4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2016 Nexenta Systems, Inc.
27 * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC.
28 * Copyright (c) 2015, 2017, Intel Corporation.
29 * Copyright (c) 2020 Datto Inc.
30 * Copyright (c) 2020, The FreeBSD Foundation [1]
32 * [1] Portions of this software were developed by Allan Jude
33 * under sponsorship from the FreeBSD Foundation.
34 * Copyright (c) 2021 Allan Jude
35 * Copyright (c) 2021 Toomas Soome <tsoome@me.com>
36 * Copyright (c) 2023, Klara Inc.
37 * Copyright (c) 2023, Rob Norris <robn@despairlabs.com>
45 #include <openssl/evp.h>
46 #include <sys/zfs_context.h>
48 #include <sys/spa_impl.h>
51 #include <sys/fs/zfs.h>
52 #include <sys/zfs_znode.h>
53 #include <sys/zfs_sa.h>
55 #include <sys/sa_impl.h>
57 #include <sys/vdev_impl.h>
58 #include <sys/metaslab_impl.h>
59 #include <sys/dmu_objset.h>
60 #include <sys/dsl_dir.h>
61 #include <sys/dsl_dataset.h>
62 #include <sys/dsl_pool.h>
63 #include <sys/dsl_bookmark.h>
66 #include <sys/zil_impl.h>
68 #include <sys/resource.h>
69 #include <sys/dmu_send.h>
70 #include <sys/dmu_traverse.h>
71 #include <sys/zio_checksum.h>
72 #include <sys/zio_compress.h>
73 #include <sys/zfs_fuid.h>
75 #include <sys/arc_impl.h>
77 #include <sys/zfeature.h>
79 #include <sys/blkptr.h>
80 #include <sys/dsl_crypt.h>
81 #include <sys/dsl_scan.h>
82 #include <sys/btree.h>
84 #include <sys/brt_impl.h>
85 #include <zfs_comutil.h>
86 #include <sys/zstd/zstd.h>
88 #include <libnvpair.h>
93 #define ZDB_COMPRESS_NAME(idx) ((idx) < ZIO_COMPRESS_FUNCTIONS ? \
94 zio_compress_table[(idx)].ci_name : "UNKNOWN")
95 #define ZDB_CHECKSUM_NAME(idx) ((idx) < ZIO_CHECKSUM_FUNCTIONS ? \
96 zio_checksum_table[(idx)].ci_name : "UNKNOWN")
97 #define ZDB_OT_TYPE(idx) ((idx) < DMU_OT_NUMTYPES ? (idx) : \
98 (idx) == DMU_OTN_ZAP_DATA || (idx) == DMU_OTN_ZAP_METADATA ? \
100 (idx) == DMU_OTN_UINT64_DATA || (idx) == DMU_OTN_UINT64_METADATA ? \
101 DMU_OT_UINT64_OTHER : DMU_OT_NUMTYPES)
103 /* Some platforms require part of inode IDs to be remapped */
105 #define ZDB_MAP_OBJECT_ID(obj) INO_XNUTOZFS(obj, 2)
107 #define ZDB_MAP_OBJECT_ID(obj) (obj)
111 zdb_ot_name(dmu_object_type_t type
)
113 if (type
< DMU_OT_NUMTYPES
)
114 return (dmu_ot
[type
].ot_name
);
115 else if ((type
& DMU_OT_NEWTYPE
) &&
116 ((type
& DMU_OT_BYTESWAP_MASK
) < DMU_BSWAP_NUMFUNCS
))
117 return (dmu_ot_byteswap
[type
& DMU_OT_BYTESWAP_MASK
].ob_name
);
122 extern int reference_tracking_enable
;
123 extern int zfs_recover
;
124 extern uint_t zfs_vdev_async_read_max_active
;
125 extern boolean_t spa_load_verify_dryrun
;
126 extern boolean_t spa_mode_readable_spacemaps
;
127 extern uint_t zfs_reconstruct_indirect_combinations_max
;
128 extern uint_t zfs_btree_verify_intensity
;
130 static const char cmdname
[] = "zdb";
131 uint8_t dump_opt
[256];
133 typedef void object_viewer_t(objset_t
*, uint64_t, void *data
, size_t size
);
135 static uint64_t *zopt_metaslab
= NULL
;
136 static unsigned zopt_metaslab_args
= 0;
138 typedef struct zopt_object_range
{
139 uint64_t zor_obj_start
;
140 uint64_t zor_obj_end
;
142 } zopt_object_range_t
;
144 static zopt_object_range_t
*zopt_object_ranges
= NULL
;
145 static unsigned zopt_object_args
= 0;
147 static int flagbits
[256];
149 #define ZOR_FLAG_PLAIN_FILE 0x0001
150 #define ZOR_FLAG_DIRECTORY 0x0002
151 #define ZOR_FLAG_SPACE_MAP 0x0004
152 #define ZOR_FLAG_ZAP 0x0008
153 #define ZOR_FLAG_ALL_TYPES -1
154 #define ZOR_SUPPORTED_FLAGS (ZOR_FLAG_PLAIN_FILE | \
155 ZOR_FLAG_DIRECTORY | \
156 ZOR_FLAG_SPACE_MAP | \
159 #define ZDB_FLAG_CHECKSUM 0x0001
160 #define ZDB_FLAG_DECOMPRESS 0x0002
161 #define ZDB_FLAG_BSWAP 0x0004
162 #define ZDB_FLAG_GBH 0x0008
163 #define ZDB_FLAG_INDIRECT 0x0010
164 #define ZDB_FLAG_RAW 0x0020
165 #define ZDB_FLAG_PRINT_BLKPTR 0x0040
166 #define ZDB_FLAG_VERBOSE 0x0080
168 static uint64_t max_inflight_bytes
= 256 * 1024 * 1024; /* 256MB */
169 static int leaked_objects
= 0;
170 static range_tree_t
*mos_refd_objs
;
172 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t
*,
174 static void mos_obj_refd(uint64_t);
175 static void mos_obj_refd_multiple(uint64_t);
176 static int dump_bpobj_cb(void *arg
, const blkptr_t
*bp
, boolean_t free
,
179 typedef struct sublivelist_verify
{
180 /* FREE's that haven't yet matched to an ALLOC, in one sub-livelist */
183 /* ALLOC's without a matching FREE, accumulates across sub-livelists */
184 zfs_btree_t sv_leftover
;
185 } sublivelist_verify_t
;
188 livelist_compare(const void *larg
, const void *rarg
)
190 const blkptr_t
*l
= larg
;
191 const blkptr_t
*r
= rarg
;
193 /* Sort them according to dva[0] */
194 uint64_t l_dva0_vdev
, r_dva0_vdev
;
195 l_dva0_vdev
= DVA_GET_VDEV(&l
->blk_dva
[0]);
196 r_dva0_vdev
= DVA_GET_VDEV(&r
->blk_dva
[0]);
197 if (l_dva0_vdev
< r_dva0_vdev
)
199 else if (l_dva0_vdev
> r_dva0_vdev
)
202 /* if vdevs are equal, sort by offsets. */
203 uint64_t l_dva0_offset
;
204 uint64_t r_dva0_offset
;
205 l_dva0_offset
= DVA_GET_OFFSET(&l
->blk_dva
[0]);
206 r_dva0_offset
= DVA_GET_OFFSET(&r
->blk_dva
[0]);
207 if (l_dva0_offset
< r_dva0_offset
) {
209 } else if (l_dva0_offset
> r_dva0_offset
) {
214 * Since we're storing blkptrs without cancelling FREE/ALLOC pairs,
215 * it's possible the offsets are equal. In that case, sort by txg
217 if (l
->blk_birth
< r
->blk_birth
) {
219 } else if (l
->blk_birth
> r
->blk_birth
) {
225 typedef struct sublivelist_verify_block
{
229 * We need this to check if the block marked as allocated
230 * in the livelist was freed (and potentially reallocated)
231 * in the metaslab spacemaps at a later TXG.
233 uint64_t svb_allocated_txg
;
234 } sublivelist_verify_block_t
;
236 static void zdb_print_blkptr(const blkptr_t
*bp
, int flags
);
238 typedef struct sublivelist_verify_block_refcnt
{
239 /* block pointer entry in livelist being verified */
243 * Refcount gets incremented to 1 when we encounter the first
244 * FREE entry for the svfbr block pointer and a node for it
245 * is created in our ZDB verification/tracking metadata.
247 * As we encounter more FREE entries we increment this counter
248 * and similarly decrement it whenever we find the respective
249 * ALLOC entries for this block.
251 * When the refcount gets to 0 it means that all the FREE and
252 * ALLOC entries of this block have paired up and we no longer
253 * need to track it in our verification logic (e.g. the node
254 * containing this struct in our verification data structure
257 * [refer to sublivelist_verify_blkptr() for the actual code]
259 uint32_t svbr_refcnt
;
260 } sublivelist_verify_block_refcnt_t
;
263 sublivelist_block_refcnt_compare(const void *larg
, const void *rarg
)
265 const sublivelist_verify_block_refcnt_t
*l
= larg
;
266 const sublivelist_verify_block_refcnt_t
*r
= rarg
;
267 return (livelist_compare(&l
->svbr_blk
, &r
->svbr_blk
));
271 sublivelist_verify_blkptr(void *arg
, const blkptr_t
*bp
, boolean_t free
,
274 ASSERT3P(tx
, ==, NULL
);
275 struct sublivelist_verify
*sv
= arg
;
276 sublivelist_verify_block_refcnt_t current
= {
280 * Start with 1 in case this is the first free entry.
281 * This field is not used for our B-Tree comparisons
287 zfs_btree_index_t where
;
288 sublivelist_verify_block_refcnt_t
*pair
=
289 zfs_btree_find(&sv
->sv_pair
, ¤t
, &where
);
292 /* first free entry for this block pointer */
293 zfs_btree_add(&sv
->sv_pair
, ¤t
);
299 /* block that is currently marked as allocated */
300 for (int i
= 0; i
< SPA_DVAS_PER_BP
; i
++) {
301 if (DVA_IS_EMPTY(&bp
->blk_dva
[i
]))
303 sublivelist_verify_block_t svb
= {
304 .svb_dva
= bp
->blk_dva
[i
],
305 .svb_allocated_txg
= bp
->blk_birth
308 if (zfs_btree_find(&sv
->sv_leftover
, &svb
,
310 zfs_btree_add_idx(&sv
->sv_leftover
,
315 /* alloc matches a free entry */
317 if (pair
->svbr_refcnt
== 0) {
318 /* all allocs and frees have been matched */
319 zfs_btree_remove_idx(&sv
->sv_pair
, &where
);
328 sublivelist_verify_func(void *args
, dsl_deadlist_entry_t
*dle
)
331 struct sublivelist_verify
*sv
= args
;
333 zfs_btree_create(&sv
->sv_pair
, sublivelist_block_refcnt_compare
, NULL
,
334 sizeof (sublivelist_verify_block_refcnt_t
));
336 err
= bpobj_iterate_nofree(&dle
->dle_bpobj
, sublivelist_verify_blkptr
,
339 sublivelist_verify_block_refcnt_t
*e
;
340 zfs_btree_index_t
*cookie
= NULL
;
341 while ((e
= zfs_btree_destroy_nodes(&sv
->sv_pair
, &cookie
)) != NULL
) {
342 char blkbuf
[BP_SPRINTF_LEN
];
343 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
),
344 &e
->svbr_blk
, B_TRUE
);
345 (void) printf("\tERROR: %d unmatched FREE(s): %s\n",
346 e
->svbr_refcnt
, blkbuf
);
348 zfs_btree_destroy(&sv
->sv_pair
);
354 livelist_block_compare(const void *larg
, const void *rarg
)
356 const sublivelist_verify_block_t
*l
= larg
;
357 const sublivelist_verify_block_t
*r
= rarg
;
359 if (DVA_GET_VDEV(&l
->svb_dva
) < DVA_GET_VDEV(&r
->svb_dva
))
361 else if (DVA_GET_VDEV(&l
->svb_dva
) > DVA_GET_VDEV(&r
->svb_dva
))
364 if (DVA_GET_OFFSET(&l
->svb_dva
) < DVA_GET_OFFSET(&r
->svb_dva
))
366 else if (DVA_GET_OFFSET(&l
->svb_dva
) > DVA_GET_OFFSET(&r
->svb_dva
))
369 if (DVA_GET_ASIZE(&l
->svb_dva
) < DVA_GET_ASIZE(&r
->svb_dva
))
371 else if (DVA_GET_ASIZE(&l
->svb_dva
) > DVA_GET_ASIZE(&r
->svb_dva
))
378 * Check for errors in a livelist while tracking all unfreed ALLOCs in the
379 * sublivelist_verify_t: sv->sv_leftover
382 livelist_verify(dsl_deadlist_t
*dl
, void *arg
)
384 sublivelist_verify_t
*sv
= arg
;
385 dsl_deadlist_iterate(dl
, sublivelist_verify_func
, sv
);
389 * Check for errors in the livelist entry and discard the intermediary
393 sublivelist_verify_lightweight(void *args
, dsl_deadlist_entry_t
*dle
)
396 sublivelist_verify_t sv
;
397 zfs_btree_create(&sv
.sv_leftover
, livelist_block_compare
, NULL
,
398 sizeof (sublivelist_verify_block_t
));
399 int err
= sublivelist_verify_func(&sv
, dle
);
400 zfs_btree_clear(&sv
.sv_leftover
);
401 zfs_btree_destroy(&sv
.sv_leftover
);
405 typedef struct metaslab_verify
{
407 * Tree containing all the leftover ALLOCs from the livelists
408 * that are part of this metaslab.
410 zfs_btree_t mv_livelist_allocs
;
413 * Metaslab information.
421 * What's currently allocated for this metaslab.
423 range_tree_t
*mv_allocated
;
426 typedef void ll_iter_t(dsl_deadlist_t
*ll
, void *arg
);
428 typedef int (*zdb_log_sm_cb_t
)(spa_t
*spa
, space_map_entry_t
*sme
, uint64_t txg
,
431 typedef struct unflushed_iter_cb_arg
{
435 zdb_log_sm_cb_t uic_cb
;
436 } unflushed_iter_cb_arg_t
;
439 iterate_through_spacemap_logs_cb(space_map_entry_t
*sme
, void *arg
)
441 unflushed_iter_cb_arg_t
*uic
= arg
;
442 return (uic
->uic_cb(uic
->uic_spa
, sme
, uic
->uic_txg
, uic
->uic_arg
));
446 iterate_through_spacemap_logs(spa_t
*spa
, zdb_log_sm_cb_t cb
, void *arg
)
448 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
451 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
452 for (spa_log_sm_t
*sls
= avl_first(&spa
->spa_sm_logs_by_txg
);
453 sls
; sls
= AVL_NEXT(&spa
->spa_sm_logs_by_txg
, sls
)) {
454 space_map_t
*sm
= NULL
;
455 VERIFY0(space_map_open(&sm
, spa_meta_objset(spa
),
456 sls
->sls_sm_obj
, 0, UINT64_MAX
, SPA_MINBLOCKSHIFT
));
458 unflushed_iter_cb_arg_t uic
= {
460 .uic_txg
= sls
->sls_txg
,
464 VERIFY0(space_map_iterate(sm
, space_map_length(sm
),
465 iterate_through_spacemap_logs_cb
, &uic
));
468 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
472 verify_livelist_allocs(metaslab_verify_t
*mv
, uint64_t txg
,
473 uint64_t offset
, uint64_t size
)
475 sublivelist_verify_block_t svb
= {{{0}}};
476 DVA_SET_VDEV(&svb
.svb_dva
, mv
->mv_vdid
);
477 DVA_SET_OFFSET(&svb
.svb_dva
, offset
);
478 DVA_SET_ASIZE(&svb
.svb_dva
, size
);
479 zfs_btree_index_t where
;
480 uint64_t end_offset
= offset
+ size
;
483 * Look for an exact match for spacemap entry in the livelist entries.
484 * Then, look for other livelist entries that fall within the range
485 * of the spacemap entry as it may have been condensed
487 sublivelist_verify_block_t
*found
=
488 zfs_btree_find(&mv
->mv_livelist_allocs
, &svb
, &where
);
490 found
= zfs_btree_next(&mv
->mv_livelist_allocs
, &where
, &where
);
492 for (; found
!= NULL
&& DVA_GET_VDEV(&found
->svb_dva
) == mv
->mv_vdid
&&
493 DVA_GET_OFFSET(&found
->svb_dva
) < end_offset
;
494 found
= zfs_btree_next(&mv
->mv_livelist_allocs
, &where
, &where
)) {
495 if (found
->svb_allocated_txg
<= txg
) {
496 (void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
497 "from TXG %llx FREED at TXG %llx\n",
498 (u_longlong_t
)DVA_GET_OFFSET(&found
->svb_dva
),
499 (u_longlong_t
)DVA_GET_ASIZE(&found
->svb_dva
),
500 (u_longlong_t
)found
->svb_allocated_txg
,
507 metaslab_spacemap_validation_cb(space_map_entry_t
*sme
, void *arg
)
509 metaslab_verify_t
*mv
= arg
;
510 uint64_t offset
= sme
->sme_offset
;
511 uint64_t size
= sme
->sme_run
;
512 uint64_t txg
= sme
->sme_txg
;
514 if (sme
->sme_type
== SM_ALLOC
) {
515 if (range_tree_contains(mv
->mv_allocated
,
517 (void) printf("ERROR: DOUBLE ALLOC: "
519 "%llu:%llu LOG_SM\n",
520 (u_longlong_t
)txg
, (u_longlong_t
)offset
,
521 (u_longlong_t
)size
, (u_longlong_t
)mv
->mv_vdid
,
522 (u_longlong_t
)mv
->mv_msid
);
524 range_tree_add(mv
->mv_allocated
,
528 if (!range_tree_contains(mv
->mv_allocated
,
530 (void) printf("ERROR: DOUBLE FREE: "
532 "%llu:%llu LOG_SM\n",
533 (u_longlong_t
)txg
, (u_longlong_t
)offset
,
534 (u_longlong_t
)size
, (u_longlong_t
)mv
->mv_vdid
,
535 (u_longlong_t
)mv
->mv_msid
);
537 range_tree_remove(mv
->mv_allocated
,
542 if (sme
->sme_type
!= SM_ALLOC
) {
544 * If something is freed in the spacemap, verify that
545 * it is not listed as allocated in the livelist.
547 verify_livelist_allocs(mv
, txg
, offset
, size
);
553 spacemap_check_sm_log_cb(spa_t
*spa
, space_map_entry_t
*sme
,
554 uint64_t txg
, void *arg
)
556 metaslab_verify_t
*mv
= arg
;
557 uint64_t offset
= sme
->sme_offset
;
558 uint64_t vdev_id
= sme
->sme_vdev
;
560 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
562 /* skip indirect vdevs */
563 if (!vdev_is_concrete(vd
))
566 if (vdev_id
!= mv
->mv_vdid
)
569 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
570 if (ms
->ms_id
!= mv
->mv_msid
)
573 if (txg
< metaslab_unflushed_txg(ms
))
577 ASSERT3U(txg
, ==, sme
->sme_txg
);
578 return (metaslab_spacemap_validation_cb(sme
, mv
));
582 spacemap_check_sm_log(spa_t
*spa
, metaslab_verify_t
*mv
)
584 iterate_through_spacemap_logs(spa
, spacemap_check_sm_log_cb
, mv
);
588 spacemap_check_ms_sm(space_map_t
*sm
, metaslab_verify_t
*mv
)
593 VERIFY0(space_map_iterate(sm
, space_map_length(sm
),
594 metaslab_spacemap_validation_cb
, mv
));
597 static void iterate_deleted_livelists(spa_t
*spa
, ll_iter_t func
, void *arg
);
600 * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
601 * they are part of that metaslab (mv_msid).
604 mv_populate_livelist_allocs(metaslab_verify_t
*mv
, sublivelist_verify_t
*sv
)
606 zfs_btree_index_t where
;
607 sublivelist_verify_block_t
*svb
;
608 ASSERT3U(zfs_btree_numnodes(&mv
->mv_livelist_allocs
), ==, 0);
609 for (svb
= zfs_btree_first(&sv
->sv_leftover
, &where
);
611 svb
= zfs_btree_next(&sv
->sv_leftover
, &where
, &where
)) {
612 if (DVA_GET_VDEV(&svb
->svb_dva
) != mv
->mv_vdid
)
615 if (DVA_GET_OFFSET(&svb
->svb_dva
) < mv
->mv_start
&&
616 (DVA_GET_OFFSET(&svb
->svb_dva
) +
617 DVA_GET_ASIZE(&svb
->svb_dva
)) > mv
->mv_start
) {
618 (void) printf("ERROR: Found block that crosses "
619 "metaslab boundary: <%llu:%llx:%llx>\n",
620 (u_longlong_t
)DVA_GET_VDEV(&svb
->svb_dva
),
621 (u_longlong_t
)DVA_GET_OFFSET(&svb
->svb_dva
),
622 (u_longlong_t
)DVA_GET_ASIZE(&svb
->svb_dva
));
626 if (DVA_GET_OFFSET(&svb
->svb_dva
) < mv
->mv_start
)
629 if (DVA_GET_OFFSET(&svb
->svb_dva
) >= mv
->mv_end
)
632 if ((DVA_GET_OFFSET(&svb
->svb_dva
) +
633 DVA_GET_ASIZE(&svb
->svb_dva
)) > mv
->mv_end
) {
634 (void) printf("ERROR: Found block that crosses "
635 "metaslab boundary: <%llu:%llx:%llx>\n",
636 (u_longlong_t
)DVA_GET_VDEV(&svb
->svb_dva
),
637 (u_longlong_t
)DVA_GET_OFFSET(&svb
->svb_dva
),
638 (u_longlong_t
)DVA_GET_ASIZE(&svb
->svb_dva
));
642 zfs_btree_add(&mv
->mv_livelist_allocs
, svb
);
645 for (svb
= zfs_btree_first(&mv
->mv_livelist_allocs
, &where
);
647 svb
= zfs_btree_next(&mv
->mv_livelist_allocs
, &where
, &where
)) {
648 zfs_btree_remove(&sv
->sv_leftover
, svb
);
654 * Iterate through all the sublivelists and:
655 * - report leftover frees (**)
656 * - record leftover ALLOCs together with their TXG [see Cross Check]
658 * (**) Note: Double ALLOCs are valid in datasets that have dedup
659 * enabled. Similarly double FREEs are allowed as well but
660 * only if they pair up with a corresponding ALLOC entry once
661 * we our done with our sublivelist iteration.
665 * - iterate over spacemap and then the metaslab's entries in the
666 * spacemap log, then report any double FREEs and ALLOCs (do not
670 * After finishing the Livelist Check phase and while being in the
671 * Spacemap Check phase, we find all the recorded leftover ALLOCs
672 * of the livelist check that are part of the metaslab that we are
673 * currently looking at in the Spacemap Check. We report any entries
674 * that are marked as ALLOCs in the livelists but have been actually
675 * freed (and potentially allocated again) after their TXG stamp in
676 * the spacemaps. Also report any ALLOCs from the livelists that
677 * belong to indirect vdevs (e.g. their vdev completed removal).
679 * Note that this will miss Log Spacemap entries that cancelled each other
680 * out before being flushed to the metaslab, so we are not guaranteed
681 * to match all erroneous ALLOCs.
684 livelist_metaslab_validate(spa_t
*spa
)
686 (void) printf("Verifying deleted livelist entries\n");
688 sublivelist_verify_t sv
;
689 zfs_btree_create(&sv
.sv_leftover
, livelist_block_compare
, NULL
,
690 sizeof (sublivelist_verify_block_t
));
691 iterate_deleted_livelists(spa
, livelist_verify
, &sv
);
693 (void) printf("Verifying metaslab entries\n");
694 vdev_t
*rvd
= spa
->spa_root_vdev
;
695 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
696 vdev_t
*vd
= rvd
->vdev_child
[c
];
698 if (!vdev_is_concrete(vd
))
701 for (uint64_t mid
= 0; mid
< vd
->vdev_ms_count
; mid
++) {
702 metaslab_t
*m
= vd
->vdev_ms
[mid
];
704 (void) fprintf(stderr
,
705 "\rverifying concrete vdev %llu, "
706 "metaslab %llu of %llu ...",
707 (longlong_t
)vd
->vdev_id
,
709 (longlong_t
)vd
->vdev_ms_count
);
711 uint64_t shift
, start
;
712 range_seg_type_t type
=
713 metaslab_calculate_range_tree_type(vd
, m
,
715 metaslab_verify_t mv
;
716 mv
.mv_allocated
= range_tree_create(NULL
,
717 type
, NULL
, start
, shift
);
718 mv
.mv_vdid
= vd
->vdev_id
;
719 mv
.mv_msid
= m
->ms_id
;
720 mv
.mv_start
= m
->ms_start
;
721 mv
.mv_end
= m
->ms_start
+ m
->ms_size
;
722 zfs_btree_create(&mv
.mv_livelist_allocs
,
723 livelist_block_compare
, NULL
,
724 sizeof (sublivelist_verify_block_t
));
726 mv_populate_livelist_allocs(&mv
, &sv
);
728 spacemap_check_ms_sm(m
->ms_sm
, &mv
);
729 spacemap_check_sm_log(spa
, &mv
);
731 range_tree_vacate(mv
.mv_allocated
, NULL
, NULL
);
732 range_tree_destroy(mv
.mv_allocated
);
733 zfs_btree_clear(&mv
.mv_livelist_allocs
);
734 zfs_btree_destroy(&mv
.mv_livelist_allocs
);
737 (void) fprintf(stderr
, "\n");
740 * If there are any segments in the leftover tree after we walked
741 * through all the metaslabs in the concrete vdevs then this means
742 * that we have segments in the livelists that belong to indirect
743 * vdevs and are marked as allocated.
745 if (zfs_btree_numnodes(&sv
.sv_leftover
) == 0) {
746 zfs_btree_destroy(&sv
.sv_leftover
);
749 (void) printf("ERROR: Found livelist blocks marked as allocated "
750 "for indirect vdevs:\n");
752 zfs_btree_index_t
*where
= NULL
;
753 sublivelist_verify_block_t
*svb
;
754 while ((svb
= zfs_btree_destroy_nodes(&sv
.sv_leftover
, &where
)) !=
756 int vdev_id
= DVA_GET_VDEV(&svb
->svb_dva
);
757 ASSERT3U(vdev_id
, <, rvd
->vdev_children
);
758 vdev_t
*vd
= rvd
->vdev_child
[vdev_id
];
759 ASSERT(!vdev_is_concrete(vd
));
760 (void) printf("<%d:%llx:%llx> TXG %llx\n",
761 vdev_id
, (u_longlong_t
)DVA_GET_OFFSET(&svb
->svb_dva
),
762 (u_longlong_t
)DVA_GET_ASIZE(&svb
->svb_dva
),
763 (u_longlong_t
)svb
->svb_allocated_txg
);
766 zfs_btree_destroy(&sv
.sv_leftover
);
770 * These libumem hooks provide a reasonable set of defaults for the allocator's
771 * debugging facilities.
774 _umem_debug_init(void)
776 return ("default,verbose"); /* $UMEM_DEBUG setting */
780 _umem_logging_init(void)
782 return ("fail,contents"); /* $UMEM_LOGGING setting */
788 (void) fprintf(stderr
,
789 "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
790 "[-I <inflight I/Os>]\n"
791 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
793 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
794 "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
795 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
796 "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
797 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
798 "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
799 "\t%s [-v] <bookmark>\n"
800 "\t%s -C [-A] [-U <cache>] [<poolname>]\n"
801 "\t%s -l [-Aqu] <device>\n"
802 "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
803 "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
804 "\t%s -O [-K <key>] <dataset> <path>\n"
805 "\t%s -r [-K <key>] <dataset> <path> <destination>\n"
806 "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
807 "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
808 "\t%s -E [-A] word0:word1:...:word15\n"
809 "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
811 cmdname
, cmdname
, cmdname
, cmdname
, cmdname
, cmdname
, cmdname
,
812 cmdname
, cmdname
, cmdname
, cmdname
, cmdname
);
814 (void) fprintf(stderr
, " Dataset name must include at least one "
815 "separator character '/' or '@'\n");
816 (void) fprintf(stderr
, " If dataset name is specified, only that "
817 "dataset is dumped\n");
818 (void) fprintf(stderr
, " If object numbers or object number "
819 "ranges are specified, only those\n"
820 " objects or ranges are dumped.\n\n");
821 (void) fprintf(stderr
,
822 " Object ranges take the form <start>:<end>[:<flags>]\n"
823 " start Starting object number\n"
824 " end Ending object number, or -1 for no upper bound\n"
825 " flags Optional flags to select object types:\n"
826 " A All objects (this is the default)\n"
827 " d ZFS directories\n"
829 " m SPA space maps\n"
831 " - Negate effect of next flag\n\n");
832 (void) fprintf(stderr
, " Options to control amount of output:\n");
833 (void) fprintf(stderr
, " -b --block-stats "
834 "block statistics\n");
835 (void) fprintf(stderr
, " -B --backup "
837 (void) fprintf(stderr
, " -c --checksum "
838 "checksum all metadata (twice for all data) blocks\n");
839 (void) fprintf(stderr
, " -C --config "
840 "config (or cachefile if alone)\n");
841 (void) fprintf(stderr
, " -d --datasets "
843 (void) fprintf(stderr
, " -D --dedup-stats "
844 "dedup statistics\n");
845 (void) fprintf(stderr
, " -E --embedded-block-pointer=INTEGER\n"
846 " decode and display block "
847 "from an embedded block pointer\n");
848 (void) fprintf(stderr
, " -h --history "
850 (void) fprintf(stderr
, " -i --intent-logs "
852 (void) fprintf(stderr
, " -l --label "
853 "read label contents\n");
854 (void) fprintf(stderr
, " -k --checkpointed-state "
855 "examine the checkpointed state of the pool\n");
856 (void) fprintf(stderr
, " -L --disable-leak-tracking "
857 "disable leak tracking (do not load spacemaps)\n");
858 (void) fprintf(stderr
, " -m --metaslabs "
860 (void) fprintf(stderr
, " -M --metaslab-groups "
861 "metaslab groups\n");
862 (void) fprintf(stderr
, " -O --object-lookups "
863 "perform object lookups by path\n");
864 (void) fprintf(stderr
, " -r --copy-object "
865 "copy an object by path to file\n");
866 (void) fprintf(stderr
, " -R --read-block "
867 "read and display block from a device\n");
868 (void) fprintf(stderr
, " -s --io-stats "
869 "report stats on zdb's I/O\n");
870 (void) fprintf(stderr
, " -S --simulate-dedup "
871 "simulate dedup to measure effect\n");
872 (void) fprintf(stderr
, " -v --verbose "
873 "verbose (applies to all others)\n");
874 (void) fprintf(stderr
, " -y --livelist "
875 "perform livelist and metaslab validation on any livelists being "
877 (void) fprintf(stderr
, " Below options are intended for use "
878 "with other options:\n");
879 (void) fprintf(stderr
, " -A --ignore-assertions "
880 "ignore assertions (-A), enable panic recovery (-AA) or both "
882 (void) fprintf(stderr
, " -e --exported "
883 "pool is exported/destroyed/has altroot/not in a cachefile\n");
884 (void) fprintf(stderr
, " -F --automatic-rewind "
885 "attempt automatic rewind within safe range of transaction "
887 (void) fprintf(stderr
, " -G --dump-debug-msg "
888 "dump zfs_dbgmsg buffer before exiting\n");
889 (void) fprintf(stderr
, " -I --inflight=INTEGER "
890 "specify the maximum number of checksumming I/Os "
891 "[default is 200]\n");
892 (void) fprintf(stderr
, " -K --key=KEY "
893 "decryption key for encrypted dataset\n");
894 (void) fprintf(stderr
, " -o --option=\"OPTION=INTEGER\" "
895 "set global variable to an unsigned 32-bit integer\n");
896 (void) fprintf(stderr
, " -p --path==PATH "
897 "use one or more with -e to specify path to vdev dir\n");
898 (void) fprintf(stderr
, " -P --parseable "
899 "print numbers in parseable form\n");
900 (void) fprintf(stderr
, " -q --skip-label "
901 "don't print label contents\n");
902 (void) fprintf(stderr
, " -t --txg=INTEGER "
903 "highest txg to use when searching for uberblocks\n");
904 (void) fprintf(stderr
, " -T --brt-stats "
906 (void) fprintf(stderr
, " -u --uberblock "
908 (void) fprintf(stderr
, " -U --cachefile=PATH "
909 "use alternate cachefile\n");
910 (void) fprintf(stderr
, " -V --verbatim "
911 "do verbatim import\n");
912 (void) fprintf(stderr
, " -x --dump-blocks=PATH "
913 "dump all read blocks into specified directory\n");
914 (void) fprintf(stderr
, " -X --extreme-rewind "
915 "attempt extreme rewind (does not work with dataset)\n");
916 (void) fprintf(stderr
, " -Y --all-reconstruction "
917 "attempt all reconstruction combinations for split blocks\n");
918 (void) fprintf(stderr
, " -Z --zstd-headers "
919 "show ZSTD headers \n");
920 (void) fprintf(stderr
, "Specify an option more than once (e.g. -bb) "
921 "to make only that option verbose\n");
922 (void) fprintf(stderr
, "Default is to dump everything non-verbosely\n");
927 dump_debug_buffer(void)
931 (void) fflush(stdout
);
932 zfs_dbgmsg_print("zdb");
937 * Called for usage errors that are discovered after a call to spa_open(),
938 * dmu_bonus_hold(), or pool_match(). abort() is called for other errors.
942 fatal(const char *fmt
, ...)
947 (void) fprintf(stderr
, "%s: ", cmdname
);
948 (void) vfprintf(stderr
, fmt
, ap
);
950 (void) fprintf(stderr
, "\n");
958 dump_packed_nvlist(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
962 size_t nvsize
= *(uint64_t *)data
;
963 char *packed
= umem_alloc(nvsize
, UMEM_NOFAIL
);
965 VERIFY(0 == dmu_read(os
, object
, 0, nvsize
, packed
, DMU_READ_PREFETCH
));
967 VERIFY(nvlist_unpack(packed
, nvsize
, &nv
, 0) == 0);
969 umem_free(packed
, nvsize
);
977 dump_history_offsets(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
979 (void) os
, (void) object
, (void) size
;
980 spa_history_phys_t
*shp
= data
;
985 (void) printf("\t\tpool_create_len = %llu\n",
986 (u_longlong_t
)shp
->sh_pool_create_len
);
987 (void) printf("\t\tphys_max_off = %llu\n",
988 (u_longlong_t
)shp
->sh_phys_max_off
);
989 (void) printf("\t\tbof = %llu\n",
990 (u_longlong_t
)shp
->sh_bof
);
991 (void) printf("\t\teof = %llu\n",
992 (u_longlong_t
)shp
->sh_eof
);
993 (void) printf("\t\trecords_lost = %llu\n",
994 (u_longlong_t
)shp
->sh_records_lost
);
998 zdb_nicenum(uint64_t num
, char *buf
, size_t buflen
)
1001 (void) snprintf(buf
, buflen
, "%llu", (longlong_t
)num
);
1003 nicenum(num
, buf
, buflen
);
1007 zdb_nicebytes(uint64_t bytes
, char *buf
, size_t buflen
)
1010 (void) snprintf(buf
, buflen
, "%llu", (longlong_t
)bytes
);
1012 zfs_nicebytes(bytes
, buf
, buflen
);
1015 static const char histo_stars
[] = "****************************************";
1016 static const uint64_t histo_width
= sizeof (histo_stars
) - 1;
1019 dump_histogram(const uint64_t *histo
, int size
, int offset
)
1022 int minidx
= size
- 1;
1026 for (i
= 0; i
< size
; i
++) {
1037 if (max
< histo_width
)
1040 for (i
= minidx
; i
<= maxidx
; i
++) {
1041 (void) printf("\t\t\t%3u: %6llu %s\n",
1042 i
+ offset
, (u_longlong_t
)histo
[i
],
1043 &histo_stars
[(max
- histo
[i
]) * histo_width
/ max
]);
1048 dump_zap_stats(objset_t
*os
, uint64_t object
)
1053 error
= zap_get_stats(os
, object
, &zs
);
1057 if (zs
.zs_ptrtbl_len
== 0) {
1058 ASSERT(zs
.zs_num_blocks
== 1);
1059 (void) printf("\tmicrozap: %llu bytes, %llu entries\n",
1060 (u_longlong_t
)zs
.zs_blocksize
,
1061 (u_longlong_t
)zs
.zs_num_entries
);
1065 (void) printf("\tFat ZAP stats:\n");
1067 (void) printf("\t\tPointer table:\n");
1068 (void) printf("\t\t\t%llu elements\n",
1069 (u_longlong_t
)zs
.zs_ptrtbl_len
);
1070 (void) printf("\t\t\tzt_blk: %llu\n",
1071 (u_longlong_t
)zs
.zs_ptrtbl_zt_blk
);
1072 (void) printf("\t\t\tzt_numblks: %llu\n",
1073 (u_longlong_t
)zs
.zs_ptrtbl_zt_numblks
);
1074 (void) printf("\t\t\tzt_shift: %llu\n",
1075 (u_longlong_t
)zs
.zs_ptrtbl_zt_shift
);
1076 (void) printf("\t\t\tzt_blks_copied: %llu\n",
1077 (u_longlong_t
)zs
.zs_ptrtbl_blks_copied
);
1078 (void) printf("\t\t\tzt_nextblk: %llu\n",
1079 (u_longlong_t
)zs
.zs_ptrtbl_nextblk
);
1081 (void) printf("\t\tZAP entries: %llu\n",
1082 (u_longlong_t
)zs
.zs_num_entries
);
1083 (void) printf("\t\tLeaf blocks: %llu\n",
1084 (u_longlong_t
)zs
.zs_num_leafs
);
1085 (void) printf("\t\tTotal blocks: %llu\n",
1086 (u_longlong_t
)zs
.zs_num_blocks
);
1087 (void) printf("\t\tzap_block_type: 0x%llx\n",
1088 (u_longlong_t
)zs
.zs_block_type
);
1089 (void) printf("\t\tzap_magic: 0x%llx\n",
1090 (u_longlong_t
)zs
.zs_magic
);
1091 (void) printf("\t\tzap_salt: 0x%llx\n",
1092 (u_longlong_t
)zs
.zs_salt
);
1094 (void) printf("\t\tLeafs with 2^n pointers:\n");
1095 dump_histogram(zs
.zs_leafs_with_2n_pointers
, ZAP_HISTOGRAM_SIZE
, 0);
1097 (void) printf("\t\tBlocks with n*5 entries:\n");
1098 dump_histogram(zs
.zs_blocks_with_n5_entries
, ZAP_HISTOGRAM_SIZE
, 0);
1100 (void) printf("\t\tBlocks n/10 full:\n");
1101 dump_histogram(zs
.zs_blocks_n_tenths_full
, ZAP_HISTOGRAM_SIZE
, 0);
1103 (void) printf("\t\tEntries with n chunks:\n");
1104 dump_histogram(zs
.zs_entries_using_n_chunks
, ZAP_HISTOGRAM_SIZE
, 0);
1106 (void) printf("\t\tBuckets with n entries:\n");
1107 dump_histogram(zs
.zs_buckets_with_n_entries
, ZAP_HISTOGRAM_SIZE
, 0);
1111 dump_none(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1113 (void) os
, (void) object
, (void) data
, (void) size
;
1117 dump_unknown(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1119 (void) os
, (void) object
, (void) data
, (void) size
;
1120 (void) printf("\tUNKNOWN OBJECT TYPE\n");
1124 dump_uint8(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1126 (void) os
, (void) object
, (void) data
, (void) size
;
1130 dump_uint64(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1134 if (dump_opt
['d'] < 6)
1138 dmu_object_info_t doi
;
1140 VERIFY0(dmu_object_info(os
, object
, &doi
));
1141 size
= doi
.doi_max_offset
;
1143 * We cap the size at 1 mebibyte here to prevent
1144 * allocation failures and nigh-infinite printing if the
1145 * object is extremely large.
1147 oursize
= MIN(size
, 1 << 20);
1148 arr
= kmem_alloc(oursize
, KM_SLEEP
);
1150 int err
= dmu_read(os
, object
, 0, oursize
, arr
, 0);
1152 (void) printf("got error %u from dmu_read\n", err
);
1153 kmem_free(arr
, oursize
);
1158 * Even though the allocation is already done in this code path,
1159 * we still cap the size to prevent excessive printing.
