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>
43 #include <sys/zfs_context.h>
45 #include <sys/spa_impl.h>
48 #include <sys/fs/zfs.h>
49 #include <sys/zfs_znode.h>
50 #include <sys/zfs_sa.h>
52 #include <sys/sa_impl.h>
54 #include <sys/vdev_impl.h>
55 #include <sys/metaslab_impl.h>
56 #include <sys/dmu_objset.h>
57 #include <sys/dsl_dir.h>
58 #include <sys/dsl_dataset.h>
59 #include <sys/dsl_pool.h>
60 #include <sys/dsl_bookmark.h>
63 #include <sys/zil_impl.h>
65 #include <sys/resource.h>
66 #include <sys/dmu_send.h>
67 #include <sys/dmu_traverse.h>
68 #include <sys/zio_checksum.h>
69 #include <sys/zio_compress.h>
70 #include <sys/zfs_fuid.h>
72 #include <sys/arc_impl.h>
74 #include <sys/zfeature.h>
76 #include <sys/blkptr.h>
77 #include <sys/dsl_crypt.h>
78 #include <sys/dsl_scan.h>
79 #include <sys/btree.h>
80 #include <zfs_comutil.h>
81 #include <sys/zstd/zstd.h>
83 #include <libnvpair.h>
88 #define ZDB_COMPRESS_NAME(idx) ((idx) < ZIO_COMPRESS_FUNCTIONS ? \
89 zio_compress_table[(idx)].ci_name : "UNKNOWN")
90 #define ZDB_CHECKSUM_NAME(idx) ((idx) < ZIO_CHECKSUM_FUNCTIONS ? \
91 zio_checksum_table[(idx)].ci_name : "UNKNOWN")
92 #define ZDB_OT_TYPE(idx) ((idx) < DMU_OT_NUMTYPES ? (idx) : \
93 (idx) == DMU_OTN_ZAP_DATA || (idx) == DMU_OTN_ZAP_METADATA ? \
95 (idx) == DMU_OTN_UINT64_DATA || (idx) == DMU_OTN_UINT64_METADATA ? \
96 DMU_OT_UINT64_OTHER : DMU_OT_NUMTYPES)
98 /* Some platforms require part of inode IDs to be remapped */
100 #define ZDB_MAP_OBJECT_ID(obj) INO_XNUTOZFS(obj, 2)
102 #define ZDB_MAP_OBJECT_ID(obj) (obj)
106 zdb_ot_name(dmu_object_type_t type
)
108 if (type
< DMU_OT_NUMTYPES
)
109 return (dmu_ot
[type
].ot_name
);
110 else if ((type
& DMU_OT_NEWTYPE
) &&
111 ((type
& DMU_OT_BYTESWAP_MASK
) < DMU_BSWAP_NUMFUNCS
))
112 return (dmu_ot_byteswap
[type
& DMU_OT_BYTESWAP_MASK
].ob_name
);
117 extern int reference_tracking_enable
;
118 extern int zfs_recover
;
119 extern unsigned long zfs_arc_meta_min
, zfs_arc_meta_limit
;
120 extern int zfs_vdev_async_read_max_active
;
121 extern boolean_t spa_load_verify_dryrun
;
122 extern boolean_t spa_mode_readable_spacemaps
;
123 extern int zfs_reconstruct_indirect_combinations_max
;
124 extern int zfs_btree_verify_intensity
;
126 static const char cmdname
[] = "zdb";
127 uint8_t dump_opt
[256];
129 typedef void object_viewer_t(objset_t
*, uint64_t, void *data
, size_t size
);
131 uint64_t *zopt_metaslab
= NULL
;
132 static unsigned zopt_metaslab_args
= 0;
134 typedef struct zopt_object_range
{
135 uint64_t zor_obj_start
;
136 uint64_t zor_obj_end
;
138 } zopt_object_range_t
;
139 zopt_object_range_t
*zopt_object_ranges
= NULL
;
140 static unsigned zopt_object_args
= 0;
142 static int flagbits
[256];
144 #define ZOR_FLAG_PLAIN_FILE 0x0001
145 #define ZOR_FLAG_DIRECTORY 0x0002
146 #define ZOR_FLAG_SPACE_MAP 0x0004
147 #define ZOR_FLAG_ZAP 0x0008
148 #define ZOR_FLAG_ALL_TYPES -1
149 #define ZOR_SUPPORTED_FLAGS (ZOR_FLAG_PLAIN_FILE | \
150 ZOR_FLAG_DIRECTORY | \
151 ZOR_FLAG_SPACE_MAP | \
154 #define ZDB_FLAG_CHECKSUM 0x0001
155 #define ZDB_FLAG_DECOMPRESS 0x0002
156 #define ZDB_FLAG_BSWAP 0x0004
157 #define ZDB_FLAG_GBH 0x0008
158 #define ZDB_FLAG_INDIRECT 0x0010
159 #define ZDB_FLAG_RAW 0x0020
160 #define ZDB_FLAG_PRINT_BLKPTR 0x0040
161 #define ZDB_FLAG_VERBOSE 0x0080
163 uint64_t max_inflight_bytes
= 256 * 1024 * 1024; /* 256MB */
164 static int leaked_objects
= 0;
165 static range_tree_t
*mos_refd_objs
;
167 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t
*,
169 static void mos_obj_refd(uint64_t);
170 static void mos_obj_refd_multiple(uint64_t);
171 static int dump_bpobj_cb(void *arg
, const blkptr_t
*bp
, boolean_t free
,
174 typedef struct sublivelist_verify
{
175 /* FREE's that haven't yet matched to an ALLOC, in one sub-livelist */
178 /* ALLOC's without a matching FREE, accumulates across sub-livelists */
179 zfs_btree_t sv_leftover
;
180 } sublivelist_verify_t
;
183 livelist_compare(const void *larg
, const void *rarg
)
185 const blkptr_t
*l
= larg
;
186 const blkptr_t
*r
= rarg
;
188 /* Sort them according to dva[0] */
189 uint64_t l_dva0_vdev
, r_dva0_vdev
;
190 l_dva0_vdev
= DVA_GET_VDEV(&l
->blk_dva
[0]);
191 r_dva0_vdev
= DVA_GET_VDEV(&r
->blk_dva
[0]);
192 if (l_dva0_vdev
< r_dva0_vdev
)
194 else if (l_dva0_vdev
> r_dva0_vdev
)
197 /* if vdevs are equal, sort by offsets. */
198 uint64_t l_dva0_offset
;
199 uint64_t r_dva0_offset
;
200 l_dva0_offset
= DVA_GET_OFFSET(&l
->blk_dva
[0]);
201 r_dva0_offset
= DVA_GET_OFFSET(&r
->blk_dva
[0]);
202 if (l_dva0_offset
< r_dva0_offset
) {
204 } else if (l_dva0_offset
> r_dva0_offset
) {
209 * Since we're storing blkptrs without cancelling FREE/ALLOC pairs,
210 * it's possible the offsets are equal. In that case, sort by txg
212 if (l
->blk_birth
< r
->blk_birth
) {
214 } else if (l
->blk_birth
> r
->blk_birth
) {
220 typedef struct sublivelist_verify_block
{
224 * We need this to check if the block marked as allocated
225 * in the livelist was freed (and potentially reallocated)
226 * in the metaslab spacemaps at a later TXG.
228 uint64_t svb_allocated_txg
;
229 } sublivelist_verify_block_t
;
231 static void zdb_print_blkptr(const blkptr_t
*bp
, int flags
);
233 typedef struct sublivelist_verify_block_refcnt
{
234 /* block pointer entry in livelist being verified */
238 * Refcount gets incremented to 1 when we encounter the first
239 * FREE entry for the svfbr block pointer and a node for it
240 * is created in our ZDB verification/tracking metadata.
242 * As we encounter more FREE entries we increment this counter
243 * and similarly decrement it whenever we find the respective
244 * ALLOC entries for this block.
246 * When the refcount gets to 0 it means that all the FREE and
247 * ALLOC entries of this block have paired up and we no longer
248 * need to track it in our verification logic (e.g. the node
249 * containing this struct in our verification data structure
252 * [refer to sublivelist_verify_blkptr() for the actual code]
254 uint32_t svbr_refcnt
;
255 } sublivelist_verify_block_refcnt_t
;
258 sublivelist_block_refcnt_compare(const void *larg
, const void *rarg
)
260 const sublivelist_verify_block_refcnt_t
*l
= larg
;
261 const sublivelist_verify_block_refcnt_t
*r
= rarg
;
262 return (livelist_compare(&l
->svbr_blk
, &r
->svbr_blk
));
266 sublivelist_verify_blkptr(void *arg
, const blkptr_t
*bp
, boolean_t free
,
269 ASSERT3P(tx
, ==, NULL
);
270 struct sublivelist_verify
*sv
= arg
;
271 sublivelist_verify_block_refcnt_t current
= {
275 * Start with 1 in case this is the first free entry.
276 * This field is not used for our B-Tree comparisons
282 zfs_btree_index_t where
;
283 sublivelist_verify_block_refcnt_t
*pair
=
284 zfs_btree_find(&sv
->sv_pair
, ¤t
, &where
);
287 /* first free entry for this block pointer */
288 zfs_btree_add(&sv
->sv_pair
, ¤t
);
294 /* block that is currently marked as allocated */
295 for (int i
= 0; i
< SPA_DVAS_PER_BP
; i
++) {
296 if (DVA_IS_EMPTY(&bp
->blk_dva
[i
]))
298 sublivelist_verify_block_t svb
= {
299 .svb_dva
= bp
->blk_dva
[i
],
300 .svb_allocated_txg
= bp
->blk_birth
303 if (zfs_btree_find(&sv
->sv_leftover
, &svb
,
305 zfs_btree_add_idx(&sv
->sv_leftover
,
310 /* alloc matches a free entry */
312 if (pair
->svbr_refcnt
== 0) {
313 /* all allocs and frees have been matched */
314 zfs_btree_remove_idx(&sv
->sv_pair
, &where
);
323 sublivelist_verify_func(void *args
, dsl_deadlist_entry_t
*dle
)
326 struct sublivelist_verify
*sv
= args
;
328 zfs_btree_create(&sv
->sv_pair
, sublivelist_block_refcnt_compare
,
329 sizeof (sublivelist_verify_block_refcnt_t
));
331 err
= bpobj_iterate_nofree(&dle
->dle_bpobj
, sublivelist_verify_blkptr
,
334 sublivelist_verify_block_refcnt_t
*e
;
335 zfs_btree_index_t
*cookie
= NULL
;
336 while ((e
= zfs_btree_destroy_nodes(&sv
->sv_pair
, &cookie
)) != NULL
) {
337 char blkbuf
[BP_SPRINTF_LEN
];
338 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
),
339 &e
->svbr_blk
, B_TRUE
);
340 (void) printf("\tERROR: %d unmatched FREE(s): %s\n",
341 e
->svbr_refcnt
, blkbuf
);
343 zfs_btree_destroy(&sv
->sv_pair
);
349 livelist_block_compare(const void *larg
, const void *rarg
)
351 const sublivelist_verify_block_t
*l
= larg
;
352 const sublivelist_verify_block_t
*r
= rarg
;
354 if (DVA_GET_VDEV(&l
->svb_dva
) < DVA_GET_VDEV(&r
->svb_dva
))
356 else if (DVA_GET_VDEV(&l
->svb_dva
) > DVA_GET_VDEV(&r
->svb_dva
))
359 if (DVA_GET_OFFSET(&l
->svb_dva
) < DVA_GET_OFFSET(&r
->svb_dva
))
361 else if (DVA_GET_OFFSET(&l
->svb_dva
) > DVA_GET_OFFSET(&r
->svb_dva
))
364 if (DVA_GET_ASIZE(&l
->svb_dva
) < DVA_GET_ASIZE(&r
->svb_dva
))
366 else if (DVA_GET_ASIZE(&l
->svb_dva
) > DVA_GET_ASIZE(&r
->svb_dva
))
373 * Check for errors in a livelist while tracking all unfreed ALLOCs in the
374 * sublivelist_verify_t: sv->sv_leftover
377 livelist_verify(dsl_deadlist_t
*dl
, void *arg
)
379 sublivelist_verify_t
*sv
= arg
;
380 dsl_deadlist_iterate(dl
, sublivelist_verify_func
, sv
);
384 * Check for errors in the livelist entry and discard the intermediary
388 sublivelist_verify_lightweight(void *args
, dsl_deadlist_entry_t
*dle
)
391 sublivelist_verify_t sv
;
392 zfs_btree_create(&sv
.sv_leftover
, livelist_block_compare
,
393 sizeof (sublivelist_verify_block_t
));
394 int err
= sublivelist_verify_func(&sv
, dle
);
395 zfs_btree_clear(&sv
.sv_leftover
);
396 zfs_btree_destroy(&sv
.sv_leftover
);
400 typedef struct metaslab_verify
{
402 * Tree containing all the leftover ALLOCs from the livelists
403 * that are part of this metaslab.
405 zfs_btree_t mv_livelist_allocs
;
408 * Metaslab information.
416 * What's currently allocated for this metaslab.
418 range_tree_t
*mv_allocated
;
421 typedef void ll_iter_t(dsl_deadlist_t
*ll
, void *arg
);
423 typedef int (*zdb_log_sm_cb_t
)(spa_t
*spa
, space_map_entry_t
*sme
, uint64_t txg
,
426 typedef struct unflushed_iter_cb_arg
{
430 zdb_log_sm_cb_t uic_cb
;
431 } unflushed_iter_cb_arg_t
;
434 iterate_through_spacemap_logs_cb(space_map_entry_t
*sme
, void *arg
)
436 unflushed_iter_cb_arg_t
*uic
= arg
;
437 return (uic
->uic_cb(uic
->uic_spa
, sme
, uic
->uic_txg
, uic
->uic_arg
));
441 iterate_through_spacemap_logs(spa_t
*spa
, zdb_log_sm_cb_t cb
, void *arg
)
443 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
446 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
447 for (spa_log_sm_t
*sls
= avl_first(&spa
->spa_sm_logs_by_txg
);
448 sls
; sls
= AVL_NEXT(&spa
->spa_sm_logs_by_txg
, sls
)) {
449 space_map_t
*sm
= NULL
;
450 VERIFY0(space_map_open(&sm
, spa_meta_objset(spa
),
451 sls
->sls_sm_obj
, 0, UINT64_MAX
, SPA_MINBLOCKSHIFT
));
453 unflushed_iter_cb_arg_t uic
= {
455 .uic_txg
= sls
->sls_txg
,
459 VERIFY0(space_map_iterate(sm
, space_map_length(sm
),
460 iterate_through_spacemap_logs_cb
, &uic
));
463 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
467 verify_livelist_allocs(metaslab_verify_t
*mv
, uint64_t txg
,
468 uint64_t offset
, uint64_t size
)
470 sublivelist_verify_block_t svb
;
471 DVA_SET_VDEV(&svb
.svb_dva
, mv
->mv_vdid
);
472 DVA_SET_OFFSET(&svb
.svb_dva
, offset
);
473 DVA_SET_ASIZE(&svb
.svb_dva
, size
);
474 zfs_btree_index_t where
;
475 uint64_t end_offset
= offset
+ size
;
478 * Look for an exact match for spacemap entry in the livelist entries.
479 * Then, look for other livelist entries that fall within the range
480 * of the spacemap entry as it may have been condensed
482 sublivelist_verify_block_t
*found
=
483 zfs_btree_find(&mv
->mv_livelist_allocs
, &svb
, &where
);
485 found
= zfs_btree_next(&mv
->mv_livelist_allocs
, &where
, &where
);
487 for (; found
!= NULL
&& DVA_GET_VDEV(&found
->svb_dva
) == mv
->mv_vdid
&&
488 DVA_GET_OFFSET(&found
->svb_dva
) < end_offset
;
489 found
= zfs_btree_next(&mv
->mv_livelist_allocs
, &where
, &where
)) {
490 if (found
->svb_allocated_txg
<= txg
) {
491 (void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
492 "from TXG %llx FREED at TXG %llx\n",
493 (u_longlong_t
)DVA_GET_OFFSET(&found
->svb_dva
),
494 (u_longlong_t
)DVA_GET_ASIZE(&found
->svb_dva
),
495 (u_longlong_t
)found
->svb_allocated_txg
,
502 metaslab_spacemap_validation_cb(space_map_entry_t
*sme
, void *arg
)
504 metaslab_verify_t
*mv
= arg
;
505 uint64_t offset
= sme
->sme_offset
;
506 uint64_t size
= sme
->sme_run
;
507 uint64_t txg
= sme
->sme_txg
;
509 if (sme
->sme_type
== SM_ALLOC
) {
510 if (range_tree_contains(mv
->mv_allocated
,
512 (void) printf("ERROR: DOUBLE ALLOC: "
514 "%llu:%llu LOG_SM\n",
515 (u_longlong_t
)txg
, (u_longlong_t
)offset
,
516 (u_longlong_t
)size
, (u_longlong_t
)mv
->mv_vdid
,
517 (u_longlong_t
)mv
->mv_msid
);
519 range_tree_add(mv
->mv_allocated
,
523 if (!range_tree_contains(mv
->mv_allocated
,
525 (void) printf("ERROR: DOUBLE FREE: "
527 "%llu:%llu LOG_SM\n",
528 (u_longlong_t
)txg
, (u_longlong_t
)offset
,
529 (u_longlong_t
)size
, (u_longlong_t
)mv
->mv_vdid
,
530 (u_longlong_t
)mv
->mv_msid
);
532 range_tree_remove(mv
->mv_allocated
,
537 if (sme
->sme_type
!= SM_ALLOC
) {
539 * If something is freed in the spacemap, verify that
540 * it is not listed as allocated in the livelist.
542 verify_livelist_allocs(mv
, txg
, offset
, size
);
548 spacemap_check_sm_log_cb(spa_t
*spa
, space_map_entry_t
*sme
,
549 uint64_t txg
, void *arg
)
551 metaslab_verify_t
*mv
= arg
;
552 uint64_t offset
= sme
->sme_offset
;
553 uint64_t vdev_id
= sme
->sme_vdev
;
555 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
557 /* skip indirect vdevs */
558 if (!vdev_is_concrete(vd
))
561 if (vdev_id
!= mv
->mv_vdid
)
564 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
565 if (ms
->ms_id
!= mv
->mv_msid
)
568 if (txg
< metaslab_unflushed_txg(ms
))
572 ASSERT3U(txg
, ==, sme
->sme_txg
);
573 return (metaslab_spacemap_validation_cb(sme
, mv
));
577 spacemap_check_sm_log(spa_t
*spa
, metaslab_verify_t
*mv
)
579 iterate_through_spacemap_logs(spa
, spacemap_check_sm_log_cb
, mv
);
583 spacemap_check_ms_sm(space_map_t
*sm
, metaslab_verify_t
*mv
)
588 VERIFY0(space_map_iterate(sm
, space_map_length(sm
),
589 metaslab_spacemap_validation_cb
, mv
));
592 static void iterate_deleted_livelists(spa_t
*spa
, ll_iter_t func
, void *arg
);
595 * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
596 * they are part of that metaslab (mv_msid).
599 mv_populate_livelist_allocs(metaslab_verify_t
*mv
, sublivelist_verify_t
*sv
)
601 zfs_btree_index_t where
;
602 sublivelist_verify_block_t
*svb
;
603 ASSERT3U(zfs_btree_numnodes(&mv
->mv_livelist_allocs
), ==, 0);
604 for (svb
= zfs_btree_first(&sv
->sv_leftover
, &where
);
606 svb
= zfs_btree_next(&sv
->sv_leftover
, &where
, &where
)) {
607 if (DVA_GET_VDEV(&svb
->svb_dva
) != mv
->mv_vdid
)
610 if (DVA_GET_OFFSET(&svb
->svb_dva
) < mv
->mv_start
&&
611 (DVA_GET_OFFSET(&svb
->svb_dva
) +
612 DVA_GET_ASIZE(&svb
->svb_dva
)) > mv
->mv_start
) {
613 (void) printf("ERROR: Found block that crosses "
614 "metaslab boundary: <%llu:%llx:%llx>\n",
615 (u_longlong_t
)DVA_GET_VDEV(&svb
->svb_dva
),
616 (u_longlong_t
)DVA_GET_OFFSET(&svb
->svb_dva
),
617 (u_longlong_t
)DVA_GET_ASIZE(&svb
->svb_dva
));
621 if (DVA_GET_OFFSET(&svb
->svb_dva
) < mv
->mv_start
)
624 if (DVA_GET_OFFSET(&svb
->svb_dva
) >= mv
->mv_end
)
627 if ((DVA_GET_OFFSET(&svb
->svb_dva
) +
628 DVA_GET_ASIZE(&svb
->svb_dva
)) > mv
->mv_end
) {
629 (void) printf("ERROR: Found block that crosses "
630 "metaslab boundary: <%llu:%llx:%llx>\n",
631 (u_longlong_t
)DVA_GET_VDEV(&svb
->svb_dva
),
632 (u_longlong_t
)DVA_GET_OFFSET(&svb
->svb_dva
),
633 (u_longlong_t
)DVA_GET_ASIZE(&svb
->svb_dva
));
637 zfs_btree_add(&mv
->mv_livelist_allocs
, svb
);
640 for (svb
= zfs_btree_first(&mv
->mv_livelist_allocs
, &where
);
642 svb
= zfs_btree_next(&mv
->mv_livelist_allocs
, &where
, &where
)) {
643 zfs_btree_remove(&sv
->sv_leftover
, svb
);
649 * Iterate through all the sublivelists and:
650 * - report leftover frees (**)
651 * - record leftover ALLOCs together with their TXG [see Cross Check]
653 * (**) Note: Double ALLOCs are valid in datasets that have dedup
654 * enabled. Similarly double FREEs are allowed as well but
655 * only if they pair up with a corresponding ALLOC entry once
656 * we our done with our sublivelist iteration.
660 * - iterate over spacemap and then the metaslab's entries in the
661 * spacemap log, then report any double FREEs and ALLOCs (do not
665 * After finishing the Livelist Check phase and while being in the
666 * Spacemap Check phase, we find all the recorded leftover ALLOCs
667 * of the livelist check that are part of the metaslab that we are
668 * currently looking at in the Spacemap Check. We report any entries
669 * that are marked as ALLOCs in the livelists but have been actually
670 * freed (and potentially allocated again) after their TXG stamp in
671 * the spacemaps. Also report any ALLOCs from the livelists that
672 * belong to indirect vdevs (e.g. their vdev completed removal).
674 * Note that this will miss Log Spacemap entries that cancelled each other
675 * out before being flushed to the metaslab, so we are not guaranteed
676 * to match all erroneous ALLOCs.
679 livelist_metaslab_validate(spa_t
*spa
)
681 (void) printf("Verifying deleted livelist entries\n");
683 sublivelist_verify_t sv
;
684 zfs_btree_create(&sv
.sv_leftover
, livelist_block_compare
,
685 sizeof (sublivelist_verify_block_t
));
686 iterate_deleted_livelists(spa
, livelist_verify
, &sv
);
688 (void) printf("Verifying metaslab entries\n");
689 vdev_t
*rvd
= spa
->spa_root_vdev
;
690 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
691 vdev_t
*vd
= rvd
->vdev_child
[c
];
693 if (!vdev_is_concrete(vd
))
696 for (uint64_t mid
= 0; mid
< vd
->vdev_ms_count
; mid
++) {
697 metaslab_t
*m
= vd
->vdev_ms
[mid
];
699 (void) fprintf(stderr
,
700 "\rverifying concrete vdev %llu, "
701 "metaslab %llu of %llu ...",
702 (longlong_t
)vd
->vdev_id
,
704 (longlong_t
)vd
->vdev_ms_count
);
706 uint64_t shift
, start
;
707 range_seg_type_t type
=
708 metaslab_calculate_range_tree_type(vd
, m
,
710 metaslab_verify_t mv
;
711 mv
.mv_allocated
= range_tree_create(NULL
,
712 type
, NULL
, start
, shift
);
713 mv
.mv_vdid
= vd
->vdev_id
;
714 mv
.mv_msid
= m
->ms_id
;
715 mv
.mv_start
= m
->ms_start
;
716 mv
.mv_end
= m
->ms_start
+ m
->ms_size
;
717 zfs_btree_create(&mv
.mv_livelist_allocs
,
718 livelist_block_compare
,
719 sizeof (sublivelist_verify_block_t
));
721 mv_populate_livelist_allocs(&mv
, &sv
);
723 spacemap_check_ms_sm(m
->ms_sm
, &mv
);
724 spacemap_check_sm_log(spa
, &mv
);
726 range_tree_vacate(mv
.mv_allocated
, NULL
, NULL
);
727 range_tree_destroy(mv
.mv_allocated
);
728 zfs_btree_clear(&mv
.mv_livelist_allocs
);
729 zfs_btree_destroy(&mv
.mv_livelist_allocs
);
732 (void) fprintf(stderr
, "\n");
735 * If there are any segments in the leftover tree after we walked
736 * through all the metaslabs in the concrete vdevs then this means
737 * that we have segments in the livelists that belong to indirect
738 * vdevs and are marked as allocated.
740 if (zfs_btree_numnodes(&sv
.sv_leftover
) == 0) {
741 zfs_btree_destroy(&sv
.sv_leftover
);
744 (void) printf("ERROR: Found livelist blocks marked as allocated "
745 "for indirect vdevs:\n");
747 zfs_btree_index_t
*where
= NULL
;
748 sublivelist_verify_block_t
*svb
;
749 while ((svb
= zfs_btree_destroy_nodes(&sv
.sv_leftover
, &where
)) !=
751 int vdev_id
= DVA_GET_VDEV(&svb
->svb_dva
);
752 ASSERT3U(vdev_id
, <, rvd
->vdev_children
);
753 vdev_t
*vd
= rvd
->vdev_child
[vdev_id
];
754 ASSERT(!vdev_is_concrete(vd
));
755 (void) printf("<%d:%llx:%llx> TXG %llx\n",
756 vdev_id
, (u_longlong_t
)DVA_GET_OFFSET(&svb
->svb_dva
),
757 (u_longlong_t
)DVA_GET_ASIZE(&svb
->svb_dva
),
758 (u_longlong_t
)svb
->svb_allocated_txg
);
761 zfs_btree_destroy(&sv
.sv_leftover
);
765 * These libumem hooks provide a reasonable set of defaults for the allocator's
766 * debugging facilities.
769 _umem_debug_init(void)
771 return ("default,verbose"); /* $UMEM_DEBUG setting */
775 _umem_logging_init(void)
777 return ("fail,contents"); /* $UMEM_LOGGING setting */
783 (void) fprintf(stderr
,
784 "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
785 "[-I <inflight I/Os>]\n"
786 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
787 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
788 "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
789 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
790 "\t%s [-v] <bookmark>\n"
791 "\t%s -C [-A] [-U <cache>]\n"
792 "\t%s -l [-Aqu] <device>\n"
793 "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
794 "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
795 "\t%s -O <dataset> <path>\n"
796 "\t%s -r <dataset> <path> <destination>\n"
797 "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
798 "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
799 "\t%s -E [-A] word0:word1:...:word15\n"
800 "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
802 cmdname
, cmdname
, cmdname
, cmdname
, cmdname
, cmdname
, cmdname
,
803 cmdname
, cmdname
, cmdname
, cmdname
);
805 (void) fprintf(stderr
, " Dataset name must include at least one "
806 "separator character '/' or '@'\n");
807 (void) fprintf(stderr
, " If dataset name is specified, only that "
808 "dataset is dumped\n");
809 (void) fprintf(stderr
, " If object numbers or object number "
810 "ranges are specified, only those\n"
811 " objects or ranges are dumped.\n\n");
812 (void) fprintf(stderr
,
813 " Object ranges take the form <start>:<end>[:<flags>]\n"
814 " start Starting object number\n"
815 " end Ending object number, or -1 for no upper bound\n"
816 " flags Optional flags to select object types:\n"
817 " A All objects (this is the default)\n"
818 " d ZFS directories\n"
820 " m SPA space maps\n"
822 " - Negate effect of next flag\n\n");
823 (void) fprintf(stderr
, " Options to control amount of output:\n");
824 (void) fprintf(stderr
, " -b --block-stats "
825 "block statistics\n");
826 (void) fprintf(stderr
, " -c --checksum "
827 "checksum all metadata (twice for all data) blocks\n");
828 (void) fprintf(stderr
, " -C --config "
829 "config (or cachefile if alone)\n");
830 (void) fprintf(stderr
, " -d --datasets "
832 (void) fprintf(stderr
, " -D --dedup-stats "
833 "dedup statistics\n");
834 (void) fprintf(stderr
, " -E --embedded-block-pointer=INTEGER\n"
835 " decode and display block "
836 "from an embedded block pointer\n");
837 (void) fprintf(stderr
, " -h --history "
839 (void) fprintf(stderr
, " -i --intent-logs "
841 (void) fprintf(stderr
, " -l --label "
842 "read label contents\n");
843 (void) fprintf(stderr
, " -k --checkpointed-state "
844 "examine the checkpointed state of the pool\n");
845 (void) fprintf(stderr
, " -L --disable-leak-tracking "
846 "disable leak tracking (do not load spacemaps)\n");
847 (void) fprintf(stderr
, " -m --metaslabs "
849 (void) fprintf(stderr
, " -M --metaslab-groups "
850 "metaslab groups\n");
851 (void) fprintf(stderr
, " -O --object-lookups "
852 "perform object lookups by path\n");
853 (void) fprintf(stderr
, " -r --copy-object "
854 "copy an object by path to file\n");
855 (void) fprintf(stderr
, " -R --read-block "
856 "read and display block from a device\n");
857 (void) fprintf(stderr
, " -s --io-stats "
858 "report stats on zdb's I/O\n");
859 (void) fprintf(stderr
, " -S --simulate-dedup "
860 "simulate dedup to measure effect\n");
861 (void) fprintf(stderr
, " -v --verbose "
862 "verbose (applies to all others)\n");
863 (void) fprintf(stderr
, " -y --livelist "
864 "perform livelist and metaslab validation on any livelists being "
866 (void) fprintf(stderr
, " Below options are intended for use "
867 "with other options:\n");
868 (void) fprintf(stderr
, " -A --ignore-assertions "
869 "ignore assertions (-A), enable panic recovery (-AA) or both "
871 (void) fprintf(stderr
, " -e --exported "
872 "pool is exported/destroyed/has altroot/not in a cachefile\n");
873 (void) fprintf(stderr
, " -F --automatic-rewind "
874 "attempt automatic rewind within safe range of transaction "
876 (void) fprintf(stderr
, " -G --dump-debug-msg "
877 "dump zfs_dbgmsg buffer before exiting\n");
878 (void) fprintf(stderr
, " -I --inflight=INTEGER "
879 "specify the maximum number of checksumming I/Os "
880 "[default is 200]\n");
881 (void) fprintf(stderr
, " -o --option=\"OPTION=INTEGER\" "
882 "set global variable to an unsigned 32-bit integer\n");
883 (void) fprintf(stderr
, " -p --path==PATH "
884 "use one or more with -e to specify path to vdev dir\n");
885 (void) fprintf(stderr
, " -P --parseable "
886 "print numbers in parseable form\n");
887 (void) fprintf(stderr
, " -q --skip-label "
888 "don't print label contents\n");
889 (void) fprintf(stderr
, " -t --txg=INTEGER "
890 "highest txg to use when searching for uberblocks\n");
891 (void) fprintf(stderr
, " -u --uberblock "
893 (void) fprintf(stderr
, " -U --cachefile=PATH "
894 "use alternate cachefile\n");
895 (void) fprintf(stderr
, " -V --verbatim "
896 "do verbatim import\n");
897 (void) fprintf(stderr
, " -x --dump-blocks=PATH "
898 "dump all read blocks into specified directory\n");
899 (void) fprintf(stderr
, " -X --extreme-rewind "
900 "attempt extreme rewind (does not work with dataset)\n");
901 (void) fprintf(stderr
, " -Y --all-reconstruction "
902 "attempt all reconstruction combinations for split blocks\n");
903 (void) fprintf(stderr
, " -Z --zstd-headers "
904 "show ZSTD headers \n");
905 (void) fprintf(stderr
, "Specify an option more than once (e.g. -bb) "
906 "to make only that option verbose\n");
907 (void) fprintf(stderr
, "Default is to dump everything non-verbosely\n");
912 dump_debug_buffer(void)
916 (void) fflush(stdout
);
917 zfs_dbgmsg_print("zdb");
922 * Called for usage errors that are discovered after a call to spa_open(),
923 * dmu_bonus_hold(), or pool_match(). abort() is called for other errors.
927 fatal(const char *fmt
, ...)
932 (void) fprintf(stderr
, "%s: ", cmdname
);
933 (void) vfprintf(stderr
, fmt
, ap
);
935 (void) fprintf(stderr
, "\n");
943 dump_packed_nvlist(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
947 size_t nvsize
= *(uint64_t *)data
;
948 char *packed
= umem_alloc(nvsize
, UMEM_NOFAIL
);
950 VERIFY(0 == dmu_read(os
, object
, 0, nvsize
, packed
, DMU_READ_PREFETCH
));
952 VERIFY(nvlist_unpack(packed
, nvsize
, &nv
, 0) == 0);
954 umem_free(packed
, nvsize
);
962 dump_history_offsets(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
964 (void) os
, (void) object
, (void) size
;
965 spa_history_phys_t
*shp
= data
;
970 (void) printf("\t\tpool_create_len = %llu\n",
971 (u_longlong_t
)shp
->sh_pool_create_len
);
972 (void) printf("\t\tphys_max_off = %llu\n",
973 (u_longlong_t
)shp
->sh_phys_max_off
);
974 (void) printf("\t\tbof = %llu\n",
975 (u_longlong_t
)shp
->sh_bof
);
976 (void) printf("\t\teof = %llu\n",
977 (u_longlong_t
)shp
->sh_eof
);
978 (void) printf("\t\trecords_lost = %llu\n",
979 (u_longlong_t
)shp
->sh_records_lost
);
983 zdb_nicenum(uint64_t num
, char *buf
, size_t buflen
)
986 (void) snprintf(buf
, buflen
, "%llu", (longlong_t
)num
);
988 nicenum(num
, buf
, sizeof (buf
));
991 static const char histo_stars
[] = "****************************************";
992 static const uint64_t histo_width
= sizeof (histo_stars
) - 1;
995 dump_histogram(const uint64_t *histo
, int size
, int offset
)
998 int minidx
= size
- 1;
1002 for (i
= 0; i
< size
; i
++) {
1005 if (histo
[i
] > 0 && i
> maxidx
)
1007 if (histo
[i
] > 0 && i
< minidx
)
1011 if (max
< histo_width
)
1014 for (i
= minidx
; i
<= maxidx
; i
++) {
1015 (void) printf("\t\t\t%3u: %6llu %s\n",
1016 i
+ offset
, (u_longlong_t
)histo
[i
],
1017 &histo_stars
[(max
- histo
[i
]) * histo_width
/ max
]);
1022 dump_zap_stats(objset_t
*os
, uint64_t object
)
1027 error
= zap_get_stats(os
, object
, &zs
);
1031 if (zs
.zs_ptrtbl_len
== 0) {
1032 ASSERT(zs
.zs_num_blocks
== 1);
1033 (void) printf("\tmicrozap: %llu bytes, %llu entries\n",
1034 (u_longlong_t
)zs
.zs_blocksize
,
1035 (u_longlong_t
)zs
.zs_num_entries
);
1039 (void) printf("\tFat ZAP stats:\n");
1041 (void) printf("\t\tPointer table:\n");
1042 (void) printf("\t\t\t%llu elements\n",
1043 (u_longlong_t
)zs
.zs_ptrtbl_len
);
1044 (void) printf("\t\t\tzt_blk: %llu\n",
1045 (u_longlong_t
)zs
.zs_ptrtbl_zt_blk
);
1046 (void) printf("\t\t\tzt_numblks: %llu\n",
1047 (u_longlong_t
)zs
.zs_ptrtbl_zt_numblks
);
1048 (void) printf("\t\t\tzt_shift: %llu\n",
1049 (u_longlong_t
)zs
.zs_ptrtbl_zt_shift
);
1050 (void) printf("\t\t\tzt_blks_copied: %llu\n",
1051 (u_longlong_t
)zs
.zs_ptrtbl_blks_copied
);
1052 (void) printf("\t\t\tzt_nextblk: %llu\n",
1053 (u_longlong_t
)zs
.zs_ptrtbl_nextblk
);
1055 (void) printf("\t\tZAP entries: %llu\n",
1056 (u_longlong_t
)zs
.zs_num_entries
);
1057 (void) printf("\t\tLeaf blocks: %llu\n",
1058 (u_longlong_t
)zs
.zs_num_leafs
);
1059 (void) printf("\t\tTotal blocks: %llu\n",
1060 (u_longlong_t
)zs
.zs_num_blocks
);
1061 (void) printf("\t\tzap_block_type: 0x%llx\n",
1062 (u_longlong_t
)zs
.zs_block_type
);
1063 (void) printf("\t\tzap_magic: 0x%llx\n",
1064 (u_longlong_t
)zs
.zs_magic
);
1065 (void) printf("\t\tzap_salt: 0x%llx\n",
1066 (u_longlong_t
)zs
.zs_salt
);
1068 (void) printf("\t\tLeafs with 2^n pointers:\n");
1069 dump_histogram(zs
.zs_leafs_with_2n_pointers
, ZAP_HISTOGRAM_SIZE
, 0);
1071 (void) printf("\t\tBlocks with n*5 entries:\n");
1072 dump_histogram(zs
.zs_blocks_with_n5_entries
, ZAP_HISTOGRAM_SIZE
, 0);
1074 (void) printf("\t\tBlocks n/10 full:\n");
1075 dump_histogram(zs
.zs_blocks_n_tenths_full
, ZAP_HISTOGRAM_SIZE
, 0);
1077 (void) printf("\t\tEntries with n chunks:\n");
1078 dump_histogram(zs
.zs_entries_using_n_chunks
, ZAP_HISTOGRAM_SIZE
, 0);
1080 (void) printf("\t\tBuckets with n entries:\n");
1081 dump_histogram(zs
.zs_buckets_with_n_entries
, ZAP_HISTOGRAM_SIZE
, 0);
1085 dump_none(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1087 (void) os
, (void) object
, (void) data
, (void) size
;
1091 dump_unknown(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1093 (void) os
, (void) object
, (void) data
, (void) size
;
1094 (void) printf("\tUNKNOWN OBJECT TYPE\n");
1098 dump_uint8(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1100 (void) os
, (void) object
, (void) data
, (void) size
;
1104 dump_uint64(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1108 if (dump_opt
['d'] < 6)
1112 dmu_object_info_t doi
;
1114 VERIFY0(dmu_object_info(os
, object
, &doi
));
1115 size
= doi
.doi_max_offset
;
1117 * We cap the size at 1 mebibyte here to prevent
1118 * allocation failures and nigh-infinite printing if the
1119 * object is extremely large.
1121 oursize
= MIN(size
, 1 << 20);
1122 arr
= kmem_alloc(oursize
, KM_SLEEP
);
1124 int err
= dmu_read(os
, object
, 0, oursize
, arr
, 0);
1126 (void) printf("got error %u from dmu_read\n", err
);
1127 kmem_free(arr
, oursize
);
1132 * Even though the allocation is already done in this code path,
1133 * we still cap the size to prevent excessive printing.
