xref: /freebsd/sys/contrib/openzfs/cmd/zdb/zdb.c (revision 61145dc2b94f12f6a47344fb9aac702321880e43)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 
23 /*
24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2016 Nexenta Systems, Inc.
28  * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC.
29  * Copyright (c) 2015, 2017, Intel Corporation.
30  * Copyright (c) 2020 Datto Inc.
31  * Copyright (c) 2020, The FreeBSD Foundation [1]
32  *
33  * [1] Portions of this software were developed by Allan Jude
34  *     under sponsorship from the FreeBSD Foundation.
35  * Copyright (c) 2021 Allan Jude
36  * Copyright (c) 2021 Toomas Soome <tsoome@me.com>
37  * Copyright (c) 2023, 2024, Klara Inc.
38  * Copyright (c) 2023, Rob Norris <robn@despairlabs.com>
39  */
40 
41 #include <stdio.h>
42 #include <unistd.h>
43 #include <stdlib.h>
44 #include <ctype.h>
45 #include <getopt.h>
46 #include <openssl/evp.h>
47 #include <sys/zfs_context.h>
48 #include <sys/spa.h>
49 #include <sys/spa_impl.h>
50 #include <sys/dmu.h>
51 #include <sys/zap.h>
52 #include <sys/zap_impl.h>
53 #include <sys/fs/zfs.h>
54 #include <sys/zfs_znode.h>
55 #include <sys/zfs_sa.h>
56 #include <sys/sa.h>
57 #include <sys/sa_impl.h>
58 #include <sys/vdev.h>
59 #include <sys/vdev_impl.h>
60 #include <sys/metaslab_impl.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/dsl_dir.h>
63 #include <sys/dsl_dataset.h>
64 #include <sys/dsl_pool.h>
65 #include <sys/dsl_bookmark.h>
66 #include <sys/dbuf.h>
67 #include <sys/zil.h>
68 #include <sys/zil_impl.h>
69 #include <sys/stat.h>
70 #include <sys/resource.h>
71 #include <sys/dmu_send.h>
72 #include <sys/dmu_traverse.h>
73 #include <sys/zio_checksum.h>
74 #include <sys/zio_compress.h>
75 #include <sys/zfs_fuid.h>
76 #include <sys/arc.h>
77 #include <sys/arc_impl.h>
78 #include <sys/ddt.h>
79 #include <sys/ddt_impl.h>
80 #include <sys/zfeature.h>
81 #include <sys/abd.h>
82 #include <sys/blkptr.h>
83 #include <sys/dsl_crypt.h>
84 #include <sys/dsl_scan.h>
85 #include <sys/btree.h>
86 #include <sys/brt.h>
87 #include <sys/brt_impl.h>
88 #include <zfs_comutil.h>
89 #include <sys/zstd/zstd.h>
90 #include <sys/backtrace.h>
91 
92 #include <libnvpair.h>
93 #include <libzutil.h>
94 #include <libzfs_core.h>
95 
96 #include <libzdb.h>
97 
98 #include "zdb.h"
99 
100 
101 extern int reference_tracking_enable;
102 extern int zfs_recover;
103 extern uint_t zfs_vdev_async_read_max_active;
104 extern boolean_t spa_load_verify_dryrun;
105 extern boolean_t spa_mode_readable_spacemaps;
106 extern uint_t zfs_reconstruct_indirect_combinations_max;
107 extern uint_t zfs_btree_verify_intensity;
108 
109 static const char cmdname[] = "zdb";
110 uint8_t dump_opt[256];
111 
112 typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size);
113 
114 static uint64_t *zopt_metaslab = NULL;
115 static unsigned zopt_metaslab_args = 0;
116 
117 
118 static zopt_object_range_t *zopt_object_ranges = NULL;
119 static unsigned zopt_object_args = 0;
120 
121 static int flagbits[256];
122 
123 
124 static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */
125 static int leaked_objects = 0;
126 static zfs_range_tree_t *mos_refd_objs;
127 static spa_t *spa;
128 static objset_t *os;
129 static boolean_t kernel_init_done;
130 
131 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *,
132     boolean_t);
133 static void mos_obj_refd(uint64_t);
134 static void mos_obj_refd_multiple(uint64_t);
135 static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free,
136     dmu_tx_t *tx);
137 
138 
139 
140 static void zdb_print_blkptr(const blkptr_t *bp, int flags);
141 static void zdb_exit(int reason);
142 
143 typedef struct sublivelist_verify_block_refcnt {
144 	/* block pointer entry in livelist being verified */
145 	blkptr_t svbr_blk;
146 
147 	/*
148 	 * Refcount gets incremented to 1 when we encounter the first
149 	 * FREE entry for the svfbr block pointer and a node for it
150 	 * is created in our ZDB verification/tracking metadata.
151 	 *
152 	 * As we encounter more FREE entries we increment this counter
153 	 * and similarly decrement it whenever we find the respective
154 	 * ALLOC entries for this block.
155 	 *
156 	 * When the refcount gets to 0 it means that all the FREE and
157 	 * ALLOC entries of this block have paired up and we no longer
158 	 * need to track it in our verification logic (e.g. the node
159 	 * containing this struct in our verification data structure
160 	 * should be freed).
161 	 *
162 	 * [refer to sublivelist_verify_blkptr() for the actual code]
163 	 */
164 	uint32_t svbr_refcnt;
165 } sublivelist_verify_block_refcnt_t;
166 
167 static int
sublivelist_block_refcnt_compare(const void * larg,const void * rarg)168 sublivelist_block_refcnt_compare(const void *larg, const void *rarg)
169 {
170 	const sublivelist_verify_block_refcnt_t *l = larg;
171 	const sublivelist_verify_block_refcnt_t *r = rarg;
172 	return (livelist_compare(&l->svbr_blk, &r->svbr_blk));
173 }
174 
175 static int
sublivelist_verify_blkptr(void * arg,const blkptr_t * bp,boolean_t free,dmu_tx_t * tx)176 sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free,
177     dmu_tx_t *tx)
178 {
179 	ASSERT3P(tx, ==, NULL);
180 	struct sublivelist_verify *sv = arg;
181 	sublivelist_verify_block_refcnt_t current = {
182 			.svbr_blk = *bp,
183 
184 			/*
185 			 * Start with 1 in case this is the first free entry.
186 			 * This field is not used for our B-Tree comparisons
187 			 * anyway.
188 			 */
189 			.svbr_refcnt = 1,
190 	};
191 
192 	zfs_btree_index_t where;
193 	sublivelist_verify_block_refcnt_t *pair =
194 	    zfs_btree_find(&sv->sv_pair, &current, &where);
195 	if (free) {
196 		if (pair == NULL) {
197 			/* first free entry for this block pointer */
198 			zfs_btree_add(&sv->sv_pair, &current);
199 		} else {
200 			pair->svbr_refcnt++;
201 		}
202 	} else {
203 		if (pair == NULL) {
204 			/* block that is currently marked as allocated */
205 			for (int i = 0; i < SPA_DVAS_PER_BP; i++) {
206 				if (DVA_IS_EMPTY(&bp->blk_dva[i]))
207 					break;
208 				sublivelist_verify_block_t svb = {
209 				    .svb_dva = bp->blk_dva[i],
210 				    .svb_allocated_txg =
211 				    BP_GET_LOGICAL_BIRTH(bp)
212 				};
213 
214 				if (zfs_btree_find(&sv->sv_leftover, &svb,
215 				    &where) == NULL) {
216 					zfs_btree_add_idx(&sv->sv_leftover,
217 					    &svb, &where);
218 				}
219 			}
220 		} else {
221 			/* alloc matches a free entry */
222 			pair->svbr_refcnt--;
223 			if (pair->svbr_refcnt == 0) {
224 				/* all allocs and frees have been matched */
225 				zfs_btree_remove_idx(&sv->sv_pair, &where);
226 			}
227 		}
228 	}
229 
230 	return (0);
231 }
232 
233 static int
sublivelist_verify_func(void * args,dsl_deadlist_entry_t * dle)234 sublivelist_verify_func(void *args, dsl_deadlist_entry_t *dle)
235 {
236 	int err;
237 	struct sublivelist_verify *sv = args;
238 
239 	zfs_btree_create(&sv->sv_pair, sublivelist_block_refcnt_compare, NULL,
240 	    sizeof (sublivelist_verify_block_refcnt_t));
241 
242 	err = bpobj_iterate_nofree(&dle->dle_bpobj, sublivelist_verify_blkptr,
243 	    sv, NULL);
244 
245 	sublivelist_verify_block_refcnt_t *e;
246 	zfs_btree_index_t *cookie = NULL;
247 	while ((e = zfs_btree_destroy_nodes(&sv->sv_pair, &cookie)) != NULL) {
248 		char blkbuf[BP_SPRINTF_LEN];
249 		snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
250 		    &e->svbr_blk, B_TRUE);
251 		(void) printf("\tERROR: %d unmatched FREE(s): %s\n",
252 		    e->svbr_refcnt, blkbuf);
253 	}
254 	zfs_btree_destroy(&sv->sv_pair);
255 
256 	return (err);
257 }
258 
259 static int
livelist_block_compare(const void * larg,const void * rarg)260 livelist_block_compare(const void *larg, const void *rarg)
261 {
262 	const sublivelist_verify_block_t *l = larg;
263 	const sublivelist_verify_block_t *r = rarg;
264 
265 	if (DVA_GET_VDEV(&l->svb_dva) < DVA_GET_VDEV(&r->svb_dva))
266 		return (-1);
267 	else if (DVA_GET_VDEV(&l->svb_dva) > DVA_GET_VDEV(&r->svb_dva))
268 		return (+1);
269 
270 	if (DVA_GET_OFFSET(&l->svb_dva) < DVA_GET_OFFSET(&r->svb_dva))
271 		return (-1);
272 	else if (DVA_GET_OFFSET(&l->svb_dva) > DVA_GET_OFFSET(&r->svb_dva))
273 		return (+1);
274 
275 	if (DVA_GET_ASIZE(&l->svb_dva) < DVA_GET_ASIZE(&r->svb_dva))
276 		return (-1);
277 	else if (DVA_GET_ASIZE(&l->svb_dva) > DVA_GET_ASIZE(&r->svb_dva))
278 		return (+1);
279 
280 	return (0);
281 }
282 
283 /*
284  * Check for errors in a livelist while tracking all unfreed ALLOCs in the
285  * sublivelist_verify_t: sv->sv_leftover
286  */
287 static void
livelist_verify(dsl_deadlist_t * dl,void * arg)288 livelist_verify(dsl_deadlist_t *dl, void *arg)
289 {
290 	sublivelist_verify_t *sv = arg;
291 	dsl_deadlist_iterate(dl, sublivelist_verify_func, sv);
292 }
293 
294 /*
295  * Check for errors in the livelist entry and discard the intermediary
296  * data structures
297  */
298 static int
sublivelist_verify_lightweight(void * args,dsl_deadlist_entry_t * dle)299 sublivelist_verify_lightweight(void *args, dsl_deadlist_entry_t *dle)
300 {
301 	(void) args;
302 	sublivelist_verify_t sv;
303 	zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
304 	    sizeof (sublivelist_verify_block_t));
305 	int err = sublivelist_verify_func(&sv, dle);
306 	zfs_btree_clear(&sv.sv_leftover);
307 	zfs_btree_destroy(&sv.sv_leftover);
308 	return (err);
309 }
310 
311 typedef struct metaslab_verify {
312 	/*
313 	 * Tree containing all the leftover ALLOCs from the livelists
314 	 * that are part of this metaslab.
315 	 */
316 	zfs_btree_t mv_livelist_allocs;
317 
318 	/*
319 	 * Metaslab information.
320 	 */
321 	uint64_t mv_vdid;
322 	uint64_t mv_msid;
323 	uint64_t mv_start;
324 	uint64_t mv_end;
325 
326 	/*
327 	 * What's currently allocated for this metaslab.
328 	 */
329 	zfs_range_tree_t *mv_allocated;
330 } metaslab_verify_t;
331 
332 typedef void ll_iter_t(dsl_deadlist_t *ll, void *arg);
333 
334 typedef int (*zdb_log_sm_cb_t)(spa_t *spa, space_map_entry_t *sme, uint64_t txg,
335     void *arg);
336 
337 typedef struct unflushed_iter_cb_arg {
338 	spa_t *uic_spa;
339 	uint64_t uic_txg;
340 	void *uic_arg;
341 	zdb_log_sm_cb_t uic_cb;
342 } unflushed_iter_cb_arg_t;
343 
344 static int
iterate_through_spacemap_logs_cb(space_map_entry_t * sme,void * arg)345 iterate_through_spacemap_logs_cb(space_map_entry_t *sme, void *arg)
346 {
347 	unflushed_iter_cb_arg_t *uic = arg;
348 	return (uic->uic_cb(uic->uic_spa, sme, uic->uic_txg, uic->uic_arg));
349 }
350 
351 static void
iterate_through_spacemap_logs(spa_t * spa,zdb_log_sm_cb_t cb,void * arg)352 iterate_through_spacemap_logs(spa_t *spa, zdb_log_sm_cb_t cb, void *arg)
353 {
354 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
355 		return;
356 
357 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
358 	for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
359 	    sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
360 		space_map_t *sm = NULL;
361 		VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
362 		    sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
363 
364 		unflushed_iter_cb_arg_t uic = {
365 			.uic_spa = spa,
366 			.uic_txg = sls->sls_txg,
367 			.uic_arg = arg,
368 			.uic_cb = cb
369 		};
370 		VERIFY0(space_map_iterate(sm, space_map_length(sm),
371 		    iterate_through_spacemap_logs_cb, &uic));
372 		space_map_close(sm);
373 	}
374 	spa_config_exit(spa, SCL_CONFIG, FTAG);
375 }
376 
377 static void
verify_livelist_allocs(metaslab_verify_t * mv,uint64_t txg,uint64_t offset,uint64_t size)378 verify_livelist_allocs(metaslab_verify_t *mv, uint64_t txg,
379     uint64_t offset, uint64_t size)
380 {
381 	sublivelist_verify_block_t svb = {{{0}}};
382 	DVA_SET_VDEV(&svb.svb_dva, mv->mv_vdid);
383 	DVA_SET_OFFSET(&svb.svb_dva, offset);
384 	DVA_SET_ASIZE(&svb.svb_dva, size);
385 	zfs_btree_index_t where;
386 	uint64_t end_offset = offset + size;
387 
388 	/*
389 	 *  Look for an exact match for spacemap entry in the livelist entries.
390 	 *  Then, look for other livelist entries that fall within the range
391 	 *  of the spacemap entry as it may have been condensed
392 	 */
393 	sublivelist_verify_block_t *found =
394 	    zfs_btree_find(&mv->mv_livelist_allocs, &svb, &where);
395 	if (found == NULL) {
396 		found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where);
397 	}
398 	for (; found != NULL && DVA_GET_VDEV(&found->svb_dva) == mv->mv_vdid &&
399 	    DVA_GET_OFFSET(&found->svb_dva) < end_offset;
400 	    found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
401 		if (found->svb_allocated_txg <= txg) {
402 			(void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
403 			    "from TXG %llx FREED at TXG %llx\n",
404 			    (u_longlong_t)DVA_GET_OFFSET(&found->svb_dva),
405 			    (u_longlong_t)DVA_GET_ASIZE(&found->svb_dva),
406 			    (u_longlong_t)found->svb_allocated_txg,
407 			    (u_longlong_t)txg);
408 		}
409 	}
410 }
411 
412 static int
metaslab_spacemap_validation_cb(space_map_entry_t * sme,void * arg)413 metaslab_spacemap_validation_cb(space_map_entry_t *sme, void *arg)
414 {
415 	metaslab_verify_t *mv = arg;
416 	uint64_t offset = sme->sme_offset;
417 	uint64_t size = sme->sme_run;
418 	uint64_t txg = sme->sme_txg;
419 
420 	if (sme->sme_type == SM_ALLOC) {
421 		if (zfs_range_tree_contains(mv->mv_allocated,
422 		    offset, size)) {
423 			(void) printf("ERROR: DOUBLE ALLOC: "
424 			    "%llu [%llx:%llx] "
425 			    "%llu:%llu LOG_SM\n",
426 			    (u_longlong_t)txg, (u_longlong_t)offset,
427 			    (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
428 			    (u_longlong_t)mv->mv_msid);
429 		} else {
430 			zfs_range_tree_add(mv->mv_allocated,
431 			    offset, size);
432 		}
433 	} else {
434 		if (!zfs_range_tree_contains(mv->mv_allocated,
435 		    offset, size)) {
436 			(void) printf("ERROR: DOUBLE FREE: "
437 			    "%llu [%llx:%llx] "
438 			    "%llu:%llu LOG_SM\n",
439 			    (u_longlong_t)txg, (u_longlong_t)offset,
440 			    (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
441 			    (u_longlong_t)mv->mv_msid);
442 		} else {
443 			zfs_range_tree_remove(mv->mv_allocated,
444 			    offset, size);
445 		}
446 	}
447 
448 	if (sme->sme_type != SM_ALLOC) {
449 		/*
450 		 * If something is freed in the spacemap, verify that
451 		 * it is not listed as allocated in the livelist.
452 		 */
453 		verify_livelist_allocs(mv, txg, offset, size);
454 	}
455 	return (0);
456 }
457 
458 static int
spacemap_check_sm_log_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)459 spacemap_check_sm_log_cb(spa_t *spa, space_map_entry_t *sme,
460     uint64_t txg, void *arg)
461 {
462 	metaslab_verify_t *mv = arg;
463 	uint64_t offset = sme->sme_offset;
464 	uint64_t vdev_id = sme->sme_vdev;
465 
466 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
467 
468 	/* skip indirect vdevs */
469 	if (!vdev_is_concrete(vd))
470 		return (0);
471 
472 	if (vdev_id != mv->mv_vdid)
473 		return (0);
474 
475 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
476 	if (ms->ms_id != mv->mv_msid)
477 		return (0);
478 
479 	if (txg < metaslab_unflushed_txg(ms))
480 		return (0);
481 
482 
483 	ASSERT3U(txg, ==, sme->sme_txg);
484 	return (metaslab_spacemap_validation_cb(sme, mv));
485 }
486 
487 static void
spacemap_check_sm_log(spa_t * spa,metaslab_verify_t * mv)488 spacemap_check_sm_log(spa_t *spa, metaslab_verify_t *mv)
489 {
490 	iterate_through_spacemap_logs(spa, spacemap_check_sm_log_cb, mv);
491 }
492 
493 static void
spacemap_check_ms_sm(space_map_t * sm,metaslab_verify_t * mv)494 spacemap_check_ms_sm(space_map_t  *sm, metaslab_verify_t *mv)
495 {
496 	if (sm == NULL)
497 		return;
498 
499 	VERIFY0(space_map_iterate(sm, space_map_length(sm),
500 	    metaslab_spacemap_validation_cb, mv));
501 }
502 
503 static void iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg);
504 
505 /*
506  * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
507  * they are part of that metaslab (mv_msid).
508  */
509 static void
mv_populate_livelist_allocs(metaslab_verify_t * mv,sublivelist_verify_t * sv)510 mv_populate_livelist_allocs(metaslab_verify_t *mv, sublivelist_verify_t *sv)
511 {
512 	zfs_btree_index_t where;
513 	sublivelist_verify_block_t *svb;
514 	ASSERT3U(zfs_btree_numnodes(&mv->mv_livelist_allocs), ==, 0);
515 	for (svb = zfs_btree_first(&sv->sv_leftover, &where);
516 	    svb != NULL;
517 	    svb = zfs_btree_next(&sv->sv_leftover, &where, &where)) {
518 		if (DVA_GET_VDEV(&svb->svb_dva) != mv->mv_vdid)
519 			continue;
520 
521 		if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start &&
522 		    (DVA_GET_OFFSET(&svb->svb_dva) +
523 		    DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_start) {
524 			(void) printf("ERROR: Found block that crosses "
525 			    "metaslab boundary: <%llu:%llx:%llx>\n",
526 			    (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
527 			    (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
528 			    (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
529 			continue;
530 		}
531 
532 		if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start)
533 			continue;
534 
535 		if (DVA_GET_OFFSET(&svb->svb_dva) >= mv->mv_end)
536 			continue;
537 
538 		if ((DVA_GET_OFFSET(&svb->svb_dva) +
539 		    DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_end) {
540 			(void) printf("ERROR: Found block that crosses "
541 			    "metaslab boundary: <%llu:%llx:%llx>\n",
542 			    (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
543 			    (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
544 			    (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
545 			continue;
546 		}
547 
548 		zfs_btree_add(&mv->mv_livelist_allocs, svb);
549 	}
550 
551 	for (svb = zfs_btree_first(&mv->mv_livelist_allocs, &where);
552 	    svb != NULL;
553 	    svb = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
554 		zfs_btree_remove(&sv->sv_leftover, svb);
555 	}
556 }
557 
558 /*
559  * [Livelist Check]
560  * Iterate through all the sublivelists and:
561  * - report leftover frees (**)
562  * - record leftover ALLOCs together with their TXG [see Cross Check]
563  *
564  * (**) Note: Double ALLOCs are valid in datasets that have dedup
565  *      enabled. Similarly double FREEs are allowed as well but
566  *      only if they pair up with a corresponding ALLOC entry once
567  *      we our done with our sublivelist iteration.
568  *
569  * [Spacemap Check]
570  * for each metaslab:
571  * - iterate over spacemap and then the metaslab's entries in the
572  *   spacemap log, then report any double FREEs and ALLOCs (do not
573  *   blow up).
574  *
575  * [Cross Check]
576  * After finishing the Livelist Check phase and while being in the
577  * Spacemap Check phase, we find all the recorded leftover ALLOCs
578  * of the livelist check that are part of the metaslab that we are
579  * currently looking at in the Spacemap Check. We report any entries
580  * that are marked as ALLOCs in the livelists but have been actually
581  * freed (and potentially allocated again) after their TXG stamp in
582  * the spacemaps. Also report any ALLOCs from the livelists that
583  * belong to indirect vdevs (e.g. their vdev completed removal).
584  *
585  * Note that this will miss Log Spacemap entries that cancelled each other
586  * out before being flushed to the metaslab, so we are not guaranteed
587  * to match all erroneous ALLOCs.
588  */
589 static void
livelist_metaslab_validate(spa_t * spa)590 livelist_metaslab_validate(spa_t *spa)
591 {
592 	(void) printf("Verifying deleted livelist entries\n");
593 
594 	sublivelist_verify_t sv;
595 	zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
596 	    sizeof (sublivelist_verify_block_t));
597 	iterate_deleted_livelists(spa, livelist_verify, &sv);
598 
599 	(void) printf("Verifying metaslab entries\n");
600 	vdev_t *rvd = spa->spa_root_vdev;
601 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
602 		vdev_t *vd = rvd->vdev_child[c];
603 
604 		if (!vdev_is_concrete(vd))
605 			continue;
606 
607 		for (uint64_t mid = 0; mid < vd->vdev_ms_count; mid++) {
608 			metaslab_t *m = vd->vdev_ms[mid];
609 
610 			(void) fprintf(stderr,
611 			    "\rverifying concrete vdev %llu, "
612 			    "metaslab %llu of %llu ...",
613 			    (longlong_t)vd->vdev_id,
614 			    (longlong_t)mid,
615 			    (longlong_t)vd->vdev_ms_count);
616 
617 			uint64_t shift, start;
618 			zfs_range_seg_type_t type =
619 			    metaslab_calculate_range_tree_type(vd, m,
620 			    &start, &shift);
621 			metaslab_verify_t mv;
622 			mv.mv_allocated = zfs_range_tree_create(NULL,
623 			    type, NULL, start, shift);
624 			mv.mv_vdid = vd->vdev_id;
625 			mv.mv_msid = m->ms_id;
626 			mv.mv_start = m->ms_start;
627 			mv.mv_end = m->ms_start + m->ms_size;
628 			zfs_btree_create(&mv.mv_livelist_allocs,
629 			    livelist_block_compare, NULL,
630 			    sizeof (sublivelist_verify_block_t));
631 
632 			mv_populate_livelist_allocs(&mv, &sv);
633 
634 			spacemap_check_ms_sm(m->ms_sm, &mv);
635 			spacemap_check_sm_log(spa, &mv);
636 
637 			zfs_range_tree_vacate(mv.mv_allocated, NULL, NULL);
638 			zfs_range_tree_destroy(mv.mv_allocated);
639 			zfs_btree_clear(&mv.mv_livelist_allocs);
640 			zfs_btree_destroy(&mv.mv_livelist_allocs);
641 		}
642 	}
643 	(void) fprintf(stderr, "\n");
644 
645 	/*
646 	 * If there are any segments in the leftover tree after we walked
647 	 * through all the metaslabs in the concrete vdevs then this means
648 	 * that we have segments in the livelists that belong to indirect
649 	 * vdevs and are marked as allocated.
650 	 */
651 	if (zfs_btree_numnodes(&sv.sv_leftover) == 0) {
652 		zfs_btree_destroy(&sv.sv_leftover);
653 		return;
654 	}
655 	(void) printf("ERROR: Found livelist blocks marked as allocated "
656 	    "for indirect vdevs:\n");
657 
658 	zfs_btree_index_t *where = NULL;
659 	sublivelist_verify_block_t *svb;
660 	while ((svb = zfs_btree_destroy_nodes(&sv.sv_leftover, &where)) !=
661 	    NULL) {
662 		int vdev_id = DVA_GET_VDEV(&svb->svb_dva);
663 		ASSERT3U(vdev_id, <, rvd->vdev_children);
664 		vdev_t *vd = rvd->vdev_child[vdev_id];
665 		ASSERT(!vdev_is_concrete(vd));
666 		(void) printf("<%d:%llx:%llx> TXG %llx\n",
667 		    vdev_id, (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
668 		    (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva),
669 		    (u_longlong_t)svb->svb_allocated_txg);
670 	}
671 	(void) printf("\n");
672 	zfs_btree_destroy(&sv.sv_leftover);
673 }
674 
675 /*
676  * These libumem hooks provide a reasonable set of defaults for the allocator's
677  * debugging facilities.
678  */
679 const char *
_umem_debug_init(void)680 _umem_debug_init(void)
681 {
682 	return ("default,verbose"); /* $UMEM_DEBUG setting */
683 }
684 
685 const char *
_umem_logging_init(void)686 _umem_logging_init(void)
687 {
688 	return ("fail,contents"); /* $UMEM_LOGGING setting */
689 }
690 
691 static void
usage(void)692 usage(void)
693 {
694 	(void) fprintf(stderr,
695 	    "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
696 	    "[-I <inflight I/Os>]\n"
697 	    "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
698 	    "\t\t[-K <key>]\n"
699 	    "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
700 	    "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
701 	    "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
702 	    "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
703 	    "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
704 	    "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
705 	    "\t%s [-v] <bookmark>\n"
706 	    "\t%s -C [-A] [-U <cache>] [<poolname>]\n"
707 	    "\t%s -l [-Aqu] <device>\n"
708 	    "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
709 	    "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
710 	    "\t%s -O [-K <key>] <dataset> <path>\n"
711 	    "\t%s -r [-K <key>] <dataset> <path> <destination>\n"
712 	    "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
713 	    "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
714 	    "\t%s -E [-A] word0:word1:...:word15\n"
715 	    "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
716 	    "<poolname>\n\n",
717 	    cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
718 	    cmdname, cmdname, cmdname, cmdname, cmdname);
719 
720 	(void) fprintf(stderr, "    Dataset name must include at least one "
721 	    "separator character '/' or '@'\n");
722 	(void) fprintf(stderr, "    If dataset name is specified, only that "
723 	    "dataset is dumped\n");
724 	(void) fprintf(stderr,  "    If object numbers or object number "
725 	    "ranges are specified, only those\n"
726 	    "    objects or ranges are dumped.\n\n");
727 	(void) fprintf(stderr,
728 	    "    Object ranges take the form <start>:<end>[:<flags>]\n"
729 	    "        start    Starting object number\n"
730 	    "        end      Ending object number, or -1 for no upper bound\n"
731 	    "        flags    Optional flags to select object types:\n"
732 	    "            A     All objects (this is the default)\n"
733 	    "            d     ZFS directories\n"
734 	    "            f     ZFS files \n"
735 	    "            m     SPA space maps\n"
736 	    "            z     ZAPs\n"
737 	    "            -     Negate effect of next flag\n\n");
738 	(void) fprintf(stderr, "    Options to control amount of output:\n");
739 	(void) fprintf(stderr, "        -b --block-stats             "
740 	    "block statistics\n");
741 	(void) fprintf(stderr, "        -B --backup                  "
742 	    "backup stream\n");
743 	(void) fprintf(stderr, "        -c --checksum                "
744 	    "checksum all metadata (twice for all data) blocks\n");
745 	(void) fprintf(stderr, "        -C --config                  "
746 	    "config (or cachefile if alone)\n");
747 	(void) fprintf(stderr, "        -d --datasets                "
748 	    "dataset(s)\n");
749 	(void) fprintf(stderr, "        -D --dedup-stats             "
750 	    "dedup statistics\n");
751 	(void) fprintf(stderr, "        -E --embedded-block-pointer=INTEGER\n"
752 	    "                                     decode and display block "
753 	    "from an embedded block pointer\n");
754 	(void) fprintf(stderr, "        -h --history                 "
755 	    "pool history\n");
756 	(void) fprintf(stderr, "        -i --intent-logs             "
757 	    "intent logs\n");
758 	(void) fprintf(stderr, "        -l --label                   "
759 	    "read label contents\n");
760 	(void) fprintf(stderr, "        -k --checkpointed-state      "
761 	    "examine the checkpointed state of the pool\n");
762 	(void) fprintf(stderr, "        -L --disable-leak-tracking   "
763 	    "disable leak tracking (do not load spacemaps)\n");
764 	(void) fprintf(stderr, "        -m --metaslabs               "
765 	    "metaslabs\n");
766 	(void) fprintf(stderr, "        -M --metaslab-groups         "
767 	    "metaslab groups\n");
768 	(void) fprintf(stderr, "        -O --object-lookups          "
769 	    "perform object lookups by path\n");
770 	(void) fprintf(stderr, "        -r --copy-object             "
771 	    "copy an object by path to file\n");
772 	(void) fprintf(stderr, "        -R --read-block              "
773 	    "read and display block from a device\n");
774 	(void) fprintf(stderr, "        -s --io-stats                "
775 	    "report stats on zdb's I/O\n");
776 	(void) fprintf(stderr, "        -S --simulate-dedup          "
777 	    "simulate dedup to measure effect\n");
778 	(void) fprintf(stderr, "        -v --verbose                 "
779 	    "verbose (applies to all others)\n");
780 	(void) fprintf(stderr, "        -y --livelist                "
781 	    "perform livelist and metaslab validation on any livelists being "
782 	    "deleted\n\n");
783 	(void) fprintf(stderr, "    Below options are intended for use "
784 	    "with other options:\n");
785 	(void) fprintf(stderr, "        -A --ignore-assertions       "
786 	    "ignore assertions (-A), enable panic recovery (-AA) or both "
787 	    "(-AAA)\n");
788 	(void) fprintf(stderr, "        -e --exported                "
789 	    "pool is exported/destroyed/has altroot/not in a cachefile\n");
790 	(void) fprintf(stderr, "        -F --automatic-rewind        "
791 	    "attempt automatic rewind within safe range of transaction "
792 	    "groups\n");
793 	(void) fprintf(stderr, "        -G --dump-debug-msg          "
794 	    "dump zfs_dbgmsg buffer before exiting\n");
795 	(void) fprintf(stderr, "        -I --inflight=INTEGER        "
796 	    "specify the maximum number of checksumming I/Os "
797 	    "[default is 200]\n");
798 	(void) fprintf(stderr, "        -K --key=KEY                 "
799 	    "decryption key for encrypted dataset\n");
800 	(void) fprintf(stderr, "        -o --option=\"OPTION=INTEGER\" "
801 	    "set global variable to an unsigned 32-bit integer\n");
802 	(void) fprintf(stderr, "        -p --path==PATH              "
803 	    "use one or more with -e to specify path to vdev dir\n");
804 	(void) fprintf(stderr, "        -P --parseable               "
805 	    "print numbers in parseable form\n");
806 	(void) fprintf(stderr, "        -q --skip-label              "
807 	    "don't print label contents\n");
808 	(void) fprintf(stderr, "        -t --txg=INTEGER             "
809 	    "highest txg to use when searching for uberblocks\n");
810 	(void) fprintf(stderr, "        -T --brt-stats               "
811 	    "BRT statistics\n");
812 	(void) fprintf(stderr, "        -u --uberblock               "
813 	    "uberblock\n");
814 	(void) fprintf(stderr, "        -U --cachefile=PATH          "
815 	    "use alternate cachefile\n");
816 	(void) fprintf(stderr, "        -V --verbatim                "
817 	    "do verbatim import\n");
818 	(void) fprintf(stderr, "        -x --dump-blocks=PATH        "
819 	    "dump all read blocks into specified directory\n");
820 	(void) fprintf(stderr, "        -X --extreme-rewind          "
821 	    "attempt extreme rewind (does not work with dataset)\n");
822 	(void) fprintf(stderr, "        -Y --all-reconstruction      "
823 	    "attempt all reconstruction combinations for split blocks\n");
824 	(void) fprintf(stderr, "        -Z --zstd-headers            "
825 	    "show ZSTD headers \n");
826 	(void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
827 	    "to make only that option verbose\n");
828 	(void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
829 	zdb_exit(1);
830 }
831 
832 static void
dump_debug_buffer(void)833 dump_debug_buffer(void)
834 {
835 	ssize_t ret __attribute__((unused));
836 
837 	if (!dump_opt['G'])
838 		return;
839 	/*
840 	 * We use write() instead of printf() so that this function
841 	 * is safe to call from a signal handler.
842 	 */
843 	ret = write(STDERR_FILENO, "\n", 1);
844 	zfs_dbgmsg_print(STDERR_FILENO, "zdb");
845 }
846 
sig_handler(int signo)847 static void sig_handler(int signo)
848 {
849 	struct sigaction action;
850 
851 	libspl_backtrace(STDERR_FILENO);
852 	dump_debug_buffer();
853 
854 	/*
855 	 * Restore default action and re-raise signal so SIGSEGV and
856 	 * SIGABRT can trigger a core dump.
857 	 */
858 	action.sa_handler = SIG_DFL;
859 	sigemptyset(&action.sa_mask);
860 	action.sa_flags = 0;
861 	(void) sigaction(signo, &action, NULL);
862 	raise(signo);
863 }
864 
865 /*
866  * Called for usage errors that are discovered after a call to spa_open(),
867  * dmu_bonus_hold(), or pool_match().  abort() is called for other errors.
868  */
869 
870 static void
fatal(const char * fmt,...)871 fatal(const char *fmt, ...)
872 {
873 	va_list ap;
874 
875 	va_start(ap, fmt);
876 	(void) fprintf(stderr, "%s: ", cmdname);
877 	(void) vfprintf(stderr, fmt, ap);
878 	va_end(ap);
879 	(void) fprintf(stderr, "\n");
880 
881 	dump_debug_buffer();
882 
883 	zdb_exit(1);
884 }
885 
886 static void
dump_packed_nvlist(objset_t * os,uint64_t object,void * data,size_t size)887 dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size)
888 {
889 	(void) size;
890 	nvlist_t *nv;
891 	size_t nvsize = *(uint64_t *)data;
892 	char *packed = umem_alloc(nvsize, UMEM_NOFAIL);
893 
894 	VERIFY(0 == dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH));
895 
896 	VERIFY(nvlist_unpack(packed, nvsize, &nv, 0) == 0);
897 
898 	umem_free(packed, nvsize);
899 
900 	dump_nvlist(nv, 8);
901 
902 	nvlist_free(nv);
903 }
904 
905 static void
dump_history_offsets(objset_t * os,uint64_t object,void * data,size_t size)906 dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size)
907 {
908 	(void) os, (void) object, (void) size;
909 	spa_history_phys_t *shp = data;
910 
911 	if (shp == NULL)
912 		return;
913 
914 	(void) printf("\t\tpool_create_len = %llu\n",
915 	    (u_longlong_t)shp->sh_pool_create_len);
916 	(void) printf("\t\tphys_max_off = %llu\n",
917 	    (u_longlong_t)shp->sh_phys_max_off);
918 	(void) printf("\t\tbof = %llu\n",
919 	    (u_longlong_t)shp->sh_bof);
920 	(void) printf("\t\teof = %llu\n",
921 	    (u_longlong_t)shp->sh_eof);
922 	(void) printf("\t\trecords_lost = %llu\n",
923 	    (u_longlong_t)shp->sh_records_lost);
924 }
925 
926 static void
zdb_nicenum(uint64_t num,char * buf,size_t buflen)927 zdb_nicenum(uint64_t num, char *buf, size_t buflen)
928 {
929 	if (dump_opt['P'])
930 		(void) snprintf(buf, buflen, "%llu", (longlong_t)num);
931 	else
932 		nicenum(num, buf, buflen);
933 }
934 
935 static void
zdb_nicebytes(uint64_t bytes,char * buf,size_t buflen)936 zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen)
937 {
938 	if (dump_opt['P'])
939 		(void) snprintf(buf, buflen, "%llu", (longlong_t)bytes);
940 	else
941 		zfs_nicebytes(bytes, buf, buflen);
942 }
943 
944 static const char histo_stars[] = "****************************************";
945 static const uint64_t histo_width = sizeof (histo_stars) - 1;
946 
947 static void
dump_histogram(const uint64_t * histo,int size,int offset)948 dump_histogram(const uint64_t *histo, int size, int offset)
949 {
950 	int i;
951 	int minidx = size - 1;
952 	int maxidx = 0;
953 	uint64_t max = 0;
954 
955 	for (i = 0; i < size; i++) {
956 		if (histo[i] == 0)
957 			continue;
958 		if (histo[i] > max)
959 			max = histo[i];
960 		if (i > maxidx)
961 			maxidx = i;
962 		if (i < minidx)
963 			minidx = i;
964 	}
965 
966 	if (max < histo_width)
967 		max = histo_width;
968 
969 	for (i = minidx; i <= maxidx; i++) {
970 		(void) printf("\t\t\t%3u: %6llu %s\n",
971 		    i + offset, (u_longlong_t)histo[i],
972 		    &histo_stars[(max - histo[i]) * histo_width / max]);
973 	}
974 }
975 
976 static void
dump_zap_stats(objset_t * os,uint64_t object)977 dump_zap_stats(objset_t *os, uint64_t object)
978 {
979 	int error;
980 	zap_stats_t zs;
981 
982 	error = zap_get_stats(os, object, &zs);
983 	if (error)
984 		return;
985 
986 	if (zs.zs_ptrtbl_len == 0) {
987 		ASSERT(zs.zs_num_blocks == 1);
988 		(void) printf("\tmicrozap: %llu bytes, %llu entries\n",
989 		    (u_longlong_t)zs.zs_blocksize,
990 		    (u_longlong_t)zs.zs_num_entries);
991 		return;
992 	}
993 
994 	(void) printf("\tFat ZAP stats:\n");
995 
996 	(void) printf("\t\tPointer table:\n");
997 	(void) printf("\t\t\t%llu elements\n",
998 	    (u_longlong_t)zs.zs_ptrtbl_len);
999 	(void) printf("\t\t\tzt_blk: %llu\n",
1000 	    (u_longlong_t)zs.zs_ptrtbl_zt_blk);
1001 	(void) printf("\t\t\tzt_numblks: %llu\n",
1002 	    (u_longlong_t)zs.zs_ptrtbl_zt_numblks);
1003 	(void) printf("\t\t\tzt_shift: %llu\n",
1004 	    (u_longlong_t)zs.zs_ptrtbl_zt_shift);
1005 	(void) printf("\t\t\tzt_blks_copied: %llu\n",
1006 	    (u_longlong_t)zs.zs_ptrtbl_blks_copied);
1007 	(void) printf("\t\t\tzt_nextblk: %llu\n",
1008 	    (u_longlong_t)zs.zs_ptrtbl_nextblk);
1009 
1010 	(void) printf("\t\tZAP entries: %llu\n",
1011 	    (u_longlong_t)zs.zs_num_entries);
1012 	(void) printf("\t\tLeaf blocks: %llu\n",
1013 	    (u_longlong_t)zs.zs_num_leafs);
1014 	(void) printf("\t\tTotal blocks: %llu\n",
1015 	    (u_longlong_t)zs.zs_num_blocks);
1016 	(void) printf("\t\tzap_block_type: 0x%llx\n",
1017 	    (u_longlong_t)zs.zs_block_type);
1018 	(void) printf("\t\tzap_magic: 0x%llx\n",
1019 	    (u_longlong_t)zs.zs_magic);
1020 	(void) printf("\t\tzap_salt: 0x%llx\n",
1021 	    (u_longlong_t)zs.zs_salt);
1022 
1023 	(void) printf("\t\tLeafs with 2^n pointers:\n");
1024 	dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);
1025 
1026 	(void) printf("\t\tBlocks with n*5 entries:\n");
1027 	dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);
1028 
1029 	(void) printf("\t\tBlocks n/10 full:\n");
1030 	dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);
1031 
1032 	(void) printf("\t\tEntries with n chunks:\n");
1033 	dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);
1034 
1035 	(void) printf("\t\tBuckets with n entries:\n");
1036 	dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
1037 }
1038 
1039 static void
dump_none(objset_t * os,uint64_t object,void * data,size_t size)1040 dump_none(objset_t *os, uint64_t object, void *data, size_t size)
1041 {
1042 	(void) os, (void) object, (void) data, (void) size;
1043 }
1044 
1045 static void
dump_unknown(objset_t * os,uint64_t object,void * data,size_t size)1046 dump_unknown(objset_t *os, uint64_t object, void *data, size_t size)
1047 {
1048 	(void) os, (void) object, (void) data, (void) size;
1049 	(void) printf("\tUNKNOWN OBJECT TYPE\n");
1050 }
1051 
1052 static void
dump_uint8(objset_t * os,uint64_t object,void * data,size_t size)1053 dump_uint8(objset_t *os, uint64_t object, void *data, size_t size)
1054 {
1055 	(void) os, (void) object, (void) data, (void) size;
1056 }
1057 
1058 static void
dump_uint64(objset_t * os,uint64_t object,void * data,size_t size)1059 dump_uint64(objset_t *os, uint64_t object, void *data, size_t size)
1060 {
1061 	uint64_t *arr;
1062 	uint64_t oursize;
1063 	if (dump_opt['d'] < 6)
1064 		return;
1065 
1066 	if (data == NULL) {
1067 		dmu_object_info_t doi;
1068 
1069 		VERIFY0(dmu_object_info(os, object, &doi));
1070 		size = doi.doi_max_offset;
1071 		/*
1072 		 * We cap the size at 1 mebibyte here to prevent
1073 		 * allocation failures and nigh-infinite printing if the
1074 		 * object is extremely large.
1075 		 */
1076 		oursize = MIN(size, 1 << 20);
1077 		arr = kmem_alloc(oursize, KM_SLEEP);
1078 
1079 		int err = dmu_read(os, object, 0, oursize, arr, 0);
1080 		if (err != 0) {
1081 			(void) printf("got error %u from dmu_read\n", err);
1082 			kmem_free(arr, oursize);
1083 			return;
1084 		}
1085 	} else {
1086 		/*
1087 		 * Even though the allocation is already done in this code path,
1088 		 * we still cap the size to prevent excessive printing.