1161 oursize
= MIN(size
, 1 << 20);
1167 kmem_free(arr
, oursize
);
1168 (void) printf("\t\t[]\n");
1172 (void) printf("\t\t[%0llx", (u_longlong_t
)arr
[0]);
1173 for (size_t i
= 1; i
* sizeof (uint64_t) < oursize
; i
++) {
1175 (void) printf(", %0llx", (u_longlong_t
)arr
[i
]);
1177 (void) printf(",\n\t\t%0llx", (u_longlong_t
)arr
[i
]);
1179 if (oursize
!= size
)
1180 (void) printf(", ... ");
1181 (void) printf("]\n");
1184 kmem_free(arr
, oursize
);
1188 dump_zap(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1190 (void) data
, (void) size
;
1192 zap_attribute_t attr
;
1196 dump_zap_stats(os
, object
);
1197 (void) printf("\n");
1199 for (zap_cursor_init(&zc
, os
, object
);
1200 zap_cursor_retrieve(&zc
, &attr
) == 0;
1201 zap_cursor_advance(&zc
)) {
1202 (void) printf("\t\t%s = ", attr
.za_name
);
1203 if (attr
.za_num_integers
== 0) {
1204 (void) printf("\n");
1207 prop
= umem_zalloc(attr
.za_num_integers
*
1208 attr
.za_integer_length
, UMEM_NOFAIL
);
1209 (void) zap_lookup(os
, object
, attr
.za_name
,
1210 attr
.za_integer_length
, attr
.za_num_integers
, prop
);
1211 if (attr
.za_integer_length
== 1) {
1212 if (strcmp(attr
.za_name
,
1213 DSL_CRYPTO_KEY_MASTER_KEY
) == 0 ||
1214 strcmp(attr
.za_name
,
1215 DSL_CRYPTO_KEY_HMAC_KEY
) == 0 ||
1216 strcmp(attr
.za_name
, DSL_CRYPTO_KEY_IV
) == 0 ||
1217 strcmp(attr
.za_name
, DSL_CRYPTO_KEY_MAC
) == 0 ||
1218 strcmp(attr
.za_name
, DMU_POOL_CHECKSUM_SALT
) == 0) {
1221 for (i
= 0; i
< attr
.za_num_integers
; i
++) {
1222 (void) printf("%02x", u8
[i
]);
1225 (void) printf("%s", (char *)prop
);
1228 for (i
= 0; i
< attr
.za_num_integers
; i
++) {
1229 switch (attr
.za_integer_length
) {
1231 (void) printf("%u ",
1232 ((uint16_t *)prop
)[i
]);
1235 (void) printf("%u ",
1236 ((uint32_t *)prop
)[i
]);
1239 (void) printf("%lld ",
1240 (u_longlong_t
)((int64_t *)prop
)[i
]);
1245 (void) printf("\n");
1246 umem_free(prop
, attr
.za_num_integers
* attr
.za_integer_length
);
1248 zap_cursor_fini(&zc
);
1252 dump_bpobj(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1254 bpobj_phys_t
*bpop
= data
;
1256 char bytes
[32], comp
[32], uncomp
[32];
1258 /* make sure the output won't get truncated */
1259 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
1260 _Static_assert(sizeof (comp
) >= NN_NUMBUF_SZ
, "comp truncated");
1261 _Static_assert(sizeof (uncomp
) >= NN_NUMBUF_SZ
, "uncomp truncated");
1266 zdb_nicenum(bpop
->bpo_bytes
, bytes
, sizeof (bytes
));
1267 zdb_nicenum(bpop
->bpo_comp
, comp
, sizeof (comp
));
1268 zdb_nicenum(bpop
->bpo_uncomp
, uncomp
, sizeof (uncomp
));
1270 (void) printf("\t\tnum_blkptrs = %llu\n",
1271 (u_longlong_t
)bpop
->bpo_num_blkptrs
);
1272 (void) printf("\t\tbytes = %s\n", bytes
);
1273 if (size
>= BPOBJ_SIZE_V1
) {
1274 (void) printf("\t\tcomp = %s\n", comp
);
1275 (void) printf("\t\tuncomp = %s\n", uncomp
);
1277 if (size
>= BPOBJ_SIZE_V2
) {
1278 (void) printf("\t\tsubobjs = %llu\n",
1279 (u_longlong_t
)bpop
->bpo_subobjs
);
1280 (void) printf("\t\tnum_subobjs = %llu\n",
1281 (u_longlong_t
)bpop
->bpo_num_subobjs
);
1283 if (size
>= sizeof (*bpop
)) {
1284 (void) printf("\t\tnum_freed = %llu\n",
1285 (u_longlong_t
)bpop
->bpo_num_freed
);
1288 if (dump_opt
['d'] < 5)
1291 for (i
= 0; i
< bpop
->bpo_num_blkptrs
; i
++) {
1292 char blkbuf
[BP_SPRINTF_LEN
];
1295 int err
= dmu_read(os
, object
,
1296 i
* sizeof (bp
), sizeof (bp
), &bp
, 0);
1298 (void) printf("got error %u from dmu_read\n", err
);
1301 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
), &bp
,
1303 (void) printf("\t%s\n", blkbuf
);
1308 dump_bpobj_subobjs(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1310 (void) data
, (void) size
;
1311 dmu_object_info_t doi
;
1314 VERIFY0(dmu_object_info(os
, object
, &doi
));
1315 uint64_t *subobjs
= kmem_alloc(doi
.doi_max_offset
, KM_SLEEP
);
1317 int err
= dmu_read(os
, object
, 0, doi
.doi_max_offset
, subobjs
, 0);
1319 (void) printf("got error %u from dmu_read\n", err
);
1320 kmem_free(subobjs
, doi
.doi_max_offset
);
1324 int64_t last_nonzero
= -1;
1325 for (i
= 0; i
< doi
.doi_max_offset
/ 8; i
++) {
1326 if (subobjs
[i
] != 0)
1330 for (i
= 0; i
<= last_nonzero
; i
++) {
1331 (void) printf("\t%llu\n", (u_longlong_t
)subobjs
[i
]);
1333 kmem_free(subobjs
, doi
.doi_max_offset
);
1337 dump_ddt_zap(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1339 (void) data
, (void) size
;
1340 dump_zap_stats(os
, object
);
1341 /* contents are printed elsewhere, properly decoded */
1345 dump_sa_attrs(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1347 (void) data
, (void) size
;
1349 zap_attribute_t attr
;
1351 dump_zap_stats(os
, object
);
1352 (void) printf("\n");
1354 for (zap_cursor_init(&zc
, os
, object
);
1355 zap_cursor_retrieve(&zc
, &attr
) == 0;
1356 zap_cursor_advance(&zc
)) {
1357 (void) printf("\t\t%s = ", attr
.za_name
);
1358 if (attr
.za_num_integers
== 0) {
1359 (void) printf("\n");
1362 (void) printf(" %llx : [%d:%d:%d]\n",
1363 (u_longlong_t
)attr
.za_first_integer
,
1364 (int)ATTR_LENGTH(attr
.za_first_integer
),
1365 (int)ATTR_BSWAP(attr
.za_first_integer
),
1366 (int)ATTR_NUM(attr
.za_first_integer
));
1368 zap_cursor_fini(&zc
);
1372 dump_sa_layouts(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1374 (void) data
, (void) size
;
1376 zap_attribute_t attr
;
1377 uint16_t *layout_attrs
;
1380 dump_zap_stats(os
, object
);
1381 (void) printf("\n");
1383 for (zap_cursor_init(&zc
, os
, object
);
1384 zap_cursor_retrieve(&zc
, &attr
) == 0;
1385 zap_cursor_advance(&zc
)) {
1386 (void) printf("\t\t%s = [", attr
.za_name
);
1387 if (attr
.za_num_integers
== 0) {
1388 (void) printf("\n");
1392 VERIFY(attr
.za_integer_length
== 2);
1393 layout_attrs
= umem_zalloc(attr
.za_num_integers
*
1394 attr
.za_integer_length
, UMEM_NOFAIL
);
1396 VERIFY(zap_lookup(os
, object
, attr
.za_name
,
1397 attr
.za_integer_length
,
1398 attr
.za_num_integers
, layout_attrs
) == 0);
1400 for (i
= 0; i
!= attr
.za_num_integers
; i
++)
1401 (void) printf(" %d ", (int)layout_attrs
[i
]);
1402 (void) printf("]\n");
1403 umem_free(layout_attrs
,
1404 attr
.za_num_integers
* attr
.za_integer_length
);
1406 zap_cursor_fini(&zc
);
1410 dump_zpldir(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
1412 (void) data
, (void) size
;
1414 zap_attribute_t attr
;
1415 const char *typenames
[] = {
1416 /* 0 */ "not specified",
1418 /* 2 */ "Character Device",
1419 /* 3 */ "3 (invalid)",
1420 /* 4 */ "Directory",
1421 /* 5 */ "5 (invalid)",
1422 /* 6 */ "Block Device",
1423 /* 7 */ "7 (invalid)",
1424 /* 8 */ "Regular File",
1425 /* 9 */ "9 (invalid)",
1426 /* 10 */ "Symbolic Link",
1427 /* 11 */ "11 (invalid)",
1430 /* 14 */ "Event Port",
1431 /* 15 */ "15 (invalid)",
1434 dump_zap_stats(os
, object
);
1435 (void) printf("\n");
1437 for (zap_cursor_init(&zc
, os
, object
);
1438 zap_cursor_retrieve(&zc
, &attr
) == 0;
1439 zap_cursor_advance(&zc
)) {
1440 (void) printf("\t\t%s = %lld (type: %s)\n",
1441 attr
.za_name
, ZFS_DIRENT_OBJ(attr
.za_first_integer
),
1442 typenames
[ZFS_DIRENT_TYPE(attr
.za_first_integer
)]);
1444 zap_cursor_fini(&zc
);
1448 get_dtl_refcount(vdev_t
*vd
)
1452 if (vd
->vdev_ops
->vdev_op_leaf
) {
1453 space_map_t
*sm
= vd
->vdev_dtl_sm
;
1456 sm
->sm_dbuf
->db_size
== sizeof (space_map_phys_t
))
1461 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++)
1462 refcount
+= get_dtl_refcount(vd
->vdev_child
[c
]);
1467 get_metaslab_refcount(vdev_t
*vd
)
1471 if (vd
->vdev_top
== vd
) {
1472 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
1473 space_map_t
*sm
= vd
->vdev_ms
[m
]->ms_sm
;
1476 sm
->sm_dbuf
->db_size
== sizeof (space_map_phys_t
))
1480 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++)
1481 refcount
+= get_metaslab_refcount(vd
->vdev_child
[c
]);
1487 get_obsolete_refcount(vdev_t
*vd
)
1489 uint64_t obsolete_sm_object
;
1492 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
1493 if (vd
->vdev_top
== vd
&& obsolete_sm_object
!= 0) {
1494 dmu_object_info_t doi
;
1495 VERIFY0(dmu_object_info(vd
->vdev_spa
->spa_meta_objset
,
1496 obsolete_sm_object
, &doi
));
1497 if (doi
.doi_bonus_size
== sizeof (space_map_phys_t
)) {
1501 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
1502 ASSERT3U(obsolete_sm_object
, ==, 0);
1504 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++) {
1505 refcount
+= get_obsolete_refcount(vd
->vdev_child
[c
]);
1512 get_prev_obsolete_spacemap_refcount(spa_t
*spa
)
1515 spa
->spa_condensing_indirect_phys
.scip_prev_obsolete_sm_object
;
1516 if (prev_obj
!= 0) {
1517 dmu_object_info_t doi
;
1518 VERIFY0(dmu_object_info(spa
->spa_meta_objset
, prev_obj
, &doi
));
1519 if (doi
.doi_bonus_size
== sizeof (space_map_phys_t
)) {
1527 get_checkpoint_refcount(vdev_t
*vd
)
1531 if (vd
->vdev_top
== vd
&& vd
->vdev_top_zap
!= 0 &&
1532 zap_contains(spa_meta_objset(vd
->vdev_spa
),
1533 vd
->vdev_top_zap
, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) == 0)
1536 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1537 refcount
+= get_checkpoint_refcount(vd
->vdev_child
[c
]);
1543 get_log_spacemap_refcount(spa_t
*spa
)
1545 return (avl_numnodes(&spa
->spa_sm_logs_by_txg
));
1549 verify_spacemap_refcounts(spa_t
*spa
)
1551 uint64_t expected_refcount
= 0;
1552 uint64_t actual_refcount
;
1554 (void) feature_get_refcount(spa
,
1555 &spa_feature_table
[SPA_FEATURE_SPACEMAP_HISTOGRAM
],
1556 &expected_refcount
);
1557 actual_refcount
= get_dtl_refcount(spa
->spa_root_vdev
);
1558 actual_refcount
+= get_metaslab_refcount(spa
->spa_root_vdev
);
1559 actual_refcount
+= get_obsolete_refcount(spa
->spa_root_vdev
);
1560 actual_refcount
+= get_prev_obsolete_spacemap_refcount(spa
);
1561 actual_refcount
+= get_checkpoint_refcount(spa
->spa_root_vdev
);
1562 actual_refcount
+= get_log_spacemap_refcount(spa
);
1564 if (expected_refcount
!= actual_refcount
) {
1565 (void) printf("space map refcount mismatch: expected %lld != "
1567 (longlong_t
)expected_refcount
,
1568 (longlong_t
)actual_refcount
);
1575 dump_spacemap(objset_t
*os
, space_map_t
*sm
)
1577 const char *ddata
[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1578 "INVALID", "INVALID", "INVALID", "INVALID" };
1583 (void) printf("space map object %llu:\n",
1584 (longlong_t
)sm
->sm_object
);
1585 (void) printf(" smp_length = 0x%llx\n",
1586 (longlong_t
)sm
->sm_phys
->smp_length
);
1587 (void) printf(" smp_alloc = 0x%llx\n",
1588 (longlong_t
)sm
->sm_phys
->smp_alloc
);
1590 if (dump_opt
['d'] < 6 && dump_opt
['m'] < 4)
1594 * Print out the freelist entries in both encoded and decoded form.
1596 uint8_t mapshift
= sm
->sm_shift
;
1598 uint64_t word
, entry_id
= 0;
1599 for (uint64_t offset
= 0; offset
< space_map_length(sm
);
1600 offset
+= sizeof (word
)) {
1602 VERIFY0(dmu_read(os
, space_map_object(sm
), offset
,
1603 sizeof (word
), &word
, DMU_READ_PREFETCH
));
1605 if (sm_entry_is_debug(word
)) {
1606 uint64_t de_txg
= SM_DEBUG_TXG_DECODE(word
);
1607 uint64_t de_sync_pass
= SM_DEBUG_SYNCPASS_DECODE(word
);
1610 "\t [%6llu] PADDING\n",
1611 (u_longlong_t
)entry_id
);
1614 "\t [%6llu] %s: txg %llu pass %llu\n",
1615 (u_longlong_t
)entry_id
,
1616 ddata
[SM_DEBUG_ACTION_DECODE(word
)],
1617 (u_longlong_t
)de_txg
,
1618 (u_longlong_t
)de_sync_pass
);
1626 uint64_t entry_off
, entry_run
, entry_vdev
= SM_NO_VDEVID
;
1628 if (sm_entry_is_single_word(word
)) {
1629 entry_type
= (SM_TYPE_DECODE(word
) == SM_ALLOC
) ?
1631 entry_off
= (SM_OFFSET_DECODE(word
) << mapshift
) +
1633 entry_run
= SM_RUN_DECODE(word
) << mapshift
;
1636 /* it is a two-word entry so we read another word */
1637 ASSERT(sm_entry_is_double_word(word
));
1639 uint64_t extra_word
;
1640 offset
+= sizeof (extra_word
);
1641 VERIFY0(dmu_read(os
, space_map_object(sm
), offset
,
1642 sizeof (extra_word
), &extra_word
,
1643 DMU_READ_PREFETCH
));
1645 ASSERT3U(offset
, <=, space_map_length(sm
));
1647 entry_run
= SM2_RUN_DECODE(word
) << mapshift
;
1648 entry_vdev
= SM2_VDEV_DECODE(word
);
1649 entry_type
= (SM2_TYPE_DECODE(extra_word
) == SM_ALLOC
) ?
1651 entry_off
= (SM2_OFFSET_DECODE(extra_word
) <<
1652 mapshift
) + sm
->sm_start
;
1656 (void) printf("\t [%6llu] %c range:"
1657 " %010llx-%010llx size: %06llx vdev: %06llu words: %u\n",
1658 (u_longlong_t
)entry_id
,
1659 entry_type
, (u_longlong_t
)entry_off
,
1660 (u_longlong_t
)(entry_off
+ entry_run
),
1661 (u_longlong_t
)entry_run
,
1662 (u_longlong_t
)entry_vdev
, words
);
1664 if (entry_type
== 'A')
1670 if (alloc
!= space_map_allocated(sm
)) {
1671 (void) printf("space_map_object alloc (%lld) INCONSISTENT "
1672 "with space map summary (%lld)\n",
1673 (longlong_t
)space_map_allocated(sm
), (longlong_t
)alloc
);
1678 dump_metaslab_stats(metaslab_t
*msp
)
1681 range_tree_t
*rt
= msp
->ms_allocatable
;
1682 zfs_btree_t
*t
= &msp
->ms_allocatable_by_size
;
1683 int free_pct
= range_tree_space(rt
) * 100 / msp
->ms_size
;
1685 /* max sure nicenum has enough space */
1686 _Static_assert(sizeof (maxbuf
) >= NN_NUMBUF_SZ
, "maxbuf truncated");
1688 zdb_nicenum(metaslab_largest_allocatable(msp
), maxbuf
, sizeof (maxbuf
));
1690 (void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
1691 "segments", zfs_btree_numnodes(t
), "maxsize", maxbuf
,
1692 "freepct", free_pct
);
1693 (void) printf("\tIn-memory histogram:\n");
1694 dump_histogram(rt
->rt_histogram
, RANGE_TREE_HISTOGRAM_SIZE
, 0);
1698 dump_metaslab(metaslab_t
*msp
)
1700 vdev_t
*vd
= msp
->ms_group
->mg_vd
;
1701 spa_t
*spa
= vd
->vdev_spa
;
1702 space_map_t
*sm
= msp
->ms_sm
;
1705 zdb_nicenum(msp
->ms_size
- space_map_allocated(sm
), freebuf
,
1709 "\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
1710 (u_longlong_t
)msp
->ms_id
, (u_longlong_t
)msp
->ms_start
,
1711 (u_longlong_t
)space_map_object(sm
), freebuf
);
1713 if (dump_opt
['m'] > 2 && !dump_opt
['L']) {
1714 mutex_enter(&msp
->ms_lock
);
1715 VERIFY0(metaslab_load(msp
));
1716 range_tree_stat_verify(msp
->ms_allocatable
);
1717 dump_metaslab_stats(msp
);
1718 metaslab_unload(msp
);
1719 mutex_exit(&msp
->ms_lock
);
1722 if (dump_opt
['m'] > 1 && sm
!= NULL
&&
1723 spa_feature_is_active(spa
, SPA_FEATURE_SPACEMAP_HISTOGRAM
)) {
1725 * The space map histogram represents free space in chunks
1726 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1728 (void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1729 (u_longlong_t
)msp
->ms_fragmentation
);
1730 dump_histogram(sm
->sm_phys
->smp_histogram
,
1731 SPACE_MAP_HISTOGRAM_SIZE
, sm
->sm_shift
);
1734 if (vd
->vdev_ops
== &vdev_draid_ops
)
1735 ASSERT3U(msp
->ms_size
, <=, 1ULL << vd
->vdev_ms_shift
);
1737 ASSERT3U(msp
->ms_size
, ==, 1ULL << vd
->vdev_ms_shift
);
1739 dump_spacemap(spa
->spa_meta_objset
, msp
->ms_sm
);
1741 if (spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
)) {
1742 (void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1743 (u_longlong_t
)metaslab_unflushed_txg(msp
));
1748 print_vdev_metaslab_header(vdev_t
*vd
)
1750 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
1751 const char *bias_str
= "";
1752 if (alloc_bias
== VDEV_BIAS_LOG
|| vd
->vdev_islog
) {
1753 bias_str
= VDEV_ALLOC_BIAS_LOG
;
1754 } else if (alloc_bias
== VDEV_BIAS_SPECIAL
) {
1755 bias_str
= VDEV_ALLOC_BIAS_SPECIAL
;
1756 } else if (alloc_bias
== VDEV_BIAS_DEDUP
) {
1757 bias_str
= VDEV_ALLOC_BIAS_DEDUP
;
1760 uint64_t ms_flush_data_obj
= 0;
1761 if (vd
->vdev_top_zap
!= 0) {
1762 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
),
1763 vd
->vdev_top_zap
, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
,
1764 sizeof (uint64_t), 1, &ms_flush_data_obj
);
1765 if (error
!= ENOENT
) {
1770 (void) printf("\tvdev %10llu %s",
1771 (u_longlong_t
)vd
->vdev_id
, bias_str
);
1773 if (ms_flush_data_obj
!= 0) {
1774 (void) printf(" ms_unflushed_phys object %llu",
1775 (u_longlong_t
)ms_flush_data_obj
);
1778 (void) printf("\n\t%-10s%5llu %-19s %-15s %-12s\n",
1779 "metaslabs", (u_longlong_t
)vd
->vdev_ms_count
,
1780 "offset", "spacemap", "free");
1781 (void) printf("\t%15s %19s %15s %12s\n",
1782 "---------------", "-------------------",
1783 "---------------", "------------");
1787 dump_metaslab_groups(spa_t
*spa
, boolean_t show_special
)
1789 vdev_t
*rvd
= spa
->spa_root_vdev
;
1790 metaslab_class_t
*mc
= spa_normal_class(spa
);
1791 metaslab_class_t
*smc
= spa_special_class(spa
);
1792 uint64_t fragmentation
;
1794 metaslab_class_histogram_verify(mc
);
1796 for (unsigned c
= 0; c
< rvd
->vdev_children
; c
++) {
1797 vdev_t
*tvd
= rvd
->vdev_child
[c
];
1798 metaslab_group_t
*mg
= tvd
->vdev_mg
;
1800 if (mg
== NULL
|| (mg
->mg_class
!= mc
&&
1801 (!show_special
|| mg
->mg_class
!= smc
)))
1804 metaslab_group_histogram_verify(mg
);
1805 mg
->mg_fragmentation
= metaslab_group_fragmentation(mg
);
1807 (void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1809 (u_longlong_t
)tvd
->vdev_id
,
1810 (u_longlong_t
)tvd
->vdev_ms_count
);
1811 if (mg
->mg_fragmentation
== ZFS_FRAG_INVALID
) {
1812 (void) printf("%3s\n", "-");
1814 (void) printf("%3llu%%\n",
1815 (u_longlong_t
)mg
->mg_fragmentation
);
1817 dump_histogram(mg
->mg_histogram
, RANGE_TREE_HISTOGRAM_SIZE
, 0);
1820 (void) printf("\tpool %s\tfragmentation", spa_name(spa
));
1821 fragmentation
= metaslab_class_fragmentation(mc
);
1822 if (fragmentation
== ZFS_FRAG_INVALID
)
1823 (void) printf("\t%3s\n", "-");
1825 (void) printf("\t%3llu%%\n", (u_longlong_t
)fragmentation
);
1826 dump_histogram(mc
->mc_histogram
, RANGE_TREE_HISTOGRAM_SIZE
, 0);
1830 print_vdev_indirect(vdev_t
*vd
)
1832 vdev_indirect_config_t
*vic
= &vd
->vdev_indirect_config
;
1833 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
1834 vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
;
1837 ASSERT3P(vib
, ==, NULL
);
1841 ASSERT3U(vdev_indirect_mapping_object(vim
), ==,
1842 vic
->vic_mapping_object
);
1843 ASSERT3U(vdev_indirect_births_object(vib
), ==,
1844 vic
->vic_births_object
);
1846 (void) printf("indirect births obj %llu:\n",
1847 (longlong_t
)vic
->vic_births_object
);
1848 (void) printf(" vib_count = %llu\n",
1849 (longlong_t
)vdev_indirect_births_count(vib
));
1850 for (uint64_t i
= 0; i
< vdev_indirect_births_count(vib
); i
++) {
1851 vdev_indirect_birth_entry_phys_t
*cur_vibe
=
1852 &vib
->vib_entries
[i
];
1853 (void) printf("\toffset %llx -> txg %llu\n",
1854 (longlong_t
)cur_vibe
->vibe_offset
,
1855 (longlong_t
)cur_vibe
->vibe_phys_birth_txg
);
1857 (void) printf("\n");
1859 (void) printf("indirect mapping obj %llu:\n",
1860 (longlong_t
)vic
->vic_mapping_object
);
1861 (void) printf(" vim_max_offset = 0x%llx\n",
1862 (longlong_t
)vdev_indirect_mapping_max_offset(vim
));
1863 (void) printf(" vim_bytes_mapped = 0x%llx\n",
1864 (longlong_t
)vdev_indirect_mapping_bytes_mapped(vim
));
1865 (void) printf(" vim_count = %llu\n",
1866 (longlong_t
)vdev_indirect_mapping_num_entries(vim
));
1868 if (dump_opt
['d'] <= 5 && dump_opt
['m'] <= 3)
1871 uint32_t *counts
= vdev_indirect_mapping_load_obsolete_counts(vim
);
1873 for (uint64_t i
= 0; i
< vdev_indirect_mapping_num_entries(vim
); i
++) {
1874 vdev_indirect_mapping_entry_phys_t
*vimep
=
1875 &vim
->vim_entries
[i
];
1876 (void) printf("\t<%llx:%llx:%llx> -> "
1877 "<%llx:%llx:%llx> (%x obsolete)\n",
1878 (longlong_t
)vd
->vdev_id
,
1879 (longlong_t
)DVA_MAPPING_GET_SRC_OFFSET(vimep
),
1880 (longlong_t
)DVA_GET_ASIZE(&vimep
->vimep_dst
),
1881 (longlong_t
)DVA_GET_VDEV(&vimep
->vimep_dst
),
1882 (longlong_t
)DVA_GET_OFFSET(&vimep
->vimep_dst
),
1883 (longlong_t
)DVA_GET_ASIZE(&vimep
->vimep_dst
),
1886 (void) printf("\n");
1888 uint64_t obsolete_sm_object
;
1889 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
1890 if (obsolete_sm_object
!= 0) {
1891 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
1892 (void) printf("obsolete space map object %llu:\n",
1893 (u_longlong_t
)obsolete_sm_object
);
1894 ASSERT(vd
->vdev_obsolete_sm
!= NULL
);
1895 ASSERT3U(space_map_object(vd
->vdev_obsolete_sm
), ==,
1896 obsolete_sm_object
);
1897 dump_spacemap(mos
, vd
->vdev_obsolete_sm
);
1898 (void) printf("\n");
1903 dump_metaslabs(spa_t
*spa
)
1905 vdev_t
*vd
, *rvd
= spa
->spa_root_vdev
;
1906 uint64_t m
, c
= 0, children
= rvd
->vdev_children
;
1908 (void) printf("\nMetaslabs:\n");
1910 if (!dump_opt
['d'] && zopt_metaslab_args
> 0) {
1911 c
= zopt_metaslab
[0];
1914 (void) fatal("bad vdev id: %llu", (u_longlong_t
)c
);
1916 if (zopt_metaslab_args
> 1) {
1917 vd
= rvd
->vdev_child
[c
];
1918 print_vdev_metaslab_header(vd
);
1920 for (m
= 1; m
< zopt_metaslab_args
; m
++) {
1921 if (zopt_metaslab
[m
] < vd
->vdev_ms_count
)
1923 vd
->vdev_ms
[zopt_metaslab
[m
]]);
1925 (void) fprintf(stderr
, "bad metaslab "
1927 (u_longlong_t
)zopt_metaslab
[m
]);
1929 (void) printf("\n");
1934 for (; c
< children
; c
++) {
1935 vd
= rvd
->vdev_child
[c
];
1936 print_vdev_metaslab_header(vd
);
1938 print_vdev_indirect(vd
);
1940 for (m
= 0; m
< vd
->vdev_ms_count
; m
++)
1941 dump_metaslab(vd
->vdev_ms
[m
]);
1942 (void) printf("\n");
1947 dump_log_spacemaps(spa_t
*spa
)
1949 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
1952 (void) printf("\nLog Space Maps in Pool:\n");
1953 for (spa_log_sm_t
*sls
= avl_first(&spa
->spa_sm_logs_by_txg
);
1954 sls
; sls
= AVL_NEXT(&spa
->spa_sm_logs_by_txg
, sls
)) {
1955 space_map_t
*sm
= NULL
;
1956 VERIFY0(space_map_open(&sm
, spa_meta_objset(spa
),
1957 sls
->sls_sm_obj
, 0, UINT64_MAX
, SPA_MINBLOCKSHIFT
));
1959 (void) printf("Log Spacemap object %llu txg %llu\n",
1960 (u_longlong_t
)sls
->sls_sm_obj
, (u_longlong_t
)sls
->sls_txg
);
1961 dump_spacemap(spa
->spa_meta_objset
, sm
);
1962 space_map_close(sm
);
1964 (void) printf("\n");
1968 dump_dde(const ddt_t
*ddt
, const ddt_entry_t
*dde
, uint64_t index
)
1970 const ddt_phys_t
*ddp
= dde
->dde_phys
;
1971 const ddt_key_t
*ddk
= &dde
->dde_key
;
1972 const char *types
[4] = { "ditto", "single", "double", "triple" };
1973 char blkbuf
[BP_SPRINTF_LEN
];
1977 for (p
= 0; p
< DDT_PHYS_TYPES
; p
++, ddp
++) {
1978 if (ddp
->ddp_phys_birth
== 0)
1980 ddt_bp_create(ddt
->ddt_checksum
, ddk
, ddp
, &blk
);
1981 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), &blk
);
1982 (void) printf("index %llx refcnt %llu %s %s\n",
1983 (u_longlong_t
)index
, (u_longlong_t
)ddp
->ddp_refcnt
,
1989 dump_dedup_ratio(const ddt_stat_t
*dds
)
1991 double rL
, rP
, rD
, D
, dedup
, compress
, copies
;
1993 if (dds
->dds_blocks
== 0)
1996 rL
= (double)dds
->dds_ref_lsize
;
1997 rP
= (double)dds
->dds_ref_psize
;
1998 rD
= (double)dds
->dds_ref_dsize
;
1999 D
= (double)dds
->dds_dsize
;
2005 (void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
2006 "dedup * compress / copies = %.2f\n\n",
2007 dedup
, compress
, copies
, dedup
* compress
/ copies
);
2011 dump_ddt(ddt_t
*ddt
, enum ddt_type type
, enum ddt_class
class)
2013 char name
[DDT_NAMELEN
];
2016 dmu_object_info_t doi
;
2017 uint64_t count
, dspace
, mspace
;
2020 error
= ddt_object_info(ddt
, type
, class, &doi
);
2022 if (error
== ENOENT
)
2026 error
= ddt_object_count(ddt
, type
, class, &count
);
2031 dspace
= doi
.doi_physical_blocks_512
<< 9;
2032 mspace
= doi
.doi_fill_count
* doi
.doi_data_block_size
;
2034 ddt_object_name(ddt
, type
, class, name
);
2036 (void) printf("%s: %llu entries, size %llu on disk, %llu in core\n",
2038 (u_longlong_t
)count
,
2039 (u_longlong_t
)(dspace
/ count
),
2040 (u_longlong_t
)(mspace
/ count
));
2042 if (dump_opt
['D'] < 3)
2045 zpool_dump_ddt(NULL
, &ddt
->ddt_histogram
[type
][class]);
2047 if (dump_opt
['D'] < 4)
2050 if (dump_opt
['D'] < 5 && class == DDT_CLASS_UNIQUE
)
2053 (void) printf("%s contents:\n\n", name
);
2055 while ((error
= ddt_object_walk(ddt
, type
, class, &walk
, &dde
)) == 0)
2056 dump_dde(ddt
, &dde
, walk
);
2058 ASSERT3U(error
, ==, ENOENT
);
2060 (void) printf("\n");
2064 dump_all_ddts(spa_t
*spa
)
2066 ddt_histogram_t ddh_total
= {{{0}}};
2067 ddt_stat_t dds_total
= {0};
2069 for (enum zio_checksum c
= 0; c
< ZIO_CHECKSUM_FUNCTIONS
; c
++) {
2070 ddt_t
*ddt
= spa
->spa_ddt
[c
];
2071 for (enum ddt_type type
= 0; type
< DDT_TYPES
; type
++) {
2072 for (enum ddt_class
class = 0; class < DDT_CLASSES
;
2074 dump_ddt(ddt
, type
, class);
2079 ddt_get_dedup_stats(spa
, &dds_total
);
2081 if (dds_total
.dds_blocks
== 0) {
2082 (void) printf("All DDTs are empty\n");
2086 (void) printf("\n");
2088 if (dump_opt
['D'] > 1) {
2089 (void) printf("DDT histogram (aggregated over all DDTs):\n");
2090 ddt_get_dedup_histogram(spa
, &ddh_total
);
2091 zpool_dump_ddt(&dds_total
, &ddh_total
);
2094 dump_dedup_ratio(&dds_total
);
2098 dump_brt(spa_t
*spa
)
2100 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_BLOCK_CLONING
)) {
2101 printf("BRT: unsupported on this pool\n");
2105 if (!spa_feature_is_active(spa
, SPA_FEATURE_BLOCK_CLONING
)) {
2106 printf("BRT: empty\n");
2110 brt_t
*brt
= spa
->spa_brt
;
2113 char count
[32], used
[32], saved
[32];
2114 zdb_nicebytes(brt_get_used(spa
), used
, sizeof (used
));
2115 zdb_nicebytes(brt_get_saved(spa
), saved
, sizeof (saved
));
2116 uint64_t ratio
= brt_get_ratio(spa
);
2117 printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used
, saved
,
2118 (u_longlong_t
)(ratio
/ 100), (u_longlong_t
)(ratio
% 100));
2120 if (dump_opt
['T'] < 2)
2123 for (uint64_t vdevid
= 0; vdevid
< brt
->brt_nvdevs
; vdevid
++) {
2124 brt_vdev_t
*brtvd
= &brt
->brt_vdevs
[vdevid
];
2128 if (!brtvd
->bv_initiated
) {
2129 printf("BRT: vdev %" PRIu64
": empty\n", vdevid
);
2133 zdb_nicenum(brtvd
->bv_totalcount
, count
, sizeof (count
));
2134 zdb_nicebytes(brtvd
->bv_usedspace
, used
, sizeof (used
));
2135 zdb_nicebytes(brtvd
->bv_savedspace
, saved
, sizeof (saved
));
2136 printf("BRT: vdev %" PRIu64
": refcnt %s; used %s; saved %s\n",
2137 vdevid
, count
, used
, saved
);
2140 if (dump_opt
['T'] < 3)
2144 printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2146 for (uint64_t vdevid
= 0; vdevid
< brt
->brt_nvdevs
; vdevid
++) {
2147 brt_vdev_t
*brtvd
= &brt
->brt_vdevs
[vdevid
];
2148 if (brtvd
== NULL
|| !brtvd
->bv_initiated
)
2153 for (zap_cursor_init(&zc
, brt
->brt_mos
, brtvd
->bv_mos_entries
);
2154 zap_cursor_retrieve(&zc
, &za
) == 0;
2155 zap_cursor_advance(&zc
)) {
2156 uint64_t offset
= *(uint64_t *)za
.za_name
;
2157 uint64_t refcnt
= za
.za_first_integer
;
2159 snprintf(dva
, sizeof (dva
), "%" PRIu64
":%llx", vdevid
,
2160 (u_longlong_t
)offset
);
2161 printf("%-16s %-10llu\n", dva
, (u_longlong_t
)refcnt
);
2163 zap_cursor_fini(&zc
);
2168 dump_dtl_seg(void *arg
, uint64_t start
, uint64_t size
)
2172 (void) printf("%s [%llu,%llu) length %llu\n",
2174 (u_longlong_t
)start
,
2175 (u_longlong_t
)(start
+ size
),
2176 (u_longlong_t
)(size
));
2180 dump_dtl(vdev_t
*vd
, int indent
)
2182 spa_t
*spa
= vd
->vdev_spa
;
2184 const char *name
[DTL_TYPES
] = { "missing", "partial", "scrub",
2188 spa_vdev_state_enter(spa
, SCL_NONE
);
2189 required
= vdev_dtl_required(vd
);
2190 (void) spa_vdev_state_exit(spa
, NULL
, 0);
2193 (void) printf("\nDirty time logs:\n\n");
2195 (void) printf("\t%*s%s [%s]\n", indent
, "",
2196 vd
->vdev_path
? vd
->vdev_path
:
2197 vd
->vdev_parent
? vd
->vdev_ops
->vdev_op_type
: spa_name(spa
),
2198 required
? "DTL-required" : "DTL-expendable");
2200 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2201 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2202 if (range_tree_space(rt
) == 0)
2204 (void) snprintf(prefix
, sizeof (prefix
), "\t%*s%s",
2205 indent
+ 2, "", name
[t
]);
2206 range_tree_walk(rt
, dump_dtl_seg
, prefix
);
2207 if (dump_opt
['d'] > 5 && vd
->vdev_children
== 0)
2208 dump_spacemap(spa
->spa_meta_objset
,
2212 for (unsigned c
= 0; c
< vd
->vdev_children
; c
++)
2213 dump_dtl(vd
->vdev_child
[c
], indent
+ 4);
2217 dump_history(spa_t
*spa
)
2219 nvlist_t
**events
= NULL
;
2221 uint64_t resid
, len
, off
= 0;
2226 if ((buf
= malloc(SPA_OLD_MAXBLOCKSIZE
)) == NULL
) {
2227 (void) fprintf(stderr
, "%s: unable to allocate I/O buffer\n",
2233 len
= SPA_OLD_MAXBLOCKSIZE
;
2235 if ((error
= spa_history_get(spa
, &off
, &len
, buf
)) != 0) {
2236 (void) fprintf(stderr
, "Unable to read history: "
2237 "error %d\n", error
);
2242 if (zpool_history_unpack(buf
, len
, &resid
, &events
, &num
) != 0)
2248 (void) printf("\nHistory:\n");
2249 for (unsigned i
= 0; i
< num
; i
++) {
2250 boolean_t printed
= B_FALSE
;
2252 if (nvlist_exists(events
[i
], ZPOOL_HIST_TIME
)) {
2256 tsec
= fnvlist_lookup_uint64(events
[i
],
2258 (void) localtime_r(&tsec
, &t
);
2259 (void) strftime(tbuf
, sizeof (tbuf
), "%F.%T", &t
);
2264 if (nvlist_exists(events
[i
], ZPOOL_HIST_CMD
)) {
2265 (void) printf("%s %s\n", tbuf
,
2266 fnvlist_lookup_string(events
[i
], ZPOOL_HIST_CMD
));
2267 } else if (nvlist_exists(events
[i
], ZPOOL_HIST_INT_EVENT
)) {
2270 ievent
= fnvlist_lookup_uint64(events
[i
],
2271 ZPOOL_HIST_INT_EVENT
);
2272 if (ievent
>= ZFS_NUM_LEGACY_HISTORY_EVENTS
)
2275 (void) printf(" %s [internal %s txg:%ju] %s\n",
2277 zfs_history_event_names
[ievent
],
2278 fnvlist_lookup_uint64(events
[i
],
2280 fnvlist_lookup_string(events
[i
],
2281 ZPOOL_HIST_INT_STR
));
2282 } else if (nvlist_exists(events
[i
], ZPOOL_HIST_INT_NAME
)) {
2283 (void) printf("%s [txg:%ju] %s", tbuf
,
2284 fnvlist_lookup_uint64(events
[i
],
2286 fnvlist_lookup_string(events
[i
],
2287 ZPOOL_HIST_INT_NAME
));
2289 if (nvlist_exists(events
[i
], ZPOOL_HIST_DSNAME
)) {
2290 (void) printf(" %s (%llu)",
2291 fnvlist_lookup_string(events
[i
],
2293 (u_longlong_t
)fnvlist_lookup_uint64(
2298 (void) printf(" %s\n", fnvlist_lookup_string(events
[i
],
2299 ZPOOL_HIST_INT_STR
));
2300 } else if (nvlist_exists(events
[i
], ZPOOL_HIST_IOCTL
)) {
2301 (void) printf("%s ioctl %s\n", tbuf
,
2302 fnvlist_lookup_string(events
[i
],
2305 if (nvlist_exists(events
[i
], ZPOOL_HIST_INPUT_NVL
)) {
2306 (void) printf(" input:\n");
2307 dump_nvlist(fnvlist_lookup_nvlist(events
[i
],
2308 ZPOOL_HIST_INPUT_NVL
), 8);
2310 if (nvlist_exists(events
[i
], ZPOOL_HIST_OUTPUT_NVL
)) {
2311 (void) printf(" output:\n");
2312 dump_nvlist(fnvlist_lookup_nvlist(events
[i
],
2313 ZPOOL_HIST_OUTPUT_NVL
), 8);
2315 if (nvlist_exists(events
[i
], ZPOOL_HIST_ERRNO
)) {
2316 (void) printf(" errno: %lld\n",
2317 (longlong_t
)fnvlist_lookup_int64(events
[i
],
2326 if (dump_opt
['h'] > 1) {
2328 (void) printf("unrecognized record:\n");
2329 dump_nvlist(events
[i
], 2);
2336 dump_dnode(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
2338 (void) os
, (void) object
, (void) data
, (void) size
;
2342 blkid2offset(const dnode_phys_t
*dnp
, const blkptr_t
*bp
,
2343 const zbookmark_phys_t
*zb
)
2346 ASSERT(zb
->zb_level
< 0);
2347 if (zb
->zb_object
== 0)
2348 return (zb
->zb_blkid
);
2349 return (zb
->zb_blkid
* BP_GET_LSIZE(bp
));
2352 ASSERT(zb
->zb_level
>= 0);
2354 return ((zb
->zb_blkid
<<
2355 (zb
->zb_level
* (dnp
->dn_indblkshift
- SPA_BLKPTRSHIFT
))) *
2356 dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
2360 snprintf_zstd_header(spa_t
*spa
, char *blkbuf
, size_t buflen
,
2366 zfs_zstdhdr_t zstd_hdr
;
2369 if (BP_GET_COMPRESS(bp
) != ZIO_COMPRESS_ZSTD
)
2375 if (BP_IS_EMBEDDED(bp
)) {
2376 buf
= malloc(SPA_MAXBLOCKSIZE
);
2378 (void) fprintf(stderr
, "out of memory\n");
2381 decode_embedded_bp_compressed(bp
, buf
);
2382 memcpy(&zstd_hdr
, buf
, sizeof (zstd_hdr
));
2384 zstd_hdr
.c_len
= BE_32(zstd_hdr
.c_len
);
2385 zstd_hdr
.raw_version_level
= BE_32(zstd_hdr
.raw_version_level
);
2386 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2387 buflen
- strlen(blkbuf
),
2388 " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2389 zstd_hdr
.c_len
, zfs_get_hdrversion(&zstd_hdr
),
2390 zfs_get_hdrlevel(&zstd_hdr
));
2394 pabd
= abd_alloc_for_io(SPA_MAXBLOCKSIZE
, B_FALSE
);
2395 zio
= zio_root(spa
, NULL
, NULL
, 0);
2397 /* Decrypt but don't decompress so we can read the compression header */
2398 zio_nowait(zio_read(zio
, spa
, bp
, pabd
, BP_GET_PSIZE(bp
), NULL
, NULL
,
2399 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW_COMPRESS
,
2401 error
= zio_wait(zio
);
2403 (void) fprintf(stderr
, "read failed: %d\n", error
);
2406 buf
= abd_borrow_buf_copy(pabd
, BP_GET_LSIZE(bp
));
2407 memcpy(&zstd_hdr
, buf
, sizeof (zstd_hdr
));
2408 zstd_hdr
.c_len
= BE_32(zstd_hdr
.c_len
);
2409 zstd_hdr
.raw_version_level
= BE_32(zstd_hdr
.raw_version_level
);
2411 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2412 buflen
- strlen(blkbuf
),
2413 " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2414 zstd_hdr
.c_len
, zfs_get_hdrversion(&zstd_hdr
),
2415 zfs_get_hdrlevel(&zstd_hdr
));
2417 abd_return_buf_copy(pabd
, buf
, BP_GET_LSIZE(bp
));
2421 snprintf_blkptr_compact(char *blkbuf
, size_t buflen
, const blkptr_t
*bp
,
2424 const dva_t
*dva
= bp
->blk_dva
;
2425 int ndvas
= dump_opt
['d'] > 5 ? BP_GET_NDVAS(bp
) : 1;
2428 if (dump_opt
['b'] >= 6) {
2429 snprintf_blkptr(blkbuf
, buflen
, bp
);
2431 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2432 buflen
- strlen(blkbuf
), " %s", "FREE");
2437 if (BP_IS_EMBEDDED(bp
)) {
2438 (void) sprintf(blkbuf
,
2439 "EMBEDDED et=%u %llxL/%llxP B=%llu",
2440 (int)BPE_GET_ETYPE(bp
),
2441 (u_longlong_t
)BPE_GET_LSIZE(bp
),
2442 (u_longlong_t
)BPE_GET_PSIZE(bp
),
2443 (u_longlong_t
)bp
->blk_birth
);
2449 for (i
= 0; i
< ndvas
; i
++)
2450 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2451 buflen
- strlen(blkbuf
), "%llu:%llx:%llx ",
2452 (u_longlong_t
)DVA_GET_VDEV(&dva
[i
]),
2453 (u_longlong_t
)DVA_GET_OFFSET(&dva
[i
]),
2454 (u_longlong_t
)DVA_GET_ASIZE(&dva
[i
]));
2456 if (BP_IS_HOLE(bp
)) {
2457 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2458 buflen
- strlen(blkbuf
),
2460 (u_longlong_t
)BP_GET_LSIZE(bp
),
2461 (u_longlong_t
)bp
->blk_birth
);
2463 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2464 buflen
- strlen(blkbuf
),
2465 "%llxL/%llxP F=%llu B=%llu/%llu",
2466 (u_longlong_t
