metaslab.c revision 30beaff42d8240ebf5386e8b7a14e3d137a1631f
fa9e4066f08beec538e775443c5be79dd423fcabahrens * CDDL HEADER START
fa9e4066f08beec538e775443c5be79dd423fcabahrens * The contents of this file are subject to the terms of the
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * Common Development and Distribution License (the "License").
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * You may not use this file except in compliance with the License.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
fa9e4066f08beec538e775443c5be79dd423fcabahrens * See the License for the specific language governing permissions
fa9e4066f08beec538e775443c5be79dd423fcabahrens * and limitations under the License.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * When distributing Covered Code, include this CDDL HEADER in each
fa9e4066f08beec538e775443c5be79dd423fcabahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If applicable, add the following below this CDDL HEADER, with the
fa9e4066f08beec538e775443c5be79dd423fcabahrens * fields enclosed by brackets "[]" replaced with your own identifying
fa9e4066f08beec538e775443c5be79dd423fcabahrens * information: Portions Copyright [yyyy] [name of copyright owner]
fa9e4066f08beec538e775443c5be79dd423fcabahrens * CDDL HEADER END
8d18220deb04ec7b12410cd90deb4d45e66d49bfMark J Musante * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
bf16b11e8deb633dd6c4296d46e92399d1582df4Matthew Ahrens * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
9dc3941c735ef88de46e850f745aa556d3a071a5Sašo Kiselkov * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * Allow allocations to switch to gang blocks quickly. We do this to
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * avoid having to load lots of space_maps in a given txg. There are,
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * however, some cases where we want to avoid "fast" ganging and instead
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * we want to do an exhaustive search of all metaslabs on this device.
b6240e830b871f59c22a3918aebb3b36c872edbaGeorge Wilson * Currently we don't allow any gang, slog, or dump device related allocations
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * to "fast" gang.
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson (!((flags) & (METASLAB_GANG_CHILD | METASLAB_GANG_HEADER | \
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson (METASLAB_WEIGHT_PRIMARY | METASLAB_WEIGHT_SECONDARY)
e05725b117836db173257fae43fb0746eb857fb5bonwickuint64_t metaslab_gang_bang = SPA_MAXBLOCKSIZE + 1; /* force gang blocks */
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * The in-core space map representation is more compact than its on-disk form.
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * The zfs_condense_pct determines how much more compact the in-core
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * space_map representation must be before we compact it on-disk.
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * Values should be greater than or equal to 100.
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * The zfs_mg_noalloc_threshold defines which metaslab groups should
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * be eligible for allocation. The value is defined as a percentage of
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * a free space. Metaslab groups that have more free space than
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * zfs_mg_noalloc_threshold are always eligible for allocations. Once
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * a metaslab group's free space is less than or equal to the
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * zfs_mg_noalloc_threshold the allocator will avoid allocating to that
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * group unless all groups in the pool have reached zfs_mg_noalloc_threshold.
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * Once all groups in the pool reach zfs_mg_noalloc_threshold then all
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * groups are allowed to accept allocations. Gang blocks are always
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * eligible to allocate on any metaslab group. The default value of 0 means
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * no metaslab group will be excluded based on this criterion.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * When set will load all metaslabs when pool is first opened.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * When set will prevent metaslabs from being unloaded.
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson * Minimum size which forces the dynamic allocator to change
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * it's allocation strategy. Once the space map cannot satisfy
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson * an allocation of this size then it switches to using more
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson * aggressive strategy (i.e search by size rather than offset).
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilsonuint64_t metaslab_df_alloc_threshold = SPA_MAXBLOCKSIZE;
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson * The minimum free space, in percent, which must be available
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson * in a space map to continue allocations in a first-fit fashion.
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson * Once the space_map's free space drops below this level we dynamically
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson * switch to using best-fit allocations.
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * A metaslab is considered "free" if it contains a contiguous
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * segment which is greater than metaslab_min_alloc_size.
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilsonuint64_t metaslab_min_alloc_size = DMU_MAX_ACCESS;
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Percentage of all cpus that can be used by the metaslab taskq.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Determines how many txgs a metaslab may remain loaded without having any
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * allocations from it. As long as a metaslab continues to be used we will
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * keep it loaded.
9dc3941c735ef88de46e850f745aa556d3a071a5Sašo Kiselkov * Should we be willing to write data to degraded vdevs?
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Max number of metaslabs per group to preload.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Enable/disable preloading of metaslab.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Enable/disable additional weight factor for each metaslab.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonboolean_t metaslab_weight_factor_enable = B_FALSE;
fa9e4066f08beec538e775443c5be79dd423fcabahrens * ==========================================================================
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Metaslab classes
fa9e4066f08beec538e775443c5be79dd423fcabahrens * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_class_create(spa_t *spa, metaslab_ops_t *ops)
fa9e4066f08beec538e775443c5be79dd423fcabahrens mc = kmem_zalloc(sizeof (metaslab_class_t), KM_SLEEP);
88ecc943b4eb72f7c4fbbd8435997b85ef171fc3George Wilson * Must hold one of the spa_config locks.