1135 oursize
= MIN(size
, 1 << 20);
1140 (void) printf("\t\t[]\n");
1144 (void) printf("\t\t[%0llx", (u_longlong_t
)arr
[0]);
1145 for (size_t i
= 1; i
* sizeof (uint64_t) < oursize
; i
++) {
1147 (void) printf(", %0llx", (u_longlong_t
)arr
[i
]);
1149 (void) printf(",\n\t\t%0llx", (u_longlong_t
)arr
[i
]);
1151 if (oursize
!= size
)
1152 (void) printf(", ... ");
1153 (void) printf("]\n");
1156 kmem_free(arr
, oursize
);
1160 dump_zap(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1162 (void) data
, (void) size
;
1164 zap_attribute_t attr
;
1168 dump_zap_stats(os
, object
);
1169 (void) printf("\n");
1171 for (zap_cursor_init(&zc
, os
, object
);
1172 zap_cursor_retrieve(&zc
, &attr
) == 0;
1173 zap_cursor_advance(&zc
)) {
1174 (void) printf("\t\t%s = ", attr
.za_name
);
1175 if (attr
.za_num_integers
== 0) {
1176 (void) printf("\n");
1179 prop
= umem_zalloc(attr
.za_num_integers
*
1180 attr
.za_integer_length
, UMEM_NOFAIL
);
1181 (void) zap_lookup(os
, object
, attr
.za_name
,
1182 attr
.za_integer_length
, attr
.za_num_integers
, prop
);
1183 if (attr
.za_integer_length
== 1) {
1184 if (strcmp(attr
.za_name
,
1185 DSL_CRYPTO_KEY_MASTER_KEY
) == 0 ||
1186 strcmp(attr
.za_name
,
1187 DSL_CRYPTO_KEY_HMAC_KEY
) == 0 ||
1188 strcmp(attr
.za_name
, DSL_CRYPTO_KEY_IV
) == 0 ||
1189 strcmp(attr
.za_name
, DSL_CRYPTO_KEY_MAC
) == 0 ||
1190 strcmp(attr
.za_name
, DMU_POOL_CHECKSUM_SALT
) == 0) {
1193 for (i
= 0; i
< attr
.za_num_integers
; i
++) {
1194 (void) printf("%02x", u8
[i
]);
1197 (void) printf("%s", (char *)prop
);
1200 for (i
= 0; i
< attr
.za_num_integers
; i
++) {
1201 switch (attr
.za_integer_length
) {
1203 (void) printf("%u ",
1204 ((uint16_t *)prop
)[i
]);
1207 (void) printf("%u ",
1208 ((uint32_t *)prop
)[i
]);
1211 (void) printf("%lld ",
1212 (u_longlong_t
)((int64_t *)prop
)[i
]);
1217 (void) printf("\n");
1218 umem_free(prop
, attr
.za_num_integers
* attr
.za_integer_length
);
1220 zap_cursor_fini(&zc
);
1224 dump_bpobj(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1226 bpobj_phys_t
*bpop
= data
;
1228 char bytes
[32], comp
[32], uncomp
[32];
1230 /* make sure the output won't get truncated */
1231 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
1232 _Static_assert(sizeof (comp
) >= NN_NUMBUF_SZ
, "comp truncated");
1233 _Static_assert(sizeof (uncomp
) >= NN_NUMBUF_SZ
, "uncomp truncated");
1238 zdb_nicenum(bpop
->bpo_bytes
, bytes
, sizeof (bytes
));
1239 zdb_nicenum(bpop
->bpo_comp
, comp
, sizeof (comp
));
1240 zdb_nicenum(bpop
->bpo_uncomp
, uncomp
, sizeof (uncomp
));
1242 (void) printf("\t\tnum_blkptrs = %llu\n",
1243 (u_longlong_t
)bpop
->bpo_num_blkptrs
);
1244 (void) printf("\t\tbytes = %s\n", bytes
);
1245 if (size
>= BPOBJ_SIZE_V1
) {
1246 (void) printf("\t\tcomp = %s\n", comp
);
1247 (void) printf("\t\tuncomp = %s\n", uncomp
);
1249 if (size
>= BPOBJ_SIZE_V2
) {
1250 (void) printf("\t\tsubobjs = %llu\n",
1251 (u_longlong_t
)bpop
->bpo_subobjs
);
1252 (void) printf("\t\tnum_subobjs = %llu\n",
1253 (u_longlong_t
)bpop
->bpo_num_subobjs
);
1255 if (size
>= sizeof (*bpop
)) {
1256 (void) printf("\t\tnum_freed = %llu\n",
1257 (u_longlong_t
)bpop
->bpo_num_freed
);
1260 if (dump_opt
['d'] < 5)
1263 for (i
= 0; i
< bpop
->bpo_num_blkptrs
; i
++) {
1264 char blkbuf
[BP_SPRINTF_LEN
];
1267 int err
= dmu_read(os
, object
,
1268 i
* sizeof (bp
), sizeof (bp
), &bp
, 0);
1270 (void) printf("got error %u from dmu_read\n", err
);
1273 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
), &bp
,
1275 (void) printf("\t%s\n", blkbuf
);
1280 dump_bpobj_subobjs(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1282 (void) data
, (void) size
;
1283 dmu_object_info_t doi
;
1286 VERIFY0(dmu_object_info(os
, object
, &doi
));
1287 uint64_t *subobjs
= kmem_alloc(doi
.doi_max_offset
, KM_SLEEP
);
1289 int err
= dmu_read(os
, object
, 0, doi
.doi_max_offset
, subobjs
, 0);
1291 (void) printf("got error %u from dmu_read\n", err
);
1292 kmem_free(subobjs
, doi
.doi_max_offset
);
1296 int64_t last_nonzero
= -1;
1297 for (i
= 0; i
< doi
.doi_max_offset
/ 8; i
++) {
1298 if (subobjs
[i
] != 0)
1302 for (i
= 0; i
<= last_nonzero
; i
++) {
1303 (void) printf("\t%llu\n", (u_longlong_t
)subobjs
[i
]);
1305 kmem_free(subobjs
, doi
.doi_max_offset
);
1309 dump_ddt_zap(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1311 (void) data
, (void) size
;
1312 dump_zap_stats(os
, object
);
1313 /* contents are printed elsewhere, properly decoded */
1317 dump_sa_attrs(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1319 (void) data
, (void) size
;
1321 zap_attribute_t attr
;
1323 dump_zap_stats(os
, object
);
1324 (void) printf("\n");
1326 for (zap_cursor_init(&zc
, os
, object
);
1327 zap_cursor_retrieve(&zc
, &attr
) == 0;
1328 zap_cursor_advance(&zc
)) {
1329 (void) printf("\t\t%s = ", attr
.za_name
);
1330 if (attr
.za_num_integers
== 0) {
1331 (void) printf("\n");
1334 (void) printf(" %llx : [%d:%d:%d]\n",
1335 (u_longlong_t
)attr
.za_first_integer
,
1336 (int)ATTR_LENGTH(attr
.za_first_integer
),
1337 (int)ATTR_BSWAP(attr
.za_first_integer
),
1338 (int)ATTR_NUM(attr
.za_first_integer
));
1340 zap_cursor_fini(&zc
);
1344 dump_sa_layouts(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1346 (void) data
, (void) size
;
1348 zap_attribute_t attr
;
1349 uint16_t *layout_attrs
;
1352 dump_zap_stats(os
, object
);
1353 (void) printf("\n");
1355 for (zap_cursor_init(&zc
, os
, object
);
1356 zap_cursor_retrieve(&zc
, &attr
) == 0;
1357 zap_cursor_advance(&zc
)) {
1358 (void) printf("\t\t%s = [", attr
.za_name
);
1359 if (attr
.za_num_integers
== 0) {
1360 (void) printf("\n");
1364 VERIFY(attr
.za_integer_length
== 2);
1365 layout_attrs
= umem_zalloc(attr
.za_num_integers
*
1366 attr
.za_integer_length
, UMEM_NOFAIL
);
1368 VERIFY(zap_lookup(os
, object
, attr
.za_name
,
1369 attr
.za_integer_length
,
1370 attr
.za_num_integers
, layout_attrs
) == 0);
1372 for (i
= 0; i
!= attr
.za_num_integers
; i
++)
1373 (void) printf(" %d ", (int)layout_attrs
[i
]);
1374 (void) printf("]\n");
1375 umem_free(layout_attrs
,
1376 attr
.za_num_integers
* attr
.za_integer_length
);
1378 zap_cursor_fini(&zc
);
1382 dump_zpldir(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1384 (void) data
, (void) size
;
1386 zap_attribute_t attr
;
1387 const char *typenames
[] = {
1388 /* 0 */ "not specified",
1390 /* 2 */ "Character Device",
1391 /* 3 */ "3 (invalid)",
1392 /* 4 */ "Directory",
1393 /* 5 */ "5 (invalid)",
1394 /* 6 */ "Block Device",
1395 /* 7 */ "7 (invalid)",
1396 /* 8 */ "Regular File",
1397 /* 9 */ "9 (invalid)",
1398 /* 10 */ "Symbolic Link",
1399 /* 11 */ "11 (invalid)",
1402 /* 14 */ "Event Port",
1403 /* 15 */ "15 (invalid)",
1406 dump_zap_stats(os
, object
);
1407 (void) printf("\n");
1409 for (zap_cursor_init(&zc
, os
, object
);
1410 zap_cursor_retrieve(&zc
, &attr
) == 0;
1411 zap_cursor_advance(&zc
)) {
1412 (void) printf("\t\t%s = %lld (type: %s)\n",
1413 attr
.za_name
, ZFS_DIRENT_OBJ(attr
.za_first_integer
),
1414 typenames
[ZFS_DIRENT_TYPE(attr
.za_first_integer
)]);
1416 zap_cursor_fini(&zc
);
1420 get_dtl_refcount(vdev_t
*vd
)
1424 if (vd
->vdev_ops
->vdev_op_leaf
) {
1425 space_map_t
*sm
= vd
->vdev_dtl_sm
;
1428 sm
->sm_dbuf
->db_size
== sizeof (space_map_phys_t
))
1433 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++)
1434 refcount
+= get_dtl_refcount(vd
->vdev_child
[c
]);
1439 get_metaslab_refcount(vdev_t
*vd
)
1443 if (vd
->vdev_top
== vd
) {
1444 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
1445 space_map_t
*sm
= vd
->vdev_ms
[m
]->ms_sm
;
1448 sm
->sm_dbuf
->db_size
== sizeof (space_map_phys_t
))
1452 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++)
1453 refcount
+= get_metaslab_refcount(vd
->vdev_child
[c
]);
1459 get_obsolete_refcount(vdev_t
*vd
)
1461 uint64_t obsolete_sm_object
;
1464 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
1465 if (vd
->vdev_top
== vd
&& obsolete_sm_object
!= 0) {
1466 dmu_object_info_t doi
;
1467 VERIFY0(dmu_object_info(vd
->vdev_spa
->spa_meta_objset
,
1468 obsolete_sm_object
, &doi
));
1469 if (doi
.doi_bonus_size
== sizeof (space_map_phys_t
)) {
1473 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
1474 ASSERT3U(obsolete_sm_object
, ==, 0);
1476 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++) {
1477 refcount
+= get_obsolete_refcount(vd
->vdev_child
[c
]);
1484 get_prev_obsolete_spacemap_refcount(spa_t
*spa
)
1487 spa
->spa_condensing_indirect_phys
.scip_prev_obsolete_sm_object
;
1488 if (prev_obj
!= 0) {
1489 dmu_object_info_t doi
;
1490 VERIFY0(dmu_object_info(spa
->spa_meta_objset
, prev_obj
, &doi
));
1491 if (doi
.doi_bonus_size
== sizeof (space_map_phys_t
)) {
1499 get_checkpoint_refcount(vdev_t
*vd
)
1503 if (vd
->vdev_top
== vd
&& vd
->vdev_top_zap
!= 0 &&
1504 zap_contains(spa_meta_objset(vd
->vdev_spa
),
1505 vd
->vdev_top_zap
, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) == 0)
1508 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1509 refcount
+= get_checkpoint_refcount(vd
->vdev_child
[c
]);
1515 get_log_spacemap_refcount(spa_t
*spa
)
1517 return (avl_numnodes(&spa
->spa_sm_logs_by_txg
));
1521 verify_spacemap_refcounts(spa_t
*spa
)
1523 uint64_t expected_refcount
= 0;
1524 uint64_t actual_refcount
;
1526 (void) feature_get_refcount(spa
,
1527 &spa_feature_table
[SPA_FEATURE_SPACEMAP_HISTOGRAM
],
1528 &expected_refcount
);
1529 actual_refcount
= get_dtl_refcount(spa
->spa_root_vdev
);
1530 actual_refcount
+= get_metaslab_refcount(spa
->spa_root_vdev
);
1531 actual_refcount
+= get_obsolete_refcount(spa
->spa_root_vdev
);
1532 actual_refcount
+= get_prev_obsolete_spacemap_refcount(spa
);
1533 actual_refcount
+= get_checkpoint_refcount(spa
->spa_root_vdev
);
1534 actual_refcount
+= get_log_spacemap_refcount(spa
);
1536 if (expected_refcount
!= actual_refcount
) {
1537 (void) printf("space map refcount mismatch: expected %lld != "
1539 (longlong_t
)expected_refcount
,
1540 (longlong_t
)actual_refcount
);
1547 dump_spacemap(objset_t
*os
, space_map_t
*sm
)
1549 const char *ddata
[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1550 "INVALID", "INVALID", "INVALID", "INVALID" };
1555 (void) printf("space map object %llu:\n",
1556 (longlong_t
)sm
->sm_object
);
1557 (void) printf(" smp_length = 0x%llx\n",
1558 (longlong_t
)sm
->sm_phys
->smp_length
);
1559 (void) printf(" smp_alloc = 0x%llx\n",
1560 (longlong_t
)sm
->sm_phys
->smp_alloc
);
1562 if (dump_opt
['d'] < 6 && dump_opt
['m'] < 4)
1566 * Print out the freelist entries in both encoded and decoded form.
1568 uint8_t mapshift
= sm
->sm_shift
;
1570 uint64_t word
, entry_id
= 0;
1571 for (uint64_t offset
= 0; offset
< space_map_length(sm
);
1572 offset
+= sizeof (word
)) {
1574 VERIFY0(dmu_read(os
, space_map_object(sm
), offset
,
1575 sizeof (word
), &word
, DMU_READ_PREFETCH
));
1577 if (sm_entry_is_debug(word
)) {
1578 uint64_t de_txg
= SM_DEBUG_TXG_DECODE(word
);
1579 uint64_t de_sync_pass
= SM_DEBUG_SYNCPASS_DECODE(word
);
1582 "\t [%6llu] PADDING\n",
1583 (u_longlong_t
)entry_id
);
1586 "\t [%6llu] %s: txg %llu pass %llu\n",
1587 (u_longlong_t
)entry_id
,
1588 ddata
[SM_DEBUG_ACTION_DECODE(word
)],
1589 (u_longlong_t
)de_txg
,
1590 (u_longlong_t
)de_sync_pass
);
1598 uint64_t entry_off
, entry_run
, entry_vdev
= SM_NO_VDEVID
;
1600 if (sm_entry_is_single_word(word
)) {
1601 entry_type
= (SM_TYPE_DECODE(word
) == SM_ALLOC
) ?
1603 entry_off
= (SM_OFFSET_DECODE(word
) << mapshift
) +
1605 entry_run
= SM_RUN_DECODE(word
) << mapshift
;
1608 /* it is a two-word entry so we read another word */
1609 ASSERT(sm_entry_is_double_word(word
));
1611 uint64_t extra_word
;
1612 offset
+= sizeof (extra_word
);
1613 VERIFY0(dmu_read(os
, space_map_object(sm
), offset
,
1614 sizeof (extra_word
), &extra_word
,
1615 DMU_READ_PREFETCH
));
1617 ASSERT3U(offset
, <=, space_map_length(sm
));
1619 entry_run
= SM2_RUN_DECODE(word
) << mapshift
;
1620 entry_vdev
= SM2_VDEV_DECODE(word
);
1621 entry_type
= (SM2_TYPE_DECODE(extra_word
) == SM_ALLOC
) ?
1623 entry_off
= (SM2_OFFSET_DECODE(extra_word
) <<
1624 mapshift
) + sm
->sm_start
;
1628 (void) printf("\t [%6llu] %c range:"
1629 " %010llx-%010llx size: %06llx vdev: %06llu words: %u\n",
1630 (u_longlong_t
)entry_id
,
1631 entry_type
, (u_longlong_t
)entry_off
,
1632 (u_longlong_t
)(entry_off
+ entry_run
),
1633 (u_longlong_t
)entry_run
,
1634 (u_longlong_t
)entry_vdev
, words
);
1636 if (entry_type
== 'A')
1642 if (alloc
!= space_map_allocated(sm
)) {
1643 (void) printf("space_map_object alloc (%lld) INCONSISTENT "
1644 "with space map summary (%lld)\n",
1645 (longlong_t
)space_map_allocated(sm
), (longlong_t
)alloc
);
1650 dump_metaslab_stats(metaslab_t
*msp
)
1653 range_tree_t
*rt
= msp
->ms_allocatable
;
1654 zfs_btree_t
*t
= &msp
->ms_allocatable_by_size
;
1655 int free_pct
= range_tree_space(rt
) * 100 / msp
->ms_size
;
1657 /* max sure nicenum has enough space */
1658 _Static_assert(sizeof (maxbuf
) >= NN_NUMBUF_SZ
, "maxbuf truncated");
1660 zdb_nicenum(metaslab_largest_allocatable(msp
), maxbuf
, sizeof (maxbuf
));
1662 (void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
1663 "segments", zfs_btree_numnodes(t
), "maxsize", maxbuf
,
1664 "freepct", free_pct
);
1665 (void) printf("\tIn-memory histogram:\n");
1666 dump_histogram(rt
->rt_histogram
, RANGE_TREE_HISTOGRAM_SIZE
, 0);
1670 dump_metaslab(metaslab_t
*msp
)
1672 vdev_t
*vd
= msp
->ms_group
->mg_vd
;
1673 spa_t
*spa
= vd
->vdev_spa
;
1674 space_map_t
*sm
= msp
->ms_sm
;
1677 zdb_nicenum(msp
->ms_size
- space_map_allocated(sm
), freebuf
,
1681 "\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
1682 (u_longlong_t
)msp
->ms_id
, (u_longlong_t
)msp
->ms_start
,
1683 (u_longlong_t
)space_map_object(sm
), freebuf
);
1685 if (dump_opt
['m'] > 2 && !dump_opt
['L']) {
1686 mutex_enter(&msp
->ms_lock
);
1687 VERIFY0(metaslab_load(msp
));
1688 range_tree_stat_verify(msp
->ms_allocatable
);
1689 dump_metaslab_stats(msp
);
1690 metaslab_unload(msp
);
1691 mutex_exit(&msp
->ms_lock
);
1694 if (dump_opt
['m'] > 1 && sm
!= NULL
&&
1695 spa_feature_is_active(spa
, SPA_FEATURE_SPACEMAP_HISTOGRAM
)) {
1697 * The space map histogram represents free space in chunks
1698 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1700 (void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1701 (u_longlong_t
)msp
->ms_fragmentation
);
1702 dump_histogram(sm
->sm_phys
->smp_histogram
,
1703 SPACE_MAP_HISTOGRAM_SIZE
, sm
->sm_shift
);
1706 if (vd
->vdev_ops
== &vdev_draid_ops
)
1707 ASSERT3U(msp
->ms_size
, <=, 1ULL << vd
->vdev_ms_shift
);
1709 ASSERT3U(msp
->ms_size
, ==, 1ULL << vd
->vdev_ms_shift
);
1711 dump_spacemap(spa
->spa_meta_objset
, msp
->ms_sm
);
1713 if (spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
)) {
1714 (void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1715 (u_longlong_t
)metaslab_unflushed_txg(msp
));
1720 print_vdev_metaslab_header(vdev_t
*vd
)
1722 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
1723 const char *bias_str
= "";
1724 if (alloc_bias
== VDEV_BIAS_LOG
|| vd
->vdev_islog
) {
1725 bias_str
= VDEV_ALLOC_BIAS_LOG
;
1726 } else if (alloc_bias
== VDEV_BIAS_SPECIAL
) {
1727 bias_str
= VDEV_ALLOC_BIAS_SPECIAL
;
1728 } else if (alloc_bias
== VDEV_BIAS_DEDUP
) {
1729 bias_str
= VDEV_ALLOC_BIAS_DEDUP
;
1732 uint64_t ms_flush_data_obj
= 0;
1733 if (vd
->vdev_top_zap
!= 0) {
1734 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
),
1735 vd
->vdev_top_zap
, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
,
1736 sizeof (uint64_t), 1, &ms_flush_data_obj
);
1737 if (error
!= ENOENT
) {
1742 (void) printf("\tvdev %10llu %s",
1743 (u_longlong_t
)vd
->vdev_id
, bias_str
);
1745 if (ms_flush_data_obj
!= 0) {
1746 (void) printf(" ms_unflushed_phys object %llu",
1747 (u_longlong_t
)ms_flush_data_obj
);
1750 (void) printf("\n\t%-10s%5llu %-19s %-15s %-12s\n",
1751 "metaslabs", (u_longlong_t
)vd
->vdev_ms_count
,
1752 "offset", "spacemap", "free");
1753 (void) printf("\t%15s %19s %15s %12s\n",
1754 "---------------", "-------------------",
1755 "---------------", "------------");
1759 dump_metaslab_groups(spa_t
*spa
, boolean_t show_special
)
1761 vdev_t
*rvd
= spa
->spa_root_vdev
;
1762 metaslab_class_t
*mc
= spa_normal_class(spa
);
1763 metaslab_class_t
*smc
= spa_special_class(spa
);
1764 uint64_t fragmentation
;
1766 metaslab_class_histogram_verify(mc
);
1768 for (unsigned c
= 0; c
< rvd
->vdev_children
; c
++) {
1769 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1770 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1772 if (mg
== NULL
|| (mg
->mg_class
!= mc
&&
1773 (!show_special
|| mg
->mg_class
!= smc
)))
1776 metaslab_group_histogram_verify(mg
);
1777 mg
->mg_fragmentation
= metaslab_group_fragmentation(mg
);
1779 (void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1781 (u_longlong_t
)tvd
->vdev_id
,
1782 (u_longlong_t
)tvd
->vdev_ms_count
);
1783 if (mg
->mg_fragmentation
== ZFS_FRAG_INVALID
) {
1784 (void) printf("%3s\n", "-");
1786 (void) printf("%3llu%%\n",
1787 (u_longlong_t
)mg
->mg_fragmentation
);
1789 dump_histogram(mg
->mg_histogram
, RANGE_TREE_HISTOGRAM_SIZE
, 0);
1792 (void) printf("\tpool %s\tfragmentation", spa_name(spa
));
1793 fragmentation
= metaslab_class_fragmentation(mc
);
1794 if (fragmentation
== ZFS_FRAG_INVALID
)
1795 (void) printf("\t%3s\n", "-");
1797 (void) printf("\t%3llu%%\n", (u_longlong_t
)fragmentation
);
1798 dump_histogram(mc
->mc_histogram
, RANGE_TREE_HISTOGRAM_SIZE
, 0);
1802 print_vdev_indirect(vdev_t
*vd
)
1804 vdev_indirect_config_t
*vic
= &vd
->vdev_indirect_config
;
1805 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
1806 vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
;
1809 ASSERT3P(vib
, ==, NULL
);
1813 ASSERT3U(vdev_indirect_mapping_object(vim
), ==,
1814 vic
->vic_mapping_object
);
1815 ASSERT3U(vdev_indirect_births_object(vib
), ==,
1816 vic
->vic_births_object
);
1818 (void) printf("indirect births obj %llu:\n",
1819 (longlong_t
)vic
->vic_births_object
);
1820 (void) printf(" vib_count = %llu\n",
1821 (longlong_t
)vdev_indirect_births_count(vib
));
1822 for (uint64_t i
= 0; i
< vdev_indirect_births_count(vib
); i
++) {
1823 vdev_indirect_birth_entry_phys_t
*cur_vibe
=
1824 &vib
->vib_entries
[i
];
1825 (void) printf("\toffset %llx -> txg %llu\n",
1826 (longlong_t
)cur_vibe
->vibe_offset
,
1827 (longlong_t
)cur_vibe
->vibe_phys_birth_txg
);
1829 (void) printf("\n");
1831 (void) printf("indirect mapping obj %llu:\n",
1832 (longlong_t
)vic
->vic_mapping_object
);
1833 (void) printf(" vim_max_offset = 0x%llx\n",
1834 (longlong_t
)vdev_indirect_mapping_max_offset(vim
));
1835 (void) printf(" vim_bytes_mapped = 0x%llx\n",
1836 (longlong_t
)vdev_indirect_mapping_bytes_mapped(vim
));
1837 (void) printf(" vim_count = %llu\n",
1838 (longlong_t
)vdev_indirect_mapping_num_entries(vim
));
1840 if (dump_opt
['d'] <= 5 && dump_opt
['m'] <= 3)
1843 uint32_t *counts
= vdev_indirect_mapping_load_obsolete_counts(vim
);
1845 for (uint64_t i
= 0; i
< vdev_indirect_mapping_num_entries(vim
); i
++) {
1846 vdev_indirect_mapping_entry_phys_t
*vimep
=
1847 &vim
->vim_entries
[i
];
1848 (void) printf("\t<%llx:%llx:%llx> -> "
1849 "<%llx:%llx:%llx> (%x obsolete)\n",
1850 (longlong_t
)vd
->vdev_id
,
1851 (longlong_t
)DVA_MAPPING_GET_SRC_OFFSET(vimep
),
1852 (longlong_t
)DVA_GET_ASIZE(&vimep
->vimep_dst
),
1853 (longlong_t
)DVA_GET_VDEV(&vimep
->vimep_dst
),
1854 (longlong_t
)DVA_GET_OFFSET(&vimep
->vimep_dst
),
1855 (longlong_t
)DVA_GET_ASIZE(&vimep
->vimep_dst
),
1858 (void) printf("\n");
1860 uint64_t obsolete_sm_object
;
1861 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
1862 if (obsolete_sm_object
!= 0) {
1863 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
1864 (void) printf("obsolete space map object %llu:\n",
1865 (u_longlong_t
)obsolete_sm_object
);
1866 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
1867 ASSERT3U(space_map_object(vd
->vdev_obsolete_sm
), ==,
1868 obsolete_sm_object
);
1869 dump_spacemap(mos
, vd
->vdev_obsolete_sm
);
1870 (void) printf("\n");
1875 dump_metaslabs(spa_t
*spa
)
1877 vdev_t
*vd
, *rvd
= spa
->spa_root_vdev
;
1878 uint64_t m
, c
= 0, children
= rvd
->vdev_children
;
1880 (void) printf("\nMetaslabs:\n");
1882 if (!dump_opt
['d'] && zopt_metaslab_args
> 0) {
1883 c
= zopt_metaslab
[0];
1886 (void) fatal("bad vdev id: %llu", (u_longlong_t
)c
);
1888 if (zopt_metaslab_args
> 1) {
1889 vd
= rvd
->vdev_child
[c
];
1890 print_vdev_metaslab_header(vd
);
1892 for (m
= 1; m
< zopt_metaslab_args
; m
++) {
1893 if (zopt_metaslab
[m
] < vd
->vdev_ms_count
)
1895 vd
->vdev_ms
[zopt_metaslab
[m
]]);
1897 (void) fprintf(stderr
, "bad metaslab "
1899 (u_longlong_t
)zopt_metaslab
[m
]);
1901 (void) printf("\n");
1906 for (; c
< children
; c
++) {
1907 vd
= rvd
->vdev_child
[c
];
1908 print_vdev_metaslab_header(vd
);
1910 print_vdev_indirect(vd
);
1912 for (m
= 0; m
< vd
->vdev_ms_count
; m
++)
1913 dump_metaslab(vd
->vdev_ms
[m
]);
1914 (void) printf("\n");
1919 dump_log_spacemaps(spa_t
*spa
)
1921 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
1924 (void) printf("\nLog Space Maps in Pool:\n");
1925 for (spa_log_sm_t
*sls
= avl_first(&spa
->spa_sm_logs_by_txg
);
1926 sls
; sls
= AVL_NEXT(&spa
->spa_sm_logs_by_txg
, sls
)) {
1927 space_map_t
*sm
= NULL
;
1928 VERIFY0(space_map_open(&sm
, spa_meta_objset(spa
),
1929 sls
->sls_sm_obj
, 0, UINT64_MAX
, SPA_MINBLOCKSHIFT
));
1931 (void) printf("Log Spacemap object %llu txg %llu\n",
1932 (u_longlong_t
)sls
->sls_sm_obj
, (u_longlong_t
)sls
->sls_txg
);
1933 dump_spacemap(spa
->spa_meta_objset
, sm
);
1934 space_map_close(sm
);
1936 (void) printf("\n");
1940 dump_dde(const ddt_t
*ddt
, const ddt_entry_t
*dde
, uint64_t index
)
1942 const ddt_phys_t
*ddp
= dde
->dde_phys
;
1943 const ddt_key_t
*ddk
= &dde
->dde_key
;
1944 const char *types
[4] = { "ditto", "single", "double", "triple" };
1945 char blkbuf
[BP_SPRINTF_LEN
];
1949 for (p
= 0; p
< DDT_PHYS_TYPES
; p
++, ddp
++) {
1950 if (ddp
->ddp_phys_birth
== 0)
1952 ddt_bp_create(ddt
->ddt_checksum
, ddk
, ddp
, &blk
);
1953 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), &blk
);
1954 (void) printf("index %llx refcnt %llu %s %s\n",
1955 (u_longlong_t
)index
, (u_longlong_t
)ddp
->ddp_refcnt
,
1961 dump_dedup_ratio(const ddt_stat_t
*dds
)
1963 double rL
, rP
, rD
, D
, dedup
, compress
, copies
;
1965 if (dds
->dds_blocks
== 0)
1968 rL
= (double)dds
->dds_ref_lsize
;
1969 rP
= (double)dds
->dds_ref_psize
;
1970 rD
= (double)dds
->dds_ref_dsize
;
1971 D
= (double)dds
->dds_dsize
;
1977 (void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
1978 "dedup * compress / copies = %.2f\n\n",
1979 dedup
, compress
, copies
, dedup
* compress
/ copies
);
1983 dump_ddt(ddt_t
*ddt
, enum ddt_type type
, enum ddt_class
class)
1985 char name
[DDT_NAMELEN
];
1988 dmu_object_info_t doi
;
1989 uint64_t count
, dspace
, mspace
;
1992 error
= ddt_object_info(ddt
, type
, class, &doi
);
1994 if (error
== ENOENT
)
1998 error
= ddt_object_count(ddt
, type
, class, &count
);
2003 dspace
= doi
.doi_physical_blocks_512
<< 9;
2004 mspace
= doi
.doi_fill_count
* doi
.doi_data_block_size
;
2006 ddt_object_name(ddt
, type
, class, name
);
2008 (void) printf("%s: %llu entries, size %llu on disk, %llu in core\n",
2010 (u_longlong_t
)count
,
2011 (u_longlong_t
)(dspace
/ count
),
2012 (u_longlong_t
)(mspace
/ count
));
2014 if (dump_opt
['D'] < 3)
2017 zpool_dump_ddt(NULL
, &ddt
->ddt_histogram
[type
][class]);
2019 if (dump_opt
['D'] < 4)
2022 if (dump_opt
['D'] < 5 && class == DDT_CLASS_UNIQUE
)
2025 (void) printf("%s contents:\n\n", name
);
2027 while ((error
= ddt_object_walk(ddt
, type
, class, &walk
, &dde
)) == 0)
2028 dump_dde(ddt
, &dde
, walk
);
2030 ASSERT3U(error
, ==, ENOENT
);
2032 (void) printf("\n");
2036 dump_all_ddts(spa_t
*spa
)
2038 ddt_histogram_t ddh_total
= {{{0}}};
2039 ddt_stat_t dds_total
= {0};
2041 for (enum zio_checksum c
= 0; c
< ZIO_CHECKSUM_FUNCTIONS
; c
++) {
2042 ddt_t
*ddt
= spa
->spa_ddt
[c
];
2043 for (enum ddt_type type
= 0; type
< DDT_TYPES
; type
++) {
2044 for (enum ddt_class
class = 0; class < DDT_CLASSES
;
2046 dump_ddt(ddt
, type
, class);
2051 ddt_get_dedup_stats(spa
, &dds_total
);
2053 if (dds_total
.dds_blocks
== 0) {
2054 (void) printf("All DDTs are empty\n");
2058 (void) printf("\n");
2060 if (dump_opt
['D'] > 1) {
2061 (void) printf("DDT histogram (aggregated over all DDTs):\n");
2062 ddt_get_dedup_histogram(spa
, &ddh_total
);
2063 zpool_dump_ddt(&dds_total
, &ddh_total
);
2066 dump_dedup_ratio(&dds_total
);
2070 dump_dtl_seg(void *arg
, uint64_t start
, uint64_t size
)
2074 (void) printf("%s [%llu,%llu) length %llu\n",
2076 (u_longlong_t
)start
,
2077 (u_longlong_t
)(start
+ size
),
2078 (u_longlong_t
)(size
));
2082 dump_dtl(vdev_t
*vd
, int indent
)
2084 spa_t
*spa
= vd
->vdev_spa
;
2086 const char *name
[DTL_TYPES
] = { "missing", "partial", "scrub",
2090 spa_vdev_state_enter(spa
, SCL_NONE
);
2091 required
= vdev_dtl_required(vd
);
2092 (void) spa_vdev_state_exit(spa
, NULL
, 0);
2095 (void) printf("\nDirty time logs:\n\n");
2097 (void) printf("\t%*s%s [%s]\n", indent
, "",
2098 vd
->vdev_path
? vd
->vdev_path
:
2099 vd
->vdev_parent
? vd
->vdev_ops
->vdev_op_type
: spa_name(spa
),
2100 required
? "DTL-required" : "DTL-expendable");
2102 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2103 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2104 if (range_tree_space(rt
) == 0)
2106 (void) snprintf(prefix
, sizeof (prefix
), "\t%*s%s",
2107 indent
+ 2, "", name
[t
]);
2108 range_tree_walk(rt
, dump_dtl_seg
, prefix
);
2109 if (dump_opt
['d'] > 5 && vd
->vdev_children
== 0)
2110 dump_spacemap(spa
->spa_meta_objset
,
2114 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++)
2115 dump_dtl(vd
->vdev_child
[c
], indent
+ 4);
2119 dump_history(spa_t
*spa
)
2121 nvlist_t
**events
= NULL
;
2123 uint64_t resid
, len
, off
= 0;
2128 if ((buf
= malloc(SPA_OLD_MAXBLOCKSIZE
)) == NULL
) {
2129 (void) fprintf(stderr
, "%s: unable to allocate I/O buffer\n",
2135 len
= SPA_OLD_MAXBLOCKSIZE
;
2137 if ((error
= spa_history_get(spa
, &off
, &len
, buf
)) != 0) {
2138 (void) fprintf(stderr
, "Unable to read history: "
2139 "error %d\n", error
);
2144 if (zpool_history_unpack(buf
, len
, &resid
, &events
, &num
) != 0)
2150 (void) printf("\nHistory:\n");
2151 for (unsigned i
= 0; i
< num
; i
++) {
2152 boolean_t printed
= B_FALSE
;
2154 if (nvlist_exists(events
[i
], ZPOOL_HIST_TIME
)) {
2158 tsec
= fnvlist_lookup_uint64(events
[i
],
2160 (void) localtime_r(&tsec
, &t
);
2161 (void) strftime(tbuf
, sizeof (tbuf
), "%F.%T", &t
);
2166 if (nvlist_exists(events
[i
], ZPOOL_HIST_CMD
)) {
2167 (void) printf("%s %s\n", tbuf
,
2168 fnvlist_lookup_string(events
[i
], ZPOOL_HIST_CMD
));
2169 } else if (nvlist_exists(events
[i
], ZPOOL_HIST_INT_EVENT
)) {
2172 ievent
= fnvlist_lookup_uint64(events
[i
],
2173 ZPOOL_HIST_INT_EVENT
);
2174 if (ievent
>= ZFS_NUM_LEGACY_HISTORY_EVENTS
)
2177 (void) printf(" %s [internal %s txg:%ju] %s\n",
2179 zfs_history_event_names
[ievent
],
2180 fnvlist_lookup_uint64(events
[i
],
2182 fnvlist_lookup_string(events
[i
],
2183 ZPOOL_HIST_INT_STR
));
2184 } else if (nvlist_exists(events
[i
], ZPOOL_HIST_INT_NAME
)) {
2185 (void) printf("%s [txg:%ju] %s", tbuf
,
2186 fnvlist_lookup_uint64(events
[i
],
2188 fnvlist_lookup_string(events
[i
],
2189 ZPOOL_HIST_INT_NAME
));
2191 if (nvlist_exists(events
[i
], ZPOOL_HIST_DSNAME
)) {
2192 (void) printf(" %s (%llu)",
2193 fnvlist_lookup_string(events
[i
],
2195 (u_longlong_t
)fnvlist_lookup_uint64(
2200 (void) printf(" %s\n", fnvlist_lookup_string(events
[i
],
2201 ZPOOL_HIST_INT_STR
));
2202 } else if (nvlist_exists(events
[i
], ZPOOL_HIST_IOCTL
)) {
2203 (void) printf("%s ioctl %s\n", tbuf
,
2204 fnvlist_lookup_string(events
[i
],
2207 if (nvlist_exists(events
[i
], ZPOOL_HIST_INPUT_NVL
)) {
2208 (void) printf(" input:\n");
2209 dump_nvlist(fnvlist_lookup_nvlist(events
[i
],
2210 ZPOOL_HIST_INPUT_NVL
), 8);
2212 if (nvlist_exists(events
[i
], ZPOOL_HIST_OUTPUT_NVL
)) {
2213 (void) printf(" output:\n");
2214 dump_nvlist(fnvlist_lookup_nvlist(events
[i
],
2215 ZPOOL_HIST_OUTPUT_NVL
), 8);
2217 if (nvlist_exists(events
[i
], ZPOOL_HIST_ERRNO
)) {
2218 (void) printf(" errno: %lld\n",
2219 (longlong_t
)fnvlist_lookup_int64(events
[i
],
2228 if (dump_opt
['h'] > 1) {
2230 (void) printf("unrecognized record:\n");
2231 dump_nvlist(events
[i
], 2);
2238 dump_dnode(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
2240 (void) os
, (void) object
, (void) data
, (void) size
;
2244 blkid2offset(const dnode_phys_t
*dnp
, const blkptr_t
*bp
,
2245 const zbookmark_phys_t
*zb
)
2248 ASSERT(zb
->zb_level
< 0);
2249 if (zb
->zb_object
== 0)
2250 return (zb
->zb_blkid
);
2251 return (zb
->zb_blkid
* BP_GET_LSIZE(bp
));
2254 ASSERT(zb
->zb_level
>= 0);
2256 return ((zb
->zb_blkid
<<
2257 (zb
->zb_level
* (dnp
->dn_indblkshift
- SPA_BLKPTRSHIFT
))) *
2258 dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
2262 snprintf_zstd_header(spa_t
*spa
, char *blkbuf
, size_t buflen
,
2268 zfs_zstdhdr_t zstd_hdr
;
2271 if (BP_GET_COMPRESS(bp
) != ZIO_COMPRESS_ZSTD
)
2277 if (BP_IS_EMBEDDED(bp
)) {
2278 buf
= malloc(SPA_MAXBLOCKSIZE
);
2280 (void) fprintf(stderr
, "out of memory\n");
2283 decode_embedded_bp_compressed(bp
, buf
);
2284 memcpy(&zstd_hdr
, buf
, sizeof (zstd_hdr
));
2286 zstd_hdr
.c_len
= BE_32(zstd_hdr
.c_len
);
2287 zstd_hdr
.raw_version_level
= BE_32(zstd_hdr
.raw_version_level
);
2288 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2289 buflen
- strlen(blkbuf
),
2290 " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2291 zstd_hdr
.c_len
, zfs_get_hdrversion(&zstd_hdr
),
2292 zfs_get_hdrlevel(&zstd_hdr
));
2296 pabd
= abd_alloc_for_io(SPA_MAXBLOCKSIZE
, B_FALSE
);
2297 zio
= zio_root(spa
, NULL
, NULL
, 0);
2299 /* Decrypt but don't decompress so we can read the compression header */
2300 zio_nowait(zio_read(zio
, spa
, bp
, pabd
, BP_GET_PSIZE(bp
), NULL
, NULL
,
2301 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW_COMPRESS
,
2303 error
= zio_wait(zio
);
2305 (void) fprintf(stderr
, "read failed: %d\n", error
);
2308 buf
= abd_borrow_buf_copy(pabd
, BP_GET_LSIZE(bp
));
2309 memcpy(&zstd_hdr
, buf
, sizeof (zstd_hdr
));
2310 zstd_hdr
.c_len
= BE_32(zstd_hdr
.c_len
);
2311 zstd_hdr
.raw_version_level
= BE_32(zstd_hdr
.raw_version_level
);
2313 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2314 buflen
- strlen(blkbuf
),
2315 " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2316 zstd_hdr
.c_len
, zfs_get_hdrversion(&zstd_hdr
),
2317 zfs_get_hdrlevel(&zstd_hdr
));
2319 abd_return_buf_copy(pabd
, buf
, BP_GET_LSIZE(bp
));
2323 snprintf_blkptr_compact(char *blkbuf
, size_t buflen
, const blkptr_t
*bp
,
2326 const dva_t
*dva
= bp
->blk_dva
;
2327 int ndvas
= dump_opt
['d'] > 5 ? BP_GET_NDVAS(bp
) : 1;
2330 if (dump_opt
['b'] >= 6) {
2331 snprintf_blkptr(blkbuf
, buflen
, bp
);
2333 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2334 buflen
- strlen(blkbuf
), " %s", "FREE");
