Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
freebsd
GitHub Repository: freebsd/freebsd-src
Path: blob/main/sys/contrib/openzfs/cmd/zdb/zdb.c
107074 views
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 <[email protected]>
37
* Copyright (c) 2023, 2024, Klara Inc.
38
* Copyright (c) 2023, Rob Norris <[email protected]>
39
*/
40
41
#include <stdio.h>
42
#include <unistd.h>
43
#include <stdlib.h>
44
#include <ctype.h>
45
#include <getopt.h>
46
#include <openssl/evp.h>
47
#include <sys/zfs_context.h>
48
#include <sys/spa.h>
49
#include <sys/spa_impl.h>
50
#include <sys/dmu.h>
51
#include <sys/zap.h>
52
#include <sys/zap_impl.h>
53
#include <sys/fs/zfs.h>
54
#include <sys/zfs_znode.h>
55
#include <sys/zfs_sa.h>
56
#include <sys/sa.h>
57
#include <sys/sa_impl.h>
58
#include <sys/vdev.h>
59
#include <sys/vdev_impl.h>
60
#include <sys/metaslab_impl.h>
61
#include <sys/dmu_objset.h>
62
#include <sys/dsl_dir.h>
63
#include <sys/dsl_dataset.h>
64
#include <sys/dsl_pool.h>
65
#include <sys/dsl_bookmark.h>
66
#include <sys/dbuf.h>
67
#include <sys/zil.h>
68
#include <sys/zil_impl.h>
69
#include <sys/stat.h>
70
#include <sys/resource.h>
71
#include <sys/dmu_send.h>
72
#include <sys/dmu_traverse.h>
73
#include <sys/zio_checksum.h>
74
#include <sys/zio_compress.h>
75
#include <sys/zfs_fuid.h>
76
#include <sys/arc.h>
77
#include <sys/arc_impl.h>
78
#include <sys/ddt.h>
79
#include <sys/ddt_impl.h>
80
#include <sys/zfeature.h>
81
#include <sys/abd.h>
82
#include <sys/blkptr.h>
83
#include <sys/dsl_crypt.h>
84
#include <sys/dsl_scan.h>
85
#include <sys/btree.h>
86
#include <sys/brt.h>
87
#include <sys/brt_impl.h>
88
#include <zfs_comutil.h>
89
#include <sys/zstd/zstd.h>
90
#include <sys/backtrace.h>
91
92
#include <libzpool.h>
93
#include <libnvpair.h>
94
#include <libzutil.h>
95
#include <libzfs_core.h>
96
97
#include <libzdb.h>
98
99
#include "zdb.h"
100
101
102
extern int reference_tracking_enable;
103
extern int zfs_recover;
104
extern uint_t zfs_vdev_async_read_max_active;
105
extern boolean_t spa_load_verify_dryrun;
106
extern boolean_t spa_mode_readable_spacemaps;
107
extern uint_t zfs_reconstruct_indirect_combinations_max;
108
extern uint_t zfs_btree_verify_intensity;
109
110
enum {
111
ARG_ALLOCATED = 256,
112
ARG_BLOCK_BIN_MODE,
113
ARG_BLOCK_CLASSES,
114
};
115
116
static const char cmdname[] = "zdb";
117
uint8_t dump_opt[512];
118
119
typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size);
120
121
static uint64_t *zopt_metaslab = NULL;
122
static unsigned zopt_metaslab_args = 0;
123
124
125
static zopt_object_range_t *zopt_object_ranges = NULL;
126
static unsigned zopt_object_args = 0;
127
128
static int flagbits[256];
129
130
131
static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */
132
static int leaked_objects = 0;
133
static zfs_range_tree_t *mos_refd_objs;
134
static spa_t *spa;
135
static objset_t *os;
136
static boolean_t kernel_init_done;
137
static boolean_t corruption_found = B_FALSE;
138
139
static enum {
140
BIN_AUTO = 0,
141
BIN_PSIZE,
142
BIN_LSIZE,
143
BIN_ASIZE,
144
} block_bin_mode = BIN_AUTO;
145
146
static enum {
147
CLASS_NORMAL = 1 << 1,
148
CLASS_SPECIAL = 1 << 2,
149
CLASS_DEDUP = 1 << 3,
150
CLASS_OTHER = 1 << 4,
151
} block_classes = 0;
152
153
static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *,
154
boolean_t);
155
static void mos_obj_refd(uint64_t);
156
static void mos_obj_refd_multiple(uint64_t);
157
static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free,
158
dmu_tx_t *tx);
159
160
161
162
static void zdb_print_blkptr(const blkptr_t *bp, int flags);
163
static void zdb_exit(int reason);
164
165
typedef struct sublivelist_verify_block_refcnt {
166
/* block pointer entry in livelist being verified */
167
blkptr_t svbr_blk;
168
169
/*
170
* Refcount gets incremented to 1 when we encounter the first
171
* FREE entry for the svfbr block pointer and a node for it
172
* is created in our ZDB verification/tracking metadata.
173
*
174
* As we encounter more FREE entries we increment this counter
175
* and similarly decrement it whenever we find the respective
176
* ALLOC entries for this block.
177
*
178
* When the refcount gets to 0 it means that all the FREE and
179
* ALLOC entries of this block have paired up and we no longer
180
* need to track it in our verification logic (e.g. the node
181
* containing this struct in our verification data structure
182
* should be freed).
183
*
184
* [refer to sublivelist_verify_blkptr() for the actual code]
185
*/
186
uint32_t svbr_refcnt;
187
} sublivelist_verify_block_refcnt_t;
188
189
static int
190
sublivelist_block_refcnt_compare(const void *larg, const void *rarg)
191
{
192
const sublivelist_verify_block_refcnt_t *l = larg;
193
const sublivelist_verify_block_refcnt_t *r = rarg;
194
return (livelist_compare(&l->svbr_blk, &r->svbr_blk));
195
}
196
197
static int
198
sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free,
199
dmu_tx_t *tx)
200
{
201
ASSERT0P(tx);
202
struct sublivelist_verify *sv = arg;
203
sublivelist_verify_block_refcnt_t current = {
204
.svbr_blk = *bp,
205
206
/*
207
* Start with 1 in case this is the first free entry.
208
* This field is not used for our B-Tree comparisons
209
* anyway.
210
*/
211
.svbr_refcnt = 1,
212
};
213
214
zfs_btree_index_t where;
215
sublivelist_verify_block_refcnt_t *pair =
216
zfs_btree_find(&sv->sv_pair, &current, &where);
217
if (free) {
218
if (pair == NULL) {
219
/* first free entry for this block pointer */
220
zfs_btree_add(&sv->sv_pair, &current);
221
} else {
222
pair->svbr_refcnt++;
223
}
224
} else {
225
if (pair == NULL) {
226
/* block that is currently marked as allocated */
227
for (int i = 0; i < SPA_DVAS_PER_BP; i++) {
228
if (DVA_IS_EMPTY(&bp->blk_dva[i]))
229
break;
230
sublivelist_verify_block_t svb = {
231
.svb_dva = bp->blk_dva[i],
232
.svb_allocated_txg =
233
BP_GET_BIRTH(bp)
234
};
235
236
if (zfs_btree_find(&sv->sv_leftover, &svb,
237
&where) == NULL) {
238
zfs_btree_add_idx(&sv->sv_leftover,
239
&svb, &where);
240
}
241
}
242
} else {
243
/* alloc matches a free entry */
244
pair->svbr_refcnt--;
245
if (pair->svbr_refcnt == 0) {
246
/* all allocs and frees have been matched */
247
zfs_btree_remove_idx(&sv->sv_pair, &where);
248
}
249
}
250
}
251
252
return (0);
253
}
254
255
static int
256
sublivelist_verify_func(void *args, dsl_deadlist_entry_t *dle)
257
{
258
int err;
259
struct sublivelist_verify *sv = args;
260
261
zfs_btree_create(&sv->sv_pair, sublivelist_block_refcnt_compare, NULL,
262
sizeof (sublivelist_verify_block_refcnt_t));
263
264
err = bpobj_iterate_nofree(&dle->dle_bpobj, sublivelist_verify_blkptr,
265
sv, NULL);
266
267
sublivelist_verify_block_refcnt_t *e;
268
zfs_btree_index_t *cookie = NULL;
269
while ((e = zfs_btree_destroy_nodes(&sv->sv_pair, &cookie)) != NULL) {
270
char blkbuf[BP_SPRINTF_LEN];
271
snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
272
&e->svbr_blk, B_TRUE);
273
(void) printf("\tERROR: %d unmatched FREE(s): %s\n",
274
e->svbr_refcnt, blkbuf);
275
corruption_found = B_TRUE;
276
}
277
zfs_btree_destroy(&sv->sv_pair);
278
279
return (err);
280
}
281
282
static int
283
livelist_block_compare(const void *larg, const void *rarg)
284
{
285
const sublivelist_verify_block_t *l = larg;
286
const sublivelist_verify_block_t *r = rarg;
287
288
if (DVA_GET_VDEV(&l->svb_dva) < DVA_GET_VDEV(&r->svb_dva))
289
return (-1);
290
else if (DVA_GET_VDEV(&l->svb_dva) > DVA_GET_VDEV(&r->svb_dva))
291
return (+1);
292
293
if (DVA_GET_OFFSET(&l->svb_dva) < DVA_GET_OFFSET(&r->svb_dva))
294
return (-1);
295
else if (DVA_GET_OFFSET(&l->svb_dva) > DVA_GET_OFFSET(&r->svb_dva))
296
return (+1);
297
298
if (DVA_GET_ASIZE(&l->svb_dva) < DVA_GET_ASIZE(&r->svb_dva))
299
return (-1);
300
else if (DVA_GET_ASIZE(&l->svb_dva) > DVA_GET_ASIZE(&r->svb_dva))
301
return (+1);
302
303
return (0);
304
}
305
306
/*
307
* Check for errors in a livelist while tracking all unfreed ALLOCs in the
308
* sublivelist_verify_t: sv->sv_leftover
309
*/
310
static void
311
livelist_verify(dsl_deadlist_t *dl, void *arg)
312
{
313
sublivelist_verify_t *sv = arg;
314
dsl_deadlist_iterate(dl, sublivelist_verify_func, sv);
315
}
316
317
/*
318
* Check for errors in the livelist entry and discard the intermediary
319
* data structures
320
*/
321
static int
322
sublivelist_verify_lightweight(void *args, dsl_deadlist_entry_t *dle)
323
{
324
(void) args;
325
sublivelist_verify_t sv;
326
zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
327
sizeof (sublivelist_verify_block_t));
328
int err = sublivelist_verify_func(&sv, dle);
329
zfs_btree_clear(&sv.sv_leftover);
330
zfs_btree_destroy(&sv.sv_leftover);
331
return (err);
332
}
333
334
typedef struct metaslab_verify {
335
/*
336
* Tree containing all the leftover ALLOCs from the livelists
337
* that are part of this metaslab.
338
*/
339
zfs_btree_t mv_livelist_allocs;
340
341
/*
342
* Metaslab information.
343
*/
344
uint64_t mv_vdid;
345
uint64_t mv_msid;
346
uint64_t mv_start;
347
uint64_t mv_end;
348
349
/*
350
* What's currently allocated for this metaslab.
351
*/
352
zfs_range_tree_t *mv_allocated;
353
} metaslab_verify_t;
354
355
typedef void ll_iter_t(dsl_deadlist_t *ll, void *arg);
356
357
typedef int (*zdb_log_sm_cb_t)(spa_t *spa, space_map_entry_t *sme, uint64_t txg,
358
void *arg);
359
360
typedef struct unflushed_iter_cb_arg {
361
spa_t *uic_spa;
362
uint64_t uic_txg;
363
void *uic_arg;
364
zdb_log_sm_cb_t uic_cb;
365
} unflushed_iter_cb_arg_t;
366
367
static int
368
iterate_through_spacemap_logs_cb(space_map_entry_t *sme, void *arg)
369
{
370
unflushed_iter_cb_arg_t *uic = arg;
371
return (uic->uic_cb(uic->uic_spa, sme, uic->uic_txg, uic->uic_arg));
372
}
373
374
static void
375
iterate_through_spacemap_logs(spa_t *spa, zdb_log_sm_cb_t cb, void *arg)
376
{
377
if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
378
return;
379
380
spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
381
for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
382
sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
383
space_map_t *sm = NULL;
384
VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
385
sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
386
387
unflushed_iter_cb_arg_t uic = {
388
.uic_spa = spa,
389
.uic_txg = sls->sls_txg,
390
.uic_arg = arg,
391
.uic_cb = cb
392
};
393
VERIFY0(space_map_iterate(sm, space_map_length(sm),
394
iterate_through_spacemap_logs_cb, &uic));
395
space_map_close(sm);
396
}
397
spa_config_exit(spa, SCL_CONFIG, FTAG);
398
}
399
400
static void
401
verify_livelist_allocs(metaslab_verify_t *mv, uint64_t txg,
402
uint64_t offset, uint64_t size)
403
{
404
sublivelist_verify_block_t svb = {{{0}}};
405
DVA_SET_VDEV(&svb.svb_dva, mv->mv_vdid);
406
DVA_SET_OFFSET(&svb.svb_dva, offset);
407
DVA_SET_ASIZE(&svb.svb_dva, 0);
408
zfs_btree_index_t where;
409
uint64_t end_offset = offset + size;
410
411
/*
412
* Look for an exact match for spacemap entry in the livelist entries.
413
* Then, look for other livelist entries that fall within the range
414
* of the spacemap entry as it may have been condensed
415
*/
416
sublivelist_verify_block_t *found =
417
zfs_btree_find(&mv->mv_livelist_allocs, &svb, &where);
418
if (found == NULL) {
419
found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where);
420
}
421
for (; found != NULL && DVA_GET_VDEV(&found->svb_dva) == mv->mv_vdid &&
422
DVA_GET_OFFSET(&found->svb_dva) < end_offset;
423
found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
424
if (found->svb_allocated_txg <= txg) {
425
(void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
426
"from TXG %llx FREED at TXG %llx\n",
427
(u_longlong_t)DVA_GET_OFFSET(&found->svb_dva),
428
(u_longlong_t)DVA_GET_ASIZE(&found->svb_dva),
429
(u_longlong_t)found->svb_allocated_txg,
430
(u_longlong_t)txg);
431
corruption_found = B_TRUE;
432
}
433
}
434
}
435
436
static int
437
metaslab_spacemap_validation_cb(space_map_entry_t *sme, void *arg)
438
{
439
metaslab_verify_t *mv = arg;
440
uint64_t offset = sme->sme_offset;
441
uint64_t size = sme->sme_run;
442
uint64_t txg = sme->sme_txg;
443
444
if (sme->sme_type == SM_ALLOC) {
445
if (zfs_range_tree_contains(mv->mv_allocated,
446
offset, size)) {
447
(void) printf("ERROR: DOUBLE ALLOC: "
448
"%llu [%llx:%llx] "
449
"%llu:%llu LOG_SM\n",
450
(u_longlong_t)txg, (u_longlong_t)offset,
451
(u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
452
(u_longlong_t)mv->mv_msid);
453
corruption_found = B_TRUE;
454
} else {
455
zfs_range_tree_add(mv->mv_allocated,
456
offset, size);
457
}
458
} else {
459
if (!zfs_range_tree_contains(mv->mv_allocated,
460
offset, size)) {
461
(void) printf("ERROR: DOUBLE FREE: "
462
"%llu [%llx:%llx] "
463
"%llu:%llu LOG_SM\n",
464
(u_longlong_t)txg, (u_longlong_t)offset,
465
(u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
466
(u_longlong_t)mv->mv_msid);
467
corruption_found = B_TRUE;
468
} else {
469
zfs_range_tree_remove(mv->mv_allocated,
470
offset, size);
471
}
472
}
473
474
if (sme->sme_type != SM_ALLOC) {
475
/*
476
* If something is freed in the spacemap, verify that
477
* it is not listed as allocated in the livelist.
478
*/
479
verify_livelist_allocs(mv, txg, offset, size);
480
}
481
return (0);
482
}
483
484
static int
485
spacemap_check_sm_log_cb(spa_t *spa, space_map_entry_t *sme,
486
uint64_t txg, void *arg)
487
{
488
metaslab_verify_t *mv = arg;
489
uint64_t offset = sme->sme_offset;
490
uint64_t vdev_id = sme->sme_vdev;
491
492
vdev_t *vd = vdev_lookup_top(spa, vdev_id);
493
494
/* skip indirect vdevs */
495
if (!vdev_is_concrete(vd))
496
return (0);
497
498
if (vdev_id != mv->mv_vdid)
499
return (0);
500
501
metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
502
if (ms->ms_id != mv->mv_msid)
503
return (0);
504
505
if (txg < metaslab_unflushed_txg(ms))
506
return (0);
507
508
509
ASSERT3U(txg, ==, sme->sme_txg);
510
return (metaslab_spacemap_validation_cb(sme, mv));
511
}
512
513
static void
514
spacemap_check_sm_log(spa_t *spa, metaslab_verify_t *mv)
515
{
516
iterate_through_spacemap_logs(spa, spacemap_check_sm_log_cb, mv);
517
}
518
519
static void
520
spacemap_check_ms_sm(space_map_t *sm, metaslab_verify_t *mv)
521
{
522
if (sm == NULL)
523
return;
524
525
VERIFY0(space_map_iterate(sm, space_map_length(sm),
526
metaslab_spacemap_validation_cb, mv));
527
}
528
529
static void iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg);
530
531
/*
532
* Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
533
* they are part of that metaslab (mv_msid).
534
*/
535
static void
536
mv_populate_livelist_allocs(metaslab_verify_t *mv, sublivelist_verify_t *sv)
537
{
538
zfs_btree_index_t where;
539
sublivelist_verify_block_t *svb;
540
ASSERT3U(zfs_btree_numnodes(&mv->mv_livelist_allocs), ==, 0);
541
for (svb = zfs_btree_first(&sv->sv_leftover, &where);
542
svb != NULL;
543
svb = zfs_btree_next(&sv->sv_leftover, &where, &where)) {
544
if (DVA_GET_VDEV(&svb->svb_dva) != mv->mv_vdid)
545
continue;
546
547
if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start &&
548
(DVA_GET_OFFSET(&svb->svb_dva) +
549
DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_start) {
550
(void) printf("ERROR: Found block that crosses "
551
"metaslab boundary: <%llu:%llx:%llx>\n",
552
(u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
553
(u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
554
(u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
555
corruption_found = B_TRUE;
556
continue;
557
}
558
559
if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start)
560
continue;
561
562
if (DVA_GET_OFFSET(&svb->svb_dva) >= mv->mv_end)
563
continue;
564
565
if ((DVA_GET_OFFSET(&svb->svb_dva) +
566
DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_end) {
567
(void) printf("ERROR: Found block that crosses "
568
"metaslab boundary: <%llu:%llx:%llx>\n",
569
(u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
570
(u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
571
(u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
572
corruption_found = B_TRUE;
573
continue;
574
}
575
576
zfs_btree_add(&mv->mv_livelist_allocs, svb);
577
}
578
579
for (svb = zfs_btree_first(&mv->mv_livelist_allocs, &where);
580
svb != NULL;
581
svb = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
582
zfs_btree_remove(&sv->sv_leftover, svb);
583
}
584
}
585
586
/*
587
* [Livelist Check]
588
* Iterate through all the sublivelists and:
589
* - report leftover frees (**)
590
* - record leftover ALLOCs together with their TXG [see Cross Check]
591
*
592
* (**) Note: Double ALLOCs are valid in datasets that have dedup
593
* enabled. Similarly double FREEs are allowed as well but
594
* only if they pair up with a corresponding ALLOC entry once
595
* we our done with our sublivelist iteration.
596
*
597
* [Spacemap Check]
598
* for each metaslab:
599
* - iterate over spacemap and then the metaslab's entries in the
600
* spacemap log, then report any double FREEs and ALLOCs (do not
601
* blow up).
602
*
603
* [Cross Check]
604
* After finishing the Livelist Check phase and while being in the
605
* Spacemap Check phase, we find all the recorded leftover ALLOCs
606
* of the livelist check that are part of the metaslab that we are
607
* currently looking at in the Spacemap Check. We report any entries
608
* that are marked as ALLOCs in the livelists but have been actually
609
* freed (and potentially allocated again) after their TXG stamp in
610
* the spacemaps. Also report any ALLOCs from the livelists that
611
* belong to indirect vdevs (e.g. their vdev completed removal).
612
*
613
* Note that this will miss Log Spacemap entries that cancelled each other
614
* out before being flushed to the metaslab, so we are not guaranteed
615
* to match all erroneous ALLOCs.
616
*/
617
static void
618
livelist_metaslab_validate(spa_t *spa)
619
{
620
(void) printf("Verifying deleted livelist entries\n");
621
622
sublivelist_verify_t sv;
623
zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
624
sizeof (sublivelist_verify_block_t));
625
iterate_deleted_livelists(spa, livelist_verify, &sv);
626
627
(void) printf("Verifying metaslab entries\n");
628
vdev_t *rvd = spa->spa_root_vdev;
629
for (uint64_t c = 0; c < rvd->vdev_children; c++) {
630
vdev_t *vd = rvd->vdev_child[c];
631
632
if (!vdev_is_concrete(vd))
633
continue;
634
635
for (uint64_t mid = 0; mid < vd->vdev_ms_count; mid++) {
636
metaslab_t *m = vd->vdev_ms[mid];
637
638
(void) fprintf(stderr,
639
"\rverifying concrete vdev %llu, "
640
"metaslab %llu of %llu ...",
641
(longlong_t)vd->vdev_id,
642
(longlong_t)mid,
643
(longlong_t)vd->vdev_ms_count);
644
645
uint64_t shift, start;
646
zfs_range_seg_type_t type =
647
metaslab_calculate_range_tree_type(vd, m,
648
&start, &shift);
649
metaslab_verify_t mv;
650
mv.mv_allocated = zfs_range_tree_create_flags(
651
NULL, type, NULL, start, shift,
652
0, "livelist_metaslab_validate:mv_allocated");
653
mv.mv_vdid = vd->vdev_id;
654
mv.mv_msid = m->ms_id;
655
mv.mv_start = m->ms_start;
656
mv.mv_end = m->ms_start + m->ms_size;
657
zfs_btree_create(&mv.mv_livelist_allocs,
658
livelist_block_compare, NULL,
659
sizeof (sublivelist_verify_block_t));
660
661
mv_populate_livelist_allocs(&mv, &sv);
662
663
spacemap_check_ms_sm(m->ms_sm, &mv);
664
spacemap_check_sm_log(spa, &mv);
665
666
zfs_range_tree_vacate(mv.mv_allocated, NULL, NULL);
667
zfs_range_tree_destroy(mv.mv_allocated);
668
zfs_btree_clear(&mv.mv_livelist_allocs);
669
zfs_btree_destroy(&mv.mv_livelist_allocs);
670
}
671
}
672
(void) fprintf(stderr, "\n");
673
674
/*
675
* If there are any segments in the leftover tree after we walked
676
* through all the metaslabs in the concrete vdevs then this means
677
* that we have segments in the livelists that belong to indirect
678
* vdevs and are marked as allocated.
679
*/
680
if (zfs_btree_numnodes(&sv.sv_leftover) == 0) {
681
zfs_btree_destroy(&sv.sv_leftover);
682
return;
683
}
684
(void) printf("ERROR: Found livelist blocks marked as allocated "
685
"for indirect vdevs:\n");
686
corruption_found = B_TRUE;
687
688
zfs_btree_index_t *where = NULL;
689
sublivelist_verify_block_t *svb;
690
while ((svb = zfs_btree_destroy_nodes(&sv.sv_leftover, &where)) !=
691
NULL) {
692
int vdev_id = DVA_GET_VDEV(&svb->svb_dva);
693
ASSERT3U(vdev_id, <, rvd->vdev_children);
694
vdev_t *vd = rvd->vdev_child[vdev_id];
695
ASSERT(!vdev_is_concrete(vd));
696
(void) printf("<%d:%llx:%llx> TXG %llx\n",
697
vdev_id, (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
698
(u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva),
699
(u_longlong_t)svb->svb_allocated_txg);
700
}
701
(void) printf("\n");
702
zfs_btree_destroy(&sv.sv_leftover);
703
}
704
705
/*
706
* These libumem hooks provide a reasonable set of defaults for the allocator's
707
* debugging facilities.
708
*/
709
const char *
710
_umem_debug_init(void)
711
{
712
return ("default,verbose"); /* $UMEM_DEBUG setting */
713
}
714
715
const char *
716
_umem_logging_init(void)
717
{
718
return ("fail,contents"); /* $UMEM_LOGGING setting */
719
}
720
721
static void
722
usage(void)
723
{
724
(void) fprintf(stderr,
725
"Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
726
"[-I <inflight I/Os>]\n"
727
"\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
728
"\t\t[-K <key>]\n"
729
"\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
730
"\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
731
"\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
732
"\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
733
"\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
734
"\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
735
"\t%s [-v] <bookmark>\n"
736
"\t%s -C [-A] [-U <cache>] [<poolname>]\n"
737
"\t%s -l [-Aqu] <device>\n"
738
"\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
739
"[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
740
"\t%s -O [-K <key>] <dataset> <path>\n"
741
"\t%s -r [-K <key>] <dataset> <path> <destination>\n"
742
"\t%s -r [-K <key>] -O <dataset> <object-id> <destination>\n"
743
"\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
744
"\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
745
"\t%s -E [-A] word0:word1:...:word15\n"
746
"\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
747
"<poolname>\n\n",
748
cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
749
cmdname, cmdname, cmdname, cmdname, cmdname, cmdname);
750
751
(void) fprintf(stderr, " Dataset name must include at least one "
752
"separator character '/' or '@'\n");
753
(void) fprintf(stderr, " If dataset name is specified, only that "
754
"dataset is dumped\n");
755
(void) fprintf(stderr, " If object numbers or object number "
756
"ranges are specified, only those\n"
757
" objects or ranges are dumped.\n\n");
758
(void) fprintf(stderr,
759
" Object ranges take the form <start>:<end>[:<flags>]\n"
760
" start Starting object number\n"
761
" end Ending object number, or -1 for no upper bound\n"
762
" flags Optional flags to select object types:\n"
763
" A All objects (this is the default)\n"
764
" d ZFS directories\n"
765
" f ZFS files \n"
766
" m SPA space maps\n"
767
" z ZAPs\n"
768
" - Negate effect of next flag\n\n");
769
(void) fprintf(stderr, " Options to control amount of output:\n");
770
(void) fprintf(stderr, " -b --block-stats "
771
"block statistics\n");
772
(void) fprintf(stderr, " --bin=(lsize|psize|asize) "
773
"bin blocks based on this size in all three columns\n");
774
(void) fprintf(stderr,
775
" --class=(normal|special|dedup|other)[,...]\n"
776
" only consider blocks from "
777
"these allocation classes\n");
778
(void) fprintf(stderr, " -B --backup "
779
"backup stream\n");
780
(void) fprintf(stderr, " -c --checksum "
781
"checksum all metadata (twice for all data) blocks\n");
782
(void) fprintf(stderr, " -C --config "
783
"config (or cachefile if alone)\n");
784
(void) fprintf(stderr, " -d --datasets "
785
"dataset(s)\n");
786
(void) fprintf(stderr, " -D --dedup-stats "
787
"dedup statistics\n");
788
(void) fprintf(stderr, " -E --embedded-block-pointer=INTEGER\n"
789
" decode and display block "
790
"from an embedded block pointer\n");
791
(void) fprintf(stderr, " -h --history "
792
"pool history\n");
793
(void) fprintf(stderr, " -i --intent-logs "
794
"intent logs\n");
795
(void) fprintf(stderr, " -l --label "
796
"read label contents\n");
797
(void) fprintf(stderr, " -k --checkpointed-state "
798
"examine the checkpointed state of the pool\n");
799
(void) fprintf(stderr, " -L --disable-leak-tracking "
800
"disable leak tracking (do not load spacemaps)\n");
801
(void) fprintf(stderr, " -m --metaslabs "
802
"metaslabs\n");
803
(void) fprintf(stderr, " -M --metaslab-groups "
804
"metaslab groups\n");
805
(void) fprintf(stderr, " -O --object-lookups "
806
"perform object lookups by path\n");
807
(void) fprintf(stderr, " -r --copy-object "
808
"copy an object by path to file\n");
809
(void) fprintf(stderr, " -R --read-block "
810
"read and display block from a device\n");
811
(void) fprintf(stderr, " -s --io-stats "
812
"report stats on zdb's I/O\n");
813
(void) fprintf(stderr, " -S --simulate-dedup "
814
"simulate dedup to measure effect\n");
815
(void) fprintf(stderr, " -v --verbose "
816
"verbose (applies to all others)\n");
817
(void) fprintf(stderr, " -y --livelist "
818
"perform livelist and metaslab validation on any livelists being "
819
"deleted\n\n");
820
(void) fprintf(stderr, " Below options are intended for use "
821
"with other options:\n");
822
(void) fprintf(stderr, " -A --ignore-assertions "
823
"ignore assertions (-A), enable panic recovery (-AA) or both "
824
"(-AAA)\n");
825
(void) fprintf(stderr, " -e --exported "
826
"pool is exported/destroyed/has altroot/not in a cachefile\n");
827
(void) fprintf(stderr, " -F --automatic-rewind "
828
"attempt automatic rewind within safe range of transaction "
829
"groups\n");
830
(void) fprintf(stderr, " -G --dump-debug-msg "
831
"dump zfs_dbgmsg buffer before exiting\n");
832
(void) fprintf(stderr, " -I --inflight=INTEGER "
833
"specify the maximum number of checksumming I/Os "
834
"[default is 200]\n");
835
(void) fprintf(stderr, " -K --key=KEY "
836
"decryption key for encrypted dataset\n");
837
(void) fprintf(stderr, " -o --option=\"NAME=VALUE\" "
838
"set the named tunable to the given value\n");
839
(void) fprintf(stderr, " -p --path==PATH "
840
"use one or more with -e to specify path to vdev dir\n");
841
(void) fprintf(stderr, " -P --parseable "
842
"print numbers in parseable form\n");
843
(void) fprintf(stderr, " -q --skip-label "
844
"don't print label contents\n");
845
(void) fprintf(stderr, " -t --txg=INTEGER "
846
"highest txg to use when searching for uberblocks\n");
847
(void) fprintf(stderr, " -T --brt-stats "
848
"BRT statistics\n");
849
(void) fprintf(stderr, " -u --uberblock "
850
"uberblock\n");
851
(void) fprintf(stderr, " -U --cachefile=PATH "
852
"use alternate cachefile\n");
853
(void) fprintf(stderr, " -V --verbatim "
854
"do verbatim import\n");
855
(void) fprintf(stderr, " -x --dump-blocks=PATH "
856
"dump all read blocks into specified directory\n");
857
(void) fprintf(stderr, " -X --extreme-rewind "
858
"attempt extreme rewind (does not work with dataset)\n");
859
(void) fprintf(stderr, " -Y --all-reconstruction "
860
"attempt all reconstruction combinations for split blocks\n");
861
(void) fprintf(stderr, " -Z --zstd-headers "
862
"show ZSTD headers \n");
863
(void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
864
"to make only that option verbose\n");
865
(void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
866
zdb_exit(2);
867
}
868
869
static void
870
dump_debug_buffer(void)
871
{
872
ssize_t ret __attribute__((unused));
873
874
if (!dump_opt['G'])
875
return;
876
/*
877
* We use write() instead of printf() so that this function
878
* is safe to call from a signal handler.
879
*/
880
ret = write(STDERR_FILENO, "\n", 1);
881
zfs_dbgmsg_print(STDERR_FILENO, "zdb");
882
}
883
884
static void sig_handler(int signo)
885
{
886
struct sigaction action;
887
888
libspl_backtrace(STDERR_FILENO);
889
dump_debug_buffer();
890
891
/*
892
* Restore default action and re-raise signal so SIGSEGV and
893
* SIGABRT can trigger a core dump.
894
*/
895
action.sa_handler = SIG_DFL;
896
sigemptyset(&action.sa_mask);
897
action.sa_flags = 0;
898
(void) sigaction(signo, &action, NULL);
899
raise(signo);
900
}
901
902
/*
903
* Called for usage errors that are discovered after a call to spa_open(),
904
* dmu_bonus_hold(), or pool_match(). abort() is called for other errors.
905
*/
906
907
static void
908
fatal(const char *fmt, ...)
909
{
910
va_list ap;
911
912
va_start(ap, fmt);
913
(void) fprintf(stderr, "%s: ", cmdname);
914
(void) vfprintf(stderr, fmt, ap);
915
va_end(ap);
916
(void) fprintf(stderr, "\n");
917
918
dump_debug_buffer();
919
920
zdb_exit(1);
921
}
922
923
static void
924
dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size)
925
{
926
(void) size;
927
nvlist_t *nv;
928
size_t nvsize = *(uint64_t *)data;
929
char *packed = umem_alloc(nvsize, UMEM_NOFAIL);
930
931
VERIFY0(dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH));
932
933
VERIFY0(nvlist_unpack(packed, nvsize, &nv, 0));
934
935
umem_free(packed, nvsize);
936
937
dump_nvlist(nv, 8);
938
939
nvlist_free(nv);
940
}
941
942
static void
943
dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size)
944
{
945
(void) os, (void) object, (void) size;
946
spa_history_phys_t *shp = data;
947
948
if (shp == NULL)
949
return;
950
951
(void) printf("\t\tpool_create_len = %llu\n",
952
(u_longlong_t)shp->sh_pool_create_len);
953
(void) printf("\t\tphys_max_off = %llu\n",
954
(u_longlong_t)shp->sh_phys_max_off);
955
(void) printf("\t\tbof = %llu\n",
956
(u_longlong_t)shp->sh_bof);
957
(void) printf("\t\teof = %llu\n",
958
(u_longlong_t)shp->sh_eof);
959
(void) printf("\t\trecords_lost = %llu\n",
960
(u_longlong_t)shp->sh_records_lost);
961
}
962
963
static void
964
zdb_nicenum(uint64_t num, char *buf, size_t buflen)
965
{
966
if (dump_opt['P'])
967
(void) snprintf(buf, buflen, "%llu", (longlong_t)num);
968
else
969
nicenum(num, buf, buflen);
970
}
971
972
static void
973
zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen)
974
{
975
if (dump_opt['P'])
976
(void) snprintf(buf, buflen, "%llu", (longlong_t)bytes);
977
else
978
zfs_nicebytes(bytes, buf, buflen);
979
}
980
981
static const char histo_stars[] = "****************************************";
982
static const uint64_t histo_width = sizeof (histo_stars) - 1;
983
984
static void
985
dump_histogram(const uint64_t *histo, int size, int offset)
986
{
987
int i;
988
int minidx = size - 1;
989
int maxidx = 0;
990
uint64_t max = 0;
991
992
for (i = 0; i < size; i++) {
993
if (histo[i] == 0)
994
continue;
995
if (histo[i] > max)
996
max = histo[i];
997
if (i > maxidx)
998
maxidx = i;
999
if (i < minidx)
1000
minidx = i;
1001
}
1002
1003
if (max < histo_width)
1004
max = histo_width;
1005
1006
for (i = minidx; i <= maxidx; i++) {
1007
(void) printf("\t\t\t%3u: %6llu %s\n",
1008
i + offset, (u_longlong_t)histo[i],
1009
&histo_stars[(max - histo[i]) * histo_width / max]);
1010
}
1011
}
1012
1013
static void
1014
dump_zap_stats(objset_t *os, uint64_t object)
1015
{
1016
int error;
1017
zap_stats_t zs;
1018
1019
error = zap_get_stats(os, object, &zs);
1020
if (error)
1021
return;
1022
1023
if (zs.zs_ptrtbl_len == 0) {
1024
ASSERT(zs.zs_num_blocks == 1);
1025
(void) printf("\tmicrozap: %llu bytes, %llu entries\n",
1026
(u_longlong_t)zs.zs_blocksize,
1027
(u_longlong_t)zs.zs_num_entries);
1028
return;
1029
}
1030
1031
(void) printf("\tFat ZAP stats:\n");
1032
1033
(void) printf("\t\tPointer table:\n");
1034
(void) printf("\t\t\t%llu elements\n",
1035
(u_longlong_t)zs.zs_ptrtbl_len);
1036
(void) printf("\t\t\tzt_blk: %llu\n",
1037
(u_longlong_t)zs.zs_ptrtbl_zt_blk);
1038
(void) printf("\t\t\tzt_numblks: %llu\n",
1039
(u_longlong_t)zs.zs_ptrtbl_zt_numblks);
1040
(void) printf("\t\t\tzt_shift: %llu\n",
1041
(u_longlong_t)zs.zs_ptrtbl_zt_shift);
1042
(void) printf("\t\t\tzt_blks_copied: %llu\n",
1043
(u_longlong_t)zs.zs_ptrtbl_blks_copied);
1044
(void) printf("\t\t\tzt_nextblk: %llu\n",
1045
(u_longlong_t)zs.zs_ptrtbl_nextblk);
1046
1047
(void) printf("\t\tZAP entries: %llu\n",
1048
(u_longlong_t)zs.zs_num_entries);
1049
(void) printf("\t\tLeaf blocks: %llu\n",
1050
(u_longlong_t)zs.zs_num_leafs);
1051
(void) printf("\t\tTotal blocks: %llu\n",
1052
(u_longlong_t)zs.zs_num_blocks);
1053
(void) printf("\t\tzap_block_type: 0x%llx\n",
1054
(u_longlong_t)zs.zs_block_type);
1055
(void) printf("\t\tzap_magic: 0x%llx\n",
1056
(u_longlong_t)zs.zs_magic);
1057
(void) printf("\t\tzap_salt: 0x%llx\n",
1058
(u_longlong_t)zs.zs_salt);
1059
1060
(void) printf("\t\tLeafs with 2^n pointers:\n");
1061
dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);
1062
1063
(void) printf("\t\tBlocks with n*5 entries:\n");
1064
dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);
1065
1066
(void) printf("\t\tBlocks n/10 full:\n");
1067
dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);
1068
1069
(void) printf("\t\tEntries with n chunks:\n");
1070
dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);
1071
1072
(void) printf("\t\tBuckets with n entries:\n");
1073
dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
1074
}
1075
1076
static void
1077
dump_none(objset_t *os, uint64_t object, void *data, size_t size)
1078
{
1079
(void) os, (void) object, (void) data, (void) size;
1080
}
1081
1082
static void
1083
dump_unknown(objset_t *os, uint64_t object, void *data, size_t size)
1084
{
1085
(void) os, (void) object, (void) data, (void) size;
1086
(void) printf("\tUNKNOWN OBJECT TYPE\n");
1087
}
1088
1089
static void
1090
dump_uint8(objset_t *os, uint64_t object, void *data, size_t size)
1091
{
1092
(void) os, (void) object, (void) data, (void) size;
1093
}
1094
1095
static void
1096
dump_uint64(objset_t *os, uint64_t object, void *data, size_t size)
1097
{
1098
uint64_t *arr;
1099
uint64_t oursize;
1100
if (dump_opt['d'] < 6)
1101
return;
1102
1103
if (data == NULL) {
1104
dmu_object_info_t doi;
1105
1106
VERIFY0(dmu_object_info(os, object, &doi));
1107
size = doi.doi_max_offset;
1108
/*
1109
* We cap the size at 1 mebibyte here to prevent
1110
* allocation failures and nigh-infinite printing if the
1111
* object is extremely large.
1112
*/
1113
oursize = MIN(size, 1 << 20);
1114
arr = kmem_alloc(oursize, KM_SLEEP);
1115
1116
int err = dmu_read(os, object, 0, oursize, arr, 0);
1117
if (err != 0) {
1118
(void) printf("got error %u from dmu_read\n", err);
1119
kmem_free(arr, oursize);
1120
return;
1121
}
1122
} else {
1123
/*
1124
* Even though the allocation is already done in this code path,
1125
* we still cap the size to prevent excessive printing.