88ecc943b4eb72f7c4fbbd8435997b85ef171fc3George Wilson ASSERT(spa_config_held(mc->mc_spa, SCL_ALL, RW_READER) ||
88ecc943b4eb72f7c4fbbd8435997b85ef171fc3George Wilson spa_config_held(mc->mc_spa, SCL_ALL, RW_WRITER));
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwickmetaslab_class_space_update(metaslab_class_t *mc, int64_t alloc_delta,
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick int64_t defer_delta, int64_t space_delta, int64_t dspace_delta)
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwickmetaslab_class_get_deferred(metaslab_class_t *mc)
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick return (spa_deflate(mc->mc_spa) ? mc->mc_dspace : mc->mc_space);
fa9e4066f08beec538e775443c5be79dd423fcabahrens * ==========================================================================
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Metaslab groups
fa9e4066f08beec538e775443c5be79dd423fcabahrens * ==========================================================================
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (1);
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (-1);
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If the weights are identical, use the offset to force uniqueness.
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (-1);
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (1);
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (0);
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * Update the allocatable flag and the metaslab group's capacity.
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * The allocatable flag is set to true if the capacity is below
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * the zfs_mg_noalloc_threshold. If a metaslab group transitions
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * from allocatable to non-allocatable or vice versa then the metaslab
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * group's class is updated to reflect the transition.
22e30981d82a0b6dc89253596ededafae8655e00George Wilsonmetaslab_group_alloc_update(metaslab_group_t *mg)
22e30981d82a0b6dc89253596ededafae8655e00George Wilson mg->mg_free_capacity = ((vs->vs_space - vs->vs_alloc) * 100) /
22e30981d82a0b6dc89253596ededafae8655e00George Wilson mg->mg_allocatable = (mg->mg_free_capacity > zfs_mg_noalloc_threshold);
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * The mc_alloc_groups maintains a count of the number of
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * groups in this metaslab class that are still above the
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * zfs_mg_noalloc_threshold. This is used by the allocating
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * threads to determine if they should avoid allocations to
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * a given group. The allocator will avoid allocations to a group
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * if that group has reached or is below the zfs_mg_noalloc_threshold
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * and there are still other groups that are above the threshold.
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * When a group transitions from allocatable to non-allocatable or
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * vice versa we update the metaslab class to reflect that change.
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * When the mc_alloc_groups value drops to 0 that means that all
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * groups have reached the zfs_mg_noalloc_threshold making all groups
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * eligible for allocations. This effectively means that all devices
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * are balanced again.
22e30981d82a0b6dc89253596ededafae8655e00George Wilson else if (!was_allocatable && mg->mg_allocatable)
fa9e4066f08beec538e775443c5be79dd423fcabahrensmetaslab_group_create(metaslab_class_t *mc, vdev_t *vd)
fa9e4066f08beec538e775443c5be79dd423fcabahrens mg = kmem_zalloc(sizeof (metaslab_group_t), KM_SLEEP);
fa9e4066f08beec538e775443c5be79dd423fcabahrens sizeof (metaslab_t), offsetof(struct metaslab, ms_group_node));
be082110c08433beadb738ad3be035a73d995ea8George Wilson mg->mg_taskq = taskq_create("metaslab_group_taskq", metaslab_load_pct,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson minclsyspri, 10, INT_MAX, TASKQ_THREADS_CPU_PCT);
a33cae9802e94744efee12a7a77c89360645eae8Tim Haley * We may have gone below zero with the activation count
a33cae9802e94744efee12a7a77c89360645eae8Tim Haley * either because we never activated in the first place or
a33cae9802e94744efee12a7a77c89360645eae8Tim Haley * because we're done, and possibly removing the vdev.
a15215608b8bd90f714f6db21ee623b584607cb6Jeff Bonwick ASSERT(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_WRITER));
a15215608b8bd90f714f6db21ee623b584607cb6Jeff Bonwick mg->mg_aliquot = metaslab_aliquot * MAX(1, mg->mg_vd->vdev_children);
a15215608b8bd90f714f6db21ee623b584607cb6Jeff Bonwick ASSERT(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_WRITER));
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwickmetaslab_group_add(metaslab_group_t *mg, metaslab_t *msp)
fa9e4066f08beec538e775443c5be79dd423fcabahrensmetaslab_group_remove(metaslab_group_t *mg, metaslab_t *msp)
fa9e4066f08beec538e775443c5be79dd423fcabahrensmetaslab_group_sort(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight)
5f5f7a6f9c8e9c1587a54e690556d756ec67558cahrens * Although in principle the weight can be any value, in
5f5f7a6f9c8e9c1587a54e690556d756ec67558cahrens * practice we do not use values in the range [1, 510].
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * Determine if a given metaslab group should skip allocations. A metaslab
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * group should avoid allocations if its used capacity has crossed the
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * zfs_mg_noalloc_threshold and there is at least one metaslab group
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * that can still handle allocations.
22e30981d82a0b6dc89253596ededafae8655e00George Wilsonmetaslab_group_allocatable(metaslab_group_t *mg)
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * A metaslab group is considered allocatable if its free capacity
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * is greater than the set value of zfs_mg_noalloc_threshold, it's
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * associated with a slog, or there are no other metaslab groups
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * with free capacity greater than zfs_mg_noalloc_threshold.