2339 if (BP_IS_EMBEDDED(bp
)) {
2340 (void) sprintf(blkbuf
,
2341 "EMBEDDED et=%u %llxL/%llxP B=%llu",
2342 (int)BPE_GET_ETYPE(bp
),
2343 (u_longlong_t
)BPE_GET_LSIZE(bp
),
2344 (u_longlong_t
)BPE_GET_PSIZE(bp
),
2345 (u_longlong_t
)bp
->blk_birth
);
2351 for (i
= 0; i
< ndvas
; i
++)
2352 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2353 buflen
- strlen(blkbuf
), "%llu:%llx:%llx ",
2354 (u_longlong_t
)DVA_GET_VDEV(&dva
[i
]),
2355 (u_longlong_t
)DVA_GET_OFFSET(&dva
[i
]),
2356 (u_longlong_t
)DVA_GET_ASIZE(&dva
[i
]));
2358 if (BP_IS_HOLE(bp
)) {
2359 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2360 buflen
- strlen(blkbuf
),
2362 (u_longlong_t
)BP_GET_LSIZE(bp
),
2363 (u_longlong_t
)bp
->blk_birth
);
2365 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2366 buflen
- strlen(blkbuf
),
2367 "%llxL/%llxP F=%llu B=%llu/%llu",
2368 (u_longlong_t
)BP_GET_LSIZE(bp
),
2369 (u_longlong_t
)BP_GET_PSIZE(bp
),
2370 (u_longlong_t
)BP_GET_FILL(bp
),
2371 (u_longlong_t
)bp
->blk_birth
,
2372 (u_longlong_t
)BP_PHYSICAL_BIRTH(bp
));
2374 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2375 buflen
- strlen(blkbuf
), " %s", "FREE");
2376 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2377 buflen
- strlen(blkbuf
), " cksum=%llx:%llx:%llx:%llx",
2378 (u_longlong_t
)bp
->blk_cksum
.zc_word
[0],
2379 (u_longlong_t
)bp
->blk_cksum
.zc_word
[1],
2380 (u_longlong_t
)bp
->blk_cksum
.zc_word
[2],
2381 (u_longlong_t
)bp
->blk_cksum
.zc_word
[3]);
2386 print_indirect(spa_t
*spa
, blkptr_t
*bp
, const zbookmark_phys_t
*zb
,
2387 const dnode_phys_t
*dnp
)
2389 char blkbuf
[BP_SPRINTF_LEN
];
2392 if (!BP_IS_EMBEDDED(bp
)) {
2393 ASSERT3U(BP_GET_TYPE(bp
), ==, dnp
->dn_type
);
2394 ASSERT3U(BP_GET_LEVEL(bp
), ==, zb
->zb_level
);
2397 (void) printf("%16llx ", (u_longlong_t
)blkid2offset(dnp
, bp
, zb
));
2399 ASSERT(zb
->zb_level
>= 0);
2401 for (l
= dnp
->dn_nlevels
- 1; l
>= -1; l
--) {
2402 if (l
== zb
->zb_level
) {
2403 (void) printf("L%llx", (u_longlong_t
)zb
->zb_level
);
2409 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
), bp
, B_FALSE
);
2410 if (dump_opt
['Z'] && BP_GET_COMPRESS(bp
) == ZIO_COMPRESS_ZSTD
)
2411 snprintf_zstd_header(spa
, blkbuf
, sizeof (blkbuf
), bp
);
2412 (void) printf("%s\n", blkbuf
);
2416 visit_indirect(spa_t
*spa
, const dnode_phys_t
*dnp
,
2417 blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
2421 if (bp
->blk_birth
== 0)
2424 print_indirect(spa
, bp
, zb
, dnp
);
2426 if (BP_GET_LEVEL(bp
) > 0 && !BP_IS_HOLE(bp
)) {
2427 arc_flags_t flags
= ARC_FLAG_WAIT
;
2430 int epb
= BP_GET_LSIZE(bp
) >> SPA_BLKPTRSHIFT
;
2433 ASSERT(!BP_IS_REDACTED(bp
));
2435 err
= arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &buf
,
2436 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
2439 ASSERT(buf
->b_data
);
2441 /* recursively visit blocks below this */
2443 for (i
= 0; i
< epb
; i
++, cbp
++) {
2444 zbookmark_phys_t czb
;
2446 SET_BOOKMARK(&czb
, zb
->zb_objset
, zb
->zb_object
,
2448 zb
->zb_blkid
* epb
+ i
);
2449 err
= visit_indirect(spa
, dnp
, cbp
, &czb
);
2452 fill
+= BP_GET_FILL(cbp
);
2455 ASSERT3U(fill
, ==, BP_GET_FILL(bp
));
2456 arc_buf_destroy(buf
, &buf
);
2463 dump_indirect(dnode_t
*dn
)
2465 dnode_phys_t
*dnp
= dn
->dn_phys
;
2466 zbookmark_phys_t czb
;
2468 (void) printf("Indirect blocks:\n");
2470 SET_BOOKMARK(&czb
, dmu_objset_id(dn
->dn_objset
),
2471 dn
->dn_object
, dnp
->dn_nlevels
- 1, 0);
2472 for (int j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
2474 (void) visit_indirect(dmu_objset_spa(dn
->dn_objset
), dnp
,
2475 &dnp
->dn_blkptr
[j
], &czb
);
2478 (void) printf("\n");
2482 dump_dsl_dir(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
2484 (void) os
, (void) object
;
2485 dsl_dir_phys_t
*dd
= data
;
2489 /* make sure nicenum has enough space */
2490 _Static_assert(sizeof (nice
) >= NN_NUMBUF_SZ
, "nice truncated");
2495 ASSERT3U(size
, >=, sizeof (dsl_dir_phys_t
));
2497 crtime
= dd
->dd_creation_time
;
2498 (void) printf("\t\tcreation_time = %s", ctime(&crtime
));
2499 (void) printf("\t\thead_dataset_obj = %llu\n",
2500 (u_longlong_t
)dd
->dd_head_dataset_obj
);
2501 (void) printf("\t\tparent_dir_obj = %llu\n",
2502 (u_longlong_t
)dd
->dd_parent_obj
);
2503 (void) printf("\t\torigin_obj = %llu\n",
2504 (u_longlong_t
)dd
->dd_origin_obj
);
2505 (void) printf("\t\tchild_dir_zapobj = %llu\n",
2506 (u_longlong_t
)dd
->dd_child_dir_zapobj
);
2507 zdb_nicenum(dd
->dd_used_bytes
, nice
, sizeof (nice
));
2508 (void) printf("\t\tused_bytes = %s\n", nice
);
2509 zdb_nicenum(dd
->dd_compressed_bytes
, nice
, sizeof (nice
));
2510 (void) printf("\t\tcompressed_bytes = %s\n", nice
);
2511 zdb_nicenum(dd
->dd_uncompressed_bytes
, nice
, sizeof (nice
));
2512 (void) printf("\t\tuncompressed_bytes = %s\n", nice
);
2513 zdb_nicenum(dd
->dd_quota
, nice
, sizeof (nice
));
2514 (void) printf("\t\tquota = %s\n", nice
);
2515 zdb_nicenum(dd
->dd_reserved
, nice
, sizeof (nice
));
2516 (void) printf("\t\treserved = %s\n", nice
);
2517 (void) printf("\t\tprops_zapobj = %llu\n",
2518 (u_longlong_t
)dd
->dd_props_zapobj
);
2519 (void) printf("\t\tdeleg_zapobj = %llu\n",
2520 (u_longlong_t
)dd
->dd_deleg_zapobj
);
2521 (void) printf("\t\tflags = %llx\n",
2522 (u_longlong_t
)dd
->dd_flags
);
2525 zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2527 (void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2534 (void) printf("\t\tclones = %llu\n",
2535 (u_longlong_t
)dd
->dd_clones
);
2539 dump_dsl_dataset(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
2541 (void) os
, (void) object
;
2542 dsl_dataset_phys_t
*ds
= data
;
2544 char used
[32], compressed
[32], uncompressed
[32], unique
[32];
2545 char blkbuf
[BP_SPRINTF_LEN
];
2547 /* make sure nicenum has enough space */
2548 _Static_assert(sizeof (used
) >= NN_NUMBUF_SZ
, "used truncated");
2549 _Static_assert(sizeof (compressed
) >= NN_NUMBUF_SZ
,
2550 "compressed truncated");
2551 _Static_assert(sizeof (uncompressed
) >= NN_NUMBUF_SZ
,
2552 "uncompressed truncated");
2553 _Static_assert(sizeof (unique
) >= NN_NUMBUF_SZ
, "unique truncated");
2558 ASSERT(size
== sizeof (*ds
));
2559 crtime
= ds
->ds_creation_time
;
2560 zdb_nicenum(ds
->ds_referenced_bytes
, used
, sizeof (used
));
2561 zdb_nicenum(ds
->ds_compressed_bytes
, compressed
, sizeof (compressed
));
2562 zdb_nicenum(ds
->ds_uncompressed_bytes
, uncompressed
,
2563 sizeof (uncompressed
));
2564 zdb_nicenum(ds
->ds_unique_bytes
, unique
, sizeof (unique
));
2565 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), &ds
->ds_bp
);
2567 (void) printf("\t\tdir_obj = %llu\n",
2568 (u_longlong_t
)ds
->ds_dir_obj
);
2569 (void) printf("\t\tprev_snap_obj = %llu\n",
2570 (u_longlong_t
)ds
->ds_prev_snap_obj
);
2571 (void) printf("\t\tprev_snap_txg = %llu\n",
2572 (u_longlong_t
)ds
->ds_prev_snap_txg
);
2573 (void) printf("\t\tnext_snap_obj = %llu\n",
2574 (u_longlong_t
)ds
->ds_next_snap_obj
);
2575 (void) printf("\t\tsnapnames_zapobj = %llu\n",
2576 (u_longlong_t
)ds
->ds_snapnames_zapobj
);
2577 (void) printf("\t\tnum_children = %llu\n",
2578 (u_longlong_t
)ds
->ds_num_children
);
2579 (void) printf("\t\tuserrefs_obj = %llu\n",
2580 (u_longlong_t
)ds
->ds_userrefs_obj
);
2581 (void) printf("\t\tcreation_time = %s", ctime(&crtime
));
2582 (void) printf("\t\tcreation_txg = %llu\n",
2583 (u_longlong_t
)ds
->ds_creation_txg
);
2584 (void) printf("\t\tdeadlist_obj = %llu\n",
2585 (u_longlong_t
)ds
->ds_deadlist_obj
);
2586 (void) printf("\t\tused_bytes = %s\n", used
);
2587 (void) printf("\t\tcompressed_bytes = %s\n", compressed
);
2588 (void) printf("\t\tuncompressed_bytes = %s\n", uncompressed
);
2589 (void) printf("\t\tunique = %s\n", unique
);
2590 (void) printf("\t\tfsid_guid = %llu\n",
2591 (u_longlong_t
)ds
->ds_fsid_guid
);
2592 (void) printf("\t\tguid = %llu\n",
2593 (u_longlong_t
)ds
->ds_guid
);
2594 (void) printf("\t\tflags = %llx\n",
2595 (u_longlong_t
)ds
->ds_flags
);
2596 (void) printf("\t\tnext_clones_obj = %llu\n",
2597 (u_longlong_t
)ds
->ds_next_clones_obj
);
2598 (void) printf("\t\tprops_obj = %llu\n",
2599 (u_longlong_t
)ds
->ds_props_obj
);
2600 (void) printf("\t\tbp = %s\n", blkbuf
);
2604 dump_bptree_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
2606 (void) arg
, (void) tx
;
2607 char blkbuf
[BP_SPRINTF_LEN
];
2609 if (bp
->blk_birth
!= 0) {
2610 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
2611 (void) printf("\t%s\n", blkbuf
);
2617 dump_bptree(objset_t
*os
, uint64_t obj
, const char *name
)
2623 /* make sure nicenum has enough space */
2624 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
2626 if (dump_opt
['d'] < 3)
2629 VERIFY3U(0, ==, dmu_bonus_hold(os
, obj
, FTAG
, &db
));
2631 zdb_nicenum(bt
->bt_bytes
, bytes
, sizeof (bytes
));
2632 (void) printf("\n %s: %llu datasets, %s\n",
2633 name
, (unsigned long long)(bt
->bt_end
- bt
->bt_begin
), bytes
);
2634 dmu_buf_rele(db
, FTAG
);
2636 if (dump_opt
['d'] < 5)
2639 (void) printf("\n");
2641 (void) bptree_iterate(os
, obj
, B_FALSE
, dump_bptree_cb
, NULL
, NULL
);
2645 dump_bpobj_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
, dmu_tx_t
*tx
)
2647 (void) arg
, (void) tx
;
2648 char blkbuf
[BP_SPRINTF_LEN
];
2650 ASSERT(bp
->blk_birth
!= 0);
2651 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
), bp
, bp_freed
);
2652 (void) printf("\t%s\n", blkbuf
);
2657 dump_full_bpobj(bpobj_t
*bpo
, const char *name
, int indent
)
2664 /* make sure nicenum has enough space */
2665 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
2666 _Static_assert(sizeof (comp
) >= NN_NUMBUF_SZ
, "comp truncated");
2667 _Static_assert(sizeof (uncomp
) >= NN_NUMBUF_SZ
, "uncomp truncated");
2669 if (dump_opt
['d'] < 3)
2672 zdb_nicenum(bpo
->bpo_phys
->bpo_bytes
, bytes
, sizeof (bytes
));
2673 if (bpo
->bpo_havesubobj
&& bpo
->bpo_phys
->bpo_subobjs
!= 0) {
2674 zdb_nicenum(bpo
->bpo_phys
->bpo_comp
, comp
, sizeof (comp
));
2675 zdb_nicenum(bpo
->bpo_phys
->bpo_uncomp
, uncomp
, sizeof (uncomp
));
2676 if (bpo
->bpo_havefreed
) {
2677 (void) printf(" %*s: object %llu, %llu local "
2678 "blkptrs, %llu freed, %llu subobjs in object %llu, "
2679 "%s (%s/%s comp)\n",
2681 (u_longlong_t
)bpo
->bpo_object
,
2682 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2683 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_freed
,
2684 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_subobjs
,
2685 (u_longlong_t
)bpo
->bpo_phys
->bpo_subobjs
,
2686 bytes
, comp
, uncomp
);
2688 (void) printf(" %*s: object %llu, %llu local "
2689 "blkptrs, %llu subobjs in object %llu, "
2690 "%s (%s/%s comp)\n",
2692 (u_longlong_t
)bpo
->bpo_object
,
2693 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2694 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_subobjs
,
2695 (u_longlong_t
)bpo
->bpo_phys
->bpo_subobjs
,
2696 bytes
, comp
, uncomp
);
2699 for (i
= 0; i
< bpo
->bpo_phys
->bpo_num_subobjs
; i
++) {
2703 VERIFY0(dmu_read(bpo
->bpo_os
,
2704 bpo
->bpo_phys
->bpo_subobjs
,
2705 i
* sizeof (subobj
), sizeof (subobj
), &subobj
, 0));
2706 error
= bpobj_open(&subbpo
, bpo
->bpo_os
, subobj
);
2708 (void) printf("ERROR %u while trying to open "
2710 error
, (u_longlong_t
)subobj
);
2713 dump_full_bpobj(&subbpo
, "subobj", indent
+ 1);
2714 bpobj_close(&subbpo
);
2717 if (bpo
->bpo_havefreed
) {
2718 (void) printf(" %*s: object %llu, %llu blkptrs, "
2721 (u_longlong_t
)bpo
->bpo_object
,
2722 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2723 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_freed
,
2726 (void) printf(" %*s: object %llu, %llu blkptrs, "
2729 (u_longlong_t
)bpo
->bpo_object
,
2730 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2735 if (dump_opt
['d'] < 5)
2740 (void) bpobj_iterate_nofree(bpo
, dump_bpobj_cb
, NULL
, NULL
);
2741 (void) printf("\n");
2746 dump_bookmark(dsl_pool_t
*dp
, char *name
, boolean_t print_redact
,
2747 boolean_t print_list
)
2750 zfs_bookmark_phys_t prop
;
2751 objset_t
*mos
= dp
->dp_spa
->spa_meta_objset
;
2752 err
= dsl_bookmark_lookup(dp
, name
, NULL
, &prop
);
2758 (void) printf("\t#%s: ", strchr(name
, '#') + 1);
2759 (void) printf("{guid: %llx creation_txg: %llu creation_time: "
2760 "%llu redaction_obj: %llu}\n", (u_longlong_t
)prop
.zbm_guid
,
2761 (u_longlong_t
)prop
.zbm_creation_txg
,
2762 (u_longlong_t
)prop
.zbm_creation_time
,
2763 (u_longlong_t
)prop
.zbm_redaction_obj
);
2765 IMPLY(print_list
, print_redact
);
2766 if (!print_redact
|| prop
.zbm_redaction_obj
== 0)
2769 redaction_list_t
*rl
;
2770 VERIFY0(dsl_redaction_list_hold_obj(dp
,
2771 prop
.zbm_redaction_obj
, FTAG
, &rl
));
2773 redaction_list_phys_t
*rlp
= rl
->rl_phys
;
2774 (void) printf("\tRedacted:\n\t\tProgress: ");
2775 if (rlp
->rlp_last_object
!= UINT64_MAX
||
2776 rlp
->rlp_last_blkid
!= UINT64_MAX
) {
2777 (void) printf("%llu %llu (incomplete)\n",
2778 (u_longlong_t
)rlp
->rlp_last_object
,
2779 (u_longlong_t
)rlp
->rlp_last_blkid
);
2781 (void) printf("complete\n");
2783 (void) printf("\t\tSnapshots: [");
2784 for (unsigned int i
= 0; i
< rlp
->rlp_num_snaps
; i
++) {
2786 (void) printf(", ");
2787 (void) printf("%0llu",
2788 (u_longlong_t
)rlp
->rlp_snaps
[i
]);
2790 (void) printf("]\n\t\tLength: %llu\n",
2791 (u_longlong_t
)rlp
->rlp_num_entries
);
2794 dsl_redaction_list_rele(rl
, FTAG
);
2798 if (rlp
->rlp_num_entries
== 0) {
2799 dsl_redaction_list_rele(rl
, FTAG
);
2800 (void) printf("\t\tRedaction List: []\n\n");
2804 redact_block_phys_t
*rbp_buf
;
2806 dmu_object_info_t doi
;
2808 VERIFY0(dmu_object_info(mos
, prop
.zbm_redaction_obj
, &doi
));
2809 size
= doi
.doi_max_offset
;
2810 rbp_buf
= kmem_alloc(size
, KM_SLEEP
);
2812 err
= dmu_read(mos
, prop
.zbm_redaction_obj
, 0, size
,
2815 dsl_redaction_list_rele(rl
, FTAG
);
2816 kmem_free(rbp_buf
, size
);
2820 (void) printf("\t\tRedaction List: [{object: %llx, offset: "
2821 "%llx, blksz: %x, count: %llx}",
2822 (u_longlong_t
)rbp_buf
[0].rbp_object
,
2823 (u_longlong_t
)rbp_buf
[0].rbp_blkid
,
2824 (uint_t
)(redact_block_get_size(&rbp_buf
[0])),
2825 (u_longlong_t
)redact_block_get_count(&rbp_buf
[0]));
2827 for (size_t i
= 1; i
< rlp
->rlp_num_entries
; i
++) {
2828 (void) printf(",\n\t\t{object: %llx, offset: %llx, "
2829 "blksz: %x, count: %llx}",
2830 (u_longlong_t
)rbp_buf
[i
].rbp_object
,
2831 (u_longlong_t
)rbp_buf
[i
].rbp_blkid
,
2832 (uint_t
)(redact_block_get_size(&rbp_buf
[i
])),
2833 (u_longlong_t
)redact_block_get_count(&rbp_buf
[i
]));
2835 dsl_redaction_list_rele(rl
, FTAG
);
2836 kmem_free(rbp_buf
, size
);
2837 (void) printf("]\n\n");
2842 dump_bookmarks(objset_t
*os
, int verbosity
)
2845 zap_attribute_t attr
;
2846 dsl_dataset_t
*ds
= dmu_objset_ds(os
);
2847 dsl_pool_t
*dp
= spa_get_dsl(os
->os_spa
);
2848 objset_t
*mos
= os
->os_spa
->spa_meta_objset
;
2851 dsl_pool_config_enter(dp
, FTAG
);
2853 for (zap_cursor_init(&zc
, mos
, ds
->ds_bookmarks_obj
);
2854 zap_cursor_retrieve(&zc
, &attr
) == 0;
2855 zap_cursor_advance(&zc
)) {
2856 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
2857 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
2858 dmu_objset_name(os
, osname
);
2859 VERIFY3S(0, <=, snprintf(buf
, sizeof (buf
), "%s#%s", osname
,
2861 (void) dump_bookmark(dp
, buf
, verbosity
>= 5, verbosity
>= 6);
2863 zap_cursor_fini(&zc
);
2864 dsl_pool_config_exit(dp
, FTAG
);
2868 bpobj_count_refd(bpobj_t
*bpo
)
2870 mos_obj_refd(bpo
->bpo_object
);
2872 if (bpo
->bpo_havesubobj
&& bpo
->bpo_phys
->bpo_subobjs
!= 0) {
2873 mos_obj_refd(bpo
->bpo_phys
->bpo_subobjs
);
2874 for (uint64_t i
= 0; i
< bpo
->bpo_phys
->bpo_num_subobjs
; i
++) {
2878 VERIFY0(dmu_read(bpo
->bpo_os
,
2879 bpo
->bpo_phys
->bpo_subobjs
,
2880 i
* sizeof (subobj
), sizeof (subobj
), &subobj
, 0));
2881 error
= bpobj_open(&subbpo
, bpo
->bpo_os
, subobj
);
2883 (void) printf("ERROR %u while trying to open "
2885 error
, (u_longlong_t
)subobj
);
2888 bpobj_count_refd(&subbpo
);
2889 bpobj_close(&subbpo
);
2895 dsl_deadlist_entry_count_refd(void *arg
, dsl_deadlist_entry_t
*dle
)
2898 uint64_t empty_bpobj
= spa
->spa_dsl_pool
->dp_empty_bpobj
;
2899 if (dle
->dle_bpobj
.bpo_object
!= empty_bpobj
)
2900 bpobj_count_refd(&dle
->dle_bpobj
);
2905 dsl_deadlist_entry_dump(void *arg
, dsl_deadlist_entry_t
*dle
)
2907 ASSERT(arg
== NULL
);
2908 if (dump_opt
['d'] >= 5) {
2910 (void) snprintf(buf
, sizeof (buf
),
2911 "mintxg %llu -> obj %llu",
2912 (longlong_t
)dle
->dle_mintxg
,
2913 (longlong_t
)dle
->dle_bpobj
.bpo_object
);
2915 dump_full_bpobj(&dle
->dle_bpobj
, buf
, 0);
2917 (void) printf("mintxg %llu -> obj %llu\n",
2918 (longlong_t
)dle
->dle_mintxg
,
2919 (longlong_t
)dle
->dle_bpobj
.bpo_object
);
2925 dump_blkptr_list(dsl_deadlist_t
*dl
, const char *name
)
2931 spa_t
*spa
= dmu_objset_spa(dl
->dl_os
);
2932 uint64_t empty_bpobj
= spa
->spa_dsl_pool
->dp_empty_bpobj
;
2934 if (dl
->dl_oldfmt
) {
2935 if (dl
->dl_bpobj
.bpo_object
!= empty_bpobj
)
2936 bpobj_count_refd(&dl
->dl_bpobj
);
2938 mos_obj_refd(dl
->dl_object
);
2939 dsl_deadlist_iterate(dl
, dsl_deadlist_entry_count_refd
, spa
);
2942 /* make sure nicenum has enough space */
2943 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
2944 _Static_assert(sizeof (comp
) >= NN_NUMBUF_SZ
, "comp truncated");
2945 _Static_assert(sizeof (uncomp
) >= NN_NUMBUF_SZ
, "uncomp truncated");
2946 _Static_assert(sizeof (entries
) >= NN_NUMBUF_SZ
, "entries truncated");
2948 if (dump_opt
['d'] < 3)
2951 if (dl
->dl_oldfmt
) {
2952 dump_full_bpobj(&dl
->dl_bpobj
, "old-format deadlist", 0);
2956 zdb_nicenum(dl
->dl_phys
->dl_used
, bytes
, sizeof (bytes
));
2957 zdb_nicenum(dl
->dl_phys
->dl_comp
, comp
, sizeof (comp
));
2958 zdb_nicenum(dl
->dl_phys
->dl_uncomp
, uncomp
, sizeof (uncomp
));
2959 zdb_nicenum(avl_numnodes(&dl
->dl_tree
), entries
, sizeof (entries
));
2960 (void) printf("\n %s: %s (%s/%s comp), %s entries\n",
2961 name
, bytes
, comp
, uncomp
, entries
);
2963 if (dump_opt
['d'] < 4)
2966 (void) putchar('\n');
2968 dsl_deadlist_iterate(dl
, dsl_deadlist_entry_dump
, NULL
);
2972 verify_dd_livelist(objset_t
*os
)
2974 uint64_t ll_used
, used
, ll_comp
, comp
, ll_uncomp
, uncomp
;
2975 dsl_pool_t
*dp
= spa_get_dsl(os
->os_spa
);
2976 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
2978 ASSERT(!dmu_objset_is_snapshot(os
));
2979 if (!dsl_deadlist_is_open(&dd
->dd_livelist
))
2982 /* Iterate through the livelist to check for duplicates */
2983 dsl_deadlist_iterate(&dd
->dd_livelist
, sublivelist_verify_lightweight
,
2986 dsl_pool_config_enter(dp
, FTAG
);
2987 dsl_deadlist_space(&dd
->dd_livelist
, &ll_used
,
2988 &ll_comp
, &ll_uncomp
);
2990 dsl_dataset_t
*origin_ds
;
2991 ASSERT(dsl_pool_config_held(dp
));
2992 VERIFY0(dsl_dataset_hold_obj(dp
,
2993 dsl_dir_phys(dd
)->dd_origin_obj
, FTAG
, &origin_ds
));
2994 VERIFY0(dsl_dataset_space_written(origin_ds
, os
->os_dsl_dataset
,
2995 &used
, &comp
, &uncomp
));
2996 dsl_dataset_rele(origin_ds
, FTAG
);
2997 dsl_pool_config_exit(dp
, FTAG
);
2999 * It's possible that the dataset's uncomp space is larger than the
3000 * livelist's because livelists do not track embedded block pointers
3002 if (used
!= ll_used
|| comp
!= ll_comp
|| uncomp
< ll_uncomp
) {
3003 char nice_used
[32], nice_comp
[32], nice_uncomp
[32];
3004 (void) printf("Discrepancy in space accounting:\n");
3005 zdb_nicenum(used
, nice_used
, sizeof (nice_used
));
3006 zdb_nicenum(comp
, nice_comp
, sizeof (nice_comp
));
3007 zdb_nicenum(uncomp
, nice_uncomp
, sizeof (nice_uncomp
));
3008 (void) printf("dir: used %s, comp %s, uncomp %s\n",
3009 nice_used
, nice_comp
, nice_uncomp
);
3010 zdb_nicenum(ll_used
, nice_used
, sizeof (nice_used
));
3011 zdb_nicenum(ll_comp
, nice_comp
, sizeof (nice_comp
));
3012 zdb_nicenum(ll_uncomp
, nice_uncomp
, sizeof (nice_uncomp
));
3013 (void) printf("livelist: used %s, comp %s, uncomp %s\n",
3014 nice_used
, nice_comp
, nice_uncomp
);
3020 static avl_tree_t idx_tree
;
3021 static avl_tree_t domain_tree
;
3022 static boolean_t fuid_table_loaded
;
3023 static objset_t
*sa_os
= NULL
;
3024 static sa_attr_type_t
*sa_attr_table
= NULL
;
3027 open_objset(const char *path
, const void *tag
, objset_t
**osp
)
3030 uint64_t sa_attrs
= 0;
3031 uint64_t version
= 0;
3033 VERIFY3P(sa_os
, ==, NULL
);
3035 * We can't own an objset if it's redacted. Therefore, we do this
3036 * dance: hold the objset, then acquire a long hold on its dataset, then
3037 * release the pool (which is held as part of holding the objset).
3039 err
= dmu_objset_hold(path
, tag
, osp
);
3041 (void) fprintf(stderr
, "failed to hold dataset '%s': %s\n",
3042 path
, strerror(err
));
3045 dsl_dataset_long_hold(dmu_objset_ds(*osp
), tag
);
3046 dsl_pool_rele(dmu_objset_pool(*osp
), tag
);
3048 if (dmu_objset_type(*osp
) == DMU_OST_ZFS
&& !(*osp
)->os_encrypted
) {
3049 (void) zap_lookup(*osp
, MASTER_NODE_OBJ
, ZPL_VERSION_STR
,
3051 if (version
>= ZPL_VERSION_SA
) {
3052 (void) zap_lookup(*osp
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
,
3055 err
= sa_setup(*osp
, sa_attrs
, zfs_attr_table
, ZPL_END
,
3058 (void) fprintf(stderr
, "sa_setup failed: %s\n",
3060 dsl_dataset_long_rele(dmu_objset_ds(*osp
), tag
);
3061 dsl_dataset_rele(dmu_objset_ds(*osp
), tag
);
3071 close_objset(objset_t
*os
, const void *tag
)
3073 VERIFY3P(os
, ==, sa_os
);
3074 if (os
->os_sa
!= NULL
)
3076 dsl_dataset_long_rele(dmu_objset_ds(os
), tag
);
3077 dsl_dataset_rele(dmu_objset_ds(os
), tag
);
3078 sa_attr_table
= NULL
;
3083 fuid_table_destroy(void)
3085 if (fuid_table_loaded
) {
3086 zfs_fuid_table_destroy(&idx_tree
, &domain_tree
);
3087 fuid_table_loaded
= B_FALSE
;
3092 * print uid or gid information.
3093 * For normal POSIX id just the id is printed in decimal format.
3094 * For CIFS files with FUID the fuid is printed in hex followed by
3095 * the domain-rid string.
3098 print_idstr(uint64_t id
, const char *id_type
)
3100 if (FUID_INDEX(id
)) {
3101 const char *domain
=
3102 zfs_fuid_idx_domain(&idx_tree
, FUID_INDEX(id
));
3103 (void) printf("\t%s %llx [%s-%d]\n", id_type
,
3104 (u_longlong_t
)id
, domain
, (int)FUID_RID(id
));
3106 (void) printf("\t%s %llu\n", id_type
, (u_longlong_t
)id
);
3112 dump_uidgid(objset_t
*os
, uint64_t uid
, uint64_t gid
)
3114 uint32_t uid_idx
, gid_idx
;
3116 uid_idx
= FUID_INDEX(uid
);
3117 gid_idx
= FUID_INDEX(gid
);
3119 /* Load domain table, if not already loaded */
3120 if (!fuid_table_loaded
&& (uid_idx
|| gid_idx
)) {
3123 /* first find the fuid object. It lives in the master node */
3124 VERIFY(zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_FUID_TABLES
,
3125 8, 1, &fuid_obj
) == 0);
3126 zfs_fuid_avl_tree_create(&idx_tree
, &domain_tree
);
3127 (void) zfs_fuid_table_load(os
, fuid_obj
,
3128 &idx_tree
, &domain_tree
);
3129 fuid_table_loaded
= B_TRUE
;
3132 print_idstr(uid
, "uid");
3133 print_idstr(gid
, "gid");
3137 dump_znode_sa_xattr(sa_handle_t
*hdl
)
3140 nvpair_t
*elem
= NULL
;
3141 int sa_xattr_size
= 0;
3142 int sa_xattr_entries
= 0;
3144 char *sa_xattr_packed
;
3146 error
= sa_size(hdl
, sa_attr_table
[ZPL_DXATTR
], &sa_xattr_size
);
3147 if (error
|| sa_xattr_size
== 0)
3150 sa_xattr_packed
= malloc(sa_xattr_size
);
3151 if (sa_xattr_packed
== NULL
)
3154 error
= sa_lookup(hdl
, sa_attr_table
[ZPL_DXATTR
],
3155 sa_xattr_packed
, sa_xattr_size
);
3157 free(sa_xattr_packed
);
3161 error
= nvlist_unpack(sa_xattr_packed
, sa_xattr_size
, &sa_xattr
, 0);
3163 free(sa_xattr_packed
);
3167 while ((elem
= nvlist_next_nvpair(sa_xattr
, elem
)) != NULL
)
3170 (void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3171 sa_xattr_size
, sa_xattr_entries
);
3172 while ((elem
= nvlist_next_nvpair(sa_xattr
, elem
)) != NULL
) {
3176 (void) printf("\t\t%s = ", nvpair_name(elem
));
3177 nvpair_value_byte_array(elem
, &value
, &cnt
);
3178 for (idx
= 0; idx
< cnt
; ++idx
) {
3179 if (isprint(value
[idx
]))
3180 (void) putchar(value
[idx
]);
3182 (void) printf("\\%3.3o", value
[idx
]);
3184 (void) putchar('\n');
3187 nvlist_free(sa_xattr
);
3188 free(sa_xattr_packed
);
3192 dump_znode_symlink(sa_handle_t
*hdl
)
3194 int sa_symlink_size
= 0;
3195 char linktarget
[MAXPATHLEN
];
3198 error
= sa_size(hdl
, sa_attr_table
[ZPL_SYMLINK
], &sa_symlink_size
);
3199 if (error
|| sa_symlink_size
== 0) {
3202 if (sa_symlink_size
>= sizeof (linktarget
)) {
3203 (void) printf("symlink size %d is too large\n",
3207 linktarget
[sa_symlink_size
] = '\0';
3208 if (sa_lookup(hdl
, sa_attr_table
[ZPL_SYMLINK
],
3209 &linktarget
, sa_symlink_size
) == 0)
3210 (void) printf("\ttarget %s\n", linktarget
);
3214 dump_znode(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
3216 (void) data
, (void) size
;
3217 char path
[MAXPATHLEN
* 2]; /* allow for xattr and failure prefix */
3219 uint64_t xattr
, rdev
, gen
;
3220 uint64_t uid
, gid
, mode
, fsize
, parent
, links
;
3222 uint64_t acctm
[2], modtm
[2], chgtm
[2], crtm
[2];
3223 time_t z_crtime
, z_atime
, z_mtime
, z_ctime
;
3224 sa_bulk_attr_t bulk
[12];
3228 VERIFY3P(os
, ==, sa_os
);
3229 if (sa_handle_get(os
, object
, NULL
, SA_HDL_PRIVATE
, &hdl
)) {
3230 (void) printf("Failed to get handle for SA znode\n");
3234 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_UID
], NULL
, &uid
, 8);
3235 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_GID
], NULL
, &gid
, 8);
3236 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_LINKS
], NULL
,
3238 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_GEN
], NULL
, &gen
, 8);
3239 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_MODE
], NULL
,
3241 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_PARENT
],
3243 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_SIZE
], NULL
,
3245 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_ATIME
], NULL
,
3247 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_MTIME
], NULL
,
3249 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_CRTIME
], NULL
,
3251 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_CTIME
], NULL
,
3253 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_FLAGS
], NULL
,
3256 if (sa_bulk_lookup(hdl
, bulk
, idx
)) {
3257 (void) sa_handle_destroy(hdl
);
3261 z_crtime
= (time_t)crtm
[0];
3262 z_atime
= (time_t)acctm
[0];
3263 z_mtime
= (time_t)modtm
[0];
3264 z_ctime
= (time_t)chgtm
[0];
3266 if (dump_opt
['d'] > 4) {
3267 error
= zfs_obj_to_path(os
, object
, path
, sizeof (path
));
3268 if (error
== ESTALE
) {
3269 (void) snprintf(path
, sizeof (path
), "on delete queue");
3270 } else if (error
!= 0) {
3272 (void) snprintf(path
, sizeof (path
),
3273 "path not found, possibly leaked");
3275 (void) printf("\tpath %s\n", path
);
3279 dump_znode_symlink(hdl
);
3280 dump_uidgid(os
, uid
, gid
);
3281 (void) printf("\tatime %s", ctime(&z_atime
));
3282 (void) printf("\tmtime %s", ctime(&z_mtime
));
3283 (void) printf("\tctime %s", ctime(&z_ctime
));
3284 (void) printf("\tcrtime %s", ctime(&z_crtime
));
3285 (void) printf("\tgen %llu\n", (u_longlong_t
)gen
);
3286 (void) printf("\tmode %llo\n", (u_longlong_t
)mode
);
3287 (void) printf("\tsize %llu\n", (u_longlong_t
)fsize
);
3288 (void) printf("\tparent %llu\n", (u_longlong_t
)parent
);
3289 (void) printf("\tlinks %llu\n", (u_longlong_t
)links
);
3290 (void) printf("\tpflags %llx\n", (u_longlong_t
)pflags
);
3291 if (dmu_objset_projectquota_enabled(os
) && (pflags
& ZFS_PROJID
)) {
3294 if (sa_lookup(hdl
, sa_attr_table
[ZPL_PROJID
], &projid
,
3295 sizeof (uint64_t)) == 0)
3296 (void) printf("\tprojid %llu\n", (u_longlong_t
)projid
);
3298 if (sa_lookup(hdl
, sa_attr_table
[ZPL_XATTR
], &xattr
,
3299 sizeof (uint64_t)) == 0)
3300 (void) printf("\txattr %llu\n", (u_longlong_t
)xattr
);
3301 if (sa_lookup(hdl
, sa_attr_table
[ZPL_RDEV
], &rdev
,
3302 sizeof (uint64_t)) == 0)
3303 (void) printf("\trdev 0x%016llx\n", (u_longlong_t
)rdev
);
3304 dump_znode_sa_xattr(hdl
);
3305 sa_handle_destroy(hdl
);
3309 dump_acl(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
3311 (void) os
, (void) object
, (void) data
, (void) size
;
3315 dump_dmu_objset(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
3317 (void) os
, (void) object
, (void) data
, (void) size
;
3320 static object_viewer_t
*object_viewer
[DMU_OT_NUMTYPES
+ 1] = {
3321 dump_none
, /* unallocated */
3322 dump_zap
, /* object directory */
3323 dump_uint64
, /* object array */
3324 dump_none
, /* packed nvlist */
3325 dump_packed_nvlist
, /* packed nvlist size */
3326 dump_none
, /* bpobj */
3327 dump_bpobj
, /* bpobj header */
3328 dump_none
, /* SPA space map header */
3329 dump_none
, /* SPA space map */
3330 dump_none
, /* ZIL intent log */
3331 dump_dnode
, /* DMU dnode */
3332 dump_dmu_objset
, /* DMU objset */
3333 dump_dsl_dir
, /* DSL directory */
3334 dump_zap
, /* DSL directory child map */
3335 dump_zap
, /* DSL dataset snap map */
3336 dump_zap
, /* DSL props */
3337 dump_dsl_dataset
, /* DSL dataset */
3338 dump_znode
, /* ZFS znode */
3339 dump_acl
, /* ZFS V0 ACL */
3340 dump_uint8
, /* ZFS plain file */
3341 dump_zpldir
, /* ZFS directory */
3342 dump_zap
, /* ZFS master node */
3343 dump_zap
, /* ZFS delete queue */
3344 dump_uint8
, /* zvol object */
3345 dump_zap
, /* zvol prop */
3346 dump_uint8
, /* other uint8[] */
3347 dump_uint64
, /* other uint64[] */
3348 dump_zap
, /* other ZAP */
3349 dump_zap
, /* persistent error log */
3350 dump_uint8
, /* SPA history */
3351 dump_history_offsets
, /* SPA history offsets */
3352 dump_zap
, /* Pool properties */
3353 dump_zap
, /* DSL permissions */
3354 dump_acl
, /* ZFS ACL */
3355 dump_uint8
, /* ZFS SYSACL */
3356 dump_none
, /* FUID nvlist */
3357 dump_packed_nvlist
, /* FUID nvlist size */
3358 dump_zap
, /* DSL dataset next clones */
3359 dump_zap
, /* DSL scrub queue */
3360 dump_zap
, /* ZFS user/group/project used */
3361 dump_zap
, /* ZFS user/group/project quota */
3362 dump_zap
, /* snapshot refcount tags */
3363 dump_ddt_zap
, /* DDT ZAP object */
3364 dump_zap
, /* DDT statistics */
3365 dump_znode
, /* SA object */
3366 dump_zap
, /* SA Master Node */
3367 dump_sa_attrs
, /* SA attribute registration */
3368 dump_sa_layouts
, /* SA attribute layouts */
3369 dump_zap
, /* DSL scrub translations */
3370 dump_none
, /* fake dedup BP */
3371 dump_zap
, /* deadlist */
3372 dump_none
, /* deadlist hdr */
3373 dump_zap
, /* dsl clones */
3374 dump_bpobj_subobjs
, /* bpobj subobjs */
3375 dump_unknown
, /* Unknown type, must be last */
3379 match_object_type(dmu_object_type_t obj_type
, uint64_t flags
)
3381 boolean_t match
= B_TRUE
;
3384 case DMU_OT_DIRECTORY_CONTENTS
:
3385 if (!(flags
& ZOR_FLAG_DIRECTORY
))
3388 case DMU_OT_PLAIN_FILE_CONTENTS
:
3389 if (!(flags
& ZOR_FLAG_PLAIN_FILE
))
3392 case DMU_OT_SPACE_MAP
:
3393 if (!(flags
& ZOR_FLAG_SPACE_MAP
))
3397 if (strcmp(zdb_ot_name(obj_type
), "zap") == 0) {
3398 if (!(flags
& ZOR_FLAG_ZAP
))
3404 * If all bits except some of the supported flags are
3405 * set, the user combined the all-types flag (A) with
3406 * a negated flag to exclude some types (e.g. A-f to
3407 * show all object types except plain files).