)BP_GET_LSIZE(bp
),
2467 (u_longlong_t
)BP_GET_PSIZE(bp
),
2468 (u_longlong_t
)BP_GET_FILL(bp
),
2469 (u_longlong_t
)bp
->blk_birth
,
2470 (u_longlong_t
)BP_PHYSICAL_BIRTH(bp
));
2472 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2473 buflen
- strlen(blkbuf
), " %s", "FREE");
2474 (void) snprintf(blkbuf
+ strlen(blkbuf
),
2475 buflen
- strlen(blkbuf
),
2476 " cksum=%016llx:%016llx:%016llx:%016llx",
2477 (u_longlong_t
)bp
->blk_cksum
.zc_word
[0],
2478 (u_longlong_t
)bp
->blk_cksum
.zc_word
[1],
2479 (u_longlong_t
)bp
->blk_cksum
.zc_word
[2],
2480 (u_longlong_t
)bp
->blk_cksum
.zc_word
[3]);
2485 print_indirect(spa_t
*spa
, blkptr_t
*bp
, const zbookmark_phys_t
*zb
,
2486 const dnode_phys_t
*dnp
)
2488 char blkbuf
[BP_SPRINTF_LEN
];
2491 if (!BP_IS_EMBEDDED(bp
)) {
2492 ASSERT3U(BP_GET_TYPE(bp
), ==, dnp
->dn_type
);
2493 ASSERT3U(BP_GET_LEVEL(bp
), ==, zb
->zb_level
);
2496 (void) printf("%16llx ", (u_longlong_t
)blkid2offset(dnp
, bp
, zb
));
2498 ASSERT(zb
->zb_level
>= 0);
2500 for (l
= dnp
->dn_nlevels
- 1; l
>= -1; l
--) {
2501 if (l
== zb
->zb_level
) {
2502 (void) printf("L%llx", (u_longlong_t
)zb
->zb_level
);
2508 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
), bp
, B_FALSE
);
2509 if (dump_opt
['Z'] && BP_GET_COMPRESS(bp
) == ZIO_COMPRESS_ZSTD
)
2510 snprintf_zstd_header(spa
, blkbuf
, sizeof (blkbuf
), bp
);
2511 (void) printf("%s\n", blkbuf
);
2515 visit_indirect(spa_t
*spa
, const dnode_phys_t
*dnp
,
2516 blkptr_t
*bp
, const zbookmark_phys_t
*zb
)
2520 if (bp
->blk_birth
== 0)
2523 print_indirect(spa
, bp
, zb
, dnp
);
2525 if (BP_GET_LEVEL(bp
) > 0 && !BP_IS_HOLE(bp
)) {
2526 arc_flags_t flags
= ARC_FLAG_WAIT
;
2529 int epb
= BP_GET_LSIZE(bp
) >> SPA_BLKPTRSHIFT
;
2532 ASSERT(!BP_IS_REDACTED(bp
));
2534 err
= arc_read(NULL
, spa
, bp
, arc_getbuf_func
, &buf
,
2535 ZIO_PRIORITY_ASYNC_READ
, ZIO_FLAG_CANFAIL
, &flags
, zb
);
2538 ASSERT(buf
->b_data
);
2540 /* recursively visit blocks below this */
2542 for (i
= 0; i
< epb
; i
++, cbp
++) {
2543 zbookmark_phys_t czb
;
2545 SET_BOOKMARK(&czb
, zb
->zb_objset
, zb
->zb_object
,
2547 zb
->zb_blkid
* epb
+ i
);
2548 err
= visit_indirect(spa
, dnp
, cbp
, &czb
);
2551 fill
+= BP_GET_FILL(cbp
);
2554 ASSERT3U(fill
, ==, BP_GET_FILL(bp
));
2555 arc_buf_destroy(buf
, &buf
);
2562 dump_indirect(dnode_t
*dn
)
2564 dnode_phys_t
*dnp
= dn
->dn_phys
;
2565 zbookmark_phys_t czb
;
2567 (void) printf("Indirect blocks:\n");
2569 SET_BOOKMARK(&czb
, dmu_objset_id(dn
->dn_objset
),
2570 dn
->dn_object
, dnp
->dn_nlevels
- 1, 0);
2571 for (int j
= 0; j
< dnp
->dn_nblkptr
; j
++) {
2573 (void) visit_indirect(dmu_objset_spa(dn
->dn_objset
), dnp
,
2574 &dnp
->dn_blkptr
[j
], &czb
);
2577 (void) printf("\n");
2581 dump_dsl_dir(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
2583 (void) os
, (void) object
;
2584 dsl_dir_phys_t
*dd
= data
;
2588 /* make sure nicenum has enough space */
2589 _Static_assert(sizeof (nice
) >= NN_NUMBUF_SZ
, "nice truncated");
2594 ASSERT3U(size
, >=, sizeof (dsl_dir_phys_t
));
2596 crtime
= dd
->dd_creation_time
;
2597 (void) printf("\t\tcreation_time = %s", ctime(&crtime
));
2598 (void) printf("\t\thead_dataset_obj = %llu\n",
2599 (u_longlong_t
)dd
->dd_head_dataset_obj
);
2600 (void) printf("\t\tparent_dir_obj = %llu\n",
2601 (u_longlong_t
)dd
->dd_parent_obj
);
2602 (void) printf("\t\torigin_obj = %llu\n",
2603 (u_longlong_t
)dd
->dd_origin_obj
);
2604 (void) printf("\t\tchild_dir_zapobj = %llu\n",
2605 (u_longlong_t
)dd
->dd_child_dir_zapobj
);
2606 zdb_nicenum(dd
->dd_used_bytes
, nice
, sizeof (nice
));
2607 (void) printf("\t\tused_bytes = %s\n", nice
);
2608 zdb_nicenum(dd
->dd_compressed_bytes
, nice
, sizeof (nice
));
2609 (void) printf("\t\tcompressed_bytes = %s\n", nice
);
2610 zdb_nicenum(dd
->dd_uncompressed_bytes
, nice
, sizeof (nice
));
2611 (void) printf("\t\tuncompressed_bytes = %s\n", nice
);
2612 zdb_nicenum(dd
->dd_quota
, nice
, sizeof (nice
));
2613 (void) printf("\t\tquota = %s\n", nice
);
2614 zdb_nicenum(dd
->dd_reserved
, nice
, sizeof (nice
));
2615 (void) printf("\t\treserved = %s\n", nice
);
2616 (void) printf("\t\tprops_zapobj = %llu\n",
2617 (u_longlong_t
)dd
->dd_props_zapobj
);
2618 (void) printf("\t\tdeleg_zapobj = %llu\n",
2619 (u_longlong_t
)dd
->dd_deleg_zapobj
);
2620 (void) printf("\t\tflags = %llx\n",
2621 (u_longlong_t
)dd
->dd_flags
);
2624 zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2626 (void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2633 (void) printf("\t\tclones = %llu\n",
2634 (u_longlong_t
)dd
->dd_clones
);
2638 dump_dsl_dataset(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
2640 (void) os
, (void) object
;
2641 dsl_dataset_phys_t
*ds
= data
;
2643 char used
[32], compressed
[32], uncompressed
[32], unique
[32];
2644 char blkbuf
[BP_SPRINTF_LEN
];
2646 /* make sure nicenum has enough space */
2647 _Static_assert(sizeof (used
) >= NN_NUMBUF_SZ
, "used truncated");
2648 _Static_assert(sizeof (compressed
) >= NN_NUMBUF_SZ
,
2649 "compressed truncated");
2650 _Static_assert(sizeof (uncompressed
) >= NN_NUMBUF_SZ
,
2651 "uncompressed truncated");
2652 _Static_assert(sizeof (unique
) >= NN_NUMBUF_SZ
, "unique truncated");
2657 ASSERT(size
== sizeof (*ds
));
2658 crtime
= ds
->ds_creation_time
;
2659 zdb_nicenum(ds
->ds_referenced_bytes
, used
, sizeof (used
));
2660 zdb_nicenum(ds
->ds_compressed_bytes
, compressed
, sizeof (compressed
));
2661 zdb_nicenum(ds
->ds_uncompressed_bytes
, uncompressed
,
2662 sizeof (uncompressed
));
2663 zdb_nicenum(ds
->ds_unique_bytes
, unique
, sizeof (unique
));
2664 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), &ds
->ds_bp
);
2666 (void) printf("\t\tdir_obj = %llu\n",
2667 (u_longlong_t
)ds
->ds_dir_obj
);
2668 (void) printf("\t\tprev_snap_obj = %llu\n",
2669 (u_longlong_t
)ds
->ds_prev_snap_obj
);
2670 (void) printf("\t\tprev_snap_txg = %llu\n",
2671 (u_longlong_t
)ds
->ds_prev_snap_txg
);
2672 (void) printf("\t\tnext_snap_obj = %llu\n",
2673 (u_longlong_t
)ds
->ds_next_snap_obj
);
2674 (void) printf("\t\tsnapnames_zapobj = %llu\n",
2675 (u_longlong_t
)ds
->ds_snapnames_zapobj
);
2676 (void) printf("\t\tnum_children = %llu\n",
2677 (u_longlong_t
)ds
->ds_num_children
);
2678 (void) printf("\t\tuserrefs_obj = %llu\n",
2679 (u_longlong_t
)ds
->ds_userrefs_obj
);
2680 (void) printf("\t\tcreation_time = %s", ctime(&crtime
));
2681 (void) printf("\t\tcreation_txg = %llu\n",
2682 (u_longlong_t
)ds
->ds_creation_txg
);
2683 (void) printf("\t\tdeadlist_obj = %llu\n",
2684 (u_longlong_t
)ds
->ds_deadlist_obj
);
2685 (void) printf("\t\tused_bytes = %s\n", used
);
2686 (void) printf("\t\tcompressed_bytes = %s\n", compressed
);
2687 (void) printf("\t\tuncompressed_bytes = %s\n", uncompressed
);
2688 (void) printf("\t\tunique = %s\n", unique
);
2689 (void) printf("\t\tfsid_guid = %llu\n",
2690 (u_longlong_t
)ds
->ds_fsid_guid
);
2691 (void) printf("\t\tguid = %llu\n",
2692 (u_longlong_t
)ds
->ds_guid
);
2693 (void) printf("\t\tflags = %llx\n",
2694 (u_longlong_t
)ds
->ds_flags
);
2695 (void) printf("\t\tnext_clones_obj = %llu\n",
2696 (u_longlong_t
)ds
->ds_next_clones_obj
);
2697 (void) printf("\t\tprops_obj = %llu\n",
2698 (u_longlong_t
)ds
->ds_props_obj
);
2699 (void) printf("\t\tbp = %s\n", blkbuf
);
2703 dump_bptree_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
2705 (void) arg
, (void) tx
;
2706 char blkbuf
[BP_SPRINTF_LEN
];
2708 if (bp
->blk_birth
!= 0) {
2709 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
2710 (void) printf("\t%s\n", blkbuf
);
2716 dump_bptree(objset_t
*os
, uint64_t obj
, const char *name
)
2722 /* make sure nicenum has enough space */
2723 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
2725 if (dump_opt
['d'] < 3)
2728 VERIFY3U(0, ==, dmu_bonus_hold(os
, obj
, FTAG
, &db
));
2730 zdb_nicenum(bt
->bt_bytes
, bytes
, sizeof (bytes
));
2731 (void) printf("\n %s: %llu datasets, %s\n",
2732 name
, (unsigned long long)(bt
->bt_end
- bt
->bt_begin
), bytes
);
2733 dmu_buf_rele(db
, FTAG
);
2735 if (dump_opt
['d'] < 5)
2738 (void) printf("\n");
2740 (void) bptree_iterate(os
, obj
, B_FALSE
, dump_bptree_cb
, NULL
, NULL
);
2744 dump_bpobj_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
, dmu_tx_t
*tx
)
2746 (void) arg
, (void) tx
;
2747 char blkbuf
[BP_SPRINTF_LEN
];
2749 ASSERT(bp
->blk_birth
!= 0);
2750 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
), bp
, bp_freed
);
2751 (void) printf("\t%s\n", blkbuf
);
2756 dump_full_bpobj(bpobj_t
*bpo
, const char *name
, int indent
)
2763 /* make sure nicenum has enough space */
2764 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
2765 _Static_assert(sizeof (comp
) >= NN_NUMBUF_SZ
, "comp truncated");
2766 _Static_assert(sizeof (uncomp
) >= NN_NUMBUF_SZ
, "uncomp truncated");
2768 if (dump_opt
['d'] < 3)
2771 zdb_nicenum(bpo
->bpo_phys
->bpo_bytes
, bytes
, sizeof (bytes
));
2772 if (bpo
->bpo_havesubobj
&& bpo
->bpo_phys
->bpo_subobjs
!= 0) {
2773 zdb_nicenum(bpo
->bpo_phys
->bpo_comp
, comp
, sizeof (comp
));
2774 zdb_nicenum(bpo
->bpo_phys
->bpo_uncomp
, uncomp
, sizeof (uncomp
));
2775 if (bpo
->bpo_havefreed
) {
2776 (void) printf(" %*s: object %llu, %llu local "
2777 "blkptrs, %llu freed, %llu subobjs in object %llu, "
2778 "%s (%s/%s comp)\n",
2780 (u_longlong_t
)bpo
->bpo_object
,
2781 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2782 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_freed
,
2783 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_subobjs
,
2784 (u_longlong_t
)bpo
->bpo_phys
->bpo_subobjs
,
2785 bytes
, comp
, uncomp
);
2787 (void) printf(" %*s: object %llu, %llu local "
2788 "blkptrs, %llu subobjs in object %llu, "
2789 "%s (%s/%s comp)\n",
2791 (u_longlong_t
)bpo
->bpo_object
,
2792 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2793 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_subobjs
,
2794 (u_longlong_t
)bpo
->bpo_phys
->bpo_subobjs
,
2795 bytes
, comp
, uncomp
);
2798 for (i
= 0; i
< bpo
->bpo_phys
->bpo_num_subobjs
; i
++) {
2802 VERIFY0(dmu_read(bpo
->bpo_os
,
2803 bpo
->bpo_phys
->bpo_subobjs
,
2804 i
* sizeof (subobj
), sizeof (subobj
), &subobj
, 0));
2805 error
= bpobj_open(&subbpo
, bpo
->bpo_os
, subobj
);
2807 (void) printf("ERROR %u while trying to open "
2809 error
, (u_longlong_t
)subobj
);
2812 dump_full_bpobj(&subbpo
, "subobj", indent
+ 1);
2813 bpobj_close(&subbpo
);
2816 if (bpo
->bpo_havefreed
) {
2817 (void) printf(" %*s: object %llu, %llu blkptrs, "
2820 (u_longlong_t
)bpo
->bpo_object
,
2821 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2822 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_freed
,
2825 (void) printf(" %*s: object %llu, %llu blkptrs, "
2828 (u_longlong_t
)bpo
->bpo_object
,
2829 (u_longlong_t
)bpo
->bpo_phys
->bpo_num_blkptrs
,
2834 if (dump_opt
['d'] < 5)
2839 (void) bpobj_iterate_nofree(bpo
, dump_bpobj_cb
, NULL
, NULL
);
2840 (void) printf("\n");
2845 dump_bookmark(dsl_pool_t
*dp
, char *name
, boolean_t print_redact
,
2846 boolean_t print_list
)
2849 zfs_bookmark_phys_t prop
;
2850 objset_t
*mos
= dp
->dp_spa
->spa_meta_objset
;
2851 err
= dsl_bookmark_lookup(dp
, name
, NULL
, &prop
);
2857 (void) printf("\t#%s: ", strchr(name
, '#') + 1);
2858 (void) printf("{guid: %llx creation_txg: %llu creation_time: "
2859 "%llu redaction_obj: %llu}\n", (u_longlong_t
)prop
.zbm_guid
,
2860 (u_longlong_t
)prop
.zbm_creation_txg
,
2861 (u_longlong_t
)prop
.zbm_creation_time
,
2862 (u_longlong_t
)prop
.zbm_redaction_obj
);
2864 IMPLY(print_list
, print_redact
);
2865 if (!print_redact
|| prop
.zbm_redaction_obj
== 0)
2868 redaction_list_t
*rl
;
2869 VERIFY0(dsl_redaction_list_hold_obj(dp
,
2870 prop
.zbm_redaction_obj
, FTAG
, &rl
));
2872 redaction_list_phys_t
*rlp
= rl
->rl_phys
;
2873 (void) printf("\tRedacted:\n\t\tProgress: ");
2874 if (rlp
->rlp_last_object
!= UINT64_MAX
||
2875 rlp
->rlp_last_blkid
!= UINT64_MAX
) {
2876 (void) printf("%llu %llu (incomplete)\n",
2877 (u_longlong_t
)rlp
->rlp_last_object
,
2878 (u_longlong_t
)rlp
->rlp_last_blkid
);
2880 (void) printf("complete\n");
2882 (void) printf("\t\tSnapshots: [");
2883 for (unsigned int i
= 0; i
< rlp
->rlp_num_snaps
; i
++) {
2885 (void) printf(", ");
2886 (void) printf("%0llu",
2887 (u_longlong_t
)rlp
->rlp_snaps
[i
]);
2889 (void) printf("]\n\t\tLength: %llu\n",
2890 (u_longlong_t
)rlp
->rlp_num_entries
);
2893 dsl_redaction_list_rele(rl
, FTAG
);
2897 if (rlp
->rlp_num_entries
== 0) {
2898 dsl_redaction_list_rele(rl
, FTAG
);
2899 (void) printf("\t\tRedaction List: []\n\n");
2903 redact_block_phys_t
*rbp_buf
;
2905 dmu_object_info_t doi
;
2907 VERIFY0(dmu_object_info(mos
, prop
.zbm_redaction_obj
, &doi
));
2908 size
= doi
.doi_max_offset
;
2909 rbp_buf
= kmem_alloc(size
, KM_SLEEP
);
2911 err
= dmu_read(mos
, prop
.zbm_redaction_obj
, 0, size
,
2914 dsl_redaction_list_rele(rl
, FTAG
);
2915 kmem_free(rbp_buf
, size
);
2919 (void) printf("\t\tRedaction List: [{object: %llx, offset: "
2920 "%llx, blksz: %x, count: %llx}",
2921 (u_longlong_t
)rbp_buf
[0].rbp_object
,
2922 (u_longlong_t
)rbp_buf
[0].rbp_blkid
,
2923 (uint_t
)(redact_block_get_size(&rbp_buf
[0])),
2924 (u_longlong_t
)redact_block_get_count(&rbp_buf
[0]));
2926 for (size_t i
= 1; i
< rlp
->rlp_num_entries
; i
++) {
2927 (void) printf(",\n\t\t{object: %llx, offset: %llx, "
2928 "blksz: %x, count: %llx}",
2929 (u_longlong_t
)rbp_buf
[i
].rbp_object
,
2930 (u_longlong_t
)rbp_buf
[i
].rbp_blkid
,
2931 (uint_t
)(redact_block_get_size(&rbp_buf
[i
])),
2932 (u_longlong_t
)redact_block_get_count(&rbp_buf
[i
]));
2934 dsl_redaction_list_rele(rl
, FTAG
);
2935 kmem_free(rbp_buf
, size
);
2936 (void) printf("]\n\n");
2941 dump_bookmarks(objset_t
*os
, int verbosity
)
2944 zap_attribute_t attr
;
2945 dsl_dataset_t
*ds
= dmu_objset_ds(os
);
2946 dsl_pool_t
*dp
= spa_get_dsl(os
->os_spa
);
2947 objset_t
*mos
= os
->os_spa
->spa_meta_objset
;
2950 dsl_pool_config_enter(dp
, FTAG
);
2952 for (zap_cursor_init(&zc
, mos
, ds
->ds_bookmarks_obj
);
2953 zap_cursor_retrieve(&zc
, &attr
) == 0;
2954 zap_cursor_advance(&zc
)) {
2955 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
2956 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
2958 dmu_objset_name(os
, osname
);
2959 len
= snprintf(buf
, sizeof (buf
), "%s#%s", osname
,
2961 VERIFY3S(len
, <, ZFS_MAX_DATASET_NAME_LEN
);
2962 (void) dump_bookmark(dp
, buf
, verbosity
>= 5, verbosity
>= 6);
2964 zap_cursor_fini(&zc
);
2965 dsl_pool_config_exit(dp
, FTAG
);
2969 bpobj_count_refd(bpobj_t
*bpo
)
2971 mos_obj_refd(bpo
->bpo_object
);
2973 if (bpo
->bpo_havesubobj
&& bpo
->bpo_phys
->bpo_subobjs
!= 0) {
2974 mos_obj_refd(bpo
->bpo_phys
->bpo_subobjs
);
2975 for (uint64_t i
= 0; i
< bpo
->bpo_phys
->bpo_num_subobjs
; i
++) {
2979 VERIFY0(dmu_read(bpo
->bpo_os
,
2980 bpo
->bpo_phys
->bpo_subobjs
,
2981 i
* sizeof (subobj
), sizeof (subobj
), &subobj
, 0));
2982 error
= bpobj_open(&subbpo
, bpo
->bpo_os
, subobj
);
2984 (void) printf("ERROR %u while trying to open "
2986 error
, (u_longlong_t
)subobj
);
2989 bpobj_count_refd(&subbpo
);
2990 bpobj_close(&subbpo
);
2996 dsl_deadlist_entry_count_refd(void *arg
, dsl_deadlist_entry_t
*dle
)
2999 uint64_t empty_bpobj
= spa
->spa_dsl_pool
->dp_empty_bpobj
;
3000 if (dle
->dle_bpobj
.bpo_object
!= empty_bpobj
)
3001 bpobj_count_refd(&dle
->dle_bpobj
);
3006 dsl_deadlist_entry_dump(void *arg
, dsl_deadlist_entry_t
*dle
)
3008 ASSERT(arg
== NULL
);
3009 if (dump_opt
['d'] >= 5) {
3011 (void) snprintf(buf
, sizeof (buf
),
3012 "mintxg %llu -> obj %llu",
3013 (longlong_t
)dle
->dle_mintxg
,
3014 (longlong_t
)dle
->dle_bpobj
.bpo_object
);
3016 dump_full_bpobj(&dle
->dle_bpobj
, buf
, 0);
3018 (void) printf("mintxg %llu -> obj %llu\n",
3019 (longlong_t
)dle
->dle_mintxg
,
3020 (longlong_t
)dle
->dle_bpobj
.bpo_object
);
3026 dump_blkptr_list(dsl_deadlist_t
*dl
, const char *name
)
3032 spa_t
*spa
= dmu_objset_spa(dl
->dl_os
);
3033 uint64_t empty_bpobj
= spa
->spa_dsl_pool
->dp_empty_bpobj
;
3035 if (dl
->dl_oldfmt
) {
3036 if (dl
->dl_bpobj
.bpo_object
!= empty_bpobj
)
3037 bpobj_count_refd(&dl
->dl_bpobj
);
3039 mos_obj_refd(dl
->dl_object
);
3040 dsl_deadlist_iterate(dl
, dsl_deadlist_entry_count_refd
, spa
);
3043 /* make sure nicenum has enough space */
3044 _Static_assert(sizeof (bytes
) >= NN_NUMBUF_SZ
, "bytes truncated");
3045 _Static_assert(sizeof (comp
) >= NN_NUMBUF_SZ
, "comp truncated");
3046 _Static_assert(sizeof (uncomp
) >= NN_NUMBUF_SZ
, "uncomp truncated");
3047 _Static_assert(sizeof (entries
) >= NN_NUMBUF_SZ
, "entries truncated");
3049 if (dump_opt
['d'] < 3)
3052 if (dl
->dl_oldfmt
) {
3053 dump_full_bpobj(&dl
->dl_bpobj
, "old-format deadlist", 0);
3057 zdb_nicenum(dl
->dl_phys
->dl_used
, bytes
, sizeof (bytes
));
3058 zdb_nicenum(dl
->dl_phys
->dl_comp
, comp
, sizeof (comp
));
3059 zdb_nicenum(dl
->dl_phys
->dl_uncomp
, uncomp
, sizeof (uncomp
));
3060 zdb_nicenum(avl_numnodes(&dl
->dl_tree
), entries
, sizeof (entries
));
3061 (void) printf("\n %s: %s (%s/%s comp), %s entries\n",
3062 name
, bytes
, comp
, uncomp
, entries
);
3064 if (dump_opt
['d'] < 4)
3067 (void) putchar('\n');
3069 dsl_deadlist_iterate(dl
, dsl_deadlist_entry_dump
, NULL
);
3073 verify_dd_livelist(objset_t
*os
)
3075 uint64_t ll_used
, used
, ll_comp
, comp
, ll_uncomp
, uncomp
;
3076 dsl_pool_t
*dp
= spa_get_dsl(os
->os_spa
);
3077 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
3079 ASSERT(!dmu_objset_is_snapshot(os
));
3080 if (!dsl_deadlist_is_open(&dd
->dd_livelist
))
3083 /* Iterate through the livelist to check for duplicates */
3084 dsl_deadlist_iterate(&dd
->dd_livelist
, sublivelist_verify_lightweight
,
3087 dsl_pool_config_enter(dp
, FTAG
);
3088 dsl_deadlist_space(&dd
->dd_livelist
, &ll_used
,
3089 &ll_comp
, &ll_uncomp
);
3091 dsl_dataset_t
*origin_ds
;
3092 ASSERT(dsl_pool_config_held(dp
));
3093 VERIFY0(dsl_dataset_hold_obj(dp
,
3094 dsl_dir_phys(dd
)->dd_origin_obj
, FTAG
, &origin_ds
));
3095 VERIFY0(dsl_dataset_space_written(origin_ds
, os
->os_dsl_dataset
,
3096 &used
, &comp
, &uncomp
));
3097 dsl_dataset_rele(origin_ds
, FTAG
);
3098 dsl_pool_config_exit(dp
, FTAG
);
3100 * It's possible that the dataset's uncomp space is larger than the
3101 * livelist's because livelists do not track embedded block pointers
3103 if (used
!= ll_used
|| comp
!= ll_comp
|| uncomp
< ll_uncomp
) {
3104 char nice_used
[32], nice_comp
[32], nice_uncomp
[32];
3105 (void) printf("Discrepancy in space accounting:\n");
3106 zdb_nicenum(used
, nice_used
, sizeof (nice_used
));
3107 zdb_nicenum(comp
, nice_comp
, sizeof (nice_comp
));
3108 zdb_nicenum(uncomp
, nice_uncomp
, sizeof (nice_uncomp
));
3109 (void) printf("dir: used %s, comp %s, uncomp %s\n",
3110 nice_used
, nice_comp
, nice_uncomp
);
3111 zdb_nicenum(ll_used
, nice_used
, sizeof (nice_used
));
3112 zdb_nicenum(ll_comp
, nice_comp
, sizeof (nice_comp
));
3113 zdb_nicenum(ll_uncomp
, nice_uncomp
, sizeof (nice_uncomp
));
3114 (void) printf("livelist: used %s, comp %s, uncomp %s\n",
3115 nice_used
, nice_comp
, nice_uncomp
);
3121 static char *key_material
= NULL
;
3124 zdb_derive_key(dsl_dir_t
*dd
, uint8_t *key_out
)
3126 uint64_t keyformat
, salt
, iters
;
3130 VERIFY0(zap_lookup(dd
->dd_pool
->dp_meta_objset
, dd
->dd_crypto_obj
,
3131 zfs_prop_to_name(ZFS_PROP_KEYFORMAT
), sizeof (uint64_t),
3134 switch (keyformat
) {
3135 case ZFS_KEYFORMAT_HEX
:
3136 for (i
= 0; i
< WRAPPING_KEY_LEN
* 2; i
+= 2) {
3137 if (!isxdigit(key_material
[i
]) ||
3138 !isxdigit(key_material
[i
+1]))
3140 if (sscanf(&key_material
[i
], "%02hhx", &c
) != 1)
3146 case ZFS_KEYFORMAT_PASSPHRASE
:
3147 VERIFY0(zap_lookup(dd
->dd_pool
->dp_meta_objset
,
3148 dd
->dd_crypto_obj
, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT
),
3149 sizeof (uint64_t), 1, &salt
));
3150 VERIFY0(zap_lookup(dd
->dd_pool
->dp_meta_objset
,
3151 dd
->dd_crypto_obj
, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS
),
3152 sizeof (uint64_t), 1, &iters
));
3154 if (PKCS5_PBKDF2_HMAC_SHA1(key_material
, strlen(key_material
),
3155 ((uint8_t *)&salt
), sizeof (uint64_t), iters
,
3156 WRAPPING_KEY_LEN
, key_out
) != 1)
3162 fatal("no support for key format %u\n",
3163 (unsigned int) keyformat
);
3169 static char encroot
[ZFS_MAX_DATASET_NAME_LEN
];
3170 static boolean_t key_loaded
= B_FALSE
;
3173 zdb_load_key(objset_t
*os
)
3176 dsl_dir_t
*dd
, *rdd
;
3177 uint8_t key
[WRAPPING_KEY_LEN
];
3181 dp
= spa_get_dsl(os
->os_spa
);
3182 dd
= os
->os_dsl_dataset
->ds_dir
;
3184 dsl_pool_config_enter(dp
, FTAG
);
3185 VERIFY0(zap_lookup(dd
->dd_pool
->dp_meta_objset
, dd
->dd_crypto_obj
,
3186 DSL_CRYPTO_KEY_ROOT_DDOBJ
, sizeof (uint64_t), 1, &rddobj
));
3187 VERIFY0(dsl_dir_hold_obj(dd
->dd_pool
, rddobj
, NULL
, FTAG
, &rdd
));
3188 dsl_dir_name(rdd
, encroot
);
3189 dsl_dir_rele(rdd
, FTAG
);
3191 if (!zdb_derive_key(dd
, key
))
3192 fatal("couldn't derive encryption key");
3194 dsl_pool_config_exit(dp
, FTAG
);
3196 ASSERT3U(dsl_dataset_get_keystatus(dd
), ==, ZFS_KEYSTATUS_UNAVAILABLE
);
3198 dsl_crypto_params_t
*dcp
;
3199 nvlist_t
*crypto_args
;
3201 crypto_args
= fnvlist_alloc();
3202 fnvlist_add_uint8_array(crypto_args
, "wkeydata",
3203 (uint8_t *)key
, WRAPPING_KEY_LEN
);
3204 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE
,
3205 NULL
, crypto_args
, &dcp
));
3206 err
= spa_keystore_load_wkey(encroot
, dcp
, B_FALSE
);
3208 dsl_crypto_params_free(dcp
, (err
!= 0));
3209 fnvlist_free(crypto_args
);
3213 "couldn't load encryption key for %s: %s",
3214 encroot
, err
== ZFS_ERR_CRYPTO_NOTSUP
?
3215 "crypto params not supported" : strerror(err
));
3217 ASSERT3U(dsl_dataset_get_keystatus(dd
), ==, ZFS_KEYSTATUS_AVAILABLE
);
3219 printf("Unlocked encryption root: %s\n", encroot
);
3220 key_loaded
= B_TRUE
;
3224 zdb_unload_key(void)
3229 VERIFY0(spa_keystore_unload_wkey(encroot
));
3230 key_loaded
= B_FALSE
;
3233 static avl_tree_t idx_tree
;
3234 static avl_tree_t domain_tree
;
3235 static boolean_t fuid_table_loaded
;
3236 static objset_t
*sa_os
= NULL
;
3237 static sa_attr_type_t
*sa_attr_table
= NULL
;
3240 open_objset(const char *path
, const void *tag
, objset_t
**osp
)
3243 uint64_t sa_attrs
= 0;
3244 uint64_t version
= 0;
3246 VERIFY3P(sa_os
, ==, NULL
);
3249 * We can't own an objset if it's redacted. Therefore, we do this
3250 * dance: hold the objset, then acquire a long hold on its dataset, then
3251 * release the pool (which is held as part of holding the objset).
3254 if (dump_opt
['K']) {
3255 /* decryption requested, try to load keys */
3256 err
= dmu_objset_hold(path
, tag
, osp
);
3258 (void) fprintf(stderr
, "failed to hold dataset "
3260 path
, strerror(err
));
3263 dsl_dataset_long_hold(dmu_objset_ds(*osp
), tag
);
3264 dsl_pool_rele(dmu_objset_pool(*osp
), tag
);
3266 /* succeeds or dies */
3269 /* release it all */
3270 dsl_dataset_long_rele(dmu_objset_ds(*osp
), tag
);
3271 dsl_dataset_rele(dmu_objset_ds(*osp
), tag
);
3274 int ds_hold_flags
= key_loaded
? DS_HOLD_FLAG_DECRYPT
: 0;
3276 err
= dmu_objset_hold_flags(path
, ds_hold_flags
, tag
, osp
);
3278 (void) fprintf(stderr
, "failed to hold dataset '%s': %s\n",
3279 path
, strerror(err
));
3282 dsl_dataset_long_hold(dmu_objset_ds(*osp
), tag
);
3283 dsl_pool_rele(dmu_objset_pool(*osp
), tag
);
3285 if (dmu_objset_type(*osp
) == DMU_OST_ZFS
&&
3286 (key_loaded
|| !(*osp
)->os_encrypted
)) {
3287 (void) zap_lookup(*osp
, MASTER_NODE_OBJ
, ZPL_VERSION_STR
,
3289 if (version
>= ZPL_VERSION_SA
) {
3290 (void) zap_lookup(*osp
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
,
3293 err
= sa_setup(*osp
, sa_attrs
, zfs_attr_table
, ZPL_END
,
3296 (void) fprintf(stderr
, "sa_setup failed: %s\n",
3298 dsl_dataset_long_rele(dmu_objset_ds(*osp
), tag
);
3299 dsl_dataset_rele_flags(dmu_objset_ds(*osp
),
3300 ds_hold_flags
, tag
);
3310 close_objset(objset_t
*os
, const void *tag
)
3312 VERIFY3P(os
, ==, sa_os
);
3313 if (os
->os_sa
!= NULL
)
3315 dsl_dataset_long_rele(dmu_objset_ds(os
), tag
);
3316 dsl_dataset_rele_flags(dmu_objset_ds(os
),
3317 key_loaded
? DS_HOLD_FLAG_DECRYPT
: 0, tag
);
3318 sa_attr_table
= NULL
;
3325 fuid_table_destroy(void)
3327 if (fuid_table_loaded
) {
3328 zfs_fuid_table_destroy(&idx_tree
, &domain_tree
);
3329 fuid_table_loaded
= B_FALSE
;
3334 * print uid or gid information.
3335 * For normal POSIX id just the id is printed in decimal format.
3336 * For CIFS files with FUID the fuid is printed in hex followed by
3337 * the domain-rid string.
3340 print_idstr(uint64_t id
, const char *id_type
)
3342 if (FUID_INDEX(id
)) {
3343 const char *domain
=
3344 zfs_fuid_idx_domain(&idx_tree
, FUID_INDEX(id
));
3345 (void) printf("\t%s %llx [%s-%d]\n", id_type
,
3346 (u_longlong_t
)id
, domain
, (int)FUID_RID(id
));
3348 (void) printf("\t%s %llu\n", id_type
, (u_longlong_t
)id
);
3354 dump_uidgid(objset_t
*os
, uint64_t uid
, uint64_t gid
)
3356 uint32_t uid_idx
, gid_idx
;
3358 uid_idx
= FUID_INDEX(uid
);
3359 gid_idx
= FUID_INDEX(gid
);
3361 /* Load domain table, if not already loaded */
3362 if (!fuid_table_loaded
&& (uid_idx
|| gid_idx
)) {
3365 /* first find the fuid object. It lives in the master node */
3366 VERIFY(zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_FUID_TABLES
,
3367 8, 1, &fuid_obj
) == 0);
3368 zfs_fuid_avl_tree_create(&idx_tree
, &domain_tree
);
3369 (void) zfs_fuid_table_load(os
, fuid_obj
,
3370 &idx_tree
, &domain_tree
);
3371 fuid_table_loaded
= B_TRUE
;
3374 print_idstr(uid
, "uid");
3375 print_idstr(gid
, "gid");
3379 dump_znode_sa_xattr(sa_handle_t
*hdl
)
3382 nvpair_t
*elem
= NULL
;
3383 int sa_xattr_size
= 0;
3384 int sa_xattr_entries
= 0;
3386 char *sa_xattr_packed
;
3388 error
= sa_size(hdl
, sa_attr_table
[ZPL_DXATTR
], &sa_xattr_size
);
3389 if (error
|| sa_xattr_size
== 0)
3392 sa_xattr_packed
= malloc(sa_xattr_size
);
3393 if (sa_xattr_packed
== NULL
)
3396 error
= sa_lookup(hdl
, sa_attr_table
[ZPL_DXATTR
],
3397 sa_xattr_packed
, sa_xattr_size
);
3399 free(sa_xattr_packed
);
3403 error
= nvlist_unpack(sa_xattr_packed
, sa_xattr_size
, &sa_xattr
, 0);
3405 free(sa_xattr_packed
);
3409 while ((elem
= nvlist_next_nvpair(sa_xattr
, elem
)) != NULL
)
3412 (void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3413 sa_xattr_size
, sa_xattr_entries
);
3414 while ((elem
= nvlist_next_nvpair(sa_xattr
, elem
)) != NULL
) {
3415 boolean_t can_print
= !dump_opt
['P'];
3419 (void) printf("\t\t%s = ", nvpair_name(elem
));
3420 nvpair_value_byte_array(elem
, &value
, &cnt
);
3422 for (idx
= 0; idx
< cnt
; ++idx
) {
3423 if (!isprint(value
[idx
])) {
3424 can_print
= B_FALSE
;
3429 for (idx
= 0; idx
< cnt
; ++idx
) {
3431 (void) putchar(value
[idx
]);
3433 (void) printf("\\%3.3o", value
[idx
]);
3435 (void) putchar('\n');
3438 nvlist_free(sa_xattr
);
3439 free(sa_xattr_packed
);
3443 dump_znode_symlink(sa_handle_t
*hdl
)
3445 int sa_symlink_size
= 0;
3446 char linktarget
[MAXPATHLEN
];
3449 error
= sa_size(hdl
, sa_attr_table
[ZPL_SYMLINK
], &sa_symlink_size
);
3450 if (error
|| sa_symlink_size
== 0) {
3453 if (sa_symlink_size
>= sizeof (linktarget
)) {
3454 (void) printf("symlink size %d is too large\n",
3458 linktarget
[sa_symlink_size
] = '\0';
3459 if (sa_lookup(hdl
, sa_attr_table
[ZPL_SYMLINK
],
3460 &linktarget
, sa_symlink_size
) == 0)
3461 (void) printf("\ttarget %s\n", linktarget
);
3465 dump_znode(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
3467 (void) data
, (void) size
;
3468 char path
[MAXPATHLEN
* 2]; /* allow for xattr and failure prefix */
3470 uint64_t xattr
, rdev
, gen
;
3471 uint64_t uid
, gid
, mode
, fsize
, parent
, links
;
3473 uint64_t acctm
[2], modtm
[2], chgtm
[2], crtm
[2];
3474 time_t z_crtime
, z_atime
, z_mtime
, z_ctime
;
3475 sa_bulk_attr_t bulk
[12];
3479 VERIFY3P(os
, ==, sa_os
);
3480 if (sa_handle_get(os
, object
, NULL
, SA_HDL_PRIVATE
, &hdl
)) {
3481 (void) printf("Failed to get handle for SA znode\n");
3485 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_UID
], NULL
, &uid
, 8);
3486 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_GID
], NULL
, &gid
, 8);
3487 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_LINKS
], NULL
,
3489 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_GEN
], NULL
, &gen
, 8);
3490 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_MODE
], NULL
,
3492 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_PARENT
],
3494 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_SIZE
], NULL
,
3496 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_ATIME
], NULL
,
3498 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_MTIME
], NULL
,
3500 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_CRTIME
], NULL
,
3502 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_CTIME
], NULL
,
3504 SA_ADD_BULK_ATTR(bulk
, idx
, sa_attr_table
[ZPL_FLAGS
], NULL
,
3507 if (sa_bulk_lookup(hdl
, bulk
, idx
)) {
3508 (void) sa_handle_destroy(hdl
);
3512 z_crtime
= (time_t)crtm
[0];
3513 z_atime
= (time_t)acctm
[0];
3514 z_mtime
= (time_t)modtm
[0];
3515 z_ctime
= (time_t)chgtm
[0];
3517 if (dump_opt
['d'] > 4) {
3518 error
= zfs_obj_to_path(os
, object
, path
, sizeof (path
));
3519 if (error
== ESTALE
) {
3520 (void) snprintf(path
, sizeof (path
), "on delete queue");
3521 } else if (error
!= 0) {
3523 (void) snprintf(path
, sizeof (path
),
3524 "path not found, possibly leaked");
3526 (void) printf("\tpath %s\n", path
);
3530 dump_znode_symlink(hdl
);
3531 dump_uidgid(os
, uid
, gid
);
3532 (void) printf("\tatime %s", ctime(&z_atime
));
3533 (void) printf("\tmtime %s", ctime(&z_mtime
));
3534 (void) printf("\tctime %s", ctime(&z_ctime
));
3535 (void) printf("\tcrtime %s", ctime(&z_crtime
));
3536 (void) printf("\tgen %llu\n", (u_longlong_t
)gen
);
3537 (void) printf("\tmode %llo\n", (u_longlong_t
)mode
);
3538 (void) printf("\tsize %llu\n", (u_longlong_t
)fsize
);
3539 (void) printf("\tparent %llu\n", (u_longlong_t
)parent
);
3540 (void) printf("\tlinks %llu\n", (u_longlong_t
)links
);
3541 (void) printf("\tpflags %llx\n", (u_longlong_t
)pflags
);
3542 if (dmu_objset_projectquota_enabled(os
) && (pflags
& ZFS_PROJID
)) {
3545 if (sa_lookup(hdl
, sa_attr_table
[ZPL_PROJID
], &projid
,
3546 sizeof (uint64_t)) == 0)
3547 (void) printf("\tprojid %llu\n", (u_longlong_t
)projid
);
3549 if (sa_lookup(hdl
, sa_attr_table
[ZPL_XATTR
], &xattr
,
3550 sizeof (uint64_t)) == 0)
3551 (void) printf("\txattr %llu\n", (u_longlong_t
)xattr
);
3552 if (sa_lookup(hdl
, sa_attr_table
[ZPL_RDEV
], &rdev
,
3553 sizeof (uint64_t)) == 0)
3554 (void) printf("\trdev 0x%016llx\n", (u_longlong_t
)rdev
);
3555 dump_znode_sa_xattr(hdl
);
3556 sa_handle_destroy(hdl
);
3560 dump_acl(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
3562 (void) os
, (void) object
, (void) data
, (void) size
;
3566 dump_dmu_objset(objset_t
*os
, uint64_t object
, void *data
, size_t size
)
3568 (void) os
, (void) object
, (void) data
, (void) size
;
3571 static object_viewer_t
*object_viewer
[DMU_OT_NUMTYPES
+ 1] = {
3572 dump_none
, /* unallocated */
3573 dump_zap
, /* object directory */
3574 dump_uint64
, /* object array */
3575 dump_none
, /* packed nvlist */
3576 dump_packed_nvlist
, /* packed nvlist size */
3577 dump_none
, /* bpobj */
3578 dump_bpobj
, /* bpobj header */
3579 dump_none
, /* SPA space map header */
3580 dump_none
, /* SPA space map */
3581 dump_none
, /* ZIL intent log */
3582 dump_dnode
, /* DMU dnode */
3583 dump_dmu_objset
, /* DMU objset */
3584 dump_dsl_dir
, /* DSL directory */
3585 dump_zap
, /* DSL directory child map */
3586 dump_zap
, /* DSL dataset snap map */
3587 dump_zap
, /* DSL props */
3588 dump_dsl_dataset
, /* DSL dataset */
3589 dump_znode
, /* ZFS znode */
3590 dump_acl
, /* ZFS V0 ACL */
3591 dump_uint8
, /* ZFS plain file */
3592 dump_zpldir
, /* ZFS directory */
3593 dump_zap
, /* ZFS master node */
3594 dump_zap
, /* ZFS delete queue */
3595 dump_uint8
, /* zvol object */
3596 dump_zap
, /* zvol prop */
3597 dump_uint8
, /* other uint8[] */
3598 dump_uint64
, /* other uint64[] */
3599 dump_zap
, /* other ZAP */
3600 dump_zap
, /* persistent error log */
3601 dump_uint8
, /* SPA history */
3602 dump_history_offsets
, /* SPA history offsets */
3603 dump_zap
, /* Pool properties */
3604 dump_zap
, /* DSL permissions */
3605 dump_acl
, /* ZFS ACL */
3606 dump_uint8
, /* ZFS SYSACL */
3607 dump_none
, /* FUID nvlist */
3608 dump_packed_nvlist
, /* FUID nvlist size */
3609 dump_zap
, /* DSL dataset next clones */
3610 dump_zap
, /* DSL scrub queue */
3611 dump_zap
, /* ZFS user/group/project used */
3612 dump_zap
, /* ZFS user/group/project quota */
3613 dump_zap
, /* snapshot refcount tags */
3614 dump_ddt_zap
, /* DDT ZAP object */
3615 dump_zap
, /* DDT statistics */
3616 dump_znode
, /* SA object */
3617 dump_zap
, /* SA Master Node */
3618 dump_sa_attrs
, /* SA attribute registration */
3619 dump_sa_layouts
, /* SA attribute layouts */
3620 dump_zap
, /* DSL scrub translations */
3621 dump_none
, /* fake dedup BP */
3622 dump_zap
, /* deadlist */
3623 dump_none
, /* deadlist hdr */
3624 dump_zap
, /* dsl clones */
3625 dump_bpobj_subobjs
, /* bpobj subobjs */
3626 dump_unknown
, /* Unknown type, must be last */
3630 match_object_type(dmu_object_type_t obj_type
, uint64_t flags
)
3632 boolean_t match
= B_TRUE
;
3635 case DMU_OT_DIRECTORY_CONTENTS
:
3636 if (!(flags
& ZOR_FLAG_DIRECTORY
))
3639 case DMU_OT_PLAIN_FILE_CONTENTS
:
3640 if (!(flags
& ZOR_FLAG_PLAIN_FILE
))
3643 case DMU_OT_SPACE_MAP
:
3644 if (!(flags
& ZOR_FLAG_SPACE_MAP
))
3648 if (strcmp(zdb_ot_name(obj_type
), "zap") == 0) {
3649 if (!(flags
& ZOR_FLAG_ZAP
))
3655 * If all bits except some of the supported flags are
3656 * set, the user combined the all-types flag (A) with
3657 * a negated flag to exclude some types (e.g. A-f to
3658 * show all object types except plain files).
