) printf("%s %s\n", tbuf,
2285 fnvlist_lookup_string(events[i], ZPOOL_HIST_CMD));
2286 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_EVENT)) {
2287 uint64_t ievent;
2288
2289 ievent = fnvlist_lookup_uint64(events[i],
2290 ZPOOL_HIST_INT_EVENT);
2291 if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS)
2292 goto next;
2293
2294 (void) printf(" %s [internal %s txg:%ju] %s\n",
2295 tbuf,
2296 zfs_history_event_names[ievent],
2297 fnvlist_lookup_uint64(events[i],
2298 ZPOOL_HIST_TXG),
2299 fnvlist_lookup_string(events[i],
2300 ZPOOL_HIST_INT_STR));
2301 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_NAME)) {
2302 (void) printf("%s [txg:%ju] %s", tbuf,
2303 fnvlist_lookup_uint64(events[i],
2304 ZPOOL_HIST_TXG),
2305 fnvlist_lookup_string(events[i],
2306 ZPOOL_HIST_INT_NAME));
2307
2308 if (nvlist_exists(events[i], ZPOOL_HIST_DSNAME)) {
2309 (void) printf(" %s (%llu)",
2310 fnvlist_lookup_string(events[i],
2311 ZPOOL_HIST_DSNAME),
2312 (u_longlong_t)fnvlist_lookup_uint64(
2313 events[i],
2314 ZPOOL_HIST_DSID));
2315 }
2316
2317 (void) printf(" %s\n", fnvlist_lookup_string(events[i],
2318 ZPOOL_HIST_INT_STR));
2319 } else if (nvlist_exists(events[i], ZPOOL_HIST_IOCTL)) {
2320 (void) printf("%s ioctl %s\n", tbuf,
2321 fnvlist_lookup_string(events[i],
2322 ZPOOL_HIST_IOCTL));
2323
2324 if (nvlist_exists(events[i], ZPOOL_HIST_INPUT_NVL)) {
2325 (void) printf(" input:\n");
2326 dump_nvlist(fnvlist_lookup_nvlist(events[i],
2327 ZPOOL_HIST_INPUT_NVL), 8);
2328 }
2329 if (nvlist_exists(events[i], ZPOOL_HIST_OUTPUT_NVL)) {
2330 (void) printf(" output:\n");
2331 dump_nvlist(fnvlist_lookup_nvlist(events[i],
2332 ZPOOL_HIST_OUTPUT_NVL), 8);
2333 }
2334 if (nvlist_exists(events[i], ZPOOL_HIST_ERRNO)) {
2335 (void) printf(" errno: %lld\n",
2336 (longlong_t)fnvlist_lookup_int64(events[i],
2337 ZPOOL_HIST_ERRNO));
2338 }
2339 } else {
2340 goto next;
2341 }
2342
2343 printed = B_TRUE;
2344 next:
2345 if (dump_opt['h'] > 1) {
2346 if (!printed)
2347 (void) printf("unrecognized record:\n");
2348 dump_nvlist(events[i], 2);
2349 }
2350 }
2351 free(buf);
2352 }
2353
2354 static void
2355 dump_dnode(objset_t *os, uint64_t object, void *data, size_t size)
2356 {
2357 (void) os, (void) object, (void) data, (void) size;
2358 }
2359
2360 static uint64_t
2361 blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
2362 const zbookmark_phys_t *zb)
2363 {
2364 if (dnp == NULL) {
2365 ASSERT(zb->zb_level < 0);
2366 if (zb->zb_object == 0)
2367 return (zb->zb_blkid);
2368 return (zb->zb_blkid * BP_GET_LSIZE(bp));
2369 }
2370
2371 ASSERT(zb->zb_level >= 0);
2372
2373 return ((zb->zb_blkid <<
2374 (zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) *
2375 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
2376 }
2377
2378 static void
2379 snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
2380 const blkptr_t *bp)
2381 {
2382 static abd_t *pabd = NULL;
2383 void *buf;
2384 zio_t *zio;
2385 zfs_zstdhdr_t zstd_hdr;
2386 int error;
2387
2388 if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD)
2389 return;
2390
2391 if (BP_IS_HOLE(bp))
2392 return;
2393
2394 if (BP_IS_EMBEDDED(bp)) {
2395 buf = malloc(SPA_MAXBLOCKSIZE);
2396 if (buf == NULL) {
2397 (void) fprintf(stderr, "out of memory\n");
2398 zdb_exit(1);
2399 }
2400 decode_embedded_bp_compressed(bp, buf);
2401 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2402 free(buf);
2403 zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2404 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2405 (void) snprintf(blkbuf + strlen(blkbuf),
2406 buflen - strlen(blkbuf),
2407 " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2408 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2409 zfs_get_hdrlevel(&zstd_hdr));
2410 return;
2411 }
2412
2413 if (!pabd)
2414 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
2415 zio = zio_root(spa, NULL, NULL, 0);
2416
2417 /* Decrypt but don't decompress so we can read the compression header */
2418 zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL,
2419 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS,
2420 NULL));
2421 error = zio_wait(zio);
2422 if (error) {
2423 (void) fprintf(stderr, "read failed: %d\n", error);
2424 return;
2425 }
2426 buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp));
2427 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2428 zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2429 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2430
2431 (void) snprintf(blkbuf + strlen(blkbuf),
2432 buflen - strlen(blkbuf),
2433 " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2434 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2435 zfs_get_hdrlevel(&zstd_hdr));
2436
2437 abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp));
2438 }
2439
2440 static void
2441 snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
2442 boolean_t bp_freed)
2443 {
2444 const dva_t *dva = bp->blk_dva;
2445 int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1;
2446 int i;
2447
2448 if (dump_opt['b'] >= 6) {
2449 snprintf_blkptr(blkbuf, buflen, bp);
2450 if (bp_freed) {
2451 (void) snprintf(blkbuf + strlen(blkbuf),
2452 buflen - strlen(blkbuf), " %s", "FREE");
2453 }
2454 return;
2455 }
2456
2457 if (BP_IS_EMBEDDED(bp)) {
2458 (void) sprintf(blkbuf,
2459 "EMBEDDED et=%u %llxL/%llxP B=%llu",
2460 (int)BPE_GET_ETYPE(bp),
2461 (u_longlong_t)BPE_GET_LSIZE(bp),
2462 (u_longlong_t)BPE_GET_PSIZE(bp),
2463 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2464 return;
2465 }
2466
2467 blkbuf[0] = '\0';
2468
2469 for (i = 0; i < ndvas; i++)
2470 (void) snprintf(blkbuf + strlen(blkbuf),
2471 buflen - strlen(blkbuf), "%llu:%llx:%llx ",
2472 (u_longlong_t)DVA_GET_VDEV(&dva[i]),
2473 (u_longlong_t)DVA_GET_OFFSET(&dva[i]),
2474 (u_longlong_t)DVA_GET_ASIZE(&dva[i]));
2475
2476 if (BP_IS_HOLE(bp)) {
2477 (void) snprintf(blkbuf + strlen(blkbuf),
2478 buflen - strlen(blkbuf),
2479 "%llxL B=%llu",
2480 (u_longlong_t)BP_GET_LSIZE(bp),
2481 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2482 } else {
2483 (void) snprintf(blkbuf + strlen(blkbuf),
2484 buflen - strlen(blkbuf),
2485 "%llxL/%llxP F=%llu B=%llu/%llu",
2486 (u_longlong_t)BP_GET_LSIZE(bp),
2487 (u_longlong_t)BP_GET_PSIZE(bp),
2488 (u_longlong_t)BP_GET_FILL(bp),
2489 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),
2490 (u_longlong_t)BP_GET_BIRTH(bp));
2491 if (bp_freed)
2492 (void) snprintf(blkbuf + strlen(blkbuf),
2493 buflen - strlen(blkbuf), " %s", "FREE");
2494 (void) snprintf(blkbuf + strlen(blkbuf),
2495 buflen - strlen(blkbuf),
2496 " cksum=%016llx:%016llx:%016llx:%016llx",
2497 (u_longlong_t)bp->blk_cksum.zc_word[0],
2498 (u_longlong_t)bp->blk_cksum.zc_word[1],
2499 (u_longlong_t)bp->blk_cksum.zc_word[2],
2500 (u_longlong_t)bp->blk_cksum.zc_word[3]);
2501 }
2502 }
2503
2504 static void
2505 print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
2506 const dnode_phys_t *dnp)
2507 {
2508 char blkbuf[BP_SPRINTF_LEN];
2509 int l;
2510
2511 if (!BP_IS_EMBEDDED(bp)) {
2512 ASSERT3U(BP_GET_TYPE(bp), ==, dnp->dn_type);
2513 ASSERT3U(BP_GET_LEVEL(bp), ==, zb->zb_level);
2514 }
2515
2516 (void) printf("%16llx ", (u_longlong_t)blkid2offset(dnp, bp, zb));
2517
2518 ASSERT(zb->zb_level >= 0);
2519
2520 for (l = dnp->dn_nlevels - 1; l >= -1; l--) {
2521 if (l == zb->zb_level) {
2522 (void) printf("L%llx", (u_longlong_t)zb->zb_level);
2523 } else {
2524 (void) printf(" ");
2525 }
2526 }
2527
2528 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE);
2529 if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD)
2530 snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp);
2531 (void) printf("%s\n", blkbuf);
2532 }
2533
2534 static int
2535 visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
2536 blkptr_t *bp, const zbookmark_phys_t *zb)
2537 {
2538 int err = 0;
2539
2540 if (BP_GET_LOGICAL_BIRTH(bp) == 0)
2541 return (0);
2542
2543 print_indirect(spa, bp, zb, dnp);
2544
2545 if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
2546 arc_flags_t flags = ARC_FLAG_WAIT;
2547 int i;
2548 blkptr_t *cbp;
2549 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
2550 arc_buf_t *buf;
2551 uint64_t fill = 0;
2552 ASSERT(!BP_IS_REDACTED(bp));
2553
2554 err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
2555 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
2556 if (err)
2557 return (err);
2558 ASSERT(buf->b_data);
2559
2560 /* recursively visit blocks below this */
2561 cbp = buf->b_data;
2562 for (i = 0; i < epb; i++, cbp++) {
2563 zbookmark_phys_t czb;
2564
2565 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
2566 zb->zb_level - 1,
2567 zb->zb_blkid * epb + i);
2568 err = visit_indirect(spa, dnp, cbp, &czb);
2569 if (err)
2570 break;
2571 fill += BP_GET_FILL(cbp);
2572 }
2573 if (!err)
2574 ASSERT3U(fill, ==, BP_GET_FILL(bp));
2575 arc_buf_destroy(buf, &buf);
2576 }
2577
2578 return (err);
2579 }
2580
2581 static void
2582 dump_indirect(dnode_t *dn)
2583 {
2584 dnode_phys_t *dnp = dn->dn_phys;
2585 zbookmark_phys_t czb;
2586
2587 (void) printf("Indirect blocks:\n");
2588
2589 SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
2590 dn->dn_object, dnp->dn_nlevels - 1, 0);
2591 for (int j = 0; j < dnp->dn_nblkptr; j++) {
2592 czb.zb_blkid = j;
2593 (void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
2594 &dnp->dn_blkptr[j], &czb);
2595 }
2596
2597 (void) printf("\n");
2598 }
2599
2600 static void
2601 dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size)
2602 {
2603 (void) os, (void) object;
2604 dsl_dir_phys_t *dd = data;
2605 time_t crtime;
2606 char nice[32];
2607
2608 /* make sure nicenum has enough space */
2609 _Static_assert(sizeof (nice) >= NN_NUMBUF_SZ, "nice truncated");
2610
2611 if (dd == NULL)
2612 return;
2613
2614 ASSERT3U(size, >=, sizeof (dsl_dir_phys_t));
2615
2616 crtime = dd->dd_creation_time;
2617 (void) printf("\t\tcreation_time = %s", ctime(&crtime));
2618 (void) printf("\t\thead_dataset_obj = %llu\n",
2619 (u_longlong_t)dd->dd_head_dataset_obj);
2620 (void) printf("\t\tparent_dir_obj = %llu\n",
2621 (u_longlong_t)dd->dd_parent_obj);
2622 (void) printf("\t\torigin_obj = %llu\n",
2623 (u_longlong_t)dd->dd_origin_obj);
2624 (void) printf("\t\tchild_dir_zapobj = %llu\n",
2625 (u_longlong_t)dd->dd_child_dir_zapobj);
2626 zdb_nicenum(dd->dd_used_bytes, nice, sizeof (nice));
2627 (void) printf("\t\tused_bytes = %s\n", nice);
2628 zdb_nicenum(dd->dd_compressed_bytes, nice, sizeof (nice));
2629 (void) printf("\t\tcompressed_bytes = %s\n", nice);
2630 zdb_nicenum(dd->dd_uncompressed_bytes, nice, sizeof (nice));
2631 (void) printf("\t\tuncompressed_bytes = %s\n", nice);
2632 zdb_nicenum(dd->dd_quota, nice, sizeof (nice));
2633 (void) printf("\t\tquota = %s\n", nice);
2634 zdb_nicenum(dd->dd_reserved, nice, sizeof (nice));
2635 (void) printf("\t\treserved = %s\n", nice);
2636 (void) printf("\t\tprops_zapobj = %llu\n",
2637 (u_longlong_t)dd->dd_props_zapobj);
2638 (void) printf("\t\tdeleg_zapobj = %llu\n",
2639 (u_longlong_t)dd->dd_deleg_zapobj);
2640 (void) printf("\t\tflags = %llx\n",
2641 (u_longlong_t)dd->dd_flags);
2642
2643 #define DO(which) \
2644 zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2645 sizeof (nice)); \
2646 (void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2647 DO(HEAD);
2648 DO(SNAP);
2649 DO(CHILD);
2650 DO(CHILD_RSRV);
2651 DO(REFRSRV);
2652 #undef DO
2653 (void) printf("\t\tclones = %llu\n",
2654 (u_longlong_t)dd->dd_clones);
2655 }
2656
2657 static void
2658 dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
2659 {
2660 (void) os, (void) object;
2661 dsl_dataset_phys_t *ds = data;
2662 time_t crtime;
2663 char used[32], compressed[32], uncompressed[32], unique[32];
2664 char blkbuf[BP_SPRINTF_LEN];
2665
2666 /* make sure nicenum has enough space */
2667 _Static_assert(sizeof (used) >= NN_NUMBUF_SZ, "used truncated");
2668 _Static_assert(sizeof (compressed) >= NN_NUMBUF_SZ,
2669 "compressed truncated");
2670 _Static_assert(sizeof (uncompressed) >= NN_NUMBUF_SZ,
2671 "uncompressed truncated");
2672 _Static_assert(sizeof (unique) >= NN_NUMBUF_SZ, "unique truncated");
2673
2674 if (ds == NULL)
2675 return;
2676
2677 ASSERT(size == sizeof (*ds));
2678 crtime = ds->ds_creation_time;
2679 zdb_nicenum(ds->ds_referenced_bytes, used, sizeof (used));
2680 zdb_nicenum(ds->ds_compressed_bytes, compressed, sizeof (compressed));
2681 zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed,
2682 sizeof (uncompressed));
2683 zdb_nicenum(ds->ds_unique_bytes, unique, sizeof (unique));
2684 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp);
2685
2686 (void) printf("\t\tdir_obj = %llu\n",
2687 (u_longlong_t)ds->ds_dir_obj);
2688 (void) printf("\t\tprev_snap_obj = %llu\n",
2689 (u_longlong_t)ds->ds_prev_snap_obj);
2690 (void) printf("\t\tprev_snap_txg = %llu\n",
2691 (u_longlong_t)ds->ds_prev_snap_txg);
2692 (void) printf("\t\tnext_snap_obj = %llu\n",
2693 (u_longlong_t)ds->ds_next_snap_obj);
2694 (void) printf("\t\tsnapnames_zapobj = %llu\n",
2695 (u_longlong_t)ds->ds_snapnames_zapobj);
2696 (void) printf("\t\tnum_children = %llu\n",
2697 (u_longlong_t)ds->ds_num_children);
2698 (void) printf("\t\tuserrefs_obj = %llu\n",
2699 (u_longlong_t)ds->ds_userrefs_obj);
2700 (void) printf("\t\tcreation_time = %s", ctime(&crtime));
2701 (void) printf("\t\tcreation_txg = %llu\n",
2702 (u_longlong_t)ds->ds_creation_txg);
2703 (void) printf("\t\tdeadlist_obj = %llu\n",
2704 (u_longlong_t)ds->ds_deadlist_obj);
2705 (void) printf("\t\tused_bytes = %s\n", used);
2706 (void) printf("\t\tcompressed_bytes = %s\n", compressed);
2707 (void) printf("\t\tuncompressed_bytes = %s\n", uncompressed);
2708 (void) printf("\t\tunique = %s\n", unique);
2709 (void) printf("\t\tfsid_guid = %llu\n",
2710 (u_longlong_t)ds->ds_fsid_guid);
2711 (void) printf("\t\tguid = %llu\n",
2712 (u_longlong_t)ds->ds_guid);
2713 (void) printf("\t\tflags = %llx\n",
2714 (u_longlong_t)ds->ds_flags);
2715 (void) printf("\t\tnext_clones_obj = %llu\n",
2716 (u_longlong_t)ds->ds_next_clones_obj);
2717 (void) printf("\t\tprops_obj = %llu\n",
2718 (u_longlong_t)ds->ds_props_obj);
2719 (void) printf("\t\tbp = %s\n", blkbuf);
2720 }
2721
2722 static int
2723 dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2724 {
2725 (void) arg, (void) tx;
2726 char blkbuf[BP_SPRINTF_LEN];
2727
2728 if (BP_GET_LOGICAL_BIRTH(bp) != 0) {
2729 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
2730 (void) printf("\t%s\n", blkbuf);
2731 }
2732 return (0);
2733 }
2734
2735 static void
2736 dump_bptree(objset_t *os, uint64_t obj, const char *name)
2737 {
2738 char bytes[32];
2739 bptree_phys_t *bt;
2740 dmu_buf_t *db;
2741
2742 /* make sure nicenum has enough space */
2743 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2744
2745 if (dump_opt['d'] < 3)
2746 return;
2747
2748 VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
2749 bt = db->db_data;
2750 zdb_nicenum(bt->bt_bytes, bytes, sizeof (bytes));
2751 (void) printf("\n %s: %llu datasets, %s\n",
2752 name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes);
2753 dmu_buf_rele(db, FTAG);
2754
2755 if (dump_opt['d'] < 5)
2756 return;
2757
2758 (void) printf("\n");
2759
2760 (void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL);
2761 }
2762
2763 static int
2764 dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
2765 {
2766 (void) arg, (void) tx;
2767 char blkbuf[BP_SPRINTF_LEN];
2768
2769 ASSERT(BP_GET_LOGICAL_BIRTH(bp) != 0);
2770 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed);
2771 (void) printf("\t%s\n", blkbuf);
2772 return (0);
2773 }
2774
2775 static void
2776 dump_full_bpobj(bpobj_t *bpo, const char *name, int indent)
2777 {
2778 char bytes[32];
2779 char comp[32];
2780 char uncomp[32];
2781 uint64_t i;
2782
2783 /* make sure nicenum has enough space */
2784 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2785 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
2786 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
2787
2788 if (dump_opt['d'] < 3)
2789 return;
2790
2791 zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes, sizeof (bytes));
2792 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
2793 zdb_nicenum(bpo->bpo_phys->bpo_comp, comp, sizeof (comp));
2794 zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp, sizeof (uncomp));
2795 if (bpo->bpo_havefreed) {
2796 (void) printf(" %*s: object %llu, %llu local "
2797 "blkptrs, %llu freed, %llu subobjs in object %llu, "
2798 "%s (%s/%s comp)\n",
2799 indent * 8, name,
2800 (u_longlong_t)bpo->bpo_object,
2801 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2802 (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2803 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2804 (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2805 bytes, comp, uncomp);
2806 } else {
2807 (void) printf(" %*s: object %llu, %llu local "
2808 "blkptrs, %llu subobjs in object %llu, "
2809 "%s (%s/%s comp)\n",
2810 indent * 8, name,
2811 (u_longlong_t)bpo->bpo_object,
2812 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2813 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2814 (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2815 bytes, comp, uncomp);
2816 }
2817
2818 for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
2819 uint64_t subobj;
2820 bpobj_t subbpo;
2821 int error;
2822 VERIFY0(dmu_read(bpo->bpo_os,
2823 bpo->bpo_phys->bpo_subobjs,
2824 i * sizeof (subobj), sizeof (subobj), &subobj, 0));
2825 error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
2826 if (error != 0) {
2827 (void) printf("ERROR %u while trying to open "
2828 "subobj id %llu\n",
2829 error, (u_longlong_t)subobj);
2830 continue;
2831 }
2832 dump_full_bpobj(&subbpo, "subobj", indent + 1);
2833 bpobj_close(&subbpo);
2834 }
2835 } else {
2836 if (bpo->bpo_havefreed) {
2837 (void) printf(" %*s: object %llu, %llu blkptrs, "
2838 "%llu freed, %s\n",
2839 indent * 8, name,
2840 (u_longlong_t)bpo->bpo_object,
2841 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2842 (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2843 bytes);
2844 } else {
2845 (void) printf(" %*s: object %llu, %llu blkptrs, "
2846 "%s\n",
2847 indent * 8, name,
2848 (u_longlong_t)bpo->bpo_object,
2849 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2850 bytes);
2851 }
2852 }
2853
2854 if (dump_opt['d'] < 5)
2855 return;
2856
2857
2858 if (indent == 0) {
2859 (void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL);
2860 (void) printf("\n");
2861 }
2862 }
2863
2864 static int
2865 dump_bookmark(dsl_pool_t *dp, char *name, boolean_t print_redact,
2866 boolean_t print_list)
2867 {
2868 int err = 0;
2869 zfs_bookmark_phys_t prop;
2870 objset_t *mos = dp->dp_spa->spa_meta_objset;
2871 err = dsl_bookmark_lookup(dp, name, NULL, &prop);
2872
2873 if (err != 0) {
2874 return (err);
2875 }
2876
2877 (void) printf("\t#%s: ", strchr(name, '#') + 1);
2878 (void) printf("{guid: %llx creation_txg: %llu creation_time: "
2879 "%llu redaction_obj: %llu}\n", (u_longlong_t)prop.zbm_guid,
2880 (u_longlong_t)prop.zbm_creation_txg,
2881 (u_longlong_t)prop.zbm_creation_time,
2882 (u_longlong_t)prop.zbm_redaction_obj);
2883
2884 IMPLY(print_list, print_redact);
2885 if (!print_redact || prop.zbm_redaction_obj == 0)
2886 return (0);
2887
2888 redaction_list_t *rl;
2889 VERIFY0(dsl_redaction_list_hold_obj(dp,
2890 prop.zbm_redaction_obj, FTAG, &rl));
2891
2892 redaction_list_phys_t *rlp = rl->rl_phys;
2893 (void) printf("\tRedacted:\n\t\tProgress: ");
2894 if (rlp->rlp_last_object != UINT64_MAX ||
2895 rlp->rlp_last_blkid != UINT64_MAX) {
2896 (void) printf("%llu %llu (incomplete)\n",
2897 (u_longlong_t)rlp->rlp_last_object,
2898 (u_longlong_t)rlp->rlp_last_blkid);
2899 } else {
2900 (void) printf("complete\n");
2901 }
2902 (void) printf("\t\tSnapshots: [");
2903 for (unsigned int i = 0; i < rlp->rlp_num_snaps; i++) {
2904 if (i > 0)
2905 (void) printf(", ");
2906 (void) printf("%0llu",
2907 (u_longlong_t)rlp->rlp_snaps[i]);
2908 }
2909 (void) printf("]\n\t\tLength: %llu\n",
2910 (u_longlong_t)rlp->rlp_num_entries);
2911
2912 if (!print_list) {
2913 dsl_redaction_list_rele(rl, FTAG);
2914 return (0);
2915 }
2916
2917 if (rlp->rlp_num_entries == 0) {
2918 dsl_redaction_list_rele(rl, FTAG);
2919 (void) printf("\t\tRedaction List: []\n\n");
2920 return (0);
2921 }
2922
2923 redact_block_phys_t *rbp_buf;
2924 uint64_t size;
2925 dmu_object_info_t doi;
2926
2927 VERIFY0(dmu_object_info(mos, prop.zbm_redaction_obj, &doi));
2928 size = doi.doi_max_offset;
2929 rbp_buf = kmem_alloc(size, KM_SLEEP);
2930
2931 err = dmu_read(mos, prop.zbm_redaction_obj, 0, size,
2932 rbp_buf, 0);
2933 if (err != 0) {
2934 dsl_redaction_list_rele(rl, FTAG);
2935 kmem_free(rbp_buf, size);
2936 return (err);
2937 }
2938
2939 (void) printf("\t\tRedaction List: [{object: %llx, offset: "
2940 "%llx, blksz: %x, count: %llx}",
2941 (u_longlong_t)rbp_buf[0].rbp_object,
2942 (u_longlong_t)rbp_buf[0].rbp_blkid,
2943 (uint_t)(redact_block_get_size(&rbp_buf[0])),
2944 (u_longlong_t)redact_block_get_count(&rbp_buf[0]));
2945
2946 for (size_t i = 1; i < rlp->rlp_num_entries; i++) {
2947 (void) printf(",\n\t\t{object: %llx, offset: %llx, "
2948 "blksz: %x, count: %llx}",
2949 (u_longlong_t)rbp_buf[i].rbp_object,
2950 (u_longlong_t)rbp_buf[i].rbp_blkid,
2951 (uint_t)(redact_block_get_size(&rbp_buf[i])),
2952 (u_longlong_t)redact_block_get_count(&rbp_buf[i]));
2953 }
2954 dsl_redaction_list_rele(rl, FTAG);
2955 kmem_free(rbp_buf, size);
2956 (void) printf("]\n\n");
2957 return (0);
2958 }
2959
2960 static void
2961 dump_bookmarks(objset_t *os, int verbosity)
2962 {
2963 zap_cursor_t zc;
2964 zap_attribute_t *attrp;
2965 dsl_dataset_t *ds = dmu_objset_ds(os);
2966 dsl_pool_t *dp = spa_get_dsl(os->os_spa);
2967 objset_t *mos = os->os_spa->spa_meta_objset;
2968 if (verbosity < 4)
2969 return;
2970 attrp = zap_attribute_alloc();
2971 dsl_pool_config_enter(dp, FTAG);
2972
2973 for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj);
2974 zap_cursor_retrieve(&zc, attrp) == 0;
2975 zap_cursor_advance(&zc)) {
2976 char osname[ZFS_MAX_DATASET_NAME_LEN];
2977 char buf[ZFS_MAX_DATASET_NAME_LEN];
2978 int len;
2979 dmu_objset_name(os, osname);
2980 len = snprintf(buf, sizeof (buf), "%s#%s", osname,
2981 attrp->za_name);
2982 VERIFY3S(len, <, ZFS_MAX_DATASET_NAME_LEN);
2983 (void) dump_bookmark(dp, buf, verbosity >= 5, verbosity >= 6);
2984 }
2985 zap_cursor_fini(&zc);
2986 dsl_pool_config_exit(dp, FTAG);
2987 zap_attribute_free(attrp);
2988 }
2989
2990 static void
2991 bpobj_count_refd(bpobj_t *bpo)
2992 {
2993 mos_obj_refd(bpo->bpo_object);
2994
2995 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
2996 mos_obj_refd(bpo->bpo_phys->bpo_subobjs);
2997 for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
2998 uint64_t subobj;
2999 bpobj_t subbpo;
3000 int error;
3001 VERIFY0(dmu_read(bpo->bpo_os,
3002 bpo->bpo_phys->bpo_subobjs,
3003 i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3004 error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3005 if (error != 0) {
3006 (void) printf("ERROR %u while trying to open "
3007 "subobj id %llu\n",
3008 error, (u_longlong_t)subobj);
3009 continue;
3010 }
3011 bpobj_count_refd(&subbpo);
3012 bpobj_close(&subbpo);
3013 }
3014 }
3015 }
3016
3017 static int
3018 dsl_deadlist_entry_count_refd(void *arg, dsl_deadlist_entry_t *dle)
3019 {
3020 spa_t *spa = arg;
3021 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3022 if (dle->dle_bpobj.bpo_object != empty_bpobj)
3023 bpobj_count_refd(&dle->dle_bpobj);
3024 return (0);
3025 }
3026
3027 static int
3028 dsl_deadlist_entry_dump(void *arg, dsl_deadlist_entry_t *dle)
3029 {
3030 ASSERT(arg == NULL);
3031 if (dump_opt['d'] >= 5) {
3032 char buf[128];
3033 (void) snprintf(buf, sizeof (buf),
3034 "mintxg %llu -> obj %llu",
3035 (longlong_t)dle->dle_mintxg,
3036 (longlong_t)dle->dle_bpobj.bpo_object);
3037
3038 dump_full_bpobj(&dle->dle_bpobj, buf, 0);
3039 } else {
3040 (void) printf("mintxg %llu -> obj %llu\n",
3041 (longlong_t)dle->dle_mintxg,
3042 (longlong_t)dle->dle_bpobj.bpo_object);
3043 }
3044 return (0);
3045 }
3046
3047 static void
3048 dump_blkptr_list(dsl_deadlist_t *dl, const char *name)
3049 {
3050 char bytes[32];
3051 char comp[32];
3052 char uncomp[32];
3053 char entries[32];
3054 spa_t *spa = dmu_objset_spa(dl->dl_os);
3055 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3056
3057 if (dl->dl_oldfmt) {
3058 if (dl->dl_bpobj.bpo_object != empty_bpobj)
3059 bpobj_count_refd(&dl->dl_bpobj);
3060 } else {
3061 mos_obj_refd(dl->dl_object);
3062 dsl_deadlist_iterate(dl, dsl_deadlist_entry_count_refd, spa);
3063 }
3064
3065 /* make sure nicenum has enough space */
3066 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3067 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
3068 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
3069 _Static_assert(sizeof (entries) >= NN_NUMBUF_SZ, "entries truncated");
3070
3071 if (dump_opt['d'] < 3)
3072 return;
3073
3074 if (dl->dl_oldfmt) {
3075 dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0);
3076 return;
3077 }
3078
3079 zdb_nicenum(dl->dl_phys->dl_used, bytes, sizeof (bytes));
3080 zdb_nicenum(dl->dl_phys->dl_comp, comp, sizeof (comp));
3081 zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp, sizeof (uncomp));
3082 zdb_nicenum(avl_numnodes(&dl->dl_tree), entries, sizeof (entries));
3083 (void) printf("\n %s: %s (%s/%s comp), %s entries\n",
3084 name, bytes, comp, uncomp, entries);
3085
3086 if (dump_opt['d'] < 4)
3087 return;
3088
3089 (void) putchar('\n');
3090
3091 dsl_deadlist_iterate(dl, dsl_deadlist_entry_dump, NULL);
3092 }
3093
3094 static int
3095 verify_dd_livelist(objset_t *os)
3096 {
3097 uint64_t ll_used, used, ll_comp, comp, ll_uncomp, uncomp;
3098 dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3099 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
3100
3101 ASSERT(!dmu_objset_is_snapshot(os));
3102 if (!dsl_deadlist_is_open(&dd->dd_livelist))
3103 return (0);
3104
3105 /* Iterate through the livelist to check for duplicates */
3106 dsl_deadlist_iterate(&dd->dd_livelist, sublivelist_verify_lightweight,
3107 NULL);
3108
3109 dsl_pool_config_enter(dp, FTAG);
3110 dsl_deadlist_space(&dd->dd_livelist, &ll_used,
3111 &ll_comp, &ll_uncomp);
3112
3113 dsl_dataset_t *origin_ds;
3114 ASSERT(dsl_pool_config_held(dp));
3115 VERIFY0(dsl_dataset_hold_obj(dp,
3116 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &origin_ds));
3117 VERIFY0(dsl_dataset_space_written(origin_ds, os->os_dsl_dataset,
3118 &used, &comp, &uncomp));
3119 dsl_dataset_rele(origin_ds, FTAG);
3120 dsl_pool_config_exit(dp, FTAG);
3121 /*
3122 * It's possible that the dataset's uncomp space is larger than the
3123 * livelist's because livelists do not track embedded block pointers
3124 */
3125 if (used != ll_used || comp != ll_comp || uncomp < ll_uncomp) {
3126 char nice_used[32], nice_comp[32], nice_uncomp[32];
3127 (void) printf("Discrepancy in space accounting:\n");
3128 zdb_nicenum(used, nice_used, sizeof (nice_used));
3129 zdb_nicenum(comp, nice_comp, sizeof (nice_comp));
3130 zdb_nicenum(uncomp, nice_uncomp, sizeof (nice_uncomp));
3131 (void) printf("dir: used %s, comp %s, uncomp %s\n",
3132 nice_used, nice_comp, nice_uncomp);
3133 zdb_nicenum(ll_used, nice_used, sizeof (nice_used));
3134 zdb_nicenum(ll_comp, nice_comp, sizeof (nice_comp));
3135 zdb_nicenum(ll_uncomp, nice_uncomp, sizeof (nice_uncomp));
3136 (void) printf("livelist: used %s, comp %s, uncomp %s\n",
3137 nice_used, nice_comp, nice_uncomp);
3138 return (1);
3139 }
3140 return (0);
3141 }
3142
3143 static char *key_material = NULL;
3144
3145 static boolean_t
3146 zdb_derive_key(dsl_dir_t *dd, uint8_t *key_out)
3147 {
3148 uint64_t keyformat, salt, iters;
3149 int i;
3150 unsigned char c;
3151
3152 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3153 zfs_prop_to_name(ZFS_PROP_KEYFORMAT), sizeof (uint64_t),
3154 1, &keyformat));
3155
3156 switch (keyformat) {
3157 case ZFS_KEYFORMAT_HEX:
3158 for (i = 0; i < WRAPPING_KEY_LEN * 2; i += 2) {
3159 if (!isxdigit(key_material[i]) ||
3160 !isxdigit(key_material[i+1]))
3161 return (B_FALSE);
3162 if (sscanf(&key_material[i], "%02hhx", &c) != 1)
3163 return (B_FALSE);
3164 key_out[i / 2] = c;
3165 }
3166 break;
3167
3168 case ZFS_KEYFORMAT_PASSPHRASE:
3169 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3170 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT),
3171 sizeof (uint64_t), 1, &salt));
3172 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3173 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS),
3174 sizeof (uint64_t), 1, &iters));
3175
3176 if (PKCS5_PBKDF2_HMAC_SHA1(key_material, strlen(key_material),
3177 ((uint8_t *)&salt), sizeof (uint64_t), iters,
3178 WRAPPING_KEY_LEN, key_out) != 1)
3179 return (B_FALSE);
3180
3181 break;
3182
3183 default:
3184 fatal("no support for key format %u\n",
3185 (unsigned int) keyformat);
3186 }
3187
3188 return (B_TRUE);
3189 }
3190
3191 static char encroot[ZFS_MAX_DATASET_NAME_LEN];
3192 static boolean_t key_loaded = B_FALSE;
3193
3194 static void
3195 zdb_load_key(objset_t *os)
3196 {
3197 dsl_pool_t *dp;
3198 dsl_dir_t *dd, *rdd;
3199 uint8_t key[WRAPPING_KEY_LEN];
3200 uint64_t rddobj;
3201 int err;
3202
3203 dp = spa_get_dsl(os->os_spa);
3204 dd = os->os_dsl_dataset->ds_dir;
3205
3206 dsl_pool_config_enter(dp, FTAG);
3207 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3208 DSL_CRYPTO_KEY_ROOT_DDOBJ, sizeof (uint64_t), 1, &rddobj));
3209 VERIFY0(dsl_dir_hold_obj(dd->dd_pool, rddobj, NULL, FTAG, &rdd));
3210 dsl_dir_name(rdd, encroot);
3211 dsl_dir_rele(rdd, FTAG);
3212
3213 if (!zdb_derive_key(dd, key))
3214 fatal("couldn't derive encryption key");
3215
3216 dsl_pool_config_exit(dp, FTAG);
3217
3218 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_UNAVAILABLE);
3219
3220 dsl_crypto_params_t *dcp;
3221 nvlist_t *crypto_args;
3222
3223 crypto_args = fnvlist_alloc();
3224 fnvlist_add_uint8_array(crypto_args, "wkeydata",
3225 (uint8_t *)key, WRAPPING_KEY_LEN);
3226 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE,
3227 NULL, crypto_args, &dcp));
3228 err = spa_keystore_load_wkey(encroot, dcp, B_FALSE);
3229
3230 dsl_crypto_params_free(dcp, (err != 0));
3231 fnvlist_free(crypto_args);
3232
3233 if (err != 0)
3234 fatal(
3235 "couldn't load encryption key for %s: %s",
3236 encroot, err == ZFS_ERR_CRYPTO_NOTSUP ?
