Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/metaslab.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/metaslab.c (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/metaslab.c (revision 336949) @@ -1,3929 +1,4207 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011, 2015 by Delphix. All rights reserved. + * Copyright (c) 2011, 2018 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #include #include #include #include #include #include #include #include #include #include #include SYSCTL_DECL(_vfs_zfs); SYSCTL_NODE(_vfs_zfs, OID_AUTO, metaslab, CTLFLAG_RW, 0, "ZFS metaslab"); #define GANG_ALLOCATION(flags) \ ((flags) & (METASLAB_GANG_CHILD | METASLAB_GANG_HEADER)) uint64_t metaslab_aliquot = 512ULL << 10; uint64_t metaslab_force_ganging = SPA_MAXBLOCKSIZE + 1; /* force gang blocks */ SYSCTL_QUAD(_vfs_zfs_metaslab, OID_AUTO, force_ganging, CTLFLAG_RWTUN, &metaslab_force_ganging, 0, "Force gang block allocation for blocks larger than or equal to this value"); /* * Since we can touch multiple metaslabs (and their respective space maps) * with each transaction group, we benefit from having a smaller space map * block size since it allows us to issue more I/O operations scattered * around the disk. */ int zfs_metaslab_sm_blksz = (1 << 12); SYSCTL_INT(_vfs_zfs, OID_AUTO, metaslab_sm_blksz, CTLFLAG_RDTUN, &zfs_metaslab_sm_blksz, 0, "Block size for metaslab DTL space map. Power of 2 and greater than 4096."); /* * The in-core space map representation is more compact than its on-disk form. * The zfs_condense_pct determines how much more compact the in-core * space map representation must be before we compact it on-disk. * Values should be greater than or equal to 100. */ int zfs_condense_pct = 200; SYSCTL_INT(_vfs_zfs, OID_AUTO, condense_pct, CTLFLAG_RWTUN, &zfs_condense_pct, 0, "Condense on-disk spacemap when it is more than this many percents" " of in-memory counterpart"); /* * Condensing a metaslab is not guaranteed to actually reduce the amount of * space used on disk. In particular, a space map uses data in increments of * MAX(1 << ashift, space_map_blksize), so a metaslab might use the * same number of blocks after condensing. Since the goal of condensing is to * reduce the number of IOPs required to read the space map, we only want to * condense when we can be sure we will reduce the number of blocks used by the * space map. Unfortunately, we cannot precisely compute whether or not this is * the case in metaslab_should_condense since we are holding ms_lock. Instead, * we apply the following heuristic: do not condense a spacemap unless the * uncondensed size consumes greater than zfs_metaslab_condense_block_threshold * blocks. */ int zfs_metaslab_condense_block_threshold = 4; /* * The zfs_mg_noalloc_threshold defines which metaslab groups should * be eligible for allocation. The value is defined as a percentage of * free space. Metaslab groups that have more free space than * zfs_mg_noalloc_threshold are always eligible for allocations. Once * a metaslab group's free space is less than or equal to the * zfs_mg_noalloc_threshold the allocator will avoid allocating to that * group unless all groups in the pool have reached zfs_mg_noalloc_threshold. * Once all groups in the pool reach zfs_mg_noalloc_threshold then all * groups are allowed to accept allocations. Gang blocks are always * eligible to allocate on any metaslab group. The default value of 0 means * no metaslab group will be excluded based on this criterion. */ int zfs_mg_noalloc_threshold = 0; SYSCTL_INT(_vfs_zfs, OID_AUTO, mg_noalloc_threshold, CTLFLAG_RWTUN, &zfs_mg_noalloc_threshold, 0, "Percentage of metaslab group size that should be free" " to make it eligible for allocation"); /* * Metaslab groups are considered eligible for allocations if their * fragmenation metric (measured as a percentage) is less than or equal to * zfs_mg_fragmentation_threshold. If a metaslab group exceeds this threshold * then it will be skipped unless all metaslab groups within the metaslab * class have also crossed this threshold. */ int zfs_mg_fragmentation_threshold = 85; SYSCTL_INT(_vfs_zfs, OID_AUTO, mg_fragmentation_threshold, CTLFLAG_RWTUN, &zfs_mg_fragmentation_threshold, 0, "Percentage of metaslab group size that should be considered " "eligible for allocations unless all metaslab groups within the metaslab class " "have also crossed this threshold"); /* * Allow metaslabs to keep their active state as long as their fragmentation * percentage is less than or equal to zfs_metaslab_fragmentation_threshold. An * active metaslab that exceeds this threshold will no longer keep its active * status allowing better metaslabs to be selected. */ int zfs_metaslab_fragmentation_threshold = 70; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, fragmentation_threshold, CTLFLAG_RWTUN, &zfs_metaslab_fragmentation_threshold, 0, "Maximum percentage of metaslab fragmentation level to keep their active state"); /* * When set will load all metaslabs when pool is first opened. */ int metaslab_debug_load = 0; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, debug_load, CTLFLAG_RWTUN, &metaslab_debug_load, 0, "Load all metaslabs when pool is first opened"); /* * When set will prevent metaslabs from being unloaded. */ int metaslab_debug_unload = 0; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, debug_unload, CTLFLAG_RWTUN, &metaslab_debug_unload, 0, "Prevent metaslabs from being unloaded"); /* * Minimum size which forces the dynamic allocator to change * it's allocation strategy. Once the space map cannot satisfy * an allocation of this size then it switches to using more * aggressive strategy (i.e search by size rather than offset). */ uint64_t metaslab_df_alloc_threshold = SPA_OLD_MAXBLOCKSIZE; SYSCTL_QUAD(_vfs_zfs_metaslab, OID_AUTO, df_alloc_threshold, CTLFLAG_RWTUN, &metaslab_df_alloc_threshold, 0, "Minimum size which forces the dynamic allocator to change it's allocation strategy"); /* * The minimum free space, in percent, which must be available * in a space map to continue allocations in a first-fit fashion. * Once the space map's free space drops below this level we dynamically * switch to using best-fit allocations. */ int metaslab_df_free_pct = 4; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, df_free_pct, CTLFLAG_RWTUN, &metaslab_df_free_pct, 0, "The minimum free space, in percent, which must be available in a " "space map to continue allocations in a first-fit fashion"); /* * A metaslab is considered "free" if it contains a contiguous * segment which is greater than metaslab_min_alloc_size. */ uint64_t metaslab_min_alloc_size = DMU_MAX_ACCESS; SYSCTL_QUAD(_vfs_zfs_metaslab, OID_AUTO, min_alloc_size, CTLFLAG_RWTUN, &metaslab_min_alloc_size, 0, "A metaslab is considered \"free\" if it contains a contiguous " "segment which is greater than vfs.zfs.metaslab.min_alloc_size"); /* * Percentage of all cpus that can be used by the metaslab taskq. */ int metaslab_load_pct = 50; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, load_pct, CTLFLAG_RWTUN, &metaslab_load_pct, 0, "Percentage of cpus that can be used by the metaslab taskq"); /* * Determines how many txgs a metaslab may remain loaded without having any * allocations from it. As long as a metaslab continues to be used we will * keep it loaded. */ int metaslab_unload_delay = TXG_SIZE * 2; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, unload_delay, CTLFLAG_RWTUN, &metaslab_unload_delay, 0, "Number of TXGs that an unused metaslab can be kept in memory"); /* * Max number of metaslabs per group to preload. */ int metaslab_preload_limit = SPA_DVAS_PER_BP; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, preload_limit, CTLFLAG_RWTUN, &metaslab_preload_limit, 0, "Max number of metaslabs per group to preload"); /* * Enable/disable preloading of metaslab. */ boolean_t metaslab_preload_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, preload_enabled, CTLFLAG_RWTUN, &metaslab_preload_enabled, 0, "Max number of metaslabs per group to preload"); /* * Enable/disable fragmentation weighting on metaslabs. */ boolean_t metaslab_fragmentation_factor_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, fragmentation_factor_enabled, CTLFLAG_RWTUN, &metaslab_fragmentation_factor_enabled, 0, "Enable fragmentation weighting on metaslabs"); /* * Enable/disable lba weighting (i.e. outer tracks are given preference). */ boolean_t metaslab_lba_weighting_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, lba_weighting_enabled, CTLFLAG_RWTUN, &metaslab_lba_weighting_enabled, 0, "Enable LBA weighting (i.e. outer tracks are given preference)"); /* * Enable/disable metaslab group biasing. */ boolean_t metaslab_bias_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_metaslab, OID_AUTO, bias_enabled, CTLFLAG_RWTUN, &metaslab_bias_enabled, 0, "Enable metaslab group biasing"); /* * Enable/disable remapping of indirect DVAs to their concrete vdevs. */ boolean_t zfs_remap_blkptr_enable = B_TRUE; /* * Enable/disable segment-based metaslab selection. */ boolean_t zfs_metaslab_segment_weight_enabled = B_TRUE; /* * When using segment-based metaslab selection, we will continue * allocating from the active metaslab until we have exhausted * zfs_metaslab_switch_threshold of its buckets. */ int zfs_metaslab_switch_threshold = 2; /* * Internal switch to enable/disable the metaslab allocation tracing * facility. */ boolean_t metaslab_trace_enabled = B_TRUE; /* * Maximum entries that the metaslab allocation tracing facility will keep * in a given list when running in non-debug mode. We limit the number * of entries in non-debug mode to prevent us from using up too much memory. * The limit should be sufficiently large that we don't expect any allocation * to every exceed this value. In debug mode, the system will panic if this * limit is ever reached allowing for further investigation. */ uint64_t metaslab_trace_max_entries = 5000; static uint64_t metaslab_weight(metaslab_t *); static void metaslab_set_fragmentation(metaslab_t *); static void metaslab_free_impl(vdev_t *, uint64_t, uint64_t, boolean_t); static void metaslab_check_free_impl(vdev_t *, uint64_t, uint64_t); +static void metaslab_passivate(metaslab_t *msp, uint64_t weight); +static uint64_t metaslab_weight_from_range_tree(metaslab_t *msp); kmem_cache_t *metaslab_alloc_trace_cache; /* * ========================================================================== * Metaslab classes * ========================================================================== */ metaslab_class_t * metaslab_class_create(spa_t *spa, metaslab_ops_t *ops) { metaslab_class_t *mc; mc = kmem_zalloc(sizeof (metaslab_class_t), KM_SLEEP); mc->mc_spa = spa; mc->mc_rotor = NULL; mc->mc_ops = ops; mutex_init(&mc->mc_lock, NULL, MUTEX_DEFAULT, NULL); - refcount_create_tracked(&mc->mc_alloc_slots); + mc->mc_alloc_slots = kmem_zalloc(spa->spa_alloc_count * + sizeof (refcount_t), KM_SLEEP); + mc->mc_alloc_max_slots = kmem_zalloc(spa->spa_alloc_count * + sizeof (uint64_t), KM_SLEEP); + for (int i = 0; i < spa->spa_alloc_count; i++) + refcount_create_tracked(&mc->mc_alloc_slots[i]); return (mc); } void metaslab_class_destroy(metaslab_class_t *mc) { ASSERT(mc->mc_rotor == NULL); ASSERT(mc->mc_alloc == 0); ASSERT(mc->mc_deferred == 0); ASSERT(mc->mc_space == 0); ASSERT(mc->mc_dspace == 0); - refcount_destroy(&mc->mc_alloc_slots); + for (int i = 0; i < mc->mc_spa->spa_alloc_count; i++) + refcount_destroy(&mc->mc_alloc_slots[i]); + kmem_free(mc->mc_alloc_slots, mc->mc_spa->spa_alloc_count * + sizeof (refcount_t)); + kmem_free(mc->mc_alloc_max_slots, mc->mc_spa->spa_alloc_count * + sizeof (uint64_t)); mutex_destroy(&mc->mc_lock); kmem_free(mc, sizeof (metaslab_class_t)); } int metaslab_class_validate(metaslab_class_t *mc) { metaslab_group_t *mg; vdev_t *vd; /* * Must hold one of the spa_config locks. */ ASSERT(spa_config_held(mc->mc_spa, SCL_ALL, RW_READER) || spa_config_held(mc->mc_spa, SCL_ALL, RW_WRITER)); if ((mg = mc->mc_rotor) == NULL) return (0); do { vd = mg->mg_vd; ASSERT(vd->vdev_mg != NULL); ASSERT3P(vd->vdev_top, ==, vd); ASSERT3P(mg->mg_class, ==, mc); ASSERT3P(vd->vdev_ops, !=, &vdev_hole_ops); } while ((mg = mg->mg_next) != mc->mc_rotor); return (0); } void metaslab_class_space_update(metaslab_class_t *mc, int64_t alloc_delta, int64_t defer_delta, int64_t space_delta, int64_t dspace_delta) { atomic_add_64(&mc->mc_alloc, alloc_delta); atomic_add_64(&mc->mc_deferred, defer_delta); atomic_add_64(&mc->mc_space, space_delta); atomic_add_64(&mc->mc_dspace, dspace_delta); } void metaslab_class_minblocksize_update(metaslab_class_t *mc) { metaslab_group_t *mg; vdev_t *vd; uint64_t minashift = UINT64_MAX; if ((mg = mc->mc_rotor) == NULL) { mc->mc_minblocksize = SPA_MINBLOCKSIZE; return; } do { vd = mg->mg_vd; if (vd->vdev_ashift < minashift) minashift = vd->vdev_ashift; } while ((mg = mg->mg_next) != mc->mc_rotor); mc->mc_minblocksize = 1ULL << minashift; } uint64_t metaslab_class_get_alloc(metaslab_class_t *mc) { return (mc->mc_alloc); } uint64_t metaslab_class_get_deferred(metaslab_class_t *mc) { return (mc->mc_deferred); } uint64_t metaslab_class_get_space(metaslab_class_t *mc) { return (mc->mc_space); } uint64_t metaslab_class_get_dspace(metaslab_class_t *mc) { return (spa_deflate(mc->mc_spa) ? mc->mc_dspace : mc->mc_space); } uint64_t metaslab_class_get_minblocksize(metaslab_class_t *mc) { return (mc->mc_minblocksize); } void metaslab_class_histogram_verify(metaslab_class_t *mc) { vdev_t *rvd = mc->mc_spa->spa_root_vdev; uint64_t *mc_hist; int i; if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0) return; mc_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE, KM_SLEEP); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; /* * Skip any holes, uninitialized top-levels, or * vdevs that are not in this metalab class. */ if (!vdev_is_concrete(tvd) || tvd->vdev_ms_shift == 0 || mg->mg_class != mc) { continue; } for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) mc_hist[i] += mg->mg_histogram[i]; } for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) VERIFY3U(mc_hist[i], ==, mc->mc_histogram[i]); kmem_free(mc_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE); } /* * Calculate the metaslab class's fragmentation metric. The metric * is weighted based on the space contribution of each metaslab group. * The return value will be a number between 0 and 100 (inclusive), or * ZFS_FRAG_INVALID if the metric has not been set. See comment above the * zfs_frag_table for more information about the metric. */ uint64_t metaslab_class_fragmentation(metaslab_class_t *mc) { vdev_t *rvd = mc->mc_spa->spa_root_vdev; uint64_t fragmentation = 0; spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; /* * Skip any holes, uninitialized top-levels, * or vdevs that are not in this metalab class. */ if (!vdev_is_concrete(tvd) || tvd->vdev_ms_shift == 0 || mg->mg_class != mc) { continue; } /* * If a metaslab group does not contain a fragmentation * metric then just bail out. */ if (mg->mg_fragmentation == ZFS_FRAG_INVALID) { spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG); return (ZFS_FRAG_INVALID); } /* * Determine how much this metaslab_group is contributing * to the overall pool fragmentation metric. */ fragmentation += mg->mg_fragmentation * metaslab_group_get_space(mg); } fragmentation /= metaslab_class_get_space(mc); ASSERT3U(fragmentation, <=, 100); spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG); return (fragmentation); } /* * Calculate the amount of expandable space that is available in * this metaslab class. If a device is expanded then its expandable * space will be the amount of allocatable space that is currently not * part of this metaslab class. */ uint64_t metaslab_class_expandable_space(metaslab_class_t *mc) { vdev_t *rvd = mc->mc_spa->spa_root_vdev; uint64_t space = 0; spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER); for (int c = 0; c < rvd->vdev_children; c++) { uint64_t tspace; vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; if (!vdev_is_concrete(tvd) || tvd->vdev_ms_shift == 0 || mg->mg_class != mc) { continue; } /* * Calculate if we have enough space to add additional * metaslabs. We report the expandable space in terms * of the metaslab size since that's the unit of expansion. * Adjust by efi system partition size. */ tspace = tvd->vdev_max_asize - tvd->vdev_asize; if (tspace > mc->mc_spa->spa_bootsize) { tspace -= mc->mc_spa->spa_bootsize; } space += P2ALIGN(tspace, 1ULL << tvd->vdev_ms_shift); } spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG); return (space); } static int metaslab_compare(const void *x1, const void *x2) { const metaslab_t *m1 = x1; const metaslab_t *m2 = x2; + int sort1 = 0; + int sort2 = 0; + if (m1->ms_allocator != -1 && m1->ms_primary) + sort1 = 1; + else if (m1->ms_allocator != -1 && !m1->ms_primary) + sort1 = 2; + if (m2->ms_allocator != -1 && m2->ms_primary) + sort2 = 1; + else if (m2->ms_allocator != -1 && !m2->ms_primary) + sort2 = 2; + + /* + * Sort inactive metaslabs first, then primaries, then secondaries. When + * selecting a metaslab to allocate from, an allocator first tries its + * primary, then secondary active metaslab. If it doesn't have active + * metaslabs, or can't allocate from them, it searches for an inactive + * metaslab to activate. If it can't find a suitable one, it will steal + * a primary or secondary metaslab from another allocator. + */ + if (sort1 < sort2) + return (-1); + if (sort1 > sort2) + return (1); + if (m1->ms_weight < m2->ms_weight) return (1); if (m1->ms_weight > m2->ms_weight) return (-1); /* * If the weights are identical, use the offset to force uniqueness. */ if (m1->ms_start < m2->ms_start) return (-1); if (m1->ms_start > m2->ms_start) return (1); ASSERT3P(m1, ==, m2); return (0); } /* * Verify that the space accounting on disk matches the in-core range_trees. */ void metaslab_verify_space(metaslab_t *msp, uint64_t txg) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; uint64_t allocated = 0; uint64_t sm_free_space, msp_free_space; ASSERT(MUTEX_HELD(&msp->ms_lock)); if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0) return; /* * We can only verify the metaslab space when we're called * from syncing context with a loaded metaslab that has an allocated * space map. Calling this in non-syncing context does not * provide a consistent view of the metaslab since we're performing * allocations in the future. */ if (txg != spa_syncing_txg(spa) || msp->ms_sm == NULL || !msp->ms_loaded) return; sm_free_space = msp->ms_size - space_map_allocated(msp->ms_sm) - space_map_alloc_delta(msp->ms_sm); /* * Account for future allocations since we would have already * deducted that space from the ms_freetree. */ for (int t = 0; t < TXG_CONCURRENT_STATES; t++) { allocated += range_tree_space(msp->ms_allocating[(txg + t) & TXG_MASK]); } msp_free_space = range_tree_space(msp->ms_allocatable) + allocated + msp->ms_deferspace + range_tree_space(msp->ms_freed); VERIFY3U(sm_free_space, ==, msp_free_space); } /* * ========================================================================== * Metaslab groups * ========================================================================== */ /* * Update the allocatable flag and the metaslab group's capacity. * The allocatable flag is set to true if the capacity is below * the zfs_mg_noalloc_threshold or has a fragmentation value that is * greater than zfs_mg_fragmentation_threshold. If a metaslab group * transitions from allocatable to non-allocatable or vice versa then the * metaslab group's class is updated to reflect the transition. */ static void metaslab_group_alloc_update(metaslab_group_t *mg) { vdev_t *vd = mg->mg_vd; metaslab_class_t *mc = mg->mg_class; vdev_stat_t *vs = &vd->vdev_stat; boolean_t was_allocatable; boolean_t was_initialized; ASSERT(vd == vd->vdev_top); ASSERT3U(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_READER), ==, SCL_ALLOC); mutex_enter(&mg->mg_lock); was_allocatable = mg->mg_allocatable; was_initialized = mg->mg_initialized; mg->mg_free_capacity = ((vs->vs_space - vs->vs_alloc) * 100) / (vs->vs_space + 1); mutex_enter(&mc->mc_lock); /* * If the metaslab group was just added then it won't * have any space until we finish syncing out this txg. * At that point we will consider it initialized and available * for allocations. We also don't consider non-activated * metaslab groups (e.g. vdevs that are in the middle of being removed) * to be initialized, because they can't be used for allocation. */ mg->mg_initialized = metaslab_group_initialized(mg); if (!was_initialized && mg->mg_initialized) { mc->mc_groups++; } else if (was_initialized && !mg->mg_initialized) { ASSERT3U(mc->mc_groups, >, 0); mc->mc_groups--; } if (mg->mg_initialized) mg->mg_no_free_space = B_FALSE; /* * A metaslab group is considered allocatable if it has plenty * of free space or is not heavily fragmented. We only take * fragmentation into account if the metaslab group has a valid * fragmentation metric (i.e. a value between 0 and 100). */ mg->mg_allocatable = (mg->mg_activation_count > 0 && mg->mg_free_capacity > zfs_mg_noalloc_threshold && (mg->mg_fragmentation == ZFS_FRAG_INVALID || mg->mg_fragmentation <= zfs_mg_fragmentation_threshold)); /* * The mc_alloc_groups maintains a count of the number of * groups in this metaslab class that are still above the * zfs_mg_noalloc_threshold. This is used by the allocating * threads to determine if they should avoid allocations to * a given group. The allocator will avoid allocations to a group * if that group has reached or is below the zfs_mg_noalloc_threshold * and there are still other groups that are above the threshold. * When a group transitions from allocatable to non-allocatable or * vice versa we update the metaslab class to reflect that change. * When the mc_alloc_groups value drops to 0 that means that all * groups have reached the zfs_mg_noalloc_threshold making all groups * eligible for allocations. This effectively means that all devices * are balanced again. */ if (was_allocatable && !mg->mg_allocatable) mc->mc_alloc_groups--; else if (!was_allocatable && mg->mg_allocatable) mc->mc_alloc_groups++; mutex_exit(&mc->mc_lock); mutex_exit(&mg->mg_lock); } metaslab_group_t * -metaslab_group_create(metaslab_class_t *mc, vdev_t *vd) +metaslab_group_create(metaslab_class_t *mc, vdev_t *vd, int allocators) { metaslab_group_t *mg; mg = kmem_zalloc(sizeof (metaslab_group_t), KM_SLEEP); mutex_init(&mg->mg_lock, NULL, MUTEX_DEFAULT, NULL); + mg->mg_primaries = kmem_zalloc(allocators * sizeof (metaslab_t *), + KM_SLEEP); + mg->mg_secondaries = kmem_zalloc(allocators * sizeof (metaslab_t *), + KM_SLEEP); avl_create(&mg->mg_metaslab_tree, metaslab_compare, sizeof (metaslab_t), offsetof(struct metaslab, ms_group_node)); mg->mg_vd = vd; mg->mg_class = mc; mg->mg_activation_count = 0; mg->mg_initialized = B_FALSE; mg->mg_no_free_space = B_TRUE; - refcount_create_tracked(&mg->mg_alloc_queue_depth); + mg->mg_allocators = allocators; + mg->mg_alloc_queue_depth = kmem_zalloc(allocators * sizeof (refcount_t), + KM_SLEEP); + mg->mg_cur_max_alloc_queue_depth = kmem_zalloc(allocators * + sizeof (uint64_t), KM_SLEEP); + for (int i = 0; i < allocators; i++) { + refcount_create_tracked(&mg->mg_alloc_queue_depth[i]); + mg->mg_cur_max_alloc_queue_depth[i] = 0; + } + mg->mg_taskq = taskq_create("metaslab_group_taskq", metaslab_load_pct, minclsyspri, 10, INT_MAX, TASKQ_THREADS_CPU_PCT); return (mg); } void metaslab_group_destroy(metaslab_group_t *mg) { ASSERT(mg->mg_prev == NULL); ASSERT(mg->mg_next == NULL); /* * We may have gone below zero with the activation count * either because we never activated in the first place or * because we're done, and possibly removing the vdev. */ ASSERT(mg->mg_activation_count <= 0); taskq_destroy(mg->mg_taskq); avl_destroy(&mg->mg_metaslab_tree); + kmem_free(mg->mg_primaries, mg->mg_allocators * sizeof (metaslab_t *)); + kmem_free(mg->mg_secondaries, mg->mg_allocators * + sizeof (metaslab_t *)); mutex_destroy(&mg->mg_lock); - refcount_destroy(&mg->mg_alloc_queue_depth); + + for (int i = 0; i < mg->mg_allocators; i++) { + refcount_destroy(&mg->mg_alloc_queue_depth[i]); + mg->mg_cur_max_alloc_queue_depth[i] = 0; + } + kmem_free(mg->mg_alloc_queue_depth, mg->mg_allocators * + sizeof (refcount_t)); + kmem_free(mg->mg_cur_max_alloc_queue_depth, mg->mg_allocators * + sizeof (uint64_t)); + kmem_free(mg, sizeof (metaslab_group_t)); } void metaslab_group_activate(metaslab_group_t *mg) { metaslab_class_t *mc = mg->mg_class; metaslab_group_t *mgprev, *mgnext; ASSERT3U(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_WRITER), !=, 0); ASSERT(mc->mc_rotor != mg); ASSERT(mg->mg_prev == NULL); ASSERT(mg->mg_next == NULL); ASSERT(mg->mg_activation_count <= 0); if (++mg->mg_activation_count <= 0) return; mg->mg_aliquot = metaslab_aliquot * MAX(1, mg->mg_vd->vdev_children); metaslab_group_alloc_update(mg); if ((mgprev = mc->mc_rotor) == NULL) { mg->mg_prev = mg; mg->mg_next = mg; } else { mgnext = mgprev->mg_next; mg->mg_prev = mgprev; mg->mg_next = mgnext; mgprev->mg_next = mg; mgnext->mg_prev = mg; } mc->mc_rotor = mg; metaslab_class_minblocksize_update(mc); } /* * Passivate a metaslab group and remove it from the allocation rotor. * Callers must hold both the SCL_ALLOC and SCL_ZIO lock prior to passivating * a metaslab group. This function will momentarily drop spa_config_locks * that are lower than the SCL_ALLOC lock (see comment below). */ void metaslab_group_passivate(metaslab_group_t *mg) { metaslab_class_t *mc = mg->mg_class; spa_t *spa = mc->mc_spa; metaslab_group_t *mgprev, *mgnext; int locks = spa_config_held(spa, SCL_ALL, RW_WRITER); ASSERT3U(spa_config_held(spa, SCL_ALLOC | SCL_ZIO, RW_WRITER), ==, (SCL_ALLOC | SCL_ZIO)); if (--mg->mg_activation_count != 0) { ASSERT(mc->mc_rotor != mg); ASSERT(mg->mg_prev == NULL); ASSERT(mg->mg_next == NULL); ASSERT(mg->mg_activation_count < 0); return; } /* * The spa_config_lock is an array of rwlocks, ordered as * follows (from highest to lowest): * SCL_CONFIG > SCL_STATE > SCL_L2ARC > SCL_ALLOC > * SCL_ZIO > SCL_FREE > SCL_VDEV * (For more information about the spa_config_lock see spa_misc.c) * The higher the lock, the broader its coverage. When we passivate * a metaslab group, we must hold both the SCL_ALLOC and the SCL_ZIO * config locks. However, the metaslab group's taskq might be trying * to preload metaslabs so we must drop the SCL_ZIO lock and any * lower locks to allow the I/O to complete. At a minimum, * we continue to hold the SCL_ALLOC lock, which prevents any future * allocations from taking place and any changes to the vdev tree. */ spa_config_exit(spa, locks & ~(SCL_ZIO - 1), spa); taskq_wait(mg->mg_taskq); spa_config_enter(spa, locks & ~(SCL_ZIO - 1), spa, RW_WRITER); metaslab_group_alloc_update(mg); + for (int i = 0; i < mg->mg_allocators; i++) { + metaslab_t *msp = mg->mg_primaries[i]; + if (msp != NULL) { + mutex_enter(&msp->ms_lock); + metaslab_passivate(msp, + metaslab_weight_from_range_tree(msp)); + mutex_exit(&msp->ms_lock); + } + msp = mg->mg_secondaries[i]; + if (msp != NULL) { + mutex_enter(&msp->ms_lock); + metaslab_passivate(msp, + metaslab_weight_from_range_tree(msp)); + mutex_exit(&msp->ms_lock); + } + } mgprev = mg->mg_prev; mgnext = mg->mg_next; if (mg == mgnext) { mc->mc_rotor = NULL; } else { mc->mc_rotor = mgnext; mgprev->mg_next = mgnext; mgnext->mg_prev = mgprev; } mg->mg_prev = NULL; mg->mg_next = NULL; metaslab_class_minblocksize_update(mc); } boolean_t metaslab_group_initialized(metaslab_group_t *mg) { vdev_t *vd = mg->mg_vd; vdev_stat_t *vs = &vd->vdev_stat; return (vs->vs_space != 0 && mg->mg_activation_count > 0); } uint64_t metaslab_group_get_space(metaslab_group_t *mg) { return ((1ULL << mg->mg_vd->vdev_ms_shift) * mg->mg_vd->vdev_ms_count); } void metaslab_group_histogram_verify(metaslab_group_t *mg) { uint64_t *mg_hist; vdev_t *vd = mg->mg_vd; uint64_t ashift = vd->vdev_ashift; int i; if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0) return; mg_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE, KM_SLEEP); ASSERT3U(RANGE_TREE_HISTOGRAM_SIZE, >=, SPACE_MAP_HISTOGRAM_SIZE + ashift); for (int m = 0; m < vd->vdev_ms_count; m++) { metaslab_t *msp = vd->vdev_ms[m]; if (msp->ms_sm == NULL) continue; for (i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) mg_hist[i + ashift] += msp->ms_sm->sm_phys->smp_histogram[i]; } for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i ++) VERIFY3U(mg_hist[i], ==, mg->mg_histogram[i]); kmem_free(mg_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE); } static void metaslab_group_histogram_add(metaslab_group_t *mg, metaslab_t *msp) { metaslab_class_t *mc = mg->mg_class; uint64_t ashift = mg->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); if (msp->ms_sm == NULL) return; mutex_enter(&mg->mg_lock); for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { mg->mg_histogram[i + ashift] += msp->ms_sm->sm_phys->smp_histogram[i]; mc->mc_histogram[i + ashift] += msp->ms_sm->sm_phys->smp_histogram[i]; } mutex_exit(&mg->mg_lock); } void metaslab_group_histogram_remove(metaslab_group_t *mg, metaslab_t *msp) { metaslab_class_t *mc = mg->mg_class; uint64_t ashift = mg->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); if (msp->ms_sm == NULL) return; mutex_enter(&mg->mg_lock); for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { ASSERT3U(mg->mg_histogram[i + ashift], >=, msp->ms_sm->sm_phys->smp_histogram[i]); ASSERT3U(mc->mc_histogram[i + ashift], >=, msp->ms_sm->sm_phys->smp_histogram[i]); mg->mg_histogram[i + ashift] -= msp->ms_sm->sm_phys->smp_histogram[i]; mc->mc_histogram[i + ashift] -= msp->ms_sm->sm_phys->smp_histogram[i]; } mutex_exit(&mg->mg_lock); } static void metaslab_group_add(metaslab_group_t *mg, metaslab_t *msp) { ASSERT(msp->ms_group == NULL); mutex_enter(&mg->mg_lock); msp->ms_group = mg; msp->ms_weight = 0; avl_add(&mg->mg_metaslab_tree, msp); mutex_exit(&mg->mg_lock); mutex_enter(&msp->ms_lock); metaslab_group_histogram_add(mg, msp); mutex_exit(&msp->ms_lock); } static void metaslab_group_remove(metaslab_group_t *mg, metaslab_t *msp) { mutex_enter(&msp->ms_lock); metaslab_group_histogram_remove(mg, msp); mutex_exit(&msp->ms_lock); mutex_enter(&mg->mg_lock); ASSERT(msp->ms_group == mg); avl_remove(&mg->mg_metaslab_tree, msp); msp->ms_group = NULL; mutex_exit(&mg->mg_lock); } static void +metaslab_group_sort_impl(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight) +{ + ASSERT(MUTEX_HELD(&mg->mg_lock)); + ASSERT(msp->ms_group == mg); + avl_remove(&mg->mg_metaslab_tree, msp); + msp->ms_weight = weight; + avl_add(&mg->mg_metaslab_tree, msp); + +} + +static void metaslab_group_sort(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight) { /* * Although in principle the weight can be any value, in * practice we do not use values in the range [1, 511]. */ ASSERT(weight >= SPA_MINBLOCKSIZE || weight == 0); ASSERT(MUTEX_HELD(&msp->ms_lock)); mutex_enter(&mg->mg_lock); - ASSERT(msp->ms_group == mg); - avl_remove(&mg->mg_metaslab_tree, msp); - msp->ms_weight = weight; - avl_add(&mg->mg_metaslab_tree, msp); + metaslab_group_sort_impl(mg, msp, weight); mutex_exit(&mg->mg_lock); } /* * Calculate the fragmentation for a given metaslab group. We can use * a simple average here since all metaslabs within the group must have * the same size. The return value will be a value between 0 and 100 * (inclusive), or ZFS_FRAG_INVALID if less than half of the metaslab in this * group have a fragmentation metric. */ uint64_t metaslab_group_fragmentation(metaslab_group_t *mg) { vdev_t *vd = mg->mg_vd; uint64_t fragmentation = 0; uint64_t valid_ms = 0; for (int m = 0; m < vd->vdev_ms_count; m++) { metaslab_t *msp = vd->vdev_ms[m]; if (msp->ms_fragmentation == ZFS_FRAG_INVALID) continue; valid_ms++; fragmentation += msp->ms_fragmentation; } if (valid_ms <= vd->vdev_ms_count / 2) return (ZFS_FRAG_INVALID); fragmentation /= valid_ms; ASSERT3U(fragmentation, <=, 100); return (fragmentation); } /* * Determine if a given metaslab group should skip allocations. A metaslab * group should avoid allocations if its free capacity is less than the * zfs_mg_noalloc_threshold or its fragmentation metric is greater than * zfs_mg_fragmentation_threshold and there is at least one metaslab group * that can still handle allocations. If the allocation throttle is enabled * then we skip allocations to devices that have reached their maximum * allocation queue depth unless the selected metaslab group is the only * eligible group remaining. */ static boolean_t metaslab_group_allocatable(metaslab_group_t *mg, metaslab_group_t *rotor, - uint64_t psize) + uint64_t psize, int allocator) { spa_t *spa = mg->mg_vd->vdev_spa; metaslab_class_t *mc = mg->mg_class; /* * We can only consider skipping this metaslab group if it's * in the normal metaslab class and there are other metaslab * groups to select from. Otherwise, we always consider it eligible * for allocations. */ if (mc != spa_normal_class(spa) || mc->mc_groups <= 1) return (B_TRUE); /* * If the metaslab group's mg_allocatable flag is set (see comments * in metaslab_group_alloc_update() for more information) and * the allocation throttle is disabled then allow allocations to this * device. However, if the allocation throttle is enabled then * check if we have reached our allocation limit (mg_alloc_queue_depth) * to determine if we should allow allocations to this metaslab group. * If all metaslab groups are no longer considered allocatable * (mc_alloc_groups == 0) or we're trying to allocate the smallest * gang block size then we allow allocations on this metaslab group * regardless of the mg_allocatable or throttle settings. */ if (mg->mg_allocatable) { metaslab_group_t *mgp; int64_t qdepth; - uint64_t qmax = mg->mg_max_alloc_queue_depth; + uint64_t qmax = mg->mg_cur_max_alloc_queue_depth[allocator]; if (!mc->mc_alloc_throttle_enabled) return (B_TRUE); /* * If this metaslab group does not have any free space, then * there is no point in looking further. */ if (mg->mg_no_free_space) return (B_FALSE); - qdepth = refcount_count(&mg->mg_alloc_queue_depth); + qdepth = refcount_count(&mg->mg_alloc_queue_depth[allocator]); /* * If this metaslab group is below its qmax or it's * the only allocatable metasable group, then attempt * to allocate from it. */ if (qdepth < qmax || mc->mc_alloc_groups == 1) return (B_TRUE); ASSERT3U(mc->mc_alloc_groups, >, 1); /* * Since this metaslab group is at or over its qmax, we * need to determine if there are metaslab groups after this * one that might be able to handle this allocation. This is * racy since we can't hold the locks for all metaslab * groups at the same time when we make this check. */ for (mgp = mg->mg_next; mgp != rotor; mgp = mgp->mg_next) { - qmax = mgp->mg_max_alloc_queue_depth; + qmax = mgp->mg_cur_max_alloc_queue_depth[allocator]; - qdepth = refcount_count(&mgp->mg_alloc_queue_depth); + qdepth = refcount_count( + &mgp->mg_alloc_queue_depth[allocator]); /* * If there is another metaslab group that * might be able to handle the allocation, then * we return false so that we skip this group. */ if (qdepth < qmax && !mgp->mg_no_free_space) return (B_FALSE); } /* * We didn't find another group to handle the allocation * so we can't skip this metaslab group even though * we are at or over our qmax. */ return (B_TRUE); } else if (mc->mc_alloc_groups == 0 || psize == SPA_MINBLOCKSIZE) { return (B_TRUE); } return (B_FALSE); } /* * ========================================================================== * Range tree callbacks * ========================================================================== */ /* * Comparison function for the private size-ordered tree. Tree is sorted * by size, larger sizes at the end of the tree. */ static int metaslab_rangesize_compare(const void *x1, const void *x2) { const range_seg_t *r1 = x1; const range_seg_t *r2 = x2; uint64_t rs_size1 = r1->rs_end - r1->rs_start; uint64_t rs_size2 = r2->rs_end - r2->rs_start; if (rs_size1 < rs_size2) return (-1); if (rs_size1 > rs_size2) return (1); if (r1->rs_start < r2->rs_start) return (-1); if (r1->rs_start > r2->rs_start) return (1); return (0); } /* * ========================================================================== * Common allocator routines * ========================================================================== */ /* * Return the maximum contiguous segment within the metaslab. */ uint64_t metaslab_block_maxsize(metaslab_t *msp) { avl_tree_t *t = &msp->ms_allocatable_by_size; range_seg_t *rs; if (t == NULL || (rs = avl_last(t)) == NULL) return (0ULL); return (rs->rs_end - rs->rs_start); } static range_seg_t * metaslab_block_find(avl_tree_t *t, uint64_t start, uint64_t size) { range_seg_t *rs, rsearch; avl_index_t where; rsearch.rs_start = start; rsearch.rs_end = start + size; rs = avl_find(t, &rsearch, &where); if (rs == NULL) { rs = avl_nearest(t, where, AVL_AFTER); } return (rs); } /* * This is a helper function that can be used by the allocator to find * a suitable block to allocate. This will search the specified AVL * tree looking for a block that matches the specified criteria. */ static uint64_t metaslab_block_picker(avl_tree_t *t, uint64_t *cursor, uint64_t size, uint64_t align) { range_seg_t *rs = metaslab_block_find(t, *cursor, size); while (rs != NULL) { uint64_t offset = P2ROUNDUP(rs->rs_start, align); if (offset + size <= rs->rs_end) { *cursor = offset + size; return (offset); } rs = AVL_NEXT(t, rs); } /* * If we know we've searched the whole map (*cursor == 0), give up. * Otherwise, reset the cursor to the beginning and try again. */ if (*cursor == 0) return (-1ULL); *cursor = 0; return (metaslab_block_picker(t, cursor, size, align)); } /* * ========================================================================== * The first-fit block allocator * ========================================================================== */ static uint64_t metaslab_ff_alloc(metaslab_t *msp, uint64_t size) { /* * Find the largest power of 2 block size that evenly divides the * requested size. This is used to try to allocate blocks with similar * alignment from the same area of the metaslab (i.e. same cursor * bucket) but it does not guarantee that other allocations sizes * may exist in the same region. */ uint64_t align = size & -size; uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1]; avl_tree_t *t = &msp->ms_allocatable->rt_root; return (metaslab_block_picker(t, cursor, size, align)); } static metaslab_ops_t metaslab_ff_ops = { metaslab_ff_alloc }; /* * ========================================================================== * Dynamic block allocator - * Uses the first fit allocation scheme until space get low and then * adjusts to a best fit allocation method. Uses metaslab_df_alloc_threshold * and metaslab_df_free_pct to determine when to switch the allocation scheme. * ========================================================================== */ static uint64_t metaslab_df_alloc(metaslab_t *msp, uint64_t size) { /* * Find the largest power of 2 block size that evenly divides the * requested size. This is used to try to allocate blocks with similar * alignment from the same area of the metaslab (i.e. same cursor * bucket) but it does not guarantee that other allocations sizes * may exist in the same region. */ uint64_t align = size & -size; uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1]; range_tree_t *rt = msp->ms_allocatable; avl_tree_t *t = &rt->rt_root; uint64_t max_size = metaslab_block_maxsize(msp); int free_pct = range_tree_space(rt) * 100 / msp->ms_size; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_allocatable_by_size)); if (max_size < size) return (-1ULL); /* * If we're running low on space switch to using the size * sorted AVL tree (best-fit). */ if (max_size < metaslab_df_alloc_threshold || free_pct < metaslab_df_free_pct) { t = &msp->ms_allocatable_by_size; *cursor = 0; } return (metaslab_block_picker(t, cursor, size, 1ULL)); } static metaslab_ops_t metaslab_df_ops = { metaslab_df_alloc }; /* * ========================================================================== * Cursor fit block allocator - * Select the largest region in the metaslab, set the cursor to the beginning * of the range and the cursor_end to the end of the range. As allocations * are made advance the cursor. Continue allocating from the cursor until * the range is exhausted and then find a new range. * ========================================================================== */ static uint64_t metaslab_cf_alloc(metaslab_t *msp, uint64_t size) { range_tree_t *rt = msp->ms_allocatable; avl_tree_t *t = &msp->ms_allocatable_by_size; uint64_t *cursor = &msp->ms_lbas[0]; uint64_t *cursor_end = &msp->ms_lbas[1]; uint64_t offset = 0; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&rt->rt_root)); ASSERT3U(*cursor_end, >=, *cursor); if ((*cursor + size) > *cursor_end) { range_seg_t *rs; rs = avl_last(&msp->ms_allocatable_by_size); if (rs == NULL || (rs->rs_end - rs->rs_start) < size) return (-1ULL); *cursor = rs->rs_start; *cursor_end = rs->rs_end; } offset = *cursor; *cursor += size; return (offset); } static metaslab_ops_t metaslab_cf_ops = { metaslab_cf_alloc }; /* * ========================================================================== * New dynamic fit allocator - * Select a region that is large enough to allocate 2^metaslab_ndf_clump_shift * contiguous blocks. If no region is found then just use the largest segment * that remains. * ========================================================================== */ /* * Determines desired number of contiguous blocks (2^metaslab_ndf_clump_shift) * to request from the allocator. */ uint64_t metaslab_ndf_clump_shift = 4; static uint64_t metaslab_ndf_alloc(metaslab_t *msp, uint64_t size) { avl_tree_t *t = &msp->ms_allocatable->rt_root; avl_index_t where; range_seg_t *rs, rsearch; uint64_t hbit = highbit64(size); uint64_t *cursor = &msp->ms_lbas[hbit - 1]; uint64_t max_size = metaslab_block_maxsize(msp); ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_allocatable_by_size)); if (max_size < size) return (-1ULL); rsearch.rs_start = *cursor; rsearch.rs_end = *cursor + size; rs = avl_find(t, &rsearch, &where); if (rs == NULL || (rs->rs_end - rs->rs_start) < size) { t = &msp->ms_allocatable_by_size; rsearch.rs_start = 0; rsearch.rs_end = MIN(max_size, 1ULL << (hbit + metaslab_ndf_clump_shift)); rs = avl_find(t, &rsearch, &where); if (rs == NULL) rs = avl_nearest(t, where, AVL_AFTER); ASSERT(rs != NULL); } if ((rs->rs_end - rs->rs_start) >= size) { *cursor = rs->rs_start + size; return (rs->rs_start); } return (-1ULL); } static metaslab_ops_t metaslab_ndf_ops = { metaslab_ndf_alloc }; metaslab_ops_t *zfs_metaslab_ops = &metaslab_df_ops; /* * ========================================================================== * Metaslabs * ========================================================================== */ /* * Wait for any in-progress metaslab loads to complete. */ void metaslab_load_wait(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); while (msp->ms_loading) { ASSERT(!msp->ms_loaded); cv_wait(&msp->ms_load_cv, &msp->ms_lock); } } int metaslab_load(metaslab_t *msp) { int error = 0; boolean_t success = B_FALSE; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(!msp->ms_loaded); ASSERT(!msp->ms_loading); msp->ms_loading = B_TRUE; /* * Nobody else can manipulate a loading metaslab, so it's now safe * to drop the lock. This way we don't have to hold the lock while * reading the spacemap from disk. */ mutex_exit(&msp->ms_lock); /* * If the space map has not been allocated yet, then treat * all the space in the metaslab as free and add it to ms_allocatable. */ if (msp->ms_sm != NULL) { error = space_map_load(msp->ms_sm, msp->ms_allocatable, SM_FREE); } else { range_tree_add(msp->ms_allocatable, msp->ms_start, msp->ms_size); } success = (error == 0); mutex_enter(&msp->ms_lock); msp->ms_loading = B_FALSE; if (success) { ASSERT3P(msp->ms_group, !=, NULL); msp->ms_loaded = B_TRUE; /* * If the metaslab already has a spacemap, then we need to * remove all segments from the defer tree; otherwise, the * metaslab is completely empty and we can skip this. */ if (msp->ms_sm != NULL) { for (int t = 0; t < TXG_DEFER_SIZE; t++) { range_tree_walk(msp->ms_defer[t], range_tree_remove, msp->ms_allocatable); } } msp->ms_max_size = metaslab_block_maxsize(msp); } cv_broadcast(&msp->ms_load_cv); return (error); } void metaslab_unload(metaslab_t *msp) { ASSERT(MUTEX_HELD(&msp->ms_lock)); range_tree_vacate(msp->ms_allocatable, NULL, NULL); msp->ms_loaded = B_FALSE; msp->ms_weight &= ~METASLAB_ACTIVE_MASK; msp->ms_max_size = 0; } int metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object, uint64_t txg, metaslab_t **msp) { vdev_t *vd = mg->mg_vd; objset_t *mos = vd->vdev_spa->spa_meta_objset; metaslab_t *ms; int error; ms = kmem_zalloc(sizeof (metaslab_t), KM_SLEEP); mutex_init(&ms->ms_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&ms->ms_sync_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&ms->ms_load_cv, NULL, CV_DEFAULT, NULL); ms->ms_id = id; ms->ms_start = id << vd->vdev_ms_shift; ms->ms_size = 1ULL << vd->vdev_ms_shift; + ms->ms_allocator = -1; + ms->ms_new = B_TRUE; /* * We only open space map objects that already exist. All others * will be opened when we finally allocate an object for it. */ if (object != 0) { error = space_map_open(&ms->ms_sm, mos, object, ms->ms_start, ms->ms_size, vd->vdev_ashift); if (error != 0) { kmem_free(ms, sizeof (metaslab_t)); return (error); } ASSERT(ms->ms_sm != NULL); } /* * We create the main range tree here, but we don't create the * other range trees until metaslab_sync_done(). This serves * two purposes: it allows metaslab_sync_done() to detect the * addition of new space; and for debugging, it ensures that we'd * data fault on any attempt to use this metaslab before it's ready. */ ms->ms_allocatable = range_tree_create_impl(&rt_avl_ops, &ms->ms_allocatable_by_size, metaslab_rangesize_compare, 0); metaslab_group_add(mg, ms); metaslab_set_fragmentation(ms); /* * If we're opening an existing pool (txg == 0) or creating * a new one (txg == TXG_INITIAL), all space is available now. * If we're adding space to an existing pool, the new space * does not become available until after this txg has synced. * The metaslab's weight will also be initialized when we sync * out this txg. This ensures that we don't attempt to allocate * from it before we have initialized it completely. */ if (txg <= TXG_INITIAL) metaslab_sync_done(ms, 0); /* * If metaslab_debug_load is set and we're initializing a metaslab * that has an allocated space map object then load the its space * map so that can verify frees. */ if (metaslab_debug_load && ms->ms_sm != NULL) { mutex_enter(&ms->ms_lock); VERIFY0(metaslab_load(ms)); mutex_exit(&ms->ms_lock); } if (txg != 0) { vdev_dirty(vd, 0, NULL, txg); vdev_dirty(vd, VDD_METASLAB, ms, txg); } *msp = ms; return (0); } void metaslab_fini(metaslab_t *msp) { metaslab_group_t *mg = msp->ms_group; metaslab_group_remove(mg, msp); mutex_enter(&msp->ms_lock); VERIFY(msp->ms_group == NULL); vdev_space_update(mg->mg_vd, -space_map_allocated(msp->ms_sm), 0, -msp->ms_size); space_map_close(msp->ms_sm); metaslab_unload(msp); range_tree_destroy(msp->ms_allocatable); range_tree_destroy(msp->ms_freeing); range_tree_destroy(msp->ms_freed); for (int t = 0; t < TXG_SIZE; t++) { range_tree_destroy(msp->ms_allocating[t]); } for (int t = 0; t < TXG_DEFER_SIZE; t++) { range_tree_destroy(msp->ms_defer[t]); } ASSERT0(msp->ms_deferspace); range_tree_destroy(msp->ms_checkpointing); mutex_exit(&msp->ms_lock); cv_destroy(&msp->ms_load_cv); mutex_destroy(&msp->ms_lock); mutex_destroy(&msp->ms_sync_lock); + ASSERT3U(msp->ms_allocator, ==, -1); kmem_free(msp, sizeof (metaslab_t)); } #define FRAGMENTATION_TABLE_SIZE 17 /* * This table defines a segment size based fragmentation metric that will * allow each metaslab to derive its own fragmentation value. This is done * by calculating the space in each bucket of the spacemap histogram and * multiplying that by the fragmetation metric in this table. Doing * this for all buckets and dividing it by the total amount of free * space in this metaslab (i.e. the total free space in all buckets) gives * us the fragmentation metric. This means that a high fragmentation metric * equates to most of the free space being comprised of small segments. * Conversely, if the metric is low, then most of the free space is in * large segments. A 10% change in fragmentation equates to approximately * double the number of segments. * * This table defines 0% fragmented space using 16MB segments. Testing has * shown that segments that are greater than or equal to 16MB do not suffer * from drastic performance problems. Using this value, we derive the rest * of the table. Since the fragmentation value is never stored on disk, it * is possible to change these calculations in the future. */ int zfs_frag_table[FRAGMENTATION_TABLE_SIZE] = { 100, /* 512B */ 100, /* 1K */ 98, /* 2K */ 95, /* 4K */ 90, /* 8K */ 80, /* 16K */ 70, /* 32K */ 60, /* 64K */ 50, /* 128K */ 40, /* 256K */ 30, /* 512K */ 20, /* 1M */ 15, /* 2M */ 10, /* 4M */ 5, /* 8M */ 0 /* 16M */ }; /* * Calclate the metaslab's fragmentation metric. A return value * of ZFS_FRAG_INVALID means that the metaslab has not been upgraded and does * not support this metric. Otherwise, the return value should be in the * range [0, 100]. */ static void metaslab_set_fragmentation(metaslab_t *msp) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; uint64_t fragmentation = 0; uint64_t total = 0; boolean_t feature_enabled = spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM); if (!feature_enabled) { msp->ms_fragmentation = ZFS_FRAG_INVALID; return; } /* * A null space map means that the entire metaslab is free * and thus is not fragmented. */ if (msp->ms_sm == NULL) { msp->ms_fragmentation = 0; return; } /* * If this metaslab's space map has not been upgraded, flag it * so that we upgrade next time we encounter it. */ if (msp->ms_sm->sm_dbuf->db_size != sizeof (space_map_phys_t)) { uint64_t txg = spa_syncing_txg(spa); vdev_t *vd = msp->ms_group->mg_vd; /* * If we've reached the final dirty txg, then we must * be shutting down the pool. We don't want to dirty * any data past this point so skip setting the condense * flag. We can retry this action the next time the pool * is imported. */ if (spa_writeable(spa) && txg < spa_final_dirty_txg(spa)) { msp->ms_condense_wanted = B_TRUE; vdev_dirty(vd, VDD_METASLAB, msp, txg + 1); spa_dbgmsg(spa, "txg %llu, requesting force condense: " "ms_id %llu, vdev_id %llu", txg, msp->ms_id, vd->vdev_id); } msp->ms_fragmentation = ZFS_FRAG_INVALID; return; } for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) { uint64_t space = 0; uint8_t shift = msp->ms_sm->sm_shift; int idx = MIN(shift - SPA_MINBLOCKSHIFT + i, FRAGMENTATION_TABLE_SIZE - 1); if (msp->ms_sm->sm_phys->smp_histogram[i] == 0) continue; space = msp->ms_sm->sm_phys->smp_histogram[i] << (i + shift); total += space; ASSERT3U(idx, <, FRAGMENTATION_TABLE_SIZE); fragmentation += space * zfs_frag_table[idx]; } if (total > 0) fragmentation /= total; ASSERT3U(fragmentation, <=, 100); msp->ms_fragmentation = fragmentation; } /* * Compute a weight -- a selection preference value -- for the given metaslab. * This is based on the amount of free space, the level of fragmentation, * the LBA range, and whether the metaslab is loaded. */ static uint64_t metaslab_space_weight(metaslab_t *msp) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; uint64_t weight, space; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(!vd->vdev_removing); /* * The baseline weight is the metaslab's free space. */ space = msp->ms_size - space_map_allocated(msp->ms_sm); if (metaslab_fragmentation_factor_enabled && msp->ms_fragmentation != ZFS_FRAG_INVALID) { /* * Use the fragmentation information to inversely scale * down the baseline weight. We need to ensure that we * don't exclude this metaslab completely when it's 100% * fragmented. To avoid this we reduce the fragmented value * by 1. */ space = (space * (100 - (msp->ms_fragmentation - 1))) / 100; /* * If space < SPA_MINBLOCKSIZE, then we will not allocate from * this metaslab again. The fragmentation metric may have * decreased the space to something smaller than * SPA_MINBLOCKSIZE, so reset the space to SPA_MINBLOCKSIZE * so that we can consume any remaining space. */ if (space > 0 && space < SPA_MINBLOCKSIZE) space = SPA_MINBLOCKSIZE; } weight = space; /* * Modern disks have uniform bit density and constant angular velocity. * Therefore, the outer recording zones are faster (higher bandwidth) * than the inner zones by the ratio of outer to inner track diameter, * which is typically around 2:1. We account for this by assigning * higher weight to lower metaslabs (multiplier ranging from 2x to 1x). * In effect, this means that we'll select the metaslab with the most * free bandwidth rather than simply the one with the most free space. */ if (metaslab_lba_weighting_enabled) { weight = 2 * weight - (msp->ms_id * weight) / vd->vdev_ms_count; ASSERT(weight >= space && weight <= 2 * space); } /* * If this metaslab is one we're actively using, adjust its * weight to make it preferable to any inactive metaslab so * we'll polish it off. If the fragmentation on this metaslab * has exceed our threshold, then don't mark it active. */ if (msp->ms_loaded && msp->ms_fragmentation != ZFS_FRAG_INVALID && msp->ms_fragmentation <= zfs_metaslab_fragmentation_threshold) { weight |= (msp->ms_weight & METASLAB_ACTIVE_MASK); } WEIGHT_SET_SPACEBASED(weight); return (weight); } /* * Return the weight of the specified metaslab, according to the segment-based * weighting algorithm. The metaslab must be loaded. This function can * be called within a sync pass since it relies only on the metaslab's * range tree which is always accurate when the metaslab is loaded. */ static uint64_t metaslab_weight_from_range_tree(metaslab_t *msp) { uint64_t weight = 0; uint32_t segments = 0; ASSERT(msp->ms_loaded); for (int i = RANGE_TREE_HISTOGRAM_SIZE - 1; i >= SPA_MINBLOCKSHIFT; i--) { uint8_t shift = msp->ms_group->mg_vd->vdev_ashift; int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1; segments <<= 1; segments += msp->ms_allocatable->rt_histogram[i]; /* * The range tree provides more precision than the space map * and must be downgraded so that all values fit within the * space map's histogram. This allows us to compare loaded * vs. unloaded metaslabs to determine which metaslab is * considered "best". */ if (i > max_idx) continue; if (segments != 0) { WEIGHT_SET_COUNT(weight, segments); WEIGHT_SET_INDEX(weight, i); WEIGHT_SET_ACTIVE(weight, 0); break; } } return (weight); } /* * Calculate the weight based on the on-disk histogram. This should only * be called after a sync pass has completely finished since the on-disk * information is updated in metaslab_sync(). */ static uint64_t metaslab_weight_from_spacemap(metaslab_t *msp) { uint64_t weight = 0; for (int i = SPACE_MAP_HISTOGRAM_SIZE - 1; i >= 0; i--) { if (msp->ms_sm->sm_phys->smp_histogram[i] != 0) { WEIGHT_SET_COUNT(weight, msp->ms_sm->sm_phys->smp_histogram[i]); WEIGHT_SET_INDEX(weight, i + msp->ms_sm->sm_shift); WEIGHT_SET_ACTIVE(weight, 0); break; } } return (weight); } /* * Compute a segment-based weight for the specified metaslab. The weight * is determined by highest bucket in the histogram. The information * for the highest bucket is encoded into the weight value. */ static uint64_t metaslab_segment_weight(metaslab_t *msp) { metaslab_group_t *mg = msp->ms_group; uint64_t weight = 0; uint8_t shift = mg->mg_vd->vdev_ashift; ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * The metaslab is completely free. */ if (space_map_allocated(msp->ms_sm) == 0) { int idx = highbit64(msp->ms_size) - 1; int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1; if (idx < max_idx) { WEIGHT_SET_COUNT(weight, 1ULL); WEIGHT_SET_INDEX(weight, idx); } else { WEIGHT_SET_COUNT(weight, 1ULL << (idx - max_idx)); WEIGHT_SET_INDEX(weight, max_idx); } WEIGHT_SET_ACTIVE(weight, 0); ASSERT(!WEIGHT_IS_SPACEBASED(weight)); return (weight); } ASSERT3U(msp->ms_sm->sm_dbuf->db_size, ==, sizeof (space_map_phys_t)); /* * If the metaslab is fully allocated then just make the weight 0. */ if (space_map_allocated(msp->ms_sm) == msp->ms_size) return (0); /* * If the metaslab is already loaded, then use the range tree to * determine the weight. Otherwise, we rely on the space map information * to generate the weight. */ if (msp->ms_loaded) { weight = metaslab_weight_from_range_tree(msp); } else { weight = metaslab_weight_from_spacemap(msp); } /* * If the metaslab was active the last time we calculated its weight * then keep it active. We want to consume the entire region that * is associated with this weight. */ if (msp->ms_activation_weight != 0 && weight != 0) WEIGHT_SET_ACTIVE(weight, WEIGHT_GET_ACTIVE(msp->ms_weight)); return (weight); } /* * Determine if we should attempt to allocate from this metaslab. If the * metaslab has a maximum size then we can quickly determine if the desired * allocation size can be satisfied. Otherwise, if we're using segment-based * weighting then we can determine the maximum allocation that this metaslab * can accommodate based on the index encoded in the weight. If we're using * space-based weights then rely on the entire weight (excluding the weight * type bit). */ boolean_t metaslab_should_allocate(metaslab_t *msp, uint64_t asize) { boolean_t should_allocate; if (msp->ms_max_size != 0) return (msp->ms_max_size >= asize); if (!WEIGHT_IS_SPACEBASED(msp->ms_weight)) { /* * The metaslab segment weight indicates segments in the * range [2^i, 2^(i+1)), where i is the index in the weight. * Since the asize might be in the middle of the range, we * should attempt the allocation if asize < 2^(i+1). */ should_allocate = (asize < 1ULL << (WEIGHT_GET_INDEX(msp->ms_weight) + 1)); } else { should_allocate = (asize <= (msp->ms_weight & ~METASLAB_WEIGHT_TYPE)); } return (should_allocate); } static uint64_t metaslab_weight(metaslab_t *msp) { vdev_t *vd = msp->ms_group->mg_vd; spa_t *spa = vd->vdev_spa; uint64_t weight; ASSERT(MUTEX_HELD(&msp->ms_lock)); /* * If this vdev is in the process of being removed, there is nothing * for us to do here. */ if (vd->vdev_removing) return (0); metaslab_set_fragmentation(msp); /* * Update the maximum size if the metaslab is loaded. This will * ensure that we get an accurate maximum size if newly freed space * has been added back into the free tree. */ if (msp->ms_loaded) msp->ms_max_size = metaslab_block_maxsize(msp); /* * Segment-based weighting requires space map histogram support. */ if (zfs_metaslab_segment_weight_enabled && spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM) && (msp->ms_sm == NULL || msp->ms_sm->sm_dbuf->db_size == sizeof (space_map_phys_t))) { weight = metaslab_segment_weight(msp); } else { weight = metaslab_space_weight(msp); } return (weight); } static int -metaslab_activate(metaslab_t *msp, uint64_t activation_weight) +metaslab_activate_allocator(metaslab_group_t *mg, metaslab_t *msp, + int allocator, uint64_t activation_weight) { + /* + * If we're activating for the claim code, we don't want to actually + * set the metaslab up for a specific allocator. + */ + if (activation_weight == METASLAB_WEIGHT_CLAIM) + return (0); + metaslab_t **arr = (activation_weight == METASLAB_WEIGHT_PRIMARY ? + mg->mg_primaries : mg->mg_secondaries); + ASSERT(MUTEX_HELD(&msp->ms_lock)); + mutex_enter(&mg->mg_lock); + if (arr[allocator] != NULL) { + mutex_exit(&mg->mg_lock); + return (EEXIST); + } + arr[allocator] = msp; + ASSERT3S(msp->ms_allocator, ==, -1); + msp->ms_allocator = allocator; + msp->ms_primary = (activation_weight == METASLAB_WEIGHT_PRIMARY); + mutex_exit(&mg->mg_lock); + + return (0); +} + +static int +metaslab_activate(metaslab_t *msp, int allocator, uint64_t activation_weight) +{ + ASSERT(MUTEX_HELD(&msp->ms_lock)); + if ((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0) { + int error = 0; metaslab_load_wait(msp); if (!msp->ms_loaded) { - int error = metaslab_load(msp); - if (error) { + if ((error = metaslab_load(msp)) != 0) { metaslab_group_sort(msp->ms_group, msp, 0); return (error); } } + if ((msp->ms_weight & METASLAB_ACTIVE_MASK) != 0) { + /* + * The metaslab was activated for another allocator + * while we were waiting, we should reselect. + */ + return (EBUSY); + } + if ((error = metaslab_activate_allocator(msp->ms_group, msp, + allocator, activation_weight)) != 0) { + return (error); + } msp->ms_activation_weight = msp->ms_weight; metaslab_group_sort(msp->ms_group, msp, msp->ms_weight | activation_weight); } ASSERT(msp->ms_loaded); ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK); return (0); } static void +metaslab_passivate_allocator(metaslab_group_t *mg, metaslab_t *msp, + uint64_t weight) +{ + ASSERT(MUTEX_HELD(&msp->ms_lock)); + if (msp->ms_weight & METASLAB_WEIGHT_CLAIM) { + metaslab_group_sort(mg, msp, weight); + return; + } + + mutex_enter(&mg->mg_lock); + ASSERT3P(msp->ms_group, ==, mg); + if (msp->ms_primary) { + ASSERT3U(0, <=, msp->ms_allocator); + ASSERT3U(msp->ms_allocator, <, mg->mg_allocators); + ASSERT3P(mg->mg_primaries[msp->ms_allocator], ==, msp); + ASSERT(msp->ms_weight & METASLAB_WEIGHT_PRIMARY); + mg->mg_primaries[msp->ms_allocator] = NULL; + } else { + ASSERT(msp->ms_weight & METASLAB_WEIGHT_SECONDARY); + ASSERT3P(mg->mg_secondaries[msp->ms_allocator], ==, msp); + mg->mg_secondaries[msp->ms_allocator] = NULL; + } + msp->ms_allocator = -1; + metaslab_group_sort_impl(mg, msp, weight); + mutex_exit(&mg->mg_lock); +} + +static void metaslab_passivate(metaslab_t *msp, uint64_t weight) { uint64_t size = weight & ~METASLAB_WEIGHT_TYPE; /* * If size < SPA_MINBLOCKSIZE, then we will not allocate from * this metaslab again. In that case, it had better be empty, * or we would be leaving space on the table. */ ASSERT(size >= SPA_MINBLOCKSIZE || range_tree_is_empty(msp->ms_allocatable)); ASSERT0(weight & METASLAB_ACTIVE_MASK); msp->ms_activation_weight = 0; - metaslab_group_sort(msp->ms_group, msp, weight); + metaslab_passivate_allocator(msp->ms_group, msp, weight); ASSERT((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0); } /* * Segment-based metaslabs are activated once and remain active until * we either fail an allocation attempt (similar to space-based metaslabs) * or have exhausted the free space in zfs_metaslab_switch_threshold * buckets since the metaslab was activated. This function checks to see * if we've exhaused the zfs_metaslab_switch_threshold buckets in the * metaslab and passivates it proactively. This will allow us to select a * metaslabs with larger contiguous region if any remaining within this * metaslab group. If we're in sync pass > 1, then we continue using this * metaslab so that we don't dirty more block and cause more sync passes. */ void metaslab_segment_may_passivate(metaslab_t *msp) { spa_t *spa = msp->ms_group->mg_vd->vdev_spa; if (WEIGHT_IS_SPACEBASED(msp->ms_weight) || spa_sync_pass(spa) > 1) return; /* * Since we are in the middle of a sync pass, the most accurate * information that is accessible to us is the in-core range tree * histogram; calculate the new weight based on that information. */ uint64_t weight = metaslab_weight_from_range_tree(msp); int activation_idx = WEIGHT_GET_INDEX(msp->ms_activation_weight); int current_idx = WEIGHT_GET_INDEX(weight); if (current_idx <= activation_idx - zfs_metaslab_switch_threshold) metaslab_passivate(msp, weight); } static void metaslab_preload(void *arg) { metaslab_t *msp = arg; spa_t *spa = msp->ms_group->mg_vd->vdev_spa; ASSERT(!MUTEX_HELD(&msp->ms_group->mg_lock)); mutex_enter(&msp->ms_lock); metaslab_load_wait(msp); if (!msp->ms_loaded) (void) metaslab_load(msp); msp->ms_selected_txg = spa_syncing_txg(spa); mutex_exit(&msp->ms_lock); } static void metaslab_group_preload(metaslab_group_t *mg) { spa_t *spa = mg->mg_vd->vdev_spa; metaslab_t *msp; avl_tree_t *t = &mg->mg_metaslab_tree; int m = 0; if (spa_shutting_down(spa) || !metaslab_preload_enabled) { taskq_wait(mg->mg_taskq); return; } mutex_enter(&mg->mg_lock); /* * Load the next potential metaslabs */ for (msp = avl_first(t); msp != NULL; msp = AVL_NEXT(t, msp)) { ASSERT3P(msp->ms_group, ==, mg); /* * We preload only the maximum number of metaslabs specified * by metaslab_preload_limit. If a metaslab is being forced * to condense then we preload it too. This will ensure * that force condensing happens in the next txg. */ if (++m > metaslab_preload_limit && !msp->ms_condense_wanted) { continue; } VERIFY(taskq_dispatch(mg->mg_taskq, metaslab_preload, msp, TQ_SLEEP) != 0); } mutex_exit(&mg->mg_lock); } /* * Determine if the space map's on-disk footprint is past our tolerance * for inefficiency. We would like to use the following criteria to make * our decision: * * 1. The size of the space map object should not dramatically increase as a * result of writing out the free space range tree. * * 2. The minimal on-disk space map representation is zfs_condense_pct/100 * times the size than the free space range tree representation * (i.e. zfs_condense_pct = 110 and in-core = 1MB, minimal = 1.1MB). * * 3. The on-disk size of the space map should actually decrease. * * Unfortunately, we cannot compute the on-disk size of the space map in this * context because we cannot accurately compute the effects of compression, etc. * Instead, we apply the heuristic described in the block comment for * zfs_metaslab_condense_block_threshold - we only condense if the space used * is greater than a threshold number of blocks. */ static boolean_t metaslab_should_condense(metaslab_t *msp) { space_map_t *sm = msp->ms_sm; vdev_t *vd = msp->ms_group->mg_vd; uint64_t vdev_blocksize = 1 << vd->vdev_ashift; uint64_t current_txg = spa_syncing_txg(vd->vdev_spa); ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(msp->ms_loaded); /* * Allocations and frees in early passes are generally more space * efficient (in terms of blocks described in space map entries) * than the ones in later passes (e.g. we don't compress after * sync pass 5) and condensing a metaslab multiple times in a txg * could degrade performance. * * Thus we prefer condensing each metaslab at most once every txg at * the earliest sync pass possible. If a metaslab is eligible for * condensing again after being considered for condensing within the * same txg, it will hopefully be dirty in the next txg where it will * be condensed at an earlier pass. */ if (msp->ms_condense_checked_txg == current_txg) return (B_FALSE); msp->ms_condense_checked_txg = current_txg; /* * We always condense metaslabs that are empty and metaslabs for * which a condense request has been made. */ if (avl_is_empty(&msp->ms_allocatable_by_size) || msp->ms_condense_wanted) return (B_TRUE); uint64_t object_size = space_map_length(msp->ms_sm); uint64_t optimal_size = space_map_estimate_optimal_size(sm, msp->ms_allocatable, SM_NO_VDEVID); dmu_object_info_t doi; dmu_object_info_from_db(sm->sm_dbuf, &doi); uint64_t record_size = MAX(doi.doi_data_block_size, vdev_blocksize); return (object_size >= (optimal_size * zfs_condense_pct / 100) && object_size > zfs_metaslab_condense_block_threshold * record_size); } /* * Condense the on-disk space map representation to its minimized form. * The minimized form consists of a small number of allocations followed by * the entries of the free range tree. */ static void metaslab_condense(metaslab_t *msp, uint64_t txg, dmu_tx_t *tx) { range_tree_t *condense_tree; space_map_t *sm = msp->ms_sm; ASSERT(MUTEX_HELD(&msp->ms_lock)); ASSERT(msp->ms_loaded); zfs_dbgmsg("condensing: txg %llu, msp[%llu] %p, vdev id %llu, " "spa %s, smp size %llu, segments %lu, forcing condense=%s", txg, msp->ms_id, msp, msp->ms_group->mg_vd->vdev_id, msp->ms_group->mg_vd->vdev_spa->spa_name, space_map_length(msp->ms_sm), avl_numnodes(&msp->ms_allocatable->rt_root), msp->ms_condense_wanted ? "TRUE" : "FALSE"); msp->ms_condense_wanted = B_FALSE; /* * Create an range tree that is 100% allocated. We remove segments * that have been freed in this txg, any deferred frees that exist, * and any allocation in the future. Removing segments should be * a relatively inexpensive operation since we expect these trees to * have a small number of nodes. */ condense_tree = range_tree_create(NULL, NULL); range_tree_add(condense_tree, msp->ms_start, msp->ms_size); range_tree_walk(msp->ms_freeing, range_tree_remove, condense_tree); range_tree_walk(msp->ms_freed, range_tree_remove, condense_tree); for (int t = 0; t < TXG_DEFER_SIZE; t++) { range_tree_walk(msp->ms_defer[t], range_tree_remove, condense_tree); } for (int t = 1; t < TXG_CONCURRENT_STATES; t++) { range_tree_walk(msp->ms_allocating[(txg + t) & TXG_MASK], range_tree_remove, condense_tree); } /* * We're about to drop the metaslab's lock thus allowing * other consumers to change it's content. Set the * metaslab's ms_condensing flag to ensure that * allocations on this metaslab do not occur while we're * in the middle of committing it to disk. This is only critical * for ms_allocatable as all other range trees use per txg * views of their content. */ msp->ms_condensing = B_TRUE; mutex_exit(&msp->ms_lock); space_map_truncate(sm, zfs_metaslab_sm_blksz, tx); /* * While we would ideally like to create a space map representation * that consists only of allocation records, doing so can be * prohibitively expensive because the in-core free tree can be * large, and therefore computationally expensive to subtract * from the condense_tree. Instead we sync out two trees, a cheap * allocation only tree followed by the in-core free tree. While not * optimal, this is typically close to optimal, and much cheaper to * compute. */ space_map_write(sm, condense_tree, SM_ALLOC, SM_NO_VDEVID, tx); range_tree_vacate(condense_tree, NULL, NULL); range_tree_destroy(condense_tree); space_map_write(sm, msp->ms_allocatable, SM_FREE, SM_NO_VDEVID, tx); mutex_enter(&msp->ms_lock); msp->ms_condensing = B_FALSE; } /* * Write a metaslab to disk in the context of the specified transaction group. */ void metaslab_sync(metaslab_t *msp, uint64_t txg) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; spa_t *spa = vd->vdev_spa; objset_t *mos = spa_meta_objset(spa); range_tree_t *alloctree = msp->ms_allocating[txg & TXG_MASK]; dmu_tx_t *tx; uint64_t object = space_map_object(msp->ms_sm); ASSERT(!vd->vdev_ishole); /* * This metaslab has just been added so there's no work to do now. */ if (msp->ms_freeing == NULL) { ASSERT3P(alloctree, ==, NULL); return; } ASSERT3P(alloctree, !=, NULL); ASSERT3P(msp->ms_freeing, !=, NULL); ASSERT3P(msp->ms_freed, !=, NULL); ASSERT3P(msp->ms_checkpointing, !=, NULL); /* * Normally, we don't want to process a metaslab if there are no * allocations or frees to perform. However, if the metaslab is being * forced to condense and it's loaded, we need to let it through. */ if (range_tree_is_empty(alloctree) && range_tree_is_empty(msp->ms_freeing) && range_tree_is_empty(msp->ms_checkpointing) && !(msp->ms_loaded && msp->ms_condense_wanted)) return; VERIFY(txg <= spa_final_dirty_txg(spa)); /* * The only state that can actually be changing concurrently with * metaslab_sync() is the metaslab's ms_allocatable. No other * thread can be modifying this txg's alloc, freeing, * freed, or space_map_phys_t. We drop ms_lock whenever we * could call into the DMU, because the DMU can call down to us * (e.g. via zio_free()) at any time. * * The spa_vdev_remove_thread() can be reading metaslab state * concurrently, and it is locked out by the ms_sync_lock. Note * that the ms_lock is insufficient for this, because it is dropped * by space_map_write(). */ tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); if (msp->ms_sm == NULL) { uint64_t new_object; new_object = space_map_alloc(mos, zfs_metaslab_sm_blksz, tx); VERIFY3U(new_object, !=, 0); VERIFY0(space_map_open(&msp->ms_sm, mos, new_object, msp->ms_start, msp->ms_size, vd->vdev_ashift)); ASSERT(msp->ms_sm != NULL); } if (!range_tree_is_empty(msp->ms_checkpointing) && vd->vdev_checkpoint_sm == NULL) { ASSERT(spa_has_checkpoint(spa)); uint64_t new_object = space_map_alloc(mos, vdev_standard_sm_blksz, tx); VERIFY3U(new_object, !=, 0); VERIFY0(space_map_open(&vd->vdev_checkpoint_sm, mos, new_object, 0, vd->vdev_asize, vd->vdev_ashift)); ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); /* * We save the space map object as an entry in vdev_top_zap * so it can be retrieved when the pool is reopened after an * export or through zdb. */ VERIFY0(zap_add(vd->vdev_spa->spa_meta_objset, vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (new_object), 1, &new_object, tx)); } mutex_enter(&msp->ms_sync_lock); mutex_enter(&msp->ms_lock); /* * Note: metaslab_condense() clears the space map's histogram. * Therefore we must verify and remove this histogram before * condensing. */ metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); metaslab_group_histogram_remove(mg, msp); if (msp->ms_loaded && metaslab_should_condense(msp)) { metaslab_condense(msp, txg, tx); } else { mutex_exit(&msp->ms_lock); space_map_write(msp->ms_sm, alloctree, SM_ALLOC, SM_NO_VDEVID, tx); space_map_write(msp->ms_sm, msp->ms_freeing, SM_FREE, SM_NO_VDEVID, tx); mutex_enter(&msp->ms_lock); } if (!range_tree_is_empty(msp->ms_checkpointing)) { ASSERT(spa_has_checkpoint(spa)); ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); /* * Since we are doing writes to disk and the ms_checkpointing * tree won't be changing during that time, we drop the * ms_lock while writing to the checkpoint space map. */ mutex_exit(&msp->ms_lock); space_map_write(vd->vdev_checkpoint_sm, msp->ms_checkpointing, SM_FREE, SM_NO_VDEVID, tx); mutex_enter(&msp->ms_lock); space_map_update(vd->vdev_checkpoint_sm); spa->spa_checkpoint_info.sci_dspace += range_tree_space(msp->ms_checkpointing); vd->vdev_stat.vs_checkpoint_space += range_tree_space(msp->ms_checkpointing); ASSERT3U(vd->vdev_stat.vs_checkpoint_space, ==, -vd->vdev_checkpoint_sm->sm_alloc); range_tree_vacate(msp->ms_checkpointing, NULL, NULL); } if (msp->ms_loaded) { /* * When the space map is loaded, we have an accurate * histogram in the range tree. This gives us an opportunity * to bring the space map's histogram up-to-date so we clear * it first before updating it. */ space_map_histogram_clear(msp->ms_sm); space_map_histogram_add(msp->ms_sm, msp->ms_allocatable, tx); /* * Since we've cleared the histogram we need to add back * any free space that has already been processed, plus * any deferred space. This allows the on-disk histogram * to accurately reflect all free space even if some space * is not yet available for allocation (i.e. deferred). */ space_map_histogram_add(msp->ms_sm, msp->ms_freed, tx); /* * Add back any deferred free space that has not been * added back into the in-core free tree yet. This will * ensure that we don't end up with a space map histogram * that is completely empty unless the metaslab is fully * allocated. */ for (int t = 0; t < TXG_DEFER_SIZE; t++) { space_map_histogram_add(msp->ms_sm, msp->ms_defer[t], tx); } } /* * Always add the free space from this sync pass to the space * map histogram. We want to make sure that the on-disk histogram * accounts for all free space. If the space map is not loaded, * then we will lose some accuracy but will correct it the next * time we load the space map. */ space_map_histogram_add(msp->ms_sm, msp->ms_freeing, tx); metaslab_group_histogram_add(mg, msp); metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); /* * For sync pass 1, we avoid traversing this txg's free range tree * and instead will just swap the pointers for freeing and * freed. We can safely do this since the freed_tree is * guaranteed to be empty on the initial pass. */ if (spa_sync_pass(spa) == 1) { range_tree_swap(&msp->ms_freeing, &msp->ms_freed); } else { range_tree_vacate(msp->ms_freeing, range_tree_add, msp->ms_freed); } range_tree_vacate(alloctree, NULL, NULL); ASSERT0(range_tree_space(msp->ms_allocating[txg & TXG_MASK])); ASSERT0(range_tree_space(msp->ms_allocating[TXG_CLEAN(txg) & TXG_MASK])); ASSERT0(range_tree_space(msp->ms_freeing)); ASSERT0(range_tree_space(msp->ms_checkpointing)); mutex_exit(&msp->ms_lock); if (object != space_map_object(msp->ms_sm)) { object = space_map_object(msp->ms_sm); dmu_write(mos, vd->vdev_ms_array, sizeof (uint64_t) * msp->ms_id, sizeof (uint64_t), &object, tx); } mutex_exit(&msp->ms_sync_lock); dmu_tx_commit(tx); } /* * Called after a transaction group has completely synced to mark * all of the metaslab's free space as usable. */ void metaslab_sync_done(metaslab_t *msp, uint64_t txg) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; spa_t *spa = vd->vdev_spa; range_tree_t **defer_tree; int64_t alloc_delta, defer_delta; boolean_t defer_allowed = B_TRUE; ASSERT(!vd->vdev_ishole); mutex_enter(&msp->ms_lock); /* * If this metaslab is just becoming available, initialize its * range trees and add its capacity to the vdev. */ if (msp->ms_freed == NULL) { for (int t = 0; t < TXG_SIZE; t++) { ASSERT(msp->ms_allocating[t] == NULL); msp->ms_allocating[t] = range_tree_create(NULL, NULL); } ASSERT3P(msp->ms_freeing, ==, NULL); msp->ms_freeing = range_tree_create(NULL, NULL); ASSERT3P(msp->ms_freed, ==, NULL); msp->ms_freed = range_tree_create(NULL, NULL); for (int t = 0; t < TXG_DEFER_SIZE; t++) { ASSERT(msp->ms_defer[t] == NULL); msp->ms_defer[t] = range_tree_create(NULL, NULL); } ASSERT3P(msp->ms_checkpointing, ==, NULL); msp->ms_checkpointing = range_tree_create(NULL, NULL); vdev_space_update(vd, 0, 0, msp->ms_size); } ASSERT0(range_tree_space(msp->ms_freeing)); ASSERT0(range_tree_space(msp->ms_checkpointing)); defer_tree = &msp->ms_defer[txg % TXG_DEFER_SIZE]; uint64_t free_space = metaslab_class_get_space(spa_normal_class(spa)) - metaslab_class_get_alloc(spa_normal_class(spa)); if (free_space <= spa_get_slop_space(spa) || vd->vdev_removing) { defer_allowed = B_FALSE; } defer_delta = 0; alloc_delta = space_map_alloc_delta(msp->ms_sm); if (defer_allowed) { defer_delta = range_tree_space(msp->ms_freed) - range_tree_space(*defer_tree); } else { defer_delta -= range_tree_space(*defer_tree); } vdev_space_update(vd, alloc_delta + defer_delta, defer_delta, 0); /* * If there's a metaslab_load() in progress, wait for it to complete * so that we have a consistent view of the in-core space map. */ metaslab_load_wait(msp); /* * Move the frees from the defer_tree back to the free * range tree (if it's loaded). Swap the freed_tree and * the defer_tree -- this is safe to do because we've * just emptied out the defer_tree. */ range_tree_vacate(*defer_tree, msp->ms_loaded ? range_tree_add : NULL, msp->ms_allocatable); if (defer_allowed) { range_tree_swap(&msp->ms_freed, defer_tree); } else { range_tree_vacate(msp->ms_freed, msp->ms_loaded ? range_tree_add : NULL, msp->ms_allocatable); } space_map_update(msp->ms_sm); msp->ms_deferspace += defer_delta; ASSERT3S(msp->ms_deferspace, >=, 0); ASSERT3S(msp->ms_deferspace, <=, msp->ms_size); if (msp->ms_deferspace != 0) { /* * Keep syncing this metaslab until all deferred frees * are back in circulation. */ vdev_dirty(vd, VDD_METASLAB, msp, txg + 1); } + if (msp->ms_new) { + msp->ms_new = B_FALSE; + mutex_enter(&mg->mg_lock); + mg->mg_ms_ready++; + mutex_exit(&mg->mg_lock); + } /* * Calculate the new weights before unloading any metaslabs. * This will give us the most accurate weighting. */ - metaslab_group_sort(mg, msp, metaslab_weight(msp)); + metaslab_group_sort(mg, msp, metaslab_weight(msp) | + (msp->ms_weight & METASLAB_ACTIVE_MASK)); /* * If the metaslab is loaded and we've not tried to load or allocate * from it in 'metaslab_unload_delay' txgs, then unload it. */ if (msp->ms_loaded && msp->ms_selected_txg + metaslab_unload_delay < txg) { for (int t = 1; t < TXG_CONCURRENT_STATES; t++) { VERIFY0(range_tree_space( msp->ms_allocating[(txg + t) & TXG_MASK])); } + if (msp->ms_allocator != -1) { + metaslab_passivate(msp, msp->ms_weight & + ~METASLAB_ACTIVE_MASK); + } if (!metaslab_debug_unload) metaslab_unload(msp); } ASSERT0(range_tree_space(msp->ms_allocating[txg & TXG_MASK])); ASSERT0(range_tree_space(msp->ms_freeing)); ASSERT0(range_tree_space(msp->ms_freed)); ASSERT0(range_tree_space(msp->ms_checkpointing)); mutex_exit(&msp->ms_lock); } void metaslab_sync_reassess(metaslab_group_t *mg) { spa_t *spa = mg->mg_class->mc_spa; spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER); metaslab_group_alloc_update(mg); mg->mg_fragmentation = metaslab_group_fragmentation(mg); /* * Preload the next potential metaslabs but only on active * metaslab groups. We can get into a state where the metaslab * is no longer active since we dirty metaslabs as we remove a * a device, thus potentially making the metaslab group eligible * for preloading. */ if (mg->mg_activation_count > 0) { metaslab_group_preload(mg); } spa_config_exit(spa, SCL_ALLOC, FTAG); } static uint64_t metaslab_distance(metaslab_t *msp, dva_t *dva) { uint64_t ms_shift = msp->ms_group->mg_vd->vdev_ms_shift; uint64_t offset = DVA_GET_OFFSET(dva) >> ms_shift; uint64_t start = msp->ms_id; if (msp->ms_group->mg_vd->vdev_id != DVA_GET_VDEV(dva)) return (1ULL << 63); if (offset < start) return ((start - offset) << ms_shift); if (offset > start) return ((offset - start) << ms_shift); return (0); } /* * ========================================================================== * Metaslab allocation tracing facility * ========================================================================== */ kstat_t *metaslab_trace_ksp; kstat_named_t metaslab_trace_over_limit; void metaslab_alloc_trace_init(void) { ASSERT(metaslab_alloc_trace_cache == NULL); metaslab_alloc_trace_cache = kmem_cache_create( "metaslab_alloc_trace_cache", sizeof (metaslab_alloc_trace_t), 0, NULL, NULL, NULL, NULL, NULL, 0); metaslab_trace_ksp = kstat_create("zfs", 0, "metaslab_trace_stats", "misc", KSTAT_TYPE_NAMED, 1, KSTAT_FLAG_VIRTUAL); if (metaslab_trace_ksp != NULL) { metaslab_trace_ksp->ks_data = &metaslab_trace_over_limit; kstat_named_init(&metaslab_trace_over_limit, "metaslab_trace_over_limit", KSTAT_DATA_UINT64); kstat_install(metaslab_trace_ksp); } } void metaslab_alloc_trace_fini(void) { if (metaslab_trace_ksp != NULL) { kstat_delete(metaslab_trace_ksp); metaslab_trace_ksp = NULL; } kmem_cache_destroy(metaslab_alloc_trace_cache); metaslab_alloc_trace_cache = NULL; } /* * Add an allocation trace element to the allocation tracing list. */ static void metaslab_trace_add(zio_alloc_list_t *zal, metaslab_group_t *mg, - metaslab_t *msp, uint64_t psize, uint32_t dva_id, uint64_t offset) + metaslab_t *msp, uint64_t psize, uint32_t dva_id, uint64_t offset, + int allocator) { if (!metaslab_trace_enabled) return; /* * When the tracing list reaches its maximum we remove * the second element in the list before adding a new one. * By removing the second element we preserve the original * entry as a clue to what allocations steps have already been * performed. */ if (zal->zal_size == metaslab_trace_max_entries) { metaslab_alloc_trace_t *mat_next; #ifdef DEBUG panic("too many entries in allocation list"); #endif atomic_inc_64(&metaslab_trace_over_limit.value.ui64); zal->zal_size--; mat_next = list_next(&zal->zal_list, list_head(&zal->zal_list)); list_remove(&zal->zal_list, mat_next); kmem_cache_free(metaslab_alloc_trace_cache, mat_next); } metaslab_alloc_trace_t *mat = kmem_cache_alloc(metaslab_alloc_trace_cache, KM_SLEEP); list_link_init(&mat->mat_list_node); mat->mat_mg = mg; mat->mat_msp = msp; mat->mat_size = psize; mat->mat_dva_id = dva_id; mat->mat_offset = offset; mat->mat_weight = 0; + mat->mat_allocator = allocator; if (msp != NULL) mat->mat_weight = msp->ms_weight; /* * The list is part of the zio so locking is not required. Only * a single thread will perform allocations for a given zio. */ list_insert_tail(&zal->zal_list, mat); zal->zal_size++; ASSERT3U(zal->zal_size, <=, metaslab_trace_max_entries); } void metaslab_trace_init(zio_alloc_list_t *zal) { list_create(&zal->zal_list, sizeof (metaslab_alloc_trace_t), offsetof(metaslab_alloc_trace_t, mat_list_node)); zal->zal_size = 0; } void metaslab_trace_fini(zio_alloc_list_t *zal) { metaslab_alloc_trace_t *mat; while ((mat = list_remove_head(&zal->zal_list)) != NULL) kmem_cache_free(metaslab_alloc_trace_cache, mat); list_destroy(&zal->zal_list); zal->zal_size = 0; } /* * ========================================================================== * Metaslab block operations * ========================================================================== */ static void -metaslab_group_alloc_increment(spa_t *spa, uint64_t vdev, void *tag, int flags) +metaslab_group_alloc_increment(spa_t *spa, uint64_t vdev, void *tag, int flags, + int allocator) { if (!(flags & METASLAB_ASYNC_ALLOC) || - flags & METASLAB_DONT_THROTTLE) + (flags & METASLAB_DONT_THROTTLE)) return; metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg; if (!mg->mg_class->mc_alloc_throttle_enabled) return; - (void) refcount_add(&mg->mg_alloc_queue_depth, tag); + (void) refcount_add(&mg->mg_alloc_queue_depth[allocator], tag); } +static void +metaslab_group_increment_qdepth(metaslab_group_t *mg, int allocator) +{ + uint64_t max = mg->mg_max_alloc_queue_depth; + uint64_t cur = mg->mg_cur_max_alloc_queue_depth[allocator]; + while (cur < max) { + if (atomic_cas_64(&mg->mg_cur_max_alloc_queue_depth[allocator], + cur, cur + 1) == cur) { + atomic_inc_64( + &mg->mg_class->mc_alloc_max_slots[allocator]); + return; + } + cur = mg->mg_cur_max_alloc_queue_depth[allocator]; + } +} + void -metaslab_group_alloc_decrement(spa_t *spa, uint64_t vdev, void *tag, int flags) +metaslab_group_alloc_decrement(spa_t *spa, uint64_t vdev, void *tag, int flags, + int allocator, boolean_t io_complete) { if (!(flags & METASLAB_ASYNC_ALLOC) || - flags & METASLAB_DONT_THROTTLE) + (flags & METASLAB_DONT_THROTTLE)) return; metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg; if (!mg->mg_class->mc_alloc_throttle_enabled) return; - (void) refcount_remove(&mg->mg_alloc_queue_depth, tag); + (void) refcount_remove(&mg->mg_alloc_queue_depth[allocator], tag); + if (io_complete) + metaslab_group_increment_qdepth(mg, allocator); } void -metaslab_group_alloc_verify(spa_t *spa, const blkptr_t *bp, void *tag) +metaslab_group_alloc_verify(spa_t *spa, const blkptr_t *bp, void *tag, + int allocator) { #ifdef ZFS_DEBUG const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); for (int d = 0; d < ndvas; d++) { uint64_t vdev = DVA_GET_VDEV(&dva[d]); metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg; - VERIFY(refcount_not_held(&mg->mg_alloc_queue_depth, tag)); + VERIFY(refcount_not_held(&mg->mg_alloc_queue_depth[allocator], + tag)); } #endif } static uint64_t metaslab_block_alloc(metaslab_t *msp, uint64_t size, uint64_t txg) { uint64_t start; range_tree_t *rt = msp->ms_allocatable; metaslab_class_t *mc = msp->ms_group->mg_class; VERIFY(!msp->ms_condensing); start = mc->mc_ops->msop_alloc(msp, size); if (start != -1ULL) { metaslab_group_t *mg = msp->ms_group; vdev_t *vd = mg->mg_vd; VERIFY0(P2PHASE(start, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift)); VERIFY3U(range_tree_space(rt) - size, <=, msp->ms_size); range_tree_remove(rt, start, size); if (range_tree_is_empty(msp->ms_allocating[txg & TXG_MASK])) vdev_dirty(mg->mg_vd, VDD_METASLAB, msp, txg); range_tree_add(msp->ms_allocating[txg & TXG_MASK], start, size); /* Track the last successful allocation */ msp->ms_alloc_txg = txg; metaslab_verify_space(msp, txg); } /* * Now that we've attempted the allocation we need to update the * metaslab's maximum block size since it may have changed. */ msp->ms_max_size = metaslab_block_maxsize(msp); return (start); } +/* + * Find the metaslab with the highest weight that is less than what we've + * already tried. In the common case, this means that we will examine each + * metaslab at most once. Note that concurrent callers could reorder metaslabs + * by activation/passivation once we have dropped the mg_lock. If a metaslab is + * activated by another thread, and we fail to allocate from the metaslab we + * have selected, we may not try the newly-activated metaslab, and instead + * activate another metaslab. This is not optimal, but generally does not cause + * any problems (a possible exception being if every metaslab is completely full + * except for the the newly-activated metaslab which we fail to examine). + */ +static metaslab_t * +find_valid_metaslab(metaslab_group_t *mg, uint64_t activation_weight, + dva_t *dva, int d, uint64_t min_distance, uint64_t asize, int allocator, + zio_alloc_list_t *zal, metaslab_t *search, boolean_t *was_active) +{ + avl_index_t idx; + avl_tree_t *t = &mg->mg_metaslab_tree; + metaslab_t *msp = avl_find(t, search, &idx); + if (msp == NULL) + msp = avl_nearest(t, idx, AVL_AFTER); + + for (; msp != NULL; msp = AVL_NEXT(t, msp)) { + int i; + if (!metaslab_should_allocate(msp, asize)) { + metaslab_trace_add(zal, mg, msp, asize, d, + TRACE_TOO_SMALL, allocator); + continue; + } + + /* + * If the selected metaslab is condensing, skip it. + */ + if (msp->ms_condensing) + continue; + + *was_active = msp->ms_allocator != -1; + /* + * If we're activating as primary, this is our first allocation + * from this disk, so we don't need to check how close we are. + * If the metaslab under consideration was already active, + * we're getting desperate enough to steal another allocator's + * metaslab, so we still don't care about distances. + */ + if (activation_weight == METASLAB_WEIGHT_PRIMARY || *was_active) + break; + + uint64_t target_distance = min_distance + + (space_map_allocated(msp->ms_sm) != 0 ? 0 : + min_distance >> 1); + + for (i = 0; i < d; i++) { + if (metaslab_distance(msp, &dva[i]) < target_distance) + break; + } + if (i == d) + break; + } + + if (msp != NULL) { + search->ms_weight = msp->ms_weight; + search->ms_start = msp->ms_start + 1; + search->ms_allocator = msp->ms_allocator; + search->ms_primary = msp->ms_primary; + } + return (msp); +} + +/* ARGSUSED */ static uint64_t metaslab_group_alloc_normal(metaslab_group_t *mg, zio_alloc_list_t *zal, - uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d) + uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d, + int allocator) { metaslab_t *msp = NULL; uint64_t offset = -1ULL; uint64_t activation_weight; - uint64_t target_distance; - int i; + boolean_t tertiary = B_FALSE; activation_weight = METASLAB_WEIGHT_PRIMARY; - for (i = 0; i < d; i++) { - if (DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) { + for (int i = 0; i < d; i++) { + if (activation_weight == METASLAB_WEIGHT_PRIMARY && + DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) { activation_weight = METASLAB_WEIGHT_SECONDARY; + } else if (activation_weight == METASLAB_WEIGHT_SECONDARY && + DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) { + tertiary = B_TRUE; break; } } + /* + * If we don't have enough metaslabs active to fill the entire array, we + * just use the 0th slot. + */ + if (mg->mg_ms_ready < mg->mg_allocators * 2) { + tertiary = B_FALSE; + allocator = 0; + } + + ASSERT3U(mg->mg_vd->vdev_ms_count, >=, 2); + metaslab_t *search = kmem_alloc(sizeof (*search), KM_SLEEP); search->ms_weight = UINT64_MAX; search->ms_start = 0; + /* + * At the end of the metaslab tree are the already-active metaslabs, + * first the primaries, then the secondaries. When we resume searching + * through the tree, we need to consider ms_allocator and ms_primary so + * we start in the location right after where we left off, and don't + * accidentally loop forever considering the same metaslabs. + */ + search->ms_allocator = -1; + search->ms_primary = B_TRUE; for (;;) { - boolean_t was_active; - avl_tree_t *t = &mg->mg_metaslab_tree; - avl_index_t idx; + boolean_t was_active = B_FALSE; mutex_enter(&mg->mg_lock); - /* - * Find the metaslab with the highest weight that is less - * than what we've already tried. In the common case, this - * means that we will examine each metaslab at most once. - * Note that concurrent callers could reorder metaslabs - * by activation/passivation once we have dropped the mg_lock. - * If a metaslab is activated by another thread, and we fail - * to allocate from the metaslab we have selected, we may - * not try the newly-activated metaslab, and instead activate - * another metaslab. This is not optimal, but generally - * does not cause any problems (a possible exception being - * if every metaslab is completely full except for the - * the newly-activated metaslab which we fail to examine). - */ - msp = avl_find(t, search, &idx); - if (msp == NULL) - msp = avl_nearest(t, idx, AVL_AFTER); - for (; msp != NULL; msp = AVL_NEXT(t, msp)) { - - if (!metaslab_should_allocate(msp, asize)) { - metaslab_trace_add(zal, mg, msp, asize, d, - TRACE_TOO_SMALL); - continue; - } - - /* - * If the selected metaslab is condensing, skip it. - */ - if (msp->ms_condensing) - continue; - - was_active = msp->ms_weight & METASLAB_ACTIVE_MASK; - if (activation_weight == METASLAB_WEIGHT_PRIMARY) - break; - - target_distance = min_distance + - (space_map_allocated(msp->ms_sm) != 0 ? 0 : - min_distance >> 1); - - for (i = 0; i < d; i++) { - if (metaslab_distance(msp, &dva[i]) < - target_distance) - break; - } - if (i == d) - break; + if (activation_weight == METASLAB_WEIGHT_PRIMARY && + mg->mg_primaries[allocator] != NULL) { + msp = mg->mg_primaries[allocator]; + was_active = B_TRUE; + } else if (activation_weight == METASLAB_WEIGHT_SECONDARY && + mg->mg_secondaries[allocator] != NULL && !tertiary) { + msp = mg->mg_secondaries[allocator]; + was_active = B_TRUE; + } else { + msp = find_valid_metaslab(mg, activation_weight, dva, d, + min_distance, asize, allocator, zal, search, + &was_active); } + mutex_exit(&mg->mg_lock); if (msp == NULL) { kmem_free(search, sizeof (*search)); return (-1ULL); } - search->ms_weight = msp->ms_weight; - search->ms_start = msp->ms_start + 1; mutex_enter(&msp->ms_lock); - /* * Ensure that the metaslab we have selected is still * capable of handling our request. It's possible that * another thread may have changed the weight while we * were blocked on the metaslab lock. We check the * active status first to see if we need to reselect * a new metaslab. */ if (was_active && !(msp->ms_weight & METASLAB_ACTIVE_MASK)) { mutex_exit(&msp->ms_lock); continue; } - if ((msp->ms_weight & METASLAB_WEIGHT_SECONDARY) && - activation_weight == METASLAB_WEIGHT_PRIMARY) { - metaslab_passivate(msp, - msp->ms_weight & ~METASLAB_ACTIVE_MASK); + /* + * If the metaslab is freshly activated for an allocator that + * isn't the one we're allocating from, or if it's a primary and + * we're seeking a secondary (or vice versa), we go back and + * select a new metaslab. + */ + if (!was_active && (msp->ms_weight & METASLAB_ACTIVE_MASK) && + (msp->ms_allocator != -1) && + (msp->ms_allocator != allocator || ((activation_weight == + METASLAB_WEIGHT_PRIMARY) != msp->ms_primary))) { mutex_exit(&msp->ms_lock); continue; } - if (metaslab_activate(msp, activation_weight) != 0) { + if (msp->ms_weight & METASLAB_WEIGHT_CLAIM) { + metaslab_passivate(msp, msp->ms_weight & + ~METASLAB_WEIGHT_CLAIM); mutex_exit(&msp->ms_lock); continue; } + + if (metaslab_activate(msp, allocator, activation_weight) != 0) { + mutex_exit(&msp->ms_lock); + continue; + } + msp->ms_selected_txg = txg; /* * Now that we have the lock, recheck to see if we should * continue to use this metaslab for this allocation. The * the metaslab is now loaded so metaslab_should_allocate() can * accurately determine if the allocation attempt should * proceed. */ if (!metaslab_should_allocate(msp, asize)) { /* Passivate this metaslab and select a new one. */ metaslab_trace_add(zal, mg, msp, asize, d, - TRACE_TOO_SMALL); + TRACE_TOO_SMALL, allocator); goto next; } /* * If this metaslab is currently condensing then pick again as * we can't manipulate this metaslab until it's committed * to disk. */ if (msp->ms_condensing) { metaslab_trace_add(zal, mg, msp, asize, d, - TRACE_CONDENSING); + TRACE_CONDENSING, allocator); + metaslab_passivate(msp, msp->ms_weight & + ~METASLAB_ACTIVE_MASK); mutex_exit(&msp->ms_lock); continue; } offset = metaslab_block_alloc(msp, asize, txg); - metaslab_trace_add(zal, mg, msp, asize, d, offset); + metaslab_trace_add(zal, mg, msp, asize, d, offset, allocator); if (offset != -1ULL) { /* Proactively passivate the metaslab, if needed */ metaslab_segment_may_passivate(msp); break; } next: ASSERT(msp->ms_loaded); /* * We were unable to allocate from this metaslab so determine * a new weight for this metaslab. Now that we have loaded * the metaslab we can provide a better hint to the metaslab * selector. * * For space-based metaslabs, we use the maximum block size. * This information is only available when the metaslab * is loaded and is more accurate than the generic free * space weight that was calculated by metaslab_weight(). * This information allows us to quickly compare the maximum * available allocation in the metaslab to the allocation * size being requested. * * For segment-based metaslabs, determine the new weight * based on the highest bucket in the range tree. We * explicitly use the loaded segment weight (i.e. the range * tree histogram) since it contains the space that is * currently available for allocation and is accurate * even within a sync pass. */ if (WEIGHT_IS_SPACEBASED(msp->ms_weight)) { uint64_t weight = metaslab_block_maxsize(msp); WEIGHT_SET_SPACEBASED(weight); metaslab_passivate(msp, weight); } else { metaslab_passivate(msp, metaslab_weight_from_range_tree(msp)); } /* * We have just failed an allocation attempt, check * that metaslab_should_allocate() agrees. Otherwise, * we may end up in an infinite loop retrying the same * metaslab. */ ASSERT(!metaslab_should_allocate(msp, asize)); mutex_exit(&msp->ms_lock); } mutex_exit(&msp->ms_lock); kmem_free(search, sizeof (*search)); return (offset); } static uint64_t metaslab_group_alloc(metaslab_group_t *mg, zio_alloc_list_t *zal, - uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d) + uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d, + int allocator) { uint64_t offset; ASSERT(mg->mg_initialized); offset = metaslab_group_alloc_normal(mg, zal, asize, txg, - min_distance, dva, d); + min_distance, dva, d, allocator); mutex_enter(&mg->mg_lock); if (offset == -1ULL) { mg->mg_failed_allocations++; metaslab_trace_add(zal, mg, NULL, asize, d, - TRACE_GROUP_FAILURE); + TRACE_GROUP_FAILURE, allocator); if (asize == SPA_GANGBLOCKSIZE) { /* * This metaslab group was unable to allocate * the minimum gang block size so it must be out of * space. We must notify the allocation throttle * to start skipping allocation attempts to this * metaslab group until more space becomes available. * Note: this failure cannot be caused by the * allocation throttle since the allocation throttle * is only responsible for skipping devices and * not failing block allocations. */ mg->mg_no_free_space = B_TRUE; } } mg->mg_allocations++; mutex_exit(&mg->mg_lock); return (offset); } /* * If we have to write a ditto block (i.e. more than one DVA for a given BP) * on the same vdev as an existing DVA of this BP, then try to allocate it * at least (vdev_asize / (2 ^ ditto_same_vdev_distance_shift)) away from the * existing DVAs. */ int ditto_same_vdev_distance_shift = 3; /* * Allocate a block for the specified i/o. */ int metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize, dva_t *dva, int d, dva_t *hintdva, uint64_t txg, int flags, - zio_alloc_list_t *zal) + zio_alloc_list_t *zal, int allocator) { metaslab_group_t *mg, *rotor; vdev_t *vd; boolean_t try_hard = B_FALSE; ASSERT(!DVA_IS_VALID(&dva[d])); /* * For testing, make some blocks above a certain size be gang blocks. */ if (psize >= metaslab_force_ganging && (ddi_get_lbolt() & 3) == 0) { - metaslab_trace_add(zal, NULL, NULL, psize, d, TRACE_FORCE_GANG); + metaslab_trace_add(zal, NULL, NULL, psize, d, TRACE_FORCE_GANG, + allocator); return (SET_ERROR(ENOSPC)); } /* * Start at the rotor and loop through all mgs until we find something. * Note that there's no locking on mc_rotor or mc_aliquot because * nothing actually breaks if we miss a few updates -- we just won't * allocate quite as evenly. It all balances out over time. * * If we are doing ditto or log blocks, try to spread them across * consecutive vdevs. If we're forced to reuse a vdev before we've * allocated all of our ditto blocks, then try and spread them out on * that vdev as much as possible. If it turns out to not be possible, * gradually lower our standards until anything becomes acceptable. * Also, allocating on consecutive vdevs (as opposed to random vdevs) * gives us hope of containing our fault domains to something we're * able to reason about. Otherwise, any two top-level vdev failures * will guarantee the loss of data. With consecutive allocation, * only two adjacent top-level vdev failures will result in data loss. * * If we are doing gang blocks (hintdva is non-NULL), try to keep * ourselves on the same vdev as our gang block header. That * way, we can hope for locality in vdev_cache, plus it makes our * fault domains something tractable. */ if (hintdva) { vd = vdev_lookup_top(spa, DVA_GET_VDEV(&hintdva[d])); /* * It's possible the vdev we're using as the hint no * longer exists or its mg has been closed (e.g. by * device removal). Consult the rotor when * all else fails. */ if (vd != NULL && vd->vdev_mg != NULL) { mg = vd->vdev_mg; if (flags & METASLAB_HINTBP_AVOID && mg->mg_next != NULL) mg = mg->mg_next; } else { mg = mc->mc_rotor; } } else if (d != 0) { vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d - 1])); mg = vd->vdev_mg->mg_next; } else { mg = mc->mc_rotor; } /* * If the hint put us into the wrong metaslab class, or into a * metaslab group that has been passivated, just follow the rotor. */ if (mg->mg_class != mc || mg->mg_activation_count <= 0) mg = mc->mc_rotor; rotor = mg; top: do { boolean_t allocatable; ASSERT(mg->mg_activation_count == 1); vd = mg->mg_vd; /* * Don't allocate from faulted devices. */ if (try_hard) { spa_config_enter(spa, SCL_ZIO, FTAG, RW_READER); allocatable = vdev_allocatable(vd); spa_config_exit(spa, SCL_ZIO, FTAG); } else { allocatable = vdev_allocatable(vd); } /* * Determine if the selected metaslab group is eligible * for allocations. If we're ganging then don't allow * this metaslab group to skip allocations since that would * inadvertently return ENOSPC and suspend the pool * even though space is still available. */ if (allocatable && !GANG_ALLOCATION(flags) && !try_hard) { allocatable = metaslab_group_allocatable(mg, rotor, - psize); + psize, allocator); } if (!allocatable) { metaslab_trace_add(zal, mg, NULL, psize, d, - TRACE_NOT_ALLOCATABLE); + TRACE_NOT_ALLOCATABLE, allocator); goto next; } ASSERT(mg->mg_initialized); /* * Avoid writing single-copy data to a failing, * non-redundant vdev, unless we've already tried all * other vdevs. */ if ((vd->vdev_stat.vs_write_errors > 0 || vd->vdev_state < VDEV_STATE_HEALTHY) && d == 0 && !try_hard && vd->vdev_children == 0) { metaslab_trace_add(zal, mg, NULL, psize, d, - TRACE_VDEV_ERROR); + TRACE_VDEV_ERROR, allocator); goto next; } ASSERT(mg->mg_class == mc); /* * If we don't need to try hard, then require that the * block be 1/8th of the device away from any other DVAs * in this BP. If we are trying hard, allow any offset * to be used (distance=0). */ uint64_t distance = 0; if (!try_hard) { distance = vd->vdev_asize >> ditto_same_vdev_distance_shift; if (distance <= (1ULL << vd->vdev_ms_shift)) distance = 0; } uint64_t asize = vdev_psize_to_asize(vd, psize); ASSERT(P2PHASE(asize, 1ULL << vd->vdev_ashift) == 0); uint64_t offset = metaslab_group_alloc(mg, zal, asize, txg, - distance, dva, d); + distance, dva, d, allocator); if (offset != -1ULL) { /* * If we've just selected this metaslab group, * figure out whether the corresponding vdev is * over- or under-used relative to the pool, * and set an allocation bias to even it out. */ if (mc->mc_aliquot == 0 && metaslab_bias_enabled) { vdev_stat_t *vs = &vd->vdev_stat; int64_t vu, cu; vu = (vs->vs_alloc * 100) / (vs->vs_space + 1); cu = (mc->mc_alloc * 100) / (mc->mc_space + 1); /* * Calculate how much more or less we should * try to allocate from this device during * this iteration around the rotor. * For example, if a device is 80% full * and the pool is 20% full then we should * reduce allocations by 60% on this device. * * mg_bias = (20 - 80) * 512K / 100 = -307K * * This reduces allocations by 307K for this * iteration. */ mg->mg_bias = ((cu - vu) * (int64_t)mg->mg_aliquot) / 100; } else if (!metaslab_bias_enabled) { mg->mg_bias = 0; } if (atomic_add_64_nv(&mc->mc_aliquot, asize) >= mg->mg_aliquot + mg->mg_bias) { mc->mc_rotor = mg->mg_next; mc->mc_aliquot = 0; } DVA_SET_VDEV(&dva[d], vd->vdev_id); DVA_SET_OFFSET(&dva[d], offset); DVA_SET_GANG(&dva[d], !!(flags & METASLAB_GANG_HEADER)); DVA_SET_ASIZE(&dva[d], asize); return (0); } next: mc->mc_rotor = mg->mg_next; mc->mc_aliquot = 0; } while ((mg = mg->mg_next) != rotor); /* * If we haven't tried hard, do so now. */ if (!try_hard) { try_hard = B_TRUE; goto top; } bzero(&dva[d], sizeof (dva_t)); - metaslab_trace_add(zal, rotor, NULL, psize, d, TRACE_ENOSPC); + metaslab_trace_add(zal, rotor, NULL, psize, d, TRACE_ENOSPC, allocator); return (SET_ERROR(ENOSPC)); } void metaslab_free_concrete(vdev_t *vd, uint64_t offset, uint64_t asize, boolean_t checkpoint) { metaslab_t *msp; spa_t *spa = vd->vdev_spa; ASSERT(vdev_is_concrete(vd)); ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); ASSERT3U(offset >> vd->vdev_ms_shift, <, vd->vdev_ms_count); msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; VERIFY(!msp->ms_condensing); VERIFY3U(offset, >=, msp->ms_start); VERIFY3U(offset + asize, <=, msp->ms_start + msp->ms_size); VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(asize, 1ULL << vd->vdev_ashift)); metaslab_check_free_impl(vd, offset, asize); mutex_enter(&msp->ms_lock); if (range_tree_is_empty(msp->ms_freeing) && range_tree_is_empty(msp->ms_checkpointing)) { vdev_dirty(vd, VDD_METASLAB, msp, spa_syncing_txg(spa)); } if (checkpoint) { ASSERT(spa_has_checkpoint(spa)); range_tree_add(msp->ms_checkpointing, offset, asize); } else { range_tree_add(msp->ms_freeing, offset, asize); } mutex_exit(&msp->ms_lock); } /* ARGSUSED */ void metaslab_free_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset, uint64_t size, void *arg) { boolean_t *checkpoint = arg; ASSERT3P(checkpoint, !