xref: /freebsd/sys/contrib/openzfs/cmd/zdb/zdb.c (revision 17aab35a77a1b1bf02fc85bb8ffadccb0ca5006d)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or https://opensource.org/licenses/CDDL-1.0.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
25  * Copyright (c) 2014 Integros [integros.com]
26  * Copyright 2016 Nexenta Systems, Inc.
27  * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC.
28  * Copyright (c) 2015, 2017, Intel Corporation.
29  * Copyright (c) 2020 Datto Inc.
30  * Copyright (c) 2020, The FreeBSD Foundation [1]
31  *
32  * [1] Portions of this software were developed by Allan Jude
33  *     under sponsorship from the FreeBSD Foundation.
34  * Copyright (c) 2021 Allan Jude
35  * Copyright (c) 2021 Toomas Soome <tsoome@me.com>
36  * Copyright (c) 2023, 2024, Klara Inc.
37  * Copyright (c) 2023, Rob Norris <robn@despairlabs.com>
38  */
39 
40 #include <stdio.h>
41 #include <unistd.h>
42 #include <stdlib.h>
43 #include <ctype.h>
44 #include <getopt.h>
45 #include <openssl/evp.h>
46 #include <sys/zfs_context.h>
47 #include <sys/spa.h>
48 #include <sys/spa_impl.h>
49 #include <sys/dmu.h>
50 #include <sys/zap.h>
51 #include <sys/zap_impl.h>
52 #include <sys/fs/zfs.h>
53 #include <sys/zfs_znode.h>
54 #include <sys/zfs_sa.h>
55 #include <sys/sa.h>
56 #include <sys/sa_impl.h>
57 #include <sys/vdev.h>
58 #include <sys/vdev_impl.h>
59 #include <sys/metaslab_impl.h>
60 #include <sys/dmu_objset.h>
61 #include <sys/dsl_dir.h>
62 #include <sys/dsl_dataset.h>
63 #include <sys/dsl_pool.h>
64 #include <sys/dsl_bookmark.h>
65 #include <sys/dbuf.h>
66 #include <sys/zil.h>
67 #include <sys/zil_impl.h>
68 #include <sys/stat.h>
69 #include <sys/resource.h>
70 #include <sys/dmu_send.h>
71 #include <sys/dmu_traverse.h>
72 #include <sys/zio_checksum.h>
73 #include <sys/zio_compress.h>
74 #include <sys/zfs_fuid.h>
75 #include <sys/arc.h>
76 #include <sys/arc_impl.h>
77 #include <sys/ddt.h>
78 #include <sys/ddt_impl.h>
79 #include <sys/zfeature.h>
80 #include <sys/abd.h>
81 #include <sys/blkptr.h>
82 #include <sys/dsl_crypt.h>
83 #include <sys/dsl_scan.h>
84 #include <sys/btree.h>
85 #include <sys/brt.h>
86 #include <sys/brt_impl.h>
87 #include <zfs_comutil.h>
88 #include <sys/zstd/zstd.h>
89 #include <sys/backtrace.h>
90 
91 #include <libnvpair.h>
92 #include <libzutil.h>
93 #include <libzfs_core.h>
94 
95 #include <libzdb.h>
96 
97 #include "zdb.h"
98 
99 
100 extern int reference_tracking_enable;
101 extern int zfs_recover;
102 extern uint_t zfs_vdev_async_read_max_active;
103 extern boolean_t spa_load_verify_dryrun;
104 extern boolean_t spa_mode_readable_spacemaps;
105 extern uint_t zfs_reconstruct_indirect_combinations_max;
106 extern uint_t zfs_btree_verify_intensity;
107 
108 static const char cmdname[] = "zdb";
109 uint8_t dump_opt[256];
110 
111 typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size);
112 
113 static uint64_t *zopt_metaslab = NULL;
114 static unsigned zopt_metaslab_args = 0;
115 
116 
117 static zopt_object_range_t *zopt_object_ranges = NULL;
118 static unsigned zopt_object_args = 0;
119 
120 static int flagbits[256];
121 
122 
123 static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */
124 static int leaked_objects = 0;
125 static range_tree_t *mos_refd_objs;
126 static spa_t *spa;
127 static objset_t *os;
128 static boolean_t kernel_init_done;
129 
130 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *,
131     boolean_t);
132 static void mos_obj_refd(uint64_t);
133 static void mos_obj_refd_multiple(uint64_t);
134 static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free,
135     dmu_tx_t *tx);
136 
137 
138 
139 static void zdb_print_blkptr(const blkptr_t *bp, int flags);
140 static void zdb_exit(int reason);
141 
142 typedef struct sublivelist_verify_block_refcnt {
143 	/* block pointer entry in livelist being verified */
144 	blkptr_t svbr_blk;
145 
146 	/*
147 	 * Refcount gets incremented to 1 when we encounter the first
148 	 * FREE entry for the svfbr block pointer and a node for it
149 	 * is created in our ZDB verification/tracking metadata.
150 	 *
151 	 * As we encounter more FREE entries we increment this counter
152 	 * and similarly decrement it whenever we find the respective
153 	 * ALLOC entries for this block.
154 	 *
155 	 * When the refcount gets to 0 it means that all the FREE and
156 	 * ALLOC entries of this block have paired up and we no longer
157 	 * need to track it in our verification logic (e.g. the node
158 	 * containing this struct in our verification data structure
159 	 * should be freed).
160 	 *
161 	 * [refer to sublivelist_verify_blkptr() for the actual code]
162 	 */
163 	uint32_t svbr_refcnt;
164 } sublivelist_verify_block_refcnt_t;
165 
166 static int
sublivelist_block_refcnt_compare(const void * larg,const void * rarg)167 sublivelist_block_refcnt_compare(const void *larg, const void *rarg)
168 {
169 	const sublivelist_verify_block_refcnt_t *l = larg;
170 	const sublivelist_verify_block_refcnt_t *r = rarg;
171 	return (livelist_compare(&l->svbr_blk, &r->svbr_blk));
172 }
173 
174 static int
sublivelist_verify_blkptr(void * arg,const blkptr_t * bp,boolean_t free,dmu_tx_t * tx)175 sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free,
176     dmu_tx_t *tx)
177 {
178 	ASSERT3P(tx, ==, NULL);
179 	struct sublivelist_verify *sv = arg;
180 	sublivelist_verify_block_refcnt_t current = {
181 			.svbr_blk = *bp,
182 
183 			/*
184 			 * Start with 1 in case this is the first free entry.
185 			 * This field is not used for our B-Tree comparisons
186 			 * anyway.
187 			 */
188 			.svbr_refcnt = 1,
189 	};
190 
191 	zfs_btree_index_t where;
192 	sublivelist_verify_block_refcnt_t *pair =
193 	    zfs_btree_find(&sv->sv_pair, &current, &where);
194 	if (free) {
195 		if (pair == NULL) {
196 			/* first free entry for this block pointer */
197 			zfs_btree_add(&sv->sv_pair, &current);
198 		} else {
199 			pair->svbr_refcnt++;
200 		}
201 	} else {
202 		if (pair == NULL) {
203 			/* block that is currently marked as allocated */
204 			for (int i = 0; i < SPA_DVAS_PER_BP; i++) {
205 				if (DVA_IS_EMPTY(&bp->blk_dva[i]))
206 					break;
207 				sublivelist_verify_block_t svb = {
208 				    .svb_dva = bp->blk_dva[i],
209 				    .svb_allocated_txg =
210 				    BP_GET_LOGICAL_BIRTH(bp)
211 				};
212 
213 				if (zfs_btree_find(&sv->sv_leftover, &svb,
214 				    &where) == NULL) {
215 					zfs_btree_add_idx(&sv->sv_leftover,
216 					    &svb, &where);
217 				}
218 			}
219 		} else {
220 			/* alloc matches a free entry */
221 			pair->svbr_refcnt--;
222 			if (pair->svbr_refcnt == 0) {
223 				/* all allocs and frees have been matched */
224 				zfs_btree_remove_idx(&sv->sv_pair, &where);
225 			}
226 		}
227 	}
228 
229 	return (0);
230 }
231 
232 static int
sublivelist_verify_func(void * args,dsl_deadlist_entry_t * dle)233 sublivelist_verify_func(void *args, dsl_deadlist_entry_t *dle)
234 {
235 	int err;
236 	struct sublivelist_verify *sv = args;
237 
238 	zfs_btree_create(&sv->sv_pair, sublivelist_block_refcnt_compare, NULL,
239 	    sizeof (sublivelist_verify_block_refcnt_t));
240 
241 	err = bpobj_iterate_nofree(&dle->dle_bpobj, sublivelist_verify_blkptr,
242 	    sv, NULL);
243 
244 	sublivelist_verify_block_refcnt_t *e;
245 	zfs_btree_index_t *cookie = NULL;
246 	while ((e = zfs_btree_destroy_nodes(&sv->sv_pair, &cookie)) != NULL) {
247 		char blkbuf[BP_SPRINTF_LEN];
248 		snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
249 		    &e->svbr_blk, B_TRUE);
250 		(void) printf("\tERROR: %d unmatched FREE(s): %s\n",
251 		    e->svbr_refcnt, blkbuf);
252 	}
253 	zfs_btree_destroy(&sv->sv_pair);
254 
255 	return (err);
256 }
257 
258 static int
livelist_block_compare(const void * larg,const void * rarg)259 livelist_block_compare(const void *larg, const void *rarg)
260 {
261 	const sublivelist_verify_block_t *l = larg;
262 	const sublivelist_verify_block_t *r = rarg;
263 
264 	if (DVA_GET_VDEV(&l->svb_dva) < DVA_GET_VDEV(&r->svb_dva))
265 		return (-1);
266 	else if (DVA_GET_VDEV(&l->svb_dva) > DVA_GET_VDEV(&r->svb_dva))
267 		return (+1);
268 
269 	if (DVA_GET_OFFSET(&l->svb_dva) < DVA_GET_OFFSET(&r->svb_dva))
270 		return (-1);
271 	else if (DVA_GET_OFFSET(&l->svb_dva) > DVA_GET_OFFSET(&r->svb_dva))
272 		return (+1);
273 
274 	if (DVA_GET_ASIZE(&l->svb_dva) < DVA_GET_ASIZE(&r->svb_dva))
275 		return (-1);
276 	else if (DVA_GET_ASIZE(&l->svb_dva) > DVA_GET_ASIZE(&r->svb_dva))
277 		return (+1);
278 
279 	return (0);
280 }
281 
282 /*
283  * Check for errors in a livelist while tracking all unfreed ALLOCs in the
284  * sublivelist_verify_t: sv->sv_leftover
285  */
286 static void
livelist_verify(dsl_deadlist_t * dl,void * arg)287 livelist_verify(dsl_deadlist_t *dl, void *arg)
288 {
289 	sublivelist_verify_t *sv = arg;
290 	dsl_deadlist_iterate(dl, sublivelist_verify_func, sv);
291 }
292 
293 /*
294  * Check for errors in the livelist entry and discard the intermediary
295  * data structures
296  */
297 static int
sublivelist_verify_lightweight(void * args,dsl_deadlist_entry_t * dle)298 sublivelist_verify_lightweight(void *args, dsl_deadlist_entry_t *dle)
299 {
300 	(void) args;
301 	sublivelist_verify_t sv;
302 	zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
303 	    sizeof (sublivelist_verify_block_t));
304 	int err = sublivelist_verify_func(&sv, dle);
305 	zfs_btree_clear(&sv.sv_leftover);
306 	zfs_btree_destroy(&sv.sv_leftover);
307 	return (err);
308 }
309 
310 typedef struct metaslab_verify {
311 	/*
312 	 * Tree containing all the leftover ALLOCs from the livelists
313 	 * that are part of this metaslab.
314 	 */
315 	zfs_btree_t mv_livelist_allocs;
316 
317 	/*
318 	 * Metaslab information.
319 	 */
320 	uint64_t mv_vdid;
321 	uint64_t mv_msid;
322 	uint64_t mv_start;
323 	uint64_t mv_end;
324 
325 	/*
326 	 * What's currently allocated for this metaslab.
327 	 */
328 	range_tree_t *mv_allocated;
329 } metaslab_verify_t;
330 
331 typedef void ll_iter_t(dsl_deadlist_t *ll, void *arg);
332 
333 typedef int (*zdb_log_sm_cb_t)(spa_t *spa, space_map_entry_t *sme, uint64_t txg,
334     void *arg);
335 
336 typedef struct unflushed_iter_cb_arg {
337 	spa_t *uic_spa;
338 	uint64_t uic_txg;
339 	void *uic_arg;
340 	zdb_log_sm_cb_t uic_cb;
341 } unflushed_iter_cb_arg_t;
342 
343 static int
iterate_through_spacemap_logs_cb(space_map_entry_t * sme,void * arg)344 iterate_through_spacemap_logs_cb(space_map_entry_t *sme, void *arg)
345 {
346 	unflushed_iter_cb_arg_t *uic = arg;
347 	return (uic->uic_cb(uic->uic_spa, sme, uic->uic_txg, uic->uic_arg));
348 }
349 
350 static void
iterate_through_spacemap_logs(spa_t * spa,zdb_log_sm_cb_t cb,void * arg)351 iterate_through_spacemap_logs(spa_t *spa, zdb_log_sm_cb_t cb, void *arg)
352 {
353 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
354 		return;
355 
356 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
357 	for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
358 	    sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
359 		space_map_t *sm = NULL;
360 		VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
361 		    sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
362 
363 		unflushed_iter_cb_arg_t uic = {
364 			.uic_spa = spa,
365 			.uic_txg = sls->sls_txg,
366 			.uic_arg = arg,
367 			.uic_cb = cb
368 		};
369 		VERIFY0(space_map_iterate(sm, space_map_length(sm),
370 		    iterate_through_spacemap_logs_cb, &uic));
371 		space_map_close(sm);
372 	}
373 	spa_config_exit(spa, SCL_CONFIG, FTAG);
374 }
375 
376 static void
verify_livelist_allocs(metaslab_verify_t * mv,uint64_t txg,uint64_t offset,uint64_t size)377 verify_livelist_allocs(metaslab_verify_t *mv, uint64_t txg,
378     uint64_t offset, uint64_t size)
379 {
380 	sublivelist_verify_block_t svb = {{{0}}};
381 	DVA_SET_VDEV(&svb.svb_dva, mv->mv_vdid);
382 	DVA_SET_OFFSET(&svb.svb_dva, offset);
383 	DVA_SET_ASIZE(&svb.svb_dva, size);
384 	zfs_btree_index_t where;
385 	uint64_t end_offset = offset + size;
386 
387 	/*
388 	 *  Look for an exact match for spacemap entry in the livelist entries.
389 	 *  Then, look for other livelist entries that fall within the range
390 	 *  of the spacemap entry as it may have been condensed
391 	 */
392 	sublivelist_verify_block_t *found =
393 	    zfs_btree_find(&mv->mv_livelist_allocs, &svb, &where);
394 	if (found == NULL) {
395 		found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where);
396 	}
397 	for (; found != NULL && DVA_GET_VDEV(&found->svb_dva) == mv->mv_vdid &&
398 	    DVA_GET_OFFSET(&found->svb_dva) < end_offset;
399 	    found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
400 		if (found->svb_allocated_txg <= txg) {
401 			(void) printf("ERROR: Livelist ALLOC [%llx:%llx] "
402 			    "from TXG %llx FREED at TXG %llx\n",
403 			    (u_longlong_t)DVA_GET_OFFSET(&found->svb_dva),
404 			    (u_longlong_t)DVA_GET_ASIZE(&found->svb_dva),
405 			    (u_longlong_t)found->svb_allocated_txg,
406 			    (u_longlong_t)txg);
407 		}
408 	}
409 }
410 
411 static int
metaslab_spacemap_validation_cb(space_map_entry_t * sme,void * arg)412 metaslab_spacemap_validation_cb(space_map_entry_t *sme, void *arg)
413 {
414 	metaslab_verify_t *mv = arg;
415 	uint64_t offset = sme->sme_offset;
416 	uint64_t size = sme->sme_run;
417 	uint64_t txg = sme->sme_txg;
418 
419 	if (sme->sme_type == SM_ALLOC) {
420 		if (range_tree_contains(mv->mv_allocated,
421 		    offset, size)) {
422 			(void) printf("ERROR: DOUBLE ALLOC: "
423 			    "%llu [%llx:%llx] "
424 			    "%llu:%llu LOG_SM\n",
425 			    (u_longlong_t)txg, (u_longlong_t)offset,
426 			    (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
427 			    (u_longlong_t)mv->mv_msid);
428 		} else {
429 			range_tree_add(mv->mv_allocated,
430 			    offset, size);
431 		}
432 	} else {
433 		if (!range_tree_contains(mv->mv_allocated,
434 		    offset, size)) {
435 			(void) printf("ERROR: DOUBLE FREE: "
436 			    "%llu [%llx:%llx] "
437 			    "%llu:%llu LOG_SM\n",
438 			    (u_longlong_t)txg, (u_longlong_t)offset,
439 			    (u_longlong_t)size, (u_longlong_t)mv->mv_vdid,
440 			    (u_longlong_t)mv->mv_msid);
441 		} else {
442 			range_tree_remove(mv->mv_allocated,
443 			    offset, size);
444 		}
445 	}
446 
447 	if (sme->sme_type != SM_ALLOC) {
448 		/*
449 		 * If something is freed in the spacemap, verify that
450 		 * it is not listed as allocated in the livelist.
451 		 */
452 		verify_livelist_allocs(mv, txg, offset, size);
453 	}
454 	return (0);
455 }
456 
457 static int
spacemap_check_sm_log_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)458 spacemap_check_sm_log_cb(spa_t *spa, space_map_entry_t *sme,
459     uint64_t txg, void *arg)
460 {
461 	metaslab_verify_t *mv = arg;
462 	uint64_t offset = sme->sme_offset;
463 	uint64_t vdev_id = sme->sme_vdev;
464 
465 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
466 
467 	/* skip indirect vdevs */
468 	if (!vdev_is_concrete(vd))
469 		return (0);
470 
471 	if (vdev_id != mv->mv_vdid)
472 		return (0);
473 
474 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
475 	if (ms->ms_id != mv->mv_msid)
476 		return (0);
477 
478 	if (txg < metaslab_unflushed_txg(ms))
479 		return (0);
480 
481 
482 	ASSERT3U(txg, ==, sme->sme_txg);
483 	return (metaslab_spacemap_validation_cb(sme, mv));
484 }
485 
486 static void
spacemap_check_sm_log(spa_t * spa,metaslab_verify_t * mv)487 spacemap_check_sm_log(spa_t *spa, metaslab_verify_t *mv)
488 {
489 	iterate_through_spacemap_logs(spa, spacemap_check_sm_log_cb, mv);
490 }
491 
492 static void
spacemap_check_ms_sm(space_map_t * sm,metaslab_verify_t * mv)493 spacemap_check_ms_sm(space_map_t  *sm, metaslab_verify_t *mv)
494 {
495 	if (sm == NULL)
496 		return;
497 
498 	VERIFY0(space_map_iterate(sm, space_map_length(sm),
499 	    metaslab_spacemap_validation_cb, mv));
500 }
501 
502 static void iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg);
503 
504 /*
505  * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if
506  * they are part of that metaslab (mv_msid).
507  */
508 static void
mv_populate_livelist_allocs(metaslab_verify_t * mv,sublivelist_verify_t * sv)509 mv_populate_livelist_allocs(metaslab_verify_t *mv, sublivelist_verify_t *sv)
510 {
511 	zfs_btree_index_t where;
512 	sublivelist_verify_block_t *svb;
513 	ASSERT3U(zfs_btree_numnodes(&mv->mv_livelist_allocs), ==, 0);
514 	for (svb = zfs_btree_first(&sv->sv_leftover, &where);
515 	    svb != NULL;
516 	    svb = zfs_btree_next(&sv->sv_leftover, &where, &where)) {
517 		if (DVA_GET_VDEV(&svb->svb_dva) != mv->mv_vdid)
518 			continue;
519 
520 		if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start &&
521 		    (DVA_GET_OFFSET(&svb->svb_dva) +
522 		    DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_start) {
523 			(void) printf("ERROR: Found block that crosses "
524 			    "metaslab boundary: <%llu:%llx:%llx>\n",
525 			    (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
526 			    (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
527 			    (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
528 			continue;
529 		}
530 
531 		if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start)
532 			continue;
533 
534 		if (DVA_GET_OFFSET(&svb->svb_dva) >= mv->mv_end)
535 			continue;
536 
537 		if ((DVA_GET_OFFSET(&svb->svb_dva) +
538 		    DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_end) {
539 			(void) printf("ERROR: Found block that crosses "
540 			    "metaslab boundary: <%llu:%llx:%llx>\n",
541 			    (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva),
542 			    (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
543 			    (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva));
544 			continue;
545 		}
546 
547 		zfs_btree_add(&mv->mv_livelist_allocs, svb);
548 	}
549 
550 	for (svb = zfs_btree_first(&mv->mv_livelist_allocs, &where);
551 	    svb != NULL;
552 	    svb = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) {
553 		zfs_btree_remove(&sv->sv_leftover, svb);
554 	}
555 }
556 
557 /*
558  * [Livelist Check]
559  * Iterate through all the sublivelists and:
560  * - report leftover frees (**)
561  * - record leftover ALLOCs together with their TXG [see Cross Check]
562  *
563  * (**) Note: Double ALLOCs are valid in datasets that have dedup
564  *      enabled. Similarly double FREEs are allowed as well but
565  *      only if they pair up with a corresponding ALLOC entry once
566  *      we our done with our sublivelist iteration.
567  *
568  * [Spacemap Check]
569  * for each metaslab:
570  * - iterate over spacemap and then the metaslab's entries in the
571  *   spacemap log, then report any double FREEs and ALLOCs (do not
572  *   blow up).
573  *
574  * [Cross Check]
575  * After finishing the Livelist Check phase and while being in the
576  * Spacemap Check phase, we find all the recorded leftover ALLOCs
577  * of the livelist check that are part of the metaslab that we are
578  * currently looking at in the Spacemap Check. We report any entries
579  * that are marked as ALLOCs in the livelists but have been actually
580  * freed (and potentially allocated again) after their TXG stamp in
581  * the spacemaps. Also report any ALLOCs from the livelists that
582  * belong to indirect vdevs (e.g. their vdev completed removal).
583  *
584  * Note that this will miss Log Spacemap entries that cancelled each other
585  * out before being flushed to the metaslab, so we are not guaranteed
586  * to match all erroneous ALLOCs.
587  */
588 static void
livelist_metaslab_validate(spa_t * spa)589 livelist_metaslab_validate(spa_t *spa)
590 {
591 	(void) printf("Verifying deleted livelist entries\n");
592 
593 	sublivelist_verify_t sv;
594 	zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL,
595 	    sizeof (sublivelist_verify_block_t));
596 	iterate_deleted_livelists(spa, livelist_verify, &sv);
597 
598 	(void) printf("Verifying metaslab entries\n");
599 	vdev_t *rvd = spa->spa_root_vdev;
600 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
601 		vdev_t *vd = rvd->vdev_child[c];
602 
603 		if (!vdev_is_concrete(vd))
604 			continue;
605 
606 		for (uint64_t mid = 0; mid < vd->vdev_ms_count; mid++) {
607 			metaslab_t *m = vd->vdev_ms[mid];
608 
609 			(void) fprintf(stderr,
610 			    "\rverifying concrete vdev %llu, "
611 			    "metaslab %llu of %llu ...",
612 			    (longlong_t)vd->vdev_id,
613 			    (longlong_t)mid,
614 			    (longlong_t)vd->vdev_ms_count);
615 
616 			uint64_t shift, start;
617 			range_seg_type_t type =
618 			    metaslab_calculate_range_tree_type(vd, m,
619 			    &start, &shift);
620 			metaslab_verify_t mv;
621 			mv.mv_allocated = range_tree_create(NULL,
622 			    type, NULL, start, shift);
623 			mv.mv_vdid = vd->vdev_id;
624 			mv.mv_msid = m->ms_id;
625 			mv.mv_start = m->ms_start;
626 			mv.mv_end = m->ms_start + m->ms_size;
627 			zfs_btree_create(&mv.mv_livelist_allocs,
628 			    livelist_block_compare, NULL,
629 			    sizeof (sublivelist_verify_block_t));
630 
631 			mv_populate_livelist_allocs(&mv, &sv);
632 
633 			spacemap_check_ms_sm(m->ms_sm, &mv);
634 			spacemap_check_sm_log(spa, &mv);
635 
636 			range_tree_vacate(mv.mv_allocated, NULL, NULL);
637 			range_tree_destroy(mv.mv_allocated);
638 			zfs_btree_clear(&mv.mv_livelist_allocs);
639 			zfs_btree_destroy(&mv.mv_livelist_allocs);
640 		}
641 	}
642 	(void) fprintf(stderr, "\n");
643 
644 	/*
645 	 * If there are any segments in the leftover tree after we walked
646 	 * through all the metaslabs in the concrete vdevs then this means
647 	 * that we have segments in the livelists that belong to indirect
648 	 * vdevs and are marked as allocated.
649 	 */
650 	if (zfs_btree_numnodes(&sv.sv_leftover) == 0) {
651 		zfs_btree_destroy(&sv.sv_leftover);
652 		return;
653 	}
654 	(void) printf("ERROR: Found livelist blocks marked as allocated "
655 	    "for indirect vdevs:\n");
656 
657 	zfs_btree_index_t *where = NULL;
658 	sublivelist_verify_block_t *svb;
659 	while ((svb = zfs_btree_destroy_nodes(&sv.sv_leftover, &where)) !=
660 	    NULL) {
661 		int vdev_id = DVA_GET_VDEV(&svb->svb_dva);
662 		ASSERT3U(vdev_id, <, rvd->vdev_children);
663 		vdev_t *vd = rvd->vdev_child[vdev_id];
664 		ASSERT(!vdev_is_concrete(vd));
665 		(void) printf("<%d:%llx:%llx> TXG %llx\n",
666 		    vdev_id, (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva),
667 		    (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva),
668 		    (u_longlong_t)svb->svb_allocated_txg);
669 	}
670 	(void) printf("\n");
671 	zfs_btree_destroy(&sv.sv_leftover);
672 }
673 
674 /*
675  * These libumem hooks provide a reasonable set of defaults for the allocator's
676  * debugging facilities.
677  */
678 const char *
_umem_debug_init(void)679 _umem_debug_init(void)
680 {
681 	return ("default,verbose"); /* $UMEM_DEBUG setting */
682 }
683 
684 const char *
_umem_logging_init(void)685 _umem_logging_init(void)
686 {
687 	return ("fail,contents"); /* $UMEM_LOGGING setting */
688 }
689 
690 static void
usage(void)691 usage(void)
692 {
693 	(void) fprintf(stderr,
694 	    "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] "
695 	    "[-I <inflight I/Os>]\n"
696 	    "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
697 	    "\t\t[-K <key>]\n"
698 	    "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n"
699 	    "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n"
700 	    "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n"
701 	    "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n"
702 	    "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
703 	    "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n"
704 	    "\t%s [-v] <bookmark>\n"
705 	    "\t%s -C [-A] [-U <cache>] [<poolname>]\n"
706 	    "\t%s -l [-Aqu] <device>\n"
707 	    "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
708 	    "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
709 	    "\t%s -O [-K <key>] <dataset> <path>\n"
710 	    "\t%s -r [-K <key>] <dataset> <path> <destination>\n"
711 	    "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
712 	    "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
713 	    "\t%s -E [-A] word0:word1:...:word15\n"
714 	    "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
715 	    "<poolname>\n\n",
716 	    cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
717 	    cmdname, cmdname, cmdname, cmdname, cmdname);
718 
719 	(void) fprintf(stderr, "    Dataset name must include at least one "
720 	    "separator character '/' or '@'\n");
721 	(void) fprintf(stderr, "    If dataset name is specified, only that "
722 	    "dataset is dumped\n");
723 	(void) fprintf(stderr,  "    If object numbers or object number "
724 	    "ranges are specified, only those\n"
725 	    "    objects or ranges are dumped.\n\n");
726 	(void) fprintf(stderr,
727 	    "    Object ranges take the form <start>:<end>[:<flags>]\n"
728 	    "        start    Starting object number\n"
729 	    "        end      Ending object number, or -1 for no upper bound\n"
730 	    "        flags    Optional flags to select object types:\n"
731 	    "            A     All objects (this is the default)\n"
732 	    "            d     ZFS directories\n"
733 	    "            f     ZFS files \n"
734 	    "            m     SPA space maps\n"
735 	    "            z     ZAPs\n"
736 	    "            -     Negate effect of next flag\n\n");
737 	(void) fprintf(stderr, "    Options to control amount of output:\n");
738 	(void) fprintf(stderr, "        -b --block-stats             "
739 	    "block statistics\n");
740 	(void) fprintf(stderr, "        -B --backup                  "
741 	    "backup stream\n");
742 	(void) fprintf(stderr, "        -c --checksum                "
743 	    "checksum all metadata (twice for all data) blocks\n");
744 	(void) fprintf(stderr, "        -C --config                  "
745 	    "config (or cachefile if alone)\n");
746 	(void) fprintf(stderr, "        -d --datasets                "
747 	    "dataset(s)\n");
748 	(void) fprintf(stderr, "        -D --dedup-stats             "
749 	    "dedup statistics\n");
750 	(void) fprintf(stderr, "        -E --embedded-block-pointer=INTEGER\n"
751 	    "                                     decode and display block "
752 	    "from an embedded block pointer\n");
753 	(void) fprintf(stderr, "        -h --history                 "
754 	    "pool history\n");
755 	(void) fprintf(stderr, "        -i --intent-logs             "
756 	    "intent logs\n");
757 	(void) fprintf(stderr, "        -l --label                   "
758 	    "read label contents\n");
759 	(void) fprintf(stderr, "        -k --checkpointed-state      "
760 	    "examine the checkpointed state of the pool\n");
761 	(void) fprintf(stderr, "        -L --disable-leak-tracking   "
762 	    "disable leak tracking (do not load spacemaps)\n");
763 	(void) fprintf(stderr, "        -m --metaslabs               "
764 	    "metaslabs\n");
765 	(void) fprintf(stderr, "        -M --metaslab-groups         "
766 	    "metaslab groups\n");
767 	(void) fprintf(stderr, "        -O --object-lookups          "
768 	    "perform object lookups by path\n");
769 	(void) fprintf(stderr, "        -r --copy-object             "
770 	    "copy an object by path to file\n");
771 	(void) fprintf(stderr, "        -R --read-block              "
772 	    "read and display block from a device\n");
773 	(void) fprintf(stderr, "        -s --io-stats                "
774 	    "report stats on zdb's I/O\n");
775 	(void) fprintf(stderr, "        -S --simulate-dedup          "
776 	    "simulate dedup to measure effect\n");
777 	(void) fprintf(stderr, "        -v --verbose                 "
778 	    "verbose (applies to all others)\n");
779 	(void) fprintf(stderr, "        -y --livelist                "
780 	    "perform livelist and metaslab validation on any livelists being "
781 	    "deleted\n\n");
782 	(void) fprintf(stderr, "    Below options are intended for use "
783 	    "with other options:\n");
784 	(void) fprintf(stderr, "        -A --ignore-assertions       "
785 	    "ignore assertions (-A), enable panic recovery (-AA) or both "
786 	    "(-AAA)\n");
787 	(void) fprintf(stderr, "        -e --exported                "
788 	    "pool is exported/destroyed/has altroot/not in a cachefile\n");
789 	(void) fprintf(stderr, "        -F --automatic-rewind        "
790 	    "attempt automatic rewind within safe range of transaction "
791 	    "groups\n");
792 	(void) fprintf(stderr, "        -G --dump-debug-msg          "
793 	    "dump zfs_dbgmsg buffer before exiting\n");
794 	(void) fprintf(stderr, "        -I --inflight=INTEGER        "
795 	    "specify the maximum number of checksumming I/Os "
796 	    "[default is 200]\n");
797 	(void) fprintf(stderr, "        -K --key=KEY                 "
798 	    "decryption key for encrypted dataset\n");
799 	(void) fprintf(stderr, "        -o --option=\"OPTION=INTEGER\" "
800 	    "set global variable to an unsigned 32-bit integer\n");
801 	(void) fprintf(stderr, "        -p --path==PATH              "
802 	    "use one or more with -e to specify path to vdev dir\n");
803 	(void) fprintf(stderr, "        -P --parseable               "
804 	    "print numbers in parseable form\n");
805 	(void) fprintf(stderr, "        -q --skip-label              "
806 	    "don't print label contents\n");
807 	(void) fprintf(stderr, "        -t --txg=INTEGER             "
808 	    "highest txg to use when searching for uberblocks\n");
809 	(void) fprintf(stderr, "        -T --brt-stats               "
810 	    "BRT statistics\n");
811 	(void) fprintf(stderr, "        -u --uberblock               "
812 	    "uberblock\n");
813 	(void) fprintf(stderr, "        -U --cachefile=PATH          "
814 	    "use alternate cachefile\n");
815 	(void) fprintf(stderr, "        -V --verbatim                "
816 	    "do verbatim import\n");
817 	(void) fprintf(stderr, "        -x --dump-blocks=PATH        "
818 	    "dump all read blocks into specified directory\n");
819 	(void) fprintf(stderr, "        -X --extreme-rewind          "
820 	    "attempt extreme rewind (does not work with dataset)\n");
821 	(void) fprintf(stderr, "        -Y --all-reconstruction      "
822 	    "attempt all reconstruction combinations for split blocks\n");
823 	(void) fprintf(stderr, "        -Z --zstd-headers            "
824 	    "show ZSTD headers \n");
825 	(void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
826 	    "to make only that option verbose\n");
827 	(void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
828 	zdb_exit(1);
829 }
830 
831 static void
dump_debug_buffer(void)832 dump_debug_buffer(void)
833 {
834 	ssize_t ret __attribute__((unused));
835 
836 	if (!dump_opt['G'])
837 		return;
838 	/*
839 	 * We use write() instead of printf() so that this function
840 	 * is safe to call from a signal handler.
841 	 */
842 	ret = write(STDERR_FILENO, "\n", 1);
843 	zfs_dbgmsg_print(STDERR_FILENO, "zdb");
844 }
845 
sig_handler(int signo)846 static void sig_handler(int signo)
847 {
848 	struct sigaction action;
849 
850 	libspl_backtrace(STDERR_FILENO);
851 	dump_debug_buffer();
852 
853 	/*
854 	 * Restore default action and re-raise signal so SIGSEGV and
855 	 * SIGABRT can trigger a core dump.
856 	 */
857 	action.sa_handler = SIG_DFL;
858 	sigemptyset(&action.sa_mask);
859 	action.sa_flags = 0;
860 	(void) sigaction(signo, &action, NULL);
861 	raise(signo);
862 }
863 
864 /*
865  * Called for usage errors that are discovered after a call to spa_open(),
866  * dmu_bonus_hold(), or pool_match().  abort() is called for other errors.
867  */
868 
869 static void
fatal(const char * fmt,...)870 fatal(const char *fmt, ...)
871 {
872 	va_list ap;
873 
874 	va_start(ap, fmt);
875 	(void) fprintf(stderr, "%s: ", cmdname);
876 	(void) vfprintf(stderr, fmt, ap);
877 	va_end(ap);
878 	(void) fprintf(stderr, "\n");
879 
880 	dump_debug_buffer();
881 
882 	zdb_exit(1);
883 }
884 
885 static void
dump_packed_nvlist(objset_t * os,uint64_t object,void * data,size_t size)886 dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size)
887 {
888 	(void) size;
889 	nvlist_t *nv;
890 	size_t nvsize = *(uint64_t *)data;
891 	char *packed = umem_alloc(nvsize, UMEM_NOFAIL);
892 
893 	VERIFY(0 == dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH));
894 
895 	VERIFY(nvlist_unpack(packed, nvsize, &nv, 0) == 0);
896 
897 	umem_free(packed, nvsize);
898 
899 	dump_nvlist(nv, 8);
900 
901 	nvlist_free(nv);
902 }
903 
904 static void
dump_history_offsets(objset_t * os,uint64_t object,void * data,size_t size)905 dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size)
906 {
907 	(void) os, (void) object, (void) size;
908 	spa_history_phys_t *shp = data;
909 
910 	if (shp == NULL)
911 		return;
912 
913 	(void) printf("\t\tpool_create_len = %llu\n",
914 	    (u_longlong_t)shp->sh_pool_create_len);
915 	(void) printf("\t\tphys_max_off = %llu\n",
916 	    (u_longlong_t)shp->sh_phys_max_off);
917 	(void) printf("\t\tbof = %llu\n",
918 	    (u_longlong_t)shp->sh_bof);
919 	(void) printf("\t\teof = %llu\n",
920 	    (u_longlong_t)shp->sh_eof);
921 	(void) printf("\t\trecords_lost = %llu\n",
922 	    (u_longlong_t)shp->sh_records_lost);
923 }
924 
925 static void
zdb_nicenum(uint64_t num,char * buf,size_t buflen)926 zdb_nicenum(uint64_t num, char *buf, size_t buflen)
927 {
928 	if (dump_opt['P'])
929 		(void) snprintf(buf, buflen, "%llu", (longlong_t)num);
930 	else
931 		nicenum(num, buf, buflen);
932 }
933 
934 static void
zdb_nicebytes(uint64_t bytes,char * buf,size_t buflen)935 zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen)
936 {
937 	if (dump_opt['P'])
938 		(void) snprintf(buf, buflen, "%llu", (longlong_t)bytes);
939 	else
940 		zfs_nicebytes(bytes, buf, buflen);
941 }
942 
943 static const char histo_stars[] = "****************************************";
944 static const uint64_t histo_width = sizeof (histo_stars) - 1;
945 
946 static void
dump_histogram(const uint64_t * histo,int size,int offset)947 dump_histogram(const uint64_t *histo, int size, int offset)
948 {
949 	int i;
950 	int minidx = size - 1;
951 	int maxidx = 0;
952 	uint64_t max = 0;
953 
954 	for (i = 0; i < size; i++) {
955 		if (histo[i] == 0)
956 			continue;
957 		if (histo[i] > max)
958 			max = histo[i];
959 		if (i > maxidx)
960 			maxidx = i;
961 		if (i < minidx)
962 			minidx = i;
963 	}
964 
965 	if (max < histo_width)
966 		max = histo_width;
967 
968 	for (i = minidx; i <= maxidx; i++) {
969 		(void) printf("\t\t\t%3u: %6llu %s\n",
970 		    i + offset, (u_longlong_t)histo[i],
971 		    &histo_stars[(max - histo[i]) * histo_width / max]);
972 	}
973 }
974 
975 static void
dump_zap_stats(objset_t * os,uint64_t object)976 dump_zap_stats(objset_t *os, uint64_t object)
977 {
978 	int error;
979 	zap_stats_t zs;
980 
981 	error = zap_get_stats(os, object, &zs);
982 	if (error)
983 		return;
984 
985 	if (zs.zs_ptrtbl_len == 0) {
986 		ASSERT(zs.zs_num_blocks == 1);
987 		(void) printf("\tmicrozap: %llu bytes, %llu entries\n",
988 		    (u_longlong_t)zs.zs_blocksize,
989 		    (u_longlong_t)zs.zs_num_entries);
990 		return;
991 	}
992 
993 	(void) printf("\tFat ZAP stats:\n");
994 
995 	(void) printf("\t\tPointer table:\n");
996 	(void) printf("\t\t\t%llu elements\n",
997 	    (u_longlong_t)zs.zs_ptrtbl_len);
998 	(void) printf("\t\t\tzt_blk: %llu\n",
999 	    (u_longlong_t)zs.zs_ptrtbl_zt_blk);
1000 	(void) printf("\t\t\tzt_numblks: %llu\n",
1001 	    (u_longlong_t)zs.zs_ptrtbl_zt_numblks);
1002 	(void) printf("\t\t\tzt_shift: %llu\n",
1003 	    (u_longlong_t)zs.zs_ptrtbl_zt_shift);
1004 	(void) printf("\t\t\tzt_blks_copied: %llu\n",
1005 	    (u_longlong_t)zs.zs_ptrtbl_blks_copied);
1006 	(void) printf("\t\t\tzt_nextblk: %llu\n",
1007 	    (u_longlong_t)zs.zs_ptrtbl_nextblk);
1008 
1009 	(void) printf("\t\tZAP entries: %llu\n",
1010 	    (u_longlong_t)zs.zs_num_entries);
1011 	(void) printf("\t\tLeaf blocks: %llu\n",
1012 	    (u_longlong_t)zs.zs_num_leafs);
1013 	(void) printf("\t\tTotal blocks: %llu\n",
1014 	    (u_longlong_t)zs.zs_num_blocks);
1015 	(void) printf("\t\tzap_block_type: 0x%llx\n",
1016 	    (u_longlong_t)zs.zs_block_type);
1017 	(void) printf("\t\tzap_magic: 0x%llx\n",
1018 	    (u_longlong_t)zs.zs_magic);
1019 	(void) printf("\t\tzap_salt: 0x%llx\n",
1020 	    (u_longlong_t)zs.zs_salt);
1021 
1022 	(void) printf("\t\tLeafs with 2^n pointers:\n");
1023 	dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);
1024 
1025 	(void) printf("\t\tBlocks with n*5 entries:\n");
1026 	dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);
1027 
1028 	(void) printf("\t\tBlocks n/10 full:\n");
1029 	dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);
1030 
1031 	(void) printf("\t\tEntries with n chunks:\n");
1032 	dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);
1033 
1034 	(void) printf("\t\tBuckets with n entries:\n");
1035 	dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
1036 }
1037 
1038 static void
dump_none(objset_t * os,uint64_t object,void * data,size_t size)1039 dump_none(objset_t *os, uint64_t object, void *data, size_t size)
1040 {
1041 	(void) os, (void) object, (void) data, (void) size;
1042 }
1043 
1044 static void
dump_unknown(objset_t * os,uint64_t object,void * data,size_t size)1045 dump_unknown(objset_t *os, uint64_t object, void *data, size_t size)
1046 {
1047 	(void) os, (void) object, (void) data, (void) size;
1048 	(void) printf("\tUNKNOWN OBJECT TYPE\n");
1049 }
1050 
1051 static void
dump_uint8(objset_t * os,uint64_t object,void * data,size_t size)1052 dump_uint8(objset_t *os, uint64_t object, void *data, size_t size)
1053 {
1054 	(void) os, (void) object, (void) data, (void) size;
1055 }
1056 
1057 static void
dump_uint64(objset_t * os,uint64_t object,void * data,size_t size)1058 dump_uint64(objset_t *os, uint64_t object, void *data, size_t size)
1059 {
1060 	uint64_t *arr;
1061 	uint64_t oursize;
1062 	if (dump_opt['d'] < 6)
1063 		return;
1064 
1065 	if (data == NULL) {
1066 		dmu_object_info_t doi;
1067 
1068 		VERIFY0(dmu_object_info(os, object, &doi));
1069 		size = doi.doi_max_offset;
1070 		/*
1071 		 * We cap the size at 1 mebibyte here to prevent
1072 		 * allocation failures and nigh-infinite printing if the
1073 		 * object is extremely large.
1074 		 */
1075 		oursize = MIN(size, 1 << 20);
1076 		arr = kmem_alloc(oursize, KM_SLEEP);
1077 
1078 		int err = dmu_read(os, object, 0, oursize, arr, 0);
1079 		if (err != 0) {
1080 			(void) printf("got error %u from dmu_read\n", err);
1081 			kmem_free(arr, oursize);
1082 			return;
1083 		}
1084 	} else {
1085 		/*
1086 		 * Even though the allocation is already done in this code path,
1087 		 * we still cap the size to prevent excessive printing.