1089 		 */
1090 		oursize = MIN(size, 1 << 20);
1091 		arr = data;
1092 	}
1093 
1094 	if (size == 0) {
1095 		if (data == NULL)
1096 			kmem_free(arr, oursize);
1097 		(void) printf("\t\t[]\n");
1098 		return;
1099 	}
1100 
1101 	(void) printf("\t\t[%0llx", (u_longlong_t)arr[0]);
1102 	for (size_t i = 1; i * sizeof (uint64_t) < oursize; i++) {
1103 		if (i % 4 != 0)
1104 			(void) printf(", %0llx", (u_longlong_t)arr[i]);
1105 		else
1106 			(void) printf(",\n\t\t%0llx", (u_longlong_t)arr[i]);
1107 	}
1108 	if (oursize != size)
1109 		(void) printf(", ... ");
1110 	(void) printf("]\n");
1111 
1112 	if (data == NULL)
1113 		kmem_free(arr, oursize);
1114 }
1115 
1116 static void
dump_zap(objset_t * os,uint64_t object,void * data,size_t size)1117 dump_zap(objset_t *os, uint64_t object, void *data, size_t size)
1118 {
1119 	(void) data, (void) size;
1120 	zap_cursor_t zc;
1121 	zap_attribute_t *attrp = zap_attribute_long_alloc();
1122 	void *prop;
1123 	unsigned i;
1124 
1125 	dump_zap_stats(os, object);
1126 	(void) printf("\n");
1127 
1128 	for (zap_cursor_init(&zc, os, object);
1129 	    zap_cursor_retrieve(&zc, attrp) == 0;
1130 	    zap_cursor_advance(&zc)) {
1131 		boolean_t key64 =
1132 		    !!(zap_getflags(zc.zc_zap) & ZAP_FLAG_UINT64_KEY);
1133 
1134 		if (key64)
1135 			(void) printf("\t\t0x%010" PRIu64 "x = ",
1136 			    *(uint64_t *)attrp->za_name);
1137 		else
1138 			(void) printf("\t\t%s = ", attrp->za_name);
1139 
1140 		if (attrp->za_num_integers == 0) {
1141 			(void) printf("\n");
1142 			continue;
1143 		}
1144 		prop = umem_zalloc(attrp->za_num_integers *
1145 		    attrp->za_integer_length, UMEM_NOFAIL);
1146 
1147 		if (key64)
1148 			(void) zap_lookup_uint64(os, object,
1149 			    (const uint64_t *)attrp->za_name, 1,
1150 			    attrp->za_integer_length, attrp->za_num_integers,
1151 			    prop);
1152 		else
1153 			(void) zap_lookup(os, object, attrp->za_name,
1154 			    attrp->za_integer_length, attrp->za_num_integers,
1155 			    prop);
1156 
1157 		if (attrp->za_integer_length == 1 && !key64) {
1158 			if (strcmp(attrp->za_name,
1159 			    DSL_CRYPTO_KEY_MASTER_KEY) == 0 ||
1160 			    strcmp(attrp->za_name,
1161 			    DSL_CRYPTO_KEY_HMAC_KEY) == 0 ||
1162 			    strcmp(attrp->za_name, DSL_CRYPTO_KEY_IV) == 0 ||
1163 			    strcmp(attrp->za_name, DSL_CRYPTO_KEY_MAC) == 0 ||
1164 			    strcmp(attrp->za_name,
1165 			    DMU_POOL_CHECKSUM_SALT) == 0) {
1166 				uint8_t *u8 = prop;
1167 
1168 				for (i = 0; i < attrp->za_num_integers; i++) {
1169 					(void) printf("%02x", u8[i]);
1170 				}
1171 			} else {
1172 				(void) printf("%s", (char *)prop);
1173 			}
1174 		} else {
1175 			for (i = 0; i < attrp->za_num_integers; i++) {
1176 				switch (attrp->za_integer_length) {
1177 				case 1:
1178 					(void) printf("%u ",
1179 					    ((uint8_t *)prop)[i]);
1180 					break;
1181 				case 2:
1182 					(void) printf("%u ",
1183 					    ((uint16_t *)prop)[i]);
1184 					break;
1185 				case 4:
1186 					(void) printf("%u ",
1187 					    ((uint32_t *)prop)[i]);
1188 					break;
1189 				case 8:
1190 					(void) printf("%lld ",
1191 					    (u_longlong_t)((int64_t *)prop)[i]);
1192 					break;
1193 				}
1194 			}
1195 		}
1196 		(void) printf("\n");
1197 		umem_free(prop,
1198 		    attrp->za_num_integers * attrp->za_integer_length);
1199 	}
1200 	zap_cursor_fini(&zc);
1201 	zap_attribute_free(attrp);
1202 }
1203 
1204 static void
dump_bpobj(objset_t * os,uint64_t object,void * data,size_t size)1205 dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size)
1206 {
1207 	bpobj_phys_t *bpop = data;
1208 	uint64_t i;
1209 	char bytes[32], comp[32], uncomp[32];
1210 
1211 	/* make sure the output won't get truncated */
1212 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
1213 	_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
1214 	_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
1215 
1216 	if (bpop == NULL)
1217 		return;
1218 
1219 	zdb_nicenum(bpop->bpo_bytes, bytes, sizeof (bytes));
1220 	zdb_nicenum(bpop->bpo_comp, comp, sizeof (comp));
1221 	zdb_nicenum(bpop->bpo_uncomp, uncomp, sizeof (uncomp));
1222 
1223 	(void) printf("\t\tnum_blkptrs = %llu\n",
1224 	    (u_longlong_t)bpop->bpo_num_blkptrs);
1225 	(void) printf("\t\tbytes = %s\n", bytes);
1226 	if (size >= BPOBJ_SIZE_V1) {
1227 		(void) printf("\t\tcomp = %s\n", comp);
1228 		(void) printf("\t\tuncomp = %s\n", uncomp);
1229 	}
1230 	if (size >= BPOBJ_SIZE_V2) {
1231 		(void) printf("\t\tsubobjs = %llu\n",
1232 		    (u_longlong_t)bpop->bpo_subobjs);
1233 		(void) printf("\t\tnum_subobjs = %llu\n",
1234 		    (u_longlong_t)bpop->bpo_num_subobjs);
1235 	}
1236 	if (size >= sizeof (*bpop)) {
1237 		(void) printf("\t\tnum_freed = %llu\n",
1238 		    (u_longlong_t)bpop->bpo_num_freed);
1239 	}
1240 
1241 	if (dump_opt['d'] < 5)
1242 		return;
1243 
1244 	for (i = 0; i < bpop->bpo_num_blkptrs; i++) {
1245 		char blkbuf[BP_SPRINTF_LEN];
1246 		blkptr_t bp;
1247 
1248 		int err = dmu_read(os, object,
1249 		    i * sizeof (bp), sizeof (bp), &bp, 0);
1250 		if (err != 0) {
1251 			(void) printf("got error %u from dmu_read\n", err);
1252 			break;
1253 		}
1254 		snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp,
1255 		    BP_GET_FREE(&bp));
1256 		(void) printf("\t%s\n", blkbuf);
1257 	}
1258 }
1259 
1260 static void
dump_bpobj_subobjs(objset_t * os,uint64_t object,void * data,size_t size)1261 dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size)
1262 {
1263 	(void) data, (void) size;
1264 	dmu_object_info_t doi;
1265 	int64_t i;
1266 
1267 	VERIFY0(dmu_object_info(os, object, &doi));
1268 	uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP);
1269 
1270 	int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0);
1271 	if (err != 0) {
1272 		(void) printf("got error %u from dmu_read\n", err);
1273 		kmem_free(subobjs, doi.doi_max_offset);
1274 		return;
1275 	}
1276 
1277 	int64_t last_nonzero = -1;
1278 	for (i = 0; i < doi.doi_max_offset / 8; i++) {
1279 		if (subobjs[i] != 0)
1280 			last_nonzero = i;
1281 	}
1282 
1283 	for (i = 0; i <= last_nonzero; i++) {
1284 		(void) printf("\t%llu\n", (u_longlong_t)subobjs[i]);
1285 	}
1286 	kmem_free(subobjs, doi.doi_max_offset);
1287 }
1288 
1289 static void
dump_ddt_zap(objset_t * os,uint64_t object,void * data,size_t size)1290 dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size)
1291 {
1292 	(void) data, (void) size;
1293 	dump_zap_stats(os, object);
1294 	/* contents are printed elsewhere, properly decoded */
1295 }
1296 
1297 static void
dump_sa_attrs(objset_t * os,uint64_t object,void * data,size_t size)1298 dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size)
1299 {
1300 	(void) data, (void) size;
1301 	zap_cursor_t zc;
1302 	zap_attribute_t *attrp = zap_attribute_alloc();
1303 
1304 	dump_zap_stats(os, object);
1305 	(void) printf("\n");
1306 
1307 	for (zap_cursor_init(&zc, os, object);
1308 	    zap_cursor_retrieve(&zc, attrp) == 0;
1309 	    zap_cursor_advance(&zc)) {
1310 		(void) printf("\t\t%s = ", attrp->za_name);
1311 		if (attrp->za_num_integers == 0) {
1312 			(void) printf("\n");
1313 			continue;
1314 		}
1315 		(void) printf(" %llx : [%d:%d:%d]\n",
1316 		    (u_longlong_t)attrp->za_first_integer,
1317 		    (int)ATTR_LENGTH(attrp->za_first_integer),
1318 		    (int)ATTR_BSWAP(attrp->za_first_integer),
1319 		    (int)ATTR_NUM(attrp->za_first_integer));
1320 	}
1321 	zap_cursor_fini(&zc);
1322 	zap_attribute_free(attrp);
1323 }
1324 
1325 static void
dump_sa_layouts(objset_t * os,uint64_t object,void * data,size_t size)1326 dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size)
1327 {
1328 	(void) data, (void) size;
1329 	zap_cursor_t zc;
1330 	zap_attribute_t *attrp = zap_attribute_alloc();
1331 	uint16_t *layout_attrs;
1332 	unsigned i;
1333 
1334 	dump_zap_stats(os, object);
1335 	(void) printf("\n");
1336 
1337 	for (zap_cursor_init(&zc, os, object);
1338 	    zap_cursor_retrieve(&zc, attrp) == 0;
1339 	    zap_cursor_advance(&zc)) {
1340 		(void) printf("\t\t%s = [", attrp->za_name);
1341 		if (attrp->za_num_integers == 0) {
1342 			(void) printf("\n");
1343 			continue;
1344 		}
1345 
1346 		VERIFY(attrp->za_integer_length == 2);
1347 		layout_attrs = umem_zalloc(attrp->za_num_integers *
1348 		    attrp->za_integer_length, UMEM_NOFAIL);
1349 
1350 		VERIFY(zap_lookup(os, object, attrp->za_name,
1351 		    attrp->za_integer_length,
1352 		    attrp->za_num_integers, layout_attrs) == 0);
1353 
1354 		for (i = 0; i != attrp->za_num_integers; i++)
1355 			(void) printf(" %d ", (int)layout_attrs[i]);
1356 		(void) printf("]\n");
1357 		umem_free(layout_attrs,
1358 		    attrp->za_num_integers * attrp->za_integer_length);
1359 	}
1360 	zap_cursor_fini(&zc);
1361 	zap_attribute_free(attrp);
1362 }
1363 
1364 static void
dump_zpldir(objset_t * os,uint64_t object,void * data,size_t size)1365 dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
1366 {
1367 	(void) data, (void) size;
1368 	zap_cursor_t zc;
1369 	zap_attribute_t *attrp = zap_attribute_long_alloc();
1370 	const char *typenames[] = {
1371 		/* 0 */ "not specified",
1372 		/* 1 */ "FIFO",
1373 		/* 2 */ "Character Device",
1374 		/* 3 */ "3 (invalid)",
1375 		/* 4 */ "Directory",
1376 		/* 5 */ "5 (invalid)",
1377 		/* 6 */ "Block Device",
1378 		/* 7 */ "7 (invalid)",
1379 		/* 8 */ "Regular File",
1380 		/* 9 */ "9 (invalid)",
1381 		/* 10 */ "Symbolic Link",
1382 		/* 11 */ "11 (invalid)",
1383 		/* 12 */ "Socket",
1384 		/* 13 */ "Door",
1385 		/* 14 */ "Event Port",
1386 		/* 15 */ "15 (invalid)",
1387 	};
1388 
1389 	dump_zap_stats(os, object);
1390 	(void) printf("\n");
1391 
1392 	for (zap_cursor_init(&zc, os, object);
1393 	    zap_cursor_retrieve(&zc, attrp) == 0;
1394 	    zap_cursor_advance(&zc)) {
1395 		(void) printf("\t\t%s = %lld (type: %s)\n",
1396 		    attrp->za_name, ZFS_DIRENT_OBJ(attrp->za_first_integer),
1397 		    typenames[ZFS_DIRENT_TYPE(attrp->za_first_integer)]);
1398 	}
1399 	zap_cursor_fini(&zc);
1400 	zap_attribute_free(attrp);
1401 }
1402 
1403 static int
get_dtl_refcount(vdev_t * vd)1404 get_dtl_refcount(vdev_t *vd)
1405 {
1406 	int refcount = 0;
1407 
1408 	if (vd->vdev_ops->vdev_op_leaf) {
1409 		space_map_t *sm = vd->vdev_dtl_sm;
1410 
1411 		if (sm != NULL &&
1412 		    sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1413 			return (1);
1414 		return (0);
1415 	}
1416 
1417 	for (unsigned c = 0; c < vd->vdev_children; c++)
1418 		refcount += get_dtl_refcount(vd->vdev_child[c]);
1419 	return (refcount);
1420 }
1421 
1422 static int
get_metaslab_refcount(vdev_t * vd)1423 get_metaslab_refcount(vdev_t *vd)
1424 {
1425 	int refcount = 0;
1426 
1427 	if (vd->vdev_top == vd) {
1428 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
1429 			space_map_t *sm = vd->vdev_ms[m]->ms_sm;
1430 
1431 			if (sm != NULL &&
1432 			    sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1433 				refcount++;
1434 		}
1435 	}
1436 	for (unsigned c = 0; c < vd->vdev_children; c++)
1437 		refcount += get_metaslab_refcount(vd->vdev_child[c]);
1438 
1439 	return (refcount);
1440 }
1441 
1442 static int
get_obsolete_refcount(vdev_t * vd)1443 get_obsolete_refcount(vdev_t *vd)
1444 {
1445 	uint64_t obsolete_sm_object;
1446 	int refcount = 0;
1447 
1448 	VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1449 	if (vd->vdev_top == vd && obsolete_sm_object != 0) {
1450 		dmu_object_info_t doi;
1451 		VERIFY0(dmu_object_info(vd->vdev_spa->spa_meta_objset,
1452 		    obsolete_sm_object, &doi));
1453 		if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1454 			refcount++;
1455 		}
1456 	} else {
1457 		ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
1458 		ASSERT3U(obsolete_sm_object, ==, 0);
1459 	}
1460 	for (unsigned c = 0; c < vd->vdev_children; c++) {
1461 		refcount += get_obsolete_refcount(vd->vdev_child[c]);
1462 	}
1463 
1464 	return (refcount);
1465 }
1466 
1467 static int
get_prev_obsolete_spacemap_refcount(spa_t * spa)1468 get_prev_obsolete_spacemap_refcount(spa_t *spa)
1469 {
1470 	uint64_t prev_obj =
1471 	    spa->spa_condensing_indirect_phys.scip_prev_obsolete_sm_object;
1472 	if (prev_obj != 0) {
1473 		dmu_object_info_t doi;
1474 		VERIFY0(dmu_object_info(spa->spa_meta_objset, prev_obj, &doi));
1475 		if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1476 			return (1);
1477 		}
1478 	}
1479 	return (0);
1480 }
1481 
1482 static int
get_checkpoint_refcount(vdev_t * vd)1483 get_checkpoint_refcount(vdev_t *vd)
1484 {
1485 	int refcount = 0;
1486 
1487 	if (vd->vdev_top == vd && vd->vdev_top_zap != 0 &&
1488 	    zap_contains(spa_meta_objset(vd->vdev_spa),
1489 	    vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) == 0)
1490 		refcount++;
1491 
1492 	for (uint64_t c = 0; c < vd->vdev_children; c++)
1493 		refcount += get_checkpoint_refcount(vd->vdev_child[c]);
1494 
1495 	return (refcount);
1496 }
1497 
1498 static int
get_log_spacemap_refcount(spa_t * spa)1499 get_log_spacemap_refcount(spa_t *spa)
1500 {
1501 	return (avl_numnodes(&spa->spa_sm_logs_by_txg));
1502 }
1503 
1504 static int
verify_spacemap_refcounts(spa_t * spa)1505 verify_spacemap_refcounts(spa_t *spa)
1506 {
1507 	uint64_t expected_refcount = 0;
1508 	uint64_t actual_refcount;
1509 
1510 	(void) feature_get_refcount(spa,
1511 	    &spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM],
1512 	    &expected_refcount);
1513 	actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
1514 	actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);
1515 	actual_refcount += get_obsolete_refcount(spa->spa_root_vdev);
1516 	actual_refcount += get_prev_obsolete_spacemap_refcount(spa);
1517 	actual_refcount += get_checkpoint_refcount(spa->spa_root_vdev);
1518 	actual_refcount += get_log_spacemap_refcount(spa);
1519 
1520 	if (expected_refcount != actual_refcount) {
1521 		(void) printf("space map refcount mismatch: expected %lld != "
1522 		    "actual %lld\n",
1523 		    (longlong_t)expected_refcount,
1524 		    (longlong_t)actual_refcount);
1525 		return (2);
1526 	}
1527 	return (0);
1528 }
1529 
1530 static void
dump_spacemap(objset_t * os,space_map_t * sm)1531 dump_spacemap(objset_t *os, space_map_t *sm)
1532 {
1533 	const char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1534 	    "INVALID", "INVALID", "INVALID", "INVALID" };
1535 
1536 	if (sm == NULL)
1537 		return;
1538 
1539 	(void) printf("space map object %llu:\n",
1540 	    (longlong_t)sm->sm_object);
1541 	(void) printf("  smp_length = 0x%llx\n",
1542 	    (longlong_t)sm->sm_phys->smp_length);
1543 	(void) printf("  smp_alloc = 0x%llx\n",
1544 	    (longlong_t)sm->sm_phys->smp_alloc);
1545 
1546 	if (dump_opt['d'] < 6 && dump_opt['m'] < 4)
1547 		return;
1548 
1549 	/*
1550 	 * Print out the freelist entries in both encoded and decoded form.
1551 	 */
1552 	uint8_t mapshift = sm->sm_shift;
1553 	int64_t alloc = 0;
1554 	uint64_t word, entry_id = 0;
1555 	for (uint64_t offset = 0; offset < space_map_length(sm);
1556 	    offset += sizeof (word)) {
1557 
1558 		VERIFY0(dmu_read(os, space_map_object(sm), offset,
1559 		    sizeof (word), &word, DMU_READ_PREFETCH));
1560 
1561 		if (sm_entry_is_debug(word)) {
1562 			uint64_t de_txg = SM_DEBUG_TXG_DECODE(word);
1563 			uint64_t de_sync_pass = SM_DEBUG_SYNCPASS_DECODE(word);
1564 			if (de_txg == 0) {
1565 				(void) printf(
1566 				    "\t    [%6llu] PADDING\n",
1567 				    (u_longlong_t)entry_id);
1568 			} else {
1569 				(void) printf(
1570 				    "\t    [%6llu] %s: txg %llu pass %llu\n",
1571 				    (u_longlong_t)entry_id,
1572 				    ddata[SM_DEBUG_ACTION_DECODE(word)],
1573 				    (u_longlong_t)de_txg,
1574 				    (u_longlong_t)de_sync_pass);
1575 			}
1576 			entry_id++;
1577 			continue;
1578 		}
1579 
1580 		uint8_t words;
1581 		char entry_type;
1582 		uint64_t entry_off, entry_run, entry_vdev = SM_NO_VDEVID;
1583 
1584 		if (sm_entry_is_single_word(word)) {
1585 			entry_type = (SM_TYPE_DECODE(word) == SM_ALLOC) ?
1586 			    'A' : 'F';
1587 			entry_off = (SM_OFFSET_DECODE(word) << mapshift) +
1588 			    sm->sm_start;
1589 			entry_run = SM_RUN_DECODE(word) << mapshift;
1590 			words = 1;
1591 		} else {
1592 			/* it is a two-word entry so we read another word */
1593 			ASSERT(sm_entry_is_double_word(word));
1594 
1595 			uint64_t extra_word;
1596 			offset += sizeof (extra_word);
1597 			VERIFY0(dmu_read(os, space_map_object(sm), offset,
1598 			    sizeof (extra_word), &extra_word,
1599 			    DMU_READ_PREFETCH));
1600 
1601 			ASSERT3U(offset, <=, space_map_length(sm));
1602 
1603 			entry_run = SM2_RUN_DECODE(word) << mapshift;
1604 			entry_vdev = SM2_VDEV_DECODE(word);
1605 			entry_type = (SM2_TYPE_DECODE(extra_word) == SM_ALLOC) ?
1606 			    'A' : 'F';
1607 			entry_off = (SM2_OFFSET_DECODE(extra_word) <<
1608 			    mapshift) + sm->sm_start;
1609 			words = 2;
1610 		}
1611 
1612 		(void) printf("\t    [%6llu]    %c  range:"
1613 		    " %010llx-%010llx  size: %06llx vdev: %06llu words: %u\n",
1614 		    (u_longlong_t)entry_id,
1615 		    entry_type, (u_longlong_t)entry_off,
1616 		    (u_longlong_t)(entry_off + entry_run),
1617 		    (u_longlong_t)entry_run,
1618 		    (u_longlong_t)entry_vdev, words);
1619 
1620 		if (entry_type == 'A')
1621 			alloc += entry_run;
1622 		else
1623 			alloc -= entry_run;
1624 		entry_id++;
1625 	}
1626 	if (alloc != space_map_allocated(sm)) {
1627 		(void) printf("space_map_object alloc (%lld) INCONSISTENT "
1628 		    "with space map summary (%lld)\n",
1629 		    (longlong_t)space_map_allocated(sm), (longlong_t)alloc);
1630 	}
1631 }
1632 
1633 static void
dump_metaslab_stats(metaslab_t * msp)1634 dump_metaslab_stats(metaslab_t *msp)
1635 {
1636 	char maxbuf[32];
1637 	zfs_range_tree_t *rt = msp->ms_allocatable;
1638 	zfs_btree_t *t = &msp->ms_allocatable_by_size;
1639 	int free_pct = zfs_range_tree_space(rt) * 100 / msp->ms_size;
1640 
1641 	/* max sure nicenum has enough space */
1642 	_Static_assert(sizeof (maxbuf) >= NN_NUMBUF_SZ, "maxbuf truncated");
1643 
1644 	zdb_nicenum(metaslab_largest_allocatable(msp), maxbuf, sizeof (maxbuf));
1645 
1646 	(void) printf("\t %25s %10lu   %7s  %6s   %4s %4d%%\n",
1647 	    "segments", zfs_btree_numnodes(t), "maxsize", maxbuf,
1648 	    "freepct", free_pct);
1649 	(void) printf("\tIn-memory histogram:\n");
1650 	dump_histogram(rt->rt_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1651 }
1652 
1653 static void
dump_metaslab(metaslab_t * msp)1654 dump_metaslab(metaslab_t *msp)
1655 {
1656 	vdev_t *vd = msp->ms_group->mg_vd;
1657 	spa_t *spa = vd->vdev_spa;
1658 	space_map_t *sm = msp->ms_sm;
1659 	char freebuf[32];
1660 
1661 	zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf,
1662 	    sizeof (freebuf));
1663 
1664 	(void) printf(
1665 	    "\tmetaslab %6llu   offset %12llx   spacemap %6llu   free    %5s\n",
1666 	    (u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
1667 	    (u_longlong_t)space_map_object(sm), freebuf);
1668 
1669 	if (dump_opt['m'] > 2 && !dump_opt['L']) {
1670 		mutex_enter(&msp->ms_lock);
1671 		VERIFY0(metaslab_load(msp));
1672 		zfs_range_tree_stat_verify(msp->ms_allocatable);
1673 		dump_metaslab_stats(msp);
1674 		metaslab_unload(msp);
1675 		mutex_exit(&msp->ms_lock);
1676 	}
1677 
1678 	if (dump_opt['m'] > 1 && sm != NULL &&
1679 	    spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) {
1680 		/*
1681 		 * The space map histogram represents free space in chunks
1682 		 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1683 		 */
1684 		(void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1685 		    (u_longlong_t)msp->ms_fragmentation);
1686 		dump_histogram(sm->sm_phys->smp_histogram,
1687 		    SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift);
1688 	}
1689 
1690 	if (vd->vdev_ops == &vdev_draid_ops)
1691 		ASSERT3U(msp->ms_size, <=, 1ULL << vd->vdev_ms_shift);
1692 	else
1693 		ASSERT3U(msp->ms_size, ==, 1ULL << vd->vdev_ms_shift);
1694 
1695 	dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
1696 
1697 	if (spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
1698 		(void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1699 		    (u_longlong_t)metaslab_unflushed_txg(msp));
1700 	}
1701 }
1702 
1703 static void
print_vdev_metaslab_header(vdev_t * vd)1704 print_vdev_metaslab_header(vdev_t *vd)
1705 {
1706 	vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
1707 	const char *bias_str = "";
1708 	if (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) {
1709 		bias_str = VDEV_ALLOC_BIAS_LOG;
1710 	} else if (alloc_bias == VDEV_BIAS_SPECIAL) {
1711 		bias_str = VDEV_ALLOC_BIAS_SPECIAL;
1712 	} else if (alloc_bias == VDEV_BIAS_DEDUP) {
1713 		bias_str = VDEV_ALLOC_BIAS_DEDUP;
1714 	}
1715 
1716 	uint64_t ms_flush_data_obj = 0;
1717 	if (vd->vdev_top_zap != 0) {
1718 		int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
1719 		    vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
1720 		    sizeof (uint64_t), 1, &ms_flush_data_obj);
1721 		if (error != ENOENT) {
1722 			ASSERT0(error);
1723 		}
1724 	}
1725 
1726 	(void) printf("\tvdev %10llu   %s",
1727 	    (u_longlong_t)vd->vdev_id, bias_str);
1728 
1729 	if (ms_flush_data_obj != 0) {
1730 		(void) printf("   ms_unflushed_phys object %llu",
1731 		    (u_longlong_t)ms_flush_data_obj);
1732 	}
1733 
1734 	(void) printf("\n\t%-10s%5llu   %-19s   %-15s   %-12s\n",
1735 	    "metaslabs", (u_longlong_t)vd->vdev_ms_count,
1736 	    "offset", "spacemap", "free");
1737 	(void) printf("\t%15s   %19s   %15s   %12s\n",
1738 	    "---------------", "-------------------",
1739 	    "---------------", "------------");
1740 }
1741 
1742 static void
dump_metaslab_groups(spa_t * spa,boolean_t show_special)1743 dump_metaslab_groups(spa_t *spa, boolean_t show_special)
1744 {
1745 	vdev_t *rvd = spa->spa_root_vdev;
1746 	metaslab_class_t *mc = spa_normal_class(spa);
1747 	metaslab_class_t *smc = spa_special_class(spa);
1748 	uint64_t fragmentation;
1749 
1750 	metaslab_class_histogram_verify(mc);
1751 
1752 	for (unsigned c = 0; c < rvd->vdev_children; c++) {
1753 		vdev_t *tvd = rvd->vdev_child[c];
1754 		metaslab_group_t *mg = tvd->vdev_mg;
1755 
1756 		if (mg == NULL || (mg->mg_class != mc &&
1757 		    (!show_special || mg->mg_class != smc)))
1758 			continue;
1759 
1760 		metaslab_group_histogram_verify(mg);
1761 		mg->mg_fragmentation = metaslab_group_fragmentation(mg);
1762 
1763 		(void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1764 		    "fragmentation",
1765 		    (u_longlong_t)tvd->vdev_id,
1766 		    (u_longlong_t)tvd->vdev_ms_count);
1767 		if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
1768 			(void) printf("%3s\n", "-");
1769 		} else {
1770 			(void) printf("%3llu%%\n",
1771 			    (u_longlong_t)mg->mg_fragmentation);
1772 		}
1773 		dump_histogram(mg->mg_histogram,
1774 		    ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1775 	}
1776 
1777 	(void) printf("\tpool %s\tfragmentation", spa_name(spa));
1778 	fragmentation = metaslab_class_fragmentation(mc);
1779 	if (fragmentation == ZFS_FRAG_INVALID)
1780 		(void) printf("\t%3s\n", "-");
1781 	else
1782 		(void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation);
1783 	dump_histogram(mc->mc_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0);
1784 }
1785 
1786 static void
print_vdev_indirect(vdev_t * vd)1787 print_vdev_indirect(vdev_t *vd)
1788 {
1789 	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
1790 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
1791 	vdev_indirect_births_t *vib = vd->vdev_indirect_births;
1792 
1793 	if (vim == NULL) {
1794 		ASSERT3P(vib, ==, NULL);
1795 		return;
1796 	}
1797 
1798 	ASSERT3U(vdev_indirect_mapping_object(vim), ==,
1799 	    vic->vic_mapping_object);
1800 	ASSERT3U(vdev_indirect_births_object(vib), ==,
1801 	    vic->vic_births_object);
1802 
1803 	(void) printf("indirect births obj %llu:\n",
1804 	    (longlong_t)vic->vic_births_object);
1805 	(void) printf("    vib_count = %llu\n",
1806 	    (longlong_t)vdev_indirect_births_count(vib));
1807 	for (uint64_t i = 0; i < vdev_indirect_births_count(vib); i++) {
1808 		vdev_indirect_birth_entry_phys_t *cur_vibe =
1809 		    &vib->vib_entries[i];
1810 		(void) printf("\toffset %llx -> txg %llu\n",
1811 		    (longlong_t)cur_vibe->vibe_offset,
1812 		    (longlong_t)cur_vibe->vibe_phys_birth_txg);
1813 	}
1814 	(void) printf("\n");
1815 
1816 	(void) printf("indirect mapping obj %llu:\n",
1817 	    (longlong_t)vic->vic_mapping_object);
1818 	(void) printf("    vim_max_offset = 0x%llx\n",
1819 	    (longlong_t)vdev_indirect_mapping_max_offset(vim));
1820 	(void) printf("    vim_bytes_mapped = 0x%llx\n",
1821 	    (longlong_t)vdev_indirect_mapping_bytes_mapped(vim));
1822 	(void) printf("    vim_count = %llu\n",
1823 	    (longlong_t)vdev_indirect_mapping_num_entries(vim));
1824 
1825 	if (dump_opt['d'] <= 5 && dump_opt['m'] <= 3)
1826 		return;
1827 
1828 	uint32_t *counts = vdev_indirect_mapping_load_obsolete_counts(vim);
1829 
1830 	for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
1831 		vdev_indirect_mapping_entry_phys_t *vimep =
1832 		    &vim->vim_entries[i];
1833 		(void) printf("\t<%llx:%llx:%llx> -> "
1834 		    "<%llx:%llx:%llx> (%x obsolete)\n",
1835 		    (longlong_t)vd->vdev_id,
1836 		    (longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
1837 		    (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1838 		    (longlong_t)DVA_GET_VDEV(&vimep->vimep_dst),
1839 		    (longlong_t)DVA_GET_OFFSET(&vimep->vimep_dst),
1840 		    (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1841 		    counts[i]);
1842 	}
1843 	(void) printf("\n");
1844 
1845 	uint64_t obsolete_sm_object;
1846 	VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1847 	if (obsolete_sm_object != 0) {
1848 		objset_t *mos = vd->vdev_spa->spa_meta_objset;
1849 		(void) printf("obsolete space map object %llu:\n",
1850 		    (u_longlong_t)obsolete_sm_object);
1851 		ASSERT(vd->vdev_obsolete_sm != NULL);
1852 		ASSERT3U(space_map_object(vd->vdev_obsolete_sm), ==,
1853 		    obsolete_sm_object);
1854 		dump_spacemap(mos, vd->vdev_obsolete_sm);
1855 		(void) printf("\n");
1856 	}
1857 }
1858 
1859 static void
dump_metaslabs(spa_t * spa)1860 dump_metaslabs(spa_t *spa)
1861 {
1862 	vdev_t *vd, *rvd = spa->spa_root_vdev;
1863 	uint64_t m, c = 0, children = rvd->vdev_children;
1864 
1865 	(void) printf("\nMetaslabs:\n");
1866 
1867 	if (!dump_opt['d'] && zopt_metaslab_args > 0) {
1868 		c = zopt_metaslab[0];
1869 
1870 		if (c >= children)
1871 			(void) fatal("bad vdev id: %llu", (u_longlong_t)c);
1872 
1873 		if (zopt_metaslab_args > 1) {
1874 			vd = rvd->vdev_child[c];
1875 			print_vdev_metaslab_header(vd);
1876 
1877 			for (m = 1; m < zopt_metaslab_args; m++) {
1878 				if (zopt_metaslab[m] < vd->vdev_ms_count)
1879 					dump_metaslab(
1880 					    vd->vdev_ms[zopt_metaslab[m]]);
1881 				else
1882 					(void) fprintf(stderr, "bad metaslab "
1883 					    "number %llu\n",
1884 					    (u_longlong_t)zopt_metaslab[m]);
1885 			}
1886 			(void) printf("\n");
1887 			return;
1888 		}
1889 		children = c + 1;
1890 	}
1891 	for (; c < children; c++) {
1892 		vd = rvd->vdev_child[c];
1893 		print_vdev_metaslab_header(vd);
1894 
1895 		print_vdev_indirect(vd);
1896 
1897 		for (m = 0; m < vd->vdev_ms_count; m++)
1898 			dump_metaslab(vd->vdev_ms[m]);
1899 		(void) printf("\n");
1900 	}
1901 }
1902 
1903 static void
dump_log_spacemaps(spa_t * spa)1904 dump_log_spacemaps(spa_t *spa)
1905 {
1906 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1907 		return;
1908 
1909 	(void) printf("\nLog Space Maps in Pool:\n");
1910 	for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
1911 	    sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1912 		space_map_t *sm = NULL;
1913 		VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
1914 		    sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
1915 
1916 		(void) printf("Log Spacemap object %llu txg %llu\n",
1917 		    (u_longlong_t)sls->sls_sm_obj, (u_longlong_t)sls->sls_txg);
1918 		dump_spacemap(spa->spa_meta_objset, sm);
1919 		space_map_close(sm);
1920 	}
1921 	(void) printf("\n");
1922 }
1923 
1924 static void
dump_ddt_entry(const ddt_t * ddt,const ddt_lightweight_entry_t * ddlwe,uint64_t index)1925 dump_ddt_entry(const ddt_t *ddt, const ddt_lightweight_entry_t *ddlwe,
1926     uint64_t index)
1927 {
1928 	const ddt_key_t *ddk = &ddlwe->ddlwe_key;
1929 	char blkbuf[BP_SPRINTF_LEN];
1930 	blkptr_t blk;
1931 	int p;
1932 
1933 	for (p = 0; p < DDT_NPHYS(ddt); p++) {
1934 		const ddt_univ_phys_t *ddp = &ddlwe->ddlwe_phys;
1935 		ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p);
1936 
1937 		if (ddt_phys_birth(ddp, v) == 0)
1938 			continue;
1939 		ddt_bp_create(ddt->ddt_checksum, ddk, ddp, v, &blk);
1940 		snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
1941 		(void) printf("index %llx refcnt %llu phys %d %s\n",
1942 		    (u_longlong_t)index, (u_longlong_t)ddt_phys_refcnt(ddp, v),
1943 		    p, blkbuf);
1944 	}
1945 }
1946 
1947 static void
dump_dedup_ratio(const ddt_stat_t * dds)1948 dump_dedup_ratio(const ddt_stat_t *dds)
1949 {
1950 	double rL, rP, rD, D, dedup, compress, copies;
1951 
1952 	if (dds->dds_blocks == 0)
1953 		return;
1954 
1955 	rL = (double)dds->dds_ref_lsize;
1956 	rP = (double)dds->dds_ref_psize;
1957 	rD = (double)dds->dds_ref_dsize;
1958 	D = (double)dds->dds_dsize;
1959 
1960 	dedup = rD / D;
1961 	compress = rL / rP;
1962 	copies = rD / rP;
1963 
1964 	(void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
1965 	    "dedup * compress / copies = %.2f\n\n",
1966 	    dedup, compress, copies, dedup * compress / copies);
1967 }
1968 
1969 static void
dump_ddt_log(ddt_t * ddt)1970 dump_ddt_log(ddt_t *ddt)
1971 {
1972 	if (ddt->ddt_version != DDT_VERSION_FDT ||
1973 	    !(ddt->ddt_flags & DDT_FLAG_LOG))
1974 		return;
1975 
1976 	for (int n = 0; n < 2; n++) {
1977 		ddt_log_t *ddl = &ddt->ddt_log[n];
1978 
1979 		char flagstr[64] = {0};
1980 		if (ddl->ddl_flags > 0) {
1981 			flagstr[0] = ' ';
1982 			int c = 1;
1983 			if (ddl->ddl_flags & DDL_FLAG_FLUSHING)
1984 				c += strlcpy(&flagstr[c], " FLUSHING",
1985 				    sizeof (flagstr) - c);
1986 			if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT)
1987 				c += strlcpy(&flagstr[c], " CHECKPOINT",
1988 				    sizeof (flagstr) - c);
1989 			if (ddl->ddl_flags &
1990 			    ~(DDL_FLAG_FLUSHING|DDL_FLAG_CHECKPOINT))
1991 				c += strlcpy(&flagstr[c], " UNKNOWN",
1992 				    sizeof (flagstr) - c);
1993 			flagstr[1] = '[';
1994 			flagstr[c++] = ']';
1995 		}
1996 
1997 		uint64_t count = avl_numnodes(&ddl->ddl_tree);
1998 
1999 		printf(DMU_POOL_DDT_LOG ": flags=0x%02x%s; obj=%llu; "
2000 		    "len=%llu; txg=%llu; entries=%llu\n",
2001 		    zio_checksum_table[ddt->ddt_checksum].ci_name, n,
2002 		    ddl->ddl_flags, flagstr,
2003 		    (u_longlong_t)ddl->ddl_object,
2004 		    (u_longlong_t)ddl->ddl_length,
2005 		    (u_longlong_t)ddl->ddl_first_txg, (u_longlong_t)count);
2006 
2007 		if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT) {
2008 			const ddt_key_t *ddk = &ddl->ddl_checkpoint;
2009 			printf("    checkpoint: "
2010 			    "%016llx:%016llx:%016llx:%016llx:%016llx\n",
2011 			    (u_longlong_t)ddk->ddk_cksum.zc_word[0],
2012 			    (u_longlong_t)ddk->ddk_cksum.zc_word[1],
2013 			    (u_longlong_t)ddk->ddk_cksum.zc_word[2],
2014 			    (u_longlong_t)ddk->ddk_cksum.zc_word[3],
2015 			    (u_longlong_t)ddk->ddk_prop);
2016 		}
2017 
2018 		if (count == 0 || dump_opt['D'] < 4)
2019 			continue;
2020 
2021 		ddt_lightweight_entry_t ddlwe;
2022 		uint64_t index = 0;
2023 		for (ddt_log_entry_t *ddle = avl_first(&ddl->ddl_tree);
2024 		    ddle; ddle = AVL_NEXT(&ddl->ddl_tree, ddle)) {
2025 			DDT_LOG_ENTRY_TO_LIGHTWEIGHT(ddt, ddle, &ddlwe);
2026 			dump_ddt_entry(ddt, &ddlwe, index++);
2027 		}
2028 	}
2029 }
2030 
2031 static void
dump_ddt_object(ddt_t * ddt,ddt_type_t type,ddt_class_t class)2032 dump_ddt_object(ddt_t *ddt, ddt_type_t type, ddt_class_t class)
2033 {
2034 	char name[DDT_NAMELEN];
2035 	ddt_lightweight_entry_t ddlwe;
2036 	uint64_t walk = 0;
2037 	dmu_object_info_t doi;
2038 	uint64_t count, dspace, mspace;
2039 	int error;
2040 
2041 	error = ddt_object_info(ddt, type, class, &doi);
2042 
2043 	if (error == ENOENT)
2044 		return;
2045 	ASSERT(error == 0);
2046 
2047 	error = ddt_object_count(ddt, type, class, &count);
2048 	ASSERT(error == 0);
2049 	if (count == 0)
2050 		return;
2051 
2052 	dspace = doi.doi_physical_blocks_512 << 9;
2053 	mspace = doi.doi_fill_count * doi.doi_data_block_size;
2054 
2055 	ddt_object_name(ddt, type, class, name);
2056 
2057 	(void) printf("%s: dspace=%llu; mspace=%llu; entries=%llu\n", name,
2058 	    (u_longlong_t)dspace, (u_longlong_t)mspace, (u_longlong_t)count);
2059 
2060 	if (dump_opt['D'] < 3)
2061 		return;
2062 
2063 	(void) printf("%s: object=%llu\n", name,
2064 	    (u_longlong_t)ddt->ddt_object[type][class]);
2065 	zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class]);
2066 
2067 	if (dump_opt['D'] < 4)
2068 		return;
2069 
2070 	if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE)
2071 		return;
2072 
2073 	(void) printf("%s contents:\n\n", name);
2074 
2075 	while ((error = ddt_object_walk(ddt, type, class, &walk, &ddlwe)) == 0)
2076 		dump_ddt_entry(ddt, &ddlwe, walk);
2077 
2078 	ASSERT3U(error, ==, ENOENT);
2079 
2080 	(void) printf("\n");
2081 }
2082 
2083 static void
dump_ddt(ddt_t * ddt)2084 dump_ddt(ddt_t *ddt)
2085 {
2086 	if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
2087 		return;
2088 
2089 	char flagstr[64] = {0};
2090 	if (ddt->ddt_flags > 0) {
2091 		flagstr[0] = ' ';
2092 		int c = 1;
2093 		if (ddt->ddt_flags & DDT_FLAG_FLAT)
2094 			c += strlcpy(&flagstr[c], " FLAT",
2095 			    sizeof (flagstr) - c);
2096 		if (ddt->ddt_flags & DDT_FLAG_LOG)
2097 			c += strlcpy(&flagstr[c], " LOG",
2098 			    sizeof (flagstr) - c);
2099 		if (ddt->ddt_flags & ~DDT_FLAG_MASK)
2100 			c += strlcpy(&flagstr[c], " UNKNOWN",
2101 			    sizeof (flagstr) - c);
2102 		flagstr[1] = '[';
2103 		flagstr[c] = ']';
2104 	}
2105 
2106 	printf("DDT-%s: version=%llu [%s]; flags=0x%02llx%s; rootobj=%llu\n",
2107 	    zio_checksum_table[ddt->ddt_checksum].ci_name,
2108 	    (u_longlong_t)ddt->ddt_version,
2109 	    (ddt->ddt_version == 0) ? "LEGACY" :
2110 	    (ddt->ddt_version == 1) ? "FDT" : "UNKNOWN",
2111 	    (u_longlong_t)ddt->ddt_flags, flagstr,
2112 	    (u_longlong_t)ddt->ddt_dir_object);
2113 
2114 	for (ddt_type_t type = 0; type < DDT_TYPES; type++)
2115 		for (ddt_class_t class = 0; class < DDT_CLASSES; class++)
2116 			dump_ddt_object(ddt, type, class);
2117 
2118 	dump_ddt_log(ddt);
2119 }
2120 
2121 static void
dump_all_ddts(spa_t * spa)2122 dump_all_ddts(spa_t *spa)
2123 {
2124 	ddt_histogram_t ddh_total = {{{0}}};
2125 	ddt_stat_t dds_total = {0};
2126 
2127 	for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++)
2128 		dump_ddt(spa->spa_ddt[c]);
2129 
2130 	ddt_get_dedup_stats(spa, &dds_total);
2131 
2132 	if (dds_total.dds_blocks == 0) {
2133 		(void) printf("All DDTs are empty\n");
2134 		return;
2135 	}
2136 
2137 	(void) printf("\n");
2138 
2139 	if (dump_opt['D'] > 1) {
2140 		(void) printf("DDT histogram (aggregated over all DDTs):\n");
2141 		ddt_get_dedup_histogram(spa, &ddh_total);
2142 		zpool_dump_ddt(&dds_total, &ddh_total);
2143 	}
2144 
2145 	dump_dedup_ratio(&dds_total);
2146 
2147 	/*
2148 	 * Dump a histogram of unique class entry age
2149 	 */
2150 	if (dump_opt['D'] == 3 && getenv("ZDB_DDT_UNIQUE_AGE_HIST") != NULL) {
2151 		ddt_age_histo_t histogram;
2152 
2153 		(void) printf("DDT walk unique, building age histogram...\n");
2154 		ddt_prune_walk(spa, 0, &histogram);
2155 
2156 		/*
2157 		 * print out histogram for unique entry class birth
2158 		 */
2159 		if (histogram.dah_entries > 0) {
2160 			(void) printf("%5s  %9s  %4s\n",
2161 			    "age", "blocks", "amnt");
2162 			(void) printf("%5s  %9s  %4s\n",
2163 			    "-----", "---------", "----");
2164 			for (int i = 0; i < HIST_BINS; i++) {
2165 				(void) printf("%5d  %9d %4d%%\n", 1 << i,
2166 				    (int)histogram.dah_age_histo[i],
2167 				    (int)((histogram.dah_age_histo[i] * 100) /
2168 				    histogram.dah_entries));
2169 			}
2170 		}
2171 	}
2172 }
2173 
2174 static void
dump_brt(spa_t * spa)2175 dump_brt(spa_t *spa)
2176 {
2177 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) {
2178 		printf("BRT: unsupported on this pool\n");
2179 		return;
2180 	}
2181 
2182 	if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
2183 		printf("BRT: empty\n");
2184 		return;
2185 	}
2186 
2187 	char count[32], used[32], saved[32];
2188 	zdb_nicebytes(brt_get_used(spa), used, sizeof (used));
2189 	zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved));
2190 	uint64_t ratio = brt_get_ratio(spa);
2191 	printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved,
2192 	    (u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100));
2193 
2194 	if (dump_opt['T'] < 2)
2195 		return;
2196 
2197 	for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2198 		brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2199 		if (!brtvd->bv_initiated) {
2200 			printf("BRT: vdev %" PRIu64 ": empty\n", vdevid);
2201 			continue;
2202 		}
2203 
2204 		zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count));
2205 		zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used));
2206 		zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved));
2207 		printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n",
2208 		    vdevid, count, used, saved);
2209 	}
2210 
2211 	if (dump_opt['T'] < 3)
2212 		return;
2213 
2214 	/* -TTT shows a per-vdev histograms; -TTTT shows all entries */
2215 	boolean_t do_histo = dump_opt['T'] == 3;
2216 
2217 	char dva[64];
2218 
2219 	if (!do_histo)
2220 		printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2221 
2222 	for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2223 		brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2224 		if (!brtvd->bv_initiated)
2225 			continue;
2226 
2227 		uint64_t counts[64] = {};
2228 
2229 		zap_cursor_t zc;
2230 		zap_attribute_t *za = zap_attribute_alloc();
2231 		for (zap_cursor_init(&zc, spa->spa_meta_objset,
2232 		    brtvd->bv_mos_entries);
2233 		    zap_cursor_retrieve(&zc, za) == 0;
2234 		    zap_cursor_advance(&zc)) {
2235 			uint64_t refcnt;
2236 			VERIFY0(zap_lookup_uint64(spa->spa_meta_objset,
2237 			    brtvd->bv_mos_entries,
2238 			    (const uint64_t *)za->za_name, 1,
2239 			    za->za_integer_length, za->za_num_integers,
2240 			    &refcnt));
2241 
2242 			if (do_histo)
2243 				counts[highbit64(refcnt)]++;
2244 			else {
2245 				uint64_t offset =
2246 				    *(const uint64_t *)za->za_name;
2247 
2248 				snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx",
2249 				    vdevid, (u_longlong_t)offset);
2250 				printf("%-16s %-10llu\n", dva,
2251 				    (u_longlong_t)refcnt);
2252 			}
2253 		}
2254 		zap_cursor_fini(&zc);
2255 		zap_attribute_free(za);
2256 
2257 		if (do_histo) {
2258 			printf("\nBRT: vdev %" PRIu64
2259 			    ": DVAs with 2^n refcnts:\n", vdevid);
2260 			dump_histogram(counts, 64, 0);
2261 		}
2262 	}
2263 }
2264 
2265 static void
dump_dtl_seg(void * arg,uint64_t start,uint64_t size)2266 dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
2267 {
2268 	char *prefix = arg;
2269 
2270 	(void) printf("%s [%llu,%llu) length %llu\n",
2271 	    prefix,
2272 	    (u_longlong_t)start,
2273 	    (u_longlong_t)(start + size),
2274 	    (u_longlong_t)(size));
2275 }
2276 
2277 static void
dump_dtl(vdev_t * vd,int indent)2278 dump_dtl(vdev_t *vd, int indent)
2279 {
2280 	spa_t *spa = vd->vdev_spa;
2281 	boolean_t required;
2282 	const char *name[DTL_TYPES] = { "missing", "partial", "scrub",
2283 		"outage" };
2284 	char prefix[256];
2285 
2286 	spa_vdev_state_enter(spa, SCL_NONE);
2287 	required = vdev_dtl_required(vd);
2288 	(void) spa_vdev_state_exit(spa, NULL, 0);
2289 
2290 	if (indent == 0)
2291 		(void) printf("\nDirty time logs:\n\n");
2292 
2293 	(void) printf("\t%*s%s [%s]\n", indent, "",
2294 	    vd->vdev_path ? vd->vdev_path :
2295 	    vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa),
2296 	    required ? "DTL-required" : "DTL-expendable");
2297 
2298 	for (int t = 0; t < DTL_TYPES; t++) {
2299 		zfs_range_tree_t *rt = vd->vdev_dtl[t];
2300 		if (zfs_range_tree_space(rt) == 0)
2301 			continue;
2302 		(void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
2303 		    indent + 2, "", name[t]);
2304 		zfs_range_tree_walk(rt, dump_dtl_seg, prefix);
2305 		if (dump_opt['d'] > 5 && vd->vdev_children == 0)
2306 			dump_spacemap(spa->spa_meta_objset,
2307 			    vd->vdev_dtl_sm);
2308 	}
2309 
2310 	for (unsigned c = 0; c < vd->vdev_children; c++)
2311 		dump_dtl(vd->vdev_child[c], indent + 4);
2312 }
2313 
2314 static void
dump_history(spa_t * spa)2315 dump_history(spa_t *spa)
2316 {
2317 	nvlist_t **events = NULL;
2318 	char *buf;
2319 	uint64_t resid, len, off = 0;
2320 	uint_t num = 0;
2321 	int error;
2322 	char tbuf[30];
2323 
2324 	if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) {
2325 		(void) fprintf(stderr, "%s: unable to allocate I/O buffer\n",
2326 		    __func__);
2327 		return;
2328 	}
2329 
2330 	do {
2331 		len = SPA_OLD_MAXBLOCKSIZE;
2332 
2333 		if ((error = spa_history_get(spa, &off, &len, buf)) != 0) {
2334 			(void) fprintf(stderr, "Unable to read history: "
2335 			    "error %d\n", error);
2336 			free(buf);
2337 			return;
2338 		}
2339 
2340 		if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0)
2341 			break;
2342 
2343 		off -= resid;
2344 	} while (len != 0);
2345 
2346 	(void) printf("\nHistory:\n");
2347 	for (unsigned i = 0; i < num; i++) {
2348 		boolean_t printed = B_FALSE;
2349 
2350 		if (nvlist_exists(events[i], ZPOOL_HIST_TIME)) {
2351 			time_t tsec;
2352 			struct tm t;
2353 
2354 			tsec = fnvlist_lookup_uint64(events[i],
2355 			    ZPOOL_HIST_TIME);
2356 			(void) localtime_r(&tsec, &t);
2357 			(void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t);
2358 		} else {
2359 			tbuf[0] = '\0';
2360 		}
2361 
2362 		if (nvlist_exists(events[i], ZPOOL_HIST_CMD)) {
2363 			(void) printf("%s %s\n", tbuf,
2364 			    fnvlist_lookup_string(events[i], ZPOOL_HIST_CMD));
2365 		} else if (nvlist_exists(events[i], ZPOOL_HIST_INT_EVENT)) {
2366 			uint64_t ievent;
2367 
2368 			ievent = fnvlist_lookup_uint64(events[i],
2369 			    ZPOOL_HIST_INT_EVENT);
2370 			if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS)
2371 				goto next;
2372 
2373 			(void) printf(" %s [internal %s txg:%ju] %s\n",
2374 			    tbuf,
2375 			    zfs_history_event_names[ievent],
2376 			    fnvlist_lookup_uint64(events[i],
2377 			    ZPOOL_HIST_TXG),
2378 			    fnvlist_lookup_string(events[i],
2379 			    ZPOOL_HIST_INT_STR));
2380 		} else if (nvlist_exists(events[i], ZPOOL_HIST_INT_NAME)) {
2381 			(void) printf("%s [txg:%ju] %s", tbuf,
2382 			    fnvlist_lookup_uint64(events[i],
2383 			    ZPOOL_HIST_TXG),
2384 			    fnvlist_lookup_string(events[i],
2385 			    ZPOOL_HIST_INT_NAME));
2386 
2387 			if (nvlist_exists(events[i], ZPOOL_HIST_DSNAME)) {
2388 				(void) printf(" %s (%llu)",
2389 				    fnvlist_lookup_string(events[i],
2390 				    ZPOOL_HIST_DSNAME),
2391 				    (u_longlong_t)fnvlist_lookup_uint64(
2392 				    events[i],
2393 				    ZPOOL_HIST_DSID));
2394 			}
2395 
2396 			(void) printf(" %s\n", fnvlist_lookup_string(events[i],
2397 			    ZPOOL_HIST_INT_STR));
2398 		} else if (nvlist_exists(events[i], ZPOOL_HIST_IOCTL)) {
2399 			(void) printf("%s ioctl %s\n", tbuf,
2400 			    fnvlist_lookup_string(events[i],
2401 			    ZPOOL_HIST_IOCTL));
2402 
2403 			if (nvlist_exists(events[i], ZPOOL_HIST_INPUT_NVL)) {
2404 				(void) printf("    input:\n");
2405 				dump_nvlist(fnvlist_lookup_nvlist(events[i],
2406 				    ZPOOL_HIST_INPUT_NVL), 8);
2407 			}
2408 			if (nvlist_exists(events[i], ZPOOL_HIST_OUTPUT_NVL)) {
2409 				(void) printf("    output:\n");
2410 				dump_nvlist(fnvlist_lookup_nvlist(events[i],
2411 				    ZPOOL_HIST_OUTPUT_NVL), 8);
2412 			}
2413 			if (nvlist_exists(events[i], ZPOOL_HIST_ERRNO)) {
2414 				(void) printf("    errno: %lld\n",
2415 				    (longlong_t)fnvlist_lookup_int64(events[i],
2416 				    ZPOOL_HIST_ERRNO));
2417 			}
2418 		} else {
2419 			goto next;
2420 		}
2421 
2422 		printed = B_TRUE;
2423 next:
2424 		if (dump_opt['h'] > 1) {
2425 			if (!printed)
2426 				(void) printf("unrecognized record:\n");
2427 			dump_nvlist(events[i], 2);
2428 		}
2429 	}
2430 	free(buf);
2431 }
2432 
2433 static void
dump_dnode(objset_t * os,uint64_t object,void * data,size_t size)2434 dump_dnode(objset_t *os, uint64_t object, void *data, size_t size)
2435 {
2436 	(void) os, (void) object, (void) data, (void) size;
2437 }
2438 
2439 static uint64_t
blkid2offset(const dnode_phys_t * dnp,const blkptr_t * bp,const zbookmark_phys_t * zb)2440 blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
2441     const zbookmark_phys_t *zb)
2442 {
2443 	if (dnp == NULL) {
2444 		ASSERT(zb->zb_level < 0);
2445 		if (zb->zb_object == 0)
2446 			return (zb->zb_blkid);
2447 		return (zb->zb_blkid * BP_GET_LSIZE(bp));
2448 	}
2449 
2450 	ASSERT(zb->zb_level >= 0);
2451 
2452 	return ((zb->zb_blkid <<
2453 	    (zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) *
2454 	    dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
2455 }
2456 
2457 static void
snprintf_zstd_header(spa_t * spa,char * blkbuf,size_t buflen,const blkptr_t * bp)2458 snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
2459     const blkptr_t *bp)
2460 {
2461 	static abd_t *pabd = NULL;
2462 	void *buf;
2463 	zio_t *zio;
2464 	zfs_zstdhdr_t zstd_hdr;
2465 	int error;
2466 
2467 	if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD)
2468 		return;
2469 
2470 	if (BP_IS_HOLE(bp))
2471 		return;
2472 
2473 	if (BP_IS_EMBEDDED(bp)) {
2474 		buf = malloc(SPA_MAXBLOCKSIZE);
2475 		if (buf == NULL) {
2476 			(void) fprintf(stderr, "out of memory\n");
2477 			zdb_exit(1);
2478 		}
2479 		decode_embedded_bp_compressed(bp, buf);
2480 		memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2481 		free(buf);
2482 		zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2483 		zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2484 		(void) snprintf(blkbuf + strlen(blkbuf),
2485 		    buflen - strlen(blkbuf),