=, NULL); if (vd->vdev_ops->vdev_op_remap != NULL) vdev_indirect_mark_obsolete(vd, offset, size); else metaslab_free_impl(vd, offset, size, *checkpoint); } static void metaslab_free_impl(vdev_t *vd, uint64_t offset, uint64_t size, boolean_t checkpoint) { spa_t *spa = vd->vdev_spa; ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); if (spa_syncing_txg(spa) > spa_freeze_txg(spa)) return; if (spa->spa_vdev_removal != NULL && spa->spa_vdev_removal->svr_vdev == vd && vdev_is_concrete(vd)) { /* * Note: we check if the vdev is concrete because when * we complete the removal, we first change the vdev to be * an indirect vdev (in open context), and then (in syncing * context) clear spa_vdev_removal. */ free_from_removing_vdev(vd, offset, size); } else if (vd->vdev_ops->vdev_op_remap != NULL) { vdev_indirect_mark_obsolete(vd, offset, size); vd->vdev_ops->vdev_op_remap(vd, offset, size, metaslab_free_impl_cb, &checkpoint); } else { metaslab_free_concrete(vd, offset, size, checkpoint); } } typedef struct remap_blkptr_cb_arg { blkptr_t *rbca_bp; spa_remap_cb_t rbca_cb; vdev_t *rbca_remap_vd; uint64_t rbca_remap_offset; void *rbca_cb_arg; } remap_blkptr_cb_arg_t; void remap_blkptr_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset, uint64_t size, void *arg) { remap_blkptr_cb_arg_t *rbca = arg; blkptr_t *bp = rbca->rbca_bp; /* We can not remap split blocks. */ if (size != DVA_GET_ASIZE(&bp->blk_dva[0])) return; ASSERT0(inner_offset); if (rbca->rbca_cb != NULL) { /* * At this point we know that we are not handling split * blocks and we invoke the callback on the previous * vdev which must be indirect. */ ASSERT3P(rbca->rbca_remap_vd->vdev_ops, ==, &vdev_indirect_ops); rbca->rbca_cb(rbca->rbca_remap_vd->vdev_id, rbca->rbca_remap_offset, size, rbca->rbca_cb_arg); /* set up remap_blkptr_cb_arg for the next call */ rbca->rbca_remap_vd = vd; rbca->rbca_remap_offset = offset; } /* * The phys birth time is that of dva[0]. This ensures that we know * when each dva was written, so that resilver can determine which * blocks need to be scrubbed (i.e. those written during the time * the vdev was offline). It also ensures that the key used in * the ARC hash table is unique (i.e. dva[0] + phys_birth). If * we didn't change the phys_birth, a lookup in the ARC for a * remapped BP could find the data that was previously stored at * this vdev + offset. */ vdev_t *oldvd = vdev_lookup_top(vd->vdev_spa, DVA_GET_VDEV(&bp->blk_dva[0])); vdev_indirect_births_t *vib = oldvd->vdev_indirect_births; bp->blk_phys_birth = vdev_indirect_births_physbirth(vib, DVA_GET_OFFSET(&bp->blk_dva[0]), DVA_GET_ASIZE(&bp->blk_dva[0])); DVA_SET_VDEV(&bp->blk_dva[0], vd->vdev_id); DVA_SET_OFFSET(&bp->blk_dva[0], offset); } /* * If the block pointer contains any indirect DVAs, modify them to refer to * concrete DVAs. Note that this will sometimes not be possible, leaving * the indirect DVA in place. This happens if the indirect DVA spans multiple * segments in the mapping (i.e. it is a "split block"). * * If the BP was remapped, calls the callback on the original dva (note the * callback can be called multiple times if the original indirect DVA refers * to another indirect DVA, etc). * * Returns TRUE if the BP was remapped. */ boolean_t spa_remap_blkptr(spa_t *spa, blkptr_t *bp, spa_remap_cb_t callback, void *arg) { remap_blkptr_cb_arg_t rbca; if (!zfs_remap_blkptr_enable) return (B_FALSE); if (!spa_feature_is_enabled(spa, SPA_FEATURE_OBSOLETE_COUNTS)) return (B_FALSE); /* * Dedup BP's can not be remapped, because ddt_phys_select() depends * on DVA[0] being the same in the BP as in the DDT (dedup table). */ if (BP_GET_DEDUP(bp)) return (B_FALSE); /* * Gang blocks can not be remapped, because * zio_checksum_gang_verifier() depends on the DVA[0] that's in * the BP used to read the gang block header (GBH) being the same * as the DVA[0] that we allocated for the GBH. */ if (BP_IS_GANG(bp)) return (B_FALSE); /* * Embedded BP's have no DVA to remap. */ if (BP_GET_NDVAS(bp) < 1) return (B_FALSE); /* * Note: we only remap dva[0]. If we remapped other dvas, we * would no longer know what their phys birth txg is. */ dva_t *dva = &bp->blk_dva[0]; uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); if (vd->vdev_ops->vdev_op_remap == NULL) return (B_FALSE); rbca.rbca_bp = bp; rbca.rbca_cb = callback; rbca.rbca_remap_vd = vd; rbca.rbca_remap_offset = offset; rbca.rbca_cb_arg = arg; /* * remap_blkptr_cb() will be called in order for each level of * indirection, until a concrete vdev is reached or a split block is * encountered. old_vd and old_offset are updated within the callback * as we go from the one indirect vdev to the next one (either concrete * or indirect again) in that order. */ vd->vdev_ops->vdev_op_remap(vd, offset, size, remap_blkptr_cb, &rbca); /* Check if the DVA wasn't remapped because it is a split block */ if (DVA_GET_VDEV(&rbca.rbca_bp->blk_dva[0]) == vd->vdev_id) return (B_FALSE); return (B_TRUE); } /* * Undo the allocation of a DVA which happened in the given transaction group. */ void metaslab_unalloc_dva(spa_t *spa, const dva_t *dva, uint64_t txg) { metaslab_t *msp; vdev_t *vd; uint64_t vdev = DVA_GET_VDEV(dva); uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); ASSERT(DVA_IS_VALID(dva)); ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); if (txg > spa_freeze_txg(spa)) return; if ((vd = vdev_lookup_top(spa, vdev)) == NULL || (offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count) { cmn_err(CE_WARN, "metaslab_free_dva(): bad DVA %llu:%llu", (u_longlong_t)vdev, (u_longlong_t)offset); ASSERT(0); return; } ASSERT(!vd->vdev_removing); ASSERT(vdev_is_concrete(vd)); ASSERT0(vd->vdev_indirect_config.vic_mapping_object); ASSERT3P(vd->vdev_indirect_mapping, ==, NULL); if (DVA_GET_GANG(dva)) size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; mutex_enter(&msp->ms_lock); range_tree_remove(msp->ms_allocating[txg & TXG_MASK], offset, size); VERIFY(!msp->ms_condensing); VERIFY3U(offset, >=, msp->ms_start); VERIFY3U(offset + size, <=, msp->ms_start + msp->ms_size); VERIFY3U(range_tree_space(msp->ms_allocatable) + size, <=, msp->ms_size); VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift)); range_tree_add(msp->ms_allocatable, offset, size); mutex_exit(&msp->ms_lock); } /* * Free the block represented by the given DVA. */ void metaslab_free_dva(spa_t *spa, const dva_t *dva, boolean_t checkpoint) { uint64_t vdev = DVA_GET_VDEV(dva); uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); vdev_t *vd = vdev_lookup_top(spa, vdev); ASSERT(DVA_IS_VALID(dva)); ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); if (DVA_GET_GANG(dva)) { size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); } metaslab_free_impl(vd, offset, size, checkpoint); } /* * Reserve some allocation slots. The reservation system must be called * before we call into the allocator. If there aren't any available slots * then the I/O will be throttled until an I/O completes and its slots are * freed up. The function returns true if it was successful in placing * the reservation. */ boolean_t -metaslab_class_throttle_reserve(metaslab_class_t *mc, int slots, zio_t *zio, - int flags) +metaslab_class_throttle_reserve(metaslab_class_t *mc, int slots, int allocator, + zio_t *zio, int flags) { uint64_t available_slots = 0; boolean_t slot_reserved = B_FALSE; + uint64_t max = mc->mc_alloc_max_slots[allocator]; ASSERT(mc->mc_alloc_throttle_enabled); mutex_enter(&mc->mc_lock); - uint64_t reserved_slots = refcount_count(&mc->mc_alloc_slots); - if (reserved_slots < mc->mc_alloc_max_slots) - available_slots = mc->mc_alloc_max_slots - reserved_slots; + uint64_t reserved_slots = + refcount_count(&mc->mc_alloc_slots[allocator]); + if (reserved_slots < max) + available_slots = max - reserved_slots; if (slots <= available_slots || GANG_ALLOCATION(flags)) { /* * We reserve the slots individually so that we can unreserve * them individually when an I/O completes. */ for (int d = 0; d < slots; d++) { - reserved_slots = refcount_add(&mc->mc_alloc_slots, zio); + reserved_slots = + refcount_add(&mc->mc_alloc_slots[allocator], + zio); } zio->io_flags |= ZIO_FLAG_IO_ALLOCATING; slot_reserved = B_TRUE; } mutex_exit(&mc->mc_lock); return (slot_reserved); } void -metaslab_class_throttle_unreserve(metaslab_class_t *mc, int slots, zio_t *zio) +metaslab_class_throttle_unreserve(metaslab_class_t *mc, int slots, + int allocator, zio_t *zio) { ASSERT(mc->mc_alloc_throttle_enabled); mutex_enter(&mc->mc_lock); for (int d = 0; d < slots; d++) { - (void) refcount_remove(&mc->mc_alloc_slots, zio); + (void) refcount_remove(&mc->mc_alloc_slots[allocator], + zio); } mutex_exit(&mc->mc_lock); } static int metaslab_claim_concrete(vdev_t *vd, uint64_t offset, uint64_t size, uint64_t txg) { metaslab_t *msp; spa_t *spa = vd->vdev_spa; int error = 0; if (offset >> vd->vdev_ms_shift >= vd->vdev_ms_count) return (ENXIO); ASSERT3P(vd->vdev_ms, !=, NULL); msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; mutex_enter(&msp->ms_lock); if ((txg != 0 && spa_writeable(spa)) || !msp->ms_loaded) - error = metaslab_activate(msp, METASLAB_WEIGHT_SECONDARY); + error = metaslab_activate(msp, 0, METASLAB_WEIGHT_CLAIM); + /* + * No need to fail in that case; someone else has activated the + * metaslab, but that doesn't preclude us from using it. + */ + if (error == EBUSY) + error = 0; if (error == 0 && !range_tree_contains(msp->ms_allocatable, offset, size)) error = SET_ERROR(ENOENT); if (error || txg == 0) { /* txg == 0 indicates dry run */ mutex_exit(&msp->ms_lock); return (error); } VERIFY(!msp->ms_condensing); VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift)); VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift)); VERIFY3U(range_tree_space(msp->ms_allocatable) - size, <=, msp->ms_size); range_tree_remove(msp->ms_allocatable, offset, size); if (spa_writeable(spa)) { /* don't dirty if we're zdb(1M) */ if (range_tree_is_empty(msp->ms_allocating[txg & TXG_MASK])) vdev_dirty(vd, VDD_METASLAB, msp, txg); range_tree_add(msp->ms_allocating[txg & TXG_MASK], offset, size); } mutex_exit(&msp->ms_lock); return (0); } typedef struct metaslab_claim_cb_arg_t { uint64_t mcca_txg; int mcca_error; } metaslab_claim_cb_arg_t; /* ARGSUSED */ static void metaslab_claim_impl_cb(uint64_t inner_offset, vdev_t *vd, uint64_t offset, uint64_t size, void *arg) { metaslab_claim_cb_arg_t *mcca_arg = arg; if (mcca_arg->mcca_error == 0) { mcca_arg->mcca_error = metaslab_claim_concrete(vd, offset, size, mcca_arg->mcca_txg); } } int metaslab_claim_impl(vdev_t *vd, uint64_t offset, uint64_t size, uint64_t txg) { if (vd->vdev_ops->vdev_op_remap != NULL) { metaslab_claim_cb_arg_t arg; /* * Only zdb(1M) can claim on indirect vdevs. This is used * to detect leaks of mapped space (that are not accounted * for in the obsolete counts, spacemap, or bpobj). */ ASSERT(!spa_writeable(vd->vdev_spa)); arg.mcca_error = 0; arg.mcca_txg = txg; vd->vdev_ops->vdev_op_remap(vd, offset, size, metaslab_claim_impl_cb, &arg); if (arg.mcca_error == 0) { arg.mcca_error = metaslab_claim_concrete(vd, offset, size, txg); } return (arg.mcca_error); } else { return (metaslab_claim_concrete(vd, offset, size, txg)); } } /* * Intent log support: upon opening the pool after a crash, notify the SPA * of blocks that the intent log has allocated for immediate write, but * which are still considered free by the SPA because the last transaction * group didn't commit yet. */ static int metaslab_claim_dva(spa_t *spa, const dva_t *dva, uint64_t txg) { uint64_t vdev = DVA_GET_VDEV(dva); uint64_t offset = DVA_GET_OFFSET(dva); uint64_t size = DVA_GET_ASIZE(dva); vdev_t *vd; if ((vd = vdev_lookup_top(spa, vdev)) == NULL) { return (SET_ERROR(ENXIO)); } ASSERT(DVA_IS_VALID(dva)); if (DVA_GET_GANG(dva)) size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); return (metaslab_claim_impl(vd, offset, size, txg)); } int metaslab_alloc(spa_t *spa, metaslab_class_t *mc, uint64_t psize, blkptr_t *bp, int ndvas, uint64_t txg, blkptr_t *hintbp, int flags, - zio_alloc_list_t *zal, zio_t *zio) + zio_alloc_list_t *zal, zio_t *zio, int allocator) { dva_t *dva = bp->blk_dva; dva_t *hintdva = hintbp->blk_dva; int error = 0; ASSERT(bp->blk_birth == 0); ASSERT(BP_PHYSICAL_BIRTH(bp) == 0); spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER); if (mc->mc_rotor == NULL) { /* no vdevs in this class */ spa_config_exit(spa, SCL_ALLOC, FTAG); return (SET_ERROR(ENOSPC)); } ASSERT(ndvas > 0 && ndvas <= spa_max_replication(spa)); ASSERT(BP_GET_NDVAS(bp) == 0); ASSERT(hintbp == NULL || ndvas <= BP_GET_NDVAS(hintbp)); ASSERT3P(zal, !=, NULL); for (int d = 0; d < ndvas; d++) { error = metaslab_alloc_dva(spa, mc, psize, dva, d, hintdva, - txg, flags, zal); + txg, flags, zal, allocator); if (error != 0) { for (d--; d >= 0; d--) { metaslab_unalloc_dva(spa, &dva[d], txg); metaslab_group_alloc_decrement(spa, - DVA_GET_VDEV(&dva[d]), zio, flags); + DVA_GET_VDEV(&dva[d]), zio, flags, + allocator, B_FALSE); bzero(&dva[d], sizeof (dva_t)); } spa_config_exit(spa, SCL_ALLOC, FTAG); return (error); } else { /* * Update the metaslab group's queue depth * based on the newly allocated dva. */ metaslab_group_alloc_increment(spa, - DVA_GET_VDEV(&dva[d]), zio, flags); + DVA_GET_VDEV(&dva[d]), zio, flags, allocator); } } ASSERT(error == 0); ASSERT(BP_GET_NDVAS(bp) == ndvas); spa_config_exit(spa, SCL_ALLOC, FTAG); BP_SET_BIRTH(bp, txg, txg); return (0); } void metaslab_free(spa_t *spa, const blkptr_t *bp, uint64_t txg, boolean_t now) { const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); ASSERT(!BP_IS_HOLE(bp)); ASSERT(!now || bp->blk_birth >= spa_syncing_txg(spa)); /* * If we have a checkpoint for the pool we need to make sure that * the blocks that we free that are part of the checkpoint won't be * reused until the checkpoint is discarded or we revert to it. * * The checkpoint flag is passed down the metaslab_free code path * and is set whenever we want to add a block to the checkpoint's * accounting. That is, we "checkpoint" blocks that existed at the * time the checkpoint was created and are therefore referenced by * the checkpointed uberblock. * * Note that, we don't checkpoint any blocks if the current * syncing txg <= spa_checkpoint_txg. We want these frees to sync * normally as they will be referenced by the checkpointed uberblock. */ boolean_t checkpoint = B_FALSE; if (bp->blk_birth <= spa->spa_checkpoint_txg && spa_syncing_txg(spa) > spa->spa_checkpoint_txg) { /* * At this point, if the block is part of the checkpoint * there is no way it was created in the current txg. */ ASSERT(!now); ASSERT3U(spa_syncing_txg(spa), ==, txg); checkpoint = B_TRUE; } spa_config_enter(spa, SCL_FREE, FTAG, RW_READER); for (int d = 0; d < ndvas; d++) { if (now) { metaslab_unalloc_dva(spa, &dva[d], txg); } else { ASSERT3U(txg, ==, spa_syncing_txg(spa)); metaslab_free_dva(spa, &dva[d], checkpoint); } } spa_config_exit(spa, SCL_FREE, FTAG); } int metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg) { const dva_t *dva = bp->blk_dva; int ndvas = BP_GET_NDVAS(bp); int error = 0; ASSERT(!BP_IS_HOLE(bp)); if (txg != 0) { /* * First do a dry run to make sure all DVAs are claimable, * so we don't have to unwind from partial failures below. */ if ((error = metaslab_claim(spa, bp, 0)) != 0) return (error); } spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER); for (int d = 0; d < ndvas; d++) if ((error = metaslab_claim_dva(spa, &dva[d], txg)) != 0) break; spa_config_exit(spa, SCL_ALLOC, FTAG); ASSERT(error == 0 || txg == 0); return (error); } /* ARGSUSED */ static void metaslab_check_free_impl_cb(uint64_t inner, vdev_t *vd, uint64_t offset, uint64_t size, void *arg) { if (vd->vdev_ops == &vdev_indirect_ops) return; metaslab_check_free_impl(vd, offset, size); } static void metaslab_check_free_impl(vdev_t *vd, uint64_t offset, uint64_t size) { metaslab_t *msp; spa_t *spa = vd->vdev_spa; if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0) return; if (vd->vdev_ops->vdev_op_remap != NULL) { vd->vdev_ops->vdev_op_remap(vd, offset, size, metaslab_check_free_impl_cb, NULL); return; } ASSERT(vdev_is_concrete(vd)); ASSERT3U(offset >> vd->vdev_ms_shift, <, vd->vdev_ms_count); ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0); msp = vd->vdev_ms[offset >> vd->vdev_ms_shift]; mutex_enter(&msp->ms_lock); if (msp->ms_loaded) range_tree_verify(msp->ms_allocatable, offset, size); range_tree_verify(msp->ms_freeing, offset, size); range_tree_verify(msp->ms_checkpointing, offset, size); range_tree_verify(msp->ms_freed, offset, size); for (int j = 0; j < TXG_DEFER_SIZE; j++) range_tree_verify(msp->ms_defer[j], offset, size); mutex_exit(&msp->ms_lock); } void metaslab_check_free(spa_t *spa, const blkptr_t *bp) { if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0) return; spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); for (int i = 0; i < BP_GET_NDVAS(bp); i++) { uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[i]); vdev_t *vd = vdev_lookup_top(spa, vdev); uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]); uint64_t size = DVA_GET_ASIZE(&bp->blk_dva[i]); if (DVA_GET_GANG(&bp->blk_dva[i])) size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); ASSERT3P(vd, !=, NULL); metaslab_check_free_impl(vd, offset, size); } spa_config_exit(spa, SCL_VDEV, FTAG); } Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c (revision 336949) @@ -1,8335 +1,8344 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2018 by Delphix. All rights reserved. * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2013 Martin Matuska . All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright 2013 Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Integros [integros.com] * Copyright 2016 Toomas Soome * Copyright 2017 Joyent, Inc. * Copyright (c) 2017 Datto Inc. * Copyright 2018 OmniOS Community Edition (OmniOSce) Association. */ /* * SPA: Storage Pool Allocator * * This file contains all the routines used when modifying on-disk SPA state. * This includes opening, importing, destroying, exporting a pool, and syncing a * pool. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #include #include #endif /* _KERNEL */ #include "zfs_prop.h" #include "zfs_comutil.h" /* Check hostid on import? */ static int check_hostid = 1; /* * The interval, in seconds, at which failed configuration cache file writes * should be retried. */ int zfs_ccw_retry_interval = 300; SYSCTL_DECL(_vfs_zfs); SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, "Check hostid on import?"); TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, &zfs_ccw_retry_interval, 0, "Configuration cache file write, retry after failure, interval (seconds)"); typedef enum zti_modes { ZTI_MODE_FIXED, /* value is # of threads (min 1) */ ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ ZTI_MODE_NULL, /* don't create a taskq */ ZTI_NMODES } zti_modes_t; #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } #define ZTI_N(n) ZTI_P(n, 1) #define ZTI_ONE ZTI_N(1) typedef struct zio_taskq_info { zti_modes_t zti_mode; uint_t zti_value; uint_t zti_count; } zio_taskq_info_t; static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { "issue", "issue_high", "intr", "intr_high" }; /* * This table defines the taskq settings for each ZFS I/O type. When * initializing a pool, we use this table to create an appropriately sized * taskq. Some operations are low volume and therefore have a small, static * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE * macros. Other operations process a large amount of data; the ZTI_BATCH * macro causes us to create a taskq oriented for throughput. Some operations * are so high frequency and short-lived that the taskq itself can become a a * point of lock contention. The ZTI_P(#, #) macro indicates that we need an * additional degree of parallelism specified by the number of threads per- * taskq and the number of taskqs; when dispatching an event in this case, the * particular taskq is chosen at random. * * The different taskq priorities are to handle the different contexts (issue * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that * need to be handled with minimum delay. */ const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ }; static void spa_sync_version(void *arg, dmu_tx_t *tx); static void spa_sync_props(void *arg, dmu_tx_t *tx); static boolean_t spa_has_active_shared_spare(spa_t *spa); static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport); static void spa_vdev_resilver_done(spa_t *spa); uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ #ifdef PSRSET_BIND id_t zio_taskq_psrset_bind = PS_NONE; #endif #ifdef SYSDC boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ uint_t zio_taskq_basedc = 80; /* base duty cycle */ #endif boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ extern int zfs_sync_pass_deferred_free; /* * Report any spa_load_verify errors found, but do not fail spa_load. * This is used by zdb to analyze non-idle pools. */ boolean_t spa_load_verify_dryrun = B_FALSE; /* * This (illegal) pool name is used when temporarily importing a spa_t in order * to get the vdev stats associated with the imported devices. */ #define TRYIMPORT_NAME "$import" /* * For debugging purposes: print out vdev tree during pool import. */ int spa_load_print_vdev_tree = B_FALSE; /* * A non-zero value for zfs_max_missing_tvds means that we allow importing * pools with missing top-level vdevs. This is strictly intended for advanced * pool recovery cases since missing data is almost inevitable. Pools with * missing devices can only be imported read-only for safety reasons, and their * fail-mode will be automatically set to "continue". * * With 1 missing vdev we should be able to import the pool and mount all * datasets. User data that was not modified after the missing device has been * added should be recoverable. This means that snapshots created prior to the * addition of that device should be completely intact. * * With 2 missing vdevs, some datasets may fail to mount since there are * dataset statistics that are stored as regular metadata. Some data might be * recoverable if those vdevs were added recently. * * With 3 or more missing vdevs, the pool is severely damaged and MOS entries * may be missing entirely. Chances of data recovery are very low. Note that * there are also risks of performing an inadvertent rewind as we might be * missing all the vdevs with the latest uberblocks. */ uint64_t zfs_max_missing_tvds = 0; /* * The parameters below are similar to zfs_max_missing_tvds but are only * intended for a preliminary open of the pool with an untrusted config which * might be incomplete or out-dated. * * We are more tolerant for pools opened from a cachefile since we could have * an out-dated cachefile where a device removal was not registered. * We could have set the limit arbitrarily high but in the case where devices * are really missing we would want to return the proper error codes; we chose * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available * and we get a chance to retrieve the trusted config. */ uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1; /* * In the case where config was assembled by scanning device paths (/dev/dsks * by default) we are less tolerant since all the existing devices should have * been detected and we want spa_load to return the right error codes. */ uint64_t zfs_max_missing_tvds_scan = 0; SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_print_vdev_tree, CTLFLAG_RWTUN, &spa_load_print_vdev_tree, 0, "print out vdev tree during pool import"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds, CTLFLAG_RWTUN, &zfs_max_missing_tvds, 0, "allow importing pools with missing top-level vdevs"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_cachefile, CTLFLAG_RWTUN, &zfs_max_missing_tvds_cachefile, 0, "allow importing pools with missing top-level vdevs in cache file"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_scan, CTLFLAG_RWTUN, &zfs_max_missing_tvds_scan, 0, "allow importing pools with missing top-level vdevs during scan"); /* * Debugging aid that pauses spa_sync() towards the end. */ boolean_t zfs_pause_spa_sync = B_FALSE; /* * ========================================================================== * SPA properties routines * ========================================================================== */ /* * Add a (source=src, propname=propval) list to an nvlist. */ static void spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, uint64_t intval, zprop_source_t src) { const char *propname = zpool_prop_to_name(prop); nvlist_t *propval; VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); if (strval != NULL) VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); else VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); nvlist_free(propval); } /* * Get property values from the spa configuration. */ static void spa_prop_get_config(spa_t *spa, nvlist_t **nvp) { vdev_t *rvd = spa->spa_root_vdev; dsl_pool_t *pool = spa->spa_dsl_pool; uint64_t size, alloc, cap, version; zprop_source_t src = ZPROP_SRC_NONE; spa_config_dirent_t *dp; metaslab_class_t *mc = spa_normal_class(spa); ASSERT(MUTEX_HELD(&spa->spa_props_lock)); if (rvd != NULL) { alloc = metaslab_class_get_alloc(spa_normal_class(spa)); size = metaslab_class_get_space(spa_normal_class(spa)); spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, size - alloc, src); spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL, spa->spa_checkpoint_info.sci_dspace, src); spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, metaslab_class_fragmentation(mc), src); spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, metaslab_class_expandable_space(mc), src); spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, (spa_mode(spa) == FREAD), src); cap = (size == 0) ? 0 : (alloc * 100 / size); spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, ddt_get_pool_dedup_ratio(spa), src); spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, rvd->vdev_state, src); version = spa_version(spa); if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) src = ZPROP_SRC_DEFAULT; else src = ZPROP_SRC_LOCAL; spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); } if (pool != NULL) { /* * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, * when opening pools before this version freedir will be NULL. */ if (pool->dp_free_dir != NULL) { spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, src); } else { spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 0, src); } if (pool->dp_leak_dir != NULL) { spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, src); } else { spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 0, src); } } spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); if (spa->spa_comment != NULL) { spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 0, ZPROP_SRC_LOCAL); } if (spa->spa_root != NULL) spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 0, ZPROP_SRC_LOCAL); if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); } else { spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); } if ((dp = list_head(&spa->spa_config_list)) != NULL) { if (dp->scd_path == NULL) { spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, "none", 0, ZPROP_SRC_LOCAL); } else if (strcmp(dp->scd_path, spa_config_path) != 0) { spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, dp->scd_path, 0, ZPROP_SRC_LOCAL); } } } /* * Get zpool property values. */ int spa_prop_get(spa_t *spa, nvlist_t **nvp) { objset_t *mos = spa->spa_meta_objset; zap_cursor_t zc; zap_attribute_t za; int err; VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); mutex_enter(&spa->spa_props_lock); /* * Get properties from the spa config. */ spa_prop_get_config(spa, nvp); /* If no pool property object, no more prop to get. */ if (mos == NULL || spa->spa_pool_props_object == 0) { mutex_exit(&spa->spa_props_lock); return (0); } /* * Get properties from the MOS pool property object. */ for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); (err = zap_cursor_retrieve(&zc, &za)) == 0; zap_cursor_advance(&zc)) { uint64_t intval = 0; char *strval = NULL; zprop_source_t src = ZPROP_SRC_DEFAULT; zpool_prop_t prop; if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL) continue; switch (za.za_integer_length) { case 8: /* integer property */ if (za.za_first_integer != zpool_prop_default_numeric(prop)) src = ZPROP_SRC_LOCAL; if (prop == ZPOOL_PROP_BOOTFS) { dsl_pool_t *dp; dsl_dataset_t *ds = NULL; dp = spa_get_dsl(spa); dsl_pool_config_enter(dp, FTAG); if (err = dsl_dataset_hold_obj(dp, za.za_first_integer, FTAG, &ds)) { dsl_pool_config_exit(dp, FTAG); break; } strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); dsl_dataset_name(ds, strval); dsl_dataset_rele(ds, FTAG); dsl_pool_config_exit(dp, FTAG); } else { strval = NULL; intval = za.za_first_integer; } spa_prop_add_list(*nvp, prop, strval, intval, src); if (strval != NULL) kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); break; case 1: /* string property */ strval = kmem_alloc(za.za_num_integers, KM_SLEEP); err = zap_lookup(mos, spa->spa_pool_props_object, za.za_name, 1, za.za_num_integers, strval); if (err) { kmem_free(strval, za.za_num_integers); break; } spa_prop_add_list(*nvp, prop, strval, 0, src); kmem_free(strval, za.za_num_integers); break; default: break; } } zap_cursor_fini(&zc); mutex_exit(&spa->spa_props_lock); out: if (err && err != ENOENT) { nvlist_free(*nvp); *nvp = NULL; return (err); } return (0); } /* * Validate the given pool properties nvlist and modify the list * for the property values to be set. */ static int spa_prop_validate(spa_t *spa, nvlist_t *props) { nvpair_t *elem; int error = 0, reset_bootfs = 0; uint64_t objnum = 0; boolean_t has_feature = B_FALSE; elem = NULL; while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { uint64_t intval; char *strval, *slash, *check, *fname; const char *propname = nvpair_name(elem); zpool_prop_t prop = zpool_name_to_prop(propname); switch (prop) { case ZPOOL_PROP_INVAL: if (!zpool_prop_feature(propname)) { error = SET_ERROR(EINVAL); break; } /* * Sanitize the input. */ if (nvpair_type(elem) != DATA_TYPE_UINT64) { error = SET_ERROR(EINVAL); break; } if (nvpair_value_uint64(elem, &intval) != 0) { error = SET_ERROR(EINVAL); break; } if (intval != 0) { error = SET_ERROR(EINVAL); break; } fname = strchr(propname, '@') + 1; if (zfeature_lookup_name(fname, NULL) != 0) { error = SET_ERROR(EINVAL); break; } has_feature = B_TRUE; break; case ZPOOL_PROP_VERSION: error = nvpair_value_uint64(elem, &intval); if (!error && (intval < spa_version(spa) || intval > SPA_VERSION_BEFORE_FEATURES || has_feature)) error = SET_ERROR(EINVAL); break; case ZPOOL_PROP_DELEGATION: case ZPOOL_PROP_AUTOREPLACE: case ZPOOL_PROP_LISTSNAPS: case ZPOOL_PROP_AUTOEXPAND: error = nvpair_value_uint64(elem, &intval); if (!error && intval > 1) error = SET_ERROR(EINVAL); break; case ZPOOL_PROP_BOOTFS: /* * If the pool version is less than SPA_VERSION_BOOTFS, * or the pool is still being created (version == 0), * the bootfs property cannot be set. */ if (spa_version(spa) < SPA_VERSION_BOOTFS) { error = SET_ERROR(ENOTSUP); break; } /* * Make sure the vdev config is bootable */ if (!vdev_is_bootable(spa->spa_root_vdev)) { error = SET_ERROR(ENOTSUP); break; } reset_bootfs = 1; error = nvpair_value_string(elem, &strval); if (!error) { objset_t *os; uint64_t propval; if (strval == NULL || strval[0] == '\0') { objnum = zpool_prop_default_numeric( ZPOOL_PROP_BOOTFS); break; } if (error = dmu_objset_hold(strval, FTAG, &os)) break; /* * Must be ZPL, and its property settings * must be supported by GRUB (compression * is not gzip, and large blocks are not used). */ if (dmu_objset_type(os) != DMU_OST_ZFS) { error = SET_ERROR(ENOTSUP); } else if ((error = dsl_prop_get_int_ds(dmu_objset_ds(os), zfs_prop_to_name(ZFS_PROP_COMPRESSION), &propval)) == 0 && !BOOTFS_COMPRESS_VALID(propval)) { error = SET_ERROR(ENOTSUP); } else { objnum = dmu_objset_id(os); } dmu_objset_rele(os, FTAG); } break; case ZPOOL_PROP_FAILUREMODE: error = nvpair_value_uint64(elem, &intval); if (!error && (intval < ZIO_FAILURE_MODE_WAIT || intval > ZIO_FAILURE_MODE_PANIC)) error = SET_ERROR(EINVAL); /* * This is a special case which only occurs when * the pool has completely failed. This allows * the user to change the in-core failmode property * without syncing it out to disk (I/Os might * currently be blocked). We do this by returning * EIO to the caller (spa_prop_set) to trick it * into thinking we encountered a property validation * error. */ if (!error && spa_suspended(spa)) { spa->spa_failmode = intval; error = SET_ERROR(EIO); } break; case ZPOOL_PROP_CACHEFILE: if ((error = nvpair_value_string(elem, &strval)) != 0) break; if (strval[0] == '\0') break; if (strcmp(strval, "none") == 0) break; if (strval[0] != '/') { error = SET_ERROR(EINVAL); break; } slash = strrchr(strval, '/'); ASSERT(slash != NULL); if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || strcmp(slash, "/..") == 0) error = SET_ERROR(EINVAL); break; case ZPOOL_PROP_COMMENT: if ((error = nvpair_value_string(elem, &strval)) != 0) break; for (check = strval; *check != '\0'; check++) { /* * The kernel doesn't have an easy isprint() * check. For this kernel check, we merely * check ASCII apart from DEL. Fix this if * there is an easy-to-use kernel isprint(). */ if (*check >= 0x7f) { error = SET_ERROR(EINVAL); break; } } if (strlen(strval) > ZPROP_MAX_COMMENT) error = E2BIG; break; case ZPOOL_PROP_DEDUPDITTO: if (spa_version(spa) < SPA_VERSION_DEDUP) error = SET_ERROR(ENOTSUP); else error = nvpair_value_uint64(elem, &intval); if (error == 0 && intval != 0 && intval < ZIO_DEDUPDITTO_MIN) error = SET_ERROR(EINVAL); break; } if (error) break; } if (!error && reset_bootfs) { error = nvlist_remove(props, zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); if (!error) { error = nvlist_add_uint64(props, zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); } } return (error); } void spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) { char *cachefile; spa_config_dirent_t *dp; if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), &cachefile) != 0) return; dp = kmem_alloc(sizeof (spa_config_dirent_t), KM_SLEEP); if (cachefile[0] == '\0') dp->scd_path = spa_strdup(spa_config_path); else if (strcmp(cachefile, "none") == 0) dp->scd_path = NULL; else dp->scd_path = spa_strdup(cachefile); list_insert_head(&spa->spa_config_list, dp); if (need_sync) spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); } int spa_prop_set(spa_t *spa, nvlist_t *nvp) { int error; nvpair_t *elem = NULL; boolean_t need_sync = B_FALSE; if ((error = spa_prop_validate(spa, nvp)) != 0) return (error); while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT || prop == ZPOOL_PROP_READONLY) continue; if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { uint64_t ver; if (prop == ZPOOL_PROP_VERSION) { VERIFY(nvpair_value_uint64(elem, &ver) == 0); } else { ASSERT(zpool_prop_feature(nvpair_name(elem))); ver = SPA_VERSION_FEATURES; need_sync = B_TRUE; } /* Save time if the version is already set. */ if (ver == spa_version(spa)) continue; /* * In addition to the pool directory object, we might * create the pool properties object, the features for * read object, the features for write object, or the * feature descriptions object. */ error = dsl_sync_task(spa->spa_name, NULL, spa_sync_version, &ver, 6, ZFS_SPACE_CHECK_RESERVED); if (error) return (error); continue; } need_sync = B_TRUE; break; } if (need_sync) { return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, nvp, 6, ZFS_SPACE_CHECK_RESERVED)); } return (0); } /* * If the bootfs property value is dsobj, clear it. */ void spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) { if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { VERIFY(zap_remove(spa->spa_meta_objset, spa->spa_pool_props_object, zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); spa->spa_bootfs = 0; } } /*ARGSUSED*/ static int spa_change_guid_check(void *arg, dmu_tx_t *tx) { uint64_t *newguid = arg; spa_t *spa = dmu_tx_pool(tx)->dp_spa; vdev_t *rvd = spa->spa_root_vdev; uint64_t vdev_state; if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { int error = (spa_has_checkpoint(spa)) ? ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; return (SET_ERROR(error)); } spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); vdev_state = rvd->vdev_state; spa_config_exit(spa, SCL_STATE, FTAG); if (vdev_state != VDEV_STATE_HEALTHY) return (SET_ERROR(ENXIO)); ASSERT3U(spa_guid(spa), !=, *newguid); return (0); } static void spa_change_guid_sync(void *arg, dmu_tx_t *tx) { uint64_t *newguid = arg; spa_t *spa = dmu_tx_pool(tx)->dp_spa; uint64_t oldguid; vdev_t *rvd = spa->spa_root_vdev; oldguid = spa_guid(spa); spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); rvd->vdev_guid = *newguid; rvd->vdev_guid_sum += (*newguid - oldguid); vdev_config_dirty(rvd); spa_config_exit(spa, SCL_STATE, FTAG); spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", oldguid, *newguid); } /* * Change the GUID for the pool. This is done so that we can later * re-import a pool built from a clone of our own vdevs. We will modify * the root vdev's guid, our own pool guid, and then mark all of our * vdevs dirty. Note that we must make sure that all our vdevs are * online when we do this, or else any vdevs that weren't present * would be orphaned from our pool. We are also going to issue a * sysevent to update any watchers. */ int spa_change_guid(spa_t *spa) { int error; uint64_t guid; mutex_enter(&spa->spa_vdev_top_lock); mutex_enter(&spa_namespace_lock); guid = spa_generate_guid(NULL); error = dsl_sync_task(spa->spa_name, spa_change_guid_check, spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); if (error == 0) { spa_write_cachefile(spa, B_FALSE, B_TRUE); spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); } mutex_exit(&spa_namespace_lock); mutex_exit(&spa->spa_vdev_top_lock); return (error); } /* * ========================================================================== * SPA state manipulation (open/create/destroy/import/export) * ========================================================================== */ static int spa_error_entry_compare(const void *a, const void *b) { spa_error_entry_t *sa = (spa_error_entry_t *)a; spa_error_entry_t *sb = (spa_error_entry_t *)b; int ret; ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, sizeof (zbookmark_phys_t)); if (ret < 0) return (-1); else if (ret > 0) return (1); else return (0); } /* * Utility function which retrieves copies of the current logs and * re-initializes them in the process. */ void spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) { ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); avl_create(&spa->spa_errlist_scrub, spa_error_entry_compare, sizeof (spa_error_entry_t), offsetof(spa_error_entry_t, se_avl)); avl_create(&spa->spa_errlist_last, spa_error_entry_compare, sizeof (spa_error_entry_t), offsetof(spa_error_entry_t, se_avl)); } static void spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) { const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; enum zti_modes mode = ztip->zti_mode; uint_t value = ztip->zti_value; uint_t count = ztip->zti_count; spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; char name[32]; uint_t flags = 0; boolean_t batch = B_FALSE; if (mode == ZTI_MODE_NULL) { tqs->stqs_count = 0; tqs->stqs_taskq = NULL; return; } ASSERT3U(count, >, 0); tqs->stqs_count = count; tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); switch (mode) { case ZTI_MODE_FIXED: ASSERT3U(value, >=, 1); value = MAX(value, 1); break; case ZTI_MODE_BATCH: batch = B_TRUE; flags |= TASKQ_THREADS_CPU_PCT; value = zio_taskq_batch_pct; break; default: panic("unrecognized mode for %s_%s taskq (%u:%u) in " "spa_activate()", zio_type_name[t], zio_taskq_types[q], mode, value); break; } for (uint_t i = 0; i < count; i++) { taskq_t *tq; if (count > 1) { (void) snprintf(name, sizeof (name), "%s_%s_%u", zio_type_name[t], zio_taskq_types[q], i); } else { (void) snprintf(name, sizeof (name), "%s_%s", zio_type_name[t], zio_taskq_types[q]); } #ifdef SYSDC if (zio_taskq_sysdc && spa->spa_proc != &p0) { if (batch) flags |= TASKQ_DC_BATCH; tq = taskq_create_sysdc(name, value, 50, INT_MAX, spa->spa_proc, zio_taskq_basedc, flags); } else { #endif pri_t pri = maxclsyspri; /* * The write issue taskq can be extremely CPU * intensive. Run it at slightly lower priority * than the other taskqs. * FreeBSD notes: * - numerically higher priorities are lower priorities; * - if priorities divided by four (RQ_PPQ) are equal * then a difference between them is insignificant. */ if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) #ifdef illumos pri--; #else pri += 4; #endif tq = taskq_create_proc(name, value, pri, 50, INT_MAX, spa->spa_proc, flags); #ifdef SYSDC } #endif tqs->stqs_taskq[i] = tq; } } static void spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) { spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; if (tqs->stqs_taskq == NULL) { ASSERT0(tqs->stqs_count); return; } for (uint_t i = 0; i < tqs->stqs_count; i++) { ASSERT3P(tqs->stqs_taskq[i], !=, NULL); taskq_destroy(tqs->stqs_taskq[i]); } kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); tqs->stqs_taskq = NULL; } /* * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. * Note that a type may have multiple discrete taskqs to avoid lock contention * on the taskq itself. In that case we choose which taskq at random by using * the low bits of gethrtime(). */ void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) { spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; taskq_t *tq; ASSERT3P(tqs->stqs_taskq, !=, NULL); ASSERT3U(tqs->stqs_count, !=, 0); if (tqs->stqs_count == 1) { tq = tqs->stqs_taskq[0]; } else { #ifdef _KERNEL tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; #else tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; #endif } taskq_dispatch_ent(tq, func, arg, flags, ent); } static void spa_create_zio_taskqs(spa_t *spa) { for (int t = 0; t < ZIO_TYPES; t++) { for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { spa_taskqs_init(spa, t, q); } } } #ifdef _KERNEL #ifdef SPA_PROCESS static void spa_thread(void *arg) { callb_cpr_t cprinfo; spa_t *spa = arg; user_t *pu = PTOU(curproc); CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, spa->spa_name); ASSERT(curproc != &p0); (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), "zpool-%s", spa->spa_name); (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); #ifdef PSRSET_BIND /* bind this thread to the requested psrset */ if (zio_taskq_psrset_bind != PS_NONE) { pool_lock(); mutex_enter(&cpu_lock); mutex_enter(&pidlock); mutex_enter(&curproc->p_lock); if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 0, NULL, NULL) == 0) { curthread->t_bind_pset = zio_taskq_psrset_bind; } else { cmn_err(CE_WARN, "Couldn't bind process for zfs pool \"%s\" to " "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); } mutex_exit(&curproc->p_lock); mutex_exit(&pidlock); mutex_exit(&cpu_lock); pool_unlock(); } #endif #ifdef SYSDC if (zio_taskq_sysdc) { sysdc_thread_enter(curthread, 100, 0); } #endif spa->spa_proc = curproc; spa->spa_did = curthread->t_did; spa_create_zio_taskqs(spa); mutex_enter(&spa->spa_proc_lock); ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); spa->spa_proc_state = SPA_PROC_ACTIVE; cv_broadcast(&spa->spa_proc_cv); CALLB_CPR_SAFE_BEGIN(&cprinfo); while (spa->spa_proc_state == SPA_PROC_ACTIVE) cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); spa->spa_proc_state = SPA_PROC_GONE; spa->spa_proc = &p0; cv_broadcast(&spa->spa_proc_cv); CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ mutex_enter(&curproc->p_lock); lwp_exit(); } #endif /* SPA_PROCESS */ #endif /* * Activate an uninitialized pool. */ static void spa_activate(spa_t *spa, int mode) { ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); spa->spa_state = POOL_STATE_ACTIVE; spa->spa_mode = mode; spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); /* Try to create a covering process */ mutex_enter(&spa->spa_proc_lock); ASSERT(spa->spa_proc_state == SPA_PROC_NONE); ASSERT(spa->spa_proc == &p0); spa->spa_did = 0; #ifdef SPA_PROCESS /* Only create a process if we're going to be around a while. */ if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, NULL, 0) == 0) { spa->spa_proc_state = SPA_PROC_CREATED; while (spa->spa_proc_state == SPA_PROC_CREATED) { cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); } ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); ASSERT(spa->spa_proc != &p0); ASSERT(spa->spa_did != 0); } else { #ifdef _KERNEL cmn_err(CE_WARN, "Couldn't create process for zfs pool \"%s\"\n", spa->spa_name); #endif } } #endif /* SPA_PROCESS */ mutex_exit(&spa->spa_proc_lock); /* If we didn't create a process, we need to create our taskqs. */ ASSERT(spa->spa_proc == &p0); if (spa->spa_proc == &p0) { spa_create_zio_taskqs(spa); } /* * Start TRIM thread. */ trim_thread_create(spa); for (size_t i = 0; i < TXG_SIZE; i++) spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 0); list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), offsetof(vdev_t, vdev_config_dirty_node)); list_create(&spa->spa_evicting_os_list, sizeof (objset_t), offsetof(objset_t, os_evicting_node)); list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), offsetof(vdev_t, vdev_state_dirty_node)); txg_list_create(&spa->spa_vdev_txg_list, spa, offsetof(struct vdev, vdev_txg_node)); avl_create(&spa->spa_errlist_scrub, spa_error_entry_compare, sizeof (spa_error_entry_t), offsetof(spa_error_entry_t, se_avl)); avl_create(&spa->spa_errlist_last, spa_error_entry_compare, sizeof (spa_error_entry_t), offsetof(spa_error_entry_t, se_avl)); } /* * Opposite of spa_activate(). */ static void spa_deactivate(spa_t *spa) { ASSERT(spa->spa_sync_on == B_FALSE); ASSERT(spa->spa_dsl_pool == NULL); ASSERT(spa->spa_root_vdev == NULL); ASSERT(spa->spa_async_zio_root == NULL); ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); /* * Stop TRIM thread in case spa_unload() wasn't called directly * before spa_deactivate(). */ trim_thread_destroy(spa); spa_evicting_os_wait(spa); txg_list_destroy(&spa->spa_vdev_txg_list); list_destroy(&spa->spa_config_dirty_list); list_destroy(&spa->spa_evicting_os_list); list_destroy(&spa->spa_state_dirty_list); for (int t = 0; t < ZIO_TYPES; t++) { for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { spa_taskqs_fini(spa, t, q); } } for (size_t i = 0; i < TXG_SIZE; i++) { ASSERT3P(spa->spa_txg_zio[i], !=, NULL); VERIFY0(zio_wait(spa->spa_txg_zio[i])); spa->spa_txg_zio[i] = NULL; } metaslab_class_destroy(spa->spa_normal_class); spa->spa_normal_class = NULL; metaslab_class_destroy(spa->spa_log_class); spa->spa_log_class = NULL; /* * If this was part of an import or the open otherwise failed, we may * still have errors left in the queues. Empty them just in case. */ spa_errlog_drain(spa); avl_destroy(&spa->spa_errlist_scrub); avl_destroy(&spa->spa_errlist_last); spa->spa_state = POOL_STATE_UNINITIALIZED; mutex_enter(&spa->spa_proc_lock); if (spa->spa_proc_state != SPA_PROC_NONE) { ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); spa->spa_proc_state = SPA_PROC_DEACTIVATE; cv_broadcast(&spa->spa_proc_cv); while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { ASSERT(spa->spa_proc != &p0); cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); } ASSERT(spa->spa_proc_state == SPA_PROC_GONE); spa->spa_proc_state = SPA_PROC_NONE; } ASSERT(spa->spa_proc == &p0); mutex_exit(&spa->spa_proc_lock); #ifdef SPA_PROCESS /* * We want to make sure spa_thread() has actually exited the ZFS * module, so that the module can't be unloaded out from underneath * it. */ if (spa->spa_did != 0) { thread_join(spa->spa_did); spa->spa_did = 0; } #endif /* SPA_PROCESS */ } /* * Verify a pool configuration, and construct the vdev tree appropriately. This * will create all the necessary vdevs in the appropriate layout, with each vdev * in the CLOSED state. This will prep the pool before open/creation/import. * All vdev validation is done by the vdev_alloc() routine. */ static int spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, int atype) { nvlist_t **child; uint_t children; int error; if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) return (error); if ((*vdp)->vdev_ops->vdev_op_leaf) return (0); error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, &child, &children); if (error == ENOENT) return (0); if (error) { vdev_free(*vdp); *vdp = NULL; return (SET_ERROR(EINVAL)); } for (int c = 0; c < children; c++) { vdev_t *vd; if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, atype)) != 0) { vdev_free(*vdp); *vdp = NULL; return (error); } } ASSERT(*vdp != NULL); return (0); } /* * Opposite of spa_load(). */ static void spa_unload(spa_t *spa) { int i; ASSERT(MUTEX_HELD(&spa_namespace_lock)); spa_load_note(spa, "UNLOADING"); /* * Stop TRIM thread. */ trim_thread_destroy(spa); /* * Stop async tasks. */ spa_async_suspend(spa); /* * Stop syncing. */ if (spa->spa_sync_on) { txg_sync_stop(spa->spa_dsl_pool); spa->spa_sync_on = B_FALSE; } /* * Even though vdev_free() also calls vdev_metaslab_fini, we need * to call it earlier, before we wait for async i/o to complete. * This ensures that there is no async metaslab prefetching, by * calling taskq_wait(mg_taskq). */ if (spa->spa_root_vdev != NULL) { spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); spa_config_exit(spa, SCL_ALL, FTAG); } /* * Wait for any outstanding async I/O to complete. */ if (spa->spa_async_zio_root != NULL) { for (int i = 0; i < max_ncpus; i++) (void) zio_wait(spa->spa_async_zio_root[i]); kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); spa->spa_async_zio_root = NULL; } if (spa->spa_vdev_removal != NULL) { spa_vdev_removal_destroy(spa->spa_vdev_removal); spa->spa_vdev_removal = NULL; } if (spa->spa_condense_zthr != NULL) { ASSERT(!zthr_isrunning(spa->spa_condense_zthr)); zthr_destroy(spa->spa_condense_zthr); spa->spa_condense_zthr = NULL; } if (spa->spa_checkpoint_discard_zthr != NULL) { ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr)); zthr_destroy(spa->spa_checkpoint_discard_zthr); spa->spa_checkpoint_discard_zthr = NULL; } spa_condense_fini(spa); bpobj_close(&spa->spa_deferred_bpobj); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); /* * Close all vdevs. */ if (spa->spa_root_vdev) vdev_free(spa->spa_root_vdev); ASSERT(spa->spa_root_vdev == NULL); /* * Close the dsl pool. */ if (spa->spa_dsl_pool) { dsl_pool_close(spa->spa_dsl_pool); spa->spa_dsl_pool = NULL; spa->spa_meta_objset = NULL; } ddt_unload(spa); /* * Drop and purge level 2 cache */ spa_l2cache_drop(spa); for (i = 0; i < spa->spa_spares.sav_count; i++) vdev_free(spa->spa_spares.sav_vdevs[i]); if (spa->spa_spares.sav_vdevs) { kmem_free(spa->spa_spares.sav_vdevs, spa->spa_spares.sav_count * sizeof (void *)); spa->spa_spares.sav_vdevs = NULL; } if (spa->spa_spares.sav_config) { nvlist_free(spa->spa_spares.sav_config); spa->spa_spares.sav_config = NULL; } spa->spa_spares.sav_count = 0; for (i = 0; i < spa->spa_l2cache.sav_count; i++) { vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); vdev_free(spa->spa_l2cache.sav_vdevs[i]); } if (spa->spa_l2cache.sav_vdevs) { kmem_free(spa->spa_l2cache.sav_vdevs, spa->spa_l2cache.sav_count * sizeof (void *)); spa->spa_l2cache.sav_vdevs = NULL; } if (spa->spa_l2cache.sav_config) { nvlist_free(spa->spa_l2cache.sav_config); spa->spa_l2cache.sav_config = NULL; } spa->spa_l2cache.sav_count = 0; spa->spa_async_suspended = 0; spa->spa_indirect_vdevs_loaded = B_FALSE; if (spa->spa_comment != NULL) { spa_strfree(spa->spa_comment); spa->spa_comment = NULL; } spa_config_exit(spa, SCL_ALL, FTAG); } /* * Load (or re-load) the current list of vdevs describing the active spares for * this pool. When this is called, we have some form of basic information in * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and * then re-generate a more complete list including status information. */ void spa_load_spares(spa_t *spa) { nvlist_t **spares; uint_t nspares; int i; vdev_t *vd, *tvd; #ifndef _KERNEL /* * zdb opens both the current state of the pool and the * checkpointed state (if present), with a different spa_t. * * As spare vdevs are shared among open pools, we skip loading * them when we load the checkpointed state of the pool. */ if (!spa_writeable(spa)) return; #endif ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); /* * First, close and free any existing spare vdevs. */ for (i = 0; i < spa->spa_spares.sav_count; i++) { vd = spa->spa_spares.sav_vdevs[i]; /* Undo the call to spa_activate() below */ if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, B_FALSE)) != NULL && tvd->vdev_isspare) spa_spare_remove(tvd); vdev_close(vd); vdev_free(vd); } if (spa->spa_spares.sav_vdevs) kmem_free(spa->spa_spares.sav_vdevs, spa->spa_spares.sav_count * sizeof (void *)); if (spa->spa_spares.sav_config == NULL) nspares = 0; else VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); spa->spa_spares.sav_count = (int)nspares; spa->spa_spares.sav_vdevs = NULL; if (nspares == 0) return; /* * Construct the array of vdevs, opening them to get status in the * process. For each spare, there is potentially two different vdev_t * structures associated with it: one in the list of spares (used only * for basic validation purposes) and one in the active vdev * configuration (if it's spared in). During this phase we open and * validate each vdev on the spare list. If the vdev also exists in the * active configuration, then we also mark this vdev as an active spare. */ spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), KM_SLEEP); for (i = 0; i < spa->spa_spares.sav_count; i++) { VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, VDEV_ALLOC_SPARE) == 0); ASSERT(vd != NULL); spa->spa_spares.sav_vdevs[i] = vd; if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, B_FALSE)) != NULL) { if (!tvd->vdev_isspare) spa_spare_add(tvd); /* * We only mark the spare active if we were successfully * able to load the vdev. Otherwise, importing a pool * with a bad active spare would result in strange * behavior, because multiple pool would think the spare * is actively in use. * * There is a vulnerability here to an equally bizarre * circumstance, where a dead active spare is later * brought back to life (onlined or otherwise). Given * the rarity of this scenario, and the extra complexity * it adds, we ignore the possibility. */ if (!vdev_is_dead(tvd)) spa_spare_activate(tvd); } vd->vdev_top = vd; vd->vdev_aux = &spa->spa_spares; if (vdev_open(vd) != 0) continue; if (vdev_validate_aux(vd) == 0) spa_spare_add(vd); } /* * Recompute the stashed list of spares, with status information * this time. */ VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), KM_SLEEP); for (i = 0; i < spa->spa_spares.sav_count; i++) spares[i] = vdev_config_generate(spa, spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); for (i = 0; i < spa->spa_spares.sav_count; i++) nvlist_free(spares[i]); kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); } /* * Load (or re-load) the current list of vdevs describing the active l2cache for * this pool. When this is called, we have some form of basic information in * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and * then re-generate a more complete list including status information. * Devices which are already active have their details maintained, and are * not re-opened. */ void spa_load_l2cache(spa_t *spa) { nvlist_t **l2cache; uint_t nl2cache; int i, j, oldnvdevs; uint64_t guid; vdev_t *vd, **oldvdevs, **newvdevs; spa_aux_vdev_t *sav = &spa->spa_l2cache; #ifndef _KERNEL /* * zdb opens both the current state of the pool and the * checkpointed state (if present), with a different spa_t. * * As L2 caches are part of the ARC which is shared among open * pools, we skip loading them when we load the checkpointed * state of the pool. */ if (!spa_writeable(spa)) return; #endif ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); if (sav->sav_config != NULL) { VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); } else { nl2cache = 0; newvdevs = NULL; } oldvdevs = sav->sav_vdevs; oldnvdevs = sav->sav_count; sav->sav_vdevs = NULL; sav->sav_count = 0; /* * Process new nvlist of vdevs. */ for (i = 0; i < nl2cache; i++) { VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, &guid) == 0); newvdevs[i] = NULL; for (j = 0; j < oldnvdevs; j++) { vd = oldvdevs[j]; if (vd != NULL && guid == vd->vdev_guid) { /* * Retain previous vdev for add/remove ops. */ newvdevs[i] = vd; oldvdevs[j] = NULL; break; } } if (newvdevs[i] == NULL) { /* * Create new vdev */ VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, VDEV_ALLOC_L2CACHE) == 0); ASSERT(vd != NULL); newvdevs[i] = vd; /* * Commit this vdev as an l2cache device, * even if it fails to open. */ spa_l2cache_add(vd); vd->vdev_top = vd; vd->vdev_aux = sav; spa_l2cache_activate(vd); if (vdev_open(vd) != 0) continue; (void) vdev_validate_aux(vd); if (!vdev_is_dead(vd)) l2arc_add_vdev(spa, vd); } } /* * Purge vdevs that were dropped */ for (i = 0; i < oldnvdevs; i++) { uint64_t pool; vd = oldvdevs[i]; if (vd != NULL) { ASSERT(vd->vdev_isl2cache); if (spa_l2cache_exists(vd->vdev_guid, &pool) && pool != 0ULL && l2arc_vdev_present(vd)) l2arc_remove_vdev(vd); vdev_clear_stats(vd); vdev_free(vd); } } if (oldvdevs) kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); if (sav->sav_config == NULL) goto out; sav->sav_vdevs = newvdevs; sav->sav_count = (int)nl2cache; /* * Recompute the stashed list of l2cache devices, with status * information this time. */ VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); for (i = 0; i < sav->sav_count; i++) l2cache[i] = vdev_config_generate(spa, sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); VERIFY(nvlist_add_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); out: for (i = 0; i < sav->sav_count; i++) nvlist_free(l2cache[i]); if (sav->sav_count) kmem_free(l2cache, sav->sav_count * sizeof (void *)); } static int load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) { dmu_buf_t *db; char *packed = NULL; size_t nvsize = 0; int error; *value = NULL; error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); if (error != 0) return (error); nvsize = *(uint64_t *)db->db_data; dmu_buf_rele(db, FTAG); packed = kmem_alloc(nvsize, KM_SLEEP); error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, DMU_READ_PREFETCH); if (error == 0) error = nvlist_unpack(packed, nvsize, value, 0); kmem_free(packed, nvsize); return (error); } /* * Concrete top-level vdevs that are not missing and are not logs. At every * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds. */ static uint64_t spa_healthy_core_tvds(spa_t *spa) { vdev_t *rvd = spa->spa_root_vdev; uint64_t tvds = 0; for (uint64_t i = 0; i < rvd->vdev_children; i++) { vdev_t *vd = rvd->vdev_child[i]; if (vd->vdev_islog) continue; if (vdev_is_concrete(vd) && !vdev_is_dead(vd)) tvds++; } return (tvds); } /* * Checks to see if the given vdev could not be opened, in which case we post a * sysevent to notify the autoreplace code that the device has been removed. */ static void spa_check_removed(vdev_t *vd) { for (uint64_t c = 0; c < vd->vdev_children; c++) spa_check_removed(vd->vdev_child[c]); if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && vdev_is_concrete(vd)) { zfs_post_autoreplace(vd->vdev_spa, vd); spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); } } static int spa_check_for_missing_logs(spa_t *spa) { vdev_t *rvd = spa->spa_root_vdev; /* * If we're doing a normal import, then build up any additional * diagnostic information about missing log devices. * We'll pass this up to the user for further processing. */ if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { nvlist_t **child, *nv; uint64_t idx = 0; child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), KM_SLEEP); VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); for (uint64_t c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; /* * We consider a device as missing only if it failed * to open (i.e. offline or faulted is not considered * as missing). */ if (tvd->vdev_islog && tvd->vdev_state == VDEV_STATE_CANT_OPEN) { child[idx++] = vdev_config_generate(spa, tvd, B_FALSE, VDEV_CONFIG_MISSING); } } if (idx > 0) { fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, child, idx); fnvlist_add_nvlist(spa->spa_load_info, ZPOOL_CONFIG_MISSING_DEVICES, nv); for (uint64_t i = 0; i < idx; i++) nvlist_free(child[i]); } nvlist_free(nv); kmem_free(child, rvd->vdev_children * sizeof (char **)); if (idx > 0) { spa_load_failed(spa, "some log devices are missing"); vdev_dbgmsg_print_tree(rvd, 2); return (SET_ERROR(ENXIO)); } } else { for (uint64_t c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; if (tvd->vdev_islog && tvd->vdev_state == VDEV_STATE_CANT_OPEN) { spa_set_log_state(spa, SPA_LOG_CLEAR); spa_load_note(spa, "some log devices are " "missing, ZIL is dropped."); vdev_dbgmsg_print_tree(rvd, 2); break; } } } return (0); } /* * Check for missing log devices */ static boolean_t spa_check_logs(spa_t *spa) { boolean_t rv = B_FALSE; dsl_pool_t *dp = spa_get_dsl(spa); switch (spa->spa_log_state) { case SPA_LOG_MISSING: /* need to recheck in case slog has been restored */ case SPA_LOG_UNKNOWN: rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); if (rv) spa_set_log_state(spa, SPA_LOG_MISSING); break; } return (rv); } static boolean_t spa_passivate_log(spa_t *spa) { vdev_t *rvd = spa->spa_root_vdev; boolean_t slog_found = B_FALSE; ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); if (!spa_has_slogs(spa)) return (B_FALSE); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; if (tvd->vdev_islog) { metaslab_group_passivate(mg); slog_found = B_TRUE; } } return (slog_found); } static void spa_activate_log(spa_t *spa) { vdev_t *rvd = spa->spa_root_vdev; ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; if (tvd->vdev_islog) metaslab_group_activate(mg); } } int spa_reset_logs(spa_t *spa) { int error; error = dmu_objset_find(spa_name(spa), zil_reset, NULL, DS_FIND_CHILDREN); if (error == 0) { /* * We successfully offlined the log device, sync out the * current txg so that the "stubby" block can be removed * by zil_sync(). */ txg_wait_synced(spa->spa_dsl_pool, 0); } return (error); } static void spa_aux_check_removed(spa_aux_vdev_t *sav) { int i; for (i = 0; i < sav->sav_count; i++) spa_check_removed(sav->sav_vdevs[i]); } void spa_claim_notify(zio_t *zio) { spa_t *spa = zio->io_spa; if (zio->io_error) return; mutex_enter(&spa->spa_props_lock); /* any mutex will do */ if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) spa->spa_claim_max_txg = zio->io_bp->blk_birth; mutex_exit(&spa->spa_props_lock); } typedef struct spa_load_error { uint64_t sle_meta_count; uint64_t sle_data_count; } spa_load_error_t; static void spa_load_verify_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; spa_load_error_t *sle = zio->io_private; dmu_object_type_t type = BP_GET_TYPE(bp); int error = zio->io_error; spa_t *spa = zio->io_spa; abd_free(zio->io_abd); if (error) { if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && type != DMU_OT_INTENT_LOG) atomic_inc_64(&sle->sle_meta_count); else atomic_inc_64(&sle->sle_data_count); } mutex_enter(&spa->spa_scrub_lock); spa->spa_load_verify_ios--; cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&spa->spa_scrub_lock); } /* * Maximum number of concurrent scrub i/os to create while verifying * a pool while importing it. */ int spa_load_verify_maxinflight = 10000; boolean_t spa_load_verify_metadata = B_TRUE; boolean_t spa_load_verify_data = B_TRUE; SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, &spa_load_verify_maxinflight, 0, "Maximum number of concurrent scrub I/Os to create while verifying a " "pool while importing it"); SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, &spa_load_verify_metadata, 0, "Check metadata on import?"); SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, &spa_load_verify_data, 0, "Check user data on import?"); /*ARGSUSED*/ static int spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) { if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) return (0); /* * Note: normally this routine will not be called if * spa_load_verify_metadata is not set. However, it may be useful * to manually set the flag after the traversal has begun. */ if (!spa_load_verify_metadata) return (0); if (!BP_IS_METADATA(bp) && !spa_load_verify_data) return (0); zio_t *rio = arg; size_t size = BP_GET_PSIZE(bp); mutex_enter(&spa->spa_scrub_lock); while (spa->spa_load_verify_ios >= spa_load_verify_maxinflight) cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); spa->spa_load_verify_ios++; mutex_exit(&spa->spa_scrub_lock); zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); return (0); } /* ARGSUSED */ int verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) { if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) return (SET_ERROR(ENAMETOOLONG)); return (0); } static int spa_load_verify(spa_t *spa) { zio_t *rio; spa_load_error_t sle = { 0 }; zpool_load_policy_t policy; boolean_t verify_ok = B_FALSE; int error = 0; zpool_get_load_policy(spa->spa_config, &policy); if (policy.zlp_rewind & ZPOOL_NEVER_REWIND) return (0); dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); error = dmu_objset_find_dp(spa->spa_dsl_pool, spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, DS_FIND_CHILDREN); dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); if (error != 0) return (error); rio = zio_root(spa, NULL, &sle, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); if (spa_load_verify_metadata) { if (spa->spa_extreme_rewind) { spa_load_note(spa, "performing a complete scan of the " "pool since extreme rewind is on. This may take " "a very long time.\n (spa_load_verify_data=%u, " "spa_load_verify_metadata=%u)", spa_load_verify_data, spa_load_verify_metadata); } error = traverse_pool(spa, spa->spa_verify_min_txg, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, spa_load_verify_cb, rio); } (void) zio_wait(rio); spa->spa_load_meta_errors = sle.sle_meta_count; spa->spa_load_data_errors = sle.sle_data_count; if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) { spa_load_note(spa, "spa_load_verify found %llu metadata errors " "and %llu data errors", (u_longlong_t)sle.sle_meta_count, (u_longlong_t)sle.sle_data_count); } if (spa_load_verify_dryrun || (!error && sle.sle_meta_count <= policy.zlp_maxmeta && sle.sle_data_count <= policy.zlp_maxdata)) { int64_t loss = 0; verify_ok = B_TRUE; spa->spa_load_txg = spa->spa_uberblock.ub_txg; spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; VERIFY(nvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); VERIFY(nvlist_add_int64(spa->spa_load_info, ZPOOL_CONFIG_REWIND_TIME, loss) == 0); VERIFY(nvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); } else { spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; } if (spa_load_verify_dryrun) return (0); if (error) { if (error != ENXIO && error != EIO) error = SET_ERROR(EIO); return (error); } return (verify_ok ? 0 : EIO); } /* * Find a value in the pool props object. */ static void spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) { (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); } /* * Find a value in the pool directory object. */ static int spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent) { int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, name, sizeof (uint64_t), 1, val); if (error != 0 && (error != ENOENT || log_enoent)) { spa_load_failed(spa, "couldn't get '%s' value in MOS directory " "[error=%d]", name, error); } return (error); } static int spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) { vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); return (SET_ERROR(err)); } static void spa_spawn_aux_threads(spa_t *spa) { ASSERT(spa_writeable(spa)); ASSERT(MUTEX_HELD(&spa_namespace_lock)); spa_start_indirect_condensing_thread(spa); ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL); spa->spa_checkpoint_discard_zthr = zthr_create(spa_checkpoint_discard_thread_check, spa_checkpoint_discard_thread, spa); } /* * Fix up config after a partly-completed split. This is done with the * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off * pool have that entry in their config, but only the splitting one contains * a list of all the guids of the vdevs that are being split off. * * This function determines what to do with that list: either rejoin * all the disks to the pool, or complete the splitting process. To attempt * the rejoin, each disk that is offlined is marked online again, and * we do a reopen() call. If the vdev label for every disk that was * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) * then we call vdev_split() on each disk, and complete the split. * * Otherwise we leave the config alone, with all the vdevs in place in * the original pool. */ static void spa_try_repair(spa_t *spa, nvlist_t *config) { uint_t extracted; uint64_t *glist; uint_t i, gcount; nvlist_t *nvl; vdev_t **vd; boolean_t attempt_reopen; if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) return; /* check that the config is complete */ if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, &glist, &gcount) != 0) return; vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); /* attempt to online all the vdevs & validate */ attempt_reopen = B_TRUE; for (i = 0; i < gcount; i++) { if (glist[i] == 0) /* vdev is hole */ continue; vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); if (vd[i] == NULL) { /* * Don't bother attempting to reopen the disks; * just do the split. */ attempt_reopen = B_FALSE; } else { /* attempt to re-online it */ vd[i]->vdev_offline = B_FALSE; } } if (attempt_reopen) { vdev_reopen(spa->spa_root_vdev); /* check each device to see what state it's in */ for (extracted = 0, i = 0; i < gcount; i++) { if (vd[i] != NULL && vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) break; ++extracted; } } /* * If every disk has been moved to the new pool, or if we never * even attempted to look at them, then we split them off for * good. */ if (!attempt_reopen || gcount == extracted) { for (i = 0; i < gcount; i++) if (vd[i] != NULL) vdev_split(vd[i]); vdev_reopen(spa->spa_root_vdev); } kmem_free(vd, gcount * sizeof (vdev_t *)); } static int spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type) { char *ereport = FM_EREPORT_ZFS_POOL; int error; spa->spa_load_state = state; gethrestime(&spa->spa_loaded_ts); error = spa_load_impl(spa, type, &ereport); /* * Don't count references from objsets that are already closed * and are making their way through the eviction process. */ spa_evicting_os_wait(spa); spa->spa_minref = refcount_count(&spa->spa_refcount); if (error) { if (error != EEXIST) { spa->spa_loaded_ts.tv_sec = 0; spa->spa_loaded_ts.tv_nsec = 0; } if (error != EBADF) { zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); } } spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; spa->spa_ena = 0; return (error); } /* * Count the number of per-vdev ZAPs associated with all of the vdevs in the * vdev tree rooted in the given vd, and ensure that each ZAP is present in the * spa's per-vdev ZAP list. */ static uint64_t vdev_count_verify_zaps(vdev_t *vd) { spa_t *spa = vd->vdev_spa; uint64_t total = 0; if (vd->vdev_top_zap != 0) { total++; ASSERT0(zap_lookup_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, vd->vdev_top_zap)); } if (vd->vdev_leaf_zap != 0) { total++; ASSERT0(zap_lookup_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); } for (uint64_t i = 0; i < vd->vdev_children; i++) { total += vdev_count_verify_zaps(vd->vdev_child[i]); } return (total); } static int spa_verify_host(spa_t *spa, nvlist_t *mos_config) { uint64_t hostid; char *hostname; uint64_t myhostid = 0; if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, ZPOOL_CONFIG_HOSTID, &hostid) == 0) { hostname = fnvlist_lookup_string(mos_config, ZPOOL_CONFIG_HOSTNAME); myhostid = zone_get_hostid(NULL); if (hostid != 0 && myhostid != 0 && hostid != myhostid) { cmn_err(CE_WARN, "pool '%s' could not be " "loaded as it was last accessed by " "another system (host: %s hostid: 0x%llx). " "See: http://illumos.org/msg/ZFS-8000-EY", spa_name(spa), hostname, (u_longlong_t)hostid); spa_load_failed(spa, "hostid verification failed: pool " "last accessed by host: %s (hostid: 0x%llx)", hostname, (u_longlong_t)hostid); return (SET_ERROR(EBADF)); } } return (0); } static int spa_ld_parse_config(spa_t *spa, spa_import_type_t type) { int error = 0; nvlist_t *nvtree, *nvl, *config = spa->spa_config; int parse; vdev_t *rvd; uint64_t pool_guid; char *comment; /* * Versioning wasn't explicitly added to the label until later, so if * it's not present treat it as the initial version. */ if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &spa->spa_ubsync.ub_version) != 0) spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { spa_load_failed(spa, "invalid config provided: '%s' missing", ZPOOL_CONFIG_POOL_GUID); return (SET_ERROR(EINVAL)); } /* * If we are doing an import, ensure that the pool is not already * imported by checking if its pool guid already exists in the * spa namespace. * * The only case that we allow an already imported pool to be * imported again, is when the pool is checkpointed and we want to * look at its checkpointed state from userland tools like zdb. */ #ifdef _KERNEL if ((spa->spa_load_state == SPA_LOAD_IMPORT || spa->spa_load_state == SPA_LOAD_TRYIMPORT) && spa_guid_exists(pool_guid, 0)) { #else if ((spa->spa_load_state == SPA_LOAD_IMPORT || spa->spa_load_state == SPA_LOAD_TRYIMPORT) && spa_guid_exists(pool_guid, 0) && !spa_importing_readonly_checkpoint(spa)) { #endif spa_load_failed(spa, "a pool with guid %llu is already open", (u_longlong_t)pool_guid); return (SET_ERROR(EEXIST)); } spa->spa_config_guid = pool_guid; nvlist_free(spa->spa_load_info); spa->spa_load_info = fnvlist_alloc(); ASSERT(spa->spa_comment == NULL); if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) spa->spa_comment = spa_strdup(comment); (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &spa->spa_config_txg); if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) spa->spa_config_splitting = fnvlist_dup(nvl); if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) { spa_load_failed(spa, "invalid config provided: '%s' missing", ZPOOL_CONFIG_VDEV_TREE); return (SET_ERROR(EINVAL)); } /* * Create "The Godfather" zio to hold all async IOs */ spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), KM_SLEEP); for (int i = 0; i < max_ncpus; i++) { spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); } /* * Parse the configuration into a vdev tree. We explicitly set the * value that will be returned by spa_version() since parsing the * configuration requires knowing the version number. */ spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); parse = (type == SPA_IMPORT_EXISTING ? VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse); spa_config_exit(spa, SCL_ALL, FTAG); if (error != 0) { spa_load_failed(spa, "unable to parse config [error=%d]", error); return (error); } ASSERT(spa->spa_root_vdev == rvd); ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); if (type != SPA_IMPORT_ASSEMBLE) { ASSERT(spa_guid(spa) == pool_guid); } return (0); } /* * Recursively open all vdevs in the vdev tree. This function is called twice: * first with the untrusted config, then with the trusted config. */ static int spa_ld_open_vdevs(spa_t *spa) { int error = 0; /* * spa_missing_tvds_allowed defines how many top-level vdevs can be * missing/unopenable for the root vdev to be still considered openable. */ if (spa->spa_trust_config) { spa->spa_missing_tvds_allowed = zfs_max_missing_tvds; } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) { spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile; } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) { spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan; } else { spa->spa_missing_tvds_allowed = 0; } spa->spa_missing_tvds_allowed = MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = vdev_open(spa->spa_root_vdev); spa_config_exit(spa, SCL_ALL, FTAG); if (spa->spa_missing_tvds != 0) { spa_load_note(spa, "vdev tree has %lld missing top-level " "vdevs.", (u_longlong_t)spa->spa_missing_tvds); if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) { /* * Although theoretically we could allow users to open * incomplete pools in RW mode, we'd need to add a lot * of extra logic (e.g. adjust pool space to account * for missing vdevs). * This limitation also prevents users from accidentally * opening the pool in RW mode during data recovery and * damaging it further. */ spa_load_note(spa, "pools with missing top-level " "vdevs can only be opened in read-only mode."); error = SET_ERROR(ENXIO); } else { spa_load_note(spa, "current settings allow for maximum " "%lld missing top-level vdevs at this stage.", (u_longlong_t)spa->spa_missing_tvds_allowed); } } if (error != 0) { spa_load_failed(spa, "unable to open vdev tree [error=%d]", error); } if (spa->spa_missing_tvds != 0 || error != 0) vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2); return (error); } /* * We need to validate the vdev labels against the configuration that * we have in hand. This function is called twice: first with an untrusted * config, then with a trusted config. The validation is more strict when the * config is trusted. */ static int spa_ld_validate_vdevs(spa_t *spa) { int error = 0; vdev_t *rvd = spa->spa_root_vdev; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = vdev_validate(rvd); spa_config_exit(spa, SCL_ALL, FTAG); if (error != 0) { spa_load_failed(spa, "vdev_validate failed [error=%d]", error); return (error); } if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { spa_load_failed(spa, "cannot open vdev tree after invalidating " "some vdevs"); vdev_dbgmsg_print_tree(rvd, 2); return (SET_ERROR(ENXIO)); } return (0); } static void spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub) { spa->spa_state = POOL_STATE_ACTIVE; spa->spa_ubsync = spa->spa_uberblock; spa->spa_verify_min_txg = spa->spa_extreme_rewind ? TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; spa->spa_first_txg = spa->spa_last_ubsync_txg ? spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; spa->spa_claim_max_txg = spa->spa_first_txg; spa->spa_prev_software_version = ub->ub_software_version; } static int spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type) { vdev_t *rvd = spa->spa_root_vdev; nvlist_t *label; uberblock_t *ub = &spa->spa_uberblock; /* * If we are opening the checkpointed state of the pool by * rewinding to it, at this point we will have written the * checkpointed uberblock to the vdev labels, so searching * the labels will find the right uberblock. However, if * we are opening the checkpointed state read-only, we have * not modified the labels. Therefore, we must ignore the * labels and continue using the spa_uberblock that was set * by spa_ld_checkpoint_rewind. * * Note that it would be fine to ignore the labels when * rewinding (opening writeable) as well. However, if we * crash just after writing the labels, we will end up * searching the labels. Doing so in the common case means * that this code path gets exercised normally, rather than * just in the edge case. */ if (ub->ub_checkpoint_txg != 0 && spa_importing_readonly_checkpoint(spa)) { spa_ld_select_uberblock_done(spa, ub); return (0); } /* * Find the best uberblock. */ vdev_uberblock_load(rvd, ub, &label); /* * If we weren't able to find a single valid uberblock, return failure. */ if (ub->ub_txg == 0) { nvlist_free(label); spa_load_failed(spa, "no valid uberblock found"); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); } spa_load_note(spa, "using uberblock with txg=%llu", (u_longlong_t)ub->ub_txg); /* * If the pool has an unsupported version we can't open it. */ if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { nvlist_free(label); spa_load_failed(spa, "version %llu is not supported", (u_longlong_t)ub->ub_version); return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); } if (ub->ub_version >= SPA_VERSION_FEATURES) { nvlist_t *features; /* * If we weren't able to find what's necessary for reading the * MOS in the label, return failure. */ if (label == NULL) { spa_load_failed(spa, "label config unavailable"); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); } if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { nvlist_free(label); spa_load_failed(spa, "invalid label: '%s' missing", ZPOOL_CONFIG_FEATURES_FOR_READ); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); } /* * Update our in-core representation with the definitive values * from the label. */ nvlist_free(spa->spa_label_features); VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); } nvlist_free(label); /* * Look through entries in the label nvlist's features_for_read. If * there is a feature listed there which we don't understand then we * cannot open a pool. */ if (ub->ub_version >= SPA_VERSION_FEATURES) { nvlist_t *unsup_feat; VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 0); for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, NULL); nvp != NULL; nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { if (!zfeature_is_supported(nvpair_name(nvp))) { VERIFY(nvlist_add_string(unsup_feat, nvpair_name(nvp), "") == 0); } } if (!nvlist_empty(unsup_feat)) { VERIFY(nvlist_add_nvlist(spa->spa_load_info, ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); nvlist_free(unsup_feat); spa_load_failed(spa, "some features are unsupported"); return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); } nvlist_free(unsup_feat); } if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_try_repair(spa, spa->spa_config); spa_config_exit(spa, SCL_ALL, FTAG); nvlist_free(spa->spa_config_splitting); spa->spa_config_splitting = NULL; } /* * Initialize internal SPA structures. */ spa_ld_select_uberblock_done(spa, ub); return (0); } static int spa_ld_open_rootbp(spa_t *spa) { int error = 0; vdev_t *rvd = spa->spa_root_vdev; error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); if (error != 0) { spa_load_failed(spa, "unable to open rootbp in dsl_pool_init " "[error=%d]", error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; return (0); } static int spa_ld_trusted_config(spa_t *spa, spa_import_type_t type, boolean_t reloading) { vdev_t *mrvd, *rvd = spa->spa_root_vdev; nvlist_t *nv, *mos_config, *policy; int error = 0, copy_error; uint64_t healthy_tvds, healthy_tvds_mos; uint64_t mos_config_txg; if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); /* * If we're assembling a pool from a split, the config provided is * already trusted so there is nothing to do. */ if (type == SPA_IMPORT_ASSEMBLE) return (0); healthy_tvds = spa_healthy_core_tvds(spa); if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { spa_load_failed(spa, "unable to retrieve MOS config"); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } /* * If we are doing an open, pool owner wasn't verified yet, thus do * the verification here. */ if (spa->spa_load_state == SPA_LOAD_OPEN) { error = spa_verify_host(spa, mos_config); if (error != 0) { nvlist_free(mos_config); return (error); } } nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); /* * Build a new vdev tree from the trusted config */ VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); /* * Vdev paths in the MOS may be obsolete. If the untrusted config was * obtained by scanning /dev/dsk, then it will have the right vdev * paths. We update the trusted MOS config with this information. * We first try to copy the paths with vdev_copy_path_strict, which * succeeds only when both configs have exactly the same vdev tree. * If that fails, we fall back to a more flexible method that has a * best effort policy. */ copy_error = vdev_copy_path_strict(rvd, mrvd); if (copy_error != 0 || spa_load_print_vdev_tree) { spa_load_note(spa, "provided vdev tree:"); vdev_dbgmsg_print_tree(rvd, 2); spa_load_note(spa, "MOS vdev tree:"); vdev_dbgmsg_print_tree(mrvd, 2); } if (copy_error != 0) { spa_load_note(spa, "vdev_copy_path_strict failed, falling " "back to vdev_copy_path_relaxed"); vdev_copy_path_relaxed(rvd, mrvd); } vdev_close(rvd); vdev_free(rvd); spa->spa_root_vdev = mrvd; rvd = mrvd; spa_config_exit(spa, SCL_ALL, FTAG); /* * We will use spa_config if we decide to reload the spa or if spa_load * fails and we rewind. We must thus regenerate the config using the * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to * pass settings on how to load the pool and is not stored in the MOS. * We copy it over to our new, trusted config. */ mos_config_txg = fnvlist_lookup_uint64(mos_config, ZPOOL_CONFIG_POOL_TXG); nvlist_free(mos_config); mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE); if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY, &policy) == 0) fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy); spa_config_set(spa, mos_config); spa->spa_config_source = SPA_CONFIG_SRC_MOS; /* * Now that we got the config from the MOS, we should be more strict * in checking blkptrs and can make assumptions about the consistency * of the vdev tree. spa_trust_config must be set to true before opening * vdevs in order for them to be writeable. */ spa->spa_trust_config = B_TRUE; /* * Open and validate the new vdev tree */ error = spa_ld_open_vdevs(spa); if (error != 0) return (error); error = spa_ld_validate_vdevs(spa); if (error != 0) return (error); if (copy_error != 0 || spa_load_print_vdev_tree) { spa_load_note(spa, "final vdev tree:"); vdev_dbgmsg_print_tree(rvd, 2); } if (spa->spa_load_state != SPA_LOAD_TRYIMPORT && !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) { /* * Sanity check to make sure that we are indeed loading the * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds * in the config provided and they happened to be the only ones * to have the latest uberblock, we could involuntarily perform * an extreme rewind. */ healthy_tvds_mos = spa_healthy_core_tvds(spa); if (healthy_tvds_mos - healthy_tvds >= SPA_SYNC_MIN_VDEVS) { spa_load_note(spa, "config provided misses too many " "top-level vdevs compared to MOS (%lld vs %lld). ", (u_longlong_t)healthy_tvds, (u_longlong_t)healthy_tvds_mos); spa_load_note(spa, "vdev tree:"); vdev_dbgmsg_print_tree(rvd, 2); if (reloading) { spa_load_failed(spa, "config was already " "provided from MOS. Aborting."); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } spa_load_note(spa, "spa must be reloaded using MOS " "config"); return (SET_ERROR(EAGAIN)); } } error = spa_check_for_missing_logs(spa); if (error != 0) return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) { spa_load_failed(spa, "uberblock guid sum doesn't match MOS " "guid sum (%llu != %llu)", (u_longlong_t)spa->spa_uberblock.ub_guid_sum, (u_longlong_t)rvd->vdev_guid_sum); return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); } return (0); } static int spa_ld_open_indirect_vdev_metadata(spa_t *spa) { int error = 0; vdev_t *rvd = spa->spa_root_vdev; /* * Everything that we read before spa_remove_init() must be stored * on concreted vdevs. Therefore we do this as early as possible. */ error = spa_remove_init(spa); if (error != 0) { spa_load_failed(spa, "spa_remove_init failed [error=%d]", error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } /* * Retrieve information needed to condense indirect vdev mappings. */ error = spa_condense_init(spa); if (error != 0) { spa_load_failed(spa, "spa_condense_init failed [error=%d]", error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); } return (0); } static int spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep) { int error = 0; vdev_t *rvd = spa->spa_root_vdev; if (spa_version(spa) >= SPA_VERSION_FEATURES) { boolean_t missing_feat_read = B_FALSE; nvlist_t *unsup_feat, *enabled_feat; if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, &spa->spa_feat_for_read_obj, B_TRUE) != 0) { return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, &spa->spa_feat_for_write_obj, B_TRUE) != 0) { return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, &spa->spa_feat_desc_obj, B_TRUE) != 0) { return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } enabled_feat = fnvlist_alloc(); unsup_feat = fnvlist_alloc(); if (!spa_features_check(spa, B_FALSE, unsup_feat, enabled_feat)) missing_feat_read = B_TRUE; if (spa_writeable(spa) || spa->spa_load_state == SPA_LOAD_TRYIMPORT) { if (!spa_features_check(spa, B_TRUE, unsup_feat, enabled_feat)) { *missing_feat_writep = B_TRUE; } } fnvlist_add_nvlist(spa->spa_load_info, ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); if (!nvlist_empty(unsup_feat)) { fnvlist_add_nvlist(spa->spa_load_info, ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); } fnvlist_free(enabled_feat); fnvlist_free(unsup_feat); if (!missing_feat_read) { fnvlist_add_boolean(spa->spa_load_info, ZPOOL_CONFIG_CAN_RDONLY); } /* * If the state is SPA_LOAD_TRYIMPORT, our objective is * twofold: to determine whether the pool is available for * import in read-write mode and (if it is not) whether the * pool is available for import in read-only mode. If the pool * is available for import in read-write mode, it is displayed * as available in userland; if it is not available for import * in read-only mode, it is displayed as unavailable in * userland. If the pool is available for import in read-only * mode but not read-write mode, it is displayed as unavailable * in userland with a special note that the pool is actually * available for open in read-only mode. * * As a result, if the state is SPA_LOAD_TRYIMPORT and we are * missing a feature for write, we must first determine whether * the pool can be opened read-only before returning to * userland in order to know whether to display the * abovementioned note. */ if (missing_feat_read || (*missing_feat_writep && spa_writeable(spa))) { spa_load_failed(spa, "pool uses unsupported features"); return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); } /* * Load refcounts for ZFS features from disk into an in-memory * cache during SPA initialization. */ for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { uint64_t refcount; error = feature_get_refcount_from_disk(spa, &spa_feature_table[i], &refcount); if (error == 0) { spa->spa_feat_refcount_cache[i] = refcount; } else if (error == ENOTSUP) { spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; } else { spa_load_failed(spa, "error getting refcount " "for feature %s [error=%d]", spa_feature_table[i].fi_guid, error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } } } if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } return (0); } static int spa_ld_load_special_directories(spa_t *spa) { int error = 0; vdev_t *rvd = spa->spa_root_vdev; spa->spa_is_initializing = B_TRUE; error = dsl_pool_open(spa->spa_dsl_pool); spa->spa_is_initializing = B_FALSE; if (error != 0) { spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } return (0); } static int spa_ld_get_props(spa_t *spa) { int error = 0; uint64_t obj; vdev_t *rvd = spa->spa_root_vdev; /* Grab the secret checksum salt from the MOS. */ error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, sizeof (spa->spa_cksum_salt.zcs_bytes), spa->spa_cksum_salt.zcs_bytes); if (error == ENOENT) { /* Generate a new salt for subsequent use */ (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, sizeof (spa->spa_cksum_salt.zcs_bytes)); } else if (error != 0) { spa_load_failed(spa, "unable to retrieve checksum salt from " "MOS [error=%d]", error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); if (error != 0) { spa_load_failed(spa, "error opening deferred-frees bpobj " "[error=%d]", error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } /* * Load the bit that tells us to use the new accounting function * (raid-z deflation). If we have an older pool, this will not * be present. */ error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, &spa->spa_creation_version, B_FALSE); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); /* * Load the persistent error log. If we have an older pool, this will * not be present. */ error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last, B_FALSE); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, &spa->spa_errlog_scrub, B_FALSE); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); /* * Load the history object. If we have an older pool, this * will not be present. */ error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); /* * Load the per-vdev ZAP map. If we have an older pool, this will not * be present; in this case, defer its creation to a later time to * avoid dirtying the MOS this early / out of sync context. See * spa_sync_config_object. */ /* The sentinel is only available in the MOS config. */ nvlist_t *mos_config; if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { spa_load_failed(spa, "unable to retrieve MOS config"); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, &spa->spa_all_vdev_zaps, B_FALSE); if (error == ENOENT) { VERIFY(!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); spa->spa_avz_action = AVZ_ACTION_INITIALIZE; ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); } else if (error != 0) { return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { /* * An older version of ZFS overwrote the sentinel value, so * we have orphaned per-vdev ZAPs in the MOS. Defer their * destruction to later; see spa_sync_config_object. */ spa->spa_avz_action = AVZ_ACTION_DESTROY; /* * We're assuming that no vdevs have had their ZAPs created * before this. Better be sure of it. */ ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); } nvlist_free(mos_config); spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object, B_FALSE); if (error && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (error == 0) { uint64_t autoreplace; spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, &spa->spa_dedup_ditto); spa->spa_autoreplace = (autoreplace != 0); } /* * If we are importing a pool with missing top-level vdevs, * we enforce that the pool doesn't panic or get suspended on * error since the likelihood of missing data is extremely high. */ if (spa->spa_missing_tvds > 0 && spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { spa_load_note(spa, "forcing failmode to 'continue' " "as some top level vdevs are missing"); spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE; } return (0); } static int spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type) { int error = 0; vdev_t *rvd = spa->spa_root_vdev; /* * If we're assembling the pool from the split-off vdevs of * an existing pool, we don't want to attach the spares & cache * devices. */ /* * Load any hot spares for this pool. */ error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object, B_FALSE); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); if (load_nvlist(spa, spa->spa_spares.sav_object, &spa->spa_spares.sav_config) != 0) { spa_load_failed(spa, "error loading spares nvlist"); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_spares(spa); spa_config_exit(spa, SCL_ALL, FTAG); } else if (error == 0) { spa->spa_spares.sav_sync = B_TRUE; } /* * Load any level 2 ARC devices for this pool. */ error = spa_dir_prop(spa, DMU_POOL_L2CACHE, &spa->spa_l2cache.sav_object, B_FALSE); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); if (load_nvlist(spa, spa->spa_l2cache.sav_object, &spa->spa_l2cache.sav_config) != 0) { spa_load_failed(spa, "error loading l2cache nvlist"); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_l2cache(spa); spa_config_exit(spa, SCL_ALL, FTAG); } else if (error == 0) { spa->spa_l2cache.sav_sync = B_TRUE; } return (0); } static int spa_ld_load_vdev_metadata(spa_t *spa) { int error = 0; vdev_t *rvd = spa->spa_root_vdev; /* * If the 'autoreplace' property is set, then post a resource notifying * the ZFS DE that it should not issue any faults for unopenable * devices. We also iterate over the vdevs, and post a sysevent for any * unopenable vdevs so that the normal autoreplace handler can take * over. */ if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { spa_check_removed(spa->spa_root_vdev); /* * For the import case, this is done in spa_import(), because * at this point we're using the spare definitions from * the MOS config, not necessarily from the userland config. */ if (spa->spa_load_state != SPA_LOAD_IMPORT) { spa_aux_check_removed(&spa->spa_spares); spa_aux_check_removed(&spa->spa_l2cache); } } /* * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc. */ error = vdev_load(rvd); if (error != 0) { spa_load_failed(spa, "vdev_load failed [error=%d]", error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); } /* * Propagate the leaf DTLs we just loaded all the way up the vdev tree. */ spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); vdev_dtl_reassess(rvd, 0, 0, B_FALSE); spa_config_exit(spa, SCL_ALL, FTAG); return (0); } static int spa_ld_load_dedup_tables(spa_t *spa) { int error = 0; vdev_t *rvd = spa->spa_root_vdev; error = ddt_load(spa); if (error != 0) { spa_load_failed(spa, "ddt_load failed [error=%d]", error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } return (0); } static int spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport) { vdev_t *rvd = spa->spa_root_vdev; if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) { boolean_t missing = spa_check_logs(spa); if (missing) { if (spa->spa_missing_tvds != 0) { spa_load_note(spa, "spa_check_logs failed " "so dropping the logs"); } else { *ereport = FM_EREPORT_ZFS_LOG_REPLAY; spa_load_failed(spa, "spa_check_logs failed"); return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); } } } return (0); } static int spa_ld_verify_pool_data(spa_t *spa) { int error = 0; vdev_t *rvd = spa->spa_root_vdev; /* * We've successfully opened the pool, verify that we're ready * to start pushing transactions. */ if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) { error = spa_load_verify(spa); if (error != 0) { spa_load_failed(spa, "spa_load_verify failed " "[error=%d]", error); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); } } return (0); } static void spa_ld_claim_log_blocks(spa_t *spa) { dmu_tx_t *tx; dsl_pool_t *dp = spa_get_dsl(spa); /* * Claim log blocks that haven't been committed yet. * This must all happen in a single txg. * Note: spa_claim_max_txg is updated by spa_claim_notify(), * invoked from zil_claim_log_block()'s i/o done callback. * Price of rollback is that we abandon the log. */ spa->spa_claiming = B_TRUE; tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, zil_claim, tx, DS_FIND_CHILDREN); dmu_tx_commit(tx); spa->spa_claiming = B_FALSE; spa_set_log_state(spa, SPA_LOG_GOOD); } static void spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg, boolean_t update_config_cache) { vdev_t *rvd = spa->spa_root_vdev; int need_update = B_FALSE; /* * If the config cache is stale, or we have uninitialized * metaslabs (see spa_vdev_add()), then update the config. * * If this is a verbatim import, trust the current * in-core spa_config and update the disk labels. */ if (update_config_cache || config_cache_txg != spa->spa_config_txg || spa->spa_load_state == SPA_LOAD_IMPORT || spa->spa_load_state == SPA_LOAD_RECOVER || (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) need_update = B_TRUE; for (int c = 0; c < rvd->vdev_children; c++) if (rvd->vdev_child[c]->vdev_ms_array == 0) need_update = B_TRUE; /* * Update the config cache asychronously in case we're the * root pool, in which case the config cache isn't writable yet. */ if (need_update) spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); } static void spa_ld_prepare_for_reload(spa_t *spa) { int mode = spa->spa_mode; int async_suspended = spa->spa_async_suspended; spa_unload(spa); spa_deactivate(spa); spa_activate(spa, mode); /* * We save the value of spa_async_suspended as it gets reset to 0 by * spa_unload(). We want to restore it back to the original value before * returning as we might be calling spa_async_resume() later. */ spa->spa_async_suspended = async_suspended; } static int spa_ld_read_checkpoint_txg(spa_t *spa) { uberblock_t checkpoint; int error = 0; ASSERT0(spa->spa_checkpoint_txg); ASSERT(MUTEX_HELD(&spa_namespace_lock)); error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); if (error == ENOENT) return (0); if (error != 0) return (error); ASSERT3U(checkpoint.ub_txg, !=, 0); ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0); ASSERT3U(checkpoint.ub_timestamp, !