3237 "crypto params not supported" : strerror(err));
3238
3239 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_AVAILABLE);
3240
3241 printf("Unlocked encryption root: %s\n", encroot);
3242 key_loaded = B_TRUE;
3243 }
3244
3245 static void
3246 zdb_unload_key(void)
3247 {
3248 if (!key_loaded)
3249 return;
3250
3251 VERIFY0(spa_keystore_unload_wkey(encroot));
3252 key_loaded = B_FALSE;
3253 }
3254
3255 static avl_tree_t idx_tree;
3256 static avl_tree_t domain_tree;
3257 static boolean_t fuid_table_loaded;
3258 static objset_t *sa_os = NULL;
3259 static sa_attr_type_t *sa_attr_table = NULL;
3260
3261 static int
3262 open_objset(const char *path, const void *tag, objset_t **osp)
3263 {
3264 int err;
3265 uint64_t sa_attrs = 0;
3266 uint64_t version = 0;
3267
3268 VERIFY3P(sa_os, ==, NULL);
3269
3270 /*
3271 * We can't own an objset if it's redacted. Therefore, we do this
3272 * dance: hold the objset, then acquire a long hold on its dataset, then
3273 * release the pool (which is held as part of holding the objset).
3274 */
3275
3276 if (dump_opt['K']) {
3277 /* decryption requested, try to load keys */
3278 err = dmu_objset_hold(path, tag, osp);
3279 if (err != 0) {
3280 (void) fprintf(stderr, "failed to hold dataset "
3281 "'%s': %s\n",
3282 path, strerror(err));
3283 return (err);
3284 }
3285 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3286 dsl_pool_rele(dmu_objset_pool(*osp), tag);
3287
3288 /* succeeds or dies */
3289 zdb_load_key(*osp);
3290
3291 /* release it all */
3292 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3293 dsl_dataset_rele(dmu_objset_ds(*osp), tag);
3294 }
3295
3296 int ds_hold_flags = key_loaded ? DS_HOLD_FLAG_DECRYPT : 0;
3297
3298 err = dmu_objset_hold_flags(path, ds_hold_flags, tag, osp);
3299 if (err != 0) {
3300 (void) fprintf(stderr, "failed to hold dataset '%s': %s\n",
3301 path, strerror(err));
3302 return (err);
3303 }
3304 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3305 dsl_pool_rele(dmu_objset_pool(*osp), tag);
3306
3307 if (dmu_objset_type(*osp) == DMU_OST_ZFS &&
3308 (key_loaded || !(*osp)->os_encrypted)) {
3309 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR,
3310 8, 1, &version);
3311 if (version >= ZPL_VERSION_SA) {
3312 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS,
3313 8, 1, &sa_attrs);
3314 }
3315 err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END,
3316 &sa_attr_table);
3317 if (err != 0) {
3318 (void) fprintf(stderr, "sa_setup failed: %s\n",
3319 strerror(err));
3320 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3321 dsl_dataset_rele_flags(dmu_objset_ds(*osp),
3322 ds_hold_flags, tag);
3323 *osp = NULL;
3324 }
3325 }
3326 sa_os = *osp;
3327
3328 return (err);
3329 }
3330
3331 static void
3332 close_objset(objset_t *os, const void *tag)
3333 {
3334 VERIFY3P(os, ==, sa_os);
3335 if (os->os_sa != NULL)
3336 sa_tear_down(os);
3337 dsl_dataset_long_rele(dmu_objset_ds(os), tag);
3338 dsl_dataset_rele_flags(dmu_objset_ds(os),
3339 key_loaded ? DS_HOLD_FLAG_DECRYPT : 0, tag);
3340 sa_attr_table = NULL;
3341 sa_os = NULL;
3342
3343 zdb_unload_key();
3344 }
3345
3346 static void
3347 fuid_table_destroy(void)
3348 {
3349 if (fuid_table_loaded) {
3350 zfs_fuid_table_destroy(&idx_tree, &domain_tree);
3351 fuid_table_loaded = B_FALSE;
3352 }
3353 }
3354
3355 /*
3356 * Clean up DDT internal state. ddt_lookup() adds entries to ddt_tree, which on
3357 * a live pool are normally cleaned up during ddt_sync(). We can't do that (and
3358 * wouldn't want to anyway), but if we don't clean up the presence of stuff on
3359 * ddt_tree will trip asserts in ddt_table_free(). So, we clean up ourselves.
3360 *
3361 * Note that this is not a particularly efficient way to do this, but
3362 * ddt_remove() is the only public method that can do the work we need, and it
3363 * requires the right locks and etc to do the job. This is only ever called
3364 * during zdb shutdown so efficiency is not especially important.
3365 */
3366 static void
3367 zdb_ddt_cleanup(spa_t *spa)
3368 {
3369 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
3370 ddt_t *ddt = spa->spa_ddt[c];
3371 if (!ddt)
3372 continue;
3373
3374 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3375 ddt_enter(ddt);
3376 ddt_entry_t *dde = avl_first(&ddt->ddt_tree), *next;
3377 while (dde) {
3378 next = AVL_NEXT(&ddt->ddt_tree, dde);
3379 dde->dde_io = NULL;
3380 ddt_remove(ddt, dde);
3381 dde = next;
3382 }
3383 ddt_exit(ddt);
3384 spa_config_exit(spa, SCL_CONFIG, FTAG);
3385 }
3386 }
3387
3388 static void
3389 zdb_exit(int reason)
3390 {
3391 if (spa != NULL)
3392 zdb_ddt_cleanup(spa);
3393
3394 if (os != NULL) {
3395 close_objset(os, FTAG);
3396 } else if (spa != NULL) {
3397 spa_close(spa, FTAG);
3398 }
3399
3400 fuid_table_destroy();
3401
3402 if (kernel_init_done)
3403 kernel_fini();
3404
3405 exit(reason);
3406 }
3407
3408 /*
3409 * print uid or gid information.
3410 * For normal POSIX id just the id is printed in decimal format.
3411 * For CIFS files with FUID the fuid is printed in hex followed by
3412 * the domain-rid string.
3413 */
3414 static void
3415 print_idstr(uint64_t id, const char *id_type)
3416 {
3417 if (FUID_INDEX(id)) {
3418 const char *domain =
3419 zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id));
3420 (void) printf("\t%s %llx [%s-%d]\n", id_type,
3421 (u_longlong_t)id, domain, (int)FUID_RID(id));
3422 } else {
3423 (void) printf("\t%s %llu\n", id_type, (u_longlong_t)id);
3424 }
3425
3426 }
3427
3428 static void
3429 dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid)
3430 {
3431 uint32_t uid_idx, gid_idx;
3432
3433 uid_idx = FUID_INDEX(uid);
3434 gid_idx = FUID_INDEX(gid);
3435
3436 /* Load domain table, if not already loaded */
3437 if (!fuid_table_loaded && (uid_idx || gid_idx)) {
3438 uint64_t fuid_obj;
3439
3440 /* first find the fuid object. It lives in the master node */
3441 VERIFY(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES,
3442 8, 1, &fuid_obj) == 0);
3443 zfs_fuid_avl_tree_create(&idx_tree, &domain_tree);
3444 (void) zfs_fuid_table_load(os, fuid_obj,
3445 &idx_tree, &domain_tree);
3446 fuid_table_loaded = B_TRUE;
3447 }
3448
3449 print_idstr(uid, "uid");
3450 print_idstr(gid, "gid");
3451 }
3452
3453 static void
3454 dump_znode_sa_xattr(sa_handle_t *hdl)
3455 {
3456 nvlist_t *sa_xattr;
3457 nvpair_t *elem = NULL;
3458 int sa_xattr_size = 0;
3459 int sa_xattr_entries = 0;
3460 int error;
3461 char *sa_xattr_packed;
3462
3463 error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size);
3464 if (error || sa_xattr_size == 0)
3465 return;
3466
3467 sa_xattr_packed = malloc(sa_xattr_size);
3468 if (sa_xattr_packed == NULL)
3469 return;
3470
3471 error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR],
3472 sa_xattr_packed, sa_xattr_size);
3473 if (error) {
3474 free(sa_xattr_packed);
3475 return;
3476 }
3477
3478 error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0);
3479 if (error) {
3480 free(sa_xattr_packed);
3481 return;
3482 }
3483
3484 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL)
3485 sa_xattr_entries++;
3486
3487 (void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3488 sa_xattr_size, sa_xattr_entries);
3489 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) {
3490 boolean_t can_print = !dump_opt['P'];
3491 uchar_t *value;
3492 uint_t cnt, idx;
3493
3494 (void) printf("\t\t%s = ", nvpair_name(elem));
3495 nvpair_value_byte_array(elem, &value, &cnt);
3496
3497 for (idx = 0; idx < cnt; ++idx) {
3498 if (!isprint(value[idx])) {
3499 can_print = B_FALSE;
3500 break;
3501 }
3502 }
3503
3504 for (idx = 0; idx < cnt; ++idx) {
3505 if (can_print)
3506 (void) putchar(value[idx]);
3507 else
3508 (void) printf("\\%3.3o", value[idx]);
3509 }
3510 (void) putchar('\n');
3511 }
3512
3513 nvlist_free(sa_xattr);
3514 free(sa_xattr_packed);
3515 }
3516
3517 static void
3518 dump_znode_symlink(sa_handle_t *hdl)
3519 {
3520 int sa_symlink_size = 0;
3521 char linktarget[MAXPATHLEN];
3522 int error;
3523
3524 error = sa_size(hdl, sa_attr_table[ZPL_SYMLINK], &sa_symlink_size);
3525 if (error || sa_symlink_size == 0) {
3526 return;
3527 }
3528 if (sa_symlink_size >= sizeof (linktarget)) {
3529 (void) printf("symlink size %d is too large\n",
3530 sa_symlink_size);
3531 return;
3532 }
3533 linktarget[sa_symlink_size] = '\0';
3534 if (sa_lookup(hdl, sa_attr_table[ZPL_SYMLINK],
3535 &linktarget, sa_symlink_size) == 0)
3536 (void) printf("\ttarget %s\n", linktarget);
3537 }
3538
3539 static void
3540 dump_znode(objset_t *os, uint64_t object, void *data, size_t size)
3541 {
3542 (void) data, (void) size;
3543 char path[MAXPATHLEN * 2]; /* allow for xattr and failure prefix */
3544 sa_handle_t *hdl;
3545 uint64_t xattr, rdev, gen;
3546 uint64_t uid, gid, mode, fsize, parent, links;
3547 uint64_t pflags;
3548 uint64_t acctm[2], modtm[2], chgtm[2], crtm[2];
3549 time_t z_crtime, z_atime, z_mtime, z_ctime;
3550 sa_bulk_attr_t bulk[12];
3551 int idx = 0;
3552 int error;
3553
3554 VERIFY3P(os, ==, sa_os);
3555 if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) {
3556 (void) printf("Failed to get handle for SA znode\n");
3557 return;
3558 }
3559
3560 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8);
3561 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8);
3562 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL,
3563 &links, 8);
3564 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8);
3565 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL,
3566 &mode, 8);
3567 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT],
3568 NULL, &parent, 8);
3569 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL,
3570 &fsize, 8);
3571 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL,
3572 acctm, 16);
3573 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL,
3574 modtm, 16);
3575 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL,
3576 crtm, 16);
3577 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL,
3578 chgtm, 16);
3579 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL,
3580 &pflags, 8);
3581
3582 if (sa_bulk_lookup(hdl, bulk, idx)) {
3583 (void) sa_handle_destroy(hdl);
3584 return;
3585 }
3586
3587 z_crtime = (time_t)crtm[0];
3588 z_atime = (time_t)acctm[0];
3589 z_mtime = (time_t)modtm[0];
3590 z_ctime = (time_t)chgtm[0];
3591
3592 if (dump_opt['d'] > 4) {
3593 error = zfs_obj_to_path(os, object, path, sizeof (path));
3594 if (error == ESTALE) {
3595 (void) snprintf(path, sizeof (path), "on delete queue");
3596 } else if (error != 0) {
3597 leaked_objects++;
3598 (void) snprintf(path, sizeof (path),
3599 "path not found, possibly leaked");
3600 }
3601 (void) printf("\tpath %s\n", path);
3602 }
3603
3604 if (S_ISLNK(mode))
3605 dump_znode_symlink(hdl);
3606 dump_uidgid(os, uid, gid);
3607 (void) printf("\tatime %s", ctime(&z_atime));
3608 (void) printf("\tmtime %s", ctime(&z_mtime));
3609 (void) printf("\tctime %s", ctime(&z_ctime));
3610 (void) printf("\tcrtime %s", ctime(&z_crtime));
3611 (void) printf("\tgen %llu\n", (u_longlong_t)gen);
3612 (void) printf("\tmode %llo\n", (u_longlong_t)mode);
3613 (void) printf("\tsize %llu\n", (u_longlong_t)fsize);
3614 (void) printf("\tparent %llu\n", (u_longlong_t)parent);
3615 (void) printf("\tlinks %llu\n", (u_longlong_t)links);
3616 (void) printf("\tpflags %llx\n", (u_longlong_t)pflags);
3617 if (dmu_objset_projectquota_enabled(os) && (pflags & ZFS_PROJID)) {
3618 uint64_t projid;
3619
3620 if (sa_lookup(hdl, sa_attr_table[ZPL_PROJID], &projid,
3621 sizeof (uint64_t)) == 0)
3622 (void) printf("\tprojid %llu\n", (u_longlong_t)projid);
3623 }
3624 if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr,
3625 sizeof (uint64_t)) == 0)
3626 (void) printf("\txattr %llu\n", (u_longlong_t)xattr);
3627 if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev,
3628 sizeof (uint64_t)) == 0)
3629 (void) printf("\trdev 0x%016llx\n", (u_longlong_t)rdev);
3630 dump_znode_sa_xattr(hdl);
3631 sa_handle_destroy(hdl);
3632 }
3633
3634 static void
3635 dump_acl(objset_t *os, uint64_t object, void *data, size_t size)
3636 {
3637 (void) os, (void) object, (void) data, (void) size;
3638 }
3639
3640 static void
3641 dump_dmu_objset(objset_t *os, uint64_t object, void *data, size_t size)
3642 {
3643 (void) os, (void) object, (void) data, (void) size;
3644 }
3645
3646 static object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = {
3647 dump_none, /* unallocated */
3648 dump_zap, /* object directory */
3649 dump_uint64, /* object array */
3650 dump_none, /* packed nvlist */
3651 dump_packed_nvlist, /* packed nvlist size */
3652 dump_none, /* bpobj */
3653 dump_bpobj, /* bpobj header */
3654 dump_none, /* SPA space map header */
3655 dump_none, /* SPA space map */
3656 dump_none, /* ZIL intent log */
3657 dump_dnode, /* DMU dnode */
3658 dump_dmu_objset, /* DMU objset */
3659 dump_dsl_dir, /* DSL directory */
3660 dump_zap, /* DSL directory child map */
3661 dump_zap, /* DSL dataset snap map */
3662 dump_zap, /* DSL props */
3663 dump_dsl_dataset, /* DSL dataset */
3664 dump_znode, /* ZFS znode */
3665 dump_acl, /* ZFS V0 ACL */
3666 dump_uint8, /* ZFS plain file */
3667 dump_zpldir, /* ZFS directory */
3668 dump_zap, /* ZFS master node */
3669 dump_zap, /* ZFS delete queue */
3670 dump_uint8, /* zvol object */
3671 dump_zap, /* zvol prop */
3672 dump_uint8, /* other uint8[] */
3673 dump_uint64, /* other uint64[] */
3674 dump_zap, /* other ZAP */
3675 dump_zap, /* persistent error log */
3676 dump_uint8, /* SPA history */
3677 dump_history_offsets, /* SPA history offsets */
3678 dump_zap, /* Pool properties */
3679 dump_zap, /* DSL permissions */
3680 dump_acl, /* ZFS ACL */
3681 dump_uint8, /* ZFS SYSACL */
3682 dump_none, /* FUID nvlist */
3683 dump_packed_nvlist, /* FUID nvlist size */
3684 dump_zap, /* DSL dataset next clones */
3685 dump_zap, /* DSL scrub queue */
3686 dump_zap, /* ZFS user/group/project used */
3687 dump_zap, /* ZFS user/group/project quota */
3688 dump_zap, /* snapshot refcount tags */
3689 dump_ddt_zap, /* DDT ZAP object */
3690 dump_zap, /* DDT statistics */
3691 dump_znode, /* SA object */
3692 dump_zap, /* SA Master Node */
3693 dump_sa_attrs, /* SA attribute registration */
3694 dump_sa_layouts, /* SA attribute layouts */
3695 dump_zap, /* DSL scrub translations */
3696 dump_none, /* fake dedup BP */
3697 dump_zap, /* deadlist */
3698 dump_none, /* deadlist hdr */
3699 dump_zap, /* dsl clones */
3700 dump_bpobj_subobjs, /* bpobj subobjs */
3701 dump_unknown, /* Unknown type, must be last */
3702 };
3703
3704 static boolean_t
3705 match_object_type(dmu_object_type_t obj_type, uint64_t flags)
3706 {
3707 boolean_t match = B_TRUE;
3708
3709 switch (obj_type) {
3710 case DMU_OT_DIRECTORY_CONTENTS:
3711 if (!(flags & ZOR_FLAG_DIRECTORY))
3712 match = B_FALSE;
3713 break;
3714 case DMU_OT_PLAIN_FILE_CONTENTS:
3715 if (!(flags & ZOR_FLAG_PLAIN_FILE))
3716 match = B_FALSE;
3717 break;
3718 case DMU_OT_SPACE_MAP:
3719 if (!(flags & ZOR_FLAG_SPACE_MAP))
3720 match = B_FALSE;
3721 break;
3722 default:
3723 if (strcmp(zdb_ot_name(obj_type), "zap") == 0) {
3724 if (!(flags & ZOR_FLAG_ZAP))
3725 match = B_FALSE;
3726 break;
3727 }
3728
3729 /*
3730 * If all bits except some of the supported flags are
3731 * set, the user combined the all-types flag (A) with
3732 * a negated flag to exclude some types (e.g. A-f to
3733 * show all object types except plain files).
3734 */
3735 if ((flags | ZOR_SUPPORTED_FLAGS) != ZOR_FLAG_ALL_TYPES)
3736 match = B_FALSE;
3737
3738 break;
3739 }
3740
3741 return (match);
3742 }
3743
3744 static void
3745 dump_object(objset_t *os, uint64_t object, int verbosity,
3746 boolean_t *print_header, uint64_t *dnode_slots_used, uint64_t flags)
3747 {
3748 dmu_buf_t *db = NULL;
3749 dmu_object_info_t doi;
3750 dnode_t *dn;
3751 boolean_t dnode_held = B_FALSE;
3752 void *bonus = NULL;
3753 size_t bsize = 0;
3754 char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
3755 char bonus_size[32];
3756 char aux[50];
3757 int error;
3758
3759 /* make sure nicenum has enough space */
3760 _Static_assert(sizeof (iblk) >= NN_NUMBUF_SZ, "iblk truncated");
3761 _Static_assert(sizeof (dblk) >= NN_NUMBUF_SZ, "dblk truncated");
3762 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, "lsize truncated");
3763 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, "asize truncated");
3764 _Static_assert(sizeof (bonus_size) >= NN_NUMBUF_SZ,
3765 "bonus_size truncated");
3766
3767 if (*print_header) {
3768 (void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n",
3769 "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3770 "lsize", "%full", "type");
3771 *print_header = 0;
3772 }
3773
3774 if (object == 0) {
3775 dn = DMU_META_DNODE(os);
3776 dmu_object_info_from_dnode(dn, &doi);
3777 } else {
3778 /*
3779 * Encrypted datasets will have sensitive bonus buffers
3780 * encrypted. Therefore we cannot hold the bonus buffer and
3781 * must hold the dnode itself instead.
3782 */
3783 error = dmu_object_info(os, object, &doi);
3784 if (error)
3785 fatal("dmu_object_info() failed, errno %u", error);
3786
3787 if (!key_loaded && os->os_encrypted &&
3788 DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) {
3789 error = dnode_hold(os, object, FTAG, &dn);
3790 if (error)
3791 fatal("dnode_hold() failed, errno %u", error);
3792 dnode_held = B_TRUE;
3793 } else {
3794 error = dmu_bonus_hold(os, object, FTAG, &db);
3795 if (error)
3796 fatal("dmu_bonus_hold(%llu) failed, errno %u",
3797 object, error);
3798 bonus = db->db_data;
3799 bsize = db->db_size;
3800 dn = DB_DNODE((dmu_buf_impl_t *)db);
3801 }
3802 }
3803
3804 /*
3805 * Default to showing all object types if no flags were specified.
3806 */
3807 if (flags != 0 && flags != ZOR_FLAG_ALL_TYPES &&
3808 !match_object_type(doi.doi_type, flags))
3809 goto out;
3810
3811 if (dnode_slots_used)
3812 *dnode_slots_used = doi.doi_dnodesize / DNODE_MIN_SIZE;
3813
3814 zdb_nicenum(doi.doi_metadata_block_size, iblk, sizeof (iblk));
3815 zdb_nicenum(doi.doi_data_block_size, dblk, sizeof (dblk));
3816 zdb_nicenum(doi.doi_max_offset, lsize, sizeof (lsize));
3817 zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize, sizeof (asize));
3818 zdb_nicenum(doi.doi_bonus_size, bonus_size, sizeof (bonus_size));
3819 zdb_nicenum(doi.doi_dnodesize, dnsize, sizeof (dnsize));
3820 (void) snprintf(fill, sizeof (fill), "%6.2f", 100.0 *
3821 doi.doi_fill_count * doi.doi_data_block_size / (object == 0 ?
3822 DNODES_PER_BLOCK : 1) / doi.doi_max_offset);
3823
3824 aux[0] = '\0';
3825
3826 if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) {
3827 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3828 " (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum));
3829 }
3830
3831 if (doi.doi_compress == ZIO_COMPRESS_INHERIT &&
3832 ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) {
3833 const char *compname = NULL;
3834 if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION,
3835 ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel),
3836 &compname) == 0) {
3837 (void) snprintf(aux + strlen(aux),
3838 sizeof (aux) - strlen(aux), " (Z=inherit=%s)",
3839 compname);
3840 } else {
3841 (void) snprintf(aux + strlen(aux),
3842 sizeof (aux) - strlen(aux),
3843 " (Z=inherit=%s-unknown)",
3844 ZDB_COMPRESS_NAME(os->os_compress));
3845 }
3846 } else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) {
3847 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3848 " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress));
3849 } else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
3850 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3851 " (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress));
3852 }
3853
3854 (void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n",
3855 (u_longlong_t)object, doi.doi_indirection, iblk, dblk,
3856 asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux);
3857
3858 if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
3859 (void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n",
3860 "", "", "", "", "", "", bonus_size, "bonus",
3861 zdb_ot_name(doi.doi_bonus_type));
3862 }
3863
3864 if (verbosity >= 4) {
3865 (void) printf("\tdnode flags: %s%s%s%s\n",
3866 (dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ?
3867 "USED_BYTES " : "",
3868 (dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ?
3869 "USERUSED_ACCOUNTED " : "",
3870 (dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ?
3871 "USEROBJUSED_ACCOUNTED " : "",
3872 (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ?
3873 "SPILL_BLKPTR" : "");
3874 (void) printf("\tdnode maxblkid: %llu\n",
3875 (longlong_t)dn->dn_phys->dn_maxblkid);
3876
3877 if (!dnode_held) {
3878 object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os,
3879 object, bonus, bsize);
3880 } else {
3881 (void) printf("\t\t(bonus encrypted)\n");
3882 }
3883
3884 if (key_loaded ||
3885 (!os->os_encrypted || !DMU_OT_IS_ENCRYPTED(doi.doi_type))) {
3886 object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object,
3887 NULL, 0);
3888 } else {
3889 (void) printf("\t\t(object encrypted)\n");
3890 }
3891
3892 *print_header = B_TRUE;
3893 }
3894
3895 if (verbosity >= 5) {
3896 if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
3897 char blkbuf[BP_SPRINTF_LEN];
3898 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
3899 DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE);
3900 (void) printf("\nSpill block: %s\n", blkbuf);
3901 }
3902 dump_indirect(dn);
3903 }
3904
3905 if (verbosity >= 5) {
3906 /*
3907 * Report the list of segments that comprise the object.
3908 */
3909 uint64_t start = 0;
3910 uint64_t end;
3911 uint64_t blkfill = 1;
3912 int minlvl = 1;
3913
3914 if (dn->dn_type == DMU_OT_DNODE) {
3915 minlvl = 0;
3916 blkfill = DNODES_PER_BLOCK;
3917 }
3918
3919 for (;;) {
3920 char segsize[32];
3921 /* make sure nicenum has enough space */
3922 _Static_assert(sizeof (segsize) >= NN_NUMBUF_SZ,
3923 "segsize truncated");
3924 error = dnode_next_offset(dn,
3925 0, &start, minlvl, blkfill, 0);
3926 if (error)
3927 break;
3928 end = start;
3929 error = dnode_next_offset(dn,
3930 DNODE_FIND_HOLE, &end, minlvl, blkfill, 0);
3931 zdb_nicenum(end - start, segsize, sizeof (segsize));
3932 (void) printf("\t\tsegment [%016llx, %016llx)"
3933 " size %5s\n", (u_longlong_t)start,
3934 (u_longlong_t)end, segsize);
3935 if (error)
3936 break;
3937 start = end;
3938 }
3939 }
3940
3941 out:
3942 if (db != NULL)
3943 dmu_buf_rele(db, FTAG);
3944 if (dnode_held)
3945 dnode_rele(dn, FTAG);
3946 }
3947
3948 static void
3949 count_dir_mos_objects(dsl_dir_t *dd)
3950 {
3951 mos_obj_refd(dd->dd_object);
3952 mos_obj_refd(dsl_dir_phys(dd)->dd_child_dir_zapobj);
3953 mos_obj_refd(dsl_dir_phys(dd)->dd_deleg_zapobj);
3954 mos_obj_refd(dsl_dir_phys(dd)->dd_props_zapobj);
3955 mos_obj_refd(dsl_dir_phys(dd)->dd_clones);
3956
3957 /*
3958 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
3959 * Ignore the references after the first one.
3960 */
3961 mos_obj_refd_multiple(dd->dd_crypto_obj);
3962 }
3963
3964 static void
3965 count_ds_mos_objects(dsl_dataset_t *ds)
3966 {
3967 mos_obj_refd(ds->ds_object);
3968 mos_obj_refd(dsl_dataset_phys(ds)->ds_next_clones_obj);
3969 mos_obj_refd(dsl_dataset_phys(ds)->ds_props_obj);
3970 mos_obj_refd(dsl_dataset_phys(ds)->ds_userrefs_obj);
3971 mos_obj_refd(dsl_dataset_phys(ds)->ds_snapnames_zapobj);
3972 mos_obj_refd(ds->ds_bookmarks_obj);
3973
3974 if (!dsl_dataset_is_snapshot(ds)) {
3975 count_dir_mos_objects(ds->ds_dir);
3976 }
3977 }
3978
3979 static const char *const objset_types[DMU_OST_NUMTYPES] = {
3980 "NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
3981
3982 /*
3983 * Parse a string denoting a range of object IDs of the form
3984 * <start>[:<end>[:flags]], and store the results in zor.
3985 * Return 0 on success. On error, return 1 and update the msg
3986 * pointer to point to a descriptive error message.