1088 		 */
1089 		oursize = MIN(size, 1 << 20);
1090 		arr = data;
1091 	}
1092 
1093 	if (size == 0) {
1094 		if (data == NULL)
1095 			kmem_free(arr, oursize);
1096 		(void) printf("\t\t[]\n");
1097 		return;
1098 	}
1099 
1100 	(void) printf("\t\t[%0llx", (u_longlong_t)arr[0]);
1101 	for (size_t i = 1; i * sizeof (uint64_t) < oursize; i++) {
1102 		if (i % 4 != 0)
1103 			(void) printf(", %0llx", (u_longlong_t)arr[i]);
1104 		else
1105 			(void) printf(",\n\t\t%0llx", (u_longlong_t)arr[i]);
1106 	}
1107 	if (oursize != size)
1108 		(void) printf(", ... ");
1109 	(void) printf("]\n");
1110 
1111 	if (data == NULL)
1112 		kmem_free(arr, oursize);
1113 }
1114 
1115 static void
dump_zap(objset_t * os,uint64_t object,void * data,size_t size)1116 dump_zap(objset_t *os, uint64_t object, void *data, size_t size)
1117 {
1118 	(void) data, (void) size;
1119 	zap_cursor_t zc;
1120 	zap_attribute_t *attrp = zap_attribute_long_alloc();
1121 	void *prop;
1122 	unsigned i;
1123 
1124 	dump_zap_stats(os, object);
1125 	(void) printf("\n");
1126 
1127 	for (zap_cursor_init(&zc, os, object);
1128 	    zap_cursor_retrieve(&zc, attrp) == 0;
1129 	    zap_cursor_advance(&zc)) {
1130 		boolean_t key64 =
1131 		    !!(zap_getflags(zc.zc_zap) & ZAP_FLAG_UINT64_KEY);
1132 
1133 		if (key64)
1134 			(void) printf("\t\t0x%010" PRIu64 "x = ",
1135 			    *(uint64_t *)attrp->za_name);
1136 		else
1137 			(void) printf("\t\t%s = ", attrp->za_name);
1138 
1139 		if (attrp->za_num_integers == 0) {
1140 			(void) printf("\n");
1141 			continue;
1142 		}
1143 		prop = umem_zalloc(attrp->za_num_integers *
1144 		    attrp->za_integer_length, UMEM_NOFAIL);
1145 
1146 		if (key64)
1147 			(void) zap_lookup_uint64(os, object,
1148 			    (const uint64_t *)attrp->za_name, 1,
1149 			    attrp->za_integer_length, attrp->za_num_integers,
1150 			    prop);
1151 		else
1152 			(void) zap_lookup(os, object, attrp->za_name,
1153 			    attrp->za_integer_length, attrp->za_num_integers,
1154 			    prop);
1155 
1156 		if (attrp->za_integer_length == 1 && !key64) {
1157 			if (strcmp(attrp->za_name,
1158 			    DSL_CRYPTO_KEY_MASTER_KEY) == 0 ||
1159 			    strcmp(attrp->za_name,
1160 			    DSL_CRYPTO_KEY_HMAC_KEY) == 0 ||
1161 			    strcmp(attrp->za_name, DSL_CRYPTO_KEY_IV) == 0 ||
1162 			    strcmp(attrp->za_name, DSL_CRYPTO_KEY_MAC) == 0 ||
1163 			    strcmp(attrp->za_name,
1164 			    DMU_POOL_CHECKSUM_SALT) == 0) {
1165 				uint8_t *u8 = prop;
1166 
1167 				for (i = 0; i < attrp->za_num_integers; i++) {
1168 					(void) printf("%02x", u8[i]);
1169 				}
1170 			} else {
1171 				(void) printf("%s", (char *)prop);
1172 			}
1173 		} else {
1174 			for (i = 0; i < attrp->za_num_integers; i++) {
1175 				switch (attrp->za_integer_length) {
1176 				case 1:
1177 					(void) printf("%u ",
1178 					    ((uint8_t *)prop)[i]);
1179 					break;
1180 				case 2:
1181 					(void) printf("%u ",
1182 					    ((uint16_t *)prop)[i]);
1183 					break;
1184 				case 4:
1185 					(void) printf("%u ",
1186 					    ((uint32_t *)prop)[i]);
1187 					break;
1188 				case 8:
1189 					(void) printf("%lld ",
1190 					    (u_longlong_t)((int64_t *)prop)[i]);
1191 					break;
1192 				}
1193 			}
1194 		}
1195 		(void) printf("\n");
1196 		umem_free(prop,
1197 		    attrp->za_num_integers * attrp->za_integer_length);
1198 	}
1199 	zap_cursor_fini(&zc);
1200 	zap_attribute_free(attrp);
1201 }
1202 
1203 static void
dump_bpobj(objset_t * os,uint64_t object,void * data,size_t size)1204 dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size)
1205 {
1206 	bpobj_phys_t *bpop = data;
1207 	uint64_t i;
1208 	char bytes[32], comp[32], uncomp[32];
1209 
1210 	/* make sure the output won't get truncated */
1211 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
1212 	_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
1213 	_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
1214 
1215 	if (bpop == NULL)
1216 		return;
1217 
1218 	zdb_nicenum(bpop->bpo_bytes, bytes, sizeof (bytes));
1219 	zdb_nicenum(bpop->bpo_comp, comp, sizeof (comp));
1220 	zdb_nicenum(bpop->bpo_uncomp, uncomp, sizeof (uncomp));
1221 
1222 	(void) printf("\t\tnum_blkptrs = %llu\n",
1223 	    (u_longlong_t)bpop->bpo_num_blkptrs);
1224 	(void) printf("\t\tbytes = %s\n", bytes);
1225 	if (size >= BPOBJ_SIZE_V1) {
1226 		(void) printf("\t\tcomp = %s\n", comp);
1227 		(void) printf("\t\tuncomp = %s\n", uncomp);
1228 	}
1229 	if (size >= BPOBJ_SIZE_V2) {
1230 		(void) printf("\t\tsubobjs = %llu\n",
1231 		    (u_longlong_t)bpop->bpo_subobjs);
1232 		(void) printf("\t\tnum_subobjs = %llu\n",
1233 		    (u_longlong_t)bpop->bpo_num_subobjs);
1234 	}
1235 	if (size >= sizeof (*bpop)) {
1236 		(void) printf("\t\tnum_freed = %llu\n",
1237 		    (u_longlong_t)bpop->bpo_num_freed);
1238 	}
1239 
1240 	if (dump_opt['d'] < 5)
1241 		return;
1242 
1243 	for (i = 0; i < bpop->bpo_num_blkptrs; i++) {
1244 		char blkbuf[BP_SPRINTF_LEN];
1245 		blkptr_t bp;
1246 
1247 		int err = dmu_read(os, object,
1248 		    i * sizeof (bp), sizeof (bp), &bp, 0);
1249 		if (err != 0) {
1250 			(void) printf("got error %u from dmu_read\n", err);
1251 			break;
1252 		}
1253 		snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp,
1254 		    BP_GET_FREE(&bp));
1255 		(void) printf("\t%s\n", blkbuf);
1256 	}
1257 }
1258 
1259 static void
dump_bpobj_subobjs(objset_t * os,uint64_t object,void * data,size_t size)1260 dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size)
1261 {
1262 	(void) data, (void) size;
1263 	dmu_object_info_t doi;
1264 	int64_t i;
1265 
1266 	VERIFY0(dmu_object_info(os, object, &doi));
1267 	uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP);
1268 
1269 	int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0);
1270 	if (err != 0) {
1271 		(void) printf("got error %u from dmu_read\n", err);
1272 		kmem_free(subobjs, doi.doi_max_offset);
1273 		return;
1274 	}
1275 
1276 	int64_t last_nonzero = -1;
1277 	for (i = 0; i < doi.doi_max_offset / 8; i++) {
1278 		if (subobjs[i] != 0)
1279 			last_nonzero = i;
1280 	}
1281 
1282 	for (i = 0; i <= last_nonzero; i++) {
1283 		(void) printf("\t%llu\n", (u_longlong_t)subobjs[i]);
1284 	}
1285 	kmem_free(subobjs, doi.doi_max_offset);
1286 }
1287 
1288 static void
dump_ddt_zap(objset_t * os,uint64_t object,void * data,size_t size)1289 dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size)
1290 {
1291 	(void) data, (void) size;
1292 	dump_zap_stats(os, object);
1293 	/* contents are printed elsewhere, properly decoded */
1294 }
1295 
1296 static void
dump_sa_attrs(objset_t * os,uint64_t object,void * data,size_t size)1297 dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size)
1298 {
1299 	(void) data, (void) size;
1300 	zap_cursor_t zc;
1301 	zap_attribute_t *attrp = zap_attribute_alloc();
1302 
1303 	dump_zap_stats(os, object);
1304 	(void) printf("\n");
1305 
1306 	for (zap_cursor_init(&zc, os, object);
1307 	    zap_cursor_retrieve(&zc, attrp) == 0;
1308 	    zap_cursor_advance(&zc)) {
1309 		(void) printf("\t\t%s = ", attrp->za_name);
1310 		if (attrp->za_num_integers == 0) {
1311 			(void) printf("\n");
1312 			continue;
1313 		}
1314 		(void) printf(" %llx : [%d:%d:%d]\n",
1315 		    (u_longlong_t)attrp->za_first_integer,
1316 		    (int)ATTR_LENGTH(attrp->za_first_integer),
1317 		    (int)ATTR_BSWAP(attrp->za_first_integer),
1318 		    (int)ATTR_NUM(attrp->za_first_integer));
1319 	}
1320 	zap_cursor_fini(&zc);
1321 	zap_attribute_free(attrp);
1322 }
1323 
1324 static void
dump_sa_layouts(objset_t * os,uint64_t object,void * data,size_t size)1325 dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size)
1326 {
1327 	(void) data, (void) size;
1328 	zap_cursor_t zc;
1329 	zap_attribute_t *attrp = zap_attribute_alloc();
1330 	uint16_t *layout_attrs;
1331 	unsigned i;
1332 
1333 	dump_zap_stats(os, object);
1334 	(void) printf("\n");
1335 
1336 	for (zap_cursor_init(&zc, os, object);
1337 	    zap_cursor_retrieve(&zc, attrp) == 0;
1338 	    zap_cursor_advance(&zc)) {
1339 		(void) printf("\t\t%s = [", attrp->za_name);
1340 		if (attrp->za_num_integers == 0) {
1341 			(void) printf("\n");
1342 			continue;
1343 		}
1344 
1345 		VERIFY(attrp->za_integer_length == 2);
1346 		layout_attrs = umem_zalloc(attrp->za_num_integers *
1347 		    attrp->za_integer_length, UMEM_NOFAIL);
1348 
1349 		VERIFY(zap_lookup(os, object, attrp->za_name,
1350 		    attrp->za_integer_length,
1351 		    attrp->za_num_integers, layout_attrs) == 0);
1352 
1353 		for (i = 0; i != attrp->za_num_integers; i++)
1354 			(void) printf(" %d ", (int)layout_attrs[i]);
1355 		(void) printf("]\n");
1356 		umem_free(layout_attrs,
1357 		    attrp->za_num_integers * attrp->za_integer_length);
1358 	}
1359 	zap_cursor_fini(&zc);
1360 	zap_attribute_free(attrp);
1361 }
1362 
1363 static void
dump_zpldir(objset_t * os,uint64_t object,void * data,size_t size)1364 dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
1365 {
1366 	(void) data, (void) size;
1367 	zap_cursor_t zc;
1368 	zap_attribute_t *attrp = zap_attribute_long_alloc();
1369 	const char *typenames[] = {
1370 		/* 0 */ "not specified",
1371 		/* 1 */ "FIFO",
1372 		/* 2 */ "Character Device",
1373 		/* 3 */ "3 (invalid)",
1374 		/* 4 */ "Directory",
1375 		/* 5 */ "5 (invalid)",
1376 		/* 6 */ "Block Device",
1377 		/* 7 */ "7 (invalid)",
1378 		/* 8 */ "Regular File",
1379 		/* 9 */ "9 (invalid)",
1380 		/* 10 */ "Symbolic Link",
1381 		/* 11 */ "11 (invalid)",
1382 		/* 12 */ "Socket",
1383 		/* 13 */ "Door",
1384 		/* 14 */ "Event Port",
1385 		/* 15 */ "15 (invalid)",
1386 	};
1387 
1388 	dump_zap_stats(os, object);
1389 	(void) printf("\n");
1390 
1391 	for (zap_cursor_init(&zc, os, object);
1392 	    zap_cursor_retrieve(&zc, attrp) == 0;
1393 	    zap_cursor_advance(&zc)) {
1394 		(void) printf("\t\t%s = %lld (type: %s)\n",
1395 		    attrp->za_name, ZFS_DIRENT_OBJ(attrp->za_first_integer),
1396 		    typenames[ZFS_DIRENT_TYPE(attrp->za_first_integer)]);
1397 	}
1398 	zap_cursor_fini(&zc);
1399 	zap_attribute_free(attrp);
1400 }
1401 
1402 static int
get_dtl_refcount(vdev_t * vd)1403 get_dtl_refcount(vdev_t *vd)
1404 {
1405 	int refcount = 0;
1406 
1407 	if (vd->vdev_ops->vdev_op_leaf) {
1408 		space_map_t *sm = vd->vdev_dtl_sm;
1409 
1410 		if (sm != NULL &&
1411 		    sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1412 			return (1);
1413 		return (0);
1414 	}
1415 
1416 	for (unsigned c = 0; c < vd->vdev_children; c++)
1417 		refcount += get_dtl_refcount(vd->vdev_child[c]);
1418 	return (refcount);
1419 }
1420 
1421 static int
get_metaslab_refcount(vdev_t * vd)1422 get_metaslab_refcount(vdev_t *vd)
1423 {
1424 	int refcount = 0;
1425 
1426 	if (vd->vdev_top == vd) {
1427 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
1428 			space_map_t *sm = vd->vdev_ms[m]->ms_sm;
1429 
1430 			if (sm != NULL &&
1431 			    sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
1432 				refcount++;
1433 		}
1434 	}
1435 	for (unsigned c = 0; c < vd->vdev_children; c++)
1436 		refcount += get_metaslab_refcount(vd->vdev_child[c]);
1437 
1438 	return (refcount);
1439 }
1440 
1441 static int
get_obsolete_refcount(vdev_t * vd)1442 get_obsolete_refcount(vdev_t *vd)
1443 {
1444 	uint64_t obsolete_sm_object;
1445 	int refcount = 0;
1446 
1447 	VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1448 	if (vd->vdev_top == vd && obsolete_sm_object != 0) {
1449 		dmu_object_info_t doi;
1450 		VERIFY0(dmu_object_info(vd->vdev_spa->spa_meta_objset,
1451 		    obsolete_sm_object, &doi));
1452 		if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1453 			refcount++;
1454 		}
1455 	} else {
1456 		ASSERT3P(vd->vdev_obsolete_sm, ==, NULL);
1457 		ASSERT3U(obsolete_sm_object, ==, 0);
1458 	}
1459 	for (unsigned c = 0; c < vd->vdev_children; c++) {
1460 		refcount += get_obsolete_refcount(vd->vdev_child[c]);
1461 	}
1462 
1463 	return (refcount);
1464 }
1465 
1466 static int
get_prev_obsolete_spacemap_refcount(spa_t * spa)1467 get_prev_obsolete_spacemap_refcount(spa_t *spa)
1468 {
1469 	uint64_t prev_obj =
1470 	    spa->spa_condensing_indirect_phys.scip_prev_obsolete_sm_object;
1471 	if (prev_obj != 0) {
1472 		dmu_object_info_t doi;
1473 		VERIFY0(dmu_object_info(spa->spa_meta_objset, prev_obj, &doi));
1474 		if (doi.doi_bonus_size == sizeof (space_map_phys_t)) {
1475 			return (1);
1476 		}
1477 	}
1478 	return (0);
1479 }
1480 
1481 static int
get_checkpoint_refcount(vdev_t * vd)1482 get_checkpoint_refcount(vdev_t *vd)
1483 {
1484 	int refcount = 0;
1485 
1486 	if (vd->vdev_top == vd && vd->vdev_top_zap != 0 &&
1487 	    zap_contains(spa_meta_objset(vd->vdev_spa),
1488 	    vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) == 0)
1489 		refcount++;
1490 
1491 	for (uint64_t c = 0; c < vd->vdev_children; c++)
1492 		refcount += get_checkpoint_refcount(vd->vdev_child[c]);
1493 
1494 	return (refcount);
1495 }
1496 
1497 static int
get_log_spacemap_refcount(spa_t * spa)1498 get_log_spacemap_refcount(spa_t *spa)
1499 {
1500 	return (avl_numnodes(&spa->spa_sm_logs_by_txg));
1501 }
1502 
1503 static int
verify_spacemap_refcounts(spa_t * spa)1504 verify_spacemap_refcounts(spa_t *spa)
1505 {
1506 	uint64_t expected_refcount = 0;
1507 	uint64_t actual_refcount;
1508 
1509 	(void) feature_get_refcount(spa,
1510 	    &spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM],
1511 	    &expected_refcount);
1512 	actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
1513 	actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);
1514 	actual_refcount += get_obsolete_refcount(spa->spa_root_vdev);
1515 	actual_refcount += get_prev_obsolete_spacemap_refcount(spa);
1516 	actual_refcount += get_checkpoint_refcount(spa->spa_root_vdev);
1517 	actual_refcount += get_log_spacemap_refcount(spa);
1518 
1519 	if (expected_refcount != actual_refcount) {
1520 		(void) printf("space map refcount mismatch: expected %lld != "
1521 		    "actual %lld\n",
1522 		    (longlong_t)expected_refcount,
1523 		    (longlong_t)actual_refcount);
1524 		return (2);
1525 	}
1526 	return (0);
1527 }
1528 
1529 static void
dump_spacemap(objset_t * os,space_map_t * sm)1530 dump_spacemap(objset_t *os, space_map_t *sm)
1531 {
1532 	const char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
1533 	    "INVALID", "INVALID", "INVALID", "INVALID" };
1534 
1535 	if (sm == NULL)
1536 		return;
1537 
1538 	(void) printf("space map object %llu:\n",
1539 	    (longlong_t)sm->sm_object);
1540 	(void) printf("  smp_length = 0x%llx\n",
1541 	    (longlong_t)sm->sm_phys->smp_length);
1542 	(void) printf("  smp_alloc = 0x%llx\n",
1543 	    (longlong_t)sm->sm_phys->smp_alloc);
1544 
1545 	if (dump_opt['d'] < 6 && dump_opt['m'] < 4)
1546 		return;
1547 
1548 	/*
1549 	 * Print out the freelist entries in both encoded and decoded form.
1550 	 */
1551 	uint8_t mapshift = sm->sm_shift;
1552 	int64_t alloc = 0;
1553 	uint64_t word, entry_id = 0;
1554 	for (uint64_t offset = 0; offset < space_map_length(sm);
1555 	    offset += sizeof (word)) {
1556 
1557 		VERIFY0(dmu_read(os, space_map_object(sm), offset,
1558 		    sizeof (word), &word, DMU_READ_PREFETCH));
1559 
1560 		if (sm_entry_is_debug(word)) {
1561 			uint64_t de_txg = SM_DEBUG_TXG_DECODE(word);
1562 			uint64_t de_sync_pass = SM_DEBUG_SYNCPASS_DECODE(word);
1563 			if (de_txg == 0) {
1564 				(void) printf(
1565 				    "\t    [%6llu] PADDING\n",
1566 				    (u_longlong_t)entry_id);
1567 			} else {
1568 				(void) printf(
1569 				    "\t    [%6llu] %s: txg %llu pass %llu\n",
1570 				    (u_longlong_t)entry_id,
1571 				    ddata[SM_DEBUG_ACTION_DECODE(word)],
1572 				    (u_longlong_t)de_txg,
1573 				    (u_longlong_t)de_sync_pass);
1574 			}
1575 			entry_id++;
1576 			continue;
1577 		}
1578 
1579 		uint8_t words;
1580 		char entry_type;
1581 		uint64_t entry_off, entry_run, entry_vdev = SM_NO_VDEVID;
1582 
1583 		if (sm_entry_is_single_word(word)) {
1584 			entry_type = (SM_TYPE_DECODE(word) == SM_ALLOC) ?
1585 			    'A' : 'F';
1586 			entry_off = (SM_OFFSET_DECODE(word) << mapshift) +
1587 			    sm->sm_start;
1588 			entry_run = SM_RUN_DECODE(word) << mapshift;
1589 			words = 1;
1590 		} else {
1591 			/* it is a two-word entry so we read another word */
1592 			ASSERT(sm_entry_is_double_word(word));
1593 
1594 			uint64_t extra_word;
1595 			offset += sizeof (extra_word);
1596 			VERIFY0(dmu_read(os, space_map_object(sm), offset,
1597 			    sizeof (extra_word), &extra_word,
1598 			    DMU_READ_PREFETCH));
1599 
1600 			ASSERT3U(offset, <=, space_map_length(sm));
1601 
1602 			entry_run = SM2_RUN_DECODE(word) << mapshift;
1603 			entry_vdev = SM2_VDEV_DECODE(word);
1604 			entry_type = (SM2_TYPE_DECODE(extra_word) == SM_ALLOC) ?
1605 			    'A' : 'F';
1606 			entry_off = (SM2_OFFSET_DECODE(extra_word) <<
1607 			    mapshift) + sm->sm_start;
1608 			words = 2;
1609 		}
1610 
1611 		(void) printf("\t    [%6llu]    %c  range:"
1612 		    " %010llx-%010llx  size: %06llx vdev: %06llu words: %u\n",
1613 		    (u_longlong_t)entry_id,
1614 		    entry_type, (u_longlong_t)entry_off,
1615 		    (u_longlong_t)(entry_off + entry_run),
1616 		    (u_longlong_t)entry_run,
1617 		    (u_longlong_t)entry_vdev, words);
1618 
1619 		if (entry_type == 'A')
1620 			alloc += entry_run;
1621 		else
1622 			alloc -= entry_run;
1623 		entry_id++;
1624 	}
1625 	if (alloc != space_map_allocated(sm)) {
1626 		(void) printf("space_map_object alloc (%lld) INCONSISTENT "
1627 		    "with space map summary (%lld)\n",
1628 		    (longlong_t)space_map_allocated(sm), (longlong_t)alloc);
1629 	}
1630 }
1631 
1632 static void
dump_metaslab_stats(metaslab_t * msp)1633 dump_metaslab_stats(metaslab_t *msp)
1634 {
1635 	char maxbuf[32];
1636 	range_tree_t *rt = msp->ms_allocatable;
1637 	zfs_btree_t *t = &msp->ms_allocatable_by_size;
1638 	int free_pct = range_tree_space(rt) * 100 / msp->ms_size;
1639 
1640 	/* max sure nicenum has enough space */
1641 	_Static_assert(sizeof (maxbuf) >= NN_NUMBUF_SZ, "maxbuf truncated");
1642 
1643 	zdb_nicenum(metaslab_largest_allocatable(msp), maxbuf, sizeof (maxbuf));
1644 
1645 	(void) printf("\t %25s %10lu   %7s  %6s   %4s %4d%%\n",
1646 	    "segments", zfs_btree_numnodes(t), "maxsize", maxbuf,
1647 	    "freepct", free_pct);
1648 	(void) printf("\tIn-memory histogram:\n");
1649 	dump_histogram(rt->rt_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1650 }
1651 
1652 static void
dump_metaslab(metaslab_t * msp)1653 dump_metaslab(metaslab_t *msp)
1654 {
1655 	vdev_t *vd = msp->ms_group->mg_vd;
1656 	spa_t *spa = vd->vdev_spa;
1657 	space_map_t *sm = msp->ms_sm;
1658 	char freebuf[32];
1659 
1660 	zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf,
1661 	    sizeof (freebuf));
1662 
1663 	(void) printf(
1664 	    "\tmetaslab %6llu   offset %12llx   spacemap %6llu   free    %5s\n",
1665 	    (u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
1666 	    (u_longlong_t)space_map_object(sm), freebuf);
1667 
1668 	if (dump_opt['m'] > 2 && !dump_opt['L']) {
1669 		mutex_enter(&msp->ms_lock);
1670 		VERIFY0(metaslab_load(msp));
1671 		range_tree_stat_verify(msp->ms_allocatable);
1672 		dump_metaslab_stats(msp);
1673 		metaslab_unload(msp);
1674 		mutex_exit(&msp->ms_lock);
1675 	}
1676 
1677 	if (dump_opt['m'] > 1 && sm != NULL &&
1678 	    spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) {
1679 		/*
1680 		 * The space map histogram represents free space in chunks
1681 		 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
1682 		 */
1683 		(void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
1684 		    (u_longlong_t)msp->ms_fragmentation);
1685 		dump_histogram(sm->sm_phys->smp_histogram,
1686 		    SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift);
1687 	}
1688 
1689 	if (vd->vdev_ops == &vdev_draid_ops)
1690 		ASSERT3U(msp->ms_size, <=, 1ULL << vd->vdev_ms_shift);
1691 	else
1692 		ASSERT3U(msp->ms_size, ==, 1ULL << vd->vdev_ms_shift);
1693 
1694 	dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
1695 
1696 	if (spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) {
1697 		(void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n",
1698 		    (u_longlong_t)metaslab_unflushed_txg(msp));
1699 	}
1700 }
1701 
1702 static void
print_vdev_metaslab_header(vdev_t * vd)1703 print_vdev_metaslab_header(vdev_t *vd)
1704 {
1705 	vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
1706 	const char *bias_str = "";
1707 	if (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) {
1708 		bias_str = VDEV_ALLOC_BIAS_LOG;
1709 	} else if (alloc_bias == VDEV_BIAS_SPECIAL) {
1710 		bias_str = VDEV_ALLOC_BIAS_SPECIAL;
1711 	} else if (alloc_bias == VDEV_BIAS_DEDUP) {
1712 		bias_str = VDEV_ALLOC_BIAS_DEDUP;
1713 	}
1714 
1715 	uint64_t ms_flush_data_obj = 0;
1716 	if (vd->vdev_top_zap != 0) {
1717 		int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
1718 		    vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
1719 		    sizeof (uint64_t), 1, &ms_flush_data_obj);
1720 		if (error != ENOENT) {
1721 			ASSERT0(error);
1722 		}
1723 	}
1724 
1725 	(void) printf("\tvdev %10llu   %s",
1726 	    (u_longlong_t)vd->vdev_id, bias_str);
1727 
1728 	if (ms_flush_data_obj != 0) {
1729 		(void) printf("   ms_unflushed_phys object %llu",
1730 		    (u_longlong_t)ms_flush_data_obj);
1731 	}
1732 
1733 	(void) printf("\n\t%-10s%5llu   %-19s   %-15s   %-12s\n",
1734 	    "metaslabs", (u_longlong_t)vd->vdev_ms_count,
1735 	    "offset", "spacemap", "free");
1736 	(void) printf("\t%15s   %19s   %15s   %12s\n",
1737 	    "---------------", "-------------------",
1738 	    "---------------", "------------");
1739 }
1740 
1741 static void
dump_metaslab_groups(spa_t * spa,boolean_t show_special)1742 dump_metaslab_groups(spa_t *spa, boolean_t show_special)
1743 {
1744 	vdev_t *rvd = spa->spa_root_vdev;
1745 	metaslab_class_t *mc = spa_normal_class(spa);
1746 	metaslab_class_t *smc = spa_special_class(spa);
1747 	uint64_t fragmentation;
1748 
1749 	metaslab_class_histogram_verify(mc);
1750 
1751 	for (unsigned c = 0; c < rvd->vdev_children; c++) {
1752 		vdev_t *tvd = rvd->vdev_child[c];
1753 		metaslab_group_t *mg = tvd->vdev_mg;
1754 
1755 		if (mg == NULL || (mg->mg_class != mc &&
1756 		    (!show_special || mg->mg_class != smc)))
1757 			continue;
1758 
1759 		metaslab_group_histogram_verify(mg);
1760 		mg->mg_fragmentation = metaslab_group_fragmentation(mg);
1761 
1762 		(void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
1763 		    "fragmentation",
1764 		    (u_longlong_t)tvd->vdev_id,
1765 		    (u_longlong_t)tvd->vdev_ms_count);
1766 		if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
1767 			(void) printf("%3s\n", "-");
1768 		} else {
1769 			(void) printf("%3llu%%\n",
1770 			    (u_longlong_t)mg->mg_fragmentation);
1771 		}
1772 		dump_histogram(mg->mg_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1773 	}
1774 
1775 	(void) printf("\tpool %s\tfragmentation", spa_name(spa));
1776 	fragmentation = metaslab_class_fragmentation(mc);
1777 	if (fragmentation == ZFS_FRAG_INVALID)
1778 		(void) printf("\t%3s\n", "-");
1779 	else
1780 		(void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation);
1781 	dump_histogram(mc->mc_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
1782 }
1783 
1784 static void
print_vdev_indirect(vdev_t * vd)1785 print_vdev_indirect(vdev_t *vd)
1786 {
1787 	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
1788 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
1789 	vdev_indirect_births_t *vib = vd->vdev_indirect_births;
1790 
1791 	if (vim == NULL) {
1792 		ASSERT3P(vib, ==, NULL);
1793 		return;
1794 	}
1795 
1796 	ASSERT3U(vdev_indirect_mapping_object(vim), ==,
1797 	    vic->vic_mapping_object);
1798 	ASSERT3U(vdev_indirect_births_object(vib), ==,
1799 	    vic->vic_births_object);
1800 
1801 	(void) printf("indirect births obj %llu:\n",
1802 	    (longlong_t)vic->vic_births_object);
1803 	(void) printf("    vib_count = %llu\n",
1804 	    (longlong_t)vdev_indirect_births_count(vib));
1805 	for (uint64_t i = 0; i < vdev_indirect_births_count(vib); i++) {
1806 		vdev_indirect_birth_entry_phys_t *cur_vibe =
1807 		    &vib->vib_entries[i];
1808 		(void) printf("\toffset %llx -> txg %llu\n",
1809 		    (longlong_t)cur_vibe->vibe_offset,
1810 		    (longlong_t)cur_vibe->vibe_phys_birth_txg);
1811 	}
1812 	(void) printf("\n");
1813 
1814 	(void) printf("indirect mapping obj %llu:\n",
1815 	    (longlong_t)vic->vic_mapping_object);
1816 	(void) printf("    vim_max_offset = 0x%llx\n",
1817 	    (longlong_t)vdev_indirect_mapping_max_offset(vim));
1818 	(void) printf("    vim_bytes_mapped = 0x%llx\n",
1819 	    (longlong_t)vdev_indirect_mapping_bytes_mapped(vim));
1820 	(void) printf("    vim_count = %llu\n",
1821 	    (longlong_t)vdev_indirect_mapping_num_entries(vim));
1822 
1823 	if (dump_opt['d'] <= 5 && dump_opt['m'] <= 3)
1824 		return;
1825 
1826 	uint32_t *counts = vdev_indirect_mapping_load_obsolete_counts(vim);
1827 
1828 	for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
1829 		vdev_indirect_mapping_entry_phys_t *vimep =
1830 		    &vim->vim_entries[i];
1831 		(void) printf("\t<%llx:%llx:%llx> -> "
1832 		    "<%llx:%llx:%llx> (%x obsolete)\n",
1833 		    (longlong_t)vd->vdev_id,
1834 		    (longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
1835 		    (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1836 		    (longlong_t)DVA_GET_VDEV(&vimep->vimep_dst),
1837 		    (longlong_t)DVA_GET_OFFSET(&vimep->vimep_dst),
1838 		    (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
1839 		    counts[i]);
1840 	}
1841 	(void) printf("\n");
1842 
1843 	uint64_t obsolete_sm_object;
1844 	VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
1845 	if (obsolete_sm_object != 0) {
1846 		objset_t *mos = vd->vdev_spa->spa_meta_objset;
1847 		(void) printf("obsolete space map object %llu:\n",
1848 		    (u_longlong_t)obsolete_sm_object);
1849 		ASSERT(vd->vdev_obsolete_sm != NULL);
1850 		ASSERT3U(space_map_object(vd->vdev_obsolete_sm), ==,
1851 		    obsolete_sm_object);
1852 		dump_spacemap(mos, vd->vdev_obsolete_sm);
1853 		(void) printf("\n");
1854 	}
1855 }
1856 
1857 static void
dump_metaslabs(spa_t * spa)1858 dump_metaslabs(spa_t *spa)
1859 {
1860 	vdev_t *vd, *rvd = spa->spa_root_vdev;
1861 	uint64_t m, c = 0, children = rvd->vdev_children;
1862 
1863 	(void) printf("\nMetaslabs:\n");
1864 
1865 	if (!dump_opt['d'] && zopt_metaslab_args > 0) {
1866 		c = zopt_metaslab[0];
1867 
1868 		if (c >= children)
1869 			(void) fatal("bad vdev id: %llu", (u_longlong_t)c);
1870 
1871 		if (zopt_metaslab_args > 1) {
1872 			vd = rvd->vdev_child[c];
1873 			print_vdev_metaslab_header(vd);
1874 
1875 			for (m = 1; m < zopt_metaslab_args; m++) {
1876 				if (zopt_metaslab[m] < vd->vdev_ms_count)
1877 					dump_metaslab(
1878 					    vd->vdev_ms[zopt_metaslab[m]]);
1879 				else
1880 					(void) fprintf(stderr, "bad metaslab "
1881 					    "number %llu\n",
1882 					    (u_longlong_t)zopt_metaslab[m]);
1883 			}
1884 			(void) printf("\n");
1885 			return;
1886 		}
1887 		children = c + 1;
1888 	}
1889 	for (; c < children; c++) {
1890 		vd = rvd->vdev_child[c];
1891 		print_vdev_metaslab_header(vd);
1892 
1893 		print_vdev_indirect(vd);
1894 
1895 		for (m = 0; m < vd->vdev_ms_count; m++)
1896 			dump_metaslab(vd->vdev_ms[m]);
1897 		(void) printf("\n");
1898 	}
1899 }
1900 
1901 static void
dump_log_spacemaps(spa_t * spa)1902 dump_log_spacemaps(spa_t *spa)
1903 {
1904 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1905 		return;
1906 
1907 	(void) printf("\nLog Space Maps in Pool:\n");
1908 	for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
1909 	    sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) {
1910 		space_map_t *sm = NULL;
1911 		VERIFY0(space_map_open(&sm, spa_meta_objset(spa),
1912 		    sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT));
1913 
1914 		(void) printf("Log Spacemap object %llu txg %llu\n",
1915 		    (u_longlong_t)sls->sls_sm_obj, (u_longlong_t)sls->sls_txg);
1916 		dump_spacemap(spa->spa_meta_objset, sm);
1917 		space_map_close(sm);
1918 	}
1919 	(void) printf("\n");
1920 }
1921 
1922 static void
dump_ddt_entry(const ddt_t * ddt,const ddt_lightweight_entry_t * ddlwe,uint64_t index)1923 dump_ddt_entry(const ddt_t *ddt, const ddt_lightweight_entry_t *ddlwe,
1924     uint64_t index)
1925 {
1926 	const ddt_key_t *ddk = &ddlwe->ddlwe_key;
1927 	char blkbuf[BP_SPRINTF_LEN];
1928 	blkptr_t blk;
1929 	int p;
1930 
1931 	for (p = 0; p < DDT_NPHYS(ddt); p++) {
1932 		const ddt_univ_phys_t *ddp = &ddlwe->ddlwe_phys;
1933 		ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p);
1934 
1935 		if (ddt_phys_birth(ddp, v) == 0)
1936 			continue;
1937 		ddt_bp_create(ddt->ddt_checksum, ddk, ddp, v, &blk);
1938 		snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
1939 		(void) printf("index %llx refcnt %llu phys %d %s\n",
1940 		    (u_longlong_t)index, (u_longlong_t)ddt_phys_refcnt(ddp, v),
1941 		    p, blkbuf);
1942 	}
1943 }
1944 
1945 static void
dump_dedup_ratio(const ddt_stat_t * dds)1946 dump_dedup_ratio(const ddt_stat_t *dds)
1947 {
1948 	double rL, rP, rD, D, dedup, compress, copies;
1949 
1950 	if (dds->dds_blocks == 0)
1951 		return;
1952 
1953 	rL = (double)dds->dds_ref_lsize;
1954 	rP = (double)dds->dds_ref_psize;
1955 	rD = (double)dds->dds_ref_dsize;
1956 	D = (double)dds->dds_dsize;
1957 
1958 	dedup = rD / D;
1959 	compress = rL / rP;
1960 	copies = rD / rP;
1961 
1962 	(void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
1963 	    "dedup * compress / copies = %.2f\n\n",
1964 	    dedup, compress, copies, dedup * compress / copies);
1965 }
1966 
1967 static void
dump_ddt_log(ddt_t * ddt)1968 dump_ddt_log(ddt_t *ddt)
1969 {
1970 	if (ddt->ddt_version != DDT_VERSION_FDT ||
1971 	    !(ddt->ddt_flags & DDT_FLAG_LOG))
1972 		return;
1973 
1974 	for (int n = 0; n < 2; n++) {
1975 		ddt_log_t *ddl = &ddt->ddt_log[n];
1976 
1977 		char flagstr[64] = {0};
1978 		if (ddl->ddl_flags > 0) {
1979 			flagstr[0] = ' ';
1980 			int c = 1;
1981 			if (ddl->ddl_flags & DDL_FLAG_FLUSHING)
1982 				c += strlcpy(&flagstr[c], " FLUSHING",
1983 				    sizeof (flagstr) - c);
1984 			if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT)
1985 				c += strlcpy(&flagstr[c], " CHECKPOINT",
1986 				    sizeof (flagstr) - c);
1987 			if (ddl->ddl_flags &
1988 			    ~(DDL_FLAG_FLUSHING|DDL_FLAG_CHECKPOINT))
1989 				c += strlcpy(&flagstr[c], " UNKNOWN",
1990 				    sizeof (flagstr) - c);
1991 			flagstr[1] = '[';
1992 			flagstr[c++] = ']';
1993 		}
1994 
1995 		uint64_t count = avl_numnodes(&ddl->ddl_tree);
1996 
1997 		printf(DMU_POOL_DDT_LOG ": flags=0x%02x%s; obj=%llu; "
1998 		    "len=%llu; txg=%llu; entries=%llu\n",
1999 		    zio_checksum_table[ddt->ddt_checksum].ci_name, n,
2000 		    ddl->ddl_flags, flagstr,
2001 		    (u_longlong_t)ddl->ddl_object,
2002 		    (u_longlong_t)ddl->ddl_length,
2003 		    (u_longlong_t)ddl->ddl_first_txg, (u_longlong_t)count);
2004 
2005 		if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT) {
2006 			const ddt_key_t *ddk = &ddl->ddl_checkpoint;
2007 			printf("    checkpoint: "
2008 			    "%016llx:%016llx:%016llx:%016llx:%016llx\n",
2009 			    (u_longlong_t)ddk->ddk_cksum.zc_word[0],
2010 			    (u_longlong_t)ddk->ddk_cksum.zc_word[1],
2011 			    (u_longlong_t)ddk->ddk_cksum.zc_word[2],
2012 			    (u_longlong_t)ddk->ddk_cksum.zc_word[3],
2013 			    (u_longlong_t)ddk->ddk_prop);
2014 		}
2015 
2016 		if (count == 0 || dump_opt['D'] < 4)
2017 			continue;
2018 
2019 		ddt_lightweight_entry_t ddlwe;
2020 		uint64_t index = 0;
2021 		for (ddt_log_entry_t *ddle = avl_first(&ddl->ddl_tree);
2022 		    ddle; ddle = AVL_NEXT(&ddl->ddl_tree, ddle)) {
2023 			DDT_LOG_ENTRY_TO_LIGHTWEIGHT(ddt, ddle, &ddlwe);
2024 			dump_ddt_entry(ddt, &ddlwe, index++);
2025 		}
2026 	}
2027 }
2028 
2029 static void
dump_ddt_object(ddt_t * ddt,ddt_type_t type,ddt_class_t class)2030 dump_ddt_object(ddt_t *ddt, ddt_type_t type, ddt_class_t class)
2031 {
2032 	char name[DDT_NAMELEN];
2033 	ddt_lightweight_entry_t ddlwe;
2034 	uint64_t walk = 0;
2035 	dmu_object_info_t doi;
2036 	uint64_t count, dspace, mspace;
2037 	int error;
2038 
2039 	error = ddt_object_info(ddt, type, class, &doi);
2040 
2041 	if (error == ENOENT)
2042 		return;
2043 	ASSERT(error == 0);
2044 
2045 	error = ddt_object_count(ddt, type, class, &count);
2046 	ASSERT(error == 0);
2047 	if (count == 0)
2048 		return;
2049 
2050 	dspace = doi.doi_physical_blocks_512 << 9;
2051 	mspace = doi.doi_fill_count * doi.doi_data_block_size;
2052 
2053 	ddt_object_name(ddt, type, class, name);
2054 
2055 	(void) printf("%s: dspace=%llu; mspace=%llu; entries=%llu\n", name,
2056 	    (u_longlong_t)dspace, (u_longlong_t)mspace, (u_longlong_t)count);
2057 
2058 	if (dump_opt['D'] < 3)
2059 		return;
2060 
2061 	zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class]);
2062 
2063 	if (dump_opt['D'] < 4)
2064 		return;
2065 
2066 	if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE)
2067 		return;
2068 
2069 	(void) printf("%s contents:\n\n", name);
2070 
2071 	while ((error = ddt_object_walk(ddt, type, class, &walk, &ddlwe)) == 0)
2072 		dump_ddt_entry(ddt, &ddlwe, walk);
2073 
2074 	ASSERT3U(error, ==, ENOENT);
2075 
2076 	(void) printf("\n");
2077 }
2078 
2079 static void
dump_ddt(ddt_t * ddt)2080 dump_ddt(ddt_t *ddt)
2081 {
2082 	if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
2083 		return;
2084 
2085 	char flagstr[64] = {0};
2086 	if (ddt->ddt_flags > 0) {
2087 		flagstr[0] = ' ';
2088 		int c = 1;
2089 		if (ddt->ddt_flags & DDT_FLAG_FLAT)
2090 			c += strlcpy(&flagstr[c], " FLAT",
2091 			    sizeof (flagstr) - c);
2092 		if (ddt->ddt_flags & DDT_FLAG_LOG)
2093 			c += strlcpy(&flagstr[c], " LOG",
2094 			    sizeof (flagstr) - c);
2095 		if (ddt->ddt_flags & ~DDT_FLAG_MASK)
2096 			c += strlcpy(&flagstr[c], " UNKNOWN",
2097 			    sizeof (flagstr) - c);
2098 		flagstr[1] = '[';
2099 		flagstr[c] = ']';
2100 	}
2101 
2102 	printf("DDT-%s: version=%llu [%s]; flags=0x%02llx%s; rootobj=%llu\n",
2103 	    zio_checksum_table[ddt->ddt_checksum].ci_name,
2104 	    (u_longlong_t)ddt->ddt_version,
2105 	    (ddt->ddt_version == 0) ? "LEGACY" :
2106 	    (ddt->ddt_version == 1) ? "FDT" : "UNKNOWN",
2107 	    (u_longlong_t)ddt->ddt_flags, flagstr,
2108 	    (u_longlong_t)ddt->ddt_dir_object);
2109 
2110 	for (ddt_type_t type = 0; type < DDT_TYPES; type++)
2111 		for (ddt_class_t class = 0; class < DDT_CLASSES; class++)
2112 			dump_ddt_object(ddt, type, class);
2113 
2114 	dump_ddt_log(ddt);
2115 }
2116 
2117 static void
dump_all_ddts(spa_t * spa)2118 dump_all_ddts(spa_t *spa)
2119 {
2120 	ddt_histogram_t ddh_total = {{{0}}};
2121 	ddt_stat_t dds_total = {0};
2122 
2123 	for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++)
2124 		dump_ddt(spa->spa_ddt[c]);
2125 
2126 	ddt_get_dedup_stats(spa, &dds_total);
2127 
2128 	if (dds_total.dds_blocks == 0) {
2129 		(void) printf("All DDTs are empty\n");
2130 		return;
2131 	}
2132 
2133 	(void) printf("\n");
2134 
2135 	if (dump_opt['D'] > 1) {
2136 		(void) printf("DDT histogram (aggregated over all DDTs):\n");
2137 		ddt_get_dedup_histogram(spa, &ddh_total);
2138 		zpool_dump_ddt(&dds_total, &ddh_total);
2139 	}
2140 
2141 	dump_dedup_ratio(&dds_total);
2142 
2143 	/*
2144 	 * Dump a histogram of unique class entry age
2145 	 */
2146 	if (dump_opt['D'] == 3 && getenv("ZDB_DDT_UNIQUE_AGE_HIST") != NULL) {
2147 		ddt_age_histo_t histogram;
2148 
2149 		(void) printf("DDT walk unique, building age histogram...\n");
2150 		ddt_prune_walk(spa, 0, &histogram);
2151 
2152 		/*
2153 		 * print out histogram for unique entry class birth
2154 		 */
2155 		if (histogram.dah_entries > 0) {
2156 			(void) printf("%5s  %9s  %4s\n",
2157 			    "age", "blocks", "amnt");
2158 			(void) printf("%5s  %9s  %4s\n",
2159 			    "-----", "---------", "----");
2160 			for (int i = 0; i < HIST_BINS; i++) {
2161 				(void) printf("%5d  %9d %4d%%\n", 1 << i,
2162 				    (int)histogram.dah_age_histo[i],
2163 				    (int)((histogram.dah_age_histo[i] * 100) /
2164 				    histogram.dah_entries));
2165 			}
2166 		}
2167 	}
2168 }
2169 
2170 static void
dump_brt(spa_t * spa)2171 dump_brt(spa_t *spa)
2172 {
2173 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) {
2174 		printf("BRT: unsupported on this pool\n");
2175 		return;
2176 	}
2177 
2178 	if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
2179 		printf("BRT: empty\n");
2180 		return;
2181 	}
2182 
2183 	char count[32], used[32], saved[32];
2184 	zdb_nicebytes(brt_get_used(spa), used, sizeof (used));
2185 	zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved));
2186 	uint64_t ratio = brt_get_ratio(spa);
2187 	printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved,
2188 	    (u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100));
2189 
2190 	if (dump_opt['T'] < 2)
2191 		return;
2192 
2193 	for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2194 		brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2195 		if (!brtvd->bv_initiated) {
2196 			printf("BRT: vdev %" PRIu64 ": empty\n", vdevid);
2197 			continue;
2198 		}
2199 
2200 		zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count));
2201 		zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used));
2202 		zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved));
2203 		printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n",
2204 		    vdevid, count, used, saved);
2205 	}
2206 
2207 	if (dump_opt['T'] < 3)
2208 		return;
2209 
2210 	/* -TTT shows a per-vdev histograms; -TTTT shows all entries */
2211 	boolean_t do_histo = dump_opt['T'] == 3;
2212 
2213 	char dva[64];
2214 
2215 	if (!do_histo)
2216 		printf("\n%-16s %-10s\n", "DVA", "REFCNT");
2217 
2218 	for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
2219 		brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
2220 		if (!brtvd->bv_initiated)
2221 			continue;
2222 
2223 		uint64_t counts[64] = {};
2224 
2225 		zap_cursor_t zc;
2226 		zap_attribute_t *za = zap_attribute_alloc();
2227 		for (zap_cursor_init(&zc, spa->spa_meta_objset,
2228 		    brtvd->bv_mos_entries);
2229 		    zap_cursor_retrieve(&zc, za) == 0;
2230 		    zap_cursor_advance(&zc)) {
2231 			uint64_t refcnt;
2232 			VERIFY0(zap_lookup_uint64(spa->spa_meta_objset,
2233 			    brtvd->bv_mos_entries,
2234 			    (const uint64_t *)za->za_name, 1,
2235 			    za->za_integer_length, za->za_num_integers,
2236 			    &refcnt));
2237 
2238 			if (do_histo)
2239 				counts[highbit64(refcnt)]++;
2240 			else {
2241 				uint64_t offset =
2242 				    *(const uint64_t *)za->za_name;
2243 
2244 				snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx",
2245 				    vdevid, (u_longlong_t)offset);
2246 				printf("%-16s %-10llu\n", dva,
2247 				    (u_longlong_t)refcnt);
2248 			}
2249 		}
2250 		zap_cursor_fini(&zc);
2251 		zap_attribute_free(za);
2252 
2253 		if (do_histo) {
2254 			printf("\nBRT: vdev %" PRIu64
2255 			    ": DVAs with 2^n refcnts:\n", vdevid);
2256 			dump_histogram(counts, 64, 0);
2257 		}
2258 	}
2259 }
2260 
2261 static void
dump_dtl_seg(void * arg,uint64_t start,uint64_t size)2262 dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
2263 {
2264 	char *prefix = arg;
2265 
2266 	(void) printf("%s [%llu,%llu) length %llu\n",
2267 	    prefix,
2268 	    (u_longlong_t)start,
2269 	    (u_longlong_t)(start + size),
2270 	    (u_longlong_t)(size));
2271 }
2272 
2273 static void
dump_dtl(vdev_t * vd,int indent)2274 dump_dtl(vdev_t *vd, int indent)
2275 {
2276 	spa_t *spa = vd->vdev_spa;
2277 	boolean_t required;
2278 	const char *name[DTL_TYPES] = { "missing", "partial", "scrub",
2279 		"outage" };
2280 	char prefix[256];
2281 
2282 	spa_vdev_state_enter(spa, SCL_NONE);
2283 	required = vdev_dtl_required(vd);
2284 	(void) spa_vdev_state_exit(spa, NULL, 0);
2285 
2286 	if (indent == 0)
2287 		(void) printf("\nDirty time logs:\n\n");
2288 
2289 	(void) printf("\t%*s%s [%s]\n", indent, "",
2290 	    vd->vdev_path ? vd->vdev_path :
2291 	    vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa),
2292 	    required ? "DTL-required" : "DTL-expendable");
2293 
2294 	for (int t = 0; t < DTL_TYPES; t++) {
2295 		range_tree_t *rt = vd->vdev_dtl[t];
2296 		if (range_tree_space(rt) == 0)
2297 			continue;
2298 		(void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
2299 		    indent + 2, "", name[t]);
2300 		range_tree_walk(rt, dump_dtl_seg, prefix);
2301 		if (dump_opt['d'] > 5 && vd->vdev_children == 0)
2302 			dump_spacemap(spa->spa_meta_objset,
2303 			    vd->vdev_dtl_sm);
2304 	}
2305 
2306 	for (unsigned c = 0; c < vd->vdev_children; c++)
2307 		dump_dtl(vd->vdev_child[c], indent + 4);
2308 }
2309 
2310 static void
dump_history(spa_t * spa)2311 dump_history(spa_t *spa)
2312 {
2313 	nvlist_t **events = NULL;
2314 	char *buf;
2315 	uint64_t resid, len, off = 0;
2316 	uint_t num = 0;
2317 	int error;
2318 	char tbuf[30];
2319 
2320 	if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) {
2321 		(void) fprintf(stderr, "%s: unable to allocate I/O buffer\n",
2322 		    __func__);
2323 		return;
2324 	}
2325 
2326 	do {
2327 		len = SPA_OLD_MAXBLOCKSIZE;
2328 
2329 		if ((error = spa_history_get(spa, &off, &len, buf)) != 0) {
2330 			(void) fprintf(stderr, "Unable to read history: "
2331 			    "error %d\n", error);
2332 			free(buf);
2333 			return;
2334 		}
2335 
2336 		if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0)
2337 			break;
2338 
2339 		off -= resid;
2340 	} while (len != 0);
2341 
2342 	(void) printf("\nHistory:\n");
2343 	for (unsigned i = 0; i < num; i++) {
2344 		boolean_t printed = B_FALSE;
2345 
2346 		if (nvlist_exists(events[i], ZPOOL_HIST_TIME)) {
2347 			time_t tsec;
2348 			struct tm t;
2349 
2350 			tsec = fnvlist_lookup_uint64(events[i],
2351 			    ZPOOL_HIST_TIME);
2352 			(void) localtime_r(&tsec, &t);
2353 			(void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t);
2354 		} else {
2355 			tbuf[0] = '\0';
2356 		}
2357 
2358 		if (nvlist_exists(events[i], ZPOOL_HIST_CMD)) {
2359 			(void) printf("%s %s\n", tbuf,
2360 			    fnvlist_lookup_string(events[i], ZPOOL_HIST_CMD));
2361 		} else if (nvlist_exists(events[i], ZPOOL_HIST_INT_EVENT)) {
2362 			uint64_t ievent;
2363 
2364 			ievent = fnvlist_lookup_uint64(events[i],
2365 			    ZPOOL_HIST_INT_EVENT);
2366 			if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS)
2367 				goto next;
2368 
2369 			(void) printf(" %s [internal %s txg:%ju] %s\n",
2370 			    tbuf,
2371 			    zfs_history_event_names[ievent],
2372 			    fnvlist_lookup_uint64(events[i],
2373 			    ZPOOL_HIST_TXG),
2374 			    fnvlist_lookup_string(events[i],
2375 			    ZPOOL_HIST_INT_STR));
2376 		} else if (nvlist_exists(events[i], ZPOOL_HIST_INT_NAME)) {
2377 			(void) printf("%s [txg:%ju] %s", tbuf,
2378 			    fnvlist_lookup_uint64(events[i],
2379 			    ZPOOL_HIST_TXG),
2380 			    fnvlist_lookup_string(events[i],
2381 			    ZPOOL_HIST_INT_NAME));
2382 
2383 			if (nvlist_exists(events[i], ZPOOL_HIST_DSNAME)) {
2384 				(void) printf(" %s (%llu)",
2385 				    fnvlist_lookup_string(events[i],
2386 				    ZPOOL_HIST_DSNAME),
2387 				    (u_longlong_t)fnvlist_lookup_uint64(
2388 				    events[i],
2389 				    ZPOOL_HIST_DSID));
2390 			}
2391 
2392 			(void) printf(" %s\n", fnvlist_lookup_string(events[i],
2393 			    ZPOOL_HIST_INT_STR));
2394 		} else if (nvlist_exists(events[i], ZPOOL_HIST_IOCTL)) {
2395 			(void) printf("%s ioctl %s\n", tbuf,
2396 			    fnvlist_lookup_string(events[i],
2397 			    ZPOOL_HIST_IOCTL));
2398 
2399 			if (nvlist_exists(events[i], ZPOOL_HIST_INPUT_NVL)) {
2400 				(void) printf("    input:\n");
2401 				dump_nvlist(fnvlist_lookup_nvlist(events[i],
2402 				    ZPOOL_HIST_INPUT_NVL), 8);
2403 			}
2404 			if (nvlist_exists(events[i], ZPOOL_HIST_OUTPUT_NVL)) {
2405 				(void) printf("    output:\n");
2406 				dump_nvlist(fnvlist_lookup_nvlist(events[i],
2407 				    ZPOOL_HIST_OUTPUT_NVL), 8);
2408 			}
2409 			if (nvlist_exists(events[i], ZPOOL_HIST_ERRNO)) {
2410 				(void) printf("    errno: %lld\n",
2411 				    (longlong_t)fnvlist_lookup_int64(events[i],
2412 				    ZPOOL_HIST_ERRNO));
2413 			}
2414 		} else {
2415 			goto next;
2416 		}
2417 
2418 		printed = B_TRUE;
2419 next:
2420 		if (dump_opt['h'] > 1) {
2421 			if (!printed)
2422 				(void) printf("unrecognized record:\n");
2423 			dump_nvlist(events[i], 2);
2424 		}
2425 	}
2426 	free(buf);
2427 }
2428 
2429 static void
dump_dnode(objset_t * os,uint64_t object,void * data,size_t size)2430 dump_dnode(objset_t *os, uint64_t object, void *data, size_t size)
2431 {
2432 	(void) os, (void) object, (void) data, (void) size;
2433 }
2434 
2435 static uint64_t
blkid2offset(const dnode_phys_t * dnp,const blkptr_t * bp,const zbookmark_phys_t * zb)2436 blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
2437     const zbookmark_phys_t *zb)
2438 {
2439 	if (dnp == NULL) {
2440 		ASSERT(zb->zb_level < 0);
2441 		if (zb->zb_object == 0)
2442 			return (zb->zb_blkid);
2443 		return (zb->zb_blkid * BP_GET_LSIZE(bp));
2444 	}
2445 
2446 	ASSERT(zb->zb_level >= 0);
2447 
2448 	return ((zb->zb_blkid <<
2449 	    (zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) *
2450 	    dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
2451 }
2452 
2453 static void
snprintf_zstd_header(spa_t * spa,char * blkbuf,size_t buflen,const blkptr_t * bp)2454 snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
2455     const blkptr_t *bp)
2456 {
2457 	static abd_t *pabd = NULL;
2458 	void *buf;
2459 	zio_t *zio;
2460 	zfs_zstdhdr_t zstd_hdr;
2461 	int error;
2462 
2463 	if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD)
2464 		return;
2465 
2466 	if (BP_IS_HOLE(bp))
2467 		return;
2468 
2469 	if (BP_IS_EMBEDDED(bp)) {
2470 		buf = malloc(SPA_MAXBLOCKSIZE);
2471 		if (buf == NULL) {
2472 			(void) fprintf(stderr, "out of memory\n");
2473 			zdb_exit(1);
2474 		}
2475 		decode_embedded_bp_compressed(bp, buf);
2476 		memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2477 		free(buf);
2478 		zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2479 		zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2480 		(void) snprintf(blkbuf + strlen(blkbuf),
2481 		    buflen - strlen(blkbuf),
2482 		    " ZSTD:size=%u:version=%u:level=%u:EMBEDDED",
