1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or https://opensource.org/licenses/CDDL-1.0.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22
23 /*
24 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Nexenta Systems, Inc.
28 * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC.
29 * Copyright (c) 2015, 2017, Intel Corporation.
30 * Copyright (c) 2020 Datto Inc.
31 * Copyright (c) 2020, The FreeBSD Foundation [1]
32 *
33 * [1] Portions of this software were developed by Allan Jude
34 * under sponsorship from the FreeBSD Foundation.
35 * Copyright (c) 2021 Allan Jude
36 * Copyright (c) 2021 Toomas Soome <tsoome@me.com>
37 * Copyright (c) 2023, 2024, Klara Inc.
38 * Copyright (c) 2023, Rob Norris <robn@despairlabs.com>
39 */
40
41 #include <stdio.h>
42 #include <unistd.h>
43 #include <stdlib.h>
44 #include <ctype.h>
45 #include <getopt.h>
46 #include <openssl/evp.h>
47 #include <sys/zfs_context.h>
48 #include <sys/spa.h>
49 #include <sys/spa_impl.h>
50 #include <sys/dmu.h>
51 #include <sys/zap.h>
52 #include <sys/zap_impl.h>
53 #include <sys/fs/zfs.h>
54 #include <sys/zfs_znode.h>
55 #include <sys/zfs_sa.h>
56 #include <sys/sa.h>
57 #include <sys/sa_impl.h>
58 #include <sys/vdev.h>
59 #include <sys/vdev_impl.h>
60 #include <sys/metaslab_impl.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/dsl_dir.h>
63 #include <sys/dsl_dataset.h>
64 #include <sys/dsl_pool.h>
65 #include <sys/dsl_bookmark.h>
66 #include <sys/dbuf.h>
67 #include <sys/zil.h>
68 #include <sys/zil_impl.h>
69 #include <sys/stat.h>
70 #include <sys/resource.h>
71 #include <sys/dmu_send.h>
72 #include <sys/dmu_traverse.h>
73 #include <sys/zio_checksum.h>
74 #include <sys/zio_compress.h>
75 #include <sys/zfs_fuid.h>
76 #include <sys/arc.h>
77 #include <sys/arc_impl.h>
78 #include <sys/ddt.h>
79 #include <sys/ddt_impl.h>
80 #include <sys/zfeature.h>
81 #include <sys/abd.h>
82 #include <sys/blkptr.h>
83 #include <sys/dsl_crypt.h>
84 #include <sys/dsl_scan.h>
85 #include <sys/btree.h>
86 #include <sys/brt.h>
87 #include <sys/brt_impl.h>
88 #include <zfs_comutil.h>
89 #include <sys/zstd/zstd.h>
90 #include <sys/backtrace.h>
91
92 #include <libnvpair.h>
93 #include <libzutil.h>
94 #include <libzfs_core.h>
95
96 #include <libzdb.h>
97
98 #include "zdb.h"
99
100
101 extern int reference_tracking_enable;
102 extern int zfs_recover;
103 extern uint_t zfs_vdev_async_read_max_active;
104 extern boolean_t spa_load_verify_dryrun;
105 extern boolean_t spa_mode_readable_spacemaps;
106 extern uint_t zfs_reconstruct_indirect_combinations_max;
107 extern uint_t zfs_btree_verify_intensity;
108
109 static const char cmdname[] = "zdb";
110 uint8_t dump_opt[256];
111
112 typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size);
113
114 static uint64_t *zopt_metaslab = NULL;
115 static unsigned zopt_metaslab_args = 0;
116
117
118 static zopt_object_range_t *zopt_object_ranges = NULL;
119 static unsigned zopt_object_args = 0;
120
121 static int flagbits[256];
122
123
124 static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */
125 static int leaked_objects = 0;
126 static zfs_range_tree_t *mos_refd_objs;
127 static spa_t *spa;
128 static objset_t *os;
129 static boolean_t kernel_init_done;
130
131 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *,
132 boolean_t);
133 static void mos_obj_refd(uint64_t);
134 static void mos_obj_refd_multiple(uint64_t);
135 static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free,
136 dmu_tx_t *tx);
137
138
139
140 static void zdb_print_blkptr(const blkptr_t *bp, int flags);
141 static void zdb_exit(int reason);
142
143 typedef struct sublivelist_verify_block_refcnt {
144 /* block pointer entry in livelist being verified */
145 blkptr_t svbr_blk;
146
147 /*
148 * Refcount gets incremented to 1 when we encounter the first
149 * FREE entry for the svfbr block pointer and a node for it
150 * is created in our ZDB verification/tracking metadata.
151 *
152 * As we encounter more FREE entries we increment this counter
153 * and similarly decrement it whenever we find the respective
154 * ALLOC entries for this block.
155 *
156 * When the refcount gets to 0 it means that all the FREE and
157 * ALLOC entries of this block have paired up and we no longer
158 * need to track it in our verification logic (e.g. the node
159 * containing this struct in our verification data structure
160 * should be freed).
161 *
162 * [refer to sublivelist_verify_blkptr() for the actual code]
163 */
164 uint32_t svbr_refcnt;
165 } sublivelist_verify_block_refcnt_t;
166
167 static int
sublivelist_block_refcnt_compare(const void * larg,const void * rarg)168 sublivelist_block_refcnt_compare(const void *larg, const void *rarg)
169 {
170 const sublivelist_verify_block_refcnt_t *l = larg;
171 const sublivelist_verify_block_refcnt_t *r = rarg;
172 return (livelist_compare(&l->svbr_blk, &r->svbr_blk));
173 }
174
175 static int
sublivelist_verify_blkptr(void * arg,const blkptr_t * bp,boolean_t free,dmu_tx_t * tx)176 sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free,
177 dmu_tx_t *tx)
178 {
179 ASSERT3P(tx, ==, NULL);
180 struct sublivelist_verify *sv = arg;
181 sublivelist_verify_block_refcnt_t current = {
182 .svbr_blk = *bp,
183
184 /*
185 * Start with 1 in case this is the first free entry.
186 * This field is not used for our B-Tree comparisons
187 * anyway.
188 */
189 .svbr_refcnt = 1,
190 };
191
192 zfs_btree_index_t where;
193 sublivelist_verify_block_refcnt_t *pair =
194 zfs_btree_find(&sv->sv_pair, ¤t, &where);
195 if (free) {
196 if (pair == NULL) {
197 /* first free entry for this block pointer */
198 zfs_btree_add(&sv->sv_pair, ¤t);
199 } else {
200 pair->svbr_refcnt++;
201 }
202 } else {
203 if (pair == NULL) {
204 /* block that is currently marked as allocated */
205 for (int i = 0; i < SPA_DVAS_PER_BP; i++) {
206 if (DVA_IS_EMPTY(&bp->blk_dva[i]))
207 break;
208 sublivelist_verify_block_t svb = {
209 .svb_dva = bp->blk_dva[i],
210 .svb_allocated_txg =
211 BP_GET_BIRTH(bp)
212 };
213
214 if (zfs_btree_find(&sv->sv_leftover, &svb,
215 &where) == NULL) {
216 zfs_btree_add_idx(&sv->sv_leftover,
217 &svb, &where);
218 }
219 }
220 } else {
221 /* alloc matches a free entry */
222 pair->svbr_refcnt--;
223 if (pair->svbr_refcnt == 0) {
224 /* all allocs and frees have been matched */
225 zfs_btree_remove_idx(&sv->sv_pair, &where);
226 }
227 }
228 }
229
230 return (0);
231 }
232
233 static int
sublivelist_verify_func(void * args,dsl_deadlist_entry_t * dle)234 sublivelist_verify_func(void *args, dsl_deadlist_entry_t *dle)
235 {
236 int err;
237 struct sublivelist_verify *sv = args;
238
239 zfs_btree_create(&sv->sv_pair, sublivelist_block_refcnt_compare, NULL,
240 sizeof (sublivelist_verify_block_refcnt_t));
241
242 err = bpobj_iterate_nofree(&dle->dle_bpobj, sublivelist_verify_blkptr,
243 sv, NULL);
244
245 sublivelist_verify_block_refcnt_t *e;
246 zfs_btree_index_t *cookie = NULL;
247 while ((e = zfs_btree_destroy_nodes(&sv->sv_pair, &cookie)) != NULL) {
248 char blkbuf[BP_SPRINTF_LEN];
249 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
250 &e->svbr_blk, B_TRUE);
251 (void) printf("\tERROR: %d unmatched FREE(s): %s\n",
252 e->svbr_refcnt, blkbuf);
253 }
254 zfs_btree_destroy(&sv->sv_pair);
255
256 return (err);
257 }
258
259 static int
livelist_block_compare(const void * larg,const void * rarg)260 livelist_block_compare(const void *larg, const void *rarg)
261 {
262 const sublivelist_verify_block_t *l = larg;
263 const sublivelist_verify_block_t *r = rarg;
264
265 if (DVA_GET_VDEV(&l->svb_dva) < DVA_GET_VDEV(&r->svb_dva))
266 return (-1);
267 else if (DVA_GET_VDEV(&l->svb_dva) > DVA_GET_VDEV(&r->svb_dva))
268 return (+1);
269
270 if (DVA_GET_OFFSET(&l->svb_dva) < DVA_GET_OFFSET(&r->svb_dva))
271 return (-1);
272 else if (DVA_GET_OFFSET(&l->svb_dva) > DVA_GET_OFFSET(&r->svb_dva))
273 return (+1);
274
275 if (DVA_GET_ASIZE(&l->svb_dva) < DVA_GET_ASIZE(&r->svb_dva))
276 return (-1);
277 else if (DVA_GET_ASIZE(&l->svb_dva) > DVA_GET_ASIZE(&r->svb_dva))
278 return (+1);
279
280 return (0);
281 }
282
283 /*
284 * Check for errors in a livelist while tracking all unfreed ALLOCs in the
285 * sublivelist_verify_t: sv->sv_leftover
286 */
287 static void
livelist_verify(dsl_deadlist_t * dl,void * arg)288 livelist_verify(dsl_deadlist_t *dl, void *arg)
289 {
290 sublivelist_verify_t *sv = arg;
291 dsl_deadlist_iterate(dl, sublivelist_verify_func, sv);
292 }
293
294 /*
295 * Check for errors in the livelist entry and discard the intermediary
296 * data structures
297 */
298 static int
sublivelist_verify_lightweight(void * args,dsl_deadlist_entry_t * dle)299 sublivelist_verify_lightweight(void *args, dsl_deadlist_entry_t *dle)
300 {
301 (void) args;
302 sublivelist_verify_t sv;
303 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
304 sizeof (sublivelist_verify_block_t));
305 int err = sublivelist_verify_func(&sv, dle);
306 zfs_btree_clear(&sv.sv_leftover);
307 zfs_btree_destroy(&sv.sv_leftover);
308 return (err);
309 }
310
311 typedef struct metaslab_verify {
312 /*
313 * Tree containing all the leftover ALLOCs from the livelists
314 * that are part of this metaslab.
315 */
316 zfs_btree_t mv_livelist_allocs;
317
318 /*
319 * Metaslab information.
320 */
321 uint64_t mv_vdid;
322 uint64_t mv_msid;
323 uint64_t mv_start;
324 uint64_t mv_end;
325
326 /*
327 * What's currently allocated for this metaslab.
328 */
329 zfs_range_tree_t *mv_allocated;
330 } metaslab_verify_t;
331
332 typedef void ll_iter_t(dsl_deadlist_t *ll, void *arg);
333
334 typedef int (*zdb_log_sm_cb_t)(spa_t *spa, space_map_entry_t *sme, uint64_t txg,
335 void *arg);
336
337 typedef struct unflushed_iter_cb_arg {
338 spa_t *uic_spa;
339 uint64_t uic_txg;
340 void *uic_arg;
341 zdb_log_sm_cb_t uic_cb;
342 } unflushed_iter_cb_arg_t;
343
344 static int
iterate_through_spacemap_logs_cb(space_map_entry_t * sme,void * arg)345 iterate_through_spacemap_logs_cb(space_map_entry_t *sme, void *arg)
346 {
347 unflushed_iter_cb_arg_t *uic = arg;
348 return (uic->uic_cb(uic->uic_spa, sme, uic->uic_txg, uic->uic_arg));
349 }
350
351 static void
iterate_through_spacemap_logs(spa_t * spa,zdb_log_sm_cb_t cb,void * arg)352 iterate_through_spacemap_logs(spa_t *spa, zdb_log_sm_cb_t cb, void *arg)
353 {
354 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
355 return;
356
357 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
358 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
359 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
360 space_map_t *sm = NULL;
361 VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
362 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
363
364 unflushed_iter_cb_arg_t uic = {
365 .uic_spa = spa,
366 .uic_txg = sls->sls_txg,
367 .uic_arg = arg,
368 .uic_cb = cb
369 };
370 VERIFY0(space_map_iterate(sm, space_map_length(sm),
371 iterate_through_spacemap_logs_cb, &uic));
372 space_map_close(sm);
373 }
374 spa_config_exit(spa, SCL_CONFIG, FTAG);
375 }
376
377 static void
verify_livelist_allocs(metaslab_verify_t * mv,uint64_t txg,uint64_t offset,uint64_t size)378 verify_livelist_allocs(metaslab_verify_t *mv, uint64_t txg,
379 uint64_t offset, uint64_t size)
380 {
381 sublivelist_verify_block_t svb = {{{0}}};
382 DVA_SET_VDEV(&svb.svb_dva, mv->mv_vdid);
383 DVA_SET_OFFSET(&svb.svb_dva, offset);
384 DVA_SET_ASIZE(&svb.svb_dva, size);
385 zfs_btree_index_t where;
386 uint64_t end_offset = offset + size;
387
388 /*
389 * Look for an exact match for spacemap entry in the livelist entries.
390 * Then, look for other livelist entries that fall within the range
391 * of the spacemap entry as it may have been condensed
392 */
393 sublivelist_verify_block_t *found =
394 zfs_btree_find(&mv->mv_livelist_allocs, &svb, &where);
395 if (found == NULL) {
396 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where);
397 }
398 for (; found != NULL && DVA_GET_VDEV(&found->svb_dva) == mv->mv_vdid &&
399 DVA_GET_OFFSET(&found->svb_dva) < end_offset;
400 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
401 if (found->svb_allocated_txg <= txg) {
402 (void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
403 "from TXG %llx FREED at TXG %llx\n",
404 (u_longlong_t)DVA_GET_OFFSET(&found->svb_dva),
405 (u_longlong_t)DVA_GET_ASIZE(&found->svb_dva),
406 (u_longlong_t)found->svb_allocated_txg,
407 (u_longlong_t)txg);
408 }
409 }
410 }
411
412 static int
metaslab_spacemap_validation_cb(space_map_entry_t * sme,void * arg)413 metaslab_spacemap_validation_cb(space_map_entry_t *sme, void *arg)
414 {
415 metaslab_verify_t *mv = arg;
416 uint64_t offset = sme->sme_offset;
417 uint64_t size = sme->sme_run;
418 uint64_t txg = sme->sme_txg;
419
420 if (sme->sme_type == SM_ALLOC) {
421 if (zfs_range_tree_contains(mv->mv_allocated,
422 offset, size)) {
423 (void) printf("ERROR: DOUBLE ALLOC: "
424 "%llu [%llx:%llx] "
425 "%llu:%llu LOG_SM\n",
426 (u_longlong_t)txg, (u_longlong_t)offset,
427 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
428 (u_longlong_t)mv->mv_msid);
429 } else {
430 zfs_range_tree_add(mv->mv_allocated,
431 offset, size);
432 }
433 } else {
434 if (!zfs_range_tree_contains(mv->mv_allocated,
435 offset, size)) {
436 (void) printf("ERROR: DOUBLE FREE: "
437 "%llu [%llx:%llx] "
438 "%llu:%llu LOG_SM\n",
439 (u_longlong_t)txg, (u_longlong_t)offset,
440 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
441 (u_longlong_t)mv->mv_msid);
442 } else {
443 zfs_range_tree_remove(mv->mv_allocated,
444 offset, size);
445 }
446 }
447
448 if (sme->sme_type != SM_ALLOC) {
449 /*
450 * If something is freed in the spacemap, verify that
451 * it is not listed as allocated in the livelist.
452 */
453 verify_livelist_allocs(mv, txg, offset, size);
454 }
455 return (0);
456 }
457
458 static int
spacemap_check_sm_log_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)459 spacemap_check_sm_log_cb(spa_t *spa, space_map_entry_t *sme,
460 uint64_t txg, void *arg)
461 {
462 metaslab_verify_t *mv = arg;
463 uint64_t offset = sme->sme_offset;
464 uint64_t vdev_id = sme->sme_vdev;
465
466 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
467
468 /* skip indirect vdevs */
469 if (!vdev_is_concrete(vd))
470 return (0);
471
472 if (vdev_id != mv->mv_vdid)
473 return (0);
474
475 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
476 if (ms->ms_id != mv->mv_msid)
477 return (0);
478
479 if (txg < metaslab_unflushed_txg(ms))
480 return (0);
481
482
483 ASSERT3U(txg, ==, sme->sme_txg);
484 return (metaslab_spacemap_validation_cb(sme, mv));
485 }
486
487 static void
spacemap_check_sm_log(spa_t * spa,metaslab_verify_t * mv)488 spacemap_check_sm_log(spa_t *spa, metaslab_verify_t *mv)
489 {
490 iterate_through_spacemap_logs(spa, spacemap_check_sm_log_cb, mv);
491 }
492
493 static void
spacemap_check_ms_sm(space_map_t * sm,metaslab_verify_t * mv)494 spacemap_check_ms_sm(space_map_t *sm, metaslab_verify_t *mv)
495 {
496 if (sm == NULL)
497 return;
498
499 VERIFY0(space_map_iterate(sm, space_map_length(sm),
500 metaslab_spacemap_validation_cb, mv));
501 }
502
503 static void iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg);
504
505 /*
506 * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
507 * they are part of that metaslab (mv_msid).
508 */
509 static void
mv_populate_livelist_allocs(metaslab_verify_t * mv,sublivelist_verify_t * sv)510 mv_populate_livelist_allocs(metaslab_verify_t *mv, sublivelist_verify_t *sv)
511 {
512 zfs_btree_index_t where;
513 sublivelist_verify_block_t *svb;
514 ASSERT3U(zfs_btree_numnodes(&mv->mv_livelist_allocs), ==, 0);
515 for (svb = zfs_btree_first(&sv->sv_leftover, &where);
516 svb != NULL;
517 svb = zfs_btree_next(&sv->sv_leftover, &where, &where)) {
518 if (DVA_GET_VDEV(&svb->svb_dva) != mv->mv_vdid)
519 continue;
520
521 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start &&
522 (DVA_GET_OFFSET(&svb->svb_dva) +
523 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_start) {
524 (void) printf("ERROR: Found block that crosses "
525 "metaslab boundary: <%llu:%llx:%llx>\n",
526 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
527 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
528 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
529 continue;
530 }
531
532 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start)
533 continue;
534
535 if (DVA_GET_OFFSET(&svb->svb_dva) >= mv->mv_end)
536 continue;
537
538 if ((DVA_GET_OFFSET(&svb->svb_dva) +
539 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_end) {
540 (void) printf("ERROR: Found block that crosses "
541 "metaslab boundary: <%llu:%llx:%llx>\n",
542 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
543 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
544 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
545 continue;
546 }
547
548 zfs_btree_add(&mv->mv_livelist_allocs, svb);
549 }
550
551 for (svb = zfs_btree_first(&mv->mv_livelist_allocs, &where);
552 svb != NULL;
553 svb = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
554 zfs_btree_remove(&sv->sv_leftover, svb);
555 }
556 }
557
558 /*
559 * [Livelist Check]
560 * Iterate through all the sublivelists and:
561 * - report leftover frees (**)
562 * - record leftover ALLOCs together with their TXG [see Cross Check]
563 *
564 * (**) Note: Double ALLOCs are valid in datasets that have dedup
565 * enabled. Similarly double FREEs are allowed as well but
566 * only if they pair up with a corresponding ALLOC entry once
567 * we our done with our sublivelist iteration.
568 *
569 * [Spacemap Check]
570 * for each metaslab:
571 * - iterate over spacemap and then the metaslab's entries in the
572 * spacemap log, then report any double FREEs and ALLOCs (do not
573 * blow up).
574 *
575 * [Cross Check]
576 * After finishing the Livelist Check phase and while being in the
577 * Spacemap Check phase, we find all the recorded leftover ALLOCs
578 * of the livelist check that are part of the metaslab that we are
579 * currently looking at in the Spacemap Check. We report any entries
580 * that are marked as ALLOCs in the livelists but have been actually
581 * freed (and potentially allocated again) after their TXG stamp in
582 * the spacemaps. Also report any ALLOCs from the livelists that
583 * belong to indirect vdevs (e.g. their vdev completed removal).
584 *
585 * Note that this will miss Log Spacemap entries that cancelled each other
586 * out before being flushed to the metaslab, so we are not guaranteed
587 * to match all erroneous ALLOCs.
588 */
589 static void
livelist_metaslab_validate(spa_t * spa)590 livelist_metaslab_validate(spa_t *spa)
591 {
592 (void) printf("Verifying deleted livelist entries\n");
593
594 sublivelist_verify_t sv;
595 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
596 sizeof (sublivelist_verify_block_t));
597 iterate_deleted_livelists(spa, livelist_verify, &sv);
598
599 (void) printf("Verifying metaslab entries\n");
600 vdev_t *rvd = spa->spa_root_vdev;
601 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
602 vdev_t *vd = rvd->vdev_child[c];
603
604 if (!vdev_is_concrete(vd))
605 continue;
606
607 for (uint64_t mid = 0; mid < vd->vdev_ms_count; mid++) {
608 metaslab_t *m = vd->vdev_ms[mid];
609
610 (void) fprintf(stderr,
611 "\rverifying concrete vdev %llu, "
612 "metaslab %llu of %llu ...",
613 (longlong_t)vd->vdev_id,
614 (longlong_t)mid,
615 (longlong_t)vd->vdev_ms_count);
616
617 uint64_t shift, start;
618 zfs_range_seg_type_t type =
619 metaslab_calculate_range_tree_type(vd, m,
620 &start, &shift);
621 metaslab_verify_t mv;
622 mv.mv_allocated = zfs_range_tree_create_flags(
623 NULL, type, NULL, start, shift,
624 0, "livelist_metaslab_validate:mv_allocated");
625 mv.mv_vdid = vd->vdev_id;
626 mv.mv_msid = m->ms_id;
627 mv.mv_start = m->ms_start;
628 mv.mv_end = m->ms_start + m->ms_size;
629 zfs_btree_create(&mv.mv_livelist_allocs,
630 livelist_block_compare, NULL,
631 sizeof (sublivelist_verify_block_t));
632
633 mv_populate_livelist_allocs(&mv, &sv);
634
635 spacemap_check_ms_sm(m->ms_sm, &mv);
636 spacemap_check_sm_log(spa, &mv);
637
638 zfs_range_tree_vacate(mv.mv_allocated, NULL, NULL);
639 zfs_range_tree_destroy(mv.mv_allocated);
640 zfs_btree_clear(&mv.mv_livelist_allocs);
641 zfs_btree_destroy(&mv.mv_livelist_allocs);
642 }
643 }
644 (void) fprintf(stderr, "\n");
645
646 /*
647 * If there are any segments in the leftover tree after we walked
648 * through all the metaslabs in the concrete vdevs then this means
649 * that we have segments in the livelists that belong to indirect
650 * vdevs and are marked as allocated.
651 */
652 if (zfs_btree_numnodes(&sv.sv_leftover) == 0) {
653 zfs_btree_destroy(&sv.sv_leftover);
654 return;
655 }
656 (void) printf("ERROR: Found livelist blocks marked as allocated "
657 "for indirect vdevs:\n");
658
659 zfs_btree_index_t *where = NULL;
660 sublivelist_verify_block_t *svb;
661 while ((svb = zfs_btree_destroy_nodes(&sv.sv_leftover, &where)) !=
662 NULL) {
663 int vdev_id = DVA_GET_VDEV(&svb->svb_dva);
664 ASSERT3U(vdev_id, <, rvd->vdev_children);
665 vdev_t *vd = rvd->vdev_child[vdev_id];
666 ASSERT(!vdev_is_concrete(vd));
667 (void) printf("<%d:%llx:%llx> TXG %llx\n",
668 vdev_id, (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
669 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva),
670 (u_longlong_t)svb->svb_allocated_txg);
671 }
672 (void) printf("\n");
673 zfs_btree_destroy(&sv.sv_leftover);
674 }
675
676 /*
677 * These libumem hooks provide a reasonable set of defaults for the allocator's
678 * debugging facilities.
679 */
680 const char *
_umem_debug_init(void)681 _umem_debug_init(void)
682 {
683 return ("default,verbose"); /* $UMEM_DEBUG setting */
684 }
685
686 const char *
_umem_logging_init(void)687 _umem_logging_init(void)
688 {
689 return ("fail,contents"); /* $UMEM_LOGGING setting */
690 }
691
692 static void
usage(void)693 usage(void)
694 {
695 (void) fprintf(stderr,
696 "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
697 "[-I <inflight I/Os>]\n"
698 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
699 "\t\t[-K <key>]\n"
700 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
701 "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
702 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
703 "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
704 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
705 "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
706 "\t%s [-v] <bookmark>\n"
707 "\t%s -C [-A] [-U <cache>] [<poolname>]\n"
708 "\t%s -l [-Aqu] <device>\n"
709 "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
710 "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
711 "\t%s -O [-K <key>] <dataset> <path>\n"
712 "\t%s -r [-K <key>] <dataset> <path> <destination>\n"
713 "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
714 "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
715 "\t%s -E [-A] word0:word1:...:word15\n"
716 "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
717 "<poolname>\n\n",
718 cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
719 cmdname, cmdname, cmdname, cmdname, cmdname);
720
721 (void) fprintf(stderr, " Dataset name must include at least one "
722 "separator character '/' or '@'\n");
723 (void) fprintf(stderr, " If dataset name is specified, only that "
724 "dataset is dumped\n");
725 (void) fprintf(stderr, " If object numbers or object number "
726 "ranges are specified, only those\n"
727 " objects or ranges are dumped.\n\n");
728 (void) fprintf(stderr,
729 " Object ranges take the form <start>:<end>[:<flags>]\n"
730 " start Starting object number\n"
731 " end Ending object number, or -1 for no upper bound\n"
732 " flags Optional flags to select object types:\n"
733 " A All objects (this is the default)\n"
734 " d ZFS directories\n"
735 " f ZFS files \n"
736 " m SPA space maps\n"
737 " z ZAPs\n"
738 " - Negate effect of next flag\n\n");
739 (void) fprintf(stderr, " Options to control amount of output:\n");
740 (void) fprintf(stderr, " -b --block-stats "
741 "block statistics\n");
742 (void) fprintf(stderr, " -B --backup "
743 "backup stream\n");
744 (void) fprintf(stderr, " -c --checksum "
745 "checksum all metadata (twice for all data) blocks\n");
746 (void) fprintf(stderr, " -C --config "
747 "config (or cachefile if alone)\n");
748 (void) fprintf(stderr, " -d --datasets "
749 "dataset(s)\n");
750 (void) fprintf(stderr, " -D --dedup-stats "
751 "dedup statistics\n");
752 (void) fprintf(stderr, " -E --embedded-block-pointer=INTEGER\n"
753 " decode and display block "
754 "from an embedded block pointer\n");
755 (void) fprintf(stderr, " -h --history "
756 "pool history\n");
757 (void) fprintf(stderr, " -i --intent-logs "
758 "intent logs\n");
759 (void) fprintf(stderr, " -l --label "
760 "read label contents\n");
761 (void) fprintf(stderr, " -k --checkpointed-state "
762 "examine the checkpointed state of the pool\n");
763 (void) fprintf(stderr, " -L --disable-leak-tracking "
764 "disable leak tracking (do not load spacemaps)\n");
765 (void) fprintf(stderr, " -m --metaslabs "
766 "metaslabs\n");
767 (void) fprintf(stderr, " -M --metaslab-groups "
768 "metaslab groups\n");
769 (void) fprintf(stderr, " -O --object-lookups "
770 "perform object lookups by path\n");
771 (void) fprintf(stderr, " -r --copy-object "
772 "copy an object by path to file\n");
773 (void) fprintf(stderr, " -R --read-block "
774 "read and display block from a device\n");
775 (void) fprintf(stderr, " -s --io-stats "
776 "report stats on zdb's I/O\n");
777 (void) fprintf(stderr, " -S --simulate-dedup "
778 "simulate dedup to measure effect\n");
779 (void) fprintf(stderr, " -v --verbose "
780 "verbose (applies to all others)\n");
781 (void) fprintf(stderr, " -y --livelist "
782 "perform livelist and metaslab validation on any livelists being "
783 "deleted\n\n");
784 (void) fprintf(stderr, " Below options are intended for use "
785 "with other options:\n");
786 (void) fprintf(stderr, " -A --ignore-assertions "
787 "ignore assertions (-A), enable panic recovery (-AA) or both "
788 "(-AAA)\n");
789 (void) fprintf(stderr, " -e --exported "
790 "pool is exported/destroyed/has altroot/not in a cachefile\n");
791 (void) fprintf(stderr, " -F --automatic-rewind "
792 "attempt automatic rewind within safe range of transaction "
793 "groups\n");
794 (void) fprintf(stderr, " -G --dump-debug-msg "
795 "dump zfs_dbgmsg buffer before exiting\n");
796 (void) fprintf(stderr, " -I --inflight=INTEGER "
797 "specify the maximum number of checksumming I/Os "
798 "[default is 200]\n");
799 (void) fprintf(stderr, " -K --key=KEY "
800 "decryption key for encrypted dataset\n");
801 (void) fprintf(stderr, " -o --option=\"NAME=VALUE\" "
802 "set the named tunable to the given value\n");
803 (void) fprintf(stderr, " -p --path==PATH "
804 "use one or more with -e to specify path to vdev dir\n");
805 (void) fprintf(stderr, " -P --parseable "
806 "print numbers in parseable form\n");
807 (void) fprintf(stderr, " -q --skip-label "
808 "don't print label contents\n");
809 (void) fprintf(stderr, " -t --txg=INTEGER "
810 "highest txg to use when searching for uberblocks\n");
811 (void) fprintf(stderr, " -T --brt-stats "
812 "BRT statistics\n");
813 (void) fprintf(stderr, " -u --uberblock "
814 "uberblock\n");
815 (void) fprintf(stderr, " -U --cachefile=PATH "
816 "use alternate cachefile\n");
817 (void) fprintf(stderr, " -V --verbatim "
818 "do verbatim import\n");
819 (void) fprintf(stderr, " -x --dump-blocks=PATH "
820 "dump all read blocks into specified directory\n");
821 (void) fprintf(stderr, " -X --extreme-rewind "
822 "attempt extreme rewind (does not work with dataset)\n");
823 (void) fprintf(stderr, " -Y --all-reconstruction "
824 "attempt all reconstruction combinations for split blocks\n");
825 (void) fprintf(stderr, " -Z --zstd-headers "
826 "show ZSTD headers \n");
827 (void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
828 "to make only that option verbose\n");
829 (void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
830 zdb_exit(1);
831 }
832
833 static void
dump_debug_buffer(void)834 dump_debug_buffer(void)
835 {
836 ssize_t ret __attribute__((unused));
837
838 if (!dump_opt['G'])
839 return;
840 /*
841 * We use write() instead of printf() so that this function
842 * is safe to call from a signal handler.
843 */
844 ret = write(STDERR_FILENO, "\n", 1);
845 zfs_dbgmsg_print(STDERR_FILENO, "zdb");
846 }
847
sig_handler(int signo)848 static void sig_handler(int signo)
849 {
850 struct sigaction action;
851
852 libspl_backtrace(STDERR_FILENO);
853 dump_debug_buffer();
854
855 /*
856 * Restore default action and re-raise signal so SIGSEGV and
857 * SIGABRT can trigger a core dump.
858 */
859 action.sa_handler = SIG_DFL;
860 sigemptyset(&action.sa_mask);
861 action.sa_flags = 0;
862 (void) sigaction(signo, &action, NULL);
863 raise(signo);
864 }
865
866 /*
867 * Called for usage errors that are discovered after a call to spa_open(),
868 * dmu_bonus_hold(), or pool_match(). abort() is called for other errors.
869 */
870
871 static void
fatal(const char * fmt,...)872 fatal(const char *fmt, ...)
873 {
874 va_list ap;
875
876 va_start(ap, fmt);
877 (void) fprintf(stderr, "%s: ", cmdname);
878 (void) vfprintf(stderr, fmt, ap);
879 va_end(ap);
880 (void) fprintf(stderr, "\n");
881
882 dump_debug_buffer();
883
884 zdb_exit(1);
885 }
886
887 static void
dump_packed_nvlist(objset_t * os,uint64_t object,void * data,size_t size)888 dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size)
889 {
890 (void) size;
891 nvlist_t *nv;
892 size_t nvsize = *(uint64_t *)data;
893 char *packed = umem_alloc(nvsize, UMEM_NOFAIL);
894
895 VERIFY(0 == dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH));
896
897 VERIFY(nvlist_unpack(packed, nvsize, &nv, 0) == 0);
898
899 umem_free(packed, nvsize);
900
901 dump_nvlist(nv, 8);
902
903 nvlist_free(nv);
904 }
905
906 static void
dump_history_offsets(objset_t * os,uint64_t object,void * data,size_t size)907 dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size)
908 {
909 (void) os, (void) object, (void) size;
910 spa_history_phys_t *shp = data;
911
912 if (shp == NULL)
913 return;
914
915 (void) printf("\t\tpool_create_len = %llu\n",
916 (u_longlong_t)shp->sh_pool_create_len);
917 (void) printf("\t\tphys_max_off = %llu\n",
918 (u_longlong_t)shp->sh_phys_max_off);
919 (void) printf("\t\tbof = %llu\n",
920 (u_longlong_t)shp->sh_bof);
921 (void) printf("\t\teof = %llu\n",
922 (u_longlong_t)shp->sh_eof);
923 (void) printf("\t\trecords_lost = %llu\n",
924 (u_longlong_t)shp->sh_records_lost);
925 }
926
927 static void
zdb_nicenum(uint64_t num,char * buf,size_t buflen)928 zdb_nicenum(uint64_t num, char *buf, size_t buflen)
929 {
930 if (dump_opt['P'])
931 (void) snprintf(buf, buflen, "%llu", (longlong_t)num);
932 else
933 nicenum(num, buf, buflen);
934 }
935
936 static void
zdb_nicebytes(uint64_t bytes,char * buf,size_t buflen)937 zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen)
938 {
939 if (dump_opt['P'])
940 (void) snprintf(buf, buflen, "%llu", (longlong_t)bytes);
941 else
942 zfs_nicebytes(bytes, buf, buflen);
943 }
944
945 static const char histo_stars[] = "****************************************";
946 static const uint64_t histo_width = sizeof (histo_stars) - 1;
947
948 static void
dump_histogram(const uint64_t * histo,int size,int offset)949 dump_histogram(const uint64_t *histo, int size, int offset)
950 {
951 int i;
952 int minidx = size - 1;
953 int maxidx = 0;
954 uint64_t max = 0;
955
956 for (i = 0; i < size; i++) {
957 if (histo[i] == 0)
958 continue;
959 if (histo[i] > max)
960 max = histo[i];
961 if (i > maxidx)
962 maxidx = i;
963 if (i < minidx)
964 minidx = i;
965 }
966
967 if (max < histo_width)
968 max = histo_width;
969
970 for (i = minidx; i <= maxidx; i++) {
971 (void) printf("\t\t\t%3u: %6llu %s\n",
972 i + offset, (u_longlong_t)histo[i],
973 &histo_stars[(max - histo[i]) * histo_width / max]);
974 }
975 }
976
977 static void
dump_zap_stats(objset_t * os,uint64_t object)978 dump_zap_stats(objset_t *os, uint64_t object)
979 {
980 int error;
981 zap_stats_t zs;
982
983 error = zap_get_stats(os, object, &zs);
984 if (error)
985 return;
986
987 if (zs.zs_ptrtbl_len == 0) {
988 ASSERT(zs.zs_num_blocks == 1);
989 (void) printf("\tmicrozap: %llu bytes, %llu entries\n",
990 (u_longlong_t)zs.zs_blocksize,
991 (u_longlong_t)zs.zs_num_entries);
992 return;
993 }
994
995 (void) printf("\tFat ZAP stats:\n");
996
997 (void) printf("\t\tPointer table:\n");
998 (void) printf("\t\t\t%llu elements\n",
999 (u_longlong_t)zs.zs_ptrtbl_len);
1000 (void) printf("\t\t\tzt_blk: %llu\n",
1001 (u_longlong_t)zs.zs_ptrtbl_zt_blk);
1002 (void) printf("\t\t\tzt_numblks: %llu\n",
1003 (u_longlong_t)zs.zs_ptrtbl_zt_numblks);
1004 (void) printf("\t\t\tzt_shift: %llu\n",
1005 (u_longlong_t)zs.zs_ptrtbl_zt_shift);
1006 (void) printf("\t\t\tzt_blks_copied: %llu\n",
1007 (u_longlong_t)zs.zs_ptrtbl_blks_copied);
1008 (void) printf("\t\t\tzt_nextblk: %llu\n",
1009 (u_longlong_t)zs.zs_ptrtbl_nextblk);
1010
1011 (void) printf("\t\tZAP entries: %llu\n",
1012 (u_longlong_t)zs.zs_num_entries);
1013 (void) printf("\t\tLeaf blocks: %llu\n",
1014 (u_longlong_t)zs.zs_num_leafs);
1015 (void) printf("\t\tTotal blocks: %llu\n",
1016 (u_longlong_t)zs.zs_num_blocks);
1017 (void) printf("\t\tzap_block_type: 0x%llx\n",
1018 (u_longlong_t)zs.zs_block_type);
1019 (void) printf("\t\tzap_magic: 0x%llx\n",
1020 (u_longlong_t)zs.zs_magic);
1021 (void) printf("\t\tzap_salt: 0x%llx\n",
1022 (u_longlong_t)zs.zs_salt);
1023
1024 (void) printf("\t\tLeafs with 2^n pointers:\n");
1025 dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);
1026
1027 (void) printf("\t\tBlocks with n*5 entries:\n");
1028 dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);
1029
1030 (void) printf("\t\tBlocks n/10 full:\n");
1031 dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);
1032
1033 (void) printf("\t\tEntries with n chunks:\n");
1034 dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);
1035
1036 (void) printf("\t\tBuckets with n entries:\n");
1037 dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
1038 }
1039
1040 static void
dump_none(objset_t * os,uint64_t object,void * data,size_t size)1041 dump_none(objset_t *os, uint64_t object, void *data, size_t size)
1042 {
1043 (void) os, (void) object, (void) data, (void) size;
1044 }
1045
1046 static void
dump_unknown(objset_t * os,uint64_t object,void * data,size_t size)1047 dump_unknown(objset_t *os, uint64_t object, void *data, size_t size)
1048 {
1049 (void) os, (void) object, (void) data, (void) size;
1050 (void) printf("\tUNKNOWN OBJECT TYPE\n");
1051 }
1052
1053 static void
dump_uint8(objset_t * os,uint64_t object,void * data,size_t size)1054 dump_uint8(objset_t *os, uint64_t object, void *data, size_t size)
1055 {
1056 (void) os, (void) object, (void) data, (void) size;
1057 }
1058
1059 static void
dump_uint64(objset_t * os,uint64_t object,void * data,size_t size)1060 dump_uint64(objset_t *os, uint64_t object, void *data, size_t size)
1061 {
1062 uint64_t *arr;
1063 uint64_t oursize;
1064 if (dump_opt['d'] < 6)
1065 return;
1066
1067 if (data == NULL) {
1068 dmu_object_info_t doi;
1069
1070 VERIFY0(dmu_object_info(os, object, &doi));
1071 size = doi.doi_max_offset;
1072 /*
1073 * We cap the size at 1 mebibyte here to prevent
1074 * allocation failures and nigh-infinite printing if the
1075 * object is extremely large.
1076 */
1077 oursize = MIN(size, 1 << 20);
1078 arr = kmem_alloc(oursize, KM_SLEEP);
1079
1080 int err = dmu_read(os, object, 0, oursize, arr, 0);
1081 if (err != 0) {
1082 (void) printf("got error %u from dmu_read\n", err);
1083 kmem_free(arr, oursize);
1084 return;
1085 }
1086 } else {
1087 /*
1088 * Even though the allocation is already done in this code path,
1089 * we still cap the size to prevent excessive printing.
