4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2016 Nexenta Systems, Inc.
27 * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC.
28 * Copyright (c) 2015, 2017, Intel Corporation.
29 * Copyright (c) 2020 Datto Inc.
30 * Copyright (c) 2020, The FreeBSD Foundation [1]
32 * [1] Portions of this software were developed by Allan Jude
33 * under sponsorship from the FreeBSD Foundation.
34 * Copyright (c) 2021 Allan Jude
35 * Copyright (c) 2021 Toomas Soome <tsoome@me.com>
36 * Copyright (c) 2023, Klara Inc.
37 * Copyright (c) 2023, Rob Norris <robn@despairlabs.com>
45 #include <openssl/evp.h>
46 #include <sys/zfs_context.h>
48 #include <sys/spa_impl.h>
51 #include <sys/fs/zfs.h>
52 #include <sys/zfs_znode.h>
53 #include <sys/zfs_sa.h>
55 #include <sys/sa_impl.h>
57 #include <sys/vdev_impl.h>
58 #include <sys/metaslab_impl.h>
59 #include <sys/dmu_objset.h>
60 #include <sys/dsl_dir.h>
61 #include <sys/dsl_dataset.h>
62 #include <sys/dsl_pool.h>
63 #include <sys/dsl_bookmark.h>
66 #include <sys/zil_impl.h>
68 #include <sys/resource.h>
69 #include <sys/dmu_send.h>
70 #include <sys/dmu_traverse.h>
71 #include <sys/zio_checksum.h>
72 #include <sys/zio_compress.h>
73 #include <sys/zfs_fuid.h>
75 #include <sys/arc_impl.h>
77 #include <sys/ddt_impl.h>
78 #include <sys/zfeature.h>
80 #include <sys/blkptr.h>
81 #include <sys/dsl_crypt.h>
82 #include <sys/dsl_scan.h>
83 #include <sys/btree.h>
85 #include <sys/brt_impl.h>
86 #include <zfs_comutil.h>
87 #include <sys/zstd/zstd.h>
89 #include <libnvpair.h>
97 extern int reference_tracking_enable
;
98 extern int zfs_recover
;
99 extern uint_t zfs_vdev_async_read_max_active
;
100 extern boolean_t spa_load_verify_dryrun
;
101 extern boolean_t spa_mode_readable_spacemaps
;
102 extern uint_t zfs_reconstruct_indirect_combinations_max
;
103 extern uint_t zfs_btree_verify_intensity
;
105 static const char cmdname
[] = "zdb";
106 uint8_t dump_opt
[256];
108 typedef void object_viewer_t(objset_t
*, uint64_t, void *data
, size_t size
);
110 static uint64_t *zopt_metaslab
= NULL
;
111 static unsigned zopt_metaslab_args
= 0;
114 static zopt_object_range_t
*zopt_object_ranges
= NULL
;
115 static unsigned zopt_object_args
= 0;
117 static int flagbits
[256];
120 static uint64_t max_inflight_bytes
= 256 * 1024 * 1024; /* 256MB */
121 static int leaked_objects
= 0;
122 static range_tree_t
*mos_refd_objs
;
124 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t
*,
126 static void mos_obj_refd(uint64_t);
127 static void mos_obj_refd_multiple(uint64_t);
128 static int dump_bpobj_cb(void *arg
, const blkptr_t
*bp
, boolean_t free
,
133 static void zdb_print_blkptr(const blkptr_t
*bp
, int flags
);
135 typedef struct sublivelist_verify_block_refcnt
{
136 /* block pointer entry in livelist being verified */
140 * Refcount gets incremented to 1 when we encounter the first
141 * FREE entry for the svfbr block pointer and a node for it
142 * is created in our ZDB verification/tracking metadata.
144 * As we encounter more FREE entries we increment this counter
145 * and similarly decrement it whenever we find the respective
146 * ALLOC entries for this block.
148 * When the refcount gets to 0 it means that all the FREE and
149 * ALLOC entries of this block have paired up and we no longer
150 * need to track it in our verification logic (e.g. the node
151 * containing this struct in our verification data structure
154 * [refer to sublivelist_verify_blkptr() for the actual code]
156 uint32_t svbr_refcnt
;
157 } sublivelist_verify_block_refcnt_t
;
160 sublivelist_block_refcnt_compare(const void *larg
, const void *rarg
)
162 const sublivelist_verify_block_refcnt_t
*l
= larg
;
163 const sublivelist_verify_block_refcnt_t
*r
= rarg
;
164 return (livelist_compare(&l
->svbr_blk
, &r
->svbr_blk
));
168 sublivelist_verify_blkptr(void *arg
, const blkptr_t
*bp
, boolean_t free
,
171 ASSERT3P(tx
, ==, NULL
);
172 struct sublivelist_verify
*sv
= arg
;
173 sublivelist_verify_block_refcnt_t current
= {
177 * Start with 1 in case this is the first free entry.
178 * This field is not used for our B-Tree comparisons
184 zfs_btree_index_t where
;
185 sublivelist_verify_block_refcnt_t
*pair
=
186 zfs_btree_find(&sv
->sv_pair
, ¤t
, &where
);
189 /* first free entry for this block pointer */
190 zfs_btree_add(&sv
->sv_pair
, ¤t
);
196 /* block that is currently marked as allocated */
197 for (int i
= 0; i
< SPA_DVAS_PER_BP
; i
++) {
198 if (DVA_IS_EMPTY(&bp
->blk_dva
[i
]))
200 sublivelist_verify_block_t svb
= {
201 .svb_dva
= bp
->blk_dva
[i
],
202 .svb_allocated_txg
= bp
->blk_birth
205 if (zfs_btree_find(&sv
->sv_leftover
, &svb
,
207 zfs_btree_add_idx(&sv
->sv_leftover
,
212 /* alloc matches a free entry */
214 if (pair
->svbr_refcnt
== 0) {
215 /* all allocs and frees have been matched */
216 zfs_btree_remove_idx(&sv
->sv_pair
, &where
);
225 sublivelist_verify_func(void *args
, dsl_deadlist_entry_t
*dle
)
228 struct sublivelist_verify
*sv
= args
;
230 zfs_btree_create(&sv
->sv_pair
, sublivelist_block_refcnt_compare
, NULL
,
231 sizeof (sublivelist_verify_block_refcnt_t
));
233 err
= bpobj_iterate_nofree(&dle
->dle_bpobj
, sublivelist_verify_blkptr
,
236 sublivelist_verify_block_refcnt_t
*e
;
237 zfs_btree_index_t
*cookie
= NULL
;
238 while ((e
= zfs_btree_destroy_nodes(&sv
->sv_pair
, &cookie
)) != NULL
) {
239 char blkbuf
[BP_SPRINTF_LEN
];
240 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
),
241 &e
->svbr_blk
, B_TRUE
);
242 (void) printf("\tERROR: %d unmatched FREE(s): %s\n",
243 e
->svbr_refcnt
, blkbuf
);
245 zfs_btree_destroy(&sv
->sv_pair
);
251 livelist_block_compare(const void *larg
, const void *rarg
)
253 const sublivelist_verify_block_t
*l
= larg
;
254 const sublivelist_verify_block_t
*r
= rarg
;
256 if (DVA_GET_VDEV(&l
->svb_dva
) < DVA_GET_VDEV(&r
->svb_dva
))
258 else if (DVA_GET_VDEV(&l
->svb_dva
) > DVA_GET_VDEV(&r
->svb_dva
))
261 if (DVA_GET_OFFSET(&l
->svb_dva
) < DVA_GET_OFFSET(&r
->svb_dva
))
263 else if (DVA_GET_OFFSET(&l
->svb_dva
) > DVA_GET_OFFSET(&r
->svb_dva
))
266 if (DVA_GET_ASIZE(&l
->svb_dva
) < DVA_GET_ASIZE(&r
->svb_dva
))
268 else if (DVA_GET_ASIZE(&l
->svb_dva
) > DVA_GET_ASIZE(&r
->svb_dva
))
275 * Check for errors in a livelist while tracking all unfreed ALLOCs in the
276 * sublivelist_verify_t: sv->sv_leftover
279 livelist_verify(dsl_deadlist_t
*dl
, void *arg
)
281 sublivelist_verify_t
*sv
= arg
;
282 dsl_deadlist_iterate(dl
, sublivelist_verify_func
, sv
);
286 * Check for errors in the livelist entry and discard the intermediary
290 sublivelist_verify_lightweight(void *args
, dsl_deadlist_entry_t
*dle
)
293 sublivelist_verify_t sv
;
294 zfs_btree_create(&sv
.sv_leftover
, livelist_block_compare
, NULL
,
295 sizeof (sublivelist_verify_block_t
));
296 int err
= sublivelist_verify_func(&sv
, dle
);
297 zfs_btree_clear(&sv
.sv_leftover
);
298 zfs_btree_destroy(&sv
.sv_leftover
);
302 typedef struct metaslab_verify
{
304 * Tree containing all the leftover ALLOCs from the livelists
305 * that are part of this metaslab.
307 zfs_btree_t mv_livelist_allocs
;
310 * Metaslab information.
318 * What's currently allocated for this metaslab.
320 range_tree_t
*mv_allocated
;
323 typedef void ll_iter_t(dsl_deadlist_t
*ll
, void *arg
);
325 typedef int (*zdb_log_sm_cb_t
)(spa_t
*spa
, space_map_entry_t
*sme
, uint64_t txg
,
328 typedef struct unflushed_iter_cb_arg
{
332 zdb_log_sm_cb_t uic_cb
;
333 } unflushed_iter_cb_arg_t
;
336 iterate_through_spacemap_logs_cb(space_map_entry_t
*sme
, void *arg
)
338 unflushed_iter_cb_arg_t
*uic
= arg
;
339 return (uic
->uic_cb(uic
->uic_spa
, sme
, uic
->uic_txg
, uic
->uic_arg
));
343 iterate_through_spacemap_logs(spa_t
*spa
, zdb_log_sm_cb_t cb
, void *arg
)
345 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
348 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
349 for (spa_log_sm_t
*sls
= avl_first(&spa
->spa_sm_logs_by_txg
);
350 sls
; sls
= AVL_NEXT(&spa
->spa_sm_logs_by_txg
, sls
)) {
351 space_map_t
*sm
= NULL
;
352 VERIFY0(space_map_open(&sm
, spa_meta_objset(spa
),
353 sls
->sls_sm_obj
, 0, UINT64_MAX
, SPA_MINBLOCKSHIFT
));
355 unflushed_iter_cb_arg_t uic
= {
357 .uic_txg
= sls
->sls_txg
,
361 VERIFY0(space_map_iterate(sm
, space_map_length(sm
),
362 iterate_through_spacemap_logs_cb
, &uic
));
365 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
369 verify_livelist_allocs(metaslab_verify_t
*mv
, uint64_t txg
,
370 uint64_t offset
, uint64_t size
)
372 sublivelist_verify_block_t svb
= {{{0}}};
373 DVA_SET_VDEV(&svb
.svb_dva
, mv
->mv_vdid
);
374 DVA_SET_OFFSET(&svb
.svb_dva
, offset
);
375 DVA_SET_ASIZE(&svb
.svb_dva
, size
);
376 zfs_btree_index_t where
;
377 uint64_t end_offset
= offset
+ size
;
380 * Look for an exact match for spacemap entry in the livelist entries.
381 * Then, look for other livelist entries that fall within the range
382 * of the spacemap entry as it may have been condensed
384 sublivelist_verify_block_t
*found
=
385 zfs_btree_find(&mv
->mv_livelist_allocs
, &svb
, &where
);
387 found
= zfs_btree_next(&mv
->mv_livelist_allocs
, &where
, &where
);
389 for (; found
!= NULL
&& DVA_GET_VDEV(&found
->svb_dva
) == mv
->mv_vdid
&&
390 DVA_GET_OFFSET(&found
->svb_dva
) < end_offset
;
391 found
= zfs_btree_next(&mv
->mv_livelist_allocs
, &where
, &where
)) {
392 if (found
->svb_allocated_txg
<= txg
) {
393 (void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
394 "from TXG %llx FREED at TXG %llx\n",
395 (u_longlong_t
)DVA_GET_OFFSET(&found
->svb_dva
),
396 (u_longlong_t
)DVA_GET_ASIZE(&found
->svb_dva
),
397 (u_longlong_t
)found
->svb_allocated_txg
,
404 metaslab_spacemap_validation_cb(space_map_entry_t
*sme
, void *arg
)
406 metaslab_verify_t
*mv
= arg
;
407 uint64_t offset
= sme
->sme_offset
;
408 uint64_t size
= sme
->sme_run
;
409 uint64_t txg
= sme
->sme_txg
;
411 if (sme
->sme_type
== SM_ALLOC
) {
412 if (range_tree_contains(mv
->mv_allocated
,
414 (void) printf("ERROR: DOUBLE ALLOC: "
416 "%llu:%llu LOG_SM\n",
417 (u_longlong_t
)txg
, (u_longlong_t
)offset
,
418 (u_longlong_t
)size
, (u_longlong_t
)mv
->mv_vdid
,
419 (u_longlong_t
)mv
->mv_msid
);
421 range_tree_add(mv
->mv_allocated
,
425 if (!range_tree_contains(mv
->mv_allocated
,
427 (void) printf("ERROR: DOUBLE FREE: "
429 "%llu:%llu LOG_SM\n",
430 (u_longlong_t
)txg
, (u_longlong_t
)offset
,
431 (u_longlong_t
)size
, (u_longlong_t
)mv
->mv_vdid
,
432 (u_longlong_t
)mv
->mv_msid
);
434 range_tree_remove(mv
->mv_allocated
,
439 if (sme
->sme_type
!= SM_ALLOC
) {
441 * If something is freed in the spacemap, verify that
442 * it is not listed as allocated in the livelist.
444 verify_livelist_allocs(mv
, txg
, offset
, size
);
450 spacemap_check_sm_log_cb(spa_t
*spa
, space_map_entry_t
*sme
,
451 uint64_t txg
, void *arg
)
453 metaslab_verify_t
*mv
= arg
;
454 uint64_t offset
= sme
->sme_offset
;
455 uint64_t vdev_id
= sme
->sme_vdev
;
457 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
459 /* skip indirect vdevs */
460 if (!vdev_is_concrete(vd
))
463 if (vdev_id
!= mv
->mv_vdid
)
466 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
467 if (ms
->ms_id
!= mv
->mv_msid
)
470 if (txg
< metaslab_unflushed_txg(ms
))
474 ASSERT3U(txg
, ==, sme
->sme_txg
);
475 return (metaslab_spacemap_validation_cb(sme
, mv
));
479 spacemap_check_sm_log(spa_t
*spa
, metaslab_verify_t
*mv
)
481 iterate_through_spacemap_logs(spa
, spacemap_check_sm_log_cb
, mv
);
485 spacemap_check_ms_sm(space_map_t
*sm
, metaslab_verify_t
*mv
)
490 VERIFY0(space_map_iterate(sm
, space_map_length(sm
),
491 metaslab_spacemap_validation_cb
, mv
));
494 static void iterate_deleted_livelists(spa_t
*spa
, ll_iter_t func
, void *arg
);
497 * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
498 * they are part of that metaslab (mv_msid).
501 mv_populate_livelist_allocs(metaslab_verify_t
*mv
, sublivelist_verify_t
*sv
)
503 zfs_btree_index_t where
;
504 sublivelist_verify_block_t
*svb
;
505 ASSERT3U(zfs_btree_numnodes(&mv
->mv_livelist_allocs
), ==, 0);
506 for (svb
= zfs_btree_first(&sv
->sv_leftover
, &where
);
508 svb
= zfs_btree_next(&sv
->sv_leftover
, &where
, &where
)) {
509 if (DVA_GET_VDEV(&svb
->svb_dva
) != mv
->mv_vdid
)
512 if (DVA_GET_OFFSET(&svb
->svb_dva
) < mv
->mv_start
&&
513 (DVA_GET_OFFSET(&svb
->svb_dva
) +
514 DVA_GET_ASIZE(&svb
->svb_dva
)) > mv
->mv_start
) {
515 (void) printf("ERROR: Found block that crosses "
516 "metaslab boundary: <%llu:%llx:%llx>\n",
517 (u_longlong_t
)DVA_GET_VDEV(&svb
->svb_dva
),
518 (u_longlong_t
)DVA_GET_OFFSET(&svb
->svb_dva
),
519 (u_longlong_t
)DVA_GET_ASIZE(&svb
->svb_dva
));
523 if (DVA_GET_OFFSET(&svb
->svb_dva
) < mv
->mv_start
)
526 if (DVA_GET_OFFSET(&svb
->svb_dva
) >= mv
->mv_end
)
529 if ((DVA_GET_OFFSET(&svb
->svb_dva
) +
530 DVA_GET_ASIZE(&svb
->svb_dva
)) > mv
->mv_end
) {
531 (void) printf("ERROR: Found block that crosses "
532 "metaslab boundary: <%llu:%llx:%llx>\n",
533 (u_longlong_t
)DVA_GET_VDEV(&svb
->svb_dva
),
534 (u_longlong_t
)DVA_GET_OFFSET(&svb
->svb_dva
),
535 (u_longlong_t
)DVA_GET_ASIZE(&svb
->svb_dva
));
539 zfs_btree_add(&mv
->mv_livelist_allocs
, svb
);
542 for (svb
= zfs_btree_first(&mv
->mv_livelist_allocs
, &where
);
544 svb
= zfs_btree_next(&mv
->mv_livelist_allocs
, &where
, &where
)) {
545 zfs_btree_remove(&sv
->sv_leftover
, svb
);
551 * Iterate through all the sublivelists and:
552 * - report leftover frees (**)
553 * - record leftover ALLOCs together with their TXG [see Cross Check]
555 * (**) Note: Double ALLOCs are valid in datasets that have dedup
556 * enabled. Similarly double FREEs are allowed as well but
557 * only if they pair up with a corresponding ALLOC entry once
558 * we our done with our sublivelist iteration.
562 * - iterate over spacemap and then the metaslab's entries in the
563 * spacemap log, then report any double FREEs and ALLOCs (do not
567 * After finishing the Livelist Check phase and while being in the
568 * Spacemap Check phase, we find all the recorded leftover ALLOCs
569 * of the livelist check that are part of the metaslab that we are
570 * currently looking at in the Spacemap Check. We report any entries
571 * that are marked as ALLOCs in the livelists but have been actually
572 * freed (and potentially allocated again) after their TXG stamp in
573 * the spacemaps. Also report any ALLOCs from the livelists that
574 * belong to indirect vdevs (e.g. their vdev completed removal).
576 * Note that this will miss Log Spacemap entries that cancelled each other
577 * out before being flushed to the metaslab, so we are not guaranteed
578 * to match all erroneous ALLOCs.
581 livelist_metaslab_validate(spa_t
*spa
)
583 (void) printf("Verifying deleted livelist entries\n");
585 sublivelist_verify_t sv
;
586 zfs_btree_create(&sv
.sv_leftover
, livelist_block_compare
, NULL
,
587 sizeof (sublivelist_verify_block_t
));
588 iterate_deleted_livelists(spa
, livelist_verify
, &sv
);
590 (void) printf("Verifying metaslab entries\n");
591 vdev_t
*rvd
= spa
->spa_root_vdev
;
592 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
593 vdev_t
*vd
= rvd
->vdev_child
[c
];
595 if (!vdev_is_concrete(vd
))
598 for (uint64_t mid
= 0; mid
< vd
->vdev_ms_count
; mid
++) {
599 metaslab_t
*m
= vd
->vdev_ms
[mid
];
601 (void) fprintf(stderr
,
602 "\rverifying concrete vdev %llu, "
603 "metaslab %llu of %llu ...",
604 (longlong_t
)vd
->vdev_id
,
606 (longlong_t
)vd
->vdev_ms_count
);
608 uint64_t shift
, start
;
609 range_seg_type_t type
=
610 metaslab_calculate_range_tree_type(vd
, m
,
612 metaslab_verify_t mv
;
613 mv
.mv_allocated
= range_tree_create(NULL
,
614 type
, NULL
, start
, shift
);
615 mv
.mv_vdid
= vd
->vdev_id
;
616 mv
.mv_msid
= m
->ms_id
;
617 mv
.mv_start
= m
->ms_start
;
618 mv
.mv_end
= m
->ms_start
+ m
->ms_size
;
619 zfs_btree_create(&mv
.mv_livelist_allocs
,
620 livelist_block_compare
, NULL
,
621 sizeof (sublivelist_verify_block_t
));
623 mv_populate_livelist_allocs(&mv
, &sv
);
625 spacemap_check_ms_sm(m
->ms_sm
, &mv
);
626 spacemap_check_sm_log(spa
, &mv
);
628 range_tree_vacate(mv
.mv_allocated
, NULL
, NULL
);
629 range_tree_destroy(mv
.mv_allocated
);
630 zfs_btree_clear(&mv
.mv_livelist_allocs
);
631 zfs_btree_destroy(&mv
.mv_livelist_allocs
);
634 (void) fprintf(stderr
, "\n");
637 * If there are any segments in the leftover tree after we walked
638 * through all the metaslabs in the concrete vdevs then this means
639 * that we have segments in the livelists that belong to indirect
640 * vdevs and are marked as allocated.
642 if (zfs_btree_numnodes(&sv
.sv_leftover
) == 0) {
643 zfs_btree_destroy(&sv
.sv_leftover
);
646 (void) printf("ERROR: Found livelist blocks marked as allocated "
647 "for indirect vdevs:\n");
649 zfs_btree_index_t
*where
= NULL
;
650 sublivelist_verify_block_t
*svb
;
651 while ((svb
= zfs_btree_destroy_nodes(&sv
.sv_leftover
, &where
)) !=
653 int vdev_id
= DVA_GET_VDEV(&svb
->svb_dva
);
654 ASSERT3U(vdev_id
, <, rvd
->vdev_children
);
655 vdev_t
*vd
= rvd
->vdev_child
[vdev_id
];
656 ASSERT(!vdev_is_concrete(vd
));
657 (void) printf("<%d:%llx:%llx> TXG %llx\n",
658 vdev_id
, (u_longlong_t
)DVA_GET_OFFSET(&svb
->svb_dva
),
659 (u_longlong_t
)DVA_GET_ASIZE(&svb
->svb_dva
),
660 (u_longlong_t
)svb
->svb_allocated_txg
);
663 zfs_btree_destroy(&sv
.sv_leftover
);
667 * These libumem hooks provide a reasonable set of defaults for the allocator's
668 * debugging facilities.
671 _umem_debug_init(void)
673 return ("default,verbose"); /* $UMEM_DEBUG setting */
677 _umem_logging_init(void)
679 return ("fail,contents"); /* $UMEM_LOGGING setting */
685 (void) fprintf(stderr
,
686 "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
687 "[-I <inflight I/Os>]\n"
688 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
690 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
691 "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
692 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
693 "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
694 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
695 "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
696 "\t%s [-v] <bookmark>\n"
697 "\t%s -C [-A] [-U <cache>] [<poolname>]\n"
698 "\t%s -l [-Aqu] <device>\n"
699 "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
700 "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
701 "\t%s -O [-K <key>] <dataset> <path>\n"
702 "\t%s -r [-K <key>] <dataset> <path> <destination>\n"
703 "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
704 "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
705 "\t%s -E [-A] word0:word1:...:word15\n"
706 "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
708 cmdname
, cmdname
, cmdname
, cmdname
, cmdname
, cmdname
, cmdname
,
709 cmdname
, cmdname
, cmdname
, cmdname
, cmdname
);
711 (void) fprintf(stderr
, " Dataset name must include at least one "
712 "separator character '/' or '@'\n");
713 (void) fprintf(stderr
, " If dataset name is specified, only that "
714 "dataset is dumped\n");
715 (void) fprintf(stderr
, " If object numbers or object number "
716 "ranges are specified, only those\n"
717 " objects or ranges are dumped.\n\n");
718 (void) fprintf(stderr
,
719 " Object ranges take the form <start>:<end>[:<flags>]\n"
720 " start Starting object number\n"
721 " end Ending object number, or -1 for no upper bound\n"
722 " flags Optional flags to select object types:\n"
723 " A All objects (this is the default)\n"
724 " d ZFS directories\n"
726 " m SPA space maps\n"
728 " - Negate effect of next flag\n\n");
729 (void) fprintf(stderr
, " Options to control amount of output:\n");
730 (void) fprintf(stderr
, " -b --block-stats "
731 "block statistics\n");
732 (void) fprintf(stderr
, " -B --backup "
734 (void) fprintf(stderr
, " -c --checksum "
735 "checksum all metadata (twice for all data) blocks\n");
736 (void) fprintf(stderr
, " -C --config "
737 "config (or cachefile if alone)\n");
738 (void) fprintf(stderr
, " -d --datasets "
740 (void) fprintf(stderr
, " -D --dedup-stats "
741 "dedup statistics\n");
742 (void) fprintf(stderr
, " -E --embedded-block-pointer=INTEGER\n"
743 " decode and display block "
744 "from an embedded block pointer\n");
745 (void) fprintf(stderr
, " -h --history "
747 (void) fprintf(stderr
, " -i --intent-logs "
749 (void) fprintf(stderr
, " -l --label "
750 "read label contents\n");
751 (void) fprintf(stderr
, " -k --checkpointed-state "
752 "examine the checkpointed state of the pool\n");
753 (void) fprintf(stderr
, " -L --disable-leak-tracking "
754 "disable leak tracking (do not load spacemaps)\n");
755 (void) fprintf(stderr
, " -m --metaslabs "
757 (void) fprintf(stderr
, " -M --metaslab-groups "
758 "metaslab groups\n");
759 (void) fprintf(stderr
, " -O --object-lookups "
760 "perform object lookups by path\n");
761 (void) fprintf(stderr
, " -r --copy-object "
762 "copy an object by path to file\n");
763 (void) fprintf(stderr
, " -R --read-block "
764 "read and display block from a device\n");
765 (void) fprintf(stderr
, " -s --io-stats "
766 "report stats on zdb's I/O\n");
767 (void) fprintf(stderr
, " -S --simulate-dedup "
768 "simulate dedup to measure effect\n");
769 (void) fprintf(stderr
, " -v --verbose "
770 "verbose (applies to all others)\n");
771 (void) fprintf(stderr
, " -y --livelist "
772 "perform livelist and metaslab validation on any livelists being "
774 (void) fprintf(stderr
, " Below options are intended for use "
775 "with other options:\n");
776 (void) fprintf(stderr
, " -A --ignore-assertions "
777 "ignore assertions (-A), enable panic recovery (-AA) or both "
779 (void) fprintf(stderr
, " -e --exported "
780 "pool is exported/destroyed/has altroot/not in a cachefile\n");
781 (void) fprintf(stderr
, " -F --automatic-rewind "
782 "attempt automatic rewind within safe range of transaction "
784 (void) fprintf(stderr
, " -G --dump-debug-msg "
785 "dump zfs_dbgmsg buffer before exiting\n");
786 (void) fprintf(stderr
, " -I --inflight=INTEGER "
787 "specify the maximum number of checksumming I/Os "
788 "[default is 200]\n");
789 (void) fprintf(stderr
, " -K --key=KEY "
790 "decryption key for encrypted dataset\n");
791 (void) fprintf(stderr
, " -o --option=\"OPTION=INTEGER\" "
792 "set global variable to an unsigned 32-bit integer\n");
793 (void) fprintf(stderr
, " -p --path==PATH "
794 "use one or more with -e to specify path to vdev dir\n");
795 (void) fprintf(stderr
, " -P --parseable "
796 "print numbers in parseable form\n");
797 (void) fprintf(stderr
, " -q --skip-label "
798 "don't print label contents\n");
799 (void) fprintf(stderr
, " -t --txg=INTEGER "
800 "highest txg to use when searching for uberblocks\n");
801 (void) fprintf(stderr
, " -T --brt-stats "
803 (void) fprintf(stderr
, " -u --uberblock "
805 (void) fprintf(stderr
, " -U --cachefile=PATH "
806 "use alternate cachefile\n");
807 (void) fprintf(stderr
, " -V --verbatim "
808 "do verbatim import\n");
809 (void) fprintf(stderr
, " -x --dump-blocks=PATH "
810 "dump all read blocks into specified directory\n");
811 (void) fprintf(stderr
, " -X --extreme-rewind "
812 "attempt extreme rewind (does not work with dataset)\n");
813 (void) fprintf(stderr
, " -Y --all-reconstruction "
814 "attempt all reconstruction combinations for split blocks\n");
815 (void) fprintf(stderr
, " -Z --zstd-headers "
816 "show ZSTD headers \n");
817 (void) fprintf(stderr
, "Specify an option more than once (e.g. -bb) "
818 "to make only that option verbose\n");
819 (void) fprintf(stderr
, "Default is to dump everything non-verbosely\n");
824 dump_debug_buffer(void)
828 (void) fflush(stdout
);
829 zfs_dbgmsg_print("zdb");
834 * Called for usage errors that are discovered after a call to spa_open(),
835 * dmu_bonus_hold(), or pool_match(). abort() is called for other errors.
839 fatal(const char *fmt
, ...)
844 (void) fprintf(stderr
, "%s: ", cmdname
);
845 (void) vfprintf(stderr
, fmt
, ap
);
847 (void) fprintf(stderr
, "\n");
855 dump_packed_nvlist(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
859 size_t nvsize
= *(uint64_t *)data
;
860 char *packed
= umem_alloc(nvsize
, UMEM_NOFAIL
);
862 VERIFY(0 == dmu_read(os
, object
, 0, nvsize
, packed
, DMU_READ_PREFETCH
));
864 VERIFY(nvlist_unpack(packed
, nvsize
, &nv
, 0) == 0);
866 umem_free(packed
, nvsize
);
874 dump_history_offsets(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
876 (void) os
, (void) object
, (void) size
;
877 spa_history_phys_t
*shp
= data
;
882 (void) printf("\t\tpool_create_len = %llu\n",
883 (u_longlong_t
)shp
->sh_pool_create_len
);
884 (void) printf("\t\tphys_max_off = %llu\n",
885 (u_longlong_t
)shp
->sh_phys_max_off
);
886 (void) printf("\t\tbof = %llu\n",
887 (u_longlong_t
)shp
->sh_bof
);
888 (void) printf("\t\teof = %llu\n",
889 (u_longlong_t
)shp
->sh_eof
);
890 (void) printf("\t\trecords_lost = %llu\n",
891 (u_longlong_t
)shp
->sh_records_lost
);
895 zdb_nicenum(uint64_t num
, char *buf
, size_t buflen
)
898 (void) snprintf(buf
, buflen
, "%llu", (longlong_t
)num
);
900 nicenum(num
, buf
, buflen
);
904 zdb_nicebytes(uint64_t bytes
, char *buf
, size_t buflen
)
907 (void) snprintf(buf
, buflen
, "%llu", (longlong_t
)bytes
);
909 zfs_nicebytes(bytes
, buf
, buflen
);
912 static const char histo_stars
[] = "****************************************";
913 static const uint64_t histo_width
= sizeof (histo_stars
) - 1;
916 dump_histogram(const uint64_t *histo
, int size
, int offset
)
919 int minidx
= size
- 1;
923 for (i
= 0; i
< size
; i
++) {
934 if (max
< histo_width
)
937 for (i
= minidx
; i
<= maxidx
; i
++) {
938 (void) printf("\t\t\t%3u: %6llu %s\n",
939 i
+ offset
, (u_longlong_t
)histo
[i
],
940 &histo_stars
[(max
- histo
[i
]) * histo_width
/ max
]);
945 dump_zap_stats(objset_t
*os
, uint64_t object
)
950 error
= zap_get_stats(os
, object
, &zs
);
954 if (zs
.zs_ptrtbl_len
== 0) {
955 ASSERT(zs
.zs_num_blocks
== 1);
956 (void) printf("\tmicrozap: %llu bytes, %llu entries\n",
957 (u_longlong_t
)zs
.zs_blocksize
,
958 (u_longlong_t
)zs
.zs_num_entries
);
962 (void) printf("\tFat ZAP stats:\n");
964 (void) printf("\t\tPointer table:\n");
965 (void) printf("\t\t\t%llu elements\n",
966 (u_longlong_t
)zs
.zs_ptrtbl_len
);
967 (void) printf("\t\t\tzt_blk: %llu\n",
968 (u_longlong_t
)zs
.zs_ptrtbl_zt_blk
);
969 (void) printf("\t\t\tzt_numblks: %llu\n",
970 (u_longlong_t
)zs
.zs_ptrtbl_zt_numblks
);
971 (void) printf("\t\t\tzt_shift: %llu\n",
972 (u_longlong_t
)zs
.zs_ptrtbl_zt_shift
);
973 (void) printf("\t\t\tzt_blks_copied: %llu\n",
974 (u_longlong_t
)zs
.zs_ptrtbl_blks_copied
);
975 (void) printf("\t\t\tzt_nextblk: %llu\n",
976 (u_longlong_t
)zs
.zs_ptrtbl_nextblk
);
978 (void) printf("\t\tZAP entries: %llu\n",
979 (u_longlong_t
)zs
.zs_num_entries
);
980 (void) printf("\t\tLeaf blocks: %llu\n",
981 (u_longlong_t
)zs
.zs_num_leafs
);
982 (void) printf("\t\tTotal blocks: %llu\n",
983 (u_longlong_t
)zs
.zs_num_blocks
);
984 (void) printf("\t\tzap_block_type: 0x%llx\n",
985 (u_longlong_t
)zs
.zs_block_type
);
986 (void) printf("\t\tzap_magic: 0x%llx\n",
987 (u_longlong_t
)zs
.zs_magic
);
988 (void) printf("\t\tzap_salt: 0x%llx\n",
989 (u_longlong_t
)zs
.zs_salt
);
991 (void) printf("\t\tLeafs with 2^n pointers:\n");
992 dump_histogram(zs
.zs_leafs_with_2n_pointers
, ZAP_HISTOGRAM_SIZE
, 0);
994 (void) printf("\t\tBlocks with n*5 entries:\n");
995 dump_histogram(zs
.zs_blocks_with_n5_entries
, ZAP_HISTOGRAM_SIZE
, 0);
997 (void) printf("\t\tBlocks n/10 full:\n");
998 dump_histogram(zs
.zs_blocks_n_tenths_full
, ZAP_HISTOGRAM_SIZE
, 0);
1000 (void) printf("\t\tEntries with n chunks:\n");
1001 dump_histogram(zs
.zs_entries_using_n_chunks
, ZAP_HISTOGRAM_SIZE
, 0);
1003 (void) printf("\t\tBuckets with n entries:\n");
1004 dump_histogram(zs
.zs_buckets_with_n_entries
, ZAP_HISTOGRAM_SIZE
, 0);
1008 dump_none(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1010 (void) os
, (void) object
, (void) data
, (void) size
;
1014 dump_unknown(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1016 (void) os
, (void) object
, (void) data
, (void) size
;
1017 (void) printf("\tUNKNOWN OBJECT TYPE\n");
1021 dump_uint8(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1023 (void) os
, (void) object
, (void) data
, (void) size
;
1027 dump_uint64(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1031 if (dump_opt
['d'] < 6)
1035 dmu_object_info_t doi
;
1037 VERIFY0(dmu_object_info(os
, object
, &doi
));
1038 size
= doi
.doi_max_offset
;
1040 * We cap the size at 1 mebibyte here to prevent
1041 * allocation failures and nigh-infinite printing if the
1042 * object is extremely large.
1044 oursize
= MIN(size
, 1 << 20);
1045 arr
= kmem_alloc(oursize
, KM_SLEEP
);
1047 int err
= dmu_read(os
, object
, 0, oursize
, arr
, 0);
1049 (void) printf("got error %u from dmu_read\n", err
);
1050 kmem_free(arr
, oursize
);
1055 * Even though the allocation is already done in this code path,
1056 * we still cap the size to prevent excessive printing.