3409 if ((flags
| ZOR_SUPPORTED_FLAGS
) != ZOR_FLAG_ALL_TYPES
)
3419 dump_object(objset_t
*os
, uint64_t object
, int verbosity
,
3420 boolean_t
*print_header
, uint64_t *dnode_slots_used
, uint64_t flags
)
3422 dmu_buf_t
*db
= NULL
;
3423 dmu_object_info_t doi
;
3425 boolean_t dnode_held
= B_FALSE
;
3428 char iblk
[32], dblk
[32], lsize
[32], asize
[32], fill
[32], dnsize
[32];
3429 char bonus_size
[32];
3433 /* make sure nicenum has enough space */
3434 _Static_assert(sizeof (iblk
) >= NN_NUMBUF_SZ
, "iblk truncated");
3435 _Static_assert(sizeof (dblk
) >= NN_NUMBUF_SZ
, "dblk truncated");
3436 _Static_assert(sizeof (lsize
) >= NN_NUMBUF_SZ
, "lsize truncated");
3437 _Static_assert(sizeof (asize
) >= NN_NUMBUF_SZ
, "asize truncated");
3438 _Static_assert(sizeof (bonus_size
) >= NN_NUMBUF_SZ
,
3439 "bonus_size truncated");
3441 if (*print_header
) {
3442 (void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n",
3443 "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3444 "lsize", "%full", "type");
3449 dn
= DMU_META_DNODE(os
);
3450 dmu_object_info_from_dnode(dn
, &doi
);
3453 * Encrypted datasets will have sensitive bonus buffers
3454 * encrypted. Therefore we cannot hold the bonus buffer and
3455 * must hold the dnode itself instead.
3457 error
= dmu_object_info(os
, object
, &doi
);
3459 fatal("dmu_object_info() failed, errno %u", error
);
3461 if (os
->os_encrypted
&&
3462 DMU_OT_IS_ENCRYPTED(doi
.doi_bonus_type
)) {
3463 error
= dnode_hold(os
, object
, FTAG
, &dn
);
3465 fatal("dnode_hold() failed, errno %u", error
);
3466 dnode_held
= B_TRUE
;
3468 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
3470 fatal("dmu_bonus_hold(%llu) failed, errno %u",
3472 bonus
= db
->db_data
;
3473 bsize
= db
->db_size
;
3474 dn
= DB_DNODE((dmu_buf_impl_t
*)db
);
3479 * Default to showing all object types if no flags were specified.
3481 if (flags
!= 0 && flags
!= ZOR_FLAG_ALL_TYPES
&&
3482 !match_object_type(doi
.doi_type
, flags
))
3485 if (dnode_slots_used
)
3486 *dnode_slots_used
= doi
.doi_dnodesize
/ DNODE_MIN_SIZE
;
3488 zdb_nicenum(doi
.doi_metadata_block_size
, iblk
, sizeof (iblk
));
3489 zdb_nicenum(doi
.doi_data_block_size
, dblk
, sizeof (dblk
));
3490 zdb_nicenum(doi
.doi_max_offset
, lsize
, sizeof (lsize
));
3491 zdb_nicenum(doi
.doi_physical_blocks_512
<< 9, asize
, sizeof (asize
));
3492 zdb_nicenum(doi
.doi_bonus_size
, bonus_size
, sizeof (bonus_size
));
3493 zdb_nicenum(doi
.doi_dnodesize
, dnsize
, sizeof (dnsize
));
3494 (void) sprintf(fill
, "%6.2f", 100.0 * doi
.doi_fill_count
*
3495 doi
.doi_data_block_size
/ (object
== 0 ? DNODES_PER_BLOCK
: 1) /
3496 doi
.doi_max_offset
);
3500 if (doi
.doi_checksum
!= ZIO_CHECKSUM_INHERIT
|| verbosity
>= 6) {
3501 (void) snprintf(aux
+ strlen(aux
), sizeof (aux
) - strlen(aux
),
3502 " (K=%s)", ZDB_CHECKSUM_NAME(doi
.doi_checksum
));
3505 if (doi
.doi_compress
== ZIO_COMPRESS_INHERIT
&&
3506 ZIO_COMPRESS_HASLEVEL(os
->os_compress
) && verbosity
>= 6) {
3507 const char *compname
= NULL
;
3508 if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION
,
3509 ZIO_COMPRESS_RAW(os
->os_compress
, os
->os_complevel
),
3511 (void) snprintf(aux
+ strlen(aux
),
3512 sizeof (aux
) - strlen(aux
), " (Z=inherit=%s)",
3515 (void) snprintf(aux
+ strlen(aux
),
3516 sizeof (aux
) - strlen(aux
),
3517 " (Z=inherit=%s-unknown)",
3518 ZDB_COMPRESS_NAME(os
->os_compress
));
3520 } else if (doi
.doi_compress
== ZIO_COMPRESS_INHERIT
&& verbosity
>= 6) {
3521 (void) snprintf(aux
+ strlen(aux
), sizeof (aux
) - strlen(aux
),
3522 " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os
->os_compress
));
3523 } else if (doi
.doi_compress
!= ZIO_COMPRESS_INHERIT
|| verbosity
>= 6) {
3524 (void) snprintf(aux
+ strlen(aux
), sizeof (aux
) - strlen(aux
),
3525 " (Z=%s)", ZDB_COMPRESS_NAME(doi
.doi_compress
));
3528 (void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n",
3529 (u_longlong_t
)object
, doi
.doi_indirection
, iblk
, dblk
,
3530 asize
, dnsize
, lsize
, fill
, zdb_ot_name(doi
.doi_type
), aux
);
3532 if (doi
.doi_bonus_type
!= DMU_OT_NONE
&& verbosity
> 3) {
3533 (void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n",
3534 "", "", "", "", "", "", bonus_size
, "bonus",
3535 zdb_ot_name(doi
.doi_bonus_type
));
3538 if (verbosity
>= 4) {
3539 (void) printf("\tdnode flags: %s%s%s%s\n",
3540 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_USED_BYTES
) ?
3542 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_USERUSED_ACCOUNTED
) ?
3543 "USERUSED_ACCOUNTED " : "",
3544 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_USEROBJUSED_ACCOUNTED
) ?
3545 "USEROBJUSED_ACCOUNTED " : "",
3546 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) ?
3547 "SPILL_BLKPTR" : "");
3548 (void) printf("\tdnode maxblkid: %llu\n",
3549 (longlong_t
)dn
->dn_phys
->dn_maxblkid
);
3552 object_viewer
[ZDB_OT_TYPE(doi
.doi_bonus_type
)](os
,
3553 object
, bonus
, bsize
);
3555 (void) printf("\t\t(bonus encrypted)\n");
3558 if (!os
->os_encrypted
|| !DMU_OT_IS_ENCRYPTED(doi
.doi_type
)) {
3559 object_viewer
[ZDB_OT_TYPE(doi
.doi_type
)](os
, object
,
3562 (void) printf("\t\t(object encrypted)\n");
3565 *print_header
= B_TRUE
;
3568 if (verbosity
>= 5) {
3569 if (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
3570 char blkbuf
[BP_SPRINTF_LEN
];
3571 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
),
3572 DN_SPILL_BLKPTR(dn
->dn_phys
), B_FALSE
);
3573 (void) printf("\nSpill block: %s\n", blkbuf
);
3578 if (verbosity
>= 5) {
3580 * Report the list of segments that comprise the object.
3584 uint64_t blkfill
= 1;
3587 if (dn
->dn_type
== DMU_OT_DNODE
) {
3589 blkfill
= DNODES_PER_BLOCK
;
3594 /* make sure nicenum has enough space */
3595 _Static_assert(sizeof (segsize
) >= NN_NUMBUF_SZ
,
3596 "segsize truncated");
3597 error
= dnode_next_offset(dn
,
3598 0, &start
, minlvl
, blkfill
, 0);
3602 error
= dnode_next_offset(dn
,
3603 DNODE_FIND_HOLE
, &end
, minlvl
, blkfill
, 0);
3604 zdb_nicenum(end
- start
, segsize
, sizeof (segsize
));
3605 (void) printf("\t\tsegment [%016llx, %016llx)"
3606 " size %5s\n", (u_longlong_t
)start
,
3607 (u_longlong_t
)end
, segsize
);
3616 dmu_buf_rele(db
, FTAG
);
3618 dnode_rele(dn
, FTAG
);
3622 count_dir_mos_objects(dsl_dir_t
*dd
)
3624 mos_obj_refd(dd
->dd_object
);
3625 mos_obj_refd(dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
3626 mos_obj_refd(dsl_dir_phys(dd
)->dd_deleg_zapobj
);
3627 mos_obj_refd(dsl_dir_phys(dd
)->dd_props_zapobj
);
3628 mos_obj_refd(dsl_dir_phys(dd
)->dd_clones
);
3631 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
3632 * Ignore the references after the first one.
3634 mos_obj_refd_multiple(dd
->dd_crypto_obj
);
3638 count_ds_mos_objects(dsl_dataset_t
*ds
)
3640 mos_obj_refd(ds
->ds_object
);
3641 mos_obj_refd(dsl_dataset_phys(ds
)->ds_next_clones_obj
);
3642 mos_obj_refd(dsl_dataset_phys(ds
)->ds_props_obj
);
3643 mos_obj_refd(dsl_dataset_phys(ds
)->ds_userrefs_obj
);
3644 mos_obj_refd(dsl_dataset_phys(ds
)->ds_snapnames_zapobj
);
3645 mos_obj_refd(ds
->ds_bookmarks_obj
);
3647 if (!dsl_dataset_is_snapshot(ds
)) {
3648 count_dir_mos_objects(ds
->ds_dir
);
3652 static const char *const objset_types
[DMU_OST_NUMTYPES
] = {
3653 "NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
3656 * Parse a string denoting a range of object IDs of the form
3657 * <start>[:<end>[:flags]], and store the results in zor.
3658 * Return 0 on success. On error, return 1 and update the msg
3659 * pointer to point to a descriptive error message.
3662 parse_object_range(char *range
, zopt_object_range_t
*zor
, const char **msg
)
3665 char *p
, *s
, *dup
, *flagstr
, *tmp
= NULL
;
3670 if (strchr(range
, ':') == NULL
) {
3671 zor
->zor_obj_start
= strtoull(range
, &p
, 0);
3673 *msg
= "Invalid characters in object ID";
3676 zor
->zor_obj_start
= ZDB_MAP_OBJECT_ID(zor
->zor_obj_start
);
3677 zor
->zor_obj_end
= zor
->zor_obj_start
;
3681 if (strchr(range
, ':') == range
) {
3682 *msg
= "Invalid leading colon";
3687 len
= strlen(range
);
3688 if (range
[len
- 1] == ':') {
3689 *msg
= "Invalid trailing colon";
3694 dup
= strdup(range
);
3695 s
= strtok_r(dup
, ":", &tmp
);
3696 zor
->zor_obj_start
= strtoull(s
, &p
, 0);
3699 *msg
= "Invalid characters in start object ID";
3704 s
= strtok_r(NULL
, ":", &tmp
);
3705 zor
->zor_obj_end
= strtoull(s
, &p
, 0);
3708 *msg
= "Invalid characters in end object ID";
3713 if (zor
->zor_obj_start
> zor
->zor_obj_end
) {
3714 *msg
= "Start object ID may not exceed end object ID";
3719 s
= strtok_r(NULL
, ":", &tmp
);
3721 zor
->zor_flags
= ZOR_FLAG_ALL_TYPES
;
3723 } else if (strtok_r(NULL
, ":", &tmp
) != NULL
) {
3724 *msg
= "Invalid colon-delimited field after flags";
3730 for (i
= 0; flagstr
[i
]; i
++) {
3732 boolean_t negation
= (flagstr
[i
] == '-');
3736 if (flagstr
[i
] == '\0') {
3737 *msg
= "Invalid trailing negation operator";
3742 bit
= flagbits
[(uchar_t
)flagstr
[i
]];
3744 *msg
= "Invalid flag";
3753 zor
->zor_flags
= flags
;
3755 zor
->zor_obj_start
= ZDB_MAP_OBJECT_ID(zor
->zor_obj_start
);
3756 zor
->zor_obj_end
= ZDB_MAP_OBJECT_ID(zor
->zor_obj_end
);
3764 dump_objset(objset_t
*os
)
3766 dmu_objset_stats_t dds
= { 0 };
3767 uint64_t object
, object_count
;
3768 uint64_t refdbytes
, usedobjs
, scratch
;
3770 char blkbuf
[BP_SPRINTF_LEN
+ 20];
3771 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
3772 const char *type
= "UNKNOWN";
3773 int verbosity
= dump_opt
['d'];
3774 boolean_t print_header
;
3777 uint64_t total_slots_used
= 0;
3778 uint64_t max_slot_used
= 0;
3779 uint64_t dnode_slots
;
3784 /* make sure nicenum has enough space */
3785 _Static_assert(sizeof (numbuf
) >= NN_NUMBUF_SZ
, "numbuf truncated");
3787 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
3788 dmu_objset_fast_stat(os
, &dds
);
3789 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
3791 print_header
= B_TRUE
;
3793 if (dds
.dds_type
< DMU_OST_NUMTYPES
)
3794 type
= objset_types
[dds
.dds_type
];
3796 if (dds
.dds_type
== DMU_OST_META
) {
3797 dds
.dds_creation_txg
= TXG_INITIAL
;
3798 usedobjs
= BP_GET_FILL(os
->os_rootbp
);
3799 refdbytes
= dsl_dir_phys(os
->os_spa
->spa_dsl_pool
->dp_mos_dir
)->
3802 dmu_objset_space(os
, &refdbytes
, &scratch
, &usedobjs
, &scratch
);
3805 ASSERT3U(usedobjs
, ==, BP_GET_FILL(os
->os_rootbp
));
3807 zdb_nicenum(refdbytes
, numbuf
, sizeof (numbuf
));
3809 if (verbosity
>= 4) {
3810 (void) snprintf(blkbuf
, sizeof (blkbuf
), ", rootbp ");
3811 (void) snprintf_blkptr(blkbuf
+ strlen(blkbuf
),
3812 sizeof (blkbuf
) - strlen(blkbuf
), os
->os_rootbp
);
3817 dmu_objset_name(os
, osname
);
3819 (void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
3820 "%s, %llu objects%s%s\n",
3821 osname
, type
, (u_longlong_t
)dmu_objset_id(os
),
3822 (u_longlong_t
)dds
.dds_creation_txg
,
3823 numbuf
, (u_longlong_t
)usedobjs
, blkbuf
,
3824 (dds
.dds_inconsistent
) ? " (inconsistent)" : "");
3826 for (i
= 0; i
< zopt_object_args
; i
++) {
3827 obj_start
= zopt_object_ranges
[i
].zor_obj_start
;
3828 obj_end
= zopt_object_ranges
[i
].zor_obj_end
;
3829 flags
= zopt_object_ranges
[i
].zor_flags
;
3832 if (object
== 0 || obj_start
== obj_end
)
3833 dump_object(os
, object
, verbosity
, &print_header
, NULL
,
3838 while ((dmu_object_next(os
, &object
, B_FALSE
, 0) == 0) &&
3839 object
<= obj_end
) {
3840 dump_object(os
, object
, verbosity
, &print_header
, NULL
,
3845 if (zopt_object_args
> 0) {
3846 (void) printf("\n");
3850 if (dump_opt
['i'] != 0 || verbosity
>= 2)
3851 dump_intent_log(dmu_objset_zil(os
));
3853 if (dmu_objset_ds(os
) != NULL
) {
3854 dsl_dataset_t
*ds
= dmu_objset_ds(os
);
3855 dump_blkptr_list(&ds
->ds_deadlist
, "Deadlist");
3856 if (dsl_deadlist_is_open(&ds
->ds_dir
->dd_livelist
) &&
3857 !dmu_objset_is_snapshot(os
)) {
3858 dump_blkptr_list(&ds
->ds_dir
->dd_livelist
, "Livelist");
3859 if (verify_dd_livelist(os
) != 0)
3860 fatal("livelist is incorrect");
3863 if (dsl_dataset_remap_deadlist_exists(ds
)) {
3864 (void) printf("ds_remap_deadlist:\n");
3865 dump_blkptr_list(&ds
->ds_remap_deadlist
, "Deadlist");
3867 count_ds_mos_objects(ds
);
3870 if (dmu_objset_ds(os
) != NULL
)
3871 dump_bookmarks(os
, verbosity
);
3876 if (BP_IS_HOLE(os
->os_rootbp
))
3879 dump_object(os
, 0, verbosity
, &print_header
, NULL
, 0);
3881 if (DMU_USERUSED_DNODE(os
) != NULL
&&
3882 DMU_USERUSED_DNODE(os
)->dn_type
!= 0) {
3883 dump_object(os
, DMU_USERUSED_OBJECT
, verbosity
, &print_header
,
3885 dump_object(os
, DMU_GROUPUSED_OBJECT
, verbosity
, &print_header
,
3889 if (DMU_PROJECTUSED_DNODE(os
) != NULL
&&
3890 DMU_PROJECTUSED_DNODE(os
)->dn_type
!= 0)
3891 dump_object(os
, DMU_PROJECTUSED_OBJECT
, verbosity
,
3892 &print_header
, NULL
, 0);
3895 while ((error
= dmu_object_next(os
, &object
, B_FALSE
, 0)) == 0) {
3896 dump_object(os
, object
, verbosity
, &print_header
, &dnode_slots
,
3899 total_slots_used
+= dnode_slots
;
3900 max_slot_used
= object
+ dnode_slots
- 1;
3903 (void) printf("\n");
3905 (void) printf(" Dnode slots:\n");
3906 (void) printf("\tTotal used: %10llu\n",
3907 (u_longlong_t
)total_slots_used
);
3908 (void) printf("\tMax used: %10llu\n",
3909 (u_longlong_t
)max_slot_used
);
3910 (void) printf("\tPercent empty: %10lf\n",
3911 (double)(max_slot_used
- total_slots_used
)*100 /
3912 (double)max_slot_used
);
3913 (void) printf("\n");
3915 if (error
!= ESRCH
) {
3916 (void) fprintf(stderr
, "dmu_object_next() = %d\n", error
);
3920 ASSERT3U(object_count
, ==, usedobjs
);
3922 if (leaked_objects
!= 0) {
3923 (void) printf("%d potentially leaked objects detected\n",
3930 dump_uberblock(uberblock_t
*ub
, const char *header
, const char *footer
)
3932 time_t timestamp
= ub
->ub_timestamp
;
3934 (void) printf("%s", header
? header
: "");
3935 (void) printf("\tmagic = %016llx\n", (u_longlong_t
)ub
->ub_magic
);
3936 (void) printf("\tversion = %llu\n", (u_longlong_t
)ub
->ub_version
);
3937 (void) printf("\ttxg = %llu\n", (u_longlong_t
)ub
->ub_txg
);
3938 (void) printf("\tguid_sum = %llu\n", (u_longlong_t
)ub
->ub_guid_sum
);
3939 (void) printf("\ttimestamp = %llu UTC = %s",
3940 (u_longlong_t
)ub
->ub_timestamp
, ctime(×tamp
));
3942 (void) printf("\tmmp_magic = %016llx\n",
3943 (u_longlong_t
)ub
->ub_mmp_magic
);
3944 if (MMP_VALID(ub
)) {
3945 (void) printf("\tmmp_delay = %0llu\n",
3946 (u_longlong_t
)ub
->ub_mmp_delay
);
3947 if (MMP_SEQ_VALID(ub
))
3948 (void) printf("\tmmp_seq = %u\n",
3949 (unsigned int) MMP_SEQ(ub
));
3950 if (MMP_FAIL_INT_VALID(ub
))
3951 (void) printf("\tmmp_fail = %u\n",
3952 (unsigned int) MMP_FAIL_INT(ub
));
3953 if (MMP_INTERVAL_VALID(ub
))
3954 (void) printf("\tmmp_write = %u\n",
3955 (unsigned int) MMP_INTERVAL(ub
));
3956 /* After MMP_* to make summarize_uberblock_mmp cleaner */
3957 (void) printf("\tmmp_valid = %x\n",
3958 (unsigned int) ub
->ub_mmp_config
& 0xFF);
3961 if (dump_opt
['u'] >= 4) {
3962 char blkbuf
[BP_SPRINTF_LEN
];
3963 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), &ub
->ub_rootbp
);
3964 (void) printf("\trootbp = %s\n", blkbuf
);
3966 (void) printf("\tcheckpoint_txg = %llu\n",
3967 (u_longlong_t
)ub
->ub_checkpoint_txg
);
3968 (void) printf("%s", footer
? footer
: "");
3972 dump_config(spa_t
*spa
)
3979 error
= dmu_bonus_hold(spa
->spa_meta_objset
,
3980 spa
->spa_config_object
, FTAG
, &db
);
3983 nvsize
= *(uint64_t *)db
->db_data
;
3984 dmu_buf_rele(db
, FTAG
);
3986 (void) printf("\nMOS Configuration:\n");
3987 dump_packed_nvlist(spa
->spa_meta_objset
,
3988 spa
->spa_config_object
, (void *)&nvsize
, 1);
3990 (void) fprintf(stderr
, "dmu_bonus_hold(%llu) failed, errno %d",
3991 (u_longlong_t
)spa
->spa_config_object
, error
);
3996 dump_cachefile(const char *cachefile
)
3999 struct stat64 statbuf
;
4003 if ((fd
= open64(cachefile
, O_RDONLY
)) < 0) {
4004 (void) printf("cannot open '%s': %s\n", cachefile
,
4009 if (fstat64(fd
, &statbuf
) != 0) {
4010 (void) printf("failed to stat '%s': %s\n", cachefile
,
4015 if ((buf
= malloc(statbuf
.st_size
)) == NULL
) {
4016 (void) fprintf(stderr
, "failed to allocate %llu bytes\n",
4017 (u_longlong_t
)statbuf
.st_size
);
4021 if (read(fd
, buf
, statbuf
.st_size
) != statbuf
.st_size
) {
4022 (void) fprintf(stderr
, "failed to read %llu bytes\n",
4023 (u_longlong_t
)statbuf
.st_size
);
4029 if (nvlist_unpack(buf
, statbuf
.st_size
, &config
, 0) != 0) {
4030 (void) fprintf(stderr
, "failed to unpack nvlist\n");
4036 dump_nvlist(config
, 0);
4038 nvlist_free(config
);
4042 * ZFS label nvlist stats
4044 typedef struct zdb_nvl_stats
{
4047 size_t zns_leaf_largest
;
4048 size_t zns_leaf_total
;
4049 nvlist_t
*zns_string
;
4050 nvlist_t
*zns_uint64
;
4051 nvlist_t
*zns_boolean
;
4055 collect_nvlist_stats(nvlist_t
*nvl
, zdb_nvl_stats_t
*stats
)
4057 nvlist_t
*list
, **array
;
4058 nvpair_t
*nvp
= NULL
;
4062 stats
->zns_list_count
++;
4064 while ((nvp
= nvlist_next_nvpair(nvl
, nvp
)) != NULL
) {
4065 name
= nvpair_name(nvp
);
4067 switch (nvpair_type(nvp
)) {
4068 case DATA_TYPE_STRING
:
4069 fnvlist_add_string(stats
->zns_string
, name
,
4070 fnvpair_value_string(nvp
));
4072 case DATA_TYPE_UINT64
:
4073 fnvlist_add_uint64(stats
->zns_uint64
, name
,
4074 fnvpair_value_uint64(nvp
));
4076 case DATA_TYPE_BOOLEAN
:
4077 fnvlist_add_boolean(stats
->zns_boolean
, name
);
4079 case DATA_TYPE_NVLIST
:
4080 if (nvpair_value_nvlist(nvp
, &list
) == 0)
4081 collect_nvlist_stats(list
, stats
);
4083 case DATA_TYPE_NVLIST_ARRAY
:
4084 if (nvpair_value_nvlist_array(nvp
, &array
, &items
) != 0)
4087 for (i
= 0; i
< items
; i
++) {
4088 collect_nvlist_stats(array
[i
], stats
);
4090 /* collect stats on leaf vdev */
4091 if (strcmp(name
, "children") == 0) {
4094 (void) nvlist_size(array
[i
], &size
,
4096 stats
->zns_leaf_total
+= size
;
4097 if (size
> stats
->zns_leaf_largest
)
4098 stats
->zns_leaf_largest
= size
;
4099 stats
->zns_leaf_count
++;
4104 (void) printf("skip type %d!\n", (int)nvpair_type(nvp
));
4110 dump_nvlist_stats(nvlist_t
*nvl
, size_t cap
)
4112 zdb_nvl_stats_t stats
= { 0 };
4113 size_t size
, sum
= 0, total
;
4116 /* requires nvlist with non-unique names for stat collection */
4117 VERIFY0(nvlist_alloc(&stats
.zns_string
, 0, 0));
4118 VERIFY0(nvlist_alloc(&stats
.zns_uint64
, 0, 0));
4119 VERIFY0(nvlist_alloc(&stats
.zns_boolean
, 0, 0));
4120 VERIFY0(nvlist_size(stats
.zns_boolean
, &noise
, NV_ENCODE_XDR
));
4122 (void) printf("\n\nZFS Label NVList Config Stats:\n");
4124 VERIFY0(nvlist_size(nvl
, &total
, NV_ENCODE_XDR
));
4125 (void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4126 (int)total
, (int)(cap
- total
), 100.0 * total
/ cap
);
4128 collect_nvlist_stats(nvl
, &stats
);
4130 VERIFY0(nvlist_size(stats
.zns_uint64
, &size
, NV_ENCODE_XDR
));
4133 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4134 (int)fnvlist_num_pairs(stats
.zns_uint64
),
4135 (int)size
, 100.0 * size
/ total
);
4137 VERIFY0(nvlist_size(stats
.zns_string
, &size
, NV_ENCODE_XDR
));
4140 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4141 (int)fnvlist_num_pairs(stats
.zns_string
),
4142 (int)size
, 100.0 * size
/ total
);
4144 VERIFY0(nvlist_size(stats
.zns_boolean
, &size
, NV_ENCODE_XDR
));
4147 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4148 (int)fnvlist_num_pairs(stats
.zns_boolean
),
4149 (int)size
, 100.0 * size
/ total
);
4151 size
= total
- sum
; /* treat remainder as nvlist overhead */
4152 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4153 stats
.zns_list_count
, (int)size
, 100.0 * size
/ total
);
4155 if (stats
.zns_leaf_count
> 0) {
4156 size_t average
= stats
.zns_leaf_total
/ stats
.zns_leaf_count
;
4158 (void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4159 stats
.zns_leaf_count
, (int)average
);
4160 (void) printf("%24d bytes largest\n",
4161 (int)stats
.zns_leaf_largest
);
4163 if (dump_opt
['l'] >= 3 && average
> 0)
4164 (void) printf(" space for %d additional leaf vdevs\n",
4165 (int)((cap
- total
) / average
));
4167 (void) printf("\n");
4169 nvlist_free(stats
.zns_string
);
4170 nvlist_free(stats
.zns_uint64
);
4171 nvlist_free(stats
.zns_boolean
);
4174 typedef struct cksum_record
{
4176 boolean_t labels
[VDEV_LABELS
];
4181 cksum_record_compare(const void *x1
, const void *x2
)
4183 const cksum_record_t
*l
= (cksum_record_t
*)x1
;
4184 const cksum_record_t
*r
= (cksum_record_t
*)x2
;
4185 int arraysize
= ARRAY_SIZE(l
->cksum
.zc_word
);
4188 for (int i
= 0; i
< arraysize
; i
++) {
4189 difference
= TREE_CMP(l
->cksum
.zc_word
[i
], r
->cksum
.zc_word
[i
]);
4194 return (difference
);
4197 static cksum_record_t
*
4198 cksum_record_alloc(zio_cksum_t
*cksum
, int l
)
4200 cksum_record_t
*rec
;
4202 rec
= umem_zalloc(sizeof (*rec
), UMEM_NOFAIL
);
4203 rec
->cksum
= *cksum
;
4204 rec
->labels
[l
] = B_TRUE
;
4209 static cksum_record_t
*
4210 cksum_record_lookup(avl_tree_t
*tree
, zio_cksum_t
*cksum
)
4212 cksum_record_t lookup
= { .cksum
= *cksum
};
4215 return (avl_find(tree
, &lookup
, &where
));
4218 static cksum_record_t
*
4219 cksum_record_insert(avl_tree_t
*tree
, zio_cksum_t
*cksum
, int l
)
4221 cksum_record_t
*rec
;
4223 rec
= cksum_record_lookup(tree
, cksum
);
4225 rec
->labels
[l
] = B_TRUE
;
4227 rec
= cksum_record_alloc(cksum
, l
);
4235 first_label(cksum_record_t
*rec
)
4237 for (int i
= 0; i
< VDEV_LABELS
; i
++)
4245 print_label_numbers(const char *prefix
, const cksum_record_t
*rec
)
4247 fputs(prefix
, stdout
);
4248 for (int i
= 0; i
< VDEV_LABELS
; i
++)
4249 if (rec
->labels
[i
] == B_TRUE
)
4254 #define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4256 typedef struct zdb_label
{
4258 uint64_t label_offset
;
4259 nvlist_t
*config_nv
;
4260 cksum_record_t
*config
;
4261 cksum_record_t
*uberblocks
[MAX_UBERBLOCK_COUNT
];
4262 boolean_t header_printed
;
4263 boolean_t read_failed
;
4264 boolean_t cksum_valid
;
4268 print_label_header(zdb_label_t
*label
, int l
)
4274 if (label
->header_printed
== B_TRUE
)
4277 (void) printf("------------------------------------\n");
4278 (void) printf("LABEL %d %s\n", l
,
4279 label
->cksum_valid
? "" : "(Bad label cksum)");
4280 (void) printf("------------------------------------\n");
4282 label
->header_printed
= B_TRUE
;
4286 print_l2arc_header(void)
4288 (void) printf("------------------------------------\n");
4289 (void) printf("L2ARC device header\n");
4290 (void) printf("------------------------------------\n");
4294 print_l2arc_log_blocks(void)
4296 (void) printf("------------------------------------\n");
4297 (void) printf("L2ARC device log blocks\n");
4298 (void) printf("------------------------------------\n");
4302 dump_l2arc_log_entries(uint64_t log_entries
,
4303 l2arc_log_ent_phys_t
*le
, uint64_t i
)
4305 for (int j
= 0; j
< log_entries
; j
++) {
4306 dva_t dva
= le
[j
].le_dva
;
4307 (void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4308 "vdev: %llu, offset: %llu\n",
4309 (u_longlong_t
)i
, j
+ 1,
4310 (u_longlong_t
)DVA_GET_ASIZE(&dva
),
4311 (u_longlong_t
)DVA_GET_VDEV(&dva
),
4312 (u_longlong_t
)DVA_GET_OFFSET(&dva
));
4313 (void) printf("|\t\t\t\tbirth: %llu\n",
4314 (u_longlong_t
)le
[j
].le_birth
);
4315 (void) printf("|\t\t\t\tlsize: %llu\n",
4316 (u_longlong_t
)L2BLK_GET_LSIZE((&le
[j
])->le_prop
));
4317 (void) printf("|\t\t\t\tpsize: %llu\n",
4318 (u_longlong_t
)L2BLK_GET_PSIZE((&le
[j
])->le_prop
));
4319 (void) printf("|\t\t\t\tcompr: %llu\n",
4320 (u_longlong_t
)L2BLK_GET_COMPRESS((&le
[j
])->le_prop
));
4321 (void) printf("|\t\t\t\tcomplevel: %llu\n",
4322 (u_longlong_t
)(&le
[j
])->le_complevel
);
4323 (void) printf("|\t\t\t\ttype: %llu\n",
4324 (u_longlong_t
)L2BLK_GET_TYPE((&le
[j
])->le_prop
));
4325 (void) printf("|\t\t\t\tprotected: %llu\n",
4326 (u_longlong_t
)L2BLK_GET_PROTECTED((&le
[j
])->le_prop
));
4327 (void) printf("|\t\t\t\tprefetch: %llu\n",
4328 (u_longlong_t
)L2BLK_GET_PREFETCH((&le
[j
])->le_prop
));
4329 (void) printf("|\t\t\t\taddress: %llu\n",
4330 (u_longlong_t
)le
[j
].le_daddr
);
4331 (void) printf("|\t\t\t\tARC state: %llu\n",
4332 (u_longlong_t
)L2BLK_GET_STATE((&le
[j
])->le_prop
));
4333 (void) printf("|\n");
4335 (void) printf("\n");
4339 dump_l2arc_log_blkptr(l2arc_log_blkptr_t lbps
)
4341 (void) printf("|\t\tdaddr: %llu\n", (u_longlong_t
)lbps
.lbp_daddr
);
4342 (void) printf("|\t\tpayload_asize: %llu\n",
4343 (u_longlong_t
)lbps
.lbp_payload_asize
);
4344 (void) printf("|\t\tpayload_start: %llu\n",
4345 (u_longlong_t
)lbps
.lbp_payload_start
);
4346 (void) printf("|\t\tlsize: %llu\n",
4347 (u_longlong_t
)L2BLK_GET_LSIZE((&lbps
)->lbp_prop
));
4348 (void) printf("|\t\tasize: %llu\n",
4349 (u_longlong_t
)L2BLK_GET_PSIZE((&lbps
)->lbp_prop
));
4350 (void) printf("|\t\tcompralgo: %llu\n",
4351 (u_longlong_t
)L2BLK_GET_COMPRESS((&lbps
)->lbp_prop
));
4352 (void) printf("|\t\tcksumalgo: %llu\n",
4353 (u_longlong_t
)L2BLK_GET_CHECKSUM((&lbps
)->lbp_prop
));
4354 (void) printf("|\n\n");
4358 dump_l2arc_log_blocks(int fd
, l2arc_dev_hdr_phys_t l2dhdr
,
4359 l2arc_dev_hdr_phys_t
*rebuild
)
4361 l2arc_log_blk_phys_t this_lb
;
4363 l2arc_log_blkptr_t lbps
[2];
4370 print_l2arc_log_blocks();
4371 memcpy(lbps
, l2dhdr
.dh_start_lbps
, sizeof (lbps
));
4373 dev
.l2ad_evict
= l2dhdr
.dh_evict
;
4374 dev
.l2ad_start
= l2dhdr
.dh_start
;
4375 dev
.l2ad_end
= l2dhdr
.dh_end
;
4377 if (l2dhdr
.dh_start_lbps
[0].lbp_daddr
== 0) {
4378 /* no log blocks to read */
4379 if (!dump_opt
['q']) {
4380 (void) printf("No log blocks to read\n");
4381 (void) printf("\n");
4385 dev
.l2ad_hand
= lbps
[0].lbp_daddr
+
4386 L2BLK_GET_PSIZE((&lbps
[0])->lbp_prop
);
4389 dev
.l2ad_first
= !!(l2dhdr
.dh_flags
& L2ARC_DEV_HDR_EVICT_FIRST
);
4392 if (!l2arc_log_blkptr_valid(&dev
, &lbps
[0]))
4395 /* L2BLK_GET_PSIZE returns aligned size for log blocks */
4396 asize
= L2BLK_GET_PSIZE((&lbps
[0])->lbp_prop
);
4397 if (pread64(fd
, &this_lb
, asize
, lbps
[0].lbp_daddr
) != asize
) {
4398 if (!dump_opt
['q']) {
4399 (void) printf("Error while reading next log "
4405 fletcher_4_native_varsize(&this_lb
, asize
, &cksum
);
4406 if (!ZIO_CHECKSUM_EQUAL(cksum
, lbps
[0].lbp_cksum
)) {
4408 if (!dump_opt
['q']) {
4409 (void) printf("Invalid cksum\n");
4410 dump_l2arc_log_blkptr(lbps
[0]);
4415 switch (L2BLK_GET_COMPRESS((&lbps
[0])->lbp_prop
)) {
4416 case ZIO_COMPRESS_OFF
:
4419 abd
= abd_alloc_for_io(asize
, B_TRUE
);
4420 abd_copy_from_buf_off(abd
, &this_lb
, 0, asize
);
4421 zio_decompress_data(L2BLK_GET_COMPRESS(
4422 (&lbps
[0])->lbp_prop
), abd
, &this_lb
,
4423 asize
, sizeof (this_lb
), NULL
);
4428 if (this_lb
.lb_magic
== BSWAP_64(L2ARC_LOG_BLK_MAGIC
))
4429 byteswap_uint64_array(&this_lb
, sizeof (this_lb
));
4430 if (this_lb
.lb_magic
!= L2ARC_LOG_BLK_MAGIC
) {
4432 (void) printf("Invalid log block magic\n\n");
4436 rebuild
->dh_lb_count
++;
4437 rebuild
->dh_lb_asize
+= asize
;
4438 if (dump_opt
['l'] > 1 && !dump_opt
['q']) {
4439 (void) printf("lb[%4llu]\tmagic: %llu\n",
4440 (u_longlong_t
)rebuild
->dh_lb_count
,
4441 (u_longlong_t
)this_lb
.lb_magic
);
4442 dump_l2arc_log_blkptr(lbps
[0]);
4445 if (dump_opt
['l'] > 2 && !dump_opt
['q'])
4446 dump_l2arc_log_entries(l2dhdr
.dh_log_entries
,
4448 rebuild
->dh_lb_count
);
4450 if (l2arc_range_check_overlap(lbps
[1].lbp_payload_start
,
4451 lbps
[0].lbp_payload_start
, dev
.l2ad_evict
) &&
4456 lbps
[1] = this_lb
.lb_prev_lbp
;
4459 if (!dump_opt
['q']) {
4460 (void) printf("log_blk_count:\t %llu with valid cksum\n",
4461 (u_longlong_t
)rebuild
->dh_lb_count
);
4462 (void) printf("\t\t %d with invalid cksum\n", failed
);
4463 (void) printf("log_blk_asize:\t %llu\n\n",
4464 (u_longlong_t
)rebuild
->dh_lb_asize
);
4469 dump_l2arc_header(int fd
)
4471 l2arc_dev_hdr_phys_t l2dhdr
= {0}, rebuild
= {0};
4472 int error
= B_FALSE
;
4474 if (pread64(fd
, &l2dhdr
, sizeof (l2dhdr
),
4475 VDEV_LABEL_START_SIZE
) != sizeof (l2dhdr
)) {
4478 if (l2dhdr
.dh_magic
== BSWAP_64(L2ARC_DEV_HDR_MAGIC
))
4479 byteswap_uint64_array(&l2dhdr
, sizeof (l2dhdr
));
4481 if (l2dhdr
.dh_magic
!= L2ARC_DEV_HDR_MAGIC
)
4486 (void) printf("L2ARC device header not found\n\n");
4487 /* Do not return an error here for backward compatibility */
4489 } else if (!dump_opt
['q']) {
4490 print_l2arc_header();
4492 (void) printf(" magic: %llu\n",
4493 (u_longlong_t
)l2dhdr
.dh_magic
);
4494 (void) printf(" version: %llu\n",
4495 (u_longlong_t
)l2dhdr
.dh_version
);
4496 (void) printf(" pool_guid: %llu\n",
4497 (u_longlong_t
)l2dhdr
.dh_spa_guid
);
4498 (void) printf(" flags: %llu\n",
4499 (u_longlong_t
)l2dhdr
.dh_flags
);
4500 (void) printf(" start_lbps[0]: %llu\n",
4502 l2dhdr
.dh_start_lbps
[0].lbp_daddr
);
4503 (void) printf(" start_lbps[1]: %llu\n",
4505 l2dhdr
.dh_start_lbps
[1].lbp_daddr
);
4506 (void) printf(" log_blk_ent: %llu\n",
4507 (u_longlong_t
)l2dhdr
.dh_log_entries
);
4508 (void) printf(" start: %llu\n",
4509 (u_longlong_t
)l2dhdr
.dh_start
);
4510 (void) printf(" end: %llu\n",
4511 (u_longlong_t
)l2dhdr
.dh_end
);
4512 (void) printf(" evict: %llu\n",
4513 (u_longlong_t
)l2dhdr
.dh_evict
);
4514 (void) printf(" lb_asize_refcount: %llu\n",
4515 (u_longlong_t
)l2dhdr
.dh_lb_asize
);
4516 (void) printf(" lb_count_refcount: %llu\n",
4517 (u_longlong_t
)l2dhdr
.dh_lb_count
);
4518 (void) printf(" trim_action_time: %llu\n",
4519 (u_longlong_t
)l2dhdr
.dh_trim_action_time
);
4520 (void) printf(" trim_state: %llu\n\n",
4521 (u_longlong_t
)l2dhdr
.dh_trim_state
);
4524 dump_l2arc_log_blocks(fd
, l2dhdr
, &rebuild
);
4526 * The total aligned size of log blocks and the number of log blocks
4527 * reported in the header of the device may be less than what zdb
4528 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
4529 * This happens because dump_l2arc_log_blocks() lacks the memory
4530 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
4531 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
4532 * and dh_lb_count will be lower to begin with than what exists on the
4533 * device. This is normal and zdb should not exit with an error. The
4534 * opposite case should never happen though, the values reported in the
4535 * header should never be higher than what dump_l2arc_log_blocks() and
4536 * l2arc_rebuild() report. If this happens there is a leak in the
4537 * accounting of log blocks.