1090 */
1091 oursize = MIN(size, 1 << 20);
1092 arr = data;
1093 }
1094
1095 if (size == 0) {
1096 if (data == NULL)
1097 kmem_free(arr, oursize);
1098 (void) printf("\t\t[]\n");
1099 return;
1100 }
1101
1102 (void) printf("\t\t[%0llx", (u_longlong_t)arr[0]);
1103 for (size_t i = 1; i * sizeof (uint64_t) < oursize; i++) {
1104 if (i % 4 != 0)
1105 (void) printf(", %0llx", (u_longlong_t)arr[i]);
1106 else
1107 (void) printf(",\n\t\t%0llx", (u_longlong_t)arr[i]);
1108 }
1109 if (oursize != size)
1110 (void) printf(", ... ");
1111 (void) printf("]\n");
1112
1113 if (data == NULL)
1114 kmem_free(arr, oursize);
1115 }
1116
1117 static void
dump_zap(objset_t * os,uint64_t object,void * data,size_t size)1118 dump_zap(objset_t *os, uint64_t object, void *data, size_t size)
1119 {
1120 (void) data, (void) size;
1121 zap_cursor_t zc;
1122 zap_attribute_t *attrp = zap_attribute_long_alloc();
1123 void *prop;
1124 unsigned i;
1125
1126 dump_zap_stats(os, object);
1127 (void) printf("\n");
1128
1129 for (zap_cursor_init(&zc, os, object);
1130 zap_cursor_retrieve(&zc, attrp) == 0;
1131 zap_cursor_advance(&zc)) {
1132 boolean_t key64 =
1133 !!(zap_getflags(zc.zc_zap) & ZAP_FLAG_UINT64_KEY);
1134
1135 if (key64)
1136 (void) printf("\t\t0x%010" PRIu64 "x = ",
1137 *(uint64_t *)attrp->za_name);
1138 else
1139 (void) printf("\t\t%s = ", attrp->za_name);
1140
1141 if (attrp->za_num_integers == 0) {
1142 (void) printf("\n");
1143 continue;
1144 }
1145 prop = umem_zalloc(attrp->za_num_integers *
1146 attrp->za_integer_length, UMEM_NOFAIL);
1147
1148 if (key64)
1149 (void) zap_lookup_uint64(os, object,
1150 (const uint64_t *)attrp->za_name, 1,
1151 attrp->za_integer_length, attrp->za_num_integers,
1152 prop);
1153 else
1154 (void) zap_lookup(os, object, attrp->za_name,
1155 attrp->za_integer_length, attrp->za_num_integers,
1156 prop);
1157
1158 if (attrp->za_integer_length == 1 && !key64) {
1159 if (strcmp(attrp->za_name,
1160 DSL_CRYPTO_KEY_MASTER_KEY) == 0 ||
1161 strcmp(attrp->za_name,
1162 DSL_CRYPTO_KEY_HMAC_KEY) == 0 ||
1163 strcmp(attrp->za_name, DSL_CRYPTO_KEY_IV) == 0 ||
1164 strcmp(attrp->za_name, DSL_CRYPTO_KEY_MAC) == 0 ||
1165 strcmp(attrp->za_name,
1166 DMU_POOL_CHECKSUM_SALT) == 0) {
1167 uint8_t *u8 = prop;
1168
1169 for (i = 0; i < attrp->za_num_integers; i++) {
1170 (void) printf("%02x", u8[i]);
1171 }
1172 } else {
1173 (void) printf("%s", (char *)prop);
1174 }
1175 } else {
1176 for (i = 0; i < attrp->za_num_integers; i++) {
1177 switch (attrp->za_integer_length) {
1178 case 1:
1179 (void) printf("%u ",
1180 ((uint8_t *)prop)[i]);
1181 break;
1182 case 2:
1183 (void) printf("%u ",
1184 ((uint16_t *)prop)[i]);
1185 break;
1186 case 4:
1187 (void) printf("%u ",
1188 ((uint32_t *)prop)[i]);
1189 break;
1190 case 8:
1191 (void) printf("%lld ",
1192 (u_longlong_t)((int64_t *)prop)[i]);
1193 break;
1194 }
1195 }
1196 }
1197 (void) printf("\n");
1198 umem_free(prop,
1199 attrp->za_num_integers * attrp->za_integer_length);
1200 }
1201 zap_cursor_fini(&zc);
1202 zap_attribute_free(attrp);
1203 }
1204
1205 static void
dump_bpobj(objset_t * os,uint64_t object,void * data,size_t size)1206 dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size)
1207 {
1208 bpobj_phys_t *bpop = data;
1209 uint64_t i;
1210 char bytes[32], comp[32], uncomp[32];
1211
1212 /* make sure the output won't get truncated */
1213 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
1214 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
1215 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
1216
1217 if (bpop == NULL)
1218 return;
1219
1220 zdb_nicenum(bpop->bpo_bytes, bytes, sizeof (bytes));
1221 zdb_nicenum(bpop->bpo_comp, comp, sizeof (comp));
1222 zdb_nicenum(bpop->bpo_uncomp, uncomp, sizeof (uncomp));
1223
1224 (void) printf("\t\tnum_blkptrs = %llu\n",
1225 (u_longlong_t)bpop->bpo_num_blkptrs);
1226 (void) printf("\t\tbytes = %s\n", bytes);
1227 if (size >= BPOBJ_SIZE_V1) {
1228 (void) printf("\t\tcomp = %s\n", comp);
1229 (void) printf("\t\tuncomp = %s\n", uncomp);
1230 }
1231 if (size >= BPOBJ_SIZE_V2) {
1232 (void) printf("\t\tsubobjs = %llu\n",
1233 (u_longlong_t)bpop->bpo_subobjs);
1234 (void) printf("\t\tnum_subobjs = %llu\n",
1235 (u_longlong_t)bpop->bpo_num_subobjs);
1236 }
1237 if (size >= sizeof (*bpop)) {
1238 (void) printf("\t\tnum_freed = %llu\n",
1239 (u_longlong_t)bpop->bpo_num_freed);
1240 }
1241
1242 if (dump_opt['d'] < 5)
1243 return;
1244
1245 for (i = 0; i < bpop->bpo_num_blkptrs; i++) {
1246 char blkbuf[BP_SPRINTF_LEN];
1247 blkptr_t bp;
1248
1249 int err = dmu_read(os, object,
1250 i * sizeof (bp), sizeof (bp), &bp, 0);
1251 if (err != 0) {
1252 (void) printf("got error %u from dmu_read\n", err);
1253 break;
1254 }
1255 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp,
1256 BP_GET_FREE(&bp));
1257 (void) printf("\t%s\n", blkbuf);
1258 }
1259 }
1260
1261 static void
dump_bpobj_subobjs(objset_t * os,uint64_t object,void * data,size_t size)1262 dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size)
1263 {
1264 (void) data, (void) size;
1265 dmu_object_info_t doi;
1266 int64_t i;
1267
1268 VERIFY0(dmu_object_info(os, object, &doi));
1269 uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP);
1270
1271 int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0);
1272 if (err != 0) {
1273 (void) printf("got error %u from dmu_read\n", err);
1274 kmem_free(subobjs, doi.doi_max_offset);
1275 return;
1276 }
1277
1278 int64_t last_nonzero = -1;
1279 for (i = 0; i < doi.doi_max_offset / 8; i++) {
1280 if (subobjs[i] != 0)
1281 last_nonzero = i;
1282 }
1283
1284 for (i = 0; i <= last_nonzero; i++) {
1285 (void) printf("\t%llu\n", (u_longlong_t)subobjs[i]);
1286 }
1287 kmem_free(subobjs, doi.doi_max_offset);
1288 }
1289
1290 static void
dump_ddt_zap(objset_t * os,uint64_t object,void * data,size_t size)1291 dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size)
1292 {
1293 (void) data, (void) size;
1294 dump_zap_stats(os, object);
1295 /* contents are printed elsewhere, properly decoded */
1296 }
1297
1298 static void
dump_sa_attrs(objset_t * os,uint64_t object,void * data,size_t size)1299 dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size)
1300 {
1301 (void) data, (void) size;
1302 zap_cursor_t zc;
1303 zap_attribute_t *attrp = zap_attribute_alloc();
1304
1305 dump_zap_stats(os, object);
1306 (void) printf("\n");
1307
1308 for (zap_cursor_init(&zc, os, object);
1309 zap_cursor_retrieve(&zc, attrp) == 0;
1310 zap_cursor_advance(&zc)) {
1311 (void) printf("\t\t%s = ", attrp->za_name);
1312 if (attrp->za_num_integers == 0) {
1313 (void) printf("\n");
1314 continue;
1315 }
1316 (void) printf(" %llx : [%d:%d:%d]\n",
1317 (u_longlong_t)attrp->za_first_integer,
1318 (int)ATTR_LENGTH(attrp->za_first_integer),
1319 (int)ATTR_BSWAP(attrp->za_first_integer),
1320 (int)ATTR_NUM(attrp->za_first_integer));
1321 }
1322 zap_cursor_fini(&zc);
1323 zap_attribute_free(attrp);
1324 }
1325
1326 static void
dump_sa_layouts(objset_t * os,uint64_t object,void * data,size_t size)1327 dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size)
1328 {
1329 (void) data, (void) size;
1330 zap_cursor_t zc;
1331 zap_attribute_t *attrp = zap_attribute_alloc();
1332 uint16_t *layout_attrs;
1333 unsigned i;
1334
1335 dump_zap_stats(os, object);
1336 (void) printf("\n");
1337
1338 for (zap_cursor_init(&zc, os, object);
1339 zap_cursor_retrieve(&zc, attrp) == 0;
1340 zap_cursor_advance(&zc)) {
1341 (void) printf("\t\t%s = [", attrp->za_name);
1342 if (attrp->za_num_integers == 0) {
1343 (void) printf("\n");
1344 continue;
1345 }
1346
1347 VERIFY(attrp->za_integer_length == 2);
1348 layout_attrs = umem_zalloc(attrp->za_num_integers *
1349 attrp->za_integer_length, UMEM_NOFAIL);
1350
1351 VERIFY(zap_lookup(os, object, attrp->za_name,
1352 attrp->za_integer_length,
1353 attrp->za_num_integers, layout_attrs) == 0);
1354
1355 for (i = 0; i != attrp->za_num_integers; i++)
1356 (void) printf(" %d ", (int)layout_attrs[i]);
1357 (void) printf("]\n");
1358 umem_free(layout_attrs,
1359 attrp->za_num_integers * attrp->za_integer_length);
1360 }
1361 zap_cursor_fini(&zc);
1362 zap_attribute_free(attrp);
1363 }
1364
1365 static void
dump_zpldir(objset_t * os,uint64_t object,void * data,size_t size)1366 dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
1367 {
1368 (void) data, (void) size;
1369 zap_cursor_t zc;
1370 zap_attribute_t *attrp = zap_attribute_long_alloc();
1371 const char *typenames[] = {
1372 /* 0 */ "not specified",
1373 /* 1 */ "FIFO",
1374 /* 2 */ "Character Device",
1375 /* 3 */ "3 (invalid)",
1376 /* 4 */ "Directory",
1377 /* 5 */ "5 (invalid)",
1378 /* 6 */ "Block Device",
1379 /* 7 */ "7 (invalid)",
1380 /* 8 */ "Regular File",
1381 /* 9 */ "9 (invalid)",
1382 /* 10 */ "Symbolic Link",
1383 /* 11 */ "11 (invalid)",
1384 /* 12 */ "Socket",
1385 /* 13 */ "Door",
1386 /* 14 */ "Event Port",
1387 /* 15 */ "15 (invalid)",
1388 };
1389
1390 dump_zap_stats(os, object);
1391 (void) printf("\n");
1392
1393 for (zap_cursor_init(&zc, os, object);
1394 zap_cursor_retrieve(&zc, attrp) == 0;
1395 zap_cursor_advance(&zc)) {
1396 (void) printf("\t\t%s = %lld (type: %s)\n",
1397 attrp->za_name, ZFS_DIRENT_OBJ(attrp->za_first_integer),
1398 typenames[ZFS_DIRENT_TYPE(attrp->za_first_integer)]);
1399 }
1400 zap_cursor_fini(&zc);
1401 zap_attribute_free(attrp);
1402 }
1403
1404 static int
get_dtl_refcount(vdev_t * vd)1405 get_dtl_refcount(vdev_t *vd)
1406 {
1407 int refcount = 0;
1408
1409 if (vd->vdev_ops->vdev_op_leaf) {
1410 space_map_t *sm = vd->vdev_dtl_sm;
1411
1412 if (sm != NULL &&
1413 sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1414 return (1);
1415 return (0);
1416 }
1417
1418 for (unsigned c = 0; c < vd->vdev_children; c++)
1419 refcount += get_dtl_refcount(vd->vdev_child[c]);
1420 return (refcount);
1421 }
1422
1423 static int
get_metaslab_refcount(vdev_t * vd)1424 get_metaslab_refcount(vdev_t *vd)
1425 {
1426 int refcount = 0;
1427
1428 if (vd->vdev_top == vd) {
1429 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
1430 space_map_t *sm = vd->vdev_ms[m]->ms_sm;
1431
1432 if (sm != NULL &&
1433 sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1434 refcount++;
1435 }
1436 }
1437 for (unsigned c = 0; c < vd->vdev_children; c++)
1438 refcount += get_metaslab_refcount(vd->vdev_child[c]);
1439
1440 return (refcount);
1441 }
1442
1443 static int
get_obsolete_refcount(vdev_t * vd)1444 get_obsolete_refcount(vdev_t *vd)
1445 {
1446 uint64_t obsolete_sm_object;
1447 int refcount = 0;
1448
1449 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1450 if (vd->vdev_top == vd && obsolete_sm_object != 0) {
1451 dmu_object_info_t doi;
1452 VERIFY0(dmu_object_info(vd->vdev_spa->spa_meta_objset,
1453 obsolete_sm_object, &doi));
1454 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1455 refcount++;
1456 }
1457 } else {
1458 ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
1459 ASSERT3U(obsolete_sm_object, ==, 0);
1460 }
1461 for (unsigned c = 0; c < vd->vdev_children; c++) {
1462 refcount += get_obsolete_refcount(vd->vdev_child[c]);
1463 }
1464
1465 return (refcount);
1466 }
1467
1468 static int
get_prev_obsolete_spacemap_refcount(spa_t * spa)1469 get_prev_obsolete_spacemap_refcount(spa_t *spa)
1470 {
1471 uint64_t prev_obj =
1472 spa->spa_condensing_indirect_phys.scip_prev_obsolete_sm_object;
1473 if (prev_obj != 0) {
1474 dmu_object_info_t doi;
1475 VERIFY0(dmu_object_info(spa->spa_meta_objset, prev_obj, &doi));
1476 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1477 return (1);
1478 }
1479 }
1480 return (0);
1481 }
1482
1483 static int
get_checkpoint_refcount(vdev_t * vd)1484 get_checkpoint_refcount(vdev_t *vd)
1485 {
1486 int refcount = 0;
1487
1488 if (vd->vdev_top == vd && vd->vdev_top_zap != 0 &&
1489 zap_contains(spa_meta_objset(vd->vdev_spa),
1490 vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) == 0)
1491 refcount++;
1492
1493 for (uint64_t c = 0; c < vd->vdev_children; c++)
1494 refcount += get_checkpoint_refcount(vd->vdev_child[c]);
1495
1496 return (refcount);
1497 }
1498
1499 static int
get_log_spacemap_refcount(spa_t * spa)1500 get_log_spacemap_refcount(spa_t *spa)
1501 {
1502 return (avl_numnodes(&spa->spa_sm_logs_by_txg));
1503 }
1504
1505 static int
verify_spacemap_refcounts(spa_t * spa)1506 verify_spacemap_refcounts(spa_t *spa)
1507 {
1508 uint64_t expected_refcount = 0;
1509 uint64_t actual_refcount;
1510
1511 (void) feature_get_refcount(spa,
1512 &spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM],
1513 &expected_refcount);
1514 actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
1515 actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);
1516 actual_refcount += get_obsolete_refcount(spa->spa_root_vdev);
1517 actual_refcount += get_prev_obsolete_spacemap_refcount(spa);
1518 actual_refcount += get_checkpoint_refcount(spa->spa_root_vdev);
1519 actual_refcount += get_log_spacemap_refcount(spa);
1520
1521 if (expected_refcount != actual_refcount) {
1522 (void) printf("space map refcount mismatch: expected %lld != "
1523 "actual %lld\n",
1524 (longlong_t)expected_refcount,
1525 (longlong_t)actual_refcount);
1526 return (2);
1527 }
1528 return (0);
1529 }
1530
1531 static void
dump_spacemap(objset_t * os,space_map_t * sm)1532 dump_spacemap(objset_t *os, space_map_t *sm)
1533 {
1534 const char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1535 "INVALID", "INVALID", "INVALID", "INVALID" };
1536
1537 if (sm == NULL)
1538 return;
1539
1540 (void) printf("space map object %llu:\n",
1541 (longlong_t)sm->sm_object);
1542 (void) printf(" smp_length = 0x%llx\n",
1543 (longlong_t)sm->sm_phys->smp_length);
1544 (void) printf(" smp_alloc = 0x%llx\n",
1545 (longlong_t)sm->sm_phys->smp_alloc);
1546
1547 if (dump_opt['d'] < 6 && dump_opt['m'] < 4)
1548 return;
1549
1550 /*
1551 * Print out the freelist entries in both encoded and decoded form.
1552 */
1553 uint8_t mapshift = sm->sm_shift;
1554 int64_t alloc = 0;
1555 uint64_t word, entry_id = 0;
1556 for (uint64_t offset = 0; offset < space_map_length(sm);
1557 offset += sizeof (word)) {
1558
1559 VERIFY0(dmu_read(os, space_map_object(sm), offset,
1560 sizeof (word), &word, DMU_READ_PREFETCH));
1561
1562 if (sm_entry_is_debug(word)) {
1563 uint64_t de_txg = SM_DEBUG_TXG_DECODE(word);
1564 uint64_t de_sync_pass = SM_DEBUG_SYNCPASS_DECODE(word);
1565 if (de_txg == 0) {
1566 (void) printf(
1567 "\t [%6llu] PADDING\n",
1568 (u_longlong_t)entry_id);
1569 } else {
1570 (void) printf(
1571 "\t [%6llu] %s: txg %llu pass %llu\n",
1572 (u_longlong_t)entry_id,
1573 ddata[SM_DEBUG_ACTION_DECODE(word)],
1574 (u_longlong_t)de_txg,
1575 (u_longlong_t)de_sync_pass);
1576 }
1577 entry_id++;
1578 continue;
1579 }
1580
1581 uint8_t words;
1582 char entry_type;
1583 uint64_t entry_off, entry_run, entry_vdev = SM_NO_VDEVID;
1584
1585 if (sm_entry_is_single_word(word)) {
1586 entry_type = (SM_TYPE_DECODE(word) == SM_ALLOC) ?
1587 'A' : 'F';
1588 entry_off = (SM_OFFSET_DECODE(word) << mapshift) +
1589 sm->sm_start;
1590 entry_run = SM_RUN_DECODE(word) << mapshift;
1591 words = 1;
1592 } else {
1593 /* it is a two-word entry so we read another word */
1594 ASSERT(sm_entry_is_double_word(word));
1595
1596 uint64_t extra_word;
1597 offset += sizeof (extra_word);
1598 VERIFY0(dmu_read(os, space_map_object(sm), offset,
1599 sizeof (extra_word), &extra_word,
1600 DMU_READ_PREFETCH));
1601
1602 ASSERT3U(offset, <=, space_map_length(sm));
1603
1604 entry_run = SM2_RUN_DECODE(word) << mapshift;
1605 entry_vdev = SM2_VDEV_DECODE(word);
1606 entry_type = (SM2_TYPE_DECODE(extra_word) == SM_ALLOC) ?
1607 'A' : 'F';
1608 entry_off = (SM2_OFFSET_DECODE(extra_word) <<
1609 mapshift) + sm->sm_start;
1610 words = 2;
1611 }
1612
1613 (void) printf("\t [%6llu] %c range:"
1614 " %010llx-%010llx size: %06llx vdev: %06llu words: %u\n",
1615 (u_longlong_t)entry_id,
1616 entry_type, (u_longlong_t)entry_off,
1617 (u_longlong_t)(entry_off + entry_run),
1618 (u_longlong_t)entry_run,
1619 (u_longlong_t)entry_vdev, words);
1620
1621 if (entry_type == 'A')
1622 alloc += entry_run;
1623 else
1624 alloc -= entry_run;
1625 entry_id++;
1626 }
1627 if (alloc != space_map_allocated(sm)) {
1628 (void) printf("space_map_object alloc (%lld) INCONSISTENT "
1629 "with space map summary (%lld)\n",
1630 (longlong_t)space_map_allocated(sm), (longlong_t)alloc);
1631 }
1632 }
1633
1634 static void
dump_metaslab_stats(metaslab_t * msp)1635 dump_metaslab_stats(metaslab_t *msp)
1636 {
1637 char maxbuf[32];
1638 zfs_range_tree_t *rt = msp->ms_allocatable;
1639 zfs_btree_t *t = &msp->ms_allocatable_by_size;
1640 int free_pct = zfs_range_tree_space(rt) * 100 / msp->ms_size;
1641
1642 /* max sure nicenum has enough space */
1643 _Static_assert(sizeof (maxbuf) >= NN_NUMBUF_SZ, "maxbuf truncated");
1644
1645 zdb_nicenum(metaslab_largest_allocatable(msp), maxbuf, sizeof (maxbuf));
1646
1647 (void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
1648 "segments", zfs_btree_numnodes(t), "maxsize", maxbuf,
1649 "freepct", free_pct);
1650 (void) printf("\tIn-memory histogram:\n");
1651 dump_histogram(rt->rt_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1652 }
1653
1654 static void
dump_metaslab(metaslab_t * msp)1655 dump_metaslab(metaslab_t *msp)
1656 {
1657 vdev_t *vd = msp->ms_group->mg_vd;
1658 spa_t *spa = vd->vdev_spa;
1659 space_map_t *sm = msp->ms_sm;
1660 char freebuf[32];
1661
1662 zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf,
1663 sizeof (freebuf));
1664
1665 (void) printf(
1666 "\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
1667 (u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
1668 (u_longlong_t)space_map_object(sm), freebuf);
1669
1670 if (dump_opt['m'] > 2 && !dump_opt['L']) {
1671 mutex_enter(&msp->ms_lock);
1672 VERIFY0(metaslab_load(msp));
1673 zfs_range_tree_stat_verify(msp->ms_allocatable);
1674 dump_metaslab_stats(msp);
1675 metaslab_unload(msp);
1676 mutex_exit(&msp->ms_lock);
1677 }
1678
1679 if (dump_opt['m'] > 1 && sm != NULL &&
1680 spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) {
1681 /*
1682 * The space map histogram represents free space in chunks
1683 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1684 */
1685 (void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1686 (u_longlong_t)msp->ms_fragmentation);
1687 dump_histogram(sm->sm_phys->smp_histogram,
1688 SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift);
1689 }
1690
1691 if (vd->vdev_ops == &vdev_draid_ops)
1692 ASSERT3U(msp->ms_size, <=, 1ULL << vd->vdev_ms_shift);
1693 else
1694 ASSERT3U(msp->ms_size, ==, 1ULL << vd->vdev_ms_shift);
1695
1696 dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
1697
1698 if (spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
1699 (void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1700 (u_longlong_t)metaslab_unflushed_txg(msp));
1701 }
1702 }
1703
1704 static void
print_vdev_metaslab_header(vdev_t * vd)1705 print_vdev_metaslab_header(vdev_t *vd)
1706 {
1707 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
1708 const char *bias_str = "";
1709 if (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) {
1710 bias_str = VDEV_ALLOC_BIAS_LOG;
1711 } else if (alloc_bias == VDEV_BIAS_SPECIAL) {
1712 bias_str = VDEV_ALLOC_BIAS_SPECIAL;
1713 } else if (alloc_bias == VDEV_BIAS_DEDUP) {
1714 bias_str = VDEV_ALLOC_BIAS_DEDUP;
1715 }
1716
1717 uint64_t ms_flush_data_obj = 0;
1718 if (vd->vdev_top_zap != 0) {
1719 int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
1720 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
1721 sizeof (uint64_t), 1, &ms_flush_data_obj);
1722 if (error != ENOENT) {
1723 ASSERT0(error);
1724 }
1725 }
1726
1727 (void) printf("\tvdev %10llu %s",
1728 (u_longlong_t)vd->vdev_id, bias_str);
1729
1730 if (ms_flush_data_obj != 0) {
1731 (void) printf(" ms_unflushed_phys object %llu",
1732 (u_longlong_t)ms_flush_data_obj);
1733 }
1734
1735 (void) printf("\n\t%-10s%5llu %-19s %-15s %-12s\n",
1736 "metaslabs", (u_longlong_t)vd->vdev_ms_count,
1737 "offset", "spacemap", "free");
1738 (void) printf("\t%15s %19s %15s %12s\n",
1739 "---------------", "-------------------",
1740 "---------------", "------------");
1741 }
1742
1743 static void
dump_metaslab_groups(spa_t * spa,boolean_t show_special)1744 dump_metaslab_groups(spa_t *spa, boolean_t show_special)
1745 {
1746 vdev_t *rvd = spa->spa_root_vdev;
1747 metaslab_class_t *mc = spa_normal_class(spa);
1748 metaslab_class_t *smc = spa_special_class(spa);
1749 uint64_t fragmentation;
1750
1751 metaslab_class_histogram_verify(mc);
1752
1753 for (unsigned c = 0; c < rvd->vdev_children; c++) {
1754 vdev_t *tvd = rvd->vdev_child[c];
1755 metaslab_group_t *mg = tvd->vdev_mg;
1756
1757 if (mg == NULL || (mg->mg_class != mc &&
1758 (!show_special || mg->mg_class != smc)))
1759 continue;
1760
1761 metaslab_group_histogram_verify(mg);
1762 mg->mg_fragmentation = metaslab_group_fragmentation(mg);
1763
1764 (void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1765 "fragmentation",
1766 (u_longlong_t)tvd->vdev_id,
1767 (u_longlong_t)tvd->vdev_ms_count);
1768 if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
1769 (void) printf("%3s\n", "-");
1770 } else {
1771 (void) printf("%3llu%%\n",
1772 (u_longlong_t)mg->mg_fragmentation);
1773 }
1774 dump_histogram(mg->mg_histogram,
1775 ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1776 }
1777
1778 (void) printf("\tpool %s\tfragmentation", spa_name(spa));
1779 fragmentation = metaslab_class_fragmentation(mc);
1780 if (fragmentation == ZFS_FRAG_INVALID)
1781 (void) printf("\t%3s\n", "-");
1782 else
1783 (void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation);
1784 dump_histogram(mc->mc_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1785 }
1786
1787 static void
print_vdev_indirect(vdev_t * vd)1788 print_vdev_indirect(vdev_t *vd)
1789 {
1790 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
1791 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
1792 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
1793
1794 if (vim == NULL) {
1795 ASSERT3P(vib, ==, NULL);
1796 return;
1797 }
1798
1799 ASSERT3U(vdev_indirect_mapping_object(vim), ==,
1800 vic->vic_mapping_object);
1801 ASSERT3U(vdev_indirect_births_object(vib), ==,
1802 vic->vic_births_object);
1803
1804 (void) printf("indirect births obj %llu:\n",
1805 (longlong_t)vic->vic_births_object);
1806 (void) printf(" vib_count = %llu\n",
1807 (longlong_t)vdev_indirect_births_count(vib));
1808 for (uint64_t i = 0; i < vdev_indirect_births_count(vib); i++) {
1809 vdev_indirect_birth_entry_phys_t *cur_vibe =
1810 &vib->vib_entries[i];
1811 (void) printf("\toffset %llx -> txg %llu\n",
1812 (longlong_t)cur_vibe->vibe_offset,
1813 (longlong_t)cur_vibe->vibe_phys_birth_txg);
1814 }
1815 (void) printf("\n");
1816
1817 (void) printf("indirect mapping obj %llu:\n",
1818 (longlong_t)vic->vic_mapping_object);
1819 (void) printf(" vim_max_offset = 0x%llx\n",
1820 (longlong_t)vdev_indirect_mapping_max_offset(vim));
1821 (void) printf(" vim_bytes_mapped = 0x%llx\n",
1822 (longlong_t)vdev_indirect_mapping_bytes_mapped(vim));
1823 (void) printf(" vim_count = %llu\n",
1824 (longlong_t)vdev_indirect_mapping_num_entries(vim));
1825
1826 if (dump_opt['d'] <= 5 && dump_opt['m'] <= 3)
1827 return;
1828
1829 uint32_t *counts = vdev_indirect_mapping_load_obsolete_counts(vim);
1830
1831 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
1832 vdev_indirect_mapping_entry_phys_t *vimep =
1833 &vim->vim_entries[i];
1834 (void) printf("\t<%llx:%llx:%llx> -> "
1835 "<%llx:%llx:%llx> (%x obsolete)\n",
1836 (longlong_t)vd->vdev_id,
1837 (longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
1838 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1839 (longlong_t)DVA_GET_VDEV(&vimep->vimep_dst),
1840 (longlong_t)DVA_GET_OFFSET(&vimep->vimep_dst),
1841 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1842 counts[i]);
1843 }
1844 (void) printf("\n");
1845
1846 uint64_t obsolete_sm_object;
1847 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1848 if (obsolete_sm_object != 0) {
1849 objset_t *mos = vd->vdev_spa->spa_meta_objset;
1850 (void) printf("obsolete space map object %llu:\n",
1851 (u_longlong_t)obsolete_sm_object);
1852 ASSERT(vd->vdev_obsolete_sm != NULL);
1853 ASSERT3U(space_map_object(vd->vdev_obsolete_sm), ==,
1854 obsolete_sm_object);
1855 dump_spacemap(mos, vd->vdev_obsolete_sm);
1856 (void) printf("\n");
1857 }
1858 }
1859
1860 static void
dump_metaslabs(spa_t * spa)1861 dump_metaslabs(spa_t *spa)
1862 {
1863 vdev_t *vd, *rvd = spa->spa_root_vdev;
1864 uint64_t m, c = 0, children = rvd->vdev_children;
1865
1866 (void) printf("\nMetaslabs:\n");
1867
1868 if (!dump_opt['d'] && zopt_metaslab_args > 0) {
1869 c = zopt_metaslab[0];
1870
1871 if (c >= children)
1872 (void) fatal("bad vdev id: %llu", (u_longlong_t)c);
1873
1874 if (zopt_metaslab_args > 1) {
1875 vd = rvd->vdev_child[c];
1876 print_vdev_metaslab_header(vd);
1877
1878 for (m = 1; m < zopt_metaslab_args; m++) {
1879 if (zopt_metaslab[m] < vd->vdev_ms_count)
1880 dump_metaslab(
1881 vd->vdev_ms[zopt_metaslab[m]]);
1882 else
1883 (void) fprintf(stderr, "bad metaslab "
1884 "number %llu\n",
1885 (u_longlong_t)zopt_metaslab[m]);
1886 }
1887 (void) printf("\n");
1888 return;
1889 }
1890 children = c + 1;
1891 }
1892 for (; c < children; c++) {
1893 vd = rvd->vdev_child[c];
1894 print_vdev_metaslab_header(vd);
1895
1896 print_vdev_indirect(vd);
1897
1898 for (m = 0; m < vd->vdev_ms_count; m++)
1899 dump_metaslab(vd->vdev_ms[m]);
1900 (void) printf("\n");
1901 }
1902 }
1903
1904 static void
dump_log_spacemaps(spa_t * spa)1905 dump_log_spacemaps(spa_t *spa)
1906 {
1907 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1908 return;
1909
1910 (void) printf("\nLog Space Maps in Pool:\n");
1911 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
1912 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1913 space_map_t *sm = NULL;
1914 VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
1915 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
1916
1917 (void) printf("Log Spacemap object %llu txg %llu\n",
1918 (u_longlong_t)sls->sls_sm_obj, (u_longlong_t)sls->sls_txg);
1919 dump_spacemap(spa->spa_meta_objset, sm);
1920 space_map_close(sm);
1921 }
1922 (void) printf("\n");
1923 }
1924
1925 static void
dump_ddt_entry(const ddt_t * ddt,const ddt_lightweight_entry_t * ddlwe,uint64_t index)1926 dump_ddt_entry(const ddt_t *ddt, const ddt_lightweight_entry_t *ddlwe,
1927 uint64_t index)
1928 {
1929 const ddt_key_t *ddk = &ddlwe->ddlwe_key;
1930 char blkbuf[BP_SPRINTF_LEN];
1931 blkptr_t blk;
1932 int p;
1933
1934 for (p = 0; p < DDT_NPHYS(ddt); p++) {
1935 const ddt_univ_phys_t *ddp = &ddlwe->ddlwe_phys;
1936 ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p);
1937
1938 if (ddt_phys_birth(ddp, v) == 0)
1939 continue;
1940 ddt_bp_create(ddt->ddt_checksum, ddk, ddp, v, &blk);
1941 snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
1942 (void) printf("index %llx refcnt %llu phys %d %s\n",
1943 (u_longlong_t)index, (u_longlong_t)ddt_phys_refcnt(ddp, v),
1944 p, blkbuf);
1945 }
1946 }
1947
1948 static void
dump_dedup_ratio(const ddt_stat_t * dds)1949 dump_dedup_ratio(const ddt_stat_t *dds)
1950 {
1951 double rL, rP, rD, D, dedup, compress, copies;
1952
1953 if (dds->dds_blocks == 0)
1954 return;
1955
1956 rL = (double)dds->dds_ref_lsize;
1957 rP = (double)dds->dds_ref_psize;
1958 rD = (double)dds->dds_ref_dsize;
1959 D = (double)dds->dds_dsize;
1960
1961 dedup = rD / D;
1962 compress = rL / rP;
1963 copies = rD / rP;
1964
1965 (void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
1966 "dedup * compress / copies = %.2f\n\n",
1967 dedup, compress, copies, dedup * compress / copies);
1968 }
1969
1970 static void
dump_ddt_log(ddt_t * ddt)1971 dump_ddt_log(ddt_t *ddt)
1972 {
1973 if (ddt->ddt_version != DDT_VERSION_FDT ||
1974 !(ddt->ddt_flags & DDT_FLAG_LOG))
1975 return;
1976
1977 for (int n = 0; n < 2; n++) {
1978 ddt_log_t *ddl = &ddt->ddt_log[n];
1979
1980 char flagstr[64] = {0};
1981 if (ddl->ddl_flags > 0) {
1982 flagstr[0] = ' ';
1983 int c = 1;
1984 if (ddl->ddl_flags & DDL_FLAG_FLUSHING)
1985 c += strlcpy(&flagstr[c], " FLUSHING",
1986 sizeof (flagstr) - c);
1987 if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT)
1988 c += strlcpy(&flagstr[c], " CHECKPOINT",
1989 sizeof (flagstr) - c);
1990 if (ddl->ddl_flags &
1991 ~(DDL_FLAG_FLUSHING|DDL_FLAG_CHECKPOINT))
1992 c += strlcpy(&flagstr[c], " UNKNOWN",
1993 sizeof (flagstr) - c);
1994 flagstr[1] = '[';
1995 flagstr[c] = ']';
1996 }
1997
1998 uint64_t count = avl_numnodes(&ddl->ddl_tree);
1999
2000 printf(DMU_POOL_DDT_LOG ": flags=0x%02x%s; obj=%llu; "
2001 "len=%llu; txg=%llu; entries=%llu\n",
2002 zio_checksum_table[ddt->ddt_checksum].ci_name, n,
2003 ddl->ddl_flags, flagstr,
2004 (u_longlong_t)ddl->ddl_object,
2005 (u_longlong_t)ddl->ddl_length,
2006 (u_longlong_t)ddl->ddl_first_txg, (u_longlong_t)count);
2007
2008 if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT) {
2009 const ddt_key_t *ddk = &ddl->ddl_checkpoint;
2010 printf(" checkpoint: "
2011 "%016llx:%016llx:%016llx:%016llx:%016llx\n",
2012 (u_longlong_t)ddk->ddk_cksum.zc_word[0],
2013 (u_longlong_t)ddk->ddk_cksum.zc_word[1],
2014 (u_longlong_t)ddk->ddk_cksum.zc_word[2],
2015 (u_longlong_t)ddk->ddk_cksum.zc_word[3],
2016 (u_longlong_t)ddk->ddk_prop);
2017 }
2018
2019 if (count == 0 || dump_opt['D'] < 4)
2020 continue;
2021
2022 ddt_lightweight_entry_t ddlwe;
2023 uint64_t index = 0;
2024 for (ddt_log_entry_t *ddle = avl_first(&ddl->ddl_tree);
2025 ddle; ddle = AVL_NEXT(&ddl->ddl_tree, ddle)) {
2026 DDT_LOG_ENTRY_TO_LIGHTWEIGHT(ddt, ddle, &ddlwe);
2027 dump_ddt_entry(ddt, &ddlwe, index++);
2028 }
2029 }
2030 }
2031
2032 static void
dump_ddt_object(ddt_t * ddt,ddt_type_t type,ddt_class_t class)2033 dump_ddt_object(ddt_t *ddt, ddt_type_t type, ddt_class_t class)
2034 {
2035 char name[DDT_NAMELEN];
2036 ddt_lightweight_entry_t ddlwe;
2037 uint64_t walk = 0;
2038 dmu_object_info_t doi;
2039 uint64_t count, dspace, mspace;
2040 int error;
2041
2042 error = ddt_object_info(ddt, type, class, &doi);
2043
2044 if (error == ENOENT)
2045 return;
2046 ASSERT(error == 0);
2047
2048 error = ddt_object_count(ddt, type, class, &count);
2049 ASSERT(error == 0);
2050 if (count == 0)
2051 return;
2052
2053 dspace = doi.doi_physical_blocks_512 << 9;
2054 mspace = doi.doi_fill_count * doi.doi_data_block_size;
2055
2056 ddt_object_name(ddt, type, class, name);
2057
2058 (void) printf("%s: dspace=%llu; mspace=%llu; entries=%llu\n", name,
2059 (u_longlong_t)dspace, (u_longlong_t)mspace, (u_longlong_t)count);
2060
2061 if (dump_opt['D'] < 3)
2062 return;
2063
2064 (void) printf("%s: object=%llu\n", name,
2065 (u_longlong_t)ddt->ddt_object[type][class]);
2066 zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class]);
2067
2068 if (dump_opt['D'] < 4)
2069 return;
2070
2071 if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE)
2072 return;
2073
2074 (void) printf("%s contents:\n\n", name);
2075
2076 while ((error = ddt_object_walk(ddt, type, class, &walk, &ddlwe)) == 0)
2077 dump_ddt_entry(ddt, &ddlwe, walk);
2078
2079 ASSERT3U(error, ==, ENOENT);
2080
2081 (void) printf("\n");
2082 }
2083
2084 static void
dump_ddt(ddt_t * ddt)2085 dump_ddt(ddt_t *ddt)
2086 {
2087 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
2088 return;
2089
2090 char flagstr[64] = {0};
2091 if (ddt->ddt_flags > 0) {
2092 flagstr[0] = ' ';
2093 int c = 1;
2094 if (ddt->ddt_flags & DDT_FLAG_FLAT)
2095 c += strlcpy(&flagstr[c], " FLAT",
2096 sizeof (flagstr) - c);
2097 if (ddt->ddt_flags & DDT_FLAG_LOG)
2098 c += strlcpy(&flagstr[c], " LOG",
2099 sizeof (flagstr) - c);
2100 if (ddt->ddt_flags & ~DDT_FLAG_MASK)
2101 c += strlcpy(&flagstr[c], " UNKNOWN",
2102 sizeof (flagstr) - c);
2103 flagstr[1] = '[';
2104 flagstr[c] = ']';
2105 }
2106
2107 printf("DDT-%s: version=%llu [%s]; flags=0x%02llx%s; rootobj=%llu\n",
2108 zio_checksum_table[ddt->ddt_checksum].ci_name,
2109 (u_longlong_t)ddt->ddt_version,
2110 (ddt->ddt_version == 0) ? "LEGACY" :
2111 (ddt->ddt_version == 1) ? "FDT" : "UNKNOWN",
2112 (u_longlong_t)ddt->ddt_flags, flagstr,
2113 (u_longlong_t)ddt->ddt_dir_object);
2114
2115 for (ddt_type_t type = 0; type < DDT_TYPES; type++)
2116 for (ddt_class_t class = 0; class < DDT_CLASSES; class++)
2117 dump_ddt_object(ddt, type, class);
2118
2119 dump_ddt_log(ddt);
2120 }
2121
2122 static void
dump_all_ddts(spa_t * spa)2123 dump_all_ddts(spa_t *spa)
2124 {
2125 ddt_histogram_t ddh_total = {{{0}}};
2126 ddt_stat_t dds_total = {0};
2127
2128 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++)
2129 dump_ddt(spa->spa_ddt[c]);
2130
2131 ddt_get_dedup_stats(spa, &dds_total);
2132
2133 if (dds_total.dds_blocks == 0) {
2134 (void) printf("All DDTs are empty\n");
2135 return;
2136 }
2137
2138 (void) printf("\n");
2139
2140 if (dump_opt['D'] > 1) {
2141 (void) printf("DDT histogram (aggregated over all DDTs):\n");
2142 ddt_get_dedup_histogram(spa, &ddh_total);
2143 zpool_dump_ddt(&dds_total, &ddh_total);
2144 }
2145
2146 dump_dedup_ratio(&dds_total);
2147
2148 /*
2149 * Dump a histogram of unique class entry age
2150 */
2151 if (dump_opt['D'] == 3 && getenv("ZDB_DDT_UNIQUE_AGE_HIST") != NULL) {
2152 ddt_age_histo_t histogram;
2153
2154 (void) printf("DDT walk unique, building age histogram...\n");
2155 ddt_prune_walk(spa, 0, &histogram);
2156
2157 /*
2158 * print out histogram for unique entry class birth
2159 */
2160 if (histogram.dah_entries > 0) {
2161 (void) printf("%5s %9s %4s\n",
2162 "age", "blocks", "amnt");
2163 (void) printf("%5s %9s %4s\n",
2164 "-----", "---------", "----");
2165 for (int i = 0; i < HIST_BINS; i++) {
2166 (void) printf("%5d %9d %4d%%\n", 1 << i,
2167 (int)histogram.dah_age_histo[i],
2168 (int)((histogram.dah_age_histo[i] * 100) /
2169 histogram.dah_entries));
2170 }
2171 }
2172 }
2173 }
2174
2175 static void
dump_brt(spa_t * spa)2176 dump_brt(spa_t *spa)
2177 {
2178 if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) {
2179 printf("BRT: unsupported on this pool\n");
2180 return;
2181 }
2182
2183 if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
2184 printf("BRT: empty\n");
2185 return;
2186 }
2187
2188 char count[32], used[32], saved[32];
2189 zdb_nicebytes(brt_get_used(spa), used, sizeof (used));
2190 zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved));
2191 uint64_t ratio = brt_get_ratio(spa);
2192 printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved,
2193 (u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100));
2194
2195 if (dump_opt['T'] < 2)
2196 return;
2197
2198 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2199 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2200 if (!brtvd->bv_initiated) {
2201 printf("BRT: vdev %" PRIu64 ": empty\n", vdevid);
2202 continue;
2203 }
2204
2205 zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count));
2206 zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used));
2207 zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved));
2208 printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n",
2209 vdevid, count, used, saved);
2210 }
2211
2212 if (dump_opt['T'] < 3)
2213 return;
2214
2215 /* -TTT shows a per-vdev histograms; -TTTT shows all entries */
2216 boolean_t do_histo = dump_opt['T'] == 3;
2217
2218 char dva[64];
2219
2220 if (!do_histo)
2221 printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2222
2223 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2224 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2225 if (!brtvd->bv_initiated)
2226 continue;
2227
2228 uint64_t counts[64] = {};
2229
2230 zap_cursor_t zc;
2231 zap_attribute_t *za = zap_attribute_alloc();
2232 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2233 brtvd->bv_mos_entries);
2234 zap_cursor_retrieve(&zc, za) == 0;
2235 zap_cursor_advance(&zc)) {
2236 uint64_t refcnt;
2237 VERIFY0(zap_lookup_uint64(spa->spa_meta_objset,
2238 brtvd->bv_mos_entries,
2239 (const uint64_t *)za->za_name, 1,
2240 za->za_integer_length, za->za_num_integers,
2241 &refcnt));
2242
2243 if (do_histo)
2244 counts[highbit64(refcnt)]++;
2245 else {
2246 uint64_t offset =
2247 *(const uint64_t *)za->za_name;
2248
2249 snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx",
2250 vdevid, (u_longlong_t)offset);
2251 printf("%-16s %-10llu\n", dva,
2252 (u_longlong_t)refcnt);
2253 }
2254 }
2255 zap_cursor_fini(&zc);
2256 zap_attribute_free(za);
2257
2258 if (do_histo) {
2259 printf("\nBRT: vdev %" PRIu64
2260 ": DVAs with 2^n refcnts:\n", vdevid);
2261 dump_histogram(counts, 64, 0);
2262 }
2263 }
2264 }
2265
2266 static void
dump_dtl_seg(void * arg,uint64_t start,uint64_t size)2267 dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
2268 {
2269 char *prefix = arg;
2270
2271 (void) printf("%s [%llu,%llu) length %llu\n",
2272 prefix,
2273 (u_longlong_t)start,
2274 (u_longlong_t)(start + size),
2275 (u_longlong_t)(size));
2276 }
2277
2278 static void
dump_dtl(vdev_t * vd,int indent)2279 dump_dtl(vdev_t *vd, int indent)
2280 {
2281 spa_t *spa = vd->vdev_spa;
2282 boolean_t required;
2283 const char *name[DTL_TYPES] = { "missing", "partial", "scrub",
2284 "outage" };
2285 char prefix[256];
2286
2287 spa_vdev_state_enter(spa, SCL_NONE);
2288 required = vdev_dtl_required(vd);
2289 (void) spa_vdev_state_exit(spa, NULL, 0);
2290
2291 if (indent == 0)
2292 (void) printf("\nDirty time logs:\n\n");
2293
2294 (void) printf("\t%*s%s [%s]\n", indent, "",
2295 vd->vdev_path ? vd->vdev_path :
2296 vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa),
2297 required ? "DTL-required" : "DTL-expendable");
2298
2299 for (int t = 0; t < DTL_TYPES; t++) {
2300 zfs_range_tree_t *rt = vd->vdev_dtl[t];
2301 if (zfs_range_tree_space(rt) == 0)
2302 continue;
2303 (void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
2304 indent + 2, "", name[t]);
2305 zfs_range_tree_walk(rt, dump_dtl_seg, prefix);
2306 if (dump_opt['d'] > 5 && vd->vdev_children == 0)
2307 dump_spacemap(spa->spa_meta_objset,
2308 vd->vdev_dtl_sm);
2309 }
2310
2311 for (unsigned c = 0; c < vd->vdev_children; c++)
2312 dump_dtl(vd->vdev_child[c], indent + 4);
2313 }
2314
2315 static void
dump_history(spa_t * spa)2316 dump_history(spa_t *spa)
2317 {
2318 nvlist_t **events = NULL;
2319 char *buf;
2320 uint64_t resid, len, off = 0;
2321 uint_t num = 0;
2322 int error;
2323 char tbuf[30];
2324
2325 if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) {
2326 (void) fprintf(stderr, "%s: unable to allocate I/O buffer\n",
2327 __func__);
2328 return;
2329 }
2330
2331 do {
2332 len = SPA_OLD_MAXBLOCKSIZE;
2333
2334 if ((error = spa_history_get(spa, &off, &len, buf)) != 0) {
2335 (void) fprintf(stderr, "Unable to read history: "
2336 "error %d\n", error);
2337 free(buf);
2338 return;
2339 }
2340
2341 if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0)
2342 break;
2343
2344 off -= resid;
2345 } while (len != 0);
2346
2347 (void) printf("\nHistory:\n");
2348 for (unsigned i = 0; i < num; i++) {
2349 boolean_t printed = B_FALSE;
2350
2351 if (nvlist_exists(events[i], ZPOOL_HIST_TIME)) {
2352 time_t tsec;
2353 struct tm t;
2354
2355 tsec = fnvlist_lookup_uint64(events[i],
2356 ZPOOL_HIST_TIME);
2357 (void) localtime_r(&tsec, &t);
2358 (void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t);
2359 } else {
2360 tbuf[0] = '\0';
2361 }
2362
2363 if (nvlist_exists(events[i], ZPOOL_HIST_CMD)) {
2364 (void) printf("%s %s\n", tbuf,
2365 fnvlist_lookup_string(events[i], ZPOOL_HIST_CMD));
2366 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_EVENT)) {
2367 uint64_t ievent;
2368
2369 ievent = fnvlist_lookup_uint64(events[i],
2370 ZPOOL_HIST_INT_EVENT);
2371 if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS)
2372 goto next;
2373
2374 (void) printf(" %s [internal %s txg:%ju] %s\n",
2375 tbuf,
2376 zfs_history_event_names[ievent],
2377 fnvlist_lookup_uint64(events[i],
2378 ZPOOL_HIST_TXG),
2379 fnvlist_lookup_string(events[i],
2380 ZPOOL_HIST_INT_STR));
2381 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_NAME)) {
2382 (void) printf("%s [txg:%ju] %s", tbuf,
2383 fnvlist_lookup_uint64(events[i],
2384 ZPOOL_HIST_TXG),
2385 fnvlist_lookup_string(events[i],
2386 ZPOOL_HIST_INT_NAME));
2387
2388 if (nvlist_exists(events[i], ZPOOL_HIST_DSNAME)) {
2389 (void) printf(" %s (%llu)",
2390 fnvlist_lookup_string(events[i],
2391 ZPOOL_HIST_DSNAME),
2392 (u_longlong_t)fnvlist_lookup_uint64(
2393 events[i],
2394 ZPOOL_HIST_DSID));
2395 }
2396
2397 (void) printf(" %s\n", fnvlist_lookup_string(events[i],
2398 ZPOOL_HIST_INT_STR));
2399 } else if (nvlist_exists(events[i], ZPOOL_HIST_IOCTL)) {
2400 (void) printf("%s ioctl %s\n", tbuf,
2401 fnvlist_lookup_string(events[i],
2402 ZPOOL_HIST_IOCTL));
2403
2404 if (nvlist_exists(events[i], ZPOOL_HIST_INPUT_NVL)) {
2405 (void) printf(" input:\n");
2406 dump_nvlist(fnvlist_lookup_nvlist(events[i],
2407 ZPOOL_HIST_INPUT_NVL), 8);
2408 }
2409 if (nvlist_exists(events[i], ZPOOL_HIST_OUTPUT_NVL)) {
2410 (void) printf(" output:\n");
2411 dump_nvlist(fnvlist_lookup_nvlist(events[i],
2412 ZPOOL_HIST_OUTPUT_NVL), 8);
2413 }
2414 if (nvlist_exists(events[i], ZPOOL_HIST_ERRNO)) {
2415 (void) printf(" errno: %lld\n",
2416 (longlong_t)fnvlist_lookup_int64(events[i],
2417 ZPOOL_HIST_ERRNO));
2418 }
2419 } else {
2420 goto next;
2421 }
2422
2423 printed = B_TRUE;
2424 next:
2425 if (dump_opt['h'] > 1) {
2426 if (!printed)
2427 (void) printf("unrecognized record:\n");
2428 dump_nvlist(events[i], 2);
2429 }
2430 }
2431 free(buf);
2432 }
2433
2434 static void
dump_dnode(objset_t * os,uint64_t object,void * data,size_t size)2435 dump_dnode(objset_t *os, uint64_t object, void *data, size_t size)
2436 {
2437 (void) os, (void) object, (void) data, (void) size;
2438 }
2439
2440 static uint64_t
blkid2offset(const dnode_phys_t * dnp,const blkptr_t * bp,const zbookmark_phys_t * zb)2441 blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
2442 const zbookmark_phys_t *zb)
2443 {
2444 if (dnp == NULL) {
2445 ASSERT(zb->zb_level < 0);
2446 if (zb->zb_object == 0)
2447 return (zb->zb_blkid);
2448 return (zb->zb_blkid * BP_GET_LSIZE(bp));
2449 }
2450
2451 ASSERT(zb->zb_level >= 0);
2452
2453 return ((zb->zb_blkid <<
2454 (zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) *
2455 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
2456 }
2457
2458 static void
snprintf_zstd_header(spa_t * spa,char * blkbuf,size_t buflen,const blkptr_t * bp)2459 snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
2460 const blkptr_t *bp)
2461 {
2462 static abd_t *pabd = NULL;
2463 void *buf;
2464 zio_t *zio;
2465 zfs_zstdhdr_t zstd_hdr;
2466 int error;
2467
2468 if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD)
2469 return;
2470
2471 if (BP_IS_HOLE(bp))
2472 return;
2473
2474 if (BP_IS_EMBEDDED(bp)) {
2475 buf = malloc(SPA_MAXBLOCKSIZE);
2476 if (buf == NULL) {
2477 (void) fprintf(stderr, "out of memory\n");
2478 zdb_exit(1);
2479 }
2480 decode_embedded_bp_compressed(bp, buf);
2481 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2482 free(buf);
2483 zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2484 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2485 (void) snprintf(blkbuf + strlen(blkbuf),
2486 buflen - strlen(blkbuf),
2487 " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2488 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2489 zfs_get_hdrlevel(&zstd_hdr));
2490 return;
2491 }
2492
2493 if (!pabd)
2494 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
2495 zio = zio_root(spa, NULL, NULL, 0);
2496
2497 /* Decrypt but don't decompress so we can read the compression header */
