xref: /freebsd/sys/contrib/openzfs/cmd/zdb/zdb.c (revision 718519f4efc71096422fc71dab90b2a3369871ff)
1 /*
2  * CDDL HEADER START
3  *
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.
7  *
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.
12  *
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]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
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]
31  *
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, 2024, Klara Inc.
37  * Copyright (c) 2023, Rob Norris <robn@despairlabs.com>
38  */
39 
40 #include <stdio.h>
41 #include <unistd.h>
42 #include <stdlib.h>
43 #include <ctype.h>
44 #include <getopt.h>
45 #include <openssl/evp.h>
46 #include <sys/zfs_context.h>
47 #include <sys/spa.h>
48 #include <sys/spa_impl.h>
49 #include <sys/dmu.h>
50 #include <sys/zap.h>
51 #include <sys/zap_impl.h>
52 #include <sys/fs/zfs.h>
53 #include <sys/zfs_znode.h>
54 #include <sys/zfs_sa.h>
55 #include <sys/sa.h>
56 #include <sys/sa_impl.h>
57 #include <sys/vdev.h>
58 #include <sys/vdev_impl.h>
59 #include <sys/metaslab_impl.h>
60 #include <sys/dmu_objset.h>
61 #include <sys/dsl_dir.h>
62 #include <sys/dsl_dataset.h>
63 #include <sys/dsl_pool.h>
64 #include <sys/dsl_bookmark.h>
65 #include <sys/dbuf.h>
66 #include <sys/zil.h>
67 #include <sys/zil_impl.h>
68 #include <sys/stat.h>
69 #include <sys/resource.h>
70 #include <sys/dmu_send.h>
71 #include <sys/dmu_traverse.h>
72 #include <sys/zio_checksum.h>
73 #include <sys/zio_compress.h>
74 #include <sys/zfs_fuid.h>
75 #include <sys/arc.h>
76 #include <sys/arc_impl.h>
77 #include <sys/ddt.h>
78 #include <sys/ddt_impl.h>
79 #include <sys/zfeature.h>
80 #include <sys/abd.h>
81 #include <sys/blkptr.h>
82 #include <sys/dsl_crypt.h>
83 #include <sys/dsl_scan.h>
84 #include <sys/btree.h>
85 #include <sys/brt.h>
86 #include <sys/brt_impl.h>
87 #include <zfs_comutil.h>
88 #include <sys/zstd/zstd.h>
89 #include <sys/backtrace.h>
90 
91 #include <libnvpair.h>
92 #include <libzutil.h>
93 #include <libzfs_core.h>
94 
95 #include <libzdb.h>
96 
97 #include "zdb.h"
98 
99 
100 extern int reference_tracking_enable;
101 extern int zfs_recover;
102 extern uint_t zfs_vdev_async_read_max_active;
103 extern boolean_t spa_load_verify_dryrun;
104 extern boolean_t spa_mode_readable_spacemaps;
105 extern uint_t zfs_reconstruct_indirect_combinations_max;
106 extern uint_t zfs_btree_verify_intensity;
107 
108 static const char cmdname[] = "zdb";
109 uint8_t dump_opt[256];
110 
111 typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size);
112 
113 static uint64_t *zopt_metaslab = NULL;
114 static unsigned zopt_metaslab_args = 0;
115 
116 
117 static zopt_object_range_t *zopt_object_ranges = NULL;
118 static unsigned zopt_object_args = 0;
119 
120 static int flagbits[256];
121 
122 
123 static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */
124 static int leaked_objects = 0;
125 static range_tree_t *mos_refd_objs;
126 static spa_t *spa;
127 static objset_t *os;
128 static boolean_t kernel_init_done;
129 
130 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *,
131     boolean_t);
132 static void mos_obj_refd(uint64_t);
133 static void mos_obj_refd_multiple(uint64_t);
134 static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free,
135     dmu_tx_t *tx);
136 
137 
138 
139 static void zdb_print_blkptr(const blkptr_t *bp, int flags);
140 static void zdb_exit(int reason);
141 
142 typedef struct sublivelist_verify_block_refcnt {
143 	/* block pointer entry in livelist being verified */
144 	blkptr_t svbr_blk;
145 
146 	/*
147 	 * Refcount gets incremented to 1 when we encounter the first
148 	 * FREE entry for the svfbr block pointer and a node for it
149 	 * is created in our ZDB verification/tracking metadata.
150 	 *
151 	 * As we encounter more FREE entries we increment this counter
152 	 * and similarly decrement it whenever we find the respective
153 	 * ALLOC entries for this block.
154 	 *
155 	 * When the refcount gets to 0 it means that all the FREE and
156 	 * ALLOC entries of this block have paired up and we no longer
157 	 * need to track it in our verification logic (e.g. the node
158 	 * containing this struct in our verification data structure
159 	 * should be freed).
160 	 *
161 	 * [refer to sublivelist_verify_blkptr() for the actual code]
162 	 */
163 	uint32_t svbr_refcnt;
164 } sublivelist_verify_block_refcnt_t;
165 
166 static int
sublivelist_block_refcnt_compare(const void * larg,const void * rarg)167 sublivelist_block_refcnt_compare(const void *larg, const void *rarg)
168 {
169 	const sublivelist_verify_block_refcnt_t *l = larg;
170 	const sublivelist_verify_block_refcnt_t *r = rarg;
171 	return (livelist_compare(&l->svbr_blk, &r->svbr_blk));
172 }
173 
174 static int
sublivelist_verify_blkptr(void * arg,const blkptr_t * bp,boolean_t free,dmu_tx_t * tx)175 sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free,
176     dmu_tx_t *tx)
177 {
178 	ASSERT3P(tx, ==, NULL);
179 	struct sublivelist_verify *sv = arg;
180 	sublivelist_verify_block_refcnt_t current = {
181 			.svbr_blk = *bp,
182 
183 			/*
184 			 * Start with 1 in case this is the first free entry.
185 			 * This field is not used for our B-Tree comparisons
186 			 * anyway.
187 			 */
188 			.svbr_refcnt = 1,
189 	};
190 
191 	zfs_btree_index_t where;
192 	sublivelist_verify_block_refcnt_t *pair =
193 	    zfs_btree_find(&sv->sv_pair, &current, &where);
194 	if (free) {
195 		if (pair == NULL) {
196 			/* first free entry for this block pointer */
197 			zfs_btree_add(&sv->sv_pair, &current);
198 		} else {
199 			pair->svbr_refcnt++;
200 		}
201 	} else {
202 		if (pair == NULL) {
203 			/* block that is currently marked as allocated */
204 			for (int i = 0; i < SPA_DVAS_PER_BP; i++) {
205 				if (DVA_IS_EMPTY(&bp->blk_dva[i]))
206 					break;
207 				sublivelist_verify_block_t svb = {
208 				    .svb_dva = bp->blk_dva[i],
209 				    .svb_allocated_txg =
210 				    BP_GET_LOGICAL_BIRTH(bp)
211 				};
212 
213 				if (zfs_btree_find(&sv->sv_leftover, &svb,
214 				    &where) == NULL) {
215 					zfs_btree_add_idx(&sv->sv_leftover,
216 					    &svb, &where);
217 				}
218 			}
219 		} else {
220 			/* alloc matches a free entry */
221 			pair->svbr_refcnt--;
222 			if (pair->svbr_refcnt == 0) {
223 				/* all allocs and frees have been matched */
224 				zfs_btree_remove_idx(&sv->sv_pair, &where);
225 			}
226 		}
227 	}
228 
229 	return (0);
230 }
231 
232 static int
sublivelist_verify_func(void * args,dsl_deadlist_entry_t * dle)233 sublivelist_verify_func(void *args, dsl_deadlist_entry_t *dle)
234 {
235 	int err;
236 	struct sublivelist_verify *sv = args;
237 
238 	zfs_btree_create(&sv->sv_pair, sublivelist_block_refcnt_compare, NULL,
239 	    sizeof (sublivelist_verify_block_refcnt_t));
240 
241 	err = bpobj_iterate_nofree(&dle->dle_bpobj, sublivelist_verify_blkptr,
242 	    sv, NULL);
243 
244 	sublivelist_verify_block_refcnt_t *e;
245 	zfs_btree_index_t *cookie = NULL;
246 	while ((e = zfs_btree_destroy_nodes(&sv->sv_pair, &cookie)) != NULL) {
247 		char blkbuf[BP_SPRINTF_LEN];
248 		snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
249 		    &e->svbr_blk, B_TRUE);
250 		(void) printf("\tERROR: %d unmatched FREE(s): %s\n",
251 		    e->svbr_refcnt, blkbuf);
252 	}
253 	zfs_btree_destroy(&sv->sv_pair);
254 
255 	return (err);
256 }
257 
258 static int
livelist_block_compare(const void * larg,const void * rarg)259 livelist_block_compare(const void *larg, const void *rarg)
260 {
261 	const sublivelist_verify_block_t *l = larg;
262 	const sublivelist_verify_block_t *r = rarg;
263 
264 	if (DVA_GET_VDEV(&l->svb_dva) < DVA_GET_VDEV(&r->svb_dva))
265 		return (-1);
266 	else if (DVA_GET_VDEV(&l->svb_dva) > DVA_GET_VDEV(&r->svb_dva))
267 		return (+1);
268 
269 	if (DVA_GET_OFFSET(&l->svb_dva) < DVA_GET_OFFSET(&r->svb_dva))
270 		return (-1);
271 	else if (DVA_GET_OFFSET(&l->svb_dva) > DVA_GET_OFFSET(&r->svb_dva))
272 		return (+1);
273 
274 	if (DVA_GET_ASIZE(&l->svb_dva) < DVA_GET_ASIZE(&r->svb_dva))
275 		return (-1);
276 	else if (DVA_GET_ASIZE(&l->svb_dva) > DVA_GET_ASIZE(&r->svb_dva))
277 		return (+1);
278 
279 	return (0);
280 }
281 
282 /*
283  * Check for errors in a livelist while tracking all unfreed ALLOCs in the
284  * sublivelist_verify_t: sv->sv_leftover
285  */
286 static void
livelist_verify(dsl_deadlist_t * dl,void * arg)287 livelist_verify(dsl_deadlist_t *dl, void *arg)
288 {
289 	sublivelist_verify_t *sv = arg;
290 	dsl_deadlist_iterate(dl, sublivelist_verify_func, sv);
291 }
292 
293 /*
294  * Check for errors in the livelist entry and discard the intermediary
295  * data structures
296  */
297 static int
sublivelist_verify_lightweight(void * args,dsl_deadlist_entry_t * dle)298 sublivelist_verify_lightweight(void *args, dsl_deadlist_entry_t *dle)
299 {
300 	(void) args;
301 	sublivelist_verify_t sv;
302 	zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
303 	    sizeof (sublivelist_verify_block_t));
304 	int err = sublivelist_verify_func(&sv, dle);
305 	zfs_btree_clear(&sv.sv_leftover);
306 	zfs_btree_destroy(&sv.sv_leftover);
307 	return (err);
308 }
309 
310 typedef struct metaslab_verify {
311 	/*
312 	 * Tree containing all the leftover ALLOCs from the livelists
313 	 * that are part of this metaslab.
314 	 */
315 	zfs_btree_t mv_livelist_allocs;
316 
317 	/*
318 	 * Metaslab information.
319 	 */
320 	uint64_t mv_vdid;
321 	uint64_t mv_msid;
322 	uint64_t mv_start;
323 	uint64_t mv_end;
324 
325 	/*
326 	 * What's currently allocated for this metaslab.
327 	 */
328 	range_tree_t *mv_allocated;
329 } metaslab_verify_t;
330 
331 typedef void ll_iter_t(dsl_deadlist_t *ll, void *arg);
332 
333 typedef int (*zdb_log_sm_cb_t)(spa_t *spa, space_map_entry_t *sme, uint64_t txg,
334     void *arg);
335 
336 typedef struct unflushed_iter_cb_arg {
337 	spa_t *uic_spa;
338 	uint64_t uic_txg;
339 	void *uic_arg;
340 	zdb_log_sm_cb_t uic_cb;
341 } unflushed_iter_cb_arg_t;
342 
343 static int
iterate_through_spacemap_logs_cb(space_map_entry_t * sme,void * arg)344 iterate_through_spacemap_logs_cb(space_map_entry_t *sme, void *arg)
345 {
346 	unflushed_iter_cb_arg_t *uic = arg;
347 	return (uic->uic_cb(uic->uic_spa, sme, uic->uic_txg, uic->uic_arg));
348 }
349 
350 static void
iterate_through_spacemap_logs(spa_t * spa,zdb_log_sm_cb_t cb,void * arg)351 iterate_through_spacemap_logs(spa_t *spa, zdb_log_sm_cb_t cb, void *arg)
352 {
353 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
354 		return;
355 
356 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
357 	for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
358 	    sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
359 		space_map_t *sm = NULL;
360 		VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
361 		    sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
362 
363 		unflushed_iter_cb_arg_t uic = {
364 			.uic_spa = spa,
365 			.uic_txg = sls->sls_txg,
366 			.uic_arg = arg,
367 			.uic_cb = cb
368 		};
369 		VERIFY0(space_map_iterate(sm, space_map_length(sm),
370 		    iterate_through_spacemap_logs_cb, &uic));
371 		space_map_close(sm);
372 	}
373 	spa_config_exit(spa, SCL_CONFIG, FTAG);
374 }
375 
376 static void
verify_livelist_allocs(metaslab_verify_t * mv,uint64_t txg,uint64_t offset,uint64_t size)377 verify_livelist_allocs(metaslab_verify_t *mv, uint64_t txg,
378     uint64_t offset, uint64_t size)
379 {
380 	sublivelist_verify_block_t svb = {{{0}}};
381 	DVA_SET_VDEV(&svb.svb_dva, mv->mv_vdid);
382 	DVA_SET_OFFSET(&svb.svb_dva, offset);
383 	DVA_SET_ASIZE(&svb.svb_dva, size);
384 	zfs_btree_index_t where;
385 	uint64_t end_offset = offset + size;
386 
387 	/*
388 	 *  Look for an exact match for spacemap entry in the livelist entries.
389 	 *  Then, look for other livelist entries that fall within the range
390 	 *  of the spacemap entry as it may have been condensed
391 	 */
392 	sublivelist_verify_block_t *found =
393 	    zfs_btree_find(&mv->mv_livelist_allocs, &svb, &where);
394 	if (found == NULL) {
395 		found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where);
396 	}
397 	for (; found != NULL && DVA_GET_VDEV(&found->svb_dva) == mv->mv_vdid &&
398 	    DVA_GET_OFFSET(&found->svb_dva) < end_offset;
399 	    found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
400 		if (found->svb_allocated_txg <= txg) {
401 			(void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
402 			    "from TXG %llx FREED at TXG %llx\n",
403 			    (u_longlong_t)DVA_GET_OFFSET(&found->svb_dva),
404 			    (u_longlong_t)DVA_GET_ASIZE(&found->svb_dva),
405 			    (u_longlong_t)found->svb_allocated_txg,
406 			    (u_longlong_t)txg);
407 		}
408 	}
409 }
410 
411 static int
metaslab_spacemap_validation_cb(space_map_entry_t * sme,void * arg)412 metaslab_spacemap_validation_cb(space_map_entry_t *sme, void *arg)
413 {
414 	metaslab_verify_t *mv = arg;
415 	uint64_t offset = sme->sme_offset;
416 	uint64_t size = sme->sme_run;
417 	uint64_t txg = sme->sme_txg;
418 
419 	if (sme->sme_type == SM_ALLOC) {
420 		if (range_tree_contains(mv->mv_allocated,
421 		    offset, size)) {
422 			(void) printf("ERROR: DOUBLE ALLOC: "
423 			    "%llu [%llx:%llx] "
424 			    "%llu:%llu LOG_SM\n",
425 			    (u_longlong_t)txg, (u_longlong_t)offset,
426 			    (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
427 			    (u_longlong_t)mv->mv_msid);
428 		} else {
429 			range_tree_add(mv->mv_allocated,
430 			    offset, size);
431 		}
432 	} else {
433 		if (!range_tree_contains(mv->mv_allocated,
434 		    offset, size)) {
435 			(void) printf("ERROR: DOUBLE FREE: "
436 			    "%llu [%llx:%llx] "
437 			    "%llu:%llu LOG_SM\n",
438 			    (u_longlong_t)txg, (u_longlong_t)offset,
439 			    (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
440 			    (u_longlong_t)mv->mv_msid);
441 		} else {
442 			range_tree_remove(mv->mv_allocated,
443 			    offset, size);
444 		}
445 	}
446 
447 	if (sme->sme_type != SM_ALLOC) {
448 		/*
449 		 * If something is freed in the spacemap, verify that
450 		 * it is not listed as allocated in the livelist.
451 		 */
452 		verify_livelist_allocs(mv, txg, offset, size);
453 	}
454 	return (0);
455 }
456 
457 static int
spacemap_check_sm_log_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)458 spacemap_check_sm_log_cb(spa_t *spa, space_map_entry_t *sme,
459     uint64_t txg, void *arg)
460 {
461 	metaslab_verify_t *mv = arg;
462 	uint64_t offset = sme->sme_offset;
463 	uint64_t vdev_id = sme->sme_vdev;
464 
465 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
466 
467 	/* skip indirect vdevs */
468 	if (!vdev_is_concrete(vd))
469 		return (0);
470 
471 	if (vdev_id != mv->mv_vdid)
472 		return (0);
473 
474 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
475 	if (ms->ms_id != mv->mv_msid)
476 		return (0);
477 
478 	if (txg < metaslab_unflushed_txg(ms))
479 		return (0);
480 
481 
482 	ASSERT3U(txg, ==, sme->sme_txg);
483 	return (metaslab_spacemap_validation_cb(sme, mv));
484 }
485 
486 static void
spacemap_check_sm_log(spa_t * spa,metaslab_verify_t * mv)487 spacemap_check_sm_log(spa_t *spa, metaslab_verify_t *mv)
488 {
489 	iterate_through_spacemap_logs(spa, spacemap_check_sm_log_cb, mv);
490 }
491 
492 static void
spacemap_check_ms_sm(space_map_t * sm,metaslab_verify_t * mv)493 spacemap_check_ms_sm(space_map_t  *sm, metaslab_verify_t *mv)
494 {
495 	if (sm == NULL)
496 		return;
497 
498 	VERIFY0(space_map_iterate(sm, space_map_length(sm),
499 	    metaslab_spacemap_validation_cb, mv));
500 }
501 
502 static void iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg);
503 
504 /*
505  * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
506  * they are part of that metaslab (mv_msid).
507  */
508 static void
mv_populate_livelist_allocs(metaslab_verify_t * mv,sublivelist_verify_t * sv)509 mv_populate_livelist_allocs(metaslab_verify_t *mv, sublivelist_verify_t *sv)
510 {
511 	zfs_btree_index_t where;
512 	sublivelist_verify_block_t *svb;
513 	ASSERT3U(zfs_btree_numnodes(&mv->mv_livelist_allocs), ==, 0);
514 	for (svb = zfs_btree_first(&sv->sv_leftover, &where);
515 	    svb != NULL;
516 	    svb = zfs_btree_next(&sv->sv_leftover, &where, &where)) {
517 		if (DVA_GET_VDEV(&svb->svb_dva) != mv->mv_vdid)
518 			continue;
519 
520 		if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start &&
521 		    (DVA_GET_OFFSET(&svb->svb_dva) +
522 		    DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_start) {
523 			(void) printf("ERROR: Found block that crosses "
524 			    "metaslab boundary: <%llu:%llx:%llx>\n",
525 			    (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
526 			    (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
527 			    (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
528 			continue;
529 		}
530 
531 		if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start)
532 			continue;
533 
534 		if (DVA_GET_OFFSET(&svb->svb_dva) >= mv->mv_end)
535 			continue;
536 
537 		if ((DVA_GET_OFFSET(&svb->svb_dva) +
538 		    DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_end) {
539 			(void) printf("ERROR: Found block that crosses "
540 			    "metaslab boundary: <%llu:%llx:%llx>\n",
541 			    (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
542 			    (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
543 			    (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
544 			continue;
545 		}
546 
547 		zfs_btree_add(&mv->mv_livelist_allocs, svb);
548 	}
549 
550 	for (svb = zfs_btree_first(&mv->mv_livelist_allocs, &where);
551 	    svb != NULL;
552 	    svb = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
553 		zfs_btree_remove(&sv->sv_leftover, svb);
554 	}
555 }
556 
557 /*
558  * [Livelist Check]
559  * Iterate through all the sublivelists and:
560  * - report leftover frees (**)
561  * - record leftover ALLOCs together with their TXG [see Cross Check]
562  *
563  * (**) Note: Double ALLOCs are valid in datasets that have dedup
564  *      enabled. Similarly double FREEs are allowed as well but
565  *      only if they pair up with a corresponding ALLOC entry once
566  *      we our done with our sublivelist iteration.
567  *
568  * [Spacemap Check]
569  * for each metaslab:
570  * - iterate over spacemap and then the metaslab's entries in the
571  *   spacemap log, then report any double FREEs and ALLOCs (do not
572  *   blow up).
573  *
574  * [Cross Check]
575  * After finishing the Livelist Check phase and while being in the
576  * Spacemap Check phase, we find all the recorded leftover ALLOCs
577  * of the livelist check that are part of the metaslab that we are
578  * currently looking at in the Spacemap Check. We report any entries
579  * that are marked as ALLOCs in the livelists but have been actually
580  * freed (and potentially allocated again) after their TXG stamp in
581  * the spacemaps. Also report any ALLOCs from the livelists that
582  * belong to indirect vdevs (e.g. their vdev completed removal).
583  *
584  * Note that this will miss Log Spacemap entries that cancelled each other
585  * out before being flushed to the metaslab, so we are not guaranteed
586  * to match all erroneous ALLOCs.
587  */
588 static void
livelist_metaslab_validate(spa_t * spa)589 livelist_metaslab_validate(spa_t *spa)
590 {
591 	(void) printf("Verifying deleted livelist entries\n");
592 
593 	sublivelist_verify_t sv;
594 	zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
595 	    sizeof (sublivelist_verify_block_t));
596 	iterate_deleted_livelists(spa, livelist_verify, &sv);
597 
598 	(void) printf("Verifying metaslab entries\n");
599 	vdev_t *rvd = spa->spa_root_vdev;
600 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
601 		vdev_t *vd = rvd->vdev_child[c];
602 
603 		if (!vdev_is_concrete(vd))
604 			continue;
605 
606 		for (uint64_t mid = 0; mid < vd->vdev_ms_count; mid++) {
607 			metaslab_t *m = vd->vdev_ms[mid];
608 
609 			(void) fprintf(stderr,
610 			    "\rverifying concrete vdev %llu, "
611 			    "metaslab %llu of %llu ...",
612 			    (longlong_t)vd->vdev_id,
613 			    (longlong_t)mid,
614 			    (longlong_t)vd->vdev_ms_count);
615 
616 			uint64_t shift, start;
617 			range_seg_type_t type =
618 			    metaslab_calculate_range_tree_type(vd, m,
619 			    &start, &shift);
620 			metaslab_verify_t mv;
621 			mv.mv_allocated = range_tree_create(NULL,
622 			    type, NULL, start, shift);
623 			mv.mv_vdid = vd->vdev_id;
624 			mv.mv_msid = m->ms_id;
625 			mv.mv_start = m->ms_start;
626 			mv.mv_end = m->ms_start + m->ms_size;
627 			zfs_btree_create(&mv.mv_livelist_allocs,
628 			    livelist_block_compare, NULL,
629 			    sizeof (sublivelist_verify_block_t));
630 
631 			mv_populate_livelist_allocs(&mv, &sv);
632 
633 			spacemap_check_ms_sm(m->ms_sm, &mv);
634 			spacemap_check_sm_log(spa, &mv);
635 
636 			range_tree_vacate(mv.mv_allocated, NULL, NULL);
637 			range_tree_destroy(mv.mv_allocated);
638 			zfs_btree_clear(&mv.mv_livelist_allocs);
639 			zfs_btree_destroy(&mv.mv_livelist_allocs);
640 		}
641 	}
642 	(void) fprintf(stderr, "\n");
643 
644 	/*
645 	 * If there are any segments in the leftover tree after we walked
646 	 * through all the metaslabs in the concrete vdevs then this means
647 	 * that we have segments in the livelists that belong to indirect
648 	 * vdevs and are marked as allocated.
649 	 */
650 	if (zfs_btree_numnodes(&sv.sv_leftover) == 0) {
651 		zfs_btree_destroy(&sv.sv_leftover);
652 		return;
653 	}
654 	(void) printf("ERROR: Found livelist blocks marked as allocated "
655 	    "for indirect vdevs:\n");
656 
657 	zfs_btree_index_t *where = NULL;
658 	sublivelist_verify_block_t *svb;
659 	while ((svb = zfs_btree_destroy_nodes(&sv.sv_leftover, &where)) !=
660 	    NULL) {
661 		int vdev_id = DVA_GET_VDEV(&svb->svb_dva);
662 		ASSERT3U(vdev_id, <, rvd->vdev_children);
663 		vdev_t *vd = rvd->vdev_child[vdev_id];
664 		ASSERT(!vdev_is_concrete(vd));
665 		(void) printf("<%d:%llx:%llx> TXG %llx\n",
666 		    vdev_id, (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
667 		    (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva),
668 		    (u_longlong_t)svb->svb_allocated_txg);
669 	}
670 	(void) printf("\n");
671 	zfs_btree_destroy(&sv.sv_leftover);
672 }
673 
674 /*
675  * These libumem hooks provide a reasonable set of defaults for the allocator's
676  * debugging facilities.
677  */
678 const char *
_umem_debug_init(void)679 _umem_debug_init(void)
680 {
681 	return ("default,verbose"); /* $UMEM_DEBUG setting */
682 }
683 
684 const char *
_umem_logging_init(void)685 _umem_logging_init(void)
686 {
687 	return ("fail,contents"); /* $UMEM_LOGGING setting */
688 }
689 
690 static void
usage(void)691 usage(void)
692 {
693 	(void) fprintf(stderr,
694 	    "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
695 	    "[-I <inflight I/Os>]\n"
696 	    "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
697 	    "\t\t[-K <key>]\n"
698 	    "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
699 	    "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
700 	    "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
701 	    "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
702 	    "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
703 	    "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
704 	    "\t%s [-v] <bookmark>\n"
705 	    "\t%s -C [-A] [-U <cache>] [<poolname>]\n"
706 	    "\t%s -l [-Aqu] <device>\n"
707 	    "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
708 	    "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
709 	    "\t%s -O [-K <key>] <dataset> <path>\n"
710 	    "\t%s -r [-K <key>] <dataset> <path> <destination>\n"
711 	    "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
712 	    "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
713 	    "\t%s -E [-A] word0:word1:...:word15\n"
714 	    "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
715 	    "<poolname>\n\n",
716 	    cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
717 	    cmdname, cmdname, cmdname, cmdname, cmdname);
718 
719 	(void) fprintf(stderr, "    Dataset name must include at least one "
720 	    "separator character '/' or '@'\n");
721 	(void) fprintf(stderr, "    If dataset name is specified, only that "
722 	    "dataset is dumped\n");
723 	(void) fprintf(stderr,  "    If object numbers or object number "
724 	    "ranges are specified, only those\n"
725 	    "    objects or ranges are dumped.\n\n");
726 	(void) fprintf(stderr,
727 	    "    Object ranges take the form <start>:<end>[:<flags>]\n"
728 	    "        start    Starting object number\n"
729 	    "        end      Ending object number, or -1 for no upper bound\n"
730 	    "        flags    Optional flags to select object types:\n"
731 	    "            A     All objects (this is the default)\n"
732 	    "            d     ZFS directories\n"
733 	    "            f     ZFS files \n"
734 	    "            m     SPA space maps\n"
735 	    "            z     ZAPs\n"
736 	    "            -     Negate effect of next flag\n\n");
737 	(void) fprintf(stderr, "    Options to control amount of output:\n");
738 	(void) fprintf(stderr, "        -b --block-stats             "
739 	    "block statistics\n");
740 	(void) fprintf(stderr, "        -B --backup                  "
741 	    "backup stream\n");
742 	(void) fprintf(stderr, "        -c --checksum                "
743 	    "checksum all metadata (twice for all data) blocks\n");
744 	(void) fprintf(stderr, "        -C --config                  "
745 	    "config (or cachefile if alone)\n");
746 	(void) fprintf(stderr, "        -d --datasets                "
747 	    "dataset(s)\n");
748 	(void) fprintf(stderr, "        -D --dedup-stats             "
749 	    "dedup statistics\n");
750 	(void) fprintf(stderr, "        -E --embedded-block-pointer=INTEGER\n"
751 	    "                                     decode and display block "
752 	    "from an embedded block pointer\n");
753 	(void) fprintf(stderr, "        -h --history                 "
754 	    "pool history\n");
755 	(void) fprintf(stderr, "        -i --intent-logs             "
756 	    "intent logs\n");
757 	(void) fprintf(stderr, "        -l --label                   "
758 	    "read label contents\n");
759 	(void) fprintf(stderr, "        -k --checkpointed-state      "
760 	    "examine the checkpointed state of the pool\n");
761 	(void) fprintf(stderr, "        -L --disable-leak-tracking   "
762 	    "disable leak tracking (do not load spacemaps)\n");
763 	(void) fprintf(stderr, "        -m --metaslabs               "
764 	    "metaslabs\n");
765 	(void) fprintf(stderr, "        -M --metaslab-groups         "
766 	    "metaslab groups\n");
767 	(void) fprintf(stderr, "        -O --object-lookups          "
768 	    "perform object lookups by path\n");
769 	(void) fprintf(stderr, "        -r --copy-object             "
770 	    "copy an object by path to file\n");
771 	(void) fprintf(stderr, "        -R --read-block              "
772 	    "read and display block from a device\n");
773 	(void) fprintf(stderr, "        -s --io-stats                "
774 	    "report stats on zdb's I/O\n");
775 	(void) fprintf(stderr, "        -S --simulate-dedup          "
776 	    "simulate dedup to measure effect\n");
777 	(void) fprintf(stderr, "        -v --verbose                 "
778 	    "verbose (applies to all others)\n");
779 	(void) fprintf(stderr, "        -y --livelist                "
780 	    "perform livelist and metaslab validation on any livelists being "
781 	    "deleted\n\n");
782 	(void) fprintf(stderr, "    Below options are intended for use "
783 	    "with other options:\n");
784 	(void) fprintf(stderr, "        -A --ignore-assertions       "
785 	    "ignore assertions (-A), enable panic recovery (-AA) or both "
786 	    "(-AAA)\n");
787 	(void) fprintf(stderr, "        -e --exported                "
788 	    "pool is exported/destroyed/has altroot/not in a cachefile\n");
789 	(void) fprintf(stderr, "        -F --automatic-rewind        "
790 	    "attempt automatic rewind within safe range of transaction "
791 	    "groups\n");
792 	(void) fprintf(stderr, "        -G --dump-debug-msg          "
793 	    "dump zfs_dbgmsg buffer before exiting\n");
794 	(void) fprintf(stderr, "        -I --inflight=INTEGER        "
795 	    "specify the maximum number of checksumming I/Os "
796 	    "[default is 200]\n");
797 	(void) fprintf(stderr, "        -K --key=KEY                 "
798 	    "decryption key for encrypted dataset\n");
799 	(void) fprintf(stderr, "        -o --option=\"OPTION=INTEGER\" "
800 	    "set global variable to an unsigned 32-bit integer\n");
801 	(void) fprintf(stderr, "        -p --path==PATH              "
802 	    "use one or more with -e to specify path to vdev dir\n");
803 	(void) fprintf(stderr, "        -P --parseable               "
804 	    "print numbers in parseable form\n");
805 	(void) fprintf(stderr, "        -q --skip-label              "
806 	    "don't print label contents\n");
807 	(void) fprintf(stderr, "        -t --txg=INTEGER             "
808 	    "highest txg to use when searching for uberblocks\n");
809 	(void) fprintf(stderr, "        -T --brt-stats               "
810 	    "BRT statistics\n");
811 	(void) fprintf(stderr, "        -u --uberblock               "
812 	    "uberblock\n");
813 	(void) fprintf(stderr, "        -U --cachefile=PATH          "
814 	    "use alternate cachefile\n");
815 	(void) fprintf(stderr, "        -V --verbatim                "
816 	    "do verbatim import\n");
817 	(void) fprintf(stderr, "        -x --dump-blocks=PATH        "
818 	    "dump all read blocks into specified directory\n");
819 	(void) fprintf(stderr, "        -X --extreme-rewind          "
820 	    "attempt extreme rewind (does not work with dataset)\n");
821 	(void) fprintf(stderr, "        -Y --all-reconstruction      "
822 	    "attempt all reconstruction combinations for split blocks\n");
823 	(void) fprintf(stderr, "        -Z --zstd-headers            "
824 	    "show ZSTD headers \n");
825 	(void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
826 	    "to make only that option verbose\n");
827 	(void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
828 	zdb_exit(1);
829 }
830 
831 static void
dump_debug_buffer(void)832 dump_debug_buffer(void)
833 {
834 	ssize_t ret __attribute__((unused));
835 
836 	if (!dump_opt['G'])
837 		return;
838 	/*
839 	 * We use write() instead of printf() so that this function
840 	 * is safe to call from a signal handler.
841 	 */
842 	ret = write(STDERR_FILENO, "\n", 1);
843 	zfs_dbgmsg_print(STDERR_FILENO, "zdb");
844 }
845 
sig_handler(int signo)846 static void sig_handler(int signo)
847 {
848 	struct sigaction action;
849 
850 	libspl_backtrace(STDERR_FILENO);
851 	dump_debug_buffer();
852 
853 	/*
854 	 * Restore default action and re-raise signal so SIGSEGV and
855 	 * SIGABRT can trigger a core dump.
856 	 */
857 	action.sa_handler = SIG_DFL;
858 	sigemptyset(&action.sa_mask);
859 	action.sa_flags = 0;
860 	(void) sigaction(signo, &action, NULL);
861 	raise(signo);
862 }
863 
864 /*
865  * Called for usage errors that are discovered after a call to spa_open(),
866  * dmu_bonus_hold(), or pool_match().  abort() is called for other errors.
867  */
868 
869 static void
fatal(const char * fmt,...)870 fatal(const char *fmt, ...)
871 {
872 	va_list ap;
873 
874 	va_start(ap, fmt);
875 	(void) fprintf(stderr, "%s: ", cmdname);
876 	(void) vfprintf(stderr, fmt, ap);
877 	va_end(ap);
878 	(void) fprintf(stderr, "\n");
879 
880 	dump_debug_buffer();
881 
882 	zdb_exit(1);
883 }
884 
885 static void
dump_packed_nvlist(objset_t * os,uint64_t object,void * data,size_t size)886 dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size)
887 {
888 	(void) size;
889 	nvlist_t *nv;
890 	size_t nvsize = *(uint64_t *)data;
891 	char *packed = umem_alloc(nvsize, UMEM_NOFAIL);
892 
893 	VERIFY(0 == dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH));
894 
895 	VERIFY(nvlist_unpack(packed, nvsize, &nv, 0) == 0);
896 
897 	umem_free(packed, nvsize);
898 
899 	dump_nvlist(nv, 8);
900 
901 	nvlist_free(nv);
902 }
903 
904 static void
dump_history_offsets(objset_t * os,uint64_t object,void * data,size_t size)905 dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size)
906 {
907 	(void) os, (void) object, (void) size;
908 	spa_history_phys_t *shp = data;
909 
910 	if (shp == NULL)
911 		return;
912 
913 	(void) printf("\t\tpool_create_len = %llu\n",
914 	    (u_longlong_t)shp->sh_pool_create_len);
915 	(void) printf("\t\tphys_max_off = %llu\n",
916 	    (u_longlong_t)shp->sh_phys_max_off);
917 	(void) printf("\t\tbof = %llu\n",
918 	    (u_longlong_t)shp->sh_bof);
919 	(void) printf("\t\teof = %llu\n",
920 	    (u_longlong_t)shp->sh_eof);
921 	(void) printf("\t\trecords_lost = %llu\n",
922 	    (u_longlong_t)shp->sh_records_lost);
923 }
924 
925 static void
zdb_nicenum(uint64_t num,char * buf,size_t buflen)926 zdb_nicenum(uint64_t num, char *buf, size_t buflen)
927 {
928 	if (dump_opt['P'])
929 		(void) snprintf(buf, buflen, "%llu", (longlong_t)num);
930 	else
931 		nicenum(num, buf, buflen);
932 }
933 
934 static void
zdb_nicebytes(uint64_t bytes,char * buf,size_t buflen)935 zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen)
936 {
937 	if (dump_opt['P'])
938 		(void) snprintf(buf, buflen, "%llu", (longlong_t)bytes);
939 	else
940 		zfs_nicebytes(bytes, buf, buflen);
941 }
942 
943 static const char histo_stars[] = "****************************************";
944 static const uint64_t histo_width = sizeof (histo_stars) - 1;
945 
946 static void
dump_histogram(const uint64_t * histo,int size,int offset)947 dump_histogram(const uint64_t *histo, int size, int offset)
948 {
949 	int i;
950 	int minidx = size - 1;
951 	int maxidx = 0;
952 	uint64_t max = 0;
953 
954 	for (i = 0; i < size; i++) {
955 		if (histo[i] == 0)
956 			continue;
957 		if (histo[i] > max)
958 			max = histo[i];
959 		if (i > maxidx)
960 			maxidx = i;
961 		if (i < minidx)
962 			minidx = i;
963 	}
964 
965 	if (max < histo_width)
966 		max = histo_width;
967 
968 	for (i = minidx; i <= maxidx; i++) {
969 		(void) printf("\t\t\t%3u: %6llu %s\n",
970 		    i + offset, (u_longlong_t)histo[i],
971 		    &histo_stars[(max - histo[i]) * histo_width / max]);
972 	}
973 }
974 
975 static void
dump_zap_stats(objset_t * os,uint64_t object)976 dump_zap_stats(objset_t *os, uint64_t object)
977 {
978 	int error;
979 	zap_stats_t zs;
980 
981 	error = zap_get_stats(os, object, &zs);
982 	if (error)
983 		return;
984 
985 	if (zs.zs_ptrtbl_len == 0) {
986 		ASSERT(zs.zs_num_blocks == 1);
987 		(void) printf("\tmicrozap: %llu bytes, %llu entries\n",
988 		    (u_longlong_t)zs.zs_blocksize,
989 		    (u_longlong_t)zs.zs_num_entries);
990 		return;
991 	}
992 
993 	(void) printf("\tFat ZAP stats:\n");
994 
995 	(void) printf("\t\tPointer table:\n");
996 	(void) printf("\t\t\t%llu elements\n",
997 	    (u_longlong_t)zs.zs_ptrtbl_len);
998 	(void) printf("\t\t\tzt_blk: %llu\n",
999 	    (u_longlong_t)zs.zs_ptrtbl_zt_blk);
1000 	(void) printf("\t\t\tzt_numblks: %llu\n",
1001 	    (u_longlong_t)zs.zs_ptrtbl_zt_numblks);
1002 	(void) printf("\t\t\tzt_shift: %llu\n",
1003 	    (u_longlong_t)zs.zs_ptrtbl_zt_shift);
1004 	(void) printf("\t\t\tzt_blks_copied: %llu\n",
1005 	    (u_longlong_t)zs.zs_ptrtbl_blks_copied);
1006 	(void) printf("\t\t\tzt_nextblk: %llu\n",
1007 	    (u_longlong_t)zs.zs_ptrtbl_nextblk);
1008 
1009 	(void) printf("\t\tZAP entries: %llu\n",
1010 	    (u_longlong_t)zs.zs_num_entries);
1011 	(void) printf("\t\tLeaf blocks: %llu\n",
1012 	    (u_longlong_t)zs.zs_num_leafs);
1013 	(void) printf("\t\tTotal blocks: %llu\n",
1014 	    (u_longlong_t)zs.zs_num_blocks);
1015 	(void) printf("\t\tzap_block_type: 0x%llx\n",
1016 	    (u_longlong_t)zs.zs_block_type);
1017 	(void) printf("\t\tzap_magic: 0x%llx\n",
1018 	    (u_longlong_t)zs.zs_magic);
1019 	(void) printf("\t\tzap_salt: 0x%llx\n",
1020 	    (u_longlong_t)zs.zs_salt);
1021 
1022 	(void) printf("\t\tLeafs with 2^n pointers:\n");
1023 	dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);
1024 
1025 	(void) printf("\t\tBlocks with n*5 entries:\n");
1026 	dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);
1027 
1028 	(void) printf("\t\tBlocks n/10 full:\n");
1029 	dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);
1030 
1031 	(void) printf("\t\tEntries with n chunks:\n");
1032 	dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);
1033 
1034 	(void) printf("\t\tBuckets with n entries:\n");
1035 	dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
1036 }
1037 
1038 static void
dump_none(objset_t * os,uint64_t object,void * data,size_t size)1039 dump_none(objset_t *os, uint64_t object, void *data, size_t size)
1040 {
1041 	(void) os, (void) object, (void) data, (void) size;
1042 }
1043 
1044 static void
dump_unknown(objset_t * os,uint64_t object,void * data,size_t size)1045 dump_unknown(objset_t *os, uint64_t object, void *data, size_t size)
1046 {
1047 	(void) os, (void) object, (void) data, (void) size;
1048 	(void) printf("\tUNKNOWN OBJECT TYPE\n");
1049 }
1050 
1051 static void
dump_uint8(objset_t * os,uint64_t object,void * data,size_t size)1052 dump_uint8(objset_t *os, uint64_t object, void *data, size_t size)
1053 {
1054 	(void) os, (void) object, (void) data, (void) size;
1055 }
1056 
1057 static void
dump_uint64(objset_t * os,uint64_t object,void * data,size_t size)1058 dump_uint64(objset_t *os, uint64_t object, void *data, size_t size)
1059 {
1060 	uint64_t *arr;
1061 	uint64_t oursize;
1062 	if (dump_opt['d'] < 6)
1063 		return;
1064 
1065 	if (data == NULL) {
1066 		dmu_object_info_t doi;
1067 
1068 		VERIFY0(dmu_object_info(os, object, &doi));
1069 		size = doi.doi_max_offset;
1070 		/*
1071 		 * We cap the size at 1 mebibyte here to prevent
1072 		 * allocation failures and nigh-infinite printing if the
1073 		 * object is extremely large.
1074 		 */
1075 		oursize = MIN(size, 1 << 20);
1076 		arr = kmem_alloc(oursize, KM_SLEEP);
1077 
1078 		int err = dmu_read(os, object, 0, oursize, arr, 0);
1079 		if (err != 0) {
1080 			(void) printf("got error %u from dmu_read\n", err);
1081 			kmem_free(arr, oursize);
1082 			return;
1083 		}
1084 	} else {
1085 		/*
1086 		 * Even though the allocation is already done in this code path,
1087 		 * we still cap the size to prevent excessive printing.