1126
*/
1127
oursize = MIN(size, 1 << 20);
1128
arr = data;
1129
}
1130
1131
if (size == 0) {
1132
if (data == NULL)
1133
kmem_free(arr, oursize);
1134
(void) printf("\t\t[]\n");
1135
return;
1136
}
1137
1138
(void) printf("\t\t[%0llx", (u_longlong_t)arr[0]);
1139
for (size_t i = 1; i * sizeof (uint64_t) < oursize; i++) {
1140
if (i % 4 != 0)
1141
(void) printf(", %0llx", (u_longlong_t)arr[i]);
1142
else
1143
(void) printf(",\n\t\t%0llx", (u_longlong_t)arr[i]);
1144
}
1145
if (oursize != size)
1146
(void) printf(", ... ");
1147
(void) printf("]\n");
1148
1149
if (data == NULL)
1150
kmem_free(arr, oursize);
1151
}
1152
1153
static void
1154
dump_zap(objset_t *os, uint64_t object, void *data, size_t size)
1155
{
1156
(void) data, (void) size;
1157
zap_cursor_t zc;
1158
zap_attribute_t *attrp = zap_attribute_long_alloc();
1159
void *prop;
1160
unsigned i;
1161
1162
dump_zap_stats(os, object);
1163
(void) printf("\n");
1164
1165
for (zap_cursor_init(&zc, os, object);
1166
zap_cursor_retrieve(&zc, attrp) == 0;
1167
zap_cursor_advance(&zc)) {
1168
boolean_t key64 =
1169
!!(zap_getflags(zc.zc_zap) & ZAP_FLAG_UINT64_KEY);
1170
1171
if (key64)
1172
(void) printf("\t\t0x%010" PRIu64 "x = ",
1173
*(uint64_t *)attrp->za_name);
1174
else
1175
(void) printf("\t\t%s = ", attrp->za_name);
1176
1177
if (attrp->za_num_integers == 0) {
1178
(void) printf("\n");
1179
continue;
1180
}
1181
prop = umem_zalloc(attrp->za_num_integers *
1182
attrp->za_integer_length, UMEM_NOFAIL);
1183
1184
if (key64)
1185
(void) zap_lookup_uint64(os, object,
1186
(const uint64_t *)attrp->za_name, 1,
1187
attrp->za_integer_length, attrp->za_num_integers,
1188
prop);
1189
else
1190
(void) zap_lookup(os, object, attrp->za_name,
1191
attrp->za_integer_length, attrp->za_num_integers,
1192
prop);
1193
1194
if (attrp->za_integer_length == 1 && !key64) {
1195
if (strcmp(attrp->za_name,
1196
DSL_CRYPTO_KEY_MASTER_KEY) == 0 ||
1197
strcmp(attrp->za_name,
1198
DSL_CRYPTO_KEY_HMAC_KEY) == 0 ||
1199
strcmp(attrp->za_name, DSL_CRYPTO_KEY_IV) == 0 ||
1200
strcmp(attrp->za_name, DSL_CRYPTO_KEY_MAC) == 0 ||
1201
strcmp(attrp->za_name,
1202
DMU_POOL_CHECKSUM_SALT) == 0) {
1203
uint8_t *u8 = prop;
1204
1205
for (i = 0; i < attrp->za_num_integers; i++) {
1206
(void) printf("%02x", u8[i]);
1207
}
1208
} else {
1209
(void) printf("%s", (char *)prop);
1210
}
1211
} else {
1212
for (i = 0; i < attrp->za_num_integers; i++) {
1213
switch (attrp->za_integer_length) {
1214
case 1:
1215
(void) printf("%u ",
1216
((uint8_t *)prop)[i]);
1217
break;
1218
case 2:
1219
(void) printf("%u ",
1220
((uint16_t *)prop)[i]);
1221
break;
1222
case 4:
1223
(void) printf("%u ",
1224
((uint32_t *)prop)[i]);
1225
break;
1226
case 8:
1227
(void) printf("%lld ",
1228
(u_longlong_t)((int64_t *)prop)[i]);
1229
break;
1230
}
1231
}
1232
}
1233
(void) printf("\n");
1234
umem_free(prop,
1235
attrp->za_num_integers * attrp->za_integer_length);
1236
}
1237
zap_cursor_fini(&zc);
1238
zap_attribute_free(attrp);
1239
}
1240
1241
static void
1242
dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size)
1243
{
1244
bpobj_phys_t *bpop = data;
1245
uint64_t i;
1246
char bytes[32], comp[32], uncomp[32];
1247
1248
/* make sure the output won't get truncated */
1249
_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
1250
_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
1251
_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
1252
1253
if (bpop == NULL)
1254
return;
1255
1256
zdb_nicenum(bpop->bpo_bytes, bytes, sizeof (bytes));
1257
zdb_nicenum(bpop->bpo_comp, comp, sizeof (comp));
1258
zdb_nicenum(bpop->bpo_uncomp, uncomp, sizeof (uncomp));
1259
1260
(void) printf("\t\tnum_blkptrs = %llu\n",
1261
(u_longlong_t)bpop->bpo_num_blkptrs);
1262
(void) printf("\t\tbytes = %s\n", bytes);
1263
if (size >= BPOBJ_SIZE_V1) {
1264
(void) printf("\t\tcomp = %s\n", comp);
1265
(void) printf("\t\tuncomp = %s\n", uncomp);
1266
}
1267
if (size >= BPOBJ_SIZE_V2) {
1268
(void) printf("\t\tsubobjs = %llu\n",
1269
(u_longlong_t)bpop->bpo_subobjs);
1270
(void) printf("\t\tnum_subobjs = %llu\n",
1271
(u_longlong_t)bpop->bpo_num_subobjs);
1272
}
1273
if (size >= sizeof (*bpop)) {
1274
(void) printf("\t\tnum_freed = %llu\n",
1275
(u_longlong_t)bpop->bpo_num_freed);
1276
}
1277
1278
if (dump_opt['d'] < 5)
1279
return;
1280
1281
for (i = 0; i < bpop->bpo_num_blkptrs; i++) {
1282
char blkbuf[BP_SPRINTF_LEN];
1283
blkptr_t bp;
1284
1285
int err = dmu_read(os, object,
1286
i * sizeof (bp), sizeof (bp), &bp, 0);
1287
if (err != 0) {
1288
(void) printf("got error %u from dmu_read\n", err);
1289
break;
1290
}
1291
snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp,
1292
BP_GET_FREE(&bp));
1293
(void) printf("\t%s\n", blkbuf);
1294
}
1295
}
1296
1297
static void
1298
dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size)
1299
{
1300
(void) data, (void) size;
1301
dmu_object_info_t doi;
1302
int64_t i;
1303
1304
VERIFY0(dmu_object_info(os, object, &doi));
1305
uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP);
1306
1307
int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0);
1308
if (err != 0) {
1309
(void) printf("got error %u from dmu_read\n", err);
1310
kmem_free(subobjs, doi.doi_max_offset);
1311
return;
1312
}
1313
1314
int64_t last_nonzero = -1;
1315
for (i = 0; i < doi.doi_max_offset / 8; i++) {
1316
if (subobjs[i] != 0)
1317
last_nonzero = i;
1318
}
1319
1320
for (i = 0; i <= last_nonzero; i++) {
1321
(void) printf("\t%llu\n", (u_longlong_t)subobjs[i]);
1322
}
1323
kmem_free(subobjs, doi.doi_max_offset);
1324
}
1325
1326
static void
1327
dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size)
1328
{
1329
(void) data, (void) size;
1330
dump_zap_stats(os, object);
1331
/* contents are printed elsewhere, properly decoded */
1332
}
1333
1334
static void
1335
dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size)
1336
{
1337
(void) data, (void) size;
1338
zap_cursor_t zc;
1339
zap_attribute_t *attrp = zap_attribute_alloc();
1340
1341
dump_zap_stats(os, object);
1342
(void) printf("\n");
1343
1344
for (zap_cursor_init(&zc, os, object);
1345
zap_cursor_retrieve(&zc, attrp) == 0;
1346
zap_cursor_advance(&zc)) {
1347
(void) printf("\t\t%s = ", attrp->za_name);
1348
if (attrp->za_num_integers == 0) {
1349
(void) printf("\n");
1350
continue;
1351
}
1352
(void) printf(" %llx : [%d:%d:%d]\n",
1353
(u_longlong_t)attrp->za_first_integer,
1354
(int)ATTR_LENGTH(attrp->za_first_integer),
1355
(int)ATTR_BSWAP(attrp->za_first_integer),
1356
(int)ATTR_NUM(attrp->za_first_integer));
1357
}
1358
zap_cursor_fini(&zc);
1359
zap_attribute_free(attrp);
1360
}
1361
1362
static void
1363
dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size)
1364
{
1365
(void) data, (void) size;
1366
zap_cursor_t zc;
1367
zap_attribute_t *attrp = zap_attribute_alloc();
1368
uint16_t *layout_attrs;
1369
unsigned i;
1370
1371
dump_zap_stats(os, object);
1372
(void) printf("\n");
1373
1374
for (zap_cursor_init(&zc, os, object);
1375
zap_cursor_retrieve(&zc, attrp) == 0;
1376
zap_cursor_advance(&zc)) {
1377
(void) printf("\t\t%s = [", attrp->za_name);
1378
if (attrp->za_num_integers == 0) {
1379
(void) printf("\n");
1380
continue;
1381
}
1382
1383
VERIFY(attrp->za_integer_length == 2);
1384
layout_attrs = umem_zalloc(attrp->za_num_integers *
1385
attrp->za_integer_length, UMEM_NOFAIL);
1386
1387
VERIFY(zap_lookup(os, object, attrp->za_name,
1388
attrp->za_integer_length,
1389
attrp->za_num_integers, layout_attrs) == 0);
1390
1391
for (i = 0; i != attrp->za_num_integers; i++)
1392
(void) printf(" %d ", (int)layout_attrs[i]);
1393
(void) printf("]\n");
1394
umem_free(layout_attrs,
1395
attrp->za_num_integers * attrp->za_integer_length);
1396
}
1397
zap_cursor_fini(&zc);
1398
zap_attribute_free(attrp);
1399
}
1400
1401
static void
1402
dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
1403
{
1404
(void) data, (void) size;
1405
zap_cursor_t zc;
1406
zap_attribute_t *attrp = zap_attribute_long_alloc();
1407
const char *typenames[] = {
1408
/* 0 */ "not specified",
1409
/* 1 */ "FIFO",
1410
/* 2 */ "Character Device",
1411
/* 3 */ "3 (invalid)",
1412
/* 4 */ "Directory",
1413
/* 5 */ "5 (invalid)",
1414
/* 6 */ "Block Device",
1415
/* 7 */ "7 (invalid)",
1416
/* 8 */ "Regular File",
1417
/* 9 */ "9 (invalid)",
1418
/* 10 */ "Symbolic Link",
1419
/* 11 */ "11 (invalid)",
1420
/* 12 */ "Socket",
1421
/* 13 */ "Door",
1422
/* 14 */ "Event Port",
1423
/* 15 */ "15 (invalid)",
1424
};
1425
1426
dump_zap_stats(os, object);
1427
(void) printf("\n");
1428
1429
for (zap_cursor_init(&zc, os, object);
1430
zap_cursor_retrieve(&zc, attrp) == 0;
1431
zap_cursor_advance(&zc)) {
1432
(void) printf("\t\t%s = %lld (type: %s)\n",
1433
attrp->za_name, ZFS_DIRENT_OBJ(attrp->za_first_integer),
1434
typenames[ZFS_DIRENT_TYPE(attrp->za_first_integer)]);
1435
}
1436
zap_cursor_fini(&zc);
1437
zap_attribute_free(attrp);
1438
}
1439
1440
static int
1441
get_dtl_refcount(vdev_t *vd)
1442
{
1443
int refcount = 0;
1444
1445
if (vd->vdev_ops->vdev_op_leaf) {
1446
space_map_t *sm = vd->vdev_dtl_sm;
1447
1448
if (sm != NULL &&
1449
sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1450
return (1);
1451
return (0);
1452
}
1453
1454
for (unsigned c = 0; c < vd->vdev_children; c++)
1455
refcount += get_dtl_refcount(vd->vdev_child[c]);
1456
return (refcount);
1457
}
1458
1459
static int
1460
get_metaslab_refcount(vdev_t *vd)
1461
{
1462
int refcount = 0;
1463
1464
if (vd->vdev_top == vd) {
1465
for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
1466
space_map_t *sm = vd->vdev_ms[m]->ms_sm;
1467
1468
if (sm != NULL &&
1469
sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1470
refcount++;
1471
}
1472
}
1473
for (unsigned c = 0; c < vd->vdev_children; c++)
1474
refcount += get_metaslab_refcount(vd->vdev_child[c]);
1475
1476
return (refcount);
1477
}
1478
1479
static int
1480
get_obsolete_refcount(vdev_t *vd)
1481
{
1482
uint64_t obsolete_sm_object;
1483
int refcount = 0;
1484
1485
VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1486
if (vd->vdev_top == vd && obsolete_sm_object != 0) {
1487
dmu_object_info_t doi;
1488
VERIFY0(dmu_object_info(vd->vdev_spa->spa_meta_objset,
1489
obsolete_sm_object, &doi));
1490
if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1491
refcount++;
1492
}
1493
} else {
1494
ASSERT0P(vd->vdev_obsolete_sm);
1495
ASSERT0(obsolete_sm_object);
1496
}
1497
for (unsigned c = 0; c < vd->vdev_children; c++) {
1498
refcount += get_obsolete_refcount(vd->vdev_child[c]);
1499
}
1500
1501
return (refcount);
1502
}
1503
1504
static int
1505
get_prev_obsolete_spacemap_refcount(spa_t *spa)
1506
{
1507
uint64_t prev_obj =
1508
spa->spa_condensing_indirect_phys.scip_prev_obsolete_sm_object;
1509
if (prev_obj != 0) {
1510
dmu_object_info_t doi;
1511
VERIFY0(dmu_object_info(spa->spa_meta_objset, prev_obj, &doi));
1512
if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1513
return (1);
1514
}
1515
}
1516
return (0);
1517
}
1518
1519
static int
1520
get_checkpoint_refcount(vdev_t *vd)
1521
{
1522
int refcount = 0;
1523
1524
if (vd->vdev_top == vd && vd->vdev_top_zap != 0 &&
1525
zap_contains(spa_meta_objset(vd->vdev_spa),
1526
vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) == 0)
1527
refcount++;
1528
1529
for (uint64_t c = 0; c < vd->vdev_children; c++)
1530
refcount += get_checkpoint_refcount(vd->vdev_child[c]);
1531
1532
return (refcount);
1533
}
1534
1535
static int
1536
get_log_spacemap_refcount(spa_t *spa)
1537
{
1538
return (avl_numnodes(&spa->spa_sm_logs_by_txg));
1539
}
1540
1541
static int
1542
verify_spacemap_refcounts(spa_t *spa)
1543
{
1544
uint64_t expected_refcount = 0;
1545
uint64_t actual_refcount;
1546
1547
(void) feature_get_refcount(spa,
1548
&spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM],
1549
&expected_refcount);
1550
actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
1551
actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);
1552
actual_refcount += get_obsolete_refcount(spa->spa_root_vdev);
1553
actual_refcount += get_prev_obsolete_spacemap_refcount(spa);
1554
actual_refcount += get_checkpoint_refcount(spa->spa_root_vdev);
1555
actual_refcount += get_log_spacemap_refcount(spa);
1556
1557
if (expected_refcount != actual_refcount) {
1558
(void) printf("space map refcount mismatch: expected %lld != "
1559
"actual %lld\n",
1560
(longlong_t)expected_refcount,
1561
(longlong_t)actual_refcount);
1562
return (2);
1563
}
1564
return (0);
1565
}
1566
1567
static void
1568
dump_spacemap(objset_t *os, space_map_t *sm)
1569
{
1570
const char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1571
"INVALID", "INVALID", "INVALID", "INVALID" };
1572
1573
if (sm == NULL)
1574
return;
1575
1576
(void) printf("space map object %llu:\n",
1577
(longlong_t)sm->sm_object);
1578
(void) printf(" smp_length = 0x%llx\n",
1579
(longlong_t)sm->sm_phys->smp_length);
1580
(void) printf(" smp_alloc = 0x%llx\n",
1581
(longlong_t)sm->sm_phys->smp_alloc);
1582
1583
if (dump_opt['d'] < 6 && dump_opt['m'] < 4)
1584
return;
1585
1586
/*
1587
* Print out the freelist entries in both encoded and decoded form.
1588
*/
1589
uint8_t mapshift = sm->sm_shift;
1590
int64_t alloc = 0;
1591
uint64_t word, entry_id = 0;
1592
for (uint64_t offset = 0; offset < space_map_length(sm);
1593
offset += sizeof (word)) {
1594
1595
VERIFY0(dmu_read(os, space_map_object(sm), offset,
1596
sizeof (word), &word, DMU_READ_PREFETCH));
1597
1598
if (sm_entry_is_debug(word)) {
1599
uint64_t de_txg = SM_DEBUG_TXG_DECODE(word);
1600
uint64_t de_sync_pass = SM_DEBUG_SYNCPASS_DECODE(word);
1601
if (de_txg == 0) {
1602
(void) printf(
1603
"\t [%6llu] PADDING\n",
1604
(u_longlong_t)entry_id);
1605
} else {
1606
(void) printf(
1607
"\t [%6llu] %s: txg %llu pass %llu\n",
1608
(u_longlong_t)entry_id,
1609
ddata[SM_DEBUG_ACTION_DECODE(word)],
1610
(u_longlong_t)de_txg,
1611
(u_longlong_t)de_sync_pass);
1612
}
1613
entry_id++;
1614
continue;
1615
}
1616
1617
char entry_type;
1618
uint64_t entry_off, entry_run, entry_vdev;
1619
1620
if (sm_entry_is_single_word(word)) {
1621
entry_type = (SM_TYPE_DECODE(word) == SM_ALLOC) ?
1622
'A' : 'F';
1623
entry_off = (SM_OFFSET_DECODE(word) << mapshift) +
1624
sm->sm_start;
1625
entry_run = SM_RUN_DECODE(word) << mapshift;
1626
1627
(void) printf("\t [%6llu] %c "
1628
"range: %012llx-%012llx size: %08llx\n",
1629
(u_longlong_t)entry_id, entry_type,
1630
(u_longlong_t)entry_off,
1631
(u_longlong_t)(entry_off + entry_run - 1),
1632
(u_longlong_t)entry_run);
1633
} else {
1634
/* it is a two-word entry so we read another word */
1635
ASSERT(sm_entry_is_double_word(word));
1636
1637
uint64_t extra_word;
1638
offset += sizeof (extra_word);
1639
ASSERT3U(offset, <, space_map_length(sm));
1640
VERIFY0(dmu_read(os, space_map_object(sm), offset,
1641
sizeof (extra_word), &extra_word,
1642
DMU_READ_PREFETCH));
1643
1644
entry_run = SM2_RUN_DECODE(word) << mapshift;
1645
entry_vdev = SM2_VDEV_DECODE(word);
1646
entry_type = (SM2_TYPE_DECODE(extra_word) == SM_ALLOC) ?
1647
'A' : 'F';
1648
entry_off = (SM2_OFFSET_DECODE(extra_word) <<
1649
mapshift) + sm->sm_start;
1650
1651
if (zopt_metaslab_args == 0 ||
1652
zopt_metaslab[0] == entry_vdev) {
1653
(void) printf("\t [%6llu] %c "
1654
"range: %012llx-%012llx size: %08llx "
1655
"vdev: %llu\n",
1656
(u_longlong_t)entry_id, entry_type,
1657
(u_longlong_t)entry_off,
1658
(u_longlong_t)(entry_off + entry_run - 1),
1659
(u_longlong_t)entry_run,
1660
(u_longlong_t)entry_vdev);
1661
}
1662
}
1663
1664
if (entry_type == 'A')
1665
alloc += entry_run;
1666
else
1667
alloc -= entry_run;
1668
entry_id++;
1669
}
1670
if (alloc != space_map_allocated(sm)) {
1671
(void) printf("space_map_object alloc (%lld) INCONSISTENT "
1672
"with space map summary (%lld)\n",
1673
(longlong_t)space_map_allocated(sm), (longlong_t)alloc);
1674
}
1675
}
1676
1677
static void
1678
dump_metaslab_stats(metaslab_t *msp)
1679
{
1680
char maxbuf[32];
1681
zfs_range_tree_t *rt = msp->ms_allocatable;
1682
zfs_btree_t *t = &msp->ms_allocatable_by_size;
1683
int free_pct = zfs_range_tree_space(rt) * 100 / msp->ms_size;
1684
1685
/* max sure nicenum has enough space */
1686
_Static_assert(sizeof (maxbuf) >= NN_NUMBUF_SZ, "maxbuf truncated");
1687
1688
zdb_nicenum(metaslab_largest_allocatable(msp), maxbuf, sizeof (maxbuf));
1689
1690
(void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
1691
"segments", zfs_btree_numnodes(t), "maxsize", maxbuf,
1692
"freepct", free_pct);
1693
(void) printf("\tIn-memory histogram:\n");
1694
dump_histogram(rt->rt_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1695
}
1696
1697
static void
1698
dump_allocated(void *arg, uint64_t start, uint64_t size)
1699
{
1700
uint64_t *off = arg;
1701
if (*off != start)
1702
(void) printf("ALLOC: %"PRIu64" %"PRIu64"\n", *off,
1703
start - *off);
1704
*off = start + size;
1705
}
1706
1707
static void
1708
dump_metaslab(metaslab_t *msp)
1709
{
1710
vdev_t *vd = msp->ms_group->mg_vd;
1711
spa_t *spa = vd->vdev_spa;
1712
space_map_t *sm = msp->ms_sm;
1713
char freebuf[32];
1714
1715
zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf,
1716
sizeof (freebuf));
1717
1718
(void) printf(
1719
"\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
1720
(u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
1721
(u_longlong_t)space_map_object(sm), freebuf);
1722
1723
if (dump_opt[ARG_ALLOCATED] ||
1724
(dump_opt['m'] > 2 && !dump_opt['L'])) {
1725
mutex_enter(&msp->ms_lock);
1726
VERIFY0(metaslab_load(msp));
1727
}
1728
1729
if (dump_opt['m'] > 2 && !dump_opt['L']) {
1730
zfs_range_tree_stat_verify(msp->ms_allocatable);
1731
dump_metaslab_stats(msp);
1732
}
1733
1734
if (dump_opt[ARG_ALLOCATED]) {
1735
uint64_t off = msp->ms_start;
1736
zfs_range_tree_walk(msp->ms_allocatable, dump_allocated,
1737
&off);
1738
if (off != msp->ms_start + msp->ms_size)
1739
(void) printf("ALLOC: %"PRIu64" %"PRIu64"\n", off,
1740
msp->ms_size - off);
1741
}
1742
1743
if (dump_opt['m'] > 1 && sm != NULL &&
1744
spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) {
1745
/*
1746
* The space map histogram represents free space in chunks
1747
* of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1748
*/
1749
(void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1750
(u_longlong_t)msp->ms_fragmentation);
1751
dump_histogram(sm->sm_phys->smp_histogram,
1752
SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift);
1753
}
1754
1755
if (dump_opt[ARG_ALLOCATED] ||
1756
(dump_opt['m'] > 2 && !dump_opt['L'])) {
1757
metaslab_unload(msp);
1758
mutex_exit(&msp->ms_lock);
1759
}
1760
1761
if (vd->vdev_ops == &vdev_draid_ops)
1762
ASSERT3U(msp->ms_size, <=, 1ULL << vd->vdev_ms_shift);
1763
else
1764
ASSERT3U(msp->ms_size, ==, 1ULL << vd->vdev_ms_shift);
1765
1766
dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
1767
1768
if (spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
1769
(void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1770
(u_longlong_t)metaslab_unflushed_txg(msp));
1771
}
1772
}
1773
1774
static void
1775
print_vdev_metaslab_header(vdev_t *vd)
1776
{
1777
vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
1778
const char *bias_str = "";
1779
if (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) {
1780
bias_str = VDEV_ALLOC_BIAS_LOG;
1781
} else if (alloc_bias == VDEV_BIAS_SPECIAL) {
1782
bias_str = VDEV_ALLOC_BIAS_SPECIAL;
1783
} else if (alloc_bias == VDEV_BIAS_DEDUP) {
1784
bias_str = VDEV_ALLOC_BIAS_DEDUP;
1785
}
1786
1787
uint64_t ms_flush_data_obj = 0;
1788
if (vd->vdev_top_zap != 0) {
1789
int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
1790
vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
1791
sizeof (uint64_t), 1, &ms_flush_data_obj);
1792
if (error != ENOENT) {
1793
ASSERT0(error);
1794
}
1795
}
1796
1797
(void) printf("\tvdev %10llu\t%s metaslab shift %4llu",
1798
(u_longlong_t)vd->vdev_id, bias_str,
1799
(u_longlong_t)vd->vdev_ms_shift);
1800
1801
if (ms_flush_data_obj != 0) {
1802
(void) printf(" ms_unflushed_phys object %llu",
1803
(u_longlong_t)ms_flush_data_obj);
1804
}
1805
1806
(void) printf("\n\t%-10s%5llu %-19s %-15s %-12s\n",
1807
"metaslabs", (u_longlong_t)vd->vdev_ms_count,
1808
"offset", "spacemap", "free");
1809
(void) printf("\t%15s %19s %15s %12s\n",
1810
"---------------", "-------------------",
1811
"---------------", "------------");
1812
}
1813
1814
static void
1815
dump_metaslab_groups(spa_t *spa, boolean_t show_special)
1816
{
1817
vdev_t *rvd = spa->spa_root_vdev;
1818
metaslab_class_t *mc = spa_normal_class(spa);
1819
metaslab_class_t *smc = spa_special_class(spa);
1820
uint64_t fragmentation;
1821
1822
metaslab_class_histogram_verify(mc);
1823
1824
for (unsigned c = 0; c < rvd->vdev_children; c++) {
1825
vdev_t *tvd = rvd->vdev_child[c];
1826
metaslab_group_t *mg = tvd->vdev_mg;
1827
1828
if (mg == NULL || (mg->mg_class != mc &&
1829
(!show_special || mg->mg_class != smc)))
1830
continue;
1831
1832
metaslab_group_histogram_verify(mg);
1833
mg->mg_fragmentation = metaslab_group_fragmentation(mg);
1834
1835
(void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1836
"fragmentation",
1837
(u_longlong_t)tvd->vdev_id,
1838
(u_longlong_t)tvd->vdev_ms_count);
1839
if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
1840
(void) printf("%3s\n", "-");
1841
} else {
1842
(void) printf("%3llu%%\n",
1843
(u_longlong_t)mg->mg_fragmentation);
1844
}
1845
dump_histogram(mg->mg_histogram,
1846
ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1847
}
1848
1849
(void) printf("\tpool %s\tfragmentation", spa_name(spa));
1850
fragmentation = metaslab_class_fragmentation(mc);
1851
if (fragmentation == ZFS_FRAG_INVALID)
1852
(void) printf("\t%3s\n", "-");
1853
else
1854
(void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation);
1855
dump_histogram(mc->mc_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1856
}
1857
1858
static void
1859
print_vdev_indirect(vdev_t *vd)
1860
{
1861
vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
1862
vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
1863
vdev_indirect_births_t *vib = vd->vdev_indirect_births;
1864
1865
if (vim == NULL) {
1866
ASSERT0P(vib);
1867
return;
1868
}
1869
1870
ASSERT3U(vdev_indirect_mapping_object(vim), ==,
1871
vic->vic_mapping_object);
1872
ASSERT3U(vdev_indirect_births_object(vib), ==,
1873
vic->vic_births_object);
1874
1875
(void) printf("indirect births obj %llu:\n",
1876
(longlong_t)vic->vic_births_object);
1877
(void) printf(" vib_count = %llu\n",
1878
(longlong_t)vdev_indirect_births_count(vib));
1879
for (uint64_t i = 0; i < vdev_indirect_births_count(vib); i++) {
1880
vdev_indirect_birth_entry_phys_t *cur_vibe =
1881
&vib->vib_entries[i];
1882
(void) printf("\toffset %llx -> txg %llu\n",
1883
(longlong_t)cur_vibe->vibe_offset,
1884
(longlong_t)cur_vibe->vibe_phys_birth_txg);
1885
}
1886
(void) printf("\n");
1887
1888
(void) printf("indirect mapping obj %llu:\n",
1889
(longlong_t)vic->vic_mapping_object);
1890
(void) printf(" vim_max_offset = 0x%llx\n",
1891
(longlong_t)vdev_indirect_mapping_max_offset(vim));
1892
(void) printf(" vim_bytes_mapped = 0x%llx\n",
1893
(longlong_t)vdev_indirect_mapping_bytes_mapped(vim));
1894
(void) printf(" vim_count = %llu\n",
1895
(longlong_t)vdev_indirect_mapping_num_entries(vim));
1896
1897
if (dump_opt['d'] <= 5 && dump_opt['m'] <= 3)
1898
return;
1899
1900
uint32_t *counts = vdev_indirect_mapping_load_obsolete_counts(vim);
1901
1902
for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
1903
vdev_indirect_mapping_entry_phys_t *vimep =
1904
&vim->vim_entries[i];
1905
(void) printf("\t<%llx:%llx:%llx> -> "
1906
"<%llx:%llx:%llx> (%x obsolete)\n",
1907
(longlong_t)vd->vdev_id,
1908
(longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
1909
(longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1910
(longlong_t)DVA_GET_VDEV(&vimep->vimep_dst),
1911
(longlong_t)DVA_GET_OFFSET(&vimep->vimep_dst),
1912
(longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1913
counts[i]);
1914
}
1915
(void) printf("\n");
1916
1917
uint64_t obsolete_sm_object;
1918
VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1919
if (obsolete_sm_object != 0) {
1920
objset_t *mos = vd->vdev_spa->spa_meta_objset;
1921
(void) printf("obsolete space map object %llu:\n",
1922
(u_longlong_t)obsolete_sm_object);
1923
ASSERT(vd->vdev_obsolete_sm != NULL);
1924
ASSERT3U(space_map_object(vd->vdev_obsolete_sm), ==,
1925
obsolete_sm_object);
1926
dump_spacemap(mos, vd->vdev_obsolete_sm);
1927
(void) printf("\n");
1928
}
1929
}
1930
1931
static void
1932
dump_metaslabs(spa_t *spa)
1933
{
1934
vdev_t *vd, *rvd = spa->spa_root_vdev;
1935
uint64_t m, c = 0, children = rvd->vdev_children;
1936
1937
(void) printf("\nMetaslabs:\n");
1938
1939
if (zopt_metaslab_args > 0) {
1940
c = zopt_metaslab[0];
1941
1942
if (c >= children)
1943
(void) fatal("bad vdev id: %llu", (u_longlong_t)c);
1944
1945
if (zopt_metaslab_args > 1) {
1946
vd = rvd->vdev_child[c];
1947
print_vdev_metaslab_header(vd);
1948
1949
for (m = 1; m < zopt_metaslab_args; m++) {
1950
if (zopt_metaslab[m] < vd->vdev_ms_count)
1951
dump_metaslab(
1952
vd->vdev_ms[zopt_metaslab[m]]);
1953
else
1954
(void) fprintf(stderr, "bad metaslab "
1955
"number %llu\n",
1956
(u_longlong_t)zopt_metaslab[m]);
1957
}
1958
(void) printf("\n");
1959
return;
1960
}
1961
children = c + 1;
1962
}
1963
for (; c < children; c++) {
1964
vd = rvd->vdev_child[c];
1965
print_vdev_metaslab_header(vd);
1966
1967
print_vdev_indirect(vd);
1968
1969
for (m = 0; m < vd->vdev_ms_count; m++)
1970
dump_metaslab(vd->vdev_ms[m]);
1971
(void) printf("\n");
1972
}
1973
}
1974
1975
static void
1976
dump_log_spacemaps(spa_t *spa)
1977
{
1978
if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1979
return;
1980
1981
(void) printf("\nLog Space Maps in Pool:\n");
1982
for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
1983
sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1984
space_map_t *sm = NULL;
1985
VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
1986
sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
1987
1988
(void) printf("Log Spacemap object %llu txg %llu\n",
1989
(u_longlong_t)sls->sls_sm_obj, (u_longlong_t)sls->sls_txg);
1990
dump_spacemap(spa->spa_meta_objset, sm);
1991
space_map_close(sm);
1992
}
1993
(void) printf("\n");
1994
}
1995
1996
static void
1997
dump_ddt_entry(const ddt_t *ddt, const ddt_lightweight_entry_t *ddlwe,
1998
uint64_t index)
1999
{
2000
const ddt_key_t *ddk = &ddlwe->ddlwe_key;
2001
char blkbuf[BP_SPRINTF_LEN];
2002
blkptr_t blk;
2003
int p;
2004
2005
for (p = 0; p < DDT_NPHYS(ddt); p++) {
2006
const ddt_univ_phys_t *ddp = &ddlwe->ddlwe_phys;
2007
ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p);
2008
2009
if (ddt_phys_birth(ddp, v) == 0)
2010
continue;
2011
ddt_bp_create(ddt->ddt_checksum, ddk, ddp, v, &blk);
2012
snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
2013
(void) printf("index %llx refcnt %llu phys %d %s\n",
2014
(u_longlong_t)index, (u_longlong_t)ddt_phys_refcnt(ddp, v),
2015
p, blkbuf);
2016
}
2017
}
2018
2019
static void
2020
dump_dedup_ratio(const ddt_stat_t *dds)
2021
{
2022
double rL, rP, rD, D, dedup, compress, copies;
2023
2024
if (dds->dds_blocks == 0)
2025
return;
2026
2027
rL = (double)dds->dds_ref_lsize;
2028
rP = (double)dds->dds_ref_psize;
2029
rD = (double)dds->dds_ref_dsize;
2030
D = (double)dds->dds_dsize;
2031
2032
dedup = rD / D;
2033
compress = rL / rP;
2034
copies = rD / rP;
2035
2036
(void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
2037
"dedup * compress / copies = %.2f\n\n",
2038
dedup, compress, copies, dedup * compress / copies);
2039
}
2040
2041
static void
2042
dump_ddt_log(ddt_t *ddt)
2043
{
2044
if (ddt->ddt_version != DDT_VERSION_FDT ||
2045
!(ddt->ddt_flags & DDT_FLAG_LOG))
2046
return;
2047
2048
for (int n = 0; n < 2; n++) {
2049
ddt_log_t *ddl = &ddt->ddt_log[n];
2050
2051
char flagstr[64] = {0};
2052
if (ddl->ddl_flags > 0) {
2053
flagstr[0] = ' ';
2054
int c = 1;
2055
if (ddl->ddl_flags & DDL_FLAG_FLUSHING)
2056
c += strlcpy(&flagstr[c], " FLUSHING",
2057
sizeof (flagstr) - c);
2058
if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT)
2059
c += strlcpy(&flagstr[c], " CHECKPOINT",
2060
sizeof (flagstr) - c);
2061
if (ddl->ddl_flags &
2062
~(DDL_FLAG_FLUSHING|DDL_FLAG_CHECKPOINT))
2063
c += strlcpy(&flagstr[c], " UNKNOWN",
2064
sizeof (flagstr) - c);
2065
flagstr[1] = '[';
2066
flagstr[c] = ']';
2067
}
2068
2069
uint64_t count = avl_numnodes(&ddl->ddl_tree);
2070
2071
printf(DMU_POOL_DDT_LOG ": flags=0x%02x%s; obj=%llu; "
2072
"len=%llu; txg=%llu; entries=%llu\n",
2073
zio_checksum_table[ddt->ddt_checksum].ci_name, n,
2074
ddl->ddl_flags, flagstr,
2075
(u_longlong_t)ddl->ddl_object,
2076
(u_longlong_t)ddl->ddl_length,
2077
(u_longlong_t)ddl->ddl_first_txg, (u_longlong_t)count);
2078
2079
if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT) {
2080
const ddt_key_t *ddk = &ddl->ddl_checkpoint;
2081
printf(" checkpoint: "
2082
"%016llx:%016llx:%016llx:%016llx:%016llx\n",
2083
(u_longlong_t)ddk->ddk_cksum.zc_word[0],
2084
(u_longlong_t)ddk->ddk_cksum.zc_word[1],
2085
(u_longlong_t)ddk->ddk_cksum.zc_word[2],
2086
(u_longlong_t)ddk->ddk_cksum.zc_word[3],
2087
(u_longlong_t)ddk->ddk_prop);
2088
}
2089
2090
if (count == 0 || dump_opt['D'] < 4)
2091
continue;
2092
2093
ddt_lightweight_entry_t ddlwe;
2094
uint64_t index = 0;
2095
for (ddt_log_entry_t *ddle = avl_first(&ddl->ddl_tree);
2096
ddle; ddle = AVL_NEXT(&ddl->ddl_tree, ddle)) {
2097
DDT_LOG_ENTRY_TO_LIGHTWEIGHT(ddt, ddle, &ddlwe);
2098
dump_ddt_entry(ddt, &ddlwe, index++);
2099
}
2100
}
2101
}
2102
2103
static void
2104
dump_ddt_object(ddt_t *ddt, ddt_type_t type, ddt_class_t class)
2105
{
2106
char name[DDT_NAMELEN];
2107
ddt_lightweight_entry_t ddlwe;
2108
uint64_t walk = 0;
2109
dmu_object_info_t doi;
2110
uint64_t count, dspace, mspace;
2111
int error;
2112
2113
error = ddt_object_info(ddt, type, class, &doi);
2114
2115
if (error == ENOENT)
2116
return;
2117
ASSERT0(error);
2118
2119
error = ddt_object_count(ddt, type, class, &count);
2120
ASSERT0(error);
2121
if (count == 0)
2122
return;
2123
2124
dspace = doi.doi_physical_blocks_512 << 9;
2125
mspace = doi.doi_fill_count * doi.doi_data_block_size;
2126
2127
ddt_object_name(ddt, type, class, name);
2128
2129
(void) printf("%s: dspace=%llu; mspace=%llu; entries=%llu\n", name,
2130
(u_longlong_t)dspace, (u_longlong_t)mspace, (u_longlong_t)count);
2131
2132
if (dump_opt['D'] < 3)
2133
return;
2134
2135
(void) printf("%s: object=%llu\n", name,
2136
(u_longlong_t)ddt->ddt_object[type][class]);
2137
zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class]);
2138
2139
if (dump_opt['D'] < 4)
2140
return;
2141
2142
if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE)
2143
return;
2144
2145
(void) printf("%s contents:\n\n", name);
2146
2147
while ((error = ddt_object_walk(ddt, type, class, &walk, &ddlwe)) == 0)
2148
dump_ddt_entry(ddt, &ddlwe, walk);
2149
2150
ASSERT3U(error, ==, ENOENT);
2151
2152
(void) printf("\n");
2153
}
2154
2155
static void
2156
dump_ddt(ddt_t *ddt)
2157
{
2158
if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
2159
return;
2160
2161
char flagstr[64] = {0};
2162
if (ddt->ddt_flags > 0) {
2163
flagstr[0] = ' ';
2164
int c = 1;
2165
if (ddt->ddt_flags & DDT_FLAG_FLAT)
2166
c += strlcpy(&flagstr[c], " FLAT",
2167
sizeof (flagstr) - c);
2168
if (ddt->ddt_flags & DDT_FLAG_LOG)
2169
c += strlcpy(&flagstr[c], " LOG",
2170
sizeof (flagstr) - c);
2171
if (ddt->ddt_flags & ~DDT_FLAG_MASK)
2172
c += strlcpy(&flagstr[c], " UNKNOWN",
2173
sizeof (flagstr) - c);
2174
flagstr[1] = '[';
2175
flagstr[c] = ']';
2176
}
2177
2178
printf("DDT-%s: version=%llu [%s]; flags=0x%02llx%s; rootobj=%llu\n",
2179
zio_checksum_table[ddt->ddt_checksum].ci_name,
2180
(u_longlong_t)ddt->ddt_version,
2181
(ddt->ddt_version == 0) ? "LEGACY" :
2182
(ddt->ddt_version == 1) ? "FDT" : "UNKNOWN",
2183
(u_longlong_t)ddt->ddt_flags, flagstr,
2184
(u_longlong_t)ddt->ddt_dir_object);
2185
2186
for (ddt_type_t type = 0; type < DDT_TYPES; type++)
2187
for (ddt_class_t class = 0; class < DDT_CLASSES; class++)
2188
dump_ddt_object(ddt, type, class);
2189
2190
dump_ddt_log(ddt);
2191
}
2192
2193
static void
2194
dump_all_ddts(spa_t *spa)
2195
{
2196
ddt_histogram_t ddh_total = {{{0}}};
2197
ddt_stat_t dds_total = {0};
2198
2199
for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++)
2200
dump_ddt(spa->spa_ddt[c]);
2201
2202
ddt_get_dedup_stats(spa, &dds_total);
2203
2204
if (dds_total.dds_blocks == 0) {
2205
(void) printf("All DDTs are empty\n");
2206
return;
2207
}
2208
2209
(void) printf("\n");
2210
2211
if (dump_opt['D'] > 1) {
2212
(void) printf("DDT histogram (aggregated over all DDTs):\n");
2213
ddt_get_dedup_histogram(spa, &ddh_total);
2214
zpool_dump_ddt(&dds_total, &ddh_total);
2215
}
2216
2217
dump_dedup_ratio(&dds_total);
2218
2219
/*
2220
* Dump a histogram of unique class entry age
2221
*/
2222
if (dump_opt['D'] == 3 && getenv("ZDB_DDT_UNIQUE_AGE_HIST") != NULL) {
2223
ddt_age_histo_t histogram;
2224
2225
(void) printf("DDT walk unique, building age histogram...\n");
2226
ddt_prune_walk(spa, 0, &histogram);
2227
2228
/*
2229
* print out histogram for unique entry class birth
2230
*/
2231
if (histogram.dah_entries > 0) {
2232
(void) printf("%5s %9s %4s\n",
2233
"age", "blocks", "amnt");
2234
(void) printf("%5s %9s %4s\n",
2235
"-----", "---------", "----");
2236
for (int i = 0; i < HIST_BINS; i++) {
2237
(void) printf("%5d %9d %4d%%\n", 1 << i,
2238
(int)histogram.dah_age_histo[i],
2239
(int)((histogram.dah_age_histo[i] * 100) /
2240
histogram.dah_entries));
2241
}
2242
}
2243
}
2244
}
2245
2246
static void
2247
dump_brt(spa_t *spa)
2248
{
2249
if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) {
2250
printf("BRT: unsupported on this pool\n");
2251
return;
2252
}
2253
2254
if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
2255
printf("BRT: empty\n");
2256
return;
2257
}
2258
2259
char count[32], used[32], saved[32];
2260
zdb_nicebytes(brt_get_used(spa), used, sizeof (used));
2261
zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved));
2262
uint64_t ratio = brt_get_ratio(spa);
2263
printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved,
2264
(u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100));
2265
2266
if (dump_opt['T'] < 2)
2267
return;
2268
2269
for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2270
brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2271
if (!brtvd->bv_initiated) {
2272
printf("BRT: vdev %" PRIu64 ": empty\n", vdevid);
2273
continue;
2274
}
2275
2276
zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count));
2277
zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used));
2278
zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved));
2279
printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n",
2280
vdevid, count, used, saved);
2281
}
2282
2283
if (dump_opt['T'] < 3)
2284
return;
2285
2286
/* -TTT shows a per-vdev histograms; -TTTT shows all entries */
2287
boolean_t do_histo = dump_opt['T'] == 3;
2288
2289
char dva[64];
2290
2291
if (!do_histo)
2292
printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2293
2294
for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2295
brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2296
if (!brtvd->bv_initiated)
2297
continue;
2298
2299
uint64_t counts[64] = {};
2300
2301
zap_cursor_t zc;
2302
zap_attribute_t *za = zap_attribute_alloc();
2303
for (zap_cursor_init(&zc, spa->spa_meta_objset,
2304
brtvd->bv_mos_entries);
2305
zap_cursor_retrieve(&zc, za) == 0;
2306
zap_cursor_advance(&zc)) {
2307
uint64_t refcnt;
2308
VERIFY0(zap_lookup_uint64(spa->spa_meta_objset,
2309
brtvd->bv_mos_entries,
2310
(const uint64_t *)za->za_name, 1,
2311
za->za_integer_length, za->za_num_integers,
2312
&refcnt));
2313
2314
if (do_histo)
2315
counts[highbit64(refcnt)]++;
2316
else {
2317
uint64_t offset =
2318
*(const uint64_t *)za->za_name;
2319
2320
snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx",
2321
vdevid, (u_longlong_t)offset);
2322
printf("%-16s %-10llu\n", dva,
2323
(u_longlong_t)refcnt);
2324
}
2325
}
2326
zap_cursor_fini(&zc);
2327
zap_attribute_free(za);
2328
2329
if (do_histo) {
2330
printf("\nBRT: vdev %" PRIu64
2331
": DVAs with 2^n refcnts:\n", vdevid);
2332
dump_histogram(counts, 64, 0);
2333
}
2334
}
2335
}
2336
2337
static void
2338
dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
2339
{
2340
char *prefix = arg;
2341
2342
(void) printf("%s [%llu,%llu) length %llu\n",
2343
prefix,
2344
(u_longlong_t)start,
2345
(u_longlong_t)(start + size),
2346
(u_longlong_t)(size));
2347
}
2348
2349
static void
2350
dump_dtl(vdev_t *vd, int indent)
2351
{
2352
spa_t *spa = vd->vdev_spa;
2353
boolean_t required;
2354
const char *name[DTL_TYPES] = { "missing", "partial", "scrub",
2355
"outage" };
2356
char prefix[256];
2357
2358
spa_vdev_state_enter(spa, SCL_NONE);
2359
required = vdev_dtl_required(vd);
2360
(void) spa_vdev_state_exit(spa, NULL, 0);
2361
2362
if (indent == 0)
2363
(void) printf("\nDirty time logs:\n\n");
2364
2365
(void) printf("\t%*s%s [%s]\n", indent, "",
2366
vd->vdev_path ? vd->vdev_path :
2367
vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa),
2368
required ? "DTL-required" : "DTL-expendable");
2369
2370
for (int t = 0; t < DTL_TYPES; t++) {
2371
zfs_range_tree_t *rt = vd->vdev_dtl[t];
2372
if (zfs_range_tree_space(rt) == 0)
2373
continue;
2374
(void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
2375
indent + 2, "", name[t]);
2376
zfs_range_tree_walk(rt, dump_dtl_seg, prefix);
2377
if (dump_opt['d'] > 5 && vd->vdev_children == 0)
2378
dump_spacemap(spa->spa_meta_objset,
2379
vd->vdev_dtl_sm);
2380
}
2381
2382
for (unsigned c = 0; c < vd->vdev_children; c++)
2383
dump_dtl(vd->vdev_child[c], indent + 4);
2384
}
2385
2386
static void
2387
dump_history(spa_t *spa)
2388
{
2389
nvlist_t **events = NULL;
2390
char *buf;
2391
uint64_t resid, len, off = 0;
2392
uint_t num = 0;
2393
int error;
2394
char tbuf[30];
2395
2396
if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) {
2397
(void) fprintf(stderr, "%s: unable to allocate I/O buffer\n",
2398
__func__);
2399
return;
2400
}
2401
2402
do {
2403
len = SPA_OLD_MAXBLOCKSIZE;
2404
2405
if ((error = spa_history_get(spa, &off, &len, buf)) != 0) {
2406
(void) fprintf(stderr, "Unable to read history: "
2407
"error %d\n", error);
2408
free(buf);
2409
return;
2410
}
2411
2412
if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0)
2413
break;
2414
2415
off -= resid;
2416
} while (len != 0);
2417
2418
(void) printf("\nHistory:\n");
2419
for (unsigned i = 0; i < num; i++) {
2420
boolean_t printed = B_FALSE;
2421
2422
if (nvlist_exists(events[i], ZPOOL_HIST_TIME)) {
2423
time_t tsec;
2424
struct tm t;
2425
2426
tsec = fnvlist_lookup_uint64(events[i],
2427
ZPOOL_HIST_TIME);
2428
(void) localtime_r(&tsec, &t);
2429
(void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t);
2430
} else {
2431
tbuf[0] = '\0';
2432
}
2433
2434
if (nvlist_exists(events[i], ZPOOL_HIST_CMD)) {
2435
(void) printf("%s %s\n", tbuf,
2436
fnvlist_lookup_string(events[i], ZPOOL_HIST_CMD));
2437
} else if (nvlist_exists(events[i], ZPOOL_HIST_INT_EVENT)) {
2438
uint64_t ievent;
2439
2440
ievent = fnvlist_lookup_uint64(events[i],
2441
ZPOOL_HIST_INT_EVENT);
2442
if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS)
2443
goto next;
2444
2445
(void) printf(" %s [internal %s txg:%ju] %s\n",
2446
tbuf,
2447
zfs_history_event_names[ievent],
2448
fnvlist_lookup_uint64(events[i],
2449
ZPOOL_HIST_TXG),
2450
fnvlist_lookup_string(events[i],
2451
ZPOOL_HIST_INT_STR));
2452
} else if (nvlist_exists(events[i], ZPOOL_HIST_INT_NAME)) {
2453
(void) printf("%s [txg:%ju] %s", tbuf,
2454
fnvlist_lookup_uint64(events[i],
2455
ZPOOL_HIST_TXG),
2456
fnvlist_lookup_string(events[i],
2457
ZPOOL_HIST_INT_NAME));
2458
2459
if (nvlist_exists(events[i], ZPOOL_HIST_DSNAME)) {
2460
(void) printf(" %s (%llu)",
2461
fnvlist_lookup_string(events[i],
2462
ZPOOL_HIST_DSNAME),
2463
(u_longlong_t)fnvlist_lookup_uint64(
2464
events[i],
2465
ZPOOL_HIST_DSID));
2466
}
2467
2468
(void) printf(" %s\n", fnvlist_lookup_string(events[i],
2469
ZPOOL_HIST_INT_STR));
2470
} else if (nvlist_exists(events[i], ZPOOL_HIST_IOCTL)) {
2471
(void) printf("%s ioctl %s\n", tbuf,
2472
fnvlist_lookup_string(events[i],
2473
ZPOOL_HIST_IOCTL));
2474
2475
if (nvlist_exists(events[i], ZPOOL_HIST_INPUT_NVL)) {
2476
(void) printf(" input:\n");
2477
dump_nvlist(fnvlist_lookup_nvlist(events[i],
2478
ZPOOL_HIST_INPUT_NVL), 8);
2479
}
2480
if (nvlist_exists(events[i], ZPOOL_HIST_OUTPUT_NVL)) {
2481
(void) printf(" output:\n");
2482
dump_nvlist(fnvlist_lookup_nvlist(events[i],
2483
ZPOOL_HIST_OUTPUT_NVL), 8);
2484
}
2485
if (nvlist_exists(events[i], ZPOOL_HIST_ERRNO)) {
2486
(void) printf(" errno: %lld\n",
2487
(longlong_t)fnvlist_lookup_int64(events[i],
2488
ZPOOL_HIST_ERRNO));
2489
}
2490
} else {
2491
goto next;
2492
}
2493
2494
printed = B_TRUE;
2495
next:
2496
if (dump_opt['h'] > 1) {
2497
if (!printed)
2498
(void) printf("unrecognized record:\n");
2499
dump_nvlist(events[i], 2);
2500
}
2501
}
2502
free(buf);
2503
}
2504
2505
static void
2506
dump_dnode(objset_t *os, uint64_t object, void *data, size_t size)
2507
{
2508
(void) os, (void) object, (void) data, (void) size;
2509
}
2510
2511
static uint64_t
2512
blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
2513
const zbookmark_phys_t *zb)
2514
{
2515
if (dnp == NULL) {
2516
ASSERT(zb->zb_level < 0);
2517
if (zb->zb_object == 0)
2518
return (zb->zb_blkid);
2519
return (zb->zb_blkid * BP_GET_LSIZE(bp));
2520
}
2521
2522
ASSERT(zb->zb_level >= 0);
2523
2524
return ((zb->zb_blkid <<
2525
(zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) *
2526
dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
2527
}
2528
2529
static void
2530
snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
2531
const blkptr_t *bp)
2532
{
2533
static abd_t *pabd = NULL;
2534
void *buf;
2535
zio_t *zio;
2536
zfs_zstdhdr_t zstd_hdr;
2537
int error;
2538
2539
if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD)
2540
return;
2541
2542
if (BP_IS_HOLE(bp))
2543
return;
2544
2545
if (BP_IS_EMBEDDED(bp)) {
2546
buf = malloc(SPA_MAXBLOCKSIZE);
2547
if (buf == NULL) {
2548
(void) fprintf(stderr, "out of memory\n");
2549
zdb_exit(1);
2550
}
2551
decode_embedded_bp_compressed(bp, buf);
2552
memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2553
free(buf);
2554
zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2555
zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2556
(void) snprintf(blkbuf + strlen(blkbuf),
2557
buflen - strlen(blkbuf),
2558
" ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2559
zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2560
zfs_get_hdrlevel(&zstd_hdr));
2561
return;
2562
}
2563
2564
if (!pabd)
2565
pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
2566
zio = zio_root(spa, NULL, NULL, 0);
2567
2568
/* Decrypt but don't decompress so we can read the compression header */
2569
zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL,