22e30981d82a0b6dc89253596ededafae8655e00George Wilson return (mg->mg_free_capacity > zfs_mg_noalloc_threshold ||
22e30981d82a0b6dc89253596ededafae8655e00George Wilson mc != spa_normal_class(spa) || mc->mc_alloc_groups == 0);
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Range tree callbacks
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Comparison function for the private size-ordered tree. Tree is sorted
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * by size, larger sizes at the end of the tree.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_rangesize_compare(const void *x1, const void *x2)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson uint64_t rs_size1 = r1->rs_end - r1->rs_start;
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson uint64_t rs_size2 = r2->rs_end - r2->rs_start;
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Create any block allocator specific components. The current allocators
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * rely on using both a size-ordered range_tree_t and an array of uint64_t's.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_rt_create(range_tree_t *rt, void *arg)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson avl_create(&msp->ms_size_tree, metaslab_rangesize_compare,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson sizeof (range_seg_t), offsetof(range_seg_t, rs_pp_node));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Destroy the block allocator specific components.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_rt_destroy(range_tree_t *rt, void *arg)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_rt_add(range_tree_t *rt, range_seg_t *rs, void *arg)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_rt_remove(range_tree_t *rt, range_seg_t *rs, void *arg)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_rt_vacate(range_tree_t *rt, void *arg)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Normally one would walk the tree freeing nodes along the way.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Since the nodes are shared with the range trees we can avoid
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * walking all nodes and just reinitialize the avl tree. The nodes
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * will be freed by the range tree, so we don't want to free them here.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson avl_create(&msp->ms_size_tree, metaslab_rangesize_compare,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson sizeof (range_seg_t), offsetof(range_seg_t, rs_pp_node));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Metaslab block operations
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * ==========================================================================
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * Return the maximum contiguous segment within the metaslab.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_block_alloc(metaslab_t *msp, uint64_t size)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY0(P2PHASE(start, 1ULL << vd->vdev_ashift));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY3U(range_tree_space(rt) - size, <=, msp->ms_size);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Common allocator routines
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * This is a helper function that can be used by the allocator to find
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * a suitable block to allocate. This will search the specified AVL
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * tree looking for a block that matches the specified criteria.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_block_picker(avl_tree_t *t, uint64_t *cursor, uint64_t size,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson uint64_t offset = P2ROUNDUP(rs->rs_start, align);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * If we know we've searched the whole map (*cursor == 0), give up.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Otherwise, reset the cursor to the beginning and try again.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson return (-1ULL);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson return (metaslab_block_picker(t, cursor, size, align));
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * ==========================================================================
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * The first-fit block allocator
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_ff_alloc(metaslab_t *msp, uint64_t size)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Find the largest power of 2 block size that evenly divides the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * requested size. This is used to try to allocate blocks with similar
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * alignment from the same area of the metaslab (i.e. same cursor
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * bucket) but it does not guarantee that other allocations sizes
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * may exist in the same region.
bf16b11e8deb633dd6c4296d46e92399d1582df4Matthew Ahrens uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1];
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson return (metaslab_block_picker(t, cursor, size, align));
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson/* ARGSUSED */
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * ==========================================================================
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * Dynamic block allocator -
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * Uses the first fit allocation scheme until space get low and then
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * adjusts to a best fit allocation method. Uses metaslab_df_alloc_threshold
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * and metaslab_df_free_pct to determine when to switch the allocation scheme.
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_df_alloc(metaslab_t *msp, uint64_t size)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Find the largest power of 2 block size that evenly divides the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * requested size. This is used to try to allocate blocks with similar
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * alignment from the same area of the metaslab (i.e. same cursor
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * bucket) but it does not guarantee that other allocations sizes
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * may exist in the same region.
bf16b11e8deb633dd6c4296d46e92399d1582df4Matthew Ahrens uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1];
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson uint64_t max_size = metaslab_block_maxsize(msp);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson int free_pct = range_tree_space(rt) * 100 / msp->ms_size;
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_size_tree));
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson return (-1ULL);
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson * If we're running low on space switch to using the size
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson * sorted AVL tree (best-fit).
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson return (metaslab_block_picker(t, cursor, size, 1ULL));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson uint64_t max_size = metaslab_block_maxsize(msp);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson int free_pct = range_tree_space(rt) * 100 / msp->ms_size;
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson if (max_size >= metaslab_df_alloc_threshold &&
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Cursor fit block allocator -
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Select the largest region in the metaslab, set the cursor to the beginning
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * of the range and the cursor_end to the end of the range. As allocations
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * are made advance the cursor. Continue allocating from the cursor until
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * the range is exhausted and then find a new range.