1088 		 */
1089 		oursize = MIN(size, 1 << 20);
1090 		arr = data;
1091 	}
1092 
1093 	if (size == 0) {
1094 		if (data == NULL)
1095 			kmem_free(arr, oursize);
1096 		(void) printf("\t\t[]\n");
1097 		return;
1098 	}
1099 
1100 	(void) printf("\t\t[%0llx", (u_longlong_t)arr[0]);
1101 	for (size_t i = 1; i * sizeof (uint64_t) < oursize; i++) {
1102 		if (i % 4 != 0)
1103 			(void) printf(", %0llx", (u_longlong_t)arr[i]);
1104 		else
1105 			(void) printf(",\n\t\t%0llx", (u_longlong_t)arr[i]);
1106 	}
1107 	if (oursize != size)
1108 		(void) printf(", ... ");
1109 	(void) printf("]\n");
1110 
1111 	if (data == NULL)
1112 		kmem_free(arr, oursize);
1113 }
1114 
1115 static void
dump_zap(objset_t * os,uint64_t object,void * data,size_t size)1116 dump_zap(objset_t *os, uint64_t object, void *data, size_t size)
1117 {
1118 	(void) data, (void) size;
1119 	zap_cursor_t zc;
1120 	zap_attribute_t *attrp = zap_attribute_long_alloc();
1121 	void *prop;
1122 	unsigned i;
1123 
1124 	dump_zap_stats(os, object);
1125 	(void) printf("\n");
1126 
1127 	for (zap_cursor_init(&zc, os, object);
1128 	    zap_cursor_retrieve(&zc, attrp) == 0;
1129 	    zap_cursor_advance(&zc)) {
1130 		boolean_t key64 =
1131 		    !!(zap_getflags(zc.zc_zap) & ZAP_FLAG_UINT64_KEY);
1132 
1133 		if (key64)
1134 			(void) printf("\t\t0x%010" PRIu64 "x = ",
1135 			    *(uint64_t *)attrp->za_name);
1136 		else
1137 			(void) printf("\t\t%s = ", attrp->za_name);
1138 
1139 		if (attrp->za_num_integers == 0) {
1140 			(void) printf("\n");
1141 			continue;
1142 		}
1143 		prop = umem_zalloc(attrp->za_num_integers *
1144 		    attrp->za_integer_length, UMEM_NOFAIL);
1145 
1146 		if (key64)
1147 			(void) zap_lookup_uint64(os, object,
1148 			    (const uint64_t *)attrp->za_name, 1,
1149 			    attrp->za_integer_length, attrp->za_num_integers,
1150 			    prop);
1151 		else
1152 			(void) zap_lookup(os, object, attrp->za_name,
1153 			    attrp->za_integer_length, attrp->za_num_integers,
1154 			    prop);
1155 
1156 		if (attrp->za_integer_length == 1 && !key64) {
1157 			if (strcmp(attrp->za_name,
1158 			    DSL_CRYPTO_KEY_MASTER_KEY) == 0 ||
1159 			    strcmp(attrp->za_name,
1160 			    DSL_CRYPTO_KEY_HMAC_KEY) == 0 ||
1161 			    strcmp(attrp->za_name, DSL_CRYPTO_KEY_IV) == 0 ||
1162 			    strcmp(attrp->za_name, DSL_CRYPTO_KEY_MAC) == 0 ||
1163 			    strcmp(attrp->za_name,
1164 			    DMU_POOL_CHECKSUM_SALT) == 0) {
1165 				uint8_t *u8 = prop;
1166 
1167 				for (i = 0; i < attrp->za_num_integers; i++) {
1168 					(void) printf("%02x", u8[i]);
1169 				}
1170 			} else {
1171 				(void) printf("%s", (char *)prop);
1172 			}
1173 		} else {
1174 			for (i = 0; i < attrp->za_num_integers; i++) {
1175 				switch (attrp->za_integer_length) {
1176 				case 1:
1177 					(void) printf("%u ",
1178 					    ((uint8_t *)prop)[i]);
1179 					break;
1180 				case 2:
1181 					(void) printf("%u ",
1182 					    ((uint16_t *)prop)[i]);
1183 					break;
1184 				case 4:
1185 					(void) printf("%u ",
1186 					    ((uint32_t *)prop)[i]);
1187 					break;
1188 				case 8:
1189 					(void) printf("%lld ",
1190 					    (u_longlong_t)((int64_t *)prop)[i]);
1191 					break;
1192 				}
1193 			}
1194 		}
1195 		(void) printf("\n");
1196 		umem_free(prop,
1197 		    attrp->za_num_integers * attrp->za_integer_length);
1198 	}
1199 	zap_cursor_fini(&zc);
1200 	zap_attribute_free(attrp);
1201 }
1202 
1203 static void
dump_bpobj(objset_t * os,uint64_t object,void * data,size_t size)1204 dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size)
1205 {
1206 	bpobj_phys_t *bpop = data;
1207 	uint64_t i;
1208 	char bytes[32], comp[32], uncomp[32];
1209 
1210 	/* make sure the output won't get truncated */
1211 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
1212 	_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
1213 	_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
1214 
1215 	if (bpop == NULL)
1216 		return;
1217 
1218 	zdb_nicenum(bpop->bpo_bytes, bytes, sizeof (bytes));
1219 	zdb_nicenum(bpop->bpo_comp, comp, sizeof (comp));
1220 	zdb_nicenum(bpop->bpo_uncomp, uncomp, sizeof (uncomp));
1221 
1222 	(void) printf("\t\tnum_blkptrs = %llu\n",
1223 	    (u_longlong_t)bpop->bpo_num_blkptrs);
1224 	(void) printf("\t\tbytes = %s\n", bytes);
1225 	if (size >= BPOBJ_SIZE_V1) {
1226 		(void) printf("\t\tcomp = %s\n", comp);
1227 		(void) printf("\t\tuncomp = %s\n", uncomp);
1228 	}
1229 	if (size >= BPOBJ_SIZE_V2) {
1230 		(void) printf("\t\tsubobjs = %llu\n",
1231 		    (u_longlong_t)bpop->bpo_subobjs);
1232 		(void) printf("\t\tnum_subobjs = %llu\n",
1233 		    (u_longlong_t)bpop->bpo_num_subobjs);
1234 	}
1235 	if (size >= sizeof (*bpop)) {
1236 		(void) printf("\t\tnum_freed = %llu\n",
1237 		    (u_longlong_t)bpop->bpo_num_freed);
1238 	}
1239 
1240 	if (dump_opt['d'] < 5)
1241 		return;
1242 
1243 	for (i = 0; i < bpop->bpo_num_blkptrs; i++) {
1244 		char blkbuf[BP_SPRINTF_LEN];
1245 		blkptr_t bp;
1246 
1247 		int err = dmu_read(os, object,
1248 		    i * sizeof (bp), sizeof (bp), &bp, 0);
1249 		if (err != 0) {
1250 			(void) printf("got error %u from dmu_read\n", err);
1251 			break;
1252 		}
1253 		snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp,
1254 		    BP_GET_FREE(&bp));
1255 		(void) printf("\t%s\n", blkbuf);
1256 	}
1257 }
1258 
1259 static void
dump_bpobj_subobjs(objset_t * os,uint64_t object,void * data,size_t size)1260 dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size)
1261 {
1262 	(void) data, (void) size;
1263 	dmu_object_info_t doi;
1264 	int64_t i;
1265 
1266 	VERIFY0(dmu_object_info(os, object, &doi));
1267 	uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP);
1268 
1269 	int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0);
1270 	if (err != 0) {
1271 		(void) printf("got error %u from dmu_read\n", err);
1272 		kmem_free(subobjs, doi.doi_max_offset);
1273 		return;
1274 	}
1275 
1276 	int64_t last_nonzero = -1;
1277 	for (i = 0; i < doi.doi_max_offset / 8; i++) {
1278 		if (subobjs[i] != 0)
1279 			last_nonzero = i;
1280 	}
1281 
1282 	for (i = 0; i <= last_nonzero; i++) {
1283 		(void) printf("\t%llu\n", (u_longlong_t)subobjs[i]);
1284 	}
1285 	kmem_free(subobjs, doi.doi_max_offset);
1286 }
1287 
1288 static void
dump_ddt_zap(objset_t * os,uint64_t object,void * data,size_t size)1289 dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size)
1290 {
1291 	(void) data, (void) size;
1292 	dump_zap_stats(os, object);
1293 	/* contents are printed elsewhere, properly decoded */
1294 }
1295 
1296 static void
dump_sa_attrs(objset_t * os,uint64_t object,void * data,size_t size)1297 dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size)
1298 {
1299 	(void) data, (void) size;
1300 	zap_cursor_t zc;
1301 	zap_attribute_t *attrp = zap_attribute_alloc();
1302 
1303 	dump_zap_stats(os, object);
1304 	(void) printf("\n");
1305 
1306 	for (zap_cursor_init(&zc, os, object);
1307 	    zap_cursor_retrieve(&zc, attrp) == 0;
1308 	    zap_cursor_advance(&zc)) {
1309 		(void) printf("\t\t%s = ", attrp->za_name);
1310 		if (attrp->za_num_integers == 0) {
1311 			(void) printf("\n");
1312 			continue;
1313 		}
1314 		(void) printf(" %llx : [%d:%d:%d]\n",
1315 		    (u_longlong_t)attrp->za_first_integer,
1316 		    (int)ATTR_LENGTH(attrp->za_first_integer),
1317 		    (int)ATTR_BSWAP(attrp->za_first_integer),
1318 		    (int)ATTR_NUM(attrp->za_first_integer));
1319 	}
1320 	zap_cursor_fini(&zc);
1321 	zap_attribute_free(attrp);
1322 }
1323 
1324 static void
dump_sa_layouts(objset_t * os,uint64_t object,void * data,size_t size)1325 dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size)
1326 {
1327 	(void) data, (void) size;
1328 	zap_cursor_t zc;
1329 	zap_attribute_t *attrp = zap_attribute_alloc();
1330 	uint16_t *layout_attrs;
1331 	unsigned i;
1332 
1333 	dump_zap_stats(os, object);
1334 	(void) printf("\n");
1335 
1336 	for (zap_cursor_init(&zc, os, object);
1337 	    zap_cursor_retrieve(&zc, attrp) == 0;
1338 	    zap_cursor_advance(&zc)) {
1339 		(void) printf("\t\t%s = [", attrp->za_name);
1340 		if (attrp->za_num_integers == 0) {
1341 			(void) printf("\n");
1342 			continue;
1343 		}
1344 
1345 		VERIFY(attrp->za_integer_length == 2);
1346 		layout_attrs = umem_zalloc(attrp->za_num_integers *
1347 		    attrp->za_integer_length, UMEM_NOFAIL);
1348 
1349 		VERIFY(zap_lookup(os, object, attrp->za_name,
1350 		    attrp->za_integer_length,
1351 		    attrp->za_num_integers, layout_attrs) == 0);
1352 
1353 		for (i = 0; i != attrp->za_num_integers; i++)
1354 			(void) printf(" %d ", (int)layout_attrs[i]);
1355 		(void) printf("]\n");
1356 		umem_free(layout_attrs,
1357 		    attrp->za_num_integers * attrp->za_integer_length);
1358 	}
1359 	zap_cursor_fini(&zc);
1360 	zap_attribute_free(attrp);
1361 }
1362 
1363 static void
dump_zpldir(objset_t * os,uint64_t object,void * data,size_t size)1364 dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
1365 {
1366 	(void) data, (void) size;
1367 	zap_cursor_t zc;
1368 	zap_attribute_t *attrp = zap_attribute_long_alloc();
1369 	const char *typenames[] = {
1370 		/* 0 */ "not specified",
1371 		/* 1 */ "FIFO",
1372 		/* 2 */ "Character Device",
1373 		/* 3 */ "3 (invalid)",
1374 		/* 4 */ "Directory",
1375 		/* 5 */ "5 (invalid)",
1376 		/* 6 */ "Block Device",
1377 		/* 7 */ "7 (invalid)",
1378 		/* 8 */ "Regular File",
1379 		/* 9 */ "9 (invalid)",
1380 		/* 10 */ "Symbolic Link",
1381 		/* 11 */ "11 (invalid)",
1382 		/* 12 */ "Socket",
1383 		/* 13 */ "Door",
1384 		/* 14 */ "Event Port",
1385 		/* 15 */ "15 (invalid)",
1386 	};
1387 
1388 	dump_zap_stats(os, object);
1389 	(void) printf("\n");
1390 
1391 	for (zap_cursor_init(&zc, os, object);
1392 	    zap_cursor_retrieve(&zc, attrp) == 0;
1393 	    zap_cursor_advance(&zc)) {
1394 		(void) printf("\t\t%s = %lld (type: %s)\n",
1395 		    attrp->za_name, ZFS_DIRENT_OBJ(attrp->za_first_integer),
1396 		    typenames[ZFS_DIRENT_TYPE(attrp->za_first_integer)]);
1397 	}
1398 	zap_cursor_fini(&zc);
1399 	zap_attribute_free(attrp);
1400 }
1401 
1402 static int
get_dtl_refcount(vdev_t * vd)1403 get_dtl_refcount(vdev_t *vd)
1404 {
1405 	int refcount = 0;
1406 
1407 	if (vd->vdev_ops->vdev_op_leaf) {
1408 		space_map_t *sm = vd->vdev_dtl_sm;
1409 
1410 		if (sm != NULL &&
1411 		    sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1412 			return (1);
1413 		return (0);
1414 	}
1415 
1416 	for (unsigned c = 0; c < vd->vdev_children; c++)
1417 		refcount += get_dtl_refcount(vd->vdev_child[c]);
1418 	return (refcount);
1419 }
1420 
1421 static int
get_metaslab_refcount(vdev_t * vd)1422 get_metaslab_refcount(vdev_t *vd)
1423 {
1424 	int refcount = 0;
1425 
1426 	if (vd->vdev_top == vd) {
1427 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
1428 			space_map_t *sm = vd->vdev_ms[m]->ms_sm;
1429 
1430 			if (sm != NULL &&
1431 			    sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1432 				refcount++;
1433 		}
1434 	}
1435 	for (unsigned c = 0; c < vd->vdev_children; c++)
1436 		refcount += get_metaslab_refcount(vd->vdev_child[c]);
1437 
1438 	return (refcount);
1439 }
1440 
1441 static int
get_obsolete_refcount(vdev_t * vd)1442 get_obsolete_refcount(vdev_t *vd)
1443 {
1444 	uint64_t obsolete_sm_object;
1445 	int refcount = 0;
1446 
1447 	VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1448 	if (vd->vdev_top == vd && obsolete_sm_object != 0) {
1449 		dmu_object_info_t doi;
1450 		VERIFY0(dmu_object_info(vd->vdev_spa->spa_meta_objset,
1451 		    obsolete_sm_object, &doi));
1452 		if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1453 			refcount++;
1454 		}
1455 	} else {
1456 		ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
1457 		ASSERT3U(obsolete_sm_object, ==, 0);
1458 	}
1459 	for (unsigned c = 0; c < vd->vdev_children; c++) {
1460 		refcount += get_obsolete_refcount(vd->vdev_child[c]);
1461 	}
1462 
1463 	return (refcount);
1464 }
1465 
1466 static int
get_prev_obsolete_spacemap_refcount(spa_t * spa)1467 get_prev_obsolete_spacemap_refcount(spa_t *spa)
1468 {
1469 	uint64_t prev_obj =
1470 	    spa->spa_condensing_indirect_phys.scip_prev_obsolete_sm_object;
1471 	if (prev_obj != 0) {
1472 		dmu_object_info_t doi;
1473 		VERIFY0(dmu_object_info(spa->spa_meta_objset, prev_obj, &doi));
1474 		if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1475 			return (1);
1476 		}
1477 	}
1478 	return (0);
1479 }
1480 
1481 static int
get_checkpoint_refcount(vdev_t * vd)1482 get_checkpoint_refcount(vdev_t *vd)
1483 {
1484 	int refcount = 0;
1485 
1486 	if (vd->vdev_top == vd && vd->vdev_top_zap != 0 &&
1487 	    zap_contains(spa_meta_objset(vd->vdev_spa),
1488 	    vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) == 0)
1489 		refcount++;
1490 
1491 	for (uint64_t c = 0; c < vd->vdev_children; c++)
1492 		refcount += get_checkpoint_refcount(vd->vdev_child[c]);
1493 
1494 	return (refcount);
1495 }
1496 
1497 static int
get_log_spacemap_refcount(spa_t * spa)1498 get_log_spacemap_refcount(spa_t *spa)
1499 {
1500 	return (avl_numnodes(&spa->spa_sm_logs_by_txg));
1501 }
1502 
1503 static int
verify_spacemap_refcounts(spa_t * spa)1504 verify_spacemap_refcounts(spa_t *spa)
1505 {
1506 	uint64_t expected_refcount = 0;
1507 	uint64_t actual_refcount;
1508 
1509 	(void) feature_get_refcount(spa,
1510 	    &spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM],
1511 	    &expected_refcount);
1512 	actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
1513 	actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);
1514 	actual_refcount += get_obsolete_refcount(spa->spa_root_vdev);
1515 	actual_refcount += get_prev_obsolete_spacemap_refcount(spa);
1516 	actual_refcount += get_checkpoint_refcount(spa->spa_root_vdev);
1517 	actual_refcount += get_log_spacemap_refcount(spa);
1518 
1519 	if (expected_refcount != actual_refcount) {
1520 		(void) printf("space map refcount mismatch: expected %lld != "
1521 		    "actual %lld\n",
1522 		    (longlong_t)expected_refcount,
1523 		    (longlong_t)actual_refcount);
1524 		return (2);
1525 	}
1526 	return (0);
1527 }
1528 
1529 static void
dump_spacemap(objset_t * os,space_map_t * sm)1530 dump_spacemap(objset_t *os, space_map_t *sm)
1531 {
1532 	const char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1533 	    "INVALID", "INVALID", "INVALID", "INVALID" };
1534 
1535 	if (sm == NULL)
1536 		return;
1537 
1538 	(void) printf("space map object %llu:\n",
1539 	    (longlong_t)sm->sm_object);
1540 	(void) printf("  smp_length = 0x%llx\n",
1541 	    (longlong_t)sm->sm_phys->smp_length);
1542 	(void) printf("  smp_alloc = 0x%llx\n",
1543 	    (longlong_t)sm->sm_phys->smp_alloc);
1544 
1545 	if (dump_opt['d'] < 6 && dump_opt['m'] < 4)
1546 		return;
1547 
1548 	/*
1549 	 * Print out the freelist entries in both encoded and decoded form.
1550 	 */
1551 	uint8_t mapshift = sm->sm_shift;
1552 	int64_t alloc = 0;
1553 	uint64_t word, entry_id = 0;
1554 	for (uint64_t offset = 0; offset < space_map_length(sm);
1555 	    offset += sizeof (word)) {
1556 
1557 		VERIFY0(dmu_read(os, space_map_object(sm), offset,
1558 		    sizeof (word), &word, DMU_READ_PREFETCH));
1559 
1560 		if (sm_entry_is_debug(word)) {
1561 			uint64_t de_txg = SM_DEBUG_TXG_DECODE(word);
1562 			uint64_t de_sync_pass = SM_DEBUG_SYNCPASS_DECODE(word);
1563 			if (de_txg == 0) {
1564 				(void) printf(
1565 				    "\t    [%6llu] PADDING\n",
1566 				    (u_longlong_t)entry_id);
1567 			} else {
1568 				(void) printf(
1569 				    "\t    [%6llu] %s: txg %llu pass %llu\n",
1570 				    (u_longlong_t)entry_id,
1571 				    ddata[SM_DEBUG_ACTION_DECODE(word)],
1572 				    (u_longlong_t)de_txg,
1573 				    (u_longlong_t)de_sync_pass);
1574 			}
1575 			entry_id++;
1576 			continue;
1577 		}
1578 
1579 		uint8_t words;
1580 		char entry_type;
1581 		uint64_t entry_off, entry_run, entry_vdev = SM_NO_VDEVID;
1582 
1583 		if (sm_entry_is_single_word(word)) {
1584 			entry_type = (SM_TYPE_DECODE(word) == SM_ALLOC) ?
1585 			    'A' : 'F';
1586 			entry_off = (SM_OFFSET_DECODE(word) << mapshift) +
1587 			    sm->sm_start;
1588 			entry_run = SM_RUN_DECODE(word) << mapshift;
1589 			words = 1;
1590 		} else {
1591 			/* it is a two-word entry so we read another word */
1592 			ASSERT(sm_entry_is_double_word(word));
1593 
1594 			uint64_t extra_word;
1595 			offset += sizeof (extra_word);
1596 			VERIFY0(dmu_read(os, space_map_object(sm), offset,
1597 			    sizeof (extra_word), &extra_word,
1598 			    DMU_READ_PREFETCH));
1599 
1600 			ASSERT3U(offset, <=, space_map_length(sm));
1601 
1602 			entry_run = SM2_RUN_DECODE(word) << mapshift;
1603 			entry_vdev = SM2_VDEV_DECODE(word);
1604 			entry_type = (SM2_TYPE_DECODE(extra_word) == SM_ALLOC) ?
1605 			    'A' : 'F';
1606 			entry_off = (SM2_OFFSET_DECODE(extra_word) <<
1607 			    mapshift) + sm->sm_start;
1608 			words = 2;
1609 		}
1610 
1611 		(void) printf("\t    [%6llu]    %c  range:"
1612 		    " %010llx-%010llx  size: %06llx vdev: %06llu words: %u\n",
1613 		    (u_longlong_t)entry_id,
1614 		    entry_type, (u_longlong_t)entry_off,
1615 		    (u_longlong_t)(entry_off + entry_run),
1616 		    (u_longlong_t)entry_run,
1617 		    (u_longlong_t)entry_vdev, words);
1618 
1619 		if (entry_type == 'A')
1620 			alloc += entry_run;
1621 		else
1622 			alloc -= entry_run;
1623 		entry_id++;
1624 	}
1625 	if (alloc != space_map_allocated(sm)) {
1626 		(void) printf("space_map_object alloc (%lld) INCONSISTENT "
1627 		    "with space map summary (%lld)\n",
1628 		    (longlong_t)space_map_allocated(sm), (longlong_t)alloc);
1629 	}
1630 }
1631 
1632 static void
dump_metaslab_stats(metaslab_t * msp)1633 dump_metaslab_stats(metaslab_t *msp)
1634 {
1635 	char maxbuf[32];
1636 	range_tree_t *rt = msp->ms_allocatable;
1637 	zfs_btree_t *t = &msp->ms_allocatable_by_size;
1638 	int free_pct = range_tree_space(rt) * 100 / msp->ms_size;
1639 
1640 	/* max sure nicenum has enough space */
1641 	_Static_assert(sizeof (maxbuf) >= NN_NUMBUF_SZ, "maxbuf truncated");
1642 
1643 	zdb_nicenum(metaslab_largest_allocatable(msp), maxbuf, sizeof (maxbuf));
1644 
1645 	(void) printf("\t %25s %10lu   %7s  %6s   %4s %4d%%\n",
1646 	    "segments", zfs_btree_numnodes(t), "maxsize", maxbuf,
1647 	    "freepct", free_pct);
1648 	(void) printf("\tIn-memory histogram:\n");
1649 	dump_histogram(rt->rt_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1650 }
1651 
1652 static void
dump_metaslab(metaslab_t * msp)1653 dump_metaslab(metaslab_t *msp)
1654 {
1655 	vdev_t *vd = msp->ms_group->mg_vd;
1656 	spa_t *spa = vd->vdev_spa;
1657 	space_map_t *sm = msp->ms_sm;
1658 	char freebuf[32];
1659 
1660 	zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf,
1661 	    sizeof (freebuf));
1662 
1663 	(void) printf(
1664 	    "\tmetaslab %6llu   offset %12llx   spacemap %6llu   free    %5s\n",
1665 	    (u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
1666 	    (u_longlong_t)space_map_object(sm), freebuf);
1667 
1668 	if (dump_opt['m'] > 2 && !dump_opt['L']) {
1669 		mutex_enter(&msp->ms_lock);
1670 		VERIFY0(metaslab_load(msp));
1671 		range_tree_stat_verify(msp->ms_allocatable);
1672 		dump_metaslab_stats(msp);
1673 		metaslab_unload(msp);
1674 		mutex_exit(&msp->ms_lock);
1675 	}
1676 
1677 	if (dump_opt['m'] > 1 && sm != NULL &&
1678 	    spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) {
1679 		/*
1680 		 * The space map histogram represents free space in chunks
1681 		 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1682 		 */
1683 		(void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1684 		    (u_longlong_t)msp->ms_fragmentation);
1685 		dump_histogram(sm->sm_phys->smp_histogram,
1686 		    SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift);
1687 	}
1688 
1689 	if (vd->vdev_ops == &vdev_draid_ops)
1690 		ASSERT3U(msp->ms_size, <=, 1ULL << vd->vdev_ms_shift);
1691 	else
1692 		ASSERT3U(msp->ms_size, ==, 1ULL << vd->vdev_ms_shift);
1693 
1694 	dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
1695 
1696 	if (spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
1697 		(void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1698 		    (u_longlong_t)metaslab_unflushed_txg(msp));
1699 	}
1700 }
1701 
1702 static void
print_vdev_metaslab_header(vdev_t * vd)1703 print_vdev_metaslab_header(vdev_t *vd)
1704 {
1705 	vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
1706 	const char *bias_str = "";
1707 	if (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) {
1708 		bias_str = VDEV_ALLOC_BIAS_LOG;
1709 	} else if (alloc_bias == VDEV_BIAS_SPECIAL) {
1710 		bias_str = VDEV_ALLOC_BIAS_SPECIAL;
1711 	} else if (alloc_bias == VDEV_BIAS_DEDUP) {
1712 		bias_str = VDEV_ALLOC_BIAS_DEDUP;
1713 	}
1714 
1715 	uint64_t ms_flush_data_obj = 0;
1716 	if (vd->vdev_top_zap != 0) {
1717 		int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
1718 		    vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
1719 		    sizeof (uint64_t), 1, &ms_flush_data_obj);
1720 		if (error != ENOENT) {
1721 			ASSERT0(error);
1722 		}
1723 	}
1724 
1725 	(void) printf("\tvdev %10llu   %s",
1726 	    (u_longlong_t)vd->vdev_id, bias_str);
1727 
1728 	if (ms_flush_data_obj != 0) {
1729 		(void) printf("   ms_unflushed_phys object %llu",
1730 		    (u_longlong_t)ms_flush_data_obj);
1731 	}
1732 
1733 	(void) printf("\n\t%-10s%5llu   %-19s   %-15s   %-12s\n",
1734 	    "metaslabs", (u_longlong_t)vd->vdev_ms_count,
1735 	    "offset", "spacemap", "free");
1736 	(void) printf("\t%15s   %19s   %15s   %12s\n",
1737 	    "---------------", "-------------------",
1738 	    "---------------", "------------");
1739 }
1740 
1741 static void
dump_metaslab_groups(spa_t * spa,boolean_t show_special)1742 dump_metaslab_groups(spa_t *spa, boolean_t show_special)
1743 {
1744 	vdev_t *rvd = spa->spa_root_vdev;
1745 	metaslab_class_t *mc = spa_normal_class(spa);
1746 	metaslab_class_t *smc = spa_special_class(spa);
1747 	uint64_t fragmentation;
1748 
1749 	metaslab_class_histogram_verify(mc);
1750 
1751 	for (unsigned c = 0; c < rvd->vdev_children; c++) {
1752 		vdev_t *tvd = rvd->vdev_child[c];
1753 		metaslab_group_t *mg = tvd->vdev_mg;
1754 
1755 		if (mg == NULL || (mg->mg_class != mc &&
1756 		    (!show_special || mg->mg_class != smc)))
1757 			continue;
1758 
1759 		metaslab_group_histogram_verify(mg);
1760 		mg->mg_fragmentation = metaslab_group_fragmentation(mg);
1761 
1762 		(void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1763 		    "fragmentation",
1764 		    (u_longlong_t)tvd->vdev_id,
1765 		    (u_longlong_t)tvd->vdev_ms_count);
1766 		if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
1767 			(void) printf("%3s\n", "-");
1768 		} else {
1769 			(void) printf("%3llu%%\n",
1770 			    (u_longlong_t)mg->mg_fragmentation);
1771 		}
1772 		dump_histogram(mg->mg_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1773 	}
1774 
1775 	(void) printf("\tpool %s\tfragmentation", spa_name(spa));
1776 	fragmentation = metaslab_class_fragmentation(mc);
1777 	if (fragmentation == ZFS_FRAG_INVALID)
1778 		(void) printf("\t%3s\n", "-");
1779 	else
1780 		(void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation);
1781 	dump_histogram(mc->mc_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1782 }
1783 
1784 static void
print_vdev_indirect(vdev_t * vd)1785 print_vdev_indirect(vdev_t *vd)
1786 {
1787 	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
1788 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
1789 	vdev_indirect_births_t *vib = vd->vdev_indirect_births;
1790 
1791 	if (vim == NULL) {
1792 		ASSERT3P(vib, ==, NULL);
1793 		return;
1794 	}
1795 
1796 	ASSERT3U(vdev_indirect_mapping_object(vim), ==,
1797 	    vic->vic_mapping_object);
1798 	ASSERT3U(vdev_indirect_births_object(vib), ==,
1799 	    vic->vic_births_object);
1800 
1801 	(void) printf("indirect births obj %llu:\n",
1802 	    (longlong_t)vic->vic_births_object);
1803 	(void) printf("    vib_count = %llu\n",
1804 	    (longlong_t)vdev_indirect_births_count(vib));
1805 	for (uint64_t i = 0; i < vdev_indirect_births_count(vib); i++) {
1806 		vdev_indirect_birth_entry_phys_t *cur_vibe =
1807 		    &vib->vib_entries[i];
1808 		(void) printf("\toffset %llx -> txg %llu\n",
1809 		    (longlong_t)cur_vibe->vibe_offset,
1810 		    (longlong_t)cur_vibe->vibe_phys_birth_txg);
1811 	}
1812 	(void) printf("\n");
1813 
1814 	(void) printf("indirect mapping obj %llu:\n",
1815 	    (longlong_t)vic->vic_mapping_object);
1816 	(void) printf("    vim_max_offset = 0x%llx\n",
1817 	    (longlong_t)vdev_indirect_mapping_max_offset(vim));
1818 	(void) printf("    vim_bytes_mapped = 0x%llx\n",
1819 	    (longlong_t)vdev_indirect_mapping_bytes_mapped(vim));
1820 	(void) printf("    vim_count = %llu\n",
1821 	    (longlong_t)vdev_indirect_mapping_num_entries(vim));
1822 
1823 	if (dump_opt['d'] <= 5 && dump_opt['m'] <= 3)
1824 		return;
1825 
1826 	uint32_t *counts = vdev_indirect_mapping_load_obsolete_counts(vim);
1827 
1828 	for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
1829 		vdev_indirect_mapping_entry_phys_t *vimep =
1830 		    &vim->vim_entries[i];
1831 		(void) printf("\t<%llx:%llx:%llx> -> "
1832 		    "<%llx:%llx:%llx> (%x obsolete)\n",
1833 		    (longlong_t)vd->vdev_id,
1834 		    (longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
1835 		    (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1836 		    (longlong_t)DVA_GET_VDEV(&vimep->vimep_dst),
1837 		    (longlong_t)DVA_GET_OFFSET(&vimep->vimep_dst),
1838 		    (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1839 		    counts[i]);
1840 	}
1841 	(void) printf("\n");
1842 
1843 	uint64_t obsolete_sm_object;
1844 	VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1845 	if (obsolete_sm_object != 0) {
1846 		objset_t *mos = vd->vdev_spa->spa_meta_objset;
1847 		(void) printf("obsolete space map object %llu:\n",
1848 		    (u_longlong_t)obsolete_sm_object);
1849 		ASSERT(vd->vdev_obsolete_sm != NULL);
1850 		ASSERT3U(space_map_object(vd->vdev_obsolete_sm), ==,
1851 		    obsolete_sm_object);
1852 		dump_spacemap(mos, vd->vdev_obsolete_sm);
1853 		(void) printf("\n");
1854 	}
1855 }
1856 
1857 static void
dump_metaslabs(spa_t * spa)1858 dump_metaslabs(spa_t *spa)
1859 {
1860 	vdev_t *vd, *rvd = spa->spa_root_vdev;
1861 	uint64_t m, c = 0, children = rvd->vdev_children;
1862 
1863 	(void) printf("\nMetaslabs:\n");
1864 
1865 	if (!dump_opt['d'] && zopt_metaslab_args > 0) {
1866 		c = zopt_metaslab[0];
1867 
1868 		if (c >= children)
1869 			(void) fatal("bad vdev id: %llu", (u_longlong_t)c);
1870 
1871 		if (zopt_metaslab_args > 1) {
1872 			vd = rvd->vdev_child[c];
1873 			print_vdev_metaslab_header(vd);
1874 
1875 			for (m = 1; m < zopt_metaslab_args; m++) {
1876 				if (zopt_metaslab[m] < vd->vdev_ms_count)
1877 					dump_metaslab(
1878 					    vd->vdev_ms[zopt_metaslab[m]]);
1879 				else
1880 					(void) fprintf(stderr, "bad metaslab "
1881 					    "number %llu\n",
1882 					    (u_longlong_t)zopt_metaslab[m]);
1883 			}
1884 			(void) printf("\n");
1885 			return;
1886 		}
1887 		children = c + 1;
1888 	}
1889 	for (; c < children; c++) {
1890 		vd = rvd->vdev_child[c];
1891 		print_vdev_metaslab_header(vd);
1892 
1893 		print_vdev_indirect(vd);
1894 
1895 		for (m = 0; m < vd->vdev_ms_count; m++)
1896 			dump_metaslab(vd->vdev_ms[m]);
1897 		(void) printf("\n");
1898 	}
1899 }
1900 
1901 static void
dump_log_spacemaps(spa_t * spa)1902 dump_log_spacemaps(spa_t *spa)
1903 {
1904 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1905 		return;
1906 
1907 	(void) printf("\nLog Space Maps in Pool:\n");
1908 	for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
1909 	    sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1910 		space_map_t *sm = NULL;
1911 		VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
1912 		    sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
1913 
1914 		(void) printf("Log Spacemap object %llu txg %llu\n",
1915 		    (u_longlong_t)sls->sls_sm_obj, (u_longlong_t)sls->sls_txg);
1916 		dump_spacemap(spa->spa_meta_objset, sm);
1917 		space_map_close(sm);
1918 	}
1919 	(void) printf("\n");
1920 }
1921 
1922 static void
dump_ddt_entry(const ddt_t * ddt,const ddt_lightweight_entry_t * ddlwe,uint64_t index)1923 dump_ddt_entry(const ddt_t *ddt, const ddt_lightweight_entry_t *ddlwe,
1924     uint64_t index)
1925 {
1926 	const ddt_key_t *ddk = &ddlwe->ddlwe_key;
1927 	char blkbuf[BP_SPRINTF_LEN];
1928 	blkptr_t blk;
1929 	int p;
1930 
1931 	for (p = 0; p < DDT_NPHYS(ddt); p++) {
1932 		const ddt_univ_phys_t *ddp = &ddlwe->ddlwe_phys;
1933 		ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p);
1934 
1935 		if (ddt_phys_birth(ddp, v) == 0)
1936 			continue;
1937 		ddt_bp_create(ddt->ddt_checksum, ddk, ddp, v, &blk);
1938 		snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
1939 		(void) printf("index %llx refcnt %llu phys %d %s\n",
1940 		    (u_longlong_t)index, (u_longlong_t)ddt_phys_refcnt(ddp, v),
1941 		    p, blkbuf);
1942 	}
1943 }
1944 
1945 static void
dump_dedup_ratio(const ddt_stat_t * dds)1946 dump_dedup_ratio(const ddt_stat_t *dds)
1947 {
1948 	double rL, rP, rD, D, dedup, compress, copies;
1949 
1950 	if (dds->dds_blocks == 0)
1951 		return;
1952 
1953 	rL = (double)dds->dds_ref_lsize;
1954 	rP = (double)dds->dds_ref_psize;
1955 	rD = (double)dds->dds_ref_dsize;
1956 	D = (double)dds->dds_dsize;
1957 
1958 	dedup = rD / D;
1959 	compress = rL / rP;
1960 	copies = rD / rP;
1961 
1962 	(void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
1963 	    "dedup * compress / copies = %.2f\n\n",
1964 	    dedup, compress, copies, dedup * compress / copies);
1965 }
1966 
1967 static void
dump_ddt_log(ddt_t * ddt)1968 dump_ddt_log(ddt_t *ddt)
1969 {
1970 	for (int n = 0; n < 2; n++) {
1971 		ddt_log_t *ddl = &ddt->ddt_log[n];
1972 
1973 		uint64_t count = avl_numnodes(&ddl->ddl_tree);
1974 		if (count == 0)
1975 			continue;
1976 
1977 		printf(DMU_POOL_DDT_LOG ": %" PRIu64 " log entries\n",
1978 		    zio_checksum_table[ddt->ddt_checksum].ci_name, n, count);
1979 
1980 		if (dump_opt['D'] < 4)
1981 			continue;
1982 
1983 		ddt_lightweight_entry_t ddlwe;
1984 		uint64_t index = 0;
1985 		for (ddt_log_entry_t *ddle = avl_first(&ddl->ddl_tree);
1986 		    ddle; ddle = AVL_NEXT(&ddl->ddl_tree, ddle)) {
1987 			DDT_LOG_ENTRY_TO_LIGHTWEIGHT(ddt, ddle, &ddlwe);
1988 			dump_ddt_entry(ddt, &ddlwe, index++);
1989 		}
1990 	}
1991 }
1992 
1993 static void
dump_ddt(ddt_t * ddt,ddt_type_t type,ddt_class_t class)1994 dump_ddt(ddt_t *ddt, ddt_type_t type, ddt_class_t class)
1995 {
1996 	char name[DDT_NAMELEN];
1997 	ddt_lightweight_entry_t ddlwe;
1998 	uint64_t walk = 0;
1999 	dmu_object_info_t doi;
2000 	uint64_t count, dspace, mspace;
2001 	int error;
2002 
2003 	error = ddt_object_info(ddt, type, class, &doi);
2004 
2005 	if (error == ENOENT)
2006 		return;
2007 	ASSERT(error == 0);
2008 
2009 	error = ddt_object_count(ddt, type, class, &count);
2010 	ASSERT(error == 0);
2011 	if (count == 0)
2012 		return;
2013 
2014 	dspace = doi.doi_physical_blocks_512 << 9;
2015 	mspace = doi.doi_fill_count * doi.doi_data_block_size;
2016 
2017 	ddt_object_name(ddt, type, class, name);
2018 
2019 	(void) printf("%s: %llu entries, size %llu on disk, %llu in core\n",
2020 	    name,
2021 	    (u_longlong_t)count,
2022 	    (u_longlong_t)dspace,
2023 	    (u_longlong_t)mspace);
2024 
2025 	if (dump_opt['D'] < 3)
2026 		return;
2027 
2028 	zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class]);
2029 
2030 	if (dump_opt['D'] < 4)
2031 		return;
2032 
2033 	if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE)
2034 		return;
2035 
2036 	(void) printf("%s contents:\n\n", name);
2037 
2038 	while ((error = ddt_object_walk(ddt, type, class, &walk, &ddlwe)) == 0)
2039 		dump_ddt_entry(ddt, &ddlwe, walk);
2040 
2041 	ASSERT3U(error, ==, ENOENT);
2042 
2043 	(void) printf("\n");
2044 }
2045 
2046 static void
dump_all_ddts(spa_t * spa)2047 dump_all_ddts(spa_t *spa)
2048 {
2049 	ddt_histogram_t ddh_total = {{{0}}};
2050 	ddt_stat_t dds_total = {0};
2051 
2052 	for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
2053 		ddt_t *ddt = spa->spa_ddt[c];
2054 		if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
2055 			continue;
2056 		for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
2057 			for (ddt_class_t class = 0; class < DDT_CLASSES;
2058 			    class++) {
2059 				dump_ddt(ddt, type, class);
2060 			}
2061 		}
2062 		dump_ddt_log(ddt);
2063 	}
2064 
2065 	ddt_get_dedup_stats(spa, &dds_total);
2066 
2067 	if (dds_total.dds_blocks == 0) {
2068 		(void) printf("All DDTs are empty\n");
2069 		return;
2070 	}
2071 
2072 	(void) printf("\n");
2073 
2074 	if (dump_opt['D'] > 1) {
2075 		(void) printf("DDT histogram (aggregated over all DDTs):\n");
2076 		ddt_get_dedup_histogram(spa, &ddh_total);
2077 		zpool_dump_ddt(&dds_total, &ddh_total);
2078 	}
2079 
2080 	dump_dedup_ratio(&dds_total);
2081 
2082 	/*
2083 	 * Dump a histogram of unique class entry age
2084 	 */
2085 	if (dump_opt['D'] == 3 && getenv("ZDB_DDT_UNIQUE_AGE_HIST") != NULL) {
2086 		ddt_age_histo_t histogram;
2087 
2088 		(void) printf("DDT walk unique, building age histogram...\n");
2089 		ddt_prune_walk(spa, 0, &histogram);
2090 
2091 		/*
2092 		 * print out histogram for unique entry class birth
2093 		 */
2094 		if (histogram.dah_entries > 0) {
2095 			(void) printf("%5s  %9s  %4s\n",
2096 			    "age", "blocks", "amnt");
2097 			(void) printf("%5s  %9s  %4s\n",
2098 			    "-----", "---------", "----");
2099 			for (int i = 0; i < HIST_BINS; i++) {
2100 				(void) printf("%5d  %9d %4d%%\n", 1 << i,
2101 				    (int)histogram.dah_age_histo[i],
2102 				    (int)((histogram.dah_age_histo[i] * 100) /
2103 				    histogram.dah_entries));
2104 			}
2105 		}
2106 	}
2107 }
2108 
2109 static void
dump_brt(spa_t * spa)2110 dump_brt(spa_t *spa)
2111 {
2112 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) {
2113 		printf("BRT: unsupported on this pool\n");
2114 		return;
2115 	}
2116 
2117 	if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
2118 		printf("BRT: empty\n");
2119 		return;
2120 	}
2121 
2122 	char count[32], used[32], saved[32];
2123 	zdb_nicebytes(brt_get_used(spa), used, sizeof (used));
2124 	zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved));
2125 	uint64_t ratio = brt_get_ratio(spa);
2126 	printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved,
2127 	    (u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100));
2128 
2129 	if (dump_opt['T'] < 2)
2130 		return;
2131 
2132 	for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2133 		brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2134 		if (!brtvd->bv_initiated) {
2135 			printf("BRT: vdev %" PRIu64 ": empty\n", vdevid);
2136 			continue;
2137 		}
2138 
2139 		zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count));
2140 		zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used));
2141 		zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved));
2142 		printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n",
2143 		    vdevid, count, used, saved);
2144 	}
2145 
2146 	if (dump_opt['T'] < 3)
2147 		return;
2148 
2149 	/* -TTT shows a per-vdev histograms; -TTTT shows all entries */
2150 	boolean_t do_histo = dump_opt['T'] == 3;
2151 
2152 	char dva[64];
2153 
2154 	if (!do_histo)
2155 		printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2156 
2157 	for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2158 		brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2159 		if (!brtvd->bv_initiated)
2160 			continue;
2161 
2162 		uint64_t counts[64] = {};
2163 
2164 		zap_cursor_t zc;
2165 		zap_attribute_t *za = zap_attribute_alloc();
2166 		for (zap_cursor_init(&zc, spa->spa_meta_objset,
2167 		    brtvd->bv_mos_entries);
2168 		    zap_cursor_retrieve(&zc, za) == 0;
2169 		    zap_cursor_advance(&zc)) {
2170 			uint64_t refcnt;
2171 			VERIFY0(zap_lookup_uint64(spa->spa_meta_objset,
2172 			    brtvd->bv_mos_entries,
2173 			    (const uint64_t *)za->za_name, 1,
2174 			    za->za_integer_length, za->za_num_integers,
2175 			    &refcnt));
2176 
2177 			if (do_histo)
2178 				counts[highbit64(refcnt)]++;
2179 			else {
2180 				uint64_t offset =
2181 				    *(const uint64_t *)za->za_name;
2182 
2183 				snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx",
2184 				    vdevid, (u_longlong_t)offset);
2185 				printf("%-16s %-10llu\n", dva,
2186 				    (u_longlong_t)refcnt);
2187 			}
2188 		}
2189 		zap_cursor_fini(&zc);
2190 		zap_attribute_free(za);
2191 
2192 		if (do_histo) {
2193 			printf("\nBRT: vdev %" PRIu64
2194 			    ": DVAs with 2^n refcnts:\n", vdevid);
2195 			dump_histogram(counts, 64, 0);
2196 		}
2197 	}
2198 }
2199 
2200 static void
dump_dtl_seg(void * arg,uint64_t start,uint64_t size)2201 dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
2202 {
2203 	char *prefix = arg;
2204 
2205 	(void) printf("%s [%llu,%llu) length %llu\n",
2206 	    prefix,
2207 	    (u_longlong_t)start,
2208 	    (u_longlong_t)(start + size),
2209 	    (u_longlong_t)(size));
2210 }
2211 
2212 static void
dump_dtl(vdev_t * vd,int indent)2213 dump_dtl(vdev_t *vd, int indent)
2214 {
2215 	spa_t *spa = vd->vdev_spa;
2216 	boolean_t required;
2217 	const char *name[DTL_TYPES] = { "missing", "partial", "scrub",
2218 		"outage" };
2219 	char prefix[256];
2220 
2221 	spa_vdev_state_enter(spa, SCL_NONE);
2222 	required = vdev_dtl_required(vd);
2223 	(void) spa_vdev_state_exit(spa, NULL, 0);
2224 
2225 	if (indent == 0)
2226 		(void) printf("\nDirty time logs:\n\n");
2227 
2228 	(void) printf("\t%*s%s [%s]\n", indent, "",
2229 	    vd->vdev_path ? vd->vdev_path :
2230 	    vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa),
2231 	    required ? "DTL-required" : "DTL-expendable");
2232 
2233 	for (int t = 0; t < DTL_TYPES; t++) {
2234 		range_tree_t *rt = vd->vdev_dtl[t];
2235 		if (range_tree_space(rt) == 0)
2236 			continue;
2237 		(void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
2238 		    indent + 2, "", name[t]);
2239 		range_tree_walk(rt, dump_dtl_seg, prefix);
2240 		if (dump_opt['d'] > 5 && vd->vdev_children == 0)
2241 			dump_spacemap(spa->spa_meta_objset,
2242 			    vd->vdev_dtl_sm);
2243 	}
2244 
2245 	for (unsigned c = 0; c < vd->vdev_children; c++)
2246 		dump_dtl(vd->vdev_child[c], indent + 4);
2247 }
2248 
2249 static void
dump_history(spa_t * spa)2250 dump_history(spa_t *spa)
2251 {
2252 	nvlist_t **events = NULL;
2253 	char *buf;
2254 	uint64_t resid, len, off = 0;
2255 	uint_t num = 0;
2256 	int error;
2257 	char tbuf[30];
2258 
2259 	if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) {
2260 		(void) fprintf(stderr, "%s: unable to allocate I/O buffer\n",
2261 		    __func__);
2262 		return;
2263 	}
2264 
2265 	do {
2266 		len = SPA_OLD_MAXBLOCKSIZE;
2267 
2268 		if ((error = spa_history_get(spa, &off, &len, buf)) != 0) {
2269 			(void) fprintf(stderr, "Unable to read history: "
2270 			    "error %d\n", error);
2271 			free(buf);
2272 			return;
2273 		}
2274 
2275 		if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0)
2276 			break;
2277 
2278 		off -= resid;
2279 	} while (len != 0);
2280 
2281 	(void) printf("\nHistory:\n");
2282 	for (unsigned i = 0; i < num; i++) {
2283 		boolean_t printed = B_FALSE;
2284 
2285 		if (nvlist_exists(events[i], ZPOOL_HIST_TIME)) {
2286 			time_t tsec;
2287 			struct tm t;
2288 
2289 			tsec = fnvlist_lookup_uint64(events[i],
2290 			    ZPOOL_HIST_TIME);
2291 			(void) localtime_r(&tsec, &t);
2292 			(void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t);
2293 		} else {
2294 			tbuf[0] = '\0';
2295 		}
2296 
2297 		if (nvlist_exists(events[i], ZPOOL_HIST_CMD)) {
2298 			(void) printf("%s %s\n", tbuf,
2299 			    fnvlist_lookup_string(events[i], ZPOOL_HIST_CMD));
2300 		} else if (nvlist_exists(events[i], ZPOOL_HIST_INT_EVENT)) {
2301 			uint64_t ievent;
2302 
2303 			ievent = fnvlist_lookup_uint64(events[i],
2304 			    ZPOOL_HIST_INT_EVENT);
2305 			if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS)
2306 				goto next;
2307 
2308 			(void) printf(" %s [internal %s txg:%ju] %s\n",
2309 			    tbuf,
2310 			    zfs_history_event_names[ievent],
2311 			    fnvlist_lookup_uint64(events[i],
2312 			    ZPOOL_HIST_TXG),
2313 			    fnvlist_lookup_string(events[i],
2314 			    ZPOOL_HIST_INT_STR));
2315 		} else if (nvlist_exists(events[i], ZPOOL_HIST_INT_NAME)) {
2316 			(void) printf("%s [txg:%ju] %s", tbuf,
2317 			    fnvlist_lookup_uint64(events[i],
2318 			    ZPOOL_HIST_TXG),
2319 			    fnvlist_lookup_string(events[i],
2320 			    ZPOOL_HIST_INT_NAME));
2321 
2322 			if (nvlist_exists(events[i], ZPOOL_HIST_DSNAME)) {
2323 				(void) printf(" %s (%llu)",
2324 				    fnvlist_lookup_string(events[i],
2325 				    ZPOOL_HIST_DSNAME),
2326 				    (u_longlong_t)fnvlist_lookup_uint64(
2327 				    events[i],
2328 				    ZPOOL_HIST_DSID));
2329 			}
2330 
2331 			(void) printf(" %s\n", fnvlist_lookup_string(events[i],
2332 			    ZPOOL_HIST_INT_STR));
2333 		} else if (nvlist_exists(events[i], ZPOOL_HIST_IOCTL)) {
2334 			(void) printf("%s ioctl %s\n", tbuf,
2335 			    fnvlist_lookup_string(events[i],
2336 			    ZPOOL_HIST_IOCTL));
2337 
2338 			if (nvlist_exists(events[i], ZPOOL_HIST_INPUT_NVL)) {
2339 				(void) printf("    input:\n");
2340 				dump_nvlist(fnvlist_lookup_nvlist(events[i],
2341 				    ZPOOL_HIST_INPUT_NVL), 8);
2342 			}
2343 			if (nvlist_exists(events[i], ZPOOL_HIST_OUTPUT_NVL)) {
2344 				(void) printf("    output:\n");
2345 				dump_nvlist(fnvlist_lookup_nvlist(events[i],
2346 				    ZPOOL_HIST_OUTPUT_NVL), 8);
2347 			}
2348 			if (nvlist_exists(events[i], ZPOOL_HIST_ERRNO)) {
2349 				(void) printf("    errno: %lld\n",
2350 				    (longlong_t)fnvlist_lookup_int64(events[i],
2351 				    ZPOOL_HIST_ERRNO));
2352 			}
2353 		} else {
2354 			goto next;
2355 		}
2356 
2357 		printed = B_TRUE;
2358 next:
2359 		if (dump_opt['h'] > 1) {
2360 			if (!printed)
2361 				(void) printf("unrecognized record:\n");
2362 			dump_nvlist(events[i], 2);
2363 		}
2364 	}
2365 	free(buf);
2366 }
2367 
2368 static void
dump_dnode(objset_t * os,uint64_t object,void * data,size_t size)2369 dump_dnode(objset_t *os, uint64_t object, void *data, size_t size)
2370 {
2371 	(void) os, (void) object, (void) data, (void) size;
2372 }
2373 
2374 static uint64_t
blkid2offset(const dnode_phys_t * dnp,const blkptr_t * bp,const zbookmark_phys_t * zb)2375 blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
2376     const zbookmark_phys_t *zb)
2377 {
2378 	if (dnp == NULL) {
2379 		ASSERT(zb->zb_level < 0);
2380 		if (zb->zb_object == 0)
2381 			return (zb->zb_blkid);
2382 		return (zb->zb_blkid * BP_GET_LSIZE(bp));
2383 	}
2384 
2385 	ASSERT(zb->zb_level >= 0);
2386 
2387 	return ((zb->zb_blkid <<
2388 	    (zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) *
2389 	    dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
2390 }
2391 
2392 static void
snprintf_zstd_header(spa_t * spa,char * blkbuf,size_t buflen,const blkptr_t * bp)2393 snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
2394     const blkptr_t *bp)
2395 {
2396 	static abd_t *pabd = NULL;
2397 	void *buf;
2398 	zio_t *zio;
2399 	zfs_zstdhdr_t zstd_hdr;
2400 	int error;
2401 
2402 	if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD)
2403 		return;
2404 
2405 	if (BP_IS_HOLE(bp))
2406 		return;
2407 
2408 	if (BP_IS_EMBEDDED(bp)) {
2409 		buf = malloc(SPA_MAXBLOCKSIZE);
2410 		if (buf == NULL) {
2411 			(void) fprintf(stderr, "out of memory\n");
2412 			zdb_exit(1);
2413 		}
2414 		decode_embedded_bp_compressed(bp, buf);
2415 		memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2416 		free(buf);
2417 		zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2418 		zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2419 		(void) snprintf(blkbuf + strlen(blkbuf),
2420 		    buflen - strlen(blkbuf),
2421 		    " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2422 		    zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2423 		    zfs_get_hdrlevel(&zstd_hdr));
2424 		return;
2425 	}
2426 
2427 	if (!pabd)
2428 		pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
2429 	zio = zio_root(spa, NULL, NULL, 0);
2430 
2431 	/* Decrypt but don't decompress so we can read the compression header */
2432 	zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL,
2433 	    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS,
2434 	    NULL));
2435 	error = zio_wait(zio);
2436 	if (error) {
2437 		(void) fprintf(stderr, "read failed: %d\n", error);
2438 		return;
2439 	}
2440 	buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp));
2441 	memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2442 	zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2443 	zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2444 
2445 	(void) snprintf(blkbuf + strlen(blkbuf),
2446 	    buflen - strlen(blkbuf),
2447 	    " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2448 	    zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2449 	    zfs_get_hdrlevel(&zstd_hdr));
2450 
2451 	abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp));
2452 }
2453 
2454 static void