3987 */
3988 static int
3989 parse_object_range(char *range, zopt_object_range_t *zor, const char **msg)
3990 {
3991 uint64_t flags = 0;
3992 char *p, *s, *dup, *flagstr, *tmp = NULL;
3993 size_t len;
3994 int i;
3995 int rc = 0;
3996
3997 if (strchr(range, ':') == NULL) {
3998 zor->zor_obj_start = strtoull(range, &p, 0);
3999 if (*p != '\0') {
4000 *msg = "Invalid characters in object ID";
4001 rc = 1;
4002 }
4003 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4004 zor->zor_obj_end = zor->zor_obj_start;
4005 return (rc);
4006 }
4007
4008 if (strchr(range, ':') == range) {
4009 *msg = "Invalid leading colon";
4010 rc = 1;
4011 return (rc);
4012 }
4013
4014 len = strlen(range);
4015 if (range[len - 1] == ':') {
4016 *msg = "Invalid trailing colon";
4017 rc = 1;
4018 return (rc);
4019 }
4020
4021 dup = strdup(range);
4022 s = strtok_r(dup, ":", &tmp);
4023 zor->zor_obj_start = strtoull(s, &p, 0);
4024
4025 if (*p != '\0') {
4026 *msg = "Invalid characters in start object ID";
4027 rc = 1;
4028 goto out;
4029 }
4030
4031 s = strtok_r(NULL, ":", &tmp);
4032 zor->zor_obj_end = strtoull(s, &p, 0);
4033
4034 if (*p != '\0') {
4035 *msg = "Invalid characters in end object ID";
4036 rc = 1;
4037 goto out;
4038 }
4039
4040 if (zor->zor_obj_start > zor->zor_obj_end) {
4041 *msg = "Start object ID may not exceed end object ID";
4042 rc = 1;
4043 goto out;
4044 }
4045
4046 s = strtok_r(NULL, ":", &tmp);
4047 if (s == NULL) {
4048 zor->zor_flags = ZOR_FLAG_ALL_TYPES;
4049 goto out;
4050 } else if (strtok_r(NULL, ":", &tmp) != NULL) {
4051 *msg = "Invalid colon-delimited field after flags";
4052 rc = 1;
4053 goto out;
4054 }
4055
4056 flagstr = s;
4057 for (i = 0; flagstr[i]; i++) {
4058 int bit;
4059 boolean_t negation = (flagstr[i] == '-');
4060
4061 if (negation) {
4062 i++;
4063 if (flagstr[i] == '\0') {
4064 *msg = "Invalid trailing negation operator";
4065 rc = 1;
4066 goto out;
4067 }
4068 }
4069 bit = flagbits[(uchar_t)flagstr[i]];
4070 if (bit == 0) {
4071 *msg = "Invalid flag";
4072 rc = 1;
4073 goto out;
4074 }
4075 if (negation)
4076 flags &= ~bit;
4077 else
4078 flags |= bit;
4079 }
4080 zor->zor_flags = flags;
4081
4082 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4083 zor->zor_obj_end = ZDB_MAP_OBJECT_ID(zor->zor_obj_end);
4084
4085 out:
4086 free(dup);
4087 return (rc);
4088 }
4089
4090 static void
4091 dump_objset(objset_t *os)
4092 {
4093 dmu_objset_stats_t dds = { 0 };
4094 uint64_t object, object_count;
4095 uint64_t refdbytes, usedobjs, scratch;
4096 char numbuf[32];
4097 char blkbuf[BP_SPRINTF_LEN + 20];
4098 char osname[ZFS_MAX_DATASET_NAME_LEN];
4099 const char *type = "UNKNOWN";
4100 int verbosity = dump_opt['d'];
4101 boolean_t print_header;
4102 unsigned i;
4103 int error;
4104 uint64_t total_slots_used = 0;
4105 uint64_t max_slot_used = 0;
4106 uint64_t dnode_slots;
4107 uint64_t obj_start;
4108 uint64_t obj_end;
4109 uint64_t flags;
4110
4111 /* make sure nicenum has enough space */
4112 _Static_assert(sizeof (numbuf) >= NN_NUMBUF_SZ, "numbuf truncated");
4113
4114 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
4115 dmu_objset_fast_stat(os, &dds);
4116 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
4117
4118 print_header = B_TRUE;
4119
4120 if (dds.dds_type < DMU_OST_NUMTYPES)
4121 type = objset_types[dds.dds_type];
4122
4123 if (dds.dds_type == DMU_OST_META) {
4124 dds.dds_creation_txg = TXG_INITIAL;
4125 usedobjs = BP_GET_FILL(os->os_rootbp);
4126 refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)->
4127 dd_used_bytes;
4128 } else {
4129 dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
4130 }
4131
4132 ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
4133
4134 zdb_nicenum(refdbytes, numbuf, sizeof (numbuf));
4135
4136 if (verbosity >= 4) {
4137 (void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp ");
4138 (void) snprintf_blkptr(blkbuf + strlen(blkbuf),
4139 sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp);
4140 } else {
4141 blkbuf[0] = '\0';
4142 }
4143
4144 dmu_objset_name(os, osname);
4145
4146 (void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
4147 "%s, %llu objects%s%s\n",
4148 osname, type, (u_longlong_t)dmu_objset_id(os),
4149 (u_longlong_t)dds.dds_creation_txg,
4150 numbuf, (u_longlong_t)usedobjs, blkbuf,
4151 (dds.dds_inconsistent) ? " (inconsistent)" : "");
4152
4153 for (i = 0; i < zopt_object_args; i++) {
4154 obj_start = zopt_object_ranges[i].zor_obj_start;
4155 obj_end = zopt_object_ranges[i].zor_obj_end;
4156 flags = zopt_object_ranges[i].zor_flags;
4157
4158 object = obj_start;
4159 if (object == 0 || obj_start == obj_end)
4160 dump_object(os, object, verbosity, &print_header, NULL,
4161 flags);
4162 else
4163 object--;
4164
4165 while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) &&
4166 object <= obj_end) {
4167 dump_object(os, object, verbosity, &print_header, NULL,
4168 flags);
4169 }
4170 }
4171
4172 if (zopt_object_args > 0) {
4173 (void) printf("\n");
4174 return;
4175 }
4176
4177 if (dump_opt['i'] != 0 || verbosity >= 2)
4178 dump_intent_log(dmu_objset_zil(os));
4179
4180 if (dmu_objset_ds(os) != NULL) {
4181 dsl_dataset_t *ds = dmu_objset_ds(os);
4182 dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
4183 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
4184 !dmu_objset_is_snapshot(os)) {
4185 dump_blkptr_list(&ds->ds_dir->dd_livelist, "Livelist");
4186 if (verify_dd_livelist(os) != 0)
4187 fatal("livelist is incorrect");
4188 }
4189
4190 if (dsl_dataset_remap_deadlist_exists(ds)) {
4191 (void) printf("ds_remap_deadlist:\n");
4192 dump_blkptr_list(&ds->ds_remap_deadlist, "Deadlist");
4193 }
4194 count_ds_mos_objects(ds);
4195 }
4196
4197 if (dmu_objset_ds(os) != NULL)
4198 dump_bookmarks(os, verbosity);
4199
4200 if (verbosity < 2)
4201 return;
4202
4203 if (BP_IS_HOLE(os->os_rootbp))
4204 return;
4205
4206 dump_object(os, 0, verbosity, &print_header, NULL, 0);
4207 object_count = 0;
4208 if (DMU_USERUSED_DNODE(os) != NULL &&
4209 DMU_USERUSED_DNODE(os)->dn_type != 0) {
4210 dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header,
4211 NULL, 0);
4212 dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header,
4213 NULL, 0);
4214 }
4215
4216 if (DMU_PROJECTUSED_DNODE(os) != NULL &&
4217 DMU_PROJECTUSED_DNODE(os)->dn_type != 0)
4218 dump_object(os, DMU_PROJECTUSED_OBJECT, verbosity,
4219 &print_header, NULL, 0);
4220
4221 object = 0;
4222 while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
4223 dump_object(os, object, verbosity, &print_header, &dnode_slots,
4224 0);
4225 object_count++;
4226 total_slots_used += dnode_slots;
4227 max_slot_used = object + dnode_slots - 1;
4228 }
4229
4230 (void) printf("\n");
4231
4232 (void) printf(" Dnode slots:\n");
4233 (void) printf("\tTotal used: %10llu\n",
4234 (u_longlong_t)total_slots_used);
4235 (void) printf("\tMax used: %10llu\n",
4236 (u_longlong_t)max_slot_used);
4237 (void) printf("\tPercent empty: %10lf\n",
4238 (double)(max_slot_used - total_slots_used)*100 /
4239 (double)max_slot_used);
4240 (void) printf("\n");
4241
4242 if (error != ESRCH) {
4243 (void) fprintf(stderr, "dmu_object_next() = %d\n", error);
4244 abort();
4245 }
4246
4247 ASSERT3U(object_count, ==, usedobjs);
4248
4249 if (leaked_objects != 0) {
4250 (void) printf("%d potentially leaked objects detected\n",
4251 leaked_objects);
4252 leaked_objects = 0;
4253 }
4254 }
4255
4256 static void
4257 dump_uberblock(uberblock_t *ub, const char *header, const char *footer)
4258 {
4259 time_t timestamp = ub->ub_timestamp;
4260
4261 (void) printf("%s", header ? header : "");
4262 (void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic);
4263 (void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version);
4264 (void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg);
4265 (void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum);
4266 (void) printf("\ttimestamp = %llu UTC = %s",
4267 (u_longlong_t)ub->ub_timestamp, ctime(&timestamp));
4268
4269 char blkbuf[BP_SPRINTF_LEN];
4270 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4271 (void) printf("\tbp = %s\n", blkbuf);
4272
4273 (void) printf("\tmmp_magic = %016llx\n",
4274 (u_longlong_t)ub->ub_mmp_magic);
4275 if (MMP_VALID(ub)) {
4276 (void) printf("\tmmp_delay = %0llu\n",
4277 (u_longlong_t)ub->ub_mmp_delay);
4278 if (MMP_SEQ_VALID(ub))
4279 (void) printf("\tmmp_seq = %u\n",
4280 (unsigned int) MMP_SEQ(ub));
4281 if (MMP_FAIL_INT_VALID(ub))
4282 (void) printf("\tmmp_fail = %u\n",
4283 (unsigned int) MMP_FAIL_INT(ub));
4284 if (MMP_INTERVAL_VALID(ub))
4285 (void) printf("\tmmp_write = %u\n",
4286 (unsigned int) MMP_INTERVAL(ub));
4287 /* After MMP_* to make summarize_uberblock_mmp cleaner */
4288 (void) printf("\tmmp_valid = %x\n",
4289 (unsigned int) ub->ub_mmp_config & 0xFF);
4290 }
4291
4292 if (dump_opt['u'] >= 4) {
4293 char blkbuf[BP_SPRINTF_LEN];
4294 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4295 (void) printf("\trootbp = %s\n", blkbuf);
4296 }
4297 (void) printf("\tcheckpoint_txg = %llu\n",
4298 (u_longlong_t)ub->ub_checkpoint_txg);
4299
4300 (void) printf("\traidz_reflow state=%u off=%llu\n",
4301 (int)RRSS_GET_STATE(ub),
4302 (u_longlong_t)RRSS_GET_OFFSET(ub));
4303
4304 (void) printf("%s", footer ? footer : "");
4305 }
4306
4307 static void
4308 dump_config(spa_t *spa)
4309 {
4310 dmu_buf_t *db;
4311 size_t nvsize = 0;
4312 int error = 0;
4313
4314
4315 error = dmu_bonus_hold(spa->spa_meta_objset,
4316 spa->spa_config_object, FTAG, &db);
4317
4318 if (error == 0) {
4319 nvsize = *(uint64_t *)db->db_data;
4320 dmu_buf_rele(db, FTAG);
4321
4322 (void) printf("\nMOS Configuration:\n");
4323 dump_packed_nvlist(spa->spa_meta_objset,
4324 spa->spa_config_object, (void *)&nvsize, 1);
4325 } else {
4326 (void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d",
4327 (u_longlong_t)spa->spa_config_object, error);
4328 }
4329 }
4330
4331 static void
4332 dump_cachefile(const char *cachefile)
4333 {
4334 int fd;
4335 struct stat64 statbuf;
4336 char *buf;
4337 nvlist_t *config;
4338
4339 if ((fd = open64(cachefile, O_RDONLY)) < 0) {
4340 (void) printf("cannot open '%s': %s\n", cachefile,
4341 strerror(errno));
4342 zdb_exit(1);
4343 }
4344
4345 if (fstat64(fd, &statbuf) != 0) {
4346 (void) printf("failed to stat '%s': %s\n", cachefile,
4347 strerror(errno));
4348 zdb_exit(1);
4349 }
4350
4351 if ((buf = malloc(statbuf.st_size)) == NULL) {
4352 (void) fprintf(stderr, "failed to allocate %llu bytes\n",
4353 (u_longlong_t)statbuf.st_size);
4354 zdb_exit(1);
4355 }
4356
4357 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
4358 (void) fprintf(stderr, "failed to read %llu bytes\n",
4359 (u_longlong_t)statbuf.st_size);
4360 zdb_exit(1);
4361 }
4362
4363 (void) close(fd);
4364
4365 if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) {
4366 (void) fprintf(stderr, "failed to unpack nvlist\n");
4367 zdb_exit(1);
4368 }
4369
4370 free(buf);
4371
4372 dump_nvlist(config, 0);
4373
4374 nvlist_free(config);
4375 }
4376
4377 /*
4378 * ZFS label nvlist stats
4379 */
4380 typedef struct zdb_nvl_stats {
4381 int zns_list_count;
4382 int zns_leaf_count;
4383 size_t zns_leaf_largest;
4384 size_t zns_leaf_total;
4385 nvlist_t *zns_string;
4386 nvlist_t *zns_uint64;
4387 nvlist_t *zns_boolean;
4388 } zdb_nvl_stats_t;
4389
4390 static void
4391 collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats)
4392 {
4393 nvlist_t *list, **array;
4394 nvpair_t *nvp = NULL;
4395 const char *name;
4396 uint_t i, items;
4397
4398 stats->zns_list_count++;
4399
4400 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4401 name = nvpair_name(nvp);
4402
4403 switch (nvpair_type(nvp)) {
4404 case DATA_TYPE_STRING:
4405 fnvlist_add_string(stats->zns_string, name,
4406 fnvpair_value_string(nvp));
4407 break;
4408 case DATA_TYPE_UINT64:
4409 fnvlist_add_uint64(stats->zns_uint64, name,
4410 fnvpair_value_uint64(nvp));
4411 break;
4412 case DATA_TYPE_BOOLEAN:
4413 fnvlist_add_boolean(stats->zns_boolean, name);
4414 break;
4415 case DATA_TYPE_NVLIST:
4416 if (nvpair_value_nvlist(nvp, &list) == 0)
4417 collect_nvlist_stats(list, stats);
4418 break;
4419 case DATA_TYPE_NVLIST_ARRAY:
4420 if (nvpair_value_nvlist_array(nvp, &array, &items) != 0)
4421 break;
4422
4423 for (i = 0; i < items; i++) {
4424 collect_nvlist_stats(array[i], stats);
4425
4426 /* collect stats on leaf vdev */
4427 if (strcmp(name, "children") == 0) {
4428 size_t size;
4429
4430 (void) nvlist_size(array[i], &size,
4431 NV_ENCODE_XDR);
4432 stats->zns_leaf_total += size;
4433 if (size > stats->zns_leaf_largest)
4434 stats->zns_leaf_largest = size;
4435 stats->zns_leaf_count++;
4436 }
4437 }
4438 break;
4439 default:
4440 (void) printf("skip type %d!\n", (int)nvpair_type(nvp));
4441 }
4442 }
4443 }
4444
4445 static void
4446 dump_nvlist_stats(nvlist_t *nvl, size_t cap)
4447 {
4448 zdb_nvl_stats_t stats = { 0 };
4449 size_t size, sum = 0, total;
4450 size_t noise;
4451
4452 /* requires nvlist with non-unique names for stat collection */
4453 VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0));
4454 VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0));
4455 VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0));
4456 VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR));
4457
4458 (void) printf("\n\nZFS Label NVList Config Stats:\n");
4459
4460 VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR));
4461 (void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4462 (int)total, (int)(cap - total), 100.0 * total / cap);
4463
4464 collect_nvlist_stats(nvl, &stats);
4465
4466 VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR));
4467 size -= noise;
4468 sum += size;
4469 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4470 (int)fnvlist_num_pairs(stats.zns_uint64),
4471 (int)size, 100.0 * size / total);
4472
4473 VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR));
4474 size -= noise;
4475 sum += size;
4476 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4477 (int)fnvlist_num_pairs(stats.zns_string),
4478 (int)size, 100.0 * size / total);
4479
4480 VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR));
4481 size -= noise;
4482 sum += size;
4483 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4484 (int)fnvlist_num_pairs(stats.zns_boolean),
4485 (int)size, 100.0 * size / total);
4486
4487 size = total - sum; /* treat remainder as nvlist overhead */
4488 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4489 stats.zns_list_count, (int)size, 100.0 * size / total);
4490
4491 if (stats.zns_leaf_count > 0) {
4492 size_t average = stats.zns_leaf_total / stats.zns_leaf_count;
4493
4494 (void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4495 stats.zns_leaf_count, (int)average);
4496 (void) printf("%24d bytes largest\n",
4497 (int)stats.zns_leaf_largest);
4498
4499 if (dump_opt['l'] >= 3 && average > 0)
4500 (void) printf(" space for %d additional leaf vdevs\n",
4501 (int)((cap - total) / average));
4502 }
4503 (void) printf("\n");
4504
4505 nvlist_free(stats.zns_string);
4506 nvlist_free(stats.zns_uint64);
4507 nvlist_free(stats.zns_boolean);
4508 }
4509
4510 typedef struct cksum_record {
4511 zio_cksum_t cksum;
4512 boolean_t labels[VDEV_LABELS];
4513 avl_node_t link;
4514 } cksum_record_t;
4515
4516 static int
4517 cksum_record_compare(const void *x1, const void *x2)
4518 {
4519 const cksum_record_t *l = (cksum_record_t *)x1;
4520 const cksum_record_t *r = (cksum_record_t *)x2;
4521 int arraysize = ARRAY_SIZE(l->cksum.zc_word);
4522 int difference = 0;
4523
4524 for (int i = 0; i < arraysize; i++) {
4525 difference = TREE_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]);
4526 if (difference)
4527 break;
4528 }
4529
4530 return (difference);
4531 }
4532
4533 static cksum_record_t *
4534 cksum_record_alloc(zio_cksum_t *cksum, int l)
4535 {
4536 cksum_record_t *rec;
4537
4538 rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL);
4539 rec->cksum = *cksum;
4540 rec->labels[l] = B_TRUE;
4541
4542 return (rec);
4543 }
4544
4545 static cksum_record_t *
4546 cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum)
4547 {
4548 cksum_record_t lookup = { .cksum = *cksum };
4549 avl_index_t where;
4550
4551 return (avl_find(tree, &lookup, &where));
4552 }
4553
4554 static cksum_record_t *
4555 cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l)
4556 {
4557 cksum_record_t *rec;
4558
4559 rec = cksum_record_lookup(tree, cksum);
4560 if (rec) {
4561 rec->labels[l] = B_TRUE;
4562 } else {
4563 rec = cksum_record_alloc(cksum, l);
4564 avl_add(tree, rec);
4565 }
4566
4567 return (rec);
4568 }
4569
4570 static int
4571 first_label(cksum_record_t *rec)
4572 {
4573 for (int i = 0; i < VDEV_LABELS; i++)
4574 if (rec->labels[i])
4575 return (i);
4576
4577 return (-1);
4578 }
4579
4580 static void
4581 print_label_numbers(const char *prefix, const cksum_record_t *rec)
4582 {
4583 fputs(prefix, stdout);
4584 for (int i = 0; i < VDEV_LABELS; i++)
4585 if (rec->labels[i] == B_TRUE)
4586 printf("%d ", i);
4587 putchar('\n');
4588 }
4589
4590 #define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4591
4592 typedef struct zdb_label {
4593 vdev_label_t label;
4594 uint64_t label_offset;
4595 nvlist_t *config_nv;
4596 cksum_record_t *config;
4597 cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT];
4598 boolean_t header_printed;
4599 boolean_t read_failed;
4600 boolean_t cksum_valid;
4601 } zdb_label_t;
4602
4603 static void
4604 print_label_header(zdb_label_t *label, int l)
4605 {
4606
4607 if (dump_opt['q'])
4608 return;
4609
4610 if (label->header_printed == B_TRUE)
4611 return;
4612
4613 (void) printf("------------------------------------\n");
4614 (void) printf("LABEL %d %s\n", l,
4615 label->cksum_valid ? "" : "(Bad label cksum)");
4616 (void) printf("------------------------------------\n");
4617
4618 label->header_printed = B_TRUE;
4619 }
4620
4621 static void
4622 print_l2arc_header(void)
4623 {
4624 (void) printf("------------------------------------\n");
4625 (void) printf("L2ARC device header\n");
4626 (void) printf("------------------------------------\n");
4627 }
4628
4629 static void
4630 print_l2arc_log_blocks(void)
4631 {
4632 (void) printf("------------------------------------\n");
4633 (void) printf("L2ARC device log blocks\n");
4634 (void) printf("------------------------------------\n");
4635 }
4636
4637 static void
4638 dump_l2arc_log_entries(uint64_t log_entries,
4639 l2arc_log_ent_phys_t *le, uint64_t i)
4640 {
4641 for (int j = 0; j < log_entries; j++) {
4642 dva_t dva = le[j].le_dva;
4643 (void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4644 "vdev: %llu, offset: %llu\n",
4645 (u_longlong_t)i, j + 1,
4646 (u_longlong_t)DVA_GET_ASIZE(&dva),
4647 (u_longlong_t)DVA_GET_VDEV(&dva),
4648 (u_longlong_t)DVA_GET_OFFSET(&dva));
4649 (void) printf("|\t\t\t\tbirth: %llu\n",
4650 (u_longlong_t)le[j].le_birth);
4651 (void) printf("|\t\t\t\tlsize: %llu\n",
4652 (u_longlong_t)L2BLK_GET_LSIZE((&le[j])->le_prop));
4653 (void) printf("|\t\t\t\tpsize: %llu\n",
4654 (u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop));
4655 (void) printf("|\t\t\t\tcompr: %llu\n",
4656 (u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop));
4657 (void) printf("|\t\t\t\tcomplevel: %llu\n",
4658 (u_longlong_t)(&le[j])->le_complevel);
4659 (void) printf("|\t\t\t\ttype: %llu\n",
4660 (u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop));
4661 (void) printf("|\t\t\t\tprotected: %llu\n",
4662 (u_longlong_t)L2BLK_GET_PROTECTED((&le[j])->le_prop));
4663 (void) printf("|\t\t\t\tprefetch: %llu\n",
4664 (u_longlong_t)L2BLK_GET_PREFETCH((&le[j])->le_prop));
4665 (void) printf("|\t\t\t\taddress: %llu\n",
4666 (u_longlong_t)le[j].le_daddr);
4667 (void) printf("|\t\t\t\tARC state: %llu\n",
4668 (u_longlong_t)L2BLK_GET_STATE((&le[j])->le_prop));
4669 (void) printf("|\n");
4670 }
4671 (void) printf("\n");
4672 }
4673
4674 static void
4675 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t *lbps)
4676 {
4677 (void) printf("|\t\tdaddr: %llu\n", (u_longlong_t)lbps->lbp_daddr);
4678 (void) printf("|\t\tpayload_asize: %llu\n",
4679 (u_longlong_t)lbps->lbp_payload_asize);
4680 (void) printf("|\t\tpayload_start: %llu\n",
4681 (u_longlong_t)lbps->lbp_payload_start);
4682 (void) printf("|\t\tlsize: %llu\n",
4683 (u_longlong_t)L2BLK_GET_LSIZE(lbps->lbp_prop));
4684 (void) printf("|\t\tasize: %llu\n",
4685 (u_longlong_t)L2BLK_GET_PSIZE(lbps->lbp_prop));
4686 (void) printf("|\t\tcompralgo: %llu\n",
4687 (u_longlong_t)L2BLK_GET_COMPRESS(lbps->lbp_prop));
4688 (void) printf("|\t\tcksumalgo: %llu\n",
4689 (u_longlong_t)L2BLK_GET_CHECKSUM(lbps->lbp_prop));
4690 (void) printf("|\n\n");
4691 }
4692
4693 static void
4694 dump_l2arc_log_blocks(int fd, const l2arc_dev_hdr_phys_t *l2dhdr,
4695 l2arc_dev_hdr_phys_t *rebuild)
4696 {
4697 l2arc_log_blk_phys_t this_lb;
4698 uint64_t asize;
4699 l2arc_log_blkptr_t lbps[2];
4700 zio_cksum_t cksum;
4701 int failed = 0;
4702 l2arc_dev_t dev;
4703
4704 if (!dump_opt['q'])
4705 print_l2arc_log_blocks();
4706 memcpy(lbps, l2dhdr->dh_start_lbps, sizeof (lbps));
4707
4708 dev.l2ad_evict = l2dhdr->dh_evict;
4709 dev.l2ad_start = l2dhdr->dh_start;
4710 dev.l2ad_end = l2dhdr->dh_end;
4711
4712 if (l2dhdr->dh_start_lbps[0].lbp_daddr == 0) {
4713 /* no log blocks to read */
4714 if (!dump_opt['q']) {
4715 (void) printf("No log blocks to read\n");
4716 (void) printf("\n");
4717 }
4718 return;
4719 } else {
4720 dev.l2ad_hand = lbps[0].lbp_daddr +
4721 L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4722 }
4723
4724 dev.l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST);
4725
4726 for (;;) {
4727 if (!l2arc_log_blkptr_valid(&dev, &lbps[0]))
4728 break;
4729
4730 /* L2BLK_GET_PSIZE returns aligned size for log blocks */
4731 asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4732 if (pread64(fd, &this_lb, asize, lbps[0].lbp_daddr) != asize) {
4733 if (!dump_opt['q']) {
4734 (void) printf("Error while reading next log "
4735 "block\n\n");
4736 }
4737 break;
4738 }
4739
4740 fletcher_4_native_varsize(&this_lb, asize, &cksum);
4741 if (!ZIO_CHECKSUM_EQUAL(cksum, lbps[0].lbp_cksum)) {
4742 failed++;
4743 if (!dump_opt['q']) {
4744 (void) printf("Invalid cksum\n");
4745 dump_l2arc_log_blkptr(&lbps[0]);
4746 }
4747 break;
4748 }
4749
4750 switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) {
4751 case ZIO_COMPRESS_OFF:
4752 break;
4753 default: {
4754 abd_t *abd = abd_alloc_linear(asize, B_TRUE);
4755 abd_copy_from_buf_off(abd, &this_lb, 0, asize);
4756 abd_t dabd;
4757 abd_get_from_buf_struct(&dabd, &this_lb,
4758 sizeof (this_lb));
4759 int err = zio_decompress_data(L2BLK_GET_COMPRESS(
4760 (&lbps[0])->lbp_prop), abd, &dabd,
4761 asize, sizeof (this_lb), NULL);
4762 abd_free(&dabd);
4763 abd_free(abd);
4764 if (err != 0) {
4765 (void) printf("L2ARC block decompression "
4766 "failed\n");
4767 goto out;
4768 }
4769 break;
4770 }
4771 }
4772
4773 if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
4774 byteswap_uint64_array(&this_lb, sizeof (this_lb));
4775 if (this_lb.lb_magic != L2ARC_LOG_BLK_MAGIC) {
4776 if (!dump_opt['q'])
4777 (void) printf("Invalid log block magic\n\n");
4778 break;
4779 }
4780
4781 rebuild->dh_lb_count++;
4782 rebuild->dh_lb_asize += asize;
4783 if (dump_opt['l'] > 1 && !dump_opt['q']) {
4784 (void) printf("lb[%4llu]\tmagic: %llu\n",
4785 (u_longlong_t)rebuild->dh_lb_count,
4786 (u_longlong_t)this_lb.lb_magic);
4787 dump_l2arc_log_blkptr(&lbps[0]);
4788 }
4789
4790 if (dump_opt['l'] > 2 && !dump_opt['q'])
4791 dump_l2arc_log_entries(l2dhdr->dh_log_entries,
4792 this_lb.lb_entries,
4793 rebuild->dh_lb_count);
4794
4795 if (l2arc_range_check_overlap(lbps[1].lbp_payload_start,
4796 lbps[0].lbp_payload_start, dev.l2ad_evict) &&
4797 !dev.l2ad_first)
4798 break;
4799
4800 lbps[0] = lbps[1];
4801 lbps[1] = this_lb.lb_prev_lbp;
4802 }
4803 out:
4804 if (!dump_opt['q']) {
4805 (void) printf("log_blk_count:\t %llu with valid cksum\n",
4806 (u_longlong_t)rebuild->dh_lb_count);
4807 (void) printf("\t\t %d with invalid cksum\n", failed);
4808 (void) printf("log_blk_asize:\t %llu\n\n",
4809 (u_longlong_t)rebuild->dh_lb_asize);
4810 }
4811 }
4812
4813 static int
4814 dump_l2arc_header(int fd)
4815 {
4816 l2arc_dev_hdr_phys_t l2dhdr = {0}, rebuild = {0};
4817 int error = B_FALSE;
4818
4819 if (pread64(fd, &l2dhdr, sizeof (l2dhdr),
4820 VDEV_LABEL_START_SIZE) != sizeof (l2dhdr)) {
4821 error = B_TRUE;
4822 } else {
4823 if (l2dhdr.dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC))
4824 byteswap_uint64_array(&l2dhdr, sizeof (l2dhdr));
4825
4826 if (l2dhdr.dh_magic != L2ARC_DEV_HDR_MAGIC)
4827 error = B_TRUE;
4828 }
4829
4830 if (error) {
4831 (void) printf("L2ARC device header not found\n\n");
4832 /* Do not return an error here for backward compatibility */
4833 return (0);
4834 } else if (!dump_opt['q']) {
4835 print_l2arc_header();
4836
4837 (void) printf(" magic: %llu\n",
4838 (u_longlong_t)l2dhdr.dh_magic);
4839 (void) printf(" version: %llu\n",
4840 (u_longlong_t)l2dhdr.dh_version);
4841 (void) printf(" pool_guid: %llu\n",
4842 (u_longlong_t)l2dhdr.dh_spa_guid);
4843 (void) printf(" flags: %llu\n",
4844 (u_longlong_t)l2dhdr.dh_flags);
4845 (void) printf(" start_lbps[0]: %llu\n",
4846 (u_longlong_t)
4847 l2dhdr.dh_start_lbps[0].lbp_daddr);
4848 (void) printf(" start_lbps[1]: %llu\n",
4849 (u_longlong_t)
4850 l2dhdr.dh_start_lbps[1].lbp_daddr);
4851 (void) printf(" log_blk_ent: %llu\n",
4852 (u_longlong_t)l2dhdr.dh_log_entries);
4853 (void) printf(" start: %llu\n",
4854 (u_longlong_t)l2dhdr.dh_start);
4855 (void) printf(" end: %llu\n",
4856 (u_longlong_t)l2dhdr.dh_end);
4857 (void) printf(" evict: %llu\n",
4858 (u_longlong_t)l2dhdr.dh_evict);
4859 (void) printf(" lb_asize_refcount: %llu\n",
4860 (u_longlong_t)l2dhdr.dh_lb_asize);
4861 (void) printf(" lb_count_refcount: %llu\n",
4862 (u_longlong_t)l2dhdr.dh_lb_count);
4863 (void) printf(" trim_action_time: %llu\n",
4864 (u_longlong_t)l2dhdr.dh_trim_action_time);
4865 (void) printf(" trim_state: %llu\n\n",
4866 (u_longlong_t)l2dhdr.dh_trim_state);
4867 }
4868
4869 dump_l2arc_log_blocks(fd, &l2dhdr, &rebuild);
4870 /*
4871 * The total aligned size of log blocks and the number of log blocks
4872 * reported in the header of the device may be less than what zdb
4873 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
4874 * This happens because dump_l2arc_log_blocks() lacks the memory
4875 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
4876 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
4877 * and dh_lb_count will be lower to begin with than what exists on the
4878 * device. This is normal and zdb should not exit with an error. The
4879 * opposite case should never happen though, the values reported in the
4880 * header should never be higher than what dump_l2arc_log_blocks() and
4881 * l2arc_rebuild() report. If this happens there is a leak in the
4882 * accounting of log blocks.
4883 */
4884 if (l2dhdr.dh_lb_asize > rebuild.dh_lb_asize ||
4885 l2dhdr.dh_lb_count > rebuild.dh_lb_count)
4886 return (1);
4887
4888 return (0);
4889 }
4890
4891 static void
4892 dump_config_from_label(zdb_label_t *label, size_t buflen, int l)
4893 {
4894 if (dump_opt['q'])
4895 return;
4896
4897 if ((dump_opt['l'] < 3) && (first_label(label->config) != l))
4898 return;
4899
4900 print_label_header(label, l);
4901 dump_nvlist(label->config_nv, 4);
4902 print_label_numbers(" labels = ", label->config);
4903
4904 if (dump_opt['l'] >= 2)
4905 dump_nvlist_stats(label->config_nv, buflen);
4906 }
4907
4908 #define ZDB_MAX_UB_HEADER_SIZE 32
4909
4910 static void
4911 dump_label_uberblocks(zdb_label_t *label, uint64_t ashift, int label_num)
4912 {
4913
4914 vdev_t vd;
4915 char header[ZDB_MAX_UB_HEADER_SIZE];
4916
4917 vd.vdev_ashift = ashift;
4918 vd.vdev_top = &vd;
4919
4920 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
4921 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
4922 uberblock_t *ub = (void *)((char *)&label->label + uoff);
4923 cksum_record_t *rec = label->uberblocks[i];
4924
4925 if (rec == NULL) {
4926 if (dump_opt['u'] >= 2) {
4927 print_label_header(label, label_num);
4928 (void) printf(" Uberblock[%d] invalid\n", i);
4929 }
4930 continue;
4931 }
4932
4933 if ((dump_opt['u'] < 3) && (first_label(rec) != label_num))
4934 continue;
4935
4936 if ((dump_opt['u'] < 4) &&
4937 (ub->ub_mmp_magic == MMP_MAGIC) && ub->ub_mmp_delay &&
4938 (i >= VDEV_UBERBLOCK_COUNT(&vd) - MMP_BLOCKS_PER_LABEL))
4939 continue;
4940
4941 print_label_header(label, label_num);
4942 (void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE,
4943 " Uberblock[%d]\n", i);
4944 dump_uberblock(ub, header, "");
4945 print_label_numbers(" labels = ", rec);
4946 }
4947 }
4948
4949 static char curpath[PATH_MAX];
4950
4951 /*
4952 * Iterate through the path components, recursively passing
4953 * current one's obj and remaining path until we find the obj
4954 * for the last one.
4955 */
4956 static int
4957 dump_path_impl(objset_t *os, uint64_t obj, char *name, uint64_t *retobj)
4958 {
4959 int err;
4960 boolean_t header = B_TRUE;
4961 uint64_t child_obj;
4962 char *s;
4963 dmu_buf_t *db;
4964 dmu_object_info_t doi;
4965
4966 if ((s = strchr(name, '/')) != NULL)
4967 *s = '\0';
4968 err = zap_lookup(os, obj, name, 8, 1, &child_obj);
4969
4970 (void) strlcat(curpath, name, sizeof (curpath));
4971
4972 if (err != 0) {
4973 (void) fprintf(stderr, "failed to lookup %s: %s\n",
4974 curpath, strerror(err));
4975 return (err);
4976 }
4977
4978 child_obj = ZFS_DIRENT_OBJ(child_obj);
4979 err = sa_buf_hold(os, child_obj, FTAG, &db);
4980 if (err != 0) {
4981 (void) fprintf(stderr,
4982 "failed to get SA dbuf for obj %llu: %s\n",
4983 (u_longlong_t)child_obj, strerror(err));
4984 return (EINVAL);
4985 }
4986 dmu_object_info_from_db(db, &doi);
4987 sa_buf_rele(db, FTAG);
4988
4989 if (doi.doi_bonus_type != DMU_OT_SA &&
4990 doi.doi_bonus_type != DMU_OT_ZNODE) {
4991 (void) fprintf(stderr, "invalid bonus type %d for obj %llu\n",
4992 doi.doi_bonus_type, (u_longlong_t)child_obj);
4993 return (EINVAL);
4994 }
4995
4996 if (dump_opt['v'] > 6) {
4997 (void) printf("obj=%llu %s type=%d bonustype=%d\n",
4998 (u_longlong_t)child_obj, curpath, doi.doi_type,
4999 doi.doi_bonus_type);
5000 }
5001
5002 (void) strlcat(curpath, "/", sizeof (curpath));
5003
5004 switch (doi.doi_type) {
5005 case DMU_OT_DIRECTORY_CONTENTS:
5006 if (s != NULL && *(s + 1) != '\0')
5007 return (dump_path_impl(os, child_obj, s + 1, retobj));
5008 zfs_fallthrough;
5009 case DMU_OT_PLAIN_FILE_CONTENTS:
5010 if (retobj != NULL) {
5011 *retobj = child_obj;
5012 } else {
5013 dump_object(os, child_obj, dump_opt['v'], &header,
5014 NULL, 0);
5015 }
5016 return (0);
5017 default:
5018 (void) fprintf(stderr, "object %llu has non-file/directory "
5019 "type %d\n", (u_longlong_t)obj, doi.doi_type);
5020 break;
5021 }
5022
5023 return (EINVAL);
5024 }
5025
5026 /*
5027 * Dump the blocks for the object specified by path inside the dataset.
