1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * CDDL HEADER START 4 * 5 * The contents of this file are subject to the terms of the 6 * Common Development and Distribution License (the "License"). 7 * You may not use this file except in compliance with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or https://opensource.org/licenses/CDDL-1.0. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 23 /* 24 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 25 * Copyright (c) 2011, 2019 by Delphix. All rights reserved. 26 * Copyright (c) 2014 Integros [integros.com] 27 * Copyright 2016 Nexenta Systems, Inc. 28 * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC. 29 * Copyright (c) 2015, 2017, Intel Corporation. 30 * Copyright (c) 2020 Datto Inc. 31 * Copyright (c) 2020, The FreeBSD Foundation [1] 32 * 33 * [1] Portions of this software were developed by Allan Jude 34 * under sponsorship from the FreeBSD Foundation. 35 * Copyright (c) 2021 Allan Jude 36 * Copyright (c) 2021 Toomas Soome <tsoome@me.com> 37 * Copyright (c) 2023, 2024, Klara Inc. 38 * Copyright (c) 2023, Rob Norris <robn@despairlabs.com> 39 */ 40 41 #include <stdio.h> 42 #include <unistd.h> 43 #include <stdlib.h> 44 #include <ctype.h> 45 #include <getopt.h> 46 #include <openssl/evp.h> 47 #include <sys/zfs_context.h> 48 #include <sys/spa.h> 49 #include <sys/spa_impl.h> 50 #include <sys/dmu.h> 51 #include <sys/zap.h> 52 #include <sys/zap_impl.h> 53 #include <sys/fs/zfs.h> 54 #include <sys/zfs_znode.h> 55 #include <sys/zfs_sa.h> 56 #include <sys/sa.h> 57 #include <sys/sa_impl.h> 58 #include <sys/vdev.h> 59 #include <sys/vdev_impl.h> 60 #include <sys/metaslab_impl.h> 61 #include <sys/dmu_objset.h> 62 #include <sys/dsl_dir.h> 63 #include <sys/dsl_dataset.h> 64 #include <sys/dsl_pool.h> 65 #include <sys/dsl_bookmark.h> 66 #include <sys/dbuf.h> 67 #include <sys/zil.h> 68 #include <sys/zil_impl.h> 69 #include <sys/stat.h> 70 #include <sys/resource.h> 71 #include <sys/dmu_send.h> 72 #include <sys/dmu_traverse.h> 73 #include <sys/zio_checksum.h> 74 #include <sys/zio_compress.h> 75 #include <sys/zfs_fuid.h> 76 #include <sys/arc.h> 77 #include <sys/arc_impl.h> 78 #include <sys/ddt.h> 79 #include <sys/ddt_impl.h> 80 #include <sys/zfeature.h> 81 #include <sys/abd.h> 82 #include <sys/blkptr.h> 83 #include <sys/dsl_crypt.h> 84 #include <sys/dsl_scan.h> 85 #include <sys/btree.h> 86 #include <sys/brt.h> 87 #include <sys/brt_impl.h> 88 #include <zfs_comutil.h> 89 #include <sys/zstd/zstd.h> 90 #include <sys/backtrace.h> 91 92 #include <libnvpair.h> 93 #include <libzutil.h> 94 #include <libzfs_core.h> 95 96 #include <libzdb.h> 97 98 #include "zdb.h" 99 100 101 extern int reference_tracking_enable; 102 extern int zfs_recover; 103 extern uint_t zfs_vdev_async_read_max_active; 104 extern boolean_t spa_load_verify_dryrun; 105 extern boolean_t spa_mode_readable_spacemaps; 106 extern uint_t zfs_reconstruct_indirect_combinations_max; 107 extern uint_t zfs_btree_verify_intensity; 108 109 static const char cmdname[] = "zdb"; 110 uint8_t dump_opt[512]; 111 112 #define ALLOCATED_OPT 256 113 114 typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size); 115 116 static uint64_t *zopt_metaslab = NULL; 117 static unsigned zopt_metaslab_args = 0; 118 119 120 static zopt_object_range_t *zopt_object_ranges = NULL; 121 static unsigned zopt_object_args = 0; 122 123 static int flagbits[256]; 124 125 126 static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */ 127 static int leaked_objects = 0; 128 static zfs_range_tree_t *mos_refd_objs; 129 static spa_t *spa; 130 static objset_t *os; 131 static boolean_t kernel_init_done; 132 static boolean_t corruption_found = B_FALSE; 133 134 static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *, 135 boolean_t); 136 static void mos_obj_refd(uint64_t); 137 static void mos_obj_refd_multiple(uint64_t); 138 static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free, 139 dmu_tx_t *tx); 140 141 142 143 static void zdb_print_blkptr(const blkptr_t *bp, int flags); 144 static void zdb_exit(int reason); 145 146 typedef struct sublivelist_verify_block_refcnt { 147 /* block pointer entry in livelist being verified */ 148 blkptr_t svbr_blk; 149 150 /* 151 * Refcount gets incremented to 1 when we encounter the first 152 * FREE entry for the svfbr block pointer and a node for it 153 * is created in our ZDB verification/tracking metadata. 154 * 155 * As we encounter more FREE entries we increment this counter 156 * and similarly decrement it whenever we find the respective 157 * ALLOC entries for this block. 158 * 159 * When the refcount gets to 0 it means that all the FREE and 160 * ALLOC entries of this block have paired up and we no longer 161 * need to track it in our verification logic (e.g. the node 162 * containing this struct in our verification data structure 163 * should be freed). 164 * 165 * [refer to sublivelist_verify_blkptr() for the actual code] 166 */ 167 uint32_t svbr_refcnt; 168 } sublivelist_verify_block_refcnt_t; 169 170 static int 171 sublivelist_block_refcnt_compare(const void *larg, const void *rarg) 172 { 173 const sublivelist_verify_block_refcnt_t *l = larg; 174 const sublivelist_verify_block_refcnt_t *r = rarg; 175 return (livelist_compare(&l->svbr_blk, &r->svbr_blk)); 176 } 177 178 static int 179 sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free, 180 dmu_tx_t *tx) 181 { 182 ASSERT0P(tx); 183 struct sublivelist_verify *sv = arg; 184 sublivelist_verify_block_refcnt_t current = { 185 .svbr_blk = *bp, 186 187 /* 188 * Start with 1 in case this is the first free entry. 189 * This field is not used for our B-Tree comparisons 190 * anyway. 191 */ 192 .svbr_refcnt = 1, 193 }; 194 195 zfs_btree_index_t where; 196 sublivelist_verify_block_refcnt_t *pair = 197 zfs_btree_find(&sv->sv_pair, ¤t, &where); 198 if (free) { 199 if (pair == NULL) { 200 /* first free entry for this block pointer */ 201 zfs_btree_add(&sv->sv_pair, ¤t); 202 } else { 203 pair->svbr_refcnt++; 204 } 205 } else { 206 if (pair == NULL) { 207 /* block that is currently marked as allocated */ 208 for (int i = 0; i < SPA_DVAS_PER_BP; i++) { 209 if (DVA_IS_EMPTY(&bp->blk_dva[i])) 210 break; 211 sublivelist_verify_block_t svb = { 212 .svb_dva = bp->blk_dva[i], 213 .svb_allocated_txg = 214 BP_GET_BIRTH(bp) 215 }; 216 217 if (zfs_btree_find(&sv->sv_leftover, &svb, 218 &where) == NULL) { 219 zfs_btree_add_idx(&sv->sv_leftover, 220 &svb, &where); 221 } 222 } 223 } else { 224 /* alloc matches a free entry */ 225 pair->svbr_refcnt--; 226 if (pair->svbr_refcnt == 0) { 227 /* all allocs and frees have been matched */ 228 zfs_btree_remove_idx(&sv->sv_pair, &where); 229 } 230 } 231 } 232 233 return (0); 234 } 235 236 static int 237 sublivelist_verify_func(void *args, dsl_deadlist_entry_t *dle) 238 { 239 int err; 240 struct sublivelist_verify *sv = args; 241 242 zfs_btree_create(&sv->sv_pair, sublivelist_block_refcnt_compare, NULL, 243 sizeof (sublivelist_verify_block_refcnt_t)); 244 245 err = bpobj_iterate_nofree(&dle->dle_bpobj, sublivelist_verify_blkptr, 246 sv, NULL); 247 248 sublivelist_verify_block_refcnt_t *e; 249 zfs_btree_index_t *cookie = NULL; 250 while ((e = zfs_btree_destroy_nodes(&sv->sv_pair, &cookie)) != NULL) { 251 char blkbuf[BP_SPRINTF_LEN]; 252 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), 253 &e->svbr_blk, B_TRUE); 254 (void) printf("\tERROR: %d unmatched FREE(s): %s\n", 255 e->svbr_refcnt, blkbuf); 256 corruption_found = B_TRUE; 257 } 258 zfs_btree_destroy(&sv->sv_pair); 259 260 return (err); 261 } 262 263 static int 264 livelist_block_compare(const void *larg, const void *rarg) 265 { 266 const sublivelist_verify_block_t *l = larg; 267 const sublivelist_verify_block_t *r = rarg; 268 269 if (DVA_GET_VDEV(&l->svb_dva) < DVA_GET_VDEV(&r->svb_dva)) 270 return (-1); 271 else if (DVA_GET_VDEV(&l->svb_dva) > DVA_GET_VDEV(&r->svb_dva)) 272 return (+1); 273 274 if (DVA_GET_OFFSET(&l->svb_dva) < DVA_GET_OFFSET(&r->svb_dva)) 275 return (-1); 276 else if (DVA_GET_OFFSET(&l->svb_dva) > DVA_GET_OFFSET(&r->svb_dva)) 277 return (+1); 278 279 if (DVA_GET_ASIZE(&l->svb_dva) < DVA_GET_ASIZE(&r->svb_dva)) 280 return (-1); 281 else if (DVA_GET_ASIZE(&l->svb_dva) > DVA_GET_ASIZE(&r->svb_dva)) 282 return (+1); 283 284 return (0); 285 } 286 287 /* 288 * Check for errors in a livelist while tracking all unfreed ALLOCs in the 289 * sublivelist_verify_t: sv->sv_leftover 290 */ 291 static void 292 livelist_verify(dsl_deadlist_t *dl, void *arg) 293 { 294 sublivelist_verify_t *sv = arg; 295 dsl_deadlist_iterate(dl, sublivelist_verify_func, sv); 296 } 297 298 /* 299 * Check for errors in the livelist entry and discard the intermediary 300 * data structures 301 */ 302 static int 303 sublivelist_verify_lightweight(void *args, dsl_deadlist_entry_t *dle) 304 { 305 (void) args; 306 sublivelist_verify_t sv; 307 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL, 308 sizeof (sublivelist_verify_block_t)); 309 int err = sublivelist_verify_func(&sv, dle); 310 zfs_btree_clear(&sv.sv_leftover); 311 zfs_btree_destroy(&sv.sv_leftover); 312 return (err); 313 } 314 315 typedef struct metaslab_verify { 316 /* 317 * Tree containing all the leftover ALLOCs from the livelists 318 * that are part of this metaslab. 319 */ 320 zfs_btree_t mv_livelist_allocs; 321 322 /* 323 * Metaslab information. 324 */ 325 uint64_t mv_vdid; 326 uint64_t mv_msid; 327 uint64_t mv_start; 328 uint64_t mv_end; 329 330 /* 331 * What's currently allocated for this metaslab. 332 */ 333 zfs_range_tree_t *mv_allocated; 334 } metaslab_verify_t; 335 336 typedef void ll_iter_t(dsl_deadlist_t *ll, void *arg); 337 338 typedef int (*zdb_log_sm_cb_t)(spa_t *spa, space_map_entry_t *sme, uint64_t txg, 339 void *arg); 340 341 typedef struct unflushed_iter_cb_arg { 342 spa_t *uic_spa; 343 uint64_t uic_txg; 344 void *uic_arg; 345 zdb_log_sm_cb_t uic_cb; 346 } unflushed_iter_cb_arg_t; 347 348 static int 349 iterate_through_spacemap_logs_cb(space_map_entry_t *sme, void *arg) 350 { 351 unflushed_iter_cb_arg_t *uic = arg; 352 return (uic->uic_cb(uic->uic_spa, sme, uic->uic_txg, uic->uic_arg)); 353 } 354 355 static void 356 iterate_through_spacemap_logs(spa_t *spa, zdb_log_sm_cb_t cb, void *arg) 357 { 358 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) 359 return; 360 361 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 362 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg); 363 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) { 364 space_map_t *sm = NULL; 365 VERIFY0(space_map_open(&sm, spa_meta_objset(spa), 366 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT)); 367 368 unflushed_iter_cb_arg_t uic = { 369 .uic_spa = spa, 370 .uic_txg = sls->sls_txg, 371 .uic_arg = arg, 372 .uic_cb = cb 373 }; 374 VERIFY0(space_map_iterate(sm, space_map_length(sm), 375 iterate_through_spacemap_logs_cb, &uic)); 376 space_map_close(sm); 377 } 378 spa_config_exit(spa, SCL_CONFIG, FTAG); 379 } 380 381 static void 382 verify_livelist_allocs(metaslab_verify_t *mv, uint64_t txg, 383 uint64_t offset, uint64_t size) 384 { 385 sublivelist_verify_block_t svb = {{{0}}}; 386 DVA_SET_VDEV(&svb.svb_dva, mv->mv_vdid); 387 DVA_SET_OFFSET(&svb.svb_dva, offset); 388 DVA_SET_ASIZE(&svb.svb_dva, 0); 389 zfs_btree_index_t where; 390 uint64_t end_offset = offset + size; 391 392 /* 393 * Look for an exact match for spacemap entry in the livelist entries. 394 * Then, look for other livelist entries that fall within the range 395 * of the spacemap entry as it may have been condensed 396 */ 397 sublivelist_verify_block_t *found = 398 zfs_btree_find(&mv->mv_livelist_allocs, &svb, &where); 399 if (found == NULL) { 400 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where); 401 } 402 for (; found != NULL && DVA_GET_VDEV(&found->svb_dva) == mv->mv_vdid && 403 DVA_GET_OFFSET(&found->svb_dva) < end_offset; 404 found = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) { 405 if (found->svb_allocated_txg <= txg) { 406 (void) printf("ERROR: Livelist ALLOC [%llx:%llx] " 407 "from TXG %llx FREED at TXG %llx\n", 408 (u_longlong_t)DVA_GET_OFFSET(&found->svb_dva), 409 (u_longlong_t)DVA_GET_ASIZE(&found->svb_dva), 410 (u_longlong_t)found->svb_allocated_txg, 411 (u_longlong_t)txg); 412 corruption_found = B_TRUE; 413 } 414 } 415 } 416 417 static int 418 metaslab_spacemap_validation_cb(space_map_entry_t *sme, void *arg) 419 { 420 metaslab_verify_t *mv = arg; 421 uint64_t offset = sme->sme_offset; 422 uint64_t size = sme->sme_run; 423 uint64_t txg = sme->sme_txg; 424 425 if (sme->sme_type == SM_ALLOC) { 426 if (zfs_range_tree_contains(mv->mv_allocated, 427 offset, size)) { 428 (void) printf("ERROR: DOUBLE ALLOC: " 429 "%llu [%llx:%llx] " 430 "%llu:%llu LOG_SM\n", 431 (u_longlong_t)txg, (u_longlong_t)offset, 432 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid, 433 (u_longlong_t)mv->mv_msid); 434 corruption_found = B_TRUE; 435 } else { 436 zfs_range_tree_add(mv->mv_allocated, 437 offset, size); 438 } 439 } else { 440 if (!zfs_range_tree_contains(mv->mv_allocated, 441 offset, size)) { 442 (void) printf("ERROR: DOUBLE FREE: " 443 "%llu [%llx:%llx] " 444 "%llu:%llu LOG_SM\n", 445 (u_longlong_t)txg, (u_longlong_t)offset, 446 (u_longlong_t)size, (u_longlong_t)mv->mv_vdid, 447 (u_longlong_t)mv->mv_msid); 448 corruption_found = B_TRUE; 449 } else { 450 zfs_range_tree_remove(mv->mv_allocated, 451 offset, size); 452 } 453 } 454 455 if (sme->sme_type != SM_ALLOC) { 456 /* 457 * If something is freed in the spacemap, verify that 458 * it is not listed as allocated in the livelist. 459 */ 460 verify_livelist_allocs(mv, txg, offset, size); 461 } 462 return (0); 463 } 464 465 static int 466 spacemap_check_sm_log_cb(spa_t *spa, space_map_entry_t *sme, 467 uint64_t txg, void *arg) 468 { 469 metaslab_verify_t *mv = arg; 470 uint64_t offset = sme->sme_offset; 471 uint64_t vdev_id = sme->sme_vdev; 472 473 vdev_t *vd = vdev_lookup_top(spa, vdev_id); 474 475 /* skip indirect vdevs */ 476 if (!vdev_is_concrete(vd)) 477 return (0); 478 479 if (vdev_id != mv->mv_vdid) 480 return (0); 481 482 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift]; 483 if (ms->ms_id != mv->mv_msid) 484 return (0); 485 486 if (txg < metaslab_unflushed_txg(ms)) 487 return (0); 488 489 490 ASSERT3U(txg, ==, sme->sme_txg); 491 return (metaslab_spacemap_validation_cb(sme, mv)); 492 } 493 494 static void 495 spacemap_check_sm_log(spa_t *spa, metaslab_verify_t *mv) 496 { 497 iterate_through_spacemap_logs(spa, spacemap_check_sm_log_cb, mv); 498 } 499 500 static void 501 spacemap_check_ms_sm(space_map_t *sm, metaslab_verify_t *mv) 502 { 503 if (sm == NULL) 504 return; 505 506 VERIFY0(space_map_iterate(sm, space_map_length(sm), 507 metaslab_spacemap_validation_cb, mv)); 508 } 509 510 static void iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg); 511 512 /* 513 * Transfer blocks from sv_leftover tree to the mv_livelist_allocs if 514 * they are part of that metaslab (mv_msid). 515 */ 516 static void 517 mv_populate_livelist_allocs(metaslab_verify_t *mv, sublivelist_verify_t *sv) 518 { 519 zfs_btree_index_t where; 520 sublivelist_verify_block_t *svb; 521 ASSERT3U(zfs_btree_numnodes(&mv->mv_livelist_allocs), ==, 0); 522 for (svb = zfs_btree_first(&sv->sv_leftover, &where); 523 svb != NULL; 524 svb = zfs_btree_next(&sv->sv_leftover, &where, &where)) { 525 if (DVA_GET_VDEV(&svb->svb_dva) != mv->mv_vdid) 526 continue; 527 528 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start && 529 (DVA_GET_OFFSET(&svb->svb_dva) + 530 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_start) { 531 (void) printf("ERROR: Found block that crosses " 532 "metaslab boundary: <%llu:%llx:%llx>\n", 533 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva), 534 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva), 535 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva)); 536 corruption_found = B_TRUE; 537 continue; 538 } 539 540 if (DVA_GET_OFFSET(&svb->svb_dva) < mv->mv_start) 541 continue; 542 543 if (DVA_GET_OFFSET(&svb->svb_dva) >= mv->mv_end) 544 continue; 545 546 if ((DVA_GET_OFFSET(&svb->svb_dva) + 547 DVA_GET_ASIZE(&svb->svb_dva)) > mv->mv_end) { 548 (void) printf("ERROR: Found block that crosses " 549 "metaslab boundary: <%llu:%llx:%llx>\n", 550 (u_longlong_t)DVA_GET_VDEV(&svb->svb_dva), 551 (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva), 552 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva)); 553 corruption_found = B_TRUE; 554 continue; 555 } 556 557 zfs_btree_add(&mv->mv_livelist_allocs, svb); 558 } 559 560 for (svb = zfs_btree_first(&mv->mv_livelist_allocs, &where); 561 svb != NULL; 562 svb = zfs_btree_next(&mv->mv_livelist_allocs, &where, &where)) { 563 zfs_btree_remove(&sv->sv_leftover, svb); 564 } 565 } 566 567 /* 568 * [Livelist Check] 569 * Iterate through all the sublivelists and: 570 * - report leftover frees (**) 571 * - record leftover ALLOCs together with their TXG [see Cross Check] 572 * 573 * (**) Note: Double ALLOCs are valid in datasets that have dedup 574 * enabled. Similarly double FREEs are allowed as well but 575 * only if they pair up with a corresponding ALLOC entry once 576 * we our done with our sublivelist iteration. 577 * 578 * [Spacemap Check] 579 * for each metaslab: 580 * - iterate over spacemap and then the metaslab's entries in the 581 * spacemap log, then report any double FREEs and ALLOCs (do not 582 * blow up). 583 * 584 * [Cross Check] 585 * After finishing the Livelist Check phase and while being in the 586 * Spacemap Check phase, we find all the recorded leftover ALLOCs 587 * of the livelist check that are part of the metaslab that we are 588 * currently looking at in the Spacemap Check. We report any entries 589 * that are marked as ALLOCs in the livelists but have been actually 590 * freed (and potentially allocated again) after their TXG stamp in 591 * the spacemaps. Also report any ALLOCs from the livelists that 592 * belong to indirect vdevs (e.g. their vdev completed removal). 593 * 594 * Note that this will miss Log Spacemap entries that cancelled each other 595 * out before being flushed to the metaslab, so we are not guaranteed 596 * to match all erroneous ALLOCs. 597 */ 598 static void 599 livelist_metaslab_validate(spa_t *spa) 600 { 601 (void) printf("Verifying deleted livelist entries\n"); 602 603 sublivelist_verify_t sv; 604 zfs_btree_create(&sv.sv_leftover, livelist_block_compare, NULL, 605 sizeof (sublivelist_verify_block_t)); 606 iterate_deleted_livelists(spa, livelist_verify, &sv); 607 608 (void) printf("Verifying metaslab entries\n"); 609 vdev_t *rvd = spa->spa_root_vdev; 610 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 611 vdev_t *vd = rvd->vdev_child[c]; 612 613 if (!vdev_is_concrete(vd)) 614 continue; 615 616 for (uint64_t mid = 0; mid < vd->vdev_ms_count; mid++) { 617 metaslab_t *m = vd->vdev_ms[mid]; 618 619 (void) fprintf(stderr, 620 "\rverifying concrete vdev %llu, " 621 "metaslab %llu of %llu ...", 622 (longlong_t)vd->vdev_id, 623 (longlong_t)mid, 624 (longlong_t)vd->vdev_ms_count); 625 626 uint64_t shift, start; 627 zfs_range_seg_type_t type = 628 metaslab_calculate_range_tree_type(vd, m, 629 &start, &shift); 630 metaslab_verify_t mv; 631 mv.mv_allocated = zfs_range_tree_create_flags( 632 NULL, type, NULL, start, shift, 633 0, "livelist_metaslab_validate:mv_allocated"); 634 mv.mv_vdid = vd->vdev_id; 635 mv.mv_msid = m->ms_id; 636 mv.mv_start = m->ms_start; 637 mv.mv_end = m->ms_start + m->ms_size; 638 zfs_btree_create(&mv.mv_livelist_allocs, 639 livelist_block_compare, NULL, 640 sizeof (sublivelist_verify_block_t)); 641 642 mv_populate_livelist_allocs(&mv, &sv); 643 644 spacemap_check_ms_sm(m->ms_sm, &mv); 645 spacemap_check_sm_log(spa, &mv); 646 647 zfs_range_tree_vacate(mv.mv_allocated, NULL, NULL); 648 zfs_range_tree_destroy(mv.mv_allocated); 649 zfs_btree_clear(&mv.mv_livelist_allocs); 650 zfs_btree_destroy(&mv.mv_livelist_allocs); 651 } 652 } 653 (void) fprintf(stderr, "\n"); 654 655 /* 656 * If there are any segments in the leftover tree after we walked 657 * through all the metaslabs in the concrete vdevs then this means 658 * that we have segments in the livelists that belong to indirect 659 * vdevs and are marked as allocated. 660 */ 661 if (zfs_btree_numnodes(&sv.sv_leftover) == 0) { 662 zfs_btree_destroy(&sv.sv_leftover); 663 return; 664 } 665 (void) printf("ERROR: Found livelist blocks marked as allocated " 666 "for indirect vdevs:\n"); 667 corruption_found = B_TRUE; 668 669 zfs_btree_index_t *where = NULL; 670 sublivelist_verify_block_t *svb; 671 while ((svb = zfs_btree_destroy_nodes(&sv.sv_leftover, &where)) != 672 NULL) { 673 int vdev_id = DVA_GET_VDEV(&svb->svb_dva); 674 ASSERT3U(vdev_id, <, rvd->vdev_children); 675 vdev_t *vd = rvd->vdev_child[vdev_id]; 676 ASSERT(!vdev_is_concrete(vd)); 677 (void) printf("<%d:%llx:%llx> TXG %llx\n", 678 vdev_id, (u_longlong_t)DVA_GET_OFFSET(&svb->svb_dva), 679 (u_longlong_t)DVA_GET_ASIZE(&svb->svb_dva), 680 (u_longlong_t)svb->svb_allocated_txg); 681 } 682 (void) printf("\n"); 683 zfs_btree_destroy(&sv.sv_leftover); 684 } 685 686 /* 687 * These libumem hooks provide a reasonable set of defaults for the allocator's 688 * debugging facilities. 689 */ 690 const char * 691 _umem_debug_init(void) 692 { 693 return ("default,verbose"); /* $UMEM_DEBUG setting */ 694 } 695 696 const char * 697 _umem_logging_init(void) 698 { 699 return ("fail,contents"); /* $UMEM_LOGGING setting */ 700 } 701 702 static void 703 usage(void) 704 { 705 (void) fprintf(stderr, 706 "Usage:\t%s [-AbcdDFGhikLMPsvXy] [-e [-V] [-p <path> ...]] " 707 "[-I <inflight I/Os>]\n" 708 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n" 709 "\t\t[-K <key>]\n" 710 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]]\n" 711 "\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] [-K <key>]\n" 712 "\t\t[<poolname>[/<dataset | objset id>] [<object | range> ...]\n" 713 "\t%s -B [-e [-V] [-p <path> ...]] [-I <inflight I/Os>]\n" 714 "\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n" 715 "\t\t[-K <key>] <poolname>/<objset id> [<backupflags>]\n" 716 "\t%s [-v] <bookmark>\n" 717 "\t%s -C [-A] [-U <cache>] [<poolname>]\n" 718 "\t%s -l [-Aqu] <device>\n" 719 "\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] " 720 "[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n" 721 "\t%s -O [-K <key>] <dataset> <path>\n" 722 "\t%s -r [-K <key>] <dataset> <path> <destination>\n" 723 "\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n" 724 "\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n" 725 "\t%s -E [-A] word0:word1:...:word15\n" 726 "\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] " 727 "<poolname>\n\n", 728 cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, 729 cmdname, cmdname, cmdname, cmdname, cmdname); 730 731 (void) fprintf(stderr, " Dataset name must include at least one " 732 "separator character '/' or '@'\n"); 733 (void) fprintf(stderr, " If dataset name is specified, only that " 734 "dataset is dumped\n"); 735 (void) fprintf(stderr, " If object numbers or object number " 736 "ranges are specified, only those\n" 737 " objects or ranges are dumped.\n\n"); 738 (void) fprintf(stderr, 739 " Object ranges take the form <start>:<end>[:<flags>]\n" 740 " start Starting object number\n" 741 " end Ending object number, or -1 for no upper bound\n" 742 " flags Optional flags to select object types:\n" 743 " A All objects (this is the default)\n" 744 " d ZFS directories\n" 745 " f ZFS files \n" 746 " m SPA space maps\n" 747 " z ZAPs\n" 748 " - Negate effect of next flag\n\n"); 749 (void) fprintf(stderr, " Options to control amount of output:\n"); 750 (void) fprintf(stderr, " -b --block-stats " 751 "block statistics\n"); 752 (void) fprintf(stderr, " -B --backup " 753 "backup stream\n"); 754 (void) fprintf(stderr, " -c --checksum " 755 "checksum all metadata (twice for all data) blocks\n"); 756 (void) fprintf(stderr, " -C --config " 757 "config (or cachefile if alone)\n"); 758 (void) fprintf(stderr, " -d --datasets " 759 "dataset(s)\n"); 760 (void) fprintf(stderr, " -D --dedup-stats " 761 "dedup statistics\n"); 762 (void) fprintf(stderr, " -E --embedded-block-pointer=INTEGER\n" 763 " decode and display block " 764 "from an embedded block pointer\n"); 765 (void) fprintf(stderr, " -h --history " 766 "pool history\n"); 767 (void) fprintf(stderr, " -i --intent-logs " 768 "intent logs\n"); 769 (void) fprintf(stderr, " -l --label " 770 "read label contents\n"); 771 (void) fprintf(stderr, " -k --checkpointed-state " 772 "examine the checkpointed state of the pool\n"); 773 (void) fprintf(stderr, " -L --disable-leak-tracking " 774 "disable leak tracking (do not load spacemaps)\n"); 775 (void) fprintf(stderr, " -m --metaslabs " 776 "metaslabs\n"); 777 (void) fprintf(stderr, " -M --metaslab-groups " 778 "metaslab groups\n"); 779 (void) fprintf(stderr, " -O --object-lookups " 780 "perform object lookups by path\n"); 781 (void) fprintf(stderr, " -r --copy-object " 782 "copy an object by path to file\n"); 783 (void) fprintf(stderr, " -R --read-block " 784 "read and display block from a device\n"); 785 (void) fprintf(stderr, " -s --io-stats " 786 "report stats on zdb's I/O\n"); 787 (void) fprintf(stderr, " -S --simulate-dedup " 788 "simulate dedup to measure effect\n"); 789 (void) fprintf(stderr, " -v --verbose " 790 "verbose (applies to all others)\n"); 791 (void) fprintf(stderr, " -y --livelist " 792 "perform livelist and metaslab validation on any livelists being " 793 "deleted\n\n"); 794 (void) fprintf(stderr, " Below options are intended for use " 795 "with other options:\n"); 796 (void) fprintf(stderr, " -A --ignore-assertions " 797 "ignore assertions (-A), enable panic recovery (-AA) or both " 798 "(-AAA)\n"); 799 (void) fprintf(stderr, " -e --exported " 800 "pool is exported/destroyed/has altroot/not in a cachefile\n"); 801 (void) fprintf(stderr, " -F --automatic-rewind " 802 "attempt automatic rewind within safe range of transaction " 803 "groups\n"); 804 (void) fprintf(stderr, " -G --dump-debug-msg " 805 "dump zfs_dbgmsg buffer before exiting\n"); 806 (void) fprintf(stderr, " -I --inflight=INTEGER " 807 "specify the maximum number of checksumming I/Os " 808 "[default is 200]\n"); 809 (void) fprintf(stderr, " -K --key=KEY " 810 "decryption key for encrypted dataset\n"); 811 (void) fprintf(stderr, " -o --option=\"NAME=VALUE\" " 812 "set the named tunable to the given value\n"); 813 (void) fprintf(stderr, " -p --path==PATH " 814 "use one or more with -e to specify path to vdev dir\n"); 815 (void) fprintf(stderr, " -P --parseable " 816 "print numbers in parseable form\n"); 817 (void) fprintf(stderr, " -q --skip-label " 818 "don't print label contents\n"); 819 (void) fprintf(stderr, " -t --txg=INTEGER " 820 "highest txg to use when searching for uberblocks\n"); 821 (void) fprintf(stderr, " -T --brt-stats " 822 "BRT statistics\n"); 823 (void) fprintf(stderr, " -u --uberblock " 824 "uberblock\n"); 825 (void) fprintf(stderr, " -U --cachefile=PATH " 826 "use alternate cachefile\n"); 827 (void) fprintf(stderr, " -V --verbatim " 828 "do verbatim import\n"); 829 (void) fprintf(stderr, " -x --dump-blocks=PATH " 830 "dump all read blocks into specified directory\n"); 831 (void) fprintf(stderr, " -X --extreme-rewind " 832 "attempt extreme rewind (does not work with dataset)\n"); 833 (void) fprintf(stderr, " -Y --all-reconstruction " 834 "attempt all reconstruction combinations for split blocks\n"); 835 (void) fprintf(stderr, " -Z --zstd-headers " 836 "show ZSTD headers \n"); 837 (void) fprintf(stderr, "Specify an option more than once (e.g. -bb) " 838 "to make only that option verbose\n"); 839 (void) fprintf(stderr, "Default is to dump everything non-verbosely\n"); 840 zdb_exit(2); 841 } 842 843 static void 844 dump_debug_buffer(void) 845 { 846 ssize_t ret __attribute__((unused)); 847 848 if (!dump_opt['G']) 849 return; 850 /* 851 * We use write() instead of printf() so that this function 852 * is safe to call from a signal handler. 853 */ 854 ret = write(STDERR_FILENO, "\n", 1); 855 zfs_dbgmsg_print(STDERR_FILENO, "zdb"); 856 } 857 858 static void sig_handler(int signo) 859 { 860 struct sigaction action; 861 862 libspl_backtrace(STDERR_FILENO); 863 dump_debug_buffer(); 864 865 /* 866 * Restore default action and re-raise signal so SIGSEGV and 867 * SIGABRT can trigger a core dump. 868 */ 869 action.sa_handler = SIG_DFL; 870 sigemptyset(&action.sa_mask); 871 action.sa_flags = 0; 872 (void) sigaction(signo, &action, NULL); 873 raise(signo); 874 } 875 876 /* 877 * Called for usage errors that are discovered after a call to spa_open(), 878 * dmu_bonus_hold(), or pool_match(). abort() is called for other errors. 879 */ 880 881 static void 882 fatal(const char *fmt, ...) 883 { 884 va_list ap; 885 886 va_start(ap, fmt); 887 (void) fprintf(stderr, "%s: ", cmdname); 888 (void) vfprintf(stderr, fmt, ap); 889 va_end(ap); 890 (void) fprintf(stderr, "\n"); 891 892 dump_debug_buffer(); 893 894 zdb_exit(1); 895 } 896 897 static void 898 dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size) 899 { 900 (void) size; 901 nvlist_t *nv; 902 size_t nvsize = *(uint64_t *)data; 903 char *packed = umem_alloc(nvsize, UMEM_NOFAIL); 904 905 VERIFY0(dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH)); 906 907 VERIFY0(nvlist_unpack(packed, nvsize, &nv, 0)); 908 909 umem_free(packed, nvsize); 910 911 dump_nvlist(nv, 8); 912 913 nvlist_free(nv); 914 } 915 916 static void 917 dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size) 918 { 919 (void) os, (void) object, (void) size; 920 spa_history_phys_t *shp = data; 921 922 if (shp == NULL) 923 return; 924 925 (void) printf("\t\tpool_create_len = %llu\n", 926 (u_longlong_t)shp->sh_pool_create_len); 927 (void) printf("\t\tphys_max_off = %llu\n", 928 (u_longlong_t)shp->sh_phys_max_off); 929 (void) printf("\t\tbof = %llu\n", 930 (u_longlong_t)shp->sh_bof); 931 (void) printf("\t\teof = %llu\n", 932 (u_longlong_t)shp->sh_eof); 933 (void) printf("\t\trecords_lost = %llu\n", 934 (u_longlong_t)shp->sh_records_lost); 935 } 936 937 static void 938 zdb_nicenum(uint64_t num, char *buf, size_t buflen) 939 { 940 if (dump_opt['P']) 941 (void) snprintf(buf, buflen, "%llu", (longlong_t)num); 942 else 943 nicenum(num, buf, buflen); 944 } 945 946 static void 947 zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen) 948 { 949 if (dump_opt['P']) 950 (void) snprintf(buf, buflen, "%llu", (longlong_t)bytes); 951 else 952 zfs_nicebytes(bytes, buf, buflen); 953 } 954 955 static const char histo_stars[] = "****************************************"; 956 static const uint64_t histo_width = sizeof (histo_stars) - 1; 957 958 static void 959 dump_histogram(const uint64_t *histo, int size, int offset) 960 { 961 int i; 962 int minidx = size - 1; 963 int maxidx = 0; 964 uint64_t max = 0; 965 966 for (i = 0; i < size; i++) { 967 if (histo[i] == 0) 968 continue; 969 if (histo[i] > max) 970 max = histo[i]; 971 if (i > maxidx) 972 maxidx = i; 973 if (i < minidx) 974 minidx = i; 975 } 976 977 if (max < histo_width) 978 max = histo_width; 979 980 for (i = minidx; i <= maxidx; i++) { 981 (void) printf("\t\t\t%3u: %6llu %s\n", 982 i + offset, (u_longlong_t)histo[i], 983 &histo_stars[(max - histo[i]) * histo_width / max]); 984 } 985 } 986 987 static void 988 dump_zap_stats(objset_t *os, uint64_t object) 989 { 990 int error; 991 zap_stats_t zs; 992 993 error = zap_get_stats(os, object, &zs); 994 if (error) 995 return; 996 997 if (zs.zs_ptrtbl_len == 0) { 998 ASSERT(zs.zs_num_blocks == 1); 999 (void) printf("\tmicrozap: %llu bytes, %llu entries\n", 1000 (u_longlong_t)zs.zs_blocksize, 1001 (u_longlong_t)zs.zs_num_entries); 1002 return; 1003 } 1004 1005 (void) printf("\tFat ZAP stats:\n"); 1006 1007 (void) printf("\t\tPointer table:\n"); 1008 (void) printf("\t\t\t%llu elements\n", 1009 (u_longlong_t)zs.zs_ptrtbl_len); 1010 (void) printf("\t\t\tzt_blk: %llu\n", 1011 (u_longlong_t)zs.zs_ptrtbl_zt_blk); 1012 (void) printf("\t\t\tzt_numblks: %llu\n", 1013 (u_longlong_t)zs.zs_ptrtbl_zt_numblks); 1014 (void) printf("\t\t\tzt_shift: %llu\n", 1015 (u_longlong_t)zs.zs_ptrtbl_zt_shift); 1016 (void) printf("\t\t\tzt_blks_copied: %llu\n", 1017 (u_longlong_t)zs.zs_ptrtbl_blks_copied); 1018 (void) printf("\t\t\tzt_nextblk: %llu\n", 1019 (u_longlong_t)zs.zs_ptrtbl_nextblk); 1020 1021 (void) printf("\t\tZAP entries: %llu\n", 1022 (u_longlong_t)zs.zs_num_entries); 1023 (void) printf("\t\tLeaf blocks: %llu\n", 1024 (u_longlong_t)zs.zs_num_leafs); 1025 (void) printf("\t\tTotal blocks: %llu\n", 1026 (u_longlong_t)zs.zs_num_blocks); 1027 (void) printf("\t\tzap_block_type: 0x%llx\n", 1028 (u_longlong_t)zs.zs_block_type); 1029 (void) printf("\t\tzap_magic: 0x%llx\n", 1030 (u_longlong_t)zs.zs_magic); 1031 (void) printf("\t\tzap_salt: 0x%llx\n", 1032 (u_longlong_t)zs.zs_salt); 1033 1034 (void) printf("\t\tLeafs with 2^n pointers:\n"); 1035 dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0); 1036 1037 (void) printf("\t\tBlocks with n*5 entries:\n"); 1038 dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0); 1039 1040 (void) printf("\t\tBlocks n/10 full:\n"); 1041 dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0); 1042 1043 (void) printf("\t\tEntries with n chunks:\n"); 1044 dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0); 1045 1046 (void) printf("\t\tBuckets with n entries:\n"); 1047 dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0); 1048 } 1049 1050 static void 1051 dump_none(objset_t *os, uint64_t object, void *data, size_t size) 1052 { 1053 (void) os, (void) object, (void) data, (void) size; 1054 } 1055 1056 static void 1057 dump_unknown(objset_t *os, uint64_t object, void *data, size_t size) 1058 { 1059 (void) os, (void) object, (void) data, (void) size; 1060 (void) printf("\tUNKNOWN OBJECT TYPE\n"); 1061 } 1062 1063 static void 1064 dump_uint8(objset_t *os, uint64_t object, void *data, size_t size) 1065 { 1066 (void) os, (void) object, (void) data, (void) size; 1067 } 1068 1069 static void 1070 dump_uint64(objset_t *os, uint64_t object, void *data, size_t size) 1071 { 1072 uint64_t *arr; 1073 uint64_t oursize; 1074 if (dump_opt['d'] < 6) 1075 return; 1076 1077 if (data == NULL) { 1078 dmu_object_info_t doi; 1079 1080 VERIFY0(dmu_object_info(os, object, &doi)); 1081 size = doi.doi_max_offset; 1082 /* 1083 * We cap the size at 1 mebibyte here to prevent 1084 * allocation failures and nigh-infinite printing if the 1085 * object is extremely large. 1086 */ 1087 oursize = MIN(size, 1 << 20); 1088 arr = kmem_alloc(oursize, KM_SLEEP); 1089 1090 int err = dmu_read(os, object, 0, oursize, arr, 0); 1091 if (err != 0) { 1092 (void) printf("got error %u from dmu_read\n", err); 1093 kmem_free(arr, oursize); 1094 return; 1095 } 1096 } else { 1097 /* 1098 * Even though the allocation is already done in this code path, 1099 * we still cap the size to prevent excessive printing. 1100 */ 1101 oursize = MIN(size, 1 << 20); 1102 arr = data; 1103 } 1104 1105 if (size == 0) { 1106 if (data == NULL) 1107 kmem_free(arr, oursize); 1108 (void) printf("\t\t[]\n"); 1109 return; 1110 } 1111 1112 (void) printf("\t\t[%0llx", (u_longlong_t)arr[0]); 1113 for (size_t i = 1; i * sizeof (uint64_t) < oursize; i++) { 1114 if (i % 4 != 0) 1115 (void) printf(", %0llx", (u_longlong_t)arr[i]); 1116 else 1117 (void) printf(",\n\t\t%0llx", (u_longlong_t)arr[i]); 1118 } 1119 if (oursize != size) 1120 (void) printf(", ... "); 1121 (void) printf("]\n"); 1122 1123 if (data == NULL) 1124 kmem_free(arr, oursize); 1125 } 1126 1127 static void 1128 dump_zap(objset_t *os, uint64_t object, void *data, size_t size) 1129 { 1130 (void) data, (void) size; 1131 zap_cursor_t zc; 1132 zap_attribute_t *attrp = zap_attribute_long_alloc(); 1133 void *prop; 1134 unsigned i; 1135 1136 dump_zap_stats(os, object); 1137 (void) printf("\n"); 1138 1139 for (zap_cursor_init(&zc, os, object); 1140 zap_cursor_retrieve(&zc, attrp) == 0; 1141 zap_cursor_advance(&zc)) { 1142 boolean_t key64 = 1143 !!