2486 		    " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2487 		    zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2488 		    zfs_get_hdrlevel(&zstd_hdr));
2489 		return;
2490 	}
2491 
2492 	if (!pabd)
2493 		pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
2494 	zio = zio_root(spa, NULL, NULL, 0);
2495 
2496 	/* Decrypt but don't decompress so we can read the compression header */
2497 	zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL,
2498 	    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS,
2499 	    NULL));
2500 	error = zio_wait(zio);
2501 	if (error) {
2502 		(void) fprintf(stderr, "read failed: %d\n", error);
2503 		return;
2504 	}
2505 	buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp));
2506 	memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2507 	zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2508 	zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2509 
2510 	(void) snprintf(blkbuf + strlen(blkbuf),
2511 	    buflen - strlen(blkbuf),
2512 	    " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2513 	    zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2514 	    zfs_get_hdrlevel(&zstd_hdr));
2515 
2516 	abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp));
2517 }
2518 
2519 static void
snprintf_blkptr_compact(char * blkbuf,size_t buflen,const blkptr_t * bp,boolean_t bp_freed)2520 snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
2521     boolean_t bp_freed)
2522 {
2523 	const dva_t *dva = bp->blk_dva;
2524 	int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1;
2525 	int i;
2526 
2527 	if (dump_opt['b'] >= 6) {
2528 		snprintf_blkptr(blkbuf, buflen, bp);
2529 		if (bp_freed) {
2530 			(void) snprintf(blkbuf + strlen(blkbuf),
2531 			    buflen - strlen(blkbuf), " %s", "FREE");
2532 		}
2533 		return;
2534 	}
2535 
2536 	if (BP_IS_EMBEDDED(bp)) {
2537 		(void) sprintf(blkbuf,
2538 		    "EMBEDDED et=%u %llxL/%llxP B=%llu",
2539 		    (int)BPE_GET_ETYPE(bp),
2540 		    (u_longlong_t)BPE_GET_LSIZE(bp),
2541 		    (u_longlong_t)BPE_GET_PSIZE(bp),
2542 		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2543 		return;
2544 	}
2545 
2546 	blkbuf[0] = '\0';
2547 
2548 	for (i = 0; i < ndvas; i++) {
2549 		(void) snprintf(blkbuf + strlen(blkbuf),
2550 		    buflen - strlen(blkbuf), "%llu:%llx:%llx%s ",
2551 		    (u_longlong_t)DVA_GET_VDEV(&dva[i]),
2552 		    (u_longlong_t)DVA_GET_OFFSET(&dva[i]),
2553 		    (u_longlong_t)DVA_GET_ASIZE(&dva[i]),
2554 		    (DVA_GET_GANG(&dva[i]) ? "G" : ""));
2555 	}
2556 
2557 	if (BP_IS_HOLE(bp)) {
2558 		(void) snprintf(blkbuf + strlen(blkbuf),
2559 		    buflen - strlen(blkbuf),
2560 		    "%llxL B=%llu",
2561 		    (u_longlong_t)BP_GET_LSIZE(bp),
2562 		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2563 	} else {
2564 		(void) snprintf(blkbuf + strlen(blkbuf),
2565 		    buflen - strlen(blkbuf),
2566 		    "%llxL/%llxP F=%llu B=%llu/%llu",
2567 		    (u_longlong_t)BP_GET_LSIZE(bp),
2568 		    (u_longlong_t)BP_GET_PSIZE(bp),
2569 		    (u_longlong_t)BP_GET_FILL(bp),
2570 		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),
2571 		    (u_longlong_t)BP_GET_BIRTH(bp));
2572 		if (bp_freed)
2573 			(void) snprintf(blkbuf + strlen(blkbuf),
2574 			    buflen - strlen(blkbuf), " %s", "FREE");
2575 		(void) snprintf(blkbuf + strlen(blkbuf),
2576 		    buflen - strlen(blkbuf),
2577 		    " cksum=%016llx:%016llx:%016llx:%016llx",
2578 		    (u_longlong_t)bp->blk_cksum.zc_word[0],
2579 		    (u_longlong_t)bp->blk_cksum.zc_word[1],
2580 		    (u_longlong_t)bp->blk_cksum.zc_word[2],
2581 		    (u_longlong_t)bp->blk_cksum.zc_word[3]);
2582 	}
2583 }
2584 
2585 static void
print_indirect(spa_t * spa,blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp)2586 print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
2587     const dnode_phys_t *dnp)
2588 {
2589 	char blkbuf[BP_SPRINTF_LEN];
2590 	int l;
2591 
2592 	if (!BP_IS_EMBEDDED(bp)) {
2593 		ASSERT3U(BP_GET_TYPE(bp), ==, dnp->dn_type);
2594 		ASSERT3U(BP_GET_LEVEL(bp), ==, zb->zb_level);
2595 	}
2596 
2597 	(void) printf("%16llx ", (u_longlong_t)blkid2offset(dnp, bp, zb));
2598 
2599 	ASSERT(zb->zb_level >= 0);
2600 
2601 	for (l = dnp->dn_nlevels - 1; l >= -1; l--) {
2602 		if (l == zb->zb_level) {
2603 			(void) printf("L%llx", (u_longlong_t)zb->zb_level);
2604 		} else {
2605 			(void) printf(" ");
2606 		}
2607 	}
2608 
2609 	snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE);
2610 	if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD)
2611 		snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp);
2612 	(void) printf("%s\n", blkbuf);
2613 }
2614 
2615 static int
visit_indirect(spa_t * spa,const dnode_phys_t * dnp,blkptr_t * bp,const zbookmark_phys_t * zb)2616 visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
2617     blkptr_t *bp, const zbookmark_phys_t *zb)
2618 {
2619 	int err = 0;
2620 
2621 	if (BP_GET_LOGICAL_BIRTH(bp) == 0)
2622 		return (0);
2623 
2624 	print_indirect(spa, bp, zb, dnp);
2625 
2626 	if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
2627 		arc_flags_t flags = ARC_FLAG_WAIT;
2628 		int i;
2629 		blkptr_t *cbp;
2630 		int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
2631 		arc_buf_t *buf;
2632 		uint64_t fill = 0;
2633 		ASSERT(!BP_IS_REDACTED(bp));
2634 
2635 		err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
2636 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
2637 		if (err)
2638 			return (err);
2639 		ASSERT(buf->b_data);
2640 
2641 		/* recursively visit blocks below this */
2642 		cbp = buf->b_data;
2643 		for (i = 0; i < epb; i++, cbp++) {
2644 			zbookmark_phys_t czb;
2645 
2646 			SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
2647 			    zb->zb_level - 1,
2648 			    zb->zb_blkid * epb + i);
2649 			err = visit_indirect(spa, dnp, cbp, &czb);
2650 			if (err)
2651 				break;
2652 			fill += BP_GET_FILL(cbp);
2653 		}
2654 		if (!err)
2655 			ASSERT3U(fill, ==, BP_GET_FILL(bp));
2656 		arc_buf_destroy(buf, &buf);
2657 	}
2658 
2659 	return (err);
2660 }
2661 
2662 static void
dump_indirect(dnode_t * dn)2663 dump_indirect(dnode_t *dn)
2664 {
2665 	dnode_phys_t *dnp = dn->dn_phys;
2666 	zbookmark_phys_t czb;
2667 
2668 	(void) printf("Indirect blocks:\n");
2669 
2670 	SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
2671 	    dn->dn_object, dnp->dn_nlevels - 1, 0);
2672 	for (int j = 0; j < dnp->dn_nblkptr; j++) {
2673 		czb.zb_blkid = j;
2674 		(void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
2675 		    &dnp->dn_blkptr[j], &czb);
2676 	}
2677 
2678 	(void) printf("\n");
2679 }
2680 
2681 static void
dump_dsl_dir(objset_t * os,uint64_t object,void * data,size_t size)2682 dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size)
2683 {
2684 	(void) os, (void) object;
2685 	dsl_dir_phys_t *dd = data;
2686 	time_t crtime;
2687 	char nice[32];
2688 
2689 	/* make sure nicenum has enough space */
2690 	_Static_assert(sizeof (nice) >= NN_NUMBUF_SZ, "nice truncated");
2691 
2692 	if (dd == NULL)
2693 		return;
2694 
2695 	ASSERT3U(size, >=, sizeof (dsl_dir_phys_t));
2696 
2697 	crtime = dd->dd_creation_time;
2698 	(void) printf("\t\tcreation_time = %s", ctime(&crtime));
2699 	(void) printf("\t\thead_dataset_obj = %llu\n",
2700 	    (u_longlong_t)dd->dd_head_dataset_obj);
2701 	(void) printf("\t\tparent_dir_obj = %llu\n",
2702 	    (u_longlong_t)dd->dd_parent_obj);
2703 	(void) printf("\t\torigin_obj = %llu\n",
2704 	    (u_longlong_t)dd->dd_origin_obj);
2705 	(void) printf("\t\tchild_dir_zapobj = %llu\n",
2706 	    (u_longlong_t)dd->dd_child_dir_zapobj);
2707 	zdb_nicenum(dd->dd_used_bytes, nice, sizeof (nice));
2708 	(void) printf("\t\tused_bytes = %s\n", nice);
2709 	zdb_nicenum(dd->dd_compressed_bytes, nice, sizeof (nice));
2710 	(void) printf("\t\tcompressed_bytes = %s\n", nice);
2711 	zdb_nicenum(dd->dd_uncompressed_bytes, nice, sizeof (nice));
2712 	(void) printf("\t\tuncompressed_bytes = %s\n", nice);
2713 	zdb_nicenum(dd->dd_quota, nice, sizeof (nice));
2714 	(void) printf("\t\tquota = %s\n", nice);
2715 	zdb_nicenum(dd->dd_reserved, nice, sizeof (nice));
2716 	(void) printf("\t\treserved = %s\n", nice);
2717 	(void) printf("\t\tprops_zapobj = %llu\n",
2718 	    (u_longlong_t)dd->dd_props_zapobj);
2719 	(void) printf("\t\tdeleg_zapobj = %llu\n",
2720 	    (u_longlong_t)dd->dd_deleg_zapobj);
2721 	(void) printf("\t\tflags = %llx\n",
2722 	    (u_longlong_t)dd->dd_flags);
2723 
2724 #define	DO(which) \
2725 	zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2726 	    sizeof (nice)); \
2727 	(void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2728 	DO(HEAD);
2729 	DO(SNAP);
2730 	DO(CHILD);
2731 	DO(CHILD_RSRV);
2732 	DO(REFRSRV);
2733 #undef DO
2734 	(void) printf("\t\tclones = %llu\n",
2735 	    (u_longlong_t)dd->dd_clones);
2736 }
2737 
2738 static void
dump_dsl_dataset(objset_t * os,uint64_t object,void * data,size_t size)2739 dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
2740 {
2741 	(void) os, (void) object;
2742 	dsl_dataset_phys_t *ds = data;
2743 	time_t crtime;
2744 	char used[32], compressed[32], uncompressed[32], unique[32];
2745 	char blkbuf[BP_SPRINTF_LEN];
2746 
2747 	/* make sure nicenum has enough space */
2748 	_Static_assert(sizeof (used) >= NN_NUMBUF_SZ, "used truncated");
2749 	_Static_assert(sizeof (compressed) >= NN_NUMBUF_SZ,
2750 	    "compressed truncated");
2751 	_Static_assert(sizeof (uncompressed) >= NN_NUMBUF_SZ,
2752 	    "uncompressed truncated");
2753 	_Static_assert(sizeof (unique) >= NN_NUMBUF_SZ, "unique truncated");
2754 
2755 	if (ds == NULL)
2756 		return;
2757 
2758 	ASSERT(size == sizeof (*ds));
2759 	crtime = ds->ds_creation_time;
2760 	zdb_nicenum(ds->ds_referenced_bytes, used, sizeof (used));
2761 	zdb_nicenum(ds->ds_compressed_bytes, compressed, sizeof (compressed));
2762 	zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed,
2763 	    sizeof (uncompressed));
2764 	zdb_nicenum(ds->ds_unique_bytes, unique, sizeof (unique));
2765 	snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp);
2766 
2767 	(void) printf("\t\tdir_obj = %llu\n",
2768 	    (u_longlong_t)ds->ds_dir_obj);
2769 	(void) printf("\t\tprev_snap_obj = %llu\n",
2770 	    (u_longlong_t)ds->ds_prev_snap_obj);
2771 	(void) printf("\t\tprev_snap_txg = %llu\n",
2772 	    (u_longlong_t)ds->ds_prev_snap_txg);
2773 	(void) printf("\t\tnext_snap_obj = %llu\n",
2774 	    (u_longlong_t)ds->ds_next_snap_obj);
2775 	(void) printf("\t\tsnapnames_zapobj = %llu\n",
2776 	    (u_longlong_t)ds->ds_snapnames_zapobj);
2777 	(void) printf("\t\tnum_children = %llu\n",
2778 	    (u_longlong_t)ds->ds_num_children);
2779 	(void) printf("\t\tuserrefs_obj = %llu\n",
2780 	    (u_longlong_t)ds->ds_userrefs_obj);
2781 	(void) printf("\t\tcreation_time = %s", ctime(&crtime));
2782 	(void) printf("\t\tcreation_txg = %llu\n",
2783 	    (u_longlong_t)ds->ds_creation_txg);
2784 	(void) printf("\t\tdeadlist_obj = %llu\n",
2785 	    (u_longlong_t)ds->ds_deadlist_obj);
2786 	(void) printf("\t\tused_bytes = %s\n", used);
2787 	(void) printf("\t\tcompressed_bytes = %s\n", compressed);
2788 	(void) printf("\t\tuncompressed_bytes = %s\n", uncompressed);
2789 	(void) printf("\t\tunique = %s\n", unique);
2790 	(void) printf("\t\tfsid_guid = %llu\n",
2791 	    (u_longlong_t)ds->ds_fsid_guid);
2792 	(void) printf("\t\tguid = %llu\n",
2793 	    (u_longlong_t)ds->ds_guid);
2794 	(void) printf("\t\tflags = %llx\n",
2795 	    (u_longlong_t)ds->ds_flags);
2796 	(void) printf("\t\tnext_clones_obj = %llu\n",
2797 	    (u_longlong_t)ds->ds_next_clones_obj);
2798 	(void) printf("\t\tprops_obj = %llu\n",
2799 	    (u_longlong_t)ds->ds_props_obj);
2800 	(void) printf("\t\tbp = %s\n", blkbuf);
2801 }
2802 
2803 static int
dump_bptree_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)2804 dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2805 {
2806 	(void) arg, (void) tx;
2807 	char blkbuf[BP_SPRINTF_LEN];
2808 
2809 	if (BP_GET_LOGICAL_BIRTH(bp) != 0) {
2810 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
2811 		(void) printf("\t%s\n", blkbuf);
2812 	}
2813 	return (0);
2814 }
2815 
2816 static void
dump_bptree(objset_t * os,uint64_t obj,const char * name)2817 dump_bptree(objset_t *os, uint64_t obj, const char *name)
2818 {
2819 	char bytes[32];
2820 	bptree_phys_t *bt;
2821 	dmu_buf_t *db;
2822 
2823 	/* make sure nicenum has enough space */
2824 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2825 
2826 	if (dump_opt['d'] < 3)
2827 		return;
2828 
2829 	VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
2830 	bt = db->db_data;
2831 	zdb_nicenum(bt->bt_bytes, bytes, sizeof (bytes));
2832 	(void) printf("\n    %s: %llu datasets, %s\n",
2833 	    name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes);
2834 	dmu_buf_rele(db, FTAG);
2835 
2836 	if (dump_opt['d'] < 5)
2837 		return;
2838 
2839 	(void) printf("\n");
2840 
2841 	(void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL);
2842 }
2843 
2844 static int
dump_bpobj_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)2845 dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
2846 {
2847 	(void) arg, (void) tx;
2848 	char blkbuf[BP_SPRINTF_LEN];
2849 
2850 	ASSERT(BP_GET_LOGICAL_BIRTH(bp) != 0);
2851 	snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed);
2852 	(void) printf("\t%s\n", blkbuf);
2853 	return (0);
2854 }
2855 
2856 static void
dump_full_bpobj(bpobj_t * bpo,const char * name,int indent)2857 dump_full_bpobj(bpobj_t *bpo, const char *name, int indent)
2858 {
2859 	char bytes[32];
2860 	char comp[32];
2861 	char uncomp[32];
2862 	uint64_t i;
2863 
2864 	/* make sure nicenum has enough space */
2865 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2866 	_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
2867 	_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
2868 
2869 	if (dump_opt['d'] < 3)
2870 		return;
2871 
2872 	zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes, sizeof (bytes));
2873 	if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
2874 		zdb_nicenum(bpo->bpo_phys->bpo_comp, comp, sizeof (comp));
2875 		zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp, sizeof (uncomp));
2876 		if (bpo->bpo_havefreed) {
2877 			(void) printf("    %*s: object %llu, %llu local "
2878 			    "blkptrs, %llu freed, %llu subobjs in object %llu, "
2879 			    "%s (%s/%s comp)\n",
2880 			    indent * 8, name,
2881 			    (u_longlong_t)bpo->bpo_object,
2882 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2883 			    (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2884 			    (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2885 			    (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2886 			    bytes, comp, uncomp);
2887 		} else {
2888 			(void) printf("    %*s: object %llu, %llu local "
2889 			    "blkptrs, %llu subobjs in object %llu, "
2890 			    "%s (%s/%s comp)\n",
2891 			    indent * 8, name,
2892 			    (u_longlong_t)bpo->bpo_object,
2893 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2894 			    (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2895 			    (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2896 			    bytes, comp, uncomp);
2897 		}
2898 
2899 		for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
2900 			uint64_t subobj;
2901 			bpobj_t subbpo;
2902 			int error;
2903 			VERIFY0(dmu_read(bpo->bpo_os,
2904 			    bpo->bpo_phys->bpo_subobjs,
2905 			    i * sizeof (subobj), sizeof (subobj), &subobj, 0));
2906 			error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
2907 			if (error != 0) {
2908 				(void) printf("ERROR %u while trying to open "
2909 				    "subobj id %llu\n",
2910 				    error, (u_longlong_t)subobj);
2911 				continue;
2912 			}
2913 			dump_full_bpobj(&subbpo, "subobj", indent + 1);
2914 			bpobj_close(&subbpo);
2915 		}
2916 	} else {
2917 		if (bpo->bpo_havefreed) {
2918 			(void) printf("    %*s: object %llu, %llu blkptrs, "
2919 			    "%llu freed, %s\n",
2920 			    indent * 8, name,
2921 			    (u_longlong_t)bpo->bpo_object,
2922 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2923 			    (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2924 			    bytes);
2925 		} else {
2926 			(void) printf("    %*s: object %llu, %llu blkptrs, "
2927 			    "%s\n",
2928 			    indent * 8, name,
2929 			    (u_longlong_t)bpo->bpo_object,
2930 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2931 			    bytes);
2932 		}
2933 	}
2934 
2935 	if (dump_opt['d'] < 5)
2936 		return;
2937 
2938 
2939 	if (indent == 0) {
2940 		(void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL);
2941 		(void) printf("\n");
2942 	}
2943 }
2944 
2945 static int
dump_bookmark(dsl_pool_t * dp,char * name,boolean_t print_redact,boolean_t print_list)2946 dump_bookmark(dsl_pool_t *dp, char *name, boolean_t print_redact,
2947     boolean_t print_list)
2948 {
2949 	int err = 0;
2950 	zfs_bookmark_phys_t prop;
2951 	objset_t *mos = dp->dp_spa->spa_meta_objset;
2952 	err = dsl_bookmark_lookup(dp, name, NULL, &prop);
2953 
2954 	if (err != 0) {
2955 		return (err);
2956 	}
2957 
2958 	(void) printf("\t#%s: ", strchr(name, '#') + 1);
2959 	(void) printf("{guid: %llx creation_txg: %llu creation_time: "
2960 	    "%llu redaction_obj: %llu}\n", (u_longlong_t)prop.zbm_guid,
2961 	    (u_longlong_t)prop.zbm_creation_txg,
2962 	    (u_longlong_t)prop.zbm_creation_time,
2963 	    (u_longlong_t)prop.zbm_redaction_obj);
2964 
2965 	IMPLY(print_list, print_redact);
2966 	if (!print_redact || prop.zbm_redaction_obj == 0)
2967 		return (0);
2968 
2969 	redaction_list_t *rl;
2970 	VERIFY0(dsl_redaction_list_hold_obj(dp,
2971 	    prop.zbm_redaction_obj, FTAG, &rl));
2972 
2973 	redaction_list_phys_t *rlp = rl->rl_phys;
2974 	(void) printf("\tRedacted:\n\t\tProgress: ");
2975 	if (rlp->rlp_last_object != UINT64_MAX ||
2976 	    rlp->rlp_last_blkid != UINT64_MAX) {
2977 		(void) printf("%llu %llu (incomplete)\n",
2978 		    (u_longlong_t)rlp->rlp_last_object,
2979 		    (u_longlong_t)rlp->rlp_last_blkid);
2980 	} else {
2981 		(void) printf("complete\n");
2982 	}
2983 	(void) printf("\t\tSnapshots: [");
2984 	for (unsigned int i = 0; i < rlp->rlp_num_snaps; i++) {
2985 		if (i > 0)
2986 			(void) printf(", ");
2987 		(void) printf("%0llu",
2988 		    (u_longlong_t)rlp->rlp_snaps[i]);
2989 	}
2990 	(void) printf("]\n\t\tLength: %llu\n",
2991 	    (u_longlong_t)rlp->rlp_num_entries);
2992 
2993 	if (!print_list) {
2994 		dsl_redaction_list_rele(rl, FTAG);
2995 		return (0);
2996 	}
2997 
2998 	if (rlp->rlp_num_entries == 0) {
2999 		dsl_redaction_list_rele(rl, FTAG);
3000 		(void) printf("\t\tRedaction List: []\n\n");
3001 		return (0);
3002 	}
3003 
3004 	redact_block_phys_t *rbp_buf;
3005 	uint64_t size;
3006 	dmu_object_info_t doi;
3007 
3008 	VERIFY0(dmu_object_info(mos, prop.zbm_redaction_obj, &doi));
3009 	size = doi.doi_max_offset;
3010 	rbp_buf = kmem_alloc(size, KM_SLEEP);
3011 
3012 	err = dmu_read(mos, prop.zbm_redaction_obj, 0, size,
3013 	    rbp_buf, 0);
3014 	if (err != 0) {
3015 		dsl_redaction_list_rele(rl, FTAG);
3016 		kmem_free(rbp_buf, size);
3017 		return (err);
3018 	}
3019 
3020 	(void) printf("\t\tRedaction List: [{object: %llx, offset: "
3021 	    "%llx, blksz: %x, count: %llx}",
3022 	    (u_longlong_t)rbp_buf[0].rbp_object,
3023 	    (u_longlong_t)rbp_buf[0].rbp_blkid,
3024 	    (uint_t)(redact_block_get_size(&rbp_buf[0])),
3025 	    (u_longlong_t)redact_block_get_count(&rbp_buf[0]));
3026 
3027 	for (size_t i = 1; i < rlp->rlp_num_entries; i++) {
3028 		(void) printf(",\n\t\t{object: %llx, offset: %llx, "
3029 		    "blksz: %x, count: %llx}",
3030 		    (u_longlong_t)rbp_buf[i].rbp_object,
3031 		    (u_longlong_t)rbp_buf[i].rbp_blkid,
3032 		    (uint_t)(redact_block_get_size(&rbp_buf[i])),
3033 		    (u_longlong_t)redact_block_get_count(&rbp_buf[i]));
3034 	}
3035 	dsl_redaction_list_rele(rl, FTAG);
3036 	kmem_free(rbp_buf, size);
3037 	(void) printf("]\n\n");
3038 	return (0);
3039 }
3040 
3041 static void
dump_bookmarks(objset_t * os,int verbosity)3042 dump_bookmarks(objset_t *os, int verbosity)
3043 {
3044 	zap_cursor_t zc;
3045 	zap_attribute_t *attrp;
3046 	dsl_dataset_t *ds = dmu_objset_ds(os);
3047 	dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3048 	objset_t *mos = os->os_spa->spa_meta_objset;
3049 	if (verbosity < 4)
3050 		return;
3051 	attrp = zap_attribute_alloc();
3052 	dsl_pool_config_enter(dp, FTAG);
3053 
3054 	for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj);
3055 	    zap_cursor_retrieve(&zc, attrp) == 0;
3056 	    zap_cursor_advance(&zc)) {
3057 		char osname[ZFS_MAX_DATASET_NAME_LEN];
3058 		char buf[ZFS_MAX_DATASET_NAME_LEN];
3059 		int len;
3060 		dmu_objset_name(os, osname);
3061 		len = snprintf(buf, sizeof (buf), "%s#%s", osname,
3062 		    attrp->za_name);
3063 		VERIFY3S(len, <, ZFS_MAX_DATASET_NAME_LEN);
3064 		(void) dump_bookmark(dp, buf, verbosity >= 5, verbosity >= 6);
3065 	}
3066 	zap_cursor_fini(&zc);
3067 	dsl_pool_config_exit(dp, FTAG);
3068 	zap_attribute_free(attrp);
3069 }
3070 
3071 static void
bpobj_count_refd(bpobj_t * bpo)3072 bpobj_count_refd(bpobj_t *bpo)
3073 {
3074 	mos_obj_refd(bpo->bpo_object);
3075 
3076 	if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
3077 		mos_obj_refd(bpo->bpo_phys->bpo_subobjs);
3078 		for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
3079 			uint64_t subobj;
3080 			bpobj_t subbpo;
3081 			int error;
3082 			VERIFY0(dmu_read(bpo->bpo_os,
3083 			    bpo->bpo_phys->bpo_subobjs,
3084 			    i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3085 			error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3086 			if (error != 0) {
3087 				(void) printf("ERROR %u while trying to open "
3088 				    "subobj id %llu\n",
3089 				    error, (u_longlong_t)subobj);
3090 				continue;
3091 			}
3092 			bpobj_count_refd(&subbpo);
3093 			bpobj_close(&subbpo);
3094 		}
3095 	}
3096 }
3097 
3098 static int
dsl_deadlist_entry_count_refd(void * arg,dsl_deadlist_entry_t * dle)3099 dsl_deadlist_entry_count_refd(void *arg, dsl_deadlist_entry_t *dle)
3100 {
3101 	spa_t *spa = arg;
3102 	uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3103 	if (dle->dle_bpobj.bpo_object != empty_bpobj)
3104 		bpobj_count_refd(&dle->dle_bpobj);
3105 	return (0);
3106 }
3107 
3108 static int
dsl_deadlist_entry_dump(void * arg,dsl_deadlist_entry_t * dle)3109 dsl_deadlist_entry_dump(void *arg, dsl_deadlist_entry_t *dle)
3110 {
3111 	ASSERT(arg == NULL);
3112 	if (dump_opt['d'] >= 5) {
3113 		char buf[128];
3114 		(void) snprintf(buf, sizeof (buf),
3115 		    "mintxg %llu -> obj %llu",
3116 		    (longlong_t)dle->dle_mintxg,
3117 		    (longlong_t)dle->dle_bpobj.bpo_object);
3118 
3119 		dump_full_bpobj(&dle->dle_bpobj, buf, 0);
3120 	} else {
3121 		(void) printf("mintxg %llu -> obj %llu\n",
3122 		    (longlong_t)dle->dle_mintxg,
3123 		    (longlong_t)dle->dle_bpobj.bpo_object);
3124 	}
3125 	return (0);
3126 }
3127 
3128 static void
dump_blkptr_list(dsl_deadlist_t * dl,const char * name)3129 dump_blkptr_list(dsl_deadlist_t *dl, const char *name)
3130 {
3131 	char bytes[32];
3132 	char comp[32];
3133 	char uncomp[32];
3134 	char entries[32];
3135 	spa_t *spa = dmu_objset_spa(dl->dl_os);
3136 	uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3137 
3138 	if (dl->dl_oldfmt) {
3139 		if (dl->dl_bpobj.bpo_object != empty_bpobj)
3140 			bpobj_count_refd(&dl->dl_bpobj);
3141 	} else {
3142 		mos_obj_refd(dl->dl_object);
3143 		dsl_deadlist_iterate(dl, dsl_deadlist_entry_count_refd, spa);
3144 	}
3145 
3146 	/* make sure nicenum has enough space */
3147 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3148 	_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
3149 	_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
3150 	_Static_assert(sizeof (entries) >= NN_NUMBUF_SZ, "entries truncated");
3151 
3152 	if (dump_opt['d'] < 3)
3153 		return;
3154 
3155 	if (dl->dl_oldfmt) {
3156 		dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0);
3157 		return;
3158 	}
3159 
3160 	zdb_nicenum(dl->dl_phys->dl_used, bytes, sizeof (bytes));
3161 	zdb_nicenum(dl->dl_phys->dl_comp, comp, sizeof (comp));
3162 	zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp, sizeof (uncomp));
3163 	zdb_nicenum(avl_numnodes(&dl->dl_tree), entries, sizeof (entries));
3164 	(void) printf("\n    %s: %s (%s/%s comp), %s entries\n",
3165 	    name, bytes, comp, uncomp, entries);
3166 
3167 	if (dump_opt['d'] < 4)
3168 		return;
3169 
3170 	(void) putchar('\n');
3171 
3172 	dsl_deadlist_iterate(dl, dsl_deadlist_entry_dump, NULL);
3173 }
3174 
3175 static int
verify_dd_livelist(objset_t * os)3176 verify_dd_livelist(objset_t *os)
3177 {
3178 	uint64_t ll_used, used, ll_comp, comp, ll_uncomp, uncomp;
3179 	dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3180 	dsl_dir_t  *dd = os->os_dsl_dataset->ds_dir;
3181 
3182 	ASSERT(!dmu_objset_is_snapshot(os));
3183 	if (!dsl_deadlist_is_open(&dd->dd_livelist))
3184 		return (0);
3185 
3186 	/* Iterate through the livelist to check for duplicates */
3187 	dsl_deadlist_iterate(&dd->dd_livelist, sublivelist_verify_lightweight,
3188 	    NULL);
3189 
3190 	dsl_pool_config_enter(dp, FTAG);
3191 	dsl_deadlist_space(&dd->dd_livelist, &ll_used,
3192 	    &ll_comp, &ll_uncomp);
3193 
3194 	dsl_dataset_t *origin_ds;
3195 	ASSERT(dsl_pool_config_held(dp));
3196 	VERIFY0(dsl_dataset_hold_obj(dp,
3197 	    dsl_dir_phys(dd)->dd_origin_obj, FTAG, &origin_ds));
3198 	VERIFY0(dsl_dataset_space_written(origin_ds, os->os_dsl_dataset,
3199 	    &used, &comp, &uncomp));
3200 	dsl_dataset_rele(origin_ds, FTAG);
3201 	dsl_pool_config_exit(dp, FTAG);
3202 	/*
3203 	 *  It's possible that the dataset's uncomp space is larger than the
3204 	 *  livelist's because livelists do not track embedded block pointers
3205 	 */
3206 	if (used != ll_used || comp != ll_comp || uncomp < ll_uncomp) {
3207 		char nice_used[32], nice_comp[32], nice_uncomp[32];
3208 		(void) printf("Discrepancy in space accounting:\n");
3209 		zdb_nicenum(used, nice_used, sizeof (nice_used));
3210 		zdb_nicenum(comp, nice_comp, sizeof (nice_comp));
3211 		zdb_nicenum(uncomp, nice_uncomp, sizeof (nice_uncomp));
3212 		(void) printf("dir: used %s, comp %s, uncomp %s\n",
3213 		    nice_used, nice_comp, nice_uncomp);
3214 		zdb_nicenum(ll_used, nice_used, sizeof (nice_used));
3215 		zdb_nicenum(ll_comp, nice_comp, sizeof (nice_comp));
3216 		zdb_nicenum(ll_uncomp, nice_uncomp, sizeof (nice_uncomp));
3217 		(void) printf("livelist: used %s, comp %s, uncomp %s\n",
3218 		    nice_used, nice_comp, nice_uncomp);
3219 		return (1);
3220 	}
3221 	return (0);
3222 }
3223 
3224 static char *key_material = NULL;
3225 
3226 static boolean_t
zdb_derive_key(dsl_dir_t * dd,uint8_t * key_out)3227 zdb_derive_key(dsl_dir_t *dd, uint8_t *key_out)
3228 {
3229 	uint64_t keyformat, salt, iters;
3230 	int i;
3231 	unsigned char c;
3232 
3233 	VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3234 	    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), sizeof (uint64_t),
3235 	    1, &keyformat));
3236 
3237 	switch (keyformat) {
3238 	case ZFS_KEYFORMAT_HEX:
3239 		for (i = 0; i < WRAPPING_KEY_LEN * 2; i += 2) {
3240 			if (!isxdigit(key_material[i]) ||
3241 			    !isxdigit(key_material[i+1]))
3242 				return (B_FALSE);
3243 			if (sscanf(&key_material[i], "%02hhx", &c) != 1)
3244 				return (B_FALSE);
3245 			key_out[i / 2] = c;
3246 		}
3247 		break;
3248 
3249 	case ZFS_KEYFORMAT_PASSPHRASE:
3250 		VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3251 		    dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT),
3252 		    sizeof (uint64_t), 1, &salt));
3253 		VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3254 		    dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS),
3255 		    sizeof (uint64_t), 1, &iters));
3256 
3257 		if (PKCS5_PBKDF2_HMAC_SHA1(key_material, strlen(key_material),
3258 		    ((uint8_t *)&salt), sizeof (uint64_t), iters,
3259 		    WRAPPING_KEY_LEN, key_out) != 1)
3260 			return (B_FALSE);
3261 
3262 		break;
3263 
3264 	default:
3265 		fatal("no support for key format %u\n",
3266 		    (unsigned int) keyformat);
3267 	}
3268 
3269 	return (B_TRUE);
3270 }
3271 
3272 static char encroot[ZFS_MAX_DATASET_NAME_LEN];
3273 static boolean_t key_loaded = B_FALSE;
3274 
3275 static void
zdb_load_key(objset_t * os)3276 zdb_load_key(objset_t *os)
3277 {
3278 	dsl_pool_t *dp;
3279 	dsl_dir_t *dd, *rdd;
3280 	uint8_t key[WRAPPING_KEY_LEN];
3281 	uint64_t rddobj;
3282 	int err;
3283 
3284 	dp = spa_get_dsl(os->os_spa);
3285 	dd = os->os_dsl_dataset->ds_dir;
3286 
3287 	dsl_pool_config_enter(dp, FTAG);
3288 	VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3289 	    DSL_CRYPTO_KEY_ROOT_DDOBJ, sizeof (uint64_t), 1, &rddobj));
3290 	VERIFY0(dsl_dir_hold_obj(dd->dd_pool, rddobj, NULL, FTAG, &rdd));
3291 	dsl_dir_name(rdd, encroot);
3292 	dsl_dir_rele(rdd, FTAG);
3293 
3294 	if (!zdb_derive_key(dd, key))
3295 		fatal("couldn't derive encryption key");
3296 
3297 	dsl_pool_config_exit(dp, FTAG);
3298 
3299 	ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_UNAVAILABLE);
3300 
3301 	dsl_crypto_params_t *dcp;
3302 	nvlist_t *crypto_args;
3303 
3304 	crypto_args = fnvlist_alloc();
3305 	fnvlist_add_uint8_array(crypto_args, "wkeydata",
3306 	    (uint8_t *)key, WRAPPING_KEY_LEN);
3307 	VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE,
3308 	    NULL, crypto_args, &dcp));
3309 	err = spa_keystore_load_wkey(encroot, dcp, B_FALSE);
3310 
3311 	dsl_crypto_params_free(dcp, (err != 0));
3312 	fnvlist_free(crypto_args);
3313 
3314 	if (err != 0)
3315 		fatal(
3316 		    "couldn't load encryption key for %s: %s",
3317 		    encroot, err == ZFS_ERR_CRYPTO_NOTSUP ?
3318 		    "crypto params not supported" : strerror(err));
3319 
3320 	ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_AVAILABLE);
3321 
3322 	printf("Unlocked encryption root: %s\n", encroot);
3323 	key_loaded = B_TRUE;
3324 }
3325 
3326 static void
zdb_unload_key(void)3327 zdb_unload_key(void)
3328 {
3329 	if (!key_loaded)
3330 		return;
3331 
3332 	VERIFY0(spa_keystore_unload_wkey(encroot));
3333 	key_loaded = B_FALSE;
3334 }
3335 
3336 static avl_tree_t idx_tree;
3337 static avl_tree_t domain_tree;
3338 static boolean_t fuid_table_loaded;
3339 static objset_t *sa_os = NULL;
3340 static sa_attr_type_t *sa_attr_table = NULL;
3341 
3342 static int
open_objset(const char * path,const void * tag,objset_t ** osp)3343 open_objset(const char *path, const void *tag, objset_t **osp)
3344 {
3345 	int err;
3346 	uint64_t sa_attrs = 0;
3347 	uint64_t version = 0;
3348 
3349 	VERIFY3P(sa_os, ==, NULL);
3350 
3351 	/*
3352 	 * We can't own an objset if it's redacted.  Therefore, we do this
3353 	 * dance: hold the objset, then acquire a long hold on its dataset, then
3354 	 * release the pool (which is held as part of holding the objset).
3355 	 */
3356 
3357 	if (dump_opt['K']) {
3358 		/* decryption requested, try to load keys */
3359 		err = dmu_objset_hold(path, tag, osp);
3360 		if (err != 0) {
3361 			(void) fprintf(stderr, "failed to hold dataset "
3362 			    "'%s': %s\n",
3363 			    path, strerror(err));
3364 			return (err);
3365 		}
3366 		dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3367 		dsl_pool_rele(dmu_objset_pool(*osp), tag);
3368 
3369 		/* succeeds or dies */
3370 		zdb_load_key(*osp);
3371 
3372 		/* release it all */
3373 		dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3374 		dsl_dataset_rele(dmu_objset_ds(*osp), tag);
3375 	}
3376 
3377 	int ds_hold_flags = key_loaded ? DS_HOLD_FLAG_DECRYPT : 0;
3378 
3379 	err = dmu_objset_hold_flags(path, ds_hold_flags, tag, osp);
3380 	if (err != 0) {
3381 		(void) fprintf(stderr, "failed to hold dataset '%s': %s\n",
3382 		    path, strerror(err));
3383 		return (err);
3384 	}
3385 	dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3386 	dsl_pool_rele(dmu_objset_pool(*osp), tag);
3387 
3388 	if (dmu_objset_type(*osp) == DMU_OST_ZFS &&
3389 	    (key_loaded || !(*osp)->os_encrypted)) {
3390 		(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR,
3391 		    8, 1, &version);
3392 		if (version >= ZPL_VERSION_SA) {
3393 			(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS,
3394 			    8, 1, &sa_attrs);
3395 		}
3396 		err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END,
3397 		    &sa_attr_table);
3398 		if (err != 0) {
3399 			(void) fprintf(stderr, "sa_setup failed: %s\n",
3400 			    strerror(err));
3401 			dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3402 			dsl_dataset_rele_flags(dmu_objset_ds(*osp),
3403 			    ds_hold_flags, tag);
3404 			*osp = NULL;
3405 		}
3406 	}
3407 	sa_os = *osp;
3408 
3409 	return (err);
3410 }
3411 
3412 static void
close_objset(objset_t * os,const void * tag)3413 close_objset(objset_t *os, const void *tag)
3414 {
3415 	VERIFY3P(os, ==, sa_os);
3416 	if (os->os_sa != NULL)
3417 		sa_tear_down(os);
3418 	dsl_dataset_long_rele(dmu_objset_ds(os), tag);
3419 	dsl_dataset_rele_flags(dmu_objset_ds(os),
3420 	    key_loaded ? DS_HOLD_FLAG_DECRYPT : 0, tag);
3421 	sa_attr_table = NULL;
3422 	sa_os = NULL;
3423 
3424 	zdb_unload_key();
3425 }
3426 
3427 static void
fuid_table_destroy(void)3428 fuid_table_destroy(void)
3429 {
3430 	if (fuid_table_loaded) {
3431 		zfs_fuid_table_destroy(&idx_tree, &domain_tree);
3432 		fuid_table_loaded = B_FALSE;
3433 	}
3434 }
3435 
3436 /*
3437  * Clean up DDT internal state. ddt_lookup() adds entries to ddt_tree, which on
3438  * a live pool are normally cleaned up during ddt_sync(). We can't do that (and
3439  * wouldn't want to anyway), but if we don't clean up the presence of stuff on
3440  * ddt_tree will trip asserts in ddt_table_free(). So, we clean up ourselves.
3441  *
3442  * Note that this is not a particularly efficient way to do this, but
3443  * ddt_remove() is the only public method that can do the work we need, and it
3444  * requires the right locks and etc to do the job. This is only ever called
3445  * during zdb shutdown so efficiency is not especially important.
3446  */
3447 static void
zdb_ddt_cleanup(spa_t * spa)3448 zdb_ddt_cleanup(spa_t *spa)
3449 {
3450 	for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
3451 		ddt_t *ddt = spa->spa_ddt[c];
3452 		if (!ddt)
3453 			continue;
3454 
3455 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3456 		ddt_enter(ddt);
3457 		ddt_entry_t *dde = avl_first(&ddt->ddt_tree), *next;
3458 		while (dde) {
3459 			next = AVL_NEXT(&ddt->ddt_tree, dde);
3460 			dde->dde_io = NULL;
3461 			ddt_remove(ddt, dde);
3462 			dde = next;
3463 		}
3464 		ddt_exit(ddt);
3465 		spa_config_exit(spa, SCL_CONFIG, FTAG);
3466 	}
3467 }
3468 
3469 static void
zdb_exit(int reason)3470 zdb_exit(int reason)
3471 {
3472 	if (spa != NULL)
3473 		zdb_ddt_cleanup(spa);
3474 
3475 	if (os != NULL) {
3476 		close_objset(os, FTAG);
3477 	} else if (spa != NULL) {
3478 		spa_close(spa, FTAG);
3479 	}
3480 
3481 	fuid_table_destroy();
3482 
3483 	if (kernel_init_done)
3484 		kernel_fini();
3485 
3486 	exit(reason);
3487 }
3488 
3489 /*
3490  * print uid or gid information.
3491  * For normal POSIX id just the id is printed in decimal format.
3492  * For CIFS files with FUID the fuid is printed in hex followed by
3493  * the domain-rid string.
3494  */
3495 static void
print_idstr(uint64_t id,const char * id_type)3496 print_idstr(uint64_t id, const char *id_type)
3497 {
3498 	if (FUID_INDEX(id)) {
3499 		const char *domain =
3500 		    zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id));
3501 		(void) printf("\t%s     %llx [%s-%d]\n", id_type,
3502 		    (u_longlong_t)id, domain, (int)FUID_RID(id));
3503 	} else {
3504 		(void) printf("\t%s     %llu\n", id_type, (u_longlong_t)id);
3505 	}
3506 
3507 }
3508 
3509 static void
dump_uidgid(objset_t * os,uint64_t uid,uint64_t gid)3510 dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid)
3511 {
3512 	uint32_t uid_idx, gid_idx;
3513 
3514 	uid_idx = FUID_INDEX(uid);
3515 	gid_idx = FUID_INDEX(gid);
3516 
3517 	/* Load domain table, if not already loaded */
3518 	if (!fuid_table_loaded && (uid_idx || gid_idx)) {
3519 		uint64_t fuid_obj;
3520 
3521 		/* first find the fuid object.  It lives in the master node */
3522 		VERIFY(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES,
3523 		    8, 1, &fuid_obj) == 0);
3524 		zfs_fuid_avl_tree_create(&idx_tree, &domain_tree);
3525 		(void) zfs_fuid_table_load(os, fuid_obj,
3526 		    &idx_tree, &domain_tree);
3527 		fuid_table_loaded = B_TRUE;
3528 	}
3529 
3530 	print_idstr(uid, "uid");
3531 	print_idstr(gid, "gid");
3532 }
3533 
3534 static void
dump_znode_sa_xattr(sa_handle_t * hdl)3535 dump_znode_sa_xattr(sa_handle_t *hdl)
3536 {
3537 	nvlist_t *sa_xattr;
3538 	nvpair_t *elem = NULL;
3539 	int sa_xattr_size = 0;
3540 	int sa_xattr_entries = 0;
3541 	int error;
3542 	char *sa_xattr_packed;
3543 
3544 	error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size);
3545 	if (error || sa_xattr_size == 0)
3546 		return;
3547 
3548 	sa_xattr_packed = malloc(sa_xattr_size);
3549 	if (sa_xattr_packed == NULL)
3550 		return;
3551 
3552 	error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR],
3553 	    sa_xattr_packed, sa_xattr_size);
3554 	if (error) {
3555 		free(sa_xattr_packed);
3556 		return;
3557 	}
3558 
3559 	error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0);
3560 	if (error) {
3561 		free(sa_xattr_packed);
3562 		return;
3563 	}
3564 
3565 	while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL)
3566 		sa_xattr_entries++;
3567 
3568 	(void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3569 	    sa_xattr_size, sa_xattr_entries);
3570 	while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) {
3571 		boolean_t can_print = !dump_opt['P'];
3572 		uchar_t *value;
3573 		uint_t cnt, idx;
3574 
3575 		(void) printf("\t\t%s = ", nvpair_name(elem));
3576 		nvpair_value_byte_array(elem, &value, &cnt);
3577 
3578 		for (idx = 0; idx < cnt; ++idx) {
3579 			if (!isprint(value[idx])) {
3580 				can_print = B_FALSE;
3581 				break;
3582 			}
3583 		}
3584 
3585 		for (idx = 0; idx < cnt; ++idx) {
3586 			if (can_print)
3587 				(void) putchar(value[idx]);
3588 			else
3589 				(void) printf("\\%3.3o", value[idx]);
3590 		}
3591 		(void) putchar('\n');
3592 	}
3593 
3594 	nvlist_free(sa_xattr);
3595 	free(sa_xattr_packed);
3596 }
3597 
3598 static void
dump_znode_symlink(sa_handle_t * hdl)3599 dump_znode_symlink(sa_handle_t *hdl)
3600 {
3601 	int sa_symlink_size = 0;
3602 	char linktarget[MAXPATHLEN];
3603 	int error;
3604 
3605 	error = sa_size(hdl, sa_attr_table[ZPL_SYMLINK], &sa_symlink_size);
3606 	if (error || sa_symlink_size == 0) {
3607 		return;
3608 	}
3609 	if (sa_symlink_size >= sizeof (linktarget)) {
3610 		(void) printf("symlink size %d is too large\n",
3611 		    sa_symlink_size);
3612 		return;
3613 	}
3614 	linktarget[sa_symlink_size] = '\0';
3615 	if (sa_lookup(hdl, sa_attr_table[ZPL_SYMLINK],
3616 	    &linktarget, sa_symlink_size) == 0)
3617 		(void) printf("\ttarget	%s\n", linktarget);
3618 }
3619 
3620 static void
dump_znode(objset_t * os,uint64_t object,void * data,size_t size)3621 dump_znode(objset_t *os, uint64_t object, void *data, size_t size)
3622 {
3623 	(void) data, (void) size;
3624 	char path[MAXPATHLEN * 2];	/* allow for xattr and failure prefix */
3625 	sa_handle_t *hdl;
3626 	uint64_t xattr, rdev, gen;
3627 	uint64_t uid, gid, mode, fsize, parent, links;
3628 	uint64_t pflags;
3629 	uint64_t acctm[2], modtm[2], chgtm[2], crtm[2];
3630 	time_t z_crtime, z_atime, z_mtime, z_ctime;
3631 	sa_bulk_attr_t bulk[12];
3632 	int idx = 0;
3633 	int error;
3634 
3635 	VERIFY3P(os, ==, sa_os);
3636 	if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) {
3637 		(void) printf("Failed to get handle for SA znode\n");
3638 		return;
3639 	}
3640 
3641 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8);
3642 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8);
3643 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL,
3644 	    &links, 8);
3645 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8);
3646 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL,
3647 	    &mode, 8);
3648 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT],
3649 	    NULL, &parent, 8);
3650 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL,
3651 	    &fsize, 8);
3652 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL,
3653 	    acctm, 16);
3654 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL,
3655 	    modtm, 16);
3656 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL,
3657 	    crtm, 16);
3658 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL,
3659 	    chgtm, 16);
3660 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL,
3661 	    &pflags, 8);
3662 
3663 	if (sa_bulk_lookup(hdl, bulk, idx)) {
3664 		(void) sa_handle_destroy(hdl);
3665 		return;
3666 	}
3667 
3668 	z_crtime = (time_t)crtm[0];
3669 	z_atime = (time_t)acctm[0];
3670 	z_mtime = (time_t)modtm[0];
3671 	z_ctime = (time_t)chgtm[0];
3672 
3673 	if (dump_opt['d'] > 4) {
3674 		error = zfs_obj_to_path(os, object, path, sizeof (path));
3675 		if (error == ESTALE) {
3676 			(void) snprintf(path, sizeof (path), "on delete queue");
3677 		} else if (error != 0) {
3678 			leaked_objects++;
3679 			(void) snprintf(path, sizeof (path),
3680 			    "path not found, possibly leaked");
3681 		}
3682 		(void) printf("\tpath	%s\n", path);
3683 	}
3684 
3685 	if (S_ISLNK(mode))
3686 		dump_znode_symlink(hdl);
3687 	dump_uidgid(os, uid, gid);
3688 	(void) printf("\tatime	%s", ctime(&z_atime));
3689 	(void) printf("\tmtime	%s", ctime(&z_mtime));
3690 	(void) printf("\tctime	%s", ctime(&z_ctime));
3691 	(void) printf("\tcrtime	%s", ctime(&z_crtime));
3692 	(void) printf("\tgen	%llu\n", (u_longlong_t)gen);
3693 	(void) printf("\tmode	%llo\n", (u_longlong_t)mode);
3694 	(void) printf("\tsize	%llu\n", (u_longlong_t)fsize);
3695 	(void) printf("\tparent	%llu\n", (u_longlong_t)parent);
3696 	(void) printf("\tlinks	%llu\n", (u_longlong_t)links);
3697 	(void) printf("\tpflags	%llx\n", (u_longlong_t)pflags);
3698 	if (dmu_objset_projectquota_enabled(os) && (pflags & ZFS_PROJID)) {
3699 		uint64_t projid;
3700 
3701 		if (sa_lookup(hdl, sa_attr_table[ZPL_PROJID], &projid,
3702 		    sizeof (uint64_t)) == 0)
3703 			(void) printf("\tprojid	%llu\n", (u_longlong_t)projid);
3704 	}
3705 	if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr,
3706 	    sizeof (uint64_t)) == 0)
3707 		(void) printf("\txattr	%llu\n", (u_longlong_t)xattr);
3708 	if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev,
3709 	    sizeof (uint64_t)) == 0)
3710 		(void) printf("\trdev	0x%016llx\n", (u_longlong_t)rdev);
3711 	dump_znode_sa_xattr(hdl);
3712 	sa_handle_destroy(hdl);
3713 }
3714 
3715 static void
dump_acl(objset_t * os,uint64_t object,void * data,size_t size)3716 dump_acl(objset_t *os, uint64_t object, void *data, size_t size)
3717 {
3718 	(void) os, (void) object, (void) data, (void) size;
3719 }
3720 
3721 static void
dump_dmu_objset(objset_t * os,uint64_t object,void * data,size_t size)3722 dump_dmu_objset(objset_t *os, uint64_t object, void *data, size_t size)
3723 {
3724 	(void) os, (void) object, (void) data, (void) size;
3725 }
3726 
3727 static object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = {
3728 	dump_none,		/* unallocated			*/
3729 	dump_zap,		/* object directory		*/
3730 	dump_uint64,		/* object array			*/
3731 	dump_none,		/* packed nvlist		*/
3732 	dump_packed_nvlist,	/* packed nvlist size		*/
3733 	dump_none,		/* bpobj			*/
3734 	dump_bpobj,		/* bpobj header			*/
3735 	dump_none,		/* SPA space map header		*/
3736 	dump_none,		/* SPA space map		*/
3737 	dump_none,		/* ZIL intent log		*/
3738 	dump_dnode,		/* DMU dnode			*/
3739 	dump_dmu_objset,	/* DMU objset			*/
3740 	dump_dsl_dir,		/* DSL directory		*/
3741 	dump_zap,		/* DSL directory child map	*/
3742 	dump_zap,		/* DSL dataset snap map		*/
3743 	dump_zap,		/* DSL props			*/
3744 	dump_dsl_dataset,	/* DSL dataset			*/
3745 	dump_znode,		/* ZFS znode			*/
3746 	dump_acl,		/* ZFS V0 ACL			*/
3747 	dump_uint8,		/* ZFS plain file		*/
3748 	dump_zpldir,		/* ZFS directory		*/
3749 	dump_zap,		/* ZFS master node		*/
3750 	dump_zap,		/* ZFS delete queue		*/
3751 	dump_uint8,		/* zvol object			*/
3752 	dump_zap,		/* zvol prop			*/
3753 	dump_uint8,		/* other uint8[]		*/
3754 	dump_uint64,		/* other uint64[]		*/
3755 	dump_zap,		/* other ZAP			*/
3756 	dump_zap,		/* persistent error log		*/
3757 	dump_uint8,		/* SPA history			*/
3758 	dump_history_offsets,	/* SPA history offsets		*/
3759 	dump_zap,		/* Pool properties		*/
3760 	dump_zap,		/* DSL permissions		*/
3761 	dump_acl,		/* ZFS ACL			*/
3762 	dump_uint8,		/* ZFS SYSACL			*/
3763 	dump_none,		/* FUID nvlist			*/
3764 	dump_packed_nvlist,	/* FUID nvlist size		*/
3765 	dump_zap,		/* DSL dataset next clones	*/
3766 	dump_zap,		/* DSL scrub queue		*/
3767 	dump_zap,		/* ZFS user/group/project used	*/
3768 	dump_zap,		/* ZFS user/group/project quota	*/
3769 	dump_zap,		/* snapshot refcount tags	*/
3770 	dump_ddt_zap,		/* DDT ZAP object		*/
3771 	dump_zap,		/* DDT statistics		*/
3772 	dump_znode,		/* SA object			*/
3773 	dump_zap,		/* SA Master Node		*/
3774 	dump_sa_attrs,		/* SA attribute registration	*/
3775 	dump_sa_layouts,	/* SA attribute layouts		*/
3776 	dump_zap,		/* DSL scrub translations	*/
3777 	dump_none,		/* fake dedup BP		*/
3778 	dump_zap,		/* deadlist			*/
3779 	dump_none,		/* deadlist hdr			*/
3780 	dump_zap,		/* dsl clones			*/
3781 	dump_bpobj_subobjs,	/* bpobj subobjs		*/
3782 	dump_unknown,		/* Unknown type, must be last	*/
3783 };
3784 
3785 static boolean_t
match_object_type(dmu_object_type_t obj_type,uint64_t flags)3786 match_object_type(dmu_object_type_t obj_type, uint64_t flags)
3787 {
3788 	boolean_t match = B_TRUE;
3789 
3790 	switch (obj_type) {
3791 	case DMU_OT_DIRECTORY_CONTENTS:
3792 		if (!(flags & ZOR_FLAG_DIRECTORY))
3793 			match = B_FALSE;
3794 		break;
3795 	case DMU_OT_PLAIN_FILE_CONTENTS:
3796 		if (!(flags & ZOR_FLAG_PLAIN_FILE))
3797 			match = B_FALSE;
3798 		break;
3799 	case DMU_OT_SPACE_MAP:
3800 		if (!(flags & ZOR_FLAG_SPACE_MAP))
3801 			match = B_FALSE;
3802 		break;
3803 	default:
3804 		if (strcmp(zdb_ot_name(obj_type), "zap") == 0) {
3805 			if (!(flags & ZOR_FLAG_ZAP))
3806 				match = B_FALSE;
3807 			break;
3808 		}
3809 
3810 		/*
3811 		 * If all bits except some of the supported flags are
3812 		 * set, the user combined the all-types flag (A) with
3813 		 * a negated flag to exclude some types (e.g. A-f to
3814 		 * show all object types except plain files).