2570
ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS,
2571
NULL));
2572
error = zio_wait(zio);
2573
if (error) {
2574
(void) fprintf(stderr, "read failed: %d\n", error);
2575
return;
2576
}
2577
buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp));
2578
memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2579
zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2580
zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2581
2582
(void) snprintf(blkbuf + strlen(blkbuf),
2583
buflen - strlen(blkbuf),
2584
" ZSTD:size=%u:version=%u:level=%u:NORMAL",
2585
zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2586
zfs_get_hdrlevel(&zstd_hdr));
2587
2588
abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp));
2589
}
2590
2591
static void
2592
snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
2593
boolean_t bp_freed)
2594
{
2595
const dva_t *dva = bp->blk_dva;
2596
int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1;
2597
int i;
2598
2599
if (dump_opt['b'] >= 6) {
2600
snprintf_blkptr(blkbuf, buflen, bp);
2601
if (bp_freed) {
2602
(void) snprintf(blkbuf + strlen(blkbuf),
2603
buflen - strlen(blkbuf), " %s", "FREE");
2604
}
2605
return;
2606
}
2607
2608
if (BP_IS_EMBEDDED(bp)) {
2609
(void) sprintf(blkbuf,
2610
"EMBEDDED et=%u %llxL/%llxP B=%llu",
2611
(int)BPE_GET_ETYPE(bp),
2612
(u_longlong_t)BPE_GET_LSIZE(bp),
2613
(u_longlong_t)BPE_GET_PSIZE(bp),
2614
(u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2615
return;
2616
}
2617
2618
blkbuf[0] = '\0';
2619
2620
for (i = 0; i < ndvas; i++) {
2621
(void) snprintf(blkbuf + strlen(blkbuf),
2622
buflen - strlen(blkbuf), "%llu:%llx:%llx%s ",
2623
(u_longlong_t)DVA_GET_VDEV(&dva[i]),
2624
(u_longlong_t)DVA_GET_OFFSET(&dva[i]),
2625
(u_longlong_t)DVA_GET_ASIZE(&dva[i]),
2626
(DVA_GET_GANG(&dva[i]) ? "G" : ""));
2627
}
2628
2629
if (BP_IS_HOLE(bp)) {
2630
(void) snprintf(blkbuf + strlen(blkbuf),
2631
buflen - strlen(blkbuf),
2632
"%llxL B=%llu",
2633
(u_longlong_t)BP_GET_LSIZE(bp),
2634
(u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2635
} else {
2636
(void) snprintf(blkbuf + strlen(blkbuf),
2637
buflen - strlen(blkbuf),
2638
"%llxL/%llxP F=%llu B=%llu/%llu",
2639
(u_longlong_t)BP_GET_LSIZE(bp),
2640
(u_longlong_t)BP_GET_PSIZE(bp),
2641
(u_longlong_t)BP_GET_FILL(bp),
2642
(u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),
2643
(u_longlong_t)BP_GET_PHYSICAL_BIRTH(bp));
2644
if (bp_freed)
2645
(void) snprintf(blkbuf + strlen(blkbuf),
2646
buflen - strlen(blkbuf), " %s", "FREE");
2647
(void) snprintf(blkbuf + strlen(blkbuf),
2648
buflen - strlen(blkbuf),
2649
" cksum=%016llx:%016llx:%016llx:%016llx",
2650
(u_longlong_t)bp->blk_cksum.zc_word[0],
2651
(u_longlong_t)bp->blk_cksum.zc_word[1],
2652
(u_longlong_t)bp->blk_cksum.zc_word[2],
2653
(u_longlong_t)bp->blk_cksum.zc_word[3]);
2654
}
2655
}
2656
2657
static u_longlong_t
2658
print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
2659
const dnode_phys_t *dnp)
2660
{
2661
char blkbuf[BP_SPRINTF_LEN];
2662
u_longlong_t offset;
2663
int l;
2664
2665
offset = (u_longlong_t)blkid2offset(dnp, bp, zb);
2666
2667
(void) printf("%16llx ", offset);
2668
2669
ASSERT(zb->zb_level >= 0);
2670
2671
for (l = dnp->dn_nlevels - 1; l >= -1; l--) {
2672
if (l == zb->zb_level) {
2673
(void) printf("L%llx", (u_longlong_t)zb->zb_level);
2674
} else {
2675
(void) printf(" ");
2676
}
2677
}
2678
2679
snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE);
2680
if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD)
2681
snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp);
2682
(void) printf("%s", blkbuf);
2683
2684
if (!BP_IS_EMBEDDED(bp)) {
2685
if (BP_GET_TYPE(bp) != dnp->dn_type) {
2686
(void) printf(" (ERROR: Block pointer type "
2687
"(%llu) does not match dnode type (%hhu))",
2688
BP_GET_TYPE(bp), dnp->dn_type);
2689
corruption_found = B_TRUE;
2690
}
2691
if (BP_GET_LEVEL(bp) != zb->zb_level) {
2692
(void) printf(" (ERROR: Block pointer level "
2693
"(%llu) does not match bookmark level (%lld))",
2694
BP_GET_LEVEL(bp), (longlong_t)zb->zb_level);
2695
corruption_found = B_TRUE;
2696
}
2697
}
2698
(void) printf("\n");
2699
2700
return (offset);
2701
}
2702
2703
static int
2704
visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
2705
blkptr_t *bp, const zbookmark_phys_t *zb)
2706
{
2707
u_longlong_t offset;
2708
int err = 0;
2709
2710
if (BP_GET_BIRTH(bp) == 0)
2711
return (0);
2712
2713
offset = print_indirect(spa, bp, zb, dnp);
2714
2715
if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
2716
arc_flags_t flags = ARC_FLAG_WAIT;
2717
int i;
2718
blkptr_t *cbp;
2719
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
2720
arc_buf_t *buf;
2721
uint64_t fill = 0;
2722
ASSERT(!BP_IS_REDACTED(bp));
2723
2724
err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
2725
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
2726
if (err)
2727
return (err);
2728
ASSERT(buf->b_data);
2729
2730
/* recursively visit blocks below this */
2731
cbp = buf->b_data;
2732
for (i = 0; i < epb; i++, cbp++) {
2733
zbookmark_phys_t czb;
2734
2735
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
2736
zb->zb_level - 1,
2737
zb->zb_blkid * epb + i);
2738
err = visit_indirect(spa, dnp, cbp, &czb);
2739
if (err)
2740
break;
2741
fill += BP_GET_FILL(cbp);
2742
}
2743
if (!err) {
2744
if (fill != BP_GET_FILL(bp)) {
2745
(void) printf("%16llx: Block pointer "
2746
"fill (%llu) does not match calculated "
2747
"value (%llu)\n", offset, BP_GET_FILL(bp),
2748
(u_longlong_t)fill);
2749
corruption_found = B_TRUE;
2750
}
2751
}
2752
arc_buf_destroy(buf, &buf);
2753
}
2754
2755
return (err);
2756
}
2757
2758
static void
2759
dump_indirect(dnode_t *dn)
2760
{
2761
dnode_phys_t *dnp = dn->dn_phys;
2762
zbookmark_phys_t czb;
2763
2764
(void) printf("Indirect blocks:\n");
2765
2766
SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
2767
dn->dn_object, dnp->dn_nlevels - 1, 0);
2768
for (int j = 0; j < dnp->dn_nblkptr; j++) {
2769
czb.zb_blkid = j;
2770
(void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
2771
&dnp->dn_blkptr[j], &czb);
2772
}
2773
2774
(void) printf("\n");
2775
}
2776
2777
static void
2778
dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size)
2779
{
2780
(void) os, (void) object;
2781
dsl_dir_phys_t *dd = data;
2782
time_t crtime;
2783
char nice[32];
2784
2785
/* make sure nicenum has enough space */
2786
_Static_assert(sizeof (nice) >= NN_NUMBUF_SZ, "nice truncated");
2787
2788
if (dd == NULL)
2789
return;
2790
2791
ASSERT3U(size, >=, sizeof (dsl_dir_phys_t));
2792
2793
crtime = dd->dd_creation_time;
2794
(void) printf("\t\tcreation_time = %s", ctime(&crtime));
2795
(void) printf("\t\thead_dataset_obj = %llu\n",
2796
(u_longlong_t)dd->dd_head_dataset_obj);
2797
(void) printf("\t\tparent_dir_obj = %llu\n",
2798
(u_longlong_t)dd->dd_parent_obj);
2799
(void) printf("\t\torigin_obj = %llu\n",
2800
(u_longlong_t)dd->dd_origin_obj);
2801
(void) printf("\t\tchild_dir_zapobj = %llu\n",
2802
(u_longlong_t)dd->dd_child_dir_zapobj);
2803
zdb_nicenum(dd->dd_used_bytes, nice, sizeof (nice));
2804
(void) printf("\t\tused_bytes = %s\n", nice);
2805
zdb_nicenum(dd->dd_compressed_bytes, nice, sizeof (nice));
2806
(void) printf("\t\tcompressed_bytes = %s\n", nice);
2807
zdb_nicenum(dd->dd_uncompressed_bytes, nice, sizeof (nice));
2808
(void) printf("\t\tuncompressed_bytes = %s\n", nice);
2809
zdb_nicenum(dd->dd_quota, nice, sizeof (nice));
2810
(void) printf("\t\tquota = %s\n", nice);
2811
zdb_nicenum(dd->dd_reserved, nice, sizeof (nice));
2812
(void) printf("\t\treserved = %s\n", nice);
2813
(void) printf("\t\tprops_zapobj = %llu\n",
2814
(u_longlong_t)dd->dd_props_zapobj);
2815
(void) printf("\t\tdeleg_zapobj = %llu\n",
2816
(u_longlong_t)dd->dd_deleg_zapobj);
2817
(void) printf("\t\tflags = %llx\n",
2818
(u_longlong_t)dd->dd_flags);
2819
2820
#define DO(which) \
2821
zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2822
sizeof (nice)); \
2823
(void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2824
DO(HEAD);
2825
DO(SNAP);
2826
DO(CHILD);
2827
DO(CHILD_RSRV);
2828
DO(REFRSRV);
2829
#undef DO
2830
(void) printf("\t\tclones = %llu\n",
2831
(u_longlong_t)dd->dd_clones);
2832
}
2833
2834
static void
2835
dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
2836
{
2837
(void) os, (void) object;
2838
dsl_dataset_phys_t *ds = data;
2839
time_t crtime;
2840
char used[32], compressed[32], uncompressed[32], unique[32];
2841
char blkbuf[BP_SPRINTF_LEN];
2842
2843
/* make sure nicenum has enough space */
2844
_Static_assert(sizeof (used) >= NN_NUMBUF_SZ, "used truncated");
2845
_Static_assert(sizeof (compressed) >= NN_NUMBUF_SZ,
2846
"compressed truncated");
2847
_Static_assert(sizeof (uncompressed) >= NN_NUMBUF_SZ,
2848
"uncompressed truncated");
2849
_Static_assert(sizeof (unique) >= NN_NUMBUF_SZ, "unique truncated");
2850
2851
if (ds == NULL)
2852
return;
2853
2854
ASSERT(size == sizeof (*ds));
2855
crtime = ds->ds_creation_time;
2856
zdb_nicenum(ds->ds_referenced_bytes, used, sizeof (used));
2857
zdb_nicenum(ds->ds_compressed_bytes, compressed, sizeof (compressed));
2858
zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed,
2859
sizeof (uncompressed));
2860
zdb_nicenum(ds->ds_unique_bytes, unique, sizeof (unique));
2861
snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp);
2862
2863
(void) printf("\t\tdir_obj = %llu\n",
2864
(u_longlong_t)ds->ds_dir_obj);
2865
(void) printf("\t\tprev_snap_obj = %llu\n",
2866
(u_longlong_t)ds->ds_prev_snap_obj);
2867
(void) printf("\t\tprev_snap_txg = %llu\n",
2868
(u_longlong_t)ds->ds_prev_snap_txg);
2869
(void) printf("\t\tnext_snap_obj = %llu\n",
2870
(u_longlong_t)ds->ds_next_snap_obj);
2871
(void) printf("\t\tsnapnames_zapobj = %llu\n",
2872
(u_longlong_t)ds->ds_snapnames_zapobj);
2873
(void) printf("\t\tnum_children = %llu\n",
2874
(u_longlong_t)ds->ds_num_children);
2875
(void) printf("\t\tuserrefs_obj = %llu\n",
2876
(u_longlong_t)ds->ds_userrefs_obj);
2877
(void) printf("\t\tcreation_time = %s", ctime(&crtime));
2878
(void) printf("\t\tcreation_txg = %llu\n",
2879
(u_longlong_t)ds->ds_creation_txg);
2880
(void) printf("\t\tdeadlist_obj = %llu\n",
2881
(u_longlong_t)ds->ds_deadlist_obj);
2882
(void) printf("\t\tused_bytes = %s\n", used);
2883
(void) printf("\t\tcompressed_bytes = %s\n", compressed);
2884
(void) printf("\t\tuncompressed_bytes = %s\n", uncompressed);
2885
(void) printf("\t\tunique = %s\n", unique);
2886
(void) printf("\t\tfsid_guid = %llu\n",
2887
(u_longlong_t)ds->ds_fsid_guid);
2888
(void) printf("\t\tguid = %llu\n",
2889
(u_longlong_t)ds->ds_guid);
2890
(void) printf("\t\tflags = %llx\n",
2891
(u_longlong_t)ds->ds_flags);
2892
(void) printf("\t\tnext_clones_obj = %llu\n",
2893
(u_longlong_t)ds->ds_next_clones_obj);
2894
(void) printf("\t\tprops_obj = %llu\n",
2895
(u_longlong_t)ds->ds_props_obj);
2896
(void) printf("\t\tbp = %s\n", blkbuf);
2897
}
2898
2899
static int
2900
dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2901
{
2902
(void) arg, (void) tx;
2903
char blkbuf[BP_SPRINTF_LEN];
2904
2905
if (BP_GET_BIRTH(bp) != 0) {
2906
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
2907
(void) printf("\t%s\n", blkbuf);
2908
}
2909
return (0);
2910
}
2911
2912
static void
2913
dump_bptree(objset_t *os, uint64_t obj, const char *name)
2914
{
2915
char bytes[32];
2916
bptree_phys_t *bt;
2917
dmu_buf_t *db;
2918
2919
/* make sure nicenum has enough space */
2920
_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2921
2922
if (dump_opt['d'] < 3)
2923
return;
2924
2925
VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
2926
bt = db->db_data;
2927
zdb_nicenum(bt->bt_bytes, bytes, sizeof (bytes));
2928
(void) printf("\n %s: %llu datasets, %s\n",
2929
name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes);
2930
dmu_buf_rele(db, FTAG);
2931
2932
if (dump_opt['d'] < 5)
2933
return;
2934
2935
(void) printf("\n");
2936
2937
(void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL);
2938
}
2939
2940
static int
2941
dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
2942
{
2943
(void) arg, (void) tx;
2944
char blkbuf[BP_SPRINTF_LEN];
2945
2946
ASSERT(BP_GET_BIRTH(bp) != 0);
2947
snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed);
2948
(void) printf("\t%s\n", blkbuf);
2949
return (0);
2950
}
2951
2952
static void
2953
dump_full_bpobj(bpobj_t *bpo, const char *name, int indent)
2954
{
2955
char bytes[32];
2956
char comp[32];
2957
char uncomp[32];
2958
uint64_t i;
2959
2960
/* make sure nicenum has enough space */
2961
_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2962
_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
2963
_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
2964
2965
if (dump_opt['d'] < 3)
2966
return;
2967
2968
zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes, sizeof (bytes));
2969
if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
2970
zdb_nicenum(bpo->bpo_phys->bpo_comp, comp, sizeof (comp));
2971
zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp, sizeof (uncomp));
2972
if (bpo->bpo_havefreed) {
2973
(void) printf(" %*s: object %llu, %llu local "
2974
"blkptrs, %llu freed, %llu subobjs in object %llu, "
2975
"%s (%s/%s comp)\n",
2976
indent * 8, name,
2977
(u_longlong_t)bpo->bpo_object,
2978
(u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2979
(u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2980
(u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2981
(u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2982
bytes, comp, uncomp);
2983
} else {
2984
(void) printf(" %*s: object %llu, %llu local "
2985
"blkptrs, %llu subobjs in object %llu, "
2986
"%s (%s/%s comp)\n",
2987
indent * 8, name,
2988
(u_longlong_t)bpo->bpo_object,
2989
(u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2990
(u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2991
(u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2992
bytes, comp, uncomp);
2993
}
2994
2995
for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
2996
uint64_t subobj;
2997
bpobj_t subbpo;
2998
int error;
2999
VERIFY0(dmu_read(bpo->bpo_os,
3000
bpo->bpo_phys->bpo_subobjs,
3001
i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3002
error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3003
if (error != 0) {
3004
(void) printf("ERROR %u while trying to open "
3005
"subobj id %llu\n",
3006
error, (u_longlong_t)subobj);
3007
corruption_found = B_TRUE;
3008
continue;
3009
}
3010
dump_full_bpobj(&subbpo, "subobj", indent + 1);
3011
bpobj_close(&subbpo);
3012
}
3013
} else {
3014
if (bpo->bpo_havefreed) {
3015
(void) printf(" %*s: object %llu, %llu blkptrs, "
3016
"%llu freed, %s\n",
3017
indent * 8, name,
3018
(u_longlong_t)bpo->bpo_object,
3019
(u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
3020
(u_longlong_t)bpo->bpo_phys->bpo_num_freed,
3021
bytes);
3022
} else {
3023
(void) printf(" %*s: object %llu, %llu blkptrs, "
3024
"%s\n",
3025
indent * 8, name,
3026
(u_longlong_t)bpo->bpo_object,
3027
(u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
3028
bytes);
3029
}
3030
}
3031
3032
if (dump_opt['d'] < 5)
3033
return;
3034
3035
3036
if (indent == 0) {
3037
(void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL);
3038
(void) printf("\n");
3039
}
3040
}
3041
3042
static int
3043
dump_bookmark(dsl_pool_t *dp, char *name, boolean_t print_redact,
3044
boolean_t print_list)
3045
{
3046
int err = 0;
3047
zfs_bookmark_phys_t prop;
3048
objset_t *mos = dp->dp_spa->spa_meta_objset;
3049
err = dsl_bookmark_lookup(dp, name, NULL, &prop);
3050
3051
if (err != 0) {
3052
return (err);
3053
}
3054
3055
(void) printf("\t#%s: ", strchr(name, '#') + 1);
3056
(void) printf("{guid: %llx creation_txg: %llu creation_time: "
3057
"%llu redaction_obj: %llu}\n", (u_longlong_t)prop.zbm_guid,
3058
(u_longlong_t)prop.zbm_creation_txg,
3059
(u_longlong_t)prop.zbm_creation_time,
3060
(u_longlong_t)prop.zbm_redaction_obj);
3061
3062
IMPLY(print_list, print_redact);
3063
if (!print_redact || prop.zbm_redaction_obj == 0)
3064
return (0);
3065
3066
redaction_list_t *rl;
3067
VERIFY0(dsl_redaction_list_hold_obj(dp,
3068
prop.zbm_redaction_obj, FTAG, &rl));
3069
3070
redaction_list_phys_t *rlp = rl->rl_phys;
3071
(void) printf("\tRedacted:\n\t\tProgress: ");
3072
if (rlp->rlp_last_object != UINT64_MAX ||
3073
rlp->rlp_last_blkid != UINT64_MAX) {
3074
(void) printf("%llu %llu (incomplete)\n",
3075
(u_longlong_t)rlp->rlp_last_object,
3076
(u_longlong_t)rlp->rlp_last_blkid);
3077
} else {
3078
(void) printf("complete\n");
3079
}
3080
(void) printf("\t\tSnapshots: [");
3081
for (unsigned int i = 0; i < rlp->rlp_num_snaps; i++) {
3082
if (i > 0)
3083
(void) printf(", ");
3084
(void) printf("%0llu",
3085
(u_longlong_t)rlp->rlp_snaps[i]);
3086
}
3087
(void) printf("]\n\t\tLength: %llu\n",
3088
(u_longlong_t)rlp->rlp_num_entries);
3089
3090
if (!print_list) {
3091
dsl_redaction_list_rele(rl, FTAG);
3092
return (0);
3093
}
3094
3095
if (rlp->rlp_num_entries == 0) {
3096
dsl_redaction_list_rele(rl, FTAG);
3097
(void) printf("\t\tRedaction List: []\n\n");
3098
return (0);
3099
}
3100
3101
redact_block_phys_t *rbp_buf;
3102
uint64_t size;
3103
dmu_object_info_t doi;
3104
3105
VERIFY0(dmu_object_info(mos, prop.zbm_redaction_obj, &doi));
3106
size = doi.doi_max_offset;
3107
rbp_buf = kmem_alloc(size, KM_SLEEP);
3108
3109
err = dmu_read(mos, prop.zbm_redaction_obj, 0, size,
3110
rbp_buf, 0);
3111
if (err != 0) {
3112
dsl_redaction_list_rele(rl, FTAG);
3113
kmem_free(rbp_buf, size);
3114
return (err);
3115
}
3116
3117
(void) printf("\t\tRedaction List: [{object: %llx, offset: "
3118
"%llx, blksz: %x, count: %llx}",
3119
(u_longlong_t)rbp_buf[0].rbp_object,
3120
(u_longlong_t)rbp_buf[0].rbp_blkid,
3121
(uint_t)(redact_block_get_size(&rbp_buf[0])),
3122
(u_longlong_t)redact_block_get_count(&rbp_buf[0]));
3123
3124
for (size_t i = 1; i < rlp->rlp_num_entries; i++) {
3125
(void) printf(",\n\t\t{object: %llx, offset: %llx, "
3126
"blksz: %x, count: %llx}",
3127
(u_longlong_t)rbp_buf[i].rbp_object,
3128
(u_longlong_t)rbp_buf[i].rbp_blkid,
3129
(uint_t)(redact_block_get_size(&rbp_buf[i])),
3130
(u_longlong_t)redact_block_get_count(&rbp_buf[i]));
3131
}
3132
dsl_redaction_list_rele(rl, FTAG);
3133
kmem_free(rbp_buf, size);
3134
(void) printf("]\n\n");
3135
return (0);
3136
}
3137
3138
static void
3139
dump_bookmarks(objset_t *os, int verbosity)
3140
{
3141
zap_cursor_t zc;
3142
zap_attribute_t *attrp;
3143
dsl_dataset_t *ds = dmu_objset_ds(os);
3144
dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3145
objset_t *mos = os->os_spa->spa_meta_objset;
3146
if (verbosity < 4)
3147
return;
3148
attrp = zap_attribute_alloc();
3149
dsl_pool_config_enter(dp, FTAG);
3150
3151
for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj);
3152
zap_cursor_retrieve(&zc, attrp) == 0;
3153
zap_cursor_advance(&zc)) {
3154
char osname[ZFS_MAX_DATASET_NAME_LEN];
3155
char buf[ZFS_MAX_DATASET_NAME_LEN];
3156
int len;
3157
dmu_objset_name(os, osname);
3158
len = snprintf(buf, sizeof (buf), "%s#%s", osname,
3159
attrp->za_name);
3160
VERIFY3S(len, <, ZFS_MAX_DATASET_NAME_LEN);
3161
(void) dump_bookmark(dp, buf, verbosity >= 5, verbosity >= 6);
3162
}
3163
zap_cursor_fini(&zc);
3164
dsl_pool_config_exit(dp, FTAG);
3165
zap_attribute_free(attrp);
3166
}
3167
3168
static void
3169
bpobj_count_refd(bpobj_t *bpo)
3170
{
3171
mos_obj_refd(bpo->bpo_object);
3172
3173
if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
3174
mos_obj_refd(bpo->bpo_phys->bpo_subobjs);
3175
for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
3176
uint64_t subobj;
3177
bpobj_t subbpo;
3178
int error;
3179
VERIFY0(dmu_read(bpo->bpo_os,
3180
bpo->bpo_phys->bpo_subobjs,
3181
i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3182
error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3183
if (error != 0) {
3184
(void) printf("ERROR %u while trying to open "
3185
"subobj id %llu\n",
3186
error, (u_longlong_t)subobj);
3187
corruption_found = B_TRUE;
3188
continue;
3189
}
3190
bpobj_count_refd(&subbpo);
3191
bpobj_close(&subbpo);
3192
}
3193
}
3194
}
3195
3196
static int
3197
dsl_deadlist_entry_count_refd(void *arg, dsl_deadlist_entry_t *dle)
3198
{
3199
spa_t *spa = arg;
3200
uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3201
if (dle->dle_bpobj.bpo_object != empty_bpobj)
3202
bpobj_count_refd(&dle->dle_bpobj);
3203
return (0);
3204
}
3205
3206
static int
3207
dsl_deadlist_entry_dump(void *arg, dsl_deadlist_entry_t *dle)
3208
{
3209
ASSERT0P(arg);
3210
if (dump_opt['d'] >= 5) {
3211
char buf[128];
3212
(void) snprintf(buf, sizeof (buf),
3213
"mintxg %llu -> obj %llu",
3214
(longlong_t)dle->dle_mintxg,
3215
(longlong_t)dle->dle_bpobj.bpo_object);
3216
3217
dump_full_bpobj(&dle->dle_bpobj, buf, 0);
3218
} else {
3219
(void) printf("mintxg %llu -> obj %llu\n",
3220
(longlong_t)dle->dle_mintxg,
3221
(longlong_t)dle->dle_bpobj.bpo_object);
3222
}
3223
return (0);
3224
}
3225
3226
static void
3227
dump_blkptr_list(dsl_deadlist_t *dl, const char *name)
3228
{
3229
char bytes[32];
3230
char comp[32];
3231
char uncomp[32];
3232
char entries[32];
3233
spa_t *spa = dmu_objset_spa(dl->dl_os);
3234
uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3235
3236
if (dl->dl_oldfmt) {
3237
if (dl->dl_bpobj.bpo_object != empty_bpobj)
3238
bpobj_count_refd(&dl->dl_bpobj);
3239
} else {
3240
mos_obj_refd(dl->dl_object);
3241
dsl_deadlist_iterate(dl, dsl_deadlist_entry_count_refd, spa);
3242
}
3243
3244
/* make sure nicenum has enough space */
3245
_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3246
_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
3247
_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
3248
_Static_assert(sizeof (entries) >= NN_NUMBUF_SZ, "entries truncated");
3249
3250
if (dump_opt['d'] < 3)
3251
return;
3252
3253
if (dl->dl_oldfmt) {
3254
dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0);
3255
return;
3256
}
3257
3258
zdb_nicenum(dl->dl_phys->dl_used, bytes, sizeof (bytes));
3259
zdb_nicenum(dl->dl_phys->dl_comp, comp, sizeof (comp));
3260
zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp, sizeof (uncomp));
3261
zdb_nicenum(avl_numnodes(&dl->dl_tree), entries, sizeof (entries));
3262
(void) printf("\n %s: %s (%s/%s comp), %s entries\n",
3263
name, bytes, comp, uncomp, entries);
3264
3265
if (dump_opt['d'] < 4)
3266
return;
3267
3268
(void) putchar('\n');
3269
3270
dsl_deadlist_iterate(dl, dsl_deadlist_entry_dump, NULL);
3271
}
3272
3273
static int
3274
verify_dd_livelist(objset_t *os)
3275
{
3276
uint64_t ll_used, used, ll_comp, comp, ll_uncomp, uncomp;
3277
dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3278
dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
3279
3280
ASSERT(!dmu_objset_is_snapshot(os));
3281
if (!dsl_deadlist_is_open(&dd->dd_livelist))
3282
return (0);
3283
3284
/* Iterate through the livelist to check for duplicates */
3285
dsl_deadlist_iterate(&dd->dd_livelist, sublivelist_verify_lightweight,
3286
NULL);
3287
3288
dsl_pool_config_enter(dp, FTAG);
3289
dsl_deadlist_space(&dd->dd_livelist, &ll_used,
3290
&ll_comp, &ll_uncomp);
3291
3292
dsl_dataset_t *origin_ds;
3293
ASSERT(dsl_pool_config_held(dp));
3294
VERIFY0(dsl_dataset_hold_obj(dp,
3295
dsl_dir_phys(dd)->dd_origin_obj, FTAG, &origin_ds));
3296
VERIFY0(dsl_dataset_space_written(origin_ds, os->os_dsl_dataset,
3297
&used, &comp, &uncomp));
3298
dsl_dataset_rele(origin_ds, FTAG);
3299
dsl_pool_config_exit(dp, FTAG);
3300
/*
3301
* It's possible that the dataset's uncomp space is larger than the
3302
* livelist's because livelists do not track embedded block pointers
3303
*/
3304
if (used != ll_used || comp != ll_comp || uncomp < ll_uncomp) {
3305
char nice_used[32], nice_comp[32], nice_uncomp[32];
3306
(void) printf("Discrepancy in space accounting:\n");
3307
zdb_nicenum(used, nice_used, sizeof (nice_used));
3308
zdb_nicenum(comp, nice_comp, sizeof (nice_comp));
3309
zdb_nicenum(uncomp, nice_uncomp, sizeof (nice_uncomp));
3310
(void) printf("dir: used %s, comp %s, uncomp %s\n",
3311
nice_used, nice_comp, nice_uncomp);
3312
zdb_nicenum(ll_used, nice_used, sizeof (nice_used));
3313
zdb_nicenum(ll_comp, nice_comp, sizeof (nice_comp));
3314
zdb_nicenum(ll_uncomp, nice_uncomp, sizeof (nice_uncomp));
3315
(void) printf("livelist: used %s, comp %s, uncomp %s\n",
3316
nice_used, nice_comp, nice_uncomp);
3317
return (1);
3318
}
3319
return (0);
3320
}
3321
3322
static char *key_material = NULL;
3323
3324
static boolean_t
3325
zdb_derive_key(dsl_dir_t *dd, uint8_t *key_out)
3326
{
3327
uint64_t keyformat, salt, iters;
3328
int i;
3329
unsigned char c;
3330
FILE *f;
3331
3332
VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3333
zfs_prop_to_name(ZFS_PROP_KEYFORMAT), sizeof (uint64_t),
3334
1, &keyformat));
3335
3336
switch (keyformat) {
3337
case ZFS_KEYFORMAT_HEX:
3338
for (i = 0; i < WRAPPING_KEY_LEN * 2; i += 2) {
3339
if (!isxdigit(key_material[i]) ||
3340
!isxdigit(key_material[i+1]))
3341
return (B_FALSE);
3342
if (sscanf(&key_material[i], "%02hhx", &c) != 1)
3343
return (B_FALSE);
3344
key_out[i / 2] = c;
3345
}
3346
break;
3347
3348
case ZFS_KEYFORMAT_PASSPHRASE:
3349
VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3350
dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT),
3351
sizeof (uint64_t), 1, &salt));
3352
VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3353
dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS),
3354
sizeof (uint64_t), 1, &iters));
3355
3356
if (PKCS5_PBKDF2_HMAC_SHA1(key_material, strlen(key_material),
3357
((uint8_t *)&salt), sizeof (uint64_t), iters,
3358
WRAPPING_KEY_LEN, key_out) != 1)
3359
return (B_FALSE);
3360
3361
break;
3362
3363
case ZFS_KEYFORMAT_RAW:
3364
if ((f = fopen(key_material, "r")) == NULL)
3365
return (B_FALSE);
3366
3367
if (fread(key_out, 1, WRAPPING_KEY_LEN, f) !=
3368
WRAPPING_KEY_LEN) {
3369
(void) fclose(f);
3370
return (B_FALSE);
3371
}
3372
3373
/* Check the key length */
3374
if (fgetc(f) != EOF) {
3375
(void) fclose(f);
3376
return (B_FALSE);
3377
}
3378
3379
(void) fclose(f);
3380
break;
3381
3382
default:
3383
fatal("no support for key format %u\n",
3384
(unsigned int) keyformat);
3385
}
3386
3387
return (B_TRUE);
3388
}
3389
3390
static char encroot[ZFS_MAX_DATASET_NAME_LEN];
3391
static boolean_t key_loaded = B_FALSE;
3392
3393
static void
3394
zdb_load_key(objset_t *os)
3395
{
3396
dsl_pool_t *dp;
3397
dsl_dir_t *dd, *rdd;
3398
uint8_t key[WRAPPING_KEY_LEN];
3399
uint64_t rddobj;
3400
int err;
3401
3402
dp = spa_get_dsl(os->os_spa);
3403
dd = os->os_dsl_dataset->ds_dir;
3404
3405
dsl_pool_config_enter(dp, FTAG);
3406
VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3407
DSL_CRYPTO_KEY_ROOT_DDOBJ, sizeof (uint64_t), 1, &rddobj));
3408
VERIFY0(dsl_dir_hold_obj(dd->dd_pool, rddobj, NULL, FTAG, &rdd));
3409
dsl_dir_name(rdd, encroot);
3410
dsl_dir_rele(rdd, FTAG);
3411
3412
if (!zdb_derive_key(dd, key))
3413
fatal("couldn't derive encryption key");
3414
3415
dsl_pool_config_exit(dp, FTAG);
3416
3417
ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_UNAVAILABLE);
3418
3419
dsl_crypto_params_t *dcp;
3420
nvlist_t *crypto_args;
3421
3422
crypto_args = fnvlist_alloc();
3423
fnvlist_add_uint8_array(crypto_args, "wkeydata",
3424
(uint8_t *)key, WRAPPING_KEY_LEN);
3425
VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE,
3426
NULL, crypto_args, &dcp));
3427
err = spa_keystore_load_wkey(encroot, dcp, B_FALSE);
3428
3429
dsl_crypto_params_free(dcp, (err != 0));
3430
fnvlist_free(crypto_args);
3431
3432
if (err != 0)
3433
fatal(
3434
"couldn't load encryption key for %s: %s",
3435
encroot, err == ZFS_ERR_CRYPTO_NOTSUP ?
3436
"crypto params not supported" : strerror(err));
3437
3438
ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_AVAILABLE);
3439
3440
printf("Unlocked encryption root: %s\n", encroot);
3441
key_loaded = B_TRUE;
3442
}
3443
3444
static void
3445
zdb_unload_key(void)
3446
{
3447
if (!key_loaded)
3448
return;
3449
3450
VERIFY0(spa_keystore_unload_wkey(encroot));
3451
key_loaded = B_FALSE;
3452
}
3453
3454
static avl_tree_t idx_tree;
3455
static avl_tree_t domain_tree;
3456
static boolean_t fuid_table_loaded;
3457
static objset_t *sa_os = NULL;
3458
static sa_attr_type_t *sa_attr_table = NULL;
3459
3460
static int
3461
open_objset(const char *path, const void *tag, objset_t **osp)
3462
{
3463
int err;
3464
uint64_t sa_attrs = 0;
3465
uint64_t version = 0;
3466
3467
VERIFY0P(sa_os);
3468
3469
/*
3470
* We can't own an objset if it's redacted. Therefore, we do this
3471
* dance: hold the objset, then acquire a long hold on its dataset, then
3472
* release the pool (which is held as part of holding the objset).
3473
*/
3474
3475
if (dump_opt['K']) {
3476
/* decryption requested, try to load keys */
3477
err = dmu_objset_hold(path, tag, osp);
3478
if (err != 0) {
3479
(void) fprintf(stderr, "failed to hold dataset "
3480
"'%s': %s\n",
3481
path, strerror(err));
3482
return (err);
3483
}
3484
dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3485
dsl_pool_rele(dmu_objset_pool(*osp), tag);
3486
3487
/* succeeds or dies */
3488
zdb_load_key(*osp);
3489
3490
/* release it all */
3491
dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3492
dsl_dataset_rele(dmu_objset_ds(*osp), tag);
3493
}
3494
3495
int ds_hold_flags = key_loaded ? DS_HOLD_FLAG_DECRYPT : 0;
3496
3497
err = dmu_objset_hold_flags(path, ds_hold_flags, tag, osp);
3498
if (err != 0) {
3499
(void) fprintf(stderr, "failed to hold dataset '%s': %s\n",
3500
path, strerror(err));
3501
return (err);
3502
}
3503
dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3504
dsl_pool_rele(dmu_objset_pool(*osp), tag);
3505
3506
if (dmu_objset_type(*osp) == DMU_OST_ZFS &&
3507
(key_loaded || !(*osp)->os_encrypted)) {
3508
(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR,
3509
8, 1, &version);
3510
if (version >= ZPL_VERSION_SA) {
3511
(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS,
3512
8, 1, &sa_attrs);
3513
}
3514
err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END,
3515
&sa_attr_table);
3516
if (err != 0) {
3517
(void) fprintf(stderr, "sa_setup failed: %s\n",
3518
strerror(err));
3519
dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3520
dsl_dataset_rele_flags(dmu_objset_ds(*osp),
3521
ds_hold_flags, tag);
3522
*osp = NULL;
3523
}
3524
}
3525
sa_os = *osp;
3526
3527
return (err);
3528
}
3529
3530
static void
3531
close_objset(objset_t *os, const void *tag)
3532
{
3533
VERIFY3P(os, ==, sa_os);
3534
if (os->os_sa != NULL)
3535
sa_tear_down(os);
3536
dsl_dataset_long_rele(dmu_objset_ds(os), tag);
3537
dsl_dataset_rele_flags(dmu_objset_ds(os),
3538
key_loaded ? DS_HOLD_FLAG_DECRYPT : 0, tag);
3539
sa_attr_table = NULL;
3540
sa_os = NULL;
3541
3542
zdb_unload_key();
3543
}
3544
3545
static void
3546
fuid_table_destroy(void)
3547
{
3548
if (fuid_table_loaded) {
3549
zfs_fuid_table_destroy(&idx_tree, &domain_tree);
3550
fuid_table_loaded = B_FALSE;
3551
}
3552
}
3553
3554
/*
3555
* Clean up DDT internal state. ddt_lookup() adds entries to ddt_tree, which on
3556
* a live pool are normally cleaned up during ddt_sync(). We can't do that (and
3557
* wouldn't want to anyway), but if we don't clean up the presence of stuff on
3558
* ddt_tree will trip asserts in ddt_table_free(). So, we clean up ourselves.
3559
*
3560
* Note that this is not a particularly efficient way to do this, but
3561
* ddt_remove() is the only public method that can do the work we need, and it
3562
* requires the right locks and etc to do the job. This is only ever called
3563
* during zdb shutdown so efficiency is not especially important.
3564
*/
3565
static void
3566
zdb_ddt_cleanup(spa_t *spa)
3567
{
3568
for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
3569
ddt_t *ddt = spa->spa_ddt[c];
3570
if (!ddt)
3571
continue;
3572
3573
spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3574
ddt_enter(ddt);
3575
ddt_entry_t *dde = avl_first(&ddt->ddt_tree), *next;
3576
while (dde) {
3577
next = AVL_NEXT(&ddt->ddt_tree, dde);
3578
dde->dde_io = NULL;
3579
ddt_remove(ddt, dde);
3580
dde = next;
3581
}
3582
ddt_exit(ddt);
3583
spa_config_exit(spa, SCL_CONFIG, FTAG);
3584
}
3585
}
3586
3587
static void
3588
zdb_exit(int reason)
3589
{
3590
if (spa != NULL)
3591
zdb_ddt_cleanup(spa);
3592
3593
if (os != NULL) {
3594
close_objset(os, FTAG);
3595
} else if (spa != NULL) {
3596
spa_close(spa, FTAG);
3597
}
3598
3599
fuid_table_destroy();
3600
3601
if (kernel_init_done)
3602
kernel_fini();
3603
3604
exit(reason);
3605
}
3606
3607
/*
3608
* print uid or gid information.
3609
* For normal POSIX id just the id is printed in decimal format.
3610
* For CIFS files with FUID the fuid is printed in hex followed by
3611
* the domain-rid string.
3612
*/
3613
static void
3614
print_idstr(uint64_t id, const char *id_type)
3615
{
3616
if (FUID_INDEX(id)) {
3617
const char *domain =
3618
zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id));
3619
(void) printf("\t%s %llx [%s-%d]\n", id_type,
3620
(u_longlong_t)id, domain, (int)FUID_RID(id));
3621
} else {
3622
(void) printf("\t%s %llu\n", id_type, (u_longlong_t)id);
3623
}
3624
3625
}
3626
3627
static void
3628
dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid)
3629
{
3630
uint32_t uid_idx, gid_idx;
3631
3632
uid_idx = FUID_INDEX(uid);
3633
gid_idx = FUID_INDEX(gid);
3634
3635
/* Load domain table, if not already loaded */
3636
if (!fuid_table_loaded && (uid_idx || gid_idx)) {
3637
uint64_t fuid_obj;
3638
3639
/* first find the fuid object. It lives in the master node */
3640
VERIFY0(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES,
3641
8, 1, &fuid_obj));
3642
zfs_fuid_avl_tree_create(&idx_tree, &domain_tree);
3643
(void) zfs_fuid_table_load(os, fuid_obj,
3644
&idx_tree, &domain_tree);
3645
fuid_table_loaded = B_TRUE;
3646
}
3647
3648
print_idstr(uid, "uid");
3649
print_idstr(gid, "gid");
3650
}
3651
3652
static void
3653
dump_znode_sa_xattr(sa_handle_t *hdl)
3654
{
3655
nvlist_t *sa_xattr;
3656
nvpair_t *elem = NULL;
3657
int sa_xattr_size = 0;
3658
int sa_xattr_entries = 0;
3659
int error;
3660
char *sa_xattr_packed;
3661
3662
error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size);
3663
if (error || sa_xattr_size == 0)
3664
return;
3665
3666
sa_xattr_packed = malloc(sa_xattr_size);
3667
if (sa_xattr_packed == NULL)
3668
return;
3669
3670
error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR],
3671
sa_xattr_packed, sa_xattr_size);
3672
if (error) {
3673
free(sa_xattr_packed);
3674
return;
3675
}
3676
3677
error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0);
3678
if (error) {
3679
free(sa_xattr_packed);
3680
return;
3681
}
3682
3683
while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL)
3684
sa_xattr_entries++;
3685
3686
(void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3687
sa_xattr_size, sa_xattr_entries);
3688
while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) {
3689
boolean_t can_print = !dump_opt['P'];
3690
uchar_t *value;
3691
uint_t cnt, idx;
3692
3693
(void) printf("\t\t%s = ", nvpair_name(elem));
3694
nvpair_value_byte_array(elem, &value, &cnt);
3695
3696
for (idx = 0; idx < cnt; ++idx) {
3697
if (!isprint(value[idx])) {
3698
can_print = B_FALSE;
3699
break;
3700
}
3701
}
3702
3703
for (idx = 0; idx < cnt; ++idx) {
3704
if (can_print)
3705
(void) putchar(value[idx]);
3706
else
3707
(void) printf("\\%3.3o", value[idx]);
3708
}
3709
(void) putchar('\n');
3710
}
3711
3712
nvlist_free(sa_xattr);
3713
free(sa_xattr_packed);
3714
}
3715
3716
static void
3717
dump_znode_symlink(sa_handle_t *hdl)
3718
{
3719
int sa_symlink_size = 0;
3720
char linktarget[MAXPATHLEN];
3721
int error;
3722
3723
error = sa_size(hdl, sa_attr_table[ZPL_SYMLINK], &sa_symlink_size);
3724
if (error || sa_symlink_size == 0) {
3725
return;
3726
}
3727
if (sa_symlink_size >= sizeof (linktarget)) {
3728
(void) printf("symlink size %d is too large\n",
3729
sa_symlink_size);
3730
return;
3731
}
3732
linktarget[sa_symlink_size] = '\0';
3733
if (sa_lookup(hdl, sa_attr_table[ZPL_SYMLINK],
3734
&linktarget, sa_symlink_size) == 0)
3735
(void) printf("\ttarget %s\n", linktarget);
3736
}
3737
3738
static void
3739
dump_znode(objset_t *os, uint64_t object, void *data, size_t size)
3740
{
3741
(void) data, (void) size;
3742
char path[MAXPATHLEN * 2]; /* allow for xattr and failure prefix */
3743
sa_handle_t *hdl;
3744
uint64_t xattr, rdev, gen;
3745
uint64_t uid, gid, mode, fsize, parent, links;
3746
uint64_t pflags;
3747
uint64_t acctm[2], modtm[2], chgtm[2], crtm[2];
3748
time_t z_crtime, z_atime, z_mtime, z_ctime;
3749
sa_bulk_attr_t bulk[12];
3750
int idx = 0;
3751
int error;
3752
3753
VERIFY3P(os, ==, sa_os);
3754
if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) {
3755
(void) printf("Failed to get handle for SA znode\n");
3756
return;
3757
}
3758
3759
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8);
3760
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8);
3761
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL,
3762
&links, 8);
3763
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8);
3764
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL,
3765
&mode, 8);
3766
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT],
3767
NULL, &parent, 8);
3768
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL,
3769
&fsize, 8);
3770
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL,
3771
acctm, 16);
3772
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL,
3773
modtm, 16);
3774
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL,
3775
crtm, 16);
3776
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL,
3777
chgtm, 16);
3778
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL,
3779
&pflags, 8);
3780
3781
if (sa_bulk_lookup(hdl, bulk, idx)) {
3782
(void) sa_handle_destroy(hdl);
3783
return;
3784
}
3785
3786
z_crtime = (time_t)crtm[0];
3787
z_atime = (time_t)acctm[0];
3788
z_mtime = (time_t)modtm[0];
3789
z_ctime = (time_t)chgtm[0];
3790
3791
if (dump_opt['d'] > 4) {
3792
error = zfs_obj_to_path(os, object, path, sizeof (path));
3793
if (error == ESTALE) {
3794
(void) snprintf(path, sizeof (path), "on delete queue");
3795
} else if (error != 0) {
3796
leaked_objects++;
3797
(void) snprintf(path, sizeof (path),
3798
"path not found, possibly leaked");
3799
}
3800
(void) printf("\tpath %s\n", path);
3801
}
3802
3803
if (S_ISLNK(mode))
3804
dump_znode_symlink(hdl);
3805
dump_uidgid(os, uid, gid);
3806
(void) printf("\tatime %s", ctime(&z_atime));
3807
(void) printf("\tmtime %s", ctime(&z_mtime));
3808
(void) printf("\tctime %s", ctime(&z_ctime));
3809
(void) printf("\tcrtime %s", ctime(&z_crtime));
3810
(void) printf("\tgen %llu\n", (u_longlong_t)gen);
3811
(void) printf("\tmode %llo\n", (u_longlong_t)mode);
3812
(void) printf("\tsize %llu\n", (u_longlong_t)fsize);
3813
(void) printf("\tparent %llu\n", (u_longlong_t)parent);
3814
(void) printf("\tlinks %llu\n", (u_longlong_t)links);
3815
(void) printf("\tpflags %llx\n", (u_longlong_t)pflags);
3816
if (dmu_objset_projectquota_enabled(os) && (pflags & ZFS_PROJID)) {
3817
uint64_t projid;
3818
3819
if (sa_lookup(hdl, sa_attr_table[ZPL_PROJID], &projid,
3820
sizeof (uint64_t)) == 0)
3821
(void) printf("\tprojid %llu\n", (u_longlong_t)projid);
3822
}
3823
if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr,
3824
sizeof (uint64_t)) == 0)
3825
(void) printf("\txattr %llu\n", (u_longlong_t)xattr);
3826
if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev,
3827
sizeof (uint64_t)) == 0)
3828
(void) printf("\trdev 0x%016llx\n", (u_longlong_t)rdev);
3829
dump_znode_sa_xattr(hdl);
3830
sa_handle_destroy(hdl);
3831
}
3832
3833
static void
3834
dump_acl(objset_t *os, uint64_t object, void *data, size_t size)
3835
{
3836
(void) os, (void) object, (void) data, (void) size;
3837
}
3838
3839
static void
3840
dump_dmu_objset(objset_t *os, uint64_t object, void *data, size_t size)
3841
{
3842
(void) os, (void) object, (void) data, (void) size;
3843
}
3844
3845
static object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = {
3846
dump_none, /* unallocated */
3847
dump_zap, /* object directory */
3848
dump_uint64, /* object array */
3849
dump_none, /* packed nvlist */
3850
dump_packed_nvlist, /* packed nvlist size */
3851
dump_none, /* bpobj */
3852
dump_bpobj, /* bpobj header */
3853
dump_none, /* SPA space map header */
3854
dump_none, /* SPA space map */
3855
dump_none, /* ZIL intent log */
3856
dump_dnode, /* DMU dnode */
3857
dump_dmu_objset, /* DMU objset */
3858
dump_dsl_dir, /* DSL directory */
3859
dump_zap, /* DSL directory child map */
3860
dump_zap, /* DSL dataset snap map */
3861
dump_zap, /* DSL props */
3862
dump_dsl_dataset, /* DSL dataset */
3863
dump_znode, /* ZFS znode */
3864
dump_acl, /* ZFS V0 ACL */
3865
dump_uint8, /* ZFS plain file */
3866
dump_zpldir, /* ZFS directory */
3867
dump_zap, /* ZFS master node */
3868
dump_zap, /* ZFS delete queue */
3869
dump_uint8, /* zvol object */
3870
dump_zap, /* zvol prop */
3871
dump_uint8, /* other uint8[] */
3872
dump_uint64, /* other uint64[] */
3873
dump_zap, /* other ZAP */
3874
dump_zap, /* persistent error log */
3875
dump_uint8, /* SPA history */
3876
dump_history_offsets, /* SPA history offsets */
3877
dump_zap, /* Pool properties */
3878
dump_zap, /* DSL permissions */
3879
dump_acl, /* ZFS ACL */
3880
dump_uint8, /* ZFS SYSACL */
3881
dump_none, /* FUID nvlist */
3882
dump_packed_nvlist, /* FUID nvlist size */
3883
dump_zap, /* DSL dataset next clones */
3884
dump_zap, /* DSL scrub queue */
3885
dump_zap, /* ZFS user/group/project used */
3886
dump_zap, /* ZFS user/group/project quota */
3887
dump_zap, /* snapshot refcount tags */
3888
dump_ddt_zap, /* DDT ZAP object */
3889
dump_zap, /* DDT statistics */
3890
dump_znode, /* SA object */
3891
dump_zap, /* SA Master Node */
3892
dump_sa_attrs, /* SA attribute registration */
3893
dump_sa_layouts, /* SA attribute layouts */
3894
dump_zap, /* DSL scrub translations */
3895
dump_none, /* fake dedup BP */
3896
dump_zap, /* deadlist */
3897
dump_none, /* deadlist hdr */
3898
dump_zap, /* dsl clones */
3899
dump_bpobj_subobjs, /* bpobj subobjs */
3900
dump_unknown, /* Unknown type, must be last */
3901
};
3902
3903
static boolean_t
3904
match_object_type(dmu_object_type_t obj_type, uint64_t flags)
3905
{
3906
boolean_t match = B_TRUE;
3907
3908
switch (obj_type) {
3909
case DMU_OT_DIRECTORY_CONTENTS:
3910
if (!(flags & ZOR_FLAG_DIRECTORY))
3911
match = B_FALSE;
3912
break;
3913
case DMU_OT_PLAIN_FILE_CONTENTS:
3914
if (!(flags & ZOR_FLAG_PLAIN_FILE))
3915
match = B_FALSE;
3916
break;
3917
case DMU_OT_SPACE_MAP:
3918
if (!(flags & ZOR_FLAG_SPACE_MAP))
3919
match = B_FALSE;
3920
break;
3921
default:
3922
if (strcmp(zdb_ot_name(obj_type), "zap") == 0) {
3923
if (!(flags & ZOR_FLAG_ZAP))
3924
match = B_FALSE;
3925
break;
3926
}
3927
3928
/*
3929
* If all bits except some of the supported flags are
3930
* set, the user combined the all-types flag (A) with
3931
* a negated flag to exclude some types (e.g. A-f to
3932
* show all object types except plain files).