3660 if ((flags
| ZOR_SUPPORTED_FLAGS
) != ZOR_FLAG_ALL_TYPES
)
3670 dump_object(objset_t
*os
, uint64_t object
, int verbosity
,
3671 boolean_t
*print_header
, uint64_t *dnode_slots_used
, uint64_t flags
)
3673 dmu_buf_t
*db
= NULL
;
3674 dmu_object_info_t doi
;
3676 boolean_t dnode_held
= B_FALSE
;
3679 char iblk
[32], dblk
[32], lsize
[32], asize
[32], fill
[32], dnsize
[32];
3680 char bonus_size
[32];
3684 /* make sure nicenum has enough space */
3685 _Static_assert(sizeof (iblk
) >= NN_NUMBUF_SZ
, "iblk truncated");
3686 _Static_assert(sizeof (dblk
) >= NN_NUMBUF_SZ
, "dblk truncated");
3687 _Static_assert(sizeof (lsize
) >= NN_NUMBUF_SZ
, "lsize truncated");
3688 _Static_assert(sizeof (asize
) >= NN_NUMBUF_SZ
, "asize truncated");
3689 _Static_assert(sizeof (bonus_size
) >= NN_NUMBUF_SZ
,
3690 "bonus_size truncated");
3692 if (*print_header
) {
3693 (void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n",
3694 "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3695 "lsize", "%full", "type");
3700 dn
= DMU_META_DNODE(os
);
3701 dmu_object_info_from_dnode(dn
, &doi
);
3704 * Encrypted datasets will have sensitive bonus buffers
3705 * encrypted. Therefore we cannot hold the bonus buffer and
3706 * must hold the dnode itself instead.
3708 error
= dmu_object_info(os
, object
, &doi
);
3710 fatal("dmu_object_info() failed, errno %u", error
);
3712 if (!key_loaded
&& os
->os_encrypted
&&
3713 DMU_OT_IS_ENCRYPTED(doi
.doi_bonus_type
)) {
3714 error
= dnode_hold(os
, object
, FTAG
, &dn
);
3716 fatal("dnode_hold() failed, errno %u", error
);
3717 dnode_held
= B_TRUE
;
3719 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
3721 fatal("dmu_bonus_hold(%llu) failed, errno %u",
3723 bonus
= db
->db_data
;
3724 bsize
= db
->db_size
;
3725 dn
= DB_DNODE((dmu_buf_impl_t
*)db
);
3730 * Default to showing all object types if no flags were specified.
3732 if (flags
!= 0 && flags
!= ZOR_FLAG_ALL_TYPES
&&
3733 !match_object_type(doi
.doi_type
, flags
))
3736 if (dnode_slots_used
)
3737 *dnode_slots_used
= doi
.doi_dnodesize
/ DNODE_MIN_SIZE
;
3739 zdb_nicenum(doi
.doi_metadata_block_size
, iblk
, sizeof (iblk
));
3740 zdb_nicenum(doi
.doi_data_block_size
, dblk
, sizeof (dblk
));
3741 zdb_nicenum(doi
.doi_max_offset
, lsize
, sizeof (lsize
));
3742 zdb_nicenum(doi
.doi_physical_blocks_512
<< 9, asize
, sizeof (asize
));
3743 zdb_nicenum(doi
.doi_bonus_size
, bonus_size
, sizeof (bonus_size
));
3744 zdb_nicenum(doi
.doi_dnodesize
, dnsize
, sizeof (dnsize
));
3745 (void) snprintf(fill
, sizeof (fill
), "%6.2f", 100.0 *
3746 doi
.doi_fill_count
* doi
.doi_data_block_size
/ (object
== 0 ?
3747 DNODES_PER_BLOCK
: 1) / doi
.doi_max_offset
);
3751 if (doi
.doi_checksum
!= ZIO_CHECKSUM_INHERIT
|| verbosity
>= 6) {
3752 (void) snprintf(aux
+ strlen(aux
), sizeof (aux
) - strlen(aux
),
3753 " (K=%s)", ZDB_CHECKSUM_NAME(doi
.doi_checksum
));
3756 if (doi
.doi_compress
== ZIO_COMPRESS_INHERIT
&&
3757 ZIO_COMPRESS_HASLEVEL(os
->os_compress
) && verbosity
>= 6) {
3758 const char *compname
= NULL
;
3759 if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION
,
3760 ZIO_COMPRESS_RAW(os
->os_compress
, os
->os_complevel
),
3762 (void) snprintf(aux
+ strlen(aux
),
3763 sizeof (aux
) - strlen(aux
), " (Z=inherit=%s)",
3766 (void) snprintf(aux
+ strlen(aux
),
3767 sizeof (aux
) - strlen(aux
),
3768 " (Z=inherit=%s-unknown)",
3769 ZDB_COMPRESS_NAME(os
->os_compress
));
3771 } else if (doi
.doi_compress
== ZIO_COMPRESS_INHERIT
&& verbosity
>= 6) {
3772 (void) snprintf(aux
+ strlen(aux
), sizeof (aux
) - strlen(aux
),
3773 " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os
->os_compress
));
3774 } else if (doi
.doi_compress
!= ZIO_COMPRESS_INHERIT
|| verbosity
>= 6) {
3775 (void) snprintf(aux
+ strlen(aux
), sizeof (aux
) - strlen(aux
),
3776 " (Z=%s)", ZDB_COMPRESS_NAME(doi
.doi_compress
));
3779 (void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n",
3780 (u_longlong_t
)object
, doi
.doi_indirection
, iblk
, dblk
,
3781 asize
, dnsize
, lsize
, fill
, zdb_ot_name(doi
.doi_type
), aux
);
3783 if (doi
.doi_bonus_type
!= DMU_OT_NONE
&& verbosity
> 3) {
3784 (void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n",
3785 "", "", "", "", "", "", bonus_size
, "bonus",
3786 zdb_ot_name(doi
.doi_bonus_type
));
3789 if (verbosity
>= 4) {
3790 (void) printf("\tdnode flags: %s%s%s%s\n",
3791 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_USED_BYTES
) ?
3793 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_USERUSED_ACCOUNTED
) ?
3794 "USERUSED_ACCOUNTED " : "",
3795 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_USEROBJUSED_ACCOUNTED
) ?
3796 "USEROBJUSED_ACCOUNTED " : "",
3797 (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) ?
3798 "SPILL_BLKPTR" : "");
3799 (void) printf("\tdnode maxblkid: %llu\n",
3800 (longlong_t
)dn
->dn_phys
->dn_maxblkid
);
3803 object_viewer
[ZDB_OT_TYPE(doi
.doi_bonus_type
)](os
,
3804 object
, bonus
, bsize
);
3806 (void) printf("\t\t(bonus encrypted)\n");
3810 (!os
->os_encrypted
|| !DMU_OT_IS_ENCRYPTED(doi
.doi_type
))) {
3811 object_viewer
[ZDB_OT_TYPE(doi
.doi_type
)](os
, object
,
3814 (void) printf("\t\t(object encrypted)\n");
3817 *print_header
= B_TRUE
;
3820 if (verbosity
>= 5) {
3821 if (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
3822 char blkbuf
[BP_SPRINTF_LEN
];
3823 snprintf_blkptr_compact(blkbuf
, sizeof (blkbuf
),
3824 DN_SPILL_BLKPTR(dn
->dn_phys
), B_FALSE
);
3825 (void) printf("\nSpill block: %s\n", blkbuf
);
3830 if (verbosity
>= 5) {
3832 * Report the list of segments that comprise the object.
3836 uint64_t blkfill
= 1;
3839 if (dn
->dn_type
== DMU_OT_DNODE
) {
3841 blkfill
= DNODES_PER_BLOCK
;
3846 /* make sure nicenum has enough space */
3847 _Static_assert(sizeof (segsize
) >= NN_NUMBUF_SZ
,
3848 "segsize truncated");
3849 error
= dnode_next_offset(dn
,
3850 0, &start
, minlvl
, blkfill
, 0);
3854 error
= dnode_next_offset(dn
,
3855 DNODE_FIND_HOLE
, &end
, minlvl
, blkfill
, 0);
3856 zdb_nicenum(end
- start
, segsize
, sizeof (segsize
));
3857 (void) printf("\t\tsegment [%016llx, %016llx)"
3858 " size %5s\n", (u_longlong_t
)start
,
3859 (u_longlong_t
)end
, segsize
);
3868 dmu_buf_rele(db
, FTAG
);
3870 dnode_rele(dn
, FTAG
);
3874 count_dir_mos_objects(dsl_dir_t
*dd
)
3876 mos_obj_refd(dd
->dd_object
);
3877 mos_obj_refd(dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
3878 mos_obj_refd(dsl_dir_phys(dd
)->dd_deleg_zapobj
);
3879 mos_obj_refd(dsl_dir_phys(dd
)->dd_props_zapobj
);
3880 mos_obj_refd(dsl_dir_phys(dd
)->dd_clones
);
3883 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
3884 * Ignore the references after the first one.
3886 mos_obj_refd_multiple(dd
->dd_crypto_obj
);
3890 count_ds_mos_objects(dsl_dataset_t
*ds
)
3892 mos_obj_refd(ds
->ds_object
);
3893 mos_obj_refd(dsl_dataset_phys(ds
)->ds_next_clones_obj
);
3894 mos_obj_refd(dsl_dataset_phys(ds
)->ds_props_obj
);
3895 mos_obj_refd(dsl_dataset_phys(ds
)->ds_userrefs_obj
);
3896 mos_obj_refd(dsl_dataset_phys(ds
)->ds_snapnames_zapobj
);
3897 mos_obj_refd(ds
->ds_bookmarks_obj
);
3899 if (!dsl_dataset_is_snapshot(ds
)) {
3900 count_dir_mos_objects(ds
->ds_dir
);
3904 static const char *const objset_types
[DMU_OST_NUMTYPES
] = {
3905 "NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
3908 * Parse a string denoting a range of object IDs of the form
3909 * <start>[:<end>[:flags]], and store the results in zor.
3910 * Return 0 on success. On error, return 1 and update the msg
3911 * pointer to point to a descriptive error message.
3914 parse_object_range(char *range
, zopt_object_range_t
*zor
, const char **msg
)
3917 char *p
, *s
, *dup
, *flagstr
, *tmp
= NULL
;
3922 if (strchr(range
, ':') == NULL
) {
3923 zor
->zor_obj_start
= strtoull(range
, &p
, 0);
3925 *msg
= "Invalid characters in object ID";
3928 zor
->zor_obj_start
= ZDB_MAP_OBJECT_ID(zor
->zor_obj_start
);
3929 zor
->zor_obj_end
= zor
->zor_obj_start
;
3933 if (strchr(range
, ':') == range
) {
3934 *msg
= "Invalid leading colon";
3939 len
= strlen(range
);
3940 if (range
[len
- 1] == ':') {
3941 *msg
= "Invalid trailing colon";
3946 dup
= strdup(range
);
3947 s
= strtok_r(dup
, ":", &tmp
);
3948 zor
->zor_obj_start
= strtoull(s
, &p
, 0);
3951 *msg
= "Invalid characters in start object ID";
3956 s
= strtok_r(NULL
, ":", &tmp
);
3957 zor
->zor_obj_end
= strtoull(s
, &p
, 0);
3960 *msg
= "Invalid characters in end object ID";
3965 if (zor
->zor_obj_start
> zor
->zor_obj_end
) {
3966 *msg
= "Start object ID may not exceed end object ID";
3971 s
= strtok_r(NULL
, ":", &tmp
);
3973 zor
->zor_flags
= ZOR_FLAG_ALL_TYPES
;
3975 } else if (strtok_r(NULL
, ":", &tmp
) != NULL
) {
3976 *msg
= "Invalid colon-delimited field after flags";
3982 for (i
= 0; flagstr
[i
]; i
++) {
3984 boolean_t negation
= (flagstr
[i
] == '-');
3988 if (flagstr
[i
] == '\0') {
3989 *msg
= "Invalid trailing negation operator";
3994 bit
= flagbits
[(uchar_t
)flagstr
[i
]];
3996 *msg
= "Invalid flag";
4005 zor
->zor_flags
= flags
;
4007 zor
->zor_obj_start
= ZDB_MAP_OBJECT_ID(zor
->zor_obj_start
);
4008 zor
->zor_obj_end
= ZDB_MAP_OBJECT_ID(zor
->zor_obj_end
);
4016 dump_objset(objset_t
*os
)
4018 dmu_objset_stats_t dds
= { 0 };
4019 uint64_t object
, object_count
;
4020 uint64_t refdbytes
, usedobjs
, scratch
;
4022 char blkbuf
[BP_SPRINTF_LEN
+ 20];
4023 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
4024 const char *type
= "UNKNOWN";
4025 int verbosity
= dump_opt
['d'];
4026 boolean_t print_header
;
4029 uint64_t total_slots_used
= 0;
4030 uint64_t max_slot_used
= 0;
4031 uint64_t dnode_slots
;
4036 /* make sure nicenum has enough space */
4037 _Static_assert(sizeof (numbuf
) >= NN_NUMBUF_SZ
, "numbuf truncated");
4039 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
4040 dmu_objset_fast_stat(os
, &dds
);
4041 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
4043 print_header
= B_TRUE
;
4045 if (dds
.dds_type
< DMU_OST_NUMTYPES
)
4046 type
= objset_types
[dds
.dds_type
];
4048 if (dds
.dds_type
== DMU_OST_META
) {
4049 dds
.dds_creation_txg
= TXG_INITIAL
;
4050 usedobjs
= BP_GET_FILL(os
->os_rootbp
);
4051 refdbytes
= dsl_dir_phys(os
->os_spa
->spa_dsl_pool
->dp_mos_dir
)->
4054 dmu_objset_space(os
, &refdbytes
, &scratch
, &usedobjs
, &scratch
);
4057 ASSERT3U(usedobjs
, ==, BP_GET_FILL(os
->os_rootbp
));
4059 zdb_nicenum(refdbytes
, numbuf
, sizeof (numbuf
));
4061 if (verbosity
>= 4) {
4062 (void) snprintf(blkbuf
, sizeof (blkbuf
), ", rootbp ");
4063 (void) snprintf_blkptr(blkbuf
+ strlen(blkbuf
),
4064 sizeof (blkbuf
) - strlen(blkbuf
), os
->os_rootbp
);
4069 dmu_objset_name(os
, osname
);
4071 (void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
4072 "%s, %llu objects%s%s\n",
4073 osname
, type
, (u_longlong_t
)dmu_objset_id(os
),
4074 (u_longlong_t
)dds
.dds_creation_txg
,
4075 numbuf
, (u_longlong_t
)usedobjs
, blkbuf
,
4076 (dds
.dds_inconsistent
) ? " (inconsistent)" : "");
4078 for (i
= 0; i
< zopt_object_args
; i
++) {
4079 obj_start
= zopt_object_ranges
[i
].zor_obj_start
;
4080 obj_end
= zopt_object_ranges
[i
].zor_obj_end
;
4081 flags
= zopt_object_ranges
[i
].zor_flags
;
4084 if (object
== 0 || obj_start
== obj_end
)
4085 dump_object(os
, object
, verbosity
, &print_header
, NULL
,
4090 while ((dmu_object_next(os
, &object
, B_FALSE
, 0) == 0) &&
4091 object
<= obj_end
) {
4092 dump_object(os
, object
, verbosity
, &print_header
, NULL
,
4097 if (zopt_object_args
> 0) {
4098 (void) printf("\n");
4102 if (dump_opt
['i'] != 0 || verbosity
>= 2)
4103 dump_intent_log(dmu_objset_zil(os
));
4105 if (dmu_objset_ds(os
) != NULL
) {
4106 dsl_dataset_t
*ds
= dmu_objset_ds(os
);
4107 dump_blkptr_list(&ds
->ds_deadlist
, "Deadlist");
4108 if (dsl_deadlist_is_open(&ds
->ds_dir
->dd_livelist
) &&
4109 !dmu_objset_is_snapshot(os
)) {
4110 dump_blkptr_list(&ds
->ds_dir
->dd_livelist
, "Livelist");
4111 if (verify_dd_livelist(os
) != 0)
4112 fatal("livelist is incorrect");
4115 if (dsl_dataset_remap_deadlist_exists(ds
)) {
4116 (void) printf("ds_remap_deadlist:\n");
4117 dump_blkptr_list(&ds
->ds_remap_deadlist
, "Deadlist");
4119 count_ds_mos_objects(ds
);
4122 if (dmu_objset_ds(os
) != NULL
)
4123 dump_bookmarks(os
, verbosity
);
4128 if (BP_IS_HOLE(os
->os_rootbp
))
4131 dump_object(os
, 0, verbosity
, &print_header
, NULL
, 0);
4133 if (DMU_USERUSED_DNODE(os
) != NULL
&&
4134 DMU_USERUSED_DNODE(os
)->dn_type
!= 0) {
4135 dump_object(os
, DMU_USERUSED_OBJECT
, verbosity
, &print_header
,
4137 dump_object(os
, DMU_GROUPUSED_OBJECT
, verbosity
, &print_header
,
4141 if (DMU_PROJECTUSED_DNODE(os
) != NULL
&&
4142 DMU_PROJECTUSED_DNODE(os
)->dn_type
!= 0)
4143 dump_object(os
, DMU_PROJECTUSED_OBJECT
, verbosity
,
4144 &print_header
, NULL
, 0);
4147 while ((error
= dmu_object_next(os
, &object
, B_FALSE
, 0)) == 0) {
4148 dump_object(os
, object
, verbosity
, &print_header
, &dnode_slots
,
4151 total_slots_used
+= dnode_slots
;
4152 max_slot_used
= object
+ dnode_slots
- 1;
4155 (void) printf("\n");
4157 (void) printf(" Dnode slots:\n");
4158 (void) printf("\tTotal used: %10llu\n",
4159 (u_longlong_t
)total_slots_used
);
4160 (void) printf("\tMax used: %10llu\n",
4161 (u_longlong_t
)max_slot_used
);
4162 (void) printf("\tPercent empty: %10lf\n",
4163 (double)(max_slot_used
- total_slots_used
)*100 /
4164 (double)max_slot_used
);
4165 (void) printf("\n");
4167 if (error
!= ESRCH
) {
4168 (void) fprintf(stderr
, "dmu_object_next() = %d\n", error
);
4172 ASSERT3U(object_count
, ==, usedobjs
);
4174 if (leaked_objects
!= 0) {
4175 (void) printf("%d potentially leaked objects detected\n",
4182 dump_uberblock(uberblock_t
*ub
, const char *header
, const char *footer
)
4184 time_t timestamp
= ub
->ub_timestamp
;
4186 (void) printf("%s", header
? header
: "");
4187 (void) printf("\tmagic = %016llx\n", (u_longlong_t
)ub
->ub_magic
);
4188 (void) printf("\tversion = %llu\n", (u_longlong_t
)ub
->ub_version
);
4189 (void) printf("\ttxg = %llu\n", (u_longlong_t
)ub
->ub_txg
);
4190 (void) printf("\tguid_sum = %llu\n", (u_longlong_t
)ub
->ub_guid_sum
);
4191 (void) printf("\ttimestamp = %llu UTC = %s",
4192 (u_longlong_t
)ub
->ub_timestamp
, ctime(×tamp
));
4194 (void) printf("\tmmp_magic = %016llx\n",
4195 (u_longlong_t
)ub
->ub_mmp_magic
);
4196 if (MMP_VALID(ub
)) {
4197 (void) printf("\tmmp_delay = %0llu\n",
4198 (u_longlong_t
)ub
->ub_mmp_delay
);
4199 if (MMP_SEQ_VALID(ub
))
4200 (void) printf("\tmmp_seq = %u\n",
4201 (unsigned int) MMP_SEQ(ub
));
4202 if (MMP_FAIL_INT_VALID(ub
))
4203 (void) printf("\tmmp_fail = %u\n",
4204 (unsigned int) MMP_FAIL_INT(ub
));
4205 if (MMP_INTERVAL_VALID(ub
))
4206 (void) printf("\tmmp_write = %u\n",
4207 (unsigned int) MMP_INTERVAL(ub
));
4208 /* After MMP_* to make summarize_uberblock_mmp cleaner */
4209 (void) printf("\tmmp_valid = %x\n",
4210 (unsigned int) ub
->ub_mmp_config
& 0xFF);
4213 if (dump_opt
['u'] >= 4) {
4214 char blkbuf
[BP_SPRINTF_LEN
];
4215 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), &ub
->ub_rootbp
);
4216 (void) printf("\trootbp = %s\n", blkbuf
);
4218 (void) printf("\tcheckpoint_txg = %llu\n",
4219 (u_longlong_t
)ub
->ub_checkpoint_txg
);
4220 (void) printf("%s", footer
? footer
: "");
4224 dump_config(spa_t
*spa
)
4231 error
= dmu_bonus_hold(spa
->spa_meta_objset
,
4232 spa
->spa_config_object
, FTAG
, &db
);
4235 nvsize
= *(uint64_t *)db
->db_data
;
4236 dmu_buf_rele(db
, FTAG
);
4238 (void) printf("\nMOS Configuration:\n");
4239 dump_packed_nvlist(spa
->spa_meta_objset
,
4240 spa
->spa_config_object
, (void *)&nvsize
, 1);
4242 (void) fprintf(stderr
, "dmu_bonus_hold(%llu) failed, errno %d",
4243 (u_longlong_t
)spa
->spa_config_object
, error
);
4248 dump_cachefile(const char *cachefile
)
4251 struct stat64 statbuf
;
4255 if ((fd
= open64(cachefile
, O_RDONLY
)) < 0) {
4256 (void) printf("cannot open '%s': %s\n", cachefile
,
4261 if (fstat64(fd
, &statbuf
) != 0) {
4262 (void) printf("failed to stat '%s': %s\n", cachefile
,
4267 if ((buf
= malloc(statbuf
.st_size
)) == NULL
) {
4268 (void) fprintf(stderr
, "failed to allocate %llu bytes\n",
4269 (u_longlong_t
)statbuf
.st_size
);
4273 if (read(fd
, buf
, statbuf
.st_size
) != statbuf
.st_size
) {
4274 (void) fprintf(stderr
, "failed to read %llu bytes\n",
4275 (u_longlong_t
)statbuf
.st_size
);
4281 if (nvlist_unpack(buf
, statbuf
.st_size
, &config
, 0) != 0) {
4282 (void) fprintf(stderr
, "failed to unpack nvlist\n");
4288 dump_nvlist(config
, 0);
4290 nvlist_free(config
);
4294 * ZFS label nvlist stats
4296 typedef struct zdb_nvl_stats
{
4299 size_t zns_leaf_largest
;
4300 size_t zns_leaf_total
;
4301 nvlist_t
*zns_string
;
4302 nvlist_t
*zns_uint64
;
4303 nvlist_t
*zns_boolean
;
4307 collect_nvlist_stats(nvlist_t
*nvl
, zdb_nvl_stats_t
*stats
)
4309 nvlist_t
*list
, **array
;
4310 nvpair_t
*nvp
= NULL
;
4314 stats
->zns_list_count
++;
4316 while ((nvp
= nvlist_next_nvpair(nvl
, nvp
)) != NULL
) {
4317 name
= nvpair_name(nvp
);
4319 switch (nvpair_type(nvp
)) {
4320 case DATA_TYPE_STRING
:
4321 fnvlist_add_string(stats
->zns_string
, name
,
4322 fnvpair_value_string(nvp
));
4324 case DATA_TYPE_UINT64
:
4325 fnvlist_add_uint64(stats
->zns_uint64
, name
,
4326 fnvpair_value_uint64(nvp
));
4328 case DATA_TYPE_BOOLEAN
:
4329 fnvlist_add_boolean(stats
->zns_boolean
, name
);
4331 case DATA_TYPE_NVLIST
:
4332 if (nvpair_value_nvlist(nvp
, &list
) == 0)
4333 collect_nvlist_stats(list
, stats
);
4335 case DATA_TYPE_NVLIST_ARRAY
:
4336 if (nvpair_value_nvlist_array(nvp
, &array
, &items
) != 0)
4339 for (i
= 0; i
< items
; i
++) {
4340 collect_nvlist_stats(array
[i
], stats
);
4342 /* collect stats on leaf vdev */
4343 if (strcmp(name
, "children") == 0) {
4346 (void) nvlist_size(array
[i
], &size
,
4348 stats
->zns_leaf_total
+= size
;
4349 if (size
> stats
->zns_leaf_largest
)
4350 stats
->zns_leaf_largest
= size
;
4351 stats
->zns_leaf_count
++;
4356 (void) printf("skip type %d!\n", (int)nvpair_type(nvp
));
4362 dump_nvlist_stats(nvlist_t
*nvl
, size_t cap
)
4364 zdb_nvl_stats_t stats
= { 0 };
4365 size_t size
, sum
= 0, total
;
4368 /* requires nvlist with non-unique names for stat collection */
4369 VERIFY0(nvlist_alloc(&stats
.zns_string
, 0, 0));
4370 VERIFY0(nvlist_alloc(&stats
.zns_uint64
, 0, 0));
4371 VERIFY0(nvlist_alloc(&stats
.zns_boolean
, 0, 0));
4372 VERIFY0(nvlist_size(stats
.zns_boolean
, &noise
, NV_ENCODE_XDR
));
4374 (void) printf("\n\nZFS Label NVList Config Stats:\n");
4376 VERIFY0(nvlist_size(nvl
, &total
, NV_ENCODE_XDR
));
4377 (void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4378 (int)total
, (int)(cap
- total
), 100.0 * total
/ cap
);
4380 collect_nvlist_stats(nvl
, &stats
);
4382 VERIFY0(nvlist_size(stats
.zns_uint64
, &size
, NV_ENCODE_XDR
));
4385 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4386 (int)fnvlist_num_pairs(stats
.zns_uint64
),
4387 (int)size
, 100.0 * size
/ total
);
4389 VERIFY0(nvlist_size(stats
.zns_string
, &size
, NV_ENCODE_XDR
));
4392 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4393 (int)fnvlist_num_pairs(stats
.zns_string
),
4394 (int)size
, 100.0 * size
/ total
);
4396 VERIFY0(nvlist_size(stats
.zns_boolean
, &size
, NV_ENCODE_XDR
));
4399 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4400 (int)fnvlist_num_pairs(stats
.zns_boolean
),
4401 (int)size
, 100.0 * size
/ total
);
4403 size
= total
- sum
; /* treat remainder as nvlist overhead */
4404 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4405 stats
.zns_list_count
, (int)size
, 100.0 * size
/ total
);
4407 if (stats
.zns_leaf_count
> 0) {
4408 size_t average
= stats
.zns_leaf_total
/ stats
.zns_leaf_count
;
4410 (void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4411 stats
.zns_leaf_count
, (int)average
);
4412 (void) printf("%24d bytes largest\n",
4413 (int)stats
.zns_leaf_largest
);
4415 if (dump_opt
['l'] >= 3 && average
> 0)
4416 (void) printf(" space for %d additional leaf vdevs\n",
4417 (int)((cap
- total
) / average
));
4419 (void) printf("\n");
4421 nvlist_free(stats
.zns_string
);
4422 nvlist_free(stats
.zns_uint64
);
4423 nvlist_free(stats
.zns_boolean
);
4426 typedef struct cksum_record
{
4428 boolean_t labels
[VDEV_LABELS
];
4433 cksum_record_compare(const void *x1
, const void *x2
)
4435 const cksum_record_t
*l
= (cksum_record_t
*)x1
;
4436 const cksum_record_t
*r
= (cksum_record_t
*)x2
;
4437 int arraysize
= ARRAY_SIZE(l
->cksum
.zc_word
);
4440 for (int i
= 0; i
< arraysize
; i
++) {
4441 difference
= TREE_CMP(l
->cksum
.zc_word
[i
], r
->cksum
.zc_word
[i
]);
4446 return (difference
);
4449 static cksum_record_t
*
4450 cksum_record_alloc(zio_cksum_t
*cksum
, int l
)
4452 cksum_record_t
*rec
;
4454 rec
= umem_zalloc(sizeof (*rec
), UMEM_NOFAIL
);
4455 rec
->cksum
= *cksum
;
4456 rec
->labels
[l
] = B_TRUE
;
4461 static cksum_record_t
*
4462 cksum_record_lookup(avl_tree_t
*tree
, zio_cksum_t
*cksum
)
4464 cksum_record_t lookup
= { .cksum
= *cksum
};
4467 return (avl_find(tree
, &lookup
, &where
));
4470 static cksum_record_t
*
4471 cksum_record_insert(avl_tree_t
*tree
, zio_cksum_t
*cksum
, int l
)
4473 cksum_record_t
*rec
;
4475 rec
= cksum_record_lookup(tree
, cksum
);
4477 rec
->labels
[l
] = B_TRUE
;
4479 rec
= cksum_record_alloc(cksum
, l
);
4487 first_label(cksum_record_t
*rec
)
4489 for (int i
= 0; i
< VDEV_LABELS
; i
++)
4497 print_label_numbers(const char *prefix
, const cksum_record_t
*rec
)
4499 fputs(prefix
, stdout
);
4500 for (int i
= 0; i
< VDEV_LABELS
; i
++)
4501 if (rec
->labels
[i
] == B_TRUE
)
4506 #define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4508 typedef struct zdb_label
{
4510 uint64_t label_offset
;
4511 nvlist_t
*config_nv
;
4512 cksum_record_t
*config
;
4513 cksum_record_t
*uberblocks
[MAX_UBERBLOCK_COUNT
];
4514 boolean_t header_printed
;
4515 boolean_t read_failed
;
4516 boolean_t cksum_valid
;
4520 print_label_header(zdb_label_t
*label
, int l
)
4526 if (label
->header_printed
== B_TRUE
)
4529 (void) printf("------------------------------------\n");
4530 (void) printf("LABEL %d %s\n", l
,
4531 label
->cksum_valid
? "" : "(Bad label cksum)");
4532 (void) printf("------------------------------------\n");
4534 label
->header_printed
= B_TRUE
;
4538 print_l2arc_header(void)
4540 (void) printf("------------------------------------\n");
4541 (void) printf("L2ARC device header\n");
4542 (void) printf("------------------------------------\n");
4546 print_l2arc_log_blocks(void)
4548 (void) printf("------------------------------------\n");
4549 (void) printf("L2ARC device log blocks\n");
4550 (void) printf("------------------------------------\n");
4554 dump_l2arc_log_entries(uint64_t log_entries
,
4555 l2arc_log_ent_phys_t
*le
, uint64_t i
)
4557 for (int j
= 0; j
< log_entries
; j
++) {
4558 dva_t dva
= le
[j
].le_dva
;
4559 (void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4560 "vdev: %llu, offset: %llu\n",
4561 (u_longlong_t
)i
, j
+ 1,
4562 (u_longlong_t
)DVA_GET_ASIZE(&dva
),
4563 (u_longlong_t
)DVA_GET_VDEV(&dva
),
4564 (u_longlong_t
)DVA_GET_OFFSET(&dva
));
4565 (void) printf("|\t\t\t\tbirth: %llu\n",
4566 (u_longlong_t
)le
[j
].le_birth
);
4567 (void) printf("|\t\t\t\tlsize: %llu\n",
4568 (u_longlong_t
)L2BLK_GET_LSIZE((&le
[j
])->le_prop
));
4569 (void) printf("|\t\t\t\tpsize: %llu\n",
4570 (u_longlong_t
)L2BLK_GET_PSIZE((&le
[j
])->le_prop
));
4571 (void) printf("|\t\t\t\tcompr: %llu\n",
4572 (u_longlong_t
)L2BLK_GET_COMPRESS((&le
[j
])->le_prop
));
4573 (void) printf("|\t\t\t\tcomplevel: %llu\n",
4574 (u_longlong_t
)(&le
[j
])->le_complevel
);
4575 (void) printf("|\t\t\t\ttype: %llu\n",
4576 (u_longlong_t
)L2BLK_GET_TYPE((&le
[j
])->le_prop
));
4577 (void) printf("|\t\t\t\tprotected: %llu\n",
4578 (u_longlong_t
)L2BLK_GET_PROTECTED((&le
[j
])->le_prop
));
4579 (void) printf("|\t\t\t\tprefetch: %llu\n",
4580 (u_longlong_t
)L2BLK_GET_PREFETCH((&le
[j
])->le_prop
));
4581 (void) printf("|\t\t\t\taddress: %llu\n",
4582 (u_longlong_t
)le
[j
].le_daddr
);
4583 (void) printf("|\t\t\t\tARC state: %llu\n",
4584 (u_longlong_t
)L2BLK_GET_STATE((&le
[j
])->le_prop
));
4585 (void) printf("|\n");
4587 (void) printf("\n");
4591 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t
*lbps
)
4593 (void) printf("|\t\tdaddr: %llu\n", (u_longlong_t
)lbps
->lbp_daddr
);
4594 (void) printf("|\t\tpayload_asize: %llu\n",
4595 (u_longlong_t
)lbps
->lbp_payload_asize
);
4596 (void) printf("|\t\tpayload_start: %llu\n",
4597 (u_longlong_t
)lbps
->lbp_payload_start
);
4598 (void) printf("|\t\tlsize: %llu\n",
4599 (u_longlong_t
)L2BLK_GET_LSIZE(lbps
->lbp_prop
));
4600 (void) printf("|\t\tasize: %llu\n",
4601 (u_longlong_t
)L2BLK_GET_PSIZE(lbps
->lbp_prop
));
4602 (void) printf("|\t\tcompralgo: %llu\n",
4603 (u_longlong_t
)L2BLK_GET_COMPRESS(lbps
->lbp_prop
));
4604 (void) printf("|\t\tcksumalgo: %llu\n",
4605 (u_longlong_t
)L2BLK_GET_CHECKSUM(lbps
->lbp_prop
));
4606 (void) printf("|\n\n");
4610 dump_l2arc_log_blocks(int fd
, const l2arc_dev_hdr_phys_t
*l2dhdr
,
4611 l2arc_dev_hdr_phys_t
*rebuild
)
4613 l2arc_log_blk_phys_t this_lb
;
4615 l2arc_log_blkptr_t lbps
[2];
4622 print_l2arc_log_blocks();
4623 memcpy(lbps
, l2dhdr
->dh_start_lbps
, sizeof (lbps
));
4625 dev
.l2ad_evict
= l2dhdr
->dh_evict
;
4626 dev
.l2ad_start
= l2dhdr
->dh_start
;
4627 dev
.l2ad_end
= l2dhdr
->dh_end
;
4629 if (l2dhdr
->dh_start_lbps
[0].lbp_daddr
== 0) {
4630 /* no log blocks to read */
4631 if (!dump_opt
['q']) {
4632 (void) printf("No log blocks to read\n");
4633 (void) printf("\n");
4637 dev
.l2ad_hand
= lbps
[0].lbp_daddr
+
4638 L2BLK_GET_PSIZE((&lbps
[0])->lbp_prop
);
4641 dev
.l2ad_first
= !!(l2dhdr
->dh_flags
& L2ARC_DEV_HDR_EVICT_FIRST
);
4644 if (!l2arc_log_blkptr_valid(&dev
, &lbps
[0]))
4647 /* L2BLK_GET_PSIZE returns aligned size for log blocks */
4648 asize
= L2BLK_GET_PSIZE((&lbps
[0])->lbp_prop
);
4649 if (pread64(fd
, &this_lb
, asize
, lbps
[0].lbp_daddr
) != asize
) {
4650 if (!dump_opt
['q']) {
4651 (void) printf("Error while reading next log "
4657 fletcher_4_native_varsize(&this_lb
, asize
, &cksum
);
4658 if (!ZIO_CHECKSUM_EQUAL(cksum
, lbps
[0].lbp_cksum
)) {
4660 if (!dump_opt
['q']) {
4661 (void) printf("Invalid cksum\n");
4662 dump_l2arc_log_blkptr(&lbps
[0]);
4667 switch (L2BLK_GET_COMPRESS((&lbps
[0])->lbp_prop
)) {
4668 case ZIO_COMPRESS_OFF
:
4671 abd
= abd_alloc_for_io(asize
, B_TRUE
);
4672 abd_copy_from_buf_off(abd
, &this_lb
, 0, asize
);
4673 if (zio_decompress_data(L2BLK_GET_COMPRESS(
4674 (&lbps
[0])->lbp_prop
), abd
, &this_lb
,
4675 asize
, sizeof (this_lb
), NULL
) != 0) {
4676 (void) printf("L2ARC block decompression "
4685 if (this_lb
.lb_magic
== BSWAP_64(L2ARC_LOG_BLK_MAGIC
))
4686 byteswap_uint64_array(&this_lb
, sizeof (this_lb
));
4687 if (this_lb
.lb_magic
!= L2ARC_LOG_BLK_MAGIC
) {
4689 (void) printf("Invalid log block magic\n\n");
4693 rebuild
->dh_lb_count
++;
4694 rebuild
->dh_lb_asize
+= asize
;
4695 if (dump_opt
['l'] > 1 && !dump_opt
['q']) {
4696 (void) printf("lb[%4llu]\tmagic: %llu\n",
4697 (u_longlong_t
)rebuild
->dh_lb_count
,
4698 (u_longlong_t
)this_lb
.lb_magic
);
4699 dump_l2arc_log_blkptr(&lbps
[0]);
4702 if (dump_opt
['l'] > 2 && !dump_opt
['q'])
4703 dump_l2arc_log_entries(l2dhdr
->dh_log_entries
,
4705 rebuild
->dh_lb_count
);
4707 if (l2arc_range_check_overlap(lbps
[1].lbp_payload_start
,
4708 lbps
[0].lbp_payload_start
, dev
.l2ad_evict
) &&
4713 lbps
[1] = this_lb
.lb_prev_lbp
;
4716 if (!dump_opt
['q']) {
4717 (void) printf("log_blk_count:\t %llu with valid cksum\n",
4718 (u_longlong_t
)rebuild
->dh_lb_count
);
4719 (void) printf("\t\t %d with invalid cksum\n", failed
);
4720 (void) printf("log_blk_asize:\t %llu\n\n",
4721 (u_longlong_t
)rebuild
->dh_lb_asize
);
4726 dump_l2arc_header(int fd
)
4728 l2arc_dev_hdr_phys_t l2dhdr
= {0}, rebuild
= {0};
4729 int error
= B_FALSE
;
4731 if (pread64(fd
, &l2dhdr
, sizeof (l2dhdr
),
4732 VDEV_LABEL_START_SIZE
) != sizeof (l2dhdr
)) {
4735 if (l2dhdr
.dh_magic
== BSWAP_64(L2ARC_DEV_HDR_MAGIC
))
4736 byteswap_uint64_array(&l2dhdr
, sizeof (l2dhdr
));
4738 if (l2dhdr
.dh_magic
!= L2ARC_DEV_HDR_MAGIC
)
4743 (void) printf("L2ARC device header not found\n\n");
4744 /* Do not return an error here for backward compatibility */
4746 } else if (!dump_opt
['q']) {
4747 print_l2arc_header();
4749 (void) printf(" magic: %llu\n",
4750 (u_longlong_t
)l2dhdr
.dh_magic
);
4751 (void) printf(" version: %llu\n",
4752 (u_longlong_t
)l2dhdr
.dh_version
);
4753 (void) printf(" pool_guid: %llu\n",
4754 (u_longlong_t
)l2dhdr
.dh_spa_guid
);
4755 (void) printf(" flags: %llu\n",
4756 (u_longlong_t
)l2dhdr
.dh_flags
);
4757 (void) printf(" start_lbps[0]: %llu\n",
4759 l2dhdr
.dh_start_lbps
[0].lbp_daddr
);
4760 (void) printf(" start_lbps[1]: %llu\n",
4762 l2dhdr
.dh_start_lbps
[1].lbp_daddr
);
4763 (void) printf(" log_blk_ent: %llu\n",
4764 (u_longlong_t
)l2dhdr
.dh_log_entries
);
4765 (void) printf(" start: %llu\n",
4766 (u_longlong_t
)l2dhdr
.dh_start
);
4767 (void) printf(" end: %llu\n",
4768 (u_longlong_t
)l2dhdr
.dh_end
);
4769 (void) printf(" evict: %llu\n",
4770 (u_longlong_t
)l2dhdr
.dh_evict
);
4771 (void) printf(" lb_asize_refcount: %llu\n",
4772 (u_longlong_t
)l2dhdr
.dh_lb_asize
);
4773 (void) printf(" lb_count_refcount: %llu\n",
4774 (u_longlong_t
)l2dhdr
.dh_lb_count
);
4775 (void) printf(" trim_action_time: %llu\n",
4776 (u_longlong_t
)l2dhdr
.dh_trim_action_time
);
4777 (void) printf(" trim_state: %llu\n\n",
4778 (u_longlong_t
)l2dhdr
.dh_trim_state
);
4781 dump_l2arc_log_blocks(fd
, &l2dhdr
, &rebuild
);
4783 * The total aligned size of log blocks and the number of log blocks
4784 * reported in the header of the device may be less than what zdb
4785 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
4786 * This happens because dump_l2arc_log_blocks() lacks the memory
4787 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
4788 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
4789 * and dh_lb_count will be lower to begin with than what exists on the
4790 * device. This is normal and zdb should not exit with an error. The
4791 * opposite case should never happen though, the values reported in the
4792 * header should never be higher than what dump_l2arc_log_blocks() and
4793 * l2arc_rebuild() report. If this happens there is a leak in the
4794 * accounting of log blocks.