2498 zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL,
2499 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS,
2500 NULL));
2501 error = zio_wait(zio);
2502 if (error) {
2503 (void) fprintf(stderr, "read failed: %d\n", error);
2504 return;
2505 }
2506 buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp));
2507 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2508 zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2509 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2510
2511 (void) snprintf(blkbuf + strlen(blkbuf),
2512 buflen - strlen(blkbuf),
2513 " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2514 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2515 zfs_get_hdrlevel(&zstd_hdr));
2516
2517 abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp));
2518 }
2519
2520 static void
snprintf_blkptr_compact(char * blkbuf,size_t buflen,const blkptr_t * bp,boolean_t bp_freed)2521 snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
2522 boolean_t bp_freed)
2523 {
2524 const dva_t *dva = bp->blk_dva;
2525 int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1;
2526 int i;
2527
2528 if (dump_opt['b'] >= 6) {
2529 snprintf_blkptr(blkbuf, buflen, bp);
2530 if (bp_freed) {
2531 (void) snprintf(blkbuf + strlen(blkbuf),
2532 buflen - strlen(blkbuf), " %s", "FREE");
2533 }
2534 return;
2535 }
2536
2537 if (BP_IS_EMBEDDED(bp)) {
2538 (void) sprintf(blkbuf,
2539 "EMBEDDED et=%u %llxL/%llxP B=%llu",
2540 (int)BPE_GET_ETYPE(bp),
2541 (u_longlong_t)BPE_GET_LSIZE(bp),
2542 (u_longlong_t)BPE_GET_PSIZE(bp),
2543 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2544 return;
2545 }
2546
2547 blkbuf[0] = '\0';
2548
2549 for (i = 0; i < ndvas; i++) {
2550 (void) snprintf(blkbuf + strlen(blkbuf),
2551 buflen - strlen(blkbuf), "%llu:%llx:%llx%s ",
2552 (u_longlong_t)DVA_GET_VDEV(&dva[i]),
2553 (u_longlong_t)DVA_GET_OFFSET(&dva[i]),
2554 (u_longlong_t)DVA_GET_ASIZE(&dva[i]),
2555 (DVA_GET_GANG(&dva[i]) ? "G" : ""));
2556 }
2557
2558 if (BP_IS_HOLE(bp)) {
2559 (void) snprintf(blkbuf + strlen(blkbuf),
2560 buflen - strlen(blkbuf),
2561 "%llxL B=%llu",
2562 (u_longlong_t)BP_GET_LSIZE(bp),
2563 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2564 } else {
2565 (void) snprintf(blkbuf + strlen(blkbuf),
2566 buflen - strlen(blkbuf),
2567 "%llxL/%llxP F=%llu B=%llu/%llu",
2568 (u_longlong_t)BP_GET_LSIZE(bp),
2569 (u_longlong_t)BP_GET_PSIZE(bp),
2570 (u_longlong_t)BP_GET_FILL(bp),
2571 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),
2572 (u_longlong_t)BP_GET_PHYSICAL_BIRTH(bp));
2573 if (bp_freed)
2574 (void) snprintf(blkbuf + strlen(blkbuf),
2575 buflen - strlen(blkbuf), " %s", "FREE");
2576 (void) snprintf(blkbuf + strlen(blkbuf),
2577 buflen - strlen(blkbuf),
2578 " cksum=%016llx:%016llx:%016llx:%016llx",
2579 (u_longlong_t)bp->blk_cksum.zc_word[0],
2580 (u_longlong_t)bp->blk_cksum.zc_word[1],
2581 (u_longlong_t)bp->blk_cksum.zc_word[2],
2582 (u_longlong_t)bp->blk_cksum.zc_word[3]);
2583 }
2584 }
2585
2586 static void
print_indirect(spa_t * spa,blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp)2587 print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
2588 const dnode_phys_t *dnp)
2589 {
2590 char blkbuf[BP_SPRINTF_LEN];
2591 int l;
2592
2593 if (!BP_IS_EMBEDDED(bp)) {
2594 ASSERT3U(BP_GET_TYPE(bp), ==, dnp->dn_type);
2595 ASSERT3U(BP_GET_LEVEL(bp), ==, zb->zb_level);
2596 }
2597
2598 (void) printf("%16llx ", (u_longlong_t)blkid2offset(dnp, bp, zb));
2599
2600 ASSERT(zb->zb_level >= 0);
2601
2602 for (l = dnp->dn_nlevels - 1; l >= -1; l--) {
2603 if (l == zb->zb_level) {
2604 (void) printf("L%llx", (u_longlong_t)zb->zb_level);
2605 } else {
2606 (void) printf(" ");
2607 }
2608 }
2609
2610 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE);
2611 if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD)
2612 snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp);
2613 (void) printf("%s\n", blkbuf);
2614 }
2615
2616 static int
visit_indirect(spa_t * spa,const dnode_phys_t * dnp,blkptr_t * bp,const zbookmark_phys_t * zb)2617 visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
2618 blkptr_t *bp, const zbookmark_phys_t *zb)
2619 {
2620 int err = 0;
2621
2622 if (BP_GET_BIRTH(bp) == 0)
2623 return (0);
2624
2625 print_indirect(spa, bp, zb, dnp);
2626
2627 if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
2628 arc_flags_t flags = ARC_FLAG_WAIT;
2629 int i;
2630 blkptr_t *cbp;
2631 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
2632 arc_buf_t *buf;
2633 uint64_t fill = 0;
2634 ASSERT(!BP_IS_REDACTED(bp));
2635
2636 err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
2637 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
2638 if (err)
2639 return (err);
2640 ASSERT(buf->b_data);
2641
2642 /* recursively visit blocks below this */
2643 cbp = buf->b_data;
2644 for (i = 0; i < epb; i++, cbp++) {
2645 zbookmark_phys_t czb;
2646
2647 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
2648 zb->zb_level - 1,
2649 zb->zb_blkid * epb + i);
2650 err = visit_indirect(spa, dnp, cbp, &czb);
2651 if (err)
2652 break;
2653 fill += BP_GET_FILL(cbp);
2654 }
2655 if (!err)
2656 ASSERT3U(fill, ==, BP_GET_FILL(bp));
2657 arc_buf_destroy(buf, &buf);
2658 }
2659
2660 return (err);
2661 }
2662
2663 static void
dump_indirect(dnode_t * dn)2664 dump_indirect(dnode_t *dn)
2665 {
2666 dnode_phys_t *dnp = dn->dn_phys;
2667 zbookmark_phys_t czb;
2668
2669 (void) printf("Indirect blocks:\n");
2670
2671 SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
2672 dn->dn_object, dnp->dn_nlevels - 1, 0);
2673 for (int j = 0; j < dnp->dn_nblkptr; j++) {
2674 czb.zb_blkid = j;
2675 (void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
2676 &dnp->dn_blkptr[j], &czb);
2677 }
2678
2679 (void) printf("\n");
2680 }
2681
2682 static void
dump_dsl_dir(objset_t * os,uint64_t object,void * data,size_t size)2683 dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size)
2684 {
2685 (void) os, (void) object;
2686 dsl_dir_phys_t *dd = data;
2687 time_t crtime;
2688 char nice[32];
2689
2690 /* make sure nicenum has enough space */
2691 _Static_assert(sizeof (nice) >= NN_NUMBUF_SZ, "nice truncated");
2692
2693 if (dd == NULL)
2694 return;
2695
2696 ASSERT3U(size, >=, sizeof (dsl_dir_phys_t));
2697
2698 crtime = dd->dd_creation_time;
2699 (void) printf("\t\tcreation_time = %s", ctime(&crtime));
2700 (void) printf("\t\thead_dataset_obj = %llu\n",
2701 (u_longlong_t)dd->dd_head_dataset_obj);
2702 (void) printf("\t\tparent_dir_obj = %llu\n",
2703 (u_longlong_t)dd->dd_parent_obj);
2704 (void) printf("\t\torigin_obj = %llu\n",
2705 (u_longlong_t)dd->dd_origin_obj);
2706 (void) printf("\t\tchild_dir_zapobj = %llu\n",
2707 (u_longlong_t)dd->dd_child_dir_zapobj);
2708 zdb_nicenum(dd->dd_used_bytes, nice, sizeof (nice));
2709 (void) printf("\t\tused_bytes = %s\n", nice);
2710 zdb_nicenum(dd->dd_compressed_bytes, nice, sizeof (nice));
2711 (void) printf("\t\tcompressed_bytes = %s\n", nice);
2712 zdb_nicenum(dd->dd_uncompressed_bytes, nice, sizeof (nice));
2713 (void) printf("\t\tuncompressed_bytes = %s\n", nice);
2714 zdb_nicenum(dd->dd_quota, nice, sizeof (nice));
2715 (void) printf("\t\tquota = %s\n", nice);
2716 zdb_nicenum(dd->dd_reserved, nice, sizeof (nice));
2717 (void) printf("\t\treserved = %s\n", nice);
2718 (void) printf("\t\tprops_zapobj = %llu\n",
2719 (u_longlong_t)dd->dd_props_zapobj);
2720 (void) printf("\t\tdeleg_zapobj = %llu\n",
2721 (u_longlong_t)dd->dd_deleg_zapobj);
2722 (void) printf("\t\tflags = %llx\n",
2723 (u_longlong_t)dd->dd_flags);
2724
2725 #define DO(which) \
2726 zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2727 sizeof (nice)); \
2728 (void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2729 DO(HEAD);
2730 DO(SNAP);
2731 DO(CHILD);
2732 DO(CHILD_RSRV);
2733 DO(REFRSRV);
2734 #undef DO
2735 (void) printf("\t\tclones = %llu\n",
2736 (u_longlong_t)dd->dd_clones);
2737 }
2738
2739 static void
dump_dsl_dataset(objset_t * os,uint64_t object,void * data,size_t size)2740 dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
2741 {
2742 (void) os, (void) object;
2743 dsl_dataset_phys_t *ds = data;
2744 time_t crtime;
2745 char used[32], compressed[32], uncompressed[32], unique[32];
2746 char blkbuf[BP_SPRINTF_LEN];
2747
2748 /* make sure nicenum has enough space */
2749 _Static_assert(sizeof (used) >= NN_NUMBUF_SZ, "used truncated");
2750 _Static_assert(sizeof (compressed) >= NN_NUMBUF_SZ,
2751 "compressed truncated");
2752 _Static_assert(sizeof (uncompressed) >= NN_NUMBUF_SZ,
2753 "uncompressed truncated");
2754 _Static_assert(sizeof (unique) >= NN_NUMBUF_SZ, "unique truncated");
2755
2756 if (ds == NULL)
2757 return;
2758
2759 ASSERT(size == sizeof (*ds));
2760 crtime = ds->ds_creation_time;
2761 zdb_nicenum(ds->ds_referenced_bytes, used, sizeof (used));
2762 zdb_nicenum(ds->ds_compressed_bytes, compressed, sizeof (compressed));
2763 zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed,
2764 sizeof (uncompressed));
2765 zdb_nicenum(ds->ds_unique_bytes, unique, sizeof (unique));
2766 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp);
2767
2768 (void) printf("\t\tdir_obj = %llu\n",
2769 (u_longlong_t)ds->ds_dir_obj);
2770 (void) printf("\t\tprev_snap_obj = %llu\n",
2771 (u_longlong_t)ds->ds_prev_snap_obj);
2772 (void) printf("\t\tprev_snap_txg = %llu\n",
2773 (u_longlong_t)ds->ds_prev_snap_txg);
2774 (void) printf("\t\tnext_snap_obj = %llu\n",
2775 (u_longlong_t)ds->ds_next_snap_obj);
2776 (void) printf("\t\tsnapnames_zapobj = %llu\n",
2777 (u_longlong_t)ds->ds_snapnames_zapobj);
2778 (void) printf("\t\tnum_children = %llu\n",
2779 (u_longlong_t)ds->ds_num_children);
2780 (void) printf("\t\tuserrefs_obj = %llu\n",
2781 (u_longlong_t)ds->ds_userrefs_obj);
2782 (void) printf("\t\tcreation_time = %s", ctime(&crtime));
2783 (void) printf("\t\tcreation_txg = %llu\n",
2784 (u_longlong_t)ds->ds_creation_txg);
2785 (void) printf("\t\tdeadlist_obj = %llu\n",
2786 (u_longlong_t)ds->ds_deadlist_obj);
2787 (void) printf("\t\tused_bytes = %s\n", used);
2788 (void) printf("\t\tcompressed_bytes = %s\n", compressed);
2789 (void) printf("\t\tuncompressed_bytes = %s\n", uncompressed);
2790 (void) printf("\t\tunique = %s\n", unique);
2791 (void) printf("\t\tfsid_guid = %llu\n",
2792 (u_longlong_t)ds->ds_fsid_guid);
2793 (void) printf("\t\tguid = %llu\n",
2794 (u_longlong_t)ds->ds_guid);
2795 (void) printf("\t\tflags = %llx\n",
2796 (u_longlong_t)ds->ds_flags);
2797 (void) printf("\t\tnext_clones_obj = %llu\n",
2798 (u_longlong_t)ds->ds_next_clones_obj);
2799 (void) printf("\t\tprops_obj = %llu\n",
2800 (u_longlong_t)ds->ds_props_obj);
2801 (void) printf("\t\tbp = %s\n", blkbuf);
2802 }
2803
2804 static int
dump_bptree_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)2805 dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2806 {
2807 (void) arg, (void) tx;
2808 char blkbuf[BP_SPRINTF_LEN];
2809
2810 if (BP_GET_BIRTH(bp) != 0) {
2811 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
2812 (void) printf("\t%s\n", blkbuf);
2813 }
2814 return (0);
2815 }
2816
2817 static void
dump_bptree(objset_t * os,uint64_t obj,const char * name)2818 dump_bptree(objset_t *os, uint64_t obj, const char *name)
2819 {
2820 char bytes[32];
2821 bptree_phys_t *bt;
2822 dmu_buf_t *db;
2823
2824 /* make sure nicenum has enough space */
2825 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2826
2827 if (dump_opt['d'] < 3)
2828 return;
2829
2830 VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
2831 bt = db->db_data;
2832 zdb_nicenum(bt->bt_bytes, bytes, sizeof (bytes));
2833 (void) printf("\n %s: %llu datasets, %s\n",
2834 name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes);
2835 dmu_buf_rele(db, FTAG);
2836
2837 if (dump_opt['d'] < 5)
2838 return;
2839
2840 (void) printf("\n");
2841
2842 (void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL);
2843 }
2844
2845 static int
dump_bpobj_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)2846 dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
2847 {
2848 (void) arg, (void) tx;
2849 char blkbuf[BP_SPRINTF_LEN];
2850
2851 ASSERT(BP_GET_BIRTH(bp) != 0);
2852 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed);
2853 (void) printf("\t%s\n", blkbuf);
2854 return (0);
2855 }
2856
2857 static void
dump_full_bpobj(bpobj_t * bpo,const char * name,int indent)2858 dump_full_bpobj(bpobj_t *bpo, const char *name, int indent)
2859 {
2860 char bytes[32];
2861 char comp[32];
2862 char uncomp[32];
2863 uint64_t i;
2864
2865 /* make sure nicenum has enough space */
2866 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2867 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
2868 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
2869
2870 if (dump_opt['d'] < 3)
2871 return;
2872
2873 zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes, sizeof (bytes));
2874 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
2875 zdb_nicenum(bpo->bpo_phys->bpo_comp, comp, sizeof (comp));
2876 zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp, sizeof (uncomp));
2877 if (bpo->bpo_havefreed) {
2878 (void) printf(" %*s: object %llu, %llu local "
2879 "blkptrs, %llu freed, %llu subobjs in object %llu, "
2880 "%s (%s/%s comp)\n",
2881 indent * 8, name,
2882 (u_longlong_t)bpo->bpo_object,
2883 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2884 (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2885 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2886 (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2887 bytes, comp, uncomp);
2888 } else {
2889 (void) printf(" %*s: object %llu, %llu local "
2890 "blkptrs, %llu subobjs in object %llu, "
2891 "%s (%s/%s comp)\n",
2892 indent * 8, name,
2893 (u_longlong_t)bpo->bpo_object,
2894 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2895 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2896 (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2897 bytes, comp, uncomp);
2898 }
2899
2900 for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
2901 uint64_t subobj;
2902 bpobj_t subbpo;
2903 int error;
2904 VERIFY0(dmu_read(bpo->bpo_os,
2905 bpo->bpo_phys->bpo_subobjs,
2906 i * sizeof (subobj), sizeof (subobj), &subobj, 0));
2907 error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
2908 if (error != 0) {
2909 (void) printf("ERROR %u while trying to open "
2910 "subobj id %llu\n",
2911 error, (u_longlong_t)subobj);
2912 continue;
2913 }
2914 dump_full_bpobj(&subbpo, "subobj", indent + 1);
2915 bpobj_close(&subbpo);
2916 }
2917 } else {
2918 if (bpo->bpo_havefreed) {
2919 (void) printf(" %*s: object %llu, %llu blkptrs, "
2920 "%llu freed, %s\n",
2921 indent * 8, name,
2922 (u_longlong_t)bpo->bpo_object,
2923 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2924 (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2925 bytes);
2926 } else {
2927 (void) printf(" %*s: object %llu, %llu blkptrs, "
2928 "%s\n",
2929 indent * 8, name,
2930 (u_longlong_t)bpo->bpo_object,
2931 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2932 bytes);
2933 }
2934 }
2935
2936 if (dump_opt['d'] < 5)
2937 return;
2938
2939
2940 if (indent == 0) {
2941 (void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL);
2942 (void) printf("\n");
2943 }
2944 }
2945
2946 static int
dump_bookmark(dsl_pool_t * dp,char * name,boolean_t print_redact,boolean_t print_list)2947 dump_bookmark(dsl_pool_t *dp, char *name, boolean_t print_redact,
2948 boolean_t print_list)
2949 {
2950 int err = 0;
2951 zfs_bookmark_phys_t prop;
2952 objset_t *mos = dp->dp_spa->spa_meta_objset;
2953 err = dsl_bookmark_lookup(dp, name, NULL, &prop);
2954
2955 if (err != 0) {
2956 return (err);
2957 }
2958
2959 (void) printf("\t#%s: ", strchr(name, '#') + 1);
2960 (void) printf("{guid: %llx creation_txg: %llu creation_time: "
2961 "%llu redaction_obj: %llu}\n", (u_longlong_t)prop.zbm_guid,
2962 (u_longlong_t)prop.zbm_creation_txg,
2963 (u_longlong_t)prop.zbm_creation_time,
2964 (u_longlong_t)prop.zbm_redaction_obj);
2965
2966 IMPLY(print_list, print_redact);
2967 if (!print_redact || prop.zbm_redaction_obj == 0)
2968 return (0);
2969
2970 redaction_list_t *rl;
2971 VERIFY0(dsl_redaction_list_hold_obj(dp,
2972 prop.zbm_redaction_obj, FTAG, &rl));
2973
2974 redaction_list_phys_t *rlp = rl->rl_phys;
2975 (void) printf("\tRedacted:\n\t\tProgress: ");
2976 if (rlp->rlp_last_object != UINT64_MAX ||
2977 rlp->rlp_last_blkid != UINT64_MAX) {
2978 (void) printf("%llu %llu (incomplete)\n",
2979 (u_longlong_t)rlp->rlp_last_object,
2980 (u_longlong_t)rlp->rlp_last_blkid);
2981 } else {
2982 (void) printf("complete\n");
2983 }
2984 (void) printf("\t\tSnapshots: [");
2985 for (unsigned int i = 0; i < rlp->rlp_num_snaps; i++) {
2986 if (i > 0)
2987 (void) printf(", ");
2988 (void) printf("%0llu",
2989 (u_longlong_t)rlp->rlp_snaps[i]);
2990 }
2991 (void) printf("]\n\t\tLength: %llu\n",
2992 (u_longlong_t)rlp->rlp_num_entries);
2993
2994 if (!print_list) {
2995 dsl_redaction_list_rele(rl, FTAG);
2996 return (0);
2997 }
2998
2999 if (rlp->rlp_num_entries == 0) {
3000 dsl_redaction_list_rele(rl, FTAG);
3001 (void) printf("\t\tRedaction List: []\n\n");
3002 return (0);
3003 }
3004
3005 redact_block_phys_t *rbp_buf;
3006 uint64_t size;
3007 dmu_object_info_t doi;
3008
3009 VERIFY0(dmu_object_info(mos, prop.zbm_redaction_obj, &doi));
3010 size = doi.doi_max_offset;
3011 rbp_buf = kmem_alloc(size, KM_SLEEP);
3012
3013 err = dmu_read(mos, prop.zbm_redaction_obj, 0, size,
3014 rbp_buf, 0);
3015 if (err != 0) {
3016 dsl_redaction_list_rele(rl, FTAG);
3017 kmem_free(rbp_buf, size);
3018 return (err);
3019 }
3020
3021 (void) printf("\t\tRedaction List: [{object: %llx, offset: "
3022 "%llx, blksz: %x, count: %llx}",
3023 (u_longlong_t)rbp_buf[0].rbp_object,
3024 (u_longlong_t)rbp_buf[0].rbp_blkid,
3025 (uint_t)(redact_block_get_size(&rbp_buf[0])),
3026 (u_longlong_t)redact_block_get_count(&rbp_buf[0]));
3027
3028 for (size_t i = 1; i < rlp->rlp_num_entries; i++) {
3029 (void) printf(",\n\t\t{object: %llx, offset: %llx, "
3030 "blksz: %x, count: %llx}",
3031 (u_longlong_t)rbp_buf[i].rbp_object,
3032 (u_longlong_t)rbp_buf[i].rbp_blkid,
3033 (uint_t)(redact_block_get_size(&rbp_buf[i])),
3034 (u_longlong_t)redact_block_get_count(&rbp_buf[i]));
3035 }
3036 dsl_redaction_list_rele(rl, FTAG);
3037 kmem_free(rbp_buf, size);
3038 (void) printf("]\n\n");
3039 return (0);
3040 }
3041
3042 static void
dump_bookmarks(objset_t * os,int verbosity)3043 dump_bookmarks(objset_t *os, int verbosity)
3044 {
3045 zap_cursor_t zc;
3046 zap_attribute_t *attrp;
3047 dsl_dataset_t *ds = dmu_objset_ds(os);
3048 dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3049 objset_t *mos = os->os_spa->spa_meta_objset;
3050 if (verbosity < 4)
3051 return;
3052 attrp = zap_attribute_alloc();
3053 dsl_pool_config_enter(dp, FTAG);
3054
3055 for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj);
3056 zap_cursor_retrieve(&zc, attrp) == 0;
3057 zap_cursor_advance(&zc)) {
3058 char osname[ZFS_MAX_DATASET_NAME_LEN];
3059 char buf[ZFS_MAX_DATASET_NAME_LEN];
3060 int len;
3061 dmu_objset_name(os, osname);
3062 len = snprintf(buf, sizeof (buf), "%s#%s", osname,
3063 attrp->za_name);
3064 VERIFY3S(len, <, ZFS_MAX_DATASET_NAME_LEN);
3065 (void) dump_bookmark(dp, buf, verbosity >= 5, verbosity >= 6);
3066 }
3067 zap_cursor_fini(&zc);
3068 dsl_pool_config_exit(dp, FTAG);
3069 zap_attribute_free(attrp);
3070 }
3071
3072 static void
bpobj_count_refd(bpobj_t * bpo)3073 bpobj_count_refd(bpobj_t *bpo)
3074 {
3075 mos_obj_refd(bpo->bpo_object);
3076
3077 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
3078 mos_obj_refd(bpo->bpo_phys->bpo_subobjs);
3079 for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
3080 uint64_t subobj;
3081 bpobj_t subbpo;
3082 int error;
3083 VERIFY0(dmu_read(bpo->bpo_os,
3084 bpo->bpo_phys->bpo_subobjs,
3085 i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3086 error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3087 if (error != 0) {
3088 (void) printf("ERROR %u while trying to open "
3089 "subobj id %llu\n",
3090 error, (u_longlong_t)subobj);
3091 continue;
3092 }
3093 bpobj_count_refd(&subbpo);
3094 bpobj_close(&subbpo);
3095 }
3096 }
3097 }
3098
3099 static int
dsl_deadlist_entry_count_refd(void * arg,dsl_deadlist_entry_t * dle)3100 dsl_deadlist_entry_count_refd(void *arg, dsl_deadlist_entry_t *dle)
3101 {
3102 spa_t *spa = arg;
3103 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3104 if (dle->dle_bpobj.bpo_object != empty_bpobj)
3105 bpobj_count_refd(&dle->dle_bpobj);
3106 return (0);
3107 }
3108
3109 static int
dsl_deadlist_entry_dump(void * arg,dsl_deadlist_entry_t * dle)3110 dsl_deadlist_entry_dump(void *arg, dsl_deadlist_entry_t *dle)
3111 {
3112 ASSERT(arg == NULL);
3113 if (dump_opt['d'] >= 5) {
3114 char buf[128];
3115 (void) snprintf(buf, sizeof (buf),
3116 "mintxg %llu -> obj %llu",
3117 (longlong_t)dle->dle_mintxg,
3118 (longlong_t)dle->dle_bpobj.bpo_object);
3119
3120 dump_full_bpobj(&dle->dle_bpobj, buf, 0);
3121 } else {
3122 (void) printf("mintxg %llu -> obj %llu\n",
3123 (longlong_t)dle->dle_mintxg,
3124 (longlong_t)dle->dle_bpobj.bpo_object);
3125 }
3126 return (0);
3127 }
3128
3129 static void
dump_blkptr_list(dsl_deadlist_t * dl,const char * name)3130 dump_blkptr_list(dsl_deadlist_t *dl, const char *name)
3131 {
3132 char bytes[32];
3133 char comp[32];
3134 char uncomp[32];
3135 char entries[32];
3136 spa_t *spa = dmu_objset_spa(dl->dl_os);
3137 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3138
3139 if (dl->dl_oldfmt) {
3140 if (dl->dl_bpobj.bpo_object != empty_bpobj)
3141 bpobj_count_refd(&dl->dl_bpobj);
3142 } else {
3143 mos_obj_refd(dl->dl_object);
3144 dsl_deadlist_iterate(dl, dsl_deadlist_entry_count_refd, spa);
3145 }
3146
3147 /* make sure nicenum has enough space */
3148 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3149 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
3150 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
3151 _Static_assert(sizeof (entries) >= NN_NUMBUF_SZ, "entries truncated");
3152
3153 if (dump_opt['d'] < 3)
3154 return;
3155
3156 if (dl->dl_oldfmt) {
3157 dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0);
3158 return;
3159 }
3160
3161 zdb_nicenum(dl->dl_phys->dl_used, bytes, sizeof (bytes));
3162 zdb_nicenum(dl->dl_phys->dl_comp, comp, sizeof (comp));
3163 zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp, sizeof (uncomp));
3164 zdb_nicenum(avl_numnodes(&dl->dl_tree), entries, sizeof (entries));
3165 (void) printf("\n %s: %s (%s/%s comp), %s entries\n",
3166 name, bytes, comp, uncomp, entries);
3167
3168 if (dump_opt['d'] < 4)
3169 return;
3170
3171 (void) putchar('\n');
3172
3173 dsl_deadlist_iterate(dl, dsl_deadlist_entry_dump, NULL);
3174 }
3175
3176 static int
verify_dd_livelist(objset_t * os)3177 verify_dd_livelist(objset_t *os)
3178 {
3179 uint64_t ll_used, used, ll_comp, comp, ll_uncomp, uncomp;
3180 dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3181 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
3182
3183 ASSERT(!dmu_objset_is_snapshot(os));
3184 if (!dsl_deadlist_is_open(&dd->dd_livelist))
3185 return (0);
3186
3187 /* Iterate through the livelist to check for duplicates */
3188 dsl_deadlist_iterate(&dd->dd_livelist, sublivelist_verify_lightweight,
3189 NULL);
3190
3191 dsl_pool_config_enter(dp, FTAG);
3192 dsl_deadlist_space(&dd->dd_livelist, &ll_used,
3193 &ll_comp, &ll_uncomp);
3194
3195 dsl_dataset_t *origin_ds;
3196 ASSERT(dsl_pool_config_held(dp));
3197 VERIFY0(dsl_dataset_hold_obj(dp,
3198 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &origin_ds));
3199 VERIFY0(dsl_dataset_space_written(origin_ds, os->os_dsl_dataset,
3200 &used, &comp, &uncomp));
3201 dsl_dataset_rele(origin_ds, FTAG);
3202 dsl_pool_config_exit(dp, FTAG);
3203 /*
3204 * It's possible that the dataset's uncomp space is larger than the
3205 * livelist's because livelists do not track embedded block pointers
3206 */
3207 if (used != ll_used || comp != ll_comp || uncomp < ll_uncomp) {
3208 char nice_used[32], nice_comp[32], nice_uncomp[32];
3209 (void) printf("Discrepancy in space accounting:\n");
3210 zdb_nicenum(used, nice_used, sizeof (nice_used));
3211 zdb_nicenum(comp, nice_comp, sizeof (nice_comp));
3212 zdb_nicenum(uncomp, nice_uncomp, sizeof (nice_uncomp));
3213 (void) printf("dir: used %s, comp %s, uncomp %s\n",
3214 nice_used, nice_comp, nice_uncomp);
3215 zdb_nicenum(ll_used, nice_used, sizeof (nice_used));
3216 zdb_nicenum(ll_comp, nice_comp, sizeof (nice_comp));
3217 zdb_nicenum(ll_uncomp, nice_uncomp, sizeof (nice_uncomp));
3218 (void) printf("livelist: used %s, comp %s, uncomp %s\n",
3219 nice_used, nice_comp, nice_uncomp);
3220 return (1);
3221 }
3222 return (0);
3223 }
3224
3225 static char *key_material = NULL;
3226
3227 static boolean_t
zdb_derive_key(dsl_dir_t * dd,uint8_t * key_out)3228 zdb_derive_key(dsl_dir_t *dd, uint8_t *key_out)
3229 {
3230 uint64_t keyformat, salt, iters;
3231 int i;
3232 unsigned char c;
3233
3234 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3235 zfs_prop_to_name(ZFS_PROP_KEYFORMAT), sizeof (uint64_t),
3236 1, &keyformat));
3237
3238 switch (keyformat) {
3239 case ZFS_KEYFORMAT_HEX:
3240 for (i = 0; i < WRAPPING_KEY_LEN * 2; i += 2) {
3241 if (!isxdigit(key_material[i]) ||
3242 !isxdigit(key_material[i+1]))
3243 return (B_FALSE);
3244 if (sscanf(&key_material[i], "%02hhx", &c) != 1)
3245 return (B_FALSE);
3246 key_out[i / 2] = c;
3247 }
3248 break;
3249
3250 case ZFS_KEYFORMAT_PASSPHRASE:
3251 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3252 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT),
3253 sizeof (uint64_t), 1, &salt));
3254 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3255 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS),
3256 sizeof (uint64_t), 1, &iters));
3257
3258 if (PKCS5_PBKDF2_HMAC_SHA1(key_material, strlen(key_material),
3259 ((uint8_t *)&salt), sizeof (uint64_t), iters,
3260 WRAPPING_KEY_LEN, key_out) != 1)
3261 return (B_FALSE);
3262
3263 break;
3264
3265 default:
3266 fatal("no support for key format %u\n",
3267 (unsigned int) keyformat);
3268 }
3269
3270 return (B_TRUE);
3271 }
3272
3273 static char encroot[ZFS_MAX_DATASET_NAME_LEN];
3274 static boolean_t key_loaded = B_FALSE;
3275
3276 static void
zdb_load_key(objset_t * os)3277 zdb_load_key(objset_t *os)
3278 {
3279 dsl_pool_t *dp;
3280 dsl_dir_t *dd, *rdd;
3281 uint8_t key[WRAPPING_KEY_LEN];
3282 uint64_t rddobj;
3283 int err;
3284
3285 dp = spa_get_dsl(os->os_spa);
3286 dd = os->os_dsl_dataset->ds_dir;
3287
3288 dsl_pool_config_enter(dp, FTAG);
3289 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3290 DSL_CRYPTO_KEY_ROOT_DDOBJ, sizeof (uint64_t), 1, &rddobj));
3291 VERIFY0(dsl_dir_hold_obj(dd->dd_pool, rddobj, NULL, FTAG, &rdd));
3292 dsl_dir_name(rdd, encroot);
3293 dsl_dir_rele(rdd, FTAG);
3294
3295 if (!zdb_derive_key(dd, key))
3296 fatal("couldn't derive encryption key");
3297
3298 dsl_pool_config_exit(dp, FTAG);
3299
3300 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_UNAVAILABLE);
3301
3302 dsl_crypto_params_t *dcp;
3303 nvlist_t *crypto_args;
3304
3305 crypto_args = fnvlist_alloc();
3306 fnvlist_add_uint8_array(crypto_args, "wkeydata",
3307 (uint8_t *)key, WRAPPING_KEY_LEN);
3308 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE,
3309 NULL, crypto_args, &dcp));
3310 err = spa_keystore_load_wkey(encroot, dcp, B_FALSE);
3311
3312 dsl_crypto_params_free(dcp, (err != 0));
3313 fnvlist_free(crypto_args);
3314
3315 if (err != 0)
3316 fatal(
3317 "couldn't load encryption key for %s: %s",
3318 encroot, err == ZFS_ERR_CRYPTO_NOTSUP ?
3319 "crypto params not supported" : strerror(err));
3320
3321 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_AVAILABLE);
3322
3323 printf("Unlocked encryption root: %s\n", encroot);
3324 key_loaded = B_TRUE;
3325 }
3326
3327 static void
zdb_unload_key(void)3328 zdb_unload_key(void)
3329 {
3330 if (!key_loaded)
3331 return;
3332
3333 VERIFY0(spa_keystore_unload_wkey(encroot));
3334 key_loaded = B_FALSE;
3335 }
3336
3337 static avl_tree_t idx_tree;
3338 static avl_tree_t domain_tree;
3339 static boolean_t fuid_table_loaded;
3340 static objset_t *sa_os = NULL;
3341 static sa_attr_type_t *sa_attr_table = NULL;
3342
3343 static int
open_objset(const char * path,const void * tag,objset_t ** osp)3344 open_objset(const char *path, const void *tag, objset_t **osp)
3345 {
3346 int err;
3347 uint64_t sa_attrs = 0;
3348 uint64_t version = 0;
3349
3350 VERIFY3P(sa_os, ==, NULL);
3351
3352 /*
3353 * We can't own an objset if it's redacted. Therefore, we do this
3354 * dance: hold the objset, then acquire a long hold on its dataset, then
3355 * release the pool (which is held as part of holding the objset).
3356 */
3357
3358 if (dump_opt['K']) {
3359 /* decryption requested, try to load keys */
3360 err = dmu_objset_hold(path, tag, osp);
3361 if (err != 0) {
3362 (void) fprintf(stderr, "failed to hold dataset "
3363 "'%s': %s\n",
3364 path, strerror(err));
3365 return (err);
3366 }
3367 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3368 dsl_pool_rele(dmu_objset_pool(*osp), tag);
3369
3370 /* succeeds or dies */
3371 zdb_load_key(*osp);
3372
3373 /* release it all */
3374 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3375 dsl_dataset_rele(dmu_objset_ds(*osp), tag);
3376 }
3377
3378 int ds_hold_flags = key_loaded ? DS_HOLD_FLAG_DECRYPT : 0;
3379
3380 err = dmu_objset_hold_flags(path, ds_hold_flags, tag, osp);
3381 if (err != 0) {
3382 (void) fprintf(stderr, "failed to hold dataset '%s': %s\n",
3383 path, strerror(err));
3384 return (err);
3385 }
3386 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3387 dsl_pool_rele(dmu_objset_pool(*osp), tag);
3388
3389 if (dmu_objset_type(*osp) == DMU_OST_ZFS &&
3390 (key_loaded || !(*osp)->os_encrypted)) {
3391 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR,
3392 8, 1, &version);
3393 if (version >= ZPL_VERSION_SA) {
3394 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS,
3395 8, 1, &sa_attrs);
3396 }
3397 err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END,
3398 &sa_attr_table);
3399 if (err != 0) {
3400 (void) fprintf(stderr, "sa_setup failed: %s\n",
3401 strerror(err));
3402 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3403 dsl_dataset_rele_flags(dmu_objset_ds(*osp),
3404 ds_hold_flags, tag);
3405 *osp = NULL;
3406 }
3407 }
3408 sa_os = *osp;
3409
3410 return (err);
3411 }
3412
3413 static void
close_objset(objset_t * os,const void * tag)3414 close_objset(objset_t *os, const void *tag)
3415 {
3416 VERIFY3P(os, ==, sa_os);
3417 if (os->os_sa != NULL)
3418 sa_tear_down(os);
3419 dsl_dataset_long_rele(dmu_objset_ds(os), tag);
3420 dsl_dataset_rele_flags(dmu_objset_ds(os),
3421 key_loaded ? DS_HOLD_FLAG_DECRYPT : 0, tag);
3422 sa_attr_table = NULL;
3423 sa_os = NULL;
3424
3425 zdb_unload_key();
3426 }
3427
3428 static void
fuid_table_destroy(void)3429 fuid_table_destroy(void)
3430 {
3431 if (fuid_table_loaded) {
3432 zfs_fuid_table_destroy(&idx_tree, &domain_tree);
3433 fuid_table_loaded = B_FALSE;
3434 }
3435 }
3436
3437 /*
3438 * Clean up DDT internal state. ddt_lookup() adds entries to ddt_tree, which on
3439 * a live pool are normally cleaned up during ddt_sync(). We can't do that (and
3440 * wouldn't want to anyway), but if we don't clean up the presence of stuff on
3441 * ddt_tree will trip asserts in ddt_table_free(). So, we clean up ourselves.
3442 *
3443 * Note that this is not a particularly efficient way to do this, but
3444 * ddt_remove() is the only public method that can do the work we need, and it
3445 * requires the right locks and etc to do the job. This is only ever called
3446 * during zdb shutdown so efficiency is not especially important.
3447 */
3448 static void
zdb_ddt_cleanup(spa_t * spa)3449 zdb_ddt_cleanup(spa_t *spa)
3450 {
3451 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
3452 ddt_t *ddt = spa->spa_ddt[c];
3453 if (!ddt)
3454 continue;
3455
3456 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3457 ddt_enter(ddt);
3458 ddt_entry_t *dde = avl_first(&ddt->ddt_tree), *next;
3459 while (dde) {
3460 next = AVL_NEXT(&ddt->ddt_tree, dde);
3461 dde->dde_io = NULL;
3462 ddt_remove(ddt, dde);
3463 dde = next;
3464 }
3465 ddt_exit(ddt);
3466 spa_config_exit(spa, SCL_CONFIG, FTAG);
3467 }
3468 }
3469
3470 static void
zdb_exit(int reason)3471 zdb_exit(int reason)
3472 {
3473 if (spa != NULL)
3474 zdb_ddt_cleanup(spa);
3475
3476 if (os != NULL) {
3477 close_objset(os, FTAG);
3478 } else if (spa != NULL) {
3479 spa_close(spa, FTAG);
3480 }
3481
3482 fuid_table_destroy();
3483
3484 if (kernel_init_done)
3485 kernel_fini();
3486
3487 exit(reason);
3488 }
3489
3490 /*
3491 * print uid or gid information.
3492 * For normal POSIX id just the id is printed in decimal format.
3493 * For CIFS files with FUID the fuid is printed in hex followed by
3494 * the domain-rid string.
3495 */
3496 static void
print_idstr(uint64_t id,const char * id_type)3497 print_idstr(uint64_t id, const char *id_type)
3498 {
3499 if (FUID_INDEX(id)) {
3500 const char *domain =
3501 zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id));
3502 (void) printf("\t%s %llx [%s-%d]\n", id_type,
3503 (u_longlong_t)id, domain, (int)FUID_RID(id));
3504 } else {
3505 (void) printf("\t%s %llu\n", id_type, (u_longlong_t)id);
3506 }
3507
3508 }
3509
3510 static void
dump_uidgid(objset_t * os,uint64_t uid,uint64_t gid)3511 dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid)
3512 {
3513 uint32_t uid_idx, gid_idx;
3514
3515 uid_idx = FUID_INDEX(uid);
3516 gid_idx = FUID_INDEX(gid);
3517
3518 /* Load domain table, if not already loaded */
3519 if (!fuid_table_loaded && (uid_idx || gid_idx)) {
3520 uint64_t fuid_obj;
3521
3522 /* first find the fuid object. It lives in the master node */
3523 VERIFY(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES,
3524 8, 1, &fuid_obj) == 0);
3525 zfs_fuid_avl_tree_create(&idx_tree, &domain_tree);
3526 (void) zfs_fuid_table_load(os, fuid_obj,
3527 &idx_tree, &domain_tree);
3528 fuid_table_loaded = B_TRUE;
3529 }
3530
3531 print_idstr(uid, "uid");
3532 print_idstr(gid, "gid");
3533 }
3534
3535 static void
dump_znode_sa_xattr(sa_handle_t * hdl)3536 dump_znode_sa_xattr(sa_handle_t *hdl)
3537 {
3538 nvlist_t *sa_xattr;
3539 nvpair_t *elem = NULL;
3540 int sa_xattr_size = 0;
3541 int sa_xattr_entries = 0;
3542 int error;
3543 char *sa_xattr_packed;
3544
3545 error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size);
3546 if (error || sa_xattr_size == 0)
3547 return;
3548
3549 sa_xattr_packed = malloc(sa_xattr_size);
3550 if (sa_xattr_packed == NULL)
3551 return;
3552
3553 error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR],
3554 sa_xattr_packed, sa_xattr_size);
3555 if (error) {
3556 free(sa_xattr_packed);
3557 return;
3558 }
3559
3560 error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0);
3561 if (error) {
3562 free(sa_xattr_packed);
3563 return;
3564 }
3565
3566 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL)
3567 sa_xattr_entries++;
3568
3569 (void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3570 sa_xattr_size, sa_xattr_entries);
3571 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) {
3572 boolean_t can_print = !dump_opt['P'];
3573 uchar_t *value;
3574 uint_t cnt, idx;
3575
3576 (void) printf("\t\t%s = ", nvpair_name(elem));
3577 nvpair_value_byte_array(elem, &value, &cnt);
3578
3579 for (idx = 0; idx < cnt; ++idx) {
3580 if (!isprint(value[idx])) {
3581 can_print = B_FALSE;
3582 break;
3583 }
3584 }
3585
3586 for (idx = 0; idx < cnt; ++idx) {
3587 if (can_print)
3588 (void) putchar(value[idx]);
3589 else
3590 (void) printf("\\%3.3o", value[idx]);
3591 }
3592 (void) putchar('\n');
3593 }
3594
3595 nvlist_free(sa_xattr);
3596 free(sa_xattr_packed);
3597 }
3598
3599 static void
dump_znode_symlink(sa_handle_t * hdl)3600 dump_znode_symlink(sa_handle_t *hdl)
3601 {
3602 int sa_symlink_size = 0;
3603 char linktarget[MAXPATHLEN];
3604 int error;
3605
3606 error = sa_size(hdl, sa_attr_table[ZPL_SYMLINK], &sa_symlink_size);
3607 if (error || sa_symlink_size == 0) {
3608 return;
3609 }
3610 if (sa_symlink_size >= sizeof (linktarget)) {
3611 (void) printf("symlink size %d is too large\n",
3612 sa_symlink_size);
3613 return;
3614 }
3615 linktarget[sa_symlink_size] = '\0';
3616 if (sa_lookup(hdl, sa_attr_table[ZPL_SYMLINK],
3617 &linktarget, sa_symlink_size) == 0)
3618 (void) printf("\ttarget %s\n", linktarget);
3619 }
3620
3621 static void
dump_znode(objset_t * os,uint64_t object,void * data,size_t size)3622 dump_znode(objset_t *os, uint64_t object, void *data, size_t size)
3623 {
3624 (void) data, (void) size;
3625 char path[MAXPATHLEN * 2]; /* allow for xattr and failure prefix */
3626 sa_handle_t *hdl;
3627 uint64_t xattr, rdev, gen;
3628 uint64_t uid, gid, mode, fsize, parent, links;
3629 uint64_t pflags;
3630 uint64_t acctm[2], modtm[2], chgtm[2], crtm[2];
3631 time_t z_crtime, z_atime, z_mtime, z_ctime;
3632 sa_bulk_attr_t bulk[12];
3633 int idx = 0;
3634 int error;
3635
3636 VERIFY3P(os, ==, sa_os);
3637 if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) {
3638 (void) printf("Failed to get handle for SA znode\n");
3639 return;
3640 }
3641
3642 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8);
3643 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8);
3644 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL,
3645 &links, 8);
3646 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8);
3647 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL,
3648 &mode, 8);
3649 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT],
3650 NULL, &parent, 8);
3651 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL,
3652 &fsize, 8);
3653 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL,
3654 acctm, 16);
3655 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL,
3656 modtm, 16);
3657 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL,
3658 crtm, 16);
3659 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL,
3660 chgtm, 16);
3661 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL,
3662 &pflags, 8);
3663
3664 if (sa_bulk_lookup(hdl, bulk, idx)) {
3665 (void) sa_handle_destroy(hdl);
3666 return;
3667 }
3668
3669 z_crtime = (time_t)crtm[0];
3670 z_atime = (time_t)acctm[0];
3671 z_mtime = (time_t)modtm[0];
3672 z_ctime = (time_t)chgtm[0];
3673
3674 if (dump_opt['d'] > 4) {
3675 error = zfs_obj_to_path(os, object, path, sizeof (path));
3676 if (error == ESTALE) {
3677 (void) snprintf(path, sizeof (path), "on delete queue");
3678 } else if (error != 0) {
3679 leaked_objects++;
3680 (void) snprintf(path, sizeof (path),
3681 "path not found, possibly leaked");
3682 }
3683 (void) printf("\tpath %s\n", path);
3684 }
3685
3686 if (S_ISLNK(mode))
3687 dump_znode_symlink(hdl);
3688 dump_uidgid(os, uid, gid);
3689 (void) printf("\tatime %s", ctime(&z_atime));
3690 (void) printf("\tmtime %s", ctime(&z_mtime));
3691 (void) printf("\tctime %s", ctime(&z_ctime));
3692 (void) printf("\tcrtime %s", ctime(&z_crtime));
3693 (void) printf("\tgen %llu\n", (u_longlong_t)gen);
3694 (void) printf("\tmode %llo\n", (u_longlong_t)mode);
3695 (void) printf("\tsize %llu\n", (u_longlong_t)fsize);
3696 (void) printf("\tparent %llu\n", (u_longlong_t)parent);
3697 (void) printf("\tlinks %llu\n", (u_longlong_t)links);
3698 (void) printf("\tpflags %llx\n", (u_longlong_t)pflags);
3699 if (dmu_objset_projectquota_enabled(os) && (pflags & ZFS_PROJID)) {
3700 uint64_t projid;
3701
3702 if (sa_lookup(hdl, sa_attr_table[ZPL_PROJID], &projid,
3703 sizeof (uint64_t)) == 0)
3704 (void) printf("\tprojid %llu\n", (u_longlong_t)projid);
3705 }
3706 if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr,
3707 sizeof (uint64_t)) == 0)
3708 (void) printf("\txattr %llu\n", (u_longlong_t)xattr);
3709 if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev,
3710 sizeof (uint64_t)) == 0)
3711 (void) printf("\trdev 0x%016llx\n", (u_longlong_t)rdev);
3712 dump_znode_sa_xattr(hdl);
3713 sa_handle_destroy(hdl);
3714 }
3715
3716 static void
dump_acl(objset_t * os,uint64_t object,void * data,size_t size)3717 dump_acl(objset_t *os, uint64_t object, void *data, size_t size)
3718 {
3719 (void) os, (void) object, (void) data, (void) size;
3720 }
3721
3722 static void
dump_dmu_objset(objset_t * os,uint64_t object,void * data,size_t size)3723 dump_dmu_objset(objset_t *os, uint64_t object, void *data, size_t size)
3724 {
3725 (void) os, (void) object, (void) data, (void) size;
3726 }
3727
3728 static object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = {
3729 dump_none, /* unallocated */
3730 dump_zap, /* object directory */
3731 dump_uint64, /* object array */
3732 dump_none, /* packed nvlist */
3733 dump_packed_nvlist, /* packed nvlist size */
3734 dump_none, /* bpobj */
3735 dump_bpobj, /* bpobj header */
3736 dump_none, /* SPA space map header */
3737 dump_none, /* SPA space map */
3738 dump_none, /* ZIL intent log */
3739 dump_dnode, /* DMU dnode */
3740 dump_dmu_objset, /* DMU objset */
3741 dump_dsl_dir, /* DSL directory */
3742 dump_zap, /* DSL directory child map */
3743 dump_zap, /* DSL dataset snap map */
3744 dump_zap, /* DSL props */
3745 dump_dsl_dataset, /* DSL dataset */
3746 dump_znode, /* ZFS znode */
3747 dump_acl, /* ZFS V0 ACL */
3748 dump_uint8, /* ZFS plain file */
3749 dump_zpldir, /* ZFS directory */
3750 dump_zap, /* ZFS master node */
3751 dump_zap, /* ZFS delete queue */
3752 dump_uint8, /* zvol object */
3753 dump_zap, /* zvol prop */
3754 dump_uint8, /* other uint8[] */
3755 dump_uint64, /* other uint64[] */
3756 dump_zap, /* other ZAP */
3757 dump_zap, /* persistent error log */
3758 dump_uint8, /* SPA history */
3759 dump_history_offsets, /* SPA history offsets */
3760 dump_zap, /* Pool properties */
3761 dump_zap, /* DSL permissions */
3762 dump_acl, /* ZFS ACL */
3763 dump_uint8, /* ZFS SYSACL */
3764 dump_none, /* FUID nvlist */
3765 dump_packed_nvlist, /* FUID nvlist size */
3766 dump_zap, /* DSL dataset next clones */
3767 dump_zap, /* DSL scrub queue */
3768 dump_zap, /* ZFS user/group/project used */
3769 dump_zap, /* ZFS user/group/project quota */
3770 dump_zap, /* snapshot refcount tags */
3771 dump_ddt_zap, /* DDT ZAP object */
3772 dump_zap, /* DDT statistics */
3773 dump_znode, /* SA object */
3774 dump_zap, /* SA Master Node */
3775 dump_sa_attrs, /* SA attribute registration */
3776 dump_sa_layouts, /* SA attribute layouts */
3777 dump_zap, /* DSL scrub translations */
3778 dump_none, /* fake dedup BP */
3779 dump_zap, /* deadlist */
3780 dump_none, /* deadlist hdr */
3781 dump_zap, /* dsl clones */
3782 dump_bpobj_subobjs, /* bpobj subobjs */
3783 dump_unknown, /* Unknown type, must be last */
3784 };
3785
3786 static boolean_t
match_object_type(dmu_object_type_t obj_type,uint64_t flags)3787 match_object_type(dmu_object_type_t obj_type, uint64_t flags)
3788 {
3789 boolean_t match = B_TRUE;
3790
3791 switch (obj_type) {
3792 case DMU_OT_DIRECTORY_CONTENTS:
3793 if (!(flags & ZOR_FLAG_DIRECTORY))
3794 match = B_FALSE;
3795 break;
3796 case DMU_OT_PLAIN_FILE_CONTENTS:
3797 if (!(flags & ZOR_FLAG_PLAIN_FILE))
3798 match = B_FALSE;
3799 break;
3800 case DMU_OT_SPACE_MAP:
3801 if (!(flags & ZOR_FLAG_SPACE_MAP))
3802 match = B_FALSE;
3803 break;
3804 default:
3805 if (strcmp(zdb_ot_name(obj_type), "zap") == 0) {
3806 if (!(flags & ZOR_FLAG_ZAP))
3807 match = B_FALSE;
3808 break;
3809 }
3810
3811 /*
3812 * If all bits except some of the supported flags are
3813 * set, the user combined the all-types flag (A) with
3814 * a negated flag to exclude some types (e.g. A-f to
3815 * show all object types except plain files).