(zap_getflags(zc.zc_zap) & ZAP_FLAG_UINT64_KEY); 1144 1145 if (key64) 1146 (void) printf("\t\t0x%010" PRIu64 "x = ", 1147 *(uint64_t *)attrp->za_name); 1148 else 1149 (void) printf("\t\t%s = ", attrp->za_name); 1150 1151 if (attrp->za_num_integers == 0) { 1152 (void) printf("\n"); 1153 continue; 1154 } 1155 prop = umem_zalloc(attrp->za_num_integers * 1156 attrp->za_integer_length, UMEM_NOFAIL); 1157 1158 if (key64) 1159 (void) zap_lookup_uint64(os, object, 1160 (const uint64_t *)attrp->za_name, 1, 1161 attrp->za_integer_length, attrp->za_num_integers, 1162 prop); 1163 else 1164 (void) zap_lookup(os, object, attrp->za_name, 1165 attrp->za_integer_length, attrp->za_num_integers, 1166 prop); 1167 1168 if (attrp->za_integer_length == 1 && !key64) { 1169 if (strcmp(attrp->za_name, 1170 DSL_CRYPTO_KEY_MASTER_KEY) == 0 || 1171 strcmp(attrp->za_name, 1172 DSL_CRYPTO_KEY_HMAC_KEY) == 0 || 1173 strcmp(attrp->za_name, DSL_CRYPTO_KEY_IV) == 0 || 1174 strcmp(attrp->za_name, DSL_CRYPTO_KEY_MAC) == 0 || 1175 strcmp(attrp->za_name, 1176 DMU_POOL_CHECKSUM_SALT) == 0) { 1177 uint8_t *u8 = prop; 1178 1179 for (i = 0; i < attrp->za_num_integers; i++) { 1180 (void) printf("%02x", u8[i]); 1181 } 1182 } else { 1183 (void) printf("%s", (char *)prop); 1184 } 1185 } else { 1186 for (i = 0; i < attrp->za_num_integers; i++) { 1187 switch (attrp->za_integer_length) { 1188 case 1: 1189 (void) printf("%u ", 1190 ((uint8_t *)prop)[i]); 1191 break; 1192 case 2: 1193 (void) printf("%u ", 1194 ((uint16_t *)prop)[i]); 1195 break; 1196 case 4: 1197 (void) printf("%u ", 1198 ((uint32_t *)prop)[i]); 1199 break; 1200 case 8: 1201 (void) printf("%lld ", 1202 (u_longlong_t)((int64_t *)prop)[i]); 1203 break; 1204 } 1205 } 1206 } 1207 (void) printf("\n"); 1208 umem_free(prop, 1209 attrp->za_num_integers * attrp->za_integer_length); 1210 } 1211 zap_cursor_fini(&zc); 1212 zap_attribute_free(attrp); 1213 } 1214 1215 static void 1216 dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size) 1217 { 1218 bpobj_phys_t *bpop = data; 1219 uint64_t i; 1220 char bytes[32], comp[32], uncomp[32]; 1221 1222 /* make sure the output won't get truncated */ 1223 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated"); 1224 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated"); 1225 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated"); 1226 1227 if (bpop == NULL) 1228 return; 1229 1230 zdb_nicenum(bpop->bpo_bytes, bytes, sizeof (bytes)); 1231 zdb_nicenum(bpop->bpo_comp, comp, sizeof (comp)); 1232 zdb_nicenum(bpop->bpo_uncomp, uncomp, sizeof (uncomp)); 1233 1234 (void) printf("\t\tnum_blkptrs = %llu\n", 1235 (u_longlong_t)bpop->bpo_num_blkptrs); 1236 (void) printf("\t\tbytes = %s\n", bytes); 1237 if (size >= BPOBJ_SIZE_V1) { 1238 (void) printf("\t\tcomp = %s\n", comp); 1239 (void) printf("\t\tuncomp = %s\n", uncomp); 1240 } 1241 if (size >= BPOBJ_SIZE_V2) { 1242 (void) printf("\t\tsubobjs = %llu\n", 1243 (u_longlong_t)bpop->bpo_subobjs); 1244 (void) printf("\t\tnum_subobjs = %llu\n", 1245 (u_longlong_t)bpop->bpo_num_subobjs); 1246 } 1247 if (size >= sizeof (*bpop)) { 1248 (void) printf("\t\tnum_freed = %llu\n", 1249 (u_longlong_t)bpop->bpo_num_freed); 1250 } 1251 1252 if (dump_opt['d'] < 5) 1253 return; 1254 1255 for (i = 0; i < bpop->bpo_num_blkptrs; i++) { 1256 char blkbuf[BP_SPRINTF_LEN]; 1257 blkptr_t bp; 1258 1259 int err = dmu_read(os, object, 1260 i * sizeof (bp), sizeof (bp), &bp, 0); 1261 if (err != 0) { 1262 (void) printf("got error %u from dmu_read\n", err); 1263 break; 1264 } 1265 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp, 1266 BP_GET_FREE(&bp)); 1267 (void) printf("\t%s\n", blkbuf); 1268 } 1269 } 1270 1271 static void 1272 dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size) 1273 { 1274 (void) data, (void) size; 1275 dmu_object_info_t doi; 1276 int64_t i; 1277 1278 VERIFY0(dmu_object_info(os, object, &doi)); 1279 uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP); 1280 1281 int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0); 1282 if (err != 0) { 1283 (void) printf("got error %u from dmu_read\n", err); 1284 kmem_free(subobjs, doi.doi_max_offset); 1285 return; 1286 } 1287 1288 int64_t last_nonzero = -1; 1289 for (i = 0; i < doi.doi_max_offset / 8; i++) { 1290 if (subobjs[i] != 0) 1291 last_nonzero = i; 1292 } 1293 1294 for (i = 0; i <= last_nonzero; i++) { 1295 (void) printf("\t%llu\n", (u_longlong_t)subobjs[i]); 1296 } 1297 kmem_free(subobjs, doi.doi_max_offset); 1298 } 1299 1300 static void 1301 dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size) 1302 { 1303 (void) data, (void) size; 1304 dump_zap_stats(os, object); 1305 /* contents are printed elsewhere, properly decoded */ 1306 } 1307 1308 static void 1309 dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size) 1310 { 1311 (void) data, (void) size; 1312 zap_cursor_t zc; 1313 zap_attribute_t *attrp = zap_attribute_alloc(); 1314 1315 dump_zap_stats(os, object); 1316 (void) printf("\n"); 1317 1318 for (zap_cursor_init(&zc, os, object); 1319 zap_cursor_retrieve(&zc, attrp) == 0; 1320 zap_cursor_advance(&zc)) { 1321 (void) printf("\t\t%s = ", attrp->za_name); 1322 if (attrp->za_num_integers == 0) { 1323 (void) printf("\n"); 1324 continue; 1325 } 1326 (void) printf(" %llx : [%d:%d:%d]\n", 1327 (u_longlong_t)attrp->za_first_integer, 1328 (int)ATTR_LENGTH(attrp->za_first_integer), 1329 (int)ATTR_BSWAP(attrp->za_first_integer), 1330 (int)ATTR_NUM(attrp->za_first_integer)); 1331 } 1332 zap_cursor_fini(&zc); 1333 zap_attribute_free(attrp); 1334 } 1335 1336 static void 1337 dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size) 1338 { 1339 (void) data, (void) size; 1340 zap_cursor_t zc; 1341 zap_attribute_t *attrp = zap_attribute_alloc(); 1342 uint16_t *layout_attrs; 1343 unsigned i; 1344 1345 dump_zap_stats(os, object); 1346 (void) printf("\n"); 1347 1348 for (zap_cursor_init(&zc, os, object); 1349 zap_cursor_retrieve(&zc, attrp) == 0; 1350 zap_cursor_advance(&zc)) { 1351 (void) printf("\t\t%s = [", attrp->za_name); 1352 if (attrp->za_num_integers == 0) { 1353 (void) printf("\n"); 1354 continue; 1355 } 1356 1357 VERIFY(attrp->za_integer_length == 2); 1358 layout_attrs = umem_zalloc(attrp->za_num_integers * 1359 attrp->za_integer_length, UMEM_NOFAIL); 1360 1361 VERIFY(zap_lookup(os, object, attrp->za_name, 1362 attrp->za_integer_length, 1363 attrp->za_num_integers, layout_attrs) == 0); 1364 1365 for (i = 0; i != attrp->za_num_integers; i++) 1366 (void) printf(" %d ", (int)layout_attrs[i]); 1367 (void) printf("]\n"); 1368 umem_free(layout_attrs, 1369 attrp->za_num_integers * attrp->za_integer_length); 1370 } 1371 zap_cursor_fini(&zc); 1372 zap_attribute_free(attrp); 1373 } 1374 1375 static void 1376 dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size) 1377 { 1378 (void) data, (void) size; 1379 zap_cursor_t zc; 1380 zap_attribute_t *attrp = zap_attribute_long_alloc(); 1381 const char *typenames[] = { 1382 /* 0 */ "not specified", 1383 /* 1 */ "FIFO", 1384 /* 2 */ "Character Device", 1385 /* 3 */ "3 (invalid)", 1386 /* 4 */ "Directory", 1387 /* 5 */ "5 (invalid)", 1388 /* 6 */ "Block Device", 1389 /* 7 */ "7 (invalid)", 1390 /* 8 */ "Regular File", 1391 /* 9 */ "9 (invalid)", 1392 /* 10 */ "Symbolic Link", 1393 /* 11 */ "11 (invalid)", 1394 /* 12 */ "Socket", 1395 /* 13 */ "Door", 1396 /* 14 */ "Event Port", 1397 /* 15 */ "15 (invalid)", 1398 }; 1399 1400 dump_zap_stats(os, object); 1401 (void) printf("\n"); 1402 1403 for (zap_cursor_init(&zc, os, object); 1404 zap_cursor_retrieve(&zc, attrp) == 0; 1405 zap_cursor_advance(&zc)) { 1406 (void) printf("\t\t%s = %lld (type: %s)\n", 1407 attrp->za_name, ZFS_DIRENT_OBJ(attrp->za_first_integer), 1408 typenames[ZFS_DIRENT_TYPE(attrp->za_first_integer)]); 1409 } 1410 zap_cursor_fini(&zc); 1411 zap_attribute_free(attrp); 1412 } 1413 1414 static int 1415 get_dtl_refcount(vdev_t *vd) 1416 { 1417 int refcount = 0; 1418 1419 if (vd->vdev_ops->vdev_op_leaf) { 1420 space_map_t *sm = vd->vdev_dtl_sm; 1421 1422 if (sm != NULL && 1423 sm->sm_dbuf->db_size == sizeof (space_map_phys_t)) 1424 return (1); 1425 return (0); 1426 } 1427 1428 for (unsigned c = 0; c < vd->vdev_children; c++) 1429 refcount += get_dtl_refcount(vd->vdev_child[c]); 1430 return (refcount); 1431 } 1432 1433 static int 1434 get_metaslab_refcount(vdev_t *vd) 1435 { 1436 int refcount = 0; 1437 1438 if (vd->vdev_top == vd) { 1439 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) { 1440 space_map_t *sm = vd->vdev_ms[m]->ms_sm; 1441 1442 if (sm != NULL && 1443 sm->sm_dbuf->db_size == sizeof (space_map_phys_t)) 1444 refcount++; 1445 } 1446 } 1447 for (unsigned c = 0; c < vd->vdev_children; c++) 1448 refcount += get_metaslab_refcount(vd->vdev_child[c]); 1449 1450 return (refcount); 1451 } 1452 1453 static int 1454 get_obsolete_refcount(vdev_t *vd) 1455 { 1456 uint64_t obsolete_sm_object; 1457 int refcount = 0; 1458 1459 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object)); 1460 if (vd->vdev_top == vd && obsolete_sm_object != 0) { 1461 dmu_object_info_t doi; 1462 VERIFY0(dmu_object_info(vd->vdev_spa->spa_meta_objset, 1463 obsolete_sm_object, &doi)); 1464 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) { 1465 refcount++; 1466 } 1467 } else { 1468 ASSERT0P(vd->vdev_obsolete_sm); 1469 ASSERT0(obsolete_sm_object); 1470 } 1471 for (unsigned c = 0; c < vd->vdev_children; c++) { 1472 refcount += get_obsolete_refcount(vd->vdev_child[c]); 1473 } 1474 1475 return (refcount); 1476 } 1477 1478 static int 1479 get_prev_obsolete_spacemap_refcount(spa_t *spa) 1480 { 1481 uint64_t prev_obj = 1482 spa->spa_condensing_indirect_phys.scip_prev_obsolete_sm_object; 1483 if (prev_obj != 0) { 1484 dmu_object_info_t doi; 1485 VERIFY0(dmu_object_info(spa->spa_meta_objset, prev_obj, &doi)); 1486 if (doi.doi_bonus_size == sizeof (space_map_phys_t)) { 1487 return (1); 1488 } 1489 } 1490 return (0); 1491 } 1492 1493 static int 1494 get_checkpoint_refcount(vdev_t *vd) 1495 { 1496 int refcount = 0; 1497 1498 if (vd->vdev_top == vd && vd->vdev_top_zap != 0 && 1499 zap_contains(spa_meta_objset(vd->vdev_spa), 1500 vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) == 0) 1501 refcount++; 1502 1503 for (uint64_t c = 0; c < vd->vdev_children; c++) 1504 refcount += get_checkpoint_refcount(vd->vdev_child[c]); 1505 1506 return (refcount); 1507 } 1508 1509 static int 1510 get_log_spacemap_refcount(spa_t *spa) 1511 { 1512 return (avl_numnodes(&spa->spa_sm_logs_by_txg)); 1513 } 1514 1515 static int 1516 verify_spacemap_refcounts(spa_t *spa) 1517 { 1518 uint64_t expected_refcount = 0; 1519 uint64_t actual_refcount; 1520 1521 (void) feature_get_refcount(spa, 1522 &spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM], 1523 &expected_refcount); 1524 actual_refcount = get_dtl_refcount(spa->spa_root_vdev); 1525 actual_refcount += get_metaslab_refcount(spa->spa_root_vdev); 1526 actual_refcount += get_obsolete_refcount(spa->spa_root_vdev); 1527 actual_refcount += get_prev_obsolete_spacemap_refcount(spa); 1528 actual_refcount += get_checkpoint_refcount(spa->spa_root_vdev); 1529 actual_refcount += get_log_spacemap_refcount(spa); 1530 1531 if (expected_refcount != actual_refcount) { 1532 (void) printf("space map refcount mismatch: expected %lld != " 1533 "actual %lld\n", 1534 (longlong_t)expected_refcount, 1535 (longlong_t)actual_refcount); 1536 return (2); 1537 } 1538 return (0); 1539 } 1540 1541 static void 1542 dump_spacemap(objset_t *os, space_map_t *sm) 1543 { 1544 const char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID", 1545 "INVALID", "INVALID", "INVALID", "INVALID" }; 1546 1547 if (sm == NULL) 1548 return; 1549 1550 (void) printf("space map object %llu:\n", 1551 (longlong_t)sm->sm_object); 1552 (void) printf(" smp_length = 0x%llx\n", 1553 (longlong_t)sm->sm_phys->smp_length); 1554 (void) printf(" smp_alloc = 0x%llx\n", 1555 (longlong_t)sm->sm_phys->smp_alloc); 1556 1557 if (dump_opt['d'] < 6 && dump_opt['m'] < 4) 1558 return; 1559 1560 /* 1561 * Print out the freelist entries in both encoded and decoded form. 1562 */ 1563 uint8_t mapshift = sm->sm_shift; 1564 int64_t alloc = 0; 1565 uint64_t word, entry_id = 0; 1566 for (uint64_t offset = 0; offset < space_map_length(sm); 1567 offset += sizeof (word)) { 1568 1569 VERIFY0(dmu_read(os, space_map_object(sm), offset, 1570 sizeof (word), &word, DMU_READ_PREFETCH)); 1571 1572 if (sm_entry_is_debug(word)) { 1573 uint64_t de_txg = SM_DEBUG_TXG_DECODE(word); 1574 uint64_t de_sync_pass = SM_DEBUG_SYNCPASS_DECODE(word); 1575 if (de_txg == 0) { 1576 (void) printf( 1577 "\t [%6llu] PADDING\n", 1578 (u_longlong_t)entry_id); 1579 } else { 1580 (void) printf( 1581 "\t [%6llu] %s: txg %llu pass %llu\n", 1582 (u_longlong_t)entry_id, 1583 ddata[SM_DEBUG_ACTION_DECODE(word)], 1584 (u_longlong_t)de_txg, 1585 (u_longlong_t)de_sync_pass); 1586 } 1587 entry_id++; 1588 continue; 1589 } 1590 1591 char entry_type; 1592 uint64_t entry_off, entry_run, entry_vdev; 1593 1594 if (sm_entry_is_single_word(word)) { 1595 entry_type = (SM_TYPE_DECODE(word) == SM_ALLOC) ? 1596 'A' : 'F'; 1597 entry_off = (SM_OFFSET_DECODE(word) << mapshift) + 1598 sm->sm_start; 1599 entry_run = SM_RUN_DECODE(word) << mapshift; 1600 1601 (void) printf("\t [%6llu] %c " 1602 "range: %012llx-%012llx size: %08llx\n", 1603 (u_longlong_t)entry_id, entry_type, 1604 (u_longlong_t)entry_off, 1605 (u_longlong_t)(entry_off + entry_run - 1), 1606 (u_longlong_t)entry_run); 1607 } else { 1608 /* it is a two-word entry so we read another word */ 1609 ASSERT(sm_entry_is_double_word(word)); 1610 1611 uint64_t extra_word; 1612 offset += sizeof (extra_word); 1613 ASSERT3U(offset, <, space_map_length(sm)); 1614 VERIFY0(dmu_read(os, space_map_object(sm), offset, 1615 sizeof (extra_word), &extra_word, 1616 DMU_READ_PREFETCH)); 1617 1618 entry_run = SM2_RUN_DECODE(word) << mapshift; 1619 entry_vdev = SM2_VDEV_DECODE(word); 1620 entry_type = (SM2_TYPE_DECODE(extra_word) == SM_ALLOC) ? 1621 'A' : 'F'; 1622 entry_off = (SM2_OFFSET_DECODE(extra_word) << 1623 mapshift) + sm->sm_start; 1624 1625 if (zopt_metaslab_args == 0 || 1626 zopt_metaslab[0] == entry_vdev) { 1627 (void) printf("\t [%6llu] %c " 1628 "range: %012llx-%012llx size: %08llx " 1629 "vdev: %llu\n", 1630 (u_longlong_t)entry_id, entry_type, 1631 (u_longlong_t)entry_off, 1632 (u_longlong_t)(entry_off + entry_run - 1), 1633 (u_longlong_t)entry_run, 1634 (u_longlong_t)entry_vdev); 1635 } 1636 } 1637 1638 if (entry_type == 'A') 1639 alloc += entry_run; 1640 else 1641 alloc -= entry_run; 1642 entry_id++; 1643 } 1644 if (alloc != space_map_allocated(sm)) { 1645 (void) printf("space_map_object alloc (%lld) INCONSISTENT " 1646 "with space map summary (%lld)\n", 1647 (longlong_t)space_map_allocated(sm), (longlong_t)alloc); 1648 } 1649 } 1650 1651 static void 1652 dump_metaslab_stats(metaslab_t *msp) 1653 { 1654 char maxbuf[32]; 1655 zfs_range_tree_t *rt = msp->ms_allocatable; 1656 zfs_btree_t *t = &msp->ms_allocatable_by_size; 1657 int free_pct = zfs_range_tree_space(rt) * 100 / msp->ms_size; 1658 1659 /* max sure nicenum has enough space */ 1660 _Static_assert(sizeof (maxbuf) >= NN_NUMBUF_SZ, "maxbuf truncated"); 1661 1662 zdb_nicenum(metaslab_largest_allocatable(msp), maxbuf, sizeof (maxbuf)); 1663 1664 (void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n", 1665 "segments", zfs_btree_numnodes(t), "maxsize", maxbuf, 1666 "freepct", free_pct); 1667 (void) printf("\tIn-memory histogram:\n"); 1668 dump_histogram(rt->rt_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0); 1669 } 1670 1671 static void 1672 dump_allocated(void *arg, uint64_t start, uint64_t size) 1673 { 1674 uint64_t *off = arg; 1675 if (*off != start) 1676 (void) printf("ALLOC: %"PRIu64" %"PRIu64"\n", *off, 1677 start - *off); 1678 *off = start + size; 1679 } 1680 1681 static void 1682 dump_metaslab(metaslab_t *msp) 1683 { 1684 vdev_t *vd = msp->ms_group->mg_vd; 1685 spa_t *spa = vd->vdev_spa; 1686 space_map_t *sm = msp->ms_sm; 1687 char freebuf[32]; 1688 1689 zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf, 1690 sizeof (freebuf)); 1691 1692 (void) printf( 1693 "\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n", 1694 (u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start, 1695 (u_longlong_t)space_map_object(sm), freebuf); 1696 1697 if (dump_opt[ALLOCATED_OPT] || 1698 (dump_opt['m'] > 2 && !dump_opt['L'])) { 1699 mutex_enter(&msp->ms_lock); 1700 VERIFY0(metaslab_load(msp)); 1701 } 1702 1703 if (dump_opt['m'] > 2 && !dump_opt['L']) { 1704 zfs_range_tree_stat_verify(msp->ms_allocatable); 1705 dump_metaslab_stats(msp); 1706 } 1707 1708 if (dump_opt[ALLOCATED_OPT]) { 1709 uint64_t off = msp->ms_start; 1710 zfs_range_tree_walk(msp->ms_allocatable, dump_allocated, 1711 &off); 1712 if (off != msp->ms_start + msp->ms_size) 1713 (void) printf("ALLOC: %"PRIu64" %"PRIu64"\n", off, 1714 msp->ms_size - off); 1715 } 1716 1717 if (dump_opt['m'] > 1 && sm != NULL && 1718 spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) { 1719 /* 1720 * The space map histogram represents free space in chunks 1721 * of sm_shift (i.e. bucket 0 refers to 2^sm_shift). 1722 */ 1723 (void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n", 1724 (u_longlong_t)msp->ms_fragmentation); 1725 dump_histogram(sm->sm_phys->smp_histogram, 1726 SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift); 1727 } 1728 1729 if (dump_opt[ALLOCATED_OPT] || 1730 (dump_opt['m'] > 2 && !dump_opt['L'])) { 1731 metaslab_unload(msp); 1732 mutex_exit(&msp->ms_lock); 1733 } 1734 1735 if (vd->vdev_ops == &vdev_draid_ops) 1736 ASSERT3U(msp->ms_size, <=, 1ULL << vd->vdev_ms_shift); 1737 else 1738 ASSERT3U(msp->ms_size, ==, 1ULL << vd->vdev_ms_shift); 1739 1740 dump_spacemap(spa->spa_meta_objset, msp->ms_sm); 1741 1742 if (spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) { 1743 (void) printf("\tFlush data:\n\tunflushed txg=%llu\n\n", 1744 (u_longlong_t)metaslab_unflushed_txg(msp)); 1745 } 1746 } 1747 1748 static void 1749 print_vdev_metaslab_header(vdev_t *vd) 1750 { 1751 vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias; 1752 const char *bias_str = ""; 1753 if (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) { 1754 bias_str = VDEV_ALLOC_BIAS_LOG; 1755 } else if (alloc_bias == VDEV_BIAS_SPECIAL) { 1756 bias_str = VDEV_ALLOC_BIAS_SPECIAL; 1757 } else if (alloc_bias == VDEV_BIAS_DEDUP) { 1758 bias_str = VDEV_ALLOC_BIAS_DEDUP; 1759 } 1760 1761 uint64_t ms_flush_data_obj = 0; 1762 if (vd->vdev_top_zap != 0) { 1763 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), 1764 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, 1765 sizeof (uint64_t), 1, &ms_flush_data_obj); 1766 if (error != ENOENT) { 1767 ASSERT0(error); 1768 } 1769 } 1770 1771 (void) printf("\tvdev %10llu\t%s metaslab shift %4llu", 1772 (u_longlong_t)vd->vdev_id, bias_str, 1773 (u_longlong_t)vd->vdev_ms_shift); 1774 1775 if (ms_flush_data_obj != 0) { 1776 (void) printf(" ms_unflushed_phys object %llu", 1777 (u_longlong_t)ms_flush_data_obj); 1778 } 1779 1780 (void) printf("\n\t%-10s%5llu %-19s %-15s %-12s\n", 1781 "metaslabs", (u_longlong_t)vd->vdev_ms_count, 1782 "offset", "spacemap", "free"); 1783 (void) printf("\t%15s %19s %15s %12s\n", 1784 "---------------", "-------------------", 1785 "---------------", "------------"); 1786 } 1787 1788 static void 1789 dump_metaslab_groups(spa_t *spa, boolean_t show_special) 1790 { 1791 vdev_t *rvd = spa->spa_root_vdev; 1792 metaslab_class_t *mc = spa_normal_class(spa); 1793 metaslab_class_t *smc = spa_special_class(spa); 1794 uint64_t fragmentation; 1795 1796 metaslab_class_histogram_verify(mc); 1797 1798 for (unsigned c = 0; c < rvd->vdev_children; c++) { 1799 vdev_t *tvd = rvd->vdev_child[c]; 1800 metaslab_group_t *mg = tvd->vdev_mg; 1801 1802 if (mg == NULL || (mg->mg_class != mc && 1803 (!show_special || mg->mg_class != smc))) 1804 continue; 1805 1806 metaslab_group_histogram_verify(mg); 1807 mg->mg_fragmentation = metaslab_group_fragmentation(mg); 1808 1809 (void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t" 1810 "fragmentation", 1811 (u_longlong_t)tvd->vdev_id, 1812 (u_longlong_t)tvd->vdev_ms_count); 1813 if (mg->mg_fragmentation == ZFS_FRAG_INVALID) { 1814 (void) printf("%3s\n", "-"); 1815 } else { 1816 (void) printf("%3llu%%\n", 1817 (u_longlong_t)mg->mg_fragmentation); 1818 } 1819 dump_histogram(mg->mg_histogram, 1820 ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0); 1821 } 1822 1823 (void) printf("\tpool %s\tfragmentation", spa_name(spa)); 1824 fragmentation = metaslab_class_fragmentation(mc); 1825 if (fragmentation == ZFS_FRAG_INVALID) 1826 (void) printf("\t%3s\n", "-"); 1827 else 1828 (void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation); 1829 dump_histogram(mc->mc_histogram, ZFS_RANGE_TREE_HISTOGRAM_SIZE, 0); 1830 } 1831 1832 static void 1833 print_vdev_indirect(vdev_t *vd) 1834 { 1835 vdev_indirect_config_t *vic = &vd->vdev_indirect_config; 1836 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 1837 vdev_indirect_births_t *vib = vd->vdev_indirect_births; 1838 1839 if (vim == NULL) { 1840 ASSERT0P(vib); 1841 return; 1842 } 1843 1844 ASSERT3U(vdev_indirect_mapping_object(vim), ==, 1845 vic->vic_mapping_object); 1846 ASSERT3U(vdev_indirect_births_object(vib), ==, 1847 vic->vic_births_object); 1848 1849 (void) printf("indirect births obj %llu:\n", 1850 (longlong_t)vic->vic_births_object); 1851 (void) printf(" vib_count = %llu\n", 1852 (longlong_t)vdev_indirect_births_count(vib)); 1853 for (uint64_t i = 0; i < vdev_indirect_births_count(vib); i++) { 1854 vdev_indirect_birth_entry_phys_t *cur_vibe = 1855 &vib->vib_entries[i]; 1856 (void) printf("\toffset %llx -> txg %llu\n", 1857 (longlong_t)cur_vibe->vibe_offset, 1858 (longlong_t)cur_vibe->vibe_phys_birth_txg); 1859 } 1860 (void) printf("\n"); 1861 1862 (void) printf("indirect mapping obj %llu:\n", 1863 (longlong_t)vic->vic_mapping_object); 1864 (void) printf(" vim_max_offset = 0x%llx\n", 1865 (longlong_t)vdev_indirect_mapping_max_offset(vim)); 1866 (void) printf(" vim_bytes_mapped = 0x%llx\n", 1867 (longlong_t)vdev_indirect_mapping_bytes_mapped(vim)); 1868 (void) printf(" vim_count = %llu\n", 1869 (longlong_t)vdev_indirect_mapping_num_entries(vim)); 1870 1871 if (dump_opt['d'] <= 5 && dump_opt['m'] <= 3) 1872 return; 1873 1874 uint32_t *counts = vdev_indirect_mapping_load_obsolete_counts(vim); 1875 1876 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) { 1877 vdev_indirect_mapping_entry_phys_t *vimep = 1878 &vim->vim_entries[i]; 1879 (void) printf("\t<%llx:%llx:%llx> -> " 1880 "<%llx:%llx:%llx> (%x obsolete)\n", 1881 (longlong_t)vd->vdev_id, 1882 (longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep), 1883 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst), 1884 (longlong_t)DVA_GET_VDEV(&vimep->vimep_dst), 1885 (longlong_t)DVA_GET_OFFSET(&vimep->vimep_dst), 1886 (longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst), 1887 counts[i]); 1888 } 1889 (void) printf("\n"); 1890 1891 uint64_t obsolete_sm_object; 1892 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object)); 1893 if (obsolete_sm_object != 0) { 1894 objset_t *mos = vd->vdev_spa->spa_meta_objset; 1895 (void) printf("obsolete space map object %llu:\n", 1896 (u_longlong_t)obsolete_sm_object); 1897 ASSERT(vd->vdev_obsolete_sm != NULL); 1898 ASSERT3U(space_map_object(vd->vdev_obsolete_sm), ==, 1899 obsolete_sm_object); 1900 dump_spacemap(mos, vd->vdev_obsolete_sm); 1901 (void) printf("\n"); 1902 } 1903 } 1904 1905 static void 1906 dump_metaslabs(spa_t *spa) 1907 { 1908 vdev_t *vd, *rvd = spa->spa_root_vdev; 1909 uint64_t m, c = 0, children = rvd->vdev_children; 1910 1911 (void) printf("\nMetaslabs:\n"); 1912 1913 if (zopt_metaslab_args > 0) { 1914 c = zopt_metaslab[0]; 1915 1916 if (c >= children) 1917 (void) fatal("bad vdev id: %llu", (u_longlong_t)c); 1918 1919 if (zopt_metaslab_args > 1) { 1920 vd = rvd->vdev_child[c]; 1921 print_vdev_metaslab_header(vd); 1922 1923 for (m = 1; m < zopt_metaslab_args; m++) { 1924 if (zopt_metaslab[m] < vd->vdev_ms_count) 1925 dump_metaslab( 1926 vd->vdev_ms[zopt_metaslab[m]]); 1927 else 1928 (void) fprintf(stderr, "bad metaslab " 1929 "number %llu\n", 1930 (u_longlong_t)zopt_metaslab[m]); 1931 } 1932 (void) printf("\n"); 1933 return; 1934 } 1935 children = c + 1; 1936 } 1937 for (; c < children; c++) { 1938 vd = rvd->vdev_child[c]; 1939 print_vdev_metaslab_header(vd); 1940 1941 print_vdev_indirect(vd); 1942 1943 for (m = 0; m < vd->vdev_ms_count; m++) 1944 dump_metaslab(vd->vdev_ms[m]); 1945 (void) printf("\n"); 1946 } 1947 } 1948 1949 static void 1950 dump_log_spacemaps(spa_t *spa) 1951 { 1952 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) 1953 return; 1954 1955 (void) printf("\nLog Space Maps in Pool:\n"); 1956 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg); 1957 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) { 1958 space_map_t *sm = NULL; 1959 VERIFY0(space_map_open(&sm, spa_meta_objset(spa), 1960 sls->sls_sm_obj, 0, UINT64_MAX, SPA_MINBLOCKSHIFT)); 1961 1962 (void) printf("Log Spacemap object %llu txg %llu\n", 1963 (u_longlong_t)sls->sls_sm_obj, (u_longlong_t)sls->sls_txg); 1964 dump_spacemap(spa->spa_meta_objset, sm); 1965 space_map_close(sm); 1966 } 1967 (void) printf("\n"); 1968 } 1969 1970 static void 1971 dump_ddt_entry(const ddt_t *ddt, const ddt_lightweight_entry_t *ddlwe, 1972 uint64_t index) 1973 { 1974 const ddt_key_t *ddk = &ddlwe->ddlwe_key; 1975 char blkbuf[BP_SPRINTF_LEN]; 1976 blkptr_t blk; 1977 int p; 1978 1979 for (p = 0; p < DDT_NPHYS(ddt); p++) { 1980 const ddt_univ_phys_t *ddp = &ddlwe->ddlwe_phys; 1981 ddt_phys_variant_t v = DDT_PHYS_VARIANT(ddt, p); 1982 1983 if (ddt_phys_birth(ddp, v) == 0) 1984 continue; 1985 ddt_bp_create(ddt->ddt_checksum, ddk, ddp, v, &blk); 1986 snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk); 1987 (void) printf("index %llx refcnt %llu phys %d %s\n", 1988 (u_longlong_t)index, (u_longlong_t)ddt_phys_refcnt(ddp, v), 1989 p, blkbuf); 1990 } 1991 } 1992 1993 static void 1994 dump_dedup_ratio(const ddt_stat_t *dds) 1995 { 1996 double rL, rP, rD, D, dedup, compress, copies; 1997 1998 if (dds->dds_blocks == 0) 1999 return; 2000 2001 rL = (double)dds->dds_ref_lsize; 2002 rP = (double)dds->dds_ref_psize; 2003 rD = (double)dds->dds_ref_dsize; 2004 D = (double)dds->dds_dsize; 2005 2006 dedup = rD / D; 2007 compress = rL / rP; 2008 copies = rD / rP; 2009 2010 (void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, " 2011 "dedup * compress / copies = %.2f\n\n", 2012 dedup, compress, copies, dedup * compress / copies); 2013 } 2014 2015 static void 2016 dump_ddt_log(ddt_t *ddt) 2017 { 2018 if (ddt->ddt_version != DDT_VERSION_FDT || 2019 !(ddt->ddt_flags & DDT_FLAG_LOG)) 2020 return; 2021 2022 for (int n = 0; n < 2; n++) { 2023 ddt_log_t *ddl = &ddt->ddt_log[n]; 2024 2025 char flagstr[64] = {0}; 2026 if (ddl->ddl_flags > 0) { 2027 flagstr[0] = ' '; 2028 int c = 1; 2029 if (ddl->ddl_flags & DDL_FLAG_FLUSHING) 2030 c += strlcpy(&flagstr[c], " FLUSHING", 2031 sizeof (flagstr) - c); 2032 if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT) 2033 c += strlcpy(&flagstr[c], " CHECKPOINT", 2034 sizeof (flagstr) - c); 2035 if (ddl->ddl_flags & 2036 ~(DDL_FLAG_FLUSHING|DDL_FLAG_CHECKPOINT)) 2037 c += strlcpy(&flagstr[c], " UNKNOWN", 2038 sizeof (flagstr) - c); 2039 flagstr[1] = '['; 2040 flagstr[c] = ']'; 2041 } 2042 2043 uint64_t count = avl_numnodes(&ddl->ddl_tree); 2044 2045 printf(DMU_POOL_DDT_LOG ": flags=0x%02x%s; obj=%llu; " 2046 "len=%llu; txg=%llu; entries=%llu\n", 2047 zio_checksum_table[ddt->ddt_checksum].ci_name, n, 2048 ddl->ddl_flags, flagstr, 2049 (u_longlong_t)ddl->ddl_object, 2050 (u_longlong_t)ddl->ddl_length, 2051 (u_longlong_t)ddl->ddl_first_txg, (u_longlong_t)count); 2052 2053 if (ddl->ddl_flags & DDL_FLAG_CHECKPOINT) { 2054 const ddt_key_t *ddk = &ddl->ddl_checkpoint; 2055 printf(" checkpoint: " 2056 "%016llx:%016llx:%016llx:%016llx:%016llx\n", 2057 (u_longlong_t)ddk->ddk_cksum.zc_word[0], 2058 (u_longlong_t)ddk->ddk_cksum.zc_word[1], 2059 (u_longlong_t)ddk->ddk_cksum.zc_word[2], 2060 (u_longlong_t)ddk->ddk_cksum.zc_word[3], 2061 (u_longlong_t)ddk->ddk_prop); 2062 } 2063 2064 if (count == 0 || dump_opt['D'] < 4) 2065 continue; 2066 2067 ddt_lightweight_entry_t ddlwe; 2068 uint64_t index = 0; 2069 for (ddt_log_entry_t *ddle = avl_first(&ddl->ddl_tree); 2070 ddle; ddle = AVL_NEXT(&ddl->ddl_tree, ddle)) { 2071 DDT_LOG_ENTRY_TO_LIGHTWEIGHT(ddt, ddle, &ddlwe); 2072 dump_ddt_entry(ddt, &ddlwe, index++); 2073 } 2074 } 2075 } 2076 2077 static void 2078 dump_ddt_object(ddt_t *ddt, ddt_type_t type, ddt_class_t class) 2079 { 2080 char name[DDT_NAMELEN]; 2081 ddt_lightweight_entry_t ddlwe; 2082 uint64_t walk = 0; 2083 dmu_object_info_t doi; 2084 uint64_t count, dspace, mspace; 2085 int error; 2086 2087 error = ddt_object_info(ddt, type, class, &doi); 2088 2089 if (error == ENOENT) 2090 return; 2091 ASSERT0(error); 2092 2093 error = ddt_object_count(ddt, type, class, &count); 2094 ASSERT0(error); 2095 if (count == 0) 2096 return; 2097 2098 dspace = doi.doi_physical_blocks_512 << 9; 2099 mspace = doi.doi_fill_count * doi.doi_data_block_size; 2100 2101 ddt_object_name(ddt, type, class, name); 2102 2103 (void) printf("%s: dspace=%llu; mspace=%llu; entries=%llu\n", name, 2104 (u_longlong_t)dspace, (u_longlong_t)mspace, (u_longlong_t)count); 2105 2106 if (dump_opt['D'] < 3) 2107 return; 2108 2109 (void) printf("%s: object=%llu\n", name, 2110 (u_longlong_t)ddt->ddt_object[type][class]); 2111 zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class]); 2112 2113 if (dump_opt['D'] < 4) 2114 return; 2115 2116 if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE) 2117 return; 2118 2119 (void) printf("%s contents:\n\n", name); 2120 2121 while ((error = ddt_object_walk(ddt, type, class, &walk, &ddlwe)) == 0) 2122 dump_ddt_entry(ddt, &ddlwe, walk); 2123 2124 ASSERT3U(error, ==, ENOENT); 2125 2126 (void) printf("\n"); 2127 } 2128 2129 static void 2130 dump_ddt(ddt_t *ddt) 2131 { 2132 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED) 2133 return; 2134 2135 char flagstr[64] = {0}; 2136 if (ddt->ddt_flags > 0) { 2137 flagstr[0] = ' '; 2138 int c = 1; 2139 if (ddt->ddt_flags & DDT_FLAG_FLAT) 2140 c += strlcpy(&flagstr[c], " FLAT", 2141 sizeof (flagstr) - c); 2142 if (ddt->ddt_flags & DDT_FLAG_LOG) 2143 c += strlcpy(&flagstr[c], " LOG", 2144 sizeof (flagstr) - c); 2145 if (ddt->ddt_flags & ~DDT_FLAG_MASK) 2146 c += strlcpy(&flagstr[c], " UNKNOWN", 2147 sizeof (flagstr) - c); 2148 flagstr[1] = '['; 2149 flagstr[c] = ']'; 2150 } 2151 2152 printf("DDT-%s: version=%llu [%s]; flags=0x%02llx%s; rootobj=%llu\n", 2153 zio_checksum_table[ddt->ddt_checksum].ci_name, 2154 (u_longlong_t)ddt->ddt_version, 2155 (ddt->ddt_version == 0) ? "LEGACY" : 2156 (ddt->ddt_version == 1) ? "FDT" : "UNKNOWN", 2157 (u_longlong_t)ddt->ddt_flags, flagstr, 2158 (u_longlong_t)ddt->ddt_dir_object); 2159 2160 for (ddt_type_t type = 0; type < DDT_TYPES; type++) 2161 for (ddt_class_t class = 0; class < DDT_CLASSES; class++) 2162 dump_ddt_object(ddt, type, class); 2163 2164 dump_ddt_log(ddt); 2165 } 2166 2167 static void 2168 dump_all_ddts(spa_t *spa) 2169 { 2170 ddt_histogram_t ddh_total = {{{0}}}; 2171 ddt_stat_t dds_total = {0}; 2172 2173 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) 2174 dump_ddt(spa->spa_ddt[c]); 2175 2176 ddt_get_dedup_stats(spa, &dds_total); 2177 2178 if (dds_total.dds_blocks == 0) { 2179 (void) printf("All DDTs are empty\n"); 2180 return; 2181 } 2182 2183 (void) printf("\n"); 2184 2185 if (dump_opt['D'] > 1) { 2186 (void) printf("DDT histogram (aggregated over all DDTs):\n"); 2187 ddt_get_dedup_histogram(spa, &ddh_total); 2188 zpool_dump_ddt(&dds_total, &ddh_total); 2189 } 2190 2191 dump_dedup_ratio(&dds_total); 2192 2193 /* 2194 * Dump a histogram of unique class entry age 2195 */ 2196 if (dump_opt['D'] == 3 && getenv("ZDB_DDT_UNIQUE_AGE_HIST") != NULL) { 2197 ddt_age_histo_t histogram; 2198 2199 (void) printf("DDT walk unique, building age histogram...\n"); 2200 ddt_prune_walk(spa, 0, &histogram); 2201 2202 /* 2203 * print out histogram for unique entry class birth 2204 */ 2205 if (histogram.dah_entries > 0) { 2206 (void) printf("%5s %9s %4s\n", 2207 "age", "blocks", "amnt"); 2208 (void) printf("%5s %9s %4s\n", 2209 "-----", "---------", "----"); 2210 for (int i = 0; i < HIST_BINS; i++) { 2211 (void) printf("%5d %9d %4d%%\n", 1 << i, 2212 (int)histogram.dah_age_histo[i], 2213 (int)((histogram.dah_age_histo[i] * 100) / 2214 histogram.dah_entries)); 2215 } 2216 } 2217 } 2218 } 2219 2220 static void 2221 dump_brt(spa_t *spa) 2222 { 2223 if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) { 2224 printf("BRT: unsupported on this pool\n"); 2225 return; 2226 } 2227 2228 if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) { 2229 printf("BRT: empty\n"); 2230 return; 2231 } 2232 2233 char count[32], used[32], saved[32]; 2234 zdb_nicebytes(brt_get_used(spa), used, sizeof (used)); 2235 zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved)); 2236 uint64_t ratio = brt_get_ratio(spa); 2237 printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved, 2238 (u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100)); 2239 2240 if (dump_opt['T'] < 2) 2241 return; 2242 2243 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) { 2244 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid]; 2245 if (!brtvd->bv_initiated) { 2246 printf("BRT: vdev %" PRIu64 ": empty\n", vdevid); 2247 continue; 2248 } 2249 2250 zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count)); 2251 zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used)); 2252 zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved)); 2253 printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n", 2254 vdevid, count, used, saved); 2255 } 2256 2257 if (dump_opt['T'] < 3) 2258 return; 2259 2260 /* -TTT shows a per-vdev histograms; -TTTT shows all entries */ 2261 boolean_t do_histo = dump_opt['T'] == 3; 2262 2263 char dva[64]; 2264 2265 if (!do_histo) 2266 printf("\n%-16s %-10s\n", "DVA", "REFCNT"); 2267 2268 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) { 2269 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid]; 2270 if (!brtvd->bv_initiated) 2271 continue; 2272 2273 uint64_t counts[64] = {}; 2274 2275 zap_cursor_t zc; 2276 zap_attribute_t *za = zap_attribute_alloc(); 2277 for (zap_cursor_init(&zc, spa->spa_meta_objset, 2278 brtvd->bv_mos_entries); 2279 zap_cursor_retrieve(&zc, za) == 0; 2280 zap_cursor_advance(&zc)) { 2281 uint64_t refcnt; 2282 VERIFY0(zap_lookup_uint64(spa->spa_meta_objset, 2283 brtvd->bv_mos_entries, 2284 (const uint64_t *)za->za_name, 1, 2285 za->za_integer_length, za->za_num_integers, 2286 &refcnt)); 2287 2288 if (do_histo) 2289 counts[highbit64(refcnt)]++; 2290 else { 2291 uint64_t offset = 2292 *(const uint64_t *)za->za_name; 2293 2294 snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx", 2295 vdevid, (u_longlong_t)offset); 2296 printf("%-16s %-10llu\n", dva, 2297 (u_longlong_t)refcnt); 2298 } 2299 } 2300 zap_cursor_fini(&zc); 2301 zap_attribute_free(za); 2302 2303 if (do_histo) { 2304 printf("\nBRT: vdev %" PRIu64 2305 ": DVAs with 2^n refcnts:\n", vdevid); 2306 dump_histogram(counts, 64, 0); 2307 } 2308 } 2309 } 2310 2311 static void 2312 dump_dtl_seg(void *arg, uint64_t start, uint64_t size) 2313 { 2314 char *prefix = arg; 2315 2316 (void) printf("%s [%llu,%llu) length %llu\n", 2317 prefix, 2318 (u_longlong_t)start, 2319 (u_longlong_t)(start + size), 2320 (u_longlong_t)(size)); 2321 } 2322 2323 static void 2324 dump_dtl(vdev_t *vd, int indent) 2325 { 2326 spa_t *spa = vd->vdev_spa; 2327 boolean_t required; 2328 const char *name[DTL_TYPES] = { "missing", "partial", "scrub", 2329 "outage" }; 2330 char prefix[256]; 2331 2332 spa_vdev_state_enter(spa, SCL_NONE); 2333 required = vdev_dtl_required(vd); 2334 (void) spa_vdev_state_exit(spa, NULL, 0); 2335 2336 if (indent == 0) 2337 (void) printf("\nDirty time logs:\n\n"); 2338 2339 (void) printf("\t%*s%s [%s]\n", indent, "", 2340 vd->vdev_path ? vd->vdev_path : 2341 vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa), 2342 required ? "DTL-required" : "DTL-expendable"); 2343 2344 for (int t = 0; t < DTL_TYPES; t++) { 2345 zfs_range_tree_t *rt = vd->vdev_dtl[t]; 2346 if (zfs_range_tree_space(rt) == 0) 2347 continue; 2348 (void) snprintf(prefix, sizeof (prefix), "\t%*s%s", 2349 indent + 2, "", name[t]); 2350 zfs_range_tree_walk(rt, dump_dtl_seg, prefix); 2351 if (dump_opt['d'] > 5 && vd->vdev_children == 0) 2352 dump_spacemap(spa->spa_meta_objset, 2353 vd->vdev_dtl_sm); 2354 } 2355 2356 for (unsigned c = 0; c < vd->vdev_children; c++) 2357 dump_dtl(vd->vdev_child[c], indent + 4); 2358 } 2359 2360 static void 2361 dump_history(spa_t *spa) 2362 { 2363 nvlist_t **events = NULL; 2364 char *buf; 2365 uint64_t resid, len, off = 0; 2366 uint_t num = 0; 2367 int error; 2368 char tbuf[30]; 2369 2370 if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) { 2371 (void) fprintf(stderr, "%s: unable to allocate I/O buffer\n", 2372 __func__); 2373 return; 2374 } 2375 2376 do { 2377 len = SPA_OLD_MAXBLOCKSIZE; 2378 2379 if ((error = spa_history_get(spa, &off, &len, buf)) != 0) { 2380 (void) fprintf(stderr, "Unable to read history: " 2381 "error %d\n", error); 2382 free(buf); 2383 return; 2384 } 2385 2386 if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0) 2387 break; 2388 2389 off -= resid; 2390 } while (len != 0); 2391 2392 (void) printf("\nHistory:\n"); 2393 for (unsigned i = 0; i < num; i++) { 2394 boolean_t printed = B_FALSE; 2395 2396 if (nvlist_exists(events[i], ZPOOL_HIST_TIME)) { 2397 time_t tsec; 2398 struct tm t; 2399 2400 tsec = fnvlist_lookup_uint64(events[i], 2401 ZPOOL_HIST_TIME); 2402 (void) localtime_r(&tsec, &t); 2403 (void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t); 2404 } else { 2405 tbuf[0] = '\0'; 2406 } 2407 2408 if (nvlist_exists(events[i], ZPOOL_HIST_CMD)) { 2409 (void) printf("%s %s\n", tbuf, 2410 fnvlist_lookup_string(events[i], ZPOOL_HIST_CMD)); 2411 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_EVENT)) { 2412 uint64_t ievent; 2413 2414 ievent = fnvlist_lookup_uint64(events[i], 2415 ZPOOL_HIST_INT_EVENT); 2416 if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS) 2417 goto next; 2418 2419 (void) printf(" %s [internal %s txg:%ju] %s\n", 2420 tbuf, 2421 zfs_history_event_names[ievent], 2422 fnvlist_lookup_uint64(events[i], 2423 ZPOOL_HIST_TXG), 2424 fnvlist_lookup_string(events[i], 2425 ZPOOL_HIST_INT_STR)); 2426 } else if (nvlist_exists(events[i], ZPOOL_HIST_INT_NAME)) { 2427 (void) printf("%s [txg:%ju] %s", tbuf, 2428 fnvlist_lookup_uint64(events[i], 2429 ZPOOL_HIST_TXG), 2430 fnvlist_lookup_string(events[i], 2431 ZPOOL_HIST_INT_NAME)); 2432 2433 if (nvlist_exists(events[i], ZPOOL_HIST_DSNAME)) { 2434 (void) printf(" %s (%llu)", 2435 fnvlist_lookup_string(events[i], 2436 ZPOOL_HIST_DSNAME), 2437 (u_longlong_t)fnvlist_lookup_uint64( 2438 events[i], 2439 ZPOOL_HIST_DSID)); 2440 } 2441 2442 (void) printf(" %s\n", fnvlist_lookup_string(events[i], 2443 ZPOOL_HIST_INT_STR)); 2444 } else if (nvlist_exists(events[i], ZPOOL_HIST_IOCTL)) { 2445 (void) printf("%s ioctl %s\n", tbuf, 2446 fnvlist_lookup_string(events[i], 2447 ZPOOL_HIST_IOCTL)); 2448 2449 if (nvlist_exists(events[i], ZPOOL_HIST_INPUT_NVL)) { 2450 (void) printf(" input:\n"); 2451 dump_nvlist(fnvlist_lookup_nvlist(events[i], 2452 ZPOOL_HIST_INPUT_NVL), 8); 2453 } 2454 if (nvlist_exists(events[i], ZPOOL_HIST_OUTPUT_NVL)) { 2455 (void) printf(" output:\n"); 2456 dump_nvlist(fnvlist_lookup_nvlist(events[i], 2457 ZPOOL_HIST_OUTPUT_NVL), 8); 2458 } 2459 if (nvlist_exists(events[i], ZPOOL_HIST_ERRNO)) { 2460 (void) printf(" errno: %lld\n", 2461 (longlong_t)fnvlist_lookup_int64(events[i], 2462 ZPOOL_HIST_ERRNO)); 2463 } 2464 } else { 2465 goto next; 2466 } 2467 2468 printed = B_TRUE; 2469 next: 2470 if (dump_opt['h'] > 1) { 2471 if (!printed) 2472 (void) printf("unrecognized record:\n"); 2473 dump_nvlist(events[i], 2); 2474 } 2475 } 2476 free(buf); 2477 } 2478 2479 static void 2480 dump_dnode(objset_t *os, uint64_t object, void *data, size_t size) 2481 { 2482 (void) os, (void) object, (void) data, (void) size; 2483 } 2484 2485 static uint64_t 2486 blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp, 2487 const zbookmark_phys_t *zb) 2488 { 2489 if (dnp == NULL) { 2490 ASSERT(zb->zb_level < 0); 2491 if (zb->zb_object == 0) 2492 return (zb->zb_blkid); 2493 return (zb->zb_blkid * BP_GET_LSIZE(bp)); 2494 } 2495 2496 ASSERT(zb->zb_level >= 0); 2497 2498 return ((zb->zb_blkid << 2499 (zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) * 2500 dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT); 2501 } 2502 2503 static void 2504 snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen, 2505 const blkptr_t *bp) 2506 { 2507 static abd_t *pabd = NULL; 2508 void *buf; 2509 zio_t *zio; 2510 zfs_zstdhdr_t zstd_hdr; 2511 int error; 2512 2513 if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_ZSTD) 2514 return; 2515 2516 if (BP_IS_HOLE(bp)) 2517 return; 2518 2519 if (BP_IS_EMBEDDED(bp)) { 2520 buf = malloc(SPA_MAXBLOCKSIZE); 2521 if (buf == NULL) { 2522 (void) fprintf(stderr, "out of memory\n"); 2523 zdb_exit(1); 2524 } 2525 decode_embedded_bp_compressed(bp, buf); 2526 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr)); 2527 free(buf); 2528 zstd_hdr.c_len = BE_32(zstd_hdr.c_len); 2529 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level); 2530 (void) snprintf(blkbuf + strlen(blkbuf), 2531 buflen - strlen(blkbuf), 2532 " ZSTD:size=%u:version=%u:level=%u:EMBEDDED", 2533 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr), 2534 zfs_get_hdrlevel(&zstd_hdr)); 2535 return; 2536 } 2537 2538 if (!pabd) 2539 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE); 2540 zio = zio_root(spa, NULL, NULL, 0); 2541 2542 /* Decrypt but don't decompress so we can read the compression header */ 2543 zio_nowait(zio_read(zio, spa, bp, pabd, BP_GET_PSIZE(bp), NULL, NULL, 2544 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW_COMPRESS, 2545 NULL)); 2546 error = zio_wait(zio); 2547 if (error) { 2548 (void) fprintf(stderr, "read failed: %d\n", error); 2549 return; 2550 } 2551 buf = abd_borrow_buf_copy(pabd, BP_GET_LSIZE(bp)); 2552 memcpy(&zstd_hdr, buf, sizeof (zstd_hdr)); 2553 zstd_hdr.c_len = BE_32(zstd_hdr.c_len); 2554 zstd_hdr.raw_version_level = BE_32(zstd_hdr.raw_version_level); 2555 2556 (void) snprintf(blkbuf + strlen(blkbuf), 2557 buflen - strlen(blkbuf), 2558 " ZSTD:size=%u:version=%u:level=%u:NORMAL", 2559 zstd_hdr.c_len, zfs_get_hdrversion(&zstd_hdr), 2560 zfs_get_hdrlevel(&zstd_hdr)); 2561 2562 abd_return_buf_copy(pabd, buf, BP_GET_LSIZE(bp)); 2563 } 2564 2565 static void 2566 snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp, 2567 boolean_t bp_freed) 2568 { 2569 const dva_t *dva = bp->blk_dva; 2570 int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1; 2571 int i; 2572 2573 if (dump_opt['b'] >= 6) { 2574 snprintf_blkptr(blkbuf, buflen, bp); 2575 if (bp_freed) { 2576 (void) snprintf(blkbuf + strlen(blkbuf), 2577 buflen - strlen(blkbuf), " %s", "FREE"); 2578 } 2579 return; 2580 } 2581 2582 if (BP_IS_EMBEDDED(bp)) { 2583 (void) sprintf(blkbuf, 2584 "EMBEDDED et=%u %llxL/%llxP B=%llu", 2585 (int)BPE_GET_ETYPE(bp), 2586 (u_longlong_t)BPE_GET_LSIZE(bp), 2587 (u_longlong_t)BPE_GET_PSIZE(bp), 2588 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp)); 2589 return; 2590 } 2591 2592 blkbuf[0] = '\0'; 2593 2594 for (i = 0; i < ndvas; i++) { 2595 (void) snprintf(blkbuf + strlen(blkbuf), 2596 buflen - strlen(blkbuf), "%llu:%llx:%llx%s ", 2597 (u_longlong_t)DVA_GET_VDEV(&dva[i]), 2598 (u_longlong_t)DVA_GET_OFFSET(&dva[i]), 2599 (u_longlong_t)DVA_GET_ASIZE(&dva[i]), 2600 (DVA_GET_GANG(&dva[i]) ? "G" : "")); 2601 } 2602 2603 if (BP_IS_HOLE(bp)) { 2604 (void) snprintf(blkbuf + strlen(blkbuf), 2605 buflen - strlen(blkbuf), 2606 "%llxL B=%llu", 2607 (u_longlong_t)BP_GET_LSIZE(bp), 2608 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp)); 2609 } else { 2610 (void) snprintf(blkbuf + strlen(blkbuf), 2611 buflen - strlen(blkbuf), 2612 "%llxL/%llxP F=%llu B=%llu/%llu", 2613 (u_longlong_t)BP_GET_LSIZE(bp), 2614 (u_longlong_t)BP_GET_PSIZE(bp), 2615 (u_longlong_t)BP_GET_FILL(bp), 2616 (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp), 2617 (u_longlong_t)BP_GET_PHYSICAL_BIRTH(bp)); 2618 if (bp_freed) 2619 (void) snprintf(blkbuf + strlen(blkbuf), 2620 buflen - strlen(blkbuf), " %s", "FREE"); 2621 (void) snprintf(blkbuf + strlen(blkbuf), 2622 buflen - strlen(blkbuf), 2623 " cksum=%016llx:%016llx:%016llx:%016llx", 2624 (u_longlong_t)bp->blk_cksum.zc_word[0], 2625 (u_longlong_t)bp->blk_cksum.zc_word[1], 2626 (u_longlong_t)bp->blk_cksum.zc_word[2], 2627 (u_longlong_t)bp->blk_cksum.zc_word[3]); 2628 } 2629 } 2630 2631 static u_longlong_t 2632 print_indirect(spa_t *spa, blkptr_t *bp, const zbookmark_phys_t *zb, 2633 const dnode_phys_t *dnp) 2634 { 2635 char blkbuf[BP_SPRINTF_LEN]; 2636 u_longlong_t offset; 2637 int l; 2638 2639 offset = (u_longlong_t)blkid2offset(dnp, bp, zb); 2640 2641 (void) printf("%16llx ", offset); 2642 2643 ASSERT(zb->zb_level >= 0); 2644 2645 for (l = dnp->dn_nlevels - 1; l >= -1; l--) { 2646 if (l == zb->zb_level) { 2647 (void) printf("L%llx", (u_longlong_t)zb->zb_level); 2648 } else { 2649 (void) printf(" "); 2650 } 2651 } 2652 2653 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, B_FALSE); 2654 if (dump_opt['Z'] && BP_GET_COMPRESS(bp) == ZIO_COMPRESS_ZSTD) 2655 snprintf_zstd_header(spa, blkbuf, sizeof (blkbuf), bp); 2656 (void) printf("%s", blkbuf); 2657 2658 if (!BP_IS_EMBEDDED(bp)) { 2659 if (BP_GET_TYPE(bp) != dnp->dn_type) { 2660 (void) printf(" (ERROR: Block pointer type " 2661 "(%llu) does not match dnode type (%hhu))", 2662 BP_GET_TYPE(bp), dnp->dn_type); 2663 corruption_found = B_TRUE; 2664 } 2665 if (BP_GET_LEVEL(bp) != zb->zb_level) { 2666 (void) printf(" (ERROR: Block pointer level " 2667 "(%llu) does not match bookmark level (%lld))", 2668 BP_GET_LEVEL(bp), (longlong_t)zb->zb_level); 2669 corruption_found = B_TRUE; 2670 } 2671 } 2672 (void) printf("\n"); 2673 2674 return (offset); 2675 } 2676 2677 static int 2678 visit_indirect(spa_t *spa, const dnode_phys_t *dnp, 2679 blkptr_t *bp, const zbookmark_phys_t *zb) 2680 { 2681 u_longlong_t offset; 2682 int err = 0; 2683 2684 if (BP_GET_BIRTH(bp) == 0) 2685 return (0); 2686 2687 offset = print_indirect(spa, bp, zb, dnp); 2688 2689 if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) { 2690 arc_flags_t flags = ARC_FLAG_WAIT; 2691 int i; 2692 blkptr_t *cbp; 2693 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; 2694 arc_buf_t *buf; 2695 uint64_t fill = 0; 2696 ASSERT(!BP_IS_REDACTED(bp)); 2697 2698 err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf, 2699 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); 2700 if (err) 2701 return (err); 2702 ASSERT(buf->b_data); 2703 2704 /* recursively visit blocks below this */ 2705 cbp = buf->b_data; 2706 for (i = 0; i < epb; i++, cbp++) { 2707 zbookmark_phys_t czb; 2708 2709 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, 2710 zb->zb_level - 1, 2711 zb->zb_blkid * epb + i); 2712 err = visit_indirect(spa, dnp, cbp, &czb); 2713 if (err) 2714 break; 2715 fill += BP_GET_FILL(cbp); 2716 } 2717 if (!err) { 2718 if (fill != BP_GET_FILL(bp)) { 2719 (void) printf("%16llx: Block pointer " 2720 "fill (%llu) does not match calculated " 2721 "value (%llu)\n", offset, BP_GET_FILL(bp), 2722 (u_longlong_t)fill); 2723 corruption_found = B_TRUE; 2724 } 2725 } 2726 arc_buf_destroy(buf, &buf); 2727 } 2728 2729 return (err); 2730 } 2731 2732 static void 2733 dump_indirect(dnode_t *dn) 2734 { 2735 dnode_phys_t *dnp = dn->dn_phys; 2736 zbookmark_phys_t czb; 2737 2738 (void) printf("Indirect blocks:\n"); 2739 2740 SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset), 2741 dn->dn_object, dnp->dn_nlevels - 1, 0); 2742 for (int j = 0; j < dnp->dn_nblkptr; j++) { 2743 czb.zb_blkid = j; 2744 (void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp, 2745 &dnp->dn_blkptr[j], &czb); 2746 } 2747 2748 (void) printf("\n"); 2749 } 2750 2751 static void 2752 dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size) 2753 { 2754 (void) os, (void) object; 2755 dsl_dir_phys_t *dd = data; 2756 time_t crtime; 2757 char nice[32]; 2758 2759 /* make sure nicenum has enough space */ 2760 _Static_assert(sizeof (nice) >= NN_NUMBUF_SZ, "nice truncated"); 2761 2762 if (dd == NULL) 2763 return; 2764 2765 ASSERT3U(size, >=, sizeof (dsl_dir_phys_t)); 2766 2767 crtime = dd->dd_creation_time; 2768 (void) printf("\t\tcreation_time = %s", ctime(&crtime)); 2769 (void) printf("\t\thead_dataset_obj = %llu\n", 2770 (u_longlong_t)dd->dd_head_dataset_obj); 2771 (void) printf("\t\tparent_dir_obj = %llu\n", 2772 (u_longlong_t)dd->dd_parent_obj); 2773 (void) printf("\t\torigin_obj = %llu\n", 2774 (u_longlong_t)dd->dd_origin_obj); 2775 (void) printf("\t\tchild_dir_zapobj = %llu\n", 2776 (u_longlong_t)dd->dd_child_dir_zapobj); 2777 zdb_nicenum(dd->dd_used_bytes, nice, sizeof (nice)); 2778 (void) printf("\t\tused_bytes = %s\n", nice); 2779 zdb_nicenum(dd->dd_compressed_bytes, nice, sizeof (nice)); 2780 (void) printf("\t\tcompressed_bytes = %s\n", nice); 2781 zdb_nicenum(dd->dd_uncompressed_bytes, nice, sizeof (nice)); 2782 (void) printf("\t\tuncompressed_bytes = %s\n", nice); 2783 zdb_nicenum(dd->dd_quota, nice, sizeof (nice)); 2784 (void) printf("\t\tquota = %s\n", nice); 2785 zdb_nicenum(dd->dd_reserved, nice, sizeof (nice)); 2786 (void) printf("\t\treserved = %s\n", nice); 2787 (void) printf("\t\tprops_zapobj = %llu\n", 2788 (u_longlong_t)dd->dd_props_zapobj); 2789 (void) printf("\t\tdeleg_zapobj = %llu\n", 2790 (u_longlong_t)dd->dd_deleg_zapobj); 2791 (void) printf("\t\tflags = %llx\n", 2792 (u_longlong_t)dd->dd_flags); 2793 2794 #define DO(which) \ 2795 zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice, \ 2796 sizeof (nice)); \ 2797 (void) printf("\t\tused_breakdown[" #which "] = %s\n", nice) 2798 DO(HEAD); 2799 DO(SNAP); 2800 DO(CHILD); 2801 DO(CHILD_RSRV); 2802 DO(REFRSRV); 2803 #undef DO 2804 (void) printf("\t\tclones = %llu\n", 2805 (u_longlong_t)dd->dd_clones); 2806 } 2807 2808 static void 2809 dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size) 2810 { 2811 (void) os, (void) object; 2812 dsl_dataset_phys_t *ds = data; 2813 time_t crtime; 2814 char used[32], compressed[32], uncompressed[32], unique[32]; 2815 char blkbuf[BP_SPRINTF_LEN]; 2816 2817 /* make sure nicenum has enough space */ 2818 _Static_assert(sizeof (used) >= NN_NUMBUF_SZ, "used truncated"); 2819 _Static_assert(sizeof (compressed) >= NN_NUMBUF_SZ, 2820 "compressed truncated"); 2821 _Static_assert(sizeof (uncompressed) >= NN_NUMBUF_SZ, 2822 "uncompressed truncated"); 2823 _Static_assert(sizeof (unique) >= NN_NUMBUF_SZ, "unique truncated"); 2824 2825 if (ds == NULL) 2826 return; 2827 2828 ASSERT(size == sizeof (*ds)); 2829 crtime = ds->ds_creation_time; 2830 zdb_nicenum(ds->ds_referenced_bytes, used, sizeof (used)); 2831 zdb_nicenum(ds->ds_compressed_bytes, compressed, sizeof (compressed)); 2832 zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed, 2833 sizeof (uncompressed)); 2834 zdb_nicenum(ds->ds_unique_bytes, unique, sizeof (unique)); 2835 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp); 2836 2837 (void) printf("\t\tdir_obj = %llu\n", 2838 (u_longlong_t)ds->ds_dir_obj); 2839 (void) printf("\t\tprev_snap_obj = %llu\n", 2840 (u_longlong_t)ds->ds_prev_snap_obj); 2841 (void) printf("\t\tprev_snap_txg = %llu\n", 2842 (u_longlong_t)ds->ds_prev_snap_txg); 2843 (void) printf("\t\tnext_snap_obj = %llu\n", 2844 (u_longlong_t)ds->ds_next_snap_obj); 2845 (void) printf("\t\tsnapnames_zapobj = %llu\n", 2846 (u_longlong_t)ds->ds_snapnames_zapobj); 2847 (void) printf("\t\tnum_children = %llu\n", 2848 (u_longlong_t)ds->ds_num_children); 2849 (void) printf("\t\tuserrefs_obj = %llu\n", 2850 (u_longlong_t)ds->ds_userrefs_obj); 2851 (void) printf("\t\tcreation_time = %s", ctime(&crtime)); 2852 (void) printf("\t\tcreation_txg = %llu\n", 2853 (u_longlong_t)ds->ds_creation_txg); 2854 (void) printf("\t\tdeadlist_obj = %llu\n", 2855 (u_longlong_t)ds->ds_deadlist_obj); 2856 (void) printf("\t\tused_bytes = %s\n", used); 2857 (void) printf("\t\tcompressed_bytes = %s\n", compressed); 2858 (void) printf("\t\tuncompressed_bytes = %s\n", uncompressed); 2859 (void) printf("\t\tunique = %s\n", unique); 2860 (void) printf("\t\tfsid_guid = %llu\n", 2861 (u_longlong_t)ds->ds_fsid_guid); 2862 (void) printf("\t\tguid = %llu\n", 2863 (u_longlong_t)ds->ds_guid); 2864 (void) printf("\t\tflags = %llx\n", 2865 (u_longlong_t)ds->ds_flags); 2866 (void) printf("\t\tnext_clones_obj = %llu\n", 2867 (u_longlong_t)ds->ds_next_clones_obj); 2868 (void) printf("\t\tprops_obj = %llu\n", 2869 (u_longlong_t)ds->ds_props_obj); 2870 (void) printf("\t\tbp = %s\n", blkbuf); 2871 } 2872 2873 static int 2874 dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 2875 { 2876 (void) arg, (void) tx; 2877 char blkbuf[BP_SPRINTF_LEN]; 2878 2879 if (BP_GET_BIRTH(bp) != 0) { 2880 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp); 2881 (void) printf("\t%s\n", blkbuf); 2882 } 2883 return (0); 2884 } 2885 2886 static void 2887 dump_bptree(objset_t *os, uint64_t obj, const char *name) 2888 { 2889 char bytes[32]; 2890 bptree_phys_t *bt; 2891 dmu_buf_t *db; 2892 2893 /* make sure nicenum has enough space */ 2894 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated"); 2895 2896 if (dump_opt['d'] < 3) 2897 return; 2898 2899 VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db)); 2900 bt = db->db_data; 2901 zdb_nicenum(bt->bt_bytes, bytes, sizeof (bytes)); 2902 (void) printf("\n %s: %llu datasets, %s\n", 2903 name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes); 2904 dmu_buf_rele(db, FTAG); 2905 2906 if (dump_opt['d'] < 5) 2907 return; 2908 2909 (void) printf("\n"); 2910 2911 (void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL); 2912 } 2913 2914 static int 2915 dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx) 2916 { 2917 (void) arg, (void) tx; 2918 char blkbuf[BP_SPRINTF_LEN]; 2919 2920 ASSERT(BP_GET_BIRTH(bp) != 0); 2921 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed); 2922 (void) printf("\t%s\n", blkbuf); 2923 return (0); 2924 } 2925 2926 static void 2927 dump_full_bpobj(bpobj_t *bpo, const char *name, int indent) 2928 { 2929 char bytes[32]; 2930 char comp[32]; 2931 char uncomp[32]; 2932 uint64_t i; 2933 2934 /* make sure nicenum has enough space */ 2935 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated"); 2936 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated"); 2937 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated"); 2938 2939 if (dump_opt['d'] < 3) 2940 return; 2941 2942 zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes, sizeof (bytes)); 2943 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) { 2944 zdb_nicenum(bpo->bpo_phys->bpo_comp, comp, sizeof (comp)); 2945 zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp, sizeof (uncomp)); 2946 if (bpo->bpo_havefreed) { 2947 (void) printf(" %*s: object %llu, %llu local " 2948 "blkptrs, %llu freed, %llu subobjs in object %llu, " 2949 "%s (%s/%s comp)\n", 2950 indent * 8, name, 2951 (u_longlong_t)bpo->bpo_object, 2952 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs, 2953 (u_longlong_t)bpo->bpo_phys->bpo_num_freed, 2954 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs, 2955 (u_longlong_t)bpo->bpo_phys->bpo_subobjs, 2956 bytes, comp, uncomp); 2957 } else { 2958 (void) printf(" %*s: object %llu, %llu local " 2959 "blkptrs, %llu subobjs in object %llu, " 2960 "%s (%s/%s comp)\n", 2961 indent * 8, name, 2962 (u_longlong_t)bpo->bpo_object, 2963 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs, 2964 (u_longlong_t)bpo->bpo_phys->bpo_num_subobjs, 2965 (u_longlong_t)bpo->bpo_phys->bpo_subobjs, 2966 bytes, comp, uncomp); 2967 } 2968 2969 for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) { 2970 uint64_t subobj; 2971 bpobj_t subbpo; 2972 int error; 2973 VERIFY0(dmu_read(bpo->bpo_os, 2974 bpo->bpo_phys->bpo_subobjs, 2975 i * sizeof (subobj), sizeof (subobj), &subobj, 0)); 2976 error = bpobj_open(&subbpo, bpo->bpo_os, subobj); 2977 if (error != 0) { 2978 (void) printf("ERROR %u while trying to open " 2979 "subobj id %llu\n", 2980 error, (u_longlong_t)subobj); 2981 corruption_found = B_TRUE; 2982 continue; 2983 } 2984 dump_full_bpobj(&subbpo, "subobj", indent + 1); 2985 bpobj_close(&subbpo); 2986 } 2987 } else { 2988 if (bpo->bpo_havefreed) { 2989 (void) printf(" %*s: object %llu, %llu blkptrs, " 2990 "%llu freed, %s\n", 2991 indent * 8, name, 2992 (u_longlong_t)bpo->bpo_object, 2993 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs, 2994 (u_longlong_t)bpo->bpo_phys->bpo_num_freed, 2995 bytes); 2996 } else { 2997 (void) printf(" %*s: object %llu, %llu blkptrs, " 2998 "%s\n", 2999 indent * 8, name, 3000 (u_longlong_t)bpo->bpo_object, 3001 (u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs, 3002 bytes); 3003 } 3004 } 3005 3006 if (dump_opt['d'] < 5) 3007 return; 3008 3009 3010 if (indent == 0) { 3011 (void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL); 3012 (void) printf("\n"); 3013 } 3014 } 3015 3016 static int 3017 dump_bookmark(dsl_pool_t *dp, char *name, boolean_t print_redact, 3018 boolean_t print_list) 3019 { 3020 int err = 0; 3021 zfs_bookmark_phys_t prop; 3022 objset_t *mos = dp->dp_spa->spa_meta_objset; 3023 err = dsl_bookmark_lookup(dp, name, NULL, &prop); 3024 3025 if (err != 0) { 3026 return (err); 3027 } 3028 3029 (void) printf("\t#%s: ", strchr(name, '#') + 1); 3030 (void) printf("{guid: %llx creation_txg: %llu creation_time: " 3031 "%llu redaction_obj: %llu}\n", (u_longlong_t)prop.zbm_guid, 3032 (u_longlong_t)prop.zbm_creation_txg, 3033 (u_longlong_t)prop.zbm_creation_time, 3034 (u_longlong_t)prop.zbm_redaction_obj); 3035 3036 IMPLY(print_list, print_redact); 3037 if (!print_redact || prop.zbm_redaction_obj == 0) 3038 return (0); 3039 3040 redaction_list_t *rl; 3041 VERIFY0(dsl_redaction_list_hold_obj(dp, 3042 prop.zbm_redaction_obj, FTAG, &rl)); 3043 3044 redaction_list_phys_t *rlp = rl->rl_phys; 3045 (void) printf("\tRedacted:\n\t\tProgress: "); 3046 if (rlp->rlp_last_object != UINT64_MAX || 3047 rlp->rlp_last_blkid != UINT64_MAX) { 3048 (void) printf("%llu %llu (incomplete)\n", 3049 (u_longlong_t)rlp->rlp_last_object, 3050 (u_longlong_t)rlp->rlp_last_blkid); 3051 } else { 3052 (void) printf("complete\n"); 3053 } 3054 (void) printf("\t\tSnapshots: ["); 3055 for (unsigned int i = 0; i < rlp->rlp_num_snaps; i++) { 3056 if (i > 0) 3057 (void) printf(", "); 3058 (void) printf("%0llu", 3059 (u_longlong_t)rlp->rlp_snaps[i]); 3060 } 3061 (void) printf("]\n\t\tLength: %llu\n", 3062 (u_longlong_t)rlp->rlp_num_entries); 3063 3064 if (!print_list) { 3065 dsl_redaction_list_rele(rl, FTAG); 3066 return (0); 3067 } 3068 3069 if (rlp->rlp_num_entries == 0) { 3070 dsl_redaction_list_rele(rl, FTAG); 3071 (void) printf("\t\tRedaction List: []\n\n"); 3072 return (0); 3073 } 3074 3075 redact_block_phys_t *rbp_buf; 3076 uint64_t size; 3077 dmu_object_info_t doi; 3078 3079 VERIFY0(dmu_object_info(mos, prop.zbm_redaction_obj, &doi)); 3080 size = doi.doi_max_offset; 3081 rbp_buf = kmem_alloc(size, KM_SLEEP); 3082 3083 err = dmu_read(mos, prop.zbm_redaction_obj, 0, size, 3084 rbp_buf, 0); 3085 if (err != 0) { 3086 dsl_redaction_list_rele(rl, FTAG); 3087 kmem_free(rbp_buf, size); 3088 return (err); 3089 } 3090 3091 (void) printf("\t\tRedaction List: [{object: %llx, offset: " 3092 "%llx, blksz: %x, count: %llx}", 3093 (u_longlong_t)rbp_buf[0].rbp_object, 3094 (u_longlong_t)rbp_buf[0].rbp_blkid, 3095 (uint_t)(redact_block_get_size(&rbp_buf[0])), 3096 (u_longlong_t)redact_block_get_count(&rbp_buf[0])); 3097 3098 for (size_t i = 1; i < rlp->rlp_num_entries; i++) { 3099 (void) printf(",\n\t\t{object: %llx, offset: %llx, " 3100 "blksz: %x, count: %llx}", 3101 (u_longlong_t)rbp_buf[i].rbp_object, 3102 (u_longlong_t)rbp_buf[i].rbp_blkid, 3103 (uint_t)(redact_block_get_size(&rbp_buf[i])), 3104 (u_longlong_t)redact_block_get_count(&rbp_buf[i])); 3105 } 3106 dsl_redaction_list_rele(rl, FTAG); 3107 kmem_free(rbp_buf, size); 3108 (void) printf("]\n\n"); 3109 return (0); 3110 } 3111 3112 static void 3113 dump_bookmarks(objset_t *os, int verbosity) 3114 { 3115 zap_cursor_t zc; 3116 zap_attribute_t *attrp; 3117 dsl_dataset_t *ds = dmu_objset_ds(os); 3118 dsl_pool_t *dp = spa_get_dsl(os->os_spa); 3119 objset_t *mos = os->os_spa->spa_meta_objset; 3120 if (verbosity < 4) 3121 return; 3122 attrp = zap_attribute_alloc(); 3123 dsl_pool_config_enter(dp, FTAG); 3124 3125 for (zap_cursor_init(&zc, mos, ds->ds_bookmarks_obj); 3126 zap_cursor_retrieve(&zc, attrp) == 0; 3127 zap_cursor_advance(&zc)) { 3128 char osname[ZFS_MAX_DATASET_NAME_LEN]; 3129 char buf[ZFS_MAX_DATASET_NAME_LEN]; 3130 int len; 3131 dmu_objset_name(os, osname); 3132 len = snprintf(buf, sizeof (buf), "%s#%s", osname, 3133 attrp->za_name); 3134 VERIFY3S(len, <, ZFS_MAX_DATASET_NAME_LEN); 3135 (void) dump_bookmark(dp, buf, verbosity >= 5, verbosity >= 6); 3136 } 3137 zap_cursor_fini(&zc); 3138 dsl_pool_config_exit(dp, FTAG); 3139 zap_attribute_free(attrp); 3140 } 3141 3142 static void 3143 bpobj_count_refd(bpobj_t *bpo) 3144 { 3145 mos_obj_refd(bpo->bpo_object); 3146 3147 if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) { 3148 mos_obj_refd(bpo->bpo_phys->bpo_subobjs); 3149 for (uint64_t i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) { 3150 uint64_t subobj; 3151 bpobj_t subbpo; 3152 int error; 3153 VERIFY0(dmu_read(bpo->bpo_os, 3154 bpo->bpo_phys->bpo_subobjs, 3155 i * sizeof (subobj), sizeof (subobj), &subobj, 0)); 3156 error = bpobj_open(&subbpo, bpo->bpo_os, subobj); 3157 if (error != 0) { 3158 (void) printf("ERROR %u while trying to open " 3159 "subobj id %llu\n", 3160 error, (u_longlong_t)subobj); 3161 corruption_found = B_TRUE; 3162 continue; 3163 } 3164 bpobj_count_refd(&subbpo); 3165 bpobj_close(&subbpo); 3166 } 3167 } 3168 } 3169 3170 static int 3171 dsl_deadlist_entry_count_refd(void *arg, dsl_deadlist_entry_t *dle) 3172 { 3173 spa_t *spa = arg; 3174 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj; 3175 if (dle->dle_bpobj.bpo_object != empty_bpobj) 3176 bpobj_count_refd(&dle->dle_bpobj); 3177 return (0); 3178 } 3179 3180 static int 3181 dsl_deadlist_entry_dump(void *arg, dsl_deadlist_entry_t *dle) 3182 { 3183 ASSERT0P(arg); 3184 if (dump_opt['d'] >= 5) { 3185 char buf[128]; 3186 (void) snprintf(buf, sizeof (buf), 3187 "mintxg %llu -> obj %llu", 3188 (longlong_t)dle->dle_mintxg, 3189 (longlong_t)dle->dle_bpobj.bpo_object); 3190 3191 dump_full_bpobj(&dle->dle_bpobj, buf, 0); 3192 } else { 3193 (void) printf("mintxg %llu -> obj %llu\n", 3194 (longlong_t)dle->dle_mintxg, 3195 (longlong_t)dle->dle_bpobj.bpo_object); 3196 } 3197 return (0); 3198 } 3199 3200 static void 3201 dump_blkptr_list(dsl_deadlist_t *dl, const char *name) 3202 { 3203 char bytes[32]; 3204 char comp[32]; 3205 char uncomp[32]; 3206 char entries[32]; 3207 spa_t *spa = dmu_objset_spa(dl->dl_os); 3208 uint64_t empty_bpobj = spa->spa_dsl_pool->dp_empty_bpobj; 3209 3210 if (dl->dl_oldfmt) { 3211 if (dl->dl_bpobj.bpo_object != empty_bpobj) 3212 bpobj_count_refd(&dl->dl_bpobj); 3213 } else { 3214 mos_obj_refd(dl->dl_object); 3215 dsl_deadlist_iterate(dl, dsl_deadlist_entry_count_refd, spa); 3216 } 3217 3218 /* make sure nicenum has enough space */ 3219 _Static_assert(sizeof (bytes) >= NN_NUMBUF_SZ, "bytes truncated"); 3220 _Static_assert(sizeof (comp) >= NN_NUMBUF_SZ, "comp truncated"); 3221 _Static_assert(sizeof (uncomp) >= NN_NUMBUF_SZ, "uncomp truncated"); 3222 _Static_assert(sizeof (entries) >= NN_NUMBUF_SZ, "entries truncated"); 3223 3224 if (dump_opt['d'] < 3) 3225 return; 3226 3227 if (dl->dl_oldfmt) { 3228 dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0); 3229 return; 3230 } 3231 3232 zdb_nicenum(dl->dl_phys->dl_used, bytes, sizeof (bytes)); 3233 zdb_nicenum(dl->dl_phys->dl_comp, comp, sizeof (comp)); 3234 zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp, sizeof (uncomp)); 3235 zdb_nicenum(avl_numnodes(&dl->dl_tree), entries, sizeof (entries)); 3236 (void) printf("\n %s: %s (%s/%s comp), %s entries\n", 3237 name, bytes, comp, uncomp, entries); 3238 3239 if (dump_opt['d'] < 4) 3240 return; 3241 3242 (void) putchar('\n'); 3243 3244 dsl_deadlist_iterate(dl, dsl_deadlist_entry_dump, NULL); 3245 } 3246 3247 static int 3248 verify_dd_livelist(objset_t *os) 3249 { 3250 uint64_t ll_used, used, ll_comp, comp, ll_uncomp, uncomp; 3251 dsl_pool_t *dp = spa_get_dsl(os->os_spa); 3252 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 3253 3254 ASSERT(!dmu_objset_is_snapshot(os)); 3255 if (!dsl_deadlist_is_open(&dd->dd_livelist)) 3256 return (0); 3257 3258 /* Iterate through the livelist to check for duplicates */ 3259 dsl_deadlist_iterate(&dd->dd_livelist, sublivelist_verify_lightweight, 3260 NULL); 3261 3262 dsl_pool_config_enter(dp, FTAG); 3263 dsl_deadlist_space(&dd->dd_livelist, &ll_used, 3264 &ll_comp, &ll_uncomp); 3265 3266 dsl_dataset_t *origin_ds; 3267 ASSERT(dsl_pool_config_held(dp)); 3268 VERIFY0(dsl_dataset_hold_obj(dp, 3269 dsl_dir_phys(dd)->dd_origin_obj, FTAG, &origin_ds)); 3270 VERIFY0(dsl_dataset_space_written(origin_ds, os->os_dsl_dataset, 3271 &used, &comp, &uncomp)); 3272 dsl_dataset_rele(origin_ds, FTAG); 3273 dsl_pool_config_exit(dp, FTAG); 3274 /* 3275 * It's possible that the dataset's uncomp space is larger than the 3276 * livelist's because livelists do not track embedded block pointers 3277 */ 3278 if (used != ll_used || comp != ll_comp || uncomp < ll_uncomp) { 3279 char nice_used[32], nice_comp[32], nice_uncomp[32]; 3280 (void) printf("Discrepancy in space accounting:\n"); 3281 zdb_nicenum(used, nice_used, sizeof (nice_used)); 3282 zdb_nicenum(comp, nice_comp, sizeof (nice_comp)); 3283 zdb_nicenum(uncomp, nice_uncomp, sizeof (nice_uncomp)); 3284 (void) printf("dir: used %s, comp %s, uncomp %s\n", 3285 nice_used, nice_comp, nice_uncomp); 3286 zdb_nicenum(ll_used, nice_used, sizeof (nice_used)); 3287 zdb_nicenum(ll_comp, nice_comp, sizeof (nice_comp)); 3288 zdb_nicenum(ll_uncomp, nice_uncomp, sizeof (nice_uncomp)); 3289 (void) printf("livelist: used %s, comp %s, uncomp %s\n", 3290 nice_used, nice_comp, nice_uncomp); 3291 return (1); 3292 } 3293 return (0); 3294 } 3295 3296 static char *key_material = NULL; 3297 3298 static boolean_t 3299 zdb_derive_key(dsl_dir_t *dd, uint8_t *key_out) 3300 { 3301 uint64_t keyformat, salt, iters; 3302 int i; 3303 unsigned char c; 3304 FILE *f; 3305 3306 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj, 3307 zfs_prop_to_name(ZFS_PROP_KEYFORMAT), sizeof (uint64_t), 3308 1, &keyformat)); 3309 3310 switch (keyformat) { 3311 case ZFS_KEYFORMAT_HEX: 3312 for (i = 0; i < WRAPPING_KEY_LEN * 2; i += 2) { 3313 if (!isxdigit(key_material[i]) || 3314 !isxdigit(key_material[i+1])) 3315 return (B_FALSE); 3316 if (sscanf(&key_material[i], "%02hhx", &c) != 1) 3317 return (B_FALSE); 3318 key_out[i / 2] = c; 3319 } 3320 break; 3321 3322 case ZFS_KEYFORMAT_PASSPHRASE: 3323 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, 3324 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 3325 sizeof (uint64_t), 1, &salt)); 3326 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, 3327 dd->dd_crypto_obj, zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 3328 sizeof (uint64_t), 1, &iters)); 3329 3330 if (PKCS5_PBKDF2_HMAC_SHA1(key_material, strlen(key_material), 3331 ((uint8_t *)&salt), sizeof (uint64_t), iters, 3332 WRAPPING_KEY_LEN, key_out) != 1) 3333 return (B_FALSE); 3334 3335 break; 3336 3337 case ZFS_KEYFORMAT_RAW: 3338 if ((f = fopen(key_material, "r")) == NULL) 3339 return (B_FALSE); 3340 3341 if (fread(key_out, 1, WRAPPING_KEY_LEN, f) != 3342 WRAPPING_KEY_LEN) { 3343 (void) fclose(f); 3344 return (B_FALSE); 3345 } 3346 3347 /* Check the key length */ 3348 if (fgetc(f) != EOF) { 3349 (void) fclose(f); 3350 return (B_FALSE); 3351 } 3352 3353 (void) fclose(f); 3354 break; 3355 3356 default: 3357 fatal("no support for key format %u\n", 3358 (unsigned int) keyformat); 3359 } 3360 3361 return (B_TRUE); 3362 } 3363 3364 static char encroot[ZFS_MAX_DATASET_NAME_LEN]; 3365 static boolean_t key_loaded = B_FALSE; 3366 3367 static void 3368 zdb_load_key(objset_t *os) 3369 { 3370 dsl_pool_t *dp; 3371 dsl_dir_t *dd, *rdd; 3372 uint8_t key[WRAPPING_KEY_LEN]; 3373 uint64_t rddobj; 3374 int err; 3375 3376 dp = spa_get_dsl(os->os_spa); 3377 dd = os->os_dsl_dataset->ds_dir; 3378 3379 dsl_pool_config_enter(dp, FTAG); 3380 VERIFY0(zap_lookup(dd->dd_pool->dp_meta_objset, dd->dd_crypto_obj, 3381 DSL_CRYPTO_KEY_ROOT_DDOBJ, sizeof (uint64_t), 1, &rddobj)); 3382 VERIFY0(dsl_dir_hold_obj(dd->dd_pool, rddobj, NULL, FTAG, &rdd)); 3383 dsl_dir_name(rdd, encroot); 3384 dsl_dir_rele(rdd, FTAG); 3385 3386 if (!zdb_derive_key(dd, key)) 3387 fatal("couldn't derive encryption key"); 3388 3389 dsl_pool_config_exit(dp, FTAG); 3390 3391 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_UNAVAILABLE); 3392 3393 dsl_crypto_params_t *dcp; 3394 nvlist_t *crypto_args; 3395 3396 crypto_args = fnvlist_alloc(); 3397 fnvlist_add_uint8_array(crypto_args, "wkeydata", 3398 (uint8_t *)key, WRAPPING_KEY_LEN); 3399 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, 3400 NULL, crypto_args, &dcp)); 3401 err = spa_keystore_load_wkey(encroot, dcp, B_FALSE); 3402 3403 dsl_crypto_params_free(dcp, (err != 0)); 3404 fnvlist_free(crypto_args); 3405 3406 if (err != 0) 3407 fatal( 3408 "couldn't load encryption key for %s: %s", 3409 encroot, err == ZFS_ERR_CRYPTO_NOTSUP ? 3410 "crypto params not supported" : strerror(err)); 3411 3412 ASSERT3U(dsl_dataset_get_keystatus(dd), ==, ZFS_KEYSTATUS_AVAILABLE); 3413 3414 printf("Unlocked encryption root: %s\n", encroot); 3415 key_loaded = B_TRUE; 3416 } 3417 3418 static void 3419 zdb_unload_key(void) 3420 { 3421 if (!key_loaded) 3422 return; 3423 3424 VERIFY0(spa_keystore_unload_wkey(encroot)); 3425 key_loaded = B_FALSE; 3426 } 3427 3428 static avl_tree_t idx_tree; 3429 static avl_tree_t domain_tree; 3430 static boolean_t fuid_table_loaded; 3431 static objset_t *sa_os = NULL; 3432 static sa_attr_type_t *sa_attr_table = NULL; 3433 3434 static int 3435 open_objset(const char *path, const void *tag, objset_t **osp) 3436 { 3437 int err; 3438 uint64_t sa_attrs = 0; 3439 uint64_t version = 0; 3440 3441 VERIFY0P(sa_os); 3442 3443 /* 3444 * We can't own an objset if it's redacted. Therefore, we do this 3445 * dance: hold the objset, then acquire a long hold on its dataset, then 3446 * release the pool (which is held as part of holding the objset). 3447 */ 3448 3449 if (dump_opt['K']) { 3450 /* decryption requested, try to load keys */ 3451 err = dmu_objset_hold(path, tag, osp); 3452 if (err != 0) { 3453 (void) fprintf(stderr, "failed to hold dataset " 3454 "'%s': %s\n", 3455 path, strerror(err)); 3456 return (err); 3457 } 3458 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag); 3459 dsl_pool_rele(dmu_objset_pool(*osp), tag); 3460 3461 /* succeeds or dies */ 3462 zdb_load_key(*osp); 3463 3464 /* release it all */ 3465 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag); 3466 dsl_dataset_rele(dmu_objset_ds(*osp), tag); 3467 } 3468 3469 int ds_hold_flags = key_loaded ? DS_HOLD_FLAG_DECRYPT : 0; 3470 3471 err = dmu_objset_hold_flags(path, ds_hold_flags, tag, osp); 3472 if (err != 0) { 3473 (void) fprintf(stderr, "failed to hold dataset '%s': %s\n", 3474 path, strerror(err)); 3475 return (err); 3476 } 3477 dsl_dataset_long_hold(dmu_objset_ds(*osp), tag); 3478 dsl_pool_rele(dmu_objset_pool(*osp), tag); 3479 3480 if (dmu_objset_type(*osp) == DMU_OST_ZFS && 3481 (key_loaded || !(*osp)->os_encrypted)) { 3482 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR, 3483 8, 1, &version); 3484 if (version >= ZPL_VERSION_SA) { 3485 (void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 3486 8, 1, &sa_attrs); 3487 } 3488 err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END, 3489 &sa_attr_table); 3490 if (err != 0) { 3491 (void) fprintf(stderr, "sa_setup failed: %s\n", 3492 strerror(err)); 3493 dsl_dataset_long_rele(dmu_objset_ds(*osp), tag); 3494 dsl_dataset_rele_flags(dmu_objset_ds(*osp), 3495 ds_hold_flags, tag); 3496 *osp = NULL; 3497 } 3498 } 3499 sa_os = *osp; 3500 3501 return (err); 3502 } 3503 3504 static void 3505 close_objset(objset_t *os, const void *tag) 3506 { 3507 VERIFY3P(os, ==, sa_os); 3508 if (os->os_sa != NULL) 3509 sa_tear_down(os); 3510 dsl_dataset_long_rele(dmu_objset_ds(os), tag); 3511 dsl_dataset_rele_flags(dmu_objset_ds(os), 3512 key_loaded ? DS_HOLD_FLAG_DECRYPT : 0, tag); 3513 sa_attr_table = NULL; 3514 sa_os = NULL; 3515 3516 zdb_unload_key(); 3517 } 3518 3519 static void 3520 fuid_table_destroy(void) 3521 { 3522 if (fuid_table_loaded) { 3523 zfs_fuid_table_destroy(&idx_tree, &domain_tree); 3524 fuid_table_loaded = B_FALSE; 3525 } 3526 } 3527 3528 /* 3529 * Clean up DDT internal state. ddt_lookup() adds entries to ddt_tree, which on 3530 * a live pool are normally cleaned up during ddt_sync(). We can't do that (and 3531 * wouldn't want to anyway), but if we don't clean up the presence of stuff on 3532 * ddt_tree will trip asserts in ddt_table_free(). So, we clean up ourselves. 3533 * 3534 * Note that this is not a particularly efficient way to do this, but 3535 * ddt_remove() is the only public method that can do the work we need, and it 3536 * requires the right locks and etc to do the job. This is only ever called 3537 * during zdb shutdown so efficiency is not especially important. 3538 */ 3539 static void 3540 zdb_ddt_cleanup(spa_t *spa) 3541 { 3542 for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { 3543 ddt_t *ddt = spa->spa_ddt[c]; 3544 if (!ddt) 3545 continue; 3546 3547 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3548 ddt_enter(ddt); 3549 ddt_entry_t *dde = avl_first(&ddt->ddt_tree), *next; 3550 while (dde) { 3551 next = AVL_NEXT(&ddt->ddt_tree, dde); 3552 dde->dde_io = NULL; 3553 ddt_remove(ddt, dde); 3554 dde = next; 3555 } 3556 ddt_exit(ddt); 3557 spa_config_exit(spa, SCL_CONFIG, FTAG); 3558 } 3559 } 3560 3561 static void 3562 zdb_exit(int reason) 3563 { 3564 if (spa != NULL) 3565 zdb_ddt_cleanup(spa); 3566 3567 if (os != NULL) { 3568 close_objset(os, FTAG); 3569 } else if (spa != NULL) { 3570 spa_close(spa, FTAG); 3571 } 3572 3573 fuid_table_destroy(); 3574 3575 if (kernel_init_done) 3576 kernel_fini(); 3577 3578 exit(reason); 3579 } 3580 3581 /* 3582 * print uid or gid information. 3583 * For normal POSIX id just the id is printed in decimal format. 3584 * For CIFS files with FUID the fuid is printed in hex followed by 3585 * the domain-rid string. 3586 */ 3587 static void 3588 print_idstr(uint64_t id, const char *id_type) 3589 { 3590 if (FUID_INDEX(id)) { 3591 const char *domain = 3592 zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id)); 3593 (void) printf("\t%s %llx [%s-%d]\n", id_type, 3594 (u_longlong_t)id, domain, (int)FUID_RID(id)); 3595 } else { 3596 (void) printf("\t%s %llu\n", id_type, (u_longlong_t)id); 3597 } 3598 3599 } 3600 3601 static void 3602 dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid) 3603 { 3604 uint32_t uid_idx, gid_idx; 3605 3606 uid_idx = FUID_INDEX(uid); 3607 gid_idx = FUID_INDEX(gid); 3608 3609 /* Load domain table, if not already loaded */ 3610 if (!fuid_table_loaded && (uid_idx || gid_idx)) { 3611 uint64_t fuid_obj; 3612 3613 /* first find the fuid object. It lives in the master node */ 3614 VERIFY0(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 3615 8, 1, &fuid_obj)); 3616 zfs_fuid_avl_tree_create(&idx_tree, &domain_tree); 3617 (void) zfs_fuid_table_load(os, fuid_obj, 3618 &idx_tree, &domain_tree); 3619 fuid_table_loaded = B_TRUE; 3620 } 3621 3622 print_idstr(uid, "uid"); 3623 print_idstr(gid, "gid"); 3624 } 3625 3626 static void 3627 dump_znode_sa_xattr(sa_handle_t *hdl) 3628 { 3629 nvlist_t *sa_xattr; 3630 nvpair_t *elem = NULL; 3631 int sa_xattr_size = 0; 3632 int sa_xattr_entries = 0; 3633 int error; 3634 char *sa_xattr_packed; 3635 3636 error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size); 3637 if (error || sa_xattr_size == 0) 3638 return; 3639 3640 sa_xattr_packed = malloc(sa_xattr_size); 3641 if (sa_xattr_packed == NULL) 3642 return; 3643 3644 error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR], 3645 sa_xattr_packed, sa_xattr_size); 3646 if (error) { 3647 free(sa_xattr_packed); 3648 return; 3649 } 3650 3651 error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0); 3652 if (error) { 3653 free(sa_xattr_packed); 3654 return; 3655 } 3656 3657 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) 3658 sa_xattr_entries++; 3659 3660 (void) printf("\tSA xattrs: %d bytes, %d entries\n\n", 3661 sa_xattr_size, sa_xattr_entries); 3662 while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) { 3663 boolean_t can_print = !dump_opt['P']; 3664 uchar_t *value; 3665 uint_t cnt, idx; 3666 3667 (void) printf("\t\t%s = ", nvpair_name(elem)); 3668 nvpair_value_byte_array(elem, &value, &cnt); 3669 3670 for (idx = 0; idx < cnt; ++idx) { 3671 if (!isprint(value[idx])) { 3672 can_print = B_FALSE; 3673 break; 3674 } 3675 } 3676 3677 for (idx = 0; idx < cnt; ++idx) { 3678 if (can_print) 3679 (void) putchar(value[idx]); 3680 else 3681 (void) printf("\\%3.3o", value[idx]); 3682 } 3683 (void) putchar('\n'); 3684 } 3685 3686 nvlist_free(sa_xattr); 3687 free(sa_xattr_packed); 3688 } 3689 3690 static void 3691 dump_znode_symlink(sa_handle_t *hdl) 3692 { 3693 int sa_symlink_size = 0; 3694 char linktarget[MAXPATHLEN]; 3695 int error; 3696 3697 error = sa_size(hdl, sa_attr_table[ZPL_SYMLINK], &sa_symlink_size); 3698 if (error || sa_symlink_size == 0) { 3699 return; 3700 } 3701 if (sa_symlink_size >= sizeof (linktarget)) { 3702 (void) printf("symlink size %d is too large\n", 3703 sa_symlink_size); 3704 return; 3705 } 3706 linktarget[sa_symlink_size] = '\0'; 3707 if (sa_lookup(hdl, sa_attr_table[ZPL_SYMLINK], 3708 &linktarget, sa_symlink_size) == 0) 3709 (void) printf("\ttarget %s\n", linktarget); 3710 } 3711 3712 static void 3713 dump_znode(objset_t *os, uint64_t object, void *data, size_t size) 3714 { 3715 (void) data, (void) size; 3716 char path[MAXPATHLEN * 2]; /* allow for xattr and failure prefix */ 3717 sa_handle_t *hdl; 3718 uint64_t xattr, rdev, gen; 3719 uint64_t uid, gid, mode, fsize, parent, links; 3720 uint64_t pflags; 3721 uint64_t acctm[2], modtm[2], chgtm[2], crtm[2]; 3722 time_t z_crtime, z_atime, z_mtime, z_ctime; 3723 sa_bulk_attr_t bulk[12]; 3724 int idx = 0; 3725 int error; 3726 3727 VERIFY3P(os, ==, sa_os); 3728 if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) { 3729 (void) printf("Failed to get handle for SA znode\n"); 3730 return; 3731 } 3732 3733 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8); 3734 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8); 3735 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL, 3736 &links, 8); 3737 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8); 3738 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL, 3739 &mode, 8); 3740 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT], 3741 NULL, &parent, 8); 3742 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL, 3743 &fsize, 8); 3744 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL, 3745 acctm, 16); 3746 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL, 3747 modtm, 16); 3748 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL, 3749 crtm, 16); 3750 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL, 3751 chgtm, 16); 3752 SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL, 3753 &pflags, 8); 3754 3755 if (sa_bulk_lookup(hdl, bulk, idx)) { 3756 (void) sa_handle_destroy(hdl); 3757 return; 3758 } 3759 3760 z_crtime = (time_t)crtm[0]; 3761 z_atime = (time_t)acctm[0]; 3762 z_mtime = (time_t)modtm[0]; 3763 z_ctime = (time_t)chgtm[0]; 3764 3765 if (dump_opt['d'] > 4) { 3766 error = zfs_obj_to_path(os, object, path, sizeof (path)); 3767 if (error == ESTALE) { 3768 (void) snprintf(path, sizeof (path), "on delete queue"); 3769 } else if (error != 0) { 3770 leaked_objects++; 3771 (void) snprintf(path, sizeof (path), 3772 "path not found, possibly leaked"); 3773 } 3774 (void) printf("\tpath %s\n", path); 3775 } 3776 3777 if (S_ISLNK(mode)) 3778 dump_znode_symlink(hdl); 3779 dump_uidgid(os, uid, gid); 3780 (void) printf("\tatime %s", ctime(&z_atime)); 3781 (void) printf("\tmtime %s", ctime(&z_mtime)); 3782 (void) printf("\tctime %s", ctime(&z_ctime)); 3783 (void) printf("\tcrtime %s", ctime(&z_crtime)); 3784 (void) printf("\tgen %llu\n", (u_longlong_t)gen); 3785 (void) printf("\tmode %llo\n", (u_longlong_t)mode); 3786 (void) printf("\tsize %llu\n", (u_longlong_t)fsize); 3787 (void) printf("\tparent %llu\n", (u_longlong_t)parent); 3788 (void) printf("\tlinks %llu\n", (u_longlong_t)links); 3789 (void) printf("\tpflags %llx\n", (u_longlong_t)pflags); 3790 if (dmu_objset_projectquota_enabled(os) && (pflags & ZFS_PROJID)) { 3791 uint64_t projid; 3792 3793 if (sa_lookup(hdl, sa_attr_table[ZPL_PROJID], &projid, 3794 sizeof (uint64_t)) == 0) 3795 (void) printf("\tprojid %llu\n", (u_longlong_t)projid); 3796 } 3797 if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr, 3798 sizeof (uint64_t)) == 0) 3799 (void) printf("\txattr %llu\n", (u_longlong_t)xattr); 3800 if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev, 3801 sizeof (uint64_t)) == 0) 3802 (void) printf("\trdev 0x%016llx\n", (u_longlong_t)rdev); 3803 dump_znode_sa_xattr(hdl); 3804 sa_handle_destroy(hdl); 3805 } 3806 3807 static void 3808 dump_acl(objset_t *os, uint64_t object, void *data, size_t size) 3809 { 3810 (void) os, (void) object, (void) data, (void) size; 3811 } 3812 3813 static void 3814 dump_dmu_objset(objset_t *os, uint64_t object, void *data, size_t size) 3815 { 3816 (void) os, (void) object, (void) data, (void) size; 3817 } 3818 3819 static object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = { 3820 dump_none, /* unallocated */ 3821 dump_zap, /* object directory */ 3822 dump_uint64, /* object array */ 3823 dump_none, /* packed nvlist */ 3824 dump_packed_nvlist, /* packed nvlist size */ 3825 dump_none, /* bpobj */ 3826 dump_bpobj, /* bpobj header */ 3827 dump_none, /* SPA space map header */ 3828 dump_none, /* SPA space map */ 3829 dump_none, /* ZIL intent log */ 3830 dump_dnode, /* DMU dnode */ 3831 dump_dmu_objset, /* DMU objset */ 3832 dump_dsl_dir, /* DSL directory */ 3833 dump_zap, /* DSL directory child map */ 3834 dump_zap, /* DSL dataset snap map */ 3835 dump_zap, /* DSL props */ 3836 dump_dsl_dataset, /* DSL dataset */ 3837 dump_znode, /* ZFS znode */ 3838 dump_acl, /* ZFS V0 ACL */ 3839 dump_uint8, /* ZFS plain file */ 3840 dump_zpldir, /* ZFS directory */ 3841 dump_zap, /* ZFS master node */ 3842 dump_zap, /* ZFS delete queue */ 3843 dump_uint8, /* zvol object */ 3844 dump_zap, /* zvol prop */ 3845 dump_uint8, /* other uint8[] */ 3846 dump_uint64, /* other uint64[] */ 3847 dump_zap, /* other ZAP */ 3848 dump_zap, /* persistent error log */ 3849 dump_uint8, /* SPA history */ 3850 dump_history_offsets, /* SPA history offsets */ 3851 dump_zap, /* Pool properties */ 3852 dump_zap, /* DSL permissions */ 3853 dump_acl, /* ZFS ACL */ 3854 dump_uint8, /* ZFS SYSACL */ 3855 dump_none, /* FUID nvlist */ 3856 dump_packed_nvlist, /* FUID nvlist size */ 3857 dump_zap, /* DSL dataset next clones */ 3858 dump_zap, /* DSL scrub queue */ 3859 dump_zap, /* ZFS user/group/project used */ 3860 dump_zap, /* ZFS user/group/project quota */ 3861 dump_zap, /* snapshot refcount tags */ 3862 dump_ddt_zap, /* DDT ZAP object */ 3863 dump_zap, /* DDT statistics */ 3864 dump_znode, /* SA object */ 3865 dump_zap, /* SA Master Node */ 3866 dump_sa_attrs, /* SA attribute registration */ 3867 dump_sa_layouts, /* SA attribute layouts */ 3868 dump_zap, /* DSL scrub translations */ 3869 dump_none, /* fake dedup BP */ 3870 dump_zap, /* deadlist */ 3871 dump_none, /* deadlist hdr */ 3872 dump_zap, /* dsl clones */ 3873 dump_bpobj_subobjs, /* bpobj subobjs */ 3874 dump_unknown, /* Unknown type, must be last */ 3875 }; 3876 3877 static boolean_t 3878 match_object_type(dmu_object_type_t obj_type, uint64_t flags) 3879 { 3880 boolean_t match = B_TRUE; 3881 3882 switch (obj_type) { 3883 case DMU_OT_DIRECTORY_CONTENTS: 3884 if (!