snprintf_blkptr_compact(char * blkbuf,size_t buflen,const blkptr_t * bp,boolean_t bp_freed)2455 snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
2456     boolean_t bp_freed)
2457 {
2458 	const dva_t *dva = bp->blk_dva;
2459 	int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1;
2460 	int i;
2461 
2462 	if (dump_opt['b'] >= 6) {
2463 		snprintf_blkptr(blkbuf, buflen, bp);
2464 		if (bp_freed) {
2465 			(void) snprintf(blkbuf + strlen(blkbuf),
2466 			    buflen - strlen(blkbuf), " %s", "FREE");
2467 		}
2468 		return;
2469 	}
2470 
2471 	if (BP_IS_EMBEDDED(bp)) {
2472 		(void) sprintf(blkbuf,
2473 		    "EMBEDDED et=%u %llxL/%llxP B=%llu",
2474 		    (int)BPE_GET_ETYPE(bp),
2475 		    (u_longlong_t)BPE_GET_LSIZE(bp),
2476 		    (u_longlong_t)BPE_GET_PSIZE(bp),
2477 		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2478 		return;
2479 	}
2480 
2481 	blkbuf[0] = '\0';
2482 
2483 	for (i = 0; i < ndvas; i++)
2484 		(void) snprintf(blkbuf + strlen(blkbuf),
2485 		    buflen - strlen(blkbuf), "%llu:%llx:%llx ",
2486 		    (u_longlong_t)DVA_GET_VDEV(&dva[i]),
2487 		    (u_longlong_t)DVA_GET_OFFSET(&dva[i]),
2488 		    (u_longlong_t)DVA_GET_ASIZE(&dva[i]));
2489 
2490 	if (BP_IS_HOLE(bp)) {
2491 		(void) snprintf(blkbuf + strlen(blkbuf),
2492 		    buflen - strlen(blkbuf),
2493 		    "%llxL B=%llu",
2494 		    (u_longlong_t)BP_GET_LSIZE(bp),
2495 		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2496 	} else {
2497 		(void) snprintf(blkbuf + strlen(blkbuf),
2498 		    buflen - strlen(blkbuf),
2499 		    "%llxL/%llxP F=%llu B=%llu/%llu",
2500 		    (u_longlong_t)BP_GET_LSIZE(bp),
2501 		    (u_longlong_t)BP_GET_PSIZE(bp),
2502 		    (u_longlong_t)BP_GET_FILL(bp),
2503 		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),
2504 		    (u_longlong_t)BP_GET_BIRTH(bp));
2505 		if (bp_freed)
2506 			(void) snprintf(blkbuf + strlen(blkbuf),
2507 			    buflen - strlen(blkbuf), " %s", "FREE");
2508 		(void) snprintf(blkbuf + strlen(blkbuf),
2509 		    buflen - strlen(blkbuf),
2510 		    " cksum=%016llx:%016llx:%016llx:%016llx",
2511 		    (u_longlong_t)bp->blk_cksum.zc_word[0],
2512 		    (u_longlong_t)bp->blk_cksum.zc_word[1],
2513 		    (u_longlong_t)bp->blk_cksum.zc_word[2],
2514 		    (u_longlong_t)bp->blk_cksum.zc_word[3]);
2515 	}
2516 }
2517 
2518 static void
print_indirect(spa_t * spa,blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp)2519 print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
2520     const dnode_phys_t *dnp)
2521 {
2522 	char blkbuf[BP_SPRINTF_LEN];
2523 	int l;
2524 
2525 	if (!BP_IS_EMBEDDED(bp)) {
2526 		ASSERT3U(BP_GET_TYPE(bp), ==, dnp->dn_type);
2527 		ASSERT3U(BP_GET_LEVEL(bp), ==, zb->zb_level);
2528 	}
2529 
2530 	(void) printf("%16llx ", (u_longlong_t)blkid2offset(dnp, bp, zb));
2531 
2532 	ASSERT(zb->zb_level >= 0);
2533 
2534 	for (l = dnp->dn_nlevels - 1; l >= -1; l--) {
2535 		if (l == zb->zb_level) {
2536 			(void) printf("L%llx", (u_longlong_t)zb->zb_level);
2537 		} else {
2538 			(void) printf(" ");
2539 		}
2540 	}
2541 
2542 	snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE);
2543 	if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD)
2544 		snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp);
2545 	(void) printf("%s\n", blkbuf);
2546 }
2547 
2548 static int
visit_indirect(spa_t * spa,const dnode_phys_t * dnp,blkptr_t * bp,const zbookmark_phys_t * zb)2549 visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
2550     blkptr_t *bp, const zbookmark_phys_t *zb)
2551 {
2552 	int err = 0;
2553 
2554 	if (BP_GET_LOGICAL_BIRTH(bp) == 0)
2555 		return (0);
2556 
2557 	print_indirect(spa, bp, zb, dnp);
2558 
2559 	if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
2560 		arc_flags_t flags = ARC_FLAG_WAIT;
2561 		int i;
2562 		blkptr_t *cbp;
2563 		int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
2564 		arc_buf_t *buf;
2565 		uint64_t fill = 0;
2566 		ASSERT(!BP_IS_REDACTED(bp));
2567 
2568 		err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
2569 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
2570 		if (err)
2571 			return (err);
2572 		ASSERT(buf->b_data);
2573 
2574 		/* recursively visit blocks below this */
2575 		cbp = buf->b_data;
2576 		for (i = 0; i < epb; i++, cbp++) {
2577 			zbookmark_phys_t czb;
2578 
2579 			SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
2580 			    zb->zb_level - 1,
2581 			    zb->zb_blkid * epb + i);
2582 			err = visit_indirect(spa, dnp, cbp, &czb);
2583 			if (err)
2584 				break;
2585 			fill += BP_GET_FILL(cbp);
2586 		}
2587 		if (!err)
2588 			ASSERT3U(fill, ==, BP_GET_FILL(bp));
2589 		arc_buf_destroy(buf, &buf);
2590 	}
2591 
2592 	return (err);
2593 }
2594 
2595 static void
dump_indirect(dnode_t * dn)2596 dump_indirect(dnode_t *dn)
2597 {
2598 	dnode_phys_t *dnp = dn->dn_phys;
2599 	zbookmark_phys_t czb;
2600 
2601 	(void) printf("Indirect blocks:\n");
2602 
2603 	SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
2604 	    dn->dn_object, dnp->dn_nlevels - 1, 0);
2605 	for (int j = 0; j < dnp->dn_nblkptr; j++) {
2606 		czb.zb_blkid = j;
2607 		(void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
2608 		    &dnp->dn_blkptr[j], &czb);
2609 	}
2610 
2611 	(void) printf("\n");
2612 }
2613 
2614 static void
dump_dsl_dir(objset_t * os,uint64_t object,void * data,size_t size)2615 dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size)
2616 {
2617 	(void) os, (void) object;
2618 	dsl_dir_phys_t *dd = data;
2619 	time_t crtime;
2620 	char nice[32];
2621 
2622 	/* make sure nicenum has enough space */
2623 	_Static_assert(sizeof (nice) >= NN_NUMBUF_SZ, "nice truncated");
2624 
2625 	if (dd == NULL)
2626 		return;
2627 
2628 	ASSERT3U(size, >=, sizeof (dsl_dir_phys_t));
2629 
2630 	crtime = dd->dd_creation_time;
2631 	(void) printf("\t\tcreation_time = %s", ctime(&crtime));
2632 	(void) printf("\t\thead_dataset_obj = %llu\n",
2633 	    (u_longlong_t)dd->dd_head_dataset_obj);
2634 	(void) printf("\t\tparent_dir_obj = %llu\n",
2635 	    (u_longlong_t)dd->dd_parent_obj);
2636 	(void) printf("\t\torigin_obj = %llu\n",
2637 	    (u_longlong_t)dd->dd_origin_obj);
2638 	(void) printf("\t\tchild_dir_zapobj = %llu\n",
2639 	    (u_longlong_t)dd->dd_child_dir_zapobj);
2640 	zdb_nicenum(dd->dd_used_bytes, nice, sizeof (nice));
2641 	(void) printf("\t\tused_bytes = %s\n", nice);
2642 	zdb_nicenum(dd->dd_compressed_bytes, nice, sizeof (nice));
2643 	(void) printf("\t\tcompressed_bytes = %s\n", nice);
2644 	zdb_nicenum(dd->dd_uncompressed_bytes, nice, sizeof (nice));
2645 	(void) printf("\t\tuncompressed_bytes = %s\n", nice);
2646 	zdb_nicenum(dd->dd_quota, nice, sizeof (nice));
2647 	(void) printf("\t\tquota = %s\n", nice);
2648 	zdb_nicenum(dd->dd_reserved, nice, sizeof (nice));
2649 	(void) printf("\t\treserved = %s\n", nice);
2650 	(void) printf("\t\tprops_zapobj = %llu\n",
2651 	    (u_longlong_t)dd->dd_props_zapobj);
2652 	(void) printf("\t\tdeleg_zapobj = %llu\n",
2653 	    (u_longlong_t)dd->dd_deleg_zapobj);
2654 	(void) printf("\t\tflags = %llx\n",
2655 	    (u_longlong_t)dd->dd_flags);
2656 
2657 #define	DO(which) \
2658 	zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2659 	    sizeof (nice)); \
2660 	(void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2661 	DO(HEAD);
2662 	DO(SNAP);
2663 	DO(CHILD);
2664 	DO(CHILD_RSRV);
2665 	DO(REFRSRV);
2666 #undef DO
2667 	(void) printf("\t\tclones = %llu\n",
2668 	    (u_longlong_t)dd->dd_clones);
2669 }
2670 
2671 static void
dump_dsl_dataset(objset_t * os,uint64_t object,void * data,size_t size)2672 dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
2673 {
2674 	(void) os, (void) object;
2675 	dsl_dataset_phys_t *ds = data;
2676 	time_t crtime;
2677 	char used[32], compressed[32], uncompressed[32], unique[32];
2678 	char blkbuf[BP_SPRINTF_LEN];
2679 
2680 	/* make sure nicenum has enough space */
2681 	_Static_assert(sizeof (used) >= NN_NUMBUF_SZ, "used truncated");
2682 	_Static_assert(sizeof (compressed) >= NN_NUMBUF_SZ,
2683 	    "compressed truncated");
2684 	_Static_assert(sizeof (uncompressed) >= NN_NUMBUF_SZ,
2685 	    "uncompressed truncated");
2686 	_Static_assert(sizeof (unique) >= NN_NUMBUF_SZ, "unique truncated");
2687 
2688 	if (ds == NULL)
2689 		return;
2690 
2691 	ASSERT(size == sizeof (*ds));
2692 	crtime = ds->ds_creation_time;
2693 	zdb_nicenum(ds->ds_referenced_bytes, used, sizeof (used));
2694 	zdb_nicenum(ds->ds_compressed_bytes, compressed, sizeof (compressed));
2695 	zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed,
2696 	    sizeof (uncompressed));
2697 	zdb_nicenum(ds->ds_unique_bytes, unique, sizeof (unique));
2698 	snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp);
2699 
2700 	(void) printf("\t\tdir_obj = %llu\n",
2701 	    (u_longlong_t)ds->ds_dir_obj);
2702 	(void) printf("\t\tprev_snap_obj = %llu\n",
2703 	    (u_longlong_t)ds->ds_prev_snap_obj);
2704 	(void) printf("\t\tprev_snap_txg = %llu\n",
2705 	    (u_longlong_t)ds->ds_prev_snap_txg);
2706 	(void) printf("\t\tnext_snap_obj = %llu\n",
2707 	    (u_longlong_t)ds->ds_next_snap_obj);
2708 	(void) printf("\t\tsnapnames_zapobj = %llu\n",
2709 	    (u_longlong_t)ds->ds_snapnames_zapobj);
2710 	(void) printf("\t\tnum_children = %llu\n",
2711 	    (u_longlong_t)ds->ds_num_children);
2712 	(void) printf("\t\tuserrefs_obj = %llu\n",
2713 	    (u_longlong_t)ds->ds_userrefs_obj);
2714 	(void) printf("\t\tcreation_time = %s", ctime(&crtime));
2715 	(void) printf("\t\tcreation_txg = %llu\n",
2716 	    (u_longlong_t)ds->ds_creation_txg);
2717 	(void) printf("\t\tdeadlist_obj = %llu\n",
2718 	    (u_longlong_t)ds->ds_deadlist_obj);
2719 	(void) printf("\t\tused_bytes = %s\n", used);
2720 	(void) printf("\t\tcompressed_bytes = %s\n", compressed);
2721 	(void) printf("\t\tuncompressed_bytes = %s\n", uncompressed);
2722 	(void) printf("\t\tunique = %s\n", unique);
2723 	(void) printf("\t\tfsid_guid = %llu\n",
2724 	    (u_longlong_t)ds->ds_fsid_guid);
2725 	(void) printf("\t\tguid = %llu\n",
2726 	    (u_longlong_t)ds->ds_guid);
2727 	(void) printf("\t\tflags = %llx\n",
2728 	    (u_longlong_t)ds->ds_flags);
2729 	(void) printf("\t\tnext_clones_obj = %llu\n",
2730 	    (u_longlong_t)ds->ds_next_clones_obj);
2731 	(void) printf("\t\tprops_obj = %llu\n",
2732 	    (u_longlong_t)ds->ds_props_obj);
2733 	(void) printf("\t\tbp = %s\n", blkbuf);
2734 }
2735 
2736 static int
dump_bptree_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)2737 dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2738 {
2739 	(void) arg, (void) tx;
2740 	char blkbuf[BP_SPRINTF_LEN];
2741 
2742 	if (BP_GET_LOGICAL_BIRTH(bp) != 0) {
2743 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
2744 		(void) printf("\t%s\n", blkbuf);
2745 	}
2746 	return (0);
2747 }
2748 
2749 static void
dump_bptree(objset_t * os,uint64_t obj,const char * name)2750 dump_bptree(objset_t *os, uint64_t obj, const char *name)
2751 {
2752 	char bytes[32];
2753 	bptree_phys_t *bt;
2754 	dmu_buf_t *db;
2755 
2756 	/* make sure nicenum has enough space */
2757 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2758 
2759 	if (dump_opt['d'] < 3)
2760 		return;
2761 
2762 	VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
2763 	bt = db->db_data;
2764 	zdb_nicenum(bt->bt_bytes, bytes, sizeof (bytes));
2765 	(void) printf("\n    %s: %llu datasets, %s\n",
2766 	    name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes);
2767 	dmu_buf_rele(db, FTAG);
2768 
2769 	if (dump_opt['d'] < 5)
2770 		return;
2771 
2772 	(void) printf("\n");
2773 
2774 	(void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL);
2775 }
2776 
2777 static int
dump_bpobj_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)2778 dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
2779 {
2780 	(void) arg, (void) tx;
2781 	char blkbuf[BP_SPRINTF_LEN];
2782 
2783 	ASSERT(BP_GET_LOGICAL_BIRTH(bp) != 0);
2784 	snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed);
2785 	(void) printf("\t%s\n", blkbuf);
2786 	return (0);
2787 }
2788 
2789 static void
dump_full_bpobj(bpobj_t * bpo,const char * name,int indent)2790 dump_full_bpobj(bpobj_t *bpo, const char *name, int indent)
2791 {
2792 	char bytes[32];
2793 	char comp[32];
2794 	char uncomp[32];
2795 	uint64_t i;
2796 
2797 	/* make sure nicenum has enough space */
2798 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2799 	_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
2800 	_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
2801 
2802 	if (dump_opt['d'] < 3)
2803 		return;
2804 
2805 	zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes, sizeof (bytes));
2806 	if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
2807 		zdb_nicenum(bpo->bpo_phys->bpo_comp, comp, sizeof (comp));
2808 		zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp, sizeof (uncomp));
2809 		if (bpo->bpo_havefreed) {
2810 			(void) printf("    %*s: object %llu, %llu local "
2811 			    "blkptrs, %llu freed, %llu subobjs in object %llu, "
2812 			    "%s (%s/%s comp)\n",
2813 			    indent * 8, name,
2814 			    (u_longlong_t)bpo->bpo_object,
2815 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2816 			    (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2817 			    (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2818 			    (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2819 			    bytes, comp, uncomp);
2820 		} else {
2821 			(void) printf("    %*s: object %llu, %llu local "
2822 			    "blkptrs, %llu subobjs in object %llu, "
2823 			    "%s (%s/%s comp)\n",
2824 			    indent * 8, name,
2825 			    (u_longlong_t)bpo->bpo_object,
2826 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2827 			    (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2828 			    (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2829 			    bytes, comp, uncomp);
2830 		}
2831 
2832 		for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
2833 			uint64_t subobj;
2834 			bpobj_t subbpo;
2835 			int error;
2836 			VERIFY0(dmu_read(bpo->bpo_os,
2837 			    bpo->bpo_phys->bpo_subobjs,
2838 			    i * sizeof (subobj), sizeof (subobj), &subobj, 0));
2839 			error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
2840 			if (error != 0) {
2841 				(void) printf("ERROR %u while trying to open "
2842 				    "subobj id %llu\n",
2843 				    error, (u_longlong_t)subobj);
2844 				continue;
2845 			}
2846 			dump_full_bpobj(&subbpo, "subobj", indent + 1);
2847 			bpobj_close(&subbpo);
2848 		}
2849 	} else {
2850 		if (bpo->bpo_havefreed) {
2851 			(void) printf("    %*s: object %llu, %llu blkptrs, "
2852 			    "%llu freed, %s\n",
2853 			    indent * 8, name,
2854 			    (u_longlong_t)bpo->bpo_object,
2855 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2856 			    (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2857 			    bytes);
2858 		} else {
2859 			(void) printf("    %*s: object %llu, %llu blkptrs, "
2860 			    "%s\n",
2861 			    indent * 8, name,
2862 			    (u_longlong_t)bpo->bpo_object,
2863 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2864 			    bytes);
2865 		}
2866 	}
2867 
2868 	if (dump_opt['d'] < 5)
2869 		return;
2870 
2871 
2872 	if (indent == 0) {
2873 		(void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL);
2874 		(void) printf("\n");
2875 	}
2876 }
2877 
2878 static int
dump_bookmark(dsl_pool_t * dp,char * name,boolean_t print_redact,boolean_t print_list)2879 dump_bookmark(dsl_pool_t *dp, char *name, boolean_t print_redact,
2880     boolean_t print_list)
2881 {
2882 	int err = 0;
2883 	zfs_bookmark_phys_t prop;
2884 	objset_t *mos = dp->dp_spa->spa_meta_objset;
2885 	err = dsl_bookmark_lookup(dp, name, NULL, &prop);
2886 
2887 	if (err != 0) {
2888 		return (err);
2889 	}
2890 
2891 	(void) printf("\t#%s: ", strchr(name, '#') + 1);
2892 	(void) printf("{guid: %llx creation_txg: %llu creation_time: "
2893 	    "%llu redaction_obj: %llu}\n", (u_longlong_t)prop.zbm_guid,
2894 	    (u_longlong_t)prop.zbm_creation_txg,
2895 	    (u_longlong_t)prop.zbm_creation_time,
2896 	    (u_longlong_t)prop.zbm_redaction_obj);
2897 
2898 	IMPLY(print_list, print_redact);
2899 	if (!print_redact || prop.zbm_redaction_obj == 0)
2900 		return (0);
2901 
2902 	redaction_list_t *rl;
2903 	VERIFY0(dsl_redaction_list_hold_obj(dp,
2904 	    prop.zbm_redaction_obj, FTAG, &rl));
2905 
2906 	redaction_list_phys_t *rlp = rl->rl_phys;
2907 	(void) printf("\tRedacted:\n\t\tProgress: ");
2908 	if (rlp->rlp_last_object != UINT64_MAX ||
2909 	    rlp->rlp_last_blkid != UINT64_MAX) {
2910 		(void) printf("%llu %llu (incomplete)\n",
2911 		    (u_longlong_t)rlp->rlp_last_object,
2912 		    (u_longlong_t)rlp->rlp_last_blkid);
2913 	} else {
2914 		(void) printf("complete\n");
2915 	}
2916 	(void) printf("\t\tSnapshots: [");
2917 	for (unsigned int i = 0; i < rlp->rlp_num_snaps; i++) {
2918 		if (i > 0)
2919 			(void) printf(", ");
2920 		(void) printf("%0llu",
2921 		    (u_longlong_t)rlp->rlp_snaps[i]);
2922 	}
2923 	(void) printf("]\n\t\tLength: %llu\n",
2924 	    (u_longlong_t)rlp->rlp_num_entries);
2925 
2926 	if (!print_list) {
2927 		dsl_redaction_list_rele(rl, FTAG);
2928 		return (0);
2929 	}
2930 
2931 	if (rlp->rlp_num_entries == 0) {
2932 		dsl_redaction_list_rele(rl, FTAG);
2933 		(void) printf("\t\tRedaction List: []\n\n");
2934 		return (0);
2935 	}
2936 
2937 	redact_block_phys_t *rbp_buf;
2938 	uint64_t size;
2939 	dmu_object_info_t doi;
2940 
2941 	VERIFY0(dmu_object_info(mos, prop.zbm_redaction_obj, &doi));
2942 	size = doi.doi_max_offset;
2943 	rbp_buf = kmem_alloc(size, KM_SLEEP);
2944 
2945 	err = dmu_read(mos, prop.zbm_redaction_obj, 0, size,
2946 	    rbp_buf, 0);
2947 	if (err != 0) {
2948 		dsl_redaction_list_rele(rl, FTAG);
2949 		kmem_free(rbp_buf, size);
2950 		return (err);
2951 	}
2952 
2953 	(void) printf("\t\tRedaction List: [{object: %llx, offset: "
2954 	    "%llx, blksz: %x, count: %llx}",
2955 	    (u_longlong_t)rbp_buf[0].rbp_object,
2956 	    (u_longlong_t)rbp_buf[0].rbp_blkid,
2957 	    (uint_t)(redact_block_get_size(&rbp_buf[0])),
2958 	    (u_longlong_t)redact_block_get_count(&rbp_buf[0]));
2959 
2960 	for (size_t i = 1; i < rlp->rlp_num_entries; i++) {
2961 		(void) printf(",\n\t\t{object: %llx, offset: %llx, "
2962 		    "blksz: %x, count: %llx}",
2963 		    (u_longlong_t)rbp_buf[i].rbp_object,
2964 		    (u_longlong_t)rbp_buf[i].rbp_blkid,
2965 		    (uint_t)(redact_block_get_size(&rbp_buf[i])),
2966 		    (u_longlong_t)redact_block_get_count(&rbp_buf[i]));
2967 	}
2968 	dsl_redaction_list_rele(rl, FTAG);
2969 	kmem_free(rbp_buf, size);
2970 	(void) printf("]\n\n");
2971 	return (0);
2972 }
2973 
2974 static void
dump_bookmarks(objset_t * os,int verbosity)2975 dump_bookmarks(objset_t *os, int verbosity)
2976 {
2977 	zap_cursor_t zc;
2978 	zap_attribute_t *attrp;
2979 	dsl_dataset_t *ds = dmu_objset_ds(os);
2980 	dsl_pool_t *dp = spa_get_dsl(os->os_spa);
2981 	objset_t *mos = os->os_spa->spa_meta_objset;
2982 	if (verbosity < 4)
2983 		return;
2984 	attrp = zap_attribute_alloc();
2985 	dsl_pool_config_enter(dp, FTAG);
2986 
2987 	for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj);
2988 	    zap_cursor_retrieve(&zc, attrp) == 0;
2989 	    zap_cursor_advance(&zc)) {
2990 		char osname[ZFS_MAX_DATASET_NAME_LEN];
2991 		char buf[ZFS_MAX_DATASET_NAME_LEN];
2992 		int len;
2993 		dmu_objset_name(os, osname);
2994 		len = snprintf(buf, sizeof (buf), "%s#%s", osname,
2995 		    attrp->za_name);
2996 		VERIFY3S(len, <, ZFS_MAX_DATASET_NAME_LEN);
2997 		(void) dump_bookmark(dp, buf, verbosity >= 5, verbosity >= 6);
2998 	}
2999 	zap_cursor_fini(&zc);
3000 	dsl_pool_config_exit(dp, FTAG);
3001 	zap_attribute_free(attrp);
3002 }
3003 
3004 static void
bpobj_count_refd(bpobj_t * bpo)3005 bpobj_count_refd(bpobj_t *bpo)
3006 {
3007 	mos_obj_refd(bpo->bpo_object);
3008 
3009 	if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
3010 		mos_obj_refd(bpo->bpo_phys->bpo_subobjs);
3011 		for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
3012 			uint64_t subobj;
3013 			bpobj_t subbpo;
3014 			int error;
3015 			VERIFY0(dmu_read(bpo->bpo_os,
3016 			    bpo->bpo_phys->bpo_subobjs,
3017 			    i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3018 			error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3019 			if (error != 0) {
3020 				(void) printf("ERROR %u while trying to open "
3021 				    "subobj id %llu\n",
3022 				    error, (u_longlong_t)subobj);
3023 				continue;
3024 			}
3025 			bpobj_count_refd(&subbpo);
3026 			bpobj_close(&subbpo);
3027 		}
3028 	}
3029 }
3030 
3031 static int
dsl_deadlist_entry_count_refd(void * arg,dsl_deadlist_entry_t * dle)3032 dsl_deadlist_entry_count_refd(void *arg, dsl_deadlist_entry_t *dle)
3033 {
3034 	spa_t *spa = arg;
3035 	uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3036 	if (dle->dle_bpobj.bpo_object != empty_bpobj)
3037 		bpobj_count_refd(&dle->dle_bpobj);
3038 	return (0);
3039 }
3040 
3041 static int
dsl_deadlist_entry_dump(void * arg,dsl_deadlist_entry_t * dle)3042 dsl_deadlist_entry_dump(void *arg, dsl_deadlist_entry_t *dle)
3043 {
3044 	ASSERT(arg == NULL);
3045 	if (dump_opt['d'] >= 5) {
3046 		char buf[128];
3047 		(void) snprintf(buf, sizeof (buf),
3048 		    "mintxg %llu -> obj %llu",
3049 		    (longlong_t)dle->dle_mintxg,
3050 		    (longlong_t)dle->dle_bpobj.bpo_object);
3051 
3052 		dump_full_bpobj(&dle->dle_bpobj, buf, 0);
3053 	} else {
3054 		(void) printf("mintxg %llu -> obj %llu\n",
3055 		    (longlong_t)dle->dle_mintxg,
3056 		    (longlong_t)dle->dle_bpobj.bpo_object);
3057 	}
3058 	return (0);
3059 }
3060 
3061 static void
dump_blkptr_list(dsl_deadlist_t * dl,const char * name)3062 dump_blkptr_list(dsl_deadlist_t *dl, const char *name)
3063 {
3064 	char bytes[32];
3065 	char comp[32];
3066 	char uncomp[32];
3067 	char entries[32];
3068 	spa_t *spa = dmu_objset_spa(dl->dl_os);
3069 	uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3070 
3071 	if (dl->dl_oldfmt) {
3072 		if (dl->dl_bpobj.bpo_object != empty_bpobj)
3073 			bpobj_count_refd(&dl->dl_bpobj);
3074 	} else {
3075 		mos_obj_refd(dl->dl_object);
3076 		dsl_deadlist_iterate(dl, dsl_deadlist_entry_count_refd, spa);
3077 	}
3078 
3079 	/* make sure nicenum has enough space */
3080 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3081 	_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
3082 	_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
3083 	_Static_assert(sizeof (entries) >= NN_NUMBUF_SZ, "entries truncated");
3084 
3085 	if (dump_opt['d'] < 3)
3086 		return;
3087 
3088 	if (dl->dl_oldfmt) {
3089 		dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0);
3090 		return;
3091 	}
3092 
3093 	zdb_nicenum(dl->dl_phys->dl_used, bytes, sizeof (bytes));
3094 	zdb_nicenum(dl->dl_phys->dl_comp, comp, sizeof (comp));
3095 	zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp, sizeof (uncomp));
3096 	zdb_nicenum(avl_numnodes(&dl->dl_tree), entries, sizeof (entries));
3097 	(void) printf("\n    %s: %s (%s/%s comp), %s entries\n",
3098 	    name, bytes, comp, uncomp, entries);
3099 
3100 	if (dump_opt['d'] < 4)
3101 		return;
3102 
3103 	(void) putchar('\n');
3104 
3105 	dsl_deadlist_iterate(dl, dsl_deadlist_entry_dump, NULL);
3106 }
3107 
3108 static int
verify_dd_livelist(objset_t * os)3109 verify_dd_livelist(objset_t *os)
3110 {
3111 	uint64_t ll_used, used, ll_comp, comp, ll_uncomp, uncomp;
3112 	dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3113 	dsl_dir_t  *dd = os->os_dsl_dataset->ds_dir;
3114 
3115 	ASSERT(!dmu_objset_is_snapshot(os));
3116 	if (!dsl_deadlist_is_open(&dd->dd_livelist))
3117 		return (0);
3118 
3119 	/* Iterate through the livelist to check for duplicates */
3120 	dsl_deadlist_iterate(&dd->dd_livelist, sublivelist_verify_lightweight,
3121 	    NULL);
3122 
3123 	dsl_pool_config_enter(dp, FTAG);
3124 	dsl_deadlist_space(&dd->dd_livelist, &ll_used,
3125 	    &ll_comp, &ll_uncomp);
3126 
3127 	dsl_dataset_t *origin_ds;
3128 	ASSERT(dsl_pool_config_held(dp));
3129 	VERIFY0(dsl_dataset_hold_obj(dp,
3130 	    dsl_dir_phys(dd)->dd_origin_obj, FTAG, &origin_ds));
3131 	VERIFY0(dsl_dataset_space_written(origin_ds, os->os_dsl_dataset,
3132 	    &used, &comp, &uncomp));
3133 	dsl_dataset_rele(origin_ds, FTAG);
3134 	dsl_pool_config_exit(dp, FTAG);
3135 	/*
3136 	 *  It's possible that the dataset's uncomp space is larger than the
3137 	 *  livelist's because livelists do not track embedded block pointers
3138 	 */
3139 	if (used != ll_used || comp != ll_comp || uncomp < ll_uncomp) {
3140 		char nice_used[32], nice_comp[32], nice_uncomp[32];
3141 		(void) printf("Discrepancy in space accounting:\n");
3142 		zdb_nicenum(used, nice_used, sizeof (nice_used));
3143 		zdb_nicenum(comp, nice_comp, sizeof (nice_comp));
3144 		zdb_nicenum(uncomp, nice_uncomp, sizeof (nice_uncomp));
3145 		(void) printf("dir: used %s, comp %s, uncomp %s\n",
3146 		    nice_used, nice_comp, nice_uncomp);
3147 		zdb_nicenum(ll_used, nice_used, sizeof (nice_used));
3148 		zdb_nicenum(ll_comp, nice_comp, sizeof (nice_comp));
3149 		zdb_nicenum(ll_uncomp, nice_uncomp, sizeof (nice_uncomp));
3150 		(void) printf("livelist: used %s, comp %s, uncomp %s\n",
3151 		    nice_used, nice_comp, nice_uncomp);
3152 		return (1);
3153 	}
3154 	return (0);
3155 }
3156 
3157 static char *key_material = NULL;
3158 
3159 static boolean_t
zdb_derive_key(dsl_dir_t * dd,uint8_t * key_out)3160 zdb_derive_key(dsl_dir_t *dd, uint8_t *key_out)
3161 {
3162 	uint64_t keyformat, salt, iters;
3163 	int i;
3164 	unsigned char c;
3165 
3166 	VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3167 	    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), sizeof (uint64_t),
3168 	    1, &keyformat));
3169 
3170 	switch (keyformat) {
3171 	case ZFS_KEYFORMAT_HEX:
3172 		for (i = 0; i < WRAPPING_KEY_LEN * 2; i += 2) {
3173 			if (!isxdigit(key_material[i]) ||
3174 			    !isxdigit(key_material[i+1]))
3175 				return (B_FALSE);
3176 			if (sscanf(&key_material[i], "%02hhx", &c) != 1)
3177 				return (B_FALSE);
3178 			key_out[i / 2] = c;
3179 		}
3180 		break;
3181 
3182 	case ZFS_KEYFORMAT_PASSPHRASE:
3183 		VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3184 		    dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT),
3185 		    sizeof (uint64_t), 1, &salt));
3186 		VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3187 		    dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS),
3188 		    sizeof (uint64_t), 1, &iters));
3189 
3190 		if (PKCS5_PBKDF2_HMAC_SHA1(key_material, strlen(key_material),
3191 		    ((uint8_t *)&salt), sizeof (uint64_t), iters,
3192 		    WRAPPING_KEY_LEN, key_out) != 1)
3193 			return (B_FALSE);
3194 
3195 		break;
3196 
3197 	default:
3198 		fatal("no support for key format %u\n",
3199 		    (unsigned int) keyformat);
3200 	}
3201 
3202 	return (B_TRUE);
3203 }
3204 
3205 static char encroot[ZFS_MAX_DATASET_NAME_LEN];
3206 static boolean_t key_loaded = B_FALSE;
3207 
3208 static void
zdb_load_key(objset_t * os)3209 zdb_load_key(objset_t *os)
3210 {
3211 	dsl_pool_t *dp;
3212 	dsl_dir_t *dd, *rdd;
3213 	uint8_t key[WRAPPING_KEY_LEN];
3214 	uint64_t rddobj;
3215 	int err;
3216 
3217 	dp = spa_get_dsl(os->os_spa);
3218 	dd = os->os_dsl_dataset->ds_dir;
3219 
3220 	dsl_pool_config_enter(dp, FTAG);
3221 	VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3222 	    DSL_CRYPTO_KEY_ROOT_DDOBJ, sizeof (uint64_t), 1, &rddobj));
3223 	VERIFY0(dsl_dir_hold_obj(dd->dd_pool, rddobj, NULL, FTAG, &rdd));
3224 	dsl_dir_name(rdd, encroot);
3225 	dsl_dir_rele(rdd, FTAG);
3226 
3227 	if (!zdb_derive_key(dd, key))
3228 		fatal("couldn't derive encryption key");
3229 
3230 	dsl_pool_config_exit(dp, FTAG);
3231 
3232 	ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_UNAVAILABLE);
3233 
3234 	dsl_crypto_params_t *dcp;
3235 	nvlist_t *crypto_args;
3236 
3237 	crypto_args = fnvlist_alloc();
3238 	fnvlist_add_uint8_array(crypto_args, "wkeydata",
3239 	    (uint8_t *)key, WRAPPING_KEY_LEN);
3240 	VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE,
3241 	    NULL, crypto_args, &dcp));
3242 	err = spa_keystore_load_wkey(encroot, dcp, B_FALSE);
3243 
3244 	dsl_crypto_params_free(dcp, (err != 0));
3245 	fnvlist_free(crypto_args);
3246 
3247 	if (err != 0)
3248 		fatal(
3249 		    "couldn't load encryption key for %s: %s",
3250 		    encroot, err == ZFS_ERR_CRYPTO_NOTSUP ?
3251 		    "crypto params not supported" : strerror(err));
3252 
3253 	ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_AVAILABLE);
3254 
3255 	printf("Unlocked encryption root: %s\n", encroot);
3256 	key_loaded = B_TRUE;
3257 }
3258 
3259 static void
zdb_unload_key(void)3260 zdb_unload_key(void)
3261 {
3262 	if (!key_loaded)
3263 		return;
3264 
3265 	VERIFY0(spa_keystore_unload_wkey(encroot));
3266 	key_loaded = B_FALSE;
3267 }
3268 
3269 static avl_tree_t idx_tree;
3270 static avl_tree_t domain_tree;
3271 static boolean_t fuid_table_loaded;
3272 static objset_t *sa_os = NULL;
3273 static sa_attr_type_t *sa_attr_table = NULL;
3274 
3275 static int
open_objset(const char * path,const void * tag,objset_t ** osp)3276 open_objset(const char *path, const void *tag, objset_t **osp)
3277 {
3278 	int err;
3279 	uint64_t sa_attrs = 0;
3280 	uint64_t version = 0;
3281 
3282 	VERIFY3P(sa_os, ==, NULL);
3283 
3284 	/*
3285 	 * We can't own an objset if it's redacted.  Therefore, we do this
3286 	 * dance: hold the objset, then acquire a long hold on its dataset, then
3287 	 * release the pool (which is held as part of holding the objset).
3288 	 */
3289 
3290 	if (dump_opt['K']) {
3291 		/* decryption requested, try to load keys */
3292 		err = dmu_objset_hold(path, tag, osp);
3293 		if (err != 0) {
3294 			(void) fprintf(stderr, "failed to hold dataset "
3295 			    "'%s': %s\n",
3296 			    path, strerror(err));
3297 			return (err);
3298 		}
3299 		dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3300 		dsl_pool_rele(dmu_objset_pool(*osp), tag);
3301 
3302 		/* succeeds or dies */
3303 		zdb_load_key(*osp);
3304 
3305 		/* release it all */
3306 		dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3307 		dsl_dataset_rele(dmu_objset_ds(*osp), tag);
3308 	}
3309 
3310 	int ds_hold_flags = key_loaded ? DS_HOLD_FLAG_DECRYPT : 0;
3311 
3312 	err = dmu_objset_hold_flags(path, ds_hold_flags, tag, osp);
3313 	if (err != 0) {
3314 		(void) fprintf(stderr, "failed to hold dataset '%s': %s\n",
3315 		    path, strerror(err));
3316 		return (err);
3317 	}
3318 	dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3319 	dsl_pool_rele(dmu_objset_pool(*osp), tag);
3320 
3321 	if (dmu_objset_type(*osp) == DMU_OST_ZFS &&
3322 	    (key_loaded || !(*osp)->os_encrypted)) {
3323 		(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR,
3324 		    8, 1, &version);
3325 		if (version >= ZPL_VERSION_SA) {
3326 			(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS,
3327 			    8, 1, &sa_attrs);
3328 		}
3329 		err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END,
3330 		    &sa_attr_table);
3331 		if (err != 0) {
3332 			(void) fprintf(stderr, "sa_setup failed: %s\n",
3333 			    strerror(err));
3334 			dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3335 			dsl_dataset_rele_flags(dmu_objset_ds(*osp),
3336 			    ds_hold_flags, tag);
3337 			*osp = NULL;
3338 		}
3339 	}
3340 	sa_os = *osp;
3341 
3342 	return (err);
3343 }
3344 
3345 static void
close_objset(objset_t * os,const void * tag)3346 close_objset(objset_t *os, const void *tag)
3347 {
3348 	VERIFY3P(os, ==, sa_os);
3349 	if (os->os_sa != NULL)
3350 		sa_tear_down(os);
3351 	dsl_dataset_long_rele(dmu_objset_ds(os), tag);
3352 	dsl_dataset_rele_flags(dmu_objset_ds(os),
3353 	    key_loaded ? DS_HOLD_FLAG_DECRYPT : 0, tag);
3354 	sa_attr_table = NULL;
3355 	sa_os = NULL;
3356 
3357 	zdb_unload_key();
3358 }
3359 
3360 static void
fuid_table_destroy(void)3361 fuid_table_destroy(void)
3362 {
3363 	if (fuid_table_loaded) {
3364 		zfs_fuid_table_destroy(&idx_tree, &domain_tree);
3365 		fuid_table_loaded = B_FALSE;
3366 	}
3367 }
3368 
3369 /*
3370  * Clean up DDT internal state. ddt_lookup() adds entries to ddt_tree, which on
3371  * a live pool are normally cleaned up during ddt_sync(). We can't do that (and
3372  * wouldn't want to anyway), but if we don't clean up the presence of stuff on
3373  * ddt_tree will trip asserts in ddt_table_free(). So, we clean up ourselves.
3374  *
3375  * Note that this is not a particularly efficient way to do this, but
3376  * ddt_remove() is the only public method that can do the work we need, and it
3377  * requires the right locks and etc to do the job. This is only ever called
3378  * during zdb shutdown so efficiency is not especially important.
3379  */
3380 static void
zdb_ddt_cleanup(spa_t * spa)3381 zdb_ddt_cleanup(spa_t *spa)
3382 {
3383 	for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
3384 		ddt_t *ddt = spa->spa_ddt[c];
3385 		if (!ddt)
3386 			continue;
3387 
3388 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3389 		ddt_enter(ddt);
3390 		ddt_entry_t *dde = avl_first(&ddt->ddt_tree), *next;
3391 		while (dde) {
3392 			next = AVL_NEXT(&ddt->ddt_tree, dde);
3393 			dde->dde_io = NULL;
3394 			ddt_remove(ddt, dde);
3395 			dde = next;
3396 		}
3397 		ddt_exit(ddt);
3398 		spa_config_exit(spa, SCL_CONFIG, FTAG);
3399 	}
3400 }
3401 
3402 static void
zdb_exit(int reason)3403 zdb_exit(int reason)
3404 {
3405 	if (spa != NULL)
3406 		zdb_ddt_cleanup(spa);
3407 
3408 	if (os != NULL) {
3409 		close_objset(os, FTAG);
3410 	} else if (spa != NULL) {
3411 		spa_close(spa, FTAG);
3412 	}
3413 
3414 	fuid_table_destroy();
3415 
3416 	if (kernel_init_done)
3417 		kernel_fini();
3418 
3419 	exit(reason);
3420 }
3421 
3422 /*
3423  * print uid or gid information.
3424  * For normal POSIX id just the id is printed in decimal format.
3425  * For CIFS files with FUID the fuid is printed in hex followed by
3426  * the domain-rid string.
3427  */
3428 static void
print_idstr(uint64_t id,const char * id_type)3429 print_idstr(uint64_t id, const char *id_type)
3430 {
3431 	if (FUID_INDEX(id)) {
3432 		const char *domain =
3433 		    zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id));
3434 		(void) printf("\t%s     %llx [%s-%d]\n", id_type,
3435 		    (u_longlong_t)id, domain, (int)FUID_RID(id));
3436 	} else {
3437 		(void) printf("\t%s     %llu\n", id_type, (u_longlong_t)id);
3438 	}
3439 
3440 }
3441 
3442 static void
dump_uidgid(objset_t * os,uint64_t uid,uint64_t gid)3443 dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid)
3444 {
3445 	uint32_t uid_idx, gid_idx;
3446 
3447 	uid_idx = FUID_INDEX(uid);
3448 	gid_idx = FUID_INDEX(gid);
3449 
3450 	/* Load domain table, if not already loaded */
3451 	if (!fuid_table_loaded && (uid_idx || gid_idx)) {
3452 		uint64_t fuid_obj;
3453 
3454 		/* first find the fuid object.  It lives in the master node */
3455 		VERIFY(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES,
3456 		    8, 1, &fuid_obj) == 0);
3457 		zfs_fuid_avl_tree_create(&idx_tree, &domain_tree);
3458 		(void) zfs_fuid_table_load(os, fuid_obj,
3459 		    &idx_tree, &domain_tree);
3460 		fuid_table_loaded = B_TRUE;
3461 	}
3462 
3463 	print_idstr(uid, "uid");
3464 	print_idstr(gid, "gid");
3465 }
3466 
3467 static void
dump_znode_sa_xattr(sa_handle_t * hdl)3468 dump_znode_sa_xattr(sa_handle_t *hdl)
3469 {
3470 	nvlist_t *sa_xattr;
3471 	nvpair_t *elem = NULL;
3472 	int sa_xattr_size = 0;
3473 	int sa_xattr_entries = 0;
3474 	int error;
3475 	char *sa_xattr_packed;
3476 
3477 	error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size);
3478 	if (error || sa_xattr_size == 0)
3479 		return;
3480 
3481 	sa_xattr_packed = malloc(sa_xattr_size);
3482 	if (sa_xattr_packed == NULL)
3483 		return;
3484 
3485 	error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR],
3486 	    sa_xattr_packed, sa_xattr_size);
3487 	if (error) {
3488 		free(sa_xattr_packed);
3489 		return;
3490 	}
3491 
3492 	error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0);
3493 	if (error) {
3494 		free(sa_xattr_packed);
3495 		return;
3496 	}
3497 
3498 	while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL)
3499 		sa_xattr_entries++;
3500 
3501 	(void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3502 	    sa_xattr_size, sa_xattr_entries);
3503 	while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) {
3504 		boolean_t can_print = !dump_opt['P'];
3505 		uchar_t *value;
3506 		uint_t cnt, idx;
3507 
3508 		(void) printf("\t\t%s = ", nvpair_name(elem));
3509 		nvpair_value_byte_array(elem, &value, &cnt);
3510 
3511 		for (idx = 0; idx < cnt; ++idx) {
3512 			if (!isprint(value[idx])) {
3513 				can_print = B_FALSE;
3514 				break;
3515 			}
3516 		}
3517 
3518 		for (idx = 0; idx < cnt; ++idx) {
3519 			if (can_print)
3520 				(void) putchar(value[idx]);
3521 			else
3522 				(void) printf("\\%3.3o", value[idx]);
3523 		}
3524 		(void) putchar('\n');
3525 	}
3526 
3527 	nvlist_free(sa_xattr);
3528 	free(sa_xattr_packed);
3529 }
3530 
3531 static void
dump_znode_symlink(sa_handle_t * hdl)3532 dump_znode_symlink(sa_handle_t *hdl)
3533 {
3534 	int sa_symlink_size = 0;
3535 	char linktarget[MAXPATHLEN];
3536 	int error;
3537 
3538 	error = sa_size(hdl, sa_attr_table[ZPL_SYMLINK], &sa_symlink_size);
3539 	if (error || sa_symlink_size == 0) {
3540 		return;
3541 	}
3542 	if (sa_symlink_size >= sizeof (linktarget)) {
3543 		(void) printf("symlink size %d is too large\n",
3544 		    sa_symlink_size);
3545 		return;
3546 	}
3547 	linktarget[sa_symlink_size] = '\0';
3548 	if (sa_lookup(hdl, sa_attr_table[ZPL_SYMLINK],
3549 	    &linktarget, sa_symlink_size) == 0)
3550 		(void) printf("\ttarget	%s\n", linktarget);
3551 }
3552 
3553 static void
dump_znode(objset_t * os,uint64_t object,void * data,size_t size)3554 dump_znode(objset_t *os, uint64_t object, void *data, size_t size)
3555 {
3556 	(void) data, (void) size;
3557 	char path[MAXPATHLEN * 2];	/* allow for xattr and failure prefix */
3558 	sa_handle_t *hdl;
3559 	uint64_t xattr, rdev, gen;
3560 	uint64_t uid, gid, mode, fsize, parent, links;
3561 	uint64_t pflags;
3562 	uint64_t acctm[2], modtm[2], chgtm[2], crtm[2];
3563 	time_t z_crtime, z_atime, z_mtime, z_ctime;
3564 	sa_bulk_attr_t bulk[12];
3565 	int idx = 0;
3566 	int error;
3567 
3568 	VERIFY3P(os, ==, sa_os);
3569 	if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) {
3570 		(void) printf("Failed to get handle for SA znode\n");
3571 		return;
3572 	}
3573 
3574 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8);
3575 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8);
3576 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL,
3577 	    &links, 8);
3578 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8);
3579 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL,
3580 	    &mode, 8);
3581 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT],
3582 	    NULL, &parent, 8);
3583 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL,
3584 	    &fsize, 8);
3585 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL,
3586 	    acctm, 16);
3587 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL,
3588 	    modtm, 16);
3589 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL,
3590 	    crtm, 16);
3591 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL,
3592 	    chgtm, 16);
3593 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL,
3594 	    &pflags, 8);
3595 
3596 	if (sa_bulk_lookup(hdl, bulk, idx)) {
3597 		(void) sa_handle_destroy(hdl);
3598 		return;
3599 	}
3600 
3601 	z_crtime = (time_t)crtm[0];
3602 	z_atime = (time_t)acctm[0];
3603 	z_mtime = (time_t)modtm[0];
3604 	z_ctime = (time_t)chgtm[0];
3605 
3606 	if (dump_opt['d'] > 4) {
3607 		error = zfs_obj_to_path(os, object, path, sizeof (path));
3608 		if (error == ESTALE) {
3609 			(void) snprintf(path, sizeof (path), "on delete queue");
3610 		} else if (error != 0) {
3611 			leaked_objects++;
3612 			(void) snprintf(path, sizeof (path),
3613 			    "path not found, possibly leaked");
3614 		}
3615 		(void) printf("\tpath	%s\n", path);
3616 	}
3617 
3618 	if (S_ISLNK(mode))
3619 		dump_znode_symlink(hdl);
3620 	dump_uidgid(os, uid, gid);
3621 	(void) printf("\tatime	%s", ctime(&z_atime));
3622 	(void) printf("\tmtime	%s", ctime(&z_mtime));
3623 	(void) printf("\tctime	%s", ctime(&z_ctime));
3624 	(void) printf("\tcrtime	%s", ctime(&z_crtime));
3625 	(void) printf("\tgen	%llu\n", (u_longlong_t)gen);
3626 	(void) printf("\tmode	%llo\n", (u_longlong_t)mode);
3627 	(void) printf("\tsize	%llu\n", (u_longlong_t)fsize);
3628 	(void) printf("\tparent	%llu\n", (u_longlong_t)parent);
3629 	(void) printf("\tlinks	%llu\n", (u_longlong_t)links);
3630 	(void) printf("\tpflags	%llx\n", (u_longlong_t)pflags);
3631 	if (dmu_objset_projectquota_enabled(os) && (pflags & ZFS_PROJID)) {
3632 		uint64_t projid;
3633 
3634 		if (sa_lookup(hdl, sa_attr_table[ZPL_PROJID], &projid,
3635 		    sizeof (uint64_t)) == 0)
3636 			(void) printf("\tprojid	%llu\n", (u_longlong_t)projid);
3637 	}
3638 	if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr,
3639 	    sizeof (uint64_t)) == 0)
3640 		(void) printf("\txattr	%llu\n", (u_longlong_t)xattr);
3641 	if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev,
3642 	    sizeof (uint64_t)) == 0)
3643 		(void) printf("\trdev	0x%016llx\n", (u_longlong_t)rdev);
3644 	dump_znode_sa_xattr(hdl);
3645 	sa_handle_destroy(hdl);
3646 }
3647 
3648 static void
dump_acl(objset_t * os,uint64_t object,void * data,size_t size)3649 dump_acl(objset_t *os, uint64_t object, void *data, size_t size)
3650 {
3651 	(void) os, (void) object, (void) data, (void) size;
3652 }
3653 
3654 static void
dump_dmu_objset(objset_t * os,uint64_t object,void * data,size_t size)3655 dump_dmu_objset(objset_t *os, uint64_t object, void *data, size_t size)
3656 {
3657 	(void) os, (void) object, (void) data, (void) size;
3658 }
3659 
3660 static object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = {
3661 	dump_none,		/* unallocated			*/
3662 	dump_zap,		/* object directory		*/
3663 	dump_uint64,		/* object array			*/
3664 	dump_none,		/* packed nvlist		*/
3665 	dump_packed_nvlist,	/* packed nvlist size		*/
3666 	dump_none,		/* bpobj			*/
3667 	dump_bpobj,		/* bpobj header			*/
3668 	dump_none,		/* SPA space map header		*/
3669 	dump_none,		/* SPA space map		*/
3670 	dump_none,		/* ZIL intent log		*/
3671 	dump_dnode,		/* DMU dnode			*/
3672 	dump_dmu_objset,	/* DMU objset			*/
3673 	dump_dsl_dir,		/* DSL directory		*/
3674 	dump_zap,		/* DSL directory child map	*/
3675 	dump_zap,		/* DSL dataset snap map		*/
3676 	dump_zap,		/* DSL props			*/
3677 	dump_dsl_dataset,	/* DSL dataset			*/
3678 	dump_znode,		/* ZFS znode			*/
3679 	dump_acl,		/* ZFS V0 ACL			*/
3680 	dump_uint8,		/* ZFS plain file		*/
3681 	dump_zpldir,		/* ZFS directory		*/
3682 	dump_zap,		/* ZFS master node		*/
3683 	dump_zap,		/* ZFS delete queue		*/
3684 	dump_uint8,		/* zvol object			*/
3685 	dump_zap,		/* zvol prop			*/
3686 	dump_uint8,		/* other uint8[]		*/
3687 	dump_uint64,		/* other uint64[]		*/
3688 	dump_zap,		/* other ZAP			*/
3689 	dump_zap,		/* persistent error log		*/
3690 	dump_uint8,		/* SPA history			*/
3691 	dump_history_offsets,	/* SPA history offsets		*/
3692 	dump_zap,		/* Pool properties		*/
3693 	dump_zap,		/* DSL permissions		*/
3694 	dump_acl,		/* ZFS ACL			*/
3695 	dump_uint8,		/* ZFS SYSACL			*/
3696 	dump_none,		/* FUID nvlist			*/
3697 	dump_packed_nvlist,	/* FUID nvlist size		*/
3698 	dump_zap,		/* DSL dataset next clones	*/
3699 	dump_zap,		/* DSL scrub queue		*/
3700 	dump_zap,		/* ZFS user/group/project used	*/
3701 	dump_zap,		/* ZFS user/group/project quota	*/
3702 	dump_zap,		/* snapshot refcount tags	*/
3703 	dump_ddt_zap,		/* DDT ZAP object		*/
3704 	dump_zap,		/* DDT statistics		*/
3705 	dump_znode,		/* SA object			*/
3706 	dump_zap,		/* SA Master Node		*/
3707 	dump_sa_attrs,		/* SA attribute registration	*/
3708 	dump_sa_layouts,	/* SA attribute layouts		*/
3709 	dump_zap,		/* DSL scrub translations	*/
3710 	dump_none,		/* fake dedup BP		*/
3711 	dump_zap,		/* deadlist			*/
3712 	dump_none,		/* deadlist hdr			*/
3713 	dump_zap,		/* dsl clones			*/
3714 	dump_bpobj_subobjs,	/* bpobj subobjs		*/
3715 	dump_unknown,		/* Unknown type, must be last	*/
3716 };
3717 
3718 static boolean_t
match_object_type(dmu_object_type_t obj_type,uint64_t flags)3719 match_object_type(dmu_object_type_t obj_type, uint64_t flags)
3720 {
3721 	boolean_t match = B_TRUE;
3722 
3723 	switch (obj_type) {
3724 	case DMU_OT_DIRECTORY_CONTENTS:
3725 		if (!(flags & ZOR_FLAG_DIRECTORY))
3726 			match = B_FALSE;
3727 		break;
3728 	case DMU_OT_PLAIN_FILE_CONTENTS:
3729 		if (!(flags & ZOR_FLAG_PLAIN_FILE))
3730 			match = B_FALSE;
3731 		break;
3732 	case DMU_OT_SPACE_MAP:
3733 		if (!(flags & ZOR_FLAG_SPACE_MAP))
3734 			match = B_FALSE;
3735 		break;
3736 	default:
3737 		if (strcmp(zdb_ot_name(obj_type), "zap") == 0) {
3738 			if (!(flags & ZOR_FLAG_ZAP))
3739 				match = B_FALSE;
3740 			break;
3741 		}
3742 
3743 		/*
3744 		 * If all bits except some of the supported flags are
3745 		 * set, the user combined the all-types flag (A) with
3746 		 * a negated flag to exclude some types (e.g. A-f to
3747 		 * show all object types except plain files).
