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 * Copyright (c) 2026, TrueNAS.
40 */
41
42 #include <stdio.h>
43 #include <unistd.h>
44 #include <stdlib.h>
45 #include <ctype.h>
46 #include <getopt.h>
47 #include <openssl/evp.h>
48 #include <sys/zfs_context.h>
49 #include <sys/spa.h>
50 #include <sys/spa_impl.h>
51 #include <sys/dmu.h>
52 #include <sys/zap.h>
53 #include <sys/zap_impl.h>
54 #include <sys/fs/zfs.h>
55 #include <sys/zfs_znode.h>
56 #include <sys/zfs_sa.h>
57 #include <sys/sa.h>
58 #include <sys/sa_impl.h>
59 #include <sys/vdev.h>
60 #include <sys/vdev_impl.h>
61 #include <sys/metaslab_impl.h>
62 #include <sys/dmu_objset.h>
63 #include <sys/dsl_dir.h>
64 #include <sys/dsl_dataset.h>
65 #include <sys/dsl_pool.h>
66 #include <sys/dsl_bookmark.h>
67 #include <sys/dbuf.h>
68 #include <sys/zil.h>
69 #include <sys/zil_impl.h>
70 #include <sys/stat.h>
71 #include <sys/resource.h>
72 #include <sys/dmu_send.h>
73 #include <sys/dmu_traverse.h>
74 #include <sys/zio_checksum.h>
75 #include <sys/zio_compress.h>
76 #include <sys/zfs_fuid.h>
77 #include <sys/arc.h>
78 #include <sys/arc_impl.h>
79 #include <sys/ddt.h>
80 #include <sys/ddt_impl.h>
81 #include <sys/zfeature.h>
82 #include <sys/abd.h>
83 #include <sys/blkptr.h>
84 #include <sys/dsl_crypt.h>
85 #include <sys/dsl_scan.h>
86 #include <sys/btree.h>
87 #include <sys/brt.h>
88 #include <sys/brt_impl.h>
89 #include <zfs_comutil.h>
90 #include <sys/vdev_raidz.h>
91 #include <sys/vdev_raidz_impl.h>
92 #include <sys/zstd/zstd.h>
93 #include <sys/backtrace.h>
94
95 #include <libzpool.h>
96 #include <libnvpair.h>
97 #include <libzutil.h>
98 #include <libzfs_core.h>
99
100 #include <libzdb.h>
101
102 #include "zdb.h"
103
104
105 extern int reference_tracking_enable;
106 extern int zfs_recover;
107 extern uint_t zfs_vdev_async_read_max_active;
108 extern boolean_t spa_load_verify_dryrun;
109 extern boolean_t spa_mode_readable_spacemaps;
110 extern uint_t zfs_reconstruct_indirect_combinations_max;
111 extern uint_t zfs_btree_verify_intensity;
112
113 enum {
114 ARG_ALLOCATED = 256,
115 ARG_BLOCK_BIN_MODE,
116 ARG_BLOCK_CLASSES,
117 };
118
119 static const char cmdname[] = "zdb";
120 uint8_t dump_opt[512];
121
122 typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size);
123
124 static uint64_t *zopt_metaslab = NULL;
125 static unsigned zopt_metaslab_args = 0;
126
127
128 static zopt_object_range_t *zopt_object_ranges = NULL;
129 static unsigned zopt_object_args = 0;
130
131 static int flagbits[256];
132
133
134 static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */
135 static int leaked_objects = 0;
136 static zfs_range_tree_t *mos_refd_objs;
137 static spa_t *spa;
138 static objset_t *os;
139 static boolean_t kernel_init_done;
140 static boolean_t corruption_found = B_FALSE;
141
142 static enum {
143 BIN_AUTO = 0,
144 BIN_PSIZE,
145 BIN_LSIZE,
146 BIN_ASIZE,
147 } block_bin_mode = BIN_AUTO;
148
149 static enum {
150 CLASS_NORMAL = 1 << 1,
151 CLASS_SPECIAL = 1 << 2,
152 CLASS_DEDUP = 1 << 3,
153 CLASS_OTHER = 1 << 4,
154 } block_classes = 0;
155
156 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *,
157 boolean_t);
158 static void mos_obj_refd(uint64_t);
159 static void mos_obj_refd_multiple(uint64_t);
160 static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free,
161 dmu_tx_t *tx);
162
163
164
165 static void zdb_print_blkptr(const blkptr_t *bp, int flags);
166 static void zdb_exit(int reason);
167
168 typedef struct sublivelist_verify_block_refcnt {
169 /* block pointer entry in livelist being verified */
170 blkptr_t svbr_blk;
171
172 /*
173 * Refcount gets incremented to 1 when we encounter the first
174 * FREE entry for the svfbr block pointer and a node for it
175 * is created in our ZDB verification/tracking metadata.
176 *
177 * As we encounter more FREE entries we increment this counter
178 * and similarly decrement it whenever we find the respective
179 * ALLOC entries for this block.
180 *
181 * When the refcount gets to 0 it means that all the FREE and
182 * ALLOC entries of this block have paired up and we no longer
183 * need to track it in our verification logic (e.g. the node
184 * containing this struct in our verification data structure
185 * should be freed).
186 *
187 * [refer to sublivelist_verify_blkptr() for the actual code]
188 */
189 uint32_t svbr_refcnt;
190 } sublivelist_verify_block_refcnt_t;
191
192 static int
sublivelist_block_refcnt_compare(const void * larg,const void * rarg)193 sublivelist_block_refcnt_compare(const void *larg, const void *rarg)
194 {
195 const sublivelist_verify_block_refcnt_t *l = larg;
196 const sublivelist_verify_block_refcnt_t *r = rarg;
197 return (livelist_compare(&l->svbr_blk, &r->svbr_blk));
198 }
199
200 static int
sublivelist_verify_blkptr(void * arg,const blkptr_t * bp,boolean_t free,dmu_tx_t * tx)201 sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free,
202 dmu_tx_t *tx)
203 {
204 ASSERT0P(tx);
205 struct sublivelist_verify *sv = arg;
206 sublivelist_verify_block_refcnt_t current = {
207 .svbr_blk = *bp,
208
209 /*
210 * Start with 1 in case this is the first free entry.
211 * This field is not used for our B-Tree comparisons
212 * anyway.
213 */
214 .svbr_refcnt = 1,
215 };
216
217 zfs_btree_index_t where;
218 sublivelist_verify_block_refcnt_t *pair =
219 zfs_btree_find(&sv->sv_pair, ¤t, &where);
220 if (free) {
221 if (pair == NULL) {
222 /* first free entry for this block pointer */
223 zfs_btree_add(&sv->sv_pair, ¤t);
224 } else {
225 pair->svbr_refcnt++;
226 }
227 } else {
228 if (pair == NULL) {
229 /* block that is currently marked as allocated */
230 for (int i = 0; i < SPA_DVAS_PER_BP; i++) {
231 if (DVA_IS_EMPTY(&bp->blk_dva[i]))
232 break;
233 sublivelist_verify_block_t svb = {
234 .svb_dva = bp->blk_dva[i],
235 .svb_allocated_txg =
236 BP_GET_BIRTH(bp)
237 };
238
239 if (zfs_btree_find(&sv->sv_leftover, &svb,
240 &where) == NULL) {
241 zfs_btree_add_idx(&sv->sv_leftover,
242 &svb, &where);
243 }
244 }
245 } else {
246 /* alloc matches a free entry */
247 pair->svbr_refcnt--;
248 if (pair->svbr_refcnt == 0) {
249 /* all allocs and frees have been matched */
250 zfs_btree_remove_idx(&sv->sv_pair, &where);
251 }
252 }
253 }
254
255 return (0);
256 }
257
258 static int
sublivelist_verify_func(void * args,dsl_deadlist_entry_t * dle)259 sublivelist_verify_func(void *args, dsl_deadlist_entry_t *dle)
260 {
261 int err;
262 struct sublivelist_verify *sv = args;
263
264 zfs_btree_create(&sv->sv_pair, sublivelist_block_refcnt_compare, NULL,
265 sizeof (sublivelist_verify_block_refcnt_t));
266
267 err = bpobj_iterate_nofree(&dle->dle_bpobj, sublivelist_verify_blkptr,
268 sv, NULL);
269
270 sublivelist_verify_block_refcnt_t *e;
271 zfs_btree_index_t *cookie = NULL;
272 while ((e = zfs_btree_destroy_nodes(&sv->sv_pair, &cookie)) != NULL) {
273 char blkbuf[BP_SPRINTF_LEN];
274 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
275 &e->svbr_blk, B_TRUE);
276 (void) printf("\tERROR: %d unmatched FREE(s): %s\n",
277 e->svbr_refcnt, blkbuf);
278 corruption_found = B_TRUE;
279 }
280 zfs_btree_destroy(&sv->sv_pair);
281
282 return (err);
283 }
284
285 static int
livelist_block_compare(const void * larg,const void * rarg)286 livelist_block_compare(const void *larg, const void *rarg)
287 {
288 const sublivelist_verify_block_t *l = larg;
289 const sublivelist_verify_block_t *r = rarg;
290
291 if (DVA_GET_VDEV(&l->svb_dva) < DVA_GET_VDEV(&r->svb_dva))
292 return (-1);
293 else if (DVA_GET_VDEV(&l->svb_dva) > DVA_GET_VDEV(&r->svb_dva))
294 return (+1);
295
296 if (DVA_GET_OFFSET(&l->svb_dva) < DVA_GET_OFFSET(&r->svb_dva))
297 return (-1);
298 else if (DVA_GET_OFFSET(&l->svb_dva) > DVA_GET_OFFSET(&r->svb_dva))
299 return (+1);
300
301 if (DVA_GET_ASIZE(&l->svb_dva) < DVA_GET_ASIZE(&r->svb_dva))
302 return (-1);
303 else if (DVA_GET_ASIZE(&l->svb_dva) > DVA_GET_ASIZE(&r->svb_dva))
304 return (+1);
305
306 return (0);
307 }
308
309 /*
310 * Check for errors in a livelist while tracking all unfreed ALLOCs in the
311 * sublivelist_verify_t: sv->sv_leftover
312 */
313 static void
livelist_verify(dsl_deadlist_t * dl,void * arg)314 livelist_verify(dsl_deadlist_t *dl, void *arg)
315 {
316 sublivelist_verify_t *sv = arg;
317 dsl_deadlist_iterate(dl, sublivelist_verify_func, sv);
318 }
319
320 /*
321 * Check for errors in the livelist entry and discard the intermediary
322 * data structures
323 */
324 static int
sublivelist_verify_lightweight(void * args,dsl_deadlist_entry_t * dle)325 sublivelist_verify_lightweight(void *args, dsl_deadlist_entry_t *dle)
326 {
327 (void) args;
328 sublivelist_verify_t sv;
329 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
330 sizeof (sublivelist_verify_block_t));
331 int err = sublivelist_verify_func(&sv, dle);
332 zfs_btree_clear(&sv.sv_leftover);
333 zfs_btree_destroy(&sv.sv_leftover);
334 return (err);
335 }
336
337 typedef struct metaslab_verify {
338 /*
339 * Tree containing all the leftover ALLOCs from the livelists
340 * that are part of this metaslab.
341 */
342 zfs_btree_t mv_livelist_allocs;
343
344 /*
345 * Metaslab information.
346 */
347 uint64_t mv_vdid;
348 uint64_t mv_msid;
349 uint64_t mv_start;
350 uint64_t mv_end;
351
352 /*
353 * What's currently allocated for this metaslab.
354 */
355 zfs_range_tree_t *mv_allocated;
356 } metaslab_verify_t;
357
358 typedef void ll_iter_t(dsl_deadlist_t *ll, void *arg);
359
360 typedef int (*zdb_log_sm_cb_t)(spa_t *spa, space_map_entry_t *sme, uint64_t txg,
361 void *arg);
362
363 typedef struct unflushed_iter_cb_arg {
364 spa_t *uic_spa;
365 uint64_t uic_txg;
366 void *uic_arg;
367 zdb_log_sm_cb_t uic_cb;
368 } unflushed_iter_cb_arg_t;
369
370 static int
iterate_through_spacemap_logs_cb(space_map_entry_t * sme,void * arg)371 iterate_through_spacemap_logs_cb(space_map_entry_t *sme, void *arg)
372 {
373 unflushed_iter_cb_arg_t *uic = arg;
374 return (uic->uic_cb(uic->uic_spa, sme, uic->uic_txg, uic->uic_arg));
375 }
376
377 static void
iterate_through_spacemap_logs(spa_t * spa,zdb_log_sm_cb_t cb,void * arg)378 iterate_through_spacemap_logs(spa_t *spa, zdb_log_sm_cb_t cb, void *arg)
379 {
380 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
381 return;
382
383 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
384 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
385 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
386 space_map_t *sm = NULL;
387 VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
388 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
389
390 unflushed_iter_cb_arg_t uic = {
391 .uic_spa = spa,
392 .uic_txg = sls->sls_txg,
393 .uic_arg = arg,
394 .uic_cb = cb
395 };
396 VERIFY0(space_map_iterate(sm, space_map_length(sm),
397 iterate_through_spacemap_logs_cb, &uic));
398 space_map_close(sm);
399 }
400 spa_config_exit(spa, SCL_CONFIG, FTAG);
401 }
402
403 static void
verify_livelist_allocs(metaslab_verify_t * mv,uint64_t txg,uint64_t offset,uint64_t size)404 verify_livelist_allocs(metaslab_verify_t *mv, uint64_t txg,
405 uint64_t offset, uint64_t size)
406 {
407 sublivelist_verify_block_t svb = {{{0}}};
408 DVA_SET_VDEV(&svb.svb_dva, mv->mv_vdid);
409 DVA_SET_OFFSET(&svb.svb_dva, offset);
410 DVA_SET_ASIZE(&svb.svb_dva, 0);
411 zfs_btree_index_t where;
412 uint64_t end_offset = offset + size;
413
414 /*
415 * Look for an exact match for spacemap entry in the livelist entries.
416 * Then, look for other livelist entries that fall within the range
417 * of the spacemap entry as it may have been condensed
418 */
419 sublivelist_verify_block_t *found =
420 zfs_btree_find(&mv->mv_livelist_allocs, &svb, &where);
421 if (found == NULL) {
422 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where);
423 }
424 for (; found != NULL && DVA_GET_VDEV(&found->svb_dva) == mv->mv_vdid &&
425 DVA_GET_OFFSET(&found->svb_dva) < end_offset;
426 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
427 if (found->svb_allocated_txg <= txg) {
428 (void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
429 "from TXG %llx FREED at TXG %llx\n",
430 (u_longlong_t)DVA_GET_OFFSET(&found->svb_dva),
431 (u_longlong_t)DVA_GET_ASIZE(&found->svb_dva),
432 (u_longlong_t)found->svb_allocated_txg,
433 (u_longlong_t)txg);
434 corruption_found = B_TRUE;
435 }
436 }
437 }
438
439 static int
metaslab_spacemap_validation_cb(space_map_entry_t * sme,void * arg)440 metaslab_spacemap_validation_cb(space_map_entry_t *sme, void *arg)
441 {
442 metaslab_verify_t *mv = arg;
443 uint64_t offset = sme->sme_offset;
444 uint64_t size = sme->sme_run;
445 uint64_t txg = sme->sme_txg;
446
447 if (sme->sme_type == SM_ALLOC) {
448 if (zfs_range_tree_contains(mv->mv_allocated,
449 offset, size)) {
450 (void) printf("ERROR: DOUBLE ALLOC: "
451 "%llu [%llx:%llx] "
452 "%llu:%llu LOG_SM\n",
453 (u_longlong_t)txg, (u_longlong_t)offset,
454 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
455 (u_longlong_t)mv->mv_msid);
456 corruption_found = B_TRUE;
457 } else {
458 zfs_range_tree_add(mv->mv_allocated,
459 offset, size);
460 }
461 } else {
462 if (!zfs_range_tree_contains(mv->mv_allocated,
463 offset, size)) {
464 (void) printf("ERROR: DOUBLE FREE: "
465 "%llu [%llx:%llx] "
466 "%llu:%llu LOG_SM\n",
467 (u_longlong_t)txg, (u_longlong_t)offset,
468 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
469 (u_longlong_t)mv->mv_msid);
470 corruption_found = B_TRUE;
471 } else {
472 zfs_range_tree_remove(mv->mv_allocated,
473 offset, size);
474 }
475 }
476
477 if (sme->sme_type != SM_ALLOC) {
478 /*
479 * If something is freed in the spacemap, verify that
480 * it is not listed as allocated in the livelist.
481 */
482 verify_livelist_allocs(mv, txg, offset, size);
483 }
484 return (0);
485 }
486
487 static int
spacemap_check_sm_log_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)488 spacemap_check_sm_log_cb(spa_t *spa, space_map_entry_t *sme,
489 uint64_t txg, void *arg)
490 {
491 metaslab_verify_t *mv = arg;
492 uint64_t offset = sme->sme_offset;
493 uint64_t vdev_id = sme->sme_vdev;
494
495 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
496
497 /* skip indirect vdevs */
498 if (!vdev_is_concrete(vd))
499 return (0);
500
501 if (vdev_id != mv->mv_vdid)
502 return (0);
503
504 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
505 if (ms->ms_id != mv->mv_msid)
506 return (0);
507
508 if (txg < metaslab_unflushed_txg(ms))
509 return (0);
510
511
512 ASSERT3U(txg, ==, sme->sme_txg);
513 return (metaslab_spacemap_validation_cb(sme, mv));
514 }
515
516 static void
spacemap_check_sm_log(spa_t * spa,metaslab_verify_t * mv)517 spacemap_check_sm_log(spa_t *spa, metaslab_verify_t *mv)
518 {
519 iterate_through_spacemap_logs(spa, spacemap_check_sm_log_cb, mv);
520 }
521
522 static void
spacemap_check_ms_sm(space_map_t * sm,metaslab_verify_t * mv)523 spacemap_check_ms_sm(space_map_t *sm, metaslab_verify_t *mv)
524 {
525 if (sm == NULL)
526 return;
527
528 VERIFY0(space_map_iterate(sm, space_map_length(sm),
529 metaslab_spacemap_validation_cb, mv));
530 }
531
532 static void iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg);
533
534 /*
535 * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
536 * they are part of that metaslab (mv_msid).
537 */
538 static void
mv_populate_livelist_allocs(metaslab_verify_t * mv,sublivelist_verify_t * sv)539 mv_populate_livelist_allocs(metaslab_verify_t *mv, sublivelist_verify_t *sv)
540 {
541 zfs_btree_index_t where;
542 sublivelist_verify_block_t *svb;
543 ASSERT3U(zfs_btree_numnodes(&mv->mv_livelist_allocs), ==, 0);
544 for (svb = zfs_btree_first(&sv->sv_leftover, &where);
545 svb != NULL;
546 svb = zfs_btree_next(&sv->sv_leftover, &where, &where)) {
547 if (DVA_GET_VDEV(&svb->svb_dva) != mv->mv_vdid)
548 continue;
549
550 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start &&
551 (DVA_GET_OFFSET(&svb->svb_dva) +
552 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_start) {
553 (void) printf("ERROR: Found block that crosses "
554 "metaslab boundary: <%llu:%llx:%llx>\n",
555 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
556 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
557 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
558 corruption_found = B_TRUE;
559 continue;
560 }
561
562 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start)
563 continue;
564
565 if (DVA_GET_OFFSET(&svb->svb_dva) >= mv->mv_end)
566 continue;
567
568 if ((DVA_GET_OFFSET(&svb->svb_dva) +
569 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_end) {
570 (void) printf("ERROR: Found block that crosses "
571 "metaslab boundary: <%llu:%llx:%llx>\n",
572 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
573 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
574 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
575 corruption_found = B_TRUE;
576 continue;
577 }
578
579 zfs_btree_add(&mv->mv_livelist_allocs, svb);
580 }
581
582 for (svb = zfs_btree_first(&mv->mv_livelist_allocs, &where);
583 svb != NULL;
584 svb = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
585 zfs_btree_remove(&sv->sv_leftover, svb);
586 }
587 }
588
589 /*
590 * [Livelist Check]
591 * Iterate through all the sublivelists and:
592 * - report leftover frees (**)
593 * - record leftover ALLOCs together with their TXG [see Cross Check]
594 *
595 * (**) Note: Double ALLOCs are valid in datasets that have dedup
596 * enabled. Similarly double FREEs are allowed as well but
597 * only if they pair up with a corresponding ALLOC entry once
598 * we our done with our sublivelist iteration.
599 *
600 * [Spacemap Check]
601 * for each metaslab:
602 * - iterate over spacemap and then the metaslab's entries in the
603 * spacemap log, then report any double FREEs and ALLOCs (do not
604 * blow up).
605 *
606 * [Cross Check]
607 * After finishing the Livelist Check phase and while being in the
608 * Spacemap Check phase, we find all the recorded leftover ALLOCs
609 * of the livelist check that are part of the metaslab that we are
610 * currently looking at in the Spacemap Check. We report any entries
611 * that are marked as ALLOCs in the livelists but have been actually
612 * freed (and potentially allocated again) after their TXG stamp in
613 * the spacemaps. Also report any ALLOCs from the livelists that
614 * belong to indirect vdevs (e.g. their vdev completed removal).
615 *
616 * Note that this will miss Log Spacemap entries that cancelled each other
617 * out before being flushed to the metaslab, so we are not guaranteed
618 * to match all erroneous ALLOCs.
619 */
620 static void
livelist_metaslab_validate(spa_t * spa)621 livelist_metaslab_validate(spa_t *spa)
622 {
623 (void) printf("Verifying deleted livelist entries\n");
624
625 sublivelist_verify_t sv;
626 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
627 sizeof (sublivelist_verify_block_t));
628 iterate_deleted_livelists(spa, livelist_verify, &sv);
629
630 (void) printf("Verifying metaslab entries\n");
631 vdev_t *rvd = spa->spa_root_vdev;
632 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
633 vdev_t *vd = rvd->vdev_child[c];
634
635 if (!vdev_is_concrete(vd))
636 continue;
637
638 for (uint64_t mid = 0; mid < vd->vdev_ms_count; mid++) {
639 metaslab_t *m = vd->vdev_ms[mid];
640
641 (void) fprintf(stderr,
642 "\rverifying concrete vdev %llu, "
643 "metaslab %llu of %llu ...",
644 (longlong_t)vd->vdev_id,
645 (longlong_t)mid,
646 (longlong_t)vd->vdev_ms_count);
647
648 uint64_t shift, start;
649 zfs_range_seg_type_t type =
650 metaslab_calculate_range_tree_type(vd, m,
651 &start, &shift);
652 metaslab_verify_t mv;
653 mv.mv_allocated = zfs_range_tree_create_flags(
654 NULL, type, NULL, start, shift,
655 0, "livelist_metaslab_validate:mv_allocated");
656 mv.mv_vdid = vd->vdev_id;
657 mv.mv_msid = m->ms_id;
658 mv.mv_start = m->ms_start;
659 mv.mv_end = m->ms_start + m->ms_size;
660 zfs_btree_create(&mv.mv_livelist_allocs,
661 livelist_block_compare, NULL,
662 sizeof (sublivelist_verify_block_t));
663
664 mv_populate_livelist_allocs(&mv, &sv);
665
666 spacemap_check_ms_sm(m->ms_sm, &mv);
667 spacemap_check_sm_log(spa, &mv);
668
669 zfs_range_tree_vacate(mv.mv_allocated, NULL, NULL);
670 zfs_range_tree_destroy(mv.mv_allocated);
671 zfs_btree_clear(&mv.mv_livelist_allocs);
672 zfs_btree_destroy(&mv.mv_livelist_allocs);
673 }
674 }
675 (void) fprintf(stderr, "\n");
676
677 /*
678 * If there are any segments in the leftover tree after we walked
679 * through all the metaslabs in the concrete vdevs then this means
680 * that we have segments in the livelists that belong to indirect
681 * vdevs and are marked as allocated.
682 */
683 if (zfs_btree_numnodes(&sv.sv_leftover) == 0) {
684 zfs_btree_destroy(&sv.sv_leftover);
685 return;
686 }
687 (void) printf("ERROR: Found livelist blocks marked as allocated "
688 "for indirect vdevs:\n");
689 corruption_found = B_TRUE;
690
691 zfs_btree_index_t *where = NULL;
692 sublivelist_verify_block_t *svb;
693 while ((svb = zfs_btree_destroy_nodes(&sv.sv_leftover, &where)) !=
694 NULL) {
695 int vdev_id = DVA_GET_VDEV(&svb->svb_dva);
696 ASSERT3U(vdev_id, <, rvd->vdev_children);
697 vdev_t *vd = rvd->vdev_child[vdev_id];
698 ASSERT(!vdev_is_concrete(vd));
699 (void) printf("<%d:%llx:%llx> TXG %llx\n",
700 vdev_id, (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
701 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva),
702 (u_longlong_t)svb->svb_allocated_txg);
703 }
704 (void) printf("\n");
705 zfs_btree_destroy(&sv.sv_leftover);
706 }
707
708 /*
709 * These libumem hooks provide a reasonable set of defaults for the allocator's
710 * debugging facilities.
711 */
712 const char *
_umem_debug_init(void)713 _umem_debug_init(void)
714 {
715 return ("default,verbose"); /* $UMEM_DEBUG setting */
716 }
717
718 const char *
_umem_logging_init(void)719 _umem_logging_init(void)
720 {
721 return ("fail,contents"); /* $UMEM_LOGGING setting */
722 }
723
724 static void
usage(void)725 usage(void)
726 {
727 (void) fprintf(stderr,
728 "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
729 "[-I <inflight I/Os>]\n"
730 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
731 "\t\t[-K <key>]\n"
732 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
733 "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
734 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
735 "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
736 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
737 "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
738 "\t%s [-v] <bookmark>\n"
739 "\t%s -C [-A] [-U <cache>] [<poolname>]\n"
740 "\t%s -l [-Aqu] <device>\n"
741 "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
742 "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
743 "\t%s -O [-K <key>] <dataset> <path>\n"
744 "\t%s -r [-K <key>] <dataset> <path> <destination>\n"
745 "\t%s -r [-K <key>] -O <dataset> <object-id> <destination>\n"
746 "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
747 "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
748 "\t%s -f [-H] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
749 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
750 "\t%s -E [-A] word0:word1:...:word15\n"
751 "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
752 "<poolname>\n\n",
753 cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
754 cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname);
755
756 (void) fprintf(stderr, " Dataset name must include at least one "
757 "separator character '/' or '@'\n");
758 (void) fprintf(stderr, " If dataset name is specified, only that "
759 "dataset is dumped\n");
760 (void) fprintf(stderr, " If object numbers or object number "
761 "ranges are specified, only those\n"
762 " objects or ranges are dumped.\n\n");
763 (void) fprintf(stderr,
764 " Object ranges take the form <start>:<end>[:<flags>]\n"
765 " start Starting object number\n"
766 " end Ending object number, or -1 for no upper bound\n"
767 " flags Optional flags to select object types:\n"
768 " A All objects (this is the default)\n"
769 " d ZFS directories\n"
770 " f ZFS files \n"
771 " m SPA space maps\n"
772 " z ZAPs\n"
773 " - Negate effect of next flag\n\n");
774 (void) fprintf(stderr, " Options to control amount of output:\n");
775 (void) fprintf(stderr, " -b --block-stats "
776 "block statistics\n");
777 (void) fprintf(stderr, " --bin=(lsize|psize|asize) "
778 "bin blocks based on this size in all three columns\n");
779 (void) fprintf(stderr,
780 " --class=(normal|special|dedup|other)[,...]\n"
781 " only consider blocks from "
782 "these allocation classes\n");
783 (void) fprintf(stderr, " -B --backup "
784 "backup stream\n");
785 (void) fprintf(stderr, " -c --checksum "
786 "checksum all metadata (twice for all data) blocks\n");
787 (void) fprintf(stderr, " -C --config "
788 "config (or cachefile if alone)\n");
789 (void) fprintf(stderr, " -d --datasets "
790 "dataset(s)\n");
791 (void) fprintf(stderr, " -D --dedup-stats "
792 "dedup statistics\n");
793 (void) fprintf(stderr, " -f --file-layout "
794 "display a file's layout across raidz disks\n");
795 (void) fprintf(stderr, " -E --embedded-block-pointer=INTEGER\n"
796 " decode and display block "
797 "from an embedded block pointer\n");
798 (void) fprintf(stderr, " -h --history "
799 "pool history\n");
800 (void) fprintf(stderr, " -i --intent-logs "
801 "intent logs\n");
802 (void) fprintf(stderr, " -l --label "
803 "read label contents\n");
804 (void) fprintf(stderr, " -k --checkpointed-state "
805 "examine the checkpointed state of the pool\n");
806 (void) fprintf(stderr, " -L --disable-leak-tracking "
807 "disable leak tracking (do not load spacemaps)\n");
808 (void) fprintf(stderr, " -m --metaslabs "
809 "metaslabs\n");
810 (void) fprintf(stderr, " -M --metaslab-groups "
811 "metaslab groups\n");
812 (void) fprintf(stderr, " -O --object-lookups "
813 "perform object lookups by path\n");
814 (void) fprintf(stderr, " -r --copy-object "
815 "copy an object by path to file\n");
816 (void) fprintf(stderr, " -R --read-block "
817 "read and display block from a device\n");
818 (void) fprintf(stderr, " -s --io-stats "
819 "report stats on zdb's I/O\n");
820 (void) fprintf(stderr, " -S --simulate-dedup "
821 "simulate dedup to measure effect\n");
822 (void) fprintf(stderr, " -v --verbose "
823 "verbose (applies to all others)\n");
824 (void) fprintf(stderr, " -y --livelist "
825 "perform livelist and metaslab validation on any livelists being "
826 "deleted\n\n");
827
828 (void) fprintf(stderr, " Below options are intended for use "
829 "with other options:\n");
830 (void) fprintf(stderr, " -A --ignore-assertions "
831 "ignore assertions (-A), enable panic recovery (-AA) or both "
832 "(-AAA)\n");
833 (void) fprintf(stderr, " -e --exported "
834 "pool is exported/destroyed/has altroot/not in a cachefile\n");
835 (void) fprintf(stderr, " -F --automatic-rewind "
836 "attempt automatic rewind within safe range of transaction "
837 "groups\n");
838 (void) fprintf(stderr, " -G --dump-debug-msg "
839 "dump zfs_dbgmsg buffer before exiting\n");
840 (void) fprintf(stderr, " -H --scripting-mode "
841 "dump output in a format that is scripting friendly\n");
842 (void) fprintf(stderr, " -I --inflight=INTEGER "
843 "specify the maximum number of checksumming I/Os "
844 "[default is 200]\n");
845 (void) fprintf(stderr, " -K --key=KEY "
846 "decryption key for encrypted dataset\n");
847 (void) fprintf(stderr, " -o --option=\"NAME=VALUE\" "
848 "set the named tunable to the given value\n");
849 (void) fprintf(stderr, " -p --path==PATH "
850 "use one or more with -e to specify path to vdev dir\n");
851 (void) fprintf(stderr, " -P --parseable "
852 "print numbers in parseable form\n");
853 (void) fprintf(stderr, " -q --skip-label "
854 "don't print label contents\n");
855 (void) fprintf(stderr, " -t --txg=INTEGER "
856 "highest txg to use when searching for uberblocks\n");
857 (void) fprintf(stderr, " -T --brt-stats "
858 "BRT statistics\n");
859 (void) fprintf(stderr, " -u --uberblock "
860 "uberblock\n");
861 (void) fprintf(stderr, " -U --cachefile=PATH "
862 "use alternate cachefile\n");
863 (void) fprintf(stderr, " -V --verbatim "
864 "do verbatim import\n");
865 (void) fprintf(stderr, " -x --dump-blocks=PATH "
866 "dump all read blocks into specified directory\n");
867 (void) fprintf(stderr, " -X --extreme-rewind "
868 "attempt extreme rewind (does not work with dataset)\n");
869 (void) fprintf(stderr, " -Y --all-reconstruction "
870 "attempt all reconstruction combinations for split blocks\n");
871 (void) fprintf(stderr, " -Z --zstd-headers "
872 "show ZSTD headers \n");
873 (void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
874 "to make only that option verbose\n");
875 (void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
876 zdb_exit(2);
877 }
878
879 static void
dump_debug_buffer(void)880 dump_debug_buffer(void)
881 {
882 ssize_t ret __attribute__((unused));
883
884 if (!dump_opt['G'])
885 return;
886 /*
887 * We use write() instead of printf() so that this function
888 * is safe to call from a signal handler.
889 */
890 ret = write(STDERR_FILENO, "\n", 1);
891 zfs_dbgmsg_print(STDERR_FILENO, "zdb");
892 }
893
sig_handler(int signo)894 static void sig_handler(int signo)
895 {
896 struct sigaction action;
897
898 libspl_backtrace(STDERR_FILENO);
899 dump_debug_buffer();
900
901 /*
902 * Restore default action and re-raise signal so SIGSEGV and
903 * SIGABRT can trigger a core dump.
904 */
905 action.sa_handler = SIG_DFL;
906 sigemptyset(&action.sa_mask);
907 action.sa_flags = 0;
908 (void) sigaction(signo, &action, NULL);
909 raise(signo);
910 }
911
912 /*
913 * Called for usage errors that are discovered after a call to spa_open(),
914 * dmu_bonus_hold(), or pool_match(). abort() is called for other errors.
915 */
916
917 static void
fatal(const char * fmt,...)918 fatal(const char *fmt, ...)
919 {
920 va_list ap;
921
922 va_start(ap, fmt);
923 (void) fprintf(stderr, "%s: ", cmdname);
924 (void) vfprintf(stderr, fmt, ap);
925 va_end(ap);
926 (void) fprintf(stderr, "\n");
927
928 dump_debug_buffer();
929
930 zdb_exit(1);
931 }
932
933 static void
dump_packed_nvlist(objset_t * os,uint64_t object,void * data,size_t size)934 dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size)
935 {
936 (void) size;
937 nvlist_t *nv;
938 size_t nvsize = *(uint64_t *)data;
939 char *packed = umem_alloc(nvsize, UMEM_NOFAIL);
940
941 VERIFY0(dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH));
942
943 VERIFY0(nvlist_unpack(packed, nvsize, &nv, 0));
944
945 umem_free(packed, nvsize);
946
947 dump_nvlist(nv, 8);
948
949 nvlist_free(nv);
950 }
951
952 static void
dump_history_offsets(objset_t * os,uint64_t object,void * data,size_t size)953 dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size)
954 {
955 (void) os, (void) object, (void) size;
956 spa_history_phys_t *shp = data;
957
958 if (shp == NULL)
959 return;
960
961 (void) printf("\t\tpool_create_len = %llu\n",
962 (u_longlong_t)shp->sh_pool_create_len);
963 (void) printf("\t\tphys_max_off = %llu\n",
964 (u_longlong_t)shp->sh_phys_max_off);
965 (void) printf("\t\tbof = %llu\n",
966 (u_longlong_t)shp->sh_bof);
967 (void) printf("\t\teof = %llu\n",
968 (u_longlong_t)shp->sh_eof);
969 (void) printf("\t\trecords_lost = %llu\n",
970 (u_longlong_t)shp->sh_records_lost);
971 }
972
973 static void
zdb_nicenum(uint64_t num,char * buf,size_t buflen)974 zdb_nicenum(uint64_t num, char *buf, size_t buflen)
975 {
976 if (dump_opt['P'])
977 (void) snprintf(buf, buflen, "%llu", (longlong_t)num);
978 else
979 nicenum(num, buf, buflen);
980 }
981
982 static void
zdb_nicebytes(uint64_t bytes,char * buf,size_t buflen)983 zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen)
984 {
985 if (dump_opt['P'])
986 (void) snprintf(buf, buflen, "%llu", (longlong_t)bytes);
987 else
988 zfs_nicebytes(bytes, buf, buflen);
989 }
990
991 static const char histo_stars[] = "****************************************";
992 static const uint64_t histo_width = sizeof (histo_stars) - 1;
993
994 static void
dump_histogram(const uint64_t * histo,int size,int offset)995 dump_histogram(const uint64_t *histo, int size, int offset)
996 {
997 int i;
998 int minidx = size - 1;
999 int maxidx = 0;
1000 uint64_t max = 0;
1001
1002 for (i = 0; i < size; i++) {
1003 if (histo[i] == 0)
1004 continue;
1005 if (histo[i] > max)
1006 max = histo[i];
1007 if (i > maxidx)
1008 maxidx = i;
1009 if (i < minidx)
1010 minidx = i;
1011 }
1012
1013 if (max < histo_width)
1014 max = histo_width;
1015
1016 for (i = minidx; i <= maxidx; i++) {
1017 (void) printf("\t\t\t%3u: %6llu %s\n",
1018 i + offset, (u_longlong_t)histo[i],
1019 &histo_stars[(max - histo[i]) * histo_width / max]);
1020 }
1021 }
1022
1023 static void
dump_zap_stats(objset_t * os,uint64_t object)1024 dump_zap_stats(objset_t *os, uint64_t object)
1025 {
1026 int error;
1027 zap_stats_t zs;
1028
1029 error = zap_get_stats(os, object, &zs);
1030 if (error)
1031 return;
1032
1033 if (zs.zs_ptrtbl_len == 0) {
1034 ASSERT(zs.zs_num_blocks == 1);
1035 (void) printf("\tmicrozap: %llu bytes, %llu entries\n",
1036 (u_longlong_t)zs.zs_blocksize,
1037 (u_longlong_t)zs.zs_num_entries);
1038 return;
1039 }
1040
1041 (void) printf("\tFat ZAP stats:\n");
1042
1043 (void) printf("\t\tPointer table:\n");
1044 (void) printf("\t\t\t%llu elements\n",
1045 (u_longlong_t)zs.zs_ptrtbl_len);
1046 (void) printf("\t\t\tzt_blk: %llu\n",
1047 (u_longlong_t)zs.zs_ptrtbl_zt_blk);
1048 (void) printf("\t\t\tzt_numblks: %llu\n",
1049 (u_longlong_t)zs.zs_ptrtbl_zt_numblks);
1050 (void) printf("\t\t\tzt_shift: %llu\n",
1051 (u_longlong_t)zs.zs_ptrtbl_zt_shift);
1052 (void) printf("\t\t\tzt_blks_copied: %llu\n",
1053 (u_longlong_t)zs.zs_ptrtbl_blks_copied);
1054 (void) printf("\t\t\tzt_nextblk: %llu\n",
1055 (u_longlong_t)zs.zs_ptrtbl_nextblk);
1056
1057 (void) printf("\t\tZAP entries: %llu\n",
1058 (u_longlong_t)zs.zs_num_entries);
1059 (void) printf("\t\tLeaf blocks: %llu\n",
1060 (u_longlong_t)zs.zs_num_leafs);
1061 (void) printf("\t\tTotal blocks: %llu\n",
1062 (u_longlong_t)zs.zs_num_blocks);
1063 (void) printf("\t\tzap_block_type: 0x%llx\n",
1064 (u_longlong_t)zs.zs_block_type);
1065 (void) printf("\t\tzap_magic: 0x%llx\n",
1066 (u_longlong_t)zs.zs_magic);
1067 (void) printf("\t\tzap_salt: 0x%llx\n",
1068 (u_longlong_t)zs.zs_salt);
1069
1070 (void) printf("\t\tLeafs with 2^n pointers:\n");
1071 dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);
1072
1073 (void) printf("\t\tBlocks with n*5 entries:\n");
1074 dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);
1075
1076 (void) printf("\t\tBlocks n/10 full:\n");
1077 dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);
1078
1079 (void) printf("\t\tEntries with n chunks:\n");
1080 dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);
1081
1082 (void) printf("\t\tBuckets with n entries:\n");
1083 dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
1084 }
1085
1086 static void
dump_none(objset_t * os,uint64_t object,void * data,size_t size)1087 dump_none(objset_t *os, uint64_t object, void *data, size_t size)
1088 {
1089 (void) os, (void) object, (void) data, (void) size;
1090 }
1091
1092 static void
dump_unknown(objset_t * os,uint64_t object,void * data,size_t size)1093 dump_unknown(objset_t *os, uint64_t object, void *data, size_t size)
1094 {
1095 (void) os, (void) object, (void) data, (void) size;
1096 (void) printf("\tUNKNOWN OBJECT TYPE\n");
1097 }
1098
1099 static void
dump_uint8(objset_t * os,uint64_t object,void * data,size_t size)1100 dump_uint8(objset_t *os, uint64_t object, void *data, size_t size)
1101 {
1102 (void) os, (void) object, (void) data, (void) size;
1103 }
1104
1105 static void
dump_uint64(objset_t * os,uint64_t object,void * data,size_t size)1106 dump_uint64(objset_t *os, uint64_t object, void *data, size_t size)
1107 {
1108 uint64_t *arr;
1109 uint64_t oursize;
1110 if (dump_opt['d'] < 6)
1111 return;
1112
1113 if (data == NULL) {
1114 dmu_object_info_t doi;
1115
1116 VERIFY0(dmu_object_info(os, object, &doi));
1117 size = doi.doi_max_offset;
1118 /*
1119 * We cap the size at 1 mebibyte here to prevent
1120 * allocation failures and nigh-infinite printing if the
1121 * object is extremely large.
1122 */
1123 oursize = MIN(size, 1 << 20);
1124 arr = kmem_alloc(oursize, KM_SLEEP);
1125
1126 int err = dmu_read(os, object, 0, oursize, arr, 0);
1127 if (err != 0) {
1128 (void) printf("got error %u from dmu_read\n", err);
1129 kmem_free(arr, oursize);
1130 return;
1131 }
1132 } else {
1133 /*
1134 * Even though the allocation is already done in this code path,
1135 * we still cap the size to prevent excessive printing.
1136 */
1137 oursize = MIN(size, 1 << 20);
1138 arr = data;
1139 }
1140
1141 if (size == 0) {
1142 if (data == NULL)
1143 kmem_free(arr, oursize);
1144 (void) printf("\t\t[]\n");
1145 return;
1146 }
1147
1148 (void) printf("\t\t[%0llx", (u_longlong_t)arr[0]);
1149 for (size_t i = 1; i * sizeof (uint64_t) < oursize; i++) {
1150 if (i % 4 != 0)
1151 (void) printf(", %0llx", (u_longlong_t)arr[i]);
1152 else
1153 (void) printf(",\n\t\t%0llx", (u_longlong_t)arr[i]);
1154 }
1155 if (oursize != size)
1156 (void) printf(", ... ");
1157 (void) printf("]\n");
1158
1159 if (data == NULL)
1160 kmem_free(arr, oursize);
1161 }
1162
1163 static void
dump_zap(objset_t * os,uint64_t object,void * data,size_t size)1164 dump_zap(objset_t *os, uint64_t object, void *data, size_t size)
1165 {
1166 (void) data, (void) size;
1167 zap_cursor_t zc;
1168 zap_attribute_t *attrp = zap_attribute_long_alloc();
1169 void *prop;
1170 unsigned i;
1171
1172 dump_zap_stats(os, object);
1173 (void) printf("\n");
1174
1175 for (zap_cursor_init(&zc, os, object);
1176 zap_cursor_retrieve(&zc, attrp) == 0;
1177 zap_cursor_advance(&zc)) {
1178 boolean_t key64 =
1179 !!(zap_getflags(zc.zc_zap) & ZAP_FLAG_UINT64_KEY);
1180
1181 if (key64)
1182 (void) printf("\t\t0x%010" PRIu64 "x = ",
1183 *(uint64_t *)attrp->za_name);
1184 else
1185 (void) printf("\t\t%s = ", attrp->za_name);
1186
1187 if (attrp->za_num_integers == 0) {
1188 (void) printf("\n");
1189 continue;
1190 }
1191 prop = umem_zalloc(attrp->za_num_integers *
1192 attrp->za_integer_length, UMEM_NOFAIL);
1193
1194 if (key64)
1195 (void) zap_lookup_uint64(os, object,
1196 (const uint64_t *)attrp->za_name, 1,
1197 attrp->za_integer_length, attrp->za_num_integers,
1198 prop);
1199 else
1200 (void) zap_lookup(os, object, attrp->za_name,
1201 attrp->za_integer_length, attrp->za_num_integers,
1202 prop);
1203
1204 if (attrp->za_integer_length == 1 && !key64) {
1205 if (strcmp(attrp->za_name,
1206 DSL_CRYPTO_KEY_MASTER_KEY) == 0 ||
1207 strcmp(attrp->za_name,
1208 DSL_CRYPTO_KEY_HMAC_KEY) == 0 ||
1209 strcmp(attrp->za_name, DSL_CRYPTO_KEY_IV) == 0 ||
1210 strcmp(attrp->za_name, DSL_CRYPTO_KEY_MAC) == 0 ||
1211 strcmp(attrp->za_name,
1212 DMU_POOL_CHECKSUM_SALT) == 0) {
1213 uint8_t *u8 = prop;
1214
1215 for (i = 0; i < attrp->za_num_integers; i++) {
1216 (void) printf("%02x", u8[i]);
1217 }
1218 } else {
1219 (void) printf("%s", (char *)prop);
1220 }
1221 } else {
1222 for (i = 0; i < attrp->za_num_integers; i++) {
1223 switch (attrp->za_integer_length) {
1224 case 1:
1225 (void) printf("%u ",
1226 ((uint8_t *)prop)[i]);
1227 break;
1228 case 2:
1229 (void) printf("%u ",
1230 ((uint16_t *)prop)[i]);
1231 break;
1232 case 4:
1233 (void) printf("%u ",
1234 ((uint32_t *)prop)[i]);
1235 break;
1236 case 8:
1237 (void) printf("%lld ",
1238 (u_longlong_t)((int64_t *)prop)[i]);
1239 break;
1240 }
1241 }
1242 }
1243 (void) printf("\n");
1244 umem_free(prop,
1245 attrp->za_num_integers * attrp->za_integer_length);
1246 }
1247 zap_cursor_fini(&zc);
1248 zap_attribute_free(attrp);
1249 }
1250
1251 static void
dump_bpobj(objset_t * os,uint64_t object,void * data,size_t size)1252 dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size)
1253 {
1254 bpobj_phys_t *bpop = data;
1255 uint64_t i;
1256 char bytes[32], comp[32], uncomp[32];
1257
1258 /* make sure the output won't get truncated */
1259 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
1260 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
1261 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
1262
1263 if (bpop == NULL)
1264 return;
1265
1266 zdb_nicenum(bpop->bpo_bytes, bytes, sizeof (bytes));
1267 zdb_nicenum(bpop->bpo_comp, comp, sizeof (comp));
1268 zdb_nicenum(bpop->bpo_uncomp, uncomp, sizeof (uncomp));
1269
1270 (void) printf("\t\tnum_blkptrs = %llu\n",
1271 (u_longlong_t)bpop->bpo_num_blkptrs);
1272 (void) printf("\t\tbytes = %s\n", bytes);
1273 if (size >= BPOBJ_SIZE_V1) {
1274 (void) printf("\t\tcomp = %s\n", comp);
1275 (void) printf("\t\tuncomp = %s\n", uncomp);
1276 }
1277 if (size >= BPOBJ_SIZE_V2) {
1278 (void) printf("\t\tsubobjs = %llu\n",
1279 (u_longlong_t)bpop->bpo_subobjs);
1280 (void) printf("\t\tnum_subobjs = %llu\n",
1281 (u_longlong_t)bpop->bpo_num_subobjs);
1282 }
1283 if (size >= sizeof (*bpop)) {
1284 (void) printf("\t\tnum_freed = %llu\n",
1285 (u_longlong_t)bpop->bpo_num_freed);
1286 }
1287
1288 if (dump_opt['d'] < 5)
1289 return;
1290
1291 for (i = 0; i < bpop->bpo_num_blkptrs; i++) {
1292 char blkbuf[BP_SPRINTF_LEN];
1293 blkptr_t bp;
1294
1295 int err = dmu_read(os, object,
1296 i * sizeof (bp), sizeof (bp), &bp, 0);
1297 if (err != 0) {
1298 (void) printf("got error %u from dmu_read\n", err);
1299 break;
1300 }
1301 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp,
1302 BP_GET_FREE(&bp));
1303 (void) printf("\t%s\n", blkbuf);
1304 }
1305 }
1306
1307 static void
dump_bpobj_subobjs(objset_t * os,uint64_t object,void * data,size_t size)1308 dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size)
1309 {
1310 (void) data, (void) size;
1311 dmu_object_info_t doi;
1312 int64_t i;
1313
1314 VERIFY0(dmu_object_info(os, object, &doi));
1315 uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP);
1316
1317 int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0);
1318 if (err != 0) {
1319 (void) printf("got error %u from dmu_read\n", err);
1320 kmem_free(subobjs, doi.doi_max_offset);
1321 return;
1322 }
1323
1324 int64_t last_nonzero = -1;
1325 for (i = 0; i < doi.doi_max_offset / 8; i++) {
1326 if (subobjs[i] != 0)
1327 last_nonzero = i;
1328 }
1329
1330 for (i = 0; i <= last_nonzero; i++) {
1331 (void) printf("\t%llu\n", (u_longlong_t)subobjs[i]);
1332 }
1333 kmem_free(subobjs, doi.doi_max_offset);
1334 }
1335
1336 static void
dump_ddt_zap(objset_t * os,uint64_t object,void * data,size_t size)1337 dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size)
1338 {
1339 (void) data, (void) size;
1340 dump_zap_stats(os, object);
1341 /* contents are printed elsewhere, properly decoded */
1342 }
1343
1344 static void
dump_sa_attrs(objset_t * os,uint64_t object,void * data,size_t size)1345 dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size)
1346 {
1347 (void) data, (void) size;
1348 zap_cursor_t zc;
1349 zap_attribute_t *attrp = zap_attribute_alloc();
1350
1351 dump_zap_stats(os, object);
1352 (void) printf("\n");
1353
1354 for (zap_cursor_init(&zc, os, object);
1355 zap_cursor_retrieve(&zc, attrp) == 0;
1356 zap_cursor_advance(&zc)) {
1357 (void) printf("\t\t%s = ", attrp->za_name);
1358 if (attrp->za_num_integers == 0) {
1359 (void) printf("\n");
1360 continue;
1361 }
1362 (void) printf(" %llx : [%d:%d:%d]\n",
1363 (u_longlong_t)attrp->za_first_integer,
1364 (int)ATTR_LENGTH(attrp->za_first_integer),
1365 (int)ATTR_BSWAP(attrp->za_first_integer),
1366 (int)ATTR_NUM(attrp->za_first_integer));
1367 }
1368 zap_cursor_fini(&zc);
1369 zap_attribute_free(attrp);
1370 }
1371
1372 static void
dump_sa_layouts(objset_t * os,uint64_t object,void * data,size_t size)1373 dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size)
1374 {
1375 (void) data, (void) size;
1376 zap_cursor_t zc;
1377 zap_attribute_t *attrp = zap_attribute_alloc();
1378 uint16_t *layout_attrs;
1379 unsigned i;
1380
1381 dump_zap_stats(os, object);
1382 (void) printf("\n");
1383
1384 for (zap_cursor_init(&zc, os, object);
1385 zap_cursor_retrieve(&zc, attrp) == 0;
1386 zap_cursor_advance(&zc)) {
1387 (void) printf("\t\t%s = [", attrp->za_name);
1388 if (attrp->za_num_integers == 0) {
1389 (void) printf("\n");
1390 continue;
1391 }
1392
1393 VERIFY(attrp->za_integer_length == 2);
1394 layout_attrs = umem_zalloc(attrp->za_num_integers *
1395 attrp->za_integer_length, UMEM_NOFAIL);
1396
1397 VERIFY(zap_lookup(os, object, attrp->za_name,
1398 attrp->za_integer_length,
1399 attrp->za_num_integers, layout_attrs) == 0);
1400
1401 for (i = 0; i != attrp->za_num_integers; i++)
1402 (void) printf(" %d ", (int)layout_attrs[i]);
1403 (void) printf("]\n");
1404 umem_free(layout_attrs,
1405 attrp->za_num_integers * attrp->za_integer_length);
1406 }
1407 zap_cursor_fini(&zc);
1408 zap_attribute_free(attrp);
1409 }
1410
1411 static void
dump_zpldir(objset_t * os,uint64_t object,void * data,size_t size)1412 dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
1413 {
1414 (void) data, (void) size;
1415 zap_cursor_t zc;
1416 zap_attribute_t *attrp = zap_attribute_long_alloc();
1417 const char *typenames[] = {
1418 /* 0 */ "not specified",
1419 /* 1 */ "FIFO",
1420 /* 2 */ "Character Device",
1421 /* 3 */ "3 (invalid)",
1422 /* 4 */ "Directory",
1423 /* 5 */ "5 (invalid)",
1424 /* 6 */ "Block Device",
1425 /* 7 */ "7 (invalid)",
1426 /* 8 */ "Regular File",
1427 /* 9 */ "9 (invalid)",
1428 /* 10 */ "Symbolic Link",
1429 /* 11 */ "11 (invalid)",
1430 /* 12 */ "Socket",
1431 /* 13 */ "Door",
1432 /* 14 */ "Event Port",
1433 /* 15 */ "15 (invalid)",
1434 };
1435
1436 dump_zap_stats(os, object);
1437 (void) printf("\n");
1438
1439 for (zap_cursor_init(&zc, os, object);
1440 zap_cursor_retrieve(&zc, attrp) == 0;
1441 zap_cursor_advance(&zc)) {
1442 (void) printf("\t\t%s = %lld (type: %s)\n",
1443 attrp->za_name, ZFS_DIRENT_OBJ(attrp->za_first_integer),
1444 typenames[ZFS_DIRENT_TYPE(attrp->za_first_integer)]);
1445 }
1446 zap_cursor_fini(&zc);
1447 zap_attribute_free(attrp);
1448 }
1449
1450 static int
get_dtl_refcount(vdev_t * vd)1451 get_dtl_refcount(vdev_t *vd)
1452 {
1453 int refcount = 0;
1454
1455 if (vd->vdev_ops->vdev_op_leaf) {
1456 space_map_t *sm = vd->vdev_dtl_sm;
1457
1458 if (sm != NULL &&
1459 sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1460 return (1);
1461 return (0);
1462 }
1463
1464 for (unsigned c = 0; c < vd->vdev_children; c++)
1465 refcount += get_dtl_refcount(vd->vdev_child[c]);
1466 return (refcount);
1467 }
1468
1469 static int
get_metaslab_refcount(vdev_t * vd)1470 get_metaslab_refcount(vdev_t *vd)
1471 {
1472 int refcount = 0;
1473
1474 if (vd->vdev_top == vd) {
1475 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
1476 space_map_t *sm = vd->vdev_ms[m]->ms_sm;
1477
1478 if (sm != NULL &&
1479 sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1480 refcount++;
1481 }
1482 }
1483 for (unsigned c = 0; c < vd->vdev_children; c++)
1484 refcount += get_metaslab_refcount(vd->vdev_child[c]);
1485
1486 return (refcount);
1487 }
1488
1489 static int
get_obsolete_refcount(vdev_t * vd)1490 get_obsolete_refcount(vdev_t *vd)
1491 {
1492 uint64_t obsolete_sm_object;
1493 int refcount = 0;
1494
1495 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1496 if (vd->vdev_top == vd && obsolete_sm_object != 0) {
1497 dmu_object_info_t doi;
1498 VERIFY0(dmu_object_info(vd->vdev_spa->spa_meta_objset,
1499 obsolete_sm_object, &doi));
1500 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1501 refcount++;
1502 }
1503 } else {
1504 ASSERT0P(vd->vdev_obsolete_sm);
1505 ASSERT0(obsolete_sm_object);
1506 }
1507 for (unsigned c = 0; c < vd->vdev_children; c++) {
1508 refcount += get_obsolete_refcount(vd->vdev_child[c]);
1509 }
1510
1511 return (refcount);
1512 }
1513
1514 static int
get_prev_obsolete_spacemap_refcount(spa_t * spa)1515 get_prev_obsolete_spacemap_refcount(spa_t *spa)
1516 {
1517 uint64_t prev_obj =
1518 spa->spa_condensing_indirect_phys.scip_prev_obsolete_sm_object;
1519 if (prev_obj != 0) {
1520 dmu_object_info_t doi;
1521 VERIFY0(dmu_object_info(spa->spa_meta_objset, prev_obj, &doi));
1522 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1523 return (1);
1524 }
1525 }
1526 return (0);
1527 }
1528
1529 static int
get_checkpoint_refcount(vdev_t * vd)1530 get_checkpoint_refcount(vdev_t *vd)
1531 {
1532 int refcount = 0;
1533
1534 if (vd->vdev_top == vd && vd->vdev_top_zap != 0 &&
1535 zap_contains(spa_meta_objset(vd->vdev_spa),
1536 vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) == 0)
1537 refcount++;
1538
1539 for (uint64_t c = 0; c < vd->vdev_children; c++)
1540 refcount += get_checkpoint_refcount(vd->vdev_child[c]);
1541
1542 return (refcount);
1543 }
1544
1545 static int
get_log_spacemap_refcount(spa_t * spa)1546 get_log_spacemap_refcount(spa_t *spa)
1547 {
1548 return (avl_numnodes(&spa->spa_sm_logs_by_txg));
1549 }
1550
1551 static int
verify_spacemap_refcounts(spa_t * spa)1552 verify_spacemap_refcounts(spa_t *spa)
1553 {
1554 uint64_t expected_refcount = 0;
1555 uint64_t actual_refcount;
1556
1557 (void) feature_get_refcount(spa,
1558 &spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM],
1559 &expected_refcount);
1560 actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
1561 actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);
1562 actual_refcount += get_obsolete_refcount(spa->spa_root_vdev);
1563 actual_refcount += get_prev_obsolete_spacemap_refcount(spa);
1564 actual_refcount += get_checkpoint_refcount(spa->spa_root_vdev);
1565 actual_refcount += get_log_spacemap_refcount(spa);
1566
1567 if (expected_refcount != actual_refcount) {
1568 (void) printf("space map refcount mismatch: expected %lld != "
1569 "actual %lld\n",
1570 (longlong_t)expected_refcount,
1571 (longlong_t)actual_refcount);
1572 return (2);
1573 }
1574 return (0);
1575 }
1576
1577 static void
dump_spacemap(objset_t * os,space_map_t * sm)1578 dump_spacemap(objset_t *os, space_map_t *sm)
1579 {
1580 const char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1581 "INVALID", "INVALID", "INVALID", "INVALID" };
1582
1583 if (sm == NULL)
1584 return;
1585
1586 (void) printf("space map object %llu:\n",
1587 (longlong_t)sm->sm_object);
1588 (void) printf(" smp_length = 0x%llx\n",
1589 (longlong_t)sm->sm_phys->smp_length);
1590 (void) printf(" smp_alloc = 0x%llx\n",
1591 (longlong_t)sm->sm_phys->smp_alloc);
1592
1593 if (dump_opt['d'] < 6 && dump_opt['m'] < 4)
1594 return;
1595
1596 /*
1597 * Print out the freelist entries in both encoded and decoded form.