4539 if (l2dhdr
.dh_lb_asize
> rebuild
.dh_lb_asize
||
4540 l2dhdr
.dh_lb_count
> rebuild
.dh_lb_count
)
4547 dump_config_from_label(zdb_label_t
*label
, size_t buflen
, int l
)
4552 if ((dump_opt
['l'] < 3) && (first_label(label
->config
) != l
))
4555 print_label_header(label
, l
);
4556 dump_nvlist(label
->config_nv
, 4);
4557 print_label_numbers(" labels = ", label
->config
);
4559 if (dump_opt
['l'] >= 2)
4560 dump_nvlist_stats(label
->config_nv
, buflen
);
4563 #define ZDB_MAX_UB_HEADER_SIZE 32
4566 dump_label_uberblocks(zdb_label_t
*label
, uint64_t ashift
, int label_num
)
4570 char header
[ZDB_MAX_UB_HEADER_SIZE
];
4572 vd
.vdev_ashift
= ashift
;
4575 for (int i
= 0; i
< VDEV_UBERBLOCK_COUNT(&vd
); i
++) {
4576 uint64_t uoff
= VDEV_UBERBLOCK_OFFSET(&vd
, i
);
4577 uberblock_t
*ub
= (void *)((char *)&label
->label
+ uoff
);
4578 cksum_record_t
*rec
= label
->uberblocks
[i
];
4581 if (dump_opt
['u'] >= 2) {
4582 print_label_header(label
, label_num
);
4583 (void) printf(" Uberblock[%d] invalid\n", i
);
4588 if ((dump_opt
['u'] < 3) && (first_label(rec
) != label_num
))
4591 if ((dump_opt
['u'] < 4) &&
4592 (ub
->ub_mmp_magic
== MMP_MAGIC
) && ub
->ub_mmp_delay
&&
4593 (i
>= VDEV_UBERBLOCK_COUNT(&vd
) - MMP_BLOCKS_PER_LABEL
))
4596 print_label_header(label
, label_num
);
4597 (void) snprintf(header
, ZDB_MAX_UB_HEADER_SIZE
,
4598 " Uberblock[%d]\n", i
);
4599 dump_uberblock(ub
, header
, "");
4600 print_label_numbers(" labels = ", rec
);
4604 static char curpath
[PATH_MAX
];
4607 * Iterate through the path components, recursively passing
4608 * current one's obj and remaining path until we find the obj
4612 dump_path_impl(objset_t
*os
, uint64_t obj
, char *name
, uint64_t *retobj
)
4615 boolean_t header
= B_TRUE
;
4619 dmu_object_info_t doi
;
4621 if ((s
= strchr(name
, '/')) != NULL
)
4623 err
= zap_lookup(os
, obj
, name
, 8, 1, &child_obj
);
4625 (void) strlcat(curpath
, name
, sizeof (curpath
));
4628 (void) fprintf(stderr
, "failed to lookup %s: %s\n",
4629 curpath
, strerror(err
));
4633 child_obj
= ZFS_DIRENT_OBJ(child_obj
);
4634 err
= sa_buf_hold(os
, child_obj
, FTAG
, &db
);
4636 (void) fprintf(stderr
,
4637 "failed to get SA dbuf for obj %llu: %s\n",
4638 (u_longlong_t
)child_obj
, strerror(err
));
4641 dmu_object_info_from_db(db
, &doi
);
4642 sa_buf_rele(db
, FTAG
);
4644 if (doi
.doi_bonus_type
!= DMU_OT_SA
&&
4645 doi
.doi_bonus_type
!= DMU_OT_ZNODE
) {
4646 (void) fprintf(stderr
, "invalid bonus type %d for obj %llu\n",
4647 doi
.doi_bonus_type
, (u_longlong_t
)child_obj
);
4651 if (dump_opt
['v'] > 6) {
4652 (void) printf("obj=%llu %s type=%d bonustype=%d\n",
4653 (u_longlong_t
)child_obj
, curpath
, doi
.doi_type
,
4654 doi
.doi_bonus_type
);
4657 (void) strlcat(curpath
, "/", sizeof (curpath
));
4659 switch (doi
.doi_type
) {
4660 case DMU_OT_DIRECTORY_CONTENTS
:
4661 if (s
!= NULL
&& *(s
+ 1) != '\0')
4662 return (dump_path_impl(os
, child_obj
, s
+ 1, retobj
));
4664 case DMU_OT_PLAIN_FILE_CONTENTS
:
4665 if (retobj
!= NULL
) {
4666 *retobj
= child_obj
;
4668 dump_object(os
, child_obj
, dump_opt
['v'], &header
,
4673 (void) fprintf(stderr
, "object %llu has non-file/directory "
4674 "type %d\n", (u_longlong_t
)obj
, doi
.doi_type
);
4682 * Dump the blocks for the object specified by path inside the dataset.
4685 dump_path(char *ds
, char *path
, uint64_t *retobj
)
4691 err
= open_objset(ds
, FTAG
, &os
);
4695 err
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_ROOT_OBJ
, 8, 1, &root_obj
);
4697 (void) fprintf(stderr
, "can't lookup root znode: %s\n",
4699 close_objset(os
, FTAG
);
4703 (void) snprintf(curpath
, sizeof (curpath
), "dataset=%s path=/", ds
);
4705 err
= dump_path_impl(os
, root_obj
, path
, retobj
);
4707 close_objset(os
, FTAG
);
4712 zdb_copy_object(objset_t
*os
, uint64_t srcobj
, char *destfile
)
4715 uint64_t size
, readsize
, oursize
, offset
;
4719 (void) printf("Copying object %" PRIu64
" to file %s\n", srcobj
,
4722 VERIFY3P(os
, ==, sa_os
);
4723 if ((err
= sa_handle_get(os
, srcobj
, NULL
, SA_HDL_PRIVATE
, &hdl
))) {
4724 (void) printf("Failed to get handle for SA znode\n");
4727 if ((err
= sa_lookup(hdl
, sa_attr_table
[ZPL_SIZE
], &size
, 8))) {
4728 (void) sa_handle_destroy(hdl
);
4731 (void) sa_handle_destroy(hdl
);
4733 (void) printf("Object %" PRIu64
" is %" PRIu64
" bytes\n", srcobj
,
4739 int fd
= open(destfile
, O_WRONLY
| O_CREAT
| O_TRUNC
, 0644);
4743 * We cap the size at 1 mebibyte here to prevent
4744 * allocation failures and nigh-infinite printing if the
4745 * object is extremely large.
4747 oursize
= MIN(size
, 1 << 20);
4749 char *buf
= kmem_alloc(oursize
, KM_NOSLEEP
);
4755 while (offset
< size
) {
4756 readsize
= MIN(size
- offset
, 1 << 20);
4757 err
= dmu_read(os
, srcobj
, offset
, readsize
, buf
, 0);
4759 (void) printf("got error %u from dmu_read\n", err
);
4760 kmem_free(buf
, oursize
);
4764 if (dump_opt
['v'] > 3) {
4765 (void) printf("Read offset=%" PRIu64
" size=%" PRIu64
4766 " error=%d\n", offset
, readsize
, err
);
4769 writesize
= write(fd
, buf
, readsize
);
4770 if (writesize
< 0) {
4773 } else if (writesize
!= readsize
) {
4774 /* Incomplete write */
4775 (void) fprintf(stderr
, "Short write, only wrote %llu of"
4776 " %" PRIu64
" bytes, exiting...\n",
4777 (u_longlong_t
)writesize
, readsize
);
4787 kmem_free(buf
, oursize
);
4793 label_cksum_valid(vdev_label_t
*label
, uint64_t offset
)
4795 zio_checksum_info_t
*ci
= &zio_checksum_table
[ZIO_CHECKSUM_LABEL
];
4796 zio_cksum_t expected_cksum
;
4797 zio_cksum_t actual_cksum
;
4798 zio_cksum_t verifier
;
4802 void *data
= (char *)label
+ offsetof(vdev_label_t
, vl_vdev_phys
);
4803 eck
= (zio_eck_t
*)((char *)(data
) + VDEV_PHYS_SIZE
) - 1;
4805 offset
+= offsetof(vdev_label_t
, vl_vdev_phys
);
4806 ZIO_SET_CHECKSUM(&verifier
, offset
, 0, 0, 0);
4808 byteswap
= (eck
->zec_magic
== BSWAP_64(ZEC_MAGIC
));
4810 byteswap_uint64_array(&verifier
, sizeof (zio_cksum_t
));
4812 expected_cksum
= eck
->zec_cksum
;
4813 eck
->zec_cksum
= verifier
;
4815 abd_t
*abd
= abd_get_from_buf(data
, VDEV_PHYS_SIZE
);
4816 ci
->ci_func
[byteswap
](abd
, VDEV_PHYS_SIZE
, NULL
, &actual_cksum
);
4820 byteswap_uint64_array(&expected_cksum
, sizeof (zio_cksum_t
));
4822 if (ZIO_CHECKSUM_EQUAL(actual_cksum
, expected_cksum
))
4829 dump_label(const char *dev
)
4831 char path
[MAXPATHLEN
];
4832 zdb_label_t labels
[VDEV_LABELS
] = {{{{0}}}};
4833 uint64_t psize
, ashift
, l2cache
;
4834 struct stat64 statbuf
;
4835 boolean_t config_found
= B_FALSE
;
4836 boolean_t error
= B_FALSE
;
4837 boolean_t read_l2arc_header
= B_FALSE
;
4838 avl_tree_t config_tree
;
4839 avl_tree_t uberblock_tree
;
4840 void *node
, *cookie
;
4844 * Check if we were given absolute path and use it as is.
4845 * Otherwise if the provided vdev name doesn't point to a file,
4846 * try prepending expected disk paths and partition numbers.
4848 (void) strlcpy(path
, dev
, sizeof (path
));
4849 if (dev
[0] != '/' && stat64(path
, &statbuf
) != 0) {
4852 error
= zfs_resolve_shortname(dev
, path
, MAXPATHLEN
);
4853 if (error
== 0 && zfs_dev_is_whole_disk(path
)) {
4854 if (zfs_append_partition(path
, MAXPATHLEN
) == -1)
4858 if (error
|| (stat64(path
, &statbuf
) != 0)) {
4859 (void) printf("failed to find device %s, try "
4860 "specifying absolute path instead\n", dev
);
4865 if ((fd
= open64(path
, O_RDONLY
)) < 0) {
4866 (void) printf("cannot open '%s': %s\n", path
, strerror(errno
));
4870 if (fstat64_blk(fd
, &statbuf
) != 0) {
4871 (void) printf("failed to stat '%s': %s\n", path
,
4877 if (S_ISBLK(statbuf
.st_mode
) && zfs_dev_flush(fd
) != 0)
4878 (void) printf("failed to invalidate cache '%s' : %s\n", path
,
4881 avl_create(&config_tree
, cksum_record_compare
,
4882 sizeof (cksum_record_t
), offsetof(cksum_record_t
, link
));
4883 avl_create(&uberblock_tree
, cksum_record_compare
,
4884 sizeof (cksum_record_t
), offsetof(cksum_record_t
, link
));
4886 psize
= statbuf
.st_size
;
4887 psize
= P2ALIGN(psize
, (uint64_t)sizeof (vdev_label_t
));
4888 ashift
= SPA_MINBLOCKSHIFT
;
4891 * 1. Read the label from disk
4892 * 2. Verify label cksum
4893 * 3. Unpack the configuration and insert in config tree.
4894 * 4. Traverse all uberblocks and insert in uberblock tree.
4896 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
4897 zdb_label_t
*label
= &labels
[l
];
4898 char *buf
= label
->label
.vl_vdev_phys
.vp_nvlist
;
4899 size_t buflen
= sizeof (label
->label
.vl_vdev_phys
.vp_nvlist
);
4901 cksum_record_t
*rec
;
4905 label
->label_offset
= vdev_label_offset(psize
, l
, 0);
4907 if (pread64(fd
, &label
->label
, sizeof (label
->label
),
4908 label
->label_offset
) != sizeof (label
->label
)) {
4910 (void) printf("failed to read label %d\n", l
);
4911 label
->read_failed
= B_TRUE
;
4916 label
->read_failed
= B_FALSE
;
4917 label
->cksum_valid
= label_cksum_valid(&label
->label
,
4918 label
->label_offset
);
4920 if (nvlist_unpack(buf
, buflen
, &config
, 0) == 0) {
4921 nvlist_t
*vdev_tree
= NULL
;
4924 if ((nvlist_lookup_nvlist(config
,
4925 ZPOOL_CONFIG_VDEV_TREE
, &vdev_tree
) != 0) ||
4926 (nvlist_lookup_uint64(vdev_tree
,
4927 ZPOOL_CONFIG_ASHIFT
, &ashift
) != 0))
4928 ashift
= SPA_MINBLOCKSHIFT
;
4930 if (nvlist_size(config
, &size
, NV_ENCODE_XDR
) != 0)
4933 /* If the device is a cache device clear the header. */
4934 if (!read_l2arc_header
) {
4935 if (nvlist_lookup_uint64(config
,
4936 ZPOOL_CONFIG_POOL_STATE
, &l2cache
) == 0 &&
4937 l2cache
== POOL_STATE_L2CACHE
) {
4938 read_l2arc_header
= B_TRUE
;
4942 fletcher_4_native_varsize(buf
, size
, &cksum
);
4943 rec
= cksum_record_insert(&config_tree
, &cksum
, l
);
4945 label
->config
= rec
;
4946 label
->config_nv
= config
;
4947 config_found
= B_TRUE
;
4952 vd
.vdev_ashift
= ashift
;
4955 for (int i
= 0; i
< VDEV_UBERBLOCK_COUNT(&vd
); i
++) {
4956 uint64_t uoff
= VDEV_UBERBLOCK_OFFSET(&vd
, i
);
4957 uberblock_t
*ub
= (void *)((char *)label
+ uoff
);
4959 if (uberblock_verify(ub
))
4962 fletcher_4_native_varsize(ub
, sizeof (*ub
), &cksum
);
4963 rec
= cksum_record_insert(&uberblock_tree
, &cksum
, l
);
4965 label
->uberblocks
[i
] = rec
;
4970 * Dump the label and uberblocks.
4972 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
4973 zdb_label_t
*label
= &labels
[l
];
4974 size_t buflen
= sizeof (label
->label
.vl_vdev_phys
.vp_nvlist
);
4976 if (label
->read_failed
== B_TRUE
)
4979 if (label
->config_nv
) {
4980 dump_config_from_label(label
, buflen
, l
);
4983 (void) printf("failed to unpack label %d\n", l
);
4987 dump_label_uberblocks(label
, ashift
, l
);
4989 nvlist_free(label
->config_nv
);
4993 * Dump the L2ARC header, if existent.
4995 if (read_l2arc_header
)
4996 error
|= dump_l2arc_header(fd
);
4999 while ((node
= avl_destroy_nodes(&config_tree
, &cookie
)) != NULL
)
5000 umem_free(node
, sizeof (cksum_record_t
));
5003 while ((node
= avl_destroy_nodes(&uberblock_tree
, &cookie
)) != NULL
)
5004 umem_free(node
, sizeof (cksum_record_t
));
5006 avl_destroy(&config_tree
);
5007 avl_destroy(&uberblock_tree
);
5011 return (config_found
== B_FALSE
? 2 :
5012 (error
== B_TRUE
? 1 : 0));
5015 static uint64_t dataset_feature_count
[SPA_FEATURES
];
5016 static uint64_t global_feature_count
[SPA_FEATURES
];
5017 static uint64_t remap_deadlist_count
= 0;
5020 dump_one_objset(const char *dsname
, void *arg
)
5027 error
= open_objset(dsname
, FTAG
, &os
);
5031 for (f
= 0; f
< SPA_FEATURES
; f
++) {
5032 if (!dsl_dataset_feature_is_active(dmu_objset_ds(os
), f
))
5034 ASSERT(spa_feature_table
[f
].fi_flags
&
5035 ZFEATURE_FLAG_PER_DATASET
);
5036 dataset_feature_count
[f
]++;
5039 if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os
))) {
5040 remap_deadlist_count
++;
5043 for (dsl_bookmark_node_t
*dbn
=
5044 avl_first(&dmu_objset_ds(os
)->ds_bookmarks
); dbn
!= NULL
;
5045 dbn
= AVL_NEXT(&dmu_objset_ds(os
)->ds_bookmarks
, dbn
)) {
5046 mos_obj_refd(dbn
->dbn_phys
.zbm_redaction_obj
);
5047 if (dbn
->dbn_phys
.zbm_redaction_obj
!= 0)
5048 global_feature_count
[SPA_FEATURE_REDACTION_BOOKMARKS
]++;
5049 if (dbn
->dbn_phys
.zbm_flags
& ZBM_FLAG_HAS_FBN
)
5050 global_feature_count
[SPA_FEATURE_BOOKMARK_WRITTEN
]++;
5053 if (dsl_deadlist_is_open(&dmu_objset_ds(os
)->ds_dir
->dd_livelist
) &&
5054 !dmu_objset_is_snapshot(os
)) {
5055 global_feature_count
[SPA_FEATURE_LIVELIST
]++;
5059 close_objset(os
, FTAG
);
5060 fuid_table_destroy();
5067 #define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5068 typedef struct zdb_blkstats
{
5074 uint64_t zb_ditto_samevdev
;
5075 uint64_t zb_ditto_same_ms
;
5076 uint64_t zb_psize_histogram
[PSIZE_HISTO_SIZE
];
5080 * Extended object types to report deferred frees and dedup auto-ditto blocks.
5082 #define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0)
5083 #define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1)
5084 #define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2)
5085 #define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3)
5087 static const char *zdb_ot_extname
[] = {
5094 #define ZB_TOTAL DN_MAX_LEVELS
5095 #define SPA_MAX_FOR_16M (SPA_MAXBLOCKSHIFT+1)
5097 typedef struct zdb_cb
{
5098 zdb_blkstats_t zcb_type
[ZB_TOTAL
+ 1][ZDB_OT_TOTAL
+ 1];
5099 uint64_t zcb_removing_size
;
5100 uint64_t zcb_checkpoint_size
;
5101 uint64_t zcb_dedup_asize
;
5102 uint64_t zcb_dedup_blocks
;
5103 uint64_t zcb_psize_count
[SPA_MAX_FOR_16M
];
5104 uint64_t zcb_lsize_count
[SPA_MAX_FOR_16M
];
5105 uint64_t zcb_asize_count
[SPA_MAX_FOR_16M
];
5106 uint64_t zcb_psize_len
[SPA_MAX_FOR_16M
];
5107 uint64_t zcb_lsize_len
[SPA_MAX_FOR_16M
];
5108 uint64_t zcb_asize_len
[SPA_MAX_FOR_16M
];
5109 uint64_t zcb_psize_total
;
5110 uint64_t zcb_lsize_total
;
5111 uint64_t zcb_asize_total
;
5112 uint64_t zcb_embedded_blocks
[NUM_BP_EMBEDDED_TYPES
];
5113 uint64_t zcb_embedded_histogram
[NUM_BP_EMBEDDED_TYPES
]
5114 [BPE_PAYLOAD_SIZE
+ 1];
5116 hrtime_t zcb_lastprint
;
5117 uint64_t zcb_totalasize
;
5118 uint64_t zcb_errors
[256];
5122 uint32_t **zcb_vd_obsolete_counts
;
5125 /* test if two DVA offsets from same vdev are within the same metaslab */
5127 same_metaslab(spa_t
*spa
, uint64_t vdev
, uint64_t off1
, uint64_t off2
)
5129 vdev_t
*vd
= vdev_lookup_top(spa
, vdev
);
5130 uint64_t ms_shift
= vd
->vdev_ms_shift
;
5132 return ((off1
>> ms_shift
) == (off2
>> ms_shift
));
5136 * Used to simplify reporting of the histogram data.
5138 typedef struct one_histo
{
5142 uint64_t cumulative
;
5146 * The number of separate histograms processed for psize, lsize and asize.
5151 * This routine will create a fixed column size output of three different
5152 * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5153 * the count, length and cumulative length of the psize, lsize and
5156 * All three types of blocks are listed on a single line
5158 * By default the table is printed in nicenumber format (e.g. 123K) but
5159 * if the '-P' parameter is specified then the full raw number (parseable)
5163 dump_size_histograms(zdb_cb_t
*zcb
)
5166 * A temporary buffer that allows us to convert a number into
5167 * a string using zdb_nicenumber to allow either raw or human
5168 * readable numbers to be output.
5173 * Define titles which are used in the headers of the tables
5174 * printed by this routine.
5176 const char blocksize_title1
[] = "block";
5177 const char blocksize_title2
[] = "size";
5178 const char count_title
[] = "Count";
5179 const char length_title
[] = "Size";
5180 const char cumulative_title
[] = "Cum.";
5183 * Setup the histogram arrays (psize, lsize, and asize).
5185 one_histo_t parm_histo
[NUM_HISTO
];
5187 parm_histo
[0].name
= "psize";
5188 parm_histo
[0].count
= zcb
->zcb_psize_count
;
5189 parm_histo
[0].len
= zcb
->zcb_psize_len
;
5190 parm_histo
[0].cumulative
= 0;
5192 parm_histo
[1].name
= "lsize";
5193 parm_histo
[1].count
= zcb
->zcb_lsize_count
;
5194 parm_histo
[1].len
= zcb
->zcb_lsize_len
;
5195 parm_histo
[1].cumulative
= 0;
5197 parm_histo
[2].name
= "asize";
5198 parm_histo
[2].count
= zcb
->zcb_asize_count
;
5199 parm_histo
[2].len
= zcb
->zcb_asize_len
;
5200 parm_histo
[2].cumulative
= 0;
5203 (void) printf("\nBlock Size Histogram\n");
5205 * Print the first line titles
5208 (void) printf("\n%s\t", blocksize_title1
);
5210 (void) printf("\n%7s ", blocksize_title1
);
5212 for (int j
= 0; j
< NUM_HISTO
; j
++) {
5213 if (dump_opt
['P']) {
5214 if (j
< NUM_HISTO
- 1) {
5215 (void) printf("%s\t\t\t", parm_histo
[j
].name
);
5217 /* Don't print trailing spaces */
5218 (void) printf(" %s", parm_histo
[j
].name
);
5221 if (j
< NUM_HISTO
- 1) {
5222 /* Left aligned strings in the output */
5223 (void) printf("%-7s ",
5224 parm_histo
[j
].name
);
5226 /* Don't print trailing spaces */
5227 (void) printf("%s", parm_histo
[j
].name
);
5231 (void) printf("\n");
5234 * Print the second line titles
5236 if (dump_opt
['P']) {
5237 (void) printf("%s\t", blocksize_title2
);
5239 (void) printf("%7s ", blocksize_title2
);
5242 for (int i
= 0; i
< NUM_HISTO
; i
++) {
5243 if (dump_opt
['P']) {
5244 (void) printf("%s\t%s\t%s\t",
5245 count_title
, length_title
, cumulative_title
);
5247 (void) printf("%7s%7s%7s",
5248 count_title
, length_title
, cumulative_title
);
5251 (void) printf("\n");
5256 for (int i
= SPA_MINBLOCKSHIFT
; i
< SPA_MAX_FOR_16M
; i
++) {
5259 * Print the first column showing the blocksize
5261 zdb_nicenum((1ULL << i
), numbuf
, sizeof (numbuf
));
5263 if (dump_opt
['P']) {
5264 printf("%s", numbuf
);
5266 printf("%7s:", numbuf
);
5270 * Print the remaining set of 3 columns per size:
5271 * for psize, lsize and asize
5273 for (int j
= 0; j
< NUM_HISTO
; j
++) {
5274 parm_histo
[j
].cumulative
+= parm_histo
[j
].len
[i
];
5276 zdb_nicenum(parm_histo
[j
].count
[i
],
5277 numbuf
, sizeof (numbuf
));
5279 (void) printf("\t%s", numbuf
);
5281 (void) printf("%7s", numbuf
);
5283 zdb_nicenum(parm_histo
[j
].len
[i
],
5284 numbuf
, sizeof (numbuf
));
5286 (void) printf("\t%s", numbuf
);
5288 (void) printf("%7s", numbuf
);
5290 zdb_nicenum(parm_histo
[j
].cumulative
,
5291 numbuf
, sizeof (numbuf
));
5293 (void) printf("\t%s", numbuf
);
5295 (void) printf("%7s", numbuf
);
5297 (void) printf("\n");
5302 zdb_count_block(zdb_cb_t
*zcb
, zilog_t
*zilog
, const blkptr_t
*bp
,
5303 dmu_object_type_t type
)
5305 uint64_t refcnt
= 0;
5308 ASSERT(type
< ZDB_OT_TOTAL
);
5310 if (zilog
&& zil_bp_tree_add(zilog
, bp
) != 0)
5313 spa_config_enter(zcb
->zcb_spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5315 for (i
= 0; i
< 4; i
++) {
5316 int l
= (i
< 2) ? BP_GET_LEVEL(bp
) : ZB_TOTAL
;
5317 int t
= (i
& 1) ? type
: ZDB_OT_TOTAL
;
5319 zdb_blkstats_t
*zb
= &zcb
->zcb_type
[l
][t
];
5321 zb
->zb_asize
+= BP_GET_ASIZE(bp
);
5322 zb
->zb_lsize
+= BP_GET_LSIZE(bp
);
5323 zb
->zb_psize
+= BP_GET_PSIZE(bp
);
5327 * The histogram is only big enough to record blocks up to
5328 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
5331 unsigned idx
= BP_GET_PSIZE(bp
) >> SPA_MINBLOCKSHIFT
;
5332 idx
= MIN(idx
, SPA_OLD_MAXBLOCKSIZE
/ SPA_MINBLOCKSIZE
+ 1);
5333 zb
->zb_psize_histogram
[idx
]++;
5335 zb
->zb_gangs
+= BP_COUNT_GANG(bp
);
5337 switch (BP_GET_NDVAS(bp
)) {
5339 if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5340 DVA_GET_VDEV(&bp
->blk_dva
[1])) {
5341 zb
->zb_ditto_samevdev
++;
5343 if (same_metaslab(zcb
->zcb_spa
,
5344 DVA_GET_VDEV(&bp
->blk_dva
[0]),
5345 DVA_GET_OFFSET(&bp
->blk_dva
[0]),
5346 DVA_GET_OFFSET(&bp
->blk_dva
[1])))
5347 zb
->zb_ditto_same_ms
++;
5351 equal
= (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5352 DVA_GET_VDEV(&bp
->blk_dva
[1])) +
5353 (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5354 DVA_GET_VDEV(&bp
->blk_dva
[2])) +
5355 (DVA_GET_VDEV(&bp
->blk_dva
[1]) ==
5356 DVA_GET_VDEV(&bp
->blk_dva
[2]));
5358 zb
->zb_ditto_samevdev
++;
5360 if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5361 DVA_GET_VDEV(&bp
->blk_dva
[1]) &&
5362 same_metaslab(zcb
->zcb_spa
,
5363 DVA_GET_VDEV(&bp
->blk_dva
[0]),
5364 DVA_GET_OFFSET(&bp
->blk_dva
[0]),
5365 DVA_GET_OFFSET(&bp
->blk_dva
[1])))
5366 zb
->zb_ditto_same_ms
++;
5367 else if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5368 DVA_GET_VDEV(&bp
->blk_dva
[2]) &&
5369 same_metaslab(zcb
->zcb_spa
,
5370 DVA_GET_VDEV(&bp
->blk_dva
[0]),
5371 DVA_GET_OFFSET(&bp
->blk_dva
[0]),
5372 DVA_GET_OFFSET(&bp
->blk_dva
[2])))
5373 zb
->zb_ditto_same_ms
++;
5374 else if (DVA_GET_VDEV(&bp
->blk_dva
[1]) ==
5375 DVA_GET_VDEV(&bp
->blk_dva
[2]) &&
5376 same_metaslab(zcb
->zcb_spa
,
5377 DVA_GET_VDEV(&bp
->blk_dva
[1]),
5378 DVA_GET_OFFSET(&bp
->blk_dva
[1]),
5379 DVA_GET_OFFSET(&bp
->blk_dva
[2])))
5380 zb
->zb_ditto_same_ms
++;
5386 spa_config_exit(zcb
->zcb_spa
, SCL_CONFIG
, FTAG
);
5388 if (BP_IS_EMBEDDED(bp
)) {
5389 zcb
->zcb_embedded_blocks
[BPE_GET_ETYPE(bp
)]++;
5390 zcb
->zcb_embedded_histogram
[BPE_GET_ETYPE(bp
)]
5391 [BPE_GET_PSIZE(bp
)]++;
5395 * The binning histogram bins by powers of two up to
5396 * SPA_MAXBLOCKSIZE rather than creating bins for
5397 * every possible blocksize found in the pool.
5399 int bin
= highbit64(BP_GET_PSIZE(bp
)) - 1;
5401 zcb
->zcb_psize_count
[bin
]++;
5402 zcb
->zcb_psize_len
[bin
] += BP_GET_PSIZE(bp
);
5403 zcb
->zcb_psize_total
+= BP_GET_PSIZE(bp
);
5405 bin
= highbit64(BP_GET_LSIZE(bp
)) - 1;
5407 zcb
->zcb_lsize_count
[bin
]++;
5408 zcb
->zcb_lsize_len
[bin
] += BP_GET_LSIZE(bp
);
5409 zcb
->zcb_lsize_total
+= BP_GET_LSIZE(bp
);
5411 bin
= highbit64(BP_GET_ASIZE(bp
)) - 1;
5413 zcb
->zcb_asize_count
[bin
]++;
5414 zcb
->zcb_asize_len
[bin
] += BP_GET_ASIZE(bp
);
5415 zcb
->zcb_asize_total
+= BP_GET_ASIZE(bp
);
5420 if (BP_GET_DEDUP(bp
)) {
5424 ddt
= ddt_select(zcb
->zcb_spa
, bp
);
5426 dde
= ddt_lookup(ddt
, bp
, B_FALSE
);
5431 ddt_phys_t
*ddp
= ddt_phys_select(dde
, bp
);
5432 ddt_phys_decref(ddp
);
5433 refcnt
= ddp
->ddp_refcnt
;
5434 if (ddt_phys_total_refcnt(dde
) == 0)
5435 ddt_remove(ddt
, dde
);
5440 VERIFY3U(zio_wait(zio_claim(NULL
, zcb
->zcb_spa
,
5441 refcnt
? 0 : spa_min_claim_txg(zcb
->zcb_spa
),
5442 bp
, NULL
, NULL
, ZIO_FLAG_CANFAIL
)), ==, 0);
5446 zdb_blkptr_done(zio_t
*zio
)
5448 spa_t
*spa
= zio
->io_spa
;
5449 blkptr_t
*bp
= zio
->io_bp
;
5450 int ioerr
= zio
->io_error
;
5451 zdb_cb_t
*zcb
= zio
->io_private
;
5452 zbookmark_phys_t
*zb
= &zio
->io_bookmark
;
5454 mutex_enter(&spa
->spa_scrub_lock
);
5455 spa
->spa_load_verify_bytes
-= BP_GET_PSIZE(bp
);
5456 cv_broadcast(&spa
->spa_scrub_io_cv
);
5458 if (ioerr
&& !(zio
->io_flags
& ZIO_FLAG_SPECULATIVE
)) {
5459 char blkbuf
[BP_SPRINTF_LEN
];
5461 zcb
->zcb_haderrors
= 1;
5462 zcb
->zcb_errors
[ioerr
]++;
5464 if (dump_opt
['b'] >= 2)
5465 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
5469 (void) printf("zdb_blkptr_cb: "
5470 "Got error %d reading "
5471 "<%llu, %llu, %lld, %llx> %s -- skipping\n",
5473 (u_longlong_t
)zb
->zb_objset
,
5474 (u_longlong_t
)zb
->zb_object
,
5475 (u_longlong_t
)zb
->zb_level
,
5476 (u_longlong_t
)zb
->zb_blkid
,
5479 mutex_exit(&spa
->spa_scrub_lock
);
5481 abd_free(zio
->io_abd
);
5485 zdb_blkptr_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
5486 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
5488 zdb_cb_t
*zcb
= arg
;
5489 dmu_object_type_t type
;
5490 boolean_t is_metadata
;
5492 if (zb
->zb_level
== ZB_DNODE_LEVEL
)
5495 if (dump_opt
['b'] >= 5 && bp
->blk_birth
> 0) {
5496 char blkbuf
[BP_SPRINTF_LEN
];
5497 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
5498 (void) printf("objset %llu object %llu "
5499 "level %lld offset 0x%llx %s\n",
5500 (u_longlong_t
)zb
->zb_objset
,
5501 (u_longlong_t
)zb
->zb_object
,
5502 (longlong_t
)zb
->zb_level
,
5503 (u_longlong_t
)blkid2offset(dnp
, bp
, zb
),
5507 if (BP_IS_HOLE(bp
) || BP_IS_REDACTED(bp
))
5510 type
= BP_GET_TYPE(bp
);
5512 zdb_count_block(zcb
, zilog
, bp
,
5513 (type
& DMU_OT_NEWTYPE
) ? ZDB_OT_OTHER
: type
);
5515 is_metadata
= (BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
));
5517 if (!BP_IS_EMBEDDED(bp
) &&
5518 (dump_opt
['c'] > 1 || (dump_opt
['c'] && is_metadata
))) {
5519 size_t size
= BP_GET_PSIZE(bp
);
5520 abd_t
*abd
= abd_alloc(size
, B_FALSE
);
5521 int flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
;
5523 /* If it's an intent log block, failure is expected. */
5524 if (zb
->zb_level
== ZB_ZIL_LEVEL
)
5525 flags
|= ZIO_FLAG_SPECULATIVE
;
5527 mutex_enter(&spa
->spa_scrub_lock
);
5528 while (spa
->spa_load_verify_bytes
> max_inflight_bytes
)
5529 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
5530 spa
->spa_load_verify_bytes
+= size
;
5531 mutex_exit(&spa
->spa_scrub_lock
);
5533 zio_nowait(zio_read(NULL
, spa
, bp
, abd
, size
,
5534 zdb_blkptr_done
, zcb
, ZIO_PRIORITY_ASYNC_READ
, flags
, zb
));
5537 zcb
->zcb_readfails
= 0;
5539 /* only call gethrtime() every 100 blocks */
5546 if (dump_opt
['b'] < 5 && gethrtime() > zcb
->zcb_lastprint
+ NANOSEC
) {
5547 uint64_t now
= gethrtime();
5549 uint64_t bytes
= zcb
->zcb_type
[ZB_TOTAL
][ZDB_OT_TOTAL
].zb_asize
;
5550 uint64_t kb_per_sec
=
5551 1 + bytes
/ (1 + ((now
- zcb
->zcb_start
) / 1000 / 1000));
5552 uint64_t sec_remaining
=
5553 (zcb
->zcb_totalasize
- bytes
) / 1024 / kb_per_sec
;
5555 /* make sure nicenum has enough space */
5556 _Static_assert(sizeof (buf
) >= NN_NUMBUF_SZ
, "buf truncated");
5558 zfs_nicebytes(bytes
, buf
, sizeof (buf
));
5559 (void) fprintf(stderr
,
5560 "\r%5s completed (%4"PRIu64
"MB/s) "
5561 "estimated time remaining: "
5562 "%"PRIu64
"hr %02"PRIu64
"min %02"PRIu64
"sec ",
5563 buf
, kb_per_sec
/ 1024,
5564 sec_remaining
/ 60 / 60,
5565 sec_remaining
/ 60 % 60,
5566 sec_remaining
% 60);
5568 zcb
->zcb_lastprint
= now
;
5575 zdb_leak(void *arg
, uint64_t start
, uint64_t size
)
5579 (void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
5580 (u_longlong_t
)vd
->vdev_id
, (u_longlong_t
)start
, (u_longlong_t
)size
);
5583 static metaslab_ops_t zdb_metaslab_ops
= {
5588 load_unflushed_svr_segs_cb(spa_t
*spa
, space_map_entry_t
*sme
,
5589 uint64_t txg
, void *arg
)
5591 spa_vdev_removal_t
*svr
= arg
;
5593 uint64_t offset
= sme
->sme_offset
;
5594 uint64_t size
= sme
->sme_run
;
5596 /* skip vdevs we don't care about */
5597 if (sme
->sme_vdev
!= svr
->svr_vdev_id
)
5600 vdev_t
*vd
= vdev_lookup_top(spa
, sme
->sme_vdev
);
5601 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
5602 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
5604 if (txg
< metaslab_unflushed_txg(ms
))
5607 if (sme
->sme_type
== SM_ALLOC
)
5608 range_tree_add(svr
->svr_allocd_segs
, offset
, size
);
5610 range_tree_remove(svr
->svr_allocd_segs
, offset
, size
);
5616 claim_segment_impl_cb(uint64_t inner_offset
, vdev_t
*vd
, uint64_t offset
,
5617 uint64_t size
, void *arg
)
5619 (void) inner_offset
, (void) arg
;
5622 * This callback was called through a remap from
5623 * a device being removed. Therefore, the vdev that
5624 * this callback is applied to is a concrete
5627 ASSERT(vdev_is_concrete(vd
));
5629 VERIFY0(metaslab_claim_impl(vd
, offset
, size
,
5630 spa_min_claim_txg(vd
->vdev_spa
)));
5634 claim_segment_cb(void *arg
, uint64_t offset
, uint64_t size
)
5638 vdev_indirect_ops
.vdev_op_remap(vd
, offset
, size
,
5639 claim_segment_impl_cb
, NULL
);
5643 * After accounting for all allocated blocks that are directly referenced,
5644 * we might have missed a reference to a block from a partially complete
5645 * (and thus unused) indirect mapping object. We perform a secondary pass
5646 * through the metaslabs we have already mapped and claim the destination
5650 zdb_claim_removing(spa_t
*spa
, zdb_cb_t
*zcb
)
5655 if (spa
->spa_vdev_removal
== NULL
)
5658 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5660 spa_vdev_removal_t
*svr
= spa
->spa_vdev_removal
;
5661 vdev_t
*vd
= vdev_lookup_top(spa
, svr
->svr_vdev_id
);
5662 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
5664 ASSERT0(range_tree_space(svr
->svr_allocd_segs
));
5666 range_tree_t
*allocs
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
5667 for (uint64_t msi
= 0; msi
< vd
->vdev_ms_count
; msi
++) {
5668 metaslab_t
*msp
= vd
->vdev_ms
[msi
];
5670 ASSERT0(range_tree_space(allocs
));
5671 if (msp
->ms_sm
!= NULL
)
5672 VERIFY0(space_map_load(msp
->ms_sm
, allocs
, SM_ALLOC
));
5673 range_tree_vacate(allocs
, range_tree_add
, svr
->svr_allocd_segs
);
5675 range_tree_destroy(allocs
);
5677 iterate_through_spacemap_logs(spa
, load_unflushed_svr_segs_cb
, svr
);
5680 * Clear everything past what has been synced,
5681 * because we have not allocated mappings for
5684 range_tree_clear(svr
->svr_allocd_segs
,
5685 vdev_indirect_mapping_max_offset(vim
),
5686 vd
->vdev_asize
- vdev_indirect_mapping_max_offset(vim
));
5688 zcb
->zcb_removing_size
+= range_tree_space(svr
->svr_allocd_segs
);
5689 range_tree_vacate(svr
->svr_allocd_segs
, claim_segment_cb
, vd
);
5691 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
5695 increment_indirect_mapping_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
5699 zdb_cb_t
*zcb
= arg
;
5700 spa_t
*spa
= zcb
->zcb_spa
;
5702 const dva_t
*dva
= &bp
->blk_dva
[0];
5705 ASSERT(!dump_opt
['L']);
5706 ASSERT3U(BP_GET_NDVAS(bp
), ==, 1);
5708 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
5709 vd
= vdev_lookup_top(zcb
->zcb_spa
, DVA_GET_VDEV(dva
));
5710 ASSERT3P(vd
, !=, NULL
);
5711 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
5713 ASSERT(vd
->vdev_indirect_config
.vic_mapping_object
!= 0);
5714 ASSERT3P(zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
], !=, NULL
);
5716 vdev_indirect_mapping_increment_obsolete_count(
5717 vd
->vdev_indirect_mapping
,
5718 DVA_GET_OFFSET(dva
), DVA_GET_ASIZE(dva
),
5719 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
]);
5725 zdb_load_obsolete_counts(vdev_t
*vd
)
5727 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
5728 spa_t
*spa
= vd
->vdev_spa
;
5729 spa_condensing_indirect_phys_t
*scip
=
5730 &spa
->spa_condensing_indirect_phys
;
5731 uint64_t obsolete_sm_object
;
5734 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
5735 EQUIV(obsolete_sm_object
!= 0, vd
->vdev_obsolete_sm
!= NULL
);
5736 counts
= vdev_indirect_mapping_load_obsolete_counts(vim
);
5737 if (vd
->vdev_obsolete_sm
!= NULL
) {
5738 vdev_indirect_mapping_load_obsolete_spacemap(vim
, counts
,
5739 vd
->vdev_obsolete_sm
);
5741 if (scip
->scip_vdev
== vd
->vdev_id
&&
5742 scip
->scip_prev_obsolete_sm_object
!= 0) {
5743 space_map_t
*prev_obsolete_sm
= NULL
;
5744 VERIFY0(space_map_open(&prev_obsolete_sm
, spa
->spa_meta_objset
,
5745 scip
->scip_prev_obsolete_sm_object
, 0, vd
->vdev_asize
, 0));
5746 vdev_indirect_mapping_load_obsolete_spacemap(vim
, counts
,
5748 space_map_close(prev_obsolete_sm
);
5754 zdb_ddt_leak_init(spa_t
*spa
, zdb_cb_t
*zcb
)
5756 ddt_bookmark_t ddb
= {0};
5761 ASSERT(!dump_opt
['L']);
5763 while ((error
= ddt_walk(spa
, &ddb
, &dde
)) == 0) {
5765 ddt_phys_t
*ddp
= dde
.dde_phys
;
5767 if (ddb
.ddb_class
== DDT_CLASS_UNIQUE
)
5770 ASSERT(ddt_phys_total_refcnt(&dde
) > 1);
5772 for (p
= 0; p
< DDT_PHYS_TYPES
; p
++, ddp
++) {
5773 if (ddp
->ddp_phys_birth
== 0)
5775 ddt_bp_create(ddb
.ddb_checksum
,
5776 &dde
.dde_key
, ddp
, &blk
);
5777 if (p
== DDT_PHYS_DITTO
) {
5778 zdb_count_block(zcb
, NULL
, &blk
, ZDB_OT_DITTO
);
5780 zcb
->zcb_dedup_asize
+=
5781 BP_GET_ASIZE(&blk
) * (ddp
->ddp_refcnt
- 1);
5782 zcb
->zcb_dedup_blocks
++;
5785 ddt_t
*ddt
= spa
->spa_ddt
[ddb
.ddb_checksum
];
5787 VERIFY(ddt_lookup(ddt
, &blk
, B_TRUE
) != NULL
);
5791 ASSERT(error
== ENOENT
);
5794 typedef struct checkpoint_sm_exclude_entry_arg
{
5796 uint64_t cseea_checkpoint_size
;
5797 } checkpoint_sm_exclude_entry_arg_t
;
5800 checkpoint_sm_exclude_entry_cb(space_map_entry_t
*sme
, void *arg
)
5802 checkpoint_sm_exclude_entry_arg_t
*cseea
= arg
;
5803 vdev_t
*vd
= cseea
->cseea_vd
;
5804 metaslab_t
*ms
= vd
->vdev_ms
[sme
->sme_offset
>> vd
->vdev_ms_shift
];
5805 uint64_t end
= sme
->sme_offset
+ sme
->sme_run
;
5807 ASSERT(sme
->sme_type
== SM_FREE
);
5810 * Since the vdev_checkpoint_sm exists in the vdev level
5811 * and the ms_sm space maps exist in the metaslab level,
5812 * an entry in the checkpoint space map could theoretically
5813 * cross the boundaries of the metaslab that it belongs.