3933
*/
3934
if ((flags | ZOR_SUPPORTED_FLAGS) != ZOR_FLAG_ALL_TYPES)
3935
match = B_FALSE;
3936
3937
break;
3938
}
3939
3940
return (match);
3941
}
3942
3943
static void
3944
dump_object(objset_t *os, uint64_t object, int verbosity,
3945
boolean_t *print_header, uint64_t *dnode_slots_used, uint64_t flags)
3946
{
3947
dmu_buf_t *db = NULL;
3948
dmu_object_info_t doi;
3949
dnode_t *dn;
3950
boolean_t dnode_held = B_FALSE;
3951
void *bonus = NULL;
3952
size_t bsize = 0;
3953
char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
3954
char bonus_size[32];
3955
char aux[50];
3956
int error;
3957
3958
/* make sure nicenum has enough space */
3959
_Static_assert(sizeof (iblk) >= NN_NUMBUF_SZ, "iblk truncated");
3960
_Static_assert(sizeof (dblk) >= NN_NUMBUF_SZ, "dblk truncated");
3961
_Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, "lsize truncated");
3962
_Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, "asize truncated");
3963
_Static_assert(sizeof (bonus_size) >= NN_NUMBUF_SZ,
3964
"bonus_size truncated");
3965
3966
if (*print_header) {
3967
(void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n",
3968
"Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3969
"lsize", "%full", "type");
3970
*print_header = 0;
3971
}
3972
3973
if (object == 0) {
3974
dn = DMU_META_DNODE(os);
3975
dmu_object_info_from_dnode(dn, &doi);
3976
} else {
3977
/*
3978
* Encrypted datasets will have sensitive bonus buffers
3979
* encrypted. Therefore we cannot hold the bonus buffer and
3980
* must hold the dnode itself instead.
3981
*/
3982
error = dmu_object_info(os, object, &doi);
3983
if (error)
3984
fatal("dmu_object_info() failed, errno %u", error);
3985
3986
if (!key_loaded && os->os_encrypted &&
3987
DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) {
3988
error = dnode_hold(os, object, FTAG, &dn);
3989
if (error)
3990
fatal("dnode_hold() failed, errno %u", error);
3991
dnode_held = B_TRUE;
3992
} else {
3993
error = dmu_bonus_hold(os, object, FTAG, &db);
3994
if (error)
3995
fatal("dmu_bonus_hold(%llu) failed, errno %u",
3996
object, error);
3997
bonus = db->db_data;
3998
bsize = db->db_size;
3999
dn = DB_DNODE((dmu_buf_impl_t *)db);
4000
}
4001
}
4002
4003
/*
4004
* Default to showing all object types if no flags were specified.
4005
*/
4006
if (flags != 0 && flags != ZOR_FLAG_ALL_TYPES &&
4007
!match_object_type(doi.doi_type, flags))
4008
goto out;
4009
4010
if (dnode_slots_used)
4011
*dnode_slots_used = doi.doi_dnodesize / DNODE_MIN_SIZE;
4012
4013
zdb_nicenum(doi.doi_metadata_block_size, iblk, sizeof (iblk));
4014
zdb_nicenum(doi.doi_data_block_size, dblk, sizeof (dblk));
4015
zdb_nicenum(doi.doi_max_offset, lsize, sizeof (lsize));
4016
zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize, sizeof (asize));
4017
zdb_nicenum(doi.doi_bonus_size, bonus_size, sizeof (bonus_size));
4018
zdb_nicenum(doi.doi_dnodesize, dnsize, sizeof (dnsize));
4019
(void) snprintf(fill, sizeof (fill), "%6.2f", 100.0 *
4020
doi.doi_fill_count * doi.doi_data_block_size / (object == 0 ?
4021
DNODES_PER_BLOCK : 1) / doi.doi_max_offset);
4022
4023
aux[0] = '\0';
4024
4025
if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) {
4026
(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
4027
" (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum));
4028
}
4029
4030
if (doi.doi_compress == ZIO_COMPRESS_INHERIT &&
4031
ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) {
4032
const char *compname = NULL;
4033
if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION,
4034
ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel),
4035
&compname) == 0) {
4036
(void) snprintf(aux + strlen(aux),
4037
sizeof (aux) - strlen(aux), " (Z=inherit=%s)",
4038
compname);
4039
} else {
4040
(void) snprintf(aux + strlen(aux),
4041
sizeof (aux) - strlen(aux),
4042
" (Z=inherit=%s-unknown)",
4043
ZDB_COMPRESS_NAME(os->os_compress));
4044
}
4045
} else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) {
4046
(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
4047
" (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress));
4048
} else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
4049
(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
4050
" (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress));
4051
}
4052
4053
(void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n",
4054
(u_longlong_t)object, doi.doi_indirection, iblk, dblk,
4055
asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux);
4056
4057
if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
4058
(void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n",
4059
"", "", "", "", "", "", bonus_size, "bonus",
4060
zdb_ot_name(doi.doi_bonus_type));
4061
}
4062
4063
if (verbosity >= 4) {
4064
(void) printf("\tdnode flags: %s%s%s%s\n",
4065
(dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ?
4066
"USED_BYTES " : "",
4067
(dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ?
4068
"USERUSED_ACCOUNTED " : "",
4069
(dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ?
4070
"USEROBJUSED_ACCOUNTED " : "",
4071
(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ?
4072
"SPILL_BLKPTR" : "");
4073
(void) printf("\tdnode maxblkid: %llu\n",
4074
(longlong_t)dn->dn_phys->dn_maxblkid);
4075
4076
if (!dnode_held) {
4077
object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os,
4078
object, bonus, bsize);
4079
} else {
4080
(void) printf("\t\t(bonus encrypted)\n");
4081
}
4082
4083
if (key_loaded ||
4084
(!os->os_encrypted || !DMU_OT_IS_ENCRYPTED(doi.doi_type))) {
4085
object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object,
4086
NULL, 0);
4087
} else {
4088
(void) printf("\t\t(object encrypted)\n");
4089
}
4090
4091
*print_header = B_TRUE;
4092
}
4093
4094
if (verbosity >= 5) {
4095
if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
4096
char blkbuf[BP_SPRINTF_LEN];
4097
snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
4098
DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE);
4099
(void) printf("\nSpill block: %s\n", blkbuf);
4100
}
4101
dump_indirect(dn);
4102
}
4103
4104
if (verbosity >= 5) {
4105
/*
4106
* Report the list of segments that comprise the object.
4107
*/
4108
uint64_t start = 0;
4109
uint64_t end;
4110
uint64_t blkfill = 1;
4111
int minlvl = 1;
4112
4113
if (dn->dn_type == DMU_OT_DNODE) {
4114
minlvl = 0;
4115
blkfill = DNODES_PER_BLOCK;
4116
}
4117
4118
for (;;) {
4119
char segsize[32];
4120
/* make sure nicenum has enough space */
4121
_Static_assert(sizeof (segsize) >= NN_NUMBUF_SZ,
4122
"segsize truncated");
4123
error = dnode_next_offset(dn,
4124
0, &start, minlvl, blkfill, 0);
4125
if (error)
4126
break;
4127
end = start;
4128
error = dnode_next_offset(dn,
4129
DNODE_FIND_HOLE, &end, minlvl, blkfill, 0);
4130
zdb_nicenum(end - start, segsize, sizeof (segsize));
4131
(void) printf("\t\tsegment [%016llx, %016llx)"
4132
" size %5s\n", (u_longlong_t)start,
4133
(u_longlong_t)end, segsize);
4134
if (error)
4135
break;
4136
start = end;
4137
}
4138
}
4139
4140
out:
4141
if (db != NULL)
4142
dmu_buf_rele(db, FTAG);
4143
if (dnode_held)
4144
dnode_rele(dn, FTAG);
4145
}
4146
4147
static void
4148
count_dir_mos_objects(dsl_dir_t *dd)
4149
{
4150
mos_obj_refd(dd->dd_object);
4151
mos_obj_refd(dsl_dir_phys(dd)->dd_child_dir_zapobj);
4152
mos_obj_refd(dsl_dir_phys(dd)->dd_deleg_zapobj);
4153
mos_obj_refd(dsl_dir_phys(dd)->dd_props_zapobj);
4154
mos_obj_refd(dsl_dir_phys(dd)->dd_clones);
4155
4156
/*
4157
* The dd_crypto_obj can be referenced by multiple dsl_dir's.
4158
* Ignore the references after the first one.
4159
*/
4160
mos_obj_refd_multiple(dd->dd_crypto_obj);
4161
}
4162
4163
static void
4164
count_ds_mos_objects(dsl_dataset_t *ds)
4165
{
4166
mos_obj_refd(ds->ds_object);
4167
mos_obj_refd(dsl_dataset_phys(ds)->ds_next_clones_obj);
4168
mos_obj_refd(dsl_dataset_phys(ds)->ds_props_obj);
4169
mos_obj_refd(dsl_dataset_phys(ds)->ds_userrefs_obj);
4170
mos_obj_refd(dsl_dataset_phys(ds)->ds_snapnames_zapobj);
4171
mos_obj_refd(ds->ds_bookmarks_obj);
4172
4173
if (!dsl_dataset_is_snapshot(ds)) {
4174
count_dir_mos_objects(ds->ds_dir);
4175
}
4176
}
4177
4178
static const char *const objset_types[DMU_OST_NUMTYPES] = {
4179
"NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
4180
4181
/*
4182
* Parse a string denoting a range of object IDs of the form
4183
* <start>[:<end>[:flags]], and store the results in zor.
4184
* Return 0 on success. On error, return 1 and update the msg
4185
* pointer to point to a descriptive error message.
4186
*/
4187
static int
4188
parse_object_range(char *range, zopt_object_range_t *zor, const char **msg)
4189
{
4190
uint64_t flags = 0;
4191
char *p, *s, *dup, *flagstr, *tmp = NULL;
4192
size_t len;
4193
int i;
4194
int rc = 0;
4195
4196
if (strchr(range, ':') == NULL) {
4197
zor->zor_obj_start = strtoull(range, &p, 0);
4198
if (*p != '\0') {
4199
*msg = "Invalid characters in object ID";
4200
rc = 1;
4201
}
4202
zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4203
zor->zor_obj_end = zor->zor_obj_start;
4204
return (rc);
4205
}
4206
4207
if (strchr(range, ':') == range) {
4208
*msg = "Invalid leading colon";
4209
rc = 1;
4210
return (rc);
4211
}
4212
4213
len = strlen(range);
4214
if (range[len - 1] == ':') {
4215
*msg = "Invalid trailing colon";
4216
rc = 1;
4217
return (rc);
4218
}
4219
4220
dup = strdup(range);
4221
s = strtok_r(dup, ":", &tmp);
4222
zor->zor_obj_start = strtoull(s, &p, 0);
4223
4224
if (*p != '\0') {
4225
*msg = "Invalid characters in start object ID";
4226
rc = 1;
4227
goto out;
4228
}
4229
4230
s = strtok_r(NULL, ":", &tmp);
4231
zor->zor_obj_end = strtoull(s, &p, 0);
4232
4233
if (*p != '\0') {
4234
*msg = "Invalid characters in end object ID";
4235
rc = 1;
4236
goto out;
4237
}
4238
4239
if (zor->zor_obj_start > zor->zor_obj_end) {
4240
*msg = "Start object ID may not exceed end object ID";
4241
rc = 1;
4242
goto out;
4243
}
4244
4245
s = strtok_r(NULL, ":", &tmp);
4246
if (s == NULL) {
4247
zor->zor_flags = ZOR_FLAG_ALL_TYPES;
4248
goto out;
4249
} else if (strtok_r(NULL, ":", &tmp) != NULL) {
4250
*msg = "Invalid colon-delimited field after flags";
4251
rc = 1;
4252
goto out;
4253
}
4254
4255
flagstr = s;
4256
for (i = 0; flagstr[i]; i++) {
4257
int bit;
4258
boolean_t negation = (flagstr[i] == '-');
4259
4260
if (negation) {
4261
i++;
4262
if (flagstr[i] == '\0') {
4263
*msg = "Invalid trailing negation operator";
4264
rc = 1;
4265
goto out;
4266
}
4267
}
4268
bit = flagbits[(uchar_t)flagstr[i]];
4269
if (bit == 0) {
4270
*msg = "Invalid flag";
4271
rc = 1;
4272
goto out;
4273
}
4274
if (negation)
4275
flags &= ~bit;
4276
else
4277
flags |= bit;
4278
}
4279
zor->zor_flags = flags;
4280
4281
zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4282
zor->zor_obj_end = ZDB_MAP_OBJECT_ID(zor->zor_obj_end);
4283
4284
out:
4285
free(dup);
4286
return (rc);
4287
}
4288
4289
static void
4290
dump_objset(objset_t *os)
4291
{
4292
dmu_objset_stats_t dds = { 0 };
4293
uint64_t object, object_count;
4294
uint64_t refdbytes, usedobjs, scratch;
4295
char numbuf[32];
4296
char blkbuf[BP_SPRINTF_LEN + 20];
4297
char osname[ZFS_MAX_DATASET_NAME_LEN];
4298
const char *type = "UNKNOWN";
4299
int verbosity = dump_opt['d'];
4300
boolean_t print_header;
4301
unsigned i;
4302
int error;
4303
uint64_t total_slots_used = 0;
4304
uint64_t max_slot_used = 0;
4305
uint64_t dnode_slots;
4306
uint64_t obj_start;
4307
uint64_t obj_end;
4308
uint64_t flags;
4309
4310
/* make sure nicenum has enough space */
4311
_Static_assert(sizeof (numbuf) >= NN_NUMBUF_SZ, "numbuf truncated");
4312
4313
dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
4314
dmu_objset_fast_stat(os, &dds);
4315
dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
4316
4317
print_header = B_TRUE;
4318
4319
if (dds.dds_type < DMU_OST_NUMTYPES)
4320
type = objset_types[dds.dds_type];
4321
4322
if (dds.dds_type == DMU_OST_META) {
4323
dds.dds_creation_txg = TXG_INITIAL;
4324
usedobjs = BP_GET_FILL(os->os_rootbp);
4325
refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)->
4326
dd_used_bytes;
4327
} else {
4328
dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
4329
}
4330
4331
ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
4332
4333
zdb_nicenum(refdbytes, numbuf, sizeof (numbuf));
4334
4335
if (verbosity >= 4) {
4336
(void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp ");
4337
(void) snprintf_blkptr(blkbuf + strlen(blkbuf),
4338
sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp);
4339
} else {
4340
blkbuf[0] = '\0';
4341
}
4342
4343
dmu_objset_name(os, osname);
4344
4345
(void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
4346
"%s, %llu objects%s%s\n",
4347
osname, type, (u_longlong_t)dmu_objset_id(os),
4348
(u_longlong_t)dds.dds_creation_txg,
4349
numbuf, (u_longlong_t)usedobjs, blkbuf,
4350
(dds.dds_inconsistent) ? " (inconsistent)" : "");
4351
4352
for (i = 0; i < zopt_object_args; i++) {
4353
obj_start = zopt_object_ranges[i].zor_obj_start;
4354
obj_end = zopt_object_ranges[i].zor_obj_end;
4355
flags = zopt_object_ranges[i].zor_flags;
4356
4357
object = obj_start;
4358
if (object == 0 || obj_start == obj_end)
4359
dump_object(os, object, verbosity, &print_header, NULL,
4360
flags);
4361
else
4362
object--;
4363
4364
while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) &&
4365
object <= obj_end) {
4366
dump_object(os, object, verbosity, &print_header, NULL,
4367
flags);
4368
}
4369
}
4370
4371
if (zopt_object_args > 0) {
4372
(void) printf("\n");
4373
return;
4374
}
4375
4376
if (dump_opt['i'] != 0 || verbosity >= 2)
4377
dump_intent_log(dmu_objset_zil(os));
4378
4379
if (dmu_objset_ds(os) != NULL) {
4380
dsl_dataset_t *ds = dmu_objset_ds(os);
4381
dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
4382
if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
4383
!dmu_objset_is_snapshot(os)) {
4384
dump_blkptr_list(&ds->ds_dir->dd_livelist, "Livelist");
4385
if (verify_dd_livelist(os) != 0)
4386
fatal("livelist is incorrect");
4387
}
4388
4389
if (dsl_dataset_remap_deadlist_exists(ds)) {
4390
(void) printf("ds_remap_deadlist:\n");
4391
dump_blkptr_list(&ds->ds_remap_deadlist, "Deadlist");
4392
}
4393
count_ds_mos_objects(ds);
4394
}
4395
4396
if (dmu_objset_ds(os) != NULL)
4397
dump_bookmarks(os, verbosity);
4398
4399
if (verbosity < 2)
4400
return;
4401
4402
if (BP_IS_HOLE(os->os_rootbp))
4403
return;
4404
4405
dump_object(os, 0, verbosity, &print_header, NULL, 0);
4406
object_count = 0;
4407
if (DMU_USERUSED_DNODE(os) != NULL &&
4408
DMU_USERUSED_DNODE(os)->dn_type != 0) {
4409
dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header,
4410
NULL, 0);
4411
dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header,
4412
NULL, 0);
4413
}
4414
4415
if (DMU_PROJECTUSED_DNODE(os) != NULL &&
4416
DMU_PROJECTUSED_DNODE(os)->dn_type != 0)
4417
dump_object(os, DMU_PROJECTUSED_OBJECT, verbosity,
4418
&print_header, NULL, 0);
4419
4420
object = 0;
4421
while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
4422
dump_object(os, object, verbosity, &print_header, &dnode_slots,
4423
0);
4424
object_count++;
4425
total_slots_used += dnode_slots;
4426
max_slot_used = object + dnode_slots - 1;
4427
}
4428
4429
(void) printf("\n");
4430
4431
(void) printf(" Dnode slots:\n");
4432
(void) printf("\tTotal used: %10llu\n",
4433
(u_longlong_t)total_slots_used);
4434
(void) printf("\tMax used: %10llu\n",
4435
(u_longlong_t)max_slot_used);
4436
(void) printf("\tPercent empty: %10lf\n",
4437
(double)(max_slot_used - total_slots_used)*100 /
4438
(double)max_slot_used);
4439
(void) printf("\n");
4440
4441
if (error != ESRCH) {
4442
(void) fprintf(stderr, "dmu_object_next() = %d\n", error);
4443
abort();
4444
}
4445
4446
ASSERT3U(object_count, ==, usedobjs);
4447
4448
if (leaked_objects != 0) {
4449
(void) printf("%d potentially leaked objects detected\n",
4450
leaked_objects);
4451
leaked_objects = 0;
4452
}
4453
}
4454
4455
static void
4456
dump_uberblock(uberblock_t *ub, const char *header, const char *footer)
4457
{
4458
time_t timestamp = ub->ub_timestamp;
4459
4460
(void) printf("%s", header ? header : "");
4461
(void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic);
4462
(void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version);
4463
(void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg);
4464
(void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum);
4465
(void) printf("\ttimestamp = %llu UTC = %s",
4466
(u_longlong_t)ub->ub_timestamp, ctime(&timestamp));
4467
4468
char blkbuf[BP_SPRINTF_LEN];
4469
snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4470
(void) printf("\tbp = %s\n", blkbuf);
4471
4472
(void) printf("\tmmp_magic = %016llx\n",
4473
(u_longlong_t)ub->ub_mmp_magic);
4474
if (MMP_VALID(ub)) {
4475
(void) printf("\tmmp_delay = %0llu\n",
4476
(u_longlong_t)ub->ub_mmp_delay);
4477
if (MMP_SEQ_VALID(ub))
4478
(void) printf("\tmmp_seq = %u\n",
4479
(unsigned int) MMP_SEQ(ub));
4480
if (MMP_FAIL_INT_VALID(ub))
4481
(void) printf("\tmmp_fail = %u\n",
4482
(unsigned int) MMP_FAIL_INT(ub));
4483
if (MMP_INTERVAL_VALID(ub))
4484
(void) printf("\tmmp_write = %u\n",
4485
(unsigned int) MMP_INTERVAL(ub));
4486
/* After MMP_* to make summarize_uberblock_mmp cleaner */
4487
(void) printf("\tmmp_valid = %x\n",
4488
(unsigned int) ub->ub_mmp_config & 0xFF);
4489
}
4490
4491
if (dump_opt['u'] >= 4) {
4492
char blkbuf[BP_SPRINTF_LEN];
4493
snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4494
(void) printf("\trootbp = %s\n", blkbuf);
4495
}
4496
(void) printf("\tcheckpoint_txg = %llu\n",
4497
(u_longlong_t)ub->ub_checkpoint_txg);
4498
4499
(void) printf("\traidz_reflow state=%u off=%llu\n",
4500
(int)RRSS_GET_STATE(ub),
4501
(u_longlong_t)RRSS_GET_OFFSET(ub));
4502
4503
(void) printf("%s", footer ? footer : "");
4504
}
4505
4506
static void
4507
dump_config(spa_t *spa)
4508
{
4509
dmu_buf_t *db;
4510
size_t nvsize = 0;
4511
int error = 0;
4512
4513
4514
error = dmu_bonus_hold(spa->spa_meta_objset,
4515
spa->spa_config_object, FTAG, &db);
4516
4517
if (error == 0) {
4518
nvsize = *(uint64_t *)db->db_data;
4519
dmu_buf_rele(db, FTAG);
4520
4521
(void) printf("\nMOS Configuration:\n");
4522
dump_packed_nvlist(spa->spa_meta_objset,
4523
spa->spa_config_object, (void *)&nvsize, 1);
4524
} else {
4525
(void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d",
4526
(u_longlong_t)spa->spa_config_object, error);
4527
}
4528
}
4529
4530
static void
4531
dump_cachefile(const char *cachefile)
4532
{
4533
int fd;
4534
struct stat64 statbuf;
4535
char *buf;
4536
nvlist_t *config;
4537
4538
if ((fd = open64(cachefile, O_RDONLY)) < 0) {
4539
(void) printf("cannot open '%s': %s\n", cachefile,
4540
strerror(errno));
4541
zdb_exit(1);
4542
}
4543
4544
if (fstat64(fd, &statbuf) != 0) {
4545
(void) printf("failed to stat '%s': %s\n", cachefile,
4546
strerror(errno));
4547
zdb_exit(1);
4548
}
4549
4550
if ((buf = malloc(statbuf.st_size)) == NULL) {
4551
(void) fprintf(stderr, "failed to allocate %llu bytes\n",
4552
(u_longlong_t)statbuf.st_size);
4553
zdb_exit(1);
4554
}
4555
4556
if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
4557
(void) fprintf(stderr, "failed to read %llu bytes\n",
4558
(u_longlong_t)statbuf.st_size);
4559
zdb_exit(1);
4560
}
4561
4562
(void) close(fd);
4563
4564
if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) {
4565
(void) fprintf(stderr, "failed to unpack nvlist\n");
4566
zdb_exit(1);
4567
}
4568
4569
free(buf);
4570
4571
dump_nvlist(config, 0);
4572
4573
nvlist_free(config);
4574
}
4575
4576
/*
4577
* ZFS label nvlist stats
4578
*/
4579
typedef struct zdb_nvl_stats {
4580
int zns_list_count;
4581
int zns_leaf_count;
4582
size_t zns_leaf_largest;
4583
size_t zns_leaf_total;
4584
nvlist_t *zns_string;
4585
nvlist_t *zns_uint64;
4586
nvlist_t *zns_boolean;
4587
} zdb_nvl_stats_t;
4588
4589
static void
4590
collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats)
4591
{
4592
nvlist_t *list, **array;
4593
nvpair_t *nvp = NULL;
4594
const char *name;
4595
uint_t i, items;
4596
4597
stats->zns_list_count++;
4598
4599
while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4600
name = nvpair_name(nvp);
4601
4602
switch (nvpair_type(nvp)) {
4603
case DATA_TYPE_STRING:
4604
fnvlist_add_string(stats->zns_string, name,
4605
fnvpair_value_string(nvp));
4606
break;
4607
case DATA_TYPE_UINT64:
4608
fnvlist_add_uint64(stats->zns_uint64, name,
4609
fnvpair_value_uint64(nvp));
4610
break;
4611
case DATA_TYPE_BOOLEAN:
4612
fnvlist_add_boolean(stats->zns_boolean, name);
4613
break;
4614
case DATA_TYPE_NVLIST:
4615
if (nvpair_value_nvlist(nvp, &list) == 0)
4616
collect_nvlist_stats(list, stats);
4617
break;
4618
case DATA_TYPE_NVLIST_ARRAY:
4619
if (nvpair_value_nvlist_array(nvp, &array, &items) != 0)
4620
break;
4621
4622
for (i = 0; i < items; i++) {
4623
collect_nvlist_stats(array[i], stats);
4624
4625
/* collect stats on leaf vdev */
4626
if (strcmp(name, "children") == 0) {
4627
size_t size;
4628
4629
(void) nvlist_size(array[i], &size,
4630
NV_ENCODE_XDR);
4631
stats->zns_leaf_total += size;
4632
if (size > stats->zns_leaf_largest)
4633
stats->zns_leaf_largest = size;
4634
stats->zns_leaf_count++;
4635
}
4636
}
4637
break;
4638
default:
4639
(void) printf("skip type %d!\n", (int)nvpair_type(nvp));
4640
}
4641
}
4642
}
4643
4644
static void
4645
dump_nvlist_stats(nvlist_t *nvl, size_t cap)
4646
{
4647
zdb_nvl_stats_t stats = { 0 };
4648
size_t size, sum = 0, total;
4649
size_t noise;
4650
4651
/* requires nvlist with non-unique names for stat collection */
4652
VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0));
4653
VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0));
4654
VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0));
4655
VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR));
4656
4657
(void) printf("\n\nZFS Label NVList Config Stats:\n");
4658
4659
VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR));
4660
(void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4661
(int)total, (int)(cap - total), 100.0 * total / cap);
4662
4663
collect_nvlist_stats(nvl, &stats);
4664
4665
VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR));
4666
size -= noise;
4667
sum += size;
4668
(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4669
(int)fnvlist_num_pairs(stats.zns_uint64),
4670
(int)size, 100.0 * size / total);
4671
4672
VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR));
4673
size -= noise;
4674
sum += size;
4675
(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4676
(int)fnvlist_num_pairs(stats.zns_string),
4677
(int)size, 100.0 * size / total);
4678
4679
VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR));
4680
size -= noise;
4681
sum += size;
4682
(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4683
(int)fnvlist_num_pairs(stats.zns_boolean),
4684
(int)size, 100.0 * size / total);
4685
4686
size = total - sum; /* treat remainder as nvlist overhead */
4687
(void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4688
stats.zns_list_count, (int)size, 100.0 * size / total);
4689
4690
if (stats.zns_leaf_count > 0) {
4691
size_t average = stats.zns_leaf_total / stats.zns_leaf_count;
4692
4693
(void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4694
stats.zns_leaf_count, (int)average);
4695
(void) printf("%24d bytes largest\n",
4696
(int)stats.zns_leaf_largest);
4697
4698
if (dump_opt['l'] >= 3 && average > 0)
4699
(void) printf(" space for %d additional leaf vdevs\n",
4700
(int)((cap - total) / average));
4701
}
4702
(void) printf("\n");
4703
4704
nvlist_free(stats.zns_string);
4705
nvlist_free(stats.zns_uint64);
4706
nvlist_free(stats.zns_boolean);
4707
}
4708
4709
typedef struct cksum_record {
4710
zio_cksum_t cksum;
4711
boolean_t labels[VDEV_LABELS];
4712
avl_node_t link;
4713
} cksum_record_t;
4714
4715
static int
4716
cksum_record_compare(const void *x1, const void *x2)
4717
{
4718
const cksum_record_t *l = (cksum_record_t *)x1;
4719
const cksum_record_t *r = (cksum_record_t *)x2;
4720
int arraysize = ARRAY_SIZE(l->cksum.zc_word);
4721
int difference = 0;
4722
4723
for (int i = 0; i < arraysize; i++) {
4724
difference = TREE_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]);
4725
if (difference)
4726
break;
4727
}
4728
4729
return (difference);
4730
}
4731
4732
static cksum_record_t *
4733
cksum_record_alloc(zio_cksum_t *cksum, int l)
4734
{
4735
cksum_record_t *rec;
4736
4737
rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL);
4738
rec->cksum = *cksum;
4739
rec->labels[l] = B_TRUE;
4740
4741
return (rec);
4742
}
4743
4744
static cksum_record_t *
4745
cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum)
4746
{
4747
cksum_record_t lookup = { .cksum = *cksum };
4748
avl_index_t where;
4749
4750
return (avl_find(tree, &lookup, &where));
4751
}
4752
4753
static cksum_record_t *
4754
cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l)
4755
{
4756
cksum_record_t *rec;
4757
4758
rec = cksum_record_lookup(tree, cksum);
4759
if (rec) {
4760
rec->labels[l] = B_TRUE;
4761
} else {
4762
rec = cksum_record_alloc(cksum, l);
4763
avl_add(tree, rec);
4764
}
4765
4766
return (rec);
4767
}
4768
4769
static int
4770
first_label(cksum_record_t *rec)
4771
{
4772
for (int i = 0; i < VDEV_LABELS; i++)
4773
if (rec->labels[i])
4774
return (i);
4775
4776
return (-1);
4777
}
4778
4779
static void
4780
print_label_numbers(const char *prefix, const cksum_record_t *rec)
4781
{
4782
fputs(prefix, stdout);
4783
for (int i = 0; i < VDEV_LABELS; i++)
4784
if (rec->labels[i] == B_TRUE)
4785
printf("%d ", i);
4786
putchar('\n');
4787
}
4788
4789
#define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4790
4791
typedef struct zdb_label {
4792
vdev_label_t label;
4793
uint64_t label_offset;
4794
nvlist_t *config_nv;
4795
cksum_record_t *config;
4796
cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT];
4797
boolean_t header_printed;
4798
boolean_t read_failed;
4799
boolean_t cksum_valid;
4800
} zdb_label_t;
4801
4802
static void
4803
print_label_header(zdb_label_t *label, int l)
4804
{
4805
4806
if (dump_opt['q'])
4807
return;
4808
4809
if (label->header_printed == B_TRUE)
4810
return;
4811
4812
(void) printf("------------------------------------\n");
4813
(void) printf("LABEL %d %s\n", l,
4814
label->cksum_valid ? "" : "(Bad label cksum)");
4815
(void) printf("------------------------------------\n");
4816
4817
label->header_printed = B_TRUE;
4818
}
4819
4820
static void
4821
print_l2arc_header(void)
4822
{
4823
(void) printf("------------------------------------\n");
4824
(void) printf("L2ARC device header\n");
4825
(void) printf("------------------------------------\n");
4826
}
4827
4828
static void
4829
print_l2arc_log_blocks(void)
4830
{
4831
(void) printf("------------------------------------\n");
4832
(void) printf("L2ARC device log blocks\n");
4833
(void) printf("------------------------------------\n");
4834
}
4835
4836
static void
4837
dump_l2arc_log_entries(uint64_t log_entries,
4838
l2arc_log_ent_phys_t *le, uint64_t i)
4839
{
4840
for (int j = 0; j < log_entries; j++) {
4841
dva_t dva = le[j].le_dva;
4842
(void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4843
"vdev: %llu, offset: %llu\n",
4844
(u_longlong_t)i, j + 1,
4845
(u_longlong_t)DVA_GET_ASIZE(&dva),
4846
(u_longlong_t)DVA_GET_VDEV(&dva),
4847
(u_longlong_t)DVA_GET_OFFSET(&dva));
4848
(void) printf("|\t\t\t\tbirth: %llu\n",
4849
(u_longlong_t)le[j].le_birth);
4850
(void) printf("|\t\t\t\tlsize: %llu\n",
4851
(u_longlong_t)L2BLK_GET_LSIZE((&le[j])->le_prop));
4852
(void) printf("|\t\t\t\tpsize: %llu\n",
4853
(u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop));
4854
(void) printf("|\t\t\t\tcompr: %llu\n",
4855
(u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop));
4856
(void) printf("|\t\t\t\tcomplevel: %llu\n",
4857
(u_longlong_t)(&le[j])->le_complevel);
4858
(void) printf("|\t\t\t\ttype: %llu\n",
4859
(u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop));
4860
(void) printf("|\t\t\t\tprotected: %llu\n",
4861
(u_longlong_t)L2BLK_GET_PROTECTED((&le[j])->le_prop));
4862
(void) printf("|\t\t\t\tprefetch: %llu\n",
4863
(u_longlong_t)L2BLK_GET_PREFETCH((&le[j])->le_prop));
4864
(void) printf("|\t\t\t\taddress: %llu\n",
4865
(u_longlong_t)le[j].le_daddr);
4866
(void) printf("|\t\t\t\tARC state: %llu\n",
4867
(u_longlong_t)L2BLK_GET_STATE((&le[j])->le_prop));
4868
(void) printf("|\n");
4869
}
4870
(void) printf("\n");
4871
}
4872
4873
static void
4874
dump_l2arc_log_blkptr(const l2arc_log_blkptr_t *lbps)
4875
{
4876
(void) printf("|\t\tdaddr: %llu\n", (u_longlong_t)lbps->lbp_daddr);
4877
(void) printf("|\t\tpayload_asize: %llu\n",
4878
(u_longlong_t)lbps->lbp_payload_asize);
4879
(void) printf("|\t\tpayload_start: %llu\n",
4880
(u_longlong_t)lbps->lbp_payload_start);
4881
(void) printf("|\t\tlsize: %llu\n",
4882
(u_longlong_t)L2BLK_GET_LSIZE(lbps->lbp_prop));
4883
(void) printf("|\t\tasize: %llu\n",
4884
(u_longlong_t)L2BLK_GET_PSIZE(lbps->lbp_prop));
4885
(void) printf("|\t\tcompralgo: %llu\n",
4886
(u_longlong_t)L2BLK_GET_COMPRESS(lbps->lbp_prop));
4887
(void) printf("|\t\tcksumalgo: %llu\n",
4888
(u_longlong_t)L2BLK_GET_CHECKSUM(lbps->lbp_prop));
4889
(void) printf("|\n\n");
4890
}
4891
4892
static void
4893
dump_l2arc_log_blocks(int fd, const l2arc_dev_hdr_phys_t *l2dhdr,
4894
l2arc_dev_hdr_phys_t *rebuild)
4895
{
4896
l2arc_log_blk_phys_t this_lb;
4897
uint64_t asize;
4898
l2arc_log_blkptr_t lbps[2];
4899
zio_cksum_t cksum;
4900
int failed = 0;
4901
l2arc_dev_t dev;
4902
4903
if (!dump_opt['q'])
4904
print_l2arc_log_blocks();
4905
memcpy(lbps, l2dhdr->dh_start_lbps, sizeof (lbps));
4906
4907
dev.l2ad_evict = l2dhdr->dh_evict;
4908
dev.l2ad_start = l2dhdr->dh_start;
4909
dev.l2ad_end = l2dhdr->dh_end;
4910
4911
if (l2dhdr->dh_start_lbps[0].lbp_daddr == 0) {
4912
/* no log blocks to read */
4913
if (!dump_opt['q']) {
4914
(void) printf("No log blocks to read\n");
4915
(void) printf("\n");
4916
}
4917
return;
4918
} else {
4919
dev.l2ad_hand = lbps[0].lbp_daddr +
4920
L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4921
}
4922
4923
dev.l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST);
4924
4925
for (;;) {
4926
if (!l2arc_log_blkptr_valid(&dev, &lbps[0]))
4927
break;
4928
4929
/* L2BLK_GET_PSIZE returns aligned size for log blocks */
4930
asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4931
if (pread64(fd, &this_lb, asize, lbps[0].lbp_daddr) != asize) {
4932
if (!dump_opt['q']) {
4933
(void) printf("Error while reading next log "
4934
"block\n\n");
4935
}
4936
break;
4937
}
4938
4939
fletcher_4_native_varsize(&this_lb, asize, &cksum);
4940
if (!ZIO_CHECKSUM_EQUAL(cksum, lbps[0].lbp_cksum)) {
4941
failed++;
4942
if (!dump_opt['q']) {
4943
(void) printf("Invalid cksum\n");
4944
dump_l2arc_log_blkptr(&lbps[0]);
4945
}
4946
break;
4947
}
4948
4949
switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) {
4950
case ZIO_COMPRESS_OFF:
4951
break;
4952
default: {
4953
abd_t *abd = abd_alloc_linear(asize, B_TRUE);
4954
abd_copy_from_buf_off(abd, &this_lb, 0, asize);
4955
abd_t dabd;
4956
abd_get_from_buf_struct(&dabd, &this_lb,
4957
sizeof (this_lb));
4958
int err = zio_decompress_data(L2BLK_GET_COMPRESS(
4959
(&lbps[0])->lbp_prop), abd, &dabd,
4960
asize, sizeof (this_lb), NULL);
4961
abd_free(&dabd);
4962
abd_free(abd);
4963
if (err != 0) {
4964
(void) printf("L2ARC block decompression "
4965
"failed\n");
4966
goto out;
4967
}
4968
break;
4969
}
4970
}
4971
4972
if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
4973
byteswap_uint64_array(&this_lb, sizeof (this_lb));
4974
if (this_lb.lb_magic != L2ARC_LOG_BLK_MAGIC) {
4975
if (!dump_opt['q'])
4976
(void) printf("Invalid log block magic\n\n");
4977
break;
4978
}
4979
4980
rebuild->dh_lb_count++;
4981
rebuild->dh_lb_asize += asize;
4982
if (dump_opt['l'] > 1 && !dump_opt['q']) {
4983
(void) printf("lb[%4llu]\tmagic: %llu\n",
4984
(u_longlong_t)rebuild->dh_lb_count,
4985
(u_longlong_t)this_lb.lb_magic);
4986
dump_l2arc_log_blkptr(&lbps[0]);
4987
}
4988
4989
if (dump_opt['l'] > 2 && !dump_opt['q'])
4990
dump_l2arc_log_entries(l2dhdr->dh_log_entries,
4991
this_lb.lb_entries,
4992
rebuild->dh_lb_count);
4993
4994
if (l2arc_range_check_overlap(lbps[1].lbp_payload_start,
4995
lbps[0].lbp_payload_start, dev.l2ad_evict) &&
4996
!dev.l2ad_first)
4997
break;
4998
4999
lbps[0] = lbps[1];
5000
lbps[1] = this_lb.lb_prev_lbp;
5001
}
5002
out:
5003
if (!dump_opt['q']) {
5004
(void) printf("log_blk_count:\t %llu with valid cksum\n",
5005
(u_longlong_t)rebuild->dh_lb_count);
5006
(void) printf("\t\t %d with invalid cksum\n", failed);
5007
(void) printf("log_blk_asize:\t %llu\n\n",
5008
(u_longlong_t)rebuild->dh_lb_asize);
5009
}
5010
}
5011
5012
static int
5013
dump_l2arc_header(int fd)
5014
{
5015
l2arc_dev_hdr_phys_t l2dhdr = {0}, rebuild = {0};
5016
int error = B_FALSE;
5017
5018
if (pread64(fd, &l2dhdr, sizeof (l2dhdr),
5019
VDEV_LABEL_START_SIZE) != sizeof (l2dhdr)) {
5020
error = B_TRUE;
5021
} else {
5022
if (l2dhdr.dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC))
5023
byteswap_uint64_array(&l2dhdr, sizeof (l2dhdr));
5024
5025
if (l2dhdr.dh_magic != L2ARC_DEV_HDR_MAGIC)
5026
error = B_TRUE;
5027
}
5028
5029
if (error) {
5030
(void) printf("L2ARC device header not found\n\n");
5031
/* Do not return an error here for backward compatibility */
5032
return (0);
5033
} else if (!dump_opt['q']) {
5034
print_l2arc_header();
5035
5036
(void) printf(" magic: %llu\n",
5037
(u_longlong_t)l2dhdr.dh_magic);
5038
(void) printf(" version: %llu\n",
5039
(u_longlong_t)l2dhdr.dh_version);
5040
(void) printf(" pool_guid: %llu\n",
5041
(u_longlong_t)l2dhdr.dh_spa_guid);
5042
(void) printf(" flags: %llu\n",
5043
(u_longlong_t)l2dhdr.dh_flags);
5044
(void) printf(" start_lbps[0]: %llu\n",
5045
(u_longlong_t)
5046
l2dhdr.dh_start_lbps[0].lbp_daddr);
5047
(void) printf(" start_lbps[1]: %llu\n",
5048
(u_longlong_t)
5049
l2dhdr.dh_start_lbps[1].lbp_daddr);
5050
(void) printf(" log_blk_ent: %llu\n",
5051
(u_longlong_t)l2dhdr.dh_log_entries);
5052
(void) printf(" start: %llu\n",
5053
(u_longlong_t)l2dhdr.dh_start);
5054
(void) printf(" end: %llu\n",
5055
(u_longlong_t)l2dhdr.dh_end);
5056
(void) printf(" evict: %llu\n",
5057
(u_longlong_t)l2dhdr.dh_evict);
5058
(void) printf(" lb_asize_refcount: %llu\n",
5059
(u_longlong_t)l2dhdr.dh_lb_asize);
5060
(void) printf(" lb_count_refcount: %llu\n",
5061
(u_longlong_t)l2dhdr.dh_lb_count);
5062
(void) printf(" trim_action_time: %llu\n",
5063
(u_longlong_t)l2dhdr.dh_trim_action_time);
5064
(void) printf(" trim_state: %llu\n\n",
5065
(u_longlong_t)l2dhdr.dh_trim_state);
5066
}
5067
5068
dump_l2arc_log_blocks(fd, &l2dhdr, &rebuild);
5069
/*
5070
* The total aligned size of log blocks and the number of log blocks
5071
* reported in the header of the device may be less than what zdb
5072
* reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
5073
* This happens because dump_l2arc_log_blocks() lacks the memory
5074
* pressure valve that l2arc_rebuild() has. Thus, if we are on a system
5075
* with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
5076
* and dh_lb_count will be lower to begin with than what exists on the
5077
* device. This is normal and zdb should not exit with an error. The
5078
* opposite case should never happen though, the values reported in the
5079
* header should never be higher than what dump_l2arc_log_blocks() and
5080
* l2arc_rebuild() report. If this happens there is a leak in the
5081
* accounting of log blocks.