2483 		    zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2484 		    zfs_get_hdrlevel(&zstd_hdr));
2485 		return;
2486 	}
2487 
2488 	if (!pabd)
2489 		pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
2490 	zio = zio_root(spa, NULL, NULL, 0);
2491 
2492 	/* Decrypt but don't decompress so we can read the compression header */
2493 	zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL,
2494 	    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS,
2495 	    NULL));
2496 	error = zio_wait(zio);
2497 	if (error) {
2498 		(void) fprintf(stderr, "read failed: %d\n", error);
2499 		return;
2500 	}
2501 	buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp));
2502 	memcpy(&zstd_hdr, buf, sizeof (zstd_hdr));
2503 	zstd_hdr.c_len = BE_32(zstd_hdr.c_len);
2504 	zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level);
2505 
2506 	(void) snprintf(blkbuf + strlen(blkbuf),
2507 	    buflen - strlen(blkbuf),
2508 	    " ZSTD:size=%u:version=%u:level=%u:NORMAL",
2509 	    zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr),
2510 	    zfs_get_hdrlevel(&zstd_hdr));
2511 
2512 	abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp));
2513 }
2514 
2515 static void
snprintf_blkptr_compact(char * blkbuf,size_t buflen,const blkptr_t * bp,boolean_t bp_freed)2516 snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
2517     boolean_t bp_freed)
2518 {
2519 	const dva_t *dva = bp->blk_dva;
2520 	int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1;
2521 	int i;
2522 
2523 	if (dump_opt['b'] >= 6) {
2524 		snprintf_blkptr(blkbuf, buflen, bp);
2525 		if (bp_freed) {
2526 			(void) snprintf(blkbuf + strlen(blkbuf),
2527 			    buflen - strlen(blkbuf), " %s", "FREE");
2528 		}
2529 		return;
2530 	}
2531 
2532 	if (BP_IS_EMBEDDED(bp)) {
2533 		(void) sprintf(blkbuf,
2534 		    "EMBEDDED et=%u %llxL/%llxP B=%llu",
2535 		    (int)BPE_GET_ETYPE(bp),
2536 		    (u_longlong_t)BPE_GET_LSIZE(bp),
2537 		    (u_longlong_t)BPE_GET_PSIZE(bp),
2538 		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2539 		return;
2540 	}
2541 
2542 	blkbuf[0] = '\0';
2543 
2544 	for (i = 0; i < ndvas; i++)
2545 		(void) snprintf(blkbuf + strlen(blkbuf),
2546 		    buflen - strlen(blkbuf), "%llu:%llx:%llx ",
2547 		    (u_longlong_t)DVA_GET_VDEV(&dva[i]),
2548 		    (u_longlong_t)DVA_GET_OFFSET(&dva[i]),
2549 		    (u_longlong_t)DVA_GET_ASIZE(&dva[i]));
2550 
2551 	if (BP_IS_HOLE(bp)) {
2552 		(void) snprintf(blkbuf + strlen(blkbuf),
2553 		    buflen - strlen(blkbuf),
2554 		    "%llxL B=%llu",
2555 		    (u_longlong_t)BP_GET_LSIZE(bp),
2556 		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
2557 	} else {
2558 		(void) snprintf(blkbuf + strlen(blkbuf),
2559 		    buflen - strlen(blkbuf),
2560 		    "%llxL/%llxP F=%llu B=%llu/%llu",
2561 		    (u_longlong_t)BP_GET_LSIZE(bp),
2562 		    (u_longlong_t)BP_GET_PSIZE(bp),
2563 		    (u_longlong_t)BP_GET_FILL(bp),
2564 		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),
2565 		    (u_longlong_t)BP_GET_BIRTH(bp));
2566 		if (bp_freed)
2567 			(void) snprintf(blkbuf + strlen(blkbuf),
2568 			    buflen - strlen(blkbuf), " %s", "FREE");
2569 		(void) snprintf(blkbuf + strlen(blkbuf),
2570 		    buflen - strlen(blkbuf),
2571 		    " cksum=%016llx:%016llx:%016llx:%016llx",
2572 		    (u_longlong_t)bp->blk_cksum.zc_word[0],
2573 		    (u_longlong_t)bp->blk_cksum.zc_word[1],
2574 		    (u_longlong_t)bp->blk_cksum.zc_word[2],
2575 		    (u_longlong_t)bp->blk_cksum.zc_word[3]);
2576 	}
2577 }
2578 
2579 static void
print_indirect(spa_t * spa,blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp)2580 print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb,
2581     const dnode_phys_t *dnp)
2582 {
2583 	char blkbuf[BP_SPRINTF_LEN];
2584 	int l;
2585 
2586 	if (!BP_IS_EMBEDDED(bp)) {
2587 		ASSERT3U(BP_GET_TYPE(bp), ==, dnp->dn_type);
2588 		ASSERT3U(BP_GET_LEVEL(bp), ==, zb->zb_level);
2589 	}
2590 
2591 	(void) printf("%16llx ", (u_longlong_t)blkid2offset(dnp, bp, zb));
2592 
2593 	ASSERT(zb->zb_level >= 0);
2594 
2595 	for (l = dnp->dn_nlevels - 1; l >= -1; l--) {
2596 		if (l == zb->zb_level) {
2597 			(void) printf("L%llx", (u_longlong_t)zb->zb_level);
2598 		} else {
2599 			(void) printf(" ");
2600 		}
2601 	}
2602 
2603 	snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE);
2604 	if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD)
2605 		snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp);
2606 	(void) printf("%s\n", blkbuf);
2607 }
2608 
2609 static int
visit_indirect(spa_t * spa,const dnode_phys_t * dnp,blkptr_t * bp,const zbookmark_phys_t * zb)2610 visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
2611     blkptr_t *bp, const zbookmark_phys_t *zb)
2612 {
2613 	int err = 0;
2614 
2615 	if (BP_GET_LOGICAL_BIRTH(bp) == 0)
2616 		return (0);
2617 
2618 	print_indirect(spa, bp, zb, dnp);
2619 
2620 	if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
2621 		arc_flags_t flags = ARC_FLAG_WAIT;
2622 		int i;
2623 		blkptr_t *cbp;
2624 		int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
2625 		arc_buf_t *buf;
2626 		uint64_t fill = 0;
2627 		ASSERT(!BP_IS_REDACTED(bp));
2628 
2629 		err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
2630 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
2631 		if (err)
2632 			return (err);
2633 		ASSERT(buf->b_data);
2634 
2635 		/* recursively visit blocks below this */
2636 		cbp = buf->b_data;
2637 		for (i = 0; i < epb; i++, cbp++) {
2638 			zbookmark_phys_t czb;
2639 
2640 			SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
2641 			    zb->zb_level - 1,
2642 			    zb->zb_blkid * epb + i);
2643 			err = visit_indirect(spa, dnp, cbp, &czb);
2644 			if (err)
2645 				break;
2646 			fill += BP_GET_FILL(cbp);
2647 		}
2648 		if (!err)
2649 			ASSERT3U(fill, ==, BP_GET_FILL(bp));
2650 		arc_buf_destroy(buf, &buf);
2651 	}
2652 
2653 	return (err);
2654 }
2655 
2656 static void
dump_indirect(dnode_t * dn)2657 dump_indirect(dnode_t *dn)
2658 {
2659 	dnode_phys_t *dnp = dn->dn_phys;
2660 	zbookmark_phys_t czb;
2661 
2662 	(void) printf("Indirect blocks:\n");
2663 
2664 	SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
2665 	    dn->dn_object, dnp->dn_nlevels - 1, 0);
2666 	for (int j = 0; j < dnp->dn_nblkptr; j++) {
2667 		czb.zb_blkid = j;
2668 		(void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
2669 		    &dnp->dn_blkptr[j], &czb);
2670 	}
2671 
2672 	(void) printf("\n");
2673 }
2674 
2675 static void
dump_dsl_dir(objset_t * os,uint64_t object,void * data,size_t size)2676 dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size)
2677 {
2678 	(void) os, (void) object;
2679 	dsl_dir_phys_t *dd = data;
2680 	time_t crtime;
2681 	char nice[32];
2682 
2683 	/* make sure nicenum has enough space */
2684 	_Static_assert(sizeof (nice) >= NN_NUMBUF_SZ, "nice truncated");
2685 
2686 	if (dd == NULL)
2687 		return;
2688 
2689 	ASSERT3U(size, >=, sizeof (dsl_dir_phys_t));
2690 
2691 	crtime = dd->dd_creation_time;
2692 	(void) printf("\t\tcreation_time = %s", ctime(&crtime));
2693 	(void) printf("\t\thead_dataset_obj = %llu\n",
2694 	    (u_longlong_t)dd->dd_head_dataset_obj);
2695 	(void) printf("\t\tparent_dir_obj = %llu\n",
2696 	    (u_longlong_t)dd->dd_parent_obj);
2697 	(void) printf("\t\torigin_obj = %llu\n",
2698 	    (u_longlong_t)dd->dd_origin_obj);
2699 	(void) printf("\t\tchild_dir_zapobj = %llu\n",
2700 	    (u_longlong_t)dd->dd_child_dir_zapobj);
2701 	zdb_nicenum(dd->dd_used_bytes, nice, sizeof (nice));
2702 	(void) printf("\t\tused_bytes = %s\n", nice);
2703 	zdb_nicenum(dd->dd_compressed_bytes, nice, sizeof (nice));
2704 	(void) printf("\t\tcompressed_bytes = %s\n", nice);
2705 	zdb_nicenum(dd->dd_uncompressed_bytes, nice, sizeof (nice));
2706 	(void) printf("\t\tuncompressed_bytes = %s\n", nice);
2707 	zdb_nicenum(dd->dd_quota, nice, sizeof (nice));
2708 	(void) printf("\t\tquota = %s\n", nice);
2709 	zdb_nicenum(dd->dd_reserved, nice, sizeof (nice));
2710 	(void) printf("\t\treserved = %s\n", nice);
2711 	(void) printf("\t\tprops_zapobj = %llu\n",
2712 	    (u_longlong_t)dd->dd_props_zapobj);
2713 	(void) printf("\t\tdeleg_zapobj = %llu\n",
2714 	    (u_longlong_t)dd->dd_deleg_zapobj);
2715 	(void) printf("\t\tflags = %llx\n",
2716 	    (u_longlong_t)dd->dd_flags);
2717 
2718 #define	DO(which) \
2719 	zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \
2720 	    sizeof (nice)); \
2721 	(void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
2722 	DO(HEAD);
2723 	DO(SNAP);
2724 	DO(CHILD);
2725 	DO(CHILD_RSRV);
2726 	DO(REFRSRV);
2727 #undef DO
2728 	(void) printf("\t\tclones = %llu\n",
2729 	    (u_longlong_t)dd->dd_clones);
2730 }
2731 
2732 static void
dump_dsl_dataset(objset_t * os,uint64_t object,void * data,size_t size)2733 dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
2734 {
2735 	(void) os, (void) object;
2736 	dsl_dataset_phys_t *ds = data;
2737 	time_t crtime;
2738 	char used[32], compressed[32], uncompressed[32], unique[32];
2739 	char blkbuf[BP_SPRINTF_LEN];
2740 
2741 	/* make sure nicenum has enough space */
2742 	_Static_assert(sizeof (used) >= NN_NUMBUF_SZ, "used truncated");
2743 	_Static_assert(sizeof (compressed) >= NN_NUMBUF_SZ,
2744 	    "compressed truncated");
2745 	_Static_assert(sizeof (uncompressed) >= NN_NUMBUF_SZ,
2746 	    "uncompressed truncated");
2747 	_Static_assert(sizeof (unique) >= NN_NUMBUF_SZ, "unique truncated");
2748 
2749 	if (ds == NULL)
2750 		return;
2751 
2752 	ASSERT(size == sizeof (*ds));
2753 	crtime = ds->ds_creation_time;
2754 	zdb_nicenum(ds->ds_referenced_bytes, used, sizeof (used));
2755 	zdb_nicenum(ds->ds_compressed_bytes, compressed, sizeof (compressed));
2756 	zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed,
2757 	    sizeof (uncompressed));
2758 	zdb_nicenum(ds->ds_unique_bytes, unique, sizeof (unique));
2759 	snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp);
2760 
2761 	(void) printf("\t\tdir_obj = %llu\n",
2762 	    (u_longlong_t)ds->ds_dir_obj);
2763 	(void) printf("\t\tprev_snap_obj = %llu\n",
2764 	    (u_longlong_t)ds->ds_prev_snap_obj);
2765 	(void) printf("\t\tprev_snap_txg = %llu\n",
2766 	    (u_longlong_t)ds->ds_prev_snap_txg);
2767 	(void) printf("\t\tnext_snap_obj = %llu\n",
2768 	    (u_longlong_t)ds->ds_next_snap_obj);
2769 	(void) printf("\t\tsnapnames_zapobj = %llu\n",
2770 	    (u_longlong_t)ds->ds_snapnames_zapobj);
2771 	(void) printf("\t\tnum_children = %llu\n",
2772 	    (u_longlong_t)ds->ds_num_children);
2773 	(void) printf("\t\tuserrefs_obj = %llu\n",
2774 	    (u_longlong_t)ds->ds_userrefs_obj);
2775 	(void) printf("\t\tcreation_time = %s", ctime(&crtime));
2776 	(void) printf("\t\tcreation_txg = %llu\n",
2777 	    (u_longlong_t)ds->ds_creation_txg);
2778 	(void) printf("\t\tdeadlist_obj = %llu\n",
2779 	    (u_longlong_t)ds->ds_deadlist_obj);
2780 	(void) printf("\t\tused_bytes = %s\n", used);
2781 	(void) printf("\t\tcompressed_bytes = %s\n", compressed);
2782 	(void) printf("\t\tuncompressed_bytes = %s\n", uncompressed);
2783 	(void) printf("\t\tunique = %s\n", unique);
2784 	(void) printf("\t\tfsid_guid = %llu\n",
2785 	    (u_longlong_t)ds->ds_fsid_guid);
2786 	(void) printf("\t\tguid = %llu\n",
2787 	    (u_longlong_t)ds->ds_guid);
2788 	(void) printf("\t\tflags = %llx\n",
2789 	    (u_longlong_t)ds->ds_flags);
2790 	(void) printf("\t\tnext_clones_obj = %llu\n",
2791 	    (u_longlong_t)ds->ds_next_clones_obj);
2792 	(void) printf("\t\tprops_obj = %llu\n",
2793 	    (u_longlong_t)ds->ds_props_obj);
2794 	(void) printf("\t\tbp = %s\n", blkbuf);
2795 }
2796 
2797 static int
dump_bptree_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)2798 dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
2799 {
2800 	(void) arg, (void) tx;
2801 	char blkbuf[BP_SPRINTF_LEN];
2802 
2803 	if (BP_GET_LOGICAL_BIRTH(bp) != 0) {
2804 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
2805 		(void) printf("\t%s\n", blkbuf);
2806 	}
2807 	return (0);
2808 }
2809 
2810 static void
dump_bptree(objset_t * os,uint64_t obj,const char * name)2811 dump_bptree(objset_t *os, uint64_t obj, const char *name)
2812 {
2813 	char bytes[32];
2814 	bptree_phys_t *bt;
2815 	dmu_buf_t *db;
2816 
2817 	/* make sure nicenum has enough space */
2818 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2819 
2820 	if (dump_opt['d'] < 3)
2821 		return;
2822 
2823 	VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
2824 	bt = db->db_data;
2825 	zdb_nicenum(bt->bt_bytes, bytes, sizeof (bytes));
2826 	(void) printf("\n    %s: %llu datasets, %s\n",
2827 	    name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes);
2828 	dmu_buf_rele(db, FTAG);
2829 
2830 	if (dump_opt['d'] < 5)
2831 		return;
2832 
2833 	(void) printf("\n");
2834 
2835 	(void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL);
2836 }
2837 
2838 static int
dump_bpobj_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)2839 dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
2840 {
2841 	(void) arg, (void) tx;
2842 	char blkbuf[BP_SPRINTF_LEN];
2843 
2844 	ASSERT(BP_GET_LOGICAL_BIRTH(bp) != 0);
2845 	snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed);
2846 	(void) printf("\t%s\n", blkbuf);
2847 	return (0);
2848 }
2849 
2850 static void
dump_full_bpobj(bpobj_t * bpo,const char * name,int indent)2851 dump_full_bpobj(bpobj_t *bpo, const char *name, int indent)
2852 {
2853 	char bytes[32];
2854 	char comp[32];
2855 	char uncomp[32];
2856 	uint64_t i;
2857 
2858 	/* make sure nicenum has enough space */
2859 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
2860 	_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
2861 	_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
2862 
2863 	if (dump_opt['d'] < 3)
2864 		return;
2865 
2866 	zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes, sizeof (bytes));
2867 	if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
2868 		zdb_nicenum(bpo->bpo_phys->bpo_comp, comp, sizeof (comp));
2869 		zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp, sizeof (uncomp));
2870 		if (bpo->bpo_havefreed) {
2871 			(void) printf("    %*s: object %llu, %llu local "
2872 			    "blkptrs, %llu freed, %llu subobjs in object %llu, "
2873 			    "%s (%s/%s comp)\n",
2874 			    indent * 8, name,
2875 			    (u_longlong_t)bpo->bpo_object,
2876 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2877 			    (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2878 			    (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2879 			    (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2880 			    bytes, comp, uncomp);
2881 		} else {
2882 			(void) printf("    %*s: object %llu, %llu local "
2883 			    "blkptrs, %llu subobjs in object %llu, "
2884 			    "%s (%s/%s comp)\n",
2885 			    indent * 8, name,
2886 			    (u_longlong_t)bpo->bpo_object,
2887 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2888 			    (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
2889 			    (u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2890 			    bytes, comp, uncomp);
2891 		}
2892 
2893 		for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
2894 			uint64_t subobj;
2895 			bpobj_t subbpo;
2896 			int error;
2897 			VERIFY0(dmu_read(bpo->bpo_os,
2898 			    bpo->bpo_phys->bpo_subobjs,
2899 			    i * sizeof (subobj), sizeof (subobj), &subobj, 0));
2900 			error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
2901 			if (error != 0) {
2902 				(void) printf("ERROR %u while trying to open "
2903 				    "subobj id %llu\n",
2904 				    error, (u_longlong_t)subobj);
2905 				continue;
2906 			}
2907 			dump_full_bpobj(&subbpo, "subobj", indent + 1);
2908 			bpobj_close(&subbpo);
2909 		}
2910 	} else {
2911 		if (bpo->bpo_havefreed) {
2912 			(void) printf("    %*s: object %llu, %llu blkptrs, "
2913 			    "%llu freed, %s\n",
2914 			    indent * 8, name,
2915 			    (u_longlong_t)bpo->bpo_object,
2916 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2917 			    (u_longlong_t)bpo->bpo_phys->bpo_num_freed,
2918 			    bytes);
2919 		} else {
2920 			(void) printf("    %*s: object %llu, %llu blkptrs, "
2921 			    "%s\n",
2922 			    indent * 8, name,
2923 			    (u_longlong_t)bpo->bpo_object,
2924 			    (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
2925 			    bytes);
2926 		}
2927 	}
2928 
2929 	if (dump_opt['d'] < 5)
2930 		return;
2931 
2932 
2933 	if (indent == 0) {
2934 		(void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL);
2935 		(void) printf("\n");
2936 	}
2937 }
2938 
2939 static int
dump_bookmark(dsl_pool_t * dp,char * name,boolean_t print_redact,boolean_t print_list)2940 dump_bookmark(dsl_pool_t *dp, char *name, boolean_t print_redact,
2941     boolean_t print_list)
2942 {
2943 	int err = 0;
2944 	zfs_bookmark_phys_t prop;
2945 	objset_t *mos = dp->dp_spa->spa_meta_objset;
2946 	err = dsl_bookmark_lookup(dp, name, NULL, &prop);
2947 
2948 	if (err != 0) {
2949 		return (err);
2950 	}
2951 
2952 	(void) printf("\t#%s: ", strchr(name, '#') + 1);
2953 	(void) printf("{guid: %llx creation_txg: %llu creation_time: "
2954 	    "%llu redaction_obj: %llu}\n", (u_longlong_t)prop.zbm_guid,
2955 	    (u_longlong_t)prop.zbm_creation_txg,
2956 	    (u_longlong_t)prop.zbm_creation_time,
2957 	    (u_longlong_t)prop.zbm_redaction_obj);
2958 
2959 	IMPLY(print_list, print_redact);
2960 	if (!print_redact || prop.zbm_redaction_obj == 0)
2961 		return (0);
2962 
2963 	redaction_list_t *rl;
2964 	VERIFY0(dsl_redaction_list_hold_obj(dp,
2965 	    prop.zbm_redaction_obj, FTAG, &rl));
2966 
2967 	redaction_list_phys_t *rlp = rl->rl_phys;
2968 	(void) printf("\tRedacted:\n\t\tProgress: ");
2969 	if (rlp->rlp_last_object != UINT64_MAX ||
2970 	    rlp->rlp_last_blkid != UINT64_MAX) {
2971 		(void) printf("%llu %llu (incomplete)\n",
2972 		    (u_longlong_t)rlp->rlp_last_object,
2973 		    (u_longlong_t)rlp->rlp_last_blkid);
2974 	} else {
2975 		(void) printf("complete\n");
2976 	}
2977 	(void) printf("\t\tSnapshots: [");
2978 	for (unsigned int i = 0; i < rlp->rlp_num_snaps; i++) {
2979 		if (i > 0)
2980 			(void) printf(", ");
2981 		(void) printf("%0llu",
2982 		    (u_longlong_t)rlp->rlp_snaps[i]);
2983 	}
2984 	(void) printf("]\n\t\tLength: %llu\n",
2985 	    (u_longlong_t)rlp->rlp_num_entries);
2986 
2987 	if (!print_list) {
2988 		dsl_redaction_list_rele(rl, FTAG);
2989 		return (0);
2990 	}
2991 
2992 	if (rlp->rlp_num_entries == 0) {
2993 		dsl_redaction_list_rele(rl, FTAG);
2994 		(void) printf("\t\tRedaction List: []\n\n");
2995 		return (0);
2996 	}
2997 
2998 	redact_block_phys_t *rbp_buf;
2999 	uint64_t size;
3000 	dmu_object_info_t doi;
3001 
3002 	VERIFY0(dmu_object_info(mos, prop.zbm_redaction_obj, &doi));
3003 	size = doi.doi_max_offset;
3004 	rbp_buf = kmem_alloc(size, KM_SLEEP);
3005 
3006 	err = dmu_read(mos, prop.zbm_redaction_obj, 0, size,
3007 	    rbp_buf, 0);
3008 	if (err != 0) {
3009 		dsl_redaction_list_rele(rl, FTAG);
3010 		kmem_free(rbp_buf, size);
3011 		return (err);
3012 	}
3013 
3014 	(void) printf("\t\tRedaction List: [{object: %llx, offset: "
3015 	    "%llx, blksz: %x, count: %llx}",
3016 	    (u_longlong_t)rbp_buf[0].rbp_object,
3017 	    (u_longlong_t)rbp_buf[0].rbp_blkid,
3018 	    (uint_t)(redact_block_get_size(&rbp_buf[0])),
3019 	    (u_longlong_t)redact_block_get_count(&rbp_buf[0]));
3020 
3021 	for (size_t i = 1; i < rlp->rlp_num_entries; i++) {
3022 		(void) printf(",\n\t\t{object: %llx, offset: %llx, "
3023 		    "blksz: %x, count: %llx}",
3024 		    (u_longlong_t)rbp_buf[i].rbp_object,
3025 		    (u_longlong_t)rbp_buf[i].rbp_blkid,
3026 		    (uint_t)(redact_block_get_size(&rbp_buf[i])),
3027 		    (u_longlong_t)redact_block_get_count(&rbp_buf[i]));
3028 	}
3029 	dsl_redaction_list_rele(rl, FTAG);
3030 	kmem_free(rbp_buf, size);
3031 	(void) printf("]\n\n");
3032 	return (0);
3033 }
3034 
3035 static void
dump_bookmarks(objset_t * os,int verbosity)3036 dump_bookmarks(objset_t *os, int verbosity)
3037 {
3038 	zap_cursor_t zc;
3039 	zap_attribute_t *attrp;
3040 	dsl_dataset_t *ds = dmu_objset_ds(os);
3041 	dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3042 	objset_t *mos = os->os_spa->spa_meta_objset;
3043 	if (verbosity < 4)
3044 		return;
3045 	attrp = zap_attribute_alloc();
3046 	dsl_pool_config_enter(dp, FTAG);
3047 
3048 	for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj);
3049 	    zap_cursor_retrieve(&zc, attrp) == 0;
3050 	    zap_cursor_advance(&zc)) {
3051 		char osname[ZFS_MAX_DATASET_NAME_LEN];
3052 		char buf[ZFS_MAX_DATASET_NAME_LEN];
3053 		int len;
3054 		dmu_objset_name(os, osname);
3055 		len = snprintf(buf, sizeof (buf), "%s#%s", osname,
3056 		    attrp->za_name);
3057 		VERIFY3S(len, <, ZFS_MAX_DATASET_NAME_LEN);
3058 		(void) dump_bookmark(dp, buf, verbosity >= 5, verbosity >= 6);
3059 	}
3060 	zap_cursor_fini(&zc);
3061 	dsl_pool_config_exit(dp, FTAG);
3062 	zap_attribute_free(attrp);
3063 }
3064 
3065 static void
bpobj_count_refd(bpobj_t * bpo)3066 bpobj_count_refd(bpobj_t *bpo)
3067 {
3068 	mos_obj_refd(bpo->bpo_object);
3069 
3070 	if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
3071 		mos_obj_refd(bpo->bpo_phys->bpo_subobjs);
3072 		for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
3073 			uint64_t subobj;
3074 			bpobj_t subbpo;
3075 			int error;
3076 			VERIFY0(dmu_read(bpo->bpo_os,
3077 			    bpo->bpo_phys->bpo_subobjs,
3078 			    i * sizeof (subobj), sizeof (subobj), &subobj, 0));
3079 			error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
3080 			if (error != 0) {
3081 				(void) printf("ERROR %u while trying to open "
3082 				    "subobj id %llu\n",
3083 				    error, (u_longlong_t)subobj);
3084 				continue;
3085 			}
3086 			bpobj_count_refd(&subbpo);
3087 			bpobj_close(&subbpo);
3088 		}
3089 	}
3090 }
3091 
3092 static int
dsl_deadlist_entry_count_refd(void * arg,dsl_deadlist_entry_t * dle)3093 dsl_deadlist_entry_count_refd(void *arg, dsl_deadlist_entry_t *dle)
3094 {
3095 	spa_t *spa = arg;
3096 	uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3097 	if (dle->dle_bpobj.bpo_object != empty_bpobj)
3098 		bpobj_count_refd(&dle->dle_bpobj);
3099 	return (0);
3100 }
3101 
3102 static int
dsl_deadlist_entry_dump(void * arg,dsl_deadlist_entry_t * dle)3103 dsl_deadlist_entry_dump(void *arg, dsl_deadlist_entry_t *dle)
3104 {
3105 	ASSERT(arg == NULL);
3106 	if (dump_opt['d'] >= 5) {
3107 		char buf[128];
3108 		(void) snprintf(buf, sizeof (buf),
3109 		    "mintxg %llu -> obj %llu",
3110 		    (longlong_t)dle->dle_mintxg,
3111 		    (longlong_t)dle->dle_bpobj.bpo_object);
3112 
3113 		dump_full_bpobj(&dle->dle_bpobj, buf, 0);
3114 	} else {
3115 		(void) printf("mintxg %llu -> obj %llu\n",
3116 		    (longlong_t)dle->dle_mintxg,
3117 		    (longlong_t)dle->dle_bpobj.bpo_object);
3118 	}
3119 	return (0);
3120 }
3121 
3122 static void
dump_blkptr_list(dsl_deadlist_t * dl,const char * name)3123 dump_blkptr_list(dsl_deadlist_t *dl, const char *name)
3124 {
3125 	char bytes[32];
3126 	char comp[32];
3127 	char uncomp[32];
3128 	char entries[32];
3129 	spa_t *spa = dmu_objset_spa(dl->dl_os);
3130 	uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj;
3131 
3132 	if (dl->dl_oldfmt) {
3133 		if (dl->dl_bpobj.bpo_object != empty_bpobj)
3134 			bpobj_count_refd(&dl->dl_bpobj);
3135 	} else {
3136 		mos_obj_refd(dl->dl_object);
3137 		dsl_deadlist_iterate(dl, dsl_deadlist_entry_count_refd, spa);
3138 	}
3139 
3140 	/* make sure nicenum has enough space */
3141 	_Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated");
3142 	_Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated");
3143 	_Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated");
3144 	_Static_assert(sizeof (entries) >= NN_NUMBUF_SZ, "entries truncated");
3145 
3146 	if (dump_opt['d'] < 3)
3147 		return;
3148 
3149 	if (dl->dl_oldfmt) {
3150 		dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0);
3151 		return;
3152 	}
3153 
3154 	zdb_nicenum(dl->dl_phys->dl_used, bytes, sizeof (bytes));
3155 	zdb_nicenum(dl->dl_phys->dl_comp, comp, sizeof (comp));
3156 	zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp, sizeof (uncomp));
3157 	zdb_nicenum(avl_numnodes(&dl->dl_tree), entries, sizeof (entries));
3158 	(void) printf("\n    %s: %s (%s/%s comp), %s entries\n",
3159 	    name, bytes, comp, uncomp, entries);
3160 
3161 	if (dump_opt['d'] < 4)
3162 		return;
3163 
3164 	(void) putchar('\n');
3165 
3166 	dsl_deadlist_iterate(dl, dsl_deadlist_entry_dump, NULL);
3167 }
3168 
3169 static int
verify_dd_livelist(objset_t * os)3170 verify_dd_livelist(objset_t *os)
3171 {
3172 	uint64_t ll_used, used, ll_comp, comp, ll_uncomp, uncomp;
3173 	dsl_pool_t *dp = spa_get_dsl(os->os_spa);
3174 	dsl_dir_t  *dd = os->os_dsl_dataset->ds_dir;
3175 
3176 	ASSERT(!dmu_objset_is_snapshot(os));
3177 	if (!dsl_deadlist_is_open(&dd->dd_livelist))
3178 		return (0);
3179 
3180 	/* Iterate through the livelist to check for duplicates */
3181 	dsl_deadlist_iterate(&dd->dd_livelist, sublivelist_verify_lightweight,
3182 	    NULL);
3183 
3184 	dsl_pool_config_enter(dp, FTAG);
3185 	dsl_deadlist_space(&dd->dd_livelist, &ll_used,
3186 	    &ll_comp, &ll_uncomp);
3187 
3188 	dsl_dataset_t *origin_ds;
3189 	ASSERT(dsl_pool_config_held(dp));
3190 	VERIFY0(dsl_dataset_hold_obj(dp,
3191 	    dsl_dir_phys(dd)->dd_origin_obj, FTAG, &origin_ds));
3192 	VERIFY0(dsl_dataset_space_written(origin_ds, os->os_dsl_dataset,
3193 	    &used, &comp, &uncomp));
3194 	dsl_dataset_rele(origin_ds, FTAG);
3195 	dsl_pool_config_exit(dp, FTAG);
3196 	/*
3197 	 *  It's possible that the dataset's uncomp space is larger than the
3198 	 *  livelist's because livelists do not track embedded block pointers
3199 	 */
3200 	if (used != ll_used || comp != ll_comp || uncomp < ll_uncomp) {
3201 		char nice_used[32], nice_comp[32], nice_uncomp[32];
3202 		(void) printf("Discrepancy in space accounting:\n");
3203 		zdb_nicenum(used, nice_used, sizeof (nice_used));
3204 		zdb_nicenum(comp, nice_comp, sizeof (nice_comp));
3205 		zdb_nicenum(uncomp, nice_uncomp, sizeof (nice_uncomp));
3206 		(void) printf("dir: used %s, comp %s, uncomp %s\n",
3207 		    nice_used, nice_comp, nice_uncomp);
3208 		zdb_nicenum(ll_used, nice_used, sizeof (nice_used));
3209 		zdb_nicenum(ll_comp, nice_comp, sizeof (nice_comp));
3210 		zdb_nicenum(ll_uncomp, nice_uncomp, sizeof (nice_uncomp));
3211 		(void) printf("livelist: used %s, comp %s, uncomp %s\n",
3212 		    nice_used, nice_comp, nice_uncomp);
3213 		return (1);
3214 	}
3215 	return (0);
3216 }
3217 
3218 static char *key_material = NULL;
3219 
3220 static boolean_t
zdb_derive_key(dsl_dir_t * dd,uint8_t * key_out)3221 zdb_derive_key(dsl_dir_t *dd, uint8_t *key_out)
3222 {
3223 	uint64_t keyformat, salt, iters;
3224 	int i;
3225 	unsigned char c;
3226 
3227 	VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3228 	    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), sizeof (uint64_t),
3229 	    1, &keyformat));
3230 
3231 	switch (keyformat) {
3232 	case ZFS_KEYFORMAT_HEX:
3233 		for (i = 0; i < WRAPPING_KEY_LEN * 2; i += 2) {
3234 			if (!isxdigit(key_material[i]) ||
3235 			    !isxdigit(key_material[i+1]))
3236 				return (B_FALSE);
3237 			if (sscanf(&key_material[i], "%02hhx", &c) != 1)
3238 				return (B_FALSE);
3239 			key_out[i / 2] = c;
3240 		}
3241 		break;
3242 
3243 	case ZFS_KEYFORMAT_PASSPHRASE:
3244 		VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3245 		    dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT),
3246 		    sizeof (uint64_t), 1, &salt));
3247 		VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset,
3248 		    dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS),
3249 		    sizeof (uint64_t), 1, &iters));
3250 
3251 		if (PKCS5_PBKDF2_HMAC_SHA1(key_material, strlen(key_material),
3252 		    ((uint8_t *)&salt), sizeof (uint64_t), iters,
3253 		    WRAPPING_KEY_LEN, key_out) != 1)
3254 			return (B_FALSE);
3255 
3256 		break;
3257 
3258 	default:
3259 		fatal("no support for key format %u\n",
3260 		    (unsigned int) keyformat);
3261 	}
3262 
3263 	return (B_TRUE);
3264 }
3265 
3266 static char encroot[ZFS_MAX_DATASET_NAME_LEN];
3267 static boolean_t key_loaded = B_FALSE;
3268 
3269 static void
zdb_load_key(objset_t * os)3270 zdb_load_key(objset_t *os)
3271 {
3272 	dsl_pool_t *dp;
3273 	dsl_dir_t *dd, *rdd;
3274 	uint8_t key[WRAPPING_KEY_LEN];
3275 	uint64_t rddobj;
3276 	int err;
3277 
3278 	dp = spa_get_dsl(os->os_spa);
3279 	dd = os->os_dsl_dataset->ds_dir;
3280 
3281 	dsl_pool_config_enter(dp, FTAG);
3282 	VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj,
3283 	    DSL_CRYPTO_KEY_ROOT_DDOBJ, sizeof (uint64_t), 1, &rddobj));
3284 	VERIFY0(dsl_dir_hold_obj(dd->dd_pool, rddobj, NULL, FTAG, &rdd));
3285 	dsl_dir_name(rdd, encroot);
3286 	dsl_dir_rele(rdd, FTAG);
3287 
3288 	if (!zdb_derive_key(dd, key))
3289 		fatal("couldn't derive encryption key");
3290 
3291 	dsl_pool_config_exit(dp, FTAG);
3292 
3293 	ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_UNAVAILABLE);
3294 
3295 	dsl_crypto_params_t *dcp;
3296 	nvlist_t *crypto_args;
3297 
3298 	crypto_args = fnvlist_alloc();
3299 	fnvlist_add_uint8_array(crypto_args, "wkeydata",
3300 	    (uint8_t *)key, WRAPPING_KEY_LEN);
3301 	VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE,
3302 	    NULL, crypto_args, &dcp));
3303 	err = spa_keystore_load_wkey(encroot, dcp, B_FALSE);
3304 
3305 	dsl_crypto_params_free(dcp, (err != 0));
3306 	fnvlist_free(crypto_args);
3307 
3308 	if (err != 0)
3309 		fatal(
3310 		    "couldn't load encryption key for %s: %s",
3311 		    encroot, err == ZFS_ERR_CRYPTO_NOTSUP ?
3312 		    "crypto params not supported" : strerror(err));
3313 
3314 	ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_AVAILABLE);
3315 
3316 	printf("Unlocked encryption root: %s\n", encroot);
3317 	key_loaded = B_TRUE;
3318 }
3319 
3320 static void
zdb_unload_key(void)3321 zdb_unload_key(void)
3322 {
3323 	if (!key_loaded)
3324 		return;
3325 
3326 	VERIFY0(spa_keystore_unload_wkey(encroot));
3327 	key_loaded = B_FALSE;
3328 }
3329 
3330 static avl_tree_t idx_tree;
3331 static avl_tree_t domain_tree;
3332 static boolean_t fuid_table_loaded;
3333 static objset_t *sa_os = NULL;
3334 static sa_attr_type_t *sa_attr_table = NULL;
3335 
3336 static int
open_objset(const char * path,const void * tag,objset_t ** osp)3337 open_objset(const char *path, const void *tag, objset_t **osp)
3338 {
3339 	int err;
3340 	uint64_t sa_attrs = 0;
3341 	uint64_t version = 0;
3342 
3343 	VERIFY3P(sa_os, ==, NULL);
3344 
3345 	/*
3346 	 * We can't own an objset if it's redacted.  Therefore, we do this
3347 	 * dance: hold the objset, then acquire a long hold on its dataset, then
3348 	 * release the pool (which is held as part of holding the objset).
3349 	 */
3350 
3351 	if (dump_opt['K']) {
3352 		/* decryption requested, try to load keys */
3353 		err = dmu_objset_hold(path, tag, osp);
3354 		if (err != 0) {
3355 			(void) fprintf(stderr, "failed to hold dataset "
3356 			    "'%s': %s\n",
3357 			    path, strerror(err));
3358 			return (err);
3359 		}
3360 		dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3361 		dsl_pool_rele(dmu_objset_pool(*osp), tag);
3362 
3363 		/* succeeds or dies */
3364 		zdb_load_key(*osp);
3365 
3366 		/* release it all */
3367 		dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3368 		dsl_dataset_rele(dmu_objset_ds(*osp), tag);
3369 	}
3370 
3371 	int ds_hold_flags = key_loaded ? DS_HOLD_FLAG_DECRYPT : 0;
3372 
3373 	err = dmu_objset_hold_flags(path, ds_hold_flags, tag, osp);
3374 	if (err != 0) {
3375 		(void) fprintf(stderr, "failed to hold dataset '%s': %s\n",
3376 		    path, strerror(err));
3377 		return (err);
3378 	}
3379 	dsl_dataset_long_hold(dmu_objset_ds(*osp), tag);
3380 	dsl_pool_rele(dmu_objset_pool(*osp), tag);
3381 
3382 	if (dmu_objset_type(*osp) == DMU_OST_ZFS &&
3383 	    (key_loaded || !(*osp)->os_encrypted)) {
3384 		(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR,
3385 		    8, 1, &version);
3386 		if (version >= ZPL_VERSION_SA) {
3387 			(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS,
3388 			    8, 1, &sa_attrs);
3389 		}
3390 		err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END,
3391 		    &sa_attr_table);
3392 		if (err != 0) {
3393 			(void) fprintf(stderr, "sa_setup failed: %s\n",
3394 			    strerror(err));
3395 			dsl_dataset_long_rele(dmu_objset_ds(*osp), tag);
3396 			dsl_dataset_rele_flags(dmu_objset_ds(*osp),
3397 			    ds_hold_flags, tag);
3398 			*osp = NULL;
3399 		}
3400 	}
3401 	sa_os = *osp;
3402 
3403 	return (err);
3404 }
3405 
3406 static void
close_objset(objset_t * os,const void * tag)3407 close_objset(objset_t *os, const void *tag)
3408 {
3409 	VERIFY3P(os, ==, sa_os);
3410 	if (os->os_sa != NULL)
3411 		sa_tear_down(os);
3412 	dsl_dataset_long_rele(dmu_objset_ds(os), tag);
3413 	dsl_dataset_rele_flags(dmu_objset_ds(os),
3414 	    key_loaded ? DS_HOLD_FLAG_DECRYPT : 0, tag);
3415 	sa_attr_table = NULL;
3416 	sa_os = NULL;
3417 
3418 	zdb_unload_key();
3419 }
3420 
3421 static void
fuid_table_destroy(void)3422 fuid_table_destroy(void)
3423 {
3424 	if (fuid_table_loaded) {
3425 		zfs_fuid_table_destroy(&idx_tree, &domain_tree);
3426 		fuid_table_loaded = B_FALSE;
3427 	}
3428 }
3429 
3430 /*
3431  * Clean up DDT internal state. ddt_lookup() adds entries to ddt_tree, which on
3432  * a live pool are normally cleaned up during ddt_sync(). We can't do that (and
3433  * wouldn't want to anyway), but if we don't clean up the presence of stuff on
3434  * ddt_tree will trip asserts in ddt_table_free(). So, we clean up ourselves.
3435  *
3436  * Note that this is not a particularly efficient way to do this, but
3437  * ddt_remove() is the only public method that can do the work we need, and it
3438  * requires the right locks and etc to do the job. This is only ever called
3439  * during zdb shutdown so efficiency is not especially important.
3440  */
3441 static void
zdb_ddt_cleanup(spa_t * spa)3442 zdb_ddt_cleanup(spa_t *spa)
3443 {
3444 	for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
3445 		ddt_t *ddt = spa->spa_ddt[c];
3446 		if (!ddt)
3447 			continue;
3448 
3449 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3450 		ddt_enter(ddt);
3451 		ddt_entry_t *dde = avl_first(&ddt->ddt_tree), *next;
3452 		while (dde) {
3453 			next = AVL_NEXT(&ddt->ddt_tree, dde);
3454 			dde->dde_io = NULL;
3455 			ddt_remove(ddt, dde);
3456 			dde = next;
3457 		}
3458 		ddt_exit(ddt);
3459 		spa_config_exit(spa, SCL_CONFIG, FTAG);
3460 	}
3461 }
3462 
3463 static void
zdb_exit(int reason)3464 zdb_exit(int reason)
3465 {
3466 	if (spa != NULL)
3467 		zdb_ddt_cleanup(spa);
3468 
3469 	if (os != NULL) {
3470 		close_objset(os, FTAG);
3471 	} else if (spa != NULL) {
3472 		spa_close(spa, FTAG);
3473 	}
3474 
3475 	fuid_table_destroy();
3476 
3477 	if (kernel_init_done)
3478 		kernel_fini();
3479 
3480 	exit(reason);
3481 }
3482 
3483 /*
3484  * print uid or gid information.
3485  * For normal POSIX id just the id is printed in decimal format.
3486  * For CIFS files with FUID the fuid is printed in hex followed by
3487  * the domain-rid string.
3488  */
3489 static void
print_idstr(uint64_t id,const char * id_type)3490 print_idstr(uint64_t id, const char *id_type)
3491 {
3492 	if (FUID_INDEX(id)) {
3493 		const char *domain =
3494 		    zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id));
3495 		(void) printf("\t%s     %llx [%s-%d]\n", id_type,
3496 		    (u_longlong_t)id, domain, (int)FUID_RID(id));
3497 	} else {
3498 		(void) printf("\t%s     %llu\n", id_type, (u_longlong_t)id);
3499 	}
3500 
3501 }
3502 
3503 static void
dump_uidgid(objset_t * os,uint64_t uid,uint64_t gid)3504 dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid)
3505 {
3506 	uint32_t uid_idx, gid_idx;
3507 
3508 	uid_idx = FUID_INDEX(uid);
3509 	gid_idx = FUID_INDEX(gid);
3510 
3511 	/* Load domain table, if not already loaded */
3512 	if (!fuid_table_loaded && (uid_idx || gid_idx)) {
3513 		uint64_t fuid_obj;
3514 
3515 		/* first find the fuid object.  It lives in the master node */
3516 		VERIFY(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES,
3517 		    8, 1, &fuid_obj) == 0);
3518 		zfs_fuid_avl_tree_create(&idx_tree, &domain_tree);
3519 		(void) zfs_fuid_table_load(os, fuid_obj,
3520 		    &idx_tree, &domain_tree);
3521 		fuid_table_loaded = B_TRUE;
3522 	}
3523 
3524 	print_idstr(uid, "uid");
3525 	print_idstr(gid, "gid");
3526 }
3527 
3528 static void
dump_znode_sa_xattr(sa_handle_t * hdl)3529 dump_znode_sa_xattr(sa_handle_t *hdl)
3530 {
3531 	nvlist_t *sa_xattr;
3532 	nvpair_t *elem = NULL;
3533 	int sa_xattr_size = 0;
3534 	int sa_xattr_entries = 0;
3535 	int error;
3536 	char *sa_xattr_packed;
3537 
3538 	error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size);
3539 	if (error || sa_xattr_size == 0)
3540 		return;
3541 
3542 	sa_xattr_packed = malloc(sa_xattr_size);
3543 	if (sa_xattr_packed == NULL)
3544 		return;
3545 
3546 	error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR],
3547 	    sa_xattr_packed, sa_xattr_size);
3548 	if (error) {
3549 		free(sa_xattr_packed);
3550 		return;
3551 	}
3552 
3553 	error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0);
3554 	if (error) {
3555 		free(sa_xattr_packed);
3556 		return;
3557 	}
3558 
3559 	while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL)
3560 		sa_xattr_entries++;
3561 
3562 	(void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
3563 	    sa_xattr_size, sa_xattr_entries);
3564 	while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) {
3565 		boolean_t can_print = !dump_opt['P'];
3566 		uchar_t *value;
3567 		uint_t cnt, idx;
3568 
3569 		(void) printf("\t\t%s = ", nvpair_name(elem));
3570 		nvpair_value_byte_array(elem, &value, &cnt);
3571 
3572 		for (idx = 0; idx < cnt; ++idx) {
3573 			if (!isprint(value[idx])) {
3574 				can_print = B_FALSE;
3575 				break;
3576 			}
3577 		}
3578 
3579 		for (idx = 0; idx < cnt; ++idx) {
3580 			if (can_print)
3581 				(void) putchar(value[idx]);
3582 			else
3583 				(void) printf("\\%3.3o", value[idx]);
3584 		}
3585 		(void) putchar('\n');
3586 	}
3587 
3588 	nvlist_free(sa_xattr);
3589 	free(sa_xattr_packed);
3590 }
3591 
3592 static void
dump_znode_symlink(sa_handle_t * hdl)3593 dump_znode_symlink(sa_handle_t *hdl)
3594 {
3595 	int sa_symlink_size = 0;
3596 	char linktarget[MAXPATHLEN];
3597 	int error;
3598 
3599 	error = sa_size(hdl, sa_attr_table[ZPL_SYMLINK], &sa_symlink_size);
3600 	if (error || sa_symlink_size == 0) {
3601 		return;
3602 	}
3603 	if (sa_symlink_size >= sizeof (linktarget)) {
3604 		(void) printf("symlink size %d is too large\n",
3605 		    sa_symlink_size);
3606 		return;
3607 	}
3608 	linktarget[sa_symlink_size] = '\0';
3609 	if (sa_lookup(hdl, sa_attr_table[ZPL_SYMLINK],
3610 	    &linktarget, sa_symlink_size) == 0)
3611 		(void) printf("\ttarget	%s\n", linktarget);
3612 }
3613 
3614 static void
dump_znode(objset_t * os,uint64_t object,void * data,size_t size)3615 dump_znode(objset_t *os, uint64_t object, void *data, size_t size)
3616 {
3617 	(void) data, (void) size;
3618 	char path[MAXPATHLEN * 2];	/* allow for xattr and failure prefix */
3619 	sa_handle_t *hdl;
3620 	uint64_t xattr, rdev, gen;
3621 	uint64_t uid, gid, mode, fsize, parent, links;
3622 	uint64_t pflags;
3623 	uint64_t acctm[2], modtm[2], chgtm[2], crtm[2];
3624 	time_t z_crtime, z_atime, z_mtime, z_ctime;
3625 	sa_bulk_attr_t bulk[12];
3626 	int idx = 0;
3627 	int error;
3628 
3629 	VERIFY3P(os, ==, sa_os);
3630 	if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) {
3631 		(void) printf("Failed to get handle for SA znode\n");
3632 		return;
3633 	}
3634 
3635 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8);
3636 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8);
3637 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL,
3638 	    &links, 8);
3639 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8);
3640 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL,
3641 	    &mode, 8);
3642 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT],
3643 	    NULL, &parent, 8);
3644 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL,
3645 	    &fsize, 8);
3646 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL,
3647 	    acctm, 16);
3648 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL,
3649 	    modtm, 16);
3650 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL,
3651 	    crtm, 16);
3652 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL,
3653 	    chgtm, 16);
3654 	SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL,
3655 	    &pflags, 8);
3656 
3657 	if (sa_bulk_lookup(hdl, bulk, idx)) {
3658 		(void) sa_handle_destroy(hdl);
3659 		return;
3660 	}
3661 
3662 	z_crtime = (time_t)crtm[0];
3663 	z_atime = (time_t)acctm[0];
3664 	z_mtime = (time_t)modtm[0];
3665 	z_ctime = (time_t)chgtm[0];
3666 
3667 	if (dump_opt['d'] > 4) {
3668 		error = zfs_obj_to_path(os, object, path, sizeof (path));
3669 		if (error == ESTALE) {
3670 			(void) snprintf(path, sizeof (path), "on delete queue");
3671 		} else if (error != 0) {
3672 			leaked_objects++;
3673 			(void) snprintf(path, sizeof (path),
3674 			    "path not found, possibly leaked");
3675 		}
3676 		(void) printf("\tpath	%s\n", path);
3677 	}
3678 
3679 	if (S_ISLNK(mode))
3680 		dump_znode_symlink(hdl);
3681 	dump_uidgid(os, uid, gid);
3682 	(void) printf("\tatime	%s", ctime(&z_atime));
3683 	(void) printf("\tmtime	%s", ctime(&z_mtime));
3684 	(void) printf("\tctime	%s", ctime(&z_ctime));
3685 	(void) printf("\tcrtime	%s", ctime(&z_crtime));
3686 	(void) printf("\tgen	%llu\n", (u_longlong_t)gen);
3687 	(void) printf("\tmode	%llo\n", (u_longlong_t)mode);
3688 	(void) printf("\tsize	%llu\n", (u_longlong_t)fsize);
3689 	(void) printf("\tparent	%llu\n", (u_longlong_t)parent);
3690 	(void) printf("\tlinks	%llu\n", (u_longlong_t)links);
3691 	(void) printf("\tpflags	%llx\n", (u_longlong_t)pflags);
3692 	if (dmu_objset_projectquota_enabled(os) && (pflags & ZFS_PROJID)) {
3693 		uint64_t projid;
3694 
3695 		if (sa_lookup(hdl, sa_attr_table[ZPL_PROJID], &projid,
3696 		    sizeof (uint64_t)) == 0)
3697 			(void) printf("\tprojid	%llu\n", (u_longlong_t)projid);
3698 	}
3699 	if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr,
3700 	    sizeof (uint64_t)) == 0)
3701 		(void) printf("\txattr	%llu\n", (u_longlong_t)xattr);
3702 	if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev,
3703 	    sizeof (uint64_t)) == 0)
3704 		(void) printf("\trdev	0x%016llx\n", (u_longlong_t)rdev);
3705 	dump_znode_sa_xattr(hdl);
3706 	sa_handle_destroy(hdl);
3707 }
3708 
3709 static void
dump_acl(objset_t * os,uint64_t object,void * data,size_t size)3710 dump_acl(objset_t *os, uint64_t object, void *data, size_t size)
3711 {
3712 	(void) os, (void) object, (void) data, (void) size;
3713 }
3714 
3715 static void
dump_dmu_objset(objset_t * os,uint64_t object,void * data,size_t size)3716 dump_dmu_objset(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 object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = {
3722 	dump_none,		/* unallocated			*/
3723 	dump_zap,		/* object directory		*/
3724 	dump_uint64,		/* object array			*/
3725 	dump_none,		/* packed nvlist		*/
3726 	dump_packed_nvlist,	/* packed nvlist size		*/
3727 	dump_none,		/* bpobj			*/
3728 	dump_bpobj,		/* bpobj header			*/
3729 	dump_none,		/* SPA space map header		*/
3730 	dump_none,		/* SPA space map		*/
3731 	dump_none,		/* ZIL intent log		*/
3732 	dump_dnode,		/* DMU dnode			*/
3733 	dump_dmu_objset,	/* DMU objset			*/
3734 	dump_dsl_dir,		/* DSL directory		*/
3735 	dump_zap,		/* DSL directory child map	*/
3736 	dump_zap,		/* DSL dataset snap map		*/
3737 	dump_zap,		/* DSL props			*/
3738 	dump_dsl_dataset,	/* DSL dataset			*/
3739 	dump_znode,		/* ZFS znode			*/
3740 	dump_acl,		/* ZFS V0 ACL			*/
3741 	dump_uint8,		/* ZFS plain file		*/
3742 	dump_zpldir,		/* ZFS directory		*/
3743 	dump_zap,		/* ZFS master node		*/
3744 	dump_zap,		/* ZFS delete queue		*/
3745 	dump_uint8,		/* zvol object			*/
3746 	dump_zap,		/* zvol prop			*/
3747 	dump_uint8,		/* other uint8[]		*/
3748 	dump_uint64,		/* other uint64[]		*/
3749 	dump_zap,		/* other ZAP			*/
3750 	dump_zap,		/* persistent error log		*/
3751 	dump_uint8,		/* SPA history			*/
3752 	dump_history_offsets,	/* SPA history offsets		*/
3753 	dump_zap,		/* Pool properties		*/
3754 	dump_zap,		/* DSL permissions		*/
3755 	dump_acl,		/* ZFS ACL			*/
3756 	dump_uint8,		/* ZFS SYSACL			*/
3757 	dump_none,		/* FUID nvlist			*/
3758 	dump_packed_nvlist,	/* FUID nvlist size		*/
3759 	dump_zap,		/* DSL dataset next clones	*/
3760 	dump_zap,		/* DSL scrub queue		*/
3761 	dump_zap,		/* ZFS user/group/project used	*/
3762 	dump_zap,		/* ZFS user/group/project quota	*/
3763 	dump_zap,		/* snapshot refcount tags	*/
3764 	dump_ddt_zap,		/* DDT ZAP object		*/
3765 	dump_zap,		/* DDT statistics		*/
3766 	dump_znode,		/* SA object			*/
3767 	dump_zap,		/* SA Master Node		*/
3768 	dump_sa_attrs,		/* SA attribute registration	*/
3769 	dump_sa_layouts,	/* SA attribute layouts		*/
3770 	dump_zap,		/* DSL scrub translations	*/
3771 	dump_none,		/* fake dedup BP		*/
3772 	dump_zap,		/* deadlist			*/
3773 	dump_none,		/* deadlist hdr			*/
3774 	dump_zap,		/* dsl clones			*/
3775 	dump_bpobj_subobjs,	/* bpobj subobjs		*/
3776 	dump_unknown,		/* Unknown type, must be last	*/
3777 };
3778 
3779 static boolean_t
match_object_type(dmu_object_type_t obj_type,uint64_t flags)3780 match_object_type(dmu_object_type_t obj_type, uint64_t flags)
3781 {
3782 	boolean_t match = B_TRUE;
3783 
3784 	switch (obj_type) {
3785 	case DMU_OT_DIRECTORY_CONTENTS:
3786 		if (!(flags & ZOR_FLAG_DIRECTORY))
3787 			match = B_FALSE;
3788 		break;
3789 	case DMU_OT_PLAIN_FILE_CONTENTS:
3790 		if (!(flags & ZOR_FLAG_PLAIN_FILE))
3791 			match = B_FALSE;
3792 		break;
3793 	case DMU_OT_SPACE_MAP:
3794 		if (!(flags & ZOR_FLAG_SPACE_MAP))
3795 			match = B_FALSE;
3796 		break;
3797 	default:
3798 		if (strcmp(zdb_ot_name(obj_type), "zap") == 0) {
3799 			if (!(flags & ZOR_FLAG_ZAP))
3800 				match = B_FALSE;
3801 			break;
3802 		}
3803 
3804 		/*
3805 		 * If all bits except some of the supported flags are
3806 		 * set, the user combined the all-types flag (A) with
3807 		 * a negated flag to exclude some types (e.g. A-f to
3808 		 * show all object types except plain files).