=, 0); spa->spa_checkpoint_txg = checkpoint.ub_txg; spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp; return (0); } static int spa_ld_mos_init(spa_t *spa, spa_import_type_t type) { int error = 0; ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); /* * Never trust the config that is provided unless we are assembling * a pool following a split. * This means don't trust blkptrs and the vdev tree in general. This * also effectively puts the spa in read-only mode since * spa_writeable() checks for spa_trust_config to be true. * We will later load a trusted config from the MOS. */ if (type != SPA_IMPORT_ASSEMBLE) spa->spa_trust_config = B_FALSE; /* * Parse the config provided to create a vdev tree. */ error = spa_ld_parse_config(spa, type); if (error != 0) return (error); /* * Now that we have the vdev tree, try to open each vdev. This involves * opening the underlying physical device, retrieving its geometry and * probing the vdev with a dummy I/O. The state of each vdev will be set * based on the success of those operations. After this we'll be ready * to read from the vdevs. */ error = spa_ld_open_vdevs(spa); if (error != 0) return (error); /* * Read the label of each vdev and make sure that the GUIDs stored * there match the GUIDs in the config provided. * If we're assembling a new pool that's been split off from an * existing pool, the labels haven't yet been updated so we skip * validation for now. */ if (type != SPA_IMPORT_ASSEMBLE) { error = spa_ld_validate_vdevs(spa); if (error != 0) return (error); } /* * Read all vdev labels to find the best uberblock (i.e. latest, * unless spa_load_max_txg is set) and store it in spa_uberblock. We * get the list of features required to read blkptrs in the MOS from * the vdev label with the best uberblock and verify that our version * of zfs supports them all. */ error = spa_ld_select_uberblock(spa, type); if (error != 0) return (error); /* * Pass that uberblock to the dsl_pool layer which will open the root * blkptr. This blkptr points to the latest version of the MOS and will * allow us to read its contents. */ error = spa_ld_open_rootbp(spa); if (error != 0) return (error); return (0); } static int spa_ld_checkpoint_rewind(spa_t *spa) { uberblock_t checkpoint; int error = 0; ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); if (error != 0) { spa_load_failed(spa, "unable to retrieve checkpointed " "uberblock from the MOS config [error=%d]", error); if (error == ENOENT) error = ZFS_ERR_NO_CHECKPOINT; return (error); } ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg); ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg); /* * We need to update the txg and timestamp of the checkpointed * uberblock to be higher than the latest one. This ensures that * the checkpointed uberblock is selected if we were to close and * reopen the pool right after we've written it in the vdev labels. * (also see block comment in vdev_uberblock_compare) */ checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1; checkpoint.ub_timestamp = gethrestime_sec(); /* * Set current uberblock to be the checkpointed uberblock. */ spa->spa_uberblock = checkpoint; /* * If we are doing a normal rewind, then the pool is open for * writing and we sync the "updated" checkpointed uberblock to * disk. Once this is done, we've basically rewound the whole * pool and there is no way back. * * There are cases when we don't want to attempt and sync the * checkpointed uberblock to disk because we are opening a * pool as read-only. Specifically, verifying the checkpointed * state with zdb, and importing the checkpointed state to get * a "preview" of its content. */ if (spa_writeable(spa)) { vdev_t *rvd = spa->spa_root_vdev; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; int svdcount = 0; int children = rvd->vdev_children; int c0 = spa_get_random(children); for (int c = 0; c < children; c++) { vdev_t *vd = rvd->vdev_child[(c0 + c) % children]; /* Stop when revisiting the first vdev */ if (c > 0 && svd[0] == vd) break; if (vd->vdev_ms_array == 0 || vd->vdev_islog || !vdev_is_concrete(vd)) continue; svd[svdcount++] = vd; if (svdcount == SPA_SYNC_MIN_VDEVS) break; } error = vdev_config_sync(svd, svdcount, spa->spa_first_txg); if (error == 0) spa->spa_last_synced_guid = rvd->vdev_guid; spa_config_exit(spa, SCL_ALL, FTAG); if (error != 0) { spa_load_failed(spa, "failed to write checkpointed " "uberblock to the vdev labels [error=%d]", error); return (error); } } return (0); } static int spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type, boolean_t *update_config_cache) { int error; /* * Parse the config for pool, open and validate vdevs, * select an uberblock, and use that uberblock to open * the MOS. */ error = spa_ld_mos_init(spa, type); if (error != 0) return (error); /* * Retrieve the trusted config stored in the MOS and use it to create * a new, exact version of the vdev tree, then reopen all vdevs. */ error = spa_ld_trusted_config(spa, type, B_FALSE); if (error == EAGAIN) { if (update_config_cache != NULL) *update_config_cache = B_TRUE; /* * Redo the loading process with the trusted config if it is * too different from the untrusted config. */ spa_ld_prepare_for_reload(spa); spa_load_note(spa, "RELOADING"); error = spa_ld_mos_init(spa, type); if (error != 0) return (error); error = spa_ld_trusted_config(spa, type, B_TRUE); if (error != 0) return (error); } else if (error != 0) { return (error); } return (0); } /* * Load an existing storage pool, using the config provided. This config * describes which vdevs are part of the pool and is later validated against * partial configs present in each vdev's label and an entire copy of the * config stored in the MOS. */ static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport) { int error = 0; boolean_t missing_feat_write = B_FALSE; boolean_t checkpoint_rewind = (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); boolean_t update_config_cache = B_FALSE; ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); spa_load_note(spa, "LOADING"); error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache); if (error != 0) return (error); /* * If we are rewinding to the checkpoint then we need to repeat * everything we've done so far in this function but this time * selecting the checkpointed uberblock and using that to open * the MOS. */ if (checkpoint_rewind) { /* * If we are rewinding to the checkpoint update config cache * anyway. */ update_config_cache = B_TRUE; /* * Extract the checkpointed uberblock from the current MOS * and use this as the pool's uberblock from now on. If the * pool is imported as writeable we also write the checkpoint * uberblock to the labels, making the rewind permanent. */ error = spa_ld_checkpoint_rewind(spa); if (error != 0) return (error); /* * Redo the loading process process again with the * checkpointed uberblock. */ spa_ld_prepare_for_reload(spa); spa_load_note(spa, "LOADING checkpointed uberblock"); error = spa_ld_mos_with_trusted_config(spa, type, NULL); if (error != 0) return (error); } /* * Retrieve the checkpoint txg if the pool has a checkpoint. */ error = spa_ld_read_checkpoint_txg(spa); if (error != 0) return (error); /* * Retrieve the mapping of indirect vdevs. Those vdevs were removed * from the pool and their contents were re-mapped to other vdevs. Note * that everything that we read before this step must have been * rewritten on concrete vdevs after the last device removal was * initiated. Otherwise we could be reading from indirect vdevs before * we have loaded their mappings. */ error = spa_ld_open_indirect_vdev_metadata(spa); if (error != 0) return (error); /* * Retrieve the full list of active features from the MOS and check if * they are all supported. */ error = spa_ld_check_features(spa, &missing_feat_write); if (error != 0) return (error); /* * Load several special directories from the MOS needed by the dsl_pool * layer. */ error = spa_ld_load_special_directories(spa); if (error != 0) return (error); /* * Retrieve pool properties from the MOS. */ error = spa_ld_get_props(spa); if (error != 0) return (error); /* * Retrieve the list of auxiliary devices - cache devices and spares - * and open them. */ error = spa_ld_open_aux_vdevs(spa, type); if (error != 0) return (error); /* * Load the metadata for all vdevs. Also check if unopenable devices * should be autoreplaced. */ error = spa_ld_load_vdev_metadata(spa); if (error != 0) return (error); error = spa_ld_load_dedup_tables(spa); if (error != 0) return (error); /* * Verify the logs now to make sure we don't have any unexpected errors * when we claim log blocks later. */ error = spa_ld_verify_logs(spa, type, ereport); if (error != 0) return (error); if (missing_feat_write) { ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT); /* * At this point, we know that we can open the pool in * read-only mode but not read-write mode. We now have enough * information and can return to userland. */ return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); } /* * Traverse the last txgs to make sure the pool was left off in a safe * state. When performing an extreme rewind, we verify the whole pool, * which can take a very long time. */ error = spa_ld_verify_pool_data(spa); if (error != 0) return (error); /* * Calculate the deflated space for the pool. This must be done before * we write anything to the pool because we'd need to update the space * accounting using the deflated sizes. */ spa_update_dspace(spa); /* * We have now retrieved all the information we needed to open the * pool. If we are importing the pool in read-write mode, a few * additional steps must be performed to finish the import. */ if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER || spa->spa_load_max_txg == UINT64_MAX)) { uint64_t config_cache_txg = spa->spa_config_txg; ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT); /* * In case of a checkpoint rewind, log the original txg * of the checkpointed uberblock. */ if (checkpoint_rewind) { spa_history_log_internal(spa, "checkpoint rewind", NULL, "rewound state to txg=%llu", (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg); } /* * Traverse the ZIL and claim all blocks. */ spa_ld_claim_log_blocks(spa); /* * Kick-off the syncing thread. */ spa->spa_sync_on = B_TRUE; txg_sync_start(spa->spa_dsl_pool); /* * Wait for all claims to sync. We sync up to the highest * claimed log block birth time so that claimed log blocks * don't appear to be from the future. spa_claim_max_txg * will have been set for us by ZIL traversal operations * performed above. */ txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); /* * Check if we need to request an update of the config. On the * next sync, we would update the config stored in vdev labels * and the cachefile (by default /etc/zfs/zpool.cache). */ spa_ld_check_for_config_update(spa, config_cache_txg, update_config_cache); /* * Check all DTLs to see if anything needs resilvering. */ if (!dsl_scan_resilvering(spa->spa_dsl_pool) && vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) spa_async_request(spa, SPA_ASYNC_RESILVER); /* * Log the fact that we booted up (so that we can detect if * we rebooted in the middle of an operation). */ spa_history_log_version(spa, "open"); /* * Delete any inconsistent datasets. */ (void) dmu_objset_find(spa_name(spa), dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); /* * Clean up any stale temporary dataset userrefs. */ dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); spa_restart_removal(spa); spa_spawn_aux_threads(spa); } spa_load_note(spa, "LOADED"); return (0); } static int spa_load_retry(spa_t *spa, spa_load_state_t state) { int mode = spa->spa_mode; spa_unload(spa); spa_deactivate(spa); spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; spa_activate(spa, mode); spa_async_suspend(spa); spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu", (u_longlong_t)spa->spa_load_max_txg); return (spa_load(spa, state, SPA_IMPORT_EXISTING)); } /* * If spa_load() fails this function will try loading prior txg's. If * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this * function will not rewind the pool and will return the same error as * spa_load(). */ static int spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request, int rewind_flags) { nvlist_t *loadinfo = NULL; nvlist_t *config = NULL; int load_error, rewind_error; uint64_t safe_rewind_txg; uint64_t min_txg; if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { spa->spa_load_max_txg = spa->spa_load_txg; spa_set_log_state(spa, SPA_LOG_CLEAR); } else { spa->spa_load_max_txg = max_request; if (max_request != UINT64_MAX) spa->spa_extreme_rewind = B_TRUE; } load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING); if (load_error == 0) return (0); if (load_error == ZFS_ERR_NO_CHECKPOINT) { /* * When attempting checkpoint-rewind on a pool with no * checkpoint, we should not attempt to load uberblocks * from previous txgs when spa_load fails. */ ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); return (load_error); } if (spa->spa_root_vdev != NULL) config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; if (rewind_flags & ZPOOL_NEVER_REWIND) { nvlist_free(config); return (load_error); } if (state == SPA_LOAD_RECOVER) { /* Price of rolling back is discarding txgs, including log */ spa_set_log_state(spa, SPA_LOG_CLEAR); } else { /* * If we aren't rolling back save the load info from our first * import attempt so that we can restore it after attempting * to rewind. */ loadinfo = spa->spa_load_info; spa->spa_load_info = fnvlist_alloc(); } spa->spa_load_max_txg = spa->spa_last_ubsync_txg; safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? TXG_INITIAL : safe_rewind_txg; /* * Continue as long as we're finding errors, we're still within * the acceptable rewind range, and we're still finding uberblocks */ while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { if (spa->spa_load_max_txg < safe_rewind_txg) spa->spa_extreme_rewind = B_TRUE; rewind_error = spa_load_retry(spa, state); } spa->spa_extreme_rewind = B_FALSE; spa->spa_load_max_txg = UINT64_MAX; if (config && (rewind_error || state != SPA_LOAD_RECOVER)) spa_config_set(spa, config); else nvlist_free(config); if (state == SPA_LOAD_RECOVER) { ASSERT3P(loadinfo, ==, NULL); return (rewind_error); } else { /* Store the rewind info as part of the initial load info */ fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, spa->spa_load_info); /* Restore the initial load info */ fnvlist_free(spa->spa_load_info); spa->spa_load_info = loadinfo; return (load_error); } } /* * Pool Open/Import * * The import case is identical to an open except that the configuration is sent * down from userland, instead of grabbed from the configuration cache. For the * case of an open, the pool configuration will exist in the * POOL_STATE_UNINITIALIZED state. * * The stats information (gen/count/ustats) is used to gather vdev statistics at * the same time open the pool, without having to keep around the spa_t in some * ambiguous state. */ static int spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, nvlist_t **config) { spa_t *spa; spa_load_state_t state = SPA_LOAD_OPEN; int error; int locked = B_FALSE; int firstopen = B_FALSE; *spapp = NULL; /* * As disgusting as this is, we need to support recursive calls to this * function because dsl_dir_open() is called during spa_load(), and ends * up calling spa_open() again. The real fix is to figure out how to * avoid dsl_dir_open() calling this in the first place. */ if (mutex_owner(&spa_namespace_lock) != curthread) { mutex_enter(&spa_namespace_lock); locked = B_TRUE; } if ((spa = spa_lookup(pool)) == NULL) { if (locked) mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENOENT)); } if (spa->spa_state == POOL_STATE_UNINITIALIZED) { zpool_load_policy_t policy; firstopen = B_TRUE; zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config, &policy); if (policy.zlp_rewind & ZPOOL_DO_REWIND) state = SPA_LOAD_RECOVER; spa_activate(spa, spa_mode_global); if (state != SPA_LOAD_RECOVER) spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; zfs_dbgmsg("spa_open_common: opening %s", pool); error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind); if (error == EBADF) { /* * If vdev_validate() returns failure (indicated by * EBADF), it indicates that one of the vdevs indicates * that the pool has been exported or destroyed. If * this is the case, the config cache is out of sync and * we should remove the pool from the namespace. */ spa_unload(spa); spa_deactivate(spa); spa_write_cachefile(spa, B_TRUE, B_TRUE); spa_remove(spa); if (locked) mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENOENT)); } if (error) { /* * We can't open the pool, but we still have useful * information: the state of each vdev after the * attempted vdev_open(). Return this to the user. */ if (config != NULL && spa->spa_config) { VERIFY(nvlist_dup(spa->spa_config, config, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info) == 0); } spa_unload(spa); spa_deactivate(spa); spa->spa_last_open_failed = error; if (locked) mutex_exit(&spa_namespace_lock); *spapp = NULL; return (error); } } spa_open_ref(spa, tag); if (config != NULL) *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); /* * If we've recovered the pool, pass back any information we * gathered while doing the load. */ if (state == SPA_LOAD_RECOVER) { VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info) == 0); } if (locked) { spa->spa_last_open_failed = 0; spa->spa_last_ubsync_txg = 0; spa->spa_load_txg = 0; mutex_exit(&spa_namespace_lock); #ifdef __FreeBSD__ #ifdef _KERNEL if (firstopen) zvol_create_minors(spa->spa_name); #endif #endif } *spapp = spa; return (0); } int spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, nvlist_t **config) { return (spa_open_common(name, spapp, tag, policy, config)); } int spa_open(const char *name, spa_t **spapp, void *tag) { return (spa_open_common(name, spapp, tag, NULL, NULL)); } /* * Lookup the given spa_t, incrementing the inject count in the process, * preventing it from being exported or destroyed. */ spa_t * spa_inject_addref(char *name) { spa_t *spa; mutex_enter(&spa_namespace_lock); if ((spa = spa_lookup(name)) == NULL) { mutex_exit(&spa_namespace_lock); return (NULL); } spa->spa_inject_ref++; mutex_exit(&spa_namespace_lock); return (spa); } void spa_inject_delref(spa_t *spa) { mutex_enter(&spa_namespace_lock); spa->spa_inject_ref--; mutex_exit(&spa_namespace_lock); } /* * Add spares device information to the nvlist. */ static void spa_add_spares(spa_t *spa, nvlist_t *config) { nvlist_t **spares; uint_t i, nspares; nvlist_t *nvroot; uint64_t guid; vdev_stat_t *vs; uint_t vsc; uint64_t pool; ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); if (spa->spa_spares.sav_count == 0) return; VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); if (nspares != 0) { VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, spares, nspares) == 0); VERIFY(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); /* * Go through and find any spares which have since been * repurposed as an active spare. If this is the case, update * their status appropriately. */ for (i = 0; i < nspares; i++) { VERIFY(nvlist_lookup_uint64(spares[i], ZPOOL_CONFIG_GUID, &guid) == 0); if (spa_spare_exists(guid, &pool, NULL) && pool != 0ULL) { VERIFY(nvlist_lookup_uint64_array( spares[i], ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) == 0); vs->vs_state = VDEV_STATE_CANT_OPEN; vs->vs_aux = VDEV_AUX_SPARED; } } } } /* * Add l2cache device information to the nvlist, including vdev stats. */ static void spa_add_l2cache(spa_t *spa, nvlist_t *config) { nvlist_t **l2cache; uint_t i, j, nl2cache; nvlist_t *nvroot; uint64_t guid; vdev_t *vd; vdev_stat_t *vs; uint_t vsc; ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); if (spa->spa_l2cache.sav_count == 0) return; VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); if (nl2cache != 0) { VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); VERIFY(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); /* * Update level 2 cache device stats. */ for (i = 0; i < nl2cache; i++) { VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, &guid) == 0); vd = NULL; for (j = 0; j < spa->spa_l2cache.sav_count; j++) { if (guid == spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { vd = spa->spa_l2cache.sav_vdevs[j]; break; } } ASSERT(vd != NULL); VERIFY(nvlist_lookup_uint64_array(l2cache[i], ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) == 0); vdev_get_stats(vd, vs); } } } static void spa_add_feature_stats(spa_t *spa, nvlist_t *config) { nvlist_t *features; zap_cursor_t zc; zap_attribute_t za; ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); /* We may be unable to read features if pool is suspended. */ if (spa_suspended(spa)) goto out; if (spa->spa_feat_for_read_obj != 0) { for (zap_cursor_init(&zc, spa->spa_meta_objset, spa->spa_feat_for_read_obj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { ASSERT(za.za_integer_length == sizeof (uint64_t) && za.za_num_integers == 1); VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, za.za_first_integer)); } zap_cursor_fini(&zc); } if (spa->spa_feat_for_write_obj != 0) { for (zap_cursor_init(&zc, spa->spa_meta_objset, spa->spa_feat_for_write_obj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { ASSERT(za.za_integer_length == sizeof (uint64_t) && za.za_num_integers == 1); VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, za.za_first_integer)); } zap_cursor_fini(&zc); } out: VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, features) == 0); nvlist_free(features); } int spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) { int error; spa_t *spa; *config = NULL; error = spa_open_common(name, &spa, FTAG, NULL, config); if (spa != NULL) { /* * This still leaves a window of inconsistency where the spares * or l2cache devices could change and the config would be * self-inconsistent. */ spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); if (*config != NULL) { uint64_t loadtimes[2]; loadtimes[0] = spa->spa_loaded_ts.tv_sec; loadtimes[1] = spa->spa_loaded_ts.tv_nsec; VERIFY(nvlist_add_uint64_array(*config, ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT, spa_get_errlog_size(spa)) == 0); if (spa_suspended(spa)) VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_SUSPENDED, spa->spa_failmode) == 0); spa_add_spares(spa, *config); spa_add_l2cache(spa, *config); spa_add_feature_stats(spa, *config); } } /* * We want to get the alternate root even for faulted pools, so we cheat * and call spa_lookup() directly. */ if (altroot) { if (spa == NULL) { mutex_enter(&spa_namespace_lock); spa = spa_lookup(name); if (spa) spa_altroot(spa, altroot, buflen); else altroot[0] = '\0'; spa = NULL; mutex_exit(&spa_namespace_lock); } else { spa_altroot(spa, altroot, buflen); } } if (spa != NULL) { spa_config_exit(spa, SCL_CONFIG, FTAG); spa_close(spa, FTAG); } return (error); } /* * Validate that the auxiliary device array is well formed. We must have an * array of nvlists, each which describes a valid leaf vdev. If this is an * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be * specified, as long as they are well-formed. */ static int spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, spa_aux_vdev_t *sav, const char *config, uint64_t version, vdev_labeltype_t label) { nvlist_t **dev; uint_t i, ndev; vdev_t *vd; int error; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); /* * It's acceptable to have no devs specified. */ if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) return (0); if (ndev == 0) return (SET_ERROR(EINVAL)); /* * Make sure the pool is formatted with a version that supports this * device type. */ if (spa_version(spa) < version) return (SET_ERROR(ENOTSUP)); /* * Set the pending device list so we correctly handle device in-use * checking. */ sav->sav_pending = dev; sav->sav_npending = ndev; for (i = 0; i < ndev; i++) { if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, mode)) != 0) goto out; if (!vd->vdev_ops->vdev_op_leaf) { vdev_free(vd); error = SET_ERROR(EINVAL); goto out; } /* * The L2ARC currently only supports disk devices in * kernel context. For user-level testing, we allow it. */ #ifdef _KERNEL if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { error = SET_ERROR(ENOTBLK); vdev_free(vd); goto out; } #endif vd->vdev_top = vd; if ((error = vdev_open(vd)) == 0 && (error = vdev_label_init(vd, crtxg, label)) == 0) { VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); } vdev_free(vd); if (error && (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) goto out; else error = 0; } out: sav->sav_pending = NULL; sav->sav_npending = 0; return (error); } static int spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) { int error; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, VDEV_LABEL_SPARE)) != 0) { return (error); } return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, VDEV_LABEL_L2CACHE)); } static void spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, const char *config) { int i; if (sav->sav_config != NULL) { nvlist_t **olddevs; uint_t oldndevs; nvlist_t **newdevs; /* * Generate new dev list by concatentating with the * current dev list. */ VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, &olddevs, &oldndevs) == 0); newdevs = kmem_alloc(sizeof (void *) * (ndevs + oldndevs), KM_SLEEP); for (i = 0; i < oldndevs; i++) VERIFY(nvlist_dup(olddevs[i], &newdevs[i], KM_SLEEP) == 0); for (i = 0; i < ndevs; i++) VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], KM_SLEEP) == 0); VERIFY(nvlist_remove(sav->sav_config, config, DATA_TYPE_NVLIST_ARRAY) == 0); VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, newdevs, ndevs + oldndevs) == 0); for (i = 0; i < oldndevs + ndevs; i++) nvlist_free(newdevs[i]); kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); } else { /* * Generate a new dev list. */ VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, devs, ndevs) == 0); } } /* * Stop and drop level 2 ARC devices */ void spa_l2cache_drop(spa_t *spa) { vdev_t *vd; int i; spa_aux_vdev_t *sav = &spa->spa_l2cache; for (i = 0; i < sav->sav_count; i++) { uint64_t pool; vd = sav->sav_vdevs[i]; ASSERT(vd != NULL); if (spa_l2cache_exists(vd->vdev_guid, &pool) && pool != 0ULL && l2arc_vdev_present(vd)) l2arc_remove_vdev(vd); } } /* * Pool Creation */ int spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, nvlist_t *zplprops) { spa_t *spa; char *altroot = NULL; vdev_t *rvd; dsl_pool_t *dp; dmu_tx_t *tx; int error = 0; uint64_t txg = TXG_INITIAL; nvlist_t **spares, **l2cache; uint_t nspares, nl2cache; uint64_t version, obj; boolean_t has_features; char *poolname; nvlist_t *nvl; if (nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0) poolname = (char *)pool; /* * If this pool already exists, return failure. */ mutex_enter(&spa_namespace_lock); if (spa_lookup(poolname) != NULL) { mutex_exit(&spa_namespace_lock); return (SET_ERROR(EEXIST)); } /* * Allocate a new spa_t structure. */ nvl = fnvlist_alloc(); fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); (void) nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); spa = spa_add(poolname, nvl, altroot); fnvlist_free(nvl); spa_activate(spa, spa_mode_global); if (props && (error = spa_prop_validate(spa, props))) { spa_deactivate(spa); spa_remove(spa); mutex_exit(&spa_namespace_lock); return (error); } /* * Temporary pool names should never be written to disk. */ if (poolname != pool) spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; has_features = B_FALSE; for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); elem != NULL; elem = nvlist_next_nvpair(props, elem)) { if (zpool_prop_feature(nvpair_name(elem))) has_features = B_TRUE; } if (has_features || nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { version = SPA_VERSION; } ASSERT(SPA_VERSION_IS_SUPPORTED(version)); spa->spa_first_txg = txg; spa->spa_uberblock.ub_txg = txg - 1; spa->spa_uberblock.ub_version = version; spa->spa_ubsync = spa->spa_uberblock; spa->spa_load_state = SPA_LOAD_CREATE; spa->spa_removing_phys.sr_state = DSS_NONE; spa->spa_removing_phys.sr_removing_vdev = -1; spa->spa_removing_phys.sr_prev_indirect_vdev = -1; /* * Create "The Godfather" zio to hold all async IOs */ spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), KM_SLEEP); for (int i = 0; i < max_ncpus; i++) { spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); } /* * Create the root vdev. */ spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); ASSERT(error != 0 || rvd != NULL); ASSERT(error != 0 || spa->spa_root_vdev == rvd); if (error == 0 && !zfs_allocatable_devs(nvroot)) error = SET_ERROR(EINVAL); if (error == 0 && (error = vdev_create(rvd, txg, B_FALSE)) == 0 && (error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) == 0) { for (int c = 0; c < rvd->vdev_children; c++) { vdev_ashift_optimize(rvd->vdev_child[c]); vdev_metaslab_set_size(rvd->vdev_child[c]); vdev_expand(rvd->vdev_child[c], txg); } } spa_config_exit(spa, SCL_ALL, FTAG); if (error != 0) { spa_unload(spa); spa_deactivate(spa); spa_remove(spa); mutex_exit(&spa_namespace_lock); return (error); } /* * Get the list of spares, if specified. */ if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) { VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, spares, nspares) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_spares(spa); spa_config_exit(spa, SCL_ALL, FTAG); spa->spa_spares.sav_sync = B_TRUE; } /* * Get the list of level 2 cache devices, if specified. */ if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0) { VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_l2cache(spa); spa_config_exit(spa, SCL_ALL, FTAG); spa->spa_l2cache.sav_sync = B_TRUE; } spa->spa_is_initializing = B_TRUE; spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); spa->spa_meta_objset = dp->dp_meta_objset; spa->spa_is_initializing = B_FALSE; /* * Create DDTs (dedup tables). */ ddt_create(spa); spa_update_dspace(spa); tx = dmu_tx_create_assigned(dp, txg); /* * Create the pool config object. */ spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); if (zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { cmn_err(CE_PANIC, "failed to add pool config"); } if (spa_version(spa) >= SPA_VERSION_FEATURES) spa_feature_create_zap_objects(spa, tx); if (zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, sizeof (uint64_t), 1, &version, tx) != 0) { cmn_err(CE_PANIC, "failed to add pool version"); } /* Newly created pools with the right version are always deflated. */ if (version >= SPA_VERSION_RAIDZ_DEFLATE) { spa->spa_deflate = TRUE; if (zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { cmn_err(CE_PANIC, "failed to add deflate"); } } /* * Create the deferred-free bpobj. Turn off compression * because sync-to-convergence takes longer if the blocksize * keeps changing. */ obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); dmu_object_set_compress(spa->spa_meta_objset, obj, ZIO_COMPRESS_OFF, tx); if (zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, sizeof (uint64_t), 1, &obj, tx) != 0) { cmn_err(CE_PANIC, "failed to add bpobj"); } VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj)); /* * Create the pool's history object. */ if (version >= SPA_VERSION_ZPOOL_HISTORY) spa_history_create_obj(spa, tx); /* * Generate some random noise for salted checksums to operate on. */ (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, sizeof (spa->spa_cksum_salt.zcs_bytes)); /* * Set pool properties. */ spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); if (props != NULL) { spa_configfile_set(spa, props, B_FALSE); spa_sync_props(props, tx); } dmu_tx_commit(tx); spa->spa_sync_on = B_TRUE; txg_sync_start(spa->spa_dsl_pool); /* * We explicitly wait for the first transaction to complete so that our * bean counters are appropriately updated. */ txg_wait_synced(spa->spa_dsl_pool, txg); spa_spawn_aux_threads(spa); spa_write_cachefile(spa, B_FALSE, B_TRUE); spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); spa_history_log_version(spa, "create"); /* * Don't count references from objsets that are already closed * and are making their way through the eviction process. */ spa_evicting_os_wait(spa); spa->spa_minref = refcount_count(&spa->spa_refcount); spa->spa_load_state = SPA_LOAD_NONE; mutex_exit(&spa_namespace_lock); return (0); } #ifdef _KERNEL #ifdef illumos /* * Get the root pool information from the root disk, then import the root pool * during the system boot up time. */ extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); static nvlist_t * spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) { nvlist_t *config; nvlist_t *nvtop, *nvroot; uint64_t pgid; if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) return (NULL); /* * Add this top-level vdev to the child array. */ VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0); VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid) == 0); VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); /* * Put this pool's top-level vdevs into a root vdev. */ VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &nvtop, 1) == 0); /* * Replace the existing vdev_tree with the new root vdev in * this pool's configuration (remove the old, add the new). */ VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); nvlist_free(nvroot); return (config); } /* * Walk the vdev tree and see if we can find a device with "better" * configuration. A configuration is "better" if the label on that * device has a more recent txg. */ static void spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) { for (int c = 0; c < vd->vdev_children; c++) spa_alt_rootvdev(vd->vdev_child[c], avd, txg); if (vd->vdev_ops->vdev_op_leaf) { nvlist_t *label; uint64_t label_txg; if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, &label) != 0) return; VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, &label_txg) == 0); /* * Do we have a better boot device? */ if (label_txg > *txg) { *txg = label_txg; *avd = vd; } nvlist_free(label); } } /* * Import a root pool. * * For x86. devpath_list will consist of devid and/or physpath name of * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). * The GRUB "findroot" command will return the vdev we should boot. * * For Sparc, devpath_list consists the physpath name of the booting device * no matter the rootpool is a single device pool or a mirrored pool. * e.g. * "/pci@1f,0/ide@d/disk@0,0:a" */ int spa_import_rootpool(char *devpath, char *devid) { spa_t *spa; vdev_t *rvd, *bvd, *avd = NULL; nvlist_t *config, *nvtop; uint64_t guid, txg; char *pname; int error; /* * Read the label from the boot device and generate a configuration. */ config = spa_generate_rootconf(devpath, devid, &guid); #if defined(_OBP) && defined(_KERNEL) if (config == NULL) { if (strstr(devpath, "/iscsi/ssd") != NULL) { /* iscsi boot */ get_iscsi_bootpath_phy(devpath); config = spa_generate_rootconf(devpath, devid, &guid); } } #endif if (config == NULL) { cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", devpath); return (SET_ERROR(EIO)); } VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, &pname) == 0); VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); mutex_enter(&spa_namespace_lock); if ((spa = spa_lookup(pname)) != NULL) { /* * Remove the existing root pool from the namespace so that we * can replace it with the correct config we just read in. */ spa_remove(spa); } spa = spa_add(pname, config, NULL); spa->spa_is_root = B_TRUE; spa->spa_import_flags = ZFS_IMPORT_VERBATIM; if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &spa->spa_ubsync.ub_version) != 0) spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; /* * Build up a vdev tree based on the boot device's label config. */ VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, VDEV_ALLOC_ROOTPOOL); spa_config_exit(spa, SCL_ALL, FTAG); if (error) { mutex_exit(&spa_namespace_lock); nvlist_free(config); cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", pname); return (error); } /* * Get the boot vdev. */ if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", (u_longlong_t)guid); error = SET_ERROR(ENOENT); goto out; } /* * Determine if there is a better boot device. */ avd = bvd; spa_alt_rootvdev(rvd, &avd, &txg); if (avd != bvd) { cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " "try booting from '%s'", avd->vdev_path); error = SET_ERROR(EINVAL); goto out; } /* * If the boot device is part of a spare vdev then ensure that * we're booting off the active spare. */ if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && !bvd->vdev_isspare) { cmn_err(CE_NOTE, "The boot device is currently spared. Please " "try booting from '%s'", bvd->vdev_parent-> vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); error = SET_ERROR(EINVAL); goto out; } error = 0; out: spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); vdev_free(rvd); spa_config_exit(spa, SCL_ALL, FTAG); mutex_exit(&spa_namespace_lock); nvlist_free(config); return (error); } #else /* !illumos */ extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, uint64_t *count); static nvlist_t * spa_generate_rootconf(const char *name) { nvlist_t **configs, **tops; nvlist_t *config; nvlist_t *best_cfg, *nvtop, *nvroot; uint64_t *holes; uint64_t best_txg; uint64_t nchildren; uint64_t pgid; uint64_t count; uint64_t i; uint_t nholes; if (vdev_geom_read_pool_label(name, &configs, &count) != 0) return (NULL); ASSERT3U(count, !=, 0); best_txg = 0; for (i = 0; i < count; i++) { uint64_t txg; VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, &txg) == 0); if (txg > best_txg) { best_txg = txg; best_cfg = configs[i]; } } nchildren = 1; nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); holes = NULL; nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, &holes, &nholes); tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); for (i = 0; i < nchildren; i++) { if (i >= count) break; if (configs[i] == NULL) continue; VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0); nvlist_dup(nvtop, &tops[i], KM_SLEEP); } for (i = 0; holes != NULL && i < nholes; i++) { if (i >= nchildren) continue; if (tops[holes[i]] != NULL) continue; nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, holes[i]) == 0); VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 0) == 0); } for (i = 0; i < nchildren; i++) { if (tops[i] != NULL) continue; nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, VDEV_TYPE_MISSING) == 0); VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, i) == 0); VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 0) == 0); } /* * Create pool config based on the best vdev config. */ nvlist_dup(best_cfg, &config, KM_SLEEP); /* * Put this pool's top-level vdevs into a root vdev. */ VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid) == 0); VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, tops, nchildren) == 0); /* * Replace the existing vdev_tree with the new root vdev in * this pool's configuration (remove the old, add the new). */ VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); /* * Drop vdev config elements that should not be present at pool level. */ nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); for (i = 0; i < count; i++) nvlist_free(configs[i]); kmem_free(configs, count * sizeof(void *)); for (i = 0; i < nchildren; i++) nvlist_free(tops[i]); kmem_free(tops, nchildren * sizeof(void *)); nvlist_free(nvroot); return (config); } int spa_import_rootpool(const char *name) { spa_t *spa; vdev_t *rvd, *bvd, *avd = NULL; nvlist_t *config, *nvtop; uint64_t txg; char *pname; int error; /* * Read the label from the boot device and generate a configuration. */ config = spa_generate_rootconf(name); mutex_enter(&spa_namespace_lock); if (config != NULL) { VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, &pname) == 0 && strcmp(name, pname) == 0); VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); if ((spa = spa_lookup(pname)) != NULL) { /* * The pool could already be imported, * e.g., after reboot -r. */ if (spa->spa_state == POOL_STATE_ACTIVE) { mutex_exit(&spa_namespace_lock); nvlist_free(config); return (0); } /* * Remove the existing root pool from the namespace so * that we can replace it with the correct config * we just read in. */ spa_remove(spa); } spa = spa_add(pname, config, NULL); /* * Set spa_ubsync.ub_version as it can be used in vdev_alloc() * via spa_version(). */ if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &spa->spa_ubsync.ub_version) != 0) spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; } else if ((spa = spa_lookup(name)) == NULL) { mutex_exit(&spa_namespace_lock); nvlist_free(config); cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", name); return (EIO); } else { VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); } spa->spa_is_root = B_TRUE; spa->spa_import_flags = ZFS_IMPORT_VERBATIM; /* * Build up a vdev tree based on the boot device's label config. */ VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, VDEV_ALLOC_ROOTPOOL); spa_config_exit(spa, SCL_ALL, FTAG); if (error) { mutex_exit(&spa_namespace_lock); nvlist_free(config); cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", pname); return (error); } spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); vdev_free(rvd); spa_config_exit(spa, SCL_ALL, FTAG); mutex_exit(&spa_namespace_lock); nvlist_free(config); return (0); } #endif /* illumos */ #endif /* _KERNEL */ /* * Import a non-root pool into the system. */ int spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) { spa_t *spa; char *altroot = NULL; spa_load_state_t state = SPA_LOAD_IMPORT; zpool_load_policy_t policy; uint64_t mode = spa_mode_global; uint64_t readonly = B_FALSE; int error; nvlist_t *nvroot; nvlist_t **spares, **l2cache; uint_t nspares, nl2cache; /* * If a pool with this name exists, return failure. */ mutex_enter(&spa_namespace_lock); if (spa_lookup(pool) != NULL) { mutex_exit(&spa_namespace_lock); return (SET_ERROR(EEXIST)); } /* * Create and initialize the spa structure. */ (void) nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); (void) nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); if (readonly) mode = FREAD; spa = spa_add(pool, config, altroot); spa->spa_import_flags = flags; /* * Verbatim import - Take a pool and insert it into the namespace * as if it had been loaded at boot. */ if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { if (props != NULL) spa_configfile_set(spa, props, B_FALSE); spa_write_cachefile(spa, B_FALSE, B_TRUE); spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); zfs_dbgmsg("spa_import: verbatim import of %s", pool); mutex_exit(&spa_namespace_lock); return (0); } spa_activate(spa, mode); /* * Don't start async tasks until we know everything is healthy. */ spa_async_suspend(spa); zpool_get_load_policy(config, &policy); if (policy.zlp_rewind & ZPOOL_DO_REWIND) state = SPA_LOAD_RECOVER; spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT; if (state != SPA_LOAD_RECOVER) { spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; zfs_dbgmsg("spa_import: importing %s", pool); } else { zfs_dbgmsg("spa_import: importing %s, max_txg=%lld " "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg); } error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind); /* * Propagate anything learned while loading the pool and pass it * back to caller (i.e. rewind info, missing devices, etc). */ VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); /* * Toss any existing sparelist, as it doesn't have any validity * anymore, and conflicts with spa_has_spare(). */ if (spa->spa_spares.sav_config) { nvlist_free(spa->spa_spares.sav_config); spa->spa_spares.sav_config = NULL; spa_load_spares(spa); } if (spa->spa_l2cache.sav_config) { nvlist_free(spa->spa_l2cache.sav_config); spa->spa_l2cache.sav_config = NULL; spa_load_l2cache(spa); } VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); if (error == 0) error = spa_validate_aux(spa, nvroot, -1ULL, VDEV_ALLOC_SPARE); if (error == 0) error = spa_validate_aux(spa, nvroot, -1ULL, VDEV_ALLOC_L2CACHE); spa_config_exit(spa, SCL_ALL, FTAG); if (props != NULL) spa_configfile_set(spa, props, B_FALSE); if (error != 0 || (props && spa_writeable(spa) && (error = spa_prop_set(spa, props)))) { spa_unload(spa); spa_deactivate(spa); spa_remove(spa); mutex_exit(&spa_namespace_lock); return (error); } spa_async_resume(spa); /* * Override any spares and level 2 cache devices as specified by * the user, as these may have correct device names/devids, etc. */ if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) { if (spa->spa_spares.sav_config) VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); else VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, spares, nspares) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_spares(spa); spa_config_exit(spa, SCL_ALL, FTAG); spa->spa_spares.sav_sync = B_TRUE; } if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0) { if (spa->spa_l2cache.sav_config) VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); else VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_l2cache(spa); spa_config_exit(spa, SCL_ALL, FTAG); spa->spa_l2cache.sav_sync = B_TRUE; } /* * Check for any removed devices. */ if (spa->spa_autoreplace) { spa_aux_check_removed(&spa->spa_spares); spa_aux_check_removed(&spa->spa_l2cache); } if (spa_writeable(spa)) { /* * Update the config cache to include the newly-imported pool. */ spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); } /* * It's possible that the pool was expanded while it was exported. * We kick off an async task to handle this for us. */ spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); spa_history_log_version(spa, "import"); spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); mutex_exit(&spa_namespace_lock); #ifdef __FreeBSD__ #ifdef _KERNEL zvol_create_minors(pool); #endif #endif return (0); } nvlist_t * spa_tryimport(nvlist_t *tryconfig) { nvlist_t *config = NULL; char *poolname, *cachefile; spa_t *spa; uint64_t state; int error; zpool_load_policy_t policy; if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) return (NULL); if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) return (NULL); /* * Create and initialize the spa structure. */ mutex_enter(&spa_namespace_lock); spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); spa_activate(spa, FREAD); /* * Rewind pool if a max txg was provided. */ zpool_get_load_policy(spa->spa_config, &policy); if (policy.zlp_txg != UINT64_MAX) { spa->spa_load_max_txg = policy.zlp_txg; spa->spa_extreme_rewind = B_TRUE; zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld", poolname, (longlong_t)policy.zlp_txg); } else { zfs_dbgmsg("spa_tryimport: importing %s", poolname); } if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile) == 0) { zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile); spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; } else { spa->spa_config_source = SPA_CONFIG_SRC_SCAN; } error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING); /* * If 'tryconfig' was at least parsable, return the current config. */ if (spa->spa_root_vdev != NULL) { config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, poolname) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, state) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, spa->spa_uberblock.ub_timestamp) == 0); VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info) == 0); /* * If the bootfs property exists on this pool then we * copy it out so that external consumers can tell which * pools are bootable. */ if ((!error || error == EEXIST) && spa->spa_bootfs) { char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); /* * We have to play games with the name since the * pool was opened as TRYIMPORT_NAME. */ if (dsl_dsobj_to_dsname(spa_name(spa), spa->spa_bootfs, tmpname) == 0) { char *cp; char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); cp = strchr(tmpname, '/'); if (cp == NULL) { (void) strlcpy(dsname, tmpname, MAXPATHLEN); } else { (void) snprintf(dsname, MAXPATHLEN, "%s/%s", poolname, ++cp); } VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_BOOTFS, dsname) == 0); kmem_free(dsname, MAXPATHLEN); } kmem_free(tmpname, MAXPATHLEN); } /* * Add the list of hot spares and level 2 cache devices. */ spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); spa_add_spares(spa, config); spa_add_l2cache(spa, config); spa_config_exit(spa, SCL_CONFIG, FTAG); } spa_unload(spa); spa_deactivate(spa); spa_remove(spa); mutex_exit(&spa_namespace_lock); return (config); } /* * Pool export/destroy * * The act of destroying or exporting a pool is very simple. We make sure there * is no more pending I/O and any references to the pool are gone. Then, we * update the pool state and sync all the labels to disk, removing the * configuration from the cache afterwards. If the 'hardforce' flag is set, then * we don't sync the labels or remove the configuration cache. */ static int spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, boolean_t force, boolean_t hardforce) { spa_t *spa; if (oldconfig) *oldconfig = NULL; if (!(spa_mode_global & FWRITE)) return (SET_ERROR(EROFS)); mutex_enter(&spa_namespace_lock); if ((spa = spa_lookup(pool)) == NULL) { mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENOENT)); } /* * Put a hold on the pool, drop the namespace lock, stop async tasks, * reacquire the namespace lock, and see if we can export. */ spa_open_ref(spa, FTAG); mutex_exit(&spa_namespace_lock); spa_async_suspend(spa); mutex_enter(&spa_namespace_lock); spa_close(spa, FTAG); /* * The pool will be in core if it's openable, * in which case we can modify its state. */ if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { /* * Objsets may be open only because they're dirty, so we * have to force it to sync before checking spa_refcnt. */ txg_wait_synced(spa->spa_dsl_pool, 0); spa_evicting_os_wait(spa); /* * A pool cannot be exported or destroyed if there are active * references. If we are resetting a pool, allow references by * fault injection handlers. */ if (!spa_refcount_zero(spa) || (spa->spa_inject_ref != 0 && new_state != POOL_STATE_UNINITIALIZED)) { spa_async_resume(spa); mutex_exit(&spa_namespace_lock); return (SET_ERROR(EBUSY)); } /* * A pool cannot be exported if it has an active shared spare. * This is to prevent other pools stealing the active spare * from an exported pool. At user's own will, such pool can * be forcedly exported. */ if (!force && new_state == POOL_STATE_EXPORTED && spa_has_active_shared_spare(spa)) { spa_async_resume(spa); mutex_exit(&spa_namespace_lock); return (SET_ERROR(EXDEV)); } /* * We want this to be reflected on every label, * so mark them all dirty. spa_unload() will do the * final sync that pushes these changes out. */ if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa->spa_state = new_state; spa->spa_final_txg = spa_last_synced_txg(spa) + TXG_DEFER_SIZE + 1; vdev_config_dirty(spa->spa_root_vdev); spa_config_exit(spa, SCL_ALL, FTAG); } } spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); if (spa->spa_state != POOL_STATE_UNINITIALIZED) { spa_unload(spa); spa_deactivate(spa); } if (oldconfig && spa->spa_config) VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); if (new_state != POOL_STATE_UNINITIALIZED) { if (!hardforce) spa_write_cachefile(spa, B_TRUE, B_TRUE); spa_remove(spa); } mutex_exit(&spa_namespace_lock); return (0); } /* * Destroy a storage pool. */ int spa_destroy(char *pool) { return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, B_FALSE, B_FALSE)); } /* * Export a storage pool. */ int spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, boolean_t hardforce) { return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, force, hardforce)); } /* * Similar to spa_export(), this unloads the spa_t without actually removing it * from the namespace in any way. */ int spa_reset(char *pool) { return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, B_FALSE, B_FALSE)); } /* * ========================================================================== * Device manipulation * ========================================================================== */ /* * Add a device to a storage pool. */ int spa_vdev_add(spa_t *spa, nvlist_t *nvroot) { uint64_t txg, id; int error; vdev_t *rvd = spa->spa_root_vdev; vdev_t *vd, *tvd; nvlist_t **spares, **l2cache; uint_t nspares, nl2cache; ASSERT(spa_writeable(spa)); txg = spa_vdev_enter(spa); if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, VDEV_ALLOC_ADD)) != 0) return (spa_vdev_exit(spa, NULL, txg, error)); spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, &nspares) != 0) nspares = 0; if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) != 0) nl2cache = 0; if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) return (spa_vdev_exit(spa, vd, txg, EINVAL)); if (vd->vdev_children != 0 && (error = vdev_create(vd, txg, B_FALSE)) != 0) return (spa_vdev_exit(spa, vd, txg, error)); /* * We must validate the spares and l2cache devices after checking the * children. Otherwise, vdev_inuse() will blindly overwrite the spare. */ if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) return (spa_vdev_exit(spa, vd, txg, error)); /* * If we are in the middle of a device removal, we can only add * devices which match the existing devices in the pool. * If we are in the middle of a removal, or have some indirect * vdevs, we can not add raidz toplevels. */ if (spa->spa_vdev_removal != NULL || spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { for (int c = 0; c < vd->vdev_children; c++) { tvd = vd->vdev_child[c]; if (spa->spa_vdev_removal != NULL && tvd->vdev_ashift != spa->spa_vdev_removal->svr_vdev->vdev_ashift) { return (spa_vdev_exit(spa, vd, txg, EINVAL)); } /* Fail if top level vdev is raidz */ if (tvd->vdev_ops == &vdev_raidz_ops) { return (spa_vdev_exit(spa, vd, txg, EINVAL)); } /* * Need the top level mirror to be * a mirror of leaf vdevs only */ if (tvd->vdev_ops == &vdev_mirror_ops) { for (uint64_t cid = 0; cid < tvd->vdev_children; cid++) { vdev_t *cvd = tvd->vdev_child[cid]; if (!cvd->vdev_ops->vdev_op_leaf) { return (spa_vdev_exit(spa, vd, txg, EINVAL)); } } } } } for (int c = 0; c < vd->vdev_children; c++) { /* * Set the vdev id to the first hole, if one exists. */ for (id = 0; id < rvd->vdev_children; id++) { if (rvd->vdev_child[id]->vdev_ishole) { vdev_free(rvd->vdev_child[id]); break; } } tvd = vd->vdev_child[c]; vdev_remove_child(vd, tvd); tvd->vdev_id = id; vdev_add_child(rvd, tvd); vdev_config_dirty(tvd); } if (nspares != 0) { spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, ZPOOL_CONFIG_SPARES); spa_load_spares(spa); spa->spa_spares.sav_sync = B_TRUE; } if (nl2cache != 0) { spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, ZPOOL_CONFIG_L2CACHE); spa_load_l2cache(spa); spa->spa_l2cache.sav_sync = B_TRUE; } /* * We have to be careful when adding new vdevs to an existing pool. * If other threads start allocating from these vdevs before we * sync the config cache, and we lose power, then upon reboot we may * fail to open the pool because there are DVAs that the config cache * can't translate. Therefore, we first add the vdevs without * initializing metaslabs; sync the config cache (via spa_vdev_exit()); * and then let spa_config_update() initialize the new metaslabs. * * spa_load() checks for added-but-not-initialized vdevs, so that * if we lose power at any point in this sequence, the remaining * steps will be completed the next time we load the pool. */ (void) spa_vdev_exit(spa, vd, txg, 0); mutex_enter(&spa_namespace_lock); spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); mutex_exit(&spa_namespace_lock); return (0); } /* * Attach a device to a mirror. The arguments are the path to any device * in the mirror, and the nvroot for the new device. If the path specifies * a device that is not mirrored, we automatically insert the mirror vdev. * * If 'replacing' is specified, the new device is intended to replace the * existing device; in this case the two devices are made into their own * mirror using the 'replacing' vdev, which is functionally identical to * the mirror vdev (it actually reuses all the same ops) but has a few * extra rules: you can't attach to it after it's been created, and upon * completion of resilvering, the first disk (the one being replaced) * is automatically detached. */ int spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) { uint64_t txg, dtl_max_txg; vdev_t *rvd = spa->spa_root_vdev; vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; vdev_ops_t *pvops; char *oldvdpath, *newvdpath; int newvd_isspare; int error; ASSERT(spa_writeable(spa)); txg = spa_vdev_enter(spa); oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); ASSERT(MUTEX_HELD(&spa_namespace_lock)); if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { error = (spa_has_checkpoint(spa)) ? ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; return (spa_vdev_exit(spa, NULL, txg, error)); } if (spa->spa_vdev_removal != NULL || spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { return (spa_vdev_exit(spa, NULL, txg, EBUSY)); } if (oldvd == NULL) return (spa_vdev_exit(spa, NULL, txg, ENODEV)); if (!oldvd->vdev_ops->vdev_op_leaf) return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); pvd = oldvd->vdev_parent; if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, VDEV_ALLOC_ATTACH)) != 0) return (spa_vdev_exit(spa, NULL, txg, EINVAL)); if (newrootvd->vdev_children != 1) return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); newvd = newrootvd->vdev_child[0]; if (!newvd->vdev_ops->vdev_op_leaf) return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); if ((error = vdev_create(newrootvd, txg, replacing)) != 0) return (spa_vdev_exit(spa, newrootvd, txg, error)); /* * Spares can't replace logs */ if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); if (!replacing) { /* * For attach, the only allowable parent is a mirror or the root * vdev. */ if (pvd->vdev_ops != &vdev_mirror_ops && pvd->vdev_ops != &vdev_root_ops) return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); pvops = &vdev_mirror_ops; } else { /* * Active hot spares can only be replaced by inactive hot * spares. */ if (pvd->vdev_ops == &vdev_spare_ops && oldvd->vdev_isspare && !spa_has_spare(spa, newvd->vdev_guid)) return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); /* * If the source is a hot spare, and the parent isn't already a * spare, then we want to create a new hot spare. Otherwise, we * want to create a replacing vdev. The user is not allowed to * attach to a spared vdev child unless the 'isspare' state is * the same (spare replaces spare, non-spare replaces * non-spare). */ if (pvd->vdev_ops == &vdev_replacing_ops && spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); } else if (pvd->vdev_ops == &vdev_spare_ops && newvd->vdev_isspare != oldvd->vdev_isspare) { return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); } if (newvd->vdev_isspare) pvops = &vdev_spare_ops; else pvops = &vdev_replacing_ops; } /* * Make sure the new device is big enough. */ if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); /* * The new device cannot have a higher alignment requirement * than the top-level vdev. */ if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); /* * If this is an in-place replacement, update oldvd's path and devid * to make it distinguishable from newvd, and unopenable from now on. */ if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { spa_strfree(oldvd->vdev_path); oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, KM_SLEEP); (void) sprintf(oldvd->vdev_path, "%s/%s", newvd->vdev_path, "old"); if (oldvd->vdev_devid != NULL) { spa_strfree(oldvd->vdev_devid); oldvd->vdev_devid = NULL; } } /* mark the device being resilvered */ newvd->vdev_resilver_txg = txg; /* * If the parent is not a mirror, or if we're replacing, insert the new * mirror/replacing/spare vdev above oldvd. */ if (pvd->vdev_ops != pvops) pvd = vdev_add_parent(oldvd, pvops); ASSERT(pvd->vdev_top->vdev_parent == rvd); ASSERT(pvd->vdev_ops == pvops); ASSERT(oldvd->vdev_parent == pvd); /* * Extract the new device from its root and add it to pvd. */ vdev_remove_child(newrootvd, newvd); newvd->vdev_id = pvd->vdev_children; newvd->vdev_crtxg = oldvd->vdev_crtxg; vdev_add_child(pvd, newvd); tvd = newvd->vdev_top; ASSERT(pvd->vdev_top == tvd); ASSERT(tvd->vdev_parent == rvd); vdev_config_dirty(tvd); /* * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account * for any dmu_sync-ed blocks. It will propagate upward when * spa_vdev_exit() calls vdev_dtl_reassess(). */ dtl_max_txg = txg + TXG_CONCURRENT_STATES; vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, dtl_max_txg - TXG_INITIAL); if (newvd->vdev_isspare) { spa_spare_activate(newvd); spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); } oldvdpath = spa_strdup(oldvd->vdev_path); newvdpath = spa_strdup(newvd->vdev_path); newvd_isspare = newvd->vdev_isspare; /* * Mark newvd's DTL dirty in this txg. */ vdev_dirty(tvd, VDD_DTL, newvd, txg); /* * Schedule the resilver to restart in the future. We do this to * ensure that dmu_sync-ed blocks have been stitched into the * respective datasets. */ dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); if (spa->spa_bootfs) spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); /* * Commit the config */ (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); spa_history_log_internal(spa, "vdev attach", NULL, "%s vdev=%s %s vdev=%s", replacing && newvd_isspare ? "spare in" : replacing ? "replace" : "attach", newvdpath, replacing ? "for" : "to", oldvdpath); spa_strfree(oldvdpath); spa_strfree(newvdpath); return (0); } /* * Detach a device from a mirror or replacing vdev. * * If 'replace_done' is specified, only detach if the parent * is a replacing vdev. */ int spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) { uint64_t txg; int error; vdev_t *rvd = spa->spa_root_vdev; vdev_t *vd, *pvd, *cvd, *tvd; boolean_t unspare = B_FALSE; uint64_t unspare_guid = 0; char *vdpath; ASSERT(spa_writeable(spa)); txg = spa_vdev_enter(spa); vd = spa_lookup_by_guid(spa, guid, B_FALSE); /* * Besides being called directly from the userland through the * ioctl interface, spa_vdev_detach() can be potentially called * at the end of spa_vdev_resilver_done(). * * In the regular case, when we have a checkpoint this shouldn't * happen as we never empty the DTLs of a vdev during the scrub * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done() * should never get here when we have a checkpoint. * * That said, even in a case when we checkpoint the pool exactly * as spa_vdev_resilver_done() calls this function everything * should be fine as the resilver will return right away. */ ASSERT(MUTEX_HELD(&spa_namespace_lock)); if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { error = (spa_has_checkpoint(spa)) ? ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; return (spa_vdev_exit(spa, NULL, txg, error)); } if (vd == NULL) return (spa_vdev_exit(spa, NULL, txg, ENODEV)); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); pvd = vd->vdev_parent; /* * If the parent/child relationship is not as expected, don't do it. * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing * vdev that's replacing B with C. The user's intent in replacing * is to go from M(A,B) to M(A,C). If the user decides to cancel * the replace by detaching C, the expected behavior is to end up * M(A,B). But suppose that right after deciding to detach C, * the replacement of B completes. We would have M(A,C), and then * ask to detach C, which would leave us with just A -- not what * the user wanted. To prevent this, we make sure that the * parent/child relationship hasn't changed -- in this example, * that C's parent is still the replacing vdev R. */ if (pvd->vdev_guid != pguid && pguid != 0) return (spa_vdev_exit(spa, NULL, txg, EBUSY)); /* * Only 'replacing' or 'spare' vdevs can be replaced. */ if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && pvd->vdev_ops != &vdev_spare_ops) return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); ASSERT(pvd->vdev_ops != &vdev_spare_ops || spa_version(spa) >= SPA_VERSION_SPARES); /* * Only mirror, replacing, and spare vdevs support detach. */ if (pvd->vdev_ops != &vdev_replacing_ops && pvd->vdev_ops != &vdev_mirror_ops && pvd->vdev_ops != &vdev_spare_ops) return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); /* * If this device has the only valid copy of some data, * we cannot safely detach it. */ if (vdev_dtl_required(vd)) return (spa_vdev_exit(spa, NULL, txg, EBUSY)); ASSERT(pvd->vdev_children >= 2); /* * If we are detaching the second disk from a replacing vdev, then * check to see if we changed the original vdev's path to have "/old" * at the end in spa_vdev_attach(). If so, undo that change now. */ if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && vd->vdev_path != NULL) { size_t len = strlen(vd->vdev_path); for (int c = 0; c < pvd->vdev_children; c++) { cvd = pvd->vdev_child[c]; if (cvd == vd || cvd->vdev_path == NULL) continue; if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && strcmp(cvd->vdev_path + len, "/old") == 0) { spa_strfree(cvd->vdev_path); cvd->vdev_path = spa_strdup(vd->vdev_path); break; } } } /* * If we are detaching the original disk from a spare, then it implies * that the spare should become a real disk, and be removed from the * active spare list for the pool. */ if (pvd->vdev_ops == &vdev_spare_ops && vd->vdev_id == 0 && pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) unspare = B_TRUE; /* * Erase the disk labels so the disk can be used for other things. * This must be done after all other error cases are handled, * but before we disembowel vd (so we can still do I/O to it). * But if we can't do it, don't treat the error as fatal -- * it may be that the unwritability of the disk is the reason * it's being detached! */ error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); /* * Remove vd from its parent and compact the parent's children. */ vdev_remove_child(pvd, vd); vdev_compact_children(pvd); /* * Remember one of the remaining children so we can get tvd below. */ cvd = pvd->vdev_child[pvd->vdev_children - 1]; /* * If we need to remove the remaining child from the list of hot spares, * do it now, marking the vdev as no longer a spare in the process. * We must do this before vdev_remove_parent(), because that can * change the GUID if it creates a new toplevel GUID. For a similar * reason, we must remove the spare now, in the same txg as the detach; * otherwise someone could attach a new sibling, change the GUID, and * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. */ if (unspare) { ASSERT(cvd->vdev_isspare); spa_spare_remove(cvd); unspare_guid = cvd->vdev_guid; (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); cvd->vdev_unspare = B_TRUE; } /* * If the parent mirror/replacing vdev only has one child, * the parent is no longer needed. Remove it from the tree. */ if (pvd->vdev_children == 1) { if (pvd->vdev_ops == &vdev_spare_ops) cvd->vdev_unspare = B_FALSE; vdev_remove_parent(cvd); } /* * We don't set tvd until now because the parent we just removed * may have been the previous top-level vdev. */ tvd = cvd->vdev_top; ASSERT(tvd->vdev_parent == rvd); /* * Reevaluate the parent vdev state. */ vdev_propagate_state(cvd); /* * If the 'autoexpand' property is set on the pool then automatically * try to expand the size of the pool. For example if the device we * just detached was smaller than the others, it may be possible to * add metaslabs (i.e. grow the pool). We need to reopen the vdev * first so that we can obtain the updated sizes of the leaf vdevs. */ if (spa->spa_autoexpand) { vdev_reopen(tvd); vdev_expand(tvd, txg); } vdev_config_dirty(tvd); /* * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that * vd->vdev_detached is set and free vd's DTL object in syncing context. * But first make sure we're not on any *other* txg's DTL list, to * prevent vd from being accessed after it's freed. */ vdpath = spa_strdup(vd->vdev_path); for (int t = 0; t < TXG_SIZE; t++) (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); vd->vdev_detached = B_TRUE; vdev_dirty(tvd, VDD_DTL, vd, txg); spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); /* hang on to the spa before we release the lock */ spa_open_ref(spa, FTAG); error = spa_vdev_exit(spa, vd, txg, 0); spa_history_log_internal(spa, "detach", NULL, "vdev=%s", vdpath); spa_strfree(vdpath); /* * If this was the removal of the original device in a hot spare vdev, * then we want to go through and remove the device from the hot spare * list of every other pool. */ if (unspare) { spa_t *altspa = NULL; mutex_enter(&spa_namespace_lock); while ((altspa = spa_next(altspa)) != NULL) { if (altspa->spa_state != POOL_STATE_ACTIVE || altspa == spa) continue; spa_open_ref(altspa, FTAG); mutex_exit(&spa_namespace_lock); (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); mutex_enter(&spa_namespace_lock); spa_close(altspa, FTAG); } mutex_exit(&spa_namespace_lock); /* search the rest of the vdevs for spares to remove */ spa_vdev_resilver_done(spa); } /* all done with the spa; OK to release */ mutex_enter(&spa_namespace_lock); spa_close(spa, FTAG); mutex_exit(&spa_namespace_lock); return (error); } /* * Split a set of devices from their mirrors, and create a new pool from them. */ int spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, nvlist_t *props, boolean_t exp) { int error = 0; uint64_t txg, *glist; spa_t *newspa; uint_t c, children, lastlog; nvlist_t **child, *nvl, *tmp; dmu_tx_t *tx; char *altroot = NULL; vdev_t *rvd, **vml = NULL; /* vdev modify list */ boolean_t activate_slog; ASSERT(spa_writeable(spa)); txg = spa_vdev_enter(spa); ASSERT(MUTEX_HELD(&spa_namespace_lock)); if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { error = (spa_has_checkpoint(spa)) ? ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; return (spa_vdev_exit(spa, NULL, txg, error)); } /* clear the log and flush everything up to now */ activate_slog = spa_passivate_log(spa); (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); error = spa_reset_logs(spa); txg = spa_vdev_config_enter(spa); if (activate_slog) spa_activate_log(spa); if (error != 0) return (spa_vdev_exit(spa, NULL, txg, error)); /* check new spa name before going any further */ if (spa_lookup(newname) != NULL) return (spa_vdev_exit(spa, NULL, txg, EEXIST)); /* * scan through all the children to ensure they're all mirrors */ if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) return (spa_vdev_exit(spa, NULL, txg, EINVAL)); /* first, check to ensure we've got the right child count */ rvd = spa->spa_root_vdev; lastlog = 0; for (c = 0; c < rvd->vdev_children; c++) { vdev_t *vd = rvd->vdev_child[c]; /* don't count the holes & logs as children */ if (vd->vdev_islog || !vdev_is_concrete(vd)) { if (lastlog == 0) lastlog = c; continue; } lastlog = 0; } if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) return (spa_vdev_exit(spa, NULL, txg, EINVAL)); /* next, ensure no spare or cache devices are part of the split */ if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) return (spa_vdev_exit(spa, NULL, txg, EINVAL)); vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); /* then, loop over each vdev and validate it */ for (c = 0; c < children; c++) { uint64_t is_hole = 0; (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, &is_hole); if (is_hole != 0) { if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || spa->spa_root_vdev->vdev_child[c]->vdev_islog) { continue; } else { error = SET_ERROR(EINVAL); break; } } /* which disk is going to be split? */ if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, &glist[c]) != 0) { error = SET_ERROR(EINVAL); break; } /* look it up in the spa */ vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); if (vml[c] == NULL) { error = SET_ERROR(ENODEV); break; } /* make sure there's nothing stopping the split */ if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || vml[c]->vdev_islog || !vdev_is_concrete(vml[c]) || vml[c]->vdev_isspare || vml[c]->vdev_isl2cache || !vdev_writeable(vml[c]) || vml[c]->vdev_children != 0 || vml[c]->vdev_state != VDEV_STATE_HEALTHY || c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { error = SET_ERROR(EINVAL); break; } if (vdev_dtl_required(vml[c])) { error = SET_ERROR(EBUSY); break; } /* we need certain info from the top level */ VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, vml[c]->vdev_top->vdev_ms_array) == 0); VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, vml[c]->vdev_top->vdev_ms_shift) == 0); VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, vml[c]->vdev_top->vdev_asize) == 0); VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, vml[c]->vdev_top->vdev_ashift) == 0); /* transfer per-vdev ZAPs */ ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); VERIFY0(nvlist_add_uint64(child[c], ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); VERIFY0(nvlist_add_uint64(child[c], ZPOOL_CONFIG_VDEV_TOP_ZAP, vml[c]->vdev_parent->vdev_top_zap)); } if (error != 0) { kmem_free(vml, children * sizeof (vdev_t *)); kmem_free(glist, children * sizeof (uint64_t)); return (spa_vdev_exit(spa, NULL, txg, error)); } /* stop writers from using the disks */ for (c = 0; c < children; c++) { if (vml[c] != NULL) vml[c]->vdev_offline = B_TRUE; } vdev_reopen(spa->spa_root_vdev); /* * Temporarily record the splitting vdevs in the spa config. This * will disappear once the config is regenerated. */ VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, glist, children) == 0); kmem_free(glist, children * sizeof (uint64_t)); mutex_enter(&spa->spa_props_lock); VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, nvl) == 0); mutex_exit(&spa->spa_props_lock); spa->spa_config_splitting = nvl; vdev_config_dirty(spa->spa_root_vdev); /* configure and create the new pool */ VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa)) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, spa->spa_config_txg) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_generate_guid(NULL)) == 0); VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); (void) nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); /* add the new pool to the namespace */ newspa = spa_add(newname, config, altroot); newspa->spa_avz_action = AVZ_ACTION_REBUILD; newspa->spa_config_txg = spa->spa_config_txg; spa_set_log_state(newspa, SPA_LOG_CLEAR); /* release the spa config lock, retaining the namespace lock */ spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); if (zio_injection_enabled) zio_handle_panic_injection(spa, FTAG, 1); spa_activate(newspa, spa_mode_global); spa_async_suspend(newspa); #ifndef illumos /* mark that we are creating new spa by splitting */ newspa->spa_splitting_newspa = B_TRUE; #endif newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT; /* create the new pool from the disks of the original pool */ error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE); #ifndef illumos newspa->spa_splitting_newspa = B_FALSE; #endif if (error) goto out; /* if that worked, generate a real config for the new pool */ if (newspa->spa_root_vdev != NULL) { VERIFY(nvlist_alloc(&newspa->spa_config_splitting, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, B_TRUE)); } /* set the props */ if (props != NULL) { spa_configfile_set(newspa, props, B_FALSE); error = spa_prop_set(newspa, props); if (error) goto out; } /* flush everything */ txg = spa_vdev_config_enter(newspa); vdev_config_dirty(newspa->spa_root_vdev); (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); if (zio_injection_enabled) zio_handle_panic_injection(spa, FTAG, 2); spa_async_resume(newspa); /* finally, update the original pool's config */ txg = spa_vdev_config_enter(spa); tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); error = dmu_tx_assign(tx, TXG_WAIT); if (error != 0) dmu_tx_abort(tx); for (c = 0; c < children; c++) { if (vml[c] != NULL) { vdev_split(vml[c]); if (error == 0) spa_history_log_internal(spa, "detach", tx, "vdev=%s", vml[c]->vdev_path); vdev_free(vml[c]); } } spa->spa_avz_action = AVZ_ACTION_REBUILD; vdev_config_dirty(spa->spa_root_vdev); spa->spa_config_splitting = NULL; nvlist_free(nvl); if (error == 0) dmu_tx_commit(tx); (void) spa_vdev_exit(spa, NULL, txg, 0); if (zio_injection_enabled) zio_handle_panic_injection(spa, FTAG, 3); /* split is complete; log a history record */ spa_history_log_internal(newspa, "split", NULL, "from pool %s", spa_name(spa)); kmem_free(vml, children * sizeof (vdev_t *)); /* if we're not going to mount the filesystems in userland, export */ if (exp) error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, B_FALSE, B_FALSE); return (error); out: spa_unload(newspa); spa_deactivate(newspa); spa_remove(newspa); txg = spa_vdev_config_enter(spa); /* re-online all offlined disks */ for (c = 0; c < children; c++) { if (vml[c] != NULL) vml[c]->vdev_offline = B_FALSE; } vdev_reopen(spa->spa_root_vdev); nvlist_free(spa->spa_config_splitting); spa->spa_config_splitting = NULL; (void) spa_vdev_exit(spa, NULL, txg, error); kmem_free(vml, children * sizeof (vdev_t *)); return (error); } /* * Find any device that's done replacing, or a vdev marked 'unspare' that's * currently spared, so we can detach it. */ static vdev_t * spa_vdev_resilver_done_hunt(vdev_t *vd) { vdev_t *newvd, *oldvd; for (int c = 0; c < vd->vdev_children; c++) { oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); if (oldvd != NULL) return (oldvd); } /* * Check for a completed replacement. We always consider the first * vdev in the list to be the oldest vdev, and the last one to be * the newest (see spa_vdev_attach() for how that works). In * the case where the newest vdev is faulted, we will not automatically * remove it after a resilver completes. This is OK as it will require * user intervention to determine which disk the admin wishes to keep. */ if (vd->vdev_ops == &vdev_replacing_ops) { ASSERT(vd->vdev_children > 1); newvd = vd->vdev_child[vd->vdev_children - 1]; oldvd = vd->vdev_child[0]; if (vdev_dtl_empty(newvd, DTL_MISSING) && vdev_dtl_empty(newvd, DTL_OUTAGE) && !vdev_dtl_required(oldvd)) return (oldvd); } /* * Check for a completed resilver with the 'unspare' flag set. */ if (vd->vdev_ops == &vdev_spare_ops) { vdev_t *first = vd->vdev_child[0]; vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; if (last->vdev_unspare) { oldvd = first; newvd = last; } else if (first->vdev_unspare) { oldvd = last; newvd = first; } else { oldvd = NULL; } if (oldvd != NULL && vdev_dtl_empty(newvd, DTL_MISSING) && vdev_dtl_empty(newvd, DTL_OUTAGE) && !vdev_dtl_required(oldvd)) return (oldvd); /* * If there are more than two spares attached to a disk, * and those spares are not required, then we want to * attempt to free them up now so that they can be used * by other pools. Once we're back down to a single * disk+spare, we stop removing them. */ if (vd->vdev_children > 2) { newvd = vd->vdev_child[1]; if (newvd->vdev_isspare && last->vdev_isspare && vdev_dtl_empty(last, DTL_MISSING) && vdev_dtl_empty(last, DTL_OUTAGE) && !vdev_dtl_required(newvd)) return (newvd); } } return (NULL); } static void spa_vdev_resilver_done(spa_t *spa) { vdev_t *vd, *pvd, *ppvd; uint64_t guid, sguid, pguid, ppguid; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { pvd = vd->vdev_parent; ppvd = pvd->vdev_parent; guid = vd->vdev_guid; pguid = pvd->vdev_guid; ppguid = ppvd->vdev_guid; sguid = 0; /* * If we have just finished replacing a hot spared device, then * we need to detach the parent's first child (the original hot * spare) as well. */ if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && ppvd->vdev_children == 2) { ASSERT(pvd->vdev_ops == &vdev_replacing_ops); sguid = ppvd->vdev_child[1]->vdev_guid; } ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); spa_config_exit(spa, SCL_ALL, FTAG); if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) return; if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) return; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); } spa_config_exit(spa, SCL_ALL, FTAG); } /* * Update the stored path or FRU for this vdev. */ int spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, boolean_t ispath) { vdev_t *vd; boolean_t sync = B_FALSE; ASSERT(spa_writeable(spa)); spa_vdev_state_enter(spa, SCL_ALL); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, ENOENT)); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); if (ispath) { if (strcmp(value, vd->vdev_path) != 0) { spa_strfree(vd->vdev_path); vd->vdev_path = spa_strdup(value); sync = B_TRUE; } } else { if (vd->vdev_fru == NULL) { vd->vdev_fru = spa_strdup(value); sync = B_TRUE; } else if (strcmp(value, vd->vdev_fru) != 0) { spa_strfree(vd->vdev_fru); vd->vdev_fru = spa_strdup(value); sync = B_TRUE; } } return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); } int spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) { return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); } int spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) { return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); } /* * ========================================================================== * SPA Scanning * ========================================================================== */ int spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) { ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); if (dsl_scan_resilvering(spa->spa_dsl_pool)) return (SET_ERROR(EBUSY)); return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); } int spa_scan_stop(spa_t *spa) { ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); if (dsl_scan_resilvering(spa->spa_dsl_pool)) return (SET_ERROR(EBUSY)); return (dsl_scan_cancel(spa->spa_dsl_pool)); } int spa_scan(spa_t *spa, pool_scan_func_t func) { ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) return (SET_ERROR(ENOTSUP)); /* * If a resilver was requested, but there is no DTL on a * writeable leaf device, we have nothing to do. */ if (func == POOL_SCAN_RESILVER && !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); return (0); } return (dsl_scan(spa->spa_dsl_pool, func)); } /* * ========================================================================== * SPA async task processing * ========================================================================== */ static void spa_async_remove(spa_t *spa, vdev_t *vd) { if (vd->vdev_remove_wanted) { vd->vdev_remove_wanted = B_FALSE; vd->vdev_delayed_close = B_FALSE; vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); /* * We want to clear the stats, but we don't want to do a full * vdev_clear() as that will cause us to throw away * degraded/faulted state as well as attempt to reopen the * device, all of which is a waste. */ vd->vdev_stat.vs_read_errors = 0; vd->vdev_stat.vs_write_errors = 0; vd->vdev_stat.vs_checksum_errors = 0; vdev_state_dirty(vd->vdev_top); /* Tell userspace that the vdev is gone. */ zfs_post_remove(spa, vd); } for (int c = 0; c < vd->vdev_children; c++) spa_async_remove(spa, vd->vdev_child[c]); } static void spa_async_probe(spa_t *spa, vdev_t *vd) { if (vd->vdev_probe_wanted) { vd->vdev_probe_wanted = B_FALSE; vdev_reopen(vd); /* vdev_open() does the actual probe */ } for (int c = 0; c < vd->vdev_children; c++) spa_async_probe(spa, vd->vdev_child[c]); } static void spa_async_autoexpand(spa_t *spa, vdev_t *vd) { sysevent_id_t eid; nvlist_t *attr; char *physpath; if (!spa->spa_autoexpand) return; for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; spa_async_autoexpand(spa, cvd); } if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) return; physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); nvlist_free(attr); kmem_free(physpath, MAXPATHLEN); } static void spa_async_thread(void *arg) { spa_t *spa = (spa_t *)arg; int tasks; ASSERT(spa->spa_sync_on); mutex_enter(&spa->spa_async_lock); tasks = spa->spa_async_tasks; spa->spa_async_tasks &= SPA_ASYNC_REMOVE; mutex_exit(&spa->spa_async_lock); /* * See if the config needs to be updated. */ if (tasks & SPA_ASYNC_CONFIG_UPDATE) { uint64_t old_space, new_space; mutex_enter(&spa_namespace_lock); old_space = metaslab_class_get_space(spa_normal_class(spa)); spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); new_space = metaslab_class_get_space(spa_normal_class(spa)); mutex_exit(&spa_namespace_lock); /* * If the pool grew as a result of the config update, * then log an internal history event. */ if (new_space != old_space) { spa_history_log_internal(spa, "vdev online", NULL, "pool '%s' size: %llu(+%llu)", spa_name(spa), new_space, new_space - old_space); } } if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); spa_async_autoexpand(spa, spa->spa_root_vdev); spa_config_exit(spa, SCL_CONFIG, FTAG); } /* * See if any devices need to be probed. */ if (tasks & SPA_ASYNC_PROBE) { spa_vdev_state_enter(spa, SCL_NONE); spa_async_probe(spa, spa->spa_root_vdev); (void) spa_vdev_state_exit(spa, NULL, 0); } /* * If any devices are done replacing, detach them. */ if (tasks & SPA_ASYNC_RESILVER_DONE) spa_vdev_resilver_done(spa); /* * Kick off a resilver. */ if (tasks & SPA_ASYNC_RESILVER) dsl_resilver_restart(spa->spa_dsl_pool, 0); /* * Let the world know that we're done. */ mutex_enter(&spa->spa_async_lock); spa->spa_async_thread = NULL; cv_broadcast(&spa->spa_async_cv); mutex_exit(&spa->spa_async_lock); thread_exit(); } static void spa_async_thread_vd(void *arg) { spa_t *spa = arg; int tasks; mutex_enter(&spa->spa_async_lock); tasks = spa->spa_async_tasks; retry: spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; mutex_exit(&spa->spa_async_lock); /* * See if any devices need to be marked REMOVED. */ if (tasks & SPA_ASYNC_REMOVE) { spa_vdev_state_enter(spa, SCL_NONE); spa_async_remove(spa, spa->spa_root_vdev); for (int i = 0; i < spa->spa_l2cache.sav_count; i++) spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); for (int i = 0; i < spa->spa_spares.sav_count; i++) spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); (void) spa_vdev_state_exit(spa, NULL, 0); } /* * Let the world know that we're done. */ mutex_enter(&spa->spa_async_lock); tasks = spa->spa_async_tasks; if ((tasks & SPA_ASYNC_REMOVE) != 0) goto retry; spa->spa_async_thread_vd = NULL; cv_broadcast(&spa->spa_async_cv); mutex_exit(&spa->spa_async_lock); thread_exit(); } void spa_async_suspend(spa_t *spa) { mutex_enter(&spa->spa_async_lock); spa->spa_async_suspended++; while (spa->spa_async_thread != NULL || spa->spa_async_thread_vd != NULL) cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); mutex_exit(&spa->spa_async_lock); spa_vdev_remove_suspend(spa); zthr_t *condense_thread = spa->spa_condense_zthr; if (condense_thread != NULL && zthr_isrunning(condense_thread)) VERIFY0(zthr_cancel(condense_thread)); zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; if (discard_thread != NULL && zthr_isrunning(discard_thread)) VERIFY0(zthr_cancel(discard_thread)); } void spa_async_resume(spa_t *spa) { mutex_enter(&spa->spa_async_lock); ASSERT(spa->spa_async_suspended != 0); spa->spa_async_suspended--; mutex_exit(&spa->spa_async_lock); spa_restart_removal(spa); zthr_t *condense_thread = spa->spa_condense_zthr; if (condense_thread != NULL && !zthr_isrunning(condense_thread)) zthr_resume(condense_thread); zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; if (discard_thread != NULL && !zthr_isrunning(discard_thread)) zthr_resume(discard_thread); } static boolean_t spa_async_tasks_pending(spa_t *spa) { uint_t non_config_tasks; uint_t config_task; boolean_t config_task_suspended; non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | SPA_ASYNC_REMOVE); config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; if (spa->spa_ccw_fail_time == 0) { config_task_suspended = B_FALSE; } else { config_task_suspended = (gethrtime() - spa->spa_ccw_fail_time) < (zfs_ccw_retry_interval * NANOSEC); } return (non_config_tasks || (config_task && !config_task_suspended)); } static void spa_async_dispatch(spa_t *spa) { mutex_enter(&spa->spa_async_lock); if (spa_async_tasks_pending(spa) && !spa->spa_async_suspended && spa->spa_async_thread == NULL && rootdir != NULL) spa->spa_async_thread = thread_create(NULL, 0, spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); mutex_exit(&spa->spa_async_lock); } static void spa_async_dispatch_vd(spa_t *spa) { mutex_enter(&spa->spa_async_lock); if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && !spa->spa_async_suspended && spa->spa_async_thread_vd == NULL && rootdir != NULL) spa->spa_async_thread_vd = thread_create(NULL, 0, spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); mutex_exit(&spa->spa_async_lock); } void spa_async_request(spa_t *spa, int task) { zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); mutex_enter(&spa->spa_async_lock); spa->spa_async_tasks |= task; mutex_exit(&spa->spa_async_lock); spa_async_dispatch_vd(spa); } /* * ========================================================================== * SPA syncing routines * ========================================================================== */ static int bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { bpobj_t *bpo = arg; bpobj_enqueue(bpo, bp, tx); return (0); } static int spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { zio_t *zio = arg; zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, BP_GET_PSIZE(bp), zio->io_flags)); return (0); } /* * Note: this simple function is not inlined to make it easier to dtrace the * amount of time spent syncing frees. */ static void spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) { zio_t *zio = zio_root(spa, NULL, NULL, 0); bplist_iterate(bpl, spa_free_sync_cb, zio, tx); VERIFY(zio_wait(zio) == 0); } /* * Note: this simple function is not inlined to make it easier to dtrace the * amount of time spent syncing deferred frees. */ static void spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) { zio_t *zio = zio_root(spa, NULL, NULL, 0); VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, spa_free_sync_cb, zio, tx), ==, 0); VERIFY0(zio_wait(zio)); } static void spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) { char *packed = NULL; size_t bufsize; size_t nvsize = 0; dmu_buf_t *db; VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); /* * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration * information. This avoids the dmu_buf_will_dirty() path and * saves us a pre-read to get data we don't actually care about. */ bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); packed = kmem_alloc(bufsize, KM_SLEEP); VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, KM_SLEEP) == 0); bzero(packed + nvsize, bufsize - nvsize); dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); kmem_free(packed, bufsize); VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); dmu_buf_will_dirty(db, tx); *(uint64_t *)db->db_data = nvsize; dmu_buf_rele(db, FTAG); } static void spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, const char *config, const char *entry) { nvlist_t *nvroot; nvlist_t **list; int i; if (!sav->sav_sync) return; /* * Update the MOS nvlist describing the list of available devices. * spa_validate_aux() will have already made sure this nvlist is * valid and the vdevs are labeled appropriately. */ if (sav->sav_object == 0) { sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); VERIFY(zap_update(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, &sav->sav_object, tx) == 0); } VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); if (sav->sav_count == 0) { VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); } else { list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); for (i = 0; i < sav->sav_count; i++) list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], B_FALSE, VDEV_CONFIG_L2CACHE); VERIFY(nvlist_add_nvlist_array(nvroot, config, list, sav->sav_count) == 0); for (i = 0; i < sav->sav_count; i++) nvlist_free(list[i]); kmem_free(list, sav->sav_count * sizeof (void *)); } spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); nvlist_free(nvroot); sav->sav_sync = B_FALSE; } /* * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. * The all-vdev ZAP must be empty. */ static void spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) { spa_t *spa = vd->vdev_spa; if (vd->vdev_top_zap != 0) { VERIFY0(zap_add_int(spa->spa_meta_objset, avz, vd->vdev_top_zap, tx)); } if (vd->vdev_leaf_zap != 0) { VERIFY0(zap_add_int(spa->spa_meta_objset, avz, vd->vdev_leaf_zap, tx)); } for (uint64_t i = 0; i < vd->vdev_children; i++) { spa_avz_build(vd->vdev_child[i], avz, tx); } } static void spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) { nvlist_t *config; /* * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, * its config may not be dirty but we still need to build per-vdev ZAPs. * Similarly, if the pool is being assembled (e.g. after a split), we * need to rebuild the AVZ although the config may not be dirty. */ if (list_is_empty(&spa->spa_config_dirty_list) && spa->spa_avz_action == AVZ_ACTION_NONE) return; spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || spa->spa_avz_action == AVZ_ACTION_INITIALIZE || spa->spa_all_vdev_zaps != 0); if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { /* Make and build the new AVZ */ uint64_t new_avz = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); spa_avz_build(spa->spa_root_vdev, new_avz, tx); /* Diff old AVZ with new one */ zap_cursor_t zc; zap_attribute_t za; for (zap_cursor_init(&zc, spa->spa_meta_objset, spa->spa_all_vdev_zaps); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { uint64_t vdzap = za.za_first_integer; if (zap_lookup_int(spa->spa_meta_objset, new_avz, vdzap) == ENOENT) { /* * ZAP is listed in old AVZ but not in new one; * destroy it */ VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, tx)); } } zap_cursor_fini(&zc); /* Destroy the old AVZ */ VERIFY0(zap_destroy(spa->spa_meta_objset, spa->spa_all_vdev_zaps, tx)); /* Replace the old AVZ in the dir obj with the new one */ VERIFY0(zap_update(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, sizeof (new_avz), 1, &new_avz, tx)); spa->spa_all_vdev_zaps = new_avz; } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { zap_cursor_t zc; zap_attribute_t za; /* Walk through the AVZ and destroy all listed ZAPs */ for (zap_cursor_init(&zc, spa->spa_meta_objset, spa->spa_all_vdev_zaps); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { uint64_t zap = za.za_first_integer; VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); } zap_cursor_fini(&zc); /* Destroy and unlink the AVZ itself */ VERIFY0(zap_destroy(spa->spa_meta_objset, spa->spa_all_vdev_zaps, tx)); VERIFY0(zap_remove(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); spa->spa_all_vdev_zaps = 0; } if (spa->spa_all_vdev_zaps == 0) { spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx); } spa->spa_avz_action = AVZ_ACTION_NONE; /* Create ZAPs for vdevs that don't have them. */ vdev_construct_zaps(spa->spa_root_vdev, tx); config = spa_config_generate(spa, spa->spa_root_vdev, dmu_tx_get_txg(tx), B_FALSE); /* * If we're upgrading the spa version then make sure that * the config object gets updated with the correct version. */ if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa->spa_uberblock.ub_version); spa_config_exit(spa, SCL_STATE, FTAG); nvlist_free(spa->spa_config_syncing); spa->spa_config_syncing = config; spa_sync_nvlist(spa, spa->spa_config_object, config, tx); } static void spa_sync_version(void *arg, dmu_tx_t *tx) { uint64_t *versionp = arg; uint64_t version = *versionp; spa_t *spa = dmu_tx_pool(tx)->dp_spa; /* * Setting the version is special cased when first creating the pool. */ ASSERT(tx->tx_txg != TXG_INITIAL); ASSERT(SPA_VERSION_IS_SUPPORTED(version)); ASSERT(version >= spa_version(spa)); spa->spa_uberblock.ub_version = version; vdev_config_dirty(spa->spa_root_vdev); spa_history_log_internal(spa, "set", tx, "version=%lld", version); } /* * Set zpool properties. */ static void spa_sync_props(void *arg, dmu_tx_t *tx) { nvlist_t *nvp = arg; spa_t *spa = dmu_tx_pool(tx)->dp_spa; objset_t *mos = spa->spa_meta_objset; nvpair_t *elem = NULL; mutex_enter(&spa->spa_props_lock); while ((elem = nvlist_next_nvpair(nvp, elem))) { uint64_t intval; char *strval, *fname; zpool_prop_t prop; const char *propname; zprop_type_t proptype; spa_feature_t fid; switch (prop = zpool_name_to_prop(nvpair_name(elem))) { case ZPOOL_PROP_INVAL: /* * We checked this earlier in spa_prop_validate(). */ ASSERT(zpool_prop_feature(nvpair_name(elem))); fname = strchr(nvpair_name(elem), '@') + 1; VERIFY0(zfeature_lookup_name(fname, &fid)); spa_feature_enable(spa, fid, tx); spa_history_log_internal(spa, "set", tx, "%s=enabled", nvpair_name(elem)); break; case ZPOOL_PROP_VERSION: intval = fnvpair_value_uint64(elem); /* * The version is synced seperatly before other * properties and should be correct by now. */ ASSERT3U(spa_version(spa), >=, intval); break; case ZPOOL_PROP_ALTROOT: /* * 'altroot' is a non-persistent property. It should * have been set temporarily at creation or import time. */ ASSERT(spa->spa_root != NULL); break; case ZPOOL_PROP_READONLY: case ZPOOL_PROP_CACHEFILE: /* * 'readonly' and 'cachefile' are also non-persisitent * properties. */ break; case ZPOOL_PROP_COMMENT: strval = fnvpair_value_string(elem); if (spa->spa_comment != NULL) spa_strfree(spa->spa_comment); spa->spa_comment = spa_strdup(strval); /* * We need to dirty the configuration on all the vdevs * so that their labels get updated. It's unnecessary * to do this for pool creation since the vdev's * configuratoin has already been dirtied. */ if (tx->tx_txg != TXG_INITIAL) vdev_config_dirty(spa->spa_root_vdev); spa_history_log_internal(spa, "set", tx, "%s=%s", nvpair_name(elem), strval); break; default: /* * Set pool property values in the poolprops mos object. */ if (spa->spa_pool_props_object == 0) { spa->spa_pool_props_object = zap_create_link(mos, DMU_OT_POOL_PROPS, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, tx); } /* normalize the property name */ propname = zpool_prop_to_name(prop); proptype = zpool_prop_get_type(prop); if (nvpair_type(elem) == DATA_TYPE_STRING) { ASSERT(proptype == PROP_TYPE_STRING); strval = fnvpair_value_string(elem); VERIFY0(zap_update(mos, spa->spa_pool_props_object, propname, 1, strlen(strval) + 1, strval, tx)); spa_history_log_internal(spa, "set", tx, "%s=%s", nvpair_name(elem), strval); } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { intval = fnvpair_value_uint64(elem); if (proptype == PROP_TYPE_INDEX) { const char *unused; VERIFY0(zpool_prop_index_to_string( prop, intval, &unused)); } VERIFY0(zap_update(mos, spa->spa_pool_props_object, propname, 8, 1, &intval, tx)); spa_history_log_internal(spa, "set", tx, "%s=%lld", nvpair_name(elem), intval); } else { ASSERT(0); /* not allowed */ } switch (prop) { case ZPOOL_PROP_DELEGATION: spa->spa_delegation = intval; break; case ZPOOL_PROP_BOOTFS: spa->spa_bootfs = intval; break; case ZPOOL_PROP_FAILUREMODE: spa->spa_failmode = intval; break; case ZPOOL_PROP_AUTOEXPAND: spa->spa_autoexpand = intval; if (tx->tx_txg != TXG_INITIAL) spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); break; case ZPOOL_PROP_DEDUPDITTO: spa->spa_dedup_ditto = intval; break; default: break; } } } mutex_exit(&spa->spa_props_lock); } /* * Perform one-time upgrade on-disk changes. spa_version() does not * reflect the new version this txg, so there must be no changes this * txg to anything that the upgrade code depends on after it executes. * Therefore this must be called after dsl_pool_sync() does the sync * tasks. */ static void spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) { dsl_pool_t *dp = spa->spa_dsl_pool; ASSERT(spa->spa_sync_pass == 1); rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { dsl_pool_create_origin(dp, tx); /* Keeping the origin open increases spa_minref */ spa->spa_minref += 3; } if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { dsl_pool_upgrade_clones(dp, tx); } if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { dsl_pool_upgrade_dir_clones(dp, tx); /* Keeping the freedir open increases spa_minref */ spa->spa_minref += 3; } if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { spa_feature_create_zap_objects(spa, tx); } /* * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable * when possibility to use lz4 compression for metadata was added * Old pools that have this feature enabled must be upgraded to have * this feature active */ if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { boolean_t lz4_en = spa_feature_is_enabled(spa, SPA_FEATURE_LZ4_COMPRESS); boolean_t lz4_ac = spa_feature_is_active(spa, SPA_FEATURE_LZ4_COMPRESS); if (lz4_en && !lz4_ac) spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); } /* * If we haven't written the salt, do so now. Note that the * feature may not be activated yet, but that's fine since * the presence of this ZAP entry is backwards compatible. */ if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT) == ENOENT) { VERIFY0(zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, sizeof (spa->spa_cksum_salt.zcs_bytes), spa->spa_cksum_salt.zcs_bytes, tx)); } rrw_exit(&dp->dp_config_rwlock, FTAG); } static void vdev_indirect_state_sync_verify(vdev_t *vd) { vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; vdev_indirect_births_t *vib = vd->vdev_indirect_births; if (vd->vdev_ops == &vdev_indirect_ops) { ASSERT(vim != NULL); ASSERT(vib != NULL); } if (vdev_obsolete_sm_object(vd) != 0) { ASSERT(vd->vdev_obsolete_sm != NULL); ASSERT(vd->vdev_removing || vd->vdev_ops == &vdev_indirect_ops); ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); ASSERT3U(vdev_obsolete_sm_object(vd), ==, space_map_object(vd->vdev_obsolete_sm)); ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, space_map_allocated(vd->vdev_obsolete_sm)); } ASSERT(vd->vdev_obsolete_segments != NULL); /* * Since frees / remaps to an indirect vdev can only * happen in syncing context, the obsolete segments * tree must be empty when we start syncing. */ ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); } /* * Sync the specified transaction group. New blocks may be dirtied as * part of the process, so we iterate until it converges. */ void spa_sync(spa_t *spa, uint64_t txg) { dsl_pool_t *dp = spa->spa_dsl_pool; objset_t *mos = spa->spa_meta_objset; bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; vdev_t *rvd = spa->spa_root_vdev; vdev_t *vd; dmu_tx_t *tx; int error; uint32_t max_queue_depth = zfs_vdev_async_write_max_active * zfs_vdev_queue_depth_pct / 100; VERIFY(spa_writeable(spa)); /* * Wait for i/os issued in open context that need to complete * before this txg syncs. */ VERIFY0(zio_wait(spa->spa_txg_zio[txg & TXG_MASK])); spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 0); /* * Lock out configuration changes. */ spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); spa->spa_syncing_txg = txg; spa->spa_sync_pass = 0; - mutex_enter(&spa->spa_alloc_lock); - VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); - mutex_exit(&spa->spa_alloc_lock); + for (int i = 0; i < spa->spa_alloc_count; i++) { + mutex_enter(&spa->spa_alloc_locks[i]); + VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i])); + mutex_exit(&spa->spa_alloc_locks[i]); + } /* * If there are any pending vdev state changes, convert them * into config changes that go out with this transaction group. */ spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); while (list_head(&spa->spa_state_dirty_list) != NULL) { /* * We need the write lock here because, for aux vdevs, * calling vdev_config_dirty() modifies sav_config. * This is ugly and will become unnecessary when we * eliminate the aux vdev wart by integrating all vdevs * into the root vdev tree. */ spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { vdev_state_clean(vd); vdev_config_dirty(vd); } spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); } spa_config_exit(spa, SCL_STATE, FTAG); tx = dmu_tx_create_assigned(dp, txg); spa->spa_sync_starttime = gethrtime(); #ifdef illumos VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, spa->spa_sync_starttime + spa->spa_deadman_synctime)); #else /* !illumos */ #ifdef _KERNEL callout_schedule(&spa->spa_deadman_cycid, hz * spa->spa_deadman_synctime / NANOSEC); #endif #endif /* illumos */ /* * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, * set spa_deflate if we have no raid-z vdevs. */ if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { int i; for (i = 0; i < rvd->vdev_children; i++) { vd = rvd->vdev_child[i]; if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) break; } if (i == rvd->vdev_children) { spa->spa_deflate = TRUE; VERIFY(0 == zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, sizeof (uint64_t), 1, &spa->spa_deflate, tx)); } } /* * Set the top-level vdev's max queue depth. Evaluate each * top-level's async write queue depth in case it changed. * The max queue depth will not change in the middle of syncing * out this txg. */ - uint64_t queue_depth_total = 0; + uint64_t slots_per_allocator = 0; for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; if (mg == NULL || mg->mg_class != spa_normal_class(spa) || !metaslab_group_initialized(mg)) continue; /* * It is safe to do a lock-free check here because only async * allocations look at mg_max_alloc_queue_depth, and async * allocations all happen from spa_sync(). */ - ASSERT0(refcount_count(&mg->mg_alloc_queue_depth)); + for (int i = 0; i < spa->spa_alloc_count; i++) + ASSERT0(refcount_count(&(mg->mg_alloc_queue_depth[i]))); mg->mg_max_alloc_queue_depth = max_queue_depth; - queue_depth_total += mg->mg_max_alloc_queue_depth; + + for (int i = 0; i < spa->spa_alloc_count; i++) { + mg->mg_cur_max_alloc_queue_depth[i] = + zfs_vdev_def_queue_depth; + } + slots_per_allocator += zfs_vdev_def_queue_depth; } metaslab_class_t *mc = spa_normal_class(spa); - ASSERT0(refcount_count(&mc->mc_alloc_slots)); - mc->mc_alloc_max_slots = queue_depth_total; + for (int i = 0; i < spa->spa_alloc_count; i++) { + ASSERT0(refcount_count(&mc->mc_alloc_slots[i])); + mc->mc_alloc_max_slots[i] = slots_per_allocator; + } mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; - ASSERT3U(mc->mc_alloc_max_slots, <=, - max_queue_depth * rvd->vdev_children); - for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *vd = rvd->vdev_child[c]; vdev_indirect_state_sync_verify(vd); if (vdev_indirect_should_condense(vd)) { spa_condense_indirect_start_sync(vd, tx); break; } } /* * Iterate to convergence. */ do { int pass = ++spa->spa_sync_pass; spa_sync_config_object(spa, tx); spa_sync_aux_dev(spa, &spa->spa_spares, tx, ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); spa_errlog_sync(spa, txg); dsl_pool_sync(dp, txg); if (pass < zfs_sync_pass_deferred_free) { spa_sync_frees(spa, free_bpl, tx); } else { /* * We can not defer frees in pass 1, because * we sync the deferred frees later in pass 1. */ ASSERT3U(pass, >, 1); bplist_iterate(free_bpl, bpobj_enqueue_cb, &spa->spa_deferred_bpobj, tx); } ddt_sync(spa, txg); dsl_scan_sync(dp, tx); if (spa->spa_vdev_removal != NULL) svr_sync(spa, tx); while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) != NULL) vdev_sync(vd, txg); if (pass == 1) { spa_sync_upgrades(spa, tx); ASSERT3U(txg, >=, spa->spa_uberblock.ub_rootbp.blk_birth); /* * Note: We need to check if the MOS is dirty * because we could have marked the MOS dirty * without updating the uberblock (e.g. if we * have sync tasks but no dirty user data). We * need to check the uberblock's rootbp because * it is updated if we have synced out dirty * data (though in this case the MOS will most * likely also be dirty due to second order * effects, we don't want to rely on that here). */ if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && !dmu_objset_is_dirty(mos, txg)) { /* * Nothing changed on the first pass, * therefore this TXG is a no-op. Avoid * syncing deferred frees, so that we * can keep this TXG as a no-op. */ ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, txg)); break; } spa_sync_deferred_frees(spa, tx); } } while (dmu_objset_is_dirty(mos, txg)); if (!list_is_empty(&spa->spa_config_dirty_list)) { /* * Make sure that the number of ZAPs for all the vdevs matches * the number of ZAPs in the per-vdev ZAP list. This only gets * called if the config is dirty; otherwise there may be * outstanding AVZ operations that weren't completed in * spa_sync_config_object. */ uint64_t all_vdev_zap_entry_count; ASSERT0(zap_count(spa->spa_meta_objset, spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, all_vdev_zap_entry_count); } if (spa->spa_vdev_removal != NULL) { ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); } /* * Rewrite the vdev configuration (which includes the uberblock) * to commit the transaction group. * * If there are no dirty vdevs, we sync the uberblock to a few * random top-level vdevs that are known to be visible in the * config cache (see spa_vdev_add() for a complete description). * If there *are* dirty vdevs, sync the uberblock to all vdevs. */ for (;;) { /* * We hold SCL_STATE to prevent vdev open/close/etc. * while we're attempting to write the vdev labels. */ spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); if (list_is_empty(&spa->spa_config_dirty_list)) { vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; int svdcount = 0; int children = rvd->vdev_children; int c0 = spa_get_random(children); for (int c = 0; c < children; c++) { vd = rvd->vdev_child[(c0 + c) % children]; /* Stop when revisiting the first vdev */ if (c > 0 && svd[0] == vd) break; if (vd->vdev_ms_array == 0 || vd->vdev_islog || !vdev_is_concrete(vd)) continue; svd[svdcount++] = vd; if (svdcount == SPA_SYNC_MIN_VDEVS) break; } error = vdev_config_sync(svd, svdcount, txg); } else { error = vdev_config_sync(rvd->vdev_child, rvd->vdev_children, txg); } if (error == 0) spa->spa_last_synced_guid = rvd->vdev_guid; spa_config_exit(spa, SCL_STATE, FTAG); if (error == 0) break; zio_suspend(spa, NULL); zio_resume_wait(spa); } dmu_tx_commit(tx); #ifdef illumos VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); #else /* !illumos */ #ifdef _KERNEL callout_drain(&spa->spa_deadman_cycid); #endif #endif /* illumos */ /* * Clear the dirty config list. */ while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) vdev_config_clean(vd); /* * Now that the new config has synced transactionally, * let it become visible to the config cache. */ if (spa->spa_config_syncing != NULL) { spa_config_set(spa, spa->spa_config_syncing); spa->spa_config_txg = txg; spa->spa_config_syncing = NULL; } dsl_pool_sync_done(dp, txg); - mutex_enter(&spa->spa_alloc_lock); - VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); - mutex_exit(&spa->spa_alloc_lock); + for (int i = 0; i < spa->spa_alloc_count; i++) { + mutex_enter(&spa->spa_alloc_locks[i]); + VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i])); + mutex_exit(&spa->spa_alloc_locks[i]); + } /* * Update usable space statistics. */ while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) vdev_sync_done(vd, txg); spa_update_dspace(spa); /* * It had better be the case that we didn't dirty anything * since vdev_config_sync(). */ ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); while (zfs_pause_spa_sync) delay(1); spa->spa_sync_pass = 0; /* * Update the last synced uberblock here. We want to do this at * the end of spa_sync() so that consumers of spa_last_synced_txg() * will be guaranteed that all the processing associated with * that txg has been completed. */ spa->spa_ubsync = spa->spa_uberblock; spa_config_exit(spa, SCL_CONFIG, FTAG); spa_handle_ignored_writes(spa); /* * If any async tasks have been requested, kick them off. */ spa_async_dispatch(spa); spa_async_dispatch_vd(spa); } /* * Sync all pools. We don't want to hold the namespace lock across these * operations, so we take a reference on the spa_t and drop the lock during the * sync. */ void spa_sync_allpools(void) { spa_t *spa = NULL; mutex_enter(&spa_namespace_lock); while ((spa = spa_next(spa)) != NULL) { if (spa_state(spa) != POOL_STATE_ACTIVE || !spa_writeable(spa) || spa_suspended(spa)) continue; spa_open_ref(spa, FTAG); mutex_exit(&spa_namespace_lock); txg_wait_synced(spa_get_dsl(spa), 0); mutex_enter(&spa_namespace_lock); spa_close(spa, FTAG); } mutex_exit(&spa_namespace_lock); } /* * ========================================================================== * Miscellaneous routines * ========================================================================== */ /* * Remove all pools in the system. */ void spa_evict_all(void) { spa_t *spa; /* * Remove all cached state. All pools should be closed now, * so every spa in the AVL tree should be unreferenced. */ mutex_enter(&spa_namespace_lock); while ((spa = spa_next(NULL)) != NULL) { /* * Stop async tasks. The async thread may need to detach * a device that's been replaced, which requires grabbing * spa_namespace_lock, so we must drop it here. */ spa_open_ref(spa, FTAG); mutex_exit(&spa_namespace_lock); spa_async_suspend(spa); mutex_enter(&spa_namespace_lock); spa_close(spa, FTAG); if (spa->spa_state != POOL_STATE_UNINITIALIZED) { spa_unload(spa); spa_deactivate(spa); } spa_remove(spa); } mutex_exit(&spa_namespace_lock); } vdev_t * spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) { vdev_t *vd; int i; if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) return (vd); if (aux) { for (i = 0; i < spa->spa_l2cache.sav_count; i++) { vd = spa->spa_l2cache.sav_vdevs[i]; if (vd->vdev_guid == guid) return (vd); } for (i = 0; i < spa->spa_spares.sav_count; i++) { vd = spa->spa_spares.sav_vdevs[i]; if (vd->vdev_guid == guid) return (vd); } } return (NULL); } void spa_upgrade(spa_t *spa, uint64_t version) { ASSERT(spa_writeable(spa)); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); /* * This should only be called for a non-faulted pool, and since a * future version would result in an unopenable pool, this shouldn't be * possible. */ ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); ASSERT3U(version, >=, spa->spa_uberblock.ub_version); spa->spa_uberblock.ub_version = version; vdev_config_dirty(spa->spa_root_vdev); spa_config_exit(spa, SCL_ALL, FTAG); txg_wait_synced(spa_get_dsl(spa), 0); } boolean_t spa_has_spare(spa_t *spa, uint64_t guid) { int i; uint64_t spareguid; spa_aux_vdev_t *sav = &spa->spa_spares; for (i = 0; i < sav->sav_count; i++) if (sav->sav_vdevs[i]->vdev_guid == guid) return (B_TRUE); for (i = 0; i < sav->sav_npending; i++) { if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, &spareguid) == 0 && spareguid == guid) return (B_TRUE); } return (B_FALSE); } /* * Check if a pool has an active shared spare device. * Note: reference count of an active spare is 2, as a spare and as a replace */ static boolean_t spa_has_active_shared_spare(spa_t *spa) { int i, refcnt; uint64_t pool; spa_aux_vdev_t *sav = &spa->spa_spares; for (i = 0; i < sav->sav_count; i++) { if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, &refcnt) && pool != 0ULL && pool == spa_guid(spa) && refcnt > 2) return (B_TRUE); } return (B_FALSE); } sysevent_t * spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) { sysevent_t *ev = NULL; #ifdef _KERNEL sysevent_attr_list_t *attr = NULL; sysevent_value_t value; ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", SE_SLEEP); ASSERT(ev != NULL); value.value_type = SE_DATA_TYPE_STRING; value.value.sv_string = spa_name(spa); if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) goto done; value.value_type = SE_DATA_TYPE_UINT64; value.value.sv_uint64 = spa_guid(spa); if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) goto done; if (vd) { value.value_type = SE_DATA_TYPE_UINT64; value.value.sv_uint64 = vd->vdev_guid; if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, SE_SLEEP) != 0) goto done; if (vd->vdev_path) { value.value_type = SE_DATA_TYPE_STRING; value.value.sv_string = vd->vdev_path; if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, &value, SE_SLEEP) != 0) goto done; } } if (hist_nvl != NULL) { fnvlist_merge((nvlist_t *)attr, hist_nvl); } if (sysevent_attach_attributes(ev, attr) != 0) goto done; attr = NULL; done: if (attr) sysevent_free_attr(attr); #endif return (ev); } void spa_event_post(sysevent_t *ev) { #ifdef _KERNEL sysevent_id_t eid; (void) log_sysevent(ev, SE_SLEEP, &eid); sysevent_free(ev); #endif } void spa_event_discard(sysevent_t *ev) { #ifdef _KERNEL sysevent_free(ev); #endif } /* * Post a sysevent corresponding to the given event. The 'name' must be one of * the event definitions in sys/sysevent/eventdefs.h. The payload will be * filled in from the spa and (optionally) the vdev and history nvl. This * doesn't do anything in the userland libzpool, as we don't want consumers to * misinterpret ztest or zdb as real changes. */ void spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) { spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); } Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c (revision 336949) @@ -1,2394 +1,2410 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011, 2017 by Delphix. All rights reserved. + * Copyright (c) 2011, 2018 by Delphix. All rights reserved. * Copyright 2015 Nexenta Systems, Inc. All rights reserved. * Copyright 2013 Martin Matuska . All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright 2013 Saso Kiselkov. All rights reserved. * Copyright (c) 2014 Integros [integros.com] * Copyright (c) 2017 Datto Inc. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "zfs_prop.h" #include #if defined(__FreeBSD__) && defined(_KERNEL) #include #include #endif /* * SPA locking * * There are four basic locks for managing spa_t structures: * * spa_namespace_lock (global mutex) * * This lock must be acquired to do any of the following: * * - Lookup a spa_t by name * - Add or remove a spa_t from the namespace * - Increase spa_refcount from non-zero * - Check if spa_refcount is zero * - Rename a spa_t * - add/remove/attach/detach devices * - Held for the duration of create/destroy/import/export * * It does not need to handle recursion. A create or destroy may * reference objects (files or zvols) in other pools, but by * definition they must have an existing reference, and will never need * to lookup a spa_t by name. * * spa_refcount (per-spa refcount_t protected by mutex) * * This reference count keep track of any active users of the spa_t. The * spa_t cannot be destroyed or freed while this is non-zero. Internally, * the refcount is never really 'zero' - opening a pool implicitly keeps * some references in the DMU. Internally we check against spa_minref, but * present the image of a zero/non-zero value to consumers. * * spa_config_lock[] (per-spa array of rwlocks) * * This protects the spa_t from config changes, and must be held in * the following circumstances: * * - RW_READER to perform I/O to the spa * - RW_WRITER to change the vdev config * * The locking order is fairly straightforward: * * spa_namespace_lock -> spa_refcount * * The namespace lock must be acquired to increase the refcount from 0 * or to check if it is zero. * * spa_refcount -> spa_config_lock[] * * There must be at least one valid reference on the spa_t to acquire * the config lock. * * spa_namespace_lock -> spa_config_lock[] * * The namespace lock must always be taken before the config lock. * * * The spa_namespace_lock can be acquired directly and is globally visible. * * The namespace is manipulated using the following functions, all of which * require the spa_namespace_lock to be held. * * spa_lookup() Lookup a spa_t by name. * * spa_add() Create a new spa_t in the namespace. * * spa_remove() Remove a spa_t from the namespace. This also * frees up any memory associated with the spa_t. * * spa_next() Returns the next spa_t in the system, or the * first if NULL is passed. * * spa_evict_all() Shutdown and remove all spa_t structures in * the system. * * spa_guid_exists() Determine whether a pool/device guid exists. * * The spa_refcount is manipulated using the following functions: * * spa_open_ref() Adds a reference to the given spa_t. Must be * called with spa_namespace_lock held if the * refcount is currently zero. * * spa_close() Remove a reference from the spa_t. This will * not free the spa_t or remove it from the * namespace. No locking is required. * * spa_refcount_zero() Returns true if the refcount is currently * zero. Must be called with spa_namespace_lock * held. * * The spa_config_lock[] is an array of rwlocks, ordered as follows: * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV. * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}(). * * To read the configuration, it suffices to hold one of these locks as reader. * To modify the configuration, you must hold all locks as writer. To modify * vdev state without altering the vdev tree's topology (e.g. online/offline), * you must hold SCL_STATE and SCL_ZIO as writer. * * We use these distinct config locks to avoid recursive lock entry. * For example, spa_sync() (which holds SCL_CONFIG as reader) induces * block allocations (SCL_ALLOC), which may require reading space maps * from disk (dmu_read() -> zio_read() -> SCL_ZIO). * * The spa config locks cannot be normal rwlocks because we need the * ability to hand off ownership. For example, SCL_ZIO is acquired * by the issuing thread and later released by an interrupt thread. * They do, however, obey the usual write-wanted semantics to prevent * writer (i.e. system administrator) starvation. * * The lock acquisition rules are as follows: * * SCL_CONFIG * Protects changes to the vdev tree topology, such as vdev * add/remove/attach/detach. Protects the dirty config list * (spa_config_dirty_list) and the set of spares and l2arc devices. * * SCL_STATE * Protects changes to pool state and vdev state, such as vdev * online/offline/fault/degrade/clear. Protects the dirty state list * (spa_state_dirty_list) and global pool state (spa_state). * * SCL_ALLOC * Protects changes to metaslab groups and classes. * Held as reader by metaslab_alloc() and metaslab_claim(). * * SCL_ZIO * Held by bp-level zios (those which have no io_vd upon entry) * to prevent changes to the vdev tree. The bp-level zio implicitly * protects all of its vdev child zios, which do not hold SCL_ZIO. * * SCL_FREE * Protects changes to metaslab groups and classes. * Held as reader by metaslab_free(). SCL_FREE is distinct from * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free * blocks in zio_done() while another i/o that holds either * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete. * * SCL_VDEV * Held as reader to prevent changes to the vdev tree during trivial * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the * other locks, and lower than all of them, to ensure that it's safe * to acquire regardless of caller context. * * In addition, the following rules apply: * * (a) spa_props_lock protects pool properties, spa_config and spa_config_list. * The lock ordering is SCL_CONFIG > spa_props_lock. * * (b) I/O operations on leaf vdevs. For any zio operation that takes * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(), * or zio_write_phys() -- the caller must ensure that the config cannot * cannot change in the interim, and that the vdev cannot be reopened. * SCL_STATE as reader suffices for both. * * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit(). * * spa_vdev_enter() Acquire the namespace lock and the config lock * for writing. * * spa_vdev_exit() Release the config lock, wait for all I/O * to complete, sync the updated configs to the * cache, and release the namespace lock. * * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit(). * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual * locking is, always, based on spa_namespace_lock and spa_config_lock[]. * * spa_rename() is also implemented within this file since it requires * manipulation of the namespace. */ static avl_tree_t spa_namespace_avl; kmutex_t spa_namespace_lock; static kcondvar_t spa_namespace_cv; static int spa_active_count; int spa_max_replication_override = SPA_DVAS_PER_BP; static kmutex_t spa_spare_lock; static avl_tree_t spa_spare_avl; static kmutex_t spa_l2cache_lock; static avl_tree_t spa_l2cache_avl; kmem_cache_t *spa_buffer_pool; int spa_mode_global; #ifdef ZFS_DEBUG /* * Everything except dprintf, spa, and indirect_remap is on by default * in debug builds. */ int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA | ZFS_DEBUG_INDIRECT_REMAP); #else int zfs_flags = 0; #endif /* * zfs_recover can be set to nonzero to attempt to recover from * otherwise-fatal errors, typically caused by on-disk corruption. When * set, calls to zfs_panic_recover() will turn into warning messages. * This should only be used as a last resort, as it typically results * in leaked space, or worse. */ boolean_t zfs_recover = B_FALSE; /* * If destroy encounters an EIO while reading metadata (e.g. indirect * blocks), space referenced by the missing metadata can not be freed. * Normally this causes the background destroy to become "stalled", as * it is unable to make forward progress. While in this stalled state, * all remaining space to free from the error-encountering filesystem is * "temporarily leaked". Set this flag to cause it to ignore the EIO, * permanently leak the space from indirect blocks that can not be read, * and continue to free everything else that it can. * * The default, "stalling" behavior is useful if the storage partially * fails (i.e. some but not all i/os fail), and then later recovers. In * this case, we will be able to continue pool operations while it is * partially failed, and when it recovers, we can continue to free the * space, with no leaks. However, note that this case is actually * fairly rare. * * Typically pools either (a) fail completely (but perhaps temporarily, * e.g. a top-level vdev going offline), or (b) have localized, * permanent errors (e.g. disk returns the wrong data due to bit flip or * firmware bug). In case (a), this setting does not matter because the * pool will be suspended and the sync thread will not be able to make * forward progress regardless. In case (b), because the error is * permanent, the best we can do is leak the minimum amount of space, * which is what setting this flag will do. Therefore, it is reasonable * for this flag to normally be set, but we chose the more conservative * approach of not setting it, so that there is no possibility of * leaking space in the "partial temporary" failure case. */ boolean_t zfs_free_leak_on_eio = B_FALSE; /* * Expiration time in milliseconds. This value has two meanings. First it is * used to determine when the spa_deadman() logic should fire. By default the * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds. * Secondly, the value determines if an I/O is considered "hung". Any I/O that * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting * in a system panic. */ uint64_t zfs_deadman_synctime_ms = 1000000ULL; /* * Check time in milliseconds. This defines the frequency at which we check * for hung I/O. */ uint64_t zfs_deadman_checktime_ms = 5000ULL; /* * Default value of -1 for zfs_deadman_enabled is resolved in * zfs_deadman_init() */ int zfs_deadman_enabled = -1; /* * The worst case is single-sector max-parity RAID-Z blocks, in which * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1) * times the size; so just assume that. Add to this the fact that * we can have up to 3 DVAs per bp, and one more factor of 2 because * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together, * the worst case is: * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24 */ int spa_asize_inflation = 24; #if defined(__FreeBSD__) && defined(_KERNEL) SYSCTL_DECL(_vfs_zfs); SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0, "Try to recover from otherwise-fatal errors."); static int sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS) { int err, val; val = zfs_flags; err = sysctl_handle_int(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); /* * ZFS_DEBUG_MODIFY must be enabled prior to boot so all * arc buffers in the system have the necessary additional * checksum data. However, it is safe to disable at any * time. */ if (!(zfs_flags & ZFS_DEBUG_MODIFY)) val &= ~ZFS_DEBUG_MODIFY; zfs_flags = val; return (0); } SYSCTL_PROC(_vfs_zfs, OID_AUTO, debugflags, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int), sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing."); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN, &zfs_deadman_synctime_ms, 0, "Stalled ZFS I/O expiration time in milliseconds"); SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN, &zfs_deadman_checktime_ms, 0, "Period of checks for stalled ZFS I/O in milliseconds"); SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN, &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O"); SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN, &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes"); #endif #ifndef illumos #ifdef _KERNEL static void zfs_deadman_init() { /* * If we are not i386 or amd64 or in a virtual machine, * disable ZFS deadman thread by default */ if (zfs_deadman_enabled == -1) { #if defined(__amd64__) || defined(__i386__) zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0; #else zfs_deadman_enabled = 0; #endif } } #endif /* _KERNEL */ #endif /* !illumos */ /* * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in * the pool to be consumed. This ensures that we don't run the pool * completely out of space, due to unaccounted changes (e.g. to the MOS). * It also limits the worst-case time to allocate space. If we have * less than this amount of free space, most ZPL operations (e.g. write, * create) will return ENOSPC. * * Certain operations (e.g. file removal, most administrative actions) can * use half the slop space. They will only return ENOSPC if less than half * the slop space is free. Typically, once the pool has less than the slop * space free, the user will use these operations to free up space in the pool. * These are the operations that call dsl_pool_adjustedsize() with the netfree * argument set to TRUE. * * Operations that are almost guaranteed to free up space in the absence of * a pool checkpoint can use up to three quarters of the slop space * (e.g zfs destroy). * * A very restricted set of operations are always permitted, regardless of * the amount of free space. These are the operations that call * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net * increase in the amount of space used, it is possible to run the pool * completely out of space, causing it to be permanently read-only. * * Note that on very small pools, the slop space will be larger than * 3.2%, in an effort to have it be at least spa_min_slop (128MB), * but we never allow it to be more than half the pool size. * * See also the comments in zfs_space_check_t. */ int spa_slop_shift = 5; SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN, &spa_slop_shift, 0, "Shift value of reserved space (1/(2^spa_slop_shift))."); uint64_t spa_min_slop = 128 * 1024 * 1024; SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, spa_min_slop, CTLFLAG_RWTUN, &spa_min_slop, 0, "Minimal value of reserved space"); +int spa_allocators = 4; + /*PRINTFLIKE2*/ void spa_load_failed(spa_t *spa, const char *fmt, ...) { va_list adx; char buf[256]; va_start(adx, fmt); (void) vsnprintf(buf, sizeof (buf), fmt, adx); va_end(adx); zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name, spa->spa_trust_config ? "trusted" : "untrusted", buf); } /*PRINTFLIKE2*/ void spa_load_note(spa_t *spa, const char *fmt, ...) { va_list adx; char buf[256]; va_start(adx, fmt); (void) vsnprintf(buf, sizeof (buf), fmt, adx); va_end(adx); zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name, spa->spa_trust_config ? "trusted" : "untrusted", buf); } /* * ========================================================================== * SPA config locking * ========================================================================== */ static void spa_config_lock_init(spa_t *spa) { for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL); refcount_create_untracked(&scl->scl_count); scl->scl_writer = NULL; scl->scl_write_wanted = 0; } } static void spa_config_lock_destroy(spa_t *spa) { for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; mutex_destroy(&scl->scl_lock); cv_destroy(&scl->scl_cv); refcount_destroy(&scl->scl_count); ASSERT(scl->scl_writer == NULL); ASSERT(scl->scl_write_wanted == 0); } } int spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw) { for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (!(locks & (1 << i))) continue; mutex_enter(&scl->scl_lock); if (rw == RW_READER) { if (scl->scl_writer || scl->scl_write_wanted) { mutex_exit(&scl->scl_lock); spa_config_exit(spa, locks & ((1 << i) - 1), tag); return (0); } } else { ASSERT(scl->scl_writer != curthread); if (!refcount_is_zero(&scl->scl_count)) { mutex_exit(&scl->scl_lock); spa_config_exit(spa, locks & ((1 << i) - 1), tag); return (0); } scl->scl_writer = curthread; } (void) refcount_add(&scl->scl_count, tag); mutex_exit(&scl->scl_lock); } return (1); } void spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw) { int wlocks_held = 0; ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY); for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (scl->scl_writer == curthread) wlocks_held |= (1 << i); if (!(locks & (1 << i))) continue; mutex_enter(&scl->scl_lock); if (rw == RW_READER) { while (scl->scl_writer || scl->scl_write_wanted) { cv_wait(&scl->scl_cv, &scl->scl_lock); } } else { ASSERT(scl->scl_writer != curthread); while (!refcount_is_zero(&scl->scl_count)) { scl->scl_write_wanted++; cv_wait(&scl->scl_cv, &scl->scl_lock); scl->scl_write_wanted--; } scl->scl_writer = curthread; } (void) refcount_add(&scl->scl_count, tag); mutex_exit(&scl->scl_lock); } ASSERT3U(wlocks_held, <=, locks); } void spa_config_exit(spa_t *spa, int locks, void *tag) { for (int i = SCL_LOCKS - 1; i >= 0; i--) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (!(locks & (1 << i))) continue; mutex_enter(&scl->scl_lock); ASSERT(!refcount_is_zero(&scl->scl_count)); if (refcount_remove(&scl->scl_count, tag) == 0) { ASSERT(scl->scl_writer == NULL || scl->scl_writer == curthread); scl->scl_writer = NULL; /* OK in either case */ cv_broadcast(&scl->scl_cv); } mutex_exit(&scl->scl_lock); } } int spa_config_held(spa_t *spa, int locks, krw_t rw) { int locks_held = 0; for (int i = 0; i < SCL_LOCKS; i++) { spa_config_lock_t *scl = &spa->spa_config_lock[i]; if (!(locks & (1 << i))) continue; if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) || (rw == RW_WRITER && scl->scl_writer == curthread)) locks_held |= 1 << i; } return (locks_held); } /* * ========================================================================== * SPA namespace functions * ========================================================================== */ /* * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held. * Returns NULL if no matching spa_t is found. */ spa_t * spa_lookup(const char *name) { static spa_t search; /* spa_t is large; don't allocate on stack */ spa_t *spa; avl_index_t where; char *cp; ASSERT(MUTEX_HELD(&spa_namespace_lock)); (void) strlcpy(search.spa_name, name, sizeof (search.spa_name)); /* * If it's a full dataset name, figure out the pool name and * just use that. */ cp = strpbrk(search.spa_name, "/@#"); if (cp != NULL) *cp = '\0'; spa = avl_find(&spa_namespace_avl, &search, &where); return (spa); } /* * Fires when spa_sync has not completed within zfs_deadman_synctime_ms. * If the zfs_deadman_enabled flag is set then it inspects all vdev queues * looking for potentially hung I/Os. */ static void spa_deadman(void *arg, int pending) { spa_t *spa = arg; /* * Disable the deadman timer if the pool is suspended. */ if (spa_suspended(spa)) { #ifdef illumos VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); #else /* Nothing. just don't schedule any future callouts. */ #endif return; } zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu", (gethrtime() - spa->spa_sync_starttime) / NANOSEC, ++spa->spa_deadman_calls); if (zfs_deadman_enabled) vdev_deadman(spa->spa_root_vdev); #ifdef __FreeBSD__ #ifdef _KERNEL callout_schedule(&spa->spa_deadman_cycid, hz * zfs_deadman_checktime_ms / MILLISEC); #endif #endif } #if defined(__FreeBSD__) && defined(_KERNEL) static void spa_deadman_timeout(void *arg) { spa_t *spa = arg; taskqueue_enqueue(taskqueue_thread, &spa->spa_deadman_task); } #endif /* * Create an uninitialized spa_t with the given name. Requires * spa_namespace_lock. The caller must ensure that the spa_t doesn't already * exist by calling spa_lookup() first. */ spa_t * spa_add(const char *name, nvlist_t *config, const char *altroot) { spa_t *spa; spa_config_dirent_t *dp; #ifdef illumos cyc_handler_t hdlr; cyc_time_t when; #endif ASSERT(MUTEX_HELD(&spa_namespace_lock)); spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP); mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL); - mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL); cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL); for (int t = 0; t < TXG_SIZE; t++) bplist_create(&spa->spa_free_bplist[t]); (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name)); spa->spa_state = POOL_STATE_UNINITIALIZED; spa->spa_freeze_txg = UINT64_MAX; spa->spa_final_txg = UINT64_MAX; spa->spa_load_max_txg = UINT64_MAX; spa->spa_proc = &p0; spa->spa_proc_state = SPA_PROC_NONE; spa->spa_trust_config = B_TRUE; #ifdef illumos hdlr.cyh_func = spa_deadman; hdlr.cyh_arg = spa; hdlr.cyh_level = CY_LOW_LEVEL; #endif spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms); #ifdef illumos /* * This determines how often we need to check for hung I/Os after * the cyclic has already fired. Since checking for hung I/Os is * an expensive operation we don't want to check too frequently. * Instead wait for 5 seconds before checking again. */ when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms); when.cyt_when = CY_INFINITY; mutex_enter(&cpu_lock); spa->spa_deadman_cycid = cyclic_add(&hdlr, &when); mutex_exit(&cpu_lock); #else /* !illumos */ #ifdef _KERNEL /* * callout(9) does not provide a way to initialize a callout with * a function and an argument, so we use callout_reset() to schedule * the callout in the very distant future. Even if that event ever * fires, it should be okayas we won't have any active zio-s. * But normally spa_sync() will reschedule the callout with a proper * timeout. * callout(9) does not allow the callback function to sleep but * vdev_deadman() needs to acquire vq_lock and illumos mutexes are * emulated using sx(9). For this reason spa_deadman_timeout() * will schedule spa_deadman() as task on a taskqueue that allows * sleeping. */ TASK_INIT(&spa->spa_deadman_task, 0, spa_deadman, spa); callout_init(&spa->spa_deadman_cycid, 1); callout_reset_sbt(&spa->spa_deadman_cycid, SBT_MAX, 0, spa_deadman_timeout, spa, 0); #endif #endif refcount_create(&spa->spa_refcount); spa_config_lock_init(spa); avl_add(&spa_namespace_avl, spa); /* * Set the alternate root, if there is one. */ if (altroot) { spa->spa_root = spa_strdup(altroot); spa_active_count++; } - avl_create(&spa->spa_alloc_tree, zio_bookmark_compare, - sizeof (zio_t), offsetof(zio_t, io_alloc_node)); + spa->spa_alloc_count = spa_allocators; + spa->spa_alloc_locks = kmem_zalloc(spa->spa_alloc_count * + sizeof (kmutex_t), KM_SLEEP); + spa->spa_alloc_trees = kmem_zalloc(spa->spa_alloc_count * + sizeof (avl_tree_t), KM_SLEEP); + for (int i = 0; i < spa->spa_alloc_count; i++) { + mutex_init(&spa->spa_alloc_locks[i], NULL, MUTEX_DEFAULT, NULL); + avl_create(&spa->spa_alloc_trees[i], zio_bookmark_compare, + sizeof (zio_t), offsetof(zio_t, io_alloc_node)); + } /* * Every pool starts with the default cachefile */ list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t), offsetof(spa_config_dirent_t, scd_link)); dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP); dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path); list_insert_head(&spa->spa_config_list, dp); VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME, KM_SLEEP) == 0); if (config != NULL) { nvlist_t *features; if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, &features) == 0) { VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); } VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0); } if (spa->spa_label_features == NULL) { VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME, KM_SLEEP) == 0); } spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0); spa->spa_min_ashift = INT_MAX; spa->spa_max_ashift = 0; /* * As a pool is being created, treat all features as disabled by * setting SPA_FEATURE_DISABLED for all entries in the feature * refcount cache. */ for (int i = 0; i < SPA_FEATURES; i++) { spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; } return (spa); } /* * Removes a spa_t from the namespace, freeing up any memory used. Requires * spa_namespace_lock. This is called only after the spa_t has been closed and * deactivated. */ void spa_remove(spa_t *spa) { spa_config_dirent_t *dp; ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0); nvlist_free(spa->spa_config_splitting); avl_remove(&spa_namespace_avl, spa); cv_broadcast(&spa_namespace_cv); if (spa->spa_root) { spa_strfree(spa->spa_root); spa_active_count--; } while ((dp = list_head(&spa->spa_config_list)) != NULL) { list_remove(&spa->spa_config_list, dp); if (dp->scd_path != NULL) spa_strfree(dp->scd_path); kmem_free(dp, sizeof (spa_config_dirent_t)); } - avl_destroy(&spa->spa_alloc_tree); + for (int i = 0; i < spa->spa_alloc_count; i++) { + avl_destroy(&spa->spa_alloc_trees[i]); + mutex_destroy(&spa->spa_alloc_locks[i]); + } + kmem_free(spa->spa_alloc_locks, spa->spa_alloc_count * + sizeof (kmutex_t)); + kmem_free(spa->spa_alloc_trees, spa->spa_alloc_count * + sizeof (avl_tree_t)); + list_destroy(&spa->spa_config_list); nvlist_free(spa->spa_label_features); nvlist_free(spa->spa_load_info); spa_config_set(spa, NULL); #ifdef illumos mutex_enter(&cpu_lock); if (spa->spa_deadman_cycid != CYCLIC_NONE) cyclic_remove(spa->spa_deadman_cycid); mutex_exit(&cpu_lock); spa->spa_deadman_cycid = CYCLIC_NONE; #else /* !illumos */ #ifdef _KERNEL callout_drain(&spa->spa_deadman_cycid); taskqueue_drain(taskqueue_thread, &spa->spa_deadman_task); #endif #endif refcount_destroy(&spa->spa_refcount); spa_config_lock_destroy(spa); for (int t = 0; t < TXG_SIZE; t++) bplist_destroy(&spa->spa_free_bplist[t]); zio_checksum_templates_free(spa); cv_destroy(&spa->spa_async_cv); cv_destroy(&spa->spa_evicting_os_cv); cv_destroy(&spa->spa_proc_cv); cv_destroy(&spa->spa_scrub_io_cv); cv_destroy(&spa->spa_suspend_cv); - mutex_destroy(&spa->spa_alloc_lock); mutex_destroy(&spa->spa_async_lock); mutex_destroy(&spa->spa_errlist_lock); mutex_destroy(&spa->spa_errlog_lock); mutex_destroy(&spa->spa_evicting_os_lock); mutex_destroy(&spa->spa_history_lock); mutex_destroy(&spa->spa_proc_lock); mutex_destroy(&spa->spa_props_lock); mutex_destroy(&spa->spa_cksum_tmpls_lock); mutex_destroy(&spa->spa_scrub_lock); mutex_destroy(&spa->spa_suspend_lock); mutex_destroy(&spa->spa_vdev_top_lock); kmem_free(spa, sizeof (spa_t)); } /* * Given a pool, return the next pool in the namespace, or NULL if there is * none. If 'prev' is NULL, return the first pool. */ spa_t * spa_next(spa_t *prev) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); if (prev) return (AVL_NEXT(&spa_namespace_avl, prev)); else return (avl_first(&spa_namespace_avl)); } /* * ========================================================================== * SPA refcount functions * ========================================================================== */ /* * Add a reference to the given spa_t. Must have at least one reference, or * have the namespace lock held. */ void spa_open_ref(spa_t *spa, void *tag) { ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref || MUTEX_HELD(&spa_namespace_lock)); (void) refcount_add(&spa->spa_refcount, tag); } /* * Remove a reference to the given spa_t. Must have at least one reference, or * have the namespace lock held. */ void spa_close(spa_t *spa, void *tag) { ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref || MUTEX_HELD(&spa_namespace_lock)); (void) refcount_remove(&spa->spa_refcount, tag); } /* * Remove a reference to the given spa_t held by a dsl dir that is * being asynchronously released. Async releases occur from a taskq * performing eviction of dsl datasets and dirs. The namespace lock * isn't held and the hold by the object being evicted may contribute to * spa_minref (e.g. dataset or directory released during pool export), * so the asserts in spa_close() do not apply. */ void spa_async_close(spa_t *spa, void *tag) { (void) refcount_remove(&spa->spa_refcount, tag); } /* * Check to see if the spa refcount is zero. Must be called with * spa_namespace_lock held. We really compare against spa_minref, which is the * number of references acquired when opening a pool */ boolean_t spa_refcount_zero(spa_t *spa) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); return (refcount_count(&spa->spa_refcount) == spa->spa_minref); } /* * ========================================================================== * SPA spare and l2cache tracking * ========================================================================== */ /* * Hot spares and cache devices are tracked using the same code below, * for 'auxiliary' devices. */ typedef struct spa_aux { uint64_t aux_guid; uint64_t aux_pool; avl_node_t aux_avl; int aux_count; } spa_aux_t; static int spa_aux_compare(const void *a, const void *b) { const spa_aux_t *sa = a; const spa_aux_t *sb = b; if (sa->aux_guid < sb->aux_guid) return (-1); else if (sa->aux_guid > sb->aux_guid) return (1); else return (0); } void spa_aux_add(vdev_t *vd, avl_tree_t *avl) { avl_index_t where; spa_aux_t search; spa_aux_t *aux; search.aux_guid = vd->vdev_guid; if ((aux = avl_find(avl, &search, &where)) != NULL) { aux->aux_count++; } else { aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP); aux->aux_guid = vd->vdev_guid; aux->aux_count = 1; avl_insert(avl, aux, where); } } void spa_aux_remove(vdev_t *vd, avl_tree_t *avl) { spa_aux_t search; spa_aux_t *aux; avl_index_t where; search.aux_guid = vd->vdev_guid; aux = avl_find(avl, &search, &where); ASSERT(aux != NULL); if (--aux->aux_count == 0) { avl_remove(avl, aux); kmem_free(aux, sizeof (spa_aux_t)); } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) { aux->aux_pool = 0ULL; } } boolean_t spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl) { spa_aux_t search, *found; search.aux_guid = guid; found = avl_find(avl, &search, NULL); if (pool) { if (found) *pool = found->aux_pool; else *pool = 0ULL; } if (refcnt) { if (found) *refcnt = found->aux_count; else *refcnt = 0; } return (found != NULL); } void spa_aux_activate(vdev_t *vd, avl_tree_t *avl) { spa_aux_t search, *found; avl_index_t where; search.aux_guid = vd->vdev_guid; found = avl_find(avl, &search, &where); ASSERT(found != NULL); ASSERT(found->aux_pool == 0ULL); found->aux_pool = spa_guid(vd->vdev_spa); } /* * Spares are tracked globally due to the following constraints: * * - A spare may be part of multiple pools. * - A spare may be added to a pool even if it's actively in use within * another pool. * - A spare in use in any pool can only be the source of a replacement if * the target is a spare in the same pool. * * We keep track of all spares on the system through the use of a reference * counted AVL tree. When a vdev is added as a spare, or used as a replacement * spare, then we bump the reference count in the AVL tree. In addition, we set * the 'vdev_isspare' member to indicate that the device is a spare (active or * inactive). When a spare is made active (used to replace a device in the * pool), we also keep track of which pool its been made a part of. * * The 'spa_spare_lock' protects the AVL tree. These functions are normally * called under the spa_namespace lock as part of vdev reconfiguration. The * separate spare lock exists for the status query path, which does not need to * be completely consistent with respect to other vdev configuration changes. */ static int spa_spare_compare(const void *a, const void *b) { return (spa_aux_compare(a, b)); } void spa_spare_add(vdev_t *vd) { mutex_enter(&spa_spare_lock); ASSERT(!vd->vdev_isspare); spa_aux_add(vd, &spa_spare_avl); vd->vdev_isspare = B_TRUE; mutex_exit(&spa_spare_lock); } void spa_spare_remove(vdev_t *vd) { mutex_enter(&spa_spare_lock); ASSERT(vd->vdev_isspare); spa_aux_remove(vd, &spa_spare_avl); vd->vdev_isspare = B_FALSE; mutex_exit(&spa_spare_lock); } boolean_t spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt) { boolean_t found; mutex_enter(&spa_spare_lock); found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl); mutex_exit(&spa_spare_lock); return (found); } void spa_spare_activate(vdev_t *vd) { mutex_enter(&spa_spare_lock); ASSERT(vd->vdev_isspare); spa_aux_activate(vd, &spa_spare_avl); mutex_exit(&spa_spare_lock); } /* * Level 2 ARC devices are tracked globally for the same reasons as spares. * Cache devices currently only support one pool per cache device, and so * for these devices the aux reference count is currently unused beyond 1. */ static int spa_l2cache_compare(const void *a, const void *b) { return (spa_aux_compare(a, b)); } void spa_l2cache_add(vdev_t *vd) { mutex_enter(&spa_l2cache_lock); ASSERT(!vd->vdev_isl2cache); spa_aux_add(vd, &spa_l2cache_avl); vd->vdev_isl2cache = B_TRUE; mutex_exit(&spa_l2cache_lock); } void spa_l2cache_remove(vdev_t *vd) { mutex_enter(&spa_l2cache_lock); ASSERT(vd->vdev_isl2cache); spa_aux_remove(vd, &spa_l2cache_avl); vd->vdev_isl2cache = B_FALSE; mutex_exit(&spa_l2cache_lock); } boolean_t spa_l2cache_exists(uint64_t guid, uint64_t *pool) { boolean_t found; mutex_enter(&spa_l2cache_lock); found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl); mutex_exit(&spa_l2cache_lock); return (found); } void spa_l2cache_activate(vdev_t *vd) { mutex_enter(&spa_l2cache_lock); ASSERT(vd->vdev_isl2cache); spa_aux_activate(vd, &spa_l2cache_avl); mutex_exit(&spa_l2cache_lock); } /* * ========================================================================== * SPA vdev locking * ========================================================================== */ /* * Lock the given spa_t for the purpose of adding or removing a vdev. * Grabs the global spa_namespace_lock plus the spa config lock for writing. * It returns the next transaction group for the spa_t. */ uint64_t spa_vdev_enter(spa_t *spa) { mutex_enter(&spa->spa_vdev_top_lock); mutex_enter(&spa_namespace_lock); return (spa_vdev_config_enter(spa)); } /* * Internal implementation for spa_vdev_enter(). Used when a vdev * operation requires multiple syncs (i.e. removing a device) while * keeping the spa_namespace_lock held. */ uint64_t spa_vdev_config_enter(spa_t *spa) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); return (spa_last_synced_txg(spa) + 1); } /* * Used in combination with spa_vdev_config_enter() to allow the syncing * of multiple transactions without releasing the spa_namespace_lock. */ void spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag) { ASSERT(MUTEX_HELD(&spa_namespace_lock)); int config_changed = B_FALSE; ASSERT(txg > spa_last_synced_txg(spa)); spa->spa_pending_vdev = NULL; /* * Reassess the DTLs. */ vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE); if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) { config_changed = B_TRUE; spa->spa_config_generation++; } /* * Verify the metaslab classes. */ ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0); ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0); spa_config_exit(spa, SCL_ALL, spa); /* * Panic the system if the specified tag requires it. This * is useful for ensuring that configurations are updated * transactionally. */ if (zio_injection_enabled) zio_handle_panic_injection(spa, tag, 0); /* * Note: this txg_wait_synced() is important because it ensures * that there won't be more than one config change per txg. * This allows us to use the txg as the generation number. */ if (error == 0) txg_wait_synced(spa->spa_dsl_pool, txg); if (vd != NULL) { ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL); spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); vdev_free(vd); spa_config_exit(spa, SCL_ALL, spa); } /* * If the config changed, update the config cache. */ if (config_changed) spa_write_cachefile(spa, B_FALSE, B_TRUE); } /* * Unlock the spa_t after adding or removing a vdev. Besides undoing the * locking of spa_vdev_enter(), we also want make sure the transactions have * synced to disk, and then update the global configuration cache with the new * information. */ int spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error) { spa_vdev_config_exit(spa, vd, txg, error, FTAG); mutex_exit(&spa_namespace_lock); mutex_exit(&spa->spa_vdev_top_lock); return (error); } /* * Lock the given spa_t for the purpose of changing vdev state. */ void spa_vdev_state_enter(spa_t *spa, int oplocks) { int locks = SCL_STATE_ALL | oplocks; /* * Root pools may need to read of the underlying devfs filesystem * when opening up a vdev. Unfortunately if we're holding the * SCL_ZIO lock it will result in a deadlock when we try to issue * the read from the root filesystem. Instead we "prefetch" * the associated vnodes that we need prior to opening the * underlying devices and cache them so that we can prevent * any I/O when we are doing the actual open. */ if (spa_is_root(spa)) { int low = locks & ~(SCL_ZIO - 1); int high = locks & ~low; spa_config_enter(spa, high, spa, RW_WRITER); vdev_hold(spa->spa_root_vdev); spa_config_enter(spa, low, spa, RW_WRITER); } else { spa_config_enter(spa, locks, spa, RW_WRITER); } spa->spa_vdev_locks = locks; } int spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error) { boolean_t config_changed = B_FALSE; if (vd != NULL || error == 0) vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev, 0, 0, B_FALSE); if (vd != NULL) { vdev_state_dirty(vd->vdev_top); config_changed = B_TRUE; spa->spa_config_generation++; } if (spa_is_root(spa)) vdev_rele(spa->spa_root_vdev); ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL); spa_config_exit(spa, spa->spa_vdev_locks, spa); /* * If anything changed, wait for it to sync. This ensures that, * from the system administrator's perspective, zpool(1M) commands * are synchronous. This is important for things like zpool offline: * when the command completes, you expect no further I/O from ZFS. */ if (vd != NULL) txg_wait_synced(spa->spa_dsl_pool, 0); /* * If the config changed, update the config cache. */ if (config_changed) { mutex_enter(&spa_namespace_lock); spa_write_cachefile(spa, B_FALSE, B_TRUE); mutex_exit(&spa_namespace_lock); } return (error); } /* * ========================================================================== * Miscellaneous functions * ========================================================================== */ void spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx) { if (!nvlist_exists(spa->spa_label_features, feature)) { fnvlist_add_boolean(spa->spa_label_features, feature); /* * When we are creating the pool (tx_txg==TXG_INITIAL), we can't * dirty the vdev config because lock SCL_CONFIG is not held. * Thankfully, in this case we don't need to dirty the config * because it will be written out anyway when we finish * creating the pool. */ if (tx->tx_txg != TXG_INITIAL) vdev_config_dirty(spa->spa_root_vdev); } } void spa_deactivate_mos_feature(spa_t *spa, const char *feature) { if (nvlist_remove_all(spa->spa_label_features, feature) == 0) vdev_config_dirty(spa->spa_root_vdev); } /* * Rename a spa_t. */ int spa_rename(const char *name, const char *newname) { spa_t *spa; int err; /* * Lookup the spa_t and grab the config lock for writing. We need to * actually open the pool so that we can sync out the necessary labels. * It's OK to call spa_open() with the namespace lock held because we * allow recursive calls for other reasons. */ mutex_enter(&spa_namespace_lock); if ((err = spa_open(name, &spa, FTAG)) != 0) { mutex_exit(&spa_namespace_lock); return (err); } spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); avl_remove(&spa_namespace_avl, spa); (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name)); avl_add(&spa_namespace_avl, spa); /* * Sync all labels to disk with the new names by marking the root vdev * dirty and waiting for it to sync. It will pick up the new pool name * during the sync. */ vdev_config_dirty(spa->spa_root_vdev); spa_config_exit(spa, SCL_ALL, FTAG); txg_wait_synced(spa->spa_dsl_pool, 0); /* * Sync the updated config cache. */ spa_write_cachefile(spa, B_FALSE, B_TRUE); spa_close(spa, FTAG); mutex_exit(&spa_namespace_lock); return (0); } /* * Return the spa_t associated with given pool_guid, if it exists. If * device_guid is non-zero, determine whether the pool exists *and* contains * a device with the specified device_guid. */ spa_t * spa_by_guid(uint64_t pool_guid, uint64_t device_guid) { spa_t *spa; avl_tree_t *t = &spa_namespace_avl; ASSERT(MUTEX_HELD(&spa_namespace_lock)); for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) { if (spa->spa_state == POOL_STATE_UNINITIALIZED) continue; if (spa->spa_root_vdev == NULL) continue; if (spa_guid(spa) == pool_guid) { if (device_guid == 0) break; if (vdev_lookup_by_guid(spa->spa_root_vdev, device_guid) != NULL) break; /* * Check any devices we may be in the process of adding. */ if (spa->spa_pending_vdev) { if (vdev_lookup_by_guid(spa->spa_pending_vdev, device_guid) != NULL) break; } } } return (spa); } /* * Determine whether a pool with the given pool_guid exists. */ boolean_t spa_guid_exists(uint64_t pool_guid, uint64_t device_guid) { return (spa_by_guid(pool_guid, device_guid) != NULL); } char * spa_strdup(const char *s) { size_t len; char *new; len = strlen(s); new = kmem_alloc(len + 1, KM_SLEEP); bcopy(s, new, len); new[len] = '\0'; return (new); } void spa_strfree(char *s) { kmem_free(s, strlen(s) + 1); } uint64_t spa_get_random(uint64_t range) { uint64_t r; ASSERT(range != 0); (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t)); return (r % range); } uint64_t spa_generate_guid(spa_t *spa) { uint64_t guid = spa_get_random(-1ULL); if (spa != NULL) { while (guid == 0 || spa_guid_exists(spa_guid(spa), guid)) guid = spa_get_random(-1ULL); } else { while (guid == 0 || spa_guid_exists(guid, 0)) guid = spa_get_random(-1ULL); } return (guid); } void snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp) { char type[256]; char *checksum = NULL; char *compress = NULL; if (bp != NULL) { if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) { dmu_object_byteswap_t bswap = DMU_OT_BYTESWAP(BP_GET_TYPE(bp)); (void) snprintf(type, sizeof (type), "bswap %s %s", DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ? "metadata" : "data", dmu_ot_byteswap[bswap].ob_name); } else { (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name, sizeof (type)); } if (!BP_IS_EMBEDDED(bp)) { checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name; } compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name; } SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum, compress); } void spa_freeze(spa_t *spa) { uint64_t freeze_txg = 0; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); if (spa->spa_freeze_txg == UINT64_MAX) { freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE; spa->spa_freeze_txg = freeze_txg; } spa_config_exit(spa, SCL_ALL, FTAG); if (freeze_txg != 0) txg_wait_synced(spa_get_dsl(spa), freeze_txg); } void zfs_panic_recover(const char *fmt, ...) { va_list adx; va_start(adx, fmt); vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx); va_end(adx); } /* * This is a stripped-down version of strtoull, suitable only for converting * lowercase hexadecimal numbers that don't overflow. */ uint64_t zfs_strtonum(const char *str, char **nptr) { uint64_t val = 0; char c; int digit; while ((c = *str) != '\0') { if (c >= '0' && c <= '9') digit = c - '0'; else if (c >= 'a' && c <= 'f') digit = 10 + c - 'a'; else break; val *= 16; val += digit; str++; } if (nptr) *nptr = (char *)str; return (val); } /* * ========================================================================== * Accessor functions * ========================================================================== */ boolean_t spa_shutting_down(spa_t *spa) { return (spa->spa_async_suspended); } dsl_pool_t * spa_get_dsl(spa_t *spa) { return (spa->spa_dsl_pool); } boolean_t spa_is_initializing(spa_t *spa) { return (spa->spa_is_initializing); } boolean_t spa_indirect_vdevs_loaded(spa_t *spa) { return (spa->spa_indirect_vdevs_loaded); } blkptr_t * spa_get_rootblkptr(spa_t *spa) { return (&spa->spa_ubsync.ub_rootbp); } void spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp) { spa->spa_uberblock.ub_rootbp = *bp; } void spa_altroot(spa_t *spa, char *buf, size_t buflen) { if (spa->spa_root == NULL) buf[0] = '\0'; else (void) strncpy(buf, spa->spa_root, buflen); } int spa_sync_pass(spa_t *spa) { return (spa->spa_sync_pass); } char * spa_name(spa_t *spa) { return (spa->spa_name); } uint64_t spa_guid(spa_t *spa) { dsl_pool_t *dp = spa_get_dsl(spa); uint64_t guid; /* * If we fail to parse the config during spa_load(), we can go through * the error path (which posts an ereport) and end up here with no root * vdev. We stash the original pool guid in 'spa_config_guid' to handle * this case. */ if (spa->spa_root_vdev == NULL) return (spa->spa_config_guid); guid = spa->spa_last_synced_guid != 0 ? spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid; /* * Return the most recently synced out guid unless we're * in syncing context. */ if (dp && dsl_pool_sync_context(dp)) return (spa->spa_root_vdev->vdev_guid); else return (guid); } uint64_t spa_load_guid(spa_t *spa) { /* * This is a GUID that exists solely as a reference for the * purposes of the arc. It is generated at load time, and * is never written to persistent storage. */ return (spa->spa_load_guid); } uint64_t spa_last_synced_txg(spa_t *spa) { return (spa->spa_ubsync.ub_txg); } uint64_t spa_first_txg(spa_t *spa) { return (spa->spa_first_txg); } uint64_t spa_syncing_txg(spa_t *spa) { return (spa->spa_syncing_txg); } /* * Return the last txg where data can be dirtied. The final txgs * will be used to just clear out any deferred frees that remain. */ uint64_t spa_final_dirty_txg(spa_t *spa) { return (spa->spa_final_txg - TXG_DEFER_SIZE); } pool_state_t spa_state(spa_t *spa) { return (spa->spa_state); } spa_load_state_t spa_load_state(spa_t *spa) { return (spa->spa_load_state); } uint64_t spa_freeze_txg(spa_t *spa) { return (spa->spa_freeze_txg); } /* ARGSUSED */ uint64_t spa_get_worst_case_asize(spa_t *spa, uint64_t lsize) { return (lsize * spa_asize_inflation); } /* * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%), * or at least 128MB, unless that would cause it to be more than half the * pool size. * * See the comment above spa_slop_shift for details. */ uint64_t spa_get_slop_space(spa_t *spa) { uint64_t space = spa_get_dspace(spa); return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop))); } uint64_t spa_get_dspace(spa_t *spa) { return (spa->spa_dspace); } uint64_t spa_get_checkpoint_space(spa_t *spa) { return (spa->spa_checkpoint_info.sci_dspace); } void spa_update_dspace(spa_t *spa) { spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) + ddt_get_dedup_dspace(spa); if (spa->spa_vdev_removal != NULL) { /* * We can't allocate from the removing device, so * subtract its size. This prevents the DMU/DSL from * filling up the (now smaller) pool while we are in the * middle of removing the device. * * Note that the DMU/DSL doesn't actually know or care * how much space is allocated (it does its own tracking * of how much space has been logically used). So it * doesn't matter that the data we are moving may be * allocated twice (on the old device and the new * device). */ vdev_t *vd = spa->spa_vdev_removal->svr_vdev; spa->spa_dspace -= spa_deflate(spa) ? vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space; } } /* * Return the failure mode that has been set to this pool. The default * behavior will be to block all I/Os when a complete failure occurs. */ uint8_t spa_get_failmode(spa_t *spa) { return (spa->spa_failmode); } boolean_t spa_suspended(spa_t *spa) { return (spa->spa_suspended); } uint64_t spa_version(spa_t *spa) { return (spa->spa_ubsync.ub_version); } boolean_t spa_deflate(spa_t *spa) { return (spa->spa_deflate); } metaslab_class_t * spa_normal_class(spa_t *spa) { return (spa->spa_normal_class); } metaslab_class_t * spa_log_class(spa_t *spa) { return (spa->spa_log_class); } void spa_evicting_os_register(spa_t *spa, objset_t *os) { mutex_enter(&spa->spa_evicting_os_lock); list_insert_head(&spa->spa_evicting_os_list, os); mutex_exit(&spa->spa_evicting_os_lock); } void spa_evicting_os_deregister(spa_t *spa, objset_t *os) { mutex_enter(&spa->spa_evicting_os_lock); list_remove(&spa->spa_evicting_os_list, os); cv_broadcast(&spa->spa_evicting_os_cv); mutex_exit(&spa->spa_evicting_os_lock); } void spa_evicting_os_wait(spa_t *spa) { mutex_enter(&spa->spa_evicting_os_lock); while (!list_is_empty(&spa->spa_evicting_os_list)) cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock); mutex_exit(&spa->spa_evicting_os_lock); dmu_buf_user_evict_wait(); } int spa_max_replication(spa_t *spa) { /* * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to * handle BPs with more than one DVA allocated. Set our max * replication level accordingly. */ if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS) return (1); return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override)); } int spa_prev_software_version(spa_t *spa) { return (spa->spa_prev_software_version); } uint64_t spa_deadman_synctime(spa_t *spa) { return (spa->spa_deadman_synctime); } uint64_t dva_get_dsize_sync(spa_t *spa, const dva_t *dva) { uint64_t asize = DVA_GET_ASIZE(dva); uint64_t dsize = asize; ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); if (asize != 0 && spa->spa_deflate) { uint64_t vdev = DVA_GET_VDEV(dva); vdev_t *vd = vdev_lookup_top(spa, vdev); if (vd == NULL) { panic( "dva_get_dsize_sync(): bad DVA %llu:%llu", (u_longlong_t)vdev, (u_longlong_t)asize); } dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; } return (dsize); } uint64_t bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp) { uint64_t dsize = 0; for (int d = 0; d < BP_GET_NDVAS(bp); d++) dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); return (dsize); } uint64_t bp_get_dsize(spa_t *spa, const blkptr_t *bp) { uint64_t dsize = 0; spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); for (int d = 0; d < BP_GET_NDVAS(bp); d++) dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]); spa_config_exit(spa, SCL_VDEV, FTAG); return (dsize); } /* * ========================================================================== * Initialization and Termination * ========================================================================== */ static int spa_name_compare(const void *a1, const void *a2) { const spa_t *s1 = a1; const spa_t *s2 = a2; int s; s = strcmp(s1->spa_name, s2->spa_name); if (s > 0) return (1); if (s < 0) return (-1); return (0); } int spa_busy(void) { return (spa_active_count); } void spa_boot_init() { spa_config_load(); } #ifdef _KERNEL EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0); #endif void spa_init(int mode) { mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL); avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t), offsetof(spa_t, spa_avl)); avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t), offsetof(spa_aux_t, aux_avl)); avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t), offsetof(spa_aux_t, aux_avl)); spa_mode_global = mode; #ifdef illumos #ifdef _KERNEL spa_arch_init(); #else if (spa_mode_global != FREAD && dprintf_find_string("watch")) { arc_procfd = open("/proc/self/ctl", O_WRONLY); if (arc_procfd == -1) { perror("could not enable watchpoints: " "opening /proc/self/ctl failed: "); } else { arc_watch = B_TRUE; } } #endif #endif /* illumos */ refcount_sysinit(); unique_init(); range_tree_init(); metaslab_alloc_trace_init(); zio_init(); lz4_init(); dmu_init(); zil_init(); vdev_cache_stat_init(); vdev_file_init(); zfs_prop_init(); zpool_prop_init(); zpool_feature_init(); spa_config_load(); l2arc_start(); scan_init(); dsl_scan_global_init(); #ifndef illumos #ifdef _KERNEL zfs_deadman_init(); #endif #endif /* !illumos */ } void spa_fini(void) { l2arc_stop(); spa_evict_all(); vdev_file_fini(); vdev_cache_stat_fini(); zil_fini(); dmu_fini(); lz4_fini(); zio_fini(); metaslab_alloc_trace_fini(); range_tree_fini(); unique_fini(); refcount_fini(); scan_fini(); avl_destroy(&spa_namespace_avl); avl_destroy(&spa_spare_avl); avl_destroy(&spa_l2cache_avl); cv_destroy(&spa_namespace_cv); mutex_destroy(&spa_namespace_lock); mutex_destroy(&spa_spare_lock); mutex_destroy(&spa_l2cache_lock); } /* * Return whether this pool has slogs. No locking needed. * It's not a problem if the wrong answer is returned as it's only for * performance and not correctness */ boolean_t spa_has_slogs(spa_t *spa) { return (spa->spa_log_class->mc_rotor != NULL); } spa_log_state_t spa_get_log_state(spa_t *spa) { return (spa->spa_log_state); } void spa_set_log_state(spa_t *spa, spa_log_state_t state) { spa->spa_log_state = state; } boolean_t spa_is_root(spa_t *spa) { return (spa->spa_is_root); } boolean_t spa_writeable(spa_t *spa) { return (!!(spa->spa_mode & FWRITE) && spa->spa_trust_config); } /* * Returns true if there is a pending sync task in any of the current * syncing txg, the current quiescing txg, or the current open txg. */ boolean_t spa_has_pending_synctask(spa_t *spa) { return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks) || !txg_all_lists_empty(&spa->spa_dsl_pool->dp_early_sync_tasks)); } int spa_mode(spa_t *spa) { return (spa->spa_mode); } uint64_t spa_bootfs(spa_t *spa) { return (spa->spa_bootfs); } uint64_t spa_delegation(spa_t *spa) { return (spa->spa_delegation); } objset_t * spa_meta_objset(spa_t *spa) { return (spa->spa_meta_objset); } enum zio_checksum spa_dedup_checksum(spa_t *spa) { return (spa->spa_dedup_checksum); } /* * Reset pool scan stat per scan pass (or reboot). */ void spa_scan_stat_init(spa_t *spa) { /* data not stored on disk */ spa->spa_scan_pass_start = gethrestime_sec(); if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan)) spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start; else spa->spa_scan_pass_scrub_pause = 0; spa->spa_scan_pass_scrub_spent_paused = 0; spa->spa_scan_pass_exam = 0; spa->spa_scan_pass_issued = 0; vdev_scan_stat_init(spa->spa_root_vdev); } /* * Get scan stats for zpool status reports */ int spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps) { dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL; if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE) return (SET_ERROR(ENOENT)); bzero(ps, sizeof (pool_scan_stat_t)); /* data stored on disk */ ps->pss_func = scn->scn_phys.scn_func; ps->pss_state = scn->scn_phys.scn_state; ps->pss_start_time = scn->scn_phys.scn_start_time; ps->pss_end_time = scn->scn_phys.scn_end_time; ps->pss_to_examine = scn->scn_phys.scn_to_examine; ps->pss_to_process = scn->scn_phys.scn_to_process; ps->pss_processed = scn->scn_phys.scn_processed; ps->pss_errors = scn->scn_phys.scn_errors; ps->pss_examined = scn->scn_phys.scn_examined; ps->pss_issued = scn->scn_issued_before_pass + spa->spa_scan_pass_issued; /* data not stored on disk */ ps->pss_pass_start = spa->spa_scan_pass_start; ps->pss_pass_exam = spa->spa_scan_pass_exam; ps->pss_pass_issued = spa->spa_scan_pass_issued; ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause; ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused; return (0); } boolean_t spa_debug_enabled(spa_t *spa) { return (spa->spa_debug); } int spa_maxblocksize(spa_t *spa) { if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) return (SPA_MAXBLOCKSIZE); else return (SPA_OLD_MAXBLOCKSIZE); } /* * Returns the txg that the last device removal completed. No indirect mappings * have been added since this txg. */ uint64_t spa_get_last_removal_txg(spa_t *spa) { uint64_t vdevid; uint64_t ret = -1ULL; spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); /* * sr_prev_indirect_vdev is only modified while holding all the * config locks, so it is sufficient to hold SCL_VDEV as reader when * examining it. */ vdevid = spa->spa_removing_phys.sr_prev_indirect_vdev; while (vdevid != -1ULL) { vdev_t *vd = vdev_lookup_top(spa, vdevid); vdev_indirect_births_t *vib = vd->vdev_indirect_births; ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops); /* * If the removal did not remap any data, we don't care. */ if (vdev_indirect_births_count(vib) != 0) { ret = vdev_indirect_births_last_entry_txg(vib); break; } vdevid = vd->vdev_indirect_config.vic_prev_indirect_vdev; } spa_config_exit(spa, SCL_VDEV, FTAG); IMPLY(ret != -1ULL, spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL)); return (ret); } boolean_t spa_trust_config(spa_t *spa) { return (spa->spa_trust_config); } uint64_t spa_missing_tvds_allowed(spa_t *spa) { return (spa->spa_missing_tvds_allowed); } void spa_set_missing_tvds(spa_t *spa, uint64_t missing) { spa->spa_missing_tvds = missing; } boolean_t spa_top_vdevs_spacemap_addressable(spa_t *spa) { vdev_t *rvd = spa->spa_root_vdev; for (uint64_t c = 0; c < rvd->vdev_children; c++) { if (!vdev_is_spacemap_addressable(rvd->vdev_child[c])) return (B_FALSE); } return (B_TRUE); } boolean_t spa_has_checkpoint(spa_t *spa) { return (spa->spa_checkpoint_txg != 0); } boolean_t spa_importing_readonly_checkpoint(spa_t *spa) { return ((spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT) && spa->spa_mode == FREAD); } uint64_t spa_min_claim_txg(spa_t *spa) { uint64_t checkpoint_txg = spa->spa_uberblock.ub_checkpoint_txg; if (checkpoint_txg != 0) return (checkpoint_txg + 1); return (spa->spa_first_txg); } /* * If there is a checkpoint, async destroys may consume more space from * the pool instead of freeing it. In an attempt to save the pool from * getting suspended when it is about to run out of space, we stop * processing async destroys. */ boolean_t spa_suspend_async_destroy(spa_t *spa) { dsl_pool_t *dp = spa_get_dsl(spa); uint64_t unreserved = dsl_pool_unreserved_space(dp, ZFS_SPACE_CHECK_EXTRA_RESERVED); uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes; uint64_t avail = (unreserved > used) ? (unreserved - used) : 0; if (spa_has_checkpoint(spa) && avail == 0) return (B_TRUE); return (B_FALSE); } Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/metaslab.h =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/metaslab.h (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/metaslab.h (revision 336949) @@ -1,119 +1,121 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011, 2017 by Delphix. All rights reserved. + * Copyright (c) 2011, 2018 by Delphix. All rights reserved. */ #ifndef _SYS_METASLAB_H #define _SYS_METASLAB_H #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif typedef struct metaslab_ops { uint64_t (*msop_alloc)(metaslab_t *, uint64_t); } metaslab_ops_t; extern metaslab_ops_t *zfs_metaslab_ops; int metaslab_init(metaslab_group_t *, uint64_t, uint64_t, uint64_t, metaslab_t **); void metaslab_fini(metaslab_t *); void metaslab_load_wait(metaslab_t *); int metaslab_load(metaslab_t *); void metaslab_unload(metaslab_t *); void metaslab_sync(metaslab_t *, uint64_t); void metaslab_sync_done(metaslab_t *, uint64_t); void metaslab_sync_reassess(metaslab_group_t *); uint64_t metaslab_block_maxsize(metaslab_t *); #define METASLAB_HINTBP_FAVOR 0x0 #define METASLAB_HINTBP_AVOID 0x1 #define METASLAB_GANG_HEADER 0x2 #define METASLAB_GANG_CHILD 0x4 #define METASLAB_ASYNC_ALLOC 0x8 #define METASLAB_DONT_THROTTLE 0x10 int metaslab_alloc(spa_t *, metaslab_class_t *, uint64_t, - blkptr_t *, int, uint64_t, blkptr_t *, int, zio_alloc_list_t *, zio_t *); + blkptr_t *, int, uint64_t, blkptr_t *, int, zio_alloc_list_t *, zio_t *, + int); int metaslab_alloc_dva(spa_t *, metaslab_class_t *, uint64_t, - dva_t *, int, dva_t *, uint64_t, int, zio_alloc_list_t *); + dva_t *, int, dva_t *, uint64_t, int, zio_alloc_list_t *, int); void metaslab_free(spa_t *, const blkptr_t *, uint64_t, boolean_t); void metaslab_free_concrete(vdev_t *, uint64_t, uint64_t, boolean_t); void metaslab_free_dva(spa_t *, const dva_t *, boolean_t); void metaslab_free_impl_cb(uint64_t, vdev_t *, uint64_t, uint64_t, void *); void metaslab_unalloc_dva(spa_t *, const dva_t *, uint64_t); int metaslab_claim(spa_t *, const blkptr_t *, uint64_t); int metaslab_claim_impl(vdev_t *, uint64_t, uint64_t, uint64_t); void metaslab_check_free(spa_t *, const blkptr_t *); void metaslab_alloc_trace_init(void); void metaslab_alloc_trace_fini(void); void metaslab_trace_init(zio_alloc_list_t *); void metaslab_trace_fini(zio_alloc_list_t *); metaslab_class_t *metaslab_class_create(spa_t *, metaslab_ops_t *); void metaslab_class_destroy(metaslab_class_t *); int metaslab_class_validate(metaslab_class_t *); void metaslab_class_histogram_verify(metaslab_class_t *); uint64_t metaslab_class_fragmentation(metaslab_class_t *); uint64_t metaslab_class_expandable_space(metaslab_class_t *); -boolean_t metaslab_class_throttle_reserve(metaslab_class_t *, int, +boolean_t metaslab_class_throttle_reserve(metaslab_class_t *, int, int, zio_t *, int); -void metaslab_class_throttle_unreserve(metaslab_class_t *, int, zio_t *); +void metaslab_class_throttle_unreserve(metaslab_class_t *, int, int, zio_t *); void metaslab_class_space_update(metaslab_class_t *, int64_t, int64_t, int64_t, int64_t); uint64_t metaslab_class_get_alloc(metaslab_class_t *); uint64_t metaslab_class_get_space(metaslab_class_t *); uint64_t metaslab_class_get_dspace(metaslab_class_t *); uint64_t metaslab_class_get_deferred(metaslab_class_t *); uint64_t metaslab_class_get_minblocksize(metaslab_class_t *mc); -metaslab_group_t *metaslab_group_create(metaslab_class_t *, vdev_t *); +metaslab_group_t *metaslab_group_create(metaslab_class_t *, vdev_t *, int); void metaslab_group_destroy(metaslab_group_t *); void metaslab_group_activate(metaslab_group_t *); void metaslab_group_passivate(metaslab_group_t *); boolean_t metaslab_group_initialized(metaslab_group_t *); uint64_t metaslab_group_get_space(metaslab_group_t *); void metaslab_group_histogram_verify(metaslab_group_t *); uint64_t metaslab_group_fragmentation(metaslab_group_t *); void metaslab_group_histogram_remove(metaslab_group_t *, metaslab_t *); -void metaslab_group_alloc_decrement(spa_t *, uint64_t, void *, int); -void metaslab_group_alloc_verify(spa_t *, const blkptr_t *, void *); +void metaslab_group_alloc_decrement(spa_t *, uint64_t, void *, int, int, + boolean_t); +void metaslab_group_alloc_verify(spa_t *, const blkptr_t *, void *, int); #ifdef __cplusplus } #endif #endif /* _SYS_METASLAB_H */ Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/metaslab_impl.h =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/metaslab_impl.h (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/metaslab_impl.h (revision 336949) @@ -1,379 +1,414 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* - * Copyright (c) 2011, 2017 by Delphix. All rights reserved. + * Copyright (c) 2011, 2018 by Delphix. All rights reserved. */ #ifndef _SYS_METASLAB_IMPL_H #define _SYS_METASLAB_IMPL_H #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /* * Metaslab allocation tracing record. */ typedef struct metaslab_alloc_trace { list_node_t mat_list_node; metaslab_group_t *mat_mg; metaslab_t *mat_msp; uint64_t mat_size; uint64_t mat_weight; uint32_t mat_dva_id; uint64_t mat_offset; + int mat_allocator; } metaslab_alloc_trace_t; /* * Used by the metaslab allocation tracing facility to indicate * error conditions. These errors are stored to the offset member * of the metaslab_alloc_trace_t record and displayed by mdb. */ typedef enum trace_alloc_type { TRACE_ALLOC_FAILURE = -1ULL, TRACE_TOO_SMALL = -2ULL, TRACE_FORCE_GANG = -3ULL, TRACE_NOT_ALLOCATABLE = -4ULL, TRACE_GROUP_FAILURE = -5ULL, TRACE_ENOSPC = -6ULL, TRACE_CONDENSING = -7ULL, TRACE_VDEV_ERROR = -8ULL } trace_alloc_type_t; #define METASLAB_WEIGHT_PRIMARY (1ULL << 63) #define METASLAB_WEIGHT_SECONDARY (1ULL << 62) -#define METASLAB_WEIGHT_TYPE (1ULL << 61) +#define METASLAB_WEIGHT_CLAIM (1ULL << 61) +#define METASLAB_WEIGHT_TYPE (1ULL << 60) #define METASLAB_ACTIVE_MASK \ - (METASLAB_WEIGHT_PRIMARY | METASLAB_WEIGHT_SECONDARY) + (METASLAB_WEIGHT_PRIMARY | METASLAB_WEIGHT_SECONDARY | \ + METASLAB_WEIGHT_CLAIM) /* * The metaslab weight is used to encode the amount of free space in a * metaslab, such that the "best" metaslab appears first when sorting the * metaslabs by weight. The weight (and therefore the "best" metaslab) can * be determined in two different ways: by computing a weighted sum of all * the free space in the metaslab (a space based weight) or by counting only * the free segments of the largest size (a segment based weight). We prefer * the segment based weight because it reflects how the free space is * comprised, but we cannot always use it -- legacy pools do not have the * space map histogram information necessary to determine the largest * contiguous regions. Pools that have the space map histogram determine * the segment weight by looking at each bucket in the histogram and * determining the free space whose size in bytes is in the range: * [2^i, 2^(i+1)) * We then encode the largest index, i, that contains regions into the * segment-weighted value. * * Space-based weight: * * 64 56 48 40 32 24 16 8 0 * +-------+-------+-------+-------+-------+-------+-------+-------+ - * |PS1| weighted-free space | + * |PSC1| weighted-free space | * +-------+-------+-------+-------+-------+-------+-------+-------+ * * PS - indicates primary and secondary activation + * C - indicates activation for claimed block zio * space - the fragmentation-weighted space * * Segment-based weight: * * 64 56 48 40 32 24 16 8 0 * +-------+-------+-------+-------+-------+-------+-------+-------+ - * |PS0| idx| count of segments in region | + * |PSC0| idx| count of segments in region | * +-------+-------+-------+-------+-------+-------+-------+-------+ * * PS - indicates primary and secondary activation + * C - indicates activation for claimed block zio * idx - index for the highest bucket in the histogram * count - number of segments in the specified bucket */ -#define WEIGHT_GET_ACTIVE(weight) BF64_GET((weight), 62, 2) -#define WEIGHT_SET_ACTIVE(weight, x) BF64_SET((weight), 62, 2, x) +#define WEIGHT_GET_ACTIVE(weight) BF64_GET((weight), 61, 3) +#define WEIGHT_SET_ACTIVE(weight, x) BF64_SET((weight), 61, 3, x) #define WEIGHT_IS_SPACEBASED(weight) \ - ((weight) == 0 || BF64_GET((weight), 61, 1)) -#define WEIGHT_SET_SPACEBASED(weight) BF64_SET((weight), 61, 1, 1) + ((weight) == 0 || BF64_GET((weight), 60, 1)) +#define WEIGHT_SET_SPACEBASED(weight) BF64_SET((weight), 60, 1, 1) /* * These macros are only applicable to segment-based weighting. */ -#define WEIGHT_GET_INDEX(weight) BF64_GET((weight), 55, 6) -#define WEIGHT_SET_INDEX(weight, x) BF64_SET((weight), 55, 6, x) -#define WEIGHT_GET_COUNT(weight) BF64_GET((weight), 0, 55) -#define WEIGHT_SET_COUNT(weight, x) BF64_SET((weight), 0, 55, x) +#define WEIGHT_GET_INDEX(weight) BF64_GET((weight), 54, 6) +#define WEIGHT_SET_INDEX(weight, x) BF64_SET((weight), 54, 6, x) +#define WEIGHT_GET_COUNT(weight) BF64_GET((weight), 0, 54) +#define WEIGHT_SET_COUNT(weight, x) BF64_SET((weight), 0, 54, x) /* * A metaslab class encompasses a category of allocatable top-level vdevs. * Each top-level vdev is associated with a metaslab group which defines * the allocatable region for that vdev. Examples of these categories include * "normal" for data block allocations (i.e. main pool allocations) or "log" * for allocations designated for intent log devices (i.e. slog devices). * When a block allocation is requested from the SPA it is associated with a * metaslab_class_t, and only top-level vdevs (i.e. metaslab groups) belonging * to the class can be used to satisfy that request. Allocations are done * by traversing the metaslab groups that are linked off of the mc_rotor field. * This rotor points to the next metaslab group where allocations will be * attempted. Allocating a block is a 3 step process -- select the metaslab * group, select the metaslab, and then allocate the block. The metaslab * class defines the low-level block allocator that will be used as the * final step in allocation. These allocators are pluggable allowing each class * to use a block allocator that best suits that class. */ struct metaslab_class { kmutex_t mc_lock; spa_t *mc_spa; metaslab_group_t *mc_rotor; metaslab_ops_t *mc_ops; uint64_t mc_aliquot; /* * Track the number of metaslab groups that have been initialized * and can accept allocations. An initialized metaslab group is * one has been completely added to the config (i.e. we have * updated the MOS config and the space has been added to the pool). */ uint64_t mc_groups; /* * Toggle to enable/disable the allocation throttle. */ boolean_t mc_alloc_throttle_enabled; /* * The allocation throttle works on a reservation system. Whenever * an asynchronous zio wants to perform an allocation it must * first reserve the number of blocks that it wants to allocate. * If there aren't sufficient slots available for the pending zio * then that I/O is throttled until more slots free up. The current * number of reserved allocations is maintained by the mc_alloc_slots * refcount. The mc_alloc_max_slots value determines the maximum * number of allocations that the system allows. Gang blocks are * allowed to reserve slots even if we've reached the maximum * number of allocations allowed. */ - uint64_t mc_alloc_max_slots; - refcount_t mc_alloc_slots; + uint64_t *mc_alloc_max_slots; + refcount_t *mc_alloc_slots; uint64_t mc_alloc_groups; /* # of allocatable groups */ uint64_t mc_alloc; /* total allocated space */ uint64_t mc_deferred; /* total deferred frees */ uint64_t mc_space; /* total space (alloc + free) */ uint64_t mc_dspace; /* total deflated space */ uint64_t mc_minblocksize; uint64_t mc_histogram[RANGE_TREE_HISTOGRAM_SIZE]; }; /* * Metaslab groups encapsulate all the allocatable regions (i.e. metaslabs) * of a top-level vdev. They are linked togther to form a circular linked * list and can belong to only one metaslab class. Metaslab groups may become * ineligible for allocations for a number of reasons such as limited free * space, fragmentation, or going offline. When this happens the allocator will * simply find the next metaslab group in the linked list and attempt * to allocate from that group instead. */ struct metaslab_group { kmutex_t mg_lock; + metaslab_t **mg_primaries; + metaslab_t **mg_secondaries; avl_tree_t mg_metaslab_tree; uint64_t mg_aliquot; boolean_t mg_allocatable; /* can we allocate? */ + uint64_t mg_ms_ready; /* * A metaslab group is considered to be initialized only after * we have updated the MOS config and added the space to the pool. * We only allow allocation attempts to a metaslab group if it * has been initialized. */ boolean_t mg_initialized; uint64_t mg_free_capacity; /* percentage free */ int64_t mg_bias; int64_t mg_activation_count; metaslab_class_t *mg_class; vdev_t *mg_vd; taskq_t *mg_taskq; metaslab_group_t *mg_prev; metaslab_group_t *mg_next; /* - * Each metaslab group can handle mg_max_alloc_queue_depth allocations - * which are tracked by mg_alloc_queue_depth. It's possible for a - * metaslab group to handle more allocations than its max. This - * can occur when gang blocks are required or when other groups - * are unable to handle their share of allocations. + * In order for the allocation throttle to function properly, we cannot + * have too many IOs going to each disk by default; the throttle + * operates by allocating more work to disks that finish quickly, so + * allocating larger chunks to each disk reduces its effectiveness. + * However, if the number of IOs going to each allocator is too small, + * we will not perform proper aggregation at the vdev_queue layer, + * also resulting in decreased performance. Therefore, we will use a + * ramp-up strategy. + * + * Each allocator in each metaslab group has a current queue depth + * (mg_alloc_queue_depth[allocator]) and a current max queue depth + * (mg_cur_max_alloc_queue_depth[allocator]), and each metaslab group + * has an absolute max queue depth (mg_max_alloc_queue_depth). We + * add IOs to an allocator until the mg_alloc_queue_depth for that + * allocator hits the cur_max. Every time an IO completes for a given + * allocator on a given metaslab group, we increment its cur_max until + * it reaches mg_max_alloc_queue_depth. The cur_max resets every txg to + * help protect against disks that decrease in performance over time. + * + * It's possible for an allocator to handle more allocations than + * its max. This can occur when gang blocks are required or when other + * groups are unable to handle their share of allocations. */ uint64_t mg_max_alloc_queue_depth; - refcount_t mg_alloc_queue_depth; - + uint64_t *mg_cur_max_alloc_queue_depth; + refcount_t *mg_alloc_queue_depth; + int mg_allocators; /* * A metalab group that can no longer allocate the minimum block * size will set mg_no_free_space. Once a metaslab group is out * of space then its share of work must be distributed to other * groups. */ boolean_t mg_no_free_space; uint64_t mg_allocations; uint64_t mg_failed_allocations; uint64_t mg_fragmentation; uint64_t mg_histogram[RANGE_TREE_HISTOGRAM_SIZE]; }; /* * This value defines the number of elements in the ms_lbas array. The value * of 64 was chosen as it covers all power of 2 buckets up to UINT64_MAX. * This is the equivalent of highbit(UINT64_MAX). */ #define MAX_LBAS 64 /* * Each metaslab maintains a set of in-core trees to track metaslab * operations. The in-core free tree (ms_allocatable) contains the list of * free segments which are eligible for allocation. As blocks are * allocated, the allocated segment are removed from the ms_allocatable and * added to a per txg allocation tree (ms_allocating). As blocks are * freed, they are added to the free tree (ms_freeing). These trees * allow us to process all allocations and frees in syncing context * where it is safe to update the on-disk space maps. An additional set * of in-core trees is maintained to track deferred frees * (ms_defer). Once a block is freed it will move from the * ms_freed to the ms_defer tree. A deferred free means that a block * has been freed but cannot be used by the pool until TXG_DEFER_SIZE * transactions groups later. For example, a block that is freed in txg * 50 will not be available for reallocation until txg 52 (50 + * TXG_DEFER_SIZE). This provides a safety net for uberblock rollback. * A pool could be safely rolled back TXG_DEFERS_SIZE transactions * groups and ensure that no block has been reallocated. * * The simplified transition diagram looks like this: * * * ALLOCATE * | * V * free segment (ms_allocatable) -> ms_allocating[4] -> (write to space map) * ^ * | ms_freeing <--- FREE * | | * | v * | ms_freed * | | * +-------- ms_defer[2] <-------+-------> (write to space map) * * * Each metaslab's space is tracked in a single space map in the MOS, * which is only updated in syncing context. Each time we sync a txg, * we append the allocs and frees from that txg to the space map. The * pool space is only updated once all metaslabs have finished syncing. * * To load the in-core free tree we read the space map from disk. This * object contains a series of alloc and free records that are combined * to make up the list of all free segments in this metaslab. These * segments are represented in-core by the ms_allocatable and are stored * in an AVL tree. * * As the space map grows (as a result of the appends) it will * eventually become space-inefficient. When the metaslab's in-core * free tree is zfs_condense_pct/100 times the size of the minimal * on-disk representation, we rewrite it in its minimized form. If a * metaslab needs to condense then we must set the ms_condensing flag to * ensure that allocations are not performed on the metaslab that is * being written. */ struct metaslab { kmutex_t ms_lock; kmutex_t ms_sync_lock; kcondvar_t ms_load_cv; space_map_t *ms_sm; uint64_t ms_id; uint64_t ms_start; uint64_t ms_size; uint64_t ms_fragmentation; range_tree_t *ms_allocating[TXG_SIZE]; range_tree_t *ms_allocatable; /* * The following range trees are accessed only from syncing context. * ms_free*tree only have entries while syncing, and are empty * between syncs. */ range_tree_t *ms_freeing; /* to free this syncing txg */ range_tree_t *ms_freed; /* already freed this syncing txg */ range_tree_t *ms_defer[TXG_DEFER_SIZE]; range_tree_t *ms_checkpointing; /* to add to the checkpoint */ boolean_t ms_condensing; /* condensing? */ boolean_t ms_condense_wanted; uint64_t ms_condense_checked_txg; /* * We must hold both ms_lock and ms_group->mg_lock in order to * modify ms_loaded. */ boolean_t ms_loaded; boolean_t ms_loading; int64_t ms_deferspace; /* sum of ms_defermap[] space */ uint64_t ms_weight; /* weight vs. others in group */ uint64_t ms_activation_weight; /* activation weight */ /* * Track of whenever a metaslab is selected for loading or allocation. * We use this value to determine how long the metaslab should * stay cached. */ uint64_t ms_selected_txg; uint64_t ms_alloc_txg; /* last successful alloc (debug only) */ uint64_t ms_max_size; /* maximum allocatable size */ /* + * -1 if it's not active in an allocator, otherwise set to the allocator + * this metaslab is active for. + */ + int ms_allocator; + boolean_t ms_primary; /* Only valid if ms_allocator is not -1 */ + + /* * The metaslab block allocators can optionally use a size-ordered * range tree and/or an array of LBAs. Not all allocators use * this functionality. The ms_allocatable_by_size should always * contain the same number of segments as the ms_allocatable. The * only difference is that the ms_allocatable_by_size is ordered by * segment sizes. */ avl_tree_t ms_allocatable_by_size; uint64_t ms_lbas[MAX_LBAS]; metaslab_group_t *ms_group; /* metaslab group */ avl_node_t ms_group_node; /* node in metaslab group tree */ txg_node_t ms_txg_node; /* per-txg dirty metaslab links */ + + boolean_t ms_new; }; #ifdef __cplusplus } #endif #endif /* _SYS_METASLAB_IMPL_H */ Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa_impl.h =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa_impl.h (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa_impl.h (revision 336949) @@ -1,410 +1,418 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2018 by Delphix. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. * Copyright 2013 Martin Matuska . All rights reserved. * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. * Copyright 2013 Saso Kiselkov. All rights reserved. * Copyright (c) 2017 Datto Inc. */ #ifndef _SYS_SPA_IMPL_H #define _SYS_SPA_IMPL_H #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif typedef struct spa_error_entry { zbookmark_phys_t se_bookmark; char *se_name; avl_node_t se_avl; } spa_error_entry_t; typedef struct spa_history_phys { uint64_t sh_pool_create_len; /* ending offset of zpool create */ uint64_t sh_phys_max_off; /* physical EOF */ uint64_t sh_bof; /* logical BOF */ uint64_t sh_eof; /* logical EOF */ uint64_t sh_records_lost; /* num of records overwritten */ } spa_history_phys_t; /* * All members must be uint64_t, for byteswap purposes. */ typedef struct spa_removing_phys { uint64_t sr_state; /* dsl_scan_state_t */ /* * The vdev ID that we most recently attempted to remove, * or -1 if no removal has been attempted. */ uint64_t sr_removing_vdev; /* * The vdev ID that we most recently successfully removed, * or -1 if no devices have been removed. */ uint64_t sr_prev_indirect_vdev; uint64_t sr_start_time; uint64_t sr_end_time; /* * Note that we can not use the space map's or indirect mapping's * accounting as a substitute for these values, because we need to * count frees of not-yet-copied data as though it did the copy. * Otherwise, we could get into a situation where copied > to_copy, * or we complete before copied == to_copy. */ uint64_t sr_to_copy; /* bytes that need to be copied */ uint64_t sr_copied; /* bytes that have been copied or freed */ } spa_removing_phys_t; /* * This struct is stored as an entry in the DMU_POOL_DIRECTORY_OBJECT * (with key DMU_POOL_CONDENSING_INDIRECT). It is present if a condense * of an indirect vdev's mapping object is in progress. */ typedef struct spa_condensing_indirect_phys { /* * The vdev ID of the indirect vdev whose indirect mapping is * being condensed. */ uint64_t scip_vdev; /* * The vdev's old obsolete spacemap. This spacemap's contents are * being integrated into the new mapping. */ uint64_t scip_prev_obsolete_sm_object; /* * The new mapping object that is being created. */ uint64_t scip_next_mapping_object; } spa_condensing_indirect_phys_t; struct spa_aux_vdev { uint64_t sav_object; /* MOS object for device list */ nvlist_t *sav_config; /* cached device config */ vdev_t **sav_vdevs; /* devices */ int sav_count; /* number devices */ boolean_t sav_sync; /* sync the device list */ nvlist_t **sav_pending; /* pending device additions */ uint_t sav_npending; /* # pending devices */ }; typedef struct spa_config_lock { kmutex_t scl_lock; kthread_t *scl_writer; int scl_write_wanted; kcondvar_t scl_cv; refcount_t scl_count; } spa_config_lock_t; typedef struct spa_config_dirent { list_node_t scd_link; char *scd_path; } spa_config_dirent_t; typedef enum zio_taskq_type { ZIO_TASKQ_ISSUE = 0, ZIO_TASKQ_ISSUE_HIGH, ZIO_TASKQ_INTERRUPT, ZIO_TASKQ_INTERRUPT_HIGH, ZIO_TASKQ_TYPES } zio_taskq_type_t; /* * State machine for the zpool-poolname process. The states transitions * are done as follows: * * From To Routine * PROC_NONE -> PROC_CREATED spa_activate() * PROC_CREATED -> PROC_ACTIVE spa_thread() * PROC_ACTIVE -> PROC_DEACTIVATE spa_deactivate() * PROC_DEACTIVATE -> PROC_GONE spa_thread() * PROC_GONE -> PROC_NONE spa_deactivate() */ typedef enum spa_proc_state { SPA_PROC_NONE, /* spa_proc = &p0, no process created */ SPA_PROC_CREATED, /* spa_activate() has proc, is waiting */ SPA_PROC_ACTIVE, /* taskqs created, spa_proc set */ SPA_PROC_DEACTIVATE, /* spa_deactivate() requests process exit */ SPA_PROC_GONE /* spa_thread() is exiting, spa_proc = &p0 */ } spa_proc_state_t; typedef struct spa_taskqs { uint_t stqs_count; taskq_t **stqs_taskq; } spa_taskqs_t; typedef enum spa_all_vdev_zap_action { AVZ_ACTION_NONE = 0, AVZ_ACTION_DESTROY, /* Destroy all per-vdev ZAPs and the AVZ. */ AVZ_ACTION_REBUILD, /* Populate the new AVZ, see spa_avz_rebuild */ AVZ_ACTION_INITIALIZE } spa_avz_action_t; typedef enum spa_config_source { SPA_CONFIG_SRC_NONE = 0, SPA_CONFIG_SRC_SCAN, /* scan of path (default: /dev/dsk) */ SPA_CONFIG_SRC_CACHEFILE, /* any cachefile */ SPA_CONFIG_SRC_TRYIMPORT, /* returned from call to tryimport */ SPA_CONFIG_SRC_SPLIT, /* new pool in a pool split */ SPA_CONFIG_SRC_MOS /* MOS, but not always from right txg */ } spa_config_source_t; struct spa { /* * Fields protected by spa_namespace_lock. */ char spa_name[ZFS_MAX_DATASET_NAME_LEN]; /* pool name */ char *spa_comment; /* comment */ avl_node_t spa_avl; /* node in spa_namespace_avl */ nvlist_t *spa_config; /* last synced config */ nvlist_t *spa_config_syncing; /* currently syncing config */ nvlist_t *spa_config_splitting; /* config for splitting */ nvlist_t *spa_load_info; /* info and errors from load */ uint64_t spa_config_txg; /* txg of last config change */ int spa_sync_pass; /* iterate-to-convergence */ pool_state_t spa_state; /* pool state */ int spa_inject_ref; /* injection references */ uint8_t spa_sync_on; /* sync threads are running */ spa_load_state_t spa_load_state; /* current load operation */ boolean_t spa_indirect_vdevs_loaded; /* mappings loaded? */ boolean_t spa_trust_config; /* do we trust vdev tree? */ spa_config_source_t spa_config_source; /* where config comes from? */ uint64_t spa_import_flags; /* import specific flags */ spa_taskqs_t spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES]; dsl_pool_t *spa_dsl_pool; boolean_t spa_is_initializing; /* true while opening pool */ metaslab_class_t *spa_normal_class; /* normal data class */ metaslab_class_t *spa_log_class; /* intent log data class */ uint64_t spa_first_txg; /* first txg after spa_open() */ uint64_t spa_final_txg; /* txg of export/destroy */ uint64_t spa_freeze_txg; /* freeze pool at this txg */ uint64_t spa_load_max_txg; /* best initial ub_txg */ uint64_t spa_claim_max_txg; /* highest claimed birth txg */ timespec_t spa_loaded_ts; /* 1st successful open time */ objset_t *spa_meta_objset; /* copy of dp->dp_meta_objset */ kmutex_t spa_evicting_os_lock; /* Evicting objset list lock */ list_t spa_evicting_os_list; /* Objsets being evicted. */ kcondvar_t spa_evicting_os_cv; /* Objset Eviction Completion */ txg_list_t spa_vdev_txg_list; /* per-txg dirty vdev list */ vdev_t *spa_root_vdev; /* top-level vdev container */ int spa_min_ashift; /* of vdevs in normal class */ int spa_max_ashift; /* of vdevs in normal class */ uint64_t spa_config_guid; /* config pool guid */ uint64_t spa_load_guid; /* spa_load initialized guid */ uint64_t spa_last_synced_guid; /* last synced guid */ list_t spa_config_dirty_list; /* vdevs with dirty config */ list_t spa_state_dirty_list; /* vdevs with dirty state */ - kmutex_t spa_alloc_lock; - avl_tree_t spa_alloc_tree; + /* + * spa_alloc_locks and spa_alloc_trees are arrays, whose lengths are + * stored in spa_alloc_count. There is one tree and one lock for each + * allocator, to help improve allocation performance in write-heavy + * workloads. + */ + kmutex_t *spa_alloc_locks; + avl_tree_t *spa_alloc_trees; + int spa_alloc_count; + spa_aux_vdev_t spa_spares; /* hot spares */ spa_aux_vdev_t spa_l2cache; /* L2ARC cache devices */ nvlist_t *spa_label_features; /* Features for reading MOS */ uint64_t spa_config_object; /* MOS object for pool config */ uint64_t spa_config_generation; /* config generation number */ uint64_t spa_syncing_txg; /* txg currently syncing */ bpobj_t spa_deferred_bpobj; /* deferred-free bplist */ bplist_t spa_free_bplist[TXG_SIZE]; /* bplist of stuff to free */ zio_cksum_salt_t spa_cksum_salt; /* secret salt for cksum */ /* checksum context templates */ kmutex_t spa_cksum_tmpls_lock; void *spa_cksum_tmpls[ZIO_CHECKSUM_FUNCTIONS]; uberblock_t spa_ubsync; /* last synced uberblock */ uberblock_t spa_uberblock; /* current uberblock */ boolean_t spa_extreme_rewind; /* rewind past deferred frees */ uint64_t spa_last_io; /* lbolt of last non-scan I/O */ kmutex_t spa_scrub_lock; /* resilver/scrub lock */ uint64_t spa_scrub_inflight; /* in-flight scrub bytes */ uint64_t spa_load_verify_ios; /* in-flight verifications IOs */ kcondvar_t spa_scrub_io_cv; /* scrub I/O completion */ uint8_t spa_scrub_active; /* active or suspended? */ uint8_t spa_scrub_type; /* type of scrub we're doing */ uint8_t spa_scrub_finished; /* indicator to rotate logs */ uint8_t spa_scrub_started; /* started since last boot */ uint8_t spa_scrub_reopen; /* scrub doing vdev_reopen */ uint64_t spa_scan_pass_start; /* start time per pass/reboot */ uint64_t spa_scan_pass_scrub_pause; /* scrub pause time */ uint64_t spa_scan_pass_scrub_spent_paused; /* total paused */ uint64_t spa_scan_pass_exam; /* examined bytes per pass */ uint64_t spa_scan_pass_issued; /* issued bytes per pass */ kmutex_t spa_async_lock; /* protect async state */ kthread_t *spa_async_thread; /* thread doing async task */ kthread_t *spa_async_thread_vd; /* thread doing vd async task */ int spa_async_suspended; /* async tasks suspended */ kcondvar_t spa_async_cv; /* wait for thread_exit() */ uint16_t spa_async_tasks; /* async task mask */ uint64_t spa_missing_tvds; /* unopenable tvds on load */ uint64_t spa_missing_tvds_allowed; /* allow loading spa? */ spa_removing_phys_t spa_removing_phys; spa_vdev_removal_t *spa_vdev_removal; spa_condensing_indirect_phys_t spa_condensing_indirect_phys; spa_condensing_indirect_t *spa_condensing_indirect; zthr_t *spa_condense_zthr; /* zthr doing condense. */ uint64_t spa_checkpoint_txg; /* the txg of the checkpoint */ spa_checkpoint_info_t spa_checkpoint_info; /* checkpoint accounting */ zthr_t *spa_checkpoint_discard_zthr; char *spa_root; /* alternate root directory */ uint64_t spa_ena; /* spa-wide ereport ENA */ int spa_last_open_failed; /* error if last open failed */ uint64_t spa_last_ubsync_txg; /* "best" uberblock txg */ uint64_t spa_last_ubsync_txg_ts; /* timestamp from that ub */ uint64_t spa_load_txg; /* ub txg that loaded */ uint64_t spa_load_txg_ts; /* timestamp from that ub */ uint64_t spa_load_meta_errors; /* verify metadata err count */ uint64_t spa_load_data_errors; /* verify data err count */ uint64_t spa_verify_min_txg; /* start txg of verify scrub */ kmutex_t spa_errlog_lock; /* error log lock */ uint64_t spa_errlog_last; /* last error log object */ uint64_t spa_errlog_scrub; /* scrub error log object */ kmutex_t spa_errlist_lock; /* error list/ereport lock */ avl_tree_t spa_errlist_last; /* last error list */ avl_tree_t spa_errlist_scrub; /* scrub error list */ uint64_t spa_deflate; /* should we deflate? */ uint64_t spa_history; /* history object */ kmutex_t spa_history_lock; /* history lock */ vdev_t *spa_pending_vdev; /* pending vdev additions */ kmutex_t spa_props_lock; /* property lock */ uint64_t spa_pool_props_object; /* object for properties */ uint64_t spa_bootfs; /* default boot filesystem */ uint64_t spa_failmode; /* failure mode for the pool */ uint64_t spa_delegation; /* delegation on/off */ list_t spa_config_list; /* previous cache file(s) */ /* per-CPU array of root of async I/O: */ zio_t **spa_async_zio_root; zio_t *spa_suspend_zio_root; /* root of all suspended I/O */ zio_t *spa_txg_zio[TXG_SIZE]; /* spa_sync() waits for this */ kmutex_t spa_suspend_lock; /* protects suspend_zio_root */ kcondvar_t spa_suspend_cv; /* notification of resume */ uint8_t spa_suspended; /* pool is suspended */ uint8_t spa_claiming; /* pool is doing zil_claim() */ boolean_t spa_debug; /* debug enabled? */ boolean_t spa_is_root; /* pool is root */ int spa_minref; /* num refs when first opened */ int spa_mode; /* FREAD | FWRITE */ spa_log_state_t spa_log_state; /* log state */ uint64_t spa_autoexpand; /* lun expansion on/off */ uint64_t spa_bootsize; /* efi system partition size */ ddt_t *spa_ddt[ZIO_CHECKSUM_FUNCTIONS]; /* in-core DDTs */ uint64_t spa_ddt_stat_object; /* DDT statistics */ uint64_t spa_dedup_ditto; /* dedup ditto threshold */ uint64_t spa_dedup_checksum; /* default dedup checksum */ uint64_t spa_dspace; /* dspace in normal class */ kmutex_t spa_vdev_top_lock; /* dueling offline/remove */ kmutex_t spa_proc_lock; /* protects spa_proc* */ kcondvar_t spa_proc_cv; /* spa_proc_state transitions */ spa_proc_state_t spa_proc_state; /* see definition */ struct proc *spa_proc; /* "zpool-poolname" process */ uint64_t spa_did; /* if procp != p0, did of t1 */ kthread_t *spa_trim_thread; /* thread sending TRIM I/Os */ kmutex_t spa_trim_lock; /* protects spa_trim_cv */ kcondvar_t spa_trim_cv; /* used to notify TRIM thread */ boolean_t spa_autoreplace; /* autoreplace set in open */ int spa_vdev_locks; /* locks grabbed */ uint64_t spa_creation_version; /* version at pool creation */ uint64_t spa_prev_software_version; /* See ub_software_version */ uint64_t spa_feat_for_write_obj; /* required to write to pool */ uint64_t spa_feat_for_read_obj; /* required to read from pool */ uint64_t spa_feat_desc_obj; /* Feature descriptions */ uint64_t spa_feat_enabled_txg_obj; /* Feature enabled txg */ /* cache feature refcounts */ uint64_t spa_feat_refcount_cache[SPA_FEATURES]; #ifdef illumos cyclic_id_t spa_deadman_cycid; /* cyclic id */ #else /* !illumos */ #ifdef _KERNEL struct callout spa_deadman_cycid; /* callout id */ struct task spa_deadman_task; #endif #endif /* illumos */ uint64_t spa_deadman_calls; /* number of deadman calls */ hrtime_t spa_sync_starttime; /* starting time fo spa_sync */ uint64_t spa_deadman_synctime; /* deadman expiration timer */ uint64_t spa_all_vdev_zaps; /* ZAP of per-vd ZAP obj #s */ spa_avz_action_t spa_avz_action; /* destroy/rebuild AVZ? */ #ifdef illumos /* * spa_iokstat_lock protects spa_iokstat and * spa_queue_stats[]. */ kmutex_t spa_iokstat_lock; struct kstat *spa_iokstat; /* kstat of io to this pool */ struct { int spa_active; int spa_queued; } spa_queue_stats[ZIO_PRIORITY_NUM_QUEUEABLE]; #endif hrtime_t spa_ccw_fail_time; /* Conf cache write fail time */ /* * spa_refcount & spa_config_lock must be the last elements * because refcount_t changes size based on compilation options. * In order for the MDB module to function correctly, the other * fields must remain in the same location. */ spa_config_lock_t spa_config_lock[SCL_LOCKS]; /* config changes */ refcount_t spa_refcount; /* number of opens */ #ifndef illumos boolean_t spa_splitting_newspa; /* creating new spa in split */ #endif }; extern const char *spa_config_path; extern void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent); extern void spa_load_spares(spa_t *spa); extern void spa_load_l2cache(spa_t *spa); #ifdef __cplusplus } #endif #endif /* _SYS_SPA_IMPL_H */ Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/vdev_impl.h =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/vdev_impl.h (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/vdev_impl.h (revision 336949) @@ -1,513 +1,514 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011, 2017 by Delphix. All rights reserved. + * Copyright (c) 2011, 2018 by Delphix. All rights reserved. */ #ifndef _SYS_VDEV_IMPL_H #define _SYS_VDEV_IMPL_H #include #include #include #include #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /* * Virtual device descriptors. * * All storage pool operations go through the virtual device framework, * which provides data replication and I/O scheduling. */ /* * Forward declarations that lots of things need. */ typedef struct vdev_queue vdev_queue_t; typedef struct vdev_cache vdev_cache_t; typedef struct vdev_cache_entry vdev_cache_entry_t; struct abd; extern int zfs_vdev_queue_depth_pct; +extern int zfs_vdev_def_queue_depth; extern uint32_t zfs_vdev_async_write_max_active; /* * Virtual device operations */ typedef int vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size, uint64_t *logical_ashift, uint64_t *physical_ashift); typedef void vdev_close_func_t(vdev_t *vd); typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize); typedef void vdev_io_start_func_t(zio_t *zio); typedef void vdev_io_done_func_t(zio_t *zio); typedef void vdev_state_change_func_t(vdev_t *vd, int, int); typedef boolean_t vdev_need_resilver_func_t(vdev_t *vd, uint64_t, size_t); typedef void vdev_hold_func_t(vdev_t *vd); typedef void vdev_rele_func_t(vdev_t *vd); typedef void vdev_remap_cb_t(uint64_t inner_offset, vdev_t *vd, uint64_t offset, uint64_t size, void *arg); typedef void vdev_remap_func_t(vdev_t *vd, uint64_t offset, uint64_t size, vdev_remap_cb_t callback, void *arg); typedef struct vdev_ops { vdev_open_func_t *vdev_op_open; vdev_close_func_t *vdev_op_close; vdev_asize_func_t *vdev_op_asize; vdev_io_start_func_t *vdev_op_io_start; vdev_io_done_func_t *vdev_op_io_done; vdev_state_change_func_t *vdev_op_state_change; vdev_need_resilver_func_t *vdev_op_need_resilver; vdev_hold_func_t *vdev_op_hold; vdev_rele_func_t *vdev_op_rele; vdev_remap_func_t *vdev_op_remap; char vdev_op_type[16]; boolean_t vdev_op_leaf; } vdev_ops_t; /* * Virtual device properties */ struct vdev_cache_entry { struct abd *ve_abd; uint64_t ve_offset; uint64_t ve_lastused; avl_node_t ve_offset_node; avl_node_t ve_lastused_node; uint32_t ve_hits; uint16_t ve_missed_update; zio_t *ve_fill_io; }; struct vdev_cache { avl_tree_t vc_offset_tree; avl_tree_t vc_lastused_tree; kmutex_t vc_lock; }; typedef struct vdev_queue_class { uint32_t vqc_active; /* * Sorted by offset or timestamp, depending on if the queue is * LBA-ordered vs FIFO. */ avl_tree_t vqc_queued_tree; } vdev_queue_class_t; struct vdev_queue { vdev_t *vq_vdev; vdev_queue_class_t vq_class[ZIO_PRIORITY_NUM_QUEUEABLE]; avl_tree_t vq_active_tree; avl_tree_t vq_read_offset_tree; avl_tree_t vq_write_offset_tree; uint64_t vq_last_offset; hrtime_t vq_io_complete_ts; /* time last i/o completed */ kmutex_t vq_lock; uint64_t vq_lastoffset; }; /* * On-disk indirect vdev state. * * An indirect vdev is described exclusively in the MOS config of a pool. * The config for an indirect vdev includes several fields, which are * accessed in memory by a vdev_indirect_config_t. */ typedef struct vdev_indirect_config { /* * Object (in MOS) which contains the indirect mapping. This object * contains an array of vdev_indirect_mapping_entry_phys_t ordered by * vimep_src. The bonus buffer for this object is a * vdev_indirect_mapping_phys_t. This object is allocated when a vdev * removal is initiated. * * Note that this object can be empty if none of the data on the vdev * has been copied yet. */ uint64_t vic_mapping_object; /* * Object (in MOS) which contains the birth times for the mapping * entries. This object contains an array of * vdev_indirect_birth_entry_phys_t sorted by vibe_offset. The bonus * buffer for this object is a vdev_indirect_birth_phys_t. This object * is allocated when a vdev removal is initiated. * * Note that this object can be empty if none of the vdev has yet been * copied. */ uint64_t vic_births_object; /* * This is the vdev ID which was removed previous to this vdev, or * UINT64_MAX if there are no previously removed vdevs. */ uint64_t vic_prev_indirect_vdev; } vdev_indirect_config_t; /* * Virtual device descriptor */ struct vdev { /* * Common to all vdev types. */ uint64_t vdev_id; /* child number in vdev parent */ uint64_t vdev_guid; /* unique ID for this vdev */ uint64_t vdev_guid_sum; /* self guid + all child guids */ uint64_t vdev_orig_guid; /* orig. guid prior to remove */ uint64_t vdev_asize; /* allocatable device capacity */ uint64_t vdev_min_asize; /* min acceptable asize */ uint64_t vdev_max_asize; /* max acceptable asize */ uint64_t vdev_ashift; /* block alignment shift */ /* * Logical block alignment shift * * The smallest sized/aligned I/O supported by the device. */ uint64_t vdev_logical_ashift; /* * Physical block alignment shift * * The device supports logical I/Os with vdev_logical_ashift * size/alignment, but optimum performance will be achieved by * aligning/sizing requests to vdev_physical_ashift. Smaller * requests may be inflated or incur device level read-modify-write * operations. * * May be 0 to indicate no preference (i.e. use vdev_logical_ashift). */ uint64_t vdev_physical_ashift; uint64_t vdev_state; /* see VDEV_STATE_* #defines */ uint64_t vdev_prevstate; /* used when reopening a vdev */ vdev_ops_t *vdev_ops; /* vdev operations */ spa_t *vdev_spa; /* spa for this vdev */ void *vdev_tsd; /* type-specific data */ vnode_t *vdev_name_vp; /* vnode for pathname */ vnode_t *vdev_devid_vp; /* vnode for devid */ vdev_t *vdev_top; /* top-level vdev */ vdev_t *vdev_parent; /* parent vdev */ vdev_t **vdev_child; /* array of children */ uint64_t vdev_children; /* number of children */ vdev_stat_t vdev_stat; /* virtual device statistics */ boolean_t vdev_expanding; /* expand the vdev? */ boolean_t vdev_reopening; /* reopen in progress? */ int vdev_open_error; /* error on last open */ kthread_t *vdev_open_thread; /* thread opening children */ uint64_t vdev_crtxg; /* txg when top-level was added */ /* * Top-level vdev state. */ uint64_t vdev_ms_array; /* metaslab array object */ uint64_t vdev_ms_shift; /* metaslab size shift */ uint64_t vdev_ms_count; /* number of metaslabs */ metaslab_group_t *vdev_mg; /* metaslab group */ metaslab_t **vdev_ms; /* metaslab array */ txg_list_t vdev_ms_list; /* per-txg dirty metaslab lists */ txg_list_t vdev_dtl_list; /* per-txg dirty DTL lists */ txg_node_t vdev_txg_node; /* per-txg dirty vdev linkage */ boolean_t vdev_remove_wanted; /* async remove wanted? */ boolean_t vdev_probe_wanted; /* async probe wanted? */ list_node_t vdev_config_dirty_node; /* config dirty list */ list_node_t vdev_state_dirty_node; /* state dirty list */ uint64_t vdev_deflate_ratio; /* deflation ratio (x512) */ uint64_t vdev_islog; /* is an intent log device */ uint64_t vdev_removing; /* device is being removed? */ boolean_t vdev_ishole; /* is a hole in the namespace */ kmutex_t vdev_queue_lock; /* protects vdev_queue_depth */ uint64_t vdev_top_zap; /* pool checkpoint related */ space_map_t *vdev_checkpoint_sm; /* contains reserved blocks */ /* * Values stored in the config for an indirect or removing vdev. */ vdev_indirect_config_t vdev_indirect_config; /* * The vdev_indirect_rwlock protects the vdev_indirect_mapping * pointer from changing on indirect vdevs (when it is condensed). * Note that removing (not yet indirect) vdevs have different * access patterns (the mapping is not accessed from open context, * e.g. from zio_read) and locking strategy (e.g. svr_lock). */ krwlock_t vdev_indirect_rwlock; vdev_indirect_mapping_t *vdev_indirect_mapping; vdev_indirect_births_t *vdev_indirect_births; /* * In memory data structures used to manage the obsolete sm, for * indirect or removing vdevs. * * The vdev_obsolete_segments is the in-core record of the segments * that are no longer referenced anywhere in the pool (due to * being freed or remapped and not referenced by any