5082
*/
5083
if (l2dhdr.dh_lb_asize > rebuild.dh_lb_asize ||
5084
l2dhdr.dh_lb_count > rebuild.dh_lb_count)
5085
return (1);
5086
5087
return (0);
5088
}
5089
5090
static void
5091
dump_config_from_label(zdb_label_t *label, size_t buflen, int l)
5092
{
5093
if (dump_opt['q'])
5094
return;
5095
5096
if ((dump_opt['l'] < 3) && (first_label(label->config) != l))
5097
return;
5098
5099
print_label_header(label, l);
5100
dump_nvlist(label->config_nv, 4);
5101
print_label_numbers(" labels = ", label->config);
5102
5103
if (dump_opt['l'] >= 2)
5104
dump_nvlist_stats(label->config_nv, buflen);
5105
}
5106
5107
#define ZDB_MAX_UB_HEADER_SIZE 32
5108
5109
static void
5110
dump_label_uberblocks(zdb_label_t *label, uint64_t ashift, int label_num)
5111
{
5112
5113
vdev_t vd;
5114
char header[ZDB_MAX_UB_HEADER_SIZE];
5115
5116
vd.vdev_ashift = ashift;
5117
vd.vdev_top = &vd;
5118
5119
for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5120
uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5121
uberblock_t *ub = (void *)((char *)&label->label + uoff);
5122
cksum_record_t *rec = label->uberblocks[i];
5123
5124
if (rec == NULL) {
5125
if (dump_opt['u'] >= 2) {
5126
print_label_header(label, label_num);
5127
(void) printf(" Uberblock[%d] invalid\n", i);
5128
}
5129
continue;
5130
}
5131
5132
if ((dump_opt['u'] < 3) && (first_label(rec) != label_num))
5133
continue;
5134
5135
if ((dump_opt['u'] < 4) &&
5136
(ub->ub_mmp_magic == MMP_MAGIC) && ub->ub_mmp_delay &&
5137
(i >= VDEV_UBERBLOCK_COUNT(&vd) - MMP_BLOCKS_PER_LABEL))
5138
continue;
5139
5140
print_label_header(label, label_num);
5141
(void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE,
5142
" Uberblock[%d]\n", i);
5143
dump_uberblock(ub, header, "");
5144
print_label_numbers(" labels = ", rec);
5145
}
5146
}
5147
5148
static char curpath[PATH_MAX];
5149
5150
/*
5151
* Iterate through the path components, recursively passing
5152
* current one's obj and remaining path until we find the obj
5153
* for the last one.
5154
*/
5155
static int
5156
dump_path_impl(objset_t *os, uint64_t obj, char *name, uint64_t *retobj)
5157
{
5158
int err;
5159
boolean_t header = B_TRUE;
5160
uint64_t child_obj;
5161
char *s;
5162
dmu_buf_t *db;
5163
dmu_object_info_t doi;
5164
5165
if ((s = strchr(name, '/')) != NULL)
5166
*s = '\0';
5167
err = zap_lookup(os, obj, name, 8, 1, &child_obj);
5168
5169
(void) strlcat(curpath, name, sizeof (curpath));
5170
5171
if (err != 0) {
5172
(void) fprintf(stderr, "failed to lookup %s: %s\n",
5173
curpath, strerror(err));
5174
return (err);
5175
}
5176
5177
child_obj = ZFS_DIRENT_OBJ(child_obj);
5178
err = sa_buf_hold(os, child_obj, FTAG, &db);
5179
if (err != 0) {
5180
(void) fprintf(stderr,
5181
"failed to get SA dbuf for obj %llu: %s\n",
5182
(u_longlong_t)child_obj, strerror(err));
5183
return (EINVAL);
5184
}
5185
dmu_object_info_from_db(db, &doi);
5186
sa_buf_rele(db, FTAG);
5187
5188
if (doi.doi_bonus_type != DMU_OT_SA &&
5189
doi.doi_bonus_type != DMU_OT_ZNODE) {
5190
(void) fprintf(stderr, "invalid bonus type %d for obj %llu\n",
5191
doi.doi_bonus_type, (u_longlong_t)child_obj);
5192
return (EINVAL);
5193
}
5194
5195
if (dump_opt['v'] > 6) {
5196
(void) printf("obj=%llu %s type=%d bonustype=%d\n",
5197
(u_longlong_t)child_obj, curpath, doi.doi_type,
5198
doi.doi_bonus_type);
5199
}
5200
5201
(void) strlcat(curpath, "/", sizeof (curpath));
5202
5203
switch (doi.doi_type) {
5204
case DMU_OT_DIRECTORY_CONTENTS:
5205
if (s != NULL && *(s + 1) != '\0')
5206
return (dump_path_impl(os, child_obj, s + 1, retobj));
5207
zfs_fallthrough;
5208
case DMU_OT_PLAIN_FILE_CONTENTS:
5209
if (retobj != NULL) {
5210
*retobj = child_obj;
5211
} else {
5212
dump_object(os, child_obj, dump_opt['v'], &header,
5213
NULL, 0);
5214
}
5215
return (0);
5216
default:
5217
(void) fprintf(stderr, "object %llu has non-file/directory "
5218
"type %d\n", (u_longlong_t)obj, doi.doi_type);
5219
break;
5220
}
5221
5222
return (EINVAL);
5223
}
5224
5225
/*
5226
* Dump the blocks for the object specified by path inside the dataset.
5227
*/
5228
static int
5229
dump_path(char *ds, char *path, uint64_t *retobj)
5230
{
5231
int err;
5232
objset_t *os;
5233
uint64_t root_obj;
5234
5235
err = open_objset(ds, FTAG, &os);
5236
if (err != 0)
5237
return (err);
5238
5239
err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj);
5240
if (err != 0) {
5241
(void) fprintf(stderr, "can't lookup root znode: %s\n",
5242
strerror(err));
5243
close_objset(os, FTAG);
5244
return (EINVAL);
5245
}
5246
5247
(void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds);
5248
5249
err = dump_path_impl(os, root_obj, path, retobj);
5250
5251
close_objset(os, FTAG);
5252
return (err);
5253
}
5254
5255
static int
5256
dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
5257
{
5258
const char *p = (const char *)buf;
5259
ssize_t nwritten;
5260
5261
(void) os;
5262
(void) arg;
5263
5264
/* Write the data out, handling short writes and signals. */
5265
while ((nwritten = write(STDOUT_FILENO, p, len)) < len) {
5266
if (nwritten < 0) {
5267
if (errno == EINTR)
5268
continue;
5269
return (errno);
5270
}
5271
p += nwritten;
5272
len -= nwritten;
5273
}
5274
5275
return (0);
5276
}
5277
5278
static void
5279
dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
5280
{
5281
boolean_t embed = B_FALSE;
5282
boolean_t large_block = B_FALSE;
5283
boolean_t compress = B_FALSE;
5284
boolean_t raw = B_FALSE;
5285
5286
const char *c;
5287
for (c = flagstr; c != NULL && *c != '\0'; c++) {
5288
switch (*c) {
5289
case 'e':
5290
embed = B_TRUE;
5291
break;
5292
case 'L':
5293
large_block = B_TRUE;
5294
break;
5295
case 'c':
5296
compress = B_TRUE;
5297
break;
5298
case 'w':
5299
raw = B_TRUE;
5300
break;
5301
default:
5302
fprintf(stderr, "dump_backup: invalid flag "
5303
"'%c'\n", *c);
5304
return;
5305
}
5306
}
5307
5308
if (isatty(STDOUT_FILENO)) {
5309
fprintf(stderr, "dump_backup: stream cannot be written "
5310
"to a terminal\n");
5311
return;
5312
}
5313
5314
offset_t off = 0;
5315
dmu_send_outparams_t out = {
5316
.dso_outfunc = dump_backup_bytes,
5317
.dso_dryrun = B_FALSE,
5318
};
5319
5320
int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed,
5321
large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO,
5322
&off, &out);
5323
if (err != 0) {
5324
fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",
5325
strerror(err));
5326
return;
5327
}
5328
}
5329
5330
static int
5331
zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile)
5332
{
5333
int err = 0;
5334
uint64_t size, readsize, oursize, offset;
5335
ssize_t writesize;
5336
sa_handle_t *hdl;
5337
5338
(void) printf("Copying object %" PRIu64 " to file %s\n", srcobj,
5339
destfile);
5340
5341
VERIFY3P(os, ==, sa_os);
5342
if ((err = sa_handle_get(os, srcobj, NULL, SA_HDL_PRIVATE, &hdl))) {
5343
(void) printf("Failed to get handle for SA znode\n");
5344
return (err);
5345
}
5346
if ((err = sa_lookup(hdl, sa_attr_table[ZPL_SIZE], &size, 8))) {
5347
(void) sa_handle_destroy(hdl);
5348
return (err);
5349
}
5350
(void) sa_handle_destroy(hdl);
5351
5352
(void) printf("Object %" PRIu64 " is %" PRIu64 " bytes\n", srcobj,
5353
size);
5354
if (size == 0) {
5355
return (EINVAL);
5356
}
5357
5358
int fd = open(destfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
5359
if (fd == -1)
5360
return (errno);
5361
/*
5362
* We cap the size at 1 mebibyte here to prevent
5363
* allocation failures and nigh-infinite printing if the
5364
* object is extremely large.
5365
*/
5366
oursize = MIN(size, 1 << 20);
5367
offset = 0;
5368
char *buf = kmem_alloc(oursize, KM_NOSLEEP);
5369
if (buf == NULL) {
5370
(void) close(fd);
5371
return (ENOMEM);
5372
}
5373
5374
while (offset < size) {
5375
readsize = MIN(size - offset, 1 << 20);
5376
err = dmu_read(os, srcobj, offset, readsize, buf, 0);
5377
if (err != 0) {
5378
(void) printf("got error %u from dmu_read\n", err);
5379
kmem_free(buf, oursize);
5380
(void) close(fd);
5381
return (err);
5382
}
5383
if (dump_opt['v'] > 3) {
5384
(void) printf("Read offset=%" PRIu64 " size=%" PRIu64
5385
" error=%d\n", offset, readsize, err);
5386
}
5387
5388
writesize = write(fd, buf, readsize);
5389
if (writesize < 0) {
5390
err = errno;
5391
break;
5392
} else if (writesize != readsize) {
5393
/* Incomplete write */
5394
(void) fprintf(stderr, "Short write, only wrote %llu of"
5395
" %" PRIu64 " bytes, exiting...\n",
5396
(u_longlong_t)writesize, readsize);
5397
break;
5398
}
5399
5400
offset += readsize;
5401
}
5402
5403
(void) close(fd);
5404
5405
if (buf != NULL)
5406
kmem_free(buf, oursize);
5407
5408
return (err);
5409
}
5410
5411
static boolean_t
5412
label_cksum_valid(vdev_label_t *label, uint64_t offset)
5413
{
5414
zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
5415
zio_cksum_t expected_cksum;
5416
zio_cksum_t actual_cksum;
5417
zio_cksum_t verifier;
5418
zio_eck_t *eck;
5419
int byteswap;
5420
5421
void *data = (char *)label + offsetof(vdev_label_t, vl_vdev_phys);
5422
eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1;
5423
5424
offset += offsetof(vdev_label_t, vl_vdev_phys);
5425
ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
5426
5427
byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
5428
if (byteswap)
5429
byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
5430
5431
expected_cksum = eck->zec_cksum;
5432
eck->zec_cksum = verifier;
5433
5434
abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE);
5435
ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum);
5436
abd_free(abd);
5437
5438
if (byteswap)
5439
byteswap_uint64_array(&expected_cksum, sizeof (zio_cksum_t));
5440
5441
if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
5442
return (B_TRUE);
5443
5444
return (B_FALSE);
5445
}
5446
5447
static int
5448
dump_label(const char *dev)
5449
{
5450
char path[MAXPATHLEN];
5451
zdb_label_t labels[VDEV_LABELS] = {{{{0}}}};
5452
uint64_t psize, ashift, l2cache;
5453
struct stat64 statbuf;
5454
boolean_t config_found = B_FALSE;
5455
boolean_t error = B_FALSE;
5456
boolean_t read_l2arc_header = B_FALSE;
5457
avl_tree_t config_tree;
5458
avl_tree_t uberblock_tree;
5459
void *node, *cookie;
5460
int fd;
5461
5462
/*
5463
* Check if we were given absolute path and use it as is.
5464
* Otherwise if the provided vdev name doesn't point to a file,
5465
* try prepending expected disk paths and partition numbers.
5466
*/
5467
(void) strlcpy(path, dev, sizeof (path));
5468
if (dev[0] != '/' && stat64(path, &statbuf) != 0) {
5469
int error;
5470
5471
error = zfs_resolve_shortname(dev, path, MAXPATHLEN);
5472
if (error == 0 && zfs_dev_is_whole_disk(path)) {
5473
if (zfs_append_partition(path, MAXPATHLEN) == -1)
5474
error = ENOENT;
5475
}
5476
5477
if (error || (stat64(path, &statbuf) != 0)) {
5478
(void) printf("failed to find device %s, try "
5479
"specifying absolute path instead\n", dev);
5480
return (1);
5481
}
5482
}
5483
5484
if ((fd = open64(path, O_RDONLY)) < 0) {
5485
(void) printf("cannot open '%s': %s\n", path, strerror(errno));
5486
zdb_exit(1);
5487
}
5488
5489
if (fstat64_blk(fd, &statbuf) != 0) {
5490
(void) printf("failed to stat '%s': %s\n", path,
5491
strerror(errno));
5492
(void) close(fd);
5493
zdb_exit(1);
5494
}
5495
5496
if (S_ISBLK(statbuf.st_mode) && zfs_dev_flush(fd) != 0)
5497
(void) printf("failed to invalidate cache '%s' : %s\n", path,
5498
strerror(errno));
5499
5500
avl_create(&config_tree, cksum_record_compare,
5501
sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5502
avl_create(&uberblock_tree, cksum_record_compare,
5503
sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5504
5505
psize = statbuf.st_size;
5506
psize = P2ALIGN_TYPED(psize, sizeof (vdev_label_t), uint64_t);
5507
ashift = SPA_MINBLOCKSHIFT;
5508
5509
/*
5510
* 1. Read the label from disk
5511
* 2. Verify label cksum
5512
* 3. Unpack the configuration and insert in config tree.
5513
* 4. Traverse all uberblocks and insert in uberblock tree.
5514
*/
5515
for (int l = 0; l < VDEV_LABELS; l++) {
5516
zdb_label_t *label = &labels[l];
5517
char *buf = label->label.vl_vdev_phys.vp_nvlist;
5518
size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5519
nvlist_t *config;
5520
cksum_record_t *rec;
5521
zio_cksum_t cksum;
5522
vdev_t vd;
5523
5524
label->label_offset = vdev_label_offset(psize, l, 0);
5525
5526
if (pread64(fd, &label->label, sizeof (label->label),
5527
label->label_offset) != sizeof (label->label)) {
5528
if (!dump_opt['q'])
5529
(void) printf("failed to read label %d\n", l);
5530
label->read_failed = B_TRUE;
5531
error = B_TRUE;
5532
continue;
5533
}
5534
5535
label->read_failed = B_FALSE;
5536
label->cksum_valid = label_cksum_valid(&label->label,
5537
label->label_offset);
5538
5539
if (nvlist_unpack(buf, buflen, &config, 0) == 0) {
5540
nvlist_t *vdev_tree = NULL;
5541
size_t size;
5542
5543
if ((nvlist_lookup_nvlist(config,
5544
ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) ||
5545
(nvlist_lookup_uint64(vdev_tree,
5546
ZPOOL_CONFIG_ASHIFT, &ashift) != 0))
5547
ashift = SPA_MINBLOCKSHIFT;
5548
5549
if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0)
5550
size = buflen;
5551
5552
/* If the device is a cache device read the header. */
5553
if (!read_l2arc_header) {
5554
if (nvlist_lookup_uint64(config,
5555
ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 &&
5556
l2cache == POOL_STATE_L2CACHE) {
5557
read_l2arc_header = B_TRUE;
5558
}
5559
}
5560
5561
fletcher_4_native_varsize(buf, size, &cksum);
5562
rec = cksum_record_insert(&config_tree, &cksum, l);
5563
5564
label->config = rec;
5565
label->config_nv = config;
5566
config_found = B_TRUE;
5567
} else {
5568
error = B_TRUE;
5569
}
5570
5571
vd.vdev_ashift = ashift;
5572
vd.vdev_top = &vd;
5573
5574
for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5575
uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5576
uberblock_t *ub = (void *)((char *)label + uoff);
5577
5578
if (uberblock_verify(ub))
5579
continue;
5580
5581
fletcher_4_native_varsize(ub, sizeof (*ub), &cksum);
5582
rec = cksum_record_insert(&uberblock_tree, &cksum, l);
5583
5584
label->uberblocks[i] = rec;
5585
}
5586
}
5587
5588
/*
5589
* Dump the label and uberblocks.
5590
*/
5591
for (int l = 0; l < VDEV_LABELS; l++) {
5592
zdb_label_t *label = &labels[l];
5593
size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5594
5595
if (label->read_failed == B_TRUE)
5596
continue;
5597
5598
if (label->config_nv) {
5599
dump_config_from_label(label, buflen, l);
5600
} else {
5601
if (!dump_opt['q'])
5602
(void) printf("failed to unpack label %d\n", l);
5603
}
5604
5605
if (dump_opt['u'])
5606
dump_label_uberblocks(label, ashift, l);
5607
5608
nvlist_free(label->config_nv);
5609
}
5610
5611
/*
5612
* Dump the L2ARC header, if existent.
5613
*/
5614
if (read_l2arc_header)
5615
error |= dump_l2arc_header(fd);
5616
5617
cookie = NULL;
5618
while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL)
5619
umem_free(node, sizeof (cksum_record_t));
5620
5621
cookie = NULL;
5622
while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL)
5623
umem_free(node, sizeof (cksum_record_t));
5624
5625
avl_destroy(&config_tree);
5626
avl_destroy(&uberblock_tree);
5627
5628
(void) close(fd);
5629
5630
return (config_found == B_FALSE ? 2 :
5631
(error == B_TRUE ? 1 : 0));
5632
}
5633
5634
static uint64_t dataset_feature_count[SPA_FEATURES];
5635
static uint64_t global_feature_count[SPA_FEATURES];
5636
static uint64_t remap_deadlist_count = 0;
5637
5638
static int
5639
dump_one_objset(const char *dsname, void *arg)
5640
{
5641
(void) arg;
5642
int error;
5643
objset_t *os;
5644
spa_feature_t f;
5645
5646
error = open_objset(dsname, FTAG, &os);
5647
if (error != 0)
5648
return (0);
5649
5650
for (f = 0; f < SPA_FEATURES; f++) {
5651
if (!dsl_dataset_feature_is_active(dmu_objset_ds(os), f))
5652
continue;
5653
ASSERT(spa_feature_table[f].fi_flags &
5654
ZFEATURE_FLAG_PER_DATASET);
5655
dataset_feature_count[f]++;
5656
}
5657
5658
if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os))) {
5659
remap_deadlist_count++;
5660
}
5661
5662
for (dsl_bookmark_node_t *dbn =
5663
avl_first(&dmu_objset_ds(os)->ds_bookmarks); dbn != NULL;
5664
dbn = AVL_NEXT(&dmu_objset_ds(os)->ds_bookmarks, dbn)) {
5665
mos_obj_refd(dbn->dbn_phys.zbm_redaction_obj);
5666
if (dbn->dbn_phys.zbm_redaction_obj != 0) {
5667
global_feature_count[
5668
SPA_FEATURE_REDACTION_BOOKMARKS]++;
5669
objset_t *mos = os->os_spa->spa_meta_objset;
5670
dnode_t *rl;
5671
VERIFY0(dnode_hold(mos,
5672
dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl));
5673
if (rl->dn_have_spill) {
5674
global_feature_count[
5675
SPA_FEATURE_REDACTION_LIST_SPILL]++;
5676
}
5677
}
5678
if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)
5679
global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN]++;
5680
}
5681
5682
if (dsl_deadlist_is_open(&dmu_objset_ds(os)->ds_dir->dd_livelist) &&
5683
!dmu_objset_is_snapshot(os)) {
5684
global_feature_count[SPA_FEATURE_LIVELIST]++;
5685
}
5686
5687
dump_objset(os);
5688
close_objset(os, FTAG);
5689
fuid_table_destroy();
5690
return (0);
5691
}
5692
5693
/*
5694
* Block statistics.
5695
*/
5696
#define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5697
typedef struct zdb_blkstats {
5698
uint64_t zb_asize;
5699
uint64_t zb_lsize;
5700
uint64_t zb_psize;
5701
uint64_t zb_count;
5702
uint64_t zb_gangs;
5703
uint64_t zb_ditto_samevdev;
5704
uint64_t zb_ditto_same_ms;
5705
uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
5706
} zdb_blkstats_t;
5707
5708
/*
5709
* Extended object types to report deferred frees and dedup auto-ditto blocks.
5710
*/
5711
#define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0)
5712
#define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1)
5713
#define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2)
5714
#define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3)
5715
5716
static const char *zdb_ot_extname[] = {
5717
"deferred free",
5718
"dedup ditto",
5719
"other",
5720
"Total",
5721
};
5722
5723
#define ZB_TOTAL DN_MAX_LEVELS
5724
#define SPA_MAX_FOR_16M (SPA_MAXBLOCKSHIFT+1)
5725
5726
typedef struct zdb_brt_entry {
5727
dva_t zbre_dva;
5728
uint64_t zbre_refcount;
5729
avl_node_t zbre_node;
5730
} zdb_brt_entry_t;
5731
5732
typedef struct zdb_cb {
5733
zdb_blkstats_t zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
5734
uint64_t zcb_removing_size;
5735
uint64_t zcb_checkpoint_size;
5736
uint64_t zcb_dedup_asize;
5737
uint64_t zcb_dedup_blocks;
5738
uint64_t zcb_clone_asize;
5739
uint64_t zcb_clone_blocks;
5740
uint64_t zcb_psize_count[SPA_MAX_FOR_16M];
5741
uint64_t zcb_lsize_count[SPA_MAX_FOR_16M];
5742
uint64_t zcb_asize_count[SPA_MAX_FOR_16M];
5743
uint64_t zcb_psize_len[SPA_MAX_FOR_16M];
5744
uint64_t zcb_lsize_len[SPA_MAX_FOR_16M];
5745
uint64_t zcb_asize_len[SPA_MAX_FOR_16M];
5746
uint64_t zcb_psize_total;
5747
uint64_t zcb_lsize_total;
5748
uint64_t zcb_asize_total;
5749
uint64_t zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
5750
uint64_t zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
5751
[BPE_PAYLOAD_SIZE + 1];
5752
uint64_t zcb_start;
5753
hrtime_t zcb_lastprint;
5754
uint64_t zcb_totalasize;
5755
uint64_t zcb_errors[256];
5756
int zcb_readfails;
5757
int zcb_haderrors;
5758
spa_t *zcb_spa;
5759
uint32_t **zcb_vd_obsolete_counts;
5760
avl_tree_t zcb_brt;
5761
boolean_t zcb_brt_is_active;
5762
} zdb_cb_t;
5763
5764
/* test if two DVA offsets from same vdev are within the same metaslab */
5765
static boolean_t
5766
same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2)
5767
{
5768
vdev_t *vd = vdev_lookup_top(spa, vdev);
5769
uint64_t ms_shift = vd->vdev_ms_shift;
5770
5771
return ((off1 >> ms_shift) == (off2 >> ms_shift));
5772
}
5773
5774
/*
5775
* Used to simplify reporting of the histogram data.
5776
*/
5777
typedef struct one_histo {
5778
const char *name;
5779
uint64_t *count;
5780
uint64_t *len;
5781
uint64_t cumulative;
5782
} one_histo_t;
5783
5784
/*
5785
* The number of separate histograms processed for psize, lsize and asize.
5786
*/
5787
#define NUM_HISTO 3
5788
5789
/*
5790
* This routine will create a fixed column size output of three different
5791
* histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5792
* the count, length and cumulative length of the psize, lsize and
5793
* asize blocks.
5794
*
5795
* All three types of blocks are listed on a single line
5796
*
5797
* By default the table is printed in nicenumber format (e.g. 123K) but
5798
* if the '-P' parameter is specified then the full raw number (parseable)
5799
* is printed out.
5800
*/
5801
static void
5802
dump_size_histograms(zdb_cb_t *zcb)
5803
{
5804
/*
5805
* A temporary buffer that allows us to convert a number into
5806
* a string using zdb_nicenumber to allow either raw or human
5807
* readable numbers to be output.
5808
*/
5809
char numbuf[32];
5810
5811
/*
5812
* Define titles which are used in the headers of the tables
5813
* printed by this routine.
5814
*/
5815
const char blocksize_title1[] = "block";
5816
const char blocksize_title2[] = "size";
5817
const char count_title[] = "Count";
5818
const char length_title[] = "Size";
5819
const char cumulative_title[] = "Cum.";
5820
5821
/*
5822
* Setup the histogram arrays (psize, lsize, and asize).
5823
*/
5824
one_histo_t parm_histo[NUM_HISTO];
5825
5826
parm_histo[0].name = "psize";
5827
parm_histo[0].count = zcb->zcb_psize_count;
5828
parm_histo[0].len = zcb->zcb_psize_len;
5829
parm_histo[0].cumulative = 0;
5830
5831
parm_histo[1].name = "lsize";
5832
parm_histo[1].count = zcb->zcb_lsize_count;
5833
parm_histo[1].len = zcb->zcb_lsize_len;
5834
parm_histo[1].cumulative = 0;
5835
5836
parm_histo[2].name = "asize";
5837
parm_histo[2].count = zcb->zcb_asize_count;
5838
parm_histo[2].len = zcb->zcb_asize_len;
5839
parm_histo[2].cumulative = 0;
5840
5841
5842
(void) printf("\nBlock Size Histogram\n");
5843
switch (block_bin_mode) {
5844
case BIN_PSIZE:
5845
printf("(note: all categories are binned by %s)\n", "psize");
5846
break;
5847
case BIN_LSIZE:
5848
printf("(note: all categories are binned by %s)\n", "lsize");
5849
break;
5850
case BIN_ASIZE:
5851
printf("(note: all categories are binned by %s)\n", "asize");
5852
break;
5853
default:
5854
printf("(note: all categories are binned separately)\n");
5855
break;
5856
}
5857
if (block_classes != 0) {
5858
char buf[256] = "";
5859
if (block_classes & CLASS_NORMAL)
5860
strlcat(buf, "\"normal\", ", sizeof (buf));
5861
if (block_classes & CLASS_SPECIAL)
5862
strlcat(buf, "\"special\", ", sizeof (buf));
5863
if (block_classes & CLASS_DEDUP)
5864
strlcat(buf, "\"dedup\", ", sizeof (buf));
5865
if (block_classes & CLASS_OTHER)
5866
strlcat(buf, "\"other\", ", sizeof (buf));
5867
buf[strlen(buf)-2] = '\0';
5868
printf("(note: only blocks in these classes are counted: %s)\n",
5869
buf);
5870
}
5871
/*
5872
* Print the first line titles
5873
*/
5874
if (dump_opt['P'])
5875
(void) printf("\n%s\t", blocksize_title1);
5876
else
5877
(void) printf("\n%7s ", blocksize_title1);
5878
5879
for (int j = 0; j < NUM_HISTO; j++) {
5880
if (dump_opt['P']) {
5881
if (j < NUM_HISTO - 1) {
5882
(void) printf("%s\t\t\t", parm_histo[j].name);
5883
} else {
5884
/* Don't print trailing spaces */
5885
(void) printf(" %s", parm_histo[j].name);
5886
}
5887
} else {
5888
if (j < NUM_HISTO - 1) {
5889
/* Left aligned strings in the output */
5890
(void) printf("%-7s ",
5891
parm_histo[j].name);
5892
} else {
5893
/* Don't print trailing spaces */
5894
(void) printf("%s", parm_histo[j].name);
5895
}
5896
}
5897
}
5898
(void) printf("\n");
5899
5900
/*
5901
* Print the second line titles
5902
*/
5903
if (dump_opt['P']) {
5904
(void) printf("%s\t", blocksize_title2);
5905
} else {
5906
(void) printf("%7s ", blocksize_title2);
5907
}
5908
5909
for (int i = 0; i < NUM_HISTO; i++) {
5910
if (dump_opt['P']) {
5911
(void) printf("%s\t%s\t%s\t",
5912
count_title, length_title, cumulative_title);
5913
} else {
5914
(void) printf("%7s%7s%7s",
5915
count_title, length_title, cumulative_title);
5916
}
5917
}
5918
(void) printf("\n");
5919
5920
/*
5921
* Print the rows
5922
*/
5923
for (int i = SPA_MINBLOCKSHIFT; i < SPA_MAX_FOR_16M; i++) {
5924
5925
/*
5926
* Print the first column showing the blocksize
5927
*/
5928
zdb_nicenum((1ULL << i), numbuf, sizeof (numbuf));
5929
5930
if (dump_opt['P']) {
5931
printf("%s", numbuf);
5932
} else {
5933
printf("%7s:", numbuf);
5934
}
5935
5936
/*
5937
* Print the remaining set of 3 columns per size:
5938
* for psize, lsize and asize
5939
*/
5940
for (int j = 0; j < NUM_HISTO; j++) {
5941
parm_histo[j].cumulative += parm_histo[j].len[i];
5942
5943
zdb_nicenum(parm_histo[j].count[i],
5944
numbuf, sizeof (numbuf));
5945
if (dump_opt['P'])
5946
(void) printf("\t%s", numbuf);
5947
else
5948
(void) printf("%7s", numbuf);
5949
5950
zdb_nicenum(parm_histo[j].len[i],
5951
numbuf, sizeof (numbuf));
5952
if (dump_opt['P'])
5953
(void) printf("\t%s", numbuf);
5954
else
5955
(void) printf("%7s", numbuf);
5956
5957
zdb_nicenum(parm_histo[j].cumulative,
5958
numbuf, sizeof (numbuf));
5959
if (dump_opt['P'])
5960
(void) printf("\t%s", numbuf);
5961
else
5962
(void) printf("%7s", numbuf);
5963
}
5964
(void) printf("\n");
5965
}
5966
}
5967
5968
static void
5969
zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
5970
dmu_object_type_t type)
5971
{
5972
int i;
5973
5974
ASSERT(type < ZDB_OT_TOTAL);
5975
5976
if (zilog && zil_bp_tree_add(zilog, bp) != 0)
5977
return;
5978
5979
/*
5980
* This flag controls if we will issue a claim for the block while
5981
* counting it, to ensure that all blocks are referenced in space maps.
5982
* We don't issue claims if we're not doing leak tracking, because it's
5983
* expensive if the user isn't interested. We also don't claim the
5984
* second or later occurences of cloned or dedup'd blocks, because we
5985
* already claimed them the first time.
5986
*/
5987
boolean_t do_claim = !dump_opt['L'];
5988
5989
spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
5990
5991
blkptr_t tempbp;
5992
if (BP_GET_DEDUP(bp)) {
5993
/*
5994
* Dedup'd blocks are special. We need to count them, so we can
5995
* later uncount them when reporting leaked space, and we must
5996
* only claim them once.
5997
*
5998
* We use the existing dedup system to track what we've seen.
5999
* The first time we see a block, we do a ddt_lookup() to see
6000
* if it exists in the DDT. If we're doing leak tracking, we
6001
* claim the block at this time.
6002
*
6003
* Each time we see a block, we reduce the refcount in the
6004
* entry by one, and add to the size and count of dedup'd
6005
* blocks to report at the end.
6006
*/
6007
6008
ddt_t *ddt = ddt_select(zcb->zcb_spa, bp);
6009
6010
ddt_enter(ddt);
6011
6012
/*
6013
* Find the block. This will create the entry in memory, but
6014
* we'll know if that happened by its refcount.
6015
*/
6016
ddt_entry_t *dde = ddt_lookup(ddt, bp, B_TRUE);
6017
6018
/*
6019
* ddt_lookup() can return NULL if this block didn't exist
6020
* in the DDT and creating it would take the DDT over its
6021
* quota. Since we got the block from disk, it must exist in
6022
* the DDT, so this can't happen. However, when unique entries
6023
* are pruned, the dedup bit can be set with no corresponding
6024
* entry in the DDT.
6025
*/
6026
if (dde == NULL) {
6027
ddt_exit(ddt);
6028
goto skipped;
6029
}
6030
6031
/* Get the phys for this variant */
6032
ddt_phys_variant_t v = ddt_phys_select(ddt, dde, bp);
6033
6034
/*
6035
* This entry may have multiple sets of DVAs. We must claim
6036
* each set the first time we see them in a real block on disk,
6037
* or count them on subsequent occurences. We don't have a
6038
* convenient way to track the first time we see each variant,
6039
* so we repurpose dde_io as a set of "seen" flag bits. We can
6040
* do this safely in zdb because it never writes, so it will
6041
* never have a writing zio for this block in that pointer.
6042
*/
6043
boolean_t seen = !!(((uintptr_t)dde->dde_io) & (1 << v));
6044
if (!seen)
6045
dde->dde_io =
6046
(void *)(((uintptr_t)dde->dde_io) | (1 << v));
6047
6048
/* Consume a reference for this block. */
6049
if (ddt_phys_total_refcnt(ddt, dde->dde_phys) > 0)
6050
ddt_phys_decref(dde->dde_phys, v);
6051
6052
/*
6053
* If this entry has a single flat phys, it may have been
6054
* extended with additional DVAs at some time in its life.
6055
* This block might be from before it was fully extended, and
6056
* so have fewer DVAs.
6057
*
6058
* If this is the first time we've seen this block, and we
6059
* claimed it as-is, then we would miss the claim on some
6060
* number of DVAs, which would then be seen as leaked.
6061
*
6062
* In all cases, if we've had fewer DVAs, then the asize would
6063
* be too small, and would lead to the pool apparently using
6064
* more space than allocated.
6065
*
6066
* To handle this, we copy the canonical set of DVAs from the
6067
* entry back to the block pointer before we claim it.
6068
*/
6069
if (v == DDT_PHYS_FLAT) {
6070
ASSERT3U(BP_GET_PHYSICAL_BIRTH(bp), ==,
6071
ddt_phys_birth(dde->dde_phys, v));
6072
tempbp = *bp;
6073
ddt_bp_fill(dde->dde_phys, v, &tempbp,
6074
BP_GET_PHYSICAL_BIRTH(bp));
6075
bp = &tempbp;
6076
}
6077
6078
if (seen) {
6079
/*
6080
* The second or later time we see this block,
6081
* it's a duplicate and we count it.
6082
*/
6083
zcb->zcb_dedup_asize += BP_GET_ASIZE(bp);
6084
zcb->zcb_dedup_blocks++;
6085
6086
/* Already claimed, don't do it again. */
6087
do_claim = B_FALSE;
6088
}
6089
6090
ddt_exit(ddt);
6091
} else if (zcb->zcb_brt_is_active &&
6092
brt_maybe_exists(zcb->zcb_spa, bp)) {
6093
/*
6094
* Cloned blocks are special. We need to count them, so we can
6095
* later uncount them when reporting leaked space, and we must
6096
* only claim them once.
6097
*
6098
* To do this, we keep our own in-memory BRT. For each block
6099
* we haven't seen before, we look it up in the real BRT and
6100
* if its there, we note it and its refcount then proceed as
6101
* normal. If we see the block again, we count it as a clone
6102
* and then give it no further consideration.
6103
*/
6104
zdb_brt_entry_t zbre_search, *zbre;
6105
avl_index_t where;
6106
6107
zbre_search.zbre_dva = bp->blk_dva[0];
6108
zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where);
6109
if (zbre == NULL) {
6110
/* Not seen before; track it */
6111
uint64_t refcnt =
6112
brt_entry_get_refcount(zcb->zcb_spa, bp);
6113
if (refcnt > 0) {
6114
zbre = umem_zalloc(sizeof (zdb_brt_entry_t),
6115
UMEM_NOFAIL);
6116
zbre->zbre_dva = bp->blk_dva[0];
6117
zbre->zbre_refcount = refcnt;
6118
avl_insert(&zcb->zcb_brt, zbre, where);
6119
}
6120
} else {
6121
/*
6122
* Second or later occurrence, count it and take a
6123
* refcount.
6124
*/
6125
zcb->zcb_clone_asize += BP_GET_ASIZE(bp);
6126
zcb->zcb_clone_blocks++;
6127
6128
zbre->zbre_refcount--;
6129
if (zbre->zbre_refcount == 0) {
6130
avl_remove(&zcb->zcb_brt, zbre);
6131
umem_free(zbre, sizeof (zdb_brt_entry_t));
6132
}
6133
6134
/* Already claimed, don't do it again. */
6135
do_claim = B_FALSE;
6136
}
6137
}
6138
6139
skipped:
6140
for (i = 0; i < 4; i++) {
6141
int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
6142
int t = (i & 1) ? type : ZDB_OT_TOTAL;
6143
int equal;
6144
zdb_blkstats_t *zb = &zcb->zcb_type[l][t];
6145
6146
zb->zb_asize += BP_GET_ASIZE(bp);
6147
zb->zb_lsize += BP_GET_LSIZE(bp);
6148
zb->zb_psize += BP_GET_PSIZE(bp);
6149
zb->zb_count++;
6150
6151
/*
6152
* The histogram is only big enough to record blocks up to
6153
* SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
6154
* "other", bucket.
6155
*/
6156
unsigned idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT;
6157
idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1);
6158
zb->zb_psize_histogram[idx]++;
6159
6160
zb->zb_gangs += BP_COUNT_GANG(bp);
6161
6162
switch (BP_GET_NDVAS(bp)) {
6163
case 2:
6164
if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6165
DVA_GET_VDEV(&bp->blk_dva[1])) {
6166
zb->zb_ditto_samevdev++;
6167
6168
if (same_metaslab(zcb->zcb_spa,
6169
DVA_GET_VDEV(&bp->blk_dva[0]),
6170
DVA_GET_OFFSET(&bp->blk_dva[0]),
6171
DVA_GET_OFFSET(&bp->blk_dva[1])))
6172
zb->zb_ditto_same_ms++;
6173
}
6174
break;
6175
case 3:
6176
equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6177
DVA_GET_VDEV(&bp->blk_dva[1])) +
6178
(DVA_GET_VDEV(&bp->blk_dva[0]) ==
6179
DVA_GET_VDEV(&bp->blk_dva[2])) +
6180
(DVA_GET_VDEV(&bp->blk_dva[1]) ==
6181
DVA_GET_VDEV(&bp->blk_dva[2]));
6182
if (equal != 0) {
6183
zb->zb_ditto_samevdev++;
6184
6185
if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6186
DVA_GET_VDEV(&bp->blk_dva[1]) &&
6187
same_metaslab(zcb->zcb_spa,
6188
DVA_GET_VDEV(&bp->blk_dva[0]),
6189
DVA_GET_OFFSET(&bp->blk_dva[0]),
6190
DVA_GET_OFFSET(&bp->blk_dva[1])))
6191
zb->zb_ditto_same_ms++;
6192
else if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6193
DVA_GET_VDEV(&bp->blk_dva[2]) &&
6194
same_metaslab(zcb->zcb_spa,
6195
DVA_GET_VDEV(&bp->blk_dva[0]),
6196
DVA_GET_OFFSET(&bp->blk_dva[0]),
6197
DVA_GET_OFFSET(&bp->blk_dva[2])))
6198
zb->zb_ditto_same_ms++;
6199
else if (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6200
DVA_GET_VDEV(&bp->blk_dva[2]) &&
6201
same_metaslab(zcb->zcb_spa,
6202
DVA_GET_VDEV(&bp->blk_dva[1]),
6203
DVA_GET_OFFSET(&bp->blk_dva[1]),
6204
DVA_GET_OFFSET(&bp->blk_dva[2])))
6205
zb->zb_ditto_same_ms++;
6206
}
6207
break;
6208
}
6209
}
6210
6211
spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
6212
6213
if (BP_IS_EMBEDDED(bp)) {
6214
zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
6215
zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
6216
[BPE_GET_PSIZE(bp)]++;
6217
return;
6218
}
6219
6220
if (block_classes != 0) {
6221
spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
6222
6223
uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[0]);
6224
uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[0]);
6225
vdev_t *vd = vdev_lookup_top(zcb->zcb_spa, vdev);
6226
ASSERT(vd != NULL);
6227
metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6228
ASSERT(ms != NULL);
6229
metaslab_group_t *mg = ms->ms_group;
6230
ASSERT(mg != NULL);
6231
metaslab_class_t *mc = mg->mg_class;
6232
ASSERT(mc != NULL);
6233
6234
spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
6235
6236
int class;
6237
if (mc == spa_normal_class(zcb->zcb_spa)) {
6238
class = CLASS_NORMAL;
6239
} else if (mc == spa_special_class(zcb->zcb_spa)) {
6240
class = CLASS_SPECIAL;
6241
} else if (mc == spa_dedup_class(zcb->zcb_spa)) {
6242
class = CLASS_DEDUP;
6243
} else {
6244
class = CLASS_OTHER;
6245
}
6246
6247
if (!(block_classes & class)) {
6248
goto hist_skipped;
6249
}
6250
}
6251
6252
/*
6253
* The binning histogram bins by powers of two up to
6254
* SPA_MAXBLOCKSIZE rather than creating bins for
6255
* every possible blocksize found in the pool.
6256
*/
6257
int bin;
6258
6259
/*
6260
* Binning strategy: each bin includes blocks up to and including
6261
* the given size (excluding blocks that fit into the previous bin).
6262
* This way, the "4K" bin includes blocks within the (2K; 4K] range.
6263
*/
6264
#define BIN(size) (highbit64((size) - 1))
6265
6266
switch (block_bin_mode) {
6267
case BIN_PSIZE: bin = BIN(BP_GET_PSIZE(bp)); break;
6268
case BIN_LSIZE: bin = BIN(BP_GET_LSIZE(bp)); break;
6269
case BIN_ASIZE: bin = BIN(BP_GET_ASIZE(bp)); break;
6270
case BIN_AUTO: break;
6271
default: PANIC("bad block_bin_mode"); abort();
6272
}
6273
6274
if (block_bin_mode == BIN_AUTO)
6275
bin = BIN(BP_GET_PSIZE(bp));
6276
6277
zcb->zcb_psize_count[bin]++;
6278
zcb->zcb_psize_len[bin] += BP_GET_PSIZE(bp);
6279
zcb->zcb_psize_total += BP_GET_PSIZE(bp);
6280
6281
if (block_bin_mode == BIN_AUTO)
6282
bin = BIN(BP_GET_LSIZE(bp));
6283
6284
zcb->zcb_lsize_count[bin]++;
6285
zcb->zcb_lsize_len[bin] += BP_GET_LSIZE(bp);
6286
zcb->zcb_lsize_total += BP_GET_LSIZE(bp);
6287
6288
if (block_bin_mode == BIN_AUTO)
6289
bin = BIN(BP_GET_ASIZE(bp));
6290
6291
zcb->zcb_asize_count[bin]++;
6292
zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp);
6293
zcb->zcb_asize_total += BP_GET_ASIZE(bp);
6294
6295
#undef BIN
6296
6297
hist_skipped:
6298
if (!do_claim)
6299
return;
6300
6301
VERIFY0(zio_wait(zio_claim(NULL, zcb->zcb_spa,
6302
spa_min_claim_txg(zcb->zcb_spa), bp, NULL, NULL,
6303
ZIO_FLAG_CANFAIL)));
6304
}
6305
6306
static void
6307
zdb_blkptr_done(zio_t *zio)
6308
{
6309
spa_t *spa = zio->io_spa;
6310
blkptr_t *bp = zio->io_bp;
6311
int ioerr = zio->io_error;
6312
zdb_cb_t *zcb = zio->io_private;
6313
zbookmark_phys_t *zb = &zio->io_bookmark;
6314
6315
mutex_enter(&spa->spa_scrub_lock);
6316
spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
6317
cv_broadcast(&spa->spa_scrub_io_cv);
6318
6319
if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
6320
char blkbuf[BP_SPRINTF_LEN];
6321
6322
zcb->zcb_haderrors = 1;
6323
zcb->zcb_errors[ioerr]++;
6324
6325
if (dump_opt['b'] >= 2)
6326
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6327
else
6328
blkbuf[0] = '\0';
6329
6330
(void) printf("zdb_blkptr_cb: "
6331
"Got error %d reading "
6332
"<%llu, %llu, %lld, %llx> %s -- skipping\n",
6333
ioerr,
6334
(u_longlong_t)zb->zb_objset,
6335
(u_longlong_t)zb->zb_object,
6336
(u_longlong_t)zb->zb_level,
6337
(u_longlong_t)zb->zb_blkid,
6338
blkbuf);
6339
}
6340
mutex_exit(&spa->spa_scrub_lock);
6341
6342
abd_free(zio->io_abd);
6343
}
6344
6345
static int
6346
zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
6347
const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
6348
{
6349
zdb_cb_t *zcb = arg;
6350
dmu_object_type_t type;
6351
boolean_t is_metadata;
6352
6353
if (zb->zb_level == ZB_DNODE_LEVEL)
6354
return (0);
6355
6356
if (dump_opt['b'] >= 5 && BP_GET_BIRTH(bp) > 0) {
6357
char blkbuf[BP_SPRINTF_LEN];
6358
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6359
(void) printf("objset %llu object %llu "
6360
"level %lld offset 0x%llx %s\n",
6361
(u_longlong_t)zb->zb_objset,
6362
(u_longlong_t)zb->zb_object,
6363
(longlong_t)zb->zb_level,
6364
(u_longlong_t)blkid2offset(dnp, bp, zb),
6365
blkbuf);
6366
}
6367
6368
if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
6369
return (0);
6370
6371
type = BP_GET_TYPE(bp);
6372
6373
zdb_count_block(zcb, zilog, bp,
6374
(type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type);
6375
6376
is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
6377
6378
if (!BP_IS_EMBEDDED(bp) &&
6379
(dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
6380
size_t size = BP_GET_PSIZE(bp);
6381
abd_t *abd = abd_alloc(size, B_FALSE);
6382
int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
6383
6384
/* If it's an intent log block, failure is expected. */
6385
if (zb->zb_level == ZB_ZIL_LEVEL)
6386
flags |= ZIO_FLAG_SPECULATIVE;
6387
6388
mutex_enter(&spa->spa_scrub_lock);
6389
while (spa->spa_load_verify_bytes > max_inflight_bytes)
6390
cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
6391
spa->spa_load_verify_bytes += size;
6392
mutex_exit(&spa->spa_scrub_lock);
6393
6394
zio_nowait(zio_read(NULL, spa, bp, abd, size,
6395
zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
6396
}
6397
6398
zcb->zcb_readfails = 0;
6399
6400
/* only call gethrtime() every 100 blocks */
6401
static int iters;
6402
if (++iters > 100)
6403
iters = 0;
6404
else
6405
return (0);
6406
6407
if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
6408
uint64_t now = gethrtime();
6409
char buf[10];
6410
uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
6411
uint64_t kb_per_sec =
6412
1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000));
6413
uint64_t sec_remaining =
6414
(zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec;
6415
6416
/* make sure nicenum has enough space */
6417
_Static_assert(sizeof (buf) >= NN_NUMBUF_SZ, "buf truncated");
6418
6419
zfs_nicebytes(bytes, buf, sizeof (buf));
6420
(void) fprintf(stderr,
6421
"\r%5s completed (%4"PRIu64"MB/s) "
6422
"estimated time remaining: "
6423
"%"PRIu64"hr %02"PRIu64"min %02"PRIu64"sec ",
6424
buf, kb_per_sec / 1024,
6425
sec_remaining / 60 / 60,
6426
sec_remaining / 60 % 60,
6427
sec_remaining % 60);
6428
6429
zcb->zcb_lastprint = now;
6430
}
6431
6432
return (0);
6433
}
6434
6435
static void
6436
zdb_leak(void *arg, uint64_t start, uint64_t size)
6437
{
6438
vdev_t *vd = arg;
6439
6440
(void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
6441
(u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
6442
}
6443
6444
static metaslab_ops_t zdb_metaslab_ops = {
6445
NULL /* alloc */
6446
};
6447
6448
static int
6449
load_unflushed_svr_segs_cb(spa_t *spa, space_map_entry_t *sme,
6450
uint64_t txg, void *arg)
6451
{
6452
spa_vdev_removal_t *svr = arg;
6453
6454
uint64_t offset = sme->sme_offset;
6455
uint64_t size = sme->sme_run;
6456
6457
/* skip vdevs we don't care about */
6458
if (sme->sme_vdev != svr->svr_vdev_id)
6459
return (0);
6460
6461
vdev_t *vd = vdev_lookup_top(spa, sme->sme_vdev);
6462
metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6463
ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6464
6465
if (txg < metaslab_unflushed_txg(ms))
6466
return (0);
6467
6468
if (sme->sme_type == SM_ALLOC)
6469
zfs_range_tree_add(svr->svr_allocd_segs, offset, size);
6470
else
6471
zfs_range_tree_remove(svr->svr_allocd_segs, offset, size);
6472
6473
return (0);
6474
}
6475
6476
static void
6477
claim_segment_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
6478
uint64_t size, void *arg)
6479
{
6480
(void) inner_offset, (void) arg;
6481
6482
/*
6483
* This callback was called through a remap from
6484
* a device being removed. Therefore, the vdev that
6485
* this callback is applied to is a concrete
6486
* vdev.