1058 oursize
= MIN(size
, 1 << 20);
1064 kmem_free(arr
, oursize
);
1065 (void) printf("\t\t[]\n");
1069 (void) printf("\t\t[%0llx", (u_longlong_t
)arr
[0]);
1070 for (size_t i
= 1; i
* sizeof (uint64_t) < oursize
; i
++) {
1072 (void) printf(", %0llx", (u_longlong_t
)arr
[i
]);
1074 (void) printf(",\n\t\t%0llx", (u_longlong_t
)arr
[i
]);
1076 if (oursize
!= size
)
1077 (void) printf(", ... ");
1078 (void) printf("]\n");
1081 kmem_free(arr
, oursize
);
1085 dump_zap(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1087 (void) data
, (void) size
;
1089 zap_attribute_t attr
;
1093 dump_zap_stats(os
, object
);
1094 (void) printf("\n");
1096 for (zap_cursor_init(&zc
, os
, object
);
1097 zap_cursor_retrieve(&zc
, &attr
) == 0;
1098 zap_cursor_advance(&zc
)) {
1099 (void) printf("\t\t%s = ", attr
.za_name
);
1100 if (attr
.za_num_integers
== 0) {
1101 (void) printf("\n");
1104 prop
= umem_zalloc(attr
.za_num_integers
*
1105 attr
.za_integer_length
, UMEM_NOFAIL
);
1106 (void) zap_lookup(os
, object
, attr
.za_name
,
1107 attr
.za_integer_length
, attr
.za_num_integers
, prop
);
1108 if (attr
.za_integer_length
== 1) {
1109 if (strcmp(attr
.za_name
,
1110 DSL_CRYPTO_KEY_MASTER_KEY
) == 0 ||
1111 strcmp(attr
.za_name
,
1112 DSL_CRYPTO_KEY_HMAC_KEY
) == 0 ||
1113 strcmp(attr
.za_name
, DSL_CRYPTO_KEY_IV
) == 0 ||
1114 strcmp(attr
.za_name
, DSL_CRYPTO_KEY_MAC
) == 0 ||
1115 strcmp(attr
.za_name
, DMU_POOL_CHECKSUM_SALT
) == 0) {
1118 for (i
= 0; i
< attr
.za_num_integers
; i
++) {
1119 (void) printf("%02x", u8
[i
]);
1122 (void) printf("%s", (char *)prop
);
1125 for (i
= 0; i
< attr
.za_num_integers
; i
++) {
1126 switch (attr
.za_integer_length
) {
1128 (void) printf("%u ",
1129 ((uint16_t *)prop
)[i
]);
1132 (void) printf("%u ",
1133 ((uint32_t *)prop
)[i
]);
1136 (void) printf("%lld ",
1137 (u_longlong_t
)((int64_t *)prop
)[i
]);
1142 (void) printf("\n");
1143 umem_free(prop
, attr
.za_num_integers
* attr
.za_integer_length
);
1145 zap_cursor_fini(&zc
);
1149 dump_bpobj(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1151 bpobj_phys_t
*bpop
= data
;
1153 char bytes
[32], comp
[32], uncomp
[32];
1155 /* make sure the output won't get truncated */
1156 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
1157 _Static_assert(sizeof (comp
) >= NN_NUMBUF_SZ
, "comp truncated");
1158 _Static_assert(sizeof (uncomp
) >= NN_NUMBUF_SZ
, "uncomp truncated");
1163 zdb_nicenum(bpop
->bpo_bytes
, bytes
, sizeof (bytes
));
1164 zdb_nicenum(bpop
->bpo_comp
, comp
, sizeof (comp
));
1165 zdb_nicenum(bpop
->bpo_uncomp
, uncomp
, sizeof (uncomp
));
1167 (void) printf("\t\tnum_blkptrs = %llu\n",
1168 (u_longlong_t
)bpop
->bpo_num_blkptrs
);
1169 (void) printf("\t\tbytes = %s\n", bytes
);
1170 if (size
>= BPOBJ_SIZE_V1
) {
1171 (void) printf("\t\tcomp = %s\n", comp
);
1172 (void) printf("\t\tuncomp = %s\n", uncomp
);
1174 if (size
>= BPOBJ_SIZE_V2
) {
1175 (void) printf("\t\tsubobjs = %llu\n",
1176 (u_longlong_t
)bpop
->bpo_subobjs
);
1177 (void) printf("\t\tnum_subobjs = %llu\n",
1178 (u_longlong_t
)bpop
->bpo_num_subobjs
);
1180 if (size
>= sizeof (*bpop
)) {
1181 (void) printf("\t\tnum_freed = %llu\n",
1182 (u_longlong_t
)bpop
->bpo_num_freed
);
1185 if (dump_opt
['d'] < 5)
1188 for (i
= 0; i
< bpop
->bpo_num_blkptrs
; i
++) {
1189 char blkbuf
[BP_SPRINTF_LEN
];
1192 int err
= dmu_read(os
, object
,
1193 i
* sizeof (bp
), sizeof (bp
), &bp
, 0);
1195 (void) printf("got error %u from dmu_read\n", err
);
1198 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
), &bp
,
1200 (void) printf("\t%s\n", blkbuf
);
1205 dump_bpobj_subobjs(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1207 (void) data
, (void) size
;
1208 dmu_object_info_t doi
;
1211 VERIFY0(dmu_object_info(os
, object
, &doi
));
1212 uint64_t *subobjs
= kmem_alloc(doi
.doi_max_offset
, KM_SLEEP
);
1214 int err
= dmu_read(os
, object
, 0, doi
.doi_max_offset
, subobjs
, 0);
1216 (void) printf("got error %u from dmu_read\n", err
);
1217 kmem_free(subobjs
, doi
.doi_max_offset
);
1221 int64_t last_nonzero
= -1;
1222 for (i
= 0; i
< doi
.doi_max_offset
/ 8; i
++) {
1223 if (subobjs
[i
] != 0)
1227 for (i
= 0; i
<= last_nonzero
; i
++) {
1228 (void) printf("\t%llu\n", (u_longlong_t
)subobjs
[i
]);
1230 kmem_free(subobjs
, doi
.doi_max_offset
);
1234 dump_ddt_zap(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1236 (void) data
, (void) size
;
1237 dump_zap_stats(os
, object
);
1238 /* contents are printed elsewhere, properly decoded */
1242 dump_sa_attrs(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1244 (void) data
, (void) size
;
1246 zap_attribute_t attr
;
1248 dump_zap_stats(os
, object
);
1249 (void) printf("\n");
1251 for (zap_cursor_init(&zc
, os
, object
);
1252 zap_cursor_retrieve(&zc
, &attr
) == 0;
1253 zap_cursor_advance(&zc
)) {
1254 (void) printf("\t\t%s = ", attr
.za_name
);
1255 if (attr
.za_num_integers
== 0) {
1256 (void) printf("\n");
1259 (void) printf(" %llx : [%d:%d:%d]\n",
1260 (u_longlong_t
)attr
.za_first_integer
,
1261 (int)ATTR_LENGTH(attr
.za_first_integer
),
1262 (int)ATTR_BSWAP(attr
.za_first_integer
),
1263 (int)ATTR_NUM(attr
.za_first_integer
));
1265 zap_cursor_fini(&zc
);
1269 dump_sa_layouts(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1271 (void) data
, (void) size
;
1273 zap_attribute_t attr
;
1274 uint16_t *layout_attrs
;
1277 dump_zap_stats(os
, object
);
1278 (void) printf("\n");
1280 for (zap_cursor_init(&zc
, os
, object
);
1281 zap_cursor_retrieve(&zc
, &attr
) == 0;
1282 zap_cursor_advance(&zc
)) {
1283 (void) printf("\t\t%s = [", attr
.za_name
);
1284 if (attr
.za_num_integers
== 0) {
1285 (void) printf("\n");
1289 VERIFY(attr
.za_integer_length
== 2);
1290 layout_attrs
= umem_zalloc(attr
.za_num_integers
*
1291 attr
.za_integer_length
, UMEM_NOFAIL
);
1293 VERIFY(zap_lookup(os
, object
, attr
.za_name
,
1294 attr
.za_integer_length
,
1295 attr
.za_num_integers
, layout_attrs
) == 0);
1297 for (i
= 0; i
!= attr
.za_num_integers
; i
++)
1298 (void) printf(" %d ", (int)layout_attrs
[i
]);
1299 (void) printf("]\n");
1300 umem_free(layout_attrs
,
1301 attr
.za_num_integers
* attr
.za_integer_length
);
1303 zap_cursor_fini(&zc
);
1307 dump_zpldir(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1309 (void) data
, (void) size
;
1311 zap_attribute_t attr
;
1312 const char *typenames
[] = {
1313 /* 0 */ "not specified",
1315 /* 2 */ "Character Device",
1316 /* 3 */ "3 (invalid)",
1317 /* 4 */ "Directory",
1318 /* 5 */ "5 (invalid)",
1319 /* 6 */ "Block Device",
1320 /* 7 */ "7 (invalid)",
1321 /* 8 */ "Regular File",
1322 /* 9 */ "9 (invalid)",
1323 /* 10 */ "Symbolic Link",
1324 /* 11 */ "11 (invalid)",
1327 /* 14 */ "Event Port",
1328 /* 15 */ "15 (invalid)",
1331 dump_zap_stats(os
, object
);
1332 (void) printf("\n");
1334 for (zap_cursor_init(&zc
, os
, object
);
1335 zap_cursor_retrieve(&zc
, &attr
) == 0;
1336 zap_cursor_advance(&zc
)) {
1337 (void) printf("\t\t%s = %lld (type: %s)\n",
1338 attr
.za_name
, ZFS_DIRENT_OBJ(attr
.za_first_integer
),
1339 typenames
[ZFS_DIRENT_TYPE(attr
.za_first_integer
)]);
1341 zap_cursor_fini(&zc
);
1345 get_dtl_refcount(vdev_t
*vd
)
1349 if (vd
->vdev_ops
->vdev_op_leaf
) {
1350 space_map_t
*sm
= vd
->vdev_dtl_sm
;
1353 sm
->sm_dbuf
->db_size
== sizeof (space_map_phys_t
))
1358 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++)
1359 refcount
+= get_dtl_refcount(vd
->vdev_child
[c
]);
1364 get_metaslab_refcount(vdev_t
*vd
)
1368 if (vd
->vdev_top
== vd
) {
1369 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
1370 space_map_t
*sm
= vd
->vdev_ms
[m
]->ms_sm
;
1373 sm
->sm_dbuf
->db_size
== sizeof (space_map_phys_t
))
1377 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++)
1378 refcount
+= get_metaslab_refcount(vd
->vdev_child
[c
]);
1384 get_obsolete_refcount(vdev_t
*vd
)
1386 uint64_t obsolete_sm_object
;
1389 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
1390 if (vd
->vdev_top
== vd
&& obsolete_sm_object
!= 0) {
1391 dmu_object_info_t doi
;
1392 VERIFY0(dmu_object_info(vd
->vdev_spa
->spa_meta_objset
,
1393 obsolete_sm_object
, &doi
));
1394 if (doi
.doi_bonus_size
== sizeof (space_map_phys_t
)) {
1398 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
1399 ASSERT3U(obsolete_sm_object
, ==, 0);
1401 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++) {
1402 refcount
+= get_obsolete_refcount(vd
->vdev_child
[c
]);
1409 get_prev_obsolete_spacemap_refcount(spa_t
*spa
)
1412 spa
->spa_condensing_indirect_phys
.scip_prev_obsolete_sm_object
;
1413 if (prev_obj
!= 0) {
1414 dmu_object_info_t doi
;
1415 VERIFY0(dmu_object_info(spa
->spa_meta_objset
, prev_obj
, &doi
));
1416 if (doi
.doi_bonus_size
== sizeof (space_map_phys_t
)) {
1424 get_checkpoint_refcount(vdev_t
*vd
)
1428 if (vd
->vdev_top
== vd
&& vd
->vdev_top_zap
!= 0 &&
1429 zap_contains(spa_meta_objset(vd
->vdev_spa
),
1430 vd
->vdev_top_zap
, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) == 0)
1433 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1434 refcount
+= get_checkpoint_refcount(vd
->vdev_child
[c
]);
1440 get_log_spacemap_refcount(spa_t
*spa
)
1442 return (avl_numnodes(&spa
->spa_sm_logs_by_txg
));
1446 verify_spacemap_refcounts(spa_t
*spa
)
1448 uint64_t expected_refcount
= 0;
1449 uint64_t actual_refcount
;
1451 (void) feature_get_refcount(spa
,
1452 &spa_feature_table
[SPA_FEATURE_SPACEMAP_HISTOGRAM
],
1453 &expected_refcount
);
1454 actual_refcount
= get_dtl_refcount(spa
->spa_root_vdev
);
1455 actual_refcount
+= get_metaslab_refcount(spa
->spa_root_vdev
);
1456 actual_refcount
+= get_obsolete_refcount(spa
->spa_root_vdev
);
1457 actual_refcount
+= get_prev_obsolete_spacemap_refcount(spa
);
1458 actual_refcount
+= get_checkpoint_refcount(spa
->spa_root_vdev
);
1459 actual_refcount
+= get_log_spacemap_refcount(spa
);
1461 if (expected_refcount
!= actual_refcount
) {
1462 (void) printf("space map refcount mismatch: expected %lld != "
1464 (longlong_t
)expected_refcount
,
1465 (longlong_t
)actual_refcount
);
1472 dump_spacemap(objset_t
*os
, space_map_t
*sm
)
1474 const char *ddata
[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1475 "INVALID", "INVALID", "INVALID", "INVALID" };
1480 (void) printf("space map object %llu:\n",
1481 (longlong_t
)sm
->sm_object
);
1482 (void) printf(" smp_length = 0x%llx\n",
1483 (longlong_t
)sm
->sm_phys
->smp_length
);
1484 (void) printf(" smp_alloc = 0x%llx\n",
1485 (longlong_t
)sm
->sm_phys
->smp_alloc
);
1487 if (dump_opt
['d'] < 6 && dump_opt
['m'] < 4)
1491 * Print out the freelist entries in both encoded and decoded form.
1493 uint8_t mapshift
= sm
->sm_shift
;
1495 uint64_t word
, entry_id
= 0;
1496 for (uint64_t offset
= 0; offset
< space_map_length(sm
);
1497 offset
+= sizeof (word
)) {
1499 VERIFY0(dmu_read(os
, space_map_object(sm
), offset
,
1500 sizeof (word
), &word
, DMU_READ_PREFETCH
));
1502 if (sm_entry_is_debug(word
)) {
1503 uint64_t de_txg
= SM_DEBUG_TXG_DECODE(word
);
1504 uint64_t de_sync_pass
= SM_DEBUG_SYNCPASS_DECODE(word
);
1507 "\t [%6llu] PADDING\n",
1508 (u_longlong_t
)entry_id
);
1511 "\t [%6llu] %s: txg %llu pass %llu\n",
1512 (u_longlong_t
)entry_id
,
1513 ddata
[SM_DEBUG_ACTION_DECODE(word
)],
1514 (u_longlong_t
)de_txg
,
1515 (u_longlong_t
)de_sync_pass
);
1523 uint64_t entry_off
, entry_run
, entry_vdev
= SM_NO_VDEVID
;
1525 if (sm_entry_is_single_word(word
)) {
1526 entry_type
= (SM_TYPE_DECODE(word
) == SM_ALLOC
) ?
1528 entry_off
= (SM_OFFSET_DECODE(word
) << mapshift
) +
1530 entry_run
= SM_RUN_DECODE(word
) << mapshift
;
1533 /* it is a two-word entry so we read another word */
1534 ASSERT(sm_entry_is_double_word(word
));
1536 uint64_t extra_word
;
1537 offset
+= sizeof (extra_word
);
1538 VERIFY0(dmu_read(os
, space_map_object(sm
), offset
,
1539 sizeof (extra_word
), &extra_word
,
1540 DMU_READ_PREFETCH
));
1542 ASSERT3U(offset
, <=, space_map_length(sm
));
1544 entry_run
= SM2_RUN_DECODE(word
) << mapshift
;
1545 entry_vdev
= SM2_VDEV_DECODE(word
);
1546 entry_type
= (SM2_TYPE_DECODE(extra_word
) == SM_ALLOC
) ?
1548 entry_off
= (SM2_OFFSET_DECODE(extra_word
) <<
1549 mapshift
) + sm
->sm_start
;
1553 (void) printf("\t [%6llu] %c range:"
1554 " %010llx-%010llx size: %06llx vdev: %06llu words: %u\n",
1555 (u_longlong_t
)entry_id
,
1556 entry_type
, (u_longlong_t
)entry_off
,
1557 (u_longlong_t
)(entry_off
+ entry_run
),
1558 (u_longlong_t
)entry_run
,
1559 (u_longlong_t
)entry_vdev
, words
);
1561 if (entry_type
== 'A')
1567 if (alloc
!= space_map_allocated(sm
)) {
1568 (void) printf("space_map_object alloc (%lld) INCONSISTENT "
1569 "with space map summary (%lld)\n",
1570 (longlong_t
)space_map_allocated(sm
), (longlong_t
)alloc
);
1575 dump_metaslab_stats(metaslab_t
*msp
)
1578 range_tree_t
*rt
= msp
->ms_allocatable
;
1579 zfs_btree_t
*t
= &msp
->ms_allocatable_by_size
;
1580 int free_pct
= range_tree_space(rt
) * 100 / msp
->ms_size
;
1582 /* max sure nicenum has enough space */
1583 _Static_assert(sizeof (maxbuf
) >= NN_NUMBUF_SZ
, "maxbuf truncated");
1585 zdb_nicenum(metaslab_largest_allocatable(msp
), maxbuf
, sizeof (maxbuf
));
1587 (void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
1588 "segments", zfs_btree_numnodes(t
), "maxsize", maxbuf
,
1589 "freepct", free_pct
);
1590 (void) printf("\tIn-memory histogram:\n");
1591 dump_histogram(rt
->rt_histogram
, RANGE_TREE_HISTOGRAM_SIZE
, 0);
1595 dump_metaslab(metaslab_t
*msp
)
1597 vdev_t
*vd
= msp
->ms_group
->mg_vd
;
1598 spa_t
*spa
= vd
->vdev_spa
;
1599 space_map_t
*sm
= msp
->ms_sm
;
1602 zdb_nicenum(msp
->ms_size
- space_map_allocated(sm
), freebuf
,
1606 "\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
1607 (u_longlong_t
)msp
->ms_id
, (u_longlong_t
)msp
->ms_start
,
1608 (u_longlong_t
)space_map_object(sm
), freebuf
);
1610 if (dump_opt
['m'] > 2 && !dump_opt
['L']) {
1611 mutex_enter(&msp
->ms_lock
);
1612 VERIFY0(metaslab_load(msp
));
1613 range_tree_stat_verify(msp
->ms_allocatable
);
1614 dump_metaslab_stats(msp
);
1615 metaslab_unload(msp
);
1616 mutex_exit(&msp
->ms_lock
);
1619 if (dump_opt
['m'] > 1 && sm
!= NULL
&&
1620 spa_feature_is_active(spa
, SPA_FEATURE_SPACEMAP_HISTOGRAM
)) {
1622 * The space map histogram represents free space in chunks
1623 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1625 (void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1626 (u_longlong_t
)msp
->ms_fragmentation
);
1627 dump_histogram(sm
->sm_phys
->smp_histogram
,
1628 SPACE_MAP_HISTOGRAM_SIZE
, sm
->sm_shift
);
1631 if (vd
->vdev_ops
== &vdev_draid_ops
)
1632 ASSERT3U(msp
->ms_size
, <=, 1ULL << vd
->vdev_ms_shift
);
1634 ASSERT3U(msp
->ms_size
, ==, 1ULL << vd
->vdev_ms_shift
);
1636 dump_spacemap(spa
->spa_meta_objset
, msp
->ms_sm
);
1638 if (spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
)) {
1639 (void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1640 (u_longlong_t
)metaslab_unflushed_txg(msp
));
1645 print_vdev_metaslab_header(vdev_t
*vd
)
1647 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
1648 const char *bias_str
= "";
1649 if (alloc_bias
== VDEV_BIAS_LOG
|| vd
->vdev_islog
) {
1650 bias_str
= VDEV_ALLOC_BIAS_LOG
;
1651 } else if (alloc_bias
== VDEV_BIAS_SPECIAL
) {
1652 bias_str
= VDEV_ALLOC_BIAS_SPECIAL
;
1653 } else if (alloc_bias
== VDEV_BIAS_DEDUP
) {
1654 bias_str
= VDEV_ALLOC_BIAS_DEDUP
;
1657 uint64_t ms_flush_data_obj
= 0;
1658 if (vd
->vdev_top_zap
!= 0) {
1659 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
),
1660 vd
->vdev_top_zap
, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
,
1661 sizeof (uint64_t), 1, &ms_flush_data_obj
);
1662 if (error
!= ENOENT
) {
1667 (void) printf("\tvdev %10llu %s",
1668 (u_longlong_t
)vd
->vdev_id
, bias_str
);
1670 if (ms_flush_data_obj
!= 0) {
1671 (void) printf(" ms_unflushed_phys object %llu",
1672 (u_longlong_t
)ms_flush_data_obj
);
1675 (void) printf("\n\t%-10s%5llu %-19s %-15s %-12s\n",
1676 "metaslabs", (u_longlong_t
)vd
->vdev_ms_count
,
1677 "offset", "spacemap", "free");
1678 (void) printf("\t%15s %19s %15s %12s\n",
1679 "---------------", "-------------------",
1680 "---------------", "------------");
1684 dump_metaslab_groups(spa_t
*spa
, boolean_t show_special
)
1686 vdev_t
*rvd
= spa
->spa_root_vdev
;
1687 metaslab_class_t
*mc
= spa_normal_class(spa
);
1688 metaslab_class_t
*smc
= spa_special_class(spa
);
1689 uint64_t fragmentation
;
1691 metaslab_class_histogram_verify(mc
);
1693 for (unsigned c
= 0; c
< rvd
->vdev_children
; c
++) {
1694 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1695 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1697 if (mg
== NULL
|| (mg
->mg_class
!= mc
&&
1698 (!show_special
|| mg
->mg_class
!= smc
)))
1701 metaslab_group_histogram_verify(mg
);
1702 mg
->mg_fragmentation
= metaslab_group_fragmentation(mg
);
1704 (void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1706 (u_longlong_t
)tvd
->vdev_id
,
1707 (u_longlong_t
)tvd
->vdev_ms_count
);
1708 if (mg
->mg_fragmentation
== ZFS_FRAG_INVALID
) {
1709 (void) printf("%3s\n", "-");
1711 (void) printf("%3llu%%\n",
1712 (u_longlong_t
)mg
->mg_fragmentation
);
1714 dump_histogram(mg
->mg_histogram
, RANGE_TREE_HISTOGRAM_SIZE
, 0);
1717 (void) printf("\tpool %s\tfragmentation", spa_name(spa
));
1718 fragmentation
= metaslab_class_fragmentation(mc
);
1719 if (fragmentation
== ZFS_FRAG_INVALID
)
1720 (void) printf("\t%3s\n", "-");
1722 (void) printf("\t%3llu%%\n", (u_longlong_t
)fragmentation
);
1723 dump_histogram(mc
->mc_histogram
, RANGE_TREE_HISTOGRAM_SIZE
, 0);
1727 print_vdev_indirect(vdev_t
*vd
)
1729 vdev_indirect_config_t
*vic
= &vd
->vdev_indirect_config
;
1730 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
1731 vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
;
1734 ASSERT3P(vib
, ==, NULL
);
1738 ASSERT3U(vdev_indirect_mapping_object(vim
), ==,
1739 vic
->vic_mapping_object
);
1740 ASSERT3U(vdev_indirect_births_object(vib
), ==,
1741 vic
->vic_births_object
);
1743 (void) printf("indirect births obj %llu:\n",
1744 (longlong_t
)vic
->vic_births_object
);
1745 (void) printf(" vib_count = %llu\n",
1746 (longlong_t
)vdev_indirect_births_count(vib
));
1747 for (uint64_t i
= 0; i
< vdev_indirect_births_count(vib
); i
++) {
1748 vdev_indirect_birth_entry_phys_t
*cur_vibe
=
1749 &vib
->vib_entries
[i
];
1750 (void) printf("\toffset %llx -> txg %llu\n",
1751 (longlong_t
)cur_vibe
->vibe_offset
,
1752 (longlong_t
)cur_vibe
->vibe_phys_birth_txg
);
1754 (void) printf("\n");
1756 (void) printf("indirect mapping obj %llu:\n",
1757 (longlong_t
)vic
->vic_mapping_object
);
1758 (void) printf(" vim_max_offset = 0x%llx\n",
1759 (longlong_t
)vdev_indirect_mapping_max_offset(vim
));
1760 (void) printf(" vim_bytes_mapped = 0x%llx\n",
1761 (longlong_t
)vdev_indirect_mapping_bytes_mapped(vim
));
1762 (void) printf(" vim_count = %llu\n",
1763 (longlong_t
)vdev_indirect_mapping_num_entries(vim
));
1765 if (dump_opt
['d'] <= 5 && dump_opt
['m'] <= 3)
1768 uint32_t *counts
= vdev_indirect_mapping_load_obsolete_counts(vim
);
1770 for (uint64_t i
= 0; i
< vdev_indirect_mapping_num_entries(vim
); i
++) {
1771 vdev_indirect_mapping_entry_phys_t
*vimep
=
1772 &vim
->vim_entries
[i
];
1773 (void) printf("\t<%llx:%llx:%llx> -> "
1774 "<%llx:%llx:%llx> (%x obsolete)\n",
1775 (longlong_t
)vd
->vdev_id
,
1776 (longlong_t
)DVA_MAPPING_GET_SRC_OFFSET(vimep
),
1777 (longlong_t
)DVA_GET_ASIZE(&vimep
->vimep_dst
),
1778 (longlong_t
)DVA_GET_VDEV(&vimep
->vimep_dst
),
1779 (longlong_t
)DVA_GET_OFFSET(&vimep
->vimep_dst
),
1780 (longlong_t
)DVA_GET_ASIZE(&vimep
->vimep_dst
),
1783 (void) printf("\n");
1785 uint64_t obsolete_sm_object
;
1786 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
1787 if (obsolete_sm_object
!= 0) {
1788 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
1789 (void) printf("obsolete space map object %llu:\n",
1790 (u_longlong_t
)obsolete_sm_object
);
1791 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
1792 ASSERT3U(space_map_object(vd
->vdev_obsolete_sm
), ==,
1793 obsolete_sm_object
);
1794 dump_spacemap(mos
, vd
->vdev_obsolete_sm
);
1795 (void) printf("\n");
1800 dump_metaslabs(spa_t
*spa
)
1802 vdev_t
*vd
, *rvd
= spa
->spa_root_vdev
;
1803 uint64_t m
, c
= 0, children
= rvd
->vdev_children
;
1805 (void) printf("\nMetaslabs:\n");
1807 if (!dump_opt
['d'] && zopt_metaslab_args
> 0) {
1808 c
= zopt_metaslab
[0];
1811 (void) fatal("bad vdev id: %llu", (u_longlong_t
)c
);
1813 if (zopt_metaslab_args
> 1) {
1814 vd
= rvd
->vdev_child
[c
];
1815 print_vdev_metaslab_header(vd
);
1817 for (m
= 1; m
< zopt_metaslab_args
; m
++) {
1818 if (zopt_metaslab
[m
] < vd
->vdev_ms_count
)
1820 vd
->vdev_ms
[zopt_metaslab
[m
]]);
1822 (void) fprintf(stderr
, "bad metaslab "
1824 (u_longlong_t
)zopt_metaslab
[m
]);
1826 (void) printf("\n");
1831 for (; c
< children
; c
++) {
1832 vd
= rvd
->vdev_child
[c
];
1833 print_vdev_metaslab_header(vd
);
1835 print_vdev_indirect(vd
);
1837 for (m
= 0; m
< vd
->vdev_ms_count
; m
++)
1838 dump_metaslab(vd
->vdev_ms
[m
]);
1839 (void) printf("\n");
1844 dump_log_spacemaps(spa_t
*spa
)
1846 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
1849 (void) printf("\nLog Space Maps in Pool:\n");
1850 for (spa_log_sm_t
*sls
= avl_first(&spa
->spa_sm_logs_by_txg
);
1851 sls
; sls
= AVL_NEXT(&spa
->spa_sm_logs_by_txg
, sls
)) {
1852 space_map_t
*sm
= NULL
;
1853 VERIFY0(space_map_open(&sm
, spa_meta_objset(spa
),
1854 sls
->sls_sm_obj
, 0, UINT64_MAX
, SPA_MINBLOCKSHIFT
));
1856 (void) printf("Log Spacemap object %llu txg %llu\n",
1857 (u_longlong_t
)sls
->sls_sm_obj
, (u_longlong_t
)sls
->sls_txg
);
1858 dump_spacemap(spa
->spa_meta_objset
, sm
);
1859 space_map_close(sm
);
1861 (void) printf("\n");
1865 dump_dde(const ddt_t
*ddt
, const ddt_entry_t
*dde
, uint64_t index
)
1867 const ddt_phys_t
*ddp
= dde
->dde_phys
;
1868 const ddt_key_t
*ddk
= &dde
->dde_key
;
1869 const char *types
[4] = { "ditto", "single", "double", "triple" };
1870 char blkbuf
[BP_SPRINTF_LEN
];
1874 for (p
= 0; p
< DDT_PHYS_TYPES
; p
++, ddp
++) {
1875 if (ddp
->ddp_phys_birth
== 0)
1877 ddt_bp_create(ddt
->ddt_checksum
, ddk
, ddp
, &blk
);
1878 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), &blk
);
1879 (void) printf("index %llx refcnt %llu %s %s\n",
1880 (u_longlong_t
)index
, (u_longlong_t
)ddp
->ddp_refcnt
,
1886 dump_dedup_ratio(const ddt_stat_t
*dds
)
1888 double rL
, rP
, rD
, D
, dedup
, compress
, copies
;
1890 if (dds
->dds_blocks
== 0)
1893 rL
= (double)dds
->dds_ref_lsize
;
1894 rP
= (double)dds
->dds_ref_psize
;
1895 rD
= (double)dds
->dds_ref_dsize
;
1896 D
= (double)dds
->dds_dsize
;
1902 (void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
1903 "dedup * compress / copies = %.2f\n\n",
1904 dedup
, compress
, copies
, dedup
* compress
/ copies
);
1908 dump_ddt(ddt_t
*ddt
, enum ddt_type type
, enum ddt_class
class)
1910 char name
[DDT_NAMELEN
];
1913 dmu_object_info_t doi
;
1914 uint64_t count
, dspace
, mspace
;
1917 error
= ddt_object_info(ddt
, type
, class, &doi
);
1919 if (error
== ENOENT
)
1923 error
= ddt_object_count(ddt
, type
, class, &count
);
1928 dspace
= doi
.doi_physical_blocks_512
<< 9;
1929 mspace
= doi
.doi_fill_count
* doi
.doi_data_block_size
;
1931 ddt_object_name(ddt
, type
, class, name
);
1933 (void) printf("%s: %llu entries, size %llu on disk, %llu in core\n",
1935 (u_longlong_t
)count
,
1936 (u_longlong_t
)(dspace
/ count
),
1937 (u_longlong_t
)(mspace
/ count
));
1939 if (dump_opt
['D'] < 3)
1942 zpool_dump_ddt(NULL
, &ddt
->ddt_histogram
[type
][class]);
1944 if (dump_opt
['D'] < 4)
1947 if (dump_opt
['D'] < 5 && class == DDT_CLASS_UNIQUE
)
1950 (void) printf("%s contents:\n\n", name
);
1952 while ((error
= ddt_object_walk(ddt
, type
, class, &walk
, &dde
)) == 0)
1953 dump_dde(ddt
, &dde
, walk
);
1955 ASSERT3U(error
, ==, ENOENT
);
1957 (void) printf("\n");
1961 dump_all_ddts(spa_t
*spa
)
1963 ddt_histogram_t ddh_total
= {{{0}}};
1964 ddt_stat_t dds_total
= {0};
1966 for (enum zio_checksum c
= 0; c
< ZIO_CHECKSUM_FUNCTIONS
; c
++) {
1967 ddt_t
*ddt
= spa
->spa_ddt
[c
];
1968 for (enum ddt_type type
= 0; type
< DDT_TYPES
; type
++) {
1969 for (enum ddt_class
class = 0; class < DDT_CLASSES
;
1971 dump_ddt(ddt
, type
, class);
1976 ddt_get_dedup_stats(spa
, &dds_total
);
1978 if (dds_total
.dds_blocks
== 0) {
1979 (void) printf("All DDTs are empty\n");
1983 (void) printf("\n");
1985 if (dump_opt
['D'] > 1) {
1986 (void) printf("DDT histogram (aggregated over all DDTs):\n");
1987 ddt_get_dedup_histogram(spa
, &ddh_total
);
1988 zpool_dump_ddt(&dds_total
, &ddh_total
);
1991 dump_dedup_ratio(&dds_total
);
1995 dump_brt(spa_t
*spa
)
1997 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_BLOCK_CLONING
)) {
1998 printf("BRT: unsupported on this pool\n");
2002 if (!spa_feature_is_active(spa
, SPA_FEATURE_BLOCK_CLONING
)) {
2003 printf("BRT: empty\n");
2007 brt_t
*brt
= spa
->spa_brt
;
2010 char count
[32], used
[32], saved
[32];
2011 zdb_nicebytes(brt_get_used(spa
), used
, sizeof (used
));
2012 zdb_nicebytes(brt_get_saved(spa
), saved
, sizeof (saved
));
2013 uint64_t ratio
= brt_get_ratio(spa
);
2014 printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used
, saved
,
2015 (u_longlong_t
)(ratio
/ 100), (u_longlong_t
)(ratio
% 100));
2017 if (dump_opt
['T'] < 2)
2020 for (uint64_t vdevid
= 0; vdevid
< brt
->brt_nvdevs
; vdevid
++) {
2021 brt_vdev_t
*brtvd
= &brt
->brt_vdevs
[vdevid
];
2025 if (!brtvd
->bv_initiated
) {
2026 printf("BRT: vdev %" PRIu64
": empty\n", vdevid
);
2030 zdb_nicenum(brtvd
->bv_totalcount
, count
, sizeof (count
));
2031 zdb_nicebytes(brtvd
->bv_usedspace
, used
, sizeof (used
));
2032 zdb_nicebytes(brtvd
->bv_savedspace
, saved
, sizeof (saved
));
2033 printf("BRT: vdev %" PRIu64
": refcnt %s; used %s; saved %s\n",
2034 vdevid
, count
, used
, saved
);
2037 if (dump_opt
['T'] < 3)
2041 printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2043 for (uint64_t vdevid
= 0; vdevid
< brt
->brt_nvdevs
; vdevid
++) {
2044 brt_vdev_t
*brtvd
= &brt
->brt_vdevs
[vdevid
];
2045 if (brtvd
== NULL
|| !brtvd
->bv_initiated
)
2050 for (zap_cursor_init(&zc
, brt
->brt_mos
, brtvd
->bv_mos_entries
);
2051 zap_cursor_retrieve(&zc
, &za
) == 0;
2052 zap_cursor_advance(&zc
)) {
2053 uint64_t offset
= *(uint64_t *)za
.za_name
;
2054 uint64_t refcnt
= za
.za_first_integer
;
2056 snprintf(dva
, sizeof (dva
), "%" PRIu64
":%llx", vdevid
,
2057 (u_longlong_t
)offset
);
2058 printf("%-16s %-10llu\n", dva
, (u_longlong_t
)refcnt
);
2060 zap_cursor_fini(&zc
);
2065 dump_dtl_seg(void *arg
, uint64_t start
, uint64_t size
)
2069 (void) printf("%s [%llu,%llu) length %llu\n",
2071 (u_longlong_t
)start
,
2072 (u_longlong_t
)(start
+ size
),
2073 (u_longlong_t
)(size
));
2077 dump_dtl(vdev_t
*vd
, int indent
)
2079 spa_t
*spa
= vd
->vdev_spa
;
2081 const char *name
[DTL_TYPES
] = { "missing", "partial", "scrub",
2085 spa_vdev_state_enter(spa
, SCL_NONE
);
2086 required
= vdev_dtl_required(vd
);
2087 (void) spa_vdev_state_exit(spa
, NULL
, 0);
2090 (void) printf("\nDirty time logs:\n\n");
2092 (void) printf("\t%*s%s [%s]\n", indent
, "",
2093 vd
->vdev_path
? vd
->vdev_path
:
2094 vd
->vdev_parent
? vd
->vdev_ops
->vdev_op_type
: spa_name(spa
),
2095 required
? "DTL-required" : "DTL-expendable");
2097 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2098 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2099 if (range_tree_space(rt
) == 0)
2101 (void) snprintf(prefix
, sizeof (prefix
), "\t%*s%s",
2102 indent
+ 2, "", name
[t
]);
2103 range_tree_walk(rt
, dump_dtl_seg
, prefix
);
2104 if (dump_opt
['d'] > 5 && vd
->vdev_children
== 0)
2105 dump_spacemap(spa
->spa_meta_objset
,
2109 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++)
2110 dump_dtl(vd
->vdev_child
[c
], indent
+ 4);
2114 dump_history(spa_t
*spa
)
2116 nvlist_t
**events
= NULL
;
2118 uint64_t resid
, len
, off
= 0;
2123 if ((buf
= malloc(SPA_OLD_MAXBLOCKSIZE
)) == NULL
) {
2124 (void) fprintf(stderr
, "%s: unable to allocate I/O buffer\n",
2130 len
= SPA_OLD_MAXBLOCKSIZE
;
2132 if ((error
= spa_history_get(spa
, &off
, &len
, buf
)) != 0) {
2133 (void) fprintf(stderr
, "Unable to read history: "
2134 "error %d\n", error
);
2139 if (zpool_history_unpack(buf
, len
, &resid
, &events
, &num
) != 0)
2145 (void) printf("\nHistory:\n");
2146 for (unsigned i
= 0; i
< num
; i
++) {
2147 boolean_t printed
= B_FALSE
;
2149 if (nvlist_exists(events
[i
], ZPOOL_HIST_TIME
)) {
2153 tsec
= fnvlist_lookup_uint64(events
[i
],
2155 (void) localtime_r(&tsec
, &t
);
2156 (void) strftime(tbuf
, sizeof (tbuf
), "%F.%T", &t
);
2161 if (nvlist_exists(events
[i
], ZPOOL_HIST_CMD
)) {
2162 (void) printf("%s %s\n", tbuf
,
2163 fnvlist_lookup_string(events
[i
], ZPOOL_HIST_CMD
));
2164 } else if (nvlist_exists(events
[i
], ZPOOL_HIST_INT_EVENT
)) {
2167 ievent
= fnvlist_lookup_uint64(events
[i
],
2168 ZPOOL_HIST_INT_EVENT
);
2169 if (ievent
>= ZFS_NUM_LEGACY_HISTORY_EVENTS
)
2172 (void) printf(" %s [internal %s txg:%ju] %s\n",
2174 zfs_history_event_names
[ievent
],
2175 fnvlist_lookup_uint64(events
[i
],
2177 fnvlist_lookup_string(events
[i
],
2178 ZPOOL_HIST_INT_STR
));
2179 } else if (nvlist_exists(events
[i
], ZPOOL_HIST_INT_NAME
)) {
2180 (void) printf("%s [txg:%ju] %s", tbuf
,
2181 fnvlist_lookup_uint64(events
[i
],
2183 fnvlist_lookup_string(events
[i
],
2184 ZPOOL_HIST_INT_NAME
));
2186 if (nvlist_exists(events
[i
], ZPOOL_HIST_DSNAME
)) {
2187 (void) printf(" %s (%llu)",
2188 fnvlist_lookup_string(events
[i
],
2190 (u_longlong_t
)fnvlist_lookup_uint64(
2195 (void) printf(" %s\n", fnvlist_lookup_string(events
[i
],
2196 ZPOOL_HIST_INT_STR
));
2197 } else if (nvlist_exists(events
[i
], ZPOOL_HIST_IOCTL
)) {
2198 (void) printf("%s ioctl %s\n", tbuf
,
2199 fnvlist_lookup_string(events
[i
],
2202 if (nvlist_exists(events
[i
], ZPOOL_HIST_INPUT_NVL
)) {
2203 (void) printf(" input:\n");
2204 dump_nvlist(fnvlist_lookup_nvlist(events
[i
],
2205 ZPOOL_HIST_INPUT_NVL
), 8);
2207 if (nvlist_exists(events
[i
], ZPOOL_HIST_OUTPUT_NVL
)) {
2208 (void) printf(" output:\n");
2209 dump_nvlist(fnvlist_lookup_nvlist(events
[i
],
2210 ZPOOL_HIST_OUTPUT_NVL
), 8);
2212 if (nvlist_exists(events
[i
], ZPOOL_HIST_ERRNO
)) {
2213 (void) printf(" errno: %lld\n",
2214 (longlong_t
)fnvlist_lookup_int64(events
[i
],
2223 if (dump_opt
['h'] > 1) {
2225 (void) printf("unrecognized record:\n");
2226 dump_nvlist(events
[i
], 2);
2233 dump_dnode(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
2235 (void) os
, (void) object
, (void) data
, (void) size
;
2239 blkid2offset(const dnode_phys_t
*dnp
, const blkptr_t
*bp
,
2240 const zbookmark_phys_t
*zb
)
2243 ASSERT(zb
->zb_level
< 0);
2244 if (zb
->zb_object
== 0)
2245 return (zb
->zb_blkid
);
2246 return (zb
->zb_blkid
* BP_GET_LSIZE(bp
));
2249 ASSERT(zb
->zb_level
>= 0);
2251 return ((zb
->zb_blkid
<<
2252 (zb
->zb_level
* (dnp
->dn_indblkshift
- SPA_BLKPTRSHIFT
))) *
2253 dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
2257 snprintf_zstd_header(spa_t
*spa
, char *blkbuf
, size_t buflen
,
2260 static abd_t
*pabd
= NULL
;
2263 zfs_zstdhdr_t zstd_hdr
;
2266 if (BP_GET_COMPRESS(bp
) != ZIO_COMPRESS_ZSTD
)
2272 if (BP_IS_EMBEDDED(bp
)) {
2273 buf
= malloc(SPA_MAXBLOCKSIZE
);
2275 (void) fprintf(stderr
, "out of memory\n");
2278 decode_embedded_bp_compressed(bp
, buf
);
2279 memcpy(&zstd_hdr
, buf
, sizeof (zstd_hdr
));
2281 zstd_hdr
.c_len
= BE_32(zstd_hdr
.c_len
);
2282 zstd_hdr
.raw_version_level
= BE_32(zstd_hdr
.raw_version_level
);
2283 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2284 buflen
- strlen(blkbuf
),
2285 " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2286 zstd_hdr
.c_len
, zfs_get_hdrversion(&zstd_hdr
),
2287 zfs_get_hdrlevel(&zstd_hdr
));
2292 pabd
= abd_alloc_for_io(SPA_MAXBLOCKSIZE
, B_FALSE
);
2293 zio
= zio_root(spa
, NULL
, NULL
, 0);
2295 /* Decrypt but don't decompress so we can read the compression header */
2296 zio_nowait(zio_read(zio
, spa
, bp
, pabd
, BP_GET_PSIZE(bp
), NULL
, NULL
,
2297 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW_COMPRESS
,
2299 error
= zio_wait(zio
);
2301 (void) fprintf(stderr
, "read failed: %d\n", error
);
2304 buf
= abd_borrow_buf_copy(pabd
, BP_GET_LSIZE(bp
));
2305 memcpy(&zstd_hdr
, buf
, sizeof (zstd_hdr
));
2306 zstd_hdr
.c_len
= BE_32(zstd_hdr
.c_len
);
2307 zstd_hdr
.raw_version_level
= BE_32(zstd_hdr
.raw_version_level
);
2309 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2310 buflen
- strlen(blkbuf
),
2311 " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2312 zstd_hdr
.c_len
, zfs_get_hdrversion(&zstd_hdr
),
2313 zfs_get_hdrlevel(&zstd_hdr
));
2315 abd_return_buf_copy(pabd
, buf
, BP_GET_LSIZE(bp
));
2319 snprintf_blkptr_compact(char *blkbuf
, size_t buflen
, const blkptr_t
*bp
,
2322 const dva_t
*dva
= bp
->blk_dva
;
2323 int ndvas
= dump_opt
['d'] > 5 ? BP_GET_NDVAS(bp
) : 1;
2326 if (dump_opt
['b'] >= 6) {
2327 snprintf_blkptr(blkbuf
, buflen
, bp
);
2329 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2330 buflen
- strlen(blkbuf
), " %s", "FREE");
2335 if (BP_IS_EMBEDDED(bp
)) {
2336 (void) sprintf(blkbuf
,
2337 "EMBEDDED et=%u %llxL/%llxP B=%llu",
2338 (int)BPE_GET_ETYPE(bp
),
2339 (u_longlong_t
)BPE_GET_LSIZE(bp
),
2340 (u_longlong_t
)BPE_GET_PSIZE(bp
),
2341 (u_longlong_t
)bp
->blk_birth
);
2347 for (i
= 0; i
< ndvas
; i
++)
2348 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2349 buflen
- strlen(blkbuf
), "%llu:%llx:%llx ",
2350 (u_longlong_t
)DVA_GET_VDEV(&dva
[i
]),
2351 (u_longlong_t
)DVA_GET_OFFSET(&dva
[i
]),
2352 (u_longlong_t
)DVA_GET_ASIZE(&dva
[i
]));
2354 if (BP_IS_HOLE(bp
)) {
2355 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2356 buflen
- strlen(blkbuf
),
2358 (u_longlong_t
)BP_GET_LSIZE(bp
),
2359 (u_longlong_t
)bp
->blk_birth
);
2361 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2362 buflen
- strlen(blkbuf
),
2363 "%llxL/%llxP F=%llu B=%llu/%llu",