3809 		 */
3810 		if ((flags | ZOR_SUPPORTED_FLAGS) != ZOR_FLAG_ALL_TYPES)
3811 			match = B_FALSE;
3812 
3813 		break;
3814 	}
3815 
3816 	return (match);
3817 }
3818 
3819 static void
dump_object(objset_t * os,uint64_t object,int verbosity,boolean_t * print_header,uint64_t * dnode_slots_used,uint64_t flags)3820 dump_object(objset_t *os, uint64_t object, int verbosity,
3821     boolean_t *print_header, uint64_t *dnode_slots_used, uint64_t flags)
3822 {
3823 	dmu_buf_t *db = NULL;
3824 	dmu_object_info_t doi;
3825 	dnode_t *dn;
3826 	boolean_t dnode_held = B_FALSE;
3827 	void *bonus = NULL;
3828 	size_t bsize = 0;
3829 	char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
3830 	char bonus_size[32];
3831 	char aux[50];
3832 	int error;
3833 
3834 	/* make sure nicenum has enough space */
3835 	_Static_assert(sizeof (iblk) >= NN_NUMBUF_SZ, "iblk truncated");
3836 	_Static_assert(sizeof (dblk) >= NN_NUMBUF_SZ, "dblk truncated");
3837 	_Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, "lsize truncated");
3838 	_Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, "asize truncated");
3839 	_Static_assert(sizeof (bonus_size) >= NN_NUMBUF_SZ,
3840 	    "bonus_size truncated");
3841 
3842 	if (*print_header) {
3843 		(void) printf("\n%10s  %3s  %5s  %5s  %5s  %6s  %5s  %6s  %s\n",
3844 		    "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
3845 		    "lsize", "%full", "type");
3846 		*print_header = 0;
3847 	}
3848 
3849 	if (object == 0) {
3850 		dn = DMU_META_DNODE(os);
3851 		dmu_object_info_from_dnode(dn, &doi);
3852 	} else {
3853 		/*
3854 		 * Encrypted datasets will have sensitive bonus buffers
3855 		 * encrypted. Therefore we cannot hold the bonus buffer and
3856 		 * must hold the dnode itself instead.
3857 		 */
3858 		error = dmu_object_info(os, object, &doi);
3859 		if (error)
3860 			fatal("dmu_object_info() failed, errno %u", error);
3861 
3862 		if (!key_loaded && os->os_encrypted &&
3863 		    DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) {
3864 			error = dnode_hold(os, object, FTAG, &dn);
3865 			if (error)
3866 				fatal("dnode_hold() failed, errno %u", error);
3867 			dnode_held = B_TRUE;
3868 		} else {
3869 			error = dmu_bonus_hold(os, object, FTAG, &db);
3870 			if (error)
3871 				fatal("dmu_bonus_hold(%llu) failed, errno %u",
3872 				    object, error);
3873 			bonus = db->db_data;
3874 			bsize = db->db_size;
3875 			dn = DB_DNODE((dmu_buf_impl_t *)db);
3876 		}
3877 	}
3878 
3879 	/*
3880 	 * Default to showing all object types if no flags were specified.
3881 	 */
3882 	if (flags != 0 && flags != ZOR_FLAG_ALL_TYPES &&
3883 	    !match_object_type(doi.doi_type, flags))
3884 		goto out;
3885 
3886 	if (dnode_slots_used)
3887 		*dnode_slots_used = doi.doi_dnodesize / DNODE_MIN_SIZE;
3888 
3889 	zdb_nicenum(doi.doi_metadata_block_size, iblk, sizeof (iblk));
3890 	zdb_nicenum(doi.doi_data_block_size, dblk, sizeof (dblk));
3891 	zdb_nicenum(doi.doi_max_offset, lsize, sizeof (lsize));
3892 	zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize, sizeof (asize));
3893 	zdb_nicenum(doi.doi_bonus_size, bonus_size, sizeof (bonus_size));
3894 	zdb_nicenum(doi.doi_dnodesize, dnsize, sizeof (dnsize));
3895 	(void) snprintf(fill, sizeof (fill), "%6.2f", 100.0 *
3896 	    doi.doi_fill_count * doi.doi_data_block_size / (object == 0 ?
3897 	    DNODES_PER_BLOCK : 1) / doi.doi_max_offset);
3898 
3899 	aux[0] = '\0';
3900 
3901 	if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) {
3902 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3903 		    " (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum));
3904 	}
3905 
3906 	if (doi.doi_compress == ZIO_COMPRESS_INHERIT &&
3907 	    ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) {
3908 		const char *compname = NULL;
3909 		if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION,
3910 		    ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel),
3911 		    &compname) == 0) {
3912 			(void) snprintf(aux + strlen(aux),
3913 			    sizeof (aux) - strlen(aux), " (Z=inherit=%s)",
3914 			    compname);
3915 		} else {
3916 			(void) snprintf(aux + strlen(aux),
3917 			    sizeof (aux) - strlen(aux),
3918 			    " (Z=inherit=%s-unknown)",
3919 			    ZDB_COMPRESS_NAME(os->os_compress));
3920 		}
3921 	} else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) {
3922 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3923 		    " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress));
3924 	} else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
3925 		(void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux),
3926 		    " (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress));
3927 	}
3928 
3929 	(void) printf("%10lld  %3u  %5s  %5s  %5s  %6s  %5s  %6s  %s%s\n",
3930 	    (u_longlong_t)object, doi.doi_indirection, iblk, dblk,
3931 	    asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux);
3932 
3933 	if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
3934 		(void) printf("%10s  %3s  %5s  %5s  %5s  %5s  %5s  %6s  %s\n",
3935 		    "", "", "", "", "", "", bonus_size, "bonus",
3936 		    zdb_ot_name(doi.doi_bonus_type));
3937 	}
3938 
3939 	if (verbosity >= 4) {
3940 		(void) printf("\tdnode flags: %s%s%s%s\n",
3941 		    (dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ?
3942 		    "USED_BYTES " : "",
3943 		    (dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ?
3944 		    "USERUSED_ACCOUNTED " : "",
3945 		    (dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ?
3946 		    "USEROBJUSED_ACCOUNTED " : "",
3947 		    (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ?
3948 		    "SPILL_BLKPTR" : "");
3949 		(void) printf("\tdnode maxblkid: %llu\n",
3950 		    (longlong_t)dn->dn_phys->dn_maxblkid);
3951 
3952 		if (!dnode_held) {
3953 			object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os,
3954 			    object, bonus, bsize);
3955 		} else {
3956 			(void) printf("\t\t(bonus encrypted)\n");
3957 		}
3958 
3959 		if (key_loaded ||
3960 		    (!os->os_encrypted || !DMU_OT_IS_ENCRYPTED(doi.doi_type))) {
3961 			object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object,
3962 			    NULL, 0);
3963 		} else {
3964 			(void) printf("\t\t(object encrypted)\n");
3965 		}
3966 
3967 		*print_header = B_TRUE;
3968 	}
3969 
3970 	if (verbosity >= 5) {
3971 		if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
3972 			char blkbuf[BP_SPRINTF_LEN];
3973 			snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
3974 			    DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE);
3975 			(void) printf("\nSpill block: %s\n", blkbuf);
3976 		}
3977 		dump_indirect(dn);
3978 	}
3979 
3980 	if (verbosity >= 5) {
3981 		/*
3982 		 * Report the list of segments that comprise the object.
3983 		 */
3984 		uint64_t start = 0;
3985 		uint64_t end;
3986 		uint64_t blkfill = 1;
3987 		int minlvl = 1;
3988 
3989 		if (dn->dn_type == DMU_OT_DNODE) {
3990 			minlvl = 0;
3991 			blkfill = DNODES_PER_BLOCK;
3992 		}
3993 
3994 		for (;;) {
3995 			char segsize[32];
3996 			/* make sure nicenum has enough space */
3997 			_Static_assert(sizeof (segsize) >= NN_NUMBUF_SZ,
3998 			    "segsize truncated");
3999 			error = dnode_next_offset(dn,
4000 			    0, &start, minlvl, blkfill, 0);
4001 			if (error)
4002 				break;
4003 			end = start;
4004 			error = dnode_next_offset(dn,
4005 			    DNODE_FIND_HOLE, &end, minlvl, blkfill, 0);
4006 			zdb_nicenum(end - start, segsize, sizeof (segsize));
4007 			(void) printf("\t\tsegment [%016llx, %016llx)"
4008 			    " size %5s\n", (u_longlong_t)start,
4009 			    (u_longlong_t)end, segsize);
4010 			if (error)
4011 				break;
4012 			start = end;
4013 		}
4014 	}
4015 
4016 out:
4017 	if (db != NULL)
4018 		dmu_buf_rele(db, FTAG);
4019 	if (dnode_held)
4020 		dnode_rele(dn, FTAG);
4021 }
4022 
4023 static void
count_dir_mos_objects(dsl_dir_t * dd)4024 count_dir_mos_objects(dsl_dir_t *dd)
4025 {
4026 	mos_obj_refd(dd->dd_object);
4027 	mos_obj_refd(dsl_dir_phys(dd)->dd_child_dir_zapobj);
4028 	mos_obj_refd(dsl_dir_phys(dd)->dd_deleg_zapobj);
4029 	mos_obj_refd(dsl_dir_phys(dd)->dd_props_zapobj);
4030 	mos_obj_refd(dsl_dir_phys(dd)->dd_clones);
4031 
4032 	/*
4033 	 * The dd_crypto_obj can be referenced by multiple dsl_dir's.
4034 	 * Ignore the references after the first one.
4035 	 */
4036 	mos_obj_refd_multiple(dd->dd_crypto_obj);
4037 }
4038 
4039 static void
count_ds_mos_objects(dsl_dataset_t * ds)4040 count_ds_mos_objects(dsl_dataset_t *ds)
4041 {
4042 	mos_obj_refd(ds->ds_object);
4043 	mos_obj_refd(dsl_dataset_phys(ds)->ds_next_clones_obj);
4044 	mos_obj_refd(dsl_dataset_phys(ds)->ds_props_obj);
4045 	mos_obj_refd(dsl_dataset_phys(ds)->ds_userrefs_obj);
4046 	mos_obj_refd(dsl_dataset_phys(ds)->ds_snapnames_zapobj);
4047 	mos_obj_refd(ds->ds_bookmarks_obj);
4048 
4049 	if (!dsl_dataset_is_snapshot(ds)) {
4050 		count_dir_mos_objects(ds->ds_dir);
4051 	}
4052 }
4053 
4054 static const char *const objset_types[DMU_OST_NUMTYPES] = {
4055 	"NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
4056 
4057 /*
4058  * Parse a string denoting a range of object IDs of the form
4059  * <start>[:<end>[:flags]], and store the results in zor.
4060  * Return 0 on success. On error, return 1 and update the msg
4061  * pointer to point to a descriptive error message.
4062  */
4063 static int
parse_object_range(char * range,zopt_object_range_t * zor,const char ** msg)4064 parse_object_range(char *range, zopt_object_range_t *zor, const char **msg)
4065 {
4066 	uint64_t flags = 0;
4067 	char *p, *s, *dup, *flagstr, *tmp = NULL;
4068 	size_t len;
4069 	int i;
4070 	int rc = 0;
4071 
4072 	if (strchr(range, ':') == NULL) {
4073 		zor->zor_obj_start = strtoull(range, &p, 0);
4074 		if (*p != '\0') {
4075 			*msg = "Invalid characters in object ID";
4076 			rc = 1;
4077 		}
4078 		zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4079 		zor->zor_obj_end = zor->zor_obj_start;
4080 		return (rc);
4081 	}
4082 
4083 	if (strchr(range, ':') == range) {
4084 		*msg = "Invalid leading colon";
4085 		rc = 1;
4086 		return (rc);
4087 	}
4088 
4089 	len = strlen(range);
4090 	if (range[len - 1] == ':') {
4091 		*msg = "Invalid trailing colon";
4092 		rc = 1;
4093 		return (rc);
4094 	}
4095 
4096 	dup = strdup(range);
4097 	s = strtok_r(dup, ":", &tmp);
4098 	zor->zor_obj_start = strtoull(s, &p, 0);
4099 
4100 	if (*p != '\0') {
4101 		*msg = "Invalid characters in start object ID";
4102 		rc = 1;
4103 		goto out;
4104 	}
4105 
4106 	s = strtok_r(NULL, ":", &tmp);
4107 	zor->zor_obj_end = strtoull(s, &p, 0);
4108 
4109 	if (*p != '\0') {
4110 		*msg = "Invalid characters in end object ID";
4111 		rc = 1;
4112 		goto out;
4113 	}
4114 
4115 	if (zor->zor_obj_start > zor->zor_obj_end) {
4116 		*msg = "Start object ID may not exceed end object ID";
4117 		rc = 1;
4118 		goto out;
4119 	}
4120 
4121 	s = strtok_r(NULL, ":", &tmp);
4122 	if (s == NULL) {
4123 		zor->zor_flags = ZOR_FLAG_ALL_TYPES;
4124 		goto out;
4125 	} else if (strtok_r(NULL, ":", &tmp) != NULL) {
4126 		*msg = "Invalid colon-delimited field after flags";
4127 		rc = 1;
4128 		goto out;
4129 	}
4130 
4131 	flagstr = s;
4132 	for (i = 0; flagstr[i]; i++) {
4133 		int bit;
4134 		boolean_t negation = (flagstr[i] == '-');
4135 
4136 		if (negation) {
4137 			i++;
4138 			if (flagstr[i] == '\0') {
4139 				*msg = "Invalid trailing negation operator";
4140 				rc = 1;
4141 				goto out;
4142 			}
4143 		}
4144 		bit = flagbits[(uchar_t)flagstr[i]];
4145 		if (bit == 0) {
4146 			*msg = "Invalid flag";
4147 			rc = 1;
4148 			goto out;
4149 		}
4150 		if (negation)
4151 			flags &= ~bit;
4152 		else
4153 			flags |= bit;
4154 	}
4155 	zor->zor_flags = flags;
4156 
4157 	zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start);
4158 	zor->zor_obj_end = ZDB_MAP_OBJECT_ID(zor->zor_obj_end);
4159 
4160 out:
4161 	free(dup);
4162 	return (rc);
4163 }
4164 
4165 static void
dump_objset(objset_t * os)4166 dump_objset(objset_t *os)
4167 {
4168 	dmu_objset_stats_t dds = { 0 };
4169 	uint64_t object, object_count;
4170 	uint64_t refdbytes, usedobjs, scratch;
4171 	char numbuf[32];
4172 	char blkbuf[BP_SPRINTF_LEN + 20];
4173 	char osname[ZFS_MAX_DATASET_NAME_LEN];
4174 	const char *type = "UNKNOWN";
4175 	int verbosity = dump_opt['d'];
4176 	boolean_t print_header;
4177 	unsigned i;
4178 	int error;
4179 	uint64_t total_slots_used = 0;
4180 	uint64_t max_slot_used = 0;
4181 	uint64_t dnode_slots;
4182 	uint64_t obj_start;
4183 	uint64_t obj_end;
4184 	uint64_t flags;
4185 
4186 	/* make sure nicenum has enough space */
4187 	_Static_assert(sizeof (numbuf) >= NN_NUMBUF_SZ, "numbuf truncated");
4188 
4189 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
4190 	dmu_objset_fast_stat(os, &dds);
4191 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
4192 
4193 	print_header = B_TRUE;
4194 
4195 	if (dds.dds_type < DMU_OST_NUMTYPES)
4196 		type = objset_types[dds.dds_type];
4197 
4198 	if (dds.dds_type == DMU_OST_META) {
4199 		dds.dds_creation_txg = TXG_INITIAL;
4200 		usedobjs = BP_GET_FILL(os->os_rootbp);
4201 		refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)->
4202 		    dd_used_bytes;
4203 	} else {
4204 		dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
4205 	}
4206 
4207 	ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
4208 
4209 	zdb_nicenum(refdbytes, numbuf, sizeof (numbuf));
4210 
4211 	if (verbosity >= 4) {
4212 		(void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp ");
4213 		(void) snprintf_blkptr(blkbuf + strlen(blkbuf),
4214 		    sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp);
4215 	} else {
4216 		blkbuf[0] = '\0';
4217 	}
4218 
4219 	dmu_objset_name(os, osname);
4220 
4221 	(void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
4222 	    "%s, %llu objects%s%s\n",
4223 	    osname, type, (u_longlong_t)dmu_objset_id(os),
4224 	    (u_longlong_t)dds.dds_creation_txg,
4225 	    numbuf, (u_longlong_t)usedobjs, blkbuf,
4226 	    (dds.dds_inconsistent) ? " (inconsistent)" : "");
4227 
4228 	for (i = 0; i < zopt_object_args; i++) {
4229 		obj_start = zopt_object_ranges[i].zor_obj_start;
4230 		obj_end = zopt_object_ranges[i].zor_obj_end;
4231 		flags = zopt_object_ranges[i].zor_flags;
4232 
4233 		object = obj_start;
4234 		if (object == 0 || obj_start == obj_end)
4235 			dump_object(os, object, verbosity, &print_header, NULL,
4236 			    flags);
4237 		else
4238 			object--;
4239 
4240 		while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) &&
4241 		    object <= obj_end) {
4242 			dump_object(os, object, verbosity, &print_header, NULL,
4243 			    flags);
4244 		}
4245 	}
4246 
4247 	if (zopt_object_args > 0) {
4248 		(void) printf("\n");
4249 		return;
4250 	}
4251 
4252 	if (dump_opt['i'] != 0 || verbosity >= 2)
4253 		dump_intent_log(dmu_objset_zil(os));
4254 
4255 	if (dmu_objset_ds(os) != NULL) {
4256 		dsl_dataset_t *ds = dmu_objset_ds(os);
4257 		dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
4258 		if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
4259 		    !dmu_objset_is_snapshot(os)) {
4260 			dump_blkptr_list(&ds->ds_dir->dd_livelist, "Livelist");
4261 			if (verify_dd_livelist(os) != 0)
4262 				fatal("livelist is incorrect");
4263 		}
4264 
4265 		if (dsl_dataset_remap_deadlist_exists(ds)) {
4266 			(void) printf("ds_remap_deadlist:\n");
4267 			dump_blkptr_list(&ds->ds_remap_deadlist, "Deadlist");
4268 		}
4269 		count_ds_mos_objects(ds);
4270 	}
4271 
4272 	if (dmu_objset_ds(os) != NULL)
4273 		dump_bookmarks(os, verbosity);
4274 
4275 	if (verbosity < 2)
4276 		return;
4277 
4278 	if (BP_IS_HOLE(os->os_rootbp))
4279 		return;
4280 
4281 	dump_object(os, 0, verbosity, &print_header, NULL, 0);
4282 	object_count = 0;
4283 	if (DMU_USERUSED_DNODE(os) != NULL &&
4284 	    DMU_USERUSED_DNODE(os)->dn_type != 0) {
4285 		dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header,
4286 		    NULL, 0);
4287 		dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header,
4288 		    NULL, 0);
4289 	}
4290 
4291 	if (DMU_PROJECTUSED_DNODE(os) != NULL &&
4292 	    DMU_PROJECTUSED_DNODE(os)->dn_type != 0)
4293 		dump_object(os, DMU_PROJECTUSED_OBJECT, verbosity,
4294 		    &print_header, NULL, 0);
4295 
4296 	object = 0;
4297 	while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
4298 		dump_object(os, object, verbosity, &print_header, &dnode_slots,
4299 		    0);
4300 		object_count++;
4301 		total_slots_used += dnode_slots;
4302 		max_slot_used = object + dnode_slots - 1;
4303 	}
4304 
4305 	(void) printf("\n");
4306 
4307 	(void) printf("    Dnode slots:\n");
4308 	(void) printf("\tTotal used:    %10llu\n",
4309 	    (u_longlong_t)total_slots_used);
4310 	(void) printf("\tMax used:      %10llu\n",
4311 	    (u_longlong_t)max_slot_used);
4312 	(void) printf("\tPercent empty: %10lf\n",
4313 	    (double)(max_slot_used - total_slots_used)*100 /
4314 	    (double)max_slot_used);
4315 	(void) printf("\n");
4316 
4317 	if (error != ESRCH) {
4318 		(void) fprintf(stderr, "dmu_object_next() = %d\n", error);
4319 		abort();
4320 	}
4321 
4322 	ASSERT3U(object_count, ==, usedobjs);
4323 
4324 	if (leaked_objects != 0) {
4325 		(void) printf("%d potentially leaked objects detected\n",
4326 		    leaked_objects);
4327 		leaked_objects = 0;
4328 	}
4329 }
4330 
4331 static void
dump_uberblock(uberblock_t * ub,const char * header,const char * footer)4332 dump_uberblock(uberblock_t *ub, const char *header, const char *footer)
4333 {
4334 	time_t timestamp = ub->ub_timestamp;
4335 
4336 	(void) printf("%s", header ? header : "");
4337 	(void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic);
4338 	(void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version);
4339 	(void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg);
4340 	(void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum);
4341 	(void) printf("\ttimestamp = %llu UTC = %s",
4342 	    (u_longlong_t)ub->ub_timestamp, ctime(&timestamp));
4343 
4344 	char blkbuf[BP_SPRINTF_LEN];
4345 	snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4346 	(void) printf("\tbp = %s\n", blkbuf);
4347 
4348 	(void) printf("\tmmp_magic = %016llx\n",
4349 	    (u_longlong_t)ub->ub_mmp_magic);
4350 	if (MMP_VALID(ub)) {
4351 		(void) printf("\tmmp_delay = %0llu\n",
4352 		    (u_longlong_t)ub->ub_mmp_delay);
4353 		if (MMP_SEQ_VALID(ub))
4354 			(void) printf("\tmmp_seq = %u\n",
4355 			    (unsigned int) MMP_SEQ(ub));
4356 		if (MMP_FAIL_INT_VALID(ub))
4357 			(void) printf("\tmmp_fail = %u\n",
4358 			    (unsigned int) MMP_FAIL_INT(ub));
4359 		if (MMP_INTERVAL_VALID(ub))
4360 			(void) printf("\tmmp_write = %u\n",
4361 			    (unsigned int) MMP_INTERVAL(ub));
4362 		/* After MMP_* to make summarize_uberblock_mmp cleaner */
4363 		(void) printf("\tmmp_valid = %x\n",
4364 		    (unsigned int) ub->ub_mmp_config & 0xFF);
4365 	}
4366 
4367 	if (dump_opt['u'] >= 4) {
4368 		char blkbuf[BP_SPRINTF_LEN];
4369 		snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
4370 		(void) printf("\trootbp = %s\n", blkbuf);
4371 	}
4372 	(void) printf("\tcheckpoint_txg = %llu\n",
4373 	    (u_longlong_t)ub->ub_checkpoint_txg);
4374 
4375 	(void) printf("\traidz_reflow state=%u off=%llu\n",
4376 	    (int)RRSS_GET_STATE(ub),
4377 	    (u_longlong_t)RRSS_GET_OFFSET(ub));
4378 
4379 	(void) printf("%s", footer ? footer : "");
4380 }
4381 
4382 static void
dump_config(spa_t * spa)4383 dump_config(spa_t *spa)
4384 {
4385 	dmu_buf_t *db;
4386 	size_t nvsize = 0;
4387 	int error = 0;
4388 
4389 
4390 	error = dmu_bonus_hold(spa->spa_meta_objset,
4391 	    spa->spa_config_object, FTAG, &db);
4392 
4393 	if (error == 0) {
4394 		nvsize = *(uint64_t *)db->db_data;
4395 		dmu_buf_rele(db, FTAG);
4396 
4397 		(void) printf("\nMOS Configuration:\n");
4398 		dump_packed_nvlist(spa->spa_meta_objset,
4399 		    spa->spa_config_object, (void *)&nvsize, 1);
4400 	} else {
4401 		(void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d",
4402 		    (u_longlong_t)spa->spa_config_object, error);
4403 	}
4404 }
4405 
4406 static void
dump_cachefile(const char * cachefile)4407 dump_cachefile(const char *cachefile)
4408 {
4409 	int fd;
4410 	struct stat64 statbuf;
4411 	char *buf;
4412 	nvlist_t *config;
4413 
4414 	if ((fd = open64(cachefile, O_RDONLY)) < 0) {
4415 		(void) printf("cannot open '%s': %s\n", cachefile,
4416 		    strerror(errno));
4417 		zdb_exit(1);
4418 	}
4419 
4420 	if (fstat64(fd, &statbuf) != 0) {
4421 		(void) printf("failed to stat '%s': %s\n", cachefile,
4422 		    strerror(errno));
4423 		zdb_exit(1);
4424 	}
4425 
4426 	if ((buf = malloc(statbuf.st_size)) == NULL) {
4427 		(void) fprintf(stderr, "failed to allocate %llu bytes\n",
4428 		    (u_longlong_t)statbuf.st_size);
4429 		zdb_exit(1);
4430 	}
4431 
4432 	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
4433 		(void) fprintf(stderr, "failed to read %llu bytes\n",
4434 		    (u_longlong_t)statbuf.st_size);
4435 		zdb_exit(1);
4436 	}
4437 
4438 	(void) close(fd);
4439 
4440 	if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) {
4441 		(void) fprintf(stderr, "failed to unpack nvlist\n");
4442 		zdb_exit(1);
4443 	}
4444 
4445 	free(buf);
4446 
4447 	dump_nvlist(config, 0);
4448 
4449 	nvlist_free(config);
4450 }
4451 
4452 /*
4453  * ZFS label nvlist stats
4454  */
4455 typedef struct zdb_nvl_stats {
4456 	int		zns_list_count;
4457 	int		zns_leaf_count;
4458 	size_t		zns_leaf_largest;
4459 	size_t		zns_leaf_total;
4460 	nvlist_t	*zns_string;
4461 	nvlist_t	*zns_uint64;
4462 	nvlist_t	*zns_boolean;
4463 } zdb_nvl_stats_t;
4464 
4465 static void
collect_nvlist_stats(nvlist_t * nvl,zdb_nvl_stats_t * stats)4466 collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats)
4467 {
4468 	nvlist_t *list, **array;
4469 	nvpair_t *nvp = NULL;
4470 	const char *name;
4471 	uint_t i, items;
4472 
4473 	stats->zns_list_count++;
4474 
4475 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4476 		name = nvpair_name(nvp);
4477 
4478 		switch (nvpair_type(nvp)) {
4479 		case DATA_TYPE_STRING:
4480 			fnvlist_add_string(stats->zns_string, name,
4481 			    fnvpair_value_string(nvp));
4482 			break;
4483 		case DATA_TYPE_UINT64:
4484 			fnvlist_add_uint64(stats->zns_uint64, name,
4485 			    fnvpair_value_uint64(nvp));
4486 			break;
4487 		case DATA_TYPE_BOOLEAN:
4488 			fnvlist_add_boolean(stats->zns_boolean, name);
4489 			break;
4490 		case DATA_TYPE_NVLIST:
4491 			if (nvpair_value_nvlist(nvp, &list) == 0)
4492 				collect_nvlist_stats(list, stats);
4493 			break;
4494 		case DATA_TYPE_NVLIST_ARRAY:
4495 			if (nvpair_value_nvlist_array(nvp, &array, &items) != 0)
4496 				break;
4497 
4498 			for (i = 0; i < items; i++) {
4499 				collect_nvlist_stats(array[i], stats);
4500 
4501 				/* collect stats on leaf vdev */
4502 				if (strcmp(name, "children") == 0) {
4503 					size_t size;
4504 
4505 					(void) nvlist_size(array[i], &size,
4506 					    NV_ENCODE_XDR);
4507 					stats->zns_leaf_total += size;
4508 					if (size > stats->zns_leaf_largest)
4509 						stats->zns_leaf_largest = size;
4510 					stats->zns_leaf_count++;
4511 				}
4512 			}
4513 			break;
4514 		default:
4515 			(void) printf("skip type %d!\n", (int)nvpair_type(nvp));
4516 		}
4517 	}
4518 }
4519 
4520 static void
dump_nvlist_stats(nvlist_t * nvl,size_t cap)4521 dump_nvlist_stats(nvlist_t *nvl, size_t cap)
4522 {
4523 	zdb_nvl_stats_t stats = { 0 };
4524 	size_t size, sum = 0, total;
4525 	size_t noise;
4526 
4527 	/* requires nvlist with non-unique names for stat collection */
4528 	VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0));
4529 	VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0));
4530 	VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0));
4531 	VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR));
4532 
4533 	(void) printf("\n\nZFS Label NVList Config Stats:\n");
4534 
4535 	VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR));
4536 	(void) printf("  %d bytes used, %d bytes free (using %4.1f%%)\n\n",
4537 	    (int)total, (int)(cap - total), 100.0 * total / cap);
4538 
4539 	collect_nvlist_stats(nvl, &stats);
4540 
4541 	VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR));
4542 	size -= noise;
4543 	sum += size;
4544 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
4545 	    (int)fnvlist_num_pairs(stats.zns_uint64),
4546 	    (int)size, 100.0 * size / total);
4547 
4548 	VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR));
4549 	size -= noise;
4550 	sum += size;
4551 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
4552 	    (int)fnvlist_num_pairs(stats.zns_string),
4553 	    (int)size, 100.0 * size / total);
4554 
4555 	VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR));
4556 	size -= noise;
4557 	sum += size;
4558 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
4559 	    (int)fnvlist_num_pairs(stats.zns_boolean),
4560 	    (int)size, 100.0 * size / total);
4561 
4562 	size = total - sum;	/* treat remainder as nvlist overhead */
4563 	(void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
4564 	    stats.zns_list_count, (int)size, 100.0 * size / total);
4565 
4566 	if (stats.zns_leaf_count > 0) {
4567 		size_t average = stats.zns_leaf_total / stats.zns_leaf_count;
4568 
4569 		(void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
4570 		    stats.zns_leaf_count, (int)average);
4571 		(void) printf("%24d bytes largest\n",
4572 		    (int)stats.zns_leaf_largest);
4573 
4574 		if (dump_opt['l'] >= 3 && average > 0)
4575 			(void) printf("  space for %d additional leaf vdevs\n",
4576 			    (int)((cap - total) / average));
4577 	}
4578 	(void) printf("\n");
4579 
4580 	nvlist_free(stats.zns_string);
4581 	nvlist_free(stats.zns_uint64);
4582 	nvlist_free(stats.zns_boolean);
4583 }
4584 
4585 typedef struct cksum_record {
4586 	zio_cksum_t cksum;
4587 	boolean_t labels[VDEV_LABELS];
4588 	avl_node_t link;
4589 } cksum_record_t;
4590 
4591 static int
cksum_record_compare(const void * x1,const void * x2)4592 cksum_record_compare(const void *x1, const void *x2)
4593 {
4594 	const cksum_record_t *l = (cksum_record_t *)x1;
4595 	const cksum_record_t *r = (cksum_record_t *)x2;
4596 	int arraysize = ARRAY_SIZE(l->cksum.zc_word);
4597 	int difference = 0;
4598 
4599 	for (int i = 0; i < arraysize; i++) {
4600 		difference = TREE_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]);
4601 		if (difference)
4602 			break;
4603 	}
4604 
4605 	return (difference);
4606 }
4607 
4608 static cksum_record_t *
cksum_record_alloc(zio_cksum_t * cksum,int l)4609 cksum_record_alloc(zio_cksum_t *cksum, int l)
4610 {
4611 	cksum_record_t *rec;
4612 
4613 	rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL);
4614 	rec->cksum = *cksum;
4615 	rec->labels[l] = B_TRUE;
4616 
4617 	return (rec);
4618 }
4619 
4620 static cksum_record_t *
cksum_record_lookup(avl_tree_t * tree,zio_cksum_t * cksum)4621 cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum)
4622 {
4623 	cksum_record_t lookup = { .cksum = *cksum };
4624 	avl_index_t where;
4625 
4626 	return (avl_find(tree, &lookup, &where));
4627 }
4628 
4629 static cksum_record_t *
cksum_record_insert(avl_tree_t * tree,zio_cksum_t * cksum,int l)4630 cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l)
4631 {
4632 	cksum_record_t *rec;
4633 
4634 	rec = cksum_record_lookup(tree, cksum);
4635 	if (rec) {
4636 		rec->labels[l] = B_TRUE;
4637 	} else {
4638 		rec = cksum_record_alloc(cksum, l);
4639 		avl_add(tree, rec);
4640 	}
4641 
4642 	return (rec);
4643 }
4644 
4645 static int
first_label(cksum_record_t * rec)4646 first_label(cksum_record_t *rec)
4647 {
4648 	for (int i = 0; i < VDEV_LABELS; i++)
4649 		if (rec->labels[i])
4650 			return (i);
4651 
4652 	return (-1);
4653 }
4654 
4655 static void
print_label_numbers(const char * prefix,const cksum_record_t * rec)4656 print_label_numbers(const char *prefix, const cksum_record_t *rec)
4657 {
4658 	fputs(prefix, stdout);
4659 	for (int i = 0; i < VDEV_LABELS; i++)
4660 		if (rec->labels[i] == B_TRUE)
4661 			printf("%d ", i);
4662 	putchar('\n');
4663 }
4664 
4665 #define	MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
4666 
4667 typedef struct zdb_label {
4668 	vdev_label_t label;
4669 	uint64_t label_offset;
4670 	nvlist_t *config_nv;
4671 	cksum_record_t *config;
4672 	cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT];
4673 	boolean_t header_printed;
4674 	boolean_t read_failed;
4675 	boolean_t cksum_valid;
4676 } zdb_label_t;
4677 
4678 static void
print_label_header(zdb_label_t * label,int l)4679 print_label_header(zdb_label_t *label, int l)
4680 {
4681 
4682 	if (dump_opt['q'])
4683 		return;
4684 
4685 	if (label->header_printed == B_TRUE)
4686 		return;
4687 
4688 	(void) printf("------------------------------------\n");
4689 	(void) printf("LABEL %d %s\n", l,
4690 	    label->cksum_valid ? "" : "(Bad label cksum)");
4691 	(void) printf("------------------------------------\n");
4692 
4693 	label->header_printed = B_TRUE;
4694 }
4695 
4696 static void
print_l2arc_header(void)4697 print_l2arc_header(void)
4698 {
4699 	(void) printf("------------------------------------\n");
4700 	(void) printf("L2ARC device header\n");
4701 	(void) printf("------------------------------------\n");
4702 }
4703 
4704 static void
print_l2arc_log_blocks(void)4705 print_l2arc_log_blocks(void)
4706 {
4707 	(void) printf("------------------------------------\n");
4708 	(void) printf("L2ARC device log blocks\n");
4709 	(void) printf("------------------------------------\n");
4710 }
4711 
4712 static void
dump_l2arc_log_entries(uint64_t log_entries,l2arc_log_ent_phys_t * le,uint64_t i)4713 dump_l2arc_log_entries(uint64_t log_entries,
4714     l2arc_log_ent_phys_t *le, uint64_t i)
4715 {
4716 	for (int j = 0; j < log_entries; j++) {
4717 		dva_t dva = le[j].le_dva;
4718 		(void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, "
4719 		    "vdev: %llu, offset: %llu\n",
4720 		    (u_longlong_t)i, j + 1,
4721 		    (u_longlong_t)DVA_GET_ASIZE(&dva),
4722 		    (u_longlong_t)DVA_GET_VDEV(&dva),
4723 		    (u_longlong_t)DVA_GET_OFFSET(&dva));
4724 		(void) printf("|\t\t\t\tbirth: %llu\n",
4725 		    (u_longlong_t)le[j].le_birth);
4726 		(void) printf("|\t\t\t\tlsize: %llu\n",
4727 		    (u_longlong_t)L2BLK_GET_LSIZE((&le[j])->le_prop));
4728 		(void) printf("|\t\t\t\tpsize: %llu\n",
4729 		    (u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop));
4730 		(void) printf("|\t\t\t\tcompr: %llu\n",
4731 		    (u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop));
4732 		(void) printf("|\t\t\t\tcomplevel: %llu\n",
4733 		    (u_longlong_t)(&le[j])->le_complevel);
4734 		(void) printf("|\t\t\t\ttype: %llu\n",
4735 		    (u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop));
4736 		(void) printf("|\t\t\t\tprotected: %llu\n",
4737 		    (u_longlong_t)L2BLK_GET_PROTECTED((&le[j])->le_prop));
4738 		(void) printf("|\t\t\t\tprefetch: %llu\n",
4739 		    (u_longlong_t)L2BLK_GET_PREFETCH((&le[j])->le_prop));
4740 		(void) printf("|\t\t\t\taddress: %llu\n",
4741 		    (u_longlong_t)le[j].le_daddr);
4742 		(void) printf("|\t\t\t\tARC state: %llu\n",
4743 		    (u_longlong_t)L2BLK_GET_STATE((&le[j])->le_prop));
4744 		(void) printf("|\n");
4745 	}
4746 	(void) printf("\n");
4747 }
4748 
4749 static void
dump_l2arc_log_blkptr(const l2arc_log_blkptr_t * lbps)4750 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t *lbps)
4751 {
4752 	(void) printf("|\t\tdaddr: %llu\n", (u_longlong_t)lbps->lbp_daddr);
4753 	(void) printf("|\t\tpayload_asize: %llu\n",
4754 	    (u_longlong_t)lbps->lbp_payload_asize);
4755 	(void) printf("|\t\tpayload_start: %llu\n",
4756 	    (u_longlong_t)lbps->lbp_payload_start);
4757 	(void) printf("|\t\tlsize: %llu\n",
4758 	    (u_longlong_t)L2BLK_GET_LSIZE(lbps->lbp_prop));
4759 	(void) printf("|\t\tasize: %llu\n",
4760 	    (u_longlong_t)L2BLK_GET_PSIZE(lbps->lbp_prop));
4761 	(void) printf("|\t\tcompralgo: %llu\n",
4762 	    (u_longlong_t)L2BLK_GET_COMPRESS(lbps->lbp_prop));
4763 	(void) printf("|\t\tcksumalgo: %llu\n",
4764 	    (u_longlong_t)L2BLK_GET_CHECKSUM(lbps->lbp_prop));
4765 	(void) printf("|\n\n");
4766 }
4767 
4768 static void
dump_l2arc_log_blocks(int fd,const l2arc_dev_hdr_phys_t * l2dhdr,l2arc_dev_hdr_phys_t * rebuild)4769 dump_l2arc_log_blocks(int fd, const l2arc_dev_hdr_phys_t *l2dhdr,
4770     l2arc_dev_hdr_phys_t *rebuild)
4771 {
4772 	l2arc_log_blk_phys_t this_lb;
4773 	uint64_t asize;
4774 	l2arc_log_blkptr_t lbps[2];
4775 	zio_cksum_t cksum;
4776 	int failed = 0;
4777 	l2arc_dev_t dev;
4778 
4779 	if (!dump_opt['q'])
4780 		print_l2arc_log_blocks();
4781 	memcpy(lbps, l2dhdr->dh_start_lbps, sizeof (lbps));
4782 
4783 	dev.l2ad_evict = l2dhdr->dh_evict;
4784 	dev.l2ad_start = l2dhdr->dh_start;
4785 	dev.l2ad_end = l2dhdr->dh_end;
4786 
4787 	if (l2dhdr->dh_start_lbps[0].lbp_daddr == 0) {
4788 		/* no log blocks to read */
4789 		if (!dump_opt['q']) {
4790 			(void) printf("No log blocks to read\n");
4791 			(void) printf("\n");
4792 		}
4793 		return;
4794 	} else {
4795 		dev.l2ad_hand = lbps[0].lbp_daddr +
4796 		    L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4797 	}
4798 
4799 	dev.l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST);
4800 
4801 	for (;;) {
4802 		if (!l2arc_log_blkptr_valid(&dev, &lbps[0]))
4803 			break;
4804 
4805 		/* L2BLK_GET_PSIZE returns aligned size for log blocks */
4806 		asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop);
4807 		if (pread64(fd, &this_lb, asize, lbps[0].lbp_daddr) != asize) {
4808 			if (!dump_opt['q']) {
4809 				(void) printf("Error while reading next log "
4810 				    "block\n\n");
4811 			}
4812 			break;
4813 		}
4814 
4815 		fletcher_4_native_varsize(&this_lb, asize, &cksum);
4816 		if (!ZIO_CHECKSUM_EQUAL(cksum, lbps[0].lbp_cksum)) {
4817 			failed++;
4818 			if (!dump_opt['q']) {
4819 				(void) printf("Invalid cksum\n");
4820 				dump_l2arc_log_blkptr(&lbps[0]);
4821 			}
4822 			break;
4823 		}
4824 
4825 		switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) {
4826 		case ZIO_COMPRESS_OFF:
4827 			break;
4828 		default: {
4829 			abd_t *abd = abd_alloc_linear(asize, B_TRUE);
4830 			abd_copy_from_buf_off(abd, &this_lb, 0, asize);
4831 			abd_t dabd;
4832 			abd_get_from_buf_struct(&dabd, &this_lb,
4833 			    sizeof (this_lb));
4834 			int err = zio_decompress_data(L2BLK_GET_COMPRESS(
4835 			    (&lbps[0])->lbp_prop), abd, &dabd,
4836 			    asize, sizeof (this_lb), NULL);
4837 			abd_free(&dabd);
4838 			abd_free(abd);
4839 			if (err != 0) {
4840 				(void) printf("L2ARC block decompression "
4841 				    "failed\n");
4842 				goto out;
4843 			}
4844 			break;
4845 		}
4846 		}
4847 
4848 		if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC))
4849 			byteswap_uint64_array(&this_lb, sizeof (this_lb));
4850 		if (this_lb.lb_magic != L2ARC_LOG_BLK_MAGIC) {
4851 			if (!dump_opt['q'])
4852 				(void) printf("Invalid log block magic\n\n");
4853 			break;
4854 		}
4855 
4856 		rebuild->dh_lb_count++;
4857 		rebuild->dh_lb_asize += asize;
4858 		if (dump_opt['l'] > 1 && !dump_opt['q']) {
4859 			(void) printf("lb[%4llu]\tmagic: %llu\n",
4860 			    (u_longlong_t)rebuild->dh_lb_count,
4861 			    (u_longlong_t)this_lb.lb_magic);
4862 			dump_l2arc_log_blkptr(&lbps[0]);
4863 		}
4864 
4865 		if (dump_opt['l'] > 2 && !dump_opt['q'])
4866 			dump_l2arc_log_entries(l2dhdr->dh_log_entries,
4867 			    this_lb.lb_entries,
4868 			    rebuild->dh_lb_count);
4869 
4870 		if (l2arc_range_check_overlap(lbps[1].lbp_payload_start,
4871 		    lbps[0].lbp_payload_start, dev.l2ad_evict) &&
4872 		    !dev.l2ad_first)
4873 			break;
4874 
4875 		lbps[0] = lbps[1];
4876 		lbps[1] = this_lb.lb_prev_lbp;
4877 	}
4878 out:
4879 	if (!dump_opt['q']) {
4880 		(void) printf("log_blk_count:\t %llu with valid cksum\n",
4881 		    (u_longlong_t)rebuild->dh_lb_count);
4882 		(void) printf("\t\t %d with invalid cksum\n", failed);
4883 		(void) printf("log_blk_asize:\t %llu\n\n",
4884 		    (u_longlong_t)rebuild->dh_lb_asize);
4885 	}
4886 }
4887 
4888 static int
dump_l2arc_header(int fd)4889 dump_l2arc_header(int fd)
4890 {
4891 	l2arc_dev_hdr_phys_t l2dhdr = {0}, rebuild = {0};
4892 	int error = B_FALSE;
4893 
4894 	if (pread64(fd, &l2dhdr, sizeof (l2dhdr),
4895 	    VDEV_LABEL_START_SIZE) != sizeof (l2dhdr)) {
4896 		error = B_TRUE;
4897 	} else {
4898 		if (l2dhdr.dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC))
4899 			byteswap_uint64_array(&l2dhdr, sizeof (l2dhdr));
4900 
4901 		if (l2dhdr.dh_magic != L2ARC_DEV_HDR_MAGIC)
4902 			error = B_TRUE;
4903 	}
4904 
4905 	if (error) {
4906 		(void) printf("L2ARC device header not found\n\n");
4907 		/* Do not return an error here for backward compatibility */
4908 		return (0);
4909 	} else if (!dump_opt['q']) {
4910 		print_l2arc_header();
4911 
4912 		(void) printf("    magic: %llu\n",
4913 		    (u_longlong_t)l2dhdr.dh_magic);
4914 		(void) printf("    version: %llu\n",
4915 		    (u_longlong_t)l2dhdr.dh_version);
4916 		(void) printf("    pool_guid: %llu\n",
4917 		    (u_longlong_t)l2dhdr.dh_spa_guid);
4918 		(void) printf("    flags: %llu\n",
4919 		    (u_longlong_t)l2dhdr.dh_flags);
4920 		(void) printf("    start_lbps[0]: %llu\n",
4921 		    (u_longlong_t)
4922 		    l2dhdr.dh_start_lbps[0].lbp_daddr);
4923 		(void) printf("    start_lbps[1]: %llu\n",
4924 		    (u_longlong_t)
4925 		    l2dhdr.dh_start_lbps[1].lbp_daddr);
4926 		(void) printf("    log_blk_ent: %llu\n",
4927 		    (u_longlong_t)l2dhdr.dh_log_entries);
4928 		(void) printf("    start: %llu\n",
4929 		    (u_longlong_t)l2dhdr.dh_start);
4930 		(void) printf("    end: %llu\n",
4931 		    (u_longlong_t)l2dhdr.dh_end);
4932 		(void) printf("    evict: %llu\n",
4933 		    (u_longlong_t)l2dhdr.dh_evict);
4934 		(void) printf("    lb_asize_refcount: %llu\n",
4935 		    (u_longlong_t)l2dhdr.dh_lb_asize);
4936 		(void) printf("    lb_count_refcount: %llu\n",
4937 		    (u_longlong_t)l2dhdr.dh_lb_count);
4938 		(void) printf("    trim_action_time: %llu\n",
4939 		    (u_longlong_t)l2dhdr.dh_trim_action_time);
4940 		(void) printf("    trim_state: %llu\n\n",
4941 		    (u_longlong_t)l2dhdr.dh_trim_state);
4942 	}
4943 
4944 	dump_l2arc_log_blocks(fd, &l2dhdr, &rebuild);
4945 	/*
4946 	 * The total aligned size of log blocks and the number of log blocks
4947 	 * reported in the header of the device may be less than what zdb
4948 	 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild().