3816 */
3817 if ((flags | ZOR_SUPPORTED_FLAGS) != ZOR_FLAG_ALL_TYPES)
3818 match = B_FALSE;
3819
3820 break;
3821 }
3822
3823 return (match);
3824 }
3825
3826 static void
dump_object(objset_t * os,uint64_t object,int verbosity,boolean_t * print_header,uint64_t * dnode_slots_used,uint64_t flags)3827 dump_object(objset_t *os, uint64_t object, int verbosity,
3828 boolean_t *print_header, uint64_t *dnode_slots_used, uint64_t flags)
3829 {
3830 dmu_buf_t *db = NULL;
3831 dmu_object_info_t doi;
3832 dnode_t *dn;
3833 boolean_t dnode_held = B_FALSE;
3834 void *bonus = NULL;
3835 size_t bsize = 0;
3836 char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
3837 char bonus_size[32];
3838 char aux[50];
3839 int error;
3840
3841 /* make sure nicenum has enough space */
3842 _Static_assert(sizeof (iblk) >= NN_NUMBUF_SZ, "iblk truncated");
3843 _Static_assert(sizeof (dblk) >= NN_NUMBUF_SZ, "dblk truncated");
3844 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, "lsize truncated");
3845 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, "asize truncated");
3846 _Static_assert(sizeof (bonus_size) >= NN_NUMBUF_SZ,
3847 "bonus_size truncated");
3848
3849 if (*print_header) {
3850 (void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n",
3851 "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3852 "lsize", "%full", "type");
3853 *print_header = 0;
3854 }
3855
3856 if (object == 0) {
3857 dn = DMU_META_DNODE(os);
3858 dmu_object_info_from_dnode(dn, &doi);
3859 } else {
3860 /*
3861 * Encrypted datasets will have sensitive bonus buffers
3862 * encrypted. Therefore we cannot hold the bonus buffer and
3863 * must hold the dnode itself instead.
3864 */
3865 error = dmu_object_info(os, object, &doi);
3866 if (error)
3867 fatal("dmu_object_info() failed, errno %u", error);
3868
3869 if (!key_loaded && os->os_encrypted &&
3870 DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) {
3871 error = dnode_hold(os, object, FTAG, &dn);
3872 if (error)
3873 fatal("dnode_hold() failed, errno %u", error);
3874 dnode_held = B_TRUE;
3875 } else {
3876 error = dmu_bonus_hold(os, object, FTAG, &db);
3877 if (error)
3878 fatal("dmu_bonus_hold(%llu) failed, errno %u",
3879 object, error);
3880 bonus = db->db_data;
3881 bsize = db->db_size;
3882 dn = DB_DNODE((dmu_buf_impl_t *)db);
3883 }
3884 }
3885
3886 /*
3887 * Default to showing all object types if no flags were specified.
3888 */
3889 if (flags != 0 && flags != ZOR_FLAG_ALL_TYPES &&
3890 !match_object_type(doi.doi_type, flags))
3891 goto out;
3892
3893 if (dnode_slots_used)
3894 *dnode_slots_used = doi.doi_dnodesize / DNODE_MIN_SIZE;
3895
3896 zdb_nicenum(doi.doi_metadata_block_size, iblk, sizeof (iblk));
3897 zdb_nicenum(doi.doi_data_block_size, dblk, sizeof (dblk));
3898 zdb_nicenum(doi.doi_max_offset, lsize, sizeof (lsize));
3899 zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize, sizeof (asize));
3900 zdb_nicenum(doi.doi_bonus_size, bonus_size, sizeof (bonus_size));
3901 zdb_nicenum(doi.doi_dnodesize, dnsize, sizeof (dnsize));
3902 (void) snprintf(fill, sizeof (fill), "%6.2f", 100.0 *
3903 doi.doi_fill_count * doi.doi_data_block_size / (object == 0 ?
3904 DNODES_PER_BLOCK : 1) / doi.doi_max_offset);
3905
3906 aux[0] = '\0';
3907
3908 if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) {
3909 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3910 " (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum));
3911 }
3912
3913 if (doi.doi_compress == ZIO_COMPRESS_INHERIT &&
3914 ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) {
3915 const char *compname = NULL;
3916 if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION,
3917 ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel),
3918 &compname) == 0) {
3919 (void) snprintf(aux + strlen(aux),
3920 sizeof (aux) - strlen(aux), " (Z=inherit=%s)",
3921 compname);
3922 } else {
3923 (void) snprintf(aux + strlen(aux),
3924 sizeof (aux) - strlen(aux),
3925 " (Z=inherit=%s-unknown)",
3926 ZDB_COMPRESS_NAME(os->os_compress));
3927 }
3928 } else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) {
3929 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3930 " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress));
3931 } else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
3932 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3933 " (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress));
3934 }
3935
3936 (void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n",
3937 (u_longlong_t)object, doi.doi_indirection, iblk, dblk,
3938 asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux);
3939
3940 if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
3941 (void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n",
3942 "", "", "", "", "", "", bonus_size, "bonus",
3943 zdb_ot_name(doi.doi_bonus_type));
3944 }
3945
3946 if (verbosity >= 4) {
3947 (void) printf("\tdnode flags: %s%s%s%s\n",
3948 (dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ?
3949 "USED_BYTES " : "",
3950 (dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ?
3951 "USERUSED_ACCOUNTED " : "",
3952 (dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ?
3953 "USEROBJUSED_ACCOUNTED " : "",
3954 (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ?
3955 "SPILL_BLKPTR" : "");
3956 (void) printf("\tdnode maxblkid: %llu\n",
3957 (longlong_t)dn->dn_phys->dn_maxblkid);
3958
3959 if (!dnode_held) {
3960 object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os,
3961 object, bonus, bsize);
3962 } else {
3963 (void) printf("\t\t(bonus encrypted)\n");
3964 }
3965
3966 if (key_loaded ||
3967 (!os->os_encrypted || !DMU_OT_IS_ENCRYPTED(doi.doi_type))) {
3968 object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object,
3969 NULL, 0);
3970 } else {
3971 (void) printf("\t\t(object encrypted)\n");
3972 }
3973
3974 *print_header = B_TRUE;
3975 }
3976
3977 if (verbosity >= 5) {
3978 if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
3979 char blkbuf[BP_SPRINTF_LEN];
3980 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
3981 DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE);
3982 (void) printf("\nSpill block: %s\n", blkbuf);
3983 }
3984 dump_indirect(dn);
3985 }
3986
3987 if (verbosity >= 5) {
3988 /*
3989 * Report the list of segments that comprise the object.
3990 */
3991 uint64_t start = 0;
3992 uint64_t end;
3993 uint64_t blkfill = 1;
3994 int minlvl = 1;
3995
3996 if (dn->dn_type == DMU_OT_DNODE) {
3997 minlvl = 0;
3998 blkfill = DNODES_PER_BLOCK;
3999 }
4000
4001 for (;;) {
4002 char segsize[32];
4003 /* make sure nicenum has enough space */
4004 _Static_assert(sizeof (segsize) >= NN_NUMBUF_SZ,
4005 "segsize truncated");
4006 error = dnode_next_offset(dn,
4007 0, &start, minlvl, blkfill, 0);
4008 if (error)
4009 break;
4010 end = start;
4011 error = dnode_next_offset(dn,
4012 DNODE_FIND_HOLE, &end, minlvl, blkfill, 0);
4013 zdb_nicenum(end - start, segsize, sizeof (segsize));
4014 (void) printf("\t\tsegment [%016llx, %016llx)"
4015 " size %5s\n", (u_longlong_t)start,
4016 (u_longlong_t)end, segsize);
4017 if (error)
4018 break;
4019 start = end;
4020 }
4021 }
4022
4023 out:
4024 if (db != NULL)
4025 dmu_buf_rele(db, FTAG);
4026 if (dnode_held)
4027 dnode_rele(dn, FTAG);
4028 }
4029
4030 static void
count_dir_mos_objects(dsl_dir_t * dd)4031 count_dir_mos_objects(dsl_dir_t *dd)
4032 {
4033 mos_obj_refd(dd->dd_object);
4034 mos_obj_refd(dsl_dir_phys(dd)->dd_child_dir_zapobj);
4035 mos_obj_refd(dsl_dir_phys(dd)->dd_deleg_zapobj);
4036 mos_obj_refd(dsl_dir_phys(dd)->dd_props_zapobj);
4037 mos_obj_refd(dsl_dir_phys(dd)->dd_clones);
4038
4039 /*
4040 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
4041 * Ignore the references after the first one.
4042 */
4043 mos_obj_refd_multiple(dd->dd_crypto_obj);
4044 }
4045
4046 static void
count_ds_mos_objects(dsl_dataset_t * ds)4047 count_ds_mos_objects(dsl_dataset_t *ds)
4048 {
4049 mos_obj_refd(ds->ds_object);
4050 mos_obj_refd(dsl_dataset_phys(ds)->ds_next_clones_obj);
4051 mos_obj_refd(dsl_dataset_phys(ds)->ds_props_obj);
4052 mos_obj_refd(dsl_dataset_phys(ds)->ds_userrefs_obj);
4053 mos_obj_refd(dsl_dataset_phys(ds)->ds_snapnames_zapobj);
4054 mos_obj_refd(ds->ds_bookmarks_obj);
4055
4056 if (!dsl_dataset_is_snapshot(ds)) {
4057 count_dir_mos_objects(ds->ds_dir);
4058 }
4059 }
4060
4061 static const char *const objset_types[DMU_OST_NUMTYPES] = {
4062 "NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
4063
4064 /*
4065 * Parse a string denoting a range of object IDs of the form
4066 * <start>[:<end>[:flags]], and store the results in zor.
4067 * Return 0 on success. On error, return 1 and update the msg
4068 * pointer to point to a descriptive error message.
4069 */
4070 static int
parse_object_range(char * range,zopt_object_range_t * zor,const char ** msg)4071 parse_object_range(char *range, zopt_object_range_t *zor, const char **msg)
4072 {
4073 uint64_t flags = 0;
4074 char *p, *s, *dup, *flagstr, *tmp = NULL;
4075 size_t len;
4076 int i;
4077 int rc = 0;
4078
4079 if (strchr(range, ':') == NULL) {
4080 zor->zor_obj_start = strtoull(range, &p, 0);
4081 if (*p != '\0') {
4082 *msg = "Invalid characters in object ID";
4083 rc = 1;
4084 }
4085 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4086 zor->zor_obj_end = zor->zor_obj_start;
4087 return (rc);
4088 }
4089
4090 if (strchr(range, ':') == range) {
4091 *msg = "Invalid leading colon";
4092 rc = 1;
4093 return (rc);
4094 }
4095
4096 len = strlen(range);
4097 if (range[len - 1] == ':') {
4098 *msg = "Invalid trailing colon";
4099 rc = 1;
4100 return (rc);
4101 }
4102
4103 dup = strdup(range);
4104 s = strtok_r(dup, ":", &tmp);
4105 zor->zor_obj_start = strtoull(s, &p, 0);
4106
4107 if (*p != '\0') {
4108 *msg = "Invalid characters in start object ID";
4109 rc = 1;
4110 goto out;
4111 }
4112
4113 s = strtok_r(NULL, ":", &tmp);
4114 zor->zor_obj_end = strtoull(s, &p, 0);
4115
4116 if (*p != '\0') {
4117 *msg = "Invalid characters in end object ID";
4118 rc = 1;
4119 goto out;
4120 }
4121
4122 if (zor->zor_obj_start > zor->zor_obj_end) {
4123 *msg = "Start object ID may not exceed end object ID";
4124 rc = 1;
4125 goto out;
4126 }
4127
4128 s = strtok_r(NULL, ":", &tmp);
4129 if (s == NULL) {
4130 zor->zor_flags = ZOR_FLAG_ALL_TYPES;
4131 goto out;
4132 } else if (strtok_r(NULL, ":", &tmp) != NULL) {
4133 *msg = "Invalid colon-delimited field after flags";
4134 rc = 1;
4135 goto out;
4136 }
4137
4138 flagstr = s;
4139 for (i = 0; flagstr[i]; i++) {
4140 int bit;
4141 boolean_t negation = (flagstr[i] == '-');
4142
4143 if (negation) {
4144 i++;
4145 if (flagstr[i] == '\0') {
4146 *msg = "Invalid trailing negation operator";
4147 rc = 1;
4148 goto out;
4149 }
4150 }
4151 bit = flagbits[(uchar_t)flagstr[i]];
4152 if (bit == 0) {
4153 *msg = "Invalid flag";
4154 rc = 1;
4155 goto out;
4156 }
4157 if (negation)
4158 flags &= ~bit;
4159 else
4160 flags |= bit;
4161 }
4162 zor->zor_flags = flags;
4163
4164 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4165 zor->zor_obj_end = ZDB_MAP_OBJECT_ID(zor->zor_obj_end);
4166
4167 out:
4168 free(dup);
4169 return (rc);
4170 }
4171
4172 static void
dump_objset(objset_t * os)4173 dump_objset(objset_t *os)
4174 {
4175 dmu_objset_stats_t dds = { 0 };
4176 uint64_t object, object_count;
4177 uint64_t refdbytes, usedobjs, scratch;
4178 char numbuf[32];
4179 char blkbuf[BP_SPRINTF_LEN + 20];
4180 char osname[ZFS_MAX_DATASET_NAME_LEN];
4181 const char *type = "UNKNOWN";
4182 int verbosity = dump_opt['d'];
4183 boolean_t print_header;
4184 unsigned i;
4185 int error;
4186 uint64_t total_slots_used = 0;
4187 uint64_t max_slot_used = 0;
4188 uint64_t dnode_slots;
4189 uint64_t obj_start;
4190 uint64_t obj_end;
4191 uint64_t flags;
4192
4193 /* make sure nicenum has enough space */
4194 _Static_assert(sizeof (numbuf) >= NN_NUMBUF_SZ, "numbuf truncated");
4195
4196 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
4197 dmu_objset_fast_stat(os, &dds);
4198 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
4199
4200 print_header = B_TRUE;
4201
4202 if (dds.dds_type < DMU_OST_NUMTYPES)
4203 type = objset_types[dds.dds_type];
4204
4205 if (dds.dds_type == DMU_OST_META) {
4206 dds.dds_creation_txg = TXG_INITIAL;
4207 usedobjs = BP_GET_FILL(os->os_rootbp);
4208 refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)->
4209 dd_used_bytes;
4210 } else {
4211 dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
4212 }
4213
4214 ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
4215
4216 zdb_nicenum(refdbytes, numbuf, sizeof (numbuf));
4217
4218 if (verbosity >= 4) {
4219 (void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp ");
4220 (void) snprintf_blkptr(blkbuf + strlen(blkbuf),
4221 sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp);
4222 } else {
4223 blkbuf[0] = '\0';
4224 }
4225
4226 dmu_objset_name(os, osname);
4227
4228 (void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
4229 "%s, %llu objects%s%s\n",
4230 osname, type, (u_longlong_t)dmu_objset_id(os),
4231 (u_longlong_t)dds.dds_creation_txg,
4232 numbuf, (u_longlong_t)usedobjs, blkbuf,
4233 (dds.dds_inconsistent) ? " (inconsistent)" : "");
4234
4235 for (i = 0; i < zopt_object_args; i++) {
4236 obj_start = zopt_object_ranges[i].zor_obj_start;
4237 obj_end = zopt_object_ranges[i].zor_obj_end;
4238 flags = zopt_object_ranges[i].zor_flags;
4239
4240 object = obj_start;
4241 if (object == 0 || obj_start == obj_end)
4242 dump_object(os, object, verbosity, &print_header, NULL,
4243 flags);
4244 else
4245 object--;
4246
4247 while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) &&
4248 object <= obj_end) {
4249 dump_object(os, object, verbosity, &print_header, NULL,
4250 flags);
4251 }
4252 }
4253
4254 if (zopt_object_args > 0) {
4255 (void) printf("\n");
4256 return;
4257 }
4258
4259 if (dump_opt['i'] != 0 || verbosity >= 2)
4260 dump_intent_log(dmu_objset_zil(os));
4261
4262 if (dmu_objset_ds(os) != NULL) {
4263 dsl_dataset_t *ds = dmu_objset_ds(os);
4264 dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
4265 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
4266 !dmu_objset_is_snapshot(os)) {
4267 dump_blkptr_list(&ds->ds_dir->dd_livelist, "Livelist");
4268 if (verify_dd_livelist(os) != 0)
4269 fatal("livelist is incorrect");
4270 }
4271
4272 if (dsl_dataset_remap_deadlist_exists(ds)) {
4273 (void) printf("ds_remap_deadlist:\n");
4274 dump_blkptr_list(&ds->ds_remap_deadlist, "Deadlist");
4275 }
4276 count_ds_mos_objects(ds);
4277 }
4278
4279 if (dmu_objset_ds(os) != NULL)
4280 dump_bookmarks(os, verbosity);
4281
4282 if (verbosity < 2)
4283 return;
4284
4285 if (BP_IS_HOLE(os->os_rootbp))
4286 return;
4287
4288 dump_object(os, 0, verbosity, &print_header, NULL, 0);
4289 object_count = 0;
4290 if (DMU_USERUSED_DNODE(os) != NULL &&
4291 DMU_USERUSED_DNODE(os)->dn_type != 0) {
4292 dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header,
4293 NULL, 0);
4294 dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header,
4295 NULL, 0);
4296 }
4297
4298 if (DMU_PROJECTUSED_DNODE(os) != NULL &&
4299 DMU_PROJECTUSED_DNODE(os)->dn_type != 0)
4300 dump_object(os, DMU_PROJECTUSED_OBJECT, verbosity,
4301 &print_header, NULL, 0);
4302
4303 object = 0;
4304 while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
4305 dump_object(os, object, verbosity, &print_header, &dnode_slots,
4306 0);
4307 object_count++;
4308 total_slots_used += dnode_slots;
4309 max_slot_used = object + dnode_slots - 1;
4310 }
4311
4312 (void) printf("\n");
4313
4314 (void) printf(" Dnode slots:\n");
4315 (void) printf("\tTotal used: %10llu\n",
4316 (u_longlong_t)total_slots_used);
4317 (void) printf("\tMax used: %10llu\n",
4318 (u_longlong_t)max_slot_used);
4319 (void) printf("\tPercent empty: %10lf\n",
4320 (double)(max_slot_used - total_slots_used)*100 /
4321 (double)max_slot_used);
4322 (void) printf("\n");
4323
4324 if (error != ESRCH) {
4325 (void) fprintf(stderr, "dmu_object_next() = %d\n", error);
4326 abort();
4327 }
4328
4329 ASSERT3U(object_count, ==, usedobjs);
4330
4331 if (leaked_objects != 0) {
4332 (void) printf("%d potentially leaked objects detected\n",
4333 leaked_objects);
4334 leaked_objects = 0;
4335 }
4336 }
4337
4338 static void
dump_uberblock(uberblock_t * ub,const char * header,const char * footer)4339 dump_uberblock(uberblock_t *ub, const char *header, const char *footer)
4340 {
4341 time_t timestamp = ub->ub_timestamp;
4342
4343 (void) printf("%s", header ? header : "");
4344 (void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic);
4345 (void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version);
4346 (void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg);
4347 (void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum);
4348 (void) printf("\ttimestamp = %llu UTC = %s",
4349 (u_longlong_t)ub->ub_timestamp, ctime(×tamp));
4350
4351 char blkbuf[BP_SPRINTF_LEN];
4352 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4353 (void) printf("\tbp = %s\n", blkbuf);
4354
4355 (void) printf("\tmmp_magic = %016llx\n",
4356 (u_longlong_t)ub->ub_mmp_magic);
4357 if (MMP_VALID(ub)) {
4358 (void) printf("\tmmp_delay = %0llu\n",
4359 (u_longlong_t)ub->ub_mmp_delay);
4360 if (MMP_SEQ_VALID(ub))
4361 (void) printf("\tmmp_seq = %u\n",
4362 (unsigned int) MMP_SEQ(ub));
4363 if (MMP_FAIL_INT_VALID(ub))
4364 (void) printf("\tmmp_fail = %u\n",
4365 (unsigned int) MMP_FAIL_INT(ub));
4366 if (MMP_INTERVAL_VALID(ub))
4367 (void) printf("\tmmp_write = %u\n",
4368 (unsigned int) MMP_INTERVAL(ub));
4369 /* After MMP_* to make summarize_uberblock_mmp cleaner */
4370 (void) printf("\tmmp_valid = %x\n",
4371 (unsigned int) ub->ub_mmp_config & 0xFF);
4372 }
4373
4374 if (dump_opt['u'] >= 4) {
4375 char blkbuf[BP_SPRINTF_LEN];
4376 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4377 (void) printf("\trootbp = %s\n", blkbuf);
4378 }
4379 (void) printf("\tcheckpoint_txg = %llu\n",
4380 (u_longlong_t)ub->ub_checkpoint_txg);
4381
4382 (void) printf("\traidz_reflow state=%u off=%llu\n",
4383 (int)RRSS_GET_STATE(ub),
4384 (u_longlong_t)RRSS_GET_OFFSET(ub));
4385
4386 (void) printf("%s", footer ? footer : "");
4387 }
4388
4389 static void
dump_config(spa_t * spa)4390 dump_config(spa_t *spa)
4391 {
4392 dmu_buf_t *db;
4393 size_t nvsize = 0;
4394 int error = 0;
4395
4396
4397 error = dmu_bonus_hold(spa->spa_meta_objset,
4398 spa->spa_config_object, FTAG, &db);
4399
4400 if (error == 0) {
4401 nvsize = *(uint64_t *)db->db_data;
4402 dmu_buf_rele(db, FTAG);
4403
4404 (void) printf("\nMOS Configuration:\n");
4405 dump_packed_nvlist(spa->spa_meta_objset,
4406 spa->spa_config_object, (void *)&nvsize, 1);
4407 } else {
4408 (void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d",
4409 (u_longlong_t)spa->spa_config_object, error);
4410 }
4411 }
4412
4413 static void
dump_cachefile(const char * cachefile)4414 dump_cachefile(const char *cachefile)
4415 {
4416 int fd;
4417 struct stat64 statbuf;
4418 char *buf;
4419 nvlist_t *config;
4420
4421 if ((fd = open64(cachefile, O_RDONLY)) < 0) {
4422 (void) printf("cannot open '%s': %s\n", cachefile,
4423 strerror(errno));
4424 zdb_exit(1);
4425 }
4426
4427 if (fstat64(fd, &statbuf) != 0) {
4428 (void) printf("failed to stat '%s': %s\n", cachefile,
4429 strerror(errno));
4430 zdb_exit(1);
4431 }
4432
4433 if ((buf = malloc(statbuf.st_size)) == NULL) {
4434 (void) fprintf(stderr, "failed to allocate %llu bytes\n",
4435 (u_longlong_t)statbuf.st_size);
4436 zdb_exit(1);
4437 }
4438
4439 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
4440 (void) fprintf(stderr, "failed to read %llu bytes\n",
4441 (u_longlong_t)statbuf.st_size);
4442 zdb_exit(1);
4443 }
4444
4445 (void) close(fd);
4446
4447 if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) {
4448 (void) fprintf(stderr, "failed to unpack nvlist\n");
4449 zdb_exit(1);
4450 }
4451
4452 free(buf);
4453
4454 dump_nvlist(config, 0);
4455
4456 nvlist_free(config);
4457 }
4458
4459 /*
4460 * ZFS label nvlist stats
4461 */
4462 typedef struct zdb_nvl_stats {
4463 int zns_list_count;
4464 int zns_leaf_count;
4465 size_t zns_leaf_largest;
4466 size_t zns_leaf_total;
4467 nvlist_t *zns_string;
4468 nvlist_t *zns_uint64;
4469 nvlist_t *zns_boolean;
4470 } zdb_nvl_stats_t;
4471
4472 static void
collect_nvlist_stats(nvlist_t * nvl,zdb_nvl_stats_t * stats)4473 collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats)
4474 {
4475 nvlist_t *list, **array;
4476 nvpair_t *nvp = NULL;
4477 const char *name;
4478 uint_t i, items;
4479
4480 stats->zns_list_count++;
4481
4482 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4483 name = nvpair_name(nvp);
4484
4485 switch (nvpair_type(nvp)) {
4486 case DATA_TYPE_STRING:
4487 fnvlist_add_string(stats->zns_string, name,
4488 fnvpair_value_string(nvp));
4489 break;
4490 case DATA_TYPE_UINT64:
4491 fnvlist_add_uint64(stats->zns_uint64, name,
4492 fnvpair_value_uint64(nvp));
4493 break;
4494 case DATA_TYPE_BOOLEAN:
4495 fnvlist_add_boolean(stats->zns_boolean, name);
4496 break;
4497 case DATA_TYPE_NVLIST:
4498 if (nvpair_value_nvlist(nvp, &list) == 0)
4499 collect_nvlist_stats(list, stats);
4500 break;
4501 case DATA_TYPE_NVLIST_ARRAY:
4502 if (nvpair_value_nvlist_array(nvp, &array, &items) != 0)
4503 break;
4504
4505 for (i = 0; i < items; i++) {
4506 collect_nvlist_stats(array[i], stats);
4507
4508 /* collect stats on leaf vdev */
4509 if (strcmp(name, "children") == 0) {
4510 size_t size;
4511
4512 (void) nvlist_size(array[i], &size,
4513 NV_ENCODE_XDR);
4514 stats->zns_leaf_total += size;
4515 if (size > stats->zns_leaf_largest)
4516 stats->zns_leaf_largest = size;
4517 stats->zns_leaf_count++;
4518 }
4519 }
4520 break;
4521 default:
4522 (void) printf("skip type %d!\n", (int)nvpair_type(nvp));
4523 }
4524 }
4525 }
4526
4527 static void
dump_nvlist_stats(nvlist_t * nvl,size_t cap)4528 dump_nvlist_stats(nvlist_t *nvl, size_t cap)
4529 {
4530 zdb_nvl_stats_t stats = { 0 };
4531 size_t size, sum = 0, total;
4532 size_t noise;
4533
4534 /* requires nvlist with non-unique names for stat collection */
4535 VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0));
4536 VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0));
4537 VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0));
4538 VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR));
4539
4540 (void) printf("\n\nZFS Label NVList Config Stats:\n");
4541
4542 VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR));
4543 (void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4544 (int)total, (int)(cap - total), 100.0 * total / cap);
4545
4546 collect_nvlist_stats(nvl, &stats);
4547
4548 VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR));
4549 size -= noise;
4550 sum += size;
4551 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4552 (int)fnvlist_num_pairs(stats.zns_uint64),
4553 (int)size, 100.0 * size / total);
4554
4555 VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR));
4556 size -= noise;
4557 sum += size;
4558 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4559 (int)fnvlist_num_pairs(stats.zns_string),
4560 (int)size, 100.0 * size / total);
4561
4562 VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR));
4563 size -= noise;
4564 sum += size;
4565 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4566 (int)fnvlist_num_pairs(stats.zns_boolean),
4567 (int)size, 100.0 * size / total);
4568
4569 size = total - sum; /* treat remainder as nvlist overhead */
4570 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4571 stats.zns_list_count, (int)size, 100.0 * size / total);
4572
4573 if (stats.zns_leaf_count > 0) {
4574 size_t average = stats.zns_leaf_total / stats.zns_leaf_count;
4575
4576 (void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4577 stats.zns_leaf_count, (int)average);
4578 (void) printf("%24d bytes largest\n",
4579 (int)stats.zns_leaf_largest);
4580
4581 if (dump_opt['l'] >= 3 && average > 0)
4582 (void) printf(" space for %d additional leaf vdevs\n",
4583 (int)((cap - total) / average));
4584 }
4585 (void) printf("\n");
4586
4587 nvlist_free(stats.zns_string);
4588 nvlist_free(stats.zns_uint64);
4589 nvlist_free(stats.zns_boolean);
4590 }
4591
4592 typedef struct cksum_record {
4593 zio_cksum_t cksum;
4594 boolean_t labels[VDEV_LABELS];
4595 avl_node_t link;
4596 } cksum_record_t;
4597
4598 static int
cksum_record_compare(const void * x1,const void * x2)4599 cksum_record_compare(const void *x1, const void *x2)
4600 {
4601 const cksum_record_t *l = (cksum_record_t *)x1;
4602 const cksum_record_t *r = (cksum_record_t *)x2;
4603 int arraysize = ARRAY_SIZE(l->cksum.zc_word);
4604 int difference = 0;
4605
4606 for (int i = 0; i < arraysize; i++) {
4607 difference = TREE_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]);
4608 if (difference)
4609 break;
4610 }
4611
4612 return (difference);
4613 }
4614
4615 static cksum_record_t *
cksum_record_alloc(zio_cksum_t * cksum,int l)4616 cksum_record_alloc(zio_cksum_t *cksum, int l)
4617 {
4618 cksum_record_t *rec;
4619
4620 rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL);
4621 rec->cksum = *cksum;
4622 rec->labels[l] = B_TRUE;
4623
4624 return (rec);
4625 }
4626
4627 static cksum_record_t *
cksum_record_lookup(avl_tree_t * tree,zio_cksum_t * cksum)4628 cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum)
4629 {
4630 cksum_record_t lookup = { .cksum = *cksum };
4631 avl_index_t where;
4632
4633 return (avl_find(tree, &lookup, &where));
4634 }
4635
4636 static cksum_record_t *
cksum_record_insert(avl_tree_t * tree,zio_cksum_t * cksum,int l)4637 cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l)
4638 {
4639 cksum_record_t *rec;
4640
4641 rec = cksum_record_lookup(tree, cksum);
4642 if (rec) {
4643 rec->labels[l] = B_TRUE;
4644 } else {
4645 rec = cksum_record_alloc(cksum, l);
4646 avl_add(tree, rec);
4647 }
4648
4649 return (rec);
4650 }
4651
4652 static int
first_label(cksum_record_t * rec)4653 first_label(cksum_record_t *rec)
4654 {
4655 for (int i = 0; i < VDEV_LABELS; i++)
4656 if (rec->labels[i])
4657 return (i);
4658
4659 return (-1);
4660 }
4661
4662 static void
print_label_numbers(const char * prefix,const cksum_record_t * rec)4663 print_label_numbers(const char *prefix, const cksum_record_t *rec)
4664 {
4665 fputs(prefix, stdout);
4666 for (int i = 0; i < VDEV_LABELS; i++)
4667 if (rec->labels[i] == B_TRUE)
4668 printf("%d ", i);
4669 putchar('\n');
4670 }
4671
4672 #define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4673
4674 typedef struct zdb_label {
4675 vdev_label_t label;
4676 uint64_t label_offset;
4677 nvlist_t *config_nv;
4678 cksum_record_t *config;
4679 cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT];
4680 boolean_t header_printed;
4681 boolean_t read_failed;
4682 boolean_t cksum_valid;
4683 } zdb_label_t;
4684
4685 static void
print_label_header(zdb_label_t * label,int l)4686 print_label_header(zdb_label_t *label, int l)
4687 {
4688
4689 if (dump_opt['q'])
4690 return;
4691
4692 if (label->header_printed == B_TRUE)
4693 return;
4694
4695 (void) printf("------------------------------------\n");
4696 (void) printf("LABEL %d %s\n", l,
4697 label->cksum_valid ? "" : "(Bad label cksum)");
4698 (void) printf("------------------------------------\n");
4699
4700 label->header_printed = B_TRUE;
4701 }
4702
4703 static void
print_l2arc_header(void)4704 print_l2arc_header(void)
4705 {
4706 (void) printf("------------------------------------\n");
4707 (void) printf("L2ARC device header\n");
4708 (void) printf("------------------------------------\n");
4709 }
4710
4711 static void
print_l2arc_log_blocks(void)4712 print_l2arc_log_blocks(void)
4713 {
4714 (void) printf("------------------------------------\n");
4715 (void) printf("L2ARC device log blocks\n");
4716 (void) printf("------------------------------------\n");
4717 }
4718
4719 static void
dump_l2arc_log_entries(uint64_t log_entries,l2arc_log_ent_phys_t * le,uint64_t i)4720 dump_l2arc_log_entries(uint64_t log_entries,
4721 l2arc_log_ent_phys_t *le, uint64_t i)
4722 {
4723 for (int j = 0; j < log_entries; j++) {
4724 dva_t dva = le[j].le_dva;
4725 (void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4726 "vdev: %llu, offset: %llu\n",
4727 (u_longlong_t)i, j + 1,
4728 (u_longlong_t)DVA_GET_ASIZE(&dva),
4729 (u_longlong_t)DVA_GET_VDEV(&dva),
4730 (u_longlong_t)DVA_GET_OFFSET(&dva));
4731 (void) printf("|\t\t\t\tbirth: %llu\n",
4732 (u_longlong_t)le[j].le_birth);
4733 (void) printf("|\t\t\t\tlsize: %llu\n",
4734 (u_longlong_t)L2BLK_GET_LSIZE((&le[j])->le_prop));
4735 (void) printf("|\t\t\t\tpsize: %llu\n",
4736 (u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop));
4737 (void) printf("|\t\t\t\tcompr: %llu\n",
4738 (u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop));
4739 (void) printf("|\t\t\t\tcomplevel: %llu\n",
4740 (u_longlong_t)(&le[j])->le_complevel);
4741 (void) printf("|\t\t\t\ttype: %llu\n",
4742 (u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop));
4743 (void) printf("|\t\t\t\tprotected: %llu\n",
4744 (u_longlong_t)L2BLK_GET_PROTECTED((&le[j])->le_prop));
4745 (void) printf("|\t\t\t\tprefetch: %llu\n",
4746 (u_longlong_t)L2BLK_GET_PREFETCH((&le[j])->le_prop));
4747 (void) printf("|\t\t\t\taddress: %llu\n",
4748 (u_longlong_t)le[j].le_daddr);
4749 (void) printf("|\t\t\t\tARC state: %llu\n",
4750 (u_longlong_t)L2BLK_GET_STATE((&le[j])->le_prop));
4751 (void) printf("|\n");
4752 }
4753 (void) printf("\n");
4754 }
4755
4756 static void
dump_l2arc_log_blkptr(const l2arc_log_blkptr_t * lbps)4757 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t *lbps)
4758 {
4759 (void) printf("|\t\tdaddr: %llu\n", (u_longlong_t)lbps->lbp_daddr);
4760 (void) printf("|\t\tpayload_asize: %llu\n",
4761 (u_longlong_t)lbps->lbp_payload_asize);
4762 (void) printf("|\t\tpayload_start: %llu\n",
4763 (u_longlong_t)lbps->lbp_payload_start);
4764 (void) printf("|\t\tlsize: %llu\n",
4765 (u_longlong_t)L2BLK_GET_LSIZE(lbps->lbp_prop));
4766 (void) printf("|\t\tasize: %llu\n",
4767 (u_longlong_t)L2BLK_GET_PSIZE(lbps->lbp_prop));
4768 (void) printf("|\t\tcompralgo: %llu\n",
4769 (u_longlong_t)L2BLK_GET_COMPRESS(lbps->lbp_prop));
4770 (void) printf("|\t\tcksumalgo: %llu\n",
4771 (u_longlong_t)L2BLK_GET_CHECKSUM(lbps->lbp_prop));
4772 (void) printf("|\n\n");
4773 }
4774
4775 static void
dump_l2arc_log_blocks(int fd,const l2arc_dev_hdr_phys_t * l2dhdr,l2arc_dev_hdr_phys_t * rebuild)4776 dump_l2arc_log_blocks(int fd, const l2arc_dev_hdr_phys_t *l2dhdr,
4777 l2arc_dev_hdr_phys_t *rebuild)
4778 {
4779 l2arc_log_blk_phys_t this_lb;
4780 uint64_t asize;
4781 l2arc_log_blkptr_t lbps[2];
4782 zio_cksum_t cksum;
4783 int failed = 0;
4784 l2arc_dev_t dev;
4785
4786 if (!dump_opt['q'])
4787 print_l2arc_log_blocks();
4788 memcpy(lbps, l2dhdr->dh_start_lbps, sizeof (lbps));
4789
4790 dev.l2ad_evict = l2dhdr->dh_evict;
4791 dev.l2ad_start = l2dhdr->dh_start;
4792 dev.l2ad_end = l2dhdr->dh_end;
4793
4794 if (l2dhdr->dh_start_lbps[0].lbp_daddr == 0) {
4795 /* no log blocks to read */
4796 if (!dump_opt['q']) {
4797 (void) printf("No log blocks to read\n");
4798 (void) printf("\n");
4799 }
4800 return;
4801 } else {
4802 dev.l2ad_hand = lbps[0].lbp_daddr +
4803 L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4804 }
4805
4806 dev.l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST);
4807
4808 for (;;) {
4809 if (!l2arc_log_blkptr_valid(&dev, &lbps[0]))
4810 break;
4811
4812 /* L2BLK_GET_PSIZE returns aligned size for log blocks */
4813 asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4814 if (pread64(fd, &this_lb, asize, lbps[0].lbp_daddr) != asize) {
4815 if (!dump_opt['q']) {
4816 (void) printf("Error while reading next log "
4817 "block\n\n");
4818 }
4819 break;
4820 }
4821
4822 fletcher_4_native_varsize(&this_lb, asize, &cksum);
4823 if (!ZIO_CHECKSUM_EQUAL(cksum, lbps[0].lbp_cksum)) {
4824 failed++;
4825 if (!dump_opt['q']) {
4826 (void) printf("Invalid cksum\n");
4827 dump_l2arc_log_blkptr(&lbps[0]);
4828 }
4829 break;
4830 }
4831
4832 switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) {
4833 case ZIO_COMPRESS_OFF:
4834 break;
4835 default: {
4836 abd_t *abd = abd_alloc_linear(asize, B_TRUE);
4837 abd_copy_from_buf_off(abd, &this_lb, 0, asize);
4838 abd_t dabd;
4839 abd_get_from_buf_struct(&dabd, &this_lb,
4840 sizeof (this_lb));
4841 int err = zio_decompress_data(L2BLK_GET_COMPRESS(
4842 (&lbps[0])->lbp_prop), abd, &dabd,
4843 asize, sizeof (this_lb), NULL);
4844 abd_free(&dabd);
4845 abd_free(abd);
4846 if (err != 0) {
4847 (void) printf("L2ARC block decompression "
4848 "failed\n");
4849 goto out;
4850 }
4851 break;
4852 }
4853 }
4854
4855 if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
4856 byteswap_uint64_array(&this_lb, sizeof (this_lb));
4857 if (this_lb.lb_magic != L2ARC_LOG_BLK_MAGIC) {
4858 if (!dump_opt['q'])
4859 (void) printf("Invalid log block magic\n\n");
4860 break;
4861 }
4862
4863 rebuild->dh_lb_count++;
4864 rebuild->dh_lb_asize += asize;
4865 if (dump_opt['l'] > 1 && !dump_opt['q']) {
4866 (void) printf("lb[%4llu]\tmagic: %llu\n",
4867 (u_longlong_t)rebuild->dh_lb_count,
4868 (u_longlong_t)this_lb.lb_magic);
4869 dump_l2arc_log_blkptr(&lbps[0]);
4870 }
4871
4872 if (dump_opt['l'] > 2 && !dump_opt['q'])
4873 dump_l2arc_log_entries(l2dhdr->dh_log_entries,
4874 this_lb.lb_entries,
4875 rebuild->dh_lb_count);
4876
4877 if (l2arc_range_check_overlap(lbps[1].lbp_payload_start,
4878 lbps[0].lbp_payload_start, dev.l2ad_evict) &&
4879 !dev.l2ad_first)
4880 break;
4881
4882 lbps[0] = lbps[1];
4883 lbps[1] = this_lb.lb_prev_lbp;
4884 }
4885 out:
4886 if (!dump_opt['q']) {
4887 (void) printf("log_blk_count:\t %llu with valid cksum\n",
4888 (u_longlong_t)rebuild->dh_lb_count);
4889 (void) printf("\t\t %d with invalid cksum\n", failed);
4890 (void) printf("log_blk_asize:\t %llu\n\n",
4891 (u_longlong_t)rebuild->dh_lb_asize);
4892 }
4893 }
4894
4895 static int
dump_l2arc_header(int fd)4896 dump_l2arc_header(int fd)
4897 {
4898 l2arc_dev_hdr_phys_t l2dhdr = {0}, rebuild = {0};
4899 int error = B_FALSE;
4900
4901 if (pread64(fd, &l2dhdr, sizeof (l2dhdr),
4902 VDEV_LABEL_START_SIZE) != sizeof (l2dhdr)) {
4903 error = B_TRUE;
4904 } else {
4905 if (l2dhdr.dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC))
4906 byteswap_uint64_array(&l2dhdr, sizeof (l2dhdr));
4907
4908 if (l2dhdr.dh_magic != L2ARC_DEV_HDR_MAGIC)
4909 error = B_TRUE;
4910 }
4911
4912 if (error) {
4913 (void) printf("L2ARC device header not found\n\n");
4914 /* Do not return an error here for backward compatibility */
4915 return (0);
4916 } else if (!dump_opt['q']) {
4917 print_l2arc_header();
4918
4919 (void) printf(" magic: %llu\n",
4920 (u_longlong_t)l2dhdr.dh_magic);
4921 (void) printf(" version: %llu\n",
4922 (u_longlong_t)l2dhdr.dh_version);
4923 (void) printf(" pool_guid: %llu\n",
4924 (u_longlong_t)l2dhdr.dh_spa_guid);
4925 (void) printf(" flags: %llu\n",
4926 (u_longlong_t)l2dhdr.dh_flags);
4927 (void) printf(" start_lbps[0]: %llu\n",
4928 (u_longlong_t)
4929 l2dhdr.dh_start_lbps[0].lbp_daddr);
4930 (void) printf(" start_lbps[1]: %llu\n",
4931 (u_longlong_t)
4932 l2dhdr.dh_start_lbps[1].lbp_daddr);
4933 (void) printf(" log_blk_ent: %llu\n",
4934 (u_longlong_t)l2dhdr.dh_log_entries);
4935 (void) printf(" start: %llu\n",
4936 (u_longlong_t)l2dhdr.dh_start);
4937 (void) printf(" end: %llu\n",
4938 (u_longlong_t)l2dhdr.dh_end);
4939 (void) printf(" evict: %llu\n",
4940 (u_longlong_t)l2dhdr.dh_evict);
4941 (void) printf(" lb_asize_refcount: %llu\n",
4942 (u_longlong_t)l2dhdr.dh_lb_asize);
4943 (void) printf(" lb_count_refcount: %llu\n",
4944 (u_longlong_t)l2dhdr.dh_lb_count);
4945 (void) printf(" trim_action_time: %llu\n",
4946 (u_longlong_t)l2dhdr.dh_trim_action_time);
4947 (void) printf(" trim_state: %llu\n\n",
4948 (u_longlong_t)l2dhdr.dh_trim_state);
4949 }
4950
4951 dump_l2arc_log_blocks(fd, &l2dhdr, &rebuild);
4952 /*
4953 * The total aligned size of log blocks and the number of log blocks
4954 * reported in the header of the device may be less than what zdb
4955 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
4956 * This happens because dump_l2arc_log_blocks() lacks the memory
4957 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
4958 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
4959 * and dh_lb_count will be lower to begin with than what exists on the
4960 * device. This is normal and zdb should not exit with an error. The
4961 * opposite case should never happen though, the values reported in the
4962 * header should never be higher than what dump_l2arc_log_blocks() and
4963 * l2arc_rebuild() report. If this happens there is a leak in the
4964 * accounting of log blocks.