5028 */
5029 static int
5030 dump_path(char *ds, char *path, uint64_t *retobj)
5031 {
5032 int err;
5033 objset_t *os;
5034 uint64_t root_obj;
5035
5036 err = open_objset(ds, FTAG, &os);
5037 if (err != 0)
5038 return (err);
5039
5040 err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj);
5041 if (err != 0) {
5042 (void) fprintf(stderr, "can't lookup root znode: %s\n",
5043 strerror(err));
5044 close_objset(os, FTAG);
5045 return (EINVAL);
5046 }
5047
5048 (void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds);
5049
5050 err = dump_path_impl(os, root_obj, path, retobj);
5051
5052 close_objset(os, FTAG);
5053 return (err);
5054 }
5055
5056 static int
5057 dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
5058 {
5059 const char *p = (const char *)buf;
5060 ssize_t nwritten;
5061
5062 (void) os;
5063 (void) arg;
5064
5065 /* Write the data out, handling short writes and signals. */
5066 while ((nwritten = write(STDOUT_FILENO, p, len)) < len) {
5067 if (nwritten < 0) {
5068 if (errno == EINTR)
5069 continue;
5070 return (errno);
5071 }
5072 p += nwritten;
5073 len -= nwritten;
5074 }
5075
5076 return (0);
5077 }
5078
5079 static void
5080 dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
5081 {
5082 boolean_t embed = B_FALSE;
5083 boolean_t large_block = B_FALSE;
5084 boolean_t compress = B_FALSE;
5085 boolean_t raw = B_FALSE;
5086
5087 const char *c;
5088 for (c = flagstr; c != NULL && *c != '\0'; c++) {
5089 switch (*c) {
5090 case 'e':
5091 embed = B_TRUE;
5092 break;
5093 case 'L':
5094 large_block = B_TRUE;
5095 break;
5096 case 'c':
5097 compress = B_TRUE;
5098 break;
5099 case 'w':
5100 raw = B_TRUE;
5101 break;
5102 default:
5103 fprintf(stderr, "dump_backup: invalid flag "
5104 "'%c'\n", *c);
5105 return;
5106 }
5107 }
5108
5109 if (isatty(STDOUT_FILENO)) {
5110 fprintf(stderr, "dump_backup: stream cannot be written "
5111 "to a terminal\n");
5112 return;
5113 }
5114
5115 offset_t off = 0;
5116 dmu_send_outparams_t out = {
5117 .dso_outfunc = dump_backup_bytes,
5118 .dso_dryrun = B_FALSE,
5119 };
5120
5121 int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed,
5122 large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO,
5123 &off, &out);
5124 if (err != 0) {
5125 fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",
5126 strerror(err));
5127 return;
5128 }
5129 }
5130
5131 static int
5132 zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile)
5133 {
5134 int err = 0;
5135 uint64_t size, readsize, oursize, offset;
5136 ssize_t writesize;
5137 sa_handle_t *hdl;
5138
5139 (void) printf("Copying object %" PRIu64 " to file %s\n", srcobj,
5140 destfile);
5141
5142 VERIFY3P(os, ==, sa_os);
5143 if ((err = sa_handle_get(os, srcobj, NULL, SA_HDL_PRIVATE, &hdl))) {
5144 (void) printf("Failed to get handle for SA znode\n");
5145 return (err);
5146 }
5147 if ((err = sa_lookup(hdl, sa_attr_table[ZPL_SIZE], &size, 8))) {
5148 (void) sa_handle_destroy(hdl);
5149 return (err);
5150 }
5151 (void) sa_handle_destroy(hdl);
5152
5153 (void) printf("Object %" PRIu64 " is %" PRIu64 " bytes\n", srcobj,
5154 size);
5155 if (size == 0) {
5156 return (EINVAL);
5157 }
5158
5159 int fd = open(destfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
5160 if (fd == -1)
5161 return (errno);
5162 /*
5163 * We cap the size at 1 mebibyte here to prevent
5164 * allocation failures and nigh-infinite printing if the
5165 * object is extremely large.
5166 */
5167 oursize = MIN(size, 1 << 20);
5168 offset = 0;
5169 char *buf = kmem_alloc(oursize, KM_NOSLEEP);
5170 if (buf == NULL) {
5171 (void) close(fd);
5172 return (ENOMEM);
5173 }
5174
5175 while (offset < size) {
5176 readsize = MIN(size - offset, 1 << 20);
5177 err = dmu_read(os, srcobj, offset, readsize, buf, 0);
5178 if (err != 0) {
5179 (void) printf("got error %u from dmu_read\n", err);
5180 kmem_free(buf, oursize);
5181 (void) close(fd);
5182 return (err);
5183 }
5184 if (dump_opt['v'] > 3) {
5185 (void) printf("Read offset=%" PRIu64 " size=%" PRIu64
5186 " error=%d\n", offset, readsize, err);
5187 }
5188
5189 writesize = write(fd, buf, readsize);
5190 if (writesize < 0) {
5191 err = errno;
5192 break;
5193 } else if (writesize != readsize) {
5194 /* Incomplete write */
5195 (void) fprintf(stderr, "Short write, only wrote %llu of"
5196 " %" PRIu64 " bytes, exiting...\n",
5197 (u_longlong_t)writesize, readsize);
5198 break;
5199 }
5200
5201 offset += readsize;
5202 }
5203
5204 (void) close(fd);
5205
5206 if (buf != NULL)
5207 kmem_free(buf, oursize);
5208
5209 return (err);
5210 }
5211
5212 static boolean_t
5213 label_cksum_valid(vdev_label_t *label, uint64_t offset)
5214 {
5215 zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
5216 zio_cksum_t expected_cksum;
5217 zio_cksum_t actual_cksum;
5218 zio_cksum_t verifier;
5219 zio_eck_t *eck;
5220 int byteswap;
5221
5222 void *data = (char *)label + offsetof(vdev_label_t, vl_vdev_phys);
5223 eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1;
5224
5225 offset += offsetof(vdev_label_t, vl_vdev_phys);
5226 ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
5227
5228 byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
5229 if (byteswap)
5230 byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
5231
5232 expected_cksum = eck->zec_cksum;
5233 eck->zec_cksum = verifier;
5234
5235 abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE);
5236 ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum);
5237 abd_free(abd);
5238
5239 if (byteswap)
5240 byteswap_uint64_array(&expected_cksum, sizeof (zio_cksum_t));
5241
5242 if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
5243 return (B_TRUE);
5244
5245 return (B_FALSE);
5246 }
5247
5248 static int
5249 dump_label(const char *dev)
5250 {
5251 char path[MAXPATHLEN];
5252 zdb_label_t labels[VDEV_LABELS] = {{{{0}}}};
5253 uint64_t psize, ashift, l2cache;
5254 struct stat64 statbuf;
5255 boolean_t config_found = B_FALSE;
5256 boolean_t error = B_FALSE;
5257 boolean_t read_l2arc_header = B_FALSE;
5258 avl_tree_t config_tree;
5259 avl_tree_t uberblock_tree;
5260 void *node, *cookie;
5261 int fd;
5262
5263 /*
5264 * Check if we were given absolute path and use it as is.
5265 * Otherwise if the provided vdev name doesn't point to a file,
5266 * try prepending expected disk paths and partition numbers.
5267 */
5268 (void) strlcpy(path, dev, sizeof (path));
5269 if (dev[0] != '/' && stat64(path, &statbuf) != 0) {
5270 int error;
5271
5272 error = zfs_resolve_shortname(dev, path, MAXPATHLEN);
5273 if (error == 0 && zfs_dev_is_whole_disk(path)) {
5274 if (zfs_append_partition(path, MAXPATHLEN) == -1)
5275 error = ENOENT;
5276 }
5277
5278 if (error || (stat64(path, &statbuf) != 0)) {
5279 (void) printf("failed to find device %s, try "
5280 "specifying absolute path instead\n", dev);
5281 return (1);
5282 }
5283 }
5284
5285 if ((fd = open64(path, O_RDONLY)) < 0) {
5286 (void) printf("cannot open '%s': %s\n", path, strerror(errno));
5287 zdb_exit(1);
5288 }
5289
5290 if (fstat64_blk(fd, &statbuf) != 0) {
5291 (void) printf("failed to stat '%s': %s\n", path,
5292 strerror(errno));
5293 (void) close(fd);
5294 zdb_exit(1);
5295 }
5296
5297 if (S_ISBLK(statbuf.st_mode) && zfs_dev_flush(fd) != 0)
5298 (void) printf("failed to invalidate cache '%s' : %s\n", path,
5299 strerror(errno));
5300
5301 avl_create(&config_tree, cksum_record_compare,
5302 sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5303 avl_create(&uberblock_tree, cksum_record_compare,
5304 sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5305
5306 psize = statbuf.st_size;
5307 psize = P2ALIGN_TYPED(psize, sizeof (vdev_label_t), uint64_t);
5308 ashift = SPA_MINBLOCKSHIFT;
5309
5310 /*
5311 * 1. Read the label from disk
5312 * 2. Verify label cksum
5313 * 3. Unpack the configuration and insert in config tree.
5314 * 4. Traverse all uberblocks and insert in uberblock tree.
5315 */
5316 for (int l = 0; l < VDEV_LABELS; l++) {
5317 zdb_label_t *label = &labels[l];
5318 char *buf = label->label.vl_vdev_phys.vp_nvlist;
5319 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5320 nvlist_t *config;
5321 cksum_record_t *rec;
5322 zio_cksum_t cksum;
5323 vdev_t vd;
5324
5325 label->label_offset = vdev_label_offset(psize, l, 0);
5326
5327 if (pread64(fd, &label->label, sizeof (label->label),
5328 label->label_offset) != sizeof (label->label)) {
5329 if (!dump_opt['q'])
5330 (void) printf("failed to read label %d\n", l);
5331 label->read_failed = B_TRUE;
5332 error = B_TRUE;
5333 continue;
5334 }
5335
5336 label->read_failed = B_FALSE;
5337 label->cksum_valid = label_cksum_valid(&label->label,
5338 label->label_offset);
5339
5340 if (nvlist_unpack(buf, buflen, &config, 0) == 0) {
5341 nvlist_t *vdev_tree = NULL;
5342 size_t size;
5343
5344 if ((nvlist_lookup_nvlist(config,
5345 ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) ||
5346 (nvlist_lookup_uint64(vdev_tree,
5347 ZPOOL_CONFIG_ASHIFT, &ashift) != 0))
5348 ashift = SPA_MINBLOCKSHIFT;
5349
5350 if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0)
5351 size = buflen;
5352
5353 /* If the device is a cache device read the header. */
5354 if (!read_l2arc_header) {
5355 if (nvlist_lookup_uint64(config,
5356 ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 &&
5357 l2cache == POOL_STATE_L2CACHE) {
5358 read_l2arc_header = B_TRUE;
5359 }
5360 }
5361
5362 fletcher_4_native_varsize(buf, size, &cksum);
5363 rec = cksum_record_insert(&config_tree, &cksum, l);
5364
5365 label->config = rec;
5366 label->config_nv = config;
5367 config_found = B_TRUE;
5368 } else {
5369 error = B_TRUE;
5370 }
5371
5372 vd.vdev_ashift = ashift;
5373 vd.vdev_top = &vd;
5374
5375 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5376 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5377 uberblock_t *ub = (void *)((char *)label + uoff);
5378
5379 if (uberblock_verify(ub))
5380 continue;
5381
5382 fletcher_4_native_varsize(ub, sizeof (*ub), &cksum);
5383 rec = cksum_record_insert(&uberblock_tree, &cksum, l);
5384
5385 label->uberblocks[i] = rec;
5386 }
5387 }
5388
5389 /*
5390 * Dump the label and uberblocks.
5391 */
5392 for (int l = 0; l < VDEV_LABELS; l++) {
5393 zdb_label_t *label = &labels[l];
5394 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5395
5396 if (label->read_failed == B_TRUE)
5397 continue;
5398
5399 if (label->config_nv) {
5400 dump_config_from_label(label, buflen, l);
5401 } else {
5402 if (!dump_opt['q'])
5403 (void) printf("failed to unpack label %d\n", l);
5404 }
5405
5406 if (dump_opt['u'])
5407 dump_label_uberblocks(label, ashift, l);
5408
5409 nvlist_free(label->config_nv);
5410 }
5411
5412 /*
5413 * Dump the L2ARC header, if existent.
5414 */
5415 if (read_l2arc_header)
5416 error |= dump_l2arc_header(fd);
5417
5418 cookie = NULL;
5419 while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL)
5420 umem_free(node, sizeof (cksum_record_t));
5421
5422 cookie = NULL;
5423 while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL)
5424 umem_free(node, sizeof (cksum_record_t));
5425
5426 avl_destroy(&config_tree);
5427 avl_destroy(&uberblock_tree);
5428
5429 (void) close(fd);
5430
5431 return (config_found == B_FALSE ? 2 :
5432 (error == B_TRUE ? 1 : 0));
5433 }
5434
5435 static uint64_t dataset_feature_count[SPA_FEATURES];
5436 static uint64_t global_feature_count[SPA_FEATURES];
5437 static uint64_t remap_deadlist_count = 0;
5438
5439 static int
5440 dump_one_objset(const char *dsname, void *arg)
5441 {
5442 (void) arg;
5443 int error;
5444 objset_t *os;
5445 spa_feature_t f;
5446
5447 error = open_objset(dsname, FTAG, &os);
5448 if (error != 0)
5449 return (0);
5450
5451 for (f = 0; f < SPA_FEATURES; f++) {
5452 if (!dsl_dataset_feature_is_active(dmu_objset_ds(os), f))
5453 continue;
5454 ASSERT(spa_feature_table[f].fi_flags &
5455 ZFEATURE_FLAG_PER_DATASET);
5456 dataset_feature_count[f]++;
5457 }
5458
5459 if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os))) {
5460 remap_deadlist_count++;
5461 }
5462
5463 for (dsl_bookmark_node_t *dbn =
5464 avl_first(&dmu_objset_ds(os)->ds_bookmarks); dbn != NULL;
5465 dbn = AVL_NEXT(&dmu_objset_ds(os)->ds_bookmarks, dbn)) {
5466 mos_obj_refd(dbn->dbn_phys.zbm_redaction_obj);
5467 if (dbn->dbn_phys.zbm_redaction_obj != 0) {
5468 global_feature_count[
5469 SPA_FEATURE_REDACTION_BOOKMARKS]++;
5470 objset_t *mos = os->os_spa->spa_meta_objset;
5471 dnode_t *rl;
5472 VERIFY0(dnode_hold(mos,
5473 dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl));
5474 if (rl->dn_have_spill) {
5475 global_feature_count[
5476 SPA_FEATURE_REDACTION_LIST_SPILL]++;
5477 }
5478 }
5479 if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)
5480 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN]++;
5481 }
5482
5483 if (dsl_deadlist_is_open(&dmu_objset_ds(os)->ds_dir->dd_livelist) &&
5484 !dmu_objset_is_snapshot(os)) {
5485 global_feature_count[SPA_FEATURE_LIVELIST]++;
5486 }
5487
5488 dump_objset(os);
5489 close_objset(os, FTAG);
5490 fuid_table_destroy();
5491 return (0);
5492 }
5493
5494 /*
5495 * Block statistics.
5496 */
5497 #define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5498 typedef struct zdb_blkstats {
5499 uint64_t zb_asize;
5500 uint64_t zb_lsize;
5501 uint64_t zb_psize;
5502 uint64_t zb_count;
5503 uint64_t zb_gangs;
5504 uint64_t zb_ditto_samevdev;
5505 uint64_t zb_ditto_same_ms;
5506 uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
5507 } zdb_blkstats_t;
5508
5509 /*
5510 * Extended object types to report deferred frees and dedup auto-ditto blocks.
5511 */
5512 #define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0)
5513 #define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1)
5514 #define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2)
5515 #define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3)
5516
5517 static const char *zdb_ot_extname[] = {
5518 "deferred free",
5519 "dedup ditto",
5520 "other",
5521 "Total",
5522 };
5523
5524 #define ZB_TOTAL DN_MAX_LEVELS
5525 #define SPA_MAX_FOR_16M (SPA_MAXBLOCKSHIFT+1)
5526
5527 typedef struct zdb_brt_entry {
5528 dva_t zbre_dva;
5529 uint64_t zbre_refcount;
5530 avl_node_t zbre_node;
5531 } zdb_brt_entry_t;
5532
5533 typedef struct zdb_cb {
5534 zdb_blkstats_t zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
5535 uint64_t zcb_removing_size;
5536 uint64_t zcb_checkpoint_size;
5537 uint64_t zcb_dedup_asize;
5538 uint64_t zcb_dedup_blocks;
5539 uint64_t zcb_clone_asize;
5540 uint64_t zcb_clone_blocks;
5541 uint64_t zcb_psize_count[SPA_MAX_FOR_16M];
5542 uint64_t zcb_lsize_count[SPA_MAX_FOR_16M];
5543 uint64_t zcb_asize_count[SPA_MAX_FOR_16M];
5544 uint64_t zcb_psize_len[SPA_MAX_FOR_16M];
5545 uint64_t zcb_lsize_len[SPA_MAX_FOR_16M];
5546 uint64_t zcb_asize_len[SPA_MAX_FOR_16M];
5547 uint64_t zcb_psize_total;
5548 uint64_t zcb_lsize_total;
5549 uint64_t zcb_asize_total;
5550 uint64_t zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
5551 uint64_t zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
5552 [BPE_PAYLOAD_SIZE + 1];
5553 uint64_t zcb_start;
5554 hrtime_t zcb_lastprint;
5555 uint64_t zcb_totalasize;
5556 uint64_t zcb_errors[256];
5557 int zcb_readfails;
5558 int zcb_haderrors;
5559 spa_t *zcb_spa;
5560 uint32_t **zcb_vd_obsolete_counts;
5561 avl_tree_t zcb_brt;
5562 boolean_t zcb_brt_is_active;
5563 } zdb_cb_t;
5564
5565 /* test if two DVA offsets from same vdev are within the same metaslab */
5566 static boolean_t
5567 same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2)
5568 {
5569 vdev_t *vd = vdev_lookup_top(spa, vdev);
5570 uint64_t ms_shift = vd->vdev_ms_shift;
5571
5572 return ((off1 >> ms_shift) == (off2 >> ms_shift));
5573 }
5574
5575 /*
5576 * Used to simplify reporting of the histogram data.
5577 */
5578 typedef struct one_histo {
5579 const char *name;
5580 uint64_t *count;
5581 uint64_t *len;
5582 uint64_t cumulative;
5583 } one_histo_t;
5584
5585 /*
5586 * The number of separate histograms processed for psize, lsize and asize.
5587 */
5588 #define NUM_HISTO 3
5589
5590 /*
5591 * This routine will create a fixed column size output of three different
5592 * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5593 * the count, length and cumulative length of the psize, lsize and
5594 * asize blocks.
5595 *
5596 * All three types of blocks are listed on a single line
5597 *
5598 * By default the table is printed in nicenumber format (e.g. 123K) but
5599 * if the '-P' parameter is specified then the full raw number (parseable)
5600 * is printed out.
5601 */
5602 static void
5603 dump_size_histograms(zdb_cb_t *zcb)
5604 {
5605 /*
5606 * A temporary buffer that allows us to convert a number into
5607 * a string using zdb_nicenumber to allow either raw or human
5608 * readable numbers to be output.
5609 */
5610 char numbuf[32];
5611
5612 /*
5613 * Define titles which are used in the headers of the tables
5614 * printed by this routine.
5615 */
5616 const char blocksize_title1[] = "block";
5617 const char blocksize_title2[] = "size";
5618 const char count_title[] = "Count";
5619 const char length_title[] = "Size";
5620 const char cumulative_title[] = "Cum.";
5621
5622 /*
5623 * Setup the histogram arrays (psize, lsize, and asize).
5624 */
5625 one_histo_t parm_histo[NUM_HISTO];
5626
5627 parm_histo[0].name = "psize";
5628 parm_histo[0].count = zcb->zcb_psize_count;
5629 parm_histo[0].len = zcb->zcb_psize_len;
5630 parm_histo[0].cumulative = 0;
5631
5632 parm_histo[1].name = "lsize";
5633 parm_histo[1].count = zcb->zcb_lsize_count;
5634 parm_histo[1].len = zcb->zcb_lsize_len;
5635 parm_histo[1].cumulative = 0;
5636
5637 parm_histo[2].name = "asize";
5638 parm_histo[2].count = zcb->zcb_asize_count;
5639 parm_histo[2].len = zcb->zcb_asize_len;
5640 parm_histo[2].cumulative = 0;
5641
5642
5643 (void) printf("\nBlock Size Histogram\n");
5644 /*
5645 * Print the first line titles
5646 */
5647 if (dump_opt['P'])
5648 (void) printf("\n%s\t", blocksize_title1);
5649 else
5650 (void) printf("\n%7s ", blocksize_title1);
5651
5652 for (int j = 0; j < NUM_HISTO; j++) {
5653 if (dump_opt['P']) {
5654 if (j < NUM_HISTO - 1) {
5655 (void) printf("%s\t\t\t", parm_histo[j].name);
5656 } else {
5657 /* Don't print trailing spaces */
5658 (void) printf(" %s", parm_histo[j].name);
5659 }
5660 } else {
5661 if (j < NUM_HISTO - 1) {
5662 /* Left aligned strings in the output */
5663 (void) printf("%-7s ",
5664 parm_histo[j].name);
5665 } else {
5666 /* Don't print trailing spaces */
5667 (void) printf("%s", parm_histo[j].name);
5668 }
5669 }
5670 }
5671 (void) printf("\n");
5672
5673 /*
5674 * Print the second line titles
5675 */
5676 if (dump_opt['P']) {
5677 (void) printf("%s\t", blocksize_title2);
5678 } else {
5679 (void) printf("%7s ", blocksize_title2);
5680 }
5681
5682 for (int i = 0; i < NUM_HISTO; i++) {
5683 if (dump_opt['P']) {
5684 (void) printf("%s\t%s\t%s\t",
5685 count_title, length_title, cumulative_title);
5686 } else {
5687 (void) printf("%7s%7s%7s",
5688 count_title, length_title, cumulative_title);
5689 }
5690 }
5691 (void) printf("\n");
5692
5693 /*
5694 * Print the rows
5695 */
5696 for (int i = SPA_MINBLOCKSHIFT; i < SPA_MAX_FOR_16M; i++) {
5697
5698 /*
5699 * Print the first column showing the blocksize
5700 */
5701 zdb_nicenum((1ULL << i), numbuf, sizeof (numbuf));
5702
5703 if (dump_opt['P']) {
5704 printf("%s", numbuf);
5705 } else {
5706 printf("%7s:", numbuf);
5707 }
5708
5709 /*
5710 * Print the remaining set of 3 columns per size:
5711 * for psize, lsize and asize
5712 */
5713 for (int j = 0; j < NUM_HISTO; j++) {
5714 parm_histo[j].cumulative += parm_histo[j].len[i];
5715
5716 zdb_nicenum(parm_histo[j].count[i],
5717 numbuf, sizeof (numbuf));
5718 if (dump_opt['P'])
5719 (void) printf("\t%s", numbuf);
5720 else
5721 (void) printf("%7s", numbuf);
5722
5723 zdb_nicenum(parm_histo[j].len[i],
5724 numbuf, sizeof (numbuf));
5725 if (dump_opt['P'])
5726 (void) printf("\t%s", numbuf);
5727 else
5728 (void) printf("%7s", numbuf);
5729
5730 zdb_nicenum(parm_histo[j].cumulative,
5731 numbuf, sizeof (numbuf));
5732 if (dump_opt['P'])
5733 (void) printf("\t%s", numbuf);
5734 else
5735 (void) printf("%7s", numbuf);
5736 }
5737 (void) printf("\n");
5738 }
5739 }
5740
5741 static void
5742 zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
5743 dmu_object_type_t type)
5744 {
5745 int i;
5746
5747 ASSERT(type < ZDB_OT_TOTAL);
5748
5749 if (zilog && zil_bp_tree_add(zilog, bp) != 0)
5750 return;
5751
5752 /*
5753 * This flag controls if we will issue a claim for the block while
5754 * counting it, to ensure that all blocks are referenced in space maps.
5755 * We don't issue claims if we're not doing leak tracking, because it's
5756 * expensive if the user isn't interested. We also don't claim the
5757 * second or later occurences of cloned or dedup'd blocks, because we
5758 * already claimed them the first time.
5759 */
5760 boolean_t do_claim = !dump_opt['L'];
5761
5762 spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
5763
5764 blkptr_t tempbp;
5765 if (BP_GET_DEDUP(bp)) {
5766 /*
5767 * Dedup'd blocks are special. We need to count them, so we can
5768 * later uncount them when reporting leaked space, and we must
5769 * only claim them once.
5770 *
5771 * We use the existing dedup system to track what we've seen.
5772 * The first time we see a block, we do a ddt_lookup() to see
5773 * if it exists in the DDT. If we're doing leak tracking, we
5774 * claim the block at this time.
5775 *
5776 * Each time we see a block, we reduce the refcount in the
5777 * entry by one, and add to the size and count of dedup'd
5778 * blocks to report at the end.
5779 */
5780
5781 ddt_t *ddt = ddt_select(zcb->zcb_spa, bp);
5782
5783 ddt_enter(ddt);
5784
5785 /*
5786 * Find the block. This will create the entry in memory, but
5787 * we'll know if that happened by its refcount.
5788 */
5789 ddt_entry_t *dde = ddt_lookup(ddt, bp);
5790
5791 /*
5792 * ddt_lookup() can return NULL if this block didn't exist
5793 * in the DDT and creating it would take the DDT over its
5794 * quota. Since we got the block from disk, it must exist in
5795 * the DDT, so this can't happen. However, when unique entries
5796 * are pruned, the dedup bit can be set with no corresponding
5797 * entry in the DDT.
5798 */
5799 if (dde == NULL) {
5800 ddt_exit(ddt);
5801 goto skipped;
5802 }
5803
5804 /* Get the phys for this variant */
5805 ddt_phys_variant_t v = ddt_phys_select(ddt, dde, bp);
5806
5807 /*
5808 * This entry may have multiple sets of DVAs. We must claim
5809 * each set the first time we see them in a real block on disk,
5810 * or count them on subsequent occurences. We don't have a
5811 * convenient way to track the first time we see each variant,
5812 * so we repurpose dde_io as a set of "seen" flag bits. We can
5813 * do this safely in zdb because it never writes, so it will
5814 * never have a writing zio for this block in that pointer.
5815 */
5816 boolean_t seen = !!(((uintptr_t)dde->dde_io) & (1 << v));
5817 if (!seen)
5818 dde->dde_io =
5819 (void *)(((uintptr_t)dde->dde_io) | (1 << v));
5820
5821 /* Consume a reference for this block. */
5822 if (ddt_phys_total_refcnt(ddt, dde->dde_phys) > 0)
5823 ddt_phys_decref(dde->dde_phys, v);
5824
5825 /*
5826 * If this entry has a single flat phys, it may have been
5827 * extended with additional DVAs at some time in its life.
5828 * This block might be from before it was fully extended, and
5829 * so have fewer DVAs.
5830 *
5831 * If this is the first time we've seen this block, and we
5832 * claimed it as-is, then we would miss the claim on some
5833 * number of DVAs, which would then be seen as leaked.
5834 *
5835 * In all cases, if we've had fewer DVAs, then the asize would
5836 * be too small, and would lead to the pool apparently using
5837 * more space than allocated.
5838 *
5839 * To handle this, we copy the canonical set of DVAs from the
5840 * entry back to the block pointer before we claim it.
5841 */
5842 if (v == DDT_PHYS_FLAT) {
5843 ASSERT3U(BP_GET_BIRTH(bp), ==,
5844 ddt_phys_birth(dde->dde_phys, v));
5845 tempbp = *bp;
5846 ddt_bp_fill(dde->dde_phys, v, &tempbp,
5847 BP_GET_BIRTH(bp));
5848 bp = &tempbp;
5849 }
5850
5851 if (seen) {
5852 /*
5853 * The second or later time we see this block,
5854 * it's a duplicate and we count it.
5855 */
5856 zcb->zcb_dedup_asize += BP_GET_ASIZE(bp);
5857 zcb->zcb_dedup_blocks++;
5858
5859 /* Already claimed, don't do it again. */
5860 do_claim = B_FALSE;
5861 }
5862
5863 ddt_exit(ddt);
5864 } else if (zcb->zcb_brt_is_active &&
5865 brt_maybe_exists(zcb->zcb_spa, bp)) {
5866 /*
5867 * Cloned blocks are special. We need to count them, so we can
5868 * later uncount them when reporting leaked space, and we must
5869 * only claim them once.
5870 *
5871 * To do this, we keep our own in-memory BRT. For each block
5872 * we haven't seen before, we look it up in the real BRT and
5873 * if its there, we note it and its refcount then proceed as
5874 * normal. If we see the block again, we count it as a clone
5875 * and then give it no further consideration.
5876 */
5877 zdb_brt_entry_t zbre_search, *zbre;
5878 avl_index_t where;
5879
5880 zbre_search.zbre_dva = bp->blk_dva[0];
5881 zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where);
5882 if (zbre == NULL) {
5883 /* Not seen before; track it */
5884 uint64_t refcnt =
5885 brt_entry_get_refcount(zcb->zcb_spa, bp);
5886 if (refcnt > 0) {
5887 zbre = umem_zalloc(sizeof (zdb_brt_entry_t),
5888 UMEM_NOFAIL);
5889 zbre->zbre_dva = bp->blk_dva[0];
5890 zbre->zbre_refcount = refcnt;
5891 avl_insert(&zcb->zcb_brt, zbre, where);
5892 }
5893 } else {
5894 /*
5895 * Second or later occurrence, count it and take a
5896 * refcount.
5897 */
5898 zcb->zcb_clone_asize += BP_GET_ASIZE(bp);
5899 zcb->zcb_clone_blocks++;
5900
5901 zbre->zbre_refcount--;
5902 if (zbre->zbre_refcount == 0) {
5903 avl_remove(&zcb->zcb_brt, zbre);
5904 umem_free(zbre, sizeof (zdb_brt_entry_t));
5905 }
5906
5907 /* Already claimed, don't do it again. */
5908 do_claim = B_FALSE;
5909 }
5910 }
5911
5912 skipped:
5913 for (i = 0; i < 4; i++) {
5914 int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
5915 int t = (i & 1) ? type : ZDB_OT_TOTAL;
5916 int equal;
5917 zdb_blkstats_t *zb = &zcb->zcb_type[l][t];
5918
5919 zb->zb_asize += BP_GET_ASIZE(bp);
5920 zb->zb_lsize += BP_GET_LSIZE(bp);
5921 zb->zb_psize += BP_GET_PSIZE(bp);
5922 zb->zb_count++;
5923
5924 /*
5925 * The histogram is only big enough to record blocks up to
5926 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
5927 * "other", bucket.
5928 */
5929 unsigned idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT;
5930 idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1);
5931 zb->zb_psize_histogram[idx]++;
5932
5933 zb->zb_gangs += BP_COUNT_GANG(bp);
5934
5935 switch (BP_GET_NDVAS(bp)) {
5936 case 2:
5937 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5938 DVA_GET_VDEV(&bp->blk_dva[1])) {
5939 zb->zb_ditto_samevdev++;
5940
5941 if (same_metaslab(zcb->zcb_spa,
5942 DVA_GET_VDEV(&bp->blk_dva[0]),
5943 DVA_GET_OFFSET(&bp->blk_dva[0]),
5944 DVA_GET_OFFSET(&bp->blk_dva[1])))
5945 zb->zb_ditto_same_ms++;
5946 }
5947 break;
5948 case 3:
5949 equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5950 DVA_GET_VDEV(&bp->blk_dva[1])) +
5951 (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5952 DVA_GET_VDEV(&bp->blk_dva[2])) +
5953 (DVA_GET_VDEV(&bp->blk_dva[1]) ==
5954 DVA_GET_VDEV(&bp->blk_dva[2]));
5955 if (equal != 0) {
5956 zb->zb_ditto_samevdev++;
5957
5958 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5959 DVA_GET_VDEV(&bp->blk_dva[1]) &&
5960 same_metaslab(zcb->zcb_spa,
5961 DVA_GET_VDEV(&bp->blk_dva[0]),
5962 DVA_GET_OFFSET(&bp->blk_dva[0]),
5963 DVA_GET_OFFSET(&bp->blk_dva[1])))
5964 zb->zb_ditto_same_ms++;
5965 else if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5966 DVA_GET_VDEV(&bp->blk_dva[2]) &&
5967 same_metaslab(zcb->zcb_spa,
5968 DVA_GET_VDEV(&bp->blk_dva[0]),
5969 DVA_GET_OFFSET(&bp->blk_dva[0]),
5970 DVA_GET_OFFSET(&bp->blk_dva[2])))
5971 zb->zb_ditto_same_ms++;
5972 else if (DVA_GET_VDEV(&bp->blk_dva[1]) ==
5973 DVA_GET_VDEV(&bp->blk_dva[2]) &&
5974 same_metaslab(zcb->zcb_spa,
5975 DVA_GET_VDEV(&bp->blk_dva[1]),
5976 DVA_GET_OFFSET(&bp->blk_dva[1]),
5977 DVA_GET_OFFSET(&bp->blk_dva[2])))
5978 zb->zb_ditto_same_ms++;
5979 }
5980 break;
5981 }
5982 }
5983
5984 spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
5985
5986 if (BP_IS_EMBEDDED(bp)) {
5987 zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
5988 zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
5989 [BPE_GET_PSIZE(bp)]++;
5990 return;
5991 }
5992 /*
5993 * The binning histogram bins by powers of two up to
5994 * SPA_MAXBLOCKSIZE rather than creating bins for
5995 * every possible blocksize found in the pool.