3815 		 */
3816 		if ((flags | ZOR_SUPPORTED_FLAGS) != ZOR_FLAG_ALL_TYPES)
3817 			match = B_FALSE;
3818 
3819 		break;
3820 	}
3821 
3822 	return (match);
3823 }
3824 
3825 static void
dump_object(objset_t * os,uint64_t object,int verbosity,boolean_t * print_header,uint64_t * dnode_slots_used,uint64_t flags)3826 dump_object(objset_t *os, uint64_t object, int verbosity,
3827     boolean_t *print_header, uint64_t *dnode_slots_used, uint64_t flags)
3828 {
3829 	dmu_buf_t *db = NULL;
3830 	dmu_object_info_t doi;
3831 	dnode_t *dn;
3832 	boolean_t dnode_held = B_FALSE;
3833 	void *bonus = NULL;
3834 	size_t bsize = 0;
3835 	char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
3836 	char bonus_size[32];
3837 	char aux[50];
3838 	int error;
3839 
3840 	/* make sure nicenum has enough space */
3841 	_Static_assert(sizeof (iblk) >= NN_NUMBUF_SZ, "iblk truncated");
3842 	_Static_assert(sizeof (dblk) >= NN_NUMBUF_SZ, "dblk truncated");
3843 	_Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, "lsize truncated");
3844 	_Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, "asize truncated");
3845 	_Static_assert(sizeof (bonus_size) >= NN_NUMBUF_SZ,
3846 	    "bonus_size truncated");
3847 
3848 	if (*print_header) {
3849 		(void) printf("\n%10s  %3s  %5s  %5s  %5s  %6s  %5s  %6s  %s\n",
3850 		    "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3851 		    "lsize", "%full", "type");
3852 		*print_header = 0;
3853 	}
3854 
3855 	if (object == 0) {
3856 		dn = DMU_META_DNODE(os);
3857 		dmu_object_info_from_dnode(dn, &doi);
3858 	} else {
3859 		/*
3860 		 * Encrypted datasets will have sensitive bonus buffers
3861 		 * encrypted. Therefore we cannot hold the bonus buffer and
3862 		 * must hold the dnode itself instead.
3863 		 */
3864 		error = dmu_object_info(os, object, &doi);
3865 		if (error)
3866 			fatal("dmu_object_info() failed, errno %u", error);
3867 
3868 		if (!key_loaded && os->os_encrypted &&
3869 		    DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) {
3870 			error = dnode_hold(os, object, FTAG, &dn);
3871 			if (error)
3872 				fatal("dnode_hold() failed, errno %u", error);
3873 			dnode_held = B_TRUE;
3874 		} else {
3875 			error = dmu_bonus_hold(os, object, FTAG, &db);
3876 			if (error)
3877 				fatal("dmu_bonus_hold(%llu) failed, errno %u",
3878 				    object, error);
3879 			bonus = db->db_data;
3880 			bsize = db->db_size;
3881 			dn = DB_DNODE((dmu_buf_impl_t *)db);
3882 		}
3883 	}
3884 
3885 	/*
3886 	 * Default to showing all object types if no flags were specified.
3887 	 */
3888 	if (flags != 0 && flags != ZOR_FLAG_ALL_TYPES &&
3889 	    !match_object_type(doi.doi_type, flags))
3890 		goto out;
3891 
3892 	if (dnode_slots_used)
3893 		*dnode_slots_used = doi.doi_dnodesize / DNODE_MIN_SIZE;
3894 
3895 	zdb_nicenum(doi.doi_metadata_block_size, iblk, sizeof (iblk));
3896 	zdb_nicenum(doi.doi_data_block_size, dblk, sizeof (dblk));
3897 	zdb_nicenum(doi.doi_max_offset, lsize, sizeof (lsize));
3898 	zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize, sizeof (asize));
3899 	zdb_nicenum(doi.doi_bonus_size, bonus_size, sizeof (bonus_size));
3900 	zdb_nicenum(doi.doi_dnodesize, dnsize, sizeof (dnsize));
3901 	(void) snprintf(fill, sizeof (fill), "%6.2f", 100.0 *
3902 	    doi.doi_fill_count * doi.doi_data_block_size / (object == 0 ?
3903 	    DNODES_PER_BLOCK : 1) / doi.doi_max_offset);
3904 
3905 	aux[0] = '\0';
3906 
3907 	if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) {
3908 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3909 		    " (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum));
3910 	}
3911 
3912 	if (doi.doi_compress == ZIO_COMPRESS_INHERIT &&
3913 	    ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) {
3914 		const char *compname = NULL;
3915 		if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION,
3916 		    ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel),
3917 		    &compname) == 0) {
3918 			(void) snprintf(aux + strlen(aux),
3919 			    sizeof (aux) - strlen(aux), " (Z=inherit=%s)",
3920 			    compname);
3921 		} else {
3922 			(void) snprintf(aux + strlen(aux),
3923 			    sizeof (aux) - strlen(aux),
3924 			    " (Z=inherit=%s-unknown)",
3925 			    ZDB_COMPRESS_NAME(os->os_compress));
3926 		}
3927 	} else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) {
3928 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3929 		    " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress));
3930 	} else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
3931 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3932 		    " (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress));
3933 	}
3934 
3935 	(void) printf("%10lld  %3u  %5s  %5s  %5s  %6s  %5s  %6s  %s%s\n",
3936 	    (u_longlong_t)object, doi.doi_indirection, iblk, dblk,
3937 	    asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux);
3938 
3939 	if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
3940 		(void) printf("%10s  %3s  %5s  %5s  %5s  %5s  %5s  %6s  %s\n",
3941 		    "", "", "", "", "", "", bonus_size, "bonus",
3942 		    zdb_ot_name(doi.doi_bonus_type));
3943 	}
3944 
3945 	if (verbosity >= 4) {
3946 		(void) printf("\tdnode flags: %s%s%s%s\n",
3947 		    (dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ?
3948 		    "USED_BYTES " : "",
3949 		    (dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ?
3950 		    "USERUSED_ACCOUNTED " : "",
3951 		    (dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ?
3952 		    "USEROBJUSED_ACCOUNTED " : "",
3953 		    (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ?
3954 		    "SPILL_BLKPTR" : "");
3955 		(void) printf("\tdnode maxblkid: %llu\n",
3956 		    (longlong_t)dn->dn_phys->dn_maxblkid);
3957 
3958 		if (!dnode_held) {
3959 			object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os,
3960 			    object, bonus, bsize);
3961 		} else {
3962 			(void) printf("\t\t(bonus encrypted)\n");
3963 		}
3964 
3965 		if (key_loaded ||
3966 		    (!os->os_encrypted || !DMU_OT_IS_ENCRYPTED(doi.doi_type))) {
3967 			object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object,
3968 			    NULL, 0);
3969 		} else {
3970 			(void) printf("\t\t(object encrypted)\n");
3971 		}
3972 
3973 		*print_header = B_TRUE;
3974 	}
3975 
3976 	if (verbosity >= 5) {
3977 		if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
3978 			char blkbuf[BP_SPRINTF_LEN];
3979 			snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
3980 			    DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE);
3981 			(void) printf("\nSpill block: %s\n", blkbuf);
3982 		}
3983 		dump_indirect(dn);
3984 	}
3985 
3986 	if (verbosity >= 5) {
3987 		/*
3988 		 * Report the list of segments that comprise the object.
3989 		 */
3990 		uint64_t start = 0;
3991 		uint64_t end;
3992 		uint64_t blkfill = 1;
3993 		int minlvl = 1;
3994 
3995 		if (dn->dn_type == DMU_OT_DNODE) {
3996 			minlvl = 0;
3997 			blkfill = DNODES_PER_BLOCK;
3998 		}
3999 
4000 		for (;;) {
4001 			char segsize[32];
4002 			/* make sure nicenum has enough space */
4003 			_Static_assert(sizeof (segsize) >= NN_NUMBUF_SZ,
4004 			    "segsize truncated");
4005 			error = dnode_next_offset(dn,
4006 			    0, &start, minlvl, blkfill, 0);
4007 			if (error)
4008 				break;
4009 			end = start;
4010 			error = dnode_next_offset(dn,
4011 			    DNODE_FIND_HOLE, &end, minlvl, blkfill, 0);
4012 			zdb_nicenum(end - start, segsize, sizeof (segsize));
4013 			(void) printf("\t\tsegment [%016llx, %016llx)"
4014 			    " size %5s\n", (u_longlong_t)start,
4015 			    (u_longlong_t)end, segsize);
4016 			if (error)
4017 				break;
4018 			start = end;
4019 		}
4020 	}
4021 
4022 out:
4023 	if (db != NULL)
4024 		dmu_buf_rele(db, FTAG);
4025 	if (dnode_held)
4026 		dnode_rele(dn, FTAG);
4027 }
4028 
4029 static void
count_dir_mos_objects(dsl_dir_t * dd)4030 count_dir_mos_objects(dsl_dir_t *dd)
4031 {
4032 	mos_obj_refd(dd->dd_object);
4033 	mos_obj_refd(dsl_dir_phys(dd)->dd_child_dir_zapobj);
4034 	mos_obj_refd(dsl_dir_phys(dd)->dd_deleg_zapobj);
4035 	mos_obj_refd(dsl_dir_phys(dd)->dd_props_zapobj);
4036 	mos_obj_refd(dsl_dir_phys(dd)->dd_clones);
4037 
4038 	/*
4039 	 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
4040 	 * Ignore the references after the first one.
4041 	 */
4042 	mos_obj_refd_multiple(dd->dd_crypto_obj);
4043 }
4044 
4045 static void
count_ds_mos_objects(dsl_dataset_t * ds)4046 count_ds_mos_objects(dsl_dataset_t *ds)
4047 {
4048 	mos_obj_refd(ds->ds_object);
4049 	mos_obj_refd(dsl_dataset_phys(ds)->ds_next_clones_obj);
4050 	mos_obj_refd(dsl_dataset_phys(ds)->ds_props_obj);
4051 	mos_obj_refd(dsl_dataset_phys(ds)->ds_userrefs_obj);
4052 	mos_obj_refd(dsl_dataset_phys(ds)->ds_snapnames_zapobj);
4053 	mos_obj_refd(ds->ds_bookmarks_obj);
4054 
4055 	if (!dsl_dataset_is_snapshot(ds)) {
4056 		count_dir_mos_objects(ds->ds_dir);
4057 	}
4058 }
4059 
4060 static const char *const objset_types[DMU_OST_NUMTYPES] = {
4061 	"NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
4062 
4063 /*
4064  * Parse a string denoting a range of object IDs of the form
4065  * <start>[:<end>[:flags]], and store the results in zor.
4066  * Return 0 on success. On error, return 1 and update the msg
4067  * pointer to point to a descriptive error message.
4068  */
4069 static int
parse_object_range(char * range,zopt_object_range_t * zor,const char ** msg)4070 parse_object_range(char *range, zopt_object_range_t *zor, const char **msg)
4071 {
4072 	uint64_t flags = 0;
4073 	char *p, *s, *dup, *flagstr, *tmp = NULL;
4074 	size_t len;
4075 	int i;
4076 	int rc = 0;
4077 
4078 	if (strchr(range, ':') == NULL) {
4079 		zor->zor_obj_start = strtoull(range, &p, 0);
4080 		if (*p != '\0') {
4081 			*msg = "Invalid characters in object ID";
4082 			rc = 1;
4083 		}
4084 		zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4085 		zor->zor_obj_end = zor->zor_obj_start;
4086 		return (rc);
4087 	}
4088 
4089 	if (strchr(range, ':') == range) {
4090 		*msg = "Invalid leading colon";
4091 		rc = 1;
4092 		return (rc);
4093 	}
4094 
4095 	len = strlen(range);
4096 	if (range[len - 1] == ':') {
4097 		*msg = "Invalid trailing colon";
4098 		rc = 1;
4099 		return (rc);
4100 	}
4101 
4102 	dup = strdup(range);
4103 	s = strtok_r(dup, ":", &tmp);
4104 	zor->zor_obj_start = strtoull(s, &p, 0);
4105 
4106 	if (*p != '\0') {
4107 		*msg = "Invalid characters in start object ID";
4108 		rc = 1;
4109 		goto out;
4110 	}
4111 
4112 	s = strtok_r(NULL, ":", &tmp);
4113 	zor->zor_obj_end = strtoull(s, &p, 0);
4114 
4115 	if (*p != '\0') {
4116 		*msg = "Invalid characters in end object ID";
4117 		rc = 1;
4118 		goto out;
4119 	}
4120 
4121 	if (zor->zor_obj_start > zor->zor_obj_end) {
4122 		*msg = "Start object ID may not exceed end object ID";
4123 		rc = 1;
4124 		goto out;
4125 	}
4126 
4127 	s = strtok_r(NULL, ":", &tmp);
4128 	if (s == NULL) {
4129 		zor->zor_flags = ZOR_FLAG_ALL_TYPES;
4130 		goto out;
4131 	} else if (strtok_r(NULL, ":", &tmp) != NULL) {
4132 		*msg = "Invalid colon-delimited field after flags";
4133 		rc = 1;
4134 		goto out;
4135 	}
4136 
4137 	flagstr = s;
4138 	for (i = 0; flagstr[i]; i++) {
4139 		int bit;
4140 		boolean_t negation = (flagstr[i] == '-');
4141 
4142 		if (negation) {
4143 			i++;
4144 			if (flagstr[i] == '\0') {
4145 				*msg = "Invalid trailing negation operator";
4146 				rc = 1;
4147 				goto out;
4148 			}
4149 		}
4150 		bit = flagbits[(uchar_t)flagstr[i]];
4151 		if (bit == 0) {
4152 			*msg = "Invalid flag";
4153 			rc = 1;
4154 			goto out;
4155 		}
4156 		if (negation)
4157 			flags &= ~bit;
4158 		else
4159 			flags |= bit;
4160 	}
4161 	zor->zor_flags = flags;
4162 
4163 	zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4164 	zor->zor_obj_end = ZDB_MAP_OBJECT_ID(zor->zor_obj_end);
4165 
4166 out:
4167 	free(dup);
4168 	return (rc);
4169 }
4170 
4171 static void
dump_objset(objset_t * os)4172 dump_objset(objset_t *os)
4173 {
4174 	dmu_objset_stats_t dds = { 0 };
4175 	uint64_t object, object_count;
4176 	uint64_t refdbytes, usedobjs, scratch;
4177 	char numbuf[32];
4178 	char blkbuf[BP_SPRINTF_LEN + 20];
4179 	char osname[ZFS_MAX_DATASET_NAME_LEN];
4180 	const char *type = "UNKNOWN";
4181 	int verbosity = dump_opt['d'];
4182 	boolean_t print_header;
4183 	unsigned i;
4184 	int error;
4185 	uint64_t total_slots_used = 0;
4186 	uint64_t max_slot_used = 0;
4187 	uint64_t dnode_slots;
4188 	uint64_t obj_start;
4189 	uint64_t obj_end;
4190 	uint64_t flags;
4191 
4192 	/* make sure nicenum has enough space */
4193 	_Static_assert(sizeof (numbuf) >= NN_NUMBUF_SZ, "numbuf truncated");
4194 
4195 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
4196 	dmu_objset_fast_stat(os, &dds);
4197 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
4198 
4199 	print_header = B_TRUE;
4200 
4201 	if (dds.dds_type < DMU_OST_NUMTYPES)
4202 		type = objset_types[dds.dds_type];
4203 
4204 	if (dds.dds_type == DMU_OST_META) {
4205 		dds.dds_creation_txg = TXG_INITIAL;
4206 		usedobjs = BP_GET_FILL(os->os_rootbp);
4207 		refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)->
4208 		    dd_used_bytes;
4209 	} else {
4210 		dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
4211 	}
4212 
4213 	ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
4214 
4215 	zdb_nicenum(refdbytes, numbuf, sizeof (numbuf));
4216 
4217 	if (verbosity >= 4) {
4218 		(void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp ");
4219 		(void) snprintf_blkptr(blkbuf + strlen(blkbuf),
4220 		    sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp);
4221 	} else {
4222 		blkbuf[0] = '\0';
4223 	}
4224 
4225 	dmu_objset_name(os, osname);
4226 
4227 	(void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
4228 	    "%s, %llu objects%s%s\n",
4229 	    osname, type, (u_longlong_t)dmu_objset_id(os),
4230 	    (u_longlong_t)dds.dds_creation_txg,
4231 	    numbuf, (u_longlong_t)usedobjs, blkbuf,
4232 	    (dds.dds_inconsistent) ? " (inconsistent)" : "");
4233 
4234 	for (i = 0; i < zopt_object_args; i++) {
4235 		obj_start = zopt_object_ranges[i].zor_obj_start;
4236 		obj_end = zopt_object_ranges[i].zor_obj_end;
4237 		flags = zopt_object_ranges[i].zor_flags;
4238 
4239 		object = obj_start;
4240 		if (object == 0 || obj_start == obj_end)
4241 			dump_object(os, object, verbosity, &print_header, NULL,
4242 			    flags);
4243 		else
4244 			object--;
4245 
4246 		while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) &&
4247 		    object <= obj_end) {
4248 			dump_object(os, object, verbosity, &print_header, NULL,
4249 			    flags);
4250 		}
4251 	}
4252 
4253 	if (zopt_object_args > 0) {
4254 		(void) printf("\n");
4255 		return;
4256 	}
4257 
4258 	if (dump_opt['i'] != 0 || verbosity >= 2)
4259 		dump_intent_log(dmu_objset_zil(os));
4260 
4261 	if (dmu_objset_ds(os) != NULL) {
4262 		dsl_dataset_t *ds = dmu_objset_ds(os);
4263 		dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
4264 		if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
4265 		    !dmu_objset_is_snapshot(os)) {
4266 			dump_blkptr_list(&ds->ds_dir->dd_livelist, "Livelist");
4267 			if (verify_dd_livelist(os) != 0)
4268 				fatal("livelist is incorrect");
4269 		}
4270 
4271 		if (dsl_dataset_remap_deadlist_exists(ds)) {
4272 			(void) printf("ds_remap_deadlist:\n");
4273 			dump_blkptr_list(&ds->ds_remap_deadlist, "Deadlist");
4274 		}
4275 		count_ds_mos_objects(ds);
4276 	}
4277 
4278 	if (dmu_objset_ds(os) != NULL)
4279 		dump_bookmarks(os, verbosity);
4280 
4281 	if (verbosity < 2)
4282 		return;
4283 
4284 	if (BP_IS_HOLE(os->os_rootbp))
4285 		return;
4286 
4287 	dump_object(os, 0, verbosity, &print_header, NULL, 0);
4288 	object_count = 0;
4289 	if (DMU_USERUSED_DNODE(os) != NULL &&
4290 	    DMU_USERUSED_DNODE(os)->dn_type != 0) {
4291 		dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header,
4292 		    NULL, 0);
4293 		dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header,
4294 		    NULL, 0);
4295 	}
4296 
4297 	if (DMU_PROJECTUSED_DNODE(os) != NULL &&
4298 	    DMU_PROJECTUSED_DNODE(os)->dn_type != 0)
4299 		dump_object(os, DMU_PROJECTUSED_OBJECT, verbosity,
4300 		    &print_header, NULL, 0);
4301 
4302 	object = 0;
4303 	while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
4304 		dump_object(os, object, verbosity, &print_header, &dnode_slots,
4305 		    0);
4306 		object_count++;
4307 		total_slots_used += dnode_slots;
4308 		max_slot_used = object + dnode_slots - 1;
4309 	}
4310 
4311 	(void) printf("\n");
4312 
4313 	(void) printf("    Dnode slots:\n");
4314 	(void) printf("\tTotal used:    %10llu\n",
4315 	    (u_longlong_t)total_slots_used);
4316 	(void) printf("\tMax used:      %10llu\n",
4317 	    (u_longlong_t)max_slot_used);
4318 	(void) printf("\tPercent empty: %10lf\n",
4319 	    (double)(max_slot_used - total_slots_used)*100 /
4320 	    (double)max_slot_used);
4321 	(void) printf("\n");
4322 
4323 	if (error != ESRCH) {
4324 		(void) fprintf(stderr, "dmu_object_next() = %d\n", error);
4325 		abort();
4326 	}
4327 
4328 	ASSERT3U(object_count, ==, usedobjs);
4329 
4330 	if (leaked_objects != 0) {
4331 		(void) printf("%d potentially leaked objects detected\n",
4332 		    leaked_objects);
4333 		leaked_objects = 0;
4334 	}
4335 }
4336 
4337 static void
dump_uberblock(uberblock_t * ub,const char * header,const char * footer)4338 dump_uberblock(uberblock_t *ub, const char *header, const char *footer)
4339 {
4340 	time_t timestamp = ub->ub_timestamp;
4341 
4342 	(void) printf("%s", header ? header : "");
4343 	(void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic);
4344 	(void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version);
4345 	(void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg);
4346 	(void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum);
4347 	(void) printf("\ttimestamp = %llu UTC = %s",
4348 	    (u_longlong_t)ub->ub_timestamp, ctime(&timestamp));
4349 
4350 	char blkbuf[BP_SPRINTF_LEN];
4351 	snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4352 	(void) printf("\tbp = %s\n", blkbuf);
4353 
4354 	(void) printf("\tmmp_magic = %016llx\n",
4355 	    (u_longlong_t)ub->ub_mmp_magic);
4356 	if (MMP_VALID(ub)) {
4357 		(void) printf("\tmmp_delay = %0llu\n",
4358 		    (u_longlong_t)ub->ub_mmp_delay);
4359 		if (MMP_SEQ_VALID(ub))
4360 			(void) printf("\tmmp_seq = %u\n",
4361 			    (unsigned int) MMP_SEQ(ub));
4362 		if (MMP_FAIL_INT_VALID(ub))
4363 			(void) printf("\tmmp_fail = %u\n",
4364 			    (unsigned int) MMP_FAIL_INT(ub));
4365 		if (MMP_INTERVAL_VALID(ub))
4366 			(void) printf("\tmmp_write = %u\n",
4367 			    (unsigned int) MMP_INTERVAL(ub));
4368 		/* After MMP_* to make summarize_uberblock_mmp cleaner */
4369 		(void) printf("\tmmp_valid = %x\n",
4370 		    (unsigned int) ub->ub_mmp_config & 0xFF);
4371 	}
4372 
4373 	if (dump_opt['u'] >= 4) {
4374 		char blkbuf[BP_SPRINTF_LEN];
4375 		snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4376 		(void) printf("\trootbp = %s\n", blkbuf);
4377 	}
4378 	(void) printf("\tcheckpoint_txg = %llu\n",
4379 	    (u_longlong_t)ub->ub_checkpoint_txg);
4380 
4381 	(void) printf("\traidz_reflow state=%u off=%llu\n",
4382 	    (int)RRSS_GET_STATE(ub),
4383 	    (u_longlong_t)RRSS_GET_OFFSET(ub));
4384 
4385 	(void) printf("%s", footer ? footer : "");
4386 }
4387 
4388 static void
dump_config(spa_t * spa)4389 dump_config(spa_t *spa)
4390 {
4391 	dmu_buf_t *db;
4392 	size_t nvsize = 0;
4393 	int error = 0;
4394 
4395 
4396 	error = dmu_bonus_hold(spa->spa_meta_objset,
4397 	    spa->spa_config_object, FTAG, &db);
4398 
4399 	if (error == 0) {
4400 		nvsize = *(uint64_t *)db->db_data;
4401 		dmu_buf_rele(db, FTAG);
4402 
4403 		(void) printf("\nMOS Configuration:\n");
4404 		dump_packed_nvlist(spa->spa_meta_objset,
4405 		    spa->spa_config_object, (void *)&nvsize, 1);
4406 	} else {
4407 		(void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d",
4408 		    (u_longlong_t)spa->spa_config_object, error);
4409 	}
4410 }
4411 
4412 static void
dump_cachefile(const char * cachefile)4413 dump_cachefile(const char *cachefile)
4414 {
4415 	int fd;
4416 	struct stat64 statbuf;
4417 	char *buf;
4418 	nvlist_t *config;
4419 
4420 	if ((fd = open64(cachefile, O_RDONLY)) < 0) {
4421 		(void) printf("cannot open '%s': %s\n", cachefile,
4422 		    strerror(errno));
4423 		zdb_exit(1);
4424 	}
4425 
4426 	if (fstat64(fd, &statbuf) != 0) {
4427 		(void) printf("failed to stat '%s': %s\n", cachefile,
4428 		    strerror(errno));
4429 		zdb_exit(1);
4430 	}
4431 
4432 	if ((buf = malloc(statbuf.st_size)) == NULL) {
4433 		(void) fprintf(stderr, "failed to allocate %llu bytes\n",
4434 		    (u_longlong_t)statbuf.st_size);
4435 		zdb_exit(1);
4436 	}
4437 
4438 	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
4439 		(void) fprintf(stderr, "failed to read %llu bytes\n",
4440 		    (u_longlong_t)statbuf.st_size);
4441 		zdb_exit(1);
4442 	}
4443 
4444 	(void) close(fd);
4445 
4446 	if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) {
4447 		(void) fprintf(stderr, "failed to unpack nvlist\n");
4448 		zdb_exit(1);
4449 	}
4450 
4451 	free(buf);
4452 
4453 	dump_nvlist(config, 0);
4454 
4455 	nvlist_free(config);
4456 }
4457 
4458 /*
4459  * ZFS label nvlist stats
4460  */
4461 typedef struct zdb_nvl_stats {
4462 	int		zns_list_count;
4463 	int		zns_leaf_count;
4464 	size_t		zns_leaf_largest;
4465 	size_t		zns_leaf_total;
4466 	nvlist_t	*zns_string;
4467 	nvlist_t	*zns_uint64;
4468 	nvlist_t	*zns_boolean;
4469 } zdb_nvl_stats_t;
4470 
4471 static void
collect_nvlist_stats(nvlist_t * nvl,zdb_nvl_stats_t * stats)4472 collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats)
4473 {
4474 	nvlist_t *list, **array;
4475 	nvpair_t *nvp = NULL;
4476 	const char *name;
4477 	uint_t i, items;
4478 
4479 	stats->zns_list_count++;
4480 
4481 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4482 		name = nvpair_name(nvp);
4483 
4484 		switch (nvpair_type(nvp)) {
4485 		case DATA_TYPE_STRING:
4486 			fnvlist_add_string(stats->zns_string, name,
4487 			    fnvpair_value_string(nvp));
4488 			break;
4489 		case DATA_TYPE_UINT64:
4490 			fnvlist_add_uint64(stats->zns_uint64, name,
4491 			    fnvpair_value_uint64(nvp));
4492 			break;
4493 		case DATA_TYPE_BOOLEAN:
4494 			fnvlist_add_boolean(stats->zns_boolean, name);
4495 			break;
4496 		case DATA_TYPE_NVLIST:
4497 			if (nvpair_value_nvlist(nvp, &list) == 0)
4498 				collect_nvlist_stats(list, stats);
4499 			break;
4500 		case DATA_TYPE_NVLIST_ARRAY:
4501 			if (nvpair_value_nvlist_array(nvp, &array, &items) != 0)
4502 				break;
4503 
4504 			for (i = 0; i < items; i++) {
4505 				collect_nvlist_stats(array[i], stats);
4506 
4507 				/* collect stats on leaf vdev */
4508 				if (strcmp(name, "children") == 0) {
4509 					size_t size;
4510 
4511 					(void) nvlist_size(array[i], &size,
4512 					    NV_ENCODE_XDR);
4513 					stats->zns_leaf_total += size;
4514 					if (size > stats->zns_leaf_largest)
4515 						stats->zns_leaf_largest = size;
4516 					stats->zns_leaf_count++;
4517 				}
4518 			}
4519 			break;
4520 		default:
4521 			(void) printf("skip type %d!\n", (int)nvpair_type(nvp));
4522 		}
4523 	}
4524 }
4525 
4526 static void
dump_nvlist_stats(nvlist_t * nvl,size_t cap)4527 dump_nvlist_stats(nvlist_t *nvl, size_t cap)
4528 {
4529 	zdb_nvl_stats_t stats = { 0 };
4530 	size_t size, sum = 0, total;
4531 	size_t noise;
4532 
4533 	/* requires nvlist with non-unique names for stat collection */
4534 	VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0));
4535 	VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0));
4536 	VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0));
4537 	VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR));
4538 
4539 	(void) printf("\n\nZFS Label NVList Config Stats:\n");
4540 
4541 	VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR));
4542 	(void) printf("  %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4543 	    (int)total, (int)(cap - total), 100.0 * total / cap);
4544 
4545 	collect_nvlist_stats(nvl, &stats);
4546 
4547 	VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR));
4548 	size -= noise;
4549 	sum += size;
4550 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4551 	    (int)fnvlist_num_pairs(stats.zns_uint64),
4552 	    (int)size, 100.0 * size / total);
4553 
4554 	VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR));
4555 	size -= noise;
4556 	sum += size;
4557 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4558 	    (int)fnvlist_num_pairs(stats.zns_string),
4559 	    (int)size, 100.0 * size / total);
4560 
4561 	VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR));
4562 	size -= noise;
4563 	sum += size;
4564 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4565 	    (int)fnvlist_num_pairs(stats.zns_boolean),
4566 	    (int)size, 100.0 * size / total);
4567 
4568 	size = total - sum;	/* treat remainder as nvlist overhead */
4569 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4570 	    stats.zns_list_count, (int)size, 100.0 * size / total);
4571 
4572 	if (stats.zns_leaf_count > 0) {
4573 		size_t average = stats.zns_leaf_total / stats.zns_leaf_count;
4574 
4575 		(void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4576 		    stats.zns_leaf_count, (int)average);
4577 		(void) printf("%24d bytes largest\n",
4578 		    (int)stats.zns_leaf_largest);
4579 
4580 		if (dump_opt['l'] >= 3 && average > 0)
4581 			(void) printf("  space for %d additional leaf vdevs\n",
4582 			    (int)((cap - total) / average));
4583 	}
4584 	(void) printf("\n");
4585 
4586 	nvlist_free(stats.zns_string);
4587 	nvlist_free(stats.zns_uint64);
4588 	nvlist_free(stats.zns_boolean);
4589 }
4590 
4591 typedef struct cksum_record {
4592 	zio_cksum_t cksum;
4593 	boolean_t labels[VDEV_LABELS];
4594 	avl_node_t link;
4595 } cksum_record_t;
4596 
4597 static int
cksum_record_compare(const void * x1,const void * x2)4598 cksum_record_compare(const void *x1, const void *x2)
4599 {
4600 	const cksum_record_t *l = (cksum_record_t *)x1;
4601 	const cksum_record_t *r = (cksum_record_t *)x2;
4602 	int arraysize = ARRAY_SIZE(l->cksum.zc_word);
4603 	int difference = 0;
4604 
4605 	for (int i = 0; i < arraysize; i++) {
4606 		difference = TREE_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]);
4607 		if (difference)
4608 			break;
4609 	}
4610 
4611 	return (difference);
4612 }
4613 
4614 static cksum_record_t *
cksum_record_alloc(zio_cksum_t * cksum,int l)4615 cksum_record_alloc(zio_cksum_t *cksum, int l)
4616 {
4617 	cksum_record_t *rec;
4618 
4619 	rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL);
4620 	rec->cksum = *cksum;
4621 	rec->labels[l] = B_TRUE;
4622 
4623 	return (rec);
4624 }
4625 
4626 static cksum_record_t *
cksum_record_lookup(avl_tree_t * tree,zio_cksum_t * cksum)4627 cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum)
4628 {
4629 	cksum_record_t lookup = { .cksum = *cksum };
4630 	avl_index_t where;
4631 
4632 	return (avl_find(tree, &lookup, &where));
4633 }
4634 
4635 static cksum_record_t *
cksum_record_insert(avl_tree_t * tree,zio_cksum_t * cksum,int l)4636 cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l)
4637 {
4638 	cksum_record_t *rec;
4639 
4640 	rec = cksum_record_lookup(tree, cksum);
4641 	if (rec) {
4642 		rec->labels[l] = B_TRUE;
4643 	} else {
4644 		rec = cksum_record_alloc(cksum, l);
4645 		avl_add(tree, rec);
4646 	}
4647 
4648 	return (rec);
4649 }
4650 
4651 static int
first_label(cksum_record_t * rec)4652 first_label(cksum_record_t *rec)
4653 {
4654 	for (int i = 0; i < VDEV_LABELS; i++)
4655 		if (rec->labels[i])
4656 			return (i);
4657 
4658 	return (-1);
4659 }
4660 
4661 static void
print_label_numbers(const char * prefix,const cksum_record_t * rec)4662 print_label_numbers(const char *prefix, const cksum_record_t *rec)
4663 {
4664 	fputs(prefix, stdout);
4665 	for (int i = 0; i < VDEV_LABELS; i++)
4666 		if (rec->labels[i] == B_TRUE)
4667 			printf("%d ", i);
4668 	putchar('\n');
4669 }
4670 
4671 #define	MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4672 
4673 typedef struct zdb_label {
4674 	vdev_label_t label;
4675 	uint64_t label_offset;
4676 	nvlist_t *config_nv;
4677 	cksum_record_t *config;
4678 	cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT];
4679 	boolean_t header_printed;
4680 	boolean_t read_failed;
4681 	boolean_t cksum_valid;
4682 } zdb_label_t;
4683 
4684 static void
print_label_header(zdb_label_t * label,int l)4685 print_label_header(zdb_label_t *label, int l)
4686 {
4687 
4688 	if (dump_opt['q'])
4689 		return;
4690 
4691 	if (label->header_printed == B_TRUE)
4692 		return;
4693 
4694 	(void) printf("------------------------------------\n");
4695 	(void) printf("LABEL %d %s\n", l,
4696 	    label->cksum_valid ? "" : "(Bad label cksum)");
4697 	(void) printf("------------------------------------\n");
4698 
4699 	label->header_printed = B_TRUE;
4700 }
4701 
4702 static void
print_l2arc_header(void)4703 print_l2arc_header(void)
4704 {
4705 	(void) printf("------------------------------------\n");
4706 	(void) printf("L2ARC device header\n");
4707 	(void) printf("------------------------------------\n");
4708 }
4709 
4710 static void
print_l2arc_log_blocks(void)4711 print_l2arc_log_blocks(void)
4712 {
4713 	(void) printf("------------------------------------\n");
4714 	(void) printf("L2ARC device log blocks\n");
4715 	(void) printf("------------------------------------\n");
4716 }
4717 
4718 static void
dump_l2arc_log_entries(uint64_t log_entries,l2arc_log_ent_phys_t * le,uint64_t i)4719 dump_l2arc_log_entries(uint64_t log_entries,
4720     l2arc_log_ent_phys_t *le, uint64_t i)
4721 {
4722 	for (int j = 0; j < log_entries; j++) {
4723 		dva_t dva = le[j].le_dva;
4724 		(void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4725 		    "vdev: %llu, offset: %llu\n",
4726 		    (u_longlong_t)i, j + 1,
4727 		    (u_longlong_t)DVA_GET_ASIZE(&dva),
4728 		    (u_longlong_t)DVA_GET_VDEV(&dva),
4729 		    (u_longlong_t)DVA_GET_OFFSET(&dva));
4730 		(void) printf("|\t\t\t\tbirth: %llu\n",
4731 		    (u_longlong_t)le[j].le_birth);
4732 		(void) printf("|\t\t\t\tlsize: %llu\n",
4733 		    (u_longlong_t)L2BLK_GET_LSIZE((&le[j])->le_prop));
4734 		(void) printf("|\t\t\t\tpsize: %llu\n",
4735 		    (u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop));
4736 		(void) printf("|\t\t\t\tcompr: %llu\n",
4737 		    (u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop));
4738 		(void) printf("|\t\t\t\tcomplevel: %llu\n",
4739 		    (u_longlong_t)(&le[j])->le_complevel);
4740 		(void) printf("|\t\t\t\ttype: %llu\n",
4741 		    (u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop));
4742 		(void) printf("|\t\t\t\tprotected: %llu\n",
4743 		    (u_longlong_t)L2BLK_GET_PROTECTED((&le[j])->le_prop));
4744 		(void) printf("|\t\t\t\tprefetch: %llu\n",
4745 		    (u_longlong_t)L2BLK_GET_PREFETCH((&le[j])->le_prop));
4746 		(void) printf("|\t\t\t\taddress: %llu\n",
4747 		    (u_longlong_t)le[j].le_daddr);
4748 		(void) printf("|\t\t\t\tARC state: %llu\n",
4749 		    (u_longlong_t)L2BLK_GET_STATE((&le[j])->le_prop));
4750 		(void) printf("|\n");
4751 	}
4752 	(void) printf("\n");
4753 }
4754 
4755 static void
dump_l2arc_log_blkptr(const l2arc_log_blkptr_t * lbps)4756 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t *lbps)
4757 {
4758 	(void) printf("|\t\tdaddr: %llu\n", (u_longlong_t)lbps->lbp_daddr);
4759 	(void) printf("|\t\tpayload_asize: %llu\n",
4760 	    (u_longlong_t)lbps->lbp_payload_asize);
4761 	(void) printf("|\t\tpayload_start: %llu\n",
4762 	    (u_longlong_t)lbps->lbp_payload_start);
4763 	(void) printf("|\t\tlsize: %llu\n",
4764 	    (u_longlong_t)L2BLK_GET_LSIZE(lbps->lbp_prop));
4765 	(void) printf("|\t\tasize: %llu\n",
4766 	    (u_longlong_t)L2BLK_GET_PSIZE(lbps->lbp_prop));
4767 	(void) printf("|\t\tcompralgo: %llu\n",
4768 	    (u_longlong_t)L2BLK_GET_COMPRESS(lbps->lbp_prop));
4769 	(void) printf("|\t\tcksumalgo: %llu\n",
4770 	    (u_longlong_t)L2BLK_GET_CHECKSUM(lbps->lbp_prop));
4771 	(void) printf("|\n\n");
4772 }
4773 
4774 static void
dump_l2arc_log_blocks(int fd,const l2arc_dev_hdr_phys_t * l2dhdr,l2arc_dev_hdr_phys_t * rebuild)4775 dump_l2arc_log_blocks(int fd, const l2arc_dev_hdr_phys_t *l2dhdr,
4776     l2arc_dev_hdr_phys_t *rebuild)
4777 {
4778 	l2arc_log_blk_phys_t this_lb;
4779 	uint64_t asize;
4780 	l2arc_log_blkptr_t lbps[2];
4781 	zio_cksum_t cksum;
4782 	int failed = 0;
4783 	l2arc_dev_t dev;
4784 
4785 	if (!dump_opt['q'])
4786 		print_l2arc_log_blocks();
4787 	memcpy(lbps, l2dhdr->dh_start_lbps, sizeof (lbps));
4788 
4789 	dev.l2ad_evict = l2dhdr->dh_evict;
4790 	dev.l2ad_start = l2dhdr->dh_start;
4791 	dev.l2ad_end = l2dhdr->dh_end;
4792 
4793 	if (l2dhdr->dh_start_lbps[0].lbp_daddr == 0) {
4794 		/* no log blocks to read */
4795 		if (!dump_opt['q']) {
4796 			(void) printf("No log blocks to read\n");
4797 			(void) printf("\n");
4798 		}
4799 		return;
4800 	} else {
4801 		dev.l2ad_hand = lbps[0].lbp_daddr +
4802 		    L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4803 	}
4804 
4805 	dev.l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST);
4806 
4807 	for (;;) {
4808 		if (!l2arc_log_blkptr_valid(&dev, &lbps[0]))
4809 			break;
4810 
4811 		/* L2BLK_GET_PSIZE returns aligned size for log blocks */
4812 		asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4813 		if (pread64(fd, &this_lb, asize, lbps[0].lbp_daddr) != asize) {
4814 			if (!dump_opt['q']) {
4815 				(void) printf("Error while reading next log "
4816 				    "block\n\n");
4817 			}
4818 			break;
4819 		}
4820 
4821 		fletcher_4_native_varsize(&this_lb, asize, &cksum);
4822 		if (!ZIO_CHECKSUM_EQUAL(cksum, lbps[0].lbp_cksum)) {
4823 			failed++;
4824 			if (!dump_opt['q']) {
4825 				(void) printf("Invalid cksum\n");
4826 				dump_l2arc_log_blkptr(&lbps[0]);
4827 			}
4828 			break;
4829 		}
4830 
4831 		switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) {
4832 		case ZIO_COMPRESS_OFF:
4833 			break;
4834 		default: {
4835 			abd_t *abd = abd_alloc_linear(asize, B_TRUE);
4836 			abd_copy_from_buf_off(abd, &this_lb, 0, asize);
4837 			abd_t dabd;
4838 			abd_get_from_buf_struct(&dabd, &this_lb,
4839 			    sizeof (this_lb));
4840 			int err = zio_decompress_data(L2BLK_GET_COMPRESS(
4841 			    (&lbps[0])->lbp_prop), abd, &dabd,
4842 			    asize, sizeof (this_lb), NULL);
4843 			abd_free(&dabd);
4844 			abd_free(abd);
4845 			if (err != 0) {
4846 				(void) printf("L2ARC block decompression "
4847 				    "failed\n");
4848 				goto out;
4849 			}
4850 			break;
4851 		}
4852 		}
4853 
4854 		if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
4855 			byteswap_uint64_array(&this_lb, sizeof (this_lb));
4856 		if (this_lb.lb_magic != L2ARC_LOG_BLK_MAGIC) {
4857 			if (!dump_opt['q'])
4858 				(void) printf("Invalid log block magic\n\n");
4859 			break;
4860 		}
4861 
4862 		rebuild->dh_lb_count++;
4863 		rebuild->dh_lb_asize += asize;
4864 		if (dump_opt['l'] > 1 && !dump_opt['q']) {
4865 			(void) printf("lb[%4llu]\tmagic: %llu\n",
4866 			    (u_longlong_t)rebuild->dh_lb_count,
4867 			    (u_longlong_t)this_lb.lb_magic);
4868 			dump_l2arc_log_blkptr(&lbps[0]);
4869 		}
4870 
4871 		if (dump_opt['l'] > 2 && !dump_opt['q'])
4872 			dump_l2arc_log_entries(l2dhdr->dh_log_entries,
4873 			    this_lb.lb_entries,
4874 			    rebuild->dh_lb_count);
4875 
4876 		if (l2arc_range_check_overlap(lbps[1].lbp_payload_start,
4877 		    lbps[0].lbp_payload_start, dev.l2ad_evict) &&
4878 		    !dev.l2ad_first)
4879 			break;
4880 
4881 		lbps[0] = lbps[1];
4882 		lbps[1] = this_lb.lb_prev_lbp;
4883 	}
4884 out:
4885 	if (!dump_opt['q']) {
4886 		(void) printf("log_blk_count:\t %llu with valid cksum\n",
4887 		    (u_longlong_t)rebuild->dh_lb_count);
4888 		(void) printf("\t\t %d with invalid cksum\n", failed);
4889 		(void) printf("log_blk_asize:\t %llu\n\n",
4890 		    (u_longlong_t)rebuild->dh_lb_asize);
4891 	}
4892 }
4893 
4894 static int
dump_l2arc_header(int fd)4895 dump_l2arc_header(int fd)
4896 {
4897 	l2arc_dev_hdr_phys_t l2dhdr = {0}, rebuild = {0};
4898 	int error = B_FALSE;
4899 
4900 	if (pread64(fd, &l2dhdr, sizeof (l2dhdr),
4901 	    VDEV_LABEL_START_SIZE) != sizeof (l2dhdr)) {
4902 		error = B_TRUE;
4903 	} else {
4904 		if (l2dhdr.dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC))
4905 			byteswap_uint64_array(&l2dhdr, sizeof (l2dhdr));
4906 
4907 		if (l2dhdr.dh_magic != L2ARC_DEV_HDR_MAGIC)
4908 			error = B_TRUE;
4909 	}
4910 
4911 	if (error) {
4912 		(void) printf("L2ARC device header not found\n\n");
4913 		/* Do not return an error here for backward compatibility */
4914 		return (0);
4915 	} else if (!dump_opt['q']) {
4916 		print_l2arc_header();
4917 
4918 		(void) printf("    magic: %llu\n",
4919 		    (u_longlong_t)l2dhdr.dh_magic);
4920 		(void) printf("    version: %llu\n",
4921 		    (u_longlong_t)l2dhdr.dh_version);
4922 		(void) printf("    pool_guid: %llu\n",
4923 		    (u_longlong_t)l2dhdr.dh_spa_guid);
4924 		(void) printf("    flags: %llu\n",
4925 		    (u_longlong_t)l2dhdr.dh_flags);
4926 		(void) printf("    start_lbps[0]: %llu\n",
4927 		    (u_longlong_t)
4928 		    l2dhdr.dh_start_lbps[0].lbp_daddr);
4929 		(void) printf("    start_lbps[1]: %llu\n",
4930 		    (u_longlong_t)
4931 		    l2dhdr.dh_start_lbps[1].lbp_daddr);
4932 		(void) printf("    log_blk_ent: %llu\n",
4933 		    (u_longlong_t)l2dhdr.dh_log_entries);
4934 		(void) printf("    start: %llu\n",
4935 		    (u_longlong_t)l2dhdr.dh_start);
4936 		(void) printf("    end: %llu\n",
4937 		    (u_longlong_t)l2dhdr.dh_end);
4938 		(void) printf("    evict: %llu\n",
4939 		    (u_longlong_t)l2dhdr.dh_evict);
4940 		(void) printf("    lb_asize_refcount: %llu\n",
4941 		    (u_longlong_t)l2dhdr.dh_lb_asize);
4942 		(void) printf("    lb_count_refcount: %llu\n",
4943 		    (u_longlong_t)l2dhdr.dh_lb_count);
4944 		(void) printf("    trim_action_time: %llu\n",
4945 		    (u_longlong_t)l2dhdr.dh_trim_action_time);
4946 		(void) printf("    trim_state: %llu\n\n",
4947 		    (u_longlong_t)l2dhdr.dh_trim_state);
4948 	}
4949 
4950 	dump_l2arc_log_blocks(fd, &l2dhdr, &rebuild);
4951 	/*
4952 	 * The total aligned size of log blocks and the number of log blocks
4953 	 * reported in the header of the device may be less than what zdb
4954 	 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
4955 	 * This happens because dump_l2arc_log_blocks() lacks the memory
4956 	 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
4957 	 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
4958 	 * and dh_lb_count will be lower to begin with than what exists on the
4959 	 * device. This is normal and zdb should not exit with an error. The
4960 	 * opposite case should never happen though, the values reported in the
4961 	 * header should never be higher than what dump_l2arc_log_blocks() and
4962 	 * l2arc_rebuild() report. If this happens there is a leak in the
4963 	 * accounting of log blocks.