6487
*/
6488
ASSERT(vdev_is_concrete(vd));
6489
6490
VERIFY0(metaslab_claim_impl(vd, offset, size,
6491
spa_min_claim_txg(vd->vdev_spa)));
6492
}
6493
6494
static void
6495
claim_segment_cb(void *arg, uint64_t offset, uint64_t size)
6496
{
6497
vdev_t *vd = arg;
6498
6499
vdev_indirect_ops.vdev_op_remap(vd, offset, size,
6500
claim_segment_impl_cb, NULL);
6501
}
6502
6503
/*
6504
* After accounting for all allocated blocks that are directly referenced,
6505
* we might have missed a reference to a block from a partially complete
6506
* (and thus unused) indirect mapping object. We perform a secondary pass
6507
* through the metaslabs we have already mapped and claim the destination
6508
* blocks.
6509
*/
6510
static void
6511
zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb)
6512
{
6513
if (dump_opt['L'])
6514
return;
6515
6516
if (spa->spa_vdev_removal == NULL)
6517
return;
6518
6519
spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6520
6521
spa_vdev_removal_t *svr = spa->spa_vdev_removal;
6522
vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id);
6523
vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6524
6525
ASSERT0(zfs_range_tree_space(svr->svr_allocd_segs));
6526
6527
zfs_range_tree_t *allocs = zfs_range_tree_create_flags(
6528
NULL, ZFS_RANGE_SEG64, NULL, 0, 0,
6529
0, "zdb_claim_removing:allocs");
6530
for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) {
6531
metaslab_t *msp = vd->vdev_ms[msi];
6532
6533
ASSERT0(zfs_range_tree_space(allocs));
6534
if (msp->ms_sm != NULL)
6535
VERIFY0(space_map_load(msp->ms_sm, allocs, SM_ALLOC));
6536
zfs_range_tree_vacate(allocs, zfs_range_tree_add,
6537
svr->svr_allocd_segs);
6538
}
6539
zfs_range_tree_destroy(allocs);
6540
6541
iterate_through_spacemap_logs(spa, load_unflushed_svr_segs_cb, svr);
6542
6543
/*
6544
* Clear everything past what has been synced,
6545
* because we have not allocated mappings for
6546
* it yet.
6547
*/
6548
zfs_range_tree_clear(svr->svr_allocd_segs,
6549
vdev_indirect_mapping_max_offset(vim),
6550
vd->vdev_asize - vdev_indirect_mapping_max_offset(vim));
6551
6552
zcb->zcb_removing_size += zfs_range_tree_space(svr->svr_allocd_segs);
6553
zfs_range_tree_vacate(svr->svr_allocd_segs, claim_segment_cb, vd);
6554
6555
spa_config_exit(spa, SCL_CONFIG, FTAG);
6556
}
6557
6558
static int
6559
increment_indirect_mapping_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6560
dmu_tx_t *tx)
6561
{
6562
(void) tx;
6563
zdb_cb_t *zcb = arg;
6564
spa_t *spa = zcb->zcb_spa;
6565
vdev_t *vd;
6566
const dva_t *dva = &bp->blk_dva[0];
6567
6568
ASSERT(!bp_freed);
6569
ASSERT(!dump_opt['L']);
6570
ASSERT3U(BP_GET_NDVAS(bp), ==, 1);
6571
6572
spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6573
vd = vdev_lookup_top(zcb->zcb_spa, DVA_GET_VDEV(dva));
6574
ASSERT3P(vd, !=, NULL);
6575
spa_config_exit(spa, SCL_VDEV, FTAG);
6576
6577
ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0);
6578
ASSERT3P(zcb->zcb_vd_obsolete_counts[vd->vdev_id], !=, NULL);
6579
6580
vdev_indirect_mapping_increment_obsolete_count(
6581
vd->vdev_indirect_mapping,
6582
DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva),
6583
zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6584
6585
return (0);
6586
}
6587
6588
static uint32_t *
6589
zdb_load_obsolete_counts(vdev_t *vd)
6590
{
6591
vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6592
spa_t *spa = vd->vdev_spa;
6593
spa_condensing_indirect_phys_t *scip =
6594
&spa->spa_condensing_indirect_phys;
6595
uint64_t obsolete_sm_object;
6596
uint32_t *counts;
6597
6598
VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
6599
EQUIV(obsolete_sm_object != 0, vd->vdev_obsolete_sm != NULL);
6600
counts = vdev_indirect_mapping_load_obsolete_counts(vim);
6601
if (vd->vdev_obsolete_sm != NULL) {
6602
vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6603
vd->vdev_obsolete_sm);
6604
}
6605
if (scip->scip_vdev == vd->vdev_id &&
6606
scip->scip_prev_obsolete_sm_object != 0) {
6607
space_map_t *prev_obsolete_sm = NULL;
6608
VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset,
6609
scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0));
6610
vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6611
prev_obsolete_sm);
6612
space_map_close(prev_obsolete_sm);
6613
}
6614
return (counts);
6615
}
6616
6617
typedef struct checkpoint_sm_exclude_entry_arg {
6618
vdev_t *cseea_vd;
6619
uint64_t cseea_checkpoint_size;
6620
} checkpoint_sm_exclude_entry_arg_t;
6621
6622
static int
6623
checkpoint_sm_exclude_entry_cb(space_map_entry_t *sme, void *arg)
6624
{
6625
checkpoint_sm_exclude_entry_arg_t *cseea = arg;
6626
vdev_t *vd = cseea->cseea_vd;
6627
metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
6628
uint64_t end = sme->sme_offset + sme->sme_run;
6629
6630
ASSERT(sme->sme_type == SM_FREE);
6631
6632
/*
6633
* Since the vdev_checkpoint_sm exists in the vdev level
6634
* and the ms_sm space maps exist in the metaslab level,
6635
* an entry in the checkpoint space map could theoretically
6636
* cross the boundaries of the metaslab that it belongs.
6637
*
6638
* In reality, because of the way that we populate and
6639
* manipulate the checkpoint's space maps currently,
6640
* there shouldn't be any entries that cross metaslabs.
6641
* Hence the assertion below.
6642
*
6643
* That said, there is no fundamental requirement that
6644
* the checkpoint's space map entries should not cross
6645
* metaslab boundaries. So if needed we could add code
6646
* that handles metaslab-crossing segments in the future.
6647
*/
6648
VERIFY3U(sme->sme_offset, >=, ms->ms_start);
6649
VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
6650
6651
/*
6652
* By removing the entry from the allocated segments we
6653
* also verify that the entry is there to begin with.
6654
*/
6655
mutex_enter(&ms->ms_lock);
6656
zfs_range_tree_remove(ms->ms_allocatable, sme->sme_offset,
6657
sme->sme_run);
6658
mutex_exit(&ms->ms_lock);
6659
6660
cseea->cseea_checkpoint_size += sme->sme_run;
6661
return (0);
6662
}
6663
6664
static void
6665
zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_cb_t *zcb)
6666
{
6667
spa_t *spa = vd->vdev_spa;
6668
space_map_t *checkpoint_sm = NULL;
6669
uint64_t checkpoint_sm_obj;
6670
6671
/*
6672
* If there is no vdev_top_zap, we are in a pool whose
6673
* version predates the pool checkpoint feature.
6674
*/
6675
if (vd->vdev_top_zap == 0)
6676
return;
6677
6678
/*
6679
* If there is no reference of the vdev_checkpoint_sm in
6680
* the vdev_top_zap, then one of the following scenarios
6681
* is true:
6682
*
6683
* 1] There is no checkpoint
6684
* 2] There is a checkpoint, but no checkpointed blocks
6685
* have been freed yet
6686
* 3] The current vdev is indirect
6687
*
6688
* In these cases we return immediately.
6689
*/
6690
if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
6691
VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
6692
return;
6693
6694
VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
6695
VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1,
6696
&checkpoint_sm_obj));
6697
6698
checkpoint_sm_exclude_entry_arg_t cseea;
6699
cseea.cseea_vd = vd;
6700
cseea.cseea_checkpoint_size = 0;
6701
6702
VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
6703
checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
6704
6705
VERIFY0(space_map_iterate(checkpoint_sm,
6706
space_map_length(checkpoint_sm),
6707
checkpoint_sm_exclude_entry_cb, &cseea));
6708
space_map_close(checkpoint_sm);
6709
6710
zcb->zcb_checkpoint_size += cseea.cseea_checkpoint_size;
6711
}
6712
6713
static void
6714
zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb)
6715
{
6716
ASSERT(!dump_opt['L']);
6717
6718
vdev_t *rvd = spa->spa_root_vdev;
6719
for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6720
ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id);
6721
zdb_leak_init_vdev_exclude_checkpoint(rvd->vdev_child[c], zcb);
6722
}
6723
}
6724
6725
static int
6726
count_unflushed_space_cb(spa_t *spa, space_map_entry_t *sme,
6727
uint64_t txg, void *arg)
6728
{
6729
int64_t *ualloc_space = arg;
6730
6731
uint64_t offset = sme->sme_offset;
6732
uint64_t vdev_id = sme->sme_vdev;
6733
6734
vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6735
if (!vdev_is_concrete(vd))
6736
return (0);
6737
6738
metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6739
ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6740
6741
if (txg < metaslab_unflushed_txg(ms))
6742
return (0);
6743
6744
if (sme->sme_type == SM_ALLOC)
6745
*ualloc_space += sme->sme_run;
6746
else
6747
*ualloc_space -= sme->sme_run;
6748
6749
return (0);
6750
}
6751
6752
static int64_t
6753
get_unflushed_alloc_space(spa_t *spa)
6754
{
6755
if (dump_opt['L'])
6756
return (0);
6757
6758
int64_t ualloc_space = 0;
6759
iterate_through_spacemap_logs(spa, count_unflushed_space_cb,
6760
&ualloc_space);
6761
return (ualloc_space);
6762
}
6763
6764
static int
6765
load_unflushed_cb(spa_t *spa, space_map_entry_t *sme, uint64_t txg, void *arg)
6766
{
6767
maptype_t *uic_maptype = arg;
6768
6769
uint64_t offset = sme->sme_offset;
6770
uint64_t size = sme->sme_run;
6771
uint64_t vdev_id = sme->sme_vdev;
6772
6773
vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6774
6775
/* skip indirect vdevs */
6776
if (!vdev_is_concrete(vd))
6777
return (0);
6778
6779
metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6780
6781
ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6782
ASSERT(*uic_maptype == SM_ALLOC || *uic_maptype == SM_FREE);
6783
6784
if (txg < metaslab_unflushed_txg(ms))
6785
return (0);
6786
6787
if (*uic_maptype == sme->sme_type)
6788
zfs_range_tree_add(ms->ms_allocatable, offset, size);
6789
else
6790
zfs_range_tree_remove(ms->ms_allocatable, offset, size);
6791
6792
return (0);
6793
}
6794
6795
static void
6796
load_unflushed_to_ms_allocatables(spa_t *spa, maptype_t maptype)
6797
{
6798
iterate_through_spacemap_logs(spa, load_unflushed_cb, &maptype);
6799
}
6800
6801
static void
6802
load_concrete_ms_allocatable_trees(spa_t *spa, maptype_t maptype)
6803
{
6804
vdev_t *rvd = spa->spa_root_vdev;
6805
for (uint64_t i = 0; i < rvd->vdev_children; i++) {
6806
vdev_t *vd = rvd->vdev_child[i];
6807
6808
ASSERT3U(i, ==, vd->vdev_id);
6809
6810
if (vd->vdev_ops == &vdev_indirect_ops)
6811
continue;
6812
6813
for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6814
metaslab_t *msp = vd->vdev_ms[m];
6815
6816
(void) fprintf(stderr,
6817
"\rloading concrete vdev %llu, "
6818
"metaslab %llu of %llu ...",
6819
(longlong_t)vd->vdev_id,
6820
(longlong_t)msp->ms_id,
6821
(longlong_t)vd->vdev_ms_count);
6822
6823
mutex_enter(&msp->ms_lock);
6824
zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6825
6826
/*
6827
* We don't want to spend the CPU manipulating the
6828
* size-ordered tree, so clear the range_tree ops.
6829
*/
6830
msp->ms_allocatable->rt_ops = NULL;
6831
6832
if (msp->ms_sm != NULL) {
6833
VERIFY0(space_map_load(msp->ms_sm,
6834
msp->ms_allocatable, maptype));
6835
}
6836
if (!msp->ms_loaded)
6837
msp->ms_loaded = B_TRUE;
6838
mutex_exit(&msp->ms_lock);
6839
}
6840
}
6841
6842
load_unflushed_to_ms_allocatables(spa, maptype);
6843
}
6844
6845
/*
6846
* vm_idxp is an in-out parameter which (for indirect vdevs) is the
6847
* index in vim_entries that has the first entry in this metaslab.
6848
* On return, it will be set to the first entry after this metaslab.
6849
*/
6850
static void
6851
load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab_t *msp,
6852
uint64_t *vim_idxp)
6853
{
6854
vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6855
6856
mutex_enter(&msp->ms_lock);
6857
zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6858
6859
/*
6860
* We don't want to spend the CPU manipulating the
6861
* size-ordered tree, so clear the range_tree ops.
6862
*/
6863
msp->ms_allocatable->rt_ops = NULL;
6864
6865
for (; *vim_idxp < vdev_indirect_mapping_num_entries(vim);
6866
(*vim_idxp)++) {
6867
vdev_indirect_mapping_entry_phys_t *vimep =
6868
&vim->vim_entries[*vim_idxp];
6869
uint64_t ent_offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6870
uint64_t ent_len = DVA_GET_ASIZE(&vimep->vimep_dst);
6871
ASSERT3U(ent_offset, >=, msp->ms_start);
6872
if (ent_offset >= msp->ms_start + msp->ms_size)
6873
break;
6874
6875
/*
6876
* Mappings do not cross metaslab boundaries,
6877
* because we create them by walking the metaslabs.
6878
*/
6879
ASSERT3U(ent_offset + ent_len, <=,
6880
msp->ms_start + msp->ms_size);
6881
zfs_range_tree_add(msp->ms_allocatable, ent_offset, ent_len);
6882
}
6883
6884
if (!msp->ms_loaded)
6885
msp->ms_loaded = B_TRUE;
6886
mutex_exit(&msp->ms_lock);
6887
}
6888
6889
static void
6890
zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb)
6891
{
6892
ASSERT(!dump_opt['L']);
6893
6894
vdev_t *rvd = spa->spa_root_vdev;
6895
for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6896
vdev_t *vd = rvd->vdev_child[c];
6897
6898
ASSERT3U(c, ==, vd->vdev_id);
6899
6900
if (vd->vdev_ops != &vdev_indirect_ops)
6901
continue;
6902
6903
/*
6904
* Note: we don't check for mapping leaks on
6905
* removing vdevs because their ms_allocatable's
6906
* are used to look for leaks in allocated space.
6907
*/
6908
zcb->zcb_vd_obsolete_counts[c] = zdb_load_obsolete_counts(vd);
6909
6910
/*
6911
* Normally, indirect vdevs don't have any
6912
* metaslabs. We want to set them up for
6913
* zio_claim().
6914
*/
6915
vdev_metaslab_group_create(vd);
6916
VERIFY0(vdev_metaslab_init(vd, 0));
6917
6918
vdev_indirect_mapping_t *vim __maybe_unused =
6919
vd->vdev_indirect_mapping;
6920
uint64_t vim_idx = 0;
6921
for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6922
6923
(void) fprintf(stderr,
6924
"\rloading indirect vdev %llu, "
6925
"metaslab %llu of %llu ...",
6926
(longlong_t)vd->vdev_id,
6927
(longlong_t)vd->vdev_ms[m]->ms_id,
6928
(longlong_t)vd->vdev_ms_count);
6929
6930
load_indirect_ms_allocatable_tree(vd, vd->vdev_ms[m],
6931
&vim_idx);
6932
}
6933
ASSERT3U(vim_idx, ==, vdev_indirect_mapping_num_entries(vim));
6934
}
6935
}
6936
6937
static void
6938
zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
6939
{
6940
zcb->zcb_spa = spa;
6941
6942
if (dump_opt['L'])
6943
return;
6944
6945
dsl_pool_t *dp = spa->spa_dsl_pool;
6946
vdev_t *rvd = spa->spa_root_vdev;
6947
6948
/*
6949
* We are going to be changing the meaning of the metaslab's
6950
* ms_allocatable. Ensure that the allocator doesn't try to
6951
* use the tree.
6952
*/
6953
spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
6954
spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
6955
spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6956
spa->spa_special_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6957
6958
zcb->zcb_vd_obsolete_counts =
6959
umem_zalloc(rvd->vdev_children * sizeof (uint32_t *),
6960
UMEM_NOFAIL);
6961
6962
/*
6963
* For leak detection, we overload the ms_allocatable trees
6964
* to contain allocated segments instead of free segments.
6965
* As a result, we can't use the normal metaslab_load/unload
6966
* interfaces.
6967
*/
6968
zdb_leak_init_prepare_indirect_vdevs(spa, zcb);
6969
load_concrete_ms_allocatable_trees(spa, SM_ALLOC);
6970
6971
/*
6972
* On load_concrete_ms_allocatable_trees() we loaded all the
6973
* allocated entries from the ms_sm to the ms_allocatable for
6974
* each metaslab. If the pool has a checkpoint or is in the
6975
* middle of discarding a checkpoint, some of these blocks
6976
* may have been freed but their ms_sm may not have been
6977
* updated because they are referenced by the checkpoint. In
6978
* order to avoid false-positives during leak-detection, we
6979
* go through the vdev's checkpoint space map and exclude all
6980
* its entries from their relevant ms_allocatable.
6981
*
6982
* We also aggregate the space held by the checkpoint and add
6983
* it to zcb_checkpoint_size.
6984
*
6985
* Note that at this point we are also verifying that all the
6986
* entries on the checkpoint_sm are marked as allocated in
6987
* the ms_sm of their relevant metaslab.
6988
* [see comment in checkpoint_sm_exclude_entry_cb()]
6989
*/
6990
zdb_leak_init_exclude_checkpoint(spa, zcb);
6991
ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa));
6992
6993
/* for cleaner progress output */
6994
(void) fprintf(stderr, "\n");
6995
6996
if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
6997
ASSERT(spa_feature_is_enabled(spa,
6998
SPA_FEATURE_DEVICE_REMOVAL));
6999
(void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj,
7000
increment_indirect_mapping_cb, zcb, NULL);
7001
}
7002
}
7003
7004
static boolean_t
7005
zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb)
7006
{
7007
boolean_t leaks = B_FALSE;
7008
vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7009
uint64_t total_leaked = 0;
7010
boolean_t are_precise = B_FALSE;
7011
7012
ASSERT(vim != NULL);
7013
7014
for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
7015
vdev_indirect_mapping_entry_phys_t *vimep =
7016
&vim->vim_entries[i];
7017
uint64_t obsolete_bytes = 0;
7018
uint64_t offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
7019
metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
7020
7021
/*
7022
* This is not very efficient but it's easy to
7023
* verify correctness.
7024
*/
7025
for (uint64_t inner_offset = 0;
7026
inner_offset < DVA_GET_ASIZE(&vimep->vimep_dst);
7027
inner_offset += 1ULL << vd->vdev_ashift) {
7028
if (zfs_range_tree_contains(msp->ms_allocatable,
7029
offset + inner_offset, 1ULL << vd->vdev_ashift)) {
7030
obsolete_bytes += 1ULL << vd->vdev_ashift;
7031
}
7032
}
7033
7034
int64_t bytes_leaked = obsolete_bytes -
7035
zcb->zcb_vd_obsolete_counts[vd->vdev_id][i];
7036
ASSERT3U(DVA_GET_ASIZE(&vimep->vimep_dst), >=,
7037
zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]);
7038
7039
VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7040
if (bytes_leaked != 0 && (are_precise || dump_opt['d'] >= 5)) {
7041
(void) printf("obsolete indirect mapping count "
7042
"mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
7043
(u_longlong_t)vd->vdev_id,
7044
(u_longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
7045
(u_longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
7046
(u_longlong_t)bytes_leaked);
7047
}
7048
total_leaked += ABS(bytes_leaked);
7049
}
7050
7051
VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7052
if (!are_precise && total_leaked > 0) {
7053
int pct_leaked = total_leaked * 100 /
7054
vdev_indirect_mapping_bytes_mapped(vim);
7055
(void) printf("cannot verify obsolete indirect mapping "
7056
"counts of vdev %llu because precise feature was not "
7057
"enabled when it was removed: %d%% (%llx bytes) of mapping"
7058
"unreferenced\n",
7059
(u_longlong_t)vd->vdev_id, pct_leaked,
7060
(u_longlong_t)total_leaked);
7061
} else if (total_leaked > 0) {
7062
(void) printf("obsolete indirect mapping count mismatch "
7063
"for vdev %llu -- %llx total bytes mismatched\n",
7064
(u_longlong_t)vd->vdev_id,
7065
(u_longlong_t)total_leaked);
7066
leaks |= B_TRUE;
7067
}
7068
7069
vdev_indirect_mapping_free_obsolete_counts(vim,
7070
zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
7071
zcb->zcb_vd_obsolete_counts[vd->vdev_id] = NULL;
7072
7073
return (leaks);
7074
}
7075
7076
static boolean_t
7077
zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb)
7078
{
7079
if (dump_opt['L'])
7080
return (B_FALSE);
7081
7082
boolean_t leaks = B_FALSE;
7083
vdev_t *rvd = spa->spa_root_vdev;
7084
for (unsigned c = 0; c < rvd->vdev_children; c++) {
7085
vdev_t *vd = rvd->vdev_child[c];
7086
7087
if (zcb->zcb_vd_obsolete_counts[c] != NULL) {
7088
leaks |= zdb_check_for_obsolete_leaks(vd, zcb);
7089
}
7090
7091
for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
7092
metaslab_t *msp = vd->vdev_ms[m];
7093
ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class ==
7094
spa_embedded_log_class(spa) ||
7095
msp->ms_group->mg_class ==
7096
spa_special_embedded_log_class(spa)) ?
7097
vd->vdev_log_mg : vd->vdev_mg);
7098
7099
/*
7100
* ms_allocatable has been overloaded
7101
* to contain allocated segments. Now that
7102
* we finished traversing all blocks, any
7103
* block that remains in the ms_allocatable
7104
* represents an allocated block that we
7105
* did not claim during the traversal.
7106
* Claimed blocks would have been removed
7107
* from the ms_allocatable. For indirect
7108
* vdevs, space remaining in the tree
7109
* represents parts of the mapping that are
7110
* not referenced, which is not a bug.
7111
*/
7112
if (vd->vdev_ops == &vdev_indirect_ops) {
7113
zfs_range_tree_vacate(msp->ms_allocatable,
7114
NULL, NULL);
7115
} else {
7116
zfs_range_tree_vacate(msp->ms_allocatable,
7117
zdb_leak, vd);
7118
}
7119
if (msp->ms_loaded) {
7120
msp->ms_loaded = B_FALSE;
7121
}
7122
}
7123
}
7124
7125
umem_free(zcb->zcb_vd_obsolete_counts,
7126
rvd->vdev_children * sizeof (uint32_t *));
7127
zcb->zcb_vd_obsolete_counts = NULL;
7128
7129
return (leaks);
7130
}
7131
7132
static int
7133
count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7134
{
7135
(void) tx;
7136
zdb_cb_t *zcb = arg;
7137
7138
if (dump_opt['b'] >= 5) {
7139
char blkbuf[BP_SPRINTF_LEN];
7140
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
7141
(void) printf("[%s] %s\n",
7142
"deferred free", blkbuf);
7143
}
7144
zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED);
7145
return (0);
7146
}
7147
7148
/*
7149
* Iterate over livelists which have been destroyed by the user but
7150
* are still present in the MOS, waiting to be freed
7151
*/
7152
static void
7153
iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg)
7154
{
7155
objset_t *mos = spa->spa_meta_objset;
7156
uint64_t zap_obj;
7157
int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
7158
DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
7159
if (err == ENOENT)
7160
return;
7161
ASSERT0(err);
7162
7163
zap_cursor_t zc;
7164
zap_attribute_t *attrp = zap_attribute_alloc();
7165
dsl_deadlist_t ll;
7166
/* NULL out os prior to dsl_deadlist_open in case it's garbage */
7167
ll.dl_os = NULL;
7168
for (zap_cursor_init(&zc, mos, zap_obj);
7169
zap_cursor_retrieve(&zc, attrp) == 0;
7170
(void) zap_cursor_advance(&zc)) {
7171
VERIFY0(dsl_deadlist_open(&ll, mos, attrp->za_first_integer));
7172
func(&ll, arg);
7173
dsl_deadlist_close(&ll);
7174
}
7175
zap_cursor_fini(&zc);
7176
zap_attribute_free(attrp);
7177
}
7178
7179
static int
7180
bpobj_count_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
7181
dmu_tx_t *tx)
7182
{
7183
ASSERT(!bp_freed);
7184
return (count_block_cb(arg, bp, tx));
7185
}
7186
7187
static int
7188
livelist_entry_count_blocks_cb(void *args, dsl_deadlist_entry_t *dle)
7189
{
7190
zdb_cb_t *zbc = args;
7191
bplist_t blks;
7192
bplist_create(&blks);
7193
/* determine which blocks have been alloc'd but not freed */
7194
VERIFY0(dsl_process_sub_livelist(&dle->dle_bpobj, &blks, NULL, NULL));
7195
/* count those blocks */
7196
(void) bplist_iterate(&blks, count_block_cb, zbc, NULL);
7197
bplist_destroy(&blks);
7198
return (0);
7199
}
7200
7201
static void
7202
livelist_count_blocks(dsl_deadlist_t *ll, void *arg)
7203
{
7204
dsl_deadlist_iterate(ll, livelist_entry_count_blocks_cb, arg);
7205
}
7206
7207
/*
7208
* Count the blocks in the livelists that have been destroyed by the user
7209
* but haven't yet been freed.
7210
*/
7211
static void
7212
deleted_livelists_count_blocks(spa_t *spa, zdb_cb_t *zbc)
7213
{
7214
iterate_deleted_livelists(spa, livelist_count_blocks, zbc);
7215
}
7216
7217
static void
7218
dump_livelist_cb(dsl_deadlist_t *ll, void *arg)
7219
{
7220
ASSERT0P(arg);
7221
global_feature_count[SPA_FEATURE_LIVELIST]++;
7222
dump_blkptr_list(ll, "Deleted Livelist");
7223
dsl_deadlist_iterate(ll, sublivelist_verify_lightweight, NULL);
7224
}
7225
7226
/*
7227
* Print out, register object references to, and increment feature counts for
7228
* livelists that have been destroyed by the user but haven't yet been freed.
7229
*/
7230
static void
7231
deleted_livelists_dump_mos(spa_t *spa)
7232
{
7233
uint64_t zap_obj;
7234
objset_t *mos = spa->spa_meta_objset;
7235
int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
7236
DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
7237
if (err == ENOENT)
7238
return;
7239
mos_obj_refd(zap_obj);
7240
iterate_deleted_livelists(spa, dump_livelist_cb, NULL);
7241
}
7242
7243
static int
7244
zdb_brt_entry_compare(const void *zcn1, const void *zcn2)
7245
{
7246
const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva;
7247
const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva;
7248
int cmp;
7249
7250
cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
7251
if (cmp == 0)
7252
cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2));
7253
7254
return (cmp);
7255
}
7256
7257
static int
7258
dump_block_stats(spa_t *spa)
7259
{
7260
zdb_cb_t *zcb;
7261
zdb_blkstats_t *zb, *tzb;
7262
uint64_t norm_alloc, norm_space, total_alloc, total_found;
7263
int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7264
TRAVERSE_NO_DECRYPT | TRAVERSE_HARD;
7265
boolean_t leaks = B_FALSE;
7266
int e, c, err;
7267
bp_embedded_type_t i;
7268
7269
ddt_prefetch_all(spa);
7270
7271
zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL);
7272
7273
if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
7274
avl_create(&zcb->zcb_brt, zdb_brt_entry_compare,
7275
sizeof (zdb_brt_entry_t),
7276
offsetof(zdb_brt_entry_t, zbre_node));
7277
zcb->zcb_brt_is_active = B_TRUE;
7278
}
7279
7280
(void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
7281
(dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
7282
(dump_opt['c'] == 1) ? "metadata " : "",
7283
dump_opt['c'] ? "checksums " : "",
7284
(dump_opt['c'] && !dump_opt['L']) ? "and verify " : "",
7285
!dump_opt['L'] ? "nothing leaked " : "");
7286
7287
/*
7288
* When leak detection is enabled we load all space maps as SM_ALLOC
7289
* maps, then traverse the pool claiming each block we discover. If
7290
* the pool is perfectly consistent, the segment trees will be empty
7291
* when we're done. Anything left over is a leak; any block we can't
7292
* claim (because it's not part of any space map) is a double
7293
* allocation, reference to a freed block, or an unclaimed log block.
7294
*
7295
* When leak detection is disabled (-L option) we still traverse the
7296
* pool claiming each block we discover, but we skip opening any space
7297
* maps.
7298
*/
7299
zdb_leak_init(spa, zcb);
7300
7301
/*
7302
* If there's a deferred-free bplist, process that first.
7303
*/
7304
(void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
7305
bpobj_count_block_cb, zcb, NULL);
7306
7307
if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
7308
(void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
7309
bpobj_count_block_cb, zcb, NULL);
7310
}
7311
7312
zdb_claim_removing(spa, zcb);
7313
7314
if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
7315
VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,
7316
spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb,
7317
zcb, NULL));
7318
}
7319
7320
deleted_livelists_count_blocks(spa, zcb);
7321
7322
if (dump_opt['c'] > 1)
7323
flags |= TRAVERSE_PREFETCH_DATA;
7324
7325
zcb->zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
7326
zcb->zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa));
7327
zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
7328
zcb->zcb_totalasize +=
7329
metaslab_class_get_alloc(spa_embedded_log_class(spa));
7330
zcb->zcb_totalasize +=
7331
metaslab_class_get_alloc(spa_special_embedded_log_class(spa));
7332
zcb->zcb_start = zcb->zcb_lastprint = gethrtime();
7333
err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb);
7334
7335
/*
7336
* If we've traversed the data blocks then we need to wait for those
7337
* I/Os to complete. We leverage "The Godfather" zio to wait on
7338
* all async I/Os to complete.
7339
*/
7340
if (dump_opt['c']) {
7341
for (c = 0; c < max_ncpus; c++) {
7342
(void) zio_wait(spa->spa_async_zio_root[c]);
7343
spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL,
7344
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
7345
ZIO_FLAG_GODFATHER);
7346
}
7347
}
7348
ASSERT0(spa->spa_load_verify_bytes);
7349
7350
/*
7351
* Done after zio_wait() since zcb_haderrors is modified in
7352
* zdb_blkptr_done()
7353
*/
7354
zcb->zcb_haderrors |= err;
7355
7356
if (zcb->zcb_haderrors) {
7357
(void) printf("\nError counts:\n\n");
7358
(void) printf("\t%5s %s\n", "errno", "count");
7359
for (e = 0; e < 256; e++) {
7360
if (zcb->zcb_errors[e] != 0) {
7361
(void) printf("\t%5d %llu\n",
7362
e, (u_longlong_t)zcb->zcb_errors[e]);
7363
}
7364
}
7365
}
7366
7367
/*
7368
* Report any leaked segments.
7369
*/
7370
leaks |= zdb_leak_fini(spa, zcb);
7371
7372
tzb = &zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL];
7373
7374
norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7375
norm_space = metaslab_class_get_space(spa_normal_class(spa));
7376
7377
total_alloc = norm_alloc +
7378
metaslab_class_get_alloc(spa_log_class(spa)) +
7379
metaslab_class_get_alloc(spa_embedded_log_class(spa)) +
7380
metaslab_class_get_alloc(spa_special_embedded_log_class(spa)) +
7381
metaslab_class_get_alloc(spa_special_class(spa)) +
7382
metaslab_class_get_alloc(spa_dedup_class(spa)) +
7383
get_unflushed_alloc_space(spa);
7384
total_found =
7385
tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize +
7386
zcb->zcb_removing_size + zcb->zcb_checkpoint_size;
7387
7388
if (total_found == total_alloc && !dump_opt['L']) {
7389
(void) printf("\n\tNo leaks (block sum matches space"
7390
" maps exactly)\n");
7391
} else if (!dump_opt['L']) {
7392
(void) printf("block traversal size %llu != alloc %llu "
7393
"(%s %lld)\n",
7394
(u_longlong_t)total_found,
7395
(u_longlong_t)total_alloc,
7396
(dump_opt['L']) ? "unreachable" : "leaked",
7397
(longlong_t)(total_alloc - total_found));
7398
}
7399
7400
if (tzb->zb_count == 0) {
7401
umem_free(zcb, sizeof (zdb_cb_t));
7402
return (2);
7403
}
7404
7405
(void) printf("\n");
7406
(void) printf("\t%-16s %14llu\n", "bp count:",
7407
(u_longlong_t)tzb->zb_count);
7408
(void) printf("\t%-16s %14llu\n", "ganged count:",
7409
(longlong_t)tzb->zb_gangs);
7410
(void) printf("\t%-16s %14llu avg: %6llu\n", "bp logical:",
7411
(u_longlong_t)tzb->zb_lsize,
7412
(u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
7413
(void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7414
"bp physical:", (u_longlong_t)tzb->zb_psize,
7415
(u_longlong_t)(tzb->zb_psize / tzb->zb_count),
7416
(double)tzb->zb_lsize / tzb->zb_psize);
7417
(void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
7418
"bp allocated:", (u_longlong_t)tzb->zb_asize,
7419
(u_longlong_t)(tzb->zb_asize / tzb->zb_count),
7420
(double)tzb->zb_lsize / tzb->zb_asize);
7421
(void) printf("\t%-16s %14llu ref>1: %6llu deduplication: %6.2f\n",
7422
"bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize,
7423
(u_longlong_t)zcb->zcb_dedup_blocks,
7424
(double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0);
7425
(void) printf("\t%-16s %14llu count: %6llu\n",
7426
"bp cloned:", (u_longlong_t)zcb->zcb_clone_asize,
7427
(u_longlong_t)zcb->zcb_clone_blocks);
7428
(void) printf("\t%-16s %14llu used: %5.2f%%\n", "Normal class:",
7429
(u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
7430
7431
if (spa_special_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7432
uint64_t alloc = metaslab_class_get_alloc(
7433
spa_special_class(spa));
7434
uint64_t space = metaslab_class_get_space(
7435
spa_special_class(spa));
7436
7437
(void) printf("\t%-16s %14llu used: %5.2f%%\n",
7438
"Special class", (u_longlong_t)alloc,
7439
100.0 * alloc / space);
7440
}
7441
7442
if (spa_dedup_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7443
uint64_t alloc = metaslab_class_get_alloc(
7444
spa_dedup_class(spa));
7445
uint64_t space = metaslab_class_get_space(
7446
spa_dedup_class(spa));
7447
7448
(void) printf("\t%-16s %14llu used: %5.2f%%\n",
7449
"Dedup class", (u_longlong_t)alloc,
7450
100.0 * alloc / space);
7451
}
7452
7453
if (spa_embedded_log_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7454
uint64_t alloc = metaslab_class_get_alloc(
7455
spa_embedded_log_class(spa));
7456
uint64_t space = metaslab_class_get_space(
7457
spa_embedded_log_class(spa));
7458
7459
(void) printf("\t%-16s %14llu used: %5.2f%%\n",
7460
"Embedded log class", (u_longlong_t)alloc,
7461
100.0 * alloc / space);
7462
}
7463
7464
if (spa_special_embedded_log_class(spa)->mc_allocator[0].mca_rotor
7465
!= NULL) {
7466
uint64_t alloc = metaslab_class_get_alloc(
7467
spa_special_embedded_log_class(spa));
7468
uint64_t space = metaslab_class_get_space(
7469
spa_special_embedded_log_class(spa));
7470
7471
(void) printf("\t%-16s %14llu used: %5.2f%%\n",
7472
"Special embedded log", (u_longlong_t)alloc,
7473
100.0 * alloc / space);
7474
}
7475
7476
for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
7477
if (zcb->zcb_embedded_blocks[i] == 0)
7478
continue;
7479
(void) printf("\n");
7480
(void) printf("\tadditional, non-pointer bps of type %u: "
7481
"%10llu\n",
7482
i, (u_longlong_t)zcb->zcb_embedded_blocks[i]);
7483
7484
if (dump_opt['b'] >= 3) {
7485
(void) printf("\t number of (compressed) bytes: "
7486
"number of bps\n");
7487
dump_histogram(zcb->zcb_embedded_histogram[i],
7488
sizeof (zcb->zcb_embedded_histogram[i]) /
7489
sizeof (zcb->zcb_embedded_histogram[i][0]), 0);
7490
}
7491
}
7492
7493
if (tzb->zb_ditto_samevdev != 0) {
7494
(void) printf("\tDittoed blocks on same vdev: %llu\n",
7495
(longlong_t)tzb->zb_ditto_samevdev);
7496
}
7497
if (tzb->zb_ditto_same_ms != 0) {
7498
(void) printf("\tDittoed blocks in same metaslab: %llu\n",
7499
(longlong_t)tzb->zb_ditto_same_ms);
7500
}
7501
7502
for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) {
7503
vdev_t *vd = spa->spa_root_vdev->vdev_child[v];
7504
vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7505
7506
if (vim == NULL) {
7507
continue;
7508
}
7509
7510
char mem[32];
7511
zdb_nicenum(vdev_indirect_mapping_num_entries(vim),
7512
mem, vdev_indirect_mapping_size(vim));
7513
7514
(void) printf("\tindirect vdev id %llu has %llu segments "
7515
"(%s in memory)\n",
7516
(longlong_t)vd->vdev_id,
7517
(longlong_t)vdev_indirect_mapping_num_entries(vim), mem);
7518
}
7519
7520
if (dump_opt['b'] >= 2) {
7521
int l, t, level;
7522
char csize[32], lsize[32], psize[32], asize[32];
7523
char avg[32], gang[32];
7524
(void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
7525
"\t avg\t comp\t%%Total\tType\n");
7526
7527
zfs_blkstat_t *mdstats = umem_zalloc(sizeof (zfs_blkstat_t),
7528
UMEM_NOFAIL);
7529
7530
for (t = 0; t <= ZDB_OT_TOTAL; t++) {
7531
const char *typename;
7532
7533
/* make sure nicenum has enough space */
7534
_Static_assert(sizeof (csize) >= NN_NUMBUF_SZ,
7535
"csize truncated");
7536
_Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ,
7537
"lsize truncated");
7538
_Static_assert(sizeof (psize) >= NN_NUMBUF_SZ,
7539
"psize truncated");
7540
_Static_assert(sizeof (asize) >= NN_NUMBUF_SZ,
7541
"asize truncated");
7542
_Static_assert(sizeof (avg) >= NN_NUMBUF_SZ,
7543
"avg truncated");
7544
_Static_assert(sizeof (gang) >= NN_NUMBUF_SZ,
7545
"gang truncated");
7546
7547
if (t < DMU_OT_NUMTYPES)
7548
typename = dmu_ot[t].ot_name;
7549
else
7550
typename = zdb_ot_extname[t - DMU_OT_NUMTYPES];
7551
7552
if (zcb->zcb_type[ZB_TOTAL][t].zb_asize == 0) {
7553
(void) printf("%6s\t%5s\t%5s\t%5s"
7554
"\t%5s\t%5s\t%6s\t%s\n",
7555
"-",
7556
"-",
7557
"-",
7558
"-",
7559
"-",
7560
"-",
7561
"-",
7562
typename);
7563
continue;
7564
}
7565
7566
for (l = ZB_TOTAL - 1; l >= -1; l--) {
7567
level = (l == -1 ? ZB_TOTAL : l);
7568
zb = &zcb->zcb_type[level][t];
7569
7570
if (zb->zb_asize == 0)
7571
continue;
7572
7573
if (level != ZB_TOTAL && t < DMU_OT_NUMTYPES &&
7574
(level > 0 || DMU_OT_IS_METADATA(t))) {
7575
mdstats->zb_count += zb->zb_count;
7576
mdstats->zb_lsize += zb->zb_lsize;
7577
mdstats->zb_psize += zb->zb_psize;
7578
mdstats->zb_asize += zb->zb_asize;
7579
mdstats->zb_gangs += zb->zb_gangs;
7580
}
7581
7582
if (dump_opt['b'] < 3 && level != ZB_TOTAL)
7583
continue;
7584
7585
if (level == 0 && zb->zb_asize ==
7586
zcb->zcb_type[ZB_TOTAL][t].zb_asize)
7587
continue;
7588
7589
zdb_nicenum(zb->zb_count, csize,
7590
sizeof (csize));
7591
zdb_nicenum(zb->zb_lsize, lsize,
7592
sizeof (lsize));
7593
zdb_nicenum(zb->zb_psize, psize,
7594
sizeof (psize));
7595
zdb_nicenum(zb->zb_asize, asize,
7596
sizeof (asize));
7597
zdb_nicenum(zb->zb_asize / zb->zb_count, avg,
7598
sizeof (avg));
7599
zdb_nicenum(zb->zb_gangs, gang, sizeof (gang));
7600
7601
(void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7602
"\t%5.2f\t%6.2f\t",
7603
csize, lsize, psize, asize, avg,
7604
(double)zb->zb_lsize / zb->zb_psize,
7605
100.0 * zb->zb_asize / tzb->zb_asize);
7606
7607
if (level == ZB_TOTAL)
7608
(void) printf("%s\n", typename);
7609
else
7610
(void) printf(" L%d %s\n",
7611
level, typename);
7612
7613
if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) {
7614
(void) printf("\t number of ganged "
7615
"blocks: %s\n", gang);
7616
}
7617
7618
if (dump_opt['b'] >= 4) {
7619
(void) printf("psize "
7620
"(in 512-byte sectors): "
7621
"number of blocks\n");
7622
dump_histogram(zb->zb_psize_histogram,
7623
PSIZE_HISTO_SIZE, 0);
7624
}
7625
}
7626
}
7627
zdb_nicenum(mdstats->zb_count, csize,
7628
sizeof (csize));
7629
zdb_nicenum(mdstats->zb_lsize, lsize,
7630
sizeof (lsize));
7631
zdb_nicenum(mdstats->zb_psize, psize,
7632
sizeof (psize));
7633
zdb_nicenum(mdstats->zb_asize, asize,
7634
sizeof (asize));
7635
zdb_nicenum(mdstats->zb_asize / mdstats->zb_count, avg,
7636
sizeof (avg));
7637
zdb_nicenum(mdstats->zb_gangs, gang, sizeof (gang));
7638
7639
(void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7640
"\t%5.2f\t%6.2f\t",
7641
csize, lsize, psize, asize, avg,
7642
(double)mdstats->zb_lsize / mdstats->zb_psize,
7643
100.0 * mdstats->zb_asize / tzb->zb_asize);
7644
(void) printf("%s\n", "Metadata Total");
7645
7646
/* Output a table summarizing block sizes in the pool */
7647
if (dump_opt['b'] >= 2) {
7648
dump_size_histograms(zcb);
7649
}
7650
7651
umem_free(mdstats, sizeof (zfs_blkstat_t));
7652
}
7653
7654
(void) printf("\n");
7655
7656
if (leaks) {
7657
umem_free(zcb, sizeof (zdb_cb_t));
7658
return (2);
7659
}
7660
7661
if (zcb->zcb_haderrors) {
7662
umem_free(zcb, sizeof (zdb_cb_t));
7663
return (3);
7664
}
7665
7666
umem_free(zcb, sizeof (zdb_cb_t));
7667
return (0);
7668
}
7669
7670
typedef struct zdb_ddt_entry {
7671
/* key must be first for ddt_key_compare */
7672
ddt_key_t zdde_key;
7673
uint64_t zdde_ref_blocks;
7674
uint64_t zdde_ref_lsize;
7675
uint64_t zdde_ref_psize;
7676
uint64_t zdde_ref_dsize;
7677
avl_node_t zdde_node;
7678
} zdb_ddt_entry_t;
7679
7680
static int
7681
zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
7682
const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
7683
{
7684
(void) zilog, (void) dnp;
7685
avl_tree_t *t = arg;
7686
avl_index_t where;
7687
zdb_ddt_entry_t *zdde, zdde_search;
7688
7689
if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
7690
BP_IS_EMBEDDED(bp))
7691
return (0);
7692
7693
if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
7694
(void) printf("traversing objset %llu, %llu objects, "
7695
"%lu blocks so far\n",
7696
(u_longlong_t)zb->zb_objset,
7697
(u_longlong_t)BP_GET_FILL(bp),
7698
avl_numnodes(t));
7699
}
7700
7701
if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF ||
7702
BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
7703
return (0);
7704
7705
ddt_key_fill(&zdde_search.zdde_key, bp);
7706
7707
zdde = avl_find(t, &zdde_search, &where);
7708
7709
if (zdde == NULL) {
7710
zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL);
7711
zdde->zdde_key = zdde_search.zdde_key;
7712
avl_insert(t, zdde, where);
7713
}
7714
7715
zdde->zdde_ref_blocks += 1;
7716
zdde->zdde_ref_lsize += BP_GET_LSIZE(bp);
7717
zdde->zdde_ref_psize += BP_GET_PSIZE(bp);
7718
zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp);
7719
7720
return (0);
7721
}
7722
7723
static void
7724
dump_simulated_ddt(spa_t *spa)
7725
{
7726
avl_tree_t t;
7727
void *cookie = NULL;
7728
zdb_ddt_entry_t *zdde;
7729
ddt_histogram_t ddh_total = {{{0}}};
7730
ddt_stat_t dds_total = {0};
7731
7732
avl_create(&t, ddt_key_compare,
7733
sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
7734
7735
spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7736
7737
(void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7738
TRAVERSE_NO_DECRYPT, zdb_ddt_add_cb, &t);
7739
7740
spa_config_exit(spa, SCL_CONFIG, FTAG);
7741
7742
while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) {
7743
uint64_t refcnt = zdde->zdde_ref_blocks;
7744
ASSERT(refcnt != 0);
7745
7746
ddt_stat_t *dds = &ddh_total.ddh_stat[highbit64(refcnt) - 1];
7747
7748
dds->dds_blocks += zdde->zdde_ref_blocks / refcnt;
7749
dds->dds_lsize += zdde->zdde_ref_lsize / refcnt;
7750
dds->dds_psize += zdde->zdde_ref_psize / refcnt;
7751
dds->dds_dsize += zdde->zdde_ref_dsize / refcnt;
7752
7753
dds->dds_ref_blocks += zdde->zdde_ref_blocks;
7754
dds->dds_ref_lsize += zdde->zdde_ref_lsize;
7755
dds->dds_ref_psize += zdde->zdde_ref_psize;
7756
dds->dds_ref_dsize += zdde->zdde_ref_dsize;
7757
7758
umem_free(zdde, sizeof (*zdde));
7759
}
7760
7761
avl_destroy(&t);
7762
7763
ddt_histogram_total(&dds_total, &ddh_total);
7764
7765
(void) printf("Simulated DDT histogram:\n");
7766
7767
zpool_dump_ddt(&dds_total, &ddh_total);
7768
7769
dump_dedup_ratio(&dds_total);
7770
}
7771
7772
static int
7773
verify_device_removal_feature_counts(spa_t *spa)
7774
{
7775
uint64_t dr_feature_refcount = 0;
7776
uint64_t oc_feature_refcount = 0;
7777
uint64_t indirect_vdev_count = 0;
7778
uint64_t precise_vdev_count = 0;
7779
uint64_t obsolete_counts_object_count = 0;
7780
uint64_t obsolete_sm_count = 0;
7781
uint64_t obsolete_counts_count = 0;
7782
uint64_t scip_count = 0;
7783
uint64_t obsolete_bpobj_count = 0;
7784
int ret = 0;
7785
7786
spa_condensing_indirect_phys_t *scip =
7787
&spa->spa_condensing_indirect_phys;
7788
if (scip->scip_next_mapping_object != 0) {
7789
vdev_t *vd = spa->spa_root_vdev->vdev_child[scip->scip_vdev];
7790
ASSERT(scip->scip_prev_obsolete_sm_object != 0);
7791
ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
7792
7793
(void) printf("Condensing indirect vdev %llu: new mapping "
7794
"object %llu, prev obsolete sm %llu\n",
7795
(u_longlong_t)scip->scip_vdev,
7796
(u_longlong_t)scip->scip_next_mapping_object,
7797
(u_longlong_t)scip->scip_prev_obsolete_sm_object);
7798
if (scip->scip_prev_obsolete_sm_object != 0) {
7799
space_map_t *prev_obsolete_sm = NULL;
7800
VERIFY0(space_map_open(&prev_obsolete_sm,
7801
spa->spa_meta_objset,
7802
scip->scip_prev_obsolete_sm_object,
7803
0, vd->vdev_asize, 0));
7804
dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm);
7805
(void) printf("\n");
7806
space_map_close(prev_obsolete_sm);
7807
}
7808
7809
scip_count += 2;
7810
}
7811
7812
for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
7813
vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
7814
vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
7815
7816
if (vic->vic_mapping_object != 0) {
7817
ASSERT(vd->vdev_ops == &vdev_indirect_ops ||
7818
vd->vdev_removing);
7819
indirect_vdev_count++;
7820
7821
if (vd->vdev_indirect_mapping->vim_havecounts) {
7822
obsolete_counts_count++;
7823
}
7824
}
7825
7826
boolean_t are_precise;
7827
VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7828
if (are_precise) {
7829
ASSERT(vic->vic_mapping_object != 0);
7830
precise_vdev_count++;
7831
}
7832
7833
uint64_t obsolete_sm_object;
7834
VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
7835
if (obsolete_sm_object != 0) {
7836
ASSERT(vic->vic_mapping_object != 0);
7837
obsolete_sm_count++;
7838
}
7839
}
7840
7841
(void) feature_get_refcount(spa,
7842
&spa_feature_table[SPA_FEATURE_DEVICE_REMOVAL],
7843
&dr_feature_refcount);
7844
(void) feature_get_refcount(spa,
7845
&spa_feature_table[SPA_FEATURE_OBSOLETE_COUNTS],
7846
&oc_feature_refcount);
7847
7848
if (dr_feature_refcount != indirect_vdev_count) {
7849
ret = 1;
7850
(void) printf("Number of indirect vdevs (%llu) " \
7851
"does not match feature count (%llu)\n",
7852
(u_longlong_t)indirect_vdev_count,
7853
(u_longlong_t)dr_feature_refcount);
7854
} else {
7855
(void) printf("Verified device_removal feature refcount " \
7856
"of %llu is correct\n",
7857
(u_longlong_t)dr_feature_refcount);
7858
}
7859
7860
if (zap_contains(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT,
7861
DMU_POOL_OBSOLETE_BPOBJ) == 0) {
7862
obsolete_bpobj_count++;
7863
}
7864
7865
7866
obsolete_counts_object_count = precise_vdev_count;
7867
obsolete_counts_object_count += obsolete_sm_count;
7868
obsolete_counts_object_count += obsolete_counts_count;
7869
obsolete_counts_object_count += scip_count;
7870
obsolete_counts_object_count += obsolete_bpobj_count;
7871
obsolete_counts_object_count += remap_deadlist_count;
7872
7873
if (oc_feature_refcount != obsolete_counts_object_count) {
7874
ret = 1;
7875
(void) printf("Number of obsolete counts objects (%llu) " \
7876
"does not match feature count (%llu)\n",
7877
(u_longlong_t)obsolete_counts_object_count,
7878
(u_longlong_t)oc_feature_refcount);
7879
(void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
7880
"ob:%llu rd:%llu\n",
7881
(u_longlong_t)precise_vdev_count,
7882
(u_longlong_t)obsolete_sm_count,
7883
(u_longlong_t)obsolete_counts_count,
7884
(u_longlong_t)scip_count,
7885
(u_longlong_t)obsolete_bpobj_count,
7886
(u_longlong_t)remap_deadlist_count);
7887
} else {
7888
(void) printf("Verified indirect_refcount feature refcount " \
7889
"of %llu is correct\n",
7890
(u_longlong_t)oc_feature_refcount);
7891
}
7892
return (ret);
7893
}
7894
7895
static void
7896
zdb_set_skip_mmp(char *target)
7897
{
7898
spa_t *spa;
7899
7900
/*
7901
* Disable the activity check to allow examination of
7902
* active pools.