5815 * In reality, because of the way that we populate and
5816 * manipulate the checkpoint's space maps currently,
5817 * there shouldn't be any entries that cross metaslabs.
5818 * Hence the assertion below.
5820 * That said, there is no fundamental requirement that
5821 * the checkpoint's space map entries should not cross
5822 * metaslab boundaries. So if needed we could add code
5823 * that handles metaslab-crossing segments in the future.
5825 VERIFY3U(sme
->sme_offset
, >=, ms
->ms_start
);
5826 VERIFY3U(end
, <=, ms
->ms_start
+ ms
->ms_size
);
5829 * By removing the entry from the allocated segments we
5830 * also verify that the entry is there to begin with.
5832 mutex_enter(&ms
->ms_lock
);
5833 range_tree_remove(ms
->ms_allocatable
, sme
->sme_offset
, sme
->sme_run
);
5834 mutex_exit(&ms
->ms_lock
);
5836 cseea
->cseea_checkpoint_size
+= sme
->sme_run
;
5841 zdb_leak_init_vdev_exclude_checkpoint(vdev_t
*vd
, zdb_cb_t
*zcb
)
5843 spa_t
*spa
= vd
->vdev_spa
;
5844 space_map_t
*checkpoint_sm
= NULL
;
5845 uint64_t checkpoint_sm_obj
;
5848 * If there is no vdev_top_zap, we are in a pool whose
5849 * version predates the pool checkpoint feature.
5851 if (vd
->vdev_top_zap
== 0)
5855 * If there is no reference of the vdev_checkpoint_sm in
5856 * the vdev_top_zap, then one of the following scenarios
5859 * 1] There is no checkpoint
5860 * 2] There is a checkpoint, but no checkpointed blocks
5861 * have been freed yet
5862 * 3] The current vdev is indirect
5864 * In these cases we return immediately.
5866 if (zap_contains(spa_meta_objset(spa
), vd
->vdev_top_zap
,
5867 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) != 0)
5870 VERIFY0(zap_lookup(spa_meta_objset(spa
), vd
->vdev_top_zap
,
5871 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1,
5872 &checkpoint_sm_obj
));
5874 checkpoint_sm_exclude_entry_arg_t cseea
;
5875 cseea
.cseea_vd
= vd
;
5876 cseea
.cseea_checkpoint_size
= 0;
5878 VERIFY0(space_map_open(&checkpoint_sm
, spa_meta_objset(spa
),
5879 checkpoint_sm_obj
, 0, vd
->vdev_asize
, vd
->vdev_ashift
));
5881 VERIFY0(space_map_iterate(checkpoint_sm
,
5882 space_map_length(checkpoint_sm
),
5883 checkpoint_sm_exclude_entry_cb
, &cseea
));
5884 space_map_close(checkpoint_sm
);
5886 zcb
->zcb_checkpoint_size
+= cseea
.cseea_checkpoint_size
;
5890 zdb_leak_init_exclude_checkpoint(spa_t
*spa
, zdb_cb_t
*zcb
)
5892 ASSERT(!dump_opt
['L']);
5894 vdev_t
*rvd
= spa
->spa_root_vdev
;
5895 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
5896 ASSERT3U(c
, ==, rvd
->vdev_child
[c
]->vdev_id
);
5897 zdb_leak_init_vdev_exclude_checkpoint(rvd
->vdev_child
[c
], zcb
);
5902 count_unflushed_space_cb(spa_t
*spa
, space_map_entry_t
*sme
,
5903 uint64_t txg
, void *arg
)
5905 int64_t *ualloc_space
= arg
;
5907 uint64_t offset
= sme
->sme_offset
;
5908 uint64_t vdev_id
= sme
->sme_vdev
;
5910 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
5911 if (!vdev_is_concrete(vd
))
5914 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
5915 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
5917 if (txg
< metaslab_unflushed_txg(ms
))
5920 if (sme
->sme_type
== SM_ALLOC
)
5921 *ualloc_space
+= sme
->sme_run
;
5923 *ualloc_space
-= sme
->sme_run
;
5929 get_unflushed_alloc_space(spa_t
*spa
)
5934 int64_t ualloc_space
= 0;
5935 iterate_through_spacemap_logs(spa
, count_unflushed_space_cb
,
5937 return (ualloc_space
);
5941 load_unflushed_cb(spa_t
*spa
, space_map_entry_t
*sme
, uint64_t txg
, void *arg
)
5943 maptype_t
*uic_maptype
= arg
;
5945 uint64_t offset
= sme
->sme_offset
;
5946 uint64_t size
= sme
->sme_run
;
5947 uint64_t vdev_id
= sme
->sme_vdev
;
5949 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
5951 /* skip indirect vdevs */
5952 if (!vdev_is_concrete(vd
))
5955 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
5957 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
5958 ASSERT(*uic_maptype
== SM_ALLOC
|| *uic_maptype
== SM_FREE
);
5960 if (txg
< metaslab_unflushed_txg(ms
))
5963 if (*uic_maptype
== sme
->sme_type
)
5964 range_tree_add(ms
->ms_allocatable
, offset
, size
);
5966 range_tree_remove(ms
->ms_allocatable
, offset
, size
);
5972 load_unflushed_to_ms_allocatables(spa_t
*spa
, maptype_t maptype
)
5974 iterate_through_spacemap_logs(spa
, load_unflushed_cb
, &maptype
);
5978 load_concrete_ms_allocatable_trees(spa_t
*spa
, maptype_t maptype
)
5980 vdev_t
*rvd
= spa
->spa_root_vdev
;
5981 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
5982 vdev_t
*vd
= rvd
->vdev_child
[i
];
5984 ASSERT3U(i
, ==, vd
->vdev_id
);
5986 if (vd
->vdev_ops
== &vdev_indirect_ops
)
5989 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
5990 metaslab_t
*msp
= vd
->vdev_ms
[m
];
5992 (void) fprintf(stderr
,
5993 "\rloading concrete vdev %llu, "
5994 "metaslab %llu of %llu ...",
5995 (longlong_t
)vd
->vdev_id
,
5996 (longlong_t
)msp
->ms_id
,
5997 (longlong_t
)vd
->vdev_ms_count
);
5999 mutex_enter(&msp
->ms_lock
);
6000 range_tree_vacate(msp
->ms_allocatable
, NULL
, NULL
);
6003 * We don't want to spend the CPU manipulating the
6004 * size-ordered tree, so clear the range_tree ops.
6006 msp
->ms_allocatable
->rt_ops
= NULL
;
6008 if (msp
->ms_sm
!= NULL
) {
6009 VERIFY0(space_map_load(msp
->ms_sm
,
6010 msp
->ms_allocatable
, maptype
));
6012 if (!msp
->ms_loaded
)
6013 msp
->ms_loaded
= B_TRUE
;
6014 mutex_exit(&msp
->ms_lock
);
6018 load_unflushed_to_ms_allocatables(spa
, maptype
);
6022 * vm_idxp is an in-out parameter which (for indirect vdevs) is the
6023 * index in vim_entries that has the first entry in this metaslab.
6024 * On return, it will be set to the first entry after this metaslab.
6027 load_indirect_ms_allocatable_tree(vdev_t
*vd
, metaslab_t
*msp
,
6030 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6032 mutex_enter(&msp
->ms_lock
);
6033 range_tree_vacate(msp
->ms_allocatable
, NULL
, NULL
);
6036 * We don't want to spend the CPU manipulating the
6037 * size-ordered tree, so clear the range_tree ops.
6039 msp
->ms_allocatable
->rt_ops
= NULL
;
6041 for (; *vim_idxp
< vdev_indirect_mapping_num_entries(vim
);
6043 vdev_indirect_mapping_entry_phys_t
*vimep
=
6044 &vim
->vim_entries
[*vim_idxp
];
6045 uint64_t ent_offset
= DVA_MAPPING_GET_SRC_OFFSET(vimep
);
6046 uint64_t ent_len
= DVA_GET_ASIZE(&vimep
->vimep_dst
);
6047 ASSERT3U(ent_offset
, >=, msp
->ms_start
);
6048 if (ent_offset
>= msp
->ms_start
+ msp
->ms_size
)
6052 * Mappings do not cross metaslab boundaries,
6053 * because we create them by walking the metaslabs.
6055 ASSERT3U(ent_offset
+ ent_len
, <=,
6056 msp
->ms_start
+ msp
->ms_size
);
6057 range_tree_add(msp
->ms_allocatable
, ent_offset
, ent_len
);
6060 if (!msp
->ms_loaded
)
6061 msp
->ms_loaded
= B_TRUE
;
6062 mutex_exit(&msp
->ms_lock
);
6066 zdb_leak_init_prepare_indirect_vdevs(spa_t
*spa
, zdb_cb_t
*zcb
)
6068 ASSERT(!dump_opt
['L']);
6070 vdev_t
*rvd
= spa
->spa_root_vdev
;
6071 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
6072 vdev_t
*vd
= rvd
->vdev_child
[c
];
6074 ASSERT3U(c
, ==, vd
->vdev_id
);
6076 if (vd
->vdev_ops
!= &vdev_indirect_ops
)
6080 * Note: we don't check for mapping leaks on
6081 * removing vdevs because their ms_allocatable's
6082 * are used to look for leaks in allocated space.
6084 zcb
->zcb_vd_obsolete_counts
[c
] = zdb_load_obsolete_counts(vd
);
6087 * Normally, indirect vdevs don't have any
6088 * metaslabs. We want to set them up for
6091 vdev_metaslab_group_create(vd
);
6092 VERIFY0(vdev_metaslab_init(vd
, 0));
6094 vdev_indirect_mapping_t
*vim __maybe_unused
=
6095 vd
->vdev_indirect_mapping
;
6096 uint64_t vim_idx
= 0;
6097 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
6099 (void) fprintf(stderr
,
6100 "\rloading indirect vdev %llu, "
6101 "metaslab %llu of %llu ...",
6102 (longlong_t
)vd
->vdev_id
,
6103 (longlong_t
)vd
->vdev_ms
[m
]->ms_id
,
6104 (longlong_t
)vd
->vdev_ms_count
);
6106 load_indirect_ms_allocatable_tree(vd
, vd
->vdev_ms
[m
],
6109 ASSERT3U(vim_idx
, ==, vdev_indirect_mapping_num_entries(vim
));
6114 zdb_leak_init(spa_t
*spa
, zdb_cb_t
*zcb
)
6121 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6122 vdev_t
*rvd
= spa
->spa_root_vdev
;
6125 * We are going to be changing the meaning of the metaslab's
6126 * ms_allocatable. Ensure that the allocator doesn't try to
6129 spa
->spa_normal_class
->mc_ops
= &zdb_metaslab_ops
;
6130 spa
->spa_log_class
->mc_ops
= &zdb_metaslab_ops
;
6131 spa
->spa_embedded_log_class
->mc_ops
= &zdb_metaslab_ops
;
6133 zcb
->zcb_vd_obsolete_counts
=
6134 umem_zalloc(rvd
->vdev_children
* sizeof (uint32_t *),
6138 * For leak detection, we overload the ms_allocatable trees
6139 * to contain allocated segments instead of free segments.
6140 * As a result, we can't use the normal metaslab_load/unload
6143 zdb_leak_init_prepare_indirect_vdevs(spa
, zcb
);
6144 load_concrete_ms_allocatable_trees(spa
, SM_ALLOC
);
6147 * On load_concrete_ms_allocatable_trees() we loaded all the
6148 * allocated entries from the ms_sm to the ms_allocatable for
6149 * each metaslab. If the pool has a checkpoint or is in the
6150 * middle of discarding a checkpoint, some of these blocks
6151 * may have been freed but their ms_sm may not have been
6152 * updated because they are referenced by the checkpoint. In
6153 * order to avoid false-positives during leak-detection, we
6154 * go through the vdev's checkpoint space map and exclude all
6155 * its entries from their relevant ms_allocatable.
6157 * We also aggregate the space held by the checkpoint and add
6158 * it to zcb_checkpoint_size.
6160 * Note that at this point we are also verifying that all the
6161 * entries on the checkpoint_sm are marked as allocated in
6162 * the ms_sm of their relevant metaslab.
6163 * [see comment in checkpoint_sm_exclude_entry_cb()]
6165 zdb_leak_init_exclude_checkpoint(spa
, zcb
);
6166 ASSERT3U(zcb
->zcb_checkpoint_size
, ==, spa_get_checkpoint_space(spa
));
6168 /* for cleaner progress output */
6169 (void) fprintf(stderr
, "\n");
6171 if (bpobj_is_open(&dp
->dp_obsolete_bpobj
)) {
6172 ASSERT(spa_feature_is_enabled(spa
,
6173 SPA_FEATURE_DEVICE_REMOVAL
));
6174 (void) bpobj_iterate_nofree(&dp
->dp_obsolete_bpobj
,
6175 increment_indirect_mapping_cb
, zcb
, NULL
);
6178 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6179 zdb_ddt_leak_init(spa
, zcb
);
6180 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6184 zdb_check_for_obsolete_leaks(vdev_t
*vd
, zdb_cb_t
*zcb
)
6186 boolean_t leaks
= B_FALSE
;
6187 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6188 uint64_t total_leaked
= 0;
6189 boolean_t are_precise
= B_FALSE
;
6191 ASSERT(vim
!= NULL
);
6193 for (uint64_t i
= 0; i
< vdev_indirect_mapping_num_entries(vim
); i
++) {
6194 vdev_indirect_mapping_entry_phys_t
*vimep
=
6195 &vim
->vim_entries
[i
];
6196 uint64_t obsolete_bytes
= 0;
6197 uint64_t offset
= DVA_MAPPING_GET_SRC_OFFSET(vimep
);
6198 metaslab_t
*msp
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
6201 * This is not very efficient but it's easy to
6202 * verify correctness.
6204 for (uint64_t inner_offset
= 0;
6205 inner_offset
< DVA_GET_ASIZE(&vimep
->vimep_dst
);
6206 inner_offset
+= 1 << vd
->vdev_ashift
) {
6207 if (range_tree_contains(msp
->ms_allocatable
,
6208 offset
+ inner_offset
, 1 << vd
->vdev_ashift
)) {
6209 obsolete_bytes
+= 1 << vd
->vdev_ashift
;
6213 int64_t bytes_leaked
= obsolete_bytes
-
6214 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
][i
];
6215 ASSERT3U(DVA_GET_ASIZE(&vimep
->vimep_dst
), >=,
6216 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
][i
]);
6218 VERIFY0(vdev_obsolete_counts_are_precise(vd
, &are_precise
));
6219 if (bytes_leaked
!= 0 && (are_precise
|| dump_opt
['d'] >= 5)) {
6220 (void) printf("obsolete indirect mapping count "
6221 "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
6222 (u_longlong_t
)vd
->vdev_id
,
6223 (u_longlong_t
)DVA_MAPPING_GET_SRC_OFFSET(vimep
),
6224 (u_longlong_t
)DVA_GET_ASIZE(&vimep
->vimep_dst
),
6225 (u_longlong_t
)bytes_leaked
);
6227 total_leaked
+= ABS(bytes_leaked
);
6230 VERIFY0(vdev_obsolete_counts_are_precise(vd
, &are_precise
));
6231 if (!are_precise
&& total_leaked
> 0) {
6232 int pct_leaked
= total_leaked
* 100 /
6233 vdev_indirect_mapping_bytes_mapped(vim
);
6234 (void) printf("cannot verify obsolete indirect mapping "
6235 "counts of vdev %llu because precise feature was not "
6236 "enabled when it was removed: %d%% (%llx bytes) of mapping"
6238 (u_longlong_t
)vd
->vdev_id
, pct_leaked
,
6239 (u_longlong_t
)total_leaked
);
6240 } else if (total_leaked
> 0) {
6241 (void) printf("obsolete indirect mapping count mismatch "
6242 "for vdev %llu -- %llx total bytes mismatched\n",
6243 (u_longlong_t
)vd
->vdev_id
,
6244 (u_longlong_t
)total_leaked
);
6248 vdev_indirect_mapping_free_obsolete_counts(vim
,
6249 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
]);
6250 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
] = NULL
;
6256 zdb_leak_fini(spa_t
*spa
, zdb_cb_t
*zcb
)
6261 boolean_t leaks
= B_FALSE
;
6262 vdev_t
*rvd
= spa
->spa_root_vdev
;
6263 for (unsigned c
= 0; c
< rvd
->vdev_children
; c
++) {
6264 vdev_t
*vd
= rvd
->vdev_child
[c
];
6266 if (zcb
->zcb_vd_obsolete_counts
[c
] != NULL
) {
6267 leaks
|= zdb_check_for_obsolete_leaks(vd
, zcb
);
6270 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
6271 metaslab_t
*msp
= vd
->vdev_ms
[m
];
6272 ASSERT3P(msp
->ms_group
, ==, (msp
->ms_group
->mg_class
==
6273 spa_embedded_log_class(spa
)) ?
6274 vd
->vdev_log_mg
: vd
->vdev_mg
);
6277 * ms_allocatable has been overloaded
6278 * to contain allocated segments. Now that
6279 * we finished traversing all blocks, any
6280 * block that remains in the ms_allocatable
6281 * represents an allocated block that we
6282 * did not claim during the traversal.
6283 * Claimed blocks would have been removed
6284 * from the ms_allocatable. For indirect
6285 * vdevs, space remaining in the tree
6286 * represents parts of the mapping that are
6287 * not referenced, which is not a bug.
6289 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
6290 range_tree_vacate(msp
->ms_allocatable
,
6293 range_tree_vacate(msp
->ms_allocatable
,
6296 if (msp
->ms_loaded
) {
6297 msp
->ms_loaded
= B_FALSE
;
6302 umem_free(zcb
->zcb_vd_obsolete_counts
,
6303 rvd
->vdev_children
* sizeof (uint32_t *));
6304 zcb
->zcb_vd_obsolete_counts
= NULL
;
6310 count_block_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6313 zdb_cb_t
*zcb
= arg
;
6315 if (dump_opt
['b'] >= 5) {
6316 char blkbuf
[BP_SPRINTF_LEN
];
6317 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
6318 (void) printf("[%s] %s\n",
6319 "deferred free", blkbuf
);
6321 zdb_count_block(zcb
, NULL
, bp
, ZDB_OT_DEFERRED
);
6326 * Iterate over livelists which have been destroyed by the user but
6327 * are still present in the MOS, waiting to be freed
6330 iterate_deleted_livelists(spa_t
*spa
, ll_iter_t func
, void *arg
)
6332 objset_t
*mos
= spa
->spa_meta_objset
;
6334 int err
= zap_lookup(mos
, DMU_POOL_DIRECTORY_OBJECT
,
6335 DMU_POOL_DELETED_CLONES
, sizeof (uint64_t), 1, &zap_obj
);
6341 zap_attribute_t attr
;
6343 /* NULL out os prior to dsl_deadlist_open in case it's garbage */
6345 for (zap_cursor_init(&zc
, mos
, zap_obj
);
6346 zap_cursor_retrieve(&zc
, &attr
) == 0;
6347 (void) zap_cursor_advance(&zc
)) {
6348 dsl_deadlist_open(&ll
, mos
, attr
.za_first_integer
);
6350 dsl_deadlist_close(&ll
);
6352 zap_cursor_fini(&zc
);
6356 bpobj_count_block_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
6360 return (count_block_cb(arg
, bp
, tx
));
6364 livelist_entry_count_blocks_cb(void *args
, dsl_deadlist_entry_t
*dle
)
6366 zdb_cb_t
*zbc
= args
;
6368 bplist_create(&blks
);
6369 /* determine which blocks have been alloc'd but not freed */
6370 VERIFY0(dsl_process_sub_livelist(&dle
->dle_bpobj
, &blks
, NULL
, NULL
));
6371 /* count those blocks */
6372 (void) bplist_iterate(&blks
, count_block_cb
, zbc
, NULL
);
6373 bplist_destroy(&blks
);
6378 livelist_count_blocks(dsl_deadlist_t
*ll
, void *arg
)
6380 dsl_deadlist_iterate(ll
, livelist_entry_count_blocks_cb
, arg
);
6384 * Count the blocks in the livelists that have been destroyed by the user
6385 * but haven't yet been freed.
6388 deleted_livelists_count_blocks(spa_t
*spa
, zdb_cb_t
*zbc
)
6390 iterate_deleted_livelists(spa
, livelist_count_blocks
, zbc
);
6394 dump_livelist_cb(dsl_deadlist_t
*ll
, void *arg
)
6396 ASSERT3P(arg
, ==, NULL
);
6397 global_feature_count
[SPA_FEATURE_LIVELIST
]++;
6398 dump_blkptr_list(ll
, "Deleted Livelist");
6399 dsl_deadlist_iterate(ll
, sublivelist_verify_lightweight
, NULL
);
6403 * Print out, register object references to, and increment feature counts for
6404 * livelists that have been destroyed by the user but haven't yet been freed.
6407 deleted_livelists_dump_mos(spa_t
*spa
)
6410 objset_t
*mos
= spa
->spa_meta_objset
;
6411 int err
= zap_lookup(mos
, DMU_POOL_DIRECTORY_OBJECT
,
6412 DMU_POOL_DELETED_CLONES
, sizeof (uint64_t), 1, &zap_obj
);
6415 mos_obj_refd(zap_obj
);
6416 iterate_deleted_livelists(spa
, dump_livelist_cb
, NULL
);
6420 dump_block_stats(spa_t
*spa
)
6423 zdb_blkstats_t
*zb
, *tzb
;
6424 uint64_t norm_alloc
, norm_space
, total_alloc
, total_found
;
6425 int flags
= TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
6426 TRAVERSE_NO_DECRYPT
| TRAVERSE_HARD
;
6427 boolean_t leaks
= B_FALSE
;
6429 bp_embedded_type_t i
;
6431 zcb
= umem_zalloc(sizeof (zdb_cb_t
), UMEM_NOFAIL
);
6433 (void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
6434 (dump_opt
['c'] || !dump_opt
['L']) ? "to verify " : "",
6435 (dump_opt
['c'] == 1) ? "metadata " : "",
6436 dump_opt
['c'] ? "checksums " : "",
6437 (dump_opt
['c'] && !dump_opt
['L']) ? "and verify " : "",
6438 !dump_opt
['L'] ? "nothing leaked " : "");
6441 * When leak detection is enabled we load all space maps as SM_ALLOC
6442 * maps, then traverse the pool claiming each block we discover. If
6443 * the pool is perfectly consistent, the segment trees will be empty
6444 * when we're done. Anything left over is a leak; any block we can't
6445 * claim (because it's not part of any space map) is a double
6446 * allocation, reference to a freed block, or an unclaimed log block.
6448 * When leak detection is disabled (-L option) we still traverse the
6449 * pool claiming each block we discover, but we skip opening any space
6452 zdb_leak_init(spa
, zcb
);
6455 * If there's a deferred-free bplist, process that first.
6457 (void) bpobj_iterate_nofree(&spa
->spa_deferred_bpobj
,
6458 bpobj_count_block_cb
, zcb
, NULL
);
6460 if (spa_version(spa
) >= SPA_VERSION_DEADLISTS
) {
6461 (void) bpobj_iterate_nofree(&spa
->spa_dsl_pool
->dp_free_bpobj
,
6462 bpobj_count_block_cb
, zcb
, NULL
);
6465 zdb_claim_removing(spa
, zcb
);
6467 if (spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
)) {
6468 VERIFY3U(0, ==, bptree_iterate(spa
->spa_meta_objset
,
6469 spa
->spa_dsl_pool
->dp_bptree_obj
, B_FALSE
, count_block_cb
,
6473 deleted_livelists_count_blocks(spa
, zcb
);
6475 if (dump_opt
['c'] > 1)
6476 flags
|= TRAVERSE_PREFETCH_DATA
;
6478 zcb
->zcb_totalasize
= metaslab_class_get_alloc(spa_normal_class(spa
));
6479 zcb
->zcb_totalasize
+= metaslab_class_get_alloc(spa_special_class(spa
));
6480 zcb
->zcb_totalasize
+= metaslab_class_get_alloc(spa_dedup_class(spa
));
6481 zcb
->zcb_totalasize
+=
6482 metaslab_class_get_alloc(spa_embedded_log_class(spa
));
6483 zcb
->zcb_start
= zcb
->zcb_lastprint
= gethrtime();
6484 err
= traverse_pool(spa
, 0, flags
, zdb_blkptr_cb
, zcb
);
6487 * If we've traversed the data blocks then we need to wait for those
6488 * I/Os to complete. We leverage "The Godfather" zio to wait on
6489 * all async I/Os to complete.
6491 if (dump_opt
['c']) {
6492 for (c
= 0; c
< max_ncpus
; c
++) {
6493 (void) zio_wait(spa
->spa_async_zio_root
[c
]);
6494 spa
->spa_async_zio_root
[c
] = zio_root(spa
, NULL
, NULL
,
6495 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
6496 ZIO_FLAG_GODFATHER
);
6499 ASSERT0(spa
->spa_load_verify_bytes
);
6502 * Done after zio_wait() since zcb_haderrors is modified in
6505 zcb
->zcb_haderrors
|= err
;
6507 if (zcb
->zcb_haderrors
) {
6508 (void) printf("\nError counts:\n\n");
6509 (void) printf("\t%5s %s\n", "errno", "count");
6510 for (e
= 0; e
< 256; e
++) {
6511 if (zcb
->zcb_errors
[e
] != 0) {
6512 (void) printf("\t%5d %llu\n",
6513 e
, (u_longlong_t
)zcb
->zcb_errors
[e
]);
6519 * Report any leaked segments.