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_cf_alloc(metaslab_t *msp, uint64_t size)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&rt->rt_root));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (rs == NULL || (rs->rs_end - rs->rs_start) < size)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson return (-1ULL);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson return (metaslab_block_maxsize(msp) < metaslab_min_alloc_size);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * New dynamic fit allocator -
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Select a region that is large enough to allocate 2^metaslab_ndf_clump_shift
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * contiguous blocks. If no region is found then just use the largest segment
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * that remains.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Determines desired number of contiguous blocks (2^metaslab_ndf_clump_shift)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * to request from the allocator.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_ndf_alloc(metaslab_t *msp, uint64_t size)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson uint64_t max_size = metaslab_block_maxsize(msp);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_size_tree));
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson return (-1ULL);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (rs == NULL || (rs->rs_end - rs->rs_start) < size) {
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson return (-1ULL);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson (metaslab_min_alloc_size << metaslab_ndf_clump_shift));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_ops_t *zfs_metaslab_ops = &metaslab_df_ops;
fa9e4066f08beec538e775443c5be79dd423fcabahrens * ==========================================================================
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Metaslabs
fa9e4066f08beec538e775443c5be79dd423fcabahrens * ==========================================================================
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Wait for any in-progress metaslab loads to complete.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * If the space map has not been allocated yet, then treat
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * all the space in the metaslab as free and add it to the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson error = space_map_load(msp->ms_sm, msp->ms_tree, SM_FREE);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_add(msp->ms_tree, msp->ms_start, msp->ms_size);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson for (int t = 0; t < TXG_DEFER_SIZE; t++) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilsonmetaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object, uint64_t txg)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson objset_t *mos = vd->vdev_spa->spa_meta_objset;
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson cv_init(&msp->ms_load_cv, NULL, CV_DEFAULT, NULL);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * We only open space map objects that already exist. All others
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * will be opened when we finally allocate an object for it.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY0(space_map_open(&msp->ms_sm, mos, object, msp->ms_start,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson msp->ms_size, vd->vdev_ashift, &msp->ms_lock));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * We create the main range tree here, but we don't create the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * alloctree and freetree until metaslab_sync_done(). This serves
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * two purposes: it allows metaslab_sync_done() to detect the
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * addition of new space; and for debugging, it ensures that we'd
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * data fault on any attempt to use this metaslab before it's ready.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson msp->ms_tree = range_tree_create(&metaslab_rt_ops, msp, &msp->ms_lock);
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If we're opening an existing pool (txg == 0) or creating
fa9e4066f08beec538e775443c5be79dd423fcabahrens * a new one (txg == TXG_INITIAL), all space is available now.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If we're adding space to an existing pool, the new space
fa9e4066f08beec538e775443c5be79dd423fcabahrens * does not become available until after this txg has synced.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * If metaslab_debug_load is set and we're initializing a metaslab
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * that has an allocated space_map object then load the its space
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * map so that can verify frees.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (metaslab_debug_load && msp->ms_sm != NULL) {
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick if (txg != 0) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson vdev_space_update(mg->mg_vd, -space_map_allocated(msp->ms_sm),
468c413a79615e77179e8d98f22a7e513a8135bdTim Haley for (int t = 0; t < TXG_SIZE; t++) {
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson for (int t = 0; t < TXG_DEFER_SIZE; t++) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Apply a weighting factor based on the histogram information for this
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * metaslab. The current weighting factor is somewhat arbitrary and requires
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * additional investigation. The implementation provides a measure of
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * "weighted" free space and gives a higher weighting for larger contiguous
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * regions. The weighting factor is determined by counting the number of
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * sm_shift sectors that exist in each region represented by the histogram.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * That value is then multiplied by the power of 2 exponent and the sm_shift
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * For example, assume the 2^21 histogram bucket has 4 2MB regions and the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * metaslab has an sm_shift value of 9 (512B):
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * 1) calculate the number of sm_shift sectors in the region:
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * 2^21 / 2^9 = 2^12 = 4096 * 4 (number of regions) = 16384
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * 2) multiply by the power of 2 exponent and the sm_shift value:
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * 16384 * 21 * 9 = 3096576
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * This value will be added to the weighting of the metaslab.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * A null space map means that the entire metaslab is free,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * calculate a weight factor that spans the entire size of the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (msp->ms_sm->sm_dbuf->db_size != sizeof (space_map_phys_t))
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson for (i = 0; i < SPACE_MAP_HISTOGRAM_SIZE(msp->ms_sm); i++) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (msp->ms_sm->sm_phys->smp_histogram[i] == 0)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Determine the number of sm_shift sectors in the region
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * indicated by the histogram. For example, given an
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * sm_shift value of 9 (512 bytes) and i = 4 then we know
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * that we're looking at an 8K region in the histogram
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * (i.e. 9 + 4 = 13, 2^13 = 8192). To figure out the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * number of sm_shift sectors (512 bytes in this example),
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * we would take 8192 / 512 = 16. Since the histogram
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * is offset by sm_shift we can simply use the value of
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * of i to calculate this (i.e. 2^i = 16 where i = 4).
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson sectors = msp->ms_sm->sm_phys->smp_histogram[i] << i;
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson factor += (i + msp->ms_sm->sm_shift) * sectors;
9eb57f7f3fbb970d4b9b89dcd5ecf543fe2414d5George Wilson * This vdev is in the process of being removed so there is nothing
9eb57f7f3fbb970d4b9b89dcd5ecf543fe2414d5George Wilson * for us to do here.
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * The baseline weight is the metaslab's free space.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson space = msp->ms_size - space_map_allocated(msp->ms_sm);
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * Modern disks have uniform bit density and constant angular velocity.
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * Therefore, the outer recording zones are faster (higher bandwidth)
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * than the inner zones by the ratio of outer to inner track diameter,
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * which is typically around 2:1. We account for this by assigning
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * higher weight to lower metaslabs (multiplier ranging from 2x to 1x).
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * In effect, this means that we'll select the metaslab with the most
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * free bandwidth rather than simply the one with the most free space.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson weight = 2 * weight - (msp->ms_id * weight) / vd->vdev_ms_count;
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (msp->ms_loaded && !msp->ms_ops->msop_fragmented(msp)) {
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * If this metaslab is one we're actively using, adjust its
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * weight to make it preferable to any inactive metaslab so
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson * we'll polish it off.