5996 */
5997 int bin = highbit64(BP_GET_PSIZE(bp)) - 1;
5998
5999 zcb->zcb_psize_count[bin]++;
6000 zcb->zcb_psize_len[bin] += BP_GET_PSIZE(bp);
6001 zcb->zcb_psize_total += BP_GET_PSIZE(bp);
6002
6003 bin = highbit64(BP_GET_LSIZE(bp)) - 1;
6004
6005 zcb->zcb_lsize_count[bin]++;
6006 zcb->zcb_lsize_len[bin] += BP_GET_LSIZE(bp);
6007 zcb->zcb_lsize_total += BP_GET_LSIZE(bp);
6008
6009 bin = highbit64(BP_GET_ASIZE(bp)) - 1;
6010
6011 zcb->zcb_asize_count[bin]++;
6012 zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp);
6013 zcb->zcb_asize_total += BP_GET_ASIZE(bp);
6014
6015 if (!do_claim)
6016 return;
6017
6018 VERIFY0(zio_wait(zio_claim(NULL, zcb->zcb_spa,
6019 spa_min_claim_txg(zcb->zcb_spa), bp, NULL, NULL,
6020 ZIO_FLAG_CANFAIL)));
6021 }
6022
6023 static void
6024 zdb_blkptr_done(zio_t *zio)
6025 {
6026 spa_t *spa = zio->io_spa;
6027 blkptr_t *bp = zio->io_bp;
6028 int ioerr = zio->io_error;
6029 zdb_cb_t *zcb = zio->io_private;
6030 zbookmark_phys_t *zb = &zio->io_bookmark;
6031
6032 mutex_enter(&spa->spa_scrub_lock);
6033 spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
6034 cv_broadcast(&spa->spa_scrub_io_cv);
6035
6036 if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
6037 char blkbuf[BP_SPRINTF_LEN];
6038
6039 zcb->zcb_haderrors = 1;
6040 zcb->zcb_errors[ioerr]++;
6041
6042 if (dump_opt['b'] >= 2)
6043 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6044 else
6045 blkbuf[0] = '\0';
6046
6047 (void) printf("zdb_blkptr_cb: "
6048 "Got error %d reading "
6049 "<%llu, %llu, %lld, %llx> %s -- skipping\n",
6050 ioerr,
6051 (u_longlong_t)zb->zb_objset,
6052 (u_longlong_t)zb->zb_object,
6053 (u_longlong_t)zb->zb_level,
6054 (u_longlong_t)zb->zb_blkid,
6055 blkbuf);
6056 }
6057 mutex_exit(&spa->spa_scrub_lock);
6058
6059 abd_free(zio->io_abd);
6060 }
6061
6062 static int
6063 zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
6064 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
6065 {
6066 zdb_cb_t *zcb = arg;
6067 dmu_object_type_t type;
6068 boolean_t is_metadata;
6069
6070 if (zb->zb_level == ZB_DNODE_LEVEL)
6071 return (0);
6072
6073 if (dump_opt['b'] >= 5 && BP_GET_LOGICAL_BIRTH(bp) > 0) {
6074 char blkbuf[BP_SPRINTF_LEN];
6075 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6076 (void) printf("objset %llu object %llu "
6077 "level %lld offset 0x%llx %s\n",
6078 (u_longlong_t)zb->zb_objset,
6079 (u_longlong_t)zb->zb_object,
6080 (longlong_t)zb->zb_level,
6081 (u_longlong_t)blkid2offset(dnp, bp, zb),
6082 blkbuf);
6083 }
6084
6085 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
6086 return (0);
6087
6088 type = BP_GET_TYPE(bp);
6089
6090 zdb_count_block(zcb, zilog, bp,
6091 (type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type);
6092
6093 is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
6094
6095 if (!BP_IS_EMBEDDED(bp) &&
6096 (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
6097 size_t size = BP_GET_PSIZE(bp);
6098 abd_t *abd = abd_alloc(size, B_FALSE);
6099 int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
6100
6101 /* If it's an intent log block, failure is expected. */
6102 if (zb->zb_level == ZB_ZIL_LEVEL)
6103 flags |= ZIO_FLAG_SPECULATIVE;
6104
6105 mutex_enter(&spa->spa_scrub_lock);
6106 while (spa->spa_load_verify_bytes > max_inflight_bytes)
6107 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
6108 spa->spa_load_verify_bytes += size;
6109 mutex_exit(&spa->spa_scrub_lock);
6110
6111 zio_nowait(zio_read(NULL, spa, bp, abd, size,
6112 zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
6113 }
6114
6115 zcb->zcb_readfails = 0;
6116
6117 /* only call gethrtime() every 100 blocks */
6118 static int iters;
6119 if (++iters > 100)
6120 iters = 0;
6121 else
6122 return (0);
6123
6124 if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
6125 uint64_t now = gethrtime();
6126 char buf[10];
6127 uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
6128 uint64_t kb_per_sec =
6129 1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000));
6130 uint64_t sec_remaining =
6131 (zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec;
6132
6133 /* make sure nicenum has enough space */
6134 _Static_assert(sizeof (buf) >= NN_NUMBUF_SZ, "buf truncated");
6135
6136 zfs_nicebytes(bytes, buf, sizeof (buf));
6137 (void) fprintf(stderr,
6138 "\r%5s completed (%4"PRIu64"MB/s) "
6139 "estimated time remaining: "
6140 "%"PRIu64"hr %02"PRIu64"min %02"PRIu64"sec ",
6141 buf, kb_per_sec / 1024,
6142 sec_remaining / 60 / 60,
6143 sec_remaining / 60 % 60,
6144 sec_remaining % 60);
6145
6146 zcb->zcb_lastprint = now;
6147 }
6148
6149 return (0);
6150 }
6151
6152 static void
6153 zdb_leak(void *arg, uint64_t start, uint64_t size)
6154 {
6155 vdev_t *vd = arg;
6156
6157 (void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
6158 (u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
6159 }
6160
6161 static metaslab_ops_t zdb_metaslab_ops = {
6162 NULL /* alloc */
6163 };
6164
6165 static int
6166 load_unflushed_svr_segs_cb(spa_t *spa, space_map_entry_t *sme,
6167 uint64_t txg, void *arg)
6168 {
6169 spa_vdev_removal_t *svr = arg;
6170
6171 uint64_t offset = sme->sme_offset;
6172 uint64_t size = sme->sme_run;
6173
6174 /* skip vdevs we don't care about */
6175 if (sme->sme_vdev != svr->svr_vdev_id)
6176 return (0);
6177
6178 vdev_t *vd = vdev_lookup_top(spa, sme->sme_vdev);
6179 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6180 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6181
6182 if (txg < metaslab_unflushed_txg(ms))
6183 return (0);
6184
6185 if (sme->sme_type == SM_ALLOC)
6186 range_tree_add(svr->svr_allocd_segs, offset, size);
6187 else
6188 range_tree_remove(svr->svr_allocd_segs, offset, size);
6189
6190 return (0);
6191 }
6192
6193 static void
6194 claim_segment_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
6195 uint64_t size, void *arg)
6196 {
6197 (void) inner_offset, (void) arg;
6198
6199 /*
6200 * This callback was called through a remap from
6201 * a device being removed. Therefore, the vdev that
6202 * this callback is applied to is a concrete
6203 * vdev.
6204 */
6205 ASSERT(vdev_is_concrete(vd));
6206
6207 VERIFY0(metaslab_claim_impl(vd, offset, size,
6208 spa_min_claim_txg(vd->vdev_spa)));
6209 }
6210
6211 static void
6212 claim_segment_cb(void *arg, uint64_t offset, uint64_t size)
6213 {
6214 vdev_t *vd = arg;
6215
6216 vdev_indirect_ops.vdev_op_remap(vd, offset, size,
6217 claim_segment_impl_cb, NULL);
6218 }
6219
6220 /*
6221 * After accounting for all allocated blocks that are directly referenced,
6222 * we might have missed a reference to a block from a partially complete
6223 * (and thus unused) indirect mapping object. We perform a secondary pass
6224 * through the metaslabs we have already mapped and claim the destination
6225 * blocks.
6226 */
6227 static void
6228 zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb)
6229 {
6230 if (dump_opt['L'])
6231 return;
6232
6233 if (spa->spa_vdev_removal == NULL)
6234 return;
6235
6236 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6237
6238 spa_vdev_removal_t *svr = spa->spa_vdev_removal;
6239 vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id);
6240 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6241
6242 ASSERT0(range_tree_space(svr->svr_allocd_segs));
6243
6244 range_tree_t *allocs = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
6245 for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) {
6246 metaslab_t *msp = vd->vdev_ms[msi];
6247
6248 ASSERT0(range_tree_space(allocs));
6249 if (msp->ms_sm != NULL)
6250 VERIFY0(space_map_load(msp->ms_sm, allocs, SM_ALLOC));
6251 range_tree_vacate(allocs, range_tree_add, svr->svr_allocd_segs);
6252 }
6253 range_tree_destroy(allocs);
6254
6255 iterate_through_spacemap_logs(spa, load_unflushed_svr_segs_cb, svr);
6256
6257 /*
6258 * Clear everything past what has been synced,
6259 * because we have not allocated mappings for
6260 * it yet.
6261 */
6262 range_tree_clear(svr->svr_allocd_segs,
6263 vdev_indirect_mapping_max_offset(vim),
6264 vd->vdev_asize - vdev_indirect_mapping_max_offset(vim));
6265
6266 zcb->zcb_removing_size += range_tree_space(svr->svr_allocd_segs);
6267 range_tree_vacate(svr->svr_allocd_segs, claim_segment_cb, vd);
6268
6269 spa_config_exit(spa, SCL_CONFIG, FTAG);
6270 }
6271
6272 static int
6273 increment_indirect_mapping_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6274 dmu_tx_t *tx)
6275 {
6276 (void) tx;
6277 zdb_cb_t *zcb = arg;
6278 spa_t *spa = zcb->zcb_spa;
6279 vdev_t *vd;
6280 const dva_t *dva = &bp->blk_dva[0];
6281
6282 ASSERT(!bp_freed);
6283 ASSERT(!dump_opt['L']);
6284 ASSERT3U(BP_GET_NDVAS(bp), ==, 1);
6285
6286 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6287 vd = vdev_lookup_top(zcb->zcb_spa, DVA_GET_VDEV(dva));
6288 ASSERT3P(vd, !=, NULL);
6289 spa_config_exit(spa, SCL_VDEV, FTAG);
6290
6291 ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0);
6292 ASSERT3P(zcb->zcb_vd_obsolete_counts[vd->vdev_id], !=, NULL);
6293
6294 vdev_indirect_mapping_increment_obsolete_count(
6295 vd->vdev_indirect_mapping,
6296 DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva),
6297 zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6298
6299 return (0);
6300 }
6301
6302 static uint32_t *
6303 zdb_load_obsolete_counts(vdev_t *vd)
6304 {
6305 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6306 spa_t *spa = vd->vdev_spa;
6307 spa_condensing_indirect_phys_t *scip =
6308 &spa->spa_condensing_indirect_phys;
6309 uint64_t obsolete_sm_object;
6310 uint32_t *counts;
6311
6312 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
6313 EQUIV(obsolete_sm_object != 0, vd->vdev_obsolete_sm != NULL);
6314 counts = vdev_indirect_mapping_load_obsolete_counts(vim);
6315 if (vd->vdev_obsolete_sm != NULL) {
6316 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6317 vd->vdev_obsolete_sm);
6318 }
6319 if (scip->scip_vdev == vd->vdev_id &&
6320 scip->scip_prev_obsolete_sm_object != 0) {
6321 space_map_t *prev_obsolete_sm = NULL;
6322 VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset,
6323 scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0));
6324 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6325 prev_obsolete_sm);
6326 space_map_close(prev_obsolete_sm);
6327 }
6328 return (counts);
6329 }
6330
6331 typedef struct checkpoint_sm_exclude_entry_arg {
6332 vdev_t *cseea_vd;
6333 uint64_t cseea_checkpoint_size;
6334 } checkpoint_sm_exclude_entry_arg_t;
6335
6336 static int
6337 checkpoint_sm_exclude_entry_cb(space_map_entry_t *sme, void *arg)
6338 {
6339 checkpoint_sm_exclude_entry_arg_t *cseea = arg;
6340 vdev_t *vd = cseea->cseea_vd;
6341 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
6342 uint64_t end = sme->sme_offset + sme->sme_run;
6343
6344 ASSERT(sme->sme_type == SM_FREE);
6345
6346 /*
6347 * Since the vdev_checkpoint_sm exists in the vdev level
6348 * and the ms_sm space maps exist in the metaslab level,
6349 * an entry in the checkpoint space map could theoretically
6350 * cross the boundaries of the metaslab that it belongs.
6351 *
6352 * In reality, because of the way that we populate and
6353 * manipulate the checkpoint's space maps currently,
6354 * there shouldn't be any entries that cross metaslabs.
6355 * Hence the assertion below.
6356 *
6357 * That said, there is no fundamental requirement that
6358 * the checkpoint's space map entries should not cross
6359 * metaslab boundaries. So if needed we could add code
6360 * that handles metaslab-crossing segments in the future.
6361 */
6362 VERIFY3U(sme->sme_offset, >=, ms->ms_start);
6363 VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
6364
6365 /*
6366 * By removing the entry from the allocated segments we
6367 * also verify that the entry is there to begin with.
6368 */
6369 mutex_enter(&ms->ms_lock);
6370 range_tree_remove(ms->ms_allocatable, sme->sme_offset, sme->sme_run);
6371 mutex_exit(&ms->ms_lock);
6372
6373 cseea->cseea_checkpoint_size += sme->sme_run;
6374 return (0);
6375 }
6376
6377 static void
6378 zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_cb_t *zcb)
6379 {
6380 spa_t *spa = vd->vdev_spa;
6381 space_map_t *checkpoint_sm = NULL;
6382 uint64_t checkpoint_sm_obj;
6383
6384 /*
6385 * If there is no vdev_top_zap, we are in a pool whose
6386 * version predates the pool checkpoint feature.
6387 */
6388 if (vd->vdev_top_zap == 0)
6389 return;
6390
6391 /*
6392 * If there is no reference of the vdev_checkpoint_sm in
6393 * the vdev_top_zap, then one of the following scenarios
6394 * is true:
6395 *
6396 * 1] There is no checkpoint
6397 * 2] There is a checkpoint, but no checkpointed blocks
6398 * have been freed yet
6399 * 3] The current vdev is indirect
6400 *
6401 * In these cases we return immediately.
6402 */
6403 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
6404 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
6405 return;
6406
6407 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
6408 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1,
6409 &checkpoint_sm_obj));
6410
6411 checkpoint_sm_exclude_entry_arg_t cseea;
6412 cseea.cseea_vd = vd;
6413 cseea.cseea_checkpoint_size = 0;
6414
6415 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
6416 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
6417
6418 VERIFY0(space_map_iterate(checkpoint_sm,
6419 space_map_length(checkpoint_sm),
6420 checkpoint_sm_exclude_entry_cb, &cseea));
6421 space_map_close(checkpoint_sm);
6422
6423 zcb->zcb_checkpoint_size += cseea.cseea_checkpoint_size;
6424 }
6425
6426 static void
6427 zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb)
6428 {
6429 ASSERT(!dump_opt['L']);
6430
6431 vdev_t *rvd = spa->spa_root_vdev;
6432 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6433 ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id);
6434 zdb_leak_init_vdev_exclude_checkpoint(rvd->vdev_child[c], zcb);
6435 }
6436 }
6437
6438 static int
6439 count_unflushed_space_cb(spa_t *spa, space_map_entry_t *sme,
6440 uint64_t txg, void *arg)
6441 {
6442 int64_t *ualloc_space = arg;
6443
6444 uint64_t offset = sme->sme_offset;
6445 uint64_t vdev_id = sme->sme_vdev;
6446
6447 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6448 if (!vdev_is_concrete(vd))
6449 return (0);
6450
6451 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6452 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6453
6454 if (txg < metaslab_unflushed_txg(ms))
6455 return (0);
6456
6457 if (sme->sme_type == SM_ALLOC)
6458 *ualloc_space += sme->sme_run;
6459 else
6460 *ualloc_space -= sme->sme_run;
6461
6462 return (0);
6463 }
6464
6465 static int64_t
6466 get_unflushed_alloc_space(spa_t *spa)
6467 {
6468 if (dump_opt['L'])
6469 return (0);
6470
6471 int64_t ualloc_space = 0;
6472 iterate_through_spacemap_logs(spa, count_unflushed_space_cb,
6473 &ualloc_space);
6474 return (ualloc_space);
6475 }
6476
6477 static int
6478 load_unflushed_cb(spa_t *spa, space_map_entry_t *sme, uint64_t txg, void *arg)
6479 {
6480 maptype_t *uic_maptype = arg;
6481
6482 uint64_t offset = sme->sme_offset;
6483 uint64_t size = sme->sme_run;
6484 uint64_t vdev_id = sme->sme_vdev;
6485
6486 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6487
6488 /* skip indirect vdevs */
6489 if (!vdev_is_concrete(vd))
6490 return (0);
6491
6492 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6493
6494 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6495 ASSERT(*uic_maptype == SM_ALLOC || *uic_maptype == SM_FREE);
6496
6497 if (txg < metaslab_unflushed_txg(ms))
6498 return (0);
6499
6500 if (*uic_maptype == sme->sme_type)
6501 range_tree_add(ms->ms_allocatable, offset, size);
6502 else
6503 range_tree_remove(ms->ms_allocatable, offset, size);
6504
6505 return (0);
6506 }
6507
6508 static void
6509 load_unflushed_to_ms_allocatables(spa_t *spa, maptype_t maptype)
6510 {
6511 iterate_through_spacemap_logs(spa, load_unflushed_cb, &maptype);
6512 }
6513
6514 static void
6515 load_concrete_ms_allocatable_trees(spa_t *spa, maptype_t maptype)
6516 {
6517 vdev_t *rvd = spa->spa_root_vdev;
6518 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
6519 vdev_t *vd = rvd->vdev_child[i];
6520
6521 ASSERT3U(i, ==, vd->vdev_id);
6522
6523 if (vd->vdev_ops == &vdev_indirect_ops)
6524 continue;
6525
6526 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6527 metaslab_t *msp = vd->vdev_ms[m];
6528
6529 (void) fprintf(stderr,
6530 "\rloading concrete vdev %llu, "
6531 "metaslab %llu of %llu ...",
6532 (longlong_t)vd->vdev_id,
6533 (longlong_t)msp->ms_id,
6534 (longlong_t)vd->vdev_ms_count);
6535
6536 mutex_enter(&msp->ms_lock);
6537 range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6538
6539 /*
6540 * We don't want to spend the CPU manipulating the
6541 * size-ordered tree, so clear the range_tree ops.
6542 */
6543 msp->ms_allocatable->rt_ops = NULL;
6544
6545 if (msp->ms_sm != NULL) {
6546 VERIFY0(space_map_load(msp->ms_sm,
6547 msp->ms_allocatable, maptype));
6548 }
6549 if (!msp->ms_loaded)
6550 msp->ms_loaded = B_TRUE;
6551 mutex_exit(&msp->ms_lock);
6552 }
6553 }
6554
6555 load_unflushed_to_ms_allocatables(spa, maptype);
6556 }
6557
6558 /*
6559 * vm_idxp is an in-out parameter which (for indirect vdevs) is the
6560 * index in vim_entries that has the first entry in this metaslab.
6561 * On return, it will be set to the first entry after this metaslab.
6562 */
6563 static void
6564 load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab_t *msp,
6565 uint64_t *vim_idxp)
6566 {
6567 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6568
6569 mutex_enter(&msp->ms_lock);
6570 range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6571
6572 /*
6573 * We don't want to spend the CPU manipulating the
6574 * size-ordered tree, so clear the range_tree ops.
6575 */
6576 msp->ms_allocatable->rt_ops = NULL;
6577
6578 for (; *vim_idxp < vdev_indirect_mapping_num_entries(vim);
6579 (*vim_idxp)++) {
6580 vdev_indirect_mapping_entry_phys_t *vimep =
6581 &vim->vim_entries[*vim_idxp];
6582 uint64_t ent_offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6583 uint64_t ent_len = DVA_GET_ASIZE(&vimep->vimep_dst);
6584 ASSERT3U(ent_offset, >=, msp->ms_start);
6585 if (ent_offset >= msp->ms_start + msp->ms_size)
6586 break;
6587
6588 /*
6589 * Mappings do not cross metaslab boundaries,
6590 * because we create them by walking the metaslabs.
6591 */
6592 ASSERT3U(ent_offset + ent_len, <=,
6593 msp->ms_start + msp->ms_size);
6594 range_tree_add(msp->ms_allocatable, ent_offset, ent_len);
6595 }
6596
6597 if (!msp->ms_loaded)
6598 msp->ms_loaded = B_TRUE;
6599 mutex_exit(&msp->ms_lock);
6600 }
6601
6602 static void
6603 zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb)
6604 {
6605 ASSERT(!dump_opt['L']);
6606
6607 vdev_t *rvd = spa->spa_root_vdev;
6608 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6609 vdev_t *vd = rvd->vdev_child[c];
6610
6611 ASSERT3U(c, ==, vd->vdev_id);
6612
6613 if (vd->vdev_ops != &vdev_indirect_ops)
6614 continue;
6615
6616 /*
6617 * Note: we don't check for mapping leaks on
6618 * removing vdevs because their ms_allocatable's
6619 * are used to look for leaks in allocated space.
6620 */
6621 zcb->zcb_vd_obsolete_counts[c] = zdb_load_obsolete_counts(vd);
6622
6623 /*
6624 * Normally, indirect vdevs don't have any
6625 * metaslabs. We want to set them up for
6626 * zio_claim().
6627 */
6628 vdev_metaslab_group_create(vd);
6629 VERIFY0(vdev_metaslab_init(vd, 0));
6630
6631 vdev_indirect_mapping_t *vim __maybe_unused =
6632 vd->vdev_indirect_mapping;
6633 uint64_t vim_idx = 0;
6634 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6635
6636 (void) fprintf(stderr,
6637 "\rloading indirect vdev %llu, "
6638 "metaslab %llu of %llu ...",
6639 (longlong_t)vd->vdev_id,
6640 (longlong_t)vd->vdev_ms[m]->ms_id,
6641 (longlong_t)vd->vdev_ms_count);
6642
6643 load_indirect_ms_allocatable_tree(vd, vd->vdev_ms[m],
6644 &vim_idx);
6645 }
6646 ASSERT3U(vim_idx, ==, vdev_indirect_mapping_num_entries(vim));
6647 }
6648 }
6649
6650 static void
6651 zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
6652 {
6653 zcb->zcb_spa = spa;
6654
6655 if (dump_opt['L'])
6656 return;
6657
6658 dsl_pool_t *dp = spa->spa_dsl_pool;
6659 vdev_t *rvd = spa->spa_root_vdev;
6660
6661 /*
6662 * We are going to be changing the meaning of the metaslab's
6663 * ms_allocatable. Ensure that the allocator doesn't try to
6664 * use the tree.
6665 */
6666 spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
6667 spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
6668 spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6669
6670 zcb->zcb_vd_obsolete_counts =
6671 umem_zalloc(rvd->vdev_children * sizeof (uint32_t *),
6672 UMEM_NOFAIL);
6673
6674 /*
6675 * For leak detection, we overload the ms_allocatable trees
6676 * to contain allocated segments instead of free segments.
6677 * As a result, we can't use the normal metaslab_load/unload
6678 * interfaces.
6679 */
6680 zdb_leak_init_prepare_indirect_vdevs(spa, zcb);
6681 load_concrete_ms_allocatable_trees(spa, SM_ALLOC);
6682
6683 /*
6684 * On load_concrete_ms_allocatable_trees() we loaded all the
6685 * allocated entries from the ms_sm to the ms_allocatable for
6686 * each metaslab. If the pool has a checkpoint or is in the
6687 * middle of discarding a checkpoint, some of these blocks
6688 * may have been freed but their ms_sm may not have been
6689 * updated because they are referenced by the checkpoint. In
6690 * order to avoid false-positives during leak-detection, we
6691 * go through the vdev's checkpoint space map and exclude all
6692 * its entries from their relevant ms_allocatable.
6693 *
6694 * We also aggregate the space held by the checkpoint and add
6695 * it to zcb_checkpoint_size.
6696 *
6697 * Note that at this point we are also verifying that all the
6698 * entries on the checkpoint_sm are marked as allocated in
6699 * the ms_sm of their relevant metaslab.
6700 * [see comment in checkpoint_sm_exclude_entry_cb()]
6701 */
6702 zdb_leak_init_exclude_checkpoint(spa, zcb);
6703 ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa));
6704
6705 /* for cleaner progress output */
6706 (void) fprintf(stderr, "\n");
6707
6708 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
6709 ASSERT(spa_feature_is_enabled(spa,
6710 SPA_FEATURE_DEVICE_REMOVAL));
6711 (void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj,
6712 increment_indirect_mapping_cb, zcb, NULL);
6713 }
6714 }
6715
6716 static boolean_t
6717 zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb)
6718 {
6719 boolean_t leaks = B_FALSE;
6720 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6721 uint64_t total_leaked = 0;
6722 boolean_t are_precise = B_FALSE;
6723
6724 ASSERT(vim != NULL);
6725
6726 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
6727 vdev_indirect_mapping_entry_phys_t *vimep =
6728 &vim->vim_entries[i];
6729 uint64_t obsolete_bytes = 0;
6730 uint64_t offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6731 metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6732
6733 /*
6734 * This is not very efficient but it's easy to
6735 * verify correctness.
6736 */
6737 for (uint64_t inner_offset = 0;
6738 inner_offset < DVA_GET_ASIZE(&vimep->vimep_dst);
6739 inner_offset += 1ULL << vd->vdev_ashift) {
6740 if (range_tree_contains(msp->ms_allocatable,
6741 offset + inner_offset, 1ULL << vd->vdev_ashift)) {
6742 obsolete_bytes += 1ULL << vd->vdev_ashift;
6743 }
6744 }
6745
6746 int64_t bytes_leaked = obsolete_bytes -
6747 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i];
6748 ASSERT3U(DVA_GET_ASIZE(&vimep->vimep_dst), >=,
6749 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]);
6750
6751 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6752 if (bytes_leaked != 0 && (are_precise || dump_opt['d'] >= 5)) {
6753 (void) printf("obsolete indirect mapping count "
6754 "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
6755 (u_longlong_t)vd->vdev_id,
6756 (u_longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
6757 (u_longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
6758 (u_longlong_t)bytes_leaked);
6759 }
6760 total_leaked += ABS(bytes_leaked);
6761 }
6762
6763 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6764 if (!are_precise && total_leaked > 0) {
6765 int pct_leaked = total_leaked * 100 /
6766 vdev_indirect_mapping_bytes_mapped(vim);
6767 (void) printf("cannot verify obsolete indirect mapping "
6768 "counts of vdev %llu because precise feature was not "
6769 "enabled when it was removed: %d%% (%llx bytes) of mapping"
6770 "unreferenced\n",
6771 (u_longlong_t)vd->vdev_id, pct_leaked,
6772 (u_longlong_t)total_leaked);
6773 } else if (total_leaked > 0) {
6774 (void) printf("obsolete indirect mapping count mismatch "
6775 "for vdev %llu -- %llx total bytes mismatched\n",
6776 (u_longlong_t)vd->vdev_id,
6777 (u_longlong_t)total_leaked);
6778 leaks |= B_TRUE;
6779 }
6780
6781 vdev_indirect_mapping_free_obsolete_counts(vim,
6782 zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6783 zcb->zcb_vd_obsolete_counts[vd->vdev_id] = NULL;
6784
6785 return (leaks);
6786 }
6787
6788 static boolean_t
6789 zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb)
6790 {
6791 if (dump_opt['L'])
6792 return (B_FALSE);
6793
6794 boolean_t leaks = B_FALSE;
6795 vdev_t *rvd = spa->spa_root_vdev;
6796 for (unsigned c = 0; c < rvd->vdev_children; c++) {
6797 vdev_t *vd = rvd->vdev_child[c];
6798
6799 if (zcb->zcb_vd_obsolete_counts[c] != NULL) {
6800 leaks |= zdb_check_for_obsolete_leaks(vd, zcb);
6801 }
6802
6803 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6804 metaslab_t *msp = vd->vdev_ms[m];
6805 ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class ==
6806 spa_embedded_log_class(spa)) ?
6807 vd->vdev_log_mg : vd->vdev_mg);
6808
6809 /*
6810 * ms_allocatable has been overloaded
6811 * to contain allocated segments. Now that
6812 * we finished traversing all blocks, any
6813 * block that remains in the ms_allocatable
6814 * represents an allocated block that we
6815 * did not claim during the traversal.
6816 * Claimed blocks would have been removed
6817 * from the ms_allocatable. For indirect
6818 * vdevs, space remaining in the tree
6819 * represents parts of the mapping that are
6820 * not referenced, which is not a bug.
6821 */
6822 if (vd->vdev_ops == &vdev_indirect_ops) {
6823 range_tree_vacate(msp->ms_allocatable,
6824 NULL, NULL);
6825 } else {
6826 range_tree_vacate(msp->ms_allocatable,
6827 zdb_leak, vd);
6828 }
6829 if (msp->ms_loaded) {
6830 msp->ms_loaded = B_FALSE;
6831 }
6832 }
6833 }
6834
6835 umem_free(zcb->zcb_vd_obsolete_counts,
6836 rvd->vdev_children * sizeof (uint32_t *));
6837 zcb->zcb_vd_obsolete_counts = NULL;
6838
6839 return (leaks);
6840 }
6841
6842 static int
6843 count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6844 {
6845 (void) tx;
6846 zdb_cb_t *zcb = arg;
6847
6848 if (dump_opt['b'] >= 5) {
6849 char blkbuf[BP_SPRINTF_LEN];
6850 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6851 (void) printf("[%s] %s\n",
6852 "deferred free", blkbuf);
6853 }
6854 zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED);
6855 return (0);
6856 }
6857
6858 /*
6859 * Iterate over livelists which have been destroyed by the user but
6860 * are still present in the MOS, waiting to be freed
6861 */
6862 static void
6863 iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg)
6864 {
6865 objset_t *mos = spa->spa_meta_objset;
6866 uint64_t zap_obj;
6867 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
6868 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
6869 if (err == ENOENT)
6870 return;
6871 ASSERT0(err);
6872
6873 zap_cursor_t zc;
6874 zap_attribute_t *attrp = zap_attribute_alloc();
6875 dsl_deadlist_t ll;
6876 /* NULL out os prior to dsl_deadlist_open in case it's garbage */
6877 ll.dl_os = NULL;
6878 for (zap_cursor_init(&zc, mos, zap_obj);
6879 zap_cursor_retrieve(&zc, attrp) == 0;
6880 (void) zap_cursor_advance(&zc)) {
6881 dsl_deadlist_open(&ll, mos, attrp->za_first_integer);
6882 func(&ll, arg);
6883 dsl_deadlist_close(&ll);
6884 }
6885 zap_cursor_fini(&zc);
6886 zap_attribute_free(attrp);
6887 }
6888
6889 static int
6890 bpobj_count_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6891 dmu_tx_t *tx)
6892 {
6893 ASSERT(!bp_freed);
6894 return (count_block_cb(arg, bp, tx));
6895 }
6896
6897 static int
6898 livelist_entry_count_blocks_cb(void *args, dsl_deadlist_entry_t *dle)
6899 {
6900 zdb_cb_t *zbc = args;
6901 bplist_t blks;
6902 bplist_create(&blks);
6903 /* determine which blocks have been alloc'd but not freed */
6904 VERIFY0(dsl_process_sub_livelist(&dle->dle_bpobj, &blks, NULL, NULL));
6905 /* count those blocks */
6906 (void) bplist_iterate(&blks, count_block_cb, zbc, NULL);
6907 bplist_destroy(&blks);
6908 return (0);
6909 }
6910
6911 static void
6912 livelist_count_blocks(dsl_deadlist_t *ll, void *arg)
6913 {
6914 dsl_deadlist_iterate(ll, livelist_entry_count_blocks_cb, arg);
6915 }
6916
6917 /*
6918 * Count the blocks in the livelists that have been destroyed by the user
6919 * but haven't yet been freed.
6920 */
6921 static void
6922 deleted_livelists_count_blocks(spa_t *spa, zdb_cb_t *zbc)
6923 {
6924 iterate_deleted_livelists(spa, livelist_count_blocks, zbc);
6925 }
6926
6927 static void
6928 dump_livelist_cb(dsl_deadlist_t *ll, void *arg)
6929 {
6930 ASSERT3P(arg, ==, NULL);
6931 global_feature_count[SPA_FEATURE_LIVELIST]++;
6932 dump_blkptr_list(ll, "Deleted Livelist");
6933 dsl_deadlist_iterate(ll, sublivelist_verify_lightweight, NULL);
6934 }
6935
6936 /*
6937 * Print out, register object references to, and increment feature counts for
6938 * livelists that have been destroyed by the user but haven't yet been freed.
6939 */
6940 static void
6941 deleted_livelists_dump_mos(spa_t *spa)
6942 {
6943 uint64_t zap_obj;
6944 objset_t *mos = spa->spa_meta_objset;
6945 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
6946 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
6947 if (err == ENOENT)
6948 return;
6949 mos_obj_refd(zap_obj);
6950 iterate_deleted_livelists(spa, dump_livelist_cb, NULL);
6951 }
6952
6953 static int
6954 zdb_brt_entry_compare(const void *zcn1, const void *zcn2)
6955 {
6956 const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva;
6957 const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva;
6958 int cmp;
6959
6960 cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
6961 if (cmp == 0)
6962 cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2));
6963
6964 return (cmp);
6965 }
6966
6967 static int
6968 dump_block_stats(spa_t *spa)
6969 {
6970 zdb_cb_t *zcb;
6971 zdb_blkstats_t *zb, *tzb;
6972 uint64_t norm_alloc, norm_space, total_alloc, total_found;
6973 int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
6974 TRAVERSE_NO_DECRYPT | TRAVERSE_HARD;
6975 boolean_t leaks = B_FALSE;
6976 int e, c, err;
6977 bp_embedded_type_t i;
6978
6979 ddt_prefetch_all(spa);
6980
6981 zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL);
6982
6983 if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
6984 avl_create(&zcb->zcb_brt, zdb_brt_entry_compare,
6985 sizeof (zdb_brt_entry_t),
6986 offsetof(zdb_brt_entry_t, zbre_node));
6987 zcb->zcb_brt_is_active = B_TRUE;
6988 }
6989
6990 (void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
6991 (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
6992 (dump_opt['c'] == 1) ? "metadata " : "",
6993 dump_opt['c'] ? "checksums " : "",
6994 (dump_opt['c'] && !dump_opt['L']) ? "and verify " : "",
6995 !dump_opt['L'] ? "nothing leaked " : "");
6996
6997 /*
6998 * When leak detection is enabled we load all space maps as SM_ALLOC
6999 * maps, then traverse the pool claiming each block we discover. If
7000 * the pool is perfectly consistent, the segment trees will be empty
7001 * when we're done. Anything left over is a leak; any block we can't
7002 * claim (because it's not part of any space map) is a double
7003 * allocation, reference to a freed block, or an unclaimed log block.
7004 *
7005 * When leak detection is disabled (-L option) we still traverse the
7006 * pool claiming each block we discover, but we skip opening any space
7007 * maps.
7008 */
7009 zdb_leak_init(spa, zcb);
7010
7011 /*
7012 * If there's a deferred-free bplist, process that first.
7013 */
7014 (void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
7015 bpobj_count_block_cb, zcb, NULL);
7016
7017 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
7018 (void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
7019 bpobj_count_block_cb, zcb, NULL);
7020 }
7021
7022 zdb_claim_removing(spa, zcb);
7023
7024 if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
7025 VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,
7026 spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb,
7027 zcb, NULL));
7028 }
7029
7030 deleted_livelists_count_blocks(spa, zcb);
7031
7032 if (dump_opt['c'] > 1)
7033 flags |= TRAVERSE_PREFETCH_DATA;
7034
7035 zcb->zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
7036 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa));
7037 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
7038 zcb->zcb_totalasize +=
7039 metaslab_class_get_alloc(spa_embedded_log_class(spa));
7040 zcb->zcb_start = zcb->zcb_lastprint = gethrtime();
7041 err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb);
7042
7043 /*
7044 * If we've traversed the data blocks then we need to wait for those
7045 * I/Os to complete. We leverage "The Godfather" zio to wait on
7046 * all async I/Os to complete.
7047 */
7048 if (dump_opt['c']) {
7049 for (c = 0; c < max_ncpus; c++) {
7050 (void) zio_wait(spa->spa_async_zio_root[c]);
7051 spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL,
7052 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
7053 ZIO_FLAG_GODFATHER);
7054 }
7055 }
7056 ASSERT0(spa->spa_load_verify_bytes);
7057
7058 /*
7059 * Done after zio_wait() since zcb_haderrors is modified in
7060 * zdb_blkptr_done()
7061 */
7062 zcb->zcb_haderrors |= err;
7063
7064 if (zcb->zcb_haderrors) {
7065 (void) printf("\nError counts:\n\n");
7066 (void) printf("\t%5s %s\n", "errno", "count");
7067 for (e = 0; e < 256; e++) {
7068 if (zcb->zcb_errors[e] != 0) {
7069 (void) printf("\t%5d %llu\n",
7070 e, (u_longlong_t)zcb->zcb_errors[e]);
7071 }
7072 }
7073 }
7074
7075 /*
7076 * Report any leaked segments.