2364 (u_longlong_t
)BP_GET_LSIZE(bp
),
2365 (u_longlong_t
)BP_GET_PSIZE(bp
),
2366 (u_longlong_t
)BP_GET_FILL(bp
),
2367 (u_longlong_t
)bp
->blk_birth
,
2368 (u_longlong_t
)BP_PHYSICAL_BIRTH(bp
));
2370 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2371 buflen
- strlen(blkbuf
), " %s", "FREE");
2372 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2373 buflen
- strlen(blkbuf
),
2374 " cksum=%016llx:%016llx:%016llx:%016llx",
2375 (u_longlong_t
)bp
->blk_cksum
.zc_word
[0],
2376 (u_longlong_t
)bp
->blk_cksum
.zc_word
[1],
2377 (u_longlong_t
)bp
->blk_cksum
.zc_word
[2],
2378 (u_longlong_t
)bp
->blk_cksum
.zc_word
[3]);
2383 print_indirect(spa_t
*spa
, blkptr_t
*bp
, const zbookmark_phys_t
*zb
,
2384 const dnode_phys_t
*dnp
)
2386 char blkbuf
[BP_SPRINTF_LEN
];
2389 if (!BP_IS_EMBEDDED(bp
)) {
2390 ASSERT3U(BP_GET_TYPE(bp
), ==, dnp
->dn_type
);
2391 ASSERT3U(BP_GET_LEVEL(bp
), ==, zb
->zb_level
);
2394 (void) printf("%16llx ", (u_longlong_t
)blkid2offset(dnp
, bp
, zb
));
2396 ASSERT(zb
->zb_level
>= 0);
2398 for (l
= dnp
->dn_nlevels
- 1; l
>= -1; l
--) {
2399 if (l
== zb
->zb_level
) {
2400 (void) printf("L%llx", (u_longlong_t
)zb
->zb_level
);
2406 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
), bp
, B_FALSE
);
2407 if (dump_opt
['Z'] && BP_GET_COMPRESS(bp
) == ZIO_COMPRESS_ZSTD
)
2408 snprintf_zstd_header(spa
, blkbuf
, sizeof (blkbuf
), bp
);
2409 (void) printf("%s\n", blkbuf
);
2413 visit_indirect(spa_t
*spa
, const dnode_phys_t
*dnp
,
2414 blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
2418 if (bp
->blk_birth
== 0)
2421 print_indirect(spa
, bp
, zb
, dnp
);
2423 if (BP_GET_LEVEL(bp
) > 0 && !BP_IS_HOLE(bp
)) {
2424 arc_flags_t flags
= ARC_FLAG_WAIT
;
2427 int epb
= BP_GET_LSIZE(bp
) >> SPA_BLKPTRSHIFT
;
2430 ASSERT(!BP_IS_REDACTED(bp
));
2432 err
= arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &buf
,
2433 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
2436 ASSERT(buf
->b_data
);
2438 /* recursively visit blocks below this */
2440 for (i
= 0; i
< epb
; i
++, cbp
++) {
2441 zbookmark_phys_t czb
;
2443 SET_BOOKMARK(&czb
, zb
->zb_objset
, zb
->zb_object
,
2445 zb
->zb_blkid
* epb
+ i
);
2446 err
= visit_indirect(spa
, dnp
, cbp
, &czb
);
2449 fill
+= BP_GET_FILL(cbp
);
2452 ASSERT3U(fill
, ==, BP_GET_FILL(bp
));
2453 arc_buf_destroy(buf
, &buf
);
2460 dump_indirect(dnode_t
*dn
)
2462 dnode_phys_t
*dnp
= dn
->dn_phys
;
2463 zbookmark_phys_t czb
;
2465 (void) printf("Indirect blocks:\n");
2467 SET_BOOKMARK(&czb
, dmu_objset_id(dn
->dn_objset
),
2468 dn
->dn_object
, dnp
->dn_nlevels
- 1, 0);
2469 for (int j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
2471 (void) visit_indirect(dmu_objset_spa(dn
->dn_objset
), dnp
,
2472 &dnp
->dn_blkptr
[j
], &czb
);
2475 (void) printf("\n");
2479 dump_dsl_dir(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
2481 (void) os
, (void) object
;
2482 dsl_dir_phys_t
*dd
= data
;
2486 /* make sure nicenum has enough space */
2487 _Static_assert(sizeof (nice
) >= NN_NUMBUF_SZ
, "nice truncated");
2492 ASSERT3U(size
, >=, sizeof (dsl_dir_phys_t
));
2494 crtime
= dd
->dd_creation_time
;
2495 (void) printf("\t\tcreation_time = %s", ctime(&crtime
));
2496 (void) printf("\t\thead_dataset_obj = %llu\n",
2497 (u_longlong_t
)dd
->dd_head_dataset_obj
);
2498 (void) printf("\t\tparent_dir_obj = %llu\n",
2499 (u_longlong_t
)dd
->dd_parent_obj
);
2500 (void) printf("\t\torigin_obj = %llu\n",
2501 (u_longlong_t
)dd
->dd_origin_obj
);
2502 (void) printf("\t\tchild_dir_zapobj = %llu\n",
2503 (u_longlong_t
)dd
->dd_child_dir_zapobj
);
2504 zdb_nicenum(dd
->dd_used_bytes
, nice
, sizeof (nice
));
2505 (void) printf("\t\tused_bytes = %s\n", nice
);
2506 zdb_nicenum(dd
->dd_compressed_bytes
, nice
, sizeof (nice
));
2507 (void) printf("\t\tcompressed_bytes = %s\n", nice
);
2508 zdb_nicenum(dd
->dd_uncompressed_bytes
, nice
, sizeof (nice
));
2509 (void) printf("\t\tuncompressed_bytes = %s\n", nice
);
2510 zdb_nicenum(dd
->dd_quota
, nice
, sizeof (nice
));
2511 (void) printf("\t\tquota = %s\n", nice
);
2512 zdb_nicenum(dd
->dd_reserved
, nice
, sizeof (nice
));
2513 (void) printf("\t\treserved = %s\n", nice
);
2514 (void) printf("\t\tprops_zapobj = %llu\n",
2515 (u_longlong_t
)dd
->dd_props_zapobj
);
2516 (void) printf("\t\tdeleg_zapobj = %llu\n",
2517 (u_longlong_t
)dd
->dd_deleg_zapobj
);
2518 (void) printf("\t\tflags = %llx\n",
2519 (u_longlong_t
)dd
->dd_flags
);
2522 zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2524 (void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2531 (void) printf("\t\tclones = %llu\n",
2532 (u_longlong_t
)dd
->dd_clones
);
2536 dump_dsl_dataset(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
2538 (void) os
, (void) object
;
2539 dsl_dataset_phys_t
*ds
= data
;
2541 char used
[32], compressed
[32], uncompressed
[32], unique
[32];
2542 char blkbuf
[BP_SPRINTF_LEN
];
2544 /* make sure nicenum has enough space */
2545 _Static_assert(sizeof (used
) >= NN_NUMBUF_SZ
, "used truncated");
2546 _Static_assert(sizeof (compressed
) >= NN_NUMBUF_SZ
,
2547 "compressed truncated");
2548 _Static_assert(sizeof (uncompressed
) >= NN_NUMBUF_SZ
,
2549 "uncompressed truncated");
2550 _Static_assert(sizeof (unique
) >= NN_NUMBUF_SZ
, "unique truncated");
2555 ASSERT(size
== sizeof (*ds
));
2556 crtime
= ds
->ds_creation_time
;
2557 zdb_nicenum(ds
->ds_referenced_bytes
, used
, sizeof (used
));
2558 zdb_nicenum(ds
->ds_compressed_bytes
, compressed
, sizeof (compressed
));
2559 zdb_nicenum(ds
->ds_uncompressed_bytes
, uncompressed
,
2560 sizeof (uncompressed
));
2561 zdb_nicenum(ds
->ds_unique_bytes
, unique
, sizeof (unique
));
2562 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), &ds
->ds_bp
);
2564 (void) printf("\t\tdir_obj = %llu\n",
2565 (u_longlong_t
)ds
->ds_dir_obj
);
2566 (void) printf("\t\tprev_snap_obj = %llu\n",
2567 (u_longlong_t
)ds
->ds_prev_snap_obj
);
2568 (void) printf("\t\tprev_snap_txg = %llu\n",
2569 (u_longlong_t
)ds
->ds_prev_snap_txg
);
2570 (void) printf("\t\tnext_snap_obj = %llu\n",
2571 (u_longlong_t
)ds
->ds_next_snap_obj
);
2572 (void) printf("\t\tsnapnames_zapobj = %llu\n",
2573 (u_longlong_t
)ds
->ds_snapnames_zapobj
);
2574 (void) printf("\t\tnum_children = %llu\n",
2575 (u_longlong_t
)ds
->ds_num_children
);
2576 (void) printf("\t\tuserrefs_obj = %llu\n",
2577 (u_longlong_t
)ds
->ds_userrefs_obj
);
2578 (void) printf("\t\tcreation_time = %s", ctime(&crtime
));
2579 (void) printf("\t\tcreation_txg = %llu\n",
2580 (u_longlong_t
)ds
->ds_creation_txg
);
2581 (void) printf("\t\tdeadlist_obj = %llu\n",
2582 (u_longlong_t
)ds
->ds_deadlist_obj
);
2583 (void) printf("\t\tused_bytes = %s\n", used
);
2584 (void) printf("\t\tcompressed_bytes = %s\n", compressed
);
2585 (void) printf("\t\tuncompressed_bytes = %s\n", uncompressed
);
2586 (void) printf("\t\tunique = %s\n", unique
);
2587 (void) printf("\t\tfsid_guid = %llu\n",
2588 (u_longlong_t
)ds
->ds_fsid_guid
);
2589 (void) printf("\t\tguid = %llu\n",
2590 (u_longlong_t
)ds
->ds_guid
);
2591 (void) printf("\t\tflags = %llx\n",
2592 (u_longlong_t
)ds
->ds_flags
);
2593 (void) printf("\t\tnext_clones_obj = %llu\n",
2594 (u_longlong_t
)ds
->ds_next_clones_obj
);
2595 (void) printf("\t\tprops_obj = %llu\n",
2596 (u_longlong_t
)ds
->ds_props_obj
);
2597 (void) printf("\t\tbp = %s\n", blkbuf
);
2601 dump_bptree_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
2603 (void) arg
, (void) tx
;
2604 char blkbuf
[BP_SPRINTF_LEN
];
2606 if (bp
->blk_birth
!= 0) {
2607 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
2608 (void) printf("\t%s\n", blkbuf
);
2614 dump_bptree(objset_t
*os
, uint64_t obj
, const char *name
)
2620 /* make sure nicenum has enough space */
2621 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
2623 if (dump_opt
['d'] < 3)
2626 VERIFY3U(0, ==, dmu_bonus_hold(os
, obj
, FTAG
, &db
));
2628 zdb_nicenum(bt
->bt_bytes
, bytes
, sizeof (bytes
));
2629 (void) printf("\n %s: %llu datasets, %s\n",
2630 name
, (unsigned long long)(bt
->bt_end
- bt
->bt_begin
), bytes
);
2631 dmu_buf_rele(db
, FTAG
);
2633 if (dump_opt
['d'] < 5)
2636 (void) printf("\n");
2638 (void) bptree_iterate(os
, obj
, B_FALSE
, dump_bptree_cb
, NULL
, NULL
);
2642 dump_bpobj_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
, dmu_tx_t
*tx
)
2644 (void) arg
, (void) tx
;
2645 char blkbuf
[BP_SPRINTF_LEN
];
2647 ASSERT(bp
->blk_birth
!= 0);
2648 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
), bp
, bp_freed
);
2649 (void) printf("\t%s\n", blkbuf
);
2654 dump_full_bpobj(bpobj_t
*bpo
, const char *name
, int indent
)
2661 /* make sure nicenum has enough space */
2662 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
2663 _Static_assert(sizeof (comp
) >= NN_NUMBUF_SZ
, "comp truncated");
2664 _Static_assert(sizeof (uncomp
) >= NN_NUMBUF_SZ
, "uncomp truncated");
2666 if (dump_opt
['d'] < 3)
2669 zdb_nicenum(bpo
->bpo_phys
->bpo_bytes
, bytes
, sizeof (bytes
));
2670 if (bpo
->bpo_havesubobj
&& bpo
->bpo_phys
->bpo_subobjs
!= 0) {
2671 zdb_nicenum(bpo
->bpo_phys
->bpo_comp
, comp
, sizeof (comp
));
2672 zdb_nicenum(bpo
->bpo_phys
->bpo_uncomp
, uncomp
, sizeof (uncomp
));
2673 if (bpo
->bpo_havefreed
) {
2674 (void) printf(" %*s: object %llu, %llu local "
2675 "blkptrs, %llu freed, %llu subobjs in object %llu, "
2676 "%s (%s/%s comp)\n",
2678 (u_longlong_t
)bpo
->bpo_object
,
2679 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2680 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_freed
,
2681 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_subobjs
,
2682 (u_longlong_t
)bpo
->bpo_phys
->bpo_subobjs
,
2683 bytes
, comp
, uncomp
);
2685 (void) printf(" %*s: object %llu, %llu local "
2686 "blkptrs, %llu subobjs in object %llu, "
2687 "%s (%s/%s comp)\n",
2689 (u_longlong_t
)bpo
->bpo_object
,
2690 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2691 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_subobjs
,
2692 (u_longlong_t
)bpo
->bpo_phys
->bpo_subobjs
,
2693 bytes
, comp
, uncomp
);
2696 for (i
= 0; i
< bpo
->bpo_phys
->bpo_num_subobjs
; i
++) {
2700 VERIFY0(dmu_read(bpo
->bpo_os
,
2701 bpo
->bpo_phys
->bpo_subobjs
,
2702 i
* sizeof (subobj
), sizeof (subobj
), &subobj
, 0));
2703 error
= bpobj_open(&subbpo
, bpo
->bpo_os
, subobj
);
2705 (void) printf("ERROR %u while trying to open "
2707 error
, (u_longlong_t
)subobj
);
2710 dump_full_bpobj(&subbpo
, "subobj", indent
+ 1);
2711 bpobj_close(&subbpo
);
2714 if (bpo
->bpo_havefreed
) {
2715 (void) printf(" %*s: object %llu, %llu blkptrs, "
2718 (u_longlong_t
)bpo
->bpo_object
,
2719 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2720 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_freed
,
2723 (void) printf(" %*s: object %llu, %llu blkptrs, "
2726 (u_longlong_t
)bpo
->bpo_object
,
2727 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2732 if (dump_opt
['d'] < 5)
2737 (void) bpobj_iterate_nofree(bpo
, dump_bpobj_cb
, NULL
, NULL
);
2738 (void) printf("\n");
2743 dump_bookmark(dsl_pool_t
*dp
, char *name
, boolean_t print_redact
,
2744 boolean_t print_list
)
2747 zfs_bookmark_phys_t prop
;
2748 objset_t
*mos
= dp
->dp_spa
->spa_meta_objset
;
2749 err
= dsl_bookmark_lookup(dp
, name
, NULL
, &prop
);
2755 (void) printf("\t#%s: ", strchr(name
, '#') + 1);
2756 (void) printf("{guid: %llx creation_txg: %llu creation_time: "
2757 "%llu redaction_obj: %llu}\n", (u_longlong_t
)prop
.zbm_guid
,
2758 (u_longlong_t
)prop
.zbm_creation_txg
,
2759 (u_longlong_t
)prop
.zbm_creation_time
,
2760 (u_longlong_t
)prop
.zbm_redaction_obj
);
2762 IMPLY(print_list
, print_redact
);
2763 if (!print_redact
|| prop
.zbm_redaction_obj
== 0)
2766 redaction_list_t
*rl
;
2767 VERIFY0(dsl_redaction_list_hold_obj(dp
,
2768 prop
.zbm_redaction_obj
, FTAG
, &rl
));
2770 redaction_list_phys_t
*rlp
= rl
->rl_phys
;
2771 (void) printf("\tRedacted:\n\t\tProgress: ");
2772 if (rlp
->rlp_last_object
!= UINT64_MAX
||
2773 rlp
->rlp_last_blkid
!= UINT64_MAX
) {
2774 (void) printf("%llu %llu (incomplete)\n",
2775 (u_longlong_t
)rlp
->rlp_last_object
,
2776 (u_longlong_t
)rlp
->rlp_last_blkid
);
2778 (void) printf("complete\n");
2780 (void) printf("\t\tSnapshots: [");
2781 for (unsigned int i
= 0; i
< rlp
->rlp_num_snaps
; i
++) {
2783 (void) printf(", ");
2784 (void) printf("%0llu",
2785 (u_longlong_t
)rlp
->rlp_snaps
[i
]);
2787 (void) printf("]\n\t\tLength: %llu\n",
2788 (u_longlong_t
)rlp
->rlp_num_entries
);
2791 dsl_redaction_list_rele(rl
, FTAG
);
2795 if (rlp
->rlp_num_entries
== 0) {
2796 dsl_redaction_list_rele(rl
, FTAG
);
2797 (void) printf("\t\tRedaction List: []\n\n");
2801 redact_block_phys_t
*rbp_buf
;
2803 dmu_object_info_t doi
;
2805 VERIFY0(dmu_object_info(mos
, prop
.zbm_redaction_obj
, &doi
));
2806 size
= doi
.doi_max_offset
;
2807 rbp_buf
= kmem_alloc(size
, KM_SLEEP
);
2809 err
= dmu_read(mos
, prop
.zbm_redaction_obj
, 0, size
,
2812 dsl_redaction_list_rele(rl
, FTAG
);
2813 kmem_free(rbp_buf
, size
);
2817 (void) printf("\t\tRedaction List: [{object: %llx, offset: "
2818 "%llx, blksz: %x, count: %llx}",
2819 (u_longlong_t
)rbp_buf
[0].rbp_object
,
2820 (u_longlong_t
)rbp_buf
[0].rbp_blkid
,
2821 (uint_t
)(redact_block_get_size(&rbp_buf
[0])),
2822 (u_longlong_t
)redact_block_get_count(&rbp_buf
[0]));
2824 for (size_t i
= 1; i
< rlp
->rlp_num_entries
; i
++) {
2825 (void) printf(",\n\t\t{object: %llx, offset: %llx, "
2826 "blksz: %x, count: %llx}",
2827 (u_longlong_t
)rbp_buf
[i
].rbp_object
,
2828 (u_longlong_t
)rbp_buf
[i
].rbp_blkid
,
2829 (uint_t
)(redact_block_get_size(&rbp_buf
[i
])),
2830 (u_longlong_t
)redact_block_get_count(&rbp_buf
[i
]));
2832 dsl_redaction_list_rele(rl
, FTAG
);
2833 kmem_free(rbp_buf
, size
);
2834 (void) printf("]\n\n");
2839 dump_bookmarks(objset_t
*os
, int verbosity
)
2842 zap_attribute_t attr
;
2843 dsl_dataset_t
*ds
= dmu_objset_ds(os
);
2844 dsl_pool_t
*dp
= spa_get_dsl(os
->os_spa
);
2845 objset_t
*mos
= os
->os_spa
->spa_meta_objset
;
2848 dsl_pool_config_enter(dp
, FTAG
);
2850 for (zap_cursor_init(&zc
, mos
, ds
->ds_bookmarks_obj
);
2851 zap_cursor_retrieve(&zc
, &attr
) == 0;
2852 zap_cursor_advance(&zc
)) {
2853 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
2854 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
2856 dmu_objset_name(os
, osname
);
2857 len
= snprintf(buf
, sizeof (buf
), "%s#%s", osname
,
2859 VERIFY3S(len
, <, ZFS_MAX_DATASET_NAME_LEN
);
2860 (void) dump_bookmark(dp
, buf
, verbosity
>= 5, verbosity
>= 6);
2862 zap_cursor_fini(&zc
);
2863 dsl_pool_config_exit(dp
, FTAG
);
2867 bpobj_count_refd(bpobj_t
*bpo
)
2869 mos_obj_refd(bpo
->bpo_object
);
2871 if (bpo
->bpo_havesubobj
&& bpo
->bpo_phys
->bpo_subobjs
!= 0) {
2872 mos_obj_refd(bpo
->bpo_phys
->bpo_subobjs
);
2873 for (uint64_t i
= 0; i
< bpo
->bpo_phys
->bpo_num_subobjs
; i
++) {
2877 VERIFY0(dmu_read(bpo
->bpo_os
,
2878 bpo
->bpo_phys
->bpo_subobjs
,
2879 i
* sizeof (subobj
), sizeof (subobj
), &subobj
, 0));
2880 error
= bpobj_open(&subbpo
, bpo
->bpo_os
, subobj
);
2882 (void) printf("ERROR %u while trying to open "
2884 error
, (u_longlong_t
)subobj
);
2887 bpobj_count_refd(&subbpo
);
2888 bpobj_close(&subbpo
);
2894 dsl_deadlist_entry_count_refd(void *arg
, dsl_deadlist_entry_t
*dle
)
2897 uint64_t empty_bpobj
= spa
->spa_dsl_pool
->dp_empty_bpobj
;
2898 if (dle
->dle_bpobj
.bpo_object
!= empty_bpobj
)
2899 bpobj_count_refd(&dle
->dle_bpobj
);
2904 dsl_deadlist_entry_dump(void *arg
, dsl_deadlist_entry_t
*dle
)
2906 ASSERT(arg
== NULL
);
2907 if (dump_opt
['d'] >= 5) {
2909 (void) snprintf(buf
, sizeof (buf
),
2910 "mintxg %llu -> obj %llu",
2911 (longlong_t
)dle
->dle_mintxg
,
2912 (longlong_t
)dle
->dle_bpobj
.bpo_object
);
2914 dump_full_bpobj(&dle
->dle_bpobj
, buf
, 0);
2916 (void) printf("mintxg %llu -> obj %llu\n",
2917 (longlong_t
)dle
->dle_mintxg
,
2918 (longlong_t
)dle
->dle_bpobj
.bpo_object
);
2924 dump_blkptr_list(dsl_deadlist_t
*dl
, const char *name
)
2930 spa_t
*spa
= dmu_objset_spa(dl
->dl_os
);
2931 uint64_t empty_bpobj
= spa
->spa_dsl_pool
->dp_empty_bpobj
;
2933 if (dl
->dl_oldfmt
) {
2934 if (dl
->dl_bpobj
.bpo_object
!= empty_bpobj
)
2935 bpobj_count_refd(&dl
->dl_bpobj
);
2937 mos_obj_refd(dl
->dl_object
);
2938 dsl_deadlist_iterate(dl
, dsl_deadlist_entry_count_refd
, spa
);
2941 /* make sure nicenum has enough space */
2942 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
2943 _Static_assert(sizeof (comp
) >= NN_NUMBUF_SZ
, "comp truncated");
2944 _Static_assert(sizeof (uncomp
) >= NN_NUMBUF_SZ
, "uncomp truncated");
2945 _Static_assert(sizeof (entries
) >= NN_NUMBUF_SZ
, "entries truncated");
2947 if (dump_opt
['d'] < 3)
2950 if (dl
->dl_oldfmt
) {
2951 dump_full_bpobj(&dl
->dl_bpobj
, "old-format deadlist", 0);
2955 zdb_nicenum(dl
->dl_phys
->dl_used
, bytes
, sizeof (bytes
));
2956 zdb_nicenum(dl
->dl_phys
->dl_comp
, comp
, sizeof (comp
));
2957 zdb_nicenum(dl
->dl_phys
->dl_uncomp
, uncomp
, sizeof (uncomp
));
2958 zdb_nicenum(avl_numnodes(&dl
->dl_tree
), entries
, sizeof (entries
));
2959 (void) printf("\n %s: %s (%s/%s comp), %s entries\n",
2960 name
, bytes
, comp
, uncomp
, entries
);
2962 if (dump_opt
['d'] < 4)
2965 (void) putchar('\n');
2967 dsl_deadlist_iterate(dl
, dsl_deadlist_entry_dump
, NULL
);
2971 verify_dd_livelist(objset_t
*os
)
2973 uint64_t ll_used
, used
, ll_comp
, comp
, ll_uncomp
, uncomp
;
2974 dsl_pool_t
*dp
= spa_get_dsl(os
->os_spa
);
2975 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
2977 ASSERT(!dmu_objset_is_snapshot(os
));
2978 if (!dsl_deadlist_is_open(&dd
->dd_livelist
))
2981 /* Iterate through the livelist to check for duplicates */
2982 dsl_deadlist_iterate(&dd
->dd_livelist
, sublivelist_verify_lightweight
,
2985 dsl_pool_config_enter(dp
, FTAG
);
2986 dsl_deadlist_space(&dd
->dd_livelist
, &ll_used
,
2987 &ll_comp
, &ll_uncomp
);
2989 dsl_dataset_t
*origin_ds
;
2990 ASSERT(dsl_pool_config_held(dp
));
2991 VERIFY0(dsl_dataset_hold_obj(dp
,
2992 dsl_dir_phys(dd
)->dd_origin_obj
, FTAG
, &origin_ds
));
2993 VERIFY0(dsl_dataset_space_written(origin_ds
, os
->os_dsl_dataset
,
2994 &used
, &comp
, &uncomp
));
2995 dsl_dataset_rele(origin_ds
, FTAG
);
2996 dsl_pool_config_exit(dp
, FTAG
);
2998 * It's possible that the dataset's uncomp space is larger than the
2999 * livelist's because livelists do not track embedded block pointers
3001 if (used
!= ll_used
|| comp
!= ll_comp
|| uncomp
< ll_uncomp
) {
3002 char nice_used
[32], nice_comp
[32], nice_uncomp
[32];
3003 (void) printf("Discrepancy in space accounting:\n");
3004 zdb_nicenum(used
, nice_used
, sizeof (nice_used
));
3005 zdb_nicenum(comp
, nice_comp
, sizeof (nice_comp
));
3006 zdb_nicenum(uncomp
, nice_uncomp
, sizeof (nice_uncomp
));
3007 (void) printf("dir: used %s, comp %s, uncomp %s\n",
3008 nice_used
, nice_comp
, nice_uncomp
);
3009 zdb_nicenum(ll_used
, nice_used
, sizeof (nice_used
));
3010 zdb_nicenum(ll_comp
, nice_comp
, sizeof (nice_comp
));
3011 zdb_nicenum(ll_uncomp
, nice_uncomp
, sizeof (nice_uncomp
));
3012 (void) printf("livelist: used %s, comp %s, uncomp %s\n",
3013 nice_used
, nice_comp
, nice_uncomp
);
3019 static char *key_material
= NULL
;
3022 zdb_derive_key(dsl_dir_t
*dd
, uint8_t *key_out
)
3024 uint64_t keyformat
, salt
, iters
;
3028 VERIFY0(zap_lookup(dd
->dd_pool
->dp_meta_objset
, dd
->dd_crypto_obj
,
3029 zfs_prop_to_name(ZFS_PROP_KEYFORMAT
), sizeof (uint64_t),
3032 switch (keyformat
) {
3033 case ZFS_KEYFORMAT_HEX
:
3034 for (i
= 0; i
< WRAPPING_KEY_LEN
* 2; i
+= 2) {
3035 if (!isxdigit(key_material
[i
]) ||
3036 !isxdigit(key_material
[i
+1]))
3038 if (sscanf(&key_material
[i
], "%02hhx", &c
) != 1)
3044 case ZFS_KEYFORMAT_PASSPHRASE
:
3045 VERIFY0(zap_lookup(dd
->dd_pool
->dp_meta_objset
,
3046 dd
->dd_crypto_obj
, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT
),
3047 sizeof (uint64_t), 1, &salt
));
3048 VERIFY0(zap_lookup(dd
->dd_pool
->dp_meta_objset
,
3049 dd
->dd_crypto_obj
, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS
),
3050 sizeof (uint64_t), 1, &iters
));
3052 if (PKCS5_PBKDF2_HMAC_SHA1(key_material
, strlen(key_material
),
3053 ((uint8_t *)&salt
), sizeof (uint64_t), iters
,
3054 WRAPPING_KEY_LEN
, key_out
) != 1)
3060 fatal("no support for key format %u\n",
3061 (unsigned int) keyformat
);
3067 static char encroot
[ZFS_MAX_DATASET_NAME_LEN
];
3068 static boolean_t key_loaded
= B_FALSE
;
3071 zdb_load_key(objset_t
*os
)
3074 dsl_dir_t
*dd
, *rdd
;
3075 uint8_t key
[WRAPPING_KEY_LEN
];
3079 dp
= spa_get_dsl(os
->os_spa
);
3080 dd
= os
->os_dsl_dataset
->ds_dir
;
3082 dsl_pool_config_enter(dp
, FTAG
);
3083 VERIFY0(zap_lookup(dd
->dd_pool
->dp_meta_objset
, dd
->dd_crypto_obj
,
3084 DSL_CRYPTO_KEY_ROOT_DDOBJ
, sizeof (uint64_t), 1, &rddobj
));
3085 VERIFY0(dsl_dir_hold_obj(dd
->dd_pool
, rddobj
, NULL
, FTAG
, &rdd
));
3086 dsl_dir_name(rdd
, encroot
);
3087 dsl_dir_rele(rdd
, FTAG
);
3089 if (!zdb_derive_key(dd
, key
))
3090 fatal("couldn't derive encryption key");
3092 dsl_pool_config_exit(dp
, FTAG
);
3094 ASSERT3U(dsl_dataset_get_keystatus(dd
), ==, ZFS_KEYSTATUS_UNAVAILABLE
);
3096 dsl_crypto_params_t
*dcp
;
3097 nvlist_t
*crypto_args
;
3099 crypto_args
= fnvlist_alloc();
3100 fnvlist_add_uint8_array(crypto_args
, "wkeydata",
3101 (uint8_t *)key
, WRAPPING_KEY_LEN
);
3102 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE
,
3103 NULL
, crypto_args
, &dcp
));
3104 err
= spa_keystore_load_wkey(encroot
, dcp
, B_FALSE
);
3106 dsl_crypto_params_free(dcp
, (err
!= 0));
3107 fnvlist_free(crypto_args
);
3111 "couldn't load encryption key for %s: %s",
3112 encroot
, err
== ZFS_ERR_CRYPTO_NOTSUP
?
3113 "crypto params not supported" : strerror(err
));
3115 ASSERT3U(dsl_dataset_get_keystatus(dd
), ==, ZFS_KEYSTATUS_AVAILABLE
);
3117 printf("Unlocked encryption root: %s\n", encroot
);
3118 key_loaded
= B_TRUE
;
3122 zdb_unload_key(void)
3127 VERIFY0(spa_keystore_unload_wkey(encroot
));
3128 key_loaded
= B_FALSE
;
3131 static avl_tree_t idx_tree
;
3132 static avl_tree_t domain_tree
;
3133 static boolean_t fuid_table_loaded
;
3134 static objset_t
*sa_os
= NULL
;
3135 static sa_attr_type_t
*sa_attr_table
= NULL
;
3138 open_objset(const char *path
, const void *tag
, objset_t
**osp
)
3141 uint64_t sa_attrs
= 0;
3142 uint64_t version
= 0;
3144 VERIFY3P(sa_os
, ==, NULL
);
3147 * We can't own an objset if it's redacted. Therefore, we do this
3148 * dance: hold the objset, then acquire a long hold on its dataset, then
3149 * release the pool (which is held as part of holding the objset).
3152 if (dump_opt
['K']) {
3153 /* decryption requested, try to load keys */
3154 err
= dmu_objset_hold(path
, tag
, osp
);
3156 (void) fprintf(stderr
, "failed to hold dataset "
3158 path
, strerror(err
));
3161 dsl_dataset_long_hold(dmu_objset_ds(*osp
), tag
);
3162 dsl_pool_rele(dmu_objset_pool(*osp
), tag
);
3164 /* succeeds or dies */
3167 /* release it all */
3168 dsl_dataset_long_rele(dmu_objset_ds(*osp
), tag
);
3169 dsl_dataset_rele(dmu_objset_ds(*osp
), tag
);
3172 int ds_hold_flags
= key_loaded
? DS_HOLD_FLAG_DECRYPT
: 0;
3174 err
= dmu_objset_hold_flags(path
, ds_hold_flags
, tag
, osp
);
3176 (void) fprintf(stderr
, "failed to hold dataset '%s': %s\n",
3177 path
, strerror(err
));
3180 dsl_dataset_long_hold(dmu_objset_ds(*osp
), tag
);
3181 dsl_pool_rele(dmu_objset_pool(*osp
), tag
);
3183 if (dmu_objset_type(*osp
) == DMU_OST_ZFS
&&
3184 (key_loaded
|| !(*osp
)->os_encrypted
)) {
3185 (void) zap_lookup(*osp
, MASTER_NODE_OBJ
, ZPL_VERSION_STR
,
3187 if (version
>= ZPL_VERSION_SA
) {
3188 (void) zap_lookup(*osp
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
,
3191 err
= sa_setup(*osp
, sa_attrs
, zfs_attr_table
, ZPL_END
,
3194 (void) fprintf(stderr
, "sa_setup failed: %s\n",
3196 dsl_dataset_long_rele(dmu_objset_ds(*osp
), tag
);
3197 dsl_dataset_rele_flags(dmu_objset_ds(*osp
),
3198 ds_hold_flags
, tag
);
3208 close_objset(objset_t
*os
, const void *tag
)
3210 VERIFY3P(os
, ==, sa_os
);
3211 if (os
->os_sa
!= NULL
)
3213 dsl_dataset_long_rele(dmu_objset_ds(os
), tag
);
3214 dsl_dataset_rele_flags(dmu_objset_ds(os
),
3215 key_loaded
? DS_HOLD_FLAG_DECRYPT
: 0, tag
);
3216 sa_attr_table
= NULL
;
3223 fuid_table_destroy(void)
3225 if (fuid_table_loaded
) {
3226 zfs_fuid_table_destroy(&idx_tree
, &domain_tree
);
3227 fuid_table_loaded
= B_FALSE
;
3232 * print uid or gid information.
3233 * For normal POSIX id just the id is printed in decimal format.
3234 * For CIFS files with FUID the fuid is printed in hex followed by
3235 * the domain-rid string.
3238 print_idstr(uint64_t id
, const char *id_type
)
3240 if (FUID_INDEX(id
)) {
3241 const char *domain
=
3242 zfs_fuid_idx_domain(&idx_tree
, FUID_INDEX(id
));
3243 (void) printf("\t%s %llx [%s-%d]\n", id_type
,
3244 (u_longlong_t
)id
, domain
, (int)FUID_RID(id
));
3246 (void) printf("\t%s %llu\n", id_type
, (u_longlong_t
)id
);
3252 dump_uidgid(objset_t
*os
, uint64_t uid
, uint64_t gid
)
3254 uint32_t uid_idx
, gid_idx
;
3256 uid_idx
= FUID_INDEX(uid
);
3257 gid_idx
= FUID_INDEX(gid
);
3259 /* Load domain table, if not already loaded */
3260 if (!fuid_table_loaded
&& (uid_idx
|| gid_idx
)) {
3263 /* first find the fuid object. It lives in the master node */
3264 VERIFY(zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_FUID_TABLES
,
3265 8, 1, &fuid_obj
) == 0);
3266 zfs_fuid_avl_tree_create(&idx_tree
, &domain_tree
);
3267 (void) zfs_fuid_table_load(os
, fuid_obj
,
3268 &idx_tree
, &domain_tree
);
3269 fuid_table_loaded
= B_TRUE
;
3272 print_idstr(uid
, "uid");
3273 print_idstr(gid
, "gid");
3277 dump_znode_sa_xattr(sa_handle_t
*hdl
)
3280 nvpair_t
*elem
= NULL
;
3281 int sa_xattr_size
= 0;
3282 int sa_xattr_entries
= 0;
3284 char *sa_xattr_packed
;
3286 error
= sa_size(hdl
, sa_attr_table
[ZPL_DXATTR
], &sa_xattr_size
);
3287 if (error
|| sa_xattr_size
== 0)
3290 sa_xattr_packed
= malloc(sa_xattr_size
);
3291 if (sa_xattr_packed
== NULL
)
3294 error
= sa_lookup(hdl
, sa_attr_table
[ZPL_DXATTR
],
3295 sa_xattr_packed
, sa_xattr_size
);
3297 free(sa_xattr_packed
);
3301 error
= nvlist_unpack(sa_xattr_packed
, sa_xattr_size
, &sa_xattr
, 0);
3303 free(sa_xattr_packed
);
3307 while ((elem
= nvlist_next_nvpair(sa_xattr
, elem
)) != NULL
)
3310 (void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3311 sa_xattr_size
, sa_xattr_entries
);
3312 while ((elem
= nvlist_next_nvpair(sa_xattr
, elem
)) != NULL
) {
3313 boolean_t can_print
= !dump_opt
['P'];
3317 (void) printf("\t\t%s = ", nvpair_name(elem
));
3318 nvpair_value_byte_array(elem
, &value
, &cnt
);
3320 for (idx
= 0; idx
< cnt
; ++idx
) {
3321 if (!isprint(value
[idx
])) {
3322 can_print
= B_FALSE
;
3327 for (idx
= 0; idx
< cnt
; ++idx
) {
3329 (void) putchar(value
[idx
]);
3331 (void) printf("\\%3.3o", value
[idx
]);
3333 (void) putchar('\n');
3336 nvlist_free(sa_xattr
);
3337 free(sa_xattr_packed
);
3341 dump_znode_symlink(sa_handle_t
*hdl
)
3343 int sa_symlink_size
= 0;
3344 char linktarget
[MAXPATHLEN
];
3347 error
= sa_size(hdl
, sa_attr_table
[ZPL_SYMLINK
], &sa_symlink_size
);
3348 if (error
|| sa_symlink_size
== 0) {
3351 if (sa_symlink_size
>= sizeof (linktarget
)) {
3352 (void) printf("symlink size %d is too large\n",
3356 linktarget
[sa_symlink_size
] = '\0';
3357 if (sa_lookup(hdl
, sa_attr_table
[ZPL_SYMLINK
],
3358 &linktarget
, sa_symlink_size
) == 0)
3359 (void) printf("\ttarget %s\n", linktarget
);
3363 dump_znode(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
3365 (void) data
, (void) size
;
3366 char path
[MAXPATHLEN
* 2]; /* allow for xattr and failure prefix */
3368 uint64_t xattr
, rdev
, gen
;
3369 uint64_t uid
, gid
, mode
, fsize
, parent
, links
;
3371 uint64_t acctm
[2], modtm
[2], chgtm
[2], crtm
[2];
3372 time_t z_crtime
, z_atime
, z_mtime
, z_ctime
;
3373 sa_bulk_attr_t bulk
[12];
3377 VERIFY3P(os
, ==, sa_os
);
3378 if (sa_handle_get(os
, object
, NULL
, SA_HDL_PRIVATE
, &hdl
)) {
3379 (void) printf("Failed to get handle for SA znode\n");
3383 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_UID
], NULL
, &uid
, 8);
3384 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_GID
], NULL
, &gid
, 8);
3385 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_LINKS
], NULL
,
3387 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_GEN
], NULL
, &gen
, 8);
3388 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_MODE
], NULL
,
3390 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_PARENT
],
3392 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_SIZE
], NULL
,
3394 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_ATIME
], NULL
,
3396 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_MTIME
], NULL
,
3398 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_CRTIME
], NULL
,
3400 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_CTIME
], NULL
,
3402 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_FLAGS
], NULL
,
3405 if (sa_bulk_lookup(hdl
, bulk
, idx
)) {
3406 (void) sa_handle_destroy(hdl
);
3410 z_crtime
= (time_t)crtm
[0];
3411 z_atime
= (time_t)acctm
[0];
3412 z_mtime
= (time_t)modtm
[0];
3413 z_ctime
= (time_t)chgtm
[0];
3415 if (dump_opt
['d'] > 4) {
3416 error
= zfs_obj_to_path(os
, object
, path
, sizeof (path
));
3417 if (error
== ESTALE
) {
3418 (void) snprintf(path
, sizeof (path
), "on delete queue");
3419 } else if (error
!= 0) {
3421 (void) snprintf(path
, sizeof (path
),
3422 "path not found, possibly leaked");
3424 (void) printf("\tpath %s\n", path
);
3428 dump_znode_symlink(hdl
);
3429 dump_uidgid(os
, uid
, gid
);
3430 (void) printf("\tatime %s", ctime(&z_atime
));
3431 (void) printf("\tmtime %s", ctime(&z_mtime
));
3432 (void) printf("\tctime %s", ctime(&z_ctime
));
3433 (void) printf("\tcrtime %s", ctime(&z_crtime
));
3434 (void) printf("\tgen %llu\n", (u_longlong_t
)gen
);
3435 (void) printf("\tmode %llo\n", (u_longlong_t
)mode
);
3436 (void) printf("\tsize %llu\n", (u_longlong_t
)fsize
);
3437 (void) printf("\tparent %llu\n", (u_longlong_t
)parent
);
3438 (void) printf("\tlinks %llu\n", (u_longlong_t
)links
);
3439 (void) printf("\tpflags %llx\n", (u_longlong_t
)pflags
);
3440 if (dmu_objset_projectquota_enabled(os
) && (pflags
& ZFS_PROJID
)) {
3443 if (sa_lookup(hdl
, sa_attr_table
[ZPL_PROJID
], &projid
,
3444 sizeof (uint64_t)) == 0)
3445 (void) printf("\tprojid %llu\n", (u_longlong_t
)projid
);
3447 if (sa_lookup(hdl
, sa_attr_table
[ZPL_XATTR
], &xattr
,
3448 sizeof (uint64_t)) == 0)
3449 (void) printf("\txattr %llu\n", (u_longlong_t
)xattr
);
3450 if (sa_lookup(hdl
, sa_attr_table
[ZPL_RDEV
], &rdev
,
3451 sizeof (uint64_t)) == 0)
3452 (void) printf("\trdev 0x%016llx\n", (u_longlong_t
)rdev
);
3453 dump_znode_sa_xattr(hdl
);
3454 sa_handle_destroy(hdl
);
3458 dump_acl(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
3460 (void) os
, (void) object
, (void) data
, (void) size
;
3464 dump_dmu_objset(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
3466 (void) os
, (void) object
, (void) data
, (void) size
;
3469 static object_viewer_t
*object_viewer
[DMU_OT_NUMTYPES
+ 1] = {
3470 dump_none
, /* unallocated */
3471 dump_zap
, /* object directory */
3472 dump_uint64
, /* object array */
3473 dump_none
, /* packed nvlist */
3474 dump_packed_nvlist
, /* packed nvlist size */
3475 dump_none
, /* bpobj */
3476 dump_bpobj
, /* bpobj header */
3477 dump_none
, /* SPA space map header */
3478 dump_none
, /* SPA space map */
3479 dump_none
, /* ZIL intent log */
3480 dump_dnode
, /* DMU dnode */
3481 dump_dmu_objset
, /* DMU objset */
3482 dump_dsl_dir
, /* DSL directory */
3483 dump_zap
, /* DSL directory child map */
3484 dump_zap
, /* DSL dataset snap map */
3485 dump_zap
, /* DSL props */
3486 dump_dsl_dataset
, /* DSL dataset */
3487 dump_znode
, /* ZFS znode */
3488 dump_acl
, /* ZFS V0 ACL */
3489 dump_uint8
, /* ZFS plain file */
3490 dump_zpldir
, /* ZFS directory */
3491 dump_zap
, /* ZFS master node */
3492 dump_zap
, /* ZFS delete queue */
3493 dump_uint8
, /* zvol object */
3494 dump_zap
, /* zvol prop */
3495 dump_uint8
, /* other uint8[] */
3496 dump_uint64
, /* other uint64[] */
3497 dump_zap
, /* other ZAP */
3498 dump_zap
, /* persistent error log */
3499 dump_uint8
, /* SPA history */
3500 dump_history_offsets
, /* SPA history offsets */
3501 dump_zap
, /* Pool properties */
3502 dump_zap
, /* DSL permissions */
3503 dump_acl
, /* ZFS ACL */
3504 dump_uint8
, /* ZFS SYSACL */
3505 dump_none
, /* FUID nvlist */
3506 dump_packed_nvlist
, /* FUID nvlist size */
3507 dump_zap
, /* DSL dataset next clones */
3508 dump_zap
, /* DSL scrub queue */
3509 dump_zap
, /* ZFS user/group/project used */
3510 dump_zap
, /* ZFS user/group/project quota */
3511 dump_zap
, /* snapshot refcount tags */
3512 dump_ddt_zap
, /* DDT ZAP object */
3513 dump_zap
, /* DDT statistics */
3514 dump_znode
, /* SA object */
3515 dump_zap
, /* SA Master Node */
3516 dump_sa_attrs
, /* SA attribute registration */
3517 dump_sa_layouts
, /* SA attribute layouts */
3518 dump_zap
, /* DSL scrub translations */
3519 dump_none
, /* fake dedup BP */
3520 dump_zap
, /* deadlist */
3521 dump_none
, /* deadlist hdr */
3522 dump_zap
, /* dsl clones */
3523 dump_bpobj_subobjs
, /* bpobj subobjs */
3524 dump_unknown
, /* Unknown type, must be last */
3528 match_object_type(dmu_object_type_t obj_type
, uint64_t flags
)
3530 boolean_t match
= B_TRUE
;
3533 case DMU_OT_DIRECTORY_CONTENTS
:
3534 if (!(flags
& ZOR_FLAG_DIRECTORY
))
3537 case DMU_OT_PLAIN_FILE_CONTENTS
:
3538 if (!(flags
& ZOR_FLAG_PLAIN_FILE
))
3541 case DMU_OT_SPACE_MAP
:
3542 if (!(flags
& ZOR_FLAG_SPACE_MAP
))
3546 if (strcmp(zdb_ot_name(obj_type
), "zap") == 0) {
3547 if (!(flags
& ZOR_FLAG_ZAP
))
3553 * If all bits except some of the supported flags are
3554 * set, the user combined the all-types flag (A) with
3555 * a negated flag to exclude some types (e.g. A-f to
3556 * show all object types except plain files).