80eb36f241abf8c076119fb4c49a55fd61ebc710George Wilson weight |= (msp->ms_weight & METASLAB_ACTIVE_MASK);
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilsonmetaslab_activate(metaslab_t *msp, uint64_t activation_weight)
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick return (0);
5f5f7a6f9c8e9c1587a54e690556d756ec67558cahrens * If size < SPA_MINBLOCKSIZE, then we will not allocate from
5f5f7a6f9c8e9c1587a54e690556d756ec67558cahrens * this metaslab again. In that case, it had better be empty,
5f5f7a6f9c8e9c1587a54e690556d756ec67558cahrens * or we would be leaving space on the table.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson ASSERT(size >= SPA_MINBLOCKSIZE || range_tree_space(msp->ms_tree) == 0);
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm metaslab_group_sort(msp->ms_group, msp, MIN(msp->ms_weight, size));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Set the ms_access_txg value so that we don't unload it right away.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson msp->ms_access_txg = spa_syncing_txg(spa) + metaslab_unload_delay + 1;
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (spa_shutting_down(spa) || !metaslab_preload_enabled) {
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * Load the next potential metaslabs
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson /* If we have reached our preload limit then we're done */
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * We must drop the metaslab group lock here to preserve
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * lock ordering with the ms_lock (when grabbing both
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * the mg_lock and the ms_lock, the ms_lock must be taken
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * first). As a result, it is possible that the ordering
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * of the metaslabs within the avl tree may change before
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * we reacquire the lock. The metaslab cannot be removed from
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * the tree while we're in syncing context so it is safe to
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * drop the mg_lock here. If the metaslabs are reordered
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * nothing will break -- we just may end up loading a
30beaff42d8240ebf5386e8b7a14e3d137a1631fGeorge Wilson * less than optimal one.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY(taskq_dispatch(mg->mg_taskq, metaslab_preload,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Determine if the space map's on-disk footprint is past our tolerance
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * for inefficiency. We would like to use the following criteria to make
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * our decision:
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * 1. The size of the space map object should not dramatically increase as a
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * result of writing out the free space range tree.
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * 2. The minimal on-disk space map representation is zfs_condense_pct/100
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * times the size than the free space range tree representation
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * (i.e. zfs_condense_pct = 110 and in-core = 1MB, minimal = 1.1.MB).
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * Checking the first condition is tricky since we don't want to walk
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * the entire AVL tree calculating the estimated on-disk size. Instead we
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * use the size-ordered range tree in the metaslab and calculate the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * size required to write out the largest segment in our free tree. If the
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * size required to represent that segment on disk is larger than the space
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * map object then we avoid condensing this map.
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * To determine the second criterion we use a best-case estimate and assume
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * each segment can be represented on-disk as a single 64-bit entry. We refer
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * to this best-case estimate as the space map's minimal form.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Use the ms_size_tree range tree, which is ordered by size, to
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * obtain the largest segment in the free tree. If the tree is empty
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * then we should condense the map.
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * Calculate the number of 64-bit entries this segment would
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * require when written to disk. If this single segment would be
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * larger on-disk than the entire current on-disk structure, then
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * clearly condensing will increase the on-disk structure size.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson size = (rs->rs_end - rs->rs_start) >> sm->sm_shift;
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson return (segsz <= space_map_length(msp->ms_sm) &&
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson space_map_length(msp->ms_sm) >= (zfs_condense_pct *
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson sizeof (uint64_t) * avl_numnodes(&msp->ms_tree->rt_root)) / 100);
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * Condense the on-disk space map representation to its minimized form.
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * The minimized form consists of a small number of allocations followed by
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * the entries of the free range tree.
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilsonmetaslab_condense(metaslab_t *msp, uint64_t txg, dmu_tx_t *tx)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_t *freetree = msp->ms_freetree[txg & TXG_MASK];
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson spa_dbgmsg(spa, "condensing: txg %llu, msp[%llu] %p, "
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson "smp size %llu, segments %lu", txg, msp->ms_id, msp,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson space_map_length(msp->ms_sm), avl_numnodes(&msp->ms_tree->rt_root));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Create an range tree that is 100% allocated. We remove segments
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * that have been freed in this txg, any deferred frees that exist,
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * and any allocation in the future. Removing segments should be
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * a relatively inexpensive operation since we expect these trees to
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * have a small number of nodes.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson condense_tree = range_tree_create(NULL, NULL, &msp->ms_lock);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_add(condense_tree, msp->ms_start, msp->ms_size);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Remove what's been freed in this txg from the condense_tree.
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * Since we're in sync_pass 1, we know that all the frees from
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * this txg are in the freetree.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_walk(freetree, range_tree_remove, condense_tree);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson for (int t = 0; t < TXG_DEFER_SIZE; t++) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson for (int t = 1; t < TXG_CONCURRENT_STATES; t++) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_walk(msp->ms_alloctree[(txg + t) & TXG_MASK],
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * We're about to drop the metaslab's lock thus allowing
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * other consumers to change it's content. Set the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * metaslab's ms_condensing flag to ensure that
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * allocations on this metaslab do not occur while we're
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * in the middle of committing it to disk. This is only critical
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * for the ms_tree as all other range trees use per txg
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * views of their content.
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * While we would ideally like to create a space_map representation
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * that consists only of allocation records, doing so can be
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * prohibitively expensive because the in-core free tree can be
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * large, and therefore computationally expensive to subtract
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * from the condense_tree. Instead we sync out two trees, a cheap
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * allocation only tree followed by the in-core free tree. While not
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * optimal, this is typically close to optimal, and much cheaper to
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson space_map_write(sm, condense_tree, SM_ALLOC, tx);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson space_map_write(sm, msp->ms_tree, SM_FREE, tx);
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Write a metaslab to disk in the context of the specified transaction group.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_t *alloctree = msp->ms_alloctree[txg & TXG_MASK];
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_t **freetree = &msp->ms_freetree[txg & TXG_MASK];
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson uint64_t object = space_map_object(msp->ms_sm);
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * This metaslab has just been added so there's no work to do now.