4796 if (l2dhdr
.dh_lb_asize
> rebuild
.dh_lb_asize
||
4797 l2dhdr
.dh_lb_count
> rebuild
.dh_lb_count
)
4804 dump_config_from_label(zdb_label_t
*label
, size_t buflen
, int l
)
4809 if ((dump_opt
['l'] < 3) && (first_label(label
->config
) != l
))
4812 print_label_header(label
, l
);
4813 dump_nvlist(label
->config_nv
, 4);
4814 print_label_numbers(" labels = ", label
->config
);
4816 if (dump_opt
['l'] >= 2)
4817 dump_nvlist_stats(label
->config_nv
, buflen
);
4820 #define ZDB_MAX_UB_HEADER_SIZE 32
4823 dump_label_uberblocks(zdb_label_t
*label
, uint64_t ashift
, int label_num
)
4827 char header
[ZDB_MAX_UB_HEADER_SIZE
];
4829 vd
.vdev_ashift
= ashift
;
4832 for (int i
= 0; i
< VDEV_UBERBLOCK_COUNT(&vd
); i
++) {
4833 uint64_t uoff
= VDEV_UBERBLOCK_OFFSET(&vd
, i
);
4834 uberblock_t
*ub
= (void *)((char *)&label
->label
+ uoff
);
4835 cksum_record_t
*rec
= label
->uberblocks
[i
];
4838 if (dump_opt
['u'] >= 2) {
4839 print_label_header(label
, label_num
);
4840 (void) printf(" Uberblock[%d] invalid\n", i
);
4845 if ((dump_opt
['u'] < 3) && (first_label(rec
) != label_num
))
4848 if ((dump_opt
['u'] < 4) &&
4849 (ub
->ub_mmp_magic
== MMP_MAGIC
) && ub
->ub_mmp_delay
&&
4850 (i
>= VDEV_UBERBLOCK_COUNT(&vd
) - MMP_BLOCKS_PER_LABEL
))
4853 print_label_header(label
, label_num
);
4854 (void) snprintf(header
, ZDB_MAX_UB_HEADER_SIZE
,
4855 " Uberblock[%d]\n", i
);
4856 dump_uberblock(ub
, header
, "");
4857 print_label_numbers(" labels = ", rec
);
4861 static char curpath
[PATH_MAX
];
4864 * Iterate through the path components, recursively passing
4865 * current one's obj and remaining path until we find the obj
4869 dump_path_impl(objset_t
*os
, uint64_t obj
, char *name
, uint64_t *retobj
)
4872 boolean_t header
= B_TRUE
;
4876 dmu_object_info_t doi
;
4878 if ((s
= strchr(name
, '/')) != NULL
)
4880 err
= zap_lookup(os
, obj
, name
, 8, 1, &child_obj
);
4882 (void) strlcat(curpath
, name
, sizeof (curpath
));
4885 (void) fprintf(stderr
, "failed to lookup %s: %s\n",
4886 curpath
, strerror(err
));
4890 child_obj
= ZFS_DIRENT_OBJ(child_obj
);
4891 err
= sa_buf_hold(os
, child_obj
, FTAG
, &db
);
4893 (void) fprintf(stderr
,
4894 "failed to get SA dbuf for obj %llu: %s\n",
4895 (u_longlong_t
)child_obj
, strerror(err
));
4898 dmu_object_info_from_db(db
, &doi
);
4899 sa_buf_rele(db
, FTAG
);
4901 if (doi
.doi_bonus_type
!= DMU_OT_SA
&&
4902 doi
.doi_bonus_type
!= DMU_OT_ZNODE
) {
4903 (void) fprintf(stderr
, "invalid bonus type %d for obj %llu\n",
4904 doi
.doi_bonus_type
, (u_longlong_t
)child_obj
);
4908 if (dump_opt
['v'] > 6) {
4909 (void) printf("obj=%llu %s type=%d bonustype=%d\n",
4910 (u_longlong_t
)child_obj
, curpath
, doi
.doi_type
,
4911 doi
.doi_bonus_type
);
4914 (void) strlcat(curpath
, "/", sizeof (curpath
));
4916 switch (doi
.doi_type
) {
4917 case DMU_OT_DIRECTORY_CONTENTS
:
4918 if (s
!= NULL
&& *(s
+ 1) != '\0')
4919 return (dump_path_impl(os
, child_obj
, s
+ 1, retobj
));
4921 case DMU_OT_PLAIN_FILE_CONTENTS
:
4922 if (retobj
!= NULL
) {
4923 *retobj
= child_obj
;
4925 dump_object(os
, child_obj
, dump_opt
['v'], &header
,
4930 (void) fprintf(stderr
, "object %llu has non-file/directory "
4931 "type %d\n", (u_longlong_t
)obj
, doi
.doi_type
);
4939 * Dump the blocks for the object specified by path inside the dataset.
4942 dump_path(char *ds
, char *path
, uint64_t *retobj
)
4948 err
= open_objset(ds
, FTAG
, &os
);
4952 err
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_ROOT_OBJ
, 8, 1, &root_obj
);
4954 (void) fprintf(stderr
, "can't lookup root znode: %s\n",
4956 close_objset(os
, FTAG
);
4960 (void) snprintf(curpath
, sizeof (curpath
), "dataset=%s path=/", ds
);
4962 err
= dump_path_impl(os
, root_obj
, path
, retobj
);
4964 close_objset(os
, FTAG
);
4969 dump_backup_bytes(objset_t
*os
, void *buf
, int len
, void *arg
)
4971 const char *p
= (const char *)buf
;
4977 /* Write the data out, handling short writes and signals. */
4978 while ((nwritten
= write(STDOUT_FILENO
, p
, len
)) < len
) {
4992 dump_backup(const char *pool
, uint64_t objset_id
, const char *flagstr
)
4994 boolean_t embed
= B_FALSE
;
4995 boolean_t large_block
= B_FALSE
;
4996 boolean_t compress
= B_FALSE
;
4997 boolean_t raw
= B_FALSE
;
5000 for (c
= flagstr
; c
!= NULL
&& *c
!= '\0'; c
++) {
5006 large_block
= B_TRUE
;
5015 fprintf(stderr
, "dump_backup: invalid flag "
5021 if (isatty(STDOUT_FILENO
)) {
5022 fprintf(stderr
, "dump_backup: stream cannot be written "
5028 dmu_send_outparams_t out
= {
5029 .dso_outfunc
= dump_backup_bytes
,
5030 .dso_dryrun
= B_FALSE
,
5033 int err
= dmu_send_obj(pool
, objset_id
, /* fromsnap */0, embed
,
5034 large_block
, compress
, raw
, /* saved */ B_FALSE
, STDOUT_FILENO
,
5037 fprintf(stderr
, "dump_backup: dmu_send_obj: %s\n",
5044 zdb_copy_object(objset_t
*os
, uint64_t srcobj
, char *destfile
)
5047 uint64_t size
, readsize
, oursize
, offset
;
5051 (void) printf("Copying object %" PRIu64
" to file %s\n", srcobj
,
5054 VERIFY3P(os
, ==, sa_os
);
5055 if ((err
= sa_handle_get(os
, srcobj
, NULL
, SA_HDL_PRIVATE
, &hdl
))) {
5056 (void) printf("Failed to get handle for SA znode\n");
5059 if ((err
= sa_lookup(hdl
, sa_attr_table
[ZPL_SIZE
], &size
, 8))) {
5060 (void) sa_handle_destroy(hdl
);
5063 (void) sa_handle_destroy(hdl
);
5065 (void) printf("Object %" PRIu64
" is %" PRIu64
" bytes\n", srcobj
,
5071 int fd
= open(destfile
, O_WRONLY
| O_CREAT
| O_TRUNC
, 0644);
5075 * We cap the size at 1 mebibyte here to prevent
5076 * allocation failures and nigh-infinite printing if the
5077 * object is extremely large.
5079 oursize
= MIN(size
, 1 << 20);
5081 char *buf
= kmem_alloc(oursize
, KM_NOSLEEP
);
5087 while (offset
< size
) {
5088 readsize
= MIN(size
- offset
, 1 << 20);
5089 err
= dmu_read(os
, srcobj
, offset
, readsize
, buf
, 0);
5091 (void) printf("got error %u from dmu_read\n", err
);
5092 kmem_free(buf
, oursize
);
5096 if (dump_opt
['v'] > 3) {
5097 (void) printf("Read offset=%" PRIu64
" size=%" PRIu64
5098 " error=%d\n", offset
, readsize
, err
);
5101 writesize
= write(fd
, buf
, readsize
);
5102 if (writesize
< 0) {
5105 } else if (writesize
!= readsize
) {
5106 /* Incomplete write */
5107 (void) fprintf(stderr
, "Short write, only wrote %llu of"
5108 " %" PRIu64
" bytes, exiting...\n",
5109 (u_longlong_t
)writesize
, readsize
);
5119 kmem_free(buf
, oursize
);
5125 label_cksum_valid(vdev_label_t
*label
, uint64_t offset
)
5127 zio_checksum_info_t
*ci
= &zio_checksum_table
[ZIO_CHECKSUM_LABEL
];
5128 zio_cksum_t expected_cksum
;
5129 zio_cksum_t actual_cksum
;
5130 zio_cksum_t verifier
;
5134 void *data
= (char *)label
+ offsetof(vdev_label_t
, vl_vdev_phys
);
5135 eck
= (zio_eck_t
*)((char *)(data
) + VDEV_PHYS_SIZE
) - 1;
5137 offset
+= offsetof(vdev_label_t
, vl_vdev_phys
);
5138 ZIO_SET_CHECKSUM(&verifier
, offset
, 0, 0, 0);
5140 byteswap
= (eck
->zec_magic
== BSWAP_64(ZEC_MAGIC
));
5142 byteswap_uint64_array(&verifier
, sizeof (zio_cksum_t
));
5144 expected_cksum
= eck
->zec_cksum
;
5145 eck
->zec_cksum
= verifier
;
5147 abd_t
*abd
= abd_get_from_buf(data
, VDEV_PHYS_SIZE
);
5148 ci
->ci_func
[byteswap
](abd
, VDEV_PHYS_SIZE
, NULL
, &actual_cksum
);
5152 byteswap_uint64_array(&expected_cksum
, sizeof (zio_cksum_t
));
5154 if (ZIO_CHECKSUM_EQUAL(actual_cksum
, expected_cksum
))
5161 dump_label(const char *dev
)
5163 char path
[MAXPATHLEN
];
5164 zdb_label_t labels
[VDEV_LABELS
] = {{{{0}}}};
5165 uint64_t psize
, ashift
, l2cache
;
5166 struct stat64 statbuf
;
5167 boolean_t config_found
= B_FALSE
;
5168 boolean_t error
= B_FALSE
;
5169 boolean_t read_l2arc_header
= B_FALSE
;
5170 avl_tree_t config_tree
;
5171 avl_tree_t uberblock_tree
;
5172 void *node
, *cookie
;
5176 * Check if we were given absolute path and use it as is.
5177 * Otherwise if the provided vdev name doesn't point to a file,
5178 * try prepending expected disk paths and partition numbers.
5180 (void) strlcpy(path
, dev
, sizeof (path
));
5181 if (dev
[0] != '/' && stat64(path
, &statbuf
) != 0) {
5184 error
= zfs_resolve_shortname(dev
, path
, MAXPATHLEN
);
5185 if (error
== 0 && zfs_dev_is_whole_disk(path
)) {
5186 if (zfs_append_partition(path
, MAXPATHLEN
) == -1)
5190 if (error
|| (stat64(path
, &statbuf
) != 0)) {
5191 (void) printf("failed to find device %s, try "
5192 "specifying absolute path instead\n", dev
);
5197 if ((fd
= open64(path
, O_RDONLY
)) < 0) {
5198 (void) printf("cannot open '%s': %s\n", path
, strerror(errno
));
5202 if (fstat64_blk(fd
, &statbuf
) != 0) {
5203 (void) printf("failed to stat '%s': %s\n", path
,
5209 if (S_ISBLK(statbuf
.st_mode
) && zfs_dev_flush(fd
) != 0)
5210 (void) printf("failed to invalidate cache '%s' : %s\n", path
,
5213 avl_create(&config_tree
, cksum_record_compare
,
5214 sizeof (cksum_record_t
), offsetof(cksum_record_t
, link
));
5215 avl_create(&uberblock_tree
, cksum_record_compare
,
5216 sizeof (cksum_record_t
), offsetof(cksum_record_t
, link
));
5218 psize
= statbuf
.st_size
;
5219 psize
= P2ALIGN(psize
, (uint64_t)sizeof (vdev_label_t
));
5220 ashift
= SPA_MINBLOCKSHIFT
;
5223 * 1. Read the label from disk
5224 * 2. Verify label cksum
5225 * 3. Unpack the configuration and insert in config tree.
5226 * 4. Traverse all uberblocks and insert in uberblock tree.
5228 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
5229 zdb_label_t
*label
= &labels
[l
];
5230 char *buf
= label
->label
.vl_vdev_phys
.vp_nvlist
;
5231 size_t buflen
= sizeof (label
->label
.vl_vdev_phys
.vp_nvlist
);
5233 cksum_record_t
*rec
;
5237 label
->label_offset
= vdev_label_offset(psize
, l
, 0);
5239 if (pread64(fd
, &label
->label
, sizeof (label
->label
),
5240 label
->label_offset
) != sizeof (label
->label
)) {
5242 (void) printf("failed to read label %d\n", l
);
5243 label
->read_failed
= B_TRUE
;
5248 label
->read_failed
= B_FALSE
;
5249 label
->cksum_valid
= label_cksum_valid(&label
->label
,
5250 label
->label_offset
);
5252 if (nvlist_unpack(buf
, buflen
, &config
, 0) == 0) {
5253 nvlist_t
*vdev_tree
= NULL
;
5256 if ((nvlist_lookup_nvlist(config
,
5257 ZPOOL_CONFIG_VDEV_TREE
, &vdev_tree
) != 0) ||
5258 (nvlist_lookup_uint64(vdev_tree
,
5259 ZPOOL_CONFIG_ASHIFT
, &ashift
) != 0))
5260 ashift
= SPA_MINBLOCKSHIFT
;
5262 if (nvlist_size(config
, &size
, NV_ENCODE_XDR
) != 0)
5265 /* If the device is a cache device read the header. */
5266 if (!read_l2arc_header
) {
5267 if (nvlist_lookup_uint64(config
,
5268 ZPOOL_CONFIG_POOL_STATE
, &l2cache
) == 0 &&
5269 l2cache
== POOL_STATE_L2CACHE
) {
5270 read_l2arc_header
= B_TRUE
;
5274 fletcher_4_native_varsize(buf
, size
, &cksum
);
5275 rec
= cksum_record_insert(&config_tree
, &cksum
, l
);
5277 label
->config
= rec
;
5278 label
->config_nv
= config
;
5279 config_found
= B_TRUE
;
5284 vd
.vdev_ashift
= ashift
;
5287 for (int i
= 0; i
< VDEV_UBERBLOCK_COUNT(&vd
); i
++) {
5288 uint64_t uoff
= VDEV_UBERBLOCK_OFFSET(&vd
, i
);
5289 uberblock_t
*ub
= (void *)((char *)label
+ uoff
);
5291 if (uberblock_verify(ub
))
5294 fletcher_4_native_varsize(ub
, sizeof (*ub
), &cksum
);
5295 rec
= cksum_record_insert(&uberblock_tree
, &cksum
, l
);
5297 label
->uberblocks
[i
] = rec
;
5302 * Dump the label and uberblocks.
5304 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
5305 zdb_label_t
*label
= &labels
[l
];
5306 size_t buflen
= sizeof (label
->label
.vl_vdev_phys
.vp_nvlist
);
5308 if (label
->read_failed
== B_TRUE
)
5311 if (label
->config_nv
) {
5312 dump_config_from_label(label
, buflen
, l
);
5315 (void) printf("failed to unpack label %d\n", l
);
5319 dump_label_uberblocks(label
, ashift
, l
);
5321 nvlist_free(label
->config_nv
);
5325 * Dump the L2ARC header, if existent.
5327 if (read_l2arc_header
)
5328 error
|= dump_l2arc_header(fd
);
5331 while ((node
= avl_destroy_nodes(&config_tree
, &cookie
)) != NULL
)
5332 umem_free(node
, sizeof (cksum_record_t
));
5335 while ((node
= avl_destroy_nodes(&uberblock_tree
, &cookie
)) != NULL
)
5336 umem_free(node
, sizeof (cksum_record_t
));
5338 avl_destroy(&config_tree
);
5339 avl_destroy(&uberblock_tree
);
5343 return (config_found
== B_FALSE
? 2 :
5344 (error
== B_TRUE
? 1 : 0));
5347 static uint64_t dataset_feature_count
[SPA_FEATURES
];
5348 static uint64_t global_feature_count
[SPA_FEATURES
];
5349 static uint64_t remap_deadlist_count
= 0;
5352 dump_one_objset(const char *dsname
, void *arg
)
5359 error
= open_objset(dsname
, FTAG
, &os
);
5363 for (f
= 0; f
< SPA_FEATURES
; f
++) {
5364 if (!dsl_dataset_feature_is_active(dmu_objset_ds(os
), f
))
5366 ASSERT(spa_feature_table
[f
].fi_flags
&
5367 ZFEATURE_FLAG_PER_DATASET
);
5368 dataset_feature_count
[f
]++;
5371 if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os
))) {
5372 remap_deadlist_count
++;
5375 for (dsl_bookmark_node_t
*dbn
=
5376 avl_first(&dmu_objset_ds(os
)->ds_bookmarks
); dbn
!= NULL
;
5377 dbn
= AVL_NEXT(&dmu_objset_ds(os
)->ds_bookmarks
, dbn
)) {
5378 mos_obj_refd(dbn
->dbn_phys
.zbm_redaction_obj
);
5379 if (dbn
->dbn_phys
.zbm_redaction_obj
!= 0)
5380 global_feature_count
[SPA_FEATURE_REDACTION_BOOKMARKS
]++;
5381 if (dbn
->dbn_phys
.zbm_flags
& ZBM_FLAG_HAS_FBN
)
5382 global_feature_count
[SPA_FEATURE_BOOKMARK_WRITTEN
]++;
5385 if (dsl_deadlist_is_open(&dmu_objset_ds(os
)->ds_dir
->dd_livelist
) &&
5386 !dmu_objset_is_snapshot(os
)) {
5387 global_feature_count
[SPA_FEATURE_LIVELIST
]++;
5391 close_objset(os
, FTAG
);
5392 fuid_table_destroy();
5399 #define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5400 typedef struct zdb_blkstats
{
5406 uint64_t zb_ditto_samevdev
;
5407 uint64_t zb_ditto_same_ms
;
5408 uint64_t zb_psize_histogram
[PSIZE_HISTO_SIZE
];
5412 * Extended object types to report deferred frees and dedup auto-ditto blocks.
5414 #define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0)
5415 #define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1)
5416 #define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2)
5417 #define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3)
5419 static const char *zdb_ot_extname
[] = {
5426 #define ZB_TOTAL DN_MAX_LEVELS
5427 #define SPA_MAX_FOR_16M (SPA_MAXBLOCKSHIFT+1)
5429 typedef struct zdb_brt_entry
{
5431 uint64_t zbre_refcount
;
5432 avl_node_t zbre_node
;
5435 typedef struct zdb_cb
{
5436 zdb_blkstats_t zcb_type
[ZB_TOTAL
+ 1][ZDB_OT_TOTAL
+ 1];
5437 uint64_t zcb_removing_size
;
5438 uint64_t zcb_checkpoint_size
;
5439 uint64_t zcb_dedup_asize
;
5440 uint64_t zcb_dedup_blocks
;
5441 uint64_t zcb_clone_asize
;
5442 uint64_t zcb_clone_blocks
;
5443 uint64_t zcb_psize_count
[SPA_MAX_FOR_16M
];
5444 uint64_t zcb_lsize_count
[SPA_MAX_FOR_16M
];
5445 uint64_t zcb_asize_count
[SPA_MAX_FOR_16M
];
5446 uint64_t zcb_psize_len
[SPA_MAX_FOR_16M
];
5447 uint64_t zcb_lsize_len
[SPA_MAX_FOR_16M
];
5448 uint64_t zcb_asize_len
[SPA_MAX_FOR_16M
];
5449 uint64_t zcb_psize_total
;
5450 uint64_t zcb_lsize_total
;
5451 uint64_t zcb_asize_total
;
5452 uint64_t zcb_embedded_blocks
[NUM_BP_EMBEDDED_TYPES
];
5453 uint64_t zcb_embedded_histogram
[NUM_BP_EMBEDDED_TYPES
]
5454 [BPE_PAYLOAD_SIZE
+ 1];
5456 hrtime_t zcb_lastprint
;
5457 uint64_t zcb_totalasize
;
5458 uint64_t zcb_errors
[256];
5462 uint32_t **zcb_vd_obsolete_counts
;
5464 boolean_t zcb_brt_is_active
;
5467 /* test if two DVA offsets from same vdev are within the same metaslab */
5469 same_metaslab(spa_t
*spa
, uint64_t vdev
, uint64_t off1
, uint64_t off2
)
5471 vdev_t
*vd
= vdev_lookup_top(spa
, vdev
);
5472 uint64_t ms_shift
= vd
->vdev_ms_shift
;
5474 return ((off1
>> ms_shift
) == (off2
>> ms_shift
));
5478 * Used to simplify reporting of the histogram data.
5480 typedef struct one_histo
{
5484 uint64_t cumulative
;
5488 * The number of separate histograms processed for psize, lsize and asize.
5493 * This routine will create a fixed column size output of three different
5494 * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5495 * the count, length and cumulative length of the psize, lsize and
5498 * All three types of blocks are listed on a single line
5500 * By default the table is printed in nicenumber format (e.g. 123K) but
5501 * if the '-P' parameter is specified then the full raw number (parseable)
5505 dump_size_histograms(zdb_cb_t
*zcb
)
5508 * A temporary buffer that allows us to convert a number into
5509 * a string using zdb_nicenumber to allow either raw or human
5510 * readable numbers to be output.
5515 * Define titles which are used in the headers of the tables
5516 * printed by this routine.
5518 const char blocksize_title1
[] = "block";
5519 const char blocksize_title2
[] = "size";
5520 const char count_title
[] = "Count";
5521 const char length_title
[] = "Size";
5522 const char cumulative_title
[] = "Cum.";
5525 * Setup the histogram arrays (psize, lsize, and asize).
5527 one_histo_t parm_histo
[NUM_HISTO
];
5529 parm_histo
[0].name
= "psize";
5530 parm_histo
[0].count
= zcb
->zcb_psize_count
;
5531 parm_histo
[0].len
= zcb
->zcb_psize_len
;
5532 parm_histo
[0].cumulative
= 0;
5534 parm_histo
[1].name
= "lsize";
5535 parm_histo
[1].count
= zcb
->zcb_lsize_count
;
5536 parm_histo
[1].len
= zcb
->zcb_lsize_len
;
5537 parm_histo
[1].cumulative
= 0;
5539 parm_histo
[2].name
= "asize";
5540 parm_histo
[2].count
= zcb
->zcb_asize_count
;
5541 parm_histo
[2].len
= zcb
->zcb_asize_len
;
5542 parm_histo
[2].cumulative
= 0;
5545 (void) printf("\nBlock Size Histogram\n");
5547 * Print the first line titles
5550 (void) printf("\n%s\t", blocksize_title1
);
5552 (void) printf("\n%7s ", blocksize_title1
);
5554 for (int j
= 0; j
< NUM_HISTO
; j
++) {
5555 if (dump_opt
['P']) {
5556 if (j
< NUM_HISTO
- 1) {
5557 (void) printf("%s\t\t\t", parm_histo
[j
].name
);
5559 /* Don't print trailing spaces */
5560 (void) printf(" %s", parm_histo
[j
].name
);
5563 if (j
< NUM_HISTO
- 1) {
5564 /* Left aligned strings in the output */
5565 (void) printf("%-7s ",
5566 parm_histo
[j
].name
);
5568 /* Don't print trailing spaces */
5569 (void) printf("%s", parm_histo
[j
].name
);
5573 (void) printf("\n");
5576 * Print the second line titles
5578 if (dump_opt
['P']) {
5579 (void) printf("%s\t", blocksize_title2
);
5581 (void) printf("%7s ", blocksize_title2
);
5584 for (int i
= 0; i
< NUM_HISTO
; i
++) {
5585 if (dump_opt
['P']) {
5586 (void) printf("%s\t%s\t%s\t",
5587 count_title
, length_title
, cumulative_title
);
5589 (void) printf("%7s%7s%7s",
5590 count_title
, length_title
, cumulative_title
);
5593 (void) printf("\n");
5598 for (int i
= SPA_MINBLOCKSHIFT
; i
< SPA_MAX_FOR_16M
; i
++) {
5601 * Print the first column showing the blocksize
5603 zdb_nicenum((1ULL << i
), numbuf
, sizeof (numbuf
));
5605 if (dump_opt
['P']) {
5606 printf("%s", numbuf
);
5608 printf("%7s:", numbuf
);
5612 * Print the remaining set of 3 columns per size:
5613 * for psize, lsize and asize
5615 for (int j
= 0; j
< NUM_HISTO
; j
++) {
5616 parm_histo
[j
].cumulative
+= parm_histo
[j
].len
[i
];
5618 zdb_nicenum(parm_histo
[j
].count
[i
],
5619 numbuf
, sizeof (numbuf
));
5621 (void) printf("\t%s", numbuf
);
5623 (void) printf("%7s", numbuf
);
5625 zdb_nicenum(parm_histo
[j
].len
[i
],
5626 numbuf
, sizeof (numbuf
));
5628 (void) printf("\t%s", numbuf
);
5630 (void) printf("%7s", numbuf
);
5632 zdb_nicenum(parm_histo
[j
].cumulative
,
5633 numbuf
, sizeof (numbuf
));
5635 (void) printf("\t%s", numbuf
);
5637 (void) printf("%7s", numbuf
);
5639 (void) printf("\n");
5644 zdb_count_block(zdb_cb_t
*zcb
, zilog_t
*zilog
, const blkptr_t
*bp
,
5645 dmu_object_type_t type
)
5647 uint64_t refcnt
= 0;
5650 ASSERT(type
< ZDB_OT_TOTAL
);
5652 if (zilog
&& zil_bp_tree_add(zilog
, bp
) != 0)
5655 spa_config_enter(zcb
->zcb_spa
, SCL_CONFIG
, FTAG
, RW_READER
);
5657 for (i
= 0; i
< 4; i
++) {
5658 int l
= (i
< 2) ? BP_GET_LEVEL(bp
) : ZB_TOTAL
;
5659 int t
= (i
& 1) ? type
: ZDB_OT_TOTAL
;
5661 zdb_blkstats_t
*zb
= &zcb
->zcb_type
[l
][t
];
5663 zb
->zb_asize
+= BP_GET_ASIZE(bp
);
5664 zb
->zb_lsize
+= BP_GET_LSIZE(bp
);
5665 zb
->zb_psize
+= BP_GET_PSIZE(bp
);
5669 * The histogram is only big enough to record blocks up to
5670 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
5673 unsigned idx
= BP_GET_PSIZE(bp
) >> SPA_MINBLOCKSHIFT
;
5674 idx
= MIN(idx
, SPA_OLD_MAXBLOCKSIZE
/ SPA_MINBLOCKSIZE
+ 1);
5675 zb
->zb_psize_histogram
[idx
]++;
5677 zb
->zb_gangs
+= BP_COUNT_GANG(bp
);
5679 switch (BP_GET_NDVAS(bp
)) {
5681 if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5682 DVA_GET_VDEV(&bp
->blk_dva
[1])) {
5683 zb
->zb_ditto_samevdev
++;
5685 if (same_metaslab(zcb
->zcb_spa
,
5686 DVA_GET_VDEV(&bp
->blk_dva
[0]),
5687 DVA_GET_OFFSET(&bp
->blk_dva
[0]),
5688 DVA_GET_OFFSET(&bp
->blk_dva
[1])))
5689 zb
->zb_ditto_same_ms
++;
5693 equal
= (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5694 DVA_GET_VDEV(&bp
->blk_dva
[1])) +
5695 (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5696 DVA_GET_VDEV(&bp
->blk_dva
[2])) +
5697 (DVA_GET_VDEV(&bp
->blk_dva
[1]) ==
5698 DVA_GET_VDEV(&bp
->blk_dva
[2]));
5700 zb
->zb_ditto_samevdev
++;
5702 if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5703 DVA_GET_VDEV(&bp
->blk_dva
[1]) &&
5704 same_metaslab(zcb
->zcb_spa
,
5705 DVA_GET_VDEV(&bp
->blk_dva
[0]),
5706 DVA_GET_OFFSET(&bp
->blk_dva
[0]),
5707 DVA_GET_OFFSET(&bp
->blk_dva
[1])))
5708 zb
->zb_ditto_same_ms
++;
5709 else if (DVA_GET_VDEV(&bp
->blk_dva
[0]) ==
5710 DVA_GET_VDEV(&bp
->blk_dva
[2]) &&
5711 same_metaslab(zcb
->zcb_spa
,
5712 DVA_GET_VDEV(&bp
->blk_dva
[0]),
5713 DVA_GET_OFFSET(&bp
->blk_dva
[0]),
5714 DVA_GET_OFFSET(&bp
->blk_dva
[2])))
5715 zb
->zb_ditto_same_ms
++;
5716 else if (DVA_GET_VDEV(&bp
->blk_dva
[1]) ==
5717 DVA_GET_VDEV(&bp
->blk_dva
[2]) &&
5718 same_metaslab(zcb
->zcb_spa
,
5719 DVA_GET_VDEV(&bp
->blk_dva
[1]),
5720 DVA_GET_OFFSET(&bp
->blk_dva
[1]),
5721 DVA_GET_OFFSET(&bp
->blk_dva
[2])))
5722 zb
->zb_ditto_same_ms
++;
5728 spa_config_exit(zcb
->zcb_spa
, SCL_CONFIG
, FTAG
);
5730 if (BP_IS_EMBEDDED(bp
)) {
5731 zcb
->zcb_embedded_blocks
[BPE_GET_ETYPE(bp
)]++;
5732 zcb
->zcb_embedded_histogram
[BPE_GET_ETYPE(bp
)]
5733 [BPE_GET_PSIZE(bp
)]++;
5737 * The binning histogram bins by powers of two up to
5738 * SPA_MAXBLOCKSIZE rather than creating bins for
5739 * every possible blocksize found in the pool.
5741 int bin
= highbit64(BP_GET_PSIZE(bp
)) - 1;
5743 zcb
->zcb_psize_count
[bin
]++;
5744 zcb
->zcb_psize_len
[bin
] += BP_GET_PSIZE(bp
);
5745 zcb
->zcb_psize_total
+= BP_GET_PSIZE(bp
);
5747 bin
= highbit64(BP_GET_LSIZE(bp
)) - 1;
5749 zcb
->zcb_lsize_count
[bin
]++;
5750 zcb
->zcb_lsize_len
[bin
] += BP_GET_LSIZE(bp
);
5751 zcb
->zcb_lsize_total
+= BP_GET_LSIZE(bp
);
5753 bin
= highbit64(BP_GET_ASIZE(bp
)) - 1;
5755 zcb
->zcb_asize_count
[bin
]++;
5756 zcb
->zcb_asize_len
[bin
] += BP_GET_ASIZE(bp
);
5757 zcb
->zcb_asize_total
+= BP_GET_ASIZE(bp
);
5759 if (zcb
->zcb_brt_is_active
&& brt_maybe_exists(zcb
->zcb_spa
, bp
)) {
5761 * Cloned blocks are special. We need to count them, so we can
5762 * later uncount them when reporting leaked space, and we must
5763 * only claim them them once.
5765 * To do this, we keep our own in-memory BRT. For each block
5766 * we haven't seen before, we look it up in the real BRT and
5767 * if its there, we note it and its refcount then proceed as
5768 * normal. If we see the block again, we count it as a clone
5769 * and then give it no further consideration.