3558 if ((flags
| ZOR_SUPPORTED_FLAGS
) != ZOR_FLAG_ALL_TYPES
)
3568 dump_object(objset_t
*os
, uint64_t object
, int verbosity
,
3569 boolean_t
*print_header
, uint64_t *dnode_slots_used
, uint64_t flags
)
3571 dmu_buf_t
*db
= NULL
;
3572 dmu_object_info_t doi
;
3574 boolean_t dnode_held
= B_FALSE
;
3577 char iblk
[32], dblk
[32], lsize
[32], asize
[32], fill
[32], dnsize
[32];
3578 char bonus_size
[32];
3582 /* make sure nicenum has enough space */
3583 _Static_assert(sizeof (iblk
) >= NN_NUMBUF_SZ
, "iblk truncated");
3584 _Static_assert(sizeof (dblk
) >= NN_NUMBUF_SZ
, "dblk truncated");
3585 _Static_assert(sizeof (lsize
) >= NN_NUMBUF_SZ
, "lsize truncated");
3586 _Static_assert(sizeof (asize
) >= NN_NUMBUF_SZ
, "asize truncated");
3587 _Static_assert(sizeof (bonus_size
) >= NN_NUMBUF_SZ
,
3588 "bonus_size truncated");
3590 if (*print_header
) {
3591 (void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n",
3592 "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3593 "lsize", "%full", "type");
3598 dn
= DMU_META_DNODE(os
);
3599 dmu_object_info_from_dnode(dn
, &doi
);
3602 * Encrypted datasets will have sensitive bonus buffers
3603 * encrypted. Therefore we cannot hold the bonus buffer and
3604 * must hold the dnode itself instead.
3606 error
= dmu_object_info(os
, object
, &doi
);
3608 fatal("dmu_object_info() failed, errno %u", error
);
3610 if (!key_loaded
&& os
->os_encrypted
&&
3611 DMU_OT_IS_ENCRYPTED(doi
.doi_bonus_type
)) {
3612 error
= dnode_hold(os
, object
, FTAG
, &dn
);
3614 fatal("dnode_hold() failed, errno %u", error
);
3615 dnode_held
= B_TRUE
;
3617 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
3619 fatal("dmu_bonus_hold(%llu) failed, errno %u",
3621 bonus
= db
->db_data
;
3622 bsize
= db
->db_size
;
3623 dn
= DB_DNODE((dmu_buf_impl_t
*)db
);
3628 * Default to showing all object types if no flags were specified.
3630 if (flags
!= 0 && flags
!= ZOR_FLAG_ALL_TYPES
&&
3631 !match_object_type(doi
.doi_type
, flags
))
3634 if (dnode_slots_used
)
3635 *dnode_slots_used
= doi
.doi_dnodesize
/ DNODE_MIN_SIZE
;
3637 zdb_nicenum(doi
.doi_metadata_block_size
, iblk
, sizeof (iblk
));
3638 zdb_nicenum(doi
.doi_data_block_size
, dblk
, sizeof (dblk
));
3639 zdb_nicenum(doi
.doi_max_offset
, lsize
, sizeof (lsize
));
3640 zdb_nicenum(doi
.doi_physical_blocks_512
<< 9, asize
, sizeof (asize
));
3641 zdb_nicenum(doi
.doi_bonus_size
, bonus_size
, sizeof (bonus_size
));
3642 zdb_nicenum(doi
.doi_dnodesize
, dnsize
, sizeof (dnsize
));
3643 (void) snprintf(fill
, sizeof (fill
), "%6.2f", 100.0 *
3644 doi
.doi_fill_count
* doi
.doi_data_block_size
/ (object
== 0 ?
3645 DNODES_PER_BLOCK
: 1) / doi
.doi_max_offset
);
3649 if (doi
.doi_checksum
!= ZIO_CHECKSUM_INHERIT
|| verbosity
>= 6) {
3650 (void) snprintf(aux
+ strlen(aux
), sizeof (aux
) - strlen(aux
),
3651 " (K=%s)", ZDB_CHECKSUM_NAME(doi
.doi_checksum
));
3654 if (doi
.doi_compress
== ZIO_COMPRESS_INHERIT
&&
3655 ZIO_COMPRESS_HASLEVEL(os
->os_compress
) && verbosity
>= 6) {
3656 const char *compname
= NULL
;
3657 if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION
,
3658 ZIO_COMPRESS_RAW(os
->os_compress
, os
->os_complevel
),
3660 (void) snprintf(aux
+ strlen(aux
),
3661 sizeof (aux
) - strlen(aux
), " (Z=inherit=%s)",
3664 (void) snprintf(aux
+ strlen(aux
),
3665 sizeof (aux
) - strlen(aux
),
3666 " (Z=inherit=%s-unknown)",
3667 ZDB_COMPRESS_NAME(os
->os_compress
));
3669 } else if (doi
.doi_compress
== ZIO_COMPRESS_INHERIT
&& verbosity
>= 6) {
3670 (void) snprintf(aux
+ strlen(aux
), sizeof (aux
) - strlen(aux
),
3671 " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os
->os_compress
));
3672 } else if (doi
.doi_compress
!= ZIO_COMPRESS_INHERIT
|| verbosity
>= 6) {
3673 (void) snprintf(aux
+ strlen(aux
), sizeof (aux
) - strlen(aux
),
3674 " (Z=%s)", ZDB_COMPRESS_NAME(doi
.doi_compress
));
3677 (void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n",
3678 (u_longlong_t
)object
, doi
.doi_indirection
, iblk
, dblk
,
3679 asize
, dnsize
, lsize
, fill
, zdb_ot_name(doi
.doi_type
), aux
);
3681 if (doi
.doi_bonus_type
!= DMU_OT_NONE
&& verbosity
> 3) {
3682 (void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n",
3683 "", "", "", "", "", "", bonus_size
, "bonus",
3684 zdb_ot_name(doi
.doi_bonus_type
));
3687 if (verbosity
>= 4) {
3688 (void) printf("\tdnode flags: %s%s%s%s\n",
3689 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_USED_BYTES
) ?
3691 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_USERUSED_ACCOUNTED
) ?
3692 "USERUSED_ACCOUNTED " : "",
3693 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_USEROBJUSED_ACCOUNTED
) ?
3694 "USEROBJUSED_ACCOUNTED " : "",
3695 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) ?
3696 "SPILL_BLKPTR" : "");
3697 (void) printf("\tdnode maxblkid: %llu\n",
3698 (longlong_t
)dn
->dn_phys
->dn_maxblkid
);
3701 object_viewer
[ZDB_OT_TYPE(doi
.doi_bonus_type
)](os
,
3702 object
, bonus
, bsize
);
3704 (void) printf("\t\t(bonus encrypted)\n");
3708 (!os
->os_encrypted
|| !DMU_OT_IS_ENCRYPTED(doi
.doi_type
))) {
3709 object_viewer
[ZDB_OT_TYPE(doi
.doi_type
)](os
, object
,
3712 (void) printf("\t\t(object encrypted)\n");
3715 *print_header
= B_TRUE
;
3718 if (verbosity
>= 5) {
3719 if (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
3720 char blkbuf
[BP_SPRINTF_LEN
];
3721 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
),
3722 DN_SPILL_BLKPTR(dn
->dn_phys
), B_FALSE
);
3723 (void) printf("\nSpill block: %s\n", blkbuf
);
3728 if (verbosity
>= 5) {
3730 * Report the list of segments that comprise the object.
3734 uint64_t blkfill
= 1;
3737 if (dn
->dn_type
== DMU_OT_DNODE
) {
3739 blkfill
= DNODES_PER_BLOCK
;
3744 /* make sure nicenum has enough space */
3745 _Static_assert(sizeof (segsize
) >= NN_NUMBUF_SZ
,
3746 "segsize truncated");
3747 error
= dnode_next_offset(dn
,
3748 0, &start
, minlvl
, blkfill
, 0);
3752 error
= dnode_next_offset(dn
,
3753 DNODE_FIND_HOLE
, &end
, minlvl
, blkfill
, 0);
3754 zdb_nicenum(end
- start
, segsize
, sizeof (segsize
));
3755 (void) printf("\t\tsegment [%016llx, %016llx)"
3756 " size %5s\n", (u_longlong_t
)start
,
3757 (u_longlong_t
)end
, segsize
);
3766 dmu_buf_rele(db
, FTAG
);
3768 dnode_rele(dn
, FTAG
);
3772 count_dir_mos_objects(dsl_dir_t
*dd
)
3774 mos_obj_refd(dd
->dd_object
);
3775 mos_obj_refd(dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
3776 mos_obj_refd(dsl_dir_phys(dd
)->dd_deleg_zapobj
);
3777 mos_obj_refd(dsl_dir_phys(dd
)->dd_props_zapobj
);
3778 mos_obj_refd(dsl_dir_phys(dd
)->dd_clones
);
3781 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
3782 * Ignore the references after the first one.
3784 mos_obj_refd_multiple(dd
->dd_crypto_obj
);
3788 count_ds_mos_objects(dsl_dataset_t
*ds
)
3790 mos_obj_refd(ds
->ds_object
);
3791 mos_obj_refd(dsl_dataset_phys(ds
)->ds_next_clones_obj
);
3792 mos_obj_refd(dsl_dataset_phys(ds
)->ds_props_obj
);
3793 mos_obj_refd(dsl_dataset_phys(ds
)->ds_userrefs_obj
);
3794 mos_obj_refd(dsl_dataset_phys(ds
)->ds_snapnames_zapobj
);
3795 mos_obj_refd(ds
->ds_bookmarks_obj
);
3797 if (!dsl_dataset_is_snapshot(ds
)) {
3798 count_dir_mos_objects(ds
->ds_dir
);
3802 static const char *const objset_types
[DMU_OST_NUMTYPES
] = {
3803 "NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
3806 * Parse a string denoting a range of object IDs of the form
3807 * <start>[:<end>[:flags]], and store the results in zor.
3808 * Return 0 on success. On error, return 1 and update the msg
3809 * pointer to point to a descriptive error message.
3812 parse_object_range(char *range
, zopt_object_range_t
*zor
, const char **msg
)
3815 char *p
, *s
, *dup
, *flagstr
, *tmp
= NULL
;
3820 if (strchr(range
, ':') == NULL
) {
3821 zor
->zor_obj_start
= strtoull(range
, &p
, 0);
3823 *msg
= "Invalid characters in object ID";
3826 zor
->zor_obj_start
= ZDB_MAP_OBJECT_ID(zor
->zor_obj_start
);
3827 zor
->zor_obj_end
= zor
->zor_obj_start
;
3831 if (strchr(range
, ':') == range
) {
3832 *msg
= "Invalid leading colon";
3837 len
= strlen(range
);
3838 if (range
[len
- 1] == ':') {
3839 *msg
= "Invalid trailing colon";
3844 dup
= strdup(range
);
3845 s
= strtok_r(dup
, ":", &tmp
);
3846 zor
->zor_obj_start
= strtoull(s
, &p
, 0);
3849 *msg
= "Invalid characters in start object ID";
3854 s
= strtok_r(NULL
, ":", &tmp
);
3855 zor
->zor_obj_end
= strtoull(s
, &p
, 0);
3858 *msg
= "Invalid characters in end object ID";
3863 if (zor
->zor_obj_start
> zor
->zor_obj_end
) {
3864 *msg
= "Start object ID may not exceed end object ID";
3869 s
= strtok_r(NULL
, ":", &tmp
);
3871 zor
->zor_flags
= ZOR_FLAG_ALL_TYPES
;
3873 } else if (strtok_r(NULL
, ":", &tmp
) != NULL
) {
3874 *msg
= "Invalid colon-delimited field after flags";
3880 for (i
= 0; flagstr
[i
]; i
++) {
3882 boolean_t negation
= (flagstr
[i
] == '-');
3886 if (flagstr
[i
] == '\0') {
3887 *msg
= "Invalid trailing negation operator";
3892 bit
= flagbits
[(uchar_t
)flagstr
[i
]];
3894 *msg
= "Invalid flag";
3903 zor
->zor_flags
= flags
;
3905 zor
->zor_obj_start
= ZDB_MAP_OBJECT_ID(zor
->zor_obj_start
);
3906 zor
->zor_obj_end
= ZDB_MAP_OBJECT_ID(zor
->zor_obj_end
);
3914 dump_objset(objset_t
*os
)
3916 dmu_objset_stats_t dds
= { 0 };
3917 uint64_t object
, object_count
;
3918 uint64_t refdbytes
, usedobjs
, scratch
;
3920 char blkbuf
[BP_SPRINTF_LEN
+ 20];
3921 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
3922 const char *type
= "UNKNOWN";
3923 int verbosity
= dump_opt
['d'];
3924 boolean_t print_header
;
3927 uint64_t total_slots_used
= 0;
3928 uint64_t max_slot_used
= 0;
3929 uint64_t dnode_slots
;
3934 /* make sure nicenum has enough space */
3935 _Static_assert(sizeof (numbuf
) >= NN_NUMBUF_SZ
, "numbuf truncated");
3937 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
3938 dmu_objset_fast_stat(os
, &dds
);
3939 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
3941 print_header
= B_TRUE
;
3943 if (dds
.dds_type
< DMU_OST_NUMTYPES
)
3944 type
= objset_types
[dds
.dds_type
];
3946 if (dds
.dds_type
== DMU_OST_META
) {
3947 dds
.dds_creation_txg
= TXG_INITIAL
;
3948 usedobjs
= BP_GET_FILL(os
->os_rootbp
);
3949 refdbytes
= dsl_dir_phys(os
->os_spa
->spa_dsl_pool
->dp_mos_dir
)->
3952 dmu_objset_space(os
, &refdbytes
, &scratch
, &usedobjs
, &scratch
);
3955 ASSERT3U(usedobjs
, ==, BP_GET_FILL(os
->os_rootbp
));
3957 zdb_nicenum(refdbytes
, numbuf
, sizeof (numbuf
));
3959 if (verbosity
>= 4) {
3960 (void) snprintf(blkbuf
, sizeof (blkbuf
), ", rootbp ");
3961 (void) snprintf_blkptr(blkbuf
+ strlen(blkbuf
),
3962 sizeof (blkbuf
) - strlen(blkbuf
), os
->os_rootbp
);
3967 dmu_objset_name(os
, osname
);
3969 (void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
3970 "%s, %llu objects%s%s\n",
3971 osname
, type
, (u_longlong_t
)dmu_objset_id(os
),
3972 (u_longlong_t
)dds
.dds_creation_txg
,
3973 numbuf
, (u_longlong_t
)usedobjs
, blkbuf
,
3974 (dds
.dds_inconsistent
) ? " (inconsistent)" : "");
3976 for (i
= 0; i
< zopt_object_args
; i
++) {
3977 obj_start
= zopt_object_ranges
[i
].zor_obj_start
;
3978 obj_end
= zopt_object_ranges
[i
].zor_obj_end
;
3979 flags
= zopt_object_ranges
[i
].zor_flags
;
3982 if (object
== 0 || obj_start
== obj_end
)
3983 dump_object(os
, object
, verbosity
, &print_header
, NULL
,
3988 while ((dmu_object_next(os
, &object
, B_FALSE
, 0) == 0) &&
3989 object
<= obj_end
) {
3990 dump_object(os
, object
, verbosity
, &print_header
, NULL
,
3995 if (zopt_object_args
> 0) {
3996 (void) printf("\n");
4000 if (dump_opt
['i'] != 0 || verbosity
>= 2)
4001 dump_intent_log(dmu_objset_zil(os
));
4003 if (dmu_objset_ds(os
) != NULL
) {
4004 dsl_dataset_t
*ds
= dmu_objset_ds(os
);
4005 dump_blkptr_list(&ds
->ds_deadlist
, "Deadlist");
4006 if (dsl_deadlist_is_open(&ds
->ds_dir
->dd_livelist
) &&
4007 !dmu_objset_is_snapshot(os
)) {
4008 dump_blkptr_list(&ds
->ds_dir
->dd_livelist
, "Livelist");
4009 if (verify_dd_livelist(os
) != 0)
4010 fatal("livelist is incorrect");
4013 if (dsl_dataset_remap_deadlist_exists(ds
)) {
4014 (void) printf("ds_remap_deadlist:\n");
4015 dump_blkptr_list(&ds
->ds_remap_deadlist
, "Deadlist");
4017 count_ds_mos_objects(ds
);
4020 if (dmu_objset_ds(os
) != NULL
)
4021 dump_bookmarks(os
, verbosity
);
4026 if (BP_IS_HOLE(os
->os_rootbp
))
4029 dump_object(os
, 0, verbosity
, &print_header
, NULL
, 0);
4031 if (DMU_USERUSED_DNODE(os
) != NULL
&&
4032 DMU_USERUSED_DNODE(os
)->dn_type
!= 0) {
4033 dump_object(os
, DMU_USERUSED_OBJECT
, verbosity
, &print_header
,
4035 dump_object(os
, DMU_GROUPUSED_OBJECT
, verbosity
, &print_header
,
4039 if (DMU_PROJECTUSED_DNODE(os
) != NULL
&&
4040 DMU_PROJECTUSED_DNODE(os
)->dn_type
!= 0)
4041 dump_object(os
, DMU_PROJECTUSED_OBJECT
, verbosity
,
4042 &print_header
, NULL
, 0);
4045 while ((error
= dmu_object_next(os
, &object
, B_FALSE
, 0)) == 0) {
4046 dump_object(os
, object
, verbosity
, &print_header
, &dnode_slots
,
4049 total_slots_used
+= dnode_slots
;
4050 max_slot_used
= object
+ dnode_slots
- 1;
4053 (void) printf("\n");
4055 (void) printf(" Dnode slots:\n");
4056 (void) printf("\tTotal used: %10llu\n",
4057 (u_longlong_t
)total_slots_used
);
4058 (void) printf("\tMax used: %10llu\n",
4059 (u_longlong_t
)max_slot_used
);
4060 (void) printf("\tPercent empty: %10lf\n",
4061 (double)(max_slot_used
- total_slots_used
)*100 /
4062 (double)max_slot_used
);
4063 (void) printf("\n");
4065 if (error
!= ESRCH
) {
4066 (void) fprintf(stderr
, "dmu_object_next() = %d\n", error
);
4070 ASSERT3U(object_count
, ==, usedobjs
);
4072 if (leaked_objects
!= 0) {
4073 (void) printf("%d potentially leaked objects detected\n",
4080 dump_uberblock(uberblock_t
*ub
, const char *header
, const char *footer
)
4082 time_t timestamp
= ub
->ub_timestamp
;
4084 (void) printf("%s", header
? header
: "");
4085 (void) printf("\tmagic = %016llx\n", (u_longlong_t
)ub
->ub_magic
);
4086 (void) printf("\tversion = %llu\n", (u_longlong_t
)ub
->ub_version
);
4087 (void) printf("\ttxg = %llu\n", (u_longlong_t
)ub
->ub_txg
);
4088 (void) printf("\tguid_sum = %llu\n", (u_longlong_t
)ub
->ub_guid_sum
);
4089 (void) printf("\ttimestamp = %llu UTC = %s",
4090 (u_longlong_t
)ub
->ub_timestamp
, ctime(×tamp
));
4092 (void) printf("\tmmp_magic = %016llx\n",
4093 (u_longlong_t
)ub
->ub_mmp_magic
);
4094 if (MMP_VALID(ub
)) {
4095 (void) printf("\tmmp_delay = %0llu\n",
4096 (u_longlong_t
)ub
->ub_mmp_delay
);
4097 if (MMP_SEQ_VALID(ub
))
4098 (void) printf("\tmmp_seq = %u\n",
4099 (unsigned int) MMP_SEQ(ub
));
4100 if (MMP_FAIL_INT_VALID(ub
))
4101 (void) printf("\tmmp_fail = %u\n",
4102 (unsigned int) MMP_FAIL_INT(ub
));
4103 if (MMP_INTERVAL_VALID(ub
))
4104 (void) printf("\tmmp_write = %u\n",
4105 (unsigned int) MMP_INTERVAL(ub
));
4106 /* After MMP_* to make summarize_uberblock_mmp cleaner */
4107 (void) printf("\tmmp_valid = %x\n",
4108 (unsigned int) ub
->ub_mmp_config
& 0xFF);
4111 if (dump_opt
['u'] >= 4) {
4112 char blkbuf
[BP_SPRINTF_LEN
];
4113 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), &ub
->ub_rootbp
);
4114 (void) printf("\trootbp = %s\n", blkbuf
);
4116 (void) printf("\tcheckpoint_txg = %llu\n",
4117 (u_longlong_t
)ub
->ub_checkpoint_txg
);
4119 (void) printf("\traidz_reflow state=%u off=%llu\n",
4120 (int)RRSS_GET_STATE(ub
),
4121 (u_longlong_t
)RRSS_GET_OFFSET(ub
));
4123 (void) printf("%s", footer
? footer
: "");
4127 dump_config(spa_t
*spa
)
4134 error
= dmu_bonus_hold(spa
->spa_meta_objset
,
4135 spa
->spa_config_object
, FTAG
, &db
);
4138 nvsize
= *(uint64_t *)db
->db_data
;
4139 dmu_buf_rele(db
, FTAG
);
4141 (void) printf("\nMOS Configuration:\n");
4142 dump_packed_nvlist(spa
->spa_meta_objset
,
4143 spa
->spa_config_object
, (void *)&nvsize
, 1);
4145 (void) fprintf(stderr
, "dmu_bonus_hold(%llu) failed, errno %d",
4146 (u_longlong_t
)spa
->spa_config_object
, error
);
4151 dump_cachefile(const char *cachefile
)
4154 struct stat64 statbuf
;
4158 if ((fd
= open64(cachefile
, O_RDONLY
)) < 0) {
4159 (void) printf("cannot open '%s': %s\n", cachefile
,
4164 if (fstat64(fd
, &statbuf
) != 0) {
4165 (void) printf("failed to stat '%s': %s\n", cachefile
,
4170 if ((buf
= malloc(statbuf
.st_size
)) == NULL
) {
4171 (void) fprintf(stderr
, "failed to allocate %llu bytes\n",
4172 (u_longlong_t
)statbuf
.st_size
);
4176 if (read(fd
, buf
, statbuf
.st_size
) != statbuf
.st_size
) {
4177 (void) fprintf(stderr
, "failed to read %llu bytes\n",
4178 (u_longlong_t
)statbuf
.st_size
);
4184 if (nvlist_unpack(buf
, statbuf
.st_size
, &config
, 0) != 0) {
4185 (void) fprintf(stderr
, "failed to unpack nvlist\n");
4191 dump_nvlist(config
, 0);
4193 nvlist_free(config
);
4197 * ZFS label nvlist stats
4199 typedef struct zdb_nvl_stats
{
4202 size_t zns_leaf_largest
;
4203 size_t zns_leaf_total
;
4204 nvlist_t
*zns_string
;
4205 nvlist_t
*zns_uint64
;
4206 nvlist_t
*zns_boolean
;
4210 collect_nvlist_stats(nvlist_t
*nvl
, zdb_nvl_stats_t
*stats
)
4212 nvlist_t
*list
, **array
;
4213 nvpair_t
*nvp
= NULL
;
4217 stats
->zns_list_count
++;
4219 while ((nvp
= nvlist_next_nvpair(nvl
, nvp
)) != NULL
) {
4220 name
= nvpair_name(nvp
);
4222 switch (nvpair_type(nvp
)) {
4223 case DATA_TYPE_STRING
:
4224 fnvlist_add_string(stats
->zns_string
, name
,
4225 fnvpair_value_string(nvp
));
4227 case DATA_TYPE_UINT64
:
4228 fnvlist_add_uint64(stats
->zns_uint64
, name
,
4229 fnvpair_value_uint64(nvp
));
4231 case DATA_TYPE_BOOLEAN
:
4232 fnvlist_add_boolean(stats
->zns_boolean
, name
);
4234 case DATA_TYPE_NVLIST
:
4235 if (nvpair_value_nvlist(nvp
, &list
) == 0)
4236 collect_nvlist_stats(list
, stats
);
4238 case DATA_TYPE_NVLIST_ARRAY
:
4239 if (nvpair_value_nvlist_array(nvp
, &array
, &items
) != 0)
4242 for (i
= 0; i
< items
; i
++) {
4243 collect_nvlist_stats(array
[i
], stats
);
4245 /* collect stats on leaf vdev */
4246 if (strcmp(name
, "children") == 0) {
4249 (void) nvlist_size(array
[i
], &size
,
4251 stats
->zns_leaf_total
+= size
;
4252 if (size
> stats
->zns_leaf_largest
)
4253 stats
->zns_leaf_largest
= size
;
4254 stats
->zns_leaf_count
++;
4259 (void) printf("skip type %d!\n", (int)nvpair_type(nvp
));
4265 dump_nvlist_stats(nvlist_t
*nvl
, size_t cap
)
4267 zdb_nvl_stats_t stats
= { 0 };
4268 size_t size
, sum
= 0, total
;
4271 /* requires nvlist with non-unique names for stat collection */
4272 VERIFY0(nvlist_alloc(&stats
.zns_string
, 0, 0));
4273 VERIFY0(nvlist_alloc(&stats
.zns_uint64
, 0, 0));
4274 VERIFY0(nvlist_alloc(&stats
.zns_boolean
, 0, 0));
4275 VERIFY0(nvlist_size(stats
.zns_boolean
, &noise
, NV_ENCODE_XDR
));
4277 (void) printf("\n\nZFS Label NVList Config Stats:\n");
4279 VERIFY0(nvlist_size(nvl
, &total
, NV_ENCODE_XDR
));
4280 (void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4281 (int)total
, (int)(cap
- total
), 100.0 * total
/ cap
);
4283 collect_nvlist_stats(nvl
, &stats
);
4285 VERIFY0(nvlist_size(stats
.zns_uint64
, &size
, NV_ENCODE_XDR
));
4288 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4289 (int)fnvlist_num_pairs(stats
.zns_uint64
),
4290 (int)size
, 100.0 * size
/ total
);
4292 VERIFY0(nvlist_size(stats
.zns_string
, &size
, NV_ENCODE_XDR
));
4295 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4296 (int)fnvlist_num_pairs(stats
.zns_string
),
4297 (int)size
, 100.0 * size
/ total
);
4299 VERIFY0(nvlist_size(stats
.zns_boolean
, &size
, NV_ENCODE_XDR
));
4302 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4303 (int)fnvlist_num_pairs(stats
.zns_boolean
),
4304 (int)size
, 100.0 * size
/ total
);
4306 size
= total
- sum
; /* treat remainder as nvlist overhead */
4307 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4308 stats
.zns_list_count
, (int)size
, 100.0 * size
/ total
);
4310 if (stats
.zns_leaf_count
> 0) {
4311 size_t average
= stats
.zns_leaf_total
/ stats
.zns_leaf_count
;
4313 (void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4314 stats
.zns_leaf_count
, (int)average
);
4315 (void) printf("%24d bytes largest\n",
4316 (int)stats
.zns_leaf_largest
);
4318 if (dump_opt
['l'] >= 3 && average
> 0)
4319 (void) printf(" space for %d additional leaf vdevs\n",
4320 (int)((cap
- total
) / average
));
4322 (void) printf("\n");
4324 nvlist_free(stats
.zns_string
);
4325 nvlist_free(stats
.zns_uint64
);
4326 nvlist_free(stats
.zns_boolean
);
4329 typedef struct cksum_record
{
4331 boolean_t labels
[VDEV_LABELS
];
4336 cksum_record_compare(const void *x1
, const void *x2
)
4338 const cksum_record_t
*l
= (cksum_record_t
*)x1
;
4339 const cksum_record_t
*r
= (cksum_record_t
*)x2
;
4340 int arraysize
= ARRAY_SIZE(l
->cksum
.zc_word
);
4343 for (int i
= 0; i
< arraysize
; i
++) {
4344 difference
= TREE_CMP(l
->cksum
.zc_word
[i
], r
->cksum
.zc_word
[i
]);
4349 return (difference
);
4352 static cksum_record_t
*
4353 cksum_record_alloc(zio_cksum_t
*cksum
, int l
)
4355 cksum_record_t
*rec
;
4357 rec
= umem_zalloc(sizeof (*rec
), UMEM_NOFAIL
);
4358 rec
->cksum
= *cksum
;
4359 rec
->labels
[l
] = B_TRUE
;
4364 static cksum_record_t
*
4365 cksum_record_lookup(avl_tree_t
*tree
, zio_cksum_t
*cksum
)
4367 cksum_record_t lookup
= { .cksum
= *cksum
};
4370 return (avl_find(tree
, &lookup
, &where
));
4373 static cksum_record_t
*
4374 cksum_record_insert(avl_tree_t
*tree
, zio_cksum_t
*cksum
, int l
)
4376 cksum_record_t
*rec
;
4378 rec
= cksum_record_lookup(tree
, cksum
);
4380 rec
->labels
[l
] = B_TRUE
;
4382 rec
= cksum_record_alloc(cksum
, l
);
4390 first_label(cksum_record_t
*rec
)
4392 for (int i
= 0; i
< VDEV_LABELS
; i
++)
4400 print_label_numbers(const char *prefix
, const cksum_record_t
*rec
)
4402 fputs(prefix
, stdout
);
4403 for (int i
= 0; i
< VDEV_LABELS
; i
++)
4404 if (rec
->labels
[i
] == B_TRUE
)
4409 #define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4411 typedef struct zdb_label
{
4413 uint64_t label_offset
;
4414 nvlist_t
*config_nv
;
4415 cksum_record_t
*config
;
4416 cksum_record_t
*uberblocks
[MAX_UBERBLOCK_COUNT
];
4417 boolean_t header_printed
;
4418 boolean_t read_failed
;
4419 boolean_t cksum_valid
;
4423 print_label_header(zdb_label_t
*label
, int l
)
4429 if (label
->header_printed
== B_TRUE
)
4432 (void) printf("------------------------------------\n");
4433 (void) printf("LABEL %d %s\n", l
,
4434 label
->cksum_valid
? "" : "(Bad label cksum)");
4435 (void) printf("------------------------------------\n");
4437 label
->header_printed
= B_TRUE
;
4441 print_l2arc_header(void)
4443 (void) printf("------------------------------------\n");
4444 (void) printf("L2ARC device header\n");
4445 (void) printf("------------------------------------\n");
4449 print_l2arc_log_blocks(void)
4451 (void) printf("------------------------------------\n");
4452 (void) printf("L2ARC device log blocks\n");
4453 (void) printf("------------------------------------\n");
4457 dump_l2arc_log_entries(uint64_t log_entries
,
4458 l2arc_log_ent_phys_t
*le
, uint64_t i
)
4460 for (int j
= 0; j
< log_entries
; j
++) {
4461 dva_t dva
= le
[j
].le_dva
;
4462 (void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4463 "vdev: %llu, offset: %llu\n",
4464 (u_longlong_t
)i
, j
+ 1,
4465 (u_longlong_t
)DVA_GET_ASIZE(&dva
),
4466 (u_longlong_t
)DVA_GET_VDEV(&dva
),
4467 (u_longlong_t
)DVA_GET_OFFSET(&dva
));
4468 (void) printf("|\t\t\t\tbirth: %llu\n",
4469 (u_longlong_t
)le
[j
].le_birth
);
4470 (void) printf("|\t\t\t\tlsize: %llu\n",
4471 (u_longlong_t
)L2BLK_GET_LSIZE((&le
[j
])->le_prop
));
4472 (void) printf("|\t\t\t\tpsize: %llu\n",
4473 (u_longlong_t
)L2BLK_GET_PSIZE((&le
[j
])->le_prop
));
4474 (void) printf("|\t\t\t\tcompr: %llu\n",
4475 (u_longlong_t
)L2BLK_GET_COMPRESS((&le
[j
])->le_prop
));
4476 (void) printf("|\t\t\t\tcomplevel: %llu\n",
4477 (u_longlong_t
)(&le
[j
])->le_complevel
);
4478 (void) printf("|\t\t\t\ttype: %llu\n",
4479 (u_longlong_t
)L2BLK_GET_TYPE((&le
[j
])->le_prop
));
4480 (void) printf("|\t\t\t\tprotected: %llu\n",
4481 (u_longlong_t
)L2BLK_GET_PROTECTED((&le
[j
])->le_prop
));
4482 (void) printf("|\t\t\t\tprefetch: %llu\n",
4483 (u_longlong_t
)L2BLK_GET_PREFETCH((&le
[j
])->le_prop
));
4484 (void) printf("|\t\t\t\taddress: %llu\n",
4485 (u_longlong_t
)le
[j
].le_daddr
);
4486 (void) printf("|\t\t\t\tARC state: %llu\n",
4487 (u_longlong_t
)L2BLK_GET_STATE((&le
[j
])->le_prop
));
4488 (void) printf("|\n");
4490 (void) printf("\n");
4494 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t
*lbps
)
4496 (void) printf("|\t\tdaddr: %llu\n", (u_longlong_t
)lbps
->lbp_daddr
);
4497 (void) printf("|\t\tpayload_asize: %llu\n",
4498 (u_longlong_t
)lbps
->lbp_payload_asize
);
4499 (void) printf("|\t\tpayload_start: %llu\n",
4500 (u_longlong_t
)lbps
->lbp_payload_start
);
4501 (void) printf("|\t\tlsize: %llu\n",
4502 (u_longlong_t
)L2BLK_GET_LSIZE(lbps
->lbp_prop
));
4503 (void) printf("|\t\tasize: %llu\n",
4504 (u_longlong_t
)L2BLK_GET_PSIZE(lbps
->lbp_prop
));
4505 (void) printf("|\t\tcompralgo: %llu\n",
4506 (u_longlong_t
)L2BLK_GET_COMPRESS(lbps
->lbp_prop
));
4507 (void) printf("|\t\tcksumalgo: %llu\n",
4508 (u_longlong_t
)L2BLK_GET_CHECKSUM(lbps
->lbp_prop
));
4509 (void) printf("|\n\n");
4513 dump_l2arc_log_blocks(int fd
, const l2arc_dev_hdr_phys_t
*l2dhdr
,
4514 l2arc_dev_hdr_phys_t
*rebuild
)
4516 l2arc_log_blk_phys_t this_lb
;
4518 l2arc_log_blkptr_t lbps
[2];
4525 print_l2arc_log_blocks();
4526 memcpy(lbps
, l2dhdr
->dh_start_lbps
, sizeof (lbps
));
4528 dev
.l2ad_evict
= l2dhdr
->dh_evict
;
4529 dev
.l2ad_start
= l2dhdr
->dh_start
;
4530 dev
.l2ad_end
= l2dhdr
->dh_end
;
4532 if (l2dhdr
->dh_start_lbps
[0].lbp_daddr
== 0) {
4533 /* no log blocks to read */
4534 if (!dump_opt
['q']) {
4535 (void) printf("No log blocks to read\n");
4536 (void) printf("\n");
4540 dev
.l2ad_hand
= lbps
[0].lbp_daddr
+
4541 L2BLK_GET_PSIZE((&lbps
[0])->lbp_prop
);
4544 dev
.l2ad_first
= !!(l2dhdr
->dh_flags
& L2ARC_DEV_HDR_EVICT_FIRST
);
4547 if (!l2arc_log_blkptr_valid(&dev
, &lbps
[0]))
4550 /* L2BLK_GET_PSIZE returns aligned size for log blocks */
4551 asize
= L2BLK_GET_PSIZE((&lbps
[0])->lbp_prop
);
4552 if (pread64(fd
, &this_lb
, asize
, lbps
[0].lbp_daddr
) != asize
) {
4553 if (!dump_opt
['q']) {
4554 (void) printf("Error while reading next log "
4560 fletcher_4_native_varsize(&this_lb
, asize
, &cksum
);
4561 if (!ZIO_CHECKSUM_EQUAL(cksum
, lbps
[0].lbp_cksum
)) {
4563 if (!dump_opt
['q']) {
4564 (void) printf("Invalid cksum\n");
4565 dump_l2arc_log_blkptr(&lbps
[0]);
4570 switch (L2BLK_GET_COMPRESS((&lbps
[0])->lbp_prop
)) {
4571 case ZIO_COMPRESS_OFF
:
4574 abd
= abd_alloc_for_io(asize
, B_TRUE
);
4575 abd_copy_from_buf_off(abd
, &this_lb
, 0, asize
);
4576 if (zio_decompress_data(L2BLK_GET_COMPRESS(
4577 (&lbps
[0])->lbp_prop
), abd
, &this_lb
,
4578 asize
, sizeof (this_lb
), NULL
) != 0) {
4579 (void) printf("L2ARC block decompression "
4588 if (this_lb
.lb_magic
== BSWAP_64(L2ARC_LOG_BLK_MAGIC
))
4589 byteswap_uint64_array(&this_lb
, sizeof (this_lb
));
4590 if (this_lb
.lb_magic
!= L2ARC_LOG_BLK_MAGIC
) {
4592 (void) printf("Invalid log block magic\n\n");
4596 rebuild
->dh_lb_count
++;
4597 rebuild
->dh_lb_asize
+= asize
;
4598 if (dump_opt
['l'] > 1 && !dump_opt
['q']) {
4599 (void) printf("lb[%4llu]\tmagic: %llu\n",
4600 (u_longlong_t
)rebuild
->dh_lb_count
,
4601 (u_longlong_t
)this_lb
.lb_magic
);
4602 dump_l2arc_log_blkptr(&lbps
[0]);
4605 if (dump_opt
['l'] > 2 && !dump_opt
['q'])
4606 dump_l2arc_log_entries(l2dhdr
->dh_log_entries
,
4608 rebuild
->dh_lb_count
);
4610 if (l2arc_range_check_overlap(lbps
[1].lbp_payload_start
,
4611 lbps
[0].lbp_payload_start
, dev
.l2ad_evict
) &&
4616 lbps
[1] = this_lb
.lb_prev_lbp
;
4619 if (!dump_opt
['q']) {
4620 (void) printf("log_blk_count:\t %llu with valid cksum\n",
4621 (u_longlong_t
)rebuild
->dh_lb_count
);
4622 (void) printf("\t\t %d with invalid cksum\n", failed
);
4623 (void) printf("log_blk_asize:\t %llu\n\n",
4624 (u_longlong_t
)rebuild
->dh_lb_asize
);
4629 dump_l2arc_header(int fd
)
4631 l2arc_dev_hdr_phys_t l2dhdr
= {0}, rebuild
= {0};
4632 int error
= B_FALSE
;
4634 if (pread64(fd
, &l2dhdr
, sizeof (l2dhdr
),
4635 VDEV_LABEL_START_SIZE
) != sizeof (l2dhdr
)) {
4638 if (l2dhdr
.dh_magic
== BSWAP_64(L2ARC_DEV_HDR_MAGIC
))
4639 byteswap_uint64_array(&l2dhdr
, sizeof (l2dhdr
));
4641 if (l2dhdr
.dh_magic
!= L2ARC_DEV_HDR_MAGIC
)
4646 (void) printf("L2ARC device header not found\n\n");
4647 /* Do not return an error here for backward compatibility */
4649 } else if (!dump_opt
['q']) {
4650 print_l2arc_header();
4652 (void) printf(" magic: %llu\n",
4653 (u_longlong_t
)l2dhdr
.dh_magic
);
4654 (void) printf(" version: %llu\n",
4655 (u_longlong_t
)l2dhdr
.dh_version
);
4656 (void) printf(" pool_guid: %llu\n",
4657 (u_longlong_t
)l2dhdr
.dh_spa_guid
);
4658 (void) printf(" flags: %llu\n",
4659 (u_longlong_t
)l2dhdr
.dh_flags
);
4660 (void) printf(" start_lbps[0]: %llu\n",
4662 l2dhdr
.dh_start_lbps
[0].lbp_daddr
);
4663 (void) printf(" start_lbps[1]: %llu\n",
4665 l2dhdr
.dh_start_lbps
[1].lbp_daddr
);
4666 (void) printf(" log_blk_ent: %llu\n",
4667 (u_longlong_t
)l2dhdr
.dh_log_entries
);
4668 (void) printf(" start: %llu\n",
4669 (u_longlong_t
)l2dhdr
.dh_start
);
4670 (void) printf(" end: %llu\n",
4671 (u_longlong_t
)l2dhdr
.dh_end
);
4672 (void) printf(" evict: %llu\n",
4673 (u_longlong_t
)l2dhdr
.dh_evict
);
4674 (void) printf(" lb_asize_refcount: %llu\n",
4675 (u_longlong_t
)l2dhdr
.dh_lb_asize
);
4676 (void) printf(" lb_count_refcount: %llu\n",
4677 (u_longlong_t
)l2dhdr
.dh_lb_count
);
4678 (void) printf(" trim_action_time: %llu\n",
4679 (u_longlong_t
)l2dhdr
.dh_trim_action_time
);
4680 (void) printf(" trim_state: %llu\n\n",
4681 (u_longlong_t
)l2dhdr
.dh_trim_state
);
4684 dump_l2arc_log_blocks(fd
, &l2dhdr
, &rebuild
);
4686 * The total aligned size of log blocks and the number of log blocks
4687 * reported in the header of the device may be less than what zdb
4688 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
4689 * This happens because dump_l2arc_log_blocks() lacks the memory
4690 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
4691 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
4692 * and dh_lb_count will be lower to begin with than what exists on the
4693 * device. This is normal and zdb should not exit with an error. The
4694 * opposite case should never happen though, the values reported in the
4695 * header should never be higher than what dump_l2arc_log_blocks() and
4696 * l2arc_rebuild() report. If this happens there is a leak in the
4697 * accounting of log blocks.