4964 	 */
4965 	if (l2dhdr.dh_lb_asize > rebuild.dh_lb_asize ||
4966 	    l2dhdr.dh_lb_count > rebuild.dh_lb_count)
4967 		return (1);
4968 
4969 	return (0);
4970 }
4971 
4972 static void
dump_config_from_label(zdb_label_t * label,size_t buflen,int l)4973 dump_config_from_label(zdb_label_t *label, size_t buflen, int l)
4974 {
4975 	if (dump_opt['q'])
4976 		return;
4977 
4978 	if ((dump_opt['l'] < 3) && (first_label(label->config) != l))
4979 		return;
4980 
4981 	print_label_header(label, l);
4982 	dump_nvlist(label->config_nv, 4);
4983 	print_label_numbers("    labels = ", label->config);
4984 
4985 	if (dump_opt['l'] >= 2)
4986 		dump_nvlist_stats(label->config_nv, buflen);
4987 }
4988 
4989 #define	ZDB_MAX_UB_HEADER_SIZE 32
4990 
4991 static void
dump_label_uberblocks(zdb_label_t * label,uint64_t ashift,int label_num)4992 dump_label_uberblocks(zdb_label_t *label, uint64_t ashift, int label_num)
4993 {
4994 
4995 	vdev_t vd;
4996 	char header[ZDB_MAX_UB_HEADER_SIZE];
4997 
4998 	vd.vdev_ashift = ashift;
4999 	vd.vdev_top = &vd;
5000 
5001 	for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5002 		uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5003 		uberblock_t *ub = (void *)((char *)&label->label + uoff);
5004 		cksum_record_t *rec = label->uberblocks[i];
5005 
5006 		if (rec == NULL) {
5007 			if (dump_opt['u'] >= 2) {
5008 				print_label_header(label, label_num);
5009 				(void) printf("    Uberblock[%d] invalid\n", i);
5010 			}
5011 			continue;
5012 		}
5013 
5014 		if ((dump_opt['u'] < 3) && (first_label(rec) != label_num))
5015 			continue;
5016 
5017 		if ((dump_opt['u'] < 4) &&
5018 		    (ub->ub_mmp_magic == MMP_MAGIC) && ub->ub_mmp_delay &&
5019 		    (i >= VDEV_UBERBLOCK_COUNT(&vd) - MMP_BLOCKS_PER_LABEL))
5020 			continue;
5021 
5022 		print_label_header(label, label_num);
5023 		(void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE,
5024 		    "    Uberblock[%d]\n", i);
5025 		dump_uberblock(ub, header, "");
5026 		print_label_numbers("        labels = ", rec);
5027 	}
5028 }
5029 
5030 static char curpath[PATH_MAX];
5031 
5032 /*
5033  * Iterate through the path components, recursively passing
5034  * current one's obj and remaining path until we find the obj
5035  * for the last one.
5036  */
5037 static int
dump_path_impl(objset_t * os,uint64_t obj,char * name,uint64_t * retobj)5038 dump_path_impl(objset_t *os, uint64_t obj, char *name, uint64_t *retobj)
5039 {
5040 	int err;
5041 	boolean_t header = B_TRUE;
5042 	uint64_t child_obj;
5043 	char *s;
5044 	dmu_buf_t *db;
5045 	dmu_object_info_t doi;
5046 
5047 	if ((s = strchr(name, '/')) != NULL)
5048 		*s = '\0';
5049 	err = zap_lookup(os, obj, name, 8, 1, &child_obj);
5050 
5051 	(void) strlcat(curpath, name, sizeof (curpath));
5052 
5053 	if (err != 0) {
5054 		(void) fprintf(stderr, "failed to lookup %s: %s\n",
5055 		    curpath, strerror(err));
5056 		return (err);
5057 	}
5058 
5059 	child_obj = ZFS_DIRENT_OBJ(child_obj);
5060 	err = sa_buf_hold(os, child_obj, FTAG, &db);
5061 	if (err != 0) {
5062 		(void) fprintf(stderr,
5063 		    "failed to get SA dbuf for obj %llu: %s\n",
5064 		    (u_longlong_t)child_obj, strerror(err));
5065 		return (EINVAL);
5066 	}
5067 	dmu_object_info_from_db(db, &doi);
5068 	sa_buf_rele(db, FTAG);
5069 
5070 	if (doi.doi_bonus_type != DMU_OT_SA &&
5071 	    doi.doi_bonus_type != DMU_OT_ZNODE) {
5072 		(void) fprintf(stderr, "invalid bonus type %d for obj %llu\n",
5073 		    doi.doi_bonus_type, (u_longlong_t)child_obj);
5074 		return (EINVAL);
5075 	}
5076 
5077 	if (dump_opt['v'] > 6) {
5078 		(void) printf("obj=%llu %s type=%d bonustype=%d\n",
5079 		    (u_longlong_t)child_obj, curpath, doi.doi_type,
5080 		    doi.doi_bonus_type);
5081 	}
5082 
5083 	(void) strlcat(curpath, "/", sizeof (curpath));
5084 
5085 	switch (doi.doi_type) {
5086 	case DMU_OT_DIRECTORY_CONTENTS:
5087 		if (s != NULL && *(s + 1) != '\0')
5088 			return (dump_path_impl(os, child_obj, s + 1, retobj));
5089 		zfs_fallthrough;
5090 	case DMU_OT_PLAIN_FILE_CONTENTS:
5091 		if (retobj != NULL) {
5092 			*retobj = child_obj;
5093 		} else {
5094 			dump_object(os, child_obj, dump_opt['v'], &header,
5095 			    NULL, 0);
5096 		}
5097 		return (0);
5098 	default:
5099 		(void) fprintf(stderr, "object %llu has non-file/directory "
5100 		    "type %d\n", (u_longlong_t)obj, doi.doi_type);
5101 		break;
5102 	}
5103 
5104 	return (EINVAL);
5105 }
5106 
5107 /*
5108  * Dump the blocks for the object specified by path inside the dataset.
5109  */
5110 static int
dump_path(char * ds,char * path,uint64_t * retobj)5111 dump_path(char *ds, char *path, uint64_t *retobj)
5112 {
5113 	int err;
5114 	objset_t *os;
5115 	uint64_t root_obj;
5116 
5117 	err = open_objset(ds, FTAG, &os);
5118 	if (err != 0)
5119 		return (err);
5120 
5121 	err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj);
5122 	if (err != 0) {
5123 		(void) fprintf(stderr, "can't lookup root znode: %s\n",
5124 		    strerror(err));
5125 		close_objset(os, FTAG);
5126 		return (EINVAL);
5127 	}
5128 
5129 	(void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds);
5130 
5131 	err = dump_path_impl(os, root_obj, path, retobj);
5132 
5133 	close_objset(os, FTAG);
5134 	return (err);
5135 }
5136 
5137 static int
dump_backup_bytes(objset_t * os,void * buf,int len,void * arg)5138 dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
5139 {
5140 	const char *p = (const char *)buf;
5141 	ssize_t nwritten;
5142 
5143 	(void) os;
5144 	(void) arg;
5145 
5146 	/* Write the data out, handling short writes and signals. */
5147 	while ((nwritten = write(STDOUT_FILENO, p, len)) < len) {
5148 		if (nwritten < 0) {
5149 			if (errno == EINTR)
5150 				continue;
5151 			return (errno);
5152 		}
5153 		p += nwritten;
5154 		len -= nwritten;
5155 	}
5156 
5157 	return (0);
5158 }
5159 
5160 static void
dump_backup(const char * pool,uint64_t objset_id,const char * flagstr)5161 dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
5162 {
5163 	boolean_t embed = B_FALSE;
5164 	boolean_t large_block = B_FALSE;
5165 	boolean_t compress = B_FALSE;
5166 	boolean_t raw = B_FALSE;
5167 
5168 	const char *c;
5169 	for (c = flagstr; c != NULL && *c != '\0'; c++) {
5170 		switch (*c) {
5171 			case 'e':
5172 				embed = B_TRUE;
5173 				break;
5174 			case 'L':
5175 				large_block = B_TRUE;
5176 				break;
5177 			case 'c':
5178 				compress = B_TRUE;
5179 				break;
5180 			case 'w':
5181 				raw = B_TRUE;
5182 				break;
5183 			default:
5184 				fprintf(stderr, "dump_backup: invalid flag "
5185 				    "'%c'\n", *c);
5186 				return;
5187 		}
5188 	}
5189 
5190 	if (isatty(STDOUT_FILENO)) {
5191 		fprintf(stderr, "dump_backup: stream cannot be written "
5192 		    "to a terminal\n");
5193 		return;
5194 	}
5195 
5196 	offset_t off = 0;
5197 	dmu_send_outparams_t out = {
5198 	    .dso_outfunc = dump_backup_bytes,
5199 	    .dso_dryrun  = B_FALSE,
5200 	};
5201 
5202 	int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed,
5203 	    large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO,
5204 	    &off, &out);
5205 	if (err != 0) {
5206 		fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",
5207 		    strerror(err));
5208 		return;
5209 	}
5210 }
5211 
5212 static int
zdb_copy_object(objset_t * os,uint64_t srcobj,char * destfile)5213 zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile)
5214 {
5215 	int err = 0;
5216 	uint64_t size, readsize, oursize, offset;
5217 	ssize_t writesize;
5218 	sa_handle_t *hdl;
5219 
5220 	(void) printf("Copying object %" PRIu64 " to file %s\n", srcobj,
5221 	    destfile);
5222 
5223 	VERIFY3P(os, ==, sa_os);
5224 	if ((err = sa_handle_get(os, srcobj, NULL, SA_HDL_PRIVATE, &hdl))) {
5225 		(void) printf("Failed to get handle for SA znode\n");
5226 		return (err);
5227 	}
5228 	if ((err = sa_lookup(hdl, sa_attr_table[ZPL_SIZE], &size, 8))) {
5229 		(void) sa_handle_destroy(hdl);
5230 		return (err);
5231 	}
5232 	(void) sa_handle_destroy(hdl);
5233 
5234 	(void) printf("Object %" PRIu64 " is %" PRIu64 " bytes\n", srcobj,
5235 	    size);
5236 	if (size == 0) {
5237 		return (EINVAL);
5238 	}
5239 
5240 	int fd = open(destfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
5241 	if (fd == -1)
5242 		return (errno);
5243 	/*
5244 	 * We cap the size at 1 mebibyte here to prevent
5245 	 * allocation failures and nigh-infinite printing if the
5246 	 * object is extremely large.
5247 	 */
5248 	oursize = MIN(size, 1 << 20);
5249 	offset = 0;
5250 	char *buf = kmem_alloc(oursize, KM_NOSLEEP);
5251 	if (buf == NULL) {
5252 		(void) close(fd);
5253 		return (ENOMEM);
5254 	}
5255 
5256 	while (offset < size) {
5257 		readsize = MIN(size - offset, 1 << 20);
5258 		err = dmu_read(os, srcobj, offset, readsize, buf, 0);
5259 		if (err != 0) {
5260 			(void) printf("got error %u from dmu_read\n", err);
5261 			kmem_free(buf, oursize);
5262 			(void) close(fd);
5263 			return (err);
5264 		}
5265 		if (dump_opt['v'] > 3) {
5266 			(void) printf("Read offset=%" PRIu64 " size=%" PRIu64
5267 			    " error=%d\n", offset, readsize, err);
5268 		}
5269 
5270 		writesize = write(fd, buf, readsize);
5271 		if (writesize < 0) {
5272 			err = errno;
5273 			break;
5274 		} else if (writesize != readsize) {
5275 			/* Incomplete write */
5276 			(void) fprintf(stderr, "Short write, only wrote %llu of"
5277 			    " %" PRIu64 " bytes, exiting...\n",
5278 			    (u_longlong_t)writesize, readsize);
5279 			break;
5280 		}
5281 
5282 		offset += readsize;
5283 	}
5284 
5285 	(void) close(fd);
5286 
5287 	if (buf != NULL)
5288 		kmem_free(buf, oursize);
5289 
5290 	return (err);
5291 }
5292 
5293 static boolean_t
label_cksum_valid(vdev_label_t * label,uint64_t offset)5294 label_cksum_valid(vdev_label_t *label, uint64_t offset)
5295 {
5296 	zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
5297 	zio_cksum_t expected_cksum;
5298 	zio_cksum_t actual_cksum;
5299 	zio_cksum_t verifier;
5300 	zio_eck_t *eck;
5301 	int byteswap;
5302 
5303 	void *data = (char *)label + offsetof(vdev_label_t, vl_vdev_phys);
5304 	eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1;
5305 
5306 	offset += offsetof(vdev_label_t, vl_vdev_phys);
5307 	ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
5308 
5309 	byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
5310 	if (byteswap)
5311 		byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
5312 
5313 	expected_cksum = eck->zec_cksum;
5314 	eck->zec_cksum = verifier;
5315 
5316 	abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE);
5317 	ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum);
5318 	abd_free(abd);
5319 
5320 	if (byteswap)
5321 		byteswap_uint64_array(&expected_cksum, sizeof (zio_cksum_t));
5322 
5323 	if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
5324 		return (B_TRUE);
5325 
5326 	return (B_FALSE);
5327 }
5328 
5329 static int
dump_label(const char * dev)5330 dump_label(const char *dev)
5331 {
5332 	char path[MAXPATHLEN];
5333 	zdb_label_t labels[VDEV_LABELS] = {{{{0}}}};
5334 	uint64_t psize, ashift, l2cache;
5335 	struct stat64 statbuf;
5336 	boolean_t config_found = B_FALSE;
5337 	boolean_t error = B_FALSE;
5338 	boolean_t read_l2arc_header = B_FALSE;
5339 	avl_tree_t config_tree;
5340 	avl_tree_t uberblock_tree;
5341 	void *node, *cookie;
5342 	int fd;
5343 
5344 	/*
5345 	 * Check if we were given absolute path and use it as is.
5346 	 * Otherwise if the provided vdev name doesn't point to a file,
5347 	 * try prepending expected disk paths and partition numbers.
5348 	 */
5349 	(void) strlcpy(path, dev, sizeof (path));
5350 	if (dev[0] != '/' && stat64(path, &statbuf) != 0) {
5351 		int error;
5352 
5353 		error = zfs_resolve_shortname(dev, path, MAXPATHLEN);
5354 		if (error == 0 && zfs_dev_is_whole_disk(path)) {
5355 			if (zfs_append_partition(path, MAXPATHLEN) == -1)
5356 				error = ENOENT;
5357 		}
5358 
5359 		if (error || (stat64(path, &statbuf) != 0)) {
5360 			(void) printf("failed to find device %s, try "
5361 			    "specifying absolute path instead\n", dev);
5362 			return (1);
5363 		}
5364 	}
5365 
5366 	if ((fd = open64(path, O_RDONLY)) < 0) {
5367 		(void) printf("cannot open '%s': %s\n", path, strerror(errno));
5368 		zdb_exit(1);
5369 	}
5370 
5371 	if (fstat64_blk(fd, &statbuf) != 0) {
5372 		(void) printf("failed to stat '%s': %s\n", path,
5373 		    strerror(errno));
5374 		(void) close(fd);
5375 		zdb_exit(1);
5376 	}
5377 
5378 	if (S_ISBLK(statbuf.st_mode) && zfs_dev_flush(fd) != 0)
5379 		(void) printf("failed to invalidate cache '%s' : %s\n", path,
5380 		    strerror(errno));
5381 
5382 	avl_create(&config_tree, cksum_record_compare,
5383 	    sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5384 	avl_create(&uberblock_tree, cksum_record_compare,
5385 	    sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5386 
5387 	psize = statbuf.st_size;
5388 	psize = P2ALIGN_TYPED(psize, sizeof (vdev_label_t), uint64_t);
5389 	ashift = SPA_MINBLOCKSHIFT;
5390 
5391 	/*
5392 	 * 1. Read the label from disk
5393 	 * 2. Verify label cksum
5394 	 * 3. Unpack the configuration and insert in config tree.
5395 	 * 4. Traverse all uberblocks and insert in uberblock tree.
5396 	 */
5397 	for (int l = 0; l < VDEV_LABELS; l++) {
5398 		zdb_label_t *label = &labels[l];
5399 		char *buf = label->label.vl_vdev_phys.vp_nvlist;
5400 		size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5401 		nvlist_t *config;
5402 		cksum_record_t *rec;
5403 		zio_cksum_t cksum;
5404 		vdev_t vd;
5405 
5406 		label->label_offset = vdev_label_offset(psize, l, 0);
5407 
5408 		if (pread64(fd, &label->label, sizeof (label->label),
5409 		    label->label_offset) != sizeof (label->label)) {
5410 			if (!dump_opt['q'])
5411 				(void) printf("failed to read label %d\n", l);
5412 			label->read_failed = B_TRUE;
5413 			error = B_TRUE;
5414 			continue;
5415 		}
5416 
5417 		label->read_failed = B_FALSE;
5418 		label->cksum_valid = label_cksum_valid(&label->label,
5419 		    label->label_offset);
5420 
5421 		if (nvlist_unpack(buf, buflen, &config, 0) == 0) {
5422 			nvlist_t *vdev_tree = NULL;
5423 			size_t size;
5424 
5425 			if ((nvlist_lookup_nvlist(config,
5426 			    ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) ||
5427 			    (nvlist_lookup_uint64(vdev_tree,
5428 			    ZPOOL_CONFIG_ASHIFT, &ashift) != 0))
5429 				ashift = SPA_MINBLOCKSHIFT;
5430 
5431 			if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0)
5432 				size = buflen;
5433 
5434 			/* If the device is a cache device read the header. */
5435 			if (!read_l2arc_header) {
5436 				if (nvlist_lookup_uint64(config,
5437 				    ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 &&
5438 				    l2cache == POOL_STATE_L2CACHE) {
5439 					read_l2arc_header = B_TRUE;
5440 				}
5441 			}
5442 
5443 			fletcher_4_native_varsize(buf, size, &cksum);
5444 			rec = cksum_record_insert(&config_tree, &cksum, l);
5445 
5446 			label->config = rec;
5447 			label->config_nv = config;
5448 			config_found = B_TRUE;
5449 		} else {
5450 			error = B_TRUE;
5451 		}
5452 
5453 		vd.vdev_ashift = ashift;
5454 		vd.vdev_top = &vd;
5455 
5456 		for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5457 			uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5458 			uberblock_t *ub = (void *)((char *)label + uoff);
5459 
5460 			if (uberblock_verify(ub))
5461 				continue;
5462 
5463 			fletcher_4_native_varsize(ub, sizeof (*ub), &cksum);
5464 			rec = cksum_record_insert(&uberblock_tree, &cksum, l);
5465 
5466 			label->uberblocks[i] = rec;
5467 		}
5468 	}
5469 
5470 	/*
5471 	 * Dump the label and uberblocks.
5472 	 */
5473 	for (int l = 0; l < VDEV_LABELS; l++) {
5474 		zdb_label_t *label = &labels[l];
5475 		size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5476 
5477 		if (label->read_failed == B_TRUE)
5478 			continue;
5479 
5480 		if (label->config_nv) {
5481 			dump_config_from_label(label, buflen, l);
5482 		} else {
5483 			if (!dump_opt['q'])
5484 				(void) printf("failed to unpack label %d\n", l);
5485 		}
5486 
5487 		if (dump_opt['u'])
5488 			dump_label_uberblocks(label, ashift, l);
5489 
5490 		nvlist_free(label->config_nv);
5491 	}
5492 
5493 	/*
5494 	 * Dump the L2ARC header, if existent.
5495 	 */
5496 	if (read_l2arc_header)
5497 		error |= dump_l2arc_header(fd);
5498 
5499 	cookie = NULL;
5500 	while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL)
5501 		umem_free(node, sizeof (cksum_record_t));
5502 
5503 	cookie = NULL;
5504 	while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL)
5505 		umem_free(node, sizeof (cksum_record_t));
5506 
5507 	avl_destroy(&config_tree);
5508 	avl_destroy(&uberblock_tree);
5509 
5510 	(void) close(fd);
5511 
5512 	return (config_found == B_FALSE ? 2 :
5513 	    (error == B_TRUE ? 1 : 0));
5514 }
5515 
5516 static uint64_t dataset_feature_count[SPA_FEATURES];
5517 static uint64_t global_feature_count[SPA_FEATURES];
5518 static uint64_t remap_deadlist_count = 0;
5519 
5520 static int
dump_one_objset(const char * dsname,void * arg)5521 dump_one_objset(const char *dsname, void *arg)
5522 {
5523 	(void) arg;
5524 	int error;
5525 	objset_t *os;
5526 	spa_feature_t f;
5527 
5528 	error = open_objset(dsname, FTAG, &os);
5529 	if (error != 0)
5530 		return (0);
5531 
5532 	for (f = 0; f < SPA_FEATURES; f++) {
5533 		if (!dsl_dataset_feature_is_active(dmu_objset_ds(os), f))
5534 			continue;
5535 		ASSERT(spa_feature_table[f].fi_flags &
5536 		    ZFEATURE_FLAG_PER_DATASET);
5537 		dataset_feature_count[f]++;
5538 	}
5539 
5540 	if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os))) {
5541 		remap_deadlist_count++;
5542 	}
5543 
5544 	for (dsl_bookmark_node_t *dbn =
5545 	    avl_first(&dmu_objset_ds(os)->ds_bookmarks); dbn != NULL;
5546 	    dbn = AVL_NEXT(&dmu_objset_ds(os)->ds_bookmarks, dbn)) {
5547 		mos_obj_refd(dbn->dbn_phys.zbm_redaction_obj);
5548 		if (dbn->dbn_phys.zbm_redaction_obj != 0) {
5549 			global_feature_count[
5550 			    SPA_FEATURE_REDACTION_BOOKMARKS]++;
5551 			objset_t *mos = os->os_spa->spa_meta_objset;
5552 			dnode_t *rl;
5553 			VERIFY0(dnode_hold(mos,
5554 			    dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl));
5555 			if (rl->dn_have_spill) {
5556 				global_feature_count[
5557 				    SPA_FEATURE_REDACTION_LIST_SPILL]++;
5558 			}
5559 		}
5560 		if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)
5561 			global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN]++;
5562 	}
5563 
5564 	if (dsl_deadlist_is_open(&dmu_objset_ds(os)->ds_dir->dd_livelist) &&
5565 	    !dmu_objset_is_snapshot(os)) {
5566 		global_feature_count[SPA_FEATURE_LIVELIST]++;
5567 	}
5568 
5569 	dump_objset(os);
5570 	close_objset(os, FTAG);
5571 	fuid_table_destroy();
5572 	return (0);
5573 }
5574 
5575 /*
5576  * Block statistics.
5577  */
5578 #define	PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5579 typedef struct zdb_blkstats {
5580 	uint64_t zb_asize;
5581 	uint64_t zb_lsize;
5582 	uint64_t zb_psize;
5583 	uint64_t zb_count;
5584 	uint64_t zb_gangs;
5585 	uint64_t zb_ditto_samevdev;
5586 	uint64_t zb_ditto_same_ms;
5587 	uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
5588 } zdb_blkstats_t;
5589 
5590 /*
5591  * Extended object types to report deferred frees and dedup auto-ditto blocks.
5592  */
5593 #define	ZDB_OT_DEFERRED	(DMU_OT_NUMTYPES + 0)
5594 #define	ZDB_OT_DITTO	(DMU_OT_NUMTYPES + 1)
5595 #define	ZDB_OT_OTHER	(DMU_OT_NUMTYPES + 2)
5596 #define	ZDB_OT_TOTAL	(DMU_OT_NUMTYPES + 3)
5597 
5598 static const char *zdb_ot_extname[] = {
5599 	"deferred free",
5600 	"dedup ditto",
5601 	"other",
5602 	"Total",
5603 };
5604 
5605 #define	ZB_TOTAL	DN_MAX_LEVELS
5606 #define	SPA_MAX_FOR_16M	(SPA_MAXBLOCKSHIFT+1)
5607 
5608 typedef struct zdb_brt_entry {
5609 	dva_t		zbre_dva;
5610 	uint64_t	zbre_refcount;
5611 	avl_node_t	zbre_node;
5612 } zdb_brt_entry_t;
5613 
5614 typedef struct zdb_cb {
5615 	zdb_blkstats_t	zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
5616 	uint64_t	zcb_removing_size;
5617 	uint64_t	zcb_checkpoint_size;
5618 	uint64_t	zcb_dedup_asize;
5619 	uint64_t	zcb_dedup_blocks;
5620 	uint64_t	zcb_clone_asize;
5621 	uint64_t	zcb_clone_blocks;
5622 	uint64_t	zcb_psize_count[SPA_MAX_FOR_16M];
5623 	uint64_t	zcb_lsize_count[SPA_MAX_FOR_16M];
5624 	uint64_t	zcb_asize_count[SPA_MAX_FOR_16M];
5625 	uint64_t	zcb_psize_len[SPA_MAX_FOR_16M];
5626 	uint64_t	zcb_lsize_len[SPA_MAX_FOR_16M];
5627 	uint64_t	zcb_asize_len[SPA_MAX_FOR_16M];
5628 	uint64_t	zcb_psize_total;
5629 	uint64_t	zcb_lsize_total;
5630 	uint64_t	zcb_asize_total;
5631 	uint64_t	zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
5632 	uint64_t	zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
5633 	    [BPE_PAYLOAD_SIZE + 1];
5634 	uint64_t	zcb_start;
5635 	hrtime_t	zcb_lastprint;
5636 	uint64_t	zcb_totalasize;
5637 	uint64_t	zcb_errors[256];
5638 	int		zcb_readfails;
5639 	int		zcb_haderrors;
5640 	spa_t		*zcb_spa;
5641 	uint32_t	**zcb_vd_obsolete_counts;
5642 	avl_tree_t	zcb_brt;
5643 	boolean_t	zcb_brt_is_active;
5644 } zdb_cb_t;
5645 
5646 /* test if two DVA offsets from same vdev are within the same metaslab */
5647 static boolean_t
same_metaslab(spa_t * spa,uint64_t vdev,uint64_t off1,uint64_t off2)5648 same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2)
5649 {
5650 	vdev_t *vd = vdev_lookup_top(spa, vdev);
5651 	uint64_t ms_shift = vd->vdev_ms_shift;
5652 
5653 	return ((off1 >> ms_shift) == (off2 >> ms_shift));
5654 }
5655 
5656 /*
5657  * Used to simplify reporting of the histogram data.
5658  */
5659 typedef struct one_histo {
5660 	const char *name;
5661 	uint64_t *count;
5662 	uint64_t *len;
5663 	uint64_t cumulative;
5664 } one_histo_t;
5665 
5666 /*
5667  * The number of separate histograms processed for psize, lsize and asize.
5668  */
5669 #define	NUM_HISTO 3
5670 
5671 /*
5672  * This routine will create a fixed column size output of three different
5673  * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5674  * the count, length and cumulative length of the psize, lsize and
5675  * asize blocks.
5676  *
5677  * All three types of blocks are listed on a single line
5678  *
5679  * By default the table is printed in nicenumber format (e.g. 123K) but
5680  * if the '-P' parameter is specified then the full raw number (parseable)
5681  * is printed out.
5682  */
5683 static void
dump_size_histograms(zdb_cb_t * zcb)5684 dump_size_histograms(zdb_cb_t *zcb)
5685 {
5686 	/*
5687 	 * A temporary buffer that allows us to convert a number into
5688 	 * a string using zdb_nicenumber to allow either raw or human
5689 	 * readable numbers to be output.
5690 	 */
5691 	char numbuf[32];
5692 
5693 	/*
5694 	 * Define titles which are used in the headers of the tables
5695 	 * printed by this routine.
5696 	 */
5697 	const char blocksize_title1[] = "block";
5698 	const char blocksize_title2[] = "size";
5699 	const char count_title[] = "Count";
5700 	const char length_title[] = "Size";
5701 	const char cumulative_title[] = "Cum.";
5702 
5703 	/*
5704 	 * Setup the histogram arrays (psize, lsize, and asize).
5705 	 */
5706 	one_histo_t parm_histo[NUM_HISTO];
5707 
5708 	parm_histo[0].name = "psize";
5709 	parm_histo[0].count = zcb->zcb_psize_count;
5710 	parm_histo[0].len = zcb->zcb_psize_len;
5711 	parm_histo[0].cumulative = 0;
5712 
5713 	parm_histo[1].name = "lsize";
5714 	parm_histo[1].count = zcb->zcb_lsize_count;
5715 	parm_histo[1].len = zcb->zcb_lsize_len;
5716 	parm_histo[1].cumulative = 0;
5717 
5718 	parm_histo[2].name = "asize";
5719 	parm_histo[2].count = zcb->zcb_asize_count;
5720 	parm_histo[2].len = zcb->zcb_asize_len;
5721 	parm_histo[2].cumulative = 0;
5722 
5723 
5724 	(void) printf("\nBlock Size Histogram\n");
5725 	/*
5726 	 * Print the first line titles
5727 	 */
5728 	if (dump_opt['P'])
5729 		(void) printf("\n%s\t", blocksize_title1);
5730 	else
5731 		(void) printf("\n%7s   ", blocksize_title1);
5732 
5733 	for (int j = 0; j < NUM_HISTO; j++) {
5734 		if (dump_opt['P']) {
5735 			if (j < NUM_HISTO - 1) {
5736 				(void) printf("%s\t\t\t", parm_histo[j].name);
5737 			} else {
5738 				/* Don't print trailing spaces */
5739 				(void) printf("  %s", parm_histo[j].name);
5740 			}
5741 		} else {
5742 			if (j < NUM_HISTO - 1) {
5743 				/* Left aligned strings in the output */
5744 				(void) printf("%-7s              ",
5745 				    parm_histo[j].name);
5746 			} else {
5747 				/* Don't print trailing spaces */
5748 				(void) printf("%s", parm_histo[j].name);
5749 			}
5750 		}
5751 	}
5752 	(void) printf("\n");
5753 
5754 	/*
5755 	 * Print the second line titles
5756 	 */
5757 	if (dump_opt['P']) {
5758 		(void) printf("%s\t", blocksize_title2);
5759 	} else {
5760 		(void) printf("%7s ", blocksize_title2);
5761 	}
5762 
5763 	for (int i = 0; i < NUM_HISTO; i++) {
5764 		if (dump_opt['P']) {
5765 			(void) printf("%s\t%s\t%s\t",
5766 			    count_title, length_title, cumulative_title);
5767 		} else {
5768 			(void) printf("%7s%7s%7s",
5769 			    count_title, length_title, cumulative_title);
5770 		}
5771 	}
5772 	(void) printf("\n");
5773 
5774 	/*
5775 	 * Print the rows
5776 	 */
5777 	for (int i = SPA_MINBLOCKSHIFT; i < SPA_MAX_FOR_16M; i++) {
5778 
5779 		/*
5780 		 * Print the first column showing the blocksize
5781 		 */
5782 		zdb_nicenum((1ULL << i), numbuf, sizeof (numbuf));
5783 
5784 		if (dump_opt['P']) {
5785 			printf("%s", numbuf);
5786 		} else {
5787 			printf("%7s:", numbuf);
5788 		}
5789 
5790 		/*
5791 		 * Print the remaining set of 3 columns per size:
5792 		 * for psize, lsize and asize
5793 		 */
5794 		for (int j = 0; j < NUM_HISTO; j++) {
5795 			parm_histo[j].cumulative += parm_histo[j].len[i];
5796 
5797 			zdb_nicenum(parm_histo[j].count[i],
5798 			    numbuf, sizeof (numbuf));
5799 			if (dump_opt['P'])
5800 				(void) printf("\t%s", numbuf);
5801 			else
5802 				(void) printf("%7s", numbuf);
5803 
5804 			zdb_nicenum(parm_histo[j].len[i],
5805 			    numbuf, sizeof (numbuf));
5806 			if (dump_opt['P'])
5807 				(void) printf("\t%s", numbuf);
5808 			else
5809 				(void) printf("%7s", numbuf);
5810 
5811 			zdb_nicenum(parm_histo[j].cumulative,
5812 			    numbuf, sizeof (numbuf));
5813 			if (dump_opt['P'])
5814 				(void) printf("\t%s", numbuf);
5815 			else
5816 				(void) printf("%7s", numbuf);
5817 		}
5818 		(void) printf("\n");
5819 	}
5820 }
5821 
5822 static void
zdb_count_block(zdb_cb_t * zcb,zilog_t * zilog,const blkptr_t * bp,dmu_object_type_t type)5823 zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
5824     dmu_object_type_t type)
5825 {
5826 	int i;
5827 
5828 	ASSERT(type < ZDB_OT_TOTAL);
5829 
5830 	if (zilog && zil_bp_tree_add(zilog, bp) != 0)
5831 		return;
5832 
5833 	/*
5834 	 * This flag controls if we will issue a claim for the block while
5835 	 * counting it, to ensure that all blocks are referenced in space maps.
5836 	 * We don't issue claims if we're not doing leak tracking, because it's
5837 	 * expensive if the user isn't interested. We also don't claim the
5838 	 * second or later occurences of cloned or dedup'd blocks, because we
5839 	 * already claimed them the first time.
5840 	 */
5841 	boolean_t do_claim = !dump_opt['L'];
5842 
5843 	spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
5844 
5845 	blkptr_t tempbp;
5846 	if (BP_GET_DEDUP(bp)) {
5847 		/*
5848 		 * Dedup'd blocks are special. We need to count them, so we can
5849 		 * later uncount them when reporting leaked space, and we must
5850 		 * only claim them once.
5851 		 *
5852 		 * We use the existing dedup system to track what we've seen.
5853 		 * The first time we see a block, we do a ddt_lookup() to see
5854 		 * if it exists in the DDT. If we're doing leak tracking, we
5855 		 * claim the block at this time.
5856 		 *
5857 		 * Each time we see a block, we reduce the refcount in the
5858 		 * entry by one, and add to the size and count of dedup'd
5859 		 * blocks to report at the end.
5860 		 */
5861 
5862 		ddt_t *ddt = ddt_select(zcb->zcb_spa, bp);
5863 
5864 		ddt_enter(ddt);
5865 
5866 		/*
5867 		 * Find the block. This will create the entry in memory, but
5868 		 * we'll know if that happened by its refcount.
5869 		 */
5870 		ddt_entry_t *dde = ddt_lookup(ddt, bp, B_TRUE);
5871 
5872 		/*
5873 		 * ddt_lookup() can return NULL if this block didn't exist
5874 		 * in the DDT and creating it would take the DDT over its
5875 		 * quota. Since we got the block from disk, it must exist in
5876 		 * the DDT, so this can't happen. However, when unique entries
5877 		 * are pruned, the dedup bit can be set with no corresponding
5878 		 * entry in the DDT.
5879 		 */
5880 		if (dde == NULL) {
5881 			ddt_exit(ddt);
5882 			goto skipped;
5883 		}
5884 
5885 		/* Get the phys for this variant */
5886 		ddt_phys_variant_t v = ddt_phys_select(ddt, dde, bp);
5887 
5888 		/*
5889 		 * This entry may have multiple sets of DVAs. We must claim
5890 		 * each set the first time we see them in a real block on disk,
5891 		 * or count them on subsequent occurences. We don't have a
5892 		 * convenient way to track the first time we see each variant,
5893 		 * so we repurpose dde_io as a set of "seen" flag bits. We can
5894 		 * do this safely in zdb because it never writes, so it will
5895 		 * never have a writing zio for this block in that pointer.
5896 		 */
5897 		boolean_t seen = !!(((uintptr_t)dde->dde_io) & (1 << v));
5898 		if (!seen)
5899 			dde->dde_io =
5900 			    (void *)(((uintptr_t)dde->dde_io) | (1 << v));
5901 
5902 		/* Consume a reference for this block. */
5903 		if (ddt_phys_total_refcnt(ddt, dde->dde_phys) > 0)
5904 			ddt_phys_decref(dde->dde_phys, v);
5905 
5906 		/*
5907 		 * If this entry has a single flat phys, it may have been
5908 		 * extended with additional DVAs at some time in its life.
5909 		 * This block might be from before it was fully extended, and
5910 		 * so have fewer DVAs.
5911 		 *
5912 		 * If this is the first time we've seen this block, and we
5913 		 * claimed it as-is, then we would miss the claim on some
5914 		 * number of DVAs, which would then be seen as leaked.
5915 		 *
5916 		 * In all cases, if we've had fewer DVAs, then the asize would
5917 		 * be too small, and would lead to the pool apparently using
5918 		 * more space than allocated.
5919 		 *
5920 		 * To handle this, we copy the canonical set of DVAs from the
5921 		 * entry back to the block pointer before we claim it.
5922 		 */
5923 		if (v == DDT_PHYS_FLAT) {
5924 			ASSERT3U(BP_GET_BIRTH(bp), ==,
5925 			    ddt_phys_birth(dde->dde_phys, v));
5926 			tempbp = *bp;
5927 			ddt_bp_fill(dde->dde_phys, v, &tempbp,
5928 			    BP_GET_BIRTH(bp));
5929 			bp = &tempbp;
5930 		}
5931 
5932 		if (seen) {
5933 			/*
5934 			 * The second or later time we see this block,
5935 			 * it's a duplicate and we count it.
5936 			 */
5937 			zcb->zcb_dedup_asize += BP_GET_ASIZE(bp);
5938 			zcb->zcb_dedup_blocks++;
5939 
5940 			/* Already claimed, don't do it again. */
5941 			do_claim = B_FALSE;
5942 		}
5943 
5944 		ddt_exit(ddt);
5945 	} else if (zcb->zcb_brt_is_active &&
5946 	    brt_maybe_exists(zcb->zcb_spa, bp)) {
5947 		/*
5948 		 * Cloned blocks are special. We need to count them, so we can
5949 		 * later uncount them when reporting leaked space, and we must
5950 		 * only claim them once.
5951 		 *
5952 		 * To do this, we keep our own in-memory BRT. For each block
5953 		 * we haven't seen before, we look it up in the real BRT and
5954 		 * if its there, we note it and its refcount then proceed as
5955 		 * normal. If we see the block again, we count it as a clone
5956 		 * and then give it no further consideration.
5957 		 */
5958 		zdb_brt_entry_t zbre_search, *zbre;
5959 		avl_index_t where;
5960 
5961 		zbre_search.zbre_dva = bp->blk_dva[0];
5962 		zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where);
5963 		if (zbre == NULL) {
5964 			/* Not seen before; track it */
5965 			uint64_t refcnt =
5966 			    brt_entry_get_refcount(zcb->zcb_spa, bp);
5967 			if (refcnt > 0) {
5968 				zbre = umem_zalloc(sizeof (zdb_brt_entry_t),
5969 				    UMEM_NOFAIL);
5970 				zbre->zbre_dva = bp->blk_dva[0];
5971 				zbre->zbre_refcount = refcnt;
5972 				avl_insert(&zcb->zcb_brt, zbre, where);
5973 			}
5974 		} else  {
5975 			/*
5976 			 * Second or later occurrence, count it and take a
5977 			 * refcount.
5978 			 */
5979 			zcb->zcb_clone_asize += BP_GET_ASIZE(bp);
5980 			zcb->zcb_clone_blocks++;
5981 
5982 			zbre->zbre_refcount--;
5983 			if (zbre->zbre_refcount == 0) {
5984 				avl_remove(&zcb->zcb_brt, zbre);
5985 				umem_free(zbre, sizeof (zdb_brt_entry_t));
5986 			}
5987 
5988 			/* Already claimed, don't do it again. */
5989 			do_claim = B_FALSE;
5990 		}
5991 	}
5992 
5993 skipped:
5994 	for (i = 0; i < 4; i++) {
5995 		int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
5996 		int t = (i & 1) ? type : ZDB_OT_TOTAL;
5997 		int equal;
5998 		zdb_blkstats_t *zb = &zcb->zcb_type[l][t];
5999 
6000 		zb->zb_asize += BP_GET_ASIZE(bp);
6001 		zb->zb_lsize += BP_GET_LSIZE(bp);
6002 		zb->zb_psize += BP_GET_PSIZE(bp);
6003 		zb->zb_count++;
6004 
6005 		/*
6006 		 * The histogram is only big enough to record blocks up to
6007 		 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
6008 		 * "other", bucket.
6009 		 */
6010 		unsigned idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT;
6011 		idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1);
6012 		zb->zb_psize_histogram[idx]++;
6013 
6014 		zb->zb_gangs += BP_COUNT_GANG(bp);
6015 
6016 		switch (BP_GET_NDVAS(bp)) {
6017 		case 2:
6018 			if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6019 			    DVA_GET_VDEV(&bp->blk_dva[1])) {
6020 				zb->zb_ditto_samevdev++;
6021 
6022 				if (same_metaslab(zcb->zcb_spa,
6023 				    DVA_GET_VDEV(&bp->blk_dva[0]),
6024 				    DVA_GET_OFFSET(&bp->blk_dva[0]),
6025 				    DVA_GET_OFFSET(&bp->blk_dva[1])))
6026 					zb->zb_ditto_same_ms++;
6027 			}
6028 			break;
6029 		case 3:
6030 			equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6031 			    DVA_GET_VDEV(&bp->blk_dva[1])) +
6032 			    (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6033 			    DVA_GET_VDEV(&bp->blk_dva[2])) +
6034 			    (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6035 			    DVA_GET_VDEV(&bp->blk_dva[2]));
6036 			if (equal != 0) {
6037 				zb->zb_ditto_samevdev++;
6038 
6039 				if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6040 				    DVA_GET_VDEV(&bp->blk_dva[1]) &&
6041 				    same_metaslab(zcb->zcb_spa,
6042 				    DVA_GET_VDEV(&bp->blk_dva[0]),
6043 				    DVA_GET_OFFSET(&bp->blk_dva[0]),
6044 				    DVA_GET_OFFSET(&bp->blk_dva[1])))
6045 					zb->zb_ditto_same_ms++;
6046 				else if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6047 				    DVA_GET_VDEV(&bp->blk_dva[2]) &&
6048 				    same_metaslab(zcb->zcb_spa,
6049 				    DVA_GET_VDEV(&bp->blk_dva[0]),
6050 				    DVA_GET_OFFSET(&bp->blk_dva[0]),
6051 				    DVA_GET_OFFSET(&bp->blk_dva[2])))
6052 					zb->zb_ditto_same_ms++;
6053 				else if (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6054 				    DVA_GET_VDEV(&bp->blk_dva[2]) &&
6055 				    same_metaslab(zcb->zcb_spa,
6056 				    DVA_GET_VDEV(&bp->blk_dva[1]),
6057 				    DVA_GET_OFFSET(&bp->blk_dva[1]),
6058 				    DVA_GET_OFFSET(&bp->blk_dva[2])))
6059 					zb->zb_ditto_same_ms++;
6060 			}
6061 			break;
6062 		}
6063 	}
6064 
6065 	spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
6066 
6067 	if (BP_IS_EMBEDDED(bp)) {
6068 		zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
6069 		zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
6070 		    [BPE_GET_PSIZE(bp)]++;
6071 		return;
6072 	}
6073 	/*
6074 	 * The binning histogram bins by powers of two up to
6075 	 * SPA_MAXBLOCKSIZE rather than creating bins for
6076 	 * every possible blocksize found in the pool.
6077 	 */
6078 	int bin = highbit64(BP_GET_PSIZE(bp)) - 1;
6079 
6080 	zcb->zcb_psize_count[bin]++;
6081 	zcb->zcb_psize_len[bin] += BP_GET_PSIZE(bp);
6082 	zcb->zcb_psize_total += BP_GET_PSIZE(bp);
6083 
6084 	bin = highbit64(BP_GET_LSIZE(bp)) - 1;
6085 
6086 	zcb->zcb_lsize_count[bin]++;
6087 	zcb->zcb_lsize_len[bin] += BP_GET_LSIZE(bp);
6088 	zcb->zcb_lsize_total += BP_GET_LSIZE(bp);
6089 
6090 	bin = highbit64(BP_GET_ASIZE(bp)) - 1;
6091 
6092 	zcb->zcb_asize_count[bin]++;
6093 	zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp);
6094 	zcb->zcb_asize_total += BP_GET_ASIZE(bp);
6095 
6096 	if (!do_claim)
6097 		return;
6098 
6099 	VERIFY0(zio_wait(zio_claim(NULL, zcb->zcb_spa,
6100 	    spa_min_claim_txg(zcb->zcb_spa), bp, NULL, NULL,
6101 	    ZIO_FLAG_CANFAIL)));
6102 }
6103 
6104 static void
zdb_blkptr_done(zio_t * zio)6105 zdb_blkptr_done(zio_t *zio)
6106 {
6107 	spa_t *spa = zio->io_spa;
6108 	blkptr_t *bp = zio->io_bp;
6109 	int ioerr = zio->io_error;
6110 	zdb_cb_t *zcb = zio->io_private;
6111 	zbookmark_phys_t *zb = &zio->io_bookmark;
6112 
6113 	mutex_enter(&spa->spa_scrub_lock);
6114 	spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
6115 	cv_broadcast(&spa->spa_scrub_io_cv);
6116 
6117 	if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
6118 		char blkbuf[BP_SPRINTF_LEN];
6119 
6120 		zcb->zcb_haderrors = 1;
6121 		zcb->zcb_errors[ioerr]++;
6122 
6123 		if (dump_opt['b'] >= 2)
6124 			snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6125 		else
6126 			blkbuf[0] = '\0';
6127 
6128 		(void) printf("zdb_blkptr_cb: "
6129 		    "Got error %d reading "
6130 		    "<%llu, %llu, %lld, %llx> %s -- skipping\n",
6131 		    ioerr,
6132 		    (u_longlong_t)zb->zb_objset,
6133 		    (u_longlong_t)zb->zb_object,
6134 		    (u_longlong_t)zb->zb_level,
6135 		    (u_longlong_t)zb->zb_blkid,
6136 		    blkbuf);
6137 	}
6138 	mutex_exit(&spa->spa_scrub_lock);
6139 
6140 	abd_free(zio->io_abd);
6141 }
6142 
6143 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)6144 zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
6145     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
6146 {
6147 	zdb_cb_t *zcb = arg;
6148 	dmu_object_type_t type;
6149 	boolean_t is_metadata;
6150 
6151 	if (zb->zb_level == ZB_DNODE_LEVEL)
6152 		return (0);
6153 
6154 	if (dump_opt['b'] >= 5 && BP_GET_LOGICAL_BIRTH(bp) > 0) {
6155 		char blkbuf[BP_SPRINTF_LEN];
6156 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6157 		(void) printf("objset %llu object %llu "
6158 		    "level %lld offset 0x%llx %s\n",
6159 		    (u_longlong_t)zb->zb_objset,
6160 		    (u_longlong_t)zb->zb_object,
6161 		    (longlong_t)zb->zb_level,
6162 		    (u_longlong_t)blkid2offset(dnp, bp, zb),
6163 		    blkbuf);
6164 	}
6165 
6166 	if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
6167 		return (0);
6168 
6169 	type = BP_GET_TYPE(bp);
6170 
6171 	zdb_count_block(zcb, zilog, bp,
6172 	    (type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type);
6173 
6174 	is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
6175 
6176 	if (!BP_IS_EMBEDDED(bp) &&
6177 	    (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
6178 		size_t size = BP_GET_PSIZE(bp);
6179 		abd_t *abd = abd_alloc(size, B_FALSE);
6180 		int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
6181 
6182 		/* If it's an intent log block, failure is expected. */
6183 		if (zb->zb_level == ZB_ZIL_LEVEL)
6184 			flags |= ZIO_FLAG_SPECULATIVE;
6185 
6186 		mutex_enter(&spa->spa_scrub_lock);
6187 		while (spa->spa_load_verify_bytes > max_inflight_bytes)
6188 			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
6189 		spa->spa_load_verify_bytes += size;
6190 		mutex_exit(&spa->spa_scrub_lock);
6191 
6192 		zio_nowait(zio_read(NULL, spa, bp, abd, size,
6193 		    zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
6194 	}
6195 
6196 	zcb->zcb_readfails = 0;
6197 
6198 	/* only call gethrtime() every 100 blocks */
6199 	static int iters;
6200 	if (++iters > 100)
6201 		iters = 0;
6202 	else
6203 		return (0);
6204 
6205 	if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
6206 		uint64_t now = gethrtime();
6207 		char buf[10];
6208 		uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
6209 		uint64_t kb_per_sec =
6210 		    1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000));
6211 		uint64_t sec_remaining =
6212 		    (zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec;
6213 
6214 		/* make sure nicenum has enough space */
6215 		_Static_assert(sizeof (buf) >= NN_NUMBUF_SZ, "buf truncated");
6216 
6217 		zfs_nicebytes(bytes, buf, sizeof (buf));
6218 		(void) fprintf(stderr,
6219 		    "\r%5s completed (%4"PRIu64"MB/s) "
6220 		    "estimated time remaining: "
6221 		    "%"PRIu64"hr %02"PRIu64"min %02"PRIu64"sec        ",
6222 		    buf, kb_per_sec / 1024,
6223 		    sec_remaining / 60 / 60,
6224 		    sec_remaining / 60 % 60,
6225 		    sec_remaining % 60);
6226 
6227 		zcb->zcb_lastprint = now;
6228 	}
6229 
6230 	return (0);
6231 }
6232 
6233 static void
zdb_leak(void * arg,uint64_t start,uint64_t size)6234 zdb_leak(void *arg, uint64_t start, uint64_t size)
6235 {
6236 	vdev_t *vd = arg;
6237 
6238 	(void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
6239 	    (u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
6240 }
6241 
6242 static metaslab_ops_t zdb_metaslab_ops = {
6243 	NULL	/* alloc */
6244 };
6245 
6246 static int
load_unflushed_svr_segs_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6247 load_unflushed_svr_segs_cb(spa_t *spa, space_map_entry_t *sme,
6248     uint64_t txg, void *arg)
6249 {
6250 	spa_vdev_removal_t *svr = arg;
6251 
6252 	uint64_t offset = sme->sme_offset;
6253 	uint64_t size = sme->sme_run;
6254 
6255 	/* skip vdevs we don't care about */
6256 	if (sme->sme_vdev != svr->svr_vdev_id)
6257 		return (0);
6258 
6259 	vdev_t *vd = vdev_lookup_top(spa, sme->sme_vdev);
6260 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6261 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6262 
6263 	if (txg < metaslab_unflushed_txg(ms))
6264 		return (0);
6265 
6266 	if (sme->sme_type == SM_ALLOC)
6267 		zfs_range_tree_add(svr->svr_allocd_segs, offset, size);
6268 	else
6269 		zfs_range_tree_remove(svr->svr_allocd_segs, offset, size);
6270 
6271 	return (0);
6272 }
6273 
6274 static void
claim_segment_impl_cb(uint64_t inner_offset,vdev_t * vd,uint64_t offset,uint64_t size,void * arg)6275 claim_segment_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
6276     uint64_t size, void *arg)
6277 {
6278 	(void) inner_offset, (void) arg;
6279 
6280 	/*
6281 	 * This callback was called through a remap from
6282 	 * a device being removed. Therefore, the vdev that
6283 	 * this callback is applied to is a concrete
6284 	 * vdev.