4949 	 * This happens because dump_l2arc_log_blocks() lacks the memory
4950 	 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system
4951 	 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize
4952 	 * and dh_lb_count will be lower to begin with than what exists on the
4953 	 * device. This is normal and zdb should not exit with an error. The
4954 	 * opposite case should never happen though, the values reported in the
4955 	 * header should never be higher than what dump_l2arc_log_blocks() and
4956 	 * l2arc_rebuild() report. If this happens there is a leak in the
4957 	 * accounting of log blocks.
4958 	 */
4959 	if (l2dhdr.dh_lb_asize > rebuild.dh_lb_asize ||
4960 	    l2dhdr.dh_lb_count > rebuild.dh_lb_count)
4961 		return (1);
4962 
4963 	return (0);
4964 }
4965 
4966 static void
dump_config_from_label(zdb_label_t * label,size_t buflen,int l)4967 dump_config_from_label(zdb_label_t *label, size_t buflen, int l)
4968 {
4969 	if (dump_opt['q'])
4970 		return;
4971 
4972 	if ((dump_opt['l'] < 3) && (first_label(label->config) != l))
4973 		return;
4974 
4975 	print_label_header(label, l);
4976 	dump_nvlist(label->config_nv, 4);
4977 	print_label_numbers("    labels = ", label->config);
4978 
4979 	if (dump_opt['l'] >= 2)
4980 		dump_nvlist_stats(label->config_nv, buflen);
4981 }
4982 
4983 #define	ZDB_MAX_UB_HEADER_SIZE 32
4984 
4985 static void
dump_label_uberblocks(zdb_label_t * label,uint64_t ashift,int label_num)4986 dump_label_uberblocks(zdb_label_t *label, uint64_t ashift, int label_num)
4987 {
4988 
4989 	vdev_t vd;
4990 	char header[ZDB_MAX_UB_HEADER_SIZE];
4991 
4992 	vd.vdev_ashift = ashift;
4993 	vd.vdev_top = &vd;
4994 
4995 	for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
4996 		uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
4997 		uberblock_t *ub = (void *)((char *)&label->label + uoff);
4998 		cksum_record_t *rec = label->uberblocks[i];
4999 
5000 		if (rec == NULL) {
5001 			if (dump_opt['u'] >= 2) {
5002 				print_label_header(label, label_num);
5003 				(void) printf("    Uberblock[%d] invalid\n", i);
5004 			}
5005 			continue;
5006 		}
5007 
5008 		if ((dump_opt['u'] < 3) && (first_label(rec) != label_num))
5009 			continue;
5010 
5011 		if ((dump_opt['u'] < 4) &&
5012 		    (ub->ub_mmp_magic == MMP_MAGIC) && ub->ub_mmp_delay &&
5013 		    (i >= VDEV_UBERBLOCK_COUNT(&vd) - MMP_BLOCKS_PER_LABEL))
5014 			continue;
5015 
5016 		print_label_header(label, label_num);
5017 		(void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE,
5018 		    "    Uberblock[%d]\n", i);
5019 		dump_uberblock(ub, header, "");
5020 		print_label_numbers("        labels = ", rec);
5021 	}
5022 }
5023 
5024 static char curpath[PATH_MAX];
5025 
5026 /*
5027  * Iterate through the path components, recursively passing
5028  * current one's obj and remaining path until we find the obj
5029  * for the last one.
5030  */
5031 static int
dump_path_impl(objset_t * os,uint64_t obj,char * name,uint64_t * retobj)5032 dump_path_impl(objset_t *os, uint64_t obj, char *name, uint64_t *retobj)
5033 {
5034 	int err;
5035 	boolean_t header = B_TRUE;
5036 	uint64_t child_obj;
5037 	char *s;
5038 	dmu_buf_t *db;
5039 	dmu_object_info_t doi;
5040 
5041 	if ((s = strchr(name, '/')) != NULL)
5042 		*s = '\0';
5043 	err = zap_lookup(os, obj, name, 8, 1, &child_obj);
5044 
5045 	(void) strlcat(curpath, name, sizeof (curpath));
5046 
5047 	if (err != 0) {
5048 		(void) fprintf(stderr, "failed to lookup %s: %s\n",
5049 		    curpath, strerror(err));
5050 		return (err);
5051 	}
5052 
5053 	child_obj = ZFS_DIRENT_OBJ(child_obj);
5054 	err = sa_buf_hold(os, child_obj, FTAG, &db);
5055 	if (err != 0) {
5056 		(void) fprintf(stderr,
5057 		    "failed to get SA dbuf for obj %llu: %s\n",
5058 		    (u_longlong_t)child_obj, strerror(err));
5059 		return (EINVAL);
5060 	}
5061 	dmu_object_info_from_db(db, &doi);
5062 	sa_buf_rele(db, FTAG);
5063 
5064 	if (doi.doi_bonus_type != DMU_OT_SA &&
5065 	    doi.doi_bonus_type != DMU_OT_ZNODE) {
5066 		(void) fprintf(stderr, "invalid bonus type %d for obj %llu\n",
5067 		    doi.doi_bonus_type, (u_longlong_t)child_obj);
5068 		return (EINVAL);
5069 	}
5070 
5071 	if (dump_opt['v'] > 6) {
5072 		(void) printf("obj=%llu %s type=%d bonustype=%d\n",
5073 		    (u_longlong_t)child_obj, curpath, doi.doi_type,
5074 		    doi.doi_bonus_type);
5075 	}
5076 
5077 	(void) strlcat(curpath, "/", sizeof (curpath));
5078 
5079 	switch (doi.doi_type) {
5080 	case DMU_OT_DIRECTORY_CONTENTS:
5081 		if (s != NULL && *(s + 1) != '\0')
5082 			return (dump_path_impl(os, child_obj, s + 1, retobj));
5083 		zfs_fallthrough;
5084 	case DMU_OT_PLAIN_FILE_CONTENTS:
5085 		if (retobj != NULL) {
5086 			*retobj = child_obj;
5087 		} else {
5088 			dump_object(os, child_obj, dump_opt['v'], &header,
5089 			    NULL, 0);
5090 		}
5091 		return (0);
5092 	default:
5093 		(void) fprintf(stderr, "object %llu has non-file/directory "
5094 		    "type %d\n", (u_longlong_t)obj, doi.doi_type);
5095 		break;
5096 	}
5097 
5098 	return (EINVAL);
5099 }
5100 
5101 /*
5102  * Dump the blocks for the object specified by path inside the dataset.
5103  */
5104 static int
dump_path(char * ds,char * path,uint64_t * retobj)5105 dump_path(char *ds, char *path, uint64_t *retobj)
5106 {
5107 	int err;
5108 	objset_t *os;
5109 	uint64_t root_obj;
5110 
5111 	err = open_objset(ds, FTAG, &os);
5112 	if (err != 0)
5113 		return (err);
5114 
5115 	err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj);
5116 	if (err != 0) {
5117 		(void) fprintf(stderr, "can't lookup root znode: %s\n",
5118 		    strerror(err));
5119 		close_objset(os, FTAG);
5120 		return (EINVAL);
5121 	}
5122 
5123 	(void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds);
5124 
5125 	err = dump_path_impl(os, root_obj, path, retobj);
5126 
5127 	close_objset(os, FTAG);
5128 	return (err);
5129 }
5130 
5131 static int
dump_backup_bytes(objset_t * os,void * buf,int len,void * arg)5132 dump_backup_bytes(objset_t *os, void *buf, int len, void *arg)
5133 {
5134 	const char *p = (const char *)buf;
5135 	ssize_t nwritten;
5136 
5137 	(void) os;
5138 	(void) arg;
5139 
5140 	/* Write the data out, handling short writes and signals. */
5141 	while ((nwritten = write(STDOUT_FILENO, p, len)) < len) {
5142 		if (nwritten < 0) {
5143 			if (errno == EINTR)
5144 				continue;
5145 			return (errno);
5146 		}
5147 		p += nwritten;
5148 		len -= nwritten;
5149 	}
5150 
5151 	return (0);
5152 }
5153 
5154 static void
dump_backup(const char * pool,uint64_t objset_id,const char * flagstr)5155 dump_backup(const char *pool, uint64_t objset_id, const char *flagstr)
5156 {
5157 	boolean_t embed = B_FALSE;
5158 	boolean_t large_block = B_FALSE;
5159 	boolean_t compress = B_FALSE;
5160 	boolean_t raw = B_FALSE;
5161 
5162 	const char *c;
5163 	for (c = flagstr; c != NULL && *c != '\0'; c++) {
5164 		switch (*c) {
5165 			case 'e':
5166 				embed = B_TRUE;
5167 				break;
5168 			case 'L':
5169 				large_block = B_TRUE;
5170 				break;
5171 			case 'c':
5172 				compress = B_TRUE;
5173 				break;
5174 			case 'w':
5175 				raw = B_TRUE;
5176 				break;
5177 			default:
5178 				fprintf(stderr, "dump_backup: invalid flag "
5179 				    "'%c'\n", *c);
5180 				return;
5181 		}
5182 	}
5183 
5184 	if (isatty(STDOUT_FILENO)) {
5185 		fprintf(stderr, "dump_backup: stream cannot be written "
5186 		    "to a terminal\n");
5187 		return;
5188 	}
5189 
5190 	offset_t off = 0;
5191 	dmu_send_outparams_t out = {
5192 	    .dso_outfunc = dump_backup_bytes,
5193 	    .dso_dryrun  = B_FALSE,
5194 	};
5195 
5196 	int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed,
5197 	    large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO,
5198 	    &off, &out);
5199 	if (err != 0) {
5200 		fprintf(stderr, "dump_backup: dmu_send_obj: %s\n",
5201 		    strerror(err));
5202 		return;
5203 	}
5204 }
5205 
5206 static int
zdb_copy_object(objset_t * os,uint64_t srcobj,char * destfile)5207 zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile)
5208 {
5209 	int err = 0;
5210 	uint64_t size, readsize, oursize, offset;
5211 	ssize_t writesize;
5212 	sa_handle_t *hdl;
5213 
5214 	(void) printf("Copying object %" PRIu64 " to file %s\n", srcobj,
5215 	    destfile);
5216 
5217 	VERIFY3P(os, ==, sa_os);
5218 	if ((err = sa_handle_get(os, srcobj, NULL, SA_HDL_PRIVATE, &hdl))) {
5219 		(void) printf("Failed to get handle for SA znode\n");
5220 		return (err);
5221 	}
5222 	if ((err = sa_lookup(hdl, sa_attr_table[ZPL_SIZE], &size, 8))) {
5223 		(void) sa_handle_destroy(hdl);
5224 		return (err);
5225 	}
5226 	(void) sa_handle_destroy(hdl);
5227 
5228 	(void) printf("Object %" PRIu64 " is %" PRIu64 " bytes\n", srcobj,
5229 	    size);
5230 	if (size == 0) {
5231 		return (EINVAL);
5232 	}
5233 
5234 	int fd = open(destfile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
5235 	if (fd == -1)
5236 		return (errno);
5237 	/*
5238 	 * We cap the size at 1 mebibyte here to prevent
5239 	 * allocation failures and nigh-infinite printing if the
5240 	 * object is extremely large.
5241 	 */
5242 	oursize = MIN(size, 1 << 20);
5243 	offset = 0;
5244 	char *buf = kmem_alloc(oursize, KM_NOSLEEP);
5245 	if (buf == NULL) {
5246 		(void) close(fd);
5247 		return (ENOMEM);
5248 	}
5249 
5250 	while (offset < size) {
5251 		readsize = MIN(size - offset, 1 << 20);
5252 		err = dmu_read(os, srcobj, offset, readsize, buf, 0);
5253 		if (err != 0) {
5254 			(void) printf("got error %u from dmu_read\n", err);
5255 			kmem_free(buf, oursize);
5256 			(void) close(fd);
5257 			return (err);
5258 		}
5259 		if (dump_opt['v'] > 3) {
5260 			(void) printf("Read offset=%" PRIu64 " size=%" PRIu64
5261 			    " error=%d\n", offset, readsize, err);
5262 		}
5263 
5264 		writesize = write(fd, buf, readsize);
5265 		if (writesize < 0) {
5266 			err = errno;
5267 			break;
5268 		} else if (writesize != readsize) {
5269 			/* Incomplete write */
5270 			(void) fprintf(stderr, "Short write, only wrote %llu of"
5271 			    " %" PRIu64 " bytes, exiting...\n",
5272 			    (u_longlong_t)writesize, readsize);
5273 			break;
5274 		}
5275 
5276 		offset += readsize;
5277 	}
5278 
5279 	(void) close(fd);
5280 
5281 	if (buf != NULL)
5282 		kmem_free(buf, oursize);
5283 
5284 	return (err);
5285 }
5286 
5287 static boolean_t
label_cksum_valid(vdev_label_t * label,uint64_t offset)5288 label_cksum_valid(vdev_label_t *label, uint64_t offset)
5289 {
5290 	zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
5291 	zio_cksum_t expected_cksum;
5292 	zio_cksum_t actual_cksum;
5293 	zio_cksum_t verifier;
5294 	zio_eck_t *eck;
5295 	int byteswap;
5296 
5297 	void *data = (char *)label + offsetof(vdev_label_t, vl_vdev_phys);
5298 	eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1;
5299 
5300 	offset += offsetof(vdev_label_t, vl_vdev_phys);
5301 	ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
5302 
5303 	byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
5304 	if (byteswap)
5305 		byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
5306 
5307 	expected_cksum = eck->zec_cksum;
5308 	eck->zec_cksum = verifier;
5309 
5310 	abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE);
5311 	ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum);
5312 	abd_free(abd);
5313 
5314 	if (byteswap)
5315 		byteswap_uint64_array(&expected_cksum, sizeof (zio_cksum_t));
5316 
5317 	if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
5318 		return (B_TRUE);
5319 
5320 	return (B_FALSE);
5321 }
5322 
5323 static int
dump_label(const char * dev)5324 dump_label(const char *dev)
5325 {
5326 	char path[MAXPATHLEN];
5327 	zdb_label_t labels[VDEV_LABELS] = {{{{0}}}};
5328 	uint64_t psize, ashift, l2cache;
5329 	struct stat64 statbuf;
5330 	boolean_t config_found = B_FALSE;
5331 	boolean_t error = B_FALSE;
5332 	boolean_t read_l2arc_header = B_FALSE;
5333 	avl_tree_t config_tree;
5334 	avl_tree_t uberblock_tree;
5335 	void *node, *cookie;
5336 	int fd;
5337 
5338 	/*
5339 	 * Check if we were given absolute path and use it as is.
5340 	 * Otherwise if the provided vdev name doesn't point to a file,
5341 	 * try prepending expected disk paths and partition numbers.
5342 	 */
5343 	(void) strlcpy(path, dev, sizeof (path));
5344 	if (dev[0] != '/' && stat64(path, &statbuf) != 0) {
5345 		int error;
5346 
5347 		error = zfs_resolve_shortname(dev, path, MAXPATHLEN);
5348 		if (error == 0 && zfs_dev_is_whole_disk(path)) {
5349 			if (zfs_append_partition(path, MAXPATHLEN) == -1)
5350 				error = ENOENT;
5351 		}
5352 
5353 		if (error || (stat64(path, &statbuf) != 0)) {
5354 			(void) printf("failed to find device %s, try "
5355 			    "specifying absolute path instead\n", dev);
5356 			return (1);
5357 		}
5358 	}
5359 
5360 	if ((fd = open64(path, O_RDONLY)) < 0) {
5361 		(void) printf("cannot open '%s': %s\n", path, strerror(errno));
5362 		zdb_exit(1);
5363 	}
5364 
5365 	if (fstat64_blk(fd, &statbuf) != 0) {
5366 		(void) printf("failed to stat '%s': %s\n", path,
5367 		    strerror(errno));
5368 		(void) close(fd);
5369 		zdb_exit(1);
5370 	}
5371 
5372 	if (S_ISBLK(statbuf.st_mode) && zfs_dev_flush(fd) != 0)
5373 		(void) printf("failed to invalidate cache '%s' : %s\n", path,
5374 		    strerror(errno));
5375 
5376 	avl_create(&config_tree, cksum_record_compare,
5377 	    sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5378 	avl_create(&uberblock_tree, cksum_record_compare,
5379 	    sizeof (cksum_record_t), offsetof(cksum_record_t, link));
5380 
5381 	psize = statbuf.st_size;
5382 	psize = P2ALIGN_TYPED(psize, sizeof (vdev_label_t), uint64_t);
5383 	ashift = SPA_MINBLOCKSHIFT;
5384 
5385 	/*
5386 	 * 1. Read the label from disk
5387 	 * 2. Verify label cksum
5388 	 * 3. Unpack the configuration and insert in config tree.
5389 	 * 4. Traverse all uberblocks and insert in uberblock tree.
5390 	 */
5391 	for (int l = 0; l < VDEV_LABELS; l++) {
5392 		zdb_label_t *label = &labels[l];
5393 		char *buf = label->label.vl_vdev_phys.vp_nvlist;
5394 		size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5395 		nvlist_t *config;
5396 		cksum_record_t *rec;
5397 		zio_cksum_t cksum;
5398 		vdev_t vd;
5399 
5400 		label->label_offset = vdev_label_offset(psize, l, 0);
5401 
5402 		if (pread64(fd, &label->label, sizeof (label->label),
5403 		    label->label_offset) != sizeof (label->label)) {
5404 			if (!dump_opt['q'])
5405 				(void) printf("failed to read label %d\n", l);
5406 			label->read_failed = B_TRUE;
5407 			error = B_TRUE;
5408 			continue;
5409 		}
5410 
5411 		label->read_failed = B_FALSE;
5412 		label->cksum_valid = label_cksum_valid(&label->label,
5413 		    label->label_offset);
5414 
5415 		if (nvlist_unpack(buf, buflen, &config, 0) == 0) {
5416 			nvlist_t *vdev_tree = NULL;
5417 			size_t size;
5418 
5419 			if ((nvlist_lookup_nvlist(config,
5420 			    ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) ||
5421 			    (nvlist_lookup_uint64(vdev_tree,
5422 			    ZPOOL_CONFIG_ASHIFT, &ashift) != 0))
5423 				ashift = SPA_MINBLOCKSHIFT;
5424 
5425 			if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0)
5426 				size = buflen;
5427 
5428 			/* If the device is a cache device read the header. */
5429 			if (!read_l2arc_header) {
5430 				if (nvlist_lookup_uint64(config,
5431 				    ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 &&
5432 				    l2cache == POOL_STATE_L2CACHE) {
5433 					read_l2arc_header = B_TRUE;
5434 				}
5435 			}
5436 
5437 			fletcher_4_native_varsize(buf, size, &cksum);
5438 			rec = cksum_record_insert(&config_tree, &cksum, l);
5439 
5440 			label->config = rec;
5441 			label->config_nv = config;
5442 			config_found = B_TRUE;
5443 		} else {
5444 			error = B_TRUE;
5445 		}
5446 
5447 		vd.vdev_ashift = ashift;
5448 		vd.vdev_top = &vd;
5449 
5450 		for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
5451 			uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
5452 			uberblock_t *ub = (void *)((char *)label + uoff);
5453 
5454 			if (uberblock_verify(ub))
5455 				continue;
5456 
5457 			fletcher_4_native_varsize(ub, sizeof (*ub), &cksum);
5458 			rec = cksum_record_insert(&uberblock_tree, &cksum, l);
5459 
5460 			label->uberblocks[i] = rec;
5461 		}
5462 	}
5463 
5464 	/*
5465 	 * Dump the label and uberblocks.
5466 	 */
5467 	for (int l = 0; l < VDEV_LABELS; l++) {
5468 		zdb_label_t *label = &labels[l];
5469 		size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
5470 
5471 		if (label->read_failed == B_TRUE)
5472 			continue;
5473 
5474 		if (label->config_nv) {
5475 			dump_config_from_label(label, buflen, l);
5476 		} else {
5477 			if (!dump_opt['q'])
5478 				(void) printf("failed to unpack label %d\n", l);
5479 		}
5480 
5481 		if (dump_opt['u'])
5482 			dump_label_uberblocks(label, ashift, l);
5483 
5484 		nvlist_free(label->config_nv);
5485 	}
5486 
5487 	/*
5488 	 * Dump the L2ARC header, if existent.
5489 	 */
5490 	if (read_l2arc_header)
5491 		error |= dump_l2arc_header(fd);
5492 
5493 	cookie = NULL;
5494 	while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL)
5495 		umem_free(node, sizeof (cksum_record_t));
5496 
5497 	cookie = NULL;
5498 	while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL)
5499 		umem_free(node, sizeof (cksum_record_t));
5500 
5501 	avl_destroy(&config_tree);
5502 	avl_destroy(&uberblock_tree);
5503 
5504 	(void) close(fd);
5505 
5506 	return (config_found == B_FALSE ? 2 :
5507 	    (error == B_TRUE ? 1 : 0));
5508 }
5509 
5510 static uint64_t dataset_feature_count[SPA_FEATURES];
5511 static uint64_t global_feature_count[SPA_FEATURES];
5512 static uint64_t remap_deadlist_count = 0;
5513 
5514 static int
dump_one_objset(const char * dsname,void * arg)5515 dump_one_objset(const char *dsname, void *arg)
5516 {
5517 	(void) arg;
5518 	int error;
5519 	objset_t *os;
5520 	spa_feature_t f;
5521 
5522 	error = open_objset(dsname, FTAG, &os);
5523 	if (error != 0)
5524 		return (0);
5525 
5526 	for (f = 0; f < SPA_FEATURES; f++) {
5527 		if (!dsl_dataset_feature_is_active(dmu_objset_ds(os), f))
5528 			continue;
5529 		ASSERT(spa_feature_table[f].fi_flags &
5530 		    ZFEATURE_FLAG_PER_DATASET);
5531 		dataset_feature_count[f]++;
5532 	}
5533 
5534 	if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os))) {
5535 		remap_deadlist_count++;
5536 	}
5537 
5538 	for (dsl_bookmark_node_t *dbn =
5539 	    avl_first(&dmu_objset_ds(os)->ds_bookmarks); dbn != NULL;
5540 	    dbn = AVL_NEXT(&dmu_objset_ds(os)->ds_bookmarks, dbn)) {
5541 		mos_obj_refd(dbn->dbn_phys.zbm_redaction_obj);
5542 		if (dbn->dbn_phys.zbm_redaction_obj != 0) {
5543 			global_feature_count[
5544 			    SPA_FEATURE_REDACTION_BOOKMARKS]++;
5545 			objset_t *mos = os->os_spa->spa_meta_objset;
5546 			dnode_t *rl;
5547 			VERIFY0(dnode_hold(mos,
5548 			    dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl));
5549 			if (rl->dn_have_spill) {
5550 				global_feature_count[
5551 				    SPA_FEATURE_REDACTION_LIST_SPILL]++;
5552 			}
5553 		}
5554 		if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN)
5555 			global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN]++;
5556 	}
5557 
5558 	if (dsl_deadlist_is_open(&dmu_objset_ds(os)->ds_dir->dd_livelist) &&
5559 	    !dmu_objset_is_snapshot(os)) {
5560 		global_feature_count[SPA_FEATURE_LIVELIST]++;
5561 	}
5562 
5563 	dump_objset(os);
5564 	close_objset(os, FTAG);
5565 	fuid_table_destroy();
5566 	return (0);
5567 }
5568 
5569 /*
5570  * Block statistics.
5571  */
5572 #define	PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
5573 typedef struct zdb_blkstats {
5574 	uint64_t zb_asize;
5575 	uint64_t zb_lsize;
5576 	uint64_t zb_psize;
5577 	uint64_t zb_count;
5578 	uint64_t zb_gangs;
5579 	uint64_t zb_ditto_samevdev;
5580 	uint64_t zb_ditto_same_ms;
5581 	uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
5582 } zdb_blkstats_t;
5583 
5584 /*
5585  * Extended object types to report deferred frees and dedup auto-ditto blocks.
5586  */
5587 #define	ZDB_OT_DEFERRED	(DMU_OT_NUMTYPES + 0)
5588 #define	ZDB_OT_DITTO	(DMU_OT_NUMTYPES + 1)
5589 #define	ZDB_OT_OTHER	(DMU_OT_NUMTYPES + 2)
5590 #define	ZDB_OT_TOTAL	(DMU_OT_NUMTYPES + 3)
5591 
5592 static const char *zdb_ot_extname[] = {
5593 	"deferred free",
5594 	"dedup ditto",
5595 	"other",
5596 	"Total",
5597 };
5598 
5599 #define	ZB_TOTAL	DN_MAX_LEVELS
5600 #define	SPA_MAX_FOR_16M	(SPA_MAXBLOCKSHIFT+1)
5601 
5602 typedef struct zdb_brt_entry {
5603 	dva_t		zbre_dva;
5604 	uint64_t	zbre_refcount;
5605 	avl_node_t	zbre_node;
5606 } zdb_brt_entry_t;
5607 
5608 typedef struct zdb_cb {
5609 	zdb_blkstats_t	zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
5610 	uint64_t	zcb_removing_size;
5611 	uint64_t	zcb_checkpoint_size;
5612 	uint64_t	zcb_dedup_asize;
5613 	uint64_t	zcb_dedup_blocks;
5614 	uint64_t	zcb_clone_asize;
5615 	uint64_t	zcb_clone_blocks;
5616 	uint64_t	zcb_psize_count[SPA_MAX_FOR_16M];
5617 	uint64_t	zcb_lsize_count[SPA_MAX_FOR_16M];
5618 	uint64_t	zcb_asize_count[SPA_MAX_FOR_16M];
5619 	uint64_t	zcb_psize_len[SPA_MAX_FOR_16M];
5620 	uint64_t	zcb_lsize_len[SPA_MAX_FOR_16M];
5621 	uint64_t	zcb_asize_len[SPA_MAX_FOR_16M];
5622 	uint64_t	zcb_psize_total;
5623 	uint64_t	zcb_lsize_total;
5624 	uint64_t	zcb_asize_total;
5625 	uint64_t	zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
5626 	uint64_t	zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
5627 	    [BPE_PAYLOAD_SIZE + 1];
5628 	uint64_t	zcb_start;
5629 	hrtime_t	zcb_lastprint;
5630 	uint64_t	zcb_totalasize;
5631 	uint64_t	zcb_errors[256];
5632 	int		zcb_readfails;
5633 	int		zcb_haderrors;
5634 	spa_t		*zcb_spa;
5635 	uint32_t	**zcb_vd_obsolete_counts;
5636 	avl_tree_t	zcb_brt;
5637 	boolean_t	zcb_brt_is_active;
5638 } zdb_cb_t;
5639 
5640 /* test if two DVA offsets from same vdev are within the same metaslab */
5641 static boolean_t
same_metaslab(spa_t * spa,uint64_t vdev,uint64_t off1,uint64_t off2)5642 same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2)
5643 {
5644 	vdev_t *vd = vdev_lookup_top(spa, vdev);
5645 	uint64_t ms_shift = vd->vdev_ms_shift;
5646 
5647 	return ((off1 >> ms_shift) == (off2 >> ms_shift));
5648 }
5649 
5650 /*
5651  * Used to simplify reporting of the histogram data.
5652  */
5653 typedef struct one_histo {
5654 	const char *name;
5655 	uint64_t *count;
5656 	uint64_t *len;
5657 	uint64_t cumulative;
5658 } one_histo_t;
5659 
5660 /*
5661  * The number of separate histograms processed for psize, lsize and asize.
5662  */
5663 #define	NUM_HISTO 3
5664 
5665 /*
5666  * This routine will create a fixed column size output of three different
5667  * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M
5668  * the count, length and cumulative length of the psize, lsize and
5669  * asize blocks.
5670  *
5671  * All three types of blocks are listed on a single line
5672  *
5673  * By default the table is printed in nicenumber format (e.g. 123K) but
5674  * if the '-P' parameter is specified then the full raw number (parseable)
5675  * is printed out.
5676  */
5677 static void
dump_size_histograms(zdb_cb_t * zcb)5678 dump_size_histograms(zdb_cb_t *zcb)
5679 {
5680 	/*
5681 	 * A temporary buffer that allows us to convert a number into
5682 	 * a string using zdb_nicenumber to allow either raw or human
5683 	 * readable numbers to be output.
5684 	 */
5685 	char numbuf[32];
5686 
5687 	/*
5688 	 * Define titles which are used in the headers of the tables
5689 	 * printed by this routine.
5690 	 */
5691 	const char blocksize_title1[] = "block";
5692 	const char blocksize_title2[] = "size";
5693 	const char count_title[] = "Count";
5694 	const char length_title[] = "Size";
5695 	const char cumulative_title[] = "Cum.";
5696 
5697 	/*
5698 	 * Setup the histogram arrays (psize, lsize, and asize).
5699 	 */
5700 	one_histo_t parm_histo[NUM_HISTO];
5701 
5702 	parm_histo[0].name = "psize";
5703 	parm_histo[0].count = zcb->zcb_psize_count;
5704 	parm_histo[0].len = zcb->zcb_psize_len;
5705 	parm_histo[0].cumulative = 0;
5706 
5707 	parm_histo[1].name = "lsize";
5708 	parm_histo[1].count = zcb->zcb_lsize_count;
5709 	parm_histo[1].len = zcb->zcb_lsize_len;
5710 	parm_histo[1].cumulative = 0;
5711 
5712 	parm_histo[2].name = "asize";
5713 	parm_histo[2].count = zcb->zcb_asize_count;
5714 	parm_histo[2].len = zcb->zcb_asize_len;
5715 	parm_histo[2].cumulative = 0;
5716 
5717 
5718 	(void) printf("\nBlock Size Histogram\n");
5719 	/*
5720 	 * Print the first line titles
5721 	 */
5722 	if (dump_opt['P'])
5723 		(void) printf("\n%s\t", blocksize_title1);
5724 	else
5725 		(void) printf("\n%7s   ", blocksize_title1);
5726 
5727 	for (int j = 0; j < NUM_HISTO; j++) {
5728 		if (dump_opt['P']) {
5729 			if (j < NUM_HISTO - 1) {
5730 				(void) printf("%s\t\t\t", parm_histo[j].name);
5731 			} else {
5732 				/* Don't print trailing spaces */
5733 				(void) printf("  %s", parm_histo[j].name);
5734 			}
5735 		} else {
5736 			if (j < NUM_HISTO - 1) {
5737 				/* Left aligned strings in the output */
5738 				(void) printf("%-7s              ",
5739 				    parm_histo[j].name);
5740 			} else {
5741 				/* Don't print trailing spaces */
5742 				(void) printf("%s", parm_histo[j].name);
5743 			}
5744 		}
5745 	}
5746 	(void) printf("\n");
5747 
5748 	/*
5749 	 * Print the second line titles
5750 	 */
5751 	if (dump_opt['P']) {
5752 		(void) printf("%s\t", blocksize_title2);
5753 	} else {
5754 		(void) printf("%7s ", blocksize_title2);
5755 	}
5756 
5757 	for (int i = 0; i < NUM_HISTO; i++) {
5758 		if (dump_opt['P']) {
5759 			(void) printf("%s\t%s\t%s\t",
5760 			    count_title, length_title, cumulative_title);
5761 		} else {
5762 			(void) printf("%7s%7s%7s",
5763 			    count_title, length_title, cumulative_title);
5764 		}
5765 	}
5766 	(void) printf("\n");
5767 
5768 	/*
5769 	 * Print the rows
5770 	 */
5771 	for (int i = SPA_MINBLOCKSHIFT; i < SPA_MAX_FOR_16M; i++) {
5772 
5773 		/*
5774 		 * Print the first column showing the blocksize
5775 		 */
5776 		zdb_nicenum((1ULL << i), numbuf, sizeof (numbuf));
5777 
5778 		if (dump_opt['P']) {
5779 			printf("%s", numbuf);
5780 		} else {
5781 			printf("%7s:", numbuf);
5782 		}
5783 
5784 		/*
5785 		 * Print the remaining set of 3 columns per size:
5786 		 * for psize, lsize and asize
5787 		 */
5788 		for (int j = 0; j < NUM_HISTO; j++) {
5789 			parm_histo[j].cumulative += parm_histo[j].len[i];
5790 
5791 			zdb_nicenum(parm_histo[j].count[i],
5792 			    numbuf, sizeof (numbuf));
5793 			if (dump_opt['P'])
5794 				(void) printf("\t%s", numbuf);
5795 			else
5796 				(void) printf("%7s", numbuf);
5797 
5798 			zdb_nicenum(parm_histo[j].len[i],
5799 			    numbuf, sizeof (numbuf));
5800 			if (dump_opt['P'])
5801 				(void) printf("\t%s", numbuf);
5802 			else
5803 				(void) printf("%7s", numbuf);
5804 
5805 			zdb_nicenum(parm_histo[j].cumulative,
5806 			    numbuf, sizeof (numbuf));
5807 			if (dump_opt['P'])
5808 				(void) printf("\t%s", numbuf);
5809 			else
5810 				(void) printf("%7s", numbuf);
5811 		}
5812 		(void) printf("\n");
5813 	}
5814 }
5815 
5816 static void
zdb_count_block(zdb_cb_t * zcb,zilog_t * zilog,const blkptr_t * bp,dmu_object_type_t type)5817 zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
5818     dmu_object_type_t type)
5819 {
5820 	int i;
5821 
5822 	ASSERT(type < ZDB_OT_TOTAL);
5823 
5824 	if (zilog && zil_bp_tree_add(zilog, bp) != 0)
5825 		return;
5826 
5827 	/*
5828 	 * This flag controls if we will issue a claim for the block while
5829 	 * counting it, to ensure that all blocks are referenced in space maps.
5830 	 * We don't issue claims if we're not doing leak tracking, because it's
5831 	 * expensive if the user isn't interested. We also don't claim the
5832 	 * second or later occurences of cloned or dedup'd blocks, because we
5833 	 * already claimed them the first time.
5834 	 */
5835 	boolean_t do_claim = !dump_opt['L'];
5836 
5837 	spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
5838 
5839 	blkptr_t tempbp;
5840 	if (BP_GET_DEDUP(bp)) {
5841 		/*
5842 		 * Dedup'd blocks are special. We need to count them, so we can
5843 		 * later uncount them when reporting leaked space, and we must
5844 		 * only claim them once.
5845 		 *
5846 		 * We use the existing dedup system to track what we've seen.
5847 		 * The first time we see a block, we do a ddt_lookup() to see
5848 		 * if it exists in the DDT. If we're doing leak tracking, we
5849 		 * claim the block at this time.
5850 		 *
5851 		 * Each time we see a block, we reduce the refcount in the
5852 		 * entry by one, and add to the size and count of dedup'd
5853 		 * blocks to report at the end.
5854 		 */
5855 
5856 		ddt_t *ddt = ddt_select(zcb->zcb_spa, bp);
5857 
5858 		ddt_enter(ddt);
5859 
5860 		/*
5861 		 * Find the block. This will create the entry in memory, but
5862 		 * we'll know if that happened by its refcount.
5863 		 */
5864 		ddt_entry_t *dde = ddt_lookup(ddt, bp);
5865 
5866 		/*
5867 		 * ddt_lookup() can return NULL if this block didn't exist
5868 		 * in the DDT and creating it would take the DDT over its
5869 		 * quota. Since we got the block from disk, it must exist in
5870 		 * the DDT, so this can't happen. However, when unique entries
5871 		 * are pruned, the dedup bit can be set with no corresponding
5872 		 * entry in the DDT.
5873 		 */
5874 		if (dde == NULL) {
5875 			ddt_exit(ddt);
5876 			goto skipped;
5877 		}
5878 
5879 		/* Get the phys for this variant */
5880 		ddt_phys_variant_t v = ddt_phys_select(ddt, dde, bp);
5881 
5882 		/*
5883 		 * This entry may have multiple sets of DVAs. We must claim
5884 		 * each set the first time we see them in a real block on disk,
5885 		 * or count them on subsequent occurences. We don't have a
5886 		 * convenient way to track the first time we see each variant,
5887 		 * so we repurpose dde_io as a set of "seen" flag bits. We can
5888 		 * do this safely in zdb because it never writes, so it will
5889 		 * never have a writing zio for this block in that pointer.
5890 		 */
5891 		boolean_t seen = !!(((uintptr_t)dde->dde_io) & (1 << v));
5892 		if (!seen)
5893 			dde->dde_io =
5894 			    (void *)(((uintptr_t)dde->dde_io) | (1 << v));
5895 
5896 		/* Consume a reference for this block. */
5897 		if (ddt_phys_total_refcnt(ddt, dde->dde_phys) > 0)
5898 			ddt_phys_decref(dde->dde_phys, v);
5899 
5900 		/*
5901 		 * If this entry has a single flat phys, it may have been
5902 		 * extended with additional DVAs at some time in its life.
5903 		 * This block might be from before it was fully extended, and
5904 		 * so have fewer DVAs.
5905 		 *
5906 		 * If this is the first time we've seen this block, and we
5907 		 * claimed it as-is, then we would miss the claim on some
5908 		 * number of DVAs, which would then be seen as leaked.
5909 		 *
5910 		 * In all cases, if we've had fewer DVAs, then the asize would
5911 		 * be too small, and would lead to the pool apparently using
5912 		 * more space than allocated.
5913 		 *
5914 		 * To handle this, we copy the canonical set of DVAs from the
5915 		 * entry back to the block pointer before we claim it.
5916 		 */
5917 		if (v == DDT_PHYS_FLAT) {
5918 			ASSERT3U(BP_GET_BIRTH(bp), ==,
5919 			    ddt_phys_birth(dde->dde_phys, v));
5920 			tempbp = *bp;
5921 			ddt_bp_fill(dde->dde_phys, v, &tempbp,
5922 			    BP_GET_BIRTH(bp));
5923 			bp = &tempbp;
5924 		}
5925 
5926 		if (seen) {
5927 			/*
5928 			 * The second or later time we see this block,
5929 			 * it's a duplicate and we count it.
5930 			 */
5931 			zcb->zcb_dedup_asize += BP_GET_ASIZE(bp);
5932 			zcb->zcb_dedup_blocks++;
5933 
5934 			/* Already claimed, don't do it again. */
5935 			do_claim = B_FALSE;
5936 		}
5937 
5938 		ddt_exit(ddt);
5939 	} else if (zcb->zcb_brt_is_active &&
5940 	    brt_maybe_exists(zcb->zcb_spa, bp)) {
5941 		/*
5942 		 * Cloned blocks are special. We need to count them, so we can
5943 		 * later uncount them when reporting leaked space, and we must
5944 		 * only claim them once.
5945 		 *
5946 		 * To do this, we keep our own in-memory BRT. For each block
5947 		 * we haven't seen before, we look it up in the real BRT and
5948 		 * if its there, we note it and its refcount then proceed as
5949 		 * normal. If we see the block again, we count it as a clone
5950 		 * and then give it no further consideration.
5951 		 */
5952 		zdb_brt_entry_t zbre_search, *zbre;
5953 		avl_index_t where;
5954 
5955 		zbre_search.zbre_dva = bp->blk_dva[0];
5956 		zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where);
5957 		if (zbre == NULL) {
5958 			/* Not seen before; track it */
5959 			uint64_t refcnt =
5960 			    brt_entry_get_refcount(zcb->zcb_spa, bp);
5961 			if (refcnt > 0) {
5962 				zbre = umem_zalloc(sizeof (zdb_brt_entry_t),
5963 				    UMEM_NOFAIL);
5964 				zbre->zbre_dva = bp->blk_dva[0];
5965 				zbre->zbre_refcount = refcnt;
5966 				avl_insert(&zcb->zcb_brt, zbre, where);
5967 			}
5968 		} else  {
5969 			/*
5970 			 * Second or later occurrence, count it and take a
5971 			 * refcount.
5972 			 */
5973 			zcb->zcb_clone_asize += BP_GET_ASIZE(bp);
5974 			zcb->zcb_clone_blocks++;
5975 
5976 			zbre->zbre_refcount--;
5977 			if (zbre->zbre_refcount == 0) {
5978 				avl_remove(&zcb->zcb_brt, zbre);
5979 				umem_free(zbre, sizeof (zdb_brt_entry_t));
5980 			}
5981 
5982 			/* Already claimed, don't do it again. */
5983 			do_claim = B_FALSE;
5984 		}
5985 	}
5986 
5987 skipped:
5988 	for (i = 0; i < 4; i++) {
5989 		int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
5990 		int t = (i & 1) ? type : ZDB_OT_TOTAL;
5991 		int equal;
5992 		zdb_blkstats_t *zb = &zcb->zcb_type[l][t];
5993 
5994 		zb->zb_asize += BP_GET_ASIZE(bp);
5995 		zb->zb_lsize += BP_GET_LSIZE(bp);
5996 		zb->zb_psize += BP_GET_PSIZE(bp);
5997 		zb->zb_count++;
5998 
5999 		/*
6000 		 * The histogram is only big enough to record blocks up to
6001 		 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
6002 		 * "other", bucket.
6003 		 */
6004 		unsigned idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT;
6005 		idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1);
6006 		zb->zb_psize_histogram[idx]++;
6007 
6008 		zb->zb_gangs += BP_COUNT_GANG(bp);
6009 
6010 		switch (BP_GET_NDVAS(bp)) {
6011 		case 2:
6012 			if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6013 			    DVA_GET_VDEV(&bp->blk_dva[1])) {
6014 				zb->zb_ditto_samevdev++;
6015 
6016 				if (same_metaslab(zcb->zcb_spa,
6017 				    DVA_GET_VDEV(&bp->blk_dva[0]),
6018 				    DVA_GET_OFFSET(&bp->blk_dva[0]),
6019 				    DVA_GET_OFFSET(&bp->blk_dva[1])))
6020 					zb->zb_ditto_same_ms++;
6021 			}
6022 			break;
6023 		case 3:
6024 			equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6025 			    DVA_GET_VDEV(&bp->blk_dva[1])) +
6026 			    (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6027 			    DVA_GET_VDEV(&bp->blk_dva[2])) +
6028 			    (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6029 			    DVA_GET_VDEV(&bp->blk_dva[2]));
6030 			if (equal != 0) {
6031 				zb->zb_ditto_samevdev++;
6032 
6033 				if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6034 				    DVA_GET_VDEV(&bp->blk_dva[1]) &&
6035 				    same_metaslab(zcb->zcb_spa,
6036 				    DVA_GET_VDEV(&bp->blk_dva[0]),
6037 				    DVA_GET_OFFSET(&bp->blk_dva[0]),
6038 				    DVA_GET_OFFSET(&bp->blk_dva[1])))
6039 					zb->zb_ditto_same_ms++;
6040 				else if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
6041 				    DVA_GET_VDEV(&bp->blk_dva[2]) &&
6042 				    same_metaslab(zcb->zcb_spa,
6043 				    DVA_GET_VDEV(&bp->blk_dva[0]),
6044 				    DVA_GET_OFFSET(&bp->blk_dva[0]),
6045 				    DVA_GET_OFFSET(&bp->blk_dva[2])))
6046 					zb->zb_ditto_same_ms++;
6047 				else if (DVA_GET_VDEV(&bp->blk_dva[1]) ==
6048 				    DVA_GET_VDEV(&bp->blk_dva[2]) &&
6049 				    same_metaslab(zcb->zcb_spa,
6050 				    DVA_GET_VDEV(&bp->blk_dva[1]),
6051 				    DVA_GET_OFFSET(&bp->blk_dva[1]),
6052 				    DVA_GET_OFFSET(&bp->blk_dva[2])))
6053 					zb->zb_ditto_same_ms++;
6054 			}
6055 			break;
6056 		}
6057 	}
6058 
6059 	spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
6060 
6061 	if (BP_IS_EMBEDDED(bp)) {
6062 		zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
6063 		zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
6064 		    [BPE_GET_PSIZE(bp)]++;
6065 		return;
6066 	}
6067 	/*
6068 	 * The binning histogram bins by powers of two up to
6069 	 * SPA_MAXBLOCKSIZE rather than creating bins for
6070 	 * every possible blocksize found in the pool.
6071 	 */
6072 	int bin = highbit64(BP_GET_PSIZE(bp)) - 1;
6073 
6074 	zcb->zcb_psize_count[bin]++;
6075 	zcb->zcb_psize_len[bin] += BP_GET_PSIZE(bp);
6076 	zcb->zcb_psize_total += BP_GET_PSIZE(bp);
6077 
6078 	bin = highbit64(BP_GET_LSIZE(bp)) - 1;
6079 
6080 	zcb->zcb_lsize_count[bin]++;
6081 	zcb->zcb_lsize_len[bin] += BP_GET_LSIZE(bp);
6082 	zcb->zcb_lsize_total += BP_GET_LSIZE(bp);
6083 
6084 	bin = highbit64(BP_GET_ASIZE(bp)) - 1;
6085 
6086 	zcb->zcb_asize_count[bin]++;
6087 	zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp);
6088 	zcb->zcb_asize_total += BP_GET_ASIZE(bp);
6089 
6090 	if (!do_claim)
6091 		return;
6092 
6093 	VERIFY0(zio_wait(zio_claim(NULL, zcb->zcb_spa,
6094 	    spa_min_claim_txg(zcb->zcb_spa), bp, NULL, NULL,
6095 	    ZIO_FLAG_CANFAIL)));
6096 }
6097 
6098 static void
zdb_blkptr_done(zio_t * zio)6099 zdb_blkptr_done(zio_t *zio)
6100 {
6101 	spa_t *spa = zio->io_spa;
6102 	blkptr_t *bp = zio->io_bp;
6103 	int ioerr = zio->io_error;
6104 	zdb_cb_t *zcb = zio->io_private;
6105 	zbookmark_phys_t *zb = &zio->io_bookmark;
6106 
6107 	mutex_enter(&spa->spa_scrub_lock);
6108 	spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp);
6109 	cv_broadcast(&spa->spa_scrub_io_cv);
6110 
6111 	if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
6112 		char blkbuf[BP_SPRINTF_LEN];
6113 
6114 		zcb->zcb_haderrors = 1;
6115 		zcb->zcb_errors[ioerr]++;
6116 
6117 		if (dump_opt['b'] >= 2)
6118 			snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6119 		else
6120 			blkbuf[0] = '\0';
6121 
6122 		(void) printf("zdb_blkptr_cb: "
6123 		    "Got error %d reading "
6124 		    "<%llu, %llu, %lld, %llx> %s -- skipping\n",
6125 		    ioerr,
6126 		    (u_longlong_t)zb->zb_objset,
6127 		    (u_longlong_t)zb->zb_object,
6128 		    (u_longlong_t)zb->zb_level,
6129 		    (u_longlong_t)zb->zb_blkid,
6130 		    blkbuf);
6131 	}
6132 	mutex_exit(&spa->spa_scrub_lock);
6133 
6134 	abd_free(zio->io_abd);
6135 }
6136 
6137 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)6138 zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
6139     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
6140 {
6141 	zdb_cb_t *zcb = arg;
6142 	dmu_object_type_t type;
6143 	boolean_t is_metadata;
6144 
6145 	if (zb->zb_level == ZB_DNODE_LEVEL)
6146 		return (0);
6147 
6148 	if (dump_opt['b'] >= 5 && BP_GET_LOGICAL_BIRTH(bp) > 0) {
6149 		char blkbuf[BP_SPRINTF_LEN];
6150 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6151 		(void) printf("objset %llu object %llu "
6152 		    "level %lld offset 0x%llx %s\n",
6153 		    (u_longlong_t)zb->zb_objset,
6154 		    (u_longlong_t)zb->zb_object,
6155 		    (longlong_t)zb->zb_level,
6156 		    (u_longlong_t)blkid2offset(dnp, bp, zb),
6157 		    blkbuf);
6158 	}
6159 
6160 	if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
6161 		return (0);
6162 
6163 	type = BP_GET_TYPE(bp);
6164 
6165 	zdb_count_block(zcb, zilog, bp,
6166 	    (type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type);
6167 
6168 	is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
6169 
6170 	if (!BP_IS_EMBEDDED(bp) &&
6171 	    (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
6172 		size_t size = BP_GET_PSIZE(bp);
6173 		abd_t *abd = abd_alloc(size, B_FALSE);
6174 		int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
6175 
6176 		/* If it's an intent log block, failure is expected. */
6177 		if (zb->zb_level == ZB_ZIL_LEVEL)
6178 			flags |= ZIO_FLAG_SPECULATIVE;
6179 
6180 		mutex_enter(&spa->spa_scrub_lock);
6181 		while (spa->spa_load_verify_bytes > max_inflight_bytes)
6182 			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
6183 		spa->spa_load_verify_bytes += size;
6184 		mutex_exit(&spa->spa_scrub_lock);
6185 
6186 		zio_nowait(zio_read(NULL, spa, bp, abd, size,
6187 		    zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
6188 	}
6189 
6190 	zcb->zcb_readfails = 0;
6191 
6192 	/* only call gethrtime() every 100 blocks */
6193 	static int iters;
6194 	if (++iters > 100)
6195 		iters = 0;
6196 	else
6197 		return (0);
6198 
6199 	if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
6200 		uint64_t now = gethrtime();
6201 		char buf[10];
6202 		uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
6203 		uint64_t kb_per_sec =
6204 		    1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000));
6205 		uint64_t sec_remaining =
6206 		    (zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec;
6207 
6208 		/* make sure nicenum has enough space */
6209 		_Static_assert(sizeof (buf) >= NN_NUMBUF_SZ, "buf truncated");
6210 
6211 		zfs_nicebytes(bytes, buf, sizeof (buf));
6212 		(void) fprintf(stderr,
6213 		    "\r%5s completed (%4"PRIu64"MB/s) "
6214 		    "estimated time remaining: "
6215 		    "%"PRIu64"hr %02"PRIu64"min %02"PRIu64"sec        ",
6216 		    buf, kb_per_sec / 1024,
6217 		    sec_remaining / 60 / 60,
6218 		    sec_remaining / 60 % 60,
6219 		    sec_remaining % 60);
6220 
6221 		zcb->zcb_lastprint = now;
6222 	}
6223 
6224 	return (0);
6225 }
6226 
6227 static void
zdb_leak(void * arg,uint64_t start,uint64_t size)6228 zdb_leak(void *arg, uint64_t start, uint64_t size)
6229 {
6230 	vdev_t *vd = arg;
6231 
6232 	(void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
6233 	    (u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
6234 }
6235 
6236 static metaslab_ops_t zdb_metaslab_ops = {
6237 	NULL	/* alloc */
6238 };
6239 
6240 static int
load_unflushed_svr_segs_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6241 load_unflushed_svr_segs_cb(spa_t *spa, space_map_entry_t *sme,
6242     uint64_t txg, void *arg)
6243 {
6244 	spa_vdev_removal_t *svr = arg;
6245 
6246 	uint64_t offset = sme->sme_offset;
6247 	uint64_t size = sme->sme_run;
6248 
6249 	/* skip vdevs we don't care about */
6250 	if (sme->sme_vdev != svr->svr_vdev_id)
6251 		return (0);
6252 
6253 	vdev_t *vd = vdev_lookup_top(spa, sme->sme_vdev);
6254 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6255 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6256 
6257 	if (txg < metaslab_unflushed_txg(ms))
6258 		return (0);
6259 
6260 	if (sme->sme_type == SM_ALLOC)
6261 		range_tree_add(svr->svr_allocd_segs, offset, size);
6262 	else
6263 		range_tree_remove(svr->svr_allocd_segs, offset, size);
6264 
6265 	return (0);
6266 }
6267 
6268 static void
claim_segment_impl_cb(uint64_t inner_offset,vdev_t * vd,uint64_t offset,uint64_t size,void * arg)6269 claim_segment_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset,
6270     uint64_t size, void *arg)
6271 {
6272 	(void) inner_offset, (void) arg;
6273 
6274 	/*
6275 	 * This callback was called through a remap from
6276 	 * a device being removed. Therefore, the vdev that
6277 	 * this callback is applied to is a concrete
6278 	 * vdev.