6521 leaks
|= zdb_leak_fini(spa
, zcb
);
6523 tzb
= &zcb
->zcb_type
[ZB_TOTAL
][ZDB_OT_TOTAL
];
6525 norm_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
6526 norm_space
= metaslab_class_get_space(spa_normal_class(spa
));
6528 total_alloc
= norm_alloc
+
6529 metaslab_class_get_alloc(spa_log_class(spa
)) +
6530 metaslab_class_get_alloc(spa_embedded_log_class(spa
)) +
6531 metaslab_class_get_alloc(spa_special_class(spa
)) +
6532 metaslab_class_get_alloc(spa_dedup_class(spa
)) +
6533 get_unflushed_alloc_space(spa
);
6534 total_found
= tzb
->zb_asize
- zcb
->zcb_dedup_asize
+
6535 zcb
->zcb_removing_size
+ zcb
->zcb_checkpoint_size
;
6537 if (total_found
== total_alloc
&& !dump_opt
['L']) {
6538 (void) printf("\n\tNo leaks (block sum matches space"
6539 " maps exactly)\n");
6540 } else if (!dump_opt
['L']) {
6541 (void) printf("block traversal size %llu != alloc %llu "
6543 (u_longlong_t
)total_found
,
6544 (u_longlong_t
)total_alloc
,
6545 (dump_opt
['L']) ? "unreachable" : "leaked",
6546 (longlong_t
)(total_alloc
- total_found
));
6550 if (tzb
->zb_count
== 0) {
6551 umem_free(zcb
, sizeof (zdb_cb_t
));
6555 (void) printf("\n");
6556 (void) printf("\t%-16s %14llu\n", "bp count:",
6557 (u_longlong_t
)tzb
->zb_count
);
6558 (void) printf("\t%-16s %14llu\n", "ganged count:",
6559 (longlong_t
)tzb
->zb_gangs
);
6560 (void) printf("\t%-16s %14llu avg: %6llu\n", "bp logical:",
6561 (u_longlong_t
)tzb
->zb_lsize
,
6562 (u_longlong_t
)(tzb
->zb_lsize
/ tzb
->zb_count
));
6563 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
6564 "bp physical:", (u_longlong_t
)tzb
->zb_psize
,
6565 (u_longlong_t
)(tzb
->zb_psize
/ tzb
->zb_count
),
6566 (double)tzb
->zb_lsize
/ tzb
->zb_psize
);
6567 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
6568 "bp allocated:", (u_longlong_t
)tzb
->zb_asize
,
6569 (u_longlong_t
)(tzb
->zb_asize
/ tzb
->zb_count
),
6570 (double)tzb
->zb_lsize
/ tzb
->zb_asize
);
6571 (void) printf("\t%-16s %14llu ref>1: %6llu deduplication: %6.2f\n",
6572 "bp deduped:", (u_longlong_t
)zcb
->zcb_dedup_asize
,
6573 (u_longlong_t
)zcb
->zcb_dedup_blocks
,
6574 (double)zcb
->zcb_dedup_asize
/ tzb
->zb_asize
+ 1.0);
6575 (void) printf("\t%-16s %14llu used: %5.2f%%\n", "Normal class:",
6576 (u_longlong_t
)norm_alloc
, 100.0 * norm_alloc
/ norm_space
);
6578 if (spa_special_class(spa
)->mc_allocator
[0].mca_rotor
!= NULL
) {
6579 uint64_t alloc
= metaslab_class_get_alloc(
6580 spa_special_class(spa
));
6581 uint64_t space
= metaslab_class_get_space(
6582 spa_special_class(spa
));
6584 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
6585 "Special class", (u_longlong_t
)alloc
,
6586 100.0 * alloc
/ space
);
6589 if (spa_dedup_class(spa
)->mc_allocator
[0].mca_rotor
!= NULL
) {
6590 uint64_t alloc
= metaslab_class_get_alloc(
6591 spa_dedup_class(spa
));
6592 uint64_t space
= metaslab_class_get_space(
6593 spa_dedup_class(spa
));
6595 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
6596 "Dedup class", (u_longlong_t
)alloc
,
6597 100.0 * alloc
/ space
);
6600 if (spa_embedded_log_class(spa
)->mc_allocator
[0].mca_rotor
!= NULL
) {
6601 uint64_t alloc
= metaslab_class_get_alloc(
6602 spa_embedded_log_class(spa
));
6603 uint64_t space
= metaslab_class_get_space(
6604 spa_embedded_log_class(spa
));
6606 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
6607 "Embedded log class", (u_longlong_t
)alloc
,
6608 100.0 * alloc
/ space
);
6611 for (i
= 0; i
< NUM_BP_EMBEDDED_TYPES
; i
++) {
6612 if (zcb
->zcb_embedded_blocks
[i
] == 0)
6614 (void) printf("\n");
6615 (void) printf("\tadditional, non-pointer bps of type %u: "
6617 i
, (u_longlong_t
)zcb
->zcb_embedded_blocks
[i
]);
6619 if (dump_opt
['b'] >= 3) {
6620 (void) printf("\t number of (compressed) bytes: "
6622 dump_histogram(zcb
->zcb_embedded_histogram
[i
],
6623 sizeof (zcb
->zcb_embedded_histogram
[i
]) /
6624 sizeof (zcb
->zcb_embedded_histogram
[i
][0]), 0);
6628 if (tzb
->zb_ditto_samevdev
!= 0) {
6629 (void) printf("\tDittoed blocks on same vdev: %llu\n",
6630 (longlong_t
)tzb
->zb_ditto_samevdev
);
6632 if (tzb
->zb_ditto_same_ms
!= 0) {
6633 (void) printf("\tDittoed blocks in same metaslab: %llu\n",
6634 (longlong_t
)tzb
->zb_ditto_same_ms
);
6637 for (uint64_t v
= 0; v
< spa
->spa_root_vdev
->vdev_children
; v
++) {
6638 vdev_t
*vd
= spa
->spa_root_vdev
->vdev_child
[v
];
6639 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6646 zdb_nicenum(vdev_indirect_mapping_num_entries(vim
),
6647 mem
, vdev_indirect_mapping_size(vim
));
6649 (void) printf("\tindirect vdev id %llu has %llu segments "
6651 (longlong_t
)vd
->vdev_id
,
6652 (longlong_t
)vdev_indirect_mapping_num_entries(vim
), mem
);
6655 if (dump_opt
['b'] >= 2) {
6657 (void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
6658 "\t avg\t comp\t%%Total\tType\n");
6660 for (t
= 0; t
<= ZDB_OT_TOTAL
; t
++) {
6661 char csize
[32], lsize
[32], psize
[32], asize
[32];
6662 char avg
[32], gang
[32];
6663 const char *typename
;
6665 /* make sure nicenum has enough space */
6666 _Static_assert(sizeof (csize
) >= NN_NUMBUF_SZ
,
6668 _Static_assert(sizeof (lsize
) >= NN_NUMBUF_SZ
,
6670 _Static_assert(sizeof (psize
) >= NN_NUMBUF_SZ
,
6672 _Static_assert(sizeof (asize
) >= NN_NUMBUF_SZ
,
6674 _Static_assert(sizeof (avg
) >= NN_NUMBUF_SZ
,
6676 _Static_assert(sizeof (gang
) >= NN_NUMBUF_SZ
,
6679 if (t
< DMU_OT_NUMTYPES
)
6680 typename
= dmu_ot
[t
].ot_name
;
6682 typename
= zdb_ot_extname
[t
- DMU_OT_NUMTYPES
];
6684 if (zcb
->zcb_type
[ZB_TOTAL
][t
].zb_asize
== 0) {
6685 (void) printf("%6s\t%5s\t%5s\t%5s"
6686 "\t%5s\t%5s\t%6s\t%s\n",
6698 for (l
= ZB_TOTAL
- 1; l
>= -1; l
--) {
6699 level
= (l
== -1 ? ZB_TOTAL
: l
);
6700 zb
= &zcb
->zcb_type
[level
][t
];
6702 if (zb
->zb_asize
== 0)
6705 if (dump_opt
['b'] < 3 && level
!= ZB_TOTAL
)
6708 if (level
== 0 && zb
->zb_asize
==
6709 zcb
->zcb_type
[ZB_TOTAL
][t
].zb_asize
)
6712 zdb_nicenum(zb
->zb_count
, csize
,
6714 zdb_nicenum(zb
->zb_lsize
, lsize
,
6716 zdb_nicenum(zb
->zb_psize
, psize
,
6718 zdb_nicenum(zb
->zb_asize
, asize
,
6720 zdb_nicenum(zb
->zb_asize
/ zb
->zb_count
, avg
,
6722 zdb_nicenum(zb
->zb_gangs
, gang
, sizeof (gang
));
6724 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
6726 csize
, lsize
, psize
, asize
, avg
,
6727 (double)zb
->zb_lsize
/ zb
->zb_psize
,
6728 100.0 * zb
->zb_asize
/ tzb
->zb_asize
);
6730 if (level
== ZB_TOTAL
)
6731 (void) printf("%s\n", typename
);
6733 (void) printf(" L%d %s\n",
6736 if (dump_opt
['b'] >= 3 && zb
->zb_gangs
> 0) {
6737 (void) printf("\t number of ganged "
6738 "blocks: %s\n", gang
);
6741 if (dump_opt
['b'] >= 4) {
6742 (void) printf("psize "
6743 "(in 512-byte sectors): "
6744 "number of blocks\n");
6745 dump_histogram(zb
->zb_psize_histogram
,
6746 PSIZE_HISTO_SIZE
, 0);
6751 /* Output a table summarizing block sizes in the pool */
6752 if (dump_opt
['b'] >= 2) {
6753 dump_size_histograms(zcb
);
6757 (void) printf("\n");
6760 umem_free(zcb
, sizeof (zdb_cb_t
));
6764 if (zcb
->zcb_haderrors
) {
6765 umem_free(zcb
, sizeof (zdb_cb_t
));
6769 umem_free(zcb
, sizeof (zdb_cb_t
));
6773 typedef struct zdb_ddt_entry
{
6775 uint64_t zdde_ref_blocks
;
6776 uint64_t zdde_ref_lsize
;
6777 uint64_t zdde_ref_psize
;
6778 uint64_t zdde_ref_dsize
;
6779 avl_node_t zdde_node
;
6783 zdb_ddt_add_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
6784 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
6786 (void) zilog
, (void) dnp
;
6787 avl_tree_t
*t
= arg
;
6789 zdb_ddt_entry_t
*zdde
, zdde_search
;
6791 if (zb
->zb_level
== ZB_DNODE_LEVEL
|| BP_IS_HOLE(bp
) ||
6795 if (dump_opt
['S'] > 1 && zb
->zb_level
== ZB_ROOT_LEVEL
) {
6796 (void) printf("traversing objset %llu, %llu objects, "
6797 "%lu blocks so far\n",
6798 (u_longlong_t
)zb
->zb_objset
,
6799 (u_longlong_t
)BP_GET_FILL(bp
),
6803 if (BP_IS_HOLE(bp
) || BP_GET_CHECKSUM(bp
) == ZIO_CHECKSUM_OFF
||
6804 BP_GET_LEVEL(bp
) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)))
6807 ddt_key_fill(&zdde_search
.zdde_key
, bp
);
6809 zdde
= avl_find(t
, &zdde_search
, &where
);
6812 zdde
= umem_zalloc(sizeof (*zdde
), UMEM_NOFAIL
);
6813 zdde
->zdde_key
= zdde_search
.zdde_key
;
6814 avl_insert(t
, zdde
, where
);
6817 zdde
->zdde_ref_blocks
+= 1;
6818 zdde
->zdde_ref_lsize
+= BP_GET_LSIZE(bp
);
6819 zdde
->zdde_ref_psize
+= BP_GET_PSIZE(bp
);
6820 zdde
->zdde_ref_dsize
+= bp_get_dsize_sync(spa
, bp
);
6826 dump_simulated_ddt(spa_t
*spa
)
6829 void *cookie
= NULL
;
6830 zdb_ddt_entry_t
*zdde
;
6831 ddt_histogram_t ddh_total
= {{{0}}};
6832 ddt_stat_t dds_total
= {0};
6834 avl_create(&t
, ddt_entry_compare
,
6835 sizeof (zdb_ddt_entry_t
), offsetof(zdb_ddt_entry_t
, zdde_node
));
6837 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6839 (void) traverse_pool(spa
, 0, TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
6840 TRAVERSE_NO_DECRYPT
, zdb_ddt_add_cb
, &t
);
6842 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6844 while ((zdde
= avl_destroy_nodes(&t
, &cookie
)) != NULL
) {
6846 uint64_t refcnt
= zdde
->zdde_ref_blocks
;
6847 ASSERT(refcnt
!= 0);
6849 dds
.dds_blocks
= zdde
->zdde_ref_blocks
/ refcnt
;
6850 dds
.dds_lsize
= zdde
->zdde_ref_lsize
/ refcnt
;
6851 dds
.dds_psize
= zdde
->zdde_ref_psize
/ refcnt
;
6852 dds
.dds_dsize
= zdde
->zdde_ref_dsize
/ refcnt
;
6854 dds
.dds_ref_blocks
= zdde
->zdde_ref_blocks
;
6855 dds
.dds_ref_lsize
= zdde
->zdde_ref_lsize
;
6856 dds
.dds_ref_psize
= zdde
->zdde_ref_psize
;
6857 dds
.dds_ref_dsize
= zdde
->zdde_ref_dsize
;
6859 ddt_stat_add(&ddh_total
.ddh_stat
[highbit64(refcnt
) - 1],
6862 umem_free(zdde
, sizeof (*zdde
));
6867 ddt_histogram_stat(&dds_total
, &ddh_total
);
6869 (void) printf("Simulated DDT histogram:\n");
6871 zpool_dump_ddt(&dds_total
, &ddh_total
);
6873 dump_dedup_ratio(&dds_total
);
6877 verify_device_removal_feature_counts(spa_t
*spa
)
6879 uint64_t dr_feature_refcount
= 0;
6880 uint64_t oc_feature_refcount
= 0;
6881 uint64_t indirect_vdev_count
= 0;
6882 uint64_t precise_vdev_count
= 0;
6883 uint64_t obsolete_counts_object_count
= 0;
6884 uint64_t obsolete_sm_count
= 0;
6885 uint64_t obsolete_counts_count
= 0;
6886 uint64_t scip_count
= 0;
6887 uint64_t obsolete_bpobj_count
= 0;
6890 spa_condensing_indirect_phys_t
*scip
=
6891 &spa
->spa_condensing_indirect_phys
;
6892 if (scip
->scip_next_mapping_object
!= 0) {
6893 vdev_t
*vd
= spa
->spa_root_vdev
->vdev_child
[scip
->scip_vdev
];
6894 ASSERT(scip
->scip_prev_obsolete_sm_object
!= 0);
6895 ASSERT3P(vd
->vdev_ops
, ==, &vdev_indirect_ops
);
6897 (void) printf("Condensing indirect vdev %llu: new mapping "
6898 "object %llu, prev obsolete sm %llu\n",
6899 (u_longlong_t
)scip
->scip_vdev
,
6900 (u_longlong_t
)scip
->scip_next_mapping_object
,
6901 (u_longlong_t
)scip
->scip_prev_obsolete_sm_object
);
6902 if (scip
->scip_prev_obsolete_sm_object
!= 0) {
6903 space_map_t
*prev_obsolete_sm
= NULL
;
6904 VERIFY0(space_map_open(&prev_obsolete_sm
,
6905 spa
->spa_meta_objset
,
6906 scip
->scip_prev_obsolete_sm_object
,
6907 0, vd
->vdev_asize
, 0));
6908 dump_spacemap(spa
->spa_meta_objset
, prev_obsolete_sm
);
6909 (void) printf("\n");
6910 space_map_close(prev_obsolete_sm
);
6916 for (uint64_t i
= 0; i
< spa
->spa_root_vdev
->vdev_children
; i
++) {
6917 vdev_t
*vd
= spa
->spa_root_vdev
->vdev_child
[i
];
6918 vdev_indirect_config_t
*vic
= &vd
->vdev_indirect_config
;
6920 if (vic
->vic_mapping_object
!= 0) {
6921 ASSERT(vd
->vdev_ops
== &vdev_indirect_ops
||
6923 indirect_vdev_count
++;
6925 if (vd
->vdev_indirect_mapping
->vim_havecounts
) {
6926 obsolete_counts_count
++;
6930 boolean_t are_precise
;
6931 VERIFY0(vdev_obsolete_counts_are_precise(vd
, &are_precise
));
6933 ASSERT(vic
->vic_mapping_object
!= 0);
6934 precise_vdev_count
++;
6937 uint64_t obsolete_sm_object
;
6938 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
6939 if (obsolete_sm_object
!= 0) {
6940 ASSERT(vic
->vic_mapping_object
!= 0);
6941 obsolete_sm_count
++;
6945 (void) feature_get_refcount(spa
,
6946 &spa_feature_table
[SPA_FEATURE_DEVICE_REMOVAL
],
6947 &dr_feature_refcount
);
6948 (void) feature_get_refcount(spa
,
6949 &spa_feature_table
[SPA_FEATURE_OBSOLETE_COUNTS
],
6950 &oc_feature_refcount
);
6952 if (dr_feature_refcount
!= indirect_vdev_count
) {
6954 (void) printf("Number of indirect vdevs (%llu) " \
6955 "does not match feature count (%llu)\n",
6956 (u_longlong_t
)indirect_vdev_count
,
6957 (u_longlong_t
)dr_feature_refcount
);
6959 (void) printf("Verified device_removal feature refcount " \
6960 "of %llu is correct\n",
6961 (u_longlong_t
)dr_feature_refcount
);
6964 if (zap_contains(spa_meta_objset(spa
), DMU_POOL_DIRECTORY_OBJECT
,
6965 DMU_POOL_OBSOLETE_BPOBJ
) == 0) {
6966 obsolete_bpobj_count
++;
6970 obsolete_counts_object_count
= precise_vdev_count
;
6971 obsolete_counts_object_count
+= obsolete_sm_count
;
6972 obsolete_counts_object_count
+= obsolete_counts_count
;
6973 obsolete_counts_object_count
+= scip_count
;
6974 obsolete_counts_object_count
+= obsolete_bpobj_count
;
6975 obsolete_counts_object_count
+= remap_deadlist_count
;
6977 if (oc_feature_refcount
!= obsolete_counts_object_count
) {
6979 (void) printf("Number of obsolete counts objects (%llu) " \
6980 "does not match feature count (%llu)\n",
6981 (u_longlong_t
)obsolete_counts_object_count
,
6982 (u_longlong_t
)oc_feature_refcount
);
6983 (void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
6984 "ob:%llu rd:%llu\n",
6985 (u_longlong_t
)precise_vdev_count
,
6986 (u_longlong_t
)obsolete_sm_count
,
6987 (u_longlong_t
)obsolete_counts_count
,
6988 (u_longlong_t
)scip_count
,
6989 (u_longlong_t
)obsolete_bpobj_count
,
6990 (u_longlong_t
)remap_deadlist_count
);
6992 (void) printf("Verified indirect_refcount feature refcount " \
6993 "of %llu is correct\n",
6994 (u_longlong_t
)oc_feature_refcount
);
7000 zdb_set_skip_mmp(char *target
)
7005 * Disable the activity check to allow examination of
7008 mutex_enter(&spa_namespace_lock
);
7009 if ((spa
= spa_lookup(target
)) != NULL
) {
7010 spa
->spa_import_flags
|= ZFS_IMPORT_SKIP_MMP
;
7012 mutex_exit(&spa_namespace_lock
);
7015 #define BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7017 * Import the checkpointed state of the pool specified by the target
7018 * parameter as readonly. The function also accepts a pool config
7019 * as an optional parameter, else it attempts to infer the config by
7020 * the name of the target pool.
7022 * Note that the checkpointed state's pool name will be the name of
7023 * the original pool with the above suffix appended to it. In addition,
7024 * if the target is not a pool name (e.g. a path to a dataset) then
7025 * the new_path parameter is populated with the updated path to
7026 * reflect the fact that we are looking into the checkpointed state.
7028 * The function returns a newly-allocated copy of the name of the
7029 * pool containing the checkpointed state. When this copy is no
7030 * longer needed it should be freed with free(3C). Same thing
7031 * applies to the new_path parameter if allocated.
7034 import_checkpointed_state(char *target
, nvlist_t
*cfg
, char **new_path
)
7037 char *poolname
, *bogus_name
= NULL
;
7038 boolean_t freecfg
= B_FALSE
;
7040 /* If the target is not a pool, the extract the pool name */
7041 char *path_start
= strchr(target
, '/');
7042 if (path_start
!= NULL
) {
7043 size_t poolname_len
= path_start
- target
;
7044 poolname
= strndup(target
, poolname_len
);
7050 zdb_set_skip_mmp(poolname
);
7051 error
= spa_get_stats(poolname
, &cfg
, NULL
, 0);
7053 fatal("Tried to read config of pool \"%s\" but "
7054 "spa_get_stats() failed with error %d\n",
7060 if (asprintf(&bogus_name
, "%s%s", poolname
, BOGUS_SUFFIX
) == -1)
7062 fnvlist_add_string(cfg
, ZPOOL_CONFIG_POOL_NAME
, bogus_name
);
7064 error
= spa_import(bogus_name
, cfg
, NULL
,
7065 ZFS_IMPORT_MISSING_LOG
| ZFS_IMPORT_CHECKPOINT
|
7066 ZFS_IMPORT_SKIP_MMP
);
7070 fatal("Tried to import pool \"%s\" but spa_import() failed "
7071 "with error %d\n", bogus_name
, error
);
7074 if (new_path
!= NULL
&& path_start
!= NULL
) {
7075 if (asprintf(new_path
, "%s%s", bogus_name
, path_start
) == -1) {
7076 if (path_start
!= NULL
)
7082 if (target
!= poolname
)
7085 return (bogus_name
);
7088 typedef struct verify_checkpoint_sm_entry_cb_arg
{
7091 /* the following fields are only used for printing progress */
7092 uint64_t vcsec_entryid
;
7093 uint64_t vcsec_num_entries
;
7094 } verify_checkpoint_sm_entry_cb_arg_t
;
7096 #define ENTRIES_PER_PROGRESS_UPDATE 10000
7099 verify_checkpoint_sm_entry_cb(space_map_entry_t
*sme
, void *arg
)
7101 verify_checkpoint_sm_entry_cb_arg_t
*vcsec
= arg
;
7102 vdev_t
*vd
= vcsec
->vcsec_vd
;
7103 metaslab_t
*ms
= vd
->vdev_ms
[sme
->sme_offset
>> vd
->vdev_ms_shift
];
7104 uint64_t end
= sme
->sme_offset
+ sme
->sme_run
;
7106 ASSERT(sme
->sme_type
== SM_FREE
);
7108 if ((vcsec
->vcsec_entryid
% ENTRIES_PER_PROGRESS_UPDATE
) == 0) {
7109 (void) fprintf(stderr
,
7110 "\rverifying vdev %llu, space map entry %llu of %llu ...",
7111 (longlong_t
)vd
->vdev_id
,
7112 (longlong_t
)vcsec
->vcsec_entryid
,
7113 (longlong_t
)vcsec
->vcsec_num_entries
);
7115 vcsec
->vcsec_entryid
++;
7118 * See comment in checkpoint_sm_exclude_entry_cb()
7120 VERIFY3U(sme
->sme_offset
, >=, ms
->ms_start
);
7121 VERIFY3U(end
, <=, ms
->ms_start
+ ms
->ms_size
);
7124 * The entries in the vdev_checkpoint_sm should be marked as
7125 * allocated in the checkpointed state of the pool, therefore
7126 * their respective ms_allocateable trees should not contain them.
7128 mutex_enter(&ms
->ms_lock
);
7129 range_tree_verify_not_present(ms
->ms_allocatable
,
7130 sme
->sme_offset
, sme
->sme_run
);
7131 mutex_exit(&ms
->ms_lock
);
7137 * Verify that all segments in the vdev_checkpoint_sm are allocated
7138 * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
7141 * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
7142 * each vdev in the current state of the pool to the metaslab space maps
7143 * (ms_sm) of the checkpointed state of the pool.
7145 * Note that the function changes the state of the ms_allocatable
7146 * trees of the current spa_t. The entries of these ms_allocatable
7147 * trees are cleared out and then repopulated from with the free
7148 * entries of their respective ms_sm space maps.
7151 verify_checkpoint_vdev_spacemaps(spa_t
*checkpoint
, spa_t
*current
)
7153 vdev_t
*ckpoint_rvd
= checkpoint
->spa_root_vdev
;
7154 vdev_t
*current_rvd
= current
->spa_root_vdev
;
7156 load_concrete_ms_allocatable_trees(checkpoint
, SM_FREE
);
7158 for (uint64_t c
= 0; c
< ckpoint_rvd
->vdev_children
; c
++) {
7159 vdev_t
*ckpoint_vd
= ckpoint_rvd
->vdev_child
[c
];
7160 vdev_t
*current_vd
= current_rvd
->vdev_child
[c
];
7162 space_map_t
*checkpoint_sm
= NULL
;
7163 uint64_t checkpoint_sm_obj
;
7165 if (ckpoint_vd
->vdev_ops
== &vdev_indirect_ops
) {
7167 * Since we don't allow device removal in a pool
7168 * that has a checkpoint, we expect that all removed
7169 * vdevs were removed from the pool before the
7172 ASSERT3P(current_vd
->vdev_ops
, ==, &vdev_indirect_ops
);
7177 * If the checkpoint space map doesn't exist, then nothing
7178 * here is checkpointed so there's nothing to verify.
7180 if (current_vd
->vdev_top_zap
== 0 ||
7181 zap_contains(spa_meta_objset(current
),
7182 current_vd
->vdev_top_zap
,
7183 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) != 0)
7186 VERIFY0(zap_lookup(spa_meta_objset(current
),
7187 current_vd
->vdev_top_zap
, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
,
7188 sizeof (uint64_t), 1, &checkpoint_sm_obj
));
7190 VERIFY0(space_map_open(&checkpoint_sm
, spa_meta_objset(current
),
7191 checkpoint_sm_obj
, 0, current_vd
->vdev_asize
,
7192 current_vd
->vdev_ashift
));
7194 verify_checkpoint_sm_entry_cb_arg_t vcsec
;
7195 vcsec
.vcsec_vd
= ckpoint_vd
;
7196 vcsec
.vcsec_entryid
= 0;
7197 vcsec
.vcsec_num_entries
=
7198 space_map_length(checkpoint_sm
) / sizeof (uint64_t);
7199 VERIFY0(space_map_iterate(checkpoint_sm
,
7200 space_map_length(checkpoint_sm
),
7201 verify_checkpoint_sm_entry_cb
, &vcsec
));
7202 if (dump_opt
['m'] > 3)
7203 dump_spacemap(current
->spa_meta_objset
, checkpoint_sm
);
7204 space_map_close(checkpoint_sm
);
7208 * If we've added vdevs since we took the checkpoint, ensure
7209 * that their checkpoint space maps are empty.
7211 if (ckpoint_rvd
->vdev_children
< current_rvd
->vdev_children
) {
7212 for (uint64_t c
= ckpoint_rvd
->vdev_children
;
7213 c
< current_rvd
->vdev_children
; c
++) {
7214 vdev_t
*current_vd
= current_rvd
->vdev_child
[c
];
7215 VERIFY3P(current_vd
->vdev_checkpoint_sm
, ==, NULL
);
7219 /* for cleaner progress output */
7220 (void) fprintf(stderr
, "\n");
7224 * Verifies that all space that's allocated in the checkpoint is
7225 * still allocated in the current version, by checking that everything
7226 * in checkpoint's ms_allocatable (which is actually allocated, not
7227 * allocatable/free) is not present in current's ms_allocatable.
7229 * Note that the function changes the state of the ms_allocatable
7230 * trees of both spas when called. The entries of all ms_allocatable
7231 * trees are cleared out and then repopulated from their respective
7232 * ms_sm space maps. In the checkpointed state we load the allocated
7233 * entries, and in the current state we load the free entries.
7236 verify_checkpoint_ms_spacemaps(spa_t
*checkpoint
, spa_t
*current
)
7238 vdev_t
*ckpoint_rvd
= checkpoint
->spa_root_vdev
;
7239 vdev_t
*current_rvd
= current
->spa_root_vdev
;
7241 load_concrete_ms_allocatable_trees(checkpoint
, SM_ALLOC
);
7242 load_concrete_ms_allocatable_trees(current
, SM_FREE
);
7244 for (uint64_t i
= 0; i
< ckpoint_rvd
->vdev_children
; i
++) {
7245 vdev_t
*ckpoint_vd
= ckpoint_rvd
->vdev_child
[i
];
7246 vdev_t
*current_vd
= current_rvd
->vdev_child
[i
];
7248 if (ckpoint_vd
->vdev_ops
== &vdev_indirect_ops
) {
7250 * See comment in verify_checkpoint_vdev_spacemaps()
7252 ASSERT3P(current_vd
->vdev_ops
, ==, &vdev_indirect_ops
);
7256 for (uint64_t m
= 0; m
< ckpoint_vd
->vdev_ms_count
; m
++) {
7257 metaslab_t
*ckpoint_msp
= ckpoint_vd
->vdev_ms
[m
];
7258 metaslab_t
*current_msp
= current_vd
->vdev_ms
[m
];
7260 (void) fprintf(stderr
,
7261 "\rverifying vdev %llu of %llu, "
7262 "metaslab %llu of %llu ...",
7263 (longlong_t
)current_vd
->vdev_id
,
7264 (longlong_t
)current_rvd
->vdev_children
,
7265 (longlong_t
)current_vd
->vdev_ms
[m
]->ms_id
,
7266 (longlong_t
)current_vd
->vdev_ms_count
);
7269 * We walk through the ms_allocatable trees that
7270 * are loaded with the allocated blocks from the
7271 * ms_sm spacemaps of the checkpoint. For each
7272 * one of these ranges we ensure that none of them
7273 * exists in the ms_allocatable trees of the
7274 * current state which are loaded with the ranges
7275 * that are currently free.
7277 * This way we ensure that none of the blocks that
7278 * are part of the checkpoint were freed by mistake.
7280 range_tree_walk(ckpoint_msp
->ms_allocatable
,
7281 (range_tree_func_t
*)range_tree_verify_not_present
,
7282 current_msp
->ms_allocatable
);
7286 /* for cleaner progress output */
7287 (void) fprintf(stderr
, "\n");
7291 verify_checkpoint_blocks(spa_t
*spa
)
7293 ASSERT(!dump_opt
['L']);
7295 spa_t
*checkpoint_spa
;
7296 char *checkpoint_pool
;
7300 * We import the checkpointed state of the pool (under a different
7301 * name) so we can do verification on it against the current state
7304 checkpoint_pool
= import_checkpointed_state(spa
->spa_name
, NULL
,
7306 ASSERT(strcmp(spa
->spa_name
, checkpoint_pool
) != 0);
7308 error
= spa_open(checkpoint_pool
, &checkpoint_spa
, FTAG
);
7310 fatal("Tried to open pool \"%s\" but spa_open() failed with "
7311 "error %d\n", checkpoint_pool
, error
);
7315 * Ensure that ranges in the checkpoint space maps of each vdev
7316 * are allocated according to the checkpointed state's metaslab
7319 verify_checkpoint_vdev_spacemaps(checkpoint_spa
, spa
);
7322 * Ensure that allocated ranges in the checkpoint's metaslab
7323 * space maps remain allocated in the metaslab space maps of
7324 * the current state.
7326 verify_checkpoint_ms_spacemaps(checkpoint_spa
, spa
);
7329 * Once we are done, we get rid of the checkpointed state.
7331 spa_close(checkpoint_spa
, FTAG
);
7332 free(checkpoint_pool
);
7336 dump_leftover_checkpoint_blocks(spa_t
*spa
)
7338 vdev_t
*rvd
= spa
->spa_root_vdev
;
7340 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
7341 vdev_t
*vd
= rvd
->vdev_child
[i
];
7343 space_map_t
*checkpoint_sm
= NULL
;
7344 uint64_t checkpoint_sm_obj
;
7346 if (vd
->vdev_top_zap
== 0)
7349 if (zap_contains(spa_meta_objset(spa
), vd
->vdev_top_zap
,
7350 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) != 0)
7353 VERIFY0(zap_lookup(spa_meta_objset(spa
), vd
->vdev_top_zap
,
7354 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
,
7355 sizeof (uint64_t), 1, &checkpoint_sm_obj
));
7357 VERIFY0(space_map_open(&checkpoint_sm
, spa_meta_objset(spa
),
7358 checkpoint_sm_obj
, 0, vd
->vdev_asize
, vd
->vdev_ashift
));
7359 dump_spacemap(spa
->spa_meta_objset
, checkpoint_sm
);
7360 space_map_close(checkpoint_sm
);
7365 verify_checkpoint(spa_t
*spa
)
7367 uberblock_t checkpoint
;
7370 if (!spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
))
7373 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
7374 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
7375 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
7377 if (error
== ENOENT
&& !dump_opt
['L']) {
7379 * If the feature is active but the uberblock is missing
7380 * then we must be in the middle of discarding the
7383 (void) printf("\nPartially discarded checkpoint "
7385 if (dump_opt
['m'] > 3)
7386 dump_leftover_checkpoint_blocks(spa
);
7388 } else if (error
!= 0) {
7389 (void) printf("lookup error %d when looking for "
7390 "checkpointed uberblock in MOS\n", error
);
7393 dump_uberblock(&checkpoint
, "\nCheckpointed uberblock found:\n", "\n");
7395 if (checkpoint
.ub_checkpoint_txg
== 0) {
7396 (void) printf("\nub_checkpoint_txg not set in checkpointed "
7401 if (error
== 0 && !dump_opt
['L'])
7402 verify_checkpoint_blocks(spa
);
7408 mos_leaks_cb(void *arg
, uint64_t start
, uint64_t size
)
7411 for (uint64_t i
= start
; i
< size
; i
++) {
7412 (void) printf("MOS object %llu referenced but not allocated\n",
7418 mos_obj_refd(uint64_t obj
)
7420 if (obj
!= 0 && mos_refd_objs
!= NULL
)
7421 range_tree_add(mos_refd_objs
, obj
, 1);
7425 * Call on a MOS object that may already have been referenced.
7428 mos_obj_refd_multiple(uint64_t obj
)
7430 if (obj
!= 0 && mos_refd_objs
!= NULL
&&
7431 !range_tree_contains(mos_refd_objs
, obj
, 1))
7432 range_tree_add(mos_refd_objs
, obj
, 1);
7436 mos_leak_vdev_top_zap(vdev_t
*vd
)
7438 uint64_t ms_flush_data_obj
;
7439 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
),
7440 vd
->vdev_top_zap
, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
,
7441 sizeof (ms_flush_data_obj
), 1, &ms_flush_data_obj
);
7442 if (error
== ENOENT
)
7446 mos_obj_refd(ms_flush_data_obj
);
7450 mos_leak_vdev(vdev_t
*vd
)
7452 mos_obj_refd(vd
->vdev_dtl_object
);
7453 mos_obj_refd(vd
->vdev_ms_array
);
7454 mos_obj_refd(vd
->vdev_indirect_config
.vic_births_object
);
7455 mos_obj_refd(vd
->vdev_indirect_config
.vic_mapping_object
);
7456 mos_obj_refd(vd
->vdev_leaf_zap
);
7457 if (vd
->vdev_checkpoint_sm
!= NULL
)
7458 mos_obj_refd(vd
->vdev_checkpoint_sm
->sm_object
);
7459 if (vd
->vdev_indirect_mapping
!= NULL
) {
7460 mos_obj_refd(vd
->vdev_indirect_mapping
->
7461 vim_phys
->vimp_counts_object
);
7463 if (vd
->vdev_obsolete_sm
!= NULL
)
7464 mos_obj_refd(vd
->vdev_obsolete_sm
->sm_object
);
7466 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
7467 metaslab_t
*ms
= vd
->vdev_ms
[m
];
7468 mos_obj_refd(space_map_object(ms
->ms_sm
));
7471 if (vd
->vdev_top_zap
!= 0) {
7472 mos_obj_refd(vd
->vdev_top_zap
);
7473 mos_leak_vdev_top_zap(vd
);
7476 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++) {
7477 mos_leak_vdev(vd
->vdev_child
[c
]);
7482 mos_leak_log_spacemaps(spa_t
*spa
)
7484 uint64_t spacemap_zap
;
7485 int error
= zap_lookup(spa_meta_objset(spa
),
7486 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_LOG_SPACEMAP_ZAP
,
7487 sizeof (spacemap_zap
), 1, &spacemap_zap
);
7488 if (error
== ENOENT
)
7492 mos_obj_refd(spacemap_zap
);
7493 for (spa_log_sm_t
*sls
= avl_first(&spa
->spa_sm_logs_by_txg
);
7494 sls
; sls
= AVL_NEXT(&spa
->spa_sm_logs_by_txg
, sls
))
7495 mos_obj_refd(sls
->sls_sm_obj
);
7499 dump_mos_leaks(spa_t
*spa
)
7502 objset_t
*mos
= spa
->spa_meta_objset
;
7503 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7505 /* Visit and mark all referenced objects in the MOS */
7507 mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT
);
7508 mos_obj_refd(spa
->spa_pool_props_object
);
7509 mos_obj_refd(spa
->spa_config_object
);
7510 mos_obj_refd(spa
->spa_ddt_stat_object
);
7511 mos_obj_refd(spa
->spa_feat_desc_obj
);
7512 mos_obj_refd(spa
->spa_feat_enabled_txg_obj
);
7513 mos_obj_refd(spa
->spa_feat_for_read_obj
);
7514 mos_obj_refd(spa
->spa_feat_for_write_obj
);
7515 mos_obj_refd(spa
->spa_history
);
7516 mos_obj_refd(spa
->spa_errlog_last
);
7517 mos_obj_refd(spa
->spa_errlog_scrub
);
7518 mos_obj_refd(spa
->spa_all_vdev_zaps
);
7519 mos_obj_refd(spa
->spa_dsl_pool
->dp_bptree_obj
);
7520 mos_obj_refd(spa
->spa_dsl_pool
->dp_tmp_userrefs_obj
);
7521 mos_obj_refd(spa
->spa_dsl_pool
->dp_scan
->scn_phys
.scn_queue_obj
);
7522 bpobj_count_refd(&spa
->spa_deferred_bpobj
);
7523 mos_obj_refd(dp
->dp_empty_bpobj
);
7524 bpobj_count_refd(&dp
->dp_obsolete_bpobj
);
7525 bpobj_count_refd(&dp
->dp_free_bpobj
);
7526 mos_obj_refd(spa
->spa_l2cache
.sav_object
);
7527 mos_obj_refd(spa
->spa_spares
.sav_object
);
7529 if (spa
->spa_syncing_log_sm
!= NULL
)
7530 mos_obj_refd(spa
->spa_syncing_log_sm
->sm_object
);
7531 mos_leak_log_spacemaps(spa
);
7533 mos_obj_refd(spa
->spa_condensing_indirect_phys
.
7534 scip_next_mapping_object
);
7535 mos_obj_refd(spa
->spa_condensing_indirect_phys
.
7536 scip_prev_obsolete_sm_object
);
7537 if (spa
->spa_condensing_indirect_phys
.scip_next_mapping_object
!= 0) {
7538 vdev_indirect_mapping_t
*vim
=
7539 vdev_indirect_mapping_open(mos
,
7540 spa
->spa_condensing_indirect_phys
.scip_next_mapping_object
);
7541 mos_obj_refd(vim
->vim_phys
->vimp_counts_object
);
7542 vdev_indirect_mapping_close(vim
);
7544 deleted_livelists_dump_mos(spa
);
7546 if (dp
->dp_origin_snap
!= NULL
) {
7549 dsl_pool_config_enter(dp
, FTAG
);
7550 VERIFY0(dsl_dataset_hold_obj(dp
,
7551 dsl_dataset_phys(dp
->dp_origin_snap
)->ds_next_snap_obj
,
7553 count_ds_mos_objects(ds
);
7554 dump_blkptr_list(&ds
->ds_deadlist
, "Deadlist");
7555 dsl_dataset_rele(ds
, FTAG
);
7556 dsl_pool_config_exit(dp
, FTAG
);
7558 count_ds_mos_objects(dp
->dp_origin_snap
);
7559 dump_blkptr_list(&dp
->dp_origin_snap
->ds_deadlist
, "Deadlist");
7561 count_dir_mos_objects(dp
->dp_mos_dir
);
7562 if (dp
->dp_free_dir
!= NULL
)
7563 count_dir_mos_objects(dp
->dp_free_dir
);
7564 if (dp
->dp_leak_dir
!= NULL
)
7565 count_dir_mos_objects(dp
->dp_leak_dir
);
7567 mos_leak_vdev(spa
->spa_root_vdev
);
7569 for (uint64_t class = 0; class < DDT_CLASSES
; class++) {
7570 for (uint64_t type
= 0; type
< DDT_TYPES
; type
++) {
7571 for (uint64_t cksum
= 0;
7572 cksum
< ZIO_CHECKSUM_FUNCTIONS
; cksum
++) {
7573 ddt_t
*ddt
= spa
->spa_ddt
[cksum
];
7574 mos_obj_refd(ddt
->ddt_object
[type
][class]);
7580 * Visit all allocated objects and make sure they are referenced.