3748 		 */
3749 		if ((flags | ZOR_SUPPORTED_FLAGS) != ZOR_FLAG_ALL_TYPES)
3750 			match = B_FALSE;
3751 
3752 		break;
3753 	}
3754 
3755 	return (match);
3756 }
3757 
3758 static void
dump_object(objset_t * os,uint64_t object,int verbosity,boolean_t * print_header,uint64_t * dnode_slots_used,uint64_t flags)3759 dump_object(objset_t *os, uint64_t object, int verbosity,
3760     boolean_t *print_header, uint64_t *dnode_slots_used, uint64_t flags)
3761 {
3762 	dmu_buf_t *db = NULL;
3763 	dmu_object_info_t doi;
3764 	dnode_t *dn;
3765 	boolean_t dnode_held = B_FALSE;
3766 	void *bonus = NULL;
3767 	size_t bsize = 0;
3768 	char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
3769 	char bonus_size[32];
3770 	char aux[50];
3771 	int error;
3772 
3773 	/* make sure nicenum has enough space */
3774 	_Static_assert(sizeof (iblk) >= NN_NUMBUF_SZ, "iblk truncated");
3775 	_Static_assert(sizeof (dblk) >= NN_NUMBUF_SZ, "dblk truncated");
3776 	_Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, "lsize truncated");
3777 	_Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, "asize truncated");
3778 	_Static_assert(sizeof (bonus_size) >= NN_NUMBUF_SZ,
3779 	    "bonus_size truncated");
3780 
3781 	if (*print_header) {
3782 		(void) printf("\n%10s  %3s  %5s  %5s  %5s  %6s  %5s  %6s  %s\n",
3783 		    "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3784 		    "lsize", "%full", "type");
3785 		*print_header = 0;
3786 	}
3787 
3788 	if (object == 0) {
3789 		dn = DMU_META_DNODE(os);
3790 		dmu_object_info_from_dnode(dn, &doi);
3791 	} else {
3792 		/*
3793 		 * Encrypted datasets will have sensitive bonus buffers
3794 		 * encrypted. Therefore we cannot hold the bonus buffer and
3795 		 * must hold the dnode itself instead.
3796 		 */
3797 		error = dmu_object_info(os, object, &doi);
3798 		if (error)
3799 			fatal("dmu_object_info() failed, errno %u", error);
3800 
3801 		if (!key_loaded && os->os_encrypted &&
3802 		    DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) {
3803 			error = dnode_hold(os, object, FTAG, &dn);
3804 			if (error)
3805 				fatal("dnode_hold() failed, errno %u", error);
3806 			dnode_held = B_TRUE;
3807 		} else {
3808 			error = dmu_bonus_hold(os, object, FTAG, &db);
3809 			if (error)
3810 				fatal("dmu_bonus_hold(%llu) failed, errno %u",
3811 				    object, error);
3812 			bonus = db->db_data;
3813 			bsize = db->db_size;
3814 			dn = DB_DNODE((dmu_buf_impl_t *)db);
3815 		}
3816 	}
3817 
3818 	/*
3819 	 * Default to showing all object types if no flags were specified.
3820 	 */
3821 	if (flags != 0 && flags != ZOR_FLAG_ALL_TYPES &&
3822 	    !match_object_type(doi.doi_type, flags))
3823 		goto out;
3824 
3825 	if (dnode_slots_used)
3826 		*dnode_slots_used = doi.doi_dnodesize / DNODE_MIN_SIZE;
3827 
3828 	zdb_nicenum(doi.doi_metadata_block_size, iblk, sizeof (iblk));
3829 	zdb_nicenum(doi.doi_data_block_size, dblk, sizeof (dblk));
3830 	zdb_nicenum(doi.doi_max_offset, lsize, sizeof (lsize));
3831 	zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize, sizeof (asize));
3832 	zdb_nicenum(doi.doi_bonus_size, bonus_size, sizeof (bonus_size));
3833 	zdb_nicenum(doi.doi_dnodesize, dnsize, sizeof (dnsize));
3834 	(void) snprintf(fill, sizeof (fill), "%6.2f", 100.0 *
3835 	    doi.doi_fill_count * doi.doi_data_block_size / (object == 0 ?
3836 	    DNODES_PER_BLOCK : 1) / doi.doi_max_offset);
3837 
3838 	aux[0] = '\0';
3839 
3840 	if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) {
3841 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3842 		    " (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum));
3843 	}
3844 
3845 	if (doi.doi_compress == ZIO_COMPRESS_INHERIT &&
3846 	    ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) {
3847 		const char *compname = NULL;
3848 		if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION,
3849 		    ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel),
3850 		    &compname) == 0) {
3851 			(void) snprintf(aux + strlen(aux),
3852 			    sizeof (aux) - strlen(aux), " (Z=inherit=%s)",
3853 			    compname);
3854 		} else {
3855 			(void) snprintf(aux + strlen(aux),
3856 			    sizeof (aux) - strlen(aux),
3857 			    " (Z=inherit=%s-unknown)",
3858 			    ZDB_COMPRESS_NAME(os->os_compress));
3859 		}
3860 	} else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) {
3861 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3862 		    " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress));
3863 	} else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
3864 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3865 		    " (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress));
3866 	}
3867 
3868 	(void) printf("%10lld  %3u  %5s  %5s  %5s  %6s  %5s  %6s  %s%s\n",
3869 	    (u_longlong_t)object, doi.doi_indirection, iblk, dblk,
3870 	    asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux);
3871 
3872 	if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
3873 		(void) printf("%10s  %3s  %5s  %5s  %5s  %5s  %5s  %6s  %s\n",
3874 		    "", "", "", "", "", "", bonus_size, "bonus",
3875 		    zdb_ot_name(doi.doi_bonus_type));
3876 	}
3877 
3878 	if (verbosity >= 4) {
3879 		(void) printf("\tdnode flags: %s%s%s%s\n",
3880 		    (dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ?
3881 		    "USED_BYTES " : "",
3882 		    (dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ?
3883 		    "USERUSED_ACCOUNTED " : "",
3884 		    (dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ?
3885 		    "USEROBJUSED_ACCOUNTED " : "",
3886 		    (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ?
3887 		    "SPILL_BLKPTR" : "");
3888 		(void) printf("\tdnode maxblkid: %llu\n",
3889 		    (longlong_t)dn->dn_phys->dn_maxblkid);
3890 
3891 		if (!dnode_held) {
3892 			object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os,
3893 			    object, bonus, bsize);
3894 		} else {
3895 			(void) printf("\t\t(bonus encrypted)\n");
3896 		}
3897 
3898 		if (key_loaded ||
3899 		    (!os->os_encrypted || !DMU_OT_IS_ENCRYPTED(doi.doi_type))) {
3900 			object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object,
3901 			    NULL, 0);
3902 		} else {
3903 			(void) printf("\t\t(object encrypted)\n");
3904 		}
3905 
3906 		*print_header = B_TRUE;
3907 	}
3908 
3909 	if (verbosity >= 5) {
3910 		if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
3911 			char blkbuf[BP_SPRINTF_LEN];
3912 			snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
3913 			    DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE);
3914 			(void) printf("\nSpill block: %s\n", blkbuf);
3915 		}
3916 		dump_indirect(dn);
3917 	}
3918 
3919 	if (verbosity >= 5) {
3920 		/*
3921 		 * Report the list of segments that comprise the object.
3922 		 */
3923 		uint64_t start = 0;
3924 		uint64_t end;
3925 		uint64_t blkfill = 1;
3926 		int minlvl = 1;
3927 
3928 		if (dn->dn_type == DMU_OT_DNODE) {
3929 			minlvl = 0;
3930 			blkfill = DNODES_PER_BLOCK;
3931 		}
3932 
3933 		for (;;) {
3934 			char segsize[32];
3935 			/* make sure nicenum has enough space */
3936 			_Static_assert(sizeof (segsize) >= NN_NUMBUF_SZ,
3937 			    "segsize truncated");
3938 			error = dnode_next_offset(dn,
3939 			    0, &start, minlvl, blkfill, 0);
3940 			if (error)
3941 				break;
3942 			end = start;
3943 			error = dnode_next_offset(dn,
3944 			    DNODE_FIND_HOLE, &end, minlvl, blkfill, 0);
3945 			zdb_nicenum(end - start, segsize, sizeof (segsize));
3946 			(void) printf("\t\tsegment [%016llx, %016llx)"
3947 			    " size %5s\n", (u_longlong_t)start,
3948 			    (u_longlong_t)end, segsize);
3949 			if (error)
3950 				break;
3951 			start = end;
3952 		}
3953 	}
3954 
3955 out:
3956 	if (db != NULL)
3957 		dmu_buf_rele(db, FTAG);
3958 	if (dnode_held)
3959 		dnode_rele(dn, FTAG);
3960 }
3961 
3962 static void
count_dir_mos_objects(dsl_dir_t * dd)3963 count_dir_mos_objects(dsl_dir_t *dd)
3964 {
3965 	mos_obj_refd(dd->dd_object);
3966 	mos_obj_refd(dsl_dir_phys(dd)->dd_child_dir_zapobj);
3967 	mos_obj_refd(dsl_dir_phys(dd)->dd_deleg_zapobj);
3968 	mos_obj_refd(dsl_dir_phys(dd)->dd_props_zapobj);
3969 	mos_obj_refd(dsl_dir_phys(dd)->dd_clones);
3970 
3971 	/*
3972 	 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
3973 	 * Ignore the references after the first one.
3974 	 */
3975 	mos_obj_refd_multiple(dd->dd_crypto_obj);
3976 }
3977 
3978 static void
count_ds_mos_objects(dsl_dataset_t * ds)3979 count_ds_mos_objects(dsl_dataset_t *ds)
3980 {
3981 	mos_obj_refd(ds->ds_object);
3982 	mos_obj_refd(dsl_dataset_phys(ds)->ds_next_clones_obj);
3983 	mos_obj_refd(dsl_dataset_phys(ds)->ds_props_obj);
3984 	mos_obj_refd(dsl_dataset_phys(ds)->ds_userrefs_obj);
3985 	mos_obj_refd(dsl_dataset_phys(ds)->ds_snapnames_zapobj);
3986 	mos_obj_refd(ds->ds_bookmarks_obj);
3987 
3988 	if (!dsl_dataset_is_snapshot(ds)) {
3989 		count_dir_mos_objects(ds->ds_dir);
3990 	}
3991 }
3992 
3993 static const char *const objset_types[DMU_OST_NUMTYPES] = {
3994 	"NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
3995 
3996 /*
3997  * Parse a string denoting a range of object IDs of the form
3998  * <start>[:<end>[:flags]], and store the results in zor.
3999  * Return 0 on success. On error, return 1 and update the msg
4000  * pointer to point to a descriptive error message.
4001  */
4002 static int
parse_object_range(char * range,zopt_object_range_t * zor,const char ** msg)4003 parse_object_range(char *range, zopt_object_range_t *zor, const char **msg)
4004 {
4005 	uint64_t flags = 0;
4006 	char *p, *s, *dup, *flagstr, *tmp = NULL;
4007 	size_t len;
4008 	int i;
4009 	int rc = 0;
4010 
4011 	if (strchr(range, ':') == NULL) {
4012 		zor->zor_obj_start = strtoull(range, &p, 0);
4013 		if (*p != '\0') {
4014 			*msg = "Invalid characters in object ID";
4015 			rc = 1;
4016 		}
4017 		zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4018 		zor->zor_obj_end = zor->zor_obj_start;
4019 		return (rc);
4020 	}
4021 
4022 	if (strchr(range, ':') == range) {
4023 		*msg = "Invalid leading colon";
4024 		rc = 1;
4025 		return (rc);
4026 	}
4027 
4028 	len = strlen(range);
4029 	if (range[len - 1] == ':') {
4030 		*msg = "Invalid trailing colon";
4031 		rc = 1;
4032 		return (rc);
4033 	}
4034 
4035 	dup = strdup(range);
4036 	s = strtok_r(dup, ":", &tmp);
4037 	zor->zor_obj_start = strtoull(s, &p, 0);
4038 
4039 	if (*p != '\0') {
4040 		*msg = "Invalid characters in start object ID";
4041 		rc = 1;
4042 		goto out;
4043 	}
4044 
4045 	s = strtok_r(NULL, ":", &tmp);
4046 	zor->zor_obj_end = strtoull(s, &p, 0);
4047 
4048 	if (*p != '\0') {
4049 		*msg = "Invalid characters in end object ID";
4050 		rc = 1;
4051 		goto out;
4052 	}
4053 
4054 	if (zor->zor_obj_start > zor->zor_obj_end) {
4055 		*msg = "Start object ID may not exceed end object ID";
4056 		rc = 1;
4057 		goto out;
4058 	}
4059 
4060 	s = strtok_r(NULL, ":", &tmp);
4061 	if (s == NULL) {
4062 		zor->zor_flags = ZOR_FLAG_ALL_TYPES;
4063 		goto out;
4064 	} else if (strtok_r(NULL, ":", &tmp) != NULL) {
4065 		*msg = "Invalid colon-delimited field after flags";
4066 		rc = 1;
4067 		goto out;
4068 	}
4069 
4070 	flagstr = s;
4071 	for (i = 0; flagstr[i]; i++) {
4072 		int bit;
4073 		boolean_t negation = (flagstr[i] == '-');
4074 
4075 		if (negation) {
4076 			i++;
4077 			if (flagstr[i] == '\0') {
4078 				*msg = "Invalid trailing negation operator";
4079 				rc = 1;
4080 				goto out;
4081 			}
4082 		}
4083 		bit = flagbits[(uchar_t)flagstr[i]];
4084 		if (bit == 0) {
4085 			*msg = "Invalid flag";
4086 			rc = 1;
4087 			goto out;
4088 		}
4089 		if (negation)
4090 			flags &= ~bit;
4091 		else
4092 			flags |= bit;
4093 	}
4094 	zor->zor_flags = flags;
4095 
4096 	zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4097 	zor->zor_obj_end = ZDB_MAP_OBJECT_ID(zor->zor_obj_end);
4098 
4099 out:
4100 	free(dup);
4101 	return (rc);
4102 }
4103 
4104 static void
dump_objset(objset_t * os)4105 dump_objset(objset_t *os)
4106 {
4107 	dmu_objset_stats_t dds = { 0 };
4108 	uint64_t object, object_count;
4109 	uint64_t refdbytes, usedobjs, scratch;
4110 	char numbuf[32];
4111 	char blkbuf[BP_SPRINTF_LEN + 20];
4112 	char osname[ZFS_MAX_DATASET_NAME_LEN];
4113 	const char *type = "UNKNOWN";
4114 	int verbosity = dump_opt['d'];
4115 	boolean_t print_header;
4116 	unsigned i;
4117 	int error;
4118 	uint64_t total_slots_used = 0;
4119 	uint64_t max_slot_used = 0;
4120 	uint64_t dnode_slots;
4121 	uint64_t obj_start;
4122 	uint64_t obj_end;
4123 	uint64_t flags;
4124 
4125 	/* make sure nicenum has enough space */
4126 	_Static_assert(sizeof (numbuf) >= NN_NUMBUF_SZ, "numbuf truncated");
4127 
4128 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
4129 	dmu_objset_fast_stat(os, &dds);
4130 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
4131 
4132 	print_header = B_TRUE;
4133 
4134 	if (dds.dds_type < DMU_OST_NUMTYPES)
4135 		type = objset_types[dds.dds_type];
4136 
4137 	if (dds.dds_type == DMU_OST_META) {
4138 		dds.dds_creation_txg = TXG_INITIAL;
4139 		usedobjs = BP_GET_FILL(os->os_rootbp);
4140 		refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)->
4141 		    dd_used_bytes;
4142 	} else {
4143 		dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
4144 	}
4145 
4146 	ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
4147 
4148 	zdb_nicenum(refdbytes, numbuf, sizeof (numbuf));
4149 
4150 	if (verbosity >= 4) {
4151 		(void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp ");
4152 		(void) snprintf_blkptr(blkbuf + strlen(blkbuf),
4153 		    sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp);
4154 	} else {
4155 		blkbuf[0] = '\0';
4156 	}
4157 
4158 	dmu_objset_name(os, osname);
4159 
4160 	(void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
4161 	    "%s, %llu objects%s%s\n",
4162 	    osname, type, (u_longlong_t)dmu_objset_id(os),
4163 	    (u_longlong_t)dds.dds_creation_txg,
4164 	    numbuf, (u_longlong_t)usedobjs, blkbuf,
4165 	    (dds.dds_inconsistent) ? " (inconsistent)" : "");
4166 
4167 	for (i = 0; i < zopt_object_args; i++) {
4168 		obj_start = zopt_object_ranges[i].zor_obj_start;
4169 		obj_end = zopt_object_ranges[i].zor_obj_end;
4170 		flags = zopt_object_ranges[i].zor_flags;
4171 
4172 		object = obj_start;
4173 		if (object == 0 || obj_start == obj_end)
4174 			dump_object(os, object, verbosity, &print_header, NULL,
4175 			    flags);
4176 		else
4177 			object--;
4178 
4179 		while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) &&
4180 		    object <= obj_end) {
4181 			dump_object(os, object, verbosity, &print_header, NULL,
4182 			    flags);
4183 		}
4184 	}
4185 
4186 	if (zopt_object_args > 0) {
4187 		(void) printf("\n");
4188 		return;
4189 	}
4190 
4191 	if (dump_opt['i'] != 0 || verbosity >= 2)
4192 		dump_intent_log(dmu_objset_zil(os));
4193 
4194 	if (dmu_objset_ds(os) != NULL) {
4195 		dsl_dataset_t *ds = dmu_objset_ds(os);
4196 		dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
4197 		if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
4198 		    !dmu_objset_is_snapshot(os)) {
4199 			dump_blkptr_list(&ds->ds_dir->dd_livelist, "Livelist");
4200 			if (verify_dd_livelist(os) != 0)
4201 				fatal("livelist is incorrect");
4202 		}
4203 
4204 		if (dsl_dataset_remap_deadlist_exists(ds)) {
4205 			(void) printf("ds_remap_deadlist:\n");
4206 			dump_blkptr_list(&ds->ds_remap_deadlist, "Deadlist");
4207 		}
4208 		count_ds_mos_objects(ds);
4209 	}
4210 
4211 	if (dmu_objset_ds(os) != NULL)
4212 		dump_bookmarks(os, verbosity);
4213 
4214 	if (verbosity < 2)
4215 		return;
4216 
4217 	if (BP_IS_HOLE(os->os_rootbp))
4218 		return;
4219 
4220 	dump_object(os, 0, verbosity, &print_header, NULL, 0);
4221 	object_count = 0;
4222 	if (DMU_USERUSED_DNODE(os) != NULL &&
4223 	    DMU_USERUSED_DNODE(os)->dn_type != 0) {
4224 		dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header,
4225 		    NULL, 0);
4226 		dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header,
4227 		    NULL, 0);
4228 	}
4229 
4230 	if (DMU_PROJECTUSED_DNODE(os) != NULL &&
4231 	    DMU_PROJECTUSED_DNODE(os)->dn_type != 0)
4232 		dump_object(os, DMU_PROJECTUSED_OBJECT, verbosity,
4233 		    &print_header, NULL, 0);
4234 
4235 	object = 0;
4236 	while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
4237 		dump_object(os, object, verbosity, &print_header, &dnode_slots,
4238 		    0);
4239 		object_count++;
4240 		total_slots_used += dnode_slots;
4241 		max_slot_used = object + dnode_slots - 1;
4242 	}
4243 
4244 	(void) printf("\n");
4245 
4246 	(void) printf("    Dnode slots:\n");
4247 	(void) printf("\tTotal used:    %10llu\n",
4248 	    (u_longlong_t)total_slots_used);
4249 	(void) printf("\tMax used:      %10llu\n",
4250 	    (u_longlong_t)max_slot_used);
4251 	(void) printf("\tPercent empty: %10lf\n",
4252 	    (double)(max_slot_used - total_slots_used)*100 /
4253 	    (double)max_slot_used);
4254 	(void) printf("\n");
4255 
4256 	if (error != ESRCH) {
4257 		(void) fprintf(stderr, "dmu_object_next() = %d\n", error);
4258 		abort();
4259 	}
4260 
4261 	ASSERT3U(object_count, ==, usedobjs);
4262 
4263 	if (leaked_objects != 0) {
4264 		(void) printf("%d potentially leaked objects detected\n",
4265 		    leaked_objects);
4266 		leaked_objects = 0;
4267 	}
4268 }
4269 
4270 static void
dump_uberblock(uberblock_t * ub,const char * header,const char * footer)4271 dump_uberblock(uberblock_t *ub, const char *header, const char *footer)
4272 {
4273 	time_t timestamp = ub->ub_timestamp;
4274 
4275 	(void) printf("%s", header ? header : "");
4276 	(void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic);
4277 	(void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version);
4278 	(void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg);
4279 	(void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum);
4280 	(void) printf("\ttimestamp = %llu UTC = %s",
4281 	    (u_longlong_t)ub->ub_timestamp, ctime(&timestamp));
4282 
4283 	char blkbuf[BP_SPRINTF_LEN];
4284 	snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4285 	(void) printf("\tbp = %s\n", blkbuf);
4286 
4287 	(void) printf("\tmmp_magic = %016llx\n",
4288 	    (u_longlong_t)ub->ub_mmp_magic);
4289 	if (MMP_VALID(ub)) {
4290 		(void) printf("\tmmp_delay = %0llu\n",
4291 		    (u_longlong_t)ub->ub_mmp_delay);
4292 		if (MMP_SEQ_VALID(ub))
4293 			(void) printf("\tmmp_seq = %u\n",
4294 			    (unsigned int) MMP_SEQ(ub));
4295 		if (MMP_FAIL_INT_VALID(ub))
4296 			(void) printf("\tmmp_fail = %u\n",
4297 			    (unsigned int) MMP_FAIL_INT(ub));
4298 		if (MMP_INTERVAL_VALID(ub))
4299 			(void) printf("\tmmp_write = %u\n",
4300 			    (unsigned int) MMP_INTERVAL(ub));
4301 		/* After MMP_* to make summarize_uberblock_mmp cleaner */
4302 		(void) printf("\tmmp_valid = %x\n",
4303 		    (unsigned int) ub->ub_mmp_config & 0xFF);
4304 	}
4305 
4306 	if (dump_opt['u'] >= 4) {
4307 		char blkbuf[BP_SPRINTF_LEN];
4308 		snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4309 		(void) printf("\trootbp = %s\n", blkbuf);
4310 	}
4311 	(void) printf("\tcheckpoint_txg = %llu\n",
4312 	    (u_longlong_t)ub->ub_checkpoint_txg);
4313 
4314 	(void) printf("\traidz_reflow state=%u off=%llu\n",
4315 	    (int)RRSS_GET_STATE(ub),
4316 	    (u_longlong_t)RRSS_GET_OFFSET(ub));
4317 
4318 	(void) printf("%s", footer ? footer : "");
4319 }
4320 
4321 static void
dump_config(spa_t * spa)4322 dump_config(spa_t *spa)
4323 {
4324 	dmu_buf_t *db;
4325 	size_t nvsize = 0;
4326 	int error = 0;
4327 
4328 
4329 	error = dmu_bonus_hold(spa->spa_meta_objset,
4330 	    spa->spa_config_object, FTAG, &db);
4331 
4332 	if (error == 0) {
4333 		nvsize = *(uint64_t *)db->db_data;
4334 		dmu_buf_rele(db, FTAG);
4335 
4336 		(void) printf("\nMOS Configuration:\n");
4337 		dump_packed_nvlist(spa->spa_meta_objset,
4338 		    spa->spa_config_object, (void *)&nvsize, 1);
4339 	} else {
4340 		(void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d",
4341 		    (u_longlong_t)spa->spa_config_object, error);
4342 	}
4343 }
4344 
4345 static void
dump_cachefile(const char * cachefile)4346 dump_cachefile(const char *cachefile)
4347 {
4348 	int fd;
4349 	struct stat64 statbuf;
4350 	char *buf;
4351 	nvlist_t *config;
4352 
4353 	if ((fd = open64(cachefile, O_RDONLY)) < 0) {
4354 		(void) printf("cannot open '%s': %s\n", cachefile,
4355 		    strerror(errno));
4356 		zdb_exit(1);
4357 	}
4358 
4359 	if (fstat64(fd, &statbuf) != 0) {
4360 		(void) printf("failed to stat '%s': %s\n", cachefile,
4361 		    strerror(errno));
4362 		zdb_exit(1);
4363 	}
4364 
4365 	if ((buf = malloc(statbuf.st_size)) == NULL) {
4366 		(void) fprintf(stderr, "failed to allocate %llu bytes\n",
4367 		    (u_longlong_t)statbuf.st_size);
4368 		zdb_exit(1);
4369 	}
4370 
4371 	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
4372 		(void) fprintf(stderr, "failed to read %llu bytes\n",
4373 		    (u_longlong_t)statbuf.st_size);
4374 		zdb_exit(1);
4375 	}
4376 
4377 	(void) close(fd);
4378 
4379 	if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) {
4380 		(void) fprintf(stderr, "failed to unpack nvlist\n");
4381 		zdb_exit(1);
4382 	}
4383 
4384 	free(buf);
4385 
4386 	dump_nvlist(config, 0);
4387 
4388 	nvlist_free(config);
4389 }
4390 
4391 /*
4392  * ZFS label nvlist stats
4393  */
4394 typedef struct zdb_nvl_stats {
4395 	int		zns_list_count;
4396 	int		zns_leaf_count;
4397 	size_t		zns_leaf_largest;
4398 	size_t		zns_leaf_total;
4399 	nvlist_t	*zns_string;
4400 	nvlist_t	*zns_uint64;
4401 	nvlist_t	*zns_boolean;
4402 } zdb_nvl_stats_t;
4403 
4404 static void
collect_nvlist_stats(nvlist_t * nvl,zdb_nvl_stats_t * stats)4405 collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats)
4406 {
4407 	nvlist_t *list, **array;
4408 	nvpair_t *nvp = NULL;
4409 	const char *name;
4410 	uint_t i, items;
4411 
4412 	stats->zns_list_count++;
4413 
4414 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4415 		name = nvpair_name(nvp);
4416 
4417 		switch (nvpair_type(nvp)) {
4418 		case DATA_TYPE_STRING:
4419 			fnvlist_add_string(stats->zns_string, name,
4420 			    fnvpair_value_string(nvp));
4421 			break;
4422 		case DATA_TYPE_UINT64:
4423 			fnvlist_add_uint64(stats->zns_uint64, name,
4424 			    fnvpair_value_uint64(nvp));
4425 			break;
4426 		case DATA_TYPE_BOOLEAN:
4427 			fnvlist_add_boolean(stats->zns_boolean, name);
4428 			break;
4429 		case DATA_TYPE_NVLIST:
4430 			if (nvpair_value_nvlist(nvp, &list) == 0)
4431 				collect_nvlist_stats(list, stats);
4432 			break;
4433 		case DATA_TYPE_NVLIST_ARRAY:
4434 			if (nvpair_value_nvlist_array(nvp, &array, &items) != 0)
4435 				break;
4436 
4437 			for (i = 0; i < items; i++) {
4438 				collect_nvlist_stats(array[i], stats);
4439 
4440 				/* collect stats on leaf vdev */
4441 				if (strcmp(name, "children") == 0) {
4442 					size_t size;
4443 
4444 					(void) nvlist_size(array[i], &size,
4445 					    NV_ENCODE_XDR);
4446 					stats->zns_leaf_total += size;
4447 					if (size > stats->zns_leaf_largest)
4448 						stats->zns_leaf_largest = size;
4449 					stats->zns_leaf_count++;
4450 				}
4451 			}
4452 			break;
4453 		default:
4454 			(void) printf("skip type %d!\n", (int)nvpair_type(nvp));
4455 		}
4456 	}
4457 }
4458 
4459 static void
dump_nvlist_stats(nvlist_t * nvl,size_t cap)4460 dump_nvlist_stats(nvlist_t *nvl, size_t cap)
4461 {
4462 	zdb_nvl_stats_t stats = { 0 };
4463 	size_t size, sum = 0, total;
4464 	size_t noise;
4465 
4466 	/* requires nvlist with non-unique names for stat collection */
4467 	VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0));
4468 	VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0));
4469 	VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0));
4470 	VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR));
4471 
4472 	(void) printf("\n\nZFS Label NVList Config Stats:\n");
4473 
4474 	VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR));
4475 	(void) printf("  %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4476 	    (int)total, (int)(cap - total), 100.0 * total / cap);
4477 
4478 	collect_nvlist_stats(nvl, &stats);
4479 
4480 	VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR));
4481 	size -= noise;
4482 	sum += size;
4483 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4484 	    (int)fnvlist_num_pairs(stats.zns_uint64),
4485 	    (int)size, 100.0 * size / total);
4486 
4487 	VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR));
4488 	size -= noise;
4489 	sum += size;
4490 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4491 	    (int)fnvlist_num_pairs(stats.zns_string),
4492 	    (int)size, 100.0 * size / total);
4493 
4494 	VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR));
4495 	size -= noise;
4496 	sum += size;
4497 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4498 	    (int)fnvlist_num_pairs(stats.zns_boolean),
4499 	    (int)size, 100.0 * size / total);
4500 
4501 	size = total - sum;	/* treat remainder as nvlist overhead */
4502 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4503 	    stats.zns_list_count, (int)size, 100.0 * size / total);
4504 
4505 	if (stats.zns_leaf_count > 0) {
4506 		size_t average = stats.zns_leaf_total / stats.zns_leaf_count;
4507 
4508 		(void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4509 		    stats.zns_leaf_count, (int)average);
4510 		(void) printf("%24d bytes largest\n",
4511 		    (int)stats.zns_leaf_largest);
4512 
4513 		if (dump_opt['l'] >= 3 && average > 0)
4514 			(void) printf("  space for %d additional leaf vdevs\n",
4515 			    (int)((cap - total) / average));
4516 	}
4517 	(void) printf("\n");
4518 
4519 	nvlist_free(stats.zns_string);
4520 	nvlist_free(stats.zns_uint64);
4521 	nvlist_free(stats.zns_boolean);
4522 }
4523 
4524 typedef struct cksum_record {
4525 	zio_cksum_t cksum;
4526 	boolean_t labels[VDEV_LABELS];
4527 	avl_node_t link;
4528 } cksum_record_t;
4529 
4530 static int
cksum_record_compare(const void * x1,const void * x2)4531 cksum_record_compare(const void *x1, const void *x2)
4532 {
4533 	const cksum_record_t *l = (cksum_record_t *)x1;
4534 	const cksum_record_t *r = (cksum_record_t *)x2;
4535 	int arraysize = ARRAY_SIZE(l->cksum.zc_word);
4536 	int difference = 0;
4537 
4538 	for (int i = 0; i < arraysize; i++) {
4539 		difference = TREE_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]);
4540 		if (difference)
4541 			break;
4542 	}
4543 
4544 	return (difference);
4545 }
4546 
4547 static cksum_record_t *
cksum_record_alloc(zio_cksum_t * cksum,int l)4548 cksum_record_alloc(zio_cksum_t *cksum, int l)
4549 {
4550 	cksum_record_t *rec;
4551 
4552 	rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL);
4553 	rec->cksum = *cksum;
4554 	rec->labels[l] = B_TRUE;
4555 
4556 	return (rec);
4557 }
4558 
4559 static cksum_record_t *
cksum_record_lookup(avl_tree_t * tree,zio_cksum_t * cksum)4560 cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum)
4561 {
4562 	cksum_record_t lookup = { .cksum = *cksum };
4563 	avl_index_t where;
4564 
4565 	return (avl_find(tree, &lookup, &where));
4566 }
4567 
4568 static cksum_record_t *
cksum_record_insert(avl_tree_t * tree,zio_cksum_t * cksum,int l)4569 cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l)
4570 {
4571 	cksum_record_t *rec;
4572 
4573 	rec = cksum_record_lookup(tree, cksum);
4574 	if (rec) {
4575 		rec->labels[l] = B_TRUE;
4576 	} else {
4577 		rec = cksum_record_alloc(cksum, l);
4578 		avl_add(tree, rec);
4579 	}
4580 
4581 	return (rec);
4582 }
4583 
4584 static int
first_label(cksum_record_t * rec)4585 first_label(cksum_record_t *rec)
4586 {
4587 	for (int i = 0; i < VDEV_LABELS; i++)
4588 		if (rec->labels[i])
4589 			return (i);
4590 
4591 	return (-1);
4592 }
4593 
4594 static void
print_label_numbers(const char * prefix,const cksum_record_t * rec)4595 print_label_numbers(const char *prefix, const cksum_record_t *rec)
4596 {
4597 	fputs(prefix, stdout);
4598 	for (int i = 0; i < VDEV_LABELS; i++)
4599 		if (rec->labels[i] == B_TRUE)
4600 			printf("%d ", i);
4601 	putchar('\n');
4602 }
4603 
4604 #define	MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4605 
4606 typedef struct zdb_label {
4607 	vdev_label_t label;
4608 	uint64_t label_offset;
4609 	nvlist_t *config_nv;
4610 	cksum_record_t *config;
4611 	cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT];
4612 	boolean_t header_printed;
4613 	boolean_t read_failed;
4614 	boolean_t cksum_valid;
4615 } zdb_label_t;
4616 
4617 static void
print_label_header(zdb_label_t * label,int l)4618 print_label_header(zdb_label_t *label, int l)
4619 {
4620 
4621 	if (dump_opt['q'])
4622 		return;
4623 
4624 	if (label->header_printed == B_TRUE)
4625 		return;
4626 
4627 	(void) printf("------------------------------------\n");
4628 	(void) printf("LABEL %d %s\n", l,
4629 	    label->cksum_valid ? "" : "(Bad label cksum)");
4630 	(void) printf("------------------------------------\n");
4631 
4632 	label->header_printed = B_TRUE;
4633 }
4634 
4635 static void
print_l2arc_header(void)4636 print_l2arc_header(void)
4637 {
4638 	(void) printf("------------------------------------\n");
4639 	(void) printf("L2ARC device header\n");
4640 	(void) printf("------------------------------------\n");
4641 }
4642 
4643 static void
print_l2arc_log_blocks(void)4644 print_l2arc_log_blocks(void)
4645 {
4646 	(void) printf("------------------------------------\n");
4647 	(void) printf("L2ARC device log blocks\n");
4648 	(void) printf("------------------------------------\n");
4649 }
4650 
4651 static void
dump_l2arc_log_entries(uint64_t log_entries,l2arc_log_ent_phys_t * le,uint64_t i)4652 dump_l2arc_log_entries(uint64_t log_entries,
4653     l2arc_log_ent_phys_t *le, uint64_t i)
4654 {
4655 	for (int j = 0; j < log_entries; j++) {
4656 		dva_t dva = le[j].le_dva;
4657 		(void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4658 		    "vdev: %llu, offset: %llu\n",
4659 		    (u_longlong_t)i, j + 1,
4660 		    (u_longlong_t)DVA_GET_ASIZE(&dva),
4661 		    (u_longlong_t)DVA_GET_VDEV(&dva),
4662 		    (u_longlong_t)DVA_GET_OFFSET(&dva));
4663 		(void) printf("|\t\t\t\tbirth: %llu\n",
4664 		    (u_longlong_t)le[j].le_birth);
4665 		(void) printf("|\t\t\t\tlsize: %llu\n",
4666 		    (u_longlong_t)L2BLK_GET_LSIZE((&le[j])->le_prop));
4667 		(void) printf("|\t\t\t\tpsize: %llu\n",
4668 		    (u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop));
4669 		(void) printf("|\t\t\t\tcompr: %llu\n",
4670 		    (u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop));
4671 		(void) printf("|\t\t\t\tcomplevel: %llu\n",
4672 		    (u_longlong_t)(&le[j])->le_complevel);
4673 		(void) printf("|\t\t\t\ttype: %llu\n",
4674 		    (u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop));
4675 		(void) printf("|\t\t\t\tprotected: %llu\n",
4676 		    (u_longlong_t)L2BLK_GET_PROTECTED((&le[j])->le_prop));
4677 		(void) printf("|\t\t\t\tprefetch: %llu\n",
4678 		    (u_longlong_t)L2BLK_GET_PREFETCH((&le[j])->le_prop));
4679 		(void) printf("|\t\t\t\taddress: %llu\n",
4680 		    (u_longlong_t)le[j].le_daddr);
4681 		(void) printf("|\t\t\t\tARC state: %llu\n",
4682 		    (u_longlong_t)L2BLK_GET_STATE((&le[j])->le_prop));
4683 		(void) printf("|\n");
4684 	}
4685 	(void) printf("\n");
4686 }
4687 
4688 static void
dump_l2arc_log_blkptr(const l2arc_log_blkptr_t * lbps)4689 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t *lbps)
4690 {
4691 	(void) printf("|\t\tdaddr: %llu\n", (u_longlong_t)lbps->lbp_daddr);
4692 	(void) printf("|\t\tpayload_asize: %llu\n",
4693 	    (u_longlong_t)lbps->lbp_payload_asize);
4694 	(void) printf("|\t\tpayload_start: %llu\n",
4695 	    (u_longlong_t)lbps->lbp_payload_start);
4696 	(void) printf("|\t\tlsize: %llu\n",
4697 	    (u_longlong_t)L2BLK_GET_LSIZE(lbps->lbp_prop));
4698 	(void) printf("|\t\tasize: %llu\n",
4699 	    (u_longlong_t)L2BLK_GET_PSIZE(lbps->lbp_prop));
4700 	(void) printf("|\t\tcompralgo: %llu\n",
4701 	    (u_longlong_t)L2BLK_GET_COMPRESS(lbps->lbp_prop));
4702 	(void) printf("|\t\tcksumalgo: %llu\n",
4703 	    (u_longlong_t)L2BLK_GET_CHECKSUM(lbps->lbp_prop));
4704 	(void) printf("|\n\n");
4705 }
4706 
4707 static void
dump_l2arc_log_blocks(int fd,const l2arc_dev_hdr_phys_t * l2dhdr,l2arc_dev_hdr_phys_t * rebuild)4708 dump_l2arc_log_blocks(int fd, const l2arc_dev_hdr_phys_t *l2dhdr,
4709     l2arc_dev_hdr_phys_t *rebuild)
4710 {
4711 	l2arc_log_blk_phys_t this_lb;
4712 	uint64_t asize;
4713 	l2arc_log_blkptr_t lbps[2];
4714 	zio_cksum_t cksum;
4715 	int failed = 0;
4716 	l2arc_dev_t dev;
4717 
4718 	if (!dump_opt['q'])
4719 		print_l2arc_log_blocks();
4720 	memcpy(lbps, l2dhdr->dh_start_lbps, sizeof (lbps));
4721 
4722 	dev.l2ad_evict = l2dhdr->dh_evict;
4723 	dev.l2ad_start = l2dhdr->dh_start;
4724 	dev.l2ad_end = l2dhdr->dh_end;
4725 
4726 	if (l2dhdr->dh_start_lbps[0].lbp_daddr == 0) {
4727 		/* no log blocks to read */
4728 		if (!dump_opt['q']) {
4729 			(void) printf("No log blocks to read\n");
4730 			(void) printf("\n");
4731 		}
4732 		return;
4733 	} else {
4734 		dev.l2ad_hand = lbps[0].lbp_daddr +
4735 		    L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4736 	}
4737 
4738 	dev.l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST);
4739 
4740 	for (;;) {
4741 		if (!l2arc_log_blkptr_valid(&dev, &lbps[0]))
4742 			break;
4743 
4744 		/* L2BLK_GET_PSIZE returns aligned size for log blocks */
4745 		asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4746 		if (pread64(fd, &this_lb, asize, lbps[0].lbp_daddr) != asize) {
4747 			if (!dump_opt['q']) {
4748 				(void) printf("Error while reading next log "
4749 				    "block\n\n");
4750 			}
4751 			break;
4752 		}
4753 
4754 		fletcher_4_native_varsize(&this_lb, asize, &cksum);
4755 		if (!ZIO_CHECKSUM_EQUAL(cksum, lbps[0].lbp_cksum)) {
4756 			failed++;
4757 			if (!dump_opt['q']) {
4758 				(void) printf("Invalid cksum\n");
4759 				dump_l2arc_log_blkptr(&lbps[0]);
4760 			}
4761 			break;
4762 		}
4763 
4764 		switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) {
4765 		case ZIO_COMPRESS_OFF:
4766 			break;
4767 		default: {
4768 			abd_t *abd = abd_alloc_linear(asize, B_TRUE);
4769 			abd_copy_from_buf_off(abd, &this_lb, 0, asize);
4770 			abd_t dabd;
4771 			abd_get_from_buf_struct(&dabd, &this_lb,
4772 			    sizeof (this_lb));
4773 			int err = zio_decompress_data(L2BLK_GET_COMPRESS(
4774 			    (&lbps[0])->lbp_prop), abd, &dabd,
4775 			    asize, sizeof (this_lb), NULL);
4776 			abd_free(&dabd);
4777 			abd_free(abd);
4778 			if (err != 0) {
4779 				(void) printf("L2ARC block decompression "
4780 				    "failed\n");
4781 				goto out;
4782 			}
4783 			break;
4784 		}
4785 		}
4786 
4787 		if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
4788 			byteswap_uint64_array(&this_lb, sizeof (this_lb));
4789 		if (this_lb.lb_magic != L2ARC_LOG_BLK_MAGIC) {
4790 			if (!dump_opt['q'])
4791 				(void) printf("Invalid log block magic\n\n");
4792 			break;
4793 		}
4794 
4795 		rebuild->dh_lb_count++;
4796 		rebuild->dh_lb_asize += asize;
4797 		if (dump_opt['l'] > 1 && !dump_opt['q']) {
4798 			(void) printf("lb[%4llu]\tmagic: %llu\n",
4799 			    (u_longlong_t)rebuild->dh_lb_count,
4800 			    (u_longlong_t)this_lb.lb_magic);
4801 			dump_l2arc_log_blkptr(&lbps[0]);
4802 		}
4803 
4804 		if (dump_opt['l'] > 2 && !dump_opt['q'])
4805 			dump_l2arc_log_entries(l2dhdr->dh_log_entries,
4806 			    this_lb.lb_entries,
4807 			    rebuild->dh_lb_count);
4808 
4809 		if (l2arc_range_check_overlap(lbps[1].lbp_payload_start,
4810 		    lbps[0].lbp_payload_start, dev.l2ad_evict) &&
4811 		    !dev.l2ad_first)
4812 			break;
4813 
4814 		lbps[0] = lbps[1];
4815 		lbps[1] = this_lb.lb_prev_lbp;
4816 	}
4817 out:
4818 	if (!dump_opt['q']) {
4819 		(void) printf("log_blk_count:\t %llu with valid cksum\n",
4820 		    (u_longlong_t)rebuild->dh_lb_count);
4821 		(void) printf("\t\t %d with invalid cksum\n", failed);
4822 		(void) printf("log_blk_asize:\t %llu\n\n",
4823 		    (u_longlong_t)rebuild->dh_lb_asize);
4824 	}
4825 }
4826 
4827 static int
dump_l2arc_header(int fd)4828 dump_l2arc_header(int fd)
4829 {
4830 	l2arc_dev_hdr_phys_t l2dhdr = {0}, rebuild = {0};
4831 	int error = B_FALSE;
4832 
4833 	if (pread64(fd, &l2dhdr, sizeof (l2dhdr),
4834 	    VDEV_LABEL_START_SIZE) != sizeof (l2dhdr)) {
4835 		error = B_TRUE;
4836 	} else {
4837 		if (l2dhdr.dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC))
4838 			byteswap_uint64_array(&l2dhdr, sizeof (l2dhdr));
4839 
4840 		if (l2dhdr.dh_magic != L2ARC_DEV_HDR_MAGIC)
4841 			error = B_TRUE;
4842 	}
4843 
4844 	if (error) {
4845 		(void) printf("L2ARC device header not found\n\n");
4846 		/* Do not return an error here for backward compatibility */
4847 		return (0);
4848 	} else if (!dump_opt['q']) {
4849 		print_l2arc_header();
4850 
4851 		(void) printf("    magic: %llu\n",
4852 		    (u_longlong_t)l2dhdr.dh_magic);
4853 		(void) printf("    version: %llu\n",
4854 		    (u_longlong_t)l2dhdr.dh_version);
4855 		(void) printf("    pool_guid: %llu\n",
4856 		    (u_longlong_t)l2dhdr.dh_spa_guid);
4857 		(void) printf("    flags: %llu\n",
4858 		    (u_longlong_t)l2dhdr.dh_flags);
4859 		(void) printf("    start_lbps[0]: %llu\n",
4860 		    (u_longlong_t)
4861 		    l2dhdr.dh_start_lbps[0].lbp_daddr);
4862 		(void) printf("    start_lbps[1]: %llu\n",
4863 		    (u_longlong_t)
4864 		    l2dhdr.dh_start_lbps[1].lbp_daddr);
4865 		(void) printf("    log_blk_ent: %llu\n",
4866 		    (u_longlong_t)l2dhdr.dh_log_entries);
4867 		(void) printf("    start: %llu\n",
4868 		    (u_longlong_t)l2dhdr.dh_start);
4869 		(void) printf("    end: %llu\n",
4870 		    (u_longlong_t)l2dhdr.dh_end);
4871 		(void) printf("    evict: %llu\n",
4872 		    (u_longlong_t)l2dhdr.dh_evict);
4873 		(void) printf("    lb_asize_refcount: %llu\n",
4874 		    (u_longlong_t)l2dhdr.dh_lb_asize);
4875 		(void) printf("    lb_count_refcount: %llu\n",
4876 		    (u_longlong_t)l2dhdr.dh_lb_count);
4877 		(void) printf("    trim_action_time: %llu\n",
4878 		    (u_longlong_t)l2dhdr.dh_trim_action_time);
4879 		(void) printf("    trim_state: %llu\n\n",
4880 		    (u_longlong_t)l2dhdr.dh_trim_state);
4881 	}
4882 
4883 	dump_l2arc_log_blocks(fd, &l2dhdr, &rebuild);
4884 	/*
4885 	 * The total aligned size of log blocks and the number of log blocks
4886 	 * reported in the header of the device may be less than what zdb
4887 	 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
4888 	 * This happens because dump_l2arc_log_blocks() lacks the memory
4889 	 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
4890 	 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
4891 	 * and dh_lb_count will be lower to begin with than what exists on the
4892 	 * device. This is normal and zdb should not exit with an error. The
4893 	 * opposite case should never happen though, the values reported in the
4894 	 * header should never be higher than what dump_l2arc_log_blocks() and
4895 	 * l2arc_rebuild() report. If this happens there is a leak in the
4896 	 * accounting of log blocks.