6279 	 */
6280 	ASSERT(vdev_is_concrete(vd));
6281 
6282 	VERIFY0(metaslab_claim_impl(vd, offset, size,
6283 	    spa_min_claim_txg(vd->vdev_spa)));
6284 }
6285 
6286 static void
claim_segment_cb(void * arg,uint64_t offset,uint64_t size)6287 claim_segment_cb(void *arg, uint64_t offset, uint64_t size)
6288 {
6289 	vdev_t *vd = arg;
6290 
6291 	vdev_indirect_ops.vdev_op_remap(vd, offset, size,
6292 	    claim_segment_impl_cb, NULL);
6293 }
6294 
6295 /*
6296  * After accounting for all allocated blocks that are directly referenced,
6297  * we might have missed a reference to a block from a partially complete
6298  * (and thus unused) indirect mapping object. We perform a secondary pass
6299  * through the metaslabs we have already mapped and claim the destination
6300  * blocks.
6301  */
6302 static void
zdb_claim_removing(spa_t * spa,zdb_cb_t * zcb)6303 zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb)
6304 {
6305 	if (dump_opt['L'])
6306 		return;
6307 
6308 	if (spa->spa_vdev_removal == NULL)
6309 		return;
6310 
6311 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6312 
6313 	spa_vdev_removal_t *svr = spa->spa_vdev_removal;
6314 	vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id);
6315 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6316 
6317 	ASSERT0(range_tree_space(svr->svr_allocd_segs));
6318 
6319 	range_tree_t *allocs = range_tree_create(NULL, RANGE_SEG64, NULL, 0, 0);
6320 	for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) {
6321 		metaslab_t *msp = vd->vdev_ms[msi];
6322 
6323 		ASSERT0(range_tree_space(allocs));
6324 		if (msp->ms_sm != NULL)
6325 			VERIFY0(space_map_load(msp->ms_sm, allocs, SM_ALLOC));
6326 		range_tree_vacate(allocs, range_tree_add, svr->svr_allocd_segs);
6327 	}
6328 	range_tree_destroy(allocs);
6329 
6330 	iterate_through_spacemap_logs(spa, load_unflushed_svr_segs_cb, svr);
6331 
6332 	/*
6333 	 * Clear everything past what has been synced,
6334 	 * because we have not allocated mappings for
6335 	 * it yet.
6336 	 */
6337 	range_tree_clear(svr->svr_allocd_segs,
6338 	    vdev_indirect_mapping_max_offset(vim),
6339 	    vd->vdev_asize - vdev_indirect_mapping_max_offset(vim));
6340 
6341 	zcb->zcb_removing_size += range_tree_space(svr->svr_allocd_segs);
6342 	range_tree_vacate(svr->svr_allocd_segs, claim_segment_cb, vd);
6343 
6344 	spa_config_exit(spa, SCL_CONFIG, FTAG);
6345 }
6346 
6347 static int
increment_indirect_mapping_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6348 increment_indirect_mapping_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6349     dmu_tx_t *tx)
6350 {
6351 	(void) tx;
6352 	zdb_cb_t *zcb = arg;
6353 	spa_t *spa = zcb->zcb_spa;
6354 	vdev_t *vd;
6355 	const dva_t *dva = &bp->blk_dva[0];
6356 
6357 	ASSERT(!bp_freed);
6358 	ASSERT(!dump_opt['L']);
6359 	ASSERT3U(BP_GET_NDVAS(bp), ==, 1);
6360 
6361 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6362 	vd = vdev_lookup_top(zcb->zcb_spa, DVA_GET_VDEV(dva));
6363 	ASSERT3P(vd, !=, NULL);
6364 	spa_config_exit(spa, SCL_VDEV, FTAG);
6365 
6366 	ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0);
6367 	ASSERT3P(zcb->zcb_vd_obsolete_counts[vd->vdev_id], !=, NULL);
6368 
6369 	vdev_indirect_mapping_increment_obsolete_count(
6370 	    vd->vdev_indirect_mapping,
6371 	    DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva),
6372 	    zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6373 
6374 	return (0);
6375 }
6376 
6377 static uint32_t *
zdb_load_obsolete_counts(vdev_t * vd)6378 zdb_load_obsolete_counts(vdev_t *vd)
6379 {
6380 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6381 	spa_t *spa = vd->vdev_spa;
6382 	spa_condensing_indirect_phys_t *scip =
6383 	    &spa->spa_condensing_indirect_phys;
6384 	uint64_t obsolete_sm_object;
6385 	uint32_t *counts;
6386 
6387 	VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
6388 	EQUIV(obsolete_sm_object != 0, vd->vdev_obsolete_sm != NULL);
6389 	counts = vdev_indirect_mapping_load_obsolete_counts(vim);
6390 	if (vd->vdev_obsolete_sm != NULL) {
6391 		vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6392 		    vd->vdev_obsolete_sm);
6393 	}
6394 	if (scip->scip_vdev == vd->vdev_id &&
6395 	    scip->scip_prev_obsolete_sm_object != 0) {
6396 		space_map_t *prev_obsolete_sm = NULL;
6397 		VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset,
6398 		    scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0));
6399 		vdev_indirect_mapping_load_obsolete_spacemap(vim, counts,
6400 		    prev_obsolete_sm);
6401 		space_map_close(prev_obsolete_sm);
6402 	}
6403 	return (counts);
6404 }
6405 
6406 typedef struct checkpoint_sm_exclude_entry_arg {
6407 	vdev_t *cseea_vd;
6408 	uint64_t cseea_checkpoint_size;
6409 } checkpoint_sm_exclude_entry_arg_t;
6410 
6411 static int
checkpoint_sm_exclude_entry_cb(space_map_entry_t * sme,void * arg)6412 checkpoint_sm_exclude_entry_cb(space_map_entry_t *sme, void *arg)
6413 {
6414 	checkpoint_sm_exclude_entry_arg_t *cseea = arg;
6415 	vdev_t *vd = cseea->cseea_vd;
6416 	metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
6417 	uint64_t end = sme->sme_offset + sme->sme_run;
6418 
6419 	ASSERT(sme->sme_type == SM_FREE);
6420 
6421 	/*
6422 	 * Since the vdev_checkpoint_sm exists in the vdev level
6423 	 * and the ms_sm space maps exist in the metaslab level,
6424 	 * an entry in the checkpoint space map could theoretically
6425 	 * cross the boundaries of the metaslab that it belongs.
6426 	 *
6427 	 * In reality, because of the way that we populate and
6428 	 * manipulate the checkpoint's space maps currently,
6429 	 * there shouldn't be any entries that cross metaslabs.
6430 	 * Hence the assertion below.
6431 	 *
6432 	 * That said, there is no fundamental requirement that
6433 	 * the checkpoint's space map entries should not cross
6434 	 * metaslab boundaries. So if needed we could add code
6435 	 * that handles metaslab-crossing segments in the future.
6436 	 */
6437 	VERIFY3U(sme->sme_offset, >=, ms->ms_start);
6438 	VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
6439 
6440 	/*
6441 	 * By removing the entry from the allocated segments we
6442 	 * also verify that the entry is there to begin with.
6443 	 */
6444 	mutex_enter(&ms->ms_lock);
6445 	range_tree_remove(ms->ms_allocatable, sme->sme_offset, sme->sme_run);
6446 	mutex_exit(&ms->ms_lock);
6447 
6448 	cseea->cseea_checkpoint_size += sme->sme_run;
6449 	return (0);
6450 }
6451 
6452 static void
zdb_leak_init_vdev_exclude_checkpoint(vdev_t * vd,zdb_cb_t * zcb)6453 zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_cb_t *zcb)
6454 {
6455 	spa_t *spa = vd->vdev_spa;
6456 	space_map_t *checkpoint_sm = NULL;
6457 	uint64_t checkpoint_sm_obj;
6458 
6459 	/*
6460 	 * If there is no vdev_top_zap, we are in a pool whose
6461 	 * version predates the pool checkpoint feature.
6462 	 */
6463 	if (vd->vdev_top_zap == 0)
6464 		return;
6465 
6466 	/*
6467 	 * If there is no reference of the vdev_checkpoint_sm in
6468 	 * the vdev_top_zap, then one of the following scenarios
6469 	 * is true:
6470 	 *
6471 	 * 1] There is no checkpoint
6472 	 * 2] There is a checkpoint, but no checkpointed blocks
6473 	 *    have been freed yet
6474 	 * 3] The current vdev is indirect
6475 	 *
6476 	 * In these cases we return immediately.
6477 	 */
6478 	if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
6479 	    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
6480 		return;
6481 
6482 	VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
6483 	    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1,
6484 	    &checkpoint_sm_obj));
6485 
6486 	checkpoint_sm_exclude_entry_arg_t cseea;
6487 	cseea.cseea_vd = vd;
6488 	cseea.cseea_checkpoint_size = 0;
6489 
6490 	VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
6491 	    checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
6492 
6493 	VERIFY0(space_map_iterate(checkpoint_sm,
6494 	    space_map_length(checkpoint_sm),
6495 	    checkpoint_sm_exclude_entry_cb, &cseea));
6496 	space_map_close(checkpoint_sm);
6497 
6498 	zcb->zcb_checkpoint_size += cseea.cseea_checkpoint_size;
6499 }
6500 
6501 static void
zdb_leak_init_exclude_checkpoint(spa_t * spa,zdb_cb_t * zcb)6502 zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb)
6503 {
6504 	ASSERT(!dump_opt['L']);
6505 
6506 	vdev_t *rvd = spa->spa_root_vdev;
6507 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6508 		ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id);
6509 		zdb_leak_init_vdev_exclude_checkpoint(rvd->vdev_child[c], zcb);
6510 	}
6511 }
6512 
6513 static int
count_unflushed_space_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6514 count_unflushed_space_cb(spa_t *spa, space_map_entry_t *sme,
6515     uint64_t txg, void *arg)
6516 {
6517 	int64_t *ualloc_space = arg;
6518 
6519 	uint64_t offset = sme->sme_offset;
6520 	uint64_t vdev_id = sme->sme_vdev;
6521 
6522 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6523 	if (!vdev_is_concrete(vd))
6524 		return (0);
6525 
6526 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6527 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6528 
6529 	if (txg < metaslab_unflushed_txg(ms))
6530 		return (0);
6531 
6532 	if (sme->sme_type == SM_ALLOC)
6533 		*ualloc_space += sme->sme_run;
6534 	else
6535 		*ualloc_space -= sme->sme_run;
6536 
6537 	return (0);
6538 }
6539 
6540 static int64_t
get_unflushed_alloc_space(spa_t * spa)6541 get_unflushed_alloc_space(spa_t *spa)
6542 {
6543 	if (dump_opt['L'])
6544 		return (0);
6545 
6546 	int64_t ualloc_space = 0;
6547 	iterate_through_spacemap_logs(spa, count_unflushed_space_cb,
6548 	    &ualloc_space);
6549 	return (ualloc_space);
6550 }
6551 
6552 static int
load_unflushed_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)6553 load_unflushed_cb(spa_t *spa, space_map_entry_t *sme, uint64_t txg, void *arg)
6554 {
6555 	maptype_t *uic_maptype = arg;
6556 
6557 	uint64_t offset = sme->sme_offset;
6558 	uint64_t size = sme->sme_run;
6559 	uint64_t vdev_id = sme->sme_vdev;
6560 
6561 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
6562 
6563 	/* skip indirect vdevs */
6564 	if (!vdev_is_concrete(vd))
6565 		return (0);
6566 
6567 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6568 
6569 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
6570 	ASSERT(*uic_maptype == SM_ALLOC || *uic_maptype == SM_FREE);
6571 
6572 	if (txg < metaslab_unflushed_txg(ms))
6573 		return (0);
6574 
6575 	if (*uic_maptype == sme->sme_type)
6576 		range_tree_add(ms->ms_allocatable, offset, size);
6577 	else
6578 		range_tree_remove(ms->ms_allocatable, offset, size);
6579 
6580 	return (0);
6581 }
6582 
6583 static void
load_unflushed_to_ms_allocatables(spa_t * spa,maptype_t maptype)6584 load_unflushed_to_ms_allocatables(spa_t *spa, maptype_t maptype)
6585 {
6586 	iterate_through_spacemap_logs(spa, load_unflushed_cb, &maptype);
6587 }
6588 
6589 static void
load_concrete_ms_allocatable_trees(spa_t * spa,maptype_t maptype)6590 load_concrete_ms_allocatable_trees(spa_t *spa, maptype_t maptype)
6591 {
6592 	vdev_t *rvd = spa->spa_root_vdev;
6593 	for (uint64_t i = 0; i < rvd->vdev_children; i++) {
6594 		vdev_t *vd = rvd->vdev_child[i];
6595 
6596 		ASSERT3U(i, ==, vd->vdev_id);
6597 
6598 		if (vd->vdev_ops == &vdev_indirect_ops)
6599 			continue;
6600 
6601 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6602 			metaslab_t *msp = vd->vdev_ms[m];
6603 
6604 			(void) fprintf(stderr,
6605 			    "\rloading concrete vdev %llu, "
6606 			    "metaslab %llu of %llu ...",
6607 			    (longlong_t)vd->vdev_id,
6608 			    (longlong_t)msp->ms_id,
6609 			    (longlong_t)vd->vdev_ms_count);
6610 
6611 			mutex_enter(&msp->ms_lock);
6612 			range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6613 
6614 			/*
6615 			 * We don't want to spend the CPU manipulating the
6616 			 * size-ordered tree, so clear the range_tree ops.
6617 			 */
6618 			msp->ms_allocatable->rt_ops = NULL;
6619 
6620 			if (msp->ms_sm != NULL) {
6621 				VERIFY0(space_map_load(msp->ms_sm,
6622 				    msp->ms_allocatable, maptype));
6623 			}
6624 			if (!msp->ms_loaded)
6625 				msp->ms_loaded = B_TRUE;
6626 			mutex_exit(&msp->ms_lock);
6627 		}
6628 	}
6629 
6630 	load_unflushed_to_ms_allocatables(spa, maptype);
6631 }
6632 
6633 /*
6634  * vm_idxp is an in-out parameter which (for indirect vdevs) is the
6635  * index in vim_entries that has the first entry in this metaslab.
6636  * On return, it will be set to the first entry after this metaslab.
6637  */
6638 static void
load_indirect_ms_allocatable_tree(vdev_t * vd,metaslab_t * msp,uint64_t * vim_idxp)6639 load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab_t *msp,
6640     uint64_t *vim_idxp)
6641 {
6642 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6643 
6644 	mutex_enter(&msp->ms_lock);
6645 	range_tree_vacate(msp->ms_allocatable, NULL, NULL);
6646 
6647 	/*
6648 	 * We don't want to spend the CPU manipulating the
6649 	 * size-ordered tree, so clear the range_tree ops.
6650 	 */
6651 	msp->ms_allocatable->rt_ops = NULL;
6652 
6653 	for (; *vim_idxp < vdev_indirect_mapping_num_entries(vim);
6654 	    (*vim_idxp)++) {
6655 		vdev_indirect_mapping_entry_phys_t *vimep =
6656 		    &vim->vim_entries[*vim_idxp];
6657 		uint64_t ent_offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6658 		uint64_t ent_len = DVA_GET_ASIZE(&vimep->vimep_dst);
6659 		ASSERT3U(ent_offset, >=, msp->ms_start);
6660 		if (ent_offset >= msp->ms_start + msp->ms_size)
6661 			break;
6662 
6663 		/*
6664 		 * Mappings do not cross metaslab boundaries,
6665 		 * because we create them by walking the metaslabs.
6666 		 */
6667 		ASSERT3U(ent_offset + ent_len, <=,
6668 		    msp->ms_start + msp->ms_size);
6669 		range_tree_add(msp->ms_allocatable, ent_offset, ent_len);
6670 	}
6671 
6672 	if (!msp->ms_loaded)
6673 		msp->ms_loaded = B_TRUE;
6674 	mutex_exit(&msp->ms_lock);
6675 }
6676 
6677 static void
zdb_leak_init_prepare_indirect_vdevs(spa_t * spa,zdb_cb_t * zcb)6678 zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb)
6679 {
6680 	ASSERT(!dump_opt['L']);
6681 
6682 	vdev_t *rvd = spa->spa_root_vdev;
6683 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
6684 		vdev_t *vd = rvd->vdev_child[c];
6685 
6686 		ASSERT3U(c, ==, vd->vdev_id);
6687 
6688 		if (vd->vdev_ops != &vdev_indirect_ops)
6689 			continue;
6690 
6691 		/*
6692 		 * Note: we don't check for mapping leaks on
6693 		 * removing vdevs because their ms_allocatable's
6694 		 * are used to look for leaks in allocated space.
6695 		 */
6696 		zcb->zcb_vd_obsolete_counts[c] = zdb_load_obsolete_counts(vd);
6697 
6698 		/*
6699 		 * Normally, indirect vdevs don't have any
6700 		 * metaslabs.  We want to set them up for
6701 		 * zio_claim().
6702 		 */
6703 		vdev_metaslab_group_create(vd);
6704 		VERIFY0(vdev_metaslab_init(vd, 0));
6705 
6706 		vdev_indirect_mapping_t *vim __maybe_unused =
6707 		    vd->vdev_indirect_mapping;
6708 		uint64_t vim_idx = 0;
6709 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6710 
6711 			(void) fprintf(stderr,
6712 			    "\rloading indirect vdev %llu, "
6713 			    "metaslab %llu of %llu ...",
6714 			    (longlong_t)vd->vdev_id,
6715 			    (longlong_t)vd->vdev_ms[m]->ms_id,
6716 			    (longlong_t)vd->vdev_ms_count);
6717 
6718 			load_indirect_ms_allocatable_tree(vd, vd->vdev_ms[m],
6719 			    &vim_idx);
6720 		}
6721 		ASSERT3U(vim_idx, ==, vdev_indirect_mapping_num_entries(vim));
6722 	}
6723 }
6724 
6725 static void
zdb_leak_init(spa_t * spa,zdb_cb_t * zcb)6726 zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
6727 {
6728 	zcb->zcb_spa = spa;
6729 
6730 	if (dump_opt['L'])
6731 		return;
6732 
6733 	dsl_pool_t *dp = spa->spa_dsl_pool;
6734 	vdev_t *rvd = spa->spa_root_vdev;
6735 
6736 	/*
6737 	 * We are going to be changing the meaning of the metaslab's
6738 	 * ms_allocatable.  Ensure that the allocator doesn't try to
6739 	 * use the tree.
6740 	 */
6741 	spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
6742 	spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
6743 	spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops;
6744 
6745 	zcb->zcb_vd_obsolete_counts =
6746 	    umem_zalloc(rvd->vdev_children * sizeof (uint32_t *),
6747 	    UMEM_NOFAIL);
6748 
6749 	/*
6750 	 * For leak detection, we overload the ms_allocatable trees
6751 	 * to contain allocated segments instead of free segments.
6752 	 * As a result, we can't use the normal metaslab_load/unload
6753 	 * interfaces.
6754 	 */
6755 	zdb_leak_init_prepare_indirect_vdevs(spa, zcb);
6756 	load_concrete_ms_allocatable_trees(spa, SM_ALLOC);
6757 
6758 	/*
6759 	 * On load_concrete_ms_allocatable_trees() we loaded all the
6760 	 * allocated entries from the ms_sm to the ms_allocatable for
6761 	 * each metaslab. If the pool has a checkpoint or is in the
6762 	 * middle of discarding a checkpoint, some of these blocks
6763 	 * may have been freed but their ms_sm may not have been
6764 	 * updated because they are referenced by the checkpoint. In
6765 	 * order to avoid false-positives during leak-detection, we
6766 	 * go through the vdev's checkpoint space map and exclude all
6767 	 * its entries from their relevant ms_allocatable.
6768 	 *
6769 	 * We also aggregate the space held by the checkpoint and add
6770 	 * it to zcb_checkpoint_size.
6771 	 *
6772 	 * Note that at this point we are also verifying that all the
6773 	 * entries on the checkpoint_sm are marked as allocated in
6774 	 * the ms_sm of their relevant metaslab.
6775 	 * [see comment in checkpoint_sm_exclude_entry_cb()]
6776 	 */
6777 	zdb_leak_init_exclude_checkpoint(spa, zcb);
6778 	ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa));
6779 
6780 	/* for cleaner progress output */
6781 	(void) fprintf(stderr, "\n");
6782 
6783 	if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
6784 		ASSERT(spa_feature_is_enabled(spa,
6785 		    SPA_FEATURE_DEVICE_REMOVAL));
6786 		(void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj,
6787 		    increment_indirect_mapping_cb, zcb, NULL);
6788 	}
6789 }
6790 
6791 static boolean_t
zdb_check_for_obsolete_leaks(vdev_t * vd,zdb_cb_t * zcb)6792 zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb)
6793 {
6794 	boolean_t leaks = B_FALSE;
6795 	vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
6796 	uint64_t total_leaked = 0;
6797 	boolean_t are_precise = B_FALSE;
6798 
6799 	ASSERT(vim != NULL);
6800 
6801 	for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) {
6802 		vdev_indirect_mapping_entry_phys_t *vimep =
6803 		    &vim->vim_entries[i];
6804 		uint64_t obsolete_bytes = 0;
6805 		uint64_t offset = DVA_MAPPING_GET_SRC_OFFSET(vimep);
6806 		metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
6807 
6808 		/*
6809 		 * This is not very efficient but it's easy to
6810 		 * verify correctness.
6811 		 */
6812 		for (uint64_t inner_offset = 0;
6813 		    inner_offset < DVA_GET_ASIZE(&vimep->vimep_dst);
6814 		    inner_offset += 1ULL << vd->vdev_ashift) {
6815 			if (range_tree_contains(msp->ms_allocatable,
6816 			    offset + inner_offset, 1ULL << vd->vdev_ashift)) {
6817 				obsolete_bytes += 1ULL << vd->vdev_ashift;
6818 			}
6819 		}
6820 
6821 		int64_t bytes_leaked = obsolete_bytes -
6822 		    zcb->zcb_vd_obsolete_counts[vd->vdev_id][i];
6823 		ASSERT3U(DVA_GET_ASIZE(&vimep->vimep_dst), >=,
6824 		    zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]);
6825 
6826 		VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6827 		if (bytes_leaked != 0 && (are_precise || dump_opt['d'] >= 5)) {
6828 			(void) printf("obsolete indirect mapping count "
6829 			    "mismatch on %llu:%llx:%llx : %llx bytes leaked\n",
6830 			    (u_longlong_t)vd->vdev_id,
6831 			    (u_longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep),
6832 			    (u_longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst),
6833 			    (u_longlong_t)bytes_leaked);
6834 		}
6835 		total_leaked += ABS(bytes_leaked);
6836 	}
6837 
6838 	VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
6839 	if (!are_precise && total_leaked > 0) {
6840 		int pct_leaked = total_leaked * 100 /
6841 		    vdev_indirect_mapping_bytes_mapped(vim);
6842 		(void) printf("cannot verify obsolete indirect mapping "
6843 		    "counts of vdev %llu because precise feature was not "
6844 		    "enabled when it was removed: %d%% (%llx bytes) of mapping"
6845 		    "unreferenced\n",
6846 		    (u_longlong_t)vd->vdev_id, pct_leaked,
6847 		    (u_longlong_t)total_leaked);
6848 	} else if (total_leaked > 0) {
6849 		(void) printf("obsolete indirect mapping count mismatch "
6850 		    "for vdev %llu -- %llx total bytes mismatched\n",
6851 		    (u_longlong_t)vd->vdev_id,
6852 		    (u_longlong_t)total_leaked);
6853 		leaks |= B_TRUE;
6854 	}
6855 
6856 	vdev_indirect_mapping_free_obsolete_counts(vim,
6857 	    zcb->zcb_vd_obsolete_counts[vd->vdev_id]);
6858 	zcb->zcb_vd_obsolete_counts[vd->vdev_id] = NULL;
6859 
6860 	return (leaks);
6861 }
6862 
6863 static boolean_t
zdb_leak_fini(spa_t * spa,zdb_cb_t * zcb)6864 zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb)
6865 {
6866 	if (dump_opt['L'])
6867 		return (B_FALSE);
6868 
6869 	boolean_t leaks = B_FALSE;
6870 	vdev_t *rvd = spa->spa_root_vdev;
6871 	for (unsigned c = 0; c < rvd->vdev_children; c++) {
6872 		vdev_t *vd = rvd->vdev_child[c];
6873 
6874 		if (zcb->zcb_vd_obsolete_counts[c] != NULL) {
6875 			leaks |= zdb_check_for_obsolete_leaks(vd, zcb);
6876 		}
6877 
6878 		for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
6879 			metaslab_t *msp = vd->vdev_ms[m];
6880 			ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class ==
6881 			    spa_embedded_log_class(spa)) ?
6882 			    vd->vdev_log_mg : vd->vdev_mg);
6883 
6884 			/*
6885 			 * ms_allocatable has been overloaded
6886 			 * to contain allocated segments. Now that
6887 			 * we finished traversing all blocks, any
6888 			 * block that remains in the ms_allocatable
6889 			 * represents an allocated block that we
6890 			 * did not claim during the traversal.
6891 			 * Claimed blocks would have been removed
6892 			 * from the ms_allocatable.  For indirect
6893 			 * vdevs, space remaining in the tree
6894 			 * represents parts of the mapping that are
6895 			 * not referenced, which is not a bug.
6896 			 */
6897 			if (vd->vdev_ops == &vdev_indirect_ops) {
6898 				range_tree_vacate(msp->ms_allocatable,
6899 				    NULL, NULL);
6900 			} else {
6901 				range_tree_vacate(msp->ms_allocatable,
6902 				    zdb_leak, vd);
6903 			}
6904 			if (msp->ms_loaded) {
6905 				msp->ms_loaded = B_FALSE;
6906 			}
6907 		}
6908 	}
6909 
6910 	umem_free(zcb->zcb_vd_obsolete_counts,
6911 	    rvd->vdev_children * sizeof (uint32_t *));
6912 	zcb->zcb_vd_obsolete_counts = NULL;
6913 
6914 	return (leaks);
6915 }
6916 
6917 static int
count_block_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)6918 count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6919 {
6920 	(void) tx;
6921 	zdb_cb_t *zcb = arg;
6922 
6923 	if (dump_opt['b'] >= 5) {
6924 		char blkbuf[BP_SPRINTF_LEN];
6925 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
6926 		(void) printf("[%s] %s\n",
6927 		    "deferred free", blkbuf);
6928 	}
6929 	zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED);
6930 	return (0);
6931 }
6932 
6933 /*
6934  * Iterate over livelists which have been destroyed by the user but
6935  * are still present in the MOS, waiting to be freed
6936  */
6937 static void
iterate_deleted_livelists(spa_t * spa,ll_iter_t func,void * arg)6938 iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg)
6939 {
6940 	objset_t *mos = spa->spa_meta_objset;
6941 	uint64_t zap_obj;
6942 	int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
6943 	    DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
6944 	if (err == ENOENT)
6945 		return;
6946 	ASSERT0(err);
6947 
6948 	zap_cursor_t zc;
6949 	zap_attribute_t *attrp = zap_attribute_alloc();
6950 	dsl_deadlist_t ll;
6951 	/* NULL out os prior to dsl_deadlist_open in case it's garbage */
6952 	ll.dl_os = NULL;
6953 	for (zap_cursor_init(&zc, mos, zap_obj);
6954 	    zap_cursor_retrieve(&zc, attrp) == 0;
6955 	    (void) zap_cursor_advance(&zc)) {
6956 		VERIFY0(dsl_deadlist_open(&ll, mos, attrp->za_first_integer));
6957 		func(&ll, arg);
6958 		dsl_deadlist_close(&ll);
6959 	}
6960 	zap_cursor_fini(&zc);
6961 	zap_attribute_free(attrp);
6962 }
6963 
6964 static int
bpobj_count_block_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)6965 bpobj_count_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
6966     dmu_tx_t *tx)
6967 {
6968 	ASSERT(!bp_freed);
6969 	return (count_block_cb(arg, bp, tx));
6970 }
6971 
6972 static int
livelist_entry_count_blocks_cb(void * args,dsl_deadlist_entry_t * dle)6973 livelist_entry_count_blocks_cb(void *args, dsl_deadlist_entry_t *dle)
6974 {
6975 	zdb_cb_t *zbc = args;
6976 	bplist_t blks;
6977 	bplist_create(&blks);
6978 	/* determine which blocks have been alloc'd but not freed */
6979 	VERIFY0(dsl_process_sub_livelist(&dle->dle_bpobj, &blks, NULL, NULL));
6980 	/* count those blocks */
6981 	(void) bplist_iterate(&blks, count_block_cb, zbc, NULL);
6982 	bplist_destroy(&blks);
6983 	return (0);
6984 }
6985 
6986 static void
livelist_count_blocks(dsl_deadlist_t * ll,void * arg)6987 livelist_count_blocks(dsl_deadlist_t *ll, void *arg)
6988 {
6989 	dsl_deadlist_iterate(ll, livelist_entry_count_blocks_cb, arg);
6990 }
6991 
6992 /*
6993  * Count the blocks in the livelists that have been destroyed by the user
6994  * but haven't yet been freed.
6995  */
6996 static void
deleted_livelists_count_blocks(spa_t * spa,zdb_cb_t * zbc)6997 deleted_livelists_count_blocks(spa_t *spa, zdb_cb_t *zbc)
6998 {
6999 	iterate_deleted_livelists(spa, livelist_count_blocks, zbc);
7000 }
7001 
7002 static void
dump_livelist_cb(dsl_deadlist_t * ll,void * arg)7003 dump_livelist_cb(dsl_deadlist_t *ll, void *arg)
7004 {
7005 	ASSERT3P(arg, ==, NULL);
7006 	global_feature_count[SPA_FEATURE_LIVELIST]++;
7007 	dump_blkptr_list(ll, "Deleted Livelist");
7008 	dsl_deadlist_iterate(ll, sublivelist_verify_lightweight, NULL);
7009 }
7010 
7011 /*
7012  * Print out, register object references to, and increment feature counts for
7013  * livelists that have been destroyed by the user but haven't yet been freed.
7014  */
7015 static void
deleted_livelists_dump_mos(spa_t * spa)7016 deleted_livelists_dump_mos(spa_t *spa)
7017 {
7018 	uint64_t zap_obj;
7019 	objset_t *mos = spa->spa_meta_objset;
7020 	int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
7021 	    DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
7022 	if (err == ENOENT)
7023 		return;
7024 	mos_obj_refd(zap_obj);
7025 	iterate_deleted_livelists(spa, dump_livelist_cb, NULL);
7026 }
7027 
7028 static int
zdb_brt_entry_compare(const void * zcn1,const void * zcn2)7029 zdb_brt_entry_compare(const void *zcn1, const void *zcn2)
7030 {
7031 	const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva;
7032 	const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva;
7033 	int cmp;
7034 
7035 	cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
7036 	if (cmp == 0)
7037 		cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2));
7038 
7039 	return (cmp);
7040 }
7041 
7042 static int
dump_block_stats(spa_t * spa)7043 dump_block_stats(spa_t *spa)
7044 {
7045 	zdb_cb_t *zcb;
7046 	zdb_blkstats_t *zb, *tzb;
7047 	uint64_t norm_alloc, norm_space, total_alloc, total_found;
7048 	int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7049 	    TRAVERSE_NO_DECRYPT | TRAVERSE_HARD;
7050 	boolean_t leaks = B_FALSE;
7051 	int e, c, err;
7052 	bp_embedded_type_t i;
7053 
7054 	ddt_prefetch_all(spa);
7055 
7056 	zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL);
7057 
7058 	if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
7059 		avl_create(&zcb->zcb_brt, zdb_brt_entry_compare,
7060 		    sizeof (zdb_brt_entry_t),
7061 		    offsetof(zdb_brt_entry_t, zbre_node));
7062 		zcb->zcb_brt_is_active = B_TRUE;
7063 	}
7064 
7065 	(void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
7066 	    (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
7067 	    (dump_opt['c'] == 1) ? "metadata " : "",
7068 	    dump_opt['c'] ? "checksums " : "",
7069 	    (dump_opt['c'] && !dump_opt['L']) ? "and verify " : "",
7070 	    !dump_opt['L'] ? "nothing leaked " : "");
7071 
7072 	/*
7073 	 * When leak detection is enabled we load all space maps as SM_ALLOC
7074 	 * maps, then traverse the pool claiming each block we discover. If
7075 	 * the pool is perfectly consistent, the segment trees will be empty
7076 	 * when we're done. Anything left over is a leak; any block we can't
7077 	 * claim (because it's not part of any space map) is a double
7078 	 * allocation, reference to a freed block, or an unclaimed log block.
7079 	 *
7080 	 * When leak detection is disabled (-L option) we still traverse the
7081 	 * pool claiming each block we discover, but we skip opening any space
7082 	 * maps.
7083 	 */
7084 	zdb_leak_init(spa, zcb);
7085 
7086 	/*
7087 	 * If there's a deferred-free bplist, process that first.
7088 	 */
7089 	(void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
7090 	    bpobj_count_block_cb, zcb, NULL);
7091 
7092 	if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
7093 		(void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
7094 		    bpobj_count_block_cb, zcb, NULL);
7095 	}
7096 
7097 	zdb_claim_removing(spa, zcb);
7098 
7099 	if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
7100 		VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,
7101 		    spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb,
7102 		    zcb, NULL));
7103 	}
7104 
7105 	deleted_livelists_count_blocks(spa, zcb);
7106 
7107 	if (dump_opt['c'] > 1)
7108 		flags |= TRAVERSE_PREFETCH_DATA;
7109 
7110 	zcb->zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
7111 	zcb->zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa));
7112 	zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
7113 	zcb->zcb_totalasize +=
7114 	    metaslab_class_get_alloc(spa_embedded_log_class(spa));
7115 	zcb->zcb_start = zcb->zcb_lastprint = gethrtime();
7116 	err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb);
7117 
7118 	/*
7119 	 * If we've traversed the data blocks then we need to wait for those
7120 	 * I/Os to complete. We leverage "The Godfather" zio to wait on
7121 	 * all async I/Os to complete.
7122 	 */
7123 	if (dump_opt['c']) {
7124 		for (c = 0; c < max_ncpus; c++) {
7125 			(void) zio_wait(spa->spa_async_zio_root[c]);
7126 			spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL,
7127 			    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
7128 			    ZIO_FLAG_GODFATHER);
7129 		}
7130 	}
7131 	ASSERT0(spa->spa_load_verify_bytes);
7132 
7133 	/*
7134 	 * Done after zio_wait() since zcb_haderrors is modified in
7135 	 * zdb_blkptr_done()
7136 	 */
7137 	zcb->zcb_haderrors |= err;
7138 
7139 	if (zcb->zcb_haderrors) {
7140 		(void) printf("\nError counts:\n\n");
7141 		(void) printf("\t%5s  %s\n", "errno", "count");
7142 		for (e = 0; e < 256; e++) {
7143 			if (zcb->zcb_errors[e] != 0) {
7144 				(void) printf("\t%5d  %llu\n",
7145 				    e, (u_longlong_t)zcb->zcb_errors[e]);
7146 			}
7147 		}
7148 	}
7149 
7150 	/*
7151 	 * Report any leaked segments.
7152 	 */
7153 	leaks |= zdb_leak_fini(spa, zcb);
7154 
7155 	tzb = &zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL];
7156 
7157 	norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7158 	norm_space = metaslab_class_get_space(spa_normal_class(spa));
7159 
7160 	total_alloc = norm_alloc +
7161 	    metaslab_class_get_alloc(spa_log_class(spa)) +
7162 	    metaslab_class_get_alloc(spa_embedded_log_class(spa)) +
7163 	    metaslab_class_get_alloc(spa_special_class(spa)) +
7164 	    metaslab_class_get_alloc(spa_dedup_class(spa)) +
7165 	    get_unflushed_alloc_space(spa);
7166 	total_found =
7167 	    tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize +
7168 	    zcb->zcb_removing_size + zcb->zcb_checkpoint_size;
7169 
7170 	if (total_found == total_alloc && !dump_opt['L']) {
7171 		(void) printf("\n\tNo leaks (block sum matches space"
7172 		    " maps exactly)\n");
7173 	} else if (!dump_opt['L']) {
7174 		(void) printf("block traversal size %llu != alloc %llu "
7175 		    "(%s %lld)\n",
7176 		    (u_longlong_t)total_found,
7177 		    (u_longlong_t)total_alloc,
7178 		    (dump_opt['L']) ? "unreachable" : "leaked",
7179 		    (longlong_t)(total_alloc - total_found));
7180 	}
7181 
7182 	if (tzb->zb_count == 0) {
7183 		umem_free(zcb, sizeof (zdb_cb_t));
7184 		return (2);
7185 	}
7186 
7187 	(void) printf("\n");
7188 	(void) printf("\t%-16s %14llu\n", "bp count:",
7189 	    (u_longlong_t)tzb->zb_count);
7190 	(void) printf("\t%-16s %14llu\n", "ganged count:",
7191 	    (longlong_t)tzb->zb_gangs);
7192 	(void) printf("\t%-16s %14llu      avg: %6llu\n", "bp logical:",
7193 	    (u_longlong_t)tzb->zb_lsize,
7194 	    (u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
7195 	(void) printf("\t%-16s %14llu      avg: %6llu     compression: %6.2f\n",
7196 	    "bp physical:", (u_longlong_t)tzb->zb_psize,
7197 	    (u_longlong_t)(tzb->zb_psize / tzb->zb_count),
7198 	    (double)tzb->zb_lsize / tzb->zb_psize);
7199 	(void) printf("\t%-16s %14llu      avg: %6llu     compression: %6.2f\n",
7200 	    "bp allocated:", (u_longlong_t)tzb->zb_asize,
7201 	    (u_longlong_t)(tzb->zb_asize / tzb->zb_count),
7202 	    (double)tzb->zb_lsize / tzb->zb_asize);
7203 	(void) printf("\t%-16s %14llu    ref>1: %6llu   deduplication: %6.2f\n",
7204 	    "bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize,
7205 	    (u_longlong_t)zcb->zcb_dedup_blocks,
7206 	    (double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0);
7207 	(void) printf("\t%-16s %14llu    count: %6llu\n",
7208 	    "bp cloned:", (u_longlong_t)zcb->zcb_clone_asize,
7209 	    (u_longlong_t)zcb->zcb_clone_blocks);
7210 	(void) printf("\t%-16s %14llu     used: %5.2f%%\n", "Normal class:",
7211 	    (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
7212 
7213 	if (spa_special_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7214 		uint64_t alloc = metaslab_class_get_alloc(
7215 		    spa_special_class(spa));
7216 		uint64_t space = metaslab_class_get_space(
7217 		    spa_special_class(spa));
7218 
7219 		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
7220 		    "Special class", (u_longlong_t)alloc,
7221 		    100.0 * alloc / space);
7222 	}
7223 
7224 	if (spa_dedup_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7225 		uint64_t alloc = metaslab_class_get_alloc(
7226 		    spa_dedup_class(spa));
7227 		uint64_t space = metaslab_class_get_space(
7228 		    spa_dedup_class(spa));
7229 
7230 		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
7231 		    "Dedup class", (u_longlong_t)alloc,
7232 		    100.0 * alloc / space);
7233 	}
7234 
7235 	if (spa_embedded_log_class(spa)->mc_allocator[0].mca_rotor != NULL) {
7236 		uint64_t alloc = metaslab_class_get_alloc(
7237 		    spa_embedded_log_class(spa));
7238 		uint64_t space = metaslab_class_get_space(
7239 		    spa_embedded_log_class(spa));
7240 
7241 		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
7242 		    "Embedded log class", (u_longlong_t)alloc,
7243 		    100.0 * alloc / space);
7244 	}
7245 
7246 	for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
7247 		if (zcb->zcb_embedded_blocks[i] == 0)
7248 			continue;
7249 		(void) printf("\n");
7250 		(void) printf("\tadditional, non-pointer bps of type %u: "
7251 		    "%10llu\n",
7252 		    i, (u_longlong_t)zcb->zcb_embedded_blocks[i]);
7253 
7254 		if (dump_opt['b'] >= 3) {
7255 			(void) printf("\t number of (compressed) bytes:  "
7256 			    "number of bps\n");
7257 			dump_histogram(zcb->zcb_embedded_histogram[i],
7258 			    sizeof (zcb->zcb_embedded_histogram[i]) /
7259 			    sizeof (zcb->zcb_embedded_histogram[i][0]), 0);
7260 		}
7261 	}
7262 
7263 	if (tzb->zb_ditto_samevdev != 0) {
7264 		(void) printf("\tDittoed blocks on same vdev: %llu\n",
7265 		    (longlong_t)tzb->zb_ditto_samevdev);
7266 	}
7267 	if (tzb->zb_ditto_same_ms != 0) {
7268 		(void) printf("\tDittoed blocks in same metaslab: %llu\n",
7269 		    (longlong_t)tzb->zb_ditto_same_ms);
7270 	}
7271 
7272 	for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) {
7273 		vdev_t *vd = spa->spa_root_vdev->vdev_child[v];
7274 		vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7275 
7276 		if (vim == NULL) {
7277 			continue;
7278 		}
7279 
7280 		char mem[32];
7281 		zdb_nicenum(vdev_indirect_mapping_num_entries(vim),
7282 		    mem, vdev_indirect_mapping_size(vim));
7283 
7284 		(void) printf("\tindirect vdev id %llu has %llu segments "
7285 		    "(%s in memory)\n",
7286 		    (longlong_t)vd->vdev_id,
7287 		    (longlong_t)vdev_indirect_mapping_num_entries(vim), mem);
7288 	}
7289 
7290 	if (dump_opt['b'] >= 2) {
7291 		int l, t, level;
7292 		char csize[32], lsize[32], psize[32], asize[32];
7293 		char avg[32], gang[32];
7294 		(void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
7295 		    "\t  avg\t comp\t%%Total\tType\n");
7296 
7297 		zfs_blkstat_t *mdstats = umem_zalloc(sizeof (zfs_blkstat_t),
7298 		    UMEM_NOFAIL);
7299 
7300 		for (t = 0; t <= ZDB_OT_TOTAL; t++) {
7301 			const char *typename;
7302 
7303 			/* make sure nicenum has enough space */
7304 			_Static_assert(sizeof (csize) >= NN_NUMBUF_SZ,
7305 			    "csize truncated");
7306 			_Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ,
7307 			    "lsize truncated");
7308 			_Static_assert(sizeof (psize) >= NN_NUMBUF_SZ,
7309 			    "psize truncated");
7310 			_Static_assert(sizeof (asize) >= NN_NUMBUF_SZ,
7311 			    "asize truncated");
7312 			_Static_assert(sizeof (avg) >= NN_NUMBUF_SZ,
7313 			    "avg truncated");
7314 			_Static_assert(sizeof (gang) >= NN_NUMBUF_SZ,
7315 			    "gang truncated");
7316 
7317 			if (t < DMU_OT_NUMTYPES)
7318 				typename = dmu_ot[t].ot_name;
7319 			else
7320 				typename = zdb_ot_extname[t - DMU_OT_NUMTYPES];
7321 
7322 			if (zcb->zcb_type[ZB_TOTAL][t].zb_asize == 0) {
7323 				(void) printf("%6s\t%5s\t%5s\t%5s"
7324 				    "\t%5s\t%5s\t%6s\t%s\n",
7325 				    "-",
7326 				    "-",
7327 				    "-",
7328 				    "-",
7329 				    "-",
7330 				    "-",
7331 				    "-",
7332 				    typename);
7333 				continue;
7334 			}
7335 
7336 			for (l = ZB_TOTAL - 1; l >= -1; l--) {
7337 				level = (l == -1 ? ZB_TOTAL : l);
7338 				zb = &zcb->zcb_type[level][t];
7339 
7340 				if (zb->zb_asize == 0)
7341 					continue;
7342 
7343 				if (level != ZB_TOTAL && t < DMU_OT_NUMTYPES &&
7344 				    (level > 0 || DMU_OT_IS_METADATA(t))) {
7345 					mdstats->zb_count += zb->zb_count;
7346 					mdstats->zb_lsize += zb->zb_lsize;
7347 					mdstats->zb_psize += zb->zb_psize;
7348 					mdstats->zb_asize += zb->zb_asize;
7349 					mdstats->zb_gangs += zb->zb_gangs;
7350 				}
7351 
7352 				if (dump_opt['b'] < 3 && level != ZB_TOTAL)
7353 					continue;
7354 
7355 				if (level == 0 && zb->zb_asize ==
7356 				    zcb->zcb_type[ZB_TOTAL][t].zb_asize)
7357 					continue;
7358 
7359 				zdb_nicenum(zb->zb_count, csize,
7360 				    sizeof (csize));
7361 				zdb_nicenum(zb->zb_lsize, lsize,
7362 				    sizeof (lsize));
7363 				zdb_nicenum(zb->zb_psize, psize,
7364 				    sizeof (psize));
7365 				zdb_nicenum(zb->zb_asize, asize,
7366 				    sizeof (asize));
7367 				zdb_nicenum(zb->zb_asize / zb->zb_count, avg,
7368 				    sizeof (avg));
7369 				zdb_nicenum(zb->zb_gangs, gang, sizeof (gang));
7370 
7371 				(void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7372 				    "\t%5.2f\t%6.2f\t",
7373 				    csize, lsize, psize, asize, avg,
7374 				    (double)zb->zb_lsize / zb->zb_psize,
7375 				    100.0 * zb->zb_asize / tzb->zb_asize);
7376 
7377 				if (level == ZB_TOTAL)
7378 					(void) printf("%s\n", typename);
7379 				else
7380 					(void) printf("    L%d %s\n",
7381 					    level, typename);
7382 
7383 				if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) {
7384 					(void) printf("\t number of ganged "
7385 					    "blocks: %s\n", gang);
7386 				}
7387 
7388 				if (dump_opt['b'] >= 4) {
7389 					(void) printf("psize "
7390 					    "(in 512-byte sectors): "
7391 					    "number of blocks\n");
7392 					dump_histogram(zb->zb_psize_histogram,
7393 					    PSIZE_HISTO_SIZE, 0);
7394 				}
7395 			}
7396 		}
7397 		zdb_nicenum(mdstats->zb_count, csize,
7398 		    sizeof (csize));
7399 		zdb_nicenum(mdstats->zb_lsize, lsize,
7400 		    sizeof (lsize));
7401 		zdb_nicenum(mdstats->zb_psize, psize,
7402 		    sizeof (psize));
7403 		zdb_nicenum(mdstats->zb_asize, asize,
7404 		    sizeof (asize));
7405 		zdb_nicenum(mdstats->zb_asize / mdstats->zb_count, avg,
7406 		    sizeof (avg));
7407 		zdb_nicenum(mdstats->zb_gangs, gang, sizeof (gang));
7408 
7409 		(void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
7410 		    "\t%5.2f\t%6.2f\t",
7411 		    csize, lsize, psize, asize, avg,
7412 		    (double)mdstats->zb_lsize / mdstats->zb_psize,
7413 		    100.0 * mdstats->zb_asize / tzb->zb_asize);
7414 		(void) printf("%s\n", "Metadata Total");
7415 
7416 		/* Output a table summarizing block sizes in the pool */
7417 		if (dump_opt['b'] >= 2) {
7418 			dump_size_histograms(zcb);
7419 		}
7420 
7421 		umem_free(mdstats, sizeof (zfs_blkstat_t));
7422 	}
7423 
7424 	(void) printf("\n");
7425 
7426 	if (leaks) {
7427 		umem_free(zcb, sizeof (zdb_cb_t));
7428 		return (2);
7429 	}
7430 
7431 	if (zcb->zcb_haderrors) {
7432 		umem_free(zcb, sizeof (zdb_cb_t));
7433 		return (3);
7434 	}
7435 
7436 	umem_free(zcb, sizeof (zdb_cb_t));
7437 	return (0);
7438 }
7439 
7440 typedef struct zdb_ddt_entry {
7441 	/* key must be first for ddt_key_compare */
7442 	ddt_key_t	zdde_key;
7443 	uint64_t	zdde_ref_blocks;
7444 	uint64_t	zdde_ref_lsize;
7445 	uint64_t	zdde_ref_psize;
7446 	uint64_t	zdde_ref_dsize;
7447 	avl_node_t	zdde_node;
7448 } zdb_ddt_entry_t;
7449 
7450 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)7451 zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
7452     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
7453 {
7454 	(void) zilog, (void) dnp;
7455 	avl_tree_t *t = arg;
7456 	avl_index_t where;
7457 	zdb_ddt_entry_t *zdde, zdde_search;
7458 
7459 	if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
7460 	    BP_IS_EMBEDDED(bp))
7461 		return (0);
7462 
7463 	if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
7464 		(void) printf("traversing objset %llu, %llu objects, "
7465 		    "%lu blocks so far\n",
7466 		    (u_longlong_t)zb->zb_objset,
7467 		    (u_longlong_t)BP_GET_FILL(bp),
7468 		    avl_numnodes(t));
7469 	}
7470 
7471 	if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF ||
7472 	    BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
7473 		return (0);
7474 
7475 	ddt_key_fill(&zdde_search.zdde_key, bp);
7476 
7477 	zdde = avl_find(t, &zdde_search, &where);
7478 
7479 	if (zdde == NULL) {
7480 		zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL);
7481 		zdde->zdde_key = zdde_search.zdde_key;
7482 		avl_insert(t, zdde, where);
7483 	}
7484 
7485 	zdde->zdde_ref_blocks += 1;
7486 	zdde->zdde_ref_lsize += BP_GET_LSIZE(bp);
7487 	zdde->zdde_ref_psize += BP_GET_PSIZE(bp);
7488 	zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp);
7489 
7490 	return (0);
7491 }
7492 
7493 static void
dump_simulated_ddt(spa_t * spa)7494 dump_simulated_ddt(spa_t *spa)
7495 {
7496 	avl_tree_t t;
7497 	void *cookie = NULL;
7498 	zdb_ddt_entry_t *zdde;
7499 	ddt_histogram_t ddh_total = {{{0}}};
7500 	ddt_stat_t dds_total = {0};
7501 
7502 	avl_create(&t, ddt_key_compare,
7503 	    sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
7504 
7505 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7506 
7507 	(void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
7508 	    TRAVERSE_NO_DECRYPT, zdb_ddt_add_cb, &t);
7509 
7510 	spa_config_exit(spa, SCL_CONFIG, FTAG);
7511 
7512 	while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) {
7513 		uint64_t refcnt = zdde->zdde_ref_blocks;
7514 		ASSERT(refcnt != 0);
7515 
7516 		ddt_stat_t *dds = &ddh_total.ddh_stat[highbit64(refcnt) - 1];
7517 
7518 		dds->dds_blocks += zdde->zdde_ref_blocks / refcnt;
7519 		dds->dds_lsize += zdde->zdde_ref_lsize / refcnt;
7520 		dds->dds_psize += zdde->zdde_ref_psize / refcnt;
7521 		dds->dds_dsize += zdde->zdde_ref_dsize / refcnt;
7522 
7523 		dds->dds_ref_blocks += zdde->zdde_ref_blocks;
7524 		dds->dds_ref_lsize += zdde->zdde_ref_lsize;
7525 		dds->dds_ref_psize += zdde->zdde_ref_psize;
7526 		dds->dds_ref_dsize += zdde->zdde_ref_dsize;
7527 
7528 		umem_free(zdde, sizeof (*zdde));
7529 	}
7530 
7531 	avl_destroy(&t);
7532 
7533 	ddt_histogram_total(&dds_total, &ddh_total);
7534 
7535 	(void) printf("Simulated DDT histogram:\n");
7536 
7537 	zpool_dump_ddt(&dds_total, &ddh_total);
7538 
7539 	dump_dedup_ratio(&dds_total);
7540 }
7541 
7542 static int
verify_device_removal_feature_counts(spa_t * spa)7543 verify_device_removal_feature_counts(spa_t *spa)
7544 {
7545 	uint64_t dr_feature_refcount = 0;
7546 	uint64_t oc_feature_refcount = 0;
7547 	uint64_t indirect_vdev_count = 0;
7548 	uint64_t precise_vdev_count = 0;
7549 	uint64_t obsolete_counts_object_count = 0;
7550 	uint64_t obsolete_sm_count = 0;
7551 	uint64_t obsolete_counts_count = 0;
7552 	uint64_t scip_count = 0;
7553 	uint64_t obsolete_bpobj_count = 0;
7554 	int ret = 0;
7555 
7556 	spa_condensing_indirect_phys_t *scip =
7557 	    &spa->spa_condensing_indirect_phys;
7558 	if (scip->scip_next_mapping_object != 0) {
7559 		vdev_t *vd = spa->spa_root_vdev->vdev_child[scip->scip_vdev];
7560 		ASSERT(scip->scip_prev_obsolete_sm_object != 0);
7561 		ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
7562 
7563 		(void) printf("Condensing indirect vdev %llu: new mapping "
7564 		    "object %llu, prev obsolete sm %llu\n",
7565 		    (u_longlong_t)scip->scip_vdev,
7566 		    (u_longlong_t)scip->scip_next_mapping_object,
7567 		    (u_longlong_t)scip->scip_prev_obsolete_sm_object);
7568 		if (scip->scip_prev_obsolete_sm_object != 0) {
7569 			space_map_t *prev_obsolete_sm = NULL;
7570 			VERIFY0(space_map_open(&prev_obsolete_sm,
7571 			    spa->spa_meta_objset,
7572 			    scip->scip_prev_obsolete_sm_object,
7573 			    0, vd->vdev_asize, 0));
7574 			dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm);
7575 			(void) printf("\n");
7576 			space_map_close(prev_obsolete_sm);
7577 		}
7578 
7579 		scip_count += 2;
7580 	}
7581 
7582 	for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
7583 		vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
7584 		vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
7585 
7586 		if (vic->vic_mapping_object != 0) {
7587 			ASSERT(vd->vdev_ops == &vdev_indirect_ops ||
7588 			    vd->vdev_removing);
7589 			indirect_vdev_count++;
7590 
7591 			if (vd->vdev_indirect_mapping->vim_havecounts) {
7592 				obsolete_counts_count++;
7593 			}
7594 		}
7595 
7596 		boolean_t are_precise;
7597 		VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise));
7598 		if (are_precise) {
7599 			ASSERT(vic->vic_mapping_object != 0);
7600 			precise_vdev_count++;
7601 		}
7602 
7603 		uint64_t obsolete_sm_object;
7604 		VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object));
7605 		if (obsolete_sm_object != 0) {
7606 			ASSERT(vic->vic_mapping_object != 0);
7607 			obsolete_sm_count++;
7608 		}
7609 	}
7610 
7611 	(void) feature_get_refcount(spa,
7612 	    &spa_feature_table[SPA_FEATURE_DEVICE_REMOVAL],
7613 	    &dr_feature_refcount);
7614 	(void) feature_get_refcount(spa,
7615 	    &spa_feature_table[SPA_FEATURE_OBSOLETE_COUNTS],
7616 	    &oc_feature_refcount);
7617 
7618 	if (dr_feature_refcount != indirect_vdev_count) {
7619 		ret = 1;
7620 		(void) printf("Number of indirect vdevs (%llu) " \
7621 		    "does not match feature count (%llu)\n",
7622 		    (u_longlong_t)indirect_vdev_count,
7623 		    (u_longlong_t)dr_feature_refcount);
7624 	} else {
7625 		(void) printf("Verified device_removal feature refcount " \
7626 		    "of %llu is correct\n",
7627 		    (u_longlong_t)dr_feature_refcount);
7628 	}
7629 
7630 	if (zap_contains(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT,
7631 	    DMU_POOL_OBSOLETE_BPOBJ) == 0) {
7632 		obsolete_bpobj_count++;
7633 	}
7634 
7635 
7636 	obsolete_counts_object_count = precise_vdev_count;
7637 	obsolete_counts_object_count += obsolete_sm_count;
7638 	obsolete_counts_object_count += obsolete_counts_count;
7639 	obsolete_counts_object_count += scip_count;
7640 	obsolete_counts_object_count += obsolete_bpobj_count;
7641 	obsolete_counts_object_count += remap_deadlist_count;
7642 
7643 	if (oc_feature_refcount != obsolete_counts_object_count) {
7644 		ret = 1;
7645 		(void) printf("Number of obsolete counts objects (%llu) " \
7646 		    "does not match feature count (%llu)\n",
7647 		    (u_longlong_t)obsolete_counts_object_count,
7648 		    (u_longlong_t)oc_feature_refcount);
7649 		(void) printf("pv:%llu os:%llu oc:%llu sc:%llu "
7650 		    "ob:%llu rd:%llu\n",
7651 		    (u_longlong_t)precise_vdev_count,
7652 		    (u_longlong_t)obsolete_sm_count,
7653 		    (u_longlong_t)obsolete_counts_count,
7654 		    (u_longlong_t)scip_count,
7655 		    (u_longlong_t)obsolete_bpobj_count,
7656 		    (u_longlong_t)remap_deadlist_count);
7657 	} else {
7658 		(void) printf("Verified indirect_refcount feature refcount " \
7659 		    "of %llu is correct\n",
7660 		    (u_longlong_t)oc_feature_refcount);
7661 	}
7662 	return (ret);
7663 }
7664 
7665 static void
zdb_set_skip_mmp(char * target)7666 zdb_set_skip_mmp(char *target)
7667 {
7668 	spa_t *spa;
7669 
7670 	/*
7671 	 * Disable the activity check to allow examination of
7672 	 * active pools.