7903
*/
7904
spa_namespace_enter(FTAG);
7905
if ((spa = spa_lookup(target)) != NULL) {
7906
spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
7907
}
7908
spa_namespace_exit(FTAG);
7909
}
7910
7911
#define BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7912
/*
7913
* Import the checkpointed state of the pool specified by the target
7914
* parameter as readonly. The function also accepts a pool config
7915
* as an optional parameter, else it attempts to infer the config by
7916
* the name of the target pool.
7917
*
7918
* Note that the checkpointed state's pool name will be the name of
7919
* the original pool with the above suffix appended to it. In addition,
7920
* if the target is not a pool name (e.g. a path to a dataset) then
7921
* the new_path parameter is populated with the updated path to
7922
* reflect the fact that we are looking into the checkpointed state.
7923
*
7924
* The function returns a newly-allocated copy of the name of the
7925
* pool containing the checkpointed state. When this copy is no
7926
* longer needed it should be freed with free(3C). Same thing
7927
* applies to the new_path parameter if allocated.
7928
*/
7929
static char *
7930
import_checkpointed_state(char *target, nvlist_t *cfg, boolean_t target_is_spa,
7931
char **new_path)
7932
{
7933
int error = 0;
7934
char *poolname, *bogus_name = NULL;
7935
boolean_t freecfg = B_FALSE;
7936
7937
/* If the target is not a pool, the extract the pool name */
7938
char *path_start = strchr(target, '/');
7939
if (target_is_spa || path_start == NULL) {
7940
poolname = target;
7941
} else {
7942
size_t poolname_len = path_start - target;
7943
poolname = strndup(target, poolname_len);
7944
}
7945
7946
if (cfg == NULL) {
7947
zdb_set_skip_mmp(poolname);
7948
error = spa_get_stats(poolname, &cfg, NULL, 0);
7949
if (error != 0) {
7950
fatal("Tried to read config of pool \"%s\" but "
7951
"spa_get_stats() failed with error %d\n",
7952
poolname, error);
7953
}
7954
freecfg = B_TRUE;
7955
}
7956
7957
if (asprintf(&bogus_name, "%s%s", poolname, BOGUS_SUFFIX) == -1) {
7958
if (target != poolname)
7959
free(poolname);
7960
return (NULL);
7961
}
7962
fnvlist_add_string(cfg, ZPOOL_CONFIG_POOL_NAME, bogus_name);
7963
7964
error = spa_import(bogus_name, cfg, NULL,
7965
ZFS_IMPORT_MISSING_LOG | ZFS_IMPORT_CHECKPOINT |
7966
ZFS_IMPORT_SKIP_MMP);
7967
if (freecfg)
7968
nvlist_free(cfg);
7969
if (error != 0) {
7970
fatal("Tried to import pool \"%s\" but spa_import() failed "
7971
"with error %d\n", bogus_name, error);
7972
}
7973
7974
if (new_path != NULL && !target_is_spa) {
7975
if (asprintf(new_path, "%s%s", bogus_name,
7976
path_start != NULL ? path_start : "") == -1) {
7977
free(bogus_name);
7978
if (!target_is_spa && path_start != NULL)
7979
free(poolname);
7980
return (NULL);
7981
}
7982
}
7983
7984
if (target != poolname)
7985
free(poolname);
7986
7987
return (bogus_name);
7988
}
7989
7990
typedef struct verify_checkpoint_sm_entry_cb_arg {
7991
vdev_t *vcsec_vd;
7992
7993
/* the following fields are only used for printing progress */
7994
uint64_t vcsec_entryid;
7995
uint64_t vcsec_num_entries;
7996
} verify_checkpoint_sm_entry_cb_arg_t;
7997
7998
#define ENTRIES_PER_PROGRESS_UPDATE 10000
7999
8000
static int
8001
verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, void *arg)
8002
{
8003
verify_checkpoint_sm_entry_cb_arg_t *vcsec = arg;
8004
vdev_t *vd = vcsec->vcsec_vd;
8005
metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
8006
uint64_t end = sme->sme_offset + sme->sme_run;
8007
8008
ASSERT(sme->sme_type == SM_FREE);
8009
8010
if ((vcsec->vcsec_entryid % ENTRIES_PER_PROGRESS_UPDATE) == 0) {
8011
(void) fprintf(stderr,
8012
"\rverifying vdev %llu, space map entry %llu of %llu ...",
8013
(longlong_t)vd->vdev_id,
8014
(longlong_t)vcsec->vcsec_entryid,
8015
(longlong_t)vcsec->vcsec_num_entries);
8016
}
8017
vcsec->vcsec_entryid++;
8018
8019
/*
8020
* See comment in checkpoint_sm_exclude_entry_cb()
8021
*/
8022
VERIFY3U(sme->sme_offset, >=, ms->ms_start);
8023
VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
8024
8025
/*
8026
* The entries in the vdev_checkpoint_sm should be marked as
8027
* allocated in the checkpointed state of the pool, therefore
8028
* their respective ms_allocateable trees should not contain them.
8029
*/
8030
mutex_enter(&ms->ms_lock);
8031
zfs_range_tree_verify_not_present(ms->ms_allocatable,
8032
sme->sme_offset, sme->sme_run);
8033
mutex_exit(&ms->ms_lock);
8034
8035
return (0);
8036
}
8037
8038
/*
8039
* Verify that all segments in the vdev_checkpoint_sm are allocated
8040
* according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
8041
* ms_allocatable).
8042
*
8043
* Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
8044
* each vdev in the current state of the pool to the metaslab space maps
8045
* (ms_sm) of the checkpointed state of the pool.
8046
*
8047
* Note that the function changes the state of the ms_allocatable
8048
* trees of the current spa_t. The entries of these ms_allocatable
8049
* trees are cleared out and then repopulated from with the free
8050
* entries of their respective ms_sm space maps.
8051
*/
8052
static void
8053
verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, spa_t *current)
8054
{
8055
vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
8056
vdev_t *current_rvd = current->spa_root_vdev;
8057
8058
load_concrete_ms_allocatable_trees(checkpoint, SM_FREE);
8059
8060
for (uint64_t c = 0; c < ckpoint_rvd->vdev_children; c++) {
8061
vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[c];
8062
vdev_t *current_vd = current_rvd->vdev_child[c];
8063
8064
space_map_t *checkpoint_sm = NULL;
8065
uint64_t checkpoint_sm_obj;
8066
8067
if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
8068
/*
8069
* Since we don't allow device removal in a pool
8070
* that has a checkpoint, we expect that all removed
8071
* vdevs were removed from the pool before the
8072
* checkpoint.
8073
*/
8074
ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
8075
continue;
8076
}
8077
8078
/*
8079
* If the checkpoint space map doesn't exist, then nothing
8080
* here is checkpointed so there's nothing to verify.
8081
*/
8082
if (current_vd->vdev_top_zap == 0 ||
8083
zap_contains(spa_meta_objset(current),
8084
current_vd->vdev_top_zap,
8085
VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
8086
continue;
8087
8088
VERIFY0(zap_lookup(spa_meta_objset(current),
8089
current_vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
8090
sizeof (uint64_t), 1, &checkpoint_sm_obj));
8091
8092
VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current),
8093
checkpoint_sm_obj, 0, current_vd->vdev_asize,
8094
current_vd->vdev_ashift));
8095
8096
verify_checkpoint_sm_entry_cb_arg_t vcsec;
8097
vcsec.vcsec_vd = ckpoint_vd;
8098
vcsec.vcsec_entryid = 0;
8099
vcsec.vcsec_num_entries =
8100
space_map_length(checkpoint_sm) / sizeof (uint64_t);
8101
VERIFY0(space_map_iterate(checkpoint_sm,
8102
space_map_length(checkpoint_sm),
8103
verify_checkpoint_sm_entry_cb, &vcsec));
8104
if (dump_opt['m'] > 3)
8105
dump_spacemap(current->spa_meta_objset, checkpoint_sm);
8106
space_map_close(checkpoint_sm);
8107
}
8108
8109
/*
8110
* If we've added vdevs since we took the checkpoint, ensure
8111
* that their checkpoint space maps are empty.
8112
*/
8113
if (ckpoint_rvd->vdev_children < current_rvd->vdev_children) {
8114
for (uint64_t c = ckpoint_rvd->vdev_children;
8115
c < current_rvd->vdev_children; c++) {
8116
vdev_t *current_vd = current_rvd->vdev_child[c];
8117
VERIFY0P(current_vd->vdev_checkpoint_sm);
8118
}
8119
}
8120
8121
/* for cleaner progress output */
8122
(void) fprintf(stderr, "\n");
8123
}
8124
8125
/*
8126
* Verifies that all space that's allocated in the checkpoint is
8127
* still allocated in the current version, by checking that everything
8128
* in checkpoint's ms_allocatable (which is actually allocated, not
8129
* allocatable/free) is not present in current's ms_allocatable.
8130
*
8131
* Note that the function changes the state of the ms_allocatable
8132
* trees of both spas when called. The entries of all ms_allocatable
8133
* trees are cleared out and then repopulated from their respective
8134
* ms_sm space maps. In the checkpointed state we load the allocated
8135
* entries, and in the current state we load the free entries.
8136
*/
8137
static void
8138
verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_t *current)
8139
{
8140
vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
8141
vdev_t *current_rvd = current->spa_root_vdev;
8142
8143
load_concrete_ms_allocatable_trees(checkpoint, SM_ALLOC);
8144
load_concrete_ms_allocatable_trees(current, SM_FREE);
8145
8146
for (uint64_t i = 0; i < ckpoint_rvd->vdev_children; i++) {
8147
vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[i];
8148
vdev_t *current_vd = current_rvd->vdev_child[i];
8149
8150
if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
8151
/*
8152
* See comment in verify_checkpoint_vdev_spacemaps()
8153
*/
8154
ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
8155
continue;
8156
}
8157
8158
for (uint64_t m = 0; m < ckpoint_vd->vdev_ms_count; m++) {
8159
metaslab_t *ckpoint_msp = ckpoint_vd->vdev_ms[m];
8160
metaslab_t *current_msp = current_vd->vdev_ms[m];
8161
8162
(void) fprintf(stderr,
8163
"\rverifying vdev %llu of %llu, "
8164
"metaslab %llu of %llu ...",
8165
(longlong_t)current_vd->vdev_id,
8166
(longlong_t)current_rvd->vdev_children,
8167
(longlong_t)current_vd->vdev_ms[m]->ms_id,
8168
(longlong_t)current_vd->vdev_ms_count);
8169
8170
/*
8171
* We walk through the ms_allocatable trees that
8172
* are loaded with the allocated blocks from the
8173
* ms_sm spacemaps of the checkpoint. For each
8174
* one of these ranges we ensure that none of them
8175
* exists in the ms_allocatable trees of the
8176
* current state which are loaded with the ranges
8177
* that are currently free.
8178
*
8179
* This way we ensure that none of the blocks that
8180
* are part of the checkpoint were freed by mistake.
8181
*/
8182
zfs_range_tree_walk(ckpoint_msp->ms_allocatable,
8183
(zfs_range_tree_func_t *)
8184
zfs_range_tree_verify_not_present,
8185
current_msp->ms_allocatable);
8186
}
8187
}
8188
8189
/* for cleaner progress output */
8190
(void) fprintf(stderr, "\n");
8191
}
8192
8193
static void
8194
verify_checkpoint_blocks(spa_t *spa)
8195
{
8196
ASSERT(!dump_opt['L']);
8197
8198
spa_t *checkpoint_spa;
8199
char *checkpoint_pool;
8200
int error = 0;
8201
8202
/*
8203
* We import the checkpointed state of the pool (under a different
8204
* name) so we can do verification on it against the current state
8205
* of the pool.
8206
*/
8207
checkpoint_pool = import_checkpointed_state(spa->spa_name, NULL, B_TRUE,
8208
NULL);
8209
ASSERT(strcmp(spa->spa_name, checkpoint_pool) != 0);
8210
8211
error = spa_open(checkpoint_pool, &checkpoint_spa, FTAG);
8212
if (error != 0) {
8213
fatal("Tried to open pool \"%s\" but spa_open() failed with "
8214
"error %d\n", checkpoint_pool, error);
8215
}
8216
8217
/*
8218
* Ensure that ranges in the checkpoint space maps of each vdev
8219
* are allocated according to the checkpointed state's metaslab
8220
* space maps.
8221
*/
8222
verify_checkpoint_vdev_spacemaps(checkpoint_spa, spa);
8223
8224
/*
8225
* Ensure that allocated ranges in the checkpoint's metaslab
8226
* space maps remain allocated in the metaslab space maps of
8227
* the current state.
8228
*/
8229
verify_checkpoint_ms_spacemaps(checkpoint_spa, spa);
8230
8231
/*
8232
* Once we are done, we get rid of the checkpointed state.
8233
*/
8234
spa_close(checkpoint_spa, FTAG);
8235
free(checkpoint_pool);
8236
}
8237
8238
static void
8239
dump_leftover_checkpoint_blocks(spa_t *spa)
8240
{
8241
vdev_t *rvd = spa->spa_root_vdev;
8242
8243
for (uint64_t i = 0; i < rvd->vdev_children; i++) {
8244
vdev_t *vd = rvd->vdev_child[i];
8245
8246
space_map_t *checkpoint_sm = NULL;
8247
uint64_t checkpoint_sm_obj;
8248
8249
if (vd->vdev_top_zap == 0)
8250
continue;
8251
8252
if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
8253
VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
8254
continue;
8255
8256
VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
8257
VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
8258
sizeof (uint64_t), 1, &checkpoint_sm_obj));
8259
8260
VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
8261
checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
8262
dump_spacemap(spa->spa_meta_objset, checkpoint_sm);
8263
space_map_close(checkpoint_sm);
8264
}
8265
}
8266
8267
static int
8268
verify_checkpoint(spa_t *spa)
8269
{
8270
uberblock_t checkpoint;
8271
int error;
8272
8273
if (!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT))
8274
return (0);
8275
8276
error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8277
DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
8278
sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
8279
8280
if (error == ENOENT && !dump_opt['L']) {
8281
/*
8282
* If the feature is active but the uberblock is missing
8283
* then we must be in the middle of discarding the
8284
* checkpoint.
8285
*/
8286
(void) printf("\nPartially discarded checkpoint "
8287
"state found:\n");
8288
if (dump_opt['m'] > 3)
8289
dump_leftover_checkpoint_blocks(spa);
8290
return (0);
8291
} else if (error != 0) {
8292
(void) printf("lookup error %d when looking for "
8293
"checkpointed uberblock in MOS\n", error);
8294
return (error);
8295
}
8296
dump_uberblock(&checkpoint, "\nCheckpointed uberblock found:\n", "\n");
8297
8298
if (checkpoint.ub_checkpoint_txg == 0) {
8299
(void) printf("\nub_checkpoint_txg not set in checkpointed "
8300
"uberblock\n");
8301
error = 3;
8302
}
8303
8304
if (error == 0 && !dump_opt['L'])
8305
verify_checkpoint_blocks(spa);
8306
8307
return (error);
8308
}
8309
8310
static void
8311
mos_leaks_cb(void *arg, uint64_t start, uint64_t size)
8312
{
8313
(void) arg;
8314
for (uint64_t i = start; i < size; i++) {
8315
(void) printf("MOS object %llu referenced but not allocated\n",
8316
(u_longlong_t)i);
8317
}
8318
}
8319
8320
static void
8321
mos_obj_refd(uint64_t obj)
8322
{
8323
if (obj != 0 && mos_refd_objs != NULL)
8324
zfs_range_tree_add(mos_refd_objs, obj, 1);
8325
}
8326
8327
/*
8328
* Call on a MOS object that may already have been referenced.
8329
*/
8330
static void
8331
mos_obj_refd_multiple(uint64_t obj)
8332
{
8333
if (obj != 0 && mos_refd_objs != NULL &&
8334
!zfs_range_tree_contains(mos_refd_objs, obj, 1))
8335
zfs_range_tree_add(mos_refd_objs, obj, 1);
8336
}
8337
8338
static void
8339
mos_leak_vdev_top_zap(vdev_t *vd)
8340
{
8341
uint64_t ms_flush_data_obj;
8342
int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
8343
vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
8344
sizeof (ms_flush_data_obj), 1, &ms_flush_data_obj);
8345
if (error == ENOENT)
8346
return;
8347
ASSERT0(error);
8348
8349
mos_obj_refd(ms_flush_data_obj);
8350
}
8351
8352
static void
8353
mos_leak_vdev(vdev_t *vd)
8354
{
8355
mos_obj_refd(vd->vdev_dtl_object);
8356
mos_obj_refd(vd->vdev_ms_array);
8357
mos_obj_refd(vd->vdev_indirect_config.vic_births_object);
8358
mos_obj_refd(vd->vdev_indirect_config.vic_mapping_object);
8359
mos_obj_refd(vd->vdev_leaf_zap);
8360
if (vd->vdev_checkpoint_sm != NULL)
8361
mos_obj_refd(vd->vdev_checkpoint_sm->sm_object);
8362
if (vd->vdev_indirect_mapping != NULL) {
8363
mos_obj_refd(vd->vdev_indirect_mapping->
8364
vim_phys->vimp_counts_object);
8365
}
8366
if (vd->vdev_obsolete_sm != NULL)
8367
mos_obj_refd(vd->vdev_obsolete_sm->sm_object);
8368
8369
for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
8370
metaslab_t *ms = vd->vdev_ms[m];
8371
mos_obj_refd(space_map_object(ms->ms_sm));
8372
}
8373
8374
if (vd->vdev_root_zap != 0)
8375
mos_obj_refd(vd->vdev_root_zap);
8376
8377
if (vd->vdev_top_zap != 0) {
8378
mos_obj_refd(vd->vdev_top_zap);
8379
mos_leak_vdev_top_zap(vd);
8380
}
8381
8382
for (uint64_t c = 0; c < vd->vdev_children; c++) {
8383
mos_leak_vdev(vd->vdev_child[c]);
8384
}
8385
}
8386
8387
static void
8388
mos_leak_log_spacemaps(spa_t *spa)
8389
{
8390
uint64_t spacemap_zap;
8391
int error = zap_lookup(spa_meta_objset(spa),
8392
DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LOG_SPACEMAP_ZAP,
8393
sizeof (spacemap_zap), 1, &spacemap_zap);
8394
if (error == ENOENT)
8395
return;
8396
ASSERT0(error);
8397
8398
mos_obj_refd(spacemap_zap);
8399
for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
8400
sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls))
8401
mos_obj_refd(sls->sls_sm_obj);
8402
}
8403
8404
static void
8405
errorlog_count_refd(objset_t *mos, uint64_t errlog)
8406
{
8407
zap_cursor_t zc;
8408
zap_attribute_t *za = zap_attribute_alloc();
8409
for (zap_cursor_init(&zc, mos, errlog);
8410
zap_cursor_retrieve(&zc, za) == 0;
8411
zap_cursor_advance(&zc)) {
8412
mos_obj_refd(za->za_first_integer);
8413
}
8414
zap_cursor_fini(&zc);
8415
zap_attribute_free(za);
8416
}
8417
8418
static int
8419
dump_mos_leaks(spa_t *spa)
8420
{
8421
int rv = 0;
8422
objset_t *mos = spa->spa_meta_objset;
8423
dsl_pool_t *dp = spa->spa_dsl_pool;
8424
8425
/* Visit and mark all referenced objects in the MOS */
8426
8427
mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT);
8428
mos_obj_refd(spa->spa_pool_props_object);
8429
mos_obj_refd(spa->spa_config_object);
8430
mos_obj_refd(spa->spa_ddt_stat_object);
8431
mos_obj_refd(spa->spa_feat_desc_obj);
8432
mos_obj_refd(spa->spa_feat_enabled_txg_obj);
8433
mos_obj_refd(spa->spa_feat_for_read_obj);
8434
mos_obj_refd(spa->spa_feat_for_write_obj);
8435
mos_obj_refd(spa->spa_history);
8436
mos_obj_refd(spa->spa_errlog_last);
8437
mos_obj_refd(spa->spa_errlog_scrub);
8438
8439
if (spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) {
8440
errorlog_count_refd(mos, spa->spa_errlog_last);
8441
errorlog_count_refd(mos, spa->spa_errlog_scrub);
8442
}
8443
8444
mos_obj_refd(spa->spa_all_vdev_zaps);
8445
mos_obj_refd(spa->spa_dsl_pool->dp_bptree_obj);
8446
mos_obj_refd(spa->spa_dsl_pool->dp_tmp_userrefs_obj);
8447
mos_obj_refd(spa->spa_dsl_pool->dp_scan->scn_phys.scn_queue_obj);
8448
bpobj_count_refd(&spa->spa_deferred_bpobj);
8449
mos_obj_refd(dp->dp_empty_bpobj);
8450
bpobj_count_refd(&dp->dp_obsolete_bpobj);
8451
bpobj_count_refd(&dp->dp_free_bpobj);
8452
mos_obj_refd(spa->spa_l2cache.sav_object);
8453
mos_obj_refd(spa->spa_spares.sav_object);
8454
8455
if (spa->spa_syncing_log_sm != NULL)
8456
mos_obj_refd(spa->spa_syncing_log_sm->sm_object);
8457
mos_leak_log_spacemaps(spa);
8458
8459
mos_obj_refd(spa->spa_condensing_indirect_phys.
8460
scip_next_mapping_object);
8461
mos_obj_refd(spa->spa_condensing_indirect_phys.
8462
scip_prev_obsolete_sm_object);
8463
if (spa->spa_condensing_indirect_phys.scip_next_mapping_object != 0) {
8464
vdev_indirect_mapping_t *vim =
8465
vdev_indirect_mapping_open(mos,
8466
spa->spa_condensing_indirect_phys.scip_next_mapping_object);
8467
mos_obj_refd(vim->vim_phys->vimp_counts_object);
8468
vdev_indirect_mapping_close(vim);
8469
}
8470
deleted_livelists_dump_mos(spa);
8471
8472
if (dp->dp_origin_snap != NULL) {
8473
dsl_dataset_t *ds;
8474
8475
dsl_pool_config_enter(dp, FTAG);
8476
VERIFY0(dsl_dataset_hold_obj(dp,
8477
dsl_dataset_phys(dp->dp_origin_snap)->ds_next_snap_obj,
8478
FTAG, &ds));
8479
count_ds_mos_objects(ds);
8480
dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
8481
dsl_dataset_rele(ds, FTAG);
8482
dsl_pool_config_exit(dp, FTAG);
8483
8484
count_ds_mos_objects(dp->dp_origin_snap);
8485
dump_blkptr_list(&dp->dp_origin_snap->ds_deadlist, "Deadlist");
8486
}
8487
count_dir_mos_objects(dp->dp_mos_dir);
8488
if (dp->dp_free_dir != NULL)
8489
count_dir_mos_objects(dp->dp_free_dir);
8490
if (dp->dp_leak_dir != NULL)
8491
count_dir_mos_objects(dp->dp_leak_dir);
8492
8493
mos_leak_vdev(spa->spa_root_vdev);
8494
8495
for (uint64_t c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
8496
ddt_t *ddt = spa->spa_ddt[c];
8497
if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
8498
continue;
8499
8500
/* DDT store objects */
8501
for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
8502
for (ddt_class_t class = 0; class < DDT_CLASSES;
8503
class++) {
8504
mos_obj_refd(ddt->ddt_object[type][class]);
8505
}
8506
}
8507
8508
/* FDT container */
8509
if (ddt->ddt_version == DDT_VERSION_FDT)
8510
mos_obj_refd(ddt->ddt_dir_object);
8511
8512
/* FDT log objects */
8513
if (ddt->ddt_flags & DDT_FLAG_LOG) {
8514
mos_obj_refd(ddt->ddt_log[0].ddl_object);
8515
mos_obj_refd(ddt->ddt_log[1].ddl_object);
8516
}
8517
}
8518
8519
for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
8520
brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
8521
if (brtvd->bv_initiated) {
8522
mos_obj_refd(brtvd->bv_mos_brtvdev);
8523
mos_obj_refd(brtvd->bv_mos_entries);
8524
}
8525
}
8526
8527
/*
8528
* Visit all allocated objects and make sure they are referenced.
8529
*/
8530
uint64_t object = 0;
8531
while (dmu_object_next(mos, &object, B_FALSE, 0) == 0) {
8532
if (zfs_range_tree_contains(mos_refd_objs, object, 1)) {
8533
zfs_range_tree_remove(mos_refd_objs, object, 1);
8534
} else {
8535
dmu_object_info_t doi;
8536
const char *name;
8537
VERIFY0(dmu_object_info(mos, object, &doi));
8538
if (doi.doi_type & DMU_OT_NEWTYPE) {
8539
dmu_object_byteswap_t bswap =
8540
DMU_OT_BYTESWAP(doi.doi_type);
8541
name = dmu_ot_byteswap[bswap].ob_name;
8542
} else {
8543
name = dmu_ot[doi.doi_type].ot_name;
8544
}
8545
8546
(void) printf("MOS object %llu (%s) leaked\n",
8547
(u_longlong_t)object, name);
8548
rv = 2;
8549
}
8550
}
8551
(void) zfs_range_tree_walk(mos_refd_objs, mos_leaks_cb, NULL);
8552
if (!zfs_range_tree_is_empty(mos_refd_objs))
8553
rv = 2;
8554
zfs_range_tree_vacate(mos_refd_objs, NULL, NULL);
8555
zfs_range_tree_destroy(mos_refd_objs);
8556
return (rv);
8557
}
8558
8559
typedef struct log_sm_obsolete_stats_arg {
8560
uint64_t lsos_current_txg;
8561
8562
uint64_t lsos_total_entries;
8563
uint64_t lsos_valid_entries;
8564
8565
uint64_t lsos_sm_entries;
8566
uint64_t lsos_valid_sm_entries;
8567
} log_sm_obsolete_stats_arg_t;
8568
8569
static int
8570
log_spacemap_obsolete_stats_cb(spa_t *spa, space_map_entry_t *sme,
8571
uint64_t txg, void *arg)
8572
{
8573
log_sm_obsolete_stats_arg_t *lsos = arg;
8574
8575
uint64_t offset = sme->sme_offset;
8576
uint64_t vdev_id = sme->sme_vdev;
8577
8578
if (lsos->lsos_current_txg == 0) {
8579
/* this is the first log */
8580
lsos->lsos_current_txg = txg;
8581
} else if (lsos->lsos_current_txg < txg) {
8582
/* we just changed log - print stats and reset */
8583
(void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8584
(u_longlong_t)lsos->lsos_valid_sm_entries,
8585
(u_longlong_t)lsos->lsos_sm_entries,
8586
(u_longlong_t)lsos->lsos_current_txg);
8587
lsos->lsos_valid_sm_entries = 0;
8588
lsos->lsos_sm_entries = 0;
8589
lsos->lsos_current_txg = txg;
8590
}
8591
ASSERT3U(lsos->lsos_current_txg, ==, txg);
8592
8593
lsos->lsos_sm_entries++;
8594
lsos->lsos_total_entries++;
8595
8596
vdev_t *vd = vdev_lookup_top(spa, vdev_id);
8597
if (!vdev_is_concrete(vd))
8598
return (0);
8599
8600
metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
8601
ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
8602
8603
if (txg < metaslab_unflushed_txg(ms))
8604
return (0);
8605
lsos->lsos_valid_sm_entries++;
8606
lsos->lsos_valid_entries++;
8607
return (0);
8608
}
8609
8610
static void
8611
dump_log_spacemap_obsolete_stats(spa_t *spa)
8612
{
8613
if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
8614
return;
8615
8616
log_sm_obsolete_stats_arg_t lsos = {0};
8617
8618
(void) printf("Log Space Map Obsolete Entry Statistics:\n");
8619
8620
iterate_through_spacemap_logs(spa,
8621
log_spacemap_obsolete_stats_cb, &lsos);
8622
8623
/* print stats for latest log */
8624
(void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8625
(u_longlong_t)lsos.lsos_valid_sm_entries,
8626
(u_longlong_t)lsos.lsos_sm_entries,
8627
(u_longlong_t)lsos.lsos_current_txg);
8628
8629
(void) printf("%-8llu valid entries out of %-8llu - total\n\n",
8630
(u_longlong_t)lsos.lsos_valid_entries,
8631
(u_longlong_t)lsos.lsos_total_entries);
8632
}
8633
8634
static void
8635
dump_zpool(spa_t *spa)
8636
{
8637
dsl_pool_t *dp = spa_get_dsl(spa);
8638
int rc = 0;
8639
8640
if (dump_opt['y']) {
8641
livelist_metaslab_validate(spa);
8642
}
8643
8644
if (dump_opt['S']) {
8645
dump_simulated_ddt(spa);
8646
return;
8647
}
8648
8649
if (!dump_opt['e'] && dump_opt['C'] > 1) {
8650
(void) printf("\nCached configuration:\n");
8651
dump_nvlist(spa->spa_config, 8);
8652
}
8653
8654
if (dump_opt['C'])
8655
dump_config(spa);
8656
8657
if (dump_opt['u'])
8658
dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n");
8659
8660
if (dump_opt['D'])
8661
dump_all_ddts(spa);
8662
8663
if (dump_opt['T'])
8664
dump_brt(spa);
8665
8666
if (dump_opt['d'] > 2 || dump_opt['m'])
8667
dump_metaslabs(spa);
8668
if (dump_opt['M'])
8669
dump_metaslab_groups(spa, dump_opt['M'] > 1);
8670
if (dump_opt['d'] > 2 || dump_opt['m']) {
8671
dump_log_spacemaps(spa);
8672
dump_log_spacemap_obsolete_stats(spa);
8673
}
8674
8675
if (dump_opt['d'] || dump_opt['i']) {
8676
spa_feature_t f;
8677
mos_refd_objs = zfs_range_tree_create_flags(
8678
NULL, ZFS_RANGE_SEG64, NULL, 0, 0,
8679
0, "dump_zpool:mos_refd_objs");
8680
dump_objset(dp->dp_meta_objset);
8681
8682
if (dump_opt['d'] >= 3) {
8683
dsl_pool_t *dp = spa->spa_dsl_pool;
8684
dump_full_bpobj(&spa->spa_deferred_bpobj,
8685
"Deferred frees", 0);
8686
if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
8687
dump_full_bpobj(&dp->dp_free_bpobj,
8688
"Pool snapshot frees", 0);
8689
}
8690
if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
8691
ASSERT(spa_feature_is_enabled(spa,
8692
SPA_FEATURE_DEVICE_REMOVAL));
8693
dump_full_bpobj(&dp->dp_obsolete_bpobj,
8694
"Pool obsolete blocks", 0);
8695
}
8696
8697
if (spa_feature_is_active(spa,
8698
SPA_FEATURE_ASYNC_DESTROY)) {
8699
dump_bptree(spa->spa_meta_objset,
8700
dp->dp_bptree_obj,
8701
"Pool dataset frees");
8702
}
8703
dump_dtl(spa->spa_root_vdev, 0);
8704
}
8705
8706
for (spa_feature_t f = 0; f < SPA_FEATURES; f++)
8707
global_feature_count[f] = UINT64_MAX;
8708
global_feature_count[SPA_FEATURE_REDACTION_BOOKMARKS] = 0;
8709
global_feature_count[SPA_FEATURE_REDACTION_LIST_SPILL] = 0;
8710
global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN] = 0;
8711
global_feature_count[SPA_FEATURE_LIVELIST] = 0;
8712
8713
(void) dmu_objset_find(spa_name(spa), dump_one_objset,
8714
NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
8715
8716
if (rc == 0 && !dump_opt['L'])
8717
rc = dump_mos_leaks(spa);
8718
8719
for (f = 0; f < SPA_FEATURES; f++) {
8720
uint64_t refcount;
8721
8722
uint64_t *arr;
8723
if (!(spa_feature_table[f].fi_flags &
8724
ZFEATURE_FLAG_PER_DATASET)) {
8725
if (global_feature_count[f] == UINT64_MAX)
8726
continue;
8727
if (!spa_feature_is_enabled(spa, f)) {
8728
ASSERT0(global_feature_count[f]);
8729
continue;
8730
}
8731
arr = global_feature_count;
8732
} else {
8733
if (!spa_feature_is_enabled(spa, f)) {
8734
ASSERT0(dataset_feature_count[f]);
8735
continue;
8736
}
8737
arr = dataset_feature_count;
8738
}
8739
if (feature_get_refcount(spa, &spa_feature_table[f],
8740
&refcount) == ENOTSUP)
8741
continue;
8742
if (arr[f] != refcount) {
8743
(void) printf("%s feature refcount mismatch: "
8744
"%lld consumers != %lld refcount\n",
8745
spa_feature_table[f].fi_uname,
8746
(longlong_t)arr[f], (longlong_t)refcount);
8747
rc = 2;
8748
} else {
8749
(void) printf("Verified %s feature refcount "
8750
"of %llu is correct\n",
8751
spa_feature_table[f].fi_uname,
8752
(longlong_t)refcount);
8753
}
8754
}
8755
8756
if (rc == 0)
8757
rc = verify_device_removal_feature_counts(spa);
8758
}
8759
8760
if (rc == 0 && (dump_opt['b'] || dump_opt['c']))
8761
rc = dump_block_stats(spa);
8762
8763
if (rc == 0)
8764
rc = verify_spacemap_refcounts(spa);
8765
8766
if (dump_opt['s'])
8767
show_pool_stats(spa);
8768
8769
if (dump_opt['h'])
8770
dump_history(spa);
8771
8772
if (rc == 0)
8773
rc = verify_checkpoint(spa);
8774
8775
if (rc != 0) {
8776
dump_debug_buffer();
8777
zdb_exit(rc);
8778
}
8779
}
8780
8781
#define ZDB_FLAG_CHECKSUM 0x0001
8782
#define ZDB_FLAG_DECOMPRESS 0x0002
8783
#define ZDB_FLAG_BSWAP 0x0004
8784
#define ZDB_FLAG_GBH 0x0008
8785
#define ZDB_FLAG_INDIRECT 0x0010
8786
#define ZDB_FLAG_RAW 0x0020
8787
#define ZDB_FLAG_PRINT_BLKPTR 0x0040
8788
#define ZDB_FLAG_VERBOSE 0x0080
8789
8790
static int flagbits[256];
8791
static char flagbitstr[16];
8792
8793
static void
8794
zdb_print_blkptr(const blkptr_t *bp, int flags)
8795
{
8796
char blkbuf[BP_SPRINTF_LEN];
8797
8798
if (flags & ZDB_FLAG_BSWAP)
8799
byteswap_uint64_array((void *)bp, sizeof (blkptr_t));
8800
8801
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
8802
(void) printf("%s\n", blkbuf);
8803
}
8804
8805
static void
8806
zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
8807
{
8808
int i;
8809
8810
for (i = 0; i < nbps; i++)
8811
zdb_print_blkptr(&bp[i], flags);
8812
}
8813
8814
static void
8815
zdb_dump_gbh(void *buf, uint64_t size, int flags)
8816
{
8817
zdb_dump_indirect((blkptr_t *)buf, gbh_nblkptrs(size), flags);
8818
}
8819
8820
static void
8821
zdb_dump_block_raw(void *buf, uint64_t size, int flags)
8822
{
8823
if (flags & ZDB_FLAG_BSWAP)
8824
byteswap_uint64_array(buf, size);
8825
VERIFY(write(fileno(stdout), buf, size) == size);
8826
}
8827
8828
static void
8829
zdb_dump_block(char *label, void *buf, uint64_t size, int flags)
8830
{
8831
uint64_t *d = (uint64_t *)buf;
8832
unsigned nwords = size / sizeof (uint64_t);
8833
int do_bswap = !!(flags & ZDB_FLAG_BSWAP);
8834
unsigned i, j;
8835
const char *hdr;
8836
char *c;
8837
8838
8839
if (do_bswap)
8840
hdr = " 7 6 5 4 3 2 1 0 f e d c b a 9 8";
8841
else
8842
hdr = " 0 1 2 3 4 5 6 7 8 9 a b c d e f";
8843
8844
(void) printf("\n%s\n%6s %s 0123456789abcdef\n", label, "", hdr);
8845
8846
#ifdef _ZFS_LITTLE_ENDIAN
8847
/* correct the endianness */
8848
do_bswap = !do_bswap;
8849
#endif
8850
for (i = 0; i < nwords; i += 2) {
8851
(void) printf("%06llx: %016llx %016llx ",
8852
(u_longlong_t)(i * sizeof (uint64_t)),
8853
(u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]),
8854
(u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1]));
8855
8856
c = (char *)&d[i];
8857
for (j = 0; j < 2 * sizeof (uint64_t); j++)
8858
(void) printf("%c", isprint(c[j]) ? c[j] : '.');
8859
(void) printf("\n");
8860
}
8861
}
8862
8863
/*
8864
* There are two acceptable formats:
8865
* leaf_name - For example: c1t0d0 or /tmp/ztest.0a
8866
* child[.child]* - For example: 0.1.1
8867
*
8868
* The second form can be used to specify arbitrary vdevs anywhere
8869
* in the hierarchy. For example, in a pool with a mirror of
8870
* RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
8871
*/
8872
static vdev_t *
8873
zdb_vdev_lookup(vdev_t *vdev, const char *path)
8874
{
8875
char *s, *p, *q;
8876
unsigned i;
8877
8878
if (vdev == NULL)
8879
return (NULL);
8880
8881
/* First, assume the x.x.x.x format */
8882
i = strtoul(path, &s, 10);
8883
if (s == path || (s && *s != '.' && *s != '\0'))
8884
goto name;
8885
if (i >= vdev->vdev_children)
8886
return (NULL);
8887
8888
vdev = vdev->vdev_child[i];
8889
if (s && *s == '\0')
8890
return (vdev);
8891
return (zdb_vdev_lookup(vdev, s+1));
8892
8893
name:
8894
for (i = 0; i < vdev->vdev_children; i++) {
8895
vdev_t *vc = vdev->vdev_child[i];
8896
8897
if (vc->vdev_path == NULL) {
8898
vc = zdb_vdev_lookup(vc, path);
8899
if (vc == NULL)
8900
continue;
8901
else
8902
return (vc);
8903
}
8904
8905
p = strrchr(vc->vdev_path, '/');
8906
p = p ? p + 1 : vc->vdev_path;
8907
q = &vc->vdev_path[strlen(vc->vdev_path) - 2];
8908
8909
if (strcmp(vc->vdev_path, path) == 0)
8910
return (vc);
8911
if (strcmp(p, path) == 0)
8912
return (vc);
8913
if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0)
8914
return (vc);
8915
}
8916
8917
return (NULL);
8918
}
8919
8920
static int
8921
name_from_objset_id(spa_t *spa, uint64_t objset_id, char *outstr)
8922
{
8923
dsl_dataset_t *ds;
8924
8925
dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
8926
int error = dsl_dataset_hold_obj(spa->spa_dsl_pool, objset_id,
8927
NULL, &ds);
8928
if (error != 0) {
8929
(void) fprintf(stderr, "failed to hold objset %llu: %s\n",
8930
(u_longlong_t)objset_id, strerror(error));
8931
dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8932
return (error);
8933
}
8934
dsl_dataset_name(ds, outstr);
8935
dsl_dataset_rele(ds, NULL);
8936
dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8937
return (0);
8938
}
8939
8940
static boolean_t
8941
zdb_parse_block_sizes(char *sizes, uint64_t *lsize, uint64_t *psize)
8942
{
8943
char *s0, *s1, *tmp = NULL;
8944
8945
if (sizes == NULL)
8946
return (B_FALSE);
8947
8948
s0 = strtok_r(sizes, "/", &tmp);
8949
if (s0 == NULL)
8950
return (B_FALSE);
8951
s1 = strtok_r(NULL, "/", &tmp);
8952
*lsize = strtoull(s0, NULL, 16);
8953
*psize = s1 ? strtoull(s1, NULL, 16) : *lsize;
8954
return (*lsize >= *psize && *psize > 0);
8955
}
8956
8957
#define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg))
8958
8959
static boolean_t
8960
try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize,
8961
int flags, int cfunc, void *lbuf, void *lbuf2)
8962
{
8963
if (flags & ZDB_FLAG_VERBOSE) {
8964
(void) fprintf(stderr,
8965
"Trying %05llx -> %05llx (%s)\n",
8966
(u_longlong_t)psize,
8967
(u_longlong_t)lsize,
8968
zio_compress_table[cfunc].ci_name);
8969
}
8970
8971
/*
8972
* We set lbuf to all zeros and lbuf2 to all
8973
* ones, then decompress to both buffers and
8974
* compare their contents. This way we can
8975
* know if decompression filled exactly to
8976
* lsize or if it left some bytes unwritten.