(flags & ZOR_FLAG_DIRECTORY)) 3885 match = B_FALSE; 3886 break; 3887 case DMU_OT_PLAIN_FILE_CONTENTS: 3888 if (!(flags & ZOR_FLAG_PLAIN_FILE)) 3889 match = B_FALSE; 3890 break; 3891 case DMU_OT_SPACE_MAP: 3892 if (!(flags & ZOR_FLAG_SPACE_MAP)) 3893 match = B_FALSE; 3894 break; 3895 default: 3896 if (strcmp(zdb_ot_name(obj_type), "zap") == 0) { 3897 if (!(flags & ZOR_FLAG_ZAP)) 3898 match = B_FALSE; 3899 break; 3900 } 3901 3902 /* 3903 * If all bits except some of the supported flags are 3904 * set, the user combined the all-types flag (A) with 3905 * a negated flag to exclude some types (e.g. A-f to 3906 * show all object types except plain files). 3907 */ 3908 if ((flags | ZOR_SUPPORTED_FLAGS) != ZOR_FLAG_ALL_TYPES) 3909 match = B_FALSE; 3910 3911 break; 3912 } 3913 3914 return (match); 3915 } 3916 3917 static void 3918 dump_object(objset_t *os, uint64_t object, int verbosity, 3919 boolean_t *print_header, uint64_t *dnode_slots_used, uint64_t flags) 3920 { 3921 dmu_buf_t *db = NULL; 3922 dmu_object_info_t doi; 3923 dnode_t *dn; 3924 boolean_t dnode_held = B_FALSE; 3925 void *bonus = NULL; 3926 size_t bsize = 0; 3927 char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32]; 3928 char bonus_size[32]; 3929 char aux[50]; 3930 int error; 3931 3932 /* make sure nicenum has enough space */ 3933 _Static_assert(sizeof (iblk) >= NN_NUMBUF_SZ, "iblk truncated"); 3934 _Static_assert(sizeof (dblk) >= NN_NUMBUF_SZ, "dblk truncated"); 3935 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, "lsize truncated"); 3936 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, "asize truncated"); 3937 _Static_assert(sizeof (bonus_size) >= NN_NUMBUF_SZ, 3938 "bonus_size truncated"); 3939 3940 if (*print_header) { 3941 (void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n", 3942 "Object", "lvl", "iblk", "dblk", "dsize", "dnsize", 3943 "lsize", "%full", "type"); 3944 *print_header = 0; 3945 } 3946 3947 if (object == 0) { 3948 dn = DMU_META_DNODE(os); 3949 dmu_object_info_from_dnode(dn, &doi); 3950 } else { 3951 /* 3952 * Encrypted datasets will have sensitive bonus buffers 3953 * encrypted. Therefore we cannot hold the bonus buffer and 3954 * must hold the dnode itself instead. 3955 */ 3956 error = dmu_object_info(os, object, &doi); 3957 if (error) 3958 fatal("dmu_object_info() failed, errno %u", error); 3959 3960 if (!key_loaded && os->os_encrypted && 3961 DMU_OT_IS_ENCRYPTED(doi.doi_bonus_type)) { 3962 error = dnode_hold(os, object, FTAG, &dn); 3963 if (error) 3964 fatal("dnode_hold() failed, errno %u", error); 3965 dnode_held = B_TRUE; 3966 } else { 3967 error = dmu_bonus_hold(os, object, FTAG, &db); 3968 if (error) 3969 fatal("dmu_bonus_hold(%llu) failed, errno %u", 3970 object, error); 3971 bonus = db->db_data; 3972 bsize = db->db_size; 3973 dn = DB_DNODE((dmu_buf_impl_t *)db); 3974 } 3975 } 3976 3977 /* 3978 * Default to showing all object types if no flags were specified. 3979 */ 3980 if (flags != 0 && flags != ZOR_FLAG_ALL_TYPES && 3981 !match_object_type(doi.doi_type, flags)) 3982 goto out; 3983 3984 if (dnode_slots_used) 3985 *dnode_slots_used = doi.doi_dnodesize / DNODE_MIN_SIZE; 3986 3987 zdb_nicenum(doi.doi_metadata_block_size, iblk, sizeof (iblk)); 3988 zdb_nicenum(doi.doi_data_block_size, dblk, sizeof (dblk)); 3989 zdb_nicenum(doi.doi_max_offset, lsize, sizeof (lsize)); 3990 zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize, sizeof (asize)); 3991 zdb_nicenum(doi.doi_bonus_size, bonus_size, sizeof (bonus_size)); 3992 zdb_nicenum(doi.doi_dnodesize, dnsize, sizeof (dnsize)); 3993 (void) snprintf(fill, sizeof (fill), "%6.2f", 100.0 * 3994 doi.doi_fill_count * doi.doi_data_block_size / (object == 0 ? 3995 DNODES_PER_BLOCK : 1) / doi.doi_max_offset); 3996 3997 aux[0] = '\0'; 3998 3999 if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) { 4000 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux), 4001 " (K=%s)", ZDB_CHECKSUM_NAME(doi.doi_checksum)); 4002 } 4003 4004 if (doi.doi_compress == ZIO_COMPRESS_INHERIT && 4005 ZIO_COMPRESS_HASLEVEL(os->os_compress) && verbosity >= 6) { 4006 const char *compname = NULL; 4007 if (zfs_prop_index_to_string(ZFS_PROP_COMPRESSION, 4008 ZIO_COMPRESS_RAW(os->os_compress, os->os_complevel), 4009 &compname) == 0) { 4010 (void) snprintf(aux + strlen(aux), 4011 sizeof (aux) - strlen(aux), " (Z=inherit=%s)", 4012 compname); 4013 } else { 4014 (void) snprintf(aux + strlen(aux), 4015 sizeof (aux) - strlen(aux), 4016 " (Z=inherit=%s-unknown)", 4017 ZDB_COMPRESS_NAME(os->os_compress)); 4018 } 4019 } else if (doi.doi_compress == ZIO_COMPRESS_INHERIT && verbosity >= 6) { 4020 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux), 4021 " (Z=inherit=%s)", ZDB_COMPRESS_NAME(os->os_compress)); 4022 } else if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) { 4023 (void) snprintf(aux + strlen(aux), sizeof (aux) - strlen(aux), 4024 " (Z=%s)", ZDB_COMPRESS_NAME(doi.doi_compress)); 4025 } 4026 4027 (void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n", 4028 (u_longlong_t)object, doi.doi_indirection, iblk, dblk, 4029 asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux); 4030 4031 if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) { 4032 (void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n", 4033 "", "", "", "", "", "", bonus_size, "bonus", 4034 zdb_ot_name(doi.doi_bonus_type)); 4035 } 4036 4037 if (verbosity >= 4) { 4038 (void) printf("\tdnode flags: %s%s%s%s\n", 4039 (dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ? 4040 "USED_BYTES " : "", 4041 (dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ? 4042 "USERUSED_ACCOUNTED " : "", 4043 (dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ? 4044 "USEROBJUSED_ACCOUNTED " : "", 4045 (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ? 4046 "SPILL_BLKPTR" : ""); 4047 (void) printf("\tdnode maxblkid: %llu\n", 4048 (longlong_t)dn->dn_phys->dn_maxblkid); 4049 4050 if (!dnode_held) { 4051 object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os, 4052 object, bonus, bsize); 4053 } else { 4054 (void) printf("\t\t(bonus encrypted)\n"); 4055 } 4056 4057 if (key_loaded || 4058 (!os->os_encrypted || !DMU_OT_IS_ENCRYPTED(doi.doi_type))) { 4059 object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object, 4060 NULL, 0); 4061 } else { 4062 (void) printf("\t\t(object encrypted)\n"); 4063 } 4064 4065 *print_header = B_TRUE; 4066 } 4067 4068 if (verbosity >= 5) { 4069 if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { 4070 char blkbuf[BP_SPRINTF_LEN]; 4071 snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), 4072 DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE); 4073 (void) printf("\nSpill block: %s\n", blkbuf); 4074 } 4075 dump_indirect(dn); 4076 } 4077 4078 if (verbosity >= 5) { 4079 /* 4080 * Report the list of segments that comprise the object. 4081 */ 4082 uint64_t start = 0; 4083 uint64_t end; 4084 uint64_t blkfill = 1; 4085 int minlvl = 1; 4086 4087 if (dn->dn_type == DMU_OT_DNODE) { 4088 minlvl = 0; 4089 blkfill = DNODES_PER_BLOCK; 4090 } 4091 4092 for (;;) { 4093 char segsize[32]; 4094 /* make sure nicenum has enough space */ 4095 _Static_assert(sizeof (segsize) >= NN_NUMBUF_SZ, 4096 "segsize truncated"); 4097 error = dnode_next_offset(dn, 4098 0, &start, minlvl, blkfill, 0); 4099 if (error) 4100 break; 4101 end = start; 4102 error = dnode_next_offset(dn, 4103 DNODE_FIND_HOLE, &end, minlvl, blkfill, 0); 4104 zdb_nicenum(end - start, segsize, sizeof (segsize)); 4105 (void) printf("\t\tsegment [%016llx, %016llx)" 4106 " size %5s\n", (u_longlong_t)start, 4107 (u_longlong_t)end, segsize); 4108 if (error) 4109 break; 4110 start = end; 4111 } 4112 } 4113 4114 out: 4115 if (db != NULL) 4116 dmu_buf_rele(db, FTAG); 4117 if (dnode_held) 4118 dnode_rele(dn, FTAG); 4119 } 4120 4121 static void 4122 count_dir_mos_objects(dsl_dir_t *dd) 4123 { 4124 mos_obj_refd(dd->dd_object); 4125 mos_obj_refd(dsl_dir_phys(dd)->dd_child_dir_zapobj); 4126 mos_obj_refd(dsl_dir_phys(dd)->dd_deleg_zapobj); 4127 mos_obj_refd(dsl_dir_phys(dd)->dd_props_zapobj); 4128 mos_obj_refd(dsl_dir_phys(dd)->dd_clones); 4129 4130 /* 4131 * The dd_crypto_obj can be referenced by multiple dsl_dir's. 4132 * Ignore the references after the first one. 4133 */ 4134 mos_obj_refd_multiple(dd->dd_crypto_obj); 4135 } 4136 4137 static void 4138 count_ds_mos_objects(dsl_dataset_t *ds) 4139 { 4140 mos_obj_refd(ds->ds_object); 4141 mos_obj_refd(dsl_dataset_phys(ds)->ds_next_clones_obj); 4142 mos_obj_refd(dsl_dataset_phys(ds)->ds_props_obj); 4143 mos_obj_refd(dsl_dataset_phys(ds)->ds_userrefs_obj); 4144 mos_obj_refd(dsl_dataset_phys(ds)->ds_snapnames_zapobj); 4145 mos_obj_refd(ds->ds_bookmarks_obj); 4146 4147 if (!dsl_dataset_is_snapshot(ds)) { 4148 count_dir_mos_objects(ds->ds_dir); 4149 } 4150 } 4151 4152 static const char *const objset_types[DMU_OST_NUMTYPES] = { 4153 "NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" }; 4154 4155 /* 4156 * Parse a string denoting a range of object IDs of the form 4157 * <start>[:<end>[:flags]], and store the results in zor. 4158 * Return 0 on success. On error, return 1 and update the msg 4159 * pointer to point to a descriptive error message. 4160 */ 4161 static int 4162 parse_object_range(char *range, zopt_object_range_t *zor, const char **msg) 4163 { 4164 uint64_t flags = 0; 4165 char *p, *s, *dup, *flagstr, *tmp = NULL; 4166 size_t len; 4167 int i; 4168 int rc = 0; 4169 4170 if (strchr(range, ':') == NULL) { 4171 zor->zor_obj_start = strtoull(range, &p, 0); 4172 if (*p != '\0') { 4173 *msg = "Invalid characters in object ID"; 4174 rc = 1; 4175 } 4176 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start); 4177 zor->zor_obj_end = zor->zor_obj_start; 4178 return (rc); 4179 } 4180 4181 if (strchr(range, ':') == range) { 4182 *msg = "Invalid leading colon"; 4183 rc = 1; 4184 return (rc); 4185 } 4186 4187 len = strlen(range); 4188 if (range[len - 1] == ':') { 4189 *msg = "Invalid trailing colon"; 4190 rc = 1; 4191 return (rc); 4192 } 4193 4194 dup = strdup(range); 4195 s = strtok_r(dup, ":", &tmp); 4196 zor->zor_obj_start = strtoull(s, &p, 0); 4197 4198 if (*p != '\0') { 4199 *msg = "Invalid characters in start object ID"; 4200 rc = 1; 4201 goto out; 4202 } 4203 4204 s = strtok_r(NULL, ":", &tmp); 4205 zor->zor_obj_end = strtoull(s, &p, 0); 4206 4207 if (*p != '\0') { 4208 *msg = "Invalid characters in end object ID"; 4209 rc = 1; 4210 goto out; 4211 } 4212 4213 if (zor->zor_obj_start > zor->zor_obj_end) { 4214 *msg = "Start object ID may not exceed end object ID"; 4215 rc = 1; 4216 goto out; 4217 } 4218 4219 s = strtok_r(NULL, ":", &tmp); 4220 if (s == NULL) { 4221 zor->zor_flags = ZOR_FLAG_ALL_TYPES; 4222 goto out; 4223 } else if (strtok_r(NULL, ":", &tmp) != NULL) { 4224 *msg = "Invalid colon-delimited field after flags"; 4225 rc = 1; 4226 goto out; 4227 } 4228 4229 flagstr = s; 4230 for (i = 0; flagstr[i]; i++) { 4231 int bit; 4232 boolean_t negation = (flagstr[i] == '-'); 4233 4234 if (negation) { 4235 i++; 4236 if (flagstr[i] == '\0') { 4237 *msg = "Invalid trailing negation operator"; 4238 rc = 1; 4239 goto out; 4240 } 4241 } 4242 bit = flagbits[(uchar_t)flagstr[i]]; 4243 if (bit == 0) { 4244 *msg = "Invalid flag"; 4245 rc = 1; 4246 goto out; 4247 } 4248 if (negation) 4249 flags &= ~bit; 4250 else 4251 flags |= bit; 4252 } 4253 zor->zor_flags = flags; 4254 4255 zor->zor_obj_start = ZDB_MAP_OBJECT_ID(zor->zor_obj_start); 4256 zor->zor_obj_end = ZDB_MAP_OBJECT_ID(zor->zor_obj_end); 4257 4258 out: 4259 free(dup); 4260 return (rc); 4261 } 4262 4263 static void 4264 dump_objset(objset_t *os) 4265 { 4266 dmu_objset_stats_t dds = { 0 }; 4267 uint64_t object, object_count; 4268 uint64_t refdbytes, usedobjs, scratch; 4269 char numbuf[32]; 4270 char blkbuf[BP_SPRINTF_LEN + 20]; 4271 char osname[ZFS_MAX_DATASET_NAME_LEN]; 4272 const char *type = "UNKNOWN"; 4273 int verbosity = dump_opt['d']; 4274 boolean_t print_header; 4275 unsigned i; 4276 int error; 4277 uint64_t total_slots_used = 0; 4278 uint64_t max_slot_used = 0; 4279 uint64_t dnode_slots; 4280 uint64_t obj_start; 4281 uint64_t obj_end; 4282 uint64_t flags; 4283 4284 /* make sure nicenum has enough space */ 4285 _Static_assert(sizeof (numbuf) >= NN_NUMBUF_SZ, "numbuf truncated"); 4286 4287 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 4288 dmu_objset_fast_stat(os, &dds); 4289 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 4290 4291 print_header = B_TRUE; 4292 4293 if (dds.dds_type < DMU_OST_NUMTYPES) 4294 type = objset_types[dds.dds_type]; 4295 4296 if (dds.dds_type == DMU_OST_META) { 4297 dds.dds_creation_txg = TXG_INITIAL; 4298 usedobjs = BP_GET_FILL(os->os_rootbp); 4299 refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)-> 4300 dd_used_bytes; 4301 } else { 4302 dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch); 4303 } 4304 4305 ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp)); 4306 4307 zdb_nicenum(refdbytes, numbuf, sizeof (numbuf)); 4308 4309 if (verbosity >= 4) { 4310 (void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp "); 4311 (void) snprintf_blkptr(blkbuf + strlen(blkbuf), 4312 sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp); 4313 } else { 4314 blkbuf[0] = '\0'; 4315 } 4316 4317 dmu_objset_name(os, osname); 4318 4319 (void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, " 4320 "%s, %llu objects%s%s\n", 4321 osname, type, (u_longlong_t)dmu_objset_id(os), 4322 (u_longlong_t)dds.dds_creation_txg, 4323 numbuf, (u_longlong_t)usedobjs, blkbuf, 4324 (dds.dds_inconsistent) ? " (inconsistent)" : ""); 4325 4326 for (i = 0; i < zopt_object_args; i++) { 4327 obj_start = zopt_object_ranges[i].zor_obj_start; 4328 obj_end = zopt_object_ranges[i].zor_obj_end; 4329 flags = zopt_object_ranges[i].zor_flags; 4330 4331 object = obj_start; 4332 if (object == 0 || obj_start == obj_end) 4333 dump_object(os, object, verbosity, &print_header, NULL, 4334 flags); 4335 else 4336 object--; 4337 4338 while ((dmu_object_next(os, &object, B_FALSE, 0) == 0) && 4339 object <= obj_end) { 4340 dump_object(os, object, verbosity, &print_header, NULL, 4341 flags); 4342 } 4343 } 4344 4345 if (zopt_object_args > 0) { 4346 (void) printf("\n"); 4347 return; 4348 } 4349 4350 if (dump_opt['i'] != 0 || verbosity >= 2) 4351 dump_intent_log(dmu_objset_zil(os)); 4352 4353 if (dmu_objset_ds(os) != NULL) { 4354 dsl_dataset_t *ds = dmu_objset_ds(os); 4355 dump_blkptr_list(&ds->ds_deadlist, "Deadlist"); 4356 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && 4357 !dmu_objset_is_snapshot(os)) { 4358 dump_blkptr_list(&ds->ds_dir->dd_livelist, "Livelist"); 4359 if (verify_dd_livelist(os) != 0) 4360 fatal("livelist is incorrect"); 4361 } 4362 4363 if (dsl_dataset_remap_deadlist_exists(ds)) { 4364 (void) printf("ds_remap_deadlist:\n"); 4365 dump_blkptr_list(&ds->ds_remap_deadlist, "Deadlist"); 4366 } 4367 count_ds_mos_objects(ds); 4368 } 4369 4370 if (dmu_objset_ds(os) != NULL) 4371 dump_bookmarks(os, verbosity); 4372 4373 if (verbosity < 2) 4374 return; 4375 4376 if (BP_IS_HOLE(os->os_rootbp)) 4377 return; 4378 4379 dump_object(os, 0, verbosity, &print_header, NULL, 0); 4380 object_count = 0; 4381 if (DMU_USERUSED_DNODE(os) != NULL && 4382 DMU_USERUSED_DNODE(os)->dn_type != 0) { 4383 dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header, 4384 NULL, 0); 4385 dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header, 4386 NULL, 0); 4387 } 4388 4389 if (DMU_PROJECTUSED_DNODE(os) != NULL && 4390 DMU_PROJECTUSED_DNODE(os)->dn_type != 0) 4391 dump_object(os, DMU_PROJECTUSED_OBJECT, verbosity, 4392 &print_header, NULL, 0); 4393 4394 object = 0; 4395 while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) { 4396 dump_object(os, object, verbosity, &print_header, &dnode_slots, 4397 0); 4398 object_count++; 4399 total_slots_used += dnode_slots; 4400 max_slot_used = object + dnode_slots - 1; 4401 } 4402 4403 (void) printf("\n"); 4404 4405 (void) printf(" Dnode slots:\n"); 4406 (void) printf("\tTotal used: %10llu\n", 4407 (u_longlong_t)total_slots_used); 4408 (void) printf("\tMax used: %10llu\n", 4409 (u_longlong_t)max_slot_used); 4410 (void) printf("\tPercent empty: %10lf\n", 4411 (double)(max_slot_used - total_slots_used)*100 / 4412 (double)max_slot_used); 4413 (void) printf("\n"); 4414 4415 if (error != ESRCH) { 4416 (void) fprintf(stderr, "dmu_object_next() = %d\n", error); 4417 abort(); 4418 } 4419 4420 ASSERT3U(object_count, ==, usedobjs); 4421 4422 if (leaked_objects != 0) { 4423 (void) printf("%d potentially leaked objects detected\n", 4424 leaked_objects); 4425 leaked_objects = 0; 4426 } 4427 } 4428 4429 static void 4430 dump_uberblock(uberblock_t *ub, const char *header, const char *footer) 4431 { 4432 time_t timestamp = ub->ub_timestamp; 4433 4434 (void) printf("%s", header ? header : ""); 4435 (void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic); 4436 (void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version); 4437 (void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg); 4438 (void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum); 4439 (void) printf("\ttimestamp = %llu UTC = %s", 4440 (u_longlong_t)ub->ub_timestamp, ctime(×tamp)); 4441 4442 char blkbuf[BP_SPRINTF_LEN]; 4443 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp); 4444 (void) printf("\tbp = %s\n", blkbuf); 4445 4446 (void) printf("\tmmp_magic = %016llx\n", 4447 (u_longlong_t)ub->ub_mmp_magic); 4448 if (MMP_VALID(ub)) { 4449 (void) printf("\tmmp_delay = %0llu\n", 4450 (u_longlong_t)ub->ub_mmp_delay); 4451 if (MMP_SEQ_VALID(ub)) 4452 (void) printf("\tmmp_seq = %u\n", 4453 (unsigned int) MMP_SEQ(ub)); 4454 if (MMP_FAIL_INT_VALID(ub)) 4455 (void) printf("\tmmp_fail = %u\n", 4456 (unsigned int) MMP_FAIL_INT(ub)); 4457 if (MMP_INTERVAL_VALID(ub)) 4458 (void) printf("\tmmp_write = %u\n", 4459 (unsigned int) MMP_INTERVAL(ub)); 4460 /* After MMP_* to make summarize_uberblock_mmp cleaner */ 4461 (void) printf("\tmmp_valid = %x\n", 4462 (unsigned int) ub->ub_mmp_config & 0xFF); 4463 } 4464 4465 if (dump_opt['u'] >= 4) { 4466 char blkbuf[BP_SPRINTF_LEN]; 4467 snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp); 4468 (void) printf("\trootbp = %s\n", blkbuf); 4469 } 4470 (void) printf("\tcheckpoint_txg = %llu\n", 4471 (u_longlong_t)ub->ub_checkpoint_txg); 4472 4473 (void) printf("\traidz_reflow state=%u off=%llu\n", 4474 (int)RRSS_GET_STATE(ub), 4475 (u_longlong_t)RRSS_GET_OFFSET(ub)); 4476 4477 (void) printf("%s", footer ? footer : ""); 4478 } 4479 4480 static void 4481 dump_config(spa_t *spa) 4482 { 4483 dmu_buf_t *db; 4484 size_t nvsize = 0; 4485 int error = 0; 4486 4487 4488 error = dmu_bonus_hold(spa->spa_meta_objset, 4489 spa->spa_config_object, FTAG, &db); 4490 4491 if (error == 0) { 4492 nvsize = *(uint64_t *)db->db_data; 4493 dmu_buf_rele(db, FTAG); 4494 4495 (void) printf("\nMOS Configuration:\n"); 4496 dump_packed_nvlist(spa->spa_meta_objset, 4497 spa->spa_config_object, (void *)&nvsize, 1); 4498 } else { 4499 (void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d", 4500 (u_longlong_t)spa->spa_config_object, error); 4501 } 4502 } 4503 4504 static void 4505 dump_cachefile(const char *cachefile) 4506 { 4507 int fd; 4508 struct stat64 statbuf; 4509 char *buf; 4510 nvlist_t *config; 4511 4512 if ((fd = open64(cachefile, O_RDONLY)) < 0) { 4513 (void) printf("cannot open '%s': %s\n", cachefile, 4514 strerror(errno)); 4515 zdb_exit(1); 4516 } 4517 4518 if (fstat64(fd, &statbuf) != 0) { 4519 (void) printf("failed to stat '%s': %s\n", cachefile, 4520 strerror(errno)); 4521 zdb_exit(1); 4522 } 4523 4524 if ((buf = malloc(statbuf.st_size)) == NULL) { 4525 (void) fprintf(stderr, "failed to allocate %llu bytes\n", 4526 (u_longlong_t)statbuf.st_size); 4527 zdb_exit(1); 4528 } 4529 4530 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) { 4531 (void) fprintf(stderr, "failed to read %llu bytes\n", 4532 (u_longlong_t)statbuf.st_size); 4533 zdb_exit(1); 4534 } 4535 4536 (void) close(fd); 4537 4538 if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) { 4539 (void) fprintf(stderr, "failed to unpack nvlist\n"); 4540 zdb_exit(1); 4541 } 4542 4543 free(buf); 4544 4545 dump_nvlist(config, 0); 4546 4547 nvlist_free(config); 4548 } 4549 4550 /* 4551 * ZFS label nvlist stats 4552 */ 4553 typedef struct zdb_nvl_stats { 4554 int zns_list_count; 4555 int zns_leaf_count; 4556 size_t zns_leaf_largest; 4557 size_t zns_leaf_total; 4558 nvlist_t *zns_string; 4559 nvlist_t *zns_uint64; 4560 nvlist_t *zns_boolean; 4561 } zdb_nvl_stats_t; 4562 4563 static void 4564 collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats) 4565 { 4566 nvlist_t *list, **array; 4567 nvpair_t *nvp = NULL; 4568 const char *name; 4569 uint_t i, items; 4570 4571 stats->zns_list_count++; 4572 4573 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) { 4574 name = nvpair_name(nvp); 4575 4576 switch (nvpair_type(nvp)) { 4577 case DATA_TYPE_STRING: 4578 fnvlist_add_string(stats->zns_string, name, 4579 fnvpair_value_string(nvp)); 4580 break; 4581 case DATA_TYPE_UINT64: 4582 fnvlist_add_uint64(stats->zns_uint64, name, 4583 fnvpair_value_uint64(nvp)); 4584 break; 4585 case DATA_TYPE_BOOLEAN: 4586 fnvlist_add_boolean(stats->zns_boolean, name); 4587 break; 4588 case DATA_TYPE_NVLIST: 4589 if (nvpair_value_nvlist(nvp, &list) == 0) 4590 collect_nvlist_stats(list, stats); 4591 break; 4592 case DATA_TYPE_NVLIST_ARRAY: 4593 if (nvpair_value_nvlist_array(nvp, &array, &items) != 0) 4594 break; 4595 4596 for (i = 0; i < items; i++) { 4597 collect_nvlist_stats(array[i], stats); 4598 4599 /* collect stats on leaf vdev */ 4600 if (strcmp(name, "children") == 0) { 4601 size_t size; 4602 4603 (void) nvlist_size(array[i], &size, 4604 NV_ENCODE_XDR); 4605 stats->zns_leaf_total += size; 4606 if (size > stats->zns_leaf_largest) 4607 stats->zns_leaf_largest = size; 4608 stats->zns_leaf_count++; 4609 } 4610 } 4611 break; 4612 default: 4613 (void) printf("skip type %d!\n", (int)nvpair_type(nvp)); 4614 } 4615 } 4616 } 4617 4618 static void 4619 dump_nvlist_stats(nvlist_t *nvl, size_t cap) 4620 { 4621 zdb_nvl_stats_t stats = { 0 }; 4622 size_t size, sum = 0, total; 4623 size_t noise; 4624 4625 /* requires nvlist with non-unique names for stat collection */ 4626 VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0)); 4627 VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0)); 4628 VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0)); 4629 VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR)); 4630 4631 (void) printf("\n\nZFS Label NVList Config Stats:\n"); 4632 4633 VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR)); 4634 (void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n", 4635 (int)total, (int)(cap - total), 100.0 * total / cap); 4636 4637 collect_nvlist_stats(nvl, &stats); 4638 4639 VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR)); 4640 size -= noise; 4641 sum += size; 4642 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:", 4643 (int)fnvlist_num_pairs(stats.zns_uint64), 4644 (int)size, 100.0 * size / total); 4645 4646 VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR)); 4647 size -= noise; 4648 sum += size; 4649 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:", 4650 (int)fnvlist_num_pairs(stats.zns_string), 4651 (int)size, 100.0 * size / total); 4652 4653 VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR)); 4654 size -= noise; 4655 sum += size; 4656 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:", 4657 (int)fnvlist_num_pairs(stats.zns_boolean), 4658 (int)size, 100.0 * size / total); 4659 4660 size = total - sum; /* treat remainder as nvlist overhead */ 4661 (void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:", 4662 stats.zns_list_count, (int)size, 100.0 * size / total); 4663 4664 if (stats.zns_leaf_count > 0) { 4665 size_t average = stats.zns_leaf_total / stats.zns_leaf_count; 4666 4667 (void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:", 4668 stats.zns_leaf_count, (int)average); 4669 (void) printf("%24d bytes largest\n", 4670 (int)stats.zns_leaf_largest); 4671 4672 if (dump_opt['l'] >= 3 && average > 0) 4673 (void) printf(" space for %d additional leaf vdevs\n", 4674 (int)((cap - total) / average)); 4675 } 4676 (void) printf("\n"); 4677 4678 nvlist_free(stats.zns_string); 4679 nvlist_free(stats.zns_uint64); 4680 nvlist_free(stats.zns_boolean); 4681 } 4682 4683 typedef struct cksum_record { 4684 zio_cksum_t cksum; 4685 boolean_t labels[VDEV_LABELS]; 4686 avl_node_t link; 4687 } cksum_record_t; 4688 4689 static int 4690 cksum_record_compare(const void *x1, const void *x2) 4691 { 4692 const cksum_record_t *l = (cksum_record_t *)x1; 4693 const cksum_record_t *r = (cksum_record_t *)x2; 4694 int arraysize = ARRAY_SIZE(l->cksum.zc_word); 4695 int difference = 0; 4696 4697 for (int i = 0; i < arraysize; i++) { 4698 difference = TREE_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]); 4699 if (difference) 4700 break; 4701 } 4702 4703 return (difference); 4704 } 4705 4706 static cksum_record_t * 4707 cksum_record_alloc(zio_cksum_t *cksum, int l) 4708 { 4709 cksum_record_t *rec; 4710 4711 rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL); 4712 rec->cksum = *cksum; 4713 rec->labels[l] = B_TRUE; 4714 4715 return (rec); 4716 } 4717 4718 static cksum_record_t * 4719 cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum) 4720 { 4721 cksum_record_t lookup = { .cksum = *cksum }; 4722 avl_index_t where; 4723 4724 return (avl_find(tree, &lookup, &where)); 4725 } 4726 4727 static cksum_record_t * 4728 cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l) 4729 { 4730 cksum_record_t *rec; 4731 4732 rec = cksum_record_lookup(tree, cksum); 4733 if (rec) { 4734 rec->labels[l] = B_TRUE; 4735 } else { 4736 rec = cksum_record_alloc(cksum, l); 4737 avl_add(tree, rec); 4738 } 4739 4740 return (rec); 4741 } 4742 4743 static int 4744 first_label(cksum_record_t *rec) 4745 { 4746 for (int i = 0; i < VDEV_LABELS; i++) 4747 if (rec->labels[i]) 4748 return (i); 4749 4750 return (-1); 4751 } 4752 4753 static void 4754 print_label_numbers(const char *prefix, const cksum_record_t *rec) 4755 { 4756 fputs(prefix, stdout); 4757 for (int i = 0; i < VDEV_LABELS; i++) 4758 if (rec->labels[i] == B_TRUE) 4759 printf("%d ", i); 4760 putchar('\n'); 4761 } 4762 4763 #define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT) 4764 4765 typedef struct zdb_label { 4766 vdev_label_t label; 4767 uint64_t label_offset; 4768 nvlist_t *config_nv; 4769 cksum_record_t *config; 4770 cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT]; 4771 boolean_t header_printed; 4772 boolean_t read_failed; 4773 boolean_t cksum_valid; 4774 } zdb_label_t; 4775 4776 static void 4777 print_label_header(zdb_label_t *label, int l) 4778 { 4779 4780 if (dump_opt['q']) 4781 return; 4782 4783 if (label->header_printed == B_TRUE) 4784 return; 4785 4786 (void) printf("------------------------------------\n"); 4787 (void) printf("LABEL %d %s\n", l, 4788 label->cksum_valid ? "" : "(Bad label cksum)"); 4789 (void) printf("------------------------------------\n"); 4790 4791 label->header_printed = B_TRUE; 4792 } 4793 4794 static void 4795 print_l2arc_header(void) 4796 { 4797 (void) printf("------------------------------------\n"); 4798 (void) printf("L2ARC device header\n"); 4799 (void) printf("------------------------------------\n"); 4800 } 4801 4802 static void 4803 print_l2arc_log_blocks(void) 4804 { 4805 (void) printf("------------------------------------\n"); 4806 (void) printf("L2ARC device log blocks\n"); 4807 (void) printf("------------------------------------\n"); 4808 } 4809 4810 static void 4811 dump_l2arc_log_entries(uint64_t log_entries, 4812 l2arc_log_ent_phys_t *le, uint64_t i) 4813 { 4814 for (int j = 0; j < log_entries; j++) { 4815 dva_t dva = le[j].le_dva; 4816 (void) printf("lb[%4llu]\tle[%4d]\tDVA asize: %llu, " 4817 "vdev: %llu, offset: %llu\n", 4818 (u_longlong_t)i, j + 1, 4819 (u_longlong_t)DVA_GET_ASIZE(&dva), 4820 (u_longlong_t)DVA_GET_VDEV(&dva), 4821 (u_longlong_t)DVA_GET_OFFSET(&dva)); 4822 (void) printf("|\t\t\t\tbirth: %llu\n", 4823 (u_longlong_t)le[j].le_birth); 4824 (void) printf("|\t\t\t\tlsize: %llu\n", 4825 (u_longlong_t)L2BLK_GET_LSIZE((&le[j])->le_prop)); 4826 (void) printf("|\t\t\t\tpsize: %llu\n", 4827 (u_longlong_t)L2BLK_GET_PSIZE((&le[j])->le_prop)); 4828 (void) printf("|\t\t\t\tcompr: %llu\n", 4829 (u_longlong_t)L2BLK_GET_COMPRESS((&le[j])->le_prop)); 4830 (void) printf("|\t\t\t\tcomplevel: %llu\n", 4831 (u_longlong_t)(&le[j])->le_complevel); 4832 (void) printf("|\t\t\t\ttype: %llu\n", 4833 (u_longlong_t)L2BLK_GET_TYPE((&le[j])->le_prop)); 4834 (void) printf("|\t\t\t\tprotected: %llu\n", 4835 (u_longlong_t)L2BLK_GET_PROTECTED((&le[j])->le_prop)); 4836 (void) printf("|\t\t\t\tprefetch: %llu\n", 4837 (u_longlong_t)L2BLK_GET_PREFETCH((&le[j])->le_prop)); 4838 (void) printf("|\t\t\t\taddress: %llu\n", 4839 (u_longlong_t)le[j].le_daddr); 4840 (void) printf("|\t\t\t\tARC state: %llu\n", 4841 (u_longlong_t)L2BLK_GET_STATE((&le[j])->le_prop)); 4842 (void) printf("|\n"); 4843 } 4844 (void) printf("\n"); 4845 } 4846 4847 static void 4848 dump_l2arc_log_blkptr(const l2arc_log_blkptr_t *lbps) 4849 { 4850 (void) printf("|\t\tdaddr: %llu\n", (u_longlong_t)lbps->lbp_daddr); 4851 (void) printf("|\t\tpayload_asize: %llu\n", 4852 (u_longlong_t)lbps->lbp_payload_asize); 4853 (void) printf("|\t\tpayload_start: %llu\n", 4854 (u_longlong_t)lbps->lbp_payload_start); 4855 (void) printf("|\t\tlsize: %llu\n", 4856 (u_longlong_t)L2BLK_GET_LSIZE(lbps->lbp_prop)); 4857 (void) printf("|\t\tasize: %llu\n", 4858 (u_longlong_t)L2BLK_GET_PSIZE(lbps->lbp_prop)); 4859 (void) printf("|\t\tcompralgo: %llu\n", 4860 (u_longlong_t)L2BLK_GET_COMPRESS(lbps->lbp_prop)); 4861 (void) printf("|\t\tcksumalgo: %llu\n", 4862 (u_longlong_t)L2BLK_GET_CHECKSUM(lbps->lbp_prop)); 4863 (void) printf("|\n\n"); 4864 } 4865 4866 static void 4867 dump_l2arc_log_blocks(int fd, const l2arc_dev_hdr_phys_t *l2dhdr, 4868 l2arc_dev_hdr_phys_t *rebuild) 4869 { 4870 l2arc_log_blk_phys_t this_lb; 4871 uint64_t asize; 4872 l2arc_log_blkptr_t lbps[2]; 4873 zio_cksum_t cksum; 4874 int failed = 0; 4875 l2arc_dev_t dev; 4876 4877 if (!dump_opt['q']) 4878 print_l2arc_log_blocks(); 4879 memcpy(lbps, l2dhdr->dh_start_lbps, sizeof (lbps)); 4880 4881 dev.l2ad_evict = l2dhdr->dh_evict; 4882 dev.l2ad_start = l2dhdr->dh_start; 4883 dev.l2ad_end = l2dhdr->dh_end; 4884 4885 if (l2dhdr->dh_start_lbps[0].lbp_daddr == 0) { 4886 /* no log blocks to read */ 4887 if (!dump_opt['q']) { 4888 (void) printf("No log blocks to read\n"); 4889 (void) printf("\n"); 4890 } 4891 return; 4892 } else { 4893 dev.l2ad_hand = lbps[0].lbp_daddr + 4894 L2BLK_GET_PSIZE((&lbps[0])->lbp_prop); 4895 } 4896 4897 dev.l2ad_first = !!(l2dhdr->dh_flags & L2ARC_DEV_HDR_EVICT_FIRST); 4898 4899 for (;;) { 4900 if (!l2arc_log_blkptr_valid(&dev, &lbps[0])) 4901 break; 4902 4903 /* L2BLK_GET_PSIZE returns aligned size for log blocks */ 4904 asize = L2BLK_GET_PSIZE((&lbps[0])->lbp_prop); 4905 if (pread64(fd, &this_lb, asize, lbps[0].lbp_daddr) != asize) { 4906 if (!dump_opt['q']) { 4907 (void) printf("Error while reading next log " 4908 "block\n\n"); 4909 } 4910 break; 4911 } 4912 4913 fletcher_4_native_varsize(&this_lb, asize, &cksum); 4914 if (!ZIO_CHECKSUM_EQUAL(cksum, lbps[0].lbp_cksum)) { 4915 failed++; 4916 if (!dump_opt['q']) { 4917 (void) printf("Invalid cksum\n"); 4918 dump_l2arc_log_blkptr(&lbps[0]); 4919 } 4920 break; 4921 } 4922 4923 switch (L2BLK_GET_COMPRESS((&lbps[0])->lbp_prop)) { 4924 case ZIO_COMPRESS_OFF: 4925 break; 4926 default: { 4927 abd_t *abd = abd_alloc_linear(asize, B_TRUE); 4928 abd_copy_from_buf_off(abd, &this_lb, 0, asize); 4929 abd_t dabd; 4930 abd_get_from_buf_struct(&dabd, &this_lb, 4931 sizeof (this_lb)); 4932 int err = zio_decompress_data(L2BLK_GET_COMPRESS( 4933 (&lbps[0])->lbp_prop), abd, &dabd, 4934 asize, sizeof (this_lb), NULL); 4935 abd_free(&dabd); 4936 abd_free(abd); 4937 if (err != 0) { 4938 (void) printf("L2ARC block decompression " 4939 "failed\n"); 4940 goto out; 4941 } 4942 break; 4943 } 4944 } 4945 4946 if (this_lb.lb_magic == BSWAP_64(L2ARC_LOG_BLK_MAGIC)) 4947 byteswap_uint64_array(&this_lb, sizeof (this_lb)); 4948 if (this_lb.lb_magic != L2ARC_LOG_BLK_MAGIC) { 4949 if (!dump_opt['q']) 4950 (void) printf("Invalid log block magic\n\n"); 4951 break; 4952 } 4953 4954 rebuild->dh_lb_count++; 4955 rebuild->dh_lb_asize += asize; 4956 if (dump_opt['l'] > 1 && !dump_opt['q']) { 4957 (void) printf("lb[%4llu]\tmagic: %llu\n", 4958 (u_longlong_t)rebuild->dh_lb_count, 4959 (u_longlong_t)this_lb.lb_magic); 4960 dump_l2arc_log_blkptr(&lbps[0]); 4961 } 4962 4963 if (dump_opt['l'] > 2 && !dump_opt['q']) 4964 dump_l2arc_log_entries(l2dhdr->dh_log_entries, 4965 this_lb.lb_entries, 4966 rebuild->dh_lb_count); 4967 4968 if (l2arc_range_check_overlap(lbps[1].lbp_payload_start, 4969 lbps[0].lbp_payload_start, dev.l2ad_evict) && 4970 !dev.l2ad_first) 4971 break; 4972 4973 lbps[0] = lbps[1]; 4974 lbps[1] = this_lb.lb_prev_lbp; 4975 } 4976 out: 4977 if (!dump_opt['q']) { 4978 (void) printf("log_blk_count:\t %llu with valid cksum\n", 4979 (u_longlong_t)rebuild->dh_lb_count); 4980 (void) printf("\t\t %d with invalid cksum\n", failed); 4981 (void) printf("log_blk_asize:\t %llu\n\n", 4982 (u_longlong_t)rebuild->dh_lb_asize); 4983 } 4984 } 4985 4986 static int 4987 dump_l2arc_header(int fd) 4988 { 4989 l2arc_dev_hdr_phys_t l2dhdr = {0}, rebuild = {0}; 4990 int error = B_FALSE; 4991 4992 if (pread64(fd, &l2dhdr, sizeof (l2dhdr), 4993 VDEV_LABEL_START_SIZE) != sizeof (l2dhdr)) { 4994 error = B_TRUE; 4995 } else { 4996 if (l2dhdr.dh_magic == BSWAP_64(L2ARC_DEV_HDR_MAGIC)) 4997 byteswap_uint64_array(&l2dhdr, sizeof (l2dhdr)); 4998 4999 if (l2dhdr.dh_magic != L2ARC_DEV_HDR_MAGIC) 5000 error = B_TRUE; 5001 } 5002 5003 if (error) { 5004 (void) printf("L2ARC device header not found\n\n"); 5005 /* Do not return an error here for backward compatibility */ 5006 return (0); 5007 } else if (!dump_opt['q']) { 5008 print_l2arc_header(); 5009 5010 (void) printf(" magic: %llu\n", 5011 (u_longlong_t)l2dhdr.dh_magic); 5012 (void) printf(" version: %llu\n", 5013 (u_longlong_t)l2dhdr.dh_version); 5014 (void) printf(" pool_guid: %llu\n", 5015 (u_longlong_t)l2dhdr.dh_spa_guid); 5016 (void) printf(" flags: %llu\n", 5017 (u_longlong_t)l2dhdr.dh_flags); 5018 (void) printf(" start_lbps[0]: %llu\n", 5019 (u_longlong_t) 5020 l2dhdr.dh_start_lbps[0].lbp_daddr); 5021 (void) printf(" start_lbps[1]: %llu\n", 5022 (u_longlong_t) 5023 l2dhdr.dh_start_lbps[1].lbp_daddr); 5024 (void) printf(" log_blk_ent: %llu\n", 5025 (u_longlong_t)l2dhdr.dh_log_entries); 5026 (void) printf(" start: %llu\n", 5027 (u_longlong_t)l2dhdr.dh_start); 5028 (void) printf(" end: %llu\n", 5029 (u_longlong_t)l2dhdr.dh_end); 5030 (void) printf(" evict: %llu\n", 5031 (u_longlong_t)l2dhdr.dh_evict); 5032 (void) printf(" lb_asize_refcount: %llu\n", 5033 (u_longlong_t)l2dhdr.dh_lb_asize); 5034 (void) printf(" lb_count_refcount: %llu\n", 5035 (u_longlong_t)l2dhdr.dh_lb_count); 5036 (void) printf(" trim_action_time: %llu\n", 5037 (u_longlong_t)l2dhdr.dh_trim_action_time); 5038 (void) printf(" trim_state: %llu\n\n", 5039 (u_longlong_t)l2dhdr.dh_trim_state); 5040 } 5041 5042 dump_l2arc_log_blocks(fd, &l2dhdr, &rebuild); 5043 /* 5044 * The total aligned size of log blocks and the number of log blocks 5045 * reported in the header of the device may be less than what zdb 5046 * reports by dump_l2arc_log_blocks() which emulates l2arc_rebuild(). 