4897 	 */
4898 	if (l2dhdr.dh_lb_asize > rebuild.dh_lb_asize ||
4899 	    l2dhdr.dh_lb_count > rebuild.dh_lb_count)
4900 		return (1);
4901 
4902 	return (0);
4903 }
4904 
4905 static void
dump_config_from_label(zdb_label_t * label,size_t buflen,int l)4906 dump_config_from_label(zdb_label_t *label, size_t buflen, int l)
4907 {
4908 	if (dump_opt['q'])
4909 		return;
4910 
4911 	if ((dump_opt['l'] < 3) && (first_label(label->config) != l))
4912 		return;
4913 
4914 	print_label_header(label, l);
4915 	dump_nvlist(label->config_nv, 4);
4916 	print_label_numbers("    labels = ", label->config);
4917 
4918 	if (dump_opt['l'] >= 2)
4919 		dump_nvlist_stats(label->config_nv, buflen);
4920 }
4921 
4922 #define	ZDB_MAX_UB_HEADER_SIZE 32
4923 
4924 static void
dump_label_uberblocks(zdb_label_t * label,uint64_t ashift,int label_num)4925 dump_label_uberblocks(zdb_label_t *label, uint64_t ashift, int label_num)
4926 {
4927 
4928 	vdev_t vd;
4929 	char header[ZDB_MAX_UB_HEADER_SIZE];
4930 
4931 	vd.vdev_ashift = ashift;
4932 	vd.vdev_top = &vd;
4933 
4934 	for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
4935 		uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
4936 		uberblock_t *ub = (void *)((char *)&label->label + uoff);
4937 		cksum_record_t *rec = label->uberblocks[i];
4938 
4939 		if (rec == NULL) {
4940 			if (dump_opt['u'] >= 2) {
4941 				print_label_header(label, label_num);
4942 				(void) printf("    Uberblock[%d] invalid\n", i);
4943 			}
4944 			continue;
4945 		}
4946 
4947 		if ((dump_opt['u'] < 3) && (first_label(rec) != label_num))
4948 			continue;
4949 
4950 		if ((dump_opt['u'] < 4) &&
4951 		    (ub->ub_mmp_magic == MMP_MAGIC) && ub->ub_mmp_delay &&
4952 		    (i >= VDEV_UBERBLOCK_COUNT(&vd) - MMP_BLOCKS_PER_LABEL))
4953 			continue;
4954 
4955 		print_label_header(label, label_num);
4956 		(void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE,
4957 		    "    Uberblock[%d]\n", i);
4958 		dump_uberblock(ub, header, "");
4959 		print_label_numbers("        labels = ", rec);
4960 	}
4961 }
4962 
4963 static char curpath[PATH_MAX];
4964 
4965 /*
4966  * Iterate through the path components, recursively passing
4967  * current one's obj and remaining path until we find the obj
4968  * for the last one.
4969  */
4970 static int
dump_path_impl(objset_t * os,uint64_t obj,char * name,uint64_t * retobj)4971 dump_path_impl(objset_t *os, uint64_t obj, char *name, uint64_t *retobj)
4972 {
4973 	int err;
4974 	boolean_t header = B_TRUE;
4975 	uint64_t child_obj;
4976 	char *s;
4977 	dmu_buf_t *db;
4978 	dmu_object_info_t doi;
4979 
4980 	if ((s = strchr(name, '/')) != NULL)
4981 		*s = '\0';
4982 	err = zap_lookup(os, obj, name, 8, 1, &child_obj);
4983 
4984 	(void) strlcat(curpath, name, sizeof (curpath));
4985 
4986 	if (err != 0) {
4987 		(void) fprintf(stderr, "failed to lookup %s: %s\n",
4988 		    curpath, strerror(err));
4989 		return (err);
4990 	}
4991 
4992 	child_obj = ZFS_DIRENT_OBJ(child_obj);
4993 	err = sa_buf_hold(os, child_obj, FTAG, &db);
4994 	if (err != 0) {
4995 		(void) fprintf(stderr,
4996 		    "failed to get SA dbuf for obj %llu: %s\n",
4997 		    (u_longlong_t)child_obj, strerror(err));
4998 		return (EINVAL);
4999 	}
5000 	dmu_object_info_from_db(db, &doi);
5001 	sa_buf_rele(db, FTAG);
5002 
5003 	if (doi.doi_bonus_type != DMU_OT_SA &&
5004 	    doi.doi_bonus_type != DMU_OT_ZNODE) {
5005 		(void) fprintf(stderr, "invalid bonus type %d for obj %llu\n",
5006 		    doi.doi_bonus_type, (u_longlong_t)child_obj);
5007 		return (EINVAL);
5008 	}
5009 
5010 	if (dump_opt['v'] > 6) {
5011 		(void) printf("obj=%llu %s type=%d bonustype=%d\n",
5012 		    (u_longlong_t)child_obj, curpath, doi.doi_type,
5013 		    doi.doi_bonus_type);
5014 	}
5015 
5016 	(void) strlcat(curpath, "/", sizeof (curpath));
5017 
5018 	switch (doi.doi_type) {
5019 	case DMU_OT_DIRECTORY_CONTENTS:
5020 		if (s != NULL && *(s + 1) != '\0')
5021 			return (dump_path_impl(os, child_obj, s + 1, retobj));
5022 		zfs_fallthrough;
5023 	case DMU_OT_PLAIN_FILE_CONTENTS:
5024 		if (retobj != NULL) {
5025 			*retobj = child_obj;
5026 		} else {
5027 			dump_object(os, child_obj, dump_opt['v'], &header,
5028 			    NULL, 0);
5029 		}
5030 		return (0);
5031 	default:
5032 		(void) fprintf(stderr, "object %llu has non-file/directory "
5033 		    "type %d\n", (u_longlong_t)obj, doi.doi_type);
5034 		break;
5035 	}
5036 
5037 	return (EINVAL);
5038 }
5039 
5040 /*
5041  * Dump the blocks for the object specified by path inside the dataset.
5042  */
5043 static int
dump_path(char * ds,char * path,uint64_t * retobj)5044 dump_path(char *ds, char *path, uint64_t *retobj)
5045 {
5046 	int err;
5047 	objset_t *os;
5048 	uint64_t root_obj;
5049 
5050 	err = open_objset(ds, FTAG, &os);
5051 	if (err != 0)
5052 		return (err);
5053 
5054 	err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj);
5055 	if (err != 0) {
5056 		(void) fprintf(stderr, "can't lookup root znode: %s\n",
5057 		    strerror(err));
5058 		close_objset(os, FTAG);
5059 		return (EINVAL);
5060 	}
5061 
5062 	(void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds);
5063 
5064 	err = dump_path_impl(os, root_obj, path, retobj);
5065 
5066 	close_objset(os, FTAG);
5067 	return (err);
5068 }
5069 
5070 static int
dump_backup_bytes(objset_t * os,void * buf,int len,void * arg)5071 dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
5072 {
5073 	const char *p = (const char *)buf;
5074 	ssize_t nwritten;
5075 
5076 	(void) os;
5077 	(void) arg;
5078 
5079 	/* Write the data out, handling short writes and signals. */
5080 	while ((nwritten = write(STDOUT_FILENO, p, len)) < len) {
5081 		if (nwritten < 0) {
5082 			if (errno == EINTR)
5083 				continue;
5084 			return (errno);
5085 		}
5086 		p += nwritten;
5087 		len -= nwritten;
5088 	}
5089 
5090 	return (0);
5091 }
5092 
5093 static void
dump_backup(const char * pool,uint64_t objset_id,const char * flagstr)5094 dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
5095 {
5096 	boolean_t embed = B_FALSE;
5097 	boolean_t large_block = B_FALSE;
5098 	boolean_t compress = B_FALSE;
5099 	boolean_t raw = B_FALSE;
5100 
5101 	const char *c;
5102 	for (c = flagstr; c != NULL && *c != '\0'; c++) {
5103 		switch (*c) {
5104 			case 'e':
5105 				embed = B_TRUE;
5106 				break;
5107 			case 'L':
5108 				large_block = B_TRUE;
5109 				break;
5110 			case 'c':
5111 				compress = B_TRUE;
5112 				break;
5113 			case 'w':
5114 				raw = B_TRUE;
5115 				break;
5116 			default:
5117 				fprintf(stderr, "dump_backup: invalid flag "
5118 				    "'%c'\n", *c);
5119 				return;
5120 		}
5121 	}
5122 
5123 	if (isatty(STDOUT_FILENO)) {
5124 		fprintf(stderr, "dump_backup: stream cannot be written "
5125 		    "to a terminal\n");
5126 		return;
5127 	}
5128 
5129 	offset_t off = 0;
5130 	dmu_send_outparams_t out = {
5131 	    .dso_outfunc = dump_backup_bytes,
5132 	    .dso_dryrun  = B_FALSE,
5133 	};
5134 
5135 	int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed,
5136 	    large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO,
5137 	    &off, &out);
5138 	if (err != 0) {
5139 		fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",
5140 		    strerror(err));
5141 		return;
5142 	}
5143 }
5144 
5145 static int
zdb_copy_object(objset_t * os,uint64_t srcobj,char * destfile)5146 zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile)
5147 {
5148 	int err = 0;
5149 	uint64_t size, readsize, oursize, offset;
5150 	ssize_t writesize;
5151 	sa_handle_t *hdl;
5152 
5153 	(void) printf("Copying object %" PRIu64 " to file %s\n", srcobj,
5154 	    destfile);
5155 
5156 	VERIFY3P(os, ==, sa_os);
5157 	if ((err = sa_handle_get(os, srcobj, NULL, SA_HDL_PRIVATE, &hdl))) {
5158 		(void) printf("Failed to get handle for SA znode\n");
5159 		return (err);
5160 	}
5161 	if ((err = sa_lookup(hdl, sa_attr_table[ZPL_SIZE], &size, 8))) {
5162 		(void) sa_handle_destroy(hdl);
5163 		return (err);
5164 	}
5165 	(void) sa_handle_destroy(hdl);
5166 
5167 	(void) printf("Object %" PRIu64 " is %" PRIu64 " bytes\n", srcobj,
5168 	    size);
5169 	if (size == 0) {
5170 		return (EINVAL);
5171 	}
5172 
5173 	int fd = open(destfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
5174 	if (fd == -1)
5175 		return (errno);
5176 	/*
5177 	 * We cap the size at 1 mebibyte here to prevent
5178 	 * allocation failures and nigh-infinite printing if the
5179 	 * object is extremely large.
5180 	 */
5181 	oursize = MIN(size, 1 << 20);
5182 	offset = 0;
5183 	char *buf = kmem_alloc(oursize, KM_NOSLEEP);
5184 	if (buf == NULL) {
5185 		(void) close(fd);
5186 		return (ENOMEM);
5187 	}
5188 
5189 	while (offset < size) {
5190 		readsize = MIN(size - offset, 1 << 20);
5191 		err = dmu_read(os, srcobj, offset, readsize, buf, 0);
5192 		if (err != 0) {
5193 			(void) printf("got error %u from dmu_read\n", err);
5194 			kmem_free(buf, oursize);
5195 			(void) close(fd);
5196 			return (err);
5197 		}
5198 		if (dump_opt['v'] > 3) {
5199 			(void) printf("Read offset=%" PRIu64 " size=%" PRIu64
5200 			    " error=%d\n", offset, readsize, err);
5201 		}
5202 
5203 		writesize = write(fd, buf, readsize);
5204 		if (writesize < 0) {
5205 			err = errno;
5206 			break;
5207 		} else if (writesize != readsize) {
5208 			/* Incomplete write */
5209 			(void) fprintf(stderr, "Short write, only wrote %llu of"
5210 			    " %" PRIu64 " bytes, exiting...\n",
5211 			    (u_longlong_t)writesize, readsize);
5212 			break;
5213 		}
5214 
5215 		offset += readsize;
5216 	}
5217 
5218 	(void) close(fd);
5219 
5220 	if (buf != NULL)
5221 		kmem_free(buf, oursize);
5222 
5223 	return (err);
5224 }
5225 
5226 static boolean_t
label_cksum_valid(vdev_label_t * label,uint64_t offset)5227 label_cksum_valid(vdev_label_t *label, uint64_t offset)
5228 {
5229 	zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
5230 	zio_cksum_t expected_cksum;
5231 	zio_cksum_t actual_cksum;
5232 	zio_cksum_t verifier;
5233 	zio_eck_t *eck;
5234 	int byteswap;
5235 
5236 	void *data = (char *)label + offsetof(vdev_label_t, vl_vdev_phys);
5237 	eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1;
5238 
5239 	offset += offsetof(vdev_label_t, vl_vdev_phys);
5240 	ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
5241 
5242 	byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
5243 	if (byteswap)
5244 		byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
5245 
5246 	expected_cksum = eck->zec_cksum;
5247 	eck->zec_cksum = verifier;
5248 
5249 	abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE);
5250 	ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum);
5251 	abd_free(abd);
5252 
5253 	if (byteswap)
5254 		byteswap_uint64_array(&expected_cksum, sizeof (zio_cksum_t));
5255 
5256 	if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
5257 		return (B_TRUE);
5258 
5259 	return (B_FALSE);
5260 }
5261 
5262 static int
dump_label(const char * dev)5263 dump_label(const char *dev)
5264 {
5265 	char path[MAXPATHLEN];
5266 	zdb_label_t labels[VDEV_LABELS] = {{{{0}}}};
5267 	uint64_t psize, ashift, l2cache;
5268 	struct stat64 statbuf;
5269 	boolean_t config_found = B_FALSE;
5270 	boolean_t error = B_FALSE;
5271 	boolean_t read_l2arc_header = B_FALSE;
5272 	avl_tree_t config_tree;
5273 	avl_tree_t uberblock_tree;
5274 	void *node, *cookie;
5275 	int fd;
5276 
5277 	/*
5278 	 * Check if we were given absolute path and use it as is.
5279 	 * Otherwise if the provided vdev name doesn't point to a file,
5280 	 * try prepending expected disk paths and partition numbers.
5281 	 */
5282 	(void) strlcpy(path, dev, sizeof (path));
5283 	if (dev[0] != '/' && stat64(path, &statbuf) != 0) {
5284 		int error;
5285 
5286 		error = zfs_resolve_shortname(dev, path, MAXPATHLEN);
5287 		if (error == 0 && zfs_dev_is_whole_disk(path)) {
5288 			if (zfs_append_partition(path, MAXPATHLEN) == -1)
5289 				error = ENOENT;
5290 		}
5291 
5292 		if (error || (stat64(path, &statbuf) != 0)) {
5293 			(void) printf("failed to find device %s, try "
5294 			    "specifying absolute path instead\n", dev);
5295 			return (1);
5296 		}
5297 	}
5298 
5299 	if ((fd = open64(path, O_RDONLY)) < 0) {
5300 		(void) printf("cannot open '%s': %s\n", path, strerror(errno));
5301 		zdb_exit(1);
5302 	}
5303 
5304 	if (fstat64_blk(fd, &statbuf) != 0) {
5305 		(void) printf("failed to stat '%s': %s\n", path,
5306 		    strerror(errno));
5307 		(void) close(fd);
5308 		zdb_exit(1);
5309 	}
5310 
5311 	if (S_ISBLK(statbuf.st_mode) && zfs_dev_flush(fd) != 0)
5312 		(void) printf("failed to invalidate cache '%s' : %s\n", path,
5313 		    strerror(errno));
5314 
5315 	avl_create(&config_tree, cksum_record_compare,
5316 	    sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5317 	avl_create(&uberblock_tree, cksum_record_compare,
5318 	    sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5319 
5320 	psize = statbuf.st_size;
5321 	psize = P2ALIGN_TYPED(psize, sizeof (vdev_label_t), uint64_t);
5322 	ashift = SPA_MINBLOCKSHIFT;
5323 
5324 	/*
5325 	 * 1. Read the label from disk
5326 	 * 2. Verify label cksum
5327 	 * 3. Unpack the configuration and insert in config tree.
5328 	 * 4. Traverse all uberblocks and insert in uberblock tree.
5329 	 */
5330 	for (int l = 0; l < VDEV_LABELS; l++) {
5331 		zdb_label_t *label = &labels[l];
5332 		char *buf = label->label.vl_vdev_phys.vp_nvlist;
5333 		size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5334 		nvlist_t *config;
5335 		cksum_record_t *rec;
5336 		zio_cksum_t cksum;
5337 		vdev_t vd;
5338 
5339 		label->label_offset = vdev_label_offset(psize, l, 0);
5340 
5341 		if (pread64(fd, &label->label, sizeof (label->label),
5342 		    label->label_offset) != sizeof (label->label)) {
5343 			if (!dump_opt['q'])
5344 				(void) printf("failed to read label %d\n", l);
5345 			label->read_failed = B_TRUE;
5346 			error = B_TRUE;
5347 			continue;
5348 		}
5349 
5350 		label->read_failed = B_FALSE;
5351 		label->cksum_valid = label_cksum_valid(&label->label,
5352 		    label->label_offset);
5353 
5354 		if (nvlist_unpack(buf, buflen, &config, 0) == 0) {
5355 			nvlist_t *vdev_tree = NULL;
5356 			size_t size;
5357 
5358 			if ((nvlist_lookup_nvlist(config,
5359 			    ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) ||
5360 			    (nvlist_lookup_uint64(vdev_tree,
5361 			    ZPOOL_CONFIG_ASHIFT, &ashift) != 0))
5362 				ashift = SPA_MINBLOCKSHIFT;
5363 
5364 			if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0)
5365 				size = buflen;
5366 
5367 			/* If the device is a cache device read the header. */
5368 			if (!read_l2arc_header) {
5369 				if (nvlist_lookup_uint64(config,
5370 				    ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 &&
5371 				    l2cache == POOL_STATE_L2CACHE) {
5372 					read_l2arc_header = B_TRUE;
5373 				}
5374 			}
5375 
5376 			fletcher_4_native_varsize(buf, size, &cksum);
5377 			rec = cksum_record_insert(&config_tree, &cksum, l);
5378 
5379 			label->config = rec;
5380 			label->config_nv = config;
5381 			config_found = B_TRUE;
5382 		} else {
5383 			error = B_TRUE;
5384 		}
5385 
5386 		vd.vdev_ashift = ashift;
5387 		vd.vdev_top = &vd;
5388 
5389 		for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5390 			uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5391 			uberblock_t *ub = (void *)((char *)label + uoff);
5392 
5393 			if (uberblock_verify(ub))
5394 				continue;
5395 
5396 			fletcher_4_native_varsize(ub, sizeof (*ub), &cksum);
5397 			rec = cksum_record_insert(&uberblock_tree, &cksum, l);
5398 
5399 			label->uberblocks[i] = rec;
5400 		}
5401 	}
5402 
5403 	/*
5404 	 * Dump the label and uberblocks.
5405 	 */
5406 	for (int l = 0; l < VDEV_LABELS; l++) {
5407 		zdb_label_t *label = &labels[l];
5408 		size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5409 
5410 		if (label->read_failed == B_TRUE)
5411 			continue;
5412 
5413 		if (label->config_nv) {
5414 			dump_config_from_label(label, buflen, l);
5415 		} else {
5416 			if (!dump_opt['q'])
5417 				(void) printf("failed to unpack label %d\n", l);
5418 		}
5419 
5420 		if (dump_opt['u'])
5421 			dump_label_uberblocks(label, ashift, l);
5422 
5423 		nvlist_free(label->config_nv);
5424 	}
5425 
5426 	/*
5427 	 * Dump the L2ARC header, if existent.
5428 	 */
5429 	if (read_l2arc_header)
5430 		error |= dump_l2arc_header(fd);
5431 
5432 	cookie = NULL;
5433 	while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL)
5434 		umem_free(node, sizeof (cksum_record_t));
5435 
5436 	cookie = NULL;
5437 	while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL)
5438 		umem_free(node, sizeof (cksum_record_t));
5439 
5440 	avl_destroy(&config_tree);
5441 	avl_destroy(&uberblock_tree);
5442 
5443 	(void) close(fd);
5444 
5445 	return (config_found == B_FALSE ? 2 :
5446 	    (error == B_TRUE ? 1 : 0));
5447 }
5448 
5449 static uint64_t dataset_feature_count[SPA_FEATURES];
5450 static uint64_t global_feature_count[SPA_FEATURES];
5451 static uint64_t remap_deadlist_count = 0;
5452 
5453 static int
dump_one_objset(const char * dsname,void * arg)5454 dump_one_objset(const char *dsname, void *arg)
5455 {
5456 	(void) arg;
5457 	int error;
5458 	objset_t *os;
5459 	spa_feature_t f;
5460 
5461 	error = open_objset(dsname, FTAG, &os);
5462 	if (error != 0)
5463 		return (0);
5464 
5465 	for (f = 0; f < SPA_FEATURES; f++) {
5466 		if (!dsl_dataset_feature_is_active(dmu_objset_ds(os), f))
5467 			continue;
5468 		ASSERT(spa_feature_table[f].fi_flags &
5469 		    ZFEATURE_FLAG_PER_DATASET);
5470 		dataset_feature_count[f]++;
5471 	}
5472 
5473 	if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os))) {
5474 		remap_deadlist_count++;
5475 	}
5476 
5477 	for (dsl_bookmark_node_t *dbn =
5478 	    avl_first(&dmu_objset_ds(os)->ds_bookmarks); dbn != NULL;
5479 	    dbn = AVL_NEXT(&dmu_objset_ds(os)->ds_bookmarks, dbn)) {
5480 		mos_obj_refd(dbn->dbn_phys.zbm_redaction_obj);
5481 		if (dbn->dbn_phys.zbm_redaction_obj != 0) {
5482 			global_feature_count[
5483 			    SPA_FEATURE_REDACTION_BOOKMARKS]++;
5484 			objset_t *mos = os->os_spa->spa_meta_objset;
5485 			dnode_t *rl;
5486 			VERIFY0(dnode_hold(mos,
5487 			    dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl));
5488 			if (rl->dn_have_spill) {
5489 				global_feature_count[
5490 				    SPA_FEATURE_REDACTION_LIST_SPILL]++;
5491 			}
5492 		}
5493 		if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)
5494 			global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN]++;
5495 	}
5496 
5497 	if (dsl_deadlist_is_open(&dmu_objset_ds(os)->ds_dir->dd_livelist) &&
5498 	    !dmu_objset_is_snapshot(os)) {
5499 		global_feature_count[SPA_FEATURE_LIVELIST]++;
5500 	}
5501 
5502 	dump_objset(os);
5503 	close_objset(os, FTAG);
5504 	fuid_table_destroy();
5505 	return (0);
5506 }
5507 
5508 /*
5509  * Block statistics.
5510  */
5511 #define	PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5512 typedef struct zdb_blkstats {
5513 	uint64_t zb_asize;
5514 	uint64_t zb_lsize;
5515 	uint64_t zb_psize;
5516 	uint64_t zb_count;
5517 	uint64_t zb_gangs;
5518 	uint64_t zb_ditto_samevdev;
5519 	uint64_t zb_ditto_same_ms;
5520 	uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
5521 } zdb_blkstats_t;
5522 
5523 /*
5524  * Extended object types to report deferred frees and dedup auto-ditto blocks.
5525  */
5526 #define	ZDB_OT_DEFERRED	(DMU_OT_NUMTYPES + 0)
5527 #define	ZDB_OT_DITTO	(DMU_OT_NUMTYPES + 1)
5528 #define	ZDB_OT_OTHER	(DMU_OT_NUMTYPES + 2)
5529 #define	ZDB_OT_TOTAL	(DMU_OT_NUMTYPES + 3)
5530 
5531 static const char *zdb_ot_extname[] = {
5532 	"deferred free",
5533 	"dedup ditto",
5534 	"other",
5535 	"Total",
5536 };
5537 
5538 #define	ZB_TOTAL	DN_MAX_LEVELS
5539 #define	SPA_MAX_FOR_16M	(SPA_MAXBLOCKSHIFT+1)
5540 
5541 typedef struct zdb_brt_entry {
5542 	dva_t		zbre_dva;
5543 	uint64_t	zbre_refcount;
5544 	avl_node_t	zbre_node;
5545 } zdb_brt_entry_t;
5546 
5547 typedef struct zdb_cb {
5548 	zdb_blkstats_t	zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
5549 	uint64_t	zcb_removing_size;
5550 	uint64_t	zcb_checkpoint_size;
5551 	uint64_t	zcb_dedup_asize;
5552 	uint64_t	zcb_dedup_blocks;
5553 	uint64_t	zcb_clone_asize;
5554 	uint64_t	zcb_clone_blocks;
5555 	uint64_t	zcb_psize_count[SPA_MAX_FOR_16M];
5556 	uint64_t	zcb_lsize_count[SPA_MAX_FOR_16M];
5557 	uint64_t	zcb_asize_count[SPA_MAX_FOR_16M];
5558 	uint64_t	zcb_psize_len[SPA_MAX_FOR_16M];
5559 	uint64_t	zcb_lsize_len[SPA_MAX_FOR_16M];
5560 	uint64_t	zcb_asize_len[SPA_MAX_FOR_16M];
5561 	uint64_t	zcb_psize_total;
5562 	uint64_t	zcb_lsize_total;
5563 	uint64_t	zcb_asize_total;
5564 	uint64_t	zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
5565 	uint64_t	zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
5566 	    [BPE_PAYLOAD_SIZE + 1];
5567 	uint64_t	zcb_start;
5568 	hrtime_t	zcb_lastprint;
5569 	uint64_t	zcb_totalasize;
5570 	uint64_t	zcb_errors[256];
5571 	int		zcb_readfails;
5572 	int		zcb_haderrors;
5573 	spa_t		*zcb_spa;
5574 	uint32_t	**zcb_vd_obsolete_counts;
5575 	avl_tree_t	zcb_brt;
5576 	boolean_t	zcb_brt_is_active;
5577 } zdb_cb_t;
5578 
5579 /* test if two DVA offsets from same vdev are within the same metaslab */
5580 static boolean_t
same_metaslab(spa_t * spa,uint64_t vdev,uint64_t off1,uint64_t off2)5581 same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2)
5582 {
5583 	vdev_t *vd = vdev_lookup_top(spa, vdev);
5584 	uint64_t ms_shift = vd->vdev_ms_shift;
5585 
5586 	return ((off1 >> ms_shift) == (off2 >> ms_shift));
5587 }
5588 
5589 /*
5590  * Used to simplify reporting of the histogram data.
5591  */
5592 typedef struct one_histo {
5593 	const char *name;
5594 	uint64_t *count;
5595 	uint64_t *len;
5596 	uint64_t cumulative;
5597 } one_histo_t;
5598 
5599 /*
5600  * The number of separate histograms processed for psize, lsize and asize.
5601  */
5602 #define	NUM_HISTO 3
5603 
5604 /*
5605  * This routine will create a fixed column size output of three different
5606  * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5607  * the count, length and cumulative length of the psize, lsize and
5608  * asize blocks.
5609  *
5610  * All three types of blocks are listed on a single line
5611  *
5612  * By default the table is printed in nicenumber format (e.g. 123K) but
5613  * if the '-P' parameter is specified then the full raw number (parseable)
5614  * is printed out.
5615  */
5616 static void
dump_size_histograms(zdb_cb_t * zcb)5617 dump_size_histograms(zdb_cb_t *zcb)
5618 {
5619 	/*
5620 	 * A temporary buffer that allows us to convert a number into
5621 	 * a string using zdb_nicenumber to allow either raw or human
5622 	 * readable numbers to be output.
5623 	 */
5624 	char numbuf[32];
5625 
5626 	/*
5627 	 * Define titles which are used in the headers of the tables
5628 	 * printed by this routine.
5629 	 */
5630 	const char blocksize_title1[] = "block";
5631 	const char blocksize_title2[] = "size";
5632 	const char count_title[] = "Count";
5633 	const char length_title[] = "Size";
5634 	const char cumulative_title[] = "Cum.";
5635 
5636 	/*
5637 	 * Setup the histogram arrays (psize, lsize, and asize).
5638 	 */
5639 	one_histo_t parm_histo[NUM_HISTO];
5640 
5641 	parm_histo[0].name = "psize";
5642 	parm_histo[0].count = zcb->zcb_psize_count;
5643 	parm_histo[0].len = zcb->zcb_psize_len;
5644 	parm_histo[0].cumulative = 0;
5645 
5646 	parm_histo[1].name = "lsize";
5647 	parm_histo[1].count = zcb->zcb_lsize_count;
5648 	parm_histo[1].len = zcb->zcb_lsize_len;
5649 	parm_histo[1].cumulative = 0;
5650 
5651 	parm_histo[2].name = "asize";
5652 	parm_histo[2].count = zcb->zcb_asize_count;
5653 	parm_histo[2].len = zcb->zcb_asize_len;
5654 	parm_histo[2].cumulative = 0;
5655 
5656 
5657 	(void) printf("\nBlock Size Histogram\n");
5658 	/*
5659 	 * Print the first line titles
5660 	 */
5661 	if (dump_opt['P'])
5662 		(void) printf("\n%s\t", blocksize_title1);
5663 	else
5664 		(void) printf("\n%7s   ", blocksize_title1);
5665 
5666 	for (int j = 0; j < NUM_HISTO; j++) {
5667 		if (dump_opt['P']) {
5668 			if (j < NUM_HISTO - 1) {
5669 				(void) printf("%s\t\t\t", parm_histo[j].name);
5670 			} else {
5671 				/* Don't print trailing spaces */
5672 				(void) printf("  %s", parm_histo[j].name);
5673 			}
5674 		} else {
5675 			if (j < NUM_HISTO - 1) {
5676 				/* Left aligned strings in the output */
5677 				(void) printf("%-7s              ",
5678 				    parm_histo[j].name);
5679 			} else {
5680 				/* Don't print trailing spaces */
5681 				(void) printf("%s", parm_histo[j].name);
5682 			}
5683 		}
5684 	}
5685 	(void) printf("\n");
5686 
5687 	/*
5688 	 * Print the second line titles
5689 	 */
5690 	if (dump_opt['P']) {
5691 		(void) printf("%s\t", blocksize_title2);
5692 	} else {
5693 		(void) printf("%7s ", blocksize_title2);
5694 	}
5695 
5696 	for (int i = 0; i < NUM_HISTO; i++) {
5697 		if (dump_opt['P']) {
5698 			(void) printf("%s\t%s\t%s\t",
5699 			    count_title, length_title, cumulative_title);
5700 		} else {
5701 			(void) printf("%7s%7s%7s",
5702 			    count_title, length_title, cumulative_title);
5703 		}
5704 	}
5705 	(void) printf("\n");
5706 
5707 	/*
5708 	 * Print the rows
5709 	 */
5710 	for (int i = SPA_MINBLOCKSHIFT; i < SPA_MAX_FOR_16M; i++) {
5711 
5712 		/*
5713 		 * Print the first column showing the blocksize
5714 		 */
5715 		zdb_nicenum((1ULL << i), numbuf, sizeof (numbuf));
5716 
5717 		if (dump_opt['P']) {
5718 			printf("%s", numbuf);
5719 		} else {
5720 			printf("%7s:", numbuf);
5721 		}
5722 
5723 		/*
5724 		 * Print the remaining set of 3 columns per size:
5725 		 * for psize, lsize and asize
5726 		 */
5727 		for (int j = 0; j < NUM_HISTO; j++) {
5728 			parm_histo[j].cumulative += parm_histo[j].len[i];
5729 
5730 			zdb_nicenum(parm_histo[j].count[i],
5731 			    numbuf, sizeof (numbuf));
5732 			if (dump_opt['P'])
5733 				(void) printf("\t%s", numbuf);
5734 			else
5735 				(void) printf("%7s", numbuf);
5736 
5737 			zdb_nicenum(parm_histo[j].len[i],
5738 			    numbuf, sizeof (numbuf));
5739 			if (dump_opt['P'])
5740 				(void) printf("\t%s", numbuf);
5741 			else
5742 				(void) printf("%7s", numbuf);
5743 
5744 			zdb_nicenum(parm_histo[j].cumulative,
5745 			    numbuf, sizeof (numbuf));
5746 			if (dump_opt['P'])
5747 				(void) printf("\t%s", numbuf);
5748 			else
5749 				(void) printf("%7s", numbuf);
5750 		}
5751 		(void) printf("\n");
5752 	}
5753 }
5754 
5755 static void
zdb_count_block(zdb_cb_t * zcb,zilog_t * zilog,const blkptr_t * bp,dmu_object_type_t type)5756 zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
5757     dmu_object_type_t type)
5758 {
5759 	int i;
5760 
5761 	ASSERT(type < ZDB_OT_TOTAL);
5762 
5763 	if (zilog && zil_bp_tree_add(zilog, bp) != 0)
5764 		return;
5765 
5766 	/*
5767 	 * This flag controls if we will issue a claim for the block while
5768 	 * counting it, to ensure that all blocks are referenced in space maps.
5769 	 * We don't issue claims if we're not doing leak tracking, because it's
5770 	 * expensive if the user isn't interested. We also don't claim the
5771 	 * second or later occurences of cloned or dedup'd blocks, because we
5772 	 * already claimed them the first time.
5773 	 */
5774 	boolean_t do_claim = !dump_opt['L'];
5775 
5776 	spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
5777 
5778 	blkptr_t tempbp;
5779 	if (BP_GET_DEDUP(bp)) {
5780 		/*
5781 		 * Dedup'd blocks are special. We need to count them, so we can
5782 		 * later uncount them when reporting leaked space, and we must
5783 		 * only claim them once.
5784 		 *
5785 		 * We use the existing dedup system to track what we've seen.
5786 		 * The first time we see a block, we do a ddt_lookup() to see
5787 		 * if it exists in the DDT. If we're doing leak tracking, we
5788 		 * claim the block at this time.
5789 		 *
5790 		 * Each time we see a block, we reduce the refcount in the
5791 		 * entry by one, and add to the size and count of dedup'd
5792 		 * blocks to report at the end.
5793 		 */
5794 
5795 		ddt_t *ddt = ddt_select(zcb->zcb_spa, bp);
5796 
5797 		ddt_enter(ddt);
5798 
5799 		/*
5800 		 * Find the block. This will create the entry in memory, but
5801 		 * we'll know if that happened by its refcount.
5802 		 */
5803 		ddt_entry_t *dde = ddt_lookup(ddt, bp);
5804 
5805 		/*
5806 		 * ddt_lookup() can return NULL if this block didn't exist
5807 		 * in the DDT and creating it would take the DDT over its
5808 		 * quota. Since we got the block from disk, it must exist in
5809 		 * the DDT, so this can't happen. However, when unique entries
5810 		 * are pruned, the dedup bit can be set with no corresponding
5811 		 * entry in the DDT.
5812 		 */
5813 		if (dde == NULL) {
5814 			ddt_exit(ddt);
5815 			goto skipped;
5816 		}
5817 
5818 		/* Get the phys for this variant */
5819 		ddt_phys_variant_t v = ddt_phys_select(ddt, dde, bp);
5820 
5821 		/*
5822 		 * This entry may have multiple sets of DVAs. We must claim
5823 		 * each set the first time we see them in a real block on disk,
5824 		 * or count them on subsequent occurences. We don't have a
5825 		 * convenient way to track the first time we see each variant,
5826 		 * so we repurpose dde_io as a set of "seen" flag bits. We can
5827 		 * do this safely in zdb because it never writes, so it will
5828 		 * never have a writing zio for this block in that pointer.
5829 		 */
5830 		boolean_t seen = !!(((uintptr_t)dde->dde_io) & (1 << v));
5831 		if (!seen)
5832 			dde->dde_io =
5833 			    (void *)(((uintptr_t)dde->dde_io) | (1 << v));
5834 
5835 		/* Consume a reference for this block. */
5836 		if (ddt_phys_total_refcnt(ddt, dde->dde_phys) > 0)
5837 			ddt_phys_decref(dde->dde_phys, v);
5838 
5839 		/*
5840 		 * If this entry has a single flat phys, it may have been
5841 		 * extended with additional DVAs at some time in its life.
5842 		 * This block might be from before it was fully extended, and
5843 		 * so have fewer DVAs.
5844 		 *
5845 		 * If this is the first time we've seen this block, and we
5846 		 * claimed it as-is, then we would miss the claim on some
5847 		 * number of DVAs, which would then be seen as leaked.
5848 		 *
5849 		 * In all cases, if we've had fewer DVAs, then the asize would
5850 		 * be too small, and would lead to the pool apparently using
5851 		 * more space than allocated.
5852 		 *
5853 		 * To handle this, we copy the canonical set of DVAs from the
5854 		 * entry back to the block pointer before we claim it.
5855 		 */
5856 		if (v == DDT_PHYS_FLAT) {
5857 			ASSERT3U(BP_GET_BIRTH(bp), ==,
5858 			    ddt_phys_birth(dde->dde_phys, v));
5859 			tempbp = *bp;
5860 			ddt_bp_fill(dde->dde_phys, v, &tempbp,
5861 			    BP_GET_BIRTH(bp));
5862 			bp = &tempbp;
5863 		}
5864 
5865 		if (seen) {
5866 			/*
5867 			 * The second or later time we see this block,
5868 			 * it's a duplicate and we count it.
5869 			 */
5870 			zcb->zcb_dedup_asize += BP_GET_ASIZE(bp);
5871 			zcb->zcb_dedup_blocks++;
5872 
5873 			/* Already claimed, don't do it again. */
5874 			do_claim = B_FALSE;
5875 		}
5876 
5877 		ddt_exit(ddt);
5878 	} else if (zcb->zcb_brt_is_active &&
5879 	    brt_maybe_exists(zcb->zcb_spa, bp)) {
5880 		/*
5881 		 * Cloned blocks are special. We need to count them, so we can
5882 		 * later uncount them when reporting leaked space, and we must
5883 		 * only claim them once.
5884 		 *
5885 		 * To do this, we keep our own in-memory BRT. For each block
5886 		 * we haven't seen before, we look it up in the real BRT and
5887 		 * if its there, we note it and its refcount then proceed as
5888 		 * normal. If we see the block again, we count it as a clone
5889 		 * and then give it no further consideration.
5890 		 */
5891 		zdb_brt_entry_t zbre_search, *zbre;
5892 		avl_index_t where;
5893 
5894 		zbre_search.zbre_dva = bp->blk_dva[0];
5895 		zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where);
5896 		if (zbre == NULL) {
5897 			/* Not seen before; track it */
5898 			uint64_t refcnt =
5899 			    brt_entry_get_refcount(zcb->zcb_spa, bp);
5900 			if (refcnt > 0) {
5901 				zbre = umem_zalloc(sizeof (zdb_brt_entry_t),
5902 				    UMEM_NOFAIL);
5903 				zbre->zbre_dva = bp->blk_dva[0];
5904 				zbre->zbre_refcount = refcnt;
5905 				avl_insert(&zcb->zcb_brt, zbre, where);
5906 			}
5907 		} else  {
5908 			/*
5909 			 * Second or later occurrence, count it and take a
5910 			 * refcount.
5911 			 */
5912 			zcb->zcb_clone_asize += BP_GET_ASIZE(bp);
5913 			zcb->zcb_clone_blocks++;
5914 
5915 			zbre->zbre_refcount--;
5916 			if (zbre->zbre_refcount == 0) {
5917 				avl_remove(&zcb->zcb_brt, zbre);
5918 				umem_free(zbre, sizeof (zdb_brt_entry_t));
5919 			}
5920 
5921 			/* Already claimed, don't do it again. */
5922 			do_claim = B_FALSE;
5923 		}
5924 	}
5925 
5926 skipped:
5927 	for (i = 0; i < 4; i++) {
5928 		int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
5929 		int t = (i & 1) ? type : ZDB_OT_TOTAL;
5930 		int equal;
5931 		zdb_blkstats_t *zb = &zcb->zcb_type[l][t];
5932 
5933 		zb->zb_asize += BP_GET_ASIZE(bp);
5934 		zb->zb_lsize += BP_GET_LSIZE(bp);
5935 		zb->zb_psize += BP_GET_PSIZE(bp);
5936 		zb->zb_count++;
5937 
5938 		/*
5939 		 * The histogram is only big enough to record blocks up to
5940 		 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
5941 		 * "other", bucket.
5942 		 */
5943 		unsigned idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT;
5944 		idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1);
5945 		zb->zb_psize_histogram[idx]++;
5946 
5947 		zb->zb_gangs += BP_COUNT_GANG(bp);
5948 
5949 		switch (BP_GET_NDVAS(bp)) {
5950 		case 2:
5951 			if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5952 			    DVA_GET_VDEV(&bp->blk_dva[1])) {
5953 				zb->zb_ditto_samevdev++;
5954 
5955 				if (same_metaslab(zcb->zcb_spa,
5956 				    DVA_GET_VDEV(&bp->blk_dva[0]),
5957 				    DVA_GET_OFFSET(&bp->blk_dva[0]),
5958 				    DVA_GET_OFFSET(&bp->blk_dva[1])))
5959 					zb->zb_ditto_same_ms++;
5960 			}
5961 			break;
5962 		case 3:
5963 			equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5964 			    DVA_GET_VDEV(&bp->blk_dva[1])) +
5965 			    (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5966 			    DVA_GET_VDEV(&bp->blk_dva[2])) +
5967 			    (DVA_GET_VDEV(&bp->blk_dva[1]) ==
5968 			    DVA_GET_VDEV(&bp->blk_dva[2]));
5969 			if (equal != 0) {
5970 				zb->zb_ditto_samevdev++;
5971 
5972 				if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5973 				    DVA_GET_VDEV(&bp->blk_dva[1]) &&
5974 				    same_metaslab(zcb->zcb_spa,
5975 				    DVA_GET_VDEV(&bp->blk_dva[0]),
5976 				    DVA_GET_OFFSET(&bp->blk_dva[0]),
5977 				    DVA_GET_OFFSET(&bp->blk_dva[1])))
5978 					zb->zb_ditto_same_ms++;
5979 				else if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
5980 				    DVA_GET_VDEV(&bp->blk_dva[2]) &&
5981 				    same_metaslab(zcb->zcb_spa,
5982 				    DVA_GET_VDEV(&bp->blk_dva[0]),
5983 				    DVA_GET_OFFSET(&bp->blk_dva[0]),
5984 				    DVA_GET_OFFSET(&bp->blk_dva[2])))
5985 					zb->zb_ditto_same_ms++;
5986 				else if (DVA_GET_VDEV(&bp->blk_dva[1]) ==
5987 				    DVA_GET_VDEV(&bp->blk_dva[2]) &&
5988 				    same_metaslab(zcb->zcb_spa,
5989 				    DVA_GET_VDEV(&bp->blk_dva[1]),
5990 				    DVA_GET_OFFSET(&bp->blk_dva[1]),
5991 				    DVA_GET_OFFSET(&bp->blk_dva[2])))
5992 					zb->zb_ditto_same_ms++;
5993 			}
5994 			break;
5995 		}
5996 	}
5997 
5998 	spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
5999 
6000 	if (BP_IS_EMBEDDED(bp)) {
6001 		zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
6002 		zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
6003 		    [BPE_GET_PSIZE(bp)]++;
6004 		return;
6005 	}
6006 	/*
6007 	 * The binning histogram bins by powers of two up to
6008 	 * SPA_MAXBLOCKSIZE rather than creating bins for
6009 	 * every possible blocksize found in the pool.
6010 	 */
6011 	int bin = highbit64(BP_GET_PSIZE(bp)) - 1;
6012 
6013 	zcb->zcb_psize_count[bin]++;
6014 	zcb->zcb_psize_len[bin] += BP_GET_PSIZE(bp);
6015 	zcb->zcb_psize_total += BP_GET_PSIZE(bp);
6016 
6017 	bin = highbit64(BP_GET_LSIZE(bp)) - 1;
6018 
6019 	zcb->zcb_lsize_count[bin]++;
6020 	zcb->zcb_lsize_len[bin] += BP_GET_LSIZE(bp);
6021 	zcb->zcb_lsize_total += BP_GET_LSIZE(bp);
6022 
6023 	bin = highbit64(BP_GET_ASIZE(bp)) - 1;
6024 
6025 	zcb->zcb_asize_count[bin]++;
6026 	zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp);
6027 	zcb->zcb_asize_total += BP_GET_ASIZE(bp);
6028 
6029 	if (!do_claim)
6030 		return;
6031 
6032 	VERIFY0(zio_wait(zio_claim(NULL, zcb->zcb_spa,
6033 	    spa_min_claim_txg(zcb->zcb_spa), bp, NULL, NULL,
6034 	    ZIO_FLAG_CANFAIL)));
6035 }
6036 
6037 static void
zdb_blkptr_done(zio_t * zio)6038 zdb_blkptr_done(zio_t *zio)
6039 {
6040 	spa_t *spa = zio->io_spa;
6041 	blkptr_t *bp = zio->io_bp;
6042 	int ioerr = zio->io_error;
6043 	zdb_cb_t *zcb = zio->io_private;
6044 	zbookmark_phys_t *zb = &zio->io_bookmark;
6045 
6046 	mutex_enter(&spa->spa_scrub_lock);
6047 	spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
6048 	cv_broadcast(&spa->spa_scrub_io_cv);
6049 
6050 	if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
6051 		char blkbuf[BP_SPRINTF_LEN];
6052 
6053 		zcb->zcb_haderrors = 1;
6054 		zcb->zcb_errors[ioerr]++;
6055 
6056 		if (dump_opt['b'] >= 2)
6057 			snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6058 		else
6059 			blkbuf[0] = '\0';
6060 
6061 		(void) printf("zdb_blkptr_cb: "
6062 		    "Got error %d reading "
6063 		    "<%llu, %llu, %lld, %llx> %s -- skipping\n",
6064 		    ioerr,
6065 		    (u_longlong_t)zb->zb_objset,
6066 		    (u_longlong_t)zb->zb_object,
6067 		    (u_longlong_t)zb->zb_level,
6068 		    (u_longlong_t)zb->zb_blkid,
6069 		    blkbuf);
6070 	}
6071 	mutex_exit(&spa->spa_scrub_lock);
6072 
6073 	abd_free(zio->io_abd);
6074 }
6075 
6076 static int
zdb_blkptr_cb(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)6077 zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
6078     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
6079 {
6080 	zdb_cb_t *zcb = arg;
6081 	dmu_object_type_t type;
6082 	boolean_t is_metadata;
6083 
6084 	if (zb->zb_level == ZB_DNODE_LEVEL)
6085 		return (0);
6086 
6087 	if (dump_opt['b'] >= 5 && BP_GET_LOGICAL_BIRTH(bp) > 0) {
6088 		char blkbuf[BP_SPRINTF_LEN];
6089 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6090 		(void) printf("objset %llu object %llu "
6091 		    "level %lld offset 0x%llx %s\n",
6092 		    (u_longlong_t)zb->zb_objset,
6093 		    (u_longlong_t)zb->zb_object,
6094 		    (longlong_t)zb->zb_level,
6095 		    (u_longlong_t)blkid2offset(dnp, bp, zb),
6096 		    blkbuf);
6097 	}
6098 
6099 	if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
6100 		return (0);
6101 
6102 	type = BP_GET_TYPE(bp);
6103 
6104 	zdb_count_block(zcb, zilog, bp,
6105 	    (type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type);
6106 
6107 	is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
6108 
6109 	if (!BP_IS_EMBEDDED(bp) &&
6110 	    (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
6111 		size_t size = BP_GET_PSIZE(bp);
6112 		abd_t *abd = abd_alloc(size, B_FALSE);
6113 		int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
6114 
6115 		/* If it's an intent log block, failure is expected. */
6116 		if (zb->zb_level == ZB_ZIL_LEVEL)
6117 			flags |= ZIO_FLAG_SPECULATIVE;
6118 
6119 		mutex_enter(&spa->spa_scrub_lock);
6120 		while (spa->spa_load_verify_bytes > max_inflight_bytes)
6121 			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
6122 		spa->spa_load_verify_bytes += size;
6123 		mutex_exit(&spa->spa_scrub_lock);
6124 
6125 		zio_nowait(zio_read(NULL, spa, bp, abd, size,
6126 		    zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
6127 	}
6128 
6129 	zcb->zcb_readfails = 0;
6130 
6131 	/* only call gethrtime() every 100 blocks */
6132 	static int iters;
6133 	if (++iters > 100)
6134 		iters = 0;
6135 	else
6136 		return (0);
6137 
6138 	if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
6139 		uint64_t now = gethrtime();
6140 		char buf[10];
6141 		uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
6142 		uint64_t kb_per_sec =
6143 		    1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000));
6144 		uint64_t sec_remaining =
6145 		    (zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec;
6146 
6147 		/* make sure nicenum has enough space */
6148 		_Static_assert(sizeof (buf) >= NN_NUMBUF_SZ, "buf truncated");
6149 
6150 		zfs_nicebytes(bytes, buf, sizeof (buf));
6151 		(void) fprintf(stderr,
6152 		    "\r%5s completed (%4"PRIu64"MB/s) "
6153 		    "estimated time remaining: "
6154 		    "%"PRIu64"hr %02"PRIu64"min %02"PRIu64"sec        ",
6155 		    buf, kb_per_sec / 1024,
6156 		    sec_remaining / 60 / 60,
6157 		    sec_remaining / 60 % 60,
6158 		    sec_remaining % 60);
6159 
6160 		zcb->zcb_lastprint = now;
6161 	}
6162 
6163 	return (0);
6164 }
6165 
6166 static void
zdb_leak(void * arg,uint64_t start,uint64_t size)6167 zdb_leak(void *arg, uint64_t start, uint64_t size)
6168 {
6169 	vdev_t *vd = arg;
6170 
6171 	(void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
6172 	    (u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
6173 }
6174 
6175 static metaslab_ops_t zdb_metaslab_ops = {
6176 	NULL	/* alloc */
6177 };
6178 
6179 static int
load_unflushed_svr_segs_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6180 load_unflushed_svr_segs_cb(spa_t *spa, space_map_entry_t *sme,
6181     uint64_t txg, void *arg)
6182 {
6183 	spa_vdev_removal_t *svr = arg;
6184 
6185 	uint64_t offset = sme->sme_offset;
6186 	uint64_t size = sme->sme_run;
6187 
6188 	/* skip vdevs we don't care about */
6189 	if (sme->sme_vdev != svr->svr_vdev_id)
6190 		return (0);
6191 
6192 	vdev_t *vd = vdev_lookup_top(spa, sme->sme_vdev);
6193 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6194 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6195 
6196 	if (txg < metaslab_unflushed_txg(ms))
6197 		return (0);
6198 
6199 	if (sme->sme_type == SM_ALLOC)
6200 		range_tree_add(svr->svr_allocd_segs, offset, size);
6201 	else
6202 		range_tree_remove(svr->svr_allocd_segs, offset, size);
6203 
6204 	return (0);
6205 }
6206 
6207 static void
claim_segment_impl_cb(uint64_t inner_offset,vdev_t * vd,uint64_t offset,uint64_t size,void * arg)6208 claim_segment_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
6209     uint64_t size, void *arg)
6210 {
6211 	(void) inner_offset, (void) arg;
6212 
6213 	/*
6214 	 * This callback was called through a remap from
6215 	 * a device being removed. Therefore, the vdev that
6216 	 * this callback is applied to is a concrete
6217 	 * vdev.