8977
*/
8978
8979
memset(lbuf, 0x00, lsize);
8980
memset(lbuf2, 0xff, lsize);
8981
8982
abd_t labd, labd2;
8983
abd_get_from_buf_struct(&labd, lbuf, lsize);
8984
abd_get_from_buf_struct(&labd2, lbuf2, lsize);
8985
8986
boolean_t ret = B_FALSE;
8987
if (zio_decompress_data(cfunc, pabd,
8988
&labd, psize, lsize, NULL) == 0 &&
8989
zio_decompress_data(cfunc, pabd,
8990
&labd2, psize, lsize, NULL) == 0 &&
8991
memcmp(lbuf, lbuf2, lsize) == 0)
8992
ret = B_TRUE;
8993
8994
abd_free(&labd2);
8995
abd_free(&labd);
8996
8997
return (ret);
8998
}
8999
9000
static uint64_t
9001
zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
9002
uint64_t psize, int flags)
9003
{
9004
(void) buf;
9005
uint64_t orig_lsize = lsize;
9006
boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL));
9007
/*
9008
* We don't know how the data was compressed, so just try
9009
* every decompress function at every inflated blocksize.
9010
*/
9011
void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
9012
int cfuncs[ZIO_COMPRESS_FUNCTIONS] = { 0 };
9013
int *cfuncp = cfuncs;
9014
uint64_t maxlsize = SPA_MAXBLOCKSIZE;
9015
uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) |
9016
ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) |
9017
ZIO_COMPRESS_MASK(ZLE);
9018
*cfuncp++ = ZIO_COMPRESS_LZ4;
9019
*cfuncp++ = ZIO_COMPRESS_LZJB;
9020
mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB);
9021
/*
9022
* Every gzip level has the same decompressor, no need to
9023
* run it 9 times per bruteforce attempt.
9024
*/
9025
mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3);
9026
mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5);
9027
mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7);
9028
mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9);
9029
for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++)
9030
if (((1ULL << c) & mask) == 0)
9031
*cfuncp++ = c;
9032
9033
/*
9034
* On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
9035
* could take a while and we should let the user know
9036
* we are not stuck. On the other hand, printing progress
9037
* info gets old after a while. User can specify 'v' flag
9038
* to see the progression.
9039
*/
9040
if (lsize == psize)
9041
lsize += SPA_MINBLOCKSIZE;
9042
else
9043
maxlsize = lsize;
9044
9045
for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) {
9046
for (cfuncp = cfuncs; *cfuncp; cfuncp++) {
9047
if (try_decompress_block(pabd, lsize, psize, flags,
9048
*cfuncp, lbuf, lbuf2)) {
9049
tryzle = B_FALSE;
9050
break;
9051
}
9052
}
9053
if (*cfuncp != 0)
9054
break;
9055
}
9056
if (tryzle) {
9057
for (lsize = orig_lsize; lsize <= maxlsize;
9058
lsize += SPA_MINBLOCKSIZE) {
9059
if (try_decompress_block(pabd, lsize, psize, flags,
9060
ZIO_COMPRESS_ZLE, lbuf, lbuf2)) {
9061
*cfuncp = ZIO_COMPRESS_ZLE;
9062
break;
9063
}
9064
}
9065
}
9066
umem_free(lbuf2, SPA_MAXBLOCKSIZE);
9067
9068
if (*cfuncp == ZIO_COMPRESS_ZLE) {
9069
printf("\nZLE decompression was selected. If you "
9070
"suspect the results are wrong,\ntry avoiding ZLE "
9071
"by setting and exporting ZDB_NO_ZLE=\"true\"\n");
9072
}
9073
9074
return (lsize > maxlsize ? -1 : lsize);
9075
}
9076
9077
/*
9078
* Read a block from a pool and print it out. The syntax of the
9079
* block descriptor is:
9080
*
9081
* pool:vdev_specifier:offset:[lsize/]psize[:flags]
9082
*
9083
* pool - The name of the pool you wish to read from
9084
* vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
9085
* offset - offset, in hex, in bytes
9086
* size - Amount of data to read, in hex, in bytes
9087
* flags - A string of characters specifying options
9088
* b: Decode a blkptr at given offset within block
9089
* c: Calculate and display checksums
9090
* d: Decompress data before dumping
9091
* e: Byteswap data before dumping
9092
* g: Display data as a gang block header
9093
* i: Display as an indirect block
9094
* r: Dump raw data to stdout
9095
* v: Verbose
9096
*
9097
*/
9098
static void
9099
zdb_read_block(char *thing, spa_t *spa)
9100
{
9101
blkptr_t blk, *bp = &blk;
9102
dva_t *dva = bp->blk_dva;
9103
int flags = 0;
9104
uint64_t offset = 0, psize = 0, lsize = 0, blkptr_offset = 0;
9105
zio_t *zio;
9106
vdev_t *vd;
9107
abd_t *pabd;
9108
void *lbuf, *buf;
9109
char *s, *p, *dup, *flagstr, *sizes, *tmp = NULL;
9110
const char *vdev, *errmsg = NULL;
9111
int i, len, error;
9112
boolean_t borrowed = B_FALSE, found = B_FALSE;
9113
9114
dup = strdup(thing);
9115
s = strtok_r(dup, ":", &tmp);
9116
vdev = s ?: "";
9117
s = strtok_r(NULL, ":", &tmp);
9118
offset = strtoull(s ? s : "", NULL, 16);
9119
sizes = strtok_r(NULL, ":", &tmp);
9120
s = strtok_r(NULL, ":", &tmp);
9121
flagstr = strdup(s ?: "");
9122
9123
if (!zdb_parse_block_sizes(sizes, &lsize, &psize))
9124
errmsg = "invalid size(s)";
9125
if (!IS_P2ALIGNED(psize, DEV_BSIZE) || !IS_P2ALIGNED(lsize, DEV_BSIZE))
9126
errmsg = "size must be a multiple of sector size";
9127
if (!IS_P2ALIGNED(offset, DEV_BSIZE))
9128
errmsg = "offset must be a multiple of sector size";
9129
if (errmsg) {
9130
(void) printf("Invalid block specifier: %s - %s\n",
9131
thing, errmsg);
9132
goto done;
9133
}
9134
9135
tmp = NULL;
9136
for (s = strtok_r(flagstr, ":", &tmp);
9137
s != NULL;
9138
s = strtok_r(NULL, ":", &tmp)) {
9139
len = strlen(flagstr);
9140
for (i = 0; i < len; i++) {
9141
int bit = flagbits[(uchar_t)flagstr[i]];
9142
9143
if (bit == 0) {
9144
(void) printf("***Ignoring flag: %c\n",
9145
(uchar_t)flagstr[i]);
9146
continue;
9147
}
9148
found = B_TRUE;
9149
flags |= bit;
9150
9151
p = &flagstr[i + 1];
9152
if (*p != ':' && *p != '\0') {
9153
int j = 0, nextbit = flagbits[(uchar_t)*p];
9154
char *end, offstr[8] = { 0 };
9155
if ((bit == ZDB_FLAG_PRINT_BLKPTR) &&
9156
(nextbit == 0)) {
9157
/* look ahead to isolate the offset */
9158
while (nextbit == 0 &&
9159
strchr(flagbitstr, *p) == NULL) {
9160
offstr[j] = *p;
9161
j++;
9162
if (i + j > strlen(flagstr))
9163
break;
9164
p++;
9165
nextbit = flagbits[(uchar_t)*p];
9166
}
9167
blkptr_offset = strtoull(offstr, &end,
9168
16);
9169
i += j;
9170
} else if (nextbit == 0) {
9171
(void) printf("***Ignoring flag arg:"
9172
" '%c'\n", (uchar_t)*p);
9173
}
9174
}
9175
}
9176
}
9177
if (blkptr_offset % sizeof (blkptr_t)) {
9178
printf("Block pointer offset 0x%llx "
9179
"must be divisible by 0x%x\n",
9180
(longlong_t)blkptr_offset, (int)sizeof (blkptr_t));
9181
goto done;
9182
}
9183
if (found == B_FALSE && strlen(flagstr) > 0) {
9184
printf("Invalid flag arg: '%s'\n", flagstr);
9185
goto done;
9186
}
9187
9188
vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev);
9189
if (vd == NULL) {
9190
(void) printf("***Invalid vdev: %s\n", vdev);
9191
goto done;
9192
} else {
9193
if (vd->vdev_path)
9194
(void) fprintf(stderr, "Found vdev: %s\n",
9195
vd->vdev_path);
9196
else
9197
(void) fprintf(stderr, "Found vdev type: %s\n",
9198
vd->vdev_ops->vdev_op_type);
9199
}
9200
9201
pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
9202
lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
9203
9204
BP_ZERO(bp);
9205
9206
DVA_SET_VDEV(&dva[0], vd->vdev_id);
9207
DVA_SET_OFFSET(&dva[0], offset);
9208
DVA_SET_GANG(&dva[0], 0);
9209
DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
9210
9211
BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
9212
9213
BP_SET_LSIZE(bp, lsize);
9214
BP_SET_PSIZE(bp, psize);
9215
BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF);
9216
BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF);
9217
BP_SET_TYPE(bp, DMU_OT_NONE);
9218
BP_SET_LEVEL(bp, 0);
9219
BP_SET_DEDUP(bp, 0);
9220
BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
9221
9222
spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9223
zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9224
9225
if (vd == vd->vdev_top) {
9226
/*
9227
* Treat this as a normal block read.
9228
*/
9229
zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL,
9230
ZIO_PRIORITY_SYNC_READ,
9231
ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL));
9232
} else {
9233
/*
9234
* Treat this as a vdev child I/O.
9235
*/
9236
zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd,
9237
psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ,
9238
ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
9239
ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | ZIO_FLAG_OPTIONAL,
9240
NULL, NULL));
9241
}
9242
9243
error = zio_wait(zio);
9244
spa_config_exit(spa, SCL_STATE, FTAG);
9245
9246
if (error) {
9247
(void) printf("Read of %s failed, error: %d\n", thing, error);
9248
goto out;
9249
}
9250
9251
uint64_t orig_lsize = lsize;
9252
buf = lbuf;
9253
if (flags & ZDB_FLAG_DECOMPRESS) {
9254
lsize = zdb_decompress_block(pabd, buf, lbuf,
9255
lsize, psize, flags);
9256
if (lsize == -1) {
9257
(void) printf("Decompress of %s failed\n", thing);
9258
goto out;
9259
}
9260
} else {
9261
buf = abd_borrow_buf_copy(pabd, lsize);
9262
borrowed = B_TRUE;
9263
}
9264
/*
9265
* Try to detect invalid block pointer. If invalid, try
9266
* decompressing.
9267
*/
9268
if ((flags & ZDB_FLAG_PRINT_BLKPTR || flags & ZDB_FLAG_INDIRECT) &&
9269
!(flags & ZDB_FLAG_DECOMPRESS)) {
9270
const blkptr_t *b = (const blkptr_t *)(void *)
9271
((uintptr_t)buf + (uintptr_t)blkptr_offset);
9272
if (zfs_blkptr_verify(spa, b,
9273
BLK_CONFIG_NEEDED, BLK_VERIFY_ONLY)) {
9274
abd_return_buf_copy(pabd, buf, lsize);
9275
borrowed = B_FALSE;
9276
buf = lbuf;
9277
lsize = zdb_decompress_block(pabd, buf,
9278
lbuf, lsize, psize, flags);
9279
b = (const blkptr_t *)(void *)
9280
((uintptr_t)buf + (uintptr_t)blkptr_offset);
9281
if (lsize == -1 || zfs_blkptr_verify(spa, b,
9282
BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) {
9283
printf("invalid block pointer at this DVA\n");
9284
goto out;
9285
}
9286
}
9287
}
9288
9289
if (flags & ZDB_FLAG_PRINT_BLKPTR)
9290
zdb_print_blkptr((blkptr_t *)(void *)
9291
((uintptr_t)buf + (uintptr_t)blkptr_offset), flags);
9292
else if (flags & ZDB_FLAG_RAW)
9293
zdb_dump_block_raw(buf, lsize, flags);
9294
else if (flags & ZDB_FLAG_INDIRECT)
9295
zdb_dump_indirect((blkptr_t *)buf,
9296
orig_lsize / sizeof (blkptr_t), flags);
9297
else if (flags & ZDB_FLAG_GBH)
9298
zdb_dump_gbh(buf, lsize, flags);
9299
else
9300
zdb_dump_block(thing, buf, lsize, flags);
9301
9302
/*
9303
* If :c was specified, iterate through the checksum table to
9304
* calculate and display each checksum for our specified
9305
* DVA and length.
9306
*/
9307
if ((flags & ZDB_FLAG_CHECKSUM) && !(flags & ZDB_FLAG_RAW) &&
9308
!(flags & ZDB_FLAG_GBH)) {
9309
zio_t *czio;
9310
(void) printf("\n");
9311
for (enum zio_checksum ck = ZIO_CHECKSUM_LABEL;
9312
ck < ZIO_CHECKSUM_FUNCTIONS; ck++) {
9313
9314
if ((zio_checksum_table[ck].ci_flags &
9315
ZCHECKSUM_FLAG_EMBEDDED) ||
9316
ck == ZIO_CHECKSUM_NOPARITY) {
9317
continue;
9318
}
9319
BP_SET_CHECKSUM(bp, ck);
9320
spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9321
czio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9322
if (vd == vd->vdev_top) {
9323
zio_nowait(zio_read(czio, spa, bp, pabd, psize,
9324
NULL, NULL,
9325
ZIO_PRIORITY_SYNC_READ,
9326
ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9327
ZIO_FLAG_DONT_RETRY, NULL));
9328
} else {
9329
zio_nowait(zio_vdev_child_io(czio, bp, vd,
9330
offset, pabd, psize, ZIO_TYPE_READ,
9331
ZIO_PRIORITY_SYNC_READ,
9332
ZIO_FLAG_DONT_PROPAGATE |
9333
ZIO_FLAG_DONT_RETRY |
9334
ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9335
ZIO_FLAG_SPECULATIVE |
9336
ZIO_FLAG_OPTIONAL, NULL, NULL));
9337
}
9338
error = zio_wait(czio);
9339
if (error == 0 || error == ECKSUM) {
9340
zio_t *ck_zio = zio_null(NULL, spa, NULL,
9341
NULL, NULL, 0);
9342
ck_zio->io_offset =
9343
DVA_GET_OFFSET(&bp->blk_dva[0]);
9344
ck_zio->io_bp = bp;
9345
zio_checksum_compute(ck_zio, ck, pabd, psize);
9346
printf(
9347
"%12s\t"
9348
"cksum=%016llx:%016llx:%016llx:%016llx\n",
9349
zio_checksum_table[ck].ci_name,
9350
(u_longlong_t)bp->blk_cksum.zc_word[0],
9351
(u_longlong_t)bp->blk_cksum.zc_word[1],
9352
(u_longlong_t)bp->blk_cksum.zc_word[2],
9353
(u_longlong_t)bp->blk_cksum.zc_word[3]);
9354
zio_wait(ck_zio);
9355
} else {
9356
printf("error %d reading block\n", error);
9357
}
9358
spa_config_exit(spa, SCL_STATE, FTAG);
9359
}
9360
}
9361
9362
if (borrowed)
9363
abd_return_buf_copy(pabd, buf, lsize);
9364
9365
out:
9366
abd_free(pabd);
9367
umem_free(lbuf, SPA_MAXBLOCKSIZE);
9368
done:
9369
free(flagstr);
9370
free(dup);
9371
}
9372
9373
static void
9374
zdb_embedded_block(char *thing)
9375
{
9376
blkptr_t bp = {{{{0}}}};
9377
unsigned long long *words = (void *)&bp;
9378
char *buf;
9379
int err;
9380
9381
err = sscanf(thing, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
9382
"%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
9383
words + 0, words + 1, words + 2, words + 3,
9384
words + 4, words + 5, words + 6, words + 7,
9385
words + 8, words + 9, words + 10, words + 11,
9386
words + 12, words + 13, words + 14, words + 15);
9387
if (err != 16) {
9388
(void) fprintf(stderr, "invalid input format\n");
9389
zdb_exit(1);
9390
}
9391
ASSERT3U(BPE_GET_LSIZE(&bp), <=, SPA_MAXBLOCKSIZE);
9392
buf = malloc(SPA_MAXBLOCKSIZE);
9393
if (buf == NULL) {
9394
(void) fprintf(stderr, "out of memory\n");
9395
zdb_exit(1);
9396
}
9397
err = decode_embedded_bp(&bp, buf, BPE_GET_LSIZE(&bp));
9398
if (err != 0) {
9399
(void) fprintf(stderr, "decode failed: %u\n", err);
9400
zdb_exit(1);
9401
}
9402
zdb_dump_block_raw(buf, BPE_GET_LSIZE(&bp), 0);
9403
free(buf);
9404
}
9405
9406
/* check for valid hex or decimal numeric string */
9407
static boolean_t
9408
zdb_numeric(char *str)
9409
{
9410
int i = 0, len;
9411
9412
len = strlen(str);
9413
if (len == 0)
9414
return (B_FALSE);
9415
if (strncmp(str, "0x", 2) == 0 || strncmp(str, "0X", 2) == 0)
9416
i = 2;
9417
for (; i < len; i++) {
9418
if (!isxdigit(str[i]))
9419
return (B_FALSE);
9420
}
9421
return (B_TRUE);
9422
}
9423
9424
static int
9425
dummy_get_file_info(dmu_object_type_t bonustype, const void *data,
9426
zfs_file_info_t *zoi)
9427
{
9428
(void) data, (void) zoi;
9429
9430
if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
9431
return (ENOENT);
9432
9433
(void) fprintf(stderr, "dummy_get_file_info: not implemented");
9434
abort();
9435
}
9436
9437
int
9438
main(int argc, char **argv)
9439
{
9440
int c;
9441
int dump_all = 1;
9442
int verbose = 0;
9443
int error = 0;
9444
char **searchdirs = NULL;
9445
int nsearch = 0;
9446
char *target, *target_pool, dsname[ZFS_MAX_DATASET_NAME_LEN];
9447
nvlist_t *policy = NULL;
9448
uint64_t max_txg = UINT64_MAX;
9449
int64_t objset_id = -1;
9450
uint64_t object;
9451
int flags = ZFS_IMPORT_MISSING_LOG;
9452
int rewind = ZPOOL_NEVER_REWIND;
9453
char *spa_config_path_env, *objset_str;
9454
boolean_t target_is_spa = B_TRUE, dataset_lookup = B_FALSE;
9455
nvlist_t *cfg = NULL;
9456
struct sigaction action;
9457
boolean_t force_import = B_FALSE;
9458
boolean_t config_path_console = B_FALSE;
9459
char pbuf[MAXPATHLEN];
9460
9461
dprintf_setup(&argc, argv);
9462
9463
/*
9464
* Set up signal handlers, so if we crash due to bad on-disk data we
9465
* can get more info. Unlike ztest, we don't bail out if we can't set
9466
* up signal handlers, because zdb is very useful without them.
9467
*/
9468
action.sa_handler = sig_handler;
9469
sigemptyset(&action.sa_mask);
9470
action.sa_flags = 0;
9471
if (sigaction(SIGSEGV, &action, NULL) < 0) {
9472
(void) fprintf(stderr, "zdb: cannot catch SIGSEGV: %s\n",
9473
strerror(errno));
9474
}
9475
if (sigaction(SIGABRT, &action, NULL) < 0) {
9476
(void) fprintf(stderr, "zdb: cannot catch SIGABRT: %s\n",
9477
strerror(errno));
9478
}
9479
9480
/*
9481
* If there is an environment variable SPA_CONFIG_PATH it overrides
9482
* default spa_config_path setting. If -U flag is specified it will
9483
* override this environment variable settings once again.
9484
*/
9485
spa_config_path_env = getenv("SPA_CONFIG_PATH");
9486
if (spa_config_path_env != NULL)
9487
spa_config_path = spa_config_path_env;
9488
9489
/*
9490
* For performance reasons, we set this tunable down. We do so before
9491
* the arg parsing section so that the user can override this value if
9492
* they choose.
9493
*/
9494
zfs_btree_verify_intensity = 3;
9495
9496
struct option long_options[] = {
9497
{"ignore-assertions", no_argument, NULL, 'A'},
9498
{"block-stats", no_argument, NULL, 'b'},
9499
{"backup", no_argument, NULL, 'B'},
9500
{"checksum", no_argument, NULL, 'c'},
9501
{"config", no_argument, NULL, 'C'},
9502
{"datasets", no_argument, NULL, 'd'},
9503
{"dedup-stats", no_argument, NULL, 'D'},
9504
{"exported", no_argument, NULL, 'e'},
9505
{"embedded-block-pointer", no_argument, NULL, 'E'},
9506
{"automatic-rewind", no_argument, NULL, 'F'},
9507
{"dump-debug-msg", no_argument, NULL, 'G'},
9508
{"history", no_argument, NULL, 'h'},
9509
{"intent-logs", no_argument, NULL, 'i'},
9510
{"inflight", required_argument, NULL, 'I'},
9511
{"checkpointed-state", no_argument, NULL, 'k'},
9512
{"key", required_argument, NULL, 'K'},
9513
{"label", no_argument, NULL, 'l'},
9514
{"disable-leak-tracking", no_argument, NULL, 'L'},
9515
{"metaslabs", no_argument, NULL, 'm'},
9516
{"metaslab-groups", no_argument, NULL, 'M'},
9517
{"numeric", no_argument, NULL, 'N'},
9518
{"option", required_argument, NULL, 'o'},
9519
{"object-lookups", no_argument, NULL, 'O'},
9520
{"path", required_argument, NULL, 'p'},
9521
{"parseable", no_argument, NULL, 'P'},
9522
{"skip-label", no_argument, NULL, 'q'},
9523
{"copy-object", no_argument, NULL, 'r'},
9524
{"read-block", no_argument, NULL, 'R'},
9525
{"io-stats", no_argument, NULL, 's'},
9526
{"simulate-dedup", no_argument, NULL, 'S'},
9527
{"txg", required_argument, NULL, 't'},
9528
{"brt-stats", no_argument, NULL, 'T'},
9529
{"uberblock", no_argument, NULL, 'u'},
9530
{"cachefile", required_argument, NULL, 'U'},
9531
{"verbose", no_argument, NULL, 'v'},
9532
{"verbatim", no_argument, NULL, 'V'},
9533
{"dump-blocks", required_argument, NULL, 'x'},
9534
{"extreme-rewind", no_argument, NULL, 'X'},
9535
{"all-reconstruction", no_argument, NULL, 'Y'},
9536
{"livelist", no_argument, NULL, 'y'},
9537
{"zstd-headers", no_argument, NULL, 'Z'},
9538
{"allocated-map", no_argument, NULL,
9539
ARG_ALLOCATED},
9540
{"bin", required_argument, NULL,
9541
ARG_BLOCK_BIN_MODE},
9542
{"class", required_argument, NULL,
9543
ARG_BLOCK_CLASSES},
9544
{0, 0, 0, 0}
9545
};
9546
9547
while ((c = getopt_long(argc, argv,
9548
"AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
9549
long_options, NULL)) != -1) {
9550
switch (c) {
9551
case 'b':
9552
case 'B':
9553
case 'c':
9554
case 'C':
9555
case 'd':
9556
case 'D':
9557
case 'E':
9558
case 'G':
9559
case 'h':
9560
case 'i':
9561
case 'l':
9562
case 'm':
9563
case 'M':
9564
case 'N':
9565
case 'O':
9566
case 'r':
9567
case 'R':
9568
case 's':
9569
case 'S':
9570
case 'T':
9571
case 'u':
9572
case 'y':
9573
case 'Z':
9574
case ARG_ALLOCATED:
9575
dump_opt[c]++;
9576
dump_all = 0;
9577
break;
9578
case 'A':
9579
case 'e':
9580
case 'F':
9581
case 'k':
9582
case 'L':
9583
case 'P':
9584
case 'q':
9585
case 'X':
9586
dump_opt[c]++;
9587
break;
9588
case 'Y':
9589
zfs_reconstruct_indirect_combinations_max = INT_MAX;
9590
zfs_deadman_enabled = 0;
9591
break;
9592
/* NB: Sort single match options below. */
9593
case 'I':
9594
max_inflight_bytes = strtoull(optarg, NULL, 0);
9595
if (max_inflight_bytes == 0) {
9596
(void) fprintf(stderr, "maximum number "
9597
"of inflight bytes must be greater "
9598
"than 0\n");
9599
usage();
9600
}
9601
break;
9602
case 'K':
9603
dump_opt[c]++;
9604
key_material = strdup(optarg);
9605
/* redact key material in process table */
9606
while (*optarg != '\0') { *optarg++ = '*'; }
9607
break;
9608
case 'o':
9609
dump_opt[c]++;
9610
dump_all = 0;
9611
error = handle_tunable_option(optarg, B_FALSE);
9612
if (error != 0)
9613
zdb_exit(1);
9614
break;
9615
case 'p':
9616
if (searchdirs == NULL) {
9617
searchdirs = umem_alloc(sizeof (char *),
9618
UMEM_NOFAIL);
9619
} else {
9620
char **tmp = umem_alloc((nsearch + 1) *
9621
sizeof (char *), UMEM_NOFAIL);
9622
memcpy(tmp, searchdirs, nsearch *
9623
sizeof (char *));
9624
umem_free(searchdirs,
9625
nsearch * sizeof (char *));
9626
searchdirs = tmp;
9627
}
9628
searchdirs[nsearch++] = optarg;
9629
break;
9630
case 't':
9631
max_txg = strtoull(optarg, NULL, 0);
9632
if (max_txg < TXG_INITIAL) {
9633
(void) fprintf(stderr, "incorrect txg "
9634
"specified: %s\n", optarg);
9635
usage();
9636
}
9637
break;
9638
case 'U':
9639
config_path_console = B_TRUE;
9640
spa_config_path = optarg;
9641
if (spa_config_path[0] != '/') {
9642
(void) fprintf(stderr,
9643
"cachefile must be an absolute path "
9644
"(i.e. start with a slash)\n");
9645
usage();
9646
}
9647
break;
9648
case 'v':
9649
verbose++;
9650
break;
9651
case 'V':
9652
flags = ZFS_IMPORT_VERBATIM;
9653
break;
9654
case 'x':
9655
vn_dumpdir = optarg;
9656
break;
9657
case ARG_BLOCK_BIN_MODE:
9658
if (strcmp(optarg, "lsize") == 0) {
9659
block_bin_mode = BIN_LSIZE;
9660
} else if (strcmp(optarg, "psize") == 0) {
9661
block_bin_mode = BIN_PSIZE;
9662
} else if (strcmp(optarg, "asize") == 0) {
9663
block_bin_mode = BIN_ASIZE;
9664
} else {
9665
(void) fprintf(stderr,
9666
"--bin=\"%s\" must be one of \"lsize\", "
9667
"\"psize\" or \"asize\"\n", optarg);
9668
usage();
9669
}
9670
break;
9671
9672
case ARG_BLOCK_CLASSES: {
9673
char *buf = strdup(optarg), *tok = buf, *next,
9674
*save = NULL;
9675
9676
while ((next = strtok_r(tok, ",", &save)) != NULL) {
9677
tok = NULL;
9678
9679
if (strcmp(next, "normal") == 0) {
9680
block_classes |= CLASS_NORMAL;
9681
} else if (strcmp(next, "special") == 0) {
9682
block_classes |= CLASS_SPECIAL;
9683
} else if (strcmp(next, "dedup") == 0) {
9684
block_classes |= CLASS_DEDUP;
9685
} else if (strcmp(next, "other") == 0) {
9686
block_classes |= CLASS_OTHER;
9687
} else {
9688
(void) fprintf(stderr,
9689
"--class=\"%s\" must be a "
9690
"comma-separated list of either "
9691
"\"normal\", \"special\", "
9692
"\"asize\" or \"other\"; "
9693
"got \"%s\"\n",
9694
optarg, next);
9695
usage();
9696
}
9697
}
9698
9699
if (block_classes == 0) {
9700
(void) fprintf(stderr,
9701
"--class= must be a comma-separated "
9702
"list of either \"normal\", \"special\", "
9703
"\"asize\" or \"other\"; got empty\n");
9704
usage();
9705
}
9706
9707
free(buf);
9708
break;
9709
}
9710
default:
9711
usage();
9712
break;
9713
}
9714
}
9715
9716
if (!dump_opt['e'] && searchdirs != NULL) {
9717
(void) fprintf(stderr, "-p option requires use of -e\n");
9718
usage();
9719
}
9720
#if defined(_LP64)
9721
/*
9722
* ZDB does not typically re-read blocks; therefore limit the ARC
9723
* to 256 MB, which can be used entirely for metadata.
9724
*/
9725
zfs_arc_min = 2ULL << SPA_MAXBLOCKSHIFT;
9726
zfs_arc_max = 256 * 1024 * 1024;
9727
#endif
9728
9729
/*
9730
* "zdb -c" uses checksum-verifying scrub i/os which are async reads.
9731
* "zdb -b" uses traversal prefetch which uses async reads.
9732
* For good performance, let several of them be active at once.
9733
*/
9734
zfs_vdev_async_read_max_active = 10;
9735
9736
/*
9737
* Disable reference tracking for better performance.
9738
*/
9739
reference_tracking_enable = B_FALSE;
9740
9741
/*
9742
* Do not fail spa_load when spa_load_verify fails. This is needed
9743
* to load non-idle pools.
9744
*/
9745
spa_load_verify_dryrun = B_TRUE;
9746
9747
/*
9748
* ZDB should have ability to read spacemaps.
9749
*/
9750
spa_mode_readable_spacemaps = B_TRUE;
9751
9752
libspl_set_assert_ok((dump_opt['A'] == 1) || (dump_opt['A'] > 2));
9753
zfs_recover = (dump_opt['A'] > 1);
9754
9755
if (dump_all)
9756
verbose = MAX(verbose, 1);
9757
9758
for (c = 0; c < 256; c++) {
9759
if (dump_all && strchr("ABeEFkKlLNOPrRSXy", c) == NULL)
9760
dump_opt[c] = 1;
9761
if (dump_opt[c])
9762
dump_opt[c] += verbose;
9763
}
9764
9765
argc -= optind;
9766
argv += optind;
9767
if (argc < 2 && dump_opt['R'])
9768
usage();
9769
9770
target = argv[0];
9771
9772
/*
9773
* Automate cachefile
9774
*/
9775
if (!spa_config_path_env && !config_path_console && target &&
9776
libzfs_core_init() == 0) {
9777
char *pname = strdup(target);
9778
const char *value;
9779
nvlist_t *pnvl = NULL;
9780
nvlist_t *vnvl = NULL;
9781
9782
if (strpbrk(pname, "/@") != NULL)
9783
*strpbrk(pname, "/@") = '\0';
9784
9785
if (pname && lzc_get_props(pname, &pnvl) == 0) {
9786
if (nvlist_lookup_nvlist(pnvl, "cachefile",
9787
&vnvl) == 0) {
9788
value = fnvlist_lookup_string(vnvl,
9789
ZPROP_VALUE);
9790
} else {
9791
value = "-";
9792
}
9793
strlcpy(pbuf, value, sizeof (pbuf));
9794
if (pbuf[0] != '\0') {
9795
if (pbuf[0] == '/') {
9796
if (access(pbuf, F_OK) == 0)
9797
spa_config_path = pbuf;
9798
else
9799
force_import = B_TRUE;
9800
} else if ((strcmp(pbuf, "-") == 0 &&
9801
access(ZPOOL_CACHE, F_OK) != 0) ||
9802
strcmp(pbuf, "none") == 0) {
9803
force_import = B_TRUE;
9804
}
9805
}
9806
nvlist_free(vnvl);
9807
}
9808
9809
free(pname);
9810
nvlist_free(pnvl);
9811
libzfs_core_fini();
9812
}
9813
9814
dmu_objset_register_type(DMU_OST_ZFS, dummy_get_file_info);
9815
kernel_init(SPA_MODE_READ);
9816
kernel_init_done = B_TRUE;
9817
9818
if (dump_opt['E']) {
9819
if (argc != 1)
9820
usage();
9821
zdb_embedded_block(argv[0]);
9822
error = 0;
9823
goto fini;
9824
}
9825
9826
if (argc < 1) {
9827
if (!dump_opt['e'] && dump_opt['C']) {
9828
dump_cachefile(spa_config_path);
9829
error = 0;
9830
goto fini;
9831
}
9832
if (dump_opt['o'])
9833
/*
9834
* Avoid blasting tunable options off the top of the
9835
* screen.
9836
*/
9837
zdb_exit(1);
9838
usage();
9839
}
9840
9841
if (dump_opt['l']) {
9842
error = dump_label(argv[0]);
9843
goto fini;
9844
}
9845
9846
if (dump_opt['X'] || dump_opt['F'])
9847
rewind = ZPOOL_DO_REWIND |
9848
(dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
9849
9850
/* -N implies -d */
9851
if (dump_opt['N'] && dump_opt['d'] == 0)
9852
dump_opt['d'] = dump_opt['N'];
9853
9854
if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 ||
9855
nvlist_add_uint64(policy, ZPOOL_LOAD_REQUEST_TXG, max_txg) != 0 ||
9856
nvlist_add_uint32(policy, ZPOOL_LOAD_REWIND_POLICY, rewind) != 0)
9857
fatal("internal error: %s", strerror(ENOMEM));
9858
9859
error = 0;
9860
9861
if (strpbrk(target, "/@") != NULL) {
9862
size_t targetlen;
9863
9864
target_pool = strdup(target);
9865
*strpbrk(target_pool, "/@") = '\0';
9866
9867
target_is_spa = B_FALSE;
9868
targetlen = strlen(target);
9869
if (targetlen && target[targetlen - 1] == '/')
9870
target[targetlen - 1] = '\0';
9871
9872
/*
9873
* See if an objset ID was supplied (-d <pool>/<objset ID>).
9874
* To disambiguate tank/100, consider the 100 as objsetID
9875
* if -N was given, otherwise 100 is an objsetID iff
9876
* tank/100 as a named dataset fails on lookup.
9877
*/
9878
objset_str = strchr(target, '/');
9879
if (objset_str && strlen(objset_str) > 1 &&
9880
zdb_numeric(objset_str + 1)) {
9881
char *endptr;
9882
errno = 0;
9883
objset_str++;
9884
objset_id = strtoull(objset_str, &endptr, 0);
9885
/* dataset 0 is the same as opening the pool */
9886
if (errno == 0 && endptr != objset_str &&
9887
objset_id != 0) {
9888
if (dump_opt['N'])
9889
dataset_lookup = B_TRUE;
9890
}
9891
/* normal dataset name not an objset ID */
9892
if (endptr == objset_str) {
9893
objset_id = -1;
9894
}
9895
} else if (objset_str && !zdb_numeric(objset_str + 1) &&
9896
dump_opt['N']) {
9897
printf("Supply a numeric objset ID with -N\n");
9898
error = 2;
9899
goto fini;
9900
}
9901
} else {
9902
target_pool = target;
9903
}
9904
9905
if (dump_opt['e'] || force_import) {
9906
importargs_t args = { 0 };
9907
9908
/*
9909
* If path is not provided, search in /dev
9910
*/
9911
if (searchdirs == NULL) {
9912
searchdirs = umem_alloc(sizeof (char *), UMEM_NOFAIL);
9913
searchdirs[nsearch++] = (char *)ZFS_DEVDIR;
9914
}
9915
9916
args.paths = nsearch;
9917
args.path = searchdirs;
9918
args.can_be_active = B_TRUE;
9919
9920
libpc_handle_t lpch = {
9921
.lpc_lib_handle = NULL,
9922
.lpc_ops = &libzpool_config_ops,
9923
.lpc_printerr = B_TRUE
9924
};
9925
error = zpool_find_config(&lpch, target_pool, &cfg, &args);
9926
9927
if (error == 0) {
9928
9929
if (nvlist_add_nvlist(cfg,
9930
ZPOOL_LOAD_POLICY, policy) != 0) {
9931
fatal("can't open '%s': %s",
9932
target, strerror(ENOMEM));
9933
}
9934
9935
if (dump_opt['C'] > 1) {
9936
(void) printf("\nConfiguration for import:\n");
9937
dump_nvlist(cfg, 8);
9938
}
9939
9940
/*
9941
* Disable the activity check to allow examination of
9942
* active pools.
9943
*/
9944
error = spa_import(target_pool, cfg, NULL,
9945
flags | ZFS_IMPORT_SKIP_MMP);
9946
}
9947
}
9948
9949
if (searchdirs != NULL) {
9950
umem_free(searchdirs, nsearch * sizeof (char *));
9951
searchdirs = NULL;
9952
}
9953
9954
/*
9955
* We need to make sure to process -O option or call
9956
* dump_path after the -e option has been processed,
9957
* which imports the pool to the namespace if it's
9958
* not in the cachefile.
9959
*/
9960
if (dump_opt['O'] && !dump_opt['r']) {
9961
if (argc != 2)
9962
usage();
9963
dump_opt['v'] = verbose + 3;
9964
error = dump_path(argv[0], argv[1], NULL);
9965
goto fini;
9966
}
9967
9968
if (dump_opt['r']) {
9969
target_is_spa = B_FALSE;
9970
if (argc != 3)
9971
usage();
9972
dump_opt['v'] = verbose;
9973
if (dump_opt['O']) {
9974
object = strtoull(argv[1], NULL, 0);
9975
} else {
9976
error = dump_path(argv[0], argv[1], &object);
9977
}
9978
if (error != 0)
9979
fatal("internal error: %s", strerror(error));
9980
}
9981
9982
/*
9983
* import_checkpointed_state makes the assumption that the
9984
* target pool that we pass it is already part of the spa
9985
* namespace. Because of that we need to make sure to call
9986
* it always after the -e option has been processed, which
9987
* imports the pool to the namespace if it's not in the
9988
* cachefile.
9989
*/
9990
char *checkpoint_pool = NULL;
9991
char *checkpoint_target = NULL;
9992
if (dump_opt['k']) {
9993
checkpoint_pool = import_checkpointed_state(target, cfg,
9994
target_is_spa, &checkpoint_target);
9995
9996
if (checkpoint_target != NULL)
9997
target = checkpoint_target;
9998
}
9999
10000
if (cfg != NULL) {
10001
nvlist_free(cfg);
10002
cfg = NULL;
10003
}
10004
10005
if (target_pool != target)
10006
free(target_pool);
10007
10008
if (error == 0) {
10009
if (dump_opt['k'] && (target_is_spa || dump_opt['R'])) {
10010
ASSERT(checkpoint_pool != NULL);
10011
ASSERT0P(checkpoint_target);
10012
10013
error = spa_open(checkpoint_pool, &spa, FTAG);
10014
if (error != 0) {
10015
fatal("Tried to open pool \"%s\" but "
10016
"spa_open() failed with error %d\n",
10017
checkpoint_pool, error);
10018
}
10019
10020
} else if (target_is_spa || dump_opt['R'] || dump_opt['B'] ||
10021
objset_id == 0) {
10022
zdb_set_skip_mmp(target);
10023
error = spa_open_rewind(target, &spa, FTAG, policy,
10024
NULL);
10025
if (error) {
10026
/*
10027
* If we're missing the log device then
10028
* try opening the pool after clearing the
10029
* log state.
10030
*/
10031
spa_namespace_enter(FTAG);
10032
if ((spa = spa_lookup(target)) != NULL &&
10033
spa->spa_log_state == SPA_LOG_MISSING) {
10034
spa->spa_log_state = SPA_LOG_CLEAR;
10035
error = 0;
10036
}
10037
spa_namespace_exit(FTAG);
10038
10039
if (!error) {
10040
error = spa_open_rewind(target, &spa,
10041
FTAG, policy, NULL);
10042
}
10043
}
10044
} else if (strpbrk(target, "#") != NULL) {
10045
dsl_pool_t *dp;
10046
error = dsl_pool_hold(target, FTAG, &dp);
10047
if (error != 0) {
10048
fatal("can't dump '%s': %s", target,
10049
strerror(error));
10050
}
10051
error = dump_bookmark(dp, target, B_TRUE, verbose > 1);
10052
dsl_pool_rele(dp, FTAG);
10053
if (error != 0) {
10054
fatal("can't dump '%s': %s", target,
10055
strerror(error));
10056
}
10057
goto fini;
10058
} else {
10059
target_pool = strdup(target);
10060
if (strpbrk(target, "/@") != NULL)
10061
*strpbrk(target_pool, "/@") = '\0';
10062
10063
zdb_set_skip_mmp(target);
10064
/*
10065
* If -N was supplied, the user has indicated that
10066
* zdb -d <pool>/<objsetID> is in effect. Otherwise
10067
* we first assume that the dataset string is the
10068
* dataset name. If dmu_objset_hold fails with the
10069
* dataset string, and we have an objset_id, retry the
10070
* lookup with the objsetID.
10071
*/
10072
boolean_t retry = B_TRUE;
10073
retry_lookup:
10074
if (dataset_lookup == B_TRUE) {
10075
/*
10076
* Use the supplied id to get the name
10077
* for open_objset.
10078
*/
10079
error = spa_open(target_pool, &spa, FTAG);
10080
if (error == 0) {
10081
error = name_from_objset_id(spa,
10082
objset_id, dsname);
10083
spa_close(spa, FTAG);
10084
if (error == 0)
10085
target = dsname;
10086
}
10087
}
10088
if (error == 0) {
10089
if (objset_id > 0 && retry) {
10090
int err = dmu_objset_hold(target, FTAG,
10091
&os);
10092
if (err) {
10093
dataset_lookup = B_TRUE;
10094
retry = B_FALSE;
10095
goto retry_lookup;
10096
} else {
10097
dmu_objset_rele(os, FTAG);
10098
}
10099
}
10100
error = open_objset(target, FTAG, &os);
10101
}
10102
if (error == 0)
10103
spa = dmu_objset_spa(os);
10104
free(target_pool);
10105
}
10106
}
10107
nvlist_free(policy);
10108
10109
if (error)
10110
fatal("can't open '%s': %s", target, strerror(error));
10111
10112
/*
10113
* Set the pool failure mode to panic in order to prevent the pool
10114
* from suspending. A suspended I/O will have no way to resume and
10115
* can prevent the zdb(8) command from terminating as expected.
10116
*/
10117
if (spa != NULL)
10118
spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
10119
10120
argv++;
10121
argc--;
10122
if (dump_opt['r']) {
10123
error = zdb_copy_object(os, object, argv[1]);
10124
} else if (!dump_opt['R']) {
10125
flagbits['d'] = ZOR_FLAG_DIRECTORY;
10126
flagbits['f'] = ZOR_FLAG_PLAIN_FILE;
10127
flagbits['m'] = ZOR_FLAG_SPACE_MAP;
10128
flagbits['z'] = ZOR_FLAG_ZAP;
10129
flagbits['A'] = ZOR_FLAG_ALL_TYPES;
10130
10131
if (argc > 0 && dump_opt['d']) {
10132
zopt_object_args = argc;
10133
zopt_object_ranges = calloc(zopt_object_args,
10134
sizeof (zopt_object_range_t));
10135
for (unsigned i = 0; i < zopt_object_args; i++) {
10136
int err;
10137
const char *msg = NULL;
10138
10139
err = parse_object_range(argv[i],
10140
&zopt_object_ranges[i], &msg);
10141
if (err != 0)
10142
fatal("Bad object or range: '%s': %s\n",
10143
argv[i], msg ?: "");
10144
}
10145
} else if (argc > 0 && dump_opt['m']) {
10146
zopt_metaslab_args = argc;
10147
zopt_metaslab = calloc(zopt_metaslab_args,
10148
sizeof (uint64_t));
10149
for (unsigned i = 0; i < zopt_metaslab_args; i++) {
10150
errno = 0;
10151
zopt_metaslab[i] = strtoull(argv[i], NULL, 0);
10152
if (zopt_metaslab[i] == 0 && errno != 0)
10153
fatal("bad number %s: %s", argv[i],
10154
strerror(errno));
10155
}
10156
}
10157
if (dump_opt['B']) {
10158
dump_backup(target, objset_id,
10159
argc > 0 ? argv[0] : NULL);
10160
} else if (os != NULL) {
10161
dump_objset(os);
10162
} else if (zopt_object_args > 0 && !dump_opt['m']) {
10163
dump_objset(spa->spa_meta_objset);
10164
} else {
10165
dump_zpool(spa);
10166
}
10167
} else {
10168
flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR;
10169
flagbits['c'] = ZDB_FLAG_CHECKSUM;
10170
flagbits['d'] = ZDB_FLAG_DECOMPRESS;
10171
flagbits['e'] = ZDB_FLAG_BSWAP;
10172
flagbits['g'] = ZDB_FLAG_GBH;
10173
flagbits['i'] = ZDB_FLAG_INDIRECT;
10174
flagbits['r'] = ZDB_FLAG_RAW;
10175
flagbits['v'] = ZDB_FLAG_VERBOSE;
10176
10177
for (int i = 0; i < argc; i++)
10178
zdb_read_block(argv[i], spa);
10179
}
10180
10181
if (dump_opt['k']) {
10182
free(checkpoint_pool);
10183
if (!target_is_spa)
10184
free(checkpoint_target);
10185
}
10186
10187
fini:
10188
if (spa != NULL)
10189
zdb_ddt_cleanup(spa);
10190
10191
if (os != NULL) {
10192
close_objset(os, FTAG);
10193
} else if (spa != NULL) {
10194
spa_close(spa, FTAG);
10195
}
10196
10197
fuid_table_destroy();
10198
10199
dump_debug_buffer();
10200
10201
if (kernel_init_done)
10202
kernel_fini();
10203
10204
if (corruption_found && error == 0)
10205
error = 3;
10206
10207
return (error);
10208
}
10209
10210