7077 */
7078 leaks |= zdb_leak_fini(spa, zcb);
7079
7080 tzb = &zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL];
7081
7082 norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7083 norm_space = metaslab_class_get_space(spa_normal_class(spa));
7084
7085 total_alloc = norm_alloc +
7086 metaslab_class_get_alloc(spa_log_class(spa)) +
7087 metaslab_class_get_alloc(spa_embedded_log_class(spa)) +
7088 metaslab_class_get_alloc(spa_special_class(spa)) +
7089 metaslab_class_get_alloc(spa_dedup_class(spa)) +
7090 get_unflushed_alloc_space(spa);
7091 total_found =
7092 tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize +
7093 zcb->zcb_removing_size + zcb->zcb_checkpoint_size;
7094
7095 if (total_found == total_alloc && !dump_opt['L']) {
7096 (void) printf("\n\tNo leaks (block sum matches space"
7097 " maps exactly)\n");
7098 } else if (!dump_opt['L']) {
7099 (void) printf("block traversal size %llu != alloc %llu "
7100 "(%s %lld)\n",
7101 (u_longlong_t)total_found,
7102 (u_longlong_t)total_alloc,
7103 (dump_opt['L']) ? "unreachable" : "leaked",
7104 (longlong_t)(total_alloc - total_found));
7105 }
7106
7107 if (tzb->zb_count == 0) {
7108 umem_free(zcb, sizeof (zdb_cb_t));
7109 return (2);
7110 }
7111
7112 (void) printf("\n");
7113 (void) printf("\t%-16s %14llu\n", "bp count:",
7114 (u_longlong_t)tzb->zb_count);
7115 (void) printf("\t%-16s %14llu\n", "ganged count:",
7116 (longlong_t)tzb->zb_gangs);
7117 (void) printf("\t%-16s %14llu avg: %6llu\n", "bp logical:",
7118 (u_longlong_t)tzb->zb_lsize,
7119 (u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
7120 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7121 "bp physical:", (u_longlong_t)tzb->zb_psize,
7122 (u_longlong_t)(tzb->zb_psize / tzb->zb_count),
7123 (double)tzb->zb_lsize / tzb->zb_psize);
7124 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7125 "bp allocated:", (u_longlong_t)tzb->zb_asize,
7126 (u_longlong_t)(tzb->zb_asize / tzb->zb_count),
7127 (double)tzb->zb_lsize / tzb->zb_asize);
7128 (void) printf("\t%-16s %14llu ref>1: %6llu deduplication: %6.2f\n",
7129 "bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize,
7130 (u_longlong_t)zcb->zcb_dedup_blocks,
7131 (double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0);
7132 (void) printf("\t%-16s %14llu count: %6llu\n",
7133 "bp cloned:", (u_longlong_t)zcb->zcb_clone_asize,
7134 (u_longlong_t)zcb->zcb_clone_blocks);
7135 (void) printf("\t%-16s %14llu used: %5.2f%%\n", "Normal class:",
7136 (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
7137
7138 if (spa_special_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7139 uint64_t alloc = metaslab_class_get_alloc(
7140 spa_special_class(spa));
7141 uint64_t space = metaslab_class_get_space(
7142 spa_special_class(spa));
7143
7144 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7145 "Special class", (u_longlong_t)alloc,
7146 100.0 * alloc / space);
7147 }
7148
7149 if (spa_dedup_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7150 uint64_t alloc = metaslab_class_get_alloc(
7151 spa_dedup_class(spa));
7152 uint64_t space = metaslab_class_get_space(
7153 spa_dedup_class(spa));
7154
7155 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7156 "Dedup class", (u_longlong_t)alloc,
7157 100.0 * alloc / space);
7158 }
7159
7160 if (spa_embedded_log_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7161 uint64_t alloc = metaslab_class_get_alloc(
7162 spa_embedded_log_class(spa));
7163 uint64_t space = metaslab_class_get_space(
7164 spa_embedded_log_class(spa));
7165
7166 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7167 "Embedded log class", (u_longlong_t)alloc,
7168 100.0 * alloc / space);
7169 }
7170
7171 for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
7172 if (zcb->zcb_embedded_blocks[i] == 0)
7173 continue;
7174 (void) printf("\n");
7175 (void) printf("\tadditional, non-pointer bps of type %u: "
7176 "%10llu\n",
7177 i, (u_longlong_t)zcb->zcb_embedded_blocks[i]);
7178
7179 if (dump_opt['b'] >= 3) {
7180 (void) printf("\t number of (compressed) bytes: "
7181 "number of bps\n");
7182 dump_histogram(zcb->zcb_embedded_histogram[i],
7183 sizeof (zcb->zcb_embedded_histogram[i]) /
7184 sizeof (zcb->zcb_embedded_histogram[i][0]), 0);
7185 }
7186 }
7187
7188 if (tzb->zb_ditto_samevdev != 0) {
7189 (void) printf("\tDittoed blocks on same vdev: %llu\n",
7190 (longlong_t)tzb->zb_ditto_samevdev);
7191 }
7192 if (tzb->zb_ditto_same_ms != 0) {
7193 (void) printf("\tDittoed blocks in same metaslab: %llu\n",
7194 (longlong_t)tzb->zb_ditto_same_ms);
7195 }
7196
7197 for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) {
7198 vdev_t *vd = spa->spa_root_vdev->vdev_child[v];
7199 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7200
7201 if (vim == NULL) {
7202 continue;
7203 }
7204
7205 char mem[32];
7206 zdb_nicenum(vdev_indirect_mapping_num_entries(vim),
7207 mem, vdev_indirect_mapping_size(vim));
7208
7209 (void) printf("\tindirect vdev id %llu has %llu segments "
7210 "(%s in memory)\n",
7211 (longlong_t)vd->vdev_id,
7212 (longlong_t)vdev_indirect_mapping_num_entries(vim), mem);
7213 }
7214
7215 if (dump_opt['b'] >= 2) {
7216 int l, t, level;
7217 char csize[32], lsize[32], psize[32], asize[32];
7218 char avg[32], gang[32];
7219 (void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
7220 "\t avg\t comp\t%%Total\tType\n");
7221
7222 zfs_blkstat_t *mdstats = umem_zalloc(sizeof (zfs_blkstat_t),
7223 UMEM_NOFAIL);
7224
7225 for (t = 0; t <= ZDB_OT_TOTAL; t++) {
7226 const char *typename;
7227
7228 /* make sure nicenum has enough space */
7229 _Static_assert(sizeof (csize) >= NN_NUMBUF_SZ,
7230 "csize truncated");
7231 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ,
7232 "lsize truncated");
7233 _Static_assert(sizeof (psize) >= NN_NUMBUF_SZ,
7234 "psize truncated");
7235 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ,
7236 "asize truncated");
7237 _Static_assert(sizeof (avg) >= NN_NUMBUF_SZ,
7238 "avg truncated");
7239 _Static_assert(sizeof (gang) >= NN_NUMBUF_SZ,
7240 "gang truncated");
7241
7242 if (t < DMU_OT_NUMTYPES)
7243 typename = dmu_ot[t].ot_name;
7244 else
7245 typename = zdb_ot_extname[t - DMU_OT_NUMTYPES];
7246
7247 if (zcb->zcb_type[ZB_TOTAL][t].zb_asize == 0) {
7248 (void) printf("%6s\t%5s\t%5s\t%5s"
7249 "\t%5s\t%5s\t%6s\t%s\n",
7250 "-",
7251 "-",
7252 "-",
7253 "-",
7254 "-",
7255 "-",
7256 "-",
7257 typename);
7258 continue;
7259 }
7260
7261 for (l = ZB_TOTAL - 1; l >= -1; l--) {
7262 level = (l == -1 ? ZB_TOTAL : l);
7263 zb = &zcb->zcb_type[level][t];
7264
7265 if (zb->zb_asize == 0)
7266 continue;
7267
7268 if (level != ZB_TOTAL && t < DMU_OT_NUMTYPES &&
7269 (level > 0 || DMU_OT_IS_METADATA(t))) {
7270 mdstats->zb_count += zb->zb_count;
7271 mdstats->zb_lsize += zb->zb_lsize;
7272 mdstats->zb_psize += zb->zb_psize;
7273 mdstats->zb_asize += zb->zb_asize;
7274 mdstats->zb_gangs += zb->zb_gangs;
7275 }
7276
7277 if (dump_opt['b'] < 3 && level != ZB_TOTAL)
7278 continue;
7279
7280 if (level == 0 && zb->zb_asize ==
7281 zcb->zcb_type[ZB_TOTAL][t].zb_asize)
7282 continue;
7283
7284 zdb_nicenum(zb->zb_count, csize,
7285 sizeof (csize));
7286 zdb_nicenum(zb->zb_lsize, lsize,
7287 sizeof (lsize));
7288 zdb_nicenum(zb->zb_psize, psize,
7289 sizeof (psize));
7290 zdb_nicenum(zb->zb_asize, asize,
7291 sizeof (asize));
7292 zdb_nicenum(zb->zb_asize / zb->zb_count, avg,
7293 sizeof (avg));
7294 zdb_nicenum(zb->zb_gangs, gang, sizeof (gang));
7295
7296 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7297 "\t%5.2f\t%6.2f\t",
7298 csize, lsize, psize, asize, avg,
7299 (double)zb->zb_lsize / zb->zb_psize,
7300 100.0 * zb->zb_asize / tzb->zb_asize);
7301
7302 if (level == ZB_TOTAL)
7303 (void) printf("%s\n", typename);
7304 else
7305 (void) printf(" L%d %s\n",
7306 level, typename);
7307
7308 if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) {
7309 (void) printf("\t number of ganged "
7310 "blocks: %s\n", gang);
7311 }
7312
7313 if (dump_opt['b'] >= 4) {
7314 (void) printf("psize "
7315 "(in 512-byte sectors): "
7316 "number of blocks\n");
7317 dump_histogram(zb->zb_psize_histogram,
7318 PSIZE_HISTO_SIZE, 0);
7319 }
7320 }
7321 }
7322 zdb_nicenum(mdstats->zb_count, csize,
7323 sizeof (csize));
7324 zdb_nicenum(mdstats->zb_lsize, lsize,
7325 sizeof (lsize));
7326 zdb_nicenum(mdstats->zb_psize, psize,
7327 sizeof (psize));
7328 zdb_nicenum(mdstats->zb_asize, asize,
7329 sizeof (asize));
7330 zdb_nicenum(mdstats->zb_asize / mdstats->zb_count, avg,
7331 sizeof (avg));
7332 zdb_nicenum(mdstats->zb_gangs, gang, sizeof (gang));
7333
7334 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7335 "\t%5.2f\t%6.2f\t",
7336 csize, lsize, psize, asize, avg,
7337 (double)mdstats->zb_lsize / mdstats->zb_psize,
7338 100.0 * mdstats->zb_asize / tzb->zb_asize);
7339 (void) printf("%s\n", "Metadata Total");
7340
7341 /* Output a table summarizing block sizes in the pool */
7342 if (dump_opt['b'] >= 2) {
7343 dump_size_histograms(zcb);
7344 }
7345
7346 umem_free(mdstats, sizeof (zfs_blkstat_t));
7347 }
7348
7349 (void) printf("\n");
7350
7351 if (leaks) {
7352 umem_free(zcb, sizeof (zdb_cb_t));
7353 return (2);
7354 }
7355
7356 if (zcb->zcb_haderrors) {
7357 umem_free(zcb, sizeof (zdb_cb_t));
7358 return (3);
7359 }
7360
7361 umem_free(zcb, sizeof (zdb_cb_t));
7362 return (0);
7363 }
7364
7365 typedef struct zdb_ddt_entry {
7366 /* key must be first for ddt_key_compare */
7367 ddt_key_t zdde_key;
7368 uint64_t zdde_ref_blocks;
7369 uint64_t zdde_ref_lsize;
7370 uint64_t zdde_ref_psize;
7371 uint64_t zdde_ref_dsize;
7372 avl_node_t zdde_node;
7373 } zdb_ddt_entry_t;
7374
7375 static int
7376 zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
7377 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
7378 {
7379 (void) zilog, (void) dnp;
7380 avl_tree_t *t = arg;
7381 avl_index_t where;
7382 zdb_ddt_entry_t *zdde, zdde_search;
7383
7384 if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
7385 BP_IS_EMBEDDED(bp))
7386 return (0);
7387
7388 if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
7389 (void) printf("traversing objset %llu, %llu objects, "
7390 "%lu blocks so far\n",
7391 (u_longlong_t)zb->zb_objset,
7392 (u_longlong_t)BP_GET_FILL(bp),
7393 avl_numnodes(t));
7394 }
7395
7396 if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF ||
7397 BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
7398 return (0);
7399
7400 ddt_key_fill(&zdde_search.zdde_key, bp);
7401
7402 zdde = avl_find(t, &zdde_search, &where);
7403
7404 if (zdde == NULL) {
7405 zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL);
7406 zdde->zdde_key = zdde_search.zdde_key;
7407 avl_insert(t, zdde, where);
7408 }
7409
7410 zdde->zdde_ref_blocks += 1;
7411 zdde->zdde_ref_lsize += BP_GET_LSIZE(bp);
7412 zdde->zdde_ref_psize += BP_GET_PSIZE(bp);
7413 zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp);
7414
7415 return (0);
7416 }
7417
7418 static void
7419 dump_simulated_ddt(spa_t *spa)
7420 {
7421 avl_tree_t t;
7422 void *cookie = NULL;
7423 zdb_ddt_entry_t *zdde;
7424 ddt_histogram_t ddh_total = {{{0}}};
7425 ddt_stat_t dds_total = {0};
7426
7427 avl_create(&t, ddt_key_compare,
7428 sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
7429
7430 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7431
7432 (void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7433 TRAVERSE_NO_DECRYPT, zdb_ddt_add_cb, &t);
7434
7435 spa_config_exit(spa, SCL_CONFIG, FTAG);
7436
7437 while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) {
7438 uint64_t refcnt = zdde->zdde_ref_blocks;
7439 ASSERT(refcnt != 0);
7440
7441 ddt_stat_t *dds = &ddh_total.ddh_stat[highbit64(refcnt) - 1];
7442
7443 dds->dds_blocks += zdde->zdde_ref_blocks / refcnt;
7444 dds->dds_lsize += zdde->zdde_ref_lsize / refcnt;
7445 dds->dds_psize += zdde->zdde_ref_psize / refcnt;
7446 dds->dds_dsize += zdde->zdde_ref_dsize / refcnt;
7447
7448 dds->dds_ref_blocks += zdde->zdde_ref_blocks;
7449 dds->dds_ref_lsize += zdde->zdde_ref_lsize;
7450 dds->dds_ref_psize += zdde->zdde_ref_psize;
7451 dds->dds_ref_dsize += zdde->zdde_ref_dsize;
7452
7453 umem_free(zdde, sizeof (*zdde));
7454 }
7455
7456 avl_destroy(&t);
7457
7458 ddt_histogram_total(&dds_total, &ddh_total);
7459
7460 (void) printf("Simulated DDT histogram:\n");
7461
7462 zpool_dump_ddt(&dds_total, &ddh_total);
7463
7464 dump_dedup_ratio(&dds_total);
7465 }
7466
7467 static int
7468 verify_device_removal_feature_counts(spa_t *spa)
7469 {
7470 uint64_t dr_feature_refcount = 0;
7471 uint64_t oc_feature_refcount = 0;
7472 uint64_t indirect_vdev_count = 0;
7473 uint64_t precise_vdev_count = 0;
7474 uint64_t obsolete_counts_object_count = 0;
7475 uint64_t obsolete_sm_count = 0;
7476 uint64_t obsolete_counts_count = 0;
7477 uint64_t scip_count = 0;
7478 uint64_t obsolete_bpobj_count = 0;
7479 int ret = 0;
7480
7481 spa_condensing_indirect_phys_t *scip =
7482 &spa->spa_condensing_indirect_phys;
7483 if (scip->scip_next_mapping_object != 0) {
7484 vdev_t *vd = spa->spa_root_vdev->vdev_child[scip->scip_vdev];
7485 ASSERT(scip->scip_prev_obsolete_sm_object != 0);
7486 ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
7487
7488 (void) printf("Condensing indirect vdev %llu: new mapping "
7489 "object %llu, prev obsolete sm %llu\n",
7490 (u_longlong_t)scip->scip_vdev,
7491 (u_longlong_t)scip->scip_next_mapping_object,
7492 (u_longlong_t)scip->scip_prev_obsolete_sm_object);
7493 if (scip->scip_prev_obsolete_sm_object != 0) {
7494 space_map_t *prev_obsolete_sm = NULL;
7495 VERIFY0(space_map_open(&prev_obsolete_sm,
7496 spa->spa_meta_objset,
7497 scip->scip_prev_obsolete_sm_object,
7498 0, vd->vdev_asize, 0));
7499 dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm);
7500 (void) printf("\n");
7501 space_map_close(prev_obsolete_sm);
7502 }
7503
7504 scip_count += 2;
7505 }
7506
7507 for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
7508 vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
7509 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
7510
7511 if (vic->vic_mapping_object != 0) {
7512 ASSERT(vd->vdev_ops == &vdev_indirect_ops ||
7513 vd->vdev_removing);
7514 indirect_vdev_count++;
7515
7516 if (vd->vdev_indirect_mapping->vim_havecounts) {
7517 obsolete_counts_count++;
7518 }
7519 }
7520
7521 boolean_t are_precise;
7522 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7523 if (are_precise) {
7524 ASSERT(vic->vic_mapping_object != 0);
7525 precise_vdev_count++;
7526 }
7527
7528 uint64_t obsolete_sm_object;
7529 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
7530 if (obsolete_sm_object != 0) {
7531 ASSERT(vic->vic_mapping_object != 0);
7532 obsolete_sm_count++;
7533 }
7534 }
7535
7536 (void) feature_get_refcount(spa,
7537 &spa_feature_table[SPA_FEATURE_DEVICE_REMOVAL],
7538 &dr_feature_refcount);
7539 (void) feature_get_refcount(spa,
7540 &spa_feature_table[SPA_FEATURE_OBSOLETE_COUNTS],
7541 &oc_feature_refcount);
7542
7543 if (dr_feature_refcount != indirect_vdev_count) {
7544 ret = 1;
7545 (void) printf("Number of indirect vdevs (%llu) " \
7546 "does not match feature count (%llu)\n",
7547 (u_longlong_t)indirect_vdev_count,
7548 (u_longlong_t)dr_feature_refcount);
7549 } else {
7550 (void) printf("Verified device_removal feature refcount " \
7551 "of %llu is correct\n",
7552 (u_longlong_t)dr_feature_refcount);
7553 }
7554
7555 if (zap_contains(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT,
7556 DMU_POOL_OBSOLETE_BPOBJ) == 0) {
7557 obsolete_bpobj_count++;
7558 }
7559
7560
7561 obsolete_counts_object_count = precise_vdev_count;
7562 obsolete_counts_object_count += obsolete_sm_count;
7563 obsolete_counts_object_count += obsolete_counts_count;
7564 obsolete_counts_object_count += scip_count;
7565 obsolete_counts_object_count += obsolete_bpobj_count;
7566 obsolete_counts_object_count += remap_deadlist_count;
7567
7568 if (oc_feature_refcount != obsolete_counts_object_count) {
7569 ret = 1;
7570 (void) printf("Number of obsolete counts objects (%llu) " \
7571 "does not match feature count (%llu)\n",
7572 (u_longlong_t)obsolete_counts_object_count,
7573 (u_longlong_t)oc_feature_refcount);
7574 (void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
7575 "ob:%llu rd:%llu\n",
7576 (u_longlong_t)precise_vdev_count,
7577 (u_longlong_t)obsolete_sm_count,
7578 (u_longlong_t)obsolete_counts_count,
7579 (u_longlong_t)scip_count,
7580 (u_longlong_t)obsolete_bpobj_count,
7581 (u_longlong_t)remap_deadlist_count);
7582 } else {
7583 (void) printf("Verified indirect_refcount feature refcount " \
7584 "of %llu is correct\n",
7585 (u_longlong_t)oc_feature_refcount);
7586 }
7587 return (ret);
7588 }
7589
7590 static void
7591 zdb_set_skip_mmp(char *target)
7592 {
7593 spa_t *spa;
7594
7595 /*
7596 * Disable the activity check to allow examination of
7597 * active pools.
7598 */
7599 mutex_enter(&spa_namespace_lock);
7600 if ((spa = spa_lookup(target)) != NULL) {
7601 spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
7602 }
7603 mutex_exit(&spa_namespace_lock);
7604 }
7605
7606 #define BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7607 /*
7608 * Import the checkpointed state of the pool specified by the target
7609 * parameter as readonly. The function also accepts a pool config
7610 * as an optional parameter, else it attempts to infer the config by
7611 * the name of the target pool.
7612 *
7613 * Note that the checkpointed state's pool name will be the name of
7614 * the original pool with the above suffix appended to it. In addition,
7615 * if the target is not a pool name (e.g. a path to a dataset) then
7616 * the new_path parameter is populated with the updated path to
7617 * reflect the fact that we are looking into the checkpointed state.
7618 *
7619 * The function returns a newly-allocated copy of the name of the
7620 * pool containing the checkpointed state. When this copy is no
7621 * longer needed it should be freed with free(3C). Same thing
7622 * applies to the new_path parameter if allocated.
7623 */
7624 static char *
7625 import_checkpointed_state(char *target, nvlist_t *cfg, char **new_path)
7626 {
7627 int error = 0;
7628 char *poolname, *bogus_name = NULL;
7629 boolean_t freecfg = B_FALSE;
7630
7631 /* If the target is not a pool, the extract the pool name */
7632 char *path_start = strchr(target, '/');
7633 if (path_start != NULL) {
7634 size_t poolname_len = path_start - target;
7635 poolname = strndup(target, poolname_len);
7636 } else {
7637 poolname = target;
7638 }
7639
7640 if (cfg == NULL) {
7641 zdb_set_skip_mmp(poolname);
7642 error = spa_get_stats(poolname, &cfg, NULL, 0);
7643 if (error != 0) {
7644 fatal("Tried to read config of pool \"%s\" but "
7645 "spa_get_stats() failed with error %d\n",
7646 poolname, error);
7647 }
7648 freecfg = B_TRUE;
7649 }
7650
7651 if (asprintf(&bogus_name, "%s%s", poolname, BOGUS_SUFFIX) == -1) {
7652 if (target != poolname)
7653 free(poolname);
7654 return (NULL);
7655 }
7656 fnvlist_add_string(cfg, ZPOOL_CONFIG_POOL_NAME, bogus_name);
7657
7658 error = spa_import(bogus_name, cfg, NULL,
7659 ZFS_IMPORT_MISSING_LOG | ZFS_IMPORT_CHECKPOINT |
7660 ZFS_IMPORT_SKIP_MMP);
7661 if (freecfg)
7662 nvlist_free(cfg);
7663 if (error != 0) {
7664 fatal("Tried to import pool \"%s\" but spa_import() failed "
7665 "with error %d\n", bogus_name, error);
7666 }
7667
7668 if (new_path != NULL && path_start != NULL) {
7669 if (asprintf(new_path, "%s%s", bogus_name, path_start) == -1) {
7670 free(bogus_name);
7671 if (path_start != NULL)
7672 free(poolname);
7673 return (NULL);
7674 }
7675 }
7676
7677 if (target != poolname)
7678 free(poolname);
7679
7680 return (bogus_name);
7681 }
7682
7683 typedef struct verify_checkpoint_sm_entry_cb_arg {
7684 vdev_t *vcsec_vd;
7685
7686 /* the following fields are only used for printing progress */
7687 uint64_t vcsec_entryid;
7688 uint64_t vcsec_num_entries;
7689 } verify_checkpoint_sm_entry_cb_arg_t;
7690
7691 #define ENTRIES_PER_PROGRESS_UPDATE 10000
7692
7693 static int
7694 verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, void *arg)
7695 {
7696 verify_checkpoint_sm_entry_cb_arg_t *vcsec = arg;
7697 vdev_t *vd = vcsec->vcsec_vd;
7698 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
7699 uint64_t end = sme->sme_offset + sme->sme_run;
7700
7701 ASSERT(sme->sme_type == SM_FREE);
7702
7703 if ((vcsec->vcsec_entryid % ENTRIES_PER_PROGRESS_UPDATE) == 0) {
7704 (void) fprintf(stderr,
7705 "\rverifying vdev %llu, space map entry %llu of %llu ...",
7706 (longlong_t)vd->vdev_id,
7707 (longlong_t)vcsec->vcsec_entryid,
7708 (longlong_t)vcsec->vcsec_num_entries);
7709 }
7710 vcsec->vcsec_entryid++;
7711
7712 /*
7713 * See comment in checkpoint_sm_exclude_entry_cb()
7714 */
7715 VERIFY3U(sme->sme_offset, >=, ms->ms_start);
7716 VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
7717
7718 /*
7719 * The entries in the vdev_checkpoint_sm should be marked as
7720 * allocated in the checkpointed state of the pool, therefore
7721 * their respective ms_allocateable trees should not contain them.
7722 */
7723 mutex_enter(&ms->ms_lock);
7724 range_tree_verify_not_present(ms->ms_allocatable,
7725 sme->sme_offset, sme->sme_run);
7726 mutex_exit(&ms->ms_lock);
7727
7728 return (0);
7729 }
7730
7731 /*
7732 * Verify that all segments in the vdev_checkpoint_sm are allocated
7733 * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
7734 * ms_allocatable).
7735 *
7736 * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
7737 * each vdev in the current state of the pool to the metaslab space maps
7738 * (ms_sm) of the checkpointed state of the pool.
7739 *
7740 * Note that the function changes the state of the ms_allocatable
7741 * trees of the current spa_t. The entries of these ms_allocatable
7742 * trees are cleared out and then repopulated from with the free
7743 * entries of their respective ms_sm space maps.
7744 */
7745 static void
7746 verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, spa_t *current)
7747 {
7748 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7749 vdev_t *current_rvd = current->spa_root_vdev;
7750
7751 load_concrete_ms_allocatable_trees(checkpoint, SM_FREE);
7752
7753 for (uint64_t c = 0; c < ckpoint_rvd->vdev_children; c++) {
7754 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[c];
7755 vdev_t *current_vd = current_rvd->vdev_child[c];
7756
7757 space_map_t *checkpoint_sm = NULL;
7758 uint64_t checkpoint_sm_obj;
7759
7760 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7761 /*
7762 * Since we don't allow device removal in a pool
7763 * that has a checkpoint, we expect that all removed
7764 * vdevs were removed from the pool before the
7765 * checkpoint.
7766 */
7767 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7768 continue;
7769 }
7770
7771 /*
7772 * If the checkpoint space map doesn't exist, then nothing
7773 * here is checkpointed so there's nothing to verify.
7774 */
7775 if (current_vd->vdev_top_zap == 0 ||
7776 zap_contains(spa_meta_objset(current),
7777 current_vd->vdev_top_zap,
7778 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
7779 continue;
7780
7781 VERIFY0(zap_lookup(spa_meta_objset(current),
7782 current_vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
7783 sizeof (uint64_t), 1, &checkpoint_sm_obj));
7784
7785 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current),
7786 checkpoint_sm_obj, 0, current_vd->vdev_asize,
7787 current_vd->vdev_ashift));
7788
7789 verify_checkpoint_sm_entry_cb_arg_t vcsec;
7790 vcsec.vcsec_vd = ckpoint_vd;
7791 vcsec.vcsec_entryid = 0;
7792 vcsec.vcsec_num_entries =
7793 space_map_length(checkpoint_sm) / sizeof (uint64_t);
7794 VERIFY0(space_map_iterate(checkpoint_sm,
7795 space_map_length(checkpoint_sm),
7796 verify_checkpoint_sm_entry_cb, &vcsec));
7797 if (dump_opt['m'] > 3)
7798 dump_spacemap(current->spa_meta_objset, checkpoint_sm);
7799 space_map_close(checkpoint_sm);
7800 }
7801
7802 /*
7803 * If we've added vdevs since we took the checkpoint, ensure
7804 * that their checkpoint space maps are empty.
7805 */
7806 if (ckpoint_rvd->vdev_children < current_rvd->vdev_children) {
7807 for (uint64_t c = ckpoint_rvd->vdev_children;
7808 c < current_rvd->vdev_children; c++) {
7809 vdev_t *current_vd = current_rvd->vdev_child[c];
7810 VERIFY3P(current_vd->vdev_checkpoint_sm, ==, NULL);
7811 }
7812 }
7813
7814 /* for cleaner progress output */
7815 (void) fprintf(stderr, "\n");
7816 }
7817
7818 /*
7819 * Verifies that all space that's allocated in the checkpoint is
7820 * still allocated in the current version, by checking that everything
7821 * in checkpoint's ms_allocatable (which is actually allocated, not
7822 * allocatable/free) is not present in current's ms_allocatable.
7823 *
7824 * Note that the function changes the state of the ms_allocatable
7825 * trees of both spas when called. The entries of all ms_allocatable
7826 * trees are cleared out and then repopulated from their respective
7827 * ms_sm space maps. In the checkpointed state we load the allocated
7828 * entries, and in the current state we load the free entries.
7829 */
7830 static void
7831 verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_t *current)
7832 {
7833 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7834 vdev_t *current_rvd = current->spa_root_vdev;
7835
7836 load_concrete_ms_allocatable_trees(checkpoint, SM_ALLOC);
7837 load_concrete_ms_allocatable_trees(current, SM_FREE);
7838
7839 for (uint64_t i = 0; i < ckpoint_rvd->vdev_children; i++) {
7840 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[i];
7841 vdev_t *current_vd = current_rvd->vdev_child[i];
7842
7843 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7844 /*
7845 * See comment in verify_checkpoint_vdev_spacemaps()
7846 */
7847 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7848 continue;
7849 }
7850
7851 for (uint64_t m = 0; m < ckpoint_vd->vdev_ms_count; m++) {
7852 metaslab_t *ckpoint_msp = ckpoint_vd->vdev_ms[m];
7853 metaslab_t *current_msp = current_vd->vdev_ms[m];
7854
7855 (void) fprintf(stderr,
7856 "\rverifying vdev %llu of %llu, "
7857 "metaslab %llu of %llu ...",
7858 (longlong_t)current_vd->vdev_id,
7859 (longlong_t)current_rvd->vdev_children,
7860 (longlong_t)current_vd->vdev_ms[m]->ms_id,
7861 (longlong_t)current_vd->vdev_ms_count);
7862
7863 /*
7864 * We walk through the ms_allocatable trees that
7865 * are loaded with the allocated blocks from the
7866 * ms_sm spacemaps of the checkpoint. For each
7867 * one of these ranges we ensure that none of them
7868 * exists in the ms_allocatable trees of the
7869 * current state which are loaded with the ranges
7870 * that are currently free.
7871 *
7872 * This way we ensure that none of the blocks that
7873 * are part of the checkpoint were freed by mistake.
7874 */
7875 range_tree_walk(ckpoint_msp->ms_allocatable,
7876 (range_tree_func_t *)range_tree_verify_not_present,
7877 current_msp->ms_allocatable);
7878 }
7879 }
7880
7881 /* for cleaner progress output */
7882 (void) fprintf(stderr, "\n");
7883 }
7884
7885 static void
7886 verify_checkpoint_blocks(spa_t *spa)
7887 {
7888 ASSERT(!dump_opt['L']);
7889
7890 spa_t *checkpoint_spa;
7891 char *checkpoint_pool;
7892 int error = 0;
7893
7894 /*
7895 * We import the checkpointed state of the pool (under a different
7896 * name) so we can do verification on it against the current state
7897 * of the pool.
7898 */
7899 checkpoint_pool = import_checkpointed_state(spa->spa_name, NULL,
7900 NULL);
7901 ASSERT(strcmp(spa->spa_name, checkpoint_pool) != 0);
7902
7903 error = spa_open(checkpoint_pool, &checkpoint_spa, FTAG);
7904 if (error != 0) {
7905 fatal("Tried to open pool \"%s\" but spa_open() failed with "
7906 "error %d\n", checkpoint_pool, error);
7907 }
7908
7909 /*
7910 * Ensure that ranges in the checkpoint space maps of each vdev
7911 * are allocated according to the checkpointed state's metaslab
7912 * space maps.
7913 */
7914 verify_checkpoint_vdev_spacemaps(checkpoint_spa, spa);
7915
7916 /*
7917 * Ensure that allocated ranges in the checkpoint's metaslab
7918 * space maps remain allocated in the metaslab space maps of
7919 * the current state.
7920 */
7921 verify_checkpoint_ms_spacemaps(checkpoint_spa, spa);
7922
7923 /*
7924 * Once we are done, we get rid of the checkpointed state.
7925 */
7926 spa_close(checkpoint_spa, FTAG);
7927 free(checkpoint_pool);
7928 }
7929
7930 static void
7931 dump_leftover_checkpoint_blocks(spa_t *spa)
7932 {
7933 vdev_t *rvd = spa->spa_root_vdev;
7934
7935 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
7936 vdev_t *vd = rvd->vdev_child[i];
7937
7938 space_map_t *checkpoint_sm = NULL;
7939 uint64_t checkpoint_sm_obj;
7940
7941 if (vd->vdev_top_zap == 0)
7942 continue;
7943
7944 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
7945 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
7946 continue;
7947
7948 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
7949 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
7950 sizeof (uint64_t), 1, &checkpoint_sm_obj));
7951
7952 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
7953 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
7954 dump_spacemap(spa->spa_meta_objset, checkpoint_sm);
7955 space_map_close(checkpoint_sm);
7956 }
7957 }
7958
7959 static int
7960 verify_checkpoint(spa_t *spa)
7961 {
7962 uberblock_t checkpoint;
7963 int error;
7964
7965 if (!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT))
7966 return (0);
7967
7968 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
7969 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
7970 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
7971
7972 if (error == ENOENT && !dump_opt['L']) {
7973 /*
7974 * If the feature is active but the uberblock is missing
7975 * then we must be in the middle of discarding the
7976 * checkpoint.