7582 uint64_t object
= 0;
7583 while (dmu_object_next(mos
, &object
, B_FALSE
, 0) == 0) {
7584 if (range_tree_contains(mos_refd_objs
, object
, 1)) {
7585 range_tree_remove(mos_refd_objs
, object
, 1);
7587 dmu_object_info_t doi
;
7589 dmu_object_info(mos
, object
, &doi
);
7590 if (doi
.doi_type
& DMU_OT_NEWTYPE
) {
7591 dmu_object_byteswap_t bswap
=
7592 DMU_OT_BYTESWAP(doi
.doi_type
);
7593 name
= dmu_ot_byteswap
[bswap
].ob_name
;
7595 name
= dmu_ot
[doi
.doi_type
].ot_name
;
7598 (void) printf("MOS object %llu (%s) leaked\n",
7599 (u_longlong_t
)object
, name
);
7603 (void) range_tree_walk(mos_refd_objs
, mos_leaks_cb
, NULL
);
7604 if (!range_tree_is_empty(mos_refd_objs
))
7606 range_tree_vacate(mos_refd_objs
, NULL
, NULL
);
7607 range_tree_destroy(mos_refd_objs
);
7611 typedef struct log_sm_obsolete_stats_arg
{
7612 uint64_t lsos_current_txg
;
7614 uint64_t lsos_total_entries
;
7615 uint64_t lsos_valid_entries
;
7617 uint64_t lsos_sm_entries
;
7618 uint64_t lsos_valid_sm_entries
;
7619 } log_sm_obsolete_stats_arg_t
;
7622 log_spacemap_obsolete_stats_cb(spa_t
*spa
, space_map_entry_t
*sme
,
7623 uint64_t txg
, void *arg
)
7625 log_sm_obsolete_stats_arg_t
*lsos
= arg
;
7627 uint64_t offset
= sme
->sme_offset
;
7628 uint64_t vdev_id
= sme
->sme_vdev
;
7630 if (lsos
->lsos_current_txg
== 0) {
7631 /* this is the first log */
7632 lsos
->lsos_current_txg
= txg
;
7633 } else if (lsos
->lsos_current_txg
< txg
) {
7634 /* we just changed log - print stats and reset */
7635 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
7636 (u_longlong_t
)lsos
->lsos_valid_sm_entries
,
7637 (u_longlong_t
)lsos
->lsos_sm_entries
,
7638 (u_longlong_t
)lsos
->lsos_current_txg
);
7639 lsos
->lsos_valid_sm_entries
= 0;
7640 lsos
->lsos_sm_entries
= 0;
7641 lsos
->lsos_current_txg
= txg
;
7643 ASSERT3U(lsos
->lsos_current_txg
, ==, txg
);
7645 lsos
->lsos_sm_entries
++;
7646 lsos
->lsos_total_entries
++;
7648 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
7649 if (!vdev_is_concrete(vd
))
7652 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
7653 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
7655 if (txg
< metaslab_unflushed_txg(ms
))
7657 lsos
->lsos_valid_sm_entries
++;
7658 lsos
->lsos_valid_entries
++;
7663 dump_log_spacemap_obsolete_stats(spa_t
*spa
)
7665 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
7668 log_sm_obsolete_stats_arg_t lsos
= {0};
7670 (void) printf("Log Space Map Obsolete Entry Statistics:\n");
7672 iterate_through_spacemap_logs(spa
,
7673 log_spacemap_obsolete_stats_cb
, &lsos
);
7675 /* print stats for latest log */
7676 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
7677 (u_longlong_t
)lsos
.lsos_valid_sm_entries
,
7678 (u_longlong_t
)lsos
.lsos_sm_entries
,
7679 (u_longlong_t
)lsos
.lsos_current_txg
);
7681 (void) printf("%-8llu valid entries out of %-8llu - total\n\n",
7682 (u_longlong_t
)lsos
.lsos_valid_entries
,
7683 (u_longlong_t
)lsos
.lsos_total_entries
);
7687 dump_zpool(spa_t
*spa
)
7689 dsl_pool_t
*dp
= spa_get_dsl(spa
);
7692 if (dump_opt
['y']) {
7693 livelist_metaslab_validate(spa
);
7696 if (dump_opt
['S']) {
7697 dump_simulated_ddt(spa
);
7701 if (!dump_opt
['e'] && dump_opt
['C'] > 1) {
7702 (void) printf("\nCached configuration:\n");
7703 dump_nvlist(spa
->spa_config
, 8);
7710 dump_uberblock(&spa
->spa_uberblock
, "\nUberblock:\n", "\n");
7715 if (dump_opt
['d'] > 2 || dump_opt
['m'])
7716 dump_metaslabs(spa
);
7718 dump_metaslab_groups(spa
, dump_opt
['M'] > 1);
7719 if (dump_opt
['d'] > 2 || dump_opt
['m']) {
7720 dump_log_spacemaps(spa
);
7721 dump_log_spacemap_obsolete_stats(spa
);
7724 if (dump_opt
['d'] || dump_opt
['i']) {
7726 mos_refd_objs
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0,
7728 dump_objset(dp
->dp_meta_objset
);
7730 if (dump_opt
['d'] >= 3) {
7731 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7732 dump_full_bpobj(&spa
->spa_deferred_bpobj
,
7733 "Deferred frees", 0);
7734 if (spa_version(spa
) >= SPA_VERSION_DEADLISTS
) {
7735 dump_full_bpobj(&dp
->dp_free_bpobj
,
7736 "Pool snapshot frees", 0);
7738 if (bpobj_is_open(&dp
->dp_obsolete_bpobj
)) {
7739 ASSERT(spa_feature_is_enabled(spa
,
7740 SPA_FEATURE_DEVICE_REMOVAL
));
7741 dump_full_bpobj(&dp
->dp_obsolete_bpobj
,
7742 "Pool obsolete blocks", 0);
7745 if (spa_feature_is_active(spa
,
7746 SPA_FEATURE_ASYNC_DESTROY
)) {
7747 dump_bptree(spa
->spa_meta_objset
,
7749 "Pool dataset frees");
7751 dump_dtl(spa
->spa_root_vdev
, 0);
7754 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++)
7755 global_feature_count
[f
] = UINT64_MAX
;
7756 global_feature_count
[SPA_FEATURE_REDACTION_BOOKMARKS
] = 0;
7757 global_feature_count
[SPA_FEATURE_BOOKMARK_WRITTEN
] = 0;
7758 global_feature_count
[SPA_FEATURE_LIVELIST
] = 0;
7760 (void) dmu_objset_find(spa_name(spa
), dump_one_objset
,
7761 NULL
, DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
7763 if (rc
== 0 && !dump_opt
['L'])
7764 rc
= dump_mos_leaks(spa
);
7766 for (f
= 0; f
< SPA_FEATURES
; f
++) {
7770 if (!(spa_feature_table
[f
].fi_flags
&
7771 ZFEATURE_FLAG_PER_DATASET
)) {
7772 if (global_feature_count
[f
] == UINT64_MAX
)
7774 if (!spa_feature_is_enabled(spa
, f
)) {
7775 ASSERT0(global_feature_count
[f
]);
7778 arr
= global_feature_count
;
7780 if (!spa_feature_is_enabled(spa
, f
)) {
7781 ASSERT0(dataset_feature_count
[f
]);
7784 arr
= dataset_feature_count
;
7786 if (feature_get_refcount(spa
, &spa_feature_table
[f
],
7787 &refcount
) == ENOTSUP
)
7789 if (arr
[f
] != refcount
) {
7790 (void) printf("%s feature refcount mismatch: "
7791 "%lld consumers != %lld refcount\n",
7792 spa_feature_table
[f
].fi_uname
,
7793 (longlong_t
)arr
[f
], (longlong_t
)refcount
);
7796 (void) printf("Verified %s feature refcount "
7797 "of %llu is correct\n",
7798 spa_feature_table
[f
].fi_uname
,
7799 (longlong_t
)refcount
);
7804 rc
= verify_device_removal_feature_counts(spa
);
7807 if (rc
== 0 && (dump_opt
['b'] || dump_opt
['c']))
7808 rc
= dump_block_stats(spa
);
7811 rc
= verify_spacemap_refcounts(spa
);
7814 show_pool_stats(spa
);
7820 rc
= verify_checkpoint(spa
);
7823 dump_debug_buffer();
7828 #define ZDB_FLAG_CHECKSUM 0x0001
7829 #define ZDB_FLAG_DECOMPRESS 0x0002
7830 #define ZDB_FLAG_BSWAP 0x0004
7831 #define ZDB_FLAG_GBH 0x0008
7832 #define ZDB_FLAG_INDIRECT 0x0010
7833 #define ZDB_FLAG_RAW 0x0020
7834 #define ZDB_FLAG_PRINT_BLKPTR 0x0040
7835 #define ZDB_FLAG_VERBOSE 0x0080
7837 static int flagbits
[256];
7838 static char flagbitstr
[16];
7841 zdb_print_blkptr(const blkptr_t
*bp
, int flags
)
7843 char blkbuf
[BP_SPRINTF_LEN
];
7845 if (flags
& ZDB_FLAG_BSWAP
)
7846 byteswap_uint64_array((void *)bp
, sizeof (blkptr_t
));
7848 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
7849 (void) printf("%s\n", blkbuf
);
7853 zdb_dump_indirect(blkptr_t
*bp
, int nbps
, int flags
)
7857 for (i
= 0; i
< nbps
; i
++)
7858 zdb_print_blkptr(&bp
[i
], flags
);
7862 zdb_dump_gbh(void *buf
, int flags
)
7864 zdb_dump_indirect((blkptr_t
*)buf
, SPA_GBH_NBLKPTRS
, flags
);
7868 zdb_dump_block_raw(void *buf
, uint64_t size
, int flags
)
7870 if (flags
& ZDB_FLAG_BSWAP
)
7871 byteswap_uint64_array(buf
, size
);
7872 VERIFY(write(fileno(stdout
), buf
, size
) == size
);
7876 zdb_dump_block(char *label
, void *buf
, uint64_t size
, int flags
)
7878 uint64_t *d
= (uint64_t *)buf
;
7879 unsigned nwords
= size
/ sizeof (uint64_t);
7880 int do_bswap
= !!(flags
& ZDB_FLAG_BSWAP
);
7887 hdr
= " 7 6 5 4 3 2 1 0 f e d c b a 9 8";
7889 hdr
= " 0 1 2 3 4 5 6 7 8 9 a b c d e f";
7891 (void) printf("\n%s\n%6s %s 0123456789abcdef\n", label
, "", hdr
);
7893 #ifdef _LITTLE_ENDIAN
7894 /* correct the endianness */
7895 do_bswap
= !do_bswap
;
7897 for (i
= 0; i
< nwords
; i
+= 2) {
7898 (void) printf("%06llx: %016llx %016llx ",
7899 (u_longlong_t
)(i
* sizeof (uint64_t)),
7900 (u_longlong_t
)(do_bswap
? BSWAP_64(d
[i
]) : d
[i
]),
7901 (u_longlong_t
)(do_bswap
? BSWAP_64(d
[i
+ 1]) : d
[i
+ 1]));
7904 for (j
= 0; j
< 2 * sizeof (uint64_t); j
++)
7905 (void) printf("%c", isprint(c
[j
]) ? c
[j
] : '.');
7906 (void) printf("\n");
7911 * There are two acceptable formats:
7912 * leaf_name - For example: c1t0d0 or /tmp/ztest.0a
7913 * child[.child]* - For example: 0.1.1
7915 * The second form can be used to specify arbitrary vdevs anywhere
7916 * in the hierarchy. For example, in a pool with a mirror of
7917 * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
7920 zdb_vdev_lookup(vdev_t
*vdev
, const char *path
)
7928 /* First, assume the x.x.x.x format */
7929 i
= strtoul(path
, &s
, 10);
7930 if (s
== path
|| (s
&& *s
!= '.' && *s
!= '\0'))
7932 if (i
>= vdev
->vdev_children
)
7935 vdev
= vdev
->vdev_child
[i
];
7936 if (s
&& *s
== '\0')
7938 return (zdb_vdev_lookup(vdev
, s
+1));
7941 for (i
= 0; i
< vdev
->vdev_children
; i
++) {
7942 vdev_t
*vc
= vdev
->vdev_child
[i
];
7944 if (vc
->vdev_path
== NULL
) {
7945 vc
= zdb_vdev_lookup(vc
, path
);
7952 p
= strrchr(vc
->vdev_path
, '/');
7953 p
= p
? p
+ 1 : vc
->vdev_path
;
7954 q
= &vc
->vdev_path
[strlen(vc
->vdev_path
) - 2];
7956 if (strcmp(vc
->vdev_path
, path
) == 0)
7958 if (strcmp(p
, path
) == 0)
7960 if (strcmp(q
, "s0") == 0 && strncmp(p
, path
, q
- p
) == 0)
7968 name_from_objset_id(spa_t
*spa
, uint64_t objset_id
, char *outstr
)
7972 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
7973 int error
= dsl_dataset_hold_obj(spa
->spa_dsl_pool
, objset_id
,
7976 (void) fprintf(stderr
, "failed to hold objset %llu: %s\n",
7977 (u_longlong_t
)objset_id
, strerror(error
));
7978 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
7981 dsl_dataset_name(ds
, outstr
);
7982 dsl_dataset_rele(ds
, NULL
);
7983 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
7988 zdb_parse_block_sizes(char *sizes
, uint64_t *lsize
, uint64_t *psize
)
7990 char *s0
, *s1
, *tmp
= NULL
;
7995 s0
= strtok_r(sizes
, "/", &tmp
);
7998 s1
= strtok_r(NULL
, "/", &tmp
);
7999 *lsize
= strtoull(s0
, NULL
, 16);
8000 *psize
= s1
? strtoull(s1
, NULL
, 16) : *lsize
;
8001 return (*lsize
>= *psize
&& *psize
> 0);
8004 #define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg))
8007 zdb_decompress_block(abd_t
*pabd
, void *buf
, void *lbuf
, uint64_t lsize
,
8008 uint64_t psize
, int flags
)
8011 boolean_t exceeded
= B_FALSE
;
8013 * We don't know how the data was compressed, so just try
8014 * every decompress function at every inflated blocksize.
8016 void *lbuf2
= umem_alloc(SPA_MAXBLOCKSIZE
, UMEM_NOFAIL
);
8017 int cfuncs
[ZIO_COMPRESS_FUNCTIONS
] = { 0 };
8018 int *cfuncp
= cfuncs
;
8019 uint64_t maxlsize
= SPA_MAXBLOCKSIZE
;
8020 uint64_t mask
= ZIO_COMPRESS_MASK(ON
) | ZIO_COMPRESS_MASK(OFF
) |
8021 ZIO_COMPRESS_MASK(INHERIT
) | ZIO_COMPRESS_MASK(EMPTY
) |
8022 (getenv("ZDB_NO_ZLE") ? ZIO_COMPRESS_MASK(ZLE
) : 0);
8023 *cfuncp
++ = ZIO_COMPRESS_LZ4
;
8024 *cfuncp
++ = ZIO_COMPRESS_LZJB
;
8025 mask
|= ZIO_COMPRESS_MASK(LZ4
) | ZIO_COMPRESS_MASK(LZJB
);
8026 for (int c
= 0; c
< ZIO_COMPRESS_FUNCTIONS
; c
++)
8027 if (((1ULL << c
) & mask
) == 0)
8031 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
8032 * could take a while and we should let the user know
8033 * we are not stuck. On the other hand, printing progress
8034 * info gets old after a while. User can specify 'v' flag
8035 * to see the progression.
8038 lsize
+= SPA_MINBLOCKSIZE
;
8041 for (; lsize
<= maxlsize
; lsize
+= SPA_MINBLOCKSIZE
) {
8042 for (cfuncp
= cfuncs
; *cfuncp
; cfuncp
++) {
8043 if (flags
& ZDB_FLAG_VERBOSE
) {
8044 (void) fprintf(stderr
,
8045 "Trying %05llx -> %05llx (%s)\n",
8046 (u_longlong_t
)psize
,
8047 (u_longlong_t
)lsize
,
8048 zio_compress_table
[*cfuncp
].\
8053 * We randomize lbuf2, and decompress to both
8054 * lbuf and lbuf2. This way, we will know if
8055 * decompression fill exactly to lsize.
8057 VERIFY0(random_get_pseudo_bytes(lbuf2
, lsize
));
8059 if (zio_decompress_data(*cfuncp
, pabd
,
8060 lbuf
, psize
, lsize
, NULL
) == 0 &&
8061 zio_decompress_data(*cfuncp
, pabd
,
8062 lbuf2
, psize
, lsize
, NULL
) == 0 &&
8063 memcmp(lbuf
, lbuf2
, lsize
) == 0)
8069 umem_free(lbuf2
, SPA_MAXBLOCKSIZE
);
8071 if (lsize
> maxlsize
) {
8074 if (*cfuncp
== ZIO_COMPRESS_ZLE
) {
8075 printf("\nZLE decompression was selected. If you "
8076 "suspect the results are wrong,\ntry avoiding ZLE "
8077 "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
8084 * Read a block from a pool and print it out. The syntax of the
8085 * block descriptor is:
8087 * pool:vdev_specifier:offset:[lsize/]psize[:flags]
8089 * pool - The name of the pool you wish to read from
8090 * vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
8091 * offset - offset, in hex, in bytes
8092 * size - Amount of data to read, in hex, in bytes
8093 * flags - A string of characters specifying options
8094 * b: Decode a blkptr at given offset within block
8095 * c: Calculate and display checksums
8096 * d: Decompress data before dumping
8097 * e: Byteswap data before dumping
8098 * g: Display data as a gang block header
8099 * i: Display as an indirect block
8100 * r: Dump raw data to stdout
8105 zdb_read_block(char *thing
, spa_t
*spa
)
8107 blkptr_t blk
, *bp
= &blk
;
8108 dva_t
*dva
= bp
->blk_dva
;
8110 uint64_t offset
= 0, psize
= 0, lsize
= 0, blkptr_offset
= 0;
8115 char *s
, *p
, *dup
, *flagstr
, *sizes
, *tmp
= NULL
;
8116 const char *vdev
, *errmsg
= NULL
;
8118 boolean_t borrowed
= B_FALSE
, found
= B_FALSE
;
8120 dup
= strdup(thing
);
8121 s
= strtok_r(dup
, ":", &tmp
);
8123 s
= strtok_r(NULL
, ":", &tmp
);
8124 offset
= strtoull(s
? s
: "", NULL
, 16);
8125 sizes
= strtok_r(NULL
, ":", &tmp
);
8126 s
= strtok_r(NULL
, ":", &tmp
);
8127 flagstr
= strdup(s
?: "");
8129 if (!zdb_parse_block_sizes(sizes
, &lsize
, &psize
))
8130 errmsg
= "invalid size(s)";
8131 if (!IS_P2ALIGNED(psize
, DEV_BSIZE
) || !IS_P2ALIGNED(lsize
, DEV_BSIZE
))
8132 errmsg
= "size must be a multiple of sector size";
8133 if (!IS_P2ALIGNED(offset
, DEV_BSIZE
))
8134 errmsg
= "offset must be a multiple of sector size";
8136 (void) printf("Invalid block specifier: %s - %s\n",
8142 for (s
= strtok_r(flagstr
, ":", &tmp
);
8144 s
= strtok_r(NULL
, ":", &tmp
)) {
8145 for (i
= 0; i
< strlen(flagstr
); i
++) {
8146 int bit
= flagbits
[(uchar_t
)flagstr
[i
]];
8149 (void) printf("***Ignoring flag: %c\n",
8150 (uchar_t
)flagstr
[i
]);
8156 p
= &flagstr
[i
+ 1];
8157 if (*p
!= ':' && *p
!= '\0') {
8158 int j
= 0, nextbit
= flagbits
[(uchar_t
)*p
];
8159 char *end
, offstr
[8] = { 0 };
8160 if ((bit
== ZDB_FLAG_PRINT_BLKPTR
) &&
8162 /* look ahead to isolate the offset */
8163 while (nextbit
== 0 &&
8164 strchr(flagbitstr
, *p
) == NULL
) {
8167 if (i
+ j
> strlen(flagstr
))
8170 nextbit
= flagbits
[(uchar_t
)*p
];
8172 blkptr_offset
= strtoull(offstr
, &end
,
8175 } else if (nextbit
== 0) {
8176 (void) printf("***Ignoring flag arg:"
8177 " '%c'\n", (uchar_t
)*p
);
8182 if (blkptr_offset
% sizeof (blkptr_t
)) {
8183 printf("Block pointer offset 0x%llx "
8184 "must be divisible by 0x%x\n",
8185 (longlong_t
)blkptr_offset
, (int)sizeof (blkptr_t
));
8188 if (found
== B_FALSE
&& strlen(flagstr
) > 0) {
8189 printf("Invalid flag arg: '%s'\n", flagstr
);
8193 vd
= zdb_vdev_lookup(spa
->spa_root_vdev
, vdev
);
8195 (void) printf("***Invalid vdev: %s\n", vdev
);
8200 (void) fprintf(stderr
, "Found vdev: %s\n",
8203 (void) fprintf(stderr
, "Found vdev type: %s\n",
8204 vd
->vdev_ops
->vdev_op_type
);
8207 pabd
= abd_alloc_for_io(SPA_MAXBLOCKSIZE
, B_FALSE
);
8208 lbuf
= umem_alloc(SPA_MAXBLOCKSIZE
, UMEM_NOFAIL
);
8212 DVA_SET_VDEV(&dva
[0], vd
->vdev_id
);
8213 DVA_SET_OFFSET(&dva
[0], offset
);
8214 DVA_SET_GANG(&dva
[0], !!(flags
& ZDB_FLAG_GBH
));
8215 DVA_SET_ASIZE(&dva
[0], vdev_psize_to_asize(vd
, psize
));
8217 BP_SET_BIRTH(bp
, TXG_INITIAL
, TXG_INITIAL
);
8219 BP_SET_LSIZE(bp
, lsize
);
8220 BP_SET_PSIZE(bp
, psize
);
8221 BP_SET_COMPRESS(bp
, ZIO_COMPRESS_OFF
);
8222 BP_SET_CHECKSUM(bp
, ZIO_CHECKSUM_OFF
);
8223 BP_SET_TYPE(bp
, DMU_OT_NONE
);
8224 BP_SET_LEVEL(bp
, 0);
8225 BP_SET_DEDUP(bp
, 0);
8226 BP_SET_BYTEORDER(bp
, ZFS_HOST_BYTEORDER
);
8228 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8229 zio
= zio_root(spa
, NULL
, NULL
, 0);
8231 if (vd
== vd
->vdev_top
) {
8233 * Treat this as a normal block read.
8235 zio_nowait(zio_read(zio
, spa
, bp
, pabd
, psize
, NULL
, NULL
,
8236 ZIO_PRIORITY_SYNC_READ
,
8237 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
, NULL
));
8240 * Treat this as a vdev child I/O.
8242 zio_nowait(zio_vdev_child_io(zio
, bp
, vd
, offset
, pabd
,
8243 psize
, ZIO_TYPE_READ
, ZIO_PRIORITY_SYNC_READ
,
8244 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_PROPAGATE
|
8245 ZIO_FLAG_DONT_RETRY
| ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
|
8246 ZIO_FLAG_OPTIONAL
, NULL
, NULL
));
8249 error
= zio_wait(zio
);
8250 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8253 (void) printf("Read of %s failed, error: %d\n", thing
, error
);
8257 uint64_t orig_lsize
= lsize
;
8259 if (flags
& ZDB_FLAG_DECOMPRESS
) {
8260 boolean_t failed
= zdb_decompress_block(pabd
, buf
, lbuf
,
8261 lsize
, psize
, flags
);
8263 (void) printf("Decompress of %s failed\n", thing
);
8267 buf
= abd_borrow_buf_copy(pabd
, lsize
);
8271 * Try to detect invalid block pointer. If invalid, try
8274 if ((flags
& ZDB_FLAG_PRINT_BLKPTR
|| flags
& ZDB_FLAG_INDIRECT
) &&
8275 !(flags
& ZDB_FLAG_DECOMPRESS
)) {
8276 const blkptr_t
*b
= (const blkptr_t
*)(void *)
8277 ((uintptr_t)buf
+ (uintptr_t)blkptr_offset
);
8278 if (zfs_blkptr_verify(spa
, b
, B_FALSE
, BLK_VERIFY_ONLY
) ==
8280 abd_return_buf_copy(pabd
, buf
, lsize
);
8283 boolean_t failed
= zdb_decompress_block(pabd
, buf
,
8284 lbuf
, lsize
, psize
, flags
);
8285 b
= (const blkptr_t
*)(void *)
8286 ((uintptr_t)buf
+ (uintptr_t)blkptr_offset
);
8287 if (failed
|| zfs_blkptr_verify(spa
, b
, B_FALSE
,
8288 BLK_VERIFY_LOG
) == B_FALSE
) {
8289 printf("invalid block pointer at this DVA\n");
8295 if (flags
& ZDB_FLAG_PRINT_BLKPTR
)
8296 zdb_print_blkptr((blkptr_t
*)(void *)
8297 ((uintptr_t)buf
+ (uintptr_t)blkptr_offset
), flags
);
8298 else if (flags
& ZDB_FLAG_RAW
)
8299 zdb_dump_block_raw(buf
, lsize
, flags
);
8300 else if (flags
& ZDB_FLAG_INDIRECT
)
8301 zdb_dump_indirect((blkptr_t
*)buf
,
8302 orig_lsize
/ sizeof (blkptr_t
), flags
);
8303 else if (flags
& ZDB_FLAG_GBH
)
8304 zdb_dump_gbh(buf
, flags
);
8306 zdb_dump_block(thing
, buf
, lsize
, flags
);
8309 * If :c was specified, iterate through the checksum table to
8310 * calculate and display each checksum for our specified
8313 if ((flags
& ZDB_FLAG_CHECKSUM
) && !(flags
& ZDB_FLAG_RAW
) &&
8314 !(flags
& ZDB_FLAG_GBH
)) {
8316 (void) printf("\n");
8317 for (enum zio_checksum ck
= ZIO_CHECKSUM_LABEL
;
8318 ck
< ZIO_CHECKSUM_FUNCTIONS
; ck
++) {
8320 if ((zio_checksum_table
[ck
].ci_flags
&
8321 ZCHECKSUM_FLAG_EMBEDDED
) ||
8322 ck
== ZIO_CHECKSUM_NOPARITY
) {
8325 BP_SET_CHECKSUM(bp
, ck
);
8326 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8327 czio
= zio_root(spa
, NULL
, NULL
, ZIO_FLAG_CANFAIL
);
8330 if (vd
== vd
->vdev_top
) {
8331 zio_nowait(zio_read(czio
, spa
, bp
, pabd
, psize
,
8333 ZIO_PRIORITY_SYNC_READ
,
8334 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
|
8335 ZIO_FLAG_DONT_RETRY
, NULL
));
8337 zio_nowait(zio_vdev_child_io(czio
, bp
, vd
,
8338 offset
, pabd
, psize
, ZIO_TYPE_READ
,
8339 ZIO_PRIORITY_SYNC_READ
,
8340 ZIO_FLAG_DONT_CACHE
|
8341 ZIO_FLAG_DONT_PROPAGATE
|
8342 ZIO_FLAG_DONT_RETRY
|
8343 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
|
8344 ZIO_FLAG_SPECULATIVE
|
8345 ZIO_FLAG_OPTIONAL
, NULL
, NULL
));
8347 error
= zio_wait(czio
);
8348 if (error
== 0 || error
== ECKSUM
) {
8349 zio_t
*ck_zio
= zio_root(spa
, NULL
, NULL
, 0);
8351 DVA_GET_OFFSET(&bp
->blk_dva
[0]);
8353 zio_checksum_compute(ck_zio
, ck
, pabd
, lsize
);
8354 printf("%12s\tcksum=%llx:%llx:%llx:%llx\n",
8355 zio_checksum_table
[ck
].ci_name
,
8356 (u_longlong_t
)bp
->blk_cksum
.zc_word
[0],
8357 (u_longlong_t
)bp
->blk_cksum
.zc_word
[1],
8358 (u_longlong_t
)bp
->blk_cksum
.zc_word
[2],
8359 (u_longlong_t
)bp
->blk_cksum
.zc_word
[3]);
8362 printf("error %d reading block\n", error
);
8364 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8369 abd_return_buf_copy(pabd
, buf
, lsize
);
8373 umem_free(lbuf
, SPA_MAXBLOCKSIZE
);
8380 zdb_embedded_block(char *thing
)
8382 blkptr_t bp
= {{{{0}}}};
8383 unsigned long long *words
= (void *)&bp
;
8387 err
= sscanf(thing
, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
8388 "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
8389 words
+ 0, words
+ 1, words
+ 2, words
+ 3,
8390 words
+ 4, words
+ 5, words
+ 6, words
+ 7,
8391 words
+ 8, words
+ 9, words
+ 10, words
+ 11,
8392 words
+ 12, words
+ 13, words
+ 14, words
+ 15);
8394 (void) fprintf(stderr
, "invalid input format\n");
8397 ASSERT3U(BPE_GET_LSIZE(&bp
), <=, SPA_MAXBLOCKSIZE
);
8398 buf
= malloc(SPA_MAXBLOCKSIZE
);
8400 (void) fprintf(stderr
, "out of memory\n");
8403 err
= decode_embedded_bp(&bp
, buf
, BPE_GET_LSIZE(&bp
));
8405 (void) fprintf(stderr
, "decode failed: %u\n", err
);
8408 zdb_dump_block_raw(buf
, BPE_GET_LSIZE(&bp
), 0);
8412 /* check for valid hex or decimal numeric string */
8414 zdb_numeric(char *str
)
8418 if (strlen(str
) == 0)
8420 if (strncmp(str
, "0x", 2) == 0 || strncmp(str
, "0X", 2) == 0)
8422 for (; i
< strlen(str
); i
++) {
8423 if (!isxdigit(str
[i
]))
8430 main(int argc
, char **argv
)
8434 objset_t
*os
= NULL
;
8438 char **searchdirs
= NULL
;
8440 char *target
, *target_pool
, dsname
[ZFS_MAX_DATASET_NAME_LEN
];
8441 nvlist_t
*policy
= NULL
;
8442 uint64_t max_txg
= UINT64_MAX
;
8443 int64_t objset_id
= -1;
8445 int flags
= ZFS_IMPORT_MISSING_LOG
;
8446 int rewind
= ZPOOL_NEVER_REWIND
;
8447 char *spa_config_path_env
, *objset_str
;
8448 boolean_t target_is_spa
= B_TRUE
, dataset_lookup
= B_FALSE
;
8449 nvlist_t
*cfg
= NULL
;
8451 dprintf_setup(&argc
, argv
);
8454 * If there is an environment variable SPA_CONFIG_PATH it overrides
8455 * default spa_config_path setting. If -U flag is specified it will
8456 * override this environment variable settings once again.
8458 spa_config_path_env
= getenv("SPA_CONFIG_PATH");
8459 if (spa_config_path_env
!= NULL
)
8460 spa_config_path
= spa_config_path_env
;
8463 * For performance reasons, we set this tunable down. We do so before
8464 * the arg parsing section so that the user can override this value if
8467 zfs_btree_verify_intensity
= 3;
8469 struct option long_options
[] = {
8470 {"ignore-assertions", no_argument
, NULL
, 'A'},
8471 {"block-stats", no_argument
, NULL
, 'b'},
8472 {"checksum", no_argument
, NULL
, 'c'},
8473 {"config", no_argument
, NULL
, 'C'},
8474 {"datasets", no_argument
, NULL
, 'd'},
8475 {"dedup-stats", no_argument
, NULL
, 'D'},
8476 {"exported", no_argument
, NULL
, 'e'},
8477 {"embedded-block-pointer", no_argument
, NULL
, 'E'},
8478 {"automatic-rewind", no_argument
, NULL
, 'F'},
8479 {"dump-debug-msg", no_argument
, NULL
, 'G'},
8480 {"history", no_argument
, NULL
, 'h'},
8481 {"intent-logs", no_argument
, NULL
, 'i'},
8482 {"inflight", required_argument
, NULL
, 'I'},
8483 {"checkpointed-state", no_argument
, NULL
, 'k'},
8484 {"label", no_argument
, NULL
, 'l'},
8485 {"disable-leak-tracking", no_argument
, NULL
, 'L'},
8486 {"metaslabs", no_argument
, NULL
, 'm'},
8487 {"metaslab-groups", no_argument
, NULL
, 'M'},
8488 {"numeric", no_argument
, NULL
, 'N'},
8489 {"option", required_argument
, NULL
, 'o'},
8490 {"object-lookups", no_argument
, NULL
, 'O'},
8491 {"path", required_argument
, NULL
, 'p'},
8492 {"parseable", no_argument
, NULL
, 'P'},
8493 {"skip-label", no_argument
, NULL
, 'q'},
8494 {"copy-object", no_argument
, NULL
, 'r'},
8495 {"read-block", no_argument
, NULL
, 'R'},
8496 {"io-stats", no_argument
, NULL
, 's'},
8497 {"simulate-dedup", no_argument
, NULL
, 'S'},
8498 {"txg", required_argument
, NULL
, 't'},
8499 {"uberblock", no_argument
, NULL
, 'u'},
8500 {"cachefile", required_argument
, NULL
, 'U'},
8501 {"verbose", no_argument
, NULL
, 'v'},
8502 {"verbatim", no_argument
, NULL
, 'V'},
8503 {"dump-blocks", required_argument
, NULL
, 'x'},
8504 {"extreme-rewind", no_argument
, NULL
, 'X'},
8505 {"all-reconstruction", no_argument
, NULL
, 'Y'},
8506 {"livelist", no_argument
, NULL
, 'y'},
8507 {"zstd-headers", no_argument
, NULL
, 'Z'},
8511 while ((c
= getopt_long(argc
, argv
,
8512 "AbcCdDeEFGhiI:klLmMNo:Op:PqrRsSt:uU:vVx:XYyZ",
8513 long_options
, NULL
)) != -1) {
8550 zfs_reconstruct_indirect_combinations_max
= INT_MAX
;
8551 zfs_deadman_enabled
= 0;
8553 /* NB: Sort single match options below. */
8555 max_inflight_bytes
= strtoull(optarg
, NULL
, 0);
8556 if (max_inflight_bytes
== 0) {
8557 (void) fprintf(stderr
, "maximum number "
8558 "of inflight bytes must be greater "
8564 error
= set_global_var(optarg
);
8569 if (searchdirs
== NULL
) {
8570 searchdirs
= umem_alloc(sizeof (char *),
8573 char **tmp
= umem_alloc((nsearch
+ 1) *
8574 sizeof (char *), UMEM_NOFAIL
);
8575 memcpy(tmp
, searchdirs
, nsearch
*
8577 umem_free(searchdirs
,
8578 nsearch
* sizeof (char *));
8581 searchdirs
[nsearch
++] = optarg
;
8584 max_txg
= strtoull(optarg
, NULL
, 0);
8585 if (max_txg
< TXG_INITIAL
) {
8586 (void) fprintf(stderr
, "incorrect txg "
8587 "specified: %s\n", optarg
);
8592 spa_config_path
= optarg
;
8593 if (spa_config_path
[0] != '/') {
8594 (void) fprintf(stderr
,
8595 "cachefile must be an absolute path "
8596 "(i.e. start with a slash)\n");
8604 flags
= ZFS_IMPORT_VERBATIM
;
8607 vn_dumpdir
= optarg
;
8615 if (!dump_opt
['e'] && searchdirs
!= NULL
) {
8616 (void) fprintf(stderr
, "-p option requires use of -e\n");
8621 * ZDB does not typically re-read blocks; therefore limit the ARC
8622 * to 256 MB, which can be used entirely for metadata.
8624 zfs_arc_min
= zfs_arc_meta_min
= 2ULL << SPA_MAXBLOCKSHIFT
;
8625 zfs_arc_max
= zfs_arc_meta_limit
= 256 * 1024 * 1024;
8629 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
8630 * "zdb -b" uses traversal prefetch which uses async reads.
8631 * For good performance, let several of them be active at once.
8633 zfs_vdev_async_read_max_active
= 10;
8636 * Disable reference tracking for better performance.
8638 reference_tracking_enable
= B_FALSE
;
8641 * Do not fail spa_load when spa_load_verify fails. This is needed
8642 * to load non-idle pools.
8644 spa_load_verify_dryrun
= B_TRUE
;
8647 * ZDB should have ability to read spacemaps.
8649 spa_mode_readable_spacemaps
= B_TRUE
;
8651 kernel_init(SPA_MODE_READ
);
8654 verbose
= MAX(verbose
, 1);
8656 for (c
= 0; c
< 256; c
++) {
8657 if (dump_all
&& strchr("AeEFklLNOPrRSXy", c
) == NULL
)
8660 dump_opt
[c
] += verbose
;
8663 libspl_set_assert_ok((dump_opt
['A'] == 1) || (dump_opt
['A'] > 2));
8664 zfs_recover
= (dump_opt
['A'] > 1);
8668 if (argc
< 2 && dump_opt
['R'])
8671 if (dump_opt
['E']) {
8674 zdb_embedded_block(argv
[0]);
8679 if (!dump_opt
['e'] && dump_opt
['C']) {
8680 dump_cachefile(spa_config_path
);
8687 return (dump_label(argv
[0]));
8689 if (dump_opt
['O']) {
8692 dump_opt
['v'] = verbose
+ 3;
8693 return (dump_path(argv
[0], argv
[1], NULL
));
8695 if (dump_opt
['r']) {
8696 target_is_spa
= B_FALSE
;
8699 dump_opt
['v'] = verbose
;
8700 error
= dump_path(argv
[0], argv
[1], &object
);
8703 if (dump_opt
['X'] || dump_opt
['F'])
8704 rewind
= ZPOOL_DO_REWIND
|
8705 (dump_opt
['X'] ? ZPOOL_EXTREME_REWIND
: 0);
8708 if (dump_opt
['N'] && dump_opt
['d'] == 0)
8709 dump_opt
['d'] = dump_opt
['N'];
8711 if (nvlist_alloc(&policy
, NV_UNIQUE_NAME_TYPE
, 0) != 0 ||
8712 nvlist_add_uint64(policy
, ZPOOL_LOAD_REQUEST_TXG
, max_txg
) != 0 ||
8713 nvlist_add_uint32(policy
, ZPOOL_LOAD_REWIND_POLICY
, rewind
) != 0)
8714 fatal("internal error: %s", strerror(ENOMEM
));
8719 if (strpbrk(target
, "/@") != NULL
) {
8722 target_pool
= strdup(target
);
8723 *strpbrk(target_pool
, "/@") = '\0';
8725 target_is_spa
= B_FALSE
;
8726 targetlen
= strlen(target
);
8727 if (targetlen
&& target
[targetlen
- 1] == '/')
8728 target
[targetlen
- 1] = '\0';
8731 * See if an objset ID was supplied (-d <pool>/<objset ID>).
8732 * To disambiguate tank/100, consider the 100 as objsetID
8733 * if -N was given, otherwise 100 is an objsetID iff
8734 * tank/100 as a named dataset fails on lookup.
8736 objset_str
= strchr(target
, '/');
8737 if (objset_str
&& strlen(objset_str
) > 1 &&
8738 zdb_numeric(objset_str
+ 1)) {
8742 objset_id
= strtoull(objset_str
, &endptr
, 0);
8743 /* dataset 0 is the same as opening the pool */
8744 if (errno
== 0 && endptr
!= objset_str
&&
8747 dataset_lookup
= B_TRUE
;
8749 /* normal dataset name not an objset ID */
8750 if (endptr
== objset_str
) {
8753 } else if (objset_str
&& !zdb_numeric(objset_str
+ 1) &&
8755 printf("Supply a numeric objset ID with -N\n");
8759 target_pool
= target
;
8762 if (dump_opt
['e']) {
8763 importargs_t args
= { 0 };
8765 args
.paths
= nsearch
;
8766 args
.path
= searchdirs
;
8767 args
.can_be_active
= B_TRUE
;
8769 error
= zpool_find_config(NULL
, target_pool
, &cfg
, &args
,
8770 &libzpool_config_ops
);
8774 if (nvlist_add_nvlist(cfg
,
8775 ZPOOL_LOAD_POLICY
, policy
) != 0) {
8776 fatal("can't open '%s': %s",
8777 target
, strerror(ENOMEM
));
8780 if (dump_opt
['C'] > 1) {
8781 (void) printf("\nConfiguration for import:\n");
8782 dump_nvlist(cfg
, 8);
8786 * Disable the activity check to allow examination of
8789 error
= spa_import(target_pool
, cfg
, NULL
,
8790 flags
| ZFS_IMPORT_SKIP_MMP
);
8794 if (searchdirs
!= NULL
) {
8795 umem_free(searchdirs
, nsearch
* sizeof (char *));
8800 * import_checkpointed_state makes the assumption that the
8801 * target pool that we pass it is already part of the spa
8802 * namespace. Because of that we need to make sure to call
8803 * it always after the -e option has been processed, which
8804 * imports the pool to the namespace if it's not in the
8807 char *checkpoint_pool
= NULL
;
8808 char *checkpoint_target
= NULL
;
8809 if (dump_opt
['k']) {
8810 checkpoint_pool
= import_checkpointed_state(target
, cfg
,
8811 &checkpoint_target
);
8813 if (checkpoint_target
!= NULL
)
8814 target
= checkpoint_target
;
8822 if (target_pool
!= target
)
8826 if (dump_opt
['k'] && (target_is_spa
|| dump_opt
['R'])) {
8827 ASSERT(checkpoint_pool
!= NULL
);
8828 ASSERT(checkpoint_target
== NULL
);
8830 error
= spa_open(checkpoint_pool
, &spa
, FTAG
);
8832 fatal("Tried to open pool \"%s\" but "
8833 "spa_open() failed with error %d\n",
8834 checkpoint_pool
, error
);
8837 } else if (target_is_spa
|| dump_opt
['R'] || objset_id
== 0) {
8838 zdb_set_skip_mmp(target
);
8839 error
= spa_open_rewind(target
, &spa
, FTAG
, policy
,
8843 * If we're missing the log device then
8844 * try opening the pool after clearing the
8847 mutex_enter(&spa_namespace_lock
);
8848 if ((spa
= spa_lookup(target
)) != NULL
&&
8849 spa
->spa_log_state
== SPA_LOG_MISSING
) {
8850 spa
->spa_log_state
= SPA_LOG_CLEAR
;
8853 mutex_exit(&spa_namespace_lock
);
8856 error
= spa_open_rewind(target
, &spa
,
8857 FTAG
, policy
, NULL
);
8860 } else if (strpbrk(target
, "#") != NULL
) {
8862 error
= dsl_pool_hold(target
, FTAG
, &dp
);
8864 fatal("can't dump '%s': %s", target
,
8867 error
= dump_bookmark(dp
, target
, B_TRUE
, verbose
> 1);
8868 dsl_pool_rele(dp
, FTAG
);
8870 fatal("can't dump '%s': %s", target
,
8875 target_pool
= strdup(target
);
8876 if (strpbrk(target
, "/@") != NULL
)
8877 *strpbrk(target_pool
, "/@") = '\0';
8879 zdb_set_skip_mmp(target
);
8881 * If -N was supplied, the user has indicated that
8882 * zdb -d <pool>/<objsetID> is in effect. Otherwise
8883 * we first assume that the dataset string is the
8884 * dataset name. If dmu_objset_hold fails with the
8885 * dataset string, and we have an objset_id, retry the
8886 * lookup with the objsetID.
8888 boolean_t retry
= B_TRUE
;
8890 if (dataset_lookup
== B_TRUE
) {
8892 * Use the supplied id to get the name
8895 error
= spa_open(target_pool
, &spa
, FTAG
);
8897 error
= name_from_objset_id(spa
,
8899 spa_close(spa
, FTAG
);
8905 if (objset_id
> 0 && retry
) {
8906 int err
= dmu_objset_hold(target
, FTAG
,
8909 dataset_lookup
= B_TRUE
;
8913 dmu_objset_rele(os
, FTAG
);
8916 error
= open_objset(target
, FTAG
, &os
);
8919 spa
= dmu_objset_spa(os
);
8923 nvlist_free(policy
);
8926 fatal("can't open '%s': %s", target
, strerror(error
));
8929 * Set the pool failure mode to panic in order to prevent the pool
8930 * from suspending. A suspended I/O will have no way to resume and
8931 * can prevent the zdb(8) command from terminating as expected.
8934 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
8938 if (dump_opt
['r']) {
8939 error
= zdb_copy_object(os
, object
, argv
[1]);
8940 } else if (!dump_opt
['R']) {
8941 flagbits
['d'] = ZOR_FLAG_DIRECTORY
;
8942 flagbits
['f'] = ZOR_FLAG_PLAIN_FILE
;
8943 flagbits
['m'] = ZOR_FLAG_SPACE_MAP
;
8944 flagbits
['z'] = ZOR_FLAG_ZAP
;
8945 flagbits
['A'] = ZOR_FLAG_ALL_TYPES
;
8947 if (argc
> 0 && dump_opt
['d']) {
8948 zopt_object_args
= argc
;
8949 zopt_object_ranges
= calloc(zopt_object_args
,
8950 sizeof (zopt_object_range_t
));
8951 for (unsigned i
= 0; i
< zopt_object_args
; i
++) {
8953 const char *msg
= NULL
;
8955 err
= parse_object_range(argv
[i
],
8956 &zopt_object_ranges
[i
], &msg
);
8958 fatal("Bad object or range: '%s': %s\n",
8959 argv
[i
], msg
?: "");
8961 } else if (argc
> 0 && dump_opt
['m']) {
8962 zopt_metaslab_args
= argc
;
8963 zopt_metaslab
= calloc(zopt_metaslab_args
,
8965 for (unsigned i
= 0; i
< zopt_metaslab_args
; i
++) {
8967 zopt_metaslab
[i
] = strtoull(argv
[i
], NULL
, 0);
8968 if (zopt_metaslab
[i
] == 0 && errno
!= 0)
8969 fatal("bad number %s: %s", argv
[i
],
8975 } else if (zopt_object_args
> 0 && !dump_opt
['m']) {
8976 dump_objset(spa
->spa_meta_objset
);
8981 flagbits
['b'] = ZDB_FLAG_PRINT_BLKPTR
;
8982 flagbits
['c'] = ZDB_FLAG_CHECKSUM
;
8983 flagbits
['d'] = ZDB_FLAG_DECOMPRESS
;
8984 flagbits
['e'] = ZDB_FLAG_BSWAP
;
8985 flagbits
['g'] = ZDB_FLAG_GBH
;
8986 flagbits
['i'] = ZDB_FLAG_INDIRECT
;
8987 flagbits
['r'] = ZDB_FLAG_RAW
;
8988 flagbits
['v'] = ZDB_FLAG_VERBOSE
;
8990 for (int i
= 0; i
< argc
; i
++)
8991 zdb_read_block(argv
[i
], spa
);
8994 if (dump_opt
['k']) {
8995 free(checkpoint_pool
);
8997 free(checkpoint_target
);
9001 close_objset(os
, FTAG
);
9003 spa_close(spa
, FTAG
);
9006 fuid_table_destroy();
9008 dump_debug_buffer();