6218 	 */
6219 	ASSERT(vdev_is_concrete(vd));
6220 
6221 	VERIFY0(metaslab_claim_impl(vd, offset, size,
6222 	    spa_min_claim_txg(vd->vdev_spa)));
6223 }
6224 
6225 static void
claim_segment_cb(void * arg,uint64_t offset,uint64_t size)6226 claim_segment_cb(void *arg, uint64_t offset, uint64_t size)
6227 {
6228 	vdev_t *vd = arg;
6229 
6230 	vdev_indirect_ops.vdev_op_remap(vd, offset, size,
6231 	    claim_segment_impl_cb, NULL);
6232 }
6233 
6234 /*
6235  * After accounting for all allocated blocks that are directly referenced,
6236  * we might have missed a reference to a block from a partially complete
6237  * (and thus unused) indirect mapping object. We perform a secondary pass
6238  * through the metaslabs we have already mapped and claim the destination
6239  * blocks.
6240  */
6241 static void
zdb_claim_removing(spa_t * spa,zdb_cb_t * zcb)6242 zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb)
6243 {
6244 	if (dump_opt['L'])
6245 		return;
6246 
6247 	if (spa->spa_vdev_removal == NULL)
6248 		return;
6249 
6250 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6251 
6252 	spa_vdev_removal_t *svr = spa->spa_vdev_removal;
6253 	vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id);
6254 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6255 
6256 	ASSERT0(range_tree_space(svr->svr_allocd_segs));
6257 
6258 	range_tree_t *allocs = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
6259 	for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) {
6260 		metaslab_t *msp = vd->vdev_ms[msi];
6261 
6262 		ASSERT0(range_tree_space(allocs));
6263 		if (msp->ms_sm != NULL)
6264 			VERIFY0(space_map_load(msp->ms_sm, allocs, SM_ALLOC));
6265 		range_tree_vacate(allocs, range_tree_add, svr->svr_allocd_segs);
6266 	}
6267 	range_tree_destroy(allocs);
6268 
6269 	iterate_through_spacemap_logs(spa, load_unflushed_svr_segs_cb, svr);
6270 
6271 	/*
6272 	 * Clear everything past what has been synced,
6273 	 * because we have not allocated mappings for
6274 	 * it yet.
6275 	 */
6276 	range_tree_clear(svr->svr_allocd_segs,
6277 	    vdev_indirect_mapping_max_offset(vim),
6278 	    vd->vdev_asize - vdev_indirect_mapping_max_offset(vim));
6279 
6280 	zcb->zcb_removing_size += range_tree_space(svr->svr_allocd_segs);
6281 	range_tree_vacate(svr->svr_allocd_segs, claim_segment_cb, vd);
6282 
6283 	spa_config_exit(spa, SCL_CONFIG, FTAG);
6284 }
6285 
6286 static int
increment_indirect_mapping_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6287 increment_indirect_mapping_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6288     dmu_tx_t *tx)
6289 {
6290 	(void) tx;
6291 	zdb_cb_t *zcb = arg;
6292 	spa_t *spa = zcb->zcb_spa;
6293 	vdev_t *vd;
6294 	const dva_t *dva = &bp->blk_dva[0];
6295 
6296 	ASSERT(!bp_freed);
6297 	ASSERT(!dump_opt['L']);
6298 	ASSERT3U(BP_GET_NDVAS(bp), ==, 1);
6299 
6300 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6301 	vd = vdev_lookup_top(zcb->zcb_spa, DVA_GET_VDEV(dva));
6302 	ASSERT3P(vd, !=, NULL);
6303 	spa_config_exit(spa, SCL_VDEV, FTAG);
6304 
6305 	ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0);
6306 	ASSERT3P(zcb->zcb_vd_obsolete_counts[vd->vdev_id], !=, NULL);
6307 
6308 	vdev_indirect_mapping_increment_obsolete_count(
6309 	    vd->vdev_indirect_mapping,
6310 	    DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva),
6311 	    zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6312 
6313 	return (0);
6314 }
6315 
6316 static uint32_t *
zdb_load_obsolete_counts(vdev_t * vd)6317 zdb_load_obsolete_counts(vdev_t *vd)
6318 {
6319 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6320 	spa_t *spa = vd->vdev_spa;
6321 	spa_condensing_indirect_phys_t *scip =
6322 	    &spa->spa_condensing_indirect_phys;
6323 	uint64_t obsolete_sm_object;
6324 	uint32_t *counts;
6325 
6326 	VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
6327 	EQUIV(obsolete_sm_object != 0, vd->vdev_obsolete_sm != NULL);
6328 	counts = vdev_indirect_mapping_load_obsolete_counts(vim);
6329 	if (vd->vdev_obsolete_sm != NULL) {
6330 		vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6331 		    vd->vdev_obsolete_sm);
6332 	}
6333 	if (scip->scip_vdev == vd->vdev_id &&
6334 	    scip->scip_prev_obsolete_sm_object != 0) {
6335 		space_map_t *prev_obsolete_sm = NULL;
6336 		VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset,
6337 		    scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0));
6338 		vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6339 		    prev_obsolete_sm);
6340 		space_map_close(prev_obsolete_sm);
6341 	}
6342 	return (counts);
6343 }
6344 
6345 typedef struct checkpoint_sm_exclude_entry_arg {
6346 	vdev_t *cseea_vd;
6347 	uint64_t cseea_checkpoint_size;
6348 } checkpoint_sm_exclude_entry_arg_t;
6349 
6350 static int
checkpoint_sm_exclude_entry_cb(space_map_entry_t * sme,void * arg)6351 checkpoint_sm_exclude_entry_cb(space_map_entry_t *sme, void *arg)
6352 {
6353 	checkpoint_sm_exclude_entry_arg_t *cseea = arg;
6354 	vdev_t *vd = cseea->cseea_vd;
6355 	metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
6356 	uint64_t end = sme->sme_offset + sme->sme_run;
6357 
6358 	ASSERT(sme->sme_type == SM_FREE);
6359 
6360 	/*
6361 	 * Since the vdev_checkpoint_sm exists in the vdev level
6362 	 * and the ms_sm space maps exist in the metaslab level,
6363 	 * an entry in the checkpoint space map could theoretically
6364 	 * cross the boundaries of the metaslab that it belongs.
6365 	 *
6366 	 * In reality, because of the way that we populate and
6367 	 * manipulate the checkpoint's space maps currently,
6368 	 * there shouldn't be any entries that cross metaslabs.
6369 	 * Hence the assertion below.
6370 	 *
6371 	 * That said, there is no fundamental requirement that
6372 	 * the checkpoint's space map entries should not cross
6373 	 * metaslab boundaries. So if needed we could add code
6374 	 * that handles metaslab-crossing segments in the future.
6375 	 */
6376 	VERIFY3U(sme->sme_offset, >=, ms->ms_start);
6377 	VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
6378 
6379 	/*
6380 	 * By removing the entry from the allocated segments we
6381 	 * also verify that the entry is there to begin with.
6382 	 */
6383 	mutex_enter(&ms->ms_lock);
6384 	range_tree_remove(ms->ms_allocatable, sme->sme_offset, sme->sme_run);
6385 	mutex_exit(&ms->ms_lock);
6386 
6387 	cseea->cseea_checkpoint_size += sme->sme_run;
6388 	return (0);
6389 }
6390 
6391 static void
zdb_leak_init_vdev_exclude_checkpoint(vdev_t * vd,zdb_cb_t * zcb)6392 zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_cb_t *zcb)
6393 {
6394 	spa_t *spa = vd->vdev_spa;
6395 	space_map_t *checkpoint_sm = NULL;
6396 	uint64_t checkpoint_sm_obj;
6397 
6398 	/*
6399 	 * If there is no vdev_top_zap, we are in a pool whose
6400 	 * version predates the pool checkpoint feature.
6401 	 */
6402 	if (vd->vdev_top_zap == 0)
6403 		return;
6404 
6405 	/*
6406 	 * If there is no reference of the vdev_checkpoint_sm in
6407 	 * the vdev_top_zap, then one of the following scenarios
6408 	 * is true:
6409 	 *
6410 	 * 1] There is no checkpoint
6411 	 * 2] There is a checkpoint, but no checkpointed blocks
6412 	 *    have been freed yet
6413 	 * 3] The current vdev is indirect
6414 	 *
6415 	 * In these cases we return immediately.
6416 	 */
6417 	if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
6418 	    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
6419 		return;
6420 
6421 	VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
6422 	    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1,
6423 	    &checkpoint_sm_obj));
6424 
6425 	checkpoint_sm_exclude_entry_arg_t cseea;
6426 	cseea.cseea_vd = vd;
6427 	cseea.cseea_checkpoint_size = 0;
6428 
6429 	VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
6430 	    checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
6431 
6432 	VERIFY0(space_map_iterate(checkpoint_sm,
6433 	    space_map_length(checkpoint_sm),
6434 	    checkpoint_sm_exclude_entry_cb, &cseea));
6435 	space_map_close(checkpoint_sm);
6436 
6437 	zcb->zcb_checkpoint_size += cseea.cseea_checkpoint_size;
6438 }
6439 
6440 static void
zdb_leak_init_exclude_checkpoint(spa_t * spa,zdb_cb_t * zcb)6441 zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb)
6442 {
6443 	ASSERT(!dump_opt['L']);
6444 
6445 	vdev_t *rvd = spa->spa_root_vdev;
6446 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6447 		ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id);
6448 		zdb_leak_init_vdev_exclude_checkpoint(rvd->vdev_child[c], zcb);
6449 	}
6450 }
6451 
6452 static int
count_unflushed_space_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6453 count_unflushed_space_cb(spa_t *spa, space_map_entry_t *sme,
6454     uint64_t txg, void *arg)
6455 {
6456 	int64_t *ualloc_space = arg;
6457 
6458 	uint64_t offset = sme->sme_offset;
6459 	uint64_t vdev_id = sme->sme_vdev;
6460 
6461 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6462 	if (!vdev_is_concrete(vd))
6463 		return (0);
6464 
6465 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6466 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6467 
6468 	if (txg < metaslab_unflushed_txg(ms))
6469 		return (0);
6470 
6471 	if (sme->sme_type == SM_ALLOC)
6472 		*ualloc_space += sme->sme_run;
6473 	else
6474 		*ualloc_space -= sme->sme_run;
6475 
6476 	return (0);
6477 }
6478 
6479 static int64_t
get_unflushed_alloc_space(spa_t * spa)6480 get_unflushed_alloc_space(spa_t *spa)
6481 {
6482 	if (dump_opt['L'])
6483 		return (0);
6484 
6485 	int64_t ualloc_space = 0;
6486 	iterate_through_spacemap_logs(spa, count_unflushed_space_cb,
6487 	    &ualloc_space);
6488 	return (ualloc_space);
6489 }
6490 
6491 static int
load_unflushed_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6492 load_unflushed_cb(spa_t *spa, space_map_entry_t *sme, uint64_t txg, void *arg)
6493 {
6494 	maptype_t *uic_maptype = arg;
6495 
6496 	uint64_t offset = sme->sme_offset;
6497 	uint64_t size = sme->sme_run;
6498 	uint64_t vdev_id = sme->sme_vdev;
6499 
6500 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6501 
6502 	/* skip indirect vdevs */
6503 	if (!vdev_is_concrete(vd))
6504 		return (0);
6505 
6506 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6507 
6508 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6509 	ASSERT(*uic_maptype == SM_ALLOC || *uic_maptype == SM_FREE);
6510 
6511 	if (txg < metaslab_unflushed_txg(ms))
6512 		return (0);
6513 
6514 	if (*uic_maptype == sme->sme_type)
6515 		range_tree_add(ms->ms_allocatable, offset, size);
6516 	else
6517 		range_tree_remove(ms->ms_allocatable, offset, size);
6518 
6519 	return (0);
6520 }
6521 
6522 static void
load_unflushed_to_ms_allocatables(spa_t * spa,maptype_t maptype)6523 load_unflushed_to_ms_allocatables(spa_t *spa, maptype_t maptype)
6524 {
6525 	iterate_through_spacemap_logs(spa, load_unflushed_cb, &maptype);
6526 }
6527 
6528 static void
load_concrete_ms_allocatable_trees(spa_t * spa,maptype_t maptype)6529 load_concrete_ms_allocatable_trees(spa_t *spa, maptype_t maptype)
6530 {
6531 	vdev_t *rvd = spa->spa_root_vdev;
6532 	for (uint64_t i = 0; i < rvd->vdev_children; i++) {
6533 		vdev_t *vd = rvd->vdev_child[i];
6534 
6535 		ASSERT3U(i, ==, vd->vdev_id);
6536 
6537 		if (vd->vdev_ops == &vdev_indirect_ops)
6538 			continue;
6539 
6540 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6541 			metaslab_t *msp = vd->vdev_ms[m];
6542 
6543 			(void) fprintf(stderr,
6544 			    "\rloading concrete vdev %llu, "
6545 			    "metaslab %llu of %llu ...",
6546 			    (longlong_t)vd->vdev_id,
6547 			    (longlong_t)msp->ms_id,
6548 			    (longlong_t)vd->vdev_ms_count);
6549 
6550 			mutex_enter(&msp->ms_lock);
6551 			range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6552 
6553 			/*
6554 			 * We don't want to spend the CPU manipulating the
6555 			 * size-ordered tree, so clear the range_tree ops.
6556 			 */
6557 			msp->ms_allocatable->rt_ops = NULL;
6558 
6559 			if (msp->ms_sm != NULL) {
6560 				VERIFY0(space_map_load(msp->ms_sm,
6561 				    msp->ms_allocatable, maptype));
6562 			}
6563 			if (!msp->ms_loaded)
6564 				msp->ms_loaded = B_TRUE;
6565 			mutex_exit(&msp->ms_lock);
6566 		}
6567 	}
6568 
6569 	load_unflushed_to_ms_allocatables(spa, maptype);
6570 }
6571 
6572 /*
6573  * vm_idxp is an in-out parameter which (for indirect vdevs) is the
6574  * index in vim_entries that has the first entry in this metaslab.
6575  * On return, it will be set to the first entry after this metaslab.
6576  */
6577 static void
load_indirect_ms_allocatable_tree(vdev_t * vd,metaslab_t * msp,uint64_t * vim_idxp)6578 load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab_t *msp,
6579     uint64_t *vim_idxp)
6580 {
6581 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6582 
6583 	mutex_enter(&msp->ms_lock);
6584 	range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6585 
6586 	/*
6587 	 * We don't want to spend the CPU manipulating the
6588 	 * size-ordered tree, so clear the range_tree ops.
6589 	 */
6590 	msp->ms_allocatable->rt_ops = NULL;
6591 
6592 	for (; *vim_idxp < vdev_indirect_mapping_num_entries(vim);
6593 	    (*vim_idxp)++) {
6594 		vdev_indirect_mapping_entry_phys_t *vimep =
6595 		    &vim->vim_entries[*vim_idxp];
6596 		uint64_t ent_offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6597 		uint64_t ent_len = DVA_GET_ASIZE(&vimep->vimep_dst);
6598 		ASSERT3U(ent_offset, >=, msp->ms_start);
6599 		if (ent_offset >= msp->ms_start + msp->ms_size)
6600 			break;
6601 
6602 		/*
6603 		 * Mappings do not cross metaslab boundaries,
6604 		 * because we create them by walking the metaslabs.
6605 		 */
6606 		ASSERT3U(ent_offset + ent_len, <=,
6607 		    msp->ms_start + msp->ms_size);
6608 		range_tree_add(msp->ms_allocatable, ent_offset, ent_len);
6609 	}
6610 
6611 	if (!msp->ms_loaded)
6612 		msp->ms_loaded = B_TRUE;
6613 	mutex_exit(&msp->ms_lock);
6614 }
6615 
6616 static void
zdb_leak_init_prepare_indirect_vdevs(spa_t * spa,zdb_cb_t * zcb)6617 zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb)
6618 {
6619 	ASSERT(!dump_opt['L']);
6620 
6621 	vdev_t *rvd = spa->spa_root_vdev;
6622 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6623 		vdev_t *vd = rvd->vdev_child[c];
6624 
6625 		ASSERT3U(c, ==, vd->vdev_id);
6626 
6627 		if (vd->vdev_ops != &vdev_indirect_ops)
6628 			continue;
6629 
6630 		/*
6631 		 * Note: we don't check for mapping leaks on
6632 		 * removing vdevs because their ms_allocatable's
6633 		 * are used to look for leaks in allocated space.
6634 		 */
6635 		zcb->zcb_vd_obsolete_counts[c] = zdb_load_obsolete_counts(vd);
6636 
6637 		/*
6638 		 * Normally, indirect vdevs don't have any
6639 		 * metaslabs.  We want to set them up for
6640 		 * zio_claim().
6641 		 */
6642 		vdev_metaslab_group_create(vd);
6643 		VERIFY0(vdev_metaslab_init(vd, 0));
6644 
6645 		vdev_indirect_mapping_t *vim __maybe_unused =
6646 		    vd->vdev_indirect_mapping;
6647 		uint64_t vim_idx = 0;
6648 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6649 
6650 			(void) fprintf(stderr,
6651 			    "\rloading indirect vdev %llu, "
6652 			    "metaslab %llu of %llu ...",
6653 			    (longlong_t)vd->vdev_id,
6654 			    (longlong_t)vd->vdev_ms[m]->ms_id,
6655 			    (longlong_t)vd->vdev_ms_count);
6656 
6657 			load_indirect_ms_allocatable_tree(vd, vd->vdev_ms[m],
6658 			    &vim_idx);
6659 		}
6660 		ASSERT3U(vim_idx, ==, vdev_indirect_mapping_num_entries(vim));
6661 	}
6662 }
6663 
6664 static void
zdb_leak_init(spa_t * spa,zdb_cb_t * zcb)6665 zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
6666 {
6667 	zcb->zcb_spa = spa;
6668 
6669 	if (dump_opt['L'])
6670 		return;
6671 
6672 	dsl_pool_t *dp = spa->spa_dsl_pool;
6673 	vdev_t *rvd = spa->spa_root_vdev;
6674 
6675 	/*
6676 	 * We are going to be changing the meaning of the metaslab's
6677 	 * ms_allocatable.  Ensure that the allocator doesn't try to
6678 	 * use the tree.
6679 	 */
6680 	spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
6681 	spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
6682 	spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6683 
6684 	zcb->zcb_vd_obsolete_counts =
6685 	    umem_zalloc(rvd->vdev_children * sizeof (uint32_t *),
6686 	    UMEM_NOFAIL);
6687 
6688 	/*
6689 	 * For leak detection, we overload the ms_allocatable trees
6690 	 * to contain allocated segments instead of free segments.
6691 	 * As a result, we can't use the normal metaslab_load/unload
6692 	 * interfaces.
6693 	 */
6694 	zdb_leak_init_prepare_indirect_vdevs(spa, zcb);
6695 	load_concrete_ms_allocatable_trees(spa, SM_ALLOC);
6696 
6697 	/*
6698 	 * On load_concrete_ms_allocatable_trees() we loaded all the
6699 	 * allocated entries from the ms_sm to the ms_allocatable for
6700 	 * each metaslab. If the pool has a checkpoint or is in the
6701 	 * middle of discarding a checkpoint, some of these blocks
6702 	 * may have been freed but their ms_sm may not have been
6703 	 * updated because they are referenced by the checkpoint. In
6704 	 * order to avoid false-positives during leak-detection, we
6705 	 * go through the vdev's checkpoint space map and exclude all
6706 	 * its entries from their relevant ms_allocatable.
6707 	 *
6708 	 * We also aggregate the space held by the checkpoint and add
6709 	 * it to zcb_checkpoint_size.
6710 	 *
6711 	 * Note that at this point we are also verifying that all the
6712 	 * entries on the checkpoint_sm are marked as allocated in
6713 	 * the ms_sm of their relevant metaslab.
6714 	 * [see comment in checkpoint_sm_exclude_entry_cb()]
6715 	 */
6716 	zdb_leak_init_exclude_checkpoint(spa, zcb);
6717 	ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa));
6718 
6719 	/* for cleaner progress output */
6720 	(void) fprintf(stderr, "\n");
6721 
6722 	if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
6723 		ASSERT(spa_feature_is_enabled(spa,
6724 		    SPA_FEATURE_DEVICE_REMOVAL));
6725 		(void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj,
6726 		    increment_indirect_mapping_cb, zcb, NULL);
6727 	}
6728 }
6729 
6730 static boolean_t
zdb_check_for_obsolete_leaks(vdev_t * vd,zdb_cb_t * zcb)6731 zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb)
6732 {
6733 	boolean_t leaks = B_FALSE;
6734 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6735 	uint64_t total_leaked = 0;
6736 	boolean_t are_precise = B_FALSE;
6737 
6738 	ASSERT(vim != NULL);
6739 
6740 	for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
6741 		vdev_indirect_mapping_entry_phys_t *vimep =
6742 		    &vim->vim_entries[i];
6743 		uint64_t obsolete_bytes = 0;
6744 		uint64_t offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6745 		metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6746 
6747 		/*
6748 		 * This is not very efficient but it's easy to
6749 		 * verify correctness.
6750 		 */
6751 		for (uint64_t inner_offset = 0;
6752 		    inner_offset < DVA_GET_ASIZE(&vimep->vimep_dst);
6753 		    inner_offset += 1ULL << vd->vdev_ashift) {
6754 			if (range_tree_contains(msp->ms_allocatable,
6755 			    offset + inner_offset, 1ULL << vd->vdev_ashift)) {
6756 				obsolete_bytes += 1ULL << vd->vdev_ashift;
6757 			}
6758 		}
6759 
6760 		int64_t bytes_leaked = obsolete_bytes -
6761 		    zcb->zcb_vd_obsolete_counts[vd->vdev_id][i];
6762 		ASSERT3U(DVA_GET_ASIZE(&vimep->vimep_dst), >=,
6763 		    zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]);
6764 
6765 		VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6766 		if (bytes_leaked != 0 && (are_precise || dump_opt['d'] >= 5)) {
6767 			(void) printf("obsolete indirect mapping count "
6768 			    "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
6769 			    (u_longlong_t)vd->vdev_id,
6770 			    (u_longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
6771 			    (u_longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
6772 			    (u_longlong_t)bytes_leaked);
6773 		}
6774 		total_leaked += ABS(bytes_leaked);
6775 	}
6776 
6777 	VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6778 	if (!are_precise && total_leaked > 0) {
6779 		int pct_leaked = total_leaked * 100 /
6780 		    vdev_indirect_mapping_bytes_mapped(vim);
6781 		(void) printf("cannot verify obsolete indirect mapping "
6782 		    "counts of vdev %llu because precise feature was not "
6783 		    "enabled when it was removed: %d%% (%llx bytes) of mapping"
6784 		    "unreferenced\n",
6785 		    (u_longlong_t)vd->vdev_id, pct_leaked,
6786 		    (u_longlong_t)total_leaked);
6787 	} else if (total_leaked > 0) {
6788 		(void) printf("obsolete indirect mapping count mismatch "
6789 		    "for vdev %llu -- %llx total bytes mismatched\n",
6790 		    (u_longlong_t)vd->vdev_id,
6791 		    (u_longlong_t)total_leaked);
6792 		leaks |= B_TRUE;
6793 	}
6794 
6795 	vdev_indirect_mapping_free_obsolete_counts(vim,
6796 	    zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6797 	zcb->zcb_vd_obsolete_counts[vd->vdev_id] = NULL;
6798 
6799 	return (leaks);
6800 }
6801 
6802 static boolean_t
zdb_leak_fini(spa_t * spa,zdb_cb_t * zcb)6803 zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb)
6804 {
6805 	if (dump_opt['L'])
6806 		return (B_FALSE);
6807 
6808 	boolean_t leaks = B_FALSE;
6809 	vdev_t *rvd = spa->spa_root_vdev;
6810 	for (unsigned c = 0; c < rvd->vdev_children; c++) {
6811 		vdev_t *vd = rvd->vdev_child[c];
6812 
6813 		if (zcb->zcb_vd_obsolete_counts[c] != NULL) {
6814 			leaks |= zdb_check_for_obsolete_leaks(vd, zcb);
6815 		}
6816 
6817 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6818 			metaslab_t *msp = vd->vdev_ms[m];
6819 			ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class ==
6820 			    spa_embedded_log_class(spa)) ?
6821 			    vd->vdev_log_mg : vd->vdev_mg);
6822 
6823 			/*
6824 			 * ms_allocatable has been overloaded
6825 			 * to contain allocated segments. Now that
6826 			 * we finished traversing all blocks, any
6827 			 * block that remains in the ms_allocatable
6828 			 * represents an allocated block that we
6829 			 * did not claim during the traversal.
6830 			 * Claimed blocks would have been removed
6831 			 * from the ms_allocatable.  For indirect
6832 			 * vdevs, space remaining in the tree
6833 			 * represents parts of the mapping that are
6834 			 * not referenced, which is not a bug.
6835 			 */
6836 			if (vd->vdev_ops == &vdev_indirect_ops) {
6837 				range_tree_vacate(msp->ms_allocatable,
6838 				    NULL, NULL);
6839 			} else {
6840 				range_tree_vacate(msp->ms_allocatable,
6841 				    zdb_leak, vd);
6842 			}
6843 			if (msp->ms_loaded) {
6844 				msp->ms_loaded = B_FALSE;
6845 			}
6846 		}
6847 	}
6848 
6849 	umem_free(zcb->zcb_vd_obsolete_counts,
6850 	    rvd->vdev_children * sizeof (uint32_t *));
6851 	zcb->zcb_vd_obsolete_counts = NULL;
6852 
6853 	return (leaks);
6854 }
6855 
6856 static int
count_block_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)6857 count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6858 {
6859 	(void) tx;
6860 	zdb_cb_t *zcb = arg;
6861 
6862 	if (dump_opt['b'] >= 5) {
6863 		char blkbuf[BP_SPRINTF_LEN];
6864 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6865 		(void) printf("[%s] %s\n",
6866 		    "deferred free", blkbuf);
6867 	}
6868 	zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED);
6869 	return (0);
6870 }
6871 
6872 /*
6873  * Iterate over livelists which have been destroyed by the user but
6874  * are still present in the MOS, waiting to be freed
6875  */
6876 static void
iterate_deleted_livelists(spa_t * spa,ll_iter_t func,void * arg)6877 iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg)
6878 {
6879 	objset_t *mos = spa->spa_meta_objset;
6880 	uint64_t zap_obj;
6881 	int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
6882 	    DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
6883 	if (err == ENOENT)
6884 		return;
6885 	ASSERT0(err);
6886 
6887 	zap_cursor_t zc;
6888 	zap_attribute_t *attrp = zap_attribute_alloc();
6889 	dsl_deadlist_t ll;
6890 	/* NULL out os prior to dsl_deadlist_open in case it's garbage */
6891 	ll.dl_os = NULL;
6892 	for (zap_cursor_init(&zc, mos, zap_obj);
6893 	    zap_cursor_retrieve(&zc, attrp) == 0;
6894 	    (void) zap_cursor_advance(&zc)) {
6895 		dsl_deadlist_open(&ll, mos, attrp->za_first_integer);
6896 		func(&ll, arg);
6897 		dsl_deadlist_close(&ll);
6898 	}
6899 	zap_cursor_fini(&zc);
6900 	zap_attribute_free(attrp);
6901 }
6902 
6903 static int
bpobj_count_block_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6904 bpobj_count_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6905     dmu_tx_t *tx)
6906 {
6907 	ASSERT(!bp_freed);
6908 	return (count_block_cb(arg, bp, tx));
6909 }
6910 
6911 static int
livelist_entry_count_blocks_cb(void * args,dsl_deadlist_entry_t * dle)6912 livelist_entry_count_blocks_cb(void *args, dsl_deadlist_entry_t *dle)
6913 {
6914 	zdb_cb_t *zbc = args;
6915 	bplist_t blks;
6916 	bplist_create(&blks);
6917 	/* determine which blocks have been alloc'd but not freed */
6918 	VERIFY0(dsl_process_sub_livelist(&dle->dle_bpobj, &blks, NULL, NULL));
6919 	/* count those blocks */
6920 	(void) bplist_iterate(&blks, count_block_cb, zbc, NULL);
6921 	bplist_destroy(&blks);
6922 	return (0);
6923 }
6924 
6925 static void
livelist_count_blocks(dsl_deadlist_t * ll,void * arg)6926 livelist_count_blocks(dsl_deadlist_t *ll, void *arg)
6927 {
6928 	dsl_deadlist_iterate(ll, livelist_entry_count_blocks_cb, arg);
6929 }
6930 
6931 /*
6932  * Count the blocks in the livelists that have been destroyed by the user
6933  * but haven't yet been freed.
6934  */
6935 static void
deleted_livelists_count_blocks(spa_t * spa,zdb_cb_t * zbc)6936 deleted_livelists_count_blocks(spa_t *spa, zdb_cb_t *zbc)
6937 {
6938 	iterate_deleted_livelists(spa, livelist_count_blocks, zbc);
6939 }
6940 
6941 static void
dump_livelist_cb(dsl_deadlist_t * ll,void * arg)6942 dump_livelist_cb(dsl_deadlist_t *ll, void *arg)
6943 {
6944 	ASSERT3P(arg, ==, NULL);
6945 	global_feature_count[SPA_FEATURE_LIVELIST]++;
6946 	dump_blkptr_list(ll, "Deleted Livelist");
6947 	dsl_deadlist_iterate(ll, sublivelist_verify_lightweight, NULL);
6948 }
6949 
6950 /*
6951  * Print out, register object references to, and increment feature counts for
6952  * livelists that have been destroyed by the user but haven't yet been freed.
6953  */
6954 static void
deleted_livelists_dump_mos(spa_t * spa)6955 deleted_livelists_dump_mos(spa_t *spa)
6956 {
6957 	uint64_t zap_obj;
6958 	objset_t *mos = spa->spa_meta_objset;
6959 	int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
6960 	    DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
6961 	if (err == ENOENT)
6962 		return;
6963 	mos_obj_refd(zap_obj);
6964 	iterate_deleted_livelists(spa, dump_livelist_cb, NULL);
6965 }
6966 
6967 static int
zdb_brt_entry_compare(const void * zcn1,const void * zcn2)6968 zdb_brt_entry_compare(const void *zcn1, const void *zcn2)
6969 {
6970 	const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva;
6971 	const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva;
6972 	int cmp;
6973 
6974 	cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
6975 	if (cmp == 0)
6976 		cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2));
6977 
6978 	return (cmp);
6979 }
6980 
6981 static int
dump_block_stats(spa_t * spa)6982 dump_block_stats(spa_t *spa)
6983 {
6984 	zdb_cb_t *zcb;
6985 	zdb_blkstats_t *zb, *tzb;
6986 	uint64_t norm_alloc, norm_space, total_alloc, total_found;
6987 	int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
6988 	    TRAVERSE_NO_DECRYPT | TRAVERSE_HARD;
6989 	boolean_t leaks = B_FALSE;
6990 	int e, c, err;
6991 	bp_embedded_type_t i;
6992 
6993 	ddt_prefetch_all(spa);
6994 
6995 	zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL);
6996 
6997 	if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
6998 		avl_create(&zcb->zcb_brt, zdb_brt_entry_compare,
6999 		    sizeof (zdb_brt_entry_t),
7000 		    offsetof(zdb_brt_entry_t, zbre_node));
7001 		zcb->zcb_brt_is_active = B_TRUE;
7002 	}
7003 
7004 	(void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
7005 	    (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
7006 	    (dump_opt['c'] == 1) ? "metadata " : "",
7007 	    dump_opt['c'] ? "checksums " : "",
7008 	    (dump_opt['c'] && !dump_opt['L']) ? "and verify " : "",
7009 	    !dump_opt['L'] ? "nothing leaked " : "");
7010 
7011 	/*
7012 	 * When leak detection is enabled we load all space maps as SM_ALLOC
7013 	 * maps, then traverse the pool claiming each block we discover. If
7014 	 * the pool is perfectly consistent, the segment trees will be empty
7015 	 * when we're done. Anything left over is a leak; any block we can't
7016 	 * claim (because it's not part of any space map) is a double
7017 	 * allocation, reference to a freed block, or an unclaimed log block.
7018 	 *
7019 	 * When leak detection is disabled (-L option) we still traverse the
7020 	 * pool claiming each block we discover, but we skip opening any space
7021 	 * maps.
7022 	 */
7023 	zdb_leak_init(spa, zcb);
7024 
7025 	/*
7026 	 * If there's a deferred-free bplist, process that first.
7027 	 */
7028 	(void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
7029 	    bpobj_count_block_cb, zcb, NULL);
7030 
7031 	if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
7032 		(void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
7033 		    bpobj_count_block_cb, zcb, NULL);
7034 	}
7035 
7036 	zdb_claim_removing(spa, zcb);
7037 
7038 	if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
7039 		VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,
7040 		    spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb,
7041 		    zcb, NULL));
7042 	}
7043 
7044 	deleted_livelists_count_blocks(spa, zcb);
7045 
7046 	if (dump_opt['c'] > 1)
7047 		flags |= TRAVERSE_PREFETCH_DATA;
7048 
7049 	zcb->zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
7050 	zcb->zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa));
7051 	zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
7052 	zcb->zcb_totalasize +=
7053 	    metaslab_class_get_alloc(spa_embedded_log_class(spa));
7054 	zcb->zcb_start = zcb->zcb_lastprint = gethrtime();
7055 	err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb);
7056 
7057 	/*
7058 	 * If we've traversed the data blocks then we need to wait for those
7059 	 * I/Os to complete. We leverage "The Godfather" zio to wait on
7060 	 * all async I/Os to complete.
7061 	 */
7062 	if (dump_opt['c']) {
7063 		for (c = 0; c < max_ncpus; c++) {
7064 			(void) zio_wait(spa->spa_async_zio_root[c]);
7065 			spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL,
7066 			    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
7067 			    ZIO_FLAG_GODFATHER);
7068 		}
7069 	}
7070 	ASSERT0(spa->spa_load_verify_bytes);
7071 
7072 	/*
7073 	 * Done after zio_wait() since zcb_haderrors is modified in
7074 	 * zdb_blkptr_done()
7075 	 */
7076 	zcb->zcb_haderrors |= err;
7077 
7078 	if (zcb->zcb_haderrors) {
7079 		(void) printf("\nError counts:\n\n");
7080 		(void) printf("\t%5s  %s\n", "errno", "count");
7081 		for (e = 0; e < 256; e++) {
7082 			if (zcb->zcb_errors[e] != 0) {
7083 				(void) printf("\t%5d  %llu\n",
7084 				    e, (u_longlong_t)zcb->zcb_errors[e]);
7085 			}
7086 		}
7087 	}
7088 
7089 	/*
7090 	 * Report any leaked segments.
7091 	 */
7092 	leaks |= zdb_leak_fini(spa, zcb);
7093 
7094 	tzb = &zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL];
7095 
7096 	norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7097 	norm_space = metaslab_class_get_space(spa_normal_class(spa));
7098 
7099 	total_alloc = norm_alloc +
7100 	    metaslab_class_get_alloc(spa_log_class(spa)) +
7101 	    metaslab_class_get_alloc(spa_embedded_log_class(spa)) +
7102 	    metaslab_class_get_alloc(spa_special_class(spa)) +
7103 	    metaslab_class_get_alloc(spa_dedup_class(spa)) +
7104 	    get_unflushed_alloc_space(spa);
7105 	total_found =
7106 	    tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize +
7107 	    zcb->zcb_removing_size + zcb->zcb_checkpoint_size;
7108 
7109 	if (total_found == total_alloc && !dump_opt['L']) {
7110 		(void) printf("\n\tNo leaks (block sum matches space"
7111 		    " maps exactly)\n");
7112 	} else if (!dump_opt['L']) {
7113 		(void) printf("block traversal size %llu != alloc %llu "
7114 		    "(%s %lld)\n",
7115 		    (u_longlong_t)total_found,
7116 		    (u_longlong_t)total_alloc,
7117 		    (dump_opt['L']) ? "unreachable" : "leaked",
7118 		    (longlong_t)(total_alloc - total_found));
7119 	}
7120 
7121 	if (tzb->zb_count == 0) {
7122 		umem_free(zcb, sizeof (zdb_cb_t));
7123 		return (2);
7124 	}
7125 
7126 	(void) printf("\n");
7127 	(void) printf("\t%-16s %14llu\n", "bp count:",
7128 	    (u_longlong_t)tzb->zb_count);
7129 	(void) printf("\t%-16s %14llu\n", "ganged count:",
7130 	    (longlong_t)tzb->zb_gangs);
7131 	(void) printf("\t%-16s %14llu      avg: %6llu\n", "bp logical:",
7132 	    (u_longlong_t)tzb->zb_lsize,
7133 	    (u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
7134 	(void) printf("\t%-16s %14llu      avg: %6llu     compression: %6.2f\n",
7135 	    "bp physical:", (u_longlong_t)tzb->zb_psize,
7136 	    (u_longlong_t)(tzb->zb_psize / tzb->zb_count),
7137 	    (double)tzb->zb_lsize / tzb->zb_psize);
7138 	(void) printf("\t%-16s %14llu      avg: %6llu     compression: %6.2f\n",
7139 	    "bp allocated:", (u_longlong_t)tzb->zb_asize,
7140 	    (u_longlong_t)(tzb->zb_asize / tzb->zb_count),
7141 	    (double)tzb->zb_lsize / tzb->zb_asize);
7142 	(void) printf("\t%-16s %14llu    ref>1: %6llu   deduplication: %6.2f\n",
7143 	    "bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize,
7144 	    (u_longlong_t)zcb->zcb_dedup_blocks,
7145 	    (double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0);
7146 	(void) printf("\t%-16s %14llu    count: %6llu\n",
7147 	    "bp cloned:", (u_longlong_t)zcb->zcb_clone_asize,
7148 	    (u_longlong_t)zcb->zcb_clone_blocks);
7149 	(void) printf("\t%-16s %14llu     used: %5.2f%%\n", "Normal class:",
7150 	    (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
7151 
7152 	if (spa_special_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7153 		uint64_t alloc = metaslab_class_get_alloc(
7154 		    spa_special_class(spa));
7155 		uint64_t space = metaslab_class_get_space(
7156 		    spa_special_class(spa));
7157 
7158 		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
7159 		    "Special class", (u_longlong_t)alloc,
7160 		    100.0 * alloc / space);
7161 	}
7162 
7163 	if (spa_dedup_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7164 		uint64_t alloc = metaslab_class_get_alloc(
7165 		    spa_dedup_class(spa));
7166 		uint64_t space = metaslab_class_get_space(
7167 		    spa_dedup_class(spa));
7168 
7169 		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
7170 		    "Dedup class", (u_longlong_t)alloc,
7171 		    100.0 * alloc / space);
7172 	}
7173 
7174 	if (spa_embedded_log_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7175 		uint64_t alloc = metaslab_class_get_alloc(
7176 		    spa_embedded_log_class(spa));
7177 		uint64_t space = metaslab_class_get_space(
7178 		    spa_embedded_log_class(spa));
7179 
7180 		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
7181 		    "Embedded log class", (u_longlong_t)alloc,
7182 		    100.0 * alloc / space);
7183 	}
7184 
7185 	for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
7186 		if (zcb->zcb_embedded_blocks[i] == 0)
7187 			continue;
7188 		(void) printf("\n");
7189 		(void) printf("\tadditional, non-pointer bps of type %u: "
7190 		    "%10llu\n",
7191 		    i, (u_longlong_t)zcb->zcb_embedded_blocks[i]);
7192 
7193 		if (dump_opt['b'] >= 3) {
7194 			(void) printf("\t number of (compressed) bytes:  "
7195 			    "number of bps\n");
7196 			dump_histogram(zcb->zcb_embedded_histogram[i],
7197 			    sizeof (zcb->zcb_embedded_histogram[i]) /
7198 			    sizeof (zcb->zcb_embedded_histogram[i][0]), 0);
7199 		}
7200 	}
7201 
7202 	if (tzb->zb_ditto_samevdev != 0) {
7203 		(void) printf("\tDittoed blocks on same vdev: %llu\n",
7204 		    (longlong_t)tzb->zb_ditto_samevdev);
7205 	}
7206 	if (tzb->zb_ditto_same_ms != 0) {
7207 		(void) printf("\tDittoed blocks in same metaslab: %llu\n",
7208 		    (longlong_t)tzb->zb_ditto_same_ms);
7209 	}
7210 
7211 	for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) {
7212 		vdev_t *vd = spa->spa_root_vdev->vdev_child[v];
7213 		vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7214 
7215 		if (vim == NULL) {
7216 			continue;
7217 		}
7218 
7219 		char mem[32];
7220 		zdb_nicenum(vdev_indirect_mapping_num_entries(vim),
7221 		    mem, vdev_indirect_mapping_size(vim));
7222 
7223 		(void) printf("\tindirect vdev id %llu has %llu segments "
7224 		    "(%s in memory)\n",
7225 		    (longlong_t)vd->vdev_id,
7226 		    (longlong_t)vdev_indirect_mapping_num_entries(vim), mem);
7227 	}
7228 
7229 	if (dump_opt['b'] >= 2) {
7230 		int l, t, level;
7231 		char csize[32], lsize[32], psize[32], asize[32];
7232 		char avg[32], gang[32];
7233 		(void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
7234 		    "\t  avg\t comp\t%%Total\tType\n");
7235 
7236 		zfs_blkstat_t *mdstats = umem_zalloc(sizeof (zfs_blkstat_t),
7237 		    UMEM_NOFAIL);
7238 
7239 		for (t = 0; t <= ZDB_OT_TOTAL; t++) {
7240 			const char *typename;
7241 
7242 			/* make sure nicenum has enough space */
7243 			_Static_assert(sizeof (csize) >= NN_NUMBUF_SZ,
7244 			    "csize truncated");
7245 			_Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ,
7246 			    "lsize truncated");
7247 			_Static_assert(sizeof (psize) >= NN_NUMBUF_SZ,
7248 			    "psize truncated");
7249 			_Static_assert(sizeof (asize) >= NN_NUMBUF_SZ,
7250 			    "asize truncated");
7251 			_Static_assert(sizeof (avg) >= NN_NUMBUF_SZ,
7252 			    "avg truncated");
7253 			_Static_assert(sizeof (gang) >= NN_NUMBUF_SZ,
7254 			    "gang truncated");
7255 
7256 			if (t < DMU_OT_NUMTYPES)
7257 				typename = dmu_ot[t].ot_name;
7258 			else
7259 				typename = zdb_ot_extname[t - DMU_OT_NUMTYPES];
7260 
7261 			if (zcb->zcb_type[ZB_TOTAL][t].zb_asize == 0) {
7262 				(void) printf("%6s\t%5s\t%5s\t%5s"
7263 				    "\t%5s\t%5s\t%6s\t%s\n",
7264 				    "-",
7265 				    "-",
7266 				    "-",
7267 				    "-",
7268 				    "-",
7269 				    "-",
7270 				    "-",
7271 				    typename);
7272 				continue;
7273 			}
7274 
7275 			for (l = ZB_TOTAL - 1; l >= -1; l--) {
7276 				level = (l == -1 ? ZB_TOTAL : l);
7277 				zb = &zcb->zcb_type[level][t];
7278 
7279 				if (zb->zb_asize == 0)
7280 					continue;
7281 
7282 				if (level != ZB_TOTAL && t < DMU_OT_NUMTYPES &&
7283 				    (level > 0 || DMU_OT_IS_METADATA(t))) {
7284 					mdstats->zb_count += zb->zb_count;
7285 					mdstats->zb_lsize += zb->zb_lsize;
7286 					mdstats->zb_psize += zb->zb_psize;
7287 					mdstats->zb_asize += zb->zb_asize;
7288 					mdstats->zb_gangs += zb->zb_gangs;
7289 				}
7290 
7291 				if (dump_opt['b'] < 3 && level != ZB_TOTAL)
7292 					continue;
7293 
7294 				if (level == 0 && zb->zb_asize ==
7295 				    zcb->zcb_type[ZB_TOTAL][t].zb_asize)
7296 					continue;
7297 
7298 				zdb_nicenum(zb->zb_count, csize,
7299 				    sizeof (csize));
7300 				zdb_nicenum(zb->zb_lsize, lsize,
7301 				    sizeof (lsize));
7302 				zdb_nicenum(zb->zb_psize, psize,
7303 				    sizeof (psize));
7304 				zdb_nicenum(zb->zb_asize, asize,
7305 				    sizeof (asize));
7306 				zdb_nicenum(zb->zb_asize / zb->zb_count, avg,
7307 				    sizeof (avg));
7308 				zdb_nicenum(zb->zb_gangs, gang, sizeof (gang));
7309 
7310 				(void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7311 				    "\t%5.2f\t%6.2f\t",
7312 				    csize, lsize, psize, asize, avg,
7313 				    (double)zb->zb_lsize / zb->zb_psize,
7314 				    100.0 * zb->zb_asize / tzb->zb_asize);
7315 
7316 				if (level == ZB_TOTAL)
7317 					(void) printf("%s\n", typename);
7318 				else
7319 					(void) printf("    L%d %s\n",
7320 					    level, typename);
7321 
7322 				if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) {
7323 					(void) printf("\t number of ganged "
7324 					    "blocks: %s\n", gang);
7325 				}
7326 
7327 				if (dump_opt['b'] >= 4) {
7328 					(void) printf("psize "
7329 					    "(in 512-byte sectors): "
7330 					    "number of blocks\n");
7331 					dump_histogram(zb->zb_psize_histogram,
7332 					    PSIZE_HISTO_SIZE, 0);
7333 				}
7334 			}
7335 		}
7336 		zdb_nicenum(mdstats->zb_count, csize,
7337 		    sizeof (csize));
7338 		zdb_nicenum(mdstats->zb_lsize, lsize,
7339 		    sizeof (lsize));
7340 		zdb_nicenum(mdstats->zb_psize, psize,
7341 		    sizeof (psize));
7342 		zdb_nicenum(mdstats->zb_asize, asize,
7343 		    sizeof (asize));
7344 		zdb_nicenum(mdstats->zb_asize / mdstats->zb_count, avg,
7345 		    sizeof (avg));
7346 		zdb_nicenum(mdstats->zb_gangs, gang, sizeof (gang));
7347 
7348 		(void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7349 		    "\t%5.2f\t%6.2f\t",
7350 		    csize, lsize, psize, asize, avg,
7351 		    (double)mdstats->zb_lsize / mdstats->zb_psize,
7352 		    100.0 * mdstats->zb_asize / tzb->zb_asize);
7353 		(void) printf("%s\n", "Metadata Total");
7354 
7355 		/* Output a table summarizing block sizes in the pool */
7356 		if (dump_opt['b'] >= 2) {
7357 			dump_size_histograms(zcb);
7358 		}
7359 
7360 		umem_free(mdstats, sizeof (zfs_blkstat_t));
7361 	}
7362 
7363 	(void) printf("\n");
7364 
7365 	if (leaks) {
7366 		umem_free(zcb, sizeof (zdb_cb_t));
7367 		return (2);
7368 	}
7369 
7370 	if (zcb->zcb_haderrors) {
7371 		umem_free(zcb, sizeof (zdb_cb_t));
7372 		return (3);
7373 	}
7374 
7375 	umem_free(zcb, sizeof (zdb_cb_t));
7376 	return (0);
7377 }
7378 
7379 typedef struct zdb_ddt_entry {
7380 	/* key must be first for ddt_key_compare */
7381 	ddt_key_t	zdde_key;
7382 	uint64_t	zdde_ref_blocks;
7383 	uint64_t	zdde_ref_lsize;
7384 	uint64_t	zdde_ref_psize;
7385 	uint64_t	zdde_ref_dsize;
7386 	avl_node_t	zdde_node;
7387 } zdb_ddt_entry_t;
7388 
7389 static int
zdb_ddt_add_cb(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)7390 zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
7391     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
7392 {
7393 	(void) zilog, (void) dnp;
7394 	avl_tree_t *t = arg;
7395 	avl_index_t where;
7396 	zdb_ddt_entry_t *zdde, zdde_search;
7397 
7398 	if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
7399 	    BP_IS_EMBEDDED(bp))
7400 		return (0);
7401 
7402 	if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
7403 		(void) printf("traversing objset %llu, %llu objects, "
7404 		    "%lu blocks so far\n",
7405 		    (u_longlong_t)zb->zb_objset,
7406 		    (u_longlong_t)BP_GET_FILL(bp),
7407 		    avl_numnodes(t));
7408 	}
7409 
7410 	if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF ||
7411 	    BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
7412 		return (0);
7413 
7414 	ddt_key_fill(&zdde_search.zdde_key, bp);
7415 
7416 	zdde = avl_find(t, &zdde_search, &where);
7417 
7418 	if (zdde == NULL) {
7419 		zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL);
7420 		zdde->zdde_key = zdde_search.zdde_key;
7421 		avl_insert(t, zdde, where);
7422 	}
7423 
7424 	zdde->zdde_ref_blocks += 1;
7425 	zdde->zdde_ref_lsize += BP_GET_LSIZE(bp);
7426 	zdde->zdde_ref_psize += BP_GET_PSIZE(bp);
7427 	zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp);
7428 
7429 	return (0);
7430 }
7431 
7432 static void
dump_simulated_ddt(spa_t * spa)7433 dump_simulated_ddt(spa_t *spa)
7434 {
7435 	avl_tree_t t;
7436 	void *cookie = NULL;
7437 	zdb_ddt_entry_t *zdde;
7438 	ddt_histogram_t ddh_total = {{{0}}};
7439 	ddt_stat_t dds_total = {0};
7440 
7441 	avl_create(&t, ddt_key_compare,
7442 	    sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
7443 
7444 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7445 
7446 	(void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7447 	    TRAVERSE_NO_DECRYPT, zdb_ddt_add_cb, &t);
7448 
7449 	spa_config_exit(spa, SCL_CONFIG, FTAG);
7450 
7451 	while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) {
7452 		uint64_t refcnt = zdde->zdde_ref_blocks;
7453 		ASSERT(refcnt != 0);
7454 
7455 		ddt_stat_t *dds = &ddh_total.ddh_stat[highbit64(refcnt) - 1];
7456 
7457 		dds->dds_blocks += zdde->zdde_ref_blocks / refcnt;
7458 		dds->dds_lsize += zdde->zdde_ref_lsize / refcnt;
7459 		dds->dds_psize += zdde->zdde_ref_psize / refcnt;
7460 		dds->dds_dsize += zdde->zdde_ref_dsize / refcnt;
7461 
7462 		dds->dds_ref_blocks += zdde->zdde_ref_blocks;
7463 		dds->dds_ref_lsize += zdde->zdde_ref_lsize;
7464 		dds->dds_ref_psize += zdde->zdde_ref_psize;
7465 		dds->dds_ref_dsize += zdde->zdde_ref_dsize;
7466 
7467 		umem_free(zdde, sizeof (*zdde));
7468 	}
7469 
7470 	avl_destroy(&t);
7471 
7472 	ddt_histogram_total(&dds_total, &ddh_total);
7473 
7474 	(void) printf("Simulated DDT histogram:\n");
7475 
7476 	zpool_dump_ddt(&dds_total, &ddh_total);
7477 
7478 	dump_dedup_ratio(&dds_total);
7479 }
7480 
7481 static int
verify_device_removal_feature_counts(spa_t * spa)7482 verify_device_removal_feature_counts(spa_t *spa)
7483 {
7484 	uint64_t dr_feature_refcount = 0;
7485 	uint64_t oc_feature_refcount = 0;
7486 	uint64_t indirect_vdev_count = 0;
7487 	uint64_t precise_vdev_count = 0;
7488 	uint64_t obsolete_counts_object_count = 0;
7489 	uint64_t obsolete_sm_count = 0;
7490 	uint64_t obsolete_counts_count = 0;
7491 	uint64_t scip_count = 0;
7492 	uint64_t obsolete_bpobj_count = 0;
7493 	int ret = 0;
7494 
7495 	spa_condensing_indirect_phys_t *scip =
7496 	    &spa->spa_condensing_indirect_phys;
7497 	if (scip->scip_next_mapping_object != 0) {
7498 		vdev_t *vd = spa->spa_root_vdev->vdev_child[scip->scip_vdev];
7499 		ASSERT(scip->scip_prev_obsolete_sm_object != 0);
7500 		ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
7501 
7502 		(void) printf("Condensing indirect vdev %llu: new mapping "
7503 		    "object %llu, prev obsolete sm %llu\n",
7504 		    (u_longlong_t)scip->scip_vdev,
7505 		    (u_longlong_t)scip->scip_next_mapping_object,
7506 		    (u_longlong_t)scip->scip_prev_obsolete_sm_object);
7507 		if (scip->scip_prev_obsolete_sm_object != 0) {
7508 			space_map_t *prev_obsolete_sm = NULL;
7509 			VERIFY0(space_map_open(&prev_obsolete_sm,
7510 			    spa->spa_meta_objset,
7511 			    scip->scip_prev_obsolete_sm_object,
7512 			    0, vd->vdev_asize, 0));
7513 			dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm);
7514 			(void) printf("\n");
7515 			space_map_close(prev_obsolete_sm);
7516 		}
7517 
7518 		scip_count += 2;
7519 	}
7520 
7521 	for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
7522 		vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
7523 		vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
7524 
7525 		if (vic->vic_mapping_object != 0) {
7526 			ASSERT(vd->vdev_ops == &vdev_indirect_ops ||
7527 			    vd->vdev_removing);
7528 			indirect_vdev_count++;
7529 
7530 			if (vd->vdev_indirect_mapping->vim_havecounts) {
7531 				obsolete_counts_count++;
7532 			}
7533 		}
7534 
7535 		boolean_t are_precise;
7536 		VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7537 		if (are_precise) {
7538 			ASSERT(vic->vic_mapping_object != 0);
7539 			precise_vdev_count++;
7540 		}
7541 
7542 		uint64_t obsolete_sm_object;
7543 		VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
7544 		if (obsolete_sm_object != 0) {
7545 			ASSERT(vic->vic_mapping_object != 0);
7546 			obsolete_sm_count++;
7547 		}
7548 	}
7549 
7550 	(void) feature_get_refcount(spa,
7551 	    &spa_feature_table[SPA_FEATURE_DEVICE_REMOVAL],
7552 	    &dr_feature_refcount);
7553 	(void) feature_get_refcount(spa,
7554 	    &spa_feature_table[SPA_FEATURE_OBSOLETE_COUNTS],
7555 	    &oc_feature_refcount);
7556 
7557 	if (dr_feature_refcount != indirect_vdev_count) {
7558 		ret = 1;
7559 		(void) printf("Number of indirect vdevs (%llu) " \
7560 		    "does not match feature count (%llu)\n",
7561 		    (u_longlong_t)indirect_vdev_count,
7562 		    (u_longlong_t)dr_feature_refcount);
7563 	} else {
7564 		(void) printf("Verified device_removal feature refcount " \
7565 		    "of %llu is correct\n",
7566 		    (u_longlong_t)dr_feature_refcount);
7567 	}
7568 
7569 	if (zap_contains(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT,
7570 	    DMU_POOL_OBSOLETE_BPOBJ) == 0) {
7571 		obsolete_bpobj_count++;
7572 	}
7573 
7574 
7575 	obsolete_counts_object_count = precise_vdev_count;
7576 	obsolete_counts_object_count += obsolete_sm_count;
7577 	obsolete_counts_object_count += obsolete_counts_count;
7578 	obsolete_counts_object_count += scip_count;
7579 	obsolete_counts_object_count += obsolete_bpobj_count;
7580 	obsolete_counts_object_count += remap_deadlist_count;
7581 
7582 	if (oc_feature_refcount != obsolete_counts_object_count) {
7583 		ret = 1;
7584 		(void) printf("Number of obsolete counts objects (%llu) " \
7585 		    "does not match feature count (%llu)\n",
7586 		    (u_longlong_t)obsolete_counts_object_count,
7587 		    (u_longlong_t)oc_feature_refcount);
7588 		(void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
7589 		    "ob:%llu rd:%llu\n",
7590 		    (u_longlong_t)precise_vdev_count,
7591 		    (u_longlong_t)obsolete_sm_count,
7592 		    (u_longlong_t)obsolete_counts_count,
7593 		    (u_longlong_t)scip_count,
7594 		    (u_longlong_t)obsolete_bpobj_count,
7595 		    (u_longlong_t)remap_deadlist_count);
7596 	} else {
7597 		(void) printf("Verified indirect_refcount feature refcount " \
7598 		    "of %llu is correct\n",
7599 		    (u_longlong_t)oc_feature_refcount);
7600 	}
7601 	return (ret);
7602 }
7603 
7604 static void
zdb_set_skip_mmp(char * target)7605 zdb_set_skip_mmp(char *target)
7606 {
7607 	spa_t *spa;
7608 
7609 	/*
7610 	 * Disable the activity check to allow examination of
7611 	 * active pools.