5771 zdb_brt_entry_t zbre_search
, *zbre
;
5774 zbre_search
.zbre_dva
= bp
->blk_dva
[0];
5775 zbre
= avl_find(&zcb
->zcb_brt
, &zbre_search
, &where
);
5777 zcb
->zcb_clone_asize
+= BP_GET_ASIZE(bp
);
5778 zcb
->zcb_clone_blocks
++;
5780 zbre
->zbre_refcount
--;
5781 if (zbre
->zbre_refcount
== 0) {
5782 avl_remove(&zcb
->zcb_brt
, zbre
);
5783 umem_free(zbre
, sizeof (zdb_brt_entry_t
));
5788 uint64_t crefcnt
= brt_entry_get_refcount(zcb
->zcb_spa
, bp
);
5790 zbre
= umem_zalloc(sizeof (zdb_brt_entry_t
),
5792 zbre
->zbre_dva
= bp
->blk_dva
[0];
5793 zbre
->zbre_refcount
= crefcnt
;
5794 avl_insert(&zcb
->zcb_brt
, zbre
, where
);
5801 if (BP_GET_DEDUP(bp
)) {
5805 ddt
= ddt_select(zcb
->zcb_spa
, bp
);
5807 dde
= ddt_lookup(ddt
, bp
, B_FALSE
);
5812 ddt_phys_t
*ddp
= ddt_phys_select(dde
, bp
);
5813 ddt_phys_decref(ddp
);
5814 refcnt
= ddp
->ddp_refcnt
;
5815 if (ddt_phys_total_refcnt(dde
) == 0)
5816 ddt_remove(ddt
, dde
);
5821 VERIFY3U(zio_wait(zio_claim(NULL
, zcb
->zcb_spa
,
5822 refcnt
? 0 : spa_min_claim_txg(zcb
->zcb_spa
),
5823 bp
, NULL
, NULL
, ZIO_FLAG_CANFAIL
)), ==, 0);
5827 zdb_blkptr_done(zio_t
*zio
)
5829 spa_t
*spa
= zio
->io_spa
;
5830 blkptr_t
*bp
= zio
->io_bp
;
5831 int ioerr
= zio
->io_error
;
5832 zdb_cb_t
*zcb
= zio
->io_private
;
5833 zbookmark_phys_t
*zb
= &zio
->io_bookmark
;
5835 mutex_enter(&spa
->spa_scrub_lock
);
5836 spa
->spa_load_verify_bytes
-= BP_GET_PSIZE(bp
);
5837 cv_broadcast(&spa
->spa_scrub_io_cv
);
5839 if (ioerr
&& !(zio
->io_flags
& ZIO_FLAG_SPECULATIVE
)) {
5840 char blkbuf
[BP_SPRINTF_LEN
];
5842 zcb
->zcb_haderrors
= 1;
5843 zcb
->zcb_errors
[ioerr
]++;
5845 if (dump_opt
['b'] >= 2)
5846 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
5850 (void) printf("zdb_blkptr_cb: "
5851 "Got error %d reading "
5852 "<%llu, %llu, %lld, %llx> %s -- skipping\n",
5854 (u_longlong_t
)zb
->zb_objset
,
5855 (u_longlong_t
)zb
->zb_object
,
5856 (u_longlong_t
)zb
->zb_level
,
5857 (u_longlong_t
)zb
->zb_blkid
,
5860 mutex_exit(&spa
->spa_scrub_lock
);
5862 abd_free(zio
->io_abd
);
5866 zdb_blkptr_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
5867 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
5869 zdb_cb_t
*zcb
= arg
;
5870 dmu_object_type_t type
;
5871 boolean_t is_metadata
;
5873 if (zb
->zb_level
== ZB_DNODE_LEVEL
)
5876 if (dump_opt
['b'] >= 5 && bp
->blk_birth
> 0) {
5877 char blkbuf
[BP_SPRINTF_LEN
];
5878 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
5879 (void) printf("objset %llu object %llu "
5880 "level %lld offset 0x%llx %s\n",
5881 (u_longlong_t
)zb
->zb_objset
,
5882 (u_longlong_t
)zb
->zb_object
,
5883 (longlong_t
)zb
->zb_level
,
5884 (u_longlong_t
)blkid2offset(dnp
, bp
, zb
),
5888 if (BP_IS_HOLE(bp
) || BP_IS_REDACTED(bp
))
5891 type
= BP_GET_TYPE(bp
);
5893 zdb_count_block(zcb
, zilog
, bp
,
5894 (type
& DMU_OT_NEWTYPE
) ? ZDB_OT_OTHER
: type
);
5896 is_metadata
= (BP_GET_LEVEL(bp
) != 0 || DMU_OT_IS_METADATA(type
));
5898 if (!BP_IS_EMBEDDED(bp
) &&
5899 (dump_opt
['c'] > 1 || (dump_opt
['c'] && is_metadata
))) {
5900 size_t size
= BP_GET_PSIZE(bp
);
5901 abd_t
*abd
= abd_alloc(size
, B_FALSE
);
5902 int flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_SCRUB
| ZIO_FLAG_RAW
;
5904 /* If it's an intent log block, failure is expected. */
5905 if (zb
->zb_level
== ZB_ZIL_LEVEL
)
5906 flags
|= ZIO_FLAG_SPECULATIVE
;
5908 mutex_enter(&spa
->spa_scrub_lock
);
5909 while (spa
->spa_load_verify_bytes
> max_inflight_bytes
)
5910 cv_wait(&spa
->spa_scrub_io_cv
, &spa
->spa_scrub_lock
);
5911 spa
->spa_load_verify_bytes
+= size
;
5912 mutex_exit(&spa
->spa_scrub_lock
);
5914 zio_nowait(zio_read(NULL
, spa
, bp
, abd
, size
,
5915 zdb_blkptr_done
, zcb
, ZIO_PRIORITY_ASYNC_READ
, flags
, zb
));
5918 zcb
->zcb_readfails
= 0;
5920 /* only call gethrtime() every 100 blocks */
5927 if (dump_opt
['b'] < 5 && gethrtime() > zcb
->zcb_lastprint
+ NANOSEC
) {
5928 uint64_t now
= gethrtime();
5930 uint64_t bytes
= zcb
->zcb_type
[ZB_TOTAL
][ZDB_OT_TOTAL
].zb_asize
;
5931 uint64_t kb_per_sec
=
5932 1 + bytes
/ (1 + ((now
- zcb
->zcb_start
) / 1000 / 1000));
5933 uint64_t sec_remaining
=
5934 (zcb
->zcb_totalasize
- bytes
) / 1024 / kb_per_sec
;
5936 /* make sure nicenum has enough space */
5937 _Static_assert(sizeof (buf
) >= NN_NUMBUF_SZ
, "buf truncated");
5939 zfs_nicebytes(bytes
, buf
, sizeof (buf
));
5940 (void) fprintf(stderr
,
5941 "\r%5s completed (%4"PRIu64
"MB/s) "
5942 "estimated time remaining: "
5943 "%"PRIu64
"hr %02"PRIu64
"min %02"PRIu64
"sec ",
5944 buf
, kb_per_sec
/ 1024,
5945 sec_remaining
/ 60 / 60,
5946 sec_remaining
/ 60 % 60,
5947 sec_remaining
% 60);
5949 zcb
->zcb_lastprint
= now
;
5956 zdb_leak(void *arg
, uint64_t start
, uint64_t size
)
5960 (void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
5961 (u_longlong_t
)vd
->vdev_id
, (u_longlong_t
)start
, (u_longlong_t
)size
);
5964 static metaslab_ops_t zdb_metaslab_ops
= {
5969 load_unflushed_svr_segs_cb(spa_t
*spa
, space_map_entry_t
*sme
,
5970 uint64_t txg
, void *arg
)
5972 spa_vdev_removal_t
*svr
= arg
;
5974 uint64_t offset
= sme
->sme_offset
;
5975 uint64_t size
= sme
->sme_run
;
5977 /* skip vdevs we don't care about */
5978 if (sme
->sme_vdev
!= svr
->svr_vdev_id
)
5981 vdev_t
*vd
= vdev_lookup_top(spa
, sme
->sme_vdev
);
5982 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
5983 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
5985 if (txg
< metaslab_unflushed_txg(ms
))
5988 if (sme
->sme_type
== SM_ALLOC
)
5989 range_tree_add(svr
->svr_allocd_segs
, offset
, size
);
5991 range_tree_remove(svr
->svr_allocd_segs
, offset
, size
);
5997 claim_segment_impl_cb(uint64_t inner_offset
, vdev_t
*vd
, uint64_t offset
,
5998 uint64_t size
, void *arg
)
6000 (void) inner_offset
, (void) arg
;
6003 * This callback was called through a remap from
6004 * a device being removed. Therefore, the vdev that
6005 * this callback is applied to is a concrete
6008 ASSERT(vdev_is_concrete(vd
));
6010 VERIFY0(metaslab_claim_impl(vd
, offset
, size
,
6011 spa_min_claim_txg(vd
->vdev_spa
)));
6015 claim_segment_cb(void *arg
, uint64_t offset
, uint64_t size
)
6019 vdev_indirect_ops
.vdev_op_remap(vd
, offset
, size
,
6020 claim_segment_impl_cb
, NULL
);
6024 * After accounting for all allocated blocks that are directly referenced,
6025 * we might have missed a reference to a block from a partially complete
6026 * (and thus unused) indirect mapping object. We perform a secondary pass
6027 * through the metaslabs we have already mapped and claim the destination
6031 zdb_claim_removing(spa_t
*spa
, zdb_cb_t
*zcb
)
6036 if (spa
->spa_vdev_removal
== NULL
)
6039 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6041 spa_vdev_removal_t
*svr
= spa
->spa_vdev_removal
;
6042 vdev_t
*vd
= vdev_lookup_top(spa
, svr
->svr_vdev_id
);
6043 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6045 ASSERT0(range_tree_space(svr
->svr_allocd_segs
));
6047 range_tree_t
*allocs
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
6048 for (uint64_t msi
= 0; msi
< vd
->vdev_ms_count
; msi
++) {
6049 metaslab_t
*msp
= vd
->vdev_ms
[msi
];
6051 ASSERT0(range_tree_space(allocs
));
6052 if (msp
->ms_sm
!= NULL
)
6053 VERIFY0(space_map_load(msp
->ms_sm
, allocs
, SM_ALLOC
));
6054 range_tree_vacate(allocs
, range_tree_add
, svr
->svr_allocd_segs
);
6056 range_tree_destroy(allocs
);
6058 iterate_through_spacemap_logs(spa
, load_unflushed_svr_segs_cb
, svr
);
6061 * Clear everything past what has been synced,
6062 * because we have not allocated mappings for
6065 range_tree_clear(svr
->svr_allocd_segs
,
6066 vdev_indirect_mapping_max_offset(vim
),
6067 vd
->vdev_asize
- vdev_indirect_mapping_max_offset(vim
));
6069 zcb
->zcb_removing_size
+= range_tree_space(svr
->svr_allocd_segs
);
6070 range_tree_vacate(svr
->svr_allocd_segs
, claim_segment_cb
, vd
);
6072 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6076 increment_indirect_mapping_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
6080 zdb_cb_t
*zcb
= arg
;
6081 spa_t
*spa
= zcb
->zcb_spa
;
6083 const dva_t
*dva
= &bp
->blk_dva
[0];
6086 ASSERT(!dump_opt
['L']);
6087 ASSERT3U(BP_GET_NDVAS(bp
), ==, 1);
6089 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
6090 vd
= vdev_lookup_top(zcb
->zcb_spa
, DVA_GET_VDEV(dva
));
6091 ASSERT3P(vd
, !=, NULL
);
6092 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
6094 ASSERT(vd
->vdev_indirect_config
.vic_mapping_object
!= 0);
6095 ASSERT3P(zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
], !=, NULL
);
6097 vdev_indirect_mapping_increment_obsolete_count(
6098 vd
->vdev_indirect_mapping
,
6099 DVA_GET_OFFSET(dva
), DVA_GET_ASIZE(dva
),
6100 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
]);
6106 zdb_load_obsolete_counts(vdev_t
*vd
)
6108 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6109 spa_t
*spa
= vd
->vdev_spa
;
6110 spa_condensing_indirect_phys_t
*scip
=
6111 &spa
->spa_condensing_indirect_phys
;
6112 uint64_t obsolete_sm_object
;
6115 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
6116 EQUIV(obsolete_sm_object
!= 0, vd
->vdev_obsolete_sm
!= NULL
);
6117 counts
= vdev_indirect_mapping_load_obsolete_counts(vim
);
6118 if (vd
->vdev_obsolete_sm
!= NULL
) {
6119 vdev_indirect_mapping_load_obsolete_spacemap(vim
, counts
,
6120 vd
->vdev_obsolete_sm
);
6122 if (scip
->scip_vdev
== vd
->vdev_id
&&
6123 scip
->scip_prev_obsolete_sm_object
!= 0) {
6124 space_map_t
*prev_obsolete_sm
= NULL
;
6125 VERIFY0(space_map_open(&prev_obsolete_sm
, spa
->spa_meta_objset
,
6126 scip
->scip_prev_obsolete_sm_object
, 0, vd
->vdev_asize
, 0));
6127 vdev_indirect_mapping_load_obsolete_spacemap(vim
, counts
,
6129 space_map_close(prev_obsolete_sm
);
6135 zdb_ddt_leak_init(spa_t
*spa
, zdb_cb_t
*zcb
)
6137 ddt_bookmark_t ddb
= {0};
6142 ASSERT(!dump_opt
['L']);
6144 while ((error
= ddt_walk(spa
, &ddb
, &dde
)) == 0) {
6146 ddt_phys_t
*ddp
= dde
.dde_phys
;
6148 if (ddb
.ddb_class
== DDT_CLASS_UNIQUE
)
6151 ASSERT(ddt_phys_total_refcnt(&dde
) > 1);
6153 for (p
= 0; p
< DDT_PHYS_TYPES
; p
++, ddp
++) {
6154 if (ddp
->ddp_phys_birth
== 0)
6156 ddt_bp_create(ddb
.ddb_checksum
,
6157 &dde
.dde_key
, ddp
, &blk
);
6158 if (p
== DDT_PHYS_DITTO
) {
6159 zdb_count_block(zcb
, NULL
, &blk
, ZDB_OT_DITTO
);
6161 zcb
->zcb_dedup_asize
+=
6162 BP_GET_ASIZE(&blk
) * (ddp
->ddp_refcnt
- 1);
6163 zcb
->zcb_dedup_blocks
++;
6166 ddt_t
*ddt
= spa
->spa_ddt
[ddb
.ddb_checksum
];
6168 VERIFY(ddt_lookup(ddt
, &blk
, B_TRUE
) != NULL
);
6172 ASSERT(error
== ENOENT
);
6175 typedef struct checkpoint_sm_exclude_entry_arg
{
6177 uint64_t cseea_checkpoint_size
;
6178 } checkpoint_sm_exclude_entry_arg_t
;
6181 checkpoint_sm_exclude_entry_cb(space_map_entry_t
*sme
, void *arg
)
6183 checkpoint_sm_exclude_entry_arg_t
*cseea
= arg
;
6184 vdev_t
*vd
= cseea
->cseea_vd
;
6185 metaslab_t
*ms
= vd
->vdev_ms
[sme
->sme_offset
>> vd
->vdev_ms_shift
];
6186 uint64_t end
= sme
->sme_offset
+ sme
->sme_run
;
6188 ASSERT(sme
->sme_type
== SM_FREE
);
6191 * Since the vdev_checkpoint_sm exists in the vdev level
6192 * and the ms_sm space maps exist in the metaslab level,
6193 * an entry in the checkpoint space map could theoretically
6194 * cross the boundaries of the metaslab that it belongs.
6196 * In reality, because of the way that we populate and
6197 * manipulate the checkpoint's space maps currently,
6198 * there shouldn't be any entries that cross metaslabs.
6199 * Hence the assertion below.
6201 * That said, there is no fundamental requirement that
6202 * the checkpoint's space map entries should not cross
6203 * metaslab boundaries. So if needed we could add code
6204 * that handles metaslab-crossing segments in the future.
6206 VERIFY3U(sme
->sme_offset
, >=, ms
->ms_start
);
6207 VERIFY3U(end
, <=, ms
->ms_start
+ ms
->ms_size
);
6210 * By removing the entry from the allocated segments we
6211 * also verify that the entry is there to begin with.
6213 mutex_enter(&ms
->ms_lock
);
6214 range_tree_remove(ms
->ms_allocatable
, sme
->sme_offset
, sme
->sme_run
);
6215 mutex_exit(&ms
->ms_lock
);
6217 cseea
->cseea_checkpoint_size
+= sme
->sme_run
;
6222 zdb_leak_init_vdev_exclude_checkpoint(vdev_t
*vd
, zdb_cb_t
*zcb
)
6224 spa_t
*spa
= vd
->vdev_spa
;
6225 space_map_t
*checkpoint_sm
= NULL
;
6226 uint64_t checkpoint_sm_obj
;
6229 * If there is no vdev_top_zap, we are in a pool whose
6230 * version predates the pool checkpoint feature.
6232 if (vd
->vdev_top_zap
== 0)
6236 * If there is no reference of the vdev_checkpoint_sm in
6237 * the vdev_top_zap, then one of the following scenarios
6240 * 1] There is no checkpoint
6241 * 2] There is a checkpoint, but no checkpointed blocks
6242 * have been freed yet
6243 * 3] The current vdev is indirect
6245 * In these cases we return immediately.
6247 if (zap_contains(spa_meta_objset(spa
), vd
->vdev_top_zap
,
6248 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) != 0)
6251 VERIFY0(zap_lookup(spa_meta_objset(spa
), vd
->vdev_top_zap
,
6252 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1,
6253 &checkpoint_sm_obj
));
6255 checkpoint_sm_exclude_entry_arg_t cseea
;
6256 cseea
.cseea_vd
= vd
;
6257 cseea
.cseea_checkpoint_size
= 0;
6259 VERIFY0(space_map_open(&checkpoint_sm
, spa_meta_objset(spa
),
6260 checkpoint_sm_obj
, 0, vd
->vdev_asize
, vd
->vdev_ashift
));
6262 VERIFY0(space_map_iterate(checkpoint_sm
,
6263 space_map_length(checkpoint_sm
),
6264 checkpoint_sm_exclude_entry_cb
, &cseea
));
6265 space_map_close(checkpoint_sm
);
6267 zcb
->zcb_checkpoint_size
+= cseea
.cseea_checkpoint_size
;
6271 zdb_leak_init_exclude_checkpoint(spa_t
*spa
, zdb_cb_t
*zcb
)
6273 ASSERT(!dump_opt
['L']);
6275 vdev_t
*rvd
= spa
->spa_root_vdev
;
6276 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
6277 ASSERT3U(c
, ==, rvd
->vdev_child
[c
]->vdev_id
);
6278 zdb_leak_init_vdev_exclude_checkpoint(rvd
->vdev_child
[c
], zcb
);
6283 count_unflushed_space_cb(spa_t
*spa
, space_map_entry_t
*sme
,
6284 uint64_t txg
, void *arg
)
6286 int64_t *ualloc_space
= arg
;
6288 uint64_t offset
= sme
->sme_offset
;
6289 uint64_t vdev_id
= sme
->sme_vdev
;
6291 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
6292 if (!vdev_is_concrete(vd
))
6295 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
6296 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
6298 if (txg
< metaslab_unflushed_txg(ms
))
6301 if (sme
->sme_type
== SM_ALLOC
)
6302 *ualloc_space
+= sme
->sme_run
;
6304 *ualloc_space
-= sme
->sme_run
;
6310 get_unflushed_alloc_space(spa_t
*spa
)
6315 int64_t ualloc_space
= 0;
6316 iterate_through_spacemap_logs(spa
, count_unflushed_space_cb
,
6318 return (ualloc_space
);
6322 load_unflushed_cb(spa_t
*spa
, space_map_entry_t
*sme
, uint64_t txg
, void *arg
)
6324 maptype_t
*uic_maptype
= arg
;
6326 uint64_t offset
= sme
->sme_offset
;
6327 uint64_t size
= sme
->sme_run
;
6328 uint64_t vdev_id
= sme
->sme_vdev
;
6330 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
6332 /* skip indirect vdevs */
6333 if (!vdev_is_concrete(vd
))
6336 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
6338 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
6339 ASSERT(*uic_maptype
== SM_ALLOC
|| *uic_maptype
== SM_FREE
);
6341 if (txg
< metaslab_unflushed_txg(ms
))
6344 if (*uic_maptype
== sme
->sme_type
)
6345 range_tree_add(ms
->ms_allocatable
, offset
, size
);
6347 range_tree_remove(ms
->ms_allocatable
, offset
, size
);
6353 load_unflushed_to_ms_allocatables(spa_t
*spa
, maptype_t maptype
)
6355 iterate_through_spacemap_logs(spa
, load_unflushed_cb
, &maptype
);
6359 load_concrete_ms_allocatable_trees(spa_t
*spa
, maptype_t maptype
)
6361 vdev_t
*rvd
= spa
->spa_root_vdev
;
6362 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
6363 vdev_t
*vd
= rvd
->vdev_child
[i
];
6365 ASSERT3U(i
, ==, vd
->vdev_id
);
6367 if (vd
->vdev_ops
== &vdev_indirect_ops
)
6370 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
6371 metaslab_t
*msp
= vd
->vdev_ms
[m
];
6373 (void) fprintf(stderr
,
6374 "\rloading concrete vdev %llu, "
6375 "metaslab %llu of %llu ...",
6376 (longlong_t
)vd
->vdev_id
,
6377 (longlong_t
)msp
->ms_id
,
6378 (longlong_t
)vd
->vdev_ms_count
);
6380 mutex_enter(&msp
->ms_lock
);
6381 range_tree_vacate(msp
->ms_allocatable
, NULL
, NULL
);
6384 * We don't want to spend the CPU manipulating the
6385 * size-ordered tree, so clear the range_tree ops.
6387 msp
->ms_allocatable
->rt_ops
= NULL
;
6389 if (msp
->ms_sm
!= NULL
) {
6390 VERIFY0(space_map_load(msp
->ms_sm
,
6391 msp
->ms_allocatable
, maptype
));
6393 if (!msp
->ms_loaded
)
6394 msp
->ms_loaded
= B_TRUE
;
6395 mutex_exit(&msp
->ms_lock
);
6399 load_unflushed_to_ms_allocatables(spa
, maptype
);
6403 * vm_idxp is an in-out parameter which (for indirect vdevs) is the
6404 * index in vim_entries that has the first entry in this metaslab.
6405 * On return, it will be set to the first entry after this metaslab.
6408 load_indirect_ms_allocatable_tree(vdev_t
*vd
, metaslab_t
*msp
,
6411 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6413 mutex_enter(&msp
->ms_lock
);
6414 range_tree_vacate(msp
->ms_allocatable
, NULL
, NULL
);
6417 * We don't want to spend the CPU manipulating the
6418 * size-ordered tree, so clear the range_tree ops.
6420 msp
->ms_allocatable
->rt_ops
= NULL
;
6422 for (; *vim_idxp
< vdev_indirect_mapping_num_entries(vim
);
6424 vdev_indirect_mapping_entry_phys_t
*vimep
=
6425 &vim
->vim_entries
[*vim_idxp
];
6426 uint64_t ent_offset
= DVA_MAPPING_GET_SRC_OFFSET(vimep
);
6427 uint64_t ent_len
= DVA_GET_ASIZE(&vimep
->vimep_dst
);
6428 ASSERT3U(ent_offset
, >=, msp
->ms_start
);
6429 if (ent_offset
>= msp
->ms_start
+ msp
->ms_size
)
6433 * Mappings do not cross metaslab boundaries,
6434 * because we create them by walking the metaslabs.
6436 ASSERT3U(ent_offset
+ ent_len
, <=,
6437 msp
->ms_start
+ msp
->ms_size
);
6438 range_tree_add(msp
->ms_allocatable
, ent_offset
, ent_len
);
6441 if (!msp
->ms_loaded
)
6442 msp
->ms_loaded
= B_TRUE
;
6443 mutex_exit(&msp
->ms_lock
);
6447 zdb_leak_init_prepare_indirect_vdevs(spa_t
*spa
, zdb_cb_t
*zcb
)
6449 ASSERT(!dump_opt
['L']);
6451 vdev_t
*rvd
= spa
->spa_root_vdev
;
6452 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
6453 vdev_t
*vd
= rvd
->vdev_child
[c
];
6455 ASSERT3U(c
, ==, vd
->vdev_id
);
6457 if (vd
->vdev_ops
!= &vdev_indirect_ops
)
6461 * Note: we don't check for mapping leaks on
6462 * removing vdevs because their ms_allocatable's
6463 * are used to look for leaks in allocated space.
6465 zcb
->zcb_vd_obsolete_counts
[c
] = zdb_load_obsolete_counts(vd
);
6468 * Normally, indirect vdevs don't have any
6469 * metaslabs. We want to set them up for
6472 vdev_metaslab_group_create(vd
);
6473 VERIFY0(vdev_metaslab_init(vd
, 0));
6475 vdev_indirect_mapping_t
*vim __maybe_unused
=
6476 vd
->vdev_indirect_mapping
;
6477 uint64_t vim_idx
= 0;
6478 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
6480 (void) fprintf(stderr
,
6481 "\rloading indirect vdev %llu, "
6482 "metaslab %llu of %llu ...",
6483 (longlong_t
)vd
->vdev_id
,
6484 (longlong_t
)vd
->vdev_ms
[m
]->ms_id
,
6485 (longlong_t
)vd
->vdev_ms_count
);
6487 load_indirect_ms_allocatable_tree(vd
, vd
->vdev_ms
[m
],
6490 ASSERT3U(vim_idx
, ==, vdev_indirect_mapping_num_entries(vim
));
6495 zdb_leak_init(spa_t
*spa
, zdb_cb_t
*zcb
)
6502 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
6503 vdev_t
*rvd
= spa
->spa_root_vdev
;
6506 * We are going to be changing the meaning of the metaslab's
6507 * ms_allocatable. Ensure that the allocator doesn't try to
6510 spa
->spa_normal_class
->mc_ops
= &zdb_metaslab_ops
;
6511 spa
->spa_log_class
->mc_ops
= &zdb_metaslab_ops
;
6512 spa
->spa_embedded_log_class
->mc_ops
= &zdb_metaslab_ops
;
6514 zcb
->zcb_vd_obsolete_counts
=
6515 umem_zalloc(rvd
->vdev_children
* sizeof (uint32_t *),
6519 * For leak detection, we overload the ms_allocatable trees
6520 * to contain allocated segments instead of free segments.
6521 * As a result, we can't use the normal metaslab_load/unload
6524 zdb_leak_init_prepare_indirect_vdevs(spa
, zcb
);
6525 load_concrete_ms_allocatable_trees(spa
, SM_ALLOC
);
6528 * On load_concrete_ms_allocatable_trees() we loaded all the
6529 * allocated entries from the ms_sm to the ms_allocatable for
6530 * each metaslab. If the pool has a checkpoint or is in the
6531 * middle of discarding a checkpoint, some of these blocks
6532 * may have been freed but their ms_sm may not have been
6533 * updated because they are referenced by the checkpoint. In
6534 * order to avoid false-positives during leak-detection, we
6535 * go through the vdev's checkpoint space map and exclude all
6536 * its entries from their relevant ms_allocatable.
6538 * We also aggregate the space held by the checkpoint and add
6539 * it to zcb_checkpoint_size.
6541 * Note that at this point we are also verifying that all the
6542 * entries on the checkpoint_sm are marked as allocated in
6543 * the ms_sm of their relevant metaslab.
6544 * [see comment in checkpoint_sm_exclude_entry_cb()]
6546 zdb_leak_init_exclude_checkpoint(spa
, zcb
);
6547 ASSERT3U(zcb
->zcb_checkpoint_size
, ==, spa_get_checkpoint_space(spa
));
6549 /* for cleaner progress output */
6550 (void) fprintf(stderr
, "\n");
6552 if (bpobj_is_open(&dp
->dp_obsolete_bpobj
)) {
6553 ASSERT(spa_feature_is_enabled(spa
,
6554 SPA_FEATURE_DEVICE_REMOVAL
));
6555 (void) bpobj_iterate_nofree(&dp
->dp_obsolete_bpobj
,
6556 increment_indirect_mapping_cb
, zcb
, NULL
);
6559 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
6560 zdb_ddt_leak_init(spa
, zcb
);
6561 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
6565 zdb_check_for_obsolete_leaks(vdev_t
*vd
, zdb_cb_t
*zcb
)
6567 boolean_t leaks
= B_FALSE
;
6568 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
6569 uint64_t total_leaked
= 0;
6570 boolean_t are_precise
= B_FALSE
;
6572 ASSERT(vim
!= NULL
);
6574 for (uint64_t i
= 0; i
< vdev_indirect_mapping_num_entries(vim
); i
++) {
6575 vdev_indirect_mapping_entry_phys_t
*vimep
=
6576 &vim
->vim_entries
[i
];
6577 uint64_t obsolete_bytes
= 0;
6578 uint64_t offset
= DVA_MAPPING_GET_SRC_OFFSET(vimep
);
6579 metaslab_t
*msp
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
6582 * This is not very efficient but it's easy to
6583 * verify correctness.
6585 for (uint64_t inner_offset
= 0;
6586 inner_offset
< DVA_GET_ASIZE(&vimep
->vimep_dst
);
6587 inner_offset
+= 1ULL << vd
->vdev_ashift
) {
6588 if (range_tree_contains(msp
->ms_allocatable
,
6589 offset
+ inner_offset
, 1ULL << vd
->vdev_ashift
)) {
6590 obsolete_bytes
+= 1ULL << vd
->vdev_ashift
;
6594 int64_t bytes_leaked
= obsolete_bytes
-
6595 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
][i
];
6596 ASSERT3U(DVA_GET_ASIZE(&vimep
->vimep_dst
), >=,
6597 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
][i
]);
6599 VERIFY0(vdev_obsolete_counts_are_precise(vd
, &are_precise
));
6600 if (bytes_leaked
!= 0 && (are_precise
|| dump_opt
['d'] >= 5)) {
6601 (void) printf("obsolete indirect mapping count "
6602 "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
6603 (u_longlong_t
)vd
->vdev_id
,
6604 (u_longlong_t
)DVA_MAPPING_GET_SRC_OFFSET(vimep
),
6605 (u_longlong_t
)DVA_GET_ASIZE(&vimep
->vimep_dst
),
6606 (u_longlong_t
)bytes_leaked
);
6608 total_leaked
+= ABS(bytes_leaked
);
6611 VERIFY0(vdev_obsolete_counts_are_precise(vd
, &are_precise
));
6612 if (!are_precise
&& total_leaked
> 0) {
6613 int pct_leaked
= total_leaked
* 100 /
6614 vdev_indirect_mapping_bytes_mapped(vim
);
6615 (void) printf("cannot verify obsolete indirect mapping "
6616 "counts of vdev %llu because precise feature was not "
6617 "enabled when it was removed: %d%% (%llx bytes) of mapping"
6619 (u_longlong_t
)vd
->vdev_id
, pct_leaked
,
6620 (u_longlong_t
)total_leaked
);
6621 } else if (total_leaked
> 0) {
6622 (void) printf("obsolete indirect mapping count mismatch "
6623 "for vdev %llu -- %llx total bytes mismatched\n",
6624 (u_longlong_t
)vd
->vdev_id
,
6625 (u_longlong_t
)total_leaked
);
6629 vdev_indirect_mapping_free_obsolete_counts(vim
,
6630 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
]);
6631 zcb
->zcb_vd_obsolete_counts
[vd
->vdev_id
] = NULL
;
6637 zdb_leak_fini(spa_t
*spa
, zdb_cb_t
*zcb
)
6642 boolean_t leaks
= B_FALSE
;
6643 vdev_t
*rvd
= spa
->spa_root_vdev
;
6644 for (unsigned c
= 0; c
< rvd
->vdev_children
; c
++) {
6645 vdev_t
*vd
= rvd
->vdev_child
[c
];
6647 if (zcb
->zcb_vd_obsolete_counts
[c
] != NULL
) {
6648 leaks
|= zdb_check_for_obsolete_leaks(vd
, zcb
);
6651 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
6652 metaslab_t
*msp
= vd
->vdev_ms
[m
];
6653 ASSERT3P(msp
->ms_group
, ==, (msp
->ms_group
->mg_class
==
6654 spa_embedded_log_class(spa
)) ?
6655 vd
->vdev_log_mg
: vd
->vdev_mg
);
6658 * ms_allocatable has been overloaded
6659 * to contain allocated segments. Now that
6660 * we finished traversing all blocks, any
6661 * block that remains in the ms_allocatable
6662 * represents an allocated block that we
6663 * did not claim during the traversal.
6664 * Claimed blocks would have been removed
6665 * from the ms_allocatable. For indirect
6666 * vdevs, space remaining in the tree
6667 * represents parts of the mapping that are
6668 * not referenced, which is not a bug.
6670 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
6671 range_tree_vacate(msp
->ms_allocatable
,
6674 range_tree_vacate(msp
->ms_allocatable
,
6677 if (msp
->ms_loaded
) {
6678 msp
->ms_loaded
= B_FALSE
;
6683 umem_free(zcb
->zcb_vd_obsolete_counts
,
6684 rvd
->vdev_children
* sizeof (uint32_t *));
6685 zcb
->zcb_vd_obsolete_counts
= NULL
;
6691 count_block_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
6694 zdb_cb_t
*zcb
= arg
;
6696 if (dump_opt
['b'] >= 5) {
6697 char blkbuf
[BP_SPRINTF_LEN
];
6698 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
6699 (void) printf("[%s] %s\n",
6700 "deferred free", blkbuf
);
6702 zdb_count_block(zcb
, NULL
, bp
, ZDB_OT_DEFERRED
);
6707 * Iterate over livelists which have been destroyed by the user but
6708 * are still present in the MOS, waiting to be freed
6711 iterate_deleted_livelists(spa_t
*spa
, ll_iter_t func
, void *arg
)
6713 objset_t
*mos
= spa
->spa_meta_objset
;
6715 int err
= zap_lookup(mos
, DMU_POOL_DIRECTORY_OBJECT
,
6716 DMU_POOL_DELETED_CLONES
, sizeof (uint64_t), 1, &zap_obj
);
6722 zap_attribute_t attr
;
6724 /* NULL out os prior to dsl_deadlist_open in case it's garbage */
6726 for (zap_cursor_init(&zc
, mos
, zap_obj
);
6727 zap_cursor_retrieve(&zc
, &attr
) == 0;
6728 (void) zap_cursor_advance(&zc
)) {
6729 dsl_deadlist_open(&ll
, mos
, attr
.za_first_integer
);
6731 dsl_deadlist_close(&ll
);
6733 zap_cursor_fini(&zc
);
6737 bpobj_count_block_cb(void *arg
, const blkptr_t
*bp
, boolean_t bp_freed
,
6741 return (count_block_cb(arg
, bp
, tx
));
6745 livelist_entry_count_blocks_cb(void *args
, dsl_deadlist_entry_t
*dle
)
6747 zdb_cb_t
*zbc
= args
;
6749 bplist_create(&blks
);
6750 /* determine which blocks have been alloc'd but not freed */
6751 VERIFY0(dsl_process_sub_livelist(&dle
->dle_bpobj
, &blks
, NULL
, NULL
));
6752 /* count those blocks */
6753 (void) bplist_iterate(&blks
, count_block_cb
, zbc
, NULL
);
6754 bplist_destroy(&blks
);
6759 livelist_count_blocks(dsl_deadlist_t
*ll
, void *arg
)
6761 dsl_deadlist_iterate(ll
, livelist_entry_count_blocks_cb
, arg
);
6765 * Count the blocks in the livelists that have been destroyed by the user
6766 * but haven't yet been freed.
6769 deleted_livelists_count_blocks(spa_t
*spa
, zdb_cb_t
*zbc
)
6771 iterate_deleted_livelists(spa
, livelist_count_blocks
, zbc
);
6775 dump_livelist_cb(dsl_deadlist_t
*ll
, void *arg
)
6777 ASSERT3P(arg
, ==, NULL
);
6778 global_feature_count
[SPA_FEATURE_LIVELIST
]++;
6779 dump_blkptr_list(ll
, "Deleted Livelist");
6780 dsl_deadlist_iterate(ll
, sublivelist_verify_lightweight
, NULL
);
6784 * Print out, register object references to, and increment feature counts for
6785 * livelists that have been destroyed by the user but haven't yet been freed.
6788 deleted_livelists_dump_mos(spa_t
*spa
)
6791 objset_t
*mos
= spa
->spa_meta_objset
;
6792 int err
= zap_lookup(mos
, DMU_POOL_DIRECTORY_OBJECT
,
6793 DMU_POOL_DELETED_CLONES
, sizeof (uint64_t), 1, &zap_obj
);
6796 mos_obj_refd(zap_obj
);
6797 iterate_deleted_livelists(spa
, dump_livelist_cb
, NULL
);
6801 zdb_brt_entry_compare(const void *zcn1
, const void *zcn2
)
6803 const dva_t
*dva1
= &((const zdb_brt_entry_t
*)zcn1
)->zbre_dva
;
6804 const dva_t
*dva2
= &((const zdb_brt_entry_t
*)zcn2
)->zbre_dva
;
6807 cmp
= TREE_CMP(DVA_GET_VDEV(dva1
), DVA_GET_VDEV(dva2
));
6809 cmp
= TREE_CMP(DVA_GET_OFFSET(dva1
), DVA_GET_OFFSET(dva2
));
6815 dump_block_stats(spa_t
*spa
)
6818 zdb_blkstats_t
*zb
, *tzb
;
6819 uint64_t norm_alloc
, norm_space
, total_alloc
, total_found
;
6820 int flags
= TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
6821 TRAVERSE_NO_DECRYPT
| TRAVERSE_HARD
;
6822 boolean_t leaks
= B_FALSE
;
6824 bp_embedded_type_t i
;
6826 zcb
= umem_zalloc(sizeof (zdb_cb_t
), UMEM_NOFAIL
);
6828 if (spa_feature_is_active(spa
, SPA_FEATURE_BLOCK_CLONING
)) {
6829 avl_create(&zcb
->zcb_brt
, zdb_brt_entry_compare
,
6830 sizeof (zdb_brt_entry_t
),
6831 offsetof(zdb_brt_entry_t
, zbre_node
));
6832 zcb
->zcb_brt_is_active
= B_TRUE
;
6835 (void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
6836 (dump_opt
['c'] || !dump_opt
['L']) ? "to verify " : "",
6837 (dump_opt
['c'] == 1) ? "metadata " : "",
6838 dump_opt
['c'] ? "checksums " : "",
6839 (dump_opt
['c'] && !dump_opt
['L']) ? "and verify " : "",
6840 !dump_opt
['L'] ? "nothing leaked " : "");
6843 * When leak detection is enabled we load all space maps as SM_ALLOC
6844 * maps, then traverse the pool claiming each block we discover. If
6845 * the pool is perfectly consistent, the segment trees will be empty
6846 * when we're done. Anything left over is a leak; any block we can't
6847 * claim (because it's not part of any space map) is a double
6848 * allocation, reference to a freed block, or an unclaimed log block.
6850 * When leak detection is disabled (-L option) we still traverse the
6851 * pool claiming each block we discover, but we skip opening any space
6854 zdb_leak_init(spa
, zcb
);
6857 * If there's a deferred-free bplist, process that first.
6859 (void) bpobj_iterate_nofree(&spa
->spa_deferred_bpobj
,
6860 bpobj_count_block_cb
, zcb
, NULL
);
6862 if (spa_version(spa
) >= SPA_VERSION_DEADLISTS
) {
6863 (void) bpobj_iterate_nofree(&spa
->spa_dsl_pool
->dp_free_bpobj
,
6864 bpobj_count_block_cb
, zcb
, NULL
);
6867 zdb_claim_removing(spa
, zcb
);
6869 if (spa_feature_is_active(spa
, SPA_FEATURE_ASYNC_DESTROY
)) {
6870 VERIFY3U(0, ==, bptree_iterate(spa
->spa_meta_objset
,
6871 spa
->spa_dsl_pool
->dp_bptree_obj
, B_FALSE
, count_block_cb
,
6875 deleted_livelists_count_blocks(spa
, zcb
);
6877 if (dump_opt
['c'] > 1)
6878 flags
|= TRAVERSE_PREFETCH_DATA
;
6880 zcb
->zcb_totalasize
= metaslab_class_get_alloc(spa_normal_class(spa
));
6881 zcb
->zcb_totalasize
+= metaslab_class_get_alloc(spa_special_class(spa
));
6882 zcb
->zcb_totalasize
+= metaslab_class_get_alloc(spa_dedup_class(spa
));
6883 zcb
->zcb_totalasize
+=
6884 metaslab_class_get_alloc(spa_embedded_log_class(spa
));
6885 zcb
->zcb_start
= zcb
->zcb_lastprint
= gethrtime();
6886 err
= traverse_pool(spa
, 0, flags
, zdb_blkptr_cb
, zcb
);
6889 * If we've traversed the data blocks then we need to wait for those
6890 * I/Os to complete. We leverage "The Godfather" zio to wait on
6891 * all async I/Os to complete.
6893 if (dump_opt
['c']) {
6894 for (c
= 0; c
< max_ncpus
; c
++) {
6895 (void) zio_wait(spa
->spa_async_zio_root
[c
]);
6896 spa
->spa_async_zio_root
[c
] = zio_root(spa
, NULL
, NULL
,
6897 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
|
6898 ZIO_FLAG_GODFATHER
);
6901 ASSERT0(spa
->spa_load_verify_bytes
);
6904 * Done after zio_wait() since zcb_haderrors is modified in
6907 zcb
->zcb_haderrors
|= err
;
6909 if (zcb
->zcb_haderrors
) {
6910 (void) printf("\nError counts:\n\n");
6911 (void) printf("\t%5s %s\n", "errno", "count");
6912 for (e
= 0; e
< 256; e
++) {
6913 if (zcb
->zcb_errors
[e
] != 0) {
6914 (void) printf("\t%5d %llu\n",
6915 e
, (u_longlong_t
)zcb
->zcb_errors
[e
]);
6921 * Report any leaked segments.