4699 if (l2dhdr
.dh_lb_asize
> rebuild
.dh_lb_asize
||
4700 l2dhdr
.dh_lb_count
> rebuild
.dh_lb_count
)
4707 dump_config_from_label(zdb_label_t
*label
, size_t buflen
, int l
)
4712 if ((dump_opt
['l'] < 3) && (first_label(label
->config
) != l
))
4715 print_label_header(label
, l
);
4716 dump_nvlist(label
->config_nv
, 4);
4717 print_label_numbers(" labels = ", label
->config
);
4719 if (dump_opt
['l'] >= 2)
4720 dump_nvlist_stats(label
->config_nv
, buflen
);
4723 #define ZDB_MAX_UB_HEADER_SIZE 32
4726 dump_label_uberblocks(zdb_label_t
*label
, uint64_t ashift
, int label_num
)
4730 char header
[ZDB_MAX_UB_HEADER_SIZE
];
4732 vd
.vdev_ashift
= ashift
;
4735 for (int i
= 0; i
< VDEV_UBERBLOCK_COUNT(&vd
); i
++) {
4736 uint64_t uoff
= VDEV_UBERBLOCK_OFFSET(&vd
, i
);
4737 uberblock_t
*ub
= (void *)((char *)&label
->label
+ uoff
);
4738 cksum_record_t
*rec
= label
->uberblocks
[i
];
4741 if (dump_opt
['u'] >= 2) {
4742 print_label_header(label
, label_num
);
4743 (void) printf(" Uberblock[%d] invalid\n", i
);
4748 if ((dump_opt
['u'] < 3) && (first_label(rec
) != label_num
))
4751 if ((dump_opt
['u'] < 4) &&
4752 (ub
->ub_mmp_magic
== MMP_MAGIC
) && ub
->ub_mmp_delay
&&
4753 (i
>= VDEV_UBERBLOCK_COUNT(&vd
) - MMP_BLOCKS_PER_LABEL
))
4756 print_label_header(label
, label_num
);
4757 (void) snprintf(header
, ZDB_MAX_UB_HEADER_SIZE
,
4758 " Uberblock[%d]\n", i
);
4759 dump_uberblock(ub
, header
, "");
4760 print_label_numbers(" labels = ", rec
);
4764 static char curpath
[PATH_MAX
];
4767 * Iterate through the path components, recursively passing
4768 * current one's obj and remaining path until we find the obj
4772 dump_path_impl(objset_t
*os
, uint64_t obj
, char *name
, uint64_t *retobj
)
4775 boolean_t header
= B_TRUE
;
4779 dmu_object_info_t doi
;
4781 if ((s
= strchr(name
, '/')) != NULL
)
4783 err
= zap_lookup(os
, obj
, name
, 8, 1, &child_obj
);
4785 (void) strlcat(curpath
, name
, sizeof (curpath
));
4788 (void) fprintf(stderr
, "failed to lookup %s: %s\n",
4789 curpath
, strerror(err
));
4793 child_obj
= ZFS_DIRENT_OBJ(child_obj
);
4794 err
= sa_buf_hold(os
, child_obj
, FTAG
, &db
);
4796 (void) fprintf(stderr
,
4797 "failed to get SA dbuf for obj %llu: %s\n",
4798 (u_longlong_t
)child_obj
, strerror(err
));
4801 dmu_object_info_from_db(db
, &doi
);
4802 sa_buf_rele(db
, FTAG
);
4804 if (doi
.doi_bonus_type
!= DMU_OT_SA
&&
4805 doi
.doi_bonus_type
!= DMU_OT_ZNODE
) {
4806 (void) fprintf(stderr
, "invalid bonus type %d for obj %llu\n",
4807 doi
.doi_bonus_type
, (u_longlong_t
)child_obj
);
4811 if (dump_opt
['v'] > 6) {
4812 (void) printf("obj=%llu %s type=%d bonustype=%d\n",
4813 (u_longlong_t
)child_obj
, curpath
, doi
.doi_type
,
4814 doi
.doi_bonus_type
);
4817 (void) strlcat(curpath
, "/", sizeof (curpath
));
4819 switch (doi
.doi_type
) {
4820 case DMU_OT_DIRECTORY_CONTENTS
:
4821 if (s
!= NULL
&& *(s
+ 1) != '\0')
4822 return (dump_path_impl(os
, child_obj
, s
+ 1, retobj
));
4824 case DMU_OT_PLAIN_FILE_CONTENTS
:
4825 if (retobj
!= NULL
) {
4826 *retobj
= child_obj
;
4828 dump_object(os
, child_obj
, dump_opt
['v'], &header
,
4833 (void) fprintf(stderr
, "object %llu has non-file/directory "
4834 "type %d\n", (u_longlong_t
)obj
, doi
.doi_type
);
4842 * Dump the blocks for the object specified by path inside the dataset.
4845 dump_path(char *ds
, char *path
, uint64_t *retobj
)
4851 err
= open_objset(ds
, FTAG
, &os
);
4855 err
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_ROOT_OBJ
, 8, 1, &root_obj
);
4857 (void) fprintf(stderr
, "can't lookup root znode: %s\n",
4859 close_objset(os
, FTAG
);
4863 (void) snprintf(curpath
, sizeof (curpath
), "dataset=%s path=/", ds
);
4865 err
= dump_path_impl(os
, root_obj
, path
, retobj
);
4867 close_objset(os
, FTAG
);
4872 dump_backup_bytes(objset_t
*os
, void *buf
, int len
, void *arg
)
4874 const char *p
= (const char *)buf
;
4880 /* Write the data out, handling short writes and signals. */
4881 while ((nwritten
= write(STDOUT_FILENO
, p
, len
)) < len
) {
4895 dump_backup(const char *pool
, uint64_t objset_id
, const char *flagstr
)
4897 boolean_t embed
= B_FALSE
;
4898 boolean_t large_block
= B_FALSE
;
4899 boolean_t compress
= B_FALSE
;
4900 boolean_t raw
= B_FALSE
;
4903 for (c
= flagstr
; c
!= NULL
&& *c
!= '\0'; c
++) {
4909 large_block
= B_TRUE
;
4918 fprintf(stderr
, "dump_backup: invalid flag "
4924 if (isatty(STDOUT_FILENO
)) {
4925 fprintf(stderr
, "dump_backup: stream cannot be written "
4931 dmu_send_outparams_t out
= {
4932 .dso_outfunc
= dump_backup_bytes
,
4933 .dso_dryrun
= B_FALSE
,
4936 int err
= dmu_send_obj(pool
, objset_id
, /* fromsnap */0, embed
,
4937 large_block
, compress
, raw
, /* saved */ B_FALSE
, STDOUT_FILENO
,
4940 fprintf(stderr
, "dump_backup: dmu_send_obj: %s\n",
4947 zdb_copy_object(objset_t
*os
, uint64_t srcobj
, char *destfile
)
4950 uint64_t size
, readsize
, oursize
, offset
;
4954 (void) printf("Copying object %" PRIu64
" to file %s\n", srcobj
,
4957 VERIFY3P(os
, ==, sa_os
);
4958 if ((err
= sa_handle_get(os
, srcobj
, NULL
, SA_HDL_PRIVATE
, &hdl
))) {
4959 (void) printf("Failed to get handle for SA znode\n");
4962 if ((err
= sa_lookup(hdl
, sa_attr_table
[ZPL_SIZE
], &size
, 8))) {
4963 (void) sa_handle_destroy(hdl
);
4966 (void) sa_handle_destroy(hdl
);
4968 (void) printf("Object %" PRIu64
" is %" PRIu64
" bytes\n", srcobj
,
4974 int fd
= open(destfile
, O_WRONLY
| O_CREAT
| O_TRUNC
, 0644);
4978 * We cap the size at 1 mebibyte here to prevent
4979 * allocation failures and nigh-infinite printing if the
4980 * object is extremely large.
4982 oursize
= MIN(size
, 1 << 20);
4984 char *buf
= kmem_alloc(oursize
, KM_NOSLEEP
);
4990 while (offset
< size
) {
4991 readsize
= MIN(size
- offset
, 1 << 20);
4992 err
= dmu_read(os
, srcobj
, offset
, readsize
, buf
, 0);
4994 (void) printf("got error %u from dmu_read\n", err
);
4995 kmem_free(buf
, oursize
);
4999 if (dump_opt
['v'] > 3) {
5000 (void) printf("Read offset=%" PRIu64
" size=%" PRIu64
5001 " error=%d\n", offset
, readsize
, err
);
5004 writesize
= write(fd
, buf
, readsize
);
5005 if (writesize
< 0) {
5008 } else if (writesize
!= readsize
) {
5009 /* Incomplete write */
5010 (void) fprintf(stderr
, "Short write, only wrote %llu of"
5011 " %" PRIu64
" bytes, exiting...\n",
5012 (u_longlong_t
)writesize
, readsize
);
5022 kmem_free(buf
, oursize
);
5028 label_cksum_valid(vdev_label_t
*label
, uint64_t offset
)
5030 zio_checksum_info_t
*ci
= &zio_checksum_table
[ZIO_CHECKSUM_LABEL
];
5031 zio_cksum_t expected_cksum
;
5032 zio_cksum_t actual_cksum
;
5033 zio_cksum_t verifier
;
5037 void *data
= (char *)label
+ offsetof(vdev_label_t
, vl_vdev_phys
);
5038 eck
= (zio_eck_t
*)((char *)(data
) + VDEV_PHYS_SIZE
) - 1;
5040 offset
+= offsetof(vdev_label_t
, vl_vdev_phys
);
5041 ZIO_SET_CHECKSUM(&verifier
, offset
, 0, 0, 0);
5043 byteswap
= (eck
->zec_magic
== BSWAP_64(ZEC_MAGIC
));
5045 byteswap_uint64_array(&verifier
, sizeof (zio_cksum_t
));
5047 expected_cksum
= eck
->zec_cksum
;
5048 eck
->zec_cksum
= verifier
;
5050 abd_t
*abd
= abd_get_from_buf(data
, VDEV_PHYS_SIZE
);
5051 ci
->ci_func
[byteswap
](abd
, VDEV_PHYS_SIZE
, NULL
, &actual_cksum
);
5055 byteswap_uint64_array(&expected_cksum
, sizeof (zio_cksum_t
));
5057 if (ZIO_CHECKSUM_EQUAL(actual_cksum
, expected_cksum
))
5064 dump_label(const char *dev
)
5066 char path
[MAXPATHLEN
];
5067 zdb_label_t labels
[VDEV_LABELS
] = {{{{0}}}};
5068 uint64_t psize
, ashift
, l2cache
;
5069 struct stat64 statbuf
;
5070 boolean_t config_found
= B_FALSE
;
5071 boolean_t error
= B_FALSE
;
5072 boolean_t read_l2arc_header
= B_FALSE
;
5073 avl_tree_t config_tree
;
5074 avl_tree_t uberblock_tree
;
5075 void *node
, *cookie
;
5079 * Check if we were given absolute path and use it as is.
5080 * Otherwise if the provided vdev name doesn't point to a file,
5081 * try prepending expected disk paths and partition numbers.
5083 (void) strlcpy(path
, dev
, sizeof (path
));
5084 if (dev
[0] != '/' && stat64(path
, &statbuf
) != 0) {
5087 error
= zfs_resolve_shortname(dev
, path
, MAXPATHLEN
);
5088 if (error
== 0 && zfs_dev_is_whole_disk(path
)) {
5089 if (zfs_append_partition(path
, MAXPATHLEN
) == -1)
5093 if (error
|| (stat64(path
, &statbuf
) != 0)) {
5094 (void) printf("failed to find device %s, try "
5095 "specifying absolute path instead\n", dev
);
5100 if ((fd
= open64(path
, O_RDONLY
)) < 0) {
5101 (void) printf("cannot open '%s': %s\n", path
, strerror(errno
));
5105 if (fstat64_blk(fd
, &statbuf
) != 0) {
5106 (void) printf("failed to stat '%s': %s\n", path
,
5112 if (S_ISBLK(statbuf
.st_mode
) && zfs_dev_flush(fd
) != 0)
5113 (void) printf("failed to invalidate cache '%s' : %s\n", path
,
5116 avl_create(&config_tree
, cksum_record_compare
,
5117 sizeof (cksum_record_t
), offsetof(cksum_record_t
, link
));
5118 avl_create(&uberblock_tree
, cksum_record_compare
,
5119 sizeof (cksum_record_t
), offsetof(cksum_record_t
, link
));
5121 psize
= statbuf
.st_size
;
5122 psize
= P2ALIGN(psize
, (uint64_t)sizeof (vdev_label_t
));
5123 ashift
= SPA_MINBLOCKSHIFT
;
5126 * 1. Read the label from disk
5127 * 2. Verify label cksum
5128 * 3. Unpack the configuration and insert in config tree.
5129 * 4. Traverse all uberblocks and insert in uberblock tree.
5131 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
5132 zdb_label_t
*label
= &labels
[l
];
5133 char *buf
= label
->label
.vl_vdev_phys
.vp_nvlist
;
5134 size_t buflen
= sizeof (label
->label
.vl_vdev_phys
.vp_nvlist
);
5136 cksum_record_t
*rec
;
5140 label
->label_offset
= vdev_label_offset(psize
, l
, 0);
5142 if (pread64(fd
, &label
->label
, sizeof (label
->label
),
5143 label
->label_offset
) != sizeof (label
->label
)) {
5145 (void) printf("failed to read label %d\n", l
);
5146 label
->read_failed
= B_TRUE
;
5151 label
->read_failed
= B_FALSE
;
5152 label
->cksum_valid
= label_cksum_valid(&label
->label
,
5153 label
->label_offset
);
5155 if (nvlist_unpack(buf
, buflen
, &config
, 0) == 0) {
5156 nvlist_t
*vdev_tree
= NULL
;
5159 if ((nvlist_lookup_nvlist(config
,
5160 ZPOOL_CONFIG_VDEV_TREE
, &vdev_tree
) != 0) ||
5161 (nvlist_lookup_uint64(vdev_tree
,
5162 ZPOOL_CONFIG_ASHIFT
, &ashift
) != 0))
5163 ashift
= SPA_MINBLOCKSHIFT
;
5165 if (nvlist_size(config
, &size
, NV_ENCODE_XDR
) != 0)
5168 /* If the device is a cache device read the header. */
5169 if (!read_l2arc_header
) {
5170 if (nvlist_lookup_uint64(config
,
5171 ZPOOL_CONFIG_POOL_STATE
, &l2cache
) == 0 &&
5172 l2cache
== POOL_STATE_L2CACHE
) {
5173 read_l2arc_header
= B_TRUE
;
5177 fletcher_4_native_varsize(buf
, size
, &cksum
);
5178 rec
= cksum_record_insert(&config_tree
, &cksum
, l
);
5180 label
->config
= rec
;
5181 label
->config_nv
= config
;
5182 config_found
= B_TRUE
;
5187 vd
.vdev_ashift
= ashift
;
5190 for (int i
= 0; i
< VDEV_UBERBLOCK_COUNT(&vd
); i
++) {
5191 uint64_t uoff
= VDEV_UBERBLOCK_OFFSET(&vd
, i
);
5192 uberblock_t
*ub
= (void *)((char *)label
+ uoff
);
5194 if (uberblock_verify(ub
))
5197 fletcher_4_native_varsize(ub
, sizeof (*ub
), &cksum
);
5198 rec
= cksum_record_insert(&uberblock_tree
, &cksum
, l
);
5200 label
->uberblocks
[i
] = rec
;
5205 * Dump the label and uberblocks.
5207 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
5208 zdb_label_t
*label
= &labels
[l
];
5209 size_t buflen
= sizeof (label
->label
.vl_vdev_phys
.vp_nvlist
);
5211 if (label
->read_failed
== B_TRUE
)
5214 if (label
->config_nv
) {
5215 dump_config_from_label(label
, buflen
, l
);
5218 (void) printf("failed to unpack label %d\n", l
);
5222 dump_label_uberblocks(label
, ashift
, l
);
5224 nvlist_free(label
->config_nv
);
5228 * Dump the L2ARC header, if existent.
5230 if (read_l2arc_header
)
5231 error
|= dump_l2arc_header(fd
);
5234 while ((node
= avl_destroy_nodes(&config_tree
, &cookie
)) != NULL
)
5235 umem_free(node
, sizeof (cksum_record_t
));
5238 while ((node
= avl_destroy_nodes(&uberblock_tree
, &cookie
)) != NULL
)
5239 umem_free(node
, sizeof (cksum_record_t
));
5241 avl_destroy(&config_tree
);
5242 avl_destroy(&uberblock_tree
);
5246 return (config_found
== B_FALSE
? 2 :
5247 (error
== B_TRUE
? 1 : 0));
5250 static uint64_t dataset_feature_count
[SPA_FEATURES
];
5251 static uint64_t global_feature_count
[SPA_FEATURES
];
5252 static uint64_t remap_deadlist_count
= 0;
5255 dump_one_objset(const char *dsname
, void *arg
)
5262 error
= open_objset(dsname
, FTAG
, &os
);
5266 for (f
= 0; f
< SPA_FEATURES
; f
++) {
5267 if (!dsl_dataset_feature_is_active(dmu_objset_ds(os
), f
))
5269 ASSERT(spa_feature_table
[f
].fi_flags
&
5270 ZFEATURE_FLAG_PER_DATASET
);
5271 dataset_feature_count
[f
]++;
5274 if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os
))) {
5275 remap_deadlist_count
++;
5278 for (dsl_bookmark_node_t
*dbn
=
5279 avl_first(&dmu_objset_ds(os
)->ds_bookmarks
); dbn
!= NULL
;
5280 dbn
= AVL_NEXT(&dmu_objset_ds(os
)->ds_bookmarks
, dbn
)) {
5281 mos_obj_refd(dbn
->dbn_phys
.zbm_redaction_obj
);
5282 if (dbn
->dbn_phys
.zbm_redaction_obj
!= 0) {
5283 global_feature_count
[
5284 SPA_FEATURE_REDACTION_BOOKMARKS
]++;
5285 objset_t
*mos
= os
->os_spa
->spa_meta_objset
;
5287 VERIFY0(dnode_hold(mos
,
5288 dbn
->dbn_phys
.zbm_redaction_obj
, FTAG
, &rl
));
5289 if (rl
->dn_have_spill
) {
5290 global_feature_count
[
5291 SPA_FEATURE_REDACTION_LIST_SPILL
]++;
5294 if (dbn
->dbn_phys
.zbm_flags
& ZBM_FLAG_HAS_FBN
)
5295 global_feature_count
[SPA_FEATURE_BOOKMARK_WRITTEN
]++;
5298 if (dsl_deadlist_is_open(&dmu_objset_ds(os
)->ds_dir
->dd_livelist
) &&
5299 !dmu_objset_is_snapshot(os
)) {
5300 global_feature_count
[SPA_FEATURE_LIVELIST
]++;
5304 close_objset(os
, FTAG
);
5305 fuid_table_destroy();
5312 #define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5313 typedef struct zdb_blkstats
{
5319 uint64_t zb_ditto_samevdev
;
5320 uint64_t zb_ditto_same_ms
;
5321 uint64_t zb_psize_histogram
[PSIZE_HISTO_SIZE
];
5325 * Extended object types to report deferred frees and dedup auto-ditto blocks.
5327 #define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0)
5328 #define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1)
5329 #define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2)
5330 #define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3)
5332 static const char *zdb_ot_extname
[] = {
5339 #define ZB_TOTAL DN_MAX_LEVELS
5340 #define SPA_MAX_FOR_16M (SPA_MAXBLOCKSHIFT+1)
5342 typedef struct zdb_brt_entry
{
5344 uint64_t zbre_refcount
;
5345 avl_node_t zbre_node
;
5348 typedef struct zdb_cb
{
5349 zdb_blkstats_t zcb_type
[ZB_TOTAL
+ 1][ZDB_OT_TOTAL
+ 1];
5350 uint64_t zcb_removing_size
;
5351 uint64_t zcb_checkpoint_size
;
5352 uint64_t zcb_dedup_asize
;
5353 uint64_t zcb_dedup_blocks
;
5354 uint64_t zcb_clone_asize
;
5355 uint64_t zcb_clone_blocks
;
5356 uint64_t zcb_psize_count
[SPA_MAX_FOR_16M
];
5357 uint64_t zcb_lsize_count
[SPA_MAX_FOR_16M
];
5358 uint64_t zcb_asize_count
[SPA_MAX_FOR_16M
];
5359 uint64_t zcb_psize_len
[SPA_MAX_FOR_16M
];
5360 uint64_t zcb_lsize_len
[SPA_MAX_FOR_16M
];
5361 uint64_t zcb_asize_len
[SPA_MAX_FOR_16M
];
5362 uint64_t zcb_psize_total
;
5363 uint64_t zcb_lsize_total
;
5364 uint64_t zcb_asize_total
;
5365 uint64_t zcb_embedded_blocks
[NUM_BP_EMBEDDED_TYPES
];
5366 uint64_t zcb_embedded_histogram
[NUM_BP_EMBEDDED_TYPES
]
5367 [BPE_PAYLOAD_SIZE
+ 1];
5369 hrtime_t zcb_lastprint
;
5370 uint64_t zcb_totalasize
;
5371 uint64_t zcb_errors
[256];
5375 uint32_t **zcb_vd_obsolete_counts
;
5377 boolean_t zcb_brt_is_active
;
5380 /* test if two DVA offsets from same vdev are within the same metaslab */
5382 same_metaslab(spa_t
*spa
, uint64_t vdev
, uint64_t off1
, uint64_t off2
)
5384 vdev_t
*vd
= vdev_lookup_top(spa
, vdev
);
5385 uint64_t ms_shift
= vd
->vdev_ms_shift
;
5387 return ((off1
>> ms_shift
) == (off2
>> ms_shift
));
5391 * Used to simplify reporting of the histogram data.
5393 typedef struct one_histo
{
5397 uint64_t cumulative
;
5401 * The number of separate histograms processed for psize, lsize and asize.
5406 * This routine will create a fixed column size output of three different
5407 * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5408 * the count, length and cumulative length of the psize, lsize and
5411 * All three types of blocks are listed on a single line
5413 * By default the table is printed in nicenumber format (e.g. 123K) but
5414 * if the '-P' parameter is specified then the full raw number (parseable)
5418 dump_size_histograms(zdb_cb_t
*zcb
)
5421 * A temporary buffer that allows us to convert a number into
5422 * a string using zdb_nicenumber to allow either raw or human
5423 * readable numbers to be output.
5428 * Define titles which are used in the headers of the tables
5429 * printed by this routine.
5431 const char blocksize_title1
[] = "block";
5432 const char blocksize_title2
[] = "size";
5433 const char count_title
[] = "Count";
5434 const char length_title
[] = "Size";
5435 const char cumulative_title
[] = "Cum.";
5438 * Setup the histogram arrays (psize, lsize, and asize).
5440 one_histo_t parm_histo
[NUM_HISTO
];
5442 parm_histo
[0].name
= "psize";
5443 parm_histo
[0].count
= zcb
->zcb_psize_count
;
5444 parm_histo
[0].len
= zcb
->zcb_psize_len
;
5445 parm_histo
[0].cumulative
= 0;
5447 parm_histo
[1].name
= "lsize";
5448 parm_histo
[1].count
= zcb
->zcb_lsize_count
;
5449 parm_histo
[1].len
= zcb
->zcb_lsize_len
;
5450 parm_histo
[1].cumulative
= 0;
5452 parm_histo
[2].name
= "asize";
5453 parm_histo
[2].count
= zcb
->zcb_asize_count
;
5454 parm_histo
[2].len
= zcb
->zcb_asize_len
;
5455 parm_histo
[2].cumulative
= 0;
5458 (void) printf("\nBlock Size Histogram\n");
5460 * Print the first line titles
5463 (void) printf("\n%s\t", blocksize_title1
);
5465 (void) printf("\n%7s ", blocksize_title1
);
5467 for (int j
= 0; j
< NUM_HISTO
; j
++) {
5468 if (dump_opt
['P']) {
5469 if (j
< NUM_HISTO
- 1) {
5470 (void) printf("%s\t\t\t", parm_histo
[j
].name
);
5472 /* Don't print trailing spaces */
5473 (void) printf(" %s", parm_histo
[j
].name
);
5476 if (j
< NUM_HISTO
- 1) {
5477 /* Left aligned strings in the output */
5478 (void) printf("%-7s ",
5479 parm_histo
[j
].name
);
5481 /* Don't print trailing spaces */
5482 (void) printf("%s", parm_histo
[j
].name
);
5486 (void) printf("\n");
5489 * Print the second line titles
5491 if (dump_opt
['P']) {
5492 (void) printf("%s\t", blocksize_title2
);
5494 (void) printf("%7s ", blocksize_title2
);
5497 for (int i
= 0; i
< NUM_HISTO
; i
++) {
5498 if (dump_opt
['P']) {
5499 (void) printf("%s\t%s\t%s\t",
5500 count_title
, length_title
, cumulative_title
);
5502 (void) printf("%7s%7s%7s",
5503 count_title
, length_title
, cumulative_title
);
5506 (void) printf("\n");
5511 for (int i
= SPA_MINBLOCKSHIFT
; i
< SPA_MAX_FOR_16M
; i
++) {
5514 * Print the first column showing the blocksize
5516 zdb_nicenum((1ULL << i
), numbuf
, sizeof (numbuf
));
5518 if (dump_opt
['P']) {
5519 printf("%s", numbuf
);
5521 printf("%7s:", numbuf
);
5525 * Print the remaining set of 3 columns per size:
5526 * for psize, lsize and asize
5528 for (int j
= 0; j
< NUM_HISTO
; j
++) {
5529 parm_histo
[j
].cumulative
+= parm_histo
[j
].len
[i
];
5531 zdb_nicenum(parm_histo
[j
].count
[i
],
5532 numbuf
, sizeof (numbuf
));
5534 (void) printf("\t%s", numbuf
);
5536 (void) printf("%7s", numbuf
);
5538 zdb_nicenum(parm_histo
[j
].len
[i
],
5539 numbuf
, sizeof (numbuf
));
5541 (void) printf("\t%s", numbuf
);
5543 (void) printf("%7s", numbuf
);
5545 zdb_nicenum(parm_histo
[j
].cumulative
,
5546 numbuf
, sizeof (numbuf
));
5548 (void) printf("\t%s", numbuf
);
5550 (void) printf("%7s", numbuf
);
5552 (void) printf("\n");
5557 zdb_count_block(zdb_cb_t
*zcb
, zilog_t
*zilog
, const blkptr_t
*bp
,
5558 dmu_object_type_t type
)
5560 uint64_t refcnt
= 0;
5563 ASSERT(type
< ZDB_OT_TOTAL
);
5565 if (zilog
&& zil_bp_tree_add(zilog
, bp
) != 0)
5568 spa_config_enter(zcb
->zcb_spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5570 for (i
= 0; i
< 4; i
++) {
5571 int l
= (i
< 2) ? BP_GET_LEVEL(bp
) : ZB_TOTAL
;
5572 int t
= (i
& 1) ? type
: ZDB_OT_TOTAL
;
5574 zdb_blkstats_t
*zb
= &zcb
->zcb_type
[l
][t
];
5576 zb
->zb_asize
+= BP_GET_ASIZE(bp
);
5577 zb
->zb_lsize
+= BP_GET_LSIZE(bp
);
5578 zb
->zb_psize
+= BP_GET_PSIZE(bp
);
5582 * The histogram is only big enough to record blocks up to
5583 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
5586 unsigned idx
= BP_GET_PSIZE(bp
) >> SPA_MINBLOCKSHIFT
;
5587 idx
= MIN(idx
, SPA_OLD_MAXBLOCKSIZE
/ SPA_MINBLOCKSIZE
+ 1);
5588 zb
->zb_psize_histogram
[idx
]++;
5590 zb
->zb_gangs
+= BP_COUNT_GANG(bp
);
5592 switch (BP_GET_NDVAS(bp
)) {
5594 if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5595 DVA_GET_VDEV(&bp
->blk_dva
[1])) {
5596 zb
->zb_ditto_samevdev
++;
5598 if (same_metaslab(zcb
->zcb_spa
,
5599 DVA_GET_VDEV(&bp
->blk_dva
[0]),
5600 DVA_GET_OFFSET(&bp
->blk_dva
[0]),
5601 DVA_GET_OFFSET(&bp
->blk_dva
[1])))
5602 zb
->zb_ditto_same_ms
++;
5606 equal
= (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5607 DVA_GET_VDEV(&bp
->blk_dva
[1])) +
5608 (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5609 DVA_GET_VDEV(&bp
->blk_dva
[2])) +
5610 (DVA_GET_VDEV(&bp
->blk_dva
[1]) ==
5611 DVA_GET_VDEV(&bp
->blk_dva
[2]));
5613 zb
->zb_ditto_samevdev
++;
5615 if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5616 DVA_GET_VDEV(&bp
->blk_dva
[1]) &&
5617 same_metaslab(zcb
->zcb_spa
,
5618 DVA_GET_VDEV(&bp
->blk_dva
[0]),
5619 DVA_GET_OFFSET(&bp
->blk_dva
[0]),
5620 DVA_GET_OFFSET(&bp
->blk_dva
[1])))
5621 zb
->zb_ditto_same_ms
++;
5622 else if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5623 DVA_GET_VDEV(&bp
->blk_dva
[2]) &&
5624 same_metaslab(zcb
->zcb_spa
,
5625 DVA_GET_VDEV(&bp
->blk_dva
[0]),
5626 DVA_GET_OFFSET(&bp
->blk_dva
[0]),
5627 DVA_GET_OFFSET(&bp
->blk_dva
[2])))
5628 zb
->zb_ditto_same_ms
++;
5629 else if (DVA_GET_VDEV(&bp
->blk_dva
[1]) ==
5630 DVA_GET_VDEV(&bp
->blk_dva
[2]) &&
5631 same_metaslab(zcb
->zcb_spa
,
5632 DVA_GET_VDEV(&bp
->blk_dva
[1]),
5633 DVA_GET_OFFSET(&bp
->blk_dva
[1]),
5634 DVA_GET_OFFSET(&bp
->blk_dva
[2])))
5635 zb
->zb_ditto_same_ms
++;
5641 spa_config_exit(zcb
->zcb_spa
, SCL_CONFIG
, FTAG
);
5643 if (BP_IS_EMBEDDED(bp
)) {
5644 zcb
->zcb_embedded_blocks
[BPE_GET_ETYPE(bp
)]++;
5645 zcb
->zcb_embedded_histogram
[BPE_GET_ETYPE(bp
)]
5646 [BPE_GET_PSIZE(bp
)]++;
5650 * The binning histogram bins by powers of two up to
5651 * SPA_MAXBLOCKSIZE rather than creating bins for
5652 * every possible blocksize found in the pool.
5654 int bin
= highbit64(BP_GET_PSIZE(bp
)) - 1;
5656 zcb
->zcb_psize_count
[bin
]++;
5657 zcb
->zcb_psize_len
[bin
] += BP_GET_PSIZE(bp
);
5658 zcb
->zcb_psize_total
+= BP_GET_PSIZE(bp
);
5660 bin
= highbit64(BP_GET_LSIZE(bp
)) - 1;
5662 zcb
->zcb_lsize_count
[bin
]++;
5663 zcb
->zcb_lsize_len
[bin
] += BP_GET_LSIZE(bp
);
5664 zcb
->zcb_lsize_total
+= BP_GET_LSIZE(bp
);
5666 bin
= highbit64(BP_GET_ASIZE(bp
)) - 1;
5668 zcb
->zcb_asize_count
[bin
]++;
5669 zcb
->zcb_asize_len
[bin
] += BP_GET_ASIZE(bp
);
5670 zcb
->zcb_asize_total
+= BP_GET_ASIZE(bp
);
5672 if (zcb
->zcb_brt_is_active
&& brt_maybe_exists(zcb
->zcb_spa
, bp
)) {
5674 * Cloned blocks are special. We need to count them, so we can
5675 * later uncount them when reporting leaked space, and we must
5676 * only claim them them once.
5678 * To do this, we keep our own in-memory BRT. For each block
5679 * we haven't seen before, we look it up in the real BRT and
5680 * if its there, we note it and its refcount then proceed as
5681 * normal. If we see the block again, we count it as a clone
5682 * and then give it no further consideration.
5684 zdb_brt_entry_t zbre_search
, *zbre
;
5687 zbre_search
.zbre_dva
= bp
->blk_dva
[0];
5688 zbre
= avl_find(&zcb
->zcb_brt
, &zbre_search
, &where
);
5690 zcb
->zcb_clone_asize
+= BP_GET_ASIZE(bp
);
5691 zcb
->zcb_clone_blocks
++;
5693 zbre
->zbre_refcount
--;
5694 if (zbre
->zbre_refcount
== 0) {
5695 avl_remove(&zcb
->zcb_brt
, zbre
);
5696 umem_free(zbre
, sizeof (zdb_brt_entry_t
));
5701 uint64_t crefcnt
= brt_entry_get_refcount(zcb
->zcb_spa
, bp
);
5703 zbre
= umem_zalloc(sizeof (zdb_brt_entry_t
),
5705 zbre
->zbre_dva
= bp
->blk_dva
[0];
5706 zbre
->zbre_refcount
= crefcnt
;
5707 avl_insert(&zcb
->zcb_brt
, zbre
, where
);
5714 if (BP_GET_DEDUP(bp
)) {
5718 ddt
= ddt_select(zcb
->zcb_spa
, bp
);
5720 dde
= ddt_lookup(ddt
, bp
, B_FALSE
);
5725 ddt_phys_t
*ddp
= ddt_phys_select(dde
, bp
);
5726 ddt_phys_decref(ddp
);
5727 refcnt
= ddp
->ddp_refcnt
;
5728 if (ddt_phys_total_refcnt(dde
) == 0)
5729 ddt_remove(ddt
, dde
);
5734 VERIFY3U(zio_wait(zio_claim(NULL
, zcb
->zcb_spa
,
5735 refcnt
? 0 : spa_min_claim_txg(zcb
->zcb_spa
),
5736 bp
, NULL
, NULL
, ZIO_FLAG_CANFAIL
)), ==, 0);
5740 zdb_blkptr_done(zio_t
*zio
)
5742 spa_t
*spa
= zio
->io_spa
;
5743 blkptr_t
*bp
= zio
->io_bp
;
5744 int ioerr
= zio
->io_error
;
5745 zdb_cb_t
*zcb
= zio
->io_private
;
5746 zbookmark_phys_t
*zb
= &zio
->io_bookmark
;
5748 mutex_enter(&spa
->spa_scrub_lock
);
5749 spa
->spa_load_verify_bytes
-= BP_GET_PSIZE(bp
);
5750 cv_broadcast(&spa
->spa_scrub_io_cv
);
5752 if (ioerr
&& !(zio
->io_flags
& ZIO_FLAG_SPECULATIVE
)) {
5753 char blkbuf
[BP_SPRINTF_LEN
];
5755 zcb
->zcb_haderrors
= 1;
5756 zcb
->zcb_errors
[ioerr
]++;
5758 if (dump_opt
['b'] >= 2)
5759 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
5763 (void) printf("zdb_blkptr_cb: "
5764 "Got error %d reading "
5765 "<%llu, %llu, %lld, %llx> %s -- skipping\n",
5767 (u_longlong_t
)zb
->zb_objset
,
5768 (u_longlong_t
)zb
->zb_object
,
5769 (u_longlong_t
)zb
->zb_level
,
5770 (u_longlong_t
)zb
->zb_blkid
,
5773 mutex_exit(&spa
->spa_scrub_lock
);
5775 abd_free(zio
->io_abd
);
5779 zdb_blkptr_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
5780 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
5782 zdb_cb_t
*zcb
= arg
;
5783 dmu_object_type_t type
;
5784 boolean_t is_metadata
;
5786 if (zb
->zb_level
== ZB_DNODE_LEVEL
)
5789 if (dump_opt
['b'] >= 5 && bp
->blk_birth
> 0) {
5790 char blkbuf
[BP_SPRINTF_LEN
];
5791 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
5792 (void) printf("objset %llu object %llu "
5793 "level %lld offset 0x%llx %s\n",
5794 (u_longlong_t
)zb
->zb_objset
,
5795 (u_longlong_t
)zb
->zb_object
,
5796 (longlong_t
)zb
->zb_level
,
5797 (u_longlong_t
)blkid2offset(dnp
, bp
, zb
),
5801 if (BP_IS_HOLE(bp
) || BP_IS_REDACTED(bp
))
5804 type
= BP_GET_TYPE(bp
);
5806 zdb_count_block(zcb
, zilog
, bp
,
5807 (type
& DMU_OT_NEWTYPE
) ? ZDB_OT_OTHER
: type
);
5809 is_metadata
= (BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
));
5811 if (!BP_IS_EMBEDDED(bp
) &&
5812 (dump_opt
['c'] > 1 || (dump_opt
['c'] && is_metadata
))) {
5813 size_t size
= BP_GET_PSIZE(bp
);
5814 abd_t
*abd
= abd_alloc(size
, B_FALSE
);
5815 int flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
;
5817 /* If it's an intent log block, failure is expected. */
5818 if (zb
->zb_level
== ZB_ZIL_LEVEL
)
5819 flags
|= ZIO_FLAG_SPECULATIVE
;
5821 mutex_enter(&spa
->spa_scrub_lock
);
5822 while (spa
->spa_load_verify_bytes
> max_inflight_bytes
)
5823 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
5824 spa
->spa_load_verify_bytes
+= size
;
5825 mutex_exit(&spa
->spa_scrub_lock
);
5827 zio_nowait(zio_read(NULL
, spa
, bp
, abd
, size
,
5828 zdb_blkptr_done
, zcb
, ZIO_PRIORITY_ASYNC_READ
, flags
, zb
));
5831 zcb
->zcb_readfails
= 0;
5833 /* only call gethrtime() every 100 blocks */
5840 if (dump_opt
['b'] < 5 && gethrtime() > zcb
->zcb_lastprint
+ NANOSEC
) {
5841 uint64_t now
= gethrtime();
5843 uint64_t bytes
= zcb
->zcb_type
[ZB_TOTAL
][ZDB_OT_TOTAL
].zb_asize
;
5844 uint64_t kb_per_sec
=
5845 1 + bytes
/ (1 + ((now
- zcb
->zcb_start
) / 1000 / 1000));
5846 uint64_t sec_remaining
=
5847 (zcb
->zcb_totalasize
- bytes
) / 1024 / kb_per_sec
;
5849 /* make sure nicenum has enough space */
5850 _Static_assert(sizeof (buf
) >= NN_NUMBUF_SZ
, "buf truncated");
5852 zfs_nicebytes(bytes
, buf
, sizeof (buf
));
5853 (void) fprintf(stderr
,
5854 "\r%5s completed (%4"PRIu64
"MB/s) "
5855 "estimated time remaining: "
5856 "%"PRIu64
"hr %02"PRIu64
"min %02"PRIu64
"sec ",
5857 buf
, kb_per_sec
/ 1024,
5858 sec_remaining
/ 60 / 60,
5859 sec_remaining
/ 60 % 60,
5860 sec_remaining
% 60);
5862 zcb
->zcb_lastprint
= now
;
5869 zdb_leak(void *arg
, uint64_t start
, uint64_t size
)
5873 (void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
5874 (u_longlong_t
)vd
->vdev_id
, (u_longlong_t
)start
, (u_longlong_t
)size
);
5877 static metaslab_ops_t zdb_metaslab_ops
= {
5882 load_unflushed_svr_segs_cb(spa_t
*spa
, space_map_entry_t
*sme
,
5883 uint64_t txg
, void *arg
)
5885 spa_vdev_removal_t
*svr
= arg
;
5887 uint64_t offset
= sme
->sme_offset
;
5888 uint64_t size
= sme
->sme_run
;
5890 /* skip vdevs we don't care about */
5891 if (sme
->sme_vdev
!= svr
->svr_vdev_id
)
5894 vdev_t
*vd
= vdev_lookup_top(spa
, sme
->sme_vdev
);
5895 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
5896 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
5898 if (txg
< metaslab_unflushed_txg(ms
))
5901 if (sme
->sme_type
== SM_ALLOC
)
5902 range_tree_add(svr
->svr_allocd_segs
, offset
, size
);
5904 range_tree_remove(svr
->svr_allocd_segs
, offset
, size
);
5910 claim_segment_impl_cb(uint64_t inner_offset
, vdev_t
*vd
, uint64_t offset
,
5911 uint64_t size
, void *arg
)
5913 (void) inner_offset
, (void) arg
;
5916 * This callback was called through a remap from
5917 * a device being removed. Therefore, the vdev that
5918 * this callback is applied to is a concrete
5921 ASSERT(vdev_is_concrete(vd
));
5923 VERIFY0(metaslab_claim_impl(vd
, offset
, size
,
5924 spa_min_claim_txg(vd
->vdev_spa
)));
5928 claim_segment_cb(void *arg
, uint64_t offset
, uint64_t size
)
5932 vdev_indirect_ops
.vdev_op_remap(vd
, offset
, size
,
5933 claim_segment_impl_cb
, NULL
);
5937 * After accounting for all allocated blocks that are directly referenced,
5938 * we might have missed a reference to a block from a partially complete
5939 * (and thus unused) indirect mapping object. We perform a secondary pass
5940 * through the metaslabs we have already mapped and claim the destination
5944 zdb_claim_removing(spa_t
*spa
, zdb_cb_t
*zcb
)
5949 if (spa
->spa_vdev_removal
== NULL
)
5952 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5954 spa_vdev_removal_t
*svr
= spa
->spa_vdev_removal
;
5955 vdev_t
*vd
= vdev_lookup_top(spa
, svr
->svr_vdev_id
);
5956 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
5958 ASSERT0(range_tree_space(svr
->svr_allocd_segs
));
5960 range_tree_t
*allocs
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
5961 for (uint64_t msi
= 0; msi
< vd
->vdev_ms_count
; msi
++) {
5962 metaslab_t
*msp
= vd
->vdev_ms
[msi
];
5964 ASSERT0(range_tree_space(allocs
));
5965 if (msp
->ms_sm
!= NULL
)
5966 VERIFY0(space_map_load(msp
->ms_sm
, allocs
, SM_ALLOC
));
5967 range_tree_vacate(allocs
, range_tree_add
, svr
->svr_allocd_segs
);
5969 range_tree_destroy(allocs
);
5971 iterate_through_spacemap_logs(spa
, load_unflushed_svr_segs_cb
, svr
);
5974 * Clear everything past what has been synced,
5975 * because we have not allocated mappings for
5978 range_tree_clear(svr
->svr_allocd_segs
,
5979 vdev_indirect_mapping_max_offset(vim
),
5980 vd
->vdev_asize
- vdev_indirect_mapping_max_offset(vim
));
5982 zcb
->zcb_removing_size
+= range_tree_space(svr
->svr_allocd_segs
);
5983 range_tree_vacate(svr
->svr_allocd_segs
, claim_segment_cb
, vd
);
5985 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5989 increment_indirect_mapping_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
5993 zdb_cb_t
*zcb
= arg
;
5994 spa_t
*spa
= zcb
->zcb_spa
;
5996 const dva_t
*dva
= &bp
->blk_dva
[0];
5999 ASSERT(!dump_opt
['L']);
6000 ASSERT3U(BP_GET_NDVAS(bp
), ==, 1);
6002 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
6003 vd
= vdev_lookup_top(zcb
->zcb_spa
, DVA_GET_VDEV(dva
));
6004 ASSERT3P(vd
, !=, NULL
);
6005 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
6007 ASSERT(vd
->vdev_indirect_config
.vic_mapping_object
!= 0);
6008 ASSERT3P(zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
], !=, NULL
);
6010 vdev_indirect_mapping_increment_obsolete_count(
6011 vd
->vdev_indirect_mapping
,
6012 DVA_GET_OFFSET(dva
), DVA_GET_ASIZE(dva
),
6013 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
]);
6019 zdb_load_obsolete_counts(vdev_t
*vd
)
6021 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6022 spa_t
*spa
= vd
->vdev_spa
;
6023 spa_condensing_indirect_phys_t
*scip
=
6024 &spa
->spa_condensing_indirect_phys
;
6025 uint64_t obsolete_sm_object
;
6028 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
6029 EQUIV(obsolete_sm_object
!= 0, vd
->vdev_obsolete_sm
!= NULL
);
6030 counts
= vdev_indirect_mapping_load_obsolete_counts(vim
);
6031 if (vd
->vdev_obsolete_sm
!= NULL
) {
6032 vdev_indirect_mapping_load_obsolete_spacemap(vim
, counts
,
6033 vd
->vdev_obsolete_sm
);
6035 if (scip
->scip_vdev
== vd
->vdev_id
&&
6036 scip
->scip_prev_obsolete_sm_object
!= 0) {
6037 space_map_t
*prev_obsolete_sm
= NULL
;
6038 VERIFY0(space_map_open(&prev_obsolete_sm
, spa
->spa_meta_objset
,
6039 scip
->scip_prev_obsolete_sm_object
, 0, vd
->vdev_asize
, 0));
6040 vdev_indirect_mapping_load_obsolete_spacemap(vim
, counts
,
6042 space_map_close(prev_obsolete_sm
);
6048 zdb_ddt_leak_init(spa_t
*spa
, zdb_cb_t
*zcb
)
6050 ddt_bookmark_t ddb
= {0};
6055 ASSERT(!dump_opt
['L']);
6057 while ((error
= ddt_walk(spa
, &ddb
, &dde
)) == 0) {
6059 ddt_phys_t
*ddp
= dde
.dde_phys
;
6061 if (ddb
.ddb_class
== DDT_CLASS_UNIQUE
)
6064 ASSERT(ddt_phys_total_refcnt(&dde
) > 1);
6066 for (p
= 0; p
< DDT_PHYS_TYPES
; p
++, ddp
++) {
6067 if (ddp
->ddp_phys_birth
== 0)
6069 ddt_bp_create(ddb
.ddb_checksum
,
6070 &dde
.dde_key
, ddp
, &blk
);
6071 if (p
== DDT_PHYS_DITTO
) {
6072 zdb_count_block(zcb
, NULL
, &blk
, ZDB_OT_DITTO
);
6074 zcb
->zcb_dedup_asize
+=
6075 BP_GET_ASIZE(&blk
) * (ddp
->ddp_refcnt
- 1);
6076 zcb
->zcb_dedup_blocks
++;
6079 ddt_t
*ddt
= spa
->spa_ddt
[ddb
.ddb_checksum
];
6081 VERIFY(ddt_lookup(ddt
, &blk
, B_TRUE
) != NULL
);
6085 ASSERT(error
== ENOENT
);
6088 typedef struct checkpoint_sm_exclude_entry_arg
{
6090 uint64_t cseea_checkpoint_size
;
6091 } checkpoint_sm_exclude_entry_arg_t
;
6094 checkpoint_sm_exclude_entry_cb(space_map_entry_t
*sme
, void *arg
)
6096 checkpoint_sm_exclude_entry_arg_t
*cseea
= arg
;
6097 vdev_t
*vd
= cseea
->cseea_vd
;
6098 metaslab_t
*ms
= vd
->vdev_ms
[sme
->sme_offset
>> vd
->vdev_ms_shift
];
6099 uint64_t end
= sme
->sme_offset
+ sme
->sme_run
;
6101 ASSERT(sme
->sme_type
== SM_FREE
);
6104 * Since the vdev_checkpoint_sm exists in the vdev level
6105 * and the ms_sm space maps exist in the metaslab level,
6106 * an entry in the checkpoint space map could theoretically
6107 * cross the boundaries of the metaslab that it belongs.