7612 	 */
7613 	mutex_enter(&spa_namespace_lock);
7614 	if ((spa = spa_lookup(target)) != NULL) {
7615 		spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
7616 	}
7617 	mutex_exit(&spa_namespace_lock);
7618 }
7619 
7620 #define	BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7621 /*
7622  * Import the checkpointed state of the pool specified by the target
7623  * parameter as readonly. The function also accepts a pool config
7624  * as an optional parameter, else it attempts to infer the config by
7625  * the name of the target pool.
7626  *
7627  * Note that the checkpointed state's pool name will be the name of
7628  * the original pool with the above suffix appended to it. In addition,
7629  * if the target is not a pool name (e.g. a path to a dataset) then
7630  * the new_path parameter is populated with the updated path to
7631  * reflect the fact that we are looking into the checkpointed state.
7632  *
7633  * The function returns a newly-allocated copy of the name of the
7634  * pool containing the checkpointed state. When this copy is no
7635  * longer needed it should be freed with free(3C). Same thing
7636  * applies to the new_path parameter if allocated.
7637  */
7638 static char *
import_checkpointed_state(char * target,nvlist_t * cfg,char ** new_path)7639 import_checkpointed_state(char *target, nvlist_t *cfg, char **new_path)
7640 {
7641 	int error = 0;
7642 	char *poolname, *bogus_name = NULL;
7643 	boolean_t freecfg = B_FALSE;
7644 
7645 	/* If the target is not a pool, the extract the pool name */
7646 	char *path_start = strchr(target, '/');
7647 	if (path_start != NULL) {
7648 		size_t poolname_len = path_start - target;
7649 		poolname = strndup(target, poolname_len);
7650 	} else {
7651 		poolname = target;
7652 	}
7653 
7654 	if (cfg == NULL) {
7655 		zdb_set_skip_mmp(poolname);
7656 		error = spa_get_stats(poolname, &cfg, NULL, 0);
7657 		if (error != 0) {
7658 			fatal("Tried to read config of pool \"%s\" but "
7659 			    "spa_get_stats() failed with error %d\n",
7660 			    poolname, error);
7661 		}
7662 		freecfg = B_TRUE;
7663 	}
7664 
7665 	if (asprintf(&bogus_name, "%s%s", poolname, BOGUS_SUFFIX) == -1) {
7666 		if (target != poolname)
7667 			free(poolname);
7668 		return (NULL);
7669 	}
7670 	fnvlist_add_string(cfg, ZPOOL_CONFIG_POOL_NAME, bogus_name);
7671 
7672 	error = spa_import(bogus_name, cfg, NULL,
7673 	    ZFS_IMPORT_MISSING_LOG | ZFS_IMPORT_CHECKPOINT |
7674 	    ZFS_IMPORT_SKIP_MMP);
7675 	if (freecfg)
7676 		nvlist_free(cfg);
7677 	if (error != 0) {
7678 		fatal("Tried to import pool \"%s\" but spa_import() failed "
7679 		    "with error %d\n", bogus_name, error);
7680 	}
7681 
7682 	if (new_path != NULL && path_start != NULL) {
7683 		if (asprintf(new_path, "%s%s", bogus_name, path_start) == -1) {
7684 			free(bogus_name);
7685 			if (path_start != NULL)
7686 				free(poolname);
7687 			return (NULL);
7688 		}
7689 	}
7690 
7691 	if (target != poolname)
7692 		free(poolname);
7693 
7694 	return (bogus_name);
7695 }
7696 
7697 typedef struct verify_checkpoint_sm_entry_cb_arg {
7698 	vdev_t *vcsec_vd;
7699 
7700 	/* the following fields are only used for printing progress */
7701 	uint64_t vcsec_entryid;
7702 	uint64_t vcsec_num_entries;
7703 } verify_checkpoint_sm_entry_cb_arg_t;
7704 
7705 #define	ENTRIES_PER_PROGRESS_UPDATE 10000
7706 
7707 static int
verify_checkpoint_sm_entry_cb(space_map_entry_t * sme,void * arg)7708 verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, void *arg)
7709 {
7710 	verify_checkpoint_sm_entry_cb_arg_t *vcsec = arg;
7711 	vdev_t *vd = vcsec->vcsec_vd;
7712 	metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
7713 	uint64_t end = sme->sme_offset + sme->sme_run;
7714 
7715 	ASSERT(sme->sme_type == SM_FREE);
7716 
7717 	if ((vcsec->vcsec_entryid % ENTRIES_PER_PROGRESS_UPDATE) == 0) {
7718 		(void) fprintf(stderr,
7719 		    "\rverifying vdev %llu, space map entry %llu of %llu ...",
7720 		    (longlong_t)vd->vdev_id,
7721 		    (longlong_t)vcsec->vcsec_entryid,
7722 		    (longlong_t)vcsec->vcsec_num_entries);
7723 	}
7724 	vcsec->vcsec_entryid++;
7725 
7726 	/*
7727 	 * See comment in checkpoint_sm_exclude_entry_cb()
7728 	 */
7729 	VERIFY3U(sme->sme_offset, >=, ms->ms_start);
7730 	VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
7731 
7732 	/*
7733 	 * The entries in the vdev_checkpoint_sm should be marked as
7734 	 * allocated in the checkpointed state of the pool, therefore
7735 	 * their respective ms_allocateable trees should not contain them.
7736 	 */
7737 	mutex_enter(&ms->ms_lock);
7738 	range_tree_verify_not_present(ms->ms_allocatable,
7739 	    sme->sme_offset, sme->sme_run);
7740 	mutex_exit(&ms->ms_lock);
7741 
7742 	return (0);
7743 }
7744 
7745 /*
7746  * Verify that all segments in the vdev_checkpoint_sm are allocated
7747  * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
7748  * ms_allocatable).
7749  *
7750  * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
7751  * each vdev in the current state of the pool to the metaslab space maps
7752  * (ms_sm) of the checkpointed state of the pool.
7753  *
7754  * Note that the function changes the state of the ms_allocatable
7755  * trees of the current spa_t. The entries of these ms_allocatable
7756  * trees are cleared out and then repopulated from with the free
7757  * entries of their respective ms_sm space maps.
7758  */
7759 static void
verify_checkpoint_vdev_spacemaps(spa_t * checkpoint,spa_t * current)7760 verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, spa_t *current)
7761 {
7762 	vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7763 	vdev_t *current_rvd = current->spa_root_vdev;
7764 
7765 	load_concrete_ms_allocatable_trees(checkpoint, SM_FREE);
7766 
7767 	for (uint64_t c = 0; c < ckpoint_rvd->vdev_children; c++) {
7768 		vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[c];
7769 		vdev_t *current_vd = current_rvd->vdev_child[c];
7770 
7771 		space_map_t *checkpoint_sm = NULL;
7772 		uint64_t checkpoint_sm_obj;
7773 
7774 		if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7775 			/*
7776 			 * Since we don't allow device removal in a pool
7777 			 * that has a checkpoint, we expect that all removed
7778 			 * vdevs were removed from the pool before the
7779 			 * checkpoint.
7780 			 */
7781 			ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7782 			continue;
7783 		}
7784 
7785 		/*
7786 		 * If the checkpoint space map doesn't exist, then nothing
7787 		 * here is checkpointed so there's nothing to verify.
7788 		 */
7789 		if (current_vd->vdev_top_zap == 0 ||
7790 		    zap_contains(spa_meta_objset(current),
7791 		    current_vd->vdev_top_zap,
7792 		    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
7793 			continue;
7794 
7795 		VERIFY0(zap_lookup(spa_meta_objset(current),
7796 		    current_vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
7797 		    sizeof (uint64_t), 1, &checkpoint_sm_obj));
7798 
7799 		VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current),
7800 		    checkpoint_sm_obj, 0, current_vd->vdev_asize,
7801 		    current_vd->vdev_ashift));
7802 
7803 		verify_checkpoint_sm_entry_cb_arg_t vcsec;
7804 		vcsec.vcsec_vd = ckpoint_vd;
7805 		vcsec.vcsec_entryid = 0;
7806 		vcsec.vcsec_num_entries =
7807 		    space_map_length(checkpoint_sm) / sizeof (uint64_t);
7808 		VERIFY0(space_map_iterate(checkpoint_sm,
7809 		    space_map_length(checkpoint_sm),
7810 		    verify_checkpoint_sm_entry_cb, &vcsec));
7811 		if (dump_opt['m'] > 3)
7812 			dump_spacemap(current->spa_meta_objset, checkpoint_sm);
7813 		space_map_close(checkpoint_sm);
7814 	}
7815 
7816 	/*
7817 	 * If we've added vdevs since we took the checkpoint, ensure
7818 	 * that their checkpoint space maps are empty.
7819 	 */
7820 	if (ckpoint_rvd->vdev_children < current_rvd->vdev_children) {
7821 		for (uint64_t c = ckpoint_rvd->vdev_children;
7822 		    c < current_rvd->vdev_children; c++) {
7823 			vdev_t *current_vd = current_rvd->vdev_child[c];
7824 			VERIFY3P(current_vd->vdev_checkpoint_sm, ==, NULL);
7825 		}
7826 	}
7827 
7828 	/* for cleaner progress output */
7829 	(void) fprintf(stderr, "\n");
7830 }
7831 
7832 /*
7833  * Verifies that all space that's allocated in the checkpoint is
7834  * still allocated in the current version, by checking that everything
7835  * in checkpoint's ms_allocatable (which is actually allocated, not
7836  * allocatable/free) is not present in current's ms_allocatable.
7837  *
7838  * Note that the function changes the state of the ms_allocatable
7839  * trees of both spas when called. The entries of all ms_allocatable
7840  * trees are cleared out and then repopulated from their respective
7841  * ms_sm space maps. In the checkpointed state we load the allocated
7842  * entries, and in the current state we load the free entries.
7843  */
7844 static void
verify_checkpoint_ms_spacemaps(spa_t * checkpoint,spa_t * current)7845 verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_t *current)
7846 {
7847 	vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7848 	vdev_t *current_rvd = current->spa_root_vdev;
7849 
7850 	load_concrete_ms_allocatable_trees(checkpoint, SM_ALLOC);
7851 	load_concrete_ms_allocatable_trees(current, SM_FREE);
7852 
7853 	for (uint64_t i = 0; i < ckpoint_rvd->vdev_children; i++) {
7854 		vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[i];
7855 		vdev_t *current_vd = current_rvd->vdev_child[i];
7856 
7857 		if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7858 			/*
7859 			 * See comment in verify_checkpoint_vdev_spacemaps()
7860 			 */
7861 			ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7862 			continue;
7863 		}
7864 
7865 		for (uint64_t m = 0; m < ckpoint_vd->vdev_ms_count; m++) {
7866 			metaslab_t *ckpoint_msp = ckpoint_vd->vdev_ms[m];
7867 			metaslab_t *current_msp = current_vd->vdev_ms[m];
7868 
7869 			(void) fprintf(stderr,
7870 			    "\rverifying vdev %llu of %llu, "
7871 			    "metaslab %llu of %llu ...",
7872 			    (longlong_t)current_vd->vdev_id,
7873 			    (longlong_t)current_rvd->vdev_children,
7874 			    (longlong_t)current_vd->vdev_ms[m]->ms_id,
7875 			    (longlong_t)current_vd->vdev_ms_count);
7876 
7877 			/*
7878 			 * We walk through the ms_allocatable trees that
7879 			 * are loaded with the allocated blocks from the
7880 			 * ms_sm spacemaps of the checkpoint. For each
7881 			 * one of these ranges we ensure that none of them
7882 			 * exists in the ms_allocatable trees of the
7883 			 * current state which are loaded with the ranges
7884 			 * that are currently free.
7885 			 *
7886 			 * This way we ensure that none of the blocks that
7887 			 * are part of the checkpoint were freed by mistake.
7888 			 */
7889 			range_tree_walk(ckpoint_msp->ms_allocatable,
7890 			    (range_tree_func_t *)range_tree_verify_not_present,
7891 			    current_msp->ms_allocatable);
7892 		}
7893 	}
7894 
7895 	/* for cleaner progress output */
7896 	(void) fprintf(stderr, "\n");
7897 }
7898 
7899 static void
verify_checkpoint_blocks(spa_t * spa)7900 verify_checkpoint_blocks(spa_t *spa)
7901 {
7902 	ASSERT(!dump_opt['L']);
7903 
7904 	spa_t *checkpoint_spa;
7905 	char *checkpoint_pool;
7906 	int error = 0;
7907 
7908 	/*
7909 	 * We import the checkpointed state of the pool (under a different
7910 	 * name) so we can do verification on it against the current state
7911 	 * of the pool.
7912 	 */
7913 	checkpoint_pool = import_checkpointed_state(spa->spa_name, NULL,
7914 	    NULL);
7915 	ASSERT(strcmp(spa->spa_name, checkpoint_pool) != 0);
7916 
7917 	error = spa_open(checkpoint_pool, &checkpoint_spa, FTAG);
7918 	if (error != 0) {
7919 		fatal("Tried to open pool \"%s\" but spa_open() failed with "
7920 		    "error %d\n", checkpoint_pool, error);
7921 	}
7922 
7923 	/*
7924 	 * Ensure that ranges in the checkpoint space maps of each vdev
7925 	 * are allocated according to the checkpointed state's metaslab
7926 	 * space maps.
7927 	 */
7928 	verify_checkpoint_vdev_spacemaps(checkpoint_spa, spa);
7929 
7930 	/*
7931 	 * Ensure that allocated ranges in the checkpoint's metaslab
7932 	 * space maps remain allocated in the metaslab space maps of
7933 	 * the current state.
7934 	 */
7935 	verify_checkpoint_ms_spacemaps(checkpoint_spa, spa);
7936 
7937 	/*
7938 	 * Once we are done, we get rid of the checkpointed state.
7939 	 */
7940 	spa_close(checkpoint_spa, FTAG);
7941 	free(checkpoint_pool);
7942 }
7943 
7944 static void
dump_leftover_checkpoint_blocks(spa_t * spa)7945 dump_leftover_checkpoint_blocks(spa_t *spa)
7946 {
7947 	vdev_t *rvd = spa->spa_root_vdev;
7948 
7949 	for (uint64_t i = 0; i < rvd->vdev_children; i++) {
7950 		vdev_t *vd = rvd->vdev_child[i];
7951 
7952 		space_map_t *checkpoint_sm = NULL;
7953 		uint64_t checkpoint_sm_obj;
7954 
7955 		if (vd->vdev_top_zap == 0)
7956 			continue;
7957 
7958 		if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
7959 		    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
7960 			continue;
7961 
7962 		VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
7963 		    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
7964 		    sizeof (uint64_t), 1, &checkpoint_sm_obj));
7965 
7966 		VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
7967 		    checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
7968 		dump_spacemap(spa->spa_meta_objset, checkpoint_sm);
7969 		space_map_close(checkpoint_sm);
7970 	}
7971 }
7972 
7973 static int
verify_checkpoint(spa_t * spa)7974 verify_checkpoint(spa_t *spa)
7975 {
7976 	uberblock_t checkpoint;
7977 	int error;
7978 
7979 	if (!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT))
7980 		return (0);
7981 
7982 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
7983 	    DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
7984 	    sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
7985 
7986 	if (error == ENOENT && !dump_opt['L']) {
7987 		/*
7988 		 * If the feature is active but the uberblock is missing
7989 		 * then we must be in the middle of discarding the
7990 		 * checkpoint.
7991 		 */
7992 		(void) printf("\nPartially discarded checkpoint "
7993 		    "state found:\n");
7994 		if (dump_opt['m'] > 3)
7995 			dump_leftover_checkpoint_blocks(spa);
7996 		return (0);
7997 	} else if (error != 0) {
7998 		(void) printf("lookup error %d when looking for "
7999 		    "checkpointed uberblock in MOS\n", error);
8000 		return (error);
8001 	}
8002 	dump_uberblock(&checkpoint, "\nCheckpointed uberblock found:\n", "\n");
8003 
8004 	if (checkpoint.ub_checkpoint_txg == 0) {
8005 		(void) printf("\nub_checkpoint_txg not set in checkpointed "
8006 		    "uberblock\n");
8007 		error = 3;
8008 	}
8009 
8010 	if (error == 0 && !dump_opt['L'])
8011 		verify_checkpoint_blocks(spa);
8012 
8013 	return (error);
8014 }
8015 
8016 static void
mos_leaks_cb(void * arg,uint64_t start,uint64_t size)8017 mos_leaks_cb(void *arg, uint64_t start, uint64_t size)
8018 {
8019 	(void) arg;
8020 	for (uint64_t i = start; i < size; i++) {
8021 		(void) printf("MOS object %llu referenced but not allocated\n",
8022 		    (u_longlong_t)i);
8023 	}
8024 }
8025 
8026 static void
mos_obj_refd(uint64_t obj)8027 mos_obj_refd(uint64_t obj)
8028 {
8029 	if (obj != 0 && mos_refd_objs != NULL)
8030 		range_tree_add(mos_refd_objs, obj, 1);
8031 }
8032 
8033 /*
8034  * Call on a MOS object that may already have been referenced.
8035  */
8036 static void
mos_obj_refd_multiple(uint64_t obj)8037 mos_obj_refd_multiple(uint64_t obj)
8038 {
8039 	if (obj != 0 && mos_refd_objs != NULL &&
8040 	    !range_tree_contains(mos_refd_objs, obj, 1))
8041 		range_tree_add(mos_refd_objs, obj, 1);
8042 }
8043 
8044 static void
mos_leak_vdev_top_zap(vdev_t * vd)8045 mos_leak_vdev_top_zap(vdev_t *vd)
8046 {
8047 	uint64_t ms_flush_data_obj;
8048 	int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
8049 	    vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
8050 	    sizeof (ms_flush_data_obj), 1, &ms_flush_data_obj);
8051 	if (error == ENOENT)
8052 		return;
8053 	ASSERT0(error);
8054 
8055 	mos_obj_refd(ms_flush_data_obj);
8056 }
8057 
8058 static void
mos_leak_vdev(vdev_t * vd)8059 mos_leak_vdev(vdev_t *vd)
8060 {
8061 	mos_obj_refd(vd->vdev_dtl_object);
8062 	mos_obj_refd(vd->vdev_ms_array);
8063 	mos_obj_refd(vd->vdev_indirect_config.vic_births_object);
8064 	mos_obj_refd(vd->vdev_indirect_config.vic_mapping_object);
8065 	mos_obj_refd(vd->vdev_leaf_zap);
8066 	if (vd->vdev_checkpoint_sm != NULL)
8067 		mos_obj_refd(vd->vdev_checkpoint_sm->sm_object);
8068 	if (vd->vdev_indirect_mapping != NULL) {
8069 		mos_obj_refd(vd->vdev_indirect_mapping->
8070 		    vim_phys->vimp_counts_object);
8071 	}
8072 	if (vd->vdev_obsolete_sm != NULL)
8073 		mos_obj_refd(vd->vdev_obsolete_sm->sm_object);
8074 
8075 	for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
8076 		metaslab_t *ms = vd->vdev_ms[m];
8077 		mos_obj_refd(space_map_object(ms->ms_sm));
8078 	}
8079 
8080 	if (vd->vdev_root_zap != 0)
8081 		mos_obj_refd(vd->vdev_root_zap);
8082 
8083 	if (vd->vdev_top_zap != 0) {
8084 		mos_obj_refd(vd->vdev_top_zap);
8085 		mos_leak_vdev_top_zap(vd);
8086 	}
8087 
8088 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
8089 		mos_leak_vdev(vd->vdev_child[c]);
8090 	}
8091 }
8092 
8093 static void
mos_leak_log_spacemaps(spa_t * spa)8094 mos_leak_log_spacemaps(spa_t *spa)
8095 {
8096 	uint64_t spacemap_zap;
8097 	int error = zap_lookup(spa_meta_objset(spa),
8098 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LOG_SPACEMAP_ZAP,
8099 	    sizeof (spacemap_zap), 1, &spacemap_zap);
8100 	if (error == ENOENT)
8101 		return;
8102 	ASSERT0(error);
8103 
8104 	mos_obj_refd(spacemap_zap);
8105 	for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
8106 	    sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls))
8107 		mos_obj_refd(sls->sls_sm_obj);
8108 }
8109 
8110 static void
errorlog_count_refd(objset_t * mos,uint64_t errlog)8111 errorlog_count_refd(objset_t *mos, uint64_t errlog)
8112 {
8113 	zap_cursor_t zc;
8114 	zap_attribute_t *za = zap_attribute_alloc();
8115 	for (zap_cursor_init(&zc, mos, errlog);
8116 	    zap_cursor_retrieve(&zc, za) == 0;
8117 	    zap_cursor_advance(&zc)) {
8118 		mos_obj_refd(za->za_first_integer);
8119 	}
8120 	zap_cursor_fini(&zc);
8121 	zap_attribute_free(za);
8122 }
8123 
8124 static int
dump_mos_leaks(spa_t * spa)8125 dump_mos_leaks(spa_t *spa)
8126 {
8127 	int rv = 0;
8128 	objset_t *mos = spa->spa_meta_objset;
8129 	dsl_pool_t *dp = spa->spa_dsl_pool;
8130 
8131 	/* Visit and mark all referenced objects in the MOS */
8132 
8133 	mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT);
8134 	mos_obj_refd(spa->spa_pool_props_object);
8135 	mos_obj_refd(spa->spa_config_object);
8136 	mos_obj_refd(spa->spa_ddt_stat_object);
8137 	mos_obj_refd(spa->spa_feat_desc_obj);
8138 	mos_obj_refd(spa->spa_feat_enabled_txg_obj);
8139 	mos_obj_refd(spa->spa_feat_for_read_obj);
8140 	mos_obj_refd(spa->spa_feat_for_write_obj);
8141 	mos_obj_refd(spa->spa_history);
8142 	mos_obj_refd(spa->spa_errlog_last);
8143 	mos_obj_refd(spa->spa_errlog_scrub);
8144 
8145 	if (spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) {
8146 		errorlog_count_refd(mos, spa->spa_errlog_last);
8147 		errorlog_count_refd(mos, spa->spa_errlog_scrub);
8148 	}
8149 
8150 	mos_obj_refd(spa->spa_all_vdev_zaps);
8151 	mos_obj_refd(spa->spa_dsl_pool->dp_bptree_obj);
8152 	mos_obj_refd(spa->spa_dsl_pool->dp_tmp_userrefs_obj);
8153 	mos_obj_refd(spa->spa_dsl_pool->dp_scan->scn_phys.scn_queue_obj);
8154 	bpobj_count_refd(&spa->spa_deferred_bpobj);
8155 	mos_obj_refd(dp->dp_empty_bpobj);
8156 	bpobj_count_refd(&dp->dp_obsolete_bpobj);
8157 	bpobj_count_refd(&dp->dp_free_bpobj);
8158 	mos_obj_refd(spa->spa_l2cache.sav_object);
8159 	mos_obj_refd(spa->spa_spares.sav_object);
8160 
8161 	if (spa->spa_syncing_log_sm != NULL)
8162 		mos_obj_refd(spa->spa_syncing_log_sm->sm_object);
8163 	mos_leak_log_spacemaps(spa);
8164 
8165 	mos_obj_refd(spa->spa_condensing_indirect_phys.
8166 	    scip_next_mapping_object);
8167 	mos_obj_refd(spa->spa_condensing_indirect_phys.
8168 	    scip_prev_obsolete_sm_object);
8169 	if (spa->spa_condensing_indirect_phys.scip_next_mapping_object != 0) {
8170 		vdev_indirect_mapping_t *vim =
8171 		    vdev_indirect_mapping_open(mos,
8172 		    spa->spa_condensing_indirect_phys.scip_next_mapping_object);
8173 		mos_obj_refd(vim->vim_phys->vimp_counts_object);
8174 		vdev_indirect_mapping_close(vim);
8175 	}
8176 	deleted_livelists_dump_mos(spa);
8177 
8178 	if (dp->dp_origin_snap != NULL) {
8179 		dsl_dataset_t *ds;
8180 
8181 		dsl_pool_config_enter(dp, FTAG);
8182 		VERIFY0(dsl_dataset_hold_obj(dp,
8183 		    dsl_dataset_phys(dp->dp_origin_snap)->ds_next_snap_obj,
8184 		    FTAG, &ds));
8185 		count_ds_mos_objects(ds);
8186 		dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
8187 		dsl_dataset_rele(ds, FTAG);
8188 		dsl_pool_config_exit(dp, FTAG);
8189 
8190 		count_ds_mos_objects(dp->dp_origin_snap);
8191 		dump_blkptr_list(&dp->dp_origin_snap->ds_deadlist, "Deadlist");
8192 	}
8193 	count_dir_mos_objects(dp->dp_mos_dir);
8194 	if (dp->dp_free_dir != NULL)
8195 		count_dir_mos_objects(dp->dp_free_dir);
8196 	if (dp->dp_leak_dir != NULL)
8197 		count_dir_mos_objects(dp->dp_leak_dir);
8198 
8199 	mos_leak_vdev(spa->spa_root_vdev);
8200 
8201 	for (uint64_t c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
8202 		ddt_t *ddt = spa->spa_ddt[c];
8203 		if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
8204 			continue;
8205 
8206 		/* DDT store objects */
8207 		for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
8208 			for (ddt_class_t class = 0; class < DDT_CLASSES;
8209 			    class++) {
8210 				mos_obj_refd(ddt->ddt_object[type][class]);
8211 			}
8212 		}
8213 
8214 		/* FDT container */
8215 		if (ddt->ddt_version == DDT_VERSION_FDT)
8216 			mos_obj_refd(ddt->ddt_dir_object);
8217 
8218 		/* FDT log objects */
8219 		if (ddt->ddt_flags & DDT_FLAG_LOG) {
8220 			mos_obj_refd(ddt->ddt_log[0].ddl_object);
8221 			mos_obj_refd(ddt->ddt_log[1].ddl_object);
8222 		}
8223 	}
8224 
8225 	for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
8226 		brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
8227 		if (brtvd->bv_initiated) {
8228 			mos_obj_refd(brtvd->bv_mos_brtvdev);
8229 			mos_obj_refd(brtvd->bv_mos_entries);
8230 		}
8231 	}
8232 
8233 	/*
8234 	 * Visit all allocated objects and make sure they are referenced.
8235 	 */
8236 	uint64_t object = 0;
8237 	while (dmu_object_next(mos, &object, B_FALSE, 0) == 0) {
8238 		if (range_tree_contains(mos_refd_objs, object, 1)) {
8239 			range_tree_remove(mos_refd_objs, object, 1);
8240 		} else {
8241 			dmu_object_info_t doi;
8242 			const char *name;
8243 			VERIFY0(dmu_object_info(mos, object, &doi));
8244 			if (doi.doi_type & DMU_OT_NEWTYPE) {
8245 				dmu_object_byteswap_t bswap =
8246 				    DMU_OT_BYTESWAP(doi.doi_type);
8247 				name = dmu_ot_byteswap[bswap].ob_name;
8248 			} else {
8249 				name = dmu_ot[doi.doi_type].ot_name;
8250 			}
8251 
8252 			(void) printf("MOS object %llu (%s) leaked\n",
8253 			    (u_longlong_t)object, name);
8254 			rv = 2;
8255 		}
8256 	}
8257 	(void) range_tree_walk(mos_refd_objs, mos_leaks_cb, NULL);
8258 	if (!range_tree_is_empty(mos_refd_objs))
8259 		rv = 2;
8260 	range_tree_vacate(mos_refd_objs, NULL, NULL);
8261 	range_tree_destroy(mos_refd_objs);
8262 	return (rv);
8263 }
8264 
8265 typedef struct log_sm_obsolete_stats_arg {
8266 	uint64_t lsos_current_txg;
8267 
8268 	uint64_t lsos_total_entries;
8269 	uint64_t lsos_valid_entries;
8270 
8271 	uint64_t lsos_sm_entries;
8272 	uint64_t lsos_valid_sm_entries;
8273 } log_sm_obsolete_stats_arg_t;
8274 
8275 static int
log_spacemap_obsolete_stats_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)8276 log_spacemap_obsolete_stats_cb(spa_t *spa, space_map_entry_t *sme,
8277     uint64_t txg, void *arg)
8278 {
8279 	log_sm_obsolete_stats_arg_t *lsos = arg;
8280 
8281 	uint64_t offset = sme->sme_offset;
8282 	uint64_t vdev_id = sme->sme_vdev;
8283 
8284 	if (lsos->lsos_current_txg == 0) {
8285 		/* this is the first log */
8286 		lsos->lsos_current_txg = txg;
8287 	} else if (lsos->lsos_current_txg < txg) {
8288 		/* we just changed log - print stats and reset */
8289 		(void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8290 		    (u_longlong_t)lsos->lsos_valid_sm_entries,
8291 		    (u_longlong_t)lsos->lsos_sm_entries,
8292 		    (u_longlong_t)lsos->lsos_current_txg);
8293 		lsos->lsos_valid_sm_entries = 0;
8294 		lsos->lsos_sm_entries = 0;
8295 		lsos->lsos_current_txg = txg;
8296 	}
8297 	ASSERT3U(lsos->lsos_current_txg, ==, txg);
8298 
8299 	lsos->lsos_sm_entries++;
8300 	lsos->lsos_total_entries++;
8301 
8302 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
8303 	if (!vdev_is_concrete(vd))
8304 		return (0);
8305 
8306 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
8307 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
8308 
8309 	if (txg < metaslab_unflushed_txg(ms))
8310 		return (0);
8311 	lsos->lsos_valid_sm_entries++;
8312 	lsos->lsos_valid_entries++;
8313 	return (0);
8314 }
8315 
8316 static void
dump_log_spacemap_obsolete_stats(spa_t * spa)8317 dump_log_spacemap_obsolete_stats(spa_t *spa)
8318 {
8319 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
8320 		return;
8321 
8322 	log_sm_obsolete_stats_arg_t lsos = {0};
8323 
8324 	(void) printf("Log Space Map Obsolete Entry Statistics:\n");
8325 
8326 	iterate_through_spacemap_logs(spa,
8327 	    log_spacemap_obsolete_stats_cb, &lsos);
8328 
8329 	/* print stats for latest log */
8330 	(void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8331 	    (u_longlong_t)lsos.lsos_valid_sm_entries,
8332 	    (u_longlong_t)lsos.lsos_sm_entries,
8333 	    (u_longlong_t)lsos.lsos_current_txg);
8334 
8335 	(void) printf("%-8llu valid entries out of %-8llu - total\n\n",
8336 	    (u_longlong_t)lsos.lsos_valid_entries,
8337 	    (u_longlong_t)lsos.lsos_total_entries);
8338 }
8339 
8340 static void
dump_zpool(spa_t * spa)8341 dump_zpool(spa_t *spa)
8342 {
8343 	dsl_pool_t *dp = spa_get_dsl(spa);
8344 	int rc = 0;
8345 
8346 	if (dump_opt['y']) {
8347 		livelist_metaslab_validate(spa);
8348 	}
8349 
8350 	if (dump_opt['S']) {
8351 		dump_simulated_ddt(spa);
8352 		return;
8353 	}
8354 
8355 	if (!dump_opt['e'] && dump_opt['C'] > 1) {
8356 		(void) printf("\nCached configuration:\n");
8357 		dump_nvlist(spa->spa_config, 8);
8358 	}
8359 
8360 	if (dump_opt['C'])
8361 		dump_config(spa);
8362 
8363 	if (dump_opt['u'])
8364 		dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n");
8365 
8366 	if (dump_opt['D'])
8367 		dump_all_ddts(spa);
8368 
8369 	if (dump_opt['T'])
8370 		dump_brt(spa);
8371 
8372 	if (dump_opt['d'] > 2 || dump_opt['m'])
8373 		dump_metaslabs(spa);
8374 	if (dump_opt['M'])
8375 		dump_metaslab_groups(spa, dump_opt['M'] > 1);
8376 	if (dump_opt['d'] > 2 || dump_opt['m']) {
8377 		dump_log_spacemaps(spa);
8378 		dump_log_spacemap_obsolete_stats(spa);
8379 	}
8380 
8381 	if (dump_opt['d'] || dump_opt['i']) {
8382 		spa_feature_t f;
8383 		mos_refd_objs = range_tree_create(NULL, RANGE_SEG64, NULL, 0,
8384 		    0);
8385 		dump_objset(dp->dp_meta_objset);
8386 
8387 		if (dump_opt['d'] >= 3) {
8388 			dsl_pool_t *dp = spa->spa_dsl_pool;
8389 			dump_full_bpobj(&spa->spa_deferred_bpobj,
8390 			    "Deferred frees", 0);
8391 			if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
8392 				dump_full_bpobj(&dp->dp_free_bpobj,
8393 				    "Pool snapshot frees", 0);
8394 			}
8395 			if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
8396 				ASSERT(spa_feature_is_enabled(spa,
8397 				    SPA_FEATURE_DEVICE_REMOVAL));
8398 				dump_full_bpobj(&dp->dp_obsolete_bpobj,
8399 				    "Pool obsolete blocks", 0);
8400 			}
8401 
8402 			if (spa_feature_is_active(spa,
8403 			    SPA_FEATURE_ASYNC_DESTROY)) {
8404 				dump_bptree(spa->spa_meta_objset,
8405 				    dp->dp_bptree_obj,
8406 				    "Pool dataset frees");
8407 			}
8408 			dump_dtl(spa->spa_root_vdev, 0);
8409 		}
8410 
8411 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++)
8412 			global_feature_count[f] = UINT64_MAX;
8413 		global_feature_count[SPA_FEATURE_REDACTION_BOOKMARKS] = 0;
8414 		global_feature_count[SPA_FEATURE_REDACTION_LIST_SPILL] = 0;
8415 		global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN] = 0;
8416 		global_feature_count[SPA_FEATURE_LIVELIST] = 0;
8417 
8418 		(void) dmu_objset_find(spa_name(spa), dump_one_objset,
8419 		    NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
8420 
8421 		if (rc == 0 && !dump_opt['L'])
8422 			rc = dump_mos_leaks(spa);
8423 
8424 		for (f = 0; f < SPA_FEATURES; f++) {
8425 			uint64_t refcount;
8426 
8427 			uint64_t *arr;
8428 			if (!(spa_feature_table[f].fi_flags &
8429 			    ZFEATURE_FLAG_PER_DATASET)) {
8430 				if (global_feature_count[f] == UINT64_MAX)
8431 					continue;
8432 				if (!spa_feature_is_enabled(spa, f)) {
8433 					ASSERT0(global_feature_count[f]);
8434 					continue;
8435 				}
8436 				arr = global_feature_count;
8437 			} else {
8438 				if (!spa_feature_is_enabled(spa, f)) {
8439 					ASSERT0(dataset_feature_count[f]);
8440 					continue;
8441 				}
8442 				arr = dataset_feature_count;
8443 			}
8444 			if (feature_get_refcount(spa, &spa_feature_table[f],
8445 			    &refcount) == ENOTSUP)
8446 				continue;
8447 			if (arr[f] != refcount) {
8448 				(void) printf("%s feature refcount mismatch: "
8449 				    "%lld consumers != %lld refcount\n",
8450 				    spa_feature_table[f].fi_uname,
8451 				    (longlong_t)arr[f], (longlong_t)refcount);
8452 				rc = 2;
8453 			} else {
8454 				(void) printf("Verified %s feature refcount "
8455 				    "of %llu is correct\n",
8456 				    spa_feature_table[f].fi_uname,
8457 				    (longlong_t)refcount);
8458 			}
8459 		}
8460 
8461 		if (rc == 0)
8462 			rc = verify_device_removal_feature_counts(spa);
8463 	}
8464 
8465 	if (rc == 0 && (dump_opt['b'] || dump_opt['c']))
8466 		rc = dump_block_stats(spa);
8467 
8468 	if (rc == 0)
8469 		rc = verify_spacemap_refcounts(spa);
8470 
8471 	if (dump_opt['s'])
8472 		show_pool_stats(spa);
8473 
8474 	if (dump_opt['h'])
8475 		dump_history(spa);
8476 
8477 	if (rc == 0)
8478 		rc = verify_checkpoint(spa);
8479 
8480 	if (rc != 0) {
8481 		dump_debug_buffer();
8482 		zdb_exit(rc);
8483 	}
8484 }
8485 
8486 #define	ZDB_FLAG_CHECKSUM	0x0001
8487 #define	ZDB_FLAG_DECOMPRESS	0x0002
8488 #define	ZDB_FLAG_BSWAP		0x0004
8489 #define	ZDB_FLAG_GBH		0x0008
8490 #define	ZDB_FLAG_INDIRECT	0x0010
8491 #define	ZDB_FLAG_RAW		0x0020
8492 #define	ZDB_FLAG_PRINT_BLKPTR	0x0040
8493 #define	ZDB_FLAG_VERBOSE	0x0080
8494 
8495 static int flagbits[256];
8496 static char flagbitstr[16];
8497 
8498 static void
zdb_print_blkptr(const blkptr_t * bp,int flags)8499 zdb_print_blkptr(const blkptr_t *bp, int flags)
8500 {
8501 	char blkbuf[BP_SPRINTF_LEN];
8502 
8503 	if (flags & ZDB_FLAG_BSWAP)
8504 		byteswap_uint64_array((void *)bp, sizeof (blkptr_t));
8505 
8506 	snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
8507 	(void) printf("%s\n", blkbuf);
8508 }
8509 
8510 static void
zdb_dump_indirect(blkptr_t * bp,int nbps,int flags)8511 zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
8512 {
8513 	int i;
8514 
8515 	for (i = 0; i < nbps; i++)
8516 		zdb_print_blkptr(&bp[i], flags);
8517 }
8518 
8519 static void
zdb_dump_gbh(void * buf,int flags)8520 zdb_dump_gbh(void *buf, int flags)
8521 {
8522 	zdb_dump_indirect((blkptr_t *)buf, SPA_GBH_NBLKPTRS, flags);
8523 }
8524 
8525 static void
zdb_dump_block_raw(void * buf,uint64_t size,int flags)8526 zdb_dump_block_raw(void *buf, uint64_t size, int flags)
8527 {
8528 	if (flags & ZDB_FLAG_BSWAP)
8529 		byteswap_uint64_array(buf, size);
8530 	VERIFY(write(fileno(stdout), buf, size) == size);
8531 }
8532 
8533 static void
zdb_dump_block(char * label,void * buf,uint64_t size,int flags)8534 zdb_dump_block(char *label, void *buf, uint64_t size, int flags)
8535 {
8536 	uint64_t *d = (uint64_t *)buf;
8537 	unsigned nwords = size / sizeof (uint64_t);
8538 	int do_bswap = !!(flags & ZDB_FLAG_BSWAP);
8539 	unsigned i, j;
8540 	const char *hdr;
8541 	char *c;
8542 
8543 
8544 	if (do_bswap)
8545 		hdr = " 7 6 5 4 3 2 1 0   f e d c b a 9 8";
8546 	else
8547 		hdr = " 0 1 2 3 4 5 6 7   8 9 a b c d e f";
8548 
8549 	(void) printf("\n%s\n%6s   %s  0123456789abcdef\n", label, "", hdr);
8550 
8551 #ifdef _ZFS_LITTLE_ENDIAN
8552 	/* correct the endianness */
8553 	do_bswap = !do_bswap;
8554 #endif
8555 	for (i = 0; i < nwords; i += 2) {
8556 		(void) printf("%06llx:  %016llx  %016llx  ",
8557 		    (u_longlong_t)(i * sizeof (uint64_t)),
8558 		    (u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]),
8559 		    (u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1]));
8560 
8561 		c = (char *)&d[i];
8562 		for (j = 0; j < 2 * sizeof (uint64_t); j++)
8563 			(void) printf("%c", isprint(c[j]) ? c[j] : '.');
8564 		(void) printf("\n");
8565 	}
8566 }
8567 
8568 /*
8569  * There are two acceptable formats:
8570  *	leaf_name	  - For example: c1t0d0 or /tmp/ztest.0a
8571  *	child[.child]*    - For example: 0.1.1
8572  *
8573  * The second form can be used to specify arbitrary vdevs anywhere
8574  * in the hierarchy.  For example, in a pool with a mirror of
8575  * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
8576  */
8577 static vdev_t *
zdb_vdev_lookup(vdev_t * vdev,const char * path)8578 zdb_vdev_lookup(vdev_t *vdev, const char *path)
8579 {
8580 	char *s, *p, *q;
8581 	unsigned i;
8582 
8583 	if (vdev == NULL)
8584 		return (NULL);
8585 
8586 	/* First, assume the x.x.x.x format */
8587 	i = strtoul(path, &s, 10);
8588 	if (s == path || (s && *s != '.' && *s != '\0'))
8589 		goto name;
8590 	if (i >= vdev->vdev_children)
8591 		return (NULL);
8592 
8593 	vdev = vdev->vdev_child[i];
8594 	if (s && *s == '\0')
8595 		return (vdev);
8596 	return (zdb_vdev_lookup(vdev, s+1));
8597 
8598 name:
8599 	for (i = 0; i < vdev->vdev_children; i++) {
8600 		vdev_t *vc = vdev->vdev_child[i];
8601 
8602 		if (vc->vdev_path == NULL) {
8603 			vc = zdb_vdev_lookup(vc, path);
8604 			if (vc == NULL)
8605 				continue;
8606 			else
8607 				return (vc);
8608 		}
8609 
8610 		p = strrchr(vc->vdev_path, '/');
8611 		p = p ? p + 1 : vc->vdev_path;
8612 		q = &vc->vdev_path[strlen(vc->vdev_path) - 2];
8613 
8614 		if (strcmp(vc->vdev_path, path) == 0)
8615 			return (vc);
8616 		if (strcmp(p, path) == 0)
8617 			return (vc);
8618 		if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0)
8619 			return (vc);
8620 	}
8621 
8622 	return (NULL);
8623 }
8624 
8625 static int
name_from_objset_id(spa_t * spa,uint64_t objset_id,char * outstr)8626 name_from_objset_id(spa_t *spa, uint64_t objset_id, char *outstr)
8627 {
8628 	dsl_dataset_t *ds;
8629 
8630 	dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
8631 	int error = dsl_dataset_hold_obj(spa->spa_dsl_pool, objset_id,
8632 	    NULL, &ds);
8633 	if (error != 0) {
8634 		(void) fprintf(stderr, "failed to hold objset %llu: %s\n",
8635 		    (u_longlong_t)objset_id, strerror(error));
8636 		dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8637 		return (error);
8638 	}
8639 	dsl_dataset_name(ds, outstr);
8640 	dsl_dataset_rele(ds, NULL);
8641 	dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8642 	return (0);
8643 }
8644 
8645 static boolean_t
zdb_parse_block_sizes(char * sizes,uint64_t * lsize,uint64_t * psize)8646 zdb_parse_block_sizes(char *sizes, uint64_t *lsize, uint64_t *psize)
8647 {
8648 	char *s0, *s1, *tmp = NULL;
8649 
8650 	if (sizes == NULL)
8651 		return (B_FALSE);
8652 
8653 	s0 = strtok_r(sizes, "/", &tmp);
8654 	if (s0 == NULL)
8655 		return (B_FALSE);
8656 	s1 = strtok_r(NULL, "/", &tmp);
8657 	*lsize = strtoull(s0, NULL, 16);
8658 	*psize = s1 ? strtoull(s1, NULL, 16) : *lsize;
8659 	return (*lsize >= *psize && *psize > 0);
8660 }
8661 
8662 #define	ZIO_COMPRESS_MASK(alg)	(1ULL << (ZIO_COMPRESS_##alg))
8663 
8664 static boolean_t
try_decompress_block(abd_t * pabd,uint64_t lsize,uint64_t psize,int flags,int cfunc,void * lbuf,void * lbuf2)8665 try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize,
8666     int flags, int cfunc, void *lbuf, void *lbuf2)
8667 {
8668 	if (flags & ZDB_FLAG_VERBOSE) {
8669 		(void) fprintf(stderr,
8670 		    "Trying %05llx -> %05llx (%s)\n",
8671 		    (u_longlong_t)psize,
8672 		    (u_longlong_t)lsize,
8673 		    zio_compress_table[cfunc].ci_name);
8674 	}
8675 
8676 	/*
8677 	 * We set lbuf to all zeros and lbuf2 to all
8678 	 * ones, then decompress to both buffers and
8679 	 * compare their contents. This way we can
8680 	 * know if decompression filled exactly to
8681 	 * lsize or if it left some bytes unwritten.