6923 leaks
|= zdb_leak_fini(spa
, zcb
);
6925 tzb
= &zcb
->zcb_type
[ZB_TOTAL
][ZDB_OT_TOTAL
];
6927 norm_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
6928 norm_space
= metaslab_class_get_space(spa_normal_class(spa
));
6930 total_alloc
= norm_alloc
+
6931 metaslab_class_get_alloc(spa_log_class(spa
)) +
6932 metaslab_class_get_alloc(spa_embedded_log_class(spa
)) +
6933 metaslab_class_get_alloc(spa_special_class(spa
)) +
6934 metaslab_class_get_alloc(spa_dedup_class(spa
)) +
6935 get_unflushed_alloc_space(spa
);
6937 tzb
->zb_asize
- zcb
->zcb_dedup_asize
- zcb
->zcb_clone_asize
+
6938 zcb
->zcb_removing_size
+ zcb
->zcb_checkpoint_size
;
6940 if (total_found
== total_alloc
&& !dump_opt
['L']) {
6941 (void) printf("\n\tNo leaks (block sum matches space"
6942 " maps exactly)\n");
6943 } else if (!dump_opt
['L']) {
6944 (void) printf("block traversal size %llu != alloc %llu "
6946 (u_longlong_t
)total_found
,
6947 (u_longlong_t
)total_alloc
,
6948 (dump_opt
['L']) ? "unreachable" : "leaked",
6949 (longlong_t
)(total_alloc
- total_found
));
6953 if (tzb
->zb_count
== 0) {
6954 umem_free(zcb
, sizeof (zdb_cb_t
));
6958 (void) printf("\n");
6959 (void) printf("\t%-16s %14llu\n", "bp count:",
6960 (u_longlong_t
)tzb
->zb_count
);
6961 (void) printf("\t%-16s %14llu\n", "ganged count:",
6962 (longlong_t
)tzb
->zb_gangs
);
6963 (void) printf("\t%-16s %14llu avg: %6llu\n", "bp logical:",
6964 (u_longlong_t
)tzb
->zb_lsize
,
6965 (u_longlong_t
)(tzb
->zb_lsize
/ tzb
->zb_count
));
6966 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
6967 "bp physical:", (u_longlong_t
)tzb
->zb_psize
,
6968 (u_longlong_t
)(tzb
->zb_psize
/ tzb
->zb_count
),
6969 (double)tzb
->zb_lsize
/ tzb
->zb_psize
);
6970 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n",
6971 "bp allocated:", (u_longlong_t
)tzb
->zb_asize
,
6972 (u_longlong_t
)(tzb
->zb_asize
/ tzb
->zb_count
),
6973 (double)tzb
->zb_lsize
/ tzb
->zb_asize
);
6974 (void) printf("\t%-16s %14llu ref>1: %6llu deduplication: %6.2f\n",
6975 "bp deduped:", (u_longlong_t
)zcb
->zcb_dedup_asize
,
6976 (u_longlong_t
)zcb
->zcb_dedup_blocks
,
6977 (double)zcb
->zcb_dedup_asize
/ tzb
->zb_asize
+ 1.0);
6978 (void) printf("\t%-16s %14llu count: %6llu\n",
6979 "bp cloned:", (u_longlong_t
)zcb
->zcb_clone_asize
,
6980 (u_longlong_t
)zcb
->zcb_clone_blocks
);
6981 (void) printf("\t%-16s %14llu used: %5.2f%%\n", "Normal class:",
6982 (u_longlong_t
)norm_alloc
, 100.0 * norm_alloc
/ norm_space
);
6984 if (spa_special_class(spa
)->mc_allocator
[0].mca_rotor
!= NULL
) {
6985 uint64_t alloc
= metaslab_class_get_alloc(
6986 spa_special_class(spa
));
6987 uint64_t space
= metaslab_class_get_space(
6988 spa_special_class(spa
));
6990 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
6991 "Special class", (u_longlong_t
)alloc
,
6992 100.0 * alloc
/ space
);
6995 if (spa_dedup_class(spa
)->mc_allocator
[0].mca_rotor
!= NULL
) {
6996 uint64_t alloc
= metaslab_class_get_alloc(
6997 spa_dedup_class(spa
));
6998 uint64_t space
= metaslab_class_get_space(
6999 spa_dedup_class(spa
));
7001 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7002 "Dedup class", (u_longlong_t
)alloc
,
7003 100.0 * alloc
/ space
);
7006 if (spa_embedded_log_class(spa
)->mc_allocator
[0].mca_rotor
!= NULL
) {
7007 uint64_t alloc
= metaslab_class_get_alloc(
7008 spa_embedded_log_class(spa
));
7009 uint64_t space
= metaslab_class_get_space(
7010 spa_embedded_log_class(spa
));
7012 (void) printf("\t%-16s %14llu used: %5.2f%%\n",
7013 "Embedded log class", (u_longlong_t
)alloc
,
7014 100.0 * alloc
/ space
);
7017 for (i
= 0; i
< NUM_BP_EMBEDDED_TYPES
; i
++) {
7018 if (zcb
->zcb_embedded_blocks
[i
] == 0)
7020 (void) printf("\n");
7021 (void) printf("\tadditional, non-pointer bps of type %u: "
7023 i
, (u_longlong_t
)zcb
->zcb_embedded_blocks
[i
]);
7025 if (dump_opt
['b'] >= 3) {
7026 (void) printf("\t number of (compressed) bytes: "
7028 dump_histogram(zcb
->zcb_embedded_histogram
[i
],
7029 sizeof (zcb
->zcb_embedded_histogram
[i
]) /
7030 sizeof (zcb
->zcb_embedded_histogram
[i
][0]), 0);
7034 if (tzb
->zb_ditto_samevdev
!= 0) {
7035 (void) printf("\tDittoed blocks on same vdev: %llu\n",
7036 (longlong_t
)tzb
->zb_ditto_samevdev
);
7038 if (tzb
->zb_ditto_same_ms
!= 0) {
7039 (void) printf("\tDittoed blocks in same metaslab: %llu\n",
7040 (longlong_t
)tzb
->zb_ditto_same_ms
);
7043 for (uint64_t v
= 0; v
< spa
->spa_root_vdev
->vdev_children
; v
++) {
7044 vdev_t
*vd
= spa
->spa_root_vdev
->vdev_child
[v
];
7045 vdev_indirect_mapping_t
*vim
= vd
->vdev_indirect_mapping
;
7052 zdb_nicenum(vdev_indirect_mapping_num_entries(vim
),
7053 mem
, vdev_indirect_mapping_size(vim
));
7055 (void) printf("\tindirect vdev id %llu has %llu segments "
7057 (longlong_t
)vd
->vdev_id
,
7058 (longlong_t
)vdev_indirect_mapping_num_entries(vim
), mem
);
7061 if (dump_opt
['b'] >= 2) {
7063 char csize
[32], lsize
[32], psize
[32], asize
[32];
7064 char avg
[32], gang
[32];
7065 (void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
7066 "\t avg\t comp\t%%Total\tType\n");
7068 zfs_blkstat_t
*mdstats
= umem_zalloc(sizeof (zfs_blkstat_t
),
7071 for (t
= 0; t
<= ZDB_OT_TOTAL
; t
++) {
7072 const char *typename
;
7074 /* make sure nicenum has enough space */
7075 _Static_assert(sizeof (csize
) >= NN_NUMBUF_SZ
,
7077 _Static_assert(sizeof (lsize
) >= NN_NUMBUF_SZ
,
7079 _Static_assert(sizeof (psize
) >= NN_NUMBUF_SZ
,
7081 _Static_assert(sizeof (asize
) >= NN_NUMBUF_SZ
,
7083 _Static_assert(sizeof (avg
) >= NN_NUMBUF_SZ
,
7085 _Static_assert(sizeof (gang
) >= NN_NUMBUF_SZ
,
7088 if (t
< DMU_OT_NUMTYPES
)
7089 typename
= dmu_ot
[t
].ot_name
;
7091 typename
= zdb_ot_extname
[t
- DMU_OT_NUMTYPES
];
7093 if (zcb
->zcb_type
[ZB_TOTAL
][t
].zb_asize
== 0) {
7094 (void) printf("%6s\t%5s\t%5s\t%5s"
7095 "\t%5s\t%5s\t%6s\t%s\n",
7107 for (l
= ZB_TOTAL
- 1; l
>= -1; l
--) {
7108 level
= (l
== -1 ? ZB_TOTAL
: l
);
7109 zb
= &zcb
->zcb_type
[level
][t
];
7111 if (zb
->zb_asize
== 0)
7114 if (level
!= ZB_TOTAL
&& t
< DMU_OT_NUMTYPES
&&
7115 (level
> 0 || DMU_OT_IS_METADATA(t
))) {
7116 mdstats
->zb_count
+= zb
->zb_count
;
7117 mdstats
->zb_lsize
+= zb
->zb_lsize
;
7118 mdstats
->zb_psize
+= zb
->zb_psize
;
7119 mdstats
->zb_asize
+= zb
->zb_asize
;
7120 mdstats
->zb_gangs
+= zb
->zb_gangs
;
7123 if (dump_opt
['b'] < 3 && level
!= ZB_TOTAL
)
7126 if (level
== 0 && zb
->zb_asize
==
7127 zcb
->zcb_type
[ZB_TOTAL
][t
].zb_asize
)
7130 zdb_nicenum(zb
->zb_count
, csize
,
7132 zdb_nicenum(zb
->zb_lsize
, lsize
,
7134 zdb_nicenum(zb
->zb_psize
, psize
,
7136 zdb_nicenum(zb
->zb_asize
, asize
,
7138 zdb_nicenum(zb
->zb_asize
/ zb
->zb_count
, avg
,
7140 zdb_nicenum(zb
->zb_gangs
, gang
, sizeof (gang
));
7142 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7144 csize
, lsize
, psize
, asize
, avg
,
7145 (double)zb
->zb_lsize
/ zb
->zb_psize
,
7146 100.0 * zb
->zb_asize
/ tzb
->zb_asize
);
7148 if (level
== ZB_TOTAL
)
7149 (void) printf("%s\n", typename
);
7151 (void) printf(" L%d %s\n",
7154 if (dump_opt
['b'] >= 3 && zb
->zb_gangs
> 0) {
7155 (void) printf("\t number of ganged "
7156 "blocks: %s\n", gang
);
7159 if (dump_opt
['b'] >= 4) {
7160 (void) printf("psize "
7161 "(in 512-byte sectors): "
7162 "number of blocks\n");
7163 dump_histogram(zb
->zb_psize_histogram
,
7164 PSIZE_HISTO_SIZE
, 0);
7168 zdb_nicenum(mdstats
->zb_count
, csize
,
7170 zdb_nicenum(mdstats
->zb_lsize
, lsize
,
7172 zdb_nicenum(mdstats
->zb_psize
, psize
,
7174 zdb_nicenum(mdstats
->zb_asize
, asize
,
7176 zdb_nicenum(mdstats
->zb_asize
/ mdstats
->zb_count
, avg
,
7178 zdb_nicenum(mdstats
->zb_gangs
, gang
, sizeof (gang
));
7180 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7182 csize
, lsize
, psize
, asize
, avg
,
7183 (double)mdstats
->zb_lsize
/ mdstats
->zb_psize
,
7184 100.0 * mdstats
->zb_asize
/ tzb
->zb_asize
);
7185 (void) printf("%s\n", "Metadata Total");
7187 /* Output a table summarizing block sizes in the pool */
7188 if (dump_opt
['b'] >= 2) {
7189 dump_size_histograms(zcb
);
7192 umem_free(mdstats
, sizeof (zfs_blkstat_t
));
7195 (void) printf("\n");
7198 umem_free(zcb
, sizeof (zdb_cb_t
));
7202 if (zcb
->zcb_haderrors
) {
7203 umem_free(zcb
, sizeof (zdb_cb_t
));
7207 umem_free(zcb
, sizeof (zdb_cb_t
));
7211 typedef struct zdb_ddt_entry
{
7213 uint64_t zdde_ref_blocks
;
7214 uint64_t zdde_ref_lsize
;
7215 uint64_t zdde_ref_psize
;
7216 uint64_t zdde_ref_dsize
;
7217 avl_node_t zdde_node
;
7221 zdb_ddt_add_cb(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
7222 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
7224 (void) zilog
, (void) dnp
;
7225 avl_tree_t
*t
= arg
;
7227 zdb_ddt_entry_t
*zdde
, zdde_search
;
7229 if (zb
->zb_level
== ZB_DNODE_LEVEL
|| BP_IS_HOLE(bp
) ||
7233 if (dump_opt
['S'] > 1 && zb
->zb_level
== ZB_ROOT_LEVEL
) {
7234 (void) printf("traversing objset %llu, %llu objects, "
7235 "%lu blocks so far\n",
7236 (u_longlong_t
)zb
->zb_objset
,
7237 (u_longlong_t
)BP_GET_FILL(bp
),
7241 if (BP_IS_HOLE(bp
) || BP_GET_CHECKSUM(bp
) == ZIO_CHECKSUM_OFF
||
7242 BP_GET_LEVEL(bp
) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)))
7245 ddt_key_fill(&zdde_search
.zdde_key
, bp
);
7247 zdde
= avl_find(t
, &zdde_search
, &where
);
7250 zdde
= umem_zalloc(sizeof (*zdde
), UMEM_NOFAIL
);
7251 zdde
->zdde_key
= zdde_search
.zdde_key
;
7252 avl_insert(t
, zdde
, where
);
7255 zdde
->zdde_ref_blocks
+= 1;
7256 zdde
->zdde_ref_lsize
+= BP_GET_LSIZE(bp
);
7257 zdde
->zdde_ref_psize
+= BP_GET_PSIZE(bp
);
7258 zdde
->zdde_ref_dsize
+= bp_get_dsize_sync(spa
, bp
);
7264 dump_simulated_ddt(spa_t
*spa
)
7267 void *cookie
= NULL
;
7268 zdb_ddt_entry_t
*zdde
;
7269 ddt_histogram_t ddh_total
= {{{0}}};
7270 ddt_stat_t dds_total
= {0};
7272 avl_create(&t
, ddt_entry_compare
,
7273 sizeof (zdb_ddt_entry_t
), offsetof(zdb_ddt_entry_t
, zdde_node
));
7275 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
7277 (void) traverse_pool(spa
, 0, TRAVERSE_PRE
| TRAVERSE_PREFETCH_METADATA
|
7278 TRAVERSE_NO_DECRYPT
, zdb_ddt_add_cb
, &t
);
7280 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
7282 while ((zdde
= avl_destroy_nodes(&t
, &cookie
)) != NULL
) {
7284 uint64_t refcnt
= zdde
->zdde_ref_blocks
;
7285 ASSERT(refcnt
!= 0);
7287 dds
.dds_blocks
= zdde
->zdde_ref_blocks
/ refcnt
;
7288 dds
.dds_lsize
= zdde
->zdde_ref_lsize
/ refcnt
;
7289 dds
.dds_psize
= zdde
->zdde_ref_psize
/ refcnt
;
7290 dds
.dds_dsize
= zdde
->zdde_ref_dsize
/ refcnt
;
7292 dds
.dds_ref_blocks
= zdde
->zdde_ref_blocks
;
7293 dds
.dds_ref_lsize
= zdde
->zdde_ref_lsize
;
7294 dds
.dds_ref_psize
= zdde
->zdde_ref_psize
;
7295 dds
.dds_ref_dsize
= zdde
->zdde_ref_dsize
;
7297 ddt_stat_add(&ddh_total
.ddh_stat
[highbit64(refcnt
) - 1],
7300 umem_free(zdde
, sizeof (*zdde
));
7305 ddt_histogram_stat(&dds_total
, &ddh_total
);
7307 (void) printf("Simulated DDT histogram:\n");
7309 zpool_dump_ddt(&dds_total
, &ddh_total
);
7311 dump_dedup_ratio(&dds_total
);
7315 verify_device_removal_feature_counts(spa_t
*spa
)
7317 uint64_t dr_feature_refcount
= 0;
7318 uint64_t oc_feature_refcount
= 0;
7319 uint64_t indirect_vdev_count
= 0;
7320 uint64_t precise_vdev_count
= 0;
7321 uint64_t obsolete_counts_object_count
= 0;
7322 uint64_t obsolete_sm_count
= 0;
7323 uint64_t obsolete_counts_count
= 0;
7324 uint64_t scip_count
= 0;
7325 uint64_t obsolete_bpobj_count
= 0;
7328 spa_condensing_indirect_phys_t
*scip
=
7329 &spa
->spa_condensing_indirect_phys
;
7330 if (scip
->scip_next_mapping_object
!= 0) {
7331 vdev_t
*vd
= spa
->spa_root_vdev
->vdev_child
[scip
->scip_vdev
];
7332 ASSERT(scip
->scip_prev_obsolete_sm_object
!= 0);
7333 ASSERT3P(vd
->vdev_ops
, ==, &vdev_indirect_ops
);
7335 (void) printf("Condensing indirect vdev %llu: new mapping "
7336 "object %llu, prev obsolete sm %llu\n",
7337 (u_longlong_t
)scip
->scip_vdev
,
7338 (u_longlong_t
)scip
->scip_next_mapping_object
,
7339 (u_longlong_t
)scip
->scip_prev_obsolete_sm_object
);
7340 if (scip
->scip_prev_obsolete_sm_object
!= 0) {
7341 space_map_t
*prev_obsolete_sm
= NULL
;
7342 VERIFY0(space_map_open(&prev_obsolete_sm
,
7343 spa
->spa_meta_objset
,
7344 scip
->scip_prev_obsolete_sm_object
,
7345 0, vd
->vdev_asize
, 0));
7346 dump_spacemap(spa
->spa_meta_objset
, prev_obsolete_sm
);
7347 (void) printf("\n");
7348 space_map_close(prev_obsolete_sm
);
7354 for (uint64_t i
= 0; i
< spa
->spa_root_vdev
->vdev_children
; i
++) {
7355 vdev_t
*vd
= spa
->spa_root_vdev
->vdev_child
[i
];
7356 vdev_indirect_config_t
*vic
= &vd
->vdev_indirect_config
;
7358 if (vic
->vic_mapping_object
!= 0) {
7359 ASSERT(vd
->vdev_ops
== &vdev_indirect_ops
||
7361 indirect_vdev_count
++;
7363 if (vd
->vdev_indirect_mapping
->vim_havecounts
) {
7364 obsolete_counts_count
++;
7368 boolean_t are_precise
;
7369 VERIFY0(vdev_obsolete_counts_are_precise(vd
, &are_precise
));
7371 ASSERT(vic
->vic_mapping_object
!= 0);
7372 precise_vdev_count
++;
7375 uint64_t obsolete_sm_object
;
7376 VERIFY0(vdev_obsolete_sm_object(vd
, &obsolete_sm_object
));
7377 if (obsolete_sm_object
!= 0) {
7378 ASSERT(vic
->vic_mapping_object
!= 0);
7379 obsolete_sm_count
++;
7383 (void) feature_get_refcount(spa
,
7384 &spa_feature_table
[SPA_FEATURE_DEVICE_REMOVAL
],
7385 &dr_feature_refcount
);
7386 (void) feature_get_refcount(spa
,
7387 &spa_feature_table
[SPA_FEATURE_OBSOLETE_COUNTS
],
7388 &oc_feature_refcount
);
7390 if (dr_feature_refcount
!= indirect_vdev_count
) {
7392 (void) printf("Number of indirect vdevs (%llu) " \
7393 "does not match feature count (%llu)\n",
7394 (u_longlong_t
)indirect_vdev_count
,
7395 (u_longlong_t
)dr_feature_refcount
);
7397 (void) printf("Verified device_removal feature refcount " \
7398 "of %llu is correct\n",
7399 (u_longlong_t
)dr_feature_refcount
);
7402 if (zap_contains(spa_meta_objset(spa
), DMU_POOL_DIRECTORY_OBJECT
,
7403 DMU_POOL_OBSOLETE_BPOBJ
) == 0) {
7404 obsolete_bpobj_count
++;
7408 obsolete_counts_object_count
= precise_vdev_count
;
7409 obsolete_counts_object_count
+= obsolete_sm_count
;
7410 obsolete_counts_object_count
+= obsolete_counts_count
;
7411 obsolete_counts_object_count
+= scip_count
;
7412 obsolete_counts_object_count
+= obsolete_bpobj_count
;
7413 obsolete_counts_object_count
+= remap_deadlist_count
;
7415 if (oc_feature_refcount
!= obsolete_counts_object_count
) {
7417 (void) printf("Number of obsolete counts objects (%llu) " \
7418 "does not match feature count (%llu)\n",
7419 (u_longlong_t
)obsolete_counts_object_count
,
7420 (u_longlong_t
)oc_feature_refcount
);
7421 (void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
7422 "ob:%llu rd:%llu\n",
7423 (u_longlong_t
)precise_vdev_count
,
7424 (u_longlong_t
)obsolete_sm_count
,
7425 (u_longlong_t
)obsolete_counts_count
,
7426 (u_longlong_t
)scip_count
,
7427 (u_longlong_t
)obsolete_bpobj_count
,
7428 (u_longlong_t
)remap_deadlist_count
);
7430 (void) printf("Verified indirect_refcount feature refcount " \
7431 "of %llu is correct\n",
7432 (u_longlong_t
)oc_feature_refcount
);
7438 zdb_set_skip_mmp(char *target
)
7443 * Disable the activity check to allow examination of
7446 mutex_enter(&spa_namespace_lock
);
7447 if ((spa
= spa_lookup(target
)) != NULL
) {
7448 spa
->spa_import_flags
|= ZFS_IMPORT_SKIP_MMP
;
7450 mutex_exit(&spa_namespace_lock
);
7453 #define BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7455 * Import the checkpointed state of the pool specified by the target
7456 * parameter as readonly. The function also accepts a pool config
7457 * as an optional parameter, else it attempts to infer the config by
7458 * the name of the target pool.
7460 * Note that the checkpointed state's pool name will be the name of
7461 * the original pool with the above suffix appended to it. In addition,
7462 * if the target is not a pool name (e.g. a path to a dataset) then
7463 * the new_path parameter is populated with the updated path to
7464 * reflect the fact that we are looking into the checkpointed state.
7466 * The function returns a newly-allocated copy of the name of the
7467 * pool containing the checkpointed state. When this copy is no
7468 * longer needed it should be freed with free(3C). Same thing
7469 * applies to the new_path parameter if allocated.
7472 import_checkpointed_state(char *target
, nvlist_t
*cfg
, char **new_path
)
7475 char *poolname
, *bogus_name
= NULL
;
7476 boolean_t freecfg
= B_FALSE
;
7478 /* If the target is not a pool, the extract the pool name */
7479 char *path_start
= strchr(target
, '/');
7480 if (path_start
!= NULL
) {
7481 size_t poolname_len
= path_start
- target
;
7482 poolname
= strndup(target
, poolname_len
);
7488 zdb_set_skip_mmp(poolname
);
7489 error
= spa_get_stats(poolname
, &cfg
, NULL
, 0);
7491 fatal("Tried to read config of pool \"%s\" but "
7492 "spa_get_stats() failed with error %d\n",
7498 if (asprintf(&bogus_name
, "%s%s", poolname
, BOGUS_SUFFIX
) == -1) {
7499 if (target
!= poolname
)
7503 fnvlist_add_string(cfg
, ZPOOL_CONFIG_POOL_NAME
, bogus_name
);
7505 error
= spa_import(bogus_name
, cfg
, NULL
,
7506 ZFS_IMPORT_MISSING_LOG
| ZFS_IMPORT_CHECKPOINT
|
7507 ZFS_IMPORT_SKIP_MMP
);
7511 fatal("Tried to import pool \"%s\" but spa_import() failed "
7512 "with error %d\n", bogus_name
, error
);
7515 if (new_path
!= NULL
&& path_start
!= NULL
) {
7516 if (asprintf(new_path
, "%s%s", bogus_name
, path_start
) == -1) {
7518 if (path_start
!= NULL
)
7524 if (target
!= poolname
)
7527 return (bogus_name
);
7530 typedef struct verify_checkpoint_sm_entry_cb_arg
{
7533 /* the following fields are only used for printing progress */
7534 uint64_t vcsec_entryid
;
7535 uint64_t vcsec_num_entries
;
7536 } verify_checkpoint_sm_entry_cb_arg_t
;
7538 #define ENTRIES_PER_PROGRESS_UPDATE 10000
7541 verify_checkpoint_sm_entry_cb(space_map_entry_t
*sme
, void *arg
)
7543 verify_checkpoint_sm_entry_cb_arg_t
*vcsec
= arg
;
7544 vdev_t
*vd
= vcsec
->vcsec_vd
;
7545 metaslab_t
*ms
= vd
->vdev_ms
[sme
->sme_offset
>> vd
->vdev_ms_shift
];
7546 uint64_t end
= sme
->sme_offset
+ sme
->sme_run
;
7548 ASSERT(sme
->sme_type
== SM_FREE
);
7550 if ((vcsec
->vcsec_entryid
% ENTRIES_PER_PROGRESS_UPDATE
) == 0) {
7551 (void) fprintf(stderr
,
7552 "\rverifying vdev %llu, space map entry %llu of %llu ...",
7553 (longlong_t
)vd
->vdev_id
,
7554 (longlong_t
)vcsec
->vcsec_entryid
,
7555 (longlong_t
)vcsec
->vcsec_num_entries
);
7557 vcsec
->vcsec_entryid
++;
7560 * See comment in checkpoint_sm_exclude_entry_cb()
7562 VERIFY3U(sme
->sme_offset
, >=, ms
->ms_start
);
7563 VERIFY3U(end
, <=, ms
->ms_start
+ ms
->ms_size
);
7566 * The entries in the vdev_checkpoint_sm should be marked as
7567 * allocated in the checkpointed state of the pool, therefore
7568 * their respective ms_allocateable trees should not contain them.
7570 mutex_enter(&ms
->ms_lock
);
7571 range_tree_verify_not_present(ms
->ms_allocatable
,
7572 sme
->sme_offset
, sme
->sme_run
);
7573 mutex_exit(&ms
->ms_lock
);
7579 * Verify that all segments in the vdev_checkpoint_sm are allocated
7580 * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
7583 * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
7584 * each vdev in the current state of the pool to the metaslab space maps
7585 * (ms_sm) of the checkpointed state of the pool.
7587 * Note that the function changes the state of the ms_allocatable
7588 * trees of the current spa_t. The entries of these ms_allocatable
7589 * trees are cleared out and then repopulated from with the free
7590 * entries of their respective ms_sm space maps.
7593 verify_checkpoint_vdev_spacemaps(spa_t
*checkpoint
, spa_t
*current
)
7595 vdev_t
*ckpoint_rvd
= checkpoint
->spa_root_vdev
;
7596 vdev_t
*current_rvd
= current
->spa_root_vdev
;
7598 load_concrete_ms_allocatable_trees(checkpoint
, SM_FREE
);
7600 for (uint64_t c
= 0; c
< ckpoint_rvd
->vdev_children
; c
++) {
7601 vdev_t
*ckpoint_vd
= ckpoint_rvd
->vdev_child
[c
];
7602 vdev_t
*current_vd
= current_rvd
->vdev_child
[c
];
7604 space_map_t
*checkpoint_sm
= NULL
;
7605 uint64_t checkpoint_sm_obj
;
7607 if (ckpoint_vd
->vdev_ops
== &vdev_indirect_ops
) {
7609 * Since we don't allow device removal in a pool
7610 * that has a checkpoint, we expect that all removed
7611 * vdevs were removed from the pool before the
7614 ASSERT3P(current_vd
->vdev_ops
, ==, &vdev_indirect_ops
);
7619 * If the checkpoint space map doesn't exist, then nothing
7620 * here is checkpointed so there's nothing to verify.
7622 if (current_vd
->vdev_top_zap
== 0 ||
7623 zap_contains(spa_meta_objset(current
),
7624 current_vd
->vdev_top_zap
,
7625 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) != 0)
7628 VERIFY0(zap_lookup(spa_meta_objset(current
),
7629 current_vd
->vdev_top_zap
, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
,
7630 sizeof (uint64_t), 1, &checkpoint_sm_obj
));
7632 VERIFY0(space_map_open(&checkpoint_sm
, spa_meta_objset(current
),
7633 checkpoint_sm_obj
, 0, current_vd
->vdev_asize
,
7634 current_vd
->vdev_ashift
));
7636 verify_checkpoint_sm_entry_cb_arg_t vcsec
;
7637 vcsec
.vcsec_vd
= ckpoint_vd
;
7638 vcsec
.vcsec_entryid
= 0;
7639 vcsec
.vcsec_num_entries
=
7640 space_map_length(checkpoint_sm
) / sizeof (uint64_t);
7641 VERIFY0(space_map_iterate(checkpoint_sm
,
7642 space_map_length(checkpoint_sm
),
7643 verify_checkpoint_sm_entry_cb
, &vcsec
));
7644 if (dump_opt
['m'] > 3)
7645 dump_spacemap(current
->spa_meta_objset
, checkpoint_sm
);
7646 space_map_close(checkpoint_sm
);
7650 * If we've added vdevs since we took the checkpoint, ensure
7651 * that their checkpoint space maps are empty.
7653 if (ckpoint_rvd
->vdev_children
< current_rvd
->vdev_children
) {
7654 for (uint64_t c
= ckpoint_rvd
->vdev_children
;
7655 c
< current_rvd
->vdev_children
; c
++) {
7656 vdev_t
*current_vd
= current_rvd
->vdev_child
[c
];
7657 VERIFY3P(current_vd
->vdev_checkpoint_sm
, ==, NULL
);
7661 /* for cleaner progress output */
7662 (void) fprintf(stderr
, "\n");
7666 * Verifies that all space that's allocated in the checkpoint is
7667 * still allocated in the current version, by checking that everything
7668 * in checkpoint's ms_allocatable (which is actually allocated, not
7669 * allocatable/free) is not present in current's ms_allocatable.
7671 * Note that the function changes the state of the ms_allocatable
7672 * trees of both spas when called. The entries of all ms_allocatable
7673 * trees are cleared out and then repopulated from their respective
7674 * ms_sm space maps. In the checkpointed state we load the allocated
7675 * entries, and in the current state we load the free entries.
7678 verify_checkpoint_ms_spacemaps(spa_t
*checkpoint
, spa_t
*current
)
7680 vdev_t
*ckpoint_rvd
= checkpoint
->spa_root_vdev
;
7681 vdev_t
*current_rvd
= current
->spa_root_vdev
;
7683 load_concrete_ms_allocatable_trees(checkpoint
, SM_ALLOC
);
7684 load_concrete_ms_allocatable_trees(current
, SM_FREE
);
7686 for (uint64_t i
= 0; i
< ckpoint_rvd
->vdev_children
; i
++) {
7687 vdev_t
*ckpoint_vd
= ckpoint_rvd
->vdev_child
[i
];
7688 vdev_t
*current_vd
= current_rvd
->vdev_child
[i
];
7690 if (ckpoint_vd
->vdev_ops
== &vdev_indirect_ops
) {
7692 * See comment in verify_checkpoint_vdev_spacemaps()
7694 ASSERT3P(current_vd
->vdev_ops
, ==, &vdev_indirect_ops
);
7698 for (uint64_t m
= 0; m
< ckpoint_vd
->vdev_ms_count
; m
++) {
7699 metaslab_t
*ckpoint_msp
= ckpoint_vd
->vdev_ms
[m
];
7700 metaslab_t
*current_msp
= current_vd
->vdev_ms
[m
];
7702 (void) fprintf(stderr
,
7703 "\rverifying vdev %llu of %llu, "
7704 "metaslab %llu of %llu ...",
7705 (longlong_t
)current_vd
->vdev_id
,
7706 (longlong_t
)current_rvd
->vdev_children
,
7707 (longlong_t
)current_vd
->vdev_ms
[m
]->ms_id
,
7708 (longlong_t
)current_vd
->vdev_ms_count
);
7711 * We walk through the ms_allocatable trees that
7712 * are loaded with the allocated blocks from the
7713 * ms_sm spacemaps of the checkpoint. For each
7714 * one of these ranges we ensure that none of them
7715 * exists in the ms_allocatable trees of the
7716 * current state which are loaded with the ranges
7717 * that are currently free.
7719 * This way we ensure that none of the blocks that
7720 * are part of the checkpoint were freed by mistake.
7722 range_tree_walk(ckpoint_msp
->ms_allocatable
,
7723 (range_tree_func_t
*)range_tree_verify_not_present
,
7724 current_msp
->ms_allocatable
);
7728 /* for cleaner progress output */
7729 (void) fprintf(stderr
, "\n");
7733 verify_checkpoint_blocks(spa_t
*spa
)
7735 ASSERT(!dump_opt
['L']);
7737 spa_t
*checkpoint_spa
;
7738 char *checkpoint_pool
;
7742 * We import the checkpointed state of the pool (under a different
7743 * name) so we can do verification on it against the current state
7746 checkpoint_pool
= import_checkpointed_state(spa
->spa_name
, NULL
,
7748 ASSERT(strcmp(spa
->spa_name
, checkpoint_pool
) != 0);
7750 error
= spa_open(checkpoint_pool
, &checkpoint_spa
, FTAG
);
7752 fatal("Tried to open pool \"%s\" but spa_open() failed with "
7753 "error %d\n", checkpoint_pool
, error
);
7757 * Ensure that ranges in the checkpoint space maps of each vdev
7758 * are allocated according to the checkpointed state's metaslab
7761 verify_checkpoint_vdev_spacemaps(checkpoint_spa
, spa
);
7764 * Ensure that allocated ranges in the checkpoint's metaslab
7765 * space maps remain allocated in the metaslab space maps of
7766 * the current state.
7768 verify_checkpoint_ms_spacemaps(checkpoint_spa
, spa
);
7771 * Once we are done, we get rid of the checkpointed state.
7773 spa_close(checkpoint_spa
, FTAG
);
7774 free(checkpoint_pool
);
7778 dump_leftover_checkpoint_blocks(spa_t
*spa
)
7780 vdev_t
*rvd
= spa
->spa_root_vdev
;
7782 for (uint64_t i
= 0; i
< rvd
->vdev_children
; i
++) {
7783 vdev_t
*vd
= rvd
->vdev_child
[i
];
7785 space_map_t
*checkpoint_sm
= NULL
;
7786 uint64_t checkpoint_sm_obj
;
7788 if (vd
->vdev_top_zap
== 0)
7791 if (zap_contains(spa_meta_objset(spa
), vd
->vdev_top_zap
,
7792 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
) != 0)
7795 VERIFY0(zap_lookup(spa_meta_objset(spa
), vd
->vdev_top_zap
,
7796 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
,
7797 sizeof (uint64_t), 1, &checkpoint_sm_obj
));
7799 VERIFY0(space_map_open(&checkpoint_sm
, spa_meta_objset(spa
),
7800 checkpoint_sm_obj
, 0, vd
->vdev_asize
, vd
->vdev_ashift
));
7801 dump_spacemap(spa
->spa_meta_objset
, checkpoint_sm
);
7802 space_map_close(checkpoint_sm
);
7807 verify_checkpoint(spa_t
*spa
)
7809 uberblock_t checkpoint
;
7812 if (!spa_feature_is_active(spa
, SPA_FEATURE_POOL_CHECKPOINT
))
7815 error
= zap_lookup(spa
->spa_meta_objset
, DMU_POOL_DIRECTORY_OBJECT
,
7816 DMU_POOL_ZPOOL_CHECKPOINT
, sizeof (uint64_t),
7817 sizeof (uberblock_t
) / sizeof (uint64_t), &checkpoint
);
7819 if (error
== ENOENT
&& !dump_opt
['L']) {
7821 * If the feature is active but the uberblock is missing
7822 * then we must be in the middle of discarding the
7825 (void) printf("\nPartially discarded checkpoint "
7827 if (dump_opt
['m'] > 3)
7828 dump_leftover_checkpoint_blocks(spa
);
7830 } else if (error
!= 0) {
7831 (void) printf("lookup error %d when looking for "
7832 "checkpointed uberblock in MOS\n", error
);
7835 dump_uberblock(&checkpoint
, "\nCheckpointed uberblock found:\n", "\n");
7837 if (checkpoint
.ub_checkpoint_txg
== 0) {
7838 (void) printf("\nub_checkpoint_txg not set in checkpointed "
7843 if (error
== 0 && !dump_opt
['L'])
7844 verify_checkpoint_blocks(spa
);
7850 mos_leaks_cb(void *arg
, uint64_t start
, uint64_t size
)
7853 for (uint64_t i
= start
; i
< size
; i
++) {
7854 (void) printf("MOS object %llu referenced but not allocated\n",
7860 mos_obj_refd(uint64_t obj
)
7862 if (obj
!= 0 && mos_refd_objs
!= NULL
)
7863 range_tree_add(mos_refd_objs
, obj
, 1);
7867 * Call on a MOS object that may already have been referenced.
7870 mos_obj_refd_multiple(uint64_t obj
)
7872 if (obj
!= 0 && mos_refd_objs
!= NULL
&&
7873 !range_tree_contains(mos_refd_objs
, obj
, 1))
7874 range_tree_add(mos_refd_objs
, obj
, 1);
7878 mos_leak_vdev_top_zap(vdev_t
*vd
)
7880 uint64_t ms_flush_data_obj
;
7881 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
),
7882 vd
->vdev_top_zap
, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
,
7883 sizeof (ms_flush_data_obj
), 1, &ms_flush_data_obj
);
7884 if (error
== ENOENT
)
7888 mos_obj_refd(ms_flush_data_obj
);
7892 mos_leak_vdev(vdev_t
*vd
)
7894 mos_obj_refd(vd
->vdev_dtl_object
);
7895 mos_obj_refd(vd
->vdev_ms_array
);
7896 mos_obj_refd(vd
->vdev_indirect_config
.vic_births_object
);
7897 mos_obj_refd(vd
->vdev_indirect_config
.vic_mapping_object
);
7898 mos_obj_refd(vd
->vdev_leaf_zap
);
7899 if (vd
->vdev_checkpoint_sm
!= NULL
)
7900 mos_obj_refd(vd
->vdev_checkpoint_sm
->sm_object
);
7901 if (vd
->vdev_indirect_mapping
!= NULL
) {
7902 mos_obj_refd(vd
->vdev_indirect_mapping
->
7903 vim_phys
->vimp_counts_object
);
7905 if (vd
->vdev_obsolete_sm
!= NULL
)
7906 mos_obj_refd(vd
->vdev_obsolete_sm
->sm_object
);
7908 for (uint64_t m
= 0; m
< vd
->vdev_ms_count
; m
++) {
7909 metaslab_t
*ms
= vd
->vdev_ms
[m
];
7910 mos_obj_refd(space_map_object(ms
->ms_sm
));
7913 if (vd
->vdev_root_zap
!= 0)
7914 mos_obj_refd(vd
->vdev_root_zap
);
7916 if (vd
->vdev_top_zap
!= 0) {
7917 mos_obj_refd(vd
->vdev_top_zap
);
7918 mos_leak_vdev_top_zap(vd
);
7921 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++) {
7922 mos_leak_vdev(vd
->vdev_child
[c
]);
7927 mos_leak_log_spacemaps(spa_t
*spa
)
7929 uint64_t spacemap_zap
;
7930 int error
= zap_lookup(spa_meta_objset(spa
),
7931 DMU_POOL_DIRECTORY_OBJECT
, DMU_POOL_LOG_SPACEMAP_ZAP
,
7932 sizeof (spacemap_zap
), 1, &spacemap_zap
);
7933 if (error
== ENOENT
)
7937 mos_obj_refd(spacemap_zap
);
7938 for (spa_log_sm_t
*sls
= avl_first(&spa
->spa_sm_logs_by_txg
);
7939 sls
; sls
= AVL_NEXT(&spa
->spa_sm_logs_by_txg
, sls
))
7940 mos_obj_refd(sls
->sls_sm_obj
);
7944 errorlog_count_refd(objset_t
*mos
, uint64_t errlog
)
7948 for (zap_cursor_init(&zc
, mos
, errlog
);
7949 zap_cursor_retrieve(&zc
, &za
) == 0;
7950 zap_cursor_advance(&zc
)) {
7951 mos_obj_refd(za
.za_first_integer
);
7953 zap_cursor_fini(&zc
);
7957 dump_mos_leaks(spa_t
*spa
)
7960 objset_t
*mos
= spa
->spa_meta_objset
;
7961 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
7963 /* Visit and mark all referenced objects in the MOS */
7965 mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT
);
7966 mos_obj_refd(spa
->spa_pool_props_object
);
7967 mos_obj_refd(spa
->spa_config_object
);
7968 mos_obj_refd(spa
->spa_ddt_stat_object
);
7969 mos_obj_refd(spa
->spa_feat_desc_obj
);
7970 mos_obj_refd(spa
->spa_feat_enabled_txg_obj
);
7971 mos_obj_refd(spa
->spa_feat_for_read_obj
);
7972 mos_obj_refd(spa
->spa_feat_for_write_obj
);
7973 mos_obj_refd(spa
->spa_history
);
7974 mos_obj_refd(spa
->spa_errlog_last
);
7975 mos_obj_refd(spa
->spa_errlog_scrub
);
7977 if (spa_feature_is_enabled(spa
, SPA_FEATURE_HEAD_ERRLOG
)) {
7978 errorlog_count_refd(mos
, spa
->spa_errlog_last
);
7979 errorlog_count_refd(mos
, spa
->spa_errlog_scrub
);
7982 mos_obj_refd(spa
->spa_all_vdev_zaps
);
7983 mos_obj_refd(spa
->spa_dsl_pool
->dp_bptree_obj
);
7984 mos_obj_refd(spa
->spa_dsl_pool
->dp_tmp_userrefs_obj
);
7985 mos_obj_refd(spa
->spa_dsl_pool
->dp_scan
->scn_phys
.scn_queue_obj
);
7986 bpobj_count_refd(&spa
->spa_deferred_bpobj
);
7987 mos_obj_refd(dp
->dp_empty_bpobj
);
7988 bpobj_count_refd(&dp
->dp_obsolete_bpobj
);
7989 bpobj_count_refd(&dp
->dp_free_bpobj
);
7990 mos_obj_refd(spa
->spa_l2cache
.sav_object
);
7991 mos_obj_refd(spa
->spa_spares
.sav_object
);
7993 if (spa
->spa_syncing_log_sm
!= NULL
)
7994 mos_obj_refd(spa
->spa_syncing_log_sm
->sm_object
);
7995 mos_leak_log_spacemaps(spa
);
7997 mos_obj_refd(spa
->spa_condensing_indirect_phys
.
7998 scip_next_mapping_object
);
7999 mos_obj_refd(spa
->spa_condensing_indirect_phys
.
8000 scip_prev_obsolete_sm_object
);
8001 if (spa
->spa_condensing_indirect_phys
.scip_next_mapping_object
!= 0) {
8002 vdev_indirect_mapping_t
*vim
=
8003 vdev_indirect_mapping_open(mos
,
8004 spa
->spa_condensing_indirect_phys
.scip_next_mapping_object
);
8005 mos_obj_refd(vim
->vim_phys
->vimp_counts_object
);
8006 vdev_indirect_mapping_close(vim
);
8008 deleted_livelists_dump_mos(spa
);
8010 if (dp
->dp_origin_snap
!= NULL
) {
8013 dsl_pool_config_enter(dp
, FTAG
);
8014 VERIFY0(dsl_dataset_hold_obj(dp
,
8015 dsl_dataset_phys(dp
->dp_origin_snap
)->ds_next_snap_obj
,
8017 count_ds_mos_objects(ds
);
8018 dump_blkptr_list(&ds
->ds_deadlist
, "Deadlist");
8019 dsl_dataset_rele(ds
, FTAG
);
8020 dsl_pool_config_exit(dp
, FTAG
);
8022 count_ds_mos_objects(dp
->dp_origin_snap
);
8023 dump_blkptr_list(&dp
->dp_origin_snap
->ds_deadlist
, "Deadlist");
8025 count_dir_mos_objects(dp
->dp_mos_dir
);
8026 if (dp
->dp_free_dir
!= NULL
)
8027 count_dir_mos_objects(dp
->dp_free_dir
);
8028 if (dp
->dp_leak_dir
!= NULL
)
8029 count_dir_mos_objects(dp
->dp_leak_dir
);
8031 mos_leak_vdev(spa
->spa_root_vdev
);
8033 for (uint64_t class = 0; class < DDT_CLASSES
; class++) {
8034 for (uint64_t type
= 0; type
< DDT_TYPES
; type
++) {
8035 for (uint64_t cksum
= 0;
8036 cksum
< ZIO_CHECKSUM_FUNCTIONS
; cksum
++) {
8037 ddt_t
*ddt
= spa
->spa_ddt
[cksum
];
8038 mos_obj_refd(ddt
->ddt_object
[type
][class]);
8044 * Visit all allocated objects and make sure they are referenced.