snapshots). * During a sync, segments are added to vdev_obsolete_segments * via vdev_indirect_mark_obsolete(); at the end of each sync * pass, this is appended to vdev_obsolete_sm via * vdev_indirect_sync_obsolete(). The vdev_obsolete_lock * protects against concurrent modifications of vdev_obsolete_segments * from multiple zio threads. */ kmutex_t vdev_obsolete_lock; range_tree_t *vdev_obsolete_segments; space_map_t *vdev_obsolete_sm; /* * The queue depth parameters determine how many async writes are * still pending (i.e. allocated by net yet issued to disk) per * top-level (vdev_async_write_queue_depth) and the maximum allowed * (vdev_max_async_write_queue_depth). These values only apply to * top-level vdevs. */ uint64_t vdev_async_write_queue_depth; uint64_t vdev_max_async_write_queue_depth; /* * Protects the vdev_scan_io_queue field itself as well as the * structure's contents (when present). */ kmutex_t vdev_scan_io_queue_lock; struct dsl_scan_io_queue *vdev_scan_io_queue; /* * Leaf vdev state. */ range_tree_t *vdev_dtl[DTL_TYPES]; /* dirty time logs */ space_map_t *vdev_dtl_sm; /* dirty time log space map */ txg_node_t vdev_dtl_node; /* per-txg dirty DTL linkage */ uint64_t vdev_dtl_object; /* DTL object */ uint64_t vdev_psize; /* physical device capacity */ uint64_t vdev_wholedisk; /* true if this is a whole disk */ uint64_t vdev_offline; /* persistent offline state */ uint64_t vdev_faulted; /* persistent faulted state */ uint64_t vdev_degraded; /* persistent degraded state */ uint64_t vdev_removed; /* persistent removed state */ uint64_t vdev_resilver_txg; /* persistent resilvering state */ uint64_t vdev_nparity; /* number of parity devices for raidz */ char *vdev_path; /* vdev path (if any) */ char *vdev_devid; /* vdev devid (if any) */ char *vdev_physpath; /* vdev device path (if any) */ char *vdev_fru; /* physical FRU location */ uint64_t vdev_not_present; /* not present during import */ uint64_t vdev_unspare; /* unspare when resilvering done */ boolean_t vdev_nowritecache; /* true if flushwritecache failed */ boolean_t vdev_notrim; /* true if trim failed */ boolean_t vdev_checkremove; /* temporary online test */ boolean_t vdev_forcefault; /* force online fault */ boolean_t vdev_splitting; /* split or repair in progress */ boolean_t vdev_delayed_close; /* delayed device close? */ boolean_t vdev_tmpoffline; /* device taken offline temporarily? */ boolean_t vdev_detached; /* device detached? */ boolean_t vdev_cant_read; /* vdev is failing all reads */ boolean_t vdev_cant_write; /* vdev is failing all writes */ boolean_t vdev_isspare; /* was a hot spare */ boolean_t vdev_isl2cache; /* was a l2cache device */ vdev_queue_t vdev_queue; /* I/O deadline schedule queue */ vdev_cache_t vdev_cache; /* physical block cache */ spa_aux_vdev_t *vdev_aux; /* for l2cache and spares vdevs */ zio_t *vdev_probe_zio; /* root of current probe */ vdev_aux_t vdev_label_aux; /* on-disk aux state */ struct trim_map *vdev_trimmap; /* map on outstanding trims */ uint16_t vdev_rotation_rate; /* rotational rate of the media */ #define VDEV_RATE_UNKNOWN 0 #define VDEV_RATE_NON_ROTATING 1 uint64_t vdev_leaf_zap; /* * For DTrace to work in userland (libzpool) context, these fields must * remain at the end of the structure. DTrace will use the kernel's * CTF definition for 'struct vdev', and since the size of a kmutex_t is * larger in userland, the offsets for the rest of the fields would be * incorrect. */ kmutex_t vdev_dtl_lock; /* vdev_dtl_{map,resilver} */ kmutex_t vdev_stat_lock; /* vdev_stat */ kmutex_t vdev_probe_lock; /* protects vdev_probe_zio */ }; #define VDEV_RAIDZ_MAXPARITY 3 #define VDEV_PAD_SIZE (8 << 10) /* 2 padding areas (vl_pad1 and vl_pad2) to skip */ #define VDEV_SKIP_SIZE VDEV_PAD_SIZE * 2 #define VDEV_PHYS_SIZE (112 << 10) #define VDEV_UBERBLOCK_RING (128 << 10) /* The largest uberblock we support is 8k. */ #define MAX_UBERBLOCK_SHIFT (13) #define VDEV_UBERBLOCK_SHIFT(vd) \ MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \ MAX_UBERBLOCK_SHIFT) #define VDEV_UBERBLOCK_COUNT(vd) \ (VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd)) #define VDEV_UBERBLOCK_OFFSET(vd, n) \ offsetof(vdev_label_t, vl_uberblock[(n) << VDEV_UBERBLOCK_SHIFT(vd)]) #define VDEV_UBERBLOCK_SIZE(vd) (1ULL << VDEV_UBERBLOCK_SHIFT(vd)) typedef struct vdev_phys { char vp_nvlist[VDEV_PHYS_SIZE - sizeof (zio_eck_t)]; zio_eck_t vp_zbt; } vdev_phys_t; typedef struct vdev_label { char vl_pad1[VDEV_PAD_SIZE]; /* 8K */ char vl_pad2[VDEV_PAD_SIZE]; /* 8K */ vdev_phys_t vl_vdev_phys; /* 112K */ char vl_uberblock[VDEV_UBERBLOCK_RING]; /* 128K */ } vdev_label_t; /* 256K total */ /* * vdev_dirty() flags */ #define VDD_METASLAB 0x01 #define VDD_DTL 0x02 /* Offset of embedded boot loader region on each label */ #define VDEV_BOOT_OFFSET (2 * sizeof (vdev_label_t)) /* * Size of embedded boot loader region on each label. * The total size of the first two labels plus the boot area is 4MB. */ #define VDEV_BOOT_SIZE (7ULL << 19) /* 3.5M */ /* * Size of label regions at the start and end of each leaf device. */ #define VDEV_LABEL_START_SIZE (2 * sizeof (vdev_label_t) + VDEV_BOOT_SIZE) #define VDEV_LABEL_END_SIZE (2 * sizeof (vdev_label_t)) #define VDEV_LABELS 4 #define VDEV_BEST_LABEL VDEV_LABELS #define VDEV_ALLOC_LOAD 0 #define VDEV_ALLOC_ADD 1 #define VDEV_ALLOC_SPARE 2 #define VDEV_ALLOC_L2CACHE 3 #define VDEV_ALLOC_ROOTPOOL 4 #define VDEV_ALLOC_SPLIT 5 #define VDEV_ALLOC_ATTACH 6 /* * Allocate or free a vdev */ extern vdev_t *vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops); extern int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *config, vdev_t *parent, uint_t id, int alloctype); extern void vdev_free(vdev_t *vd); /* * Add or remove children and parents */ extern void vdev_add_child(vdev_t *pvd, vdev_t *cvd); extern void vdev_remove_child(vdev_t *pvd, vdev_t *cvd); extern void vdev_compact_children(vdev_t *pvd); extern vdev_t *vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops); extern void vdev_remove_parent(vdev_t *cvd); /* * vdev sync load and sync */ extern boolean_t vdev_log_state_valid(vdev_t *vd); extern int vdev_load(vdev_t *vd); extern int vdev_dtl_load(vdev_t *vd); extern void vdev_sync(vdev_t *vd, uint64_t txg); extern void vdev_sync_done(vdev_t *vd, uint64_t txg); extern void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg); extern void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg); /* * Available vdev types. */ extern vdev_ops_t vdev_root_ops; extern vdev_ops_t vdev_mirror_ops; extern vdev_ops_t vdev_replacing_ops; extern vdev_ops_t vdev_raidz_ops; #ifdef _KERNEL extern vdev_ops_t vdev_geom_ops; #else extern vdev_ops_t vdev_disk_ops; #endif extern vdev_ops_t vdev_file_ops; extern vdev_ops_t vdev_missing_ops; extern vdev_ops_t vdev_hole_ops; extern vdev_ops_t vdev_spare_ops; extern vdev_ops_t vdev_indirect_ops; /* * Common size functions */ extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize); extern uint64_t vdev_get_min_asize(vdev_t *vd); extern void vdev_set_min_asize(vdev_t *vd); /* * Global variables */ extern int vdev_standard_sm_blksz; /* zdb uses this tunable, so it must be declared here to make lint happy. */ extern int zfs_vdev_cache_size; extern uint_t zfs_geom_probe_vdev_key; /* * Functions from vdev_indirect.c */ extern void vdev_indirect_sync_obsolete(vdev_t *vd, dmu_tx_t *tx); extern boolean_t vdev_indirect_should_condense(vdev_t *vd); extern void spa_condense_indirect_start_sync(vdev_t *vd, dmu_tx_t *tx); extern int vdev_obsolete_sm_object(vdev_t *vd); extern boolean_t vdev_obsolete_counts_are_precise(vdev_t *vd); #ifdef illumos /* * Other miscellaneous functions */ int vdev_checkpoint_sm_object(vdev_t *vd); /* * The vdev_buf_t is used to translate between zio_t and buf_t, and back again. */ typedef struct vdev_buf { buf_t vb_buf; /* buffer that describes the io */ zio_t *vb_io; /* pointer back to the original zio_t */ } vdev_buf_t; #endif #ifdef __cplusplus } #endif #endif /* _SYS_VDEV_IMPL_H */ Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h (revision 336949) @@ -1,664 +1,665 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. - * Copyright (c) 2012, 2017 by Delphix. All rights reserved. + * Copyright (c) 2012, 2018 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright 2016 Toomas Soome */ #ifndef _ZIO_H #define _ZIO_H #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /* * Embedded checksum */ #define ZEC_MAGIC 0x210da7ab10c7a11ULL typedef struct zio_eck { uint64_t zec_magic; /* for validation, endianness */ zio_cksum_t zec_cksum; /* 256-bit checksum */ } zio_eck_t; /* * Gang block headers are self-checksumming and contain an array * of block pointers. */ #define SPA_GANGBLOCKSIZE SPA_MINBLOCKSIZE #define SPA_GBH_NBLKPTRS ((SPA_GANGBLOCKSIZE - \ sizeof (zio_eck_t)) / sizeof (blkptr_t)) #define SPA_GBH_FILLER ((SPA_GANGBLOCKSIZE - \ sizeof (zio_eck_t) - \ (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\ sizeof (uint64_t)) typedef struct zio_gbh { blkptr_t zg_blkptr[SPA_GBH_NBLKPTRS]; uint64_t zg_filler[SPA_GBH_FILLER]; zio_eck_t zg_tail; } zio_gbh_phys_t; enum zio_checksum { ZIO_CHECKSUM_INHERIT = 0, ZIO_CHECKSUM_ON, ZIO_CHECKSUM_OFF, ZIO_CHECKSUM_LABEL, ZIO_CHECKSUM_GANG_HEADER, ZIO_CHECKSUM_ZILOG, ZIO_CHECKSUM_FLETCHER_2, ZIO_CHECKSUM_FLETCHER_4, ZIO_CHECKSUM_SHA256, ZIO_CHECKSUM_ZILOG2, ZIO_CHECKSUM_NOPARITY, ZIO_CHECKSUM_SHA512, ZIO_CHECKSUM_SKEIN, #ifdef illumos ZIO_CHECKSUM_EDONR, #endif ZIO_CHECKSUM_FUNCTIONS }; /* * The number of "legacy" compression functions which can be set on individual * objects. */ #define ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2 #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_4 #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON #define ZIO_CHECKSUM_MASK 0xffULL #define ZIO_CHECKSUM_VERIFY (1 << 8) #define ZIO_DEDUPCHECKSUM ZIO_CHECKSUM_SHA256 #define ZIO_DEDUPDITTO_MIN 100 /* * The number of "legacy" compression functions which can be set on individual * objects. */ #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4 /* * The meaning of "compress = on" selected by the compression features enabled * on a given pool. */ #define ZIO_COMPRESS_LEGACY_ON_VALUE ZIO_COMPRESS_LZJB #define ZIO_COMPRESS_LZ4_ON_VALUE ZIO_COMPRESS_LZ4 #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_OFF #define BOOTFS_COMPRESS_VALID(compress) \ ((compress) == ZIO_COMPRESS_LZJB || \ (compress) == ZIO_COMPRESS_LZ4 || \ (compress) == ZIO_COMPRESS_ON || \ (compress) == ZIO_COMPRESS_OFF) #define ZIO_FAILURE_MODE_WAIT 0 #define ZIO_FAILURE_MODE_CONTINUE 1 #define ZIO_FAILURE_MODE_PANIC 2 enum zio_flag { /* * Flags inherited by gang, ddt, and vdev children, * and that must be equal for two zios to aggregate */ ZIO_FLAG_DONT_AGGREGATE = 1 << 0, ZIO_FLAG_IO_REPAIR = 1 << 1, ZIO_FLAG_SELF_HEAL = 1 << 2, ZIO_FLAG_RESILVER = 1 << 3, ZIO_FLAG_SCRUB = 1 << 4, ZIO_FLAG_SCAN_THREAD = 1 << 5, ZIO_FLAG_PHYSICAL = 1 << 6, #define ZIO_FLAG_AGG_INHERIT (ZIO_FLAG_CANFAIL - 1) /* * Flags inherited by ddt, gang, and vdev children. */ ZIO_FLAG_CANFAIL = 1 << 7, /* must be first for INHERIT */ ZIO_FLAG_SPECULATIVE = 1 << 8, ZIO_FLAG_CONFIG_WRITER = 1 << 9, ZIO_FLAG_DONT_RETRY = 1 << 10, ZIO_FLAG_DONT_CACHE = 1 << 11, ZIO_FLAG_NODATA = 1 << 12, ZIO_FLAG_INDUCE_DAMAGE = 1 << 13, ZIO_FLAG_IO_ALLOCATING = 1 << 14, #define ZIO_FLAG_DDT_INHERIT (ZIO_FLAG_IO_RETRY - 1) #define ZIO_FLAG_GANG_INHERIT (ZIO_FLAG_IO_RETRY - 1) /* * Flags inherited by vdev children. */ ZIO_FLAG_IO_RETRY = 1 << 15, /* must be first for INHERIT */ ZIO_FLAG_PROBE = 1 << 16, ZIO_FLAG_TRYHARD = 1 << 17, ZIO_FLAG_OPTIONAL = 1 << 18, #define ZIO_FLAG_VDEV_INHERIT (ZIO_FLAG_DONT_QUEUE - 1) /* * Flags not inherited by any children. */ ZIO_FLAG_DONT_QUEUE = 1 << 19, /* must be first for INHERIT */ ZIO_FLAG_DONT_PROPAGATE = 1 << 20, ZIO_FLAG_IO_BYPASS = 1 << 21, ZIO_FLAG_IO_REWRITE = 1 << 22, ZIO_FLAG_RAW = 1 << 23, ZIO_FLAG_GANG_CHILD = 1 << 24, ZIO_FLAG_DDT_CHILD = 1 << 25, ZIO_FLAG_GODFATHER = 1 << 26, ZIO_FLAG_NOPWRITE = 1 << 27, ZIO_FLAG_REEXECUTED = 1 << 28, ZIO_FLAG_DELEGATED = 1 << 29, }; #define ZIO_FLAG_MUSTSUCCEED 0 #define ZIO_DDT_CHILD_FLAGS(zio) \ (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) | \ ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL) #define ZIO_GANG_CHILD_FLAGS(zio) \ (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) | \ ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL) #define ZIO_VDEV_CHILD_FLAGS(zio) \ (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) | \ ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL) #define ZIO_CHILD_BIT(x) (1 << (x)) #define ZIO_CHILD_BIT_IS_SET(val, x) ((val) & (1 << (x))) enum zio_child { ZIO_CHILD_VDEV = 0, ZIO_CHILD_GANG, ZIO_CHILD_DDT, ZIO_CHILD_LOGICAL, ZIO_CHILD_TYPES }; #define ZIO_CHILD_VDEV_BIT ZIO_CHILD_BIT(ZIO_CHILD_VDEV) #define ZIO_CHILD_GANG_BIT ZIO_CHILD_BIT(ZIO_CHILD_GANG) #define ZIO_CHILD_DDT_BIT ZIO_CHILD_BIT(ZIO_CHILD_DDT) #define ZIO_CHILD_LOGICAL_BIT ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL) #define ZIO_CHILD_ALL_BITS \ (ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT | \ ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT) enum zio_wait_type { ZIO_WAIT_READY = 0, ZIO_WAIT_DONE, ZIO_WAIT_TYPES }; /* * We'll take the number 122 and 123 to indicate checksum errors and * fragmentation. Those doesn't collide with any errno values as they * are greater than ELAST. */ #define ECKSUM 122 #define EFRAGS 123 typedef void zio_done_func_t(zio_t *zio); extern boolean_t zio_dva_throttle_enabled; extern const char *zio_type_name[ZIO_TYPES]; /* * A bookmark is a four-tuple that uniquely * identifies any block in the pool. By convention, the meta-objset (MOS) * is objset 0, and the meta-dnode is object 0. This covers all blocks * except root blocks and ZIL blocks, which are defined as follows: * * Root blocks (objset_phys_t) are object 0, level -1: . * ZIL blocks are bookmarked . * dmu_sync()ed ZIL data blocks are bookmarked . * dnode visit bookmarks are . * * Note: this structure is called a bookmark because its original purpose * was to remember where to resume a pool-wide traverse. * * Note: this structure is passed between userland and the kernel, and is * stored on disk (by virtue of being incorporated into other on-disk * structures, e.g. dsl_scan_phys_t). */ typedef struct zbookmark_phys { uint64_t zb_objset; uint64_t zb_object; int64_t zb_level; uint64_t zb_blkid; } zbookmark_phys_t; #define SET_BOOKMARK(zb, objset, object, level, blkid) \ { \ (zb)->zb_objset = objset; \ (zb)->zb_object = object; \ (zb)->zb_level = level; \ (zb)->zb_blkid = blkid; \ } #define ZB_DESTROYED_OBJSET (-1ULL) #define ZB_ROOT_OBJECT (0ULL) #define ZB_ROOT_LEVEL (-1LL) #define ZB_ROOT_BLKID (0ULL) #define ZB_ZIL_OBJECT (0ULL) #define ZB_ZIL_LEVEL (-2LL) #define ZB_DNODE_LEVEL (-3LL) #define ZB_DNODE_BLKID (0ULL) #define ZB_IS_ZERO(zb) \ ((zb)->zb_objset == 0 && (zb)->zb_object == 0 && \ (zb)->zb_level == 0 && (zb)->zb_blkid == 0) #define ZB_IS_ROOT(zb) \ ((zb)->zb_object == ZB_ROOT_OBJECT && \ (zb)->zb_level == ZB_ROOT_LEVEL && \ (zb)->zb_blkid == ZB_ROOT_BLKID) typedef struct zio_prop { enum zio_checksum zp_checksum; enum zio_compress zp_compress; dmu_object_type_t zp_type; uint8_t zp_level; uint8_t zp_copies; boolean_t zp_dedup; boolean_t zp_dedup_verify; boolean_t zp_nopwrite; } zio_prop_t; typedef struct zio_cksum_report zio_cksum_report_t; typedef void zio_cksum_finish_f(zio_cksum_report_t *rep, const void *good_data); typedef void zio_cksum_free_f(void *cbdata, size_t size); struct zio_bad_cksum; /* defined in zio_checksum.h */ struct dnode_phys; struct abd; struct zio_cksum_report { struct zio_cksum_report *zcr_next; nvlist_t *zcr_ereport; nvlist_t *zcr_detector; void *zcr_cbdata; size_t zcr_cbinfo; /* passed to zcr_free() */ uint64_t zcr_align; uint64_t zcr_length; zio_cksum_finish_f *zcr_finish; zio_cksum_free_f *zcr_free; /* internal use only */ struct zio_bad_cksum *zcr_ckinfo; /* information from failure */ }; typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr, void *arg); zio_vsd_cksum_report_f zio_vsd_default_cksum_report; typedef struct zio_vsd_ops { zio_done_func_t *vsd_free; zio_vsd_cksum_report_f *vsd_cksum_report; } zio_vsd_ops_t; typedef struct zio_gang_node { zio_gbh_phys_t *gn_gbh; struct zio_gang_node *gn_child[SPA_GBH_NBLKPTRS]; } zio_gang_node_t; typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp, zio_gang_node_t *gn, struct abd *data, uint64_t offset); typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size); typedef struct zio_transform { struct abd *zt_orig_abd; uint64_t zt_orig_size; uint64_t zt_bufsize; zio_transform_func_t *zt_transform; struct zio_transform *zt_next; } zio_transform_t; typedef int zio_pipe_stage_t(zio_t *zio); /* * The io_reexecute flags are distinct from io_flags because the child must * be able to propagate them to the parent. The normal io_flags are local * to the zio, not protected by any lock, and not modifiable by children; * the reexecute flags are protected by io_lock, modifiable by children, * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set. */ #define ZIO_REEXECUTE_NOW 0x01 #define ZIO_REEXECUTE_SUSPEND 0x02 typedef struct zio_alloc_list { list_t zal_list; uint64_t zal_size; } zio_alloc_list_t; typedef struct zio_link { zio_t *zl_parent; zio_t *zl_child; list_node_t zl_parent_node; list_node_t zl_child_node; } zio_link_t; /* * Used for TRIM kstat. */ typedef struct zio_trim_stats { /* * Number of bytes successfully TRIMmed. */ kstat_named_t bytes; /* * Number of successful TRIM requests. */ kstat_named_t success; /* * Number of TRIM requests that failed because TRIM is not * supported. */ kstat_named_t unsupported; /* * Number of TRIM requests that failed for other reasons. */ kstat_named_t failed; } zio_trim_stats_t; extern zio_trim_stats_t zio_trim_stats; #define ZIO_TRIM_STAT_INCR(stat, val) \ atomic_add_64(&zio_trim_stats.stat.value.ui64, (val)); #define ZIO_TRIM_STAT_BUMP(stat) \ ZIO_TRIM_STAT_INCR(stat, 1); struct zio { /* Core information about this I/O */ zbookmark_phys_t io_bookmark; zio_prop_t io_prop; zio_type_t io_type; enum zio_child io_child_type; int io_cmd; zio_priority_t io_priority; uint8_t io_reexecute; uint8_t io_state[ZIO_WAIT_TYPES]; uint64_t io_txg; spa_t *io_spa; blkptr_t *io_bp; blkptr_t *io_bp_override; blkptr_t io_bp_copy; list_t io_parent_list; list_t io_child_list; zio_t *io_logical; zio_transform_t *io_transform_stack; /* Callback info */ zio_done_func_t *io_ready; zio_done_func_t *io_children_ready; zio_done_func_t *io_physdone; zio_done_func_t *io_done; void *io_private; int64_t io_prev_space_delta; /* DMU private */ blkptr_t io_bp_orig; /* Data represented by this I/O */ struct abd *io_abd; struct abd *io_orig_abd; uint64_t io_size; uint64_t io_orig_size; /* io_lsize != io_orig_size iff this is a raw write */ uint64_t io_lsize; /* Stuff for the vdev stack */ vdev_t *io_vd; void *io_vsd; const zio_vsd_ops_t *io_vsd_ops; uint64_t io_offset; hrtime_t io_timestamp; hrtime_t io_queued_timestamp; hrtime_t io_target_timestamp; avl_node_t io_queue_node; avl_node_t io_offset_node; avl_node_t io_alloc_node; zio_alloc_list_t io_alloc_list; #ifdef __FreeBSD__ struct bio *io_bio; #endif /* Internal pipeline state */ enum zio_flag io_flags; enum zio_stage io_stage; enum zio_stage io_pipeline; enum zio_flag io_orig_flags; enum zio_stage io_orig_stage; enum zio_stage io_orig_pipeline; enum zio_stage io_pipeline_trace; int io_error; int io_child_error[ZIO_CHILD_TYPES]; uint64_t io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES]; uint64_t io_child_count; uint64_t io_phys_children; uint64_t io_parent_count; uint64_t *io_stall; zio_t *io_gang_leader; zio_gang_node_t *io_gang_tree; void *io_executor; void *io_waiter; kmutex_t io_lock; kcondvar_t io_cv; + int io_allocator; /* FMA state */ zio_cksum_report_t *io_cksum_report; uint64_t io_ena; /* Taskq dispatching state */ taskq_ent_t io_tqent; avl_node_t io_trim_node; list_node_t io_trim_link; }; extern int zio_bookmark_compare(const void *, const void *); extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, void *priv, enum zio_flag flags); extern zio_t *zio_root(spa_t *spa, zio_done_func_t *done, void *priv, enum zio_flag flags); extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, struct abd *data, uint64_t lsize, zio_done_func_t *done, void *priv, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb); extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp, zio_done_func_t *ready, zio_done_func_t *children_ready, zio_done_func_t *physdone, zio_done_func_t *done, void *priv, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb); extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, struct abd *data, uint64_t size, zio_done_func_t *done, void *priv, zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb); extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite); extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp); extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, zio_done_func_t *done, void *priv, enum zio_flag flags); extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv, zio_priority_t priority, enum zio_flag flags); extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, struct abd *data, int checksum, zio_done_func_t *done, void *priv, zio_priority_t priority, enum zio_flag flags, boolean_t labels); extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, struct abd *data, int checksum, zio_done_func_t *done, void *priv, zio_priority_t priority, enum zio_flag flags, boolean_t labels); extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, uint64_t size, enum zio_flag flags); -extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, - blkptr_t *old_bp, uint64_t size, boolean_t *slog); +extern int zio_alloc_zil(spa_t *spa, uint64_t objset, uint64_t txg, + blkptr_t *new_bp, blkptr_t *old_bp, uint64_t size, boolean_t *slog); extern void zio_flush(zio_t *zio, vdev_t *vd); extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset, uint64_t size); extern void zio_shrink(zio_t *zio, uint64_t size); extern int zio_wait(zio_t *zio); extern void zio_nowait(zio_t *zio); extern void zio_execute(zio_t *zio); extern void zio_interrupt(zio_t *zio); extern void zio_delay_init(zio_t *zio); extern void zio_delay_interrupt(zio_t *zio); extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **); extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **); extern zio_t *zio_unique_parent(zio_t *cio); extern void zio_add_child(zio_t *pio, zio_t *cio); extern void *zio_buf_alloc(size_t size); extern void zio_buf_free(void *buf, size_t size); extern void *zio_data_buf_alloc(size_t size); extern void zio_data_buf_free(void *buf, size_t size); extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size, uint64_t bufsize, zio_transform_func_t *transform); extern void zio_pop_transforms(zio_t *zio); extern void zio_resubmit_stage_async(void *); extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd, uint64_t offset, struct abd *data, uint64_t size, int type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *priv); extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, struct abd *data, uint64_t size, int type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *priv); extern void zio_vdev_io_bypass(zio_t *zio); extern void zio_vdev_io_reissue(zio_t *zio); extern void zio_vdev_io_redone(zio_t *zio); extern void zio_change_priority(zio_t *pio, zio_priority_t priority); extern void zio_checksum_verified(zio_t *zio); extern int zio_worst_error(int e1, int e2); extern enum zio_checksum zio_checksum_select(enum zio_checksum child, enum zio_checksum parent); extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa, enum zio_checksum child, enum zio_checksum parent); extern enum zio_compress zio_compress_select(spa_t *spa, enum zio_compress child, enum zio_compress parent); extern void zio_suspend(spa_t *spa, zio_t *zio); extern int zio_resume(spa_t *spa); extern void zio_resume_wait(spa_t *spa); /* * Initial setup and teardown. */ extern void zio_init(void); extern void zio_fini(void); /* * Fault injection */ struct zinject_record; extern uint32_t zio_injection_enabled; extern int zio_inject_fault(char *name, int flags, int *id, struct zinject_record *record); extern int zio_inject_list_next(int *id, char *name, size_t buflen, struct zinject_record *record); extern int zio_clear_fault(int id); extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type); extern int zio_handle_fault_injection(zio_t *zio, int error); extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error); extern int zio_handle_label_injection(zio_t *zio, int error); extern void zio_handle_ignored_writes(zio_t *zio); extern hrtime_t zio_handle_io_delay(zio_t *zio); /* * Checksum ereport functions */ extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio, uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info); extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report, const void *good_data, const void *bad_data, boolean_t drop_if_identical); extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report); extern void zfs_ereport_free_checksum(zio_cksum_report_t *report); /* If we have the good data in hand, this function can be used */ extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd, struct zio *zio, uint64_t offset, uint64_t length, const void *good_data, const void *bad_data, struct zio_bad_cksum *info); /* Called from spa_sync(), but primarily an injection handler */ extern void spa_handle_ignored_writes(spa_t *spa); /* zbookmark_phys functions */ boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp, const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block); int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2, uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2); #ifdef __cplusplus } #endif #endif /* _ZIO_H */ Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c (revision 336949) @@ -1,4265 +1,4265 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2018 by Delphix. All rights reserved. * Copyright 2017 Nexenta Systems, Inc. * Copyright 2013 Martin Matuska . All rights reserved. * Copyright (c) 2014 Integros [integros.com] * Copyright 2016 Toomas Soome * Copyright 2017 Joyent, Inc. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include SYSCTL_DECL(_vfs_zfs); SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV"); /* * Virtual device management. */ /* * The limit for ZFS to automatically increase a top-level vdev's ashift * from logical ashift to physical ashift. * * Example: one or more 512B emulation child vdevs * child->vdev_ashift = 9 (512 bytes) * child->vdev_physical_ashift = 12 (4096 bytes) * zfs_max_auto_ashift = 11 (2048 bytes) * zfs_min_auto_ashift = 9 (512 bytes) * * On pool creation or the addition of a new top-level vdev, ZFS will * increase the ashift of the top-level vdev to 2048 as limited by * zfs_max_auto_ashift. * * Example: one or more 512B emulation child vdevs * child->vdev_ashift = 9 (512 bytes) * child->vdev_physical_ashift = 12 (4096 bytes) * zfs_max_auto_ashift = 13 (8192 bytes) * zfs_min_auto_ashift = 9 (512 bytes) * * On pool creation or the addition of a new top-level vdev, ZFS will * increase the ashift of the top-level vdev to 4096 to match the * max vdev_physical_ashift. * * Example: one or more 512B emulation child vdevs * child->vdev_ashift = 9 (512 bytes) * child->vdev_physical_ashift = 9 (512 bytes) * zfs_max_auto_ashift = 13 (8192 bytes) * zfs_min_auto_ashift = 12 (4096 bytes) * * On pool creation or the addition of a new top-level vdev, ZFS will * increase the ashift of the top-level vdev to 4096 to match the * zfs_min_auto_ashift. */ static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT; static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT; static int sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS) { uint64_t val; int err; val = zfs_max_auto_ashift; err = sysctl_handle_64(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift) return (EINVAL); zfs_max_auto_ashift = val; return (0); } SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift, CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t), sysctl_vfs_zfs_max_auto_ashift, "QU", "Max ashift used when optimising for logical -> physical sectors size on " "new top-level vdevs."); static int sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS) { uint64_t val; int err; val = zfs_min_auto_ashift; err = sysctl_handle_64(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift) return (EINVAL); zfs_min_auto_ashift = val; return (0); } SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift, CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t), sysctl_vfs_zfs_min_auto_ashift, "QU", "Min ashift used when creating new top-level vdevs."); static vdev_ops_t *vdev_ops_table[] = { &vdev_root_ops, &vdev_raidz_ops, &vdev_mirror_ops, &vdev_replacing_ops, &vdev_spare_ops, #ifdef _KERNEL &vdev_geom_ops, #else &vdev_disk_ops, #endif &vdev_file_ops, &vdev_missing_ops, &vdev_hole_ops, &vdev_indirect_ops, NULL }; /* maximum number of metaslabs per top-level vdev */ int vdev_max_ms_count = 200; SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, max_ms_count, CTLFLAG_RDTUN, &vdev_max_ms_count, 0, "Maximum number of metaslabs per top-level vdev"); /* minimum amount of metaslabs per top-level vdev */ int vdev_min_ms_count = 16; SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, min_ms_count, CTLFLAG_RDTUN, &vdev_min_ms_count, 0, "Minimum number of metaslabs per top-level vdev"); /* see comment in vdev_metaslab_set_size() */ int vdev_default_ms_shift = 29; SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, default_ms_shift, CTLFLAG_RDTUN, &vdev_default_ms_shift, 0, "Shift between vdev size and number of metaslabs"); boolean_t vdev_validate_skip = B_FALSE; /* * Since the DTL space map of a vdev is not expected to have a lot of * entries, we default its block size to 4K. */ int vdev_dtl_sm_blksz = (1 << 12); SYSCTL_INT(_vfs_zfs, OID_AUTO, dtl_sm_blksz, CTLFLAG_RDTUN, &vdev_dtl_sm_blksz, 0, "Block size for DTL space map. Power of 2 and greater than 4096."); /* * vdev-wide space maps that have lots of entries written to them at * the end of each transaction can benefit from a higher I/O bandwidth * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. */ int vdev_standard_sm_blksz = (1 << 17); SYSCTL_INT(_vfs_zfs, OID_AUTO, standard_sm_blksz, CTLFLAG_RDTUN, &vdev_standard_sm_blksz, 0, "Block size for standard space map. Power of 2 and greater than 4096."); /*PRINTFLIKE2*/ void vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) { va_list adx; char buf[256]; va_start(adx, fmt); (void) vsnprintf(buf, sizeof (buf), fmt, adx); va_end(adx); if (vd->vdev_path != NULL) { zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, vd->vdev_path, buf); } else { zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", vd->vdev_ops->vdev_op_type, (u_longlong_t)vd->vdev_id, (u_longlong_t)vd->vdev_guid, buf); } } void vdev_dbgmsg_print_tree(vdev_t *vd, int indent) { char state[20]; if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, vd->vdev_ops->vdev_op_type); return; } switch (vd->vdev_state) { case VDEV_STATE_UNKNOWN: (void) snprintf(state, sizeof (state), "unknown"); break; case VDEV_STATE_CLOSED: (void) snprintf(state, sizeof (state), "closed"); break; case VDEV_STATE_OFFLINE: (void) snprintf(state, sizeof (state), "offline"); break; case VDEV_STATE_REMOVED: (void) snprintf(state, sizeof (state), "removed"); break; case VDEV_STATE_CANT_OPEN: (void) snprintf(state, sizeof (state), "can't open"); break; case VDEV_STATE_FAULTED: (void) snprintf(state, sizeof (state), "faulted"); break; case VDEV_STATE_DEGRADED: (void) snprintf(state, sizeof (state), "degraded"); break; case VDEV_STATE_HEALTHY: (void) snprintf(state, sizeof (state), "healthy"); break; default: (void) snprintf(state, sizeof (state), "", (uint_t)vd->vdev_state); } zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, "", (int)vd->vdev_id, vd->vdev_ops->vdev_op_type, vd->vdev_islog ? " (log)" : "", (u_longlong_t)vd->vdev_guid, vd->vdev_path ? vd->vdev_path : "N/A", state); for (uint64_t i = 0; i < vd->vdev_children; i++) vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); } /* * Given a vdev type, return the appropriate ops vector. */ static vdev_ops_t * vdev_getops(const char *type) { vdev_ops_t *ops, **opspp; for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) if (strcmp(ops->vdev_op_type, type) == 0) break; return (ops); } /* * Default asize function: return the MAX of psize with the asize of * all children. This is what's used by anything other than RAID-Z. */ uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize) { uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); uint64_t csize; for (int c = 0; c < vd->vdev_children; c++) { csize = vdev_psize_to_asize(vd->vdev_child[c], psize); asize = MAX(asize, csize); } return (asize); } /* * Get the minimum allocatable size. We define the allocatable size as * the vdev's asize rounded to the nearest metaslab. This allows us to * replace or attach devices which don't have the same physical size but * can still satisfy the same number of allocations. */ uint64_t vdev_get_min_asize(vdev_t *vd) { vdev_t *pvd = vd->vdev_parent; /* * If our parent is NULL (inactive spare or cache) or is the root, * just return our own asize. */ if (pvd == NULL) return (vd->vdev_asize); /* * The top-level vdev just returns the allocatable size rounded * to the nearest metaslab. */ if (vd == vd->vdev_top) return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); /* * The allocatable space for a raidz vdev is N * sizeof(smallest child), * so each child must provide at least 1/Nth of its asize. */ if (pvd->vdev_ops == &vdev_raidz_ops) return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / pvd->vdev_children); return (pvd->vdev_min_asize); } void vdev_set_min_asize(vdev_t *vd) { vd->vdev_min_asize = vdev_get_min_asize(vd); for (int c = 0; c < vd->vdev_children; c++) vdev_set_min_asize(vd->vdev_child[c]); } vdev_t * vdev_lookup_top(spa_t *spa, uint64_t vdev) { vdev_t *rvd = spa->spa_root_vdev; ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); if (vdev < rvd->vdev_children) { ASSERT(rvd->vdev_child[vdev] != NULL); return (rvd->vdev_child[vdev]); } return (NULL); } vdev_t * vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) { vdev_t *mvd; if (vd->vdev_guid == guid) return (vd); for (int c = 0; c < vd->vdev_children; c++) if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != NULL) return (mvd); return (NULL); } static int vdev_count_leaves_impl(vdev_t *vd) { int n = 0; if (vd->vdev_ops->vdev_op_leaf) return (1); for (int c = 0; c < vd->vdev_children; c++) n += vdev_count_leaves_impl(vd->vdev_child[c]); return (n); } int vdev_count_leaves(spa_t *spa) { return (vdev_count_leaves_impl(spa->spa_root_vdev)); } void vdev_add_child(vdev_t *pvd, vdev_t *cvd) { size_t oldsize, newsize; uint64_t id = cvd->vdev_id; vdev_t **newchild; spa_t *spa = cvd->vdev_spa; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); ASSERT(cvd->vdev_parent == NULL); cvd->vdev_parent = pvd; if (pvd == NULL) return; ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); oldsize = pvd->vdev_children * sizeof (vdev_t *); pvd->vdev_children = MAX(pvd->vdev_children, id + 1); newsize = pvd->vdev_children * sizeof (vdev_t *); newchild = kmem_zalloc(newsize, KM_SLEEP); if (pvd->vdev_child != NULL) { bcopy(pvd->vdev_child, newchild, oldsize); kmem_free(pvd->vdev_child, oldsize); } pvd->vdev_child = newchild; pvd->vdev_child[id] = cvd; cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); /* * Walk up all ancestors to update guid sum. */ for (; pvd != NULL; pvd = pvd->vdev_parent) pvd->vdev_guid_sum += cvd->vdev_guid_sum; } void vdev_remove_child(vdev_t *pvd, vdev_t *cvd) { int c; uint_t id = cvd->vdev_id; ASSERT(cvd->vdev_parent == pvd); if (pvd == NULL) return; ASSERT(id < pvd->vdev_children); ASSERT(pvd->vdev_child[id] == cvd); pvd->vdev_child[id] = NULL; cvd->vdev_parent = NULL; for (c = 0; c < pvd->vdev_children; c++) if (pvd->vdev_child[c]) break; if (c == pvd->vdev_children) { kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); pvd->vdev_child = NULL; pvd->vdev_children = 0; } /* * Walk up all ancestors to update guid sum. */ for (; pvd != NULL; pvd = pvd->vdev_parent) pvd->vdev_guid_sum -= cvd->vdev_guid_sum; } /* * Remove any holes in the child array. */ void vdev_compact_children(vdev_t *pvd) { vdev_t **newchild, *cvd; int oldc = pvd->vdev_children; int newc; ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); for (int c = newc = 0; c < oldc; c++) if (pvd->vdev_child[c]) newc++; newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); for (int c = newc = 0; c < oldc; c++) { if ((cvd = pvd->vdev_child[c]) != NULL) { newchild[newc] = cvd; cvd->vdev_id = newc++; } } kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); pvd->vdev_child = newchild; pvd->vdev_children = newc; } /* * Allocate and minimally initialize a vdev_t. */ vdev_t * vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) { vdev_t *vd; vdev_indirect_config_t *vic; vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); vic = &vd->vdev_indirect_config; if (spa->spa_root_vdev == NULL) { ASSERT(ops == &vdev_root_ops); spa->spa_root_vdev = vd; spa->spa_load_guid = spa_generate_guid(NULL); } if (guid == 0 && ops != &vdev_hole_ops) { if (spa->spa_root_vdev == vd) { /* * The root vdev's guid will also be the pool guid, * which must be unique among all pools. */ guid = spa_generate_guid(NULL); } else { /* * Any other vdev's guid must be unique within the pool. */ guid = spa_generate_guid(spa); } ASSERT(!spa_guid_exists(spa_guid(spa), guid)); } vd->vdev_spa = spa; vd->vdev_id = id; vd->vdev_guid = guid; vd->vdev_guid_sum = guid; vd->vdev_ops = ops; vd->vdev_state = VDEV_STATE_CLOSED; vd->vdev_ishole = (ops == &vdev_hole_ops); vic->vic_prev_indirect_vdev = UINT64_MAX; rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); vd->vdev_obsolete_segments = range_tree_create(NULL, NULL); mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&vd->vdev_scan_io_queue_lock, NULL, MUTEX_DEFAULT, NULL); for (int t = 0; t < DTL_TYPES; t++) { vd->vdev_dtl[t] = range_tree_create(NULL, NULL); } txg_list_create(&vd->vdev_ms_list, spa, offsetof(struct metaslab, ms_txg_node)); txg_list_create(&vd->vdev_dtl_list, spa, offsetof(struct vdev, vdev_dtl_node)); vd->vdev_stat.vs_timestamp = gethrtime(); vdev_queue_init(vd); vdev_cache_init(vd); return (vd); } /* * Allocate a new vdev. The 'alloctype' is used to control whether we are * creating a new vdev or loading an existing one - the behavior is slightly * different for each case. */ int vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, int alloctype) { vdev_ops_t *ops; char *type; uint64_t guid = 0, islog, nparity; vdev_t *vd; vdev_indirect_config_t *vic; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) return (SET_ERROR(EINVAL)); if ((ops = vdev_getops(type)) == NULL) return (SET_ERROR(EINVAL)); /* * If this is a load, get the vdev guid from the nvlist. * Otherwise, vdev_alloc_common() will generate one for us. */ if (alloctype == VDEV_ALLOC_LOAD) { uint64_t label_id; if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || label_id != id) return (SET_ERROR(EINVAL)); if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) return (SET_ERROR(EINVAL)); } else if (alloctype == VDEV_ALLOC_SPARE) { if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) return (SET_ERROR(EINVAL)); } else if (alloctype == VDEV_ALLOC_L2CACHE) { if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) return (SET_ERROR(EINVAL)); } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) return (SET_ERROR(EINVAL)); } /* * The first allocated vdev must be of type 'root'. */ if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) return (SET_ERROR(EINVAL)); /* * Determine whether we're a log vdev. */ islog = 0; (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); if (islog && spa_version(spa) < SPA_VERSION_SLOGS) return (SET_ERROR(ENOTSUP)); if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) return (SET_ERROR(ENOTSUP)); /* * Set the nparity property for RAID-Z vdevs. */ nparity = -1ULL; if (ops == &vdev_raidz_ops) { if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, &nparity) == 0) { if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) return (SET_ERROR(EINVAL)); /* * Previous versions could only support 1 or 2 parity * device. */ if (nparity > 1 && spa_version(spa) < SPA_VERSION_RAIDZ2) return (SET_ERROR(ENOTSUP)); if (nparity > 2 && spa_version(spa) < SPA_VERSION_RAIDZ3) return (SET_ERROR(ENOTSUP)); } else { /* * We require the parity to be specified for SPAs that * support multiple parity levels. */ if (spa_version(spa) >= SPA_VERSION_RAIDZ2) return (SET_ERROR(EINVAL)); /* * Otherwise, we default to 1 parity device for RAID-Z. */ nparity = 1; } } else { nparity = 0; } ASSERT(nparity != -1ULL); vd = vdev_alloc_common(spa, id, guid, ops); vic = &vd->vdev_indirect_config; vd->vdev_islog = islog; vd->vdev_nparity = nparity; if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) vd->vdev_path = spa_strdup(vd->vdev_path); if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) vd->vdev_devid = spa_strdup(vd->vdev_devid); if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, &vd->vdev_physpath) == 0) vd->vdev_physpath = spa_strdup(vd->vdev_physpath); if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) vd->vdev_fru = spa_strdup(vd->vdev_fru); /* * Set the whole_disk property. If it's not specified, leave the value * as -1. */ if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &vd->vdev_wholedisk) != 0) vd->vdev_wholedisk = -1ULL; ASSERT0(vic->vic_mapping_object); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, &vic->vic_mapping_object); ASSERT0(vic->vic_births_object); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, &vic->vic_births_object); ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, &vic->vic_prev_indirect_vdev); /* * Look for the 'not present' flag. This will only be set if the device * was not present at the time of import. */ (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &vd->vdev_not_present); /* * Get the alignment requirement. */ (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); /* * Retrieve the vdev creation time. */ (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, &vd->vdev_crtxg); /* * If we're a top-level vdev, try to load the allocation parameters. */ if (parent && !parent->vdev_parent && (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, &vd->vdev_ms_array); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, &vd->vdev_ms_shift); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, &vd->vdev_asize); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, &vd->vdev_removing); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, &vd->vdev_top_zap); } else { ASSERT0(vd->vdev_top_zap); } if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { ASSERT(alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_ADD || alloctype == VDEV_ALLOC_SPLIT || alloctype == VDEV_ALLOC_ROOTPOOL); vd->vdev_mg = metaslab_group_create(islog ? - spa_log_class(spa) : spa_normal_class(spa), vd); + spa_log_class(spa) : spa_normal_class(spa), vd, + spa->spa_alloc_count); } if (vd->vdev_ops->vdev_op_leaf && (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); } else { ASSERT0(vd->vdev_leaf_zap); } /* * If we're a leaf vdev, try to load the DTL object and other state. */ if (vd->vdev_ops->vdev_op_leaf && (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || alloctype == VDEV_ALLOC_ROOTPOOL)) { if (alloctype == VDEV_ALLOC_LOAD) { (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, &vd->vdev_dtl_object); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, &vd->vdev_unspare); } if (alloctype == VDEV_ALLOC_ROOTPOOL) { uint64_t spare = 0; if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, &spare) == 0 && spare) spa_spare_add(vd); } (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &vd->vdev_offline); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, &vd->vdev_resilver_txg); /* * When importing a pool, we want to ignore the persistent fault * state, as the diagnosis made on another system may not be * valid in the current context. Local vdevs will * remain in the faulted state. */ if (spa_load_state(spa) == SPA_LOAD_OPEN) { (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &vd->vdev_faulted); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, &vd->vdev_degraded); (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &vd->vdev_removed); if (vd->vdev_faulted || vd->vdev_degraded) { char *aux; vd->vdev_label_aux = VDEV_AUX_ERR_EXCEEDED; if (nvlist_lookup_string(nv, ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && strcmp(aux, "external") == 0) vd->vdev_label_aux = VDEV_AUX_EXTERNAL; } } } /* * Add ourselves to the parent's list of children. */ vdev_add_child(parent, vd); *vdp = vd; return (0); } void vdev_free(vdev_t *vd) { spa_t *spa = vd->vdev_spa; /* * Scan queues are normally destroyed at the end of a scan. If the * queue exists here, that implies the vdev is being removed while * the scan is still running. */ if (vd->vdev_scan_io_queue != NULL) { mutex_enter(&vd->vdev_scan_io_queue_lock); dsl_scan_io_queue_destroy(vd->vdev_scan_io_queue); vd->vdev_scan_io_queue = NULL; mutex_exit(&vd->vdev_scan_io_queue_lock); } /* * vdev_free() implies closing the vdev first. This is simpler than * trying to ensure complicated semantics for all callers. */ vdev_close(vd); ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); /* * Free all children. */ for (int c = 0; c < vd->vdev_children; c++) vdev_free(vd->vdev_child[c]); ASSERT(vd->vdev_child == NULL); ASSERT(vd->vdev_guid_sum == vd->vdev_guid); /* * Discard allocation state. */ if (vd->vdev_mg != NULL) { vdev_metaslab_fini(vd); metaslab_group_destroy(vd->vdev_mg); } ASSERT0(vd->vdev_stat.vs_space); ASSERT0(vd->vdev_stat.vs_dspace); ASSERT0(vd->vdev_stat.vs_alloc); /* * Remove this vdev from its parent's child list. */ vdev_remove_child(vd->vdev_parent, vd); ASSERT(vd->vdev_parent == NULL); /* * Clean up vdev structure. */ vdev_queue_fini(vd); vdev_cache_fini(vd); if (vd->vdev_path) spa_strfree(vd->vdev_path); if (vd->vdev_devid) spa_strfree(vd->vdev_devid); if (vd->vdev_physpath) spa_strfree(vd->vdev_physpath); if (vd->vdev_fru) spa_strfree(vd->vdev_fru); if (vd->vdev_isspare) spa_spare_remove(vd); if (vd->vdev_isl2cache) spa_l2cache_remove(vd); txg_list_destroy(&vd->vdev_ms_list); txg_list_destroy(&vd->vdev_dtl_list); mutex_enter(&vd->vdev_dtl_lock); space_map_close(vd->vdev_dtl_sm); for (int t = 0; t < DTL_TYPES; t++) { range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); range_tree_destroy(vd->vdev_dtl[t]); } mutex_exit(&vd->vdev_dtl_lock); EQUIV(vd->vdev_indirect_births != NULL, vd->vdev_indirect_mapping != NULL); if (vd->vdev_indirect_births != NULL) { vdev_indirect_mapping_close(vd->vdev_indirect_mapping); vdev_indirect_births_close(vd->vdev_indirect_births); } if (vd->vdev_obsolete_sm != NULL) { ASSERT(vd->vdev_removing || vd->vdev_ops == &vdev_indirect_ops); space_map_close(vd->vdev_obsolete_sm); vd->vdev_obsolete_sm = NULL; } range_tree_destroy(vd->vdev_obsolete_segments); rw_destroy(&vd->vdev_indirect_rwlock); mutex_destroy(&vd->vdev_obsolete_lock); mutex_destroy(&vd->vdev_queue_lock); mutex_destroy(&vd->vdev_dtl_lock); mutex_destroy(&vd->vdev_stat_lock); mutex_destroy(&vd->vdev_probe_lock); mutex_destroy(&vd->vdev_scan_io_queue_lock); if (vd == spa->spa_root_vdev) spa->spa_root_vdev = NULL; kmem_free(vd, sizeof (vdev_t)); } /* * Transfer top-level vdev state from svd to tvd. */ static void vdev_top_transfer(vdev_t *svd, vdev_t *tvd) { spa_t *spa = svd->vdev_spa; metaslab_t *msp; vdev_t *vd; int t; ASSERT(tvd == tvd->vdev_top); tvd->vdev_ms_array = svd->vdev_ms_array; tvd->vdev_ms_shift = svd->vdev_ms_shift; tvd->vdev_ms_count = svd->vdev_ms_count; tvd->vdev_top_zap = svd->vdev_top_zap; svd->vdev_ms_array = 0; svd->vdev_ms_shift = 0; svd->vdev_ms_count = 0; svd->vdev_top_zap = 0; if (tvd->vdev_mg) ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); tvd->vdev_mg = svd->vdev_mg; tvd->vdev_ms = svd->vdev_ms; svd->vdev_mg = NULL; svd->vdev_ms = NULL; if (tvd->vdev_mg != NULL) tvd->vdev_mg->mg_vd = tvd; tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; svd->vdev_checkpoint_sm = NULL; tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; svd->vdev_stat.vs_alloc = 0; svd->vdev_stat.vs_space = 0; svd->vdev_stat.vs_dspace = 0; for (t = 0; t < TXG_SIZE; t++) { while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) (void) txg_list_add(&tvd->vdev_ms_list, msp, t); while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); } if (list_link_active(&svd->vdev_config_dirty_node)) { vdev_config_clean(svd); vdev_config_dirty(tvd); } if (list_link_active(&svd->vdev_state_dirty_node)) { vdev_state_clean(svd); vdev_state_dirty(tvd); } tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; svd->vdev_deflate_ratio = 0; tvd->vdev_islog = svd->vdev_islog; svd->vdev_islog = 0; dsl_scan_io_queue_vdev_xfer(svd, tvd); } static void vdev_top_update(vdev_t *tvd, vdev_t *vd) { if (vd == NULL) return; vd->vdev_top = tvd; for (int c = 0; c < vd->vdev_children; c++) vdev_top_update(tvd, vd->vdev_child[c]); } /* * Add a mirror/replacing vdev above an existing vdev. */ vdev_t * vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) { spa_t *spa = cvd->vdev_spa; vdev_t *pvd = cvd->vdev_parent; vdev_t *mvd; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); mvd->vdev_asize = cvd->vdev_asize; mvd->vdev_min_asize = cvd->vdev_min_asize; mvd->vdev_max_asize = cvd->vdev_max_asize; mvd->vdev_psize = cvd->vdev_psize; mvd->vdev_ashift = cvd->vdev_ashift; mvd->vdev_logical_ashift = cvd->vdev_logical_ashift; mvd->vdev_physical_ashift = cvd->vdev_physical_ashift; mvd->vdev_state = cvd->vdev_state; mvd->vdev_crtxg = cvd->vdev_crtxg; vdev_remove_child(pvd, cvd); vdev_add_child(pvd, mvd); cvd->vdev_id = mvd->vdev_children; vdev_add_child(mvd, cvd); vdev_top_update(cvd->vdev_top, cvd->vdev_top); if (mvd == mvd->vdev_top) vdev_top_transfer(cvd, mvd); return (mvd); } /* * Remove a 1-way mirror/replacing vdev from the tree. */ void vdev_remove_parent(vdev_t *cvd) { vdev_t *mvd = cvd->vdev_parent; vdev_t *pvd = mvd->vdev_parent; ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); ASSERT(mvd->vdev_children == 1); ASSERT(mvd->vdev_ops == &vdev_mirror_ops || mvd->vdev_ops == &vdev_replacing_ops || mvd->vdev_ops == &vdev_spare_ops); cvd->vdev_ashift = mvd->vdev_ashift; cvd->vdev_logical_ashift = mvd->vdev_logical_ashift; cvd->vdev_physical_ashift = mvd->vdev_physical_ashift; vdev_remove_child(mvd, cvd); vdev_remove_child(pvd, mvd); /* * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. * Otherwise, we could have detached an offline device, and when we * go to import the pool we'll think we have two top-level vdevs, * instead of a different version of the same top-level vdev. */ if (mvd->vdev_top == mvd) { uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; cvd->vdev_orig_guid = cvd->vdev_guid; cvd->vdev_guid += guid_delta; cvd->vdev_guid_sum += guid_delta; } cvd->vdev_id = mvd->vdev_id; vdev_add_child(pvd, cvd); vdev_top_update(cvd->vdev_top, cvd->vdev_top); if (cvd == cvd->vdev_top) vdev_top_transfer(mvd, cvd); ASSERT(mvd->vdev_children == 0); vdev_free(mvd); } int vdev_metaslab_init(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; objset_t *mos = spa->spa_meta_objset; uint64_t m; uint64_t oldc = vd->vdev_ms_count; uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; metaslab_t **mspp; int error; ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); /* * This vdev is not being allocated from yet or is a hole. */ if (vd->vdev_ms_shift == 0) return (0); ASSERT(!vd->vdev_ishole); ASSERT(oldc <= newc); mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); if (oldc != 0) { bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); } vd->vdev_ms = mspp; vd->vdev_ms_count = newc; - for (m = oldc; m < newc; m++) { uint64_t object = 0; /* * vdev_ms_array may be 0 if we are creating the "fake" * metaslabs for an indirect vdev for zdb's leak detection. * See zdb_leak_init(). */ if (txg == 0 && vd->vdev_ms_array != 0) { error = dmu_read(mos, vd->vdev_ms_array, m * sizeof (uint64_t), sizeof (uint64_t), &object, DMU_READ_PREFETCH); if (error != 0) { vdev_dbgmsg(vd, "unable to read the metaslab " "array [error=%d]", error); return (error); } } error = metaslab_init(vd->vdev_mg, m, object, txg, &(vd->vdev_ms[m])); if (error != 0) { vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", error); return (error); } } if (txg == 0) spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); /* * If the vdev is being removed we don't activate * the metaslabs since we want to ensure that no new * allocations are performed on this device. */ if (oldc == 0 && !vd->vdev_removing) metaslab_group_activate(vd->vdev_mg); if (txg == 0) spa_config_exit(spa, SCL_ALLOC, FTAG); return (0); } void vdev_metaslab_fini(vdev_t *vd) { if (vd->vdev_checkpoint_sm != NULL) { ASSERT(spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_POOL_CHECKPOINT)); space_map_close(vd->vdev_checkpoint_sm); /* * Even though we close the space map, we need to set its * pointer to NULL. The reason is that vdev_metaslab_fini() * may be called multiple times for certain operations * (i.e. when destroying a pool) so we need to ensure that * this clause never executes twice. This logic is similar * to the one used for the vdev_ms clause below. */ vd->vdev_checkpoint_sm = NULL; } if (vd->vdev_ms != NULL) { uint64_t count = vd->vdev_ms_count; metaslab_group_passivate(vd->vdev_mg); for (uint64_t m = 0; m < count; m++) { metaslab_t *msp = vd->vdev_ms[m]; if (msp != NULL) metaslab_fini(msp); } kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); vd->vdev_ms = NULL; vd->vdev_ms_count = 0; } ASSERT0(vd->vdev_ms_count); } typedef struct vdev_probe_stats { boolean_t vps_readable; boolean_t vps_writeable; int vps_flags; } vdev_probe_stats_t; static void vdev_probe_done(zio_t *zio) { spa_t *spa = zio->io_spa; vdev_t *vd = zio->io_vd; vdev_probe_stats_t *vps = zio->io_private; ASSERT(vd->vdev_probe_zio != NULL); if (zio->io_type == ZIO_TYPE_READ) { if (zio->io_error == 0) vps->vps_readable = 1; if (zio->io_error == 0 && spa_writeable(spa)) { zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, zio->io_offset, zio->io_size, zio->io_abd, ZIO_CHECKSUM_OFF, vdev_probe_done, vps, ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); } else { abd_free(zio->io_abd); } } else if (zio->io_type == ZIO_TYPE_WRITE) { if (zio->io_error == 0) vps->vps_writeable = 1; abd_free(zio->io_abd); } else if (zio->io_type == ZIO_TYPE_NULL) { zio_t *pio; vd->vdev_cant_read |= !vps->vps_readable; vd->vdev_cant_write |= !vps->vps_writeable; if (vdev_readable(vd) && (vdev_writeable(vd) || !spa_writeable(spa))) { zio->io_error = 0; } else { ASSERT(zio->io_error != 0); vdev_dbgmsg(vd, "failed probe"); zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, spa, vd, NULL, 0, 0); zio->io_error = SET_ERROR(ENXIO); } mutex_enter(&vd->vdev_probe_lock); ASSERT(vd->vdev_probe_zio == zio); vd->vdev_probe_zio = NULL; mutex_exit(&vd->vdev_probe_lock); zio_link_t *zl = NULL; while ((pio = zio_walk_parents(zio, &zl)) != NULL) if (!vdev_accessible(vd, pio)) pio->io_error = SET_ERROR(ENXIO); kmem_free(vps, sizeof (*vps)); } } /* * Determine whether this device is accessible. * * Read and write to several known locations: the pad regions of each * vdev label but the first, which we leave alone in case it contains * a VTOC. */ zio_t * vdev_probe(vdev_t *vd, zio_t *zio) { spa_t *spa = vd->vdev_spa; vdev_probe_stats_t *vps = NULL; zio_t *pio; ASSERT(vd->vdev_ops->vdev_op_leaf); /* * Don't probe the probe. */ if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) return (NULL); /* * To prevent 'probe storms' when a device fails, we create * just one probe i/o at a time. All zios that want to probe * this vdev will become parents of the probe io. */ mutex_enter(&vd->vdev_probe_lock); if ((pio = vd->vdev_probe_zio) == NULL) { vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | ZIO_FLAG_TRYHARD; if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { /* * vdev_cant_read and vdev_cant_write can only * transition from TRUE to FALSE when we have the * SCL_ZIO lock as writer; otherwise they can only * transition from FALSE to TRUE. This ensures that * any zio looking at these values can assume that * failures persist for the life of the I/O. That's * important because when a device has intermittent * connectivity problems, we want to ensure that * they're ascribed to the device (ENXIO) and not * the zio (EIO). * * Since we hold SCL_ZIO as writer here, clear both * values so the probe can reevaluate from first * principles. */ vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; vd->vdev_cant_read = B_FALSE; vd->vdev_cant_write = B_FALSE; } vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, vdev_probe_done, vps, vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); /* * We can't change the vdev state in this context, so we * kick off an async task to do it on our behalf. */ if (zio != NULL) { vd->vdev_probe_wanted = B_TRUE; spa_async_request(spa, SPA_ASYNC_PROBE); } } if (zio != NULL) zio_add_child(zio, pio); mutex_exit(&vd->vdev_probe_lock); if (vps == NULL) { ASSERT(zio != NULL); return (NULL); } for (int l = 1; l < VDEV_LABELS; l++) { zio_nowait(zio_read_phys(pio, vd, vdev_label_offset(vd->vdev_psize, l, offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), ZIO_CHECKSUM_OFF, vdev_probe_done, vps, ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); } if (zio == NULL) return (pio); zio_nowait(pio); return (NULL); } static void vdev_open_child(void *arg) { vdev_t *vd = arg; vd->vdev_open_thread = curthread; vd->vdev_open_error = vdev_open(vd); vd->vdev_open_thread = NULL; } boolean_t vdev_uses_zvols(vdev_t *vd) { if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, strlen(ZVOL_DIR)) == 0) return (B_TRUE); for (int c = 0; c < vd->vdev_children; c++) if (vdev_uses_zvols(vd->vdev_child[c])) return (B_TRUE); return (B_FALSE); } void vdev_open_children(vdev_t *vd) { taskq_t *tq; int children = vd->vdev_children; /* * in order to handle pools on top of zvols, do the opens * in a single thread so that the same thread holds the * spa_namespace_lock */ if (B_TRUE || vdev_uses_zvols(vd)) { for (int c = 0; c < children; c++) vd->vdev_child[c]->vdev_open_error = vdev_open(vd->vdev_child[c]); return; } tq = taskq_create("vdev_open", children, minclsyspri, children, children, TASKQ_PREPOPULATE); for (int c = 0; c < children; c++) VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], TQ_SLEEP) != 0); taskq_destroy(tq); } /* * Compute the raidz-deflation ratio. Note, we hard-code * in 128k (1 << 17) because it is the "typical" blocksize. * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, * otherwise it would inconsistently account for existing bp's. */ static void vdev_set_deflate_ratio(vdev_t *vd) { if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { vd->vdev_deflate_ratio = (1 << 17) / (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); } } /* * Prepare a virtual device for access. */ int vdev_open(vdev_t *vd) { spa_t *spa = vd->vdev_spa; int error; uint64_t osize = 0; uint64_t max_osize = 0; uint64_t asize, max_asize, psize; uint64_t logical_ashift = 0; uint64_t physical_ashift = 0; ASSERT(vd->vdev_open_thread == curthread || spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || vd->vdev_state == VDEV_STATE_CANT_OPEN || vd->vdev_state == VDEV_STATE_OFFLINE); vd->vdev_stat.vs_aux = VDEV_AUX_NONE; vd->vdev_cant_read = B_FALSE; vd->vdev_cant_write = B_FALSE; vd->vdev_notrim = B_FALSE; vd->vdev_min_asize = vdev_get_min_asize(vd); /* * If this vdev is not removed, check its fault status. If it's * faulted, bail out of the open. */ if (!vd->vdev_removed && vd->vdev_faulted) { ASSERT(vd->vdev_children == 0); ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || vd->vdev_label_aux == VDEV_AUX_EXTERNAL); vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, vd->vdev_label_aux); return (SET_ERROR(ENXIO)); } else if (vd->vdev_offline) { ASSERT(vd->vdev_children == 0); vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); return (SET_ERROR(ENXIO)); } error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &logical_ashift, &physical_ashift); /* * Reset the vdev_reopening flag so that we actually close * the vdev on error. */ vd->vdev_reopening = B_FALSE; if (zio_injection_enabled && error == 0) error = zio_handle_device_injection(vd, NULL, ENXIO); if (error) { if (vd->vdev_removed && vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) vd->vdev_removed = B_FALSE; if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, vd->vdev_stat.vs_aux); } else { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, vd->vdev_stat.vs_aux); } return (error); } vd->vdev_removed = B_FALSE; /* * Recheck the faulted flag now that we have confirmed that * the vdev is accessible. If we're faulted, bail. */ if (vd->vdev_faulted) { ASSERT(vd->vdev_children == 0); ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || vd->vdev_label_aux == VDEV_AUX_EXTERNAL); vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, vd->vdev_label_aux); return (SET_ERROR(ENXIO)); } if (vd->vdev_degraded) { ASSERT(vd->vdev_children == 0); vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, VDEV_AUX_ERR_EXCEEDED); } else { vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); } /* * For hole or missing vdevs we just return success. */ if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) return (0); if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf) trim_map_create(vd); for (int c = 0; c < vd->vdev_children; c++) { if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, VDEV_AUX_NONE); break; } } osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); if (vd->vdev_children == 0) { if (osize < SPA_MINDEVSIZE) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_TOO_SMALL); return (SET_ERROR(EOVERFLOW)); } psize = osize; asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); max_asize = max_osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); } else { if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_TOO_SMALL); return (SET_ERROR(EOVERFLOW)); } psize = 0; asize = osize; max_asize = max_osize; } vd->vdev_psize = psize; /* * Make sure the allocatable size hasn't shrunk too much. */ if (asize < vd->vdev_min_asize) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_BAD_LABEL); return (SET_ERROR(EINVAL)); } vd->vdev_physical_ashift = MAX(physical_ashift, vd->vdev_physical_ashift); vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift); vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift); if (vd->vdev_logical_ashift > SPA_MAXASHIFT) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_ASHIFT_TOO_BIG); return (EINVAL); } if (vd->vdev_asize == 0) { /* * This is the first-ever open, so use the computed values. * For testing purposes, a higher ashift can be requested. */ vd->vdev_asize = asize; vd->vdev_max_asize = max_asize; } else { /* * Make sure the alignment requirement hasn't increased. */ if (vd->vdev_ashift > vd->vdev_top->vdev_ashift && vd->vdev_ops->vdev_op_leaf) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_BAD_LABEL); return (EINVAL); } vd->vdev_max_asize = max_asize; } /* * If all children are healthy we update asize if either: * The asize has increased, due to a device expansion caused by dynamic * LUN growth or vdev replacement, and automatic expansion is enabled; * making the additional space available. * * The asize has decreased, due to a device shrink usually caused by a * vdev replace with a smaller device. This ensures that calculations * based of max_asize and asize e.g. esize are always valid. It's safe * to do this as we've already validated that asize is greater than * vdev_min_asize. */ if (vd->vdev_state == VDEV_STATE_HEALTHY && ((asize > vd->vdev_asize && (vd->vdev_expanding || spa->spa_autoexpand)) || (asize < vd->vdev_asize))) vd->vdev_asize = asize; vdev_set_min_asize(vd); /* * Ensure we can issue some IO before declaring the * vdev open for business. */ if (vd->vdev_ops->vdev_op_leaf && (error = zio_wait(vdev_probe(vd, NULL))) != 0) { vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, VDEV_AUX_ERR_EXCEEDED); return (error); } /* * Track the min and max ashift values for normal data devices. */ if (vd->vdev_top == vd && vd->vdev_ashift != 0 && !vd->vdev_islog && vd->vdev_aux == NULL) { if (vd->vdev_ashift > spa->spa_max_ashift) spa->spa_max_ashift = vd->vdev_ashift; if (vd->vdev_ashift < spa->spa_min_ashift) spa->spa_min_ashift = vd->vdev_ashift; } /* * If a leaf vdev has a DTL, and seems healthy, then kick off a * resilver. But don't do this if we are doing a reopen for a scrub, * since this would just restart the scrub we are already doing. */ if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && vdev_resilver_needed(vd, NULL, NULL)) spa_async_request(spa, SPA_ASYNC_RESILVER); return (0); } /* * Called once the vdevs are all opened, this routine validates the label * contents. This needs to be done before vdev_load() so that we don't * inadvertently do repair I/Os to the wrong device. * * This function will only return failure if one of the vdevs indicates that it * has since been destroyed or exported. This is only possible if * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state * will be updated but the function will return 0. */ int vdev_validate(vdev_t *vd) { spa_t *spa = vd->vdev_spa; nvlist_t *label; uint64_t guid = 0, aux_guid = 0, top_guid; uint64_t state; nvlist_t *nvl; uint64_t txg; if (vdev_validate_skip) return (0); for (uint64_t c = 0; c < vd->vdev_children; c++) if (vdev_validate(vd->vdev_child[c]) != 0) return (SET_ERROR(EBADF)); /* * If the device has already failed, or was marked offline, don't do * any further validation. Otherwise, label I/O will fail and we will * overwrite the previous state. */ if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) return (0); /* * If we are performing an extreme rewind, we allow for a label that * was modified at a point after the current txg. * If config lock is not held do not check for the txg. spa_sync could * be updating the vdev's label before updating spa_last_synced_txg. */ if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 || spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG) txg = UINT64_MAX; else txg = spa_last_synced_txg(spa); if ((label = vdev_label_read_config(vd, txg)) == NULL) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_BAD_LABEL); vdev_dbgmsg(vd, "vdev_validate: failed reading config for " "txg %llu", (u_longlong_t)txg); return (0); } /* * Determine if this vdev has been split off into another * pool. If so, then refuse to open it. */ if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, &aux_guid) == 0 && aux_guid == spa_guid(spa)) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_SPLIT_POOL); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); return (0); } if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", ZPOOL_CONFIG_POOL_GUID); return (0); } /* * If config is not trusted then ignore the spa guid check. This is * necessary because if the machine crashed during a re-guid the new * guid might have been written to all of the vdev labels, but not the * cached config. The check will be performed again once we have the * trusted config from the MOS. */ if (spa->spa_trust_config && guid != spa_guid(spa)) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " "match config (%llu != %llu)", (u_longlong_t)guid, (u_longlong_t)spa_guid(spa)); return (0); } if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, &aux_guid) != 0) aux_guid = 0; if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", ZPOOL_CONFIG_GUID); return (0); } if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", ZPOOL_CONFIG_TOP_GUID); return (0); } /* * If this vdev just became a top-level vdev because its sibling was * detached, it will have adopted the parent's vdev guid -- but the * label may or may not be on disk yet. Fortunately, either version * of the label will have the same top guid, so if we're a top-level * vdev, we can safely compare to that instead. * However, if the config comes from a cachefile that failed to update * after the detach, a top-level vdev will appear as a non top-level * vdev in the config. Also relax the constraints if we perform an * extreme rewind. * * If we split this vdev off instead, then we also check the * original pool's guid. We don't want to consider the vdev * corrupt if it is partway through a split operation. */ if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { boolean_t mismatch = B_FALSE; if (spa->spa_trust_config && !spa->spa_extreme_rewind) { if (vd != vd->vdev_top || vd->vdev_guid != top_guid) mismatch = B_TRUE; } else { if (vd->vdev_guid != top_guid && vd->vdev_top->vdev_guid != guid) mismatch = B_TRUE; } if (mismatch) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: config guid " "doesn't match label guid"); vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", (u_longlong_t)vd->vdev_guid, (u_longlong_t)vd->vdev_top->vdev_guid); vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " "aux_guid %llu", (u_longlong_t)guid, (u_longlong_t)top_guid, (u_longlong_t)aux_guid); return (0); } } if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", ZPOOL_CONFIG_POOL_STATE); return (0); } nvlist_free(label); /* * If this is a verbatim import, no need to check the * state of the pool. */ if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && spa_load_state(spa) == SPA_LOAD_OPEN && state != POOL_STATE_ACTIVE) { vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " "for spa %s", (u_longlong_t)state, spa->spa_name); return (SET_ERROR(EBADF)); } /* * If we were able to open and validate a vdev that was * previously marked permanently unavailable, clear that state * now. */ if (vd->vdev_not_present) vd->vdev_not_present = 0; return (0); } static void vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) { if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, dvd->vdev_path, svd->vdev_path); spa_strfree(dvd->vdev_path); dvd->vdev_path = spa_strdup(svd->vdev_path); } } else if (svd->vdev_path != NULL) { dvd->vdev_path = spa_strdup(svd->vdev_path); zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); } } /* * Recursively copy vdev paths from one vdev to another. Source and destination * vdev trees must have same geometry otherwise return error. Intended to copy * paths from userland config into MOS config. */ int vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) { if ((svd->vdev_ops == &vdev_missing_ops) || (svd->vdev_ishole && dvd->vdev_ishole) || (dvd->vdev_ops == &vdev_indirect_ops)) return (0); if (svd->vdev_ops != dvd->vdev_ops) { vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); return (SET_ERROR(EINVAL)); } if (svd->vdev_guid != dvd->vdev_guid) { vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " "%llu)", (u_longlong_t)svd->vdev_guid, (u_longlong_t)dvd->vdev_guid); return (SET_ERROR(EINVAL)); } if (svd->vdev_children != dvd->vdev_children) { vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " "%llu != %llu", (u_longlong_t)svd->vdev_children, (u_longlong_t)dvd->vdev_children); return (SET_ERROR(EINVAL)); } for (uint64_t i = 0; i < svd->vdev_children; i++) { int error = vdev_copy_path_strict(svd->vdev_child[i], dvd->vdev_child[i]); if (error != 0) return (error); } if (svd->vdev_ops->vdev_op_leaf) vdev_copy_path_impl(svd, dvd); return (0); } static void vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) { ASSERT(stvd->vdev_top == stvd); ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); for (uint64_t i = 0; i < dvd->vdev_children; i++) { vdev_copy_path_search(stvd, dvd->vdev_child[i]); } if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) return; /* * The idea here is that while a vdev can shift positions within * a top vdev (when replacing, attaching mirror, etc.) it cannot * step outside of it. */ vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) return; ASSERT(vd->vdev_ops->vdev_op_leaf); vdev_copy_path_impl(vd, dvd); } /* * Recursively copy vdev paths from one root vdev to another. Source and * destination vdev trees may differ in geometry. For each destination leaf * vdev, search a vdev with the same guid and top vdev id in the source. * Intended to copy paths from userland config into MOS config. */ void vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) { uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); ASSERT(srvd->vdev_ops == &vdev_root_ops); ASSERT(drvd->vdev_ops == &vdev_root_ops); for (uint64_t i = 0; i < children; i++) { vdev_copy_path_search(srvd->vdev_child[i], drvd->vdev_child[i]); } } /* * Close a virtual device. */ void vdev_close(vdev_t *vd) { spa_t *spa = vd->vdev_spa; vdev_t *pvd = vd->vdev_parent; ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); /* * If our parent is reopening, then we are as well, unless we are * going offline. */ if (pvd != NULL && pvd->vdev_reopening) vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); vd->vdev_ops->vdev_op_close(vd); vdev_cache_purge(vd); if (vd->vdev_ops->vdev_op_leaf) trim_map_destroy(vd); /* * We record the previous state before we close it, so that if we are * doing a reopen(), we don't generate FMA ereports if we notice that * it's still faulted. */ vd->vdev_prevstate = vd->vdev_state; if (vd->vdev_offline) vd->vdev_state = VDEV_STATE_OFFLINE; else vd->vdev_state = VDEV_STATE_CLOSED; vd->vdev_stat.vs_aux = VDEV_AUX_NONE; } void vdev_hold(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_is_root(spa)); if (spa->spa_state == POOL_STATE_UNINITIALIZED) return; for (int c = 0; c < vd->vdev_children; c++) vdev_hold(vd->vdev_child[c]); if (vd->vdev_ops->vdev_op_leaf) vd->vdev_ops->vdev_op_hold(vd); } void vdev_rele(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_is_root(spa)); for (int c = 0; c < vd->vdev_children; c++) vdev_rele(vd->vdev_child[c]); if (vd->vdev_ops->vdev_op_leaf) vd->vdev_ops->vdev_op_rele(vd); } /* * Reopen all interior vdevs and any unopened leaves. We don't actually * reopen leaf vdevs which had previously been opened as they might deadlock * on the spa_config_lock. Instead we only obtain the leaf's physical size. * If the leaf has never been opened then open it, as usual. */ void vdev_reopen(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); /* set the reopening flag unless we're taking the vdev offline */ vd->vdev_reopening = !vd->vdev_offline; vdev_close(vd); (void) vdev_open(vd); /* * Call vdev_validate() here to make sure we have the same device. * Otherwise, a device with an invalid label could be successfully * opened in response to vdev_reopen(). */ if (vd->vdev_aux) { (void) vdev_validate_aux(vd); if (vdev_readable(vd) && vdev_writeable(vd) && vd->vdev_aux == &spa->spa_l2cache && !l2arc_vdev_present(vd)) l2arc_add_vdev(spa, vd); } else { (void) vdev_validate(vd); } /* * Reassess parent vdev's health. */ vdev_propagate_state(vd); } int vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) { int error; /* * Normally, partial opens (e.g. of a mirror) are allowed. * For a create, however, we want to fail the request if * there are any components we can't open. */ error = vdev_open(vd); if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { vdev_close(vd); return (error ? error : ENXIO); } /* * Recursively load DTLs and initialize all labels. */ if ((error = vdev_dtl_load(vd)) != 0 || (error = vdev_label_init(vd, txg, isreplacing ? VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { vdev_close(vd); return (error); } return (0); } void vdev_metaslab_set_size(vdev_t *vd) { uint64_t asize = vd->vdev_asize; uint64_t ms_shift = 0; /* * For vdevs that are bigger than 8G the metaslab size varies in * a way that the number of metaslabs increases in powers of two, * linearly in terms of vdev_asize, starting from 16 metaslabs. * So for vdev_asize of 8G we get 16 metaslabs, for 16G, we get 32, * and so on, until we hit the maximum metaslab count limit * [vdev_max_ms_count] from which point the metaslab count stays * the same. */ ms_shift = vdev_default_ms_shift; if ((asize >> ms_shift) < vdev_min_ms_count) { /* * For devices that are less than 8G we want to have * exactly 16 metaslabs. We don't want less as integer * division rounds down, so less metaslabs mean more * wasted space. We don't want more as these vdevs are * small and in the likely event that we are running * out of space, the SPA will have a hard time finding * space due to fragmentation. */ ms_shift = highbit64(asize / vdev_min_ms_count); ms_shift = MAX(ms_shift, SPA_MAXBLOCKSHIFT); } else if ((asize >> ms_shift) > vdev_max_ms_count) { ms_shift = highbit64(asize / vdev_max_ms_count); } vd->vdev_ms_shift = ms_shift; ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); } /* * Maximize performance by inflating the configured ashift for top level * vdevs to be as close to the physical ashift as possible while maintaining * administrator defined limits and ensuring it doesn't go below the * logical ashift. */ void vdev_ashift_optimize(vdev_t *vd) { if (vd == vd->vdev_top) { if (vd->vdev_ashift < vd->vdev_physical_ashift) { vd->vdev_ashift = MIN( MAX(zfs_max_auto_ashift, vd->vdev_ashift), MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift)); } else { /* * Unusual case where logical ashift > physical ashift * so we can't cap the calculated ashift based on max * ashift as that would cause failures. * We still check if we need to increase it to match * the min ashift. */ vd->vdev_ashift = MAX(zfs_min_auto_ashift, vd->vdev_ashift); } } } void vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) { ASSERT(vd == vd->vdev_top); /* indirect vdevs don't have metaslabs or dtls */ ASSERT(vdev_is_concrete(vd) || flags == 0); ASSERT(ISP2(flags)); ASSERT(spa_writeable(vd->vdev_spa)); if (flags & VDD_METASLAB) (void) txg_list_add(&vd->vdev_ms_list, arg, txg); if (flags & VDD_DTL) (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); } void vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) { for (int c = 0; c < vd->vdev_children; c++) vdev_dirty_leaves(vd->vdev_child[c], flags, txg); if (vd->vdev_ops->vdev_op_leaf) vdev_dirty(vd->vdev_top, flags, vd, txg); } /* * DTLs. * * A vdev's DTL (dirty time log) is the set of transaction groups for which * the vdev has less than perfect replication. There are four kinds of DTL: * * DTL_MISSING: txgs for which the vdev has no valid copies of the data * * DTL_PARTIAL: txgs for which data is available, but not fully replicated * * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of * txgs that was scrubbed. * * DTL_OUTAGE: txgs which cannot currently be read, whether due to * persistent errors or just some device being offline. * Unlike the other three, the DTL_OUTAGE map is not generally * maintained; it's only computed when needed, typically to * determine whether a device can be detached. * * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device * either has the data or it doesn't. * * For interior vdevs such as mirror and RAID-Z the picture is more complex. * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because * if any child is less than fully replicated, then so is its parent. * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, * comprising only those txgs which appear in 'maxfaults' or more children; * those are the txgs we don't have enough replication to read. For example, * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); * thus, its DTL_MISSING consists of the set of txgs that appear in more than * two child DTL_MISSING maps. * * It should be clear from the above that to compute the DTLs and outage maps * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. * Therefore, that is all we keep on disk. When loading the pool, or after * a configuration change, we generate all other DTLs from first principles. */ void vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) { range_tree_t *rt = vd->vdev_dtl[t]; ASSERT(t < DTL_TYPES); ASSERT(vd != vd->vdev_spa->spa_root_vdev); ASSERT(spa_writeable(vd->vdev_spa)); mutex_enter(&vd->vdev_dtl_lock); if (!range_tree_contains(rt, txg, size)) range_tree_add(rt, txg, size); mutex_exit(&vd->vdev_dtl_lock); } boolean_t vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) { range_tree_t *rt = vd->vdev_dtl[t]; boolean_t dirty = B_FALSE; ASSERT(t < DTL_TYPES); ASSERT(vd != vd->vdev_spa->spa_root_vdev); /* * While we are loading the pool, the DTLs have not been loaded yet. * Ignore the DTLs and try all devices. This avoids a recursive * mutex enter on the vdev_dtl_lock, and also makes us try hard * when loading the pool (relying on the checksum to ensure that * we get the right data -- note that we while loading, we are * only reading the MOS, which is always checksummed). */ if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) return (B_FALSE); mutex_enter(&vd->vdev_dtl_lock); if (!range_tree_is_empty(rt)) dirty = range_tree_contains(rt, txg, size); mutex_exit(&vd->vdev_dtl_lock); return (dirty); } boolean_t vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) { range_tree_t *rt = vd->vdev_dtl[t]; boolean_t empty; mutex_enter(&vd->vdev_dtl_lock); empty = range_tree_is_empty(rt); mutex_exit(&vd->vdev_dtl_lock); return (empty); } /* * Returns B_TRUE if vdev determines offset needs to be resilvered. */ boolean_t vdev_dtl_need_resilver(vdev_t *vd, uint64_t offset, size_t psize) { ASSERT(vd != vd->vdev_spa->spa_root_vdev); if (vd->vdev_ops->vdev_op_need_resilver == NULL || vd->vdev_ops->vdev_op_leaf) return (B_TRUE); return (vd->vdev_ops->vdev_op_need_resilver(vd, offset, psize)); } /* * Returns the lowest txg in the DTL range. */ static uint64_t vdev_dtl_min(vdev_t *vd) { range_seg_t *rs; ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); ASSERT0(vd->vdev_children); rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); return (rs->rs_start - 1); } /* * Returns the highest txg in the DTL. */ static uint64_t vdev_dtl_max(vdev_t *vd) { range_seg_t *rs; ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); ASSERT0(vd->vdev_children); rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); return (rs->rs_end); } /* * Determine if a resilvering vdev should remove any DTL entries from * its range. If the vdev was resilvering for the entire duration of the * scan then it should excise that range from its DTLs. Otherwise, this * vdev is considered partially resilvered and should leave its DTL * entries intact. The comment in vdev_dtl_reassess() describes how we * excise the DTLs. */ static boolean_t vdev_dtl_should_excise(vdev_t *vd) { spa_t *spa = vd->vdev_spa; dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; ASSERT0(scn->scn_phys.scn_errors); ASSERT0(vd->vdev_children); if (vd->vdev_state < VDEV_STATE_DEGRADED) return (B_FALSE); if (vd->vdev_resilver_txg == 0 || range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) return (B_TRUE); /* * When a resilver is initiated the scan will assign the scn_max_txg * value to the highest txg value that exists in all DTLs. If this * device's max DTL is not part of this scan (i.e. it is not in * the range (scn_min_txg, scn_max_txg] then it is not eligible * for excision. */ if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); return (B_TRUE); } return (B_FALSE); } /* * Reassess DTLs after a config change or scrub completion. */ void vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) { spa_t *spa = vd->vdev_spa; avl_tree_t reftree; int minref; ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); for (int c = 0; c < vd->vdev_children; c++) vdev_dtl_reassess(vd->vdev_child[c], txg, scrub_txg, scrub_done); if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) return; if (vd->vdev_ops->vdev_op_leaf) { dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; mutex_enter(&vd->vdev_dtl_lock); /* * If we've completed a scan cleanly then determine * if this vdev should remove any DTLs. We only want to * excise regions on vdevs that were available during * the entire duration of this scan. */ if (scrub_txg != 0 && (spa->spa_scrub_started || (scn != NULL && scn->scn_phys.scn_errors == 0)) && vdev_dtl_should_excise(vd)) { /* * We completed a scrub up to scrub_txg. If we * did it without rebooting, then the scrub dtl * will be valid, so excise the old region and * fold in the scrub dtl. Otherwise, leave the * dtl as-is if there was an error. * * There's little trick here: to excise the beginning * of the DTL_MISSING map, we put it into a reference * tree and then add a segment with refcnt -1 that * covers the range [0, scrub_txg). This means * that each txg in that range has refcnt -1 or 0. * We then add DTL_SCRUB with a refcnt of 2, so that * entries in the range [0, scrub_txg) will have a * positive refcnt -- either 1 or 2. We then convert * the reference tree into the new DTL_MISSING map. */ space_reftree_create(&reftree); space_reftree_add_map(&reftree, vd->vdev_dtl[DTL_MISSING], 1); space_reftree_add_seg(&reftree, 0, scrub_txg, -1); space_reftree_add_map(&reftree, vd->vdev_dtl[DTL_SCRUB], 2); space_reftree_generate_map(&reftree, vd->vdev_dtl[DTL_MISSING], 1); space_reftree_destroy(&reftree); } range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); range_tree_walk(vd->vdev_dtl[DTL_MISSING], range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); if (scrub_done) range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); if (!vdev_readable(vd)) range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); else range_tree_walk(vd->vdev_dtl[DTL_MISSING], range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); /* * If the vdev was resilvering and no longer has any * DTLs then reset its resilvering flag and dirty * the top level so that we persist the change. */ if (vd->vdev_resilver_txg != 0 && range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) { vd->vdev_resilver_txg = 0; vdev_config_dirty(vd->vdev_top); } mutex_exit(&vd->vdev_dtl_lock); if (txg != 0) vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); return; } mutex_enter(&vd->vdev_dtl_lock); for (int t = 0; t < DTL_TYPES; t++) { /* account for child's outage in parent's missing map */ int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; if (t == DTL_SCRUB) continue; /* leaf vdevs only */ if (t == DTL_PARTIAL) minref = 1; /* i.e. non-zero */ else if (vd->vdev_nparity != 0) minref = vd->vdev_nparity + 1; /* RAID-Z */ else minref = vd->vdev_children; /* any kind of mirror */ space_reftree_create(&reftree); for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; mutex_enter(&cvd->vdev_dtl_lock); space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); mutex_exit(&cvd->vdev_dtl_lock); } space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); space_reftree_destroy(&reftree); } mutex_exit(&vd->vdev_dtl_lock); } int vdev_dtl_load(vdev_t *vd) { spa_t *spa = vd->vdev_spa; objset_t *mos = spa->spa_meta_objset; int error = 0; if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { ASSERT(vdev_is_concrete(vd)); error = space_map_open(&vd->vdev_dtl_sm, mos, vd->vdev_dtl_object, 0, -1ULL, 0); if (error) return (error); ASSERT(vd->vdev_dtl_sm != NULL); mutex_enter(&vd->vdev_dtl_lock); /* * Now that we've opened the space_map we need to update * the in-core DTL. */ space_map_update(vd->vdev_dtl_sm); error = space_map_load(vd->vdev_dtl_sm, vd->vdev_dtl[DTL_MISSING], SM_ALLOC); mutex_exit(&vd->vdev_dtl_lock); return (error); } for (int c = 0; c < vd->vdev_children; c++) { error = vdev_dtl_load(vd->vdev_child[c]); if (error != 0) break; } return (error); } void vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) { spa_t *spa = vd->vdev_spa; VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, zapobj, tx)); } uint64_t vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) { spa_t *spa = vd->vdev_spa; uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); ASSERT(zap != 0); VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, zap, tx)); return (zap); } void vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) { if (vd->vdev_ops != &vdev_hole_ops && vd->vdev_ops != &vdev_missing_ops && vd->vdev_ops != &vdev_root_ops && !vd->vdev_top->vdev_removing) { if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); } if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { vd->vdev_top_zap = vdev_create_link_zap(vd, tx); } } for (uint64_t i = 0; i < vd->vdev_children; i++) { vdev_construct_zaps(vd->vdev_child[i], tx); } } void vdev_dtl_sync(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; objset_t *mos = spa->spa_meta_objset; range_tree_t *rtsync; dmu_tx_t *tx; uint64_t object = space_map_object(vd->vdev_dtl_sm); ASSERT(vdev_is_concrete(vd)); ASSERT(vd->vdev_ops->vdev_op_leaf); tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); if (vd->vdev_detached || vd->vdev_top->vdev_removing) { mutex_enter(&vd->vdev_dtl_lock); space_map_free(vd->vdev_dtl_sm, tx); space_map_close(vd->vdev_dtl_sm); vd->vdev_dtl_sm = NULL; mutex_exit(&vd->vdev_dtl_lock); /* * We only destroy the leaf ZAP for detached leaves or for * removed log devices. Removed data devices handle leaf ZAP * cleanup later, once cancellation is no longer possible. */ if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || vd->vdev_top->vdev_islog)) { vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); vd->vdev_leaf_zap = 0; } dmu_tx_commit(tx); return; } if (vd->vdev_dtl_sm == NULL) { uint64_t new_object; new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx); VERIFY3U(new_object, !=, 0); VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 0, -1ULL, 0)); ASSERT(vd->vdev_dtl_sm != NULL); } rtsync = range_tree_create(NULL, NULL); mutex_enter(&vd->vdev_dtl_lock); range_tree_walk(rt, range_tree_add, rtsync); mutex_exit(&vd->vdev_dtl_lock); space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx); space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, SM_NO_VDEVID, tx); range_tree_vacate(rtsync, NULL, NULL); range_tree_destroy(rtsync); /* * If the object for the space map has changed then dirty * the top level so that we update the config. */ if (object != space_map_object(vd->vdev_dtl_sm)) { vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " "new object %llu", (u_longlong_t)txg, spa_name(spa), (u_longlong_t)object, (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); vdev_config_dirty(vd->vdev_top); } dmu_tx_commit(tx); mutex_enter(&vd->vdev_dtl_lock); space_map_update(vd->vdev_dtl_sm); mutex_exit(&vd->vdev_dtl_lock); } /* * Determine whether the specified vdev can be offlined/detached/removed * without losing data. */ boolean_t vdev_dtl_required(vdev_t *vd) { spa_t *spa = vd->vdev_spa; vdev_t *tvd = vd->vdev_top; uint8_t cant_read = vd->vdev_cant_read; boolean_t required; ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); if (vd == spa->spa_root_vdev || vd == tvd) return (B_TRUE); /* * Temporarily mark the device as unreadable, and then determine * whether this results in any DTL outages in the top-level vdev. * If not, we can safely offline/detach/remove the device. */ vd->vdev_cant_read = B_TRUE; vdev_dtl_reassess(tvd, 0, 0, B_FALSE); required = !vdev_dtl_empty(tvd, DTL_OUTAGE); vd->vdev_cant_read = cant_read; vdev_dtl_reassess(tvd, 0, 0, B_FALSE); if (!required && zio_injection_enabled) required = !!zio_handle_device_injection(vd, NULL, ECHILD); return (required); } /* * Determine if resilver is needed, and if so the txg range. */ boolean_t vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) { boolean_t needed = B_FALSE; uint64_t thismin = UINT64_MAX; uint64_t thismax = 0; if (vd->vdev_children == 0) { mutex_enter(&vd->vdev_dtl_lock); if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && vdev_writeable(vd)) { thismin = vdev_dtl_min(vd); thismax = vdev_dtl_max(vd); needed = B_TRUE; } mutex_exit(&vd->vdev_dtl_lock); } else { for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; uint64_t cmin, cmax; if (vdev_resilver_needed(cvd, &cmin, &cmax)) { thismin = MIN(thismin, cmin); thismax = MAX(thismax, cmax); needed = B_TRUE; } } } if (needed && minp) { *minp = thismin; *maxp = thismax; } return (needed); } /* * Gets the checkpoint space map object from the vdev's ZAP. * Returns the spacemap object, or 0 if it wasn't in the ZAP * or the ZAP doesn't exist yet. */ int vdev_checkpoint_sm_object(vdev_t *vd) { ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); if (vd->vdev_top_zap == 0) { return (0); } uint64_t sm_obj = 0; int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj); ASSERT(err == 0 || err == ENOENT); return (sm_obj); } int vdev_load(vdev_t *vd) { int error = 0; /* * Recursively load all children. */ for (int c = 0; c < vd->vdev_children; c++) { error = vdev_load(vd->vdev_child[c]); if (error != 0) { return (error); } } vdev_set_deflate_ratio(vd); /* * If this is a top-level vdev, initialize its metaslabs. */ if (vd == vd->vdev_top && vdev_is_concrete(vd)) { if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " "asize=%llu", (u_longlong_t)vd->vdev_ashift, (u_longlong_t)vd->vdev_asize); return (SET_ERROR(ENXIO)); } else if ((error = vdev_metaslab_init(vd, 0)) != 0) { vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " "[error=%d]", error); vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); return (error); } uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd); if (checkpoint_sm_obj != 0) { objset_t *mos = spa_meta_objset(vd->vdev_spa); ASSERT(vd->vdev_asize != 0); ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); if ((error = space_map_open(&vd->vdev_checkpoint_sm, mos, checkpoint_sm_obj, 0, vd->vdev_asize, vd->vdev_ashift))) { vdev_dbgmsg(vd, "vdev_load: space_map_open " "failed for checkpoint spacemap (obj %llu) " "[error=%d]", (u_longlong_t)checkpoint_sm_obj, error); return (error); } ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); space_map_update(vd->vdev_checkpoint_sm); /* * Since the checkpoint_sm contains free entries * exclusively we can use sm_alloc to indicate the * culmulative checkpointed space that has been freed. */ vd->vdev_stat.vs_checkpoint_space = -vd->vdev_checkpoint_sm->sm_alloc; vd->vdev_spa->spa_checkpoint_info.sci_dspace += vd->vdev_stat.vs_checkpoint_space; } } /* * If this is a leaf vdev, load its DTL. */ if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " "[error=%d]", error); return (error); } uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); if (obsolete_sm_object != 0) { objset_t *mos = vd->vdev_spa->spa_meta_objset; ASSERT(vd->vdev_asize != 0); ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, obsolete_sm_object, 0, vd->vdev_asize, 0))) { vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " "obsolete spacemap (obj %llu) [error=%d]", (u_longlong_t)obsolete_sm_object, error); return (error); } space_map_update(vd->vdev_obsolete_sm); } return (0); } /* * The special vdev case is used for hot spares and l2cache devices. Its * sole purpose it to set the vdev state for the associated vdev. To do this, * we make sure that we can open the underlying device, then try to read the * label, and make sure that the label is sane and that it hasn't been * repurposed to another pool. */ int vdev_validate_aux(vdev_t *vd) { nvlist_t *label; uint64_t guid, version; uint64_t state; if (!vdev_readable(vd)) return (0); if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); return (-1); } if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || !SPA_VERSION_IS_SUPPORTED(version) || nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || guid != vd->vdev_guid || nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); nvlist_free(label); return (-1); } /* * We don't actually check the pool state here. If it's in fact in * use by another pool, we update this fact on the fly when requested. */ nvlist_free(label); return (0); } /* * Free the objects used to store this vdev's spacemaps, and the array * that points to them. */ void vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) { if (vd->vdev_ms_array == 0) return; objset_t *mos = vd->vdev_spa->spa_meta_objset; uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; size_t array_bytes = array_count * sizeof (uint64_t); uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, array_bytes, smobj_array, 0)); for (uint64_t i = 0; i < array_count; i++) { uint64_t smobj = smobj_array[i]; if (smobj == 0) continue; space_map_free_obj(mos, smobj, tx); } kmem_free(smobj_array, array_bytes); VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); vd->vdev_ms_array = 0; } static void vdev_remove_empty(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; dmu_tx_t *tx; ASSERT(vd == vd->vdev_top); ASSERT3U(txg, ==, spa_syncing_txg(spa)); if (vd->vdev_ms != NULL) { metaslab_group_t *mg = vd->vdev_mg; metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); for (int m = 0; m < vd->vdev_ms_count; m++) { metaslab_t *msp = vd->vdev_ms[m]; if (msp == NULL || msp->ms_sm == NULL) continue; mutex_enter(&msp->ms_lock); /* * If the metaslab was not loaded when the vdev * was removed then the histogram accounting may * not be accurate. Update the histogram information * here so that we ensure that the metaslab group * and metaslab class are up-to-date. */ metaslab_group_histogram_remove(mg, msp); VERIFY0(space_map_allocated(msp->ms_sm)); space_map_close(msp->ms_sm); msp->ms_sm = NULL; mutex_exit(&msp->ms_lock); } if (vd->vdev_checkpoint_sm != NULL) { ASSERT(spa_has_checkpoint(spa)); space_map_close(vd->vdev_checkpoint_sm); vd->vdev_checkpoint_sm = NULL; } metaslab_group_histogram_verify(mg); metaslab_class_histogram_verify(mg->mg_class); for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) ASSERT0(mg->mg_histogram[i]); } tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); vdev_destroy_spacemaps(vd, tx); if (vd->vdev_islog && vd->vdev_top_zap != 0) { vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); vd->vdev_top_zap = 0; } dmu_tx_commit(tx); } void vdev_sync_done(vdev_t *vd, uint64_t txg) { metaslab_t *msp; boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); ASSERT(vdev_is_concrete(vd)); while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) metaslab_sync_done(msp, txg); if (reassess) metaslab_sync_reassess(vd->vdev_mg); } void vdev_sync(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; vdev_t *lvd; metaslab_t *msp; dmu_tx_t *tx; if (range_tree_space(vd->vdev_obsolete_segments) > 0) { dmu_tx_t *tx; ASSERT(vd->vdev_removing || vd->vdev_ops == &vdev_indirect_ops); tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); vdev_indirect_sync_obsolete(vd, tx); dmu_tx_commit(tx); /* * If the vdev is indirect, it can't have dirty * metaslabs or DTLs. */ if (vd->vdev_ops == &vdev_indirect_ops) { ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); return; } } ASSERT(vdev_is_concrete(vd)); if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && !vd->vdev_removing) { ASSERT(vd == vd->vdev_top); ASSERT0(vd->vdev_indirect_config.vic_mapping_object); tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); ASSERT(vd->vdev_ms_array != 0); vdev_config_dirty(vd); dmu_tx_commit(tx); } while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { metaslab_sync(msp, txg); (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); } while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) vdev_dtl_sync(lvd, txg); /* * Remove the metadata associated with this vdev once it's empty. * Note that this is typically used for log/cache device removal; * we don't empty toplevel vdevs when removing them. But if * a toplevel happens to be emptied, this is not harmful. */ if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) { vdev_remove_empty(vd, txg); } (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); } uint64_t vdev_psize_to_asize(vdev_t *vd, uint64_t psize) { return (vd->vdev_ops->vdev_op_asize(vd, psize)); } /* * Mark the given vdev faulted. A faulted vdev behaves as if the device could * not be opened, and no I/O is attempted. */ int vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) { vdev_t *vd, *tvd; spa_vdev_state_enter(spa, SCL_NONE); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, ENODEV)); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); tvd = vd->vdev_top; /* * We don't directly use the aux state here, but if we do a * vdev_reopen(), we need this value to be present to remember why we * were faulted. */ vd->vdev_label_aux = aux; /* * Faulted state takes precedence over degraded. */ vd->vdev_delayed_close = B_FALSE; vd->vdev_faulted = 1ULL; vd->vdev_degraded = 0ULL; vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); /* * If this device has the only valid copy of the data, then * back off and simply mark the vdev as degraded instead. */ if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { vd->vdev_degraded = 1ULL; vd->vdev_faulted = 0ULL; /* * If we reopen the device and it's not dead, only then do we * mark it degraded. */ vdev_reopen(tvd); if (vdev_readable(vd)) vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); } return (spa_vdev_state_exit(spa, vd, 0)); } /* * Mark the given vdev degraded. A degraded vdev is purely an indication to the * user that something is wrong. The vdev continues to operate as normal as far * as I/O is concerned. */ int vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) { vdev_t *vd; spa_vdev_state_enter(spa, SCL_NONE); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, ENODEV)); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); /* * If the vdev is already faulted, then don't do anything. */ if (vd->vdev_faulted || vd->vdev_degraded) return (spa_vdev_state_exit(spa, NULL, 0)); vd->vdev_degraded = 1ULL; if (!vdev_is_dead(vd)) vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); return (spa_vdev_state_exit(spa, vd, 0)); } /* * Online the given vdev. * * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached * spare device should be detached when the device finishes resilvering. * Second, the online should be treated like a 'test' online case, so no FMA * events are generated if the device fails to open. */ int vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) { vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; boolean_t wasoffline; vdev_state_t oldstate; spa_vdev_state_enter(spa, SCL_NONE); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, ENODEV)); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); oldstate = vd->vdev_state; tvd = vd->vdev_top; vd->vdev_offline = B_FALSE; vd->vdev_tmpoffline = B_FALSE; vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); /* XXX - L2ARC 1.0 does not support expansion */ if (!vd->vdev_aux) { for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); } vdev_reopen(tvd); vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; if (!vd->vdev_aux) { for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) pvd->vdev_expanding = B_FALSE; } if (newstate) *newstate = vd->vdev_state; if ((flags & ZFS_ONLINE_UNSPARE) && !vdev_is_dead(vd) && vd->vdev_parent && vd->vdev_parent->vdev_ops == &vdev_spare_ops && vd->vdev_parent->vdev_child[0] == vd) vd->vdev_unspare = B_TRUE; if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { /* XXX - L2ARC 1.0 does not support expansion */ if (vd->vdev_aux) return (spa_vdev_state_exit(spa, vd, ENOTSUP)); spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); } if (wasoffline || (oldstate < VDEV_STATE_DEGRADED && vd->vdev_state >= VDEV_STATE_DEGRADED)) spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); return (spa_vdev_state_exit(spa, vd, 0)); } static int vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) { vdev_t *vd, *tvd; int error = 0; uint64_t generation; metaslab_group_t *mg; top: spa_vdev_state_enter(spa, SCL_ALLOC); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, ENODEV)); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); tvd = vd->vdev_top; mg = tvd->vdev_mg; generation = spa->spa_config_generation + 1; /* * If the device isn't already offline, try to offline it. */ if (!vd->vdev_offline) { /* * If this device has the only valid copy of some data, * don't allow it to be offlined. Log devices are always * expendable. */ if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) return (spa_vdev_state_exit(spa, NULL, EBUSY)); /* * If the top-level is a slog and it has had allocations * then proceed. We check that the vdev's metaslab group * is not NULL since it's possible that we may have just * added this vdev but not yet initialized its metaslabs. */ if (tvd->vdev_islog && mg != NULL) { /* * Prevent any future allocations. */ metaslab_group_passivate(mg); (void) spa_vdev_state_exit(spa, vd, 0); error = spa_reset_logs(spa); /* * If the log device was successfully reset but has * checkpointed data, do not offline it. */ if (error == 0 && tvd->vdev_checkpoint_sm != NULL) { ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc, !