4965 */
4966 if (l2dhdr.dh_lb_asize > rebuild.dh_lb_asize ||
4967 l2dhdr.dh_lb_count > rebuild.dh_lb_count)
4968 return (1);
4969
4970 return (0);
4971 }
4972
4973 static void
dump_config_from_label(zdb_label_t * label,size_t buflen,int l)4974 dump_config_from_label(zdb_label_t *label, size_t buflen, int l)
4975 {
4976 if (dump_opt['q'])
4977 return;
4978
4979 if ((dump_opt['l'] < 3) && (first_label(label->config) != l))
4980 return;
4981
4982 print_label_header(label, l);
4983 dump_nvlist(label->config_nv, 4);
4984 print_label_numbers(" labels = ", label->config);
4985
4986 if (dump_opt['l'] >= 2)
4987 dump_nvlist_stats(label->config_nv, buflen);
4988 }
4989
4990 #define ZDB_MAX_UB_HEADER_SIZE 32
4991
4992 static void
dump_label_uberblocks(zdb_label_t * label,uint64_t ashift,int label_num)4993 dump_label_uberblocks(zdb_label_t *label, uint64_t ashift, int label_num)
4994 {
4995
4996 vdev_t vd;
4997 char header[ZDB_MAX_UB_HEADER_SIZE];
4998
4999 vd.vdev_ashift = ashift;
5000 vd.vdev_top = &vd;
5001
5002 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5003 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5004 uberblock_t *ub = (void *)((char *)&label->label + uoff);
5005 cksum_record_t *rec = label->uberblocks[i];
5006
5007 if (rec == NULL) {
5008 if (dump_opt['u'] >= 2) {
5009 print_label_header(label, label_num);
5010 (void) printf(" Uberblock[%d] invalid\n", i);
5011 }
5012 continue;
5013 }
5014
5015 if ((dump_opt['u'] < 3) && (first_label(rec) != label_num))
5016 continue;
5017
5018 if ((dump_opt['u'] < 4) &&
5019 (ub->ub_mmp_magic == MMP_MAGIC) && ub->ub_mmp_delay &&
5020 (i >= VDEV_UBERBLOCK_COUNT(&vd) - MMP_BLOCKS_PER_LABEL))
5021 continue;
5022
5023 print_label_header(label, label_num);
5024 (void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE,
5025 " Uberblock[%d]\n", i);
5026 dump_uberblock(ub, header, "");
5027 print_label_numbers(" labels = ", rec);
5028 }
5029 }
5030
5031 static char curpath[PATH_MAX];
5032
5033 /*
5034 * Iterate through the path components, recursively passing
5035 * current one's obj and remaining path until we find the obj
5036 * for the last one.
5037 */
5038 static int
dump_path_impl(objset_t * os,uint64_t obj,char * name,uint64_t * retobj)5039 dump_path_impl(objset_t *os, uint64_t obj, char *name, uint64_t *retobj)
5040 {
5041 int err;
5042 boolean_t header = B_TRUE;
5043 uint64_t child_obj;
5044 char *s;
5045 dmu_buf_t *db;
5046 dmu_object_info_t doi;
5047
5048 if ((s = strchr(name, '/')) != NULL)
5049 *s = '\0';
5050 err = zap_lookup(os, obj, name, 8, 1, &child_obj);
5051
5052 (void) strlcat(curpath, name, sizeof (curpath));
5053
5054 if (err != 0) {
5055 (void) fprintf(stderr, "failed to lookup %s: %s\n",
5056 curpath, strerror(err));
5057 return (err);
5058 }
5059
5060 child_obj = ZFS_DIRENT_OBJ(child_obj);
5061 err = sa_buf_hold(os, child_obj, FTAG, &db);
5062 if (err != 0) {
5063 (void) fprintf(stderr,
5064 "failed to get SA dbuf for obj %llu: %s\n",
5065 (u_longlong_t)child_obj, strerror(err));
5066 return (EINVAL);
5067 }
5068 dmu_object_info_from_db(db, &doi);
5069 sa_buf_rele(db, FTAG);
5070
5071 if (doi.doi_bonus_type != DMU_OT_SA &&
5072 doi.doi_bonus_type != DMU_OT_ZNODE) {
5073 (void) fprintf(stderr, "invalid bonus type %d for obj %llu\n",
5074 doi.doi_bonus_type, (u_longlong_t)child_obj);
5075 return (EINVAL);
5076 }
5077
5078 if (dump_opt['v'] > 6) {
5079 (void) printf("obj=%llu %s type=%d bonustype=%d\n",
5080 (u_longlong_t)child_obj, curpath, doi.doi_type,
5081 doi.doi_bonus_type);
5082 }
5083
5084 (void) strlcat(curpath, "/", sizeof (curpath));
5085
5086 switch (doi.doi_type) {
5087 case DMU_OT_DIRECTORY_CONTENTS:
5088 if (s != NULL && *(s + 1) != '\0')
5089 return (dump_path_impl(os, child_obj, s + 1, retobj));
5090 zfs_fallthrough;
5091 case DMU_OT_PLAIN_FILE_CONTENTS:
5092 if (retobj != NULL) {
5093 *retobj = child_obj;
5094 } else {
5095 dump_object(os, child_obj, dump_opt['v'], &header,
5096 NULL, 0);
5097 }
5098 return (0);
5099 default:
5100 (void) fprintf(stderr, "object %llu has non-file/directory "
5101 "type %d\n", (u_longlong_t)obj, doi.doi_type);
5102 break;
5103 }
5104
5105 return (EINVAL);
5106 }
5107
5108 /*
5109 * Dump the blocks for the object specified by path inside the dataset.
5110 */
5111 static int
dump_path(char * ds,char * path,uint64_t * retobj)5112 dump_path(char *ds, char *path, uint64_t *retobj)
5113 {
5114 int err;
5115 objset_t *os;
5116 uint64_t root_obj;
5117
5118 err = open_objset(ds, FTAG, &os);
5119 if (err != 0)
5120 return (err);
5121
5122 err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj);
5123 if (err != 0) {
5124 (void) fprintf(stderr, "can't lookup root znode: %s\n",
5125 strerror(err));
5126 close_objset(os, FTAG);
5127 return (EINVAL);
5128 }
5129
5130 (void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds);
5131
5132 err = dump_path_impl(os, root_obj, path, retobj);
5133
5134 close_objset(os, FTAG);
5135 return (err);
5136 }
5137
5138 static int
dump_backup_bytes(objset_t * os,void * buf,int len,void * arg)5139 dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
5140 {
5141 const char *p = (const char *)buf;
5142 ssize_t nwritten;
5143
5144 (void) os;
5145 (void) arg;
5146
5147 /* Write the data out, handling short writes and signals. */
5148 while ((nwritten = write(STDOUT_FILENO, p, len)) < len) {
5149 if (nwritten < 0) {
5150 if (errno == EINTR)
5151 continue;
5152 return (errno);
5153 }
5154 p += nwritten;
5155 len -= nwritten;
5156 }
5157
5158 return (0);
5159 }
5160
5161 static void
dump_backup(const char * pool,uint64_t objset_id,const char * flagstr)5162 dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
5163 {
5164 boolean_t embed = B_FALSE;
5165 boolean_t large_block = B_FALSE;
5166 boolean_t compress = B_FALSE;
5167 boolean_t raw = B_FALSE;
5168
5169 const char *c;
5170 for (c = flagstr; c != NULL && *c != '\0'; c++) {
5171 switch (*c) {
5172 case 'e':
5173 embed = B_TRUE;
5174 break;
5175 case 'L':
5176 large_block = B_TRUE;
5177 break;
5178 case 'c':
5179 compress = B_TRUE;
5180 break;
5181 case 'w':
5182 raw = B_TRUE;
5183 break;
5184 default:
5185 fprintf(stderr, "dump_backup: invalid flag "
5186 "'%c'\n", *c);
5187 return;
5188 }
5189 }
5190
5191 if (isatty(STDOUT_FILENO)) {
5192 fprintf(stderr, "dump_backup: stream cannot be written "
5193 "to a terminal\n");
5194 return;
5195 }
5196
5197 offset_t off = 0;
5198 dmu_send_outparams_t out = {
5199 .dso_outfunc = dump_backup_bytes,
5200 .dso_dryrun = B_FALSE,
5201 };
5202
5203 int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed,
5204 large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO,
5205 &off, &out);
5206 if (err != 0) {
5207 fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",
5208 strerror(err));
5209 return;
5210 }
5211 }
5212
5213 static int
zdb_copy_object(objset_t * os,uint64_t srcobj,char * destfile)5214 zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile)
5215 {
5216 int err = 0;
5217 uint64_t size, readsize, oursize, offset;
5218 ssize_t writesize;
5219 sa_handle_t *hdl;
5220
5221 (void) printf("Copying object %" PRIu64 " to file %s\n", srcobj,
5222 destfile);
5223
5224 VERIFY3P(os, ==, sa_os);
5225 if ((err = sa_handle_get(os, srcobj, NULL, SA_HDL_PRIVATE, &hdl))) {
5226 (void) printf("Failed to get handle for SA znode\n");
5227 return (err);
5228 }
5229 if ((err = sa_lookup(hdl, sa_attr_table[ZPL_SIZE], &size, 8))) {
5230 (void) sa_handle_destroy(hdl);
5231 return (err);
5232 }
5233 (void) sa_handle_destroy(hdl);
5234
5235 (void) printf("Object %" PRIu64 " is %" PRIu64 " bytes\n", srcobj,
5236 size);
5237 if (size == 0) {
5238 return (EINVAL);
5239 }
5240
5241 int fd = open(destfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
5242 if (fd == -1)
5243 return (errno);
5244 /*
5245 * We cap the size at 1 mebibyte here to prevent
5246 * allocation failures and nigh-infinite printing if the
5247 * object is extremely large.
5248 */
5249 oursize = MIN(size, 1 << 20);
5250 offset = 0;
5251 char *buf = kmem_alloc(oursize, KM_NOSLEEP);
5252 if (buf == NULL) {
5253 (void) close(fd);
5254 return (ENOMEM);
5255 }
5256
5257 while (offset < size) {
5258 readsize = MIN(size - offset, 1 << 20);
5259 err = dmu_read(os, srcobj, offset, readsize, buf, 0);
5260 if (err != 0) {
5261 (void) printf("got error %u from dmu_read\n", err);
5262 kmem_free(buf, oursize);
5263 (void) close(fd);
5264 return (err);
5265 }
5266 if (dump_opt['v'] > 3) {
5267 (void) printf("Read offset=%" PRIu64 " size=%" PRIu64
5268 " error=%d\n", offset, readsize, err);
5269 }
5270
5271 writesize = write(fd, buf, readsize);
5272 if (writesize < 0) {
5273 err = errno;
5274 break;
5275 } else if (writesize != readsize) {
5276 /* Incomplete write */
5277 (void) fprintf(stderr, "Short write, only wrote %llu of"
5278 " %" PRIu64 " bytes, exiting...\n",
5279 (u_longlong_t)writesize, readsize);
5280 break;
5281 }
5282
5283 offset += readsize;
5284 }
5285
5286 (void) close(fd);
5287
5288 if (buf != NULL)
5289 kmem_free(buf, oursize);
5290
5291 return (err);
5292 }
5293
5294 static boolean_t
label_cksum_valid(vdev_label_t * label,uint64_t offset)5295 label_cksum_valid(vdev_label_t *label, uint64_t offset)
5296 {
5297 zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
5298 zio_cksum_t expected_cksum;
5299 zio_cksum_t actual_cksum;
5300 zio_cksum_t verifier;
5301 zio_eck_t *eck;
5302 int byteswap;
5303
5304 void *data = (char *)label + offsetof(vdev_label_t, vl_vdev_phys);
5305 eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1;
5306
5307 offset += offsetof(vdev_label_t, vl_vdev_phys);
5308 ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
5309
5310 byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
5311 if (byteswap)
5312 byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
5313
5314 expected_cksum = eck->zec_cksum;
5315 eck->zec_cksum = verifier;
5316
5317 abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE);
5318 ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum);
5319 abd_free(abd);
5320
5321 if (byteswap)
5322 byteswap_uint64_array(&expected_cksum, sizeof (zio_cksum_t));
5323
5324 if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
5325 return (B_TRUE);
5326
5327 return (B_FALSE);
5328 }
5329
5330 static int
dump_label(const char * dev)5331 dump_label(const char *dev)
5332 {
5333 char path[MAXPATHLEN];
5334 zdb_label_t labels[VDEV_LABELS] = {{{{0}}}};
5335 uint64_t psize, ashift, l2cache;
5336 struct stat64 statbuf;
5337 boolean_t config_found = B_FALSE;
5338 boolean_t error = B_FALSE;
5339 boolean_t read_l2arc_header = B_FALSE;
5340 avl_tree_t config_tree;
5341 avl_tree_t uberblock_tree;
5342 void *node, *cookie;
5343 int fd;
5344
5345 /*
5346 * Check if we were given absolute path and use it as is.
5347 * Otherwise if the provided vdev name doesn't point to a file,
5348 * try prepending expected disk paths and partition numbers.
5349 */
5350 (void) strlcpy(path, dev, sizeof (path));
5351 if (dev[0] != '/' && stat64(path, &statbuf) != 0) {
5352 int error;
5353
5354 error = zfs_resolve_shortname(dev, path, MAXPATHLEN);
5355 if (error == 0 && zfs_dev_is_whole_disk(path)) {
5356 if (zfs_append_partition(path, MAXPATHLEN) == -1)
5357 error = ENOENT;
5358 }
5359
5360 if (error || (stat64(path, &statbuf) != 0)) {
5361 (void) printf("failed to find device %s, try "
5362 "specifying absolute path instead\n", dev);
5363 return (1);
5364 }
5365 }
5366
5367 if ((fd = open64(path, O_RDONLY)) < 0) {
5368 (void) printf("cannot open '%s': %s\n", path, strerror(errno));
5369 zdb_exit(1);
5370 }
5371
5372 if (fstat64_blk(fd, &statbuf) != 0) {
5373 (void) printf("failed to stat '%s': %s\n", path,
5374 strerror(errno));
5375 (void) close(fd);
5376 zdb_exit(1);
5377 }
5378
5379 if (S_ISBLK(statbuf.st_mode) && zfs_dev_flush(fd) != 0)
5380 (void) printf("failed to invalidate cache '%s' : %s\n", path,
5381 strerror(errno));
5382
5383 avl_create(&config_tree, cksum_record_compare,
5384 sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5385 avl_create(&uberblock_tree, cksum_record_compare,
5386 sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5387
5388 psize = statbuf.st_size;
5389 psize = P2ALIGN_TYPED(psize, sizeof (vdev_label_t), uint64_t);
5390 ashift = SPA_MINBLOCKSHIFT;
5391
5392 /*
5393 * 1. Read the label from disk
5394 * 2. Verify label cksum
5395 * 3. Unpack the configuration and insert in config tree.
5396 * 4. Traverse all uberblocks and insert in uberblock tree.
5397 */
5398 for (int l = 0; l < VDEV_LABELS; l++) {
5399 zdb_label_t *label = &labels[l];
5400 char *buf = label->label.vl_vdev_phys.vp_nvlist;
5401 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5402 nvlist_t *config;
5403 cksum_record_t *rec;
5404 zio_cksum_t cksum;
5405 vdev_t vd;
5406
5407 label->label_offset = vdev_label_offset(psize, l, 0);
5408
5409 if (pread64(fd, &label->label, sizeof (label->label),
5410 label->label_offset) != sizeof (label->label)) {
5411 if (!dump_opt['q'])
5412 (void) printf("failed to read label %d\n", l);
5413 label->read_failed = B_TRUE;
5414 error = B_TRUE;
5415 continue;
5416 }
5417
5418 label->read_failed = B_FALSE;
5419 label->cksum_valid = label_cksum_valid(&label->label,
5420 label->label_offset);
5421
5422 if (nvlist_unpack(buf, buflen, &config, 0) == 0) {
5423 nvlist_t *vdev_tree = NULL;
5424 size_t size;
5425
5426 if ((nvlist_lookup_nvlist(config,
5427 ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) ||
5428 (nvlist_lookup_uint64(vdev_tree,
5429 ZPOOL_CONFIG_ASHIFT, &ashift) != 0))
5430 ashift = SPA_MINBLOCKSHIFT;
5431
5432 if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0)
5433 size = buflen;
5434
5435 /* If the device is a cache device read the header. */
5436 if (!read_l2arc_header) {
5437 if (nvlist_lookup_uint64(config,
5438 ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 &&
5439 l2cache == POOL_STATE_L2CACHE) {
5440 read_l2arc_header = B_TRUE;
5441 }
5442 }
5443
5444 fletcher_4_native_varsize(buf, size, &cksum);
5445 rec = cksum_record_insert(&config_tree, &cksum, l);
5446
5447 label->config = rec;
5448 label->config_nv = config;
5449 config_found = B_TRUE;
5450 } else {
5451 error = B_TRUE;
5452 }
5453
5454 vd.vdev_ashift = ashift;
5455 vd.vdev_top = &vd;
5456
5457 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5458 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5459 uberblock_t *ub = (void *)((char *)label + uoff);
5460
5461 if (uberblock_verify(ub))
5462 continue;
5463
5464 fletcher_4_native_varsize(ub, sizeof (*ub), &cksum);
5465 rec = cksum_record_insert(&uberblock_tree, &cksum, l);
5466
5467 label->uberblocks[i] = rec;
5468 }
5469 }
5470
5471 /*
5472 * Dump the label and uberblocks.
5473 */
5474 for (int l = 0; l < VDEV_LABELS; l++) {
5475 zdb_label_t *label = &labels[l];
5476 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5477
5478 if (label->read_failed == B_TRUE)
5479 continue;
5480
5481 if (label->config_nv) {
5482 dump_config_from_label(label, buflen, l);
5483 } else {
5484 if (!dump_opt['q'])
5485 (void) printf("failed to unpack label %d\n", l);
5486 }
5487
5488 if (dump_opt['u'])
5489 dump_label_uberblocks(label, ashift, l);
5490
5491 nvlist_free(label->config_nv);
5492 }
5493
5494 /*
5495 * Dump the L2ARC header, if existent.
5496 */
5497 if (read_l2arc_header)
5498 error |= dump_l2arc_header(fd);
5499
5500 cookie = NULL;
5501 while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL)
5502 umem_free(node, sizeof (cksum_record_t));
5503
5504 cookie = NULL;
5505 while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL)
5506 umem_free(node, sizeof (cksum_record_t));
5507
5508 avl_destroy(&config_tree);
5509 avl_destroy(&uberblock_tree);
5510
5511 (void) close(fd);
5512
5513 return (config_found == B_FALSE ? 2 :
5514 (error == B_TRUE ? 1 : 0));
5515 }
5516
5517 static uint64_t dataset_feature_count[SPA_FEATURES];
5518 static uint64_t global_feature_count[SPA_FEATURES];
5519 static uint64_t remap_deadlist_count = 0;
5520
5521 static int
dump_one_objset(const char * dsname,void * arg)5522 dump_one_objset(const char *dsname, void *arg)
5523 {
5524 (void) arg;
5525 int error;
5526 objset_t *os;
5527 spa_feature_t f;
5528
5529 error = open_objset(dsname, FTAG, &os);
5530 if (error != 0)
5531 return (0);
5532
5533 for (f = 0; f < SPA_FEATURES; f++) {
5534 if (!dsl_dataset_feature_is_active(dmu_objset_ds(os), f))
5535 continue;
5536 ASSERT(spa_feature_table[f].fi_flags &
5537 ZFEATURE_FLAG_PER_DATASET);
5538 dataset_feature_count[f]++;
5539 }
5540
5541 if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os))) {
5542 remap_deadlist_count++;
5543 }
5544
5545 for (dsl_bookmark_node_t *dbn =
5546 avl_first(&dmu_objset_ds(os)->ds_bookmarks); dbn != NULL;
5547 dbn = AVL_NEXT(&dmu_objset_ds(os)->ds_bookmarks, dbn)) {
5548 mos_obj_refd(dbn->dbn_phys.zbm_redaction_obj);
5549 if (dbn->dbn_phys.zbm_redaction_obj != 0) {
5550 global_feature_count[
5551 SPA_FEATURE_REDACTION_BOOKMARKS]++;
5552 objset_t *mos = os->os_spa->spa_meta_objset;
5553 dnode_t *rl;
5554 VERIFY0(dnode_hold(mos,
5555 dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl));
5556 if (rl->dn_have_spill) {
5557 global_feature_count[
5558 SPA_FEATURE_REDACTION_LIST_SPILL]++;
5559 }
5560 }
5561 if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)
5562 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN]++;
5563 }
5564
5565 if (dsl_deadlist_is_open(&dmu_objset_ds(os)->ds_dir->dd_livelist) &&
5566 !dmu_objset_is_snapshot(os)) {
5567 global_feature_count[SPA_FEATURE_LIVELIST]++;
5568 }
5569
5570 dump_objset(os);
5571 close_objset(os, FTAG);
5572 fuid_table_destroy();
5573 return (0);
5574 }
5575
5576 /*
5577 * Block statistics.
5578 */
5579 #define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5580 typedef struct zdb_blkstats {
5581 uint64_t zb_asize;
5582 uint64_t zb_lsize;
5583 uint64_t zb_psize;
5584 uint64_t zb_count;
5585 uint64_t zb_gangs;
5586 uint64_t zb_ditto_samevdev;
5587 uint64_t zb_ditto_same_ms;
5588 uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
5589 } zdb_blkstats_t;
5590
5591 /*
5592 * Extended object types to report deferred frees and dedup auto-ditto blocks.
5593 */
5594 #define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0)
5595 #define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1)
5596 #define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2)
5597 #define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3)
5598
5599 static const char *zdb_ot_extname[] = {
5600 "deferred free",
5601 "dedup ditto",
5602 "other",
5603 "Total",
5604 };
5605
5606 #define ZB_TOTAL DN_MAX_LEVELS
5607 #define SPA_MAX_FOR_16M (SPA_MAXBLOCKSHIFT+1)
5608
5609 typedef struct zdb_brt_entry {
5610 dva_t zbre_dva;
5611 uint64_t zbre_refcount;
5612 avl_node_t zbre_node;
5613 } zdb_brt_entry_t;
5614
5615 typedef struct zdb_cb {
5616 zdb_blkstats_t zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
5617 uint64_t zcb_removing_size;
5618 uint64_t zcb_checkpoint_size;
5619 uint64_t zcb_dedup_asize;
5620 uint64_t zcb_dedup_blocks;
5621 uint64_t zcb_clone_asize;
5622 uint64_t zcb_clone_blocks;
5623 uint64_t zcb_psize_count[SPA_MAX_FOR_16M];
5624 uint64_t zcb_lsize_count[SPA_MAX_FOR_16M];
5625 uint64_t zcb_asize_count[SPA_MAX_FOR_16M];
5626 uint64_t zcb_psize_len[SPA_MAX_FOR_16M];
5627 uint64_t zcb_lsize_len[SPA_MAX_FOR_16M];
5628 uint64_t zcb_asize_len[SPA_MAX_FOR_16M];
5629 uint64_t zcb_psize_total;
5630 uint64_t zcb_lsize_total;
5631 uint64_t zcb_asize_total;
5632 uint64_t zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
5633 uint64_t zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
5634 [BPE_PAYLOAD_SIZE + 1];
5635 uint64_t zcb_start;
5636 hrtime_t zcb_lastprint;
5637 uint64_t zcb_totalasize;
5638 uint64_t zcb_errors[256];
5639 int zcb_readfails;
5640 int zcb_haderrors;
5641 spa_t *zcb_spa;
5642 uint32_t **zcb_vd_obsolete_counts;
5643 avl_tree_t zcb_brt;
5644 boolean_t zcb_brt_is_active;
5645 } zdb_cb_t;
5646
5647 /* test if two DVA offsets from same vdev are within the same metaslab */
5648 static boolean_t
same_metaslab(spa_t * spa,uint64_t vdev,uint64_t off1,uint64_t off2)5649 same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2)
5650 {
5651 vdev_t *vd = vdev_lookup_top(spa, vdev);
5652 uint64_t ms_shift = vd->vdev_ms_shift;
5653
5654 return ((off1 >> ms_shift) == (off2 >> ms_shift));
5655 }
5656
5657 /*
5658 * Used to simplify reporting of the histogram data.
5659 */
5660 typedef struct one_histo {
5661 const char *name;
5662 uint64_t *count;
5663 uint64_t *len;
5664 uint64_t cumulative;
5665 } one_histo_t;
5666
5667 /*
5668 * The number of separate histograms processed for psize, lsize and asize.
5669 */
5670 #define NUM_HISTO 3
5671
5672 /*
5673 * This routine will create a fixed column size output of three different
5674 * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5675 * the count, length and cumulative length of the psize, lsize and
5676 * asize blocks.
5677 *
5678 * All three types of blocks are listed on a single line
5679 *
5680 * By default the table is printed in nicenumber format (e.g. 123K) but
5681 * if the '-P' parameter is specified then the full raw number (parseable)
5682 * is printed out.
5683 */
5684 static void
dump_size_histograms(zdb_cb_t * zcb)5685 dump_size_histograms(zdb_cb_t *zcb)
5686 {
5687 /*
5688 * A temporary buffer that allows us to convert a number into
5689 * a string using zdb_nicenumber to allow either raw or human
5690 * readable numbers to be output.
5691 */
5692 char numbuf[32];
5693
5694 /*
5695 * Define titles which are used in the headers of the tables
5696 * printed by this routine.
5697 */
5698 const char blocksize_title1[] = "block";
5699 const char blocksize_title2[] = "size";
5700 const char count_title[] = "Count";
5701 const char length_title[] = "Size";
5702 const char cumulative_title[] = "Cum.";
5703
5704 /*
5705 * Setup the histogram arrays (psize, lsize, and asize).
5706 */
5707 one_histo_t parm_histo[NUM_HISTO];
5708
5709 parm_histo[0].name = "psize";
5710 parm_histo[0].count = zcb->zcb_psize_count;
5711 parm_histo[0].len = zcb->zcb_psize_len;
5712 parm_histo[0].cumulative = 0;
5713
5714 parm_histo[1].name = "lsize";
5715 parm_histo[1].count = zcb->zcb_lsize_count;
5716 parm_histo[1].len = zcb->zcb_lsize_len;
5717 parm_histo[1].cumulative = 0;
5718
5719 parm_histo[2].name = "asize";
5720 parm_histo[2].count = zcb->zcb_asize_count;
5721 parm_histo[2].len = zcb->zcb_asize_len;
5722 parm_histo[2].cumulative = 0;
5723
5724
5725 (void) printf("\nBlock Size Histogram\n");
5726 /*
5727 * Print the first line titles
5728 */
5729 if (dump_opt['P'])
5730 (void) printf("\n%s\t", blocksize_title1);
5731 else
5732 (void) printf("\n%7s ", blocksize_title1);
5733
5734 for (int j = 0; j < NUM_HISTO; j++) {
5735 if (dump_opt['P']) {
5736 if (j < NUM_HISTO - 1) {
5737 (void) printf("%s\t\t\t", parm_histo[j].name);
5738 } else {
5739 /* Don't print trailing spaces */
5740 (void) printf(" %s", parm_histo[j].name);
5741 }
5742 } else {
5743 if (j < NUM_HISTO - 1) {
5744 /* Left aligned strings in the output */
5745 (void) printf("%-7s ",
5746 parm_histo[j].name);
5747 } else {
5748 /* Don't print trailing spaces */
5749 (void) printf("%s", parm_histo[j].name);
5750 }
5751 }
5752 }
5753 (void) printf("\n");
5754
5755 /*
5756 * Print the second line titles
5757 */
5758 if (dump_opt['P']) {
5759 (void) printf("%s\t", blocksize_title2);
5760 } else {
5761 (void) printf("%7s ", blocksize_title2);
5762 }
5763
5764 for (int i = 0; i < NUM_HISTO; i++) {
5765 if (dump_opt['P']) {
5766 (void) printf("%s\t%s\t%s\t",
5767 count_title, length_title, cumulative_title);
5768 } else {
5769 (void) printf("%7s%7s%7s",
5770 count_title, length_title, cumulative_title);
5771 }
5772 }
5773 (void) printf("\n");
5774
5775 /*
5776 * Print the rows
5777 */
5778 for (int i = SPA_MINBLOCKSHIFT; i < SPA_MAX_FOR_16M; i++) {
5779
5780 /*
5781 * Print the first column showing the blocksize
5782 */
5783 zdb_nicenum((1ULL << i), numbuf, sizeof (numbuf));
5784
5785 if (dump_opt['P']) {
5786 printf("%s", numbuf);
5787 } else {
5788 printf("%7s:", numbuf);
5789 }
5790
5791 /*
5792 * Print the remaining set of 3 columns per size:
5793 * for psize, lsize and asize
5794 */
5795 for (int j = 0; j < NUM_HISTO; j++) {
5796 parm_histo[j].cumulative += parm_histo[j].len[i];
5797
5798 zdb_nicenum(parm_histo[j].count[i],
5799 numbuf, sizeof (numbuf));
5800 if (dump_opt['P'])
5801 (void) printf("\t%s", numbuf);
5802 else
5803 (void) printf("%7s", numbuf);
5804
5805 zdb_nicenum(parm_histo[j].len[i],
5806 numbuf, sizeof (numbuf));
5807 if (dump_opt['P'])
5808 (void) printf("\t%s", numbuf);
5809 else
5810 (void) printf("%7s", numbuf);
5811
5812 zdb_nicenum(parm_histo[j].cumulative,
5813 numbuf, sizeof (numbuf));
5814 if (dump_opt['P'])
5815 (void) printf("\t%s", numbuf);
5816 else
5817 (void) printf("%7s", numbuf);
5818 }
5819 (void) printf("\n");
5820 }
5821 }
5822
5823 static void
zdb_count_block(zdb_cb_t * zcb,zilog_t * zilog,const blkptr_t * bp,dmu_object_type_t type)5824 zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
5825 dmu_object_type_t type)
5826 {
5827 int i;
5828
5829 ASSERT(type < ZDB_OT_TOTAL);
5830
5831 if (zilog && zil_bp_tree_add(zilog, bp) != 0)
5832 return;
5833
5834 /*
5835 * This flag controls if we will issue a claim for the block while
5836 * counting it, to ensure that all blocks are referenced in space maps.
5837 * We don't issue claims if we're not doing leak tracking, because it's
5838 * expensive if the user isn't interested. We also don't claim the
5839 * second or later occurences of cloned or dedup'd blocks, because we
5840 * already claimed them the first time.
5841 */
5842 boolean_t do_claim = !dump_opt['L'];
5843
5844 spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
5845
5846 blkptr_t tempbp;
5847 if (BP_GET_DEDUP(bp)) {
5848 /*
5849 * Dedup'd blocks are special. We need to count them, so we can
5850 * later uncount them when reporting leaked space, and we must
5851 * only claim them once.
5852 *
5853 * We use the existing dedup system to track what we've seen.
5854 * The first time we see a block, we do a ddt_lookup() to see
5855 * if it exists in the DDT. If we're doing leak tracking, we
5856 * claim the block at this time.
5857 *
5858 * Each time we see a block, we reduce the refcount in the
5859 * entry by one, and add to the size and count of dedup'd
5860 * blocks to report at the end.
5861 */
5862
5863 ddt_t *ddt = ddt_select(zcb->zcb_spa, bp);
5864
5865 ddt_enter(ddt);
5866
5867 /*
5868 * Find the block. This will create the entry in memory, but
5869 * we'll know if that happened by its refcount.
5870 */
5871 ddt_entry_t *dde = ddt_lookup(ddt, bp, B_TRUE);
5872
5873 /*
5874 * ddt_lookup() can return NULL if this block didn't exist
5875 * in the DDT and creating it would take the DDT over its
5876 * quota. Since we got the block from disk, it must exist in
5877 * the DDT, so this can't happen. However, when unique entries
5878 * are pruned, the dedup bit can be set with no corresponding
5879 * entry in the DDT.
5880 */
5881 if (dde == NULL) {
5882 ddt_exit(ddt);
5883 goto skipped;
5884 }
5885
5886 /* Get the phys for this variant */
5887 ddt_phys_variant_t v = ddt_phys_select(ddt, dde, bp);
5888
5889 /*
5890 * This entry may have multiple sets of DVAs. We must claim
5891 * each set the first time we see them in a real block on disk,
5892 * or count them on subsequent occurences. We don't have a
5893 * convenient way to track the first time we see each variant,
5894 * so we repurpose dde_io as a set of "seen" flag bits. We can
5895 * do this safely in zdb because it never writes, so it will
5896 * never have a writing zio for this block in that pointer.
5897 */
5898 boolean_t seen = !!(((uintptr_t)dde->dde_io) & (1 << v));
5899 if (!seen)
5900 dde->dde_io =
5901 (void *)(((uintptr_t)dde->dde_io) | (1 << v));
5902
5903 /* Consume a reference for this block. */
5904 if (ddt_phys_total_refcnt(ddt, dde->dde_phys) > 0)
5905 ddt_phys_decref(dde->dde_phys, v);
5906
5907 /*
5908 * If this entry has a single flat phys, it may have been
5909 * extended with additional DVAs at some time in its life.
5910 * This block might be from before it was fully extended, and
5911 * so have fewer DVAs.
5912 *
5913 * If this is the first time we've seen this block, and we
5914 * claimed it as-is, then we would miss the claim on some
5915 * number of DVAs, which would then be seen as leaked.
5916 *
5917 * In all cases, if we've had fewer DVAs, then the asize would
5918 * be too small, and would lead to the pool apparently using
5919 * more space than allocated.
5920 *
5921 * To handle this, we copy the canonical set of DVAs from the
5922 * entry back to the block pointer before we claim it.
5923 */
5924 if (v == DDT_PHYS_FLAT) {
5925 ASSERT3U(BP_GET_PHYSICAL_BIRTH(bp), ==,
5926 ddt_phys_birth(dde->dde_phys, v));
5927 tempbp = *bp;
5928 ddt_bp_fill(dde->dde_phys, v, &tempbp,
5929 BP_GET_PHYSICAL_BIRTH(bp));
5930 bp = &tempbp;
5931 }
5932
5933 if (seen) {
5934 /*
5935 * The second or later time we see this block,
5936 * it's a duplicate and we count it.
5937 */
5938 zcb->zcb_dedup_asize += BP_GET_ASIZE(bp);
5939 zcb->zcb_dedup_blocks++;
5940
5941 /* Already claimed, don't do it again. */
5942 do_claim = B_FALSE;
5943 }
5944
5945 ddt_exit(ddt);
5946 } else if (zcb->zcb_brt_is_active &&
5947 brt_maybe_exists(zcb->zcb_spa, bp)) {
5948 /*
5949 * Cloned blocks are special. We need to count them, so we can
5950 * later uncount them when reporting leaked space, and we must
5951 * only claim them once.
5952 *
5953 * To do this, we keep our own in-memory BRT. For each block
5954 * we haven't seen before, we look it up in the real BRT and
5955 * if its there, we note it and its refcount then proceed as
5956 * normal. If we see the block again, we count it as a clone
5957 * and then give it no further consideration.
5958 */
5959 zdb_brt_entry_t zbre_search, *zbre;
5960 avl_index_t where;
5961
5962 zbre_search.zbre_dva = bp->blk_dva[0];
5963 zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where);
5964 if (zbre == NULL) {
5965 /* Not seen before; track it */
5966 uint64_t refcnt =
5967 brt_entry_get_refcount(zcb->zcb_spa, bp);
5968 if (refcnt > 0) {
5969 zbre = umem_zalloc(sizeof (zdb_brt_entry_t),
5970 UMEM_NOFAIL);
5971 zbre->zbre_dva = bp->blk_dva[0];
5972 zbre->zbre_refcount = refcnt;
5973 avl_insert(&zcb->zcb_brt, zbre, where);
5974 }
5975 } else {
5976 /*
5977 * Second or later occurrence, count it and take a
5978 * refcount.
5979 */
5980 zcb->zcb_clone_asize += BP_GET_ASIZE(bp);
5981 zcb->zcb_clone_blocks++;
5982
5983 zbre->zbre_refcount--;
5984 if (zbre->zbre_refcount == 0) {
5985 avl_remove(&zcb->zcb_brt, zbre);
5986 umem_free(zbre, sizeof (zdb_brt_entry_t));
5987 }
5988
5989 /* Already claimed, don't do it again. */
5990 do_claim = B_FALSE;
5991 }
5992 }
5993
5994 skipped:
5995 for (i = 0; i < 4; i++) {
5996 int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
5997 int t = (i & 1) ? type : ZDB_OT_TOTAL;
5998 int equal;
5999 zdb_blkstats_t *zb = &zcb->zcb_type[l][t];
6000
6001 zb->zb_asize += BP_GET_ASIZE(bp);
6002 zb->zb_lsize += BP_GET_LSIZE(bp);
6003 zb->zb_psize += BP_GET_PSIZE(bp);
6004 zb->zb_count++;
6005
6006 /*
6007 * The histogram is only big enough to record blocks up to
6008 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
6009 * "other", bucket.
6010 */
6011 unsigned idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT;
6012 idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1);
6013 zb->zb_psize_histogram[idx]++;
6014
6015 zb->zb_gangs += BP_COUNT_GANG(bp);
6016
6017 switch (BP_GET_NDVAS(bp)) {
6018 case 2:
6019 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6020 DVA_GET_VDEV(&bp->blk_dva[1])) {
6021 zb->zb_ditto_samevdev++;
6022
6023 if (same_metaslab(zcb->zcb_spa,
6024 DVA_GET_VDEV(&bp->blk_dva[0]),
6025 DVA_GET_OFFSET(&bp->blk_dva[0]),
6026 DVA_GET_OFFSET(&bp->blk_dva[1])))
6027 zb->zb_ditto_same_ms++;
6028 }
6029 break;
6030 case 3:
6031 equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6032 DVA_GET_VDEV(&bp->blk_dva[1])) +
6033 (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6034 DVA_GET_VDEV(&bp->blk_dva[2])) +
6035 (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6036 DVA_GET_VDEV(&bp->blk_dva[2]));
6037 if (equal != 0) {
6038 zb->zb_ditto_samevdev++;
6039
6040 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6041 DVA_GET_VDEV(&bp->blk_dva[1]) &&
6042 same_metaslab(zcb->zcb_spa,
6043 DVA_GET_VDEV(&bp->blk_dva[0]),
6044 DVA_GET_OFFSET(&bp->blk_dva[0]),
6045 DVA_GET_OFFSET(&bp->blk_dva[1])))
6046 zb->zb_ditto_same_ms++;
6047 else if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6048 DVA_GET_VDEV(&bp->blk_dva[2]) &&
6049 same_metaslab(zcb->zcb_spa,
6050 DVA_GET_VDEV(&bp->blk_dva[0]),
6051 DVA_GET_OFFSET(&bp->blk_dva[0]),
6052 DVA_GET_OFFSET(&bp->blk_dva[2])))
6053 zb->zb_ditto_same_ms++;
6054 else if (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6055 DVA_GET_VDEV(&bp->blk_dva[2]) &&
6056 same_metaslab(zcb->zcb_spa,
6057 DVA_GET_VDEV(&bp->blk_dva[1]),
6058 DVA_GET_OFFSET(&bp->blk_dva[1]),
6059 DVA_GET_OFFSET(&bp->blk_dva[2])))
6060 zb->zb_ditto_same_ms++;
6061 }
6062 break;
6063 }
6064 }
6065
6066 spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
6067
6068 if (BP_IS_EMBEDDED(bp)) {
6069 zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
6070 zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
6071 [BPE_GET_PSIZE(bp)]++;
6072 return;
6073 }
6074 /*
6075 * The binning histogram bins by powers of two up to
6076 * SPA_MAXBLOCKSIZE rather than creating bins for
6077 * every possible blocksize found in the pool.
6078 */
6079 int bin = highbit64(BP_GET_PSIZE(bp)) - 1;
6080
6081 zcb->zcb_psize_count[bin]++;
6082 zcb->zcb_psize_len[bin] += BP_GET_PSIZE(bp);
6083 zcb->zcb_psize_total += BP_GET_PSIZE(bp);
6084
6085 bin = highbit64(BP_GET_LSIZE(bp)) - 1;
6086
6087 zcb->zcb_lsize_count[bin]++;
6088 zcb->zcb_lsize_len[bin] += BP_GET_LSIZE(bp);
6089 zcb->zcb_lsize_total += BP_GET_LSIZE(bp);
6090
6091 bin = highbit64(BP_GET_ASIZE(bp)) - 1;
6092
6093 zcb->zcb_asize_count[bin]++;
6094 zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp);
6095 zcb->zcb_asize_total += BP_GET_ASIZE(bp);
6096
6097 if (!do_claim)
6098 return;
6099
6100 VERIFY0(zio_wait(zio_claim(NULL, zcb->zcb_spa,
6101 spa_min_claim_txg(zcb->zcb_spa), bp, NULL, NULL,
6102 ZIO_FLAG_CANFAIL)));
6103 }
6104
6105 static void
zdb_blkptr_done(zio_t * zio)6106 zdb_blkptr_done(zio_t *zio)
6107 {
6108 spa_t *spa = zio->io_spa;
6109 blkptr_t *bp = zio->io_bp;
6110 int ioerr = zio->io_error;
6111 zdb_cb_t *zcb = zio->io_private;
6112 zbookmark_phys_t *zb = &zio->io_bookmark;
6113
6114 mutex_enter(&spa->spa_scrub_lock);
6115 spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
6116 cv_broadcast(&spa->spa_scrub_io_cv);
6117
6118 if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
6119 char blkbuf[BP_SPRINTF_LEN];
6120
6121 zcb->zcb_haderrors = 1;
6122 zcb->zcb_errors[ioerr]++;
6123
6124 if (dump_opt['b'] >= 2)
6125 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6126 else
6127 blkbuf[0] = '\0';
6128
6129 (void) printf("zdb_blkptr_cb: "
6130 "Got error %d reading "
6131 "<%llu, %llu, %lld, %llx> %s -- skipping\n",
6132 ioerr,
6133 (u_longlong_t)zb->zb_objset,
6134 (u_longlong_t)zb->zb_object,
6135 (u_longlong_t)zb->zb_level,
6136 (u_longlong_t)zb->zb_blkid,
6137 blkbuf);
6138 }
6139 mutex_exit(&spa->spa_scrub_lock);
6140
6141 abd_free(zio->io_abd);
6142 }
6143
6144 static int
zdb_blkptr_cb(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)6145 zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
6146 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
6147 {
6148 zdb_cb_t *zcb = arg;
6149 dmu_object_type_t type;
6150 boolean_t is_metadata;
6151
6152 if (zb->zb_level == ZB_DNODE_LEVEL)
6153 return (0);
6154
6155 if (dump_opt['b'] >= 5 && BP_GET_BIRTH(bp) > 0) {
6156 char blkbuf[BP_SPRINTF_LEN];
6157 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6158 (void) printf("objset %llu object %llu "
6159 "level %lld offset 0x%llx %s\n",
6160 (u_longlong_t)zb->zb_objset,
6161 (u_longlong_t)zb->zb_object,
6162 (longlong_t)zb->zb_level,
6163 (u_longlong_t)blkid2offset(dnp, bp, zb),
6164 blkbuf);
6165 }
6166
6167 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
6168 return (0);
6169
6170 type = BP_GET_TYPE(bp);
6171
6172 zdb_count_block(zcb, zilog, bp,
6173 (type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type);
6174
6175 is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
6176
6177 if (!BP_IS_EMBEDDED(bp) &&
6178 (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
6179 size_t size = BP_GET_PSIZE(bp);
6180 abd_t *abd = abd_alloc(size, B_FALSE);
6181 int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
6182
6183 /* If it's an intent log block, failure is expected. */
6184 if (zb->zb_level == ZB_ZIL_LEVEL)
6185 flags |= ZIO_FLAG_SPECULATIVE;
6186
6187 mutex_enter(&spa->spa_scrub_lock);
6188 while (spa->spa_load_verify_bytes > max_inflight_bytes)
6189 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
6190 spa->spa_load_verify_bytes += size;
6191 mutex_exit(&spa->spa_scrub_lock);
6192
6193 zio_nowait(zio_read(NULL, spa, bp, abd, size,
6194 zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
6195 }
6196
6197 zcb->zcb_readfails = 0;
6198
6199 /* only call gethrtime() every 100 blocks */
6200 static int iters;
6201 if (++iters > 100)
6202 iters = 0;
6203 else
6204 return (0);
6205
6206 if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
6207 uint64_t now = gethrtime();
6208 char buf[10];
6209 uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
6210 uint64_t kb_per_sec =
6211 1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000));
6212 uint64_t sec_remaining =
6213 (zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec;
6214
6215 /* make sure nicenum has enough space */
6216 _Static_assert(sizeof (buf) >= NN_NUMBUF_SZ, "buf truncated");
6217
6218 zfs_nicebytes(bytes, buf, sizeof (buf));
6219 (void) fprintf(stderr,
6220 "\r%5s completed (%4"PRIu64"MB/s) "
6221 "estimated time remaining: "
6222 "%"PRIu64"hr %02"PRIu64"min %02"PRIu64"sec ",
6223 buf, kb_per_sec / 1024,
6224 sec_remaining / 60 / 60,
6225 sec_remaining / 60 % 60,
6226 sec_remaining % 60);
6227
6228 zcb->zcb_lastprint = now;
6229 }
6230
6231 return (0);
6232 }
6233
6234 static void
zdb_leak(void * arg,uint64_t start,uint64_t size)6235 zdb_leak(void *arg, uint64_t start, uint64_t size)
6236 {
6237 vdev_t *vd = arg;
6238
6239 (void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
6240 (u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
6241 }
6242
6243 static metaslab_ops_t zdb_metaslab_ops = {
6244 NULL /* alloc */
6245 };
6246
6247 static int
load_unflushed_svr_segs_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6248 load_unflushed_svr_segs_cb(spa_t *spa, space_map_entry_t *sme,
6249 uint64_t txg, void *arg)
6250 {
6251 spa_vdev_removal_t *svr = arg;
6252
6253 uint64_t offset = sme->sme_offset;
6254 uint64_t size = sme->sme_run;
6255
6256 /* skip vdevs we don't care about */
6257 if (sme->sme_vdev != svr->svr_vdev_id)
6258 return (0);
6259
6260 vdev_t *vd = vdev_lookup_top(spa, sme->sme_vdev);
6261 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6262 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6263
6264 if (txg < metaslab_unflushed_txg(ms))
6265 return (0);
6266
6267 if (sme->sme_type == SM_ALLOC)
6268 zfs_range_tree_add(svr->svr_allocd_segs, offset, size);
6269 else
6270 zfs_range_tree_remove(svr->svr_allocd_segs, offset, size);
6271
6272 return (0);
6273 }
6274
6275 static void
claim_segment_impl_cb(uint64_t inner_offset,vdev_t * vd,uint64_t offset,uint64_t size,void * arg)6276 claim_segment_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
6277 uint64_t size, void *arg)
6278 {
6279 (void) inner_offset, (void) arg;
6280
6281 /*
6282 * This callback was called through a remap from
6283 * a device being removed. Therefore, the vdev that
6284 * this callback is applied to is a concrete
6285 * vdev.