7977 */
7978 (void) printf("\nPartially discarded checkpoint "
7979 "state found:\n");
7980 if (dump_opt['m'] > 3)
7981 dump_leftover_checkpoint_blocks(spa);
7982 return (0);
7983 } else if (error != 0) {
7984 (void) printf("lookup error %d when looking for "
7985 "checkpointed uberblock in MOS\n", error);
7986 return (error);
7987 }
7988 dump_uberblock(&checkpoint, "\nCheckpointed uberblock found:\n", "\n");
7989
7990 if (checkpoint.ub_checkpoint_txg == 0) {
7991 (void) printf("\nub_checkpoint_txg not set in checkpointed "
7992 "uberblock\n");
7993 error = 3;
7994 }
7995
7996 if (error == 0 && !dump_opt['L'])
7997 verify_checkpoint_blocks(spa);
7998
7999 return (error);
8000 }
8001
8002 static void
8003 mos_leaks_cb(void *arg, uint64_t start, uint64_t size)
8004 {
8005 (void) arg;
8006 for (uint64_t i = start; i < size; i++) {
8007 (void) printf("MOS object %llu referenced but not allocated\n",
8008 (u_longlong_t)i);
8009 }
8010 }
8011
8012 static void
8013 mos_obj_refd(uint64_t obj)
8014 {
8015 if (obj != 0 && mos_refd_objs != NULL)
8016 range_tree_add(mos_refd_objs, obj, 1);
8017 }
8018
8019 /*
8020 * Call on a MOS object that may already have been referenced.
8021 */
8022 static void
8023 mos_obj_refd_multiple(uint64_t obj)
8024 {
8025 if (obj != 0 && mos_refd_objs != NULL &&
8026 !range_tree_contains(mos_refd_objs, obj, 1))
8027 range_tree_add(mos_refd_objs, obj, 1);
8028 }
8029
8030 static void
8031 mos_leak_vdev_top_zap(vdev_t *vd)
8032 {
8033 uint64_t ms_flush_data_obj;
8034 int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
8035 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
8036 sizeof (ms_flush_data_obj), 1, &ms_flush_data_obj);
8037 if (error == ENOENT)
8038 return;
8039 ASSERT0(error);
8040
8041 mos_obj_refd(ms_flush_data_obj);
8042 }
8043
8044 static void
8045 mos_leak_vdev(vdev_t *vd)
8046 {
8047 mos_obj_refd(vd->vdev_dtl_object);
8048 mos_obj_refd(vd->vdev_ms_array);
8049 mos_obj_refd(vd->vdev_indirect_config.vic_births_object);
8050 mos_obj_refd(vd->vdev_indirect_config.vic_mapping_object);
8051 mos_obj_refd(vd->vdev_leaf_zap);
8052 if (vd->vdev_checkpoint_sm != NULL)
8053 mos_obj_refd(vd->vdev_checkpoint_sm->sm_object);
8054 if (vd->vdev_indirect_mapping != NULL) {
8055 mos_obj_refd(vd->vdev_indirect_mapping->
8056 vim_phys->vimp_counts_object);
8057 }
8058 if (vd->vdev_obsolete_sm != NULL)
8059 mos_obj_refd(vd->vdev_obsolete_sm->sm_object);
8060
8061 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
8062 metaslab_t *ms = vd->vdev_ms[m];
8063 mos_obj_refd(space_map_object(ms->ms_sm));
8064 }
8065
8066 if (vd->vdev_root_zap != 0)
8067 mos_obj_refd(vd->vdev_root_zap);
8068
8069 if (vd->vdev_top_zap != 0) {
8070 mos_obj_refd(vd->vdev_top_zap);
8071 mos_leak_vdev_top_zap(vd);
8072 }
8073
8074 for (uint64_t c = 0; c < vd->vdev_children; c++) {
8075 mos_leak_vdev(vd->vdev_child[c]);
8076 }
8077 }
8078
8079 static void
8080 mos_leak_log_spacemaps(spa_t *spa)
8081 {
8082 uint64_t spacemap_zap;
8083 int error = zap_lookup(spa_meta_objset(spa),
8084 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LOG_SPACEMAP_ZAP,
8085 sizeof (spacemap_zap), 1, &spacemap_zap);
8086 if (error == ENOENT)
8087 return;
8088 ASSERT0(error);
8089
8090 mos_obj_refd(spacemap_zap);
8091 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
8092 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls))
8093 mos_obj_refd(sls->sls_sm_obj);
8094 }
8095
8096 static void
8097 errorlog_count_refd(objset_t *mos, uint64_t errlog)
8098 {
8099 zap_cursor_t zc;
8100 zap_attribute_t *za = zap_attribute_alloc();
8101 for (zap_cursor_init(&zc, mos, errlog);
8102 zap_cursor_retrieve(&zc, za) == 0;
8103 zap_cursor_advance(&zc)) {
8104 mos_obj_refd(za->za_first_integer);
8105 }
8106 zap_cursor_fini(&zc);
8107 zap_attribute_free(za);
8108 }
8109
8110 static int
8111 dump_mos_leaks(spa_t *spa)
8112 {
8113 int rv = 0;
8114 objset_t *mos = spa->spa_meta_objset;
8115 dsl_pool_t *dp = spa->spa_dsl_pool;
8116
8117 /* Visit and mark all referenced objects in the MOS */
8118
8119 mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT);
8120 mos_obj_refd(spa->spa_pool_props_object);
8121 mos_obj_refd(spa->spa_config_object);
8122 mos_obj_refd(spa->spa_ddt_stat_object);
8123 mos_obj_refd(spa->spa_feat_desc_obj);
8124 mos_obj_refd(spa->spa_feat_enabled_txg_obj);
8125 mos_obj_refd(spa->spa_feat_for_read_obj);
8126 mos_obj_refd(spa->spa_feat_for_write_obj);
8127 mos_obj_refd(spa->spa_history);
8128 mos_obj_refd(spa->spa_errlog_last);
8129 mos_obj_refd(spa->spa_errlog_scrub);
8130
8131 if (spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) {
8132 errorlog_count_refd(mos, spa->spa_errlog_last);
8133 errorlog_count_refd(mos, spa->spa_errlog_scrub);
8134 }
8135
8136 mos_obj_refd(spa->spa_all_vdev_zaps);
8137 mos_obj_refd(spa->spa_dsl_pool->dp_bptree_obj);
8138 mos_obj_refd(spa->spa_dsl_pool->dp_tmp_userrefs_obj);
8139 mos_obj_refd(spa->spa_dsl_pool->dp_scan->scn_phys.scn_queue_obj);
8140 bpobj_count_refd(&spa->spa_deferred_bpobj);
8141 mos_obj_refd(dp->dp_empty_bpobj);
8142 bpobj_count_refd(&dp->dp_obsolete_bpobj);
8143 bpobj_count_refd(&dp->dp_free_bpobj);
8144 mos_obj_refd(spa->spa_l2cache.sav_object);
8145 mos_obj_refd(spa->spa_spares.sav_object);
8146
8147 if (spa->spa_syncing_log_sm != NULL)
8148 mos_obj_refd(spa->spa_syncing_log_sm->sm_object);
8149 mos_leak_log_spacemaps(spa);
8150
8151 mos_obj_refd(spa->spa_condensing_indirect_phys.
8152 scip_next_mapping_object);
8153 mos_obj_refd(spa->spa_condensing_indirect_phys.
8154 scip_prev_obsolete_sm_object);
8155 if (spa->spa_condensing_indirect_phys.scip_next_mapping_object != 0) {
8156 vdev_indirect_mapping_t *vim =
8157 vdev_indirect_mapping_open(mos,
8158 spa->spa_condensing_indirect_phys.scip_next_mapping_object);
8159 mos_obj_refd(vim->vim_phys->vimp_counts_object);
8160 vdev_indirect_mapping_close(vim);
8161 }
8162 deleted_livelists_dump_mos(spa);
8163
8164 if (dp->dp_origin_snap != NULL) {
8165 dsl_dataset_t *ds;
8166
8167 dsl_pool_config_enter(dp, FTAG);
8168 VERIFY0(dsl_dataset_hold_obj(dp,
8169 dsl_dataset_phys(dp->dp_origin_snap)->ds_next_snap_obj,
8170 FTAG, &ds));
8171 count_ds_mos_objects(ds);
8172 dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
8173 dsl_dataset_rele(ds, FTAG);
8174 dsl_pool_config_exit(dp, FTAG);
8175
8176 count_ds_mos_objects(dp->dp_origin_snap);
8177 dump_blkptr_list(&dp->dp_origin_snap->ds_deadlist, "Deadlist");
8178 }
8179 count_dir_mos_objects(dp->dp_mos_dir);
8180 if (dp->dp_free_dir != NULL)
8181 count_dir_mos_objects(dp->dp_free_dir);
8182 if (dp->dp_leak_dir != NULL)
8183 count_dir_mos_objects(dp->dp_leak_dir);
8184
8185 mos_leak_vdev(spa->spa_root_vdev);
8186
8187 for (uint64_t c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
8188 ddt_t *ddt = spa->spa_ddt[c];
8189 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
8190 continue;
8191
8192 /* DDT store objects */
8193 for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
8194 for (ddt_class_t class = 0; class < DDT_CLASSES;
8195 class++) {
8196 mos_obj_refd(ddt->ddt_object[type][class]);
8197 }
8198 }
8199
8200 /* FDT container */
8201 if (ddt->ddt_version == DDT_VERSION_FDT)
8202 mos_obj_refd(ddt->ddt_dir_object);
8203
8204 /* FDT log objects */
8205 if (ddt->ddt_flags & DDT_FLAG_LOG) {
8206 mos_obj_refd(ddt->ddt_log[0].ddl_object);
8207 mos_obj_refd(ddt->ddt_log[1].ddl_object);
8208 }
8209 }
8210
8211 if (spa->spa_brt != NULL) {
8212 brt_t *brt = spa->spa_brt;
8213 for (uint64_t vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) {
8214 brt_vdev_t *brtvd = &brt->brt_vdevs[vdevid];
8215 if (brtvd != NULL && brtvd->bv_initiated) {
8216 mos_obj_refd(brtvd->bv_mos_brtvdev);
8217 mos_obj_refd(brtvd->bv_mos_entries);
8218 }
8219 }
8220 }
8221
8222 /*
8223 * Visit all allocated objects and make sure they are referenced.
8224 */
8225 uint64_t object = 0;
8226 while (dmu_object_next(mos, &object, B_FALSE, 0) == 0) {
8227 if (range_tree_contains(mos_refd_objs, object, 1)) {
8228 range_tree_remove(mos_refd_objs, object, 1);
8229 } else {
8230 dmu_object_info_t doi;
8231 const char *name;
8232 VERIFY0(dmu_object_info(mos, object, &doi));
8233 if (doi.doi_type & DMU_OT_NEWTYPE) {
8234 dmu_object_byteswap_t bswap =
8235 DMU_OT_BYTESWAP(doi.doi_type);
8236 name = dmu_ot_byteswap[bswap].ob_name;
8237 } else {
8238 name = dmu_ot[doi.doi_type].ot_name;
8239 }
8240
8241 (void) printf("MOS object %llu (%s) leaked\n",
8242 (u_longlong_t)object, name);
8243 rv = 2;
8244 }
8245 }
8246 (void) range_tree_walk(mos_refd_objs, mos_leaks_cb, NULL);
8247 if (!range_tree_is_empty(mos_refd_objs))
8248 rv = 2;
8249 range_tree_vacate(mos_refd_objs, NULL, NULL);
8250 range_tree_destroy(mos_refd_objs);
8251 return (rv);
8252 }
8253
8254 typedef struct log_sm_obsolete_stats_arg {
8255 uint64_t lsos_current_txg;
8256
8257 uint64_t lsos_total_entries;
8258 uint64_t lsos_valid_entries;
8259
8260 uint64_t lsos_sm_entries;
8261 uint64_t lsos_valid_sm_entries;
8262 } log_sm_obsolete_stats_arg_t;
8263
8264 static int
8265 log_spacemap_obsolete_stats_cb(spa_t *spa, space_map_entry_t *sme,
8266 uint64_t txg, void *arg)
8267 {
8268 log_sm_obsolete_stats_arg_t *lsos = arg;
8269
8270 uint64_t offset = sme->sme_offset;
8271 uint64_t vdev_id = sme->sme_vdev;
8272
8273 if (lsos->lsos_current_txg == 0) {
8274 /* this is the first log */
8275 lsos->lsos_current_txg = txg;
8276 } else if (lsos->lsos_current_txg < txg) {
8277 /* we just changed log - print stats and reset */
8278 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8279 (u_longlong_t)lsos->lsos_valid_sm_entries,
8280 (u_longlong_t)lsos->lsos_sm_entries,
8281 (u_longlong_t)lsos->lsos_current_txg);
8282 lsos->lsos_valid_sm_entries = 0;
8283 lsos->lsos_sm_entries = 0;
8284 lsos->lsos_current_txg = txg;
8285 }
8286 ASSERT3U(lsos->lsos_current_txg, ==, txg);
8287
8288 lsos->lsos_sm_entries++;
8289 lsos->lsos_total_entries++;
8290
8291 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
8292 if (!vdev_is_concrete(vd))
8293 return (0);
8294
8295 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
8296 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
8297
8298 if (txg < metaslab_unflushed_txg(ms))
8299 return (0);
8300 lsos->lsos_valid_sm_entries++;
8301 lsos->lsos_valid_entries++;
8302 return (0);
8303 }
8304
8305 static void
8306 dump_log_spacemap_obsolete_stats(spa_t *spa)
8307 {
8308 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
8309 return;
8310
8311 log_sm_obsolete_stats_arg_t lsos = {0};
8312
8313 (void) printf("Log Space Map Obsolete Entry Statistics:\n");
8314
8315 iterate_through_spacemap_logs(spa,
8316 log_spacemap_obsolete_stats_cb, &lsos);
8317
8318 /* print stats for latest log */
8319 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8320 (u_longlong_t)lsos.lsos_valid_sm_entries,
8321 (u_longlong_t)lsos.lsos_sm_entries,
8322 (u_longlong_t)lsos.lsos_current_txg);
8323
8324 (void) printf("%-8llu valid entries out of %-8llu - total\n\n",
8325 (u_longlong_t)lsos.lsos_valid_entries,
8326 (u_longlong_t)lsos.lsos_total_entries);
8327 }
8328
8329 static void
8330 dump_zpool(spa_t *spa)
8331 {
8332 dsl_pool_t *dp = spa_get_dsl(spa);
8333 int rc = 0;
8334
8335 if (dump_opt['y']) {
8336 livelist_metaslab_validate(spa);
8337 }
8338
8339 if (dump_opt['S']) {
8340 dump_simulated_ddt(spa);
8341 return;
8342 }
8343
8344 if (!dump_opt['e'] && dump_opt['C'] > 1) {
8345 (void) printf("\nCached configuration:\n");
8346 dump_nvlist(spa->spa_config, 8);
8347 }
8348
8349 if (dump_opt['C'])
8350 dump_config(spa);
8351
8352 if (dump_opt['u'])
8353 dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n");
8354
8355 if (dump_opt['D'])
8356 dump_all_ddts(spa);
8357
8358 if (dump_opt['T'])
8359 dump_brt(spa);
8360
8361 if (dump_opt['d'] > 2 || dump_opt['m'])
8362 dump_metaslabs(spa);
8363 if (dump_opt['M'])
8364 dump_metaslab_groups(spa, dump_opt['M'] > 1);
8365 if (dump_opt['d'] > 2 || dump_opt['m']) {
8366 dump_log_spacemaps(spa);
8367 dump_log_spacemap_obsolete_stats(spa);
8368 }
8369
8370 if (dump_opt['d'] || dump_opt['i']) {
8371 spa_feature_t f;
8372 mos_refd_objs = range_tree_create(NULL, RANGE_SEG64, NULL, 0,
8373 0);
8374 dump_objset(dp->dp_meta_objset);
8375
8376 if (dump_opt['d'] >= 3) {
8377 dsl_pool_t *dp = spa->spa_dsl_pool;
8378 dump_full_bpobj(&spa->spa_deferred_bpobj,
8379 "Deferred frees", 0);
8380 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
8381 dump_full_bpobj(&dp->dp_free_bpobj,
8382 "Pool snapshot frees", 0);
8383 }
8384 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
8385 ASSERT(spa_feature_is_enabled(spa,
8386 SPA_FEATURE_DEVICE_REMOVAL));
8387 dump_full_bpobj(&dp->dp_obsolete_bpobj,
8388 "Pool obsolete blocks", 0);
8389 }
8390
8391 if (spa_feature_is_active(spa,
8392 SPA_FEATURE_ASYNC_DESTROY)) {
8393 dump_bptree(spa->spa_meta_objset,
8394 dp->dp_bptree_obj,
8395 "Pool dataset frees");
8396 }
8397 dump_dtl(spa->spa_root_vdev, 0);
8398 }
8399
8400 for (spa_feature_t f = 0; f < SPA_FEATURES; f++)
8401 global_feature_count[f] = UINT64_MAX;
8402 global_feature_count[SPA_FEATURE_REDACTION_BOOKMARKS] = 0;
8403 global_feature_count[SPA_FEATURE_REDACTION_LIST_SPILL] = 0;
8404 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN] = 0;
8405 global_feature_count[SPA_FEATURE_LIVELIST] = 0;
8406
8407 (void) dmu_objset_find(spa_name(spa), dump_one_objset,
8408 NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
8409
8410 if (rc == 0 && !dump_opt['L'])
8411 rc = dump_mos_leaks(spa);
8412
8413 for (f = 0; f < SPA_FEATURES; f++) {
8414 uint64_t refcount;
8415
8416 uint64_t *arr;
8417 if (!(spa_feature_table[f].fi_flags &
8418 ZFEATURE_FLAG_PER_DATASET)) {
8419 if (global_feature_count[f] == UINT64_MAX)
8420 continue;
8421 if (!spa_feature_is_enabled(spa, f)) {
8422 ASSERT0(global_feature_count[f]);
8423 continue;
8424 }
8425 arr = global_feature_count;
8426 } else {
8427 if (!spa_feature_is_enabled(spa, f)) {
8428 ASSERT0(dataset_feature_count[f]);
8429 continue;
8430 }
8431 arr = dataset_feature_count;
8432 }
8433 if (feature_get_refcount(spa, &spa_feature_table[f],
8434 &refcount) == ENOTSUP)
8435 continue;
8436 if (arr[f] != refcount) {
8437 (void) printf("%s feature refcount mismatch: "
8438 "%lld consumers != %lld refcount\n",
8439 spa_feature_table[f].fi_uname,
8440 (longlong_t)arr[f], (longlong_t)refcount);
8441 rc = 2;
8442 } else {
8443 (void) printf("Verified %s feature refcount "
8444 "of %llu is correct\n",
8445 spa_feature_table[f].fi_uname,
8446 (longlong_t)refcount);
8447 }
8448 }
8449
8450 if (rc == 0)
8451 rc = verify_device_removal_feature_counts(spa);
8452 }
8453
8454 if (rc == 0 && (dump_opt['b'] || dump_opt['c']))
8455 rc = dump_block_stats(spa);
8456
8457 if (rc == 0)
8458 rc = verify_spacemap_refcounts(spa);
8459
8460 if (dump_opt['s'])
8461 show_pool_stats(spa);
8462
8463 if (dump_opt['h'])
8464 dump_history(spa);
8465
8466 if (rc == 0)
8467 rc = verify_checkpoint(spa);
8468
8469 if (rc != 0) {
8470 dump_debug_buffer();
8471 zdb_exit(rc);
8472 }
8473 }
8474
8475 #define ZDB_FLAG_CHECKSUM 0x0001
8476 #define ZDB_FLAG_DECOMPRESS 0x0002
8477 #define ZDB_FLAG_BSWAP 0x0004
8478 #define ZDB_FLAG_GBH 0x0008
8479 #define ZDB_FLAG_INDIRECT 0x0010
8480 #define ZDB_FLAG_RAW 0x0020
8481 #define ZDB_FLAG_PRINT_BLKPTR 0x0040
8482 #define ZDB_FLAG_VERBOSE 0x0080
8483
8484 static int flagbits[256];
8485 static char flagbitstr[16];
8486
8487 static void
8488 zdb_print_blkptr(const blkptr_t *bp, int flags)
8489 {
8490 char blkbuf[BP_SPRINTF_LEN];
8491
8492 if (flags & ZDB_FLAG_BSWAP)
8493 byteswap_uint64_array((void *)bp, sizeof (blkptr_t));
8494
8495 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
8496 (void) printf("%s\n", blkbuf);
8497 }
8498
8499 static void
8500 zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
8501 {
8502 int i;
8503
8504 for (i = 0; i < nbps; i++)
8505 zdb_print_blkptr(&bp[i], flags);
8506 }
8507
8508 static void
8509 zdb_dump_gbh(void *buf, int flags)
8510 {
8511 zdb_dump_indirect((blkptr_t *)buf, SPA_GBH_NBLKPTRS, flags);
8512 }
8513
8514 static void
8515 zdb_dump_block_raw(void *buf, uint64_t size, int flags)
8516 {
8517 if (flags & ZDB_FLAG_BSWAP)
8518 byteswap_uint64_array(buf, size);
8519 VERIFY(write(fileno(stdout), buf, size) == size);
8520 }
8521
8522 static void
8523 zdb_dump_block(char *label, void *buf, uint64_t size, int flags)
8524 {
8525 uint64_t *d = (uint64_t *)buf;
8526 unsigned nwords = size / sizeof (uint64_t);
8527 int do_bswap = !!(flags & ZDB_FLAG_BSWAP);
8528 unsigned i, j;
8529 const char *hdr;
8530 char *c;
8531
8532
8533 if (do_bswap)
8534 hdr = " 7 6 5 4 3 2 1 0 f e d c b a 9 8";
8535 else
8536 hdr = " 0 1 2 3 4 5 6 7 8 9 a b c d e f";
8537
8538 (void) printf("\n%s\n%6s %s 0123456789abcdef\n", label, "", hdr);
8539
8540 #ifdef _ZFS_LITTLE_ENDIAN
8541 /* correct the endianness */
8542 do_bswap = !do_bswap;
8543 #endif
8544 for (i = 0; i < nwords; i += 2) {
8545 (void) printf("%06llx: %016llx %016llx ",
8546 (u_longlong_t)(i * sizeof (uint64_t)),
8547 (u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]),
8548 (u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1]));
8549
8550 c = (char *)&d[i];
8551 for (j = 0; j < 2 * sizeof (uint64_t); j++)
8552 (void) printf("%c", isprint(c[j]) ? c[j] : '.');
8553 (void) printf("\n");
8554 }
8555 }
8556
8557 /*
8558 * There are two acceptable formats:
8559 * leaf_name - For example: c1t0d0 or /tmp/ztest.0a
8560 * child[.child]* - For example: 0.1.1
8561 *
8562 * The second form can be used to specify arbitrary vdevs anywhere
8563 * in the hierarchy. For example, in a pool with a mirror of
8564 * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
8565 */
8566 static vdev_t *
8567 zdb_vdev_lookup(vdev_t *vdev, const char *path)
8568 {
8569 char *s, *p, *q;
8570 unsigned i;
8571
8572 if (vdev == NULL)
8573 return (NULL);
8574
8575 /* First, assume the x.x.x.x format */
8576 i = strtoul(path, &s, 10);
8577 if (s == path || (s && *s != '.' && *s != '\0'))
8578 goto name;
8579 if (i >= vdev->vdev_children)
8580 return (NULL);
8581
8582 vdev = vdev->vdev_child[i];
8583 if (s && *s == '\0')
8584 return (vdev);
8585 return (zdb_vdev_lookup(vdev, s+1));
8586
8587 name:
8588 for (i = 0; i < vdev->vdev_children; i++) {
8589 vdev_t *vc = vdev->vdev_child[i];
8590
8591 if (vc->vdev_path == NULL) {
8592 vc = zdb_vdev_lookup(vc, path);
8593 if (vc == NULL)
8594 continue;
8595 else
8596 return (vc);
8597 }
8598
8599 p = strrchr(vc->vdev_path, '/');
8600 p = p ? p + 1 : vc->vdev_path;
8601 q = &vc->vdev_path[strlen(vc->vdev_path) - 2];
8602
8603 if (strcmp(vc->vdev_path, path) == 0)
8604 return (vc);
8605 if (strcmp(p, path) == 0)
8606 return (vc);
8607 if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0)
8608 return (vc);
8609 }
8610
8611 return (NULL);
8612 }
8613
8614 static int
8615 name_from_objset_id(spa_t *spa, uint64_t objset_id, char *outstr)
8616 {
8617 dsl_dataset_t *ds;
8618
8619 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
8620 int error = dsl_dataset_hold_obj(spa->spa_dsl_pool, objset_id,
8621 NULL, &ds);
8622 if (error != 0) {
8623 (void) fprintf(stderr, "failed to hold objset %llu: %s\n",
8624 (u_longlong_t)objset_id, strerror(error));
8625 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8626 return (error);
8627 }
8628 dsl_dataset_name(ds, outstr);
8629 dsl_dataset_rele(ds, NULL);
8630 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8631 return (0);
8632 }
8633
8634 static boolean_t
8635 zdb_parse_block_sizes(char *sizes, uint64_t *lsize, uint64_t *psize)
8636 {
8637 char *s0, *s1, *tmp = NULL;
8638
8639 if (sizes == NULL)
8640 return (B_FALSE);
8641
8642 s0 = strtok_r(sizes, "/", &tmp);
8643 if (s0 == NULL)
8644 return (B_FALSE);
8645 s1 = strtok_r(NULL, "/", &tmp);
8646 *lsize = strtoull(s0, NULL, 16);
8647 *psize = s1 ? strtoull(s1, NULL, 16) : *lsize;
8648 return (*lsize >= *psize && *psize > 0);
8649 }
8650
8651 #define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg))
8652
8653 static boolean_t
8654 try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize,
8655 int flags, int cfunc, void *lbuf, void *lbuf2)
8656 {
8657 if (flags & ZDB_FLAG_VERBOSE) {
8658 (void) fprintf(stderr,
8659 "Trying %05llx -> %05llx (%s)\n",
8660 (u_longlong_t)psize,
8661 (u_longlong_t)lsize,
8662 zio_compress_table[cfunc].ci_name);
8663 }
8664
8665 /*
8666 * We set lbuf to all zeros and lbuf2 to all
8667 * ones, then decompress to both buffers and
8668 * compare their contents. This way we can
8669 * know if decompression filled exactly to
8670 * lsize or if it left some bytes unwritten.
8671 */
8672
8673 memset(lbuf, 0x00, lsize);
8674 memset(lbuf2, 0xff, lsize);
8675
8676 abd_t labd, labd2;
8677 abd_get_from_buf_struct(&labd, lbuf, lsize);
8678 abd_get_from_buf_struct(&labd2, lbuf2, lsize);
8679
8680 boolean_t ret = B_FALSE;
8681 if (zio_decompress_data(cfunc, pabd,
8682 &labd, psize, lsize, NULL) == 0 &&
8683 zio_decompress_data(cfunc, pabd,
8684 &labd2, psize, lsize, NULL) == 0 &&
8685 memcmp(lbuf, lbuf2, lsize) == 0)
8686 ret = B_TRUE;
8687
8688 abd_free(&labd2);
8689 abd_free(&labd);
8690
8691 return (ret);
8692 }
8693
8694 static uint64_t
8695 zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
8696 uint64_t psize, int flags)
8697 {
8698 (void) buf;
8699 uint64_t orig_lsize = lsize;
8700 boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL));
8701 boolean_t found = B_FALSE;
8702 /*
8703 * We don't know how the data was compressed, so just try
8704 * every decompress function at every inflated blocksize.
8705 */
8706 void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
8707 int cfuncs[ZIO_COMPRESS_FUNCTIONS] = { 0 };
8708 int *cfuncp = cfuncs;
8709 uint64_t maxlsize = SPA_MAXBLOCKSIZE;
8710 uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) |
8711 ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) |
8712 ZIO_COMPRESS_MASK(ZLE);
8713 *cfuncp++ = ZIO_COMPRESS_LZ4;
8714 *cfuncp++ = ZIO_COMPRESS_LZJB;
8715 mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB);
8716 /*
8717 * Every gzip level has the same decompressor, no need to
8718 * run it 9 times per bruteforce attempt.
8719 */
8720 mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3);
8721 mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5);
8722 mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7);
8723 mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9);
8724 for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++)
8725 if (((1ULL << c) & mask) == 0)
8726 *cfuncp++ = c;
8727
8728 /*
8729 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
8730 * could take a while and we should let the user know
8731 * we are not stuck. On the other hand, printing progress
8732 * info gets old after a while. User can specify 'v' flag
8733 * to see the progression.
8734 */
8735 if (lsize == psize)
8736 lsize += SPA_MINBLOCKSIZE;
8737 else
8738 maxlsize = lsize;
8739
8740 for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) {
8741 for (cfuncp = cfuncs; *cfuncp; cfuncp++) {
8742 if (try_decompress_block(pabd, lsize, psize, flags,
8743 *cfuncp, lbuf, lbuf2)) {
8744 found = B_TRUE;
8745 break;
8746 }
8747 }
8748 if (*cfuncp != 0)
8749 break;
8750 }
8751 if (!found && tryzle) {
8752 for (lsize = orig_lsize; lsize <= maxlsize;
8753 lsize += SPA_MINBLOCKSIZE) {
8754 if (try_decompress_block(pabd, lsize, psize, flags,
8755 ZIO_COMPRESS_ZLE, lbuf, lbuf2)) {
8756 *cfuncp = ZIO_COMPRESS_ZLE;
8757 found = B_TRUE;
8758 break;
8759 }
8760 }
8761 }
8762 umem_free(lbuf2, SPA_MAXBLOCKSIZE);
8763
8764 if (*cfuncp == ZIO_COMPRESS_ZLE) {
8765 printf("\nZLE decompression was selected. If you "
8766 "suspect the results are wrong,\ntry avoiding ZLE "
8767 "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
8768 }
8769
8770 return (lsize > maxlsize ? -1 : lsize);
8771 }
8772
8773 /*
8774 * Read a block from a pool and print it out. The syntax of the
8775 * block descriptor is:
8776 *
8777 * pool:vdev_specifier:offset:[lsize/]psize[:flags]
8778 *
8779 * pool - The name of the pool you wish to read from
8780 * vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
8781 * offset - offset, in hex, in bytes
8782 * size - Amount of data to read, in hex, in bytes
8783 * flags - A string of characters specifying options
8784 * b: Decode a blkptr at given offset within block
8785 * c: Calculate and display checksums
8786 * d: Decompress data before dumping
8787 * e: Byteswap data before dumping
8788 * g: Display data as a gang block header
8789 * i: Display as an indirect block
8790 * r: Dump raw data to stdout
8791 * v: Verbose
8792 *
8793 */
8794 static void
8795 zdb_read_block(char *thing, spa_t *spa)
8796 {
8797 blkptr_t blk, *bp = &blk;
8798 dva_t *dva = bp->blk_dva;
8799 int flags = 0;
8800 uint64_t offset = 0, psize = 0, lsize = 0, blkptr_offset = 0;
8801 zio_t *zio;
8802 vdev_t *vd;
8803 abd_t *pabd;
8804 void *lbuf, *buf;
8805 char *s, *p, *dup, *flagstr, *sizes, *tmp = NULL;
8806 const char *vdev, *errmsg = NULL;
8807 int i, len, error;
8808 boolean_t borrowed = B_FALSE, found = B_FALSE;
8809
8810 dup = strdup(thing);
8811 s = strtok_r(dup, ":", &tmp);
8812 vdev = s ?: "";
8813 s = strtok_r(NULL, ":", &tmp);
8814 offset = strtoull(s ? s : "", NULL, 16);
8815 sizes = strtok_r(NULL, ":", &tmp);
8816 s = strtok_r(NULL, ":", &tmp);
8817 flagstr = strdup(s ?: "");
8818
8819 if (!zdb_parse_block_sizes(sizes, &lsize, &psize))
8820 errmsg = "invalid size(s)";
8821 if (!IS_P2ALIGNED(psize, DEV_BSIZE) || !IS_P2ALIGNED(lsize, DEV_BSIZE))
8822 errmsg = "size must be a multiple of sector size";
8823 if (!IS_P2ALIGNED(offset, DEV_BSIZE))
8824 errmsg = "offset must be a multiple of sector size";
8825 if (errmsg) {
8826 (void) printf("Invalid block specifier: %s - %s\n",
8827 thing, errmsg);
8828 goto done;
8829 }
8830
8831 tmp = NULL;
8832 for (s = strtok_r(flagstr, ":", &tmp);
8833 s != NULL;
8834 s = strtok_r(NULL, ":", &tmp)) {
8835 len = strlen(flagstr);
8836 for (i = 0; i < len; i++) {
8837 int bit = flagbits[(uchar_t)flagstr[i]];
8838
8839 if (bit == 0) {
8840 (void) printf("***Ignoring flag: %c\n",
8841 (uchar_t)flagstr[i]);
8842 continue;
8843 }
8844 found = B_TRUE;
8845 flags |= bit;
8846
8847 p = &flagstr[i + 1];
8848 if (*p != ':' && *p != '\0') {
8849 int j = 0, nextbit = flagbits[(uchar_t)*p];
8850 char *end, offstr[8] = { 0 };
8851 if ((bit == ZDB_FLAG_PRINT_BLKPTR) &&
8852 (nextbit == 0)) {
8853 /* look ahead to isolate the offset */
8854 while (nextbit == 0 &&
8855 strchr(flagbitstr, *p) == NULL) {
8856 offstr[j] = *p;
8857 j++;
8858 if (i + j > strlen(flagstr))
8859 break;
8860 p++;
8861 nextbit = flagbits[(uchar_t)*p];
8862 }
8863 blkptr_offset = strtoull(offstr, &end,
8864 16);
8865 i += j;
8866 } else if (nextbit == 0) {
8867 (void) printf("***Ignoring flag arg:"
8868 " '%c'\n", (uchar_t)*p);
8869 }
8870 }
8871 }
8872 }
8873 if (blkptr_offset % sizeof (blkptr_t)) {
8874 printf("Block pointer offset 0x%llx "
8875 "must be divisible by 0x%x\n",
8876 (longlong_t)blkptr_offset, (int)sizeof (blkptr_t));
8877 goto done;
8878 }
8879 if (found == B_FALSE && strlen(flagstr) > 0) {
8880 printf("Invalid flag arg: '%s'\n", flagstr);
8881 goto done;
8882 }
8883
8884 vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev);
8885 if (vd == NULL) {
8886 (void) printf("***Invalid vdev: %s\n", vdev);
8887 goto done;
8888 } else {
8889 if (vd->vdev_path)
8890 (void) fprintf(stderr, "Found vdev: %s\n",
8891 vd->vdev_path);
8892 else
8893 (void) fprintf(stderr, "Found vdev type: %s\n",
8894 vd->vdev_ops->vdev_op_type);
8895 }
8896
8897 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
8898 lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
8899
8900 BP_ZERO(bp);
8901
8902 DVA_SET_VDEV(&dva[0], vd->vdev_id);
8903 DVA_SET_OFFSET(&dva[0], offset);
8904 DVA_SET_GANG(&dva[0], !!(flags & ZDB_FLAG_GBH));
8905 DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
8906
8907 BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
8908
8909 BP_SET_LSIZE(bp, lsize);
8910 BP_SET_PSIZE(bp, psize);
8911 BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF);
8912 BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF);
8913 BP_SET_TYPE(bp, DMU_OT_NONE);
8914 BP_SET_LEVEL(bp, 0);
8915 BP_SET_DEDUP(bp, 0);
8916 BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
8917
8918 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8919 zio = zio_root(spa, NULL, NULL, 0);
8920
8921 if (vd == vd->vdev_top) {
8922 /*
8923 * Treat this as a normal block read.