1598 */
1599 uint8_t mapshift = sm->sm_shift;
1600 int64_t alloc = 0;
1601 uint64_t word, entry_id = 0;
1602 for (uint64_t offset = 0; offset < space_map_length(sm);
1603 offset += sizeof (word)) {
1604
1605 VERIFY0(dmu_read(os, space_map_object(sm), offset,
1606 sizeof (word), &word, DMU_READ_PREFETCH));
1607
1608 if (sm_entry_is_debug(word)) {
1609 uint64_t de_txg = SM_DEBUG_TXG_DECODE(word);
1610 uint64_t de_sync_pass = SM_DEBUG_SYNCPASS_DECODE(word);
1611 if (de_txg == 0) {
1612 (void) printf(
1613 "\t [%6llu] PADDING\n",
1614 (u_longlong_t)entry_id);
1615 } else {
1616 (void) printf(
1617 "\t [%6llu] %s: txg %llu pass %llu\n",
1618 (u_longlong_t)entry_id,
1619 ddata[SM_DEBUG_ACTION_DECODE(word)],
1620 (u_longlong_t)de_txg,
1621 (u_longlong_t)de_sync_pass);
1622 }
1623 entry_id++;
1624 continue;
1625 }
1626
1627 char entry_type;
1628 uint64_t entry_off, entry_run, entry_vdev;
1629
1630 if (sm_entry_is_single_word(word)) {
1631 entry_type = (SM_TYPE_DECODE(word) == SM_ALLOC) ?
1632 'A' : 'F';
1633 entry_off = (SM_OFFSET_DECODE(word) << mapshift) +
1634 sm->sm_start;
1635 entry_run = SM_RUN_DECODE(word) << mapshift;
1636
1637 (void) printf("\t [%6llu] %c "
1638 "range: %012llx-%012llx size: %08llx\n",
1639 (u_longlong_t)entry_id, entry_type,
1640 (u_longlong_t)entry_off,
1641 (u_longlong_t)(entry_off + entry_run - 1),
1642 (u_longlong_t)entry_run);
1643 } else {
1644 /* it is a two-word entry so we read another word */
1645 ASSERT(sm_entry_is_double_word(word));
1646
1647 uint64_t extra_word;
1648 offset += sizeof (extra_word);
1649 ASSERT3U(offset, <, space_map_length(sm));
1650 VERIFY0(dmu_read(os, space_map_object(sm), offset,
1651 sizeof (extra_word), &extra_word,
1652 DMU_READ_PREFETCH));
1653
1654 entry_run = SM2_RUN_DECODE(word) << mapshift;
1655 entry_vdev = SM2_VDEV_DECODE(word);
1656 entry_type = (SM2_TYPE_DECODE(extra_word) == SM_ALLOC) ?
1657 'A' : 'F';
1658 entry_off = (SM2_OFFSET_DECODE(extra_word) <<
1659 mapshift) + sm->sm_start;
1660
1661 if (zopt_metaslab_args == 0 ||
1662 zopt_metaslab[0] == entry_vdev) {
1663 (void) printf("\t [%6llu] %c "
1664 "range: %012llx-%012llx size: %08llx "
1665 "vdev: %llu\n",
1666 (u_longlong_t)entry_id, entry_type,
1667 (u_longlong_t)entry_off,
1668 (u_longlong_t)(entry_off + entry_run - 1),
1669 (u_longlong_t)entry_run,
1670 (u_longlong_t)entry_vdev);
1671 }
1672 }
1673
1674 if (entry_type == 'A')
1675 alloc += entry_run;
1676 else
1677 alloc -= entry_run;
1678 entry_id++;
1679 }
1680 if (alloc != space_map_allocated(sm)) {
1681 (void) printf("space_map_object alloc (%lld) INCONSISTENT "
1682 "with space map summary (%lld)\n",
1683 (longlong_t)space_map_allocated(sm), (longlong_t)alloc);
1684 }
1685 }
1686
1687 static void
dump_metaslab_stats(metaslab_t * msp)1688 dump_metaslab_stats(metaslab_t *msp)
1689 {
1690 char maxbuf[32];
1691 zfs_range_tree_t *rt = msp->ms_allocatable;
1692 zfs_btree_t *t = &msp->ms_allocatable_by_size;
1693 int free_pct = zfs_range_tree_space(rt) * 100 / msp->ms_size;
1694
1695 /* max sure nicenum has enough space */
1696 _Static_assert(sizeof (maxbuf) >= NN_NUMBUF_SZ, "maxbuf truncated");
1697
1698 zdb_nicenum(metaslab_largest_allocatable(msp), maxbuf, sizeof (maxbuf));
1699
1700 (void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
1701 "segments", zfs_btree_numnodes(t), "maxsize", maxbuf,
1702 "freepct", free_pct);
1703 (void) printf("\tIn-memory histogram:\n");
1704 dump_histogram(rt->rt_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1705 }
1706
1707 static void
dump_allocated(void * arg,uint64_t start,uint64_t size)1708 dump_allocated(void *arg, uint64_t start, uint64_t size)
1709 {
1710 uint64_t *off = arg;
1711 if (*off != start)
1712 (void) printf("ALLOC: %"PRIu64" %"PRIu64"\n", *off,
1713 start - *off);
1714 *off = start + size;
1715 }
1716
1717 static void
dump_metaslab(metaslab_t * msp)1718 dump_metaslab(metaslab_t *msp)
1719 {
1720 vdev_t *vd = msp->ms_group->mg_vd;
1721 spa_t *spa = vd->vdev_spa;
1722 space_map_t *sm = msp->ms_sm;
1723 char freebuf[32];
1724
1725 zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf,
1726 sizeof (freebuf));
1727
1728 (void) printf(
1729 "\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
1730 (u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
1731 (u_longlong_t)space_map_object(sm), freebuf);
1732
1733 if (dump_opt[ARG_ALLOCATED] ||
1734 (dump_opt['m'] > 2 && !dump_opt['L'])) {
1735 mutex_enter(&msp->ms_lock);
1736 VERIFY0(metaslab_load(msp));
1737 }
1738
1739 if (dump_opt['m'] > 2 && !dump_opt['L']) {
1740 zfs_range_tree_stat_verify(msp->ms_allocatable);
1741 dump_metaslab_stats(msp);
1742 }
1743
1744 if (dump_opt[ARG_ALLOCATED]) {
1745 uint64_t off = msp->ms_start;
1746 zfs_range_tree_walk(msp->ms_allocatable, dump_allocated,
1747 &off);
1748 if (off != msp->ms_start + msp->ms_size)
1749 (void) printf("ALLOC: %"PRIu64" %"PRIu64"\n", off,
1750 msp->ms_size - off);
1751 }
1752
1753 if (dump_opt['m'] > 1 && sm != NULL &&
1754 spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) {
1755 /*
1756 * The space map histogram represents free space in chunks
1757 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1758 */
1759 (void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1760 (u_longlong_t)msp->ms_fragmentation);
1761 dump_histogram(sm->sm_phys->smp_histogram,
1762 SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift);
1763 }
1764
1765 if (dump_opt[ARG_ALLOCATED] ||
1766 (dump_opt['m'] > 2 && !dump_opt['L'])) {
1767 metaslab_unload(msp);
1768 mutex_exit(&msp->ms_lock);
1769 }
1770
1771 if (vd->vdev_ops == &vdev_draid_ops)
1772 ASSERT3U(msp->ms_size, <=, 1ULL << vd->vdev_ms_shift);
1773 else
1774 ASSERT3U(msp->ms_size, ==, 1ULL << vd->vdev_ms_shift);
1775
1776 dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
1777
1778 if (spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
1779 (void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1780 (u_longlong_t)metaslab_unflushed_txg(msp));
1781 }
1782 }
1783
1784 static void
print_vdev_metaslab_header(vdev_t * vd)1785 print_vdev_metaslab_header(vdev_t *vd)
1786 {
1787 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
1788 const char *bias_str = "";
1789 if (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) {
1790 bias_str = VDEV_ALLOC_BIAS_LOG;
1791 } else if (alloc_bias == VDEV_BIAS_SPECIAL) {
1792 bias_str = VDEV_ALLOC_BIAS_SPECIAL;
1793 } else if (alloc_bias == VDEV_BIAS_DEDUP) {
1794 bias_str = VDEV_ALLOC_BIAS_DEDUP;
1795 }
1796
1797 uint64_t ms_flush_data_obj = 0;
1798 if (vd->vdev_top_zap != 0) {
1799 int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
1800 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
1801 sizeof (uint64_t), 1, &ms_flush_data_obj);
1802 if (error != ENOENT) {
1803 ASSERT0(error);
1804 }
1805 }
1806
1807 (void) printf("\tvdev %10llu\t%s metaslab shift %4llu",
1808 (u_longlong_t)vd->vdev_id, bias_str,
1809 (u_longlong_t)vd->vdev_ms_shift);
1810
1811 if (ms_flush_data_obj != 0) {
1812 (void) printf(" ms_unflushed_phys object %llu",
1813 (u_longlong_t)ms_flush_data_obj);
1814 }
1815
1816 (void) printf("\n\t%-10s%5llu %-19s %-15s %-12s\n",
1817 "metaslabs", (u_longlong_t)vd->vdev_ms_count,
1818 "offset", "spacemap", "free");
1819 (void) printf("\t%15s %19s %15s %12s\n",
1820 "---------------", "-------------------",
1821 "---------------", "------------");
1822 }
1823
1824 static void
dump_metaslab_groups(spa_t * spa,boolean_t show_special)1825 dump_metaslab_groups(spa_t *spa, boolean_t show_special)
1826 {
1827 vdev_t *rvd = spa->spa_root_vdev;
1828 metaslab_class_t *mc = spa_normal_class(spa);
1829 metaslab_class_t *smc = spa_special_class(spa);
1830 uint64_t fragmentation;
1831
1832 metaslab_class_histogram_verify(mc);
1833
1834 for (unsigned c = 0; c < rvd->vdev_children; c++) {
1835 vdev_t *tvd = rvd->vdev_child[c];
1836 metaslab_group_t *mg = tvd->vdev_mg;
1837
1838 if (mg == NULL || (mg->mg_class != mc &&
1839 (!show_special || mg->mg_class != smc)))
1840 continue;
1841
1842 metaslab_group_histogram_verify(mg);
1843 mg->mg_fragmentation = metaslab_group_fragmentation(mg);
1844
1845 (void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1846 "fragmentation",
1847 (u_longlong_t)tvd->vdev_id,
1848 (u_longlong_t)tvd->vdev_ms_count);
1849 if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
1850 (void) printf("%3s\n", "-");
1851 } else {
1852 (void) printf("%3llu%%\n",
1853 (u_longlong_t)mg->mg_fragmentation);
1854 }
1855 dump_histogram(mg->mg_histogram,
1856 ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1857 }
1858
1859 (void) printf("\tpool %s\tfragmentation", spa_name(spa));
1860 fragmentation = metaslab_class_fragmentation(mc);
1861 if (fragmentation == ZFS_FRAG_INVALID)
1862 (void) printf("\t%3s\n", "-");
1863 else
1864 (void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation);
1865 dump_histogram(mc->mc_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1866 }
1867
1868 static void
print_vdev_indirect(vdev_t * vd)1869 print_vdev_indirect(vdev_t *vd)
1870 {
1871 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
1872 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
1873 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
1874
1875 if (vim == NULL) {
1876 ASSERT0P(vib);
1877 return;
1878 }
1879
1880 ASSERT3U(vdev_indirect_mapping_object(vim), ==,
1881 vic->vic_mapping_object);
1882 ASSERT3U(vdev_indirect_births_object(vib), ==,
1883 vic->vic_births_object);
1884
1885 (void) printf("indirect births obj %llu:\n",
1886 (longlong_t)vic->vic_births_object);
1887 (void) printf(" vib_count = %llu\n",
1888 (longlong_t)vdev_indirect_births_count(vib));
1889 for (uint64_t i = 0; i < vdev_indirect_births_count(vib); i++) {
1890 vdev_indirect_birth_entry_phys_t *cur_vibe =
1891 &vib->vib_entries[i];
1892 (void) printf("\toffset %llx -> txg %llu\n",
1893 (longlong_t)cur_vibe->vibe_offset,
1894 (longlong_t)cur_vibe->vibe_phys_birth_txg);
1895 }
1896 (void) printf("\n");
1897
1898 (void) printf("indirect mapping obj %llu:\n",
1899 (longlong_t)vic->vic_mapping_object);
1900 (void) printf(" vim_max_offset = 0x%llx\n",
1901 (longlong_t)vdev_indirect_mapping_max_offset(vim));
1902 (void) printf(" vim_bytes_mapped = 0x%llx\n",
1903 (longlong_t)vdev_indirect_mapping_bytes_mapped(vim));
1904 (void) printf(" vim_count = %llu\n",
1905 (longlong_t)vdev_indirect_mapping_num_entries(vim));
1906
1907 if (dump_opt['d'] <= 5 && dump_opt['m'] <= 3)
1908 return;
1909
1910 uint32_t *counts = vdev_indirect_mapping_load_obsolete_counts(vim);
1911
1912 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
1913 vdev_indirect_mapping_entry_phys_t *vimep =
1914 &vim->vim_entries[i];
1915 (void) printf("\t<%llx:%llx:%llx> -> "
1916 "<%llx:%llx:%llx> (%x obsolete)\n",
1917 (longlong_t)vd->vdev_id,
1918 (longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
1919 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1920 (longlong_t)DVA_GET_VDEV(&vimep->vimep_dst),
1921 (longlong_t)DVA_GET_OFFSET(&vimep->vimep_dst),
1922 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1923 counts[i]);
1924 }
1925 (void) printf("\n");
1926
1927 uint64_t obsolete_sm_object;
1928 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1929 if (obsolete_sm_object != 0) {
1930 objset_t *mos = vd->vdev_spa->spa_meta_objset;
1931 (void) printf("obsolete space map object %llu:\n",
1932 (u_longlong_t)obsolete_sm_object);
1933 ASSERT(vd->vdev_obsolete_sm != NULL);
1934 ASSERT3U(space_map_object(vd->vdev_obsolete_sm), ==,
1935 obsolete_sm_object);
1936 dump_spacemap(mos, vd->vdev_obsolete_sm);
1937 (void) printf("\n");
1938 }
1939 }
1940
1941 static void
dump_metaslabs(spa_t * spa)1942 dump_metaslabs(spa_t *spa)
1943 {
1944 vdev_t *vd, *rvd = spa->spa_root_vdev;
1945 uint64_t m, c = 0, children = rvd->vdev_children;
1946
1947 (void) printf("\nMetaslabs:\n");
1948
1949 if (zopt_metaslab_args > 0) {
1950 c = zopt_metaslab[0];
1951
1952 if (c >= children)
1953 (void) fatal("bad vdev id: %llu", (u_longlong_t)c);
1954
1955 if (zopt_metaslab_args > 1) {
1956 vd = rvd->vdev_child[c];
1957 print_vdev_metaslab_header(vd);
1958
1959 for (m = 1; m < zopt_metaslab_args; m++) {
1960 if (zopt_metaslab[m] < vd->vdev_ms_count)
1961 dump_metaslab(
1962 vd->vdev_ms[zopt_metaslab[m]]);
1963 else
1964 (void) fprintf(stderr, "bad metaslab "
1965 "number %llu\n",
1966 (u_longlong_t)zopt_metaslab[m]);
1967 }
1968 (void) printf("\n");
1969 return;
1970 }
1971 children = c + 1;
1972 }
1973 for (; c < children; c++) {
1974 vd = rvd->vdev_child[c];
1975 print_vdev_metaslab_header(vd);
1976
1977 print_vdev_indirect(vd);
1978
1979 for (m = 0; m < vd->vdev_ms_count; m++)
1980 dump_metaslab(vd->vdev_ms[m]);
1981 (void) printf("\n");
1982 }
1983 }
1984
1985 static void
dump_log_spacemaps(spa_t * spa)1986 dump_log_spacemaps(spa_t *spa)
1987 {
1988 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1989 return;
1990
1991 (void) printf("\nLog Space Maps in Pool:\n");
1992 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
1993 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1994 space_map_t *sm = NULL;
1995 VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
1996 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
1997
1998 (void) printf("Log Spacemap object %llu txg %llu\n",
1999 (u_longlong_t)sls->sls_sm_obj, (u_longlong_t)sls->sls_txg);
2000 dump_spacemap(spa->spa_meta_objset, sm);
2001 space_map_close(sm);
2002 }
2003 (void) printf("\n");
2004 }
2005
2006 static void
dump_ddt_entry(const ddt_t * ddt,const ddt_lightweight_entry_t * ddlwe,uint64_t index)2007 dump_ddt_entry(const ddt_t *ddt, const ddt_lightweight_entry_t *ddlwe,
2008 uint64_t index)
2009 {
2010 const ddt_key_t *ddk = &ddlwe->ddlwe_key;
2011 char blkbuf[BP_SPRINTF_LEN];
2012 blkptr_t blk;
2013 int p;
2014
2015 for (p = 0; p < DDT_NPHYS(ddt); p++) {
2016 const ddt_univ_phys_t *ddp = &ddlwe->ddlwe_phys;
2017 ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p);
2018
2019 if (ddt_phys_birth(ddp, v) == 0)
2020 continue;
2021 ddt_bp_create(ddt->ddt_checksum, ddk, ddp, v, &blk);
2022 snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
2023 (void) printf("index %llx refcnt %llu phys %d %s\n",
2024 (u_longlong_t)index, (u_longlong_t)ddt_phys_refcnt(ddp, v),
2025 p, blkbuf);
2026 }
2027 }
2028
2029 static void
dump_dedup_ratio(const ddt_stat_t * dds)2030 dump_dedup_ratio(const ddt_stat_t *dds)
2031 {
2032 double rL, rP, rD, D, dedup, compress, copies;
2033
2034 if (dds->dds_blocks == 0)
2035 return;
2036
2037 rL = (double)dds->dds_ref_lsize;
2038 rP = (double)dds->dds_ref_psize;
2039 rD = (double)dds->dds_ref_dsize;
2040 D = (double)dds->dds_dsize;
2041
2042 dedup = rD / D;
2043 compress = rL / rP;
2044 copies = rD / rP;
2045
2046 (void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
2047 "dedup * compress / copies = %.2f\n\n",
2048 dedup, compress, copies, dedup * compress / copies);
2049 }
2050
2051 static void
dump_ddt_log(ddt_t * ddt)2052 dump_ddt_log(ddt_t *ddt)
2053 {
2054 if (ddt->ddt_version != DDT_VERSION_FDT ||
2055 !(ddt->ddt_flags & DDT_FLAG_LOG))
2056 return;
2057
2058 for (int n = 0; n < 2; n++) {
2059 ddt_log_t *ddl = &ddt->ddt_log[n];
2060
2061 char flagstr[64] = {0};
2062 if (ddl->ddl_flags > 0) {
2063 flagstr[0] = ' ';
2064 int c = 1;
2065 if (ddl->ddl_flags & DDL_FLAG_FLUSHING)
2066 c += strlcpy(&flagstr[c], " FLUSHING",
2067 sizeof (flagstr) - c);
2068 if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT)
2069 c += strlcpy(&flagstr[c], " CHECKPOINT",
2070 sizeof (flagstr) - c);
2071 if (ddl->ddl_flags &
2072 ~(DDL_FLAG_FLUSHING|DDL_FLAG_CHECKPOINT))
2073 c += strlcpy(&flagstr[c], " UNKNOWN",
2074 sizeof (flagstr) - c);
2075 flagstr[1] = '[';
2076 flagstr[c] = ']';
2077 }
2078
2079 uint64_t count = avl_numnodes(&ddl->ddl_tree);
2080
2081 printf(DMU_POOL_DDT_LOG ": flags=0x%02x%s; obj=%llu; "
2082 "len=%llu; txg=%llu; entries=%llu\n",
2083 zio_checksum_table[ddt->ddt_checksum].ci_name, n,
2084 ddl->ddl_flags, flagstr,
2085 (u_longlong_t)ddl->ddl_object,
2086 (u_longlong_t)ddl->ddl_length,
2087 (u_longlong_t)ddl->ddl_first_txg, (u_longlong_t)count);
2088
2089 if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT) {
2090 const ddt_key_t *ddk = &ddl->ddl_checkpoint;
2091 printf(" checkpoint: "
2092 "%016llx:%016llx:%016llx:%016llx:%016llx\n",
2093 (u_longlong_t)ddk->ddk_cksum.zc_word[0],
2094 (u_longlong_t)ddk->ddk_cksum.zc_word[1],
2095 (u_longlong_t)ddk->ddk_cksum.zc_word[2],
2096 (u_longlong_t)ddk->ddk_cksum.zc_word[3],
2097 (u_longlong_t)ddk->ddk_prop);
2098 }
2099
2100 if (count == 0 || dump_opt['D'] < 4)
2101 continue;
2102
2103 ddt_lightweight_entry_t ddlwe;
2104 uint64_t index = 0;
2105 for (ddt_log_entry_t *ddle = avl_first(&ddl->ddl_tree);
2106 ddle; ddle = AVL_NEXT(&ddl->ddl_tree, ddle)) {
2107 DDT_LOG_ENTRY_TO_LIGHTWEIGHT(ddt, ddle, &ddlwe);
2108 dump_ddt_entry(ddt, &ddlwe, index++);
2109 }
2110 }
2111 }
2112
2113 static void
dump_ddt_object(ddt_t * ddt,ddt_type_t type,ddt_class_t class)2114 dump_ddt_object(ddt_t *ddt, ddt_type_t type, ddt_class_t class)
2115 {
2116 char name[DDT_NAMELEN];
2117 ddt_lightweight_entry_t ddlwe;
2118 uint64_t walk = 0;
2119 dmu_object_info_t doi;
2120 uint64_t count, dspace, mspace;
2121 int error;
2122
2123 error = ddt_object_info(ddt, type, class, &doi);
2124
2125 if (error == ENOENT)
2126 return;
2127 ASSERT0(error);
2128
2129 error = ddt_object_count(ddt, type, class, &count);
2130 ASSERT0(error);
2131 if (count == 0)
2132 return;
2133
2134 dspace = doi.doi_physical_blocks_512 << 9;
2135 mspace = doi.doi_fill_count * doi.doi_data_block_size;
2136
2137 ddt_object_name(ddt, type, class, name);
2138
2139 (void) printf("%s: dspace=%llu; mspace=%llu; entries=%llu\n", name,
2140 (u_longlong_t)dspace, (u_longlong_t)mspace, (u_longlong_t)count);
2141
2142 if (dump_opt['D'] < 3)
2143 return;
2144
2145 (void) printf("%s: object=%llu\n", name,
2146 (u_longlong_t)ddt->ddt_object[type][class]);
2147 zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class],
2148 dump_opt['P'] > 0);
2149
2150 if (dump_opt['D'] < 4)
2151 return;
2152
2153 if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE)
2154 return;
2155
2156 (void) printf("%s contents:\n\n", name);
2157
2158 while ((error = ddt_object_walk(ddt, type, class, &walk, &ddlwe)) == 0)
2159 dump_ddt_entry(ddt, &ddlwe, walk);
2160
2161 ASSERT3U(error, ==, ENOENT);
2162
2163 (void) printf("\n");
2164 }
2165
2166 static void
dump_ddt(ddt_t * ddt)2167 dump_ddt(ddt_t *ddt)
2168 {
2169 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
2170 return;
2171
2172 char flagstr[64] = {0};
2173 if (ddt->ddt_flags > 0) {
2174 flagstr[0] = ' ';
2175 int c = 1;
2176 if (ddt->ddt_flags & DDT_FLAG_FLAT)
2177 c += strlcpy(&flagstr[c], " FLAT",
2178 sizeof (flagstr) - c);
2179 if (ddt->ddt_flags & DDT_FLAG_LOG)
2180 c += strlcpy(&flagstr[c], " LOG",
2181 sizeof (flagstr) - c);
2182 if (ddt->ddt_flags & ~DDT_FLAG_MASK)
2183 c += strlcpy(&flagstr[c], " UNKNOWN",
2184 sizeof (flagstr) - c);
2185 flagstr[1] = '[';
2186 flagstr[c] = ']';
2187 }
2188
2189 printf("DDT-%s: version=%llu [%s]; flags=0x%02llx%s; rootobj=%llu\n",
2190 zio_checksum_table[ddt->ddt_checksum].ci_name,
2191 (u_longlong_t)ddt->ddt_version,
2192 (ddt->ddt_version == 0) ? "LEGACY" :
2193 (ddt->ddt_version == 1) ? "FDT" : "UNKNOWN",
2194 (u_longlong_t)ddt->ddt_flags, flagstr,
2195 (u_longlong_t)ddt->ddt_dir_object);
2196
2197 for (ddt_type_t type = 0; type < DDT_TYPES; type++)
2198 for (ddt_class_t class = 0; class < DDT_CLASSES; class++)
2199 dump_ddt_object(ddt, type, class);
2200
2201 dump_ddt_log(ddt);
2202 }
2203
2204 static void
dump_all_ddts(spa_t * spa)2205 dump_all_ddts(spa_t *spa)
2206 {
2207 ddt_histogram_t ddh_total = {{{0}}};
2208 ddt_stat_t dds_total = {0};
2209
2210 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++)
2211 dump_ddt(spa->spa_ddt[c]);
2212
2213 ddt_get_dedup_stats(spa, &dds_total);
2214
2215 if (dds_total.dds_blocks == 0) {
2216 (void) printf("All DDTs are empty\n");
2217 return;
2218 }
2219
2220 (void) printf("\n");
2221
2222 if (dump_opt['D'] > 1) {
2223 (void) printf("DDT histogram (aggregated over all DDTs):\n");
2224 ddt_get_dedup_histogram(spa, &ddh_total);
2225 zpool_dump_ddt(&dds_total, &ddh_total, dump_opt['P'] > 0);
2226 }
2227
2228 dump_dedup_ratio(&dds_total);
2229
2230 /*
2231 * Dump a histogram of unique class entry age
2232 */
2233 if (dump_opt['D'] == 3 && getenv("ZDB_DDT_UNIQUE_AGE_HIST") != NULL) {
2234 ddt_age_histo_t histogram;
2235
2236 (void) printf("DDT walk unique, building age histogram...\n");
2237 ddt_prune_walk(spa, 0, &histogram);
2238
2239 /*
2240 * print out histogram for unique entry class birth
2241 */
2242 if (histogram.dah_entries > 0) {
2243 (void) printf("%5s %9s %4s\n",
2244 "age", "blocks", "amnt");
2245 (void) printf("%5s %9s %4s\n",
2246 "-----", "---------", "----");
2247 for (int i = 0; i < HIST_BINS; i++) {
2248 (void) printf("%5d %9d %4d%%\n", 1 << i,
2249 (int)histogram.dah_age_histo[i],
2250 (int)((histogram.dah_age_histo[i] * 100) /
2251 histogram.dah_entries));
2252 }
2253 }
2254 }
2255 }
2256
2257 static void
dump_brt(spa_t * spa)2258 dump_brt(spa_t *spa)
2259 {
2260 if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) {
2261 printf("BRT: unsupported on this pool\n");
2262 return;
2263 }
2264
2265 if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
2266 printf("BRT: empty\n");
2267 return;
2268 }
2269
2270 char count[32], used[32], saved[32];
2271 zdb_nicebytes(brt_get_used(spa), used, sizeof (used));
2272 zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved));
2273 uint64_t ratio = brt_get_ratio(spa);
2274 printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved,
2275 (u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100));
2276
2277 if (dump_opt['T'] < 2)
2278 return;
2279
2280 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2281 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2282 if (!brtvd->bv_initiated) {
2283 printf("BRT: vdev %" PRIu64 ": empty\n", vdevid);
2284 continue;
2285 }
2286
2287 zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count));
2288 zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used));
2289 zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved));
2290 printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n",
2291 vdevid, count, used, saved);
2292 }
2293
2294 if (dump_opt['T'] < 3)
2295 return;
2296
2297 /* -TTT shows a per-vdev histograms; -TTTT shows all entries */
2298 boolean_t do_histo = dump_opt['T'] == 3;
2299
2300 char dva[64];
2301
2302 if (!do_histo)
2303 printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2304
2305 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2306 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2307 if (!brtvd->bv_initiated)
2308 continue;
2309
2310 uint64_t counts[64] = {};
2311
2312 zap_cursor_t zc;
2313 zap_attribute_t *za = zap_attribute_alloc();
2314 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2315 brtvd->bv_mos_entries);
2316 zap_cursor_retrieve(&zc, za) == 0;
2317 zap_cursor_advance(&zc)) {
2318 uint64_t refcnt;
2319 VERIFY0(zap_lookup_uint64(spa->spa_meta_objset,
2320 brtvd->bv_mos_entries,
2321 (const uint64_t *)za->za_name, 1,
2322 za->za_integer_length, za->za_num_integers,
2323 &refcnt));
2324
2325 if (do_histo)
2326 counts[highbit64(refcnt)]++;
2327 else {
2328 uint64_t offset =
2329 *(const uint64_t *)za->za_name;
2330
2331 snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx",
2332 vdevid, (u_longlong_t)offset);
2333 printf("%-16s %-10llu\n", dva,
2334 (u_longlong_t)refcnt);
2335 }
2336 }
2337 zap_cursor_fini(&zc);
2338 zap_attribute_free(za);
2339
2340 if (do_histo) {
2341 printf("\nBRT: vdev %" PRIu64
2342 ": DVAs with 2^n refcnts:\n", vdevid);
2343 dump_histogram(counts, 64, 0);
2344 }
2345 }
2346 }
2347
2348 static void
dump_dtl_seg(void * arg,uint64_t start,uint64_t size)2349 dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
2350 {
2351 char *prefix = arg;
2352
2353 (void) printf("%s [%llu,%llu) length %llu\n",
2354 prefix,
2355 (u_longlong_t)start,
2356 (u_longlong_t)(start + size),
2357 (u_longlong_t)(size));
2358 }
2359
2360 static void
dump_dtl(vdev_t * vd,int indent)2361 dump_dtl(vdev_t *vd, int indent)
2362 {
2363 spa_t *spa = vd->vdev_spa;
2364 boolean_t required;
2365 const char *name[DTL_TYPES] = { "missing", "partial", "scrub",
2366 "outage" };
2367 char prefix[256];
2368
2369 spa_vdev_state_enter(spa, SCL_NONE);
2370 required = vdev_dtl_required(vd);
2371 (void) spa_vdev_state_exit(spa, NULL, 0);
2372
2373 if (indent == 0)
2374 (void) printf("\nDirty time logs:\n\n");
2375
2376 (void) printf("\t%*s%s [%s]\n", indent, "",
2377 vd->vdev_path ? vd->vdev_path :
2378 vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa),
2379 required ? "DTL-required" : "DTL-expendable");
2380
2381 for (int t = 0; t < DTL_TYPES; t++) {
2382 zfs_range_tree_t *rt = vd->vdev_dtl[t];
2383 if (zfs_range_tree_space(rt) == 0)
2384 continue;
2385 (void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
2386 indent + 2, "", name[t]);
2387 zfs_range_tree_walk(rt, dump_dtl_seg, prefix);
2388 if (dump_opt['d'] > 5 && vd->vdev_children == 0)
2389 dump_spacemap(spa->spa_meta_objset,
2390 vd->vdev_dtl_sm);
2391 }
2392
2393 for (unsigned c = 0; c < vd->vdev_children; c++)
2394 dump_dtl(vd->vdev_child[c], indent + 4);
2395 }
2396
2397 static void
dump_history(spa_t * spa)2398 dump_history(spa_t *spa)
2399 {
2400 nvlist_t **events = NULL;
2401 char *buf;
2402 uint64_t resid, len, off = 0;
2403 uint_t num = 0;
2404 int error;
2405 char tbuf[30];
2406
2407 if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) {
2408 (void) fprintf(stderr, "%s: unable to allocate I/O buffer\n",
2409 __func__);
2410 return;
2411 }
2412
2413 do {
2414 len = SPA_OLD_MAXBLOCKSIZE;
2415
2416 if ((error = spa_history_get(spa, &off, &len, buf)) != 0) {
2417 (void) fprintf(stderr, "Unable to read history: "
2418 "error %d\n", error);
2419 free(buf);
2420 return;
2421 }
2422
2423 if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0)
2424 break;
2425
2426 off -= resid;
2427 } while (len != 0);
2428
2429 (void) printf("\nHistory:\n");
2430 for (unsigned i = 0; i < num; i++) {
2431 boolean_t printed = B_FALSE;
2432
2433 if (nvlist_exists(events[i], ZPOOL_HIST_TIME)) {
2434 time_t tsec;
2435 struct tm t;
2436
2437 tsec = fnvlist_lookup_uint64(events[i],
2438 ZPOOL_HIST_TIME);
2439 (void) localtime_r(&tsec, &t);
2440 (void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t);
2441 } else {
2442 tbuf[0] = '\0';
2443 }
2444
2445 if (nvlist_exists(events[i], ZPOOL_HIST_CMD)) {
2446 (void) printf("%s %s\n", tbuf,
2447 fnvlist_lookup_string(events[i], ZPOOL_HIST_CMD));
2448 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_EVENT)) {
2449 uint64_t ievent;
2450
2451 ievent = fnvlist_lookup_uint64(events[i],
2452 ZPOOL_HIST_INT_EVENT);
2453 if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS)
2454 goto next;
2455
2456 (void) printf(" %s [internal %s txg:%ju] %s\n",
2457 tbuf,
2458 zfs_history_event_names[ievent],
2459 fnvlist_lookup_uint64(events[i],
2460 ZPOOL_HIST_TXG),
2461 fnvlist_lookup_string(events[i],
2462 ZPOOL_HIST_INT_STR));
2463 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_NAME)) {
2464 (void) printf("%s [txg:%ju] %s", tbuf,
2465 fnvlist_lookup_uint64(events[i],
2466 ZPOOL_HIST_TXG),
2467 fnvlist_lookup_string(events[i],
2468 ZPOOL_HIST_INT_NAME));
2469
2470 if (nvlist_exists(events[i], ZPOOL_HIST_DSNAME)) {
2471 (void) printf(" %s (%llu)",
2472 fnvlist_lookup_string(events[i],
2473 ZPOOL_HIST_DSNAME),
2474 (u_longlong_t)fnvlist_lookup_uint64(
2475 events[i],
2476 ZPOOL_HIST_DSID));
2477 }
2478
2479 (void) printf(" %s\n", fnvlist_lookup_string(events[i],
2480 ZPOOL_HIST_INT_STR));
2481 } else if (nvlist_exists(events[i], ZPOOL_HIST_IOCTL)) {
2482 (void) printf("%s ioctl %s\n", tbuf,
2483 fnvlist_lookup_string(events[i],
2484 ZPOOL_HIST_IOCTL));
2485
2486 if (nvlist_exists(events[i], ZPOOL_HIST_INPUT_NVL)) {
2487 (void) printf(" input:\n");
2488 dump_nvlist(fnvlist_lookup_nvlist(events[i],
2489 ZPOOL_HIST_INPUT_NVL), 8);
2490 }
2491 if (nvlist_exists(events[i], ZPOOL_HIST_OUTPUT_NVL)) {
2492 (void) printf(" output:\n");
2493 dump_nvlist(fnvlist_lookup_nvlist(events[i],
2494 ZPOOL_HIST_OUTPUT_NVL), 8);
2495 }
2496 if (nvlist_exists(events[i], ZPOOL_HIST_ERRNO)) {
2497 (void) printf(" errno: %lld\n",
2498 (longlong_t)fnvlist_lookup_int64(events[i],
2499 ZPOOL_HIST_ERRNO));
2500 }
2501 } else {
2502 goto next;
2503 }
2504
2505 printed = B_TRUE;
2506 next:
2507 if (dump_opt['h'] > 1) {
2508 if (!printed)
2509 (void) printf("unrecognized record:\n");
2510 dump_nvlist(events[i], 2);
2511 }
2512 }
2513 free(buf);
2514 }
2515
2516 static void
dump_dnode(objset_t * os,uint64_t object,void * data,size_t size)2517 dump_dnode(objset_t *os, uint64_t object, void *data, size_t size)
2518 {
2519 (void) os, (void) object, (void) data, (void) size;
2520 }
2521
2522 static uint64_t
blkid2offset(const dnode_phys_t * dnp,const blkptr_t * bp,const zbookmark_phys_t * zb)2523 blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
2524 const zbookmark_phys_t *zb)
2525 {
2526 if (dnp == NULL) {
2527 ASSERT(zb->zb_level < 0);
2528 if (zb->zb_object == 0)
2529 return (zb->zb_blkid);
2530 return (zb->zb_blkid * BP_GET_LSIZE(bp));
2531 }
2532
2533 ASSERT(zb->zb_level >= 0);
2534
2535 return ((zb->zb_blkid <<
2536 (zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) *
2537 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
2538 }
2539
2540 static void
snprintf_zstd_header(spa_t * spa,char * blkbuf,size_t buflen,const blkptr_t * bp)2541 snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
2542 const blkptr_t *bp)
2543 {
2544 static abd_t *pabd = NULL;
2545 void *buf;
2546 zio_t *zio;
2547 zfs_zstdhdr_t zstd_hdr;
2548 int error;
2549
2550 if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD)
2551 return;
2552
2553 if (BP_IS_HOLE(bp))
2554 return;
2555
2556 if (BP_IS_EMBEDDED(bp)) {
2557 buf = malloc(SPA_MAXBLOCKSIZE);
2558 if (buf == NULL) {
2559 (void) fprintf(stderr, "out of memory\n");
2560 zdb_exit(1);
2561 }
2562 decode_embedded_bp_compressed(bp, buf);
2563 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2564 free(buf);
2565 zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2566 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2567 (void) snprintf(blkbuf + strlen(blkbuf),
2568 buflen - strlen(blkbuf),
2569 " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2570 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2571 zfs_get_hdrlevel(&zstd_hdr));
2572 return;
2573 }
2574
2575 if (!pabd)
2576 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
2577 zio = zio_root(spa, NULL, NULL, 0);
2578
2579 /* Decrypt but don't decompress so we can read the compression header */
2580 zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL,
2581 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS,
2582 NULL));
2583 error = zio_wait(zio);
2584 if (error) {
2585 (void) fprintf(stderr, "read failed: %d\n", error);
2586 return;
2587 }
2588 buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp));
2589 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2590 zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2591 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2592
2593 (void) snprintf(blkbuf + strlen(blkbuf),
2594 buflen - strlen(blkbuf),
2595 " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2596 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2597 zfs_get_hdrlevel(&zstd_hdr));
2598
2599 abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp));
2600 }
2601
2602 static void
snprintf_blkptr_compact(char * blkbuf,size_t buflen,const blkptr_t * bp,boolean_t bp_freed)2603 snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
2604 boolean_t bp_freed)
2605 {
2606 const dva_t *dva = bp->blk_dva;
2607 int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1;
2608 int i;
2609
2610 if (dump_opt['b'] >= 6) {
2611 snprintf_blkptr(blkbuf, buflen, bp);
2612 if (bp_freed) {
2613 (void) snprintf(blkbuf + strlen(blkbuf),
2614 buflen - strlen(blkbuf), " %s", "FREE");
2615 }
2616 return;
2617 }
2618
2619 if (BP_IS_EMBEDDED(bp)) {
2620 (void) sprintf(blkbuf,
2621 "EMBEDDED et=%u %llxL/%llxP B=%llu",
2622 (int)BPE_GET_ETYPE(bp),
2623 (u_longlong_t)BPE_GET_LSIZE(bp),
2624 (u_longlong_t)BPE_GET_PSIZE(bp),
2625 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2626 return;
2627 }
2628
2629 blkbuf[0] = '\0';
2630
2631 for (i = 0; i < ndvas; i++) {
2632 (void) snprintf(blkbuf + strlen(blkbuf),
2633 buflen - strlen(blkbuf), "%llu:%llx:%llx%s ",
2634 (u_longlong_t)DVA_GET_VDEV(&dva[i]),
2635 (u_longlong_t)DVA_GET_OFFSET(&dva[i]),
2636 (u_longlong_t)DVA_GET_ASIZE(&dva[i]),
2637 (DVA_GET_GANG(&dva[i]) ? "G" : ""));
2638 }
2639
2640 if (BP_IS_HOLE(bp)) {
2641 (void) snprintf(blkbuf + strlen(blkbuf),
2642 buflen - strlen(blkbuf),
2643 "%llxL B=%llu",
2644 (u_longlong_t)BP_GET_LSIZE(bp),
2645 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2646 } else {
2647 (void) snprintf(blkbuf + strlen(blkbuf),
2648 buflen - strlen(blkbuf),
2649 "%llxL/%llxP F=%llu B=%llu/%llu",
2650 (u_longlong_t)BP_GET_LSIZE(bp),
2651 (u_longlong_t)BP_GET_PSIZE(bp),
2652 (u_longlong_t)BP_GET_FILL(bp),
2653 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),
2654 (u_longlong_t)BP_GET_PHYSICAL_BIRTH(bp));
2655 if (bp_freed)
2656 (void) snprintf(blkbuf + strlen(blkbuf),
2657 buflen - strlen(blkbuf), " %s", "FREE");
2658
2659 (void) snprintf(blkbuf + strlen(blkbuf),
2660 buflen - strlen(blkbuf),
2661 " cksum=%016llx:%016llx:%016llx:%016llx",
2662 (u_longlong_t)bp->blk_cksum.zc_word[0],
2663 (u_longlong_t)bp->blk_cksum.zc_word[1],
2664 (u_longlong_t)bp->blk_cksum.zc_word[2],
2665 (u_longlong_t)bp->blk_cksum.zc_word[3]);
2666 }
2667 }
2668
2669 static void
inverse_text(boolean_t on)2670 inverse_text(boolean_t on)
2671 {
2672 if (isatty(STDOUT_FILENO)) {
2673 if (on)
2674 fputs("\x1b[7m", stdout);
2675 else
2676 fputs("\x1b[m", stdout);
2677 }
2678 }
2679
2680 static void
print_file_layout_line(int line,int first_disk,int last_disk,int ashift,raidz_row_t * rr)2681 print_file_layout_line(int line, int first_disk, int last_disk, int ashift,
2682 raidz_row_t *rr)
2683 {
2684 if (first_disk != 0) {
2685 /* Account for empty columns */
2686 for (int c = 0; c < first_disk; c++) {
2687 (void) printf("%s%s", (c == 0) ? " │ " : " ",
2688 line == 0 ? "` ` ` ` ` " : " ` ` ` ` `");
2689 }
2690 }
2691
2692 /*
2693 * Check if we need to account for out of order disks in raidz1.