7673 	 */
7674 	mutex_enter(&spa_namespace_lock);
7675 	if ((spa = spa_lookup(target)) != NULL) {
7676 		spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
7677 	}
7678 	mutex_exit(&spa_namespace_lock);
7679 }
7680 
7681 #define	BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE"
7682 /*
7683  * Import the checkpointed state of the pool specified by the target
7684  * parameter as readonly. The function also accepts a pool config
7685  * as an optional parameter, else it attempts to infer the config by
7686  * the name of the target pool.
7687  *
7688  * Note that the checkpointed state's pool name will be the name of
7689  * the original pool with the above suffix appended to it. In addition,
7690  * if the target is not a pool name (e.g. a path to a dataset) then
7691  * the new_path parameter is populated with the updated path to
7692  * reflect the fact that we are looking into the checkpointed state.
7693  *
7694  * The function returns a newly-allocated copy of the name of the
7695  * pool containing the checkpointed state. When this copy is no
7696  * longer needed it should be freed with free(3C). Same thing
7697  * applies to the new_path parameter if allocated.
7698  */
7699 static char *
import_checkpointed_state(char * target,nvlist_t * cfg,char ** new_path)7700 import_checkpointed_state(char *target, nvlist_t *cfg, char **new_path)
7701 {
7702 	int error = 0;
7703 	char *poolname, *bogus_name = NULL;
7704 	boolean_t freecfg = B_FALSE;
7705 
7706 	/* If the target is not a pool, the extract the pool name */
7707 	char *path_start = strchr(target, '/');
7708 	if (path_start != NULL) {
7709 		size_t poolname_len = path_start - target;
7710 		poolname = strndup(target, poolname_len);
7711 	} else {
7712 		poolname = target;
7713 	}
7714 
7715 	if (cfg == NULL) {
7716 		zdb_set_skip_mmp(poolname);
7717 		error = spa_get_stats(poolname, &cfg, NULL, 0);
7718 		if (error != 0) {
7719 			fatal("Tried to read config of pool \"%s\" but "
7720 			    "spa_get_stats() failed with error %d\n",
7721 			    poolname, error);
7722 		}
7723 		freecfg = B_TRUE;
7724 	}
7725 
7726 	if (asprintf(&bogus_name, "%s%s", poolname, BOGUS_SUFFIX) == -1) {
7727 		if (target != poolname)
7728 			free(poolname);
7729 		return (NULL);
7730 	}
7731 	fnvlist_add_string(cfg, ZPOOL_CONFIG_POOL_NAME, bogus_name);
7732 
7733 	error = spa_import(bogus_name, cfg, NULL,
7734 	    ZFS_IMPORT_MISSING_LOG | ZFS_IMPORT_CHECKPOINT |
7735 	    ZFS_IMPORT_SKIP_MMP);
7736 	if (freecfg)
7737 		nvlist_free(cfg);
7738 	if (error != 0) {
7739 		fatal("Tried to import pool \"%s\" but spa_import() failed "
7740 		    "with error %d\n", bogus_name, error);
7741 	}
7742 
7743 	if (new_path != NULL && path_start != NULL) {
7744 		if (asprintf(new_path, "%s%s", bogus_name, path_start) == -1) {
7745 			free(bogus_name);
7746 			if (path_start != NULL)
7747 				free(poolname);
7748 			return (NULL);
7749 		}
7750 	}
7751 
7752 	if (target != poolname)
7753 		free(poolname);
7754 
7755 	return (bogus_name);
7756 }
7757 
7758 typedef struct verify_checkpoint_sm_entry_cb_arg {
7759 	vdev_t *vcsec_vd;
7760 
7761 	/* the following fields are only used for printing progress */
7762 	uint64_t vcsec_entryid;
7763 	uint64_t vcsec_num_entries;
7764 } verify_checkpoint_sm_entry_cb_arg_t;
7765 
7766 #define	ENTRIES_PER_PROGRESS_UPDATE 10000
7767 
7768 static int
verify_checkpoint_sm_entry_cb(space_map_entry_t * sme,void * arg)7769 verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, void *arg)
7770 {
7771 	verify_checkpoint_sm_entry_cb_arg_t *vcsec = arg;
7772 	vdev_t *vd = vcsec->vcsec_vd;
7773 	metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift];
7774 	uint64_t end = sme->sme_offset + sme->sme_run;
7775 
7776 	ASSERT(sme->sme_type == SM_FREE);
7777 
7778 	if ((vcsec->vcsec_entryid % ENTRIES_PER_PROGRESS_UPDATE) == 0) {
7779 		(void) fprintf(stderr,
7780 		    "\rverifying vdev %llu, space map entry %llu of %llu ...",
7781 		    (longlong_t)vd->vdev_id,
7782 		    (longlong_t)vcsec->vcsec_entryid,
7783 		    (longlong_t)vcsec->vcsec_num_entries);
7784 	}
7785 	vcsec->vcsec_entryid++;
7786 
7787 	/*
7788 	 * See comment in checkpoint_sm_exclude_entry_cb()
7789 	 */
7790 	VERIFY3U(sme->sme_offset, >=, ms->ms_start);
7791 	VERIFY3U(end, <=, ms->ms_start + ms->ms_size);
7792 
7793 	/*
7794 	 * The entries in the vdev_checkpoint_sm should be marked as
7795 	 * allocated in the checkpointed state of the pool, therefore
7796 	 * their respective ms_allocateable trees should not contain them.
7797 	 */
7798 	mutex_enter(&ms->ms_lock);
7799 	range_tree_verify_not_present(ms->ms_allocatable,
7800 	    sme->sme_offset, sme->sme_run);
7801 	mutex_exit(&ms->ms_lock);
7802 
7803 	return (0);
7804 }
7805 
7806 /*
7807  * Verify that all segments in the vdev_checkpoint_sm are allocated
7808  * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's
7809  * ms_allocatable).
7810  *
7811  * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of
7812  * each vdev in the current state of the pool to the metaslab space maps
7813  * (ms_sm) of the checkpointed state of the pool.
7814  *
7815  * Note that the function changes the state of the ms_allocatable
7816  * trees of the current spa_t. The entries of these ms_allocatable
7817  * trees are cleared out and then repopulated from with the free
7818  * entries of their respective ms_sm space maps.
7819  */
7820 static void
verify_checkpoint_vdev_spacemaps(spa_t * checkpoint,spa_t * current)7821 verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, spa_t *current)
7822 {
7823 	vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7824 	vdev_t *current_rvd = current->spa_root_vdev;
7825 
7826 	load_concrete_ms_allocatable_trees(checkpoint, SM_FREE);
7827 
7828 	for (uint64_t c = 0; c < ckpoint_rvd->vdev_children; c++) {
7829 		vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[c];
7830 		vdev_t *current_vd = current_rvd->vdev_child[c];
7831 
7832 		space_map_t *checkpoint_sm = NULL;
7833 		uint64_t checkpoint_sm_obj;
7834 
7835 		if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7836 			/*
7837 			 * Since we don't allow device removal in a pool
7838 			 * that has a checkpoint, we expect that all removed
7839 			 * vdevs were removed from the pool before the
7840 			 * checkpoint.
7841 			 */
7842 			ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7843 			continue;
7844 		}
7845 
7846 		/*
7847 		 * If the checkpoint space map doesn't exist, then nothing
7848 		 * here is checkpointed so there's nothing to verify.
7849 		 */
7850 		if (current_vd->vdev_top_zap == 0 ||
7851 		    zap_contains(spa_meta_objset(current),
7852 		    current_vd->vdev_top_zap,
7853 		    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
7854 			continue;
7855 
7856 		VERIFY0(zap_lookup(spa_meta_objset(current),
7857 		    current_vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
7858 		    sizeof (uint64_t), 1, &checkpoint_sm_obj));
7859 
7860 		VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current),
7861 		    checkpoint_sm_obj, 0, current_vd->vdev_asize,
7862 		    current_vd->vdev_ashift));
7863 
7864 		verify_checkpoint_sm_entry_cb_arg_t vcsec;
7865 		vcsec.vcsec_vd = ckpoint_vd;
7866 		vcsec.vcsec_entryid = 0;
7867 		vcsec.vcsec_num_entries =
7868 		    space_map_length(checkpoint_sm) / sizeof (uint64_t);
7869 		VERIFY0(space_map_iterate(checkpoint_sm,
7870 		    space_map_length(checkpoint_sm),
7871 		    verify_checkpoint_sm_entry_cb, &vcsec));
7872 		if (dump_opt['m'] > 3)
7873 			dump_spacemap(current->spa_meta_objset, checkpoint_sm);
7874 		space_map_close(checkpoint_sm);
7875 	}
7876 
7877 	/*
7878 	 * If we've added vdevs since we took the checkpoint, ensure
7879 	 * that their checkpoint space maps are empty.
7880 	 */
7881 	if (ckpoint_rvd->vdev_children < current_rvd->vdev_children) {
7882 		for (uint64_t c = ckpoint_rvd->vdev_children;
7883 		    c < current_rvd->vdev_children; c++) {
7884 			vdev_t *current_vd = current_rvd->vdev_child[c];
7885 			VERIFY3P(current_vd->vdev_checkpoint_sm, ==, NULL);
7886 		}
7887 	}
7888 
7889 	/* for cleaner progress output */
7890 	(void) fprintf(stderr, "\n");
7891 }
7892 
7893 /*
7894  * Verifies that all space that's allocated in the checkpoint is
7895  * still allocated in the current version, by checking that everything
7896  * in checkpoint's ms_allocatable (which is actually allocated, not
7897  * allocatable/free) is not present in current's ms_allocatable.
7898  *
7899  * Note that the function changes the state of the ms_allocatable
7900  * trees of both spas when called. The entries of all ms_allocatable
7901  * trees are cleared out and then repopulated from their respective
7902  * ms_sm space maps. In the checkpointed state we load the allocated
7903  * entries, and in the current state we load the free entries.
7904  */
7905 static void
verify_checkpoint_ms_spacemaps(spa_t * checkpoint,spa_t * current)7906 verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_t *current)
7907 {
7908 	vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev;
7909 	vdev_t *current_rvd = current->spa_root_vdev;
7910 
7911 	load_concrete_ms_allocatable_trees(checkpoint, SM_ALLOC);
7912 	load_concrete_ms_allocatable_trees(current, SM_FREE);
7913 
7914 	for (uint64_t i = 0; i < ckpoint_rvd->vdev_children; i++) {
7915 		vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[i];
7916 		vdev_t *current_vd = current_rvd->vdev_child[i];
7917 
7918 		if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) {
7919 			/*
7920 			 * See comment in verify_checkpoint_vdev_spacemaps()
7921 			 */
7922 			ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops);
7923 			continue;
7924 		}
7925 
7926 		for (uint64_t m = 0; m < ckpoint_vd->vdev_ms_count; m++) {
7927 			metaslab_t *ckpoint_msp = ckpoint_vd->vdev_ms[m];
7928 			metaslab_t *current_msp = current_vd->vdev_ms[m];
7929 
7930 			(void) fprintf(stderr,
7931 			    "\rverifying vdev %llu of %llu, "
7932 			    "metaslab %llu of %llu ...",
7933 			    (longlong_t)current_vd->vdev_id,
7934 			    (longlong_t)current_rvd->vdev_children,
7935 			    (longlong_t)current_vd->vdev_ms[m]->ms_id,
7936 			    (longlong_t)current_vd->vdev_ms_count);
7937 
7938 			/*
7939 			 * We walk through the ms_allocatable trees that
7940 			 * are loaded with the allocated blocks from the
7941 			 * ms_sm spacemaps of the checkpoint. For each
7942 			 * one of these ranges we ensure that none of them
7943 			 * exists in the ms_allocatable trees of the
7944 			 * current state which are loaded with the ranges
7945 			 * that are currently free.
7946 			 *
7947 			 * This way we ensure that none of the blocks that
7948 			 * are part of the checkpoint were freed by mistake.
7949 			 */
7950 			range_tree_walk(ckpoint_msp->ms_allocatable,
7951 			    (range_tree_func_t *)range_tree_verify_not_present,
7952 			    current_msp->ms_allocatable);
7953 		}
7954 	}
7955 
7956 	/* for cleaner progress output */
7957 	(void) fprintf(stderr, "\n");
7958 }
7959 
7960 static void
verify_checkpoint_blocks(spa_t * spa)7961 verify_checkpoint_blocks(spa_t *spa)
7962 {
7963 	ASSERT(!dump_opt['L']);
7964 
7965 	spa_t *checkpoint_spa;
7966 	char *checkpoint_pool;
7967 	int error = 0;
7968 
7969 	/*
7970 	 * We import the checkpointed state of the pool (under a different
7971 	 * name) so we can do verification on it against the current state
7972 	 * of the pool.
7973 	 */
7974 	checkpoint_pool = import_checkpointed_state(spa->spa_name, NULL,
7975 	    NULL);
7976 	ASSERT(strcmp(spa->spa_name, checkpoint_pool) != 0);
7977 
7978 	error = spa_open(checkpoint_pool, &checkpoint_spa, FTAG);
7979 	if (error != 0) {
7980 		fatal("Tried to open pool \"%s\" but spa_open() failed with "
7981 		    "error %d\n", checkpoint_pool, error);
7982 	}
7983 
7984 	/*
7985 	 * Ensure that ranges in the checkpoint space maps of each vdev
7986 	 * are allocated according to the checkpointed state's metaslab
7987 	 * space maps.
7988 	 */
7989 	verify_checkpoint_vdev_spacemaps(checkpoint_spa, spa);
7990 
7991 	/*
7992 	 * Ensure that allocated ranges in the checkpoint's metaslab
7993 	 * space maps remain allocated in the metaslab space maps of
7994 	 * the current state.
7995 	 */
7996 	verify_checkpoint_ms_spacemaps(checkpoint_spa, spa);
7997 
7998 	/*
7999 	 * Once we are done, we get rid of the checkpointed state.
8000 	 */
8001 	spa_close(checkpoint_spa, FTAG);
8002 	free(checkpoint_pool);
8003 }
8004 
8005 static void
dump_leftover_checkpoint_blocks(spa_t * spa)8006 dump_leftover_checkpoint_blocks(spa_t *spa)
8007 {
8008 	vdev_t *rvd = spa->spa_root_vdev;
8009 
8010 	for (uint64_t i = 0; i < rvd->vdev_children; i++) {
8011 		vdev_t *vd = rvd->vdev_child[i];
8012 
8013 		space_map_t *checkpoint_sm = NULL;
8014 		uint64_t checkpoint_sm_obj;
8015 
8016 		if (vd->vdev_top_zap == 0)
8017 			continue;
8018 
8019 		if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap,
8020 		    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0)
8021 			continue;
8022 
8023 		VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap,
8024 		    VDEV_TOP_ZAP_POOL_CHECKPOINT_SM,
8025 		    sizeof (uint64_t), 1, &checkpoint_sm_obj));
8026 
8027 		VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa),
8028 		    checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift));
8029 		dump_spacemap(spa->spa_meta_objset, checkpoint_sm);
8030 		space_map_close(checkpoint_sm);
8031 	}
8032 }
8033 
8034 static int
verify_checkpoint(spa_t * spa)8035 verify_checkpoint(spa_t *spa)
8036 {
8037 	uberblock_t checkpoint;
8038 	int error;
8039 
8040 	if (!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT))
8041 		return (0);
8042 
8043 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
8044 	    DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
8045 	    sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
8046 
8047 	if (error == ENOENT && !dump_opt['L']) {
8048 		/*
8049 		 * If the feature is active but the uberblock is missing
8050 		 * then we must be in the middle of discarding the
8051 		 * checkpoint.
8052 		 */
8053 		(void) printf("\nPartially discarded checkpoint "
8054 		    "state found:\n");
8055 		if (dump_opt['m'] > 3)
8056 			dump_leftover_checkpoint_blocks(spa);
8057 		return (0);
8058 	} else if (error != 0) {
8059 		(void) printf("lookup error %d when looking for "
8060 		    "checkpointed uberblock in MOS\n", error);
8061 		return (error);
8062 	}
8063 	dump_uberblock(&checkpoint, "\nCheckpointed uberblock found:\n", "\n");
8064 
8065 	if (checkpoint.ub_checkpoint_txg == 0) {
8066 		(void) printf("\nub_checkpoint_txg not set in checkpointed "
8067 		    "uberblock\n");
8068 		error = 3;
8069 	}
8070 
8071 	if (error == 0 && !dump_opt['L'])
8072 		verify_checkpoint_blocks(spa);
8073 
8074 	return (error);
8075 }
8076 
8077 static void
mos_leaks_cb(void * arg,uint64_t start,uint64_t size)8078 mos_leaks_cb(void *arg, uint64_t start, uint64_t size)
8079 {
8080 	(void) arg;
8081 	for (uint64_t i = start; i < size; i++) {
8082 		(void) printf("MOS object %llu referenced but not allocated\n",
8083 		    (u_longlong_t)i);
8084 	}
8085 }
8086 
8087 static void
mos_obj_refd(uint64_t obj)8088 mos_obj_refd(uint64_t obj)
8089 {
8090 	if (obj != 0 && mos_refd_objs != NULL)
8091 		range_tree_add(mos_refd_objs, obj, 1);
8092 }
8093 
8094 /*
8095  * Call on a MOS object that may already have been referenced.
8096  */
8097 static void
mos_obj_refd_multiple(uint64_t obj)8098 mos_obj_refd_multiple(uint64_t obj)
8099 {
8100 	if (obj != 0 && mos_refd_objs != NULL &&
8101 	    !range_tree_contains(mos_refd_objs, obj, 1))
8102 		range_tree_add(mos_refd_objs, obj, 1);
8103 }
8104 
8105 static void
mos_leak_vdev_top_zap(vdev_t * vd)8106 mos_leak_vdev_top_zap(vdev_t *vd)
8107 {
8108 	uint64_t ms_flush_data_obj;
8109 	int error = zap_lookup(spa_meta_objset(vd->vdev_spa),
8110 	    vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS,
8111 	    sizeof (ms_flush_data_obj), 1, &ms_flush_data_obj);
8112 	if (error == ENOENT)
8113 		return;
8114 	ASSERT0(error);
8115 
8116 	mos_obj_refd(ms_flush_data_obj);
8117 }
8118 
8119 static void
mos_leak_vdev(vdev_t * vd)8120 mos_leak_vdev(vdev_t *vd)
8121 {
8122 	mos_obj_refd(vd->vdev_dtl_object);
8123 	mos_obj_refd(vd->vdev_ms_array);
8124 	mos_obj_refd(vd->vdev_indirect_config.vic_births_object);
8125 	mos_obj_refd(vd->vdev_indirect_config.vic_mapping_object);
8126 	mos_obj_refd(vd->vdev_leaf_zap);
8127 	if (vd->vdev_checkpoint_sm != NULL)
8128 		mos_obj_refd(vd->vdev_checkpoint_sm->sm_object);
8129 	if (vd->vdev_indirect_mapping != NULL) {
8130 		mos_obj_refd(vd->vdev_indirect_mapping->
8131 		    vim_phys->vimp_counts_object);
8132 	}
8133 	if (vd->vdev_obsolete_sm != NULL)
8134 		mos_obj_refd(vd->vdev_obsolete_sm->sm_object);
8135 
8136 	for (uint64_t m = 0; m < vd->vdev_ms_count; m++) {
8137 		metaslab_t *ms = vd->vdev_ms[m];
8138 		mos_obj_refd(space_map_object(ms->ms_sm));
8139 	}
8140 
8141 	if (vd->vdev_root_zap != 0)
8142 		mos_obj_refd(vd->vdev_root_zap);
8143 
8144 	if (vd->vdev_top_zap != 0) {
8145 		mos_obj_refd(vd->vdev_top_zap);
8146 		mos_leak_vdev_top_zap(vd);
8147 	}
8148 
8149 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
8150 		mos_leak_vdev(vd->vdev_child[c]);
8151 	}
8152 }
8153 
8154 static void
mos_leak_log_spacemaps(spa_t * spa)8155 mos_leak_log_spacemaps(spa_t *spa)
8156 {
8157 	uint64_t spacemap_zap;
8158 	int error = zap_lookup(spa_meta_objset(spa),
8159 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LOG_SPACEMAP_ZAP,
8160 	    sizeof (spacemap_zap), 1, &spacemap_zap);
8161 	if (error == ENOENT)
8162 		return;
8163 	ASSERT0(error);
8164 
8165 	mos_obj_refd(spacemap_zap);
8166 	for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg);
8167 	    sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls))
8168 		mos_obj_refd(sls->sls_sm_obj);
8169 }
8170 
8171 static void
errorlog_count_refd(objset_t * mos,uint64_t errlog)8172 errorlog_count_refd(objset_t *mos, uint64_t errlog)
8173 {
8174 	zap_cursor_t zc;
8175 	zap_attribute_t *za = zap_attribute_alloc();
8176 	for (zap_cursor_init(&zc, mos, errlog);
8177 	    zap_cursor_retrieve(&zc, za) == 0;
8178 	    zap_cursor_advance(&zc)) {
8179 		mos_obj_refd(za->za_first_integer);
8180 	}
8181 	zap_cursor_fini(&zc);
8182 	zap_attribute_free(za);
8183 }
8184 
8185 static int
dump_mos_leaks(spa_t * spa)8186 dump_mos_leaks(spa_t *spa)
8187 {
8188 	int rv = 0;
8189 	objset_t *mos = spa->spa_meta_objset;
8190 	dsl_pool_t *dp = spa->spa_dsl_pool;
8191 
8192 	/* Visit and mark all referenced objects in the MOS */
8193 
8194 	mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT);
8195 	mos_obj_refd(spa->spa_pool_props_object);
8196 	mos_obj_refd(spa->spa_config_object);
8197 	mos_obj_refd(spa->spa_ddt_stat_object);
8198 	mos_obj_refd(spa->spa_feat_desc_obj);
8199 	mos_obj_refd(spa->spa_feat_enabled_txg_obj);
8200 	mos_obj_refd(spa->spa_feat_for_read_obj);
8201 	mos_obj_refd(spa->spa_feat_for_write_obj);
8202 	mos_obj_refd(spa->spa_history);
8203 	mos_obj_refd(spa->spa_errlog_last);
8204 	mos_obj_refd(spa->spa_errlog_scrub);
8205 
8206 	if (spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) {
8207 		errorlog_count_refd(mos, spa->spa_errlog_last);
8208 		errorlog_count_refd(mos, spa->spa_errlog_scrub);
8209 	}
8210 
8211 	mos_obj_refd(spa->spa_all_vdev_zaps);
8212 	mos_obj_refd(spa->spa_dsl_pool->dp_bptree_obj);
8213 	mos_obj_refd(spa->spa_dsl_pool->dp_tmp_userrefs_obj);
8214 	mos_obj_refd(spa->spa_dsl_pool->dp_scan->scn_phys.scn_queue_obj);
8215 	bpobj_count_refd(&spa->spa_deferred_bpobj);
8216 	mos_obj_refd(dp->dp_empty_bpobj);
8217 	bpobj_count_refd(&dp->dp_obsolete_bpobj);
8218 	bpobj_count_refd(&dp->dp_free_bpobj);
8219 	mos_obj_refd(spa->spa_l2cache.sav_object);
8220 	mos_obj_refd(spa->spa_spares.sav_object);
8221 
8222 	if (spa->spa_syncing_log_sm != NULL)
8223 		mos_obj_refd(spa->spa_syncing_log_sm->sm_object);
8224 	mos_leak_log_spacemaps(spa);
8225 
8226 	mos_obj_refd(spa->spa_condensing_indirect_phys.
8227 	    scip_next_mapping_object);
8228 	mos_obj_refd(spa->spa_condensing_indirect_phys.
8229 	    scip_prev_obsolete_sm_object);
8230 	if (spa->spa_condensing_indirect_phys.scip_next_mapping_object != 0) {
8231 		vdev_indirect_mapping_t *vim =
8232 		    vdev_indirect_mapping_open(mos,
8233 		    spa->spa_condensing_indirect_phys.scip_next_mapping_object);
8234 		mos_obj_refd(vim->vim_phys->vimp_counts_object);
8235 		vdev_indirect_mapping_close(vim);
8236 	}
8237 	deleted_livelists_dump_mos(spa);
8238 
8239 	if (dp->dp_origin_snap != NULL) {
8240 		dsl_dataset_t *ds;
8241 
8242 		dsl_pool_config_enter(dp, FTAG);
8243 		VERIFY0(dsl_dataset_hold_obj(dp,
8244 		    dsl_dataset_phys(dp->dp_origin_snap)->ds_next_snap_obj,
8245 		    FTAG, &ds));
8246 		count_ds_mos_objects(ds);
8247 		dump_blkptr_list(&ds->ds_deadlist, "Deadlist");
8248 		dsl_dataset_rele(ds, FTAG);
8249 		dsl_pool_config_exit(dp, FTAG);
8250 
8251 		count_ds_mos_objects(dp->dp_origin_snap);
8252 		dump_blkptr_list(&dp->dp_origin_snap->ds_deadlist, "Deadlist");
8253 	}
8254 	count_dir_mos_objects(dp->dp_mos_dir);
8255 	if (dp->dp_free_dir != NULL)
8256 		count_dir_mos_objects(dp->dp_free_dir);
8257 	if (dp->dp_leak_dir != NULL)
8258 		count_dir_mos_objects(dp->dp_leak_dir);
8259 
8260 	mos_leak_vdev(spa->spa_root_vdev);
8261 
8262 	for (uint64_t c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
8263 		ddt_t *ddt = spa->spa_ddt[c];
8264 		if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED)
8265 			continue;
8266 
8267 		/* DDT store objects */
8268 		for (ddt_type_t type = 0; type < DDT_TYPES; type++) {
8269 			for (ddt_class_t class = 0; class < DDT_CLASSES;
8270 			    class++) {
8271 				mos_obj_refd(ddt->ddt_object[type][class]);
8272 			}
8273 		}
8274 
8275 		/* FDT container */
8276 		if (ddt->ddt_version == DDT_VERSION_FDT)
8277 			mos_obj_refd(ddt->ddt_dir_object);
8278 
8279 		/* FDT log objects */
8280 		if (ddt->ddt_flags & DDT_FLAG_LOG) {
8281 			mos_obj_refd(ddt->ddt_log[0].ddl_object);
8282 			mos_obj_refd(ddt->ddt_log[1].ddl_object);
8283 		}
8284 	}
8285 
8286 	for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) {
8287 		brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid];
8288 		if (brtvd->bv_initiated) {
8289 			mos_obj_refd(brtvd->bv_mos_brtvdev);
8290 			mos_obj_refd(brtvd->bv_mos_entries);
8291 		}
8292 	}
8293 
8294 	/*
8295 	 * Visit all allocated objects and make sure they are referenced.
8296 	 */
8297 	uint64_t object = 0;
8298 	while (dmu_object_next(mos, &object, B_FALSE, 0) == 0) {
8299 		if (range_tree_contains(mos_refd_objs, object, 1)) {
8300 			range_tree_remove(mos_refd_objs, object, 1);
8301 		} else {
8302 			dmu_object_info_t doi;
8303 			const char *name;
8304 			VERIFY0(dmu_object_info(mos, object, &doi));
8305 			if (doi.doi_type & DMU_OT_NEWTYPE) {
8306 				dmu_object_byteswap_t bswap =
8307 				    DMU_OT_BYTESWAP(doi.doi_type);
8308 				name = dmu_ot_byteswap[bswap].ob_name;
8309 			} else {
8310 				name = dmu_ot[doi.doi_type].ot_name;
8311 			}
8312 
8313 			(void) printf("MOS object %llu (%s) leaked\n",
8314 			    (u_longlong_t)object, name);
8315 			rv = 2;
8316 		}
8317 	}
8318 	(void) range_tree_walk(mos_refd_objs, mos_leaks_cb, NULL);
8319 	if (!range_tree_is_empty(mos_refd_objs))
8320 		rv = 2;
8321 	range_tree_vacate(mos_refd_objs, NULL, NULL);
8322 	range_tree_destroy(mos_refd_objs);
8323 	return (rv);
8324 }
8325 
8326 typedef struct log_sm_obsolete_stats_arg {
8327 	uint64_t lsos_current_txg;
8328 
8329 	uint64_t lsos_total_entries;
8330 	uint64_t lsos_valid_entries;
8331 
8332 	uint64_t lsos_sm_entries;
8333 	uint64_t lsos_valid_sm_entries;
8334 } log_sm_obsolete_stats_arg_t;
8335 
8336 static int
log_spacemap_obsolete_stats_cb(spa_t * spa,space_map_entry_t * sme,uint64_t txg,void * arg)8337 log_spacemap_obsolete_stats_cb(spa_t *spa, space_map_entry_t *sme,
8338     uint64_t txg, void *arg)
8339 {
8340 	log_sm_obsolete_stats_arg_t *lsos = arg;
8341 
8342 	uint64_t offset = sme->sme_offset;
8343 	uint64_t vdev_id = sme->sme_vdev;
8344 
8345 	if (lsos->lsos_current_txg == 0) {
8346 		/* this is the first log */
8347 		lsos->lsos_current_txg = txg;
8348 	} else if (lsos->lsos_current_txg < txg) {
8349 		/* we just changed log - print stats and reset */
8350 		(void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8351 		    (u_longlong_t)lsos->lsos_valid_sm_entries,
8352 		    (u_longlong_t)lsos->lsos_sm_entries,
8353 		    (u_longlong_t)lsos->lsos_current_txg);
8354 		lsos->lsos_valid_sm_entries = 0;
8355 		lsos->lsos_sm_entries = 0;
8356 		lsos->lsos_current_txg = txg;
8357 	}
8358 	ASSERT3U(lsos->lsos_current_txg, ==, txg);
8359 
8360 	lsos->lsos_sm_entries++;
8361 	lsos->lsos_total_entries++;
8362 
8363 	vdev_t *vd = vdev_lookup_top(spa, vdev_id);
8364 	if (!vdev_is_concrete(vd))
8365 		return (0);
8366 
8367 	metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift];
8368 	ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE);
8369 
8370 	if (txg < metaslab_unflushed_txg(ms))
8371 		return (0);
8372 	lsos->lsos_valid_sm_entries++;
8373 	lsos->lsos_valid_entries++;
8374 	return (0);
8375 }
8376 
8377 static void
dump_log_spacemap_obsolete_stats(spa_t * spa)8378 dump_log_spacemap_obsolete_stats(spa_t *spa)
8379 {
8380 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
8381 		return;
8382 
8383 	log_sm_obsolete_stats_arg_t lsos = {0};
8384 
8385 	(void) printf("Log Space Map Obsolete Entry Statistics:\n");
8386 
8387 	iterate_through_spacemap_logs(spa,
8388 	    log_spacemap_obsolete_stats_cb, &lsos);
8389 
8390 	/* print stats for latest log */
8391 	(void) printf("%-8llu valid entries out of %-8llu - txg %llu\n",
8392 	    (u_longlong_t)lsos.lsos_valid_sm_entries,
8393 	    (u_longlong_t)lsos.lsos_sm_entries,
8394 	    (u_longlong_t)lsos.lsos_current_txg);
8395 
8396 	(void) printf("%-8llu valid entries out of %-8llu - total\n\n",
8397 	    (u_longlong_t)lsos.lsos_valid_entries,
8398 	    (u_longlong_t)lsos.lsos_total_entries);
8399 }
8400 
8401 static void
dump_zpool(spa_t * spa)8402 dump_zpool(spa_t *spa)
8403 {
8404 	dsl_pool_t *dp = spa_get_dsl(spa);
8405 	int rc = 0;
8406 
8407 	if (dump_opt['y']) {
8408 		livelist_metaslab_validate(spa);
8409 	}
8410 
8411 	if (dump_opt['S']) {
8412 		dump_simulated_ddt(spa);
8413 		return;
8414 	}
8415 
8416 	if (!dump_opt['e'] && dump_opt['C'] > 1) {
8417 		(void) printf("\nCached configuration:\n");
8418 		dump_nvlist(spa->spa_config, 8);
8419 	}
8420 
8421 	if (dump_opt['C'])
8422 		dump_config(spa);
8423 
8424 	if (dump_opt['u'])
8425 		dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n");
8426 
8427 	if (dump_opt['D'])
8428 		dump_all_ddts(spa);
8429 
8430 	if (dump_opt['T'])
8431 		dump_brt(spa);
8432 
8433 	if (dump_opt['d'] > 2 || dump_opt['m'])
8434 		dump_metaslabs(spa);
8435 	if (dump_opt['M'])
8436 		dump_metaslab_groups(spa, dump_opt['M'] > 1);
8437 	if (dump_opt['d'] > 2 || dump_opt['m']) {
8438 		dump_log_spacemaps(spa);
8439 		dump_log_spacemap_obsolete_stats(spa);
8440 	}
8441 
8442 	if (dump_opt['d'] || dump_opt['i']) {
8443 		spa_feature_t f;
8444 		mos_refd_objs = range_tree_create(NULL, RANGE_SEG64, NULL, 0,
8445 		    0);
8446 		dump_objset(dp->dp_meta_objset);
8447 
8448 		if (dump_opt['d'] >= 3) {
8449 			dsl_pool_t *dp = spa->spa_dsl_pool;
8450 			dump_full_bpobj(&spa->spa_deferred_bpobj,
8451 			    "Deferred frees", 0);
8452 			if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
8453 				dump_full_bpobj(&dp->dp_free_bpobj,
8454 				    "Pool snapshot frees", 0);
8455 			}
8456 			if (bpobj_is_open(&dp->dp_obsolete_bpobj)) {
8457 				ASSERT(spa_feature_is_enabled(spa,
8458 				    SPA_FEATURE_DEVICE_REMOVAL));
8459 				dump_full_bpobj(&dp->dp_obsolete_bpobj,
8460 				    "Pool obsolete blocks", 0);
8461 			}
8462 
8463 			if (spa_feature_is_active(spa,
8464 			    SPA_FEATURE_ASYNC_DESTROY)) {
8465 				dump_bptree(spa->spa_meta_objset,
8466 				    dp->dp_bptree_obj,
8467 				    "Pool dataset frees");
8468 			}
8469 			dump_dtl(spa->spa_root_vdev, 0);
8470 		}
8471 
8472 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++)
8473 			global_feature_count[f] = UINT64_MAX;
8474 		global_feature_count[SPA_FEATURE_REDACTION_BOOKMARKS] = 0;
8475 		global_feature_count[SPA_FEATURE_REDACTION_LIST_SPILL] = 0;
8476 		global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN] = 0;
8477 		global_feature_count[SPA_FEATURE_LIVELIST] = 0;
8478 
8479 		(void) dmu_objset_find(spa_name(spa), dump_one_objset,
8480 		    NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
8481 
8482 		if (rc == 0 && !dump_opt['L'])
8483 			rc = dump_mos_leaks(spa);
8484 
8485 		for (f = 0; f < SPA_FEATURES; f++) {
8486 			uint64_t refcount;
8487 
8488 			uint64_t *arr;
8489 			if (!(spa_feature_table[f].fi_flags &
8490 			    ZFEATURE_FLAG_PER_DATASET)) {
8491 				if (global_feature_count[f] == UINT64_MAX)
8492 					continue;
8493 				if (!spa_feature_is_enabled(spa, f)) {
8494 					ASSERT0(global_feature_count[f]);
8495 					continue;
8496 				}
8497 				arr = global_feature_count;
8498 			} else {
8499 				if (!spa_feature_is_enabled(spa, f)) {
8500 					ASSERT0(dataset_feature_count[f]);
8501 					continue;
8502 				}
8503 				arr = dataset_feature_count;
8504 			}
8505 			if (feature_get_refcount(spa, &spa_feature_table[f],
8506 			    &refcount) == ENOTSUP)
8507 				continue;
8508 			if (arr[f] != refcount) {
8509 				(void) printf("%s feature refcount mismatch: "
8510 				    "%lld consumers != %lld refcount\n",
8511 				    spa_feature_table[f].fi_uname,
8512 				    (longlong_t)arr[f], (longlong_t)refcount);
8513 				rc = 2;
8514 			} else {
8515 				(void) printf("Verified %s feature refcount "
8516 				    "of %llu is correct\n",
8517 				    spa_feature_table[f].fi_uname,
8518 				    (longlong_t)refcount);
8519 			}
8520 		}
8521 
8522 		if (rc == 0)
8523 			rc = verify_device_removal_feature_counts(spa);
8524 	}
8525 
8526 	if (rc == 0 && (dump_opt['b'] || dump_opt['c']))
8527 		rc = dump_block_stats(spa);
8528 
8529 	if (rc == 0)
8530 		rc = verify_spacemap_refcounts(spa);
8531 
8532 	if (dump_opt['s'])
8533 		show_pool_stats(spa);
8534 
8535 	if (dump_opt['h'])
8536 		dump_history(spa);
8537 
8538 	if (rc == 0)
8539 		rc = verify_checkpoint(spa);
8540 
8541 	if (rc != 0) {
8542 		dump_debug_buffer();
8543 		zdb_exit(rc);
8544 	}
8545 }
8546 
8547 #define	ZDB_FLAG_CHECKSUM	0x0001
8548 #define	ZDB_FLAG_DECOMPRESS	0x0002
8549 #define	ZDB_FLAG_BSWAP		0x0004
8550 #define	ZDB_FLAG_GBH		0x0008
8551 #define	ZDB_FLAG_INDIRECT	0x0010
8552 #define	ZDB_FLAG_RAW		0x0020
8553 #define	ZDB_FLAG_PRINT_BLKPTR	0x0040
8554 #define	ZDB_FLAG_VERBOSE	0x0080
8555 
8556 static int flagbits[256];
8557 static char flagbitstr[16];
8558 
8559 static void
zdb_print_blkptr(const blkptr_t * bp,int flags)8560 zdb_print_blkptr(const blkptr_t *bp, int flags)
8561 {
8562 	char blkbuf[BP_SPRINTF_LEN];
8563 
8564 	if (flags & ZDB_FLAG_BSWAP)
8565 		byteswap_uint64_array((void *)bp, sizeof (blkptr_t));
8566 
8567 	snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
8568 	(void) printf("%s\n", blkbuf);
8569 }
8570 
8571 static void
zdb_dump_indirect(blkptr_t * bp,int nbps,int flags)8572 zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
8573 {
8574 	int i;
8575 
8576 	for (i = 0; i < nbps; i++)
8577 		zdb_print_blkptr(&bp[i], flags);
8578 }
8579 
8580 static void
zdb_dump_gbh(void * buf,int flags)8581 zdb_dump_gbh(void *buf, int flags)
8582 {
8583 	zdb_dump_indirect((blkptr_t *)buf, SPA_GBH_NBLKPTRS, flags);
8584 }
8585 
8586 static void
zdb_dump_block_raw(void * buf,uint64_t size,int flags)8587 zdb_dump_block_raw(void *buf, uint64_t size, int flags)
8588 {
8589 	if (flags & ZDB_FLAG_BSWAP)
8590 		byteswap_uint64_array(buf, size);
8591 	VERIFY(write(fileno(stdout), buf, size) == size);
8592 }
8593 
8594 static void
zdb_dump_block(char * label,void * buf,uint64_t size,int flags)8595 zdb_dump_block(char *label, void *buf, uint64_t size, int flags)
8596 {
8597 	uint64_t *d = (uint64_t *)buf;
8598 	unsigned nwords = size / sizeof (uint64_t);
8599 	int do_bswap = !!(flags & ZDB_FLAG_BSWAP);
8600 	unsigned i, j;
8601 	const char *hdr;
8602 	char *c;
8603 
8604 
8605 	if (do_bswap)
8606 		hdr = " 7 6 5 4 3 2 1 0   f e d c b a 9 8";
8607 	else
8608 		hdr = " 0 1 2 3 4 5 6 7   8 9 a b c d e f";
8609 
8610 	(void) printf("\n%s\n%6s   %s  0123456789abcdef\n", label, "", hdr);
8611 
8612 #ifdef _ZFS_LITTLE_ENDIAN
8613 	/* correct the endianness */
8614 	do_bswap = !do_bswap;
8615 #endif
8616 	for (i = 0; i < nwords; i += 2) {
8617 		(void) printf("%06llx:  %016llx  %016llx  ",
8618 		    (u_longlong_t)(i * sizeof (uint64_t)),
8619 		    (u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]),
8620 		    (u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1]));
8621 
8622 		c = (char *)&d[i];
8623 		for (j = 0; j < 2 * sizeof (uint64_t); j++)
8624 			(void) printf("%c", isprint(c[j]) ? c[j] : '.');
8625 		(void) printf("\n");
8626 	}
8627 }
8628 
8629 /*
8630  * There are two acceptable formats:
8631  *	leaf_name	  - For example: c1t0d0 or /tmp/ztest.0a
8632  *	child[.child]*    - For example: 0.1.1
8633  *
8634  * The second form can be used to specify arbitrary vdevs anywhere
8635  * in the hierarchy.  For example, in a pool with a mirror of
8636  * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
8637  */
8638 static vdev_t *
zdb_vdev_lookup(vdev_t * vdev,const char * path)8639 zdb_vdev_lookup(vdev_t *vdev, const char *path)
8640 {
8641 	char *s, *p, *q;
8642 	unsigned i;
8643 
8644 	if (vdev == NULL)
8645 		return (NULL);
8646 
8647 	/* First, assume the x.x.x.x format */
8648 	i = strtoul(path, &s, 10);
8649 	if (s == path || (s && *s != '.' && *s != '\0'))
8650 		goto name;
8651 	if (i >= vdev->vdev_children)
8652 		return (NULL);
8653 
8654 	vdev = vdev->vdev_child[i];
8655 	if (s && *s == '\0')
8656 		return (vdev);
8657 	return (zdb_vdev_lookup(vdev, s+1));
8658 
8659 name:
8660 	for (i = 0; i < vdev->vdev_children; i++) {
8661 		vdev_t *vc = vdev->vdev_child[i];
8662 
8663 		if (vc->vdev_path == NULL) {
8664 			vc = zdb_vdev_lookup(vc, path);
8665 			if (vc == NULL)
8666 				continue;
8667 			else
8668 				return (vc);
8669 		}
8670 
8671 		p = strrchr(vc->vdev_path, '/');
8672 		p = p ? p + 1 : vc->vdev_path;
8673 		q = &vc->vdev_path[strlen(vc->vdev_path) - 2];
8674 
8675 		if (strcmp(vc->vdev_path, path) == 0)
8676 			return (vc);
8677 		if (strcmp(p, path) == 0)
8678 			return (vc);
8679 		if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0)
8680 			return (vc);
8681 	}
8682 
8683 	return (NULL);
8684 }
8685 
8686 static int
name_from_objset_id(spa_t * spa,uint64_t objset_id,char * outstr)8687 name_from_objset_id(spa_t *spa, uint64_t objset_id, char *outstr)
8688 {
8689 	dsl_dataset_t *ds;
8690 
8691 	dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
8692 	int error = dsl_dataset_hold_obj(spa->spa_dsl_pool, objset_id,
8693 	    NULL, &ds);
8694 	if (error != 0) {
8695 		(void) fprintf(stderr, "failed to hold objset %llu: %s\n",
8696 		    (u_longlong_t)objset_id, strerror(error));
8697 		dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8698 		return (error);
8699 	}
8700 	dsl_dataset_name(ds, outstr);
8701 	dsl_dataset_rele(ds, NULL);
8702 	dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
8703 	return (0);
8704 }
8705 
8706 static boolean_t
zdb_parse_block_sizes(char * sizes,uint64_t * lsize,uint64_t * psize)8707 zdb_parse_block_sizes(char *sizes, uint64_t *lsize, uint64_t *psize)
8708 {
8709 	char *s0, *s1, *tmp = NULL;
8710 
8711 	if (sizes == NULL)
8712 		return (B_FALSE);
8713 
8714 	s0 = strtok_r(sizes, "/", &tmp);
8715 	if (s0 == NULL)
8716 		return (B_FALSE);
8717 	s1 = strtok_r(NULL, "/", &tmp);
8718 	*lsize = strtoull(s0, NULL, 16);
8719 	*psize = s1 ? strtoull(s1, NULL, 16) : *lsize;
8720 	return (*lsize >= *psize && *psize > 0);
8721 }
8722 
8723 #define	ZIO_COMPRESS_MASK(alg)	(1ULL << (ZIO_COMPRESS_##alg))
8724 
8725 static boolean_t
try_decompress_block(abd_t * pabd,uint64_t lsize,uint64_t psize,int flags,int cfunc,void * lbuf,void * lbuf2)8726 try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize,
8727     int flags, int cfunc, void *lbuf, void *lbuf2)
8728 {
8729 	if (flags & ZDB_FLAG_VERBOSE) {
8730 		(void) fprintf(stderr,
8731 		    "Trying %05llx -> %05llx (%s)\n",
8732 		    (u_longlong_t)psize,
8733 		    (u_longlong_t)lsize,
8734 		    zio_compress_table[cfunc].ci_name);
8735 	}
8736 
8737 	/*
8738 	 * We set lbuf to all zeros and lbuf2 to all
8739 	 * ones, then decompress to both buffers and
8740 	 * compare their contents. This way we can
8741 	 * know if decompression filled exactly to
8742 	 * lsize or if it left some bytes unwritten.