5047 * This happens because dump_l2arc_log_blocks() lacks the memory 5048 * pressure valve that l2arc_rebuild() has. Thus, if we are on a system 5049 * with low memory, l2arc_rebuild will exit prematurely and dh_lb_asize 5050 * and dh_lb_count will be lower to begin with than what exists on the 5051 * device. This is normal and zdb should not exit with an error. The 5052 * opposite case should never happen though, the values reported in the 5053 * header should never be higher than what dump_l2arc_log_blocks() and 5054 * l2arc_rebuild() report. If this happens there is a leak in the 5055 * accounting of log blocks. 5056 */ 5057 if (l2dhdr.dh_lb_asize > rebuild.dh_lb_asize || 5058 l2dhdr.dh_lb_count > rebuild.dh_lb_count) 5059 return (1); 5060 5061 return (0); 5062 } 5063 5064 static void 5065 dump_config_from_label(zdb_label_t *label, size_t buflen, int l) 5066 { 5067 if (dump_opt['q']) 5068 return; 5069 5070 if ((dump_opt['l'] < 3) && (first_label(label->config) != l)) 5071 return; 5072 5073 print_label_header(label, l); 5074 dump_nvlist(label->config_nv, 4); 5075 print_label_numbers(" labels = ", label->config); 5076 5077 if (dump_opt['l'] >= 2) 5078 dump_nvlist_stats(label->config_nv, buflen); 5079 } 5080 5081 #define ZDB_MAX_UB_HEADER_SIZE 32 5082 5083 static void 5084 dump_label_uberblocks(zdb_label_t *label, uint64_t ashift, int label_num) 5085 { 5086 5087 vdev_t vd; 5088 char header[ZDB_MAX_UB_HEADER_SIZE]; 5089 5090 vd.vdev_ashift = ashift; 5091 vd.vdev_top = &vd; 5092 5093 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) { 5094 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i); 5095 uberblock_t *ub = (void *)((char *)&label->label + uoff); 5096 cksum_record_t *rec = label->uberblocks[i]; 5097 5098 if (rec == NULL) { 5099 if (dump_opt['u'] >= 2) { 5100 print_label_header(label, label_num); 5101 (void) printf(" Uberblock[%d] invalid\n", i); 5102 } 5103 continue; 5104 } 5105 5106 if ((dump_opt['u'] < 3) && (first_label(rec) != label_num)) 5107 continue; 5108 5109 if ((dump_opt['u'] < 4) && 5110 (ub->ub_mmp_magic == MMP_MAGIC) && ub->ub_mmp_delay && 5111 (i >= VDEV_UBERBLOCK_COUNT(&vd) - MMP_BLOCKS_PER_LABEL)) 5112 continue; 5113 5114 print_label_header(label, label_num); 5115 (void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE, 5116 " Uberblock[%d]\n", i); 5117 dump_uberblock(ub, header, ""); 5118 print_label_numbers(" labels = ", rec); 5119 } 5120 } 5121 5122 static char curpath[PATH_MAX]; 5123 5124 /* 5125 * Iterate through the path components, recursively passing 5126 * current one's obj and remaining path until we find the obj 5127 * for the last one. 5128 */ 5129 static int 5130 dump_path_impl(objset_t *os, uint64_t obj, char *name, uint64_t *retobj) 5131 { 5132 int err; 5133 boolean_t header = B_TRUE; 5134 uint64_t child_obj; 5135 char *s; 5136 dmu_buf_t *db; 5137 dmu_object_info_t doi; 5138 5139 if ((s = strchr(name, '/')) != NULL) 5140 *s = '\0'; 5141 err = zap_lookup(os, obj, name, 8, 1, &child_obj); 5142 5143 (void) strlcat(curpath, name, sizeof (curpath)); 5144 5145 if (err != 0) { 5146 (void) fprintf(stderr, "failed to lookup %s: %s\n", 5147 curpath, strerror(err)); 5148 return (err); 5149 } 5150 5151 child_obj = ZFS_DIRENT_OBJ(child_obj); 5152 err = sa_buf_hold(os, child_obj, FTAG, &db); 5153 if (err != 0) { 5154 (void) fprintf(stderr, 5155 "failed to get SA dbuf for obj %llu: %s\n", 5156 (u_longlong_t)child_obj, strerror(err)); 5157 return (EINVAL); 5158 } 5159 dmu_object_info_from_db(db, &doi); 5160 sa_buf_rele(db, FTAG); 5161 5162 if (doi.doi_bonus_type != DMU_OT_SA && 5163 doi.doi_bonus_type != DMU_OT_ZNODE) { 5164 (void) fprintf(stderr, "invalid bonus type %d for obj %llu\n", 5165 doi.doi_bonus_type, (u_longlong_t)child_obj); 5166 return (EINVAL); 5167 } 5168 5169 if (dump_opt['v'] > 6) { 5170 (void) printf("obj=%llu %s type=%d bonustype=%d\n", 5171 (u_longlong_t)child_obj, curpath, doi.doi_type, 5172 doi.doi_bonus_type); 5173 } 5174 5175 (void) strlcat(curpath, "/", sizeof (curpath)); 5176 5177 switch (doi.doi_type) { 5178 case DMU_OT_DIRECTORY_CONTENTS: 5179 if (s != NULL && *(s + 1) != '\0') 5180 return (dump_path_impl(os, child_obj, s + 1, retobj)); 5181 zfs_fallthrough; 5182 case DMU_OT_PLAIN_FILE_CONTENTS: 5183 if (retobj != NULL) { 5184 *retobj = child_obj; 5185 } else { 5186 dump_object(os, child_obj, dump_opt['v'], &header, 5187 NULL, 0); 5188 } 5189 return (0); 5190 default: 5191 (void) fprintf(stderr, "object %llu has non-file/directory " 5192 "type %d\n", (u_longlong_t)obj, doi.doi_type); 5193 break; 5194 } 5195 5196 return (EINVAL); 5197 } 5198 5199 /* 5200 * Dump the blocks for the object specified by path inside the dataset. 5201 */ 5202 static int 5203 dump_path(char *ds, char *path, uint64_t *retobj) 5204 { 5205 int err; 5206 objset_t *os; 5207 uint64_t root_obj; 5208 5209 err = open_objset(ds, FTAG, &os); 5210 if (err != 0) 5211 return (err); 5212 5213 err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj); 5214 if (err != 0) { 5215 (void) fprintf(stderr, "can't lookup root znode: %s\n", 5216 strerror(err)); 5217 close_objset(os, FTAG); 5218 return (EINVAL); 5219 } 5220 5221 (void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds); 5222 5223 err = dump_path_impl(os, root_obj, path, retobj); 5224 5225 close_objset(os, FTAG); 5226 return (err); 5227 } 5228 5229 static int 5230 dump_backup_bytes(objset_t *os, void *buf, int len, void *arg) 5231 { 5232 const char *p = (const char *)buf; 5233 ssize_t nwritten; 5234 5235 (void) os; 5236 (void) arg; 5237 5238 /* Write the data out, handling short writes and signals. */ 5239 while ((nwritten = write(STDOUT_FILENO, p, len)) < len) { 5240 if (nwritten < 0) { 5241 if (errno == EINTR) 5242 continue; 5243 return (errno); 5244 } 5245 p += nwritten; 5246 len -= nwritten; 5247 } 5248 5249 return (0); 5250 } 5251 5252 static void 5253 dump_backup(const char *pool, uint64_t objset_id, const char *flagstr) 5254 { 5255 boolean_t embed = B_FALSE; 5256 boolean_t large_block = B_FALSE; 5257 boolean_t compress = B_FALSE; 5258 boolean_t raw = B_FALSE; 5259 5260 const char *c; 5261 for (c = flagstr; c != NULL && *c != '\0'; c++) { 5262 switch (*c) { 5263 case 'e': 5264 embed = B_TRUE; 5265 break; 5266 case 'L': 5267 large_block = B_TRUE; 5268 break; 5269 case 'c': 5270 compress = B_TRUE; 5271 break; 5272 case 'w': 5273 raw = B_TRUE; 5274 break; 5275 default: 5276 fprintf(stderr, "dump_backup: invalid flag " 5277 "'%c'\n", *c); 5278 return; 5279 } 5280 } 5281 5282 if (isatty(STDOUT_FILENO)) { 5283 fprintf(stderr, "dump_backup: stream cannot be written " 5284 "to a terminal\n"); 5285 return; 5286 } 5287 5288 offset_t off = 0; 5289 dmu_send_outparams_t out = { 5290 .dso_outfunc = dump_backup_bytes, 5291 .dso_dryrun = B_FALSE, 5292 }; 5293 5294 int err = dmu_send_obj(pool, objset_id, /* fromsnap */0, embed, 5295 large_block, compress, raw, /* saved */ B_FALSE, STDOUT_FILENO, 5296 &off, &out); 5297 if (err != 0) { 5298 fprintf(stderr, "dump_backup: dmu_send_obj: %s\n", 5299 strerror(err)); 5300 return; 5301 } 5302 } 5303 5304 static int 5305 zdb_copy_object(objset_t *os, uint64_t srcobj, char *destfile) 5306 { 5307 int err = 0; 5308 uint64_t size, readsize, oursize, offset; 5309 ssize_t writesize; 5310 sa_handle_t *hdl; 5311 5312 (void) printf("Copying object %" PRIu64 " to file %s\n", srcobj, 5313 destfile); 5314 5315 VERIFY3P(os, ==, sa_os); 5316 if ((err = sa_handle_get(os, srcobj, NULL, SA_HDL_PRIVATE, &hdl))) { 5317 (void) printf("Failed to get handle for SA znode\n"); 5318 return (err); 5319 } 5320 if ((err = sa_lookup(hdl, sa_attr_table[ZPL_SIZE], &size, 8))) { 5321 (void) sa_handle_destroy(hdl); 5322 return (err); 5323 } 5324 (void) sa_handle_destroy(hdl); 5325 5326 (void) printf("Object %" PRIu64 " is %" PRIu64 " bytes\n", srcobj, 5327 size); 5328 if (size == 0) { 5329 return (EINVAL); 5330 } 5331 5332 int fd = open(destfile, O_WRONLY | O_CREAT | O_TRUNC, 0644); 5333 if (fd == -1) 5334 return (errno); 5335 /* 5336 * We cap the size at 1 mebibyte here to prevent 5337 * allocation failures and nigh-infinite printing if the 5338 * object is extremely large. 5339 */ 5340 oursize = MIN(size, 1 << 20); 5341 offset = 0; 5342 char *buf = kmem_alloc(oursize, KM_NOSLEEP); 5343 if (buf == NULL) { 5344 (void) close(fd); 5345 return (ENOMEM); 5346 } 5347 5348 while (offset < size) { 5349 readsize = MIN(size - offset, 1 << 20); 5350 err = dmu_read(os, srcobj, offset, readsize, buf, 0); 5351 if (err != 0) { 5352 (void) printf("got error %u from dmu_read\n", err); 5353 kmem_free(buf, oursize); 5354 (void) close(fd); 5355 return (err); 5356 } 5357 if (dump_opt['v'] > 3) { 5358 (void) printf("Read offset=%" PRIu64 " size=%" PRIu64 5359 " error=%d\n", offset, readsize, err); 5360 } 5361 5362 writesize = write(fd, buf, readsize); 5363 if (writesize < 0) { 5364 err = errno; 5365 break; 5366 } else if (writesize != readsize) { 5367 /* Incomplete write */ 5368 (void) fprintf(stderr, "Short write, only wrote %llu of" 5369 " %" PRIu64 " bytes, exiting...\n", 5370 (u_longlong_t)writesize, readsize); 5371 break; 5372 } 5373 5374 offset += readsize; 5375 } 5376 5377 (void) close(fd); 5378 5379 if (buf != NULL) 5380 kmem_free(buf, oursize); 5381 5382 return (err); 5383 } 5384 5385 static boolean_t 5386 label_cksum_valid(vdev_label_t *label, uint64_t offset) 5387 { 5388 zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL]; 5389 zio_cksum_t expected_cksum; 5390 zio_cksum_t actual_cksum; 5391 zio_cksum_t verifier; 5392 zio_eck_t *eck; 5393 int byteswap; 5394 5395 void *data = (char *)label + offsetof(vdev_label_t, vl_vdev_phys); 5396 eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1; 5397 5398 offset += offsetof(vdev_label_t, vl_vdev_phys); 5399 ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0); 5400 5401 byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC)); 5402 if (byteswap) 5403 byteswap_uint64_array(&verifier, sizeof (zio_cksum_t)); 5404 5405 expected_cksum = eck->zec_cksum; 5406 eck->zec_cksum = verifier; 5407 5408 abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE); 5409 ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum); 5410 abd_free(abd); 5411 5412 if (byteswap) 5413 byteswap_uint64_array(&expected_cksum, sizeof (zio_cksum_t)); 5414 5415 if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum)) 5416 return (B_TRUE); 5417 5418 return (B_FALSE); 5419 } 5420 5421 static int 5422 dump_label(const char *dev) 5423 { 5424 char path[MAXPATHLEN]; 5425 zdb_label_t labels[VDEV_LABELS] = {{{{0}}}}; 5426 uint64_t psize, ashift, l2cache; 5427 struct stat64 statbuf; 5428 boolean_t config_found = B_FALSE; 5429 boolean_t error = B_FALSE; 5430 boolean_t read_l2arc_header = B_FALSE; 5431 avl_tree_t config_tree; 5432 avl_tree_t uberblock_tree; 5433 void *node, *cookie; 5434 int fd; 5435 5436 /* 5437 * Check if we were given absolute path and use it as is. 5438 * Otherwise if the provided vdev name doesn't point to a file, 5439 * try prepending expected disk paths and partition numbers. 5440 */ 5441 (void) strlcpy(path, dev, sizeof (path)); 5442 if (dev[0] != '/' && stat64(path, &statbuf) != 0) { 5443 int error; 5444 5445 error = zfs_resolve_shortname(dev, path, MAXPATHLEN); 5446 if (error == 0 && zfs_dev_is_whole_disk(path)) { 5447 if (zfs_append_partition(path, MAXPATHLEN) == -1) 5448 error = ENOENT; 5449 } 5450 5451 if (error || (stat64(path, &statbuf) != 0)) { 5452 (void) printf("failed to find device %s, try " 5453 "specifying absolute path instead\n", dev); 5454 return (1); 5455 } 5456 } 5457 5458 if ((fd = open64(path, O_RDONLY)) < 0) { 5459 (void) printf("cannot open '%s': %s\n", path, strerror(errno)); 5460 zdb_exit(1); 5461 } 5462 5463 if (fstat64_blk(fd, &statbuf) != 0) { 5464 (void) printf("failed to stat '%s': %s\n", path, 5465 strerror(errno)); 5466 (void) close(fd); 5467 zdb_exit(1); 5468 } 5469 5470 if (S_ISBLK(statbuf.st_mode) && zfs_dev_flush(fd) != 0) 5471 (void) printf("failed to invalidate cache '%s' : %s\n", path, 5472 strerror(errno)); 5473 5474 avl_create(&config_tree, cksum_record_compare, 5475 sizeof (cksum_record_t), offsetof(cksum_record_t, link)); 5476 avl_create(&uberblock_tree, cksum_record_compare, 5477 sizeof (cksum_record_t), offsetof(cksum_record_t, link)); 5478 5479 psize = statbuf.st_size; 5480 psize = P2ALIGN_TYPED(psize, sizeof (vdev_label_t), uint64_t); 5481 ashift = SPA_MINBLOCKSHIFT; 5482 5483 /* 5484 * 1. Read the label from disk 5485 * 2. Verify label cksum 5486 * 3. Unpack the configuration and insert in config tree. 5487 * 4. Traverse all uberblocks and insert in uberblock tree. 5488 */ 5489 for (int l = 0; l < VDEV_LABELS; l++) { 5490 zdb_label_t *label = &labels[l]; 5491 char *buf = label->label.vl_vdev_phys.vp_nvlist; 5492 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist); 5493 nvlist_t *config; 5494 cksum_record_t *rec; 5495 zio_cksum_t cksum; 5496 vdev_t vd; 5497 5498 label->label_offset = vdev_label_offset(psize, l, 0); 5499 5500 if (pread64(fd, &label->label, sizeof (label->label), 5501 label->label_offset) != sizeof (label->label)) { 5502 if (!dump_opt['q']) 5503 (void) printf("failed to read label %d\n", l); 5504 label->read_failed = B_TRUE; 5505 error = B_TRUE; 5506 continue; 5507 } 5508 5509 label->read_failed = B_FALSE; 5510 label->cksum_valid = label_cksum_valid(&label->label, 5511 label->label_offset); 5512 5513 if (nvlist_unpack(buf, buflen, &config, 0) == 0) { 5514 nvlist_t *vdev_tree = NULL; 5515 size_t size; 5516 5517 if ((nvlist_lookup_nvlist(config, 5518 ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) || 5519 (nvlist_lookup_uint64(vdev_tree, 5520 ZPOOL_CONFIG_ASHIFT, &ashift) != 0)) 5521 ashift = SPA_MINBLOCKSHIFT; 5522 5523 if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0) 5524 size = buflen; 5525 5526 /* If the device is a cache device read the header. */ 5527 if (!read_l2arc_header) { 5528 if (nvlist_lookup_uint64(config, 5529 ZPOOL_CONFIG_POOL_STATE, &l2cache) == 0 && 5530 l2cache == POOL_STATE_L2CACHE) { 5531 read_l2arc_header = B_TRUE; 5532 } 5533 } 5534 5535 fletcher_4_native_varsize(buf, size, &cksum); 5536 rec = cksum_record_insert(&config_tree, &cksum, l); 5537 5538 label->config = rec; 5539 label->config_nv = config; 5540 config_found = B_TRUE; 5541 } else { 5542 error = B_TRUE; 5543 } 5544 5545 vd.vdev_ashift = ashift; 5546 vd.vdev_top = &vd; 5547 5548 for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) { 5549 uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i); 5550 uberblock_t *ub = (void *)((char *)label + uoff); 5551 5552 if (uberblock_verify(ub)) 5553 continue; 5554 5555 fletcher_4_native_varsize(ub, sizeof (*ub), &cksum); 5556 rec = cksum_record_insert(&uberblock_tree, &cksum, l); 5557 5558 label->uberblocks[i] = rec; 5559 } 5560 } 5561 5562 /* 5563 * Dump the label and uberblocks. 5564 */ 5565 for (int l = 0; l < VDEV_LABELS; l++) { 5566 zdb_label_t *label = &labels[l]; 5567 size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist); 5568 5569 if (label->read_failed == B_TRUE) 5570 continue; 5571 5572 if (label->config_nv) { 5573 dump_config_from_label(label, buflen, l); 5574 } else { 5575 if (!dump_opt['q']) 5576 (void) printf("failed to unpack label %d\n", l); 5577 } 5578 5579 if (dump_opt['u']) 5580 dump_label_uberblocks(label, ashift, l); 5581 5582 nvlist_free(label->config_nv); 5583 } 5584 5585 /* 5586 * Dump the L2ARC header, if existent. 5587 */ 5588 if (read_l2arc_header) 5589 error |= dump_l2arc_header(fd); 5590 5591 cookie = NULL; 5592 while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL) 5593 umem_free(node, sizeof (cksum_record_t)); 5594 5595 cookie = NULL; 5596 while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL) 5597 umem_free(node, sizeof (cksum_record_t)); 5598 5599 avl_destroy(&config_tree); 5600 avl_destroy(&uberblock_tree); 5601 5602 (void) close(fd); 5603 5604 return (config_found == B_FALSE ? 2 : 5605 (error == B_TRUE ? 1 : 0)); 5606 } 5607 5608 static uint64_t dataset_feature_count[SPA_FEATURES]; 5609 static uint64_t global_feature_count[SPA_FEATURES]; 5610 static uint64_t remap_deadlist_count = 0; 5611 5612 static int 5613 dump_one_objset(const char *dsname, void *arg) 5614 { 5615 (void) arg; 5616 int error; 5617 objset_t *os; 5618 spa_feature_t f; 5619 5620 error = open_objset(dsname, FTAG, &os); 5621 if (error != 0) 5622 return (0); 5623 5624 for (f = 0; f < SPA_FEATURES; f++) { 5625 if (!dsl_dataset_feature_is_active(dmu_objset_ds(os), f)) 5626 continue; 5627 ASSERT(spa_feature_table[f].fi_flags & 5628 ZFEATURE_FLAG_PER_DATASET); 5629 dataset_feature_count[f]++; 5630 } 5631 5632 if (dsl_dataset_remap_deadlist_exists(dmu_objset_ds(os))) { 5633 remap_deadlist_count++; 5634 } 5635 5636 for (dsl_bookmark_node_t *dbn = 5637 avl_first(&dmu_objset_ds(os)->ds_bookmarks); dbn != NULL; 5638 dbn = AVL_NEXT(&dmu_objset_ds(os)->ds_bookmarks, dbn)) { 5639 mos_obj_refd(dbn->dbn_phys.zbm_redaction_obj); 5640 if (dbn->dbn_phys.zbm_redaction_obj != 0) { 5641 global_feature_count[ 5642 SPA_FEATURE_REDACTION_BOOKMARKS]++; 5643 objset_t *mos = os->os_spa->spa_meta_objset; 5644 dnode_t *rl; 5645 VERIFY0(dnode_hold(mos, 5646 dbn->dbn_phys.zbm_redaction_obj, FTAG, &rl)); 5647 if (rl->dn_have_spill) { 5648 global_feature_count[ 5649 SPA_FEATURE_REDACTION_LIST_SPILL]++; 5650 } 5651 } 5652 if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN) 5653 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN]++; 5654 } 5655 5656 if (dsl_deadlist_is_open(&dmu_objset_ds(os)->ds_dir->dd_livelist) && 5657 !dmu_objset_is_snapshot(os)) { 5658 global_feature_count[SPA_FEATURE_LIVELIST]++; 5659 } 5660 5661 dump_objset(os); 5662 close_objset(os, FTAG); 5663 fuid_table_destroy(); 5664 return (0); 5665 } 5666 5667 /* 5668 * Block statistics. 5669 */ 5670 #define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2) 5671 typedef struct zdb_blkstats { 5672 uint64_t zb_asize; 5673 uint64_t zb_lsize; 5674 uint64_t zb_psize; 5675 uint64_t zb_count; 5676 uint64_t zb_gangs; 5677 uint64_t zb_ditto_samevdev; 5678 uint64_t zb_ditto_same_ms; 5679 uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE]; 5680 } zdb_blkstats_t; 5681 5682 /* 5683 * Extended object types to report deferred frees and dedup auto-ditto blocks. 5684 */ 5685 #define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0) 5686 #define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1) 5687 #define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2) 5688 #define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3) 5689 5690 static const char *zdb_ot_extname[] = { 5691 "deferred free", 5692 "dedup ditto", 5693 "other", 5694 "Total", 5695 }; 5696 5697 #define ZB_TOTAL DN_MAX_LEVELS 5698 #define SPA_MAX_FOR_16M (SPA_MAXBLOCKSHIFT+1) 5699 5700 typedef struct zdb_brt_entry { 5701 dva_t zbre_dva; 5702 uint64_t zbre_refcount; 5703 avl_node_t zbre_node; 5704 } zdb_brt_entry_t; 5705 5706 typedef struct zdb_cb { 5707 zdb_blkstats_t zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1]; 5708 uint64_t zcb_removing_size; 5709 uint64_t zcb_checkpoint_size; 5710 uint64_t zcb_dedup_asize; 5711 uint64_t zcb_dedup_blocks; 5712 uint64_t zcb_clone_asize; 5713 uint64_t zcb_clone_blocks; 5714 uint64_t zcb_psize_count[SPA_MAX_FOR_16M]; 5715 uint64_t zcb_lsize_count[SPA_MAX_FOR_16M]; 5716 uint64_t zcb_asize_count[SPA_MAX_FOR_16M]; 5717 uint64_t zcb_psize_len[SPA_MAX_FOR_16M]; 5718 uint64_t zcb_lsize_len[SPA_MAX_FOR_16M]; 5719 uint64_t zcb_asize_len[SPA_MAX_FOR_16M]; 5720 uint64_t zcb_psize_total; 5721 uint64_t zcb_lsize_total; 5722 uint64_t zcb_asize_total; 5723 uint64_t zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES]; 5724 uint64_t zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES] 5725 [BPE_PAYLOAD_SIZE + 1]; 5726 uint64_t zcb_start; 5727 hrtime_t zcb_lastprint; 5728 uint64_t zcb_totalasize; 5729 uint64_t zcb_errors[256]; 5730 int zcb_readfails; 5731 int zcb_haderrors; 5732 spa_t *zcb_spa; 5733 uint32_t **zcb_vd_obsolete_counts; 5734 avl_tree_t zcb_brt; 5735 boolean_t zcb_brt_is_active; 5736 } zdb_cb_t; 5737 5738 /* test if two DVA offsets from same vdev are within the same metaslab */ 5739 static boolean_t 5740 same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2) 5741 { 5742 vdev_t *vd = vdev_lookup_top(spa, vdev); 5743 uint64_t ms_shift = vd->vdev_ms_shift; 5744 5745 return ((off1 >> ms_shift) == (off2 >> ms_shift)); 5746 } 5747 5748 /* 5749 * Used to simplify reporting of the histogram data. 5750 */ 5751 typedef struct one_histo { 5752 const char *name; 5753 uint64_t *count; 5754 uint64_t *len; 5755 uint64_t cumulative; 5756 } one_histo_t; 5757 5758 /* 5759 * The number of separate histograms processed for psize, lsize and asize. 5760 */ 5761 #define NUM_HISTO 3 5762 5763 /* 5764 * This routine will create a fixed column size output of three different 5765 * histograms showing by blocksize of 512 - 2^ SPA_MAX_FOR_16M 5766 * the count, length and cumulative length of the psize, lsize and 5767 * asize blocks. 5768 * 5769 * All three types of blocks are listed on a single line 5770 * 5771 * By default the table is printed in nicenumber format (e.g. 123K) but 5772 * if the '-P' parameter is specified then the full raw number (parseable) 5773 * is printed out. 5774 */ 5775 static void 5776 dump_size_histograms(zdb_cb_t *zcb) 5777 { 5778 /* 5779 * A temporary buffer that allows us to convert a number into 5780 * a string using zdb_nicenumber to allow either raw or human 5781 * readable numbers to be output. 5782 */ 5783 char numbuf[32]; 5784 5785 /* 5786 * Define titles which are used in the headers of the tables 5787 * printed by this routine. 5788 */ 5789 const char blocksize_title1[] = "block"; 5790 const char blocksize_title2[] = "size"; 5791 const char count_title[] = "Count"; 5792 const char length_title[] = "Size"; 5793 const char cumulative_title[] = "Cum."; 5794 5795 /* 5796 * Setup the histogram arrays (psize, lsize, and asize). 5797 */ 5798 one_histo_t parm_histo[NUM_HISTO]; 5799 5800 parm_histo[0].name = "psize"; 5801 parm_histo[0].count = zcb->zcb_psize_count; 5802 parm_histo[0].len = zcb->zcb_psize_len; 5803 parm_histo[0].cumulative = 0; 5804 5805 parm_histo[1].name = "lsize"; 5806 parm_histo[1].count = zcb->zcb_lsize_count; 5807 parm_histo[1].len = zcb->zcb_lsize_len; 5808 parm_histo[1].cumulative = 0; 5809 5810 parm_histo[2].name = "asize"; 5811 parm_histo[2].count = zcb->zcb_asize_count; 5812 parm_histo[2].len = zcb->zcb_asize_len; 5813 parm_histo[2].cumulative = 0; 5814 5815 5816 (void) printf("\nBlock Size Histogram\n"); 5817 /* 5818 * Print the first line titles 5819 */ 5820 if (dump_opt['P']) 5821 (void) printf("\n%s\t", blocksize_title1); 5822 else 5823 (void) printf("\n%7s ", blocksize_title1); 5824 5825 for (int j = 0; j < NUM_HISTO; j++) { 5826 if (dump_opt['P']) { 5827 if (j < NUM_HISTO - 1) { 5828 (void) printf("%s\t\t\t", parm_histo[j].name); 5829 } else { 5830 /* Don't print trailing spaces */ 5831 (void) printf(" %s", parm_histo[j].name); 5832 } 5833 } else { 5834 if (j < NUM_HISTO - 1) { 5835 /* Left aligned strings in the output */ 5836 (void) printf("%-7s ", 5837 parm_histo[j].name); 5838 } else { 5839 /* Don't print trailing spaces */ 5840 (void) printf("%s", parm_histo[j].name); 5841 } 5842 } 5843 } 5844 (void) printf("\n"); 5845 5846 /* 5847 * Print the second line titles 5848 */ 5849 if (dump_opt['P']) { 5850 (void) printf("%s\t", blocksize_title2); 5851 } else { 5852 (void) printf("%7s ", blocksize_title2); 5853 } 5854 5855 for (int i = 0; i < NUM_HISTO; i++) { 5856 if (dump_opt['P']) { 5857 (void) printf("%s\t%s\t%s\t", 5858 count_title, length_title, cumulative_title); 5859 } else { 5860 (void) printf("%7s%7s%7s", 5861 count_title, length_title, cumulative_title); 5862 } 5863 } 5864 (void) printf("\n"); 5865 5866 /* 5867 * Print the rows 5868 */ 5869 for (int i = SPA_MINBLOCKSHIFT; i < SPA_MAX_FOR_16M; i++) { 5870 5871 /* 5872 * Print the first column showing the blocksize 5873 */ 5874 zdb_nicenum((1ULL << i), numbuf, sizeof (numbuf)); 5875 5876 if (dump_opt['P']) { 5877 printf("%s", numbuf); 5878 } else { 5879 printf("%7s:", numbuf); 5880 } 5881 5882 /* 5883 * Print the remaining set of 3 columns per size: 5884 * for psize, lsize and asize 5885 */ 5886 for (int j = 0; j < NUM_HISTO; j++) { 5887 parm_histo[j].cumulative += parm_histo[j].len[i]; 5888 5889 zdb_nicenum(parm_histo[j].count[i], 5890 numbuf, sizeof (numbuf)); 5891 if (dump_opt['P']) 5892 (void) printf("\t%s", numbuf); 5893 else 5894 (void) printf("%7s", numbuf); 5895 5896 zdb_nicenum(parm_histo[j].len[i], 5897 numbuf, sizeof (numbuf)); 5898 if (dump_opt['P']) 5899 (void) printf("\t%s", numbuf); 5900 else 5901 (void) printf("%7s", numbuf); 5902 5903 zdb_nicenum(parm_histo[j].cumulative, 5904 numbuf, sizeof (numbuf)); 5905 if (dump_opt['P']) 5906 (void) printf("\t%s", numbuf); 5907 else 5908 (void) printf("%7s", numbuf); 5909 } 5910 (void) printf("\n"); 5911 } 5912 } 5913 5914 static void 5915 zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp, 5916 dmu_object_type_t type) 5917 { 5918 int i; 5919 5920 ASSERT(type < ZDB_OT_TOTAL); 5921 5922 if (zilog && zil_bp_tree_add(zilog, bp) != 0) 5923 return; 5924 5925 /* 5926 * This flag controls if we will issue a claim for the block while 5927 * counting it, to ensure that all blocks are referenced in space maps. 5928 * We don't issue claims if we're not doing leak tracking, because it's 5929 * expensive if the user isn't interested. We also don't claim the 5930 * second or later occurences of cloned or dedup'd blocks, because we 5931 * already claimed them the first time. 5932 */ 5933 boolean_t do_claim = !dump_opt['L']; 5934 5935 spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER); 5936 5937 blkptr_t tempbp; 5938 if (BP_GET_DEDUP(bp)) { 5939 /* 5940 * Dedup'd blocks are special. We need to count them, so we can 5941 * later uncount them when reporting leaked space, and we must 5942 * only claim them once. 5943 * 5944 * We use the existing dedup system to track what we've seen. 5945 * The first time we see a block, we do a ddt_lookup() to see 5946 * if it exists in the DDT. If we're doing leak tracking, we 5947 * claim the block at this time. 5948 * 5949 * Each time we see a block, we reduce the refcount in the 5950 * entry by one, and add to the size and count of dedup'd 5951 * blocks to report at the end. 5952 */ 5953 5954 ddt_t *ddt = ddt_select(zcb->zcb_spa, bp); 5955 5956 ddt_enter(ddt); 5957 5958 /* 5959 * Find the block. This will create the entry in memory, but 5960 * we'll know if that happened by its refcount. 5961 */ 5962 ddt_entry_t *dde = ddt_lookup(ddt, bp, B_TRUE); 5963 5964 /* 5965 * ddt_lookup() can return NULL if this block didn't exist 5966 * in the DDT and creating it would take the DDT over its 5967 * quota. Since we got the block from disk, it must exist in 5968 * the DDT, so this can't happen. However, when unique entries 5969 * are pruned, the dedup bit can be set with no corresponding 5970 * entry in the DDT. 5971 */ 5972 if (dde == NULL) { 5973 ddt_exit(ddt); 5974 goto skipped; 5975 } 5976 5977 /* Get the phys for this variant */ 5978 ddt_phys_variant_t v = ddt_phys_select(ddt, dde, bp); 5979 5980 /* 5981 * This entry may have multiple sets of DVAs. We must claim 5982 * each set the first time we see them in a real block on disk, 5983 * or count them on subsequent occurences. We don't have a 5984 * convenient way to track the first time we see each variant, 5985 * so we repurpose dde_io as a set of "seen" flag bits. We can 5986 * do this safely in zdb because it never writes, so it will 5987 * never have a writing zio for this block in that pointer. 5988 */ 5989 boolean_t seen = !!(((uintptr_t)dde->dde_io) & (1 << v)); 5990 if (!seen) 5991 dde->dde_io = 5992 (void *)(((uintptr_t)dde->dde_io) | (1 << v)); 5993 5994 /* Consume a reference for this block. */ 5995 if (ddt_phys_total_refcnt(ddt, dde->dde_phys) > 0) 5996 ddt_phys_decref(dde->dde_phys, v); 5997 5998 /* 5999 * If this entry has a single flat phys, it may have been 6000 * extended with additional DVAs at some time in its life. 6001 * This block might be from before it was fully extended, and 6002 * so have fewer DVAs. 6003 * 6004 * If this is the first time we've seen this block, and we 6005 * claimed it as-is, then we would miss the claim on some 6006 * number of DVAs, which would then be seen as leaked. 6007 * 6008 * In all cases, if we've had fewer DVAs, then the asize would 6009 * be too small, and would lead to the pool apparently using 6010 * more space than allocated. 6011 * 6012 * To handle this, we copy the canonical set of DVAs from the 6013 * entry back to the block pointer before we claim it. 6014 */ 6015 if (v == DDT_PHYS_FLAT) { 6016 ASSERT3U(BP_GET_PHYSICAL_BIRTH(bp), ==, 6017 ddt_phys_birth(dde->dde_phys, v)); 6018 tempbp = *bp; 6019 ddt_bp_fill(dde->dde_phys, v, &tempbp, 6020 BP_GET_PHYSICAL_BIRTH(bp)); 6021 bp = &tempbp; 6022 } 6023 6024 if (seen) { 6025 /* 6026 * The second or later time we see this block, 6027 * it's a duplicate and we count it. 6028 */ 6029 zcb->zcb_dedup_asize += BP_GET_ASIZE(bp); 6030 zcb->zcb_dedup_blocks++; 6031 6032 /* Already claimed, don't do it again. */ 6033 do_claim = B_FALSE; 6034 } 6035 6036 ddt_exit(ddt); 6037 } else if (zcb->zcb_brt_is_active && 6038 brt_maybe_exists(zcb->zcb_spa, bp)) { 6039 /* 6040 * Cloned blocks are special. We need to count them, so we can 6041 * later uncount them when reporting leaked space, and we must 6042 * only claim them once. 6043 * 6044 * To do this, we keep our own in-memory BRT. For each block 6045 * we haven't seen before, we look it up in the real BRT and 6046 * if its there, we note it and its refcount then proceed as 6047 * normal. If we see the block again, we count it as a clone 6048 * and then give it no further consideration. 6049 */ 6050 zdb_brt_entry_t zbre_search, *zbre; 6051 avl_index_t where; 6052 6053 zbre_search.zbre_dva = bp->blk_dva[0]; 6054 zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where); 6055 if (zbre == NULL) { 6056 /* Not seen before; track it */ 6057 uint64_t refcnt = 6058 brt_entry_get_refcount(zcb->zcb_spa, bp); 6059 if (refcnt > 0) { 6060 zbre = umem_zalloc(sizeof (zdb_brt_entry_t), 6061 UMEM_NOFAIL); 6062 zbre->zbre_dva = bp->blk_dva[0]; 6063 zbre->zbre_refcount = refcnt; 6064 avl_insert(&zcb->zcb_brt, zbre, where); 6065 } 6066 } else { 6067 /* 6068 * Second or later occurrence, count it and take a 6069 * refcount. 6070 */ 6071 zcb->zcb_clone_asize += BP_GET_ASIZE(bp); 6072 zcb->zcb_clone_blocks++; 6073 6074 zbre->zbre_refcount--; 6075 if (zbre->zbre_refcount == 0) { 6076 avl_remove(&zcb->zcb_brt, zbre); 6077 umem_free(zbre, sizeof (zdb_brt_entry_t)); 6078 } 6079 6080 /* Already claimed, don't do it again. */ 6081 do_claim = B_FALSE; 6082 } 6083 } 6084 6085 skipped: 6086 for (i = 0; i < 4; i++) { 6087 int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL; 6088 int t = (i & 1) ? type : ZDB_OT_TOTAL; 6089 int equal; 6090 zdb_blkstats_t *zb = &zcb->zcb_type[l][t]; 6091 6092 zb->zb_asize += BP_GET_ASIZE(bp); 6093 zb->zb_lsize += BP_GET_LSIZE(bp); 6094 zb->zb_psize += BP_GET_PSIZE(bp); 6095 zb->zb_count++; 6096 6097 /* 6098 * The histogram is only big enough to record blocks up to 6099 * SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last, 6100 * "other", bucket. 6101 */ 6102 unsigned idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT; 6103 idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1); 6104 zb->zb_psize_histogram[idx]++; 6105 6106 zb->zb_gangs += BP_COUNT_GANG(bp); 6107 6108 switch (BP_GET_NDVAS(bp)) { 6109 case 2: 6110 if (DVA_GET_VDEV(&bp->blk_dva[0]) == 6111 DVA_GET_VDEV(&bp->blk_dva[1])) { 6112 zb->zb_ditto_samevdev++; 6113 6114 if (same_metaslab(zcb->zcb_spa, 6115 DVA_GET_VDEV(&bp->blk_dva[0]), 6116 DVA_GET_OFFSET(&bp->blk_dva[0]), 6117 DVA_GET_OFFSET(&bp->blk_dva[1]))) 6118 zb->zb_ditto_same_ms++; 6119 } 6120 break; 6121 case 3: 6122 equal = (DVA_GET_VDEV(&bp->blk_dva[0]) == 6123 DVA_GET_VDEV(&bp->blk_dva[1])) + 6124 (DVA_GET_VDEV(&bp->blk_dva[0]) == 6125 DVA_GET_VDEV(&bp->blk_dva[2])) + 6126 (DVA_GET_VDEV(&bp->blk_dva[1]) == 6127 DVA_GET_VDEV(&bp->blk_dva[2])); 6128 if (equal != 0) { 6129 zb->zb_ditto_samevdev++; 6130 6131 if (DVA_GET_VDEV(&bp->blk_dva[0]) == 6132 DVA_GET_VDEV(&bp->blk_dva[1]) && 6133 same_metaslab(zcb->zcb_spa, 6134 DVA_GET_VDEV(&bp->blk_dva[0]), 6135 DVA_GET_OFFSET(&bp->blk_dva[0]), 6136 DVA_GET_OFFSET(&bp->blk_dva[1]))) 6137 zb->zb_ditto_same_ms++; 6138 else if (DVA_GET_VDEV(&bp->blk_dva[0]) == 6139 DVA_GET_VDEV(&bp->blk_dva[2]) && 6140 same_metaslab(zcb->zcb_spa, 6141 DVA_GET_VDEV(&bp->blk_dva[0]), 6142 DVA_GET_OFFSET(&bp->blk_dva[0]), 6143 DVA_GET_OFFSET(&bp->blk_dva[2]))) 6144 zb->zb_ditto_same_ms++; 6145 else if (DVA_GET_VDEV(&bp->blk_dva[1]) == 6146 DVA_GET_VDEV(&bp->blk_dva[2]) && 6147 same_metaslab(zcb->zcb_spa, 6148 DVA_GET_VDEV(&bp->blk_dva[1]), 6149 DVA_GET_OFFSET(&bp->blk_dva[1]), 6150 DVA_GET_OFFSET(&bp->blk_dva[2]))) 6151 zb->zb_ditto_same_ms++; 6152 } 6153 break; 6154 } 6155 } 6156 6157 spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG); 6158 6159 if (BP_IS_EMBEDDED(bp)) { 6160 zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++; 6161 zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)] 6162 [BPE_GET_PSIZE(bp)]++; 6163 return; 6164 } 6165 /* 6166 * The binning histogram bins by powers of two up to 6167 * SPA_MAXBLOCKSIZE rather than creating bins for 6168 * every possible blocksize found in the pool. 6169 */ 6170 int bin = highbit64(BP_GET_PSIZE(bp)) - 1; 6171 6172 zcb->zcb_psize_count[bin]++; 6173 zcb->zcb_psize_len[bin] += BP_GET_PSIZE(bp); 6174 zcb->zcb_psize_total += BP_GET_PSIZE(bp); 6175 6176 bin = highbit64(BP_GET_LSIZE(bp)) - 1; 6177 6178 zcb->zcb_lsize_count[bin]++; 6179 zcb->zcb_lsize_len[bin] += BP_GET_LSIZE(bp); 6180 zcb->zcb_lsize_total += BP_GET_LSIZE(bp); 6181 6182 bin = highbit64(BP_GET_ASIZE(bp)) - 1; 6183 6184 zcb->zcb_asize_count[bin]++; 6185 zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp); 6186 zcb->zcb_asize_total += BP_GET_ASIZE(bp); 6187 6188 if (!do_claim) 6189 return; 6190 6191 VERIFY0(zio_wait(zio_claim(NULL, zcb->zcb_spa, 6192 spa_min_claim_txg(zcb->zcb_spa), bp, NULL, NULL, 6193 ZIO_FLAG_CANFAIL))); 6194 } 6195 6196 static void 6197 zdb_blkptr_done(zio_t *zio) 6198 { 6199 spa_t *spa = zio->io_spa; 6200 blkptr_t *bp = zio->io_bp; 6201 int ioerr = zio->io_error; 6202 zdb_cb_t *zcb = zio->io_private; 6203 zbookmark_phys_t *zb = &zio->io_bookmark; 6204 6205 mutex_enter(&spa->spa_scrub_lock); 6206 spa->spa_load_verify_bytes -= BP_GET_PSIZE(bp); 6207 cv_broadcast(&spa->spa_scrub_io_cv); 6208 6209 if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { 6210 char blkbuf[BP_SPRINTF_LEN]; 6211 6212 zcb->zcb_haderrors = 1; 6213 zcb->zcb_errors[ioerr]++; 6214 6215 if (dump_opt['b'] >= 2) 6216 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp); 6217 else 6218 blkbuf[0] = '\0'; 6219 6220 (void) printf("zdb_blkptr_cb: " 6221 "Got error %d reading " 6222 "<%llu, %llu, %lld, %llx> %s -- skipping\n", 6223 ioerr, 6224 (u_longlong_t)zb->zb_objset, 6225 (u_longlong_t)zb->zb_object, 6226 (u_longlong_t)zb->zb_level, 6227 (u_longlong_t)zb->zb_blkid, 6228 blkbuf); 6229 } 6230 mutex_exit(&spa->spa_scrub_lock); 6231 6232 abd_free(zio->io_abd); 6233 } 6234 6235 static int 6236 zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 6237 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 6238 { 6239 zdb_cb_t *zcb = arg; 6240 dmu_object_type_t type; 6241 boolean_t is_metadata; 6242 6243 if (zb->zb_level == ZB_DNODE_LEVEL) 6244 return (0); 6245 6246 if (dump_opt['b'] >= 5 && BP_GET_BIRTH(bp) > 0) { 6247 char blkbuf[BP_SPRINTF_LEN]; 6248 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp); 6249 (void) printf("objset %llu object %llu " 6250 "level %lld offset 0x%llx %s\n", 6251 (u_longlong_t)zb->zb_objset, 6252 (u_longlong_t)zb->zb_object, 6253 (longlong_t)zb->zb_level, 6254 (u_longlong_t)blkid2offset(dnp, bp, zb), 6255 blkbuf); 6256 } 6257 6258 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) 6259 return (0); 6260 6261 type = BP_GET_TYPE(bp); 6262 6263 zdb_count_block(zcb, zilog, bp, 6264 (type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type); 6265 6266 is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)); 6267 6268 if (!BP_IS_EMBEDDED(bp) && 6269 (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) { 6270 size_t size = BP_GET_PSIZE(bp); 6271 abd_t *abd = abd_alloc(size, B_FALSE); 6272 int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW; 6273 6274 /* If it's an intent log block, failure is expected. */ 6275 if (zb->zb_level == ZB_ZIL_LEVEL) 6276 flags |= ZIO_FLAG_SPECULATIVE; 6277 6278 mutex_enter(&spa->spa_scrub_lock); 6279 while (spa->spa_load_verify_bytes > max_inflight_bytes) 6280 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 6281 spa->spa_load_verify_bytes += size; 6282 mutex_exit(&spa->spa_scrub_lock); 6283 6284 zio_nowait(zio_read(NULL, spa, bp, abd, size, 6285 zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb)); 6286 } 6287 6288 zcb->zcb_readfails = 0; 6289 6290 /* only call gethrtime() every 100 blocks */ 6291 static int iters; 6292 if (++iters > 100) 6293 iters = 0; 6294 else 6295 return (0); 6296 6297 if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) { 6298 uint64_t now = gethrtime(); 6299 char buf[10]; 6300 uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize; 6301 uint64_t kb_per_sec = 6302 1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000)); 6303 uint64_t sec_remaining = 6304 (zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec; 6305 6306 /* make sure nicenum has enough space */ 6307 _Static_assert(sizeof (buf) >= NN_NUMBUF_SZ, "buf truncated"); 6308 6309 zfs_nicebytes(bytes, buf, sizeof (buf)); 6310 (void) fprintf(stderr, 6311 "\r%5s completed (%4"PRIu64"MB/s) " 6312 "estimated time remaining: " 6313 "%"PRIu64"hr %02"PRIu64"min %02"PRIu64"sec ", 6314 buf, kb_per_sec / 1024, 6315 sec_remaining / 60 / 60, 6316 sec_remaining / 60 % 60, 6317 sec_remaining % 60); 6318 6319 zcb->zcb_lastprint = now; 6320 } 6321 6322 return (0); 6323 } 6324 6325 static void 6326 zdb_leak(void *arg, uint64_t start, uint64_t size) 6327 { 6328 vdev_t *vd = arg; 6329 6330 (void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n", 6331 (u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size); 6332 } 6333 6334 static metaslab_ops_t zdb_metaslab_ops = { 6335 NULL /* alloc */ 6336 }; 6337 6338 static int 6339 load_unflushed_svr_segs_cb(spa_t *spa, space_map_entry_t *sme, 6340 uint64_t txg, void *arg) 6341 { 6342 spa_vdev_removal_t *svr = arg; 6343 6344 uint64_t offset = sme->sme_offset; 6345 uint64_t size = sme->sme_run; 6346 6347 /* skip vdevs we don't care about */ 6348 if (sme->sme_vdev != svr->svr_vdev_id) 6349 return (0); 6350 6351 vdev_t *vd = vdev_lookup_top(spa, sme->sme_vdev); 6352 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift]; 6353 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE); 6354 6355 if (txg < metaslab_unflushed_txg(ms)) 6356 return (0); 6357 6358 if (sme->sme_type == SM_ALLOC) 6359 zfs_range_tree_add(svr->svr_allocd_segs, offset, size); 6360 else 6361 zfs_range_tree_remove(svr->svr_allocd_segs, offset, size); 6362 6363 return (0); 6364 } 6365 6366 static void 6367 claim_segment_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset, 6368 uint64_t size, void *arg) 6369 { 6370 (void) inner_offset, (void) arg; 6371 6372 /* 6373 * This callback was called through a remap from 6374 * a device being removed. Therefore, the vdev that 6375 * this callback is applied to is a concrete 6376 * vdev. 6377 */ 6378 ASSERT(vdev_is_concrete(vd)); 6379 6380 VERIFY0(metaslab_claim_impl(vd, offset, size, 6381 spa_min_claim_txg(vd->vdev_spa))); 6382 } 6383 6384 static void 6385 claim_segment_cb(void *arg, uint64_t offset, uint64_t size) 6386 { 6387 vdev_t *vd = arg; 6388 6389 vdev_indirect_ops.vdev_op_remap(vd, offset, size, 6390 claim_segment_impl_cb, NULL); 6391 } 6392 6393 /* 6394 * After accounting for all allocated blocks that are directly referenced, 6395 * we might have missed a reference to a block from a partially complete 6396 * (and thus unused) indirect mapping object. We perform a secondary pass 6397 * through the metaslabs we have already mapped and claim the destination 6398 * blocks. 