2694 * Here flip means the disks in first two column were exchanged.
2695 */
2696 boolean_t flip = rr->rr_firstdatacol == 1 &&
2697 (rr->rr_offset & (1ULL << 20));
2698
2699 for (int c = 0; c < rr->rr_cols; c++) {
2700 raidz_col_t *rc = &rr->rr_col[c];
2701 char colname[8];
2702 boolean_t pcol = flip ? (c == 1) : (c < rr->rr_firstdatacol);
2703
2704 if (rc->rc_devidx > last_disk)
2705 continue;
2706
2707 (void) snprintf(colname, sizeof (colname), "%c%d",
2708 pcol ? 'P' : 'D', (flip && c < 2) ? 0 :
2709 pcol ? c : c - rr->rr_firstdatacol);
2710
2711 (void) printf("%s", (rc->rc_devidx == 0) ? " │ " : " ");
2712 if (pcol)
2713 inverse_text(B_TRUE);
2714 if (line == 0) {
2715 (void) printf("%3s %5d", colname,
2716 (int)(rc->rc_size >> ashift));
2717 } else {
2718 (void) printf("%10llx",
2719 (u_longlong_t)((rc->rc_offset +
2720 VDEV_LABEL_START_SIZE) >> ashift));
2721 }
2722 if (pcol)
2723 inverse_text(B_FALSE);
2724 if (rc->rc_devidx == last_disk) {
2725 /* empty columns */
2726 for (int c = last_disk + 1; c < rr->rr_cols; c++) {
2727 (void) printf(" %s",
2728 line == 0 ? "` ` ` ` ` " : " ` ` ` ` `");
2729 }
2730 break;
2731 }
2732 }
2733 (void) printf(" │\n");
2734 }
2735
2736
2737 static u_longlong_t
print_indirect(spa_t * spa,blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp)2738 print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
2739 const dnode_phys_t *dnp)
2740 {
2741 char blkbuf[BP_SPRINTF_LEN];
2742 u_longlong_t offset;
2743 int l;
2744
2745 offset = (u_longlong_t)blkid2offset(dnp, bp, zb);
2746
2747 (void) printf("%16llx ", offset);
2748
2749 ASSERT(zb->zb_level >= 0);
2750
2751 for (l = dnp->dn_nlevels - 1; l >= -1; l--) {
2752 if (l == zb->zb_level) {
2753 (void) printf("L%llx", (u_longlong_t)zb->zb_level);
2754 } else {
2755 (void) printf(" ");
2756 }
2757 }
2758
2759 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE);
2760 if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD)
2761 snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp);
2762 (void) printf("%s", blkbuf);
2763
2764 if (!BP_IS_EMBEDDED(bp)) {
2765 if (BP_GET_TYPE(bp) != dnp->dn_type) {
2766 (void) printf(" (ERROR: Block pointer type "
2767 "(%llu) does not match dnode type (%hhu))",
2768 BP_GET_TYPE(bp), dnp->dn_type);
2769 corruption_found = B_TRUE;
2770 }
2771 if (BP_GET_LEVEL(bp) != zb->zb_level) {
2772 (void) printf(" (ERROR: Block pointer level "
2773 "(%llu) does not match bookmark level (%lld))",
2774 BP_GET_LEVEL(bp), (longlong_t)zb->zb_level);
2775 corruption_found = B_TRUE;
2776 }
2777 }
2778 (void) printf("\n");
2779
2780 return (offset);
2781 }
2782
2783 static void
print_file_layout_raidz(vdev_t * vd,blkptr_t * bp,uint64_t file_offset,boolean_t last)2784 print_file_layout_raidz(vdev_t *vd, blkptr_t *bp, uint64_t file_offset,
2785 boolean_t last)
2786 {
2787 /*
2788 * RAIDZ file data layout
2789 *
2790 * Use vdev_raidz_map_alloc() to get the actual layout
2791 */
2792 const dva_t *dva = bp->blk_dva;
2793 zio_t zio = {0};
2794 zio.io_size = P2ROUNDUP_TYPED(BP_GET_PSIZE(bp),
2795 1ULL << vd->vdev_ashift, uint64_t);
2796 zio.io_offset = DVA_GET_OFFSET(&dva[0]);
2797 zio.io_type = ZIO_TYPE_READ;
2798 zio.io_abd = abd_alloc_for_io(zio.io_size, B_FALSE);
2799
2800 vdev_raidz_t *vdrz = vd->vdev_tsd;
2801 raidz_map_t *rm = vdev_raidz_map_alloc(&zio, vd->vdev_ashift,
2802 vd->vdev_children, vdrz->vd_nparity);
2803 raidz_row_t *rr = rm->rm_row[0];
2804
2805 if (!dump_opt['H']) {
2806 int last_disk = vd->vdev_children - 1;
2807 /*
2808 * Account for out of order disks in raidz1.
2809 * For now just reverse them back and adjust for it later.
2810 */
2811 if (rr->rr_firstdatacol == 1 &&
2812 (zio.io_offset & (1ULL << 20))) {
2813 uint64_t devidx = rr->rr_col[0].rc_devidx;
2814 rr->rr_col[0].rc_devidx = rr->rr_col[1].rc_devidx;
2815 rr->rr_col[1].rc_devidx = devidx;
2816 }
2817 int first_disk = rr->rr_col[0].rc_devidx;
2818
2819 (void) printf("%12llx", (u_longlong_t)file_offset);
2820 print_file_layout_line(0, first_disk, last_disk,
2821 vd->vdev_ashift, rr);
2822 (void) printf("%*c", 12, ' ');
2823 print_file_layout_line(1, first_disk, last_disk,
2824 vd->vdev_ashift, rr);
2825 /* Check for split row */
2826 if (first_disk != 0) {
2827 (void) printf("%*c", 12, ' ');
2828 print_file_layout_line(0, 0, first_disk - 1,
2829 vd->vdev_ashift, rr);
2830 (void) printf("%*c", 12, ' ');
2831 print_file_layout_line(1, 0, first_disk - 1,
2832 vd->vdev_ashift, rr);
2833 }
2834 /* seperate rows with a line */
2835 (void) printf("%*c", 12, ' ');
2836 for (int c = 0; c < vd->vdev_children; c++) {
2837 (void) printf("%s%10s", (c == 0) ?
2838 (last ? " └─" : " ├─") : "──",
2839 "──────────");
2840 }
2841 (void) printf("─%s\n", last ? "┘" : "┤");
2842 } else {
2843 static uint64_t next_offset = 0;
2844
2845 if (next_offset != file_offset) {
2846 (void) printf("skip hole\t-\t\t%lld\n",
2847 (u_longlong_t)((file_offset - next_offset) / 512));
2848 }
2849 next_offset = file_offset + BP_GET_LSIZE(bp);
2850 uint64_t tmp_offset = file_offset;
2851
2852
2853 for (int c = 0; c < rr->rr_cols; c++) {
2854 boolean_t pcol = c < rr->rr_firstdatacol;
2855 raidz_col_t *rc = &rr->rr_col[c];
2856 char *path = vd->vdev_child[rc->rc_devidx]->vdev_path;
2857
2858 if (rc->rc_size == 0)
2859 continue;
2860 (void) printf("%s\t\t%llu\t%d",
2861 zfs_basename(path),
2862 (u_longlong_t)(rc->rc_offset +
2863 VDEV_LABEL_START_SIZE)/512,
2864 (int)rc->rc_size/512);
2865 if (dump_opt['v']) {
2866 char label = pcol ? 'P' : 'D';
2867 int num;
2868
2869 if (c < 2) {
2870 num = 0;
2871 } else {
2872 num = pcol ? c :
2873 (c - rr->rr_firstdatacol);
2874 }
2875 printf("\t%c%d", label, num);
2876 if (dump_opt['v'] > 1) {
2877 unsigned long long off;
2878 if (pcol)
2879 off = file_offset;
2880 else
2881 off = tmp_offset;
2882 off = off / 512ULL;
2883 printf("\t%llu", off);
2884 }
2885 }
2886 if (!pcol)
2887 tmp_offset += rc->rc_size;
2888 printf("\n");
2889 }
2890 }
2891 }
2892
2893 static void
print_file_layout(spa_t * spa,blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp)2894 print_file_layout(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
2895 const dnode_phys_t *dnp)
2896 {
2897 if (!BP_IS_EMBEDDED(bp)) {
2898 ASSERT3U(BP_GET_TYPE(bp), ==, dnp->dn_type);
2899 ASSERT3U(BP_GET_LEVEL(bp), ==, zb->zb_level);
2900 }
2901 ASSERT(zb->zb_level >= 0);
2902
2903 if (BP_IS_HOLE(bp))
2904 return;
2905
2906 if (BP_IS_EMBEDDED(bp))
2907 return;
2908
2909 const dva_t *dva = bp->blk_dva;
2910 vdev_t *vd = spa->spa_root_vdev->vdev_child[DVA_GET_VDEV(&dva[0])];
2911 uint64_t file_offset = blkid2offset(dnp, bp, zb);
2912
2913 if (strcmp("raidz", vd->vdev_ops->vdev_op_type) == 0) {
2914 boolean_t last = (file_offset + BP_GET_LSIZE(bp)) >=
2915 (dnp->dn_used & ~0x03ffULL);
2916 print_file_layout_raidz(vd, bp, file_offset, last);
2917 }
2918 }
2919
2920 static int
visit_indirect(spa_t * spa,const dnode_phys_t * dnp,blkptr_t * bp,const zbookmark_phys_t * zb)2921 visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
2922 blkptr_t *bp, const zbookmark_phys_t *zb)
2923 {
2924 u_longlong_t offset = 0;
2925 int err = 0;
2926
2927 if (BP_GET_BIRTH(bp) == 0)
2928 return (0);
2929
2930 if (dump_opt['f']) {
2931 if (BP_GET_LEVEL(bp) == 0)
2932 print_file_layout(spa, bp, zb, dnp);
2933 } else {
2934 print_indirect(spa, bp, zb, dnp);
2935 }
2936
2937 if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
2938 arc_flags_t flags = ARC_FLAG_WAIT;
2939 int i;
2940 blkptr_t *cbp;
2941 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
2942 arc_buf_t *buf;
2943 uint64_t fill = 0;
2944 ASSERT(!BP_IS_REDACTED(bp));
2945
2946 err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
2947 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
2948 if (err)
2949 return (err);
2950 ASSERT(buf->b_data);
2951
2952 /* recursively visit blocks below this */
2953 cbp = buf->b_data;
2954 for (i = 0; i < epb; i++, cbp++) {
2955 zbookmark_phys_t czb;
2956
2957 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
2958 zb->zb_level - 1,
2959 zb->zb_blkid * epb + i);
2960 err = visit_indirect(spa, dnp, cbp, &czb);
2961 if (err)
2962 break;
2963 fill += BP_GET_FILL(cbp);
2964 }
2965 if (!err) {
2966 if (fill != BP_GET_FILL(bp)) {
2967 (void) printf("%16llx: Block pointer "
2968 "fill (%llu) does not match calculated "
2969 "value (%llu)\n", offset, BP_GET_FILL(bp),
2970 (u_longlong_t)fill);
2971 corruption_found = B_TRUE;
2972 }
2973 }
2974 arc_buf_destroy(buf, &buf);
2975 }
2976
2977 return (err);
2978 }
2979
2980 static void
dump_indirect(dnode_t * dn)2981 dump_indirect(dnode_t *dn)
2982 {
2983 dnode_phys_t *dnp = dn->dn_phys;
2984 zbookmark_phys_t czb;
2985
2986 (void) printf("Indirect blocks:\n");
2987 SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
2988 dn->dn_object, dnp->dn_nlevels - 1, 0);
2989 for (int j = 0; j < dnp->dn_nblkptr; j++) {
2990 czb.zb_blkid = j;
2991 (void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
2992 &dnp->dn_blkptr[j], &czb);
2993 }
2994
2995 (void) printf("\n");
2996 }
2997
2998 static int
dump_indirect_layout(dnode_t * dn)2999 dump_indirect_layout(dnode_t *dn)
3000 {
3001 dnode_phys_t *dnp = dn->dn_phys;
3002 zbookmark_phys_t czb;
3003
3004 spa_t *spa = dmu_objset_spa(dn->dn_objset);
3005 int ashift = spa->spa_root_vdev->vdev_child[0]->vdev_ashift;
3006 int children = spa->spa_root_vdev->vdev_child[0]->vdev_children;
3007
3008 if (strcmp(spa->spa_root_vdev->vdev_child[0]->vdev_ops->vdev_op_type,
3009 "raidz") != 0) {
3010 (void) fprintf(stderr, "file layout only supports raidz\n");
3011 return (ENOTSUP);
3012 }
3013
3014 /*
3015 * Start layout with a header
3016 */
3017 if (dump_opt['H']) {
3018 (void) printf("DISK\t\t\tLBA\tCOUNT");
3019 if (dump_opt['v'])
3020 (void) printf("\tTYPE");
3021 if (dump_opt['v'] > 1)
3022 (void) printf("\tOFFSET");
3023 printf("\n");
3024 } else {
3025 char diskhdr[16];
3026
3027 (void) printf("%12s: %d\n", "block size", dn->dn_datablksz);
3028 (void) printf("%12s: %s%d\n", "vdev_type",
3029 spa->spa_root_vdev->vdev_child[0]->vdev_ops->vdev_op_type,
3030 (int)vdev_get_nparity(spa->spa_root_vdev->vdev_child[0]));
3031 (void) printf("%12s: %d\n", "sector size", 1 << ashift);
3032 (void) printf("%12s: %d\n\n", "child disks", children);
3033
3034 (void) printf("%*c ", 12, ' ');
3035
3036 for (int c = 0; c < children; c++) {
3037 (void) snprintf(diskhdr, sizeof (diskhdr),
3038 "V%d:DISK-%d", 0, c);
3039 (void) printf(" %10s ", diskhdr);
3040 }
3041 (void) printf("\n");
3042
3043 (void) printf("%12s", "FILE OFFSET");
3044 for (int c = 0; c < children; c++)
3045 (void) printf("%s%10s", (c == 0) ? " ┌─" : "──",
3046 "──────────");
3047 (void) printf("─┐\n");
3048 }
3049
3050 SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
3051 dn->dn_object, dnp->dn_nlevels - 1, 0);
3052 for (int j = 0; j < dnp->dn_nblkptr; j++) {
3053 czb.zb_blkid = j;
3054 (void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
3055 &dnp->dn_blkptr[j], &czb);
3056 }
3057
3058 (void) printf("\n");
3059 return (0);
3060 }
3061
3062 static void
dump_dsl_dir(objset_t * os,uint64_t object,void * data,size_t size)3063 dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size)
3064 {
3065 (void) os, (void) object;
3066 dsl_dir_phys_t *dd = data;
3067 time_t crtime;
3068 char nice[32];
3069
3070 /* make sure nicenum has enough space */
3071 _Static_assert(sizeof (nice) >= NN_NUMBUF_SZ, "nice truncated");
3072
3073 if (dd == NULL)
3074 return;
3075
3076 ASSERT3U(size, >=, sizeof (dsl_dir_phys_t));
3077
3078 crtime = dd->dd_creation_time;
3079 (void) printf("\t\tcreation_time = %s", ctime(&crtime));
3080 (void) printf("\t\thead_dataset_obj = %llu\n",
3081 (u_longlong_t)dd->dd_head_dataset_obj);
3082 (void) printf("\t\tparent_dir_obj = %llu\n",
3083 (u_longlong_t)dd->dd_parent_obj);
3084 (void) printf("\t\torigin_obj = %llu\n",
3085 (u_longlong_t)dd->dd_origin_obj);
3086 (void) printf("\t\tchild_dir_zapobj = %llu\n",
3087 (u_longlong_t)dd->dd_child_dir_zapobj);
3088 zdb_nicenum(dd->dd_used_bytes, nice, sizeof (nice));
3089 (void) printf("\t\tused_bytes = %s\n", nice);
3090 zdb_nicenum(dd->dd_compressed_bytes, nice, sizeof (nice));
3091 (void) printf("\t\tcompressed_bytes = %s\n", nice);
3092 zdb_nicenum(dd->dd_uncompressed_bytes, nice, sizeof (nice));
3093 (void) printf("\t\tuncompressed_bytes = %s\n", nice);
3094 zdb_nicenum(dd->dd_quota, nice, sizeof (nice));
3095 (void) printf("\t\tquota = %s\n", nice);
3096 zdb_nicenum(dd->dd_reserved, nice, sizeof (nice));
3097 (void) printf("\t\treserved = %s\n", nice);
3098 (void) printf("\t\tprops_zapobj = %llu\n",
3099 (u_longlong_t)dd->dd_props_zapobj);
3100 (void) printf("\t\tdeleg_zapobj = %llu\n",
3101 (u_longlong_t)dd->dd_deleg_zapobj);
3102 (void) printf("\t\tflags = %llx\n",
3103 (u_longlong_t)dd->dd_flags);
3104
3105 #define DO(which) \
3106 zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
3107 sizeof (nice)); \
3108 (void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
3109 DO(HEAD);
3110 DO(SNAP);
3111 DO(CHILD);
3112 DO(CHILD_RSRV);
3113 DO(REFRSRV);
3114 #undef DO
3115 (void) printf("\t\tclones = %llu\n",
3116 (u_longlong_t)dd->dd_clones);
3117 }
3118
3119 static void
dump_dsl_dataset(objset_t * os,uint64_t object,void * data,size_t size)3120 dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
3121 {
3122 (void) os, (void) object;
3123 dsl_dataset_phys_t *ds = data;
3124 time_t crtime;
3125 char used[32], compressed[32], uncompressed[32], unique[32];
3126 char blkbuf[BP_SPRINTF_LEN];
3127
3128 /* make sure nicenum has enough space */
3129 _Static_assert(sizeof (used) >= NN_NUMBUF_SZ, "used truncated");
3130 _Static_assert(sizeof (compressed) >= NN_NUMBUF_SZ,
3131 "compressed truncated");
3132 _Static_assert(sizeof (uncompressed) >= NN_NUMBUF_SZ,
3133 "uncompressed truncated");
3134 _Static_assert(sizeof (unique) >= NN_NUMBUF_SZ, "unique truncated");
3135
3136 if (ds == NULL)
3137 return;
3138
3139 ASSERT(size == sizeof (*ds));
3140 crtime = ds->ds_creation_time;
3141 zdb_nicenum(ds->ds_referenced_bytes, used, sizeof (used));
3142 zdb_nicenum(ds->ds_compressed_bytes, compressed, sizeof (compressed));
3143 zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed,
3144 sizeof (uncompressed));
3145 zdb_nicenum(ds->ds_unique_bytes, unique, sizeof (unique));
3146 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp);
3147
3148 (void) printf("\t\tdir_obj = %llu\n",
3149 (u_longlong_t)ds->ds_dir_obj);
3150 (void) printf("\t\tprev_snap_obj = %llu\n",
3151 (u_longlong_t)ds->ds_prev_snap_obj);
3152 (void) printf("\t\tprev_snap_txg = %llu\n",
3153 (u_longlong_t)ds->ds_prev_snap_txg);
3154 (void) printf("\t\tnext_snap_obj = %llu\n",
3155 (u_longlong_t)ds->ds_next_snap_obj);
3156 (void) printf("\t\tsnapnames_zapobj = %llu\n",
3157 (u_longlong_t)ds->ds_snapnames_zapobj);
3158 (void) printf("\t\tnum_children = %llu\n",
3159 (u_longlong_t)ds->ds_num_children);
3160 (void) printf("\t\tuserrefs_obj = %llu\n",
3161 (u_longlong_t)ds->ds_userrefs_obj);
3162 (void) printf("\t\tcreation_time = %s", ctime(&crtime));
3163 (void) printf("\t\tcreation_txg = %llu\n",
3164 (u_longlong_t)ds->ds_creation_txg);
3165 (void) printf("\t\tdeadlist_obj = %llu\n",
3166 (u_longlong_t)ds->ds_deadlist_obj);
3167 (void) printf("\t\tused_bytes = %s\n", used);
3168 (void) printf("\t\tcompressed_bytes = %s\n", compressed);
3169 (void) printf("\t\tuncompressed_bytes = %s\n", uncompressed);
3170 (void) printf("\t\tunique = %s\n", unique);
3171 (void) printf("\t\tfsid_guid = %llu\n",
3172 (u_longlong_t)ds->ds_fsid_guid);
3173 (void) printf("\t\tguid = %llu\n",
3174 (u_longlong_t)ds->ds_guid);
3175 (void) printf("\t\tflags = %llx\n",
3176 (u_longlong_t)ds->ds_flags);
3177 (void) printf("\t\tnext_clones_obj = %llu\n",
3178 (u_longlong_t)ds->ds_next_clones_obj);
3179 (void) printf("\t\tprops_obj = %llu\n",
3180 (u_longlong_t)ds->ds_props_obj);
3181 (void) printf("\t\tbp = %s\n", blkbuf);
3182 }
3183
3184 static int
dump_bptree_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)3185 dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
3186 {
3187 (void) arg, (void) tx;
3188 char blkbuf[BP_SPRINTF_LEN];
3189
3190 if (BP_GET_BIRTH(bp) != 0) {
3191 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
3192 (void) printf("\t%s\n", blkbuf);
3193 }
3194 return (0);
3195 }
3196
3197 static void
dump_bptree(objset_t * os,uint64_t obj,const char * name)3198 dump_bptree(objset_t *os, uint64_t obj, const char *name)
3199 {
3200 char bytes[32];
3201 bptree_phys_t *bt;
3202 dmu_buf_t *db;
3203
3204 /* make sure nicenum has enough space */
3205 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3206
3207 if (dump_opt['d'] < 3)
3208 return;
3209
3210 VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
3211 bt = db->db_data;
3212 zdb_nicenum(bt->bt_bytes, bytes, sizeof (bytes));
3213 (void) printf("\n %s: %llu datasets, %s\n",
3214 name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes);
3215 dmu_buf_rele(db, FTAG);
3216
3217 if (dump_opt['d'] < 5)
3218 return;
3219
3220 (void) printf("\n");
3221
3222 (void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL);
3223 }
3224
3225 static int
dump_bpobj_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)3226 dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
3227 {
3228 (void) arg, (void) tx;
3229 char blkbuf[BP_SPRINTF_LEN];
3230
3231 ASSERT(BP_GET_BIRTH(bp) != 0);
3232 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed);
3233 (void) printf("\t%s\n", blkbuf);
3234 return (0);
3235 }
3236
3237 static void
dump_full_bpobj(bpobj_t * bpo,const char * name,int indent)3238 dump_full_bpobj(bpobj_t *bpo, const char *name, int indent)
3239 {
3240 char bytes[32];
3241 char comp[32];
3242 char uncomp[32];
3243 uint64_t i;
3244
3245 /* make sure nicenum has enough space */
3246 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3247 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
3248 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
3249
3250 if (dump_opt['d'] < 3)
3251 return;
3252
3253 zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes, sizeof (bytes));
3254 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
3255 zdb_nicenum(bpo->bpo_phys->bpo_comp, comp, sizeof (comp));
3256 zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp, sizeof (uncomp));
3257 if (bpo->bpo_havefreed) {
3258 (void) printf(" %*s: object %llu, %llu local "
3259 "blkptrs, %llu freed, %llu subobjs in object %llu, "
3260 "%s (%s/%s comp)\n",
3261 indent * 8, name,
3262 (u_longlong_t)bpo->bpo_object,
3263 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
3264 (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
3265 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
3266 (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
3267 bytes, comp, uncomp);
3268 } else {
3269 (void) printf(" %*s: object %llu, %llu local "
3270 "blkptrs, %llu subobjs in object %llu, "
3271 "%s (%s/%s comp)\n",
3272 indent * 8, name,
3273 (u_longlong_t)bpo->bpo_object,
3274 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
3275 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
3276 (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
3277 bytes, comp, uncomp);
3278 }
3279
3280 for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
3281 uint64_t subobj;
3282 bpobj_t subbpo;
3283 int error;
3284 VERIFY0(dmu_read(bpo->bpo_os,
3285 bpo->bpo_phys->bpo_subobjs,
3286 i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3287 error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3288 if (error != 0) {
3289 (void) printf("ERROR %u while trying to open "
3290 "subobj id %llu\n",
3291 error, (u_longlong_t)subobj);
3292 corruption_found = B_TRUE;
3293 continue;
3294 }
3295 dump_full_bpobj(&subbpo, "subobj", indent + 1);
3296 bpobj_close(&subbpo);
3297 }
3298 } else {
3299 if (bpo->bpo_havefreed) {
3300 (void) printf(" %*s: object %llu, %llu blkptrs, "
3301 "%llu freed, %s\n",
3302 indent * 8, name,
3303 (u_longlong_t)bpo->bpo_object,
3304 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
3305 (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
3306 bytes);
3307 } else {
3308 (void) printf(" %*s: object %llu, %llu blkptrs, "
3309 "%s\n",
3310 indent * 8, name,
3311 (u_longlong_t)bpo->bpo_object,
3312 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
3313 bytes);
3314 }
3315 }
3316
3317 if (dump_opt['d'] < 5)
3318 return;
3319
3320 if (indent == 0) {
3321 (void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL);
3322 (void) printf("\n");
3323 }
3324 }
3325
3326 static int
dump_bookmark(dsl_pool_t * dp,char * name,boolean_t print_redact,boolean_t print_list)3327 dump_bookmark(dsl_pool_t *dp, char *name, boolean_t print_redact,
3328 boolean_t print_list)
3329 {
3330 int err = 0;
3331 zfs_bookmark_phys_t prop;
3332 objset_t *mos = dp->dp_spa->spa_meta_objset;
3333 err = dsl_bookmark_lookup(dp, name, NULL, &prop);
3334
3335 if (err != 0) {
3336 return (err);
3337 }
3338
3339 (void) printf("\t#%s: ", strchr(name, '#') + 1);
3340 (void) printf("{guid: %llx creation_txg: %llu creation_time: "
3341 "%llu redaction_obj: %llu}\n", (u_longlong_t)prop.zbm_guid,
3342 (u_longlong_t)prop.zbm_creation_txg,
3343 (u_longlong_t)prop.zbm_creation_time,
3344 (u_longlong_t)prop.zbm_redaction_obj);
3345
3346 IMPLY(print_list, print_redact);
3347 if (!print_redact || prop.zbm_redaction_obj == 0)
3348 return (0);
3349
3350 redaction_list_t *rl;
3351 VERIFY0(dsl_redaction_list_hold_obj(dp,
3352 prop.zbm_redaction_obj, FTAG, &rl));
3353
3354 redaction_list_phys_t *rlp = rl->rl_phys;
3355 (void) printf("\tRedacted:\n\t\tProgress: ");
3356 if (rlp->rlp_last_object != UINT64_MAX ||
3357 rlp->rlp_last_blkid != UINT64_MAX) {
3358 (void) printf("%llu %llu (incomplete)\n",
3359 (u_longlong_t)rlp->rlp_last_object,
3360 (u_longlong_t)rlp->rlp_last_blkid);
3361 } else {
3362 (void) printf("complete\n");
3363 }
3364 (void) printf("\t\tSnapshots: [");
3365 for (unsigned int i = 0; i < rlp->rlp_num_snaps; i++) {
3366 if (i > 0)
3367 (void) printf(", ");
3368 (void) printf("%0llu",
3369 (u_longlong_t)rlp->rlp_snaps[i]);
3370 }
3371 (void) printf("]\n\t\tLength: %llu\n",
3372 (u_longlong_t)rlp->rlp_num_entries);
3373
3374 if (!print_list) {
3375 dsl_redaction_list_rele(rl, FTAG);
3376 return (0);
3377 }
3378
3379 if (rlp->rlp_num_entries == 0) {
3380 dsl_redaction_list_rele(rl, FTAG);
3381 (void) printf("\t\tRedaction List: []\n\n");
3382 return (0);
3383 }
3384
3385 redact_block_phys_t *rbp_buf;
3386 uint64_t size;
3387 dmu_object_info_t doi;
3388
3389 VERIFY0(dmu_object_info(mos, prop.zbm_redaction_obj, &doi));
3390 size = doi.doi_max_offset;
3391 rbp_buf = kmem_alloc(size, KM_SLEEP);
3392
3393 err = dmu_read(mos, prop.zbm_redaction_obj, 0, size,
3394 rbp_buf, 0);
3395 if (err != 0) {
3396 dsl_redaction_list_rele(rl, FTAG);
3397 kmem_free(rbp_buf, size);
3398 return (err);
3399 }
3400
3401 (void) printf("\t\tRedaction List: [{object: %llx, offset: "
3402 "%llx, blksz: %x, count: %llx}",
3403 (u_longlong_t)rbp_buf[0].rbp_object,
3404 (u_longlong_t)rbp_buf[0].rbp_blkid,
3405 (uint_t)(redact_block_get_size(&rbp_buf[0])),
3406 (u_longlong_t)redact_block_get_count(&rbp_buf[0]));
3407
3408 for (size_t i = 1; i < rlp->rlp_num_entries; i++) {
3409 (void) printf(",\n\t\t{object: %llx, offset: %llx, "
3410 "blksz: %x, count: %llx}",
3411 (u_longlong_t)rbp_buf[i].rbp_object,
3412 (u_longlong_t)rbp_buf[i].rbp_blkid,
3413 (uint_t)(redact_block_get_size(&rbp_buf[i])),
3414 (u_longlong_t)redact_block_get_count(&rbp_buf[i]));
3415 }
3416 dsl_redaction_list_rele(rl, FTAG);
3417 kmem_free(rbp_buf, size);
3418 (void) printf("]\n\n");
3419 return (0);
3420 }
3421
3422 static void
dump_bookmarks(objset_t * os,int verbosity)3423 dump_bookmarks(objset_t *os, int verbosity)
3424 {
3425 zap_cursor_t zc;
3426 zap_attribute_t *attrp;
3427 dsl_dataset_t *ds = dmu_objset_ds(os);
3428 dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3429 objset_t *mos = os->os_spa->spa_meta_objset;
3430 if (verbosity < 4)
3431 return;
3432 attrp = zap_attribute_alloc();
3433 dsl_pool_config_enter(dp, FTAG);
3434
3435 for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj);
3436 zap_cursor_retrieve(&zc, attrp) == 0;
3437 zap_cursor_advance(&zc)) {
3438 char osname[ZFS_MAX_DATASET_NAME_LEN];
3439 char buf[ZFS_MAX_DATASET_NAME_LEN];
3440 int len;
3441 dmu_objset_name(os, osname);
3442 len = snprintf(buf, sizeof (buf), "%s#%s", osname,
3443 attrp->za_name);
3444 VERIFY3S(len, <, ZFS_MAX_DATASET_NAME_LEN);
3445 (void) dump_bookmark(dp, buf, verbosity >= 5, verbosity >= 6);
3446 }
3447 zap_cursor_fini(&zc);
3448 dsl_pool_config_exit(dp, FTAG);
3449 zap_attribute_free(attrp);
3450 }
3451
3452 static void
bpobj_count_refd(bpobj_t * bpo)3453 bpobj_count_refd(bpobj_t *bpo)
3454 {
3455 mos_obj_refd(bpo->bpo_object);
3456
3457 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
3458 mos_obj_refd(bpo->bpo_phys->bpo_subobjs);
3459 for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
3460 uint64_t subobj;
3461 bpobj_t subbpo;
3462 int error;
3463 VERIFY0(dmu_read(bpo->bpo_os,
3464 bpo->bpo_phys->bpo_subobjs,
3465 i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3466 error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3467 if (error != 0) {
3468 (void) printf("ERROR %u while trying to open "
3469 "subobj id %llu\n",
3470 error, (u_longlong_t)subobj);
3471 corruption_found = B_TRUE;
3472 continue;
3473 }
3474 bpobj_count_refd(&subbpo);
3475 bpobj_close(&subbpo);
3476 }
3477 }
3478 }
3479
3480 static int
dsl_deadlist_entry_count_refd(void * arg,dsl_deadlist_entry_t * dle)3481 dsl_deadlist_entry_count_refd(void *arg, dsl_deadlist_entry_t *dle)
3482 {
3483 spa_t *spa = arg;
3484 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3485 if (dle->dle_bpobj.bpo_object != empty_bpobj)
3486 bpobj_count_refd(&dle->dle_bpobj);
3487 return (0);
3488 }
3489
3490 static int
dsl_deadlist_entry_dump(void * arg,dsl_deadlist_entry_t * dle)3491 dsl_deadlist_entry_dump(void *arg, dsl_deadlist_entry_t *dle)
3492 {
3493 ASSERT0P(arg);
3494 if (dump_opt['d'] >= 5) {
3495 char buf[128];
3496 (void) snprintf(buf, sizeof (buf),
3497 "mintxg %llu -> obj %llu",
3498 (longlong_t)dle->dle_mintxg,
3499 (longlong_t)dle->dle_bpobj.bpo_object);
3500
3501 dump_full_bpobj(&dle->dle_bpobj, buf, 0);
3502 } else {
3503 (void) printf("mintxg %llu -> obj %llu\n",
3504 (longlong_t)dle->dle_mintxg,
3505 (longlong_t)dle->dle_bpobj.bpo_object);
3506 }
3507 return (0);
3508 }
3509
3510 static void
dump_blkptr_list(dsl_deadlist_t * dl,const char * name)3511 dump_blkptr_list(dsl_deadlist_t *dl, const char *name)
3512 {
3513 char bytes[32];
3514 char comp[32];
3515 char uncomp[32];
3516 char entries[32];
3517 spa_t *spa = dmu_objset_spa(dl->dl_os);
3518 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3519
3520 if (dl->dl_oldfmt) {
3521 if (dl->dl_bpobj.bpo_object != empty_bpobj)
3522 bpobj_count_refd(&dl->dl_bpobj);
3523 } else {
3524 mos_obj_refd(dl->dl_object);
3525 dsl_deadlist_iterate(dl, dsl_deadlist_entry_count_refd, spa);
3526 }
3527
3528 /* make sure nicenum has enough space */
3529 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3530 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
3531 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
3532 _Static_assert(sizeof (entries) >= NN_NUMBUF_SZ, "entries truncated");
3533
3534 if (dump_opt['d'] < 3)
3535 return;
3536
3537 if (dl->dl_oldfmt) {
3538 dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0);
3539 return;
3540 }
3541
3542 zdb_nicenum(dl->dl_phys->dl_used, bytes, sizeof (bytes));
3543 zdb_nicenum(dl->dl_phys->dl_comp, comp, sizeof (comp));
3544 zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp, sizeof (uncomp));
3545 zdb_nicenum(avl_numnodes(&dl->dl_tree), entries, sizeof (entries));
3546 (void) printf("\n %s: %s (%s/%s comp), %s entries\n",
3547 name, bytes, comp, uncomp, entries);
3548
3549 if (dump_opt['d'] < 4)
3550 return;
3551
3552 (void) putchar('\n');
3553
3554 dsl_deadlist_iterate(dl, dsl_deadlist_entry_dump, NULL);
3555 }
3556
3557 static int
verify_dd_livelist(objset_t * os)3558 verify_dd_livelist(objset_t *os)
3559 {
3560 uint64_t ll_used, used, ll_comp, comp, ll_uncomp, uncomp;
3561 dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3562 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
3563
3564 ASSERT(!dmu_objset_is_snapshot(os));
3565 if (!dsl_deadlist_is_open(&dd->dd_livelist))
3566 return (0);
3567
3568 /* Iterate through the livelist to check for duplicates */
3569 dsl_deadlist_iterate(&dd->dd_livelist, sublivelist_verify_lightweight,
3570 NULL);
3571
3572 dsl_pool_config_enter(dp, FTAG);
3573 dsl_deadlist_space(&dd->dd_livelist, &ll_used,
3574 &ll_comp, &ll_uncomp);
3575
3576 dsl_dataset_t *origin_ds;
3577 ASSERT(dsl_pool_config_held(dp));
3578 VERIFY0(dsl_dataset_hold_obj(dp,
3579 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &origin_ds));
3580 VERIFY0(dsl_dataset_space_written(origin_ds, os->os_dsl_dataset,
3581 &used, &comp, &uncomp));
3582 dsl_dataset_rele(origin_ds, FTAG);
3583 dsl_pool_config_exit(dp, FTAG);
3584 /*
3585 * It's possible that the dataset's uncomp space is larger than the
3586 * livelist's because livelists do not track embedded block pointers
3587 */
3588 if (used != ll_used || comp != ll_comp || uncomp < ll_uncomp) {
3589 char nice_used[32], nice_comp[32], nice_uncomp[32];
3590 (void) printf("Discrepancy in space accounting:\n");
3591 zdb_nicenum(used, nice_used, sizeof (nice_used));
3592 zdb_nicenum(comp, nice_comp, sizeof (nice_comp));
3593 zdb_nicenum(uncomp, nice_uncomp, sizeof (nice_uncomp));
3594 (void) printf("dir: used %s, comp %s, uncomp %s\n",
3595 nice_used, nice_comp, nice_uncomp);
3596 zdb_nicenum(ll_used, nice_used, sizeof (nice_used));
3597 zdb_nicenum(ll_comp, nice_comp, sizeof (nice_comp));
3598 zdb_nicenum(ll_uncomp, nice_uncomp, sizeof (nice_uncomp));
3599 (void) printf("livelist: used %s, comp %s, uncomp %s\n",
3600 nice_used, nice_comp, nice_uncomp);
3601 return (1);
3602 }
3603 return (0);
3604 }
3605
3606 static char *key_material = NULL;
3607
3608 static boolean_t
zdb_derive_key(dsl_dir_t * dd,uint8_t * key_out)3609 zdb_derive_key(dsl_dir_t *dd, uint8_t *key_out)
3610 {
3611 uint64_t keyformat, salt, iters;
3612 int i;
3613 unsigned char c;
3614 FILE *f;
3615
3616 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3617 zfs_prop_to_name(ZFS_PROP_KEYFORMAT), sizeof (uint64_t),
3618 1, &keyformat));
3619
3620 switch (keyformat) {
3621 case ZFS_KEYFORMAT_HEX:
3622 for (i = 0; i < WRAPPING_KEY_LEN * 2; i += 2) {
3623 if (!isxdigit(key_material[i]) ||
3624 !isxdigit(key_material[i+1]))
3625 return (B_FALSE);
3626 if (sscanf(&key_material[i], "%02hhx", &c) != 1)
3627 return (B_FALSE);
3628 key_out[i / 2] = c;
3629 }
3630 break;
3631
3632 case ZFS_KEYFORMAT_PASSPHRASE:
3633 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3634 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT),
3635 sizeof (uint64_t), 1, &salt));
3636 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3637 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS),
3638 sizeof (uint64_t), 1, &iters));
3639
3640 if (PKCS5_PBKDF2_HMAC_SHA1(key_material, strlen(key_material),
3641 ((uint8_t *)&salt), sizeof (uint64_t), iters,
3642 WRAPPING_KEY_LEN, key_out) != 1)
3643 return (B_FALSE);
3644
3645 break;
3646
3647 case ZFS_KEYFORMAT_RAW:
3648 if ((f = fopen(key_material, "r")) == NULL)
3649 return (B_FALSE);
3650
3651 if (fread(key_out, 1, WRAPPING_KEY_LEN, f) !=
3652 WRAPPING_KEY_LEN) {
3653 (void) fclose(f);
3654 return (B_FALSE);
3655 }
3656
3657 /* Check the key length */
3658 if (fgetc(f) != EOF) {
3659 (void) fclose(f);
3660 return (B_FALSE);
3661 }
3662
3663 (void) fclose(f);
3664 break;
3665
3666 default:
3667 fatal("no support for key format %u\n",
3668 (unsigned int) keyformat);
3669 }
3670
3671 return (B_TRUE);
3672 }
3673
3674 static char encroot[ZFS_MAX_DATASET_NAME_LEN];
3675 static boolean_t key_loaded = B_FALSE;
3676
3677 static int
zdb_load_key(objset_t * os)3678 zdb_load_key(objset_t *os)
3679 {
3680 dsl_pool_t *dp;
3681 dsl_dir_t *dd, *rdd;
3682 uint8_t key[WRAPPING_KEY_LEN];
3683 uint64_t rddobj;
3684 int err = 0;
3685
3686 dp = spa_get_dsl(os->os_spa);
3687 dd = os->os_dsl_dataset->ds_dir;
3688
3689 dsl_pool_config_enter(dp, FTAG);
3690 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3691 DSL_CRYPTO_KEY_ROOT_DDOBJ, sizeof (uint64_t), 1, &rddobj));
3692 VERIFY0(dsl_dir_hold_obj(dd->dd_pool, rddobj, NULL, FTAG, &rdd));
3693 dsl_dir_name(rdd, encroot);
3694 dsl_dir_rele(rdd, FTAG);
3695
3696 if (!zdb_derive_key(dd, key))
3697 err = EINVAL;
3698 dsl_pool_config_exit(dp, FTAG);
3699
3700 if (err != 0) {
3701 fprintf(stderr, "couldn't derive encryption key\n");
3702 return (err);
3703 }
3704
3705 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_UNAVAILABLE);
3706
3707 dsl_crypto_params_t *dcp;
3708 nvlist_t *crypto_args;
3709
3710 crypto_args = fnvlist_alloc();
3711 fnvlist_add_uint8_array(crypto_args, "wkeydata",
3712 (uint8_t *)key, WRAPPING_KEY_LEN);
3713 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE,
3714 NULL, crypto_args, &dcp));
3715 err = spa_keystore_load_wkey(encroot, dcp, B_FALSE);
3716
3717 dsl_crypto_params_free(dcp, (err != 0));
3718 fnvlist_free(crypto_args);
3719
3720 if (err != 0) {
3721 fprintf(stderr,
3722 "couldn't load encryption key for %s: %s\n",
3723 encroot, err == ZFS_ERR_CRYPTO_NOTSUP ?
3724 "crypto params not supported" : strerror(err));
3725 return (err);
3726 }
3727
3728 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_AVAILABLE);
3729
3730 printf("Unlocked encryption root: %s\n", encroot);
3731 key_loaded = B_TRUE;
3732
3733 return (0);
3734 }
3735
3736 static void
zdb_unload_key(void)3737 zdb_unload_key(void)
3738 {
3739 if (!key_loaded)
3740 return;
3741
3742 VERIFY0(spa_keystore_unload_wkey(encroot));
3743 key_loaded = B_FALSE;
3744 }
3745
3746 static avl_tree_t idx_tree;
3747 static avl_tree_t domain_tree;
3748 static boolean_t fuid_table_loaded;
3749 static objset_t *sa_os = NULL;
3750 static sa_attr_type_t *sa_attr_table = NULL;
3751
3752 static int
open_objset(const char * path,const void * tag,objset_t ** osp)3753 open_objset(const char *path, const void *tag, objset_t **osp)
3754 {
3755 int err;
3756 uint64_t sa_attrs = 0;
3757 uint64_t version = 0;
3758
3759 VERIFY0P(sa_os);
3760
3761 /*
3762 * We can't own an objset if it's redacted. Therefore, we do this
3763 * dance: hold the objset, then acquire a long hold on its dataset, then
3764 * release the pool (which is held as part of holding the objset).
3765 */
3766
3767 if (dump_opt['K']) {
3768 /* decryption requested, try to load keys */
3769 err = dmu_objset_hold(path, tag, osp);
3770 if (err != 0) {
3771 (void) fprintf(stderr, "failed to hold dataset "
3772 "'%s': %s\n",
3773 path, strerror(err));
3774 return (err);
3775 }
3776
3777 /*
3778 * Only try to load the key and unlock the dataset if it is
3779 * actually encrypted; otherwise we'll just crash. Just
3780 * ignore the -K switch entirely otherwise; it's useful to be
3781 * able to provide even if it's not needed.
3782 */
3783 if ((*osp)->os_encrypted) {
3784 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3785 dsl_pool_rele(dmu_objset_pool(*osp), tag);
3786
3787 err = zdb_load_key(*osp);
3788
3789 /* release it all */
3790 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3791 dsl_dataset_rele(dmu_objset_ds(*osp), tag);
3792
3793 if (err != 0) {
3794 *osp = NULL;
3795 return (err);
3796 }
3797 } else {
3798 dmu_objset_rele(*osp, tag);
3799 }
3800 }
3801
3802 int ds_hold_flags = key_loaded ? DS_HOLD_FLAG_DECRYPT : 0;
3803
3804 err = dmu_objset_hold_flags(path, ds_hold_flags, tag, osp);
3805 if (err != 0) {
3806 (void) fprintf(stderr, "failed to hold dataset '%s': %s\n",
3807 path, strerror(err));
3808 *osp = NULL;
3809 return (err);
3810 }
3811 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3812 dsl_pool_rele(dmu_objset_pool(*osp), tag);
3813
3814 if (dmu_objset_type(*osp) == DMU_OST_ZFS &&
3815 (key_loaded || !(*osp)->os_encrypted)) {
3816 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR,
3817 8, 1, &version);
3818 if (version >= ZPL_VERSION_SA) {
3819 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS,
3820 8, 1, &sa_attrs);
3821 }
3822 err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END,
3823 &sa_attr_table);
3824 if (err != 0) {
3825 (void) fprintf(stderr, "sa_setup failed: %s\n",
3826 strerror(err));
3827 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3828 dsl_dataset_rele_flags(dmu_objset_ds(*osp),
3829 ds_hold_flags, tag);
3830 *osp = NULL;
3831 }
3832 }
3833 sa_os = *osp;
3834
3835 return (err);
3836 }
3837
3838 static void
close_objset(objset_t * os,const void * tag)3839 close_objset(objset_t *os, const void *tag)
3840 {
3841 VERIFY3P(os, ==, sa_os);
3842 if (os->os_sa != NULL)
3843 sa_tear_down(os);
3844 dsl_dataset_long_rele(dmu_objset_ds(os), tag);
3845 dsl_dataset_rele_flags(dmu_objset_ds(os),
3846 key_loaded ? DS_HOLD_FLAG_DECRYPT : 0, tag);
3847 sa_attr_table = NULL;
3848 sa_os = NULL;
3849
3850 zdb_unload_key();
3851 }
3852
3853 static void
fuid_table_destroy(void)3854 fuid_table_destroy(void)
3855 {
3856 if (fuid_table_loaded) {
3857 zfs_fuid_table_destroy(&idx_tree, &domain_tree);
3858 fuid_table_loaded = B_FALSE;
3859 }
3860 }
3861
3862 /*
3863 * Clean up DDT internal state. ddt_lookup() adds entries to ddt_tree, which on
3864 * a live pool are normally cleaned up during ddt_sync(). We can't do that (and
3865 * wouldn't want to anyway), but if we don't clean up the presence of stuff on
3866 * ddt_tree will trip asserts in ddt_table_free(). So, we clean up ourselves.
3867 *
3868 * Note that this is not a particularly efficient way to do this, but
3869 * ddt_remove() is the only public method that can do the work we need, and it
3870 * requires the right locks and etc to do the job. This is only ever called
3871 * during zdb shutdown so efficiency is not especially important.
3872 */
3873 static void
zdb_ddt_cleanup(spa_t * spa)3874 zdb_ddt_cleanup(spa_t *spa)
3875 {
3876 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
3877 ddt_t *ddt = spa->spa_ddt[c];
3878 if (!ddt)
3879 continue;
3880
3881 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3882 ddt_enter(ddt);
3883 ddt_entry_t *dde = avl_first(&ddt->ddt_tree), *next;
3884 while (dde) {
3885 next = AVL_NEXT(&ddt->ddt_tree, dde);
3886 dde->dde_io = NULL;
3887 ddt_remove(ddt, dde);
3888 dde = next;
3889 }
3890 ddt_exit(ddt);
3891 spa_config_exit(spa, SCL_CONFIG, FTAG);
3892 }
3893 }
3894
3895 static void
zdb_exit(int reason)3896 zdb_exit(int reason)
3897 {
3898 if (spa != NULL)
3899 zdb_ddt_cleanup(spa);
3900
3901 if (os != NULL) {
3902 close_objset(os, FTAG);
3903 } else if (spa != NULL) {
3904 spa_close(spa, FTAG);
3905 }
3906
3907 fuid_table_destroy();
3908
3909 if (kernel_init_done)
3910 kernel_fini();
3911
3912 exit(reason);
3913 }
3914
3915 /*
3916 * print uid or gid information.
3917 * For normal POSIX id just the id is printed in decimal format.
3918 * For CIFS files with FUID the fuid is printed in hex followed by
3919 * the domain-rid string.