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * The only state that can actually be changing concurrently with
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * metaslab_sync() is the metaslab's ms_tree. No other thread can
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * be modifying this txg's alloctree, freetree, freed_tree, or
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * space_map_phys_t. Therefore, we only hold ms_lock to satify
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * space_map ASSERTs. We drop it whenever we call into the DMU,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * because the DMU can call down to us (e.g. via zio_free()) at
468c413a79615e77179e8d98f22a7e513a8135bdTim Haley tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY0(space_map_open(&msp->ms_sm, mos, new_object,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (msp->ms_loaded && spa_sync_pass(spa) == 1 &&
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson space_map_write(msp->ms_sm, alloctree, SM_ALLOC, tx);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson space_map_write(msp->ms_sm, *freetree, SM_FREE, tx);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * When the space map is loaded, we have an accruate
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * histogram in the range tree. This gives us an opportunity
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * to bring the space map's histogram up-to-date so we clear
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * it first before updating it.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson space_map_histogram_add(msp->ms_sm, msp->ms_tree, tx);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Since the space map is not loaded we simply update the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * exisiting histogram with what was freed in this txg. This
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * means that the on-disk histogram may not have an accurate
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * view of the free space but it's close enough to allow
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * us to make allocation decisions.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson space_map_histogram_add(msp->ms_sm, *freetree, tx);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * For sync pass 1, we avoid traversing this txg's free range tree
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * and instead will just swap the pointers for freetree and
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * freed_tree. We can safely do this since the freed_tree is
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson * guaranteed to be empty on the initial pass.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_vacate(*freetree, range_tree_add, *freed_tree);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson ASSERT0(range_tree_space(msp->ms_alloctree[txg & TXG_MASK]));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson ASSERT0(range_tree_space(msp->ms_freetree[txg & TXG_MASK]));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson dmu_write(mos, vd->vdev_ms_array, sizeof (uint64_t) *
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Called after a transaction group has completely synced to mark
fa9e4066f08beec538e775443c5be79dd423fcabahrens * all of the metaslab's free space as usable.
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * If this metaslab is just becoming available, initialize its
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * alloctrees, freetrees, and defertree and add its capacity to
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (msp->ms_freetree[TXG_CLEAN(txg) & TXG_MASK] == NULL) {
468c413a79615e77179e8d98f22a7e513a8135bdTim Haley for (int t = 0; t < TXG_SIZE; t++) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson msp->ms_alloctree[t] = range_tree_create(NULL, msp,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson msp->ms_freetree[t] = range_tree_create(NULL, msp,
16a4a8074274d2d7cc408589cf6359f4a378c861George Wilson for (int t = 0; t < TXG_DEFER_SIZE; t++) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson msp->ms_defertree[t] = range_tree_create(NULL, msp,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson freed_tree = &msp->ms_freetree[TXG_CLEAN(txg) & TXG_MASK];
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson defer_tree = &msp->ms_defertree[txg % TXG_DEFER_SIZE];
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson alloc_delta = space_map_alloc_delta(msp->ms_sm);
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick vdev_space_update(vd, alloc_delta + defer_delta, defer_delta, 0);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson ASSERT0(range_tree_space(msp->ms_alloctree[txg & TXG_MASK]));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson ASSERT0(range_tree_space(msp->ms_freetree[txg & TXG_MASK]));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * If there's a metaslab_load() in progress, wait for it to complete
ecc2d604e885a75cc75e647b5641af99d5a6f4a6bonwick * so that we have a consistent view of the in-core space map.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Move the frees from the defer_tree back to the free
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * range tree (if it's loaded). Swap the freed_tree and the
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * defer_tree -- this is safe to do because we've just emptied out
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * the defer_tree.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson msp->ms_loaded ? range_tree_add : NULL, msp->ms_tree);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson ASSERT3S(msp->ms_deferspace, <=, msp->ms_size);
468c413a79615e77179e8d98f22a7e513a8135bdTim Haley * Keep syncing this metaslab until all deferred frees
468c413a79615e77179e8d98f22a7e513a8135bdTim Haley * are back in circulation.
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (msp->ms_loaded && msp->ms_access_txg < txg) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson for (int t = 1; t < TXG_CONCURRENT_STATES; t++) {
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson * Preload the next potential metaslabs
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm uint64_t ms_shift = msp->ms_group->mg_vd->vdev_ms_shift;
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm if (msp->ms_group->mg_vd->vdev_id != DVA_GET_VDEV(dva))
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm return (0);
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilsonmetaslab_group_alloc(metaslab_group_t *mg, uint64_t psize, uint64_t asize,
b6240e830b871f59c22a3918aebb3b36c872edbaGeorge Wilson uint64_t txg, uint64_t min_distance, dva_t *dva, int d)
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson for (i = 0; i < d; i++) {
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson if (DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) {
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm for (msp = avl_first(t); msp; msp = AVL_NEXT(t, msp)) {
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson "requirement: vdev %llu, txg %llu, mg %p, "
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson "msp %p, psize %llu, asize %llu, "
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm return (-1ULL);
03f8c366886542ed249a15d755ae78ea4e775d9dGeorge Wilson * If the selected metaslab is condensing, skip it.