6399 */ 6400 static void 6401 zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb) 6402 { 6403 if (dump_opt['L']) 6404 return; 6405 6406 if (spa->spa_vdev_removal == NULL) 6407 return; 6408 6409 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6410 6411 spa_vdev_removal_t *svr = spa->spa_vdev_removal; 6412 vdev_t *vd = vdev_lookup_top(spa, svr->svr_vdev_id); 6413 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 6414 6415 ASSERT0(zfs_range_tree_space(svr->svr_allocd_segs)); 6416 6417 zfs_range_tree_t *allocs = zfs_range_tree_create_flags( 6418 NULL, ZFS_RANGE_SEG64, NULL, 0, 0, 6419 0, "zdb_claim_removing:allocs"); 6420 for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) { 6421 metaslab_t *msp = vd->vdev_ms[msi]; 6422 6423 ASSERT0(zfs_range_tree_space(allocs)); 6424 if (msp->ms_sm != NULL) 6425 VERIFY0(space_map_load(msp->ms_sm, allocs, SM_ALLOC)); 6426 zfs_range_tree_vacate(allocs, zfs_range_tree_add, 6427 svr->svr_allocd_segs); 6428 } 6429 zfs_range_tree_destroy(allocs); 6430 6431 iterate_through_spacemap_logs(spa, load_unflushed_svr_segs_cb, svr); 6432 6433 /* 6434 * Clear everything past what has been synced, 6435 * because we have not allocated mappings for 6436 * it yet. 6437 */ 6438 zfs_range_tree_clear(svr->svr_allocd_segs, 6439 vdev_indirect_mapping_max_offset(vim), 6440 vd->vdev_asize - vdev_indirect_mapping_max_offset(vim)); 6441 6442 zcb->zcb_removing_size += zfs_range_tree_space(svr->svr_allocd_segs); 6443 zfs_range_tree_vacate(svr->svr_allocd_segs, claim_segment_cb, vd); 6444 6445 spa_config_exit(spa, SCL_CONFIG, FTAG); 6446 } 6447 6448 static int 6449 increment_indirect_mapping_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, 6450 dmu_tx_t *tx) 6451 { 6452 (void) tx; 6453 zdb_cb_t *zcb = arg; 6454 spa_t *spa = zcb->zcb_spa; 6455 vdev_t *vd; 6456 const dva_t *dva = &bp->blk_dva[0]; 6457 6458 ASSERT(!bp_freed); 6459 ASSERT(!dump_opt['L']); 6460 ASSERT3U(BP_GET_NDVAS(bp), ==, 1); 6461 6462 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 6463 vd = vdev_lookup_top(zcb->zcb_spa, DVA_GET_VDEV(dva)); 6464 ASSERT3P(vd, !=, NULL); 6465 spa_config_exit(spa, SCL_VDEV, FTAG); 6466 6467 ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0); 6468 ASSERT3P(zcb->zcb_vd_obsolete_counts[vd->vdev_id], !=, NULL); 6469 6470 vdev_indirect_mapping_increment_obsolete_count( 6471 vd->vdev_indirect_mapping, 6472 DVA_GET_OFFSET(dva), DVA_GET_ASIZE(dva), 6473 zcb->zcb_vd_obsolete_counts[vd->vdev_id]); 6474 6475 return (0); 6476 } 6477 6478 static uint32_t * 6479 zdb_load_obsolete_counts(vdev_t *vd) 6480 { 6481 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 6482 spa_t *spa = vd->vdev_spa; 6483 spa_condensing_indirect_phys_t *scip = 6484 &spa->spa_condensing_indirect_phys; 6485 uint64_t obsolete_sm_object; 6486 uint32_t *counts; 6487 6488 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object)); 6489 EQUIV(obsolete_sm_object != 0, vd->vdev_obsolete_sm != NULL); 6490 counts = vdev_indirect_mapping_load_obsolete_counts(vim); 6491 if (vd->vdev_obsolete_sm != NULL) { 6492 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts, 6493 vd->vdev_obsolete_sm); 6494 } 6495 if (scip->scip_vdev == vd->vdev_id && 6496 scip->scip_prev_obsolete_sm_object != 0) { 6497 space_map_t *prev_obsolete_sm = NULL; 6498 VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset, 6499 scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0)); 6500 vdev_indirect_mapping_load_obsolete_spacemap(vim, counts, 6501 prev_obsolete_sm); 6502 space_map_close(prev_obsolete_sm); 6503 } 6504 return (counts); 6505 } 6506 6507 typedef struct checkpoint_sm_exclude_entry_arg { 6508 vdev_t *cseea_vd; 6509 uint64_t cseea_checkpoint_size; 6510 } checkpoint_sm_exclude_entry_arg_t; 6511 6512 static int 6513 checkpoint_sm_exclude_entry_cb(space_map_entry_t *sme, void *arg) 6514 { 6515 checkpoint_sm_exclude_entry_arg_t *cseea = arg; 6516 vdev_t *vd = cseea->cseea_vd; 6517 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift]; 6518 uint64_t end = sme->sme_offset + sme->sme_run; 6519 6520 ASSERT(sme->sme_type == SM_FREE); 6521 6522 /* 6523 * Since the vdev_checkpoint_sm exists in the vdev level 6524 * and the ms_sm space maps exist in the metaslab level, 6525 * an entry in the checkpoint space map could theoretically 6526 * cross the boundaries of the metaslab that it belongs. 6527 * 6528 * In reality, because of the way that we populate and 6529 * manipulate the checkpoint's space maps currently, 6530 * there shouldn't be any entries that cross metaslabs. 6531 * Hence the assertion below. 6532 * 6533 * That said, there is no fundamental requirement that 6534 * the checkpoint's space map entries should not cross 6535 * metaslab boundaries. So if needed we could add code 6536 * that handles metaslab-crossing segments in the future. 6537 */ 6538 VERIFY3U(sme->sme_offset, >=, ms->ms_start); 6539 VERIFY3U(end, <=, ms->ms_start + ms->ms_size); 6540 6541 /* 6542 * By removing the entry from the allocated segments we 6543 * also verify that the entry is there to begin with. 6544 */ 6545 mutex_enter(&ms->ms_lock); 6546 zfs_range_tree_remove(ms->ms_allocatable, sme->sme_offset, 6547 sme->sme_run); 6548 mutex_exit(&ms->ms_lock); 6549 6550 cseea->cseea_checkpoint_size += sme->sme_run; 6551 return (0); 6552 } 6553 6554 static void 6555 zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_cb_t *zcb) 6556 { 6557 spa_t *spa = vd->vdev_spa; 6558 space_map_t *checkpoint_sm = NULL; 6559 uint64_t checkpoint_sm_obj; 6560 6561 /* 6562 * If there is no vdev_top_zap, we are in a pool whose 6563 * version predates the pool checkpoint feature. 6564 */ 6565 if (vd->vdev_top_zap == 0) 6566 return; 6567 6568 /* 6569 * If there is no reference of the vdev_checkpoint_sm in 6570 * the vdev_top_zap, then one of the following scenarios 6571 * is true: 6572 * 6573 * 1] There is no checkpoint 6574 * 2] There is a checkpoint, but no checkpointed blocks 6575 * have been freed yet 6576 * 3] The current vdev is indirect 6577 * 6578 * In these cases we return immediately. 6579 */ 6580 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap, 6581 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0) 6582 return; 6583 6584 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap, 6585 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, 6586 &checkpoint_sm_obj)); 6587 6588 checkpoint_sm_exclude_entry_arg_t cseea; 6589 cseea.cseea_vd = vd; 6590 cseea.cseea_checkpoint_size = 0; 6591 6592 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa), 6593 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift)); 6594 6595 VERIFY0(space_map_iterate(checkpoint_sm, 6596 space_map_length(checkpoint_sm), 6597 checkpoint_sm_exclude_entry_cb, &cseea)); 6598 space_map_close(checkpoint_sm); 6599 6600 zcb->zcb_checkpoint_size += cseea.cseea_checkpoint_size; 6601 } 6602 6603 static void 6604 zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb) 6605 { 6606 ASSERT(!dump_opt['L']); 6607 6608 vdev_t *rvd = spa->spa_root_vdev; 6609 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 6610 ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id); 6611 zdb_leak_init_vdev_exclude_checkpoint(rvd->vdev_child[c], zcb); 6612 } 6613 } 6614 6615 static int 6616 count_unflushed_space_cb(spa_t *spa, space_map_entry_t *sme, 6617 uint64_t txg, void *arg) 6618 { 6619 int64_t *ualloc_space = arg; 6620 6621 uint64_t offset = sme->sme_offset; 6622 uint64_t vdev_id = sme->sme_vdev; 6623 6624 vdev_t *vd = vdev_lookup_top(spa, vdev_id); 6625 if (!vdev_is_concrete(vd)) 6626 return (0); 6627 6628 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift]; 6629 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE); 6630 6631 if (txg < metaslab_unflushed_txg(ms)) 6632 return (0); 6633 6634 if (sme->sme_type == SM_ALLOC) 6635 *ualloc_space += sme->sme_run; 6636 else 6637 *ualloc_space -= sme->sme_run; 6638 6639 return (0); 6640 } 6641 6642 static int64_t 6643 get_unflushed_alloc_space(spa_t *spa) 6644 { 6645 if (dump_opt['L']) 6646 return (0); 6647 6648 int64_t ualloc_space = 0; 6649 iterate_through_spacemap_logs(spa, count_unflushed_space_cb, 6650 &ualloc_space); 6651 return (ualloc_space); 6652 } 6653 6654 static int 6655 load_unflushed_cb(spa_t *spa, space_map_entry_t *sme, uint64_t txg, void *arg) 6656 { 6657 maptype_t *uic_maptype = arg; 6658 6659 uint64_t offset = sme->sme_offset; 6660 uint64_t size = sme->sme_run; 6661 uint64_t vdev_id = sme->sme_vdev; 6662 6663 vdev_t *vd = vdev_lookup_top(spa, vdev_id); 6664 6665 /* skip indirect vdevs */ 6666 if (!vdev_is_concrete(vd)) 6667 return (0); 6668 6669 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift]; 6670 6671 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE); 6672 ASSERT(*uic_maptype == SM_ALLOC || *uic_maptype == SM_FREE); 6673 6674 if (txg < metaslab_unflushed_txg(ms)) 6675 return (0); 6676 6677 if (*uic_maptype == sme->sme_type) 6678 zfs_range_tree_add(ms->ms_allocatable, offset, size); 6679 else 6680 zfs_range_tree_remove(ms->ms_allocatable, offset, size); 6681 6682 return (0); 6683 } 6684 6685 static void 6686 load_unflushed_to_ms_allocatables(spa_t *spa, maptype_t maptype) 6687 { 6688 iterate_through_spacemap_logs(spa, load_unflushed_cb, &maptype); 6689 } 6690 6691 static void 6692 load_concrete_ms_allocatable_trees(spa_t *spa, maptype_t maptype) 6693 { 6694 vdev_t *rvd = spa->spa_root_vdev; 6695 for (uint64_t i = 0; i < rvd->vdev_children; i++) { 6696 vdev_t *vd = rvd->vdev_child[i]; 6697 6698 ASSERT3U(i, ==, vd->vdev_id); 6699 6700 if (vd->vdev_ops == &vdev_indirect_ops) 6701 continue; 6702 6703 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) { 6704 metaslab_t *msp = vd->vdev_ms[m]; 6705 6706 (void) fprintf(stderr, 6707 "\rloading concrete vdev %llu, " 6708 "metaslab %llu of %llu ...", 6709 (longlong_t)vd->vdev_id, 6710 (longlong_t)msp->ms_id, 6711 (longlong_t)vd->vdev_ms_count); 6712 6713 mutex_enter(&msp->ms_lock); 6714 zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL); 6715 6716 /* 6717 * We don't want to spend the CPU manipulating the 6718 * size-ordered tree, so clear the range_tree ops. 6719 */ 6720 msp->ms_allocatable->rt_ops = NULL; 6721 6722 if (msp->ms_sm != NULL) { 6723 VERIFY0(space_map_load(msp->ms_sm, 6724 msp->ms_allocatable, maptype)); 6725 } 6726 if (!msp->ms_loaded) 6727 msp->ms_loaded = B_TRUE; 6728 mutex_exit(&msp->ms_lock); 6729 } 6730 } 6731 6732 load_unflushed_to_ms_allocatables(spa, maptype); 6733 } 6734 6735 /* 6736 * vm_idxp is an in-out parameter which (for indirect vdevs) is the 6737 * index in vim_entries that has the first entry in this metaslab. 6738 * On return, it will be set to the first entry after this metaslab. 6739 */ 6740 static void 6741 load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab_t *msp, 6742 uint64_t *vim_idxp) 6743 { 6744 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 6745 6746 mutex_enter(&msp->ms_lock); 6747 zfs_range_tree_vacate(msp->ms_allocatable, NULL, NULL); 6748 6749 /* 6750 * We don't want to spend the CPU manipulating the 6751 * size-ordered tree, so clear the range_tree ops. 6752 */ 6753 msp->ms_allocatable->rt_ops = NULL; 6754 6755 for (; *vim_idxp < vdev_indirect_mapping_num_entries(vim); 6756 (*vim_idxp)++) { 6757 vdev_indirect_mapping_entry_phys_t *vimep = 6758 &vim->vim_entries[*vim_idxp]; 6759 uint64_t ent_offset = DVA_MAPPING_GET_SRC_OFFSET(vimep); 6760 uint64_t ent_len = DVA_GET_ASIZE(&vimep->vimep_dst); 6761 ASSERT3U(ent_offset, >=, msp->ms_start); 6762 if (ent_offset >= msp->ms_start + msp->ms_size) 6763 break; 6764 6765 /* 6766 * Mappings do not cross metaslab boundaries, 6767 * because we create them by walking the metaslabs. 6768 */ 6769 ASSERT3U(ent_offset + ent_len, <=, 6770 msp->ms_start + msp->ms_size); 6771 zfs_range_tree_add(msp->ms_allocatable, ent_offset, ent_len); 6772 } 6773 6774 if (!msp->ms_loaded) 6775 msp->ms_loaded = B_TRUE; 6776 mutex_exit(&msp->ms_lock); 6777 } 6778 6779 static void 6780 zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb) 6781 { 6782 ASSERT(!dump_opt['L']); 6783 6784 vdev_t *rvd = spa->spa_root_vdev; 6785 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 6786 vdev_t *vd = rvd->vdev_child[c]; 6787 6788 ASSERT3U(c, ==, vd->vdev_id); 6789 6790 if (vd->vdev_ops != &vdev_indirect_ops) 6791 continue; 6792 6793 /* 6794 * Note: we don't check for mapping leaks on 6795 * removing vdevs because their ms_allocatable's 6796 * are used to look for leaks in allocated space. 6797 */ 6798 zcb->zcb_vd_obsolete_counts[c] = zdb_load_obsolete_counts(vd); 6799 6800 /* 6801 * Normally, indirect vdevs don't have any 6802 * metaslabs. We want to set them up for 6803 * zio_claim(). 6804 */ 6805 vdev_metaslab_group_create(vd); 6806 VERIFY0(vdev_metaslab_init(vd, 0)); 6807 6808 vdev_indirect_mapping_t *vim __maybe_unused = 6809 vd->vdev_indirect_mapping; 6810 uint64_t vim_idx = 0; 6811 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) { 6812 6813 (void) fprintf(stderr, 6814 "\rloading indirect vdev %llu, " 6815 "metaslab %llu of %llu ...", 6816 (longlong_t)vd->vdev_id, 6817 (longlong_t)vd->vdev_ms[m]->ms_id, 6818 (longlong_t)vd->vdev_ms_count); 6819 6820 load_indirect_ms_allocatable_tree(vd, vd->vdev_ms[m], 6821 &vim_idx); 6822 } 6823 ASSERT3U(vim_idx, ==, vdev_indirect_mapping_num_entries(vim)); 6824 } 6825 } 6826 6827 static void 6828 zdb_leak_init(spa_t *spa, zdb_cb_t *zcb) 6829 { 6830 zcb->zcb_spa = spa; 6831 6832 if (dump_opt['L']) 6833 return; 6834 6835 dsl_pool_t *dp = spa->spa_dsl_pool; 6836 vdev_t *rvd = spa->spa_root_vdev; 6837 6838 /* 6839 * We are going to be changing the meaning of the metaslab's 6840 * ms_allocatable. Ensure that the allocator doesn't try to 6841 * use the tree. 6842 */ 6843 spa->spa_normal_class->mc_ops = &zdb_metaslab_ops; 6844 spa->spa_log_class->mc_ops = &zdb_metaslab_ops; 6845 spa->spa_embedded_log_class->mc_ops = &zdb_metaslab_ops; 6846 spa->spa_special_embedded_log_class->mc_ops = &zdb_metaslab_ops; 6847 6848 zcb->zcb_vd_obsolete_counts = 6849 umem_zalloc(rvd->vdev_children * sizeof (uint32_t *), 6850 UMEM_NOFAIL); 6851 6852 /* 6853 * For leak detection, we overload the ms_allocatable trees 6854 * to contain allocated segments instead of free segments. 6855 * As a result, we can't use the normal metaslab_load/unload 6856 * interfaces. 6857 */ 6858 zdb_leak_init_prepare_indirect_vdevs(spa, zcb); 6859 load_concrete_ms_allocatable_trees(spa, SM_ALLOC); 6860 6861 /* 6862 * On load_concrete_ms_allocatable_trees() we loaded all the 6863 * allocated entries from the ms_sm to the ms_allocatable for 6864 * each metaslab. If the pool has a checkpoint or is in the 6865 * middle of discarding a checkpoint, some of these blocks 6866 * may have been freed but their ms_sm may not have been 6867 * updated because they are referenced by the checkpoint. In 6868 * order to avoid false-positives during leak-detection, we 6869 * go through the vdev's checkpoint space map and exclude all 6870 * its entries from their relevant ms_allocatable. 6871 * 6872 * We also aggregate the space held by the checkpoint and add 6873 * it to zcb_checkpoint_size. 6874 * 6875 * Note that at this point we are also verifying that all the 6876 * entries on the checkpoint_sm are marked as allocated in 6877 * the ms_sm of their relevant metaslab. 6878 * [see comment in checkpoint_sm_exclude_entry_cb()] 6879 */ 6880 zdb_leak_init_exclude_checkpoint(spa, zcb); 6881 ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa)); 6882 6883 /* for cleaner progress output */ 6884 (void) fprintf(stderr, "\n"); 6885 6886 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) { 6887 ASSERT(spa_feature_is_enabled(spa, 6888 SPA_FEATURE_DEVICE_REMOVAL)); 6889 (void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj, 6890 increment_indirect_mapping_cb, zcb, NULL); 6891 } 6892 } 6893 6894 static boolean_t 6895 zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb) 6896 { 6897 boolean_t leaks = B_FALSE; 6898 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 6899 uint64_t total_leaked = 0; 6900 boolean_t are_precise = B_FALSE; 6901 6902 ASSERT(vim != NULL); 6903 6904 for (uint64_t i = 0; i < vdev_indirect_mapping_num_entries(vim); i++) { 6905 vdev_indirect_mapping_entry_phys_t *vimep = 6906 &vim->vim_entries[i]; 6907 uint64_t obsolete_bytes = 0; 6908 uint64_t offset = DVA_MAPPING_GET_SRC_OFFSET(vimep); 6909 metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; 6910 6911 /* 6912 * This is not very efficient but it's easy to 6913 * verify correctness. 6914 */ 6915 for (uint64_t inner_offset = 0; 6916 inner_offset < DVA_GET_ASIZE(&vimep->vimep_dst); 6917 inner_offset += 1ULL << vd->vdev_ashift) { 6918 if (zfs_range_tree_contains(msp->ms_allocatable, 6919 offset + inner_offset, 1ULL << vd->vdev_ashift)) { 6920 obsolete_bytes += 1ULL << vd->vdev_ashift; 6921 } 6922 } 6923 6924 int64_t bytes_leaked = obsolete_bytes - 6925 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]; 6926 ASSERT3U(DVA_GET_ASIZE(&vimep->vimep_dst), >=, 6927 zcb->zcb_vd_obsolete_counts[vd->vdev_id][i]); 6928 6929 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise)); 6930 if (bytes_leaked != 0 && (are_precise || dump_opt['d'] >= 5)) { 6931 (void) printf("obsolete indirect mapping count " 6932 "mismatch on %llu:%llx:%llx : %llx bytes leaked\n", 6933 (u_longlong_t)vd->vdev_id, 6934 (u_longlong_t)DVA_MAPPING_GET_SRC_OFFSET(vimep), 6935 (u_longlong_t)DVA_GET_ASIZE(&vimep->vimep_dst), 6936 (u_longlong_t)bytes_leaked); 6937 } 6938 total_leaked += ABS(bytes_leaked); 6939 } 6940 6941 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise)); 6942 if (!are_precise && total_leaked > 0) { 6943 int pct_leaked = total_leaked * 100 / 6944 vdev_indirect_mapping_bytes_mapped(vim); 6945 (void) printf("cannot verify obsolete indirect mapping " 6946 "counts of vdev %llu because precise feature was not " 6947 "enabled when it was removed: %d%% (%llx bytes) of mapping" 6948 "unreferenced\n", 6949 (u_longlong_t)vd->vdev_id, pct_leaked, 6950 (u_longlong_t)total_leaked); 6951 } else if (total_leaked > 0) { 6952 (void) printf("obsolete indirect mapping count mismatch " 6953 "for vdev %llu -- %llx total bytes mismatched\n", 6954 (u_longlong_t)vd->vdev_id, 6955 (u_longlong_t)total_leaked); 6956 leaks |= B_TRUE; 6957 } 6958 6959 vdev_indirect_mapping_free_obsolete_counts(vim, 6960 zcb->zcb_vd_obsolete_counts[vd->vdev_id]); 6961 zcb->zcb_vd_obsolete_counts[vd->vdev_id] = NULL; 6962 6963 return (leaks); 6964 } 6965 6966 static boolean_t 6967 zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb) 6968 { 6969 if (dump_opt['L']) 6970 return (B_FALSE); 6971 6972 boolean_t leaks = B_FALSE; 6973 vdev_t *rvd = spa->spa_root_vdev; 6974 for (unsigned c = 0; c < rvd->vdev_children; c++) { 6975 vdev_t *vd = rvd->vdev_child[c]; 6976 6977 if (zcb->zcb_vd_obsolete_counts[c] != NULL) { 6978 leaks |= zdb_check_for_obsolete_leaks(vd, zcb); 6979 } 6980 6981 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) { 6982 metaslab_t *msp = vd->vdev_ms[m]; 6983 ASSERT3P(msp->ms_group, ==, (msp->ms_group->mg_class == 6984 spa_embedded_log_class(spa) || 6985 msp->ms_group->mg_class == 6986 spa_special_embedded_log_class(spa)) ? 6987 vd->vdev_log_mg : vd->vdev_mg); 6988 6989 /* 6990 * ms_allocatable has been overloaded 6991 * to contain allocated segments. Now that 6992 * we finished traversing all blocks, any 6993 * block that remains in the ms_allocatable 6994 * represents an allocated block that we 6995 * did not claim during the traversal. 6996 * Claimed blocks would have been removed 6997 * from the ms_allocatable. For indirect 6998 * vdevs, space remaining in the tree 6999 * represents parts of the mapping that are 7000 * not referenced, which is not a bug. 7001 */ 7002 if (vd->vdev_ops == &vdev_indirect_ops) { 7003 zfs_range_tree_vacate(msp->ms_allocatable, 7004 NULL, NULL); 7005 } else { 7006 zfs_range_tree_vacate(msp->ms_allocatable, 7007 zdb_leak, vd); 7008 } 7009 if (msp->ms_loaded) { 7010 msp->ms_loaded = B_FALSE; 7011 } 7012 } 7013 } 7014 7015 umem_free(zcb->zcb_vd_obsolete_counts, 7016 rvd->vdev_children * sizeof (uint32_t *)); 7017 zcb->zcb_vd_obsolete_counts = NULL; 7018 7019 return (leaks); 7020 } 7021 7022 static int 7023 count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 7024 { 7025 (void) tx; 7026 zdb_cb_t *zcb = arg; 7027 7028 if (dump_opt['b'] >= 5) { 7029 char blkbuf[BP_SPRINTF_LEN]; 7030 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp); 7031 (void) printf("[%s] %s\n", 7032 "deferred free", blkbuf); 7033 } 7034 zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED); 7035 return (0); 7036 } 7037 7038 /* 7039 * Iterate over livelists which have been destroyed by the user but 7040 * are still present in the MOS, waiting to be freed 7041 */ 7042 static void 7043 iterate_deleted_livelists(spa_t *spa, ll_iter_t func, void *arg) 7044 { 7045 objset_t *mos = spa->spa_meta_objset; 7046 uint64_t zap_obj; 7047 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT, 7048 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj); 7049 if (err == ENOENT) 7050 return; 7051 ASSERT0(err); 7052 7053 zap_cursor_t zc; 7054 zap_attribute_t *attrp = zap_attribute_alloc(); 7055 dsl_deadlist_t ll; 7056 /* NULL out os prior to dsl_deadlist_open in case it's garbage */ 7057 ll.dl_os = NULL; 7058 for (zap_cursor_init(&zc, mos, zap_obj); 7059 zap_cursor_retrieve(&zc, attrp) == 0; 7060 (void) zap_cursor_advance(&zc)) { 7061 VERIFY0(dsl_deadlist_open(&ll, mos, attrp->za_first_integer)); 7062 func(&ll, arg); 7063 dsl_deadlist_close(&ll); 7064 } 7065 zap_cursor_fini(&zc); 7066 zap_attribute_free(attrp); 7067 } 7068 7069 static int 7070 bpobj_count_block_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, 7071 dmu_tx_t *tx) 7072 { 7073 ASSERT(!bp_freed); 7074 return (count_block_cb(arg, bp, tx)); 7075 } 7076 7077 static int 7078 livelist_entry_count_blocks_cb(void *args, dsl_deadlist_entry_t *dle) 7079 { 7080 zdb_cb_t *zbc = args; 7081 bplist_t blks; 7082 bplist_create(&blks); 7083 /* determine which blocks have been alloc'd but not freed */ 7084 VERIFY0(dsl_process_sub_livelist(&dle->dle_bpobj, &blks, NULL, NULL)); 7085 /* count those blocks */ 7086 (void) bplist_iterate(&blks, count_block_cb, zbc, NULL); 7087 bplist_destroy(&blks); 7088 return (0); 7089 } 7090 7091 static void 7092 livelist_count_blocks(dsl_deadlist_t *ll, void *arg) 7093 { 7094 dsl_deadlist_iterate(ll, livelist_entry_count_blocks_cb, arg); 7095 } 7096 7097 /* 7098 * Count the blocks in the livelists that have been destroyed by the user 7099 * but haven't yet been freed. 7100 */ 7101 static void 7102 deleted_livelists_count_blocks(spa_t *spa, zdb_cb_t *zbc) 7103 { 7104 iterate_deleted_livelists(spa, livelist_count_blocks, zbc); 7105 } 7106 7107 static void 7108 dump_livelist_cb(dsl_deadlist_t *ll, void *arg) 7109 { 7110 ASSERT0P(arg); 7111 global_feature_count[SPA_FEATURE_LIVELIST]++; 7112 dump_blkptr_list(ll, "Deleted Livelist"); 7113 dsl_deadlist_iterate(ll, sublivelist_verify_lightweight, NULL); 7114 } 7115 7116 /* 7117 * Print out, register object references to, and increment feature counts for 7118 * livelists that have been destroyed by the user but haven't yet been freed. 7119 */ 7120 static void 7121 deleted_livelists_dump_mos(spa_t *spa) 7122 { 7123 uint64_t zap_obj; 7124 objset_t *mos = spa->spa_meta_objset; 7125 int err = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT, 7126 DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj); 7127 if (err == ENOENT) 7128 return; 7129 mos_obj_refd(zap_obj); 7130 iterate_deleted_livelists(spa, dump_livelist_cb, NULL); 7131 } 7132 7133 static int 7134 zdb_brt_entry_compare(const void *zcn1, const void *zcn2) 7135 { 7136 const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva; 7137 const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva; 7138 int cmp; 7139 7140 cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2)); 7141 if (cmp == 0) 7142 cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2)); 7143 7144 return (cmp); 7145 } 7146 7147 static int 7148 dump_block_stats(spa_t *spa) 7149 { 7150 zdb_cb_t *zcb; 7151 zdb_blkstats_t *zb, *tzb; 7152 uint64_t norm_alloc, norm_space, total_alloc, total_found; 7153 int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | 7154 TRAVERSE_NO_DECRYPT | TRAVERSE_HARD; 7155 boolean_t leaks = B_FALSE; 7156 int e, c, err; 7157 bp_embedded_type_t i; 7158 7159 ddt_prefetch_all(spa); 7160 7161 zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL); 7162 7163 if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) { 7164 avl_create(&zcb->zcb_brt, zdb_brt_entry_compare, 7165 sizeof (zdb_brt_entry_t), 7166 offsetof(zdb_brt_entry_t, zbre_node)); 7167 zcb->zcb_brt_is_active = B_TRUE; 7168 } 7169 7170 (void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n", 7171 (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "", 7172 (dump_opt['c'] == 1) ? "metadata " : "", 7173 dump_opt['c'] ? "checksums " : "", 7174 (dump_opt['c'] && !dump_opt['L']) ? "and verify " : "", 7175 !dump_opt['L'] ? "nothing leaked " : ""); 7176 7177 /* 7178 * When leak detection is enabled we load all space maps as SM_ALLOC 7179 * maps, then traverse the pool claiming each block we discover. If 7180 * the pool is perfectly consistent, the segment trees will be empty 7181 * when we're done. Anything left over is a leak; any block we can't 7182 * claim (because it's not part of any space map) is a double 7183 * allocation, reference to a freed block, or an unclaimed log block. 7184 * 7185 * When leak detection is disabled (-L option) we still traverse the 7186 * pool claiming each block we discover, but we skip opening any space 7187 * maps. 7188 */ 7189 zdb_leak_init(spa, zcb); 7190 7191 /* 7192 * If there's a deferred-free bplist, process that first. 7193 */ 7194 (void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj, 7195 bpobj_count_block_cb, zcb, NULL); 7196 7197 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { 7198 (void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj, 7199 bpobj_count_block_cb, zcb, NULL); 7200 } 7201 7202 zdb_claim_removing(spa, zcb); 7203 7204 if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) { 7205 VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset, 7206 spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb, 7207 zcb, NULL)); 7208 } 7209 7210 deleted_livelists_count_blocks(spa, zcb); 7211 7212 if (dump_opt['c'] > 1) 7213 flags |= TRAVERSE_PREFETCH_DATA; 7214 7215 zcb->zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa)); 7216 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa)); 7217 zcb->zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa)); 7218 zcb->zcb_totalasize += 7219 metaslab_class_get_alloc(spa_embedded_log_class(spa)); 7220 zcb->zcb_totalasize += 7221 metaslab_class_get_alloc(spa_special_embedded_log_class(spa)); 7222 zcb->zcb_start = zcb->zcb_lastprint = gethrtime(); 7223 err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, zcb); 7224 7225 /* 7226 * If we've traversed the data blocks then we need to wait for those 7227 * I/Os to complete. We leverage "The Godfather" zio to wait on 7228 * all async I/Os to complete. 7229 */ 7230 if (dump_opt['c']) { 7231 for (c = 0; c < max_ncpus; c++) { 7232 (void) zio_wait(spa->spa_async_zio_root[c]); 7233 spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL, 7234 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 7235 ZIO_FLAG_GODFATHER); 7236 } 7237 } 7238 ASSERT0(spa->spa_load_verify_bytes); 7239 7240 /* 7241 * Done after zio_wait() since zcb_haderrors is modified in 7242 * zdb_blkptr_done() 7243 */ 7244 zcb->zcb_haderrors |= err; 7245 7246 if (zcb->zcb_haderrors) { 7247 (void) printf("\nError counts:\n\n"); 7248 (void) printf("\t%5s %s\n", "errno", "count"); 7249 for (e = 0; e < 256; e++) { 7250 if (zcb->zcb_errors[e] != 0) { 7251 (void) printf("\t%5d %llu\n", 7252 e, (u_longlong_t)zcb->zcb_errors[e]); 7253 } 7254 } 7255 } 7256 7257 /* 7258 * Report any leaked segments. 7259 */ 7260 leaks |= zdb_leak_fini(spa, zcb); 7261 7262 tzb = &zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL]; 7263 7264 norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 7265 norm_space = metaslab_class_get_space(spa_normal_class(spa)); 7266 7267 total_alloc = norm_alloc + 7268 metaslab_class_get_alloc(spa_log_class(spa)) + 7269 metaslab_class_get_alloc(spa_embedded_log_class(spa)) + 7270 metaslab_class_get_alloc(spa_special_embedded_log_class(spa)) + 7271 metaslab_class_get_alloc(spa_special_class(spa)) + 7272 metaslab_class_get_alloc(spa_dedup_class(spa)) + 7273 get_unflushed_alloc_space(spa); 7274 total_found = 7275 tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize + 7276 zcb->zcb_removing_size + zcb->zcb_checkpoint_size; 7277 7278 if (total_found == total_alloc && !dump_opt['L']) { 7279 (void) printf("\n\tNo leaks (block sum matches space" 7280 " maps exactly)\n"); 7281 } else if (!dump_opt['L']) { 7282 (void) printf("block traversal size %llu != alloc %llu " 7283 "(%s %lld)\n", 7284 (u_longlong_t)total_found, 7285 (u_longlong_t)total_alloc, 7286 (dump_opt['L']) ? "unreachable" : "leaked", 7287 (longlong_t)(total_alloc - total_found)); 7288 } 7289 7290 if (tzb->zb_count == 0) { 7291 umem_free(zcb, sizeof (zdb_cb_t)); 7292 return (2); 7293 } 7294 7295 (void) printf("\n"); 7296 (void) printf("\t%-16s %14llu\n", "bp count:", 7297 (u_longlong_t)tzb->zb_count); 7298 (void) printf("\t%-16s %14llu\n", "ganged count:", 7299 (longlong_t)tzb->zb_gangs); 7300 (void) printf("\t%-16s %14llu avg: %6llu\n", "bp logical:", 7301 (u_longlong_t)tzb->zb_lsize, 7302 (u_longlong_t)(tzb->zb_lsize / tzb->zb_count)); 7303 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n", 7304 "bp physical:", (u_longlong_t)tzb->zb_psize, 7305 (u_longlong_t)(tzb->zb_psize / tzb->zb_count), 7306 (double)tzb->zb_lsize / tzb->zb_psize); 7307 (void) printf("\t%-16s %14llu avg: %6llu compression: %6.2f\n", 7308 "bp allocated:", (u_longlong_t)tzb->zb_asize, 7309 (u_longlong_t)(tzb->zb_asize / tzb->zb_count), 7310 (double)tzb->zb_lsize / tzb->zb_asize); 7311 (void) printf("\t%-16s %14llu ref>1: %6llu deduplication: %6.2f\n", 7312 "bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize, 7313 (u_longlong_t)zcb->zcb_dedup_blocks, 7314 (double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0); 7315 (void) printf("\t%-16s %14llu count: %6llu\n", 7316 "bp cloned:", (u_longlong_t)zcb->zcb_clone_asize, 7317 (u_longlong_t)zcb->zcb_clone_blocks); 7318 (void) printf("\t%-16s %14llu used: %5.2f%%\n", "Normal class:", 7319 (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space); 7320 7321 if (spa_special_class(spa)->mc_allocator[0].mca_rotor != NULL) { 7322 uint64_t alloc = metaslab_class_get_alloc( 7323 spa_special_class(spa)); 7324 uint64_t space = metaslab_class_get_space( 7325 spa_special_class(spa)); 7326 7327 (void) printf("\t%-16s %14llu used: %5.2f%%\n", 7328 "Special class", (u_longlong_t)alloc, 7329 100.0 * alloc / space); 7330 } 7331 7332 if (spa_dedup_class(spa)->mc_allocator[0].mca_rotor != NULL) { 7333 uint64_t alloc = metaslab_class_get_alloc( 7334 spa_dedup_class(spa)); 7335 uint64_t space = metaslab_class_get_space( 7336 spa_dedup_class(spa)); 7337 7338 (void) printf("\t%-16s %14llu used: %5.2f%%\n", 7339 "Dedup class", (u_longlong_t)alloc, 7340 100.0 * alloc / space); 7341 } 7342 7343 if (spa_embedded_log_class(spa)->mc_allocator[0].mca_rotor != NULL) { 7344 uint64_t alloc = metaslab_class_get_alloc( 7345 spa_embedded_log_class(spa)); 7346 uint64_t space = metaslab_class_get_space( 7347 spa_embedded_log_class(spa)); 7348 7349 (void) printf("\t%-16s %14llu used: %5.2f%%\n", 7350 "Embedded log class", (u_longlong_t)alloc, 7351 100.0 * alloc / space); 7352 } 7353 7354 if (spa_special_embedded_log_class(spa)->mc_allocator[0].mca_rotor 7355 != NULL) { 7356 uint64_t alloc = metaslab_class_get_alloc( 7357 spa_special_embedded_log_class(spa)); 7358 uint64_t space = metaslab_class_get_space( 7359 spa_special_embedded_log_class(spa)); 7360 7361 (void) printf("\t%-16s %14llu used: %5.2f%%\n", 7362 "Special embedded log", (u_longlong_t)alloc, 7363 100.0 * alloc / space); 7364 } 7365 7366 for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) { 7367 if (zcb->zcb_embedded_blocks[i] == 0) 7368 continue; 7369 (void) printf("\n"); 7370 (void) printf("\tadditional, non-pointer bps of type %u: " 7371 "%10llu\n", 7372 i, (u_longlong_t)zcb->zcb_embedded_blocks[i]); 7373 7374 if (dump_opt['b'] >= 3) { 7375 (void) printf("\t number of (compressed) bytes: " 7376 "number of bps\n"); 7377 dump_histogram(zcb->zcb_embedded_histogram[i], 7378 sizeof (zcb->zcb_embedded_histogram[i]) / 7379 sizeof (zcb->zcb_embedded_histogram[i][0]), 0); 7380 } 7381 } 7382 7383 if (tzb->zb_ditto_samevdev != 0) { 7384 (void) printf("\tDittoed blocks on same vdev: %llu\n", 7385 (longlong_t)tzb->zb_ditto_samevdev); 7386 } 7387 if (tzb->zb_ditto_same_ms != 0) { 7388 (void) printf("\tDittoed blocks in same metaslab: %llu\n", 7389 (longlong_t)tzb->zb_ditto_same_ms); 7390 } 7391 7392 for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) { 7393 vdev_t *vd = spa->spa_root_vdev->vdev_child[v]; 7394 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 7395 7396 if (vim == NULL) { 7397 continue; 7398 } 7399 7400 char mem[32]; 7401 zdb_nicenum(vdev_indirect_mapping_num_entries(vim), 7402 mem, vdev_indirect_mapping_size(vim)); 7403 7404 (void) printf("\tindirect vdev id %llu has %llu segments " 7405 "(%s in memory)\n", 7406 (longlong_t)vd->vdev_id, 7407 (longlong_t)vdev_indirect_mapping_num_entries(vim), mem); 7408 } 7409 7410 if (dump_opt['b'] >= 2) { 7411 int l, t, level; 7412 char csize[32], lsize[32], psize[32], asize[32]; 7413 char avg[32], gang[32]; 7414 (void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE" 7415 "\t avg\t comp\t%%Total\tType\n"); 7416 7417 zfs_blkstat_t *mdstats = umem_zalloc(sizeof (zfs_blkstat_t), 7418 UMEM_NOFAIL); 7419 7420 for (t = 0; t <= ZDB_OT_TOTAL; t++) { 7421 const char *typename; 7422 7423 /* make sure nicenum has enough space */ 7424 _Static_assert(sizeof (csize) >= NN_NUMBUF_SZ, 7425 "csize truncated"); 7426 _Static_assert(sizeof (lsize) >= NN_NUMBUF_SZ, 7427 "lsize truncated"); 7428 _Static_assert(sizeof (psize) >= NN_NUMBUF_SZ, 7429 "psize truncated"); 7430 _Static_assert(sizeof (asize) >= NN_NUMBUF_SZ, 7431 "asize truncated"); 7432 _Static_assert(sizeof (avg) >= NN_NUMBUF_SZ, 7433 "avg truncated"); 7434 _Static_assert(sizeof (gang) >= NN_NUMBUF_SZ, 7435 "gang truncated"); 7436 7437 if (t < DMU_OT_NUMTYPES) 7438 typename = dmu_ot[t].ot_name; 7439 else 7440 typename = zdb_ot_extname[t - DMU_OT_NUMTYPES]; 7441 7442 if (zcb->zcb_type[ZB_TOTAL][t].zb_asize == 0) { 7443 (void) printf("%6s\t%5s\t%5s\t%5s" 7444 "\t%5s\t%5s\t%6s\t%s\n", 7445 "-", 7446 "-", 7447 "-", 7448 "-", 7449 "-", 7450 "-", 7451 "-", 7452 typename); 7453 continue; 7454 } 7455 7456 for (l = ZB_TOTAL - 1; l >= -1; l--) { 7457 level = (l == -1 ? ZB_TOTAL : l); 7458 zb = &zcb->zcb_type[level][t]; 7459 7460 if (zb->zb_asize == 0) 7461 continue; 7462 7463 if (level != ZB_TOTAL && t < DMU_OT_NUMTYPES && 7464 (level > 0 || DMU_OT_IS_METADATA(t))) { 7465 mdstats->zb_count += zb->zb_count; 7466 mdstats->zb_lsize += zb->zb_lsize; 7467 mdstats->zb_psize += zb->zb_psize; 7468 mdstats->zb_asize += zb->zb_asize; 7469 mdstats->zb_gangs += zb->zb_gangs; 7470 } 7471 7472 if (dump_opt['b'] < 3 && level != ZB_TOTAL) 7473 continue; 7474 7475 if (level == 0 && zb->zb_asize == 7476 zcb->zcb_type[ZB_TOTAL][t].zb_asize) 7477 continue; 7478 7479 zdb_nicenum(zb->zb_count, csize, 7480 sizeof (csize)); 7481 zdb_nicenum(zb->zb_lsize, lsize, 7482 sizeof (lsize)); 7483 zdb_nicenum(zb->zb_psize, psize, 7484 sizeof (psize)); 7485 zdb_nicenum(zb->zb_asize, asize, 7486 sizeof (asize)); 7487 zdb_nicenum(zb->zb_asize / zb->zb_count, avg, 7488 sizeof (avg)); 7489 zdb_nicenum(zb->zb_gangs, gang, sizeof (gang)); 