3920 */
3921 static void
print_idstr(uint64_t id,const char * id_type)3922 print_idstr(uint64_t id, const char *id_type)
3923 {
3924 if (FUID_INDEX(id)) {
3925 const char *domain =
3926 zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id));
3927 (void) printf("\t%s %llx [%s-%d]\n", id_type,
3928 (u_longlong_t)id, domain, (int)FUID_RID(id));
3929 } else {
3930 (void) printf("\t%s %llu\n", id_type, (u_longlong_t)id);
3931 }
3932
3933 }
3934
3935 static void
dump_uidgid(objset_t * os,uint64_t uid,uint64_t gid)3936 dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid)
3937 {
3938 uint32_t uid_idx, gid_idx;
3939
3940 uid_idx = FUID_INDEX(uid);
3941 gid_idx = FUID_INDEX(gid);
3942
3943 /* Load domain table, if not already loaded */
3944 if (!fuid_table_loaded && (uid_idx || gid_idx)) {
3945 uint64_t fuid_obj;
3946
3947 /* first find the fuid object. It lives in the master node */
3948 VERIFY0(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES,
3949 8, 1, &fuid_obj));
3950 zfs_fuid_avl_tree_create(&idx_tree, &domain_tree);
3951 (void) zfs_fuid_table_load(os, fuid_obj,
3952 &idx_tree, &domain_tree);
3953 fuid_table_loaded = B_TRUE;
3954 }
3955
3956 print_idstr(uid, "uid");
3957 print_idstr(gid, "gid");
3958 }
3959
3960 static void
dump_znode_sa_xattr(sa_handle_t * hdl)3961 dump_znode_sa_xattr(sa_handle_t *hdl)
3962 {
3963 nvlist_t *sa_xattr;
3964 nvpair_t *elem = NULL;
3965 int sa_xattr_size = 0;
3966 int sa_xattr_entries = 0;
3967 int error;
3968 char *sa_xattr_packed;
3969
3970 error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size);
3971 if (error || sa_xattr_size == 0)
3972 return;
3973
3974 sa_xattr_packed = malloc(sa_xattr_size);
3975 if (sa_xattr_packed == NULL)
3976 return;
3977
3978 error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR],
3979 sa_xattr_packed, sa_xattr_size);
3980 if (error) {
3981 free(sa_xattr_packed);
3982 return;
3983 }
3984
3985 error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0);
3986 if (error) {
3987 free(sa_xattr_packed);
3988 return;
3989 }
3990
3991 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL)
3992 sa_xattr_entries++;
3993
3994 (void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3995 sa_xattr_size, sa_xattr_entries);
3996 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) {
3997 boolean_t can_print = !dump_opt['P'];
3998 uchar_t *value;
3999 uint_t cnt, idx;
4000
4001 (void) printf("\t\t%s = ", nvpair_name(elem));
4002 nvpair_value_byte_array(elem, &value, &cnt);
4003
4004 for (idx = 0; idx < cnt; ++idx) {
4005 if (!isprint(value[idx])) {
4006 can_print = B_FALSE;
4007 break;
4008 }
4009 }
4010
4011 for (idx = 0; idx < cnt; ++idx) {
4012 if (can_print)
4013 (void) putchar(value[idx]);
4014 else
4015 (void) printf("\\%3.3o", value[idx]);
4016 }
4017 (void) putchar('\n');
4018 }
4019
4020 nvlist_free(sa_xattr);
4021 free(sa_xattr_packed);
4022 }
4023
4024 static void
dump_znode_symlink(sa_handle_t * hdl)4025 dump_znode_symlink(sa_handle_t *hdl)
4026 {
4027 int sa_symlink_size = 0;
4028 char linktarget[MAXPATHLEN];
4029 int error;
4030
4031 error = sa_size(hdl, sa_attr_table[ZPL_SYMLINK], &sa_symlink_size);
4032 if (error || sa_symlink_size == 0) {
4033 return;
4034 }
4035 if (sa_symlink_size >= sizeof (linktarget)) {
4036 (void) printf("symlink size %d is too large\n",
4037 sa_symlink_size);
4038 return;
4039 }
4040 linktarget[sa_symlink_size] = '\0';
4041 if (sa_lookup(hdl, sa_attr_table[ZPL_SYMLINK],
4042 &linktarget, sa_symlink_size) == 0)
4043 (void) printf("\ttarget %s\n", linktarget);
4044 }
4045
4046 static void
dump_znode(objset_t * os,uint64_t object,void * data,size_t size)4047 dump_znode(objset_t *os, uint64_t object, void *data, size_t size)
4048 {
4049 (void) data, (void) size;
4050 char path[MAXPATHLEN * 2]; /* allow for xattr and failure prefix */
4051 sa_handle_t *hdl;
4052 uint64_t xattr, rdev, gen;
4053 uint64_t uid, gid, mode, fsize, parent, links;
4054 uint64_t pflags;
4055 uint64_t acctm[2], modtm[2], chgtm[2], crtm[2];
4056 time_t z_crtime, z_atime, z_mtime, z_ctime;
4057 sa_bulk_attr_t bulk[12];
4058 int idx = 0;
4059 int error;
4060
4061 VERIFY3P(os, ==, sa_os);
4062 if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) {
4063 (void) printf("Failed to get handle for SA znode\n");
4064 return;
4065 }
4066
4067 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8);
4068 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8);
4069 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL,
4070 &links, 8);
4071 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8);
4072 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL,
4073 &mode, 8);
4074 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT],
4075 NULL, &parent, 8);
4076 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL,
4077 &fsize, 8);
4078 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL,
4079 acctm, 16);
4080 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL,
4081 modtm, 16);
4082 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL,
4083 crtm, 16);
4084 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL,
4085 chgtm, 16);
4086 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL,
4087 &pflags, 8);
4088
4089 if (sa_bulk_lookup(hdl, bulk, idx)) {
4090 (void) sa_handle_destroy(hdl);
4091 return;
4092 }
4093
4094 z_crtime = (time_t)crtm[0];
4095 z_atime = (time_t)acctm[0];
4096 z_mtime = (time_t)modtm[0];
4097 z_ctime = (time_t)chgtm[0];
4098
4099 if (dump_opt['d'] > 4) {
4100 error = zfs_obj_to_path(os, object, path, sizeof (path));
4101 if (error == ESTALE) {
4102 (void) snprintf(path, sizeof (path), "on delete queue");
4103 } else if (error != 0) {
4104 leaked_objects++;
4105 (void) snprintf(path, sizeof (path),
4106 "path not found, possibly leaked");
4107 }
4108 (void) printf("\tpath %s\n", path);
4109 }
4110
4111 if (S_ISLNK(mode))
4112 dump_znode_symlink(hdl);
4113 dump_uidgid(os, uid, gid);
4114 (void) printf("\tatime %s", ctime(&z_atime));
4115 (void) printf("\tmtime %s", ctime(&z_mtime));
4116 (void) printf("\tctime %s", ctime(&z_ctime));
4117 (void) printf("\tcrtime %s", ctime(&z_crtime));
4118 (void) printf("\tgen %llu\n", (u_longlong_t)gen);
4119 (void) printf("\tmode %llo\n", (u_longlong_t)mode);
4120 (void) printf("\tsize %llu\n", (u_longlong_t)fsize);
4121 (void) printf("\tparent %llu\n", (u_longlong_t)parent);
4122 (void) printf("\tlinks %llu\n", (u_longlong_t)links);
4123 (void) printf("\tpflags %llx\n", (u_longlong_t)pflags);
4124 if (dmu_objset_projectquota_enabled(os) && (pflags & ZFS_PROJID)) {
4125 uint64_t projid;
4126
4127 if (sa_lookup(hdl, sa_attr_table[ZPL_PROJID], &projid,
4128 sizeof (uint64_t)) == 0)
4129 (void) printf("\tprojid %llu\n", (u_longlong_t)projid);
4130 }
4131 if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr,
4132 sizeof (uint64_t)) == 0)
4133 (void) printf("\txattr %llu\n", (u_longlong_t)xattr);
4134 if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev,
4135 sizeof (uint64_t)) == 0)
4136 (void) printf("\trdev 0x%016llx\n", (u_longlong_t)rdev);
4137 dump_znode_sa_xattr(hdl);
4138 sa_handle_destroy(hdl);
4139 }
4140
4141 static void
dump_acl(objset_t * os,uint64_t object,void * data,size_t size)4142 dump_acl(objset_t *os, uint64_t object, void *data, size_t size)
4143 {
4144 (void) os, (void) object, (void) data, (void) size;
4145 }
4146
4147 static void
dump_dmu_objset(objset_t * os,uint64_t object,void * data,size_t size)4148 dump_dmu_objset(objset_t *os, uint64_t object, void *data, size_t size)
4149 {
4150 (void) os, (void) object, (void) data, (void) size;
4151 }
4152
4153 static object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = {
4154 dump_none, /* unallocated */
4155 dump_zap, /* object directory */
4156 dump_uint64, /* object array */
4157 dump_none, /* packed nvlist */
4158 dump_packed_nvlist, /* packed nvlist size */
4159 dump_none, /* bpobj */
4160 dump_bpobj, /* bpobj header */
4161 dump_none, /* SPA space map header */
4162 dump_none, /* SPA space map */
4163 dump_none, /* ZIL intent log */
4164 dump_dnode, /* DMU dnode */
4165 dump_dmu_objset, /* DMU objset */
4166 dump_dsl_dir, /* DSL directory */
4167 dump_zap, /* DSL directory child map */
4168 dump_zap, /* DSL dataset snap map */
4169 dump_zap, /* DSL props */
4170 dump_dsl_dataset, /* DSL dataset */
4171 dump_znode, /* ZFS znode */
4172 dump_acl, /* ZFS V0 ACL */
4173 dump_uint8, /* ZFS plain file */
4174 dump_zpldir, /* ZFS directory */
4175 dump_zap, /* ZFS master node */
4176 dump_zap, /* ZFS delete queue */
4177 dump_uint8, /* zvol object */
4178 dump_zap, /* zvol prop */
4179 dump_uint8, /* other uint8[] */
4180 dump_uint64, /* other uint64[] */
4181 dump_zap, /* other ZAP */
4182 dump_zap, /* persistent error log */
4183 dump_uint8, /* SPA history */
4184 dump_history_offsets, /* SPA history offsets */
4185 dump_zap, /* Pool properties */
4186 dump_zap, /* DSL permissions */
4187 dump_acl, /* ZFS ACL */
4188 dump_uint8, /* ZFS SYSACL */
4189 dump_none, /* FUID nvlist */
4190 dump_packed_nvlist, /* FUID nvlist size */
4191 dump_zap, /* DSL dataset next clones */
4192 dump_zap, /* DSL scrub queue */
4193 dump_zap, /* ZFS user/group/project used */
4194 dump_zap, /* ZFS user/group/project quota */
4195 dump_zap, /* snapshot refcount tags */
4196 dump_ddt_zap, /* DDT ZAP object */
4197 dump_zap, /* DDT statistics */
4198 dump_znode, /* SA object */
4199 dump_zap, /* SA Master Node */
4200 dump_sa_attrs, /* SA attribute registration */
4201 dump_sa_layouts, /* SA attribute layouts */
4202 dump_zap, /* DSL scrub translations */
4203 dump_none, /* fake dedup BP */
4204 dump_zap, /* deadlist */
4205 dump_none, /* deadlist hdr */
4206 dump_zap, /* dsl clones */
4207 dump_bpobj_subobjs, /* bpobj subobjs */
4208 dump_unknown, /* Unknown type, must be last */
4209 };
4210
4211 static boolean_t
match_object_type(dmu_object_type_t obj_type,uint64_t flags)4212 match_object_type(dmu_object_type_t obj_type, uint64_t flags)
4213 {
4214 boolean_t match = B_TRUE;
4215
4216 switch (obj_type) {
4217 case DMU_OT_DIRECTORY_CONTENTS:
4218 if (!(flags & ZOR_FLAG_DIRECTORY))
4219 match = B_FALSE;
4220 break;
4221 case DMU_OT_PLAIN_FILE_CONTENTS:
4222 if (!(flags & ZOR_FLAG_PLAIN_FILE))
4223 match = B_FALSE;
4224 break;
4225 case DMU_OT_SPACE_MAP:
4226 if (!(flags & ZOR_FLAG_SPACE_MAP))
4227 match = B_FALSE;
4228 break;
4229 default:
4230 if (strcmp(zdb_ot_name(obj_type), "zap") == 0) {
4231 if (!(flags & ZOR_FLAG_ZAP))
4232 match = B_FALSE;
4233 break;
4234 }
4235
4236 /*
4237 * If all bits except some of the supported flags are
4238 * set, the user combined the all-types flag (A) with
4239 * a negated flag to exclude some types (e.g. A-f to
4240 * show all object types except plain files).
4241 */
4242 if ((flags | ZOR_SUPPORTED_FLAGS) != ZOR_FLAG_ALL_TYPES)
4243 match = B_FALSE;
4244
4245 break;
4246 }
4247
4248 return (match);
4249 }
4250
4251 static void
dump_object(objset_t * os,uint64_t object,int verbosity,boolean_t * print_header,uint64_t * dnode_slots_used,uint64_t flags)4252 dump_object(objset_t *os, uint64_t object, int verbosity,
4253 boolean_t *print_header, uint64_t *dnode_slots_used, uint64_t flags)
4254 {
4255 dmu_buf_t *db = NULL;
4256 dmu_object_info_t doi;
4257 dnode_t *dn;
4258 boolean_t dnode_held = B_FALSE;
4259 void *bonus = NULL;
4260 size_t bsize = 0;
4261 char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
4262 char bonus_size[32];
4263 char aux[50];
4264 int error;
4265
4266 /* make sure nicenum has enough space */
4267 _Static_assert(sizeof (iblk) >= NN_NUMBUF_SZ, "iblk truncated");
4268 _Static_assert(sizeof (dblk) >= NN_NUMBUF_SZ, "dblk truncated");
4269 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, "lsize truncated");
4270 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, "asize truncated");
4271 _Static_assert(sizeof (bonus_size) >= NN_NUMBUF_SZ,
4272 "bonus_size truncated");
4273
4274 if (*print_header) {
4275 (void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n",
4276 "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
4277 "lsize", "%full", "type");
4278 *print_header = 0;
4279 }
4280
4281 if (object == 0) {
4282 dn = DMU_META_DNODE(os);
4283 dmu_object_info_from_dnode(dn, &doi);
4284 } else {
4285 /*
4286 * Encrypted datasets will have sensitive bonus buffers
4287 * encrypted. Therefore we cannot hold the bonus buffer and
4288 * must hold the dnode itself instead.
4289 */
4290 error = dmu_object_info(os, object, &doi);
4291 if (error)
4292 fatal("dmu_object_info() failed, errno %u", error);
4293
4294 if (!key_loaded && os->os_encrypted &&
4295 DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) {
4296 error = dnode_hold(os, object, FTAG, &dn);
4297 if (error)
4298 fatal("dnode_hold() failed, errno %u", error);
4299 dnode_held = B_TRUE;
4300 } else {
4301 error = dmu_bonus_hold(os, object, FTAG, &db);
4302 if (error)
4303 fatal("dmu_bonus_hold(%llu) failed, errno %u",
4304 object, error);
4305 bonus = db->db_data;
4306 bsize = db->db_size;
4307 dn = DB_DNODE((dmu_buf_impl_t *)db);
4308 }
4309 }
4310
4311 /*
4312 * Default to showing all object types if no flags were specified.
4313 */
4314 if (flags != 0 && flags != ZOR_FLAG_ALL_TYPES &&
4315 !match_object_type(doi.doi_type, flags))
4316 goto out;
4317
4318 if (dnode_slots_used)
4319 *dnode_slots_used = doi.doi_dnodesize / DNODE_MIN_SIZE;
4320
4321 zdb_nicenum(doi.doi_metadata_block_size, iblk, sizeof (iblk));
4322 zdb_nicenum(doi.doi_data_block_size, dblk, sizeof (dblk));
4323 zdb_nicenum(doi.doi_max_offset, lsize, sizeof (lsize));
4324 zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize, sizeof (asize));
4325 zdb_nicenum(doi.doi_bonus_size, bonus_size, sizeof (bonus_size));
4326 zdb_nicenum(doi.doi_dnodesize, dnsize, sizeof (dnsize));
4327 (void) snprintf(fill, sizeof (fill), "%6.2f", 100.0 *
4328 doi.doi_fill_count * doi.doi_data_block_size / (object == 0 ?
4329 DNODES_PER_BLOCK : 1) / doi.doi_max_offset);
4330
4331 aux[0] = '\0';
4332
4333 if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) {
4334 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
4335 " (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum));
4336 }
4337
4338 if (doi.doi_compress == ZIO_COMPRESS_INHERIT &&
4339 ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) {
4340 const char *compname = NULL;
4341 if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION,
4342 ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel),
4343 &compname) == 0) {
4344 (void) snprintf(aux + strlen(aux),
4345 sizeof (aux) - strlen(aux), " (Z=inherit=%s)",
4346 compname);
4347 } else {
4348 (void) snprintf(aux + strlen(aux),
4349 sizeof (aux) - strlen(aux),
4350 " (Z=inherit=%s-unknown)",
4351 ZDB_COMPRESS_NAME(os->os_compress));
4352 }
4353 } else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) {
4354 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
4355 " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress));
4356 } else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
4357 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
4358 " (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress));
4359 }
4360
4361 (void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n",
4362 (u_longlong_t)object, doi.doi_indirection, iblk, dblk,
4363 asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux);
4364
4365 if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
4366 (void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n",
4367 "", "", "", "", "", "", bonus_size, "bonus",
4368 zdb_ot_name(doi.doi_bonus_type));
4369 }
4370
4371 if (verbosity >= 4) {
4372 (void) printf("\tdnode flags: %s%s%s%s\n",
4373 (dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ?
4374 "USED_BYTES " : "",
4375 (dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ?
4376 "USERUSED_ACCOUNTED " : "",
4377 (dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ?
4378 "USEROBJUSED_ACCOUNTED " : "",
4379 (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ?
4380 "SPILL_BLKPTR" : "");
4381 (void) printf("\tdnode maxblkid: %llu\n",
4382 (longlong_t)dn->dn_phys->dn_maxblkid);
4383
4384 if (!dnode_held) {
4385 object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os,
4386 object, bonus, bsize);
4387 } else {
4388 (void) printf("\t\t(bonus encrypted)\n");
4389 }
4390
4391 if (key_loaded ||
4392 (!os->os_encrypted || !DMU_OT_IS_ENCRYPTED(doi.doi_type))) {
4393 object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object,
4394 NULL, 0);
4395 } else {
4396 (void) printf("\t\t(object encrypted)\n");
4397 }
4398
4399 *print_header = B_TRUE;
4400 }
4401
4402 if (verbosity >= 5) {
4403 if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
4404 char blkbuf[BP_SPRINTF_LEN];
4405 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
4406 DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE);
4407 (void) printf("\nSpill block: %s\n", blkbuf);
4408 }
4409 dump_indirect(dn);
4410 }
4411
4412 if (verbosity >= 5) {
4413 /*
4414 * Report the list of segments that comprise the object.
4415 */
4416 uint64_t start = 0;
4417 uint64_t end;
4418 uint64_t blkfill = 1;
4419 int minlvl = 1;
4420
4421 if (dn->dn_type == DMU_OT_DNODE) {
4422 minlvl = 0;
4423 blkfill = DNODES_PER_BLOCK;
4424 }
4425
4426 for (;;) {
4427 char segsize[32];
4428 /* make sure nicenum has enough space */
4429 _Static_assert(sizeof (segsize) >= NN_NUMBUF_SZ,
4430 "segsize truncated");
4431 error = dnode_next_offset(dn,
4432 0, &start, minlvl, blkfill, 0);
4433 if (error)
4434 break;
4435 end = start;
4436 error = dnode_next_offset(dn,
4437 DNODE_FIND_HOLE, &end, minlvl, blkfill, 0);
4438 zdb_nicenum(end - start, segsize, sizeof (segsize));
4439 (void) printf("\t\tsegment [%016llx, %016llx)"
4440 " size %5s\n", (u_longlong_t)start,
4441 (u_longlong_t)end, segsize);
4442 if (error)
4443 break;
4444 start = end;
4445 }
4446 }
4447 out:
4448 if (db != NULL)
4449 dmu_buf_rele(db, FTAG);
4450 if (dnode_held)
4451 dnode_rele(dn, FTAG);
4452 }
4453
4454 static void
dump_object_file_layout(objset_t * os,uint64_t object,int verbosity)4455 dump_object_file_layout(objset_t *os, uint64_t object, int verbosity)
4456 {
4457 (void) verbosity;
4458 dmu_buf_t *db = NULL;
4459 dmu_object_info_t doi;
4460 dnode_t *dn;
4461 boolean_t dnode_held = B_FALSE;
4462 int error;
4463
4464 char osname[ZFS_MAX_DATASET_NAME_LEN];
4465 dmu_objset_name(os, osname);
4466 (void) printf("%12s: '%s'\n", "objset", osname);
4467 (void) printf("%12s: %d\n", "object", (int)object);
4468
4469 if (object == 0) {
4470 dn = DMU_META_DNODE(os);
4471 dmu_object_info_from_dnode(dn, &doi);
4472 } else {
4473 /*
4474 * Encrypted datasets will have sensitive bonus buffers
4475 * encrypted. Therefore we cannot hold the bonus buffer and
4476 * must hold the dnode itself instead.
4477 */
4478 error = dmu_object_info(os, object, &doi);
4479 if (error)
4480 fatal("dmu_object_info() failed, errno %u", error);
4481
4482 if (!key_loaded && os->os_encrypted &&
4483 DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) {
4484 error = dnode_hold(os, object, FTAG, &dn);
4485 if (error)
4486 fatal("dnode_hold() failed, errno %u", error);
4487 dnode_held = B_TRUE;
4488 } else {
4489 error = dmu_bonus_hold(os, object, FTAG, &db);
4490 if (error)
4491 fatal("dmu_bonus_hold(%llu) failed, errno %u",
4492 object, error);
4493 dn = DB_DNODE((dmu_buf_impl_t *)db);
4494 }
4495 }
4496
4497 error = dump_indirect_layout(dn);
4498
4499 if (db != NULL)
4500 dmu_buf_rele(db, FTAG);
4501 if (dnode_held)
4502 dnode_rele(dn, FTAG);
4503
4504 if (error)
4505 zdb_exit(1);
4506 }
4507
4508 static void
count_dir_mos_objects(dsl_dir_t * dd)4509 count_dir_mos_objects(dsl_dir_t *dd)
4510 {
4511 mos_obj_refd(dd->dd_object);
4512 mos_obj_refd(dsl_dir_phys(dd)->dd_child_dir_zapobj);
4513 mos_obj_refd(dsl_dir_phys(dd)->dd_deleg_zapobj);
4514 mos_obj_refd(dsl_dir_phys(dd)->dd_props_zapobj);
4515 mos_obj_refd(dsl_dir_phys(dd)->dd_clones);
4516
4517 /*
4518 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
4519 * Ignore the references after the first one.
4520 */
4521 mos_obj_refd_multiple(dd->dd_crypto_obj);
4522 }
4523
4524 static void
count_ds_mos_objects(dsl_dataset_t * ds)4525 count_ds_mos_objects(dsl_dataset_t *ds)
4526 {
4527 mos_obj_refd(ds->ds_object);
4528 mos_obj_refd(dsl_dataset_phys(ds)->ds_next_clones_obj);
4529 mos_obj_refd(dsl_dataset_phys(ds)->ds_props_obj);
4530 mos_obj_refd(dsl_dataset_phys(ds)->ds_userrefs_obj);
4531 mos_obj_refd(dsl_dataset_phys(ds)->ds_snapnames_zapobj);
4532 mos_obj_refd(ds->ds_bookmarks_obj);
4533
4534 if (!dsl_dataset_is_snapshot(ds)) {
4535 count_dir_mos_objects(ds->ds_dir);
4536 }
4537 }
4538
4539 static const char *const objset_types[DMU_OST_NUMTYPES] = {
4540 "NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
4541
4542 /*
4543 * Parse a string denoting a range of object IDs of the form
4544 * <start>[:<end>[:flags]], and store the results in zor.
4545 * Return 0 on success. On error, return 1 and update the msg
4546 * pointer to point to a descriptive error message.
4547 */
4548 static int
parse_object_range(char * range,zopt_object_range_t * zor,const char ** msg)4549 parse_object_range(char *range, zopt_object_range_t *zor, const char **msg)
4550 {
4551 uint64_t flags = 0;
4552 char *p, *s, *dup, *flagstr, *tmp = NULL;
4553 size_t len;
4554 int i;
4555 int rc = 0;
4556
4557 if (strchr(range, ':') == NULL) {
4558 zor->zor_obj_start = strtoull(range, &p, 0);
4559 if (*p != '\0') {
4560 *msg = "Invalid characters in object ID";
4561 rc = 1;
4562 }
4563 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4564 zor->zor_obj_end = zor->zor_obj_start;
4565 return (rc);
4566 }
4567
4568 if (strchr(range, ':') == range) {
4569 *msg = "Invalid leading colon";
4570 rc = 1;
4571 return (rc);
4572 }
4573
4574 len = strlen(range);
4575 if (range[len - 1] == ':') {
4576 *msg = "Invalid trailing colon";
4577 rc = 1;
4578 return (rc);
4579 }
4580
4581 dup = strdup(range);
4582 s = strtok_r(dup, ":", &tmp);
4583 zor->zor_obj_start = strtoull(s, &p, 0);
4584
4585 if (*p != '\0') {
4586 *msg = "Invalid characters in start object ID";
4587 rc = 1;
4588 goto out;
4589 }
4590
4591 s = strtok_r(NULL, ":", &tmp);
4592 zor->zor_obj_end = strtoull(s, &p, 0);
4593
4594 if (*p != '\0') {
4595 *msg = "Invalid characters in end object ID";
4596 rc = 1;
4597 goto out;
4598 }
4599
4600 if (zor->zor_obj_start > zor->zor_obj_end) {
4601 *msg = "Start object ID may not exceed end object ID";
4602 rc = 1;
4603 goto out;
4604 }
4605
4606 s = strtok_r(NULL, ":", &tmp);
4607 if (s == NULL) {
4608 zor->zor_flags = ZOR_FLAG_ALL_TYPES;
4609 goto out;
4610 } else if (strtok_r(NULL, ":", &tmp) != NULL) {
4611 *msg = "Invalid colon-delimited field after flags";
4612 rc = 1;
4613 goto out;
4614 }
4615
4616 flagstr = s;
4617 for (i = 0; flagstr[i]; i++) {
4618 int bit;
4619 boolean_t negation = (flagstr[i] == '-');
4620
4621 if (negation) {
4622 i++;
4623 if (flagstr[i] == '\0') {
4624 *msg = "Invalid trailing negation operator";
4625 rc = 1;
4626 goto out;
4627 }
4628 }
4629 bit = flagbits[(uchar_t)flagstr[i]];
4630 if (bit == 0) {
4631 *msg = "Invalid flag";
4632 rc = 1;
4633 goto out;
4634 }
4635 if (negation)
4636 flags &= ~bit;
4637 else
4638 flags |= bit;
4639 }
4640 zor->zor_flags = flags;
4641
4642 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4643 zor->zor_obj_end = ZDB_MAP_OBJECT_ID(zor->zor_obj_end);
4644
4645 out:
4646 free(dup);
4647 return (rc);
4648 }
4649
4650 static void
dump_objset(objset_t * os)4651 dump_objset(objset_t *os)
4652 {
4653 dmu_objset_stats_t dds = { 0 };
4654 uint64_t object, object_count;
4655 uint64_t refdbytes, usedobjs, scratch;
4656 char numbuf[32];
4657 char blkbuf[BP_SPRINTF_LEN + 20];
4658 char osname[ZFS_MAX_DATASET_NAME_LEN];
4659 const char *type = "UNKNOWN";
4660 int verbosity = dump_opt['d'];
4661 boolean_t print_header;
4662 unsigned i;
4663 int error;
4664 uint64_t total_slots_used = 0;
4665 uint64_t max_slot_used = 0;
4666 uint64_t dnode_slots;
4667 uint64_t obj_start;
4668 uint64_t obj_end;
4669 uint64_t flags;
4670
4671 /* make sure nicenum has enough space */
4672 _Static_assert(sizeof (numbuf) >= NN_NUMBUF_SZ, "numbuf truncated");
4673
4674 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
4675 dmu_objset_fast_stat(os, &dds);
4676 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
4677
4678 print_header = B_TRUE;
4679
4680 if (dds.dds_type < DMU_OST_NUMTYPES)
4681 type = objset_types[dds.dds_type];
4682
4683 if (dds.dds_type == DMU_OST_META) {
4684 dds.dds_creation_txg = TXG_INITIAL;
4685 usedobjs = BP_GET_FILL(os->os_rootbp);
4686 refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)->
4687 dd_used_bytes;
4688 } else {
4689 dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
4690 }
4691
4692 ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
4693
4694 zdb_nicenum(refdbytes, numbuf, sizeof (numbuf));
4695
4696 if (verbosity >= 4 || dump_opt['d']) {
4697 (void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp ");
4698 (void) snprintf_blkptr(blkbuf + strlen(blkbuf),
4699 sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp);
4700 } else {
4701 blkbuf[0] = '\0';
4702 }
4703
4704 dmu_objset_name(os, osname);
4705
4706 (void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
4707 "%s, %llu objects%s%s\n",
4708 osname, type, (u_longlong_t)dmu_objset_id(os),
4709 (u_longlong_t)dds.dds_creation_txg,
4710 numbuf, (u_longlong_t)usedobjs, blkbuf,
4711 (dds.dds_inconsistent) ? " (inconsistent)" : "");
4712
4713 for (i = 0; i < zopt_object_args; i++) {
4714 obj_start = zopt_object_ranges[i].zor_obj_start;
4715 obj_end = zopt_object_ranges[i].zor_obj_end;
4716 flags = zopt_object_ranges[i].zor_flags;
4717
4718 object = obj_start;
4719 if (object == 0 || obj_start == obj_end) {
4720 dump_object(os, object, verbosity,
4721 &print_header, NULL, flags);
4722 } else {
4723 object--;
4724 }
4725
4726 while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) &&
4727 object <= obj_end) {
4728 dump_object(os, object, verbosity,
4729 &print_header, NULL, flags);
4730 }
4731 }
4732
4733 if (zopt_object_args > 0) {
4734 return;
4735 }
4736
4737 if (dump_opt['i'] != 0 || verbosity >= 2)
4738 dump_intent_log(dmu_objset_zil(os));
4739
4740 if (dmu_objset_ds(os) != NULL) {
4741 dsl_dataset_t *ds = dmu_objset_ds(os);
4742 dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
4743 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
4744 !dmu_objset_is_snapshot(os)) {
4745 dump_blkptr_list(&ds->ds_dir->dd_livelist, "Livelist");
4746 if (verify_dd_livelist(os) != 0)
4747 fatal("livelist is incorrect");
4748 }
4749
4750 if (dsl_dataset_remap_deadlist_exists(ds)) {
4751 (void) printf("ds_remap_deadlist:\n");
4752 dump_blkptr_list(&ds->ds_remap_deadlist, "Deadlist");
4753 }
4754 count_ds_mos_objects(ds);
4755 }
4756
4757 if (dmu_objset_ds(os) != NULL)
4758 dump_bookmarks(os, verbosity);
4759
4760 if (verbosity < 2)
4761 return;
4762
4763 if (BP_IS_HOLE(os->os_rootbp))
4764 return;
4765
4766 dump_object(os, 0, verbosity, &print_header, NULL, 0);
4767 object_count = 0;
4768 if (DMU_USERUSED_DNODE(os) != NULL &&
4769 DMU_USERUSED_DNODE(os)->dn_type != 0) {
4770 dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header,
4771 NULL, 0);
4772 dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header,
4773 NULL, 0);
4774 }
4775
4776 if (DMU_PROJECTUSED_DNODE(os) != NULL &&
4777 DMU_PROJECTUSED_DNODE(os)->dn_type != 0)
4778 dump_object(os, DMU_PROJECTUSED_OBJECT, verbosity,
4779 &print_header, NULL, 0);
4780
4781 object = 0;
4782 while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
4783 dump_object(os, object, verbosity, &print_header, &dnode_slots,
4784 0);
4785 object_count++;
4786 total_slots_used += dnode_slots;
4787 max_slot_used = object + dnode_slots - 1;
4788 }
4789
4790 (void) printf("\n");
4791
4792 (void) printf(" Dnode slots:\n");
4793 (void) printf("\tTotal used: %10llu\n",
4794 (u_longlong_t)total_slots_used);
4795 (void) printf("\tMax used: %10llu\n",
4796 (u_longlong_t)max_slot_used);
4797 (void) printf("\tPercent empty: %10lf\n",
4798 (double)(max_slot_used - total_slots_used)*100 /
4799 (double)max_slot_used);
4800 (void) printf("\n");
4801
4802 if (error != ESRCH) {
4803 (void) fprintf(stderr, "dmu_object_next() = %d\n", error);
4804 abort();
4805 }
4806
4807 ASSERT3U(object_count, ==, usedobjs);
4808
4809 if (leaked_objects != 0) {
4810 (void) printf("%d potentially leaked objects detected\n",
4811 leaked_objects);
4812 leaked_objects = 0;
4813 }
4814 }
4815
4816 static void
dump_file_data_layout(objset_t * os)4817 dump_file_data_layout(objset_t *os)
4818 {
4819 uint64_t object;
4820 int verbosity = dump_opt['d'];
4821 unsigned i;
4822 uint64_t obj_start;
4823 uint64_t obj_end;
4824
4825 for (i = 0; i < zopt_object_args; i++) {
4826 obj_start = zopt_object_ranges[i].zor_obj_start;
4827 obj_end = zopt_object_ranges[i].zor_obj_end;
4828
4829 object = obj_start;
4830 if (object == 0 || obj_start == obj_end)
4831 dump_object_file_layout(os, object, verbosity);
4832 else
4833 object--;
4834
4835 while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) &&
4836 object <= obj_end) {
4837 dump_object_file_layout(os, object, verbosity);
4838 }
4839 }
4840 }
4841
4842 static void
dump_uberblock(uberblock_t * ub,const char * header,const char * footer)4843 dump_uberblock(uberblock_t *ub, const char *header, const char *footer)
4844 {
4845 time_t timestamp = ub->ub_timestamp;
4846
4847 (void) printf("%s", header ? header : "");
4848 (void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic);
4849 (void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version);
4850 (void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg);
4851 (void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum);
4852 (void) printf("\ttimestamp = %llu UTC = %s",
4853 (u_longlong_t)ub->ub_timestamp, ctime(×tamp));
4854
4855 char blkbuf[BP_SPRINTF_LEN];
4856 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4857 (void) printf("\tbp = %s\n", blkbuf);
4858
4859 (void) printf("\tmmp_magic = %016llx\n",
4860 (u_longlong_t)ub->ub_mmp_magic);
4861 if (MMP_VALID(ub)) {
4862 (void) printf("\tmmp_delay = %0llu\n",
4863 (u_longlong_t)ub->ub_mmp_delay);
4864 if (MMP_SEQ_VALID(ub))
4865 (void) printf("\tmmp_seq = %u\n",
4866 (unsigned int) MMP_SEQ(ub));
4867 if (MMP_FAIL_INT_VALID(ub))
4868 (void) printf("\tmmp_fail = %u\n",
4869 (unsigned int) MMP_FAIL_INT(ub));
4870 if (MMP_INTERVAL_VALID(ub))
4871 (void) printf("\tmmp_write = %u\n",
4872 (unsigned int) MMP_INTERVAL(ub));
4873 /* After MMP_* to make summarize_uberblock_mmp cleaner */
4874 (void) printf("\tmmp_valid = %x\n",
4875 (unsigned int) ub->ub_mmp_config & 0xFF);
4876 }
4877
4878 if (dump_opt['u'] >= 4) {
4879 char blkbuf[BP_SPRINTF_LEN];
4880 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4881 (void) printf("\trootbp = %s\n", blkbuf);
4882 }
4883 (void) printf("\tcheckpoint_txg = %llu\n",
4884 (u_longlong_t)ub->ub_checkpoint_txg);
4885
4886 (void) printf("\traidz_reflow state=%u off=%llu\n",
4887 (int)RRSS_GET_STATE(ub),
4888 (u_longlong_t)RRSS_GET_OFFSET(ub));
4889
4890 (void) printf("%s", footer ? footer : "");
4891 }
4892
4893 static void
dump_config(spa_t * spa)4894 dump_config(spa_t *spa)
4895 {
4896 dmu_buf_t *db;
4897 size_t nvsize = 0;
4898 int error = 0;
4899
4900
4901 error = dmu_bonus_hold(spa->spa_meta_objset,
4902 spa->spa_config_object, FTAG, &db);
4903
4904 if (error == 0) {
4905 nvsize = *(uint64_t *)db->db_data;
4906 dmu_buf_rele(db, FTAG);
4907
4908 (void) printf("\nMOS Configuration:\n");
4909 dump_packed_nvlist(spa->spa_meta_objset,
4910 spa->spa_config_object, (void *)&nvsize, 1);
4911 } else {
4912 (void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d",
4913 (u_longlong_t)spa->spa_config_object, error);
4914 }
4915 }
4916
4917 static void
dump_cachefile(const char * cachefile)4918 dump_cachefile(const char *cachefile)
4919 {
4920 int fd;
4921 struct stat64 statbuf;
4922 char *buf;
4923 nvlist_t *config;
4924
4925 if ((fd = open64(cachefile, O_RDONLY)) < 0) {
4926 (void) printf("cannot open '%s': %s\n", cachefile,
4927 strerror(errno));
4928 zdb_exit(1);
4929 }
4930
4931 if (fstat64(fd, &statbuf) != 0) {
4932 (void) printf("failed to stat '%s': %s\n", cachefile,
4933 strerror(errno));
4934 zdb_exit(1);
4935 }
4936
4937 if ((buf = malloc(statbuf.st_size)) == NULL) {
4938 (void) fprintf(stderr, "failed to allocate %llu bytes\n",
4939 (u_longlong_t)statbuf.st_size);
4940 zdb_exit(1);
4941 }
4942
4943 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
4944 (void) fprintf(stderr, "failed to read %llu bytes\n",
4945 (u_longlong_t)statbuf.st_size);
4946 zdb_exit(1);
4947 }
4948
4949 (void) close(fd);
4950
4951 if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) {
4952 (void) fprintf(stderr, "failed to unpack nvlist\n");
4953 zdb_exit(1);
4954 }
4955
4956 free(buf);
4957
4958 dump_nvlist(config, 0);
4959
4960 nvlist_free(config);
4961 }
4962
4963 /*
4964 * ZFS label nvlist stats
4965 */
4966 typedef struct zdb_nvl_stats {
4967 int zns_list_count;
4968 int zns_leaf_count;
4969 size_t zns_leaf_largest;
4970 size_t zns_leaf_total;
4971 nvlist_t *zns_string;
4972 nvlist_t *zns_uint64;
4973 nvlist_t *zns_boolean;
4974 } zdb_nvl_stats_t;
4975
4976 static void
collect_nvlist_stats(nvlist_t * nvl,zdb_nvl_stats_t * stats)4977 collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats)
4978 {
4979 nvlist_t *list, **array;
4980 nvpair_t *nvp = NULL;
4981 const char *name;
4982 uint_t i, items;
4983
4984 stats->zns_list_count++;
4985
4986 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4987 name = nvpair_name(nvp);
4988
4989 switch (nvpair_type(nvp)) {
4990 case DATA_TYPE_STRING:
4991 fnvlist_add_string(stats->zns_string, name,
4992 fnvpair_value_string(nvp));
4993 break;
4994 case DATA_TYPE_UINT64:
4995 fnvlist_add_uint64(stats->zns_uint64, name,
4996 fnvpair_value_uint64(nvp));
4997 break;
4998 case DATA_TYPE_BOOLEAN:
4999 fnvlist_add_boolean(stats->zns_boolean, name);
5000 break;
5001 case DATA_TYPE_NVLIST:
5002 if (nvpair_value_nvlist(nvp, &list) == 0)
5003 collect_nvlist_stats(list, stats);
5004 break;
5005 case DATA_TYPE_NVLIST_ARRAY:
5006 if (nvpair_value_nvlist_array(nvp, &array, &items) != 0)
5007 break;
5008
5009 for (i = 0; i < items; i++) {
5010 collect_nvlist_stats(array[i], stats);
5011
5012 /* collect stats on leaf vdev */
5013 if (strcmp(name, "children") == 0) {
5014 size_t size;
5015
5016 (void) nvlist_size(array[i], &size,
5017 NV_ENCODE_XDR);
5018 stats->zns_leaf_total += size;
5019 if (size > stats->zns_leaf_largest)
5020 stats->zns_leaf_largest = size;
5021 stats->zns_leaf_count++;
5022 }
5023 }
5024 break;
5025 default:
5026 (void) printf("skip type %d!\n", (int)nvpair_type(nvp));
5027 }
5028 }
5029 }
5030
5031 static void
dump_nvlist_stats(nvlist_t * nvl,size_t cap)5032 dump_nvlist_stats(nvlist_t *nvl, size_t cap)
5033 {
5034 zdb_nvl_stats_t stats = { 0 };
5035 size_t size, sum = 0, total;
5036 size_t noise;
5037
5038 /* requires nvlist with non-unique names for stat collection */
5039 VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0));
5040 VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0));
5041 VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0));
5042 VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR));
5043
5044 (void) printf("\n\nZFS Label NVList Config Stats:\n");
5045
5046 VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR));
5047 (void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n",
5048 (int)total, (int)(cap - total), 100.0 * total / cap);
5049
5050 collect_nvlist_stats(nvl, &stats);
5051
5052 VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR));
5053 size -= noise;
5054 sum += size;
5055 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
5056 (int)fnvlist_num_pairs(stats.zns_uint64),
5057 (int)size, 100.0 * size / total);
5058
5059 VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR));
5060 size -= noise;
5061 sum += size;
5062 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
5063 (int)fnvlist_num_pairs(stats.zns_string),
5064 (int)size, 100.0 * size / total);
5065
5066 VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR));
5067 size -= noise;
5068 sum += size;
5069 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
5070 (int)fnvlist_num_pairs(stats.zns_boolean),
5071 (int)size, 100.0 * size / total);
5072
5073 size = total - sum; /* treat remainder as nvlist overhead */
5074 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
5075 stats.zns_list_count, (int)size, 100.0 * size / total);
5076
5077 if (stats.zns_leaf_count > 0) {
5078 size_t average = stats.zns_leaf_total / stats.zns_leaf_count;
5079
5080 (void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
5081 stats.zns_leaf_count, (int)average);
5082 (void) printf("%24d bytes largest\n",
5083 (int)stats.zns_leaf_largest);
5084
5085 if (dump_opt['l'] >= 3 && average > 0)
5086 (void) printf(" space for %d additional leaf vdevs\n",
5087 (int)((cap - total) / average));
5088 }
5089 (void) printf("\n");
5090
5091 nvlist_free(stats.zns_string);
5092 nvlist_free(stats.zns_uint64);
5093 nvlist_free(stats.zns_boolean);
5094 }
5095
5096 typedef struct cksum_record {
5097 zio_cksum_t cksum;
5098 boolean_t labels[VDEV_LABELS];
5099 avl_node_t link;
5100 } cksum_record_t;
5101
5102 static int
cksum_record_compare(const void * x1,const void * x2)5103 cksum_record_compare(const void *x1, const void *x2)
5104 {
5105 const cksum_record_t *l = (cksum_record_t *)x1;
5106 const cksum_record_t *r = (cksum_record_t *)x2;
5107 int arraysize = ARRAY_SIZE(l->cksum.zc_word);
5108 int difference = 0;
5109
5110 for (int i = 0; i < arraysize; i++) {
5111 difference = TREE_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]);
5112 if (difference)
5113 break;
5114 }
5115
5116 return (difference);
5117 }
5118
5119 static cksum_record_t *
cksum_record_alloc(zio_cksum_t * cksum,int l)5120 cksum_record_alloc(zio_cksum_t *cksum, int l)
5121 {
5122 cksum_record_t *rec;
5123
5124 rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL);
5125 rec->cksum = *cksum;
5126 rec->labels[l] = B_TRUE;
5127
5128 return (rec);
5129 }
5130
5131 static cksum_record_t *
cksum_record_lookup(avl_tree_t * tree,zio_cksum_t * cksum)5132 cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum)
5133 {
5134 cksum_record_t lookup = { .cksum = *cksum };
5135 avl_index_t where;
5136
5137 return (avl_find(tree, &lookup, &where));
5138 }
5139
5140 static cksum_record_t *
cksum_record_insert(avl_tree_t * tree,zio_cksum_t * cksum,int l)5141 cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l)
5142 {
5143 cksum_record_t *rec;
5144
5145 rec = cksum_record_lookup(tree, cksum);
5146 if (rec) {
5147 rec->labels[l] = B_TRUE;
5148 } else {
5149 rec = cksum_record_alloc(cksum, l);
5150 avl_add(tree, rec);
5151 }
5152
5153 return (rec);
5154 }
5155
5156 static int
first_label(cksum_record_t * rec)5157 first_label(cksum_record_t *rec)
5158 {
5159 for (int i = 0; i < VDEV_LABELS; i++)
5160 if (rec->labels[i])
5161 return (i);
5162
5163 return (-1);
5164 }
5165
5166 static void
print_label_numbers(const char * prefix,const cksum_record_t * rec)5167 print_label_numbers(const char *prefix, const cksum_record_t *rec)
5168 {
5169 fputs(prefix, stdout);
5170 for (int i = 0; i < VDEV_LABELS; i++)
5171 if (rec->labels[i] == B_TRUE)
5172 printf("%d ", i);
5173 putchar('\n');
5174 }
5175
5176 #define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
5177
5178 typedef struct zdb_label {
5179 vdev_label_t label;
5180 uint64_t label_offset;
5181 nvlist_t *config_nv;
5182 cksum_record_t *config;
5183 cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT];
5184 boolean_t header_printed;
5185 boolean_t read_failed;
5186 boolean_t cksum_valid;
5187 } zdb_label_t;
5188
5189 static void
print_label_header(zdb_label_t * label,int l)5190 print_label_header(zdb_label_t *label, int l)
5191 {
5192
5193 if (dump_opt['q'])
5194 return;
5195
5196 if (label->header_printed == B_TRUE)
5197 return;
5198
5199 (void) printf("------------------------------------\n");
5200 (void) printf("LABEL %d %s\n", l,
5201 label->cksum_valid ? "" : "(Bad label cksum)");
5202 (void) printf("------------------------------------\n");
5203
5204 label->header_printed = B_TRUE;
5205 }
5206
5207 static void
print_l2arc_header(void)5208 print_l2arc_header(void)
5209 {
5210 (void) printf("------------------------------------\n");
5211 (void) printf("L2ARC device header\n");
5212 (void) printf("------------------------------------\n");
5213 }
5214
5215 static void
print_l2arc_log_blocks(void)5216 print_l2arc_log_blocks(void)
5217 {
5218 (void) printf("------------------------------------\n");
5219 (void) printf("L2ARC device log blocks\n");
5220 (void) printf("------------------------------------\n");
5221 }
5222
5223 static void
dump_l2arc_log_entries(uint64_t log_entries,l2arc_log_ent_phys_t * le,uint64_t i)5224 dump_l2arc_log_entries(uint64_t log_entries,
5225 l2arc_log_ent_phys_t *le, uint64_t i)
5226 {
5227 for (int j = 0; j < log_entries; j++) {
5228 dva_t dva = le[j].le_dva;
5229 (void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
5230 "vdev: %llu, offset: %llu\n",
5231 (u_longlong_t)i, j + 1,
5232 (u_longlong_t)DVA_GET_ASIZE(&dva),
5233 (u_longlong_t)DVA_GET_VDEV(&dva),
5234 (u_longlong_t)DVA_GET_OFFSET(&dva));
5235 (void) printf("|\t\t\t\tbirth: %llu\n",
5236 (u_longlong_t)le[j].le_birth);
5237 (void) printf("|\t\t\t\tlsize: %llu\n",
5238 (u_longlong_t)L2BLK_GET_LSIZE((&le[j])->le_prop));
5239 (void) printf("|\t\t\t\tpsize: %llu\n",
5240 (u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop));
5241 (void) printf("|\t\t\t\tcompr: %llu\n",
5242 (u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop));
5243 (void) printf("|\t\t\t\tcomplevel: %llu\n",
5244 (u_longlong_t)(&le[j])->le_complevel);
5245 (void) printf("|\t\t\t\ttype: %llu\n",
5246 (u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop));
5247 (void) printf("|\t\t\t\tprotected: %llu\n",
5248 (u_longlong_t)L2BLK_GET_PROTECTED((&le[j])->le_prop));
5249 (void) printf("|\t\t\t\tprefetch: %llu\n",
5250 (u_longlong_t)L2BLK_GET_PREFETCH((&le[j])->le_prop));
5251 (void) printf("|\t\t\t\taddress: %llu\n",
5252 (u_longlong_t)le[j].le_daddr);
5253 (void) printf("|\t\t\t\tARC state: %llu\n",
5254 (u_longlong_t)L2BLK_GET_STATE((&le[j])->le_prop));
5255 (void) printf("|\n");
5256 }
5257 (void) printf("\n");
5258 }
5259
5260 static void
dump_l2arc_log_blkptr(const l2arc_log_blkptr_t * lbps)5261 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t *lbps)
5262 {
5263 (void) printf("|\t\tdaddr: %llu\n", (u_longlong_t)lbps->lbp_daddr);
5264 (void) printf("|\t\tpayload_asize: %llu\n",
5265 (u_longlong_t)lbps->lbp_payload_asize);
5266 (void) printf("|\t\tpayload_start: %llu\n",
5267 (u_longlong_t)lbps->lbp_payload_start);
5268 (void) printf("|\t\tlsize: %llu\n",
5269 (u_longlong_t)L2BLK_GET_LSIZE(lbps->lbp_prop));
5270 (void) printf("|\t\tasize: %llu\n",
5271 (u_longlong_t)L2BLK_GET_PSIZE(lbps->lbp_prop));
5272 (void) printf("|\t\tcompralgo: %llu\n",
5273 (u_longlong_t)L2BLK_GET_COMPRESS(lbps->lbp_prop));
5274 (void) printf("|\t\tcksumalgo: %llu\n",
5275 (u_longlong_t)L2BLK_GET_CHECKSUM(lbps->lbp_prop));
5276 (void) printf("|\n\n");
5277 }
5278
5279 static void
dump_l2arc_log_blocks(int fd,const l2arc_dev_hdr_phys_t * l2dhdr,l2arc_dev_hdr_phys_t * rebuild)5280 dump_l2arc_log_blocks(int fd, const l2arc_dev_hdr_phys_t *l2dhdr,
5281 l2arc_dev_hdr_phys_t *rebuild)
5282 {
5283 l2arc_log_blk_phys_t this_lb;
5284 uint64_t asize;
5285 l2arc_log_blkptr_t lbps[2];
5286 zio_cksum_t cksum;
5287 int failed = 0;
5288 l2arc_dev_t dev;
5289
5290 if (!dump_opt['q'])
5291 print_l2arc_log_blocks();
5292 memcpy(lbps, l2dhdr->dh_start_lbps, sizeof (lbps));
5293
5294 dev.l2ad_evict = l2dhdr->dh_evict;
5295 dev.l2ad_start = l2dhdr->dh_start;
5296 dev.l2ad_end = l2dhdr->dh_end;
5297
5298 if (l2dhdr->dh_start_lbps[0].lbp_daddr == 0) {
5299 /* no log blocks to read */
5300 if (!dump_opt['q']) {
5301 (void) printf("No log blocks to read\n");
5302 (void) printf("\n");
5303 }
5304 return;
5305 } else {
5306 dev.l2ad_hand = lbps[0].lbp_daddr +
5307 L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
5308 }
5309
5310 dev.l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST);
5311
5312 for (;;) {
5313 if (!l2arc_log_blkptr_valid(&dev, &lbps[0]))
5314 break;
5315
5316 /* L2BLK_GET_PSIZE returns aligned size for log blocks */
5317 asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
5318 if (pread64(fd, &this_lb, asize, lbps[0].lbp_daddr) != asize) {
5319 if (!dump_opt['q']) {
5320 (void) printf("Error while reading next log "
5321 "block\n\n");
5322 }
5323 break;
5324 }
5325
5326 fletcher_4_native_varsize(&this_lb, asize, &cksum);
5327 if (!ZIO_CHECKSUM_EQUAL(cksum, lbps[0].lbp_cksum)) {
5328 failed++;
5329 if (!dump_opt['q']) {
5330 (void) printf("Invalid cksum\n");
5331 dump_l2arc_log_blkptr(&lbps[0]);
5332 }
5333 break;
5334 }
5335
5336 switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) {
5337 case ZIO_COMPRESS_OFF:
5338 break;
5339 default: {
5340 abd_t *abd = abd_alloc_linear(asize, B_TRUE);
5341 abd_copy_from_buf_off(abd, &this_lb, 0, asize);
5342 abd_t dabd;
5343 abd_get_from_buf_struct(&dabd, &this_lb,
5344 sizeof (this_lb));
5345 int err = zio_decompress_data(L2BLK_GET_COMPRESS(
5346 (&lbps[0])->lbp_prop), abd, &dabd,
5347 asize, sizeof (this_lb), NULL);
5348 abd_free(&dabd);
5349 abd_free(abd);
5350 if (err != 0) {
5351 (void) printf("L2ARC block decompression "
5352 "failed\n");
5353 goto out;
5354 }
5355 break;
5356 }
5357 }
5358
5359 if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
5360 byteswap_uint64_array(&this_lb, sizeof (this_lb));
5361 if (this_lb.lb_magic != L2ARC_LOG_BLK_MAGIC) {
5362 if (!dump_opt['q'])
5363 (void) printf("Invalid log block magic\n\n");
5364 break;
5365 }
5366
5367 rebuild->dh_lb_count++;
5368 rebuild->dh_lb_asize += asize;
5369 if (dump_opt['l'] > 1 && !dump_opt['q']) {
5370 (void) printf("lb[%4llu]\tmagic: %llu\n",
5371 (u_longlong_t)rebuild->dh_lb_count,
5372 (u_longlong_t)this_lb.lb_magic);
5373 dump_l2arc_log_blkptr(&lbps[0]);
5374 }
5375
5376 if (dump_opt['l'] > 2 && !dump_opt['q'])
5377 dump_l2arc_log_entries(l2dhdr->dh_log_entries,
5378 this_lb.lb_entries,
5379 rebuild->dh_lb_count);
5380
5381 if (l2arc_range_check_overlap(lbps[1].lbp_payload_start,
5382 lbps[0].lbp_payload_start, dev.l2ad_evict) &&
5383 !dev.l2ad_first)
5384 break;
5385
5386 lbps[0] = lbps[1];
5387 lbps[1] = this_lb.lb_prev_lbp;
5388 }
5389 out:
5390 if (!dump_opt['q']) {
5391 (void) printf("log_blk_count:\t %llu with valid cksum\n",
5392 (u_longlong_t)rebuild->dh_lb_count);
5393 (void) printf("\t\t %d with invalid cksum\n", failed);
5394 (void) printf("log_blk_asize:\t %llu\n\n",
5395 (u_longlong_t)rebuild->dh_lb_asize);
5396 }
5397 }
5398
5399 static int
dump_l2arc_header(int fd)5400 dump_l2arc_header(int fd)
5401 {
5402 l2arc_dev_hdr_phys_t l2dhdr = {0}, rebuild = {0};
5403 int error = B_FALSE;
5404
5405 if (pread64(fd, &l2dhdr, sizeof (l2dhdr),
5406 VDEV_LABEL_START_SIZE) != sizeof (l2dhdr)) {
5407 error = B_TRUE;
5408 } else {
5409 if (l2dhdr.dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC))
5410 byteswap_uint64_array(&l2dhdr, sizeof (l2dhdr));
5411
5412 if (l2dhdr.dh_magic != L2ARC_DEV_HDR_MAGIC)
5413 error = B_TRUE;
5414 }
5415
5416 if (error) {
5417 (void) printf("L2ARC device header not found\n\n");
5418 /* Do not return an error here for backward compatibility */
5419 return (0);
5420 } else if (!dump_opt['q']) {
5421 print_l2arc_header();
5422
5423 (void) printf(" magic: %llu\n",
5424 (u_longlong_t)l2dhdr.dh_magic);
5425 (void) printf(" version: %llu\n",
5426 (u_longlong_t)l2dhdr.dh_version);
5427 (void) printf(" pool_guid: %llu\n",
5428 (u_longlong_t)l2dhdr.dh_spa_guid);
5429 (void) printf(" flags: %llu\n",
5430 (u_longlong_t)l2dhdr.dh_flags);
5431 (void) printf(" start_lbps[0]: %llu\n",
5432 (u_longlong_t)
5433 l2dhdr.dh_start_lbps[0].lbp_daddr);
5434 (void) printf(" start_lbps[1]: %llu\n",
5435 (u_longlong_t)
5436 l2dhdr.dh_start_lbps[1].lbp_daddr);
5437 (void) printf(" log_blk_ent: %llu\n",
5438 (u_longlong_t)l2dhdr.dh_log_entries);
5439 (void) printf(" start: %llu\n",
5440 (u_longlong_t)l2dhdr.dh_start);
5441 (void) printf(" end: %llu\n",
5442 (u_longlong_t)l2dhdr.dh_end);
5443 (void) printf(" evict: %llu\n",
5444 (u_longlong_t)l2dhdr.dh_evict);
5445 (void) printf(" lb_asize_refcount: %llu\n",
5446 (u_longlong_t)l2dhdr.dh_lb_asize);
5447 (void) printf(" lb_count_refcount: %llu\n",
5448 (u_longlong_t)l2dhdr.dh_lb_count);
5449 (void) printf(" trim_action_time: %llu\n",
5450 (u_longlong_t)l2dhdr.dh_trim_action_time);
5451 (void) printf(" trim_state: %llu\n\n",
5452 (u_longlong_t)l2dhdr.dh_trim_state);
5453 }
5454
5455 dump_l2arc_log_blocks(fd, &l2dhdr, &rebuild);
5456 /*
5457 * The total aligned size of log blocks and the number of log blocks
5458 * reported in the header of the device may be less than what zdb
5459 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
5460 * This happens because dump_l2arc_log_blocks() lacks the memory
5461 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
5462 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
5463 * and dh_lb_count will be lower to begin with than what exists on the
5464 * device. This is normal and zdb should not exit with an error. The
5465 * opposite case should never happen though, the values reported in the
5466 * header should never be higher than what dump_l2arc_log_blocks() and
5467 * l2arc_rebuild() report. If this happens there is a leak in the
5468 * accounting of log blocks.