6109 * In reality, because of the way that we populate and
6110 * manipulate the checkpoint's space maps currently,
6111 * there shouldn't be any entries that cross metaslabs.
6112 * Hence the assertion below.
6114 * That said, there is no fundamental requirement that
6115 * the checkpoint's space map entries should not cross
6116 * metaslab boundaries. So if needed we could add code
6117 * that handles metaslab-crossing segments in the future.
6119 VERIFY3U(sme
->sme_offset
, >=, ms
->ms_start
);
6120 VERIFY3U(end
, <=, ms
->ms_start
+ ms
->ms_size
);
6123 * By removing the entry from the allocated segments we
6124 * also verify that the entry is there to begin with.
6126 mutex_enter(&ms
->ms_lock
);
6127 range_tree_remove(ms
->ms_allocatable
, sme
->sme_offset
, sme
->sme_run
);
6128 mutex_exit(&ms
->ms_lock
);
6130 cseea
->cseea_checkpoint_size
+= sme
->sme_run
;
6135 zdb_leak_init_vdev_exclude_checkpoint(vdev_t
*vd
, zdb_cb_t
*zcb
)
6137 spa_t
*spa
= vd
->vdev_spa
;
6138 space_map_t
*checkpoint_sm
= NULL
;
6139 uint64_t checkpoint_sm_obj
;
6142 * If there is no vdev_top_zap, we are in a pool whose
6143 * version predates the pool checkpoint feature.
6145 if (vd
->vdev_top_zap
== 0)
6149 * If there is no reference of the vdev_checkpoint_sm in
6150 * the vdev_top_zap, then one of the following scenarios
6153 * 1] There is no checkpoint
6154 * 2] There is a checkpoint, but no checkpointed blocks
6155 * have been freed yet
6156 * 3] The current vdev is indirect
6158 * In these cases we return immediately.
6160 if (zap_contains(spa_meta_objset(spa
), vd
->vdev_top_zap
,
6161 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) != 0)
6164 VERIFY0(zap_lookup(spa_meta_objset(spa
), vd
->vdev_top_zap
,
6165 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1,
6166 &checkpoint_sm_obj
));
6168 checkpoint_sm_exclude_entry_arg_t cseea
;
6169 cseea
.cseea_vd
= vd
;
6170 cseea
.cseea_checkpoint_size
= 0;
6172 VERIFY0(space_map_open(&checkpoint_sm
, spa_meta_objset(spa
),
6173 checkpoint_sm_obj
, 0, vd
->vdev_asize
, vd
->vdev_ashift
));
6175 VERIFY0(space_map_iterate(checkpoint_sm
,
6176 space_map_length(checkpoint_sm
),
6177 checkpoint_sm_exclude_entry_cb
, &cseea
));
6178 space_map_close(checkpoint_sm
);
6180 zcb
->zcb_checkpoint_size
+= cseea
.cseea_checkpoint_size
;
6184 zdb_leak_init_exclude_checkpoint(spa_t
*spa
, zdb_cb_t
*zcb
)
6186 ASSERT(!dump_opt
['L']);
6188 vdev_t
*rvd
= spa
->spa_root_vdev
;
6189 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
6190 ASSERT3U(c
, ==, rvd
->vdev_child
[c
]->vdev_id
);
6191 zdb_leak_init_vdev_exclude_checkpoint(rvd
->vdev_child
[c
], zcb
);
6196 count_unflushed_space_cb(spa_t
*spa
, space_map_entry_t
*sme
,
6197 uint64_t txg
, void *arg
)
6199 int64_t *ualloc_space
= arg
;
6201 uint64_t offset
= sme
->sme_offset
;
6202 uint64_t vdev_id
= sme
->sme_vdev
;
6204 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
6205 if (!vdev_is_concrete(vd
))
6208 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
6209 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
6211 if (txg
< metaslab_unflushed_txg(ms
))
6214 if (sme
->sme_type
== SM_ALLOC
)
6215 *ualloc_space
+= sme
->sme_run
;
6217 *ualloc_space
-= sme
->sme_run
;
6223 get_unflushed_alloc_space(spa_t
*spa
)
6228 int64_t ualloc_space
= 0;
6229 iterate_through_spacemap_logs(spa
, count_unflushed_space_cb
,
6231 return (ualloc_space
);
6235 load_unflushed_cb(spa_t
*spa
, space_map_entry_t
*sme
, uint64_t txg
, void *arg
)
6237 maptype_t
*uic_maptype
= arg
;
6239 uint64_t offset
= sme
->sme_offset
;
6240 uint64_t size
= sme
->sme_run
;
6241 uint64_t vdev_id
= sme
->sme_vdev
;
6243 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
6245 /* skip indirect vdevs */
6246 if (!vdev_is_concrete(vd
))
6249 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
6251 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
6252 ASSERT(*uic_maptype
== SM_ALLOC
|| *uic_maptype
== SM_FREE
);
6254 if (txg
< metaslab_unflushed_txg(ms
))
6257 if (*uic_maptype
== sme
->sme_type
)
6258 range_tree_add(ms
->ms_allocatable
, offset
, size
);
6260 range_tree_remove(ms
->ms_allocatable
, offset
, size
);
6266 load_unflushed_to_ms_allocatables(spa_t
*spa
, maptype_t maptype
)
6268 iterate_through_spacemap_logs(spa
, load_unflushed_cb
, &maptype
);
6272 load_concrete_ms_allocatable_trees(spa_t
*spa
, maptype_t maptype
)
6274 vdev_t
*rvd
= spa
->spa_root_vdev
;
6275 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
6276 vdev_t
*vd
= rvd
->vdev_child
[i
];
6278 ASSERT3U(i
, ==, vd
->vdev_id
);
6280 if (vd
->vdev_ops
== &vdev_indirect_ops
)
6283 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
6284 metaslab_t
*msp
= vd
->vdev_ms
[m
];
6286 (void) fprintf(stderr
,
6287 "\rloading concrete vdev %llu, "
6288 "metaslab %llu of %llu ...",
6289 (longlong_t
)vd
->vdev_id
,
6290 (longlong_t
)msp
->ms_id
,
6291 (longlong_t
)vd
->vdev_ms_count
);
6293 mutex_enter(&msp
->ms_lock
);
6294 range_tree_vacate(msp
->ms_allocatable
, NULL
, NULL
);
6297 * We don't want to spend the CPU manipulating the
6298 * size-ordered tree, so clear the range_tree ops.
6300 msp
->ms_allocatable
->rt_ops
= NULL
;
6302 if (msp
->ms_sm
!= NULL
) {
6303 VERIFY0(space_map_load(msp
->ms_sm
,
6304 msp
->ms_allocatable
, maptype
));
6306 if (!msp
->ms_loaded
)
6307 msp
->ms_loaded
= B_TRUE
;
6308 mutex_exit(&msp
->ms_lock
);
6312 load_unflushed_to_ms_allocatables(spa
, maptype
);
6316 * vm_idxp is an in-out parameter which (for indirect vdevs) is the
6317 * index in vim_entries that has the first entry in this metaslab.
6318 * On return, it will be set to the first entry after this metaslab.
6321 load_indirect_ms_allocatable_tree(vdev_t
*vd
, metaslab_t
*msp
,
6324 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6326 mutex_enter(&msp
->ms_lock
);
6327 range_tree_vacate(msp
->ms_allocatable
, NULL
, NULL
);
6330 * We don't want to spend the CPU manipulating the
6331 * size-ordered tree, so clear the range_tree ops.
6333 msp
->ms_allocatable
->rt_ops
= NULL
;
6335 for (; *vim_idxp
< vdev_indirect_mapping_num_entries(vim
);
6337 vdev_indirect_mapping_entry_phys_t
*vimep
=
6338 &vim
->vim_entries
[*vim_idxp
];
6339 uint64_t ent_offset
= DVA_MAPPING_GET_SRC_OFFSET(vimep
);
6340 uint64_t ent_len
= DVA_GET_ASIZE(&vimep
->vimep_dst
);
6341 ASSERT3U(ent_offset
, >=, msp
->ms_start
);
6342 if (ent_offset
>= msp
->ms_start
+ msp
->ms_size
)
6346 * Mappings do not cross metaslab boundaries,
6347 * because we create them by walking the metaslabs.
6349 ASSERT3U(ent_offset
+ ent_len
, <=,
6350 msp
->ms_start
+ msp
->ms_size
);
6351 range_tree_add(msp
->ms_allocatable
, ent_offset
, ent_len
);
6354 if (!msp
->ms_loaded
)
6355 msp
->ms_loaded
= B_TRUE
;
6356 mutex_exit(&msp
->ms_lock
);
6360 zdb_leak_init_prepare_indirect_vdevs(spa_t
*spa
, zdb_cb_t
*zcb
)
6362 ASSERT(!dump_opt
['L']);
6364 vdev_t
*rvd
= spa
->spa_root_vdev
;
6365 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
6366 vdev_t
*vd
= rvd
->vdev_child
[c
];
6368 ASSERT3U(c
, ==, vd
->vdev_id
);
6370 if (vd
->vdev_ops
!= &vdev_indirect_ops
)
6374 * Note: we don't check for mapping leaks on
6375 * removing vdevs because their ms_allocatable's
6376 * are used to look for leaks in allocated space.
6378 zcb
->zcb_vd_obsolete_counts
[c
] = zdb_load_obsolete_counts(vd
);
6381 * Normally, indirect vdevs don't have any
6382 * metaslabs. We want to set them up for
6385 vdev_metaslab_group_create(vd
);
6386 VERIFY0(vdev_metaslab_init(vd
, 0));
6388 vdev_indirect_mapping_t
*vim __maybe_unused
=
6389 vd
->vdev_indirect_mapping
;
6390 uint64_t vim_idx
= 0;
6391 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
6393 (void) fprintf(stderr
,
6394 "\rloading indirect vdev %llu, "
6395 "metaslab %llu of %llu ...",
6396 (longlong_t
)vd
->vdev_id
,
6397 (longlong_t
)vd
->vdev_ms
[m
]->ms_id
,
6398 (longlong_t
)vd
->vdev_ms_count
);
6400 load_indirect_ms_allocatable_tree(vd
, vd
->vdev_ms
[m
],
6403 ASSERT3U(vim_idx
, ==, vdev_indirect_mapping_num_entries(vim
));
6408 zdb_leak_init(spa_t
*spa
, zdb_cb_t
*zcb
)
6415 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6416 vdev_t
*rvd
= spa
->spa_root_vdev
;
6419 * We are going to be changing the meaning of the metaslab's
6420 * ms_allocatable. Ensure that the allocator doesn't try to
6423 spa
->spa_normal_class
->mc_ops
= &zdb_metaslab_ops
;
6424 spa
->spa_log_class
->mc_ops
= &zdb_metaslab_ops
;
6425 spa
->spa_embedded_log_class
->mc_ops
= &zdb_metaslab_ops
;
6427 zcb
->zcb_vd_obsolete_counts
=
6428 umem_zalloc(rvd
->vdev_children
* sizeof (uint32_t *),
6432 * For leak detection, we overload the ms_allocatable trees
6433 * to contain allocated segments instead of free segments.
6434 * As a result, we can't use the normal metaslab_load/unload
6437 zdb_leak_init_prepare_indirect_vdevs(spa
, zcb
);
6438 load_concrete_ms_allocatable_trees(spa
, SM_ALLOC
);
6441 * On load_concrete_ms_allocatable_trees() we loaded all the
6442 * allocated entries from the ms_sm to the ms_allocatable for
6443 * each metaslab. If the pool has a checkpoint or is in the
6444 * middle of discarding a checkpoint, some of these blocks
6445 * may have been freed but their ms_sm may not have been
6446 * updated because they are referenced by the checkpoint. In
6447 * order to avoid false-positives during leak-detection, we
6448 * go through the vdev's checkpoint space map and exclude all
6449 * its entries from their relevant ms_allocatable.
6451 * We also aggregate the space held by the checkpoint and add
6452 * it to zcb_checkpoint_size.
6454 * Note that at this point we are also verifying that all the
6455 * entries on the checkpoint_sm are marked as allocated in
6456 * the ms_sm of their relevant metaslab.
6457 * [see comment in checkpoint_sm_exclude_entry_cb()]
6459 zdb_leak_init_exclude_checkpoint(spa
, zcb
);
6460 ASSERT3U(zcb
->zcb_checkpoint_size
, ==, spa_get_checkpoint_space(spa
));
6462 /* for cleaner progress output */
6463 (void) fprintf(stderr
, "\n");
6465 if (bpobj_is_open(&dp
->dp_obsolete_bpobj
)) {
6466 ASSERT(spa_feature_is_enabled(spa
,
6467 SPA_FEATURE_DEVICE_REMOVAL
));
6468 (void) bpobj_iterate_nofree(&dp
->dp_obsolete_bpobj
,
6469 increment_indirect_mapping_cb
, zcb
, NULL
);
6472 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6473 zdb_ddt_leak_init(spa
, zcb
);
6474 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6478 zdb_check_for_obsolete_leaks(vdev_t
*vd
, zdb_cb_t
*zcb
)
6480 boolean_t leaks
= B_FALSE
;
6481 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6482 uint64_t total_leaked
= 0;
6483 boolean_t are_precise
= B_FALSE
;
6485 ASSERT(vim
!= NULL
);
6487 for (uint64_t i
= 0; i
< vdev_indirect_mapping_num_entries(vim
); i
++) {
6488 vdev_indirect_mapping_entry_phys_t
*vimep
=
6489 &vim
->vim_entries
[i
];
6490 uint64_t obsolete_bytes
= 0;
6491 uint64_t offset
= DVA_MAPPING_GET_SRC_OFFSET(vimep
);
6492 metaslab_t
*msp
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
6495 * This is not very efficient but it's easy to
6496 * verify correctness.
6498 for (uint64_t inner_offset
= 0;
6499 inner_offset
< DVA_GET_ASIZE(&vimep
->vimep_dst
);
6500 inner_offset
+= 1ULL << vd
->vdev_ashift
) {
6501 if (range_tree_contains(msp
->ms_allocatable
,
6502 offset
+ inner_offset
, 1ULL << vd
->vdev_ashift
)) {
6503 obsolete_bytes
+= 1ULL << vd
->vdev_ashift
;
6507 int64_t bytes_leaked
= obsolete_bytes
-
6508 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
][i
];
6509 ASSERT3U(DVA_GET_ASIZE(&vimep
->vimep_dst
), >=,
6510 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
][i
]);
6512 VERIFY0(vdev_obsolete_counts_are_precise(vd
, &are_precise
));
6513 if (bytes_leaked
!= 0 && (are_precise
|| dump_opt
['d'] >= 5)) {
6514 (void) printf("obsolete indirect mapping count "
6515 "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
6516 (u_longlong_t
)vd
->vdev_id
,
6517 (u_longlong_t
)DVA_MAPPING_GET_SRC_OFFSET(vimep
),
6518 (u_longlong_t
)DVA_GET_ASIZE(&vimep
->vimep_dst
),
6519 (u_longlong_t
)bytes_leaked
);
6521 total_leaked
+= ABS(bytes_leaked
);
6524 VERIFY0(vdev_obsolete_counts_are_precise(vd
, &are_precise
));
6525 if (!are_precise
&& total_leaked
> 0) {
6526 int pct_leaked
= total_leaked
* 100 /
6527 vdev_indirect_mapping_bytes_mapped(vim
);
6528 (void) printf("cannot verify obsolete indirect mapping "
6529 "counts of vdev %llu because precise feature was not "
6530 "enabled when it was removed: %d%% (%llx bytes) of mapping"
6532 (u_longlong_t
)vd
->vdev_id
, pct_leaked
,
6533 (u_longlong_t
)total_leaked
);
6534 } else if (total_leaked
> 0) {
6535 (void) printf("obsolete indirect mapping count mismatch "
6536 "for vdev %llu -- %llx total bytes mismatched\n",
6537 (u_longlong_t
)vd
->vdev_id
,
6538 (u_longlong_t
)total_leaked
);
6542 vdev_indirect_mapping_free_obsolete_counts(vim
,
6543 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
]);
6544 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
] = NULL
;
6550 zdb_leak_fini(spa_t
*spa
, zdb_cb_t
*zcb
)
6555 boolean_t leaks
= B_FALSE
;
6556 vdev_t
*rvd
= spa
->spa_root_vdev
;
6557 for (unsigned c
= 0; c
< rvd
->vdev_children
; c
++) {
6558 vdev_t
*vd
= rvd
->vdev_child
[c
];
6560 if (zcb
->zcb_vd_obsolete_counts
[c
] != NULL
) {
6561 leaks
|= zdb_check_for_obsolete_leaks(vd
, zcb
);
6564 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
6565 metaslab_t
*msp
= vd
->vdev_ms
[m
];
6566 ASSERT3P(msp
->ms_group
, ==, (msp
->ms_group
->mg_class
==
6567 spa_embedded_log_class(spa
)) ?
6568 vd
->vdev_log_mg
: vd
->vdev_mg
);
6571 * ms_allocatable has been overloaded
6572 * to contain allocated segments. Now that
6573 * we finished traversing all blocks, any
6574 * block that remains in the ms_allocatable
6575 * represents an allocated block that we
6576 * did not claim during the traversal.
6577 * Claimed blocks would have been removed
6578 * from the ms_allocatable. For indirect
6579 * vdevs, space remaining in the tree
6580 * represents parts of the mapping that are
6581 * not referenced, which is not a bug.
6583 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
6584 range_tree_vacate(msp
->ms_allocatable
,
6587 range_tree_vacate(msp
->ms_allocatable
,
6590 if (msp
->ms_loaded
) {
6591 msp
->ms_loaded
= B_FALSE
;
6596 umem_free(zcb
->zcb_vd_obsolete_counts
,
6597 rvd
->vdev_children
* sizeof (uint32_t *));
6598 zcb
->zcb_vd_obsolete_counts
= NULL
;
6604 count_block_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6607 zdb_cb_t
*zcb
= arg
;
6609 if (dump_opt
['b'] >= 5) {
6610 char blkbuf
[BP_SPRINTF_LEN
];
6611 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
6612 (void) printf("[%s] %s\n",
6613 "deferred free", blkbuf
);
6615 zdb_count_block(zcb
, NULL
, bp
, ZDB_OT_DEFERRED
);
6620 * Iterate over livelists which have been destroyed by the user but
6621 * are still present in the MOS, waiting to be freed
6624 iterate_deleted_livelists(spa_t
*spa
, ll_iter_t func
, void *arg
)
6626 objset_t
*mos
= spa
->spa_meta_objset
;
6628 int err
= zap_lookup(mos
, DMU_POOL_DIRECTORY_OBJECT
,
6629 DMU_POOL_DELETED_CLONES
, sizeof (uint64_t), 1, &zap_obj
);
6635 zap_attribute_t attr
;
6637 /* NULL out os prior to dsl_deadlist_open in case it's garbage */
6639 for (zap_cursor_init(&zc
, mos
, zap_obj
);
6640 zap_cursor_retrieve(&zc
, &attr
) == 0;
6641 (void) zap_cursor_advance(&zc
)) {
6642 dsl_deadlist_open(&ll
, mos
, attr
.za_first_integer
);
6644 dsl_deadlist_close(&ll
);
6646 zap_cursor_fini(&zc
);
6650 bpobj_count_block_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
6654 return (count_block_cb(arg
, bp
, tx
));
6658 livelist_entry_count_blocks_cb(void *args
, dsl_deadlist_entry_t
*dle
)
6660 zdb_cb_t
*zbc
= args
;
6662 bplist_create(&blks
);
6663 /* determine which blocks have been alloc'd but not freed */
6664 VERIFY0(dsl_process_sub_livelist(&dle
->dle_bpobj
, &blks
, NULL
, NULL
));
6665 /* count those blocks */
6666 (void) bplist_iterate(&blks
, count_block_cb
, zbc
, NULL
);
6667 bplist_destroy(&blks
);
6672 livelist_count_blocks(dsl_deadlist_t
*ll
, void *arg
)
6674 dsl_deadlist_iterate(ll
, livelist_entry_count_blocks_cb
, arg
);
6678 * Count the blocks in the livelists that have been destroyed by the user
6679 * but haven't yet been freed.
6682 deleted_livelists_count_blocks(spa_t
*spa
, zdb_cb_t
*zbc
)
6684 iterate_deleted_livelists(spa
, livelist_count_blocks
, zbc
);
6688 dump_livelist_cb(dsl_deadlist_t
*ll
, void *arg
)
6690 ASSERT3P(arg
, ==, NULL
);
6691 global_feature_count
[SPA_FEATURE_LIVELIST
]++;
6692 dump_blkptr_list(ll
, "Deleted Livelist");
6693 dsl_deadlist_iterate(ll
, sublivelist_verify_lightweight
, NULL
);
6697 * Print out, register object references to, and increment feature counts for
6698 * livelists that have been destroyed by the user but haven't yet been freed.
6701 deleted_livelists_dump_mos(spa_t
*spa
)
6704 objset_t
*mos
= spa
->spa_meta_objset
;
6705 int err
= zap_lookup(mos
, DMU_POOL_DIRECTORY_OBJECT
,
6706 DMU_POOL_DELETED_CLONES
, sizeof (uint64_t), 1, &zap_obj
);
6709 mos_obj_refd(zap_obj
);
6710 iterate_deleted_livelists(spa
, dump_livelist_cb
, NULL
);
6714 zdb_brt_entry_compare(const void *zcn1
, const void *zcn2
)
6716 const dva_t
*dva1
= &((const zdb_brt_entry_t
*)zcn1
)->zbre_dva
;
6717 const dva_t
*dva2
= &((const zdb_brt_entry_t
*)zcn2
)->zbre_dva
;
6720 cmp
= TREE_CMP(DVA_GET_VDEV(dva1
), DVA_GET_VDEV(dva2
));
6722 cmp
= TREE_CMP(DVA_GET_OFFSET(dva1
), DVA_GET_OFFSET(dva2
));
6728 dump_block_stats(spa_t
*spa
)
6731 zdb_blkstats_t
*zb
, *tzb
;
6732 uint64_t norm_alloc
, norm_space
, total_alloc
, total_found
;
6733 int flags
= TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
6734 TRAVERSE_NO_DECRYPT
| TRAVERSE_HARD
;
6735 boolean_t leaks
= B_FALSE
;
6737 bp_embedded_type_t i
;
6739 zcb
= umem_zalloc(sizeof (zdb_cb_t
), UMEM_NOFAIL
);
6741 if (spa_feature_is_active(spa
, SPA_FEATURE_BLOCK_CLONING
)) {
6742 avl_create(&zcb
->zcb_brt
, zdb_brt_entry_compare
,
6743 sizeof (zdb_brt_entry_t
),
6744 offsetof(zdb_brt_entry_t
, zbre_node
));
6745 zcb
->zcb_brt_is_active
= B_TRUE
;
6748 (void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
6749 (dump_opt
['c'] || !dump_opt
['L']) ? "to verify " : "",
6750 (dump_opt
['c'] == 1) ? "metadata " : "",
6751 dump_opt
['c'] ? "checksums " : "",
6752 (dump_opt
['c'] && !dump_opt
['L']) ? "and verify " : "",
6753 !dump_opt
['L'] ? "nothing leaked " : "");
6756 * When leak detection is enabled we load all space maps as SM_ALLOC
6757 * maps, then traverse the pool claiming each block we discover. If
6758 * the pool is perfectly consistent, the segment trees will be empty
6759 * when we're done. Anything left over is a leak; any block we can't
6760 * claim (because it's not part of any space map) is a double
6761 * allocation, reference to a freed block, or an unclaimed log block.
6763 * When leak detection is disabled (-L option) we still traverse the
6764 * pool claiming each block we discover, but we skip opening any space
6767 zdb_leak_init(spa
, zcb
);
6770 * If there's a deferred-free bplist, process that first.
6772 (void) bpobj_iterate_nofree(&spa
->spa_deferred_bpobj
,
6773 bpobj_count_block_cb
, zcb
, NULL
);
6775 if (spa_version(spa
) >= SPA_VERSION_DEADLISTS
) {
6776 (void) bpobj_iterate_nofree(&spa
->spa_dsl_pool
->dp_free_bpobj
,
6777 bpobj_count_block_cb
, zcb
, NULL
);
6780 zdb_claim_removing(spa
, zcb
);
6782 if (spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
)) {
6783 VERIFY3U(0, ==, bptree_iterate(spa
->spa_meta_objset
,
6784 spa
->spa_dsl_pool
->dp_bptree_obj
, B_FALSE
, count_block_cb
,
6788 deleted_livelists_count_blocks(spa
, zcb
);
6790 if (dump_opt
['c'] > 1)
6791 flags
|= TRAVERSE_PREFETCH_DATA
;
6793 zcb
->zcb_totalasize
= metaslab_class_get_alloc(spa_normal_class(spa
));
6794 zcb
->zcb_totalasize
+= metaslab_class_get_alloc(spa_special_class(spa
));
6795 zcb
->zcb_totalasize
+= metaslab_class_get_alloc(spa_dedup_class(spa
));
6796 zcb
->zcb_totalasize
+=
6797 metaslab_class_get_alloc(spa_embedded_log_class(spa
));
6798 zcb
->zcb_start
= zcb
->zcb_lastprint
= gethrtime();
6799 err
= traverse_pool(spa
, 0, flags
, zdb_blkptr_cb
, zcb
);
6802 * If we've traversed the data blocks then we need to wait for those
6803 * I/Os to complete. We leverage "The Godfather" zio to wait on
6804 * all async I/Os to complete.
6806 if (dump_opt
['c']) {
6807 for (c
= 0; c
< max_ncpus
; c
++) {
6808 (void) zio_wait(spa
->spa_async_zio_root
[c
]);
6809 spa
->spa_async_zio_root
[c
] = zio_root(spa
, NULL
, NULL
,
6810 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
6811 ZIO_FLAG_GODFATHER
);
6814 ASSERT0(spa
->spa_load_verify_bytes
);
6817 * Done after zio_wait() since zcb_haderrors is modified in
6820 zcb
->zcb_haderrors
|= err
;
6822 if (zcb
->zcb_haderrors
) {
6823 (void) printf("\nError counts:\n\n");
6824 (void) printf("\t%5s %s\n", "errno", "count");
6825 for (e
= 0; e
< 256; e
++) {
6826 if (zcb
->zcb_errors
[e
] != 0) {
6827 (void) printf("\t%5d %llu\n",
6828 e
, (u_longlong_t
)zcb
->zcb_errors
[e
]);
6834 * Report any leaked segments.