8924 */
8925 zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL,
8926 ZIO_PRIORITY_SYNC_READ,
8927 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL));
8928 } else {
8929 /*
8930 * Treat this as a vdev child I/O.
8931 */
8932 zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd,
8933 psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ,
8934 ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
8935 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | ZIO_FLAG_OPTIONAL,
8936 NULL, NULL));
8937 }
8938
8939 error = zio_wait(zio);
8940 spa_config_exit(spa, SCL_STATE, FTAG);
8941
8942 if (error) {
8943 (void) printf("Read of %s failed, error: %d\n", thing, error);
8944 goto out;
8945 }
8946
8947 uint64_t orig_lsize = lsize;
8948 buf = lbuf;
8949 if (flags & ZDB_FLAG_DECOMPRESS) {
8950 lsize = zdb_decompress_block(pabd, buf, lbuf,
8951 lsize, psize, flags);
8952 if (lsize == -1) {
8953 (void) printf("Decompress of %s failed\n", thing);
8954 goto out;
8955 }
8956 } else {
8957 buf = abd_borrow_buf_copy(pabd, lsize);
8958 borrowed = B_TRUE;
8959 }
8960 /*
8961 * Try to detect invalid block pointer. If invalid, try
8962 * decompressing.
8963 */
8964 if ((flags & ZDB_FLAG_PRINT_BLKPTR || flags & ZDB_FLAG_INDIRECT) &&
8965 !(flags & ZDB_FLAG_DECOMPRESS)) {
8966 const blkptr_t *b = (const blkptr_t *)(void *)
8967 ((uintptr_t)buf + (uintptr_t)blkptr_offset);
8968 if (zfs_blkptr_verify(spa, b,
8969 BLK_CONFIG_NEEDED, BLK_VERIFY_ONLY) == B_FALSE) {
8970 abd_return_buf_copy(pabd, buf, lsize);
8971 borrowed = B_FALSE;
8972 buf = lbuf;
8973 lsize = zdb_decompress_block(pabd, buf,
8974 lbuf, lsize, psize, flags);
8975 b = (const blkptr_t *)(void *)
8976 ((uintptr_t)buf + (uintptr_t)blkptr_offset);
8977 if (lsize == -1 || zfs_blkptr_verify(spa, b,
8978 BLK_CONFIG_NEEDED, BLK_VERIFY_LOG) == B_FALSE) {
8979 printf("invalid block pointer at this DVA\n");
8980 goto out;
8981 }
8982 }
8983 }
8984
8985 if (flags & ZDB_FLAG_PRINT_BLKPTR)
8986 zdb_print_blkptr((blkptr_t *)(void *)
8987 ((uintptr_t)buf + (uintptr_t)blkptr_offset), flags);
8988 else if (flags & ZDB_FLAG_RAW)
8989 zdb_dump_block_raw(buf, lsize, flags);
8990 else if (flags & ZDB_FLAG_INDIRECT)
8991 zdb_dump_indirect((blkptr_t *)buf,
8992 orig_lsize / sizeof (blkptr_t), flags);
8993 else if (flags & ZDB_FLAG_GBH)
8994 zdb_dump_gbh(buf, flags);
8995 else
8996 zdb_dump_block(thing, buf, lsize, flags);
8997
8998 /*
8999 * If :c was specified, iterate through the checksum table to
9000 * calculate and display each checksum for our specified
9001 * DVA and length.
9002 */
9003 if ((flags & ZDB_FLAG_CHECKSUM) && !(flags & ZDB_FLAG_RAW) &&
9004 !(flags & ZDB_FLAG_GBH)) {
9005 zio_t *czio;
9006 (void) printf("\n");
9007 for (enum zio_checksum ck = ZIO_CHECKSUM_LABEL;
9008 ck < ZIO_CHECKSUM_FUNCTIONS; ck++) {
9009
9010 if ((zio_checksum_table[ck].ci_flags &
9011 ZCHECKSUM_FLAG_EMBEDDED) ||
9012 ck == ZIO_CHECKSUM_NOPARITY) {
9013 continue;
9014 }
9015 BP_SET_CHECKSUM(bp, ck);
9016 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9017 czio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9018 if (vd == vd->vdev_top) {
9019 zio_nowait(zio_read(czio, spa, bp, pabd, psize,
9020 NULL, NULL,
9021 ZIO_PRIORITY_SYNC_READ,
9022 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9023 ZIO_FLAG_DONT_RETRY, NULL));
9024 } else {
9025 zio_nowait(zio_vdev_child_io(czio, bp, vd,
9026 offset, pabd, psize, ZIO_TYPE_READ,
9027 ZIO_PRIORITY_SYNC_READ,
9028 ZIO_FLAG_DONT_PROPAGATE |
9029 ZIO_FLAG_DONT_RETRY |
9030 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9031 ZIO_FLAG_SPECULATIVE |
9032 ZIO_FLAG_OPTIONAL, NULL, NULL));
9033 }
9034 error = zio_wait(czio);
9035 if (error == 0 || error == ECKSUM) {
9036 zio_t *ck_zio = zio_null(NULL, spa, NULL,
9037 NULL, NULL, 0);
9038 ck_zio->io_offset =
9039 DVA_GET_OFFSET(&bp->blk_dva[0]);
9040 ck_zio->io_bp = bp;
9041 zio_checksum_compute(ck_zio, ck, pabd, lsize);
9042 printf(
9043 "%12s\t"
9044 "cksum=%016llx:%016llx:%016llx:%016llx\n",
9045 zio_checksum_table[ck].ci_name,
9046 (u_longlong_t)bp->blk_cksum.zc_word[0],
9047 (u_longlong_t)bp->blk_cksum.zc_word[1],
9048 (u_longlong_t)bp->blk_cksum.zc_word[2],
9049 (u_longlong_t)bp->blk_cksum.zc_word[3]);
9050 zio_wait(ck_zio);
9051 } else {
9052 printf("error %d reading block\n", error);
9053 }
9054 spa_config_exit(spa, SCL_STATE, FTAG);
9055 }
9056 }
9057
9058 if (borrowed)
9059 abd_return_buf_copy(pabd, buf, lsize);
9060
9061 out:
9062 abd_free(pabd);
9063 umem_free(lbuf, SPA_MAXBLOCKSIZE);
9064 done:
9065 free(flagstr);
9066 free(dup);
9067 }
9068
9069 static void
9070 zdb_embedded_block(char *thing)
9071 {
9072 blkptr_t bp = {{{{0}}}};
9073 unsigned long long *words = (void *)&bp;
9074 char *buf;
9075 int err;
9076
9077 err = sscanf(thing, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
9078 "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
9079 words + 0, words + 1, words + 2, words + 3,
9080 words + 4, words + 5, words + 6, words + 7,
9081 words + 8, words + 9, words + 10, words + 11,
9082 words + 12, words + 13, words + 14, words + 15);
9083 if (err != 16) {
9084 (void) fprintf(stderr, "invalid input format\n");
9085 zdb_exit(1);
9086 }
9087 ASSERT3U(BPE_GET_LSIZE(&bp), <=, SPA_MAXBLOCKSIZE);
9088 buf = malloc(SPA_MAXBLOCKSIZE);
9089 if (buf == NULL) {
9090 (void) fprintf(stderr, "out of memory\n");
9091 zdb_exit(1);
9092 }
9093 err = decode_embedded_bp(&bp, buf, BPE_GET_LSIZE(&bp));
9094 if (err != 0) {
9095 (void) fprintf(stderr, "decode failed: %u\n", err);
9096 zdb_exit(1);
9097 }
9098 zdb_dump_block_raw(buf, BPE_GET_LSIZE(&bp), 0);
9099 free(buf);
9100 }
9101
9102 /* check for valid hex or decimal numeric string */
9103 static boolean_t
9104 zdb_numeric(char *str)
9105 {
9106 int i = 0, len;
9107
9108 len = strlen(str);
9109 if (len == 0)
9110 return (B_FALSE);
9111 if (strncmp(str, "0x", 2) == 0 || strncmp(str, "0X", 2) == 0)
9112 i = 2;
9113 for (; i < len; i++) {
9114 if (!isxdigit(str[i]))
9115 return (B_FALSE);
9116 }
9117 return (B_TRUE);
9118 }
9119
9120 static int
9121 dummy_get_file_info(dmu_object_type_t bonustype, const void *data,
9122 zfs_file_info_t *zoi)
9123 {
9124 (void) data, (void) zoi;
9125
9126 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
9127 return (ENOENT);
9128
9129 (void) fprintf(stderr, "dummy_get_file_info: not implemented");
9130 abort();
9131 }
9132
9133 int
9134 main(int argc, char **argv)
9135 {
9136 int c;
9137 int dump_all = 1;
9138 int verbose = 0;
9139 int error = 0;
9140 char **searchdirs = NULL;
9141 int nsearch = 0;
9142 char *target, *target_pool, dsname[ZFS_MAX_DATASET_NAME_LEN];
9143 nvlist_t *policy = NULL;
9144 uint64_t max_txg = UINT64_MAX;
9145 int64_t objset_id = -1;
9146 uint64_t object;
9147 int flags = ZFS_IMPORT_MISSING_LOG;
9148 int rewind = ZPOOL_NEVER_REWIND;
9149 char *spa_config_path_env, *objset_str;
9150 boolean_t target_is_spa = B_TRUE, dataset_lookup = B_FALSE;
9151 nvlist_t *cfg = NULL;
9152 struct sigaction action;
9153 boolean_t force_import = B_FALSE;
9154 boolean_t config_path_console = B_FALSE;
9155 char pbuf[MAXPATHLEN];
9156
9157 dprintf_setup(&argc, argv);
9158
9159 /*
9160 * Set up signal handlers, so if we crash due to bad on-disk data we
9161 * can get more info. Unlike ztest, we don't bail out if we can't set
9162 * up signal handlers, because zdb is very useful without them.
9163 */
9164 action.sa_handler = sig_handler;
9165 sigemptyset(&action.sa_mask);
9166 action.sa_flags = 0;
9167 if (sigaction(SIGSEGV, &action, NULL) < 0) {
9168 (void) fprintf(stderr, "zdb: cannot catch SIGSEGV: %s\n",
9169 strerror(errno));
9170 }
9171 if (sigaction(SIGABRT, &action, NULL) < 0) {
9172 (void) fprintf(stderr, "zdb: cannot catch SIGABRT: %s\n",
9173 strerror(errno));
9174 }
9175
9176 /*
9177 * If there is an environment variable SPA_CONFIG_PATH it overrides
9178 * default spa_config_path setting. If -U flag is specified it will
9179 * override this environment variable settings once again.
9180 */
9181 spa_config_path_env = getenv("SPA_CONFIG_PATH");
9182 if (spa_config_path_env != NULL)
9183 spa_config_path = spa_config_path_env;
9184
9185 /*
9186 * For performance reasons, we set this tunable down. We do so before
9187 * the arg parsing section so that the user can override this value if
9188 * they choose.
9189 */
9190 zfs_btree_verify_intensity = 3;
9191
9192 struct option long_options[] = {
9193 {"ignore-assertions", no_argument, NULL, 'A'},
9194 {"block-stats", no_argument, NULL, 'b'},
9195 {"backup", no_argument, NULL, 'B'},
9196 {"checksum", no_argument, NULL, 'c'},
9197 {"config", no_argument, NULL, 'C'},
9198 {"datasets", no_argument, NULL, 'd'},
9199 {"dedup-stats", no_argument, NULL, 'D'},
9200 {"exported", no_argument, NULL, 'e'},
9201 {"embedded-block-pointer", no_argument, NULL, 'E'},
9202 {"automatic-rewind", no_argument, NULL, 'F'},
9203 {"dump-debug-msg", no_argument, NULL, 'G'},
9204 {"history", no_argument, NULL, 'h'},
9205 {"intent-logs", no_argument, NULL, 'i'},
9206 {"inflight", required_argument, NULL, 'I'},
9207 {"checkpointed-state", no_argument, NULL, 'k'},
9208 {"key", required_argument, NULL, 'K'},
9209 {"label", no_argument, NULL, 'l'},
9210 {"disable-leak-tracking", no_argument, NULL, 'L'},
9211 {"metaslabs", no_argument, NULL, 'm'},
9212 {"metaslab-groups", no_argument, NULL, 'M'},
9213 {"numeric", no_argument, NULL, 'N'},
9214 {"option", required_argument, NULL, 'o'},
9215 {"object-lookups", no_argument, NULL, 'O'},
9216 {"path", required_argument, NULL, 'p'},
9217 {"parseable", no_argument, NULL, 'P'},
9218 {"skip-label", no_argument, NULL, 'q'},
9219 {"copy-object", no_argument, NULL, 'r'},
9220 {"read-block", no_argument, NULL, 'R'},
9221 {"io-stats", no_argument, NULL, 's'},
9222 {"simulate-dedup", no_argument, NULL, 'S'},
9223 {"txg", required_argument, NULL, 't'},
9224 {"brt-stats", no_argument, NULL, 'T'},
9225 {"uberblock", no_argument, NULL, 'u'},
9226 {"cachefile", required_argument, NULL, 'U'},
9227 {"verbose", no_argument, NULL, 'v'},
9228 {"verbatim", no_argument, NULL, 'V'},
9229 {"dump-blocks", required_argument, NULL, 'x'},
9230 {"extreme-rewind", no_argument, NULL, 'X'},
9231 {"all-reconstruction", no_argument, NULL, 'Y'},
9232 {"livelist", no_argument, NULL, 'y'},
9233 {"zstd-headers", no_argument, NULL, 'Z'},
9234 {0, 0, 0, 0}
9235 };
9236
9237 while ((c = getopt_long(argc, argv,
9238 "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
9239 long_options, NULL)) != -1) {
9240 switch (c) {
9241 case 'b':
9242 case 'B':
9243 case 'c':
9244 case 'C':
9245 case 'd':
9246 case 'D':
9247 case 'E':
9248 case 'G':
9249 case 'h':
9250 case 'i':
9251 case 'l':
9252 case 'm':
9253 case 'M':
9254 case 'N':
9255 case 'O':
9256 case 'r':
9257 case 'R':
9258 case 's':
9259 case 'S':
9260 case 'T':
9261 case 'u':
9262 case 'y':
9263 case 'Z':
9264 dump_opt[c]++;
9265 dump_all = 0;
9266 break;
9267 case 'A':
9268 case 'e':
9269 case 'F':
9270 case 'k':
9271 case 'L':
9272 case 'P':
9273 case 'q':
9274 case 'X':
9275 dump_opt[c]++;
9276 break;
9277 case 'Y':
9278 zfs_reconstruct_indirect_combinations_max = INT_MAX;
9279 zfs_deadman_enabled = 0;
9280 break;
9281 /* NB: Sort single match options below. */
9282 case 'I':
9283 max_inflight_bytes = strtoull(optarg, NULL, 0);
9284 if (max_inflight_bytes == 0) {
9285 (void) fprintf(stderr, "maximum number "
9286 "of inflight bytes must be greater "
9287 "than 0\n");
9288 usage();
9289 }
9290 break;
9291 case 'K':
9292 dump_opt[c]++;
9293 key_material = strdup(optarg);
9294 /* redact key material in process table */
9295 while (*optarg != '\0') { *optarg++ = '*'; }
9296 break;
9297 case 'o':
9298 error = set_global_var(optarg);
9299 if (error != 0)
9300 usage();
9301 break;
9302 case 'p':
9303 if (searchdirs == NULL) {
9304 searchdirs = umem_alloc(sizeof (char *),
9305 UMEM_NOFAIL);
9306 } else {
9307 char **tmp = umem_alloc((nsearch + 1) *
9308 sizeof (char *), UMEM_NOFAIL);
9309 memcpy(tmp, searchdirs, nsearch *
9310 sizeof (char *));
9311 umem_free(searchdirs,
9312 nsearch * sizeof (char *));
9313 searchdirs = tmp;
9314 }
9315 searchdirs[nsearch++] = optarg;
9316 break;
9317 case 't':
9318 max_txg = strtoull(optarg, NULL, 0);
9319 if (max_txg < TXG_INITIAL) {
9320 (void) fprintf(stderr, "incorrect txg "
9321 "specified: %s\n", optarg);
9322 usage();
9323 }
9324 break;
9325 case 'U':
9326 config_path_console = B_TRUE;
9327 spa_config_path = optarg;
9328 if (spa_config_path[0] != '/') {
9329 (void) fprintf(stderr,
9330 "cachefile must be an absolute path "
9331 "(i.e. start with a slash)\n");
9332 usage();
9333 }
9334 break;
9335 case 'v':
9336 verbose++;
9337 break;
9338 case 'V':
9339 flags = ZFS_IMPORT_VERBATIM;
9340 break;
9341 case 'x':
9342 vn_dumpdir = optarg;
9343 break;
9344 default:
9345 usage();
9346 break;
9347 }
9348 }
9349
9350 if (!dump_opt['e'] && searchdirs != NULL) {
9351 (void) fprintf(stderr, "-p option requires use of -e\n");
9352 usage();
9353 }
9354 #if defined(_LP64)
9355 /*
9356 * ZDB does not typically re-read blocks; therefore limit the ARC
9357 * to 256 MB, which can be used entirely for metadata.
9358 */
9359 zfs_arc_min = 2ULL << SPA_MAXBLOCKSHIFT;
9360 zfs_arc_max = 256 * 1024 * 1024;
9361 #endif
9362
9363 /*
9364 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
9365 * "zdb -b" uses traversal prefetch which uses async reads.
9366 * For good performance, let several of them be active at once.
9367 */
9368 zfs_vdev_async_read_max_active = 10;
9369
9370 /*
9371 * Disable reference tracking for better performance.
9372 */
9373 reference_tracking_enable = B_FALSE;
9374
9375 /*
9376 * Do not fail spa_load when spa_load_verify fails. This is needed
9377 * to load non-idle pools.
9378 */
9379 spa_load_verify_dryrun = B_TRUE;
9380
9381 /*
9382 * ZDB should have ability to read spacemaps.
9383 */
9384 spa_mode_readable_spacemaps = B_TRUE;
9385
9386 if (dump_all)
9387 verbose = MAX(verbose, 1);
9388
9389 for (c = 0; c < 256; c++) {
9390 if (dump_all && strchr("ABeEFkKlLNOPrRSXy", c) == NULL)
9391 dump_opt[c] = 1;
9392 if (dump_opt[c])
9393 dump_opt[c] += verbose;
9394 }
9395
9396 libspl_set_assert_ok((dump_opt['A'] == 1) || (dump_opt['A'] > 2));
9397 zfs_recover = (dump_opt['A'] > 1);
9398
9399 argc -= optind;
9400 argv += optind;
9401 if (argc < 2 && dump_opt['R'])
9402 usage();
9403
9404 target = argv[0];
9405
9406 /*
9407 * Automate cachefile
9408 */
9409 if (!spa_config_path_env && !config_path_console && target &&
9410 libzfs_core_init() == 0) {
9411 char *pname = strdup(target);
9412 const char *value;
9413 nvlist_t *pnvl = NULL;
9414 nvlist_t *vnvl = NULL;
9415
9416 if (strpbrk(pname, "/@") != NULL)
9417 *strpbrk(pname, "/@") = '\0';
9418
9419 if (pname && lzc_get_props(pname, &pnvl) == 0) {
9420 if (nvlist_lookup_nvlist(pnvl, "cachefile",
9421 &vnvl) == 0) {
9422 value = fnvlist_lookup_string(vnvl,
9423 ZPROP_VALUE);
9424 } else {
9425 value = "-";
9426 }
9427 strlcpy(pbuf, value, sizeof (pbuf));
9428 if (pbuf[0] != '\0') {
9429 if (pbuf[0] == '/') {
9430 if (access(pbuf, F_OK) == 0)
9431 spa_config_path = pbuf;
9432 else
9433 force_import = B_TRUE;
9434 } else if ((strcmp(pbuf, "-") == 0 &&
9435 access(ZPOOL_CACHE, F_OK) != 0) ||
9436 strcmp(pbuf, "none") == 0) {
9437 force_import = B_TRUE;
9438 }
9439 }
9440 nvlist_free(vnvl);
9441 }
9442
9443 free(pname);
9444 nvlist_free(pnvl);
9445 libzfs_core_fini();
9446 }
9447
9448 dmu_objset_register_type(DMU_OST_ZFS, dummy_get_file_info);
9449 kernel_init(SPA_MODE_READ);
9450 kernel_init_done = B_TRUE;
9451
9452 if (dump_opt['E']) {
9453 if (argc != 1)
9454 usage();
9455 zdb_embedded_block(argv[0]);
9456 error = 0;
9457 goto fini;
9458 }
9459
9460 if (argc < 1) {
9461 if (!dump_opt['e'] && dump_opt['C']) {
9462 dump_cachefile(spa_config_path);
9463 error = 0;
9464 goto fini;
9465 }
9466 usage();
9467 }
9468
9469 if (dump_opt['l']) {
9470 error = dump_label(argv[0]);
9471 goto fini;
9472 }
9473
9474 if (dump_opt['X'] || dump_opt['F'])
9475 rewind = ZPOOL_DO_REWIND |
9476 (dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
9477
9478 /* -N implies -d */
9479 if (dump_opt['N'] && dump_opt['d'] == 0)
9480 dump_opt['d'] = dump_opt['N'];
9481
9482 if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 ||
9483 nvlist_add_uint64(policy, ZPOOL_LOAD_REQUEST_TXG, max_txg) != 0 ||
9484 nvlist_add_uint32(policy, ZPOOL_LOAD_REWIND_POLICY, rewind) != 0)
9485 fatal("internal error: %s", strerror(ENOMEM));
9486
9487 error = 0;
9488
9489 if (strpbrk(target, "/@") != NULL) {
9490 size_t targetlen;
9491
9492 target_pool = strdup(target);
9493 *strpbrk(target_pool, "/@") = '\0';
9494
9495 target_is_spa = B_FALSE;
9496 targetlen = strlen(target);
9497 if (targetlen && target[targetlen - 1] == '/')
9498 target[targetlen - 1] = '\0';
9499
9500 /*
9501 * See if an objset ID was supplied (-d <pool>/<objset ID>).
9502 * To disambiguate tank/100, consider the 100 as objsetID
9503 * if -N was given, otherwise 100 is an objsetID iff
9504 * tank/100 as a named dataset fails on lookup.
9505 */
9506 objset_str = strchr(target, '/');
9507 if (objset_str && strlen(objset_str) > 1 &&
9508 zdb_numeric(objset_str + 1)) {
9509 char *endptr;
9510 errno = 0;
9511 objset_str++;
9512 objset_id = strtoull(objset_str, &endptr, 0);
9513 /* dataset 0 is the same as opening the pool */
9514 if (errno == 0 && endptr != objset_str &&
9515 objset_id != 0) {
9516 if (dump_opt['N'])
9517 dataset_lookup = B_TRUE;
9518 }
9519 /* normal dataset name not an objset ID */
9520 if (endptr == objset_str) {
9521 objset_id = -1;
9522 }
9523 } else if (objset_str && !zdb_numeric(objset_str + 1) &&
9524 dump_opt['N']) {
9525 printf("Supply a numeric objset ID with -N\n");
9526 error = 1;
9527 goto fini;
9528 }
9529 } else {
9530 target_pool = target;
9531 }
9532
9533 if (dump_opt['e'] || force_import) {
9534 importargs_t args = { 0 };
9535
9536 /*
9537 * If path is not provided, search in /dev
9538 */
9539 if (searchdirs == NULL) {
9540 searchdirs = umem_alloc(sizeof (char *), UMEM_NOFAIL);
9541 searchdirs[nsearch++] = (char *)ZFS_DEVDIR;
9542 }
9543
9544 args.paths = nsearch;
9545 args.path = searchdirs;
9546 args.can_be_active = B_TRUE;
9547
9548 libpc_handle_t lpch = {
9549 .lpc_lib_handle = NULL,
9550 .lpc_ops = &libzpool_config_ops,
9551 .lpc_printerr = B_TRUE
9552 };
9553 error = zpool_find_config(&lpch, target_pool, &cfg, &args);
9554
9555 if (error == 0) {
9556
9557 if (nvlist_add_nvlist(cfg,
9558 ZPOOL_LOAD_POLICY, policy) != 0) {
9559 fatal("can't open '%s': %s",
9560 target, strerror(ENOMEM));
9561 }
9562
9563 if (dump_opt['C'] > 1) {
9564 (void) printf("\nConfiguration for import:\n");
9565 dump_nvlist(cfg, 8);
9566 }
9567
9568 /*
9569 * Disable the activity check to allow examination of
9570 * active pools.
9571 */
9572 error = spa_import(target_pool, cfg, NULL,
9573 flags | ZFS_IMPORT_SKIP_MMP);
9574 }
9575 }
9576
9577 if (searchdirs != NULL) {
9578 umem_free(searchdirs, nsearch * sizeof (char *));
9579 searchdirs = NULL;
9580 }
9581
9582 /*
9583 * We need to make sure to process -O option or call
9584 * dump_path after the -e option has been processed,
9585 * which imports the pool to the namespace if it's
9586 * not in the cachefile.
9587 */
9588 if (dump_opt['O']) {
9589 if (argc != 2)
9590 usage();
9591 dump_opt['v'] = verbose + 3;
9592 error = dump_path(argv[0], argv[1], NULL);
9593 goto fini;
9594 }
9595
9596 if (dump_opt['r']) {
9597 target_is_spa = B_FALSE;
9598 if (argc != 3)
9599 usage();
9600 dump_opt['v'] = verbose;
9601 error = dump_path(argv[0], argv[1], &object);
9602 if (error != 0)
9603 fatal("internal error: %s", strerror(error));
9604 }
9605
9606 /*
9607 * import_checkpointed_state makes the assumption that the
9608 * target pool that we pass it is already part of the spa
9609 * namespace. Because of that we need to make sure to call
9610 * it always after the -e option has been processed, which
9611 * imports the pool to the namespace if it's not in the
9612 * cachefile.
9613 */
9614 char *checkpoint_pool = NULL;
9615 char *checkpoint_target = NULL;
9616 if (dump_opt['k']) {
9617 checkpoint_pool = import_checkpointed_state(target, cfg,
9618 &checkpoint_target);
9619
9620 if (checkpoint_target != NULL)
9621 target = checkpoint_target;
9622 }
9623
9624 if (cfg != NULL) {
9625 nvlist_free(cfg);
9626 cfg = NULL;
9627 }
9628
9629 if (target_pool != target)
9630 free(target_pool);
9631
9632 if (error == 0) {
9633 if (dump_opt['k'] && (target_is_spa || dump_opt['R'])) {
9634 ASSERT(checkpoint_pool != NULL);
9635 ASSERT(checkpoint_target == NULL);
9636
9637 error = spa_open(checkpoint_pool, &spa, FTAG);
9638 if (error != 0) {
9639 fatal("Tried to open pool \"%s\" but "
9640 "spa_open() failed with error %d\n",
9641 checkpoint_pool, error);
9642 }
9643
9644 } else if (target_is_spa || dump_opt['R'] || dump_opt['B'] ||
9645 objset_id == 0) {
9646 zdb_set_skip_mmp(target);
9647 error = spa_open_rewind(target, &spa, FTAG, policy,
9648 NULL);
9649 if (error) {
9650 /*
9651 * If we're missing the log device then
9652 * try opening the pool after clearing the
9653 * log state.
9654 */
9655 mutex_enter(&spa_namespace_lock);
9656 if ((spa = spa_lookup(target)) != NULL &&
9657 spa->spa_log_state == SPA_LOG_MISSING) {
9658 spa->spa_log_state = SPA_LOG_CLEAR;
9659 error = 0;
9660 }
9661 mutex_exit(&spa_namespace_lock);
9662
9663 if (!error) {
9664 error = spa_open_rewind(target, &spa,
9665 FTAG, policy, NULL);
9666 }
9667 }
9668 } else if (strpbrk(target, "#") != NULL) {
9669 dsl_pool_t *dp;
9670 error = dsl_pool_hold(target, FTAG, &dp);
9671 if (error != 0) {
9672 fatal("can't dump '%s': %s", target,
9673 strerror(error));
9674 }
9675 error = dump_bookmark(dp, target, B_TRUE, verbose > 1);
9676 dsl_pool_rele(dp, FTAG);
9677 if (error != 0) {
9678 fatal("can't dump '%s': %s", target,
9679 strerror(error));
9680 }
9681 goto fini;
9682 } else {
9683 target_pool = strdup(target);
9684 if (strpbrk(target, "/@") != NULL)
9685 *strpbrk(target_pool, "/@") = '\0';
9686
9687 zdb_set_skip_mmp(target);
9688 /*
9689 * If -N was supplied, the user has indicated that
9690 * zdb -d <pool>/<objsetID> is in effect. Otherwise
9691 * we first assume that the dataset string is the
9692 * dataset name. If dmu_objset_hold fails with the
9693 * dataset string, and we have an objset_id, retry the
9694 * lookup with the objsetID.
9695 */
9696 boolean_t retry = B_TRUE;
9697 retry_lookup:
9698 if (dataset_lookup == B_TRUE) {
9699 /*
9700 * Use the supplied id to get the name
9701 * for open_objset.
9702 */
9703 error = spa_open(target_pool, &spa, FTAG);
9704 if (error == 0) {
9705 error = name_from_objset_id(spa,
9706 objset_id, dsname);
9707 spa_close(spa, FTAG);
9708 if (error == 0)
9709 target = dsname;
9710 }
9711 }
9712 if (error == 0) {
9713 if (objset_id > 0 && retry) {
9714 int err = dmu_objset_hold(target, FTAG,
9715 &os);
9716 if (err) {
9717 dataset_lookup = B_TRUE;
9718 retry = B_FALSE;
9719 goto retry_lookup;
9720 } else {
9721 dmu_objset_rele(os, FTAG);
9722 }
9723 }
9724 error = open_objset(target, FTAG, &os);
9725 }
9726 if (error == 0)
9727 spa = dmu_objset_spa(os);
9728 free(target_pool);
9729 }
9730 }
9731 nvlist_free(policy);
9732
9733 if (error)
9734 fatal("can't open '%s': %s", target, strerror(error));
9735
9736 /*
9737 * Set the pool failure mode to panic in order to prevent the pool
9738 * from suspending. A suspended I/O will have no way to resume and
9739 * can prevent the zdb(8) command from terminating as expected.
9740 */
9741 if (spa != NULL)
9742 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
9743
9744 argv++;
9745 argc--;
9746 if (dump_opt['r']) {
9747 error = zdb_copy_object(os, object, argv[1]);
9748 } else if (!dump_opt['R']) {
9749 flagbits['d'] = ZOR_FLAG_DIRECTORY;
9750 flagbits['f'] = ZOR_FLAG_PLAIN_FILE;
9751 flagbits['m'] = ZOR_FLAG_SPACE_MAP;
9752 flagbits['z'] = ZOR_FLAG_ZAP;
9753 flagbits['A'] = ZOR_FLAG_ALL_TYPES;
9754
9755 if (argc > 0 && dump_opt['d']) {
9756 zopt_object_args = argc;
9757 zopt_object_ranges = calloc(zopt_object_args,
9758 sizeof (zopt_object_range_t));
9759 for (unsigned i = 0; i < zopt_object_args; i++) {
9760 int err;
9761 const char *msg = NULL;
9762
9763 err = parse_object_range(argv[i],
9764 &zopt_object_ranges[i], &msg);
9765 if (err != 0)
9766 fatal("Bad object or range: '%s': %s\n",
9767 argv[i], msg ?: "");
9768 }
9769 } else if (argc > 0 && dump_opt['m']) {
9770 zopt_metaslab_args = argc;
9771 zopt_metaslab = calloc(zopt_metaslab_args,
9772 sizeof (uint64_t));
9773 for (unsigned i = 0; i < zopt_metaslab_args; i++) {
9774 errno = 0;
9775 zopt_metaslab[i] = strtoull(argv[i], NULL, 0);
9776 if (zopt_metaslab[i] == 0 && errno != 0)
9777 fatal("bad number %s: %s", argv[i],
9778 strerror(errno));
9779 }
9780 }
9781 if (dump_opt['B']) {
9782 dump_backup(target, objset_id,
9783 argc > 0 ? argv[0] : NULL);
9784 } else if (os != NULL) {
9785 dump_objset(os);
9786 } else if (zopt_object_args > 0 && !dump_opt['m']) {
9787 dump_objset(spa->spa_meta_objset);
9788 } else {
9789 dump_zpool(spa);
9790 }
9791 } else {
9792 flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR;
9793 flagbits['c'] = ZDB_FLAG_CHECKSUM;
9794 flagbits['d'] = ZDB_FLAG_DECOMPRESS;
9795 flagbits['e'] = ZDB_FLAG_BSWAP;
9796 flagbits['g'] = ZDB_FLAG_GBH;
9797 flagbits['i'] = ZDB_FLAG_INDIRECT;
9798 flagbits['r'] = ZDB_FLAG_RAW;
9799 flagbits['v'] = ZDB_FLAG_VERBOSE;
9800
9801 for (int i = 0; i < argc; i++)
9802 zdb_read_block(argv[i], spa);
9803 }
9804
9805 if (dump_opt['k']) {
9806 free(checkpoint_pool);
9807 if (!target_is_spa)
9808 free(checkpoint_target);
9809 }
9810
9811 fini:
9812 if (spa != NULL)
9813 zdb_ddt_cleanup(spa);
9814
9815 if (os != NULL) {
9816 close_objset(os, FTAG);
9817 } else if (spa != NULL) {
9818 spa_close(spa, FTAG);
9819 }
9820
9821 fuid_table_destroy();
9822
9823 dump_debug_buffer();
9824
9825 if (kernel_init_done)
9826 kernel_fini();
9827
9828 return (error);
9829 }
OpenZFS on Linux and FreeBSD