5469 */
5470 if (l2dhdr.dh_lb_asize > rebuild.dh_lb_asize ||
5471 l2dhdr.dh_lb_count > rebuild.dh_lb_count)
5472 return (1);
5473
5474 return (0);
5475 }
5476
5477 static void
dump_config_from_label(zdb_label_t * label,size_t buflen,int l)5478 dump_config_from_label(zdb_label_t *label, size_t buflen, int l)
5479 {
5480 if (dump_opt['q'])
5481 return;
5482
5483 if ((dump_opt['l'] < 3) && (first_label(label->config) != l))
5484 return;
5485
5486 print_label_header(label, l);
5487 dump_nvlist(label->config_nv, 4);
5488 print_label_numbers(" labels = ", label->config);
5489
5490 if (dump_opt['l'] >= 2)
5491 dump_nvlist_stats(label->config_nv, buflen);
5492 }
5493
5494 #define ZDB_MAX_UB_HEADER_SIZE 32
5495
5496 static void
dump_label_uberblocks(zdb_label_t * label,uint64_t ashift,int label_num)5497 dump_label_uberblocks(zdb_label_t *label, uint64_t ashift, int label_num)
5498 {
5499
5500 vdev_t vd;
5501 char header[ZDB_MAX_UB_HEADER_SIZE];
5502
5503 vd.vdev_ashift = ashift;
5504 vd.vdev_top = &vd;
5505
5506 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5507 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5508 uberblock_t *ub = (void *)((char *)&label->label + uoff);
5509 cksum_record_t *rec = label->uberblocks[i];
5510
5511 if (rec == NULL) {
5512 if (dump_opt['u'] >= 2) {
5513 print_label_header(label, label_num);
5514 (void) printf(" Uberblock[%d] invalid\n", i);
5515 }
5516 continue;
5517 }
5518
5519 if ((dump_opt['u'] < 3) && (first_label(rec) != label_num))
5520 continue;
5521
5522 if ((dump_opt['u'] < 4) &&
5523 (ub->ub_mmp_magic == MMP_MAGIC) && ub->ub_mmp_delay &&
5524 (i >= VDEV_UBERBLOCK_COUNT(&vd) - MMP_BLOCKS_PER_LABEL))
5525 continue;
5526
5527 print_label_header(label, label_num);
5528 (void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE,
5529 " Uberblock[%d]\n", i);
5530 dump_uberblock(ub, header, "");
5531 print_label_numbers(" labels = ", rec);
5532 }
5533 }
5534
5535 static char curpath[PATH_MAX];
5536
5537 /*
5538 * Iterate through the path components, recursively passing
5539 * current one's obj and remaining path until we find the obj
5540 * for the last one.
5541 */
5542 static int
dump_path_impl(objset_t * os,uint64_t obj,char * name,uint64_t * retobj)5543 dump_path_impl(objset_t *os, uint64_t obj, char *name, uint64_t *retobj)
5544 {
5545 int err;
5546 boolean_t header = B_TRUE;
5547 uint64_t child_obj;
5548 char *s;
5549 dmu_buf_t *db;
5550 dmu_object_info_t doi;
5551
5552 if ((s = strchr(name, '/')) != NULL)
5553 *s = '\0';
5554 err = zap_lookup(os, obj, name, 8, 1, &child_obj);
5555
5556 (void) strlcat(curpath, name, sizeof (curpath));
5557
5558 if (err != 0) {
5559 (void) fprintf(stderr, "failed to lookup %s: %s\n",
5560 curpath, strerror(err));
5561 return (err);
5562 }
5563
5564 child_obj = ZFS_DIRENT_OBJ(child_obj);
5565 err = sa_buf_hold(os, child_obj, FTAG, &db);
5566 if (err != 0) {
5567 (void) fprintf(stderr,
5568 "failed to get SA dbuf for obj %llu: %s\n",
5569 (u_longlong_t)child_obj, strerror(err));
5570 return (EINVAL);
5571 }
5572 dmu_object_info_from_db(db, &doi);
5573 sa_buf_rele(db, FTAG);
5574
5575 if (doi.doi_bonus_type != DMU_OT_SA &&
5576 doi.doi_bonus_type != DMU_OT_ZNODE) {
5577 (void) fprintf(stderr, "invalid bonus type %d for obj %llu\n",
5578 doi.doi_bonus_type, (u_longlong_t)child_obj);
5579 return (EINVAL);
5580 }
5581
5582 if (dump_opt['v'] > 6) {
5583 (void) printf("obj=%llu %s type=%d bonustype=%d\n",
5584 (u_longlong_t)child_obj, curpath, doi.doi_type,
5585 doi.doi_bonus_type);
5586 }
5587
5588 (void) strlcat(curpath, "/", sizeof (curpath));
5589
5590 switch (doi.doi_type) {
5591 case DMU_OT_DIRECTORY_CONTENTS:
5592 if (s != NULL && *(s + 1) != '\0')
5593 return (dump_path_impl(os, child_obj, s + 1, retobj));
5594 zfs_fallthrough;
5595 case DMU_OT_PLAIN_FILE_CONTENTS:
5596 if (retobj != NULL) {
5597 *retobj = child_obj;
5598 } else {
5599 dump_object(os, child_obj, dump_opt['v'], &header,
5600 NULL, 0);
5601 }
5602 return (0);
5603 default:
5604 (void) fprintf(stderr, "object %llu has non-file/directory "
5605 "type %d\n", (u_longlong_t)obj, doi.doi_type);
5606 break;
5607 }
5608
5609 return (EINVAL);
5610 }
5611
5612 /*
5613 * Dump the blocks for the object specified by path inside the dataset.
5614 */
5615 static int
dump_path(char * ds,char * path,uint64_t * retobj)5616 dump_path(char *ds, char *path, uint64_t *retobj)
5617 {
5618 int err;
5619 objset_t *os;
5620 uint64_t root_obj;
5621
5622 err = open_objset(ds, FTAG, &os);
5623 if (err != 0)
5624 return (err);
5625
5626 err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj);
5627 if (err != 0) {
5628 (void) fprintf(stderr, "can't lookup root znode: %s\n",
5629 strerror(err));
5630 close_objset(os, FTAG);
5631 return (EINVAL);
5632 }
5633
5634 (void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds);
5635
5636 err = dump_path_impl(os, root_obj, path, retobj);
5637
5638 close_objset(os, FTAG);
5639 return (err);
5640 }
5641
5642 static int
dump_backup_bytes(objset_t * os,void * buf,int len,void * arg)5643 dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
5644 {
5645 const char *p = (const char *)buf;
5646 ssize_t nwritten;
5647
5648 (void) os;
5649 (void) arg;
5650
5651 /* Write the data out, handling short writes and signals. */
5652 while ((nwritten = write(STDOUT_FILENO, p, len)) < len) {
5653 if (nwritten < 0) {
5654 if (errno == EINTR)
5655 continue;
5656 return (errno);
5657 }
5658 p += nwritten;
5659 len -= nwritten;
5660 }
5661
5662 return (0);
5663 }
5664
5665 static void
dump_backup(const char * pool,uint64_t objset_id,const char * flagstr)5666 dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
5667 {
5668 boolean_t embed = B_FALSE;
5669 boolean_t large_block = B_FALSE;
5670 boolean_t compress = B_FALSE;
5671 boolean_t raw = B_FALSE;
5672
5673 const char *c;
5674 for (c = flagstr; c != NULL && *c != '\0'; c++) {
5675 switch (*c) {
5676 case 'e':
5677 embed = B_TRUE;
5678 break;
5679 case 'L':
5680 large_block = B_TRUE;
5681 break;
5682 case 'c':
5683 compress = B_TRUE;
5684 break;
5685 case 'w':
5686 raw = B_TRUE;
5687 break;
5688 default:
5689 fprintf(stderr, "dump_backup: invalid flag "
5690 "'%c'\n", *c);
5691 return;
5692 }
5693 }
5694
5695 if (isatty(STDOUT_FILENO)) {
5696 fprintf(stderr, "dump_backup: stream cannot be written "
5697 "to a terminal\n");
5698 return;
5699 }
5700
5701 offset_t off = 0;
5702 dmu_send_outparams_t out = {
5703 .dso_outfunc = dump_backup_bytes,
5704 .dso_dryrun = B_FALSE,
5705 };
5706
5707 int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed,
5708 large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO,
5709 &off, &out);
5710 if (err != 0) {
5711 fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",
5712 strerror(err));
5713 return;
5714 }
5715 }
5716
5717 static int
zdb_copy_object(objset_t * os,uint64_t srcobj,char * destfile)5718 zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile)
5719 {
5720 int err = 0;
5721 uint64_t size, readsize, oursize, offset;
5722 ssize_t writesize;
5723 sa_handle_t *hdl;
5724
5725 (void) printf("Copying object %" PRIu64 " to file %s\n", srcobj,
5726 destfile);
5727
5728 VERIFY3P(os, ==, sa_os);
5729 if ((err = sa_handle_get(os, srcobj, NULL, SA_HDL_PRIVATE, &hdl))) {
5730 (void) printf("Failed to get handle for SA znode\n");
5731 return (err);
5732 }
5733 if ((err = sa_lookup(hdl, sa_attr_table[ZPL_SIZE], &size, 8))) {
5734 (void) sa_handle_destroy(hdl);
5735 return (err);
5736 }
5737 (void) sa_handle_destroy(hdl);
5738
5739 (void) printf("Object %" PRIu64 " is %" PRIu64 " bytes\n", srcobj,
5740 size);
5741 if (size == 0) {
5742 return (EINVAL);
5743 }
5744
5745 int fd = open(destfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
5746 if (fd == -1)
5747 return (errno);
5748 /*
5749 * We cap the size at 1 mebibyte here to prevent
5750 * allocation failures and nigh-infinite printing if the
5751 * object is extremely large.
5752 */
5753 oursize = MIN(size, 1 << 20);
5754 offset = 0;
5755 char *buf = kmem_alloc(oursize, KM_NOSLEEP);
5756 if (buf == NULL) {
5757 (void) close(fd);
5758 return (ENOMEM);
5759 }
5760
5761 while (offset < size) {
5762 readsize = MIN(size - offset, 1 << 20);
5763 err = dmu_read(os, srcobj, offset, readsize, buf, 0);
5764 if (err != 0) {
5765 (void) printf("got error %u from dmu_read\n", err);
5766 kmem_free(buf, oursize);
5767 (void) close(fd);
5768 return (err);
5769 }
5770 if (dump_opt['v'] > 3) {
5771 (void) printf("Read offset=%" PRIu64 " size=%" PRIu64
5772 " error=%d\n", offset, readsize, err);
5773 }
5774
5775 writesize = write(fd, buf, readsize);
5776 if (writesize < 0) {
5777 err = errno;
5778 break;
5779 } else if (writesize != readsize) {
5780 /* Incomplete write */
5781 (void) fprintf(stderr, "Short write, only wrote %llu of"
5782 " %" PRIu64 " bytes, exiting...\n",
5783 (u_longlong_t)writesize, readsize);
5784 break;
5785 }
5786
5787 offset += readsize;
5788 }
5789
5790 (void) close(fd);
5791
5792 if (buf != NULL)
5793 kmem_free(buf, oursize);
5794
5795 return (err);
5796 }
5797
5798 static boolean_t
label_cksum_valid(vdev_label_t * label,uint64_t offset)5799 label_cksum_valid(vdev_label_t *label, uint64_t offset)
5800 {
5801 zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
5802 zio_cksum_t expected_cksum;
5803 zio_cksum_t actual_cksum;
5804 zio_cksum_t verifier;
5805 zio_eck_t *eck;
5806 int byteswap;
5807
5808 void *data = (char *)label + offsetof(vdev_label_t, vl_vdev_phys);
5809 eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1;
5810
5811 offset += offsetof(vdev_label_t, vl_vdev_phys);
5812 ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
5813
5814 byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
5815 if (byteswap)
5816 byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
5817
5818 expected_cksum = eck->zec_cksum;
5819 eck->zec_cksum = verifier;
5820
5821 abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE);
5822 ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum);
5823 abd_free(abd);
5824
5825 if (byteswap)
5826 byteswap_uint64_array(&expected_cksum, sizeof (zio_cksum_t));
5827
5828 if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
5829 return (B_TRUE);
5830
5831 return (B_FALSE);
5832 }
5833
5834 static int
dump_label(const char * dev)5835 dump_label(const char *dev)
5836 {
5837 char path[MAXPATHLEN];
5838 zdb_label_t labels[VDEV_LABELS] = {{{{0}}}};
5839 uint64_t psize, ashift, l2cache;
5840 struct stat64 statbuf;
5841 boolean_t config_found = B_FALSE;
5842 boolean_t error = B_FALSE;
5843 boolean_t read_l2arc_header = B_FALSE;
5844 avl_tree_t config_tree;
5845 avl_tree_t uberblock_tree;
5846 void *node, *cookie;
5847 int fd;
5848
5849 /*
5850 * Check if we were given absolute path and use it as is.
5851 * Otherwise if the provided vdev name doesn't point to a file,
5852 * try prepending expected disk paths and partition numbers.
5853 */
5854 (void) strlcpy(path, dev, sizeof (path));
5855 if (dev[0] != '/' && stat64(path, &statbuf) != 0) {
5856 int error;
5857
5858 error = zfs_resolve_shortname(dev, path, MAXPATHLEN);
5859 if (error == 0 && zfs_dev_is_whole_disk(path)) {
5860 if (zfs_append_partition(path, MAXPATHLEN) == -1)
5861 error = ENOENT;
5862 }
5863
5864 if (error || (stat64(path, &statbuf) != 0)) {
5865 (void) printf("failed to find device %s, try "
5866 "specifying absolute path instead\n", dev);
5867 return (1);
5868 }
5869 }
5870
5871 if ((fd = open64(path, O_RDONLY)) < 0) {
5872 (void) printf("cannot open '%s': %s\n", path, strerror(errno));
5873 zdb_exit(1);
5874 }
5875
5876 if (fstat64_blk(fd, &statbuf) != 0) {
5877 (void) printf("failed to stat '%s': %s\n", path,
5878 strerror(errno));
5879 (void) close(fd);
5880 zdb_exit(1);
5881 }
5882
5883 if (S_ISBLK(statbuf.st_mode) && zfs_dev_flush(fd) != 0)
5884 (void) printf("failed to invalidate cache '%s' : %s\n", path,
5885 strerror(errno));
5886
5887 avl_create(&config_tree, cksum_record_compare,
5888 sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5889 avl_create(&uberblock_tree, cksum_record_compare,
5890 sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5891
5892 psize = statbuf.st_size;
5893 psize = P2ALIGN_TYPED(psize, sizeof (vdev_label_t), uint64_t);
5894 ashift = SPA_MINBLOCKSHIFT;
5895
5896 /*
5897 * 1. Read the label from disk
5898 * 2. Verify label cksum
5899 * 3. Unpack the configuration and insert in config tree.
5900 * 4. Traverse all uberblocks and insert in uberblock tree.
5901 */
5902 for (int l = 0; l < VDEV_LABELS; l++) {
5903 zdb_label_t *label = &labels[l];
5904 char *buf = label->label.vl_vdev_phys.vp_nvlist;
5905 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5906 nvlist_t *config;
5907 cksum_record_t *rec;
5908 zio_cksum_t cksum;
5909 vdev_t vd;
5910
5911 label->label_offset = vdev_label_offset(psize, l, 0);
5912
5913 if (pread64(fd, &label->label, sizeof (label->label),
5914 label->label_offset) != sizeof (label->label)) {
5915 if (!dump_opt['q'])
5916 (void) printf("failed to read label %d\n", l);
5917 label->read_failed = B_TRUE;
5918 error = B_TRUE;
5919 continue;
5920 }
5921
5922 label->read_failed = B_FALSE;
5923 label->cksum_valid = label_cksum_valid(&label->label,
5924 label->label_offset);
5925
5926 if (nvlist_unpack(buf, buflen, &config, 0) == 0) {
5927 nvlist_t *vdev_tree = NULL;
5928 size_t size;
5929
5930 if ((nvlist_lookup_nvlist(config,
5931 ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) ||
5932 (nvlist_lookup_uint64(vdev_tree,
5933 ZPOOL_CONFIG_ASHIFT, &ashift) != 0))
5934 ashift = SPA_MINBLOCKSHIFT;
5935
5936 if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0)
5937 size = buflen;
5938
5939 /* If the device is a cache device read the header. */
5940 if (!read_l2arc_header) {
5941 if (nvlist_lookup_uint64(config,
5942 ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 &&
5943 l2cache == POOL_STATE_L2CACHE) {
5944 read_l2arc_header = B_TRUE;
5945 }
5946 }
5947
5948 fletcher_4_native_varsize(buf, size, &cksum);
5949 rec = cksum_record_insert(&config_tree, &cksum, l);
5950
5951 label->config = rec;
5952 label->config_nv = config;
5953 config_found = B_TRUE;
5954 } else {
5955 error = B_TRUE;
5956 }
5957
5958 vd.vdev_ashift = ashift;
5959 vd.vdev_top = &vd;
5960
5961 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5962 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5963 uberblock_t *ub = (void *)((char *)label + uoff);
5964
5965 if (uberblock_verify(ub))
5966 continue;
5967
5968 fletcher_4_native_varsize(ub, sizeof (*ub), &cksum);
5969 rec = cksum_record_insert(&uberblock_tree, &cksum, l);
5970
5971 label->uberblocks[i] = rec;
5972 }
5973 }
5974
5975 /*
5976 * Dump the label and uberblocks.
5977 */
5978 for (int l = 0; l < VDEV_LABELS; l++) {
5979 zdb_label_t *label = &labels[l];
5980 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5981
5982 if (label->read_failed == B_TRUE)
5983 continue;
5984
5985 if (label->config_nv) {
5986 dump_config_from_label(label, buflen, l);
5987 } else {
5988 if (!dump_opt['q'])
5989 (void) printf("failed to unpack label %d\n", l);
5990 }
5991
5992 if (dump_opt['u'])
5993 dump_label_uberblocks(label, ashift, l);
5994
5995 nvlist_free(label->config_nv);
5996 }
5997
5998 /*
5999 * Dump the L2ARC header, if existent.
6000 */
6001 if (read_l2arc_header)
6002 error |= dump_l2arc_header(fd);
6003
6004 cookie = NULL;
6005 while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL)
6006 umem_free(node, sizeof (cksum_record_t));
6007
6008 cookie = NULL;
6009 while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL)
6010 umem_free(node, sizeof (cksum_record_t));
6011
6012 avl_destroy(&config_tree);
6013 avl_destroy(&uberblock_tree);
6014
6015 (void) close(fd);
6016
6017 return (config_found == B_FALSE ? 2 :
6018 (error == B_TRUE ? 1 : 0));
6019 }
6020
6021 static uint64_t dataset_feature_count[SPA_FEATURES];
6022 static uint64_t global_feature_count[SPA_FEATURES];
6023 static uint64_t remap_deadlist_count = 0;
6024
6025 static int
dump_one_objset(const char * dsname,void * arg)6026 dump_one_objset(const char *dsname, void *arg)
6027 {
6028 (void) arg;
6029 int error;
6030 objset_t *os;
6031 spa_feature_t f;
6032
6033 error = open_objset(dsname, FTAG, &os);
6034 if (error != 0)
6035 return (0);
6036
6037 for (f = 0; f < SPA_FEATURES; f++) {
6038 if (!dsl_dataset_feature_is_active(dmu_objset_ds(os), f))
6039 continue;
6040 ASSERT(spa_feature_table[f].fi_flags &
6041 ZFEATURE_FLAG_PER_DATASET);
6042 dataset_feature_count[f]++;
6043 }
6044
6045 if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os))) {
6046 remap_deadlist_count++;
6047 }
6048
6049 for (dsl_bookmark_node_t *dbn =
6050 avl_first(&dmu_objset_ds(os)->ds_bookmarks); dbn != NULL;
6051 dbn = AVL_NEXT(&dmu_objset_ds(os)->ds_bookmarks, dbn)) {
6052 mos_obj_refd(dbn->dbn_phys.zbm_redaction_obj);
6053 if (dbn->dbn_phys.zbm_redaction_obj != 0) {
6054 global_feature_count[
6055 SPA_FEATURE_REDACTION_BOOKMARKS]++;
6056 objset_t *mos = os->os_spa->spa_meta_objset;
6057 dnode_t *rl;
6058 VERIFY0(dnode_hold(mos,
6059 dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl));
6060 if (rl->dn_have_spill) {
6061 global_feature_count[
6062 SPA_FEATURE_REDACTION_LIST_SPILL]++;
6063 }
6064 }
6065 if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)
6066 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN]++;
6067 }
6068
6069 if (dsl_deadlist_is_open(&dmu_objset_ds(os)->ds_dir->dd_livelist) &&
6070 !dmu_objset_is_snapshot(os)) {
6071 global_feature_count[SPA_FEATURE_LIVELIST]++;
6072 }
6073 dump_objset(os);
6074 close_objset(os, FTAG);
6075 fuid_table_destroy();
6076 return (0);
6077 }
6078
6079 /*
6080 * Block statistics.
6081 */
6082 #define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
6083 typedef struct zdb_blkstats {
6084 uint64_t zb_asize;
6085 uint64_t zb_lsize;
6086 uint64_t zb_psize;
6087 uint64_t zb_count;
6088 uint64_t zb_gangs;
6089 uint64_t zb_ditto_samevdev;
6090 uint64_t zb_ditto_same_ms;
6091 uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
6092 } zdb_blkstats_t;
6093
6094 /*
6095 * Extended object types to report deferred frees and dedup auto-ditto blocks.
6096 */
6097 #define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0)
6098 #define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1)
6099 #define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2)
6100 #define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3)
6101
6102 static const char *zdb_ot_extname[] = {
6103 "deferred free",
6104 "dedup ditto",
6105 "other",
6106 "Total",
6107 };
6108
6109 #define ZB_TOTAL DN_MAX_LEVELS
6110 #define SPA_MAX_FOR_16M (SPA_MAXBLOCKSHIFT+1)
6111
6112 typedef struct zdb_brt_entry {
6113 dva_t zbre_dva;
6114 uint64_t zbre_refcount;
6115 avl_node_t zbre_node;
6116 } zdb_brt_entry_t;
6117
6118 typedef struct zdb_cb {
6119 zdb_blkstats_t zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
6120 uint64_t zcb_removing_size;
6121 uint64_t zcb_checkpoint_size;
6122 uint64_t zcb_dedup_asize;
6123 uint64_t zcb_dedup_blocks;
6124 uint64_t zcb_clone_asize;
6125 uint64_t zcb_clone_blocks;
6126 uint64_t zcb_psize_count[SPA_MAX_FOR_16M];
6127 uint64_t zcb_lsize_count[SPA_MAX_FOR_16M];
6128 uint64_t zcb_asize_count[SPA_MAX_FOR_16M];
6129 uint64_t zcb_psize_len[SPA_MAX_FOR_16M];
6130 uint64_t zcb_lsize_len[SPA_MAX_FOR_16M];
6131 uint64_t zcb_asize_len[SPA_MAX_FOR_16M];
6132 uint64_t zcb_psize_total;
6133 uint64_t zcb_lsize_total;
6134 uint64_t zcb_asize_total;
6135 uint64_t zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
6136 uint64_t zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
6137 [BPE_PAYLOAD_SIZE + 1];
6138 uint64_t zcb_start;
6139 hrtime_t zcb_lastprint;
6140 uint64_t zcb_totalasize;
6141 uint64_t zcb_errors[256];
6142 int zcb_readfails;
6143 int zcb_haderrors;
6144 spa_t *zcb_spa;
6145 uint32_t **zcb_vd_obsolete_counts;
6146 avl_tree_t zcb_brt;
6147 boolean_t zcb_brt_is_active;
6148 } zdb_cb_t;
6149
6150 /* test if two DVA offsets from same vdev are within the same metaslab */
6151 static boolean_t
same_metaslab(spa_t * spa,uint64_t vdev,uint64_t off1,uint64_t off2)6152 same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2)
6153 {
6154 vdev_t *vd = vdev_lookup_top(spa, vdev);
6155 uint64_t ms_shift = vd->vdev_ms_shift;
6156
6157 return ((off1 >> ms_shift) == (off2 >> ms_shift));
6158 }
6159
6160 /*
6161 * Used to simplify reporting of the histogram data.
6162 */
6163 typedef struct one_histo {
6164 const char *name;
6165 uint64_t *count;
6166 uint64_t *len;
6167 uint64_t cumulative;
6168 } one_histo_t;
6169
6170 /*
6171 * The number of separate histograms processed for psize, lsize and asize.
6172 */
6173 #define NUM_HISTO 3
6174
6175 /*
6176 * This routine will create a fixed column size output of three different
6177 * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
6178 * the count, length and cumulative length of the psize, lsize and
6179 * asize blocks.
6180 *
6181 * All three types of blocks are listed on a single line
6182 *
6183 * By default the table is printed in nicenumber format (e.g. 123K) but
6184 * if the '-P' parameter is specified then the full raw number (parseable)
6185 * is printed out.
6186 */
6187 static void
dump_size_histograms(zdb_cb_t * zcb)6188 dump_size_histograms(zdb_cb_t *zcb)
6189 {
6190 /*
6191 * A temporary buffer that allows us to convert a number into
6192 * a string using zdb_nicenumber to allow either raw or human
6193 * readable numbers to be output.
6194 */
6195 char numbuf[32];
6196
6197 /*
6198 * Define titles which are used in the headers of the tables
6199 * printed by this routine.
6200 */
6201 const char blocksize_title1[] = "block";
6202 const char blocksize_title2[] = "size";
6203 const char count_title[] = "Count";
6204 const char length_title[] = "Size";
6205 const char cumulative_title[] = "Cum.";
6206
6207 /*
6208 * Setup the histogram arrays (psize, lsize, and asize).
6209 */
6210 one_histo_t parm_histo[NUM_HISTO];
6211
6212 parm_histo[0].name = "psize";
6213 parm_histo[0].count = zcb->zcb_psize_count;
6214 parm_histo[0].len = zcb->zcb_psize_len;
6215 parm_histo[0].cumulative = 0;
6216
6217 parm_histo[1].name = "lsize";
6218 parm_histo[1].count = zcb->zcb_lsize_count;
6219 parm_histo[1].len = zcb->zcb_lsize_len;
6220 parm_histo[1].cumulative = 0;
6221
6222 parm_histo[2].name = "asize";
6223 parm_histo[2].count = zcb->zcb_asize_count;
6224 parm_histo[2].len = zcb->zcb_asize_len;
6225 parm_histo[2].cumulative = 0;
6226
6227
6228 (void) printf("\nBlock Size Histogram\n");
6229 switch (block_bin_mode) {
6230 case BIN_PSIZE:
6231 printf("(note: all categories are binned by %s)\n", "psize");
6232 break;
6233 case BIN_LSIZE:
6234 printf("(note: all categories are binned by %s)\n", "lsize");
6235 break;
6236 case BIN_ASIZE:
6237 printf("(note: all categories are binned by %s)\n", "asize");
6238 break;
6239 default:
6240 printf("(note: all categories are binned separately)\n");
6241 break;
6242 }
6243 if (block_classes != 0) {
6244 char buf[256] = "";
6245 if (block_classes & CLASS_NORMAL)
6246 strlcat(buf, "\"normal\", ", sizeof (buf));
6247 if (block_classes & CLASS_SPECIAL)
6248 strlcat(buf, "\"special\", ", sizeof (buf));
6249 if (block_classes & CLASS_DEDUP)
6250 strlcat(buf, "\"dedup\", ", sizeof (buf));
6251 if (block_classes & CLASS_OTHER)
6252 strlcat(buf, "\"other\", ", sizeof (buf));
6253 buf[strlen(buf)-2] = '\0';
6254 printf("(note: only blocks in these classes are counted: %s)\n",
6255 buf);
6256 }
6257 /*
6258 * Print the first line titles
6259 */
6260 if (dump_opt['P'])
6261 (void) printf("\n%s\t", blocksize_title1);
6262 else
6263 (void) printf("\n%7s ", blocksize_title1);
6264
6265 for (int j = 0; j < NUM_HISTO; j++) {
6266 if (dump_opt['P']) {
6267 if (j < NUM_HISTO - 1) {
6268 (void) printf("%s\t\t\t", parm_histo[j].name);
6269 } else {
6270 /* Don't print trailing spaces */
6271 (void) printf(" %s", parm_histo[j].name);
6272 }
6273 } else {
6274 if (j < NUM_HISTO - 1) {
6275 /* Left aligned strings in the output */
6276 (void) printf("%-7s ",
6277 parm_histo[j].name);
6278 } else {
6279 /* Don't print trailing spaces */
6280 (void) printf("%s", parm_histo[j].name);
6281 }
6282 }
6283 }
6284 (void) printf("\n");
6285
6286 /*
6287 * Print the second line titles
6288 */
6289 if (dump_opt['P']) {
6290 (void) printf("%s\t", blocksize_title2);
6291 } else {
6292 (void) printf("%7s ", blocksize_title2);
6293 }
6294
6295 for (int i = 0; i < NUM_HISTO; i++) {
6296 if (dump_opt['P']) {
6297 (void) printf("%s\t%s\t%s\t",
6298 count_title, length_title, cumulative_title);
6299 } else {
6300 (void) printf("%7s%7s%7s",
6301 count_title, length_title, cumulative_title);
6302 }
6303 }
6304 (void) printf("\n");
6305
6306 /*
6307 * Print the rows
6308 */
6309 for (int i = SPA_MINBLOCKSHIFT; i < SPA_MAX_FOR_16M; i++) {
6310
6311 /*
6312 * Print the first column showing the blocksize
6313 */
6314 zdb_nicenum((1ULL << i), numbuf, sizeof (numbuf));
6315
6316 if (dump_opt['P']) {
6317 printf("%s", numbuf);
6318 } else {
6319 printf("%7s:", numbuf);
6320 }
6321
6322 /*
6323 * Print the remaining set of 3 columns per size:
6324 * for psize, lsize and asize
6325 */
6326 for (int j = 0; j < NUM_HISTO; j++) {
6327 parm_histo[j].cumulative += parm_histo[j].len[i];
6328
6329 zdb_nicenum(parm_histo[j].count[i],
6330 numbuf, sizeof (numbuf));
6331 if (dump_opt['P'])
6332 (void) printf("\t%s", numbuf);
6333 else
6334 (void) printf("%7s", numbuf);
6335
6336 zdb_nicenum(parm_histo[j].len[i],
6337 numbuf, sizeof (numbuf));
6338 if (dump_opt['P'])
6339 (void) printf("\t%s", numbuf);
6340 else
6341 (void) printf("%7s", numbuf);
6342
6343 zdb_nicenum(parm_histo[j].cumulative,
6344 numbuf, sizeof (numbuf));
6345 if (dump_opt['P'])
6346 (void) printf("\t%s", numbuf);
6347 else
6348 (void) printf("%7s", numbuf);
6349 }
6350 (void) printf("\n");
6351 }
6352 }
6353
6354 static void
zdb_count_block(zdb_cb_t * zcb,zilog_t * zilog,const blkptr_t * bp,dmu_object_type_t type)6355 zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
6356 dmu_object_type_t type)
6357 {
6358 int i;
6359 boolean_t claimed = B_FALSE;
6360 boolean_t ddt_block = B_FALSE;
6361 boolean_t brt_block = B_FALSE;
6362
6363 ASSERT(type < ZDB_OT_TOTAL);
6364
6365 if (zilog && zil_bp_tree_add(zilog, bp) != 0)
6366 return;
6367
6368 spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
6369
6370 blkptr_t tempbp;
6371 if (BP_GET_DEDUP(bp)) {
6372 /*
6373 * Dedup'd blocks are special. We need to count them, so we can
6374 * later uncount them when reporting leaked space, and we must
6375 * only claim them once.
6376 *
6377 * We use the existing dedup system to track what we've seen.
6378 * The first time we see a block, we do a ddt_lookup() to see
6379 * if it exists in the DDT. If we're doing leak tracking, we
6380 * claim the block at this time.
6381 *
6382 * Each time we see a block, we reduce the refcount in the
6383 * entry by one, and add to the size and count of dedup'd
6384 * blocks to report at the end.
6385 */
6386
6387 ddt_t *ddt = ddt_select(zcb->zcb_spa, bp);
6388
6389 ddt_enter(ddt);
6390
6391 /*
6392 * Find the block. This will create the entry in memory, but
6393 * we'll know if that happened by its refcount.
6394 */
6395 ddt_entry_t *dde = ddt_lookup(ddt, bp, B_TRUE);
6396
6397 /*
6398 * ddt_lookup() can return NULL when unique entries are pruned
6399 * from the DDT.
6400 */
6401 if (dde == NULL) {
6402 ddt_exit(ddt);
6403 goto ddt_done;
6404 }
6405
6406 /* Get the phys for this variant */
6407 ddt_phys_variant_t v = ddt_phys_select(ddt, dde, bp);
6408
6409 /*
6410 * DDT_PHYS_NONE means the block has the dedup bit set but
6411 * its DVA doesn't match any phys in the entry. This can
6412 * happen when a DVA was evicted from the DDT and re-added
6413 * on a hash collision. The block may still have a BRT entry.
6414 */
6415 if (v == DDT_PHYS_NONE) {
6416 ddt_exit(ddt);
6417 goto ddt_done;
6418 }
6419
6420 ddt_block = B_TRUE;
6421
6422 /*
6423 * This entry may have multiple sets of DVAs. We must claim
6424 * each set the first time we see them in a real block on disk,
6425 * or count them on subsequent occurences. We don't have a
6426 * convenient way to track the first time we see each variant,
6427 * so we repurpose dde_io as a set of "seen" flag bits. We can
6428 * do this safely in zdb because it never writes, so it will
6429 * never have a writing zio for this block in that pointer.
6430 */
6431 boolean_t seen = !!(((uintptr_t)dde->dde_io) & (1 << v));
6432 if (!seen)
6433 dde->dde_io =
6434 (void *)(((uintptr_t)dde->dde_io) | (1 << v));
6435
6436 /*
6437 * Consume a reference. If this variant's refcount is already
6438 * zero, the DDT tracking is exhausted — more filesystem
6439 * references exist than the DDT accounts for.
6440 */
6441 boolean_t ddt_refcnt_exhausted =
6442 (ddt_phys_refcnt(dde->dde_phys, v) == 0);
6443 if (!ddt_refcnt_exhausted)
6444 ddt_phys_decref(dde->dde_phys, v);
6445
6446 /*
6447 * If this entry has a single flat phys, it may have been
6448 * extended with additional DVAs at some time in its life.
6449 * This block might be from before it was fully extended, and
6450 * so have fewer DVAs.
6451 *
6452 * If this is the first time we've seen this block, and we
6453 * claimed it as-is, then we would miss the claim on some
6454 * number of DVAs, which would then be seen as leaked.
6455 *
6456 * In all cases, if we've had fewer DVAs, then the asize would
6457 * be too small, and would lead to the pool apparently using
6458 * more space than allocated.
6459 *
6460 * To handle this, we copy the canonical set of DVAs from the
6461 * entry back to the block pointer before we claim it.
6462 */
6463 if (v == DDT_PHYS_FLAT) {
6464 ASSERT3U(BP_GET_PHYSICAL_BIRTH(bp), ==,
6465 ddt_phys_birth(dde->dde_phys, v));
6466 tempbp = *bp;
6467 ddt_bp_fill(dde->dde_phys, v, &tempbp,
6468 BP_GET_PHYSICAL_BIRTH(bp));
6469 bp = &tempbp;
6470 }
6471
6472 if (seen && !ddt_refcnt_exhausted) {
6473 /*
6474 * The second or later time we see this block,
6475 * it's a duplicate and we count it.
6476 */
6477 zcb->zcb_dedup_asize += BP_GET_ASIZE(bp);
6478 zcb->zcb_dedup_blocks++;
6479 claimed = B_TRUE;
6480 }
6481
6482 ddt_exit(ddt);
6483 }
6484
6485 ddt_done:
6486 if (!claimed && zcb->zcb_brt_is_active &&
6487 brt_maybe_exists(zcb->zcb_spa, bp)) {
6488 /*
6489 * Cloned blocks are special. We need to count them, so we can
6490 * later uncount them when reporting leaked space, and we must
6491 * only claim them once.
6492 *
6493 * To do this, we keep our own in-memory BRT. For each block
6494 * we haven't seen before, we look it up in the real BRT. If
6495 * we see the block again, we count it as a clone.
6496 */
6497 zdb_brt_entry_t zbre_search, *zbre;
6498 avl_index_t where;
6499
6500 zbre_search.zbre_dva = bp->blk_dva[0];
6501 zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where);
6502 if (zbre == NULL) {
6503 uint64_t refcnt =
6504 brt_entry_get_refcount(zcb->zcb_spa, bp);
6505 if (refcnt > 0) {
6506 brt_block = B_TRUE;
6507 zbre = umem_zalloc(sizeof (zdb_brt_entry_t),
6508 UMEM_NOFAIL);
6509 zbre->zbre_dva = bp->blk_dva[0];
6510 zbre->zbre_refcount = refcnt;
6511 avl_insert(&zcb->zcb_brt, zbre, where);
6512 }
6513 } else {
6514 brt_block = B_TRUE;
6515 if (zbre->zbre_refcount > 0) {
6516 zcb->zcb_clone_asize += BP_GET_ASIZE(bp);
6517 zcb->zcb_clone_blocks++;
6518 zbre->zbre_refcount--;
6519 claimed = B_TRUE;
6520 }
6521 }
6522 }
6523
6524 for (i = 0; i < 4; i++) {
6525 int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
6526 int t = (i & 1) ? type : ZDB_OT_TOTAL;
6527 int equal;
6528 zdb_blkstats_t *zb = &zcb->zcb_type[l][t];
6529
6530 zb->zb_asize += BP_GET_ASIZE(bp);
6531 zb->zb_lsize += BP_GET_LSIZE(bp);
6532 zb->zb_psize += BP_GET_PSIZE(bp);
6533 zb->zb_count++;
6534
6535 /*
6536 * The histogram is only big enough to record blocks up to
6537 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
6538 * "other", bucket.
6539 */
6540 unsigned idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT;
6541 idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1);
6542 zb->zb_psize_histogram[idx]++;
6543
6544 zb->zb_gangs += BP_COUNT_GANG(bp);
6545
6546 switch (BP_GET_NDVAS(bp)) {
6547 case 2:
6548 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6549 DVA_GET_VDEV(&bp->blk_dva[1])) {
6550 zb->zb_ditto_samevdev++;
6551
6552 if (same_metaslab(zcb->zcb_spa,
6553 DVA_GET_VDEV(&bp->blk_dva[0]),
6554 DVA_GET_OFFSET(&bp->blk_dva[0]),
6555 DVA_GET_OFFSET(&bp->blk_dva[1])))
6556 zb->zb_ditto_same_ms++;
6557 }
6558 break;
6559 case 3:
6560 equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6561 DVA_GET_VDEV(&bp->blk_dva[1])) +
6562 (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6563 DVA_GET_VDEV(&bp->blk_dva[2])) +
6564 (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6565 DVA_GET_VDEV(&bp->blk_dva[2]));
6566 if (equal != 0) {
6567 zb->zb_ditto_samevdev++;
6568
6569 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6570 DVA_GET_VDEV(&bp->blk_dva[1]) &&
6571 same_metaslab(zcb->zcb_spa,
6572 DVA_GET_VDEV(&bp->blk_dva[0]),
6573 DVA_GET_OFFSET(&bp->blk_dva[0]),
6574 DVA_GET_OFFSET(&bp->blk_dva[1])))
6575 zb->zb_ditto_same_ms++;
6576 else if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6577 DVA_GET_VDEV(&bp->blk_dva[2]) &&
6578 same_metaslab(zcb->zcb_spa,
6579 DVA_GET_VDEV(&bp->blk_dva[0]),
6580 DVA_GET_OFFSET(&bp->blk_dva[0]),
6581 DVA_GET_OFFSET(&bp->blk_dva[2])))
6582 zb->zb_ditto_same_ms++;
6583 else if (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6584 DVA_GET_VDEV(&bp->blk_dva[2]) &&
6585 same_metaslab(zcb->zcb_spa,
6586 DVA_GET_VDEV(&bp->blk_dva[1]),
6587 DVA_GET_OFFSET(&bp->blk_dva[1]),
6588 DVA_GET_OFFSET(&bp->blk_dva[2])))
6589 zb->zb_ditto_same_ms++;
6590 }
6591 break;
6592 }
6593 }
6594
6595 spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
6596
6597 if (BP_IS_EMBEDDED(bp)) {
6598 zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
6599 zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
6600 [BPE_GET_PSIZE(bp)]++;
6601 return;
6602 }
6603
6604 if (block_classes != 0) {
6605 spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
6606
6607 uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[0]);
6608 uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[0]);
6609 vdev_t *vd = vdev_lookup_top(zcb->zcb_spa, vdev);
6610 ASSERT(vd != NULL);
6611 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6612 ASSERT(ms != NULL);
6613 metaslab_group_t *mg = ms->ms_group;
6614 ASSERT(mg != NULL);
6615 metaslab_class_t *mc = mg->mg_class;
6616 ASSERT(mc != NULL);
6617
6618 spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
6619
6620 int class;
6621 if (mc == spa_normal_class(zcb->zcb_spa)) {
6622 class = CLASS_NORMAL;
6623 } else if (mc == spa_special_class(zcb->zcb_spa)) {
6624 class = CLASS_SPECIAL;
6625 } else if (mc == spa_dedup_class(zcb->zcb_spa)) {
6626 class = CLASS_DEDUP;
6627 } else {
6628 class = CLASS_OTHER;
6629 }
6630
6631 if (!(block_classes & class)) {
6632 goto hist_skipped;
6633 }
6634 }
6635
6636 /*
6637 * The binning histogram bins by powers of two up to
6638 * SPA_MAXBLOCKSIZE rather than creating bins for
6639 * every possible blocksize found in the pool.
6640 */
6641 int bin;
6642
6643 /*
6644 * Binning strategy: each bin includes blocks up to and including
6645 * the given size (excluding blocks that fit into the previous bin).
6646 * This way, the "4K" bin includes blocks within the (2K; 4K] range.