8046 uint64_t object
= 0;
8047 while (dmu_object_next(mos
, &object
, B_FALSE
, 0) == 0) {
8048 if (range_tree_contains(mos_refd_objs
, object
, 1)) {
8049 range_tree_remove(mos_refd_objs
, object
, 1);
8051 dmu_object_info_t doi
;
8053 VERIFY0(dmu_object_info(mos
, object
, &doi
));
8054 if (doi
.doi_type
& DMU_OT_NEWTYPE
) {
8055 dmu_object_byteswap_t bswap
=
8056 DMU_OT_BYTESWAP(doi
.doi_type
);
8057 name
= dmu_ot_byteswap
[bswap
].ob_name
;
8059 name
= dmu_ot
[doi
.doi_type
].ot_name
;
8062 (void) printf("MOS object %llu (%s) leaked\n",
8063 (u_longlong_t
)object
, name
);
8067 (void) range_tree_walk(mos_refd_objs
, mos_leaks_cb
, NULL
);
8068 if (!range_tree_is_empty(mos_refd_objs
))
8070 range_tree_vacate(mos_refd_objs
, NULL
, NULL
);
8071 range_tree_destroy(mos_refd_objs
);
8075 typedef struct log_sm_obsolete_stats_arg
{
8076 uint64_t lsos_current_txg
;
8078 uint64_t lsos_total_entries
;
8079 uint64_t lsos_valid_entries
;
8081 uint64_t lsos_sm_entries
;
8082 uint64_t lsos_valid_sm_entries
;
8083 } log_sm_obsolete_stats_arg_t
;
8086 log_spacemap_obsolete_stats_cb(spa_t
*spa
, space_map_entry_t
*sme
,
8087 uint64_t txg
, void *arg
)
8089 log_sm_obsolete_stats_arg_t
*lsos
= arg
;
8091 uint64_t offset
= sme
->sme_offset
;
8092 uint64_t vdev_id
= sme
->sme_vdev
;
8094 if (lsos
->lsos_current_txg
== 0) {
8095 /* this is the first log */
8096 lsos
->lsos_current_txg
= txg
;
8097 } else if (lsos
->lsos_current_txg
< txg
) {
8098 /* we just changed log - print stats and reset */
8099 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8100 (u_longlong_t
)lsos
->lsos_valid_sm_entries
,
8101 (u_longlong_t
)lsos
->lsos_sm_entries
,
8102 (u_longlong_t
)lsos
->lsos_current_txg
);
8103 lsos
->lsos_valid_sm_entries
= 0;
8104 lsos
->lsos_sm_entries
= 0;
8105 lsos
->lsos_current_txg
= txg
;
8107 ASSERT3U(lsos
->lsos_current_txg
, ==, txg
);
8109 lsos
->lsos_sm_entries
++;
8110 lsos
->lsos_total_entries
++;
8112 vdev_t
*vd
= vdev_lookup_top(spa
, vdev_id
);
8113 if (!vdev_is_concrete(vd
))
8116 metaslab_t
*ms
= vd
->vdev_ms
[offset
>> vd
->vdev_ms_shift
];
8117 ASSERT(sme
->sme_type
== SM_ALLOC
|| sme
->sme_type
== SM_FREE
);
8119 if (txg
< metaslab_unflushed_txg(ms
))
8121 lsos
->lsos_valid_sm_entries
++;
8122 lsos
->lsos_valid_entries
++;
8127 dump_log_spacemap_obsolete_stats(spa_t
*spa
)
8129 if (!spa_feature_is_active(spa
, SPA_FEATURE_LOG_SPACEMAP
))
8132 log_sm_obsolete_stats_arg_t lsos
= {0};
8134 (void) printf("Log Space Map Obsolete Entry Statistics:\n");
8136 iterate_through_spacemap_logs(spa
,
8137 log_spacemap_obsolete_stats_cb
, &lsos
);
8139 /* print stats for latest log */
8140 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8141 (u_longlong_t
)lsos
.lsos_valid_sm_entries
,
8142 (u_longlong_t
)lsos
.lsos_sm_entries
,
8143 (u_longlong_t
)lsos
.lsos_current_txg
);
8145 (void) printf("%-8llu valid entries out of %-8llu - total\n\n",
8146 (u_longlong_t
)lsos
.lsos_valid_entries
,
8147 (u_longlong_t
)lsos
.lsos_total_entries
);
8151 dump_zpool(spa_t
*spa
)
8153 dsl_pool_t
*dp
= spa_get_dsl(spa
);
8156 if (dump_opt
['y']) {
8157 livelist_metaslab_validate(spa
);
8160 if (dump_opt
['S']) {
8161 dump_simulated_ddt(spa
);
8165 if (!dump_opt
['e'] && dump_opt
['C'] > 1) {
8166 (void) printf("\nCached configuration:\n");
8167 dump_nvlist(spa
->spa_config
, 8);
8174 dump_uberblock(&spa
->spa_uberblock
, "\nUberblock:\n", "\n");
8182 if (dump_opt
['d'] > 2 || dump_opt
['m'])
8183 dump_metaslabs(spa
);
8185 dump_metaslab_groups(spa
, dump_opt
['M'] > 1);
8186 if (dump_opt
['d'] > 2 || dump_opt
['m']) {
8187 dump_log_spacemaps(spa
);
8188 dump_log_spacemap_obsolete_stats(spa
);
8191 if (dump_opt
['d'] || dump_opt
['i']) {
8193 mos_refd_objs
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0,
8195 dump_objset(dp
->dp_meta_objset
);
8197 if (dump_opt
['d'] >= 3) {
8198 dsl_pool_t
*dp
= spa
->spa_dsl_pool
;
8199 dump_full_bpobj(&spa
->spa_deferred_bpobj
,
8200 "Deferred frees", 0);
8201 if (spa_version(spa
) >= SPA_VERSION_DEADLISTS
) {
8202 dump_full_bpobj(&dp
->dp_free_bpobj
,
8203 "Pool snapshot frees", 0);
8205 if (bpobj_is_open(&dp
->dp_obsolete_bpobj
)) {
8206 ASSERT(spa_feature_is_enabled(spa
,
8207 SPA_FEATURE_DEVICE_REMOVAL
));
8208 dump_full_bpobj(&dp
->dp_obsolete_bpobj
,
8209 "Pool obsolete blocks", 0);
8212 if (spa_feature_is_active(spa
,
8213 SPA_FEATURE_ASYNC_DESTROY
)) {
8214 dump_bptree(spa
->spa_meta_objset
,
8216 "Pool dataset frees");
8218 dump_dtl(spa
->spa_root_vdev
, 0);
8221 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++)
8222 global_feature_count
[f
] = UINT64_MAX
;
8223 global_feature_count
[SPA_FEATURE_REDACTION_BOOKMARKS
] = 0;
8224 global_feature_count
[SPA_FEATURE_BOOKMARK_WRITTEN
] = 0;
8225 global_feature_count
[SPA_FEATURE_LIVELIST
] = 0;
8227 (void) dmu_objset_find(spa_name(spa
), dump_one_objset
,
8228 NULL
, DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
8230 if (rc
== 0 && !dump_opt
['L'])
8231 rc
= dump_mos_leaks(spa
);
8233 for (f
= 0; f
< SPA_FEATURES
; f
++) {
8237 if (!(spa_feature_table
[f
].fi_flags
&
8238 ZFEATURE_FLAG_PER_DATASET
)) {
8239 if (global_feature_count
[f
] == UINT64_MAX
)
8241 if (!spa_feature_is_enabled(spa
, f
)) {
8242 ASSERT0(global_feature_count
[f
]);
8245 arr
= global_feature_count
;
8247 if (!spa_feature_is_enabled(spa
, f
)) {
8248 ASSERT0(dataset_feature_count
[f
]);
8251 arr
= dataset_feature_count
;
8253 if (feature_get_refcount(spa
, &spa_feature_table
[f
],
8254 &refcount
) == ENOTSUP
)
8256 if (arr
[f
] != refcount
) {
8257 (void) printf("%s feature refcount mismatch: "
8258 "%lld consumers != %lld refcount\n",
8259 spa_feature_table
[f
].fi_uname
,
8260 (longlong_t
)arr
[f
], (longlong_t
)refcount
);
8263 (void) printf("Verified %s feature refcount "
8264 "of %llu is correct\n",
8265 spa_feature_table
[f
].fi_uname
,
8266 (longlong_t
)refcount
);
8271 rc
= verify_device_removal_feature_counts(spa
);
8274 if (rc
== 0 && (dump_opt
['b'] || dump_opt
['c']))
8275 rc
= dump_block_stats(spa
);
8278 rc
= verify_spacemap_refcounts(spa
);
8281 show_pool_stats(spa
);
8287 rc
= verify_checkpoint(spa
);
8290 dump_debug_buffer();
8295 #define ZDB_FLAG_CHECKSUM 0x0001
8296 #define ZDB_FLAG_DECOMPRESS 0x0002
8297 #define ZDB_FLAG_BSWAP 0x0004
8298 #define ZDB_FLAG_GBH 0x0008
8299 #define ZDB_FLAG_INDIRECT 0x0010
8300 #define ZDB_FLAG_RAW 0x0020
8301 #define ZDB_FLAG_PRINT_BLKPTR 0x0040
8302 #define ZDB_FLAG_VERBOSE 0x0080
8304 static int flagbits
[256];
8305 static char flagbitstr
[16];
8308 zdb_print_blkptr(const blkptr_t
*bp
, int flags
)
8310 char blkbuf
[BP_SPRINTF_LEN
];
8312 if (flags
& ZDB_FLAG_BSWAP
)
8313 byteswap_uint64_array((void *)bp
, sizeof (blkptr_t
));
8315 snprintf_blkptr(blkbuf
, sizeof (blkbuf
), bp
);
8316 (void) printf("%s\n", blkbuf
);
8320 zdb_dump_indirect(blkptr_t
*bp
, int nbps
, int flags
)
8324 for (i
= 0; i
< nbps
; i
++)
8325 zdb_print_blkptr(&bp
[i
], flags
);
8329 zdb_dump_gbh(void *buf
, int flags
)
8331 zdb_dump_indirect((blkptr_t
*)buf
, SPA_GBH_NBLKPTRS
, flags
);
8335 zdb_dump_block_raw(void *buf
, uint64_t size
, int flags
)
8337 if (flags
& ZDB_FLAG_BSWAP
)
8338 byteswap_uint64_array(buf
, size
);
8339 VERIFY(write(fileno(stdout
), buf
, size
) == size
);
8343 zdb_dump_block(char *label
, void *buf
, uint64_t size
, int flags
)
8345 uint64_t *d
= (uint64_t *)buf
;
8346 unsigned nwords
= size
/ sizeof (uint64_t);
8347 int do_bswap
= !!(flags
& ZDB_FLAG_BSWAP
);
8354 hdr
= " 7 6 5 4 3 2 1 0 f e d c b a 9 8";
8356 hdr
= " 0 1 2 3 4 5 6 7 8 9 a b c d e f";
8358 (void) printf("\n%s\n%6s %s 0123456789abcdef\n", label
, "", hdr
);
8360 #ifdef _LITTLE_ENDIAN
8361 /* correct the endianness */
8362 do_bswap
= !do_bswap
;
8364 for (i
= 0; i
< nwords
; i
+= 2) {
8365 (void) printf("%06llx: %016llx %016llx ",
8366 (u_longlong_t
)(i
* sizeof (uint64_t)),
8367 (u_longlong_t
)(do_bswap
? BSWAP_64(d
[i
]) : d
[i
]),
8368 (u_longlong_t
)(do_bswap
? BSWAP_64(d
[i
+ 1]) : d
[i
+ 1]));
8371 for (j
= 0; j
< 2 * sizeof (uint64_t); j
++)
8372 (void) printf("%c", isprint(c
[j
]) ? c
[j
] : '.');
8373 (void) printf("\n");
8378 * There are two acceptable formats:
8379 * leaf_name - For example: c1t0d0 or /tmp/ztest.0a
8380 * child[.child]* - For example: 0.1.1
8382 * The second form can be used to specify arbitrary vdevs anywhere
8383 * in the hierarchy. For example, in a pool with a mirror of
8384 * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
8387 zdb_vdev_lookup(vdev_t
*vdev
, const char *path
)
8395 /* First, assume the x.x.x.x format */
8396 i
= strtoul(path
, &s
, 10);
8397 if (s
== path
|| (s
&& *s
!= '.' && *s
!= '\0'))
8399 if (i
>= vdev
->vdev_children
)
8402 vdev
= vdev
->vdev_child
[i
];
8403 if (s
&& *s
== '\0')
8405 return (zdb_vdev_lookup(vdev
, s
+1));
8408 for (i
= 0; i
< vdev
->vdev_children
; i
++) {
8409 vdev_t
*vc
= vdev
->vdev_child
[i
];
8411 if (vc
->vdev_path
== NULL
) {
8412 vc
= zdb_vdev_lookup(vc
, path
);
8419 p
= strrchr(vc
->vdev_path
, '/');
8420 p
= p
? p
+ 1 : vc
->vdev_path
;
8421 q
= &vc
->vdev_path
[strlen(vc
->vdev_path
) - 2];
8423 if (strcmp(vc
->vdev_path
, path
) == 0)
8425 if (strcmp(p
, path
) == 0)
8427 if (strcmp(q
, "s0") == 0 && strncmp(p
, path
, q
- p
) == 0)
8435 name_from_objset_id(spa_t
*spa
, uint64_t objset_id
, char *outstr
)
8439 dsl_pool_config_enter(spa
->spa_dsl_pool
, FTAG
);
8440 int error
= dsl_dataset_hold_obj(spa
->spa_dsl_pool
, objset_id
,
8443 (void) fprintf(stderr
, "failed to hold objset %llu: %s\n",
8444 (u_longlong_t
)objset_id
, strerror(error
));
8445 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
8448 dsl_dataset_name(ds
, outstr
);
8449 dsl_dataset_rele(ds
, NULL
);
8450 dsl_pool_config_exit(spa
->spa_dsl_pool
, FTAG
);
8455 zdb_parse_block_sizes(char *sizes
, uint64_t *lsize
, uint64_t *psize
)
8457 char *s0
, *s1
, *tmp
= NULL
;
8462 s0
= strtok_r(sizes
, "/", &tmp
);
8465 s1
= strtok_r(NULL
, "/", &tmp
);
8466 *lsize
= strtoull(s0
, NULL
, 16);
8467 *psize
= s1
? strtoull(s1
, NULL
, 16) : *lsize
;
8468 return (*lsize
>= *psize
&& *psize
> 0);
8471 #define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg))
8474 zdb_decompress_block(abd_t
*pabd
, void *buf
, void *lbuf
, uint64_t lsize
,
8475 uint64_t psize
, int flags
)
8478 boolean_t exceeded
= B_FALSE
;
8480 * We don't know how the data was compressed, so just try
8481 * every decompress function at every inflated blocksize.
8483 void *lbuf2
= umem_alloc(SPA_MAXBLOCKSIZE
, UMEM_NOFAIL
);
8484 int cfuncs
[ZIO_COMPRESS_FUNCTIONS
] = { 0 };
8485 int *cfuncp
= cfuncs
;
8486 uint64_t maxlsize
= SPA_MAXBLOCKSIZE
;
8487 uint64_t mask
= ZIO_COMPRESS_MASK(ON
) | ZIO_COMPRESS_MASK(OFF
) |
8488 ZIO_COMPRESS_MASK(INHERIT
) | ZIO_COMPRESS_MASK(EMPTY
) |
8489 (getenv("ZDB_NO_ZLE") ? ZIO_COMPRESS_MASK(ZLE
) : 0);
8490 *cfuncp
++ = ZIO_COMPRESS_LZ4
;
8491 *cfuncp
++ = ZIO_COMPRESS_LZJB
;
8492 mask
|= ZIO_COMPRESS_MASK(LZ4
) | ZIO_COMPRESS_MASK(LZJB
);
8493 for (int c
= 0; c
< ZIO_COMPRESS_FUNCTIONS
; c
++)
8494 if (((1ULL << c
) & mask
) == 0)
8498 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
8499 * could take a while and we should let the user know
8500 * we are not stuck. On the other hand, printing progress
8501 * info gets old after a while. User can specify 'v' flag
8502 * to see the progression.
8505 lsize
+= SPA_MINBLOCKSIZE
;
8508 for (; lsize
<= maxlsize
; lsize
+= SPA_MINBLOCKSIZE
) {
8509 for (cfuncp
= cfuncs
; *cfuncp
; cfuncp
++) {
8510 if (flags
& ZDB_FLAG_VERBOSE
) {
8511 (void) fprintf(stderr
,
8512 "Trying %05llx -> %05llx (%s)\n",
8513 (u_longlong_t
)psize
,
8514 (u_longlong_t
)lsize
,
8515 zio_compress_table
[*cfuncp
].\
8520 * We randomize lbuf2, and decompress to both
8521 * lbuf and lbuf2. This way, we will know if
8522 * decompression fill exactly to lsize.
8524 VERIFY0(random_get_pseudo_bytes(lbuf2
, lsize
));
8526 if (zio_decompress_data(*cfuncp
, pabd
,
8527 lbuf
, psize
, lsize
, NULL
) == 0 &&
8528 zio_decompress_data(*cfuncp
, pabd
,
8529 lbuf2
, psize
, lsize
, NULL
) == 0 &&
8530 memcmp(lbuf
, lbuf2
, lsize
) == 0)
8536 umem_free(lbuf2
, SPA_MAXBLOCKSIZE
);
8538 if (lsize
> maxlsize
) {
8541 if (*cfuncp
== ZIO_COMPRESS_ZLE
) {
8542 printf("\nZLE decompression was selected. If you "
8543 "suspect the results are wrong,\ntry avoiding ZLE "
8544 "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
8551 * Read a block from a pool and print it out. The syntax of the
8552 * block descriptor is:
8554 * pool:vdev_specifier:offset:[lsize/]psize[:flags]
8556 * pool - The name of the pool you wish to read from
8557 * vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
8558 * offset - offset, in hex, in bytes
8559 * size - Amount of data to read, in hex, in bytes
8560 * flags - A string of characters specifying options
8561 * b: Decode a blkptr at given offset within block
8562 * c: Calculate and display checksums
8563 * d: Decompress data before dumping
8564 * e: Byteswap data before dumping
8565 * g: Display data as a gang block header
8566 * i: Display as an indirect block
8567 * r: Dump raw data to stdout
8572 zdb_read_block(char *thing
, spa_t
*spa
)
8574 blkptr_t blk
, *bp
= &blk
;
8575 dva_t
*dva
= bp
->blk_dva
;
8577 uint64_t offset
= 0, psize
= 0, lsize
= 0, blkptr_offset
= 0;
8582 char *s
, *p
, *dup
, *flagstr
, *sizes
, *tmp
= NULL
;
8583 const char *vdev
, *errmsg
= NULL
;
8585 boolean_t borrowed
= B_FALSE
, found
= B_FALSE
;
8587 dup
= strdup(thing
);
8588 s
= strtok_r(dup
, ":", &tmp
);
8590 s
= strtok_r(NULL
, ":", &tmp
);
8591 offset
= strtoull(s
? s
: "", NULL
, 16);
8592 sizes
= strtok_r(NULL
, ":", &tmp
);
8593 s
= strtok_r(NULL
, ":", &tmp
);
8594 flagstr
= strdup(s
?: "");
8596 if (!zdb_parse_block_sizes(sizes
, &lsize
, &psize
))
8597 errmsg
= "invalid size(s)";
8598 if (!IS_P2ALIGNED(psize
, DEV_BSIZE
) || !IS_P2ALIGNED(lsize
, DEV_BSIZE
))
8599 errmsg
= "size must be a multiple of sector size";
8600 if (!IS_P2ALIGNED(offset
, DEV_BSIZE
))
8601 errmsg
= "offset must be a multiple of sector size";
8603 (void) printf("Invalid block specifier: %s - %s\n",
8609 for (s
= strtok_r(flagstr
, ":", &tmp
);
8611 s
= strtok_r(NULL
, ":", &tmp
)) {
8612 for (i
= 0; i
< strlen(flagstr
); i
++) {
8613 int bit
= flagbits
[(uchar_t
)flagstr
[i
]];
8616 (void) printf("***Ignoring flag: %c\n",
8617 (uchar_t
)flagstr
[i
]);
8623 p
= &flagstr
[i
+ 1];
8624 if (*p
!= ':' && *p
!= '\0') {
8625 int j
= 0, nextbit
= flagbits
[(uchar_t
)*p
];
8626 char *end
, offstr
[8] = { 0 };
8627 if ((bit
== ZDB_FLAG_PRINT_BLKPTR
) &&
8629 /* look ahead to isolate the offset */
8630 while (nextbit
== 0 &&
8631 strchr(flagbitstr
, *p
) == NULL
) {
8634 if (i
+ j
> strlen(flagstr
))
8637 nextbit
= flagbits
[(uchar_t
)*p
];
8639 blkptr_offset
= strtoull(offstr
, &end
,
8642 } else if (nextbit
== 0) {
8643 (void) printf("***Ignoring flag arg:"
8644 " '%c'\n", (uchar_t
)*p
);
8649 if (blkptr_offset
% sizeof (blkptr_t
)) {
8650 printf("Block pointer offset 0x%llx "
8651 "must be divisible by 0x%x\n",
8652 (longlong_t
)blkptr_offset
, (int)sizeof (blkptr_t
));
8655 if (found
== B_FALSE
&& strlen(flagstr
) > 0) {
8656 printf("Invalid flag arg: '%s'\n", flagstr
);
8660 vd
= zdb_vdev_lookup(spa
->spa_root_vdev
, vdev
);
8662 (void) printf("***Invalid vdev: %s\n", vdev
);
8666 (void) fprintf(stderr
, "Found vdev: %s\n",
8669 (void) fprintf(stderr
, "Found vdev type: %s\n",
8670 vd
->vdev_ops
->vdev_op_type
);
8673 pabd
= abd_alloc_for_io(SPA_MAXBLOCKSIZE
, B_FALSE
);
8674 lbuf
= umem_alloc(SPA_MAXBLOCKSIZE
, UMEM_NOFAIL
);
8678 DVA_SET_VDEV(&dva
[0], vd
->vdev_id
);
8679 DVA_SET_OFFSET(&dva
[0], offset
);
8680 DVA_SET_GANG(&dva
[0], !!(flags
& ZDB_FLAG_GBH
));
8681 DVA_SET_ASIZE(&dva
[0], vdev_psize_to_asize(vd
, psize
));
8683 BP_SET_BIRTH(bp
, TXG_INITIAL
, TXG_INITIAL
);
8685 BP_SET_LSIZE(bp
, lsize
);
8686 BP_SET_PSIZE(bp
, psize
);
8687 BP_SET_COMPRESS(bp
, ZIO_COMPRESS_OFF
);
8688 BP_SET_CHECKSUM(bp
, ZIO_CHECKSUM_OFF
);
8689 BP_SET_TYPE(bp
, DMU_OT_NONE
);
8690 BP_SET_LEVEL(bp
, 0);
8691 BP_SET_DEDUP(bp
, 0);
8692 BP_SET_BYTEORDER(bp
, ZFS_HOST_BYTEORDER
);
8694 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8695 zio
= zio_root(spa
, NULL
, NULL
, 0);
8697 if (vd
== vd
->vdev_top
) {
8699 * Treat this as a normal block read.
8701 zio_nowait(zio_read(zio
, spa
, bp
, pabd
, psize
, NULL
, NULL
,
8702 ZIO_PRIORITY_SYNC_READ
,
8703 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
, NULL
));
8706 * Treat this as a vdev child I/O.
8708 zio_nowait(zio_vdev_child_io(zio
, bp
, vd
, offset
, pabd
,
8709 psize
, ZIO_TYPE_READ
, ZIO_PRIORITY_SYNC_READ
,
8710 ZIO_FLAG_DONT_PROPAGATE
| ZIO_FLAG_DONT_RETRY
|
8711 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
| ZIO_FLAG_OPTIONAL
,
8715 error
= zio_wait(zio
);
8716 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8719 (void) printf("Read of %s failed, error: %d\n", thing
, error
);
8723 uint64_t orig_lsize
= lsize
;
8725 if (flags
& ZDB_FLAG_DECOMPRESS
) {
8726 boolean_t failed
= zdb_decompress_block(pabd
, buf
, lbuf
,
8727 lsize
, psize
, flags
);
8729 (void) printf("Decompress of %s failed\n", thing
);
8733 buf
= abd_borrow_buf_copy(pabd
, lsize
);
8737 * Try to detect invalid block pointer. If invalid, try
8740 if ((flags
& ZDB_FLAG_PRINT_BLKPTR
|| flags
& ZDB_FLAG_INDIRECT
) &&
8741 !(flags
& ZDB_FLAG_DECOMPRESS
)) {
8742 const blkptr_t
*b
= (const blkptr_t
*)(void *)
8743 ((uintptr_t)buf
+ (uintptr_t)blkptr_offset
);
8744 if (zfs_blkptr_verify(spa
, b
,
8745 BLK_CONFIG_NEEDED
, BLK_VERIFY_ONLY
) == B_FALSE
) {
8746 abd_return_buf_copy(pabd
, buf
, lsize
);
8749 boolean_t failed
= zdb_decompress_block(pabd
, buf
,
8750 lbuf
, lsize
, psize
, flags
);
8751 b
= (const blkptr_t
*)(void *)
8752 ((uintptr_t)buf
+ (uintptr_t)blkptr_offset
);
8753 if (failed
|| zfs_blkptr_verify(spa
, b
,
8754 BLK_CONFIG_NEEDED
, BLK_VERIFY_LOG
) == B_FALSE
) {
8755 printf("invalid block pointer at this DVA\n");
8761 if (flags
& ZDB_FLAG_PRINT_BLKPTR
)
8762 zdb_print_blkptr((blkptr_t
*)(void *)
8763 ((uintptr_t)buf
+ (uintptr_t)blkptr_offset
), flags
);
8764 else if (flags
& ZDB_FLAG_RAW
)
8765 zdb_dump_block_raw(buf
, lsize
, flags
);
8766 else if (flags
& ZDB_FLAG_INDIRECT
)
8767 zdb_dump_indirect((blkptr_t
*)buf
,
8768 orig_lsize
/ sizeof (blkptr_t
), flags
);
8769 else if (flags
& ZDB_FLAG_GBH
)
8770 zdb_dump_gbh(buf
, flags
);
8772 zdb_dump_block(thing
, buf
, lsize
, flags
);
8775 * If :c was specified, iterate through the checksum table to
8776 * calculate and display each checksum for our specified
8779 if ((flags
& ZDB_FLAG_CHECKSUM
) && !(flags
& ZDB_FLAG_RAW
) &&
8780 !(flags
& ZDB_FLAG_GBH
)) {
8782 (void) printf("\n");
8783 for (enum zio_checksum ck
= ZIO_CHECKSUM_LABEL
;
8784 ck
< ZIO_CHECKSUM_FUNCTIONS
; ck
++) {
8786 if ((zio_checksum_table
[ck
].ci_flags
&
8787 ZCHECKSUM_FLAG_EMBEDDED
) ||
8788 ck
== ZIO_CHECKSUM_NOPARITY
) {
8791 BP_SET_CHECKSUM(bp
, ck
);
8792 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
8793 czio
= zio_root(spa
, NULL
, NULL
, ZIO_FLAG_CANFAIL
);
8796 if (vd
== vd
->vdev_top
) {
8797 zio_nowait(zio_read(czio
, spa
, bp
, pabd
, psize
,
8799 ZIO_PRIORITY_SYNC_READ
,
8800 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
|
8801 ZIO_FLAG_DONT_RETRY
, NULL
));
8803 zio_nowait(zio_vdev_child_io(czio
, bp
, vd
,
8804 offset
, pabd
, psize
, ZIO_TYPE_READ
,
8805 ZIO_PRIORITY_SYNC_READ
,
8806 ZIO_FLAG_DONT_PROPAGATE
|
8807 ZIO_FLAG_DONT_RETRY
|
8808 ZIO_FLAG_CANFAIL
| ZIO_FLAG_RAW
|
8809 ZIO_FLAG_SPECULATIVE
|
8810 ZIO_FLAG_OPTIONAL
, NULL
, NULL
));
8812 error
= zio_wait(czio
);
8813 if (error
== 0 || error
== ECKSUM
) {
8814 zio_t
*ck_zio
= zio_root(spa
, NULL
, NULL
, 0);
8816 DVA_GET_OFFSET(&bp
->blk_dva
[0]);
8818 zio_checksum_compute(ck_zio
, ck
, pabd
, lsize
);
8821 "cksum=%016llx:%016llx:%016llx:%016llx\n",
8822 zio_checksum_table
[ck
].ci_name
,
8823 (u_longlong_t
)bp
->blk_cksum
.zc_word
[0],
8824 (u_longlong_t
)bp
->blk_cksum
.zc_word
[1],
8825 (u_longlong_t
)bp
->blk_cksum
.zc_word
[2],
8826 (u_longlong_t
)bp
->blk_cksum
.zc_word
[3]);
8829 printf("error %d reading block\n", error
);
8831 spa_config_exit(spa
, SCL_STATE
, FTAG
);
8836 abd_return_buf_copy(pabd
, buf
, lsize
);
8840 umem_free(lbuf
, SPA_MAXBLOCKSIZE
);
8847 zdb_embedded_block(char *thing
)
8849 blkptr_t bp
= {{{{0}}}};
8850 unsigned long long *words
= (void *)&bp
;
8854 err
= sscanf(thing
, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
8855 "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
8856 words
+ 0, words
+ 1, words
+ 2, words
+ 3,
8857 words
+ 4, words
+ 5, words
+ 6, words
+ 7,
8858 words
+ 8, words
+ 9, words
+ 10, words
+ 11,
8859 words
+ 12, words
+ 13, words
+ 14, words
+ 15);
8861 (void) fprintf(stderr
, "invalid input format\n");
8864 ASSERT3U(BPE_GET_LSIZE(&bp
), <=, SPA_MAXBLOCKSIZE
);
8865 buf
= malloc(SPA_MAXBLOCKSIZE
);
8867 (void) fprintf(stderr
, "out of memory\n");
8870 err
= decode_embedded_bp(&bp
, buf
, BPE_GET_LSIZE(&bp
));
8872 (void) fprintf(stderr
, "decode failed: %u\n", err
);
8875 zdb_dump_block_raw(buf
, BPE_GET_LSIZE(&bp
), 0);
8879 /* check for valid hex or decimal numeric string */
8881 zdb_numeric(char *str
)
8885 if (strlen(str
) == 0)
8887 if (strncmp(str
, "0x", 2) == 0 || strncmp(str
, "0X", 2) == 0)
8889 for (; i
< strlen(str
); i
++) {
8890 if (!isxdigit(str
[i
]))
8897 main(int argc
, char **argv
)
8901 objset_t
*os
= NULL
;
8905 char **searchdirs
= NULL
;
8907 char *target
, *target_pool
, dsname
[ZFS_MAX_DATASET_NAME_LEN
];
8908 nvlist_t
*policy
= NULL
;
8909 uint64_t max_txg
= UINT64_MAX
;
8910 int64_t objset_id
= -1;
8912 int flags
= ZFS_IMPORT_MISSING_LOG
;
8913 int rewind
= ZPOOL_NEVER_REWIND
;
8914 char *spa_config_path_env
, *objset_str
;
8915 boolean_t target_is_spa
= B_TRUE
, dataset_lookup
= B_FALSE
;
8916 nvlist_t
*cfg
= NULL
;
8918 dprintf_setup(&argc
, argv
);
8921 * If there is an environment variable SPA_CONFIG_PATH it overrides
8922 * default spa_config_path setting. If -U flag is specified it will
8923 * override this environment variable settings once again.
8925 spa_config_path_env
= getenv("SPA_CONFIG_PATH");
8926 if (spa_config_path_env
!= NULL
)
8927 spa_config_path
= spa_config_path_env
;
8930 * For performance reasons, we set this tunable down. We do so before
8931 * the arg parsing section so that the user can override this value if
8934 zfs_btree_verify_intensity
= 3;
8936 struct option long_options
[] = {
8937 {"ignore-assertions", no_argument
, NULL
, 'A'},
8938 {"block-stats", no_argument
, NULL
, 'b'},
8939 {"backup", no_argument
, NULL
, 'B'},
8940 {"checksum", no_argument
, NULL
, 'c'},
8941 {"config", no_argument
, NULL
, 'C'},
8942 {"datasets", no_argument
, NULL
, 'd'},
8943 {"dedup-stats", no_argument
, NULL
, 'D'},
8944 {"exported", no_argument
, NULL
, 'e'},
8945 {"embedded-block-pointer", no_argument
, NULL
, 'E'},
8946 {"automatic-rewind", no_argument
, NULL
, 'F'},
8947 {"dump-debug-msg", no_argument
, NULL
, 'G'},
8948 {"history", no_argument
, NULL
, 'h'},
8949 {"intent-logs", no_argument
, NULL
, 'i'},
8950 {"inflight", required_argument
, NULL
, 'I'},
8951 {"checkpointed-state", no_argument
, NULL
, 'k'},
8952 {"key", required_argument
, NULL
, 'K'},
8953 {"label", no_argument
, NULL
, 'l'},
8954 {"disable-leak-tracking", no_argument
, NULL
, 'L'},
8955 {"metaslabs", no_argument
, NULL
, 'm'},
8956 {"metaslab-groups", no_argument
, NULL
, 'M'},
8957 {"numeric", no_argument
, NULL
, 'N'},
8958 {"option", required_argument
, NULL
, 'o'},
8959 {"object-lookups", no_argument
, NULL
, 'O'},
8960 {"path", required_argument
, NULL
, 'p'},
8961 {"parseable", no_argument
, NULL
, 'P'},
8962 {"skip-label", no_argument
, NULL
, 'q'},
8963 {"copy-object", no_argument
, NULL
, 'r'},
8964 {"read-block", no_argument
, NULL
, 'R'},
8965 {"io-stats", no_argument
, NULL
, 's'},
8966 {"simulate-dedup", no_argument
, NULL
, 'S'},
8967 {"txg", required_argument
, NULL
, 't'},
8968 {"brt-stats", no_argument
, NULL
, 'T'},
8969 {"uberblock", no_argument
, NULL
, 'u'},
8970 {"cachefile", required_argument
, NULL
, 'U'},
8971 {"verbose", no_argument
, NULL
, 'v'},
8972 {"verbatim", no_argument
, NULL
, 'V'},
8973 {"dump-blocks", required_argument
, NULL
, 'x'},
8974 {"extreme-rewind", no_argument
, NULL
, 'X'},
8975 {"all-reconstruction", no_argument
, NULL
, 'Y'},
8976 {"livelist", no_argument
, NULL
, 'y'},
8977 {"zstd-headers", no_argument
, NULL
, 'Z'},
8981 while ((c
= getopt_long(argc
, argv
,
8982 "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
8983 long_options
, NULL
)) != -1) {
9022 zfs_reconstruct_indirect_combinations_max
= INT_MAX
;
9023 zfs_deadman_enabled
= 0;
9025 /* NB: Sort single match options below. */
9027 max_inflight_bytes
= strtoull(optarg
, NULL
, 0);
9028 if (max_inflight_bytes
== 0) {
9029 (void) fprintf(stderr
, "maximum number "
9030 "of inflight bytes must be greater "
9037 key_material
= strdup(optarg
);
9038 /* redact key material in process table */
9039 while (*optarg
!= '\0') { *optarg
++ = '*'; }
9042 error
= set_global_var(optarg
);
9047 if (searchdirs
== NULL
) {
9048 searchdirs
= umem_alloc(sizeof (char *),
9051 char **tmp
= umem_alloc((nsearch
+ 1) *
9052 sizeof (char *), UMEM_NOFAIL
);
9053 memcpy(tmp
, searchdirs
, nsearch
*
9055 umem_free(searchdirs
,
9056 nsearch
* sizeof (char *));
9059 searchdirs
[nsearch
++] = optarg
;
9062 max_txg
= strtoull(optarg
, NULL
, 0);
9063 if (max_txg
< TXG_INITIAL
) {
9064 (void) fprintf(stderr
, "incorrect txg "
9065 "specified: %s\n", optarg
);
9070 spa_config_path
= optarg
;
9071 if (spa_config_path
[0] != '/') {
9072 (void) fprintf(stderr
,
9073 "cachefile must be an absolute path "
9074 "(i.e. start with a slash)\n");
9082 flags
= ZFS_IMPORT_VERBATIM
;
9085 vn_dumpdir
= optarg
;
9093 if (!dump_opt
['e'] && searchdirs
!= NULL
) {
9094 (void) fprintf(stderr
, "-p option requires use of -e\n");
9099 * ZDB does not typically re-read blocks; therefore limit the ARC
9100 * to 256 MB, which can be used entirely for metadata.
9102 zfs_arc_min
= 2ULL << SPA_MAXBLOCKSHIFT
;
9103 zfs_arc_max
= 256 * 1024 * 1024;
9107 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
9108 * "zdb -b" uses traversal prefetch which uses async reads.
9109 * For good performance, let several of them be active at once.
9111 zfs_vdev_async_read_max_active
= 10;
9114 * Disable reference tracking for better performance.
9116 reference_tracking_enable
= B_FALSE
;
9119 * Do not fail spa_load when spa_load_verify fails. This is needed
9120 * to load non-idle pools.
9122 spa_load_verify_dryrun
= B_TRUE
;
9125 * ZDB should have ability to read spacemaps.
9127 spa_mode_readable_spacemaps
= B_TRUE
;
9129 kernel_init(SPA_MODE_READ
);
9132 verbose
= MAX(verbose
, 1);
9134 for (c
= 0; c
< 256; c
++) {
9135 if (dump_all
&& strchr("ABeEFkKlLNOPrRSXy", c
) == NULL
)
9138 dump_opt
[c
] += verbose
;
9141 libspl_set_assert_ok((dump_opt
['A'] == 1) || (dump_opt
['A'] > 2));
9142 zfs_recover
= (dump_opt
['A'] > 1);
9146 if (argc
< 2 && dump_opt
['R'])
9149 if (dump_opt
['E']) {
9152 zdb_embedded_block(argv
[0]);
9157 if (!dump_opt
['e'] && dump_opt
['C']) {
9158 dump_cachefile(spa_config_path
);
9165 return (dump_label(argv
[0]));
9167 if (dump_opt
['X'] || dump_opt
['F'])
9168 rewind
= ZPOOL_DO_REWIND
|
9169 (dump_opt
['X'] ? ZPOOL_EXTREME_REWIND
: 0);
9172 if (dump_opt
['N'] && dump_opt
['d'] == 0)
9173 dump_opt
['d'] = dump_opt
['N'];
9175 if (nvlist_alloc(&policy
, NV_UNIQUE_NAME_TYPE
, 0) != 0 ||
9176 nvlist_add_uint64(policy
, ZPOOL_LOAD_REQUEST_TXG
, max_txg
) != 0 ||
9177 nvlist_add_uint32(policy
, ZPOOL_LOAD_REWIND_POLICY
, rewind
) != 0)
9178 fatal("internal error: %s", strerror(ENOMEM
));
9183 if (strpbrk(target
, "/@") != NULL
) {
9186 target_pool
= strdup(target
);
9187 *strpbrk(target_pool
, "/@") = '\0';
9189 target_is_spa
= B_FALSE
;
9190 targetlen
= strlen(target
);
9191 if (targetlen
&& target
[targetlen
- 1] == '/')
9192 target
[targetlen
- 1] = '\0';
9195 * See if an objset ID was supplied (-d <pool>/<objset ID>).
9196 * To disambiguate tank/100, consider the 100 as objsetID
9197 * if -N was given, otherwise 100 is an objsetID iff
9198 * tank/100 as a named dataset fails on lookup.
9200 objset_str
= strchr(target
, '/');
9201 if (objset_str
&& strlen(objset_str
) > 1 &&
9202 zdb_numeric(objset_str
+ 1)) {
9206 objset_id
= strtoull(objset_str
, &endptr
, 0);
9207 /* dataset 0 is the same as opening the pool */
9208 if (errno
== 0 && endptr
!= objset_str
&&
9211 dataset_lookup
= B_TRUE
;
9213 /* normal dataset name not an objset ID */
9214 if (endptr
== objset_str
) {
9217 } else if (objset_str
&& !zdb_numeric(objset_str
+ 1) &&
9219 printf("Supply a numeric objset ID with -N\n");
9223 target_pool
= target
;
9226 if (dump_opt
['e']) {
9227 importargs_t args
= { 0 };
9229 args
.paths
= nsearch
;
9230 args
.path
= searchdirs
;
9231 args
.can_be_active
= B_TRUE
;
9233 libpc_handle_t lpch
= {
9234 .lpc_lib_handle
= NULL
,
9235 .lpc_ops
= &libzpool_config_ops
,
9236 .lpc_printerr
= B_TRUE
9238 error
= zpool_find_config(&lpch
, target_pool
, &cfg
, &args
);
9242 if (nvlist_add_nvlist(cfg
,
9243 ZPOOL_LOAD_POLICY
, policy
) != 0) {
9244 fatal("can't open '%s': %s",
9245 target
, strerror(ENOMEM
));
9248 if (dump_opt
['C'] > 1) {
9249 (void) printf("\nConfiguration for import:\n");
9250 dump_nvlist(cfg
, 8);
9254 * Disable the activity check to allow examination of
9257 error
= spa_import(target_pool
, cfg
, NULL
,
9258 flags
| ZFS_IMPORT_SKIP_MMP
);
9262 if (searchdirs
!= NULL
) {
9263 umem_free(searchdirs
, nsearch
* sizeof (char *));
9268 * We need to make sure to process -O option or call
9269 * dump_path after the -e option has been processed,
9270 * which imports the pool to the namespace if it's
9271 * not in the cachefile.
9273 if (dump_opt
['O']) {
9276 dump_opt
['v'] = verbose
+ 3;
9277 return (dump_path(argv
[0], argv
[1], NULL
));
9280 if (dump_opt
['r']) {
9281 target_is_spa
= B_FALSE
;
9284 dump_opt
['v'] = verbose
;
9285 error
= dump_path(argv
[0], argv
[1], &object
);
9287 fatal("internal error: %s", strerror(error
));
9291 * import_checkpointed_state makes the assumption that the
9292 * target pool that we pass it is already part of the spa
9293 * namespace. Because of that we need to make sure to call
9294 * it always after the -e option has been processed, which
9295 * imports the pool to the namespace if it's not in the
9298 char *checkpoint_pool
= NULL
;
9299 char *checkpoint_target
= NULL
;
9300 if (dump_opt
['k']) {
9301 checkpoint_pool
= import_checkpointed_state(target
, cfg
,
9302 &checkpoint_target
);
9304 if (checkpoint_target
!= NULL
)
9305 target
= checkpoint_target
;
9313 if (target_pool
!= target
)
9317 if (dump_opt
['k'] && (target_is_spa
|| dump_opt
['R'])) {
9318 ASSERT(checkpoint_pool
!= NULL
);
9319 ASSERT(checkpoint_target
== NULL
);
9321 error
= spa_open(checkpoint_pool
, &spa
, FTAG
);
9323 fatal("Tried to open pool \"%s\" but "
9324 "spa_open() failed with error %d\n",
9325 checkpoint_pool
, error
);
9328 } else if (target_is_spa
|| dump_opt
['R'] || dump_opt
['B'] ||
9330 zdb_set_skip_mmp(target
);
9331 error
= spa_open_rewind(target
, &spa
, FTAG
, policy
,
9335 * If we're missing the log device then
9336 * try opening the pool after clearing the
9339 mutex_enter(&spa_namespace_lock
);
9340 if ((spa
= spa_lookup(target
)) != NULL
&&
9341 spa
->spa_log_state
== SPA_LOG_MISSING
) {
9342 spa
->spa_log_state
= SPA_LOG_CLEAR
;
9345 mutex_exit(&spa_namespace_lock
);
9348 error
= spa_open_rewind(target
, &spa
,
9349 FTAG
, policy
, NULL
);
9352 } else if (strpbrk(target
, "#") != NULL
) {
9354 error
= dsl_pool_hold(target
, FTAG
, &dp
);
9356 fatal("can't dump '%s': %s", target
,
9359 error
= dump_bookmark(dp
, target
, B_TRUE
, verbose
> 1);
9360 dsl_pool_rele(dp
, FTAG
);
9362 fatal("can't dump '%s': %s", target
,
9367 target_pool
= strdup(target
);
9368 if (strpbrk(target
, "/@") != NULL
)
9369 *strpbrk(target_pool
, "/@") = '\0';
9371 zdb_set_skip_mmp(target
);
9373 * If -N was supplied, the user has indicated that
9374 * zdb -d <pool>/<objsetID> is in effect. Otherwise
9375 * we first assume that the dataset string is the
9376 * dataset name. If dmu_objset_hold fails with the
9377 * dataset string, and we have an objset_id, retry the
9378 * lookup with the objsetID.
9380 boolean_t retry
= B_TRUE
;
9382 if (dataset_lookup
== B_TRUE
) {
9384 * Use the supplied id to get the name
9387 error
= spa_open(target_pool
, &spa
, FTAG
);
9389 error
= name_from_objset_id(spa
,
9391 spa_close(spa
, FTAG
);
9397 if (objset_id
> 0 && retry
) {
9398 int err
= dmu_objset_hold(target
, FTAG
,
9401 dataset_lookup
= B_TRUE
;
9405 dmu_objset_rele(os
, FTAG
);
9408 error
= open_objset(target
, FTAG
, &os
);
9411 spa
= dmu_objset_spa(os
);
9415 nvlist_free(policy
);
9418 fatal("can't open '%s': %s", target
, strerror(error
));
9421 * Set the pool failure mode to panic in order to prevent the pool
9422 * from suspending. A suspended I/O will have no way to resume and
9423 * can prevent the zdb(8) command from terminating as expected.
9426 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
9430 if (dump_opt
['r']) {
9431 error
= zdb_copy_object(os
, object
, argv
[1]);
9432 } else if (!dump_opt
['R']) {
9433 flagbits
['d'] = ZOR_FLAG_DIRECTORY
;
9434 flagbits
['f'] = ZOR_FLAG_PLAIN_FILE
;
9435 flagbits
['m'] = ZOR_FLAG_SPACE_MAP
;
9436 flagbits
['z'] = ZOR_FLAG_ZAP
;
9437 flagbits
['A'] = ZOR_FLAG_ALL_TYPES
;
9439 if (argc
> 0 && dump_opt
['d']) {
9440 zopt_object_args
= argc
;
9441 zopt_object_ranges
= calloc(zopt_object_args
,
9442 sizeof (zopt_object_range_t
));
9443 for (unsigned i
= 0; i
< zopt_object_args
; i
++) {
9445 const char *msg
= NULL
;
9447 err
= parse_object_range(argv
[i
],
9448 &zopt_object_ranges
[i
], &msg
);
9450 fatal("Bad object or range: '%s': %s\n",
9451 argv
[i
], msg
?: "");
9453 } else if (argc
> 0 && dump_opt
['m']) {
9454 zopt_metaslab_args
= argc
;
9455 zopt_metaslab
= calloc(zopt_metaslab_args
,
9457 for (unsigned i
= 0; i
< zopt_metaslab_args
; i
++) {
9459 zopt_metaslab
[i
] = strtoull(argv
[i
], NULL
, 0);
9460 if (zopt_metaslab
[i
] == 0 && errno
!= 0)
9461 fatal("bad number %s: %s", argv
[i
],
9465 if (dump_opt
['B']) {
9466 dump_backup(target
, objset_id
,
9467 argc
> 0 ? argv
[0] : NULL
);
9468 } else if (os
!= NULL
) {
9470 } else if (zopt_object_args
> 0 && !dump_opt
['m']) {
9471 dump_objset(spa
->spa_meta_objset
);
9476 flagbits
['b'] = ZDB_FLAG_PRINT_BLKPTR
;
9477 flagbits
['c'] = ZDB_FLAG_CHECKSUM
;
9478 flagbits
['d'] = ZDB_FLAG_DECOMPRESS
;
9479 flagbits
['e'] = ZDB_FLAG_BSWAP
;
9480 flagbits
['g'] = ZDB_FLAG_GBH
;
9481 flagbits
['i'] = ZDB_FLAG_INDIRECT
;
9482 flagbits
['r'] = ZDB_FLAG_RAW
;
9483 flagbits
['v'] = ZDB_FLAG_VERBOSE
;
9485 for (int i
= 0; i
< argc
; i
++)
9486 zdb_read_block(argv
[i
], spa
);
9489 if (dump_opt
['k']) {
9490 free(checkpoint_pool
);
9492 free(checkpoint_target
);
9496 close_objset(os
, FTAG
);
9498 spa_close(spa
, FTAG
);
9501 fuid_table_destroy();
9503 dump_debug_buffer();