7490 7491 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s" 7492 "\t%5.2f\t%6.2f\t", 7493 csize, lsize, psize, asize, avg, 7494 (double)zb->zb_lsize / zb->zb_psize, 7495 100.0 * zb->zb_asize / tzb->zb_asize); 7496 7497 if (level == ZB_TOTAL) 7498 (void) printf("%s\n", typename); 7499 else 7500 (void) printf(" L%d %s\n", 7501 level, typename); 7502 7503 if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) { 7504 (void) printf("\t number of ganged " 7505 "blocks: %s\n", gang); 7506 } 7507 7508 if (dump_opt['b'] >= 4) { 7509 (void) printf("psize " 7510 "(in 512-byte sectors): " 7511 "number of blocks\n"); 7512 dump_histogram(zb->zb_psize_histogram, 7513 PSIZE_HISTO_SIZE, 0); 7514 } 7515 } 7516 } 7517 zdb_nicenum(mdstats->zb_count, csize, 7518 sizeof (csize)); 7519 zdb_nicenum(mdstats->zb_lsize, lsize, 7520 sizeof (lsize)); 7521 zdb_nicenum(mdstats->zb_psize, psize, 7522 sizeof (psize)); 7523 zdb_nicenum(mdstats->zb_asize, asize, 7524 sizeof (asize)); 7525 zdb_nicenum(mdstats->zb_asize / mdstats->zb_count, avg, 7526 sizeof (avg)); 7527 zdb_nicenum(mdstats->zb_gangs, gang, sizeof (gang)); 7528 7529 (void) printf("%6s\t%5s\t%5s\t%5s\t%5s" 7530 "\t%5.2f\t%6.2f\t", 7531 csize, lsize, psize, asize, avg, 7532 (double)mdstats->zb_lsize / mdstats->zb_psize, 7533 100.0 * mdstats->zb_asize / tzb->zb_asize); 7534 (void) printf("%s\n", "Metadata Total"); 7535 7536 /* Output a table summarizing block sizes in the pool */ 7537 if (dump_opt['b'] >= 2) { 7538 dump_size_histograms(zcb); 7539 } 7540 7541 umem_free(mdstats, sizeof (zfs_blkstat_t)); 7542 } 7543 7544 (void) printf("\n"); 7545 7546 if (leaks) { 7547 umem_free(zcb, sizeof (zdb_cb_t)); 7548 return (2); 7549 } 7550 7551 if (zcb->zcb_haderrors) { 7552 umem_free(zcb, sizeof (zdb_cb_t)); 7553 return (3); 7554 } 7555 7556 umem_free(zcb, sizeof (zdb_cb_t)); 7557 return (0); 7558 } 7559 7560 typedef struct zdb_ddt_entry { 7561 /* key must be first for ddt_key_compare */ 7562 ddt_key_t zdde_key; 7563 uint64_t zdde_ref_blocks; 7564 uint64_t zdde_ref_lsize; 7565 uint64_t zdde_ref_psize; 7566 uint64_t zdde_ref_dsize; 7567 avl_node_t zdde_node; 7568 } zdb_ddt_entry_t; 7569 7570 static int 7571 zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 7572 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 7573 { 7574 (void) zilog, (void) dnp; 7575 avl_tree_t *t = arg; 7576 avl_index_t where; 7577 zdb_ddt_entry_t *zdde, zdde_search; 7578 7579 if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) || 7580 BP_IS_EMBEDDED(bp)) 7581 return (0); 7582 7583 if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) { 7584 (void) printf("traversing objset %llu, %llu objects, " 7585 "%lu blocks so far\n", 7586 (u_longlong_t)zb->zb_objset, 7587 (u_longlong_t)BP_GET_FILL(bp), 7588 avl_numnodes(t)); 7589 } 7590 7591 if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF || 7592 BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp))) 7593 return (0); 7594 7595 ddt_key_fill(&zdde_search.zdde_key, bp); 7596 7597 zdde = avl_find(t, &zdde_search, &where); 7598 7599 if (zdde == NULL) { 7600 zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL); 7601 zdde->zdde_key = zdde_search.zdde_key; 7602 avl_insert(t, zdde, where); 7603 } 7604 7605 zdde->zdde_ref_blocks += 1; 7606 zdde->zdde_ref_lsize += BP_GET_LSIZE(bp); 7607 zdde->zdde_ref_psize += BP_GET_PSIZE(bp); 7608 zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp); 7609 7610 return (0); 7611 } 7612 7613 static void 7614 dump_simulated_ddt(spa_t *spa) 7615 { 7616 avl_tree_t t; 7617 void *cookie = NULL; 7618 zdb_ddt_entry_t *zdde; 7619 ddt_histogram_t ddh_total = {{{0}}}; 7620 ddt_stat_t dds_total = {0}; 7621 7622 avl_create(&t, ddt_key_compare, 7623 sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node)); 7624 7625 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 7626 7627 (void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | 7628 TRAVERSE_NO_DECRYPT, zdb_ddt_add_cb, &t); 7629 7630 spa_config_exit(spa, SCL_CONFIG, FTAG); 7631 7632 while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) { 7633 uint64_t refcnt = zdde->zdde_ref_blocks; 7634 ASSERT(refcnt != 0); 7635 7636 ddt_stat_t *dds = &ddh_total.ddh_stat[highbit64(refcnt) - 1]; 7637 7638 dds->dds_blocks += zdde->zdde_ref_blocks / refcnt; 7639 dds->dds_lsize += zdde->zdde_ref_lsize / refcnt; 7640 dds->dds_psize += zdde->zdde_ref_psize / refcnt; 7641 dds->dds_dsize += zdde->zdde_ref_dsize / refcnt; 7642 7643 dds->dds_ref_blocks += zdde->zdde_ref_blocks; 7644 dds->dds_ref_lsize += zdde->zdde_ref_lsize; 7645 dds->dds_ref_psize += zdde->zdde_ref_psize; 7646 dds->dds_ref_dsize += zdde->zdde_ref_dsize; 7647 7648 umem_free(zdde, sizeof (*zdde)); 7649 } 7650 7651 avl_destroy(&t); 7652 7653 ddt_histogram_total(&dds_total, &ddh_total); 7654 7655 (void) printf("Simulated DDT histogram:\n"); 7656 7657 zpool_dump_ddt(&dds_total, &ddh_total); 7658 7659 dump_dedup_ratio(&dds_total); 7660 } 7661 7662 static int 7663 verify_device_removal_feature_counts(spa_t *spa) 7664 { 7665 uint64_t dr_feature_refcount = 0; 7666 uint64_t oc_feature_refcount = 0; 7667 uint64_t indirect_vdev_count = 0; 7668 uint64_t precise_vdev_count = 0; 7669 uint64_t obsolete_counts_object_count = 0; 7670 uint64_t obsolete_sm_count = 0; 7671 uint64_t obsolete_counts_count = 0; 7672 uint64_t scip_count = 0; 7673 uint64_t obsolete_bpobj_count = 0; 7674 int ret = 0; 7675 7676 spa_condensing_indirect_phys_t *scip = 7677 &spa->spa_condensing_indirect_phys; 7678 if (scip->scip_next_mapping_object != 0) { 7679 vdev_t *vd = spa->spa_root_vdev->vdev_child[scip->scip_vdev]; 7680 ASSERT(scip->scip_prev_obsolete_sm_object != 0); 7681 ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops); 7682 7683 (void) printf("Condensing indirect vdev %llu: new mapping " 7684 "object %llu, prev obsolete sm %llu\n", 7685 (u_longlong_t)scip->scip_vdev, 7686 (u_longlong_t)scip->scip_next_mapping_object, 7687 (u_longlong_t)scip->scip_prev_obsolete_sm_object); 7688 if (scip->scip_prev_obsolete_sm_object != 0) { 7689 space_map_t *prev_obsolete_sm = NULL; 7690 VERIFY0(space_map_open(&prev_obsolete_sm, 7691 spa->spa_meta_objset, 7692 scip->scip_prev_obsolete_sm_object, 7693 0, vd->vdev_asize, 0)); 7694 dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm); 7695 (void) printf("\n"); 7696 space_map_close(prev_obsolete_sm); 7697 } 7698 7699 scip_count += 2; 7700 } 7701 7702 for (uint64_t i = 0; i < spa->spa_root_vdev->vdev_children; i++) { 7703 vdev_t *vd = spa->spa_root_vdev->vdev_child[i]; 7704 vdev_indirect_config_t *vic = &vd->vdev_indirect_config; 7705 7706 if (vic->vic_mapping_object != 0) { 7707 ASSERT(vd->vdev_ops == &vdev_indirect_ops || 7708 vd->vdev_removing); 7709 indirect_vdev_count++; 7710 7711 if (vd->vdev_indirect_mapping->vim_havecounts) { 7712 obsolete_counts_count++; 7713 } 7714 } 7715 7716 boolean_t are_precise; 7717 VERIFY0(vdev_obsolete_counts_are_precise(vd, &are_precise)); 7718 if (are_precise) { 7719 ASSERT(vic->vic_mapping_object != 0); 7720 precise_vdev_count++; 7721 } 7722 7723 uint64_t obsolete_sm_object; 7724 VERIFY0(vdev_obsolete_sm_object(vd, &obsolete_sm_object)); 7725 if (obsolete_sm_object != 0) { 7726 ASSERT(vic->vic_mapping_object != 0); 7727 obsolete_sm_count++; 7728 } 7729 } 7730 7731 (void) feature_get_refcount(spa, 7732 &spa_feature_table[SPA_FEATURE_DEVICE_REMOVAL], 7733 &dr_feature_refcount); 7734 (void) feature_get_refcount(spa, 7735 &spa_feature_table[SPA_FEATURE_OBSOLETE_COUNTS], 7736 &oc_feature_refcount); 7737 7738 if (dr_feature_refcount != indirect_vdev_count) { 7739 ret = 1; 7740 (void) printf("Number of indirect vdevs (%llu) " \ 7741 "does not match feature count (%llu)\n", 7742 (u_longlong_t)indirect_vdev_count, 7743 (u_longlong_t)dr_feature_refcount); 7744 } else { 7745 (void) printf("Verified device_removal feature refcount " \ 7746 "of %llu is correct\n", 7747 (u_longlong_t)dr_feature_refcount); 7748 } 7749 7750 if (zap_contains(spa_meta_objset(spa), DMU_POOL_DIRECTORY_OBJECT, 7751 DMU_POOL_OBSOLETE_BPOBJ) == 0) { 7752 obsolete_bpobj_count++; 7753 } 7754 7755 7756 obsolete_counts_object_count = precise_vdev_count; 7757 obsolete_counts_object_count += obsolete_sm_count; 7758 obsolete_counts_object_count += obsolete_counts_count; 7759 obsolete_counts_object_count += scip_count; 7760 obsolete_counts_object_count += obsolete_bpobj_count; 7761 obsolete_counts_object_count += remap_deadlist_count; 7762 7763 if (oc_feature_refcount != obsolete_counts_object_count) { 7764 ret = 1; 7765 (void) printf("Number of obsolete counts objects (%llu) " \ 7766 "does not match feature count (%llu)\n", 7767 (u_longlong_t)obsolete_counts_object_count, 7768 (u_longlong_t)oc_feature_refcount); 7769 (void) printf("pv:%llu os:%llu oc:%llu sc:%llu " 7770 "ob:%llu rd:%llu\n", 7771 (u_longlong_t)precise_vdev_count, 7772 (u_longlong_t)obsolete_sm_count, 7773 (u_longlong_t)obsolete_counts_count, 7774 (u_longlong_t)scip_count, 7775 (u_longlong_t)obsolete_bpobj_count, 7776 (u_longlong_t)remap_deadlist_count); 7777 } else { 7778 (void) printf("Verified indirect_refcount feature refcount " \ 7779 "of %llu is correct\n", 7780 (u_longlong_t)oc_feature_refcount); 7781 } 7782 return (ret); 7783 } 7784 7785 static void 7786 zdb_set_skip_mmp(char *target) 7787 { 7788 spa_t *spa; 7789 7790 /* 7791 * Disable the activity check to allow examination of 7792 * active pools. 7793 */ 7794 mutex_enter(&spa_namespace_lock); 7795 if ((spa = spa_lookup(target)) != NULL) { 7796 spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP; 7797 } 7798 mutex_exit(&spa_namespace_lock); 7799 } 7800 7801 #define BOGUS_SUFFIX "_CHECKPOINTED_UNIVERSE" 7802 /* 7803 * Import the checkpointed state of the pool specified by the target 7804 * parameter as readonly. The function also accepts a pool config 7805 * as an optional parameter, else it attempts to infer the config by 7806 * the name of the target pool. 7807 * 7808 * Note that the checkpointed state's pool name will be the name of 7809 * the original pool with the above suffix appended to it. In addition, 7810 * if the target is not a pool name (e.g. a path to a dataset) then 7811 * the new_path parameter is populated with the updated path to 7812 * reflect the fact that we are looking into the checkpointed state. 7813 * 7814 * The function returns a newly-allocated copy of the name of the 7815 * pool containing the checkpointed state. When this copy is no 7816 * longer needed it should be freed with free(3C). Same thing 7817 * applies to the new_path parameter if allocated. 7818 */ 7819 static char * 7820 import_checkpointed_state(char *target, nvlist_t *cfg, boolean_t target_is_spa, 7821 char **new_path) 7822 { 7823 int error = 0; 7824 char *poolname, *bogus_name = NULL; 7825 boolean_t freecfg = B_FALSE; 7826 7827 /* If the target is not a pool, the extract the pool name */ 7828 char *path_start = strchr(target, '/'); 7829 if (target_is_spa || path_start == NULL) { 7830 poolname = target; 7831 } else { 7832 size_t poolname_len = path_start - target; 7833 poolname = strndup(target, poolname_len); 7834 } 7835 7836 if (cfg == NULL) { 7837 zdb_set_skip_mmp(poolname); 7838 error = spa_get_stats(poolname, &cfg, NULL, 0); 7839 if (error != 0) { 7840 fatal("Tried to read config of pool \"%s\" but " 7841 "spa_get_stats() failed with error %d\n", 7842 poolname, error); 7843 } 7844 freecfg = B_TRUE; 7845 } 7846 7847 if (asprintf(&bogus_name, "%s%s", poolname, BOGUS_SUFFIX) == -1) { 7848 if (target != poolname) 7849 free(poolname); 7850 return (NULL); 7851 } 7852 fnvlist_add_string(cfg, ZPOOL_CONFIG_POOL_NAME, bogus_name); 7853 7854 error = spa_import(bogus_name, cfg, NULL, 7855 ZFS_IMPORT_MISSING_LOG | ZFS_IMPORT_CHECKPOINT | 7856 ZFS_IMPORT_SKIP_MMP); 7857 if (freecfg) 7858 nvlist_free(cfg); 7859 if (error != 0) { 7860 fatal("Tried to import pool \"%s\" but spa_import() failed " 7861 "with error %d\n", bogus_name, error); 7862 } 7863 7864 if (new_path != NULL && !target_is_spa) { 7865 if (asprintf(new_path, "%s%s", bogus_name, 7866 path_start != NULL ? path_start : "") == -1) { 7867 free(bogus_name); 7868 if (!target_is_spa && path_start != NULL) 7869 free(poolname); 7870 return (NULL); 7871 } 7872 } 7873 7874 if (target != poolname) 7875 free(poolname); 7876 7877 return (bogus_name); 7878 } 7879 7880 typedef struct verify_checkpoint_sm_entry_cb_arg { 7881 vdev_t *vcsec_vd; 7882 7883 /* the following fields are only used for printing progress */ 7884 uint64_t vcsec_entryid; 7885 uint64_t vcsec_num_entries; 7886 } verify_checkpoint_sm_entry_cb_arg_t; 7887 7888 #define ENTRIES_PER_PROGRESS_UPDATE 10000 7889 7890 static int 7891 verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, void *arg) 7892 { 7893 verify_checkpoint_sm_entry_cb_arg_t *vcsec = arg; 7894 vdev_t *vd = vcsec->vcsec_vd; 7895 metaslab_t *ms = vd->vdev_ms[sme->sme_offset >> vd->vdev_ms_shift]; 7896 uint64_t end = sme->sme_offset + sme->sme_run; 7897 7898 ASSERT(sme->sme_type == SM_FREE); 7899 7900 if ((vcsec->vcsec_entryid % ENTRIES_PER_PROGRESS_UPDATE) == 0) { 7901 (void) fprintf(stderr, 7902 "\rverifying vdev %llu, space map entry %llu of %llu ...", 7903 (longlong_t)vd->vdev_id, 7904 (longlong_t)vcsec->vcsec_entryid, 7905 (longlong_t)vcsec->vcsec_num_entries); 7906 } 7907 vcsec->vcsec_entryid++; 7908 7909 /* 7910 * See comment in checkpoint_sm_exclude_entry_cb() 7911 */ 7912 VERIFY3U(sme->sme_offset, >=, ms->ms_start); 7913 VERIFY3U(end, <=, ms->ms_start + ms->ms_size); 7914 7915 /* 7916 * The entries in the vdev_checkpoint_sm should be marked as 7917 * allocated in the checkpointed state of the pool, therefore 7918 * their respective ms_allocateable trees should not contain them. 7919 */ 7920 mutex_enter(&ms->ms_lock); 7921 zfs_range_tree_verify_not_present(ms->ms_allocatable, 7922 sme->sme_offset, sme->sme_run); 7923 mutex_exit(&ms->ms_lock); 7924 7925 return (0); 7926 } 7927 7928 /* 7929 * Verify that all segments in the vdev_checkpoint_sm are allocated 7930 * according to the checkpoint's ms_sm (i.e. are not in the checkpoint's 7931 * ms_allocatable). 7932 * 7933 * Do so by comparing the checkpoint space maps (vdev_checkpoint_sm) of 7934 * each vdev in the current state of the pool to the metaslab space maps 7935 * (ms_sm) of the checkpointed state of the pool. 7936 * 7937 * Note that the function changes the state of the ms_allocatable 7938 * trees of the current spa_t. The entries of these ms_allocatable 7939 * trees are cleared out and then repopulated from with the free 7940 * entries of their respective ms_sm space maps. 7941 */ 7942 static void 7943 verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, spa_t *current) 7944 { 7945 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev; 7946 vdev_t *current_rvd = current->spa_root_vdev; 7947 7948 load_concrete_ms_allocatable_trees(checkpoint, SM_FREE); 7949 7950 for (uint64_t c = 0; c < ckpoint_rvd->vdev_children; c++) { 7951 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[c]; 7952 vdev_t *current_vd = current_rvd->vdev_child[c]; 7953 7954 space_map_t *checkpoint_sm = NULL; 7955 uint64_t checkpoint_sm_obj; 7956 7957 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) { 7958 /* 7959 * Since we don't allow device removal in a pool 7960 * that has a checkpoint, we expect that all removed 7961 * vdevs were removed from the pool before the 7962 * checkpoint. 7963 */ 7964 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops); 7965 continue; 7966 } 7967 7968 /* 7969 * If the checkpoint space map doesn't exist, then nothing 7970 * here is checkpointed so there's nothing to verify. 7971 */ 7972 if (current_vd->vdev_top_zap == 0 || 7973 zap_contains(spa_meta_objset(current), 7974 current_vd->vdev_top_zap, 7975 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0) 7976 continue; 7977 7978 VERIFY0(zap_lookup(spa_meta_objset(current), 7979 current_vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, 7980 sizeof (uint64_t), 1, &checkpoint_sm_obj)); 7981 7982 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current), 7983 checkpoint_sm_obj, 0, current_vd->vdev_asize, 7984 current_vd->vdev_ashift)); 7985 7986 verify_checkpoint_sm_entry_cb_arg_t vcsec; 7987 vcsec.vcsec_vd = ckpoint_vd; 7988 vcsec.vcsec_entryid = 0; 7989 vcsec.vcsec_num_entries = 7990 space_map_length(checkpoint_sm) / sizeof (uint64_t); 7991 VERIFY0(space_map_iterate(checkpoint_sm, 7992 space_map_length(checkpoint_sm), 7993 verify_checkpoint_sm_entry_cb, &vcsec)); 7994 if (dump_opt['m'] > 3) 7995 dump_spacemap(current->spa_meta_objset, checkpoint_sm); 7996 space_map_close(checkpoint_sm); 7997 } 7998 7999 /* 8000 * If we've added vdevs since we took the checkpoint, ensure 8001 * that their checkpoint space maps are empty. 8002 */ 8003 if (ckpoint_rvd->vdev_children < current_rvd->vdev_children) { 8004 for (uint64_t c = ckpoint_rvd->vdev_children; 8005 c < current_rvd->vdev_children; c++) { 8006 vdev_t *current_vd = current_rvd->vdev_child[c]; 8007 VERIFY0P(current_vd->vdev_checkpoint_sm); 8008 } 8009 } 8010 8011 /* for cleaner progress output */ 8012 (void) fprintf(stderr, "\n"); 8013 } 8014 8015 /* 8016 * Verifies that all space that's allocated in the checkpoint is 8017 * still allocated in the current version, by checking that everything 8018 * in checkpoint's ms_allocatable (which is actually allocated, not 8019 * allocatable/free) is not present in current's ms_allocatable. 8020 * 8021 * Note that the function changes the state of the ms_allocatable 8022 * trees of both spas when called. The entries of all ms_allocatable 8023 * trees are cleared out and then repopulated from their respective 8024 * ms_sm space maps. In the checkpointed state we load the allocated 8025 * entries, and in the current state we load the free entries. 8026 */ 8027 static void 8028 verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_t *current) 8029 { 8030 vdev_t *ckpoint_rvd = checkpoint->spa_root_vdev; 8031 vdev_t *current_rvd = current->spa_root_vdev; 8032 8033 load_concrete_ms_allocatable_trees(checkpoint, SM_ALLOC); 8034 load_concrete_ms_allocatable_trees(current, SM_FREE); 8035 8036 for (uint64_t i = 0; i < ckpoint_rvd->vdev_children; i++) { 8037 vdev_t *ckpoint_vd = ckpoint_rvd->vdev_child[i]; 8038 vdev_t *current_vd = current_rvd->vdev_child[i]; 8039 8040 if (ckpoint_vd->vdev_ops == &vdev_indirect_ops) { 8041 /* 8042 * See comment in verify_checkpoint_vdev_spacemaps() 8043 */ 8044 ASSERT3P(current_vd->vdev_ops, ==, &vdev_indirect_ops); 8045 continue; 8046 } 8047 8048 for (uint64_t m = 0; m < ckpoint_vd->vdev_ms_count; m++) { 8049 metaslab_t *ckpoint_msp = ckpoint_vd->vdev_ms[m]; 8050 metaslab_t *current_msp = current_vd->vdev_ms[m]; 8051 8052 (void) fprintf(stderr, 8053 "\rverifying vdev %llu of %llu, " 8054 "metaslab %llu of %llu ...", 8055 (longlong_t)current_vd->vdev_id, 8056 (longlong_t)current_rvd->vdev_children, 8057 (longlong_t)current_vd->vdev_ms[m]->ms_id, 8058 (longlong_t)current_vd->vdev_ms_count); 8059 8060 /* 8061 * We walk through the ms_allocatable trees that 8062 * are loaded with the allocated blocks from the 8063 * ms_sm spacemaps of the checkpoint. For each 8064 * one of these ranges we ensure that none of them 8065 * exists in the ms_allocatable trees of the 8066 * current state which are loaded with the ranges 8067 * that are currently free. 8068 * 8069 * This way we ensure that none of the blocks that 8070 * are part of the checkpoint were freed by mistake. 8071 */ 8072 zfs_range_tree_walk(ckpoint_msp->ms_allocatable, 8073 (zfs_range_tree_func_t *) 8074 zfs_range_tree_verify_not_present, 8075 current_msp->ms_allocatable); 8076 } 8077 } 8078 8079 /* for cleaner progress output */ 8080 (void) fprintf(stderr, "\n"); 8081 } 8082 8083 static void 8084 verify_checkpoint_blocks(spa_t *spa) 8085 { 8086 ASSERT(!dump_opt['L']); 8087 8088 spa_t *checkpoint_spa; 8089 char *checkpoint_pool; 8090 int error = 0; 8091 8092 /* 8093 * We import the checkpointed state of the pool (under a different 8094 * name) so we can do verification on it against the current state 8095 * of the pool. 8096 */ 8097 checkpoint_pool = import_checkpointed_state(spa->spa_name, NULL, B_TRUE, 8098 NULL); 8099 ASSERT(strcmp(spa->spa_name, checkpoint_pool) != 0); 8100 8101 error = spa_open(checkpoint_pool, &checkpoint_spa, FTAG); 8102 if (error != 0) { 8103 fatal("Tried to open pool \"%s\" but spa_open() failed with " 8104 "error %d\n", checkpoint_pool, error); 8105 } 8106 8107 /* 8108 * Ensure that ranges in the checkpoint space maps of each vdev 8109 * are allocated according to the checkpointed state's metaslab 8110 * space maps. 8111 */ 8112 verify_checkpoint_vdev_spacemaps(checkpoint_spa, spa); 8113 8114 /* 8115 * Ensure that allocated ranges in the checkpoint's metaslab 8116 * space maps remain allocated in the metaslab space maps of 8117 * the current state. 8118 */ 8119 verify_checkpoint_ms_spacemaps(checkpoint_spa, spa); 8120 8121 /* 8122 * Once we are done, we get rid of the checkpointed state. 8123 */ 8124 spa_close(checkpoint_spa, FTAG); 8125 free(checkpoint_pool); 8126 } 8127 8128 static void 8129 dump_leftover_checkpoint_blocks(spa_t *spa) 8130 { 8131 vdev_t *rvd = spa->spa_root_vdev; 8132 8133 for (uint64_t i = 0; i < rvd->vdev_children; i++) { 8134 vdev_t *vd = rvd->vdev_child[i]; 8135 8136 space_map_t *checkpoint_sm = NULL; 8137 uint64_t checkpoint_sm_obj; 8138 8139 if (vd->vdev_top_zap == 0) 8140 continue; 8141 8142 if (zap_contains(spa_meta_objset(spa), vd->vdev_top_zap, 8143 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM) != 0) 8144 continue; 8145 8146 VERIFY0(zap_lookup(spa_meta_objset(spa), vd->vdev_top_zap, 8147 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, 8148 sizeof (uint64_t), 1, &checkpoint_sm_obj)); 8149 8150 VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa), 8151 checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift)); 8152 dump_spacemap(spa->spa_meta_objset, checkpoint_sm); 8153 space_map_close(checkpoint_sm); 8154 } 8155 } 8156 8157 static int 8158 verify_checkpoint(spa_t *spa) 8159 { 8160 uberblock_t checkpoint; 8161 int error; 8162 8163 if (!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) 8164 return (0); 8165 8166 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 8167 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), 8168 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); 8169 8170 if (error == ENOENT && !dump_opt['L']) { 8171 /* 8172 * If the feature is active but the uberblock is missing 8173 * then we must be in the middle of discarding the 8174 * checkpoint. 8175 */ 8176 (void) printf("\nPartially discarded checkpoint " 8177 "state found:\n"); 8178 if (dump_opt['m'] > 3) 8179 dump_leftover_checkpoint_blocks(spa); 8180 return (0); 8181 } else if (error != 0) { 8182 (void) printf("lookup error %d when looking for " 8183 "checkpointed uberblock in MOS\n", error); 8184 return (error); 8185 } 8186 dump_uberblock(&checkpoint, "\nCheckpointed uberblock found:\n", "\n"); 8187 8188 if (checkpoint.ub_checkpoint_txg == 0) { 8189 (void) printf("\nub_checkpoint_txg not set in checkpointed " 8190 "uberblock\n"); 8191 error = 3; 8192 } 8193 8194 if (error == 0 && !dump_opt['L']) 8195 verify_checkpoint_blocks(spa); 8196 8197 return (error); 8198 } 8199 8200 static void 8201 mos_leaks_cb(void *arg, uint64_t start, uint64_t size) 8202 { 8203 (void) arg; 8204 for (uint64_t i = start; i < size; i++) { 8205 (void) printf("MOS object %llu referenced but not allocated\n", 8206 (u_longlong_t)i); 8207 } 8208 } 8209 8210 static void 8211 mos_obj_refd(uint64_t obj) 8212 { 8213 if (obj != 0 && mos_refd_objs != NULL) 8214 zfs_range_tree_add(mos_refd_objs, obj, 1); 8215 } 8216 8217 /* 8218 * Call on a MOS object that may already have been referenced. 8219 */ 8220 static void 8221 mos_obj_refd_multiple(uint64_t obj) 8222 { 8223 if (obj != 0 && mos_refd_objs != NULL && 8224 !zfs_range_tree_contains(mos_refd_objs, obj, 1)) 8225 zfs_range_tree_add(mos_refd_objs, obj, 1); 8226 } 8227 8228 static void 8229 mos_leak_vdev_top_zap(vdev_t *vd) 8230 { 8231 uint64_t ms_flush_data_obj; 8232 int error = zap_lookup(spa_meta_objset(vd->vdev_spa), 8233 vd->vdev_top_zap, VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS, 8234 sizeof (ms_flush_data_obj), 1, &ms_flush_data_obj); 8235 if (error == ENOENT) 8236 return; 8237 ASSERT0(error); 8238 8239 mos_obj_refd(ms_flush_data_obj); 8240 } 8241 8242 static void 8243 mos_leak_vdev(vdev_t *vd) 8244 { 8245 mos_obj_refd(vd->vdev_dtl_object); 8246 mos_obj_refd(vd->vdev_ms_array); 8247 mos_obj_refd(vd->vdev_indirect_config.vic_births_object); 8248 mos_obj_refd(vd->vdev_indirect_config.vic_mapping_object); 8249 mos_obj_refd(vd->vdev_leaf_zap); 8250 if (vd->vdev_checkpoint_sm != NULL) 8251 mos_obj_refd(vd->vdev_checkpoint_sm->sm_object); 8252 if (vd->vdev_indirect_mapping != NULL) { 8253 mos_obj_refd(vd->vdev_indirect_mapping-> 8254 vim_phys->vimp_counts_object); 8255 } 8256 if (vd->vdev_obsolete_sm != NULL) 8257 mos_obj_refd(vd->vdev_obsolete_sm->sm_object); 8258 8259 for (uint64_t m = 0; m < vd->vdev_ms_count; m++) { 8260 metaslab_t *ms = vd->vdev_ms[m]; 8261 mos_obj_refd(space_map_object(ms->ms_sm)); 8262 } 8263 8264 if (vd->vdev_root_zap != 0) 8265 mos_obj_refd(vd->vdev_root_zap); 8266 8267 if (vd->vdev_top_zap != 0) { 8268 mos_obj_refd(vd->vdev_top_zap); 8269 mos_leak_vdev_top_zap(vd); 8270 } 8271 8272 for (uint64_t c = 0; c < vd->vdev_children; c++) { 8273 mos_leak_vdev(vd->vdev_child[c]); 8274 } 8275 } 8276 8277 static void 8278 mos_leak_log_spacemaps(spa_t *spa) 8279 { 8280 uint64_t spacemap_zap; 8281 int error = zap_lookup(spa_meta_objset(spa), 8282 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_LOG_SPACEMAP_ZAP, 8283 sizeof (spacemap_zap), 1, &spacemap_zap); 8284 if (error == ENOENT) 8285 return; 8286 ASSERT0(error); 8287 8288 mos_obj_refd(spacemap_zap); 8289 for (spa_log_sm_t *sls = avl_first(&spa->spa_sm_logs_by_txg); 8290 sls; sls = AVL_NEXT(&spa->spa_sm_logs_by_txg, sls)) 8291 mos_obj_refd(sls->sls_sm_obj); 8292 } 8293 8294 static void 8295 errorlog_count_refd(objset_t *mos, uint64_t errlog) 8296 { 8297 zap_cursor_t zc; 8298 zap_attribute_t *za = zap_attribute_alloc(); 8299 for (zap_cursor_init(&zc, mos, errlog); 8300 zap_cursor_retrieve(&zc, za) == 0; 8301 zap_cursor_advance(&zc)) { 8302 mos_obj_refd(za->za_first_integer); 8303 } 8304 zap_cursor_fini(&zc); 8305 zap_attribute_free(za); 8306 } 8307 8308 static int 8309 dump_mos_leaks(spa_t *spa) 8310 { 8311 int rv = 0; 8312 objset_t *mos = spa->spa_meta_objset; 8313 dsl_pool_t *dp = spa->spa_dsl_pool; 8314 8315 /* Visit and mark all referenced objects in the MOS */ 8316 8317 mos_obj_refd(DMU_POOL_DIRECTORY_OBJECT); 8318 mos_obj_refd(spa->spa_pool_props_object); 8319 mos_obj_refd(spa->spa_config_object); 8320 mos_obj_refd(spa->spa_ddt_stat_object); 8321 mos_obj_refd(spa->spa_feat_desc_obj); 8322 mos_obj_refd(spa->spa_feat_enabled_txg_obj); 8323 mos_obj_refd(spa->spa_feat_for_read_obj); 8324 mos_obj_refd(spa->spa_feat_for_write_obj); 8325 mos_obj_refd(spa->spa_history); 8326 mos_obj_refd(spa->spa_errlog_last); 8327 mos_obj_refd(spa->spa_errlog_scrub); 8328 8329 if (spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG)) { 8330 errorlog_count_refd(mos, spa->spa_errlog_last); 8331 errorlog_count_refd(mos, spa->spa_errlog_scrub); 8332 } 8333 8334 mos_obj_refd(spa->spa_all_vdev_zaps); 8335 mos_obj_refd(spa->spa_dsl_pool->dp_bptree_obj); 8336 mos_obj_refd(spa->spa_dsl_pool->dp_tmp_userrefs_obj); 8337 mos_obj_refd(spa->spa_dsl_pool->dp_scan->scn_phys.scn_queue_obj); 8338 bpobj_count_refd(&spa->spa_deferred_bpobj); 8339 mos_obj_refd(dp->dp_empty_bpobj); 8340 bpobj_count_refd(&dp->dp_obsolete_bpobj); 8341 bpobj_count_refd(&dp->dp_free_bpobj); 8342 mos_obj_refd(spa->spa_l2cache.sav_object); 8343 mos_obj_refd(spa->spa_spares.sav_object); 8344 8345 if (spa->spa_syncing_log_sm != NULL) 8346 mos_obj_refd(spa->spa_syncing_log_sm->sm_object); 8347 mos_leak_log_spacemaps(spa); 8348 8349 mos_obj_refd(spa->spa_condensing_indirect_phys. 8350 scip_next_mapping_object); 8351 mos_obj_refd(spa->spa_condensing_indirect_phys. 8352 scip_prev_obsolete_sm_object); 8353 if (spa->spa_condensing_indirect_phys.scip_next_mapping_object != 0) { 8354 vdev_indirect_mapping_t *vim = 8355 vdev_indirect_mapping_open(mos, 8356 spa->spa_condensing_indirect_phys.scip_next_mapping_object); 8357 mos_obj_refd(vim->vim_phys->vimp_counts_object); 8358 vdev_indirect_mapping_close(vim); 8359 } 8360 deleted_livelists_dump_mos(spa); 8361 8362 if (dp->dp_origin_snap != NULL) { 8363 dsl_dataset_t *ds; 8364 8365 dsl_pool_config_enter(dp, FTAG); 8366 VERIFY0(dsl_dataset_hold_obj(dp, 8367 dsl_dataset_phys(dp->dp_origin_snap)->ds_next_snap_obj, 8368 FTAG, &ds)); 8369 count_ds_mos_objects(ds); 8370 dump_blkptr_list(&ds->ds_deadlist, "Deadlist"); 8371 dsl_dataset_rele(ds, FTAG); 8372 dsl_pool_config_exit(dp, FTAG); 8373 8374 count_ds_mos_objects(dp->dp_origin_snap); 8375 dump_blkptr_list(&dp->dp_origin_snap->ds_deadlist, "Deadlist"); 8376 } 8377 count_dir_mos_objects(dp->dp_mos_dir); 8378 if (dp->dp_free_dir != NULL) 8379 count_dir_mos_objects(dp->dp_free_dir); 8380 if (dp->dp_leak_dir != NULL) 8381 count_dir_mos_objects(dp->dp_leak_dir); 8382 8383 mos_leak_vdev(spa->spa_root_vdev); 8384 8385 for (uint64_t c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { 8386 ddt_t *ddt = spa->spa_ddt[c]; 8387 if (!ddt || ddt->ddt_version == DDT_VERSION_UNCONFIGURED) 8388 continue; 8389 8390 /* DDT store objects */ 8391 for (ddt_type_t type = 0; type < DDT_TYPES; type++) { 8392 for (ddt_class_t class = 0; class < DDT_CLASSES; 8393 class++) { 8394 mos_obj_refd(ddt->ddt_object[type][class]); 8395 } 8396 } 8397 8398 /* FDT container */ 8399 if (ddt->ddt_version == DDT_VERSION_FDT) 8400 mos_obj_refd(ddt->ddt_dir_object); 8401 8402 /* FDT log objects */ 8403 if (ddt->ddt_flags & DDT_FLAG_LOG) { 8404 mos_obj_refd(ddt->ddt_log[0].ddl_object); 8405 mos_obj_refd(ddt->ddt_log[1].ddl_object); 8406 } 8407 } 8408 8409 for (uint64_t vdevid = 0; vdevid < spa->spa_brt_nvdevs; vdevid++) { 8410 brt_vdev_t *brtvd = spa->spa_brt_vdevs[vdevid]; 8411 if (brtvd->bv_initiated) { 8412 mos_obj_refd(brtvd->bv_mos_brtvdev); 8413 mos_obj_refd(brtvd->bv_mos_entries); 8414 } 8415 } 8416 8417 /* 8418 * Visit all allocated objects and make sure they are referenced. 8419 */ 8420 uint64_t object = 0; 8421 while (dmu_object_next(mos, &object, B_FALSE, 0) == 0) { 8422 if (zfs_range_tree_contains(mos_refd_objs, object, 1)) { 8423 zfs_range_tree_remove(mos_refd_objs, object, 1); 8424 } else { 8425 dmu_object_info_t doi; 8426 const char *name; 8427 VERIFY0(dmu_object_info(mos, object, &doi)); 8428 if (doi.doi_type & DMU_OT_NEWTYPE) { 8429 dmu_object_byteswap_t bswap = 8430 DMU_OT_BYTESWAP(doi.doi_type); 8431 name = dmu_ot_byteswap[bswap].ob_name; 8432 } else { 8433 name = dmu_ot[doi.doi_type].ot_name; 8434 } 8435 8436 (void) printf("MOS object %llu (%s) leaked\n", 8437 (u_longlong_t)object, name); 8438 rv = 2; 8439 } 8440 } 8441 (void) zfs_range_tree_walk(mos_refd_objs, mos_leaks_cb, NULL); 8442 if (!zfs_range_tree_is_empty(mos_refd_objs)) 8443 rv = 2; 8444 zfs_range_tree_vacate(mos_refd_objs, NULL, NULL); 8445 zfs_range_tree_destroy(mos_refd_objs); 8446 return (rv); 8447 } 8448 8449 typedef struct log_sm_obsolete_stats_arg { 8450 uint64_t lsos_current_txg; 8451 8452 uint64_t lsos_total_entries; 8453 uint64_t lsos_valid_entries; 8454 8455 uint64_t lsos_sm_entries; 8456 uint64_t lsos_valid_sm_entries; 8457 } log_sm_obsolete_stats_arg_t; 8458 8459 static int 8460 log_spacemap_obsolete_stats_cb(spa_t *spa, space_map_entry_t *sme, 8461 uint64_t txg, void *arg) 8462 { 8463 log_sm_obsolete_stats_arg_t *lsos = arg; 8464 8465 uint64_t offset = sme->sme_offset; 8466 uint64_t vdev_id = sme->sme_vdev; 8467 8468 if (lsos->lsos_current_txg == 0) { 8469 /* this is the first log */ 8470 lsos->lsos_current_txg = txg; 8471 } else if (lsos->lsos_current_txg < txg) { 8472 /* we just changed log - print stats and reset */ 8473 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n", 8474 (u_longlong_t)lsos->lsos_valid_sm_entries, 8475 (u_longlong_t)lsos->lsos_sm_entries, 8476 (u_longlong_t)lsos->lsos_current_txg); 8477 lsos->lsos_valid_sm_entries = 0; 8478 lsos->lsos_sm_entries = 0; 8479 lsos->lsos_current_txg = txg; 8480 } 8481 ASSERT3U(lsos->lsos_current_txg, ==, txg); 8482 8483 lsos->lsos_sm_entries++; 8484 lsos->lsos_total_entries++; 8485 8486 vdev_t *vd = vdev_lookup_top(spa, vdev_id); 8487 if (!vdev_is_concrete(vd)) 8488 return (0); 8489 8490 metaslab_t *ms = vd->vdev_ms[offset >> vd->vdev_ms_shift]; 8491 ASSERT(sme->sme_type == SM_ALLOC || sme->sme_type == SM_FREE); 8492 8493 if (txg < metaslab_unflushed_txg(ms)) 8494 return (0); 8495 lsos->lsos_valid_sm_entries++; 8496 lsos->lsos_valid_entries++; 8497 return (0); 8498 } 8499 8500 static void 8501 dump_log_spacemap_obsolete_stats(spa_t *spa) 8502 { 8503 if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP)) 8504 return; 8505 8506 log_sm_obsolete_stats_arg_t lsos = {0}; 8507 8508 (void) printf("Log Space Map Obsolete Entry Statistics:\n"); 8509 8510 iterate_through_spacemap_logs(spa, 8511 log_spacemap_obsolete_stats_cb, &lsos); 8512 8513 /* print stats for latest log */ 8514 (void) printf("%-8llu valid entries out of %-8llu - txg %llu\n", 8515 (u_longlong_t)lsos.lsos_valid_sm_entries, 8516 (u_longlong_t)lsos.lsos_sm_entries, 8517 (u_longlong_t)lsos.lsos_current_txg); 8518 8519 (void) printf("%-8llu valid entries out of %-8llu - total\n\n", 8520 (u_longlong_t)lsos.lsos_valid_entries, 8521 (u_longlong_t)lsos.lsos_total_entries); 8522 } 8523 8524 static void 8525 dump_zpool(spa_t *spa) 8526 { 8527 dsl_pool_t *dp = spa_get_dsl(spa); 8528 int rc = 0; 8529 8530 if (dump_opt['y']) { 8531 livelist_metaslab_validate(spa); 8532 } 8533 8534 if (dump_opt['S']) { 8535 dump_simulated_ddt(spa); 8536 return; 8537 } 8538 8539 if (!dump_opt['e'] && dump_opt['C'] > 1) { 8540 (void) printf("\nCached configuration:\n"); 8541 dump_nvlist(spa->spa_config, 8); 8542 } 8543 8544 if (dump_opt['C']) 8545 dump_config(spa); 8546 8547 if (dump_opt['u']) 8548 dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n"); 8549 8550 if (dump_opt['D']) 8551 dump_all_ddts(spa); 8552 8553 if (dump_opt['T']) 8554 dump_brt(spa); 8555 8556 if (dump_opt['d'] > 2 || dump_opt['m']) 8557 dump_metaslabs(spa); 8558 if (dump_opt['M']) 8559 dump_metaslab_groups(spa, dump_opt['M'] > 1); 8560 if (dump_opt['d'] > 2 || dump_opt['m']) { 8561 dump_log_spacemaps(spa); 8562 dump_log_spacemap_obsolete_stats(spa); 8563 } 8564 8565 if (dump_opt['d'] || dump_opt['i']) { 8566 spa_feature_t f; 8567 mos_refd_objs = zfs_range_tree_create_flags( 8568 NULL, ZFS_RANGE_SEG64, NULL, 0, 0, 8569 0, "dump_zpool:mos_refd_objs"); 8570 dump_objset(dp->dp_meta_objset); 8571 8572 if (dump_opt['d'] >= 3) { 8573 dsl_pool_t *dp = spa->spa_dsl_pool; 8574 dump_full_bpobj(&spa->spa_deferred_bpobj, 8575 "Deferred frees", 0); 8576 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { 8577 dump_full_bpobj(&dp->dp_free_bpobj, 8578 "Pool snapshot frees", 0); 8579 } 8580 if (bpobj_is_open(&dp->dp_obsolete_bpobj)) { 8581 ASSERT(spa_feature_is_enabled(spa, 8582 SPA_FEATURE_DEVICE_REMOVAL)); 8583 dump_full_bpobj(&dp->dp_obsolete_bpobj, 8584 "Pool obsolete blocks", 0); 8585 } 8586 8587 if (spa_feature_is_active(spa, 8588 SPA_FEATURE_ASYNC_DESTROY)) { 8589 dump_bptree(spa->spa_meta_objset, 8590 dp->dp_bptree_obj, 8591 "Pool dataset frees"); 8592 } 8593 dump_dtl(spa->spa_root_vdev, 0); 8594 } 8595 8596 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) 8597 global_feature_count[f] = UINT64_MAX; 8598 global_feature_count[SPA_FEATURE_REDACTION_BOOKMARKS] = 0; 8599 global_feature_count[SPA_FEATURE_REDACTION_LIST_SPILL] = 0; 8600 global_feature_count[SPA_FEATURE_BOOKMARK_WRITTEN] = 0; 8601 global_feature_count[SPA_FEATURE_LIVELIST] = 0; 8602 8603 (void) dmu_objset_find(spa_name(spa), dump_one_objset, 8604 NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 8605 8606 if (rc == 0 && !dump_opt['L']) 8607 rc = dump_mos_leaks(spa); 8608 8609 for (f = 0; f < SPA_FEATURES; f++) { 8610 uint64_t refcount; 8611 8612 uint64_t *arr; 8613 if (!(spa_feature_table[f].fi_flags & 8614 ZFEATURE_FLAG_PER_DATASET)) { 8615 if (global_feature_count[f] == UINT64_MAX) 8616 continue; 8617 if (!spa_feature_is_enabled(spa, f)) { 8618 ASSERT0(global_feature_count[f]); 8619 continue; 8620 } 8621 arr = global_feature_count; 8622 } else { 8623 if (!spa_feature_is_enabled(spa, f)) { 8624 ASSERT0(dataset_feature_count[f]); 8625 continue; 8626 } 8627 arr = dataset_feature_count; 8628 } 8629 if (feature_get_refcount(spa, &spa_feature_table[f], 8630 &refcount) == ENOTSUP) 8631 continue; 8632 if (arr[f] != refcount) { 8633 (void) printf("%s feature refcount mismatch: " 8634 "%lld consumers != %lld refcount\n", 8635 spa_feature_table[f].fi_uname, 8636 (longlong_t)arr[f], (longlong_t)refcount); 8637 rc = 2; 8638 } else { 8639 (void) printf("Verified %s feature refcount " 8640 "of %llu is correct\n", 8641 spa_feature_table[f].fi_uname, 8642 (longlong_t)refcount); 8643 } 8644 } 8645 8646 if (rc == 0) 8647 rc = verify_device_removal_feature_counts(spa); 8648 } 8649 8650 if (rc == 0 && (dump_opt['b'] || dump_opt['c'])) 8651 rc = dump_block_stats(spa); 8652 8653 if (rc == 0) 8654 rc = verify_spacemap_refcounts(spa); 8655 8656 if (dump_opt['s']) 8657 show_pool_stats(spa); 8658 8659 if (dump_opt['h']) 8660 dump_history(spa); 8661 8662 if (rc == 0) 8663 rc = verify_checkpoint(spa); 8664 8665 if (rc != 0) { 8666 dump_debug_buffer(); 8667 zdb_exit(rc); 8668 } 8669 } 8670 8671 #define ZDB_FLAG_CHECKSUM 0x0001 8672 #define ZDB_FLAG_DECOMPRESS 0x0002 8673 #define ZDB_FLAG_BSWAP 0x0004 8674 #define ZDB_FLAG_GBH 0x0008 8675 #define ZDB_FLAG_INDIRECT 0x0010 8676 #define ZDB_FLAG_RAW 0x0020 8677 #define ZDB_FLAG_PRINT_BLKPTR 0x0040 8678 #define ZDB_FLAG_VERBOSE 0x0080 8679 8680 static int flagbits[256]; 8681 static char flagbitstr[16]; 8682 8683 static void 8684 zdb_print_blkptr(const blkptr_t *bp, int flags) 8685 { 8686 char blkbuf[BP_SPRINTF_LEN]; 8687 8688 if (flags & ZDB_FLAG_BSWAP) 8689 byteswap_uint64_array((void *)bp, sizeof (blkptr_t)); 8690 8691 snprintf_blkptr(blkbuf, sizeof (blkbuf), bp); 8692 (void) printf("%s\n", blkbuf); 8693 } 8694 8695 static void 8696 zdb_dump_indirect(blkptr_t *bp, int nbps, int flags) 8697 { 8698 int i; 8699 8700 for (i = 0; i < nbps; i++) 8701 zdb_print_blkptr(&bp[i], flags); 8702 } 8703 8704 static void 8705 zdb_dump_gbh(void *buf, uint64_t size, int flags) 8706 { 8707 zdb_dump_indirect((blkptr_t *)buf, gbh_nblkptrs(size), flags); 8708 } 8709 8710 static void 8711 zdb_dump_block_raw(void *buf, uint64_t size, int flags) 8712 { 8713 if (flags & ZDB_FLAG_BSWAP) 8714 byteswap_uint64_array(buf, size); 8715 VERIFY(write(fileno(stdout), buf, size) == size); 8716 } 8717 8718 static void 8719 zdb_dump_block(char *label, void *buf, uint64_t size, int flags) 8720 { 8721 uint64_t *d = (uint64_t *)buf; 8722 unsigned nwords = size / sizeof (uint64_t); 8723 int do_bswap = !!