6836 leaks
|= zdb_leak_fini(spa
, zcb
);
6838 tzb
= &zcb
->zcb_type
[ZB_TOTAL
][ZDB_OT_TOTAL
];
6840 norm_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
6841 norm_space
= metaslab_class_get_space(spa_normal_class(spa
));
6843 total_alloc
= norm_alloc
+
6844 metaslab_class_get_alloc(spa_log_class(spa
)) +
6845 metaslab_class_get_alloc(spa_embedded_log_class(spa
)) +
6846 metaslab_class_get_alloc(spa_special_class(spa
)) +
6847 metaslab_class_get_alloc(spa_dedup_class(spa
)) +
6848 get_unflushed_alloc_space(spa
);
6850 tzb
->zb_asize
- zcb
->zcb_dedup_asize
- zcb
->zcb_clone_asize
+
6851 zcb
->zcb_removing_size
+ zcb
->zcb_checkpoint_size
;
6853 if (total_found
== total_alloc
&& !dump_opt
['L']) {
6854 (void) printf("\n\tNo leaks (block sum matches space"
6855 " maps exactly)\n");
6856 } else if (!dump_opt
['L']) {
6857 (void) printf("block traversal size %llu != alloc %llu "
6859 (u_longlong_t
)total_found
,
6860 (u_longlong_t
)total_alloc
,
6861 (dump_opt
['L']) ? "unreachable" : "leaked",
6862 (longlong_t
)(total_alloc
- total_found
));
6866 if (tzb
->zb_count
== 0) {
6867 umem_free(zcb
, sizeof (zdb_cb_t
));
6871 (void) printf("\n");
6872 (void) printf("\t%-16s %14llu\n", "bp count:",
6873 (u_longlong_t
)tzb
->zb_count
);
6874 (void) printf("\t%-16s %14llu\n", "ganged count:",
6875 (longlong_t
)tzb
->zb_gangs
);
6876 (void) printf("\t%-16s %14llu avg: %6llu\n", "bp logical:",
6877 (u_longlong_t
)tzb
->zb_lsize
,
6878 (u_longlong_t
)(tzb
->zb_lsize
/ tzb
->zb_count
));
6879 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
6880 "bp physical:", (u_longlong_t
)tzb
->zb_psize
,
6881 (u_longlong_t
)(tzb
->zb_psize
/ tzb
->zb_count
),
6882 (double)tzb
->zb_lsize
/ tzb
->zb_psize
);
6883 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
6884 "bp allocated:", (u_longlong_t
)tzb
->zb_asize
,
6885 (u_longlong_t
)(tzb
->zb_asize
/ tzb
->zb_count
),
6886 (double)tzb
->zb_lsize
/ tzb
->zb_asize
);
6887 (void) printf("\t%-16s %14llu ref>1: %6llu deduplication: %6.2f\n",
6888 "bp deduped:", (u_longlong_t
)zcb
->zcb_dedup_asize
,
6889 (u_longlong_t
)zcb
->zcb_dedup_blocks
,
6890 (double)zcb
->zcb_dedup_asize
/ tzb
->zb_asize
+ 1.0);
6891 (void) printf("\t%-16s %14llu count: %6llu\n",
6892 "bp cloned:", (u_longlong_t
)zcb
->zcb_clone_asize
,
6893 (u_longlong_t
)zcb
->zcb_clone_blocks
);
6894 (void) printf("\t%-16s %14llu used: %5.2f%%\n", "Normal class:",
6895 (u_longlong_t
)norm_alloc
, 100.0 * norm_alloc
/ norm_space
);
6897 if (spa_special_class(spa
)->mc_allocator
[0].mca_rotor
!= NULL
) {
6898 uint64_t alloc
= metaslab_class_get_alloc(
6899 spa_special_class(spa
));
6900 uint64_t space
= metaslab_class_get_space(
6901 spa_special_class(spa
));
6903 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
6904 "Special class", (u_longlong_t
)alloc
,
6905 100.0 * alloc
/ space
);
6908 if (spa_dedup_class(spa
)->mc_allocator
[0].mca_rotor
!= NULL
) {
6909 uint64_t alloc
= metaslab_class_get_alloc(
6910 spa_dedup_class(spa
));
6911 uint64_t space
= metaslab_class_get_space(
6912 spa_dedup_class(spa
));
6914 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
6915 "Dedup class", (u_longlong_t
)alloc
,
6916 100.0 * alloc
/ space
);
6919 if (spa_embedded_log_class(spa
)->mc_allocator
[0].mca_rotor
!= NULL
) {
6920 uint64_t alloc
= metaslab_class_get_alloc(
6921 spa_embedded_log_class(spa
));
6922 uint64_t space
= metaslab_class_get_space(
6923 spa_embedded_log_class(spa
));
6925 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
6926 "Embedded log class", (u_longlong_t
)alloc
,
6927 100.0 * alloc
/ space
);
6930 for (i
= 0; i
< NUM_BP_EMBEDDED_TYPES
; i
++) {
6931 if (zcb
->zcb_embedded_blocks
[i
] == 0)
6933 (void) printf("\n");
6934 (void) printf("\tadditional, non-pointer bps of type %u: "
6936 i
, (u_longlong_t
)zcb
->zcb_embedded_blocks
[i
]);
6938 if (dump_opt
['b'] >= 3) {
6939 (void) printf("\t number of (compressed) bytes: "
6941 dump_histogram(zcb
->zcb_embedded_histogram
[i
],
6942 sizeof (zcb
->zcb_embedded_histogram
[i
]) /
6943 sizeof (zcb
->zcb_embedded_histogram
[i
][0]), 0);
6947 if (tzb
->zb_ditto_samevdev
!= 0) {
6948 (void) printf("\tDittoed blocks on same vdev: %llu\n",
6949 (longlong_t
)tzb
->zb_ditto_samevdev
);
6951 if (tzb
->zb_ditto_same_ms
!= 0) {
6952 (void) printf("\tDittoed blocks in same metaslab: %llu\n",
6953 (longlong_t
)tzb
->zb_ditto_same_ms
);
6956 for (uint64_t v
= 0; v
< spa
->spa_root_vdev
->vdev_children
; v
++) {
6957 vdev_t
*vd
= spa
->spa_root_vdev
->vdev_child
[v
];
6958 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6965 zdb_nicenum(vdev_indirect_mapping_num_entries(vim
),
6966 mem
, vdev_indirect_mapping_size(vim
));
6968 (void) printf("\tindirect vdev id %llu has %llu segments "
6970 (longlong_t
)vd
->vdev_id
,
6971 (longlong_t
)vdev_indirect_mapping_num_entries(vim
), mem
);
6974 if (dump_opt
['b'] >= 2) {
6976 char csize
[32], lsize
[32], psize
[32], asize
[32];
6977 char avg
[32], gang
[32];
6978 (void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
6979 "\t avg\t comp\t%%Total\tType\n");
6981 zfs_blkstat_t
*mdstats
= umem_zalloc(sizeof (zfs_blkstat_t
),
6984 for (t
= 0; t
<= ZDB_OT_TOTAL
; t
++) {
6985 const char *typename
;
6987 /* make sure nicenum has enough space */
6988 _Static_assert(sizeof (csize
) >= NN_NUMBUF_SZ
,
6990 _Static_assert(sizeof (lsize
) >= NN_NUMBUF_SZ
,
6992 _Static_assert(sizeof (psize
) >= NN_NUMBUF_SZ
,
6994 _Static_assert(sizeof (asize
) >= NN_NUMBUF_SZ
,
6996 _Static_assert(sizeof (avg
) >= NN_NUMBUF_SZ
,
6998 _Static_assert(sizeof (gang
) >= NN_NUMBUF_SZ
,
7001 if (t
< DMU_OT_NUMTYPES
)
7002 typename
= dmu_ot
[t
].ot_name
;
7004 typename
= zdb_ot_extname
[t
- DMU_OT_NUMTYPES
];
7006 if (zcb
->zcb_type
[ZB_TOTAL
][t
].zb_asize
== 0) {
7007 (void) printf("%6s\t%5s\t%5s\t%5s"
7008 "\t%5s\t%5s\t%6s\t%s\n",
7020 for (l
= ZB_TOTAL
- 1; l
>= -1; l
--) {
7021 level
= (l
== -1 ? ZB_TOTAL
: l
);
7022 zb
= &zcb
->zcb_type
[level
][t
];
7024 if (zb
->zb_asize
== 0)
7027 if (level
!= ZB_TOTAL
&& t
< DMU_OT_NUMTYPES
&&
7028 (level
> 0 || DMU_OT_IS_METADATA(t
))) {
7029 mdstats
->zb_count
+= zb
->zb_count
;
7030 mdstats
->zb_lsize
+= zb
->zb_lsize
;
7031 mdstats
->zb_psize
+= zb
->zb_psize
;
7032 mdstats
->zb_asize
+= zb
->zb_asize
;
7033 mdstats
->zb_gangs
+= zb
->zb_gangs
;
7036 if (dump_opt
['b'] < 3 && level
!= ZB_TOTAL
)
7039 if (level
== 0 && zb
->zb_asize
==
7040 zcb
->zcb_type
[ZB_TOTAL
][t
].zb_asize
)
7043 zdb_nicenum(zb
->zb_count
, csize
,
7045 zdb_nicenum(zb
->zb_lsize
, lsize
,
7047 zdb_nicenum(zb
->zb_psize
, psize
,
7049 zdb_nicenum(zb
->zb_asize
, asize
,
7051 zdb_nicenum(zb
->zb_asize
/ zb
->zb_count
, avg
,
7053 zdb_nicenum(zb
->zb_gangs
, gang
, sizeof (gang
));
7055 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7057 csize
, lsize
, psize
, asize
, avg
,
7058 (double)zb
->zb_lsize
/ zb
->zb_psize
,
7059 100.0 * zb
->zb_asize
/ tzb
->zb_asize
);
7061 if (level
== ZB_TOTAL
)
7062 (void) printf("%s\n", typename
);
7064 (void) printf(" L%d %s\n",
7067 if (dump_opt
['b'] >= 3 && zb
->zb_gangs
> 0) {
7068 (void) printf("\t number of ganged "
7069 "blocks: %s\n", gang
);
7072 if (dump_opt
['b'] >= 4) {
7073 (void) printf("psize "
7074 "(in 512-byte sectors): "
7075 "number of blocks\n");
7076 dump_histogram(zb
->zb_psize_histogram
,
7077 PSIZE_HISTO_SIZE
, 0);
7081 zdb_nicenum(mdstats
->zb_count
, csize
,
7083 zdb_nicenum(mdstats
->zb_lsize
, lsize
,
7085 zdb_nicenum(mdstats
->zb_psize
, psize
,
7087 zdb_nicenum(mdstats
->zb_asize
, asize
,
7089 zdb_nicenum(mdstats
->zb_asize
/ mdstats
->zb_count
, avg
,
7091 zdb_nicenum(mdstats
->zb_gangs
, gang
, sizeof (gang
));
7093 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7095 csize
, lsize
, psize
, asize
, avg
,
7096 (double)mdstats
->zb_lsize
/ mdstats
->zb_psize
,
7097 100.0 * mdstats
->zb_asize
/ tzb
->zb_asize
);
7098 (void) printf("%s\n", "Metadata Total");
7100 /* Output a table summarizing block sizes in the pool */
7101 if (dump_opt
['b'] >= 2) {
7102 dump_size_histograms(zcb
);
7105 umem_free(mdstats
, sizeof (zfs_blkstat_t
));
7108 (void) printf("\n");
7111 umem_free(zcb
, sizeof (zdb_cb_t
));
7115 if (zcb
->zcb_haderrors
) {
7116 umem_free(zcb
, sizeof (zdb_cb_t
));
7120 umem_free(zcb
, sizeof (zdb_cb_t
));
7124 typedef struct zdb_ddt_entry
{
7125 /* key must be first for ddt_key_compare */
7127 uint64_t zdde_ref_blocks
;
7128 uint64_t zdde_ref_lsize
;
7129 uint64_t zdde_ref_psize
;
7130 uint64_t zdde_ref_dsize
;
7131 avl_node_t zdde_node
;
7135 zdb_ddt_add_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
7136 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
7138 (void) zilog
, (void) dnp
;
7139 avl_tree_t
*t
= arg
;
7141 zdb_ddt_entry_t
*zdde
, zdde_search
;
7143 if (zb
->zb_level
== ZB_DNODE_LEVEL
|| BP_IS_HOLE(bp
) ||
7147 if (dump_opt
['S'] > 1 && zb
->zb_level
== ZB_ROOT_LEVEL
) {
7148 (void) printf("traversing objset %llu, %llu objects, "
7149 "%lu blocks so far\n",
7150 (u_longlong_t
)zb
->zb_objset
,
7151 (u_longlong_t
)BP_GET_FILL(bp
),
7155 if (BP_IS_HOLE(bp
) || BP_GET_CHECKSUM(bp
) == ZIO_CHECKSUM_OFF
||
7156 BP_GET_LEVEL(bp
) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)))
7159 ddt_key_fill(&zdde_search
.zdde_key
, bp
);
7161 zdde
= avl_find(t
, &zdde_search
, &where
);
7164 zdde
= umem_zalloc(sizeof (*zdde
), UMEM_NOFAIL
);
7165 zdde
->zdde_key
= zdde_search
.zdde_key
;
7166 avl_insert(t
, zdde
, where
);
7169 zdde
->zdde_ref_blocks
+= 1;
7170 zdde
->zdde_ref_lsize
+= BP_GET_LSIZE(bp
);
7171 zdde
->zdde_ref_psize
+= BP_GET_PSIZE(bp
);
7172 zdde
->zdde_ref_dsize
+= bp_get_dsize_sync(spa
, bp
);
7178 dump_simulated_ddt(spa_t
*spa
)
7181 void *cookie
= NULL
;
7182 zdb_ddt_entry_t
*zdde
;
7183 ddt_histogram_t ddh_total
= {{{0}}};
7184 ddt_stat_t dds_total
= {0};
7186 avl_create(&t
, ddt_key_compare
,
7187 sizeof (zdb_ddt_entry_t
), offsetof(zdb_ddt_entry_t
, zdde_node
));
7189 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
7191 (void) traverse_pool(spa
, 0, TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
7192 TRAVERSE_NO_DECRYPT
, zdb_ddt_add_cb
, &t
);
7194 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7196 while ((zdde
= avl_destroy_nodes(&t
, &cookie
)) != NULL
) {
7198 uint64_t refcnt
= zdde
->zdde_ref_blocks
;
7199 ASSERT(refcnt
!= 0);
7201 dds
.dds_blocks
= zdde
->zdde_ref_blocks
/ refcnt
;
7202 dds
.dds_lsize
= zdde
->zdde_ref_lsize
/ refcnt
;
7203 dds
.dds_psize
= zdde
->zdde_ref_psize
/ refcnt
;
7204 dds
.dds_dsize
= zdde
->zdde_ref_dsize
/ refcnt
;
7206 dds
.dds_ref_blocks
= zdde
->zdde_ref_blocks
;
7207 dds
.dds_ref_lsize
= zdde
->zdde_ref_lsize
;
7208 dds
.dds_ref_psize
= zdde
->zdde_ref_psize
;
7209 dds
.dds_ref_dsize
= zdde
->zdde_ref_dsize
;
7211 ddt_stat_add(&ddh_total
.ddh_stat
[highbit64(refcnt
) - 1],
7214 umem_free(zdde
, sizeof (*zdde
));
7219 ddt_histogram_stat(&dds_total
, &ddh_total
);
7221 (void) printf("Simulated DDT histogram:\n");
7223 zpool_dump_ddt(&dds_total
, &ddh_total
);
7225 dump_dedup_ratio(&dds_total
);
7229 verify_device_removal_feature_counts(spa_t
*spa
)
7231 uint64_t dr_feature_refcount
= 0;
7232 uint64_t oc_feature_refcount
= 0;
7233 uint64_t indirect_vdev_count
= 0;
7234 uint64_t precise_vdev_count
= 0;
7235 uint64_t obsolete_counts_object_count
= 0;
7236 uint64_t obsolete_sm_count
= 0;
7237 uint64_t obsolete_counts_count
= 0;
7238 uint64_t scip_count
= 0;
7239 uint64_t obsolete_bpobj_count
= 0;
7242 spa_condensing_indirect_phys_t
*scip
=
7243 &spa
->spa_condensing_indirect_phys
;
7244 if (scip
->scip_next_mapping_object
!= 0) {
7245 vdev_t
*vd
= spa
->spa_root_vdev
->vdev_child
[scip
->scip_vdev
];
7246 ASSERT(scip
->scip_prev_obsolete_sm_object
!= 0);
7247 ASSERT3P(vd
->vdev_ops
, ==, &vdev_indirect_ops
);
7249 (void) printf("Condensing indirect vdev %llu: new mapping "
7250 "object %llu, prev obsolete sm %llu\n",
7251 (u_longlong_t
)scip
->scip_vdev
,
7252 (u_longlong_t
)scip
->scip_next_mapping_object
,
7253 (u_longlong_t
)scip
->scip_prev_obsolete_sm_object
);
7254 if (scip
->scip_prev_obsolete_sm_object
!= 0) {
7255 space_map_t
*prev_obsolete_sm
= NULL
;
7256 VERIFY0(space_map_open(&prev_obsolete_sm
,
7257 spa
->spa_meta_objset
,
7258 scip
->scip_prev_obsolete_sm_object
,
7259 0, vd
->vdev_asize
, 0));
7260 dump_spacemap(spa
->spa_meta_objset
, prev_obsolete_sm
);
7261 (void) printf("\n");
7262 space_map_close(prev_obsolete_sm
);
7268 for (uint64_t i
= 0; i
< spa
->spa_root_vdev
->vdev_children
; i
++) {
7269 vdev_t
*vd
= spa
->spa_root_vdev
->vdev_child
[i
];
7270 vdev_indirect_config_t
*vic
= &vd
->vdev_indirect_config
;
7272 if (vic
->vic_mapping_object
!= 0) {
7273 ASSERT(vd
->vdev_ops
== &vdev_indirect_ops
||
7275 indirect_vdev_count
++;
7277 if (vd
->vdev_indirect_mapping
->vim_havecounts
) {
7278 obsolete_counts_count
++;
7282 boolean_t are_precise
;
7283 VERIFY0(vdev_obsolete_counts_are_precise(vd
, &are_precise
));
7285 ASSERT(vic
->vic_mapping_object
!= 0);
7286 precise_vdev_count
++;
7289 uint64_t obsolete_sm_object
;
7290 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
7291 if (obsolete_sm_object
!= 0) {
7292 ASSERT(vic
->vic_mapping_object
!= 0);
7293 obsolete_sm_count
++;
7297 (void) feature_get_refcount(spa
,
7298 &spa_feature_table
[SPA_FEATURE_DEVICE_REMOVAL
],
7299 &dr_feature_refcount
);
7300 (void) feature_get_refcount(spa
,
7301 &spa_feature_table
[SPA_FEATURE_OBSOLETE_COUNTS
],
7302 &oc_feature_refcount
);
7304 if (dr_feature_refcount
!= indirect_vdev_count
) {
7306 (void) printf("Number of indirect vdevs (%llu) " \
7307 "does not match feature count (%llu)\n",
7308 (u_longlong_t
)indirect_vdev_count
,
7309 (u_longlong_t
)dr_feature_refcount
);
7311 (void) printf("Verified device_removal feature refcount " \
7312 "of %llu is correct\n",
7313 (u_longlong_t
)dr_feature_refcount
);
7316 if (zap_contains(spa_meta_objset(spa
), DMU_POOL_DIRECTORY_OBJECT
,
7317 DMU_POOL_OBSOLETE_BPOBJ
) == 0) {
7318 obsolete_bpobj_count
++;
7322 obsolete_counts_object_count
= precise_vdev_count
;
7323 obsolete_counts_object_count
+= obsolete_sm_count
;
7324 obsolete_counts_object_count
+= obsolete_counts_count
;
7325 obsolete_counts_object_count
+= scip_count
;
7326 obsolete_counts_object_count
+= obsolete_bpobj_count
;
7327 obsolete_counts_object_count
+= remap_deadlist_count
;
7329 if (oc_feature_refcount
!= obsolete_counts_object_count
) {
7331 (void) printf("Number of obsolete counts objects (%llu) " \
7332 "does not match feature count (%llu)\n",
7333 (u_longlong_t
)obsolete_counts_object_count
,
7334 (u_longlong_t
)oc_feature_refcount
);
7335 (void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
7336 "ob:%llu rd:%llu\n",
7337 (u_longlong_t
)precise_vdev_count
,
7338 (u_longlong_t
)obsolete_sm_count
,
7339 (u_longlong_t
)obsolete_counts_count
,
7340 (u_longlong_t
)scip_count
,
7341 (u_longlong_t
)obsolete_bpobj_count
,
7342 (u_longlong_t
)remap_deadlist_count
);
7344 (void) printf("Verified indirect_refcount feature refcount " \
7345 "of %llu is correct\n",
7346 (u_longlong_t
)oc_feature_refcount
);
7352 zdb_set_skip_mmp(char *target
)
7357 * Disable the activity check to allow examination of
7360 mutex_enter(&spa_namespace_lock
);
7361 if ((spa
= spa_lookup(target
)) != NULL
) {
7362 spa
->spa_import_flags
|= ZFS_IMPORT_SKIP_MMP
;
7364 mutex_exit(&spa_namespace_lock
);
7367 #define BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7369 * Import the checkpointed state of the pool specified by the target
7370 * parameter as readonly. The function also accepts a pool config
7371 * as an optional parameter, else it attempts to infer the config by
7372 * the name of the target pool.
7374 * Note that the checkpointed state's pool name will be the name of
7375 * the original pool with the above suffix appended to it. In addition,
7376 * if the target is not a pool name (e.g. a path to a dataset) then
7377 * the new_path parameter is populated with the updated path to
7378 * reflect the fact that we are looking into the checkpointed state.
7380 * The function returns a newly-allocated copy of the name of the
7381 * pool containing the checkpointed state. When this copy is no
7382 * longer needed it should be freed with free(3C). Same thing
7383 * applies to the new_path parameter if allocated.
7386 import_checkpointed_state(char *target
, nvlist_t
*cfg
, char **new_path
)
7389 char *poolname
, *bogus_name
= NULL
;
7390 boolean_t freecfg
= B_FALSE
;
7392 /* If the target is not a pool, the extract the pool name */
7393 char *path_start
= strchr(target
, '/');
7394 if (path_start
!= NULL
) {
7395 size_t poolname_len
= path_start
- target
;
7396 poolname
= strndup(target
, poolname_len
);
7402 zdb_set_skip_mmp(poolname
);
7403 error
= spa_get_stats(poolname
, &cfg
, NULL
, 0);
7405 fatal("Tried to read config of pool \"%s\" but "
7406 "spa_get_stats() failed with error %d\n",
7412 if (asprintf(&bogus_name
, "%s%s", poolname
, BOGUS_SUFFIX
) == -1) {
7413 if (target
!= poolname
)
7417 fnvlist_add_string(cfg
, ZPOOL_CONFIG_POOL_NAME
, bogus_name
);
7419 error
= spa_import(bogus_name
, cfg
, NULL
,
7420 ZFS_IMPORT_MISSING_LOG
| ZFS_IMPORT_CHECKPOINT
|
7421 ZFS_IMPORT_SKIP_MMP
);
7425 fatal("Tried to import pool \"%s\" but spa_import() failed "
7426 "with error %d\n", bogus_name
, error
);
7429 if (new_path
!= NULL
&& path_start
!= NULL
) {
7430 if (asprintf(new_path
, "%s%s", bogus_name
, path_start
) == -1) {
7432 if (path_start
!= NULL
)
7438 if (target
!= poolname
)
7441 return (bogus_name
);
7444 typedef struct verify_checkpoint_sm_entry_cb_arg
{
7447 /* the following fields are only used for printing progress */
7448 uint64_t vcsec_entryid
;
7449 uint64_t vcsec_num_entries
;
7450 } verify_checkpoint_sm_entry_cb_arg_t
;
7452 #define ENTRIES_PER_PROGRESS_UPDATE 10000
7455 verify_checkpoint_sm_entry_cb(space_map_entry_t
*sme
, void *arg
)
7457 verify_checkpoint_sm_entry_cb_arg_t
*vcsec
= arg
;
7458 vdev_t
*vd
= vcsec
->vcsec_vd
;
7459 metaslab_t
*ms
= vd
->vdev_ms
[sme
->sme_offset
>> vd
->vdev_ms_shift
];
7460 uint64_t end
= sme
->sme_offset
+ sme
->sme_run
;
7462 ASSERT(sme
->sme_type
== SM_FREE
);
7464 if ((vcsec
->vcsec_entryid
% ENTRIES_PER_PROGRESS_UPDATE
) == 0) {
7465 (void) fprintf(stderr
,
7466 "\rverifying vdev %llu, space map entry %llu of %llu ...",
7467 (longlong_t
)vd
->vdev_id
,
7468 (longlong_t
)vcsec
->vcsec_entryid
,
7469 (longlong_t
)vcsec
->vcsec_num_entries
);
7471 vcsec
->vcsec_entryid
++;
7474 * See comment in checkpoint_sm_exclude_entry_cb()
7476 VERIFY3U(sme
->sme_offset
, >=, ms
->ms_start
);
7477 VERIFY3U(end
, <=, ms
->ms_start
+ ms
->ms_size
);
7480 * The entries in the vdev_checkpoint_sm should be marked as
7481 * allocated in the checkpointed state of the pool, therefore
7482 * their respective ms_allocateable trees should not contain them.
7484 mutex_enter(&ms
->ms_lock
);
7485 range_tree_verify_not_present(ms
->ms_allocatable
,
7486 sme
->sme_offset
, sme
->sme_run
);
7487 mutex_exit(&ms
->ms_lock
);
7493 * Verify that all segments in the vdev_checkpoint_sm are allocated
7494 * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
7497 * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
7498 * each vdev in the current state of the pool to the metaslab space maps
7499 * (ms_sm) of the checkpointed state of the pool.
7501 * Note that the function changes the state of the ms_allocatable
7502 * trees of the current spa_t. The entries of these ms_allocatable
7503 * trees are cleared out and then repopulated from with the free
7504 * entries of their respective ms_sm space maps.
7507 verify_checkpoint_vdev_spacemaps(spa_t
*checkpoint
, spa_t
*current
)
7509 vdev_t
*ckpoint_rvd
= checkpoint
->spa_root_vdev
;
7510 vdev_t
*current_rvd
= current
->spa_root_vdev
;
7512 load_concrete_ms_allocatable_trees(checkpoint
, SM_FREE
);
7514 for (uint64_t c
= 0; c
< ckpoint_rvd
->vdev_children
; c
++) {
7515 vdev_t
*ckpoint_vd
= ckpoint_rvd
->vdev_child
[c
];
7516 vdev_t
*current_vd
= current_rvd
->vdev_child
[c
];
7518 space_map_t
*checkpoint_sm
= NULL
;
7519 uint64_t checkpoint_sm_obj
;
7521 if (ckpoint_vd
->vdev_ops
== &vdev_indirect_ops
) {
7523 * Since we don't allow device removal in a pool
7524 * that has a checkpoint, we expect that all removed
7525 * vdevs were removed from the pool before the
7528 ASSERT3P(current_vd
->vdev_ops
, ==, &vdev_indirect_ops
);
7533 * If the checkpoint space map doesn't exist, then nothing
7534 * here is checkpointed so there's nothing to verify.
7536 if (current_vd
->vdev_top_zap
== 0 ||
7537 zap_contains(spa_meta_objset(current
),
7538 current_vd
->vdev_top_zap
,
7539 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) != 0)
7542 VERIFY0(zap_lookup(spa_meta_objset(current
),
7543 current_vd
->vdev_top_zap
, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
,
7544 sizeof (uint64_t), 1, &checkpoint_sm_obj
));
7546 VERIFY0(space_map_open(&checkpoint_sm
, spa_meta_objset(current
),
7547 checkpoint_sm_obj
, 0, current_vd
->vdev_asize
,
7548 current_vd
->vdev_ashift
));
7550 verify_checkpoint_sm_entry_cb_arg_t vcsec
;
7551 vcsec
.vcsec_vd
= ckpoint_vd
;
7552 vcsec
.vcsec_entryid
= 0;
7553 vcsec
.vcsec_num_entries
=
7554 space_map_length(checkpoint_sm
) / sizeof (uint64_t);
7555 VERIFY0(space_map_iterate(checkpoint_sm
,
7556 space_map_length(checkpoint_sm
),
7557 verify_checkpoint_sm_entry_cb
, &vcsec
));
7558 if (dump_opt
['m'] > 3)
7559 dump_spacemap(current
->spa_meta_objset
, checkpoint_sm
);
7560 space_map_close(checkpoint_sm
);
7564 * If we've added vdevs since we took the checkpoint, ensure
7565 * that their checkpoint space maps are empty.
7567 if (ckpoint_rvd
->vdev_children
< current_rvd
->vdev_children
) {
7568 for (uint64_t c
= ckpoint_rvd
->vdev_children
;
7569 c
< current_rvd
->vdev_children
; c
++) {
7570 vdev_t
*current_vd
= current_rvd
->vdev_child
[c
];
7571 VERIFY3P(current_vd
->vdev_checkpoint_sm
, ==, NULL
);
7575 /* for cleaner progress output */
7576 (void) fprintf(stderr
, "\n");
7580 * Verifies that all space that's allocated in the checkpoint is
7581 * still allocated in the current version, by checking that everything
7582 * in checkpoint's ms_allocatable (which is actually allocated, not
7583 * allocatable/free) is not present in current's ms_allocatable.
7585 * Note that the function changes the state of the ms_allocatable
7586 * trees of both spas when called. The entries of all ms_allocatable
7587 * trees are cleared out and then repopulated from their respective
7588 * ms_sm space maps. In the checkpointed state we load the allocated
7589 * entries, and in the current state we load the free entries.
7592 verify_checkpoint_ms_spacemaps(spa_t
*checkpoint
, spa_t
*current
)
7594 vdev_t
*ckpoint_rvd
= checkpoint
->spa_root_vdev
;
7595 vdev_t
*current_rvd
= current
->spa_root_vdev
;
7597 load_concrete_ms_allocatable_trees(checkpoint
, SM_ALLOC
);
7598 load_concrete_ms_allocatable_trees(current
, SM_FREE
);
7600 for (uint64_t i
= 0; i
< ckpoint_rvd
->vdev_children
; i
++) {
7601 vdev_t
*ckpoint_vd
= ckpoint_rvd
->vdev_child
[i
];
7602 vdev_t
*current_vd
= current_rvd
->vdev_child
[i
];
7604 if (ckpoint_vd
->vdev_ops
== &vdev_indirect_ops
) {
7606 * See comment in verify_checkpoint_vdev_spacemaps()
7608 ASSERT3P(current_vd
->vdev_ops
, ==, &vdev_indirect_ops
);
7612 for (uint64_t m
= 0; m
< ckpoint_vd
->vdev_ms_count
; m
++) {
7613 metaslab_t
*ckpoint_msp
= ckpoint_vd
->vdev_ms
[m
];
7614 metaslab_t
*current_msp
= current_vd
->vdev_ms
[m
];
7616 (void) fprintf(stderr
,
7617 "\rverifying vdev %llu of %llu, "
7618 "metaslab %llu of %llu ...",
7619 (longlong_t
)current_vd
->vdev_id
,
7620 (longlong_t
)current_rvd
->vdev_children
,
7621 (longlong_t
)current_vd
->vdev_ms
[m
]->ms_id
,
7622 (longlong_t
)current_vd
->vdev_ms_count
);
7625 * We walk through the ms_allocatable trees that
7626 * are loaded with the allocated blocks from the
7627 * ms_sm spacemaps of the checkpoint. For each
7628 * one of these ranges we ensure that none of them
7629 * exists in the ms_allocatable trees of the
7630 * current state which are loaded with the ranges
7631 * that are currently free.
7633 * This way we ensure that none of the blocks that
7634 * are part of the checkpoint were freed by mistake.
7636 range_tree_walk(ckpoint_msp
->ms_allocatable
,
7637 (range_tree_func_t
*)range_tree_verify_not_present
,
7638 current_msp
->ms_allocatable
);
7642 /* for cleaner progress output */
7643 (void) fprintf(stderr
, "\n");
7647 verify_checkpoint_blocks(spa_t
*spa
)
7649 ASSERT(!dump_opt
['L']);
7651 spa_t
*checkpoint_spa
;
7652 char *checkpoint_pool
;
7656 * We import the checkpointed state of the pool (under a different
7657 * name) so we can do verification on it against the current state
7660 checkpoint_pool
= import_checkpointed_state(spa
->spa_name
, NULL
,
7662 ASSERT(strcmp(spa
->spa_name
, checkpoint_pool
) != 0);
7664 error
= spa_open(checkpoint_pool
, &checkpoint_spa
, FTAG
);
7666 fatal("Tried to open pool \"%s\" but spa_open() failed with "
7667 "error %d\n", checkpoint_pool
, error
);
7671 * Ensure that ranges in the checkpoint space maps of each vdev
7672 * are allocated according to the checkpointed state's metaslab
7675 verify_checkpoint_vdev_spacemaps(checkpoint_spa
, spa
);
7678 * Ensure that allocated ranges in the checkpoint's metaslab
7679 * space maps remain allocated in the metaslab space maps of
7680 * the current state.
7682 verify_checkpoint_ms_spacemaps(checkpoint_spa
, spa
);
7685 * Once we are done, we get rid of the checkpointed state.
7687 spa_close(checkpoint_spa
, FTAG
);
7688 free(checkpoint_pool
);
7692 dump_leftover_checkpoint_blocks(spa_t
*spa
)
7694 vdev_t
*rvd
= spa
->spa_root_vdev
;
7696 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
7697 vdev_t
*vd
= rvd
->vdev_child
[i
];
7699 space_map_t
*checkpoint_sm
= NULL
;
7700 uint64_t checkpoint_sm_obj
;
7702 if (vd
->vdev_top_zap
== 0)
7705 if (zap_contains(spa_meta_objset(spa
), vd
->vdev_top_zap
,
7706 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) != 0)
7709 VERIFY0(zap_lookup(spa_meta_objset(spa
), vd
->vdev_top_zap
,
7710 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
,
7711 sizeof (uint64_t), 1, &checkpoint_sm_obj
));
7713 VERIFY0(space_map_open(&checkpoint_sm
, spa_meta_objset(spa
),
7714 checkpoint_sm_obj
, 0, vd
->vdev_asize
, vd
->vdev_ashift
));
7715 dump_spacemap(spa
->spa_meta_objset
, checkpoint_sm
);
7716 space_map_close(checkpoint_sm
);
7721 verify_checkpoint(spa_t
*spa
)
7723 uberblock_t checkpoint
;
7726 if (!spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
))
7729 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
7730 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
7731 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
7733 if (error
== ENOENT
&& !dump_opt
['L']) {
7735 * If the feature is active but the uberblock is missing
7736 * then we must be in the middle of discarding the
7739 (void) printf("\nPartially discarded checkpoint "
7741 if (dump_opt
['m'] > 3)
7742 dump_leftover_checkpoint_blocks(spa
);
7744 } else if (error
!= 0) {
7745 (void) printf("lookup error %d when looking for "
7746 "checkpointed uberblock in MOS\n", error
);
7749 dump_uberblock(&checkpoint
, "\nCheckpointed uberblock found:\n", "\n");
7751 if (checkpoint
.ub_checkpoint_txg
== 0) {
7752 (void) printf("\nub_checkpoint_txg not set in checkpointed "
7757 if (error
== 0 && !dump_opt
['L'])
7758 verify_checkpoint_blocks(spa
);
7764 mos_leaks_cb(void *arg
, uint64_t start
, uint64_t size
)
7767 for (uint64_t i
= start
; i
< size
; i
++) {
7768 (void) printf("MOS object %llu referenced but not allocated\n",
7774 mos_obj_refd(uint64_t obj
)
7776 if (obj
!= 0 && mos_refd_objs
!= NULL
)
7777 range_tree_add(mos_refd_objs
, obj
, 1);
7781 * Call on a MOS object that may already have been referenced.
7784 mos_obj_refd_multiple(uint64_t obj
)
7786 if (obj
!= 0 && mos_refd_objs
!= NULL
&&
7787 !range_tree_contains(mos_refd_objs
, obj
, 1))
7788 range_tree_add(mos_refd_objs
, obj
, 1);
7792 mos_leak_vdev_top_zap(vdev_t
*vd
)
7794 uint64_t ms_flush_data_obj
;
7795 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
),
7796 vd
->vdev_top_zap
, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
,
7797 sizeof (ms_flush_data_obj
), 1, &ms_flush_data_obj
);
7798 if (error
== ENOENT
)
7802 mos_obj_refd(ms_flush_data_obj
);
7806 mos_leak_vdev(vdev_t
*vd
)
7808 mos_obj_refd(vd
->vdev_dtl_object
);
7809 mos_obj_refd(vd
->vdev_ms_array
);
7810 mos_obj_refd(vd
->vdev_indirect_config
.vic_births_object
);
7811 mos_obj_refd(vd
->vdev_indirect_config
.vic_mapping_object
);
7812 mos_obj_refd(vd
->vdev_leaf_zap
);
7813 if (vd
->vdev_checkpoint_sm
!= NULL
)
7814 mos_obj_refd(vd
->vdev_checkpoint_sm
->sm_object
);
7815 if (vd
->vdev_indirect_mapping
!= NULL
) {
7816 mos_obj_refd(vd
->vdev_indirect_mapping
->
7817 vim_phys
->vimp_counts_object
);
7819 if (vd
->vdev_obsolete_sm
!= NULL
)
7820 mos_obj_refd(vd
->vdev_obsolete_sm
->sm_object
);
7822 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
7823 metaslab_t
*ms
= vd
->vdev_ms
[m
];
7824 mos_obj_refd(space_map_object(ms
->ms_sm
));
7827 if (vd
->vdev_root_zap
!= 0)
7828 mos_obj_refd(vd
->vdev_root_zap
);
7830 if (vd
->vdev_top_zap
!= 0) {
7831 mos_obj_refd(vd
->vdev_top_zap
);
7832 mos_leak_vdev_top_zap(vd
);
7835 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++) {
7836 mos_leak_vdev(vd
->vdev_child
[c
]);
7841 mos_leak_log_spacemaps(spa_t
*spa
)
7843 uint64_t spacemap_zap
;
7844 int error
= zap_lookup(spa_meta_objset(spa
),
7845 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_LOG_SPACEMAP_ZAP
,
7846 sizeof (spacemap_zap
), 1, &spacemap_zap
);
7847 if (error
== ENOENT
)
7851 mos_obj_refd(spacemap_zap
);
7852 for (spa_log_sm_t
*sls
= avl_first(&spa
->spa_sm_logs_by_txg
);
7853 sls
; sls
= AVL_NEXT(&spa
->spa_sm_logs_by_txg
, sls
))
7854 mos_obj_refd(sls
->sls_sm_obj
);
7858 errorlog_count_refd(objset_t
*mos
, uint64_t errlog
)
7862 for (zap_cursor_init(&zc
, mos
, errlog
);
7863 zap_cursor_retrieve(&zc
, &za
) == 0;
7864 zap_cursor_advance(&zc
)) {
7865 mos_obj_refd(za
.za_first_integer
);
7867 zap_cursor_fini(&zc
);
7871 dump_mos_leaks(spa_t
*spa
)
7874 objset_t
*mos
= spa
->spa_meta_objset
;
7875 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7877 /* Visit and mark all referenced objects in the MOS */
7879 mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT
);
7880 mos_obj_refd(spa
->spa_pool_props_object
);
7881 mos_obj_refd(spa
->spa_config_object
);
7882 mos_obj_refd(spa
->spa_ddt_stat_object
);
7883 mos_obj_refd(spa
->spa_feat_desc_obj
);
7884 mos_obj_refd(spa
->spa_feat_enabled_txg_obj
);
7885 mos_obj_refd(spa
->spa_feat_for_read_obj
);
7886 mos_obj_refd(spa
->spa_feat_for_write_obj
);
7887 mos_obj_refd(spa
->spa_history
);
7888 mos_obj_refd(spa
->spa_errlog_last
);
7889 mos_obj_refd(spa
->spa_errlog_scrub
);
7891 if (spa_feature_is_enabled(spa
, SPA_FEATURE_HEAD_ERRLOG
)) {
7892 errorlog_count_refd(mos
, spa
->spa_errlog_last
);
7893 errorlog_count_refd(mos
, spa
->spa_errlog_scrub
);
7896 mos_obj_refd(spa
->spa_all_vdev_zaps
);
7897 mos_obj_refd(spa
->spa_dsl_pool
->dp_bptree_obj
);
7898 mos_obj_refd(spa
->spa_dsl_pool
->dp_tmp_userrefs_obj
);
7899 mos_obj_refd(spa
->spa_dsl_pool
->dp_scan
->scn_phys
.scn_queue_obj
);
7900 bpobj_count_refd(&spa
->spa_deferred_bpobj
);
7901 mos_obj_refd(dp
->dp_empty_bpobj
);
7902 bpobj_count_refd(&dp
->dp_obsolete_bpobj
);
7903 bpobj_count_refd(&dp
->dp_free_bpobj
);
7904 mos_obj_refd(spa
->spa_l2cache
.sav_object
);
7905 mos_obj_refd(spa
->spa_spares
.sav_object
);
7907 if (spa
->spa_syncing_log_sm
!= NULL
)
7908 mos_obj_refd(spa
->spa_syncing_log_sm
->sm_object
);
7909 mos_leak_log_spacemaps(spa
);
7911 mos_obj_refd(spa
->spa_condensing_indirect_phys
.
7912 scip_next_mapping_object
);
7913 mos_obj_refd(spa
->spa_condensing_indirect_phys
.
7914 scip_prev_obsolete_sm_object
);
7915 if (spa
->spa_condensing_indirect_phys
.scip_next_mapping_object
!= 0) {
7916 vdev_indirect_mapping_t
*vim
=
7917 vdev_indirect_mapping_open(mos
,
7918 spa
->spa_condensing_indirect_phys
.scip_next_mapping_object
);
7919 mos_obj_refd(vim
->vim_phys
->vimp_counts_object
);
7920 vdev_indirect_mapping_close(vim
);
7922 deleted_livelists_dump_mos(spa
);
7924 if (dp
->dp_origin_snap
!= NULL
) {
7927 dsl_pool_config_enter(dp
, FTAG
);
7928 VERIFY0(dsl_dataset_hold_obj(dp
,
7929 dsl_dataset_phys(dp
->dp_origin_snap
)->ds_next_snap_obj
,
7931 count_ds_mos_objects(ds
);
7932 dump_blkptr_list(&ds
->ds_deadlist
, "Deadlist");
7933 dsl_dataset_rele(ds
, FTAG
);
7934 dsl_pool_config_exit(dp
, FTAG
);
7936 count_ds_mos_objects(dp
->dp_origin_snap
);
7937 dump_blkptr_list(&dp
->dp_origin_snap
->ds_deadlist
, "Deadlist");
7939 count_dir_mos_objects(dp
->dp_mos_dir
);
7940 if (dp
->dp_free_dir
!= NULL
)
7941 count_dir_mos_objects(dp
->dp_free_dir
);
7942 if (dp
->dp_leak_dir
!= NULL
)
7943 count_dir_mos_objects(dp
->dp_leak_dir
);
7945 mos_leak_vdev(spa
->spa_root_vdev
);
7947 for (uint64_t class = 0; class < DDT_CLASSES
; class++) {
7948 for (uint64_t type
= 0; type
< DDT_TYPES
; type
++) {
7949 for (uint64_t cksum
= 0;
7950 cksum
< ZIO_CHECKSUM_FUNCTIONS
; cksum
++) {
7951 ddt_t
*ddt
= spa
->spa_ddt
[cksum
];
7952 mos_obj_refd(ddt
->ddt_object
[type
][class]);
7957 if (spa
->spa_brt
!= NULL
) {
7958 brt_t
*brt
= spa
->spa_brt
;
7959 for (uint64_t vdevid
= 0; vdevid
< brt
->brt_nvdevs
; vdevid
++) {
7960 brt_vdev_t
*brtvd
= &brt
->brt_vdevs
[vdevid
];
7961 if (brtvd
!= NULL
&& brtvd
->bv_initiated
) {
7962 mos_obj_refd(brtvd
->bv_mos_brtvdev
);
7963 mos_obj_refd(brtvd
->bv_mos_entries
);
7969 * Visit all allocated objects and make sure they are referenced.
7971 uint64_t object
= 0;
7972 while (dmu_object_next(mos
, &object
, B_FALSE
, 0) == 0) {
7973 if (range_tree_contains(mos_refd_objs
, object
, 1)) {
7974 range_tree_remove(mos_refd_objs
, object
, 1);
7976 dmu_object_info_t doi
;
7978 VERIFY0(dmu_object_info(mos
, object
, &doi
));
7979 if (doi
.doi_type
& DMU_OT_NEWTYPE
) {
7980 dmu_object_byteswap_t bswap
=
7981 DMU_OT_BYTESWAP(doi
.doi_type
);
7982 name
= dmu_ot_byteswap
[bswap
].ob_name
;
7984 name
= dmu_ot
[doi
.doi_type
].ot_name
;
7987 (void) printf("MOS object %llu (%s) leaked\n",
7988 (u_longlong_t
)object
, name
);
7992 (void) range_tree_walk(mos_refd_objs
, mos_leaks_cb
, NULL
);
7993 if (!range_tree_is_empty(mos_refd_objs
))
7995 range_tree_vacate(mos_refd_objs
, NULL
, NULL
);
7996 range_tree_destroy(mos_refd_objs
);
8000 typedef struct log_sm_obsolete_stats_arg
{
8001 uint64_t lsos_current_txg
;
8003 uint64_t lsos_total_entries
;
8004 uint64_t lsos_valid_entries
;
8006 uint64_t lsos_sm_entries
;
8007 uint64_t lsos_valid_sm_entries
;
8008 } log_sm_obsolete_stats_arg_t
;
8011 log_spacemap_obsolete_stats_cb(spa_t
*spa
, space_map_entry_t
*sme
,
8012 uint64_t txg
, void *arg
)
8014 log_sm_obsolete_stats_arg_t
*lsos
= arg
;
8016 uint64_t offset
= sme
->sme_offset
;
8017 uint64_t vdev_id
= sme
->sme_vdev
;
8019 if (lsos
->lsos_current_txg
== 0) {
8020 /* this is the first log */
8021 lsos
->lsos_current_txg
= txg
;
8022 } else if (lsos
->lsos_current_txg
< txg
) {
8023 /* we just changed log - print stats and reset */
8024 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8025 (u_longlong_t
)lsos
->lsos_valid_sm_entries
,
8026 (u_longlong_t
)lsos
->lsos_sm_entries
,
8027 (u_longlong_t
)lsos
->lsos_current_txg
);
8028 lsos
->lsos_valid_sm_entries
= 0;
8029 lsos
->lsos_sm_entries
= 0;
8030 lsos
->lsos_current_txg
= txg
;
8032 ASSERT3U(lsos
->lsos_current_txg
, ==, txg
);
8034 lsos
->lsos_sm_entries
++;
8035 lsos
->lsos_total_entries
++;
8037 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
8038 if (!vdev_is_concrete(vd
))
8041 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
8042 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
8044 if (txg
< metaslab_unflushed_txg(ms
))
8046 lsos
->lsos_valid_sm_entries
++;
8047 lsos
->lsos_valid_entries
++;
8052 dump_log_spacemap_obsolete_stats(spa_t
*spa
)
8054 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
8057 log_sm_obsolete_stats_arg_t lsos
= {0};
8059 (void) printf("Log Space Map Obsolete Entry Statistics:\n");
8061 iterate_through_spacemap_logs(spa
,
8062 log_spacemap_obsolete_stats_cb
, &lsos
);
8064 /* print stats for latest log */
8065 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8066 (u_longlong_t
)lsos
.lsos_valid_sm_entries
,
8067 (u_longlong_t
)lsos
.lsos_sm_entries
,
8068 (u_longlong_t
)lsos
.lsos_current_txg
);
8070 (void) printf("%-8llu valid entries out of %-8llu - total\n\n",
8071 (u_longlong_t
)lsos
.lsos_valid_entries
,
8072 (u_longlong_t
)lsos
.lsos_total_entries
);
8076 dump_zpool(spa_t
*spa
)
8078 dsl_pool_t
*dp
= spa_get_dsl(spa
);
8081 if (dump_opt
['y']) {
8082 livelist_metaslab_validate(spa
);
8085 if (dump_opt
['S']) {
8086 dump_simulated_ddt(spa
);
8090 if (!dump_opt
['e'] && dump_opt
['C'] > 1) {
8091 (void) printf("\nCached configuration:\n");
8092 dump_nvlist(spa
->spa_config
, 8);
8099 dump_uberblock(&spa
->spa_uberblock
, "\nUberblock:\n", "\n");
8107 if (dump_opt
['d'] > 2 || dump_opt
['m'])
8108 dump_metaslabs(spa
);
8110 dump_metaslab_groups(spa
, dump_opt
['M'] > 1);
8111 if (dump_opt
['d'] > 2 || dump_opt
['m']) {
8112 dump_log_spacemaps(spa
);
8113 dump_log_spacemap_obsolete_stats(spa
);
8116 if (dump_opt
['d'] || dump_opt
['i']) {
8118 mos_refd_objs
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0,
8120 dump_objset(dp
->dp_meta_objset
);
8122 if (dump_opt
['d'] >= 3) {
8123 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8124 dump_full_bpobj(&spa
->spa_deferred_bpobj
,
8125 "Deferred frees", 0);
8126 if (spa_version(spa
) >= SPA_VERSION_DEADLISTS
) {
8127 dump_full_bpobj(&dp
->dp_free_bpobj
,
8128 "Pool snapshot frees", 0);
8130 if (bpobj_is_open(&dp
->dp_obsolete_bpobj
)) {
8131 ASSERT(spa_feature_is_enabled(spa
,
8132 SPA_FEATURE_DEVICE_REMOVAL
));
8133 dump_full_bpobj(&dp
->dp_obsolete_bpobj
,
8134 "Pool obsolete blocks", 0);
8137 if (spa_feature_is_active(spa
,
8138 SPA_FEATURE_ASYNC_DESTROY
)) {
8139 dump_bptree(spa
->spa_meta_objset
,
8141 "Pool dataset frees");
8143 dump_dtl(spa
->spa_root_vdev
, 0);
8146 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++)
8147 global_feature_count
[f
] = UINT64_MAX
;
8148 global_feature_count
[SPA_FEATURE_REDACTION_BOOKMARKS
] = 0;
8149 global_feature_count
[SPA_FEATURE_REDACTION_LIST_SPILL
] = 0;
8150 global_feature_count
[SPA_FEATURE_BOOKMARK_WRITTEN
] = 0;
8151 global_feature_count
[SPA_FEATURE_LIVELIST
] = 0;
8153 (void) dmu_objset_find(spa_name(spa
), dump_one_objset
,
8154 NULL
, DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
8156 if (rc
== 0 && !dump_opt
['L'])
8157 rc
= dump_mos_leaks(spa
);
8159 for (f
= 0; f
< SPA_FEATURES
; f
++) {
8163 if (!(spa_feature_table
[f
].fi_flags
&
8164 ZFEATURE_FLAG_PER_DATASET
)) {
8165 if (global_feature_count
[f
] == UINT64_MAX
)
8167 if (!spa_feature_is_enabled(spa
, f
)) {
8168 ASSERT0(global_feature_count
[f
]);
8171 arr
= global_feature_count
;
8173 if (!spa_feature_is_enabled(spa
, f
)) {
8174 ASSERT0(dataset_feature_count
[f
]);
8177 arr
= dataset_feature_count
;
8179 if (feature_get_refcount(spa
, &spa_feature_table
[f
],
8180 &refcount
) == ENOTSUP
)
8182 if (arr
[f
] != refcount
) {
8183 (void) printf("%s feature refcount mismatch: "
8184 "%lld consumers != %lld refcount\n",
8185 spa_feature_table
[f
].fi_uname
,
8186 (longlong_t
)arr
[f
], (longlong_t
)refcount
);
8189 (void) printf("Verified %s feature refcount "
8190 "of %llu is correct\n",
8191 spa_feature_table
[f
].fi_uname
,
8192 (longlong_t
)refcount
);
8197 rc
= verify_device_removal_feature_counts(spa
);
8200 if (rc
== 0 && (dump_opt
['b'] || dump_opt
['c']))
8201 rc
= dump_block_stats(spa
);
8204 rc
= verify_spacemap_refcounts(spa
);
8207 show_pool_stats(spa
);
8213 rc
= verify_checkpoint(spa
);
8216 dump_debug_buffer();
8221 #define ZDB_FLAG_CHECKSUM 0x0001
8222 #define ZDB_FLAG_DECOMPRESS 0x0002
8223 #define ZDB_FLAG_BSWAP 0x0004
8224 #define ZDB_FLAG_GBH 0x0008
8225 #define ZDB_FLAG_INDIRECT 0x0010
8226 #define ZDB_FLAG_RAW 0x0020
8227 #define ZDB_FLAG_PRINT_BLKPTR 0x0040
8228 #define ZDB_FLAG_VERBOSE 0x0080
8230 static int flagbits
[256];
8231 static char flagbitstr
[16];
8234 zdb_print_blkptr(const blkptr_t
*bp
, int flags
)
8236 char blkbuf
[BP_SPRINTF_LEN
];
8238 if (flags
& ZDB_FLAG_BSWAP
)
8239 byteswap_uint64_array((void *)bp
, sizeof (blkptr_t
));
8241 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
8242 (void) printf("%s\n", blkbuf
);
8246 zdb_dump_indirect(blkptr_t
*bp
, int nbps
, int flags
)
8250 for (i
= 0; i
< nbps
; i
++)
8251 zdb_print_blkptr(&bp
[i
], flags
);
8255 zdb_dump_gbh(void *buf
, int flags
)
8257 zdb_dump_indirect((blkptr_t
*)buf
, SPA_GBH_NBLKPTRS
, flags
);
8261 zdb_dump_block_raw(void *buf
, uint64_t size
, int flags
)
8263 if (flags
& ZDB_FLAG_BSWAP
)
8264 byteswap_uint64_array(buf
, size
);
8265 VERIFY(write(fileno(stdout
), buf
, size
) == size
);
8269 zdb_dump_block(char *label
, void *buf
, uint64_t size
, int flags
)
8271 uint64_t *d
= (uint64_t *)buf
;
8272 unsigned nwords
= size
/ sizeof (uint64_t);
8273 int do_bswap
= !!(flags
& ZDB_FLAG_BSWAP
);
8280 hdr
= " 7 6 5 4 3 2 1 0 f e d c b a 9 8";
8282 hdr
= " 0 1 2 3 4 5 6 7 8 9 a b c d e f";
8284 (void) printf("\n%s\n%6s %s 0123456789abcdef\n", label
, "", hdr
);
8286 #ifdef _LITTLE_ENDIAN
8287 /* correct the endianness */
8288 do_bswap
= !do_bswap
;
8290 for (i
= 0; i
< nwords
; i
+= 2) {
8291 (void) printf("%06llx: %016llx %016llx ",
8292 (u_longlong_t
)(i
* sizeof (uint64_t)),
8293 (u_longlong_t
)(do_bswap
? BSWAP_64(d
[i
]) : d
[i
]),
8294 (u_longlong_t
)(do_bswap
? BSWAP_64(d
[i
+ 1]) : d
[i
+ 1]));
8297 for (j
= 0; j
< 2 * sizeof (uint64_t); j
++)
8298 (void) printf("%c", isprint(c
[j
]) ? c
[j
] : '.');
8299 (void) printf("\n");
8304 * There are two acceptable formats:
8305 * leaf_name - For example: c1t0d0 or /tmp/ztest.0a
8306 * child[.child]* - For example: 0.1.1
8308 * The second form can be used to specify arbitrary vdevs anywhere
8309 * in the hierarchy. For example, in a pool with a mirror of
8310 * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
8313 zdb_vdev_lookup(vdev_t
*vdev
, const char *path
)
8321 /* First, assume the x.x.x.x format */
8322 i
= strtoul(path
, &s
, 10);
8323 if (s
== path
|| (s
&& *s
!= '.' && *s
!= '\0'))
8325 if (i
>= vdev
->vdev_children
)
8328 vdev
= vdev
->vdev_child
[i
];
8329 if (s
&& *s
== '\0')
8331 return (zdb_vdev_lookup(vdev
, s
+1));
8334 for (i
= 0; i
< vdev
->vdev_children
; i
++) {
8335 vdev_t
*vc
= vdev
->vdev_child
[i
];
8337 if (vc
->vdev_path
== NULL
) {
8338 vc
= zdb_vdev_lookup(vc
, path
);
8345 p
= strrchr(vc
->vdev_path
, '/');
8346 p
= p
? p
+ 1 : vc
->vdev_path
;
8347 q
= &vc
->vdev_path
[strlen(vc
->vdev_path
) - 2];
8349 if (strcmp(vc
->vdev_path
, path
) == 0)
8351 if (strcmp(p
, path
) == 0)
8353 if (strcmp(q
, "s0") == 0 && strncmp(p
, path
, q
- p
) == 0)
8361 name_from_objset_id(spa_t
*spa
, uint64_t objset_id
, char *outstr
)
8365 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
8366 int error
= dsl_dataset_hold_obj(spa
->spa_dsl_pool
, objset_id
,
8369 (void) fprintf(stderr
, "failed to hold objset %llu: %s\n",
8370 (u_longlong_t
)objset_id
, strerror(error
));
8371 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
8374 dsl_dataset_name(ds
, outstr
);
8375 dsl_dataset_rele(ds
, NULL
);
8376 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
8381 zdb_parse_block_sizes(char *sizes
, uint64_t *lsize
, uint64_t *psize
)
8383 char *s0
, *s1
, *tmp
= NULL
;
8388 s0
= strtok_r(sizes
, "/", &tmp
);
8391 s1
= strtok_r(NULL
, "/", &tmp
);
8392 *lsize
= strtoull(s0
, NULL
, 16);
8393 *psize
= s1
? strtoull(s1
, NULL
, 16) : *lsize
;
8394 return (*lsize
>= *psize
&& *psize
> 0);
8397 #define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg))
8400 try_decompress_block(abd_t
*pabd
, uint64_t lsize
, uint64_t psize
,
8401 int flags
, int cfunc
, void *lbuf
, void *lbuf2
)
8403 if (flags
& ZDB_FLAG_VERBOSE
) {
8404 (void) fprintf(stderr
,
8405 "Trying %05llx -> %05llx (%s)\n",
8406 (u_longlong_t
)psize
,
8407 (u_longlong_t
)lsize
,
8408 zio_compress_table
[cfunc
].ci_name
);
8412 * We set lbuf to all zeros and lbuf2 to all
8413 * ones, then decompress to both buffers and
8414 * compare their contents. This way we can
8415 * know if decompression filled exactly to
8416 * lsize or if it left some bytes unwritten.