=, 0); error = ZFS_ERR_CHECKPOINT_EXISTS; } spa_vdev_state_enter(spa, SCL_ALLOC); /* * Check to see if the config has changed. */ if (error || generation != spa->spa_config_generation) { metaslab_group_activate(mg); if (error) return (spa_vdev_state_exit(spa, vd, error)); (void) spa_vdev_state_exit(spa, vd, 0); goto top; } ASSERT0(tvd->vdev_stat.vs_alloc); } /* * Offline this device and reopen its top-level vdev. * If the top-level vdev is a log device then just offline * it. Otherwise, if this action results in the top-level * vdev becoming unusable, undo it and fail the request. */ vd->vdev_offline = B_TRUE; vdev_reopen(tvd); if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_is_dead(tvd)) { vd->vdev_offline = B_FALSE; vdev_reopen(tvd); return (spa_vdev_state_exit(spa, NULL, EBUSY)); } /* * Add the device back into the metaslab rotor so that * once we online the device it's open for business. */ if (tvd->vdev_islog && mg != NULL) metaslab_group_activate(mg); } vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); return (spa_vdev_state_exit(spa, vd, 0)); } int vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) { int error; mutex_enter(&spa->spa_vdev_top_lock); error = vdev_offline_locked(spa, guid, flags); mutex_exit(&spa->spa_vdev_top_lock); return (error); } /* * Clear the error counts associated with this vdev. Unlike vdev_online() and * vdev_offline(), we assume the spa config is locked. We also clear all * children. If 'vd' is NULL, then the user wants to clear all vdevs. */ void vdev_clear(spa_t *spa, vdev_t *vd) { vdev_t *rvd = spa->spa_root_vdev; ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); if (vd == NULL) vd = rvd; vd->vdev_stat.vs_read_errors = 0; vd->vdev_stat.vs_write_errors = 0; vd->vdev_stat.vs_checksum_errors = 0; for (int c = 0; c < vd->vdev_children; c++) vdev_clear(spa, vd->vdev_child[c]); if (vd == rvd) { for (int c = 0; c < spa->spa_l2cache.sav_count; c++) vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]); for (int c = 0; c < spa->spa_spares.sav_count; c++) vdev_clear(spa, spa->spa_spares.sav_vdevs[c]); } /* * It makes no sense to "clear" an indirect vdev. */ if (!vdev_is_concrete(vd)) return; /* * If we're in the FAULTED state or have experienced failed I/O, then * clear the persistent state and attempt to reopen the device. We * also mark the vdev config dirty, so that the new faulted state is * written out to disk. */ if (vd->vdev_faulted || vd->vdev_degraded || !vdev_readable(vd) || !vdev_writeable(vd)) { /* * When reopening in reponse to a clear event, it may be due to * a fmadm repair request. In this case, if the device is * still broken, we want to still post the ereport again. */ vd->vdev_forcefault = B_TRUE; vd->vdev_faulted = vd->vdev_degraded = 0ULL; vd->vdev_cant_read = B_FALSE; vd->vdev_cant_write = B_FALSE; vdev_reopen(vd == rvd ? rvd : vd->vdev_top); vd->vdev_forcefault = B_FALSE; if (vd != rvd && vdev_writeable(vd->vdev_top)) vdev_state_dirty(vd->vdev_top); if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) spa_async_request(spa, SPA_ASYNC_RESILVER); spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); } /* * When clearing a FMA-diagnosed fault, we always want to * unspare the device, as we assume that the original spare was * done in response to the FMA fault. */ if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && vd->vdev_parent->vdev_ops == &vdev_spare_ops && vd->vdev_parent->vdev_child[0] == vd) vd->vdev_unspare = B_TRUE; } boolean_t vdev_is_dead(vdev_t *vd) { /* * Holes and missing devices are always considered "dead". * This simplifies the code since we don't have to check for * these types of devices in the various code paths. * Instead we rely on the fact that we skip over dead devices * before issuing I/O to them. */ return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ops == &vdev_hole_ops || vd->vdev_ops == &vdev_missing_ops); } boolean_t vdev_readable(vdev_t *vd) { return (!vdev_is_dead(vd) && !vd->vdev_cant_read); } boolean_t vdev_writeable(vdev_t *vd) { return (!vdev_is_dead(vd) && !vd->vdev_cant_write && vdev_is_concrete(vd)); } boolean_t vdev_allocatable(vdev_t *vd) { uint64_t state = vd->vdev_state; /* * We currently allow allocations from vdevs which may be in the * process of reopening (i.e. VDEV_STATE_CLOSED). If the device * fails to reopen then we'll catch it later when we're holding * the proper locks. Note that we have to get the vdev state * in a local variable because although it changes atomically, * we're asking two separate questions about it. */ return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && !vd->vdev_cant_write && vdev_is_concrete(vd) && vd->vdev_mg->mg_initialized); } boolean_t vdev_accessible(vdev_t *vd, zio_t *zio) { ASSERT(zio->io_vd == vd); if (vdev_is_dead(vd) || vd->vdev_remove_wanted) return (B_FALSE); if (zio->io_type == ZIO_TYPE_READ) return (!vd->vdev_cant_read); if (zio->io_type == ZIO_TYPE_WRITE) return (!vd->vdev_cant_write); return (B_TRUE); } boolean_t vdev_is_spacemap_addressable(vdev_t *vd) { /* * Assuming 47 bits of the space map entry dedicated for the entry's * offset (see description in space_map.h), we calculate the maximum * address that can be described by a space map entry for the given * device. */ uint64_t shift = vd->vdev_ashift + 47; if (shift >= 63) /* detect potential overflow */ return (B_TRUE); return (vd->vdev_asize < (1ULL << shift)); } /* * Get statistics for the given vdev. */ void vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) { spa_t *spa = vd->vdev_spa; vdev_t *rvd = spa->spa_root_vdev; vdev_t *tvd = vd->vdev_top; ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); mutex_enter(&vd->vdev_stat_lock); bcopy(&vd->vdev_stat, vs, sizeof (*vs)); vs->vs_timestamp = gethrtime() - vs->vs_timestamp; vs->vs_state = vd->vdev_state; vs->vs_rsize = vdev_get_min_asize(vd); if (vd->vdev_ops->vdev_op_leaf) vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; /* * Report expandable space on top-level, non-auxillary devices only. * The expandable space is reported in terms of metaslab sized units * since that determines how much space the pool can expand. */ if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) { vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize - spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift); } vs->vs_configured_ashift = vd->vdev_top != NULL ? vd->vdev_top->vdev_ashift : vd->vdev_ashift; vs->vs_logical_ashift = vd->vdev_logical_ashift; vs->vs_physical_ashift = vd->vdev_physical_ashift; if (vd->vdev_aux == NULL && vd == vd->vdev_top && vdev_is_concrete(vd)) { vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; } /* * If we're getting stats on the root vdev, aggregate the I/O counts * over all top-level vdevs (i.e. the direct children of the root). */ if (vd == rvd) { for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *cvd = rvd->vdev_child[c]; vdev_stat_t *cvs = &cvd->vdev_stat; for (int t = 0; t < ZIO_TYPES; t++) { vs->vs_ops[t] += cvs->vs_ops[t]; vs->vs_bytes[t] += cvs->vs_bytes[t]; } cvs->vs_scan_removing = cvd->vdev_removing; } } mutex_exit(&vd->vdev_stat_lock); } void vdev_clear_stats(vdev_t *vd) { mutex_enter(&vd->vdev_stat_lock); vd->vdev_stat.vs_space = 0; vd->vdev_stat.vs_dspace = 0; vd->vdev_stat.vs_alloc = 0; mutex_exit(&vd->vdev_stat_lock); } void vdev_scan_stat_init(vdev_t *vd) { vdev_stat_t *vs = &vd->vdev_stat; for (int c = 0; c < vd->vdev_children; c++) vdev_scan_stat_init(vd->vdev_child[c]); mutex_enter(&vd->vdev_stat_lock); vs->vs_scan_processed = 0; mutex_exit(&vd->vdev_stat_lock); } void vdev_stat_update(zio_t *zio, uint64_t psize) { spa_t *spa = zio->io_spa; vdev_t *rvd = spa->spa_root_vdev; vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; vdev_t *pvd; uint64_t txg = zio->io_txg; vdev_stat_t *vs = &vd->vdev_stat; zio_type_t type = zio->io_type; int flags = zio->io_flags; /* * If this i/o is a gang leader, it didn't do any actual work. */ if (zio->io_gang_tree) return; if (zio->io_error == 0) { /* * If this is a root i/o, don't count it -- we've already * counted the top-level vdevs, and vdev_get_stats() will * aggregate them when asked. This reduces contention on * the root vdev_stat_lock and implicitly handles blocks * that compress away to holes, for which there is no i/o. * (Holes never create vdev children, so all the counters * remain zero, which is what we want.) * * Note: this only applies to successful i/o (io_error == 0) * because unlike i/o counts, errors are not additive. * When reading a ditto block, for example, failure of * one top-level vdev does not imply a root-level error. */ if (vd == rvd) return; ASSERT(vd == zio->io_vd); if (flags & ZIO_FLAG_IO_BYPASS) return; mutex_enter(&vd->vdev_stat_lock); if (flags & ZIO_FLAG_IO_REPAIR) { if (flags & ZIO_FLAG_SCAN_THREAD) { dsl_scan_phys_t *scn_phys = &spa->spa_dsl_pool->dp_scan->scn_phys; uint64_t *processed = &scn_phys->scn_processed; /* XXX cleanup? */ if (vd->vdev_ops->vdev_op_leaf) atomic_add_64(processed, psize); vs->vs_scan_processed += psize; } if (flags & ZIO_FLAG_SELF_HEAL) vs->vs_self_healed += psize; } vs->vs_ops[type]++; vs->vs_bytes[type] += psize; mutex_exit(&vd->vdev_stat_lock); return; } if (flags & ZIO_FLAG_SPECULATIVE) return; /* * If this is an I/O error that is going to be retried, then ignore the * error. Otherwise, the user may interpret B_FAILFAST I/O errors as * hard errors, when in reality they can happen for any number of * innocuous reasons (bus resets, MPxIO link failure, etc). */ if (zio->io_error == EIO && !(zio->io_flags & ZIO_FLAG_IO_RETRY)) return; /* * Intent logs writes won't propagate their error to the root * I/O so don't mark these types of failures as pool-level * errors. */ if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) return; mutex_enter(&vd->vdev_stat_lock); if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { if (zio->io_error == ECKSUM) vs->vs_checksum_errors++; else vs->vs_read_errors++; } if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) vs->vs_write_errors++; mutex_exit(&vd->vdev_stat_lock); if (spa->spa_load_state == SPA_LOAD_NONE && type == ZIO_TYPE_WRITE && txg != 0 && (!(flags & ZIO_FLAG_IO_REPAIR) || (flags & ZIO_FLAG_SCAN_THREAD) || spa->spa_claiming)) { /* * This is either a normal write (not a repair), or it's * a repair induced by the scrub thread, or it's a repair * made by zil_claim() during spa_load() in the first txg. * In the normal case, we commit the DTL change in the same * txg as the block was born. In the scrub-induced repair * case, we know that scrubs run in first-pass syncing context, * so we commit the DTL change in spa_syncing_txg(spa). * In the zil_claim() case, we commit in spa_first_txg(spa). * * We currently do not make DTL entries for failed spontaneous * self-healing writes triggered by normal (non-scrubbing) * reads, because we have no transactional context in which to * do so -- and it's not clear that it'd be desirable anyway. */ if (vd->vdev_ops->vdev_op_leaf) { uint64_t commit_txg = txg; if (flags & ZIO_FLAG_SCAN_THREAD) { ASSERT(flags & ZIO_FLAG_IO_REPAIR); ASSERT(spa_sync_pass(spa) == 1); vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); commit_txg = spa_syncing_txg(spa); } else if (spa->spa_claiming) { ASSERT(flags & ZIO_FLAG_IO_REPAIR); commit_txg = spa_first_txg(spa); } ASSERT(commit_txg >= spa_syncing_txg(spa)); if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) return; for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); } if (vd != rvd) vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); } } /* * Update the in-core space usage stats for this vdev, its metaslab class, * and the root vdev. */ void vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, int64_t space_delta) { int64_t dspace_delta = space_delta; spa_t *spa = vd->vdev_spa; vdev_t *rvd = spa->spa_root_vdev; metaslab_group_t *mg = vd->vdev_mg; metaslab_class_t *mc = mg ? mg->mg_class : NULL; ASSERT(vd == vd->vdev_top); /* * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion * factor. We must calculate this here and not at the root vdev * because the root vdev's psize-to-asize is simply the max of its * childrens', thus not accurate enough for us. */ ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio; mutex_enter(&vd->vdev_stat_lock); vd->vdev_stat.vs_alloc += alloc_delta; vd->vdev_stat.vs_space += space_delta; vd->vdev_stat.vs_dspace += dspace_delta; mutex_exit(&vd->vdev_stat_lock); if (mc == spa_normal_class(spa)) { mutex_enter(&rvd->vdev_stat_lock); rvd->vdev_stat.vs_alloc += alloc_delta; rvd->vdev_stat.vs_space += space_delta; rvd->vdev_stat.vs_dspace += dspace_delta; mutex_exit(&rvd->vdev_stat_lock); } if (mc != NULL) { ASSERT(rvd == vd->vdev_parent); ASSERT(vd->vdev_ms_count != 0); metaslab_class_space_update(mc, alloc_delta, defer_delta, space_delta, dspace_delta); } } /* * Mark a top-level vdev's config as dirty, placing it on the dirty list * so that it will be written out next time the vdev configuration is synced. * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. */ void vdev_config_dirty(vdev_t *vd) { spa_t *spa = vd->vdev_spa; vdev_t *rvd = spa->spa_root_vdev; int c; ASSERT(spa_writeable(spa)); /* * If this is an aux vdev (as with l2cache and spare devices), then we * update the vdev config manually and set the sync flag. */ if (vd->vdev_aux != NULL) { spa_aux_vdev_t *sav = vd->vdev_aux; nvlist_t **aux; uint_t naux; for (c = 0; c < sav->sav_count; c++) { if (sav->sav_vdevs[c] == vd) break; } if (c == sav->sav_count) { /* * We're being removed. There's nothing more to do. */ ASSERT(sav->sav_sync == B_TRUE); return; } sav->sav_sync = B_TRUE; if (nvlist_lookup_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); } ASSERT(c < naux); /* * Setting the nvlist in the middle if the array is a little * sketchy, but it will work. */ nvlist_free(aux[c]); aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); return; } /* * The dirty list is protected by the SCL_CONFIG lock. The caller * must either hold SCL_CONFIG as writer, or must be the sync thread * (which holds SCL_CONFIG as reader). There's only one sync thread, * so this is sufficient to ensure mutual exclusion. */ ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || (dsl_pool_sync_context(spa_get_dsl(spa)) && spa_config_held(spa, SCL_CONFIG, RW_READER))); if (vd == rvd) { for (c = 0; c < rvd->vdev_children; c++) vdev_config_dirty(rvd->vdev_child[c]); } else { ASSERT(vd == vd->vdev_top); if (!list_link_active(&vd->vdev_config_dirty_node) && vdev_is_concrete(vd)) { list_insert_head(&spa->spa_config_dirty_list, vd); } } } void vdev_config_clean(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || (dsl_pool_sync_context(spa_get_dsl(spa)) && spa_config_held(spa, SCL_CONFIG, RW_READER))); ASSERT(list_link_active(&vd->vdev_config_dirty_node)); list_remove(&spa->spa_config_dirty_list, vd); } /* * Mark a top-level vdev's state as dirty, so that the next pass of * spa_sync() can convert this into vdev_config_dirty(). We distinguish * the state changes from larger config changes because they require * much less locking, and are often needed for administrative actions. */ void vdev_state_dirty(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_writeable(spa)); ASSERT(vd == vd->vdev_top); /* * The state list is protected by the SCL_STATE lock. The caller * must either hold SCL_STATE as writer, or must be the sync thread * (which holds SCL_STATE as reader). There's only one sync thread, * so this is sufficient to ensure mutual exclusion. */ ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || (dsl_pool_sync_context(spa_get_dsl(spa)) && spa_config_held(spa, SCL_STATE, RW_READER))); if (!list_link_active(&vd->vdev_state_dirty_node) && vdev_is_concrete(vd)) list_insert_head(&spa->spa_state_dirty_list, vd); } void vdev_state_clean(vdev_t *vd) { spa_t *spa = vd->vdev_spa; ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || (dsl_pool_sync_context(spa_get_dsl(spa)) && spa_config_held(spa, SCL_STATE, RW_READER))); ASSERT(list_link_active(&vd->vdev_state_dirty_node)); list_remove(&spa->spa_state_dirty_list, vd); } /* * Propagate vdev state up from children to parent. */ void vdev_propagate_state(vdev_t *vd) { spa_t *spa = vd->vdev_spa; vdev_t *rvd = spa->spa_root_vdev; int degraded = 0, faulted = 0; int corrupted = 0; vdev_t *child; if (vd->vdev_children > 0) { for (int c = 0; c < vd->vdev_children; c++) { child = vd->vdev_child[c]; /* * Don't factor holes or indirect vdevs into the * decision. */ if (!vdev_is_concrete(child)) continue; if (!vdev_readable(child) || (!vdev_writeable(child) && spa_writeable(spa))) { /* * Root special: if there is a top-level log * device, treat the root vdev as if it were * degraded. */ if (child->vdev_islog && vd == rvd) degraded++; else faulted++; } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { degraded++; } if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) corrupted++; } vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); /* * Root special: if there is a top-level vdev that cannot be * opened due to corrupted metadata, then propagate the root * vdev's aux state as 'corrupt' rather than 'insufficient * replicas'. */ if (corrupted && vd == rvd && rvd->vdev_state == VDEV_STATE_CANT_OPEN) vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, VDEV_AUX_CORRUPT_DATA); } if (vd->vdev_parent) vdev_propagate_state(vd->vdev_parent); } /* * Set a vdev's state. If this is during an open, we don't update the parent * state, because we're in the process of opening children depth-first. * Otherwise, we propagate the change to the parent. * * If this routine places a device in a faulted state, an appropriate ereport is * generated. */ void vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) { uint64_t save_state; spa_t *spa = vd->vdev_spa; if (state == vd->vdev_state) { vd->vdev_stat.vs_aux = aux; return; } save_state = vd->vdev_state; vd->vdev_state = state; vd->vdev_stat.vs_aux = aux; /* * If we are setting the vdev state to anything but an open state, then * always close the underlying device unless the device has requested * a delayed close (i.e. we're about to remove or fault the device). * Otherwise, we keep accessible but invalid devices open forever. * We don't call vdev_close() itself, because that implies some extra * checks (offline, etc) that we don't want here. This is limited to * leaf devices, because otherwise closing the device will affect other * children. */ if (!vd->vdev_delayed_close && vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf) vd->vdev_ops->vdev_op_close(vd); if (vd->vdev_removed && state == VDEV_STATE_CANT_OPEN && (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { /* * If the previous state is set to VDEV_STATE_REMOVED, then this * device was previously marked removed and someone attempted to * reopen it. If this failed due to a nonexistent device, then * keep the device in the REMOVED state. We also let this be if * it is one of our special test online cases, which is only * attempting to online the device and shouldn't generate an FMA * fault. */ vd->vdev_state = VDEV_STATE_REMOVED; vd->vdev_stat.vs_aux = VDEV_AUX_NONE; } else if (state == VDEV_STATE_REMOVED) { vd->vdev_removed = B_TRUE; } else if (state == VDEV_STATE_CANT_OPEN) { /* * If we fail to open a vdev during an import or recovery, we * mark it as "not available", which signifies that it was * never there to begin with. Failure to open such a device * is not considered an error. */ if ((spa_load_state(spa) == SPA_LOAD_IMPORT || spa_load_state(spa) == SPA_LOAD_RECOVER) && vd->vdev_ops->vdev_op_leaf) vd->vdev_not_present = 1; /* * Post the appropriate ereport. If the 'prevstate' field is * set to something other than VDEV_STATE_UNKNOWN, it indicates * that this is part of a vdev_reopen(). In this case, we don't * want to post the ereport if the device was already in the * CANT_OPEN state beforehand. * * If the 'checkremove' flag is set, then this is an attempt to * online the device in response to an insertion event. If we * hit this case, then we have detected an insertion event for a * faulted or offline device that wasn't in the removed state. * In this scenario, we don't post an ereport because we are * about to replace the device, or attempt an online with * vdev_forcefault, which will generate the fault for us. */ if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && !vd->vdev_not_present && !vd->vdev_checkremove && vd != spa->spa_root_vdev) { const char *class; switch (aux) { case VDEV_AUX_OPEN_FAILED: class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; break; case VDEV_AUX_CORRUPT_DATA: class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; break; case VDEV_AUX_NO_REPLICAS: class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; break; case VDEV_AUX_BAD_GUID_SUM: class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; break; case VDEV_AUX_TOO_SMALL: class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; break; case VDEV_AUX_BAD_LABEL: class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; break; default: class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; } zfs_ereport_post(class, spa, vd, NULL, save_state, 0); } /* Erase any notion of persistent removed state */ vd->vdev_removed = B_FALSE; } else { vd->vdev_removed = B_FALSE; } /* * Notify the fmd of the state change. Be verbose and post * notifications even for stuff that's not important; the fmd agent can * sort it out. Don't emit state change events for non-leaf vdevs since * they can't change state on their own. The FMD can check their state * if it wants to when it sees that a leaf vdev had a state change. */ if (vd->vdev_ops->vdev_op_leaf) zfs_post_state_change(spa, vd); if (!isopen && vd->vdev_parent) vdev_propagate_state(vd->vdev_parent); } boolean_t vdev_children_are_offline(vdev_t *vd) { ASSERT(!vd->vdev_ops->vdev_op_leaf); for (uint64_t i = 0; i < vd->vdev_children; i++) { if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) return (B_FALSE); } return (B_TRUE); } /* * Check the vdev configuration to ensure that it's capable of supporting * a root pool. We do not support partial configuration. * In addition, only a single top-level vdev is allowed. * * FreeBSD does not have above limitations. */ boolean_t vdev_is_bootable(vdev_t *vd) { #ifdef illumos if (!vd->vdev_ops->vdev_op_leaf) { char *vdev_type = vd->vdev_ops->vdev_op_type; if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && vd->vdev_children > 1) { return (B_FALSE); } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 || strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) { return (B_FALSE); } } for (int c = 0; c < vd->vdev_children; c++) { if (!vdev_is_bootable(vd->vdev_child[c])) return (B_FALSE); } #endif /* illumos */ return (B_TRUE); } boolean_t vdev_is_concrete(vdev_t *vd) { vdev_ops_t *ops = vd->vdev_ops; if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || ops == &vdev_missing_ops || ops == &vdev_root_ops) { return (B_FALSE); } else { return (B_TRUE); } } /* * Determine if a log device has valid content. If the vdev was * removed or faulted in the MOS config then we know that * the content on the log device has already been written to the pool. */ boolean_t vdev_log_state_valid(vdev_t *vd) { if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && !vd->vdev_removed) return (B_TRUE); for (int c = 0; c < vd->vdev_children; c++) if (vdev_log_state_valid(vd->vdev_child[c])) return (B_TRUE); return (B_FALSE); } /* * Expand a vdev if possible. */ void vdev_expand(vdev_t *vd, uint64_t txg) { ASSERT(vd->vdev_top == vd); ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); vdev_set_deflate_ratio(vd); if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && vdev_is_concrete(vd)) { VERIFY(vdev_metaslab_init(vd, txg) == 0); vdev_config_dirty(vd); } } /* * Split a vdev. */ void vdev_split(vdev_t *vd) { vdev_t *cvd, *pvd = vd->vdev_parent; vdev_remove_child(pvd, vd); vdev_compact_children(pvd); cvd = pvd->vdev_child[0]; if (pvd->vdev_children == 1) { vdev_remove_parent(cvd); cvd->vdev_splitting = B_TRUE; } vdev_propagate_state(cvd); } void vdev_deadman(vdev_t *vd) { for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; vdev_deadman(cvd); } if (vd->vdev_ops->vdev_op_leaf) { vdev_queue_t *vq = &vd->vdev_queue; mutex_enter(&vq->vq_lock); if (avl_numnodes(&vq->vq_active_tree) > 0) { spa_t *spa = vd->vdev_spa; zio_t *fio; uint64_t delta; /* * Look at the head of all the pending queues, * if any I/O has been outstanding for longer than * the spa_deadman_synctime we panic the system. */ fio = avl_first(&vq->vq_active_tree); delta = gethrtime() - fio->io_timestamp; if (delta > spa_deadman_synctime(spa)) { vdev_dbgmsg(vd, "SLOW IO: zio timestamp " "%lluns, delta %lluns, last io %lluns", fio->io_timestamp, (u_longlong_t)delta, vq->vq_io_complete_ts); fm_panic("I/O to pool '%s' appears to be " "hung on vdev guid %llu at '%s'.", spa_name(spa), (long long unsigned int) vd->vdev_guid, vd->vdev_path); } } mutex_exit(&vq->vq_lock); } } Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c (revision 336949) @@ -1,1004 +1,1012 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* - * Copyright (c) 2012, 2017 by Delphix. All rights reserved. + * Copyright (c) 2012, 2018 by Delphix. All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #include #include #include #include #include #include #include #include /* * ZFS I/O Scheduler * --------------- * * ZFS issues I/O operations to leaf vdevs to satisfy and complete zios. The * I/O scheduler determines when and in what order those operations are * issued. The I/O scheduler divides operations into six I/O classes * prioritized in the following order: sync read, sync write, async read, * async write, scrub/resilver and trim. Each queue defines the minimum and * maximum number of concurrent operations that may be issued to the device. * In addition, the device has an aggregate maximum. Note that the sum of the * per-queue minimums must not exceed the aggregate maximum, and if the * aggregate maximum is equal to or greater than the sum of the per-queue * maximums, the per-queue minimum has no effect. * * For many physical devices, throughput increases with the number of * concurrent operations, but latency typically suffers. Further, physical * devices typically have a limit at which more concurrent operations have no * effect on throughput or can actually cause it to decrease. * * The scheduler selects the next operation to issue by first looking for an * I/O class whose minimum has not been satisfied. Once all are satisfied and * the aggregate maximum has not been hit, the scheduler looks for classes * whose maximum has not been satisfied. Iteration through the I/O classes is * done in the order specified above. No further operations are issued if the * aggregate maximum number of concurrent operations has been hit or if there * are no operations queued for an I/O class that has not hit its maximum. * Every time an I/O is queued or an operation completes, the I/O scheduler * looks for new operations to issue. * * All I/O classes have a fixed maximum number of outstanding operations * except for the async write class. Asynchronous writes represent the data * that is committed to stable storage during the syncing stage for * transaction groups (see txg.c). Transaction groups enter the syncing state * periodically so the number of queued async writes will quickly burst up and * then bleed down to zero. Rather than servicing them as quickly as possible, * the I/O scheduler changes the maximum number of active async write I/Os * according to the amount of dirty data in the pool (see dsl_pool.c). Since * both throughput and latency typically increase with the number of * concurrent operations issued to physical devices, reducing the burstiness * in the number of concurrent operations also stabilizes the response time of * operations from other -- and in particular synchronous -- queues. In broad * strokes, the I/O scheduler will issue more concurrent operations from the * async write queue as there's more dirty data in the pool. * * Async Writes * * The number of concurrent operations issued for the async write I/O class * follows a piece-wise linear function defined by a few adjustable points. * * | o---------| <-- zfs_vdev_async_write_max_active * ^ | /^ | * | | / | | * active | / | | * I/O | / | | * count | / | | * | / | | * |------------o | | <-- zfs_vdev_async_write_min_active * 0|____________^______|_________| * 0% | | 100% of zfs_dirty_data_max * | | * | `-- zfs_vdev_async_write_active_max_dirty_percent * `--------- zfs_vdev_async_write_active_min_dirty_percent * * Until the amount of dirty data exceeds a minimum percentage of the dirty * data allowed in the pool, the I/O scheduler will limit the number of * concurrent operations to the minimum. As that threshold is crossed, the * number of concurrent operations issued increases linearly to the maximum at * the specified maximum percentage of the dirty data allowed in the pool. * * Ideally, the amount of dirty data on a busy pool will stay in the sloped * part of the function between zfs_vdev_async_write_active_min_dirty_percent * and zfs_vdev_async_write_active_max_dirty_percent. If it exceeds the * maximum percentage, this indicates that the rate of incoming data is * greater than the rate that the backend storage can handle. In this case, we * must further throttle incoming writes (see dmu_tx_delay() for details). */ /* * The maximum number of I/Os active to each device. Ideally, this will be >= * the sum of each queue's max_active. It must be at least the sum of each * queue's min_active. */ uint32_t zfs_vdev_max_active = 1000; /* * Per-queue limits on the number of I/Os active to each device. If the * sum of the queue's max_active is < zfs_vdev_max_active, then the * min_active comes into play. We will send min_active from each queue, * and then select from queues in the order defined by zio_priority_t. * * In general, smaller max_active's will lead to lower latency of synchronous * operations. Larger max_active's may lead to higher overall throughput, * depending on underlying storage. * * The ratio of the queues' max_actives determines the balance of performance * between reads, writes, and scrubs. E.g., increasing * zfs_vdev_scrub_max_active will cause the scrub or resilver to complete * more quickly, but reads and writes to have higher latency and lower * throughput. */ uint32_t zfs_vdev_sync_read_min_active = 10; uint32_t zfs_vdev_sync_read_max_active = 10; uint32_t zfs_vdev_sync_write_min_active = 10; uint32_t zfs_vdev_sync_write_max_active = 10; uint32_t zfs_vdev_async_read_min_active = 1; uint32_t zfs_vdev_async_read_max_active = 3; uint32_t zfs_vdev_async_write_min_active = 1; uint32_t zfs_vdev_async_write_max_active = 10; uint32_t zfs_vdev_scrub_min_active = 1; uint32_t zfs_vdev_scrub_max_active = 2; uint32_t zfs_vdev_trim_min_active = 1; /* * TRIM max active is large in comparison to the other values due to the fact * that TRIM IOs are coalesced at the device layer. This value is set such * that a typical SSD can process the queued IOs in a single request. */ uint32_t zfs_vdev_trim_max_active = 64; uint32_t zfs_vdev_removal_min_active = 1; uint32_t zfs_vdev_removal_max_active = 2; /* * When the pool has less than zfs_vdev_async_write_active_min_dirty_percent * dirty data, use zfs_vdev_async_write_min_active. When it has more than * zfs_vdev_async_write_active_max_dirty_percent, use * zfs_vdev_async_write_max_active. The value is linearly interpolated * between min and max. */ int zfs_vdev_async_write_active_min_dirty_percent = 30; int zfs_vdev_async_write_active_max_dirty_percent = 60; /* * To reduce IOPs, we aggregate small adjacent I/Os into one large I/O. * For read I/Os, we also aggregate across small adjacency gaps; for writes * we include spans of optional I/Os to aid aggregation at the disk even when * they aren't able to help us aggregate at this level. */ int zfs_vdev_aggregation_limit = 1 << 20; int zfs_vdev_read_gap_limit = 32 << 10; int zfs_vdev_write_gap_limit = 4 << 10; /* * Define the queue depth percentage for each top-level. This percentage is * used in conjunction with zfs_vdev_async_max_active to determine how many * allocations a specific top-level vdev should handle. Once the queue depth * reaches zfs_vdev_queue_depth_pct * zfs_vdev_async_write_max_active / 100 * then allocator will stop allocating blocks on that top-level device. * The default kernel setting is 1000% which will yield 100 allocations per * device. For userland testing, the default setting is 300% which equates * to 30 allocations per device. */ #ifdef _KERNEL int zfs_vdev_queue_depth_pct = 1000; #else int zfs_vdev_queue_depth_pct = 300; #endif +/* + * When performing allocations for a given metaslab, we want to make sure that + * there are enough IOs to aggregate together to improve throughput. We want to + * ensure that there are at least 128k worth of IOs that can be aggregated, and + * we assume that the average allocation size is 4k, so we need the queue depth + * to be 32 per allocator to get good aggregation of sequential writes. + */ +int zfs_vdev_def_queue_depth = 32; #ifdef __FreeBSD__ #ifdef _KERNEL SYSCTL_DECL(_vfs_zfs_vdev); static int sysctl_zfs_async_write_active_min_dirty_percent(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vfs_zfs_vdev, OID_AUTO, async_write_active_min_dirty_percent, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int), sysctl_zfs_async_write_active_min_dirty_percent, "I", "Percentage of async write dirty data below which " "async_write_min_active is used."); static int sysctl_zfs_async_write_active_max_dirty_percent(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vfs_zfs_vdev, OID_AUTO, async_write_active_max_dirty_percent, CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int), sysctl_zfs_async_write_active_max_dirty_percent, "I", "Percentage of async write dirty data above which " "async_write_max_active is used."); SYSCTL_UINT(_vfs_zfs_vdev, OID_AUTO, max_active, CTLFLAG_RWTUN, &zfs_vdev_max_active, 0, "The maximum number of I/Os of all types active for each device."); #define ZFS_VDEV_QUEUE_KNOB_MIN(name) \ SYSCTL_UINT(_vfs_zfs_vdev, OID_AUTO, name ## _min_active, CTLFLAG_RWTUN,\ &zfs_vdev_ ## name ## _min_active, 0, \ "Initial number of I/O requests of type " #name \ " active for each device"); #define ZFS_VDEV_QUEUE_KNOB_MAX(name) \ SYSCTL_UINT(_vfs_zfs_vdev, OID_AUTO, name ## _max_active, CTLFLAG_RWTUN,\ &zfs_vdev_ ## name ## _max_active, 0, \ "Maximum number of I/O requests of type " #name \ " active for each device"); ZFS_VDEV_QUEUE_KNOB_MIN(sync_read); ZFS_VDEV_QUEUE_KNOB_MAX(sync_read); ZFS_VDEV_QUEUE_KNOB_MIN(sync_write); ZFS_VDEV_QUEUE_KNOB_MAX(sync_write); ZFS_VDEV_QUEUE_KNOB_MIN(async_read); ZFS_VDEV_QUEUE_KNOB_MAX(async_read); ZFS_VDEV_QUEUE_KNOB_MIN(async_write); ZFS_VDEV_QUEUE_KNOB_MAX(async_write); ZFS_VDEV_QUEUE_KNOB_MIN(scrub); ZFS_VDEV_QUEUE_KNOB_MAX(scrub); ZFS_VDEV_QUEUE_KNOB_MIN(trim); ZFS_VDEV_QUEUE_KNOB_MAX(trim); #undef ZFS_VDEV_QUEUE_KNOB SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, aggregation_limit, CTLFLAG_RWTUN, &zfs_vdev_aggregation_limit, 0, "I/O requests are aggregated up to this size"); SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, read_gap_limit, CTLFLAG_RWTUN, &zfs_vdev_read_gap_limit, 0, "Acceptable gap between two reads being aggregated"); SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, write_gap_limit, CTLFLAG_RWTUN, &zfs_vdev_write_gap_limit, 0, "Acceptable gap between two writes being aggregated"); SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, queue_depth_pct, CTLFLAG_RWTUN, &zfs_vdev_queue_depth_pct, 0, "Queue depth percentage for each top-level"); static int sysctl_zfs_async_write_active_min_dirty_percent(SYSCTL_HANDLER_ARGS) { int val, err; val = zfs_vdev_async_write_active_min_dirty_percent; err = sysctl_handle_int(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (val < 0 || val > 100 || val >= zfs_vdev_async_write_active_max_dirty_percent) return (EINVAL); zfs_vdev_async_write_active_min_dirty_percent = val; return (0); } static int sysctl_zfs_async_write_active_max_dirty_percent(SYSCTL_HANDLER_ARGS) { int val, err; val = zfs_vdev_async_write_active_max_dirty_percent; err = sysctl_handle_int(oidp, &val, 0, req); if (err != 0 || req->newptr == NULL) return (err); if (val < 0 || val > 100 || val <= zfs_vdev_async_write_active_min_dirty_percent) return (EINVAL); zfs_vdev_async_write_active_max_dirty_percent = val; return (0); } #endif #endif int vdev_queue_offset_compare(const void *x1, const void *x2) { const zio_t *z1 = x1; const zio_t *z2 = x2; if (z1->io_offset < z2->io_offset) return (-1); if (z1->io_offset > z2->io_offset) return (1); if (z1 < z2) return (-1); if (z1 > z2) return (1); return (0); } static inline avl_tree_t * vdev_queue_class_tree(vdev_queue_t *vq, zio_priority_t p) { return (&vq->vq_class[p].vqc_queued_tree); } static inline avl_tree_t * vdev_queue_type_tree(vdev_queue_t *vq, zio_type_t t) { if (t == ZIO_TYPE_READ) return (&vq->vq_read_offset_tree); else if (t == ZIO_TYPE_WRITE) return (&vq->vq_write_offset_tree); else return (NULL); } int vdev_queue_timestamp_compare(const void *x1, const void *x2) { const zio_t *z1 = x1; const zio_t *z2 = x2; if (z1->io_timestamp < z2->io_timestamp) return (-1); if (z1->io_timestamp > z2->io_timestamp) return (1); if (z1->io_offset < z2->io_offset) return (-1); if (z1->io_offset > z2->io_offset) return (1); if (z1 < z2) return (-1); if (z1 > z2) return (1); return (0); } void vdev_queue_init(vdev_t *vd) { vdev_queue_t *vq = &vd->vdev_queue; mutex_init(&vq->vq_lock, NULL, MUTEX_DEFAULT, NULL); vq->vq_vdev = vd; avl_create(&vq->vq_active_tree, vdev_queue_offset_compare, sizeof (zio_t), offsetof(struct zio, io_queue_node)); avl_create(vdev_queue_type_tree(vq, ZIO_TYPE_READ), vdev_queue_offset_compare, sizeof (zio_t), offsetof(struct zio, io_offset_node)); avl_create(vdev_queue_type_tree(vq, ZIO_TYPE_WRITE), vdev_queue_offset_compare, sizeof (zio_t), offsetof(struct zio, io_offset_node)); for (zio_priority_t p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) { int (*compfn) (const void *, const void *); /* * The synchronous i/o queues are dispatched in FIFO rather * than LBA order. This provides more consistent latency for * these i/os. */ if (p == ZIO_PRIORITY_SYNC_READ || p == ZIO_PRIORITY_SYNC_WRITE) compfn = vdev_queue_timestamp_compare; else compfn = vdev_queue_offset_compare; avl_create(vdev_queue_class_tree(vq, p), compfn, sizeof (zio_t), offsetof(struct zio, io_queue_node)); } vq->vq_lastoffset = 0; } void vdev_queue_fini(vdev_t *vd) { vdev_queue_t *vq = &vd->vdev_queue; for (zio_priority_t p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) avl_destroy(vdev_queue_class_tree(vq, p)); avl_destroy(&vq->vq_active_tree); avl_destroy(vdev_queue_type_tree(vq, ZIO_TYPE_READ)); avl_destroy(vdev_queue_type_tree(vq, ZIO_TYPE_WRITE)); mutex_destroy(&vq->vq_lock); } static void vdev_queue_io_add(vdev_queue_t *vq, zio_t *zio) { spa_t *spa = zio->io_spa; avl_tree_t *qtt; ASSERT(MUTEX_HELD(&vq->vq_lock)); ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); avl_add(vdev_queue_class_tree(vq, zio->io_priority), zio); qtt = vdev_queue_type_tree(vq, zio->io_type); if (qtt) avl_add(qtt, zio); #ifdef illumos mutex_enter(&spa->spa_iokstat_lock); spa->spa_queue_stats[zio->io_priority].spa_queued++; if (spa->spa_iokstat != NULL) kstat_waitq_enter(spa->spa_iokstat->ks_data); mutex_exit(&spa->spa_iokstat_lock); #endif } static void vdev_queue_io_remove(vdev_queue_t *vq, zio_t *zio) { spa_t *spa = zio->io_spa; avl_tree_t *qtt; ASSERT(MUTEX_HELD(&vq->vq_lock)); ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); avl_remove(vdev_queue_class_tree(vq, zio->io_priority), zio); qtt = vdev_queue_type_tree(vq, zio->io_type); if (qtt) avl_remove(qtt, zio); #ifdef illumos mutex_enter(&spa->spa_iokstat_lock); ASSERT3U(spa->spa_queue_stats[zio->io_priority].spa_queued, >, 0); spa->spa_queue_stats[zio->io_priority].spa_queued--; if (spa->spa_iokstat != NULL) kstat_waitq_exit(spa->spa_iokstat->ks_data); mutex_exit(&spa->spa_iokstat_lock); #endif } static void vdev_queue_pending_add(vdev_queue_t *vq, zio_t *zio) { spa_t *spa = zio->io_spa; ASSERT(MUTEX_HELD(&vq->vq_lock)); ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); vq->vq_class[zio->io_priority].vqc_active++; avl_add(&vq->vq_active_tree, zio); #ifdef illumos mutex_enter(&spa->spa_iokstat_lock); spa->spa_queue_stats[zio->io_priority].spa_active++; if (spa->spa_iokstat != NULL) kstat_runq_enter(spa->spa_iokstat->ks_data); mutex_exit(&spa->spa_iokstat_lock); #endif } static void vdev_queue_pending_remove(vdev_queue_t *vq, zio_t *zio) { spa_t *spa = zio->io_spa; ASSERT(MUTEX_HELD(&vq->vq_lock)); ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); vq->vq_class[zio->io_priority].vqc_active--; avl_remove(&vq->vq_active_tree, zio); #ifdef illumos mutex_enter(&spa->spa_iokstat_lock); ASSERT3U(spa->spa_queue_stats[zio->io_priority].spa_active, >, 0); spa->spa_queue_stats[zio->io_priority].spa_active--; if (spa->spa_iokstat != NULL) { kstat_io_t *ksio = spa->spa_iokstat->ks_data; kstat_runq_exit(spa->spa_iokstat->ks_data); if (zio->io_type == ZIO_TYPE_READ) { ksio->reads++; ksio->nread += zio->io_size; } else if (zio->io_type == ZIO_TYPE_WRITE) { ksio->writes++; ksio->nwritten += zio->io_size; } } mutex_exit(&spa->spa_iokstat_lock); #endif } static void vdev_queue_agg_io_done(zio_t *aio) { if (aio->io_type == ZIO_TYPE_READ) { zio_t *pio; zio_link_t *zl = NULL; while ((pio = zio_walk_parents(aio, &zl)) != NULL) { abd_copy_off(pio->io_abd, aio->io_abd, 0, pio->io_offset - aio->io_offset, pio->io_size); } } abd_free(aio->io_abd); } static int vdev_queue_class_min_active(zio_priority_t p) { switch (p) { case ZIO_PRIORITY_SYNC_READ: return (zfs_vdev_sync_read_min_active); case ZIO_PRIORITY_SYNC_WRITE: return (zfs_vdev_sync_write_min_active); case ZIO_PRIORITY_ASYNC_READ: return (zfs_vdev_async_read_min_active); case ZIO_PRIORITY_ASYNC_WRITE: return (zfs_vdev_async_write_min_active); case ZIO_PRIORITY_SCRUB: return (zfs_vdev_scrub_min_active); case ZIO_PRIORITY_TRIM: return (zfs_vdev_trim_min_active); case ZIO_PRIORITY_REMOVAL: return (zfs_vdev_removal_min_active); default: panic("invalid priority %u", p); return (0); } } static __noinline int vdev_queue_max_async_writes(spa_t *spa) { int writes; uint64_t dirty = spa->spa_dsl_pool->dp_dirty_total; uint64_t min_bytes = zfs_dirty_data_max * zfs_vdev_async_write_active_min_dirty_percent / 100; uint64_t max_bytes = zfs_dirty_data_max * zfs_vdev_async_write_active_max_dirty_percent / 100; /* * Sync tasks correspond to interactive user actions. To reduce the * execution time of those actions we push data out as fast as possible. */ if (spa_has_pending_synctask(spa)) { return (zfs_vdev_async_write_max_active); } if (dirty < min_bytes) return (zfs_vdev_async_write_min_active); if (dirty > max_bytes) return (zfs_vdev_async_write_max_active); /* * linear interpolation: * slope = (max_writes - min_writes) / (max_bytes - min_bytes) * move right by min_bytes * move up by min_writes */ writes = (dirty - min_bytes) * (zfs_vdev_async_write_max_active - zfs_vdev_async_write_min_active) / (max_bytes - min_bytes) + zfs_vdev_async_write_min_active; ASSERT3U(writes, >=, zfs_vdev_async_write_min_active); ASSERT3U(writes, <=, zfs_vdev_async_write_max_active); return (writes); } static int vdev_queue_class_max_active(spa_t *spa, zio_priority_t p) { switch (p) { case ZIO_PRIORITY_SYNC_READ: return (zfs_vdev_sync_read_max_active); case ZIO_PRIORITY_SYNC_WRITE: return (zfs_vdev_sync_write_max_active); case ZIO_PRIORITY_ASYNC_READ: return (zfs_vdev_async_read_max_active); case ZIO_PRIORITY_ASYNC_WRITE: return (vdev_queue_max_async_writes(spa)); case ZIO_PRIORITY_SCRUB: return (zfs_vdev_scrub_max_active); case ZIO_PRIORITY_TRIM: return (zfs_vdev_trim_max_active); case ZIO_PRIORITY_REMOVAL: return (zfs_vdev_removal_max_active); default: panic("invalid priority %u", p); return (0); } } /* * Return the i/o class to issue from, or ZIO_PRIORITY_MAX_QUEUEABLE if * there is no eligible class. */ static zio_priority_t vdev_queue_class_to_issue(vdev_queue_t *vq) { spa_t *spa = vq->vq_vdev->vdev_spa; zio_priority_t p; ASSERT(MUTEX_HELD(&vq->vq_lock)); if (avl_numnodes(&vq->vq_active_tree) >= zfs_vdev_max_active) return (ZIO_PRIORITY_NUM_QUEUEABLE); /* find a queue that has not reached its minimum # outstanding i/os */ for (p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) { if (avl_numnodes(vdev_queue_class_tree(vq, p)) > 0 && vq->vq_class[p].vqc_active < vdev_queue_class_min_active(p)) return (p); } /* * If we haven't found a queue, look for one that hasn't reached its * maximum # outstanding i/os. */ for (p = 0; p < ZIO_PRIORITY_NUM_QUEUEABLE; p++) { if (avl_numnodes(vdev_queue_class_tree(vq, p)) > 0 && vq->vq_class[p].vqc_active < vdev_queue_class_max_active(spa, p)) return (p); } /* No eligible queued i/os */ return (ZIO_PRIORITY_NUM_QUEUEABLE); } /* * Compute the range spanned by two i/os, which is the endpoint of the last * (lio->io_offset + lio->io_size) minus start of the first (fio->io_offset). * Conveniently, the gap between fio and lio is given by -IO_SPAN(lio, fio); * thus fio and lio are adjacent if and only if IO_SPAN(lio, fio) == 0. */ #define IO_SPAN(fio, lio) ((lio)->io_offset + (lio)->io_size - (fio)->io_offset) #define IO_GAP(fio, lio) (-IO_SPAN(lio, fio)) static zio_t * vdev_queue_aggregate(vdev_queue_t *vq, zio_t *zio) { zio_t *first, *last, *aio, *dio, *mandatory, *nio; uint64_t maxgap = 0; uint64_t size; boolean_t stretch; avl_tree_t *t; enum zio_flag flags; ASSERT(MUTEX_HELD(&vq->vq_lock)); if (zio->io_flags & ZIO_FLAG_DONT_AGGREGATE) return (NULL); first = last = zio; if (zio->io_type == ZIO_TYPE_READ) maxgap = zfs_vdev_read_gap_limit; /* * We can aggregate I/Os that are sufficiently adjacent and of * the same flavor, as expressed by the AGG_INHERIT flags. * The latter requirement is necessary so that certain * attributes of the I/O, such as whether it's a normal I/O * or a scrub/resilver, can be preserved in the aggregate. * We can include optional I/Os, but don't allow them * to begin a range as they add no benefit in that situation. */ /* * We keep track of the last non-optional I/O. */ mandatory = (first->io_flags & ZIO_FLAG_OPTIONAL) ? NULL : first; /* * Walk backwards through sufficiently contiguous I/Os * recording the last non-optional I/O. */ flags = zio->io_flags & ZIO_FLAG_AGG_INHERIT; t = vdev_queue_type_tree(vq, zio->io_type); while (t != NULL && (dio = AVL_PREV(t, first)) != NULL && (dio->io_flags & ZIO_FLAG_AGG_INHERIT) == flags && IO_SPAN(dio, last) <= zfs_vdev_aggregation_limit && IO_GAP(dio, first) <= maxgap && dio->io_type == zio->io_type) { first = dio; if (mandatory == NULL && !(first->io_flags & ZIO_FLAG_OPTIONAL)) mandatory = first; } /* * Skip any initial optional I/Os. */ while ((first->io_flags & ZIO_FLAG_OPTIONAL) && first != last) { first = AVL_NEXT(t, first); ASSERT(first != NULL); } /* * Walk forward through sufficiently contiguous I/Os. * The aggregation limit does not apply to optional i/os, so that * we can issue contiguous writes even if they are larger than the * aggregation limit. */ while ((dio = AVL_NEXT(t, last)) != NULL && (dio->io_flags & ZIO_FLAG_AGG_INHERIT) == flags && (IO_SPAN(first, dio) <= zfs_vdev_aggregation_limit || (dio->io_flags & ZIO_FLAG_OPTIONAL)) && IO_GAP(last, dio) <= maxgap && dio->io_type == zio->io_type) { last = dio; if (!(last->io_flags & ZIO_FLAG_OPTIONAL)) mandatory = last; } /* * Now that we've established the range of the I/O aggregation * we must decide what to do with trailing optional I/Os. * For reads, there's nothing to do. While we are unable to * aggregate further, it's possible that a trailing optional * I/O would allow the underlying device to aggregate with * subsequent I/Os. We must therefore determine if the next * non-optional I/O is close enough to make aggregation * worthwhile. */ stretch = B_FALSE; if (zio->io_type == ZIO_TYPE_WRITE && mandatory != NULL) { zio_t *nio = last; while ((dio = AVL_NEXT(t, nio)) != NULL && IO_GAP(nio, dio) == 0 && IO_GAP(mandatory, dio) <= zfs_vdev_write_gap_limit) { nio = dio; if (!(nio->io_flags & ZIO_FLAG_OPTIONAL)) { stretch = B_TRUE; break; } } } if (stretch) { /* * We are going to include an optional io in our aggregated * span, thus closing the write gap. Only mandatory i/os can * start aggregated spans, so make sure that the next i/o * after our span is mandatory. */ dio = AVL_NEXT(t, last); dio->io_flags &= ~ZIO_FLAG_OPTIONAL; } else { /* do not include the optional i/o */ while (last != mandatory && last != first) { ASSERT(last->io_flags & ZIO_FLAG_OPTIONAL); last = AVL_PREV(t, last); ASSERT(last != NULL); } } if (first == last) return (NULL); size = IO_SPAN(first, last); ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); aio = zio_vdev_delegated_io(first->io_vd, first->io_offset, abd_alloc_for_io(size, B_TRUE), size, first->io_type, zio->io_priority, flags | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_QUEUE, vdev_queue_agg_io_done, NULL); aio->io_timestamp = first->io_timestamp; nio = first; do { dio = nio; nio = AVL_NEXT(t, dio); ASSERT3U(dio->io_type, ==, aio->io_type); if (dio->io_flags & ZIO_FLAG_NODATA) { ASSERT3U(dio->io_type, ==, ZIO_TYPE_WRITE); abd_zero_off(aio->io_abd, dio->io_offset - aio->io_offset, dio->io_size); } else if (dio->io_type == ZIO_TYPE_WRITE) { abd_copy_off(aio->io_abd, dio->io_abd, dio->io_offset - aio->io_offset, 0, dio->io_size); } zio_add_child(dio, aio); vdev_queue_io_remove(vq, dio); zio_vdev_io_bypass(dio); zio_execute(dio); } while (dio != last); return (aio); } static zio_t * vdev_queue_io_to_issue(vdev_queue_t *vq) { zio_t *zio, *aio; zio_priority_t p; avl_index_t idx; avl_tree_t *tree; zio_t search; again: ASSERT(MUTEX_HELD(&vq->vq_lock)); p = vdev_queue_class_to_issue(vq); if (p == ZIO_PRIORITY_NUM_QUEUEABLE) { /* No eligible queued i/os */ return (NULL); } /* * For LBA-ordered queues (async / scrub), issue the i/o which follows * the most recently issued i/o in LBA (offset) order. * * For FIFO queues (sync), issue the i/o with the lowest timestamp. */ tree = vdev_queue_class_tree(vq, p); search.io_timestamp = 0; search.io_offset = vq->vq_last_offset + 1; VERIFY3P(avl_find(tree, &search, &idx), ==, NULL); zio = avl_nearest(tree, idx, AVL_AFTER); if (zio == NULL) zio = avl_first(tree); ASSERT3U(zio->io_priority, ==, p); aio = vdev_queue_aggregate(vq, zio); if (aio != NULL) zio = aio; else vdev_queue_io_remove(vq, zio); /* * If the I/O is or was optional and therefore has no data, we need to * simply discard it. We need to drop the vdev queue's lock to avoid a * deadlock that we could encounter since this I/O will complete * immediately. */ if (zio->io_flags & ZIO_FLAG_NODATA) { mutex_exit(&vq->vq_lock); zio_vdev_io_bypass(zio); zio_execute(zio); mutex_enter(&vq->vq_lock); goto again; } vdev_queue_pending_add(vq, zio); vq->vq_last_offset = zio->io_offset; return (zio); } zio_t * vdev_queue_io(zio_t *zio) { vdev_queue_t *vq = &zio->io_vd->vdev_queue; zio_t *nio; if (zio->io_flags & ZIO_FLAG_DONT_QUEUE) return (zio); /* * Children i/os inherent their parent's priority, which might * not match the child's i/o type. Fix it up here. */ if (zio->io_type == ZIO_TYPE_READ) { if (zio->io_priority != ZIO_PRIORITY_SYNC_READ && zio->io_priority != ZIO_PRIORITY_ASYNC_READ && zio->io_priority != ZIO_PRIORITY_SCRUB && zio->io_priority != ZIO_PRIORITY_REMOVAL) zio->io_priority = ZIO_PRIORITY_ASYNC_READ; } else if (zio->io_type == ZIO_TYPE_WRITE) { if (zio->io_priority != ZIO_PRIORITY_SYNC_WRITE && zio->io_priority != ZIO_PRIORITY_ASYNC_WRITE && zio->io_priority != ZIO_PRIORITY_REMOVAL) zio->io_priority = ZIO_PRIORITY_ASYNC_WRITE; } else { ASSERT(zio->io_type == ZIO_TYPE_FREE); zio->io_priority = ZIO_PRIORITY_TRIM; } zio->io_flags |= ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_QUEUE; mutex_enter(&vq->vq_lock); zio->io_timestamp = gethrtime(); vdev_queue_io_add(vq, zio); nio = vdev_queue_io_to_issue(vq); mutex_exit(&vq->vq_lock); if (nio == NULL) return (NULL); if (nio->io_done == vdev_queue_agg_io_done) { zio_nowait(nio); return (NULL); } return (nio); } void vdev_queue_io_done(zio_t *zio) { vdev_queue_t *vq = &zio->io_vd->vdev_queue; zio_t *nio; mutex_enter(&vq->vq_lock); vdev_queue_pending_remove(vq, zio); vq->vq_io_complete_ts = gethrtime(); while ((nio = vdev_queue_io_to_issue(vq)) != NULL) { mutex_exit(&vq->vq_lock); if (nio->io_done == vdev_queue_agg_io_done) { zio_nowait(nio); } else { zio_vdev_io_reissue(nio); zio_execute(nio); } mutex_enter(&vq->vq_lock); } mutex_exit(&vq->vq_lock); } void vdev_queue_change_io_priority(zio_t *zio, zio_priority_t priority) { vdev_queue_t *vq = &zio->io_vd->vdev_queue; avl_tree_t *tree; ASSERT3U(zio->io_priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); ASSERT3U(priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); if (zio->io_type == ZIO_TYPE_READ) { if (priority != ZIO_PRIORITY_SYNC_READ && priority != ZIO_PRIORITY_ASYNC_READ && priority != ZIO_PRIORITY_SCRUB) priority = ZIO_PRIORITY_ASYNC_READ; } else { ASSERT(zio->io_type == ZIO_TYPE_WRITE); if (priority != ZIO_PRIORITY_SYNC_WRITE && priority != ZIO_PRIORITY_ASYNC_WRITE) priority = ZIO_PRIORITY_ASYNC_WRITE; } mutex_enter(&vq->vq_lock); /* * If the zio is in none of the queues we can simply change * the priority. If the zio is waiting to be submitted we must * remove it from the queue and re-insert it with the new priority. * Otherwise, the zio is currently active and we cannot change its * priority. */ tree = vdev_queue_class_tree(vq, zio->io_priority); if (avl_find(tree, zio, NULL) == zio) { avl_remove(vdev_queue_class_tree(vq, zio->io_priority), zio); zio->io_priority = priority; avl_add(vdev_queue_class_tree(vq, zio->io_priority), zio); } else if (avl_find(&vq->vq_active_tree, zio, NULL) != zio) { zio->io_priority = priority; } mutex_exit(&vq->vq_lock); } /* * As these three methods are only used for load calculations we're not concerned * if we get an incorrect value on 32bit platforms due to lack of vq_lock mutex * use here, instead we prefer to keep it lock free for performance. */ int vdev_queue_length(vdev_t *vd) { return (avl_numnodes(&vd->vdev_queue.vq_active_tree)); } uint64_t vdev_queue_lastoffset(vdev_t *vd) { return (vd->vdev_queue.vq_lastoffset); } void vdev_queue_register_lastoffset(vdev_t *vd, zio_t *zio) { vd->vdev_queue.vq_lastoffset = zio->io_offset + zio->io_size; } Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_removal.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_removal.c (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_removal.c (revision 336949) @@ -1,1956 +1,1963 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2018 by Delphix. All rights reserved. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * This file contains the necessary logic to remove vdevs from a * storage pool. Currently, the only devices that can be removed * are log, cache, and spare devices; and top level vdevs from a pool * w/o raidz. (Note that members of a mirror can also be removed * by the detach operation.) * * Log vdevs are removed by evacuating them and then turning the vdev * into a hole vdev while holding spa config locks. * * Top level vdevs are removed and converted into an indirect vdev via * a multi-step process: * * - Disable allocations from this device (spa_vdev_remove_top). * * - From a new thread (spa_vdev_remove_thread), copy data from * the removing vdev to a different vdev. The copy happens in open * context (spa_vdev_copy_impl) and issues a sync task * (vdev_mapping_sync) so the sync thread can update the partial * indirect mappings in core and on disk. * * - If a free happens during a removal, it is freed from the * removing vdev, and if it has already been copied, from the new * location as well (free_from_removing_vdev). * * - After the removal is completed, the copy thread converts the vdev * into an indirect vdev (vdev_remove_complete) before instructing * the sync thread to destroy the space maps and finish the removal * (spa_finish_removal). */ typedef struct vdev_copy_arg { metaslab_t *vca_msp; uint64_t vca_outstanding_bytes; kcondvar_t vca_cv; kmutex_t vca_lock; } vdev_copy_arg_t; typedef struct vdev_copy_seg_arg { vdev_copy_arg_t *vcsa_copy_arg; uint64_t vcsa_txg; dva_t *vcsa_dest_dva; blkptr_t *vcsa_dest_bp; } vdev_copy_seg_arg_t; /* * The maximum amount of allowed data we're allowed to copy from a device * at a time when removing it. */ int zfs_remove_max_copy_bytes = 8 * 1024 * 1024; /* * The largest contiguous segment that we will attempt to allocate when * removing a device. This can be no larger than SPA_MAXBLOCKSIZE. If * there is a performance problem with attempting to allocate large blocks, * consider decreasing this. * * Note: we will issue I/Os of up to this size. The mpt driver does not * respond well to I/Os larger than 1MB, so we set this to 1MB. (When * mpt processes an I/O larger than 1MB, it needs to do an allocation of * 2 physically contiguous pages; if this allocation fails, mpt will drop * the I/O and hang the device.) */ int zfs_remove_max_segment = 1024 * 1024; /* * This is used by the test suite so that it can ensure that certain * actions happen while in the middle of a removal. */ uint64_t zfs_remove_max_bytes_pause = UINT64_MAX; #define VDEV_REMOVAL_ZAP_OBJS "lzap" static void spa_vdev_remove_thread(void *arg); static void spa_sync_removing_state(spa_t *spa, dmu_tx_t *tx) { VERIFY0(zap_update(spa->spa_dsl_pool->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_REMOVING, sizeof (uint64_t), sizeof (spa->spa_removing_phys) / sizeof (uint64_t), &spa->spa_removing_phys, tx)); } static nvlist_t * spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) { for (int i = 0; i < count; i++) { uint64_t guid = fnvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID); if (guid == target_guid) return (nvpp[i]); } return (NULL); } static void spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, nvlist_t *dev_to_remove) { nvlist_t **newdev = NULL; if (count > 1) newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); for (int i = 0, j = 0; i < count; i++) { if (dev[i] == dev_to_remove) continue; VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); } VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); for (int i = 0; i < count - 1; i++) nvlist_free(newdev[i]); if (count > 1) kmem_free(newdev, (count - 1) * sizeof (void *)); } static spa_vdev_removal_t * spa_vdev_removal_create(vdev_t *vd) { spa_vdev_removal_t *svr = kmem_zalloc(sizeof (*svr), KM_SLEEP); mutex_init(&svr->svr_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&svr->svr_cv, NULL, CV_DEFAULT, NULL); svr->svr_allocd_segs = range_tree_create(NULL, NULL); svr->svr_vdev = vd; for (int i = 0; i < TXG_SIZE; i++) { svr->svr_frees[i] = range_tree_create(NULL, NULL); list_create(&svr->svr_new_segments[i], sizeof (vdev_indirect_mapping_entry_t), offsetof(vdev_indirect_mapping_entry_t, vime_node)); } return (svr); } void spa_vdev_removal_destroy(spa_vdev_removal_t *svr) { for (int i = 0; i < TXG_SIZE; i++) { ASSERT0(svr->svr_bytes_done[i]); ASSERT0(svr->svr_max_offset_to_sync[i]); range_tree_destroy(svr->svr_frees[i]); list_destroy(&svr->svr_new_segments[i]); } range_tree_destroy(svr->svr_allocd_segs); mutex_destroy(&svr->svr_lock); cv_destroy(&svr->svr_cv); kmem_free(svr, sizeof (*svr)); } /* * This is called as a synctask in the txg in which we will mark this vdev * as removing (in the config stored in the MOS). * * It begins the evacuation of a toplevel vdev by: * - initializing the spa_removing_phys which tracks this removal * - computing the amount of space to remove for accounting purposes * - dirtying all dbufs in the spa_config_object * - creating the spa_vdev_removal * - starting the spa_vdev_remove_thread */ static void vdev_remove_initiate_sync(void *arg, dmu_tx_t *tx) { vdev_t *vd = arg; vdev_indirect_config_t *vic = &vd->vdev_indirect_config; spa_t *spa = vd->vdev_spa; objset_t *mos = spa->spa_dsl_pool->dp_meta_objset; spa_vdev_removal_t *svr = NULL; uint64_t txg = dmu_tx_get_txg(tx); ASSERT3P(vd->vdev_ops, !=, &vdev_raidz_ops); svr = spa_vdev_removal_create(vd); ASSERT(vd->vdev_removing); ASSERT3P(vd->vdev_indirect_mapping, ==, NULL); spa_feature_incr(spa, SPA_FEATURE_DEVICE_REMOVAL, tx); if (spa_feature_is_enabled(spa, SPA_FEATURE_OBSOLETE_COUNTS)) { /* * By activating the OBSOLETE_COUNTS feature, we prevent * the pool from being downgraded and ensure that the * refcounts are precise. */ spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); uint64_t one = 1; VERIFY0(zap_add(spa->spa_meta_objset, vd->vdev_top_zap, VDEV_TOP_ZAP_OBSOLETE_COUNTS_ARE_PRECISE, sizeof (one), 1, &one, tx)); ASSERT3U(vdev_obsolete_counts_are_precise(vd), !=, 0); } vic->vic_mapping_object = vdev_indirect_mapping_alloc(mos, tx); vd->vdev_indirect_mapping = vdev_indirect_mapping_open(mos, vic->vic_mapping_object); vic->vic_births_object = vdev_indirect_births_alloc(mos, tx); vd->vdev_indirect_births = vdev_indirect_births_open(mos, vic->vic_births_object); spa->spa_removing_phys.sr_removing_vdev = vd->vdev_id; spa->spa_removing_phys.sr_start_time = gethrestime_sec(); spa->spa_removing_phys.sr_end_time = 0; spa->spa_removing_phys.sr_state = DSS_SCANNING; spa->spa_removing_phys.sr_to_copy = 0; spa->spa_removing_phys.sr_copied = 0; /* * Note: We can't use vdev_stat's vs_alloc for sr_to_copy, because * there may be space in the defer tree, which is free, but still * counted in vs_alloc. */ for (uint64_t i = 0; i < vd->vdev_ms_count; i++) { metaslab_t *ms = vd->vdev_ms[i]; if (ms->ms_sm == NULL) continue; /* * Sync tasks happen before metaslab_sync(), therefore * smp_alloc and sm_alloc must be the same. */ ASSERT3U(space_map_allocated(ms->ms_sm), ==, ms->ms_sm->sm_phys->smp_alloc); spa->spa_removing_phys.sr_to_copy += space_map_allocated(ms->ms_sm); /* * Space which we are freeing this txg does not need to * be copied. */ spa->spa_removing_phys.sr_to_copy -= range_tree_space(ms->ms_freeing); ASSERT0(range_tree_space(ms->ms_freed)); for (int t = 0; t < TXG_SIZE; t++) ASSERT0(range_tree_space(ms->ms_allocating[t])); } /* * Sync tasks are called before metaslab_sync(), so there should * be no already-synced metaslabs in the TXG_CLEAN list. */ ASSERT3P(txg_list_head(&vd->vdev_ms_list, TXG_CLEAN(txg)), ==, NULL); spa_sync_removing_state(spa, tx); /* * All blocks that we need to read the most recent mapping must be * stored on concrete vdevs. Therefore, we must dirty anything that * is read before spa_remove_init(). Specifically, the * spa_config_object. (Note that although we already modified the * spa_config_object in spa_sync_removing_state, that may not have * modified all blocks of the object.) */ dmu_object_info_t doi; VERIFY0(dmu_object_info(mos, DMU_POOL_DIRECTORY_OBJECT, &doi)); for (uint64_t offset = 0; offset < doi.doi_max_offset; ) { dmu_buf_t *dbuf; VERIFY0(dmu_buf_hold(mos, DMU_POOL_DIRECTORY_OBJECT, offset, FTAG, &dbuf, 0)); dmu_buf_will_dirty(dbuf, tx); offset += dbuf->db_size; dmu_buf_rele(dbuf, FTAG); } /* * Now that we've allocated the im_object, dirty the vdev to ensure * that the object gets written to the config on disk. */ vdev_config_dirty(vd); zfs_dbgmsg("starting removal thread for vdev %llu (%p) in txg %llu " "im_obj=%llu", vd->vdev_id, vd, dmu_tx_get_txg(tx), vic->vic_mapping_object); spa_history_log_internal(spa, "vdev remove started", tx, "%s vdev %llu %s", spa_name(spa), vd->vdev_id, (vd->vdev_path != NULL) ? vd->vdev_path : "-"); /* * Setting spa_vdev_removal causes subsequent frees to call * free_from_removing_vdev(). Note that we don't need any locking * because we are the sync thread, and metaslab_free_impl() is only * called from syncing context (potentially from a zio taskq thread, * but in any case only when there are outstanding free i/os, which * there are not). */ ASSERT3P(spa->spa_vdev_removal, ==, NULL); spa->spa_vdev_removal = svr; svr->svr_thread = thread_create(NULL, 0, spa_vdev_remove_thread, vd, 0, &p0, TS_RUN, minclsyspri); } /* * When we are opening a pool, we must read the mapping for each * indirect vdev in order from most recently removed to least * recently removed. We do this because the blocks for the mapping * of older indirect vdevs may be stored on more recently removed vdevs. * In order to read each indirect mapping object, we must have * initialized all more recently removed vdevs. */ int spa_remove_init(spa_t *spa) { int error; error = zap_lookup(spa->spa_dsl_pool->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_REMOVING, sizeof (uint64_t), sizeof (spa->spa_removing_phys) / sizeof (uint64_t), &spa->spa_removing_phys); if (error == ENOENT) { spa->spa_removing_phys.sr_state = DSS_NONE; spa->spa_removing_phys.sr_removing_vdev = -1; spa->spa_removing_phys.sr_prev_indirect_vdev = -1; spa->spa_indirect_vdevs_loaded = B_TRUE; return (0); } else if (error != 0) { return (error); } if (spa->spa_removing_phys.sr_state == DSS_SCANNING) { /* * We are currently removing a vdev. Create and * initialize a spa_vdev_removal_t from the bonus * buffer of the removing vdevs vdev_im_object, and * initialize its partial mapping. */ spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); vdev_t *vd = vdev_lookup_top(spa, spa->spa_removing_phys.sr_removing_vdev); spa_config_exit(spa, SCL_STATE, FTAG); if (vd == NULL) return (EINVAL); vdev_indirect_config_t *vic = &vd->vdev_indirect_config; ASSERT(vdev_is_concrete(vd)); spa_vdev_removal_t *svr = spa_vdev_removal_create(vd); ASSERT(svr->svr_vdev->vdev_removing); vd->vdev_indirect_mapping = vdev_indirect_mapping_open( spa->spa_meta_objset, vic->vic_mapping_object); vd->vdev_indirect_births = vdev_indirect_births_open( spa->spa_meta_objset, vic->vic_births_object); spa->spa_vdev_removal = svr; } spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); uint64_t indirect_vdev_id = spa->spa_removing_phys.sr_prev_indirect_vdev; while (indirect_vdev_id != UINT64_MAX) { vdev_t *vd = vdev_lookup_top(spa, indirect_vdev_id); vdev_indirect_config_t *vic = &vd->vdev_indirect_config; ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops); vd->vdev_indirect_mapping = vdev_indirect_mapping_open( spa->spa_meta_objset, vic->vic_mapping_object); vd->vdev_indirect_births = vdev_indirect_births_open( spa->spa_meta_objset, vic->vic_births_object); indirect_vdev_id = vic->vic_prev_indirect_vdev; } spa_config_exit(spa, SCL_STATE, FTAG); /* * Now that we've loaded all the indirect mappings, we can allow * reads from other blocks (e.g. via predictive prefetch). */ spa->spa_indirect_vdevs_loaded = B_TRUE; return (0); } void spa_restart_removal(spa_t *spa) { spa_vdev_removal_t *svr = spa->spa_vdev_removal; if (svr == NULL) return; /* * In general when this function is called there is no * removal thread running. The only scenario where this * is not true is during spa_import() where this function * is called twice [once from spa_import_impl() and * spa_async_resume()]. Thus, in the scenario where we * import a pool that has an ongoing removal we don't * want to spawn a second thread. */ if (svr->svr_thread != NULL) return; if (!spa_writeable(spa)) return; vdev_t *vd = svr->svr_vdev; vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; ASSERT3P(vd, !=, NULL); ASSERT(vd->vdev_removing); zfs_dbgmsg("restarting removal of %llu at count=%llu", vd->vdev_id, vdev_indirect_mapping_num_entries(vim)); svr->svr_thread = thread_create(NULL, 0, spa_vdev_remove_thread, vd, 0, &p0, TS_RUN, minclsyspri); } /* * Process freeing from a device which is in the middle of being removed. * We must handle this carefully so that we attempt to copy freed data, * and we correctly free already-copied data. */ void free_from_removing_vdev(vdev_t *vd, uint64_t offset, uint64_t size) { spa_t *spa = vd->vdev_spa; spa_vdev_removal_t *svr = spa->spa_vdev_removal; vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; uint64_t txg = spa_syncing_txg(spa); uint64_t max_offset_yet = 0; ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0); ASSERT3U(vd->vdev_indirect_config.vic_mapping_object, ==, vdev_indirect_mapping_object(vim)); ASSERT3P(vd, ==, svr->svr_vdev); mutex_enter(&svr->svr_lock); /* * Remove the segment from the removing vdev's spacemap. This * ensures that we will not attempt to copy this space (if the * removal thread has not yet visited it), and also ensures * that we know what is actually allocated on the new vdevs * (needed if we cancel the removal). * * Note: we must do the metaslab_free_concrete() with the svr_lock * held, so that the remove_thread can not load this metaslab and then * visit this offset between the time that we metaslab_free_concrete() * and when we check to see if it has been visited. * * Note: The checkpoint flag is set to false as having/taking * a checkpoint and removing a device can't happen at the same * time. */ ASSERT(!spa_has_checkpoint(spa)); metaslab_free_concrete(vd, offset, size, B_FALSE); uint64_t synced_size = 0; uint64_t synced_offset = 0; uint64_t max_offset_synced = vdev_indirect_mapping_max_offset(vim); if (offset < max_offset_synced) { /* * The mapping for this offset is already on disk. * Free from the new location. * * Note that we use svr_max_synced_offset because it is * updated atomically with respect to the in-core mapping. * By contrast, vim_max_offset is not. * * This block may be split between a synced entry and an * in-flight or unvisited entry. Only process the synced * portion of it here. */ synced_size = MIN(size, max_offset_synced - offset); synced_offset = offset; ASSERT3U(max_offset_yet, <=, max_offset_synced); max_offset_yet = max_offset_synced; DTRACE_PROBE3(remove__free__synced, spa_t *, spa, uint64_t, offset, uint64_t, synced_size); size -= synced_size; offset += synced_size; } /* * Look at all in-flight txgs starting from the currently syncing one * and see if a section of this free is being copied. By starting from * this txg and iterating forward, we might find that this region * was copied in two different txgs and handle it appropriately. */ for (int i = 0; i < TXG_CONCURRENT_STATES; i++) { int txgoff = (txg + i) & TXG_MASK; if (size > 0 && offset < svr->svr_max_offset_to_sync[txgoff]) { /* * The mapping for this offset is in flight, and * will be synced in txg+i. */ uint64_t inflight_size = MIN(size, svr->svr_max_offset_to_sync[txgoff] - offset); DTRACE_PROBE4(remove__free__inflight, spa_t *, spa, uint64_t, offset, uint64_t, inflight_size, uint64_t, txg + i); /* * We copy data in order of increasing offset. * Therefore the max_offset_to_sync[] must increase * (or be zero, indicating that nothing is being * copied in that txg). */ if (svr->svr_max_offset_to_sync[txgoff] != 0) { ASSERT3U(svr->svr_max_offset_to_sync[txgoff], >=, max_offset_yet); max_offset_yet = svr->svr_max_offset_to_sync[txgoff]; } /* * We've already committed to copying this segment: * we have allocated space elsewhere in the pool for * it and have an IO outstanding to copy the data. We * cannot free the space before the copy has * completed, or else the copy IO might overwrite any * new data. To free that space, we record the * segment in the appropriate svr_frees tree and free * the mapped space later, in the txg where we have * completed the copy and synced the mapping (see * vdev_mapping_sync). */ range_tree_add(svr->svr_frees[txgoff], offset, inflight_size); size -= inflight_size; offset += inflight_size; /* * This space is already accounted for as being * done, because it is being copied in txg+i. * However, if i!=0, then it is being copied in * a future txg. If we crash after this txg * syncs but before txg+i syncs, then the space * will be free. Therefore we must account * for the space being done in *this* txg * (when it is freed) rather than the future txg * (when it will be copied). */ ASSERT3U(svr->svr_bytes_done[txgoff], >=, inflight_size); svr->svr_bytes_done[txgoff] -= inflight_size; svr->svr_bytes_done[txg & TXG_MASK] += inflight_size; } } ASSERT0(svr->svr_max_offset_to_sync[TXG_CLEAN(txg) & TXG_MASK]); if (size > 0) { /* * The copy thread has not yet visited this offset. Ensure * that it doesn't. */ DTRACE_PROBE3(remove__free__unvisited, spa_t *, spa, uint64_t, offset, uint64_t, size); if (svr->svr_allocd_segs != NULL) range_tree_clear(svr->svr_allocd_segs, offset, size); /* * Since we now do not need to copy this data, for * accounting purposes we have done our job and can count * it as completed. */ svr->svr_bytes_done[txg & TXG_MASK] += size; } mutex_exit(&svr->svr_lock); /* * Now that we have dropped svr_lock, process the synced portion * of this free. */ if (synced_size > 0) { vdev_indirect_mark_obsolete(vd, synced_offset, synced_size); /* * Note: this can only be called from syncing context, * and the vdev_indirect_mapping is only changed from the * sync thread, so we don't need svr_lock while doing * metaslab_free_impl_cb. */ boolean_t checkpoint = B_FALSE; vdev_indirect_ops.vdev_op_remap(vd, synced_offset, synced_size, metaslab_free_impl_cb, &checkpoint); } } /* * Stop an active removal and update the spa_removing phys. */ static void spa_finish_removal(spa_t *spa, dsl_scan_state_t state, dmu_tx_t *tx) { spa_vdev_removal_t *svr = spa->spa_vdev_removal; ASSERT3U(dmu_tx_get_txg(tx), ==, spa_syncing_txg(spa)); /* Ensure the removal thread has completed before we free the svr. */ spa_vdev_remove_suspend(spa); ASSERT(state == DSS_FINISHED || state == DSS_CANCELED); if (state == DSS_FINISHED) { spa_removing_phys_t *srp = &spa->spa_removing_phys; vdev_t *vd = svr->svr_vdev; vdev_indirect_config_t *vic = &vd->vdev_indirect_config; if (srp->sr_prev_indirect_vdev != UINT64_MAX) { vdev_t *pvd = vdev_lookup_top(spa, srp->sr_prev_indirect_vdev); ASSERT3P(pvd->vdev_ops, ==, &vdev_indirect_ops); } vic->vic_prev_indirect_vdev = srp->sr_prev_indirect_vdev; srp->sr_prev_indirect_vdev = vd->vdev_id; } spa->spa_removing_phys.sr_state = state; spa->spa_removing_phys.sr_end_time = gethrestime_sec(); spa->spa_vdev_removal = NULL; spa_vdev_removal_destroy(svr); spa_sync_removing_state(spa, tx); vdev_config_dirty(spa->spa_root_vdev); } static void free_mapped_segment_cb(void *arg, uint64_t offset, uint64_t size) { vdev_t *vd = arg; vdev_indirect_mark_obsolete(vd, offset, size); boolean_t checkpoint = B_FALSE; vdev_indirect_ops.vdev_op_remap(vd, offset, size, metaslab_free_impl_cb, &checkpoint); } /* * On behalf of the removal thread, syncs an incremental bit more of * the indirect mapping to disk and updates the in-memory mapping. * Called as a sync task in every txg that the removal thread makes progress. */ static void vdev_mapping_sync(void *arg, dmu_tx_t *tx) { spa_vdev_removal_t *svr = arg; spa_t *spa = dmu_tx_pool(tx)->dp_spa; vdev_t *vd = svr->svr_vdev; vdev_indirect_config_t *vic = &vd->vdev_indirect_config; uint64_t txg = dmu_tx_get_txg(tx); vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; ASSERT(vic->vic_mapping_object != 0); ASSERT3U(txg, ==, spa_syncing_txg(spa)); vdev_indirect_mapping_add_entries(vim, &svr->svr_new_segments[txg & TXG_MASK], tx); vdev_indirect_births_add_entry(vd->vdev_indirect_births, vdev_indirect_mapping_max_offset(vim), dmu_tx_get_txg(tx), tx); /* * Free the copied data for anything that was freed while the * mapping entries were in flight. */ mutex_enter(&svr->svr_lock); range_tree_vacate(svr->svr_frees[txg & TXG_MASK], free_mapped_segment_cb, vd); ASSERT3U(svr->svr_max_offset_to_sync[txg & TXG_MASK], >=, vdev_indirect_mapping_max_offset(vim)); svr->svr_max_offset_to_sync[txg & TXG_MASK] = 0; mutex_exit(&svr->svr_lock); spa_sync_removing_state(spa, tx); } static void spa_vdev_copy_segment_write_done(zio_t *zio) { vdev_copy_seg_arg_t *vcsa = zio->io_private; vdev_copy_arg_t *vca = vcsa->vcsa_copy_arg; spa_config_exit(zio->io_spa, SCL_STATE, FTAG); abd_free(zio->io_abd); mutex_enter(&vca->vca_lock); vca->vca_outstanding_bytes -= zio->io_size; cv_signal(&vca->vca_cv); mutex_exit(&vca->vca_lock); ASSERT0(zio->io_error); kmem_free(vcsa->vcsa_dest_bp, sizeof (blkptr_t)); kmem_free(vcsa, sizeof (vdev_copy_seg_arg_t)); } static void spa_vdev_copy_segment_read_done(zio_t *zio) { vdev_copy_seg_arg_t *vcsa = zio->io_private; dva_t *dest_dva = vcsa->vcsa_dest_dva; uint64_t txg = vcsa->vcsa_txg; spa_t *spa = zio->io_spa; vdev_t *dest_vd = vdev_lookup_top(spa, DVA_GET_VDEV(dest_dva)); blkptr_t *bp = NULL; dva_t *dva = NULL; uint64_t size = zio->io_size; ASSERT3P(dest_vd, !=, NULL); ASSERT0(zio->io_error); vcsa->vcsa_dest_bp = kmem_alloc(sizeof (blkptr_t), KM_SLEEP); bp = vcsa->vcsa_dest_bp; dva = bp->blk_dva; BP_ZERO(bp); /* initialize with dest_dva */ bcopy(dest_dva, dva, sizeof (dva_t)); BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL); BP_SET_LSIZE(bp, size); BP_SET_PSIZE(bp, size); BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF); BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF); BP_SET_TYPE(bp, DMU_OT_NONE); BP_SET_LEVEL(bp, 0); BP_SET_DEDUP(bp, 0); BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); zio_nowait(zio_rewrite(spa->spa_txg_zio[txg & TXG_MASK], spa, txg, bp, zio->io_abd, size, spa_vdev_copy_segment_write_done, vcsa, ZIO_PRIORITY_REMOVAL, 0, NULL)); } static int spa_vdev_copy_segment(vdev_t *vd, uint64_t start, uint64_t size, uint64_t txg, vdev_copy_arg_t *vca, zio_alloc_list_t *zal) { metaslab_group_t *mg = vd->vdev_mg; spa_t *spa = vd->vdev_spa; spa_vdev_removal_t *svr = spa->spa_vdev_removal; vdev_indirect_mapping_entry_t *entry; vdev_copy_seg_arg_t *private; dva_t dst = { 0 }; blkptr_t blk, *bp = &blk; dva_t *dva = bp->blk_dva; ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); + /* + * We use allocator 0 for this I/O because we don't expect device remap + * to be the steady state of the system, so parallelizing is not as + * critical as it is for other allocation types. We also want to ensure + * that the IOs are allocated together as much as possible, to reduce + * mapping sizes. + */ int error = metaslab_alloc_dva(spa, mg->mg_class, size, - &dst, 0, NULL, txg, 0, zal); + &dst, 0, NULL, txg, 0, zal, 0); if (error != 0) return (error); /* * We can't have any padding of the allocated size, otherwise we will * misunderstand what's allocated, and the size of the mapping. * The caller ensures this will be true by passing in a size that is * aligned to the worst (highest) ashift in the pool. */ ASSERT3U(DVA_GET_ASIZE(&dst), ==, size); mutex_enter(&vca->vca_lock); vca->vca_outstanding_bytes += size; mutex_exit(&vca->vca_lock); entry = kmem_zalloc(sizeof (vdev_indirect_mapping_entry_t), KM_SLEEP); DVA_MAPPING_SET_SRC_OFFSET(&entry->vime_mapping, start); entry->vime_mapping.vimep_dst = dst; private = kmem_alloc(sizeof (vdev_copy_seg_arg_t), KM_SLEEP); private->vcsa_dest_dva = &entry->vime_mapping.vimep_dst; private->vcsa_txg = txg; private->vcsa_copy_arg = vca; /* * This lock is eventually released by the donefunc for the * zio_write_phys that finishes copying the data. */ spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); /* * Do logical I/O, letting the redundancy vdevs (like mirror) * handle their own I/O instead of duplicating that code here. */ BP_ZERO(bp); DVA_SET_VDEV(&dva[0], vd->vdev_id); DVA_SET_OFFSET(&dva[0], start); DVA_SET_GANG(&dva[0], 0); DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, size)); BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL); BP_SET_LSIZE(bp, size); BP_SET_PSIZE(bp, size); BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF); BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF); BP_SET_TYPE(bp, DMU_OT_NONE); BP_SET_LEVEL(bp, 0); BP_SET_DEDUP(bp, 0); BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); zio_nowait(zio_read(spa->spa_txg_zio[txg & TXG_MASK], spa, bp, abd_alloc_for_io(size, B_FALSE), size, spa_vdev_copy_segment_read_done, private, ZIO_PRIORITY_REMOVAL, 0, NULL)); list_insert_tail(&svr->svr_new_segments[txg & TXG_MASK], entry); ASSERT3U(start + size, <=, vd->vdev_ms_count << vd->vdev_ms_shift); vdev_dirty(vd, 0, NULL, txg); return (0); } /* * Complete the removal of a toplevel vdev. This is called as a * synctask in the same txg that we will sync out the new config (to the * MOS object) which indicates that this vdev is indirect. */ static void vdev_remove_complete_sync(void *arg, dmu_tx_t *tx) { spa_vdev_removal_t *svr = arg; vdev_t *vd = svr->svr_vdev; spa_t *spa = vd->vdev_spa; ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops); for (int i = 0; i < TXG_SIZE; i++) { ASSERT0(svr->svr_bytes_done[i]); } ASSERT3U(spa->spa_removing_phys.sr_copied, ==, spa->spa_removing_phys.sr_to_copy); vdev_destroy_spacemaps(vd, tx); /* destroy leaf zaps, if any */ ASSERT3P(svr->svr_zaplist, !=, NULL); for (nvpair_t *pair = nvlist_next_nvpair(svr->svr_zaplist, NULL); pair != NULL; pair = nvlist_next_nvpair(svr->svr_zaplist, pair)) { vdev_destroy_unlink_zap(vd, fnvpair_value_uint64(pair), tx); } fnvlist_free(svr->svr_zaplist); spa_finish_removal(dmu_tx_pool(tx)->dp_spa, DSS_FINISHED, tx); /* vd->vdev_path is not available here */ spa_history_log_internal(spa, "vdev remove completed", tx, "%s vdev %llu", spa_name(spa), vd->vdev_id); } static void vdev_indirect_state_transfer(vdev_t *ivd, vdev_t *vd) { ivd->vdev_indirect_config = vd->vdev_indirect_config; ASSERT3P(ivd->vdev_indirect_mapping, ==, NULL); ASSERT(vd->vdev_indirect_mapping != NULL); ivd->vdev_indirect_mapping = vd->vdev_indirect_mapping; vd->vdev_indirect_mapping = NULL; ASSERT3P(ivd->vdev_indirect_births, ==, NULL); ASSERT(vd->vdev_indirect_births != NULL); ivd->vdev_indirect_births = vd->vdev_indirect_births; vd->vdev_indirect_births = NULL; ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); ASSERT0(range_tree_space(ivd->vdev_obsolete_segments)); if (vd->vdev_obsolete_sm != NULL) { ASSERT3U(ivd->vdev_asize, ==, vd->vdev_asize); /* * We cannot use space_map_{open,close} because we hold all * the config locks as writer. */ ASSERT3P(ivd->vdev_obsolete_sm, ==, NULL); ivd->vdev_obsolete_sm = vd->vdev_obsolete_sm; vd->vdev_obsolete_sm = NULL; } } static void vdev_remove_enlist_zaps(vdev_t *vd, nvlist_t *zlist) { ASSERT3P(zlist, !=, NULL); ASSERT3P(vd->vdev_ops, !=, &vdev_raidz_ops); if (vd->vdev_leaf_zap != 0) { char zkey[32]; (void) snprintf(zkey, sizeof (zkey), "%s-%ju", VDEV_REMOVAL_ZAP_OBJS, (uintmax_t)vd->vdev_leaf_zap); fnvlist_add_uint64(zlist, zkey, vd->vdev_leaf_zap); } for (uint64_t id = 0; id < vd->vdev_children; id++) { vdev_remove_enlist_zaps(vd->vdev_child[id], zlist); } } static void vdev_remove_replace_with_indirect(vdev_t *vd, uint64_t txg) { vdev_t *ivd; dmu_tx_t *tx; spa_t *spa = vd->vdev_spa; spa_vdev_removal_t *svr = spa->spa_vdev_removal; /* * First, build a list of leaf zaps to be destroyed. * This is passed to the sync context thread, * which does the actual unlinking. */ svr->svr_zaplist = fnvlist_alloc(); vdev_remove_enlist_zaps(vd, svr->svr_zaplist); ivd = vdev_add_parent(vd, &vdev_indirect_ops); vd->vdev_leaf_zap = 0; vdev_remove_child(ivd, vd); vdev_compact_children(ivd); vdev_indirect_state_transfer(ivd, vd); svr->svr_vdev = ivd; ASSERT(!ivd->vdev_removing); ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); dsl_sync_task_nowait(spa->spa_dsl_pool, vdev_remove_complete_sync, svr, 0, ZFS_SPACE_CHECK_NONE, tx); dmu_tx_commit(tx); /* * Indicate that this thread has exited. * After this, we can not use svr. */ mutex_enter(&svr->svr_lock); svr->svr_thread = NULL; cv_broadcast(&svr->svr_cv); mutex_exit(&svr->svr_lock); } /* * Complete the removal of a toplevel vdev. This is called in open * context by the removal thread after we have copied all vdev's data. */ static void vdev_remove_complete(vdev_t *vd) { spa_t *spa = vd->vdev_spa; uint64_t txg; /* * Wait for any deferred frees to be synced before we call * vdev_metaslab_fini() */ txg_wait_synced(spa->spa_dsl_pool, 0); txg = spa_vdev_enter(spa); zfs_dbgmsg("finishing device removal for vdev %llu in txg %llu", vd->vdev_id, txg); /* * Discard allocation state. */ if (vd->vdev_mg != NULL) { vdev_metaslab_fini(vd); metaslab_group_destroy(vd->vdev_mg); vd->vdev_mg = NULL; } ASSERT0(vd->vdev_stat.vs_space); ASSERT0(vd->vdev_stat.vs_dspace); vdev_remove_replace_with_indirect(vd, txg); /* * We now release the locks, allowing spa_sync to run and finish the * removal via vdev_remove_complete_sync in syncing context. */ (void) spa_vdev_exit(spa, NULL, txg, 0); /* * Top ZAP should have been transferred to the indirect vdev in * vdev_remove_replace_with_indirect. */ ASSERT0(vd->vdev_top_zap); /* * Leaf ZAP should have been moved in vdev_remove_replace_with_indirect. */ ASSERT0(vd->vdev_leaf_zap); txg = spa_vdev_enter(spa); (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); /* * Request to update the config and the config cachefile. */ vdev_config_dirty(spa->spa_root_vdev); (void) spa_vdev_exit(spa, vd, txg, 0); } /* * Evacuates a segment of size at most max_alloc from the vdev * via repeated calls to spa_vdev_copy_segment. If an allocation * fails, the pool is probably too fragmented to handle such a * large size, so decrease max_alloc so that the caller will not try * this size again this txg. */ static void spa_vdev_copy_impl(spa_vdev_removal_t *svr, vdev_copy_arg_t *vca, uint64_t *max_alloc, dmu_tx_t *tx) { uint64_t txg = dmu_tx_get_txg(tx); spa_t *spa = dmu_tx_pool(tx)->dp_spa; mutex_enter(&svr->svr_lock); range_seg_t *rs = avl_first(&svr->svr_allocd_segs->rt_root); if (rs == NULL) { mutex_exit(&svr->svr_lock); return; } uint64_t offset = rs->rs_start; uint64_t length = MIN(rs->rs_end - rs->rs_start, *max_alloc); range_tree_remove(svr->svr_allocd_segs, offset, length); if (svr->svr_max_offset_to_sync[txg & TXG_MASK] == 0) { dsl_sync_task_nowait(dmu_tx_pool(tx), vdev_mapping_sync, svr, 0, ZFS_SPACE_CHECK_NONE, tx); } svr->svr_max_offset_to_sync[txg & TXG_MASK] = offset + length; /* * Note: this is the amount of *allocated* space * that we are taking care of each txg. */ svr->svr_bytes_done[txg & TXG_MASK] += length; mutex_exit(&svr->svr_lock); zio_alloc_list_t zal; metaslab_trace_init(&zal); uint64_t thismax = *max_alloc; while (length > 0) { uint64_t mylen = MIN(length, thismax); int error = spa_vdev_copy_segment(svr->svr_vdev, offset, mylen, txg, vca, &zal); if (error == ENOSPC) { /* * Cut our segment in half, and don't try this * segment size again this txg. Note that the * allocation size must be aligned to the highest * ashift in the pool, so that the allocation will * not be padded out to a multiple of the ashift, * which could cause us to think that this mapping * is larger than we intended. */ ASSERT3U(spa->spa_max_ashift, >=, SPA_MINBLOCKSHIFT); ASSERT3U(spa->spa_max_ashift, ==, spa->spa_min_ashift); thismax = P2ROUNDUP(mylen / 2, 1 << spa->spa_max_ashift); ASSERT3U(thismax, <, mylen); /* * The minimum-size allocation can not fail. */ ASSERT3U(mylen, >, 1 << spa->spa_max_ashift); *max_alloc = mylen - (1 << spa->spa_max_ashift); } else { ASSERT0(error); length -= mylen; offset += mylen; /* * We've performed an allocation, so reset the * alloc trace list. */ metaslab_trace_fini(&zal); metaslab_trace_init(&zal); } } metaslab_trace_fini(&zal); } /* * The removal thread operates in open context. It iterates over all * allocated space in the vdev, by loading each metaslab's spacemap. * For each contiguous segment of allocated space (capping the segment * size at SPA_MAXBLOCKSIZE), we: * - Allocate space for it on another vdev. * - Create a new mapping from the old location to the new location * (as a record in svr_new_segments). * - Initiate a logical read zio to get the data off the removing disk. * - In the read zio's done callback, initiate a logical write zio to * write it to the new vdev. * Note that all of this will take effect when a particular TXG syncs. * The sync thread ensures that all the phys reads and writes for the syncing * TXG have completed (see spa_txg_zio) and writes the new mappings to disk * (see vdev_mapping_sync()). */ static void spa_vdev_remove_thread(void *arg) { vdev_t *vd = arg; spa_t *spa = vd->vdev_spa; spa_vdev_removal_t *svr = spa->spa_vdev_removal; vdev_copy_arg_t vca; uint64_t max_alloc = zfs_remove_max_segment; uint64_t last_txg = 0; vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; uint64_t start_offset = vdev_indirect_mapping_max_offset(vim); ASSERT3P(vd->vdev_ops, !=, &vdev_indirect_ops); ASSERT(vdev_is_concrete(vd)); ASSERT(vd->vdev_removing); ASSERT(vd->vdev_indirect_config.vic_mapping_object != 0); ASSERT3P(svr->svr_vdev, ==, vd); ASSERT(vim != NULL); mutex_init(&vca.vca_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&vca.vca_cv, NULL, CV_DEFAULT, NULL); vca.vca_outstanding_bytes = 0; mutex_enter(&svr->svr_lock); /* * Start from vim_max_offset so we pick up where we left off * if we are restarting the removal after opening the pool. */ uint64_t msi; for (msi = start_offset >> vd->vdev_ms_shift; msi < vd->vdev_ms_count && !svr->svr_thread_exit; msi++) { metaslab_t *msp = vd->vdev_ms[msi]; ASSERT3U(msi, <=, vd->vdev_ms_count); ASSERT0(range_tree_space(svr->svr_allocd_segs)); mutex_enter(&msp->ms_sync_lock); mutex_enter(&msp->ms_lock); /* * Assert nothing in flight -- ms_*tree is empty. */ for (int i = 0; i < TXG_SIZE; i++) { ASSERT0(range_tree_space(msp->ms_allocating[i])); } /* * If the metaslab has ever been allocated from (ms_sm!=NULL), * read the allocated segments from the space map object * into svr_allocd_segs. Since we do this while holding * svr_lock and ms_sync_lock, concurrent frees (which * would have modified the space map) will wait for us * to finish loading the spacemap, and then take the * appropriate action (see free_from_removing_vdev()). */ if (msp->ms_sm != NULL) { space_map_t *sm = NULL; /* * We have to open a new space map here, because * ms_sm's sm_length and sm_alloc may not reflect * what's in the object contents, if we are in between * metaslab_sync() and metaslab_sync_done(). */ VERIFY0(space_map_open(&sm, spa->spa_dsl_pool->dp_meta_objset, msp->ms_sm->sm_object, msp->ms_sm->sm_start, msp->ms_sm->sm_size, msp->ms_sm->sm_shift)); space_map_update(sm); VERIFY0(space_map_load(sm, svr->svr_allocd_segs, SM_ALLOC)); space_map_close(sm); range_tree_walk(msp->ms_freeing, range_tree_remove, svr->svr_allocd_segs); /* * When we are resuming from a paused removal (i.e. * when importing a pool with a removal in progress), * discard any state that we have already processed. */ range_tree_clear(svr->svr_allocd_segs, 0, start_offset); } mutex_exit(&msp->ms_lock); mutex_exit(&msp->ms_sync_lock); vca.vca_msp = msp; zfs_dbgmsg("copying %llu segments for metaslab %llu", avl_numnodes(&svr->svr_allocd_segs->rt_root), msp->ms_id); while (!svr->svr_thread_exit && !range_tree_is_empty(svr->svr_allocd_segs)) { mutex_exit(&svr->svr_lock); /* * This delay will pause the removal around the point * specified by zfs_remove_max_bytes_pause. We do this * solely from the test suite or during debugging. */ uint64_t bytes_copied = spa->spa_removing_phys.sr_copied; for (int i = 0; i < TXG_SIZE; i++) bytes_copied += svr->svr_bytes_done[i]; while (zfs_remove_max_bytes_pause <= bytes_copied && !svr->svr_thread_exit) delay(hz); mutex_enter(&vca.vca_lock); while (vca.vca_outstanding_bytes > zfs_remove_max_copy_bytes) { cv_wait(&vca.vca_cv, &vca.vca_lock); } mutex_exit(&vca.vca_lock); dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); uint64_t txg = dmu_tx_get_txg(tx); if (txg != last_txg) max_alloc = zfs_remove_max_segment; last_txg = txg; spa_vdev_copy_impl(svr, &vca, &max_alloc, tx); dmu_tx_commit(tx); mutex_enter(&svr->svr_lock); } } mutex_exit(&svr->svr_lock); /* * Wait for all copies to finish before cleaning up the vca. */ txg_wait_synced(spa->spa_dsl_pool, 0); ASSERT0(vca.vca_outstanding_bytes); mutex_destroy(&vca.vca_lock); cv_destroy(&vca.vca_cv); if (svr->svr_thread_exit) { mutex_enter(&svr->svr_lock); range_tree_vacate(svr->svr_allocd_segs, NULL, NULL); svr->svr_thread = NULL; cv_broadcast(&svr->svr_cv); mutex_exit(&svr->svr_lock); } else { ASSERT0(range_tree_space(svr->svr_allocd_segs)); vdev_remove_complete(vd); } thread_exit(); } void spa_vdev_remove_suspend(spa_t *spa) { spa_vdev_removal_t *svr = spa->spa_vdev_removal; if (svr == NULL) return; mutex_enter(&svr->svr_lock); svr->svr_thread_exit = B_TRUE; while (svr->svr_thread != NULL) cv_wait(&svr->svr_cv, &svr->svr_lock); svr->svr_thread_exit = B_FALSE; mutex_exit(&svr->svr_lock); } /* ARGSUSED */ static int spa_vdev_remove_cancel_check(void *arg, dmu_tx_t *tx) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; if (spa->spa_vdev_removal == NULL) return (ESRCH); return (0); } /* * Cancel a removal by freeing all entries from the partial mapping * and marking the vdev as no longer being removing. */ /* ARGSUSED */ static void spa_vdev_remove_cancel_sync(void *arg, dmu_tx_t *tx) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; spa_vdev_removal_t *svr = spa->spa_vdev_removal; vdev_t *vd = svr->svr_vdev; vdev_indirect_config_t *vic = &vd->vdev_indirect_config; vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; objset_t *mos = spa->spa_meta_objset; ASSERT3P(svr->svr_thread, ==, NULL); spa_feature_decr(spa, SPA_FEATURE_DEVICE_REMOVAL, tx); if (vdev_obsolete_counts_are_precise(vd)) { spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); VERIFY0(zap_remove(spa->spa_meta_objset, vd->vdev_top_zap, VDEV_TOP_ZAP_OBSOLETE_COUNTS_ARE_PRECISE, tx)); } if (vdev_obsolete_sm_object(vd) != 0) { ASSERT(vd->vdev_obsolete_sm != NULL); ASSERT3U(vdev_obsolete_sm_object(vd), ==, space_map_object(vd->vdev_obsolete_sm)); space_map_free(vd->vdev_obsolete_sm, tx); VERIFY0(zap_remove(spa->spa_meta_objset, vd->vdev_top_zap, VDEV_TOP_ZAP_INDIRECT_OBSOLETE_SM, tx)); space_map_close(vd->vdev_obsolete_sm); vd->vdev_obsolete_sm = NULL; spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); } for (int i = 0; i < TXG_SIZE; i++) { ASSERT(list_is_empty(&svr->svr_new_segments[i])); ASSERT3U(svr->svr_max_offset_to_sync[i], <=, vdev_indirect_mapping_max_offset(vim)); } for (uint64_t msi = 0; msi < vd->vdev_ms_count; msi++) { metaslab_t *msp = vd->vdev_ms[msi]; if (msp->ms_start >= vdev_indirect_mapping_max_offset(vim)) break; ASSERT0(range_tree_space(svr->svr_allocd_segs)); mutex_enter(&msp->ms_lock); /* * Assert nothing in flight -- ms_*tree is empty. */ for (int i = 0; i < TXG_SIZE; i++) ASSERT0(range_tree_space(msp->ms_allocating[i])); for (int i = 0; i < TXG_DEFER_SIZE; i++) ASSERT0(range_tree_space(msp->ms_defer[i])); ASSERT0(range_tree_space(msp->ms_freed)); if (msp->ms_sm != NULL) { /* * Assert that the in-core spacemap has the same * length as the on-disk one, so we can use the * existing in-core spacemap to load it from disk. */ ASSERT3U(msp->ms_sm->sm_alloc, ==, msp->ms_sm->sm_phys->smp_alloc); ASSERT3U(msp->ms_sm->sm_length, ==, msp->ms_sm->sm_phys->smp_objsize); mutex_enter(&svr->svr_lock); VERIFY0(space_map_load(msp->ms_sm, svr->svr_allocd_segs, SM_ALLOC)); range_tree_walk(msp->ms_freeing, range_tree_remove, svr->svr_allocd_segs); /* * Clear everything past what has been synced, * because we have not allocated mappings for it yet. */ uint64_t syncd = vdev_indirect_mapping_max_offset(vim); range_tree_clear(svr->svr_allocd_segs, syncd, msp->ms_sm->sm_start + msp->ms_sm->sm_size - syncd); mutex_exit(&svr->svr_lock); } mutex_exit(&msp->ms_lock); mutex_enter(&svr->svr_lock); range_tree_vacate(svr->svr_allocd_segs, free_mapped_segment_cb, vd); mutex_exit(&svr->svr_lock); } /* * Note: this must happen after we invoke free_mapped_segment_cb, * because it adds to the obsolete_segments. */ range_tree_vacate(vd->vdev_obsolete_segments, NULL, NULL); ASSERT3U(vic->vic_mapping_object, ==, vdev_indirect_mapping_object(vd->vdev_indirect_mapping)); vdev_indirect_mapping_close(vd->vdev_indirect_mapping); vd->vdev_indirect_mapping = NULL; vdev_indirect_mapping_free(mos, vic->vic_mapping_object, tx); vic->vic_mapping_object = 0; ASSERT3U(vic->vic_births_object, ==, vdev_indirect_births_object(vd->vdev_indirect_births)); vdev_indirect_births_close(vd->vdev_indirect_births); vd->vdev_indirect_births = NULL; vdev_indirect_births_free(mos, vic->vic_births_object, tx); vic->vic_births_object = 0; /* * We may have processed some frees from the removing vdev in this * txg, thus increasing svr_bytes_done; discard that here to * satisfy the assertions in spa_vdev_removal_destroy(). * Note that future txg's can not have any bytes_done, because * future TXG's are only modified from open context, and we have * already shut down the copying thread. */ svr->svr_bytes_done[dmu_tx_get_txg(tx) & TXG_MASK] = 0; spa_finish_removal(spa, DSS_CANCELED, tx); vd->vdev_removing = B_FALSE; vdev_config_dirty(vd); zfs_dbgmsg("canceled device removal for vdev %llu in %llu", vd->vdev_id, dmu_tx_get_txg(tx)); spa_history_log_internal(spa, "vdev remove canceled", tx, "%s vdev %llu %s", spa_name(spa), vd->vdev_id, (vd->vdev_path != NULL) ? vd->vdev_path : "-"); } int spa_vdev_remove_cancel(spa_t *spa) { spa_vdev_remove_suspend(spa); if (spa->spa_vdev_removal == NULL) return (ESRCH); uint64_t vdid = spa->spa_vdev_removal->svr_vdev->vdev_id; int error = dsl_sync_task(spa->spa_name, spa_vdev_remove_cancel_check, spa_vdev_remove_cancel_sync, NULL, 0, ZFS_SPACE_CHECK_EXTRA_RESERVED); if (error == 0) { spa_config_enter(spa, SCL_ALLOC | SCL_VDEV, FTAG, RW_WRITER); vdev_t *vd = vdev_lookup_top(spa, vdid); metaslab_group_activate(vd->vdev_mg); spa_config_exit(spa, SCL_ALLOC | SCL_VDEV, FTAG); } return (error); } /* * Called every sync pass of every txg if there's a svr. */ void svr_sync(spa_t *spa, dmu_tx_t *tx) { spa_vdev_removal_t *svr = spa->spa_vdev_removal; int txgoff = dmu_tx_get_txg(tx) & TXG_MASK; /* * This check is necessary so that we do not dirty the * DIRECTORY_OBJECT via spa_sync_removing_state() when there * is nothing to do. Dirtying it every time would prevent us * from syncing-to-convergence. */ if (svr->svr_bytes_done[txgoff] == 0) return; /* * Update progress accounting. */ spa->spa_removing_phys.sr_copied += svr->svr_bytes_done[txgoff]; svr->svr_bytes_done[txgoff] = 0; spa_sync_removing_state(spa, tx); } static void vdev_remove_make_hole_and_free(vdev_t *vd) { uint64_t id = vd->vdev_id; spa_t *spa = vd->vdev_spa; vdev_t *rvd = spa->spa_root_vdev; boolean_t last_vdev = (id == (rvd->vdev_children - 1)); ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); vdev_free(vd); if (last_vdev) { vdev_compact_children(rvd); } else { vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); vdev_add_child(rvd, vd); } vdev_config_dirty(rvd); /* * Reassess the health of our root vdev. */ vdev_reopen(rvd); } /* * Remove a log device. The config lock is held for the specified TXG. */ static int spa_vdev_remove_log(vdev_t *vd, uint64_t *txg) { metaslab_group_t *mg = vd->vdev_mg; spa_t *spa = vd->vdev_spa; int error = 0; ASSERT(vd->vdev_islog); ASSERT(vd == vd->vdev_top); /* * Stop allocating from this vdev. */ metaslab_group_passivate(mg); /* * Wait for the youngest allocations and frees to sync, * and then wait for the deferral of those frees to finish. */ spa_vdev_config_exit(spa, NULL, *txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); /* * Evacuate the device. We don't hold the config lock as writer * since we need to do I/O but we do keep the * spa_namespace_lock held. Once this completes the device * should no longer have any blocks allocated on it. */ if (vd->vdev_islog) { if (vd->vdev_stat.vs_alloc != 0) error = spa_reset_logs(spa); } *txg = spa_vdev_config_enter(spa); if (error != 0) { metaslab_group_activate(mg); return (error); } ASSERT0(vd->vdev_stat.vs_alloc); /* * The evacuation succeeded. Remove any remaining MOS metadata * associated with this vdev, and wait for these changes to sync. */ vd->vdev_removing = B_TRUE; vdev_dirty_leaves(vd, VDD_DTL, *txg); vdev_config_dirty(vd); spa_history_log_internal(spa, "vdev remove", NULL, "%s vdev %llu (log) %s", spa_name(spa), vd->vdev_id, (vd->vdev_path != NULL) ? vd->vdev_path : "-"); /* Make sure these changes are sync'ed */ spa_vdev_config_exit(spa, NULL, *txg, 0, FTAG); *txg = spa_vdev_config_enter(spa); sysevent_t *ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_DEV); ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); /* The top ZAP should have been destroyed by vdev_remove_empty. */ ASSERT0(vd->vdev_top_zap); /* The leaf ZAP should have been destroyed by vdev_dtl_sync. */ ASSERT0(vd->vdev_leaf_zap); (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); if (list_link_active(&vd->vdev_state_dirty_node)) vdev_state_clean(vd); if (list_link_active(&vd->vdev_config_dirty_node)) vdev_config_clean(vd); /* * Clean up the vdev namespace. */ vdev_remove_make_hole_and_free(vd); if (ev != NULL) spa_event_post(ev); return (0); } static int spa_vdev_remove_top_check(vdev_t *vd) { spa_t *spa = vd->vdev_spa; if (vd != vd->vdev_top) return (SET_ERROR(ENOTSUP)); if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REMOVAL)) return (SET_ERROR(ENOTSUP)); /* * There has to be enough free space to remove the * device and leave double the "slop" space (i.e. we * must leave at least 3% of the pool free, in addition to * the normal slop space). */ if (dsl_dir_space_available(spa->spa_dsl_pool->dp_root_dir, NULL, 0, B_TRUE) < vd->vdev_stat.vs_dspace + spa_get_slop_space(spa)) { return (SET_ERROR(ENOSPC)); } /* * There can not be a removal in progress. */ if (spa->spa_removing_phys.sr_state == DSS_SCANNING) return (SET_ERROR(EBUSY)); /* * The device must have all its data. */ if (!vdev_dtl_empty(vd, DTL_MISSING) || !vdev_dtl_empty(vd, DTL_OUTAGE)) return (SET_ERROR(EBUSY)); /* * The device must be healthy. */ if (!vdev_readable(vd)) return (SET_ERROR(EIO)); /* * All vdevs in normal class must have the same ashift. */ if (spa->spa_max_ashift != spa->spa_min_ashift) { return (SET_ERROR(EINVAL)); } /* * All vdevs in normal class must have the same ashift * and not be raidz. */ vdev_t *rvd = spa->spa_root_vdev; int num_indirect = 0; for (uint64_t id = 0; id < rvd->vdev_children; id++) { vdev_t *cvd = rvd->vdev_child[id]; if (cvd->vdev_ashift != 0 && !cvd->vdev_islog) ASSERT3U(cvd->vdev_ashift, ==, spa->spa_max_ashift); if (cvd->vdev_ops == &vdev_indirect_ops) num_indirect++; if (!vdev_is_concrete(cvd)) continue; if (cvd->vdev_ops == &vdev_raidz_ops) return (SET_ERROR(EINVAL)); /* * Need the mirror to be mirror of leaf vdevs only */ if (cvd->vdev_ops == &vdev_mirror_ops) { for (uint64_t cid = 0; cid < cvd->vdev_children; cid++) { vdev_t *tmp = cvd->vdev_child[cid]; if (!tmp->vdev_ops->vdev_op_leaf) return (SET_ERROR(EINVAL)); } } } return (0); } /* * Initiate removal of a top-level vdev, reducing the total space in the pool. * The config lock is held for the specified TXG. Once initiated, * evacuation of all allocated space (copying it to other vdevs) happens * in the background (see spa_vdev_remove_thread()), and can be canceled * (see spa_vdev_remove_cancel()). If successful, the vdev will * be transformed to an indirect vdev (see spa_vdev_remove_complete()). */ static int spa_vdev_remove_top(vdev_t *vd, uint64_t *txg) { spa_t *spa = vd->vdev_spa; int error; /* * Check for errors up-front, so that we don't waste time * passivating the metaslab group and clearing the ZIL if there * are errors. */ error = spa_vdev_remove_top_check(vd); if (error != 0) return (error); /* * Stop allocating from this vdev. Note that we must check * that this is not the only device in the pool before * passivating, otherwise we will not be able to make * progress because we can't allocate from any vdevs. * The above check for sufficient free space serves this * purpose. */ metaslab_group_t *mg = vd->vdev_mg; metaslab_group_passivate(mg); /* * Wait for the youngest allocations and frees to sync, * and then wait for the deferral of those frees to finish. */ spa_vdev_config_exit(spa, NULL, *txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); /* * We must ensure that no "stubby" log blocks are allocated * on the device to be removed. These blocks could be * written at any time, including while we are in the middle * of copying them. */ error = spa_reset_logs(spa); *txg = spa_vdev_config_enter(spa); /* * Things might have changed while the config lock was dropped * (e.g. space usage). Check for errors again. */ if (error == 0) error = spa_vdev_remove_top_check(vd); if (error != 0) { metaslab_group_activate(mg); return (error); } vd->vdev_removing = B_TRUE; vdev_dirty_leaves(vd, VDD_DTL, *txg); vdev_config_dirty(vd); dmu_tx_t *tx = dmu_tx_create_assigned(spa->spa_dsl_pool, *txg); dsl_sync_task_nowait(spa->spa_dsl_pool, vdev_remove_initiate_sync, vd, 0, ZFS_SPACE_CHECK_NONE, tx); dmu_tx_commit(tx); return (0); } /* * Remove a device from the pool. * * Removing a device from the vdev namespace requires several steps * and can take a significant amount of time. As a result we use * the spa_vdev_config_[enter/exit] functions which allow us to * grab and release the spa_config_lock while still holding the namespace * lock. During each step the configuration is synced out. */ int spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) { vdev_t *vd; nvlist_t **spares, **l2cache, *nv; uint64_t txg = 0; uint_t nspares, nl2cache; int error = 0; boolean_t locked = MUTEX_HELD(&spa_namespace_lock); sysevent_t *ev = NULL; ASSERT(spa_writeable(spa)); if (!locked) txg = spa_vdev_enter(spa); ASSERT(MUTEX_HELD(&spa_namespace_lock)); if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { error = (spa_has_checkpoint(spa)) ? ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; if (!locked) return (spa_vdev_exit(spa, NULL, txg, error)); return (error); } vd = spa_lookup_by_guid(spa, guid, B_FALSE); if (spa->spa_spares.sav_vdevs != NULL && nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { /* * Only remove the hot spare if it's not currently in use * in this pool. */ if (vd == NULL || unspare) { char *nvstr = fnvlist_lookup_string(nv, ZPOOL_CONFIG_PATH); spa_history_log_internal(spa, "vdev remove", NULL, "%s vdev (%s) %s", spa_name(spa), VDEV_TYPE_SPARE, nvstr); if (vd == NULL) vd = spa_lookup_by_guid(spa, guid, B_TRUE); ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_AUX); spa_vdev_remove_aux(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, spares, nspares, nv); spa_load_spares(spa); spa->spa_spares.sav_sync = B_TRUE; } else { error = SET_ERROR(EBUSY); } } else if (spa->spa_l2cache.sav_vdevs != NULL && nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { char *nvstr = fnvlist_lookup_string(nv, ZPOOL_CONFIG_PATH); spa_history_log_internal(spa, "vdev remove", NULL, "%s vdev (%s) %s", spa_name(spa), VDEV_TYPE_L2CACHE, nvstr); /* * Cache devices can always be removed. */ vd = spa_lookup_by_guid(spa, guid, B_TRUE); ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_AUX); spa_vdev_remove_aux(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); spa_load_l2cache(spa); spa->spa_l2cache.sav_sync = B_TRUE; } else if (vd != NULL && vd->vdev_islog) { ASSERT(!locked); error = spa_vdev_remove_log(vd, &txg); } else if (vd != NULL) { ASSERT(!locked); error = spa_vdev_remove_top(vd, &txg); } else { /* * There is no vdev of any kind with the specified guid. */ error = SET_ERROR(ENOENT); } if (!locked) error = spa_vdev_exit(spa, NULL, txg, error); if (ev != NULL) { if (error != 0) { spa_event_discard(ev); } else { spa_event_post(ev); } } return (error); } int spa_removal_get_stats(spa_t *spa, pool_removal_stat_t *prs) { prs->prs_state = spa->spa_removing_phys.sr_state; if (prs->prs_state == DSS_NONE) return (SET_ERROR(ENOENT)); prs->prs_removing_vdev = spa->spa_removing_phys.sr_removing_vdev; prs->prs_start_time = spa->spa_removing_phys.sr_start_time; prs->prs_end_time = spa->spa_removing_phys.sr_end_time; prs->prs_to_copy = spa->spa_removing_phys.sr_to_copy; prs->prs_copied = spa->spa_removing_phys.sr_copied; if (spa->spa_vdev_removal != NULL) { for (int i = 0; i < TXG_SIZE; i++) { prs->prs_copied += spa->spa_vdev_removal->svr_bytes_done[i]; } } prs->prs_mapping_memory = 0; uint64_t indirect_vdev_id = spa->spa_removing_phys.sr_prev_indirect_vdev; while (indirect_vdev_id != -1) { vdev_t *vd = spa->spa_root_vdev->vdev_child[indirect_vdev_id]; vdev_indirect_config_t *vic = &vd->vdev_indirect_config; vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops); prs->prs_mapping_memory += vdev_indirect_mapping_size(vim); indirect_vdev_id = vic->vic_prev_indirect_vdev; } return (0); } Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c (revision 336949) @@ -1,3332 +1,3334 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011, 2017 by Delphix. All rights reserved. + * Copyright (c) 2011, 2018 by Delphix. All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ /* Portions Copyright 2010 Robert Milkowski */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * The ZFS Intent Log (ZIL) saves "transaction records" (itxs) of system * calls that change the file system. Each itx has enough information to * be able to replay them after a system crash, power loss, or * equivalent failure mode. These are stored in memory until either: * * 1. they are committed to the pool by the DMU transaction group * (txg), at which point they can be discarded; or * 2. they are committed to the on-disk ZIL for the dataset being * modified (e.g. due to an fsync, O_DSYNC, or other synchronous * requirement). * * In the event of a crash or power loss, the itxs contained by each * dataset's on-disk ZIL will be replayed when that dataset is first * instantianted (e.g. if the dataset is a normal fileystem, when it is * first mounted). * * As hinted at above, there is one ZIL per dataset (both the in-memory * representation, and the on-disk representation). The on-disk format * consists of 3 parts: * * - a single, per-dataset, ZIL header; which points to a chain of * - zero or more ZIL blocks; each of which contains * - zero or more ZIL records * * A ZIL record holds the information necessary to replay a single * system call transaction. A ZIL block can hold many ZIL records, and * the blocks are chained together, similarly to a singly linked list. * * Each ZIL block contains a block pointer (blkptr_t) to the next ZIL * block in the chain, and the ZIL header points to the first block in * the chain. * * Note, there is not a fixed place in the pool to hold these ZIL * blocks; they are dynamically allocated and freed as needed from the * blocks available on the pool, though they can be preferentially * allocated from a dedicated "log" vdev. */ /* * This controls the amount of time that a ZIL block (lwb) will remain * "open" when it isn't "full", and it has a thread waiting for it to be * committed to stable storage. Please refer to the zil_commit_waiter() * function (and the comments within it) for more details. */ int zfs_commit_timeout_pct = 5; /* * Disable intent logging replay. This global ZIL switch affects all pools. */ int zil_replay_disable = 0; SYSCTL_DECL(_vfs_zfs); SYSCTL_INT(_vfs_zfs, OID_AUTO, zil_replay_disable, CTLFLAG_RWTUN, &zil_replay_disable, 0, "Disable intent logging replay"); /* * Tunable parameter for debugging or performance analysis. Setting * zfs_nocacheflush will cause corruption on power loss if a volatile * out-of-order write cache is enabled. */ boolean_t zfs_nocacheflush = B_FALSE; SYSCTL_INT(_vfs_zfs, OID_AUTO, cache_flush_disable, CTLFLAG_RDTUN, &zfs_nocacheflush, 0, "Disable cache flush"); boolean_t zfs_trim_enabled = B_TRUE; SYSCTL_DECL(_vfs_zfs_trim); SYSCTL_INT(_vfs_zfs_trim, OID_AUTO, enabled, CTLFLAG_RDTUN, &zfs_trim_enabled, 0, "Enable ZFS TRIM"); /* * Limit SLOG write size per commit executed with synchronous priority. * Any writes above that will be executed with lower (asynchronous) priority * to limit potential SLOG device abuse by single active ZIL writer. */ uint64_t zil_slog_bulk = 768 * 1024; SYSCTL_QUAD(_vfs_zfs, OID_AUTO, zil_slog_bulk, CTLFLAG_RWTUN, &zil_slog_bulk, 0, "Maximal SLOG commit size with sync priority"); static kmem_cache_t *zil_lwb_cache; static kmem_cache_t *zil_zcw_cache; #define LWB_EMPTY(lwb) ((BP_GET_LSIZE(&lwb->lwb_blk) - \ sizeof (zil_chain_t)) == (lwb->lwb_sz - lwb->lwb_nused)) static int zil_bp_compare(const void *x1, const void *x2) { const dva_t *dva1 = &((zil_bp_node_t *)x1)->zn_dva; const dva_t *dva2 = &((zil_bp_node_t *)x2)->zn_dva; if (DVA_GET_VDEV(dva1) < DVA_GET_VDEV(dva2)) return (-1); if (DVA_GET_VDEV(dva1) > DVA_GET_VDEV(dva2)) return (1); if (DVA_GET_OFFSET(dva1) < DVA_GET_OFFSET(dva2)) return (-1); if (DVA_GET_OFFSET(dva1) > DVA_GET_OFFSET(dva2)) return (1); return (0); } static void zil_bp_tree_init(zilog_t *zilog) { avl_create(&zilog->zl_bp_tree, zil_bp_compare, sizeof (zil_bp_node_t), offsetof(zil_bp_node_t, zn_node)); } static void zil_bp_tree_fini(zilog_t *zilog) { avl_tree_t *t = &zilog->zl_bp_tree; zil_bp_node_t *zn; void *cookie = NULL; while ((zn = avl_destroy_nodes(t, &cookie)) != NULL) kmem_free(zn, sizeof (zil_bp_node_t)); avl_destroy(t); } int zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp) { avl_tree_t *t = &zilog->zl_bp_tree; const dva_t *dva; zil_bp_node_t *zn; avl_index_t where; if (BP_IS_EMBEDDED(bp)) return (0); dva = BP_IDENTITY(bp); if (avl_find(t, dva, &where) != NULL) return (SET_ERROR(EEXIST)); zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP); zn->zn_dva = *dva; avl_insert(t, zn, where); return (0); } static zil_header_t * zil_header_in_syncing_context(zilog_t *zilog) { return ((zil_header_t *)zilog->zl_header); } static void zil_init_log_chain(zilog_t *zilog, blkptr_t *bp) { zio_cksum_t *zc = &bp->blk_cksum; zc->zc_word[ZIL_ZC_GUID_0] = spa_get_random(-1ULL); zc->zc_word[ZIL_ZC_GUID_1] = spa_get_random(-1ULL); zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os); zc->zc_word[ZIL_ZC_SEQ] = 1ULL; } /* * Read a log block and make sure it's valid. */ static int zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst, char **end) { enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; arc_flags_t aflags = ARC_FLAG_WAIT; arc_buf_t *abuf = NULL; zbookmark_phys_t zb; int error; if (zilog->zl_header->zh_claim_txg == 0) zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB; if (!(zilog->zl_header->zh_flags & ZIL_CLAIM_LR_SEQ_VALID)) zio_flags |= ZIO_FLAG_SPECULATIVE; SET_BOOKMARK(&zb, bp->blk_cksum.zc_word[ZIL_ZC_OBJSET], ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); if (error == 0) { zio_cksum_t cksum = bp->blk_cksum; /* * Validate the checksummed log block. * * Sequence numbers should be... sequential. The checksum * verifier for the next block should be bp's checksum plus 1. * * Also check the log chain linkage and size used. */ cksum.zc_word[ZIL_ZC_SEQ]++; if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) { zil_chain_t *zilc = abuf->b_data; char *lr = (char *)(zilc + 1); uint64_t len = zilc->zc_nused - sizeof (zil_chain_t); if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum, sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk)) { error = SET_ERROR(ECKSUM); } else { ASSERT3U(len, <=, SPA_OLD_MAXBLOCKSIZE); bcopy(lr, dst, len); *end = (char *)dst + len; *nbp = zilc->zc_next_blk; } } else { char *lr = abuf->b_data; uint64_t size = BP_GET_LSIZE(bp); zil_chain_t *zilc = (zil_chain_t *)(lr + size) - 1; if (bcmp(&cksum, &zilc->zc_next_blk.blk_cksum, sizeof (cksum)) || BP_IS_HOLE(&zilc->zc_next_blk) || (zilc->zc_nused > (size - sizeof (*zilc)))) { error = SET_ERROR(ECKSUM); } else { ASSERT3U(zilc->zc_nused, <=, SPA_OLD_MAXBLOCKSIZE); bcopy(lr, dst, zilc->zc_nused); *end = (char *)dst + zilc->zc_nused; *nbp = zilc->zc_next_blk; } } arc_buf_destroy(abuf, &abuf); } return (error); } /* * Read a TX_WRITE log data block. */ static int zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf) { enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; const blkptr_t *bp = &lr->lr_blkptr; arc_flags_t aflags = ARC_FLAG_WAIT; arc_buf_t *abuf = NULL; zbookmark_phys_t zb; int error; if (BP_IS_HOLE(bp)) { if (wbuf != NULL) bzero(wbuf, MAX(BP_GET_LSIZE(bp), lr->lr_length)); return (0); } if (zilog->zl_header->zh_claim_txg == 0) zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB; SET_BOOKMARK(&zb, dmu_objset_id(zilog->zl_os), lr->lr_foid, ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp)); error = arc_read(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); if (error == 0) { if (wbuf != NULL) bcopy(abuf->b_data, wbuf, arc_buf_size(abuf)); arc_buf_destroy(abuf, &abuf); } return (error); } /* * Parse the intent log, and call parse_func for each valid record within. */ int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg) { const zil_header_t *zh = zilog->zl_header; boolean_t claimed = !!zh->zh_claim_txg; uint64_t claim_blk_seq = claimed ? zh->zh_claim_blk_seq : UINT64_MAX; uint64_t claim_lr_seq = claimed ? zh->zh_claim_lr_seq : UINT64_MAX; uint64_t max_blk_seq = 0; uint64_t max_lr_seq = 0; uint64_t blk_count = 0; uint64_t lr_count = 0; blkptr_t blk, next_blk; char *lrbuf, *lrp; int error = 0; /* * Old logs didn't record the maximum zh_claim_lr_seq. */ if (!(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID)) claim_lr_seq = UINT64_MAX; /* * Starting at the block pointed to by zh_log we read the log chain. * For each block in the chain we strongly check that block to * ensure its validity. We stop when an invalid block is found. * For each block pointer in the chain we call parse_blk_func(). * For each record in each valid block we call parse_lr_func(). * If the log has been claimed, stop if we encounter a sequence * number greater than the highest claimed sequence number. */ lrbuf = zio_buf_alloc(SPA_OLD_MAXBLOCKSIZE); zil_bp_tree_init(zilog); for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) { uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ]; int reclen; char *end; if (blk_seq > claim_blk_seq) break; if ((error = parse_blk_func(zilog, &blk, arg, txg)) != 0) break; ASSERT3U(max_blk_seq, <, blk_seq); max_blk_seq = blk_seq; blk_count++; if (max_lr_seq == claim_lr_seq && max_blk_seq == claim_blk_seq) break; error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf, &end); if (error != 0) break; for (lrp = lrbuf; lrp < end; lrp += reclen) { lr_t *lr = (lr_t *)lrp; reclen = lr->lrc_reclen; ASSERT3U(reclen, >=, sizeof (lr_t)); if (lr->lrc_seq > claim_lr_seq) goto done; if ((error = parse_lr_func(zilog, lr, arg, txg)) != 0) goto done; ASSERT3U(max_lr_seq, <, lr->lrc_seq); max_lr_seq = lr->lrc_seq; lr_count++; } } done: zilog->zl_parse_error = error; zilog->zl_parse_blk_seq = max_blk_seq; zilog->zl_parse_lr_seq = max_lr_seq; zilog->zl_parse_blk_count = blk_count; zilog->zl_parse_lr_count = lr_count; ASSERT(!claimed || !(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID) || (max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq)); zil_bp_tree_fini(zilog); zio_buf_free(lrbuf, SPA_OLD_MAXBLOCKSIZE); return (error); } /* ARGSUSED */ static int zil_clear_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg) { ASSERT(!BP_IS_HOLE(bp)); /* * As we call this function from the context of a rewind to a * checkpoint, each ZIL block whose txg is later than the txg * that we rewind to is invalid. Thus, we return -1 so * zil_parse() doesn't attempt to read it. */ if (bp->blk_birth >= first_txg) return (-1); if (zil_bp_tree_add(zilog, bp) != 0) return (0); zio_free(zilog->zl_spa, first_txg, bp); return (0); } /* ARGSUSED */ static int zil_noop_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg) { return (0); } static int zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg) { /* * Claim log block if not already committed and not already claimed. * If tx == NULL, just verify that the block is claimable. */ if (BP_IS_HOLE(bp) || bp->blk_birth < first_txg || zil_bp_tree_add(zilog, bp) != 0) return (0); return (zio_wait(zio_claim(NULL, zilog->zl_spa, tx == NULL ? 0 : first_txg, bp, spa_claim_notify, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB))); } static int zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg) { lr_write_t *lr = (lr_write_t *)lrc; int error; if (lrc->lrc_txtype != TX_WRITE) return (0); /* * If the block is not readable, don't claim it. This can happen * in normal operation when a log block is written to disk before * some of the dmu_sync() blocks it points to. In this case, the * transaction cannot have been committed to anyone (we would have * waited for all writes to be stable first), so it is semantically * correct to declare this the end of the log. */ if (lr->lr_blkptr.blk_birth >= first_txg && (error = zil_read_log_data(zilog, lr, NULL)) != 0) return (error); return (zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg)); } /* ARGSUSED */ static int zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg) { zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp); return (0); } static int zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg) { lr_write_t *lr = (lr_write_t *)lrc; blkptr_t *bp = &lr->lr_blkptr; /* * If we previously claimed it, we need to free it. */ if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE && bp->blk_birth >= claim_txg && zil_bp_tree_add(zilog, bp) == 0 && !BP_IS_HOLE(bp)) zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp); return (0); } static int zil_lwb_vdev_compare(const void *x1, const void *x2) { const uint64_t v1 = ((zil_vdev_node_t *)x1)->zv_vdev; const uint64_t v2 = ((zil_vdev_node_t *)x2)->zv_vdev; if (v1 < v2) return (-1); if (v1 > v2) return (1); return (0); } static lwb_t * zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, boolean_t slog, uint64_t txg) { lwb_t *lwb; lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); lwb->lwb_zilog = zilog; lwb->lwb_blk = *bp; lwb->lwb_slog = slog; lwb->lwb_state = LWB_STATE_CLOSED; lwb->lwb_buf = zio_buf_alloc(BP_GET_LSIZE(bp)); lwb->lwb_max_txg = txg; lwb->lwb_write_zio = NULL; lwb->lwb_root_zio = NULL; lwb->lwb_tx = NULL; lwb->lwb_issued_timestamp = 0; if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) { lwb->lwb_nused = sizeof (zil_chain_t); lwb->lwb_sz = BP_GET_LSIZE(bp); } else { lwb->lwb_nused = 0; lwb->lwb_sz = BP_GET_LSIZE(bp) - sizeof (zil_chain_t); } mutex_enter(&zilog->zl_lock); list_insert_tail(&zilog->zl_lwb_list, lwb); mutex_exit(&zilog->zl_lock); ASSERT(!MUTEX_HELD(&lwb->lwb_vdev_lock)); ASSERT(avl_is_empty(&lwb->lwb_vdev_tree)); VERIFY(list_is_empty(&lwb->lwb_waiters)); return (lwb); } static void zil_free_lwb(zilog_t *zilog, lwb_t *lwb) { ASSERT(MUTEX_HELD(&zilog->zl_lock)); ASSERT(!MUTEX_HELD(&lwb->lwb_vdev_lock)); VERIFY(list_is_empty(&lwb->lwb_waiters)); ASSERT(avl_is_empty(&lwb->lwb_vdev_tree)); ASSERT3P(lwb->lwb_write_zio, ==, NULL); ASSERT3P(lwb->lwb_root_zio, ==, NULL); ASSERT3U(lwb->lwb_max_txg, <=, spa_syncing_txg(zilog->zl_spa)); ASSERT(lwb->lwb_state == LWB_STATE_CLOSED || lwb->lwb_state == LWB_STATE_DONE); /* * Clear the zilog's field to indicate this lwb is no longer * valid, and prevent use-after-free errors. */ if (zilog->zl_last_lwb_opened == lwb) zilog->zl_last_lwb_opened = NULL; kmem_cache_free(zil_lwb_cache, lwb); } /* * Called when we create in-memory log transactions so that we know * to cleanup the itxs at the end of spa_sync(). */ void zilog_dirty(zilog_t *zilog, uint64_t txg) { dsl_pool_t *dp = zilog->zl_dmu_pool; dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); ASSERT(spa_writeable(zilog->zl_spa)); if (ds->ds_is_snapshot) panic("dirtying snapshot!"); if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg)) { /* up the hold count until we can be written out */ dmu_buf_add_ref(ds->ds_dbuf, zilog); zilog->zl_dirty_max_txg = MAX(txg, zilog->zl_dirty_max_txg); } } /* * Determine if the zil is dirty in the specified txg. Callers wanting to * ensure that the dirty state does not change must hold the itxg_lock for * the specified txg. Holding the lock will ensure that the zil cannot be * dirtied (zil_itx_assign) or cleaned (zil_clean) while we check its current * state. */ boolean_t zilog_is_dirty_in_txg(zilog_t *zilog, uint64_t txg) { dsl_pool_t *dp = zilog->zl_dmu_pool; if (txg_list_member(&dp->dp_dirty_zilogs, zilog, txg & TXG_MASK)) return (B_TRUE); return (B_FALSE); } /* * Determine if the zil is dirty. The zil is considered dirty if it has * any pending itx records that have not been cleaned by zil_clean(). */ boolean_t zilog_is_dirty(zilog_t *zilog) { dsl_pool_t *dp = zilog->zl_dmu_pool; for (int t = 0; t < TXG_SIZE; t++) { if (txg_list_member(&dp->dp_dirty_zilogs, zilog, t)) return (B_TRUE); } return (B_FALSE); } /* * Create an on-disk intent log. */ static lwb_t * zil_create(zilog_t *zilog) { const zil_header_t *zh = zilog->zl_header; lwb_t *lwb = NULL; uint64_t txg = 0; dmu_tx_t *tx = NULL; blkptr_t blk; int error = 0; boolean_t slog = FALSE; /* * Wait for any previous destroy to complete. */ txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); ASSERT(zh->zh_claim_txg == 0); ASSERT(zh->zh_replay_seq == 0); blk = zh->zh_log; /* * Allocate an initial log block if: * - there isn't one already * - the existing block is the wrong endianess */ if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) { tx = dmu_tx_create(zilog->zl_os); VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); txg = dmu_tx_get_txg(tx); if (!BP_IS_HOLE(&blk)) { zio_free(zilog->zl_spa, txg, &blk); BP_ZERO(&blk); } - error = zio_alloc_zil(zilog->zl_spa, txg, &blk, NULL, + error = zio_alloc_zil(zilog->zl_spa, + zilog->zl_os->os_dsl_dataset->ds_object, txg, &blk, NULL, ZIL_MIN_BLKSZ, &slog); if (error == 0) zil_init_log_chain(zilog, &blk); } /* * Allocate a log write block (lwb) for the first log block. */ if (error == 0) lwb = zil_alloc_lwb(zilog, &blk, slog, txg); /* * If we just allocated the first log block, commit our transaction * and wait for zil_sync() to stuff the block poiner into zh_log. * (zh is part of the MOS, so we cannot modify it in open context.) */ if (tx != NULL) { dmu_tx_commit(tx); txg_wait_synced(zilog->zl_dmu_pool, txg); } ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0); return (lwb); } /* * In one tx, free all log blocks and clear the log header. If keep_first * is set, then we're replaying a log with no content. We want to keep the * first block, however, so that the first synchronous transaction doesn't * require a txg_wait_synced() in zil_create(). We don't need to * txg_wait_synced() here either when keep_first is set, because both * zil_create() and zil_destroy() will wait for any in-progress destroys * to complete. */ void zil_destroy(zilog_t *zilog, boolean_t keep_first) { const zil_header_t *zh = zilog->zl_header; lwb_t *lwb; dmu_tx_t *tx; uint64_t txg; /* * Wait for any previous destroy to complete. */ txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); zilog->zl_old_header = *zh; /* debugging aid */ if (BP_IS_HOLE(&zh->zh_log)) return; tx = dmu_tx_create(zilog->zl_os); VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); txg = dmu_tx_get_txg(tx); mutex_enter(&zilog->zl_lock); ASSERT3U(zilog->zl_destroy_txg, <, txg); zilog->zl_destroy_txg = txg; zilog->zl_keep_first = keep_first; if (!list_is_empty(&zilog->zl_lwb_list)) { ASSERT(zh->zh_claim_txg == 0); VERIFY(!keep_first); while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { list_remove(&zilog->zl_lwb_list, lwb); if (lwb->lwb_buf != NULL) zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); zio_free(zilog->zl_spa, txg, &lwb->lwb_blk); zil_free_lwb(zilog, lwb); } } else if (!keep_first) { zil_destroy_sync(zilog, tx); } mutex_exit(&zilog->zl_lock); dmu_tx_commit(tx); } void zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx) { ASSERT(list_is_empty(&zilog->zl_lwb_list)); (void) zil_parse(zilog, zil_free_log_block, zil_free_log_record, tx, zilog->zl_header->zh_claim_txg); } int zil_claim(dsl_pool_t *dp, dsl_dataset_t *ds, void *txarg) { dmu_tx_t *tx = txarg; zilog_t *zilog; uint64_t first_txg; zil_header_t *zh; objset_t *os; int error; error = dmu_objset_own_obj(dp, ds->ds_object, DMU_OST_ANY, B_FALSE, FTAG, &os); if (error != 0) { /* * EBUSY indicates that the objset is inconsistent, in which * case it can not have a ZIL. */ if (error != EBUSY) { cmn_err(CE_WARN, "can't open objset for %llu, error %u", (unsigned long long)ds->ds_object, error); } return (0); } zilog = dmu_objset_zil(os); zh = zil_header_in_syncing_context(zilog); ASSERT3U(tx->tx_txg, ==, spa_first_txg(zilog->zl_spa)); first_txg = spa_min_claim_txg(zilog->zl_spa); /* * If the spa_log_state is not set to be cleared, check whether * the current uberblock is a checkpoint one and if the current * header has been claimed before moving on. * * If the current uberblock is a checkpointed uberblock then * one of the following scenarios took place: * * 1] We are currently rewinding to the checkpoint of the pool. * 2] We crashed in the middle of a checkpoint rewind but we * did manage to write the checkpointed uberblock to the * vdev labels, so when we tried to import the pool again * the checkpointed uberblock was selected from the import * procedure. * * In both cases we want to zero out all the ZIL blocks, except * the ones that have been claimed at the time of the checkpoint * (their zh_claim_txg != 0). The reason is that these blocks * may be corrupted since we may have reused their locations on * disk after we took the checkpoint. * * We could try to set spa_log_state to SPA_LOG_CLEAR earlier * when we first figure out whether the current uberblock is * checkpointed or not. Unfortunately, that would discard all * the logs, including the ones that are claimed, and we would * leak space. */ if (spa_get_log_state(zilog->zl_spa) == SPA_LOG_CLEAR || (zilog->zl_spa->spa_uberblock.ub_checkpoint_txg != 0 && zh->zh_claim_txg == 0)) { if (!BP_IS_HOLE(&zh->zh_log)) { (void) zil_parse(zilog, zil_clear_log_block, zil_noop_log_record, tx, first_txg); } BP_ZERO(&zh->zh_log); dsl_dataset_dirty(dmu_objset_ds(os), tx); dmu_objset_disown(os, FTAG); return (0); } /* * If we are not rewinding and opening the pool normally, then * the min_claim_txg should be equal to the first txg of the pool. */ ASSERT3U(first_txg, ==, spa_first_txg(zilog->zl_spa)); /* * Claim all log blocks if we haven't already done so, and remember * the highest claimed sequence number. This ensures that if we can * read only part of the log now (e.g. due to a missing device), * but we can read the entire log later, we will not try to replay * or destroy beyond the last block we successfully claimed. */ ASSERT3U(zh->zh_claim_txg, <=, first_txg); if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) { (void) zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx, first_txg); zh->zh_claim_txg = first_txg; zh->zh_claim_blk_seq = zilog->zl_parse_blk_seq; zh->zh_claim_lr_seq = zilog->zl_parse_lr_seq; if (zilog->zl_parse_lr_count || zilog->zl_parse_blk_count > 1) zh->zh_flags |= ZIL_REPLAY_NEEDED; zh->zh_flags |= ZIL_CLAIM_LR_SEQ_VALID; dsl_dataset_dirty(dmu_objset_ds(os), tx); } ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1)); dmu_objset_disown(os, FTAG); return (0); } /* * Check the log by walking the log chain. * Checksum errors are ok as they indicate the end of the chain. * Any other error (no device or read failure) returns an error. */ /* ARGSUSED */ int zil_check_log_chain(dsl_pool_t *dp, dsl_dataset_t *ds, void *tx) { zilog_t *zilog; objset_t *os; blkptr_t *bp; int error; ASSERT(tx == NULL); error = dmu_objset_from_ds(ds, &os); if (error != 0) { cmn_err(CE_WARN, "can't open objset %llu, error %d", (unsigned long long)ds->ds_object, error); return (0); } zilog = dmu_objset_zil(os); bp = (blkptr_t *)&zilog->zl_header->zh_log; if (!BP_IS_HOLE(bp)) { vdev_t *vd; boolean_t valid = B_TRUE; /* * Check the first block and determine if it's on a log device * which may have been removed or faulted prior to loading this * pool. If so, there's no point in checking the rest of the * log as its content should have already been synced to the * pool. */ spa_config_enter(os->os_spa, SCL_STATE, FTAG, RW_READER); vd = vdev_lookup_top(os->os_spa, DVA_GET_VDEV(&bp->blk_dva[0])); if (vd->vdev_islog && vdev_is_dead(vd)) valid = vdev_log_state_valid(vd); spa_config_exit(os->os_spa, SCL_STATE, FTAG); if (!valid) return (0); /* * Check whether the current uberblock is checkpointed (e.g. * we are rewinding) and whether the current header has been * claimed or not. If it hasn't then skip verifying it. We * do this because its ZIL blocks may be part of the pool's * state before the rewind, which is no longer valid. */ zil_header_t *zh = zil_header_in_syncing_context(zilog); if (zilog->zl_spa->spa_uberblock.ub_checkpoint_txg != 0 && zh->zh_claim_txg == 0) return (0); } /* * Because tx == NULL, zil_claim_log_block() will not actually claim * any blocks, but just determine whether it is possible to do so. * In addition to checking the log chain, zil_claim_log_block() * will invoke zio_claim() with a done func of spa_claim_notify(), * which will update spa_max_claim_txg. See spa_load() for details. */ error = zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx, zilog->zl_header->zh_claim_txg ? -1ULL : spa_min_claim_txg(os->os_spa)); return ((error == ECKSUM || error == ENOENT) ? 0 : error); } /* * When an itx is "skipped", this function is used to properly mark the * waiter as "done, and signal any thread(s) waiting on it. An itx can * be skipped (and not committed to an lwb) for a variety of reasons, * one of them being that the itx was committed via spa_sync(), prior to * it being committed to an lwb; this can happen if a thread calling * zil_commit() is racing with spa_sync(). */ static void zil_commit_waiter_skip(zil_commit_waiter_t *zcw) { mutex_enter(&zcw->zcw_lock); ASSERT3B(zcw->zcw_done, ==, B_FALSE); zcw->zcw_done = B_TRUE; cv_broadcast(&zcw->zcw_cv); mutex_exit(&zcw->zcw_lock); } /* * This function is used when the given waiter is to be linked into an * lwb's "lwb_waiter" list; i.e. when the itx is committed to the lwb. * At this point, the waiter will no longer be referenced by the itx, * and instead, will be referenced by the lwb. */ static void zil_commit_waiter_link_lwb(zil_commit_waiter_t *zcw, lwb_t *lwb) { /* * The lwb_waiters field of the lwb is protected by the zilog's * zl_lock, thus it must be held when calling this function. */ ASSERT(MUTEX_HELD(&lwb->lwb_zilog->zl_lock)); mutex_enter(&zcw->zcw_lock); ASSERT(!list_link_active(&zcw->zcw_node)); ASSERT3P(zcw->zcw_lwb, ==, NULL); ASSERT3P(lwb, !=, NULL); ASSERT(lwb->lwb_state == LWB_STATE_OPENED || lwb->lwb_state == LWB_STATE_ISSUED); list_insert_tail(&lwb->lwb_waiters, zcw); zcw->zcw_lwb = lwb; mutex_exit(&zcw->zcw_lock); } /* * This function is used when zio_alloc_zil() fails to allocate a ZIL * block, and the given waiter must be linked to the "nolwb waiters" * list inside of zil_process_commit_list(). */ static void zil_commit_waiter_link_nolwb(zil_commit_waiter_t *zcw, list_t *nolwb) { mutex_enter(&zcw->zcw_lock); ASSERT(!list_link_active(&zcw->zcw_node)); ASSERT3P(zcw->zcw_lwb, ==, NULL); list_insert_tail(nolwb, zcw); mutex_exit(&zcw->zcw_lock); } void zil_lwb_add_block(lwb_t *lwb, const blkptr_t *bp) { avl_tree_t *t = &lwb->lwb_vdev_tree; avl_index_t where; zil_vdev_node_t *zv, zvsearch; int ndvas = BP_GET_NDVAS(bp); int i; if (zfs_nocacheflush) return; mutex_enter(&lwb->lwb_vdev_lock); for (i = 0; i < ndvas; i++) { zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]); if (avl_find(t, &zvsearch, &where) == NULL) { zv = kmem_alloc(sizeof (*zv), KM_SLEEP); zv->zv_vdev = zvsearch.zv_vdev; avl_insert(t, zv, where); } } mutex_exit(&lwb->lwb_vdev_lock); } void zil_lwb_add_txg(lwb_t *lwb, uint64_t txg) { lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg); } /* * This function is a called after all VDEVs associated with a given lwb * write have completed their DKIOCFLUSHWRITECACHE command; or as soon * as the lwb write completes, if "zfs_nocacheflush" is set. * * The intention is for this function to be called as soon as the * contents of an lwb are considered "stable" on disk, and will survive * any sudden loss of power. At this point, any threads waiting for the * lwb to reach this state are signalled, and the "waiter" structures * are marked "done". */ static void zil_lwb_flush_vdevs_done(zio_t *zio) { lwb_t *lwb = zio->io_private; zilog_t *zilog = lwb->lwb_zilog; dmu_tx_t *tx = lwb->lwb_tx; zil_commit_waiter_t *zcw; spa_config_exit(zilog->zl_spa, SCL_STATE, lwb); zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); mutex_enter(&zilog->zl_lock); /* * Ensure the lwb buffer pointer is cleared before releasing the * txg. If we have had an allocation failure and the txg is * waiting to sync then we want zil_sync() to remove the lwb so * that it's not picked up as the next new one in * zil_process_commit_list(). zil_sync() will only remove the * lwb if lwb_buf is null. */ lwb->lwb_buf = NULL; lwb->lwb_tx = NULL; ASSERT3U(lwb->lwb_issued_timestamp, >, 0); zilog->zl_last_lwb_latency = gethrtime() - lwb->lwb_issued_timestamp; lwb->lwb_root_zio = NULL; lwb->lwb_state = LWB_STATE_DONE; if (zilog->zl_last_lwb_opened == lwb) { /* * Remember the highest committed log sequence number * for ztest. We only update this value when all the log * writes succeeded, because ztest wants to ASSERT that * it got the whole log chain. */ zilog->zl_commit_lr_seq = zilog->zl_lr_seq; } while ((zcw = list_head(&lwb->lwb_waiters)) != NULL) { mutex_enter(&zcw->zcw_lock); ASSERT(list_link_active(&zcw->zcw_node)); list_remove(&lwb->lwb_waiters, zcw); ASSERT3P(zcw->zcw_lwb, ==, lwb); zcw->zcw_lwb = NULL; zcw->zcw_zio_error = zio->io_error; ASSERT3B(zcw->zcw_done, ==, B_FALSE); zcw->zcw_done = B_TRUE; cv_broadcast(&zcw->zcw_cv); mutex_exit(&zcw->zcw_lock); } mutex_exit(&zilog->zl_lock); /* * Now that we've written this log block, we have a stable pointer * to the next block in the chain, so it's OK to let the txg in * which we allocated the next block sync. */ dmu_tx_commit(tx); } /* * This is called when an lwb write completes. This means, this specific * lwb was written to disk, and all dependent lwb have also been * written to disk. * * At this point, a DKIOCFLUSHWRITECACHE command hasn't been issued to * the VDEVs involved in writing out this specific lwb. The lwb will be * "done" once zil_lwb_flush_vdevs_done() is called, which occurs in the * zio completion callback for the lwb's root zio. */ static void zil_lwb_write_done(zio_t *zio) { lwb_t *lwb = zio->io_private; spa_t *spa = zio->io_spa; zilog_t *zilog = lwb->lwb_zilog; avl_tree_t *t = &lwb->lwb_vdev_tree; void *cookie = NULL; zil_vdev_node_t *zv; ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), !=, 0); ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG); ASSERT(BP_GET_LEVEL(zio->io_bp) == 0); ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER); ASSERT(!BP_IS_GANG(zio->io_bp)); ASSERT(!BP_IS_HOLE(zio->io_bp)); ASSERT(BP_GET_FILL(zio->io_bp) == 0); abd_put(zio->io_abd); ASSERT3S(lwb->lwb_state, ==, LWB_STATE_ISSUED); mutex_enter(&zilog->zl_lock); lwb->lwb_write_zio = NULL; mutex_exit(&zilog->zl_lock); if (avl_numnodes(t) == 0) return; /* * If there was an IO error, we're not going to call zio_flush() * on these vdevs, so we simply empty the tree and free the * nodes. We avoid calling zio_flush() since there isn't any * good reason for doing so, after the lwb block failed to be * written out. */ if (zio->io_error != 0) { while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) kmem_free(zv, sizeof (*zv)); return; } while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) { vdev_t *vd = vdev_lookup_top(spa, zv->zv_vdev); if (vd != NULL) zio_flush(lwb->lwb_root_zio, vd); kmem_free(zv, sizeof (*zv)); } } /* * This function's purpose is to "open" an lwb such that it is ready to * accept new itxs being committed to it. To do this, the lwb's zio * structures are created, and linked to the lwb. This function is * idempotent; if the passed in lwb has already been opened, this * function is essentially a no-op. */ static void zil_lwb_write_open(zilog_t *zilog, lwb_t *lwb) { zbookmark_phys_t zb; zio_priority_t prio; ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock)); ASSERT3P(lwb, !=, NULL); EQUIV(lwb->lwb_root_zio == NULL, lwb->lwb_state == LWB_STATE_CLOSED); EQUIV(lwb->lwb_root_zio != NULL, lwb->lwb_state == LWB_STATE_OPENED); SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET], ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]); if (lwb->lwb_root_zio == NULL) { abd_t *lwb_abd = abd_get_from_buf(lwb->lwb_buf, BP_GET_LSIZE(&lwb->lwb_blk)); if (!lwb->lwb_slog || zilog->zl_cur_used <= zil_slog_bulk) prio = ZIO_PRIORITY_SYNC_WRITE; else prio = ZIO_PRIORITY_ASYNC_WRITE; lwb->lwb_root_zio = zio_root(zilog->zl_spa, zil_lwb_flush_vdevs_done, lwb, ZIO_FLAG_CANFAIL); ASSERT3P(lwb->lwb_root_zio, !=, NULL); lwb->lwb_write_zio = zio_rewrite(lwb->lwb_root_zio, zilog->zl_spa, 0, &lwb->lwb_blk, lwb_abd, BP_GET_LSIZE(&lwb->lwb_blk), zil_lwb_write_done, lwb, prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE, &zb); ASSERT3P(lwb->lwb_write_zio, !=, NULL); lwb->lwb_state = LWB_STATE_OPENED; mutex_enter(&zilog->zl_lock); /* * The zilog's "zl_last_lwb_opened" field is used to * build the lwb/zio dependency chain, which is used to * preserve the ordering of lwb completions that is * required by the semantics of the ZIL. Each new lwb * zio becomes a parent of the "previous" lwb zio, such * that the new lwb's zio cannot complete until the * "previous" lwb's zio completes. * * This is required by the semantics of zil_commit(); * the commit waiters attached to the lwbs will be woken * in the lwb zio's completion callback, so this zio * dependency graph ensures the waiters are woken in the * correct order (the same order the lwbs were created). */ lwb_t *last_lwb_opened = zilog->zl_last_lwb_opened; if (last_lwb_opened != NULL && last_lwb_opened->lwb_state != LWB_STATE_DONE) { ASSERT(last_lwb_opened->lwb_state == LWB_STATE_OPENED || last_lwb_opened->lwb_state == LWB_STATE_ISSUED); ASSERT3P(last_lwb_opened->lwb_root_zio, !=, NULL); zio_add_child(lwb->lwb_root_zio, last_lwb_opened->lwb_root_zio); } zilog->zl_last_lwb_opened = lwb; mutex_exit(&zilog->zl_lock); } ASSERT3P(lwb->lwb_root_zio, !=, NULL); ASSERT3P(lwb->lwb_write_zio, !=, NULL); ASSERT3S(lwb->lwb_state, ==, LWB_STATE_OPENED); } /* * Define a limited set of intent log block sizes. * * These must be a multiple of 4KB. Note only the amount used (again * aligned to 4KB) actually gets written. However, we can't always just * allocate SPA_OLD_MAXBLOCKSIZE as the slog space could be exhausted. */ uint64_t zil_block_buckets[] = { 4096, /* non TX_WRITE */ 8192+4096, /* data base */ 32*1024 + 4096, /* NFS writes */ UINT64_MAX }; /* * Start a log block write and advance to the next log block. * Calls are serialized. */ static lwb_t * zil_lwb_write_issue(zilog_t *zilog, lwb_t *lwb) { lwb_t *nlwb = NULL; zil_chain_t *zilc; spa_t *spa = zilog->zl_spa; blkptr_t *bp; dmu_tx_t *tx; uint64_t txg; uint64_t zil_blksz, wsz; int i, error; boolean_t slog; ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock)); ASSERT3P(lwb->lwb_root_zio, !=, NULL); ASSERT3P(lwb->lwb_write_zio, !=, NULL); ASSERT3S(lwb->lwb_state, ==, LWB_STATE_OPENED); if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { zilc = (zil_chain_t *)lwb->lwb_buf; bp = &zilc->zc_next_blk; } else { zilc = (zil_chain_t *)(lwb->lwb_buf + lwb->lwb_sz); bp = &zilc->zc_next_blk; } ASSERT(lwb->lwb_nused <= lwb->lwb_sz); /* * Allocate the next block and save its address in this block * before writing it in order to establish the log chain. * Note that if the allocation of nlwb synced before we wrote * the block that points at it (lwb), we'd leak it if we crashed. * Therefore, we don't do dmu_tx_commit() until zil_lwb_write_done(). * We dirty the dataset to ensure that zil_sync() will be called * to clean up in the event of allocation failure or I/O failure. */ tx = dmu_tx_create(zilog->zl_os); /* * Since we are not going to create any new dirty data, and we * can even help with clearing the existing dirty data, we * should not be subject to the dirty data based delays. We * use TXG_NOTHROTTLE to bypass the delay mechanism. */ VERIFY0(dmu_tx_assign(tx, TXG_WAIT | TXG_NOTHROTTLE)); dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); txg = dmu_tx_get_txg(tx); lwb->lwb_tx = tx; /* * Log blocks are pre-allocated. Here we select the size of the next * block, based on size used in the last block. * - first find the smallest bucket that will fit the block from a * limited set of block sizes. This is because it's faster to write * blocks allocated from the same metaslab as they are adjacent or * close. * - next find the maximum from the new suggested size and an array of * previous sizes. This lessens a picket fence effect of wrongly * guesssing the size if we have a stream of say 2k, 64k, 2k, 64k * requests. * * Note we only write what is used, but we can't just allocate * the maximum block size because we can exhaust the available * pool log space. */ zil_blksz = zilog->zl_cur_used + sizeof (zil_chain_t); for (i = 0; zil_blksz > zil_block_buckets[i]; i++) continue; zil_blksz = zil_block_buckets[i]; if (zil_blksz == UINT64_MAX) zil_blksz = SPA_OLD_MAXBLOCKSIZE; zilog->zl_prev_blks[zilog->zl_prev_rotor] = zil_blksz; for (i = 0; i < ZIL_PREV_BLKS; i++) zil_blksz = MAX(zil_blksz, zilog->zl_prev_blks[i]); zilog->zl_prev_rotor = (zilog->zl_prev_rotor + 1) & (ZIL_PREV_BLKS - 1); BP_ZERO(bp); /* pass the old blkptr in order to spread log blocks across devs */ - error = zio_alloc_zil(spa, txg, bp, &lwb->lwb_blk, zil_blksz, &slog); + error = zio_alloc_zil(spa, zilog->zl_os->os_dsl_dataset->ds_object, + txg, bp, &lwb->lwb_blk, zil_blksz, &slog); if (error == 0) { ASSERT3U(bp->blk_birth, ==, txg); bp->blk_cksum = lwb->lwb_blk.blk_cksum; bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++; /* * Allocate a new log write block (lwb). */ nlwb = zil_alloc_lwb(zilog, bp, slog, txg); } if (BP_GET_CHECKSUM(&lwb->lwb_blk) == ZIO_CHECKSUM_ZILOG2) { /* For Slim ZIL only write what is used. */ wsz = P2ROUNDUP_TYPED(lwb->lwb_nused, ZIL_MIN_BLKSZ, uint64_t); ASSERT3U(wsz, <=, lwb->lwb_sz); zio_shrink(lwb->lwb_write_zio, wsz); } else { wsz = lwb->lwb_sz; } zilc->zc_pad = 0; zilc->zc_nused = lwb->lwb_nused; zilc->zc_eck.zec_cksum = lwb->lwb_blk.blk_cksum; /* * clear unused data for security */ bzero(lwb->lwb_buf + lwb->lwb_nused, wsz - lwb->lwb_nused); spa_config_enter(zilog->zl_spa, SCL_STATE, lwb, RW_READER); zil_lwb_add_block(lwb, &lwb->lwb_blk); lwb->lwb_issued_timestamp = gethrtime(); lwb->lwb_state = LWB_STATE_ISSUED; zio_nowait(lwb->lwb_root_zio); zio_nowait(lwb->lwb_write_zio); /* * If there was an allocation failure then nlwb will be null which * forces a txg_wait_synced(). */ return (nlwb); } static lwb_t * zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb) { lr_t *lrcb, *lrc; lr_write_t *lrwb, *lrw; char *lr_buf; uint64_t dlen, dnow, lwb_sp, reclen, txg; ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock)); ASSERT3P(lwb, !=, NULL); ASSERT3P(lwb->lwb_buf, !=, NULL); zil_lwb_write_open(zilog, lwb); lrc = &itx->itx_lr; lrw = (lr_write_t *)lrc; /* * A commit itx doesn't represent any on-disk state; instead * it's simply used as a place holder on the commit list, and * provides a mechanism for attaching a "commit waiter" onto the * correct lwb (such that the waiter can be signalled upon * completion of that lwb). Thus, we don't process this itx's * log record if it's a commit itx (these itx's don't have log * records), and instead link the itx's waiter onto the lwb's * list of waiters. * * For more details, see the comment above zil_commit(). */ if (lrc->lrc_txtype == TX_COMMIT) { mutex_enter(&zilog->zl_lock); zil_commit_waiter_link_lwb(itx->itx_private, lwb); itx->itx_private = NULL; mutex_exit(&zilog->zl_lock); return (lwb); } if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY) { dlen = P2ROUNDUP_TYPED( lrw->lr_length, sizeof (uint64_t), uint64_t); } else { dlen = 0; } reclen = lrc->lrc_reclen; zilog->zl_cur_used += (reclen + dlen); txg = lrc->lrc_txg; ASSERT3U(zilog->zl_cur_used, <, UINT64_MAX - (reclen + dlen)); cont: /* * If this record won't fit in the current log block, start a new one. * For WR_NEED_COPY optimize layout for minimal number of chunks. */ lwb_sp = lwb->lwb_sz - lwb->lwb_nused; if (reclen > lwb_sp || (reclen + dlen > lwb_sp && lwb_sp < ZIL_MAX_WASTE_SPACE && (dlen % ZIL_MAX_LOG_DATA == 0 || lwb_sp < reclen + dlen % ZIL_MAX_LOG_DATA))) { lwb = zil_lwb_write_issue(zilog, lwb); if (lwb == NULL) return (NULL); zil_lwb_write_open(zilog, lwb); ASSERT(LWB_EMPTY(lwb)); lwb_sp = lwb->lwb_sz - lwb->lwb_nused; ASSERT3U(reclen + MIN(dlen, sizeof (uint64_t)), <=, lwb_sp); } dnow = MIN(dlen, lwb_sp - reclen); lr_buf = lwb->lwb_buf + lwb->lwb_nused; bcopy(lrc, lr_buf, reclen); lrcb = (lr_t *)lr_buf; /* Like lrc, but inside lwb. */ lrwb = (lr_write_t *)lrcb; /* Like lrw, but inside lwb. */ /* * If it's a write, fetch the data or get its blkptr as appropriate. */ if (lrc->lrc_txtype == TX_WRITE) { if (txg > spa_freeze_txg(zilog->zl_spa)) txg_wait_synced(zilog->zl_dmu_pool, txg); if (itx->itx_wr_state != WR_COPIED) { char *dbuf; int error; if (itx->itx_wr_state == WR_NEED_COPY) { dbuf = lr_buf + reclen; lrcb->lrc_reclen += dnow; if (lrwb->lr_length > dnow) lrwb->lr_length = dnow; lrw->lr_offset += dnow; lrw->lr_length -= dnow; } else { ASSERT(itx->itx_wr_state == WR_INDIRECT); dbuf = NULL; } /* * We pass in the "lwb_write_zio" rather than * "lwb_root_zio" so that the "lwb_write_zio" * becomes the parent of any zio's created by * the "zl_get_data" callback. The vdevs are * flushed after the "lwb_write_zio" completes, * so we want to make sure that completion * callback waits for these additional zio's, * such that the vdevs used by those zio's will * be included in the lwb's vdev tree, and those * vdevs will be properly flushed. If we passed * in "lwb_root_zio" here, then these additional * vdevs may not be flushed; e.g. if these zio's * completed after "lwb_write_zio" completed. */ error = zilog->zl_get_data(itx->itx_private, lrwb, dbuf, lwb, lwb->lwb_write_zio); if (error == EIO) { txg_wait_synced(zilog->zl_dmu_pool, txg); return (lwb); } if (error != 0) { ASSERT(error == ENOENT || error == EEXIST || error == EALREADY); return (lwb); } } } /* * We're actually making an entry, so update lrc_seq to be the * log record sequence number. Note that this is generally not * equal to the itx sequence number because not all transactions * are synchronous, and sometimes spa_sync() gets there first. */ lrcb->lrc_seq = ++zilog->zl_lr_seq; lwb->lwb_nused += reclen + dnow; zil_lwb_add_txg(lwb, txg); ASSERT3U(lwb->lwb_nused, <=, lwb->lwb_sz); ASSERT0(P2PHASE(lwb->lwb_nused, sizeof (uint64_t))); dlen -= dnow; if (dlen > 0) { zilog->zl_cur_used += reclen; goto cont; } return (lwb); } itx_t * zil_itx_create(uint64_t txtype, size_t lrsize) { itx_t *itx; lrsize = P2ROUNDUP_TYPED(lrsize, sizeof (uint64_t), size_t); itx = kmem_alloc(offsetof(itx_t, itx_lr) + lrsize, KM_SLEEP); itx->itx_lr.lrc_txtype = txtype; itx->itx_lr.lrc_reclen = lrsize; itx->itx_lr.lrc_seq = 0; /* defensive */ itx->itx_sync = B_TRUE; /* default is synchronous */ return (itx); } void zil_itx_destroy(itx_t *itx) { kmem_free(itx, offsetof(itx_t, itx_lr) + itx->itx_lr.lrc_reclen); } /* * Free up the sync and async itxs. The itxs_t has already been detached * so no locks are needed. */ static void zil_itxg_clean(itxs_t *itxs) { itx_t *itx; list_t *list; avl_tree_t *t; void *cookie; itx_async_node_t *ian; list = &itxs->i_sync_list; while ((itx = list_head(list)) != NULL) { /* * In the general case, commit itxs will not be found * here, as they'll be committed to an lwb via * zil_lwb_commit(), and free'd in that function. Having * said that, it is still possible for commit itxs to be * found here, due to the following race: * * - a thread calls zil_commit() which assigns the * commit itx to a per-txg i_sync_list * - zil_itxg_clean() is called (e.g. via spa_sync()) * while the waiter is still on the i_sync_list * * There's nothing to prevent syncing the txg while the * waiter is on the i_sync_list. This normally doesn't * happen because spa_sync() is slower than zil_commit(), * but if zil_commit() calls txg_wait_synced() (e.g. * because zil_create() or zil_commit_writer_stall() is * called) we will hit this case. */ if (itx->itx_lr.lrc_txtype == TX_COMMIT) zil_commit_waiter_skip(itx->itx_private); list_remove(list, itx); zil_itx_destroy(itx); } cookie = NULL; t = &itxs->i_async_tree; while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { list = &ian->ia_list; while ((itx = list_head(list)) != NULL) { list_remove(list, itx); /* commit itxs should never be on the async lists. */ ASSERT3U(itx->itx_lr.lrc_txtype, !=, TX_COMMIT); zil_itx_destroy(itx); } list_destroy(list); kmem_free(ian, sizeof (itx_async_node_t)); } avl_destroy(t); kmem_free(itxs, sizeof (itxs_t)); } static int zil_aitx_compare(const void *x1, const void *x2) { const uint64_t o1 = ((itx_async_node_t *)x1)->ia_foid; const uint64_t o2 = ((itx_async_node_t *)x2)->ia_foid; if (o1 < o2) return (-1); if (o1 > o2) return (1); return (0); } /* * Remove all async itx with the given oid. */ static void zil_remove_async(zilog_t *zilog, uint64_t oid) { uint64_t otxg, txg; itx_async_node_t *ian; avl_tree_t *t; avl_index_t where; list_t clean_list; itx_t *itx; ASSERT(oid != 0); list_create(&clean_list, sizeof (itx_t), offsetof(itx_t, itx_node)); if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ otxg = ZILTEST_TXG; else otxg = spa_last_synced_txg(zilog->zl_spa) + 1; for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; mutex_enter(&itxg->itxg_lock); if (itxg->itxg_txg != txg) { mutex_exit(&itxg->itxg_lock); continue; } /* * Locate the object node and append its list. */ t = &itxg->itxg_itxs->i_async_tree; ian = avl_find(t, &oid, &where); if (ian != NULL) list_move_tail(&clean_list, &ian->ia_list); mutex_exit(&itxg->itxg_lock); } while ((itx = list_head(&clean_list)) != NULL) { list_remove(&clean_list, itx); /* commit itxs should never be on the async lists. */ ASSERT3U(itx->itx_lr.lrc_txtype, !=, TX_COMMIT); zil_itx_destroy(itx); } list_destroy(&clean_list); } void zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx) { uint64_t txg; itxg_t *itxg; itxs_t *itxs, *clean = NULL; /* * Object ids can be re-instantiated in the next txg so * remove any async transactions to avoid future leaks. * This can happen if a fsync occurs on the re-instantiated * object for a WR_INDIRECT or WR_NEED_COPY write, which gets * the new file data and flushes a write record for the old object. */ if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_REMOVE) zil_remove_async(zilog, itx->itx_oid); /* * Ensure the data of a renamed file is committed before the rename. */ if ((itx->itx_lr.lrc_txtype & ~TX_CI) == TX_RENAME) zil_async_to_sync(zilog, itx->itx_oid); if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) txg = ZILTEST_TXG; else txg = dmu_tx_get_txg(tx); itxg = &zilog->zl_itxg[txg & TXG_MASK]; mutex_enter(&itxg->itxg_lock); itxs = itxg->itxg_itxs; if (itxg->itxg_txg != txg) { if (itxs != NULL) { /* * The zil_clean callback hasn't got around to cleaning * this itxg. Save the itxs for release below. * This should be rare. */ zfs_dbgmsg("zil_itx_assign: missed itx cleanup for " "txg %llu", itxg->itxg_txg); clean = itxg->itxg_itxs; } itxg->itxg_txg = txg; itxs = itxg->itxg_itxs = kmem_zalloc(sizeof (itxs_t), KM_SLEEP); list_create(&itxs->i_sync_list, sizeof (itx_t), offsetof(itx_t, itx_node)); avl_create(&itxs->i_async_tree, zil_aitx_compare, sizeof (itx_async_node_t), offsetof(itx_async_node_t, ia_node)); } if (itx->itx_sync) { list_insert_tail(&itxs->i_sync_list, itx); } else { avl_tree_t *t = &itxs->i_async_tree; uint64_t foid = ((lr_ooo_t *)&itx->itx_lr)->lr_foid; itx_async_node_t *ian; avl_index_t where; ian = avl_find(t, &foid, &where); if (ian == NULL) { ian = kmem_alloc(sizeof (itx_async_node_t), KM_SLEEP); list_create(&ian->ia_list, sizeof (itx_t), offsetof(itx_t, itx_node)); ian->ia_foid = foid; avl_insert(t, ian, where); } list_insert_tail(&ian->ia_list, itx); } itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx); /* * We don't want to dirty the ZIL using ZILTEST_TXG, because * zil_clean() will never be called using ZILTEST_TXG. Thus, we * need to be careful to always dirty the ZIL using the "real" * TXG (not itxg_txg) even when the SPA is frozen. */ zilog_dirty(zilog, dmu_tx_get_txg(tx)); mutex_exit(&itxg->itxg_lock); /* Release the old itxs now we've dropped the lock */ if (clean != NULL) zil_itxg_clean(clean); } /* * If there are any in-memory intent log transactions which have now been * synced then start up a taskq to free them. We should only do this after we * have written out the uberblocks (i.e. txg has been comitted) so that * don't inadvertently clean out in-memory log records that would be required * by zil_commit(). */ void zil_clean(zilog_t *zilog, uint64_t synced_txg) { itxg_t *itxg = &zilog->zl_itxg[synced_txg & TXG_MASK]; itxs_t *clean_me; ASSERT3U(synced_txg, <, ZILTEST_TXG); mutex_enter(&itxg->itxg_lock); if (itxg->itxg_itxs == NULL || itxg->itxg_txg == ZILTEST_TXG) { mutex_exit(&itxg->itxg_lock); return; } ASSERT3U(itxg->itxg_txg, <=, synced_txg); ASSERT3U(itxg->itxg_txg, !=, 0); clean_me = itxg->itxg_itxs; itxg->itxg_itxs = NULL; itxg->itxg_txg = 0; mutex_exit(&itxg->itxg_lock); /* * Preferably start a task queue to free up the old itxs but * if taskq_dispatch can't allocate resources to do that then * free it in-line. This should be rare. Note, using TQ_SLEEP * created a bad performance problem. */ ASSERT3P(zilog->zl_dmu_pool, !=, NULL); ASSERT3P(zilog->zl_dmu_pool->dp_zil_clean_taskq, !=, NULL); if (taskq_dispatch(zilog->zl_dmu_pool->dp_zil_clean_taskq, (void (*)(void *))zil_itxg_clean, clean_me, TQ_NOSLEEP) == 0) zil_itxg_clean(clean_me); } /* * This function will traverse the queue of itxs that need to be * committed, and move them onto the ZIL's zl_itx_commit_list. */ static void zil_get_commit_list(zilog_t *zilog) { uint64_t otxg, txg; list_t *commit_list = &zilog->zl_itx_commit_list; ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock)); if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ otxg = ZILTEST_TXG; else otxg = spa_last_synced_txg(zilog->zl_spa) + 1; /* * This is inherently racy, since there is nothing to prevent * the last synced txg from changing. That's okay since we'll * only commit things in the future. */ for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; mutex_enter(&itxg->itxg_lock); if (itxg->itxg_txg != txg) { mutex_exit(&itxg->itxg_lock); continue; } /* * If we're adding itx records to the zl_itx_commit_list, * then the zil better be dirty in this "txg". We can assert * that here since we're holding the itxg_lock which will * prevent spa_sync from cleaning it. Once we add the itxs * to the zl_itx_commit_list we must commit it to disk even * if it's unnecessary (i.e. the txg was synced). */ ASSERT(zilog_is_dirty_in_txg(zilog, txg) || spa_freeze_txg(zilog->zl_spa) != UINT64_MAX); list_move_tail(commit_list, &itxg->itxg_itxs->i_sync_list); mutex_exit(&itxg->itxg_lock); } } /* * Move the async itxs for a specified object to commit into sync lists. */ void zil_async_to_sync(zilog_t *zilog, uint64_t foid) { uint64_t otxg, txg; itx_async_node_t *ian; avl_tree_t *t; avl_index_t where; if (spa_freeze_txg(zilog->zl_spa) != UINT64_MAX) /* ziltest support */ otxg = ZILTEST_TXG; else otxg = spa_last_synced_txg(zilog->zl_spa) + 1; /* * This is inherently racy, since there is nothing to prevent * the last synced txg from changing. */ for (txg = otxg; txg < (otxg + TXG_CONCURRENT_STATES); txg++) { itxg_t *itxg = &zilog->zl_itxg[txg & TXG_MASK]; mutex_enter(&itxg->itxg_lock); if (itxg->itxg_txg != txg) { mutex_exit(&itxg->itxg_lock); continue; } /* * If a foid is specified then find that node and append its * list. Otherwise walk the tree appending all the lists * to the sync list. We add to the end rather than the * beginning to ensure the create has happened. */ t = &itxg->itxg_itxs->i_async_tree; if (foid != 0) { ian = avl_find(t, &foid, &where); if (ian != NULL) { list_move_tail(&itxg->itxg_itxs->i_sync_list, &ian->ia_list); } } else { void *cookie = NULL; while ((ian = avl_destroy_nodes(t, &cookie)) != NULL) { list_move_tail(&itxg->itxg_itxs->i_sync_list, &ian->ia_list); list_destroy(&ian->ia_list); kmem_free(ian, sizeof (itx_async_node_t)); } } mutex_exit(&itxg->itxg_lock); } } /* * This function will prune commit itxs that are at the head of the * commit list (it won't prune past the first non-commit itx), and * either: a) attach them to the last lwb that's still pending * completion, or b) skip them altogether. * * This is used as a performance optimization to prevent commit itxs * from generating new lwbs when it's unnecessary to do so. */ static void zil_prune_commit_list(zilog_t *zilog) { itx_t *itx; ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock)); while (itx = list_head(&zilog->zl_itx_commit_list)) { lr_t *lrc = &itx->itx_lr; if (lrc->lrc_txtype != TX_COMMIT) break; mutex_enter(&zilog->zl_lock); lwb_t *last_lwb = zilog->zl_last_lwb_opened; if (last_lwb == NULL || last_lwb->lwb_state == LWB_STATE_DONE) { /* * All of the itxs this waiter was waiting on * must have already completed (or there were * never any itx's for it to wait on), so it's * safe to skip this waiter and mark it done. */ zil_commit_waiter_skip(itx->itx_private); } else { zil_commit_waiter_link_lwb(itx->itx_private, last_lwb); itx->itx_private = NULL; } mutex_exit(&zilog->zl_lock); list_remove(&zilog->zl_itx_commit_list, itx); zil_itx_destroy(itx); } IMPLY(itx != NULL, itx->itx_lr.lrc_txtype != TX_COMMIT); } static void zil_commit_writer_stall(zilog_t *zilog) { /* * When zio_alloc_zil() fails to allocate the next lwb block on * disk, we must call txg_wait_synced() to ensure all of the * lwbs in the zilog's zl_lwb_list are synced and then freed (in * zil_sync()), such that any subsequent ZIL writer (i.e. a call * to zil_process_commit_list()) will have to call zil_create(), * and start a new ZIL chain. * * Since zil_alloc_zil() failed, the lwb that was previously * issued does not have a pointer to the "next" lwb on disk. * Thus, if another ZIL writer thread was to allocate the "next" * on-disk lwb, that block could be leaked in the event of a * crash (because the previous lwb on-disk would not point to * it). * * We must hold the zilog's zl_issuer_lock while we do this, to * ensure no new threads enter zil_process_commit_list() until * all lwb's in the zl_lwb_list have been synced and freed * (which is achieved via the txg_wait_synced() call). */ ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock)); txg_wait_synced(zilog->zl_dmu_pool, 0); ASSERT3P(list_tail(&zilog->zl_lwb_list), ==, NULL); } /* * This function will traverse the commit list, creating new lwbs as * needed, and committing the itxs from the commit list to these newly * created lwbs. Additionally, as a new lwb is created, the previous * lwb will be issued to the zio layer to be written to disk. */ static void zil_process_commit_list(zilog_t *zilog) { spa_t *spa = zilog->zl_spa; list_t nolwb_waiters; lwb_t *lwb; itx_t *itx; ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock)); /* * Return if there's nothing to commit before we dirty the fs by * calling zil_create(). */ if (list_head(&zilog->zl_itx_commit_list) == NULL) return; list_create(&nolwb_waiters, sizeof (zil_commit_waiter_t), offsetof(zil_commit_waiter_t, zcw_node)); lwb = list_tail(&zilog->zl_lwb_list); if (lwb == NULL) { lwb = zil_create(zilog); } else { ASSERT3S(lwb->lwb_state, !=, LWB_STATE_ISSUED); ASSERT3S(lwb->lwb_state, !=, LWB_STATE_DONE); } while (itx = list_head(&zilog->zl_itx_commit_list)) { lr_t *lrc = &itx->itx_lr; uint64_t txg = lrc->lrc_txg; ASSERT3U(txg, !=, 0); if (lrc->lrc_txtype == TX_COMMIT) { DTRACE_PROBE2(zil__process__commit__itx, zilog_t *, zilog, itx_t *, itx); } else { DTRACE_PROBE2(zil__process__normal__itx, zilog_t *, zilog, itx_t *, itx); } boolean_t synced = txg <= spa_last_synced_txg(spa); boolean_t frozen = txg > spa_freeze_txg(spa); /* * If the txg of this itx has already been synced out, then * we don't need to commit this itx to an lwb. This is * because the data of this itx will have already been * written to the main pool. This is inherently racy, and * it's still ok to commit an itx whose txg has already * been synced; this will result in a write that's * unnecessary, but will do no harm. * * With that said, we always want to commit TX_COMMIT itxs * to an lwb, regardless of whether or not that itx's txg * has been synced out. We do this to ensure any OPENED lwb * will always have at least one zil_commit_waiter_t linked * to the lwb. * * As a counter-example, if we skipped TX_COMMIT itx's * whose txg had already been synced, the following * situation could occur if we happened to be racing with * spa_sync: * * 1. we commit a non-TX_COMMIT itx to an lwb, where the * itx's txg is 10 and the last synced txg is 9. * 2. spa_sync finishes syncing out txg 10. * 3. we move to the next itx in the list, it's a TX_COMMIT * whose txg is 10, so we skip it rather than committing * it to the lwb used in (1). * * If the itx that is skipped in (3) is the last TX_COMMIT * itx in the commit list, than it's possible for the lwb * used in (1) to remain in the OPENED state indefinitely. * * To prevent the above scenario from occuring, ensuring * that once an lwb is OPENED it will transition to ISSUED * and eventually DONE, we always commit TX_COMMIT itx's to * an lwb here, even if that itx's txg has already been * synced. * * Finally, if the pool is frozen, we _always_ commit the * itx. The point of freezing the pool is to prevent data * from being written to the main pool via spa_sync, and * instead rely solely on the ZIL to persistently store the * data; i.e. when the pool is frozen, the last synced txg * value can't be trusted. */ if (frozen || !synced || lrc->lrc_txtype == TX_COMMIT) { if (lwb != NULL) { lwb = zil_lwb_commit(zilog, itx, lwb); } else if (lrc->lrc_txtype == TX_COMMIT) { ASSERT3P(lwb, ==, NULL); zil_commit_waiter_link_nolwb( itx->itx_private, &nolwb_waiters); } } list_remove(&zilog->zl_itx_commit_list, itx); zil_itx_destroy(itx); } if (lwb == NULL) { /* * This indicates zio_alloc_zil() failed to allocate the * "next" lwb on-disk. When this happens, we must stall * the ZIL write pipeline; see the comment within * zil_commit_writer_stall() for more details. */ zil_commit_writer_stall(zilog); /* * Additionally, we have to signal and mark the "nolwb" * waiters as "done" here, since without an lwb, we * can't do this via zil_lwb_flush_vdevs_done() like * normal. */ zil_commit_waiter_t *zcw; while (zcw = list_head(&nolwb_waiters)) { zil_commit_waiter_skip(zcw); list_remove(&nolwb_waiters, zcw); } } else { ASSERT(list_is_empty(&nolwb_waiters)); ASSERT3P(lwb, !=, NULL); ASSERT3S(lwb->lwb_state, !=, LWB_STATE_ISSUED); ASSERT3S(lwb->lwb_state, !=, LWB_STATE_DONE); /* * At this point, the ZIL block pointed at by the "lwb" * variable is in one of the following states: "closed" * or "open". * * If its "closed", then no itxs have been committed to * it, so there's no point in issuing its zio (i.e. * it's "empty"). * * If its "open" state, then it contains one or more * itxs that eventually need to be committed to stable * storage. In this case we intentionally do not issue * the lwb's zio to disk yet, and instead rely on one of * the following two mechanisms for issuing the zio: * * 1. Ideally, there will be more ZIL activity occuring * on the system, such that this function will be * immediately called again (not necessarily by the same * thread) and this lwb's zio will be issued via * zil_lwb_commit(). This way, the lwb is guaranteed to * be "full" when it is issued to disk, and we'll make * use of the lwb's size the best we can. * * 2. If there isn't sufficient ZIL activity occuring on * the system, such that this lwb's zio isn't issued via * zil_lwb_commit(), zil_commit_waiter() will issue the * lwb's zio. If this occurs, the lwb is not guaranteed * to be "full" by the time its zio is issued, and means * the size of the lwb was "too large" given the amount * of ZIL activity occuring on the system at that time. * * We do this for a couple of reasons: * * 1. To try and reduce the number of IOPs needed to * write the same number of itxs. If an lwb has space * available in it's buffer for more itxs, and more itxs * will be committed relatively soon (relative to the * latency of performing a write), then it's beneficial * to wait for these "next" itxs. This way, more itxs * can be committed to stable storage with fewer writes. * * 2. To try and use the largest lwb block size that the * incoming rate of itxs can support. Again, this is to * try and pack as many itxs into as few lwbs as * possible, without significantly impacting the latency * of each individual itx. */ } } /* * This function is responsible for ensuring the passed in commit waiter * (and associated commit itx) is committed to an lwb. If the waiter is * not already committed to an lwb, all itxs in the zilog's queue of * itxs will be processed. The assumption is the passed in waiter's * commit itx will found in the queue just like the other non-commit * itxs, such that when the entire queue is processed, the waiter will * have been commited to an lwb. * * The lwb associated with the passed in waiter is not guaranteed to * have been issued by the time this function completes. If the lwb is * not issued, we rely on future calls to zil_commit_writer() to issue * the lwb, or the timeout mechanism found in zil_commit_waiter(). */ static void zil_commit_writer(zilog_t *zilog, zil_commit_waiter_t *zcw) { ASSERT(!MUTEX_HELD(&zilog->zl_lock)); ASSERT(spa_writeable(zilog->zl_spa)); mutex_enter(&zilog->zl_issuer_lock); if (zcw->zcw_lwb != NULL || zcw->zcw_done) { /* * It's possible that, while we were waiting to acquire * the "zl_issuer_lock", another thread committed this * waiter to an lwb. If that occurs, we bail out early, * without processing any of the zilog's queue of itxs. * * On certain workloads and system configurations, the * "zl_issuer_lock" can become highly contended. In an * attempt to reduce this contention, we immediately drop * the lock if the waiter has already been processed. * * We've measured this optimization to reduce CPU spent * contending on this lock by up to 5%, using a system * with 32 CPUs, low latency storage (~50 usec writes), * and 1024 threads performing sync writes. */ goto out; } zil_get_commit_list(zilog); zil_prune_commit_list(zilog); zil_process_commit_list(zilog); out: mutex_exit(&zilog->zl_issuer_lock); } static void zil_commit_waiter_timeout(zilog_t *zilog, zil_commit_waiter_t *zcw) { ASSERT(!MUTEX_HELD(&zilog->zl_issuer_lock)); ASSERT(MUTEX_HELD(&zcw->zcw_lock)); ASSERT3B(zcw->zcw_done, ==, B_FALSE); lwb_t *lwb = zcw->zcw_lwb; ASSERT3P(lwb, !=, NULL); ASSERT3S(lwb->lwb_state, !=, LWB_STATE_CLOSED); /* * If the lwb has already been issued by another thread, we can * immediately return since there's no work to be done (the * point of this function is to issue the lwb). Additionally, we * do this prior to acquiring the zl_issuer_lock, to avoid * acquiring it when it's not necessary to do so. */ if (lwb->lwb_state == LWB_STATE_ISSUED || lwb->lwb_state == LWB_STATE_DONE) return; /* * In order to call zil_lwb_write_issue() we must hold the * zilog's "zl_issuer_lock". We can't simply acquire that lock, * since we're already holding the commit waiter's "zcw_lock", * and those two locks are aquired in the opposite order * elsewhere. */ mutex_exit(&zcw->zcw_lock); mutex_enter(&zilog->zl_issuer_lock); mutex_enter(&zcw->zcw_lock); /* * Since we just dropped and re-acquired the commit waiter's * lock, we have to re-check to see if the waiter was marked * "done" during that process. If the waiter was marked "done", * the "lwb" pointer is no longer valid (it can be free'd after * the waiter is marked "done"), so without this check we could * wind up with a use-after-free error below. */ if (zcw->zcw_done) goto out; ASSERT3P(lwb, ==, zcw->zcw_lwb); /* * We've already checked this above, but since we hadn't acquired * the zilog's zl_issuer_lock, we have to perform this check a * second time while holding the lock. * * We don't need to hold the zl_lock since the lwb cannot transition * from OPENED to ISSUED while we hold the zl_issuer_lock. The lwb * _can_ transition from ISSUED to DONE, but it's OK to race with * that transition since we treat the lwb the same, whether it's in * the ISSUED or DONE states. * * The important thing, is we treat the lwb differently depending on * if it's ISSUED or OPENED, and block any other threads that might * attempt to issue this lwb. For that reason we hold the * zl_issuer_lock when checking the lwb_state; we must not call * zil_lwb_write_issue() if the lwb had already been issued. * * See the comment above the lwb_state_t structure definition for * more details on the lwb states, and locking requirements. */ if (lwb->lwb_state == LWB_STATE_ISSUED || lwb->lwb_state == LWB_STATE_DONE) goto out; ASSERT3S(lwb->lwb_state, ==, LWB_STATE_OPENED); /* * As described in the comments above zil_commit_waiter() and * zil_process_commit_list(), we need to issue this lwb's zio * since we've reached the commit waiter's timeout and it still * hasn't been issued. */ lwb_t *nlwb = zil_lwb_write_issue(zilog, lwb); ASSERT3S(lwb->lwb_state, !=, LWB_STATE_OPENED); /* * Since the lwb's zio hadn't been issued by the time this thread * reached its timeout, we reset the zilog's "zl_cur_used" field * to influence the zil block size selection algorithm. * * By having to issue the lwb's zio here, it means the size of the * lwb was too large, given the incoming throughput of itxs. By * setting "zl_cur_used" to zero, we communicate this fact to the * block size selection algorithm, so it can take this informaiton * into account, and potentially select a smaller size for the * next lwb block that is allocated. */ zilog->zl_cur_used = 0; if (nlwb == NULL) { /* * When zil_lwb_write_issue() returns NULL, this * indicates zio_alloc_zil() failed to allocate the * "next" lwb on-disk. When this occurs, the ZIL write * pipeline must be stalled; see the comment within the * zil_commit_writer_stall() function for more details. * * We must drop the commit waiter's lock prior to * calling zil_commit_writer_stall() or else we can wind * up with the following deadlock: * * - This thread is waiting for the txg to sync while * holding the waiter's lock; txg_wait_synced() is * used within txg_commit_writer_stall(). * * - The txg can't sync because it is waiting for this * lwb's zio callback to call dmu_tx_commit(). * * - The lwb's zio callback can't call dmu_tx_commit() * because it's blocked trying to acquire the waiter's * lock, which occurs prior to calling dmu_tx_commit() */ mutex_exit(&zcw->zcw_lock); zil_commit_writer_stall(zilog); mutex_enter(&zcw->zcw_lock); } out: mutex_exit(&zilog->zl_issuer_lock); ASSERT(MUTEX_HELD(&zcw->zcw_lock)); } /* * This function is responsible for performing the following two tasks: * * 1. its primary responsibility is to block until the given "commit * waiter" is considered "done". * * 2. its secondary responsibility is to issue the zio for the lwb that * the given "commit waiter" is waiting on, if this function has * waited "long enough" and the lwb is still in the "open" state. * * Given a sufficient amount of itxs being generated and written using * the ZIL, the lwb's zio will be issued via the zil_lwb_commit() * function. If this does not occur, this secondary responsibility will * ensure the lwb is issued even if there is not other synchronous * activity on the system. * * For more details, see zil_process_commit_list(); more specifically, * the comment at the bottom of that function. */ static void zil_commit_waiter(zilog_t *zilog, zil_commit_waiter_t *zcw) { ASSERT(!MUTEX_HELD(&zilog->zl_lock)); ASSERT(!MUTEX_HELD(&zilog->zl_issuer_lock)); ASSERT(spa_writeable(zilog->zl_spa)); mutex_enter(&zcw->zcw_lock); /* * The timeout is scaled based on the lwb latency to avoid * significantly impacting the latency of each individual itx. * For more details, see the comment at the bottom of the * zil_process_commit_list() function. */ int pct = MAX(zfs_commit_timeout_pct, 1); #if defined(illumos) || !defined(_KERNEL) hrtime_t sleep = (zilog->zl_last_lwb_latency * pct) / 100; hrtime_t wakeup = gethrtime() + sleep; #else sbintime_t sleep = nstosbt((zilog->zl_last_lwb_latency * pct) / 100); sbintime_t wakeup = getsbinuptime() + sleep; #endif boolean_t timedout = B_FALSE; while (!zcw->zcw_done) { ASSERT(MUTEX_HELD(&zcw->zcw_lock)); lwb_t *lwb = zcw->zcw_lwb; /* * Usually, the waiter will have a non-NULL lwb field here, * but it's possible for it to be NULL as a result of * zil_commit() racing with spa_sync(). * * When zil_clean() is called, it's possible for the itxg * list (which may be cleaned via a taskq) to contain * commit itxs. When this occurs, the commit waiters linked * off of these commit itxs will not be committed to an * lwb. Additionally, these commit waiters will not be * marked done until zil_commit_waiter_skip() is called via * zil_itxg_clean(). * * Thus, it's possible for this commit waiter (i.e. the * "zcw" variable) to be found in this "in between" state; * where it's "zcw_lwb" field is NULL, and it hasn't yet * been skipped, so it's "zcw_done" field is still B_FALSE. */ IMPLY(lwb != NULL, lwb->lwb_state != LWB_STATE_CLOSED); if (lwb != NULL && lwb->lwb_state == LWB_STATE_OPENED) { ASSERT3B(timedout, ==, B_FALSE); /* * If the lwb hasn't been issued yet, then we * need to wait with a timeout, in case this * function needs to issue the lwb after the * timeout is reached; responsibility (2) from * the comment above this function. */ #if defined(illumos) || !defined(_KERNEL) clock_t timeleft = cv_timedwait_hires(&zcw->zcw_cv, &zcw->zcw_lock, wakeup, USEC2NSEC(1), CALLOUT_FLAG_ABSOLUTE); if (timeleft >= 0 || zcw->zcw_done) continue; #else int wait_err = cv_timedwait_sbt(&zcw->zcw_cv, &zcw->zcw_lock, wakeup, SBT_1NS, C_ABSOLUTE); if (wait_err != EWOULDBLOCK || zcw->zcw_done) continue; #endif timedout = B_TRUE; zil_commit_waiter_timeout(zilog, zcw); if (!zcw->zcw_done) { /* * If the commit waiter has already been * marked "done", it's possible for the * waiter's lwb structure to have already * been freed. Thus, we can only reliably * make these assertions if the waiter * isn't done. */ ASSERT3P(lwb, ==, zcw->zcw_lwb); ASSERT3S(lwb->lwb_state, !=, LWB_STATE_OPENED); } } else { /* * If the lwb isn't open, then it must have already * been issued. In that case, there's no need to * use a timeout when waiting for the lwb to * complete. * * Additionally, if the lwb is NULL, the waiter * will soon be signalled and marked done via * zil_clean() and zil_itxg_clean(), so no timeout * is required. */ IMPLY(lwb != NULL, lwb->lwb_state == LWB_STATE_ISSUED || lwb->lwb_state == LWB_STATE_DONE); cv_wait(&zcw->zcw_cv, &zcw->zcw_lock); } } mutex_exit(&zcw->zcw_lock); } static zil_commit_waiter_t * zil_alloc_commit_waiter() { zil_commit_waiter_t *zcw = kmem_cache_alloc(zil_zcw_cache, KM_SLEEP); cv_init(&zcw->zcw_cv, NULL, CV_DEFAULT, NULL); mutex_init(&zcw->zcw_lock, NULL, MUTEX_DEFAULT, NULL); list_link_init(&zcw->zcw_node); zcw->zcw_lwb = NULL; zcw->zcw_done = B_FALSE; zcw->zcw_zio_error = 0; return (zcw); } static void zil_free_commit_waiter(zil_commit_waiter_t *zcw) { ASSERT(!list_link_active(&zcw->zcw_node)); ASSERT3P(zcw->zcw_lwb, ==, NULL); ASSERT3B(zcw->zcw_done, ==, B_TRUE); mutex_destroy(&zcw->zcw_lock); cv_destroy(&zcw->zcw_cv); kmem_cache_free(zil_zcw_cache, zcw); } /* * This function is used to create a TX_COMMIT itx and assign it. This * way, it will be linked into the ZIL's list of synchronous itxs, and * then later committed to an lwb (or skipped) when * zil_process_commit_list() is called. */ static void zil_commit_itx_assign(zilog_t *zilog, zil_commit_waiter_t *zcw) { dmu_tx_t *tx = dmu_tx_create(zilog->zl_os); VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); itx_t *itx = zil_itx_create(TX_COMMIT, sizeof (lr_t)); itx->itx_sync = B_TRUE; itx->itx_private = zcw; zil_itx_assign(zilog, itx, tx); dmu_tx_commit(tx); } /* * Commit ZFS Intent Log transactions (itxs) to stable storage. * * When writing ZIL transactions to the on-disk representation of the * ZIL, the itxs are committed to a Log Write Block (lwb). Multiple * itxs can be committed to a single lwb. Once a lwb is written and * committed to stable storage (i.e. the lwb is written, and vdevs have * been flushed), each itx that was committed to that lwb is also * considered to be committed to stable storage. * * When an itx is committed to an lwb, the log record (lr_t) contained * by the itx is copied into the lwb's zio buffer, and once this buffer * is written to disk, it becomes an on-disk ZIL block. * * As itxs are generated, they're inserted into the ZIL's queue of * uncommitted itxs. The semantics of zil_commit() are such that it will * block until all itxs that were in the queue when it was called, are * committed to stable storage. * * If "foid" is zero, this means all "synchronous" and "asynchronous" * itxs, for all objects in the dataset, will be committed to stable * storage prior to zil_commit() returning. If "foid" is non-zero, all * "synchronous" itxs for all objects, but only "asynchronous" itxs * that correspond to the foid passed in, will be committed to stable * storage prior to zil_commit() returning. * * Generally speaking, when zil_commit() is called, the consumer doesn't * actually care about _all_ of the uncommitted itxs. Instead, they're * simply trying to waiting for a specific itx to be committed to disk, * but the interface(s) for interacting with the ZIL don't allow such * fine-grained communication. A better interface would allow a consumer * to create and assign an itx, and then pass a reference to this itx to * zil_commit(); such that zil_commit() would return as soon as that * specific itx was committed to disk (instead of waiting for _all_ * itxs to be committed). * * When a thread calls zil_commit() a special "commit itx" will be * generated, along with a corresponding "waiter" for this commit itx. * zil_commit() will wait on this waiter's CV, such that when the waiter * is marked done, and signalled, zil_commit() will return. * * This commit itx is inserted into the queue of uncommitted itxs. This * provides an easy mechanism for determining which itxs were in the * queue prior to zil_commit() having been called, and which itxs were * added after zil_commit() was called. * * The commit it is special; it doesn't have any on-disk representation. * When a commit itx is "committed" to an lwb, the waiter associated * with it is linked onto the lwb's list of waiters. Then, when that lwb * completes, each waiter on the lwb's list is marked done and signalled * -- allowing the thread waiting on the waiter to return from zil_commit(). * * It's important to point out a few critical factors that allow us * to make use of the commit itxs, commit waiters, per-lwb lists of * commit waiters, and zio completion callbacks like we're doing: * * 1. The list of waiters for each lwb is traversed, and each commit * waiter is marked "done" and signalled, in the zio completion * callback of the lwb's zio[*]. * * * Actually, the waiters are signalled in the zio completion * callback of the root zio for the DKIOCFLUSHWRITECACHE commands * that are sent to the vdevs upon completion of the lwb zio. * * 2. When the itxs are inserted into the ZIL's queue of uncommitted * itxs, the order in which they are inserted is preserved[*]; as * itxs are added to the queue, they are added to the tail of * in-memory linked lists. * * When committing the itxs to lwbs (to be written to disk), they * are committed in the same order in which the itxs were added to * the uncommitted queue's linked list(s); i.e. the linked list of * itxs to commit is traversed from head to tail, and each itx is * committed to an lwb in that order. * * * To clarify: * * - the order of "sync" itxs is preserved w.r.t. other * "sync" itxs, regardless of the corresponding objects. * - the order of "async" itxs is preserved w.r.t. other * "async" itxs corresponding to the same object. * - the order of "async" itxs is *not* preserved w.r.t. other * "async" itxs corresponding to different objects. * - the order of "sync" itxs w.r.t. "async" itxs (or vice * versa) is *not* preserved, even for itxs that correspond * to the same object. * * For more details, see: zil_itx_assign(), zil_async_to_sync(), * zil_get_commit_list(), and zil_process_commit_list(). * * 3. The lwbs represent a linked list of blocks on disk. Thus, any * lwb cannot be considered committed to stable storage, until its * "previous" lwb is also committed to stable storage. This fact, * coupled with the fact described above, means that itxs are * committed in (roughly) the order in which they were generated. * This is essential because itxs are dependent on prior itxs. * Thus, we *must not* deem an itx as being committed to stable * storage, until *all* prior itxs have also been committed to * stable storage. * * To enforce this ordering of lwb zio's, while still leveraging as * much of the underlying storage performance as possible, we rely * on two fundamental concepts: * * 1. The creation and issuance of lwb zio's is protected by * the zilog's "zl_issuer_lock", which ensures only a single * thread is creating and/or issuing lwb's at a time * 2. The "previous" lwb is a child of the "current" lwb * (leveraging the zio parent-child depenency graph) * * By relying on this parent-child zio relationship, we can have * many lwb zio's concurrently issued to the underlying storage, * but the order in which they complete will be the same order in * which they were created. */ void zil_commit(zilog_t *zilog, uint64_t foid) { /* * We should never attempt to call zil_commit on a snapshot for * a couple of reasons: * * 1. A snapshot may never be modified, thus it cannot have any * in-flight itxs that would have modified the dataset. * * 2. By design, when zil_commit() is called, a commit itx will * be assigned to this zilog; as a result, the zilog will be * dirtied. We must not dirty the zilog of a snapshot; there's * checks in the code that enforce this invariant, and will * cause a panic if it's not upheld. */ ASSERT3B(dmu_objset_is_snapshot(zilog->zl_os), ==, B_FALSE); if (zilog->zl_sync == ZFS_SYNC_DISABLED) return; if (!spa_writeable(zilog->zl_spa)) { /* * If the SPA is not writable, there should never be any * pending itxs waiting to be committed to disk. If that * weren't true, we'd skip writing those itxs out, and * would break the sematics of zil_commit(); thus, we're * verifying that truth before we return to the caller. */ ASSERT(list_is_empty(&zilog->zl_lwb_list)); ASSERT3P(zilog->zl_last_lwb_opened, ==, NULL); for (int i = 0; i < TXG_SIZE; i++) ASSERT3P(zilog->zl_itxg[i].itxg_itxs, ==, NULL); return; } /* * If the ZIL is suspended, we don't want to dirty it by calling * zil_commit_itx_assign() below, nor can we write out * lwbs like would be done in zil_commit_write(). Thus, we * simply rely on txg_wait_synced() to maintain the necessary * semantics, and avoid calling those functions altogether. */ if (zilog->zl_suspend > 0) { txg_wait_synced(zilog->zl_dmu_pool, 0); return; } zil_commit_impl(zilog, foid); } void zil_commit_impl(zilog_t *zilog, uint64_t foid) { /* * Move the "async" itxs for the specified foid to the "sync" * queues, such that they will be later committed (or skipped) * to an lwb when zil_process_commit_list() is called. * * Since these "async" itxs must be committed prior to this * call to zil_commit returning, we must perform this operation * before we call zil_commit_itx_assign(). */ zil_async_to_sync(zilog, foid); /* * We allocate a new "waiter" structure which will initially be * linked to the commit itx using the itx's "itx_private" field. * Since the commit itx doesn't represent any on-disk state, * when it's committed to an lwb, rather than copying the its * lr_t into the lwb's buffer, the commit itx's "waiter" will be * added to the lwb's list of waiters. Then, when the lwb is * committed to stable storage, each waiter in the lwb's list of * waiters will be marked "done", and signalled. * * We must create the waiter and assign the commit itx prior to * calling zil_commit_writer(), or else our specific commit itx * is not guaranteed to be committed to an lwb prior to calling * zil_commit_waiter(). */ zil_commit_waiter_t *zcw = zil_alloc_commit_waiter(); zil_commit_itx_assign(zilog, zcw); zil_commit_writer(zilog, zcw); zil_commit_waiter(zilog, zcw); if (zcw->zcw_zio_error != 0) { /* * If there was an error writing out the ZIL blocks that * this thread is waiting on, then we fallback to * relying on spa_sync() to write out the data this * thread is waiting on. Obviously this has performance * implications, but the expectation is for this to be * an exceptional case, and shouldn't occur often. */ DTRACE_PROBE2(zil__commit__io__error, zilog_t *, zilog, zil_commit_waiter_t *, zcw); txg_wait_synced(zilog->zl_dmu_pool, 0); } zil_free_commit_waiter(zcw); } /* * Called in syncing context to free committed log blocks and update log header. */ void zil_sync(zilog_t *zilog, dmu_tx_t *tx) { zil_header_t *zh = zil_header_in_syncing_context(zilog); uint64_t txg = dmu_tx_get_txg(tx); spa_t *spa = zilog->zl_spa; uint64_t *replayed_seq = &zilog->zl_replayed_seq[txg & TXG_MASK]; lwb_t *lwb; /* * We don't zero out zl_destroy_txg, so make sure we don't try * to destroy it twice. */ if (spa_sync_pass(spa) != 1) return; mutex_enter(&zilog->zl_lock); ASSERT(zilog->zl_stop_sync == 0); if (*replayed_seq != 0) { ASSERT(zh->zh_replay_seq < *replayed_seq); zh->zh_replay_seq = *replayed_seq; *replayed_seq = 0; } if (zilog->zl_destroy_txg == txg) { blkptr_t blk = zh->zh_log; ASSERT(list_head(&zilog->zl_lwb_list) == NULL); bzero(zh, sizeof (zil_header_t)); bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq)); if (zilog->zl_keep_first) { /* * If this block was part of log chain that couldn't * be claimed because a device was missing during * zil_claim(), but that device later returns, * then this block could erroneously appear valid. * To guard against this, assign a new GUID to the new * log chain so it doesn't matter what blk points to. */ zil_init_log_chain(zilog, &blk); zh->zh_log = blk; } } while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) { zh->zh_log = lwb->lwb_blk; if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg) break; list_remove(&zilog->zl_lwb_list, lwb); zio_free(spa, txg, &lwb->lwb_blk); zil_free_lwb(zilog, lwb); /* * If we don't have anything left in the lwb list then * we've had an allocation failure and we need to zero * out the zil_header blkptr so that we don't end * up freeing the same block twice. */ if (list_head(&zilog->zl_lwb_list) == NULL) BP_ZERO(&zh->zh_log); } mutex_exit(&zilog->zl_lock); } /* ARGSUSED */ static int zil_lwb_cons(void *vbuf, void *unused, int kmflag) { lwb_t *lwb = vbuf; list_create(&lwb->lwb_waiters, sizeof (zil_commit_waiter_t), offsetof(zil_commit_waiter_t, zcw_node)); avl_create(&lwb->lwb_vdev_tree, zil_lwb_vdev_compare, sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node)); mutex_init(&lwb->lwb_vdev_lock, NULL, MUTEX_DEFAULT, NULL); return (0); } /* ARGSUSED */ static void zil_lwb_dest(void *vbuf, void *unused) { lwb_t *lwb = vbuf; mutex_destroy(&lwb->lwb_vdev_lock); avl_destroy(&lwb->lwb_vdev_tree); list_destroy(&lwb->lwb_waiters); } void zil_init(void) { zil_lwb_cache = kmem_cache_create("zil_lwb_cache", sizeof (lwb_t), 0, zil_lwb_cons, zil_lwb_dest, NULL, NULL, NULL, 0); zil_zcw_cache = kmem_cache_create("zil_zcw_cache", sizeof (zil_commit_waiter_t), 0, NULL, NULL, NULL, NULL, NULL, 0); } void zil_fini(void) { kmem_cache_destroy(zil_zcw_cache); kmem_cache_destroy(zil_lwb_cache); } void zil_set_sync(zilog_t *zilog, uint64_t sync) { zilog->zl_sync = sync; } void zil_set_logbias(zilog_t *zilog, uint64_t logbias) { zilog->zl_logbias = logbias; } zilog_t * zil_alloc(objset_t *os, zil_header_t *zh_phys) { zilog_t *zilog; zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP); zilog->zl_header = zh_phys; zilog->zl_os = os; zilog->zl_spa = dmu_objset_spa(os); zilog->zl_dmu_pool = dmu_objset_pool(os); zilog->zl_destroy_txg = TXG_INITIAL - 1; zilog->zl_logbias = dmu_objset_logbias(os); zilog->zl_sync = dmu_objset_syncprop(os); zilog->zl_dirty_max_txg = 0; zilog->zl_last_lwb_opened = NULL; zilog->zl_last_lwb_latency = 0; mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&zilog->zl_issuer_lock, NULL, MUTEX_DEFAULT, NULL); for (int i = 0; i < TXG_SIZE; i++) { mutex_init(&zilog->zl_itxg[i].itxg_lock, NULL, MUTEX_DEFAULT, NULL); } list_create(&zilog->zl_lwb_list, sizeof (lwb_t), offsetof(lwb_t, lwb_node)); list_create(&zilog->zl_itx_commit_list, sizeof (itx_t), offsetof(itx_t, itx_node)); cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL); return (zilog); } void zil_free(zilog_t *zilog) { zilog->zl_stop_sync = 1; ASSERT0(zilog->zl_suspend); ASSERT0(zilog->zl_suspending); ASSERT(list_is_empty(&zilog->zl_lwb_list)); list_destroy(&zilog->zl_lwb_list); ASSERT(list_is_empty(&zilog->zl_itx_commit_list)); list_destroy(&zilog->zl_itx_commit_list); for (int i = 0; i < TXG_SIZE; i++) { /* * It's possible for an itx to be generated that doesn't dirty * a txg (e.g. ztest TX_TRUNCATE). So there's no zil_clean() * callback to remove the entry. We remove those here. * * Also free up the ziltest itxs. */ if (zilog->zl_itxg[i].itxg_itxs) zil_itxg_clean(zilog->zl_itxg[i].itxg_itxs); mutex_destroy(&zilog->zl_itxg[i].itxg_lock); } mutex_destroy(&zilog->zl_issuer_lock); mutex_destroy(&zilog->zl_lock); cv_destroy(&zilog->zl_cv_suspend); kmem_free(zilog, sizeof (zilog_t)); } /* * Open an intent log. */ zilog_t * zil_open(objset_t *os, zil_get_data_t *get_data) { zilog_t *zilog = dmu_objset_zil(os); ASSERT3P(zilog->zl_get_data, ==, NULL); ASSERT3P(zilog->zl_last_lwb_opened, ==, NULL); ASSERT(list_is_empty(&zilog->zl_lwb_list)); zilog->zl_get_data = get_data; return (zilog); } /* * Close an intent log. */ void zil_close(zilog_t *zilog) { lwb_t *lwb; uint64_t txg; if (!dmu_objset_is_snapshot(zilog->zl_os)) { zil_commit(zilog, 0); } else { ASSERT3P(list_tail(&zilog->zl_lwb_list), ==, NULL); ASSERT0(zilog->zl_dirty_max_txg); ASSERT3B(zilog_is_dirty(zilog), ==, B_FALSE); } mutex_enter(&zilog->zl_lock); lwb = list_tail(&zilog->zl_lwb_list); if (lwb == NULL) txg = zilog->zl_dirty_max_txg; else txg = MAX(zilog->zl_dirty_max_txg, lwb->lwb_max_txg); mutex_exit(&zilog->zl_lock); /* * We need to use txg_wait_synced() to wait long enough for the * ZIL to be clean, and to wait for all pending lwbs to be * written out. */ if (txg != 0) txg_wait_synced(zilog->zl_dmu_pool, txg); if (zilog_is_dirty(zilog)) zfs_dbgmsg("zil (%p) is dirty, txg %llu", zilog, txg); VERIFY(!zilog_is_dirty(zilog)); zilog->zl_get_data = NULL; /* * We should have only one lwb left on the list; remove it now. */ mutex_enter(&zilog->zl_lock); lwb = list_head(&zilog->zl_lwb_list); if (lwb != NULL) { ASSERT3P(lwb, ==, list_tail(&zilog->zl_lwb_list)); ASSERT3S(lwb->lwb_state, !=, LWB_STATE_ISSUED); list_remove(&zilog->zl_lwb_list, lwb); zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); zil_free_lwb(zilog, lwb); } mutex_exit(&zilog->zl_lock); } static char *suspend_tag = "zil suspending"; /* * Suspend an intent log. While in suspended mode, we still honor * synchronous semantics, but we rely on txg_wait_synced() to do it. * On old version pools, we suspend the log briefly when taking a * snapshot so that it will have an empty intent log. * * Long holds are not really intended to be used the way we do here -- * held for such a short time. A concurrent caller of dsl_dataset_long_held() * could fail. Therefore we take pains to only put a long hold if it is * actually necessary. Fortunately, it will only be necessary if the * objset is currently mounted (or the ZVOL equivalent). In that case it * will already have a long hold, so we are not really making things any worse. * * Ideally, we would locate the existing long-holder (i.e. the zfsvfs_t or * zvol_state_t), and use their mechanism to prevent their hold from being * dropped (e.g. VFS_HOLD()). However, that would be even more pain for * very little gain. * * if cookiep == NULL, this does both the suspend & resume. * Otherwise, it returns with the dataset "long held", and the cookie * should be passed into zil_resume(). */ int zil_suspend(const char *osname, void **cookiep) { objset_t *os; zilog_t *zilog; const zil_header_t *zh; int error; error = dmu_objset_hold(osname, suspend_tag, &os); if (error != 0) return (error); zilog = dmu_objset_zil(os); mutex_enter(&zilog->zl_lock); zh = zilog->zl_header; if (zh->zh_flags & ZIL_REPLAY_NEEDED) { /* unplayed log */ mutex_exit(&zilog->zl_lock); dmu_objset_rele(os, suspend_tag); return (SET_ERROR(EBUSY)); } /* * Don't put a long hold in the cases where we can avoid it. This * is when there is no cookie so we are doing a suspend & resume * (i.e. called from zil_vdev_offline()), and there's nothing to do * for the suspend because it's already suspended, or there's no ZIL. */ if (cookiep == NULL && !zilog->zl_suspending && (zilog->zl_suspend > 0 || BP_IS_HOLE(&zh->zh_log))) { mutex_exit(&zilog->zl_lock); dmu_objset_rele(os, suspend_tag); return (0); } dsl_dataset_long_hold(dmu_objset_ds(os), suspend_tag); dsl_pool_rele(dmu_objset_pool(os), suspend_tag); zilog->zl_suspend++; if (zilog->zl_suspend > 1) { /* * Someone else is already suspending it. * Just wait for them to finish. */ while (zilog->zl_suspending) cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock); mutex_exit(&zilog->zl_lock); if (cookiep == NULL) zil_resume(os); else *cookiep = os; return (0); } /* * If there is no pointer to an on-disk block, this ZIL must not * be active (e.g. filesystem not mounted), so there's nothing * to clean up. */ if (BP_IS_HOLE(&zh->zh_log)) { ASSERT(cookiep != NULL); /* fast path already handled */ *cookiep = os; mutex_exit(&zilog->zl_lock); return (0); } zilog->zl_suspending = B_TRUE; mutex_exit(&zilog->zl_lock); /* * We need to use zil_commit_impl to ensure we wait for all * LWB_STATE_OPENED and LWB_STATE_ISSUED lwb's to be committed * to disk before proceeding. If we used zil_commit instead, it * would just call txg_wait_synced(), because zl_suspend is set. * txg_wait_synced() doesn't wait for these lwb's to be * LWB_STATE_DONE before returning. */ zil_commit_impl(zilog, 0); /* * Now that we've ensured all lwb's are LWB_STATE_DONE, we use * txg_wait_synced() to ensure the data from the zilog has * migrated to the main pool before calling zil_destroy(). */ txg_wait_synced(zilog->zl_dmu_pool, 0); zil_destroy(zilog, B_FALSE); mutex_enter(&zilog->zl_lock); zilog->zl_suspending = B_FALSE; cv_broadcast(&zilog->zl_cv_suspend); mutex_exit(&zilog->zl_lock); if (cookiep == NULL) zil_resume(os); else *cookiep = os; return (0); } void zil_resume(void *cookie) { objset_t *os = cookie; zilog_t *zilog = dmu_objset_zil(os); mutex_enter(&zilog->zl_lock); ASSERT(zilog->zl_suspend != 0); zilog->zl_suspend--; mutex_exit(&zilog->zl_lock); dsl_dataset_long_rele(dmu_objset_ds(os), suspend_tag); dsl_dataset_rele(dmu_objset_ds(os), suspend_tag); } typedef struct zil_replay_arg { zil_replay_func_t **zr_replay; void *zr_arg; boolean_t zr_byteswap; char *zr_lr; } zil_replay_arg_t; static int zil_replay_error(zilog_t *zilog, lr_t *lr, int error) { char name[ZFS_MAX_DATASET_NAME_LEN]; zilog->zl_replaying_seq--; /* didn't actually replay this one */ dmu_objset_name(zilog->zl_os, name); cmn_err(CE_WARN, "ZFS replay transaction error %d, " "dataset %s, seq 0x%llx, txtype %llu %s\n", error, name, (u_longlong_t)lr->lrc_seq, (u_longlong_t)(lr->lrc_txtype & ~TX_CI), (lr->lrc_txtype & TX_CI) ? "CI" : ""); return (error); } static int zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg) { zil_replay_arg_t *zr = zra; const zil_header_t *zh = zilog->zl_header; uint64_t reclen = lr->lrc_reclen; uint64_t txtype = lr->lrc_txtype; int error = 0; zilog->zl_replaying_seq = lr->lrc_seq; if (lr->lrc_seq <= zh->zh_replay_seq) /* already replayed */ return (0); if (lr->lrc_txg < claim_txg) /* already committed */ return (0); /* Strip case-insensitive bit, still present in log record */ txtype &= ~TX_CI; if (txtype == 0 || txtype >= TX_MAX_TYPE) return (zil_replay_error(zilog, lr, EINVAL)); /* * If this record type can be logged out of order, the object * (lr_foid) may no longer exist. That's legitimate, not an error. */ if (TX_OOO(txtype)) { error = dmu_object_info(zilog->zl_os, ((lr_ooo_t *)lr)->lr_foid, NULL); if (error == ENOENT || error == EEXIST) return (0); } /* * Make a copy of the data so we can revise and extend it. */ bcopy(lr, zr->zr_lr, reclen); /* * If this is a TX_WRITE with a blkptr, suck in the data. */ if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) { error = zil_read_log_data(zilog, (lr_write_t *)lr, zr->zr_lr + reclen); if (error != 0) return (zil_replay_error(zilog, lr, error)); } /* * The log block containing this lr may have been byteswapped * so that we can easily examine common fields like lrc_txtype. * However, the log is a mix of different record types, and only the * replay vectors know how to byteswap their records. Therefore, if * the lr was byteswapped, undo it before invoking the replay vector. */ if (zr->zr_byteswap) byteswap_uint64_array(zr->zr_lr, reclen); /* * We must now do two things atomically: replay this log record, * and update the log header sequence number to reflect the fact that * we did so. At the end of each replay function the sequence number * is updated if we are in replay mode. */ error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap); if (error != 0) { /* * The DMU's dnode layer doesn't see removes until the txg * commits, so a subsequent claim can spuriously fail with * EEXIST. So if we receive any error we try syncing out * any removes then retry the transaction. Note that we * specify B_FALSE for byteswap now, so we don't do it twice. */ txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0); error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE); if (error != 0) return (zil_replay_error(zilog, lr, error)); } return (0); } /* ARGSUSED */ static int zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) { zilog->zl_replay_blks++; return (0); } /* * If this dataset has a non-empty intent log, replay it and destroy it. */ void zil_replay(objset_t *os, void *arg, zil_replay_func_t *replay_func[TX_MAX_TYPE]) { zilog_t *zilog = dmu_objset_zil(os); const zil_header_t *zh = zilog->zl_header; zil_replay_arg_t zr; if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) { zil_destroy(zilog, B_TRUE); return; } zr.zr_replay = replay_func; zr.zr_arg = arg; zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log); zr.zr_lr = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP); /* * Wait for in-progress removes to sync before starting replay. */ txg_wait_synced(zilog->zl_dmu_pool, 0); zilog->zl_replay = B_TRUE; zilog->zl_replay_time = ddi_get_lbolt(); ASSERT(zilog->zl_replay_blks == 0); (void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr, zh->zh_claim_txg); kmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE); zil_destroy(zilog, B_FALSE); txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg); zilog->zl_replay = B_FALSE; } boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx) { if (zilog->zl_sync == ZFS_SYNC_DISABLED) return (B_TRUE); if (zilog->zl_replay) { dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx); zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] = zilog->zl_replaying_seq; return (B_TRUE); } return (B_FALSE); } /* ARGSUSED */ int zil_reset(const char *osname, void *arg) { int error; error = zil_suspend(osname, NULL); if (error != 0) return (SET_ERROR(EEXIST)); return (0); } Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c (revision 336948) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c (revision 336949) @@ -1,4246 +1,4270 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. - * Copyright (c) 2011, 2017 by Delphix. All rights reserved. + * Copyright (c) 2011, 2018 by Delphix. All rights reserved. * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2014 Integros [integros.com] */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include SYSCTL_DECL(_vfs_zfs); SYSCTL_NODE(_vfs_zfs, OID_AUTO, zio, CTLFLAG_RW, 0, "ZFS ZIO"); #if defined(__amd64__) static int zio_use_uma = 1; #else static int zio_use_uma = 0; #endif SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, use_uma, CTLFLAG_RDTUN, &zio_use_uma, 0, "Use uma(9) for ZIO allocations"); static int zio_exclude_metadata = 0; SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, exclude_metadata, CTLFLAG_RDTUN, &zio_exclude_metadata, 0, "Exclude metadata buffers from dumps as well"); zio_trim_stats_t zio_trim_stats = { { "bytes", KSTAT_DATA_UINT64, "Number of bytes successfully TRIMmed" }, { "success", KSTAT_DATA_UINT64, "Number of successful TRIM requests" }, { "unsupported", KSTAT_DATA_UINT64, "Number of TRIM requests that failed because TRIM is not supported" }, { "failed", KSTAT_DATA_UINT64, "Number of TRIM requests that failed for reasons other than not supported" }, }; static kstat_t *zio_trim_ksp; /* * ========================================================================== * I/O type descriptions * ========================================================================== */ const char *zio_type_name[ZIO_TYPES] = { "zio_null", "zio_read", "zio_write", "zio_free", "zio_claim", "zio_ioctl" }; boolean_t zio_dva_throttle_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, dva_throttle_enabled, CTLFLAG_RDTUN, &zio_dva_throttle_enabled, 0, ""); /* * ========================================================================== * I/O kmem caches * ========================================================================== */ kmem_cache_t *zio_cache; kmem_cache_t *zio_link_cache; kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; #ifdef _KERNEL extern vmem_t *zio_alloc_arena; #endif #define ZIO_PIPELINE_CONTINUE 0x100 #define ZIO_PIPELINE_STOP 0x101 #define BP_SPANB(indblkshift, level) \ (((uint64_t)1) << ((level) * ((indblkshift) - SPA_BLKPTRSHIFT))) #define COMPARE_META_LEVEL 0x80000000ul /* * The following actions directly effect the spa's sync-to-convergence logic. * The values below define the sync pass when we start performing the action. * Care should be taken when changing these values as they directly impact * spa_sync() performance. Tuning these values may introduce subtle performance * pathologies and should only be done in the context of performance analysis. * These tunables will eventually be removed and replaced with #defines once * enough analysis has been done to determine optimal values. * * The 'zfs_sync_pass_deferred_free' pass must be greater than 1 to ensure that * regular blocks are not deferred. */ int zfs_sync_pass_deferred_free = 2; /* defer frees starting in this pass */ SYSCTL_INT(_vfs_zfs, OID_AUTO, sync_pass_deferred_free, CTLFLAG_RDTUN, &zfs_sync_pass_deferred_free, 0, "defer frees starting in this pass"); int zfs_sync_pass_dont_compress = 5; /* don't compress starting in this pass */ SYSCTL_INT(_vfs_zfs, OID_AUTO, sync_pass_dont_compress, CTLFLAG_RDTUN, &zfs_sync_pass_dont_compress, 0, "don't compress starting in this pass"); int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */ SYSCTL_INT(_vfs_zfs, OID_AUTO, sync_pass_rewrite, CTLFLAG_RDTUN, &zfs_sync_pass_rewrite, 0, "rewrite new bps starting in this pass"); /* * An allocating zio is one that either currently has the DVA allocate * stage set or will have it later in its lifetime. */ #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE) boolean_t zio_requeue_io_start_cut_in_line = B_TRUE; #ifdef illumos #ifdef ZFS_DEBUG int zio_buf_debug_limit = 16384; #else int zio_buf_debug_limit = 0; #endif #endif static void zio_taskq_dispatch(zio_t *, zio_taskq_type_t, boolean_t); void zio_init(void) { size_t c; zio_cache = kmem_cache_create("zio_cache", sizeof (zio_t), 0, NULL, NULL, NULL, NULL, NULL, 0); zio_link_cache = kmem_cache_create("zio_link_cache", sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, 0); if (!zio_use_uma) goto out; /* * For small buffers, we want a cache for each multiple of * SPA_MINBLOCKSIZE. For larger buffers, we want a cache * for each quarter-power of 2. */ for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { size_t size = (c + 1) << SPA_MINBLOCKSHIFT; size_t p2 = size; size_t align = 0; int cflags = zio_exclude_metadata ? KMC_NODEBUG : 0; while (!ISP2(p2)) p2 &= p2 - 1; #ifdef illumos #ifndef _KERNEL /* * If we are using watchpoints, put each buffer on its own page, * to eliminate the performance overhead of trapping to the * kernel when modifying a non-watched buffer that shares the * page with a watched buffer. */ if (arc_watch && !IS_P2ALIGNED(size, PAGESIZE)) continue; #endif #endif /* illumos */ if (size <= 4 * SPA_MINBLOCKSIZE) { align = SPA_MINBLOCKSIZE; } else if (IS_P2ALIGNED(size, p2 >> 2)) { align = MIN(p2 >> 2, PAGESIZE); } if (align != 0) { char name[36]; (void) sprintf(name, "zio_buf_%lu", (ulong_t)size); zio_buf_cache[c] = kmem_cache_create(name, size, align, NULL, NULL, NULL, NULL, NULL, cflags); /* * Since zio_data bufs do not appear in crash dumps, we * pass KMC_NOTOUCH so that no allocator metadata is * stored with the buffers. */ (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size); zio_data_buf_cache[c] = kmem_cache_create(name, size, align, NULL, NULL, NULL, NULL, NULL, cflags | KMC_NOTOUCH | KMC_NODEBUG); } } while (--c != 0) { ASSERT(zio_buf_cache[c] != NULL); if (zio_buf_cache[c - 1] == NULL) zio_buf_cache[c - 1] = zio_buf_cache[c]; ASSERT(zio_data_buf_cache[c] != NULL); if (zio_data_buf_cache[c - 1] == NULL) zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; } out: zio_inject_init(); zio_trim_ksp = kstat_create("zfs", 0, "zio_trim", "misc", KSTAT_TYPE_NAMED, sizeof(zio_trim_stats) / sizeof(kstat_named_t), KSTAT_FLAG_VIRTUAL); if (zio_trim_ksp != NULL) { zio_trim_ksp->ks_data = &zio_trim_stats; kstat_install(zio_trim_ksp); } } void zio_fini(void) { size_t c; kmem_cache_t *last_cache = NULL; kmem_cache_t *last_data_cache = NULL; for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { if (zio_buf_cache[c] != last_cache) { last_cache = zio_buf_cache[c]; kmem_cache_destroy(zio_buf_cache[c]); } zio_buf_cache[c] = NULL; if (zio_data_buf_cache[c] != last_data_cache) { last_data_cache = zio_data_buf_cache[c]; kmem_cache_destroy(zio_data_buf_cache[c]); } zio_data_buf_cache[c] = NULL; } kmem_cache_destroy(zio_link_cache); kmem_cache_destroy(zio_cache); zio_inject_fini(); if (zio_trim_ksp != NULL) { kstat_delete(zio_trim_ksp); zio_trim_ksp = NULL; } } /* * ========================================================================== * Allocate and free I/O buffers * ========================================================================== */ /* * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a * crashdump if the kernel panics, so use it judiciously. Obviously, it's * useful to inspect ZFS metadata, but if possible, we should avoid keeping * excess / transient data in-core during a crashdump. */ void * zio_buf_alloc(size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; int flags = zio_exclude_metadata ? KM_NODEBUG : 0; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); if (zio_use_uma) return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE)); else return (kmem_alloc(size, KM_SLEEP|flags)); } /* * Use zio_data_buf_alloc to allocate data. The data will not appear in a * crashdump if the kernel panics. This exists so that we will limit the amount * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount * of kernel heap dumped to disk when the kernel panics) */ void * zio_data_buf_alloc(size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); if (zio_use_uma) return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE)); else return (kmem_alloc(size, KM_SLEEP | KM_NODEBUG)); } void zio_buf_free(void *buf, size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); if (zio_use_uma) kmem_cache_free(zio_buf_cache[c], buf); else kmem_free(buf, size); } void zio_data_buf_free(void *buf, size_t size) { size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); if (zio_use_uma) kmem_cache_free(zio_data_buf_cache[c], buf); else kmem_free(buf, size); } /* * ========================================================================== * Push and pop I/O transform buffers * ========================================================================== */ void zio_push_transform(zio_t *zio, abd_t *data, uint64_t size, uint64_t bufsize, zio_transform_func_t *transform) { zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP); /* * Ensure that anyone expecting this zio to contain a linear ABD isn't * going to get a nasty surprise when they try to access the data. */ #ifdef illumos IMPLY(abd_is_linear(zio->io_abd), abd_is_linear(data)); #else IMPLY(zio->io_abd != NULL && abd_is_linear(zio->io_abd), abd_is_linear(data)); #endif zt->zt_orig_abd = zio->io_abd; zt->zt_orig_size = zio->io_size; zt->zt_bufsize = bufsize; zt->zt_transform = transform; zt->zt_next = zio->io_transform_stack; zio->io_transform_stack = zt; zio->io_abd = data; zio->io_size = size; } void zio_pop_transforms(zio_t *zio) { zio_transform_t *zt; while ((zt = zio->io_transform_stack) != NULL) { if (zt->zt_transform != NULL) zt->zt_transform(zio, zt->zt_orig_abd, zt->zt_orig_size); if (zt->zt_bufsize != 0) abd_free(zio->io_abd); zio->io_abd = zt->zt_orig_abd; zio->io_size = zt->zt_orig_size; zio->io_transform_stack = zt->zt_next; kmem_free(zt, sizeof (zio_transform_t)); } } /* * ========================================================================== * I/O transform callbacks for subblocks and decompression * ========================================================================== */ static void zio_subblock(zio_t *zio, abd_t *data, uint64_t size) { ASSERT(zio->io_size > size); if (zio->io_type == ZIO_TYPE_READ) abd_copy(data, zio->io_abd, size); } static void zio_decompress(zio_t *zio, abd_t *data, uint64_t size) { if (zio->io_error == 0) { void *tmp = abd_borrow_buf(data, size); int ret = zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), zio->io_abd, tmp, zio->io_size, size); abd_return_buf_copy(data, tmp, size); if (ret != 0) zio->io_error = SET_ERROR(EIO); } } /* * ========================================================================== * I/O parent/child relationships and pipeline interlocks * ========================================================================== */ zio_t * zio_walk_parents(zio_t *cio, zio_link_t **zl) { list_t *pl = &cio->io_parent_list; *zl = (*zl == NULL) ? list_head(pl) : list_next(pl, *zl); if (*zl == NULL) return (NULL); ASSERT((*zl)->zl_child == cio); return ((*zl)->zl_parent); } zio_t * zio_walk_children(zio_t *pio, zio_link_t **zl) { list_t *cl = &pio->io_child_list; ASSERT(MUTEX_HELD(&pio->io_lock)); *zl = (*zl == NULL) ? list_head(cl) : list_next(cl, *zl); if (*zl == NULL) return (NULL); ASSERT((*zl)->zl_parent == pio); return ((*zl)->zl_child); } zio_t * zio_unique_parent(zio_t *cio) { zio_link_t *zl = NULL; zio_t *pio = zio_walk_parents(cio, &zl); VERIFY3P(zio_walk_parents(cio, &zl), ==, NULL); return (pio); } void zio_add_child(zio_t *pio, zio_t *cio) { zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP); /* * Logical I/Os can have logical, gang, or vdev children. * Gang I/Os can have gang or vdev children. * Vdev I/Os can only have vdev children. * The following ASSERT captures all of these constraints. */ ASSERT3S(cio->io_child_type, <=, pio->io_child_type); zl->zl_parent = pio; zl->zl_child = cio; mutex_enter(&pio->io_lock); mutex_enter(&cio->io_lock); ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); for (int w = 0; w < ZIO_WAIT_TYPES; w++) pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; list_insert_head(&pio->io_child_list, zl); list_insert_head(&cio->io_parent_list, zl); pio->io_child_count++; cio->io_parent_count++; mutex_exit(&cio->io_lock); mutex_exit(&pio->io_lock); } static void zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) { ASSERT(zl->zl_parent == pio); ASSERT(zl->zl_child == cio); mutex_enter(&pio->io_lock); mutex_enter(&cio->io_lock); list_remove(&pio->io_child_list, zl); list_remove(&cio->io_parent_list, zl); pio->io_child_count--; cio->io_parent_count--; mutex_exit(&cio->io_lock); mutex_exit(&pio->io_lock); kmem_cache_free(zio_link_cache, zl); } static boolean_t zio_wait_for_children(zio_t *zio, uint8_t childbits, enum zio_wait_type wait) { boolean_t waiting = B_FALSE; mutex_enter(&zio->io_lock); ASSERT(zio->io_stall == NULL); for (int c = 0; c < ZIO_CHILD_TYPES; c++) { if (!