6647 */
6648 #define BIN(size) (highbit64((size) - 1))
6649
6650 switch (block_bin_mode) {
6651 case BIN_PSIZE: bin = BIN(BP_GET_PSIZE(bp)); break;
6652 case BIN_LSIZE: bin = BIN(BP_GET_LSIZE(bp)); break;
6653 case BIN_ASIZE: bin = BIN(BP_GET_ASIZE(bp)); break;
6654 case BIN_AUTO: break;
6655 default: PANIC("bad block_bin_mode"); abort();
6656 }
6657
6658 if (block_bin_mode == BIN_AUTO)
6659 bin = BIN(BP_GET_PSIZE(bp));
6660
6661 zcb->zcb_psize_count[bin]++;
6662 zcb->zcb_psize_len[bin] += BP_GET_PSIZE(bp);
6663 zcb->zcb_psize_total += BP_GET_PSIZE(bp);
6664
6665 if (block_bin_mode == BIN_AUTO)
6666 bin = BIN(BP_GET_LSIZE(bp));
6667
6668 zcb->zcb_lsize_count[bin]++;
6669 zcb->zcb_lsize_len[bin] += BP_GET_LSIZE(bp);
6670 zcb->zcb_lsize_total += BP_GET_LSIZE(bp);
6671
6672 if (block_bin_mode == BIN_AUTO)
6673 bin = BIN(BP_GET_ASIZE(bp));
6674
6675 zcb->zcb_asize_count[bin]++;
6676 zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp);
6677 zcb->zcb_asize_total += BP_GET_ASIZE(bp);
6678
6679 #undef BIN
6680
6681 hist_skipped:
6682 if (claimed || dump_opt['L'])
6683 return;
6684
6685 int claim_err = zio_wait(zio_claim(NULL, zcb->zcb_spa,
6686 spa_min_claim_txg(zcb->zcb_spa), bp, NULL, NULL,
6687 ZIO_FLAG_CANFAIL));
6688 if (claim_err != 0) {
6689 char blkbuf[BP_SPRINTF_LEN];
6690 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6691 (void) printf("block claim error %d%s%s: %s\n",
6692 claim_err, brt_block ? " (BRT)" : "",
6693 ddt_block ? " (DDT)" : "", blkbuf);
6694 zcb->zcb_haderrors = 1;
6695 zcb->zcb_errors[claim_err]++;
6696 }
6697 }
6698
6699 static void
zdb_blkptr_done(zio_t * zio)6700 zdb_blkptr_done(zio_t *zio)
6701 {
6702 spa_t *spa = zio->io_spa;
6703 blkptr_t *bp = zio->io_bp;
6704 int ioerr = zio->io_error;
6705 zdb_cb_t *zcb = zio->io_private;
6706 zbookmark_phys_t *zb = &zio->io_bookmark;
6707
6708 mutex_enter(&spa->spa_scrub_lock);
6709 spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
6710 cv_broadcast(&spa->spa_scrub_io_cv);
6711
6712 if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
6713 char blkbuf[BP_SPRINTF_LEN];
6714
6715 zcb->zcb_haderrors = 1;
6716 zcb->zcb_errors[ioerr]++;
6717
6718 if (dump_opt['b'] >= 2)
6719 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6720 else
6721 blkbuf[0] = '\0';
6722
6723 (void) printf("zdb_blkptr_cb: "
6724 "Got error %d reading "
6725 "<%llu, %llu, %lld, %llx> %s -- skipping\n",
6726 ioerr,
6727 (u_longlong_t)zb->zb_objset,
6728 (u_longlong_t)zb->zb_object,
6729 (u_longlong_t)zb->zb_level,
6730 (u_longlong_t)zb->zb_blkid,
6731 blkbuf);
6732 }
6733 mutex_exit(&spa->spa_scrub_lock);
6734
6735 abd_free(zio->io_abd);
6736 }
6737
6738 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)6739 zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
6740 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
6741 {
6742 zdb_cb_t *zcb = arg;
6743 dmu_object_type_t type;
6744 boolean_t is_metadata;
6745
6746 if (zb->zb_level == ZB_DNODE_LEVEL)
6747 return (0);
6748
6749 if (dump_opt['b'] >= 5 && BP_GET_BIRTH(bp) > 0) {
6750 char blkbuf[BP_SPRINTF_LEN];
6751 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6752 (void) printf("objset %llu object %llu "
6753 "level %lld offset 0x%llx %s\n",
6754 (u_longlong_t)zb->zb_objset,
6755 (u_longlong_t)zb->zb_object,
6756 (longlong_t)zb->zb_level,
6757 (u_longlong_t)blkid2offset(dnp, bp, zb),
6758 blkbuf);
6759 }
6760
6761 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
6762 return (0);
6763
6764 type = BP_GET_TYPE(bp);
6765
6766 zdb_count_block(zcb, zilog, bp,
6767 (type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type);
6768
6769 is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
6770
6771 if (!BP_IS_EMBEDDED(bp) &&
6772 (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
6773 size_t size = BP_GET_PSIZE(bp);
6774 abd_t *abd = abd_alloc(size, B_FALSE);
6775 int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
6776
6777 /* If it's an intent log block, failure is expected. */
6778 if (zb->zb_level == ZB_ZIL_LEVEL)
6779 flags |= ZIO_FLAG_SPECULATIVE;
6780
6781 mutex_enter(&spa->spa_scrub_lock);
6782 while (spa->spa_load_verify_bytes > max_inflight_bytes)
6783 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
6784 spa->spa_load_verify_bytes += size;
6785 mutex_exit(&spa->spa_scrub_lock);
6786
6787 zio_nowait(zio_read(NULL, spa, bp, abd, size,
6788 zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
6789 }
6790
6791 zcb->zcb_readfails = 0;
6792
6793 /* only call gethrtime() every 100 blocks */
6794 static int iters;
6795 if (++iters > 100)
6796 iters = 0;
6797 else
6798 return (0);
6799
6800 if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
6801 uint64_t now = gethrtime();
6802 char buf[10];
6803 uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
6804 uint64_t kb_per_sec =
6805 1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000));
6806 uint64_t sec_remaining =
6807 (zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec;
6808
6809 /* make sure nicenum has enough space */
6810 _Static_assert(sizeof (buf) >= NN_NUMBUF_SZ, "buf truncated");
6811
6812 zfs_nicebytes(bytes, buf, sizeof (buf));
6813 (void) fprintf(stderr,
6814 "\r%5s completed (%4"PRIu64"MB/s) "
6815 "estimated time remaining: "
6816 "%"PRIu64"hr %02"PRIu64"min %02"PRIu64"sec ",
6817 buf, kb_per_sec / 1024,
6818 sec_remaining / 60 / 60,
6819 sec_remaining / 60 % 60,
6820 sec_remaining % 60);
6821
6822 zcb->zcb_lastprint = now;
6823 }
6824
6825 return (0);
6826 }
6827
6828 static void
zdb_leak(void * arg,uint64_t start,uint64_t size)6829 zdb_leak(void *arg, uint64_t start, uint64_t size)
6830 {
6831 vdev_t *vd = arg;
6832
6833 (void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
6834 (u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
6835 }
6836
6837 static metaslab_ops_t zdb_metaslab_ops = {
6838 NULL /* alloc */
6839 };
6840
6841 static int
load_unflushed_svr_segs_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6842 load_unflushed_svr_segs_cb(spa_t *spa, space_map_entry_t *sme,
6843 uint64_t txg, void *arg)
6844 {
6845 spa_vdev_removal_t *svr = arg;
6846
6847 uint64_t offset = sme->sme_offset;
6848 uint64_t size = sme->sme_run;
6849
6850 /* skip vdevs we don't care about */
6851 if (sme->sme_vdev != svr->svr_vdev_id)
6852 return (0);
6853
6854 vdev_t *vd = vdev_lookup_top(spa, sme->sme_vdev);
6855 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6856 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6857
6858 if (txg < metaslab_unflushed_txg(ms))
6859 return (0);
6860
6861 if (sme->sme_type == SM_ALLOC)
6862 zfs_range_tree_add(svr->svr_allocd_segs, offset, size);
6863 else
6864 zfs_range_tree_remove(svr->svr_allocd_segs, offset, size);
6865
6866 return (0);
6867 }
6868
6869 static void
claim_segment_impl_cb(uint64_t inner_offset,vdev_t * vd,uint64_t offset,uint64_t size,void * arg)6870 claim_segment_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
6871 uint64_t size, void *arg)
6872 {
6873 (void) inner_offset, (void) arg;
6874
6875 /*
6876 * This callback was called through a remap from
6877 * a device being removed. Therefore, the vdev that
6878 * this callback is applied to is a concrete
6879 * vdev.
6880 */
6881 ASSERT(vdev_is_concrete(vd));
6882
6883 VERIFY0(metaslab_claim_impl(vd, offset, size,
6884 spa_min_claim_txg(vd->vdev_spa)));
6885 }
6886
6887 static void
claim_segment_cb(void * arg,uint64_t offset,uint64_t size)6888 claim_segment_cb(void *arg, uint64_t offset, uint64_t size)
6889 {
6890 vdev_t *vd = arg;
6891
6892 vdev_indirect_ops.vdev_op_remap(vd, offset, size,
6893 claim_segment_impl_cb, NULL);
6894 }
6895
6896 /*
6897 * After accounting for all allocated blocks that are directly referenced,
6898 * we might have missed a reference to a block from a partially complete
6899 * (and thus unused) indirect mapping object. We perform a secondary pass
6900 * through the metaslabs we have already mapped and claim the destination
6901 * blocks.
6902 */
6903 static void
zdb_claim_removing(spa_t * spa,zdb_cb_t * zcb)6904 zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb)
6905 {
6906 if (dump_opt['L'])
6907 return;
6908
6909 if (spa->spa_vdev_removal == NULL)
6910 return;
6911
6912 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6913
6914 spa_vdev_removal_t *svr = spa->spa_vdev_removal;
6915 vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id);
6916 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6917
6918 ASSERT0(zfs_range_tree_space(svr->svr_allocd_segs));
6919
6920 zfs_range_tree_t *allocs = zfs_range_tree_create_flags(
6921 NULL, ZFS_RANGE_SEG64, NULL, 0, 0,
6922 0, "zdb_claim_removing:allocs");
6923 for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) {
6924 metaslab_t *msp = vd->vdev_ms[msi];
6925
6926 ASSERT0(zfs_range_tree_space(allocs));
6927 if (msp->ms_sm != NULL)
6928 VERIFY0(space_map_load(msp->ms_sm, allocs, SM_ALLOC));
6929 zfs_range_tree_vacate(allocs, zfs_range_tree_add,
6930 svr->svr_allocd_segs);
6931 }
6932 zfs_range_tree_destroy(allocs);
6933
6934 iterate_through_spacemap_logs(spa, load_unflushed_svr_segs_cb, svr);
6935
6936 /*
6937 * Clear everything past what has been synced,
6938 * because we have not allocated mappings for
6939 * it yet.
6940 */
6941 zfs_range_tree_clear(svr->svr_allocd_segs,
6942 vdev_indirect_mapping_max_offset(vim),
6943 vd->vdev_asize - vdev_indirect_mapping_max_offset(vim));
6944
6945 zcb->zcb_removing_size += zfs_range_tree_space(svr->svr_allocd_segs);
6946 zfs_range_tree_vacate(svr->svr_allocd_segs, claim_segment_cb, vd);
6947
6948 spa_config_exit(spa, SCL_CONFIG, FTAG);
6949 }
6950
6951 static int
increment_indirect_mapping_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6952 increment_indirect_mapping_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6953 dmu_tx_t *tx)
6954 {
6955 (void) tx;
6956 zdb_cb_t *zcb = arg;
6957 spa_t *spa = zcb->zcb_spa;
6958 vdev_t *vd;
6959 const dva_t *dva = &bp->blk_dva[0];
6960
6961 ASSERT(!bp_freed);
6962 ASSERT(!dump_opt['L']);
6963 ASSERT3U(BP_GET_NDVAS(bp), ==, 1);
6964
6965 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6966 vd = vdev_lookup_top(zcb->zcb_spa, DVA_GET_VDEV(dva));
6967 ASSERT3P(vd, !=, NULL);
6968 spa_config_exit(spa, SCL_VDEV, FTAG);
6969
6970 ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0);
6971 ASSERT3P(zcb->zcb_vd_obsolete_counts[vd->vdev_id], !=, NULL);
6972
6973 vdev_indirect_mapping_increment_obsolete_count(
6974 vd->vdev_indirect_mapping,
6975 DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva),
6976 zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6977
6978 return (0);
6979 }
6980
6981 static uint32_t *
zdb_load_obsolete_counts(vdev_t * vd)6982 zdb_load_obsolete_counts(vdev_t *vd)
6983 {
6984 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6985 spa_t *spa = vd->vdev_spa;
6986 spa_condensing_indirect_phys_t *scip =
6987 &spa->spa_condensing_indirect_phys;
6988 uint64_t obsolete_sm_object;
6989 uint32_t *counts;
6990
6991 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
6992 EQUIV(obsolete_sm_object != 0, vd->vdev_obsolete_sm != NULL);
6993 counts = vdev_indirect_mapping_load_obsolete_counts(vim);
6994 if (vd->vdev_obsolete_sm != NULL) {
6995 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6996 vd->vdev_obsolete_sm);
6997 }
6998 if (scip->scip_vdev == vd->vdev_id &&
6999 scip->scip_prev_obsolete_sm_object != 0) {
7000 space_map_t *prev_obsolete_sm = NULL;
7001 VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset,
7002 scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0));
7003 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
7004 prev_obsolete_sm);
7005 space_map_close(prev_obsolete_sm);
7006 }
7007 return (counts);
7008 }
7009
7010 typedef struct checkpoint_sm_exclude_entry_arg {
7011 vdev_t *cseea_vd;
7012 uint64_t cseea_checkpoint_size;
7013 } checkpoint_sm_exclude_entry_arg_t;
7014
7015 static int
checkpoint_sm_exclude_entry_cb(space_map_entry_t * sme,void * arg)7016 checkpoint_sm_exclude_entry_cb(space_map_entry_t *sme, void *arg)
7017 {
7018 checkpoint_sm_exclude_entry_arg_t *cseea = arg;
7019 vdev_t *vd = cseea->cseea_vd;
7020 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
7021 uint64_t end = sme->sme_offset + sme->sme_run;
7022
7023 ASSERT(sme->sme_type == SM_FREE);
7024
7025 /*
7026 * Since the vdev_checkpoint_sm exists in the vdev level
7027 * and the ms_sm space maps exist in the metaslab level,
7028 * an entry in the checkpoint space map could theoretically
7029 * cross the boundaries of the metaslab that it belongs.
7030 *
7031 * In reality, because of the way that we populate and
7032 * manipulate the checkpoint's space maps currently,
7033 * there shouldn't be any entries that cross metaslabs.
7034 * Hence the assertion below.
7035 *
7036 * That said, there is no fundamental requirement that
7037 * the checkpoint's space map entries should not cross
7038 * metaslab boundaries. So if needed we could add code
7039 * that handles metaslab-crossing segments in the future.
7040 */
7041 VERIFY3U(sme->sme_offset, >=, ms->ms_start);
7042 VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
7043
7044 /*
7045 * By removing the entry from the allocated segments we
7046 * also verify that the entry is there to begin with.
7047 */
7048 mutex_enter(&ms->ms_lock);
7049 zfs_range_tree_remove(ms->ms_allocatable, sme->sme_offset,
7050 sme->sme_run);
7051 mutex_exit(&ms->ms_lock);
7052
7053 cseea->cseea_checkpoint_size += sme->sme_run;
7054 return (0);
7055 }
7056
7057 static void
zdb_leak_init_vdev_exclude_checkpoint(vdev_t * vd,zdb_cb_t * zcb)7058 zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_cb_t *zcb)
7059 {
7060 spa_t *spa = vd->vdev_spa;
7061 space_map_t *checkpoint_sm = NULL;
7062 uint64_t checkpoint_sm_obj;
7063
7064 /*
7065 * If there is no vdev_top_zap, we are in a pool whose
7066 * version predates the pool checkpoint feature.
7067 */
7068 if (vd->vdev_top_zap == 0)
7069 return;
7070
7071 /*
7072 * If there is no reference of the vdev_checkpoint_sm in
7073 * the vdev_top_zap, then one of the following scenarios
7074 * is true:
7075 *
7076 * 1] There is no checkpoint
7077 * 2] There is a checkpoint, but no checkpointed blocks
7078 * have been freed yet
7079 * 3] The current vdev is indirect
7080 *
7081 * In these cases we return immediately.
7082 */
7083 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
7084 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
7085 return;
7086
7087 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
7088 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1,
7089 &checkpoint_sm_obj));
7090
7091 checkpoint_sm_exclude_entry_arg_t cseea;
7092 cseea.cseea_vd = vd;
7093 cseea.cseea_checkpoint_size = 0;
7094
7095 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
7096 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
7097
7098 VERIFY0(space_map_iterate(checkpoint_sm,
7099 space_map_length(checkpoint_sm),
7100 checkpoint_sm_exclude_entry_cb, &cseea));
7101 space_map_close(checkpoint_sm);
7102
7103 zcb->zcb_checkpoint_size += cseea.cseea_checkpoint_size;
7104 }
7105
7106 static void
zdb_leak_init_exclude_checkpoint(spa_t * spa,zdb_cb_t * zcb)7107 zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb)
7108 {
7109 ASSERT(!dump_opt['L']);
7110
7111 vdev_t *rvd = spa->spa_root_vdev;
7112 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
7113 ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id);
7114 zdb_leak_init_vdev_exclude_checkpoint(rvd->vdev_child[c], zcb);
7115 }
7116 }
7117
7118 static int
count_unflushed_space_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)7119 count_unflushed_space_cb(spa_t *spa, space_map_entry_t *sme,
7120 uint64_t txg, void *arg)
7121 {
7122 int64_t *ualloc_space = arg;
7123
7124 uint64_t offset = sme->sme_offset;
7125 uint64_t vdev_id = sme->sme_vdev;
7126
7127 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
7128 if (!vdev_is_concrete(vd))
7129 return (0);
7130
7131 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
7132 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
7133
7134 if (txg < metaslab_unflushed_txg(ms))
7135 return (0);
7136
7137 if (sme->sme_type == SM_ALLOC)
7138 *ualloc_space += sme->sme_run;
7139 else
7140 *ualloc_space -= sme->sme_run;
7141
7142 return (0);
7143 }
7144
7145 static int64_t
get_unflushed_alloc_space(spa_t * spa)7146 get_unflushed_alloc_space(spa_t *spa)
7147 {
7148 if (dump_opt['L'])
7149 return (0);
7150
7151 int64_t ualloc_space = 0;
7152 iterate_through_spacemap_logs(spa, count_unflushed_space_cb,
7153 &ualloc_space);
7154 return (ualloc_space);
7155 }
7156
7157 static int
load_unflushed_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)7158 load_unflushed_cb(spa_t *spa, space_map_entry_t *sme, uint64_t txg, void *arg)
7159 {
7160 maptype_t *uic_maptype = arg;
7161
7162 uint64_t offset = sme->sme_offset;
7163 uint64_t size = sme->sme_run;
7164 uint64_t vdev_id = sme->sme_vdev;
7165
7166 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
7167
7168 /* skip indirect vdevs */
7169 if (!vdev_is_concrete(vd))
7170 return (0);
7171
7172 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
7173
7174 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
7175 ASSERT(*uic_maptype == SM_ALLOC || *uic_maptype == SM_FREE);
7176
7177 if (txg < metaslab_unflushed_txg(ms))
7178 return (0);
7179
7180 if (*uic_maptype == sme->sme_type)
7181 zfs_range_tree_add(ms->ms_allocatable, offset, size);
7182 else
7183 zfs_range_tree_remove(ms->ms_allocatable, offset, size);
7184
7185 return (0);
7186 }
7187
7188 static void
load_unflushed_to_ms_allocatables(spa_t * spa,maptype_t maptype)7189 load_unflushed_to_ms_allocatables(spa_t *spa, maptype_t maptype)
7190 {
7191 iterate_through_spacemap_logs(spa, load_unflushed_cb, &maptype);
7192 }
7193
7194 static void
load_concrete_ms_allocatable_trees(spa_t * spa,maptype_t maptype)7195 load_concrete_ms_allocatable_trees(spa_t *spa, maptype_t maptype)
7196 {
7197 vdev_t *rvd = spa->spa_root_vdev;
7198 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
7199 vdev_t *vd = rvd->vdev_child[i];
7200
7201 ASSERT3U(i, ==, vd->vdev_id);
7202
7203 if (vd->vdev_ops == &vdev_indirect_ops)
7204 continue;
7205
7206 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
7207 metaslab_t *msp = vd->vdev_ms[m];
7208
7209 (void) fprintf(stderr,
7210 "\rloading concrete vdev %llu, "
7211 "metaslab %llu of %llu ...",
7212 (longlong_t)vd->vdev_id,
7213 (longlong_t)msp->ms_id,
7214 (longlong_t)vd->vdev_ms_count);
7215
7216 mutex_enter(&msp->ms_lock);
7217 zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
7218
7219 /*
7220 * We don't want to spend the CPU manipulating the
7221 * size-ordered tree, so clear the range_tree ops.
7222 */
7223 msp->ms_allocatable->rt_ops = NULL;
7224
7225 if (msp->ms_sm != NULL) {
7226 VERIFY0(space_map_load(msp->ms_sm,
7227 msp->ms_allocatable, maptype));
7228 }
7229 if (!msp->ms_loaded)
7230 msp->ms_loaded = B_TRUE;
7231 mutex_exit(&msp->ms_lock);
7232 }
7233 }
7234
7235 load_unflushed_to_ms_allocatables(spa, maptype);
7236 }
7237
7238 /*
7239 * vm_idxp is an in-out parameter which (for indirect vdevs) is the
7240 * index in vim_entries that has the first entry in this metaslab.
7241 * On return, it will be set to the first entry after this metaslab.
7242 */
7243 static void
load_indirect_ms_allocatable_tree(vdev_t * vd,metaslab_t * msp,uint64_t * vim_idxp)7244 load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab_t *msp,
7245 uint64_t *vim_idxp)
7246 {
7247 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7248
7249 mutex_enter(&msp->ms_lock);
7250 zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
7251
7252 /*
7253 * We don't want to spend the CPU manipulating the
7254 * size-ordered tree, so clear the range_tree ops.
7255 */
7256 msp->ms_allocatable->rt_ops = NULL;
7257
7258 for (; *vim_idxp < vdev_indirect_mapping_num_entries(vim);
7259 (*vim_idxp)++) {
7260 vdev_indirect_mapping_entry_phys_t *vimep =
7261 &vim->vim_entries[*vim_idxp];
7262 uint64_t ent_offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
7263 uint64_t ent_len = DVA_GET_ASIZE(&vimep->vimep_dst);
7264 ASSERT3U(ent_offset, >=, msp->ms_start);
7265 if (ent_offset >= msp->ms_start + msp->ms_size)
7266 break;
7267
7268 /*
7269 * Mappings do not cross metaslab boundaries,
7270 * because we create them by walking the metaslabs.
7271 */
7272 ASSERT3U(ent_offset + ent_len, <=,
7273 msp->ms_start + msp->ms_size);
7274 zfs_range_tree_add(msp->ms_allocatable, ent_offset, ent_len);
7275 }
7276
7277 if (!msp->ms_loaded)
7278 msp->ms_loaded = B_TRUE;
7279 mutex_exit(&msp->ms_lock);
7280 }
7281
7282 static void
zdb_leak_init_prepare_indirect_vdevs(spa_t * spa,zdb_cb_t * zcb)7283 zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb)
7284 {
7285 ASSERT(!dump_opt['L']);
7286
7287 vdev_t *rvd = spa->spa_root_vdev;
7288 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
7289 vdev_t *vd = rvd->vdev_child[c];
7290
7291 ASSERT3U(c, ==, vd->vdev_id);
7292
7293 if (vd->vdev_ops != &vdev_indirect_ops)
7294 continue;
7295
7296 /*
7297 * Note: we don't check for mapping leaks on
7298 * removing vdevs because their ms_allocatable's
7299 * are used to look for leaks in allocated space.
7300 */
7301 zcb->zcb_vd_obsolete_counts[c] = zdb_load_obsolete_counts(vd);
7302
7303 /*
7304 * Normally, indirect vdevs don't have any
7305 * metaslabs. We want to set them up for
7306 * zio_claim().
7307 */
7308 vdev_metaslab_group_create(vd);
7309 VERIFY0(vdev_metaslab_init(vd, 0));
7310
7311 vdev_indirect_mapping_t *vim __maybe_unused =
7312 vd->vdev_indirect_mapping;
7313 uint64_t vim_idx = 0;
7314 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
7315
7316 (void) fprintf(stderr,
7317 "\rloading indirect vdev %llu, "
7318 "metaslab %llu of %llu ...",
7319 (longlong_t)vd->vdev_id,
7320 (longlong_t)vd->vdev_ms[m]->ms_id,
7321 (longlong_t)vd->vdev_ms_count);
7322
7323 load_indirect_ms_allocatable_tree(vd, vd->vdev_ms[m],
7324 &vim_idx);
7325 }
7326 ASSERT3U(vim_idx, ==, vdev_indirect_mapping_num_entries(vim));
7327 }
7328 }
7329
7330 static void
zdb_leak_init(spa_t * spa,zdb_cb_t * zcb)7331 zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
7332 {
7333 zcb->zcb_spa = spa;
7334
7335 if (dump_opt['L'])
7336 return;
7337
7338 dsl_pool_t *dp = spa->spa_dsl_pool;
7339 vdev_t *rvd = spa->spa_root_vdev;
7340
7341 /*
7342 * We are going to be changing the meaning of the metaslab's
7343 * ms_allocatable. Ensure that the allocator doesn't try to
7344 * use the tree.
7345 */
7346 spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
7347 spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
7348 spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops;
7349 spa->spa_special_embedded_log_class->mc_ops = &zdb_metaslab_ops;
7350
7351 zcb->zcb_vd_obsolete_counts =
7352 umem_zalloc(rvd->vdev_children * sizeof (uint32_t *),
7353 UMEM_NOFAIL);
7354
7355 /*
7356 * For leak detection, we overload the ms_allocatable trees
7357 * to contain allocated segments instead of free segments.
7358 * As a result, we can't use the normal metaslab_load/unload
7359 * interfaces.
7360 */
7361 zdb_leak_init_prepare_indirect_vdevs(spa, zcb);
7362 load_concrete_ms_allocatable_trees(spa, SM_ALLOC);
7363
7364 /*
7365 * On load_concrete_ms_allocatable_trees() we loaded all the
7366 * allocated entries from the ms_sm to the ms_allocatable for
7367 * each metaslab. If the pool has a checkpoint or is in the
7368 * middle of discarding a checkpoint, some of these blocks
7369 * may have been freed but their ms_sm may not have been
7370 * updated because they are referenced by the checkpoint. In
7371 * order to avoid false-positives during leak-detection, we
7372 * go through the vdev's checkpoint space map and exclude all
7373 * its entries from their relevant ms_allocatable.
7374 *
7375 * We also aggregate the space held by the checkpoint and add
7376 * it to zcb_checkpoint_size.
7377 *
7378 * Note that at this point we are also verifying that all the
7379 * entries on the checkpoint_sm are marked as allocated in
7380 * the ms_sm of their relevant metaslab.
7381 * [see comment in checkpoint_sm_exclude_entry_cb()]
7382 */
7383 zdb_leak_init_exclude_checkpoint(spa, zcb);
7384 ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa));
7385
7386 /* for cleaner progress output */
7387 (void) fprintf(stderr, "\n");
7388
7389 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
7390 ASSERT(spa_feature_is_enabled(spa,
7391 SPA_FEATURE_DEVICE_REMOVAL));
7392 (void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj,
7393 increment_indirect_mapping_cb, zcb, NULL);
7394 }
7395 }
7396
7397 static boolean_t
zdb_check_for_obsolete_leaks(vdev_t * vd,zdb_cb_t * zcb)7398 zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb)
7399 {
7400 boolean_t leaks = B_FALSE;
7401 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7402 uint64_t total_leaked = 0;
7403 boolean_t are_precise = B_FALSE;
7404
7405 ASSERT(vim != NULL);
7406
7407 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
7408 vdev_indirect_mapping_entry_phys_t *vimep =
7409 &vim->vim_entries[i];
7410 uint64_t obsolete_bytes = 0;
7411 uint64_t offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
7412 metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
7413
7414 /*
7415 * This is not very efficient but it's easy to
7416 * verify correctness.
7417 */
7418 for (uint64_t inner_offset = 0;
7419 inner_offset < DVA_GET_ASIZE(&vimep->vimep_dst);
7420 inner_offset += 1ULL << vd->vdev_ashift) {
7421 if (zfs_range_tree_contains(msp->ms_allocatable,
7422 offset + inner_offset, 1ULL << vd->vdev_ashift)) {
7423 obsolete_bytes += 1ULL << vd->vdev_ashift;
7424 }
7425 }
7426
7427 int64_t bytes_leaked = obsolete_bytes -
7428 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i];
7429 ASSERT3U(DVA_GET_ASIZE(&vimep->vimep_dst), >=,
7430 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]);
7431
7432 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7433 if (bytes_leaked != 0 && (are_precise || dump_opt['d'] >= 5)) {
7434 (void) printf("obsolete indirect mapping count "
7435 "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
7436 (u_longlong_t)vd->vdev_id,
7437 (u_longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
7438 (u_longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
7439 (u_longlong_t)bytes_leaked);
7440 }
7441 total_leaked += ABS(bytes_leaked);
7442 }
7443
7444 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7445 if (!are_precise && total_leaked > 0) {
7446 int pct_leaked = total_leaked * 100 /
7447 vdev_indirect_mapping_bytes_mapped(vim);
7448 (void) printf("cannot verify obsolete indirect mapping "
7449 "counts of vdev %llu because precise feature was not "
7450 "enabled when it was removed: %d%% (%llx bytes) of mapping"
7451 "unreferenced\n",
7452 (u_longlong_t)vd->vdev_id, pct_leaked,
7453 (u_longlong_t)total_leaked);
7454 } else if (total_leaked > 0) {
7455 (void) printf("obsolete indirect mapping count mismatch "
7456 "for vdev %llu -- %llx total bytes mismatched\n",
7457 (u_longlong_t)vd->vdev_id,
7458 (u_longlong_t)total_leaked);
7459 leaks |= B_TRUE;
7460 }
7461
7462 vdev_indirect_mapping_free_obsolete_counts(vim,
7463 zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
7464 zcb->zcb_vd_obsolete_counts[vd->vdev_id] = NULL;
7465
7466 return (leaks);
7467 }
7468
7469 static boolean_t
zdb_leak_fini(spa_t * spa,zdb_cb_t * zcb)7470 zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb)
7471 {
7472 boolean_t leaks = B_FALSE;
7473
7474 /*
7475 * Report leaked BRT entries whose refcount was not fully consumed by
7476 * the traversal.
7477 */
7478 if (zcb->zcb_brt_is_active) {
7479 void *cookie = NULL;
7480 zdb_brt_entry_t *zbre;
7481 while ((zbre = avl_destroy_nodes(
7482 &zcb->zcb_brt, &cookie)) != NULL) {
7483 if (!dump_opt['L'] && zbre->zbre_refcount != 0) {
7484 (void) printf("BRT leak: vdev %llu, "
7485 "offset 0x%llx, refcount %llu\n",
7486 (u_longlong_t)DVA_GET_VDEV(
7487 &zbre->zbre_dva),
7488 (u_longlong_t)DVA_GET_OFFSET(
7489 &zbre->zbre_dva),
7490 (u_longlong_t)zbre->zbre_refcount);
7491 leaks = B_TRUE;
7492 }
7493 umem_free(zbre, sizeof (zdb_brt_entry_t));
7494 }
7495 avl_destroy(&zcb->zcb_brt);
7496 }
7497
7498 if (dump_opt['L'])
7499 return (leaks);
7500
7501 /*
7502 * Report leaked DDT entries whose refcount was not fully consumed by
7503 * the traversal. Entries in the DDT ZAP that were never looked up
7504 * are not detected here.
7505 */
7506 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
7507 ddt_t *ddt = spa->spa_ddt[c];
7508 if (ddt == NULL)
7509 continue;
7510 ddt_enter(ddt);
7511 for (ddt_entry_t *dde = avl_first(&ddt->ddt_tree); dde != NULL;
7512 dde = AVL_NEXT(&ddt->ddt_tree, dde)) {
7513 for (int p = 0; p < DDT_NPHYS(ddt); p++) {
7514 ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p);
7515 uint64_t refcnt = ddt_phys_refcnt(dde->dde_phys,
7516 v);
7517 if (refcnt == 0)
7518 continue;
7519 blkptr_t blk;
7520 char blkbuf[BP_SPRINTF_LEN];
7521 ddt_bp_create(ddt->ddt_checksum, &dde->dde_key,
7522 dde->dde_phys, v, &blk);
7523 snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
7524 (void) printf("DDT leak: refcount %llu %s\n",
7525 (u_longlong_t)refcnt, blkbuf);
7526 leaks = B_TRUE;
7527 }
7528 }
7529 ddt_exit(ddt);
7530 }
7531
7532 vdev_t *rvd = spa->spa_root_vdev;
7533 for (unsigned c = 0; c < rvd->vdev_children; c++) {
7534 vdev_t *vd = rvd->vdev_child[c];
7535
7536 if (zcb->zcb_vd_obsolete_counts[c] != NULL) {
7537 leaks |= zdb_check_for_obsolete_leaks(vd, zcb);
7538 }
7539
7540 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
7541 metaslab_t *msp = vd->vdev_ms[m];
7542 ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class ==
7543 spa_embedded_log_class(spa) ||
7544 msp->ms_group->mg_class ==
7545 spa_special_embedded_log_class(spa)) ?
7546 vd->vdev_log_mg : vd->vdev_mg);
7547
7548 /*
7549 * ms_allocatable has been overloaded
7550 * to contain allocated segments. Now that
7551 * we finished traversing all blocks, any
7552 * block that remains in the ms_allocatable
7553 * represents an allocated block that we
7554 * did not claim during the traversal.
7555 * Claimed blocks would have been removed
7556 * from the ms_allocatable. For indirect
7557 * vdevs, space remaining in the tree
7558 * represents parts of the mapping that are
7559 * not referenced, which is not a bug.
7560 */
7561 if (vd->vdev_ops == &vdev_indirect_ops) {
7562 zfs_range_tree_vacate(msp->ms_allocatable,
7563 NULL, NULL);
7564 } else {
7565 zfs_range_tree_vacate(msp->ms_allocatable,
7566 zdb_leak, vd);
7567 }
7568 if (msp->ms_loaded) {
7569 msp->ms_loaded = B_FALSE;
7570 }
7571 }
7572 }
7573
7574 umem_free(zcb->zcb_vd_obsolete_counts,
7575 rvd->vdev_children * sizeof (uint32_t *));
7576 zcb->zcb_vd_obsolete_counts = NULL;
7577
7578 return (leaks);
7579 }
7580
7581 static int
count_block_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)7582 count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7583 {
7584 (void) tx;
7585 zdb_cb_t *zcb = arg;
7586
7587 if (dump_opt['b'] >= 5) {
7588 char blkbuf[BP_SPRINTF_LEN];
7589 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
7590 (void) printf("[%s] %s\n",
7591 "deferred free", blkbuf);
7592 }
7593 zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED);
7594 return (0);
7595 }
7596
7597 /*
7598 * Iterate over livelists which have been destroyed by the user but
7599 * are still present in the MOS, waiting to be freed
7600 */
7601 static void
iterate_deleted_livelists(spa_t * spa,ll_iter_t func,void * arg)7602 iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg)
7603 {
7604 objset_t *mos = spa->spa_meta_objset;
7605 uint64_t zap_obj;
7606 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
7607 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
7608 if (err == ENOENT)
7609 return;
7610 ASSERT0(err);
7611
7612 zap_cursor_t zc;
7613 zap_attribute_t *attrp = zap_attribute_alloc();
7614 dsl_deadlist_t ll;
7615 /* NULL out os prior to dsl_deadlist_open in case it's garbage */
7616 ll.dl_os = NULL;
7617 for (zap_cursor_init(&zc, mos, zap_obj);
7618 zap_cursor_retrieve(&zc, attrp) == 0;
7619 (void) zap_cursor_advance(&zc)) {
7620 VERIFY0(dsl_deadlist_open(&ll, mos, attrp->za_first_integer));
7621 func(&ll, arg);
7622 dsl_deadlist_close(&ll);
7623 }
7624 zap_cursor_fini(&zc);
7625 zap_attribute_free(attrp);
7626 }
7627
7628 static int
bpobj_count_block_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)7629 bpobj_count_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
7630 dmu_tx_t *tx)
7631 {
7632 ASSERT(!bp_freed);
7633 return (count_block_cb(arg, bp, tx));
7634 }
7635
7636 static int
livelist_entry_count_blocks_cb(void * args,dsl_deadlist_entry_t * dle)7637 livelist_entry_count_blocks_cb(void *args, dsl_deadlist_entry_t *dle)
7638 {
7639 zdb_cb_t *zbc = args;
7640 bplist_t blks;
7641 bplist_create(&blks);
7642 /* determine which blocks have been alloc'd but not freed */
7643 VERIFY0(dsl_process_sub_livelist(&dle->dle_bpobj, &blks, NULL, NULL));
7644 /* count those blocks */
7645 (void) bplist_iterate(&blks, count_block_cb, zbc, NULL);
7646 bplist_destroy(&blks);
7647 return (0);
7648 }
7649
7650 static void
livelist_count_blocks(dsl_deadlist_t * ll,void * arg)7651 livelist_count_blocks(dsl_deadlist_t *ll, void *arg)
7652 {
7653 dsl_deadlist_iterate(ll, livelist_entry_count_blocks_cb, arg);
7654 }
7655
7656 /*
7657 * Count the blocks in the livelists that have been destroyed by the user
7658 * but haven't yet been freed.
7659 */
7660 static void
deleted_livelists_count_blocks(spa_t * spa,zdb_cb_t * zbc)7661 deleted_livelists_count_blocks(spa_t *spa, zdb_cb_t *zbc)
7662 {
7663 iterate_deleted_livelists(spa, livelist_count_blocks, zbc);
7664 }
7665
7666 static void
dump_livelist_cb(dsl_deadlist_t * ll,void * arg)7667 dump_livelist_cb(dsl_deadlist_t *ll, void *arg)
7668 {
7669 ASSERT0P(arg);
7670 global_feature_count[SPA_FEATURE_LIVELIST]++;
7671 dump_blkptr_list(ll, "Deleted Livelist");
7672 dsl_deadlist_iterate(ll, sublivelist_verify_lightweight, NULL);
7673 }
7674
7675 /*
7676 * Print out, register object references to, and increment feature counts for
7677 * livelists that have been destroyed by the user but haven't yet been freed.
7678 */
7679 static void
deleted_livelists_dump_mos(spa_t * spa)7680 deleted_livelists_dump_mos(spa_t *spa)
7681 {
7682 uint64_t zap_obj;
7683 objset_t *mos = spa->spa_meta_objset;
7684 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
7685 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
7686 if (err == ENOENT)
7687 return;
7688 mos_obj_refd(zap_obj);
7689 iterate_deleted_livelists(spa, dump_livelist_cb, NULL);
7690 }
7691
7692 static int
zdb_brt_entry_compare(const void * zcn1,const void * zcn2)7693 zdb_brt_entry_compare(const void *zcn1, const void *zcn2)
7694 {
7695 const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva;
7696 const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva;
7697 int cmp;
7698
7699 cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
7700 if (cmp == 0)
7701 cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2));
7702
7703 return (cmp);
7704 }
7705
7706 static int
dump_block_stats(spa_t * spa)7707 dump_block_stats(spa_t *spa)
7708 {
7709 zdb_cb_t *zcb;
7710 zdb_blkstats_t *zb, *tzb;
7711 uint64_t norm_alloc, norm_space, total_alloc, total_found;
7712 int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7713 TRAVERSE_NO_DECRYPT | TRAVERSE_HARD;
7714 boolean_t leaks = B_FALSE;
7715 int e, c, err;
7716 bp_embedded_type_t i;
7717
7718 ddt_prefetch_all(spa);
7719
7720 zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL);
7721
7722 if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
7723 avl_create(&zcb->zcb_brt, zdb_brt_entry_compare,
7724 sizeof (zdb_brt_entry_t),
7725 offsetof(zdb_brt_entry_t, zbre_node));
7726 zcb->zcb_brt_is_active = B_TRUE;
7727 }
7728
7729 (void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
7730 (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
7731 (dump_opt['c'] == 1) ? "metadata " : "",
7732 dump_opt['c'] ? "checksums " : "",
7733 (dump_opt['c'] && !dump_opt['L']) ? "and verify " : "",
7734 !dump_opt['L'] ? "nothing leaked " : "");
7735
7736 /*
7737 * When leak detection is enabled we load all space maps as SM_ALLOC
7738 * maps, then traverse the pool claiming each block we discover. If
7739 * the pool is perfectly consistent, the segment trees will be empty
7740 * when we're done. Anything left over is a leak; any block we can't
7741 * claim (because it's not part of any space map) is a double
7742 * allocation, reference to a freed block, or an unclaimed log block.
7743 *
7744 * When leak detection is disabled (-L option) we still traverse the
7745 * pool claiming each block we discover, but we skip opening any space
7746 * maps.
7747 */
7748 zdb_leak_init(spa, zcb);
7749
7750 /*
7751 * If there's a deferred-free bplist, process that first.
7752 */
7753 (void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
7754 bpobj_count_block_cb, zcb, NULL);
7755
7756 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
7757 (void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
7758 bpobj_count_block_cb, zcb, NULL);
7759 }
7760
7761 zdb_claim_removing(spa, zcb);
7762
7763 if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
7764 VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,
7765 spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb,
7766 zcb, NULL));
7767 }
7768
7769 deleted_livelists_count_blocks(spa, zcb);
7770
7771 if (dump_opt['c'] > 1)
7772 flags |= TRAVERSE_PREFETCH_DATA;
7773
7774 zcb->zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
7775 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa));
7776 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
7777 zcb->zcb_totalasize +=
7778 metaslab_class_get_alloc(spa_embedded_log_class(spa));
7779 zcb->zcb_totalasize +=
7780 metaslab_class_get_alloc(spa_special_embedded_log_class(spa));
7781 zcb->zcb_start = zcb->zcb_lastprint = gethrtime();
7782 err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb);
7783
7784 /*
7785 * If we've traversed the data blocks then we need to wait for those
7786 * I/Os to complete. We leverage "The Godfather" zio to wait on
7787 * all async I/Os to complete.
7788 */
7789 if (dump_opt['c']) {
7790 for (c = 0; c < max_ncpus; c++) {
7791 (void) zio_wait(spa->spa_async_zio_root[c]);
7792 spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL,
7793 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
7794 ZIO_FLAG_GODFATHER);
7795 }
7796 }
7797 ASSERT0(spa->spa_load_verify_bytes);
7798
7799 /*
7800 * Done after zio_wait() since zcb_haderrors is modified in
7801 * zdb_blkptr_done()
7802 */
7803 zcb->zcb_haderrors |= err;
7804
7805 if (zcb->zcb_haderrors) {
7806 (void) printf("\nError counts:\n\n");
7807 (void) printf("\t%5s %s\n", "errno", "count");
7808 for (e = 0; e < 256; e++) {
7809 if (zcb->zcb_errors[e] != 0) {
7810 (void) printf("\t%5d %llu\n",
7811 e, (u_longlong_t)zcb->zcb_errors[e]);
7812 }
7813 }
7814 }
7815
7816 /*
7817 * Report any leaked segments.