8743 	 */
8744 
8745 	memset(lbuf, 0x00, lsize);
8746 	memset(lbuf2, 0xff, lsize);
8747 
8748 	abd_t labd, labd2;
8749 	abd_get_from_buf_struct(&labd, lbuf, lsize);
8750 	abd_get_from_buf_struct(&labd2, lbuf2, lsize);
8751 
8752 	boolean_t ret = B_FALSE;
8753 	if (zio_decompress_data(cfunc, pabd,
8754 	    &labd, psize, lsize, NULL) == 0 &&
8755 	    zio_decompress_data(cfunc, pabd,
8756 	    &labd2, psize, lsize, NULL) == 0 &&
8757 	    memcmp(lbuf, lbuf2, lsize) == 0)
8758 		ret = B_TRUE;
8759 
8760 	abd_free(&labd2);
8761 	abd_free(&labd);
8762 
8763 	return (ret);
8764 }
8765 
8766 static uint64_t
zdb_decompress_block(abd_t * pabd,void * buf,void * lbuf,uint64_t lsize,uint64_t psize,int flags)8767 zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
8768     uint64_t psize, int flags)
8769 {
8770 	(void) buf;
8771 	uint64_t orig_lsize = lsize;
8772 	boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL));
8773 	boolean_t found = B_FALSE;
8774 	/*
8775 	 * We don't know how the data was compressed, so just try
8776 	 * every decompress function at every inflated blocksize.
8777 	 */
8778 	void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
8779 	int cfuncs[ZIO_COMPRESS_FUNCTIONS] = { 0 };
8780 	int *cfuncp = cfuncs;
8781 	uint64_t maxlsize = SPA_MAXBLOCKSIZE;
8782 	uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) |
8783 	    ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) |
8784 	    ZIO_COMPRESS_MASK(ZLE);
8785 	*cfuncp++ = ZIO_COMPRESS_LZ4;
8786 	*cfuncp++ = ZIO_COMPRESS_LZJB;
8787 	mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB);
8788 	/*
8789 	 * Every gzip level has the same decompressor, no need to
8790 	 * run it 9 times per bruteforce attempt.
8791 	 */
8792 	mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3);
8793 	mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5);
8794 	mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7);
8795 	mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9);
8796 	for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++)
8797 		if (((1ULL << c) & mask) == 0)
8798 			*cfuncp++ = c;
8799 
8800 	/*
8801 	 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this
8802 	 * could take a while and we should let the user know
8803 	 * we are not stuck.  On the other hand, printing progress
8804 	 * info gets old after a while.  User can specify 'v' flag
8805 	 * to see the progression.
8806 	 */
8807 	if (lsize == psize)
8808 		lsize += SPA_MINBLOCKSIZE;
8809 	else
8810 		maxlsize = lsize;
8811 
8812 	for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) {
8813 		for (cfuncp = cfuncs; *cfuncp; cfuncp++) {
8814 			if (try_decompress_block(pabd, lsize, psize, flags,
8815 			    *cfuncp, lbuf, lbuf2)) {
8816 				found = B_TRUE;
8817 				break;
8818 			}
8819 		}
8820 		if (*cfuncp != 0)
8821 			break;
8822 	}
8823 	if (!found && tryzle) {
8824 		for (lsize = orig_lsize; lsize <= maxlsize;
8825 		    lsize += SPA_MINBLOCKSIZE) {
8826 			if (try_decompress_block(pabd, lsize, psize, flags,
8827 			    ZIO_COMPRESS_ZLE, lbuf, lbuf2)) {
8828 				*cfuncp = ZIO_COMPRESS_ZLE;
8829 				found = B_TRUE;
8830 				break;
8831 			}
8832 		}
8833 	}
8834 	umem_free(lbuf2, SPA_MAXBLOCKSIZE);
8835 
8836 	if (*cfuncp == ZIO_COMPRESS_ZLE) {
8837 		printf("\nZLE decompression was selected. If you "
8838 		    "suspect the results are wrong,\ntry avoiding ZLE "
8839 		    "by setting and exporting ZDB_NO_ZLE=\"true\"\n");
8840 	}
8841 
8842 	return (lsize > maxlsize ? -1 : lsize);
8843 }
8844 
8845 /*
8846  * Read a block from a pool and print it out.  The syntax of the
8847  * block descriptor is:
8848  *
8849  *	pool:vdev_specifier:offset:[lsize/]psize[:flags]
8850  *
8851  *	pool           - The name of the pool you wish to read from
8852  *	vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
8853  *	offset         - offset, in hex, in bytes
8854  *	size           - Amount of data to read, in hex, in bytes
8855  *	flags          - A string of characters specifying options
8856  *		 b: Decode a blkptr at given offset within block
8857  *		 c: Calculate and display checksums
8858  *		 d: Decompress data before dumping
8859  *		 e: Byteswap data before dumping
8860  *		 g: Display data as a gang block header
8861  *		 i: Display as an indirect block
8862  *		 r: Dump raw data to stdout
8863  *		 v: Verbose
8864  *
8865  */
8866 static void
zdb_read_block(char * thing,spa_t * spa)8867 zdb_read_block(char *thing, spa_t *spa)
8868 {
8869 	blkptr_t blk, *bp = &blk;
8870 	dva_t *dva = bp->blk_dva;
8871 	int flags = 0;
8872 	uint64_t offset = 0, psize = 0, lsize = 0, blkptr_offset = 0;
8873 	zio_t *zio;
8874 	vdev_t *vd;
8875 	abd_t *pabd;
8876 	void *lbuf, *buf;
8877 	char *s, *p, *dup, *flagstr, *sizes, *tmp = NULL;
8878 	const char *vdev, *errmsg = NULL;
8879 	int i, len, error;
8880 	boolean_t borrowed = B_FALSE, found = B_FALSE;
8881 
8882 	dup = strdup(thing);
8883 	s = strtok_r(dup, ":", &tmp);
8884 	vdev = s ?: "";
8885 	s = strtok_r(NULL, ":", &tmp);
8886 	offset = strtoull(s ? s : "", NULL, 16);
8887 	sizes = strtok_r(NULL, ":", &tmp);
8888 	s = strtok_r(NULL, ":", &tmp);
8889 	flagstr = strdup(s ?: "");
8890 
8891 	if (!zdb_parse_block_sizes(sizes, &lsize, &psize))
8892 		errmsg = "invalid size(s)";
8893 	if (!IS_P2ALIGNED(psize, DEV_BSIZE) || !IS_P2ALIGNED(lsize, DEV_BSIZE))
8894 		errmsg = "size must be a multiple of sector size";
8895 	if (!IS_P2ALIGNED(offset, DEV_BSIZE))
8896 		errmsg = "offset must be a multiple of sector size";
8897 	if (errmsg) {
8898 		(void) printf("Invalid block specifier: %s  - %s\n",
8899 		    thing, errmsg);
8900 		goto done;
8901 	}
8902 
8903 	tmp = NULL;
8904 	for (s = strtok_r(flagstr, ":", &tmp);
8905 	    s != NULL;
8906 	    s = strtok_r(NULL, ":", &tmp)) {
8907 		len = strlen(flagstr);
8908 		for (i = 0; i < len; i++) {
8909 			int bit = flagbits[(uchar_t)flagstr[i]];
8910 
8911 			if (bit == 0) {
8912 				(void) printf("***Ignoring flag: %c\n",
8913 				    (uchar_t)flagstr[i]);
8914 				continue;
8915 			}
8916 			found = B_TRUE;
8917 			flags |= bit;
8918 
8919 			p = &flagstr[i + 1];
8920 			if (*p != ':' && *p != '\0') {
8921 				int j = 0, nextbit = flagbits[(uchar_t)*p];
8922 				char *end, offstr[8] = { 0 };
8923 				if ((bit == ZDB_FLAG_PRINT_BLKPTR) &&
8924 				    (nextbit == 0)) {
8925 					/* look ahead to isolate the offset */
8926 					while (nextbit == 0 &&
8927 					    strchr(flagbitstr, *p) == NULL) {
8928 						offstr[j] = *p;
8929 						j++;
8930 						if (i + j > strlen(flagstr))
8931 							break;
8932 						p++;
8933 						nextbit = flagbits[(uchar_t)*p];
8934 					}
8935 					blkptr_offset = strtoull(offstr, &end,
8936 					    16);
8937 					i += j;
8938 				} else if (nextbit == 0) {
8939 					(void) printf("***Ignoring flag arg:"
8940 					    " '%c'\n", (uchar_t)*p);
8941 				}
8942 			}
8943 		}
8944 	}
8945 	if (blkptr_offset % sizeof (blkptr_t)) {
8946 		printf("Block pointer offset 0x%llx "
8947 		    "must be divisible by 0x%x\n",
8948 		    (longlong_t)blkptr_offset, (int)sizeof (blkptr_t));
8949 		goto done;
8950 	}
8951 	if (found == B_FALSE && strlen(flagstr) > 0) {
8952 		printf("Invalid flag arg: '%s'\n", flagstr);
8953 		goto done;
8954 	}
8955 
8956 	vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev);
8957 	if (vd == NULL) {
8958 		(void) printf("***Invalid vdev: %s\n", vdev);
8959 		goto done;
8960 	} else {
8961 		if (vd->vdev_path)
8962 			(void) fprintf(stderr, "Found vdev: %s\n",
8963 			    vd->vdev_path);
8964 		else
8965 			(void) fprintf(stderr, "Found vdev type: %s\n",
8966 			    vd->vdev_ops->vdev_op_type);
8967 	}
8968 
8969 	pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
8970 	lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
8971 
8972 	BP_ZERO(bp);
8973 
8974 	DVA_SET_VDEV(&dva[0], vd->vdev_id);
8975 	DVA_SET_OFFSET(&dva[0], offset);
8976 	DVA_SET_GANG(&dva[0], !!(flags & ZDB_FLAG_GBH));
8977 	DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
8978 
8979 	BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
8980 
8981 	BP_SET_LSIZE(bp, lsize);
8982 	BP_SET_PSIZE(bp, psize);
8983 	BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF);
8984 	BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF);
8985 	BP_SET_TYPE(bp, DMU_OT_NONE);
8986 	BP_SET_LEVEL(bp, 0);
8987 	BP_SET_DEDUP(bp, 0);
8988 	BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
8989 
8990 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8991 	zio = zio_root(spa, NULL, NULL, 0);
8992 
8993 	if (vd == vd->vdev_top) {
8994 		/*
8995 		 * Treat this as a normal block read.
8996 		 */
8997 		zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL,
8998 		    ZIO_PRIORITY_SYNC_READ,
8999 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL));
9000 	} else {
9001 		/*
9002 		 * Treat this as a vdev child I/O.
9003 		 */
9004 		zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd,
9005 		    psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ,
9006 		    ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
9007 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | ZIO_FLAG_OPTIONAL,
9008 		    NULL, NULL));
9009 	}
9010 
9011 	error = zio_wait(zio);
9012 	spa_config_exit(spa, SCL_STATE, FTAG);
9013 
9014 	if (error) {
9015 		(void) printf("Read of %s failed, error: %d\n", thing, error);
9016 		goto out;
9017 	}
9018 
9019 	uint64_t orig_lsize = lsize;
9020 	buf = lbuf;
9021 	if (flags & ZDB_FLAG_DECOMPRESS) {
9022 		lsize = zdb_decompress_block(pabd, buf, lbuf,
9023 		    lsize, psize, flags);
9024 		if (lsize == -1) {
9025 			(void) printf("Decompress of %s failed\n", thing);
9026 			goto out;
9027 		}
9028 	} else {
9029 		buf = abd_borrow_buf_copy(pabd, lsize);
9030 		borrowed = B_TRUE;
9031 	}
9032 	/*
9033 	 * Try to detect invalid block pointer.  If invalid, try
9034 	 * decompressing.
9035 	 */
9036 	if ((flags & ZDB_FLAG_PRINT_BLKPTR || flags & ZDB_FLAG_INDIRECT) &&
9037 	    !(flags & ZDB_FLAG_DECOMPRESS)) {
9038 		const blkptr_t *b = (const blkptr_t *)(void *)
9039 		    ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9040 		if (zfs_blkptr_verify(spa, b,
9041 		    BLK_CONFIG_NEEDED, BLK_VERIFY_ONLY) == B_FALSE) {
9042 			abd_return_buf_copy(pabd, buf, lsize);
9043 			borrowed = B_FALSE;
9044 			buf = lbuf;
9045 			lsize = zdb_decompress_block(pabd, buf,
9046 			    lbuf, lsize, psize, flags);
9047 			b = (const blkptr_t *)(void *)
9048 			    ((uintptr_t)buf + (uintptr_t)blkptr_offset);
9049 			if (lsize == -1 || zfs_blkptr_verify(spa, b,
9050 			    BLK_CONFIG_NEEDED, BLK_VERIFY_LOG) == B_FALSE) {
9051 				printf("invalid block pointer at this DVA\n");
9052 				goto out;
9053 			}
9054 		}
9055 	}
9056 
9057 	if (flags & ZDB_FLAG_PRINT_BLKPTR)
9058 		zdb_print_blkptr((blkptr_t *)(void *)
9059 		    ((uintptr_t)buf + (uintptr_t)blkptr_offset), flags);
9060 	else if (flags & ZDB_FLAG_RAW)
9061 		zdb_dump_block_raw(buf, lsize, flags);
9062 	else if (flags & ZDB_FLAG_INDIRECT)
9063 		zdb_dump_indirect((blkptr_t *)buf,
9064 		    orig_lsize / sizeof (blkptr_t), flags);
9065 	else if (flags & ZDB_FLAG_GBH)
9066 		zdb_dump_gbh(buf, flags);
9067 	else
9068 		zdb_dump_block(thing, buf, lsize, flags);
9069 
9070 	/*
9071 	 * If :c was specified, iterate through the checksum table to
9072 	 * calculate and display each checksum for our specified
9073 	 * DVA and length.
9074 	 */
9075 	if ((flags & ZDB_FLAG_CHECKSUM) && !(flags & ZDB_FLAG_RAW) &&
9076 	    !(flags & ZDB_FLAG_GBH)) {
9077 		zio_t *czio;
9078 		(void) printf("\n");
9079 		for (enum zio_checksum ck = ZIO_CHECKSUM_LABEL;
9080 		    ck < ZIO_CHECKSUM_FUNCTIONS; ck++) {
9081 
9082 			if ((zio_checksum_table[ck].ci_flags &
9083 			    ZCHECKSUM_FLAG_EMBEDDED) ||
9084 			    ck == ZIO_CHECKSUM_NOPARITY) {
9085 				continue;
9086 			}
9087 			BP_SET_CHECKSUM(bp, ck);
9088 			spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
9089 			czio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
9090 			if (vd == vd->vdev_top) {
9091 				zio_nowait(zio_read(czio, spa, bp, pabd, psize,
9092 				    NULL, NULL,
9093 				    ZIO_PRIORITY_SYNC_READ,
9094 				    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9095 				    ZIO_FLAG_DONT_RETRY, NULL));
9096 			} else {
9097 				zio_nowait(zio_vdev_child_io(czio, bp, vd,
9098 				    offset, pabd, psize, ZIO_TYPE_READ,
9099 				    ZIO_PRIORITY_SYNC_READ,
9100 				    ZIO_FLAG_DONT_PROPAGATE |
9101 				    ZIO_FLAG_DONT_RETRY |
9102 				    ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW |
9103 				    ZIO_FLAG_SPECULATIVE |
9104 				    ZIO_FLAG_OPTIONAL, NULL, NULL));
9105 			}
9106 			error = zio_wait(czio);
9107 			if (error == 0 || error == ECKSUM) {
9108 				zio_t *ck_zio = zio_null(NULL, spa, NULL,
9109 				    NULL, NULL, 0);
9110 				ck_zio->io_offset =
9111 				    DVA_GET_OFFSET(&bp->blk_dva[0]);
9112 				ck_zio->io_bp = bp;
9113 				zio_checksum_compute(ck_zio, ck, pabd, lsize);
9114 				printf(
9115 				    "%12s\t"
9116 				    "cksum=%016llx:%016llx:%016llx:%016llx\n",
9117 				    zio_checksum_table[ck].ci_name,
9118 				    (u_longlong_t)bp->blk_cksum.zc_word[0],
9119 				    (u_longlong_t)bp->blk_cksum.zc_word[1],
9120 				    (u_longlong_t)bp->blk_cksum.zc_word[2],
9121 				    (u_longlong_t)bp->blk_cksum.zc_word[3]);
9122 				zio_wait(ck_zio);
9123 			} else {
9124 				printf("error %d reading block\n", error);
9125 			}
9126 			spa_config_exit(spa, SCL_STATE, FTAG);
9127 		}
9128 	}
9129 
9130 	if (borrowed)
9131 		abd_return_buf_copy(pabd, buf, lsize);
9132 
9133 out:
9134 	abd_free(pabd);
9135 	umem_free(lbuf, SPA_MAXBLOCKSIZE);
9136 done:
9137 	free(flagstr);
9138 	free(dup);
9139 }
9140 
9141 static void
zdb_embedded_block(char * thing)9142 zdb_embedded_block(char *thing)
9143 {
9144 	blkptr_t bp = {{{{0}}}};
9145 	unsigned long long *words = (void *)&bp;
9146 	char *buf;
9147 	int err;
9148 
9149 	err = sscanf(thing, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:"
9150 	    "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx",
9151 	    words + 0, words + 1, words + 2, words + 3,
9152 	    words + 4, words + 5, words + 6, words + 7,
9153 	    words + 8, words + 9, words + 10, words + 11,
9154 	    words + 12, words + 13, words + 14, words + 15);
9155 	if (err != 16) {
9156 		(void) fprintf(stderr, "invalid input format\n");
9157 		zdb_exit(1);
9158 	}
9159 	ASSERT3U(BPE_GET_LSIZE(&bp), <=, SPA_MAXBLOCKSIZE);
9160 	buf = malloc(SPA_MAXBLOCKSIZE);
9161 	if (buf == NULL) {
9162 		(void) fprintf(stderr, "out of memory\n");
9163 		zdb_exit(1);
9164 	}
9165 	err = decode_embedded_bp(&bp, buf, BPE_GET_LSIZE(&bp));
9166 	if (err != 0) {
9167 		(void) fprintf(stderr, "decode failed: %u\n", err);
9168 		zdb_exit(1);
9169 	}
9170 	zdb_dump_block_raw(buf, BPE_GET_LSIZE(&bp), 0);
9171 	free(buf);
9172 }
9173 
9174 /* check for valid hex or decimal numeric string */
9175 static boolean_t
zdb_numeric(char * str)9176 zdb_numeric(char *str)
9177 {
9178 	int i = 0, len;
9179 
9180 	len = strlen(str);
9181 	if (len == 0)
9182 		return (B_FALSE);
9183 	if (strncmp(str, "0x", 2) == 0 || strncmp(str, "0X", 2) == 0)
9184 		i = 2;
9185 	for (; i < len; i++) {
9186 		if (!isxdigit(str[i]))
9187 			return (B_FALSE);
9188 	}
9189 	return (B_TRUE);
9190 }
9191 
9192 static int
dummy_get_file_info(dmu_object_type_t bonustype,const void * data,zfs_file_info_t * zoi)9193 dummy_get_file_info(dmu_object_type_t bonustype, const void *data,
9194     zfs_file_info_t *zoi)
9195 {
9196 	(void) data, (void) zoi;
9197 
9198 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
9199 		return (ENOENT);
9200 
9201 	(void) fprintf(stderr, "dummy_get_file_info: not implemented");
9202 	abort();
9203 }
9204 
9205 int
main(int argc,char ** argv)9206 main(int argc, char **argv)
9207 {
9208 	int c;
9209 	int dump_all = 1;
9210 	int verbose = 0;
9211 	int error = 0;
9212 	char **searchdirs = NULL;
9213 	int nsearch = 0;
9214 	char *target, *target_pool, dsname[ZFS_MAX_DATASET_NAME_LEN];
9215 	nvlist_t *policy = NULL;
9216 	uint64_t max_txg = UINT64_MAX;
9217 	int64_t objset_id = -1;
9218 	uint64_t object;
9219 	int flags = ZFS_IMPORT_MISSING_LOG;
9220 	int rewind = ZPOOL_NEVER_REWIND;
9221 	char *spa_config_path_env, *objset_str;
9222 	boolean_t target_is_spa = B_TRUE, dataset_lookup = B_FALSE;
9223 	nvlist_t *cfg = NULL;
9224 	struct sigaction action;
9225 	boolean_t force_import = B_FALSE;
9226 	boolean_t config_path_console = B_FALSE;
9227 	char pbuf[MAXPATHLEN];
9228 
9229 	dprintf_setup(&argc, argv);
9230 
9231 	/*
9232 	 * Set up signal handlers, so if we crash due to bad on-disk data we
9233 	 * can get more info. Unlike ztest, we don't bail out if we can't set
9234 	 * up signal handlers, because zdb is very useful without them.
9235 	 */
9236 	action.sa_handler = sig_handler;
9237 	sigemptyset(&action.sa_mask);
9238 	action.sa_flags = 0;
9239 	if (sigaction(SIGSEGV, &action, NULL) < 0) {
9240 		(void) fprintf(stderr, "zdb: cannot catch SIGSEGV: %s\n",
9241 		    strerror(errno));
9242 	}
9243 	if (sigaction(SIGABRT, &action, NULL) < 0) {
9244 		(void) fprintf(stderr, "zdb: cannot catch SIGABRT: %s\n",
9245 		    strerror(errno));
9246 	}
9247 
9248 	/*
9249 	 * If there is an environment variable SPA_CONFIG_PATH it overrides
9250 	 * default spa_config_path setting. If -U flag is specified it will
9251 	 * override this environment variable settings once again.
9252 	 */
9253 	spa_config_path_env = getenv("SPA_CONFIG_PATH");
9254 	if (spa_config_path_env != NULL)
9255 		spa_config_path = spa_config_path_env;
9256 
9257 	/*
9258 	 * For performance reasons, we set this tunable down. We do so before
9259 	 * the arg parsing section so that the user can override this value if
9260 	 * they choose.
9261 	 */
9262 	zfs_btree_verify_intensity = 3;
9263 
9264 	struct option long_options[] = {
9265 		{"ignore-assertions",	no_argument,		NULL, 'A'},
9266 		{"block-stats",		no_argument,		NULL, 'b'},
9267 		{"backup",		no_argument,		NULL, 'B'},
9268 		{"checksum",		no_argument,		NULL, 'c'},
9269 		{"config",		no_argument,		NULL, 'C'},
9270 		{"datasets",		no_argument,		NULL, 'd'},
9271 		{"dedup-stats",		no_argument,		NULL, 'D'},
9272 		{"exported",		no_argument,		NULL, 'e'},
9273 		{"embedded-block-pointer",	no_argument,	NULL, 'E'},
9274 		{"automatic-rewind",	no_argument,		NULL, 'F'},
9275 		{"dump-debug-msg",	no_argument,		NULL, 'G'},
9276 		{"history",		no_argument,		NULL, 'h'},
9277 		{"intent-logs",		no_argument,		NULL, 'i'},
9278 		{"inflight",		required_argument,	NULL, 'I'},
9279 		{"checkpointed-state",	no_argument,		NULL, 'k'},
9280 		{"key",			required_argument,	NULL, 'K'},
9281 		{"label",		no_argument,		NULL, 'l'},
9282 		{"disable-leak-tracking",	no_argument,	NULL, 'L'},
9283 		{"metaslabs",		no_argument,		NULL, 'm'},
9284 		{"metaslab-groups",	no_argument,		NULL, 'M'},
9285 		{"numeric",		no_argument,		NULL, 'N'},
9286 		{"option",		required_argument,	NULL, 'o'},
9287 		{"object-lookups",	no_argument,		NULL, 'O'},
9288 		{"path",		required_argument,	NULL, 'p'},
9289 		{"parseable",		no_argument,		NULL, 'P'},
9290 		{"skip-label",		no_argument,		NULL, 'q'},
9291 		{"copy-object",		no_argument,		NULL, 'r'},
9292 		{"read-block",		no_argument,		NULL, 'R'},
9293 		{"io-stats",		no_argument,		NULL, 's'},
9294 		{"simulate-dedup",	no_argument,		NULL, 'S'},
9295 		{"txg",			required_argument,	NULL, 't'},
9296 		{"brt-stats",		no_argument,		NULL, 'T'},
9297 		{"uberblock",		no_argument,		NULL, 'u'},
9298 		{"cachefile",		required_argument,	NULL, 'U'},
9299 		{"verbose",		no_argument,		NULL, 'v'},
9300 		{"verbatim",		no_argument,		NULL, 'V'},
9301 		{"dump-blocks",		required_argument,	NULL, 'x'},
9302 		{"extreme-rewind",	no_argument,		NULL, 'X'},
9303 		{"all-reconstruction",	no_argument,		NULL, 'Y'},
9304 		{"livelist",		no_argument,		NULL, 'y'},
9305 		{"zstd-headers",	no_argument,		NULL, 'Z'},
9306 		{0, 0, 0, 0}
9307 	};
9308 
9309 	while ((c = getopt_long(argc, argv,
9310 	    "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
9311 	    long_options, NULL)) != -1) {
9312 		switch (c) {
9313 		case 'b':
9314 		case 'B':
9315 		case 'c':
9316 		case 'C':
9317 		case 'd':
9318 		case 'D':
9319 		case 'E':
9320 		case 'G':
9321 		case 'h':
9322 		case 'i':
9323 		case 'l':
9324 		case 'm':
9325 		case 'M':
9326 		case 'N':
9327 		case 'O':
9328 		case 'r':
9329 		case 'R':
9330 		case 's':
9331 		case 'S':
9332 		case 'T':
9333 		case 'u':
9334 		case 'y':
9335 		case 'Z':
9336 			dump_opt[c]++;
9337 			dump_all = 0;
9338 			break;
9339 		case 'A':
9340 		case 'e':
9341 		case 'F':
9342 		case 'k':
9343 		case 'L':
9344 		case 'P':
9345 		case 'q':
9346 		case 'X':
9347 			dump_opt[c]++;
9348 			break;
9349 		case 'Y':
9350 			zfs_reconstruct_indirect_combinations_max = INT_MAX;
9351 			zfs_deadman_enabled = 0;
9352 			break;
9353 		/* NB: Sort single match options below. */
9354 		case 'I':
9355 			max_inflight_bytes = strtoull(optarg, NULL, 0);
9356 			if (max_inflight_bytes == 0) {
9357 				(void) fprintf(stderr, "maximum number "
9358 				    "of inflight bytes must be greater "
9359 				    "than 0\n");
9360 				usage();
9361 			}
9362 			break;
9363 		case 'K':
9364 			dump_opt[c]++;
9365 			key_material = strdup(optarg);
9366 			/* redact key material in process table */
9367 			while (*optarg != '\0') { *optarg++ = '*'; }
9368 			break;
9369 		case 'o':
9370 			error = set_global_var(optarg);
9371 			if (error != 0)
9372 				usage();
9373 			break;
9374 		case 'p':
9375 			if (searchdirs == NULL) {
9376 				searchdirs = umem_alloc(sizeof (char *),
9377 				    UMEM_NOFAIL);
9378 			} else {
9379 				char **tmp = umem_alloc((nsearch + 1) *
9380 				    sizeof (char *), UMEM_NOFAIL);
9381 				memcpy(tmp, searchdirs, nsearch *
9382 				    sizeof (char *));
9383 				umem_free(searchdirs,
9384 				    nsearch * sizeof (char *));
9385 				searchdirs = tmp;
9386 			}
9387 			searchdirs[nsearch++] = optarg;
9388 			break;
9389 		case 't':
9390 			max_txg = strtoull(optarg, NULL, 0);
9391 			if (max_txg < TXG_INITIAL) {
9392 				(void) fprintf(stderr, "incorrect txg "
9393 				    "specified: %s\n", optarg);
9394 				usage();
9395 			}
9396 			break;
9397 		case 'U':
9398 			config_path_console = B_TRUE;
9399 			spa_config_path = optarg;
9400 			if (spa_config_path[0] != '/') {
9401 				(void) fprintf(stderr,
9402 				    "cachefile must be an absolute path "
9403 				    "(i.e. start with a slash)\n");
9404 				usage();
9405 			}
9406 			break;
9407 		case 'v':
9408 			verbose++;
9409 			break;
9410 		case 'V':
9411 			flags = ZFS_IMPORT_VERBATIM;
9412 			break;
9413 		case 'x':
9414 			vn_dumpdir = optarg;
9415 			break;
9416 		default:
9417 			usage();
9418 			break;
9419 		}
9420 	}
9421 
9422 	if (!dump_opt['e'] && searchdirs != NULL) {
9423 		(void) fprintf(stderr, "-p option requires use of -e\n");
9424 		usage();
9425 	}
9426 #if defined(_LP64)
9427 	/*
9428 	 * ZDB does not typically re-read blocks; therefore limit the ARC
9429 	 * to 256 MB, which can be used entirely for metadata.
9430 	 */
9431 	zfs_arc_min = 2ULL << SPA_MAXBLOCKSHIFT;
9432 	zfs_arc_max = 256 * 1024 * 1024;
9433 #endif
9434 
9435 	/*
9436 	 * "zdb -c" uses checksum-verifying scrub i/os which are async reads.
9437 	 * "zdb -b" uses traversal prefetch which uses async reads.
9438 	 * For good performance, let several of them be active at once.
9439 	 */
9440 	zfs_vdev_async_read_max_active = 10;
9441 
9442 	/*
9443 	 * Disable reference tracking for better performance.
9444 	 */
9445 	reference_tracking_enable = B_FALSE;
9446 
9447 	/*
9448 	 * Do not fail spa_load when spa_load_verify fails. This is needed
9449 	 * to load non-idle pools.
9450 	 */
9451 	spa_load_verify_dryrun = B_TRUE;
9452 
9453 	/*
9454 	 * ZDB should have ability to read spacemaps.
9455 	 */
9456 	spa_mode_readable_spacemaps = B_TRUE;
9457 
9458 	if (dump_all)
9459 		verbose = MAX(verbose, 1);
9460 
9461 	for (c = 0; c < 256; c++) {
9462 		if (dump_all && strchr("ABeEFkKlLNOPrRSXy", c) == NULL)
9463 			dump_opt[c] = 1;
9464 		if (dump_opt[c])
9465 			dump_opt[c] += verbose;
9466 	}
9467 
9468 	libspl_set_assert_ok((dump_opt['A'] == 1) || (dump_opt['A'] > 2));
9469 	zfs_recover = (dump_opt['A'] > 1);
9470 
9471 	argc -= optind;
9472 	argv += optind;
9473 	if (argc < 2 && dump_opt['R'])
9474 		usage();
9475 
9476 	target = argv[0];
9477 
9478 	/*
9479 	 * Automate cachefile
9480 	 */
9481 	if (!spa_config_path_env && !config_path_console && target &&
9482 	    libzfs_core_init() == 0) {
9483 		char *pname = strdup(target);
9484 		const char *value;
9485 		nvlist_t *pnvl = NULL;
9486 		nvlist_t *vnvl = NULL;
9487 
9488 		if (strpbrk(pname, "/@") != NULL)
9489 			*strpbrk(pname, "/@") = '\0';
9490 
9491 		if (pname && lzc_get_props(pname, &pnvl) == 0) {
9492 			if (nvlist_lookup_nvlist(pnvl, "cachefile",
9493 			    &vnvl) == 0) {
9494 				value = fnvlist_lookup_string(vnvl,
9495 				    ZPROP_VALUE);
9496 			} else {
9497 				value = "-";
9498 			}
9499 			strlcpy(pbuf, value, sizeof (pbuf));
9500 			if (pbuf[0] != '\0') {
9501 				if (pbuf[0] == '/') {
9502 					if (access(pbuf, F_OK) == 0)
9503 						spa_config_path = pbuf;
9504 					else
9505 						force_import = B_TRUE;
9506 				} else if ((strcmp(pbuf, "-") == 0 &&
9507 				    access(ZPOOL_CACHE, F_OK) != 0) ||
9508 				    strcmp(pbuf, "none") == 0) {
9509 					force_import = B_TRUE;
9510 				}
9511 			}
9512 			nvlist_free(vnvl);
9513 		}
9514 
9515 		free(pname);
9516 		nvlist_free(pnvl);
9517 		libzfs_core_fini();
9518 	}
9519 
9520 	dmu_objset_register_type(DMU_OST_ZFS, dummy_get_file_info);
9521 	kernel_init(SPA_MODE_READ);
9522 	kernel_init_done = B_TRUE;
9523 
9524 	if (dump_opt['E']) {
9525 		if (argc != 1)
9526 			usage();
9527 		zdb_embedded_block(argv[0]);
9528 		error = 0;
9529 		goto fini;
9530 	}
9531 
9532 	if (argc < 1) {
9533 		if (!dump_opt['e'] && dump_opt['C']) {
9534 			dump_cachefile(spa_config_path);
9535 			error = 0;
9536 			goto fini;
9537 		}
9538 		usage();
9539 	}
9540 
9541 	if (dump_opt['l']) {
9542 		error = dump_label(argv[0]);
9543 		goto fini;
9544 	}
9545 
9546 	if (dump_opt['X'] || dump_opt['F'])
9547 		rewind = ZPOOL_DO_REWIND |
9548 		    (dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
9549 
9550 	/* -N implies -d */
9551 	if (dump_opt['N'] && dump_opt['d'] == 0)
9552 		dump_opt['d'] = dump_opt['N'];
9553 
9554 	if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 ||
9555 	    nvlist_add_uint64(policy, ZPOOL_LOAD_REQUEST_TXG, max_txg) != 0 ||
9556 	    nvlist_add_uint32(policy, ZPOOL_LOAD_REWIND_POLICY, rewind) != 0)
9557 		fatal("internal error: %s", strerror(ENOMEM));
9558 
9559 	error = 0;
9560 
9561 	if (strpbrk(target, "/@") != NULL) {
9562 		size_t targetlen;
9563 
9564 		target_pool = strdup(target);
9565 		*strpbrk(target_pool, "/@") = '\0';
9566 
9567 		target_is_spa = B_FALSE;
9568 		targetlen = strlen(target);
9569 		if (targetlen && target[targetlen - 1] == '/')
9570 			target[targetlen - 1] = '\0';
9571 
9572 		/*
9573 		 * See if an objset ID was supplied (-d <pool>/<objset ID>).
9574 		 * To disambiguate tank/100, consider the 100 as objsetID
9575 		 * if -N was given, otherwise 100 is an objsetID iff
9576 		 * tank/100 as a named dataset fails on lookup.
9577 		 */
9578 		objset_str = strchr(target, '/');
9579 		if (objset_str && strlen(objset_str) > 1 &&
9580 		    zdb_numeric(objset_str + 1)) {
9581 			char *endptr;
9582 			errno = 0;
9583 			objset_str++;
9584 			objset_id = strtoull(objset_str, &endptr, 0);
9585 			/* dataset 0 is the same as opening the pool */
9586 			if (errno == 0 && endptr != objset_str &&
9587 			    objset_id != 0) {
9588 				if (dump_opt['N'])
9589 					dataset_lookup = B_TRUE;
9590 			}
9591 			/* normal dataset name not an objset ID */
9592 			if (endptr == objset_str) {
9593 				objset_id = -1;
9594 			}
9595 		} else if (objset_str && !zdb_numeric(objset_str + 1) &&
9596 		    dump_opt['N']) {
9597 			printf("Supply a numeric objset ID with -N\n");
9598 			error = 1;
9599 			goto fini;
9600 		}
9601 	} else {
9602 		target_pool = target;
9603 	}
9604 
9605 	if (dump_opt['e'] || force_import) {
9606 		importargs_t args = { 0 };
9607 
9608 		/*
9609 		 * If path is not provided, search in /dev
9610 		 */
9611 		if (searchdirs == NULL) {
9612 			searchdirs = umem_alloc(sizeof (char *), UMEM_NOFAIL);
9613 			searchdirs[nsearch++] = (char *)ZFS_DEVDIR;
9614 		}
9615 
9616 		args.paths = nsearch;
9617 		args.path = searchdirs;
9618 		args.can_be_active = B_TRUE;
9619 
9620 		libpc_handle_t lpch = {
9621 			.lpc_lib_handle = NULL,
9622 			.lpc_ops = &libzpool_config_ops,
9623 			.lpc_printerr = B_TRUE
9624 		};
9625 		error = zpool_find_config(&lpch, target_pool, &cfg, &args);
9626 
9627 		if (error == 0) {
9628 
9629 			if (nvlist_add_nvlist(cfg,
9630 			    ZPOOL_LOAD_POLICY, policy) != 0) {
9631 				fatal("can't open '%s': %s",
9632 				    target, strerror(ENOMEM));
9633 			}
9634 
9635 			if (dump_opt['C'] > 1) {
9636 				(void) printf("\nConfiguration for import:\n");
9637 				dump_nvlist(cfg, 8);
9638 			}
9639 
9640 			/*
9641 			 * Disable the activity check to allow examination of
9642 			 * active pools.
9643 			 */
9644 			error = spa_import(target_pool, cfg, NULL,
9645 			    flags | ZFS_IMPORT_SKIP_MMP);
9646 		}
9647 	}
9648 
9649 	if (searchdirs != NULL) {
9650 		umem_free(searchdirs, nsearch * sizeof (char *));
9651 		searchdirs = NULL;
9652 	}
9653 
9654 	/*
9655 	 * We need to make sure to process -O option or call
9656 	 * dump_path after the -e option has been processed,
9657 	 * which imports the pool to the namespace if it's
9658 	 * not in the cachefile.
9659 	 */
9660 	if (dump_opt['O']) {
9661 		if (argc != 2)
9662 			usage();
9663 		dump_opt['v'] = verbose + 3;
9664 		error = dump_path(argv[0], argv[1], NULL);
9665 		goto fini;
9666 	}
9667 
9668 	if (dump_opt['r']) {
9669 		target_is_spa = B_FALSE;
9670 		if (argc != 3)
9671 			usage();
9672 		dump_opt['v'] = verbose;
9673 		error = dump_path(argv[0], argv[1], &object);
9674 		if (error != 0)
9675 			fatal("internal error: %s", strerror(error));
9676 	}
9677 
9678 	/*
9679 	 * import_checkpointed_state makes the assumption that the
9680 	 * target pool that we pass it is already part of the spa
9681 	 * namespace. Because of that we need to make sure to call
9682 	 * it always after the -e option has been processed, which
9683 	 * imports the pool to the namespace if it's not in the
9684 	 * cachefile.
9685 	 */
9686 	char *checkpoint_pool = NULL;
9687 	char *checkpoint_target = NULL;
9688 	if (dump_opt['k']) {
9689 		checkpoint_pool = import_checkpointed_state(target, cfg,
9690 		    &checkpoint_target);
9691 
9692 		if (checkpoint_target != NULL)
9693 			target = checkpoint_target;
9694 	}
9695 
9696 	if (cfg != NULL) {
9697 		nvlist_free(cfg);
9698 		cfg = NULL;
9699 	}
9700 
9701 	if (target_pool != target)
9702 		free(target_pool);
9703 
9704 	if (error == 0) {
9705 		if (dump_opt['k'] && (target_is_spa || dump_opt['R'])) {
9706 			ASSERT(checkpoint_pool != NULL);
9707 			ASSERT(checkpoint_target == NULL);
9708 
9709 			error = spa_open(checkpoint_pool, &spa, FTAG);
9710 			if (error != 0) {
9711 				fatal("Tried to open pool \"%s\" but "
9712 				    "spa_open() failed with error %d\n",
9713 				    checkpoint_pool, error);
9714 			}
9715 
9716 		} else if (target_is_spa || dump_opt['R'] || dump_opt['B'] ||
9717 		    objset_id == 0) {
9718 			zdb_set_skip_mmp(target);
9719 			error = spa_open_rewind(target, &spa, FTAG, policy,
9720 			    NULL);
9721 			if (error) {
9722 				/*
9723 				 * If we're missing the log device then
9724 				 * try opening the pool after clearing the
9725 				 * log state.
9726 				 */
9727 				mutex_enter(&spa_namespace_lock);
9728 				if ((spa = spa_lookup(target)) != NULL &&
9729 				    spa->spa_log_state == SPA_LOG_MISSING) {
9730 					spa->spa_log_state = SPA_LOG_CLEAR;
9731 					error = 0;
9732 				}
9733 				mutex_exit(&spa_namespace_lock);
9734 
9735 				if (!error) {
9736 					error = spa_open_rewind(target, &spa,
9737 					    FTAG, policy, NULL);
9738 				}
9739 			}
9740 		} else if (strpbrk(target, "#") != NULL) {
9741 			dsl_pool_t *dp;
9742 			error = dsl_pool_hold(target, FTAG, &dp);
9743 			if (error != 0) {
9744 				fatal("can't dump '%s': %s", target,
9745 				    strerror(error));
9746 			}
9747 			error = dump_bookmark(dp, target, B_TRUE, verbose > 1);
9748 			dsl_pool_rele(dp, FTAG);
9749 			if (error != 0) {
9750 				fatal("can't dump '%s': %s", target,
9751 				    strerror(error));
9752 			}
9753 			goto fini;
9754 		} else {
9755 			target_pool = strdup(target);
9756 			if (strpbrk(target, "/@") != NULL)
9757 				*strpbrk(target_pool, "/@") = '\0';
9758 
9759 			zdb_set_skip_mmp(target);
9760 			/*
9761 			 * If -N was supplied, the user has indicated that
9762 			 * zdb -d <pool>/<objsetID> is in effect.  Otherwise
9763 			 * we first assume that the dataset string is the
9764 			 * dataset name.  If dmu_objset_hold fails with the
9765 			 * dataset string, and we have an objset_id, retry the
9766 			 * lookup with the objsetID.
9767 			 */
9768 			boolean_t retry = B_TRUE;
9769 retry_lookup:
9770 			if (dataset_lookup == B_TRUE) {
9771 				/*
9772 				 * Use the supplied id to get the name
9773 				 * for open_objset.
9774 				 */
9775 				error = spa_open(target_pool, &spa, FTAG);
9776 				if (error == 0) {
9777 					error = name_from_objset_id(spa,
9778 					    objset_id, dsname);
9779 					spa_close(spa, FTAG);
9780 					if (error == 0)
9781 						target = dsname;
9782 				}
9783 			}
9784 			if (error == 0) {
9785 				if (objset_id > 0 && retry) {
9786 					int err = dmu_objset_hold(target, FTAG,
9787 					    &os);
9788 					if (err) {
9789 						dataset_lookup = B_TRUE;
9790 						retry = B_FALSE;
9791 						goto retry_lookup;
9792 					} else {
9793 						dmu_objset_rele(os, FTAG);
9794 					}
9795 				}
9796 				error = open_objset(target, FTAG, &os);
9797 			}
9798 			if (error == 0)
9799 				spa = dmu_objset_spa(os);
9800 			free(target_pool);
9801 		}
9802 	}
9803 	nvlist_free(policy);
9804 
9805 	if (error)
9806 		fatal("can't open '%s': %s", target, strerror(error));
9807 
9808 	/*
9809 	 * Set the pool failure mode to panic in order to prevent the pool
9810 	 * from suspending.  A suspended I/O will have no way to resume and
9811 	 * can prevent the zdb(8) command from terminating as expected.
9812 	 */
9813 	if (spa != NULL)
9814 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
9815 
9816 	argv++;
9817 	argc--;
9818 	if (dump_opt['r']) {
9819 		error = zdb_copy_object(os, object, argv[1]);
9820 	} else if (!dump_opt['R']) {
9821 		flagbits['d'] = ZOR_FLAG_DIRECTORY;
9822 		flagbits['f'] = ZOR_FLAG_PLAIN_FILE;
9823 		flagbits['m'] = ZOR_FLAG_SPACE_MAP;
9824 		flagbits['z'] = ZOR_FLAG_ZAP;
9825 		flagbits['A'] = ZOR_FLAG_ALL_TYPES;
9826 
9827 		if (argc > 0 && dump_opt['d']) {
9828 			zopt_object_args = argc;
9829 			zopt_object_ranges = calloc(zopt_object_args,
9830 			    sizeof (zopt_object_range_t));
9831 			for (unsigned i = 0; i < zopt_object_args; i++) {
9832 				int err;
9833 				const char *msg = NULL;
9834 
9835 				err = parse_object_range(argv[i],
9836 				    &zopt_object_ranges[i], &msg);
9837 				if (err != 0)
9838 					fatal("Bad object or range: '%s': %s\n",
9839 					    argv[i], msg ?: "");
9840 			}
9841 		} else if (argc > 0 && dump_opt['m']) {
9842 			zopt_metaslab_args = argc;
9843 			zopt_metaslab = calloc(zopt_metaslab_args,
9844 			    sizeof (uint64_t));
9845 			for (unsigned i = 0; i < zopt_metaslab_args; i++) {
9846 				errno = 0;
9847 				zopt_metaslab[i] = strtoull(argv[i], NULL, 0);
9848 				if (zopt_metaslab[i] == 0 && errno != 0)
9849 					fatal("bad number %s: %s", argv[i],
9850 					    strerror(errno));
9851 			}
9852 		}
9853 		if (dump_opt['B']) {
9854 			dump_backup(target, objset_id,
9855 			    argc > 0 ? argv[0] : NULL);
9856 		} else if (os != NULL) {
9857 			dump_objset(os);
9858 		} else if (zopt_object_args > 0 && !dump_opt['m']) {
9859 			dump_objset(spa->spa_meta_objset);
9860 		} else {
9861 			dump_zpool(spa);
9862 		}
9863 	} else {
9864 		flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR;
9865 		flagbits['c'] = ZDB_FLAG_CHECKSUM;
9866 		flagbits['d'] = ZDB_FLAG_DECOMPRESS;
9867 		flagbits['e'] = ZDB_FLAG_BSWAP;
9868 		flagbits['g'] = ZDB_FLAG_GBH;
9869 		flagbits['i'] = ZDB_FLAG_INDIRECT;
9870 		flagbits['r'] = ZDB_FLAG_RAW;
9871 		flagbits['v'] = ZDB_FLAG_VERBOSE;
9872 
9873 		for (int i = 0; i < argc; i++)
9874 			zdb_read_block(argv[i], spa);
9875 	}
9876 
9877 	if (dump_opt['k']) {
9878 		free(checkpoint_pool);
9879 		if (!target_is_spa)
9880 			free(checkpoint_target);
9881 	}
9882 
9883 fini:
9884 	if (spa != NULL)
9885 		zdb_ddt_cleanup(spa);
9886 
9887 	if (os != NULL) {
9888 		close_objset(os, FTAG);
9889 	} else if (spa != NULL) {
9890 		spa_close(spa, FTAG);
9891 	}
9892 
9893 	fuid_table_destroy();
9894 
9895 	dump_debug_buffer();
9896 
9897 	if (kernel_init_done)
9898 		kernel_fini();
9899 
9900 	return (error);
9901 }
9902