(ZIO_CHILD_BIT_IS_SET(childbits, c))) continue; uint64_t *countp = &zio->io_children[c][wait]; if (*countp != 0) { zio->io_stage >>= 1; ASSERT3U(zio->io_stage, !=, ZIO_STAGE_OPEN); zio->io_stall = countp; waiting = B_TRUE; break; } } mutex_exit(&zio->io_lock); return (waiting); } static void zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait) { uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; int *errorp = &pio->io_child_error[zio->io_child_type]; mutex_enter(&pio->io_lock); if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) *errorp = zio_worst_error(*errorp, zio->io_error); pio->io_reexecute |= zio->io_reexecute; ASSERT3U(*countp, >, 0); (*countp)--; if (*countp == 0 && pio->io_stall == countp) { zio_taskq_type_t type = pio->io_stage < ZIO_STAGE_VDEV_IO_START ? ZIO_TASKQ_ISSUE : ZIO_TASKQ_INTERRUPT; pio->io_stall = NULL; mutex_exit(&pio->io_lock); /* * Dispatch the parent zio in its own taskq so that * the child can continue to make progress. This also * prevents overflowing the stack when we have deeply nested * parent-child relationships. */ zio_taskq_dispatch(pio, type, B_FALSE); } else { mutex_exit(&pio->io_lock); } } static void zio_inherit_child_errors(zio_t *zio, enum zio_child c) { if (zio->io_child_error[c] != 0 && zio->io_error == 0) zio->io_error = zio->io_child_error[c]; } int zio_bookmark_compare(const void *x1, const void *x2) { const zio_t *z1 = x1; const zio_t *z2 = x2; if (z1->io_bookmark.zb_objset < z2->io_bookmark.zb_objset) return (-1); if (z1->io_bookmark.zb_objset > z2->io_bookmark.zb_objset) return (1); if (z1->io_bookmark.zb_object < z2->io_bookmark.zb_object) return (-1); if (z1->io_bookmark.zb_object > z2->io_bookmark.zb_object) return (1); if (z1->io_bookmark.zb_level < z2->io_bookmark.zb_level) return (-1); if (z1->io_bookmark.zb_level > z2->io_bookmark.zb_level) return (1); if (z1->io_bookmark.zb_blkid < z2->io_bookmark.zb_blkid) return (-1); if (z1->io_bookmark.zb_blkid > z2->io_bookmark.zb_blkid) return (1); if (z1 < z2) return (-1); if (z1 > z2) return (1); return (0); } /* * ========================================================================== * Create the various types of I/O (read, write, free, etc) * ========================================================================== */ static zio_t * zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, abd_t *data, uint64_t lsize, uint64_t psize, zio_done_func_t *done, void *private, zio_type_t type, zio_priority_t priority, enum zio_flag flags, vdev_t *vd, uint64_t offset, const zbookmark_phys_t *zb, enum zio_stage stage, enum zio_stage pipeline) { zio_t *zio; ASSERT3U(type == ZIO_TYPE_FREE || psize, <=, SPA_MAXBLOCKSIZE); ASSERT(P2PHASE(psize, SPA_MINBLOCKSIZE) == 0); ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); ASSERT(vd || stage == ZIO_STAGE_OPEN); IMPLY(lsize != psize, (flags & ZIO_FLAG_RAW) != 0); zio = kmem_cache_alloc(zio_cache, KM_SLEEP); bzero(zio, sizeof (zio_t)); mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL); cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL); list_create(&zio->io_parent_list, sizeof (zio_link_t), offsetof(zio_link_t, zl_parent_node)); list_create(&zio->io_child_list, sizeof (zio_link_t), offsetof(zio_link_t, zl_child_node)); metaslab_trace_init(&zio->io_alloc_list); if (vd != NULL) zio->io_child_type = ZIO_CHILD_VDEV; else if (flags & ZIO_FLAG_GANG_CHILD) zio->io_child_type = ZIO_CHILD_GANG; else if (flags & ZIO_FLAG_DDT_CHILD) zio->io_child_type = ZIO_CHILD_DDT; else zio->io_child_type = ZIO_CHILD_LOGICAL; if (bp != NULL) { zio->io_bp = (blkptr_t *)bp; zio->io_bp_copy = *bp; zio->io_bp_orig = *bp; if (type != ZIO_TYPE_WRITE || zio->io_child_type == ZIO_CHILD_DDT) zio->io_bp = &zio->io_bp_copy; /* so caller can free */ if (zio->io_child_type == ZIO_CHILD_LOGICAL) zio->io_logical = zio; if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) pipeline |= ZIO_GANG_STAGES; } zio->io_spa = spa; zio->io_txg = txg; zio->io_done = done; zio->io_private = private; zio->io_type = type; zio->io_priority = priority; zio->io_vd = vd; zio->io_offset = offset; zio->io_orig_abd = zio->io_abd = data; zio->io_orig_size = zio->io_size = psize; zio->io_lsize = lsize; zio->io_orig_flags = zio->io_flags = flags; zio->io_orig_stage = zio->io_stage = stage; zio->io_orig_pipeline = zio->io_pipeline = pipeline; zio->io_pipeline_trace = ZIO_STAGE_OPEN; zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); if (zb != NULL) zio->io_bookmark = *zb; if (pio != NULL) { if (zio->io_logical == NULL) zio->io_logical = pio->io_logical; if (zio->io_child_type == ZIO_CHILD_GANG) zio->io_gang_leader = pio->io_gang_leader; zio_add_child(pio, zio); } return (zio); } static void zio_destroy(zio_t *zio) { metaslab_trace_fini(&zio->io_alloc_list); list_destroy(&zio->io_parent_list); list_destroy(&zio->io_child_list); mutex_destroy(&zio->io_lock); cv_destroy(&zio->io_cv); kmem_cache_free(zio_cache, zio); } zio_t * zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, void *private, enum zio_flag flags) { zio_t *zio; zio = zio_create(pio, spa, 0, NULL, NULL, 0, 0, done, private, ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); return (zio); } zio_t * zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) { return (zio_null(NULL, spa, NULL, done, private, flags)); } void zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp) { if (!DMU_OT_IS_VALID(BP_GET_TYPE(bp))) { zfs_panic_recover("blkptr at %p has invalid TYPE %llu", bp, (longlong_t)BP_GET_TYPE(bp)); } if (BP_GET_CHECKSUM(bp) >= ZIO_CHECKSUM_FUNCTIONS || BP_GET_CHECKSUM(bp) <= ZIO_CHECKSUM_ON) { zfs_panic_recover("blkptr at %p has invalid CHECKSUM %llu", bp, (longlong_t)BP_GET_CHECKSUM(bp)); } if (BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_FUNCTIONS || BP_GET_COMPRESS(bp) <= ZIO_COMPRESS_ON) { zfs_panic_recover("blkptr at %p has invalid COMPRESS %llu", bp, (longlong_t)BP_GET_COMPRESS(bp)); } if (BP_GET_LSIZE(bp) > SPA_MAXBLOCKSIZE) { zfs_panic_recover("blkptr at %p has invalid LSIZE %llu", bp, (longlong_t)BP_GET_LSIZE(bp)); } if (BP_GET_PSIZE(bp) > SPA_MAXBLOCKSIZE) { zfs_panic_recover("blkptr at %p has invalid PSIZE %llu", bp, (longlong_t)BP_GET_PSIZE(bp)); } if (BP_IS_EMBEDDED(bp)) { if (BPE_GET_ETYPE(bp) > NUM_BP_EMBEDDED_TYPES) { zfs_panic_recover("blkptr at %p has invalid ETYPE %llu", bp, (longlong_t)BPE_GET_ETYPE(bp)); } } /* * Do not verify individual DVAs if the config is not trusted. This * will be done once the zio is executed in vdev_mirror_map_alloc. */ if (!spa->spa_trust_config) return; /* * Pool-specific checks. * * Note: it would be nice to verify that the blk_birth and * BP_PHYSICAL_BIRTH() are not too large. However, spa_freeze() * allows the birth time of log blocks (and dmu_sync()-ed blocks * that are in the log) to be arbitrarily large. */ for (int i = 0; i < BP_GET_NDVAS(bp); i++) { uint64_t vdevid = DVA_GET_VDEV(&bp->blk_dva[i]); if (vdevid >= spa->spa_root_vdev->vdev_children) { zfs_panic_recover("blkptr at %p DVA %u has invalid " "VDEV %llu", bp, i, (longlong_t)vdevid); continue; } vdev_t *vd = spa->spa_root_vdev->vdev_child[vdevid]; if (vd == NULL) { zfs_panic_recover("blkptr at %p DVA %u has invalid " "VDEV %llu", bp, i, (longlong_t)vdevid); continue; } if (vd->vdev_ops == &vdev_hole_ops) { zfs_panic_recover("blkptr at %p DVA %u has hole " "VDEV %llu", bp, i, (longlong_t)vdevid); continue; } if (vd->vdev_ops == &vdev_missing_ops) { /* * "missing" vdevs are valid during import, but we * don't have their detailed info (e.g. asize), so * we can't perform any more checks on them. */ continue; } uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]); uint64_t asize = DVA_GET_ASIZE(&bp->blk_dva[i]); if (BP_IS_GANG(bp)) asize = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); if (offset + asize > vd->vdev_asize) { zfs_panic_recover("blkptr at %p DVA %u has invalid " "OFFSET %llu", bp, i, (longlong_t)offset); } } } boolean_t zfs_dva_valid(spa_t *spa, const dva_t *dva, const blkptr_t *bp) { uint64_t vdevid = DVA_GET_VDEV(dva); if (vdevid >= spa->spa_root_vdev->vdev_children) return (B_FALSE); vdev_t *vd = spa->spa_root_vdev->vdev_child[vdevid]; if (vd == NULL) return (B_FALSE); if (vd->vdev_ops == &vdev_hole_ops) return (B_FALSE); if (vd->vdev_ops == &vdev_missing_ops) { return (B_FALSE); } uint64_t offset = DVA_GET_OFFSET(dva); uint64_t asize = DVA_GET_ASIZE(dva); if (BP_IS_GANG(bp)) asize = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE); if (offset + asize > vd->vdev_asize) return (B_FALSE); return (B_TRUE); } zio_t * zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, abd_t *data, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) { zio_t *zio; zfs_blkptr_verify(spa, bp); zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, data, size, size, done, private, ZIO_TYPE_READ, priority, flags, NULL, 0, zb, ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); return (zio); } zio_t * zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, abd_t *data, uint64_t lsize, uint64_t psize, const zio_prop_t *zp, zio_done_func_t *ready, zio_done_func_t *children_ready, zio_done_func_t *physdone, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) { zio_t *zio; ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && zp->zp_compress >= ZIO_COMPRESS_OFF && zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && DMU_OT_IS_VALID(zp->zp_type) && zp->zp_level < 32 && zp->zp_copies > 0 && zp->zp_copies <= spa_max_replication(spa)); zio = zio_create(pio, spa, txg, bp, data, lsize, psize, done, private, ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); zio->io_ready = ready; zio->io_children_ready = children_ready; zio->io_physdone = physdone; zio->io_prop = *zp; /* * Data can be NULL if we are going to call zio_write_override() to * provide the already-allocated BP. But we may need the data to * verify a dedup hit (if requested). In this case, don't try to * dedup (just take the already-allocated BP verbatim). */ if (data == NULL && zio->io_prop.zp_dedup_verify) { zio->io_prop.zp_dedup = zio->io_prop.zp_dedup_verify = B_FALSE; } return (zio); } zio_t * zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, abd_t *data, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb) { zio_t *zio; zio = zio_create(pio, spa, txg, bp, data, size, size, done, private, ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_IO_REWRITE, NULL, 0, zb, ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); return (zio); } void zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite) { ASSERT(zio->io_type == ZIO_TYPE_WRITE); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(zio->io_stage == ZIO_STAGE_OPEN); ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); /* * We must reset the io_prop to match the values that existed * when the bp was first written by dmu_sync() keeping in mind * that nopwrite and dedup are mutually exclusive. */ zio->io_prop.zp_dedup = nopwrite ? B_FALSE : zio->io_prop.zp_dedup; zio->io_prop.zp_nopwrite = nopwrite; zio->io_prop.zp_copies = copies; zio->io_bp_override = bp; } void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) { zfs_blkptr_verify(spa, bp); /* * The check for EMBEDDED is a performance optimization. We * process the free here (by ignoring it) rather than * putting it on the list and then processing it in zio_free_sync(). */ if (BP_IS_EMBEDDED(bp)) return; metaslab_check_free(spa, bp); /* * Frees that are for the currently-syncing txg, are not going to be * deferred, and which will not need to do a read (i.e. not GANG or * DEDUP), can be processed immediately. Otherwise, put them on the * in-memory list for later processing. */ if (zfs_trim_enabled || BP_IS_GANG(bp) || BP_GET_DEDUP(bp) || txg != spa->spa_syncing_txg || spa_sync_pass(spa) >= zfs_sync_pass_deferred_free) { bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); } else { VERIFY0(zio_wait(zio_free_sync(NULL, spa, txg, bp, BP_GET_PSIZE(bp), 0))); } } zio_t * zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, uint64_t size, enum zio_flag flags) { zio_t *zio; enum zio_stage stage = ZIO_FREE_PIPELINE; ASSERT(!BP_IS_HOLE(bp)); ASSERT(spa_syncing_txg(spa) == txg); ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free); if (BP_IS_EMBEDDED(bp)) return (zio_null(pio, spa, NULL, NULL, NULL, 0)); metaslab_check_free(spa, bp); arc_freed(spa, bp); dsl_scan_freed(spa, bp); if (zfs_trim_enabled) stage |= ZIO_STAGE_ISSUE_ASYNC | ZIO_STAGE_VDEV_IO_START | ZIO_STAGE_VDEV_IO_ASSESS; /* * GANG and DEDUP blocks can induce a read (for the gang block header, * or the DDT), so issue them asynchronously so that this thread is * not tied up. */ else if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp)) stage |= ZIO_STAGE_ISSUE_ASYNC; flags |= ZIO_FLAG_DONT_QUEUE; zio = zio_create(pio, spa, txg, bp, NULL, size, size, NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_NOW, flags, NULL, 0, NULL, ZIO_STAGE_OPEN, stage); return (zio); } zio_t * zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, zio_done_func_t *done, void *private, enum zio_flag flags) { zio_t *zio; zfs_blkptr_verify(spa, bp); if (BP_IS_EMBEDDED(bp)) return (zio_null(pio, spa, NULL, NULL, NULL, 0)); /* * A claim is an allocation of a specific block. Claims are needed * to support immediate writes in the intent log. The issue is that * immediate writes contain committed data, but in a txg that was * *not* committed. Upon opening the pool after an unclean shutdown, * the intent log claims all blocks that contain immediate write data * so that the SPA knows they're in use. * * All claims *must* be resolved in the first txg -- before the SPA * starts allocating blocks -- so that nothing is allocated twice. * If txg == 0 we just verify that the block is claimable. */ ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_min_claim_txg(spa)); ASSERT(txg == spa_min_claim_txg(spa) || txg == 0); ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */ zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), BP_GET_PSIZE(bp), done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); ASSERT0(zio->io_queued_timestamp); return (zio); } zio_t * zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, uint64_t offset, uint64_t size, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags) { zio_t *zio; int c; if (vd->vdev_children == 0) { zio = zio_create(pio, spa, 0, NULL, NULL, 0, 0, done, private, ZIO_TYPE_IOCTL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); zio->io_cmd = cmd; } else { zio = zio_null(pio, spa, NULL, NULL, NULL, flags); for (c = 0; c < vd->vdev_children; c++) zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, offset, size, done, private, priority, flags)); } return (zio); } zio_t * zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, abd_t *data, int checksum, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, boolean_t labels) { zio_t *zio; ASSERT(vd->vdev_children == 0); ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); ASSERT3U(offset + size, <=, vd->vdev_psize); zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, size, done, private, ZIO_TYPE_READ, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); zio->io_prop.zp_checksum = checksum; return (zio); } zio_t * zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, abd_t *data, int checksum, zio_done_func_t *done, void *private, zio_priority_t priority, enum zio_flag flags, boolean_t labels) { zio_t *zio; ASSERT(vd->vdev_children == 0); ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); ASSERT3U(offset + size, <=, vd->vdev_psize); zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, size, done, private, ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); zio->io_prop.zp_checksum = checksum; if (zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_EMBEDDED) { /* * zec checksums are necessarily destructive -- they modify * the end of the write buffer to hold the verifier/checksum. * Therefore, we must make a local copy in case the data is * being written to multiple places in parallel. */ abd_t *wbuf = abd_alloc_sametype(data, size); abd_copy(wbuf, data, size); zio_push_transform(zio, wbuf, size, size, NULL); } return (zio); } /* * Create a child I/O to do some work for us. */ zio_t * zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, abd_t *data, uint64_t size, int type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *private) { enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; zio_t *zio; /* * vdev child I/Os do not propagate their error to the parent. * Therefore, for correct operation the caller *must* check for * and handle the error in the child i/o's done callback. * The only exceptions are i/os that we don't care about * (OPTIONAL or REPAIR). */ ASSERT((flags & ZIO_FLAG_OPTIONAL) || (flags & ZIO_FLAG_IO_REPAIR) || done != NULL); /* * In the common case, where the parent zio was to a normal vdev, * the child zio must be to a child vdev of that vdev. Otherwise, * the child zio must be to a top-level vdev. */ if (pio->io_vd != NULL && pio->io_vd->vdev_ops != &vdev_indirect_ops) { ASSERT3P(vd->vdev_parent, ==, pio->io_vd); } else { ASSERT3P(vd, ==, vd->vdev_top); } if (type == ZIO_TYPE_READ && bp != NULL) { /* * If we have the bp, then the child should perform the * checksum and the parent need not. This pushes error * detection as close to the leaves as possible and * eliminates redundant checksums in the interior nodes. */ pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; } /* Not all IO types require vdev io done stage e.g. free */ if (!(pio->io_pipeline & ZIO_STAGE_VDEV_IO_DONE)) pipeline &= ~ZIO_STAGE_VDEV_IO_DONE; if (vd->vdev_ops->vdev_op_leaf) { ASSERT0(vd->vdev_children); offset += VDEV_LABEL_START_SIZE; } flags |= ZIO_VDEV_CHILD_FLAGS(pio); /* * If we've decided to do a repair, the write is not speculative -- * even if the original read was. */ if (flags & ZIO_FLAG_IO_REPAIR) flags &= ~ZIO_FLAG_SPECULATIVE; /* * If we're creating a child I/O that is not associated with a * top-level vdev, then the child zio is not an allocating I/O. * If this is a retried I/O then we ignore it since we will * have already processed the original allocating I/O. */ if (flags & ZIO_FLAG_IO_ALLOCATING && (vd != vd->vdev_top || (flags & ZIO_FLAG_IO_RETRY))) { metaslab_class_t *mc = spa_normal_class(pio->io_spa); ASSERT(mc->mc_alloc_throttle_enabled); ASSERT(type == ZIO_TYPE_WRITE); ASSERT(priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(!(flags & ZIO_FLAG_IO_REPAIR)); ASSERT(!(pio->io_flags & ZIO_FLAG_IO_REWRITE) || pio->io_child_type == ZIO_CHILD_GANG); flags &= ~ZIO_FLAG_IO_ALLOCATING; } zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, size, done, private, type, priority, flags, vd, offset, &pio->io_bookmark, ZIO_STAGE_VDEV_IO_START >> 1, pipeline); ASSERT3U(zio->io_child_type, ==, ZIO_CHILD_VDEV); zio->io_physdone = pio->io_physdone; if (vd->vdev_ops->vdev_op_leaf && zio->io_logical != NULL) zio->io_logical->io_phys_children++; return (zio); } zio_t * zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, abd_t *data, uint64_t size, int type, zio_priority_t priority, enum zio_flag flags, zio_done_func_t *done, void *private) { zio_t *zio; ASSERT(vd->vdev_ops->vdev_op_leaf); zio = zio_create(NULL, vd->vdev_spa, 0, NULL, data, size, size, done, private, type, priority, flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY | ZIO_FLAG_DELEGATED, vd, offset, NULL, ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); return (zio); } void zio_flush(zio_t *zio, vdev_t *vd) { zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, 0, 0, NULL, NULL, ZIO_PRIORITY_NOW, ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); } zio_t * zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset, uint64_t size) { ASSERT(vd->vdev_ops->vdev_op_leaf); return (zio_create(zio, spa, 0, NULL, NULL, size, size, NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_TRIM, ZIO_FLAG_DONT_AGGREGATE | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY, vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PHYS_PIPELINE)); } void zio_shrink(zio_t *zio, uint64_t size) { ASSERT3P(zio->io_executor, ==, NULL); ASSERT3P(zio->io_orig_size, ==, zio->io_size); ASSERT3U(size, <=, zio->io_size); /* * We don't shrink for raidz because of problems with the * reconstruction when reading back less than the block size. * Note, BP_IS_RAIDZ() assumes no compression. */ ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); if (!BP_IS_RAIDZ(zio->io_bp)) { /* we are not doing a raw write */ ASSERT3U(zio->io_size, ==, zio->io_lsize); zio->io_orig_size = zio->io_size = zio->io_lsize = size; } } /* * ========================================================================== * Prepare to read and write logical blocks * ========================================================================== */ static int zio_read_bp_init(zio_t *zio) { blkptr_t *bp = zio->io_bp; ASSERT3P(zio->io_bp, ==, &zio->io_bp_copy); if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && zio->io_child_type == ZIO_CHILD_LOGICAL && !(zio->io_flags & ZIO_FLAG_RAW)) { uint64_t psize = BP_IS_EMBEDDED(bp) ? BPE_GET_PSIZE(bp) : BP_GET_PSIZE(bp); zio_push_transform(zio, abd_alloc_sametype(zio->io_abd, psize), psize, psize, zio_decompress); } if (BP_IS_EMBEDDED(bp) && BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA) { zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; int psize = BPE_GET_PSIZE(bp); void *data = abd_borrow_buf(zio->io_abd, psize); decode_embedded_bp_compressed(bp, data); abd_return_buf_copy(zio->io_abd, data, psize); } else { ASSERT(!BP_IS_EMBEDDED(bp)); ASSERT3P(zio->io_bp, ==, &zio->io_bp_copy); } if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0) zio->io_flags |= ZIO_FLAG_DONT_CACHE; if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) zio->io_flags |= ZIO_FLAG_DONT_CACHE; if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) zio->io_pipeline = ZIO_DDT_READ_PIPELINE; return (ZIO_PIPELINE_CONTINUE); } static int zio_write_bp_init(zio_t *zio) { if (!IO_IS_ALLOCATING(zio)) return (ZIO_PIPELINE_CONTINUE); ASSERT(zio->io_child_type != ZIO_CHILD_DDT); if (zio->io_bp_override) { blkptr_t *bp = zio->io_bp; zio_prop_t *zp = &zio->io_prop; ASSERT(bp->blk_birth != zio->io_txg); ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); *bp = *zio->io_bp_override; zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; if (BP_IS_EMBEDDED(bp)) return (ZIO_PIPELINE_CONTINUE); /* * If we've been overridden and nopwrite is set then * set the flag accordingly to indicate that a nopwrite * has already occurred. */ if (!BP_IS_HOLE(bp) && zp->zp_nopwrite) { ASSERT(!zp->zp_dedup); ASSERT3U(BP_GET_CHECKSUM(bp), ==, zp->zp_checksum); zio->io_flags |= ZIO_FLAG_NOPWRITE; return (ZIO_PIPELINE_CONTINUE); } ASSERT(!zp->zp_nopwrite); if (BP_IS_HOLE(bp) || !zp->zp_dedup) return (ZIO_PIPELINE_CONTINUE); ASSERT((zio_checksum_table[zp->zp_checksum].ci_flags & ZCHECKSUM_FLAG_DEDUP) || zp->zp_dedup_verify); if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) { BP_SET_DEDUP(bp, 1); zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; return (ZIO_PIPELINE_CONTINUE); } /* * We were unable to handle this as an override bp, treat * it as a regular write I/O. */ zio->io_bp_override = NULL; *bp = zio->io_bp_orig; zio->io_pipeline = zio->io_orig_pipeline; } return (ZIO_PIPELINE_CONTINUE); } static int zio_write_compress(zio_t *zio) { spa_t *spa = zio->io_spa; zio_prop_t *zp = &zio->io_prop; enum zio_compress compress = zp->zp_compress; blkptr_t *bp = zio->io_bp; uint64_t lsize = zio->io_lsize; uint64_t psize = zio->io_size; int pass = 1; EQUIV(lsize != psize, (zio->io_flags & ZIO_FLAG_RAW) != 0); /* * If our children haven't all reached the ready stage, * wait for them and then repeat this pipeline stage. */ if (zio_wait_for_children(zio, ZIO_CHILD_LOGICAL_BIT | ZIO_CHILD_GANG_BIT, ZIO_WAIT_READY)) { return (ZIO_PIPELINE_STOP); } if (!IO_IS_ALLOCATING(zio)) return (ZIO_PIPELINE_CONTINUE); if (zio->io_children_ready != NULL) { /* * Now that all our children are ready, run the callback * associated with this zio in case it wants to modify the * data to be written. */ ASSERT3U(zp->zp_level, >, 0); zio->io_children_ready(zio); } ASSERT(zio->io_child_type != ZIO_CHILD_DDT); ASSERT(zio->io_bp_override == NULL); if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg) { /* * We're rewriting an existing block, which means we're * working on behalf of spa_sync(). For spa_sync() to * converge, it must eventually be the case that we don't * have to allocate new blocks. But compression changes * the blocksize, which forces a reallocate, and makes * convergence take longer. Therefore, after the first * few passes, stop compressing to ensure convergence. */ pass = spa_sync_pass(spa); ASSERT(zio->io_txg == spa_syncing_txg(spa)); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(!BP_GET_DEDUP(bp)); if (pass >= zfs_sync_pass_dont_compress) compress = ZIO_COMPRESS_OFF; /* Make sure someone doesn't change their mind on overwrites */ ASSERT(BP_IS_EMBEDDED(bp) || MIN(zp->zp_copies + BP_IS_GANG(bp), spa_max_replication(spa)) == BP_GET_NDVAS(bp)); } /* If it's a compressed write that is not raw, compress the buffer. */ if (compress != ZIO_COMPRESS_OFF && psize == lsize) { void *cbuf = zio_buf_alloc(lsize); psize = zio_compress_data(compress, zio->io_abd, cbuf, lsize); if (psize == 0 || psize == lsize) { compress = ZIO_COMPRESS_OFF; zio_buf_free(cbuf, lsize); } else if (!zp->zp_dedup && psize <= BPE_PAYLOAD_SIZE && zp->zp_level == 0 && !DMU_OT_HAS_FILL(zp->zp_type) && spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) { encode_embedded_bp_compressed(bp, cbuf, compress, lsize, psize); BPE_SET_ETYPE(bp, BP_EMBEDDED_TYPE_DATA); BP_SET_TYPE(bp, zio->io_prop.zp_type); BP_SET_LEVEL(bp, zio->io_prop.zp_level); zio_buf_free(cbuf, lsize); bp->blk_birth = zio->io_txg; zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; ASSERT(spa_feature_is_active(spa, SPA_FEATURE_EMBEDDED_DATA)); return (ZIO_PIPELINE_CONTINUE); } else { /* * Round up compressed size up to the ashift * of the smallest-ashift device, and zero the tail. * This ensures that the compressed size of the BP * (and thus compressratio property) are correct, * in that we charge for the padding used to fill out * the last sector. */ ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); size_t rounded = (size_t)P2ROUNDUP(psize, 1ULL << spa->spa_min_ashift); if (rounded >= lsize) { compress = ZIO_COMPRESS_OFF; zio_buf_free(cbuf, lsize); psize = lsize; } else { abd_t *cdata = abd_get_from_buf(cbuf, lsize); abd_take_ownership_of_buf(cdata, B_TRUE); abd_zero_off(cdata, psize, rounded - psize); psize = rounded; zio_push_transform(zio, cdata, psize, lsize, NULL); } } /* * We were unable to handle this as an override bp, treat * it as a regular write I/O. */ zio->io_bp_override = NULL; *bp = zio->io_bp_orig; zio->io_pipeline = zio->io_orig_pipeline; } else { ASSERT3U(psize, !=, 0); } /* * The final pass of spa_sync() must be all rewrites, but the first * few passes offer a trade-off: allocating blocks defers convergence, * but newly allocated blocks are sequential, so they can be written * to disk faster. Therefore, we allow the first few passes of * spa_sync() to allocate new blocks, but force rewrites after that. * There should only be a handful of blocks after pass 1 in any case. */ if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize && pass >= zfs_sync_pass_rewrite) { ASSERT(psize != 0); enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; zio->io_flags |= ZIO_FLAG_IO_REWRITE; } else { BP_ZERO(bp); zio->io_pipeline = ZIO_WRITE_PIPELINE; } if (psize == 0) { if (zio->io_bp_orig.blk_birth != 0 && spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) { BP_SET_LSIZE(bp, lsize); BP_SET_TYPE(bp, zp->zp_type); BP_SET_LEVEL(bp, zp->zp_level); BP_SET_BIRTH(bp, zio->io_txg, 0); } zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; } else { ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); BP_SET_LSIZE(bp, lsize); BP_SET_TYPE(bp, zp->zp_type); BP_SET_LEVEL(bp, zp->zp_level); BP_SET_PSIZE(bp, psize); BP_SET_COMPRESS(bp, compress); BP_SET_CHECKSUM(bp, zp->zp_checksum); BP_SET_DEDUP(bp, zp->zp_dedup); BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); if (zp->zp_dedup) { ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; } if (zp->zp_nopwrite) { ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); zio->io_pipeline |= ZIO_STAGE_NOP_WRITE; } } return (ZIO_PIPELINE_CONTINUE); } static int zio_free_bp_init(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (zio->io_child_type == ZIO_CHILD_LOGICAL) { if (BP_GET_DEDUP(bp)) zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; } ASSERT3P(zio->io_bp, ==, &zio->io_bp_copy); return (ZIO_PIPELINE_CONTINUE); } /* * ========================================================================== * Execute the I/O pipeline * ========================================================================== */ static void zio_taskq_dispatch(zio_t *zio, zio_taskq_type_t q, boolean_t cutinline) { spa_t *spa = zio->io_spa; zio_type_t t = zio->io_type; int flags = (cutinline ? TQ_FRONT : 0); ASSERT(q == ZIO_TASKQ_ISSUE || q == ZIO_TASKQ_INTERRUPT); /* * If we're a config writer or a probe, the normal issue and * interrupt threads may all be blocked waiting for the config lock. * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. */ if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) t = ZIO_TYPE_NULL; /* * A similar issue exists for the L2ARC write thread until L2ARC 2.0. */ if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) t = ZIO_TYPE_NULL; /* * If this is a high priority I/O, then use the high priority taskq if * available. */ if (zio->io_priority == ZIO_PRIORITY_NOW && spa->spa_zio_taskq[t][q + 1].stqs_count != 0) q++; ASSERT3U(q, <, ZIO_TASKQ_TYPES); /* * NB: We are assuming that the zio can only be dispatched * to a single taskq at a time. It would be a grievous error * to dispatch the zio to another taskq at the same time. */ #if defined(illumos) || !defined(_KERNEL) ASSERT(zio->io_tqent.tqent_next == NULL); #else ASSERT(zio->io_tqent.tqent_task.ta_pending == 0); #endif spa_taskq_dispatch_ent(spa, t, q, (task_func_t *)zio_execute, zio, flags, &zio->io_tqent); } static boolean_t zio_taskq_member(zio_t *zio, zio_taskq_type_t q) { kthread_t *executor = zio->io_executor; spa_t *spa = zio->io_spa; for (zio_type_t t = 0; t < ZIO_TYPES; t++) { spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; uint_t i; for (i = 0; i < tqs->stqs_count; i++) { if (taskq_member(tqs->stqs_taskq[i], executor)) return (B_TRUE); } } return (B_FALSE); } static int zio_issue_async(zio_t *zio) { zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); return (ZIO_PIPELINE_STOP); } void zio_interrupt(zio_t *zio) { zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); } void zio_delay_interrupt(zio_t *zio) { /* * The timeout_generic() function isn't defined in userspace, so * rather than trying to implement the function, the zio delay * functionality has been disabled for userspace builds. */ #ifdef _KERNEL /* * If io_target_timestamp is zero, then no delay has been registered * for this IO, thus jump to the end of this function and "skip" the * delay; issuing it directly to the zio layer. */ if (zio->io_target_timestamp != 0) { hrtime_t now = gethrtime(); if (now >= zio->io_target_timestamp) { /* * This IO has already taken longer than the target * delay to complete, so we don't want to delay it * any longer; we "miss" the delay and issue it * directly to the zio layer. This is likely due to * the target latency being set to a value less than * the underlying hardware can satisfy (e.g. delay * set to 1ms, but the disks take 10ms to complete an * IO request). */ DTRACE_PROBE2(zio__delay__miss, zio_t *, zio, hrtime_t, now); zio_interrupt(zio); } else { hrtime_t diff = zio->io_target_timestamp - now; DTRACE_PROBE3(zio__delay__hit, zio_t *, zio, hrtime_t, now, hrtime_t, diff); (void) timeout_generic(CALLOUT_NORMAL, (void (*)(void *))zio_interrupt, zio, diff, 1, 0); } return; } #endif DTRACE_PROBE1(zio__delay__skip, zio_t *, zio); zio_interrupt(zio); } /* * Execute the I/O pipeline until one of the following occurs: * * (1) the I/O completes * (2) the pipeline stalls waiting for dependent child I/Os * (3) the I/O issues, so we're waiting for an I/O completion interrupt * (4) the I/O is delegated by vdev-level caching or aggregation * (5) the I/O is deferred due to vdev-level queueing * (6) the I/O is handed off to another thread. * * In all cases, the pipeline stops whenever there's no CPU work; it never * burns a thread in cv_wait(). * * There's no locking on io_stage because there's no legitimate way * for multiple threads to be attempting to process the same I/O. */ static zio_pipe_stage_t *zio_pipeline[]; void zio_execute(zio_t *zio) { zio->io_executor = curthread; ASSERT3U(zio->io_queued_timestamp, >, 0); while (zio->io_stage < ZIO_STAGE_DONE) { enum zio_stage pipeline = zio->io_pipeline; enum zio_stage stage = zio->io_stage; int rv; ASSERT(!MUTEX_HELD(&zio->io_lock)); ASSERT(ISP2(stage)); ASSERT(zio->io_stall == NULL); do { stage <<= 1; } while ((stage & pipeline) == 0); ASSERT(stage <= ZIO_STAGE_DONE); /* * If we are in interrupt context and this pipeline stage * will grab a config lock that is held across I/O, * or may wait for an I/O that needs an interrupt thread * to complete, issue async to avoid deadlock. * * For VDEV_IO_START, we cut in line so that the io will * be sent to disk promptly. */ if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ? zio_requeue_io_start_cut_in_line : B_FALSE; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); return; } zio->io_stage = stage; zio->io_pipeline_trace |= zio->io_stage; rv = zio_pipeline[highbit64(stage) - 1](zio); if (rv == ZIO_PIPELINE_STOP) return; ASSERT(rv == ZIO_PIPELINE_CONTINUE); } } /* * ========================================================================== * Initiate I/O, either sync or async * ========================================================================== */ int zio_wait(zio_t *zio) { int error; ASSERT3P(zio->io_stage, ==, ZIO_STAGE_OPEN); ASSERT3P(zio->io_executor, ==, NULL); zio->io_waiter = curthread; ASSERT0(zio->io_queued_timestamp); zio->io_queued_timestamp = gethrtime(); zio_execute(zio); mutex_enter(&zio->io_lock); while (zio->io_executor != NULL) cv_wait(&zio->io_cv, &zio->io_lock); mutex_exit(&zio->io_lock); error = zio->io_error; zio_destroy(zio); return (error); } void zio_nowait(zio_t *zio) { ASSERT3P(zio->io_executor, ==, NULL); if (zio->io_child_type == ZIO_CHILD_LOGICAL && zio_unique_parent(zio) == NULL) { /* * This is a logical async I/O with no parent to wait for it. * We add it to the spa_async_root_zio "Godfather" I/O which * will ensure they complete prior to unloading the pool. */ spa_t *spa = zio->io_spa; zio_add_child(spa->spa_async_zio_root[CPU_SEQID], zio); } ASSERT0(zio->io_queued_timestamp); zio->io_queued_timestamp = gethrtime(); zio_execute(zio); } /* * ========================================================================== * Reexecute, cancel, or suspend/resume failed I/O * ========================================================================== */ static void zio_reexecute(zio_t *pio) { zio_t *cio, *cio_next; ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); ASSERT(pio->io_gang_leader == NULL); ASSERT(pio->io_gang_tree == NULL); pio->io_flags = pio->io_orig_flags; pio->io_stage = pio->io_orig_stage; pio->io_pipeline = pio->io_orig_pipeline; pio->io_reexecute = 0; pio->io_flags |= ZIO_FLAG_REEXECUTED; pio->io_pipeline_trace = 0; pio->io_error = 0; for (int w = 0; w < ZIO_WAIT_TYPES; w++) pio->io_state[w] = 0; for (int c = 0; c < ZIO_CHILD_TYPES; c++) pio->io_child_error[c] = 0; if (IO_IS_ALLOCATING(pio)) BP_ZERO(pio->io_bp); /* * As we reexecute pio's children, new children could be created. * New children go to the head of pio's io_child_list, however, * so we will (correctly) not reexecute them. The key is that * the remainder of pio's io_child_list, from 'cio_next' onward, * cannot be affected by any side effects of reexecuting 'cio'. */ zio_link_t *zl = NULL; mutex_enter(&pio->io_lock); for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) { cio_next = zio_walk_children(pio, &zl); for (int w = 0; w < ZIO_WAIT_TYPES; w++) pio->io_children[cio->io_child_type][w]++; mutex_exit(&pio->io_lock); zio_reexecute(cio); mutex_enter(&pio->io_lock); } mutex_exit(&pio->io_lock); /* * Now that all children have been reexecuted, execute the parent. * We don't reexecute "The Godfather" I/O here as it's the * responsibility of the caller to wait on it. */ if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) { pio->io_queued_timestamp = gethrtime(); zio_execute(pio); } } void zio_suspend(spa_t *spa, zio_t *zio) { if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) fm_panic("Pool '%s' has encountered an uncorrectable I/O " "failure and the failure mode property for this pool " "is set to panic.", spa_name(spa)); zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0); mutex_enter(&spa->spa_suspend_lock); if (spa->spa_suspend_zio_root == NULL) spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); spa->spa_suspended = B_TRUE; if (zio != NULL) { ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); ASSERT(zio != spa->spa_suspend_zio_root); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ASSERT(zio_unique_parent(zio) == NULL); ASSERT(zio->io_stage == ZIO_STAGE_DONE); zio_add_child(spa->spa_suspend_zio_root, zio); } mutex_exit(&spa->spa_suspend_lock); } int zio_resume(spa_t *spa) { zio_t *pio; /* * Reexecute all previously suspended i/o. */ mutex_enter(&spa->spa_suspend_lock); spa->spa_suspended = B_FALSE; cv_broadcast(&spa->spa_suspend_cv); pio = spa->spa_suspend_zio_root; spa->spa_suspend_zio_root = NULL; mutex_exit(&spa->spa_suspend_lock); if (pio == NULL) return (0); zio_reexecute(pio); return (zio_wait(pio)); } void zio_resume_wait(spa_t *spa) { mutex_enter(&spa->spa_suspend_lock); while (spa_suspended(spa)) cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); mutex_exit(&spa->spa_suspend_lock); } /* * ========================================================================== * Gang blocks. * * A gang block is a collection of small blocks that looks to the DMU * like one large block. When zio_dva_allocate() cannot find a block * of the requested size, due to either severe fragmentation or the pool * being nearly full, it calls zio_write_gang_block() to construct the * block from smaller fragments. * * A gang block consists of a gang header (zio_gbh_phys_t) and up to * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like * an indirect block: it's an array of block pointers. It consumes * only one sector and hence is allocatable regardless of fragmentation. * The gang header's bps point to its gang members, which hold the data. * * Gang blocks are self-checksumming, using the bp's * as the verifier to ensure uniqueness of the SHA256 checksum. * Critically, the gang block bp's blk_cksum is the checksum of the data, * not the gang header. This ensures that data block signatures (needed for * deduplication) are independent of how the block is physically stored. * * Gang blocks can be nested: a gang member may itself be a gang block. * Thus every gang block is a tree in which root and all interior nodes are * gang headers, and the leaves are normal blocks that contain user data. * The root of the gang tree is called the gang leader. * * To perform any operation (read, rewrite, free, claim) on a gang block, * zio_gang_assemble() first assembles the gang tree (minus data leaves) * in the io_gang_tree field of the original logical i/o by recursively * reading the gang leader and all gang headers below it. This yields * an in-core tree containing the contents of every gang header and the * bps for every constituent of the gang block. * * With the gang tree now assembled, zio_gang_issue() just walks the gang tree * and invokes a callback on each bp. To free a gang block, zio_gang_issue() * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). * zio_read_gang() is a wrapper around zio_read() that omits reading gang * headers, since we already have those in io_gang_tree. zio_rewrite_gang() * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() * of the gang header plus zio_checksum_compute() of the data to update the * gang header's blk_cksum as described above. * * The two-phase assemble/issue model solves the problem of partial failure -- * what if you'd freed part of a gang block but then couldn't read the * gang header for another part? Assembling the entire gang tree first * ensures that all the necessary gang header I/O has succeeded before * starting the actual work of free, claim, or write. Once the gang tree * is assembled, free and claim are in-memory operations that cannot fail. * * In the event that a gang write fails, zio_dva_unallocate() walks the * gang tree to immediately free (i.e. insert back into the space map) * everything we've allocated. This ensures that we don't get ENOSPC * errors during repeated suspend/resume cycles due to a flaky device. * * Gang rewrites only happen during sync-to-convergence. If we can't assemble * the gang tree, we won't modify the block, so we can safely defer the free * (knowing that the block is still intact). If we *can* assemble the gang * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free * each constituent bp and we can allocate a new block on the next sync pass. * * In all cases, the gang tree allows complete recovery from partial failure. * ========================================================================== */ static void zio_gang_issue_func_done(zio_t *zio) { abd_put(zio->io_abd); } static zio_t * zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, abd_t *data, uint64_t offset) { if (gn != NULL) return (pio); return (zio_read(pio, pio->io_spa, bp, abd_get_offset(data, offset), BP_GET_PSIZE(bp), zio_gang_issue_func_done, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark)); } static zio_t * zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, abd_t *data, uint64_t offset) { zio_t *zio; if (gn != NULL) { abd_t *gbh_abd = abd_get_from_buf(gn->gn_gbh, SPA_GANGBLOCKSIZE); zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, gbh_abd, SPA_GANGBLOCKSIZE, zio_gang_issue_func_done, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); /* * As we rewrite each gang header, the pipeline will compute * a new gang block header checksum for it; but no one will * compute a new data checksum, so we do that here. The one * exception is the gang leader: the pipeline already computed * its data checksum because that stage precedes gang assembly. * (Presently, nothing actually uses interior data checksums; * this is just good hygiene.) */ if (gn != pio->io_gang_leader->io_gang_tree) { abd_t *buf = abd_get_offset(data, offset); zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), buf, BP_GET_PSIZE(bp)); abd_put(buf); } /* * If we are here to damage data for testing purposes, * leave the GBH alone so that we can detect the damage. */ if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; } else { zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, abd_get_offset(data, offset), BP_GET_PSIZE(bp), zio_gang_issue_func_done, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); } return (zio); } /* ARGSUSED */ static zio_t * zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, abd_t *data, uint64_t offset) { return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(bp), ZIO_GANG_CHILD_FLAGS(pio))); } /* ARGSUSED */ static zio_t * zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, abd_t *data, uint64_t offset) { return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); } static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { NULL, zio_read_gang, zio_rewrite_gang, zio_free_gang, zio_claim_gang, NULL }; static void zio_gang_tree_assemble_done(zio_t *zio); static zio_gang_node_t * zio_gang_node_alloc(zio_gang_node_t **gnpp) { zio_gang_node_t *gn; ASSERT(*gnpp == NULL); gn = kmem_zalloc(sizeof (*gn), KM_SLEEP); gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); *gnpp = gn; return (gn); } static void zio_gang_node_free(zio_gang_node_t **gnpp) { zio_gang_node_t *gn = *gnpp; for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) ASSERT(gn->gn_child[g] == NULL); zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); kmem_free(gn, sizeof (*gn)); *gnpp = NULL; } static void zio_gang_tree_free(zio_gang_node_t **gnpp) { zio_gang_node_t *gn = *gnpp; if (gn == NULL) return; for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) zio_gang_tree_free(&gn->gn_child[g]); zio_gang_node_free(gnpp); } static void zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) { zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); abd_t *gbh_abd = abd_get_from_buf(gn->gn_gbh, SPA_GANGBLOCKSIZE); ASSERT(gio->io_gang_leader == gio); ASSERT(BP_IS_GANG(bp)); zio_nowait(zio_read(gio, gio->io_spa, bp, gbh_abd, SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); } static void zio_gang_tree_assemble_done(zio_t *zio) { zio_t *gio = zio->io_gang_leader; zio_gang_node_t *gn = zio->io_private; blkptr_t *bp = zio->io_bp; ASSERT(gio == zio_unique_parent(zio)); ASSERT(zio->io_child_count == 0); if (zio->io_error) return; /* this ABD was created from a linear buf in zio_gang_tree_assemble */ if (BP_SHOULD_BYTESWAP(bp)) byteswap_uint64_array(abd_to_buf(zio->io_abd), zio->io_size); ASSERT3P(abd_to_buf(zio->io_abd), ==, gn->gn_gbh); ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); abd_put(zio->io_abd); for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; if (!BP_IS_GANG(gbp)) continue; zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); } } static void zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, abd_t *data, uint64_t offset) { zio_t *gio = pio->io_gang_leader; zio_t *zio; ASSERT(BP_IS_GANG(bp) == !!gn); ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); /* * If you're a gang header, your data is in gn->gn_gbh. * If you're a gang member, your data is in 'data' and gn == NULL. */ zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data, offset); if (gn != NULL) { ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; if (BP_IS_HOLE(gbp)) continue; zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data, offset); offset += BP_GET_PSIZE(gbp); } } if (gn == gio->io_gang_tree && gio->io_abd != NULL) ASSERT3U(gio->io_size, ==, offset); if (zio != pio) zio_nowait(zio); } static int zio_gang_assemble(zio_t *zio) { blkptr_t *bp = zio->io_bp; ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); ASSERT(zio->io_child_type > ZIO_CHILD_GANG); zio->io_gang_leader = zio; zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); return (ZIO_PIPELINE_CONTINUE); } static int zio_gang_issue(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (zio_wait_for_children(zio, ZIO_CHILD_GANG_BIT, ZIO_WAIT_DONE)) { return (ZIO_PIPELINE_STOP); } ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); ASSERT(zio->io_child_type > ZIO_CHILD_GANG); if (zio->io_child_error[ZIO_CHILD_GANG] == 0) zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_abd, 0); else zio_gang_tree_free(&zio->io_gang_tree); zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; return (ZIO_PIPELINE_CONTINUE); } static void zio_write_gang_member_ready(zio_t *zio) { zio_t *pio = zio_unique_parent(zio); zio_t *gio = zio->io_gang_leader; dva_t *cdva = zio->io_bp->blk_dva; dva_t *pdva = pio->io_bp->blk_dva; uint64_t asize; if (BP_IS_HOLE(zio->io_bp)) return; ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); ASSERT(zio->io_child_type == ZIO_CHILD_GANG); ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); mutex_enter(&pio->io_lock); for (int d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { ASSERT(DVA_GET_GANG(&pdva[d])); asize = DVA_GET_ASIZE(&pdva[d]); asize += DVA_GET_ASIZE(&cdva[d]); DVA_SET_ASIZE(&pdva[d], asize); } mutex_exit(&pio->io_lock); } static void zio_write_gang_done(zio_t *zio) { abd_put(zio->io_abd); } static int zio_write_gang_block(zio_t *pio) { spa_t *spa = pio->io_spa; metaslab_class_t *mc = spa_normal_class(spa); blkptr_t *bp = pio->io_bp; zio_t *gio = pio->io_gang_leader; zio_t *zio; zio_gang_node_t *gn, **gnpp; zio_gbh_phys_t *gbh; abd_t *gbh_abd; uint64_t txg = pio->io_txg; uint64_t resid = pio->io_size; uint64_t lsize; int copies = gio->io_prop.zp_copies; int gbh_copies = MIN(copies + 1, spa_max_replication(spa)); zio_prop_t zp; int error; int flags = METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER; if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(!(pio->io_flags & ZIO_FLAG_NODATA)); flags |= METASLAB_ASYNC_ALLOC; - VERIFY(refcount_held(&mc->mc_alloc_slots, pio)); + VERIFY(refcount_held(&mc->mc_alloc_slots[pio->io_allocator], + pio)); /* * The logical zio has already placed a reservation for * 'copies' allocation slots but gang blocks may require * additional copies. These additional copies * (i.e. gbh_copies - copies) are guaranteed to succeed * since metaslab_class_throttle_reserve() always allows * additional reservations for gang blocks. */ VERIFY(metaslab_class_throttle_reserve(mc, gbh_copies - copies, - pio, flags)); + pio->io_allocator, pio, flags)); } error = metaslab_alloc(spa, mc, SPA_GANGBLOCKSIZE, bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, flags, - &pio->io_alloc_list, pio); + &pio->io_alloc_list, pio, pio->io_allocator); if (error) { if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(!(pio->io_flags & ZIO_FLAG_NODATA)); /* * If we failed to allocate the gang block header then * we remove any additional allocation reservations that * we placed here. The original reservation will * be removed when the logical I/O goes to the ready * stage. */ metaslab_class_throttle_unreserve(mc, - gbh_copies - copies, pio); + gbh_copies - copies, pio->io_allocator, pio); } pio->io_error = error; return (ZIO_PIPELINE_CONTINUE); } if (pio == gio) { gnpp = &gio->io_gang_tree; } else { gnpp = pio->io_private; ASSERT(pio->io_ready == zio_write_gang_member_ready); } gn = zio_gang_node_alloc(gnpp); gbh = gn->gn_gbh; bzero(gbh, SPA_GANGBLOCKSIZE); gbh_abd = abd_get_from_buf(gbh, SPA_GANGBLOCKSIZE); /* * Create the gang header. */ zio = zio_rewrite(pio, spa, txg, bp, gbh_abd, SPA_GANGBLOCKSIZE, zio_write_gang_done, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); /* * Create and nowait the gang children. */ for (int g = 0; resid != 0; resid -= lsize, g++) { lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), SPA_MINBLOCKSIZE); ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); zp.zp_checksum = gio->io_prop.zp_checksum; zp.zp_compress = ZIO_COMPRESS_OFF; zp.zp_type = DMU_OT_NONE; zp.zp_level = 0; zp.zp_copies = gio->io_prop.zp_copies; zp.zp_dedup = B_FALSE; zp.zp_dedup_verify = B_FALSE; zp.zp_nopwrite = B_FALSE; zio_t *cio = zio_write(zio, spa, txg, &gbh->zg_blkptr[g], abd_get_offset(pio->io_abd, pio->io_size - resid), lsize, lsize, &zp, zio_write_gang_member_ready, NULL, NULL, zio_write_gang_done, &gn->gn_child[g], pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(!(pio->io_flags & ZIO_FLAG_NODATA)); /* * Gang children won't throttle but we should * account for their work, so reserve an allocation * slot for them here. */ VERIFY(metaslab_class_throttle_reserve(mc, - zp.zp_copies, cio, flags)); + zp.zp_copies, cio->io_allocator, cio, flags)); } zio_nowait(cio); } /* * Set pio's pipeline to just wait for zio to finish. */ pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; zio_nowait(zio); return (ZIO_PIPELINE_CONTINUE); } /* * The zio_nop_write stage in the pipeline determines if allocating a * new bp is necessary. The nopwrite feature can handle writes in * either syncing or open context (i.e. zil writes) and as a result is * mutually exclusive with dedup. * * By leveraging a cryptographically secure checksum, such as SHA256, we * can compare the checksums of the new data and the old to determine if * allocating a new block is required. Note that our requirements for * cryptographic strength are fairly weak: there can't be any accidental * hash collisions, but we don't need to be secure against intentional * (malicious) collisions. To trigger a nopwrite, you have to be able * to write the file to begin with, and triggering an incorrect (hash * collision) nopwrite is no worse than simply writing to the file. * That said, there are no known attacks against the checksum algorithms * used for nopwrite, assuming that the salt and the checksums * themselves remain secret. */ static int zio_nop_write(zio_t *zio) { blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; zio_prop_t *zp = &zio->io_prop; ASSERT(BP_GET_LEVEL(bp) == 0); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); ASSERT(zp->zp_nopwrite); ASSERT(!zp->zp_dedup); ASSERT(zio->io_bp_override == NULL); ASSERT(IO_IS_ALLOCATING(zio)); /* * Check to see if the original bp and the new bp have matching * characteristics (i.e. same checksum, compression algorithms, etc). * If they don't then just continue with the pipeline which will * allocate a new bp. */ if (BP_IS_HOLE(bp_orig) || !(zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_flags & ZCHECKSUM_FLAG_NOPWRITE) || BP_GET_CHECKSUM(bp) != BP_GET_CHECKSUM(bp_orig) || BP_GET_COMPRESS(bp) != BP_GET_COMPRESS(bp_orig) || BP_GET_DEDUP(bp) != BP_GET_DEDUP(bp_orig) || zp->zp_copies != BP_GET_NDVAS(bp_orig)) return (ZIO_PIPELINE_CONTINUE); /* * If the checksums match then reset the pipeline so that we * avoid allocating a new bp and issuing any I/O. */ if (ZIO_CHECKSUM_EQUAL(bp->blk_cksum, bp_orig->blk_cksum)) { ASSERT(zio_checksum_table[zp->zp_checksum].ci_flags & ZCHECKSUM_FLAG_NOPWRITE); ASSERT3U(BP_GET_PSIZE(bp), ==, BP_GET_PSIZE(bp_orig)); ASSERT3U(BP_GET_LSIZE(bp), ==, BP_GET_LSIZE(bp_orig)); ASSERT(zp->zp_compress != ZIO_COMPRESS_OFF); ASSERT(bcmp(&bp->blk_prop, &bp_orig->blk_prop, sizeof (uint64_t)) == 0); *bp = *bp_orig; zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; zio->io_flags |= ZIO_FLAG_NOPWRITE; } return (ZIO_PIPELINE_CONTINUE); } /* * ========================================================================== * Dedup * ========================================================================== */ static void zio_ddt_child_read_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp; zio_t *pio = zio_unique_parent(zio); mutex_enter(&pio->io_lock); ddp = ddt_phys_select(dde, bp); if (zio->io_error == 0) ddt_phys_clear(ddp); /* this ddp doesn't need repair */ if (zio->io_error == 0 && dde->dde_repair_abd == NULL) dde->dde_repair_abd = zio->io_abd; else abd_free(zio->io_abd); mutex_exit(&pio->io_lock); } static int zio_ddt_read_start(zio_t *zio) { blkptr_t *bp = zio->io_bp; ASSERT(BP_GET_DEDUP(bp)); ASSERT(BP_GET_PSIZE(bp) == zio->io_size); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); if (zio->io_child_error[ZIO_CHILD_DDT]) { ddt_t *ddt = ddt_select(zio->io_spa, bp); ddt_entry_t *dde = ddt_repair_start(ddt, bp); ddt_phys_t *ddp = dde->dde_phys; ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); blkptr_t blk; ASSERT(zio->io_vsd == NULL); zio->io_vsd = dde; if (ddp_self == NULL) return (ZIO_PIPELINE_CONTINUE); for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) continue; ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, &blk); zio_nowait(zio_read(zio, zio->io_spa, &blk, abd_alloc_for_io(zio->io_size, B_TRUE), zio->io_size, zio_ddt_child_read_done, dde, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, &zio->io_bookmark)); } return (ZIO_PIPELINE_CONTINUE); } zio_nowait(zio_read(zio, zio->io_spa, bp, zio->io_abd, zio->io_size, NULL, NULL, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); return (ZIO_PIPELINE_CONTINUE); } static int zio_ddt_read_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; if (zio_wait_for_children(zio, ZIO_CHILD_DDT_BIT, ZIO_WAIT_DONE)) { return (ZIO_PIPELINE_STOP); } ASSERT(BP_GET_DEDUP(bp)); ASSERT(BP_GET_PSIZE(bp) == zio->io_size); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); if (zio->io_child_error[ZIO_CHILD_DDT]) { ddt_t *ddt = ddt_select(zio->io_spa, bp); ddt_entry_t *dde = zio->io_vsd; if (ddt == NULL) { ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); return (ZIO_PIPELINE_CONTINUE); } if (dde == NULL) { zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); return (ZIO_PIPELINE_STOP); } if (dde->dde_repair_abd != NULL) { abd_copy(zio->io_abd, dde->dde_repair_abd, zio->io_size); zio->io_child_error[ZIO_CHILD_DDT] = 0; } ddt_repair_done(ddt, dde); zio->io_vsd = NULL; } ASSERT(zio->io_vsd == NULL); return (ZIO_PIPELINE_CONTINUE); } static boolean_t zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) { spa_t *spa = zio->io_spa; boolean_t do_raw = (zio->io_flags & ZIO_FLAG_RAW); /* We should never get a raw, override zio */ ASSERT(!(zio->io_bp_override && do_raw)); /* * Note: we compare the original data, not the transformed data, * because when zio->io_bp is an override bp, we will not have * pushed the I/O transforms. That's an important optimization * because otherwise we'd compress/encrypt all dmu_sync() data twice. */ for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { zio_t *lio = dde->dde_lead_zio[p]; if (lio != NULL) { return (lio->io_orig_size != zio->io_orig_size || abd_cmp(zio->io_orig_abd, lio->io_orig_abd, zio->io_orig_size) != 0); } } for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { ddt_phys_t *ddp = &dde->dde_phys[p]; if (ddp->ddp_phys_birth != 0) { arc_buf_t *abuf = NULL; arc_flags_t aflags = ARC_FLAG_WAIT; int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE; blkptr_t blk = *zio->io_bp; int error; ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); ddt_exit(ddt); /* * Intuitively, it would make more sense to compare * io_abd than io_orig_abd in the raw case since you * don't want to look at any transformations that have * happened to the data. However, for raw I/Os the * data will actually be the same in io_abd and * io_orig_abd, so all we have to do is issue this as * a raw ARC read. */ if (do_raw) { zio_flags |= ZIO_FLAG_RAW; ASSERT3U(zio->io_size, ==, zio->io_orig_size); ASSERT0(abd_cmp(zio->io_abd, zio->io_orig_abd, zio->io_size)); ASSERT3P(zio->io_transform_stack, ==, NULL); } error = arc_read(NULL, spa, &blk, arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zio->io_bookmark); if (error == 0) { if (arc_buf_size(abuf) != zio->io_orig_size || abd_cmp_buf(zio->io_orig_abd, abuf->b_data, zio->io_orig_size) != 0) error = SET_ERROR(EEXIST); arc_buf_destroy(abuf, &abuf); } ddt_enter(ddt); return (error != 0); } } return (B_FALSE); } static void zio_ddt_child_write_ready(zio_t *zio) { int p = zio->io_prop.zp_copies; ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp = &dde->dde_phys[p]; zio_t *pio; if (zio->io_error) return; ddt_enter(ddt); ASSERT(dde->dde_lead_zio[p] == zio); ddt_phys_fill(ddp, zio->io_bp); zio_link_t *zl = NULL; while ((pio = zio_walk_parents(zio, &zl)) != NULL) ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); ddt_exit(ddt); } static void zio_ddt_child_write_done(zio_t *zio) { int p = zio->io_prop.zp_copies; ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp = &dde->dde_phys[p]; ddt_enter(ddt); ASSERT(ddp->ddp_refcnt == 0); ASSERT(dde->dde_lead_zio[p] == zio); dde->dde_lead_zio[p] = NULL; if (zio->io_error == 0) { zio_link_t *zl = NULL; while (zio_walk_parents(zio, &zl) != NULL) ddt_phys_addref(ddp); } else { ddt_phys_clear(ddp); } ddt_exit(ddt); } static void zio_ddt_ditto_write_done(zio_t *zio) { int p = DDT_PHYS_DITTO; zio_prop_t *zp = &zio->io_prop; blkptr_t *bp = zio->io_bp; ddt_t *ddt = ddt_select(zio->io_spa, bp); ddt_entry_t *dde = zio->io_private; ddt_phys_t *ddp = &dde->dde_phys[p]; ddt_key_t *ddk = &dde->dde_key; ddt_enter(ddt); ASSERT(ddp->ddp_refcnt == 0); ASSERT(dde->dde_lead_zio[p] == zio); dde->dde_lead_zio[p] = NULL; if (zio->io_error == 0) { ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum)); ASSERT(zp->zp_copies < SPA_DVAS_PER_BP); ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp)); if (ddp->ddp_phys_birth != 0) ddt_phys_free(ddt, ddk, ddp, zio->io_txg); ddt_phys_fill(ddp, bp); } ddt_exit(ddt); } static int zio_ddt_write(zio_t *zio) { spa_t *spa = zio->io_spa; blkptr_t *bp = zio->io_bp; uint64_t txg = zio->io_txg; zio_prop_t *zp = &zio->io_prop; int p = zp->zp_copies; int ditto_copies; zio_t *cio = NULL; zio_t *dio = NULL; ddt_t *ddt = ddt_select(spa, bp); ddt_entry_t *dde; ddt_phys_t *ddp; ASSERT(BP_GET_DEDUP(bp)); ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); ASSERT(!(zio->io_bp_override && (zio->io_flags & ZIO_FLAG_RAW))); ddt_enter(ddt); dde = ddt_lookup(ddt, bp, B_TRUE); ddp = &dde->dde_phys[p]; if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { /* * If we're using a weak checksum, upgrade to a strong checksum * and try again. If we're already using a strong checksum, * we can't resolve it, so just convert to an ordinary write. * (And automatically e-mail a paper to Nature?) */ if (!(zio_checksum_table[zp->zp_checksum].ci_flags & ZCHECKSUM_FLAG_DEDUP)) { zp->zp_checksum = spa_dedup_checksum(spa); zio_pop_transforms(zio); zio->io_stage = ZIO_STAGE_OPEN; BP_ZERO(bp); } else { zp->zp_dedup = B_FALSE; BP_SET_DEDUP(bp, B_FALSE); } ASSERT(!BP_GET_DEDUP(bp)); zio->io_pipeline = ZIO_WRITE_PIPELINE; ddt_exit(ddt); return (ZIO_PIPELINE_CONTINUE); } ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp); ASSERT(ditto_copies < SPA_DVAS_PER_BP); if (ditto_copies > ddt_ditto_copies_present(dde) && dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) { zio_prop_t czp = *zp; czp.zp_copies = ditto_copies; /* * If we arrived here with an override bp, we won't have run * the transform stack, so we won't have the data we need to * generate a child i/o. So, toss the override bp and restart. * This is safe, because using the override bp is just an * optimization; and it's rare, so the cost doesn't matter. */ if (zio->io_bp_override) { zio_pop_transforms(zio); zio->io_stage = ZIO_STAGE_OPEN; zio->io_pipeline = ZIO_WRITE_PIPELINE; zio->io_bp_override = NULL; BP_ZERO(bp); ddt_exit(ddt); return (ZIO_PIPELINE_CONTINUE); } dio = zio_write(zio, spa, txg, bp, zio->io_orig_abd, zio->io_orig_size, zio->io_orig_size, &czp, NULL, NULL, NULL, zio_ddt_ditto_write_done, dde, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); zio_push_transform(dio, zio->io_abd, zio->io_size, 0, NULL); dde->dde_lead_zio[DDT_PHYS_DITTO] = dio; } if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { if (ddp->ddp_phys_birth != 0) ddt_bp_fill(ddp, bp, txg); if (dde->dde_lead_zio[p] != NULL) zio_add_child(zio, dde->dde_lead_zio[p]); else ddt_phys_addref(ddp); } else if (zio->io_bp_override) { ASSERT(bp->blk_birth == txg); ASSERT(BP_EQUAL(bp, zio->io_bp_override)); ddt_phys_fill(ddp, bp); ddt_phys_addref(ddp); } else { cio = zio_write(zio, spa, txg, bp, zio->io_orig_abd, zio->io_orig_size, zio->io_orig_size, zp, zio_ddt_child_write_ready, NULL, NULL, zio_ddt_child_write_done, dde, zio->io_priority, ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); zio_push_transform(cio, zio->io_abd, zio->io_size, 0, NULL); dde->dde_lead_zio[p] = cio; } ddt_exit(ddt); if (cio) zio_nowait(cio); if (dio) zio_nowait(dio); return (ZIO_PIPELINE_CONTINUE); } ddt_entry_t *freedde; /* for debugging */ static int zio_ddt_free(zio_t *zio) { spa_t *spa = zio->io_spa; blkptr_t *bp = zio->io_bp; ddt_t *ddt = ddt_select(spa, bp); ddt_entry_t *dde; ddt_phys_t *ddp; ASSERT(BP_GET_DEDUP(bp)); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); ddt_enter(ddt); freedde = dde = ddt_lookup(ddt, bp, B_TRUE); ddp = ddt_phys_select(dde, bp); ddt_phys_decref(ddp); ddt_exit(ddt); return (ZIO_PIPELINE_CONTINUE); } /* * ========================================================================== * Allocate and free blocks * ========================================================================== */ static zio_t * -zio_io_to_allocate(spa_t *spa) +zio_io_to_allocate(spa_t *spa, int allocator) { zio_t *zio; - ASSERT(MUTEX_HELD(&spa->spa_alloc_lock)); + ASSERT(MUTEX_HELD(&spa->spa_alloc_locks[allocator])); - zio = avl_first(&spa->spa_alloc_tree); + zio = avl_first(&spa->spa_alloc_trees[allocator]); if (zio == NULL) return (NULL); ASSERT(IO_IS_ALLOCATING(zio)); /* * Try to place a reservation for this zio. If we're unable to * reserve then we throttle. */ + ASSERT3U(zio->io_allocator, ==, allocator); if (!metaslab_class_throttle_reserve(spa_normal_class(spa), - zio->io_prop.zp_copies, zio, 0)) { + zio->io_prop.zp_copies, zio->io_allocator, zio, 0)) { return (NULL); } - avl_remove(&spa->spa_alloc_tree, zio); + avl_remove(&spa->spa_alloc_trees[allocator], zio); ASSERT3U(zio->io_stage, <, ZIO_STAGE_DVA_ALLOCATE); return (zio); } static int zio_dva_throttle(zio_t *zio) { spa_t *spa = zio->io_spa; zio_t *nio; if (zio->io_priority == ZIO_PRIORITY_SYNC_WRITE || !spa_normal_class(zio->io_spa)->mc_alloc_throttle_enabled || zio->io_child_type == ZIO_CHILD_GANG || zio->io_flags & ZIO_FLAG_NODATA) { return (ZIO_PIPELINE_CONTINUE); } ASSERT(zio->io_child_type > ZIO_CHILD_GANG); ASSERT3U(zio->io_queued_timestamp, >, 0); ASSERT(zio->io_stage == ZIO_STAGE_DVA_THROTTLE); - mutex_enter(&spa->spa_alloc_lock); + zbookmark_phys_t *bm = &zio->io_bookmark; + /* + * We want to try to use as many allocators as possible to help improve + * performance, but we also want logically adjacent IOs to be physically + * adjacent to improve sequential read performance. We chunk each object + * into 2^20 block regions, and then hash based on the objset, object, + * level, and region to accomplish both of these goals. + */ + zio->io_allocator = cityhash4(bm->zb_objset, bm->zb_object, + bm->zb_level, bm->zb_blkid >> 20) % spa->spa_alloc_count; + mutex_enter(&spa->spa_alloc_locks[zio->io_allocator]); ASSERT(zio->io_type == ZIO_TYPE_WRITE); - avl_add(&spa->spa_alloc_tree, zio); + avl_add(&spa->spa_alloc_trees[zio->io_allocator], zio); - nio = zio_io_to_allocate(zio->io_spa); - mutex_exit(&spa->spa_alloc_lock); + nio = zio_io_to_allocate(zio->io_spa, zio->io_allocator); + mutex_exit(&spa->spa_alloc_locks[zio->io_allocator]); if (nio == zio) return (ZIO_PIPELINE_CONTINUE); if (nio != NULL) { ASSERT(nio->io_stage == ZIO_STAGE_DVA_THROTTLE); /* * We are passing control to a new zio so make sure that * it is processed by a different thread. We do this to * avoid stack overflows that can occur when parents are * throttled and children are making progress. We allow * it to go to the head of the taskq since it's already * been waiting. */ zio_taskq_dispatch(nio, ZIO_TASKQ_ISSUE, B_TRUE); } return (ZIO_PIPELINE_STOP); } void -zio_allocate_dispatch(spa_t *spa) +zio_allocate_dispatch(spa_t *spa, int allocator) { zio_t *zio; - mutex_enter(&spa->spa_alloc_lock); - zio = zio_io_to_allocate(spa); - mutex_exit(&spa->spa_alloc_lock); + mutex_enter(&spa->spa_alloc_locks[allocator]); + zio = zio_io_to_allocate(spa, allocator); + mutex_exit(&spa->spa_alloc_locks[allocator]); if (zio == NULL) return; ASSERT3U(zio->io_stage, ==, ZIO_STAGE_DVA_THROTTLE); ASSERT0(zio->io_error); zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_TRUE); } static int zio_dva_allocate(zio_t *zio) { spa_t *spa = zio->io_spa; metaslab_class_t *mc = spa_normal_class(spa); blkptr_t *bp = zio->io_bp; int error; int flags = 0; if (zio->io_gang_leader == NULL) { ASSERT(zio->io_child_type > ZIO_CHILD_GANG); zio->io_gang_leader = zio; } ASSERT(BP_IS_HOLE(bp)); ASSERT0(BP_GET_NDVAS(bp)); ASSERT3U(zio->io_prop.zp_copies, >, 0); ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); if (zio->io_flags & ZIO_FLAG_NODATA) { flags |= METASLAB_DONT_THROTTLE; } if (zio->io_flags & ZIO_FLAG_GANG_CHILD) { flags |= METASLAB_GANG_CHILD; } if (zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE) { flags |= METASLAB_ASYNC_ALLOC; } error = metaslab_alloc(spa, mc, zio->io_size, bp, zio->io_prop.zp_copies, zio->io_txg, NULL, flags, - &zio->io_alloc_list, zio); + &zio->io_alloc_list, zio, zio->io_allocator); if (error != 0) { spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, " "size %llu, error %d", spa_name(spa), zio, zio->io_size, error); if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) return (zio_write_gang_block(zio)); zio->io_error = error; } return (ZIO_PIPELINE_CONTINUE); } static int zio_dva_free(zio_t *zio) { metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); return (ZIO_PIPELINE_CONTINUE); } static int zio_dva_claim(zio_t *zio) { int error; error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); if (error) zio->io_error = error; return (ZIO_PIPELINE_CONTINUE); } /* * Undo an allocation. This is used by zio_done() when an I/O fails * and we want to give back the block we just allocated. * This handles both normal blocks and gang blocks. */ static void zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) { ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); ASSERT(zio->io_bp_override == NULL); if (!BP_IS_HOLE(bp)) metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); if (gn != NULL) { for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) { zio_dva_unallocate(zio, gn->gn_child[g], &gn->gn_gbh->zg_blkptr[g]); } } } /* * Try to allocate an intent log block. Return 0 on success, errno on failure. */ int -zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp, - uint64_t size, boolean_t *slog) +zio_alloc_zil(spa_t *spa, uint64_t objset, uint64_t txg, blkptr_t *new_bp, + blkptr_t *old_bp, uint64_t size, boolean_t *slog) { int error = 1; zio_alloc_list_t io_alloc_list; ASSERT(txg > spa_syncing_txg(spa)); metaslab_trace_init(&io_alloc_list); + /* + * When allocating a zil block, we don't have information about + * the final destination of the block except the objset it's part + * of, so we just hash the objset ID to pick the allocator to get + * some parallelism. + */ error = metaslab_alloc(spa, spa_log_class(spa), size, new_bp, 1, - txg, old_bp, METASLAB_HINTBP_AVOID, &io_alloc_list, NULL); + txg, old_bp, METASLAB_HINTBP_AVOID, &io_alloc_list, NULL, + cityhash4(0, 0, 0, objset) % spa->spa_alloc_count); if (error == 0) { *slog = TRUE; } else { error = metaslab_alloc(spa, spa_normal_class(spa), size, new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID, - &io_alloc_list, NULL); + &io_alloc_list, NULL, cityhash4(0, 0, 0, objset) % + spa->spa_alloc_count); if (error == 0) *slog = FALSE; } metaslab_trace_fini(&io_alloc_list); if (error == 0) { BP_SET_LSIZE(new_bp, size); BP_SET_PSIZE(new_bp, size); BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); BP_SET_CHECKSUM(new_bp, spa_version(spa) >= SPA_VERSION_SLIM_ZIL ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); BP_SET_LEVEL(new_bp, 0); BP_SET_DEDUP(new_bp, 0); BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); } else { zfs_dbgmsg("%s: zil block allocation failure: " "size %llu, error %d", spa_name(spa), size, error); } return (error); } /* * ========================================================================== * Read, write and delete to physical devices * ========================================================================== */ /* * Issue an I/O to the underlying vdev. Typically the issue pipeline * stops after this stage and will resume upon I/O completion. * However, there are instances where the vdev layer may need to * continue the pipeline when an I/O was not issued. Since the I/O * that was sent to the vdev layer might be different than the one * currently active in the pipeline (see vdev_queue_io()), we explicitly * force the underlying vdev layers to call either zio_execute() or * zio_interrupt() to ensure that the pipeline continues with the correct I/O. */ static int zio_vdev_io_start(zio_t *zio) { vdev_t *vd = zio->io_vd; uint64_t align; spa_t *spa = zio->io_spa; int ret; ASSERT(zio->io_error == 0); ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); if (vd == NULL) { if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) spa_config_enter(spa, SCL_ZIO, zio, RW_READER); /* * The mirror_ops handle multiple DVAs in a single BP. */ vdev_mirror_ops.vdev_op_io_start(zio); return (ZIO_PIPELINE_STOP); } if (vd->vdev_ops->vdev_op_leaf && zio->io_type == ZIO_TYPE_FREE && zio->io_priority == ZIO_PRIORITY_NOW) { trim_map_free(vd, zio->io_offset, zio->io_size, zio->io_txg); return (ZIO_PIPELINE_CONTINUE); } ASSERT3P(zio->io_logical, !=, zio); if (zio->io_type == ZIO_TYPE_WRITE) { ASSERT(spa->spa_trust_config); if (zio->io_vd->vdev_removing) { ASSERT(zio->io_flags & (ZIO_FLAG_PHYSICAL | ZIO_FLAG_SELF_HEAL | ZIO_FLAG_INDUCE_DAMAGE)); } } /* * We keep track of time-sensitive I/Os so that the scan thread * can quickly react to certain workloads. In particular, we care * about non-scrubbing, top-level reads and writes with the following * characteristics: * - synchronous writes of user data to non-slog devices * - any reads of user data * When these conditions are met, adjust the timestamp of spa_last_io * which allows the scan thread to adjust its workload accordingly. */ if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL && vd == vd->vdev_top && !vd->vdev_islog && zio->io_bookmark.zb_objset != DMU_META_OBJSET && zio->io_txg != spa_syncing_txg(spa)) { uint64_t old = spa->spa_last_io; uint64_t new = ddi_get_lbolt64(); if (old != new) (void) atomic_cas_64(&spa->spa_last_io, old, new); } align = 1ULL << vd->vdev_top->vdev_ashift; if (!(zio->io_flags & ZIO_FLAG_PHYSICAL) && P2PHASE(zio->io_size, align) != 0) { /* Transform logical writes to be a full physical block size. */ uint64_t asize = P2ROUNDUP(zio->io_size, align); abd_t *abuf = NULL; if (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE) abuf = abd_alloc_sametype(zio->io_abd, asize); ASSERT(vd == vd->vdev_top); if (zio->io_type == ZIO_TYPE_WRITE) { abd_copy(abuf, zio->io_abd, zio->io_size); abd_zero_off(abuf, zio->io_size, asize - zio->io_size); } zio_push_transform(zio, abuf, asize, abuf ? asize : 0, zio_subblock); } /* * If this is not a physical io, make sure that it is properly aligned * before proceeding. */ if (!(zio->io_flags & ZIO_FLAG_PHYSICAL)) { ASSERT0(P2PHASE(zio->io_offset, align)); ASSERT0(P2PHASE(zio->io_size, align)); } else { /* * For the physical io we allow alignment * to a logical block size. */ uint64_t log_align = 1ULL << vd->vdev_top->vdev_logical_ashift; ASSERT0(P2PHASE(zio->io_offset, log_align)); ASSERT0(P2PHASE(zio->io_size, log_align)); } VERIFY(zio->io_type == ZIO_TYPE_READ || spa_writeable(spa)); /* * If this is a repair I/O, and there's no self-healing involved -- * that is, we're just resilvering what we expect to resilver -- * then don't do the I/O unless zio's txg is actually in vd's DTL. * This prevents spurious resilvering with nested replication. * For example, given a mirror of mirrors, (A+B)+(C+D), if only * A is out of date, we'll read from C+D, then use the data to * resilver A+B -- but we don't actually want to resilver B, just A. * The top-level mirror has no way to know this, so instead we just * discard unnecessary repairs as we work our way down the vdev tree. * The same logic applies to any form of nested replication: * ditto + mirror, RAID-Z + replacing, etc. This covers them all. */ if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && zio->io_txg != 0 && /* not a delegated i/o */ !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { ASSERT(zio->io_type == ZIO_TYPE_WRITE); zio_vdev_io_bypass(zio); return (ZIO_PIPELINE_CONTINUE); } if (vd->vdev_ops->vdev_op_leaf) { switch (zio->io_type) { case ZIO_TYPE_READ: if (vdev_cache_read(zio)) return (ZIO_PIPELINE_CONTINUE); /* FALLTHROUGH */ case ZIO_TYPE_WRITE: case ZIO_TYPE_FREE: if ((zio = vdev_queue_io(zio)) == NULL) return (ZIO_PIPELINE_STOP); if (!vdev_accessible(vd, zio)) { zio->io_error = SET_ERROR(ENXIO); zio_interrupt(zio); return (ZIO_PIPELINE_STOP); } break; } /* * Note that we ignore repair writes for TRIM because they can * conflict with normal writes. This isn't an issue because, by * definition, we only repair blocks that aren't freed. */ if (zio->io_type == ZIO_TYPE_WRITE && !(zio->io_flags & ZIO_FLAG_IO_REPAIR) && !trim_map_write_start(zio)) return (ZIO_PIPELINE_STOP); } vd->vdev_ops->vdev_op_io_start(zio); return (ZIO_PIPELINE_STOP); } static int zio_vdev_io_done(zio_t *zio) { vdev_t *vd = zio->io_vd; vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; boolean_t unexpected_error = B_FALSE; if (zio_wait_for_children(zio, ZIO_CHILD_VDEV_BIT, ZIO_WAIT_DONE)) { return (ZIO_PIPELINE_STOP); } ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE || zio->io_type == ZIO_TYPE_FREE); if (vd != NULL && vd->vdev_ops->vdev_op_leaf && (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE || zio->io_type == ZIO_TYPE_FREE)) { if (zio->io_type == ZIO_TYPE_WRITE && !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) trim_map_write_done(zio); vdev_queue_io_done(zio); if (zio->io_type == ZIO_TYPE_WRITE) vdev_cache_write(zio); if (zio_injection_enabled && zio->io_error == 0) zio->io_error = zio_handle_device_injection(vd, zio, EIO); if (zio_injection_enabled && zio->io_error == 0) zio->io_error = zio_handle_label_injection(zio, EIO); if (zio->io_error) { if (zio->io_error == ENOTSUP && zio->io_type == ZIO_TYPE_FREE) { /* Not all devices support TRIM. */ } else if (!vdev_accessible(vd, zio)) { zio->io_error = SET_ERROR(ENXIO); } else { unexpected_error = B_TRUE; } } } ops->vdev_op_io_done(zio); if (unexpected_error) VERIFY(vdev_probe(vd, zio) == NULL); return (ZIO_PIPELINE_CONTINUE); } /* * This function is used to change the priority of an existing zio that is * currently in-flight. This is used by the arc to upgrade priority in the * event that a demand read is made for a block that is currently queued * as a scrub or async read IO. Otherwise, the high priority read request * would end up having to wait for the lower priority IO. */ void zio_change_priority(zio_t *pio, zio_priority_t priority) { zio_t *cio, *cio_next; zio_link_t *zl = NULL; ASSERT3U(priority, <, ZIO_PRIORITY_NUM_QUEUEABLE); if (pio->io_vd != NULL && pio->io_vd->vdev_ops->vdev_op_leaf) { vdev_queue_change_io_priority(pio, priority); } else { pio->io_priority = priority; } mutex_enter(&pio->io_lock); for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) { cio_next = zio_walk_children(pio, &zl); zio_change_priority(cio, priority); } mutex_exit(&pio->io_lock); } /* * For non-raidz ZIOs, we can just copy aside the bad data read from the * disk, and use that to finish the checksum ereport later. */ static void zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, const void *good_buf) { /* no processing needed */ zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); } /*ARGSUSED*/ void zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored) { void *buf = zio_buf_alloc(zio->io_size); abd_copy_to_buf(buf, zio->io_abd, zio->io_size); zcr->zcr_cbinfo = zio->io_size; zcr->zcr_cbdata = buf; zcr->zcr_finish = zio_vsd_default_cksum_finish; zcr->zcr_free = zio_buf_free; } static int zio_vdev_io_assess(zio_t *zio) { vdev_t *vd = zio->io_vd; if (zio_wait_for_children(zio, ZIO_CHILD_VDEV_BIT, ZIO_WAIT_DONE)) { return (ZIO_PIPELINE_STOP); } if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) spa_config_exit(zio->io_spa, SCL_ZIO, zio); if (zio->io_vsd != NULL) { zio->io_vsd_ops->vsd_free(zio); zio->io_vsd = NULL; } if (zio_injection_enabled && zio->io_error == 0) zio->io_error = zio_handle_fault_injection(zio, EIO); if (zio->io_type == ZIO_TYPE_FREE && zio->io_priority != ZIO_PRIORITY_NOW) { switch (zio->io_error) { case 0: ZIO_TRIM_STAT_INCR(bytes, zio->io_size); ZIO_TRIM_STAT_BUMP(success); break; case EOPNOTSUPP: ZIO_TRIM_STAT_BUMP(unsupported); break; default: ZIO_TRIM_STAT_BUMP(failed); break; } } /* * If the I/O failed, determine whether we should attempt to retry it. * * On retry, we cut in line in the issue queue, since we don't want * compression/checksumming/etc. work to prevent our (cheap) IO reissue. */ if (zio->io_error && vd == NULL && !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ zio->io_error = 0; zio->io_flags |= ZIO_FLAG_IO_RETRY | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, zio_requeue_io_start_cut_in_line); return (ZIO_PIPELINE_STOP); } /* * If we got an error on a leaf device, convert it to ENXIO * if the device is not accessible at all. */ if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && !vdev_accessible(vd, zio)) zio->io_error = SET_ERROR(ENXIO); /* * If we can't write to an interior vdev (mirror or RAID-Z), * set vdev_cant_write so that we stop trying to allocate from it. */ if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && vd != NULL && !vd->vdev_ops->vdev_op_leaf) { vd->vdev_cant_write = B_TRUE; } /* * If a cache flush returns ENOTSUP or ENOTTY, we know that no future * attempts will ever succeed. In this case we set a persistent bit so * that we don't bother with it in the future. */ if ((zio->io_error == ENOTSUP || zio->io_error == ENOTTY) && zio->io_type == ZIO_TYPE_IOCTL && zio->io_cmd == DKIOCFLUSHWRITECACHE && vd != NULL) vd->vdev_nowritecache = B_TRUE; if (zio->io_error) zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; if (vd != NULL && vd->vdev_ops->vdev_op_leaf && zio->io_physdone != NULL) { ASSERT(!(zio->io_flags & ZIO_FLAG_DELEGATED)); ASSERT(zio->io_child_type == ZIO_CHILD_VDEV); zio->io_physdone(zio->io_logical); } return (ZIO_PIPELINE_CONTINUE); } void zio_vdev_io_reissue(zio_t *zio) { ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); ASSERT(zio->io_error == 0); zio->io_stage >>= 1; } void zio_vdev_io_redone(zio_t *zio) { ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); zio->io_stage >>= 1; } void zio_vdev_io_bypass(zio_t *zio) { ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); ASSERT(zio->io_error == 0); zio->io_flags |= ZIO_FLAG_IO_BYPASS; zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; } /* * ========================================================================== * Generate and verify checksums * ========================================================================== */ static int zio_checksum_generate(zio_t *zio) { blkptr_t *bp = zio->io_bp; enum zio_checksum checksum; if (bp == NULL) { /* * This is zio_write_phys(). * We're either generating a label checksum, or none at all. */ checksum = zio->io_prop.zp_checksum; if (checksum == ZIO_CHECKSUM_OFF) return (ZIO_PIPELINE_CONTINUE); ASSERT(checksum == ZIO_CHECKSUM_LABEL); } else { if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { ASSERT(!IO_IS_ALLOCATING(zio)); checksum = ZIO_CHECKSUM_GANG_HEADER; } else { checksum = BP_GET_CHECKSUM(bp); } } zio_checksum_compute(zio, checksum, zio->io_abd, zio->io_size); return (ZIO_PIPELINE_CONTINUE); } static int zio_checksum_verify(zio_t *zio) { zio_bad_cksum_t info; blkptr_t *bp = zio->io_bp; int error; ASSERT(zio->io_vd != NULL); if (bp == NULL) { /* * This is zio_read_phys(). * We're either verifying a label checksum, or nothing at all. */ if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) return (ZIO_PIPELINE_CONTINUE); ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL); } if ((error = zio_checksum_error(zio, &info)) != 0) { zio->io_error = error; if (error == ECKSUM && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { zfs_ereport_start_checksum(zio->io_spa, zio->io_vd, zio, zio->io_offset, zio->io_size, NULL, &info); } } return (ZIO_PIPELINE_CONTINUE); } /* * Called by RAID-Z to ensure we don't compute the checksum twice. */ void zio_checksum_verified(zio_t *zio) { zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; } /* * ========================================================================== * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. * An error of 0 indicates success. ENXIO indicates whole-device failure, * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO * indicate errors that are specific to one I/O, and most likely permanent. * Any other error is presumed to be worse because we weren't expecting it. * ========================================================================== */ int zio_worst_error(int e1, int e2) { static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; int r1, r2; for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) if (e1 == zio_error_rank[r1]) break; for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) if (e2 == zio_error_rank[r2]) break; return (r1 > r2 ? e1 : e2); } /* * ========================================================================== * I/O completion * ========================================================================== */ static int zio_ready(zio_t *zio) { blkptr_t *bp = zio->io_bp; zio_t *pio, *pio_next; zio_link_t *zl = NULL; if (zio_wait_for_children(zio, ZIO_CHILD_GANG_BIT | ZIO_CHILD_DDT_BIT, ZIO_WAIT_READY)) { return (ZIO_PIPELINE_STOP); } if (zio->io_ready) { ASSERT(IO_IS_ALLOCATING(zio)); ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp) || (zio->io_flags & ZIO_FLAG_NOPWRITE)); ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); zio->io_ready(zio); } if (bp != NULL && bp != &zio->io_bp_copy) zio->io_bp_copy = *bp; if (zio->io_error != 0) { zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(IO_IS_ALLOCATING(zio)); ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); /* * We were unable to allocate anything, unreserve and * issue the next I/O to allocate. */ metaslab_class_throttle_unreserve( spa_normal_class(zio->io_spa), - zio->io_prop.zp_copies, zio); - zio_allocate_dispatch(zio->io_spa); + zio->io_prop.zp_copies, zio->io_allocator, zio); + zio_allocate_dispatch(zio->io_spa, zio->io_allocator); } } mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_READY] = 1; pio = zio_walk_parents(zio, &zl); mutex_exit(&zio->io_lock); /* * As we notify zio's parents, new parents could be added. * New parents go to the head of zio's io_parent_list, however, * so we will (correctly) not notify them. The remainder of zio's * io_parent_list, from 'pio_next' onward, cannot change because * all parents must wait for us to be done before they can be done. */ for (; pio != NULL; pio = pio_next) { pio_next = zio_walk_parents(zio, &zl); zio_notify_parent(pio, zio, ZIO_WAIT_READY); } if (zio->io_flags & ZIO_FLAG_NODATA) { if (BP_IS_GANG(bp)) { zio->io_flags &= ~ZIO_FLAG_NODATA; } else { ASSERT((uintptr_t)zio->io_abd < SPA_MAXBLOCKSIZE); zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; } } if (zio_injection_enabled && zio->io_spa->spa_syncing_txg == zio->io_txg) zio_handle_ignored_writes(zio); return (ZIO_PIPELINE_CONTINUE); } /* * Update the allocation throttle accounting. */ static void zio_dva_throttle_done(zio_t *zio) { zio_t *lio = zio->io_logical; zio_t *pio = zio_unique_parent(zio); vdev_t *vd = zio->io_vd; int flags = METASLAB_ASYNC_ALLOC; ASSERT3P(zio->io_bp, !=, NULL); ASSERT3U(zio->io_type, ==, ZIO_TYPE_WRITE); ASSERT3U(zio->io_priority, ==, ZIO_PRIORITY_ASYNC_WRITE); ASSERT3U(zio->io_child_type, ==, ZIO_CHILD_VDEV); ASSERT(vd != NULL); ASSERT3P(vd, ==, vd->vdev_top); ASSERT(!(zio->io_flags & (ZIO_FLAG_IO_REPAIR | ZIO_FLAG_IO_RETRY))); ASSERT(zio->io_flags & ZIO_FLAG_IO_ALLOCATING); ASSERT(!(lio->io_flags & ZIO_FLAG_IO_REWRITE)); ASSERT(!(lio->io_orig_flags & ZIO_FLAG_NODATA)); /* * Parents of gang children can have two flavors -- ones that * allocated the gang header (will have ZIO_FLAG_IO_REWRITE set) * and ones that allocated the constituent blocks. The allocation * throttle needs to know the allocating parent zio so we must find * it here. */ if (pio->io_child_type == ZIO_CHILD_GANG) { /* * If our parent is a rewrite gang child then our grandparent * would have been the one that performed the allocation. */ if (pio->io_flags & ZIO_FLAG_IO_REWRITE) pio = zio_unique_parent(pio); flags |= METASLAB_GANG_CHILD; } ASSERT(IO_IS_ALLOCATING(pio)); ASSERT3P(zio, !=, zio->io_logical); ASSERT(zio->io_logical != NULL); ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REPAIR)); ASSERT0(zio->io_flags & ZIO_FLAG_NOPWRITE); mutex_enter(&pio->io_lock); - metaslab_group_alloc_decrement(zio->io_spa, vd->vdev_id, pio, flags); + metaslab_group_alloc_decrement(zio->io_spa, vd->vdev_id, pio, flags, + pio->io_allocator, B_TRUE); mutex_exit(&pio->io_lock); metaslab_class_throttle_unreserve(spa_normal_class(zio->io_spa), - 1, pio); + 1, pio->io_allocator, pio); /* * Call into the pipeline to see if there is more work that * needs to be done. If there is work to be done it will be * dispatched to another taskq thread. */ - zio_allocate_dispatch(zio->io_spa); + zio_allocate_dispatch(zio->io_spa, pio->io_allocator); } static int zio_done(zio_t *zio) { spa_t *spa = zio->io_spa; zio_t *lio = zio->io_logical; blkptr_t *bp = zio->io_bp; vdev_t *vd = zio->io_vd; uint64_t psize = zio->io_size; zio_t *pio, *pio_next; metaslab_class_t *mc = spa_normal_class(spa); zio_link_t *zl = NULL; /* * If our children haven't all completed, * wait for them and then repeat this pipeline stage. */ if (zio_wait_for_children(zio, ZIO_CHILD_ALL_BITS, ZIO_WAIT_DONE)) { return (ZIO_PIPELINE_STOP); } /* * If the allocation throttle is enabled, then update the accounting. * We only track child I/Os that are part of an allocating async * write. We must do this since the allocation is performed * by the logical I/O but the actual write is done by child I/Os. */ if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING && zio->io_child_type == ZIO_CHILD_VDEV) { ASSERT(mc->mc_alloc_throttle_enabled); zio_dva_throttle_done(zio); } /* * If the allocation throttle is enabled, verify that * we have decremented the refcounts for every I/O that was throttled. */ if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING) { ASSERT(zio->io_type == ZIO_TYPE_WRITE); ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE); ASSERT(bp != NULL); - metaslab_group_alloc_verify(spa, zio->io_bp, zio); - VERIFY(refcount_not_held(&mc->mc_alloc_slots, zio)); + metaslab_group_alloc_verify(spa, zio->io_bp, zio, + zio->io_allocator); + VERIFY(refcount_not_held(&mc->mc_alloc_slots[zio->io_allocator], + zio)); } for (int c = 0; c < ZIO_CHILD_TYPES; c++) for (int w = 0; w < ZIO_WAIT_TYPES; w++) ASSERT(zio->io_children[c][w] == 0); if (bp != NULL && !BP_IS_EMBEDDED(bp)) { ASSERT(bp->blk_pad[0] == 0); ASSERT(bp->blk_pad[1] == 0); ASSERT(bcmp(bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 || (bp == zio_unique_parent(zio)->io_bp)); if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(bp) && zio->io_bp_override == NULL && !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { ASSERT(!BP_SHOULD_BYTESWAP(bp)); ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(bp)); ASSERT(BP_COUNT_GANG(bp) == 0 || (BP_COUNT_GANG(bp) == BP_GET_NDVAS(bp))); } if (zio->io_flags & ZIO_FLAG_NOPWRITE) VERIFY(BP_EQUAL(bp, &zio->io_bp_orig)); } /* * If there were child vdev/gang/ddt errors, they apply to us now. */ zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); zio_inherit_child_errors(zio, ZIO_CHILD_GANG); zio_inherit_child_errors(zio, ZIO_CHILD_DDT); /* * If the I/O on the transformed data was successful, generate any * checksum reports now while we still have the transformed data. */ if (zio->io_error == 0) { while (zio->io_cksum_report != NULL) { zio_cksum_report_t *zcr = zio->io_cksum_report; uint64_t align = zcr->zcr_align; uint64_t asize = P2ROUNDUP(psize, align); char *abuf = NULL; abd_t *adata = zio->io_abd; if (asize != psize) { adata = abd_alloc_linear(asize, B_TRUE); abd_copy(adata, zio->io_abd, psize); abd_zero_off(adata, psize, asize - psize); } if (adata != NULL) abuf = abd_borrow_buf_copy(adata, asize); zio->io_cksum_report = zcr->zcr_next; zcr->zcr_next = NULL; zcr->zcr_finish(zcr, abuf); zfs_ereport_free_checksum(zcr); if (adata != NULL) abd_return_buf(adata, abuf, asize); if (asize != psize) abd_free(adata); } } zio_pop_transforms(zio); /* note: may set zio->io_error */ vdev_stat_update(zio, psize); if (zio->io_error) { /* * If this I/O is attached to a particular vdev, * generate an error message describing the I/O failure * at the block level. We ignore these errors if the * device is currently unavailable. */ if (zio->io_error != ECKSUM && vd != NULL && !vdev_is_dead(vd)) zfs_ereport_post(FM_EREPORT_ZFS_IO, spa, vd, zio, 0, 0); if ((zio->io_error == EIO || !(zio->io_flags & (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && zio == lio) { /* * For logical I/O requests, tell the SPA to log the * error and generate a logical data ereport. */ spa_log_error(spa, zio); zfs_ereport_post(FM_EREPORT_ZFS_DATA, spa, NULL, zio, 0, 0); } } if (zio->io_error && zio == lio) { /* * Determine whether zio should be reexecuted. This will * propagate all the way to the root via zio_notify_parent(). */ ASSERT(vd == NULL && bp != NULL); ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); if (IO_IS_ALLOCATING(zio) && !(zio->io_flags & ZIO_FLAG_CANFAIL)) { if (zio->io_error != ENOSPC) zio->io_reexecute |= ZIO_REEXECUTE_NOW; else zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; } if ((zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_FREE) && !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_error == ENXIO && spa_load_state(spa) == SPA_LOAD_NONE && spa_get_failmode(spa) != ZIO_FAILURE_MODE_CONTINUE) zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; /* * Here is a possibly good place to attempt to do * either combinatorial reconstruction or error correction * based on checksums. It also might be a good place * to send out preliminary ereports before we suspend * processing. */ } /* * If there were logical child errors, they apply to us now. * We defer this until now to avoid conflating logical child * errors with errors that happened to the zio itself when * updating vdev stats and reporting FMA events above. */ zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); if ((zio->io_error || zio->io_reexecute) && IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && !(zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE))) zio_dva_unallocate(zio, zio->io_gang_tree, bp); zio_gang_tree_free(&zio->io_gang_tree); /* * Godfather I/Os should never suspend. */ if ((zio->io_flags & ZIO_FLAG_GODFATHER) && (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) zio->io_reexecute = 0; if (zio->io_reexecute) { /* * This is a logical I/O that wants to reexecute. * * Reexecute is top-down. When an i/o fails, if it's not * the root, it simply notifies its parent and sticks around. * The parent, seeing that it still has children in zio_done(), * does the same. This percolates all the way up to the root. * The root i/o will reexecute or suspend the entire tree. * * This approach ensures that zio_reexecute() honors * all the original i/o dependency relationships, e.g. * parents not executing until children are ready. */ ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); zio->io_gang_leader = NULL; mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_DONE] = 1; mutex_exit(&zio->io_lock); /* * "The Godfather" I/O monitors its children but is * not a true parent to them. It will track them through * the pipeline but severs its ties whenever they get into * trouble (e.g. suspended). This allows "The Godfather" * I/O to return status without blocking. */ zl = NULL; for (pio = zio_walk_parents(zio, &zl); pio != NULL; pio = pio_next) { zio_link_t *remove_zl = zl; pio_next = zio_walk_parents(zio, &zl); if ((pio->io_flags & ZIO_FLAG_GODFATHER) && (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { zio_remove_child(pio, zio, remove_zl); zio_notify_parent(pio, zio, ZIO_WAIT_DONE); } } if ((pio = zio_unique_parent(zio)) != NULL) { /* * We're not a root i/o, so there's nothing to do * but notify our parent. Don't propagate errors * upward since we haven't permanently failed yet. */ ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; zio_notify_parent(pio, zio, ZIO_WAIT_DONE); } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { /* * We'd fail again if we reexecuted now, so suspend * until conditions improve (e.g. device comes online). */ zio_suspend(spa, zio); } else { /* * Reexecution is potentially a huge amount of work. * Hand it off to the otherwise-unused claim taskq. */ #if defined(illumos) || !defined(_KERNEL) ASSERT(zio->io_tqent.tqent_next == NULL); #else ASSERT(zio->io_tqent.tqent_task.ta_pending == 0); #endif spa_taskq_dispatch_ent(spa, ZIO_TYPE_CLAIM, ZIO_TASKQ_ISSUE, (task_func_t *)zio_reexecute, zio, 0, &zio->io_tqent); } return (ZIO_PIPELINE_STOP); } ASSERT(zio->io_child_count == 0); ASSERT(zio->io_reexecute == 0); ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); /* * Report any checksum errors, since the I/O is complete. */ while (zio->io_cksum_report != NULL) { zio_cksum_report_t *zcr = zio->io_cksum_report; zio->io_cksum_report = zcr->zcr_next; zcr->zcr_next = NULL; zcr->zcr_finish(zcr, NULL); zfs_ereport_free_checksum(zcr); } /* * It is the responsibility of the done callback to ensure that this * particular zio is no longer discoverable for adoption, and as * such, cannot acquire any new parents. */ if (zio->io_done) zio->io_done(zio); mutex_enter(&zio->io_lock); zio->io_state[ZIO_WAIT_DONE] = 1; mutex_exit(&zio->io_lock); zl = NULL; for (pio = zio_walk_parents(zio, &zl); pio != NULL; pio = pio_next) { zio_link_t *remove_zl = zl; pio_next = zio_walk_parents(zio, &zl); zio_remove_child(pio, zio, remove_zl); zio_notify_parent(pio, zio, ZIO_WAIT_DONE); } if (zio->io_waiter != NULL) { mutex_enter(&zio->io_lock); zio->io_executor = NULL; cv_broadcast(&zio->io_cv); mutex_exit(&zio->io_lock); } else { zio_destroy(zio); } return (ZIO_PIPELINE_STOP); } /* * ========================================================================== * I/O pipeline definition * ========================================================================== */ static zio_pipe_stage_t *zio_pipeline[] = { NULL, zio_read_bp_init, zio_write_bp_init, zio_free_bp_init, zio_issue_async, zio_write_compress, zio_checksum_generate, zio_nop_write, zio_ddt_read_start, zio_ddt_read_done, zio_ddt_write, zio_ddt_free, zio_gang_assemble, zio_gang_issue, zio_dva_throttle, zio_dva_allocate, zio_dva_free, zio_dva_claim, zio_ready, zio_vdev_io_start, zio_vdev_io_done, zio_vdev_io_assess, zio_checksum_verify, zio_done }; /* * Compare two zbookmark_phys_t's to see which we would reach first in a * pre-order traversal of the object tree. * * This is simple in every case aside from the meta-dnode object. For all other * objects, we traverse them in order (object 1 before object 2, and so on). * However, all of these objects are traversed while traversing object 0, since * the data it points to is the list of objects. Thus, we need to convert to a * canonical representation so we can compare meta-dnode bookmarks to * non-meta-dnode bookmarks. * * We do this by calculating "equivalents" for each field of the zbookmark. * zbookmarks outside of the meta-dnode use their own object and level, and * calculate the level 0 equivalent (the first L0 blkid that is contained in the * blocks this bookmark refers to) by multiplying their blkid by their span * (the number of L0 blocks contained within one block at their level). * zbookmarks inside the meta-dnode calculate their object equivalent * (which is L0equiv * dnodes per data block), use 0 for their L0equiv, and use * level + 1<<31 (any value larger than a level could ever be) for their level. * This causes them to always compare before a bookmark in their object * equivalent, compare appropriately to bookmarks in other objects, and to * compare appropriately to other bookmarks in the meta-dnode. */ int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2, uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2) { /* * These variables represent the "equivalent" values for the zbookmark, * after converting zbookmarks inside the meta dnode to their * normal-object equivalents. */ uint64_t zb1obj, zb2obj; uint64_t zb1L0, zb2L0; uint64_t zb1level, zb2level; if (zb1->zb_object == zb2->zb_object && zb1->zb_level == zb2->zb_level && zb1->zb_blkid == zb2->zb_blkid) return (0); /* * BP_SPANB calculates the span in blocks. */ zb1L0 = (zb1->zb_blkid) * BP_SPANB(ibs1, zb1->zb_level); zb2L0 = (zb2->zb_blkid) * BP_SPANB(ibs2, zb2->zb_level); if (zb1->zb_object == DMU_META_DNODE_OBJECT) { zb1obj = zb1L0 * (dbss1 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT)); zb1L0 = 0; zb1level = zb1->zb_level + COMPARE_META_LEVEL; } else { zb1obj = zb1->zb_object; zb1level = zb1->zb_level; } if (zb2->zb_object == DMU_META_DNODE_OBJECT) { zb2obj = zb2L0 * (dbss2 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT)); zb2L0 = 0; zb2level = zb2->zb_level + COMPARE_META_LEVEL; } else { zb2obj = zb2->zb_object; zb2level = zb2->zb_level; } /* Now that we have a canonical representation, do the comparison. */ if (zb1obj != zb2obj) return (zb1obj < zb2obj ? -1 : 1); else if (zb1L0 != zb2L0) return (zb1L0 < zb2L0 ? -1 : 1); else if (zb1level != zb2level) return (zb1level > zb2level ? -1 : 1); /* * This can (theoretically) happen if the bookmarks have the same object * and level, but different blkids, if the block sizes are not the same. * There is presently no way to change the indirect block sizes */ return (0); } /* * This function checks the following: given that last_block is the place that * our traversal stopped last time, does that guarantee that we've visited * every node under subtree_root? Therefore, we can't just use the raw output * of zbookmark_compare. We have to pass in a modified version of * subtree_root; by incrementing the block id, and then checking whether * last_block is before or equal to that, we can tell whether or not having * visited last_block implies that all of subtree_root's children have been * visited. */ boolean_t zbookmark_subtree_completed(const dnode_phys_t *dnp, const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block) { zbookmark_phys_t mod_zb = *subtree_root; mod_zb.zb_blkid++; ASSERT(last_block->zb_level == 0); /* The objset_phys_t isn't before anything. */ if (dnp == NULL) return (B_FALSE); /* * We pass in 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT) for the * data block size in sectors, because that variable is only used if * the bookmark refers to a block in the meta-dnode. Since we don't * know without examining it what object it refers to, and there's no * harm in passing in this value in other cases, we always pass it in. * * We pass in 0 for the indirect block size shift because zb2 must be * level 0. The indirect block size is only used to calculate the span * of the bookmark, but since the bookmark must be level 0, the span is * always 1, so the math works out. * * If you make changes to how the zbookmark_compare code works, be sure * to make sure that this code still works afterwards. */ return (zbookmark_compare(dnp->dn_datablkszsec, dnp->dn_indblkshift, 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT), 0, &mod_zb, last_block) <= 0); } Index: head/sys/cddl/contrib/opensolaris =================================================================== --- head/sys/cddl/contrib/opensolaris (revision 336948) +++ head/sys/cddl/contrib/opensolaris (revision 336949) Property changes on: head/sys/cddl/contrib/opensolaris ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /vendor-sys/illumos/dist:r336948