7818 */
7819 leaks |= zdb_leak_fini(spa, zcb);
7820
7821 tzb = &zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL];
7822
7823 norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7824 norm_space = metaslab_class_get_space(spa_normal_class(spa));
7825
7826 total_alloc = norm_alloc +
7827 metaslab_class_get_alloc(spa_log_class(spa)) +
7828 metaslab_class_get_alloc(spa_embedded_log_class(spa)) +
7829 metaslab_class_get_alloc(spa_special_embedded_log_class(spa)) +
7830 metaslab_class_get_alloc(spa_special_class(spa)) +
7831 metaslab_class_get_alloc(spa_dedup_class(spa)) +
7832 get_unflushed_alloc_space(spa);
7833 total_found =
7834 tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize +
7835 zcb->zcb_removing_size + zcb->zcb_checkpoint_size;
7836
7837 if (total_found == total_alloc && !dump_opt['L']) {
7838 (void) printf("\n\tNo leaks (block sum matches space"
7839 " maps exactly)\n");
7840 } else if (!dump_opt['L']) {
7841 (void) printf("block traversal size %llu != alloc %llu "
7842 "(%s %lld)\n",
7843 (u_longlong_t)total_found,
7844 (u_longlong_t)total_alloc,
7845 (dump_opt['L']) ? "unreachable" : "leaked",
7846 (longlong_t)(total_alloc - total_found));
7847 }
7848
7849 if (tzb->zb_count == 0) {
7850 umem_free(zcb, sizeof (zdb_cb_t));
7851 return (2);
7852 }
7853
7854 (void) printf("\n");
7855 (void) printf("\t%-16s %14llu\n", "bp count:",
7856 (u_longlong_t)tzb->zb_count);
7857 (void) printf("\t%-16s %14llu\n", "ganged count:",
7858 (longlong_t)tzb->zb_gangs);
7859 (void) printf("\t%-16s %14llu avg: %6llu\n", "bp logical:",
7860 (u_longlong_t)tzb->zb_lsize,
7861 (u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
7862 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7863 "bp physical:", (u_longlong_t)tzb->zb_psize,
7864 (u_longlong_t)(tzb->zb_psize / tzb->zb_count),
7865 (double)tzb->zb_lsize / tzb->zb_psize);
7866 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7867 "bp allocated:", (u_longlong_t)tzb->zb_asize,
7868 (u_longlong_t)(tzb->zb_asize / tzb->zb_count),
7869 (double)tzb->zb_lsize / tzb->zb_asize);
7870 (void) printf("\t%-16s %14llu ref>1: %6llu deduplication: %6.2f\n",
7871 "bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize,
7872 (u_longlong_t)zcb->zcb_dedup_blocks,
7873 (double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0);
7874 (void) printf("\t%-16s %14llu count: %6llu\n",
7875 "bp cloned:", (u_longlong_t)zcb->zcb_clone_asize,
7876 (u_longlong_t)zcb->zcb_clone_blocks);
7877 (void) printf("\t%-16s %14llu used: %5.2f%%\n", "Normal class:",
7878 (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
7879
7880 if (spa_special_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7881 uint64_t alloc = metaslab_class_get_alloc(
7882 spa_special_class(spa));
7883 uint64_t space = metaslab_class_get_space(
7884 spa_special_class(spa));
7885
7886 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7887 "Special class", (u_longlong_t)alloc,
7888 100.0 * alloc / space);
7889 }
7890
7891 if (spa_dedup_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7892 uint64_t alloc = metaslab_class_get_alloc(
7893 spa_dedup_class(spa));
7894 uint64_t space = metaslab_class_get_space(
7895 spa_dedup_class(spa));
7896
7897 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7898 "Dedup class", (u_longlong_t)alloc,
7899 100.0 * alloc / space);
7900 }
7901
7902 if (spa_embedded_log_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7903 uint64_t alloc = metaslab_class_get_alloc(
7904 spa_embedded_log_class(spa));
7905 uint64_t space = metaslab_class_get_space(
7906 spa_embedded_log_class(spa));
7907
7908 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7909 "Embedded log class", (u_longlong_t)alloc,
7910 100.0 * alloc / space);
7911 }
7912
7913 if (spa_special_embedded_log_class(spa)->mc_allocator[0].mca_rotor
7914 != NULL) {
7915 uint64_t alloc = metaslab_class_get_alloc(
7916 spa_special_embedded_log_class(spa));
7917 uint64_t space = metaslab_class_get_space(
7918 spa_special_embedded_log_class(spa));
7919
7920 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7921 "Special embedded log", (u_longlong_t)alloc,
7922 100.0 * alloc / space);
7923 }
7924
7925 for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
7926 if (zcb->zcb_embedded_blocks[i] == 0)
7927 continue;
7928 (void) printf("\n");
7929 (void) printf("\tadditional, non-pointer bps of type %u: "
7930 "%10llu\n",
7931 i, (u_longlong_t)zcb->zcb_embedded_blocks[i]);
7932
7933 if (dump_opt['b'] >= 3) {
7934 (void) printf("\t number of (compressed) bytes: "
7935 "number of bps\n");
7936 dump_histogram(zcb->zcb_embedded_histogram[i],
7937 sizeof (zcb->zcb_embedded_histogram[i]) /
7938 sizeof (zcb->zcb_embedded_histogram[i][0]), 0);
7939 }
7940 }
7941
7942 if (tzb->zb_ditto_samevdev != 0) {
7943 (void) printf("\tDittoed blocks on same vdev: %llu\n",
7944 (longlong_t)tzb->zb_ditto_samevdev);
7945 }
7946 if (tzb->zb_ditto_same_ms != 0) {
7947 (void) printf("\tDittoed blocks in same metaslab: %llu\n",
7948 (longlong_t)tzb->zb_ditto_same_ms);
7949 }
7950
7951 for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) {
7952 vdev_t *vd = spa->spa_root_vdev->vdev_child[v];
7953 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7954
7955 if (vim == NULL) {
7956 continue;
7957 }
7958
7959 char mem[32];
7960 zdb_nicenum(vdev_indirect_mapping_num_entries(vim),
7961 mem, vdev_indirect_mapping_size(vim));
7962
7963 (void) printf("\tindirect vdev id %llu has %llu segments "
7964 "(%s in memory)\n",
7965 (longlong_t)vd->vdev_id,
7966 (longlong_t)vdev_indirect_mapping_num_entries(vim), mem);
7967 }
7968
7969 if (dump_opt['b'] >= 2) {
7970 int l, t, level;
7971 char csize[32], lsize[32], psize[32], asize[32];
7972 char avg[32], gang[32];
7973 (void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
7974 "\t avg\t comp\t%%Total\tType\n");
7975
7976 zfs_blkstat_t *mdstats = umem_zalloc(sizeof (zfs_blkstat_t),
7977 UMEM_NOFAIL);
7978
7979 for (t = 0; t <= ZDB_OT_TOTAL; t++) {
7980 const char *typename;
7981
7982 /* make sure nicenum has enough space */
7983 _Static_assert(sizeof (csize) >= NN_NUMBUF_SZ,
7984 "csize truncated");
7985 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ,
7986 "lsize truncated");
7987 _Static_assert(sizeof (psize) >= NN_NUMBUF_SZ,
7988 "psize truncated");
7989 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ,
7990 "asize truncated");
7991 _Static_assert(sizeof (avg) >= NN_NUMBUF_SZ,
7992 "avg truncated");
7993 _Static_assert(sizeof (gang) >= NN_NUMBUF_SZ,
7994 "gang truncated");
7995
7996 if (t < DMU_OT_NUMTYPES)
7997 typename = dmu_ot[t].ot_name;
7998 else
7999 typename = zdb_ot_extname[t - DMU_OT_NUMTYPES];
8000
8001 if (zcb->zcb_type[ZB_TOTAL][t].zb_asize == 0) {
8002 (void) printf("%6s\t%5s\t%5s\t%5s"
8003 "\t%5s\t%5s\t%6s\t%s\n",
8004 "-",
8005 "-",
8006 "-",
8007 "-",
8008 "-",
8009 "-",
8010 "-",
8011 typename);
8012 continue;
8013 }
8014
8015 for (l = ZB_TOTAL - 1; l >= -1; l--) {
8016 level = (l == -1 ? ZB_TOTAL : l);
8017 zb = &zcb->zcb_type[level][t];
8018
8019 if (zb->zb_asize == 0)
8020 continue;
8021
8022 if (level != ZB_TOTAL && t < DMU_OT_NUMTYPES &&
8023 (level > 0 || DMU_OT_IS_METADATA(t))) {
8024 mdstats->zb_count += zb->zb_count;
8025 mdstats->zb_lsize += zb->zb_lsize;
8026 mdstats->zb_psize += zb->zb_psize;
8027 mdstats->zb_asize += zb->zb_asize;
8028 mdstats->zb_gangs += zb->zb_gangs;
8029 }
8030
8031 if (dump_opt['b'] < 3 && level != ZB_TOTAL)
8032 continue;
8033
8034 if (level == 0 && zb->zb_asize ==
8035 zcb->zcb_type[ZB_TOTAL][t].zb_asize)
8036 continue;
8037
8038 zdb_nicenum(zb->zb_count, csize,
8039 sizeof (csize));
8040 zdb_nicenum(zb->zb_lsize, lsize,
8041 sizeof (lsize));
8042 zdb_nicenum(zb->zb_psize, psize,
8043 sizeof (psize));
8044 zdb_nicenum(zb->zb_asize, asize,
8045 sizeof (asize));
8046 zdb_nicenum(zb->zb_asize / zb->zb_count, avg,
8047 sizeof (avg));
8048 zdb_nicenum(zb->zb_gangs, gang, sizeof (gang));
8049
8050 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
8051 "\t%5.2f\t%6.2f\t",
8052 csize, lsize, psize, asize, avg,
8053 (double)zb->zb_lsize / zb->zb_psize,
8054 100.0 * zb->zb_asize / tzb->zb_asize);
8055
8056 if (level == ZB_TOTAL)
8057 (void) printf("%s\n", typename);
8058 else
8059 (void) printf(" L%d %s\n",
8060 level, typename);
8061
8062 if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) {
8063 (void) printf("\t number of ganged "
8064 "blocks: %s\n", gang);
8065 }
8066
8067 if (dump_opt['b'] >= 4) {
8068 (void) printf("psize "
8069 "(in 512-byte sectors): "
8070 "number of blocks\n");
8071 dump_histogram(zb->zb_psize_histogram,
8072 PSIZE_HISTO_SIZE, 0);
8073 }
8074 }
8075 }
8076 zdb_nicenum(mdstats->zb_count, csize,
8077 sizeof (csize));
8078 zdb_nicenum(mdstats->zb_lsize, lsize,
8079 sizeof (lsize));
8080 zdb_nicenum(mdstats->zb_psize, psize,
8081 sizeof (psize));
8082 zdb_nicenum(mdstats->zb_asize, asize,
8083 sizeof (asize));
8084 zdb_nicenum(mdstats->zb_asize / mdstats->zb_count, avg,
8085 sizeof (avg));
8086 zdb_nicenum(mdstats->zb_gangs, gang, sizeof (gang));
8087
8088 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
8089 "\t%5.2f\t%6.2f\t",
8090 csize, lsize, psize, asize, avg,
8091 (double)mdstats->zb_lsize / mdstats->zb_psize,
8092 100.0 * mdstats->zb_asize / tzb->zb_asize);
8093 (void) printf("%s\n", "Metadata Total");
8094
8095 /* Output a table summarizing block sizes in the pool */
8096 if (dump_opt['b'] >= 2) {
8097 dump_size_histograms(zcb);
8098 }
8099
8100 umem_free(mdstats, sizeof (zfs_blkstat_t));
8101 }
8102
8103 (void) printf("\n");
8104
8105 if (leaks) {
8106 umem_free(zcb, sizeof (zdb_cb_t));
8107 return (2);
8108 }
8109
8110 if (zcb->zcb_haderrors) {
8111 umem_free(zcb, sizeof (zdb_cb_t));
8112 return (3);
8113 }
8114
8115 umem_free(zcb, sizeof (zdb_cb_t));
8116 return (0);
8117 }
8118
8119 typedef struct zdb_ddt_entry {
8120 /* key must be first for ddt_key_compare */
8121 ddt_key_t zdde_key;
8122 uint64_t zdde_ref_blocks;
8123 uint64_t zdde_ref_lsize;
8124 uint64_t zdde_ref_psize;
8125 uint64_t zdde_ref_dsize;
8126 avl_node_t zdde_node;
8127 } zdb_ddt_entry_t;
8128
8129 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)8130 zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
8131 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
8132 {
8133 (void) zilog, (void) dnp;
8134 avl_tree_t *t = arg;
8135 avl_index_t where;
8136 zdb_ddt_entry_t *zdde, zdde_search;
8137
8138 if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
8139 BP_IS_EMBEDDED(bp))
8140 return (0);
8141
8142 if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
8143 (void) printf("traversing objset %llu, %llu objects, "
8144 "%lu blocks so far\n",
8145 (u_longlong_t)zb->zb_objset,
8146 (u_longlong_t)BP_GET_FILL(bp),
8147 avl_numnodes(t));
8148 }
8149
8150 if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF ||
8151 BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
8152 return (0);
8153
8154 ddt_key_fill(&zdde_search.zdde_key, bp);
8155
8156 zdde = avl_find(t, &zdde_search, &where);
8157
8158 if (zdde == NULL) {
8159 zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL);
8160 zdde->zdde_key = zdde_search.zdde_key;
8161 avl_insert(t, zdde, where);
8162 }
8163
8164 zdde->zdde_ref_blocks += 1;
8165 zdde->zdde_ref_lsize += BP_GET_LSIZE(bp);
8166 zdde->zdde_ref_psize += BP_GET_PSIZE(bp);
8167 zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp);
8168
8169 return (0);
8170 }
8171
8172 static void
dump_simulated_ddt(spa_t * spa)8173 dump_simulated_ddt(spa_t *spa)
8174 {
8175 avl_tree_t t;
8176 void *cookie = NULL;
8177 zdb_ddt_entry_t *zdde;
8178 ddt_histogram_t ddh_total = {{{0}}};
8179 ddt_stat_t dds_total = {0};
8180
8181 avl_create(&t, ddt_key_compare,
8182 sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
8183
8184 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8185
8186 (void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
8187 TRAVERSE_NO_DECRYPT, zdb_ddt_add_cb, &t);
8188
8189 spa_config_exit(spa, SCL_CONFIG, FTAG);
8190
8191 while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) {
8192 uint64_t refcnt = zdde->zdde_ref_blocks;
8193 ASSERT(refcnt != 0);
8194
8195 ddt_stat_t *dds = &ddh_total.ddh_stat[highbit64(refcnt) - 1];
8196
8197 dds->dds_blocks += zdde->zdde_ref_blocks / refcnt;
8198 dds->dds_lsize += zdde->zdde_ref_lsize / refcnt;
8199 dds->dds_psize += zdde->zdde_ref_psize / refcnt;
8200 dds->dds_dsize += zdde->zdde_ref_dsize / refcnt;
8201
8202 dds->dds_ref_blocks += zdde->zdde_ref_blocks;
8203 dds->dds_ref_lsize += zdde->zdde_ref_lsize;
8204 dds->dds_ref_psize += zdde->zdde_ref_psize;
8205 dds->dds_ref_dsize += zdde->zdde_ref_dsize;
8206
8207 umem_free(zdde, sizeof (*zdde));
8208 }
8209
8210 avl_destroy(&t);
8211
8212 ddt_histogram_total(&dds_total, &ddh_total);
8213
8214 (void) printf("Simulated DDT histogram:\n");
8215
8216 zpool_dump_ddt(&dds_total, &ddh_total, dump_opt['P'] > 0);
8217
8218 dump_dedup_ratio(&dds_total);
8219 }
8220
8221 static int
verify_device_removal_feature_counts(spa_t * spa)8222 verify_device_removal_feature_counts(spa_t *spa)
8223 {
8224 uint64_t dr_feature_refcount = 0;
8225 uint64_t oc_feature_refcount = 0;
8226 uint64_t indirect_vdev_count = 0;
8227 uint64_t precise_vdev_count = 0;
8228 uint64_t obsolete_counts_object_count = 0;
8229 uint64_t obsolete_sm_count = 0;
8230 uint64_t obsolete_counts_count = 0;
8231 uint64_t scip_count = 0;
8232 uint64_t obsolete_bpobj_count = 0;
8233 int ret = 0;
8234
8235 spa_condensing_indirect_phys_t *scip =
8236 &spa->spa_condensing_indirect_phys;
8237 if (scip->scip_next_mapping_object != 0) {
8238 vdev_t *vd = spa->spa_root_vdev->vdev_child[scip->scip_vdev];
8239 ASSERT(scip->scip_prev_obsolete_sm_object != 0);
8240 ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
8241
8242 (void) printf("Condensing indirect vdev %llu: new mapping "
8243 "object %llu, prev obsolete sm %llu\n",
8244 (u_longlong_t)scip->scip_vdev,
8245 (u_longlong_t)scip->scip_next_mapping_object,
8246 (u_longlong_t)scip->scip_prev_obsolete_sm_object);
8247 if (scip->scip_prev_obsolete_sm_object != 0) {
8248 space_map_t *prev_obsolete_sm = NULL;
8249 VERIFY0(space_map_open(&prev_obsolete_sm,
8250 spa->spa_meta_objset,
8251 scip->scip_prev_obsolete_sm_object,
8252 0, vd->vdev_asize, 0));
8253 dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm);
8254 (void) printf("\n");
8255 space_map_close(prev_obsolete_sm);
8256 }
8257
8258 scip_count += 2;
8259 }
8260
8261 for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
8262 vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
8263 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
8264
8265 if (vic->vic_mapping_object != 0) {
8266 ASSERT(vd->vdev_ops == &vdev_indirect_ops ||
8267 vd->vdev_removing);
8268 indirect_vdev_count++;
8269
8270 if (vd->vdev_indirect_mapping->vim_havecounts) {
8271 obsolete_counts_count++;
8272 }
8273 }
8274
8275 boolean_t are_precise;
8276 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
8277 if (are_precise) {
8278 ASSERT(vic->vic_mapping_object != 0);
8279 precise_vdev_count++;
8280 }
8281
8282 uint64_t obsolete_sm_object;
8283 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
8284 if (obsolete_sm_object != 0) {
8285 ASSERT(vic->vic_mapping_object != 0);
8286 obsolete_sm_count++;
8287 }
8288 }
8289
8290 (void) feature_get_refcount(spa,
8291 &spa_feature_table[SPA_FEATURE_DEVICE_REMOVAL],
8292 &dr_feature_refcount);
8293 (void) feature_get_refcount(spa,
8294 &spa_feature_table[SPA_FEATURE_OBSOLETE_COUNTS],
8295 &oc_feature_refcount);
8296
8297 if (dr_feature_refcount != indirect_vdev_count) {
8298 ret = 1;
8299 (void) printf("Number of indirect vdevs (%llu) " \
8300 "does not match feature count (%llu)\n",
8301 (u_longlong_t)indirect_vdev_count,
8302 (u_longlong_t)dr_feature_refcount);
8303 } else {
8304 (void) printf("Verified device_removal feature refcount " \
8305 "of %llu is correct\n",
8306 (u_longlong_t)dr_feature_refcount);
8307 }
8308
8309 if (zap_contains(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT,
8310 DMU_POOL_OBSOLETE_BPOBJ) == 0) {
8311 obsolete_bpobj_count++;
8312 }
8313
8314
8315 obsolete_counts_object_count = precise_vdev_count;
8316 obsolete_counts_object_count += obsolete_sm_count;
8317 obsolete_counts_object_count += obsolete_counts_count;
8318 obsolete_counts_object_count += scip_count;
8319 obsolete_counts_object_count += obsolete_bpobj_count;
8320 obsolete_counts_object_count += remap_deadlist_count;
8321
8322 if (oc_feature_refcount != obsolete_counts_object_count) {
8323 ret = 1;
8324 (void) printf("Number of obsolete counts objects (%llu) " \
8325 "does not match feature count (%llu)\n",
8326 (u_longlong_t)obsolete_counts_object_count,
8327 (u_longlong_t)oc_feature_refcount);
8328 (void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
8329 "ob:%llu rd:%llu\n",
8330 (u_longlong_t)precise_vdev_count,
8331 (u_longlong_t)obsolete_sm_count,
8332 (u_longlong_t)obsolete_counts_count,
8333 (u_longlong_t)scip_count,
8334 (u_longlong_t)obsolete_bpobj_count,
8335 (u_longlong_t)remap_deadlist_count);
8336 } else {
8337 (void) printf("Verified indirect_refcount feature refcount " \
8338 "of %llu is correct\n",
8339 (u_longlong_t)oc_feature_refcount);
8340 }
8341 return (ret);
8342 }
8343
8344 static void
zdb_set_skip_mmp(char * target)8345 zdb_set_skip_mmp(char *target)
8346 {
8347 spa_t *spa;
8348
8349 /*
8350 * Disable the activity check to allow examination of
8351 * active pools.
8352 */
8353 spa_namespace_enter(FTAG);
8354 if ((spa = spa_lookup(target)) != NULL) {
8355 spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
8356 }
8357 spa_namespace_exit(FTAG);
8358 }
8359
8360 #define BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
8361 /*
8362 * Import the checkpointed state of the pool specified by the target
8363 * parameter as readonly. The function also accepts a pool config
8364 * as an optional parameter, else it attempts to infer the config by
8365 * the name of the target pool.
8366 *
8367 * Note that the checkpointed state's pool name will be the name of
8368 * the original pool with the above suffix appended to it. In addition,
8369 * if the target is not a pool name (e.g. a path to a dataset) then
8370 * the new_path parameter is populated with the updated path to
8371 * reflect the fact that we are looking into the checkpointed state.
8372 *
8373 * The function returns a newly-allocated copy of the name of the
8374 * pool containing the checkpointed state. When this copy is no
8375 * longer needed it should be freed with free(3C). Same thing
8376 * applies to the new_path parameter if allocated.
8377 */
8378 static char *
import_checkpointed_state(char * target,nvlist_t * cfg,boolean_t target_is_spa,char ** new_path)8379 import_checkpointed_state(char *target, nvlist_t *cfg, boolean_t target_is_spa,
8380 char **new_path)
8381 {
8382 int error = 0;
8383 char *poolname, *bogus_name = NULL;
8384 boolean_t freecfg = B_FALSE;
8385
8386 /* If the target is not a pool, the extract the pool name */
8387 char *path_start = strchr(target, '/');
8388 if (target_is_spa || path_start == NULL) {
8389 poolname = target;
8390 } else {
8391 size_t poolname_len = path_start - target;
8392 poolname = strndup(target, poolname_len);
8393 }
8394
8395 if (cfg == NULL) {
8396 zdb_set_skip_mmp(poolname);
8397 error = spa_get_stats(poolname, &cfg, NULL, 0);
8398 if (error != 0) {
8399 fatal("Tried to read config of pool \"%s\" but "
8400 "spa_get_stats() failed with error %d\n",
8401 poolname, error);
8402 }
8403 freecfg = B_TRUE;
8404 }
8405
8406 if (asprintf(&bogus_name, "%s%s", poolname, BOGUS_SUFFIX) == -1) {
8407 if (target != poolname)
8408 free(poolname);
8409 return (NULL);
8410 }
8411 fnvlist_add_string(cfg, ZPOOL_CONFIG_POOL_NAME, bogus_name);
8412
8413 error = spa_import(bogus_name, cfg, NULL,
8414 ZFS_IMPORT_MISSING_LOG | ZFS_IMPORT_CHECKPOINT |
8415 ZFS_IMPORT_SKIP_MMP);
8416 if (freecfg)
8417 nvlist_free(cfg);
8418 if (error != 0) {
8419 fatal("Tried to import pool \"%s\" but spa_import() failed "
8420 "with error %d\n", bogus_name, error);
8421 }
8422
8423 if (new_path != NULL && !target_is_spa) {
8424 if (asprintf(new_path, "%s%s", bogus_name,
8425 path_start != NULL ? path_start : "") == -1) {
8426 free(bogus_name);
8427 if (!target_is_spa && path_start != NULL)
8428 free(poolname);
8429 return (NULL);
8430 }
8431 }
8432
8433 if (target != poolname)
8434 free(poolname);
8435
8436 return (bogus_name);
8437 }
8438
8439 typedef struct verify_checkpoint_sm_entry_cb_arg {
8440 vdev_t *vcsec_vd;
8441
8442 /* the following fields are only used for printing progress */
8443 uint64_t vcsec_entryid;
8444 uint64_t vcsec_num_entries;
8445 } verify_checkpoint_sm_entry_cb_arg_t;
8446
8447 #define ENTRIES_PER_PROGRESS_UPDATE 10000
8448
8449 static int
verify_checkpoint_sm_entry_cb(space_map_entry_t * sme,void * arg)8450 verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, void *arg)
8451 {
8452 verify_checkpoint_sm_entry_cb_arg_t *vcsec = arg;
8453 vdev_t *vd = vcsec->vcsec_vd;
8454 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
8455 uint64_t end = sme->sme_offset + sme->sme_run;
8456
8457 ASSERT(sme->sme_type == SM_FREE);
8458
8459 if ((vcsec->vcsec_entryid % ENTRIES_PER_PROGRESS_UPDATE) == 0) {
8460 (void) fprintf(stderr,
8461 "\rverifying vdev %llu, space map entry %llu of %llu ...",
8462 (longlong_t)vd->vdev_id,
8463 (longlong_t)vcsec->vcsec_entryid,
8464 (longlong_t)vcsec->vcsec_num_entries);
8465 }
8466 vcsec->vcsec_entryid++;
8467
8468 /*
8469 * See comment in checkpoint_sm_exclude_entry_cb()
8470 */
8471 VERIFY3U(sme->sme_offset, >=, ms->ms_start);
8472 VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
8473
8474 /*
8475 * The entries in the vdev_checkpoint_sm should be marked as
8476 * allocated in the checkpointed state of the pool, therefore
8477 * their respective ms_allocateable trees should not contain them.
8478 */
8479 mutex_enter(&ms->ms_lock);
8480 zfs_range_tree_verify_not_present(ms->ms_allocatable,
8481 sme->sme_offset, sme->sme_run);
8482 mutex_exit(&ms->ms_lock);
8483
8484 return (0);
8485 }
8486
8487 /*
8488 * Verify that all segments in the vdev_checkpoint_sm are allocated
8489 * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
8490 * ms_allocatable).
8491 *
8492 * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
8493 * each vdev in the current state of the pool to the metaslab space maps
8494 * (ms_sm) of the checkpointed state of the pool.
8495 *
8496 * Note that the function changes the state of the ms_allocatable
8497 * trees of the current spa_t. The entries of these ms_allocatable
8498 * trees are cleared out and then repopulated from with the free
8499 * entries of their respective ms_sm space maps.
8500 */
8501 static void
verify_checkpoint_vdev_spacemaps(spa_t * checkpoint,spa_t * current)8502 verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, spa_t *current)
8503 {
8504 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
8505 vdev_t *current_rvd = current->spa_root_vdev;
8506
8507 load_concrete_ms_allocatable_trees(checkpoint, SM_FREE);
8508
8509 for (uint64_t c = 0; c < ckpoint_rvd->vdev_children; c++) {
8510 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[c];
8511 vdev_t *current_vd = current_rvd->vdev_child[c];
8512
8513 space_map_t *checkpoint_sm = NULL;
8514 uint64_t checkpoint_sm_obj;
8515
8516 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
8517 /*
8518 * Since we don't allow device removal in a pool
8519 * that has a checkpoint, we expect that all removed
8520 * vdevs were removed from the pool before the
8521 * checkpoint.
8522 */
8523 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
8524 continue;
8525 }
8526
8527 /*
8528 * If the checkpoint space map doesn't exist, then nothing
8529 * here is checkpointed so there's nothing to verify.
8530 */
8531 if (current_vd->vdev_top_zap == 0 ||
8532 zap_contains(spa_meta_objset(current),
8533 current_vd->vdev_top_zap,
8534 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
8535 continue;
8536
8537 VERIFY0(zap_lookup(spa_meta_objset(current),
8538 current_vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
8539 sizeof (uint64_t), 1, &checkpoint_sm_obj));
8540
8541 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current),
8542 checkpoint_sm_obj, 0, current_vd->vdev_asize,
8543 current_vd->vdev_ashift));
8544
8545 verify_checkpoint_sm_entry_cb_arg_t vcsec;
8546 vcsec.vcsec_vd = ckpoint_vd;
8547 vcsec.vcsec_entryid = 0;
8548 vcsec.vcsec_num_entries =
8549 space_map_length(checkpoint_sm) / sizeof (uint64_t);
8550 VERIFY0(space_map_iterate(checkpoint_sm,
8551 space_map_length(checkpoint_sm),
8552 verify_checkpoint_sm_entry_cb, &vcsec));
8553 if (dump_opt['m'] > 3)
8554 dump_spacemap(current->spa_meta_objset, checkpoint_sm);
8555 space_map_close(checkpoint_sm);
8556 }
8557
8558 /*
8559 * If we've added vdevs since we took the checkpoint, ensure
8560 * that their checkpoint space maps are empty.
8561 */
8562 if (ckpoint_rvd->vdev_children < current_rvd->vdev_children) {
8563 for (uint64_t c = ckpoint_rvd->vdev_children;
8564 c < current_rvd->vdev_children; c++) {
8565 vdev_t *current_vd = current_rvd->vdev_child[c];
8566 VERIFY0P(current_vd->vdev_checkpoint_sm);
8567 }
8568 }
8569
8570 /* for cleaner progress output */
8571 (void) fprintf(stderr, "\n");
8572 }
8573
8574 /*
8575 * Verifies that all space that's allocated in the checkpoint is
8576 * still allocated in the current version, by checking that everything
8577 * in checkpoint's ms_allocatable (which is actually allocated, not
8578 * allocatable/free) is not present in current's ms_allocatable.
8579 *
8580 * Note that the function changes the state of the ms_allocatable
8581 * trees of both spas when called. The entries of all ms_allocatable
8582 * trees are cleared out and then repopulated from their respective
8583 * ms_sm space maps. In the checkpointed state we load the allocated
8584 * entries, and in the current state we load the free entries.
8585 */
8586 static void
verify_checkpoint_ms_spacemaps(spa_t * checkpoint,spa_t * current)8587 verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_t *current)
8588 {
8589 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
8590 vdev_t *current_rvd = current->spa_root_vdev;
8591
8592 load_concrete_ms_allocatable_trees(checkpoint, SM_ALLOC);
8593 load_concrete_ms_allocatable_trees(current, SM_FREE);
8594
8595 for (uint64_t i = 0; i < ckpoint_rvd->vdev_children; i++) {
8596 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[i];
8597 vdev_t *current_vd = current_rvd->vdev_child[i];
8598
8599 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
8600 /*
8601 * See comment in verify_checkpoint_vdev_spacemaps()
8602 */
8603 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
8604 continue;
8605 }
8606
8607 for (uint64_t m = 0; m < ckpoint_vd->vdev_ms_count; m++) {
8608 metaslab_t *ckpoint_msp = ckpoint_vd->vdev_ms[m];
8609 metaslab_t *current_msp = current_vd->vdev_ms[m];
8610
8611 (void) fprintf(stderr,
8612 "\rverifying vdev %llu of %llu, "
8613 "metaslab %llu of %llu ...",
8614 (longlong_t)current_vd->vdev_id,
8615 (longlong_t)current_rvd->vdev_children,
8616 (longlong_t)current_vd->vdev_ms[m]->ms_id,
8617 (longlong_t)current_vd->vdev_ms_count);
8618
8619 /*
8620 * We walk through the ms_allocatable trees that
8621 * are loaded with the allocated blocks from the
8622 * ms_sm spacemaps of the checkpoint. For each
8623 * one of these ranges we ensure that none of them
8624 * exists in the ms_allocatable trees of the
8625 * current state which are loaded with the ranges
8626 * that are currently free.
8627 *
8628 * This way we ensure that none of the blocks that
8629 * are part of the checkpoint were freed by mistake.
8630 */
8631 zfs_range_tree_walk(ckpoint_msp->ms_allocatable,
8632 (zfs_range_tree_func_t *)
8633 zfs_range_tree_verify_not_present,
8634 current_msp->ms_allocatable);
8635 }
8636 }
8637
8638 /* for cleaner progress output */
8639 (void) fprintf(stderr, "\n");
8640 }
8641
8642 static void
verify_checkpoint_blocks(spa_t * spa)8643 verify_checkpoint_blocks(spa_t *spa)
8644 {
8645 ASSERT(!dump_opt['L']);
8646
8647 spa_t *checkpoint_spa;
8648 char *checkpoint_pool;
8649 int error = 0;
8650
8651 /*
8652 * We import the checkpointed state of the pool (under a different
8653 * name) so we can do verification on it against the current state
8654 * of the pool.
8655 */
8656 checkpoint_pool = import_checkpointed_state(spa->spa_name, NULL, B_TRUE,
8657 NULL);
8658 ASSERT(strcmp(spa->spa_name, checkpoint_pool) != 0);
8659
8660 error = spa_open(checkpoint_pool, &checkpoint_spa, FTAG);
8661 if (error != 0) {
8662 fatal("Tried to open pool \"%s\" but spa_open() failed with "
8663 "error %d\n", checkpoint_pool, error);
8664 }
8665
8666 /*
8667 * Ensure that ranges in the checkpoint space maps of each vdev
8668 * are allocated according to the checkpointed state's metaslab
8669 * space maps.
8670 */
8671 verify_checkpoint_vdev_spacemaps(checkpoint_spa, spa);
8672
8673 /*
8674 * Ensure that allocated ranges in the checkpoint's metaslab
8675 * space maps remain allocated in the metaslab space maps of
8676 * the current state.
8677 */
8678 verify_checkpoint_ms_spacemaps(checkpoint_spa, spa);
8679
8680 /*
8681 * Once we are done, we get rid of the checkpointed state.
8682 */
8683 spa_close(checkpoint_spa, FTAG);
8684 free(checkpoint_pool);
8685 }
8686
8687 static void
dump_leftover_checkpoint_blocks(spa_t * spa)8688 dump_leftover_checkpoint_blocks(spa_t *spa)
8689 {
8690 vdev_t *rvd = spa->spa_root_vdev;
8691
8692 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
8693 vdev_t *vd = rvd->vdev_child[i];
8694
8695 space_map_t *checkpoint_sm = NULL;
8696 uint64_t checkpoint_sm_obj;
8697
8698 if (vd->vdev_top_zap == 0)
8699 continue;
8700
8701 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
8702 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
8703 continue;
8704
8705 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
8706 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
8707 sizeof (uint64_t), 1, &checkpoint_sm_obj));
8708
8709 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
8710 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
8711 dump_spacemap(spa->spa_meta_objset, checkpoint_sm);
8712 space_map_close(checkpoint_sm);
8713 }
8714 }
8715
8716 static int
verify_checkpoint(spa_t * spa)8717 verify_checkpoint(spa_t *spa)
8718 {
8719 uberblock_t checkpoint;
8720 int error;
8721
8722 if (!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT))
8723 return (0);
8724
8725 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8726 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
8727 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
8728
8729 if (error == ENOENT && !dump_opt['L']) {
8730 /*
8731 * If the feature is active but the uberblock is missing
8732 * then we must be in the middle of discarding the
8733 * checkpoint.
8734 */
8735 (void) printf("\nPartially discarded checkpoint "
8736 "state found:\n");
8737 if (dump_opt['m'] > 3)
8738 dump_leftover_checkpoint_blocks(spa);
8739 return (0);
8740 } else if (error != 0) {
8741 (void) printf("lookup error %d when looking for "
8742 "checkpointed uberblock in MOS\n", error);
8743 return (error);
8744 }
8745 dump_uberblock(&checkpoint, "\nCheckpointed uberblock found:\n", "\n");
8746
8747 if (checkpoint.ub_checkpoint_txg == 0) {
8748 (void) printf("\nub_checkpoint_txg not set in checkpointed "
8749 "uberblock\n");
8750 error = 3;
8751 }
8752
8753 if (error == 0 && !dump_opt['L'])
8754 verify_checkpoint_blocks(spa);
8755
8756 return (error);
8757 }
8758
8759 static void
mos_leaks_cb(void * arg,uint64_t start,uint64_t size)8760 mos_leaks_cb(void *arg, uint64_t start, uint64_t size)
8761 {
8762 (void) arg;
8763 for (uint64_t i = start; i < size; i++) {
8764 (void) printf("MOS object %llu referenced but not allocated\n",
8765 (u_longlong_t)i);
8766 }
8767 }
8768
8769 static void
mos_obj_refd(uint64_t obj)8770 mos_obj_refd(uint64_t obj)
8771 {
8772 if (obj != 0 && mos_refd_objs != NULL)
8773 zfs_range_tree_add(mos_refd_objs, obj, 1);
8774 }
8775
8776 /*
8777 * Call on a MOS object that may already have been referenced.
8778 */
8779 static void
mos_obj_refd_multiple(uint64_t obj)8780 mos_obj_refd_multiple(uint64_t obj)
8781 {
8782 if (obj != 0 && mos_refd_objs != NULL &&
8783 !zfs_range_tree_contains(mos_refd_objs, obj, 1))
8784 zfs_range_tree_add(mos_refd_objs, obj, 1);
8785 }
8786
8787 static void
mos_leak_vdev_top_zap(vdev_t * vd)8788 mos_leak_vdev_top_zap(vdev_t *vd)
8789 {
8790 uint64_t ms_flush_data_obj;
8791 int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
8792 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
8793 sizeof (ms_flush_data_obj), 1, &ms_flush_data_obj);
8794 if (error == ENOENT)
8795 return;
8796 ASSERT0(error);
8797
8798 mos_obj_refd(ms_flush_data_obj);
8799 }
8800
8801 static void
mos_leak_vdev(vdev_t * vd)8802 mos_leak_vdev(vdev_t *vd)
8803 {
8804 mos_obj_refd(vd->vdev_dtl_object);
8805 mos_obj_refd(vd->vdev_ms_array);
8806 mos_obj_refd(vd->vdev_indirect_config.vic_births_object);
8807 mos_obj_refd(vd->vdev_indirect_config.vic_mapping_object);
8808 mos_obj_refd(vd->vdev_leaf_zap);
8809 if (vd->vdev_checkpoint_sm != NULL)
8810 mos_obj_refd(vd->vdev_checkpoint_sm->sm_object);
8811 if (vd->vdev_indirect_mapping != NULL) {
8812 mos_obj_refd(vd->vdev_indirect_mapping->
8813 vim_phys->vimp_counts_object);
8814 }
8815 if (vd->vdev_obsolete_sm != NULL)
8816 mos_obj_refd(vd->vdev_obsolete_sm->sm_object);
8817
8818 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
8819 metaslab_t *ms = vd->vdev_ms[m];
8820 mos_obj_refd(space_map_object(ms->ms_sm));
8821 }
8822
8823 if (vd->vdev_root_zap != 0)
8824 mos_obj_refd(vd->vdev_root_zap);
8825
8826 if (vd->vdev_top_zap != 0) {
8827 mos_obj_refd(vd->vdev_top_zap);
8828 mos_leak_vdev_top_zap(vd);
8829 }
8830
8831 for (uint64_t c = 0; c < vd->vdev_children; c++) {
8832 mos_leak_vdev(vd->vdev_child[c]);
8833 }
8834 }
8835
8836 static void
mos_leak_log_spacemaps(spa_t * spa)8837 mos_leak_log_spacemaps(spa_t *spa)
8838 {
8839 uint64_t spacemap_zap;
8840 int error = zap_lookup(spa_meta_objset(spa),
8841 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LOG_SPACEMAP_ZAP,
8842 sizeof (spacemap_zap), 1, &spacemap_zap);
8843 if (error == ENOENT)
8844 return;
8845 ASSERT0(error);
8846
8847 mos_obj_refd(spacemap_zap);
8848 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
8849 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls))
8850 mos_obj_refd(sls->sls_sm_obj);
8851 }
8852
8853 static void
errorlog_count_refd(objset_t * mos,uint64_t errlog)8854 errorlog_count_refd(objset_t *mos, uint64_t errlog)
8855 {
8856 zap_cursor_t zc;
8857 zap_attribute_t *za = zap_attribute_alloc();
8858 for (zap_cursor_init(&zc, mos, errlog);
8859 zap_cursor_retrieve(&zc, za) == 0;
8860 zap_cursor_advance(&zc)) {
8861 mos_obj_refd(za->za_first_integer);
8862 }
8863 zap_cursor_fini(&zc);
8864 zap_attribute_free(za);
8865 }
8866
8867 static int
dump_mos_leaks(spa_t * spa)8868 dump_mos_leaks(spa_t *spa)
8869 {
8870 int rv = 0;
8871 objset_t *mos = spa->spa_meta_objset;
8872 dsl_pool_t *dp = spa->spa_dsl_pool;
8873
8874 /* Visit and mark all referenced objects in the MOS */
8875
8876 mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT);
8877 mos_obj_refd(spa->spa_pool_props_object);
8878 mos_obj_refd(spa->spa_config_object);
8879 mos_obj_refd(spa->spa_ddt_stat_object);
8880 mos_obj_refd(spa->spa_feat_desc_obj);
8881 mos_obj_refd(spa->spa_feat_enabled_txg_obj);
8882 mos_obj_refd(spa->spa_feat_for_read_obj);
8883 mos_obj_refd(spa->spa_feat_for_write_obj);
8884 mos_obj_refd(spa->spa_history);
8885 mos_obj_refd(spa->spa_errlog_last);
8886 mos_obj_refd(spa->spa_errlog_scrub);
8887
8888 if (spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) {
8889 errorlog_count_refd(mos, spa->spa_errlog_last);
8890 errorlog_count_refd(mos, spa->spa_errlog_scrub);
8891 }
8892
8893 mos_obj_refd(spa->spa_all_vdev_zaps);
8894 mos_obj_refd(spa->spa_dsl_pool->dp_bptree_obj);
8895 mos_obj_refd(spa->spa_dsl_pool->dp_tmp_userrefs_obj);
8896 mos_obj_refd(spa->spa_dsl_pool->dp_scan->scn_phys.scn_queue_obj);
8897 bpobj_count_refd(&spa->spa_deferred_bpobj);
8898 mos_obj_refd(dp->dp_empty_bpobj);
8899 bpobj_count_refd(&dp->dp_obsolete_bpobj);
8900 bpobj_count_refd(&dp->dp_free_bpobj);
8901 mos_obj_refd(spa->spa_l2cache.sav_object);
8902 mos_obj_refd(spa->spa_spares.sav_object);
8903
8904 if (spa->spa_syncing_log_sm != NULL)
8905 mos_obj_refd(spa->spa_syncing_log_sm->sm_object);
8906 mos_leak_log_spacemaps(spa);
8907
8908 mos_obj_refd(spa->spa_condensing_indirect_phys.
8909 scip_next_mapping_object);
8910 mos_obj_refd(spa->spa_condensing_indirect_phys.
8911 scip_prev_obsolete_sm_object);
8912 if (spa->spa_condensing_indirect_phys.scip_next_mapping_object != 0) {
8913 vdev_indirect_mapping_t *vim =
8914 vdev_indirect_mapping_open(mos,
8915 spa->spa_condensing_indirect_phys.scip_next_mapping_object);
8916 mos_obj_refd(vim->vim_phys->vimp_counts_object);
8917 vdev_indirect_mapping_close(vim);
8918 }
8919 deleted_livelists_dump_mos(spa);
8920
8921 if (dp->dp_origin_snap != NULL) {
8922 dsl_dataset_t *ds;
8923
8924 dsl_pool_config_enter(dp, FTAG);
8925 VERIFY0(dsl_dataset_hold_obj(dp,
8926 dsl_dataset_phys(dp->dp_origin_snap)->ds_next_snap_obj,
8927 FTAG, &ds));
8928 count_ds_mos_objects(ds);
8929 dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
8930 dsl_dataset_rele(ds, FTAG);
8931 dsl_pool_config_exit(dp, FTAG);
8932
8933 count_ds_mos_objects(dp->dp_origin_snap);
8934 dump_blkptr_list(&dp->dp_origin_snap->ds_deadlist, "Deadlist");
8935 }
8936 count_dir_mos_objects(dp->dp_mos_dir);
8937 if (dp->dp_free_dir != NULL)
8938 count_dir_mos_objects(dp->dp_free_dir);
8939 if (dp->dp_leak_dir != NULL)
8940 count_dir_mos_objects(dp->dp_leak_dir);
8941
8942 mos_leak_vdev(spa->spa_root_vdev);
8943
8944 for (uint64_t c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
8945 ddt_t *ddt = spa->spa_ddt[c];
8946 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
8947 continue;
8948
8949 /* DDT store objects */
8950 for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
8951 for (ddt_class_t class = 0; class < DDT_CLASSES;
8952 class++) {
8953 mos_obj_refd(ddt->ddt_object[type][class]);
8954 }
8955 }
8956
8957 /* FDT container */
8958 if (ddt->ddt_version == DDT_VERSION_FDT)
8959 mos_obj_refd(ddt->ddt_dir_object);
8960
8961 /* FDT log objects */
8962 if (ddt->ddt_flags & DDT_FLAG_LOG) {
8963 mos_obj_refd(ddt->ddt_log[0].ddl_object);
8964 mos_obj_refd(ddt->ddt_log[1].ddl_object);
8965 }
8966 }
8967
8968 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
8969 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
8970 if (brtvd->bv_initiated) {
8971 mos_obj_refd(brtvd->bv_mos_brtvdev);
8972 mos_obj_refd(brtvd->bv_mos_entries);
8973 }
8974 }
8975
8976 /*
8977 * Visit all allocated objects and make sure they are referenced.
8978 */
8979 uint64_t object = 0;
8980 while (dmu_object_next(mos, &object, B_FALSE, 0) == 0) {
8981 if (zfs_range_tree_contains(mos_refd_objs, object, 1)) {
8982 zfs_range_tree_remove(mos_refd_objs, object, 1);
8983 } else {
8984 dmu_object_info_t doi;
8985 const char *name;
8986 VERIFY0(dmu_object_info(mos, object, &doi));
8987 if (doi.doi_type & DMU_OT_NEWTYPE) {
8988 dmu_object_byteswap_t bswap =
8989 DMU_OT_BYTESWAP(doi.doi_type);
8990 name = dmu_ot_byteswap[bswap].ob_name;
8991 } else {
8992 name = dmu_ot[doi.doi_type].ot_name;
8993 }
8994
8995 (void) printf("MOS object %llu (%s) leaked\n",
8996 (u_longlong_t)object, name);
8997 rv = 2;
8998 }
8999 }
9000 (void) zfs_range_tree_walk(mos_refd_objs, mos_leaks_cb, NULL);
9001 if (!zfs_range_tree_is_empty(mos_refd_objs))
9002 rv = 2;
9003 zfs_range_tree_vacate(mos_refd_objs, NULL, NULL);
9004 zfs_range_tree_destroy(mos_refd_objs);
9005 return (rv);
9006 }
9007
9008 typedef struct log_sm_obsolete_stats_arg {
9009 uint64_t lsos_current_txg;
9010
9011 uint64_t lsos_total_entries;
9012 uint64_t lsos_valid_entries;
9013
9014 uint64_t lsos_sm_entries;
9015 uint64_t lsos_valid_sm_entries;
9016 } log_sm_obsolete_stats_arg_t;
9017
9018 static int
log_spacemap_obsolete_stats_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)9019 log_spacemap_obsolete_stats_cb(spa_t *spa, space_map_entry_t *sme,
9020 uint64_t txg, void *arg)
9021 {
9022 log_sm_obsolete_stats_arg_t *lsos = arg;
9023
9024 uint64_t offset = sme->sme_offset;
9025 uint64_t vdev_id = sme->sme_vdev;
9026
9027 if (lsos->lsos_current_txg == 0) {
9028 /* this is the first log */
9029 lsos->lsos_current_txg = txg;
9030 } else if (lsos->lsos_current_txg < txg) {
9031 /* we just changed log - print stats and reset */
9032 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
9033 (u_longlong_t)lsos->lsos_valid_sm_entries,
9034 (u_longlong_t)lsos->lsos_sm_entries,
9035 (u_longlong_t)lsos->lsos_current_txg);
9036 lsos->lsos_valid_sm_entries = 0;
9037 lsos->lsos_sm_entries = 0;
9038 lsos->lsos_current_txg = txg;
9039 }
9040 ASSERT3U(lsos->lsos_current_txg, ==, txg);
9041
9042 lsos->lsos_sm_entries++;
9043 lsos->lsos_total_entries++;
9044
9045 vdev_t *vd = vdev_lookup_top(spa, vdev_id);
9046 if (!vdev_is_concrete(vd))
9047 return (0);
9048
9049 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
9050 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
9051
9052 if (txg < metaslab_unflushed_txg(ms))
9053 return (0);
9054 lsos->lsos_valid_sm_entries++;
9055 lsos->lsos_valid_entries++;
9056 return (0);
9057 }
9058
9059 static void
dump_log_spacemap_obsolete_stats(spa_t * spa)9060 dump_log_spacemap_obsolete_stats(spa_t *spa)
9061 {
9062 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
9063 return;
9064
9065 log_sm_obsolete_stats_arg_t lsos = {0};
9066
9067 (void) printf("Log Space Map Obsolete Entry Statistics:\n");
9068
9069 iterate_through_spacemap_logs(spa,
9070 log_spacemap_obsolete_stats_cb, &lsos);
9071
9072 /* print stats for latest log */
9073 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
9074 (u_longlong_t)lsos.lsos_valid_sm_entries,
9075 (u_longlong_t)lsos.lsos_sm_entries,
9076 (u_longlong_t)lsos.lsos_current_txg);
9077
9078 (void) printf("%-8llu valid entries out of %-8llu - total\n\n",
9079 (u_longlong_t)lsos.lsos_valid_entries,
9080 (u_longlong_t)lsos.lsos_total_entries);
9081 }
9082
9083 static void
dump_zpool(spa_t * spa)9084 dump_zpool(spa_t *spa)
9085 {
9086 dsl_pool_t *dp = spa_get_dsl(spa);
9087 int rc = 0;
9088
9089 if (dump_opt['y']) {
9090 livelist_metaslab_validate(spa);
9091 }
9092
9093 if (dump_opt['S']) {
9094 dump_simulated_ddt(spa);
9095 return;
9096 }
9097
9098 if (!dump_opt['e'] && dump_opt['C'] > 1) {
9099 (void) printf("\nCached configuration:\n");
9100 dump_nvlist(spa->spa_config, 8);
9101 }
9102
9103 if (dump_opt['C'])
9104 dump_config(spa);
9105
9106 if (dump_opt['u'])
9107 dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n");
9108
9109 if (dump_opt['D'])
9110 dump_all_ddts(spa);
9111
9112 if (dump_opt['T'])
9113 dump_brt(spa);
9114
9115 if (dump_opt['d'] > 2 || dump_opt['m'])
9116 dump_metaslabs(spa);
9117 if (dump_opt['M'])
9118 dump_metaslab_groups(spa, dump_opt['M'] > 1);
9119 if (dump_opt['d'] > 2 || dump_opt['m']) {
9120 dump_log_spacemaps(spa);
9121 dump_log_spacemap_obsolete_stats(spa);
9122 }
9123
9124 if (dump_opt['d'] || dump_opt['i']) {
9125 spa_feature_t f;
9126 mos_refd_objs = zfs_range_tree_create_flags(
9127 NULL, ZFS_RANGE_SEG64, NULL, 0, 0,
9128 0, "dump_zpool:mos_refd_objs");
9129 dump_objset(dp->dp_meta_objset);
9130
9131 if (dump_opt['d'] >= 3) {
9132 dsl_pool_t *dp = spa->spa_dsl_pool;
9133 dump_full_bpobj(&spa->spa_deferred_bpobj,
9134 "Deferred frees", 0);
9135 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
9136 dump_full_bpobj(&dp->dp_free_bpobj,
9137 "Pool snapshot frees", 0);
9138 }
9139 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
9140 ASSERT(spa_feature_is_enabled(spa,
9141 SPA_FEATURE_DEVICE_REMOVAL));
9142 dump_full_bpobj(&dp->dp_obsolete_bpobj,
9143 "Pool obsolete blocks", 0);
9144 }
9145
9146 if (spa_feature_is_active(spa,
9147 SPA_FEATURE_ASYNC_DESTROY)) {
9148 dump_bptree(spa->spa_meta_objset,
9149 dp->dp_bptree_obj,
9150 "Pool dataset frees");
9151 }
9152 dump_dtl(spa->spa_root_vdev, 0);
9153 }
9154
9155 for (spa_feature_t f = 0; f < SPA_FEATURES; f++)
9156 global_feature_count[f] = UINT64_MAX;
9157 global_feature_count[SPA_FEATURE_REDACTION_BOOKMARKS] = 0;
9158 global_feature_count[SPA_FEATURE_REDACTION_LIST_SPILL] = 0;
9159 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN] = 0;
9160 global_feature_count[SPA_FEATURE_LIVELIST] = 0;
9161
9162 (void) dmu_objset_find(spa_name(spa), dump_one_objset,
9163 NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
9164
9165 if (rc == 0 && !dump_opt['L'])
9166 rc = dump_mos_leaks(spa);
9167
9168 for (f = 0; f < SPA_FEATURES; f++) {
9169 uint64_t refcount;
9170
9171 uint64_t *arr;
9172 if (!(spa_feature_table[f].fi_flags &
9173 ZFEATURE_FLAG_PER_DATASET)) {
9174 if (global_feature_count[f] == UINT64_MAX)
9175 continue;
9176 if (!spa_feature_is_enabled(spa, f)) {
9177 ASSERT0(global_feature_count[f]);
9178 continue;
9179 }
9180 arr = global_feature_count;
9181 } else {
9182 if (!spa_feature_is_enabled(spa, f)) {
9183 ASSERT0(dataset_feature_count[f]);
9184 continue;
9185 }
9186 arr = dataset_feature_count;
9187 }
9188 if (feature_get_refcount(spa, &spa_feature_table[f],
9189 &refcount) == ENOTSUP)
9190 continue;
9191 if (arr[f] != refcount) {
9192 (void) printf("%s feature refcount mismatch: "
9193 "%lld consumers != %lld refcount\n",
9194 spa_feature_table[f].fi_uname,
9195 (longlong_t)arr[f], (longlong_t)refcount);
9196 rc = 2;
9197 } else {
9198 (void) printf("Verified %s feature refcount "
9199 "of %llu is correct\n",
9200 spa_feature_table[f].fi_uname,
9201 (longlong_t)refcount);
9202 }
9203 }
9204
9205 if (rc == 0)
9206 rc = verify_device_removal_feature_counts(spa);
9207 }
9208
9209 if (rc == 0 && (dump_opt['b'] || dump_opt['c']))
9210 rc = dump_block_stats(spa);
9211
9212 if (rc == 0)
9213 rc = verify_spacemap_refcounts(spa);
9214
9215 if (dump_opt['s'])
9216 show_pool_stats(spa);
9217
9218 if (dump_opt['h'])
9219 dump_history(spa);
9220
9221 if (rc == 0)
9222 rc = verify_checkpoint(spa);
9223
9224 if (rc != 0) {
9225 dump_debug_buffer();
9226 zdb_exit(rc);
9227 }
9228 }
9229
9230 #define ZDB_FLAG_CHECKSUM 0x0001
9231 #define ZDB_FLAG_DECOMPRESS 0x0002
9232 #define ZDB_FLAG_BSWAP 0x0004
9233 #define ZDB_FLAG_GBH 0x0008
9234 #define ZDB_FLAG_INDIRECT 0x0010
9235 #define ZDB_FLAG_RAW 0x0020
9236 #define ZDB_FLAG_PRINT_BLKPTR 0x0040
9237 #define ZDB_FLAG_VERBOSE 0x0080
9238
9239 static int flagbits[256];
9240 static char flagbitstr[16];
9241
9242 static void
zdb_print_blkptr(const blkptr_t * bp,int flags)9243 zdb_print_blkptr(const blkptr_t *bp, int flags)
9244 {
9245 char blkbuf[BP_SPRINTF_LEN];
9246
9247 if (flags & ZDB_FLAG_BSWAP)
9248 byteswap_uint64_array((void *)bp, sizeof (blkptr_t));
9249
9250 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
9251 (void) printf("%s\n", blkbuf);
9252 }
9253
9254 static void
zdb_dump_indirect(blkptr_t * bp,int nbps,int flags)9255 zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
9256 {
9257 int i;
9258
9259 for (i = 0; i < nbps; i++)
9260 zdb_print_blkptr(&bp[i], flags);
9261 }
9262
9263 static void
zdb_dump_gbh(void * buf,uint64_t size,int flags)9264 zdb_dump_gbh(void *buf, uint64_t size, int flags)
9265 {
9266 zdb_dump_indirect((blkptr_t *)buf, gbh_nblkptrs(size), flags);
9267 }
9268
9269 static void
zdb_dump_block_raw(void * buf,uint64_t size,int flags)9270 zdb_dump_block_raw(void *buf, uint64_t size, int flags)
9271 {
9272 if (flags & ZDB_FLAG_BSWAP)
9273 byteswap_uint64_array(buf, size);
9274 VERIFY(write(fileno(stdout), buf, size) == size);
9275 }
9276
9277 static void
zdb_dump_block(char * label,void * buf,uint64_t size,int flags)9278 zdb_dump_block(char *label, void *buf, uint64_t size, int flags)
9279 {
9280 uint64_t *d = (uint64_t *)buf;
9281 unsigned nwords = size / sizeof (uint64_t);
9282 int do_bswap = !!(flags & ZDB_FLAG_BSWAP);
9283 unsigned i, j;
9284 const char *hdr;
9285 char *c;
9286
9287
9288 if (do_bswap)
9289 hdr = " 7 6 5 4 3 2 1 0 f e d c b a 9 8";
9290 else
9291 hdr = " 0 1 2 3 4 5 6 7 8 9 a b c d e f";
9292
9293 (void) printf("\n%s\n%6s %s 0123456789abcdef\n", label, "", hdr);
9294
9295 #ifdef _ZFS_LITTLE_ENDIAN
9296 /* correct the endianness */
9297 do_bswap = !do_bswap;
9298 #endif
9299 for (i = 0; i < nwords; i += 2) {
9300 (void) printf("%06llx: %016llx %016llx ",
9301 (u_longlong_t)(i * sizeof (uint64_t)),
9302 (u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]),
9303 (u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1]));
9304
9305 c = (char *)&d[i];
9306 for (j = 0; j < 2 * sizeof (uint64_t); j++)
9307 (void) printf("%c", isprint(c[j]) ? c[j] : '.');
9308 (void) printf("\n");
9309 }
9310 }
9311
9312 /*
9313 * There are two acceptable formats:
9314 * leaf_name - For example: c1t0d0 or /tmp/ztest.0a
9315 * child[.child]* - For example: 0.1.1
9316 *
9317 * The second form can be used to specify arbitrary vdevs anywhere
9318 * in the hierarchy. For example, in a pool with a mirror of
9319 * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
9320 */
9321 static vdev_t *
zdb_vdev_lookup(vdev_t * vdev,const char * path)9322 zdb_vdev_lookup(vdev_t *vdev, const char *path)
9323 {
9324 char *s, *p, *q;
9325 unsigned i;
9326
9327 if (vdev == NULL)
9328 return (NULL);
9329
9330 /* First, assume the x.x.x.x format */
9331 i = strtoul(path, &s, 10);
9332 if (s == path || (s && *s != '.' && *s != '\0'))
9333 goto name;
9334 if (i >= vdev->vdev_children)
9335 return (NULL);
9336
9337 vdev = vdev->vdev_child[i];
9338 if (s && *s == '\0')
9339 return (vdev);
9340 return (zdb_vdev_lookup(vdev, s+1));
9341
9342 name:
9343 for (i = 0; i < vdev->vdev_children; i++) {
9344 vdev_t *vc = vdev->vdev_child[i];
9345
9346 if (vc->vdev_path == NULL) {
9347 vc = zdb_vdev_lookup(vc, path);
9348 if (vc == NULL)
9349 continue;
9350 else
9351 return (vc);
9352 }
9353
9354 p = strrchr(vc->vdev_path, '/');
9355 p = p ? p + 1 : vc->vdev_path;
9356 q = &vc->vdev_path[strlen(vc->vdev_path) - 2];
9357
9358 if (strcmp(vc->vdev_path, path) == 0)
9359 return (vc);
9360 if (strcmp(p, path) == 0)
9361 return (vc);
9362 if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0)
9363 return (vc);
9364 }
9365
9366 return (NULL);
9367 }
9368
9369 static int
name_from_objset_id(spa_t * spa,uint64_t objset_id,char * outstr)9370 name_from_objset_id(spa_t *spa, uint64_t objset_id, char *outstr)
9371 {
9372 dsl_dataset_t *ds;
9373
9374 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
9375 int error = dsl_dataset_hold_obj(spa->spa_dsl_pool, objset_id,
9376 NULL, &ds);
9377 if (error != 0) {
9378 (void) fprintf(stderr, "failed to hold objset %llu: %s\n",
9379 (u_longlong_t)objset_id, strerror(error));
9380 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
9381 return (error);
9382 }
9383 dsl_dataset_name(ds, outstr);
9384 dsl_dataset_rele(ds, NULL);
9385 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
9386 return (0);
9387 }
9388
9389 static boolean_t
zdb_parse_block_sizes(char * sizes,uint64_t * lsize,uint64_t * psize)9390 zdb_parse_block_sizes(char *sizes, uint64_t *lsize, uint64_t *psize)
9391 {
9392 char *s0, *s1, *tmp = NULL;
9393
9394 if (sizes == NULL)
9395 return (B_FALSE);
9396
9397 s0 = strtok_r(sizes, "/", &tmp);
9398 if (s0 == NULL)
9399 return (B_FALSE);
9400 s1 = strtok_r(NULL, "/", &tmp);
9401 *lsize = strtoull(s0, NULL, 16);
9402 *psize = s1 ? strtoull(s1, NULL, 16) : *lsize;
9403 return (*lsize >= *psize && *psize > 0);
9404 }
9405
9406 #define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg))
9407
9408 static boolean_t
try_decompress_block(abd_t * pabd,uint64_t lsize,uint64_t psize,int flags,int cfunc,void * lbuf,void * lbuf2)9409 try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize,
9410 int flags, int cfunc, void *lbuf, void *lbuf2)
9411 {
9412 if (flags & ZDB_FLAG_VERBOSE) {
9413 (void) fprintf(stderr,
9414 "Trying %05llx -> %05llx (%s)\n",
9415 (u_longlong_t)psize,
9416 (u_longlong_t)lsize,
9417 zio_compress_table[cfunc].ci_name);
9418 }
9419
9420 /*
9421 * We set lbuf to all zeros and lbuf2 to all
9422 * ones, then decompress to both buffers and
9423 * compare their contents. This way we can
9424 * know if decompression filled exactly to
9425 * lsize or if it left some bytes unwritten.