6286 */
6287 ASSERT(vdev_is_concrete(vd));
6288
6289 VERIFY0(metaslab_claim_impl(vd, offset, size,
6290 spa_min_claim_txg(vd->vdev_spa)));
6291 }
6292
6293 static void
claim_segment_cb(void * arg,uint64_t offset,uint64_t size)6294 claim_segment_cb(void *arg, uint64_t offset, uint64_t size)
6295 {
6296 vdev_t *vd = arg;
6297
6298 vdev_indirect_ops.vdev_op_remap(vd, offset, size,
6299 claim_segment_impl_cb, NULL);
6300 }
6301
6302 /*
6303 * After accounting for all allocated blocks that are directly referenced,
6304 * we might have missed a reference to a block from a partially complete
6305 * (and thus unused) indirect mapping object. We perform a secondary pass
6306 * through the metaslabs we have already mapped and claim the destination
6307 * blocks.
6308 */
6309 static void
zdb_claim_removing(spa_t * spa,zdb_cb_t * zcb)6310 zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb)
6311 {
6312 if (dump_opt['L'])
6313 return;
6314
6315 if (spa->spa_vdev_removal == NULL)
6316 return;
6317
6318 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6319
6320 spa_vdev_removal_t *svr = spa->spa_vdev_removal;
6321 vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id);
6322 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6323
6324 ASSERT0(zfs_range_tree_space(svr->svr_allocd_segs));
6325
6326 zfs_range_tree_t *allocs = zfs_range_tree_create_flags(
6327 NULL, ZFS_RANGE_SEG64, NULL, 0, 0,
6328 0, "zdb_claim_removing:allocs");
6329 for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) {
6330 metaslab_t *msp = vd->vdev_ms[msi];
6331
6332 ASSERT0(zfs_range_tree_space(allocs));
6333 if (msp->ms_sm != NULL)
6334 VERIFY0(space_map_load(msp->ms_sm, allocs, SM_ALLOC));
6335 zfs_range_tree_vacate(allocs, zfs_range_tree_add,
6336 svr->svr_allocd_segs);
6337 }
6338 zfs_range_tree_destroy(allocs);
6339
6340 iterate_through_spacemap_logs(spa, load_unflushed_svr_segs_cb, svr);
6341
6342 /*
6343 * Clear everything past what has been synced,
6344 * because we have not allocated mappings for
6345 * it yet.
6346 */
6347 zfs_range_tree_clear(svr->svr_allocd_segs,
6348 vdev_indirect_mapping_max_offset(vim),
6349 vd->vdev_asize - vdev_indirect_mapping_max_offset(vim));
6350
6351 zcb->zcb_removing_size += zfs_range_tree_space(svr->svr_allocd_segs);
6352 zfs_range_tree_vacate(svr->svr_allocd_segs, claim_segment_cb, vd);
6353
6354 spa_config_exit(spa, SCL_CONFIG, FTAG);
6355 }
6356
6357 static int
increment_indirect_mapping_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6358 increment_indirect_mapping_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6359 dmu_tx_t *tx)
6360 {
6361 (void) tx;
6362 zdb_cb_t *zcb = arg;
6363 spa_t *spa = zcb->zcb_spa;
6364 vdev_t *vd;
6365 const dva_t *dva = &bp->blk_dva[0];
6366
6367 ASSERT(!bp_freed);
6368 ASSERT(!dump_opt['L']);
6369 ASSERT3U(BP_GET_NDVAS(bp), ==, 1);
6370
6371 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6372 vd = vdev_lookup_top(zcb->zcb_spa, DVA_GET_VDEV(dva));
6373 ASSERT3P(vd, !=, NULL);
6374 spa_config_exit(spa, SCL_VDEV, FTAG);
6375
6376 ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0);
6377 ASSERT3P(zcb->zcb_vd_obsolete_counts[vd->vdev_id], !=, NULL);
6378
6379 vdev_indirect_mapping_increment_obsolete_count(
6380 vd->vdev_indirect_mapping,
6381 DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva),
6382 zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6383
6384 return (0);
6385 }
6386
6387 static uint32_t *
zdb_load_obsolete_counts(vdev_t * vd)6388 zdb_load_obsolete_counts(vdev_t *vd)
6389 {
6390 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6391 spa_t *spa = vd->vdev_spa;
6392 spa_condensing_indirect_phys_t *scip =
6393 &spa->spa_condensing_indirect_phys;
6394 uint64_t obsolete_sm_object;
6395 uint32_t *counts;
6396
6397 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
6398 EQUIV(obsolete_sm_object != 0, vd->vdev_obsolete_sm != NULL);
6399 counts = vdev_indirect_mapping_load_obsolete_counts(vim);
6400 if (vd->vdev_obsolete_sm != NULL) {
6401 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6402 vd->vdev_obsolete_sm);
6403 }
6404 if (scip->scip_vdev == vd->vdev_id &&
6405 scip->scip_prev_obsolete_sm_object != 0) {
6406 space_map_t *prev_obsolete_sm = NULL;
6407 VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset,
6408 scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0));
6409 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6410 prev_obsolete_sm);
6411 space_map_close(prev_obsolete_sm);
6412 }
6413 return (counts);
6414 }
6415
6416 typedef struct checkpoint_sm_exclude_entry_arg {
6417 vdev_t *cseea_vd;
6418 uint64_t cseea_checkpoint_size;
6419 } checkpoint_sm_exclude_entry_arg_t;
6420
6421 static int
checkpoint_sm_exclude_entry_cb(space_map_entry_t * sme,void * arg)6422 checkpoint_sm_exclude_entry_cb(space_map_entry_t *sme, void *arg)
6423 {
6424 checkpoint_sm_exclude_entry_arg_t *cseea = arg;
6425 vdev_t *vd = cseea->cseea_vd;
6426 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
6427 uint64_t end = sme->sme_offset + sme->sme_run;
6428
6429 ASSERT(sme->sme_type == SM_FREE);
6430
6431 /*
6432 * Since the vdev_checkpoint_sm exists in the vdev level
6433 * and the ms_sm space maps exist in the metaslab level,
6434 * an entry in the checkpoint space map could theoretically
6435 * cross the boundaries of the metaslab that it belongs.
6436 *
6437 * In reality, because of the way that we populate and
6438 * manipulate the checkpoint's space maps currently,
6439 * there shouldn't be any entries that cross metaslabs.
6440 * Hence the assertion below.
6441 *
6442 * That said, there is no fundamental requirement that
6443 * the checkpoint's space map entries should not cross
6444 * metaslab boundaries. So if needed we could add code
6445 * that handles metaslab-crossing segments in the future.
6446 */
6447 VERIFY3U(sme->sme_offset, >=, ms->ms_start);
6448 VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
6449
6450 /*
6451 * By removing the entry from the allocated segments we
6452 * also verify that the entry is there to begin with.
6453 */
6454 mutex_enter(&ms->ms_lock);
6455 zfs_range_tree_remove(ms->ms_allocatable, sme->sme_offset,
6456 sme->sme_run);
6457 mutex_exit(&ms->ms_lock);
6458
6459 cseea->cseea_checkpoint_size += sme->sme_run;
6460 return (0);
6461 }
6462
6463 static void
zdb_leak_init_vdev_exclude_checkpoint(vdev_t * vd,zdb_cb_t * zcb)6464 zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_cb_t *zcb)
6465 {
6466 spa_t *spa = vd->vdev_spa;
6467 space_map_t *checkpoint_sm = NULL;
6468 uint64_t checkpoint_sm_obj;
6469
6470 /*
6471 * If there is no vdev_top_zap, we are in a pool whose
6472 * version predates the pool checkpoint feature.
6473 */
6474 if (vd->vdev_top_zap == 0)
6475 return;
6476
6477 /*
6478 * If there is no reference of the vdev_checkpoint_sm in
6479 * the vdev_top_zap, then one of the following scenarios
6480 * is true:
6481 *
6482 * 1] There is no checkpoint
6483 * 2] There is a checkpoint, but no checkpointed blocks
6484 * have been freed yet
6485 * 3] The current vdev is indirect
6486 *
6487 * In these cases we return immediately.
6488 */
6489 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
6490 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
6491 return;
6492
6493 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
6494 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1,
6495 &checkpoint_sm_obj));
6496
6497 checkpoint_sm_exclude_entry_arg_t cseea;
6498 cseea.cseea_vd = vd;
6499 cseea.cseea_checkpoint_size = 0;
6500
6501 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
6502 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
6503
6504 VERIFY0(space_map_iterate(checkpoint_sm,
6505 space_map_length(checkpoint_sm),
6506 checkpoint_sm_exclude_entry_cb, &cseea));
6507 space_map_close(checkpoint_sm);
6508
6509 zcb->zcb_checkpoint_size += cseea.cseea_checkpoint_size;
6510 }
6511
6512 static void
zdb_leak_init_exclude_checkpoint(spa_t * spa,zdb_cb_t * zcb)6513 zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb)
6514 {
6515 ASSERT(!dump_opt['L']);
6516
6517 vdev_t *rvd = spa->spa_root_vdev;
6518 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6519 ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id);
6520 zdb_leak_init_vdev_exclude_checkpoint(rvd->vdev_child[c], zcb);
6521 }
6522 }
6523
6524 static int
count_unflushed_space_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6525 count_unflushed_space_cb(spa_t *spa, space_map_entry_t *sme,
6526 uint64_t txg, void *arg)
6527 {
6528 int64_t *ualloc_space = arg;
6529
6530 uint64_t offset = sme->sme_offset;
6531 uint64_t vdev_id = sme->sme_vdev;
6532
6533 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6534 if (!vdev_is_concrete(vd))
6535 return (0);
6536
6537 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6538 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6539
6540 if (txg < metaslab_unflushed_txg(ms))
6541 return (0);
6542
6543 if (sme->sme_type == SM_ALLOC)
6544 *ualloc_space += sme->sme_run;
6545 else
6546 *ualloc_space -= sme->sme_run;
6547
6548 return (0);
6549 }
6550
6551 static int64_t
get_unflushed_alloc_space(spa_t * spa)6552 get_unflushed_alloc_space(spa_t *spa)
6553 {
6554 if (dump_opt['L'])
6555 return (0);
6556
6557 int64_t ualloc_space = 0;
6558 iterate_through_spacemap_logs(spa, count_unflushed_space_cb,
6559 &ualloc_space);
6560 return (ualloc_space);
6561 }
6562
6563 static int
load_unflushed_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6564 load_unflushed_cb(spa_t *spa, space_map_entry_t *sme, uint64_t txg, void *arg)
6565 {
6566 maptype_t *uic_maptype = arg;
6567
6568 uint64_t offset = sme->sme_offset;
6569 uint64_t size = sme->sme_run;
6570 uint64_t vdev_id = sme->sme_vdev;
6571
6572 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6573
6574 /* skip indirect vdevs */
6575 if (!vdev_is_concrete(vd))
6576 return (0);
6577
6578 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6579
6580 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6581 ASSERT(*uic_maptype == SM_ALLOC || *uic_maptype == SM_FREE);
6582
6583 if (txg < metaslab_unflushed_txg(ms))
6584 return (0);
6585
6586 if (*uic_maptype == sme->sme_type)
6587 zfs_range_tree_add(ms->ms_allocatable, offset, size);
6588 else
6589 zfs_range_tree_remove(ms->ms_allocatable, offset, size);
6590
6591 return (0);
6592 }
6593
6594 static void
load_unflushed_to_ms_allocatables(spa_t * spa,maptype_t maptype)6595 load_unflushed_to_ms_allocatables(spa_t *spa, maptype_t maptype)
6596 {
6597 iterate_through_spacemap_logs(spa, load_unflushed_cb, &maptype);
6598 }
6599
6600 static void
load_concrete_ms_allocatable_trees(spa_t * spa,maptype_t maptype)6601 load_concrete_ms_allocatable_trees(spa_t *spa, maptype_t maptype)
6602 {
6603 vdev_t *rvd = spa->spa_root_vdev;
6604 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
6605 vdev_t *vd = rvd->vdev_child[i];
6606
6607 ASSERT3U(i, ==, vd->vdev_id);
6608
6609 if (vd->vdev_ops == &vdev_indirect_ops)
6610 continue;
6611
6612 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6613 metaslab_t *msp = vd->vdev_ms[m];
6614
6615 (void) fprintf(stderr,
6616 "\rloading concrete vdev %llu, "
6617 "metaslab %llu of %llu ...",
6618 (longlong_t)vd->vdev_id,
6619 (longlong_t)msp->ms_id,
6620 (longlong_t)vd->vdev_ms_count);
6621
6622 mutex_enter(&msp->ms_lock);
6623 zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6624
6625 /*
6626 * We don't want to spend the CPU manipulating the
6627 * size-ordered tree, so clear the range_tree ops.
6628 */
6629 msp->ms_allocatable->rt_ops = NULL;
6630
6631 if (msp->ms_sm != NULL) {
6632 VERIFY0(space_map_load(msp->ms_sm,
6633 msp->ms_allocatable, maptype));
6634 }
6635 if (!msp->ms_loaded)
6636 msp->ms_loaded = B_TRUE;
6637 mutex_exit(&msp->ms_lock);
6638 }
6639 }
6640
6641 load_unflushed_to_ms_allocatables(spa, maptype);
6642 }
6643
6644 /*
6645 * vm_idxp is an in-out parameter which (for indirect vdevs) is the
6646 * index in vim_entries that has the first entry in this metaslab.
6647 * On return, it will be set to the first entry after this metaslab.
6648 */
6649 static void
load_indirect_ms_allocatable_tree(vdev_t * vd,metaslab_t * msp,uint64_t * vim_idxp)6650 load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab_t *msp,
6651 uint64_t *vim_idxp)
6652 {
6653 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6654
6655 mutex_enter(&msp->ms_lock);
6656 zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6657
6658 /*
6659 * We don't want to spend the CPU manipulating the
6660 * size-ordered tree, so clear the range_tree ops.
6661 */
6662 msp->ms_allocatable->rt_ops = NULL;
6663
6664 for (; *vim_idxp < vdev_indirect_mapping_num_entries(vim);
6665 (*vim_idxp)++) {
6666 vdev_indirect_mapping_entry_phys_t *vimep =
6667 &vim->vim_entries[*vim_idxp];
6668 uint64_t ent_offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6669 uint64_t ent_len = DVA_GET_ASIZE(&vimep->vimep_dst);
6670 ASSERT3U(ent_offset, >=, msp->ms_start);
6671 if (ent_offset >= msp->ms_start + msp->ms_size)
6672 break;
6673
6674 /*
6675 * Mappings do not cross metaslab boundaries,
6676 * because we create them by walking the metaslabs.
6677 */
6678 ASSERT3U(ent_offset + ent_len, <=,
6679 msp->ms_start + msp->ms_size);
6680 zfs_range_tree_add(msp->ms_allocatable, ent_offset, ent_len);
6681 }
6682
6683 if (!msp->ms_loaded)
6684 msp->ms_loaded = B_TRUE;
6685 mutex_exit(&msp->ms_lock);
6686 }
6687
6688 static void
zdb_leak_init_prepare_indirect_vdevs(spa_t * spa,zdb_cb_t * zcb)6689 zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb)
6690 {
6691 ASSERT(!dump_opt['L']);
6692
6693 vdev_t *rvd = spa->spa_root_vdev;
6694 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6695 vdev_t *vd = rvd->vdev_child[c];
6696
6697 ASSERT3U(c, ==, vd->vdev_id);
6698
6699 if (vd->vdev_ops != &vdev_indirect_ops)
6700 continue;
6701
6702 /*
6703 * Note: we don't check for mapping leaks on
6704 * removing vdevs because their ms_allocatable's
6705 * are used to look for leaks in allocated space.
6706 */
6707 zcb->zcb_vd_obsolete_counts[c] = zdb_load_obsolete_counts(vd);
6708
6709 /*
6710 * Normally, indirect vdevs don't have any
6711 * metaslabs. We want to set them up for
6712 * zio_claim().
6713 */
6714 vdev_metaslab_group_create(vd);
6715 VERIFY0(vdev_metaslab_init(vd, 0));
6716
6717 vdev_indirect_mapping_t *vim __maybe_unused =
6718 vd->vdev_indirect_mapping;
6719 uint64_t vim_idx = 0;
6720 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6721
6722 (void) fprintf(stderr,
6723 "\rloading indirect vdev %llu, "
6724 "metaslab %llu of %llu ...",
6725 (longlong_t)vd->vdev_id,
6726 (longlong_t)vd->vdev_ms[m]->ms_id,
6727 (longlong_t)vd->vdev_ms_count);
6728
6729 load_indirect_ms_allocatable_tree(vd, vd->vdev_ms[m],
6730 &vim_idx);
6731 }
6732 ASSERT3U(vim_idx, ==, vdev_indirect_mapping_num_entries(vim));
6733 }
6734 }
6735
6736 static void
zdb_leak_init(spa_t * spa,zdb_cb_t * zcb)6737 zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
6738 {
6739 zcb->zcb_spa = spa;
6740
6741 if (dump_opt['L'])
6742 return;
6743
6744 dsl_pool_t *dp = spa->spa_dsl_pool;
6745 vdev_t *rvd = spa->spa_root_vdev;
6746
6747 /*
6748 * We are going to be changing the meaning of the metaslab's
6749 * ms_allocatable. Ensure that the allocator doesn't try to
6750 * use the tree.
6751 */
6752 spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
6753 spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
6754 spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6755 spa->spa_special_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6756
6757 zcb->zcb_vd_obsolete_counts =
6758 umem_zalloc(rvd->vdev_children * sizeof (uint32_t *),
6759 UMEM_NOFAIL);
6760
6761 /*
6762 * For leak detection, we overload the ms_allocatable trees
6763 * to contain allocated segments instead of free segments.
6764 * As a result, we can't use the normal metaslab_load/unload
6765 * interfaces.
6766 */
6767 zdb_leak_init_prepare_indirect_vdevs(spa, zcb);
6768 load_concrete_ms_allocatable_trees(spa, SM_ALLOC);
6769
6770 /*
6771 * On load_concrete_ms_allocatable_trees() we loaded all the
6772 * allocated entries from the ms_sm to the ms_allocatable for
6773 * each metaslab. If the pool has a checkpoint or is in the
6774 * middle of discarding a checkpoint, some of these blocks
6775 * may have been freed but their ms_sm may not have been
6776 * updated because they are referenced by the checkpoint. In
6777 * order to avoid false-positives during leak-detection, we
6778 * go through the vdev's checkpoint space map and exclude all
6779 * its entries from their relevant ms_allocatable.
6780 *
6781 * We also aggregate the space held by the checkpoint and add
6782 * it to zcb_checkpoint_size.
6783 *
6784 * Note that at this point we are also verifying that all the
6785 * entries on the checkpoint_sm are marked as allocated in
6786 * the ms_sm of their relevant metaslab.
6787 * [see comment in checkpoint_sm_exclude_entry_cb()]
6788 */
6789 zdb_leak_init_exclude_checkpoint(spa, zcb);
6790 ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa));
6791
6792 /* for cleaner progress output */
6793 (void) fprintf(stderr, "\n");
6794
6795 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
6796 ASSERT(spa_feature_is_enabled(spa,
6797 SPA_FEATURE_DEVICE_REMOVAL));
6798 (void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj,
6799 increment_indirect_mapping_cb, zcb, NULL);
6800 }
6801 }
6802
6803 static boolean_t
zdb_check_for_obsolete_leaks(vdev_t * vd,zdb_cb_t * zcb)6804 zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb)
6805 {
6806 boolean_t leaks = B_FALSE;
6807 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6808 uint64_t total_leaked = 0;
6809 boolean_t are_precise = B_FALSE;
6810
6811 ASSERT(vim != NULL);
6812
6813 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
6814 vdev_indirect_mapping_entry_phys_t *vimep =
6815 &vim->vim_entries[i];
6816 uint64_t obsolete_bytes = 0;
6817 uint64_t offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6818 metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6819
6820 /*
6821 * This is not very efficient but it's easy to
6822 * verify correctness.
6823 */
6824 for (uint64_t inner_offset = 0;
6825 inner_offset < DVA_GET_ASIZE(&vimep->vimep_dst);
6826 inner_offset += 1ULL << vd->vdev_ashift) {
6827 if (zfs_range_tree_contains(msp->ms_allocatable,
6828 offset + inner_offset, 1ULL << vd->vdev_ashift)) {
6829 obsolete_bytes += 1ULL << vd->vdev_ashift;
6830 }
6831 }
6832
6833 int64_t bytes_leaked = obsolete_bytes -
6834 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i];
6835 ASSERT3U(DVA_GET_ASIZE(&vimep->vimep_dst), >=,
6836 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]);
6837
6838 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6839 if (bytes_leaked != 0 && (are_precise || dump_opt['d'] >= 5)) {
6840 (void) printf("obsolete indirect mapping count "
6841 "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
6842 (u_longlong_t)vd->vdev_id,
6843 (u_longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
6844 (u_longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
6845 (u_longlong_t)bytes_leaked);
6846 }
6847 total_leaked += ABS(bytes_leaked);
6848 }
6849
6850 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6851 if (!are_precise && total_leaked > 0) {
6852 int pct_leaked = total_leaked * 100 /
6853 vdev_indirect_mapping_bytes_mapped(vim);
6854 (void) printf("cannot verify obsolete indirect mapping "
6855 "counts of vdev %llu because precise feature was not "
6856 "enabled when it was removed: %d%% (%llx bytes) of mapping"
6857 "unreferenced\n",
6858 (u_longlong_t)vd->vdev_id, pct_leaked,
6859 (u_longlong_t)total_leaked);
6860 } else if (total_leaked > 0) {
6861 (void) printf("obsolete indirect mapping count mismatch "
6862 "for vdev %llu -- %llx total bytes mismatched\n",
6863 (u_longlong_t)vd->vdev_id,
6864 (u_longlong_t)total_leaked);
6865 leaks |= B_TRUE;
6866 }
6867
6868 vdev_indirect_mapping_free_obsolete_counts(vim,
6869 zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6870 zcb->zcb_vd_obsolete_counts[vd->vdev_id] = NULL;
6871
6872 return (leaks);
6873 }
6874
6875 static boolean_t
zdb_leak_fini(spa_t * spa,zdb_cb_t * zcb)6876 zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb)
6877 {
6878 if (dump_opt['L'])
6879 return (B_FALSE);
6880
6881 boolean_t leaks = B_FALSE;
6882 vdev_t *rvd = spa->spa_root_vdev;
6883 for (unsigned c = 0; c < rvd->vdev_children; c++) {
6884 vdev_t *vd = rvd->vdev_child[c];
6885
6886 if (zcb->zcb_vd_obsolete_counts[c] != NULL) {
6887 leaks |= zdb_check_for_obsolete_leaks(vd, zcb);
6888 }
6889
6890 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6891 metaslab_t *msp = vd->vdev_ms[m];
6892 ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class ==
6893 spa_embedded_log_class(spa) ||
6894 msp->ms_group->mg_class ==
6895 spa_special_embedded_log_class(spa)) ?
6896 vd->vdev_log_mg : vd->vdev_mg);
6897
6898 /*
6899 * ms_allocatable has been overloaded
6900 * to contain allocated segments. Now that
6901 * we finished traversing all blocks, any
6902 * block that remains in the ms_allocatable
6903 * represents an allocated block that we
6904 * did not claim during the traversal.
6905 * Claimed blocks would have been removed
6906 * from the ms_allocatable. For indirect
6907 * vdevs, space remaining in the tree
6908 * represents parts of the mapping that are
6909 * not referenced, which is not a bug.
6910 */
6911 if (vd->vdev_ops == &vdev_indirect_ops) {
6912 zfs_range_tree_vacate(msp->ms_allocatable,
6913 NULL, NULL);
6914 } else {
6915 zfs_range_tree_vacate(msp->ms_allocatable,
6916 zdb_leak, vd);
6917 }
6918 if (msp->ms_loaded) {
6919 msp->ms_loaded = B_FALSE;
6920 }
6921 }
6922 }
6923
6924 umem_free(zcb->zcb_vd_obsolete_counts,
6925 rvd->vdev_children * sizeof (uint32_t *));
6926 zcb->zcb_vd_obsolete_counts = NULL;
6927
6928 return (leaks);
6929 }
6930
6931 static int
count_block_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)6932 count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6933 {
6934 (void) tx;
6935 zdb_cb_t *zcb = arg;
6936
6937 if (dump_opt['b'] >= 5) {
6938 char blkbuf[BP_SPRINTF_LEN];
6939 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6940 (void) printf("[%s] %s\n",
6941 "deferred free", blkbuf);
6942 }
6943 zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED);
6944 return (0);
6945 }
6946
6947 /*
6948 * Iterate over livelists which have been destroyed by the user but
6949 * are still present in the MOS, waiting to be freed
6950 */
6951 static void
iterate_deleted_livelists(spa_t * spa,ll_iter_t func,void * arg)6952 iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg)
6953 {
6954 objset_t *mos = spa->spa_meta_objset;
6955 uint64_t zap_obj;
6956 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
6957 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
6958 if (err == ENOENT)
6959 return;
6960 ASSERT0(err);
6961
6962 zap_cursor_t zc;
6963 zap_attribute_t *attrp = zap_attribute_alloc();
6964 dsl_deadlist_t ll;
6965 /* NULL out os prior to dsl_deadlist_open in case it's garbage */
6966 ll.dl_os = NULL;
6967 for (zap_cursor_init(&zc, mos, zap_obj);
6968 zap_cursor_retrieve(&zc, attrp) == 0;
6969 (void) zap_cursor_advance(&zc)) {
6970 VERIFY0(dsl_deadlist_open(&ll, mos, attrp->za_first_integer));
6971 func(&ll, arg);
6972 dsl_deadlist_close(&ll);
6973 }
6974 zap_cursor_fini(&zc);
6975 zap_attribute_free(attrp);
6976 }
6977
6978 static int
bpobj_count_block_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6979 bpobj_count_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6980 dmu_tx_t *tx)
6981 {
6982 ASSERT(!bp_freed);
6983 return (count_block_cb(arg, bp, tx));
6984 }
6985
6986 static int
livelist_entry_count_blocks_cb(void * args,dsl_deadlist_entry_t * dle)6987 livelist_entry_count_blocks_cb(void *args, dsl_deadlist_entry_t *dle)
6988 {
6989 zdb_cb_t *zbc = args;
6990 bplist_t blks;
6991 bplist_create(&blks);
6992 /* determine which blocks have been alloc'd but not freed */
6993 VERIFY0(dsl_process_sub_livelist(&dle->dle_bpobj, &blks, NULL, NULL));
6994 /* count those blocks */
6995 (void) bplist_iterate(&blks, count_block_cb, zbc, NULL);
6996 bplist_destroy(&blks);
6997 return (0);
6998 }
6999
7000 static void
livelist_count_blocks(dsl_deadlist_t * ll,void * arg)7001 livelist_count_blocks(dsl_deadlist_t *ll, void *arg)
7002 {
7003 dsl_deadlist_iterate(ll, livelist_entry_count_blocks_cb, arg);
7004 }
7005
7006 /*
7007 * Count the blocks in the livelists that have been destroyed by the user
7008 * but haven't yet been freed.
7009 */
7010 static void
deleted_livelists_count_blocks(spa_t * spa,zdb_cb_t * zbc)7011 deleted_livelists_count_blocks(spa_t *spa, zdb_cb_t *zbc)
7012 {
7013 iterate_deleted_livelists(spa, livelist_count_blocks, zbc);
7014 }
7015
7016 static void
dump_livelist_cb(dsl_deadlist_t * ll,void * arg)7017 dump_livelist_cb(dsl_deadlist_t *ll, void *arg)
7018 {
7019 ASSERT3P(arg, ==, NULL);
7020 global_feature_count[SPA_FEATURE_LIVELIST]++;
7021 dump_blkptr_list(ll, "Deleted Livelist");
7022 dsl_deadlist_iterate(ll, sublivelist_verify_lightweight, NULL);
7023 }
7024
7025 /*
7026 * Print out, register object references to, and increment feature counts for
7027 * livelists that have been destroyed by the user but haven't yet been freed.
7028 */
7029 static void
deleted_livelists_dump_mos(spa_t * spa)7030 deleted_livelists_dump_mos(spa_t *spa)
7031 {
7032 uint64_t zap_obj;
7033 objset_t *mos = spa->spa_meta_objset;
7034 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
7035 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
7036 if (err == ENOENT)
7037 return;
7038 mos_obj_refd(zap_obj);
7039 iterate_deleted_livelists(spa, dump_livelist_cb, NULL);
7040 }
7041
7042 static int
zdb_brt_entry_compare(const void * zcn1,const void * zcn2)7043 zdb_brt_entry_compare(const void *zcn1, const void *zcn2)
7044 {
7045 const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva;
7046 const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva;
7047 int cmp;
7048
7049 cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
7050 if (cmp == 0)
7051 cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2));
7052
7053 return (cmp);
7054 }
7055
7056 static int
dump_block_stats(spa_t * spa)7057 dump_block_stats(spa_t *spa)
7058 {
7059 zdb_cb_t *zcb;
7060 zdb_blkstats_t *zb, *tzb;
7061 uint64_t norm_alloc, norm_space, total_alloc, total_found;
7062 int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7063 TRAVERSE_NO_DECRYPT | TRAVERSE_HARD;
7064 boolean_t leaks = B_FALSE;
7065 int e, c, err;
7066 bp_embedded_type_t i;
7067
7068 ddt_prefetch_all(spa);
7069
7070 zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL);
7071
7072 if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
7073 avl_create(&zcb->zcb_brt, zdb_brt_entry_compare,
7074 sizeof (zdb_brt_entry_t),
7075 offsetof(zdb_brt_entry_t, zbre_node));
7076 zcb->zcb_brt_is_active = B_TRUE;
7077 }
7078
7079 (void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
7080 (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
7081 (dump_opt['c'] == 1) ? "metadata " : "",
7082 dump_opt['c'] ? "checksums " : "",
7083 (dump_opt['c'] && !dump_opt['L']) ? "and verify " : "",
7084 !dump_opt['L'] ? "nothing leaked " : "");
7085
7086 /*
7087 * When leak detection is enabled we load all space maps as SM_ALLOC
7088 * maps, then traverse the pool claiming each block we discover. If
7089 * the pool is perfectly consistent, the segment trees will be empty
7090 * when we're done. Anything left over is a leak; any block we can't
7091 * claim (because it's not part of any space map) is a double
7092 * allocation, reference to a freed block, or an unclaimed log block.
7093 *
7094 * When leak detection is disabled (-L option) we still traverse the
7095 * pool claiming each block we discover, but we skip opening any space
7096 * maps.
7097 */
7098 zdb_leak_init(spa, zcb);
7099
7100 /*
7101 * If there's a deferred-free bplist, process that first.
7102 */
7103 (void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
7104 bpobj_count_block_cb, zcb, NULL);
7105
7106 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
7107 (void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
7108 bpobj_count_block_cb, zcb, NULL);
7109 }
7110
7111 zdb_claim_removing(spa, zcb);
7112
7113 if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
7114 VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,
7115 spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb,
7116 zcb, NULL));
7117 }
7118
7119 deleted_livelists_count_blocks(spa, zcb);
7120
7121 if (dump_opt['c'] > 1)
7122 flags |= TRAVERSE_PREFETCH_DATA;
7123
7124 zcb->zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
7125 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa));
7126 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
7127 zcb->zcb_totalasize +=
7128 metaslab_class_get_alloc(spa_embedded_log_class(spa));
7129 zcb->zcb_totalasize +=
7130 metaslab_class_get_alloc(spa_special_embedded_log_class(spa));
7131 zcb->zcb_start = zcb->zcb_lastprint = gethrtime();
7132 err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb);
7133
7134 /*
7135 * If we've traversed the data blocks then we need to wait for those
7136 * I/Os to complete. We leverage "The Godfather" zio to wait on
7137 * all async I/Os to complete.
7138 */
7139 if (dump_opt['c']) {
7140 for (c = 0; c < max_ncpus; c++) {
7141 (void) zio_wait(spa->spa_async_zio_root[c]);
7142 spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL,
7143 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
7144 ZIO_FLAG_GODFATHER);
7145 }
7146 }
7147 ASSERT0(spa->spa_load_verify_bytes);
7148
7149 /*
7150 * Done after zio_wait() since zcb_haderrors is modified in
7151 * zdb_blkptr_done()
7152 */
7153 zcb->zcb_haderrors |= err;
7154
7155 if (zcb->zcb_haderrors) {
7156 (void) printf("\nError counts:\n\n");
7157 (void) printf("\t%5s %s\n", "errno", "count");
7158 for (e = 0; e < 256; e++) {
7159 if (zcb->zcb_errors[e] != 0) {
7160 (void) printf("\t%5d %llu\n",
7161 e, (u_longlong_t)zcb->zcb_errors[e]);
7162 }
7163 }
7164 }
7165
7166 /*
7167 * Report any leaked segments.
7168 */
7169 leaks |= zdb_leak_fini(spa, zcb);
7170
7171 tzb = &zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL];
7172
7173 norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7174 norm_space = metaslab_class_get_space(spa_normal_class(spa));
7175
7176 total_alloc = norm_alloc +
7177 metaslab_class_get_alloc(spa_log_class(spa)) +
7178 metaslab_class_get_alloc(spa_embedded_log_class(spa)) +
7179 metaslab_class_get_alloc(spa_special_embedded_log_class(spa)) +
7180 metaslab_class_get_alloc(spa_special_class(spa)) +
7181 metaslab_class_get_alloc(spa_dedup_class(spa)) +
7182 get_unflushed_alloc_space(spa);
7183 total_found =
7184 tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize +
7185 zcb->zcb_removing_size + zcb->zcb_checkpoint_size;
7186
7187 if (total_found == total_alloc && !dump_opt['L']) {
7188 (void) printf("\n\tNo leaks (block sum matches space"
7189 " maps exactly)\n");
7190 } else if (!dump_opt['L']) {
7191 (void) printf("block traversal size %llu != alloc %llu "
7192 "(%s %lld)\n",
7193 (u_longlong_t)total_found,
7194 (u_longlong_t)total_alloc,
7195 (dump_opt['L']) ? "unreachable" : "leaked",
7196 (longlong_t)(total_alloc - total_found));
7197 }
7198
7199 if (tzb->zb_count == 0) {
7200 umem_free(zcb, sizeof (zdb_cb_t));
7201 return (2);
7202 }
7203
7204 (void) printf("\n");
7205 (void) printf("\t%-16s %14llu\n", "bp count:",
7206 (u_longlong_t)tzb->zb_count);
7207 (void) printf("\t%-16s %14llu\n", "ganged count:",
7208 (longlong_t)tzb->zb_gangs);
7209 (void) printf("\t%-16s %14llu avg: %6llu\n", "bp logical:",
7210 (u_longlong_t)tzb->zb_lsize,
7211 (u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
7212 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7213 "bp physical:", (u_longlong_t)tzb->zb_psize,
7214 (u_longlong_t)(tzb->zb_psize / tzb->zb_count),
7215 (double)tzb->zb_lsize / tzb->zb_psize);
7216 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7217 "bp allocated:", (u_longlong_t)tzb->zb_asize,
7218 (u_longlong_t)(tzb->zb_asize / tzb->zb_count),
7219 (double)tzb->zb_lsize / tzb->zb_asize);
7220 (void) printf("\t%-16s %14llu ref>1: %6llu deduplication: %6.2f\n",
7221 "bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize,
7222 (u_longlong_t)zcb->zcb_dedup_blocks,
7223 (double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0);
7224 (void) printf("\t%-16s %14llu count: %6llu\n",
7225 "bp cloned:", (u_longlong_t)zcb->zcb_clone_asize,
7226 (u_longlong_t)zcb->zcb_clone_blocks);
7227 (void) printf("\t%-16s %14llu used: %5.2f%%\n", "Normal class:",
7228 (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
7229
7230 if (spa_special_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7231 uint64_t alloc = metaslab_class_get_alloc(
7232 spa_special_class(spa));
7233 uint64_t space = metaslab_class_get_space(
7234 spa_special_class(spa));
7235
7236 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7237 "Special class", (u_longlong_t)alloc,
7238 100.0 * alloc / space);
7239 }
7240
7241 if (spa_dedup_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7242 uint64_t alloc = metaslab_class_get_alloc(
7243 spa_dedup_class(spa));
7244 uint64_t space = metaslab_class_get_space(
7245 spa_dedup_class(spa));
7246
7247 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7248 "Dedup class", (u_longlong_t)alloc,
7249 100.0 * alloc / space);
7250 }
7251
7252 if (spa_embedded_log_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7253 uint64_t alloc = metaslab_class_get_alloc(
7254 spa_embedded_log_class(spa));
7255 uint64_t space = metaslab_class_get_space(
7256 spa_embedded_log_class(spa));
7257
7258 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7259 "Embedded log class", (u_longlong_t)alloc,
7260 100.0 * alloc / space);
7261 }
7262
7263 if (spa_special_embedded_log_class(spa)->mc_allocator[0].mca_rotor
7264 != NULL) {
7265 uint64_t alloc = metaslab_class_get_alloc(
7266 spa_special_embedded_log_class(spa));
7267 uint64_t space = metaslab_class_get_space(
7268 spa_special_embedded_log_class(spa));
7269
7270 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7271 "Special embedded log", (u_longlong_t)alloc,
7272 100.0 * alloc / space);
7273 }
7274
7275 for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
7276 if (zcb->zcb_embedded_blocks[i] == 0)
7277 continue;
7278 (void) printf("\n");
7279 (void) printf("\tadditional, non-pointer bps of type %u: "
7280 "%10llu\n",
7281 i, (u_longlong_t)zcb->zcb_embedded_blocks[i]);
7282
7283 if (dump_opt['b'] >= 3) {
7284 (void) printf("\t number of (compressed) bytes: "
7285 "number of bps\n");
7286 dump_histogram(zcb->zcb_embedded_histogram[i],
7287 sizeof (zcb->zcb_embedded_histogram[i]) /
7288 sizeof (zcb->zcb_embedded_histogram[i][0]), 0);
7289 }
7290 }
7291
7292 if (tzb->zb_ditto_samevdev != 0) {
7293 (void) printf("\tDittoed blocks on same vdev: %llu\n",
7294 (longlong_t)tzb->zb_ditto_samevdev);
7295 }
7296 if (tzb->zb_ditto_same_ms != 0) {
7297 (void) printf("\tDittoed blocks in same metaslab: %llu\n",
7298 (longlong_t)tzb->zb_ditto_same_ms);
7299 }
7300
7301 for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) {
7302 vdev_t *vd = spa->spa_root_vdev->vdev_child[v];
7303 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7304
7305 if (vim == NULL) {
7306 continue;
7307 }
7308
7309 char mem[32];
7310 zdb_nicenum(vdev_indirect_mapping_num_entries(vim),
7311 mem, vdev_indirect_mapping_size(vim));
7312
7313 (void) printf("\tindirect vdev id %llu has %llu segments "
7314 "(%s in memory)\n",
7315 (longlong_t)vd->vdev_id,
7316 (longlong_t)vdev_indirect_mapping_num_entries(vim), mem);
7317 }
7318
7319 if (dump_opt['b'] >= 2) {
7320 int l, t, level;
7321 char csize[32], lsize[32], psize[32], asize[32];
7322 char avg[32], gang[32];
7323 (void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
7324 "\t avg\t comp\t%%Total\tType\n");
7325
7326 zfs_blkstat_t *mdstats = umem_zalloc(sizeof (zfs_blkstat_t),
7327 UMEM_NOFAIL);
7328
7329 for (t = 0; t <= ZDB_OT_TOTAL; t++) {
7330 const char *typename;
7331
7332 /* make sure nicenum has enough space */
7333 _Static_assert(sizeof (csize) >= NN_NUMBUF_SZ,
7334 "csize truncated");
7335 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ,
7336 "lsize truncated");
7337 _Static_assert(sizeof (psize) >= NN_NUMBUF_SZ,
7338 "psize truncated");
7339 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ,
7340 "asize truncated");
7341 _Static_assert(sizeof (avg) >= NN_NUMBUF_SZ,
7342 "avg truncated");
7343 _Static_assert(sizeof (gang) >= NN_NUMBUF_SZ,
7344 "gang truncated");
7345
7346 if (t < DMU_OT_NUMTYPES)
7347 typename = dmu_ot[t].ot_name;
7348 else
7349 typename = zdb_ot_extname[t - DMU_OT_NUMTYPES];
7350
7351 if (zcb->zcb_type[ZB_TOTAL][t].zb_asize == 0) {
7352 (void) printf("%6s\t%5s\t%5s\t%5s"
7353 "\t%5s\t%5s\t%6s\t%s\n",
7354 "-",
7355 "-",
7356 "-",
7357 "-",
7358 "-",
7359 "-",
7360 "-",
7361 typename);
7362 continue;
7363 }
7364
7365 for (l = ZB_TOTAL - 1; l >= -1; l--) {
7366 level = (l == -1 ? ZB_TOTAL : l);
7367 zb = &zcb->zcb_type[level][t];
7368
7369 if (zb->zb_asize == 0)
7370 continue;
7371
7372 if (level != ZB_TOTAL && t < DMU_OT_NUMTYPES &&
7373 (level > 0 || DMU_OT_IS_METADATA(t))) {
7374 mdstats->zb_count += zb->zb_count;
7375 mdstats->zb_lsize += zb->zb_lsize;
7376 mdstats->zb_psize += zb->zb_psize;
7377 mdstats->zb_asize += zb->zb_asize;
7378 mdstats->zb_gangs += zb->zb_gangs;
7379 }
7380
7381 if (dump_opt['b'] < 3 && level != ZB_TOTAL)
7382 continue;
7383
7384 if (level == 0 && zb->zb_asize ==
7385 zcb->zcb_type[ZB_TOTAL][t].zb_asize)
7386 continue;
7387
7388 zdb_nicenum(zb->zb_count, csize,
7389 sizeof (csize));
7390 zdb_nicenum(zb->zb_lsize, lsize,
7391 sizeof (lsize));
7392 zdb_nicenum(zb->zb_psize, psize,
7393 sizeof (psize));
7394 zdb_nicenum(zb->zb_asize, asize,
7395 sizeof (asize));
7396 zdb_nicenum(zb->zb_asize / zb->zb_count, avg,
7397 sizeof (avg));
7398 zdb_nicenum(zb->zb_gangs, gang, sizeof (gang));
7399
7400 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7401 "\t%5.2f\t%6.2f\t",
7402 csize, lsize, psize, asize, avg,
7403 (double)zb->zb_lsize / zb->zb_psize,
7404 100.0 * zb->zb_asize / tzb->zb_asize);
7405
7406 if (level == ZB_TOTAL)
7407 (void) printf("%s\n", typename);
7408 else
7409 (void) printf(" L%d %s\n",
7410 level, typename);
7411
7412 if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) {
7413 (void) printf("\t number of ganged "
7414 "blocks: %s\n", gang);
7415 }
7416
7417 if (dump_opt['b'] >= 4) {
7418 (void) printf("psize "
7419 "(in 512-byte sectors): "
7420 "number of blocks\n");
7421 dump_histogram(zb->zb_psize_histogram,
7422 PSIZE_HISTO_SIZE, 0);
7423 }
7424 }
7425 }
7426 zdb_nicenum(mdstats->zb_count, csize,
7427 sizeof (csize));
7428 zdb_nicenum(mdstats->zb_lsize, lsize,
7429 sizeof (lsize));
7430 zdb_nicenum(mdstats->zb_psize, psize,
7431 sizeof (psize));
7432 zdb_nicenum(mdstats->zb_asize, asize,
7433 sizeof (asize));
7434 zdb_nicenum(mdstats->zb_asize / mdstats->zb_count, avg,
7435 sizeof (avg));
7436 zdb_nicenum(mdstats->zb_gangs, gang, sizeof (gang));
7437
7438 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7439 "\t%5.2f\t%6.2f\t",
7440 csize, lsize, psize, asize, avg,
7441 (double)mdstats->zb_lsize / mdstats->zb_psize,
7442 100.0 * mdstats->zb_asize / tzb->zb_asize);
7443 (void) printf("%s\n", "Metadata Total");
7444
7445 /* Output a table summarizing block sizes in the pool */
7446 if (dump_opt['b'] >= 2) {
7447 dump_size_histograms(zcb);
7448 }
7449
7450 umem_free(mdstats, sizeof (zfs_blkstat_t));
7451 }
7452
7453 (void) printf("\n");
7454
7455 if (leaks) {
7456 umem_free(zcb, sizeof (zdb_cb_t));
7457 return (2);
7458 }
7459
7460 if (zcb->zcb_haderrors) {
7461 umem_free(zcb, sizeof (zdb_cb_t));
7462 return (3);
7463 }
7464
7465 umem_free(zcb, sizeof (zdb_cb_t));
7466 return (0);
7467 }
7468
7469 typedef struct zdb_ddt_entry {
7470 /* key must be first for ddt_key_compare */
7471 ddt_key_t zdde_key;
7472 uint64_t zdde_ref_blocks;
7473 uint64_t zdde_ref_lsize;
7474 uint64_t zdde_ref_psize;
7475 uint64_t zdde_ref_dsize;
7476 avl_node_t zdde_node;
7477 } zdb_ddt_entry_t;
7478
7479 static int
zdb_ddt_add_cb(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)7480 zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
7481 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
7482 {
7483 (void) zilog, (void) dnp;
7484 avl_tree_t *t = arg;
7485 avl_index_t where;
7486 zdb_ddt_entry_t *zdde, zdde_search;
7487
7488 if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
7489 BP_IS_EMBEDDED(bp))
7490 return (0);
7491
7492 if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
7493 (void) printf("traversing objset %llu, %llu objects, "
7494 "%lu blocks so far\n",
7495 (u_longlong_t)zb->zb_objset,
7496 (u_longlong_t)BP_GET_FILL(bp),
7497 avl_numnodes(t));
7498 }
7499
7500 if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF ||
7501 BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
7502 return (0);
7503
7504 ddt_key_fill(&zdde_search.zdde_key, bp);
7505
7506 zdde = avl_find(t, &zdde_search, &where);
7507
7508 if (zdde == NULL) {
7509 zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL);
7510 zdde->zdde_key = zdde_search.zdde_key;
7511 avl_insert(t, zdde, where);
7512 }
7513
7514 zdde->zdde_ref_blocks += 1;
7515 zdde->zdde_ref_lsize += BP_GET_LSIZE(bp);
7516 zdde->zdde_ref_psize += BP_GET_PSIZE(bp);
7517 zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp);
7518
7519 return (0);
7520 }
7521
7522 static void
dump_simulated_ddt(spa_t * spa)7523 dump_simulated_ddt(spa_t *spa)
7524 {
7525 avl_tree_t t;
7526 void *cookie = NULL;
7527 zdb_ddt_entry_t *zdde;
7528 ddt_histogram_t ddh_total = {{{0}}};
7529 ddt_stat_t dds_total = {0};
7530
7531 avl_create(&t, ddt_key_compare,
7532 sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
7533
7534 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7535
7536 (void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7537 TRAVERSE_NO_DECRYPT, zdb_ddt_add_cb, &t);
7538
7539 spa_config_exit(spa, SCL_CONFIG, FTAG);
7540
7541 while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) {
7542 uint64_t refcnt = zdde->zdde_ref_blocks;
7543 ASSERT(refcnt != 0);
7544
7545 ddt_stat_t *dds = &ddh_total.ddh_stat[highbit64(refcnt) - 1];
7546
7547 dds->dds_blocks += zdde->zdde_ref_blocks / refcnt;
7548 dds->dds_lsize += zdde->zdde_ref_lsize / refcnt;
7549 dds->dds_psize += zdde->zdde_ref_psize / refcnt;
7550 dds->dds_dsize += zdde->zdde_ref_dsize / refcnt;
7551
7552 dds->dds_ref_blocks += zdde->zdde_ref_blocks;
7553 dds->dds_ref_lsize += zdde->zdde_ref_lsize;
7554 dds->dds_ref_psize += zdde->zdde_ref_psize;
7555 dds->dds_ref_dsize += zdde->zdde_ref_dsize;
7556
7557 umem_free(zdde, sizeof (*zdde));
7558 }
7559
7560 avl_destroy(&t);
7561
7562 ddt_histogram_total(&dds_total, &ddh_total);
7563
7564 (void) printf("Simulated DDT histogram:\n");
7565
7566 zpool_dump_ddt(&dds_total, &ddh_total);
7567
7568 dump_dedup_ratio(&dds_total);
7569 }
7570
7571 static int
verify_device_removal_feature_counts(spa_t * spa)7572 verify_device_removal_feature_counts(spa_t *spa)
7573 {
7574 uint64_t dr_feature_refcount = 0;
7575 uint64_t oc_feature_refcount = 0;
7576 uint64_t indirect_vdev_count = 0;
7577 uint64_t precise_vdev_count = 0;
7578 uint64_t obsolete_counts_object_count = 0;
7579 uint64_t obsolete_sm_count = 0;
7580 uint64_t obsolete_counts_count = 0;
7581 uint64_t scip_count = 0;
7582 uint64_t obsolete_bpobj_count = 0;
7583 int ret = 0;
7584
7585 spa_condensing_indirect_phys_t *scip =
7586 &spa->spa_condensing_indirect_phys;
7587 if (scip->scip_next_mapping_object != 0) {
7588 vdev_t *vd = spa->spa_root_vdev->vdev_child[scip->scip_vdev];
7589 ASSERT(scip->scip_prev_obsolete_sm_object != 0);
7590 ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
7591
7592 (void) printf("Condensing indirect vdev %llu: new mapping "
7593 "object %llu, prev obsolete sm %llu\n",
7594 (u_longlong_t)scip->scip_vdev,
7595 (u_longlong_t)scip->scip_next_mapping_object,
7596 (u_longlong_t)scip->scip_prev_obsolete_sm_object);
7597 if (scip->scip_prev_obsolete_sm_object != 0) {
7598 space_map_t *prev_obsolete_sm = NULL;
7599 VERIFY0(space_map_open(&prev_obsolete_sm,
7600 spa->spa_meta_objset,
7601 scip->scip_prev_obsolete_sm_object,
7602 0, vd->vdev_asize, 0));
7603 dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm);
7604 (void) printf("\n");
7605 space_map_close(prev_obsolete_sm);
7606 }
7607
7608 scip_count += 2;
7609 }
7610
7611 for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
7612 vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
7613 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
7614
7615 if (vic->vic_mapping_object != 0) {
7616 ASSERT(vd->vdev_ops == &vdev_indirect_ops ||
7617 vd->vdev_removing);
7618 indirect_vdev_count++;
7619
7620 if (vd->vdev_indirect_mapping->vim_havecounts) {
7621 obsolete_counts_count++;
7622 }
7623 }
7624
7625 boolean_t are_precise;
7626 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7627 if (are_precise) {
7628 ASSERT(vic->vic_mapping_object != 0);
7629 precise_vdev_count++;
7630 }
7631
7632 uint64_t obsolete_sm_object;
7633 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
7634 if (obsolete_sm_object != 0) {
7635 ASSERT(vic->vic_mapping_object != 0);
7636 obsolete_sm_count++;
7637 }
7638 }
7639
7640 (void) feature_get_refcount(spa,
7641 &spa_feature_table[SPA_FEATURE_DEVICE_REMOVAL],
7642 &dr_feature_refcount);
7643 (void) feature_get_refcount(spa,
7644 &spa_feature_table[SPA_FEATURE_OBSOLETE_COUNTS],
7645 &oc_feature_refcount);
7646
7647 if (dr_feature_refcount != indirect_vdev_count) {
7648 ret = 1;
7649 (void) printf("Number of indirect vdevs (%llu) " \
7650 "does not match feature count (%llu)\n",
7651 (u_longlong_t)indirect_vdev_count,
7652 (u_longlong_t)dr_feature_refcount);
7653 } else {
7654 (void) printf("Verified device_removal feature refcount " \
7655 "of %llu is correct\n",
7656 (u_longlong_t)dr_feature_refcount);
7657 }
7658
7659 if (zap_contains(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT,
7660 DMU_POOL_OBSOLETE_BPOBJ) == 0) {
7661 obsolete_bpobj_count++;
7662 }
7663
7664
7665 obsolete_counts_object_count = precise_vdev_count;
7666 obsolete_counts_object_count += obsolete_sm_count;
7667 obsolete_counts_object_count += obsolete_counts_count;
7668 obsolete_counts_object_count += scip_count;
7669 obsolete_counts_object_count += obsolete_bpobj_count;
7670 obsolete_counts_object_count += remap_deadlist_count;
7671
7672 if (oc_feature_refcount != obsolete_counts_object_count) {
7673 ret = 1;
7674 (void) printf("Number of obsolete counts objects (%llu) " \
7675 "does not match feature count (%llu)\n",
7676 (u_longlong_t)obsolete_counts_object_count,
7677 (u_longlong_t)oc_feature_refcount);
7678 (void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
7679 "ob:%llu rd:%llu\n",
7680 (u_longlong_t)precise_vdev_count,
7681 (u_longlong_t)obsolete_sm_count,
7682 (u_longlong_t)obsolete_counts_count,
7683 (u_longlong_t)scip_count,
7684 (u_longlong_t)obsolete_bpobj_count,
7685 (u_longlong_t)remap_deadlist_count);
7686 } else {
7687 (void) printf("Verified indirect_refcount feature refcount " \
7688 "of %llu is correct\n",
7689 (u_longlong_t)oc_feature_refcount);
7690 }
7691 return (ret);
7692 }
7693
7694 static void
zdb_set_skip_mmp(char * target)7695 zdb_set_skip_mmp(char *target)
7696 {
7697 spa_t *spa;
7698
7699 /*
7700 * Disable the activity check to allow examination of
7701 * active pools.