(flags & ZDB_FLAG_BSWAP); 8724 unsigned i, j; 8725 const char *hdr; 8726 char *c; 8727 8728 8729 if (do_bswap) 8730 hdr = " 7 6 5 4 3 2 1 0 f e d c b a 9 8"; 8731 else 8732 hdr = " 0 1 2 3 4 5 6 7 8 9 a b c d e f"; 8733 8734 (void) printf("\n%s\n%6s %s 0123456789abcdef\n", label, "", hdr); 8735 8736 #ifdef _ZFS_LITTLE_ENDIAN 8737 /* correct the endianness */ 8738 do_bswap = !do_bswap; 8739 #endif 8740 for (i = 0; i < nwords; i += 2) { 8741 (void) printf("%06llx: %016llx %016llx ", 8742 (u_longlong_t)(i * sizeof (uint64_t)), 8743 (u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]), 8744 (u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1])); 8745 8746 c = (char *)&d[i]; 8747 for (j = 0; j < 2 * sizeof (uint64_t); j++) 8748 (void) printf("%c", isprint(c[j]) ? c[j] : '.'); 8749 (void) printf("\n"); 8750 } 8751 } 8752 8753 /* 8754 * There are two acceptable formats: 8755 * leaf_name - For example: c1t0d0 or /tmp/ztest.0a 8756 * child[.child]* - For example: 0.1.1 8757 * 8758 * The second form can be used to specify arbitrary vdevs anywhere 8759 * in the hierarchy. For example, in a pool with a mirror of 8760 * RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 . 8761 */ 8762 static vdev_t * 8763 zdb_vdev_lookup(vdev_t *vdev, const char *path) 8764 { 8765 char *s, *p, *q; 8766 unsigned i; 8767 8768 if (vdev == NULL) 8769 return (NULL); 8770 8771 /* First, assume the x.x.x.x format */ 8772 i = strtoul(path, &s, 10); 8773 if (s == path || (s && *s != '.' && *s != '\0')) 8774 goto name; 8775 if (i >= vdev->vdev_children) 8776 return (NULL); 8777 8778 vdev = vdev->vdev_child[i]; 8779 if (s && *s == '\0') 8780 return (vdev); 8781 return (zdb_vdev_lookup(vdev, s+1)); 8782 8783 name: 8784 for (i = 0; i < vdev->vdev_children; i++) { 8785 vdev_t *vc = vdev->vdev_child[i]; 8786 8787 if (vc->vdev_path == NULL) { 8788 vc = zdb_vdev_lookup(vc, path); 8789 if (vc == NULL) 8790 continue; 8791 else 8792 return (vc); 8793 } 8794 8795 p = strrchr(vc->vdev_path, '/'); 8796 p = p ? p + 1 : vc->vdev_path; 8797 q = &vc->vdev_path[strlen(vc->vdev_path) - 2]; 8798 8799 if (strcmp(vc->vdev_path, path) == 0) 8800 return (vc); 8801 if (strcmp(p, path) == 0) 8802 return (vc); 8803 if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0) 8804 return (vc); 8805 } 8806 8807 return (NULL); 8808 } 8809 8810 static int 8811 name_from_objset_id(spa_t *spa, uint64_t objset_id, char *outstr) 8812 { 8813 dsl_dataset_t *ds; 8814 8815 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 8816 int error = dsl_dataset_hold_obj(spa->spa_dsl_pool, objset_id, 8817 NULL, &ds); 8818 if (error != 0) { 8819 (void) fprintf(stderr, "failed to hold objset %llu: %s\n", 8820 (u_longlong_t)objset_id, strerror(error)); 8821 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 8822 return (error); 8823 } 8824 dsl_dataset_name(ds, outstr); 8825 dsl_dataset_rele(ds, NULL); 8826 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 8827 return (0); 8828 } 8829 8830 static boolean_t 8831 zdb_parse_block_sizes(char *sizes, uint64_t *lsize, uint64_t *psize) 8832 { 8833 char *s0, *s1, *tmp = NULL; 8834 8835 if (sizes == NULL) 8836 return (B_FALSE); 8837 8838 s0 = strtok_r(sizes, "/", &tmp); 8839 if (s0 == NULL) 8840 return (B_FALSE); 8841 s1 = strtok_r(NULL, "/", &tmp); 8842 *lsize = strtoull(s0, NULL, 16); 8843 *psize = s1 ? strtoull(s1, NULL, 16) : *lsize; 8844 return (*lsize >= *psize && *psize > 0); 8845 } 8846 8847 #define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg)) 8848 8849 static boolean_t 8850 try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize, 8851 int flags, int cfunc, void *lbuf, void *lbuf2) 8852 { 8853 if (flags & ZDB_FLAG_VERBOSE) { 8854 (void) fprintf(stderr, 8855 "Trying %05llx -> %05llx (%s)\n", 8856 (u_longlong_t)psize, 8857 (u_longlong_t)lsize, 8858 zio_compress_table[cfunc].ci_name); 8859 } 8860 8861 /* 8862 * We set lbuf to all zeros and lbuf2 to all 8863 * ones, then decompress to both buffers and 8864 * compare their contents. This way we can 8865 * know if decompression filled exactly to 8866 * lsize or if it left some bytes unwritten. 8867 */ 8868 8869 memset(lbuf, 0x00, lsize); 8870 memset(lbuf2, 0xff, lsize); 8871 8872 abd_t labd, labd2; 8873 abd_get_from_buf_struct(&labd, lbuf, lsize); 8874 abd_get_from_buf_struct(&labd2, lbuf2, lsize); 8875 8876 boolean_t ret = B_FALSE; 8877 if (zio_decompress_data(cfunc, pabd, 8878 &labd, psize, lsize, NULL) == 0 && 8879 zio_decompress_data(cfunc, pabd, 8880 &labd2, psize, lsize, NULL) == 0 && 8881 memcmp(lbuf, lbuf2, lsize) == 0) 8882 ret = B_TRUE; 8883 8884 abd_free(&labd2); 8885 abd_free(&labd); 8886 8887 return (ret); 8888 } 8889 8890 static uint64_t 8891 zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize, 8892 uint64_t psize, int flags) 8893 { 8894 (void) buf; 8895 uint64_t orig_lsize = lsize; 8896 boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL)); 8897 /* 8898 * We don't know how the data was compressed, so just try 8899 * every decompress function at every inflated blocksize. 8900 */ 8901 void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL); 8902 int cfuncs[ZIO_COMPRESS_FUNCTIONS] = { 0 }; 8903 int *cfuncp = cfuncs; 8904 uint64_t maxlsize = SPA_MAXBLOCKSIZE; 8905 uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) | 8906 ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) | 8907 ZIO_COMPRESS_MASK(ZLE); 8908 *cfuncp++ = ZIO_COMPRESS_LZ4; 8909 *cfuncp++ = ZIO_COMPRESS_LZJB; 8910 mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB); 8911 /* 8912 * Every gzip level has the same decompressor, no need to 8913 * run it 9 times per bruteforce attempt. 8914 */ 8915 mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3); 8916 mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5); 8917 mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7); 8918 mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9); 8919 for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++) 8920 if (((1ULL << c) & mask) == 0) 8921 *cfuncp++ = c; 8922 8923 /* 8924 * On the one hand, with SPA_MAXBLOCKSIZE at 16MB, this 8925 * could take a while and we should let the user know 8926 * we are not stuck. On the other hand, printing progress 8927 * info gets old after a while. User can specify 'v' flag 8928 * to see the progression. 8929 */ 8930 if (lsize == psize) 8931 lsize += SPA_MINBLOCKSIZE; 8932 else 8933 maxlsize = lsize; 8934 8935 for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) { 8936 for (cfuncp = cfuncs; *cfuncp; cfuncp++) { 8937 if (try_decompress_block(pabd, lsize, psize, flags, 8938 *cfuncp, lbuf, lbuf2)) { 8939 tryzle = B_FALSE; 8940 break; 8941 } 8942 } 8943 if (*cfuncp != 0) 8944 break; 8945 } 8946 if (tryzle) { 8947 for (lsize = orig_lsize; lsize <= maxlsize; 8948 lsize += SPA_MINBLOCKSIZE) { 8949 if (try_decompress_block(pabd, lsize, psize, flags, 8950 ZIO_COMPRESS_ZLE, lbuf, lbuf2)) { 8951 *cfuncp = ZIO_COMPRESS_ZLE; 8952 break; 8953 } 8954 } 8955 } 8956 umem_free(lbuf2, SPA_MAXBLOCKSIZE); 8957 8958 if (*cfuncp == ZIO_COMPRESS_ZLE) { 8959 printf("\nZLE decompression was selected. If you " 8960 "suspect the results are wrong,\ntry avoiding ZLE " 8961 "by setting and exporting ZDB_NO_ZLE=\"true\"\n"); 8962 } 8963 8964 return (lsize > maxlsize ? -1 : lsize); 8965 } 8966 8967 /* 8968 * Read a block from a pool and print it out. The syntax of the 8969 * block descriptor is: 8970 * 8971 * pool:vdev_specifier:offset:[lsize/]psize[:flags] 8972 * 8973 * pool - The name of the pool you wish to read from 8974 * vdev_specifier - Which vdev (see comment for zdb_vdev_lookup) 8975 * offset - offset, in hex, in bytes 8976 * size - Amount of data to read, in hex, in bytes 8977 * flags - A string of characters specifying options 8978 * b: Decode a blkptr at given offset within block 8979 * c: Calculate and display checksums 8980 * d: Decompress data before dumping 8981 * e: Byteswap data before dumping 8982 * g: Display data as a gang block header 8983 * i: Display as an indirect block 8984 * r: Dump raw data to stdout 8985 * v: Verbose 8986 * 8987 */ 8988 static void 8989 zdb_read_block(char *thing, spa_t *spa) 8990 { 8991 blkptr_t blk, *bp = &blk; 8992 dva_t *dva = bp->blk_dva; 8993 int flags = 0; 8994 uint64_t offset = 0, psize = 0, lsize = 0, blkptr_offset = 0; 8995 zio_t *zio; 8996 vdev_t *vd; 8997 abd_t *pabd; 8998 void *lbuf, *buf; 8999 char *s, *p, *dup, *flagstr, *sizes, *tmp = NULL; 9000 const char *vdev, *errmsg = NULL; 9001 int i, len, error; 9002 boolean_t borrowed = B_FALSE, found = B_FALSE; 9003 9004 dup = strdup(thing); 9005 s = strtok_r(dup, ":", &tmp); 9006 vdev = s ?: ""; 9007 s = strtok_r(NULL, ":", &tmp); 9008 offset = strtoull(s ? s : "", NULL, 16); 9009 sizes = strtok_r(NULL, ":", &tmp); 9010 s = strtok_r(NULL, ":", &tmp); 9011 flagstr = strdup(s ?: ""); 9012 9013 if (!zdb_parse_block_sizes(sizes, &lsize, &psize)) 9014 errmsg = "invalid size(s)"; 9015 if (!IS_P2ALIGNED(psize, DEV_BSIZE) || !IS_P2ALIGNED(lsize, DEV_BSIZE)) 9016 errmsg = "size must be a multiple of sector size"; 9017 if (!IS_P2ALIGNED(offset, DEV_BSIZE)) 9018 errmsg = "offset must be a multiple of sector size"; 9019 if (errmsg) { 9020 (void) printf("Invalid block specifier: %s - %s\n", 9021 thing, errmsg); 9022 goto done; 9023 } 9024 9025 tmp = NULL; 9026 for (s = strtok_r(flagstr, ":", &tmp); 9027 s != NULL; 9028 s = strtok_r(NULL, ":", &tmp)) { 9029 len = strlen(flagstr); 9030 for (i = 0; i < len; i++) { 9031 int bit = flagbits[(uchar_t)flagstr[i]]; 9032 9033 if (bit == 0) { 9034 (void) printf("***Ignoring flag: %c\n", 9035 (uchar_t)flagstr[i]); 9036 continue; 9037 } 9038 found = B_TRUE; 9039 flags |= bit; 9040 9041 p = &flagstr[i + 1]; 9042 if (*p != ':' && *p != '\0') { 9043 int j = 0, nextbit = flagbits[(uchar_t)*p]; 9044 char *end, offstr[8] = { 0 }; 9045 if ((bit == ZDB_FLAG_PRINT_BLKPTR) && 9046 (nextbit == 0)) { 9047 /* look ahead to isolate the offset */ 9048 while (nextbit == 0 && 9049 strchr(flagbitstr, *p) == NULL) { 9050 offstr[j] = *p; 9051 j++; 9052 if (i + j > strlen(flagstr)) 9053 break; 9054 p++; 9055 nextbit = flagbits[(uchar_t)*p]; 9056 } 9057 blkptr_offset = strtoull(offstr, &end, 9058 16); 9059 i += j; 9060 } else if (nextbit == 0) { 9061 (void) printf("***Ignoring flag arg:" 9062 " '%c'\n", (uchar_t)*p); 9063 } 9064 } 9065 } 9066 } 9067 if (blkptr_offset % sizeof (blkptr_t)) { 9068 printf("Block pointer offset 0x%llx " 9069 "must be divisible by 0x%x\n", 9070 (longlong_t)blkptr_offset, (int)sizeof (blkptr_t)); 9071 goto done; 9072 } 9073 if (found == B_FALSE && strlen(flagstr) > 0) { 9074 printf("Invalid flag arg: '%s'\n", flagstr); 9075 goto done; 9076 } 9077 9078 vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev); 9079 if (vd == NULL) { 9080 (void) printf("***Invalid vdev: %s\n", vdev); 9081 goto done; 9082 } else { 9083 if (vd->vdev_path) 9084 (void) fprintf(stderr, "Found vdev: %s\n", 9085 vd->vdev_path); 9086 else 9087 (void) fprintf(stderr, "Found vdev type: %s\n", 9088 vd->vdev_ops->vdev_op_type); 9089 } 9090 9091 pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE); 9092 lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL); 9093 9094 BP_ZERO(bp); 9095 9096 DVA_SET_VDEV(&dva[0], vd->vdev_id); 9097 DVA_SET_OFFSET(&dva[0], offset); 9098 DVA_SET_GANG(&dva[0], 0); 9099 DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize)); 9100 9101 BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL); 9102 9103 BP_SET_LSIZE(bp, lsize); 9104 BP_SET_PSIZE(bp, psize); 9105 BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF); 9106 BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF); 9107 BP_SET_TYPE(bp, DMU_OT_NONE); 9108 BP_SET_LEVEL(bp, 0); 9109 BP_SET_DEDUP(bp, 0); 9110 BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); 9111 9112 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 9113 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); 9114 9115 if (vd == vd->vdev_top) { 9116 /* 9117 * Treat this as a normal block read. 9118 */ 9119 zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL, 9120 ZIO_PRIORITY_SYNC_READ, 9121 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL)); 9122 } else { 9123 /* 9124 * Treat this as a vdev child I/O. 9125 */ 9126 zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd, 9127 psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ, 9128 ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY | 9129 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | ZIO_FLAG_OPTIONAL, 9130 NULL, NULL)); 9131 } 9132 9133 error = zio_wait(zio); 9134 spa_config_exit(spa, SCL_STATE, FTAG); 9135 9136 if (error) { 9137 (void) printf("Read of %s failed, error: %d\n", thing, error); 9138 goto out; 9139 } 9140 9141 uint64_t orig_lsize = lsize; 9142 buf = lbuf; 9143 if (flags & ZDB_FLAG_DECOMPRESS) { 9144 lsize = zdb_decompress_block(pabd, buf, lbuf, 9145 lsize, psize, flags); 9146 if (lsize == -1) { 9147 (void) printf("Decompress of %s failed\n", thing); 9148 goto out; 9149 } 9150 } else { 9151 buf = abd_borrow_buf_copy(pabd, lsize); 9152 borrowed = B_TRUE; 9153 } 9154 /* 9155 * Try to detect invalid block pointer. If invalid, try 9156 * decompressing. 9157 */ 9158 if ((flags & ZDB_FLAG_PRINT_BLKPTR || flags & ZDB_FLAG_INDIRECT) && 9159 !(flags & ZDB_FLAG_DECOMPRESS)) { 9160 const blkptr_t *b = (const blkptr_t *)(void *) 9161 ((uintptr_t)buf + (uintptr_t)blkptr_offset); 9162 if (zfs_blkptr_verify(spa, b, 9163 BLK_CONFIG_NEEDED, BLK_VERIFY_ONLY)) { 9164 abd_return_buf_copy(pabd, buf, lsize); 9165 borrowed = B_FALSE; 9166 buf = lbuf; 9167 lsize = zdb_decompress_block(pabd, buf, 9168 lbuf, lsize, psize, flags); 9169 b = (const blkptr_t *)(void *) 9170 ((uintptr_t)buf + (uintptr_t)blkptr_offset); 9171 if (lsize == -1 || zfs_blkptr_verify(spa, b, 9172 BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) { 9173 printf("invalid block pointer at this DVA\n"); 9174 goto out; 9175 } 9176 } 9177 } 9178 9179 if (flags & ZDB_FLAG_PRINT_BLKPTR) 9180 zdb_print_blkptr((blkptr_t *)(void *) 9181 ((uintptr_t)buf + (uintptr_t)blkptr_offset), flags); 9182 else if (flags & ZDB_FLAG_RAW) 9183 zdb_dump_block_raw(buf, lsize, flags); 9184 else if (flags & ZDB_FLAG_INDIRECT) 9185 zdb_dump_indirect((blkptr_t *)buf, 9186 orig_lsize / sizeof (blkptr_t), flags); 9187 else if (flags & ZDB_FLAG_GBH) 9188 zdb_dump_gbh(buf, lsize, flags); 9189 else 9190 zdb_dump_block(thing, buf, lsize, flags); 9191 9192 /* 9193 * If :c was specified, iterate through the checksum table to 9194 * calculate and display each checksum for our specified 9195 * DVA and length. 9196 */ 9197 if ((flags & ZDB_FLAG_CHECKSUM) && !(flags & ZDB_FLAG_RAW) && 9198 !(flags & ZDB_FLAG_GBH)) { 9199 zio_t *czio; 9200 (void) printf("\n"); 9201 for (enum zio_checksum ck = ZIO_CHECKSUM_LABEL; 9202 ck < ZIO_CHECKSUM_FUNCTIONS; ck++) { 9203 9204 if ((zio_checksum_table[ck].ci_flags & 9205 ZCHECKSUM_FLAG_EMBEDDED) || 9206 ck == ZIO_CHECKSUM_NOPARITY) { 9207 continue; 9208 } 9209 BP_SET_CHECKSUM(bp, ck); 9210 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 9211 czio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); 9212 if (vd == vd->vdev_top) { 9213 zio_nowait(zio_read(czio, spa, bp, pabd, psize, 9214 NULL, NULL, 9215 ZIO_PRIORITY_SYNC_READ, 9216 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | 9217 ZIO_FLAG_DONT_RETRY, NULL)); 9218 } else { 9219 zio_nowait(zio_vdev_child_io(czio, bp, vd, 9220 offset, pabd, psize, ZIO_TYPE_READ, 9221 ZIO_PRIORITY_SYNC_READ, 9222 ZIO_FLAG_DONT_PROPAGATE | 9223 ZIO_FLAG_DONT_RETRY | 9224 ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW | 9225 ZIO_FLAG_SPECULATIVE | 9226 ZIO_FLAG_OPTIONAL, NULL, NULL)); 9227 } 9228 error = zio_wait(czio); 9229 if (error == 0 || error == ECKSUM) { 9230 zio_t *ck_zio = zio_null(NULL, spa, NULL, 9231 NULL, NULL, 0); 9232 ck_zio->io_offset = 9233 DVA_GET_OFFSET(&bp->blk_dva[0]); 9234 ck_zio->io_bp = bp; 9235 zio_checksum_compute(ck_zio, ck, pabd, psize); 9236 printf( 9237 "%12s\t" 9238 "cksum=%016llx:%016llx:%016llx:%016llx\n", 9239 zio_checksum_table[ck].ci_name, 9240 (u_longlong_t)bp->blk_cksum.zc_word[0], 9241 (u_longlong_t)bp->blk_cksum.zc_word[1], 9242 (u_longlong_t)bp->blk_cksum.zc_word[2], 9243 (u_longlong_t)bp->blk_cksum.zc_word[3]); 9244 zio_wait(ck_zio); 9245 } else { 9246 printf("error %d reading block\n", error); 9247 } 9248 spa_config_exit(spa, SCL_STATE, FTAG); 9249 } 9250 } 9251 9252 if (borrowed) 9253 abd_return_buf_copy(pabd, buf, lsize); 9254 9255 out: 9256 abd_free(pabd); 9257 umem_free(lbuf, SPA_MAXBLOCKSIZE); 9258 done: 9259 free(flagstr); 9260 free(dup); 9261 } 9262 9263 static void 9264 zdb_embedded_block(char *thing) 9265 { 9266 blkptr_t bp = {{{{0}}}}; 9267 unsigned long long *words = (void *)&bp; 9268 char *buf; 9269 int err; 9270 9271 err = sscanf(thing, "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx:" 9272 "%llx:%llx:%llx:%llx:%llx:%llx:%llx:%llx", 9273 words + 0, words + 1, words + 2, words + 3, 9274 words + 4, words + 5, words + 6, words + 7, 9275 words + 8, words + 9, words + 10, words + 11, 9276 words + 12, words + 13, words + 14, words + 15); 9277 if (err != 16) { 9278 (void) fprintf(stderr, "invalid input format\n"); 9279 zdb_exit(1); 9280 } 9281 ASSERT3U(BPE_GET_LSIZE(&bp), <=, SPA_MAXBLOCKSIZE); 9282 buf = malloc(SPA_MAXBLOCKSIZE); 9283 if (buf == NULL) { 9284 (void) fprintf(stderr, "out of memory\n"); 9285 zdb_exit(1); 9286 } 9287 err = decode_embedded_bp(&bp, buf, BPE_GET_LSIZE(&bp)); 9288 if (err != 0) { 9289 (void) fprintf(stderr, "decode failed: %u\n", err); 9290 zdb_exit(1); 9291 } 9292 zdb_dump_block_raw(buf, BPE_GET_LSIZE(&bp), 0); 9293 free(buf); 9294 } 9295 9296 /* check for valid hex or decimal numeric string */ 9297 static boolean_t 9298 zdb_numeric(char *str) 9299 { 9300 int i = 0, len; 9301 9302 len = strlen(str); 9303 if (len == 0) 9304 return (B_FALSE); 9305 if (strncmp(str, "0x", 2) == 0 || strncmp(str, "0X", 2) == 0) 9306 i = 2; 9307 for (; i < len; i++) { 9308 if (!isxdigit(str[i])) 9309 return (B_FALSE); 9310 } 9311 return (B_TRUE); 9312 } 9313 9314 static int 9315 dummy_get_file_info(dmu_object_type_t bonustype, const void *data, 9316 zfs_file_info_t *zoi) 9317 { 9318 (void) data, (void) zoi; 9319 9320 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) 9321 return (ENOENT); 9322 9323 (void) fprintf(stderr, "dummy_get_file_info: not implemented"); 9324 abort(); 9325 } 9326 9327 int 9328 main(int argc, char **argv) 9329 { 9330 int c; 9331 int dump_all = 1; 9332 int verbose = 0; 9333 int error = 0; 9334 char **searchdirs = NULL; 9335 int nsearch = 0; 9336 char *target, *target_pool, dsname[ZFS_MAX_DATASET_NAME_LEN]; 9337 nvlist_t *policy = NULL; 9338 uint64_t max_txg = UINT64_MAX; 9339 int64_t objset_id = -1; 9340 uint64_t object; 9341 int flags = ZFS_IMPORT_MISSING_LOG; 9342 int rewind = ZPOOL_NEVER_REWIND; 9343 char *spa_config_path_env, *objset_str; 9344 boolean_t target_is_spa = B_TRUE, dataset_lookup = B_FALSE; 9345 nvlist_t *cfg = NULL; 9346 struct sigaction action; 9347 boolean_t force_import = B_FALSE; 9348 boolean_t config_path_console = B_FALSE; 9349 char pbuf[MAXPATHLEN]; 9350 9351 dprintf_setup(&argc, argv); 9352 9353 /* 9354 * Set up signal handlers, so if we crash due to bad on-disk data we 9355 * can get more info. Unlike ztest, we don't bail out if we can't set 9356 * up signal handlers, because zdb is very useful without them. 9357 */ 9358 action.sa_handler = sig_handler; 9359 sigemptyset(&action.sa_mask); 9360 action.sa_flags = 0; 9361 if (sigaction(SIGSEGV, &action, NULL) < 0) { 9362 (void) fprintf(stderr, "zdb: cannot catch SIGSEGV: %s\n", 9363 strerror(errno)); 9364 } 9365 if (sigaction(SIGABRT, &action, NULL) < 0) { 9366 (void) fprintf(stderr, "zdb: cannot catch SIGABRT: %s\n", 9367 strerror(errno)); 9368 } 9369 9370 /* 9371 * If there is an environment variable SPA_CONFIG_PATH it overrides 9372 * default spa_config_path setting. If -U flag is specified it will 9373 * override this environment variable settings once again. 9374 */ 9375 spa_config_path_env = getenv("SPA_CONFIG_PATH"); 9376 if (spa_config_path_env != NULL) 9377 spa_config_path = spa_config_path_env; 9378 9379 /* 9380 * For performance reasons, we set this tunable down. We do so before 9381 * the arg parsing section so that the user can override this value if 9382 * they choose. 9383 */ 9384 zfs_btree_verify_intensity = 3; 9385 9386 struct option long_options[] = { 9387 {"ignore-assertions", no_argument, NULL, 'A'}, 9388 {"block-stats", no_argument, NULL, 'b'}, 9389 {"backup", no_argument, NULL, 'B'}, 9390 {"checksum", no_argument, NULL, 'c'}, 9391 {"config", no_argument, NULL, 'C'}, 9392 {"datasets", no_argument, NULL, 'd'}, 9393 {"dedup-stats", no_argument, NULL, 'D'}, 9394 {"exported", no_argument, NULL, 'e'}, 9395 {"embedded-block-pointer", no_argument, NULL, 'E'}, 9396 {"automatic-rewind", no_argument, NULL, 'F'}, 9397 {"dump-debug-msg", no_argument, NULL, 'G'}, 9398 {"history", no_argument, NULL, 'h'}, 9399 {"intent-logs", no_argument, NULL, 'i'}, 9400 {"inflight", required_argument, NULL, 'I'}, 9401 {"checkpointed-state", no_argument, NULL, 'k'}, 9402 {"key", required_argument, NULL, 'K'}, 9403 {"label", no_argument, NULL, 'l'}, 9404 {"disable-leak-tracking", no_argument, NULL, 'L'}, 9405 {"metaslabs", no_argument, NULL, 'm'}, 9406 {"metaslab-groups", no_argument, NULL, 'M'}, 9407 {"numeric", no_argument, NULL, 'N'}, 9408 {"option", required_argument, NULL, 'o'}, 9409 {"object-lookups", no_argument, NULL, 'O'}, 9410 {"path", required_argument, NULL, 'p'}, 9411 {"parseable", no_argument, NULL, 'P'}, 9412 {"skip-label", no_argument, NULL, 'q'}, 9413 {"copy-object", no_argument, NULL, 'r'}, 9414 {"read-block", no_argument, NULL, 'R'}, 9415 {"io-stats", no_argument, NULL, 's'}, 9416 {"simulate-dedup", no_argument, NULL, 'S'}, 9417 {"txg", required_argument, NULL, 't'}, 9418 {"brt-stats", no_argument, NULL, 'T'}, 9419 {"uberblock", no_argument, NULL, 'u'}, 9420 {"cachefile", required_argument, NULL, 'U'}, 9421 {"verbose", no_argument, NULL, 'v'}, 9422 {"verbatim", no_argument, NULL, 'V'}, 9423 {"dump-blocks", required_argument, NULL, 'x'}, 9424 {"extreme-rewind", no_argument, NULL, 'X'}, 9425 {"all-reconstruction", no_argument, NULL, 'Y'}, 9426 {"livelist", no_argument, NULL, 'y'}, 9427 {"zstd-headers", no_argument, NULL, 'Z'}, 9428 {"allocated-map", no_argument, NULL, 9429 ALLOCATED_OPT}, 9430 {0, 0, 0, 0} 9431 }; 9432 9433 while ((c = getopt_long(argc, argv, 9434 "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ", 9435 long_options, NULL)) != -1) { 9436 switch (c) { 9437 case 'b': 9438 case 'B': 9439 case 'c': 9440 case 'C': 9441 case 'd': 9442 case 'D': 9443 case 'E': 9444 case 'G': 9445 case 'h': 9446 case 'i': 9447 case 'l': 9448 case 'm': 9449 case 'M': 9450 case 'N': 9451 case 'O': 9452 case 'r': 9453 case 'R': 9454 case 's': 9455 case 'S': 9456 case 'T': 9457 case 'u': 9458 case 'y': 9459 case 'Z': 9460 case ALLOCATED_OPT: 9461 dump_opt[c]++; 9462 dump_all = 0; 9463 break; 9464 case 'A': 9465 case 'e': 9466 case 'F': 9467 case 'k': 9468 case 'L': 9469 case 'P': 9470 case 'q': 9471 case 'X': 9472 dump_opt[c]++; 9473 break; 9474 case 'Y': 9475 zfs_reconstruct_indirect_combinations_max = INT_MAX; 9476 zfs_deadman_enabled = 0; 9477 break; 9478 /* NB: Sort single match options below. */ 9479 case 'I': 9480 max_inflight_bytes = strtoull(optarg, NULL, 0); 9481 if (max_inflight_bytes == 0) { 9482 (void) fprintf(stderr, "maximum number " 9483 "of inflight bytes must be greater " 9484 "than 0\n"); 9485 usage(); 9486 } 9487 break; 9488 case 'K': 9489 dump_opt[c]++; 9490 key_material = strdup(optarg); 9491 /* redact key material in process table */ 9492 while (*optarg != '\0') { *optarg++ = '*'; } 9493 break; 9494 case 'o': 9495 dump_opt[c]++; 9496 dump_all = 0; 9497 error = handle_tunable_option(optarg, B_FALSE); 9498 if (error != 0) 9499 zdb_exit(1); 9500 break; 9501 case 'p': 9502 if (searchdirs == NULL) { 9503 searchdirs = umem_alloc(sizeof (char *), 9504 UMEM_NOFAIL); 9505 } else { 9506 char **tmp = umem_alloc((nsearch + 1) * 9507 sizeof (char *), UMEM_NOFAIL); 9508 memcpy(tmp, searchdirs, nsearch * 9509 sizeof (char *)); 9510 umem_free(searchdirs, 9511 nsearch * sizeof (char *)); 9512 searchdirs = tmp; 9513 } 9514 searchdirs[nsearch++] = optarg; 9515 break; 9516 case 't': 9517 max_txg = strtoull(optarg, NULL, 0); 9518 if (max_txg < TXG_INITIAL) { 9519 (void) fprintf(stderr, "incorrect txg " 9520 "specified: %s\n", optarg); 9521 usage(); 9522 } 9523 break; 9524 case 'U': 9525 config_path_console = B_TRUE; 9526 spa_config_path = optarg; 9527 if (spa_config_path[0] != '/') { 9528 (void) fprintf(stderr, 9529 "cachefile must be an absolute path " 9530 "(i.e. start with a slash)\n"); 9531 usage(); 9532 } 9533 break; 9534 case 'v': 9535 verbose++; 9536 break; 9537 case 'V': 9538 flags = ZFS_IMPORT_VERBATIM; 9539 break; 9540 case 'x': 9541 vn_dumpdir = optarg; 9542 break; 9543 default: 9544 usage(); 9545 break; 9546 } 9547 } 9548 9549 if (!dump_opt['e'] && searchdirs != NULL) { 9550 (void) fprintf(stderr, "-p option requires use of -e\n"); 9551 usage(); 9552 } 9553 #if defined(_LP64) 9554 /* 9555 * ZDB does not typically re-read blocks; therefore limit the ARC 9556 * to 256 MB, which can be used entirely for metadata. 9557 */ 9558 zfs_arc_min = 2ULL << SPA_MAXBLOCKSHIFT; 9559 zfs_arc_max = 256 * 1024 * 1024; 9560 #endif 9561 9562 /* 9563 * "zdb -c" uses checksum-verifying scrub i/os which are async reads. 9564 * "zdb -b" uses traversal prefetch which uses async reads. 9565 * For good performance, let several of them be active at once. 9566 */ 9567 zfs_vdev_async_read_max_active = 10; 9568 9569 /* 9570 * Disable reference tracking for better performance. 9571 */ 9572 reference_tracking_enable = B_FALSE; 9573 9574 /* 9575 * Do not fail spa_load when spa_load_verify fails. This is needed 9576 * to load non-idle pools. 9577 */ 9578 spa_load_verify_dryrun = B_TRUE; 9579 9580 /* 9581 * ZDB should have ability to read spacemaps. 9582 */ 9583 spa_mode_readable_spacemaps = B_TRUE; 9584 9585 if (dump_all) 9586 verbose = MAX(verbose, 1); 9587 9588 for (c = 0; c < 256; c++) { 9589 if (dump_all && strchr("ABeEFkKlLNOPrRSXy", c) == NULL) 9590 dump_opt[c] = 1; 9591 if (dump_opt[c]) 9592 dump_opt[c] += verbose; 9593 } 9594 9595 libspl_set_assert_ok((dump_opt['A'] == 1) || (dump_opt['A'] > 2)); 9596 zfs_recover = (dump_opt['A'] > 1); 9597 9598 argc -= optind; 9599 argv += optind; 9600 if (argc < 2 && dump_opt['R']) 9601 usage(); 9602 9603 target = argv[0]; 9604 9605 /* 9606 * Automate cachefile 9607 */ 9608 if (!spa_config_path_env && !config_path_console && target && 9609 libzfs_core_init() == 0) { 9610 char *pname = strdup(target); 9611 const char *value; 9612 nvlist_t *pnvl = NULL; 9613 nvlist_t *vnvl = NULL; 9614 9615 if (strpbrk(pname, "/@") != NULL) 9616 *strpbrk(pname, "/@") = '\0'; 9617 9618 if (pname && lzc_get_props(pname, &pnvl) == 0) { 9619 if (nvlist_lookup_nvlist(pnvl, "cachefile", 9620 &vnvl) == 0) { 9621 value = fnvlist_lookup_string(vnvl, 9622 ZPROP_VALUE); 9623 } else { 9624 value = "-"; 9625 } 9626 strlcpy(pbuf, value, sizeof (pbuf)); 9627 if (pbuf[0] != '\0') { 9628 if (pbuf[0] == '/') { 9629 if (access(pbuf, F_OK) == 0) 9630 spa_config_path = pbuf; 9631 else 9632 force_import = B_TRUE; 9633 } else if ((strcmp(pbuf, "-") == 0 && 9634 access(ZPOOL_CACHE, F_OK) != 0) || 9635 strcmp(pbuf, "none") == 0) { 9636 force_import = B_TRUE; 9637 } 9638 } 9639 nvlist_free(vnvl); 9640 } 9641 9642 free(pname); 9643 nvlist_free(pnvl); 9644 libzfs_core_fini(); 9645 } 9646 9647 dmu_objset_register_type(DMU_OST_ZFS, dummy_get_file_info); 9648 kernel_init(SPA_MODE_READ); 9649 kernel_init_done = B_TRUE; 9650 9651 if (dump_opt['E']) { 9652 if (argc != 1) 9653 usage(); 9654 zdb_embedded_block(argv[0]); 9655 error = 0; 9656 goto fini; 9657 } 9658 9659 if (argc < 1) { 9660 if (!dump_opt['e'] && dump_opt['C']) { 9661 dump_cachefile(spa_config_path); 9662 error = 0; 9663 goto fini; 9664 } 9665 if (dump_opt['o']) 9666 /* 9667 * Avoid blasting tunable options off the top of the 9668 * screen. 9669 */ 9670 zdb_exit(1); 9671 usage(); 9672 } 9673 9674 if (dump_opt['l']) { 9675 error = dump_label(argv[0]); 9676 goto fini; 9677 } 9678 9679 if (dump_opt['X'] || dump_opt['F']) 9680 rewind = ZPOOL_DO_REWIND | 9681 (dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0); 9682 9683 /* -N implies -d */ 9684 if (dump_opt['N'] && dump_opt['d'] == 0) 9685 dump_opt['d'] = dump_opt['N']; 9686 9687 if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 || 9688 nvlist_add_uint64(policy, ZPOOL_LOAD_REQUEST_TXG, max_txg) != 0 || 9689 nvlist_add_uint32(policy, ZPOOL_LOAD_REWIND_POLICY, rewind) != 0) 9690 fatal("internal error: %s", strerror(ENOMEM)); 9691 9692 error = 0; 9693 9694 if (strpbrk(target, "/@") != NULL) { 9695 size_t targetlen; 9696 9697 target_pool = strdup(target); 9698 *strpbrk(target_pool, "/@") = '\0'; 9699 9700 target_is_spa = B_FALSE; 9701 targetlen = strlen(target); 9702 if (targetlen && target[targetlen - 1] == '/') 9703 target[targetlen - 1] = '\0'; 9704 9705 /* 9706 * See if an objset ID was supplied (-d <pool>/<objset ID>). 9707 * To disambiguate tank/100, consider the 100 as objsetID 9708 * if -N was given, otherwise 100 is an objsetID iff 9709 * tank/100 as a named dataset fails on lookup. 9710 */ 9711 objset_str = strchr(target, '/'); 9712 if (objset_str && strlen(objset_str) > 1 && 9713 zdb_numeric(objset_str + 1)) { 9714 char *endptr; 9715 errno = 0; 9716 objset_str++; 9717 objset_id = strtoull(objset_str, &endptr, 0); 9718 /* dataset 0 is the same as opening the pool */ 9719 if (errno == 0 && endptr != objset_str && 9720 objset_id != 0) { 9721 if (dump_opt['N']) 9722 dataset_lookup = B_TRUE; 9723 } 9724 /* normal dataset name not an objset ID */ 9725 if (endptr == objset_str) { 9726 objset_id = -1; 9727 } 9728 } else if (objset_str && !zdb_numeric(objset_str + 1) && 9729 dump_opt['N']) { 9730 printf("Supply a numeric objset ID with -N\n"); 9731 error = 2; 9732 goto fini; 9733 } 9734 } else { 9735 target_pool = target; 9736 } 9737 9738 if (dump_opt['e'] || force_import) { 9739 importargs_t args = { 0 }; 9740 9741 /* 9742 * If path is not provided, search in /dev 9743 */ 9744 if (searchdirs == NULL) { 9745 searchdirs = umem_alloc(sizeof (char *), UMEM_NOFAIL); 9746 searchdirs[nsearch++] = (char *)ZFS_DEVDIR; 9747 } 9748 9749 args.paths = nsearch; 9750 args.path = searchdirs; 9751 args.can_be_active = B_TRUE; 9752 9753 libpc_handle_t lpch = { 9754 .lpc_lib_handle = NULL, 9755 .lpc_ops = &libzpool_config_ops, 9756 .lpc_printerr = B_TRUE 9757 }; 9758 error = zpool_find_config(&lpch, target_pool, &cfg, &args); 9759 9760 if (error == 0) { 9761 9762 if (nvlist_add_nvlist(cfg, 9763 ZPOOL_LOAD_POLICY, policy) != 0) { 9764 fatal("can't open '%s': %s", 9765 target, strerror(ENOMEM)); 9766 } 9767 9768 if (dump_opt['C'] > 1) { 9769 (void) printf("\nConfiguration for import:\n"); 9770 dump_nvlist(cfg, 8); 9771 } 9772 9773 /* 9774 * Disable the activity check to allow examination of 9775 * active pools. 9776 */ 9777 error = spa_import(target_pool, cfg, NULL, 9778 flags | ZFS_IMPORT_SKIP_MMP); 9779 } 9780 } 9781 9782 if (searchdirs != NULL) { 9783 umem_free(searchdirs, nsearch * sizeof (char *)); 9784 searchdirs = NULL; 9785 } 9786 9787 /* 9788 * We need to make sure to process -O option or call 9789 * dump_path after the -e option has been processed, 9790 * which imports the pool to the namespace if it's 9791 * not in the cachefile. 9792 */ 9793 if (dump_opt['O']) { 9794 if (argc != 2) 9795 usage(); 9796 dump_opt['v'] = verbose + 3; 9797 error = dump_path(argv[0], argv[1], NULL); 9798 goto fini; 9799 } 9800 9801 if (dump_opt['r']) { 9802 target_is_spa = B_FALSE; 9803 if (argc != 3) 9804 usage(); 9805 dump_opt['v'] = verbose; 9806 error = dump_path(argv[0], argv[1], &object); 9807 if (error != 0) 9808 fatal("internal error: %s", strerror(error)); 9809 } 9810 9811 /* 9812 * import_checkpointed_state makes the assumption that the 9813 * target pool that we pass it is already part of the spa 9814 * namespace. Because of that we need to make sure to call 9815 * it always after the -e option has been processed, which 9816 * imports the pool to the namespace if it's not in the 9817 * cachefile. 9818 */ 9819 char *checkpoint_pool = NULL; 9820 char *checkpoint_target = NULL; 9821 if (dump_opt['k']) { 9822 checkpoint_pool = import_checkpointed_state(target, cfg, 9823 target_is_spa, &checkpoint_target); 9824 9825 if (checkpoint_target != NULL) 9826 target = checkpoint_target; 9827 } 9828 9829 if (cfg != NULL) { 9830 nvlist_free(cfg); 9831 cfg = NULL; 9832 } 9833 9834 if (target_pool != target) 9835 free(target_pool); 9836 9837 if (error == 0) { 9838 if (dump_opt['k'] && (target_is_spa || dump_opt['R'])) { 9839 ASSERT(checkpoint_pool != NULL); 9840 ASSERT0P(checkpoint_target); 9841 9842 error = spa_open(checkpoint_pool, &spa, FTAG); 9843 if (error != 0) { 9844 fatal("Tried to open pool \"%s\" but " 9845 "spa_open() failed with error %d\n", 9846 checkpoint_pool, error); 9847 } 9848 9849 } else if (target_is_spa || dump_opt['R'] || dump_opt['B'] || 9850 objset_id == 0) { 9851 zdb_set_skip_mmp(target); 9852 error = spa_open_rewind(target, &spa, FTAG, policy, 9853 NULL); 9854 if (error) { 9855 /* 9856 * If we're missing the log device then 9857 * try opening the pool after clearing the 9858 * log state. 9859 */ 9860 mutex_enter(&spa_namespace_lock); 9861 if ((spa = spa_lookup(target)) != NULL && 9862 spa->spa_log_state == SPA_LOG_MISSING) { 9863 spa->spa_log_state = SPA_LOG_CLEAR; 9864 error = 0; 9865 } 9866 mutex_exit(&spa_namespace_lock); 9867 9868 if (!error) { 9869 error = spa_open_rewind(target, &spa, 9870 FTAG, policy, NULL); 9871 } 9872 } 9873 } else if (strpbrk(target, "#") != NULL) { 9874 dsl_pool_t *dp; 9875 error = dsl_pool_hold(target, FTAG, &dp); 9876 if (error != 0) { 9877 fatal("can't dump '%s': %s", target, 9878 strerror(error)); 9879 } 9880 error = dump_bookmark(dp, target, B_TRUE, verbose > 1); 9881 dsl_pool_rele(dp, FTAG); 9882 if (error != 0) { 9883 fatal("can't dump '%s': %s", target, 9884 strerror(error)); 9885 } 9886 goto fini; 9887 } else { 9888 target_pool = strdup(target); 9889 if (strpbrk(target, "/@") != NULL) 9890 *strpbrk(target_pool, "/@") = '\0'; 9891 9892 zdb_set_skip_mmp(target); 9893 /* 9894 * If -N was supplied, the user has indicated that 9895 * zdb -d <pool>/<objsetID> is in effect. Otherwise 9896 * we first assume that the dataset string is the 9897 * dataset name. If dmu_objset_hold fails with the 9898 * dataset string, and we have an objset_id, retry the 9899 * lookup with the objsetID. 9900 */ 9901 boolean_t retry = B_TRUE; 9902 retry_lookup: 9903 if (dataset_lookup == B_TRUE) { 9904 /* 9905 * Use the supplied id to get the name 9906 * for open_objset. 9907 */ 9908 error = spa_open(target_pool, &spa, FTAG); 9909 if (error == 0) { 9910 error = name_from_objset_id(spa, 9911 objset_id, dsname); 9912 spa_close(spa, FTAG); 9913 if (error == 0) 9914 target = dsname; 9915 } 9916 } 9917 if (error == 0) { 9918 if (objset_id > 0 && retry) { 9919 int err = dmu_objset_hold(target, FTAG, 9920 &os); 9921 if (err) { 9922 dataset_lookup = B_TRUE; 9923 retry = B_FALSE; 9924 goto retry_lookup; 9925 } else { 9926 dmu_objset_rele(os, FTAG); 9927 } 9928 } 9929 error = open_objset(target, FTAG, &os); 9930 } 9931 if (error == 0) 9932 spa = dmu_objset_spa(os); 9933 free(target_pool); 9934 } 9935 } 9936 nvlist_free(policy); 9937 9938 if (error) 9939 fatal("can't open '%s': %s", target, strerror(error)); 9940 9941 /* 9942 * Set the pool failure mode to panic in order to prevent the pool 9943 * from suspending. A suspended I/O will have no way to resume and 9944 * can prevent the zdb(8) command from terminating as expected. 9945 */ 9946 if (spa != NULL) 9947 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC; 9948 9949 argv++; 9950 argc--; 9951 if (dump_opt['r']) { 9952 error = zdb_copy_object(os, object, argv[1]); 9953 } else if (!dump_opt['R']) { 9954 flagbits['d'] = ZOR_FLAG_DIRECTORY; 9955 flagbits['f'] = ZOR_FLAG_PLAIN_FILE; 9956 flagbits['m'] = ZOR_FLAG_SPACE_MAP; 9957 flagbits['z'] = ZOR_FLAG_ZAP; 9958 flagbits['A'] = ZOR_FLAG_ALL_TYPES; 9959 9960 if (argc > 0 && dump_opt['d']) { 9961 zopt_object_args = argc; 9962 zopt_object_ranges = calloc(zopt_object_args, 9963 sizeof (zopt_object_range_t)); 9964 for (unsigned i = 0; i < zopt_object_args; i++) { 9965 int err; 9966 const char *msg = NULL; 9967 9968 err = parse_object_range(argv[i], 9969 &zopt_object_ranges[i], &msg); 9970 if (err != 0) 9971 fatal("Bad object or range: '%s': %s\n", 9972 argv[i], msg ?: ""); 9973 } 9974 } else if (argc > 0 && dump_opt['m']) { 9975 zopt_metaslab_args = argc; 9976 zopt_metaslab = calloc(zopt_metaslab_args, 9977 sizeof (uint64_t)); 9978 for (unsigned i = 0; i < zopt_metaslab_args; i++) { 9979 errno = 0; 9980 zopt_metaslab[i] = strtoull(argv[i], NULL, 0); 9981 if (zopt_metaslab[i] == 0 && errno != 0) 9982 fatal("bad number %s: %s", argv[i], 9983 strerror(errno)); 9984 } 9985 } 9986 if (dump_opt['B']) { 9987 dump_backup(target, objset_id, 9988 argc > 0 ? argv[0] : NULL); 9989 } else if (os != NULL) { 9990 dump_objset(os); 9991 } else if (zopt_object_args > 0 && !dump_opt['m']) { 9992 dump_objset(spa->spa_meta_objset); 9993 } else { 9994 dump_zpool(spa); 9995 } 9996 } else { 9997 flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR; 9998 flagbits['c'] = ZDB_FLAG_CHECKSUM; 9999 flagbits['d'] = ZDB_FLAG_DECOMPRESS; 10000 flagbits['e'] = ZDB_FLAG_BSWAP; 10001 flagbits['g'] = ZDB_FLAG_GBH; 10002 flagbits['i'] = ZDB_FLAG_INDIRECT; 10003 flagbits['r'] = ZDB_FLAG_RAW; 10004 flagbits['v'] = ZDB_FLAG_VERBOSE; 10005 10006 for (int i = 0; i < argc; i++) 10007 zdb_read_block(argv[i], spa); 10008 } 10009 10010 if (dump_opt['k']) { 10011 free(checkpoint_pool); 10012 if (!target_is_spa) 10013 free(checkpoint_target); 10014 } 10015 10016 fini: 10017 if (spa != NULL) 10018 zdb_ddt_cleanup(spa); 10019 10020 if (os != NULL) { 10021 close_objset(os, FTAG); 10022 } else if (spa != NULL) { 10023 spa_close(spa, FTAG); 10024 } 10025 10026 fuid_table_destroy(); 10027 10028 dump_debug_buffer(); 10029 10030 if (kernel_init_done) 10031 kernel_fini(); 10032 10033 if (corruption_found && error == 0) 10034 error = 3; 10035 10036 return (error); 10037 } 10038