6285 	 */
6286 	ASSERT(vdev_is_concrete(vd));
6287 
6288 	VERIFY0(metaslab_claim_impl(vd, offset, size,
6289 	    spa_min_claim_txg(vd->vdev_spa)));
6290 }
6291 
6292 static void
claim_segment_cb(void * arg,uint64_t offset,uint64_t size)6293 claim_segment_cb(void *arg, uint64_t offset, uint64_t size)
6294 {
6295 	vdev_t *vd = arg;
6296 
6297 	vdev_indirect_ops.vdev_op_remap(vd, offset, size,
6298 	    claim_segment_impl_cb, NULL);
6299 }
6300 
6301 /*
6302  * After accounting for all allocated blocks that are directly referenced,
6303  * we might have missed a reference to a block from a partially complete
6304  * (and thus unused) indirect mapping object. We perform a secondary pass
6305  * through the metaslabs we have already mapped and claim the destination
6306  * blocks.
6307  */
6308 static void
zdb_claim_removing(spa_t * spa,zdb_cb_t * zcb)6309 zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb)
6310 {
6311 	if (dump_opt['L'])
6312 		return;
6313 
6314 	if (spa->spa_vdev_removal == NULL)
6315 		return;
6316 
6317 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6318 
6319 	spa_vdev_removal_t *svr = spa->spa_vdev_removal;
6320 	vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id);
6321 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6322 
6323 	ASSERT0(zfs_range_tree_space(svr->svr_allocd_segs));
6324 
6325 	zfs_range_tree_t *allocs = zfs_range_tree_create(NULL, ZFS_RANGE_SEG64,
6326 	    NULL, 0, 0);
6327 	for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) {
6328 		metaslab_t *msp = vd->vdev_ms[msi];
6329 
6330 		ASSERT0(zfs_range_tree_space(allocs));
6331 		if (msp->ms_sm != NULL)
6332 			VERIFY0(space_map_load(msp->ms_sm, allocs, SM_ALLOC));
6333 		zfs_range_tree_vacate(allocs, zfs_range_tree_add,
6334 		    svr->svr_allocd_segs);
6335 	}
6336 	zfs_range_tree_destroy(allocs);
6337 
6338 	iterate_through_spacemap_logs(spa, load_unflushed_svr_segs_cb, svr);
6339 
6340 	/*
6341 	 * Clear everything past what has been synced,
6342 	 * because we have not allocated mappings for
6343 	 * it yet.
6344 	 */
6345 	zfs_range_tree_clear(svr->svr_allocd_segs,
6346 	    vdev_indirect_mapping_max_offset(vim),
6347 	    vd->vdev_asize - vdev_indirect_mapping_max_offset(vim));
6348 
6349 	zcb->zcb_removing_size += zfs_range_tree_space(svr->svr_allocd_segs);
6350 	zfs_range_tree_vacate(svr->svr_allocd_segs, claim_segment_cb, vd);
6351 
6352 	spa_config_exit(spa, SCL_CONFIG, FTAG);
6353 }
6354 
6355 static int
increment_indirect_mapping_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6356 increment_indirect_mapping_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6357     dmu_tx_t *tx)
6358 {
6359 	(void) tx;
6360 	zdb_cb_t *zcb = arg;
6361 	spa_t *spa = zcb->zcb_spa;
6362 	vdev_t *vd;
6363 	const dva_t *dva = &bp->blk_dva[0];
6364 
6365 	ASSERT(!bp_freed);
6366 	ASSERT(!dump_opt['L']);
6367 	ASSERT3U(BP_GET_NDVAS(bp), ==, 1);
6368 
6369 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6370 	vd = vdev_lookup_top(zcb->zcb_spa, DVA_GET_VDEV(dva));
6371 	ASSERT3P(vd, !=, NULL);
6372 	spa_config_exit(spa, SCL_VDEV, FTAG);
6373 
6374 	ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0);
6375 	ASSERT3P(zcb->zcb_vd_obsolete_counts[vd->vdev_id], !=, NULL);
6376 
6377 	vdev_indirect_mapping_increment_obsolete_count(
6378 	    vd->vdev_indirect_mapping,
6379 	    DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva),
6380 	    zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6381 
6382 	return (0);
6383 }
6384 
6385 static uint32_t *
zdb_load_obsolete_counts(vdev_t * vd)6386 zdb_load_obsolete_counts(vdev_t *vd)
6387 {
6388 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6389 	spa_t *spa = vd->vdev_spa;
6390 	spa_condensing_indirect_phys_t *scip =
6391 	    &spa->spa_condensing_indirect_phys;
6392 	uint64_t obsolete_sm_object;
6393 	uint32_t *counts;
6394 
6395 	VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
6396 	EQUIV(obsolete_sm_object != 0, vd->vdev_obsolete_sm != NULL);
6397 	counts = vdev_indirect_mapping_load_obsolete_counts(vim);
6398 	if (vd->vdev_obsolete_sm != NULL) {
6399 		vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6400 		    vd->vdev_obsolete_sm);
6401 	}
6402 	if (scip->scip_vdev == vd->vdev_id &&
6403 	    scip->scip_prev_obsolete_sm_object != 0) {
6404 		space_map_t *prev_obsolete_sm = NULL;
6405 		VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset,
6406 		    scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0));
6407 		vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6408 		    prev_obsolete_sm);
6409 		space_map_close(prev_obsolete_sm);
6410 	}
6411 	return (counts);
6412 }
6413 
6414 typedef struct checkpoint_sm_exclude_entry_arg {
6415 	vdev_t *cseea_vd;
6416 	uint64_t cseea_checkpoint_size;
6417 } checkpoint_sm_exclude_entry_arg_t;
6418 
6419 static int
checkpoint_sm_exclude_entry_cb(space_map_entry_t * sme,void * arg)6420 checkpoint_sm_exclude_entry_cb(space_map_entry_t *sme, void *arg)
6421 {
6422 	checkpoint_sm_exclude_entry_arg_t *cseea = arg;
6423 	vdev_t *vd = cseea->cseea_vd;
6424 	metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
6425 	uint64_t end = sme->sme_offset + sme->sme_run;
6426 
6427 	ASSERT(sme->sme_type == SM_FREE);
6428 
6429 	/*
6430 	 * Since the vdev_checkpoint_sm exists in the vdev level
6431 	 * and the ms_sm space maps exist in the metaslab level,
6432 	 * an entry in the checkpoint space map could theoretically
6433 	 * cross the boundaries of the metaslab that it belongs.
6434 	 *
6435 	 * In reality, because of the way that we populate and
6436 	 * manipulate the checkpoint's space maps currently,
6437 	 * there shouldn't be any entries that cross metaslabs.
6438 	 * Hence the assertion below.
6439 	 *
6440 	 * That said, there is no fundamental requirement that
6441 	 * the checkpoint's space map entries should not cross
6442 	 * metaslab boundaries. So if needed we could add code
6443 	 * that handles metaslab-crossing segments in the future.
6444 	 */
6445 	VERIFY3U(sme->sme_offset, >=, ms->ms_start);
6446 	VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
6447 
6448 	/*
6449 	 * By removing the entry from the allocated segments we
6450 	 * also verify that the entry is there to begin with.
6451 	 */
6452 	mutex_enter(&ms->ms_lock);
6453 	zfs_range_tree_remove(ms->ms_allocatable, sme->sme_offset,
6454 	    sme->sme_run);
6455 	mutex_exit(&ms->ms_lock);
6456 
6457 	cseea->cseea_checkpoint_size += sme->sme_run;
6458 	return (0);
6459 }
6460 
6461 static void
zdb_leak_init_vdev_exclude_checkpoint(vdev_t * vd,zdb_cb_t * zcb)6462 zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_cb_t *zcb)
6463 {
6464 	spa_t *spa = vd->vdev_spa;
6465 	space_map_t *checkpoint_sm = NULL;
6466 	uint64_t checkpoint_sm_obj;
6467 
6468 	/*
6469 	 * If there is no vdev_top_zap, we are in a pool whose
6470 	 * version predates the pool checkpoint feature.
6471 	 */
6472 	if (vd->vdev_top_zap == 0)
6473 		return;
6474 
6475 	/*
6476 	 * If there is no reference of the vdev_checkpoint_sm in
6477 	 * the vdev_top_zap, then one of the following scenarios
6478 	 * is true:
6479 	 *
6480 	 * 1] There is no checkpoint
6481 	 * 2] There is a checkpoint, but no checkpointed blocks
6482 	 *    have been freed yet
6483 	 * 3] The current vdev is indirect
6484 	 *
6485 	 * In these cases we return immediately.
6486 	 */
6487 	if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
6488 	    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
6489 		return;
6490 
6491 	VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
6492 	    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1,
6493 	    &checkpoint_sm_obj));
6494 
6495 	checkpoint_sm_exclude_entry_arg_t cseea;
6496 	cseea.cseea_vd = vd;
6497 	cseea.cseea_checkpoint_size = 0;
6498 
6499 	VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
6500 	    checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
6501 
6502 	VERIFY0(space_map_iterate(checkpoint_sm,
6503 	    space_map_length(checkpoint_sm),
6504 	    checkpoint_sm_exclude_entry_cb, &cseea));
6505 	space_map_close(checkpoint_sm);
6506 
6507 	zcb->zcb_checkpoint_size += cseea.cseea_checkpoint_size;
6508 }
6509 
6510 static void
zdb_leak_init_exclude_checkpoint(spa_t * spa,zdb_cb_t * zcb)6511 zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb)
6512 {
6513 	ASSERT(!dump_opt['L']);
6514 
6515 	vdev_t *rvd = spa->spa_root_vdev;
6516 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6517 		ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id);
6518 		zdb_leak_init_vdev_exclude_checkpoint(rvd->vdev_child[c], zcb);
6519 	}
6520 }
6521 
6522 static int
count_unflushed_space_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6523 count_unflushed_space_cb(spa_t *spa, space_map_entry_t *sme,
6524     uint64_t txg, void *arg)
6525 {
6526 	int64_t *ualloc_space = arg;
6527 
6528 	uint64_t offset = sme->sme_offset;
6529 	uint64_t vdev_id = sme->sme_vdev;
6530 
6531 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6532 	if (!vdev_is_concrete(vd))
6533 		return (0);
6534 
6535 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6536 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6537 
6538 	if (txg < metaslab_unflushed_txg(ms))
6539 		return (0);
6540 
6541 	if (sme->sme_type == SM_ALLOC)
6542 		*ualloc_space += sme->sme_run;
6543 	else
6544 		*ualloc_space -= sme->sme_run;
6545 
6546 	return (0);
6547 }
6548 
6549 static int64_t
get_unflushed_alloc_space(spa_t * spa)6550 get_unflushed_alloc_space(spa_t *spa)
6551 {
6552 	if (dump_opt['L'])
6553 		return (0);
6554 
6555 	int64_t ualloc_space = 0;
6556 	iterate_through_spacemap_logs(spa, count_unflushed_space_cb,
6557 	    &ualloc_space);
6558 	return (ualloc_space);
6559 }
6560 
6561 static int
load_unflushed_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6562 load_unflushed_cb(spa_t *spa, space_map_entry_t *sme, uint64_t txg, void *arg)
6563 {
6564 	maptype_t *uic_maptype = arg;
6565 
6566 	uint64_t offset = sme->sme_offset;
6567 	uint64_t size = sme->sme_run;
6568 	uint64_t vdev_id = sme->sme_vdev;
6569 
6570 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6571 
6572 	/* skip indirect vdevs */
6573 	if (!vdev_is_concrete(vd))
6574 		return (0);
6575 
6576 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6577 
6578 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6579 	ASSERT(*uic_maptype == SM_ALLOC || *uic_maptype == SM_FREE);
6580 
6581 	if (txg < metaslab_unflushed_txg(ms))
6582 		return (0);
6583 
6584 	if (*uic_maptype == sme->sme_type)
6585 		zfs_range_tree_add(ms->ms_allocatable, offset, size);
6586 	else
6587 		zfs_range_tree_remove(ms->ms_allocatable, offset, size);
6588 
6589 	return (0);
6590 }
6591 
6592 static void
load_unflushed_to_ms_allocatables(spa_t * spa,maptype_t maptype)6593 load_unflushed_to_ms_allocatables(spa_t *spa, maptype_t maptype)
6594 {
6595 	iterate_through_spacemap_logs(spa, load_unflushed_cb, &maptype);
6596 }
6597 
6598 static void
load_concrete_ms_allocatable_trees(spa_t * spa,maptype_t maptype)6599 load_concrete_ms_allocatable_trees(spa_t *spa, maptype_t maptype)
6600 {
6601 	vdev_t *rvd = spa->spa_root_vdev;
6602 	for (uint64_t i = 0; i < rvd->vdev_children; i++) {
6603 		vdev_t *vd = rvd->vdev_child[i];
6604 
6605 		ASSERT3U(i, ==, vd->vdev_id);
6606 
6607 		if (vd->vdev_ops == &vdev_indirect_ops)
6608 			continue;
6609 
6610 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6611 			metaslab_t *msp = vd->vdev_ms[m];
6612 
6613 			(void) fprintf(stderr,
6614 			    "\rloading concrete vdev %llu, "
6615 			    "metaslab %llu of %llu ...",
6616 			    (longlong_t)vd->vdev_id,
6617 			    (longlong_t)msp->ms_id,
6618 			    (longlong_t)vd->vdev_ms_count);
6619 
6620 			mutex_enter(&msp->ms_lock);
6621 			zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6622 
6623 			/*
6624 			 * We don't want to spend the CPU manipulating the
6625 			 * size-ordered tree, so clear the range_tree ops.
6626 			 */
6627 			msp->ms_allocatable->rt_ops = NULL;
6628 
6629 			if (msp->ms_sm != NULL) {
6630 				VERIFY0(space_map_load(msp->ms_sm,
6631 				    msp->ms_allocatable, maptype));
6632 			}
6633 			if (!msp->ms_loaded)
6634 				msp->ms_loaded = B_TRUE;
6635 			mutex_exit(&msp->ms_lock);
6636 		}
6637 	}
6638 
6639 	load_unflushed_to_ms_allocatables(spa, maptype);
6640 }
6641 
6642 /*
6643  * vm_idxp is an in-out parameter which (for indirect vdevs) is the
6644  * index in vim_entries that has the first entry in this metaslab.
6645  * On return, it will be set to the first entry after this metaslab.
6646  */
6647 static void
load_indirect_ms_allocatable_tree(vdev_t * vd,metaslab_t * msp,uint64_t * vim_idxp)6648 load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab_t *msp,
6649     uint64_t *vim_idxp)
6650 {
6651 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6652 
6653 	mutex_enter(&msp->ms_lock);
6654 	zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6655 
6656 	/*
6657 	 * We don't want to spend the CPU manipulating the
6658 	 * size-ordered tree, so clear the range_tree ops.
6659 	 */
6660 	msp->ms_allocatable->rt_ops = NULL;
6661 
6662 	for (; *vim_idxp < vdev_indirect_mapping_num_entries(vim);
6663 	    (*vim_idxp)++) {
6664 		vdev_indirect_mapping_entry_phys_t *vimep =
6665 		    &vim->vim_entries[*vim_idxp];
6666 		uint64_t ent_offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6667 		uint64_t ent_len = DVA_GET_ASIZE(&vimep->vimep_dst);
6668 		ASSERT3U(ent_offset, >=, msp->ms_start);
6669 		if (ent_offset >= msp->ms_start + msp->ms_size)
6670 			break;
6671 
6672 		/*
6673 		 * Mappings do not cross metaslab boundaries,
6674 		 * because we create them by walking the metaslabs.
6675 		 */
6676 		ASSERT3U(ent_offset + ent_len, <=,
6677 		    msp->ms_start + msp->ms_size);
6678 		zfs_range_tree_add(msp->ms_allocatable, ent_offset, ent_len);
6679 	}
6680 
6681 	if (!msp->ms_loaded)
6682 		msp->ms_loaded = B_TRUE;
6683 	mutex_exit(&msp->ms_lock);
6684 }
6685 
6686 static void
zdb_leak_init_prepare_indirect_vdevs(spa_t * spa,zdb_cb_t * zcb)6687 zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb)
6688 {
6689 	ASSERT(!dump_opt['L']);
6690 
6691 	vdev_t *rvd = spa->spa_root_vdev;
6692 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6693 		vdev_t *vd = rvd->vdev_child[c];
6694 
6695 		ASSERT3U(c, ==, vd->vdev_id);
6696 
6697 		if (vd->vdev_ops != &vdev_indirect_ops)
6698 			continue;
6699 
6700 		/*
6701 		 * Note: we don't check for mapping leaks on
6702 		 * removing vdevs because their ms_allocatable's
6703 		 * are used to look for leaks in allocated space.
6704 		 */
6705 		zcb->zcb_vd_obsolete_counts[c] = zdb_load_obsolete_counts(vd);
6706 
6707 		/*
6708 		 * Normally, indirect vdevs don't have any
6709 		 * metaslabs.  We want to set them up for
6710 		 * zio_claim().
6711 		 */
6712 		vdev_metaslab_group_create(vd);
6713 		VERIFY0(vdev_metaslab_init(vd, 0));
6714 
6715 		vdev_indirect_mapping_t *vim __maybe_unused =
6716 		    vd->vdev_indirect_mapping;
6717 		uint64_t vim_idx = 0;
6718 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6719 
6720 			(void) fprintf(stderr,
6721 			    "\rloading indirect vdev %llu, "
6722 			    "metaslab %llu of %llu ...",
6723 			    (longlong_t)vd->vdev_id,
6724 			    (longlong_t)vd->vdev_ms[m]->ms_id,
6725 			    (longlong_t)vd->vdev_ms_count);
6726 
6727 			load_indirect_ms_allocatable_tree(vd, vd->vdev_ms[m],
6728 			    &vim_idx);
6729 		}
6730 		ASSERT3U(vim_idx, ==, vdev_indirect_mapping_num_entries(vim));
6731 	}
6732 }
6733 
6734 static void
zdb_leak_init(spa_t * spa,zdb_cb_t * zcb)6735 zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
6736 {
6737 	zcb->zcb_spa = spa;
6738 
6739 	if (dump_opt['L'])
6740 		return;
6741 
6742 	dsl_pool_t *dp = spa->spa_dsl_pool;
6743 	vdev_t *rvd = spa->spa_root_vdev;
6744 
6745 	/*
6746 	 * We are going to be changing the meaning of the metaslab's
6747 	 * ms_allocatable.  Ensure that the allocator doesn't try to
6748 	 * use the tree.
6749 	 */
6750 	spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
6751 	spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
6752 	spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6753 
6754 	zcb->zcb_vd_obsolete_counts =
6755 	    umem_zalloc(rvd->vdev_children * sizeof (uint32_t *),
6756 	    UMEM_NOFAIL);
6757 
6758 	/*
6759 	 * For leak detection, we overload the ms_allocatable trees
6760 	 * to contain allocated segments instead of free segments.
6761 	 * As a result, we can't use the normal metaslab_load/unload
6762 	 * interfaces.
6763 	 */
6764 	zdb_leak_init_prepare_indirect_vdevs(spa, zcb);
6765 	load_concrete_ms_allocatable_trees(spa, SM_ALLOC);
6766 
6767 	/*
6768 	 * On load_concrete_ms_allocatable_trees() we loaded all the
6769 	 * allocated entries from the ms_sm to the ms_allocatable for
6770 	 * each metaslab. If the pool has a checkpoint or is in the
6771 	 * middle of discarding a checkpoint, some of these blocks
6772 	 * may have been freed but their ms_sm may not have been
6773 	 * updated because they are referenced by the checkpoint. In
6774 	 * order to avoid false-positives during leak-detection, we
6775 	 * go through the vdev's checkpoint space map and exclude all
6776 	 * its entries from their relevant ms_allocatable.
6777 	 *
6778 	 * We also aggregate the space held by the checkpoint and add
6779 	 * it to zcb_checkpoint_size.
6780 	 *
6781 	 * Note that at this point we are also verifying that all the
6782 	 * entries on the checkpoint_sm are marked as allocated in
6783 	 * the ms_sm of their relevant metaslab.
6784 	 * [see comment in checkpoint_sm_exclude_entry_cb()]
6785 	 */
6786 	zdb_leak_init_exclude_checkpoint(spa, zcb);
6787 	ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa));
6788 
6789 	/* for cleaner progress output */
6790 	(void) fprintf(stderr, "\n");
6791 
6792 	if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
6793 		ASSERT(spa_feature_is_enabled(spa,
6794 		    SPA_FEATURE_DEVICE_REMOVAL));
6795 		(void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj,
6796 		    increment_indirect_mapping_cb, zcb, NULL);
6797 	}
6798 }
6799 
6800 static boolean_t
zdb_check_for_obsolete_leaks(vdev_t * vd,zdb_cb_t * zcb)6801 zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb)
6802 {
6803 	boolean_t leaks = B_FALSE;
6804 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6805 	uint64_t total_leaked = 0;
6806 	boolean_t are_precise = B_FALSE;
6807 
6808 	ASSERT(vim != NULL);
6809 
6810 	for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
6811 		vdev_indirect_mapping_entry_phys_t *vimep =
6812 		    &vim->vim_entries[i];
6813 		uint64_t obsolete_bytes = 0;
6814 		uint64_t offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6815 		metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6816 
6817 		/*
6818 		 * This is not very efficient but it's easy to
6819 		 * verify correctness.
6820 		 */
6821 		for (uint64_t inner_offset = 0;
6822 		    inner_offset < DVA_GET_ASIZE(&vimep->vimep_dst);
6823 		    inner_offset += 1ULL << vd->vdev_ashift) {
6824 			if (zfs_range_tree_contains(msp->ms_allocatable,
6825 			    offset + inner_offset, 1ULL << vd->vdev_ashift)) {
6826 				obsolete_bytes += 1ULL << vd->vdev_ashift;
6827 			}
6828 		}
6829 
6830 		int64_t bytes_leaked = obsolete_bytes -
6831 		    zcb->zcb_vd_obsolete_counts[vd->vdev_id][i];
6832 		ASSERT3U(DVA_GET_ASIZE(&vimep->vimep_dst), >=,
6833 		    zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]);
6834 
6835 		VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6836 		if (bytes_leaked != 0 && (are_precise || dump_opt['d'] >= 5)) {
6837 			(void) printf("obsolete indirect mapping count "
6838 			    "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
6839 			    (u_longlong_t)vd->vdev_id,
6840 			    (u_longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
6841 			    (u_longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
6842 			    (u_longlong_t)bytes_leaked);
6843 		}
6844 		total_leaked += ABS(bytes_leaked);
6845 	}
6846 
6847 	VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6848 	if (!are_precise && total_leaked > 0) {
6849 		int pct_leaked = total_leaked * 100 /
6850 		    vdev_indirect_mapping_bytes_mapped(vim);
6851 		(void) printf("cannot verify obsolete indirect mapping "
6852 		    "counts of vdev %llu because precise feature was not "
6853 		    "enabled when it was removed: %d%% (%llx bytes) of mapping"
6854 		    "unreferenced\n",
6855 		    (u_longlong_t)vd->vdev_id, pct_leaked,
6856 		    (u_longlong_t)total_leaked);
6857 	} else if (total_leaked > 0) {
6858 		(void) printf("obsolete indirect mapping count mismatch "
6859 		    "for vdev %llu -- %llx total bytes mismatched\n",
6860 		    (u_longlong_t)vd->vdev_id,
6861 		    (u_longlong_t)total_leaked);
6862 		leaks |= B_TRUE;
6863 	}
6864 
6865 	vdev_indirect_mapping_free_obsolete_counts(vim,
6866 	    zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6867 	zcb->zcb_vd_obsolete_counts[vd->vdev_id] = NULL;
6868 
6869 	return (leaks);
6870 }
6871 
6872 static boolean_t
zdb_leak_fini(spa_t * spa,zdb_cb_t * zcb)6873 zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb)
6874 {
6875 	if (dump_opt['L'])
6876 		return (B_FALSE);
6877 
6878 	boolean_t leaks = B_FALSE;
6879 	vdev_t *rvd = spa->spa_root_vdev;
6880 	for (unsigned c = 0; c < rvd->vdev_children; c++) {
6881 		vdev_t *vd = rvd->vdev_child[c];
6882 
6883 		if (zcb->zcb_vd_obsolete_counts[c] != NULL) {
6884 			leaks |= zdb_check_for_obsolete_leaks(vd, zcb);
6885 		}
6886 
6887 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6888 			metaslab_t *msp = vd->vdev_ms[m];
6889 			ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class ==
6890 			    spa_embedded_log_class(spa)) ?
6891 			    vd->vdev_log_mg : vd->vdev_mg);
6892 
6893 			/*
6894 			 * ms_allocatable has been overloaded
6895 			 * to contain allocated segments. Now that
6896 			 * we finished traversing all blocks, any
6897 			 * block that remains in the ms_allocatable
6898 			 * represents an allocated block that we
6899 			 * did not claim during the traversal.
6900 			 * Claimed blocks would have been removed
6901 			 * from the ms_allocatable.  For indirect
6902 			 * vdevs, space remaining in the tree
6903 			 * represents parts of the mapping that are
6904 			 * not referenced, which is not a bug.
6905 			 */
6906 			if (vd->vdev_ops == &vdev_indirect_ops) {
6907 				zfs_range_tree_vacate(msp->ms_allocatable,
6908 				    NULL, NULL);
6909 			} else {
6910 				zfs_range_tree_vacate(msp->ms_allocatable,
6911 				    zdb_leak, vd);
6912 			}
6913 			if (msp->ms_loaded) {
6914 				msp->ms_loaded = B_FALSE;
6915 			}
6916 		}
6917 	}
6918 
6919 	umem_free(zcb->zcb_vd_obsolete_counts,
6920 	    rvd->vdev_children * sizeof (uint32_t *));
6921 	zcb->zcb_vd_obsolete_counts = NULL;
6922 
6923 	return (leaks);
6924 }
6925 
6926 static int
count_block_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)6927 count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6928 {
6929 	(void) tx;
6930 	zdb_cb_t *zcb = arg;
6931 
6932 	if (dump_opt['b'] >= 5) {
6933 		char blkbuf[BP_SPRINTF_LEN];
6934 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6935 		(void) printf("[%s] %s\n",
6936 		    "deferred free", blkbuf);
6937 	}
6938 	zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED);
6939 	return (0);
6940 }
6941 
6942 /*
6943  * Iterate over livelists which have been destroyed by the user but
6944  * are still present in the MOS, waiting to be freed
6945  */
6946 static void
iterate_deleted_livelists(spa_t * spa,ll_iter_t func,void * arg)6947 iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg)
6948 {
6949 	objset_t *mos = spa->spa_meta_objset;
6950 	uint64_t zap_obj;
6951 	int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
6952 	    DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
6953 	if (err == ENOENT)
6954 		return;
6955 	ASSERT0(err);
6956 
6957 	zap_cursor_t zc;
6958 	zap_attribute_t *attrp = zap_attribute_alloc();
6959 	dsl_deadlist_t ll;
6960 	/* NULL out os prior to dsl_deadlist_open in case it's garbage */
6961 	ll.dl_os = NULL;
6962 	for (zap_cursor_init(&zc, mos, zap_obj);
6963 	    zap_cursor_retrieve(&zc, attrp) == 0;
6964 	    (void) zap_cursor_advance(&zc)) {
6965 		VERIFY0(dsl_deadlist_open(&ll, mos, attrp->za_first_integer));
6966 		func(&ll, arg);
6967 		dsl_deadlist_close(&ll);
6968 	}
6969 	zap_cursor_fini(&zc);
6970 	zap_attribute_free(attrp);
6971 }
6972 
6973 static int
bpobj_count_block_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6974 bpobj_count_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6975     dmu_tx_t *tx)
6976 {
6977 	ASSERT(!bp_freed);
6978 	return (count_block_cb(arg, bp, tx));
6979 }
6980 
6981 static int
livelist_entry_count_blocks_cb(void * args,dsl_deadlist_entry_t * dle)6982 livelist_entry_count_blocks_cb(void *args, dsl_deadlist_entry_t *dle)
6983 {
6984 	zdb_cb_t *zbc = args;
6985 	bplist_t blks;
6986 	bplist_create(&blks);
6987 	/* determine which blocks have been alloc'd but not freed */
6988 	VERIFY0(dsl_process_sub_livelist(&dle->dle_bpobj, &blks, NULL, NULL));
6989 	/* count those blocks */
6990 	(void) bplist_iterate(&blks, count_block_cb, zbc, NULL);
6991 	bplist_destroy(&blks);
6992 	return (0);
6993 }
6994 
6995 static void
livelist_count_blocks(dsl_deadlist_t * ll,void * arg)6996 livelist_count_blocks(dsl_deadlist_t *ll, void *arg)
6997 {
6998 	dsl_deadlist_iterate(ll, livelist_entry_count_blocks_cb, arg);
6999 }
7000 
7001 /*
7002  * Count the blocks in the livelists that have been destroyed by the user
7003  * but haven't yet been freed.
7004  */
7005 static void
deleted_livelists_count_blocks(spa_t * spa,zdb_cb_t * zbc)7006 deleted_livelists_count_blocks(spa_t *spa, zdb_cb_t *zbc)
7007 {
7008 	iterate_deleted_livelists(spa, livelist_count_blocks, zbc);
7009 }
7010 
7011 static void
dump_livelist_cb(dsl_deadlist_t * ll,void * arg)7012 dump_livelist_cb(dsl_deadlist_t *ll, void *arg)
7013 {
7014 	ASSERT3P(arg, ==, NULL);
7015 	global_feature_count[SPA_FEATURE_LIVELIST]++;
7016 	dump_blkptr_list(ll, "Deleted Livelist");
7017 	dsl_deadlist_iterate(ll, sublivelist_verify_lightweight, NULL);
7018 }
7019 
7020 /*
7021  * Print out, register object references to, and increment feature counts for
7022  * livelists that have been destroyed by the user but haven't yet been freed.
7023  */
7024 static void
deleted_livelists_dump_mos(spa_t * spa)7025 deleted_livelists_dump_mos(spa_t *spa)
7026 {
7027 	uint64_t zap_obj;
7028 	objset_t *mos = spa->spa_meta_objset;
7029 	int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
7030 	    DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
7031 	if (err == ENOENT)
7032 		return;
7033 	mos_obj_refd(zap_obj);
7034 	iterate_deleted_livelists(spa, dump_livelist_cb, NULL);
7035 }
7036 
7037 static int
zdb_brt_entry_compare(const void * zcn1,const void * zcn2)7038 zdb_brt_entry_compare(const void *zcn1, const void *zcn2)
7039 {
7040 	const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva;
7041 	const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva;
7042 	int cmp;
7043 
7044 	cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
7045 	if (cmp == 0)
7046 		cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2));
7047 
7048 	return (cmp);
7049 }
7050 
7051 static int
dump_block_stats(spa_t * spa)7052 dump_block_stats(spa_t *spa)
7053 {
7054 	zdb_cb_t *zcb;
7055 	zdb_blkstats_t *zb, *tzb;
7056 	uint64_t norm_alloc, norm_space, total_alloc, total_found;
7057 	int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7058 	    TRAVERSE_NO_DECRYPT | TRAVERSE_HARD;
7059 	boolean_t leaks = B_FALSE;
7060 	int e, c, err;
7061 	bp_embedded_type_t i;
7062 
7063 	ddt_prefetch_all(spa);
7064 
7065 	zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL);
7066 
7067 	if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
7068 		avl_create(&zcb->zcb_brt, zdb_brt_entry_compare,
7069 		    sizeof (zdb_brt_entry_t),
7070 		    offsetof(zdb_brt_entry_t, zbre_node));
7071 		zcb->zcb_brt_is_active = B_TRUE;
7072 	}
7073 
7074 	(void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
7075 	    (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
7076 	    (dump_opt['c'] == 1) ? "metadata " : "",
7077 	    dump_opt['c'] ? "checksums " : "",
7078 	    (dump_opt['c'] && !dump_opt['L']) ? "and verify " : "",
7079 	    !dump_opt['L'] ? "nothing leaked " : "");
7080 
7081 	/*
7082 	 * When leak detection is enabled we load all space maps as SM_ALLOC
7083 	 * maps, then traverse the pool claiming each block we discover. If
7084 	 * the pool is perfectly consistent, the segment trees will be empty
7085 	 * when we're done. Anything left over is a leak; any block we can't
7086 	 * claim (because it's not part of any space map) is a double
7087 	 * allocation, reference to a freed block, or an unclaimed log block.
7088 	 *
7089 	 * When leak detection is disabled (-L option) we still traverse the
7090 	 * pool claiming each block we discover, but we skip opening any space
7091 	 * maps.
7092 	 */
7093 	zdb_leak_init(spa, zcb);
7094 
7095 	/*
7096 	 * If there's a deferred-free bplist, process that first.
7097 	 */
7098 	(void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
7099 	    bpobj_count_block_cb, zcb, NULL);
7100 
7101 	if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
7102 		(void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
7103 		    bpobj_count_block_cb, zcb, NULL);
7104 	}
7105 
7106 	zdb_claim_removing(spa, zcb);
7107 
7108 	if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
7109 		VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,
7110 		    spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb,
7111 		    zcb, NULL));
7112 	}
7113 
7114 	deleted_livelists_count_blocks(spa, zcb);
7115 
7116 	if (dump_opt['c'] > 1)
7117 		flags |= TRAVERSE_PREFETCH_DATA;
7118 
7119 	zcb->zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
7120 	zcb->zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa));
7121 	zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
7122 	zcb->zcb_totalasize +=
7123 	    metaslab_class_get_alloc(spa_embedded_log_class(spa));
7124 	zcb->zcb_start = zcb->zcb_lastprint = gethrtime();
7125 	err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb);
7126 
7127 	/*
7128 	 * If we've traversed the data blocks then we need to wait for those
7129 	 * I/Os to complete. We leverage "The Godfather" zio to wait on
7130 	 * all async I/Os to complete.
7131 	 */
7132 	if (dump_opt['c']) {
7133 		for (c = 0; c < max_ncpus; c++) {
7134 			(void) zio_wait(spa->spa_async_zio_root[c]);
7135 			spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL,
7136 			    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
7137 			    ZIO_FLAG_GODFATHER);
7138 		}
7139 	}
7140 	ASSERT0(spa->spa_load_verify_bytes);
7141 
7142 	/*
7143 	 * Done after zio_wait() since zcb_haderrors is modified in
7144 	 * zdb_blkptr_done()
7145 	 */
7146 	zcb->zcb_haderrors |= err;
7147 
7148 	if (zcb->zcb_haderrors) {
7149 		(void) printf("\nError counts:\n\n");
7150 		(void) printf("\t%5s  %s\n", "errno", "count");
7151 		for (e = 0; e < 256; e++) {
7152 			if (zcb->zcb_errors[e] != 0) {
7153 				(void) printf("\t%5d  %llu\n",
7154 				    e, (u_longlong_t)zcb->zcb_errors[e]);
7155 			}
7156 		}
7157 	}
7158 
7159 	/*
7160 	 * Report any leaked segments.
7161 	 */
7162 	leaks |= zdb_leak_fini(spa, zcb);
7163 
7164 	tzb = &zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL];
7165 
7166 	norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7167 	norm_space = metaslab_class_get_space(spa_normal_class(spa));
7168 
7169 	total_alloc = norm_alloc +
7170 	    metaslab_class_get_alloc(spa_log_class(spa)) +
7171 	    metaslab_class_get_alloc(spa_embedded_log_class(spa)) +
7172 	    metaslab_class_get_alloc(spa_special_class(spa)) +
7173 	    metaslab_class_get_alloc(spa_dedup_class(spa)) +
7174 	    get_unflushed_alloc_space(spa);
7175 	total_found =
7176 	    tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize +
7177 	    zcb->zcb_removing_size + zcb->zcb_checkpoint_size;
7178 
7179 	if (total_found == total_alloc && !dump_opt['L']) {
7180 		(void) printf("\n\tNo leaks (block sum matches space"
7181 		    " maps exactly)\n");
7182 	} else if (!dump_opt['L']) {
7183 		(void) printf("block traversal size %llu != alloc %llu "
7184 		    "(%s %lld)\n",
7185 		    (u_longlong_t)total_found,
7186 		    (u_longlong_t)total_alloc,
7187 		    (dump_opt['L']) ? "unreachable" : "leaked",
7188 		    (longlong_t)(total_alloc - total_found));
7189 	}
7190 
7191 	if (tzb->zb_count == 0) {
7192 		umem_free(zcb, sizeof (zdb_cb_t));
7193 		return (2);
7194 	}
7195 
7196 	(void) printf("\n");
7197 	(void) printf("\t%-16s %14llu\n", "bp count:",
7198 	    (u_longlong_t)tzb->zb_count);
7199 	(void) printf("\t%-16s %14llu\n", "ganged count:",
7200 	    (longlong_t)tzb->zb_gangs);
7201 	(void) printf("\t%-16s %14llu      avg: %6llu\n", "bp logical:",
7202 	    (u_longlong_t)tzb->zb_lsize,
7203 	    (u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
7204 	(void) printf("\t%-16s %14llu      avg: %6llu     compression: %6.2f\n",
7205 	    "bp physical:", (u_longlong_t)tzb->zb_psize,
7206 	    (u_longlong_t)(tzb->zb_psize / tzb->zb_count),
7207 	    (double)tzb->zb_lsize / tzb->zb_psize);
7208 	(void) printf("\t%-16s %14llu      avg: %6llu     compression: %6.2f\n",
7209 	    "bp allocated:", (u_longlong_t)tzb->zb_asize,
7210 	    (u_longlong_t)(tzb->zb_asize / tzb->zb_count),
7211 	    (double)tzb->zb_lsize / tzb->zb_asize);
7212 	(void) printf("\t%-16s %14llu    ref>1: %6llu   deduplication: %6.2f\n",
7213 	    "bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize,
7214 	    (u_longlong_t)zcb->zcb_dedup_blocks,
7215 	    (double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0);
7216 	(void) printf("\t%-16s %14llu    count: %6llu\n",
7217 	    "bp cloned:", (u_longlong_t)zcb->zcb_clone_asize,
7218 	    (u_longlong_t)zcb->zcb_clone_blocks);
7219 	(void) printf("\t%-16s %14llu     used: %5.2f%%\n", "Normal class:",
7220 	    (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
7221 
7222 	if (spa_special_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7223 		uint64_t alloc = metaslab_class_get_alloc(
7224 		    spa_special_class(spa));
7225 		uint64_t space = metaslab_class_get_space(
7226 		    spa_special_class(spa));
7227 
7228 		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
7229 		    "Special class", (u_longlong_t)alloc,
7230 		    100.0 * alloc / space);
7231 	}
7232 
7233 	if (spa_dedup_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7234 		uint64_t alloc = metaslab_class_get_alloc(
7235 		    spa_dedup_class(spa));
7236 		uint64_t space = metaslab_class_get_space(
7237 		    spa_dedup_class(spa));
7238 
7239 		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
7240 		    "Dedup class", (u_longlong_t)alloc,
7241 		    100.0 * alloc / space);
7242 	}
7243 
7244 	if (spa_embedded_log_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7245 		uint64_t alloc = metaslab_class_get_alloc(
7246 		    spa_embedded_log_class(spa));
7247 		uint64_t space = metaslab_class_get_space(
7248 		    spa_embedded_log_class(spa));
7249 
7250 		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
7251 		    "Embedded log class", (u_longlong_t)alloc,
7252 		    100.0 * alloc / space);
7253 	}
7254 
7255 	for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
7256 		if (zcb->zcb_embedded_blocks[i] == 0)
7257 			continue;
7258 		(void) printf("\n");
7259 		(void) printf("\tadditional, non-pointer bps of type %u: "
7260 		    "%10llu\n",
7261 		    i, (u_longlong_t)zcb->zcb_embedded_blocks[i]);
7262 
7263 		if (dump_opt['b'] >= 3) {
7264 			(void) printf("\t number of (compressed) bytes:  "
7265 			    "number of bps\n");
7266 			dump_histogram(zcb->zcb_embedded_histogram[i],
7267 			    sizeof (zcb->zcb_embedded_histogram[i]) /
7268 			    sizeof (zcb->zcb_embedded_histogram[i][0]), 0);
7269 		}
7270 	}
7271 
7272 	if (tzb->zb_ditto_samevdev != 0) {
7273 		(void) printf("\tDittoed blocks on same vdev: %llu\n",
7274 		    (longlong_t)tzb->zb_ditto_samevdev);
7275 	}
7276 	if (tzb->zb_ditto_same_ms != 0) {
7277 		(void) printf("\tDittoed blocks in same metaslab: %llu\n",
7278 		    (longlong_t)tzb->zb_ditto_same_ms);
7279 	}
7280 
7281 	for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) {
7282 		vdev_t *vd = spa->spa_root_vdev->vdev_child[v];
7283 		vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7284 
7285 		if (vim == NULL) {
7286 			continue;
7287 		}
7288 
7289 		char mem[32];
7290 		zdb_nicenum(vdev_indirect_mapping_num_entries(vim),
7291 		    mem, vdev_indirect_mapping_size(vim));
7292 
7293 		(void) printf("\tindirect vdev id %llu has %llu segments "
7294 		    "(%s in memory)\n",
7295 		    (longlong_t)vd->vdev_id,
7296 		    (longlong_t)vdev_indirect_mapping_num_entries(vim), mem);
7297 	}
7298 
7299 	if (dump_opt['b'] >= 2) {
7300 		int l, t, level;
7301 		char csize[32], lsize[32], psize[32], asize[32];
7302 		char avg[32], gang[32];
7303 		(void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
7304 		    "\t  avg\t comp\t%%Total\tType\n");
7305 
7306 		zfs_blkstat_t *mdstats = umem_zalloc(sizeof (zfs_blkstat_t),
7307 		    UMEM_NOFAIL);
7308 
7309 		for (t = 0; t <= ZDB_OT_TOTAL; t++) {
7310 			const char *typename;
7311 
7312 			/* make sure nicenum has enough space */
7313 			_Static_assert(sizeof (csize) >= NN_NUMBUF_SZ,
7314 			    "csize truncated");
7315 			_Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ,
7316 			    "lsize truncated");
7317 			_Static_assert(sizeof (psize) >= NN_NUMBUF_SZ,
7318 			    "psize truncated");
7319 			_Static_assert(sizeof (asize) >= NN_NUMBUF_SZ,
7320 			    "asize truncated");
7321 			_Static_assert(sizeof (avg) >= NN_NUMBUF_SZ,
7322 			    "avg truncated");
7323 			_Static_assert(sizeof (gang) >= NN_NUMBUF_SZ,
7324 			    "gang truncated");
7325 
7326 			if (t < DMU_OT_NUMTYPES)
7327 				typename = dmu_ot[t].ot_name;
7328 			else
7329 				typename = zdb_ot_extname[t - DMU_OT_NUMTYPES];
7330 
7331 			if (zcb->zcb_type[ZB_TOTAL][t].zb_asize == 0) {
7332 				(void) printf("%6s\t%5s\t%5s\t%5s"
7333 				    "\t%5s\t%5s\t%6s\t%s\n",
7334 				    "-",
7335 				    "-",
7336 				    "-",
7337 				    "-",
7338 				    "-",
7339 				    "-",
7340 				    "-",
7341 				    typename);
7342 				continue;
7343 			}
7344 
7345 			for (l = ZB_TOTAL - 1; l >= -1; l--) {
7346 				level = (l == -1 ? ZB_TOTAL : l);
7347 				zb = &zcb->zcb_type[level][t];
7348 
7349 				if (zb->zb_asize == 0)
7350 					continue;
7351 
7352 				if (level != ZB_TOTAL && t < DMU_OT_NUMTYPES &&
7353 				    (level > 0 || DMU_OT_IS_METADATA(t))) {
7354 					mdstats->zb_count += zb->zb_count;
7355 					mdstats->zb_lsize += zb->zb_lsize;
7356 					mdstats->zb_psize += zb->zb_psize;
7357 					mdstats->zb_asize += zb->zb_asize;
7358 					mdstats->zb_gangs += zb->zb_gangs;
7359 				}
7360 
7361 				if (dump_opt['b'] < 3 && level != ZB_TOTAL)
7362 					continue;
7363 
7364 				if (level == 0 && zb->zb_asize ==
7365 				    zcb->zcb_type[ZB_TOTAL][t].zb_asize)
7366 					continue;
7367 
7368 				zdb_nicenum(zb->zb_count, csize,
7369 				    sizeof (csize));
7370 				zdb_nicenum(zb->zb_lsize, lsize,
7371 				    sizeof (lsize));
7372 				zdb_nicenum(zb->zb_psize, psize,
7373 				    sizeof (psize));
7374 				zdb_nicenum(zb->zb_asize, asize,
7375 				    sizeof (asize));
7376 				zdb_nicenum(zb->zb_asize / zb->zb_count, avg,
7377 				    sizeof (avg));
7378 				zdb_nicenum(zb->zb_gangs, gang, sizeof (gang));
7379 
7380 				(void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7381 				    "\t%5.2f\t%6.2f\t",
7382 				    csize, lsize, psize, asize, avg,
7383 				    (double)zb->zb_lsize / zb->zb_psize,
7384 				    100.0 * zb->zb_asize / tzb->zb_asize);
7385 
7386 				if (level == ZB_TOTAL)
7387 					(void) printf("%s\n", typename);
7388 				else
7389 					(void) printf("    L%d %s\n",
7390 					    level, typename);
7391 
7392 				if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) {
7393 					(void) printf("\t number of ganged "
7394 					    "blocks: %s\n", gang);
7395 				}
7396 
7397 				if (dump_opt['b'] >= 4) {
7398 					(void) printf("psize "
7399 					    "(in 512-byte sectors): "
7400 					    "number of blocks\n");
7401 					dump_histogram(zb->zb_psize_histogram,
7402 					    PSIZE_HISTO_SIZE, 0);
7403 				}
7404 			}
7405 		}
7406 		zdb_nicenum(mdstats->zb_count, csize,
7407 		    sizeof (csize));
7408 		zdb_nicenum(mdstats->zb_lsize, lsize,
7409 		    sizeof (lsize));
7410 		zdb_nicenum(mdstats->zb_psize, psize,
7411 		    sizeof (psize));
7412 		zdb_nicenum(mdstats->zb_asize, asize,
7413 		    sizeof (asize));
7414 		zdb_nicenum(mdstats->zb_asize / mdstats->zb_count, avg,
7415 		    sizeof (avg));
7416 		zdb_nicenum(mdstats->zb_gangs, gang, sizeof (gang));
7417 
7418 		(void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7419 		    "\t%5.2f\t%6.2f\t",
7420 		    csize, lsize, psize, asize, avg,
7421 		    (double)mdstats->zb_lsize / mdstats->zb_psize,
7422 		    100.0 * mdstats->zb_asize / tzb->zb_asize);
7423 		(void) printf("%s\n", "Metadata Total");
7424 
7425 		/* Output a table summarizing block sizes in the pool */
7426 		if (dump_opt['b'] >= 2) {
7427 			dump_size_histograms(zcb);
7428 		}
7429 
7430 		umem_free(mdstats, sizeof (zfs_blkstat_t));
7431 	}
7432 
7433 	(void) printf("\n");
7434 
7435 	if (leaks) {
7436 		umem_free(zcb, sizeof (zdb_cb_t));
7437 		return (2);
7438 	}
7439 
7440 	if (zcb->zcb_haderrors) {
7441 		umem_free(zcb, sizeof (zdb_cb_t));
7442 		return (3);
7443 	}
7444 
7445 	umem_free(zcb, sizeof (zdb_cb_t));
7446 	return (0);
7447 }
7448 
7449 typedef struct zdb_ddt_entry {
7450 	/* key must be first for ddt_key_compare */
7451 	ddt_key_t	zdde_key;
7452 	uint64_t	zdde_ref_blocks;
7453 	uint64_t	zdde_ref_lsize;
7454 	uint64_t	zdde_ref_psize;
7455 	uint64_t	zdde_ref_dsize;
7456 	avl_node_t	zdde_node;
7457 } zdb_ddt_entry_t;
7458 
7459 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)7460 zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
7461     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
7462 {
7463 	(void) zilog, (void) dnp;
7464 	avl_tree_t *t = arg;
7465 	avl_index_t where;
7466 	zdb_ddt_entry_t *zdde, zdde_search;
7467 
7468 	if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
7469 	    BP_IS_EMBEDDED(bp))
7470 		return (0);
7471 
7472 	if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
7473 		(void) printf("traversing objset %llu, %llu objects, "
7474 		    "%lu blocks so far\n",
7475 		    (u_longlong_t)zb->zb_objset,
7476 		    (u_longlong_t)BP_GET_FILL(bp),
7477 		    avl_numnodes(t));
7478 	}
7479 
7480 	if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF ||
7481 	    BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
7482 		return (0);
7483 
7484 	ddt_key_fill(&zdde_search.zdde_key, bp);
7485 
7486 	zdde = avl_find(t, &zdde_search, &where);
7487 
7488 	if (zdde == NULL) {
7489 		zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL);
7490 		zdde->zdde_key = zdde_search.zdde_key;
7491 		avl_insert(t, zdde, where);
7492 	}
7493 
7494 	zdde->zdde_ref_blocks += 1;
7495 	zdde->zdde_ref_lsize += BP_GET_LSIZE(bp);
7496 	zdde->zdde_ref_psize += BP_GET_PSIZE(bp);
7497 	zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp);
7498 
7499 	return (0);
7500 }
7501 
7502 static void
dump_simulated_ddt(spa_t * spa)7503 dump_simulated_ddt(spa_t *spa)
7504 {
7505 	avl_tree_t t;
7506 	void *cookie = NULL;
7507 	zdb_ddt_entry_t *zdde;
7508 	ddt_histogram_t ddh_total = {{{0}}};
7509 	ddt_stat_t dds_total = {0};
7510 
7511 	avl_create(&t, ddt_key_compare,
7512 	    sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
7513 
7514 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7515 
7516 	(void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7517 	    TRAVERSE_NO_DECRYPT, zdb_ddt_add_cb, &t);
7518 
7519 	spa_config_exit(spa, SCL_CONFIG, FTAG);
7520 
7521 	while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) {
7522 		uint64_t refcnt = zdde->zdde_ref_blocks;
7523 		ASSERT(refcnt != 0);
7524 
7525 		ddt_stat_t *dds = &ddh_total.ddh_stat[highbit64(refcnt) - 1];
7526 
7527 		dds->dds_blocks += zdde->zdde_ref_blocks / refcnt;
7528 		dds->dds_lsize += zdde->zdde_ref_lsize / refcnt;
7529 		dds->dds_psize += zdde->zdde_ref_psize / refcnt;
7530 		dds->dds_dsize += zdde->zdde_ref_dsize / refcnt;
7531 
7532 		dds->dds_ref_blocks += zdde->zdde_ref_blocks;
7533 		dds->dds_ref_lsize += zdde->zdde_ref_lsize;
7534 		dds->dds_ref_psize += zdde->zdde_ref_psize;
7535 		dds->dds_ref_dsize += zdde->zdde_ref_dsize;
7536 
7537 		umem_free(zdde, sizeof (*zdde));
7538 	}
7539 
7540 	avl_destroy(&t);
7541 
7542 	ddt_histogram_total(&dds_total, &ddh_total);
7543 
7544 	(void) printf("Simulated DDT histogram:\n");
7545 
7546 	zpool_dump_ddt(&dds_total, &ddh_total);
7547 
7548 	dump_dedup_ratio(&dds_total);
7549 }
7550 
7551 static int
verify_device_removal_feature_counts(spa_t * spa)7552 verify_device_removal_feature_counts(spa_t *spa)
7553 {
7554 	uint64_t dr_feature_refcount = 0;
7555 	uint64_t oc_feature_refcount = 0;
7556 	uint64_t indirect_vdev_count = 0;
7557 	uint64_t precise_vdev_count = 0;
7558 	uint64_t obsolete_counts_object_count = 0;
7559 	uint64_t obsolete_sm_count = 0;
7560 	uint64_t obsolete_counts_count = 0;
7561 	uint64_t scip_count = 0;
7562 	uint64_t obsolete_bpobj_count = 0;
7563 	int ret = 0;
7564 
7565 	spa_condensing_indirect_phys_t *scip =
7566 	    &spa->spa_condensing_indirect_phys;
7567 	if (scip->scip_next_mapping_object != 0) {
7568 		vdev_t *vd = spa->spa_root_vdev->vdev_child[scip->scip_vdev];
7569 		ASSERT(scip->scip_prev_obsolete_sm_object != 0);
7570 		ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
7571 
7572 		(void) printf("Condensing indirect vdev %llu: new mapping "
7573 		    "object %llu, prev obsolete sm %llu\n",
7574 		    (u_longlong_t)scip->scip_vdev,
7575 		    (u_longlong_t)scip->scip_next_mapping_object,
7576 		    (u_longlong_t)scip->scip_prev_obsolete_sm_object);
7577 		if (scip->scip_prev_obsolete_sm_object != 0) {
7578 			space_map_t *prev_obsolete_sm = NULL;
7579 			VERIFY0(space_map_open(&prev_obsolete_sm,
7580 			    spa->spa_meta_objset,
7581 			    scip->scip_prev_obsolete_sm_object,
7582 			    0, vd->vdev_asize, 0));
7583 			dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm);
7584 			(void) printf("\n");
7585 			space_map_close(prev_obsolete_sm);
7586 		}
7587 
7588 		scip_count += 2;
7589 	}
7590 
7591 	for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
7592 		vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
7593 		vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
7594 
7595 		if (vic->vic_mapping_object != 0) {
7596 			ASSERT(vd->vdev_ops == &vdev_indirect_ops ||
7597 			    vd->vdev_removing);
7598 			indirect_vdev_count++;
7599 
7600 			if (vd->vdev_indirect_mapping->vim_havecounts) {
7601 				obsolete_counts_count++;
7602 			}
7603 		}
7604 
7605 		boolean_t are_precise;
7606 		VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7607 		if (are_precise) {
7608 			ASSERT(vic->vic_mapping_object != 0);
7609 			precise_vdev_count++;
7610 		}
7611 
7612 		uint64_t obsolete_sm_object;
7613 		VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
7614 		if (obsolete_sm_object != 0) {
7615 			ASSERT(vic->vic_mapping_object != 0);
7616 			obsolete_sm_count++;
7617 		}
7618 	}
7619 
7620 	(void) feature_get_refcount(spa,
7621 	    &spa_feature_table[SPA_FEATURE_DEVICE_REMOVAL],
7622 	    &dr_feature_refcount);
7623 	(void) feature_get_refcount(spa,
7624 	    &spa_feature_table[SPA_FEATURE_OBSOLETE_COUNTS],
7625 	    &oc_feature_refcount);
7626 
7627 	if (dr_feature_refcount != indirect_vdev_count) {
7628 		ret = 1;
7629 		(void) printf("Number of indirect vdevs (%llu) " \
7630 		    "does not match feature count (%llu)\n",
7631 		    (u_longlong_t)indirect_vdev_count,
7632 		    (u_longlong_t)dr_feature_refcount);
7633 	} else {
7634 		(void) printf("Verified device_removal feature refcount " \
7635 		    "of %llu is correct\n",
7636 		    (u_longlong_t)dr_feature_refcount);
7637 	}
7638 
7639 	if (zap_contains(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT,
7640 	    DMU_POOL_OBSOLETE_BPOBJ) == 0) {
7641 		obsolete_bpobj_count++;
7642 	}
7643 
7644 
7645 	obsolete_counts_object_count = precise_vdev_count;
7646 	obsolete_counts_object_count += obsolete_sm_count;
7647 	obsolete_counts_object_count += obsolete_counts_count;
7648 	obsolete_counts_object_count += scip_count;
7649 	obsolete_counts_object_count += obsolete_bpobj_count;
7650 	obsolete_counts_object_count += remap_deadlist_count;
7651 
7652 	if (oc_feature_refcount != obsolete_counts_object_count) {
7653 		ret = 1;
7654 		(void) printf("Number of obsolete counts objects (%llu) " \
7655 		    "does not match feature count (%llu)\n",
7656 		    (u_longlong_t)obsolete_counts_object_count,
7657 		    (u_longlong_t)oc_feature_refcount);
7658 		(void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
7659 		    "ob:%llu rd:%llu\n",
7660 		    (u_longlong_t)precise_vdev_count,
7661 		    (u_longlong_t)obsolete_sm_count,
7662 		    (u_longlong_t)obsolete_counts_count,
7663 		    (u_longlong_t)scip_count,
7664 		    (u_longlong_t)obsolete_bpobj_count,
7665 		    (u_longlong_t)remap_deadlist_count);
7666 	} else {
7667 		(void) printf("Verified indirect_refcount feature refcount " \
7668 		    "of %llu is correct\n",
7669 		    (u_longlong_t)oc_feature_refcount);
7670 	}
7671 	return (ret);
7672 }
7673 
7674 static void
zdb_set_skip_mmp(char * target)7675 zdb_set_skip_mmp(char *target)
7676 {
7677 	spa_t *spa;
7678 
7679 	/*
7680 	 * Disable the activity check to allow examination of
7681 	 * active pools.