7702 */
7703 mutex_enter(&spa_namespace_lock);
7704 if ((spa = spa_lookup(target)) != NULL) {
7705 spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
7706 }
7707 mutex_exit(&spa_namespace_lock);
7708 }
7709
7710 #define BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7711 /*
7712 * Import the checkpointed state of the pool specified by the target
7713 * parameter as readonly. The function also accepts a pool config
7714 * as an optional parameter, else it attempts to infer the config by
7715 * the name of the target pool.
7716 *
7717 * Note that the checkpointed state's pool name will be the name of
7718 * the original pool with the above suffix appended to it. In addition,
7719 * if the target is not a pool name (e.g. a path to a dataset) then
7720 * the new_path parameter is populated with the updated path to
7721 * reflect the fact that we are looking into the checkpointed state.
7722 *
7723 * The function returns a newly-allocated copy of the name of the
7724 * pool containing the checkpointed state. When this copy is no
7725 * longer needed it should be freed with free(3C). Same thing
7726 * applies to the new_path parameter if allocated.
7727 */
7728 static char *
import_checkpointed_state(char * target,nvlist_t * cfg,boolean_t target_is_spa,char ** new_path)7729 import_checkpointed_state(char *target, nvlist_t *cfg, boolean_t target_is_spa,
7730 char **new_path)
7731 {
7732 int error = 0;
7733 char *poolname, *bogus_name = NULL;
7734 boolean_t freecfg = B_FALSE;
7735
7736 /* If the target is not a pool, the extract the pool name */
7737 char *path_start = strchr(target, '/');
7738 if (target_is_spa || path_start == NULL) {
7739 poolname = target;
7740 } else {
7741 size_t poolname_len = path_start - target;
7742 poolname = strndup(target, poolname_len);
7743 }
7744
7745 if (cfg == NULL) {
7746 zdb_set_skip_mmp(poolname);
7747 error = spa_get_stats(poolname, &cfg, NULL, 0);
7748 if (error != 0) {
7749 fatal("Tried to read config of pool \"%s\" but "
7750 "spa_get_stats() failed with error %d\n",
7751 poolname, error);
7752 }
7753 freecfg = B_TRUE;
7754 }
7755
7756 if (asprintf(&bogus_name, "%s%s", poolname, BOGUS_SUFFIX) == -1) {
7757 if (target != poolname)
7758 free(poolname);
7759 return (NULL);
7760 }
7761 fnvlist_add_string(cfg, ZPOOL_CONFIG_POOL_NAME, bogus_name);
7762
7763 error = spa_import(bogus_name, cfg, NULL,
7764 ZFS_IMPORT_MISSING_LOG | ZFS_IMPORT_CHECKPOINT |
7765 ZFS_IMPORT_SKIP_MMP);
7766 if (freecfg)
7767 nvlist_free(cfg);
7768 if (error != 0) {
7769 fatal("Tried to import pool \"%s\" but spa_import() failed "
7770 "with error %d\n", bogus_name, error);
7771 }
7772
7773 if (new_path != NULL && !target_is_spa) {
7774 if (asprintf(new_path, "%s%s", bogus_name,
7775 path_start != NULL ? path_start : "") == -1) {
7776 free(bogus_name);
7777 if (!target_is_spa && path_start != NULL)
7778 free(poolname);
7779 return (NULL);
7780 }
7781 }
7782
7783 if (target != poolname)
7784 free(poolname);
7785
7786 return (bogus_name);
7787 }
7788
7789 typedef struct verify_checkpoint_sm_entry_cb_arg {
7790 vdev_t *vcsec_vd;
7791
7792 /* the following fields are only used for printing progress */
7793 uint64_t vcsec_entryid;
7794 uint64_t vcsec_num_entries;
7795 } verify_checkpoint_sm_entry_cb_arg_t;
7796
7797 #define ENTRIES_PER_PROGRESS_UPDATE 10000
7798
7799 static int
verify_checkpoint_sm_entry_cb(space_map_entry_t * sme,void * arg)7800 verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, void *arg)
7801 {
7802 verify_checkpoint_sm_entry_cb_arg_t *vcsec = arg;
7803 vdev_t *vd = vcsec->vcsec_vd;
7804 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
7805 uint64_t end = sme->sme_offset + sme->sme_run;
7806
7807 ASSERT(sme->sme_type == SM_FREE);
7808
7809 if ((vcsec->vcsec_entryid % ENTRIES_PER_PROGRESS_UPDATE) == 0) {
7810 (void) fprintf(stderr,
7811 "\rverifying vdev %llu, space map entry %llu of %llu ...",
7812 (longlong_t)vd->vdev_id,
7813 (longlong_t)vcsec->vcsec_entryid,
7814 (longlong_t)vcsec->vcsec_num_entries);
7815 }
7816 vcsec->vcsec_entryid++;
7817
7818 /*
7819 * See comment in checkpoint_sm_exclude_entry_cb()
7820 */
7821 VERIFY3U(sme->sme_offset, >=, ms->ms_start);
7822 VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
7823
7824 /*
7825 * The entries in the vdev_checkpoint_sm should be marked as
7826 * allocated in the checkpointed state of the pool, therefore
7827 * their respective ms_allocateable trees should not contain them.
7828 */
7829 mutex_enter(&ms->ms_lock);
7830 zfs_range_tree_verify_not_present(ms->ms_allocatable,
7831 sme->sme_offset, sme->sme_run);
7832 mutex_exit(&ms->ms_lock);
7833
7834 return (0);
7835 }
7836
7837 /*
7838 * Verify that all segments in the vdev_checkpoint_sm are allocated
7839 * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
7840 * ms_allocatable).
7841 *
7842 * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
7843 * each vdev in the current state of the pool to the metaslab space maps
7844 * (ms_sm) of the checkpointed state of the pool.
7845 *
7846 * Note that the function changes the state of the ms_allocatable
7847 * trees of the current spa_t. The entries of these ms_allocatable
7848 * trees are cleared out and then repopulated from with the free
7849 * entries of their respective ms_sm space maps.
7850 */
7851 static void
verify_checkpoint_vdev_spacemaps(spa_t * checkpoint,spa_t * current)7852 verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, spa_t *current)
7853 {
7854 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7855 vdev_t *current_rvd = current->spa_root_vdev;
7856
7857 load_concrete_ms_allocatable_trees(checkpoint, SM_FREE);
7858
7859 for (uint64_t c = 0; c < ckpoint_rvd->vdev_children; c++) {
7860 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[c];
7861 vdev_t *current_vd = current_rvd->vdev_child[c];
7862
7863 space_map_t *checkpoint_sm = NULL;
7864 uint64_t checkpoint_sm_obj;
7865
7866 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7867 /*
7868 * Since we don't allow device removal in a pool
7869 * that has a checkpoint, we expect that all removed
7870 * vdevs were removed from the pool before the
7871 * checkpoint.
7872 */
7873 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7874 continue;
7875 }
7876
7877 /*
7878 * If the checkpoint space map doesn't exist, then nothing
7879 * here is checkpointed so there's nothing to verify.
7880 */
7881 if (current_vd->vdev_top_zap == 0 ||
7882 zap_contains(spa_meta_objset(current),
7883 current_vd->vdev_top_zap,
7884 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
7885 continue;
7886
7887 VERIFY0(zap_lookup(spa_meta_objset(current),
7888 current_vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
7889 sizeof (uint64_t), 1, &checkpoint_sm_obj));
7890
7891 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current),
7892 checkpoint_sm_obj, 0, current_vd->vdev_asize,
7893 current_vd->vdev_ashift));
7894
7895 verify_checkpoint_sm_entry_cb_arg_t vcsec;
7896 vcsec.vcsec_vd = ckpoint_vd;
7897 vcsec.vcsec_entryid = 0;
7898 vcsec.vcsec_num_entries =
7899 space_map_length(checkpoint_sm) / sizeof (uint64_t);
7900 VERIFY0(space_map_iterate(checkpoint_sm,
7901 space_map_length(checkpoint_sm),
7902 verify_checkpoint_sm_entry_cb, &vcsec));
7903 if (dump_opt['m'] > 3)
7904 dump_spacemap(current->spa_meta_objset, checkpoint_sm);
7905 space_map_close(checkpoint_sm);
7906 }
7907
7908 /*
7909 * If we've added vdevs since we took the checkpoint, ensure
7910 * that their checkpoint space maps are empty.
7911 */
7912 if (ckpoint_rvd->vdev_children < current_rvd->vdev_children) {
7913 for (uint64_t c = ckpoint_rvd->vdev_children;
7914 c < current_rvd->vdev_children; c++) {
7915 vdev_t *current_vd = current_rvd->vdev_child[c];
7916 VERIFY3P(current_vd->vdev_checkpoint_sm, ==, NULL);
7917 }
7918 }
7919
7920 /* for cleaner progress output */
7921 (void) fprintf(stderr, "\n");
7922 }
7923
7924 /*
7925 * Verifies that all space that's allocated in the checkpoint is
7926 * still allocated in the current version, by checking that everything
7927 * in checkpoint's ms_allocatable (which is actually allocated, not
7928 * allocatable/free) is not present in current's ms_allocatable.
7929 *
7930 * Note that the function changes the state of the ms_allocatable
7931 * trees of both spas when called. The entries of all ms_allocatable
7932 * trees are cleared out and then repopulated from their respective
7933 * ms_sm space maps. In the checkpointed state we load the allocated
7934 * entries, and in the current state we load the free entries.
7935 */
7936 static void
verify_checkpoint_ms_spacemaps(spa_t * checkpoint,spa_t * current)7937 verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_t *current)
7938 {
7939 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7940 vdev_t *current_rvd = current->spa_root_vdev;
7941
7942 load_concrete_ms_allocatable_trees(checkpoint, SM_ALLOC);
7943 load_concrete_ms_allocatable_trees(current, SM_FREE);
7944
7945 for (uint64_t i = 0; i < ckpoint_rvd->vdev_children; i++) {
7946 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[i];
7947 vdev_t *current_vd = current_rvd->vdev_child[i];
7948
7949 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7950 /*
7951 * See comment in verify_checkpoint_vdev_spacemaps()
7952 */
7953 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7954 continue;
7955 }
7956
7957 for (uint64_t m = 0; m < ckpoint_vd->vdev_ms_count; m++) {
7958 metaslab_t *ckpoint_msp = ckpoint_vd->vdev_ms[m];
7959 metaslab_t *current_msp = current_vd->vdev_ms[m];
7960
7961 (void) fprintf(stderr,
7962 "\rverifying vdev %llu of %llu, "
7963 "metaslab %llu of %llu ...",
7964 (longlong_t)current_vd->vdev_id,
7965 (longlong_t)current_rvd->vdev_children,
7966 (longlong_t)current_vd->vdev_ms[m]->ms_id,
7967 (longlong_t)current_vd->vdev_ms_count);
7968
7969 /*
7970 * We walk through the ms_allocatable trees that
7971 * are loaded with the allocated blocks from the
7972 * ms_sm spacemaps of the checkpoint. For each
7973 * one of these ranges we ensure that none of them
7974 * exists in the ms_allocatable trees of the
7975 * current state which are loaded with the ranges
7976 * that are currently free.
7977 *
7978 * This way we ensure that none of the blocks that
7979 * are part of the checkpoint were freed by mistake.
7980 */
7981 zfs_range_tree_walk(ckpoint_msp->ms_allocatable,
7982 (zfs_range_tree_func_t *)
7983 zfs_range_tree_verify_not_present,
7984 current_msp->ms_allocatable);
7985 }
7986 }
7987
7988 /* for cleaner progress output */
7989 (void) fprintf(stderr, "\n");
7990 }
7991
7992 static void
verify_checkpoint_blocks(spa_t * spa)7993 verify_checkpoint_blocks(spa_t *spa)
7994 {
7995 ASSERT(!dump_opt['L']);
7996
7997 spa_t *checkpoint_spa;
7998 char *checkpoint_pool;
7999 int error = 0;
8000
8001 /*
8002 * We import the checkpointed state of the pool (under a different
8003 * name) so we can do verification on it against the current state
8004 * of the pool.
8005 */
8006 checkpoint_pool = import_checkpointed_state(spa->spa_name, NULL, B_TRUE,
8007 NULL);
8008 ASSERT(strcmp(spa->spa_name, checkpoint_pool) != 0);
8009
8010 error = spa_open(checkpoint_pool, &checkpoint_spa, FTAG);
8011 if (error != 0) {
8012 fatal("Tried to open pool \"%s\" but spa_open() failed with "
8013 "error %d\n", checkpoint_pool, error);
8014 }
8015
8016 /*
8017 * Ensure that ranges in the checkpoint space maps of each vdev
8018 * are allocated according to the checkpointed state's metaslab
8019 * space maps.
8020 */
8021 verify_checkpoint_vdev_spacemaps(checkpoint_spa, spa);
8022
8023 /*
8024 * Ensure that allocated ranges in the checkpoint's metaslab
8025 * space maps remain allocated in the metaslab space maps of
8026 * the current state.
8027 */
8028 verify_checkpoint_ms_spacemaps(checkpoint_spa, spa);
8029
8030 /*
8031 * Once we are done, we get rid of the checkpointed state.
8032 */
8033 spa_close(checkpoint_spa, FTAG);
8034 free(checkpoint_pool);
8035 }
8036
8037 static void
dump_leftover_checkpoint_blocks(spa_t * spa)8038 dump_leftover_checkpoint_blocks(spa_t *spa)
8039 {
8040 vdev_t *rvd = spa->spa_root_vdev;
8041
8042 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
8043 vdev_t *vd = rvd->vdev_child[i];
8044
8045 space_map_t *checkpoint_sm = NULL;
8046 uint64_t checkpoint_sm_obj;
8047
8048 if (vd->vdev_top_zap == 0)
8049 continue;
8050
8051 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
8052 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
8053 continue;
8054
8055 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
8056 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
8057 sizeof (uint64_t), 1, &checkpoint_sm_obj));
8058
8059 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
8060 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
8061 dump_spacemap(spa->spa_meta_objset, checkpoint_sm);
8062 space_map_close(checkpoint_sm);
8063 }
8064 }
8065
8066 static int
verify_checkpoint(spa_t * spa)8067 verify_checkpoint(spa_t *spa)
8068 {
8069 uberblock_t checkpoint;
8070 int error;
8071
8072 if (!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT))
8073 return (0);
8074
8075 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8076 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
8077 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
8078
8079 if (error == ENOENT && !dump_opt['L']) {
8080 /*
8081 * If the feature is active but the uberblock is missing
8082 * then we must be in the middle of discarding the
8083 * checkpoint.
8084 */
8085 (void) printf("\nPartially discarded checkpoint "
8086 "state found:\n");
8087 if (dump_opt['m'] > 3)
8088 dump_leftover_checkpoint_blocks(spa);
8089 return (0);
8090 } else if (error != 0) {
8091 (void) printf("lookup error %d when looking for "
8092 "checkpointed uberblock in MOS\n", error);
8093 return (error);
8094 }
8095 dump_uberblock(&checkpoint, "\nCheckpointed uberblock found:\n", "\n");
8096
8097 if (checkpoint.ub_checkpoint_txg == 0) {
8098 (void) printf("\nub_checkpoint_txg not set in checkpointed "
8099 "uberblock\n");
8100 error = 3;
8101 }
8102
8103 if (error == 0 && !dump_opt['L'])
8104 verify_checkpoint_blocks(spa);
8105
8106 return (error);
8107 }
8108
8109 static void
mos_leaks_cb(void * arg,uint64_t start,uint64_t size)8110 mos_leaks_cb(void *arg, uint64_t start, uint64_t size)
8111 {
8112 (void) arg;
8113 for (uint64_t i = start; i < size; i++) {
8114 (void) printf("MOS object %llu referenced but not allocated\n",
8115 (u_longlong_t)i);
8116 }
8117 }
8118
8119 static void
mos_obj_refd(uint64_t obj)8120 mos_obj_refd(uint64_t obj)
8121 {
8122 if (obj != 0 && mos_refd_objs != NULL)
8123 zfs_range_tree_add(mos_refd_objs, obj, 1);
8124 }
8125
8126 /*
8127 * Call on a MOS object that may already have been referenced.
8128 */
8129 static void
mos_obj_refd_multiple(uint64_t obj)8130 mos_obj_refd_multiple(uint64_t obj)
8131 {
8132 if (obj != 0 && mos_refd_objs != NULL &&
8133 !zfs_range_tree_contains(mos_refd_objs, obj, 1))
8134 zfs_range_tree_add(mos_refd_objs, obj, 1);
8135 }
8136
8137 static void
mos_leak_vdev_top_zap(vdev_t * vd)8138 mos_leak_vdev_top_zap(vdev_t *vd)
8139 {
8140 uint64_t ms_flush_data_obj;
8141 int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
8142 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
8143 sizeof (ms_flush_data_obj), 1, &ms_flush_data_obj);
8144 if (error == ENOENT)
8145 return;
8146 ASSERT0(error);
8147
8148 mos_obj_refd(ms_flush_data_obj);
8149 }
8150
8151 static void
mos_leak_vdev(vdev_t * vd)8152 mos_leak_vdev(vdev_t *vd)
8153 {
8154 mos_obj_refd(vd->vdev_dtl_object);
8155 mos_obj_refd(vd->vdev_ms_array);
8156 mos_obj_refd(vd->vdev_indirect_config.vic_births_object);
8157 mos_obj_refd(vd->vdev_indirect_config.vic_mapping_object);
8158 mos_obj_refd(vd->vdev_leaf_zap);
8159 if (vd->vdev_checkpoint_sm != NULL)
8160 mos_obj_refd(vd->vdev_checkpoint_sm->sm_object);
8161 if (vd->vdev_indirect_mapping != NULL) {
8162 mos_obj_refd(vd->vdev_indirect_mapping->
8163 vim_phys->vimp_counts_object);
8164 }
8165 if (vd->vdev_obsolete_sm != NULL)
8166 mos_obj_refd(vd->vdev_obsolete_sm->sm_object);
8167
8168 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
8169 metaslab_t *ms = vd->vdev_ms[m];
8170 mos_obj_refd(space_map_object(ms->ms_sm));
8171 }
8172
8173 if (vd->vdev_root_zap != 0)
8174 mos_obj_refd(vd->vdev_root_zap);
8175
8176 if (vd->vdev_top_zap != 0) {
8177 mos_obj_refd(vd->vdev_top_zap);
8178 mos_leak_vdev_top_zap(vd);
8179 }
8180
8181 for (uint64_t c = 0; c < vd->vdev_children; c++) {
8182 mos_leak_vdev(vd->vdev_child[c]);
8183 }
8184 }
8185
8186 static void
mos_leak_log_spacemaps(spa_t * spa)8187 mos_leak_log_spacemaps(spa_t *spa)
8188 {
8189 uint64_t spacemap_zap;
8190 int error = zap_lookup(spa_meta_objset(spa),
8191 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LOG_SPACEMAP_ZAP,
8192 sizeof (spacemap_zap), 1, &spacemap_zap);
8193 if (error == ENOENT)
8194 return;
8195 ASSERT0(error);
8196
8197 mos_obj_refd(spacemap_zap);
8198 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
8199 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls))
8200 mos_obj_refd(sls->sls_sm_obj);
8201 }
8202
8203 static void
errorlog_count_refd(objset_t * mos,uint64_t errlog)8204 errorlog_count_refd(objset_t *mos, uint64_t errlog)
8205 {
8206 zap_cursor_t zc;
8207 zap_attribute_t *za = zap_attribute_alloc();
8208 for (zap_cursor_init(&zc, mos, errlog);
8209 zap_cursor_retrieve(&zc, za) == 0;
8210 zap_cursor_advance(&zc)) {
8211 mos_obj_refd(za->za_first_integer);
8212 }
8213 zap_cursor_fini(&zc);
8214 zap_attribute_free(za);
8215 }
8216
8217 static int
dump_mos_leaks(spa_t * spa)8218 dump_mos_leaks(spa_t *spa)
8219 {
8220 int rv = 0;
8221 objset_t *mos = spa->spa_meta_objset;
8222 dsl_pool_t *dp = spa->spa_dsl_pool;
8223
8224 /* Visit and mark all referenced objects in the MOS */
8225
8226 mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT);
8227 mos_obj_refd(spa->spa_pool_props_object);
8228 mos_obj_refd(spa->spa_config_object);
8229 mos_obj_refd(spa->spa_ddt_stat_object);
8230 mos_obj_refd(spa->spa_feat_desc_obj);
8231 mos_obj_refd(spa->spa_feat_enabled_txg_obj);
8232 mos_obj_refd(spa->spa_feat_for_read_obj);
8233 mos_obj_refd(spa->spa_feat_for_write_obj);
8234 mos_obj_refd(spa->spa_history);
8235 mos_obj_refd(spa->spa_errlog_last);
8236 mos_obj_refd(spa->spa_errlog_scrub);
8237
8238 if (spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) {
8239 errorlog_count_refd(mos, spa->spa_errlog_last);
8240 errorlog_count_refd(mos, spa->spa_errlog_scrub);
8241 }
8242
8243 mos_obj_refd(spa->spa_all_vdev_zaps);
8244 mos_obj_refd(spa->spa_dsl_pool->dp_bptree_obj);
8245 mos_obj_refd(spa->spa_dsl_pool->dp_tmp_userrefs_obj);
8246 mos_obj_refd(spa->spa_dsl_pool->dp_scan->scn_phys.scn_queue_obj);
8247 bpobj_count_refd(&spa->spa_deferred_bpobj);
8248 mos_obj_refd(dp->dp_empty_bpobj);
8249 bpobj_count_refd(&dp->dp_obsolete_bpobj);
8250 bpobj_count_refd(&dp->dp_free_bpobj);
8251 mos_obj_refd(spa->spa_l2cache.sav_object);
8252 mos_obj_refd(spa->spa_spares.sav_object);
8253
8254 if (spa->spa_syncing_log_sm != NULL)
8255 mos_obj_refd(spa->spa_syncing_log_sm->sm_object);
8256 mos_leak_log_spacemaps(spa);
8257
8258 mos_obj_refd(spa->spa_condensing_indirect_phys.
8259 scip_next_mapping_object);
8260 mos_obj_refd(spa->spa_condensing_indirect_phys.
8261 scip_prev_obsolete_sm_object);
8262 if (spa->spa_condensing_indirect_phys.scip_next_mapping_object != 0) {
8263 vdev_indirect_mapping_t *vim =
8264 vdev_indirect_mapping_open(mos,
8265 spa->spa_condensing_indirect_phys.scip_next_mapping_object);
8266 mos_obj_refd(vim->vim_phys->vimp_counts_object);
8267 vdev_indirect_mapping_close(vim);
8268 }
8269 deleted_livelists_dump_mos(spa);
8270
8271 if (dp->dp_origin_snap != NULL) {
8272 dsl_dataset_t *ds;
8273
8274 dsl_pool_config_enter(dp, FTAG);
8275 VERIFY0(dsl_dataset_hold_obj(dp,
8276 dsl_dataset_phys(dp->dp_origin_snap)->ds_next_snap_obj,
8277 FTAG, &ds));
8278 count_ds_mos_objects(ds);
8279 dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
8280 dsl_dataset_rele(ds, FTAG);
8281 dsl_pool_config_exit(dp, FTAG);
8282
8283 count_ds_mos_objects(dp->dp_origin_snap);
8284 dump_blkptr_list(&dp->dp_origin_snap->ds_deadlist, "Deadlist");
8285 }
8286 count_dir_mos_objects(dp->dp_mos_dir);
8287 if (dp->dp_free_dir != NULL)
8288 count_dir_mos_objects(dp->dp_free_dir);
8289 if (dp->dp_leak_dir != NULL)
8290 count_dir_mos_objects(dp->dp_leak_dir);
8291
8292 mos_leak_vdev(spa->spa_root_vdev);
8293
8294 for (uint64_t c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
8295 ddt_t *ddt = spa->spa_ddt[c];
8296 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
8297 continue;
8298
8299 /* DDT store objects */
8300 for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
8301 for (ddt_class_t class = 0; class < DDT_CLASSES;
8302 class++) {
8303 mos_obj_refd(ddt->ddt_object[type][class]);
8304 }
8305 }
8306
8307 /* FDT container */
8308 if (ddt->ddt_version == DDT_VERSION_FDT)
8309 mos_obj_refd(ddt->ddt_dir_object);
8310
8311 /* FDT log objects */
8312 if (ddt->ddt_flags & DDT_FLAG_LOG) {
8313 mos_obj_refd(ddt->ddt_log[0].ddl_object);
8314 mos_obj_refd(ddt->ddt_log[1].ddl_object);
8315 }
8316 }
8317
8318 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
8319 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
8320 if (brtvd->bv_initiated) {
8321 mos_obj_refd(brtvd->bv_mos_brtvdev);
8322 mos_obj_refd(brtvd->bv_mos_entries);
8323 }
8324 }
8325
8326 /*
8327 * Visit all allocated objects and make sure they are referenced.
8328 */
8329 uint64_t object = 0;
8330 while (dmu_object_next(mos, &object, B_FALSE, 0) == 0) {
8331 if (zfs_range_tree_contains(mos_refd_objs, object, 1)) {
8332 zfs_range_tree_remove(mos_refd_objs, object, 1);
8333 } else {
8334 dmu_object_info_t doi;
8335 const char *name;
8336 VERIFY0(dmu_object_info(mos, object, &doi));
8337 if (doi.doi_type & DMU_OT_NEWTYPE) {
8338 dmu_object_byteswap_t bswap =
8339 DMU_OT_BYTESWAP(doi.doi_type);
8340 name = dmu_ot_byteswap[bswap].ob_name;
8341 } else {
8342 name = dmu_ot[doi.doi_type].ot_name;
8343 }
8344
8345 (void) printf("MOS object %llu (%s) leaked\n",
8346 (u_longlong_t)object, name);
8347 rv = 2;
8348 }
8349 }
8350 (void) zfs_range_tree_walk(mos_refd_objs, mos_leaks_cb, NULL);
8351 if (!zfs_range_tree_is_empty(mos_refd_objs))
8352 rv = 2;
8353 zfs_range_tree_vacate(mos_refd_objs, NULL, NULL);
8354 zfs_range_tree_destroy(mos_refd_objs);
8355 return (rv);
8356 }
8357
8358 typedef struct log_sm_obsolete_stats_arg {
8359 uint64_t lsos_current_txg;
8360
8361 uint64_t lsos_total_entries;
8362 uint64_t lsos_valid_entries;
8363
8364 uint64_t lsos_sm_entries;
8365 uint64_t lsos_valid_sm_entries;
8366 } log_sm_obsolete_stats_arg_t;
8367
8368 static int
log_spacemap_obsolete_stats_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)8369 log_spacemap_obsolete_stats_cb(spa_t *spa, space_map_entry_t *sme,
8370 uint64_t txg, void *arg)
8371 {
8372 log_sm_obsolete_stats_arg_t *lsos = arg;
8373
8374 uint64_t offset = sme->sme_offset;
8375 uint64_t vdev_id = sme->sme_vdev;
8376
8377 if (lsos->lsos_current_txg == 0) {
8378 /* this is the first log */
8379 lsos->lsos_current_txg = txg;
8380 } else if (lsos->lsos_current_txg < txg) {
8381 /* we just changed log - print stats and reset */
8382 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8383 (u_longlong_t)lsos->lsos_valid_sm_entries,
8384 (u_longlong_t)lsos->lsos_sm_entries,
8385 (u_longlong_t)lsos->lsos_current_txg);
8386 lsos->lsos_valid_sm_entries = 0;
8387 lsos->lsos_sm_entries = 0;
8388 lsos->lsos_current_txg = txg;
8389 }
8390 ASSERT3U(lsos->lsos_current_txg, ==, txg);
8391
8392 lsos->lsos_sm_entries++;
8393 lsos->lsos_total_entries++;
8394
8395 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
8396 if (!vdev_is_concrete(vd))
8397 return (0);
8398
8399 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
8400 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
8401
8402 if (txg < metaslab_unflushed_txg(ms))
8403 return (0);
8404 lsos->lsos_valid_sm_entries++;
8405 lsos->lsos_valid_entries++;
8406 return (0);
8407 }
8408
8409 static void
dump_log_spacemap_obsolete_stats(spa_t * spa)8410 dump_log_spacemap_obsolete_stats(spa_t *spa)
8411 {
8412 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
8413 return;
8414
8415 log_sm_obsolete_stats_arg_t lsos = {0};
8416
8417 (void) printf("Log Space Map Obsolete Entry Statistics:\n");
8418
8419 iterate_through_spacemap_logs(spa,
8420 log_spacemap_obsolete_stats_cb, &lsos);
8421
8422 /* print stats for latest log */
8423 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8424 (u_longlong_t)lsos.lsos_valid_sm_entries,
8425 (u_longlong_t)lsos.lsos_sm_entries,
8426 (u_longlong_t)lsos.lsos_current_txg);
8427
8428 (void) printf("%-8llu valid entries out of %-8llu - total\n\n",
8429 (u_longlong_t)lsos.lsos_valid_entries,
8430 (u_longlong_t)lsos.lsos_total_entries);
8431 }
8432
8433 static void
dump_zpool(spa_t * spa)8434 dump_zpool(spa_t *spa)
8435 {
8436 dsl_pool_t *dp = spa_get_dsl(spa);
8437 int rc = 0;
8438
8439 if (dump_opt['y']) {
8440 livelist_metaslab_validate(spa);
8441 }
8442
8443 if (dump_opt['S']) {
8444 dump_simulated_ddt(spa);
8445 return;
8446 }
8447
8448 if (!dump_opt['e'] && dump_opt['C'] > 1) {
8449 (void) printf("\nCached configuration:\n");
8450 dump_nvlist(spa->spa_config, 8);
8451 }
8452
8453 if (dump_opt['C'])
8454 dump_config(spa);
8455
8456 if (dump_opt['u'])
8457 dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n");
8458
8459 if (dump_opt['D'])
8460 dump_all_ddts(spa);
8461
8462 if (dump_opt['T'])
8463 dump_brt(spa);
8464
8465 if (dump_opt['d'] > 2 || dump_opt['m'])
8466 dump_metaslabs(spa);
8467 if (dump_opt['M'])
8468 dump_metaslab_groups(spa, dump_opt['M'] > 1);
8469 if (dump_opt['d'] > 2 || dump_opt['m']) {
8470 dump_log_spacemaps(spa);
8471 dump_log_spacemap_obsolete_stats(spa);
8472 }
8473
8474 if (dump_opt['d'] || dump_opt['i']) {
8475 spa_feature_t f;
8476 mos_refd_objs = zfs_range_tree_create_flags(
8477 NULL, ZFS_RANGE_SEG64, NULL, 0, 0,
8478 0, "dump_zpool:mos_refd_objs");
8479 dump_objset(dp->dp_meta_objset);
8480
8481 if (dump_opt['d'] >= 3) {
8482 dsl_pool_t *dp = spa->spa_dsl_pool;
8483 dump_full_bpobj(&spa->spa_deferred_bpobj,
8484 "Deferred frees", 0);
8485 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
8486 dump_full_bpobj(&dp->dp_free_bpobj,
8487 "Pool snapshot frees", 0);
8488 }
8489 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
8490 ASSERT(spa_feature_is_enabled(spa,
8491 SPA_FEATURE_DEVICE_REMOVAL));
8492 dump_full_bpobj(&dp->dp_obsolete_bpobj,
8493 "Pool obsolete blocks", 0);
8494 }
8495
8496 if (spa_feature_is_active(spa,
8497 SPA_FEATURE_ASYNC_DESTROY)) {
8498 dump_bptree(spa->spa_meta_objset,
8499 dp->dp_bptree_obj,
8500 "Pool dataset frees");
8501 }
8502 dump_dtl(spa->spa_root_vdev, 0);
8503 }
8504
8505 for (spa_feature_t f = 0; f < SPA_FEATURES; f++)
8506 global_feature_count[f] = UINT64_MAX;
8507 global_feature_count[SPA_FEATURE_REDACTION_BOOKMARKS] = 0;
8508 global_feature_count[SPA_FEATURE_REDACTION_LIST_SPILL] = 0;
8509 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN] = 0;
8510 global_feature_count[SPA_FEATURE_LIVELIST] = 0;
8511
8512 (void) dmu_objset_find(spa_name(spa), dump_one_objset,
8513 NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
8514
8515 if (rc == 0 && !dump_opt['L'])
8516 rc = dump_mos_leaks(spa);
8517
8518 for (f = 0; f < SPA_FEATURES; f++) {
8519 uint64_t refcount;
8520
8521 uint64_t *arr;
8522 if (!(spa_feature_table[f].fi_flags &
8523 ZFEATURE_FLAG_PER_DATASET)) {
8524 if (global_feature_count[f] == UINT64_MAX)
8525 continue;
8526 if (!spa_feature_is_enabled(spa, f)) {
8527 ASSERT0(global_feature_count[f]);
8528 continue;
8529 }
8530 arr = global_feature_count;
8531 } else {
8532 if (!spa_feature_is_enabled(spa, f)) {
8533 ASSERT0(dataset_feature_count[f]);
8534 continue;
8535 }
8536 arr = dataset_feature_count;
8537 }
8538 if (feature_get_refcount(spa, &spa_feature_table[f],
8539 &refcount) == ENOTSUP)
8540 continue;
8541 if (arr[f] != refcount) {
8542 (void) printf("%s feature refcount mismatch: "
8543 "%lld consumers != %lld refcount\n",
8544 spa_feature_table[f].fi_uname,
8545 (longlong_t)arr[f], (longlong_t)refcount);
8546 rc = 2;
8547 } else {
8548 (void) printf("Verified %s feature refcount "
8549 "of %llu is correct\n",
8550 spa_feature_table[f].fi_uname,
8551 (longlong_t)refcount);
8552 }
8553 }
8554
8555 if (rc == 0)
8556 rc = verify_device_removal_feature_counts(spa);
8557 }
8558
8559 if (rc == 0 && (dump_opt['b'] || dump_opt['c']))
8560 rc = dump_block_stats(spa);
8561
8562 if (rc == 0)
8563 rc = verify_spacemap_refcounts(spa);
8564
8565 if (dump_opt['s'])
8566 show_pool_stats(spa);
8567
8568 if (dump_opt['h'])
8569 dump_history(spa);
8570
8571 if (rc == 0)
8572 rc = verify_checkpoint(spa);
8573
8574 if (rc != 0) {
8575 dump_debug_buffer();
8576 zdb_exit(rc);
8577 }
8578 }
8579
8580 #define ZDB_FLAG_CHECKSUM 0x0001
8581 #define ZDB_FLAG_DECOMPRESS 0x0002
8582 #define ZDB_FLAG_BSWAP 0x0004
8583 #define ZDB_FLAG_GBH 0x0008
8584 #define ZDB_FLAG_INDIRECT 0x0010
8585 #define ZDB_FLAG_RAW 0x0020
8586 #define ZDB_FLAG_PRINT_BLKPTR 0x0040
8587 #define ZDB_FLAG_VERBOSE 0x0080
8588
8589 static int flagbits[256];
8590 static char flagbitstr[16];
8591
8592 static void
zdb_print_blkptr(const blkptr_t * bp,int flags)8593 zdb_print_blkptr(const blkptr_t *bp, int flags)
8594 {
8595 char blkbuf[BP_SPRINTF_LEN];
8596
8597 if (flags & ZDB_FLAG_BSWAP)
8598 byteswap_uint64_array((void *)bp, sizeof (blkptr_t));
8599
8600 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
8601 (void) printf("%s\n", blkbuf);
8602 }
8603
8604 static void
zdb_dump_indirect(blkptr_t * bp,int nbps,int flags)8605 zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
8606 {
8607 int i;
8608
8609 for (i = 0; i < nbps; i++)
8610 zdb_print_blkptr(&bp[i], flags);
8611 }
8612
8613 static void
zdb_dump_gbh(void * buf,uint64_t size,int flags)8614 zdb_dump_gbh(void *buf, uint64_t size, int flags)
8615 {
8616 zdb_dump_indirect((blkptr_t *)buf, gbh_nblkptrs(size), flags);
8617 }
8618
8619 static void
zdb_dump_block_raw(void * buf,uint64_t size,int flags)8620 zdb_dump_block_raw(void *buf, uint64_t size, int flags)
8621 {
8622 if (flags & ZDB_FLAG_BSWAP)
8623 byteswap_uint64_array(buf, size);
8624 VERIFY(write(fileno(stdout), buf, size) == size);
8625 }
8626
8627 static void
zdb_dump_block(char * label,void * buf,uint64_t size,int flags)8628 zdb_dump_block(char *label, void *buf, uint64_t size, int flags)
8629 {
8630 uint64_t *d = (uint64_t *)buf;
8631 unsigned nwords = size / sizeof (uint64_t);
8632 int do_bswap = !!(flags & ZDB_FLAG_BSWAP);
8633 unsigned i, j;
8634 const char *hdr;
8635 char *c;
8636
8637
8638 if (do_bswap)
8639 hdr = " 7 6 5 4 3 2 1 0 f e d c b a 9 8";
8640 else
8641 hdr = " 0 1 2 3 4 5 6 7 8 9 a b c d e f";
8642
8643 (void) printf("\n%s\n%6s %s 0123456789abcdef\n", label, "", hdr);
8644
8645 #ifdef _ZFS_LITTLE_ENDIAN
8646 /* correct the endianness */
8647 do_bswap = !do_bswap;
8648 #endif
8649 for (i = 0; i < nwords; i += 2) {
8650 (void) printf("%06llx: %016llx %016llx ",
8651 (u_longlong_t)(i * sizeof (uint64_t)),
8652 (u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]),
8653 (u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1]));
8654
8655 c = (char *)&d[i];
8656 for (j = 0; j < 2 * sizeof (uint64_t); j++)
8657 (void) printf("%c", isprint(c[j]) ? c[j] : '.');
8658 (void) printf("\n");
8659 }
8660 }
8661
8662 /*
8663 * There are two acceptable formats:
8664 * leaf_name - For example: c1t0d0 or /tmp/ztest.0a
8665 * child[.child]* - For example: 0.1.1
8666 *
8667 * The second form can be used to specify arbitrary vdevs anywhere
8668 * in the hierarchy. For example, in a pool with a mirror of
8669 * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
8670 */
8671 static vdev_t *
zdb_vdev_lookup(vdev_t * vdev,const char * path)8672 zdb_vdev_lookup(vdev_t *vdev, const char *path)
8673 {
8674 char *s, *p, *q;
8675 unsigned i;
8676
8677 if (vdev == NULL)
8678 return (NULL);
8679
8680 /* First, assume the x.x.x.x format */
8681 i = strtoul(path, &s, 10);
8682 if (s == path || (s && *s != '.' && *s != '\0'))
8683 goto name;
8684 if (i >= vdev->vdev_children)
8685 return (NULL);
8686
8687 vdev = vdev->vdev_child[i];
8688 if (s && *s == '\0')
8689 return (vdev);
8690 return (zdb_vdev_lookup(vdev, s+1));
8691
8692 name:
8693 for (i = 0; i < vdev->vdev_children; i++) {
8694 vdev_t *vc = vdev->vdev_child[i];
8695
8696 if (vc->vdev_path == NULL) {
8697 vc = zdb_vdev_lookup(vc, path);
8698 if (vc == NULL)
8699 continue;
8700 else
8701 return (vc);
8702 }
8703
8704 p = strrchr(vc->vdev_path, '/');
8705 p = p ? p + 1 : vc->vdev_path;
8706 q = &vc->vdev_path[strlen(vc->vdev_path) - 2];
8707
8708 if (strcmp(vc->vdev_path, path) == 0)
8709 return (vc);
8710 if (strcmp(p, path) == 0)
8711 return (vc);
8712 if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0)
8713 return (vc);
8714 }
8715
8716 return (NULL);
8717 }
8718
8719 static int
name_from_objset_id(spa_t * spa,uint64_t objset_id,char * outstr)8720 name_from_objset_id(spa_t *spa, uint64_t objset_id, char *outstr)
8721 {
8722 dsl_dataset_t *ds;
8723
8724 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
8725 int error = dsl_dataset_hold_obj(spa->spa_dsl_pool, objset_id,
8726 NULL, &ds);
8727 if (error != 0) {
8728 (void) fprintf(stderr, "failed to hold objset %llu: %s\n",
8729 (u_longlong_t)objset_id, strerror(error));
8730 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8731 return (error);
8732 }
8733 dsl_dataset_name(ds, outstr);
8734 dsl_dataset_rele(ds, NULL);
8735 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8736 return (0);
8737 }
8738
8739 static boolean_t
zdb_parse_block_sizes(char * sizes,uint64_t * lsize,uint64_t * psize)8740 zdb_parse_block_sizes(char *sizes, uint64_t *lsize, uint64_t *psize)
8741 {
8742 char *s0, *s1, *tmp = NULL;
8743
8744 if (sizes == NULL)
8745 return (B_FALSE);
8746
8747 s0 = strtok_r(sizes, "/", &tmp);
8748 if (s0 == NULL)
8749 return (B_FALSE);
8750 s1 = strtok_r(NULL, "/", &tmp);
8751 *lsize = strtoull(s0, NULL, 16);
8752 *psize = s1 ? strtoull(s1, NULL, 16) : *lsize;
8753 return (*lsize >= *psize && *psize > 0);
8754 }
8755
8756 #define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg))
8757
8758 static boolean_t
try_decompress_block(abd_t * pabd,uint64_t lsize,uint64_t psize,int flags,int cfunc,void * lbuf,void * lbuf2)8759 try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize,
8760 int flags, int cfunc, void *lbuf, void *lbuf2)
8761 {
8762 if (flags & ZDB_FLAG_VERBOSE) {
8763 (void) fprintf(stderr,
8764 "Trying %05llx -> %05llx (%s)\n",
8765 (u_longlong_t)psize,
8766 (u_longlong_t)lsize,
8767 zio_compress_table[cfunc].ci_name);
8768 }
8769
8770 /*
8771 * We set lbuf to all zeros and lbuf2 to all
8772 * ones, then decompress to both buffers and
8773 * compare their contents. This way we can
8774 * know if decompression filled exactly to
8775 * lsize or if it left some bytes unwritten.