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson was_active = msp->ms_weight & METASLAB_ACTIVE_MASK;
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm for (i = 0; i < d; i++)
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm if (i == d)
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm return (-1ULL);
aeb1c1b609b02f03e8e7448beb88384ebc713525gw * Ensure that the metaslab we have selected is still
aeb1c1b609b02f03e8e7448beb88384ebc713525gw * capable of handling our request. It's possible that
aeb1c1b609b02f03e8e7448beb88384ebc713525gw * another thread may have changed the weight while we
aeb1c1b609b02f03e8e7448beb88384ebc713525gw * were blocked on the metaslab lock.
d6e555bdd793b8bc8fe57d5f12c3d69c813d0661George Wilson activation_weight == METASLAB_WEIGHT_PRIMARY)) {
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson if (metaslab_activate(msp, activation_weight) != 0) {
03f8c366886542ed249a15d755ae78ea4e775d9dGeorge Wilson * If this metaslab is currently condensing then pick again as
03f8c366886542ed249a15d755ae78ea4e775d9dGeorge Wilson * we can't manipulate this metaslab until it's committed
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if ((offset = metaslab_block_alloc(msp, asize)) != -1ULL)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson metaslab_passivate(msp, metaslab_block_maxsize(msp));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (range_tree_space(msp->ms_alloctree[txg & TXG_MASK]) == 0)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_add(msp->ms_alloctree[txg & TXG_MASK], offset, asize);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson msp->ms_access_txg = txg + metaslab_unload_delay;
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Allocate a block for the specified i/o.
8654d0253136055bd4cc2423d87378e8a37f2eb5perrinmetaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick dva_t *dva, int d, dva_t *hintdva, uint64_t txg, int flags)
e05725b117836db173257fae43fb0746eb857fb5bonwick * For testing, make some blocks above a certain size be gang blocks.
d3d50737e566cade9a08d73d2af95105ac7cd960Rafael Vanoni if (psize >= metaslab_gang_bang && (ddi_get_lbolt() & 3) == 0)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Start at the rotor and loop through all mgs until we find something.
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick * Note that there's no locking on mc_rotor or mc_aliquot because
fa9e4066f08beec538e775443c5be79dd423fcabahrens * nothing actually breaks if we miss a few updates -- we just won't
fa9e4066f08beec538e775443c5be79dd423fcabahrens * allocate quite as evenly. It all balances out over time.
67bd71c6cc629bab3aa0d595c624a667f1574254perrin * If we are doing ditto or log blocks, try to spread them across
67bd71c6cc629bab3aa0d595c624a667f1574254perrin * consecutive vdevs. If we're forced to reuse a vdev before we've
67bd71c6cc629bab3aa0d595c624a667f1574254perrin * allocated all of our ditto blocks, then try and spread them out on
67bd71c6cc629bab3aa0d595c624a667f1574254perrin * that vdev as much as possible. If it turns out to not be possible,
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * gradually lower our standards until anything becomes acceptable.
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * Also, allocating on consecutive vdevs (as opposed to random vdevs)
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * gives us hope of containing our fault domains to something we're
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * able to reason about. Otherwise, any two top-level vdev failures
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * will guarantee the loss of data. With consecutive allocation,
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * only two adjacent top-level vdev failures will result in data loss.
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * If we are doing gang blocks (hintdva is non-NULL), try to keep
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * ourselves on the same vdev as our gang block header. That
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * way, we can hope for locality in vdev_cache, plus it makes our
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm * fault domains something tractable.
88ecc943b4eb72f7c4fbbd8435997b85ef171fc3George Wilson * It's possible the vdev we're using as the hint no
88ecc943b4eb72f7c4fbbd8435997b85ef171fc3George Wilson * longer exists (i.e. removed). Consult the rotor when
88ecc943b4eb72f7c4fbbd8435997b85ef171fc3George Wilson * all else fails.
44cd46cadd9aab751dae6a4023c1cb5bf316d274billm } else if (d != 0) {
a15215608b8bd90f714f6db21ee623b584607cb6Jeff Bonwick * If the hint put us into the wrong metaslab class, or into a
a15215608b8bd90f714f6db21ee623b584607cb6Jeff Bonwick * metaslab group that has been passivated, just follow the rotor.
a15215608b8bd90f714f6db21ee623b584607cb6Jeff Bonwick if (mg->mg_class != mc || mg->mg_activation_count <= 0)
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick * Don't allocate from faulted devices.
8ad4d6dd86f5bc65fb3afa566c8133f3bac21648Jeff Bonwick spa_config_enter(spa, SCL_ZIO, FTAG, RW_READER);
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * Determine if the selected metaslab group is eligible
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * for allocations. If we're ganging or have requested
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * an allocation for the smallest gang block size
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * then we don't want to avoid allocating to the this
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * metaslab group. If we're in this condition we should
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * try to allocate from any device possible so that we
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * don't inadvertently return ENOSPC and suspend the pool
22e30981d82a0b6dc89253596ededafae8655e00George Wilson * even though space is still available.