7682 	 */
7683 	mutex_enter(&spa_namespace_lock);
7684 	if ((spa = spa_lookup(target)) != NULL) {
7685 		spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
7686 	}
7687 	mutex_exit(&spa_namespace_lock);
7688 }
7689 
7690 #define	BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7691 /*
7692  * Import the checkpointed state of the pool specified by the target
7693  * parameter as readonly. The function also accepts a pool config
7694  * as an optional parameter, else it attempts to infer the config by
7695  * the name of the target pool.
7696  *
7697  * Note that the checkpointed state's pool name will be the name of
7698  * the original pool with the above suffix appended to it. In addition,
7699  * if the target is not a pool name (e.g. a path to a dataset) then
7700  * the new_path parameter is populated with the updated path to
7701  * reflect the fact that we are looking into the checkpointed state.
7702  *
7703  * The function returns a newly-allocated copy of the name of the
7704  * pool containing the checkpointed state. When this copy is no
7705  * longer needed it should be freed with free(3C). Same thing
7706  * applies to the new_path parameter if allocated.
7707  */
7708 static char *
import_checkpointed_state(char * target,nvlist_t * cfg,char ** new_path)7709 import_checkpointed_state(char *target, nvlist_t *cfg, char **new_path)
7710 {
7711 	int error = 0;
7712 	char *poolname, *bogus_name = NULL;
7713 	boolean_t freecfg = B_FALSE;
7714 
7715 	/* If the target is not a pool, the extract the pool name */
7716 	char *path_start = strchr(target, '/');
7717 	if (path_start != NULL) {
7718 		size_t poolname_len = path_start - target;
7719 		poolname = strndup(target, poolname_len);
7720 	} else {
7721 		poolname = target;
7722 	}
7723 
7724 	if (cfg == NULL) {
7725 		zdb_set_skip_mmp(poolname);
7726 		error = spa_get_stats(poolname, &cfg, NULL, 0);
7727 		if (error != 0) {
7728 			fatal("Tried to read config of pool \"%s\" but "
7729 			    "spa_get_stats() failed with error %d\n",
7730 			    poolname, error);
7731 		}
7732 		freecfg = B_TRUE;
7733 	}
7734 
7735 	if (asprintf(&bogus_name, "%s%s", poolname, BOGUS_SUFFIX) == -1) {
7736 		if (target != poolname)
7737 			free(poolname);
7738 		return (NULL);
7739 	}
7740 	fnvlist_add_string(cfg, ZPOOL_CONFIG_POOL_NAME, bogus_name);
7741 
7742 	error = spa_import(bogus_name, cfg, NULL,
7743 	    ZFS_IMPORT_MISSING_LOG | ZFS_IMPORT_CHECKPOINT |
7744 	    ZFS_IMPORT_SKIP_MMP);
7745 	if (freecfg)
7746 		nvlist_free(cfg);
7747 	if (error != 0) {
7748 		fatal("Tried to import pool \"%s\" but spa_import() failed "
7749 		    "with error %d\n", bogus_name, error);
7750 	}
7751 
7752 	if (new_path != NULL && path_start != NULL) {
7753 		if (asprintf(new_path, "%s%s", bogus_name, path_start) == -1) {
7754 			free(bogus_name);
7755 			if (path_start != NULL)
7756 				free(poolname);
7757 			return (NULL);
7758 		}
7759 	}
7760 
7761 	if (target != poolname)
7762 		free(poolname);
7763 
7764 	return (bogus_name);
7765 }
7766 
7767 typedef struct verify_checkpoint_sm_entry_cb_arg {
7768 	vdev_t *vcsec_vd;
7769 
7770 	/* the following fields are only used for printing progress */
7771 	uint64_t vcsec_entryid;
7772 	uint64_t vcsec_num_entries;
7773 } verify_checkpoint_sm_entry_cb_arg_t;
7774 
7775 #define	ENTRIES_PER_PROGRESS_UPDATE 10000
7776 
7777 static int
verify_checkpoint_sm_entry_cb(space_map_entry_t * sme,void * arg)7778 verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, void *arg)
7779 {
7780 	verify_checkpoint_sm_entry_cb_arg_t *vcsec = arg;
7781 	vdev_t *vd = vcsec->vcsec_vd;
7782 	metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
7783 	uint64_t end = sme->sme_offset + sme->sme_run;
7784 
7785 	ASSERT(sme->sme_type == SM_FREE);
7786 
7787 	if ((vcsec->vcsec_entryid % ENTRIES_PER_PROGRESS_UPDATE) == 0) {
7788 		(void) fprintf(stderr,
7789 		    "\rverifying vdev %llu, space map entry %llu of %llu ...",
7790 		    (longlong_t)vd->vdev_id,
7791 		    (longlong_t)vcsec->vcsec_entryid,
7792 		    (longlong_t)vcsec->vcsec_num_entries);
7793 	}
7794 	vcsec->vcsec_entryid++;
7795 
7796 	/*
7797 	 * See comment in checkpoint_sm_exclude_entry_cb()
7798 	 */
7799 	VERIFY3U(sme->sme_offset, >=, ms->ms_start);
7800 	VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
7801 
7802 	/*
7803 	 * The entries in the vdev_checkpoint_sm should be marked as
7804 	 * allocated in the checkpointed state of the pool, therefore
7805 	 * their respective ms_allocateable trees should not contain them.
7806 	 */
7807 	mutex_enter(&ms->ms_lock);
7808 	zfs_range_tree_verify_not_present(ms->ms_allocatable,
7809 	    sme->sme_offset, sme->sme_run);
7810 	mutex_exit(&ms->ms_lock);
7811 
7812 	return (0);
7813 }
7814 
7815 /*
7816  * Verify that all segments in the vdev_checkpoint_sm are allocated
7817  * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
7818  * ms_allocatable).
7819  *
7820  * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
7821  * each vdev in the current state of the pool to the metaslab space maps
7822  * (ms_sm) of the checkpointed state of the pool.
7823  *
7824  * Note that the function changes the state of the ms_allocatable
7825  * trees of the current spa_t. The entries of these ms_allocatable
7826  * trees are cleared out and then repopulated from with the free
7827  * entries of their respective ms_sm space maps.
7828  */
7829 static void
verify_checkpoint_vdev_spacemaps(spa_t * checkpoint,spa_t * current)7830 verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, spa_t *current)
7831 {
7832 	vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7833 	vdev_t *current_rvd = current->spa_root_vdev;
7834 
7835 	load_concrete_ms_allocatable_trees(checkpoint, SM_FREE);
7836 
7837 	for (uint64_t c = 0; c < ckpoint_rvd->vdev_children; c++) {
7838 		vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[c];
7839 		vdev_t *current_vd = current_rvd->vdev_child[c];
7840 
7841 		space_map_t *checkpoint_sm = NULL;
7842 		uint64_t checkpoint_sm_obj;
7843 
7844 		if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7845 			/*
7846 			 * Since we don't allow device removal in a pool
7847 			 * that has a checkpoint, we expect that all removed
7848 			 * vdevs were removed from the pool before the
7849 			 * checkpoint.
7850 			 */
7851 			ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7852 			continue;
7853 		}
7854 
7855 		/*
7856 		 * If the checkpoint space map doesn't exist, then nothing
7857 		 * here is checkpointed so there's nothing to verify.
7858 		 */
7859 		if (current_vd->vdev_top_zap == 0 ||
7860 		    zap_contains(spa_meta_objset(current),
7861 		    current_vd->vdev_top_zap,
7862 		    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
7863 			continue;
7864 
7865 		VERIFY0(zap_lookup(spa_meta_objset(current),
7866 		    current_vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
7867 		    sizeof (uint64_t), 1, &checkpoint_sm_obj));
7868 
7869 		VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current),
7870 		    checkpoint_sm_obj, 0, current_vd->vdev_asize,
7871 		    current_vd->vdev_ashift));
7872 
7873 		verify_checkpoint_sm_entry_cb_arg_t vcsec;
7874 		vcsec.vcsec_vd = ckpoint_vd;
7875 		vcsec.vcsec_entryid = 0;
7876 		vcsec.vcsec_num_entries =
7877 		    space_map_length(checkpoint_sm) / sizeof (uint64_t);
7878 		VERIFY0(space_map_iterate(checkpoint_sm,
7879 		    space_map_length(checkpoint_sm),
7880 		    verify_checkpoint_sm_entry_cb, &vcsec));
7881 		if (dump_opt['m'] > 3)
7882 			dump_spacemap(current->spa_meta_objset, checkpoint_sm);
7883 		space_map_close(checkpoint_sm);
7884 	}
7885 
7886 	/*
7887 	 * If we've added vdevs since we took the checkpoint, ensure
7888 	 * that their checkpoint space maps are empty.
7889 	 */
7890 	if (ckpoint_rvd->vdev_children < current_rvd->vdev_children) {
7891 		for (uint64_t c = ckpoint_rvd->vdev_children;
7892 		    c < current_rvd->vdev_children; c++) {
7893 			vdev_t *current_vd = current_rvd->vdev_child[c];
7894 			VERIFY3P(current_vd->vdev_checkpoint_sm, ==, NULL);
7895 		}
7896 	}
7897 
7898 	/* for cleaner progress output */
7899 	(void) fprintf(stderr, "\n");
7900 }
7901 
7902 /*
7903  * Verifies that all space that's allocated in the checkpoint is
7904  * still allocated in the current version, by checking that everything
7905  * in checkpoint's ms_allocatable (which is actually allocated, not
7906  * allocatable/free) is not present in current's ms_allocatable.
7907  *
7908  * Note that the function changes the state of the ms_allocatable
7909  * trees of both spas when called. The entries of all ms_allocatable
7910  * trees are cleared out and then repopulated from their respective
7911  * ms_sm space maps. In the checkpointed state we load the allocated
7912  * entries, and in the current state we load the free entries.
7913  */
7914 static void
verify_checkpoint_ms_spacemaps(spa_t * checkpoint,spa_t * current)7915 verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_t *current)
7916 {
7917 	vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7918 	vdev_t *current_rvd = current->spa_root_vdev;
7919 
7920 	load_concrete_ms_allocatable_trees(checkpoint, SM_ALLOC);
7921 	load_concrete_ms_allocatable_trees(current, SM_FREE);
7922 
7923 	for (uint64_t i = 0; i < ckpoint_rvd->vdev_children; i++) {
7924 		vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[i];
7925 		vdev_t *current_vd = current_rvd->vdev_child[i];
7926 
7927 		if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7928 			/*
7929 			 * See comment in verify_checkpoint_vdev_spacemaps()
7930 			 */
7931 			ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7932 			continue;
7933 		}
7934 
7935 		for (uint64_t m = 0; m < ckpoint_vd->vdev_ms_count; m++) {
7936 			metaslab_t *ckpoint_msp = ckpoint_vd->vdev_ms[m];
7937 			metaslab_t *current_msp = current_vd->vdev_ms[m];
7938 
7939 			(void) fprintf(stderr,
7940 			    "\rverifying vdev %llu of %llu, "
7941 			    "metaslab %llu of %llu ...",
7942 			    (longlong_t)current_vd->vdev_id,
7943 			    (longlong_t)current_rvd->vdev_children,
7944 			    (longlong_t)current_vd->vdev_ms[m]->ms_id,
7945 			    (longlong_t)current_vd->vdev_ms_count);
7946 
7947 			/*
7948 			 * We walk through the ms_allocatable trees that
7949 			 * are loaded with the allocated blocks from the
7950 			 * ms_sm spacemaps of the checkpoint. For each
7951 			 * one of these ranges we ensure that none of them
7952 			 * exists in the ms_allocatable trees of the
7953 			 * current state which are loaded with the ranges
7954 			 * that are currently free.
7955 			 *
7956 			 * This way we ensure that none of the blocks that
7957 			 * are part of the checkpoint were freed by mistake.
7958 			 */
7959 			zfs_range_tree_walk(ckpoint_msp->ms_allocatable,
7960 			    (zfs_range_tree_func_t *)
7961 			    zfs_range_tree_verify_not_present,
7962 			    current_msp->ms_allocatable);
7963 		}
7964 	}
7965 
7966 	/* for cleaner progress output */
7967 	(void) fprintf(stderr, "\n");
7968 }
7969 
7970 static void
verify_checkpoint_blocks(spa_t * spa)7971 verify_checkpoint_blocks(spa_t *spa)
7972 {
7973 	ASSERT(!dump_opt['L']);
7974 
7975 	spa_t *checkpoint_spa;
7976 	char *checkpoint_pool;
7977 	int error = 0;
7978 
7979 	/*
7980 	 * We import the checkpointed state of the pool (under a different
7981 	 * name) so we can do verification on it against the current state
7982 	 * of the pool.
7983 	 */
7984 	checkpoint_pool = import_checkpointed_state(spa->spa_name, NULL,
7985 	    NULL);
7986 	ASSERT(strcmp(spa->spa_name, checkpoint_pool) != 0);
7987 
7988 	error = spa_open(checkpoint_pool, &checkpoint_spa, FTAG);
7989 	if (error != 0) {
7990 		fatal("Tried to open pool \"%s\" but spa_open() failed with "
7991 		    "error %d\n", checkpoint_pool, error);
7992 	}
7993 
7994 	/*
7995 	 * Ensure that ranges in the checkpoint space maps of each vdev
7996 	 * are allocated according to the checkpointed state's metaslab
7997 	 * space maps.
7998 	 */
7999 	verify_checkpoint_vdev_spacemaps(checkpoint_spa, spa);
8000 
8001 	/*
8002 	 * Ensure that allocated ranges in the checkpoint's metaslab
8003 	 * space maps remain allocated in the metaslab space maps of
8004 	 * the current state.
8005 	 */
8006 	verify_checkpoint_ms_spacemaps(checkpoint_spa, spa);
8007 
8008 	/*
8009 	 * Once we are done, we get rid of the checkpointed state.
8010 	 */
8011 	spa_close(checkpoint_spa, FTAG);
8012 	free(checkpoint_pool);
8013 }
8014 
8015 static void
dump_leftover_checkpoint_blocks(spa_t * spa)8016 dump_leftover_checkpoint_blocks(spa_t *spa)
8017 {
8018 	vdev_t *rvd = spa->spa_root_vdev;
8019 
8020 	for (uint64_t i = 0; i < rvd->vdev_children; i++) {
8021 		vdev_t *vd = rvd->vdev_child[i];
8022 
8023 		space_map_t *checkpoint_sm = NULL;
8024 		uint64_t checkpoint_sm_obj;
8025 
8026 		if (vd->vdev_top_zap == 0)
8027 			continue;
8028 
8029 		if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
8030 		    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
8031 			continue;
8032 
8033 		VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
8034 		    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
8035 		    sizeof (uint64_t), 1, &checkpoint_sm_obj));
8036 
8037 		VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
8038 		    checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
8039 		dump_spacemap(spa->spa_meta_objset, checkpoint_sm);
8040 		space_map_close(checkpoint_sm);
8041 	}
8042 }
8043 
8044 static int
verify_checkpoint(spa_t * spa)8045 verify_checkpoint(spa_t *spa)
8046 {
8047 	uberblock_t checkpoint;
8048 	int error;
8049 
8050 	if (!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT))
8051 		return (0);
8052 
8053 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8054 	    DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
8055 	    sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
8056 
8057 	if (error == ENOENT && !dump_opt['L']) {
8058 		/*
8059 		 * If the feature is active but the uberblock is missing
8060 		 * then we must be in the middle of discarding the
8061 		 * checkpoint.
8062 		 */
8063 		(void) printf("\nPartially discarded checkpoint "
8064 		    "state found:\n");
8065 		if (dump_opt['m'] > 3)
8066 			dump_leftover_checkpoint_blocks(spa);
8067 		return (0);
8068 	} else if (error != 0) {
8069 		(void) printf("lookup error %d when looking for "
8070 		    "checkpointed uberblock in MOS\n", error);
8071 		return (error);
8072 	}
8073 	dump_uberblock(&checkpoint, "\nCheckpointed uberblock found:\n", "\n");
8074 
8075 	if (checkpoint.ub_checkpoint_txg == 0) {
8076 		(void) printf("\nub_checkpoint_txg not set in checkpointed "
8077 		    "uberblock\n");
8078 		error = 3;
8079 	}
8080 
8081 	if (error == 0 && !dump_opt['L'])
8082 		verify_checkpoint_blocks(spa);
8083 
8084 	return (error);
8085 }
8086 
8087 static void
mos_leaks_cb(void * arg,uint64_t start,uint64_t size)8088 mos_leaks_cb(void *arg, uint64_t start, uint64_t size)
8089 {
8090 	(void) arg;
8091 	for (uint64_t i = start; i < size; i++) {
8092 		(void) printf("MOS object %llu referenced but not allocated\n",
8093 		    (u_longlong_t)i);
8094 	}
8095 }
8096 
8097 static void
mos_obj_refd(uint64_t obj)8098 mos_obj_refd(uint64_t obj)
8099 {
8100 	if (obj != 0 && mos_refd_objs != NULL)
8101 		zfs_range_tree_add(mos_refd_objs, obj, 1);
8102 }
8103 
8104 /*
8105  * Call on a MOS object that may already have been referenced.
8106  */
8107 static void
mos_obj_refd_multiple(uint64_t obj)8108 mos_obj_refd_multiple(uint64_t obj)
8109 {
8110 	if (obj != 0 && mos_refd_objs != NULL &&
8111 	    !zfs_range_tree_contains(mos_refd_objs, obj, 1))
8112 		zfs_range_tree_add(mos_refd_objs, obj, 1);
8113 }
8114 
8115 static void
mos_leak_vdev_top_zap(vdev_t * vd)8116 mos_leak_vdev_top_zap(vdev_t *vd)
8117 {
8118 	uint64_t ms_flush_data_obj;
8119 	int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
8120 	    vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
8121 	    sizeof (ms_flush_data_obj), 1, &ms_flush_data_obj);
8122 	if (error == ENOENT)
8123 		return;
8124 	ASSERT0(error);
8125 
8126 	mos_obj_refd(ms_flush_data_obj);
8127 }
8128 
8129 static void
mos_leak_vdev(vdev_t * vd)8130 mos_leak_vdev(vdev_t *vd)
8131 {
8132 	mos_obj_refd(vd->vdev_dtl_object);
8133 	mos_obj_refd(vd->vdev_ms_array);
8134 	mos_obj_refd(vd->vdev_indirect_config.vic_births_object);
8135 	mos_obj_refd(vd->vdev_indirect_config.vic_mapping_object);
8136 	mos_obj_refd(vd->vdev_leaf_zap);
8137 	if (vd->vdev_checkpoint_sm != NULL)
8138 		mos_obj_refd(vd->vdev_checkpoint_sm->sm_object);
8139 	if (vd->vdev_indirect_mapping != NULL) {
8140 		mos_obj_refd(vd->vdev_indirect_mapping->
8141 		    vim_phys->vimp_counts_object);
8142 	}
8143 	if (vd->vdev_obsolete_sm != NULL)
8144 		mos_obj_refd(vd->vdev_obsolete_sm->sm_object);
8145 
8146 	for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
8147 		metaslab_t *ms = vd->vdev_ms[m];
8148 		mos_obj_refd(space_map_object(ms->ms_sm));
8149 	}
8150 
8151 	if (vd->vdev_root_zap != 0)
8152 		mos_obj_refd(vd->vdev_root_zap);
8153 
8154 	if (vd->vdev_top_zap != 0) {
8155 		mos_obj_refd(vd->vdev_top_zap);
8156 		mos_leak_vdev_top_zap(vd);
8157 	}
8158 
8159 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
8160 		mos_leak_vdev(vd->vdev_child[c]);
8161 	}
8162 }
8163 
8164 static void
mos_leak_log_spacemaps(spa_t * spa)8165 mos_leak_log_spacemaps(spa_t *spa)
8166 {
8167 	uint64_t spacemap_zap;
8168 	int error = zap_lookup(spa_meta_objset(spa),
8169 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LOG_SPACEMAP_ZAP,
8170 	    sizeof (spacemap_zap), 1, &spacemap_zap);
8171 	if (error == ENOENT)
8172 		return;
8173 	ASSERT0(error);
8174 
8175 	mos_obj_refd(spacemap_zap);
8176 	for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
8177 	    sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls))
8178 		mos_obj_refd(sls->sls_sm_obj);
8179 }
8180 
8181 static void
errorlog_count_refd(objset_t * mos,uint64_t errlog)8182 errorlog_count_refd(objset_t *mos, uint64_t errlog)
8183 {
8184 	zap_cursor_t zc;
8185 	zap_attribute_t *za = zap_attribute_alloc();
8186 	for (zap_cursor_init(&zc, mos, errlog);
8187 	    zap_cursor_retrieve(&zc, za) == 0;
8188 	    zap_cursor_advance(&zc)) {
8189 		mos_obj_refd(za->za_first_integer);
8190 	}
8191 	zap_cursor_fini(&zc);
8192 	zap_attribute_free(za);
8193 }
8194 
8195 static int
dump_mos_leaks(spa_t * spa)8196 dump_mos_leaks(spa_t *spa)
8197 {
8198 	int rv = 0;
8199 	objset_t *mos = spa->spa_meta_objset;
8200 	dsl_pool_t *dp = spa->spa_dsl_pool;
8201 
8202 	/* Visit and mark all referenced objects in the MOS */
8203 
8204 	mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT);
8205 	mos_obj_refd(spa->spa_pool_props_object);
8206 	mos_obj_refd(spa->spa_config_object);
8207 	mos_obj_refd(spa->spa_ddt_stat_object);
8208 	mos_obj_refd(spa->spa_feat_desc_obj);
8209 	mos_obj_refd(spa->spa_feat_enabled_txg_obj);
8210 	mos_obj_refd(spa->spa_feat_for_read_obj);
8211 	mos_obj_refd(spa->spa_feat_for_write_obj);
8212 	mos_obj_refd(spa->spa_history);
8213 	mos_obj_refd(spa->spa_errlog_last);
8214 	mos_obj_refd(spa->spa_errlog_scrub);
8215 
8216 	if (spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) {
8217 		errorlog_count_refd(mos, spa->spa_errlog_last);
8218 		errorlog_count_refd(mos, spa->spa_errlog_scrub);
8219 	}
8220 
8221 	mos_obj_refd(spa->spa_all_vdev_zaps);
8222 	mos_obj_refd(spa->spa_dsl_pool->dp_bptree_obj);
8223 	mos_obj_refd(spa->spa_dsl_pool->dp_tmp_userrefs_obj);
8224 	mos_obj_refd(spa->spa_dsl_pool->dp_scan->scn_phys.scn_queue_obj);
8225 	bpobj_count_refd(&spa->spa_deferred_bpobj);
8226 	mos_obj_refd(dp->dp_empty_bpobj);
8227 	bpobj_count_refd(&dp->dp_obsolete_bpobj);
8228 	bpobj_count_refd(&dp->dp_free_bpobj);
8229 	mos_obj_refd(spa->spa_l2cache.sav_object);
8230 	mos_obj_refd(spa->spa_spares.sav_object);
8231 
8232 	if (spa->spa_syncing_log_sm != NULL)
8233 		mos_obj_refd(spa->spa_syncing_log_sm->sm_object);
8234 	mos_leak_log_spacemaps(spa);
8235 
8236 	mos_obj_refd(spa->spa_condensing_indirect_phys.
8237 	    scip_next_mapping_object);
8238 	mos_obj_refd(spa->spa_condensing_indirect_phys.
8239 	    scip_prev_obsolete_sm_object);
8240 	if (spa->spa_condensing_indirect_phys.scip_next_mapping_object != 0) {
8241 		vdev_indirect_mapping_t *vim =
8242 		    vdev_indirect_mapping_open(mos,
8243 		    spa->spa_condensing_indirect_phys.scip_next_mapping_object);
8244 		mos_obj_refd(vim->vim_phys->vimp_counts_object);
8245 		vdev_indirect_mapping_close(vim);
8246 	}
8247 	deleted_livelists_dump_mos(spa);
8248 
8249 	if (dp->dp_origin_snap != NULL) {
8250 		dsl_dataset_t *ds;
8251 
8252 		dsl_pool_config_enter(dp, FTAG);
8253 		VERIFY0(dsl_dataset_hold_obj(dp,
8254 		    dsl_dataset_phys(dp->dp_origin_snap)->ds_next_snap_obj,
8255 		    FTAG, &ds));
8256 		count_ds_mos_objects(ds);
8257 		dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
8258 		dsl_dataset_rele(ds, FTAG);
8259 		dsl_pool_config_exit(dp, FTAG);
8260 
8261 		count_ds_mos_objects(dp->dp_origin_snap);
8262 		dump_blkptr_list(&dp->dp_origin_snap->ds_deadlist, "Deadlist");
8263 	}
8264 	count_dir_mos_objects(dp->dp_mos_dir);
8265 	if (dp->dp_free_dir != NULL)
8266 		count_dir_mos_objects(dp->dp_free_dir);
8267 	if (dp->dp_leak_dir != NULL)
8268 		count_dir_mos_objects(dp->dp_leak_dir);
8269 
8270 	mos_leak_vdev(spa->spa_root_vdev);
8271 
8272 	for (uint64_t c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
8273 		ddt_t *ddt = spa->spa_ddt[c];
8274 		if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
8275 			continue;
8276 
8277 		/* DDT store objects */
8278 		for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
8279 			for (ddt_class_t class = 0; class < DDT_CLASSES;
8280 			    class++) {
8281 				mos_obj_refd(ddt->ddt_object[type][class]);
8282 			}
8283 		}
8284 
8285 		/* FDT container */
8286 		if (ddt->ddt_version == DDT_VERSION_FDT)
8287 			mos_obj_refd(ddt->ddt_dir_object);
8288 
8289 		/* FDT log objects */
8290 		if (ddt->ddt_flags & DDT_FLAG_LOG) {
8291 			mos_obj_refd(ddt->ddt_log[0].ddl_object);
8292 			mos_obj_refd(ddt->ddt_log[1].ddl_object);
8293 		}
8294 	}
8295 
8296 	for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
8297 		brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
8298 		if (brtvd->bv_initiated) {
8299 			mos_obj_refd(brtvd->bv_mos_brtvdev);
8300 			mos_obj_refd(brtvd->bv_mos_entries);
8301 		}
8302 	}
8303 
8304 	/*
8305 	 * Visit all allocated objects and make sure they are referenced.
8306 	 */
8307 	uint64_t object = 0;
8308 	while (dmu_object_next(mos, &object, B_FALSE, 0) == 0) {
8309 		if (zfs_range_tree_contains(mos_refd_objs, object, 1)) {
8310 			zfs_range_tree_remove(mos_refd_objs, object, 1);
8311 		} else {
8312 			dmu_object_info_t doi;
8313 			const char *name;
8314 			VERIFY0(dmu_object_info(mos, object, &doi));
8315 			if (doi.doi_type & DMU_OT_NEWTYPE) {
8316 				dmu_object_byteswap_t bswap =
8317 				    DMU_OT_BYTESWAP(doi.doi_type);
8318 				name = dmu_ot_byteswap[bswap].ob_name;
8319 			} else {
8320 				name = dmu_ot[doi.doi_type].ot_name;
8321 			}
8322 
8323 			(void) printf("MOS object %llu (%s) leaked\n",
8324 			    (u_longlong_t)object, name);
8325 			rv = 2;
8326 		}
8327 	}
8328 	(void) zfs_range_tree_walk(mos_refd_objs, mos_leaks_cb, NULL);
8329 	if (!zfs_range_tree_is_empty(mos_refd_objs))
8330 		rv = 2;
8331 	zfs_range_tree_vacate(mos_refd_objs, NULL, NULL);
8332 	zfs_range_tree_destroy(mos_refd_objs);
8333 	return (rv);
8334 }
8335 
8336 typedef struct log_sm_obsolete_stats_arg {
8337 	uint64_t lsos_current_txg;
8338 
8339 	uint64_t lsos_total_entries;
8340 	uint64_t lsos_valid_entries;
8341 
8342 	uint64_t lsos_sm_entries;
8343 	uint64_t lsos_valid_sm_entries;
8344 } log_sm_obsolete_stats_arg_t;
8345 
8346 static int
log_spacemap_obsolete_stats_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)8347 log_spacemap_obsolete_stats_cb(spa_t *spa, space_map_entry_t *sme,
8348     uint64_t txg, void *arg)
8349 {
8350 	log_sm_obsolete_stats_arg_t *lsos = arg;
8351 
8352 	uint64_t offset = sme->sme_offset;
8353 	uint64_t vdev_id = sme->sme_vdev;
8354 
8355 	if (lsos->lsos_current_txg == 0) {
8356 		/* this is the first log */
8357 		lsos->lsos_current_txg = txg;
8358 	} else if (lsos->lsos_current_txg < txg) {
8359 		/* we just changed log - print stats and reset */
8360 		(void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8361 		    (u_longlong_t)lsos->lsos_valid_sm_entries,
8362 		    (u_longlong_t)lsos->lsos_sm_entries,
8363 		    (u_longlong_t)lsos->lsos_current_txg);
8364 		lsos->lsos_valid_sm_entries = 0;
8365 		lsos->lsos_sm_entries = 0;
8366 		lsos->lsos_current_txg = txg;
8367 	}
8368 	ASSERT3U(lsos->lsos_current_txg, ==, txg);
8369 
8370 	lsos->lsos_sm_entries++;
8371 	lsos->lsos_total_entries++;
8372 
8373 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
8374 	if (!vdev_is_concrete(vd))
8375 		return (0);
8376 
8377 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
8378 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
8379 
8380 	if (txg < metaslab_unflushed_txg(ms))
8381 		return (0);
8382 	lsos->lsos_valid_sm_entries++;
8383 	lsos->lsos_valid_entries++;
8384 	return (0);
8385 }
8386 
8387 static void
dump_log_spacemap_obsolete_stats(spa_t * spa)8388 dump_log_spacemap_obsolete_stats(spa_t *spa)
8389 {
8390 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
8391 		return;
8392 
8393 	log_sm_obsolete_stats_arg_t lsos = {0};
8394 
8395 	(void) printf("Log Space Map Obsolete Entry Statistics:\n");
8396 
8397 	iterate_through_spacemap_logs(spa,
8398 	    log_spacemap_obsolete_stats_cb, &lsos);
8399 
8400 	/* print stats for latest log */
8401 	(void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8402 	    (u_longlong_t)lsos.lsos_valid_sm_entries,
8403 	    (u_longlong_t)lsos.lsos_sm_entries,
8404 	    (u_longlong_t)lsos.lsos_current_txg);
8405 
8406 	(void) printf("%-8llu valid entries out of %-8llu - total\n\n",
8407 	    (u_longlong_t)lsos.lsos_valid_entries,
8408 	    (u_longlong_t)lsos.lsos_total_entries);
8409 }
8410 
8411 static void
dump_zpool(spa_t * spa)8412 dump_zpool(spa_t *spa)
8413 {
8414 	dsl_pool_t *dp = spa_get_dsl(spa);
8415 	int rc = 0;
8416 
8417 	if (dump_opt['y']) {
8418 		livelist_metaslab_validate(spa);
8419 	}
8420 
8421 	if (dump_opt['S']) {
8422 		dump_simulated_ddt(spa);
8423 		return;
8424 	}
8425 
8426 	if (!dump_opt['e'] && dump_opt['C'] > 1) {
8427 		(void) printf("\nCached configuration:\n");
8428 		dump_nvlist(spa->spa_config, 8);
8429 	}
8430 
8431 	if (dump_opt['C'])
8432 		dump_config(spa);
8433 
8434 	if (dump_opt['u'])
8435 		dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n");
8436 
8437 	if (dump_opt['D'])
8438 		dump_all_ddts(spa);
8439 
8440 	if (dump_opt['T'])
8441 		dump_brt(spa);
8442 
8443 	if (dump_opt['d'] > 2 || dump_opt['m'])
8444 		dump_metaslabs(spa);
8445 	if (dump_opt['M'])
8446 		dump_metaslab_groups(spa, dump_opt['M'] > 1);
8447 	if (dump_opt['d'] > 2 || dump_opt['m']) {
8448 		dump_log_spacemaps(spa);
8449 		dump_log_spacemap_obsolete_stats(spa);
8450 	}
8451 
8452 	if (dump_opt['d'] || dump_opt['i']) {
8453 		spa_feature_t f;
8454 		mos_refd_objs = zfs_range_tree_create(NULL, ZFS_RANGE_SEG64,
8455 		    NULL, 0, 0);
8456 		dump_objset(dp->dp_meta_objset);
8457 
8458 		if (dump_opt['d'] >= 3) {
8459 			dsl_pool_t *dp = spa->spa_dsl_pool;
8460 			dump_full_bpobj(&spa->spa_deferred_bpobj,
8461 			    "Deferred frees", 0);
8462 			if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
8463 				dump_full_bpobj(&dp->dp_free_bpobj,
8464 				    "Pool snapshot frees", 0);
8465 			}
8466 			if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
8467 				ASSERT(spa_feature_is_enabled(spa,
8468 				    SPA_FEATURE_DEVICE_REMOVAL));
8469 				dump_full_bpobj(&dp->dp_obsolete_bpobj,
8470 				    "Pool obsolete blocks", 0);
8471 			}
8472 
8473 			if (spa_feature_is_active(spa,
8474 			    SPA_FEATURE_ASYNC_DESTROY)) {
8475 				dump_bptree(spa->spa_meta_objset,
8476 				    dp->dp_bptree_obj,
8477 				    "Pool dataset frees");
8478 			}
8479 			dump_dtl(spa->spa_root_vdev, 0);
8480 		}
8481 
8482 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++)
8483 			global_feature_count[f] = UINT64_MAX;
8484 		global_feature_count[SPA_FEATURE_REDACTION_BOOKMARKS] = 0;
8485 		global_feature_count[SPA_FEATURE_REDACTION_LIST_SPILL] = 0;
8486 		global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN] = 0;
8487 		global_feature_count[SPA_FEATURE_LIVELIST] = 0;
8488 
8489 		(void) dmu_objset_find(spa_name(spa), dump_one_objset,
8490 		    NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
8491 
8492 		if (rc == 0 && !dump_opt['L'])
8493 			rc = dump_mos_leaks(spa);
8494 
8495 		for (f = 0; f < SPA_FEATURES; f++) {
8496 			uint64_t refcount;
8497 
8498 			uint64_t *arr;
8499 			if (!(spa_feature_table[f].fi_flags &
8500 			    ZFEATURE_FLAG_PER_DATASET)) {
8501 				if (global_feature_count[f] == UINT64_MAX)
8502 					continue;
8503 				if (!spa_feature_is_enabled(spa, f)) {
8504 					ASSERT0(global_feature_count[f]);
8505 					continue;
8506 				}
8507 				arr = global_feature_count;
8508 			} else {
8509 				if (!spa_feature_is_enabled(spa, f)) {
8510 					ASSERT0(dataset_feature_count[f]);
8511 					continue;
8512 				}
8513 				arr = dataset_feature_count;
8514 			}
8515 			if (feature_get_refcount(spa, &spa_feature_table[f],
8516 			    &refcount) == ENOTSUP)
8517 				continue;
8518 			if (arr[f] != refcount) {
8519 				(void) printf("%s feature refcount mismatch: "
8520 				    "%lld consumers != %lld refcount\n",
8521 				    spa_feature_table[f].fi_uname,
8522 				    (longlong_t)arr[f], (longlong_t)refcount);
8523 				rc = 2;
8524 			} else {
8525 				(void) printf("Verified %s feature refcount "
8526 				    "of %llu is correct\n",
8527 				    spa_feature_table[f].fi_uname,
8528 				    (longlong_t)refcount);
8529 			}
8530 		}
8531 
8532 		if (rc == 0)
8533 			rc = verify_device_removal_feature_counts(spa);
8534 	}
8535 
8536 	if (rc == 0 && (dump_opt['b'] || dump_opt['c']))
8537 		rc = dump_block_stats(spa);
8538 
8539 	if (rc == 0)
8540 		rc = verify_spacemap_refcounts(spa);
8541 
8542 	if (dump_opt['s'])
8543 		show_pool_stats(spa);
8544 
8545 	if (dump_opt['h'])
8546 		dump_history(spa);
8547 
8548 	if (rc == 0)
8549 		rc = verify_checkpoint(spa);
8550 
8551 	if (rc != 0) {
8552 		dump_debug_buffer();
8553 		zdb_exit(rc);
8554 	}
8555 }
8556 
8557 #define	ZDB_FLAG_CHECKSUM	0x0001
8558 #define	ZDB_FLAG_DECOMPRESS	0x0002
8559 #define	ZDB_FLAG_BSWAP		0x0004
8560 #define	ZDB_FLAG_GBH		0x0008
8561 #define	ZDB_FLAG_INDIRECT	0x0010
8562 #define	ZDB_FLAG_RAW		0x0020
8563 #define	ZDB_FLAG_PRINT_BLKPTR	0x0040
8564 #define	ZDB_FLAG_VERBOSE	0x0080
8565 
8566 static int flagbits[256];
8567 static char flagbitstr[16];
8568 
8569 static void
zdb_print_blkptr(const blkptr_t * bp,int flags)8570 zdb_print_blkptr(const blkptr_t *bp, int flags)
8571 {
8572 	char blkbuf[BP_SPRINTF_LEN];
8573 
8574 	if (flags & ZDB_FLAG_BSWAP)
8575 		byteswap_uint64_array((void *)bp, sizeof (blkptr_t));
8576 
8577 	snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
8578 	(void) printf("%s\n", blkbuf);
8579 }
8580 
8581 static void
zdb_dump_indirect(blkptr_t * bp,int nbps,int flags)8582 zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
8583 {
8584 	int i;
8585 
8586 	for (i = 0; i < nbps; i++)
8587 		zdb_print_blkptr(&bp[i], flags);
8588 }
8589 
8590 static void
zdb_dump_gbh(void * buf,int flags)8591 zdb_dump_gbh(void *buf, int flags)
8592 {
8593 	zdb_dump_indirect((blkptr_t *)buf, SPA_GBH_NBLKPTRS, flags);
8594 }
8595 
8596 static void
zdb_dump_block_raw(void * buf,uint64_t size,int flags)8597 zdb_dump_block_raw(void *buf, uint64_t size, int flags)
8598 {
8599 	if (flags & ZDB_FLAG_BSWAP)
8600 		byteswap_uint64_array(buf, size);
8601 	VERIFY(write(fileno(stdout), buf, size) == size);
8602 }
8603 
8604 static void
zdb_dump_block(char * label,void * buf,uint64_t size,int flags)8605 zdb_dump_block(char *label, void *buf, uint64_t size, int flags)
8606 {
8607 	uint64_t *d = (uint64_t *)buf;
8608 	unsigned nwords = size / sizeof (uint64_t);
8609 	int do_bswap = !!(flags & ZDB_FLAG_BSWAP);
8610 	unsigned i, j;
8611 	const char *hdr;
8612 	char *c;
8613 
8614 
8615 	if (do_bswap)
8616 		hdr = " 7 6 5 4 3 2 1 0   f e d c b a 9 8";
8617 	else
8618 		hdr = " 0 1 2 3 4 5 6 7   8 9 a b c d e f";
8619 
8620 	(void) printf("\n%s\n%6s   %s  0123456789abcdef\n", label, "", hdr);
8621 
8622 #ifdef _ZFS_LITTLE_ENDIAN
8623 	/* correct the endianness */
8624 	do_bswap = !do_bswap;
8625 #endif
8626 	for (i = 0; i < nwords; i += 2) {
8627 		(void) printf("%06llx:  %016llx  %016llx  ",
8628 		    (u_longlong_t)(i * sizeof (uint64_t)),
8629 		    (u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]),
8630 		    (u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1]));
8631 
8632 		c = (char *)&d[i];
8633 		for (j = 0; j < 2 * sizeof (uint64_t); j++)
8634 			(void) printf("%c", isprint(c[j]) ? c[j] : '.');
8635 		(void) printf("\n");
8636 	}
8637 }
8638 
8639 /*
8640  * There are two acceptable formats:
8641  *	leaf_name	  - For example: c1t0d0 or /tmp/ztest.0a
8642  *	child[.child]*    - For example: 0.1.1
8643  *
8644  * The second form can be used to specify arbitrary vdevs anywhere
8645  * in the hierarchy.  For example, in a pool with a mirror of
8646  * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
8647  */
8648 static vdev_t *
zdb_vdev_lookup(vdev_t * vdev,const char * path)8649 zdb_vdev_lookup(vdev_t *vdev, const char *path)
8650 {
8651 	char *s, *p, *q;
8652 	unsigned i;
8653 
8654 	if (vdev == NULL)
8655 		return (NULL);
8656 
8657 	/* First, assume the x.x.x.x format */
8658 	i = strtoul(path, &s, 10);
8659 	if (s == path || (s && *s != '.' && *s != '\0'))
8660 		goto name;
8661 	if (i >= vdev->vdev_children)
8662 		return (NULL);
8663 
8664 	vdev = vdev->vdev_child[i];
8665 	if (s && *s == '\0')
8666 		return (vdev);
8667 	return (zdb_vdev_lookup(vdev, s+1));
8668 
8669 name:
8670 	for (i = 0; i < vdev->vdev_children; i++) {
8671 		vdev_t *vc = vdev->vdev_child[i];
8672 
8673 		if (vc->vdev_path == NULL) {
8674 			vc = zdb_vdev_lookup(vc, path);
8675 			if (vc == NULL)
8676 				continue;
8677 			else
8678 				return (vc);
8679 		}
8680 
8681 		p = strrchr(vc->vdev_path, '/');
8682 		p = p ? p + 1 : vc->vdev_path;
8683 		q = &vc->vdev_path[strlen(vc->vdev_path) - 2];
8684 
8685 		if (strcmp(vc->vdev_path, path) == 0)
8686 			return (vc);
8687 		if (strcmp(p, path) == 0)
8688 			return (vc);
8689 		if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0)
8690 			return (vc);
8691 	}
8692 
8693 	return (NULL);
8694 }
8695 
8696 static int
name_from_objset_id(spa_t * spa,uint64_t objset_id,char * outstr)8697 name_from_objset_id(spa_t *spa, uint64_t objset_id, char *outstr)
8698 {
8699 	dsl_dataset_t *ds;
8700 
8701 	dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
8702 	int error = dsl_dataset_hold_obj(spa->spa_dsl_pool, objset_id,
8703 	    NULL, &ds);
8704 	if (error != 0) {
8705 		(void) fprintf(stderr, "failed to hold objset %llu: %s\n",
8706 		    (u_longlong_t)objset_id, strerror(error));
8707 		dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8708 		return (error);
8709 	}
8710 	dsl_dataset_name(ds, outstr);
8711 	dsl_dataset_rele(ds, NULL);
8712 	dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8713 	return (0);
8714 }
8715 
8716 static boolean_t
zdb_parse_block_sizes(char * sizes,uint64_t * lsize,uint64_t * psize)8717 zdb_parse_block_sizes(char *sizes, uint64_t *lsize, uint64_t *psize)
8718 {
8719 	char *s0, *s1, *tmp = NULL;
8720 
8721 	if (sizes == NULL)
8722 		return (B_FALSE);
8723 
8724 	s0 = strtok_r(sizes, "/", &tmp);
8725 	if (s0 == NULL)
8726 		return (B_FALSE);
8727 	s1 = strtok_r(NULL, "/", &tmp);
8728 	*lsize = strtoull(s0, NULL, 16);
8729 	*psize = s1 ? strtoull(s1, NULL, 16) : *lsize;
8730 	return (*lsize >= *psize && *psize > 0);
8731 }
8732 
8733 #define	ZIO_COMPRESS_MASK(alg)	(1ULL << (ZIO_COMPRESS_##alg))
8734 
8735 static boolean_t
try_decompress_block(abd_t * pabd,uint64_t lsize,uint64_t psize,int flags,int cfunc,void * lbuf,void * lbuf2)8736 try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize,
8737     int flags, int cfunc, void *lbuf, void *lbuf2)
8738 {
8739 	if (flags & ZDB_FLAG_VERBOSE) {
8740 		(void) fprintf(stderr,
8741 		    "Trying %05llx -> %05llx (%s)\n",
8742 		    (u_longlong_t)psize,
8743 		    (u_longlong_t)lsize,
8744 		    zio_compress_table[cfunc].ci_name);
8745 	}
8746 
8747 	/*
8748 	 * We set lbuf to all zeros and lbuf2 to all
8749 	 * ones, then decompress to both buffers and
8750 	 * compare their contents. This way we can
8751 	 * know if decompression filled exactly to
8752 	 * lsize or if it left some bytes unwritten.