8776 */
8777
8778 memset(lbuf, 0x00, lsize);
8779 memset(lbuf2, 0xff, lsize);
8780
8781 abd_t labd, labd2;
8782 abd_get_from_buf_struct(&labd, lbuf, lsize);
8783 abd_get_from_buf_struct(&labd2, lbuf2, lsize);
8784
8785 boolean_t ret = B_FALSE;
8786 if (zio_decompress_data(cfunc, pabd,
8787 &labd, psize, lsize, NULL) == 0 &&
8788 zio_decompress_data(cfunc, pabd,
8789 &labd2, psize, lsize, NULL) == 0 &&
8790 memcmp(lbuf, lbuf2, lsize) == 0)
8791 ret = B_TRUE;
8792
8793 abd_free(&labd2);
8794 abd_free(&labd);
8795
8796 return (ret);
8797 }
8798
8799 static uint64_t
zdb_decompress_block(abd_t * pabd,void * buf,void * lbuf,uint64_t lsize,uint64_t psize,int flags)8800 zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
8801 uint64_t psize, int flags)
8802 {
8803 (void) buf;
8804 uint64_t orig_lsize = lsize;
8805 boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL));
8806 /*
8807 * We don't know how the data was compressed, so just try
8808 * every decompress function at every inflated blocksize.
8809 */
8810 void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
8811 int cfuncs[ZIO_COMPRESS_FUNCTIONS] = { 0 };
8812 int *cfuncp = cfuncs;
8813 uint64_t maxlsize = SPA_MAXBLOCKSIZE;
8814 uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) |
8815 ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) |
8816 ZIO_COMPRESS_MASK(ZLE);
8817 *cfuncp++ = ZIO_COMPRESS_LZ4;
8818 *cfuncp++ = ZIO_COMPRESS_LZJB;
8819 mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB);
8820 /*
8821 * Every gzip level has the same decompressor, no need to
8822 * run it 9 times per bruteforce attempt.
8823 */
8824 mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3);
8825 mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5);
8826 mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7);
8827 mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9);
8828 for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++)
8829 if (((1ULL << c) & mask) == 0)
8830 *cfuncp++ = c;
8831
8832 /*
8833 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
8834 * could take a while and we should let the user know
8835 * we are not stuck. On the other hand, printing progress
8836 * info gets old after a while. User can specify 'v' flag
8837 * to see the progression.
8838 */
8839 if (lsize == psize)
8840 lsize += SPA_MINBLOCKSIZE;
8841 else
8842 maxlsize = lsize;
8843
8844 for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) {
8845 for (cfuncp = cfuncs; *cfuncp; cfuncp++) {
8846 if (try_decompress_block(pabd, lsize, psize, flags,
8847 *cfuncp, lbuf, lbuf2)) {
8848 tryzle = B_FALSE;
8849 break;
8850 }
8851 }
8852 if (*cfuncp != 0)
8853 break;
8854 }
8855 if (tryzle) {
8856 for (lsize = orig_lsize; lsize <= maxlsize;
8857 lsize += SPA_MINBLOCKSIZE) {
8858 if (try_decompress_block(pabd, lsize, psize, flags,
8859 ZIO_COMPRESS_ZLE, lbuf, lbuf2)) {
8860 *cfuncp = ZIO_COMPRESS_ZLE;
8861 break;
8862 }
8863 }
8864 }
8865 umem_free(lbuf2, SPA_MAXBLOCKSIZE);
8866
8867 if (*cfuncp == ZIO_COMPRESS_ZLE) {
8868 printf("\nZLE decompression was selected. If you "
8869 "suspect the results are wrong,\ntry avoiding ZLE "
8870 "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
8871 }
8872
8873 return (lsize > maxlsize ? -1 : lsize);
8874 }
8875
8876 /*
8877 * Read a block from a pool and print it out. The syntax of the
8878 * block descriptor is:
8879 *
8880 * pool:vdev_specifier:offset:[lsize/]psize[:flags]
8881 *
8882 * pool - The name of the pool you wish to read from
8883 * vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
8884 * offset - offset, in hex, in bytes
8885 * size - Amount of data to read, in hex, in bytes
8886 * flags - A string of characters specifying options
8887 * b: Decode a blkptr at given offset within block
8888 * c: Calculate and display checksums
8889 * d: Decompress data before dumping
8890 * e: Byteswap data before dumping
8891 * g: Display data as a gang block header
8892 * i: Display as an indirect block
8893 * r: Dump raw data to stdout
8894 * v: Verbose
8895 *
8896 */
8897 static void
zdb_read_block(char * thing,spa_t * spa)8898 zdb_read_block(char *thing, spa_t *spa)
8899 {
8900 blkptr_t blk, *bp = &blk;
8901 dva_t *dva = bp->blk_dva;
8902 int flags = 0;
8903 uint64_t offset = 0, psize = 0, lsize = 0, blkptr_offset = 0;
8904 zio_t *zio;
8905 vdev_t *vd;
8906 abd_t *pabd;
8907 void *lbuf, *buf;
8908 char *s, *p, *dup, *flagstr, *sizes, *tmp = NULL;
8909 const char *vdev, *errmsg = NULL;
8910 int i, len, error;
8911 boolean_t borrowed = B_FALSE, found = B_FALSE;
8912
8913 dup = strdup(thing);
8914 s = strtok_r(dup, ":", &tmp);
8915 vdev = s ?: "";
8916 s = strtok_r(NULL, ":", &tmp);
8917 offset = strtoull(s ? s : "", NULL, 16);
8918 sizes = strtok_r(NULL, ":", &tmp);
8919 s = strtok_r(NULL, ":", &tmp);
8920 flagstr = strdup(s ?: "");
8921
8922 if (!zdb_parse_block_sizes(sizes, &lsize, &psize))
8923 errmsg = "invalid size(s)";
8924 if (!IS_P2ALIGNED(psize, DEV_BSIZE) || !IS_P2ALIGNED(lsize, DEV_BSIZE))
8925 errmsg = "size must be a multiple of sector size";
8926 if (!IS_P2ALIGNED(offset, DEV_BSIZE))
8927 errmsg = "offset must be a multiple of sector size";
8928 if (errmsg) {
8929 (void) printf("Invalid block specifier: %s - %s\n",
8930 thing, errmsg);
8931 goto done;
8932 }
8933
8934 tmp = NULL;
8935 for (s = strtok_r(flagstr, ":", &tmp);
8936 s != NULL;
8937 s = strtok_r(NULL, ":", &tmp)) {
8938 len = strlen(flagstr);
8939 for (i = 0; i < len; i++) {
8940 int bit = flagbits[(uchar_t)flagstr[i]];
8941
8942 if (bit == 0) {
8943 (void) printf("***Ignoring flag: %c\n",
8944 (uchar_t)flagstr[i]);
8945 continue;
8946 }
8947 found = B_TRUE;
8948 flags |= bit;
8949
8950 p = &flagstr[i + 1];
8951 if (*p != ':' && *p != '\0') {
8952 int j = 0, nextbit = flagbits[(uchar_t)*p];
8953 char *end, offstr[8] = { 0 };
8954 if ((bit == ZDB_FLAG_PRINT_BLKPTR) &&
8955 (nextbit == 0)) {
8956 /* look ahead to isolate the offset */
8957 while (nextbit == 0 &&
8958 strchr(flagbitstr, *p) == NULL) {
8959 offstr[j] = *p;
8960 j++;
8961 if (i + j > strlen(flagstr))
8962 break;
8963 p++;
8964 nextbit = flagbits[(uchar_t)*p];
8965 }
8966 blkptr_offset = strtoull(offstr, &end,
8967 16);
8968 i += j;
8969 } else if (nextbit == 0) {
8970 (void) printf("***Ignoring flag arg:"
8971 " '%c'\n", (uchar_t)*p);
8972 }
8973 }
8974 }
8975 }
8976 if (blkptr_offset % sizeof (blkptr_t)) {
8977 printf("Block pointer offset 0x%llx "
8978 "must be divisible by 0x%x\n",
8979 (longlong_t)blkptr_offset, (int)sizeof (blkptr_t));
8980 goto done;
8981 }
8982 if (found == B_FALSE && strlen(flagstr) > 0) {
8983 printf("Invalid flag arg: '%s'\n", flagstr);
8984 goto done;
8985 }
8986
8987 vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev);
8988 if (vd == NULL) {
8989 (void) printf("***Invalid vdev: %s\n", vdev);
8990 goto done;
8991 } else {
8992 if (vd->vdev_path)
8993 (void) fprintf(stderr, "Found vdev: %s\n",
8994 vd->vdev_path);
8995 else
8996 (void) fprintf(stderr, "Found vdev type: %s\n",
8997 vd->vdev_ops->vdev_op_type);
8998 }
8999
9000 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
9001 lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
9002
9003 BP_ZERO(bp);
9004
9005 DVA_SET_VDEV(&dva[0], vd->vdev_id);
9006 DVA_SET_OFFSET(&dva[0], offset);
9007 DVA_SET_GANG(&dva[0], 0);
9008 DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
9009
9010 BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
9011
9012 BP_SET_LSIZE(bp, lsize);
9013 BP_SET_PSIZE(bp, psize);
9014 BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF);
9015 BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF);
9016 BP_SET_TYPE(bp, DMU_OT_NONE);
9017 BP_SET_LEVEL(bp, 0);
9018 BP_SET_DEDUP(bp, 0);
9019 BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
9020
9021 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9022 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9023
9024 if (vd == vd->vdev_top) {
9025 /*
9026 * Treat this as a normal block read.
9027 */
9028 zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL,
9029 ZIO_PRIORITY_SYNC_READ,
9030 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL));
9031 } else {
9032 /*
9033 * Treat this as a vdev child I/O.
9034 */
9035 zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd,
9036 psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ,
9037 ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
9038 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | ZIO_FLAG_OPTIONAL,
9039 NULL, NULL));
9040 }
9041
9042 error = zio_wait(zio);
9043 spa_config_exit(spa, SCL_STATE, FTAG);
9044
9045 if (error) {
9046 (void) printf("Read of %s failed, error: %d\n", thing, error);
9047 goto out;
9048 }
9049
9050 uint64_t orig_lsize = lsize;
9051 buf = lbuf;
9052 if (flags & ZDB_FLAG_DECOMPRESS) {
9053 lsize = zdb_decompress_block(pabd, buf, lbuf,
9054 lsize, psize, flags);
9055 if (lsize == -1) {
9056 (void) printf("Decompress of %s failed\n", thing);
9057 goto out;
9058 }
9059 } else {
9060 buf = abd_borrow_buf_copy(pabd, lsize);
9061 borrowed = B_TRUE;
9062 }
9063 /*
9064 * Try to detect invalid block pointer. If invalid, try
9065 * decompressing.
9066 */
9067 if ((flags & ZDB_FLAG_PRINT_BLKPTR || flags & ZDB_FLAG_INDIRECT) &&
9068 !(flags & ZDB_FLAG_DECOMPRESS)) {
9069 const blkptr_t *b = (const blkptr_t *)(void *)
9070 ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9071 if (zfs_blkptr_verify(spa, b,
9072 BLK_CONFIG_NEEDED, BLK_VERIFY_ONLY)) {
9073 abd_return_buf_copy(pabd, buf, lsize);
9074 borrowed = B_FALSE;
9075 buf = lbuf;
9076 lsize = zdb_decompress_block(pabd, buf,
9077 lbuf, lsize, psize, flags);
9078 b = (const blkptr_t *)(void *)
9079 ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9080 if (lsize == -1 || zfs_blkptr_verify(spa, b,
9081 BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) {
9082 printf("invalid block pointer at this DVA\n");
9083 goto out;
9084 }
9085 }
9086 }
9087
9088 if (flags & ZDB_FLAG_PRINT_BLKPTR)
9089 zdb_print_blkptr((blkptr_t *)(void *)
9090 ((uintptr_t)buf + (uintptr_t)blkptr_offset), flags);
9091 else if (flags & ZDB_FLAG_RAW)
9092 zdb_dump_block_raw(buf, lsize, flags);
9093 else if (flags & ZDB_FLAG_INDIRECT)
9094 zdb_dump_indirect((blkptr_t *)buf,
9095 orig_lsize / sizeof (blkptr_t), flags);
9096 else if (flags & ZDB_FLAG_GBH)
9097 zdb_dump_gbh(buf, lsize, flags);
9098 else
9099 zdb_dump_block(thing, buf, lsize, flags);
9100
9101 /*
9102 * If :c was specified, iterate through the checksum table to
9103 * calculate and display each checksum for our specified
9104 * DVA and length.
9105 */
9106 if ((flags & ZDB_FLAG_CHECKSUM) && !(flags & ZDB_FLAG_RAW) &&
9107 !(flags & ZDB_FLAG_GBH)) {
9108 zio_t *czio;
9109 (void) printf("\n");
9110 for (enum zio_checksum ck = ZIO_CHECKSUM_LABEL;
9111 ck < ZIO_CHECKSUM_FUNCTIONS; ck++) {
9112
9113 if ((zio_checksum_table[ck].ci_flags &
9114 ZCHECKSUM_FLAG_EMBEDDED) ||
9115 ck == ZIO_CHECKSUM_NOPARITY) {
9116 continue;
9117 }
9118 BP_SET_CHECKSUM(bp, ck);
9119 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9120 czio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9121 if (vd == vd->vdev_top) {
9122 zio_nowait(zio_read(czio, spa, bp, pabd, psize,
9123 NULL, NULL,
9124 ZIO_PRIORITY_SYNC_READ,
9125 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9126 ZIO_FLAG_DONT_RETRY, NULL));
9127 } else {
9128 zio_nowait(zio_vdev_child_io(czio, bp, vd,
9129 offset, pabd, psize, ZIO_TYPE_READ,
9130 ZIO_PRIORITY_SYNC_READ,
9131 ZIO_FLAG_DONT_PROPAGATE |
9132 ZIO_FLAG_DONT_RETRY |
9133 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9134 ZIO_FLAG_SPECULATIVE |
9135 ZIO_FLAG_OPTIONAL, NULL, NULL));
9136 }
9137 error = zio_wait(czio);
9138 if (error == 0 || error == ECKSUM) {
9139 zio_t *ck_zio = zio_null(NULL, spa, NULL,
9140 NULL, NULL, 0);
9141 ck_zio->io_offset =
9142 DVA_GET_OFFSET(&bp->blk_dva[0]);
9143 ck_zio->io_bp = bp;
9144 zio_checksum_compute(ck_zio, ck, pabd, psize);
9145 printf(
9146 "%12s\t"
9147 "cksum=%016llx:%016llx:%016llx:%016llx\n",
9148 zio_checksum_table[ck].ci_name,
9149 (u_longlong_t)bp->blk_cksum.zc_word[0],
9150 (u_longlong_t)bp->blk_cksum.zc_word[1],
9151 (u_longlong_t)bp->blk_cksum.zc_word[2],
9152 (u_longlong_t)bp->blk_cksum.zc_word[3]);
9153 zio_wait(ck_zio);
9154 } else {
9155 printf("error %d reading block\n", error);
9156 }
9157 spa_config_exit(spa, SCL_STATE, FTAG);
9158 }
9159 }
9160
9161 if (borrowed)
9162 abd_return_buf_copy(pabd, buf, lsize);
9163
9164 out:
9165 abd_free(pabd);
9166 umem_free(lbuf, SPA_MAXBLOCKSIZE);
9167 done:
9168 free(flagstr);
9169 free(dup);
9170 }
9171
9172 static void
zdb_embedded_block(char * thing)9173 zdb_embedded_block(char *thing)
9174 {
9175 blkptr_t bp = {{{{0}}}};
9176 unsigned long long *words = (void *)&bp;
9177 char *buf;
9178 int err;
9179
9180 err = sscanf(thing, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
9181 "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
9182 words + 0, words + 1, words + 2, words + 3,
9183 words + 4, words + 5, words + 6, words + 7,
9184 words + 8, words + 9, words + 10, words + 11,
9185 words + 12, words + 13, words + 14, words + 15);
9186 if (err != 16) {
9187 (void) fprintf(stderr, "invalid input format\n");
9188 zdb_exit(1);
9189 }
9190 ASSERT3U(BPE_GET_LSIZE(&bp), <=, SPA_MAXBLOCKSIZE);
9191 buf = malloc(SPA_MAXBLOCKSIZE);
9192 if (buf == NULL) {
9193 (void) fprintf(stderr, "out of memory\n");
9194 zdb_exit(1);
9195 }
9196 err = decode_embedded_bp(&bp, buf, BPE_GET_LSIZE(&bp));
9197 if (err != 0) {
9198 (void) fprintf(stderr, "decode failed: %u\n", err);
9199 zdb_exit(1);
9200 }
9201 zdb_dump_block_raw(buf, BPE_GET_LSIZE(&bp), 0);
9202 free(buf);
9203 }
9204
9205 /* check for valid hex or decimal numeric string */
9206 static boolean_t
zdb_numeric(char * str)9207 zdb_numeric(char *str)
9208 {
9209 int i = 0, len;
9210
9211 len = strlen(str);
9212 if (len == 0)
9213 return (B_FALSE);
9214 if (strncmp(str, "0x", 2) == 0 || strncmp(str, "0X", 2) == 0)
9215 i = 2;
9216 for (; i < len; i++) {
9217 if (!isxdigit(str[i]))
9218 return (B_FALSE);
9219 }
9220 return (B_TRUE);
9221 }
9222
9223 static int
dummy_get_file_info(dmu_object_type_t bonustype,const void * data,zfs_file_info_t * zoi)9224 dummy_get_file_info(dmu_object_type_t bonustype, const void *data,
9225 zfs_file_info_t *zoi)
9226 {
9227 (void) data, (void) zoi;
9228
9229 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
9230 return (ENOENT);
9231
9232 (void) fprintf(stderr, "dummy_get_file_info: not implemented");
9233 abort();
9234 }
9235
9236 int
main(int argc,char ** argv)9237 main(int argc, char **argv)
9238 {
9239 int c;
9240 int dump_all = 1;
9241 int verbose = 0;
9242 int error = 0;
9243 char **searchdirs = NULL;
9244 int nsearch = 0;
9245 char *target, *target_pool, dsname[ZFS_MAX_DATASET_NAME_LEN];
9246 nvlist_t *policy = NULL;
9247 uint64_t max_txg = UINT64_MAX;
9248 int64_t objset_id = -1;
9249 uint64_t object;
9250 int flags = ZFS_IMPORT_MISSING_LOG;
9251 int rewind = ZPOOL_NEVER_REWIND;
9252 char *spa_config_path_env, *objset_str;
9253 boolean_t target_is_spa = B_TRUE, dataset_lookup = B_FALSE;
9254 nvlist_t *cfg = NULL;
9255 struct sigaction action;
9256 boolean_t force_import = B_FALSE;
9257 boolean_t config_path_console = B_FALSE;
9258 char pbuf[MAXPATHLEN];
9259
9260 dprintf_setup(&argc, argv);
9261
9262 /*
9263 * Set up signal handlers, so if we crash due to bad on-disk data we
9264 * can get more info. Unlike ztest, we don't bail out if we can't set
9265 * up signal handlers, because zdb is very useful without them.
9266 */
9267 action.sa_handler = sig_handler;
9268 sigemptyset(&action.sa_mask);
9269 action.sa_flags = 0;
9270 if (sigaction(SIGSEGV, &action, NULL) < 0) {
9271 (void) fprintf(stderr, "zdb: cannot catch SIGSEGV: %s\n",
9272 strerror(errno));
9273 }
9274 if (sigaction(SIGABRT, &action, NULL) < 0) {
9275 (void) fprintf(stderr, "zdb: cannot catch SIGABRT: %s\n",
9276 strerror(errno));
9277 }
9278
9279 /*
9280 * If there is an environment variable SPA_CONFIG_PATH it overrides
9281 * default spa_config_path setting. If -U flag is specified it will
9282 * override this environment variable settings once again.
9283 */
9284 spa_config_path_env = getenv("SPA_CONFIG_PATH");
9285 if (spa_config_path_env != NULL)
9286 spa_config_path = spa_config_path_env;
9287
9288 /*
9289 * For performance reasons, we set this tunable down. We do so before
9290 * the arg parsing section so that the user can override this value if
9291 * they choose.
9292 */
9293 zfs_btree_verify_intensity = 3;
9294
9295 struct option long_options[] = {
9296 {"ignore-assertions", no_argument, NULL, 'A'},
9297 {"block-stats", no_argument, NULL, 'b'},
9298 {"backup", no_argument, NULL, 'B'},
9299 {"checksum", no_argument, NULL, 'c'},
9300 {"config", no_argument, NULL, 'C'},
9301 {"datasets", no_argument, NULL, 'd'},
9302 {"dedup-stats", no_argument, NULL, 'D'},
9303 {"exported", no_argument, NULL, 'e'},
9304 {"embedded-block-pointer", no_argument, NULL, 'E'},
9305 {"automatic-rewind", no_argument, NULL, 'F'},
9306 {"dump-debug-msg", no_argument, NULL, 'G'},
9307 {"history", no_argument, NULL, 'h'},
9308 {"intent-logs", no_argument, NULL, 'i'},
9309 {"inflight", required_argument, NULL, 'I'},
9310 {"checkpointed-state", no_argument, NULL, 'k'},
9311 {"key", required_argument, NULL, 'K'},
9312 {"label", no_argument, NULL, 'l'},
9313 {"disable-leak-tracking", no_argument, NULL, 'L'},
9314 {"metaslabs", no_argument, NULL, 'm'},
9315 {"metaslab-groups", no_argument, NULL, 'M'},
9316 {"numeric", no_argument, NULL, 'N'},
9317 {"option", required_argument, NULL, 'o'},
9318 {"object-lookups", no_argument, NULL, 'O'},
9319 {"path", required_argument, NULL, 'p'},
9320 {"parseable", no_argument, NULL, 'P'},
9321 {"skip-label", no_argument, NULL, 'q'},
9322 {"copy-object", no_argument, NULL, 'r'},
9323 {"read-block", no_argument, NULL, 'R'},
9324 {"io-stats", no_argument, NULL, 's'},
9325 {"simulate-dedup", no_argument, NULL, 'S'},
9326 {"txg", required_argument, NULL, 't'},
9327 {"brt-stats", no_argument, NULL, 'T'},
9328 {"uberblock", no_argument, NULL, 'u'},
9329 {"cachefile", required_argument, NULL, 'U'},
9330 {"verbose", no_argument, NULL, 'v'},
9331 {"verbatim", no_argument, NULL, 'V'},
9332 {"dump-blocks", required_argument, NULL, 'x'},
9333 {"extreme-rewind", no_argument, NULL, 'X'},
9334 {"all-reconstruction", no_argument, NULL, 'Y'},
9335 {"livelist", no_argument, NULL, 'y'},
9336 {"zstd-headers", no_argument, NULL, 'Z'},
9337 {0, 0, 0, 0}
9338 };
9339
9340 while ((c = getopt_long(argc, argv,
9341 "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
9342 long_options, NULL)) != -1) {
9343 switch (c) {
9344 case 'b':
9345 case 'B':
9346 case 'c':
9347 case 'C':
9348 case 'd':
9349 case 'D':
9350 case 'E':
9351 case 'G':
9352 case 'h':
9353 case 'i':
9354 case 'l':
9355 case 'm':
9356 case 'M':
9357 case 'N':
9358 case 'O':
9359 case 'r':
9360 case 'R':
9361 case 's':
9362 case 'S':
9363 case 'T':
9364 case 'u':
9365 case 'y':
9366 case 'Z':
9367 dump_opt[c]++;
9368 dump_all = 0;
9369 break;
9370 case 'A':
9371 case 'e':
9372 case 'F':
9373 case 'k':
9374 case 'L':
9375 case 'P':
9376 case 'q':
9377 case 'X':
9378 dump_opt[c]++;
9379 break;
9380 case 'Y':
9381 zfs_reconstruct_indirect_combinations_max = INT_MAX;
9382 zfs_deadman_enabled = 0;
9383 break;
9384 /* NB: Sort single match options below. */
9385 case 'I':
9386 max_inflight_bytes = strtoull(optarg, NULL, 0);
9387 if (max_inflight_bytes == 0) {
9388 (void) fprintf(stderr, "maximum number "
9389 "of inflight bytes must be greater "
9390 "than 0\n");
9391 usage();
9392 }
9393 break;
9394 case 'K':
9395 dump_opt[c]++;
9396 key_material = strdup(optarg);
9397 /* redact key material in process table */
9398 while (*optarg != '\0') { *optarg++ = '*'; }
9399 break;
9400 case 'o':
9401 dump_opt[c]++;
9402 dump_all = 0;
9403 error = handle_tunable_option(optarg, B_FALSE);
9404 if (error != 0)
9405 zdb_exit(1);
9406 break;
9407 case 'p':
9408 if (searchdirs == NULL) {
9409 searchdirs = umem_alloc(sizeof (char *),
9410 UMEM_NOFAIL);
9411 } else {
9412 char **tmp = umem_alloc((nsearch + 1) *
9413 sizeof (char *), UMEM_NOFAIL);
9414 memcpy(tmp, searchdirs, nsearch *
9415 sizeof (char *));
9416 umem_free(searchdirs,
9417 nsearch * sizeof (char *));
9418 searchdirs = tmp;
9419 }
9420 searchdirs[nsearch++] = optarg;
9421 break;
9422 case 't':
9423 max_txg = strtoull(optarg, NULL, 0);
9424 if (max_txg < TXG_INITIAL) {
9425 (void) fprintf(stderr, "incorrect txg "
9426 "specified: %s\n", optarg);
9427 usage();
9428 }
9429 break;
9430 case 'U':
9431 config_path_console = B_TRUE;
9432 spa_config_path = optarg;
9433 if (spa_config_path[0] != '/') {
9434 (void) fprintf(stderr,
9435 "cachefile must be an absolute path "
9436 "(i.e. start with a slash)\n");
9437 usage();
9438 }
9439 break;
9440 case 'v':
9441 verbose++;
9442 break;
9443 case 'V':
9444 flags = ZFS_IMPORT_VERBATIM;
9445 break;
9446 case 'x':
9447 vn_dumpdir = optarg;
9448 break;
9449 default:
9450 usage();
9451 break;
9452 }
9453 }
9454
9455 if (!dump_opt['e'] && searchdirs != NULL) {
9456 (void) fprintf(stderr, "-p option requires use of -e\n");
9457 usage();
9458 }
9459 #if defined(_LP64)
9460 /*
9461 * ZDB does not typically re-read blocks; therefore limit the ARC
9462 * to 256 MB, which can be used entirely for metadata.
9463 */
9464 zfs_arc_min = 2ULL << SPA_MAXBLOCKSHIFT;
9465 zfs_arc_max = 256 * 1024 * 1024;
9466 #endif
9467
9468 /*
9469 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
9470 * "zdb -b" uses traversal prefetch which uses async reads.
9471 * For good performance, let several of them be active at once.
9472 */
9473 zfs_vdev_async_read_max_active = 10;
9474
9475 /*
9476 * Disable reference tracking for better performance.
9477 */
9478 reference_tracking_enable = B_FALSE;
9479
9480 /*
9481 * Do not fail spa_load when spa_load_verify fails. This is needed
9482 * to load non-idle pools.
9483 */
9484 spa_load_verify_dryrun = B_TRUE;
9485
9486 /*
9487 * ZDB should have ability to read spacemaps.
9488 */
9489 spa_mode_readable_spacemaps = B_TRUE;
9490
9491 if (dump_all)
9492 verbose = MAX(verbose, 1);
9493
9494 for (c = 0; c < 256; c++) {
9495 if (dump_all && strchr("ABeEFkKlLNOPrRSXy", c) == NULL)
9496 dump_opt[c] = 1;
9497 if (dump_opt[c])
9498 dump_opt[c] += verbose;
9499 }
9500
9501 libspl_set_assert_ok((dump_opt['A'] == 1) || (dump_opt['A'] > 2));
9502 zfs_recover = (dump_opt['A'] > 1);
9503
9504 argc -= optind;
9505 argv += optind;
9506 if (argc < 2 && dump_opt['R'])
9507 usage();
9508
9509 target = argv[0];
9510
9511 /*
9512 * Automate cachefile
9513 */
9514 if (!spa_config_path_env && !config_path_console && target &&
9515 libzfs_core_init() == 0) {
9516 char *pname = strdup(target);
9517 const char *value;
9518 nvlist_t *pnvl = NULL;
9519 nvlist_t *vnvl = NULL;
9520
9521 if (strpbrk(pname, "/@") != NULL)
9522 *strpbrk(pname, "/@") = '\0';
9523
9524 if (pname && lzc_get_props(pname, &pnvl) == 0) {
9525 if (nvlist_lookup_nvlist(pnvl, "cachefile",
9526 &vnvl) == 0) {
9527 value = fnvlist_lookup_string(vnvl,
9528 ZPROP_VALUE);
9529 } else {
9530 value = "-";
9531 }
9532 strlcpy(pbuf, value, sizeof (pbuf));
9533 if (pbuf[0] != '\0') {
9534 if (pbuf[0] == '/') {
9535 if (access(pbuf, F_OK) == 0)
9536 spa_config_path = pbuf;
9537 else
9538 force_import = B_TRUE;
9539 } else if ((strcmp(pbuf, "-") == 0 &&
9540 access(ZPOOL_CACHE, F_OK) != 0) ||
9541 strcmp(pbuf, "none") == 0) {
9542 force_import = B_TRUE;
9543 }
9544 }
9545 nvlist_free(vnvl);
9546 }
9547
9548 free(pname);
9549 nvlist_free(pnvl);
9550 libzfs_core_fini();
9551 }
9552
9553 dmu_objset_register_type(DMU_OST_ZFS, dummy_get_file_info);
9554 kernel_init(SPA_MODE_READ);
9555 kernel_init_done = B_TRUE;
9556
9557 if (dump_opt['E']) {
9558 if (argc != 1)
9559 usage();
9560 zdb_embedded_block(argv[0]);
9561 error = 0;
9562 goto fini;
9563 }
9564
9565 if (argc < 1) {
9566 if (!dump_opt['e'] && dump_opt['C']) {
9567 dump_cachefile(spa_config_path);
9568 error = 0;
9569 goto fini;
9570 }
9571 if (dump_opt['o'])
9572 /*
9573 * Avoid blasting tunable options off the top of the
9574 * screen.
9575 */
9576 zdb_exit(1);
9577 usage();
9578 }
9579
9580 if (dump_opt['l']) {
9581 error = dump_label(argv[0]);
9582 goto fini;
9583 }
9584
9585 if (dump_opt['X'] || dump_opt['F'])
9586 rewind = ZPOOL_DO_REWIND |
9587 (dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
9588
9589 /* -N implies -d */
9590 if (dump_opt['N'] && dump_opt['d'] == 0)
9591 dump_opt['d'] = dump_opt['N'];
9592
9593 if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 ||
9594 nvlist_add_uint64(policy, ZPOOL_LOAD_REQUEST_TXG, max_txg) != 0 ||
9595 nvlist_add_uint32(policy, ZPOOL_LOAD_REWIND_POLICY, rewind) != 0)
9596 fatal("internal error: %s", strerror(ENOMEM));
9597
9598 error = 0;
9599
9600 if (strpbrk(target, "/@") != NULL) {
9601 size_t targetlen;
9602
9603 target_pool = strdup(target);
9604 *strpbrk(target_pool, "/@") = '\0';
9605
9606 target_is_spa = B_FALSE;
9607 targetlen = strlen(target);
9608 if (targetlen && target[targetlen - 1] == '/')
9609 target[targetlen - 1] = '\0';
9610
9611 /*
9612 * See if an objset ID was supplied (-d <pool>/<objset ID>).
9613 * To disambiguate tank/100, consider the 100 as objsetID
9614 * if -N was given, otherwise 100 is an objsetID iff
9615 * tank/100 as a named dataset fails on lookup.
9616 */
9617 objset_str = strchr(target, '/');
9618 if (objset_str && strlen(objset_str) > 1 &&
9619 zdb_numeric(objset_str + 1)) {
9620 char *endptr;
9621 errno = 0;
9622 objset_str++;
9623 objset_id = strtoull(objset_str, &endptr, 0);
9624 /* dataset 0 is the same as opening the pool */
9625 if (errno == 0 && endptr != objset_str &&
9626 objset_id != 0) {
9627 if (dump_opt['N'])
9628 dataset_lookup = B_TRUE;
9629 }
9630 /* normal dataset name not an objset ID */
9631 if (endptr == objset_str) {
9632 objset_id = -1;
9633 }
9634 } else if (objset_str && !zdb_numeric(objset_str + 1) &&
9635 dump_opt['N']) {
9636 printf("Supply a numeric objset ID with -N\n");
9637 error = 1;
9638 goto fini;
9639 }
9640 } else {
9641 target_pool = target;
9642 }
9643
9644 if (dump_opt['e'] || force_import) {
9645 importargs_t args = { 0 };
9646
9647 /*
9648 * If path is not provided, search in /dev
9649 */
9650 if (searchdirs == NULL) {
9651 searchdirs = umem_alloc(sizeof (char *), UMEM_NOFAIL);
9652 searchdirs[nsearch++] = (char *)ZFS_DEVDIR;
9653 }
9654
9655 args.paths = nsearch;
9656 args.path = searchdirs;
9657 args.can_be_active = B_TRUE;
9658
9659 libpc_handle_t lpch = {
9660 .lpc_lib_handle = NULL,
9661 .lpc_ops = &libzpool_config_ops,
9662 .lpc_printerr = B_TRUE
9663 };
9664 error = zpool_find_config(&lpch, target_pool, &cfg, &args);
9665
9666 if (error == 0) {
9667
9668 if (nvlist_add_nvlist(cfg,
9669 ZPOOL_LOAD_POLICY, policy) != 0) {
9670 fatal("can't open '%s': %s",
9671 target, strerror(ENOMEM));
9672 }
9673
9674 if (dump_opt['C'] > 1) {
9675 (void) printf("\nConfiguration for import:\n");
9676 dump_nvlist(cfg, 8);
9677 }
9678
9679 /*
9680 * Disable the activity check to allow examination of
9681 * active pools.
9682 */
9683 error = spa_import(target_pool, cfg, NULL,
9684 flags | ZFS_IMPORT_SKIP_MMP);
9685 }
9686 }
9687
9688 if (searchdirs != NULL) {
9689 umem_free(searchdirs, nsearch * sizeof (char *));
9690 searchdirs = NULL;
9691 }
9692
9693 /*
9694 * We need to make sure to process -O option or call
9695 * dump_path after the -e option has been processed,
9696 * which imports the pool to the namespace if it's
9697 * not in the cachefile.
9698 */
9699 if (dump_opt['O']) {
9700 if (argc != 2)
9701 usage();
9702 dump_opt['v'] = verbose + 3;
9703 error = dump_path(argv[0], argv[1], NULL);
9704 goto fini;
9705 }
9706
9707 if (dump_opt['r']) {
9708 target_is_spa = B_FALSE;
9709 if (argc != 3)
9710 usage();
9711 dump_opt['v'] = verbose;
9712 error = dump_path(argv[0], argv[1], &object);
9713 if (error != 0)
9714 fatal("internal error: %s", strerror(error));
9715 }
9716
9717 /*
9718 * import_checkpointed_state makes the assumption that the
9719 * target pool that we pass it is already part of the spa
9720 * namespace. Because of that we need to make sure to call
9721 * it always after the -e option has been processed, which
9722 * imports the pool to the namespace if it's not in the
9723 * cachefile.
9724 */
9725 char *checkpoint_pool = NULL;
9726 char *checkpoint_target = NULL;
9727 if (dump_opt['k']) {
9728 checkpoint_pool = import_checkpointed_state(target, cfg,
9729 target_is_spa, &checkpoint_target);
9730
9731 if (checkpoint_target != NULL)
9732 target = checkpoint_target;
9733 }
9734
9735 if (cfg != NULL) {
9736 nvlist_free(cfg);
9737 cfg = NULL;
9738 }
9739
9740 if (target_pool != target)
9741 free(target_pool);
9742
9743 if (error == 0) {
9744 if (dump_opt['k'] && (target_is_spa || dump_opt['R'])) {
9745 ASSERT(checkpoint_pool != NULL);
9746 ASSERT(checkpoint_target == NULL);
9747
9748 error = spa_open(checkpoint_pool, &spa, FTAG);
9749 if (error != 0) {
9750 fatal("Tried to open pool \"%s\" but "
9751 "spa_open() failed with error %d\n",
9752 checkpoint_pool, error);
9753 }
9754
9755 } else if (target_is_spa || dump_opt['R'] || dump_opt['B'] ||
9756 objset_id == 0) {
9757 zdb_set_skip_mmp(target);
9758 error = spa_open_rewind(target, &spa, FTAG, policy,
9759 NULL);
9760 if (error) {
9761 /*
9762 * If we're missing the log device then
9763 * try opening the pool after clearing the
9764 * log state.
9765 */
9766 mutex_enter(&spa_namespace_lock);
9767 if ((spa = spa_lookup(target)) != NULL &&
9768 spa->spa_log_state == SPA_LOG_MISSING) {
9769 spa->spa_log_state = SPA_LOG_CLEAR;
9770 error = 0;
9771 }
9772 mutex_exit(&spa_namespace_lock);
9773
9774 if (!error) {
9775 error = spa_open_rewind(target, &spa,
9776 FTAG, policy, NULL);
9777 }
9778 }
9779 } else if (strpbrk(target, "#") != NULL) {
9780 dsl_pool_t *dp;
9781 error = dsl_pool_hold(target, FTAG, &dp);
9782 if (error != 0) {
9783 fatal("can't dump '%s': %s", target,
9784 strerror(error));
9785 }
9786 error = dump_bookmark(dp, target, B_TRUE, verbose > 1);
9787 dsl_pool_rele(dp, FTAG);
9788 if (error != 0) {
9789 fatal("can't dump '%s': %s", target,
9790 strerror(error));
9791 }
9792 goto fini;
9793 } else {
9794 target_pool = strdup(target);
9795 if (strpbrk(target, "/@") != NULL)
9796 *strpbrk(target_pool, "/@") = '\0';
9797
9798 zdb_set_skip_mmp(target);
9799 /*
9800 * If -N was supplied, the user has indicated that
9801 * zdb -d <pool>/<objsetID> is in effect. Otherwise
9802 * we first assume that the dataset string is the
9803 * dataset name. If dmu_objset_hold fails with the
9804 * dataset string, and we have an objset_id, retry the
9805 * lookup with the objsetID.
9806 */
9807 boolean_t retry = B_TRUE;
9808 retry_lookup:
9809 if (dataset_lookup == B_TRUE) {
9810 /*
9811 * Use the supplied id to get the name
9812 * for open_objset.
9813 */
9814 error = spa_open(target_pool, &spa, FTAG);
9815 if (error == 0) {
9816 error = name_from_objset_id(spa,
9817 objset_id, dsname);
9818 spa_close(spa, FTAG);
9819 if (error == 0)
9820 target = dsname;
9821 }
9822 }
9823 if (error == 0) {
9824 if (objset_id > 0 && retry) {
9825 int err = dmu_objset_hold(target, FTAG,
9826 &os);
9827 if (err) {
9828 dataset_lookup = B_TRUE;
9829 retry = B_FALSE;
9830 goto retry_lookup;
9831 } else {
9832 dmu_objset_rele(os, FTAG);
9833 }
9834 }
9835 error = open_objset(target, FTAG, &os);
9836 }
9837 if (error == 0)
9838 spa = dmu_objset_spa(os);
9839 free(target_pool);
9840 }
9841 }
9842 nvlist_free(policy);
9843
9844 if (error)
9845 fatal("can't open '%s': %s", target, strerror(error));
9846
9847 /*
9848 * Set the pool failure mode to panic in order to prevent the pool
9849 * from suspending. A suspended I/O will have no way to resume and
9850 * can prevent the zdb(8) command from terminating as expected.
9851 */
9852 if (spa != NULL)
9853 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
9854
9855 argv++;
9856 argc--;
9857 if (dump_opt['r']) {
9858 error = zdb_copy_object(os, object, argv[1]);
9859 } else if (!dump_opt['R']) {
9860 flagbits['d'] = ZOR_FLAG_DIRECTORY;
9861 flagbits['f'] = ZOR_FLAG_PLAIN_FILE;
9862 flagbits['m'] = ZOR_FLAG_SPACE_MAP;
9863 flagbits['z'] = ZOR_FLAG_ZAP;
9864 flagbits['A'] = ZOR_FLAG_ALL_TYPES;
9865
9866 if (argc > 0 && dump_opt['d']) {
9867 zopt_object_args = argc;
9868 zopt_object_ranges = calloc(zopt_object_args,
9869 sizeof (zopt_object_range_t));
9870 for (unsigned i = 0; i < zopt_object_args; i++) {
9871 int err;
9872 const char *msg = NULL;
9873
9874 err = parse_object_range(argv[i],
9875 &zopt_object_ranges[i], &msg);
9876 if (err != 0)
9877 fatal("Bad object or range: '%s': %s\n",
9878 argv[i], msg ?: "");
9879 }
9880 } else if (argc > 0 && dump_opt['m']) {
9881 zopt_metaslab_args = argc;
9882 zopt_metaslab = calloc(zopt_metaslab_args,
9883 sizeof (uint64_t));
9884 for (unsigned i = 0; i < zopt_metaslab_args; i++) {
9885 errno = 0;
9886 zopt_metaslab[i] = strtoull(argv[i], NULL, 0);
9887 if (zopt_metaslab[i] == 0 && errno != 0)
9888 fatal("bad number %s: %s", argv[i],
9889 strerror(errno));
9890 }
9891 }
9892 if (dump_opt['B']) {
9893 dump_backup(target, objset_id,
9894 argc > 0 ? argv[0] : NULL);
9895 } else if (os != NULL) {
9896 dump_objset(os);
9897 } else if (zopt_object_args > 0 && !dump_opt['m']) {
9898 dump_objset(spa->spa_meta_objset);
9899 } else {
9900 dump_zpool(spa);
9901 }
9902 } else {
9903 flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR;
9904 flagbits['c'] = ZDB_FLAG_CHECKSUM;
9905 flagbits['d'] = ZDB_FLAG_DECOMPRESS;
9906 flagbits['e'] = ZDB_FLAG_BSWAP;
9907 flagbits['g'] = ZDB_FLAG_GBH;
9908 flagbits['i'] = ZDB_FLAG_INDIRECT;
9909 flagbits['r'] = ZDB_FLAG_RAW;
9910 flagbits['v'] = ZDB_FLAG_VERBOSE;
9911
9912 for (int i = 0; i < argc; i++)
9913 zdb_read_block(argv[i], spa);
9914 }
9915
9916 if (dump_opt['k']) {
9917 free(checkpoint_pool);
9918 if (!target_is_spa)
9919 free(checkpoint_target);
9920 }
9921
9922 fini:
9923 if (spa != NULL)
9924 zdb_ddt_cleanup(spa);
9925
9926 if (os != NULL) {
9927 close_objset(os, FTAG);
9928 } else if (spa != NULL) {
9929 spa_close(spa, FTAG);
9930 }
9931
9932 fuid_table_destroy();
9933
9934 dump_debug_buffer();
9935
9936 if (kernel_init_done)
9937 kernel_fini();
9938
9939 return (error);
9940 }
9941