0a4e9518a44f226be6d39383330b5b1792d2f184gw * Avoid writing single-copy data to a failing vdev
9dc3941c735ef88de46e850f745aa556d3a071a5Sašo Kiselkov * unless the user instructs us that it is okay.
fa9e4066f08beec538e775443c5be79dd423fcabahrens ASSERT(P2PHASE(asize, 1ULL << vd->vdev_ashift) == 0);
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson offset = metaslab_group_alloc(mg, psize, asize, txg, distance,
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If we've just selected this metaslab group,
fa9e4066f08beec538e775443c5be79dd423fcabahrens * figure out whether the corresponding vdev is
fa9e4066f08beec538e775443c5be79dd423fcabahrens * over- or under-used relative to the pool,
fa9e4066f08beec538e775443c5be79dd423fcabahrens * and set an allocation bias to even it out.
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson vu = (vs->vs_alloc * 100) / (vs->vs_space + 1);
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson cu = (mc->mc_alloc * 100) / (mc->mc_space + 1);
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * Calculate how much more or less we should
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * try to allocate from this device during
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * this iteration around the rotor.
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * For example, if a device is 80% full
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * and the pool is 20% full then we should
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * reduce allocations by 60% on this device.
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * mg_bias = (20 - 80) * 512K / 100 = -307K
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson * This reduces allocations by 307K for this
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick if (atomic_add_64_nv(&mc->mc_aliquot, asize) >=
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick DVA_SET_GANG(&dva[d], !!(flags & METASLAB_GANG_HEADER));
fa9e4066f08beec538e775443c5be79dd423fcabahrens return (0);
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Free the block represented by DVA in the context of the specified
fa9e4066f08beec538e775443c5be79dd423fcabahrens * transaction group.
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwickmetaslab_free_dva(spa_t *spa, const dva_t *dva, uint64_t txg, boolean_t now)
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick (offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count) {
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick cmn_err(CE_WARN, "metaslab_free_dva(): bad DVA %llu:%llu",
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_remove(msp->ms_alloctree[txg & TXG_MASK],
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY3U(offset + size, <=, msp->ms_start + msp->ms_size);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY3U(range_tree_space(msp->ms_tree) + size, <=,
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (range_tree_space(msp->ms_freetree[txg & TXG_MASK]) == 0)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_add(msp->ms_freetree[txg & TXG_MASK],
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick * Intent log support: upon opening the pool after a crash, notify the SPA
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick * of blocks that the intent log has allocated for immediate write, but
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick * which are still considered free by the SPA because the last transaction
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick * group didn't commit yet.
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwickmetaslab_claim_dva(spa_t *spa, const dva_t *dva, uint64_t txg)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if ((txg != 0 && spa_writeable(spa)) || !msp->ms_loaded)
09c9d376e8ccb8fbba74f33cc268964464092b62George Wilson error = metaslab_activate(msp, METASLAB_WEIGHT_SECONDARY);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (error == 0 && !range_tree_contains(msp->ms_tree, offset, size))
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick if (error || txg == 0) { /* txg == 0 indicates dry run */
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson VERIFY3U(range_tree_space(msp->ms_tree) - size, <=, msp->ms_size);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_remove(msp->ms_tree, offset, size);
8ad4d6dd86f5bc65fb3afa566c8133f3bac21648Jeff Bonwick if (spa_writeable(spa)) { /* don't dirty if we're zdb(1M) */
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson if (range_tree_space(msp->ms_alloctree[txg & TXG_MASK]) == 0)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_add(msp->ms_alloctree[txg & TXG_MASK], offset, size);
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick return (0);
8654d0253136055bd4cc2423d87378e8a37f2eb5perrinmetaslab_alloc(spa_t *spa, metaslab_class_t *mc, uint64_t psize, blkptr_t *bp,
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick int ndvas, uint64_t txg, blkptr_t *hintbp, int flags)
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick if (mc->mc_rotor == NULL) { /* no vdevs in this class */
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick ASSERT(ndvas > 0 && ndvas <= spa_max_replication(spa));
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick ASSERT(hintbp == NULL || ndvas <= BP_GET_NDVAS(hintbp));
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick for (int d = 0; d < ndvas; d++) {
8654d0253136055bd4cc2423d87378e8a37f2eb5perrin error = metaslab_alloc_dva(spa, mc, psize, dva, d, hintdva,
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick for (d--; d >= 0; d--) {
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick return (0);
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwickmetaslab_free(spa_t *spa, const blkptr_t *bp, uint64_t txg, boolean_t now)
b24ab6762772a3f6a89393947930c7fa61306783Jeff Bonwick ASSERT(!now || bp->blk_birth >= spa_syncing_txg(spa));
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick spa_config_enter(spa, SCL_FREE, FTAG, RW_READER);
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick for (int d = 0; d < ndvas; d++)
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwickmetaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg)
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick * First do a dry run to make sure all DVAs are claimable,
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick * so we don't have to unwind from partial failures below.
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
e14bb3258d05c1b1077e2db7cf77088924e56919Jeff Bonwick for (int d = 0; d < ndvas; d++)
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwick if ((error = metaslab_claim_dva(spa, &dva[d], txg)) != 0)
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrensmetaslab_check_free(spa_t *spa, const blkptr_t *bp)
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[i]);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]);
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens uint64_t size = DVA_GET_ASIZE(&bp->blk_dva[i]);
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_verify(msp->ms_tree, offset, size);
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens for (int j = 0; j < TXG_SIZE; j++)
0713e232b7712cd27d99e1e935ebb8d5de61c57dGeorge Wilson range_tree_verify(msp->ms_freetree[j], offset, size);
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens for (int j = 0; j < TXG_DEFER_SIZE; j++)