8682 	 */
8683 
8684 	memset(lbuf, 0x00, lsize);
8685 	memset(lbuf2, 0xff, lsize);
8686 
8687 	abd_t labd, labd2;
8688 	abd_get_from_buf_struct(&labd, lbuf, lsize);
8689 	abd_get_from_buf_struct(&labd2, lbuf2, lsize);
8690 
8691 	boolean_t ret = B_FALSE;
8692 	if (zio_decompress_data(cfunc, pabd,
8693 	    &labd, psize, lsize, NULL) == 0 &&
8694 	    zio_decompress_data(cfunc, pabd,
8695 	    &labd2, psize, lsize, NULL) == 0 &&
8696 	    memcmp(lbuf, lbuf2, lsize) == 0)
8697 		ret = B_TRUE;
8698 
8699 	abd_free(&labd2);
8700 	abd_free(&labd);
8701 
8702 	return (ret);
8703 }
8704 
8705 static uint64_t
zdb_decompress_block(abd_t * pabd,void * buf,void * lbuf,uint64_t lsize,uint64_t psize,int flags)8706 zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
8707     uint64_t psize, int flags)
8708 {
8709 	(void) buf;
8710 	uint64_t orig_lsize = lsize;
8711 	boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL));
8712 	boolean_t found = B_FALSE;
8713 	/*
8714 	 * We don't know how the data was compressed, so just try
8715 	 * every decompress function at every inflated blocksize.
8716 	 */
8717 	void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
8718 	int cfuncs[ZIO_COMPRESS_FUNCTIONS] = { 0 };
8719 	int *cfuncp = cfuncs;
8720 	uint64_t maxlsize = SPA_MAXBLOCKSIZE;
8721 	uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) |
8722 	    ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) |
8723 	    ZIO_COMPRESS_MASK(ZLE);
8724 	*cfuncp++ = ZIO_COMPRESS_LZ4;
8725 	*cfuncp++ = ZIO_COMPRESS_LZJB;
8726 	mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB);
8727 	/*
8728 	 * Every gzip level has the same decompressor, no need to
8729 	 * run it 9 times per bruteforce attempt.
8730 	 */
8731 	mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3);
8732 	mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5);
8733 	mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7);
8734 	mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9);
8735 	for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++)
8736 		if (((1ULL << c) & mask) == 0)
8737 			*cfuncp++ = c;
8738 
8739 	/*
8740 	 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
8741 	 * could take a while and we should let the user know
8742 	 * we are not stuck.  On the other hand, printing progress
8743 	 * info gets old after a while.  User can specify 'v' flag
8744 	 * to see the progression.
8745 	 */
8746 	if (lsize == psize)
8747 		lsize += SPA_MINBLOCKSIZE;
8748 	else
8749 		maxlsize = lsize;
8750 
8751 	for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) {
8752 		for (cfuncp = cfuncs; *cfuncp; cfuncp++) {
8753 			if (try_decompress_block(pabd, lsize, psize, flags,
8754 			    *cfuncp, lbuf, lbuf2)) {
8755 				found = B_TRUE;
8756 				break;
8757 			}
8758 		}
8759 		if (*cfuncp != 0)
8760 			break;
8761 	}
8762 	if (!found && tryzle) {
8763 		for (lsize = orig_lsize; lsize <= maxlsize;
8764 		    lsize += SPA_MINBLOCKSIZE) {
8765 			if (try_decompress_block(pabd, lsize, psize, flags,
8766 			    ZIO_COMPRESS_ZLE, lbuf, lbuf2)) {
8767 				*cfuncp = ZIO_COMPRESS_ZLE;
8768 				found = B_TRUE;
8769 				break;
8770 			}
8771 		}
8772 	}
8773 	umem_free(lbuf2, SPA_MAXBLOCKSIZE);
8774 
8775 	if (*cfuncp == ZIO_COMPRESS_ZLE) {
8776 		printf("\nZLE decompression was selected. If you "
8777 		    "suspect the results are wrong,\ntry avoiding ZLE "
8778 		    "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
8779 	}
8780 
8781 	return (lsize > maxlsize ? -1 : lsize);
8782 }
8783 
8784 /*
8785  * Read a block from a pool and print it out.  The syntax of the
8786  * block descriptor is:
8787  *
8788  *	pool:vdev_specifier:offset:[lsize/]psize[:flags]
8789  *
8790  *	pool           - The name of the pool you wish to read from
8791  *	vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
8792  *	offset         - offset, in hex, in bytes
8793  *	size           - Amount of data to read, in hex, in bytes
8794  *	flags          - A string of characters specifying options
8795  *		 b: Decode a blkptr at given offset within block
8796  *		 c: Calculate and display checksums
8797  *		 d: Decompress data before dumping
8798  *		 e: Byteswap data before dumping
8799  *		 g: Display data as a gang block header
8800  *		 i: Display as an indirect block
8801  *		 r: Dump raw data to stdout
8802  *		 v: Verbose
8803  *
8804  */
8805 static void
zdb_read_block(char * thing,spa_t * spa)8806 zdb_read_block(char *thing, spa_t *spa)
8807 {
8808 	blkptr_t blk, *bp = &blk;
8809 	dva_t *dva = bp->blk_dva;
8810 	int flags = 0;
8811 	uint64_t offset = 0, psize = 0, lsize = 0, blkptr_offset = 0;
8812 	zio_t *zio;
8813 	vdev_t *vd;
8814 	abd_t *pabd;
8815 	void *lbuf, *buf;
8816 	char *s, *p, *dup, *flagstr, *sizes, *tmp = NULL;
8817 	const char *vdev, *errmsg = NULL;
8818 	int i, len, error;
8819 	boolean_t borrowed = B_FALSE, found = B_FALSE;
8820 
8821 	dup = strdup(thing);
8822 	s = strtok_r(dup, ":", &tmp);
8823 	vdev = s ?: "";
8824 	s = strtok_r(NULL, ":", &tmp);
8825 	offset = strtoull(s ? s : "", NULL, 16);
8826 	sizes = strtok_r(NULL, ":", &tmp);
8827 	s = strtok_r(NULL, ":", &tmp);
8828 	flagstr = strdup(s ?: "");
8829 
8830 	if (!zdb_parse_block_sizes(sizes, &lsize, &psize))
8831 		errmsg = "invalid size(s)";
8832 	if (!IS_P2ALIGNED(psize, DEV_BSIZE) || !IS_P2ALIGNED(lsize, DEV_BSIZE))
8833 		errmsg = "size must be a multiple of sector size";
8834 	if (!IS_P2ALIGNED(offset, DEV_BSIZE))
8835 		errmsg = "offset must be a multiple of sector size";
8836 	if (errmsg) {
8837 		(void) printf("Invalid block specifier: %s  - %s\n",
8838 		    thing, errmsg);
8839 		goto done;
8840 	}
8841 
8842 	tmp = NULL;
8843 	for (s = strtok_r(flagstr, ":", &tmp);
8844 	    s != NULL;
8845 	    s = strtok_r(NULL, ":", &tmp)) {
8846 		len = strlen(flagstr);
8847 		for (i = 0; i < len; i++) {
8848 			int bit = flagbits[(uchar_t)flagstr[i]];
8849 
8850 			if (bit == 0) {
8851 				(void) printf("***Ignoring flag: %c\n",
8852 				    (uchar_t)flagstr[i]);
8853 				continue;
8854 			}
8855 			found = B_TRUE;
8856 			flags |= bit;
8857 
8858 			p = &flagstr[i + 1];
8859 			if (*p != ':' && *p != '\0') {
8860 				int j = 0, nextbit = flagbits[(uchar_t)*p];
8861 				char *end, offstr[8] = { 0 };
8862 				if ((bit == ZDB_FLAG_PRINT_BLKPTR) &&
8863 				    (nextbit == 0)) {
8864 					/* look ahead to isolate the offset */
8865 					while (nextbit == 0 &&
8866 					    strchr(flagbitstr, *p) == NULL) {
8867 						offstr[j] = *p;
8868 						j++;
8869 						if (i + j > strlen(flagstr))
8870 							break;
8871 						p++;
8872 						nextbit = flagbits[(uchar_t)*p];
8873 					}
8874 					blkptr_offset = strtoull(offstr, &end,
8875 					    16);
8876 					i += j;
8877 				} else if (nextbit == 0) {
8878 					(void) printf("***Ignoring flag arg:"
8879 					    " '%c'\n", (uchar_t)*p);
8880 				}
8881 			}
8882 		}
8883 	}
8884 	if (blkptr_offset % sizeof (blkptr_t)) {
8885 		printf("Block pointer offset 0x%llx "
8886 		    "must be divisible by 0x%x\n",
8887 		    (longlong_t)blkptr_offset, (int)sizeof (blkptr_t));
8888 		goto done;
8889 	}
8890 	if (found == B_FALSE && strlen(flagstr) > 0) {
8891 		printf("Invalid flag arg: '%s'\n", flagstr);
8892 		goto done;
8893 	}
8894 
8895 	vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev);
8896 	if (vd == NULL) {
8897 		(void) printf("***Invalid vdev: %s\n", vdev);
8898 		goto done;
8899 	} else {
8900 		if (vd->vdev_path)
8901 			(void) fprintf(stderr, "Found vdev: %s\n",
8902 			    vd->vdev_path);
8903 		else
8904 			(void) fprintf(stderr, "Found vdev type: %s\n",
8905 			    vd->vdev_ops->vdev_op_type);
8906 	}
8907 
8908 	pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
8909 	lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
8910 
8911 	BP_ZERO(bp);
8912 
8913 	DVA_SET_VDEV(&dva[0], vd->vdev_id);
8914 	DVA_SET_OFFSET(&dva[0], offset);
8915 	DVA_SET_GANG(&dva[0], !!(flags & ZDB_FLAG_GBH));
8916 	DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
8917 
8918 	BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
8919 
8920 	BP_SET_LSIZE(bp, lsize);
8921 	BP_SET_PSIZE(bp, psize);
8922 	BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF);
8923 	BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF);
8924 	BP_SET_TYPE(bp, DMU_OT_NONE);
8925 	BP_SET_LEVEL(bp, 0);
8926 	BP_SET_DEDUP(bp, 0);
8927 	BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
8928 
8929 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8930 	zio = zio_root(spa, NULL, NULL, 0);
8931 
8932 	if (vd == vd->vdev_top) {
8933 		/*
8934 		 * Treat this as a normal block read.
8935 		 */
8936 		zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL,
8937 		    ZIO_PRIORITY_SYNC_READ,
8938 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL));
8939 	} else {
8940 		/*
8941 		 * Treat this as a vdev child I/O.
8942 		 */
8943 		zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd,
8944 		    psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ,
8945 		    ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
8946 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | ZIO_FLAG_OPTIONAL,
8947 		    NULL, NULL));
8948 	}
8949 
8950 	error = zio_wait(zio);
8951 	spa_config_exit(spa, SCL_STATE, FTAG);
8952 
8953 	if (error) {
8954 		(void) printf("Read of %s failed, error: %d\n", thing, error);
8955 		goto out;
8956 	}
8957 
8958 	uint64_t orig_lsize = lsize;
8959 	buf = lbuf;
8960 	if (flags & ZDB_FLAG_DECOMPRESS) {
8961 		lsize = zdb_decompress_block(pabd, buf, lbuf,
8962 		    lsize, psize, flags);
8963 		if (lsize == -1) {
8964 			(void) printf("Decompress of %s failed\n", thing);
8965 			goto out;
8966 		}
8967 	} else {
8968 		buf = abd_borrow_buf_copy(pabd, lsize);
8969 		borrowed = B_TRUE;
8970 	}
8971 	/*
8972 	 * Try to detect invalid block pointer.  If invalid, try
8973 	 * decompressing.
8974 	 */
8975 	if ((flags & ZDB_FLAG_PRINT_BLKPTR || flags & ZDB_FLAG_INDIRECT) &&
8976 	    !(flags & ZDB_FLAG_DECOMPRESS)) {
8977 		const blkptr_t *b = (const blkptr_t *)(void *)
8978 		    ((uintptr_t)buf + (uintptr_t)blkptr_offset);
8979 		if (zfs_blkptr_verify(spa, b,
8980 		    BLK_CONFIG_NEEDED, BLK_VERIFY_ONLY) == B_FALSE) {
8981 			abd_return_buf_copy(pabd, buf, lsize);
8982 			borrowed = B_FALSE;
8983 			buf = lbuf;
8984 			lsize = zdb_decompress_block(pabd, buf,
8985 			    lbuf, lsize, psize, flags);
8986 			b = (const blkptr_t *)(void *)
8987 			    ((uintptr_t)buf + (uintptr_t)blkptr_offset);
8988 			if (lsize == -1 || zfs_blkptr_verify(spa, b,
8989 			    BLK_CONFIG_NEEDED, BLK_VERIFY_LOG) == B_FALSE) {
8990 				printf("invalid block pointer at this DVA\n");
8991 				goto out;
8992 			}
8993 		}
8994 	}
8995 
8996 	if (flags & ZDB_FLAG_PRINT_BLKPTR)
8997 		zdb_print_blkptr((blkptr_t *)(void *)
8998 		    ((uintptr_t)buf + (uintptr_t)blkptr_offset), flags);
8999 	else if (flags & ZDB_FLAG_RAW)
9000 		zdb_dump_block_raw(buf, lsize, flags);
9001 	else if (flags & ZDB_FLAG_INDIRECT)
9002 		zdb_dump_indirect((blkptr_t *)buf,
9003 		    orig_lsize / sizeof (blkptr_t), flags);
9004 	else if (flags & ZDB_FLAG_GBH)
9005 		zdb_dump_gbh(buf, flags);
9006 	else
9007 		zdb_dump_block(thing, buf, lsize, flags);
9008 
9009 	/*
9010 	 * If :c was specified, iterate through the checksum table to
9011 	 * calculate and display each checksum for our specified
9012 	 * DVA and length.
9013 	 */
9014 	if ((flags & ZDB_FLAG_CHECKSUM) && !(flags & ZDB_FLAG_RAW) &&
9015 	    !(flags & ZDB_FLAG_GBH)) {
9016 		zio_t *czio;
9017 		(void) printf("\n");
9018 		for (enum zio_checksum ck = ZIO_CHECKSUM_LABEL;
9019 		    ck < ZIO_CHECKSUM_FUNCTIONS; ck++) {
9020 
9021 			if ((zio_checksum_table[ck].ci_flags &
9022 			    ZCHECKSUM_FLAG_EMBEDDED) ||
9023 			    ck == ZIO_CHECKSUM_NOPARITY) {
9024 				continue;
9025 			}
9026 			BP_SET_CHECKSUM(bp, ck);
9027 			spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9028 			czio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9029 			if (vd == vd->vdev_top) {
9030 				zio_nowait(zio_read(czio, spa, bp, pabd, psize,
9031 				    NULL, NULL,
9032 				    ZIO_PRIORITY_SYNC_READ,
9033 				    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9034 				    ZIO_FLAG_DONT_RETRY, NULL));
9035 			} else {
9036 				zio_nowait(zio_vdev_child_io(czio, bp, vd,
9037 				    offset, pabd, psize, ZIO_TYPE_READ,
9038 				    ZIO_PRIORITY_SYNC_READ,
9039 				    ZIO_FLAG_DONT_PROPAGATE |
9040 				    ZIO_FLAG_DONT_RETRY |
9041 				    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9042 				    ZIO_FLAG_SPECULATIVE |
9043 				    ZIO_FLAG_OPTIONAL, NULL, NULL));
9044 			}
9045 			error = zio_wait(czio);
9046 			if (error == 0 || error == ECKSUM) {
9047 				zio_t *ck_zio = zio_null(NULL, spa, NULL,
9048 				    NULL, NULL, 0);
9049 				ck_zio->io_offset =
9050 				    DVA_GET_OFFSET(&bp->blk_dva[0]);
9051 				ck_zio->io_bp = bp;
9052 				zio_checksum_compute(ck_zio, ck, pabd, lsize);
9053 				printf(
9054 				    "%12s\t"
9055 				    "cksum=%016llx:%016llx:%016llx:%016llx\n",
9056 				    zio_checksum_table[ck].ci_name,
9057 				    (u_longlong_t)bp->blk_cksum.zc_word[0],
9058 				    (u_longlong_t)bp->blk_cksum.zc_word[1],
9059 				    (u_longlong_t)bp->blk_cksum.zc_word[2],
9060 				    (u_longlong_t)bp->blk_cksum.zc_word[3]);
9061 				zio_wait(ck_zio);
9062 			} else {
9063 				printf("error %d reading block\n", error);
9064 			}
9065 			spa_config_exit(spa, SCL_STATE, FTAG);
9066 		}
9067 	}
9068 
9069 	if (borrowed)
9070 		abd_return_buf_copy(pabd, buf, lsize);
9071 
9072 out:
9073 	abd_free(pabd);
9074 	umem_free(lbuf, SPA_MAXBLOCKSIZE);
9075 done:
9076 	free(flagstr);
9077 	free(dup);
9078 }
9079 
9080 static void
zdb_embedded_block(char * thing)9081 zdb_embedded_block(char *thing)
9082 {
9083 	blkptr_t bp = {{{{0}}}};
9084 	unsigned long long *words = (void *)&bp;
9085 	char *buf;
9086 	int err;
9087 
9088 	err = sscanf(thing, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
9089 	    "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
9090 	    words + 0, words + 1, words + 2, words + 3,
9091 	    words + 4, words + 5, words + 6, words + 7,
9092 	    words + 8, words + 9, words + 10, words + 11,
9093 	    words + 12, words + 13, words + 14, words + 15);
9094 	if (err != 16) {
9095 		(void) fprintf(stderr, "invalid input format\n");
9096 		zdb_exit(1);
9097 	}
9098 	ASSERT3U(BPE_GET_LSIZE(&bp), <=, SPA_MAXBLOCKSIZE);
9099 	buf = malloc(SPA_MAXBLOCKSIZE);
9100 	if (buf == NULL) {
9101 		(void) fprintf(stderr, "out of memory\n");
9102 		zdb_exit(1);
9103 	}
9104 	err = decode_embedded_bp(&bp, buf, BPE_GET_LSIZE(&bp));
9105 	if (err != 0) {
9106 		(void) fprintf(stderr, "decode failed: %u\n", err);
9107 		zdb_exit(1);
9108 	}
9109 	zdb_dump_block_raw(buf, BPE_GET_LSIZE(&bp), 0);
9110 	free(buf);
9111 }
9112 
9113 /* check for valid hex or decimal numeric string */
9114 static boolean_t
zdb_numeric(char * str)9115 zdb_numeric(char *str)
9116 {
9117 	int i = 0, len;
9118 
9119 	len = strlen(str);
9120 	if (len == 0)
9121 		return (B_FALSE);
9122 	if (strncmp(str, "0x", 2) == 0 || strncmp(str, "0X", 2) == 0)
9123 		i = 2;
9124 	for (; i < len; i++) {
9125 		if (!isxdigit(str[i]))
9126 			return (B_FALSE);
9127 	}
9128 	return (B_TRUE);
9129 }
9130 
9131 static int
dummy_get_file_info(dmu_object_type_t bonustype,const void * data,zfs_file_info_t * zoi)9132 dummy_get_file_info(dmu_object_type_t bonustype, const void *data,
9133     zfs_file_info_t *zoi)
9134 {
9135 	(void) data, (void) zoi;
9136 
9137 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
9138 		return (ENOENT);
9139 
9140 	(void) fprintf(stderr, "dummy_get_file_info: not implemented");
9141 	abort();
9142 }
9143 
9144 int
main(int argc,char ** argv)9145 main(int argc, char **argv)
9146 {
9147 	int c;
9148 	int dump_all = 1;
9149 	int verbose = 0;
9150 	int error = 0;
9151 	char **searchdirs = NULL;
9152 	int nsearch = 0;
9153 	char *target, *target_pool, dsname[ZFS_MAX_DATASET_NAME_LEN];
9154 	nvlist_t *policy = NULL;
9155 	uint64_t max_txg = UINT64_MAX;
9156 	int64_t objset_id = -1;
9157 	uint64_t object;
9158 	int flags = ZFS_IMPORT_MISSING_LOG;
9159 	int rewind = ZPOOL_NEVER_REWIND;
9160 	char *spa_config_path_env, *objset_str;
9161 	boolean_t target_is_spa = B_TRUE, dataset_lookup = B_FALSE;
9162 	nvlist_t *cfg = NULL;
9163 	struct sigaction action;
9164 	boolean_t force_import = B_FALSE;
9165 	boolean_t config_path_console = B_FALSE;
9166 	char pbuf[MAXPATHLEN];
9167 
9168 	dprintf_setup(&argc, argv);
9169 
9170 	/*
9171 	 * Set up signal handlers, so if we crash due to bad on-disk data we
9172 	 * can get more info. Unlike ztest, we don't bail out if we can't set
9173 	 * up signal handlers, because zdb is very useful without them.
9174 	 */
9175 	action.sa_handler = sig_handler;
9176 	sigemptyset(&action.sa_mask);
9177 	action.sa_flags = 0;
9178 	if (sigaction(SIGSEGV, &action, NULL) < 0) {
9179 		(void) fprintf(stderr, "zdb: cannot catch SIGSEGV: %s\n",
9180 		    strerror(errno));
9181 	}
9182 	if (sigaction(SIGABRT, &action, NULL) < 0) {
9183 		(void) fprintf(stderr, "zdb: cannot catch SIGABRT: %s\n",
9184 		    strerror(errno));
9185 	}
9186 
9187 	/*
9188 	 * If there is an environment variable SPA_CONFIG_PATH it overrides
9189 	 * default spa_config_path setting. If -U flag is specified it will
9190 	 * override this environment variable settings once again.
9191 	 */
9192 	spa_config_path_env = getenv("SPA_CONFIG_PATH");
9193 	if (spa_config_path_env != NULL)
9194 		spa_config_path = spa_config_path_env;
9195 
9196 	/*
9197 	 * For performance reasons, we set this tunable down. We do so before
9198 	 * the arg parsing section so that the user can override this value if
9199 	 * they choose.
9200 	 */
9201 	zfs_btree_verify_intensity = 3;
9202 
9203 	struct option long_options[] = {
9204 		{"ignore-assertions",	no_argument,		NULL, 'A'},
9205 		{"block-stats",		no_argument,		NULL, 'b'},
9206 		{"backup",		no_argument,		NULL, 'B'},
9207 		{"checksum",		no_argument,		NULL, 'c'},
9208 		{"config",		no_argument,		NULL, 'C'},
9209 		{"datasets",		no_argument,		NULL, 'd'},
9210 		{"dedup-stats",		no_argument,		NULL, 'D'},
9211 		{"exported",		no_argument,		NULL, 'e'},
9212 		{"embedded-block-pointer",	no_argument,	NULL, 'E'},
9213 		{"automatic-rewind",	no_argument,		NULL, 'F'},
9214 		{"dump-debug-msg",	no_argument,		NULL, 'G'},
9215 		{"history",		no_argument,		NULL, 'h'},
9216 		{"intent-logs",		no_argument,		NULL, 'i'},
9217 		{"inflight",		required_argument,	NULL, 'I'},
9218 		{"checkpointed-state",	no_argument,		NULL, 'k'},
9219 		{"key",			required_argument,	NULL, 'K'},
9220 		{"label",		no_argument,		NULL, 'l'},
9221 		{"disable-leak-tracking",	no_argument,	NULL, 'L'},
9222 		{"metaslabs",		no_argument,		NULL, 'm'},
9223 		{"metaslab-groups",	no_argument,		NULL, 'M'},
9224 		{"numeric",		no_argument,		NULL, 'N'},
9225 		{"option",		required_argument,	NULL, 'o'},
9226 		{"object-lookups",	no_argument,		NULL, 'O'},
9227 		{"path",		required_argument,	NULL, 'p'},
9228 		{"parseable",		no_argument,		NULL, 'P'},
9229 		{"skip-label",		no_argument,		NULL, 'q'},
9230 		{"copy-object",		no_argument,		NULL, 'r'},
9231 		{"read-block",		no_argument,		NULL, 'R'},
9232 		{"io-stats",		no_argument,		NULL, 's'},
9233 		{"simulate-dedup",	no_argument,		NULL, 'S'},
9234 		{"txg",			required_argument,	NULL, 't'},
9235 		{"brt-stats",		no_argument,		NULL, 'T'},
9236 		{"uberblock",		no_argument,		NULL, 'u'},
9237 		{"cachefile",		required_argument,	NULL, 'U'},
9238 		{"verbose",		no_argument,		NULL, 'v'},
9239 		{"verbatim",		no_argument,		NULL, 'V'},
9240 		{"dump-blocks",		required_argument,	NULL, 'x'},
9241 		{"extreme-rewind",	no_argument,		NULL, 'X'},
9242 		{"all-reconstruction",	no_argument,		NULL, 'Y'},
9243 		{"livelist",		no_argument,		NULL, 'y'},
9244 		{"zstd-headers",	no_argument,		NULL, 'Z'},
9245 		{0, 0, 0, 0}
9246 	};
9247 
9248 	while ((c = getopt_long(argc, argv,
9249 	    "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
9250 	    long_options, NULL)) != -1) {
9251 		switch (c) {
9252 		case 'b':
9253 		case 'B':
9254 		case 'c':
9255 		case 'C':
9256 		case 'd':
9257 		case 'D':
9258 		case 'E':
9259 		case 'G':
9260 		case 'h':
9261 		case 'i':
9262 		case 'l':
9263 		case 'm':
9264 		case 'M':
9265 		case 'N':
9266 		case 'O':
9267 		case 'r':
9268 		case 'R':
9269 		case 's':
9270 		case 'S':
9271 		case 'T':
9272 		case 'u':
9273 		case 'y':
9274 		case 'Z':
9275 			dump_opt[c]++;
9276 			dump_all = 0;
9277 			break;
9278 		case 'A':
9279 		case 'e':
9280 		case 'F':
9281 		case 'k':
9282 		case 'L':
9283 		case 'P':
9284 		case 'q':
9285 		case 'X':
9286 			dump_opt[c]++;
9287 			break;
9288 		case 'Y':
9289 			zfs_reconstruct_indirect_combinations_max = INT_MAX;
9290 			zfs_deadman_enabled = 0;
9291 			break;
9292 		/* NB: Sort single match options below. */
9293 		case 'I':
9294 			max_inflight_bytes = strtoull(optarg, NULL, 0);
9295 			if (max_inflight_bytes == 0) {
9296 				(void) fprintf(stderr, "maximum number "
9297 				    "of inflight bytes must be greater "
9298 				    "than 0\n");
9299 				usage();
9300 			}
9301 			break;
9302 		case 'K':
9303 			dump_opt[c]++;
9304 			key_material = strdup(optarg);
9305 			/* redact key material in process table */
9306 			while (*optarg != '\0') { *optarg++ = '*'; }
9307 			break;
9308 		case 'o':
9309 			error = set_global_var(optarg);
9310 			if (error != 0)
9311 				usage();
9312 			break;
9313 		case 'p':
9314 			if (searchdirs == NULL) {
9315 				searchdirs = umem_alloc(sizeof (char *),
9316 				    UMEM_NOFAIL);
9317 			} else {
9318 				char **tmp = umem_alloc((nsearch + 1) *
9319 				    sizeof (char *), UMEM_NOFAIL);
9320 				memcpy(tmp, searchdirs, nsearch *
9321 				    sizeof (char *));
9322 				umem_free(searchdirs,
9323 				    nsearch * sizeof (char *));
9324 				searchdirs = tmp;
9325 			}
9326 			searchdirs[nsearch++] = optarg;
9327 			break;
9328 		case 't':
9329 			max_txg = strtoull(optarg, NULL, 0);
9330 			if (max_txg < TXG_INITIAL) {
9331 				(void) fprintf(stderr, "incorrect txg "
9332 				    "specified: %s\n", optarg);
9333 				usage();
9334 			}
9335 			break;
9336 		case 'U':
9337 			config_path_console = B_TRUE;
9338 			spa_config_path = optarg;
9339 			if (spa_config_path[0] != '/') {
9340 				(void) fprintf(stderr,
9341 				    "cachefile must be an absolute path "
9342 				    "(i.e. start with a slash)\n");
9343 				usage();
9344 			}
9345 			break;
9346 		case 'v':
9347 			verbose++;
9348 			break;
9349 		case 'V':
9350 			flags = ZFS_IMPORT_VERBATIM;
9351 			break;
9352 		case 'x':
9353 			vn_dumpdir = optarg;
9354 			break;
9355 		default:
9356 			usage();
9357 			break;
9358 		}
9359 	}
9360 
9361 	if (!dump_opt['e'] && searchdirs != NULL) {
9362 		(void) fprintf(stderr, "-p option requires use of -e\n");
9363 		usage();
9364 	}
9365 #if defined(_LP64)
9366 	/*
9367 	 * ZDB does not typically re-read blocks; therefore limit the ARC
9368 	 * to 256 MB, which can be used entirely for metadata.
9369 	 */
9370 	zfs_arc_min = 2ULL << SPA_MAXBLOCKSHIFT;
9371 	zfs_arc_max = 256 * 1024 * 1024;
9372 #endif
9373 
9374 	/*
9375 	 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
9376 	 * "zdb -b" uses traversal prefetch which uses async reads.
9377 	 * For good performance, let several of them be active at once.
9378 	 */
9379 	zfs_vdev_async_read_max_active = 10;
9380 
9381 	/*
9382 	 * Disable reference tracking for better performance.
9383 	 */
9384 	reference_tracking_enable = B_FALSE;
9385 
9386 	/*
9387 	 * Do not fail spa_load when spa_load_verify fails. This is needed
9388 	 * to load non-idle pools.
9389 	 */
9390 	spa_load_verify_dryrun = B_TRUE;
9391 
9392 	/*
9393 	 * ZDB should have ability to read spacemaps.
9394 	 */
9395 	spa_mode_readable_spacemaps = B_TRUE;
9396 
9397 	if (dump_all)
9398 		verbose = MAX(verbose, 1);
9399 
9400 	for (c = 0; c < 256; c++) {
9401 		if (dump_all && strchr("ABeEFkKlLNOPrRSXy", c) == NULL)
9402 			dump_opt[c] = 1;
9403 		if (dump_opt[c])
9404 			dump_opt[c] += verbose;
9405 	}
9406 
9407 	libspl_set_assert_ok((dump_opt['A'] == 1) || (dump_opt['A'] > 2));
9408 	zfs_recover = (dump_opt['A'] > 1);
9409 
9410 	argc -= optind;
9411 	argv += optind;
9412 	if (argc < 2 && dump_opt['R'])
9413 		usage();
9414 
9415 	target = argv[0];
9416 
9417 	/*
9418 	 * Automate cachefile
9419 	 */
9420 	if (!spa_config_path_env && !config_path_console && target &&
9421 	    libzfs_core_init() == 0) {
9422 		char *pname = strdup(target);
9423 		const char *value;
9424 		nvlist_t *pnvl = NULL;
9425 		nvlist_t *vnvl = NULL;
9426 
9427 		if (strpbrk(pname, "/@") != NULL)
9428 			*strpbrk(pname, "/@") = '\0';
9429 
9430 		if (pname && lzc_get_props(pname, &pnvl) == 0) {
9431 			if (nvlist_lookup_nvlist(pnvl, "cachefile",
9432 			    &vnvl) == 0) {
9433 				value = fnvlist_lookup_string(vnvl,
9434 				    ZPROP_VALUE);
9435 			} else {
9436 				value = "-";
9437 			}
9438 			strlcpy(pbuf, value, sizeof (pbuf));
9439 			if (pbuf[0] != '\0') {
9440 				if (pbuf[0] == '/') {
9441 					if (access(pbuf, F_OK) == 0)
9442 						spa_config_path = pbuf;
9443 					else
9444 						force_import = B_TRUE;
9445 				} else if ((strcmp(pbuf, "-") == 0 &&
9446 				    access(ZPOOL_CACHE, F_OK) != 0) ||
9447 				    strcmp(pbuf, "none") == 0) {
9448 					force_import = B_TRUE;
9449 				}
9450 			}
9451 			nvlist_free(vnvl);
9452 		}
9453 
9454 		free(pname);
9455 		nvlist_free(pnvl);
9456 		libzfs_core_fini();
9457 	}
9458 
9459 	dmu_objset_register_type(DMU_OST_ZFS, dummy_get_file_info);
9460 	kernel_init(SPA_MODE_READ);
9461 	kernel_init_done = B_TRUE;
9462 
9463 	if (dump_opt['E']) {
9464 		if (argc != 1)
9465 			usage();
9466 		zdb_embedded_block(argv[0]);
9467 		error = 0;
9468 		goto fini;
9469 	}
9470 
9471 	if (argc < 1) {
9472 		if (!dump_opt['e'] && dump_opt['C']) {
9473 			dump_cachefile(spa_config_path);
9474 			error = 0;
9475 			goto fini;
9476 		}
9477 		usage();
9478 	}
9479 
9480 	if (dump_opt['l']) {
9481 		error = dump_label(argv[0]);
9482 		goto fini;
9483 	}
9484 
9485 	if (dump_opt['X'] || dump_opt['F'])
9486 		rewind = ZPOOL_DO_REWIND |
9487 		    (dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
9488 
9489 	/* -N implies -d */
9490 	if (dump_opt['N'] && dump_opt['d'] == 0)
9491 		dump_opt['d'] = dump_opt['N'];
9492 
9493 	if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 ||
9494 	    nvlist_add_uint64(policy, ZPOOL_LOAD_REQUEST_TXG, max_txg) != 0 ||
9495 	    nvlist_add_uint32(policy, ZPOOL_LOAD_REWIND_POLICY, rewind) != 0)
9496 		fatal("internal error: %s", strerror(ENOMEM));
9497 
9498 	error = 0;
9499 
9500 	if (strpbrk(target, "/@") != NULL) {
9501 		size_t targetlen;
9502 
9503 		target_pool = strdup(target);
9504 		*strpbrk(target_pool, "/@") = '\0';
9505 
9506 		target_is_spa = B_FALSE;
9507 		targetlen = strlen(target);
9508 		if (targetlen && target[targetlen - 1] == '/')
9509 			target[targetlen - 1] = '\0';
9510 
9511 		/*
9512 		 * See if an objset ID was supplied (-d <pool>/<objset ID>).
9513 		 * To disambiguate tank/100, consider the 100 as objsetID
9514 		 * if -N was given, otherwise 100 is an objsetID iff
9515 		 * tank/100 as a named dataset fails on lookup.
9516 		 */
9517 		objset_str = strchr(target, '/');
9518 		if (objset_str && strlen(objset_str) > 1 &&
9519 		    zdb_numeric(objset_str + 1)) {
9520 			char *endptr;
9521 			errno = 0;
9522 			objset_str++;
9523 			objset_id = strtoull(objset_str, &endptr, 0);
9524 			/* dataset 0 is the same as opening the pool */
9525 			if (errno == 0 && endptr != objset_str &&
9526 			    objset_id != 0) {
9527 				if (dump_opt['N'])
9528 					dataset_lookup = B_TRUE;
9529 			}
9530 			/* normal dataset name not an objset ID */
9531 			if (endptr == objset_str) {
9532 				objset_id = -1;
9533 			}
9534 		} else if (objset_str && !zdb_numeric(objset_str + 1) &&
9535 		    dump_opt['N']) {
9536 			printf("Supply a numeric objset ID with -N\n");
9537 			error = 1;
9538 			goto fini;
9539 		}
9540 	} else {
9541 		target_pool = target;
9542 	}
9543 
9544 	if (dump_opt['e'] || force_import) {
9545 		importargs_t args = { 0 };
9546 
9547 		/*
9548 		 * If path is not provided, search in /dev
9549 		 */
9550 		if (searchdirs == NULL) {
9551 			searchdirs = umem_alloc(sizeof (char *), UMEM_NOFAIL);
9552 			searchdirs[nsearch++] = (char *)ZFS_DEVDIR;
9553 		}
9554 
9555 		args.paths = nsearch;
9556 		args.path = searchdirs;
9557 		args.can_be_active = B_TRUE;
9558 
9559 		libpc_handle_t lpch = {
9560 			.lpc_lib_handle = NULL,
9561 			.lpc_ops = &libzpool_config_ops,
9562 			.lpc_printerr = B_TRUE
9563 		};
9564 		error = zpool_find_config(&lpch, target_pool, &cfg, &args);
9565 
9566 		if (error == 0) {
9567 
9568 			if (nvlist_add_nvlist(cfg,
9569 			    ZPOOL_LOAD_POLICY, policy) != 0) {
9570 				fatal("can't open '%s': %s",
9571 				    target, strerror(ENOMEM));
9572 			}
9573 
9574 			if (dump_opt['C'] > 1) {
9575 				(void) printf("\nConfiguration for import:\n");
9576 				dump_nvlist(cfg, 8);
9577 			}
9578 
9579 			/*
9580 			 * Disable the activity check to allow examination of
9581 			 * active pools.
9582 			 */
9583 			error = spa_import(target_pool, cfg, NULL,
9584 			    flags | ZFS_IMPORT_SKIP_MMP);
9585 		}
9586 	}
9587 
9588 	if (searchdirs != NULL) {
9589 		umem_free(searchdirs, nsearch * sizeof (char *));
9590 		searchdirs = NULL;
9591 	}
9592 
9593 	/*
9594 	 * We need to make sure to process -O option or call
9595 	 * dump_path after the -e option has been processed,
9596 	 * which imports the pool to the namespace if it's
9597 	 * not in the cachefile.
9598 	 */
9599 	if (dump_opt['O']) {
9600 		if (argc != 2)
9601 			usage();
9602 		dump_opt['v'] = verbose + 3;
9603 		error = dump_path(argv[0], argv[1], NULL);
9604 		goto fini;
9605 	}
9606 
9607 	if (dump_opt['r']) {
9608 		target_is_spa = B_FALSE;
9609 		if (argc != 3)
9610 			usage();
9611 		dump_opt['v'] = verbose;
9612 		error = dump_path(argv[0], argv[1], &object);
9613 		if (error != 0)
9614 			fatal("internal error: %s", strerror(error));
9615 	}
9616 
9617 	/*
9618 	 * import_checkpointed_state makes the assumption that the
9619 	 * target pool that we pass it is already part of the spa
9620 	 * namespace. Because of that we need to make sure to call
9621 	 * it always after the -e option has been processed, which
9622 	 * imports the pool to the namespace if it's not in the
9623 	 * cachefile.
9624 	 */
9625 	char *checkpoint_pool = NULL;
9626 	char *checkpoint_target = NULL;
9627 	if (dump_opt['k']) {
9628 		checkpoint_pool = import_checkpointed_state(target, cfg,
9629 		    &checkpoint_target);
9630 
9631 		if (checkpoint_target != NULL)
9632 			target = checkpoint_target;
9633 	}
9634 
9635 	if (cfg != NULL) {
9636 		nvlist_free(cfg);
9637 		cfg = NULL;
9638 	}
9639 
9640 	if (target_pool != target)
9641 		free(target_pool);
9642 
9643 	if (error == 0) {
9644 		if (dump_opt['k'] && (target_is_spa || dump_opt['R'])) {
9645 			ASSERT(checkpoint_pool != NULL);
9646 			ASSERT(checkpoint_target == NULL);
9647 
9648 			error = spa_open(checkpoint_pool, &spa, FTAG);
9649 			if (error != 0) {
9650 				fatal("Tried to open pool \"%s\" but "
9651 				    "spa_open() failed with error %d\n",
9652 				    checkpoint_pool, error);
9653 			}
9654 
9655 		} else if (target_is_spa || dump_opt['R'] || dump_opt['B'] ||
9656 		    objset_id == 0) {
9657 			zdb_set_skip_mmp(target);
9658 			error = spa_open_rewind(target, &spa, FTAG, policy,
9659 			    NULL);
9660 			if (error) {
9661 				/*
9662 				 * If we're missing the log device then
9663 				 * try opening the pool after clearing the
9664 				 * log state.
9665 				 */
9666 				mutex_enter(&spa_namespace_lock);
9667 				if ((spa = spa_lookup(target)) != NULL &&
9668 				    spa->spa_log_state == SPA_LOG_MISSING) {
9669 					spa->spa_log_state = SPA_LOG_CLEAR;
9670 					error = 0;
9671 				}
9672 				mutex_exit(&spa_namespace_lock);
9673 
9674 				if (!error) {
9675 					error = spa_open_rewind(target, &spa,
9676 					    FTAG, policy, NULL);
9677 				}
9678 			}
9679 		} else if (strpbrk(target, "#") != NULL) {
9680 			dsl_pool_t *dp;
9681 			error = dsl_pool_hold(target, FTAG, &dp);
9682 			if (error != 0) {
9683 				fatal("can't dump '%s': %s", target,
9684 				    strerror(error));
9685 			}
9686 			error = dump_bookmark(dp, target, B_TRUE, verbose > 1);
9687 			dsl_pool_rele(dp, FTAG);
9688 			if (error != 0) {
9689 				fatal("can't dump '%s': %s", target,
9690 				    strerror(error));
9691 			}
9692 			goto fini;
9693 		} else {
9694 			target_pool = strdup(target);
9695 			if (strpbrk(target, "/@") != NULL)
9696 				*strpbrk(target_pool, "/@") = '\0';
9697 
9698 			zdb_set_skip_mmp(target);
9699 			/*
9700 			 * If -N was supplied, the user has indicated that
9701 			 * zdb -d <pool>/<objsetID> is in effect.  Otherwise
9702 			 * we first assume that the dataset string is the
9703 			 * dataset name.  If dmu_objset_hold fails with the
9704 			 * dataset string, and we have an objset_id, retry the
9705 			 * lookup with the objsetID.
9706 			 */
9707 			boolean_t retry = B_TRUE;
9708 retry_lookup:
9709 			if (dataset_lookup == B_TRUE) {
9710 				/*
9711 				 * Use the supplied id to get the name
9712 				 * for open_objset.
9713 				 */
9714 				error = spa_open(target_pool, &spa, FTAG);
9715 				if (error == 0) {
9716 					error = name_from_objset_id(spa,
9717 					    objset_id, dsname);
9718 					spa_close(spa, FTAG);
9719 					if (error == 0)
9720 						target = dsname;
9721 				}
9722 			}
9723 			if (error == 0) {
9724 				if (objset_id > 0 && retry) {
9725 					int err = dmu_objset_hold(target, FTAG,
9726 					    &os);
9727 					if (err) {
9728 						dataset_lookup = B_TRUE;
9729 						retry = B_FALSE;
9730 						goto retry_lookup;
9731 					} else {
9732 						dmu_objset_rele(os, FTAG);
9733 					}
9734 				}
9735 				error = open_objset(target, FTAG, &os);
9736 			}
9737 			if (error == 0)
9738 				spa = dmu_objset_spa(os);
9739 			free(target_pool);
9740 		}
9741 	}
9742 	nvlist_free(policy);
9743 
9744 	if (error)
9745 		fatal("can't open '%s': %s", target, strerror(error));
9746 
9747 	/*
9748 	 * Set the pool failure mode to panic in order to prevent the pool
9749 	 * from suspending.  A suspended I/O will have no way to resume and
9750 	 * can prevent the zdb(8) command from terminating as expected.
9751 	 */
9752 	if (spa != NULL)
9753 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
9754 
9755 	argv++;
9756 	argc--;
9757 	if (dump_opt['r']) {
9758 		error = zdb_copy_object(os, object, argv[1]);
9759 	} else if (!dump_opt['R']) {
9760 		flagbits['d'] = ZOR_FLAG_DIRECTORY;
9761 		flagbits['f'] = ZOR_FLAG_PLAIN_FILE;
9762 		flagbits['m'] = ZOR_FLAG_SPACE_MAP;
9763 		flagbits['z'] = ZOR_FLAG_ZAP;
9764 		flagbits['A'] = ZOR_FLAG_ALL_TYPES;
9765 
9766 		if (argc > 0 && dump_opt['d']) {
9767 			zopt_object_args = argc;
9768 			zopt_object_ranges = calloc(zopt_object_args,
9769 			    sizeof (zopt_object_range_t));
9770 			for (unsigned i = 0; i < zopt_object_args; i++) {
9771 				int err;
9772 				const char *msg = NULL;
9773 
9774 				err = parse_object_range(argv[i],
9775 				    &zopt_object_ranges[i], &msg);
9776 				if (err != 0)
9777 					fatal("Bad object or range: '%s': %s\n",
9778 					    argv[i], msg ?: "");
9779 			}
9780 		} else if (argc > 0 && dump_opt['m']) {
9781 			zopt_metaslab_args = argc;
9782 			zopt_metaslab = calloc(zopt_metaslab_args,
9783 			    sizeof (uint64_t));
9784 			for (unsigned i = 0; i < zopt_metaslab_args; i++) {
9785 				errno = 0;
9786 				zopt_metaslab[i] = strtoull(argv[i], NULL, 0);
9787 				if (zopt_metaslab[i] == 0 && errno != 0)
9788 					fatal("bad number %s: %s", argv[i],
9789 					    strerror(errno));
9790 			}
9791 		}
9792 		if (dump_opt['B']) {
9793 			dump_backup(target, objset_id,
9794 			    argc > 0 ? argv[0] : NULL);
9795 		} else if (os != NULL) {
9796 			dump_objset(os);
9797 		} else if (zopt_object_args > 0 && !dump_opt['m']) {
9798 			dump_objset(spa->spa_meta_objset);
9799 		} else {
9800 			dump_zpool(spa);
9801 		}
9802 	} else {
9803 		flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR;
9804 		flagbits['c'] = ZDB_FLAG_CHECKSUM;
9805 		flagbits['d'] = ZDB_FLAG_DECOMPRESS;
9806 		flagbits['e'] = ZDB_FLAG_BSWAP;
9807 		flagbits['g'] = ZDB_FLAG_GBH;
9808 		flagbits['i'] = ZDB_FLAG_INDIRECT;
9809 		flagbits['r'] = ZDB_FLAG_RAW;
9810 		flagbits['v'] = ZDB_FLAG_VERBOSE;
9811 
9812 		for (int i = 0; i < argc; i++)
9813 			zdb_read_block(argv[i], spa);
9814 	}
9815 
9816 	if (dump_opt['k']) {
9817 		free(checkpoint_pool);
9818 		if (!target_is_spa)
9819 			free(checkpoint_target);
9820 	}
9821 
9822 fini:
9823 	if (spa != NULL)
9824 		zdb_ddt_cleanup(spa);
9825 
9826 	if (os != NULL) {
9827 		close_objset(os, FTAG);
9828 	} else if (spa != NULL) {
9829 		spa_close(spa, FTAG);
9830 	}
9831 
9832 	fuid_table_destroy();
9833 
9834 	dump_debug_buffer();
9835 
9836 	if (kernel_init_done)
9837 		kernel_fini();
9838 
9839 	return (error);
9840 }
9841