8753 	 */
8754 
8755 	memset(lbuf, 0x00, lsize);
8756 	memset(lbuf2, 0xff, lsize);
8757 
8758 	abd_t labd, labd2;
8759 	abd_get_from_buf_struct(&labd, lbuf, lsize);
8760 	abd_get_from_buf_struct(&labd2, lbuf2, lsize);
8761 
8762 	boolean_t ret = B_FALSE;
8763 	if (zio_decompress_data(cfunc, pabd,
8764 	    &labd, psize, lsize, NULL) == 0 &&
8765 	    zio_decompress_data(cfunc, pabd,
8766 	    &labd2, psize, lsize, NULL) == 0 &&
8767 	    memcmp(lbuf, lbuf2, lsize) == 0)
8768 		ret = B_TRUE;
8769 
8770 	abd_free(&labd2);
8771 	abd_free(&labd);
8772 
8773 	return (ret);
8774 }
8775 
8776 static uint64_t
zdb_decompress_block(abd_t * pabd,void * buf,void * lbuf,uint64_t lsize,uint64_t psize,int flags)8777 zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
8778     uint64_t psize, int flags)
8779 {
8780 	(void) buf;
8781 	uint64_t orig_lsize = lsize;
8782 	boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL));
8783 	boolean_t found = B_FALSE;
8784 	/*
8785 	 * We don't know how the data was compressed, so just try
8786 	 * every decompress function at every inflated blocksize.
8787 	 */
8788 	void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
8789 	int cfuncs[ZIO_COMPRESS_FUNCTIONS] = { 0 };
8790 	int *cfuncp = cfuncs;
8791 	uint64_t maxlsize = SPA_MAXBLOCKSIZE;
8792 	uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) |
8793 	    ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) |
8794 	    ZIO_COMPRESS_MASK(ZLE);
8795 	*cfuncp++ = ZIO_COMPRESS_LZ4;
8796 	*cfuncp++ = ZIO_COMPRESS_LZJB;
8797 	mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB);
8798 	/*
8799 	 * Every gzip level has the same decompressor, no need to
8800 	 * run it 9 times per bruteforce attempt.
8801 	 */
8802 	mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3);
8803 	mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5);
8804 	mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7);
8805 	mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9);
8806 	for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++)
8807 		if (((1ULL << c) & mask) == 0)
8808 			*cfuncp++ = c;
8809 
8810 	/*
8811 	 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
8812 	 * could take a while and we should let the user know
8813 	 * we are not stuck.  On the other hand, printing progress
8814 	 * info gets old after a while.  User can specify 'v' flag
8815 	 * to see the progression.
8816 	 */
8817 	if (lsize == psize)
8818 		lsize += SPA_MINBLOCKSIZE;
8819 	else
8820 		maxlsize = lsize;
8821 
8822 	for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) {
8823 		for (cfuncp = cfuncs; *cfuncp; cfuncp++) {
8824 			if (try_decompress_block(pabd, lsize, psize, flags,
8825 			    *cfuncp, lbuf, lbuf2)) {
8826 				found = B_TRUE;
8827 				break;
8828 			}
8829 		}
8830 		if (*cfuncp != 0)
8831 			break;
8832 	}
8833 	if (!found && tryzle) {
8834 		for (lsize = orig_lsize; lsize <= maxlsize;
8835 		    lsize += SPA_MINBLOCKSIZE) {
8836 			if (try_decompress_block(pabd, lsize, psize, flags,
8837 			    ZIO_COMPRESS_ZLE, lbuf, lbuf2)) {
8838 				*cfuncp = ZIO_COMPRESS_ZLE;
8839 				found = B_TRUE;
8840 				break;
8841 			}
8842 		}
8843 	}
8844 	umem_free(lbuf2, SPA_MAXBLOCKSIZE);
8845 
8846 	if (*cfuncp == ZIO_COMPRESS_ZLE) {
8847 		printf("\nZLE decompression was selected. If you "
8848 		    "suspect the results are wrong,\ntry avoiding ZLE "
8849 		    "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
8850 	}
8851 
8852 	return (lsize > maxlsize ? -1 : lsize);
8853 }
8854 
8855 /*
8856  * Read a block from a pool and print it out.  The syntax of the
8857  * block descriptor is:
8858  *
8859  *	pool:vdev_specifier:offset:[lsize/]psize[:flags]
8860  *
8861  *	pool           - The name of the pool you wish to read from
8862  *	vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
8863  *	offset         - offset, in hex, in bytes
8864  *	size           - Amount of data to read, in hex, in bytes
8865  *	flags          - A string of characters specifying options
8866  *		 b: Decode a blkptr at given offset within block
8867  *		 c: Calculate and display checksums
8868  *		 d: Decompress data before dumping
8869  *		 e: Byteswap data before dumping
8870  *		 g: Display data as a gang block header
8871  *		 i: Display as an indirect block
8872  *		 r: Dump raw data to stdout
8873  *		 v: Verbose
8874  *
8875  */
8876 static void
zdb_read_block(char * thing,spa_t * spa)8877 zdb_read_block(char *thing, spa_t *spa)
8878 {
8879 	blkptr_t blk, *bp = &blk;
8880 	dva_t *dva = bp->blk_dva;
8881 	int flags = 0;
8882 	uint64_t offset = 0, psize = 0, lsize = 0, blkptr_offset = 0;
8883 	zio_t *zio;
8884 	vdev_t *vd;
8885 	abd_t *pabd;
8886 	void *lbuf, *buf;
8887 	char *s, *p, *dup, *flagstr, *sizes, *tmp = NULL;
8888 	const char *vdev, *errmsg = NULL;
8889 	int i, len, error;
8890 	boolean_t borrowed = B_FALSE, found = B_FALSE;
8891 
8892 	dup = strdup(thing);
8893 	s = strtok_r(dup, ":", &tmp);
8894 	vdev = s ?: "";
8895 	s = strtok_r(NULL, ":", &tmp);
8896 	offset = strtoull(s ? s : "", NULL, 16);
8897 	sizes = strtok_r(NULL, ":", &tmp);
8898 	s = strtok_r(NULL, ":", &tmp);
8899 	flagstr = strdup(s ?: "");
8900 
8901 	if (!zdb_parse_block_sizes(sizes, &lsize, &psize))
8902 		errmsg = "invalid size(s)";
8903 	if (!IS_P2ALIGNED(psize, DEV_BSIZE) || !IS_P2ALIGNED(lsize, DEV_BSIZE))
8904 		errmsg = "size must be a multiple of sector size";
8905 	if (!IS_P2ALIGNED(offset, DEV_BSIZE))
8906 		errmsg = "offset must be a multiple of sector size";
8907 	if (errmsg) {
8908 		(void) printf("Invalid block specifier: %s  - %s\n",
8909 		    thing, errmsg);
8910 		goto done;
8911 	}
8912 
8913 	tmp = NULL;
8914 	for (s = strtok_r(flagstr, ":", &tmp);
8915 	    s != NULL;
8916 	    s = strtok_r(NULL, ":", &tmp)) {
8917 		len = strlen(flagstr);
8918 		for (i = 0; i < len; i++) {
8919 			int bit = flagbits[(uchar_t)flagstr[i]];
8920 
8921 			if (bit == 0) {
8922 				(void) printf("***Ignoring flag: %c\n",
8923 				    (uchar_t)flagstr[i]);
8924 				continue;
8925 			}
8926 			found = B_TRUE;
8927 			flags |= bit;
8928 
8929 			p = &flagstr[i + 1];
8930 			if (*p != ':' && *p != '\0') {
8931 				int j = 0, nextbit = flagbits[(uchar_t)*p];
8932 				char *end, offstr[8] = { 0 };
8933 				if ((bit == ZDB_FLAG_PRINT_BLKPTR) &&
8934 				    (nextbit == 0)) {
8935 					/* look ahead to isolate the offset */
8936 					while (nextbit == 0 &&
8937 					    strchr(flagbitstr, *p) == NULL) {
8938 						offstr[j] = *p;
8939 						j++;
8940 						if (i + j > strlen(flagstr))
8941 							break;
8942 						p++;
8943 						nextbit = flagbits[(uchar_t)*p];
8944 					}
8945 					blkptr_offset = strtoull(offstr, &end,
8946 					    16);
8947 					i += j;
8948 				} else if (nextbit == 0) {
8949 					(void) printf("***Ignoring flag arg:"
8950 					    " '%c'\n", (uchar_t)*p);
8951 				}
8952 			}
8953 		}
8954 	}
8955 	if (blkptr_offset % sizeof (blkptr_t)) {
8956 		printf("Block pointer offset 0x%llx "
8957 		    "must be divisible by 0x%x\n",
8958 		    (longlong_t)blkptr_offset, (int)sizeof (blkptr_t));
8959 		goto done;
8960 	}
8961 	if (found == B_FALSE && strlen(flagstr) > 0) {
8962 		printf("Invalid flag arg: '%s'\n", flagstr);
8963 		goto done;
8964 	}
8965 
8966 	vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev);
8967 	if (vd == NULL) {
8968 		(void) printf("***Invalid vdev: %s\n", vdev);
8969 		goto done;
8970 	} else {
8971 		if (vd->vdev_path)
8972 			(void) fprintf(stderr, "Found vdev: %s\n",
8973 			    vd->vdev_path);
8974 		else
8975 			(void) fprintf(stderr, "Found vdev type: %s\n",
8976 			    vd->vdev_ops->vdev_op_type);
8977 	}
8978 
8979 	pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
8980 	lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
8981 
8982 	BP_ZERO(bp);
8983 
8984 	DVA_SET_VDEV(&dva[0], vd->vdev_id);
8985 	DVA_SET_OFFSET(&dva[0], offset);
8986 	DVA_SET_GANG(&dva[0], 0);
8987 	DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
8988 
8989 	BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
8990 
8991 	BP_SET_LSIZE(bp, lsize);
8992 	BP_SET_PSIZE(bp, psize);
8993 	BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF);
8994 	BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF);
8995 	BP_SET_TYPE(bp, DMU_OT_NONE);
8996 	BP_SET_LEVEL(bp, 0);
8997 	BP_SET_DEDUP(bp, 0);
8998 	BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
8999 
9000 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9001 	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9002 
9003 	if (vd == vd->vdev_top) {
9004 		/*
9005 		 * Treat this as a normal block read.
9006 		 */
9007 		zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL,
9008 		    ZIO_PRIORITY_SYNC_READ,
9009 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL));
9010 	} else {
9011 		/*
9012 		 * Treat this as a vdev child I/O.
9013 		 */
9014 		zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd,
9015 		    psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ,
9016 		    ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
9017 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | ZIO_FLAG_OPTIONAL,
9018 		    NULL, NULL));
9019 	}
9020 
9021 	error = zio_wait(zio);
9022 	spa_config_exit(spa, SCL_STATE, FTAG);
9023 
9024 	if (error) {
9025 		(void) printf("Read of %s failed, error: %d\n", thing, error);
9026 		goto out;
9027 	}
9028 
9029 	uint64_t orig_lsize = lsize;
9030 	buf = lbuf;
9031 	if (flags & ZDB_FLAG_DECOMPRESS) {
9032 		lsize = zdb_decompress_block(pabd, buf, lbuf,
9033 		    lsize, psize, flags);
9034 		if (lsize == -1) {
9035 			(void) printf("Decompress of %s failed\n", thing);
9036 			goto out;
9037 		}
9038 	} else {
9039 		buf = abd_borrow_buf_copy(pabd, lsize);
9040 		borrowed = B_TRUE;
9041 	}
9042 	/*
9043 	 * Try to detect invalid block pointer.  If invalid, try
9044 	 * decompressing.
9045 	 */
9046 	if ((flags & ZDB_FLAG_PRINT_BLKPTR || flags & ZDB_FLAG_INDIRECT) &&
9047 	    !(flags & ZDB_FLAG_DECOMPRESS)) {
9048 		const blkptr_t *b = (const blkptr_t *)(void *)
9049 		    ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9050 		if (zfs_blkptr_verify(spa, b,
9051 		    BLK_CONFIG_NEEDED, BLK_VERIFY_ONLY)) {
9052 			abd_return_buf_copy(pabd, buf, lsize);
9053 			borrowed = B_FALSE;
9054 			buf = lbuf;
9055 			lsize = zdb_decompress_block(pabd, buf,
9056 			    lbuf, lsize, psize, flags);
9057 			b = (const blkptr_t *)(void *)
9058 			    ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9059 			if (lsize == -1 || zfs_blkptr_verify(spa, b,
9060 			    BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) {
9061 				printf("invalid block pointer at this DVA\n");
9062 				goto out;
9063 			}
9064 		}
9065 	}
9066 
9067 	if (flags & ZDB_FLAG_PRINT_BLKPTR)
9068 		zdb_print_blkptr((blkptr_t *)(void *)
9069 		    ((uintptr_t)buf + (uintptr_t)blkptr_offset), flags);
9070 	else if (flags & ZDB_FLAG_RAW)
9071 		zdb_dump_block_raw(buf, lsize, flags);
9072 	else if (flags & ZDB_FLAG_INDIRECT)
9073 		zdb_dump_indirect((blkptr_t *)buf,
9074 		    orig_lsize / sizeof (blkptr_t), flags);
9075 	else if (flags & ZDB_FLAG_GBH)
9076 		zdb_dump_gbh(buf, flags);
9077 	else
9078 		zdb_dump_block(thing, buf, lsize, flags);
9079 
9080 	/*
9081 	 * If :c was specified, iterate through the checksum table to
9082 	 * calculate and display each checksum for our specified
9083 	 * DVA and length.
9084 	 */
9085 	if ((flags & ZDB_FLAG_CHECKSUM) && !(flags & ZDB_FLAG_RAW) &&
9086 	    !(flags & ZDB_FLAG_GBH)) {
9087 		zio_t *czio;
9088 		(void) printf("\n");
9089 		for (enum zio_checksum ck = ZIO_CHECKSUM_LABEL;
9090 		    ck < ZIO_CHECKSUM_FUNCTIONS; ck++) {
9091 
9092 			if ((zio_checksum_table[ck].ci_flags &
9093 			    ZCHECKSUM_FLAG_EMBEDDED) ||
9094 			    ck == ZIO_CHECKSUM_NOPARITY) {
9095 				continue;
9096 			}
9097 			BP_SET_CHECKSUM(bp, ck);
9098 			spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9099 			czio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9100 			if (vd == vd->vdev_top) {
9101 				zio_nowait(zio_read(czio, spa, bp, pabd, psize,
9102 				    NULL, NULL,
9103 				    ZIO_PRIORITY_SYNC_READ,
9104 				    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9105 				    ZIO_FLAG_DONT_RETRY, NULL));
9106 			} else {
9107 				zio_nowait(zio_vdev_child_io(czio, bp, vd,
9108 				    offset, pabd, psize, ZIO_TYPE_READ,
9109 				    ZIO_PRIORITY_SYNC_READ,
9110 				    ZIO_FLAG_DONT_PROPAGATE |
9111 				    ZIO_FLAG_DONT_RETRY |
9112 				    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9113 				    ZIO_FLAG_SPECULATIVE |
9114 				    ZIO_FLAG_OPTIONAL, NULL, NULL));
9115 			}
9116 			error = zio_wait(czio);
9117 			if (error == 0 || error == ECKSUM) {
9118 				zio_t *ck_zio = zio_null(NULL, spa, NULL,
9119 				    NULL, NULL, 0);
9120 				ck_zio->io_offset =
9121 				    DVA_GET_OFFSET(&bp->blk_dva[0]);
9122 				ck_zio->io_bp = bp;
9123 				zio_checksum_compute(ck_zio, ck, pabd, lsize);
9124 				printf(
9125 				    "%12s\t"
9126 				    "cksum=%016llx:%016llx:%016llx:%016llx\n",
9127 				    zio_checksum_table[ck].ci_name,
9128 				    (u_longlong_t)bp->blk_cksum.zc_word[0],
9129 				    (u_longlong_t)bp->blk_cksum.zc_word[1],
9130 				    (u_longlong_t)bp->blk_cksum.zc_word[2],
9131 				    (u_longlong_t)bp->blk_cksum.zc_word[3]);
9132 				zio_wait(ck_zio);
9133 			} else {
9134 				printf("error %d reading block\n", error);
9135 			}
9136 			spa_config_exit(spa, SCL_STATE, FTAG);
9137 		}
9138 	}
9139 
9140 	if (borrowed)
9141 		abd_return_buf_copy(pabd, buf, lsize);
9142 
9143 out:
9144 	abd_free(pabd);
9145 	umem_free(lbuf, SPA_MAXBLOCKSIZE);
9146 done:
9147 	free(flagstr);
9148 	free(dup);
9149 }
9150 
9151 static void
zdb_embedded_block(char * thing)9152 zdb_embedded_block(char *thing)
9153 {
9154 	blkptr_t bp = {{{{0}}}};
9155 	unsigned long long *words = (void *)&bp;
9156 	char *buf;
9157 	int err;
9158 
9159 	err = sscanf(thing, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
9160 	    "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
9161 	    words + 0, words + 1, words + 2, words + 3,
9162 	    words + 4, words + 5, words + 6, words + 7,
9163 	    words + 8, words + 9, words + 10, words + 11,
9164 	    words + 12, words + 13, words + 14, words + 15);
9165 	if (err != 16) {
9166 		(void) fprintf(stderr, "invalid input format\n");
9167 		zdb_exit(1);
9168 	}
9169 	ASSERT3U(BPE_GET_LSIZE(&bp), <=, SPA_MAXBLOCKSIZE);
9170 	buf = malloc(SPA_MAXBLOCKSIZE);
9171 	if (buf == NULL) {
9172 		(void) fprintf(stderr, "out of memory\n");
9173 		zdb_exit(1);
9174 	}
9175 	err = decode_embedded_bp(&bp, buf, BPE_GET_LSIZE(&bp));
9176 	if (err != 0) {
9177 		(void) fprintf(stderr, "decode failed: %u\n", err);
9178 		zdb_exit(1);
9179 	}
9180 	zdb_dump_block_raw(buf, BPE_GET_LSIZE(&bp), 0);
9181 	free(buf);
9182 }
9183 
9184 /* check for valid hex or decimal numeric string */
9185 static boolean_t
zdb_numeric(char * str)9186 zdb_numeric(char *str)
9187 {
9188 	int i = 0, len;
9189 
9190 	len = strlen(str);
9191 	if (len == 0)
9192 		return (B_FALSE);
9193 	if (strncmp(str, "0x", 2) == 0 || strncmp(str, "0X", 2) == 0)
9194 		i = 2;
9195 	for (; i < len; i++) {
9196 		if (!isxdigit(str[i]))
9197 			return (B_FALSE);
9198 	}
9199 	return (B_TRUE);
9200 }
9201 
9202 static int
dummy_get_file_info(dmu_object_type_t bonustype,const void * data,zfs_file_info_t * zoi)9203 dummy_get_file_info(dmu_object_type_t bonustype, const void *data,
9204     zfs_file_info_t *zoi)
9205 {
9206 	(void) data, (void) zoi;
9207 
9208 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
9209 		return (ENOENT);
9210 
9211 	(void) fprintf(stderr, "dummy_get_file_info: not implemented");
9212 	abort();
9213 }
9214 
9215 int
main(int argc,char ** argv)9216 main(int argc, char **argv)
9217 {
9218 	int c;
9219 	int dump_all = 1;
9220 	int verbose = 0;
9221 	int error = 0;
9222 	char **searchdirs = NULL;
9223 	int nsearch = 0;
9224 	char *target, *target_pool, dsname[ZFS_MAX_DATASET_NAME_LEN];
9225 	nvlist_t *policy = NULL;
9226 	uint64_t max_txg = UINT64_MAX;
9227 	int64_t objset_id = -1;
9228 	uint64_t object;
9229 	int flags = ZFS_IMPORT_MISSING_LOG;
9230 	int rewind = ZPOOL_NEVER_REWIND;
9231 	char *spa_config_path_env, *objset_str;
9232 	boolean_t target_is_spa = B_TRUE, dataset_lookup = B_FALSE;
9233 	nvlist_t *cfg = NULL;
9234 	struct sigaction action;
9235 	boolean_t force_import = B_FALSE;
9236 	boolean_t config_path_console = B_FALSE;
9237 	char pbuf[MAXPATHLEN];
9238 
9239 	dprintf_setup(&argc, argv);
9240 
9241 	/*
9242 	 * Set up signal handlers, so if we crash due to bad on-disk data we
9243 	 * can get more info. Unlike ztest, we don't bail out if we can't set
9244 	 * up signal handlers, because zdb is very useful without them.
9245 	 */
9246 	action.sa_handler = sig_handler;
9247 	sigemptyset(&action.sa_mask);
9248 	action.sa_flags = 0;
9249 	if (sigaction(SIGSEGV, &action, NULL) < 0) {
9250 		(void) fprintf(stderr, "zdb: cannot catch SIGSEGV: %s\n",
9251 		    strerror(errno));
9252 	}
9253 	if (sigaction(SIGABRT, &action, NULL) < 0) {
9254 		(void) fprintf(stderr, "zdb: cannot catch SIGABRT: %s\n",
9255 		    strerror(errno));
9256 	}
9257 
9258 	/*
9259 	 * If there is an environment variable SPA_CONFIG_PATH it overrides
9260 	 * default spa_config_path setting. If -U flag is specified it will
9261 	 * override this environment variable settings once again.
9262 	 */
9263 	spa_config_path_env = getenv("SPA_CONFIG_PATH");
9264 	if (spa_config_path_env != NULL)
9265 		spa_config_path = spa_config_path_env;
9266 
9267 	/*
9268 	 * For performance reasons, we set this tunable down. We do so before
9269 	 * the arg parsing section so that the user can override this value if
9270 	 * they choose.
9271 	 */
9272 	zfs_btree_verify_intensity = 3;
9273 
9274 	struct option long_options[] = {
9275 		{"ignore-assertions",	no_argument,		NULL, 'A'},
9276 		{"block-stats",		no_argument,		NULL, 'b'},
9277 		{"backup",		no_argument,		NULL, 'B'},
9278 		{"checksum",		no_argument,		NULL, 'c'},
9279 		{"config",		no_argument,		NULL, 'C'},
9280 		{"datasets",		no_argument,		NULL, 'd'},
9281 		{"dedup-stats",		no_argument,		NULL, 'D'},
9282 		{"exported",		no_argument,		NULL, 'e'},
9283 		{"embedded-block-pointer",	no_argument,	NULL, 'E'},
9284 		{"automatic-rewind",	no_argument,		NULL, 'F'},
9285 		{"dump-debug-msg",	no_argument,		NULL, 'G'},
9286 		{"history",		no_argument,		NULL, 'h'},
9287 		{"intent-logs",		no_argument,		NULL, 'i'},
9288 		{"inflight",		required_argument,	NULL, 'I'},
9289 		{"checkpointed-state",	no_argument,		NULL, 'k'},
9290 		{"key",			required_argument,	NULL, 'K'},
9291 		{"label",		no_argument,		NULL, 'l'},
9292 		{"disable-leak-tracking",	no_argument,	NULL, 'L'},
9293 		{"metaslabs",		no_argument,		NULL, 'm'},
9294 		{"metaslab-groups",	no_argument,		NULL, 'M'},
9295 		{"numeric",		no_argument,		NULL, 'N'},
9296 		{"option",		required_argument,	NULL, 'o'},
9297 		{"object-lookups",	no_argument,		NULL, 'O'},
9298 		{"path",		required_argument,	NULL, 'p'},
9299 		{"parseable",		no_argument,		NULL, 'P'},
9300 		{"skip-label",		no_argument,		NULL, 'q'},
9301 		{"copy-object",		no_argument,		NULL, 'r'},
9302 		{"read-block",		no_argument,		NULL, 'R'},
9303 		{"io-stats",		no_argument,		NULL, 's'},
9304 		{"simulate-dedup",	no_argument,		NULL, 'S'},
9305 		{"txg",			required_argument,	NULL, 't'},
9306 		{"brt-stats",		no_argument,		NULL, 'T'},
9307 		{"uberblock",		no_argument,		NULL, 'u'},
9308 		{"cachefile",		required_argument,	NULL, 'U'},
9309 		{"verbose",		no_argument,		NULL, 'v'},
9310 		{"verbatim",		no_argument,		NULL, 'V'},
9311 		{"dump-blocks",		required_argument,	NULL, 'x'},
9312 		{"extreme-rewind",	no_argument,		NULL, 'X'},
9313 		{"all-reconstruction",	no_argument,		NULL, 'Y'},
9314 		{"livelist",		no_argument,		NULL, 'y'},
9315 		{"zstd-headers",	no_argument,		NULL, 'Z'},
9316 		{0, 0, 0, 0}
9317 	};
9318 
9319 	while ((c = getopt_long(argc, argv,
9320 	    "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
9321 	    long_options, NULL)) != -1) {
9322 		switch (c) {
9323 		case 'b':
9324 		case 'B':
9325 		case 'c':
9326 		case 'C':
9327 		case 'd':
9328 		case 'D':
9329 		case 'E':
9330 		case 'G':
9331 		case 'h':
9332 		case 'i':
9333 		case 'l':
9334 		case 'm':
9335 		case 'M':
9336 		case 'N':
9337 		case 'O':
9338 		case 'r':
9339 		case 'R':
9340 		case 's':
9341 		case 'S':
9342 		case 'T':
9343 		case 'u':
9344 		case 'y':
9345 		case 'Z':
9346 			dump_opt[c]++;
9347 			dump_all = 0;
9348 			break;
9349 		case 'A':
9350 		case 'e':
9351 		case 'F':
9352 		case 'k':
9353 		case 'L':
9354 		case 'P':
9355 		case 'q':
9356 		case 'X':
9357 			dump_opt[c]++;
9358 			break;
9359 		case 'Y':
9360 			zfs_reconstruct_indirect_combinations_max = INT_MAX;
9361 			zfs_deadman_enabled = 0;
9362 			break;
9363 		/* NB: Sort single match options below. */
9364 		case 'I':
9365 			max_inflight_bytes = strtoull(optarg, NULL, 0);
9366 			if (max_inflight_bytes == 0) {
9367 				(void) fprintf(stderr, "maximum number "
9368 				    "of inflight bytes must be greater "
9369 				    "than 0\n");
9370 				usage();
9371 			}
9372 			break;
9373 		case 'K':
9374 			dump_opt[c]++;
9375 			key_material = strdup(optarg);
9376 			/* redact key material in process table */
9377 			while (*optarg != '\0') { *optarg++ = '*'; }
9378 			break;
9379 		case 'o':
9380 			error = set_global_var(optarg);
9381 			if (error != 0)
9382 				usage();
9383 			break;
9384 		case 'p':
9385 			if (searchdirs == NULL) {
9386 				searchdirs = umem_alloc(sizeof (char *),
9387 				    UMEM_NOFAIL);
9388 			} else {
9389 				char **tmp = umem_alloc((nsearch + 1) *
9390 				    sizeof (char *), UMEM_NOFAIL);
9391 				memcpy(tmp, searchdirs, nsearch *
9392 				    sizeof (char *));
9393 				umem_free(searchdirs,
9394 				    nsearch * sizeof (char *));
9395 				searchdirs = tmp;
9396 			}
9397 			searchdirs[nsearch++] = optarg;
9398 			break;
9399 		case 't':
9400 			max_txg = strtoull(optarg, NULL, 0);
9401 			if (max_txg < TXG_INITIAL) {
9402 				(void) fprintf(stderr, "incorrect txg "
9403 				    "specified: %s\n", optarg);
9404 				usage();
9405 			}
9406 			break;
9407 		case 'U':
9408 			config_path_console = B_TRUE;
9409 			spa_config_path = optarg;
9410 			if (spa_config_path[0] != '/') {
9411 				(void) fprintf(stderr,
9412 				    "cachefile must be an absolute path "
9413 				    "(i.e. start with a slash)\n");
9414 				usage();
9415 			}
9416 			break;
9417 		case 'v':
9418 			verbose++;
9419 			break;
9420 		case 'V':
9421 			flags = ZFS_IMPORT_VERBATIM;
9422 			break;
9423 		case 'x':
9424 			vn_dumpdir = optarg;
9425 			break;
9426 		default:
9427 			usage();
9428 			break;
9429 		}
9430 	}
9431 
9432 	if (!dump_opt['e'] && searchdirs != NULL) {
9433 		(void) fprintf(stderr, "-p option requires use of -e\n");
9434 		usage();
9435 	}
9436 #if defined(_LP64)
9437 	/*
9438 	 * ZDB does not typically re-read blocks; therefore limit the ARC
9439 	 * to 256 MB, which can be used entirely for metadata.
9440 	 */
9441 	zfs_arc_min = 2ULL << SPA_MAXBLOCKSHIFT;
9442 	zfs_arc_max = 256 * 1024 * 1024;
9443 #endif
9444 
9445 	/*
9446 	 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
9447 	 * "zdb -b" uses traversal prefetch which uses async reads.
9448 	 * For good performance, let several of them be active at once.
9449 	 */
9450 	zfs_vdev_async_read_max_active = 10;
9451 
9452 	/*
9453 	 * Disable reference tracking for better performance.
9454 	 */
9455 	reference_tracking_enable = B_FALSE;
9456 
9457 	/*
9458 	 * Do not fail spa_load when spa_load_verify fails. This is needed
9459 	 * to load non-idle pools.
9460 	 */
9461 	spa_load_verify_dryrun = B_TRUE;
9462 
9463 	/*
9464 	 * ZDB should have ability to read spacemaps.
9465 	 */
9466 	spa_mode_readable_spacemaps = B_TRUE;
9467 
9468 	if (dump_all)
9469 		verbose = MAX(verbose, 1);
9470 
9471 	for (c = 0; c < 256; c++) {
9472 		if (dump_all && strchr("ABeEFkKlLNOPrRSXy", c) == NULL)
9473 			dump_opt[c] = 1;
9474 		if (dump_opt[c])
9475 			dump_opt[c] += verbose;
9476 	}
9477 
9478 	libspl_set_assert_ok((dump_opt['A'] == 1) || (dump_opt['A'] > 2));
9479 	zfs_recover = (dump_opt['A'] > 1);
9480 
9481 	argc -= optind;
9482 	argv += optind;
9483 	if (argc < 2 && dump_opt['R'])
9484 		usage();
9485 
9486 	target = argv[0];
9487 
9488 	/*
9489 	 * Automate cachefile
9490 	 */
9491 	if (!spa_config_path_env && !config_path_console && target &&
9492 	    libzfs_core_init() == 0) {
9493 		char *pname = strdup(target);
9494 		const char *value;
9495 		nvlist_t *pnvl = NULL;
9496 		nvlist_t *vnvl = NULL;
9497 
9498 		if (strpbrk(pname, "/@") != NULL)
9499 			*strpbrk(pname, "/@") = '\0';
9500 
9501 		if (pname && lzc_get_props(pname, &pnvl) == 0) {
9502 			if (nvlist_lookup_nvlist(pnvl, "cachefile",
9503 			    &vnvl) == 0) {
9504 				value = fnvlist_lookup_string(vnvl,
9505 				    ZPROP_VALUE);
9506 			} else {
9507 				value = "-";
9508 			}
9509 			strlcpy(pbuf, value, sizeof (pbuf));
9510 			if (pbuf[0] != '\0') {
9511 				if (pbuf[0] == '/') {
9512 					if (access(pbuf, F_OK) == 0)
9513 						spa_config_path = pbuf;
9514 					else
9515 						force_import = B_TRUE;
9516 				} else if ((strcmp(pbuf, "-") == 0 &&
9517 				    access(ZPOOL_CACHE, F_OK) != 0) ||
9518 				    strcmp(pbuf, "none") == 0) {
9519 					force_import = B_TRUE;
9520 				}
9521 			}
9522 			nvlist_free(vnvl);
9523 		}
9524 
9525 		free(pname);
9526 		nvlist_free(pnvl);
9527 		libzfs_core_fini();
9528 	}
9529 
9530 	dmu_objset_register_type(DMU_OST_ZFS, dummy_get_file_info);
9531 	kernel_init(SPA_MODE_READ);
9532 	kernel_init_done = B_TRUE;
9533 
9534 	if (dump_opt['E']) {
9535 		if (argc != 1)
9536 			usage();
9537 		zdb_embedded_block(argv[0]);
9538 		error = 0;
9539 		goto fini;
9540 	}
9541 
9542 	if (argc < 1) {
9543 		if (!dump_opt['e'] && dump_opt['C']) {
9544 			dump_cachefile(spa_config_path);
9545 			error = 0;
9546 			goto fini;
9547 		}
9548 		usage();
9549 	}
9550 
9551 	if (dump_opt['l']) {
9552 		error = dump_label(argv[0]);
9553 		goto fini;
9554 	}
9555 
9556 	if (dump_opt['X'] || dump_opt['F'])
9557 		rewind = ZPOOL_DO_REWIND |
9558 		    (dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
9559 
9560 	/* -N implies -d */
9561 	if (dump_opt['N'] && dump_opt['d'] == 0)
9562 		dump_opt['d'] = dump_opt['N'];
9563 
9564 	if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 ||
9565 	    nvlist_add_uint64(policy, ZPOOL_LOAD_REQUEST_TXG, max_txg) != 0 ||
9566 	    nvlist_add_uint32(policy, ZPOOL_LOAD_REWIND_POLICY, rewind) != 0)
9567 		fatal("internal error: %s", strerror(ENOMEM));
9568 
9569 	error = 0;
9570 
9571 	if (strpbrk(target, "/@") != NULL) {
9572 		size_t targetlen;
9573 
9574 		target_pool = strdup(target);
9575 		*strpbrk(target_pool, "/@") = '\0';
9576 
9577 		target_is_spa = B_FALSE;
9578 		targetlen = strlen(target);
9579 		if (targetlen && target[targetlen - 1] == '/')
9580 			target[targetlen - 1] = '\0';
9581 
9582 		/*
9583 		 * See if an objset ID was supplied (-d <pool>/<objset ID>).
9584 		 * To disambiguate tank/100, consider the 100 as objsetID
9585 		 * if -N was given, otherwise 100 is an objsetID iff
9586 		 * tank/100 as a named dataset fails on lookup.
9587 		 */
9588 		objset_str = strchr(target, '/');
9589 		if (objset_str && strlen(objset_str) > 1 &&
9590 		    zdb_numeric(objset_str + 1)) {
9591 			char *endptr;
9592 			errno = 0;
9593 			objset_str++;
9594 			objset_id = strtoull(objset_str, &endptr, 0);
9595 			/* dataset 0 is the same as opening the pool */
9596 			if (errno == 0 && endptr != objset_str &&
9597 			    objset_id != 0) {
9598 				if (dump_opt['N'])
9599 					dataset_lookup = B_TRUE;
9600 			}
9601 			/* normal dataset name not an objset ID */
9602 			if (endptr == objset_str) {
9603 				objset_id = -1;
9604 			}
9605 		} else if (objset_str && !zdb_numeric(objset_str + 1) &&
9606 		    dump_opt['N']) {
9607 			printf("Supply a numeric objset ID with -N\n");
9608 			error = 1;
9609 			goto fini;
9610 		}
9611 	} else {
9612 		target_pool = target;
9613 	}
9614 
9615 	if (dump_opt['e'] || force_import) {
9616 		importargs_t args = { 0 };
9617 
9618 		/*
9619 		 * If path is not provided, search in /dev
9620 		 */
9621 		if (searchdirs == NULL) {
9622 			searchdirs = umem_alloc(sizeof (char *), UMEM_NOFAIL);
9623 			searchdirs[nsearch++] = (char *)ZFS_DEVDIR;
9624 		}
9625 
9626 		args.paths = nsearch;
9627 		args.path = searchdirs;
9628 		args.can_be_active = B_TRUE;
9629 
9630 		libpc_handle_t lpch = {
9631 			.lpc_lib_handle = NULL,
9632 			.lpc_ops = &libzpool_config_ops,
9633 			.lpc_printerr = B_TRUE
9634 		};
9635 		error = zpool_find_config(&lpch, target_pool, &cfg, &args);
9636 
9637 		if (error == 0) {
9638 
9639 			if (nvlist_add_nvlist(cfg,
9640 			    ZPOOL_LOAD_POLICY, policy) != 0) {
9641 				fatal("can't open '%s': %s",
9642 				    target, strerror(ENOMEM));
9643 			}
9644 
9645 			if (dump_opt['C'] > 1) {
9646 				(void) printf("\nConfiguration for import:\n");
9647 				dump_nvlist(cfg, 8);
9648 			}
9649 
9650 			/*
9651 			 * Disable the activity check to allow examination of
9652 			 * active pools.
9653 			 */
9654 			error = spa_import(target_pool, cfg, NULL,
9655 			    flags | ZFS_IMPORT_SKIP_MMP);
9656 		}
9657 	}
9658 
9659 	if (searchdirs != NULL) {
9660 		umem_free(searchdirs, nsearch * sizeof (char *));
9661 		searchdirs = NULL;
9662 	}
9663 
9664 	/*
9665 	 * We need to make sure to process -O option or call
9666 	 * dump_path after the -e option has been processed,
9667 	 * which imports the pool to the namespace if it's
9668 	 * not in the cachefile.
9669 	 */
9670 	if (dump_opt['O']) {
9671 		if (argc != 2)
9672 			usage();
9673 		dump_opt['v'] = verbose + 3;
9674 		error = dump_path(argv[0], argv[1], NULL);
9675 		goto fini;
9676 	}
9677 
9678 	if (dump_opt['r']) {
9679 		target_is_spa = B_FALSE;
9680 		if (argc != 3)
9681 			usage();
9682 		dump_opt['v'] = verbose;
9683 		error = dump_path(argv[0], argv[1], &object);
9684 		if (error != 0)
9685 			fatal("internal error: %s", strerror(error));
9686 	}
9687 
9688 	/*
9689 	 * import_checkpointed_state makes the assumption that the
9690 	 * target pool that we pass it is already part of the spa
9691 	 * namespace. Because of that we need to make sure to call
9692 	 * it always after the -e option has been processed, which
9693 	 * imports the pool to the namespace if it's not in the
9694 	 * cachefile.
9695 	 */
9696 	char *checkpoint_pool = NULL;
9697 	char *checkpoint_target = NULL;
9698 	if (dump_opt['k']) {
9699 		checkpoint_pool = import_checkpointed_state(target, cfg,
9700 		    &checkpoint_target);
9701 
9702 		if (checkpoint_target != NULL)
9703 			target = checkpoint_target;
9704 	}
9705 
9706 	if (cfg != NULL) {
9707 		nvlist_free(cfg);
9708 		cfg = NULL;
9709 	}
9710 
9711 	if (target_pool != target)
9712 		free(target_pool);
9713 
9714 	if (error == 0) {
9715 		if (dump_opt['k'] && (target_is_spa || dump_opt['R'])) {
9716 			ASSERT(checkpoint_pool != NULL);
9717 			ASSERT(checkpoint_target == NULL);
9718 
9719 			error = spa_open(checkpoint_pool, &spa, FTAG);
9720 			if (error != 0) {
9721 				fatal("Tried to open pool \"%s\" but "
9722 				    "spa_open() failed with error %d\n",
9723 				    checkpoint_pool, error);
9724 			}
9725 
9726 		} else if (target_is_spa || dump_opt['R'] || dump_opt['B'] ||
9727 		    objset_id == 0) {
9728 			zdb_set_skip_mmp(target);
9729 			error = spa_open_rewind(target, &spa, FTAG, policy,
9730 			    NULL);
9731 			if (error) {
9732 				/*
9733 				 * If we're missing the log device then
9734 				 * try opening the pool after clearing the
9735 				 * log state.
9736 				 */
9737 				mutex_enter(&spa_namespace_lock);
9738 				if ((spa = spa_lookup(target)) != NULL &&
9739 				    spa->spa_log_state == SPA_LOG_MISSING) {
9740 					spa->spa_log_state = SPA_LOG_CLEAR;
9741 					error = 0;
9742 				}
9743 				mutex_exit(&spa_namespace_lock);
9744 
9745 				if (!error) {
9746 					error = spa_open_rewind(target, &spa,
9747 					    FTAG, policy, NULL);
9748 				}
9749 			}
9750 		} else if (strpbrk(target, "#") != NULL) {
9751 			dsl_pool_t *dp;
9752 			error = dsl_pool_hold(target, FTAG, &dp);
9753 			if (error != 0) {
9754 				fatal("can't dump '%s': %s", target,
9755 				    strerror(error));
9756 			}
9757 			error = dump_bookmark(dp, target, B_TRUE, verbose > 1);
9758 			dsl_pool_rele(dp, FTAG);
9759 			if (error != 0) {
9760 				fatal("can't dump '%s': %s", target,
9761 				    strerror(error));
9762 			}
9763 			goto fini;
9764 		} else {
9765 			target_pool = strdup(target);
9766 			if (strpbrk(target, "/@") != NULL)
9767 				*strpbrk(target_pool, "/@") = '\0';
9768 
9769 			zdb_set_skip_mmp(target);
9770 			/*
9771 			 * If -N was supplied, the user has indicated that
9772 			 * zdb -d <pool>/<objsetID> is in effect.  Otherwise
9773 			 * we first assume that the dataset string is the
9774 			 * dataset name.  If dmu_objset_hold fails with the
9775 			 * dataset string, and we have an objset_id, retry the
9776 			 * lookup with the objsetID.
9777 			 */
9778 			boolean_t retry = B_TRUE;
9779 retry_lookup:
9780 			if (dataset_lookup == B_TRUE) {
9781 				/*
9782 				 * Use the supplied id to get the name
9783 				 * for open_objset.
9784 				 */
9785 				error = spa_open(target_pool, &spa, FTAG);
9786 				if (error == 0) {
9787 					error = name_from_objset_id(spa,
9788 					    objset_id, dsname);
9789 					spa_close(spa, FTAG);
9790 					if (error == 0)
9791 						target = dsname;
9792 				}
9793 			}
9794 			if (error == 0) {
9795 				if (objset_id > 0 && retry) {
9796 					int err = dmu_objset_hold(target, FTAG,
9797 					    &os);
9798 					if (err) {
9799 						dataset_lookup = B_TRUE;
9800 						retry = B_FALSE;
9801 						goto retry_lookup;
9802 					} else {
9803 						dmu_objset_rele(os, FTAG);
9804 					}
9805 				}
9806 				error = open_objset(target, FTAG, &os);
9807 			}
9808 			if (error == 0)
9809 				spa = dmu_objset_spa(os);
9810 			free(target_pool);
9811 		}
9812 	}
9813 	nvlist_free(policy);
9814 
9815 	if (error)
9816 		fatal("can't open '%s': %s", target, strerror(error));
9817 
9818 	/*
9819 	 * Set the pool failure mode to panic in order to prevent the pool
9820 	 * from suspending.  A suspended I/O will have no way to resume and
9821 	 * can prevent the zdb(8) command from terminating as expected.
9822 	 */
9823 	if (spa != NULL)
9824 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
9825 
9826 	argv++;
9827 	argc--;
9828 	if (dump_opt['r']) {
9829 		error = zdb_copy_object(os, object, argv[1]);
9830 	} else if (!dump_opt['R']) {
9831 		flagbits['d'] = ZOR_FLAG_DIRECTORY;
9832 		flagbits['f'] = ZOR_FLAG_PLAIN_FILE;
9833 		flagbits['m'] = ZOR_FLAG_SPACE_MAP;
9834 		flagbits['z'] = ZOR_FLAG_ZAP;
9835 		flagbits['A'] = ZOR_FLAG_ALL_TYPES;
9836 
9837 		if (argc > 0 && dump_opt['d']) {
9838 			zopt_object_args = argc;
9839 			zopt_object_ranges = calloc(zopt_object_args,
9840 			    sizeof (zopt_object_range_t));
9841 			for (unsigned i = 0; i < zopt_object_args; i++) {
9842 				int err;
9843 				const char *msg = NULL;
9844 
9845 				err = parse_object_range(argv[i],
9846 				    &zopt_object_ranges[i], &msg);
9847 				if (err != 0)
9848 					fatal("Bad object or range: '%s': %s\n",
9849 					    argv[i], msg ?: "");
9850 			}
9851 		} else if (argc > 0 && dump_opt['m']) {
9852 			zopt_metaslab_args = argc;
9853 			zopt_metaslab = calloc(zopt_metaslab_args,
9854 			    sizeof (uint64_t));
9855 			for (unsigned i = 0; i < zopt_metaslab_args; i++) {
9856 				errno = 0;
9857 				zopt_metaslab[i] = strtoull(argv[i], NULL, 0);
9858 				if (zopt_metaslab[i] == 0 && errno != 0)
9859 					fatal("bad number %s: %s", argv[i],
9860 					    strerror(errno));
9861 			}
9862 		}
9863 		if (dump_opt['B']) {
9864 			dump_backup(target, objset_id,
9865 			    argc > 0 ? argv[0] : NULL);
9866 		} else if (os != NULL) {
9867 			dump_objset(os);
9868 		} else if (zopt_object_args > 0 && !dump_opt['m']) {
9869 			dump_objset(spa->spa_meta_objset);
9870 		} else {
9871 			dump_zpool(spa);
9872 		}
9873 	} else {
9874 		flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR;
9875 		flagbits['c'] = ZDB_FLAG_CHECKSUM;
9876 		flagbits['d'] = ZDB_FLAG_DECOMPRESS;
9877 		flagbits['e'] = ZDB_FLAG_BSWAP;
9878 		flagbits['g'] = ZDB_FLAG_GBH;
9879 		flagbits['i'] = ZDB_FLAG_INDIRECT;
9880 		flagbits['r'] = ZDB_FLAG_RAW;
9881 		flagbits['v'] = ZDB_FLAG_VERBOSE;
9882 
9883 		for (int i = 0; i < argc; i++)
9884 			zdb_read_block(argv[i], spa);
9885 	}
9886 
9887 	if (dump_opt['k']) {
9888 		free(checkpoint_pool);
9889 		if (!target_is_spa)
9890 			free(checkpoint_target);
9891 	}
9892 
9893 fini:
9894 	if (spa != NULL)
9895 		zdb_ddt_cleanup(spa);
9896 
9897 	if (os != NULL) {
9898 		close_objset(os, FTAG);
9899 	} else if (spa != NULL) {
9900 		spa_close(spa, FTAG);
9901 	}
9902 
9903 	fuid_table_destroy();
9904 
9905 	dump_debug_buffer();
9906 
9907 	if (kernel_init_done)
9908 		kernel_fini();
9909 
9910 	return (error);
9911 }
9912