Date: Wed, 6 Nov 2019 09:00:07 +0000 (UTC) From: Andriy Gapon <avg@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-vendor@freebsd.org Subject: svn commit: r354383 - vendor-sys/illumos/dist/uts/common/fs/zfs vendor-sys/illumos/dist/uts/common/fs/zfs/sys vendor/illumos/dist/cmd/zdb vendor/illumos/dist/man/man1m Message-ID: <201911060900.xA6907VL047386@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: avg Date: Wed Nov 6 09:00:06 2019 New Revision: 354383 URL: https://svnweb.freebsd.org/changeset/base/354383 Log: 10592 misc. metaslab and vdev related ZoL bug fixes illumos/illumos-gate@555d674d5d4b8191dc83723188349d28278b2431 https://github.com/illumos/illumos-gate/commit/555d674d5d4b8191dc83723188349d28278b2431 https://www.illumos.org/issues/10592 This is a collection of recent fixes from ZoL: 8eef997679b Error path in metaslab_load_impl() forgets to drop ms_sync_lock 928e8ad47d3 Introduce auxiliary metaslab histograms 425d3237ee8 Get rid of space_map_update() for ms_synced_length 6c926f426a2 Simplify log vdev removal code 21e7cf5da89 zdb -L should skip leak detection altogether df72b8bebe0 Rename range_tree_verify to range_tree_verify_not_present 75058f33034 Remove unused vdev_t fields Portions contributed by: Jerry Jelinek <jerry.jelinek@joyent.com> Author: Serapheim Dimitropoulos <serapheim@delphix.com> Modified: vendor/illumos/dist/cmd/zdb/zdb.c vendor/illumos/dist/man/man1m/zdb.1m Changes in other areas also in this revision: Modified: vendor-sys/illumos/dist/uts/common/fs/zfs/metaslab.c vendor-sys/illumos/dist/uts/common/fs/zfs/range_tree.c vendor-sys/illumos/dist/uts/common/fs/zfs/spa_checkpoint.c vendor-sys/illumos/dist/uts/common/fs/zfs/space_map.c vendor-sys/illumos/dist/uts/common/fs/zfs/sys/metaslab.h vendor-sys/illumos/dist/uts/common/fs/zfs/sys/metaslab_impl.h vendor-sys/illumos/dist/uts/common/fs/zfs/sys/range_tree.h vendor-sys/illumos/dist/uts/common/fs/zfs/sys/space_map.h vendor-sys/illumos/dist/uts/common/fs/zfs/sys/vdev_impl.h vendor-sys/illumos/dist/uts/common/fs/zfs/vdev.c vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_indirect.c vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_indirect_mapping.c vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_initialize.c vendor-sys/illumos/dist/uts/common/fs/zfs/vdev_removal.c Modified: vendor/illumos/dist/cmd/zdb/zdb.c ============================================================================== --- vendor/illumos/dist/cmd/zdb/zdb.c Wed Nov 6 08:58:03 2019 (r354382) +++ vendor/illumos/dist/cmd/zdb/zdb.c Wed Nov 6 09:00:06 2019 (r354383) @@ -785,18 +785,21 @@ dump_spacemap(objset_t *os, space_map_t *sm) return; (void) printf("space map object %llu:\n", - (longlong_t)sm->sm_phys->smp_object); - (void) printf(" smp_objsize = 0x%llx\n", - (longlong_t)sm->sm_phys->smp_objsize); + (longlong_t)sm->sm_object); + (void) printf(" smp_length = 0x%llx\n", + (longlong_t)sm->sm_phys->smp_length); (void) printf(" smp_alloc = 0x%llx\n", (longlong_t)sm->sm_phys->smp_alloc); + if (dump_opt['d'] < 6 && dump_opt['m'] < 4) + return; + /* * Print out the freelist entries in both encoded and decoded form. */ uint8_t mapshift = sm->sm_shift; int64_t alloc = 0; - uint64_t word; + uint64_t word, entry_id = 0; for (uint64_t offset = 0; offset < space_map_length(sm); offset += sizeof (word)) { @@ -804,11 +807,12 @@ dump_spacemap(objset_t *os, space_map_t *sm) sizeof (word), &word, DMU_READ_PREFETCH)); if (sm_entry_is_debug(word)) { - (void) printf("\t [%6llu] %s: txg %llu, pass %llu\n", - (u_longlong_t)(offset / sizeof (word)), + (void) printf("\t [%6llu] %s: txg %llu pass %llu\n", + (u_longlong_t)entry_id, ddata[SM_DEBUG_ACTION_DECODE(word)], (u_longlong_t)SM_DEBUG_TXG_DECODE(word), (u_longlong_t)SM_DEBUG_SYNCPASS_DECODE(word)); + entry_id++; continue; } @@ -846,7 +850,7 @@ dump_spacemap(objset_t *os, space_map_t *sm) (void) printf("\t [%6llu] %c range:" " %010llx-%010llx size: %06llx vdev: %06llu words: %u\n", - (u_longlong_t)(offset / sizeof (word)), + (u_longlong_t)entry_id, entry_type, (u_longlong_t)entry_off, (u_longlong_t)(entry_off + entry_run), (u_longlong_t)entry_run, @@ -856,8 +860,9 @@ dump_spacemap(objset_t *os, space_map_t *sm) alloc += entry_run; else alloc -= entry_run; + entry_id++; } - if ((uint64_t)alloc != space_map_allocated(sm)) { + if (alloc != space_map_allocated(sm)) { (void) printf("space_map_object alloc (%lld) INCONSISTENT " "with space map summary (%lld)\n", (longlong_t)space_map_allocated(sm), (longlong_t)alloc); @@ -921,11 +926,8 @@ dump_metaslab(metaslab_t *msp) SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift); } - if (dump_opt['d'] > 5 || dump_opt['m'] > 3) { - ASSERT(msp->ms_size == (1ULL << vd->vdev_ms_shift)); - - dump_spacemap(spa->spa_meta_objset, msp->ms_sm); - } + ASSERT(msp->ms_size == (1ULL << vd->vdev_ms_shift)); + dump_spacemap(spa->spa_meta_objset, msp->ms_sm); } static void @@ -3096,6 +3098,8 @@ zdb_ddt_leak_init(spa_t *spa, zdb_cb_t *zcb) ddt_entry_t dde; int error; + ASSERT(!dump_opt['L']); + bzero(&ddb, sizeof (ddb)); while ((error = ddt_walk(spa, &ddb, &dde)) == 0) { blkptr_t blk; @@ -3119,12 +3123,10 @@ zdb_ddt_leak_init(spa_t *spa, zdb_cb_t *zcb) zcb->zcb_dedup_blocks++; } } - if (!dump_opt['L']) { - ddt_t *ddt = spa->spa_ddt[ddb.ddb_checksum]; - ddt_enter(ddt); - VERIFY(ddt_lookup(ddt, &blk, B_TRUE) != NULL); - ddt_exit(ddt); - } + ddt_t *ddt = spa->spa_ddt[ddb.ddb_checksum]; + ddt_enter(ddt); + VERIFY(ddt_lookup(ddt, &blk, B_TRUE) != NULL); + ddt_exit(ddt); } ASSERT(error == ENOENT); @@ -3166,6 +3168,9 @@ claim_segment_cb(void *arg, uint64_t offset, uint64_t static void zdb_claim_removing(spa_t *spa, zdb_cb_t *zcb) { + if (dump_opt['L']) + return; + if (spa->spa_vdev_removal == NULL) return; @@ -3257,7 +3262,6 @@ zdb_load_obsolete_counts(vdev_t *vd) space_map_t *prev_obsolete_sm = NULL; VERIFY0(space_map_open(&prev_obsolete_sm, spa->spa_meta_objset, scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0)); - space_map_update(prev_obsolete_sm); vdev_indirect_mapping_load_obsolete_spacemap(vim, counts, prev_obsolete_sm); space_map_close(prev_obsolete_sm); @@ -3351,9 +3355,9 @@ zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_ VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa), checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift)); - space_map_update(checkpoint_sm); VERIFY0(space_map_iterate(checkpoint_sm, + space_map_length(checkpoint_sm), checkpoint_sm_exclude_entry_cb, &cseea)); space_map_close(checkpoint_sm); @@ -3363,6 +3367,8 @@ zdb_leak_init_vdev_exclude_checkpoint(vdev_t *vd, zdb_ static void zdb_leak_init_exclude_checkpoint(spa_t *spa, zdb_cb_t *zcb) { + ASSERT(!dump_opt['L']); + vdev_t *rvd = spa->spa_root_vdev; for (uint64_t c = 0; c < rvd->vdev_children; c++) { ASSERT3U(c, ==, rvd->vdev_child[c]->vdev_id); @@ -3459,6 +3465,8 @@ load_indirect_ms_allocatable_tree(vdev_t *vd, metaslab static void zdb_leak_init_prepare_indirect_vdevs(spa_t *spa, zdb_cb_t *zcb) { + ASSERT(!dump_opt['L']); + vdev_t *rvd = spa->spa_root_vdev; for (uint64_t c = 0; c < rvd->vdev_children; c++) { vdev_t *vd = rvd->vdev_child[c]; @@ -3505,67 +3513,63 @@ zdb_leak_init(spa_t *spa, zdb_cb_t *zcb) { zcb->zcb_spa = spa; - if (!dump_opt['L']) { - dsl_pool_t *dp = spa->spa_dsl_pool; - vdev_t *rvd = spa->spa_root_vdev; + if (dump_opt['L']) + return; - /* - * We are going to be changing the meaning of the metaslab's - * ms_allocatable. Ensure that the allocator doesn't try to - * use the tree. - */ - spa->spa_normal_class->mc_ops = &zdb_metaslab_ops; - spa->spa_log_class->mc_ops = &zdb_metaslab_ops; + dsl_pool_t *dp = spa->spa_dsl_pool; + vdev_t *rvd = spa->spa_root_vdev; - zcb->zcb_vd_obsolete_counts = - umem_zalloc(rvd->vdev_children * sizeof (uint32_t *), - UMEM_NOFAIL); + /* + * We are going to be changing the meaning of the metaslab's + * ms_allocatable. Ensure that the allocator doesn't try to + * use the tree. + */ + spa->spa_normal_class->mc_ops = &zdb_metaslab_ops; + spa->spa_log_class->mc_ops = &zdb_metaslab_ops; - /* - * For leak detection, we overload the ms_allocatable trees - * to contain allocated segments instead of free segments. - * As a result, we can't use the normal metaslab_load/unload - * interfaces. - */ - zdb_leak_init_prepare_indirect_vdevs(spa, zcb); - load_concrete_ms_allocatable_trees(spa, SM_ALLOC); + zcb->zcb_vd_obsolete_counts = + umem_zalloc(rvd->vdev_children * sizeof (uint32_t *), + UMEM_NOFAIL); - /* - * On load_concrete_ms_allocatable_trees() we loaded all the - * allocated entries from the ms_sm to the ms_allocatable for - * each metaslab. If the pool has a checkpoint or is in the - * middle of discarding a checkpoint, some of these blocks - * may have been freed but their ms_sm may not have been - * updated because they are referenced by the checkpoint. In - * order to avoid false-positives during leak-detection, we - * go through the vdev's checkpoint space map and exclude all - * its entries from their relevant ms_allocatable. - * - * We also aggregate the space held by the checkpoint and add - * it to zcb_checkpoint_size. - * - * Note that at this point we are also verifying that all the - * entries on the checkpoint_sm are marked as allocated in - * the ms_sm of their relevant metaslab. - * [see comment in checkpoint_sm_exclude_entry_cb()] - */ - zdb_leak_init_exclude_checkpoint(spa, zcb); + /* + * For leak detection, we overload the ms_allocatable trees + * to contain allocated segments instead of free segments. + * As a result, we can't use the normal metaslab_load/unload + * interfaces. + */ + zdb_leak_init_prepare_indirect_vdevs(spa, zcb); + load_concrete_ms_allocatable_trees(spa, SM_ALLOC); - /* for cleaner progress output */ - (void) fprintf(stderr, "\n"); + /* + * On load_concrete_ms_allocatable_trees() we loaded all the + * allocated entries from the ms_sm to the ms_allocatable for + * each metaslab. If the pool has a checkpoint or is in the + * middle of discarding a checkpoint, some of these blocks + * may have been freed but their ms_sm may not have been + * updated because they are referenced by the checkpoint. In + * order to avoid false-positives during leak-detection, we + * go through the vdev's checkpoint space map and exclude all + * its entries from their relevant ms_allocatable. + * + * We also aggregate the space held by the checkpoint and add + * it to zcb_checkpoint_size. + * + * Note that at this point we are also verifying that all the + * entries on the checkpoint_sm are marked as allocated in + * the ms_sm of their relevant metaslab. + * [see comment in checkpoint_sm_exclude_entry_cb()] + */ + zdb_leak_init_exclude_checkpoint(spa, zcb); + ASSERT3U(zcb->zcb_checkpoint_size, ==, spa_get_checkpoint_space(spa)); - if (bpobj_is_open(&dp->dp_obsolete_bpobj)) { - ASSERT(spa_feature_is_enabled(spa, - SPA_FEATURE_DEVICE_REMOVAL)); - (void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj, - increment_indirect_mapping_cb, zcb, NULL); - } - } else { - /* - * If leak tracing is disabled, we still need to consider - * any checkpointed space in our space verification. - */ - zcb->zcb_checkpoint_size += spa_get_checkpoint_space(spa); + /* for cleaner progress output */ + (void) fprintf(stderr, "\n"); + + if (bpobj_is_open(&dp->dp_obsolete_bpobj)) { + ASSERT(spa_feature_is_enabled(spa, + SPA_FEATURE_DEVICE_REMOVAL)); + (void) bpobj_iterate_nofree(&dp->dp_obsolete_bpobj, + increment_indirect_mapping_cb, zcb, NULL); } spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); @@ -3646,52 +3650,58 @@ zdb_check_for_obsolete_leaks(vdev_t *vd, zdb_cb_t *zcb static boolean_t zdb_leak_fini(spa_t *spa, zdb_cb_t *zcb) { + if (dump_opt['L']) + return (B_FALSE); + boolean_t leaks = B_FALSE; - if (!dump_opt['L']) { - vdev_t *rvd = spa->spa_root_vdev; - for (unsigned c = 0; c < rvd->vdev_children; c++) { - vdev_t *vd = rvd->vdev_child[c]; - metaslab_group_t *mg = vd->vdev_mg; - if (zcb->zcb_vd_obsolete_counts[c] != NULL) { - leaks |= zdb_check_for_obsolete_leaks(vd, zcb); - } + vdev_t *rvd = spa->spa_root_vdev; + for (unsigned c = 0; c < rvd->vdev_children; c++) { + vdev_t *vd = rvd->vdev_child[c]; +#if DEBUG + metaslab_group_t *mg = vd->vdev_mg; +#endif - for (uint64_t m = 0; m < vd->vdev_ms_count; m++) { - metaslab_t *msp = vd->vdev_ms[m]; - ASSERT3P(mg, ==, msp->ms_group); + if (zcb->zcb_vd_obsolete_counts[c] != NULL) { + leaks |= zdb_check_for_obsolete_leaks(vd, zcb); + } - /* - * ms_allocatable has been overloaded - * to contain allocated segments. Now that - * we finished traversing all blocks, any - * block that remains in the ms_allocatable - * represents an allocated block that we - * did not claim during the traversal. - * Claimed blocks would have been removed - * from the ms_allocatable. For indirect - * vdevs, space remaining in the tree - * represents parts of the mapping that are - * not referenced, which is not a bug. - */ - if (vd->vdev_ops == &vdev_indirect_ops) { - range_tree_vacate(msp->ms_allocatable, - NULL, NULL); - } else { - range_tree_vacate(msp->ms_allocatable, - zdb_leak, vd); - } + for (uint64_t m = 0; m < vd->vdev_ms_count; m++) { + metaslab_t *msp = vd->vdev_ms[m]; + ASSERT3P(mg, ==, msp->ms_group); - if (msp->ms_loaded) { - msp->ms_loaded = B_FALSE; - } + /* + * ms_allocatable has been overloaded + * to contain allocated segments. Now that + * we finished traversing all blocks, any + * block that remains in the ms_allocatable + * represents an allocated block that we + * did not claim during the traversal. + * Claimed blocks would have been removed + * from the ms_allocatable. For indirect + * vdevs, space remaining in the tree + * represents parts of the mapping that are + * not referenced, which is not a bug. + */ + if (vd->vdev_ops == &vdev_indirect_ops) { + range_tree_vacate(msp->ms_allocatable, + NULL, NULL); + } else { + range_tree_vacate(msp->ms_allocatable, + zdb_leak, vd); } + + if (msp->ms_loaded) { + msp->ms_loaded = B_FALSE; + } } - umem_free(zcb->zcb_vd_obsolete_counts, - rvd->vdev_children * sizeof (uint32_t *)); - zcb->zcb_vd_obsolete_counts = NULL; } + + umem_free(zcb->zcb_vd_obsolete_counts, + rvd->vdev_children * sizeof (uint32_t *)); + zcb->zcb_vd_obsolete_counts = NULL; + return (leaks); } @@ -3730,13 +3740,18 @@ dump_block_stats(spa_t *spa) !dump_opt['L'] ? "nothing leaked " : ""); /* - * Load all space maps as SM_ALLOC maps, then traverse the pool - * claiming each block we discover. If the pool is perfectly - * consistent, the space maps will be empty when we're done. - * Anything left over is a leak; any block we can't claim (because - * it's not part of any space map) is a double allocation, - * reference to a freed block, or an unclaimed log block. + * When leak detection is enabled we load all space maps as SM_ALLOC + * maps, then traverse the pool claiming each block we discover. If + * the pool is perfectly consistent, the segment trees will be empty + * when we're done. Anything left over is a leak; any block we can't + * claim (because it's not part of any space map) is a double + * allocation, reference to a freed block, or an unclaimed log block. + * + * When leak detection is disabled (-L option) we still traverse the + * pool claiming each block we discover, but