8419 memset(lbuf
, 0x00, lsize
);
8420 memset(lbuf2
, 0xff, lsize
);
8422 if (zio_decompress_data(cfunc
, pabd
,
8423 lbuf
, psize
, lsize
, NULL
) == 0 &&
8424 zio_decompress_data(cfunc
, pabd
,
8425 lbuf2
, psize
, lsize
, NULL
) == 0 &&
8426 memcmp(lbuf
, lbuf2
, lsize
) == 0)
8432 zdb_decompress_block(abd_t
*pabd
, void *buf
, void *lbuf
, uint64_t lsize
,
8433 uint64_t psize
, int flags
)
8436 uint64_t orig_lsize
= lsize
;
8437 boolean_t tryzle
= ((getenv("ZDB_NO_ZLE") == NULL
));
8438 boolean_t found
= B_FALSE
;
8440 * We don't know how the data was compressed, so just try
8441 * every decompress function at every inflated blocksize.
8443 void *lbuf2
= umem_alloc(SPA_MAXBLOCKSIZE
, UMEM_NOFAIL
);
8444 int cfuncs
[ZIO_COMPRESS_FUNCTIONS
] = { 0 };
8445 int *cfuncp
= cfuncs
;
8446 uint64_t maxlsize
= SPA_MAXBLOCKSIZE
;
8447 uint64_t mask
= ZIO_COMPRESS_MASK(ON
) | ZIO_COMPRESS_MASK(OFF
) |
8448 ZIO_COMPRESS_MASK(INHERIT
) | ZIO_COMPRESS_MASK(EMPTY
) |
8449 ZIO_COMPRESS_MASK(ZLE
);
8450 *cfuncp
++ = ZIO_COMPRESS_LZ4
;
8451 *cfuncp
++ = ZIO_COMPRESS_LZJB
;
8452 mask
|= ZIO_COMPRESS_MASK(LZ4
) | ZIO_COMPRESS_MASK(LZJB
);
8454 * Every gzip level has the same decompressor, no need to
8455 * run it 9 times per bruteforce attempt.
8457 mask
|= ZIO_COMPRESS_MASK(GZIP_2
) | ZIO_COMPRESS_MASK(GZIP_3
);
8458 mask
|= ZIO_COMPRESS_MASK(GZIP_4
) | ZIO_COMPRESS_MASK(GZIP_5
);
8459 mask
|= ZIO_COMPRESS_MASK(GZIP_6
) | ZIO_COMPRESS_MASK(GZIP_7
);
8460 mask
|= ZIO_COMPRESS_MASK(GZIP_8
) | ZIO_COMPRESS_MASK(GZIP_9
);
8461 for (int c
= 0; c
< ZIO_COMPRESS_FUNCTIONS
; c
++)
8462 if (((1ULL << c
) & mask
) == 0)
8466 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
8467 * could take a while and we should let the user know
8468 * we are not stuck. On the other hand, printing progress
8469 * info gets old after a while. User can specify 'v' flag
8470 * to see the progression.
8473 lsize
+= SPA_MINBLOCKSIZE
;
8477 for (; lsize
<= maxlsize
; lsize
+= SPA_MINBLOCKSIZE
) {
8478 for (cfuncp
= cfuncs
; *cfuncp
; cfuncp
++) {
8479 if (try_decompress_block(pabd
, lsize
, psize
, flags
,
8480 *cfuncp
, lbuf
, lbuf2
)) {
8488 if (!found
&& tryzle
) {
8489 for (lsize
= orig_lsize
; lsize
<= maxlsize
;
8490 lsize
+= SPA_MINBLOCKSIZE
) {
8491 if (try_decompress_block(pabd
, lsize
, psize
, flags
,
8492 ZIO_COMPRESS_ZLE
, lbuf
, lbuf2
)) {
8493 *cfuncp
= ZIO_COMPRESS_ZLE
;
8499 umem_free(lbuf2
, SPA_MAXBLOCKSIZE
);
8501 if (*cfuncp
== ZIO_COMPRESS_ZLE
) {
8502 printf("\nZLE decompression was selected. If you "
8503 "suspect the results are wrong,\ntry avoiding ZLE "
8504 "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
8507 return (lsize
> maxlsize
? -1 : lsize
);
8511 * Read a block from a pool and print it out. The syntax of the
8512 * block descriptor is:
8514 * pool:vdev_specifier:offset:[lsize/]psize[:flags]
8516 * pool - The name of the pool you wish to read from
8517 * vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
8518 * offset - offset, in hex, in bytes
8519 * size - Amount of data to read, in hex, in bytes
8520 * flags - A string of characters specifying options
8521 * b: Decode a blkptr at given offset within block
8522 * c: Calculate and display checksums
8523 * d: Decompress data before dumping
8524 * e: Byteswap data before dumping
8525 * g: Display data as a gang block header
8526 * i: Display as an indirect block
8527 * r: Dump raw data to stdout
8532 zdb_read_block(char *thing
, spa_t
*spa
)
8534 blkptr_t blk
, *bp
= &blk
;
8535 dva_t
*dva
= bp
->blk_dva
;
8537 uint64_t offset
= 0, psize
= 0, lsize
= 0, blkptr_offset
= 0;
8542 char *s
, *p
, *dup
, *flagstr
, *sizes
, *tmp
= NULL
;
8543 const char *vdev
, *errmsg
= NULL
;
8545 boolean_t borrowed
= B_FALSE
, found
= B_FALSE
;
8547 dup
= strdup(thing
);
8548 s
= strtok_r(dup
, ":", &tmp
);
8550 s
= strtok_r(NULL
, ":", &tmp
);
8551 offset
= strtoull(s
? s
: "", NULL
, 16);
8552 sizes
= strtok_r(NULL
, ":", &tmp
);
8553 s
= strtok_r(NULL
, ":", &tmp
);
8554 flagstr
= strdup(s
?: "");
8556 if (!zdb_parse_block_sizes(sizes
, &lsize
, &psize
))
8557 errmsg
= "invalid size(s)";
8558 if (!IS_P2ALIGNED(psize
, DEV_BSIZE
) || !IS_P2ALIGNED(lsize
, DEV_BSIZE
))
8559 errmsg
= "size must be a multiple of sector size";
8560 if (!IS_P2ALIGNED(offset
, DEV_BSIZE
))
8561 errmsg
= "offset must be a multiple of sector size";
8563 (void) printf("Invalid block specifier: %s - %s\n",
8569 for (s
= strtok_r(flagstr
, ":", &tmp
);
8571 s
= strtok_r(NULL
, ":", &tmp
)) {
8572 for (i
= 0; i
< strlen(flagstr
); i
++) {
8573 int bit
= flagbits
[(uchar_t
)flagstr
[i
]];
8576 (void) printf("***Ignoring flag: %c\n",
8577 (uchar_t
)flagstr
[i
]);
8583 p
= &flagstr
[i
+ 1];
8584 if (*p
!= ':' && *p
!= '\0') {
8585 int j
= 0, nextbit
= flagbits
[(uchar_t
)*p
];
8586 char *end
, offstr
[8] = { 0 };
8587 if ((bit
== ZDB_FLAG_PRINT_BLKPTR
) &&
8589 /* look ahead to isolate the offset */
8590 while (nextbit
== 0 &&
8591 strchr(flagbitstr
, *p
) == NULL
) {
8594 if (i
+ j
> strlen(flagstr
))
8597 nextbit
= flagbits
[(uchar_t
)*p
];
8599 blkptr_offset
= strtoull(offstr
, &end
,
8602 } else if (nextbit
== 0) {
8603 (void) printf("***Ignoring flag arg:"
8604 " '%c'\n", (uchar_t
)*p
);
8609 if (blkptr_offset
% sizeof (blkptr_t
)) {
8610 printf("Block pointer offset 0x%llx "
8611 "must be divisible by 0x%x\n",
8612 (longlong_t
)blkptr_offset
, (int)sizeof (blkptr_t
));
8615 if (found
== B_FALSE
&& strlen(flagstr
) > 0) {
8616 printf("Invalid flag arg: '%s'\n", flagstr
);
8620 vd
= zdb_vdev_lookup(spa
->spa_root_vdev
, vdev
);
8622 (void) printf("***Invalid vdev: %s\n", vdev
);
8626 (void) fprintf(stderr
, "Found vdev: %s\n",
8629 (void) fprintf(stderr
, "Found vdev type: %s\n",
8630 vd
->vdev_ops
->vdev_op_type
);
8633 pabd
= abd_alloc_for_io(SPA_MAXBLOCKSIZE
, B_FALSE
);
8634 lbuf
= umem_alloc(SPA_MAXBLOCKSIZE
, UMEM_NOFAIL
);
8638 DVA_SET_VDEV(&dva
[0], vd
->vdev_id
);
8639 DVA_SET_OFFSET(&dva
[0], offset
);
8640 DVA_SET_GANG(&dva
[0], !!(flags
& ZDB_FLAG_GBH
));
8641 DVA_SET_ASIZE(&dva
[0], vdev_psize_to_asize(vd
, psize
));
8643 BP_SET_BIRTH(bp
, TXG_INITIAL
, TXG_INITIAL
);
8645 BP_SET_LSIZE(bp
, lsize
);
8646 BP_SET_PSIZE(bp
, psize
);
8647 BP_SET_COMPRESS(bp
, ZIO_COMPRESS_OFF
);
8648 BP_SET_CHECKSUM(bp
, ZIO_CHECKSUM_OFF
);
8649 BP_SET_TYPE(bp
, DMU_OT_NONE
);
8650 BP_SET_LEVEL(bp
, 0);
8651 BP_SET_DEDUP(bp
, 0);
8652 BP_SET_BYTEORDER(bp
, ZFS_HOST_BYTEORDER
);
8654 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8655 zio
= zio_root(spa
, NULL
, NULL
, 0);
8657 if (vd
== vd
->vdev_top
) {
8659 * Treat this as a normal block read.
8661 zio_nowait(zio_read(zio
, spa
, bp
, pabd
, psize
, NULL
, NULL
,
8662 ZIO_PRIORITY_SYNC_READ
,
8663 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
, NULL
));
8666 * Treat this as a vdev child I/O.
8668 zio_nowait(zio_vdev_child_io(zio
, bp
, vd
, offset
, pabd
,
8669 psize
, ZIO_TYPE_READ
, ZIO_PRIORITY_SYNC_READ
,
8670 ZIO_FLAG_DONT_PROPAGATE
| ZIO_FLAG_DONT_RETRY
|
8671 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
| ZIO_FLAG_OPTIONAL
,
8675 error
= zio_wait(zio
);
8676 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8679 (void) printf("Read of %s failed, error: %d\n", thing
, error
);
8683 uint64_t orig_lsize
= lsize
;
8685 if (flags
& ZDB_FLAG_DECOMPRESS
) {
8686 lsize
= zdb_decompress_block(pabd
, buf
, lbuf
,
8687 lsize
, psize
, flags
);
8689 (void) printf("Decompress of %s failed\n", thing
);
8693 buf
= abd_borrow_buf_copy(pabd
, lsize
);
8697 * Try to detect invalid block pointer. If invalid, try
8700 if ((flags
& ZDB_FLAG_PRINT_BLKPTR
|| flags
& ZDB_FLAG_INDIRECT
) &&
8701 !(flags
& ZDB_FLAG_DECOMPRESS
)) {
8702 const blkptr_t
*b
= (const blkptr_t
*)(void *)
8703 ((uintptr_t)buf
+ (uintptr_t)blkptr_offset
);
8704 if (zfs_blkptr_verify(spa
, b
,
8705 BLK_CONFIG_NEEDED
, BLK_VERIFY_ONLY
) == B_FALSE
) {
8706 abd_return_buf_copy(pabd
, buf
, lsize
);
8709 lsize
= zdb_decompress_block(pabd
, buf
,
8710 lbuf
, lsize
, psize
, flags
);
8711 b
= (const blkptr_t
*)(void *)
8712 ((uintptr_t)buf
+ (uintptr_t)blkptr_offset
);
8713 if (lsize
== -1 || zfs_blkptr_verify(spa
, b
,
8714 BLK_CONFIG_NEEDED
, BLK_VERIFY_LOG
) == B_FALSE
) {
8715 printf("invalid block pointer at this DVA\n");
8721 if (flags
& ZDB_FLAG_PRINT_BLKPTR
)
8722 zdb_print_blkptr((blkptr_t
*)(void *)
8723 ((uintptr_t)buf
+ (uintptr_t)blkptr_offset
), flags
);
8724 else if (flags
& ZDB_FLAG_RAW
)
8725 zdb_dump_block_raw(buf
, lsize
, flags
);
8726 else if (flags
& ZDB_FLAG_INDIRECT
)
8727 zdb_dump_indirect((blkptr_t
*)buf
,
8728 orig_lsize
/ sizeof (blkptr_t
), flags
);
8729 else if (flags
& ZDB_FLAG_GBH
)
8730 zdb_dump_gbh(buf
, flags
);
8732 zdb_dump_block(thing
, buf
, lsize
, flags
);
8735 * If :c was specified, iterate through the checksum table to
8736 * calculate and display each checksum for our specified
8739 if ((flags
& ZDB_FLAG_CHECKSUM
) && !(flags
& ZDB_FLAG_RAW
) &&
8740 !(flags
& ZDB_FLAG_GBH
)) {
8742 (void) printf("\n");
8743 for (enum zio_checksum ck
= ZIO_CHECKSUM_LABEL
;
8744 ck
< ZIO_CHECKSUM_FUNCTIONS
; ck
++) {
8746 if ((zio_checksum_table
[ck
].ci_flags
&
8747 ZCHECKSUM_FLAG_EMBEDDED
) ||
8748 ck
== ZIO_CHECKSUM_NOPARITY
) {
8751 BP_SET_CHECKSUM(bp
, ck
);
8752 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8753 czio
= zio_root(spa
, NULL
, NULL
, ZIO_FLAG_CANFAIL
);
8754 if (vd
== vd
->vdev_top
) {
8755 zio_nowait(zio_read(czio
, spa
, bp
, pabd
, psize
,
8757 ZIO_PRIORITY_SYNC_READ
,
8758 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
|
8759 ZIO_FLAG_DONT_RETRY
, NULL
));
8761 zio_nowait(zio_vdev_child_io(czio
, bp
, vd
,
8762 offset
, pabd
, psize
, ZIO_TYPE_READ
,
8763 ZIO_PRIORITY_SYNC_READ
,
8764 ZIO_FLAG_DONT_PROPAGATE
|
8765 ZIO_FLAG_DONT_RETRY
|
8766 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
|
8767 ZIO_FLAG_SPECULATIVE
|
8768 ZIO_FLAG_OPTIONAL
, NULL
, NULL
));
8770 error
= zio_wait(czio
);
8771 if (error
== 0 || error
== ECKSUM
) {
8772 zio_t
*ck_zio
= zio_null(NULL
, spa
, NULL
,
8775 DVA_GET_OFFSET(&bp
->blk_dva
[0]);
8777 zio_checksum_compute(ck_zio
, ck
, pabd
, lsize
);
8780 "cksum=%016llx:%016llx:%016llx:%016llx\n",
8781 zio_checksum_table
[ck
].ci_name
,
8782 (u_longlong_t
)bp
->blk_cksum
.zc_word
[0],
8783 (u_longlong_t
)bp
->blk_cksum
.zc_word
[1],
8784 (u_longlong_t
)bp
->blk_cksum
.zc_word
[2],
8785 (u_longlong_t
)bp
->blk_cksum
.zc_word
[3]);
8788 printf("error %d reading block\n", error
);
8790 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8795 abd_return_buf_copy(pabd
, buf
, lsize
);
8799 umem_free(lbuf
, SPA_MAXBLOCKSIZE
);
8806 zdb_embedded_block(char *thing
)
8808 blkptr_t bp
= {{{{0}}}};
8809 unsigned long long *words
= (void *)&bp
;
8813 err
= sscanf(thing
, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
8814 "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
8815 words
+ 0, words
+ 1, words
+ 2, words
+ 3,
8816 words
+ 4, words
+ 5, words
+ 6, words
+ 7,
8817 words
+ 8, words
+ 9, words
+ 10, words
+ 11,
8818 words
+ 12, words
+ 13, words
+ 14, words
+ 15);
8820 (void) fprintf(stderr
, "invalid input format\n");
8823 ASSERT3U(BPE_GET_LSIZE(&bp
), <=, SPA_MAXBLOCKSIZE
);
8824 buf
= malloc(SPA_MAXBLOCKSIZE
);
8826 (void) fprintf(stderr
, "out of memory\n");
8829 err
= decode_embedded_bp(&bp
, buf
, BPE_GET_LSIZE(&bp
));
8831 (void) fprintf(stderr
, "decode failed: %u\n", err
);
8834 zdb_dump_block_raw(buf
, BPE_GET_LSIZE(&bp
), 0);
8838 /* check for valid hex or decimal numeric string */
8840 zdb_numeric(char *str
)
8844 if (strlen(str
) == 0)
8846 if (strncmp(str
, "0x", 2) == 0 || strncmp(str
, "0X", 2) == 0)
8848 for (; i
< strlen(str
); i
++) {
8849 if (!isxdigit(str
[i
]))
8856 main(int argc
, char **argv
)
8860 objset_t
*os
= NULL
;
8864 char **searchdirs
= NULL
;
8866 char *target
, *target_pool
, dsname
[ZFS_MAX_DATASET_NAME_LEN
];
8867 nvlist_t
*policy
= NULL
;
8868 uint64_t max_txg
= UINT64_MAX
;
8869 int64_t objset_id
= -1;
8871 int flags
= ZFS_IMPORT_MISSING_LOG
;
8872 int rewind
= ZPOOL_NEVER_REWIND
;
8873 char *spa_config_path_env
, *objset_str
;
8874 boolean_t target_is_spa
= B_TRUE
, dataset_lookup
= B_FALSE
;
8875 nvlist_t
*cfg
= NULL
;
8877 dprintf_setup(&argc
, argv
);
8880 * If there is an environment variable SPA_CONFIG_PATH it overrides
8881 * default spa_config_path setting. If -U flag is specified it will
8882 * override this environment variable settings once again.
8884 spa_config_path_env
= getenv("SPA_CONFIG_PATH");
8885 if (spa_config_path_env
!= NULL
)
8886 spa_config_path
= spa_config_path_env
;
8889 * For performance reasons, we set this tunable down. We do so before
8890 * the arg parsing section so that the user can override this value if
8893 zfs_btree_verify_intensity
= 3;
8895 struct option long_options
[] = {
8896 {"ignore-assertions", no_argument
, NULL
, 'A'},
8897 {"block-stats", no_argument
, NULL
, 'b'},
8898 {"backup", no_argument
, NULL
, 'B'},
8899 {"checksum", no_argument
, NULL
, 'c'},
8900 {"config", no_argument
, NULL
, 'C'},
8901 {"datasets", no_argument
, NULL
, 'd'},
8902 {"dedup-stats", no_argument
, NULL
, 'D'},
8903 {"exported", no_argument
, NULL
, 'e'},
8904 {"embedded-block-pointer", no_argument
, NULL
, 'E'},
8905 {"automatic-rewind", no_argument
, NULL
, 'F'},
8906 {"dump-debug-msg", no_argument
, NULL
, 'G'},
8907 {"history", no_argument
, NULL
, 'h'},
8908 {"intent-logs", no_argument
, NULL
, 'i'},
8909 {"inflight", required_argument
, NULL
, 'I'},
8910 {"checkpointed-state", no_argument
, NULL
, 'k'},
8911 {"key", required_argument
, NULL
, 'K'},
8912 {"label", no_argument
, NULL
, 'l'},
8913 {"disable-leak-tracking", no_argument
, NULL
, 'L'},
8914 {"metaslabs", no_argument
, NULL
, 'm'},
8915 {"metaslab-groups", no_argument
, NULL
, 'M'},
8916 {"numeric", no_argument
, NULL
, 'N'},
8917 {"option", required_argument
, NULL
, 'o'},
8918 {"object-lookups", no_argument
, NULL
, 'O'},
8919 {"path", required_argument
, NULL
, 'p'},
8920 {"parseable", no_argument
, NULL
, 'P'},
8921 {"skip-label", no_argument
, NULL
, 'q'},
8922 {"copy-object", no_argument
, NULL
, 'r'},
8923 {"read-block", no_argument
, NULL
, 'R'},
8924 {"io-stats", no_argument
, NULL
, 's'},
8925 {"simulate-dedup", no_argument
, NULL
, 'S'},
8926 {"txg", required_argument
, NULL
, 't'},
8927 {"brt-stats", no_argument
, NULL
, 'T'},
8928 {"uberblock", no_argument
, NULL
, 'u'},
8929 {"cachefile", required_argument
, NULL
, 'U'},
8930 {"verbose", no_argument
, NULL
, 'v'},
8931 {"verbatim", no_argument
, NULL
, 'V'},
8932 {"dump-blocks", required_argument
, NULL
, 'x'},
8933 {"extreme-rewind", no_argument
, NULL
, 'X'},
8934 {"all-reconstruction", no_argument
, NULL
, 'Y'},
8935 {"livelist", no_argument
, NULL
, 'y'},
8936 {"zstd-headers", no_argument
, NULL
, 'Z'},
8940 while ((c
= getopt_long(argc
, argv
,
8941 "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
8942 long_options
, NULL
)) != -1) {
8981 zfs_reconstruct_indirect_combinations_max
= INT_MAX
;
8982 zfs_deadman_enabled
= 0;
8984 /* NB: Sort single match options below. */
8986 max_inflight_bytes
= strtoull(optarg
, NULL
, 0);
8987 if (max_inflight_bytes
== 0) {
8988 (void) fprintf(stderr
, "maximum number "
8989 "of inflight bytes must be greater "
8996 key_material
= strdup(optarg
);
8997 /* redact key material in process table */
8998 while (*optarg
!= '\0') { *optarg
++ = '*'; }
9001 error
= set_global_var(optarg
);
9006 if (searchdirs
== NULL
) {
9007 searchdirs
= umem_alloc(sizeof (char *),
9010 char **tmp
= umem_alloc((nsearch
+ 1) *
9011 sizeof (char *), UMEM_NOFAIL
);
9012 memcpy(tmp
, searchdirs
, nsearch
*
9014 umem_free(searchdirs
,
9015 nsearch
* sizeof (char *));
9018 searchdirs
[nsearch
++] = optarg
;
9021 max_txg
= strtoull(optarg
, NULL
, 0);
9022 if (max_txg
< TXG_INITIAL
) {
9023 (void) fprintf(stderr
, "incorrect txg "
9024 "specified: %s\n", optarg
);
9029 spa_config_path
= optarg
;
9030 if (spa_config_path
[0] != '/') {
9031 (void) fprintf(stderr
,
9032 "cachefile must be an absolute path "
9033 "(i.e. start with a slash)\n");
9041 flags
= ZFS_IMPORT_VERBATIM
;
9044 vn_dumpdir
= optarg
;
9052 if (!dump_opt
['e'] && searchdirs
!= NULL
) {
9053 (void) fprintf(stderr
, "-p option requires use of -e\n");
9058 * ZDB does not typically re-read blocks; therefore limit the ARC
9059 * to 256 MB, which can be used entirely for metadata.
9061 zfs_arc_min
= 2ULL << SPA_MAXBLOCKSHIFT
;
9062 zfs_arc_max
= 256 * 1024 * 1024;
9066 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
9067 * "zdb -b" uses traversal prefetch which uses async reads.
9068 * For good performance, let several of them be active at once.
9070 zfs_vdev_async_read_max_active
= 10;
9073 * Disable reference tracking for better performance.
9075 reference_tracking_enable
= B_FALSE
;
9078 * Do not fail spa_load when spa_load_verify fails. This is needed
9079 * to load non-idle pools.
9081 spa_load_verify_dryrun
= B_TRUE
;
9084 * ZDB should have ability to read spacemaps.
9086 spa_mode_readable_spacemaps
= B_TRUE
;
9088 kernel_init(SPA_MODE_READ
);
9091 verbose
= MAX(verbose
, 1);
9093 for (c
= 0; c
< 256; c
++) {
9094 if (dump_all
&& strchr("ABeEFkKlLNOPrRSXy", c
) == NULL
)
9097 dump_opt
[c
] += verbose
;
9100 libspl_set_assert_ok((dump_opt
['A'] == 1) || (dump_opt
['A'] > 2));
9101 zfs_recover
= (dump_opt
['A'] > 1);
9105 if (argc
< 2 && dump_opt
['R'])
9108 if (dump_opt
['E']) {
9111 zdb_embedded_block(argv
[0]);
9116 if (!dump_opt
['e'] && dump_opt
['C']) {
9117 dump_cachefile(spa_config_path
);
9124 return (dump_label(argv
[0]));
9126 if (dump_opt
['X'] || dump_opt
['F'])
9127 rewind
= ZPOOL_DO_REWIND
|
9128 (dump_opt
['X'] ? ZPOOL_EXTREME_REWIND
: 0);
9131 if (dump_opt
['N'] && dump_opt
['d'] == 0)
9132 dump_opt
['d'] = dump_opt
['N'];
9134 if (nvlist_alloc(&policy
, NV_UNIQUE_NAME_TYPE
, 0) != 0 ||
9135 nvlist_add_uint64(policy
, ZPOOL_LOAD_REQUEST_TXG
, max_txg
) != 0 ||
9136 nvlist_add_uint32(policy
, ZPOOL_LOAD_REWIND_POLICY
, rewind
) != 0)
9137 fatal("internal error: %s", strerror(ENOMEM
));
9142 if (strpbrk(target
, "/@") != NULL
) {
9145 target_pool
= strdup(target
);
9146 *strpbrk(target_pool
, "/@") = '\0';
9148 target_is_spa
= B_FALSE
;
9149 targetlen
= strlen(target
);
9150 if (targetlen
&& target
[targetlen
- 1] == '/')
9151 target
[targetlen
- 1] = '\0';
9154 * See if an objset ID was supplied (-d <pool>/<objset ID>).
9155 * To disambiguate tank/100, consider the 100 as objsetID
9156 * if -N was given, otherwise 100 is an objsetID iff
9157 * tank/100 as a named dataset fails on lookup.
9159 objset_str
= strchr(target
, '/');
9160 if (objset_str
&& strlen(objset_str
) > 1 &&
9161 zdb_numeric(objset_str
+ 1)) {
9165 objset_id
= strtoull(objset_str
, &endptr
, 0);
9166 /* dataset 0 is the same as opening the pool */
9167 if (errno
== 0 && endptr
!= objset_str
&&
9170 dataset_lookup
= B_TRUE
;
9172 /* normal dataset name not an objset ID */
9173 if (endptr
== objset_str
) {
9176 } else if (objset_str
&& !zdb_numeric(objset_str
+ 1) &&
9178 printf("Supply a numeric objset ID with -N\n");
9182 target_pool
= target
;
9185 if (dump_opt
['e']) {
9186 importargs_t args
= { 0 };
9188 args
.paths
= nsearch
;
9189 args
.path
= searchdirs
;
9190 args
.can_be_active
= B_TRUE
;
9192 libpc_handle_t lpch
= {
9193 .lpc_lib_handle
= NULL
,
9194 .lpc_ops
= &libzpool_config_ops
,
9195 .lpc_printerr
= B_TRUE
9197 error
= zpool_find_config(&lpch
, target_pool
, &cfg
, &args
);
9201 if (nvlist_add_nvlist(cfg
,
9202 ZPOOL_LOAD_POLICY
, policy
) != 0) {
9203 fatal("can't open '%s': %s",
9204 target
, strerror(ENOMEM
));
9207 if (dump_opt
['C'] > 1) {
9208 (void) printf("\nConfiguration for import:\n");
9209 dump_nvlist(cfg
, 8);
9213 * Disable the activity check to allow examination of
9216 error
= spa_import(target_pool
, cfg
, NULL
,
9217 flags
| ZFS_IMPORT_SKIP_MMP
);
9221 if (searchdirs
!= NULL
) {
9222 umem_free(searchdirs
, nsearch
* sizeof (char *));
9227 * We need to make sure to process -O option or call
9228 * dump_path after the -e option has been processed,
9229 * which imports the pool to the namespace if it's
9230 * not in the cachefile.
9232 if (dump_opt
['O']) {
9235 dump_opt
['v'] = verbose
+ 3;
9236 return (dump_path(argv
[0], argv
[1], NULL
));
9239 if (dump_opt
['r']) {
9240 target_is_spa
= B_FALSE
;
9243 dump_opt
['v'] = verbose
;
9244 error
= dump_path(argv
[0], argv
[1], &object
);
9246 fatal("internal error: %s", strerror(error
));
9250 * import_checkpointed_state makes the assumption that the
9251 * target pool that we pass it is already part of the spa
9252 * namespace. Because of that we need to make sure to call
9253 * it always after the -e option has been processed, which
9254 * imports the pool to the namespace if it's not in the
9257 char *checkpoint_pool
= NULL
;
9258 char *checkpoint_target
= NULL
;
9259 if (dump_opt
['k']) {
9260 checkpoint_pool
= import_checkpointed_state(target
, cfg
,
9261 &checkpoint_target
);
9263 if (checkpoint_target
!= NULL
)
9264 target
= checkpoint_target
;
9272 if (target_pool
!= target
)
9276 if (dump_opt
['k'] && (target_is_spa
|| dump_opt
['R'])) {
9277 ASSERT(checkpoint_pool
!= NULL
);
9278 ASSERT(checkpoint_target
== NULL
);
9280 error
= spa_open(checkpoint_pool
, &spa
, FTAG
);
9282 fatal("Tried to open pool \"%s\" but "
9283 "spa_open() failed with error %d\n",
9284 checkpoint_pool
, error
);
9287 } else if (target_is_spa
|| dump_opt
['R'] || dump_opt
['B'] ||
9289 zdb_set_skip_mmp(target
);
9290 error
= spa_open_rewind(target
, &spa
, FTAG
, policy
,
9294 * If we're missing the log device then
9295 * try opening the pool after clearing the
9298 mutex_enter(&spa_namespace_lock
);
9299 if ((spa
= spa_lookup(target
)) != NULL
&&
9300 spa
->spa_log_state
== SPA_LOG_MISSING
) {
9301 spa
->spa_log_state
= SPA_LOG_CLEAR
;
9304 mutex_exit(&spa_namespace_lock
);
9307 error
= spa_open_rewind(target
, &spa
,
9308 FTAG
, policy
, NULL
);
9311 } else if (strpbrk(target
, "#") != NULL
) {
9313 error
= dsl_pool_hold(target
, FTAG
, &dp
);
9315 fatal("can't dump '%s': %s", target
,
9318 error
= dump_bookmark(dp
, target
, B_TRUE
, verbose
> 1);
9319 dsl_pool_rele(dp
, FTAG
);
9321 fatal("can't dump '%s': %s", target
,
9326 target_pool
= strdup(target
);
9327 if (strpbrk(target
, "/@") != NULL
)
9328 *strpbrk(target_pool
, "/@") = '\0';
9330 zdb_set_skip_mmp(target
);
9332 * If -N was supplied, the user has indicated that
9333 * zdb -d <pool>/<objsetID> is in effect. Otherwise
9334 * we first assume that the dataset string is the
9335 * dataset name. If dmu_objset_hold fails with the
9336 * dataset string, and we have an objset_id, retry the
9337 * lookup with the objsetID.
9339 boolean_t retry
= B_TRUE
;
9341 if (dataset_lookup
== B_TRUE
) {
9343 * Use the supplied id to get the name
9346 error
= spa_open(target_pool
, &spa
, FTAG
);
9348 error
= name_from_objset_id(spa
,
9350 spa_close(spa
, FTAG
);
9356 if (objset_id
> 0 && retry
) {
9357 int err
= dmu_objset_hold(target
, FTAG
,
9360 dataset_lookup
= B_TRUE
;
9364 dmu_objset_rele(os
, FTAG
);
9367 error
= open_objset(target
, FTAG
, &os
);
9370 spa
= dmu_objset_spa(os
);
9374 nvlist_free(policy
);
9377 fatal("can't open '%s': %s", target
, strerror(error
));
9380 * Set the pool failure mode to panic in order to prevent the pool
9381 * from suspending. A suspended I/O will have no way to resume and
9382 * can prevent the zdb(8) command from terminating as expected.
9385 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
9389 if (dump_opt
['r']) {
9390 error
= zdb_copy_object(os
, object
, argv
[1]);
9391 } else if (!dump_opt
['R']) {
9392 flagbits
['d'] = ZOR_FLAG_DIRECTORY
;
9393 flagbits
['f'] = ZOR_FLAG_PLAIN_FILE
;
9394 flagbits
['m'] = ZOR_FLAG_SPACE_MAP
;
9395 flagbits
['z'] = ZOR_FLAG_ZAP
;
9396 flagbits
['A'] = ZOR_FLAG_ALL_TYPES
;
9398 if (argc
> 0 && dump_opt
['d']) {
9399 zopt_object_args
= argc
;
9400 zopt_object_ranges
= calloc(zopt_object_args
,
9401 sizeof (zopt_object_range_t
));
9402 for (unsigned i
= 0; i
< zopt_object_args
; i
++) {
9404 const char *msg
= NULL
;
9406 err
= parse_object_range(argv
[i
],
9407 &zopt_object_ranges
[i
], &msg
);
9409 fatal("Bad object or range: '%s': %s\n",
9410 argv
[i
], msg
?: "");
9412 } else if (argc
> 0 && dump_opt
['m']) {
9413 zopt_metaslab_args
= argc
;
9414 zopt_metaslab
= calloc(zopt_metaslab_args
,
9416 for (unsigned i
= 0; i
< zopt_metaslab_args
; i
++) {
9418 zopt_metaslab
[i
] = strtoull(argv
[i
], NULL
, 0);
9419 if (zopt_metaslab
[i
] == 0 && errno
!= 0)
9420 fatal("bad number %s: %s", argv
[i
],
9424 if (dump_opt
['B']) {
9425 dump_backup(target
, objset_id
,
9426 argc
> 0 ? argv
[0] : NULL
);
9427 } else if (os
!= NULL
) {
9429 } else if (zopt_object_args
> 0 && !dump_opt
['m']) {
9430 dump_objset(spa
->spa_meta_objset
);
9435 flagbits
['b'] = ZDB_FLAG_PRINT_BLKPTR
;
9436 flagbits
['c'] = ZDB_FLAG_CHECKSUM
;
9437 flagbits
['d'] = ZDB_FLAG_DECOMPRESS
;
9438 flagbits
['e'] = ZDB_FLAG_BSWAP
;
9439 flagbits
['g'] = ZDB_FLAG_GBH
;
9440 flagbits
['i'] = ZDB_FLAG_INDIRECT
;
9441 flagbits
['r'] = ZDB_FLAG_RAW
;
9442 flagbits
['v'] = ZDB_FLAG_VERBOSE
;
9444 for (int i
= 0; i
< argc
; i
++)
9445 zdb_read_block(argv
[i
], spa
);
9448 if (dump_opt
['k']) {
9449 free(checkpoint_pool
);
9451 free(checkpoint_target
);
9455 close_objset(os
, FTAG
);
9457 spa_close(spa
, FTAG
);
9460 fuid_table_destroy();
9462 dump_debug_buffer();