9426 */
9427
9428 memset(lbuf, 0x00, lsize);
9429 memset(lbuf2, 0xff, lsize);
9430
9431 abd_t labd, labd2;
9432 abd_get_from_buf_struct(&labd, lbuf, lsize);
9433 abd_get_from_buf_struct(&labd2, lbuf2, lsize);
9434
9435 boolean_t ret = B_FALSE;
9436 if (zio_decompress_data(cfunc, pabd,
9437 &labd, psize, lsize, NULL) == 0 &&
9438 zio_decompress_data(cfunc, pabd,
9439 &labd2, psize, lsize, NULL) == 0 &&
9440 memcmp(lbuf, lbuf2, lsize) == 0)
9441 ret = B_TRUE;
9442
9443 abd_free(&labd2);
9444 abd_free(&labd);
9445
9446 return (ret);
9447 }
9448
9449 static uint64_t
zdb_decompress_block(abd_t * pabd,void * buf,void * lbuf,uint64_t lsize,uint64_t psize,int flags)9450 zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
9451 uint64_t psize, int flags)
9452 {
9453 (void) buf;
9454 uint64_t orig_lsize = lsize;
9455 boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL));
9456 /*
9457 * We don't know how the data was compressed, so just try
9458 * every decompress function at every inflated blocksize.
9459 */
9460 void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
9461 int cfuncs[ZIO_COMPRESS_FUNCTIONS] = { 0 };
9462 int *cfuncp = cfuncs;
9463 uint64_t maxlsize = SPA_MAXBLOCKSIZE;
9464 uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) |
9465 ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) |
9466 ZIO_COMPRESS_MASK(ZLE);
9467 *cfuncp++ = ZIO_COMPRESS_LZ4;
9468 *cfuncp++ = ZIO_COMPRESS_LZJB;
9469 mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB);
9470 /*
9471 * Every gzip level has the same decompressor, no need to
9472 * run it 9 times per bruteforce attempt.
9473 */
9474 mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3);
9475 mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5);
9476 mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7);
9477 mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9);
9478 for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++)
9479 if (((1ULL << c) & mask) == 0)
9480 *cfuncp++ = c;
9481
9482 /*
9483 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
9484 * could take a while and we should let the user know
9485 * we are not stuck. On the other hand, printing progress
9486 * info gets old after a while. User can specify 'v' flag
9487 * to see the progression.
9488 */
9489 if (lsize == psize)
9490 lsize += SPA_MINBLOCKSIZE;
9491 else
9492 maxlsize = lsize;
9493
9494 for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) {
9495 for (cfuncp = cfuncs; *cfuncp; cfuncp++) {
9496 if (try_decompress_block(pabd, lsize, psize, flags,
9497 *cfuncp, lbuf, lbuf2)) {
9498 tryzle = B_FALSE;
9499 break;
9500 }
9501 }
9502 if (*cfuncp != 0)
9503 break;
9504 }
9505 if (tryzle) {
9506 for (lsize = orig_lsize; lsize <= maxlsize;
9507 lsize += SPA_MINBLOCKSIZE) {
9508 if (try_decompress_block(pabd, lsize, psize, flags,
9509 ZIO_COMPRESS_ZLE, lbuf, lbuf2)) {
9510 *cfuncp = ZIO_COMPRESS_ZLE;
9511 break;
9512 }
9513 }
9514 }
9515 umem_free(lbuf2, SPA_MAXBLOCKSIZE);
9516
9517 if (*cfuncp == ZIO_COMPRESS_ZLE) {
9518 printf("\nZLE decompression was selected. If you "
9519 "suspect the results are wrong,\ntry avoiding ZLE "
9520 "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
9521 }
9522
9523 return (lsize > maxlsize ? -1 : lsize);
9524 }
9525
9526 /*
9527 * Read a block from a pool and print it out. The syntax of the
9528 * block descriptor is:
9529 *
9530 * pool:vdev_specifier:offset:[lsize/]psize[:flags]
9531 *
9532 * pool - The name of the pool you wish to read from
9533 * vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
9534 * offset - offset, in hex, in bytes
9535 * size - Amount of data to read, in hex, in bytes
9536 * flags - A string of characters specifying options
9537 * b: Decode a blkptr at given offset within block
9538 * c: Calculate and display checksums
9539 * d: Decompress data before dumping
9540 * e: Byteswap data before dumping
9541 * g: Display data as a gang block header
9542 * i: Display as an indirect block
9543 * r: Dump raw data to stdout
9544 * v: Verbose
9545 *
9546 */
9547 static void
zdb_read_block(char * thing,spa_t * spa)9548 zdb_read_block(char *thing, spa_t *spa)
9549 {
9550 blkptr_t blk, *bp = &blk;
9551 dva_t *dva = bp->blk_dva;
9552 int flags = 0;
9553 uint64_t offset = 0, psize = 0, lsize = 0, blkptr_offset = 0;
9554 zio_t *zio;
9555 vdev_t *vd;
9556 abd_t *pabd;
9557 void *lbuf, *buf;
9558 char *s, *p, *dup, *flagstr, *sizes, *tmp = NULL;
9559 const char *vdev, *errmsg = NULL;
9560 int i, len, error;
9561 boolean_t borrowed = B_FALSE, found = B_FALSE;
9562
9563 dup = strdup(thing);
9564 s = strtok_r(dup, ":", &tmp);
9565 vdev = s ?: "";
9566 s = strtok_r(NULL, ":", &tmp);
9567 offset = strtoull(s ? s : "", NULL, 16);
9568 sizes = strtok_r(NULL, ":", &tmp);
9569 s = strtok_r(NULL, ":", &tmp);
9570 flagstr = strdup(s ?: "");
9571
9572 if (!zdb_parse_block_sizes(sizes, &lsize, &psize))
9573 errmsg = "invalid size(s)";
9574 if (!IS_P2ALIGNED(psize, DEV_BSIZE) || !IS_P2ALIGNED(lsize, DEV_BSIZE))
9575 errmsg = "size must be a multiple of sector size";
9576 if (!IS_P2ALIGNED(offset, DEV_BSIZE))
9577 errmsg = "offset must be a multiple of sector size";
9578 if (errmsg) {
9579 (void) printf("Invalid block specifier: %s - %s\n",
9580 thing, errmsg);
9581 goto done;
9582 }
9583
9584 tmp = NULL;
9585 for (s = strtok_r(flagstr, ":", &tmp);
9586 s != NULL;
9587 s = strtok_r(NULL, ":", &tmp)) {
9588 len = strlen(flagstr);
9589 for (i = 0; i < len; i++) {
9590 int bit = flagbits[(uchar_t)flagstr[i]];
9591
9592 if (bit == 0) {
9593 (void) printf("***Ignoring flag: %c\n",
9594 (uchar_t)flagstr[i]);
9595 continue;
9596 }
9597 found = B_TRUE;
9598 flags |= bit;
9599
9600 p = &flagstr[i + 1];
9601 if (*p != ':' && *p != '\0') {
9602 int j = 0, nextbit = flagbits[(uchar_t)*p];
9603 char *end, offstr[8] = { 0 };
9604 if ((bit == ZDB_FLAG_PRINT_BLKPTR) &&
9605 (nextbit == 0)) {
9606 /* look ahead to isolate the offset */
9607 while (nextbit == 0 &&
9608 strchr(flagbitstr, *p) == NULL) {
9609 offstr[j] = *p;
9610 j++;
9611 if (i + j > strlen(flagstr))
9612 break;
9613 p++;
9614 nextbit = flagbits[(uchar_t)*p];
9615 }
9616 blkptr_offset = strtoull(offstr, &end,
9617 16);
9618 i += j;
9619 } else if (nextbit == 0) {
9620 (void) printf("***Ignoring flag arg:"
9621 " '%c'\n", (uchar_t)*p);
9622 }
9623 }
9624 }
9625 }
9626 if (blkptr_offset % sizeof (blkptr_t)) {
9627 printf("Block pointer offset 0x%llx "
9628 "must be divisible by 0x%x\n",
9629 (longlong_t)blkptr_offset, (int)sizeof (blkptr_t));
9630 goto done;
9631 }
9632 if (found == B_FALSE && strlen(flagstr) > 0) {
9633 printf("Invalid flag arg: '%s'\n", flagstr);
9634 goto done;
9635 }
9636
9637 vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev);
9638 if (vd == NULL) {
9639 (void) printf("***Invalid vdev: %s\n", vdev);
9640 goto done;
9641 } else {
9642 if (vd->vdev_path)
9643 (void) fprintf(stderr, "Found vdev: %s\n",
9644 vd->vdev_path);
9645 else
9646 (void) fprintf(stderr, "Found vdev type: %s\n",
9647 vd->vdev_ops->vdev_op_type);
9648 }
9649
9650 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
9651 lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
9652
9653 BP_ZERO(bp);
9654
9655 DVA_SET_VDEV(&dva[0], vd->vdev_id);
9656 DVA_SET_OFFSET(&dva[0], offset);
9657 DVA_SET_GANG(&dva[0], 0);
9658 DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
9659
9660 BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
9661
9662 BP_SET_LSIZE(bp, lsize);
9663 BP_SET_PSIZE(bp, psize);
9664 BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF);
9665 BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF);
9666 BP_SET_TYPE(bp, DMU_OT_NONE);
9667 BP_SET_LEVEL(bp, 0);
9668 BP_SET_DEDUP(bp, 0);
9669 BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
9670
9671 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9672 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9673
9674 if (vd == vd->vdev_top) {
9675 /*
9676 * Treat this as a normal block read.
9677 */
9678 zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL,
9679 ZIO_PRIORITY_SYNC_READ,
9680 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL));
9681 } else {
9682 /*
9683 * Treat this as a vdev child I/O.
9684 */
9685 zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd,
9686 psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ,
9687 ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
9688 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | ZIO_FLAG_OPTIONAL,
9689 NULL, NULL));
9690 }
9691
9692 error = zio_wait(zio);
9693 spa_config_exit(spa, SCL_STATE, FTAG);
9694
9695 if (error) {
9696 (void) printf("Read of %s failed, error: %d\n", thing, error);
9697 goto out;
9698 }
9699
9700 uint64_t orig_lsize = lsize;
9701 buf = lbuf;
9702 if (flags & ZDB_FLAG_DECOMPRESS) {
9703 lsize = zdb_decompress_block(pabd, buf, lbuf,
9704 lsize, psize, flags);
9705 if (lsize == -1) {
9706 (void) printf("Decompress of %s failed\n", thing);
9707 goto out;
9708 }
9709 } else {
9710 buf = abd_borrow_buf_copy(pabd, lsize);
9711 borrowed = B_TRUE;
9712 }
9713 /*
9714 * Try to detect invalid block pointer. If invalid, try
9715 * decompressing.
9716 */
9717 if ((flags & ZDB_FLAG_PRINT_BLKPTR || flags & ZDB_FLAG_INDIRECT) &&
9718 !(flags & ZDB_FLAG_DECOMPRESS)) {
9719 const blkptr_t *b = (const blkptr_t *)(void *)
9720 ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9721 if (zfs_blkptr_verify(spa, b,
9722 BLK_CONFIG_NEEDED, BLK_VERIFY_ONLY)) {
9723 abd_return_buf_copy(pabd, buf, lsize);
9724 borrowed = B_FALSE;
9725 buf = lbuf;
9726 lsize = zdb_decompress_block(pabd, buf,
9727 lbuf, lsize, psize, flags);
9728 b = (const blkptr_t *)(void *)
9729 ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9730 if (lsize == -1 || zfs_blkptr_verify(spa, b,
9731 BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) {
9732 printf("invalid block pointer at this DVA\n");
9733 goto out;
9734 }
9735 }
9736 }
9737
9738 if (flags & ZDB_FLAG_PRINT_BLKPTR)
9739 zdb_print_blkptr((blkptr_t *)(void *)
9740 ((uintptr_t)buf + (uintptr_t)blkptr_offset), flags);
9741 else if (flags & ZDB_FLAG_RAW)
9742 zdb_dump_block_raw(buf, lsize, flags);
9743 else if (flags & ZDB_FLAG_INDIRECT)
9744 zdb_dump_indirect((blkptr_t *)buf,
9745 orig_lsize / sizeof (blkptr_t), flags);
9746 else if (flags & ZDB_FLAG_GBH)
9747 zdb_dump_gbh(buf, lsize, flags);
9748 else
9749 zdb_dump_block(thing, buf, lsize, flags);
9750
9751 /*
9752 * If :c was specified, iterate through the checksum table to
9753 * calculate and display each checksum for our specified
9754 * DVA and length.
9755 */
9756 if ((flags & ZDB_FLAG_CHECKSUM) && !(flags & ZDB_FLAG_RAW) &&
9757 !(flags & ZDB_FLAG_GBH)) {
9758 zio_t *czio;
9759 (void) printf("\n");
9760 for (enum zio_checksum ck = ZIO_CHECKSUM_LABEL;
9761 ck < ZIO_CHECKSUM_FUNCTIONS; ck++) {
9762
9763 if ((zio_checksum_table[ck].ci_flags &
9764 ZCHECKSUM_FLAG_EMBEDDED) ||
9765 ck == ZIO_CHECKSUM_NOPARITY) {
9766 continue;
9767 }
9768 BP_SET_CHECKSUM(bp, ck);
9769 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9770 czio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9771 if (vd == vd->vdev_top) {
9772 zio_nowait(zio_read(czio, spa, bp, pabd, psize,
9773 NULL, NULL,
9774 ZIO_PRIORITY_SYNC_READ,
9775 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9776 ZIO_FLAG_DONT_RETRY, NULL));
9777 } else {
9778 zio_nowait(zio_vdev_child_io(czio, bp, vd,
9779 offset, pabd, psize, ZIO_TYPE_READ,
9780 ZIO_PRIORITY_SYNC_READ,
9781 ZIO_FLAG_DONT_PROPAGATE |
9782 ZIO_FLAG_DONT_RETRY |
9783 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9784 ZIO_FLAG_SPECULATIVE |
9785 ZIO_FLAG_OPTIONAL, NULL, NULL));
9786 }
9787 error = zio_wait(czio);
9788 if (error == 0 || error == ECKSUM) {
9789 zio_t *ck_zio = zio_null(NULL, spa, NULL,
9790 NULL, NULL, 0);
9791 ck_zio->io_offset =
9792 DVA_GET_OFFSET(&bp->blk_dva[0]);
9793 ck_zio->io_bp = bp;
9794 zio_checksum_compute(ck_zio, ck, pabd, psize);
9795 printf(
9796 "%12s\t"
9797 "cksum=%016llx:%016llx:%016llx:%016llx\n",
9798 zio_checksum_table[ck].ci_name,
9799 (u_longlong_t)bp->blk_cksum.zc_word[0],
9800 (u_longlong_t)bp->blk_cksum.zc_word[1],
9801 (u_longlong_t)bp->blk_cksum.zc_word[2],
9802 (u_longlong_t)bp->blk_cksum.zc_word[3]);
9803 zio_wait(ck_zio);
9804 } else {
9805 printf("error %d reading block\n", error);
9806 }
9807 spa_config_exit(spa, SCL_STATE, FTAG);
9808 }
9809 }
9810
9811 if (borrowed)
9812 abd_return_buf_copy(pabd, buf, lsize);
9813
9814 out:
9815 abd_free(pabd);
9816 umem_free(lbuf, SPA_MAXBLOCKSIZE);
9817 done:
9818 free(flagstr);
9819 free(dup);
9820 }
9821
9822 static void
zdb_embedded_block(char * thing)9823 zdb_embedded_block(char *thing)
9824 {
9825 blkptr_t bp = {{{{0}}}};
9826 unsigned long long *words = (void *)&bp;
9827 char *buf;
9828 int err;
9829
9830 err = sscanf(thing, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
9831 "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
9832 words + 0, words + 1, words + 2, words + 3,
9833 words + 4, words + 5, words + 6, words + 7,
9834 words + 8, words + 9, words + 10, words + 11,
9835 words + 12, words + 13, words + 14, words + 15);
9836 if (err != 16) {
9837 (void) fprintf(stderr, "invalid input format\n");
9838 zdb_exit(1);
9839 }
9840 ASSERT3U(BPE_GET_LSIZE(&bp), <=, SPA_MAXBLOCKSIZE);
9841 buf = malloc(SPA_MAXBLOCKSIZE);
9842 if (buf == NULL) {
9843 (void) fprintf(stderr, "out of memory\n");
9844 zdb_exit(1);
9845 }
9846 err = decode_embedded_bp(&bp, buf, BPE_GET_LSIZE(&bp));
9847 if (err != 0) {
9848 (void) fprintf(stderr, "decode failed: %u\n", err);
9849 zdb_exit(1);
9850 }
9851 zdb_dump_block_raw(buf, BPE_GET_LSIZE(&bp), 0);
9852 free(buf);
9853 }
9854
9855 /* check for valid hex or decimal numeric string */
9856 static boolean_t
zdb_numeric(char * str)9857 zdb_numeric(char *str)
9858 {
9859 int i = 0, len;
9860
9861 len = strlen(str);
9862 if (len == 0)
9863 return (B_FALSE);
9864 if (strncmp(str, "0x", 2) == 0 || strncmp(str, "0X", 2) == 0)
9865 i = 2;
9866 for (; i < len; i++) {
9867 if (!isxdigit(str[i]))
9868 return (B_FALSE);
9869 }
9870 return (B_TRUE);
9871 }
9872
9873 static int
dummy_get_file_info(dmu_object_type_t bonustype,const void * data,zfs_file_info_t * zoi)9874 dummy_get_file_info(dmu_object_type_t bonustype, const void *data,
9875 zfs_file_info_t *zoi)
9876 {
9877 (void) data, (void) zoi;
9878
9879 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
9880 return (ENOENT);
9881
9882 (void) fprintf(stderr, "dummy_get_file_info: not implemented");
9883 abort();
9884 }
9885
9886 int
main(int argc,char ** argv)9887 main(int argc, char **argv)
9888 {
9889 int c;
9890 int dump_all = 1;
9891 int verbose = 0;
9892 int error = 0;
9893 char **searchdirs = NULL;
9894 int nsearch = 0;
9895 char *target, *target_pool, dsname[ZFS_MAX_DATASET_NAME_LEN];
9896 nvlist_t *policy = NULL;
9897 uint64_t max_txg = UINT64_MAX;
9898 int64_t objset_id = -1;
9899 uint64_t object;
9900 int flags = ZFS_IMPORT_MISSING_LOG;
9901 int rewind = ZPOOL_NEVER_REWIND;
9902 char *spa_config_path_env, *objset_str;
9903 boolean_t target_is_spa = B_TRUE, dataset_lookup = B_FALSE;
9904 nvlist_t *cfg = NULL;
9905 struct sigaction action;
9906 boolean_t force_import = B_FALSE;
9907 boolean_t config_path_console = B_FALSE;
9908 char pbuf[MAXPATHLEN];
9909
9910 dprintf_setup(&argc, argv);
9911
9912 /*
9913 * Set up signal handlers, so if we crash due to bad on-disk data we
9914 * can get more info. Unlike ztest, we don't bail out if we can't set
9915 * up signal handlers, because zdb is very useful without them.
9916 */
9917 action.sa_handler = sig_handler;
9918 sigemptyset(&action.sa_mask);
9919 action.sa_flags = 0;
9920 if (sigaction(SIGSEGV, &action, NULL) < 0) {
9921 (void) fprintf(stderr, "zdb: cannot catch SIGSEGV: %s\n",
9922 strerror(errno));
9923 }
9924 if (sigaction(SIGABRT, &action, NULL) < 0) {
9925 (void) fprintf(stderr, "zdb: cannot catch SIGABRT: %s\n",
9926 strerror(errno));
9927 }
9928
9929 /*
9930 * If there is an environment variable SPA_CONFIG_PATH it overrides
9931 * default spa_config_path setting. If -U flag is specified it will
9932 * override this environment variable settings once again.
9933 */
9934 spa_config_path_env = getenv("SPA_CONFIG_PATH");
9935 if (spa_config_path_env != NULL)
9936 spa_config_path = spa_config_path_env;
9937
9938 /*
9939 * For performance reasons, we set this tunable down. We do so before
9940 * the arg parsing section so that the user can override this value if
9941 * they choose.
9942 */
9943 zfs_btree_verify_intensity = 3;
9944
9945 struct option long_options[] = {
9946 {"ignore-assertions", no_argument, NULL, 'A'},
9947 {"block-stats", no_argument, NULL, 'b'},
9948 {"backup", no_argument, NULL, 'B'},
9949 {"checksum", no_argument, NULL, 'c'},
9950 {"config", no_argument, NULL, 'C'},
9951 {"datasets", no_argument, NULL, 'd'},
9952 {"dedup-stats", no_argument, NULL, 'D'},
9953 {"exported", no_argument, NULL, 'e'},
9954 {"embedded-block-pointer", no_argument, NULL, 'E'},
9955 {"file-layout", no_argument, NULL, 'f'},
9956 {"automatic-rewind", no_argument, NULL, 'F'},
9957 {"dump-debug-msg", no_argument, NULL, 'G'},
9958 {"history", no_argument, NULL, 'h'},
9959 {"scripting-mode", no_argument, NULL, 'H'},
9960 {"intent-logs", no_argument, NULL, 'i'},
9961 {"inflight", required_argument, NULL, 'I'},
9962 {"checkpointed-state", no_argument, NULL, 'k'},
9963 {"key", required_argument, NULL, 'K'},
9964 {"label", no_argument, NULL, 'l'},
9965 {"disable-leak-tracking", no_argument, NULL, 'L'},
9966 {"metaslabs", no_argument, NULL, 'm'},
9967 {"metaslab-groups", no_argument, NULL, 'M'},
9968 {"numeric", no_argument, NULL, 'N'},
9969 {"option", required_argument, NULL, 'o'},
9970 {"object-lookups", no_argument, NULL, 'O'},
9971 {"path", required_argument, NULL, 'p'},
9972 {"parseable", no_argument, NULL, 'P'},
9973 {"skip-label", no_argument, NULL, 'q'},
9974 {"copy-object", no_argument, NULL, 'r'},
9975 {"read-block", no_argument, NULL, 'R'},
9976 {"io-stats", no_argument, NULL, 's'},
9977 {"simulate-dedup", no_argument, NULL, 'S'},
9978 {"txg", required_argument, NULL, 't'},
9979 {"brt-stats", no_argument, NULL, 'T'},
9980 {"uberblock", no_argument, NULL, 'u'},
9981 {"cachefile", required_argument, NULL, 'U'},
9982 {"verbose", no_argument, NULL, 'v'},
9983 {"verbatim", no_argument, NULL, 'V'},
9984 {"dump-blocks", required_argument, NULL, 'x'},
9985 {"extreme-rewind", no_argument, NULL, 'X'},
9986 {"all-reconstruction", no_argument, NULL, 'Y'},
9987 {"livelist", no_argument, NULL, 'y'},
9988 {"zstd-headers", no_argument, NULL, 'Z'},
9989 {"allocated-map", no_argument, NULL,
9990 ARG_ALLOCATED},
9991 {"bin", required_argument, NULL,
9992 ARG_BLOCK_BIN_MODE},
9993 {"class", required_argument, NULL,
9994 ARG_BLOCK_CLASSES},
9995 {0, 0, 0, 0}
9996 };
9997
9998 while ((c = getopt_long(argc, argv,
9999 "AbBcCdDeEfFGhHiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
10000 long_options, NULL)) != -1) {
10001 switch (c) {
10002 case 'b':
10003 case 'B':
10004 case 'c':
10005 case 'C':
10006 case 'd':
10007 case 'D':
10008 case 'E':
10009 case 'f':
10010 case 'G':
10011 case 'h':
10012 case 'i':
10013 case 'l':
10014 case 'm':
10015 case 'M':
10016 case 'N':
10017 case 'O':
10018 case 'r':
10019 case 'R':
10020 case 's':
10021 case 'S':
10022 case 'T':
10023 case 'u':
10024 case 'y':
10025 case 'Z':
10026 case ARG_ALLOCATED:
10027 dump_opt[c]++;
10028 dump_all = 0;
10029 break;
10030 case 'A':
10031 case 'e':
10032 case 'F':
10033 case 'H':
10034 case 'k':
10035 case 'L':
10036 case 'P':
10037 case 'q':
10038 case 'X':
10039 dump_opt[c]++;
10040 break;
10041 case 'Y':
10042 zfs_reconstruct_indirect_combinations_max = INT_MAX;
10043 zfs_deadman_enabled = 0;
10044 break;
10045 /* NB: Sort single match options below. */
10046 case 'I':
10047 max_inflight_bytes = strtoull(optarg, NULL, 0);
10048 if (max_inflight_bytes == 0) {
10049 (void) fprintf(stderr, "maximum number "
10050 "of inflight bytes must be greater "
10051 "than 0\n");
10052 usage();
10053 }
10054 break;
10055 case 'K':
10056 dump_opt[c]++;
10057 key_material = strdup(optarg);
10058 /* redact key material in process table */
10059 while (*optarg != '\0') { *optarg++ = '*'; }
10060 break;
10061 case 'o':
10062 dump_opt[c]++;
10063 dump_all = 0;
10064 error = handle_tunable_option(optarg, B_FALSE);
10065 if (error != 0)
10066 zdb_exit(1);
10067 break;
10068 case 'p':
10069 if (searchdirs == NULL) {
10070 searchdirs = umem_alloc(sizeof (char *),
10071 UMEM_NOFAIL);
10072 } else {
10073 char **tmp = umem_alloc((nsearch + 1) *
10074 sizeof (char *), UMEM_NOFAIL);
10075 memcpy(tmp, searchdirs, nsearch *
10076 sizeof (char *));
10077 umem_free(searchdirs,
10078 nsearch * sizeof (char *));
10079 searchdirs = tmp;
10080 }
10081 searchdirs[nsearch++] = optarg;
10082 break;
10083 case 't':
10084 max_txg = strtoull(optarg, NULL, 0);
10085 if (max_txg < TXG_INITIAL) {
10086 (void) fprintf(stderr, "incorrect txg "
10087 "specified: %s\n", optarg);
10088 usage();
10089 }
10090 break;
10091 case 'U':
10092 config_path_console = B_TRUE;
10093 spa_config_path = optarg;
10094 if (spa_config_path[0] != '/') {
10095 (void) fprintf(stderr,
10096 "cachefile must be an absolute path "
10097 "(i.e. start with a slash)\n");
10098 usage();
10099 }
10100 break;
10101 case 'v':
10102 verbose++;
10103 break;
10104 case 'V':
10105 flags = ZFS_IMPORT_VERBATIM;
10106 break;
10107 case 'x':
10108 vn_dumpdir = optarg;
10109 break;
10110 case ARG_BLOCK_BIN_MODE:
10111 if (strcmp(optarg, "lsize") == 0) {
10112 block_bin_mode = BIN_LSIZE;
10113 } else if (strcmp(optarg, "psize") == 0) {
10114 block_bin_mode = BIN_PSIZE;
10115 } else if (strcmp(optarg, "asize") == 0) {
10116 block_bin_mode = BIN_ASIZE;
10117 } else {
10118 (void) fprintf(stderr,
10119 "--bin=\"%s\" must be one of \"lsize\", "
10120 "\"psize\" or \"asize\"\n", optarg);
10121 usage();
10122 }
10123 break;
10124
10125 case ARG_BLOCK_CLASSES: {
10126 char *buf = strdup(optarg), *tok = buf, *next,
10127 *save = NULL;
10128
10129 while ((next = strtok_r(tok, ",", &save)) != NULL) {
10130 tok = NULL;
10131
10132 if (strcmp(next, "normal") == 0) {
10133 block_classes |= CLASS_NORMAL;
10134 } else if (strcmp(next, "special") == 0) {
10135 block_classes |= CLASS_SPECIAL;
10136 } else if (strcmp(next, "dedup") == 0) {
10137 block_classes |= CLASS_DEDUP;
10138 } else if (strcmp(next, "other") == 0) {
10139 block_classes |= CLASS_OTHER;
10140 } else {
10141 (void) fprintf(stderr,
10142 "--class=\"%s\" must be a "
10143 "comma-separated list of either "
10144 "\"normal\", \"special\", "
10145 "\"asize\" or \"other\"; "
10146 "got \"%s\"\n",
10147 optarg, next);
10148 usage();
10149 }
10150 }
10151
10152 if (block_classes == 0) {
10153 (void) fprintf(stderr,
10154 "--class= must be a comma-separated "
10155 "list of either \"normal\", \"special\", "
10156 "\"asize\" or \"other\"; got empty\n");
10157 usage();
10158 }
10159
10160 free(buf);
10161 break;
10162 }
10163 default:
10164 usage();
10165 break;
10166 }
10167 }
10168
10169 if (!dump_opt['e'] && searchdirs != NULL) {
10170 (void) fprintf(stderr, "-p option requires use of -e\n");
10171 usage();
10172 }
10173 if (dump_opt['H'] && !dump_opt['f']) {
10174 (void) fprintf(stderr, "-H option requires use of -f\n");
10175 usage();
10176 }
10177 #if defined(_LP64)
10178 /*
10179 * ZDB does not typically re-read blocks; therefore limit the ARC
10180 * to 256 MB, which can be used entirely for metadata.
10181 */
10182 zfs_arc_min = 2ULL << SPA_MAXBLOCKSHIFT;
10183 zfs_arc_max = 256 * 1024 * 1024;
10184 #endif
10185
10186 /*
10187 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
10188 * "zdb -b" uses traversal prefetch which uses async reads.
10189 * For good performance, let several of them be active at once.
10190 */
10191 zfs_vdev_async_read_max_active = 10;
10192
10193 /*
10194 * Disable reference tracking for better performance.
10195 */
10196 reference_tracking_enable = B_FALSE;
10197
10198 /*
10199 * Do not fail spa_load when spa_load_verify fails. This is needed
10200 * to load non-idle pools.
10201 */
10202 spa_load_verify_dryrun = B_TRUE;
10203
10204 /*
10205 * ZDB should have ability to read spacemaps.
10206 */
10207 spa_mode_readable_spacemaps = B_TRUE;
10208
10209 libspl_set_assert_ok((dump_opt['A'] == 1) || (dump_opt['A'] > 2));
10210 zfs_recover = (dump_opt['A'] > 1);
10211
10212 if (dump_all)
10213 verbose = MAX(verbose, 1);
10214
10215 for (c = 0; c < 256; c++) {
10216 if (dump_all && strchr("ABeEfFkKlLNOPrRSXy", c) == NULL)
10217 dump_opt[c] = 1;
10218 if (dump_opt[c])
10219 dump_opt[c] += verbose;
10220 }
10221
10222 argc -= optind;
10223 argv += optind;
10224 if (argc < 2 && dump_opt['R'])
10225 usage();
10226
10227 target = argv[0];
10228
10229 /*
10230 * Automate cachefile
10231 */
10232 if (!spa_config_path_env && !config_path_console && target &&
10233 !dump_opt['l'] && libzfs_core_init() == 0) {
10234 char *pname = strdup(target);
10235 const char *value;
10236 nvlist_t *pnvl = NULL;
10237 nvlist_t *vnvl = NULL;
10238
10239 if (strpbrk(pname, "/@") != NULL)
10240 *strpbrk(pname, "/@") = '\0';
10241
10242 if (pname && lzc_get_props(pname, &pnvl) == 0) {
10243 if (nvlist_lookup_nvlist(pnvl, "cachefile",
10244 &vnvl) == 0) {
10245 value = fnvlist_lookup_string(vnvl,
10246 ZPROP_VALUE);
10247 } else {
10248 value = "-";
10249 }
10250 strlcpy(pbuf, value, sizeof (pbuf));
10251 if (pbuf[0] != '\0') {
10252 if (pbuf[0] == '/') {
10253 if (access(pbuf, F_OK) == 0)
10254 spa_config_path = pbuf;
10255 else
10256 force_import = B_TRUE;
10257 } else if ((strcmp(pbuf, "-") == 0 &&
10258 access(ZPOOL_CACHE, F_OK) != 0) ||
10259 strcmp(pbuf, "none") == 0) {
10260 force_import = B_TRUE;
10261 }
10262 }
10263 nvlist_free(vnvl);
10264 }
10265
10266 free(pname);
10267 nvlist_free(pnvl);
10268 libzfs_core_fini();
10269 }
10270
10271 dmu_objset_register_type(DMU_OST_ZFS, dummy_get_file_info);
10272 kernel_init(SPA_MODE_READ);
10273 kernel_init_done = B_TRUE;
10274
10275 if (dump_opt['E']) {
10276 if (argc != 1)
10277 usage();
10278 zdb_embedded_block(argv[0]);
10279 error = 0;
10280 goto fini;
10281 }
10282
10283 if (argc < 1) {
10284 if (!dump_opt['e'] && dump_opt['C']) {
10285 dump_cachefile(spa_config_path);
10286 error = 0;
10287 goto fini;
10288 }
10289 if (dump_opt['o'])
10290 /*
10291 * Avoid blasting tunable options off the top of the
10292 * screen.
10293 */
10294 zdb_exit(1);
10295 usage();
10296 }
10297
10298 if (dump_opt['l']) {
10299 error = dump_label(argv[0]);
10300 goto fini;
10301 }
10302
10303 if (dump_opt['X'] || dump_opt['F'])
10304 rewind = ZPOOL_DO_REWIND |
10305 (dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
10306
10307 /* -N implies -d */
10308 if (dump_opt['N'] && dump_opt['d'] == 0)
10309 dump_opt['d'] = dump_opt['N'];
10310
10311 if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 ||
10312 nvlist_add_uint64(policy, ZPOOL_LOAD_REQUEST_TXG, max_txg) != 0 ||
10313 nvlist_add_uint32(policy, ZPOOL_LOAD_REWIND_POLICY, rewind) != 0)
10314 fatal("internal error: %s", strerror(ENOMEM));
10315
10316 error = 0;
10317
10318 if (strpbrk(target, "/@") != NULL) {
10319 size_t targetlen;
10320
10321 target_pool = strdup(target);
10322 *strpbrk(target_pool, "/@") = '\0';
10323
10324 target_is_spa = B_FALSE;
10325 targetlen = strlen(target);
10326 if (targetlen && target[targetlen - 1] == '/')
10327 target[targetlen - 1] = '\0';
10328
10329 /*
10330 * See if an objset ID was supplied (-d <pool>/<objset ID>).
10331 * To disambiguate tank/100, consider the 100 as objsetID
10332 * if -N was given, otherwise 100 is an objsetID iff
10333 * tank/100 as a named dataset fails on lookup.
10334 */
10335 objset_str = strchr(target, '/');
10336 if (objset_str && strlen(objset_str) > 1 &&
10337 zdb_numeric(objset_str + 1)) {
10338 char *endptr;
10339 errno = 0;
10340 objset_str++;
10341 objset_id = strtoull(objset_str, &endptr, 0);
10342 /* dataset 0 is the same as opening the pool */
10343 if (errno == 0 && endptr != objset_str &&
10344 objset_id != 0) {
10345 if (dump_opt['N'])
10346 dataset_lookup = B_TRUE;
10347 }
10348 /* normal dataset name not an objset ID */
10349 if (endptr == objset_str) {
10350 objset_id = -1;
10351 }
10352 } else if (objset_str && !zdb_numeric(objset_str + 1) &&
10353 dump_opt['N']) {
10354 printf("Supply a numeric objset ID with -N\n");
10355 error = 2;
10356 goto fini;
10357 }
10358 } else {
10359 target_pool = target;
10360 }
10361
10362 if (dump_opt['e'] || force_import) {
10363 importargs_t args = { 0 };
10364
10365 /*
10366 * If path is not provided, search in /dev
10367 */
10368 if (searchdirs == NULL) {
10369 searchdirs = umem_alloc(sizeof (char *), UMEM_NOFAIL);
10370 searchdirs[nsearch++] = (char *)ZFS_DEVDIR;
10371 }
10372
10373 args.paths = nsearch;
10374 args.path = searchdirs;
10375 args.can_be_active = B_TRUE;
10376
10377 libpc_handle_t lpch = {
10378 .lpc_lib_handle = NULL,
10379 .lpc_ops = &libzpool_config_ops,
10380 .lpc_printerr = B_TRUE
10381 };
10382 error = zpool_find_config(&lpch, target_pool, &cfg, &args);
10383
10384 if (error == 0) {
10385
10386 if (nvlist_add_nvlist(cfg,
10387 ZPOOL_LOAD_POLICY, policy) != 0) {
10388 fatal("can't open '%s': %s",
10389 target, strerror(ENOMEM));
10390 }
10391
10392 if (dump_opt['C'] > 1) {
10393 (void) printf("\nConfiguration for import:\n");
10394 dump_nvlist(cfg, 8);
10395 }
10396
10397 /*
10398 * Disable the activity check to allow examination of
10399 * active pools.
10400 */
10401 error = spa_import(target_pool, cfg, NULL,
10402 flags | ZFS_IMPORT_SKIP_MMP);
10403 }
10404 }
10405
10406 if (searchdirs != NULL) {
10407 umem_free(searchdirs, nsearch * sizeof (char *));
10408 searchdirs = NULL;
10409 }
10410
10411 /*
10412 * We need to make sure to process -O option or call
10413 * dump_path after the -e option has been processed,
10414 * which imports the pool to the namespace if it's
10415 * not in the cachefile.
10416 */
10417 if (dump_opt['O'] && !dump_opt['r']) {
10418 if (argc != 2)
10419 usage();
10420 dump_opt['v'] = verbose + 3;
10421 error = dump_path(argv[0], argv[1], NULL);
10422 goto fini;
10423 }
10424
10425 if (dump_opt['r']) {
10426 target_is_spa = B_FALSE;
10427 if (argc != 3)
10428 usage();
10429 dump_opt['v'] = verbose;
10430 if (dump_opt['O']) {
10431 object = strtoull(argv[1], NULL, 0);
10432 } else {
10433 error = dump_path(argv[0], argv[1], &object);
10434 }
10435 if (error != 0)
10436 fatal("internal error: %s", strerror(error));
10437 }
10438
10439 /*
10440 * import_checkpointed_state makes the assumption that the
10441 * target pool that we pass it is already part of the spa
10442 * namespace. Because of that we need to make sure to call
10443 * it always after the -e option has been processed, which
10444 * imports the pool to the namespace if it's not in the
10445 * cachefile.
10446 */
10447 char *checkpoint_pool = NULL;
10448 char *checkpoint_target = NULL;
10449 if (dump_opt['k']) {
10450 checkpoint_pool = import_checkpointed_state(target, cfg,
10451 target_is_spa, &checkpoint_target);
10452
10453 if (checkpoint_target != NULL)
10454 target = checkpoint_target;
10455 }
10456
10457 if (cfg != NULL) {
10458 nvlist_free(cfg);
10459 cfg = NULL;
10460 }
10461
10462 if (target_pool != target)
10463 free(target_pool);
10464
10465 if (error == 0) {
10466 if (dump_opt['k'] && (target_is_spa || dump_opt['R'])) {
10467 ASSERT(checkpoint_pool != NULL);
10468 ASSERT0P(checkpoint_target);
10469
10470 error = spa_open(checkpoint_pool, &spa, FTAG);
10471 if (error != 0) {
10472 fatal("Tried to open pool \"%s\" but "
10473 "spa_open() failed with error %d\n",
10474 checkpoint_pool, error);
10475 }
10476
10477 } else if (target_is_spa || dump_opt['R'] || dump_opt['B'] ||
10478 objset_id == 0) {
10479 zdb_set_skip_mmp(target);
10480 error = spa_open_rewind(target, &spa, FTAG, policy,
10481 NULL);
10482 if (error) {
10483 /*
10484 * If we're missing the log device then
10485 * try opening the pool after clearing the
10486 * log state.
10487 */
10488 spa_namespace_enter(FTAG);
10489 if ((spa = spa_lookup(target)) != NULL &&
10490 spa->spa_log_state == SPA_LOG_MISSING) {
10491 spa->spa_log_state = SPA_LOG_CLEAR;
10492 error = 0;
10493 }
10494 spa_namespace_exit(FTAG);
10495
10496 if (!error) {
10497 error = spa_open_rewind(target, &spa,
10498 FTAG, policy, NULL);
10499 }
10500 }
10501 } else if (strpbrk(target, "#") != NULL) {
10502 dsl_pool_t *dp;
10503 error = dsl_pool_hold(target, FTAG, &dp);
10504 if (error != 0) {
10505 fatal("can't dump '%s': %s", target,
10506 strerror(error));
10507 }
10508 error = dump_bookmark(dp, target, B_TRUE, verbose > 1);
10509 dsl_pool_rele(dp, FTAG);
10510 if (error != 0) {
10511 fatal("can't dump '%s': %s", target,
10512 strerror(error));
10513 }
10514 goto fini;
10515 } else {
10516 target_pool = strdup(target);
10517 if (strpbrk(target, "/@") != NULL)
10518 *strpbrk(target_pool, "/@") = '\0';
10519
10520 zdb_set_skip_mmp(target);
10521 /*
10522 * If -N was supplied, the user has indicated that
10523 * zdb -d <pool>/<objsetID> is in effect. Otherwise
10524 * we first assume that the dataset string is the
10525 * dataset name. If dmu_objset_hold fails with the
10526 * dataset string, and we have an objset_id, retry the
10527 * lookup with the objsetID.
10528 */
10529 boolean_t retry = B_TRUE;
10530 retry_lookup:
10531 if (dataset_lookup == B_TRUE) {
10532 /*
10533 * Use the supplied id to get the name
10534 * for open_objset.
10535 */
10536 error = spa_open(target_pool, &spa, FTAG);
10537 if (error == 0) {
10538 error = name_from_objset_id(spa,
10539 objset_id, dsname);
10540 spa_close(spa, FTAG);
10541 if (error == 0)
10542 target = dsname;
10543 }
10544 }
10545 if (error == 0) {
10546 if (objset_id > 0 && retry) {
10547 int err = dmu_objset_hold(target, FTAG,
10548 &os);
10549 if (err) {
10550 dataset_lookup = B_TRUE;
10551 retry = B_FALSE;
10552 goto retry_lookup;
10553 } else {
10554 dmu_objset_rele(os, FTAG);
10555 }
10556 }
10557 error = open_objset(target, FTAG, &os);
10558 }
10559 if (error == 0)
10560 spa = dmu_objset_spa(os);
10561 free(target_pool);
10562 }
10563 }
10564 nvlist_free(policy);
10565
10566 if (error)
10567 fatal("can't open '%s': %s", target, strerror(error));
10568
10569 /*
10570 * Set the pool failure mode to panic in order to prevent the pool
10571 * from suspending. A suspended I/O will have no way to resume and
10572 * can prevent the zdb(8) command from terminating as expected.
10573 */
10574 if (spa != NULL)
10575 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
10576
10577 argv++;
10578 argc--;
10579 if (dump_opt['r']) {
10580 error = zdb_copy_object(os, object, argv[1]);
10581 } else if (!dump_opt['R']) {
10582 flagbits['d'] = ZOR_FLAG_DIRECTORY;
10583 flagbits['f'] = ZOR_FLAG_PLAIN_FILE;
10584 flagbits['m'] = ZOR_FLAG_SPACE_MAP;
10585 flagbits['z'] = ZOR_FLAG_ZAP;
10586 flagbits['A'] = ZOR_FLAG_ALL_TYPES;
10587
10588 if (argc > 0 && (dump_opt['d'] || dump_opt['f'])) {
10589 zopt_object_args = argc;
10590 zopt_object_ranges = calloc(zopt_object_args,
10591 sizeof (zopt_object_range_t));
10592 for (unsigned i = 0; i < zopt_object_args; i++) {
10593 int err;
10594 const char *msg = NULL;
10595
10596 err = parse_object_range(argv[i],
10597 &zopt_object_ranges[i], &msg);
10598 if (err != 0)
10599 fatal("Bad object or range: '%s': %s\n",
10600 argv[i], msg ?: "");
10601 }
10602 } else if (argc > 0 && dump_opt['m']) {
10603 zopt_metaslab_args = argc;
10604 zopt_metaslab = calloc(zopt_metaslab_args,
10605 sizeof (uint64_t));
10606 for (unsigned i = 0; i < zopt_metaslab_args; i++) {
10607 errno = 0;
10608 zopt_metaslab[i] = strtoull(argv[i], NULL, 0);
10609 if (zopt_metaslab[i] == 0 && errno != 0)
10610 fatal("bad number %s: %s", argv[i],
10611 strerror(errno));
10612 }
10613 }
10614
10615 if (dump_opt['f'] && os != NULL) {
10616 dump_opt['v'] = verbose;
10617 dump_file_data_layout(os);
10618 } else if (dump_opt['B']) {
10619 dump_backup(target, objset_id,
10620 argc > 0 ? argv[0] : NULL);
10621 } else if (os != NULL) {
10622 dump_objset(os);
10623 } else if (zopt_object_args > 0 && !dump_opt['m']) {
10624 dump_objset(spa->spa_meta_objset);
10625 } else {
10626 dump_zpool(spa);
10627 }
10628 } else {
10629 flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR;
10630 flagbits['c'] = ZDB_FLAG_CHECKSUM;
10631 flagbits['d'] = ZDB_FLAG_DECOMPRESS;
10632 flagbits['e'] = ZDB_FLAG_BSWAP;
10633 flagbits['g'] = ZDB_FLAG_GBH;
10634 flagbits['i'] = ZDB_FLAG_INDIRECT;
10635 flagbits['r'] = ZDB_FLAG_RAW;
10636 flagbits['v'] = ZDB_FLAG_VERBOSE;
10637
10638 for (int i = 0; i < argc; i++)
10639 zdb_read_block(argv[i], spa);
10640 }
10641
10642 if (dump_opt['k']) {
10643 free(checkpoint_pool);
10644 if (!target_is_spa)
10645 free(checkpoint_target);
10646 }
10647
10648 fini:
10649 if (spa != NULL)
10650 zdb_ddt_cleanup(spa);
10651
10652 if (os != NULL) {
10653 close_objset(os, FTAG);
10654 } else if (spa != NULL) {
10655 spa_close(spa, FTAG);
10656 }
10657
10658 fuid_table_destroy();
10659
10660 dump_debug_buffer();
10661
10662 if (kernel_init_done)
10663 kernel_fini();
10664
10665 if (corruption_found && error == 0)
10666 error = 3;
10667
10668 return (error);
10669 }
10670