we skip opening any space + * maps. */ + bzero(&zcb, sizeof (zdb_cb_t)); zdb_leak_init(spa, &zcb); /* @@ -3815,11 +3830,10 @@ dump_block_stats(spa_t *spa) total_found = tzb->zb_asize - zcb.zcb_dedup_asize + zcb.zcb_removing_size + zcb.zcb_checkpoint_size; - if (total_found == total_alloc) { - if (!dump_opt['L']) - (void) printf("\n\tNo leaks (block sum matches space" - " maps exactly)\n"); - } else { + if (total_found == total_alloc && !dump_opt['L']) { + (void) printf("\n\tNo leaks (block sum matches space" + " maps exactly)\n"); + } else if (!dump_opt['L']) { (void) printf("block traversal size %llu != alloc %llu " "(%s %lld)\n", (u_longlong_t)total_found, @@ -4159,7 +4173,6 @@ verify_device_removal_feature_counts(spa_t *spa) spa->spa_meta_objset, scip->scip_prev_obsolete_sm_object, 0, vd->vdev_asize, 0)); - space_map_update(prev_obsolete_sm); dump_spacemap(spa->spa_meta_objset, prev_obsolete_sm); (void) printf("\n"); space_map_close(prev_obsolete_sm); @@ -4365,7 +4378,8 @@ verify_checkpoint_sm_entry_cb(space_map_entry_t *sme, * their respective ms_allocateable trees should not contain them. */ mutex_enter(&ms->ms_lock); - range_tree_verify(ms->ms_allocatable, sme->sme_offset, sme->sme_run); + range_tree_verify_not_present(ms->ms_allocatable, + sme->sme_offset, sme->sme_run); mutex_exit(&ms->ms_lock); return (0); @@ -4428,7 +4442,6 @@ verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, sp VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(current), checkpoint_sm_obj, 0, current_vd->vdev_asize, current_vd->vdev_ashift)); - space_map_update(checkpoint_sm); verify_checkpoint_sm_entry_cb_arg_t vcsec; vcsec.vcsec_vd = ckpoint_vd; @@ -4436,6 +4449,7 @@ verify_checkpoint_vdev_spacemaps(spa_t *checkpoint, sp vcsec.vcsec_num_entries = space_map_length(checkpoint_sm) / sizeof (uint64_t); VERIFY0(space_map_iterate(checkpoint_sm, + space_map_length(checkpoint_sm), verify_checkpoint_sm_entry_cb, &vcsec)); dump_spacemap(current->spa_meta_objset, checkpoint_sm); space_map_close(checkpoint_sm); @@ -4515,7 +4529,7 @@ verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_ * are part of the checkpoint were freed by mistake. */ range_tree_walk(ckpoint_msp->ms_allocatable, - (range_tree_func_t *)range_tree_verify, + (range_tree_func_t *)range_tree_verify_not_present, current_msp->ms_allocatable); } } @@ -4527,6 +4541,8 @@ verify_checkpoint_ms_spacemaps(spa_t *checkpoint, spa_ static void verify_checkpoint_blocks(spa_t *spa) { + ASSERT(!dump_opt['L']); + spa_t *checkpoint_spa; char *checkpoint_pool; nvlist_t *config = NULL; @@ -4592,7 +4608,6 @@ dump_leftover_checkpoint_blocks(spa_t *spa) VERIFY0(space_map_open(&checkpoint_sm, spa_meta_objset(spa), checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift)); - space_map_update(checkpoint_sm); dump_spacemap(spa->spa_meta_objset, checkpoint_sm); space_map_close(checkpoint_sm); } Modified: vendor/illumos/dist/man/man1m/zdb.1m ============================================================================== --- vendor/illumos/dist/man/man1m/zdb.1m Wed Nov 6 08:58:03 2019 (r354382) +++ vendor/illumos/dist/man/man1m/zdb.1m Wed Nov 6 09:00:06 2019 (r354383) @@ -10,7 +10,7 @@ .\" .\" .\" Copyright 2012, Richard Lowe. -.\" Copyright (c) 2012, 2017 by Delphix. All rights reserved. +.\" Copyright (c) 2012, 2018 by Delphix. All rights reserved. .\" Copyright 2017 Nexenta Systems, Inc. .\" .Dd April 14, 2017 @@ -187,7 +187,7 @@ If the .Fl u option is also specified, also display the uberblocks on this device. .It Fl L -Disable leak tracing and the loading of space maps. +Disable leak detection and the loading of space maps. By default, .Nm verifies that all non-free blocks are referenced, which can be very expensive.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201911060900.xA6907VL047386>