12e405ad8SJosef Bacik // SPDX-License-Identifier: GPL-2.0 22e405ad8SJosef Bacik 352bb7a21SBoris Burkov #include <linux/sizes.h> 42ca0ec77SJohannes Thumshirn #include <linux/list_sort.h> 5784352feSDavid Sterba #include "misc.h" 62e405ad8SJosef Bacik #include "ctree.h" 72e405ad8SJosef Bacik #include "block-group.h" 83eeb3226SJosef Bacik #include "space-info.h" 99f21246dSJosef Bacik #include "disk-io.h" 109f21246dSJosef Bacik #include "free-space-cache.h" 119f21246dSJosef Bacik #include "free-space-tree.h" 12e3e0520bSJosef Bacik #include "volumes.h" 13e3e0520bSJosef Bacik #include "transaction.h" 14e3e0520bSJosef Bacik #include "ref-verify.h" 154358d963SJosef Bacik #include "sysfs.h" 164358d963SJosef Bacik #include "tree-log.h" 1777745c05SJosef Bacik #include "delalloc-space.h" 18b0643e59SDennis Zhou #include "discard.h" 1996a14336SNikolay Borisov #include "raid56.h" 2008e11a3dSNaohiro Aota #include "zoned.h" 21c7f13d42SJosef Bacik #include "fs.h" 2207e81dc9SJosef Bacik #include "accessors.h" 23a0231804SJosef Bacik #include "extent-tree.h" 242e405ad8SJosef Bacik 2506d61cb1SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2606d61cb1SJosef Bacik int btrfs_should_fragment_free_space(struct btrfs_block_group *block_group) 2706d61cb1SJosef Bacik { 2806d61cb1SJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 2906d61cb1SJosef Bacik 3006d61cb1SJosef Bacik return (btrfs_test_opt(fs_info, FRAGMENT_METADATA) && 3106d61cb1SJosef Bacik block_group->flags & BTRFS_BLOCK_GROUP_METADATA) || 3206d61cb1SJosef Bacik (btrfs_test_opt(fs_info, FRAGMENT_DATA) && 3306d61cb1SJosef Bacik block_group->flags & BTRFS_BLOCK_GROUP_DATA); 3406d61cb1SJosef Bacik } 3506d61cb1SJosef Bacik #endif 3606d61cb1SJosef Bacik 37878d7b67SJosef Bacik /* 38878d7b67SJosef Bacik * Return target flags in extended format or 0 if restripe for this chunk_type 39878d7b67SJosef Bacik * is not in progress 40878d7b67SJosef Bacik * 41878d7b67SJosef Bacik * Should be called with balance_lock held 42878d7b67SJosef Bacik */ 43e11c0406SJosef Bacik static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) 44878d7b67SJosef Bacik { 45878d7b67SJosef Bacik struct btrfs_balance_control *bctl = fs_info->balance_ctl; 46878d7b67SJosef Bacik u64 target = 0; 47878d7b67SJosef Bacik 48878d7b67SJosef Bacik if (!bctl) 49878d7b67SJosef Bacik return 0; 50878d7b67SJosef Bacik 51878d7b67SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA && 52878d7b67SJosef Bacik bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { 53878d7b67SJosef Bacik target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; 54878d7b67SJosef Bacik } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && 55878d7b67SJosef Bacik bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { 56878d7b67SJosef Bacik target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; 57878d7b67SJosef Bacik } else if (flags & BTRFS_BLOCK_GROUP_METADATA && 58878d7b67SJosef Bacik bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { 59878d7b67SJosef Bacik target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; 60878d7b67SJosef Bacik } 61878d7b67SJosef Bacik 62878d7b67SJosef Bacik return target; 63878d7b67SJosef Bacik } 64878d7b67SJosef Bacik 65878d7b67SJosef Bacik /* 66878d7b67SJosef Bacik * @flags: available profiles in extended format (see ctree.h) 67878d7b67SJosef Bacik * 68878d7b67SJosef Bacik * Return reduced profile in chunk format. If profile changing is in progress 69878d7b67SJosef Bacik * (either running or paused) picks the target profile (if it's already 70878d7b67SJosef Bacik * available), otherwise falls back to plain reducing. 71878d7b67SJosef Bacik */ 72878d7b67SJosef Bacik static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) 73878d7b67SJosef Bacik { 74878d7b67SJosef Bacik u64 num_devices = fs_info->fs_devices->rw_devices; 75878d7b67SJosef Bacik u64 target; 76878d7b67SJosef Bacik u64 raid_type; 77878d7b67SJosef Bacik u64 allowed = 0; 78878d7b67SJosef Bacik 79878d7b67SJosef Bacik /* 80878d7b67SJosef Bacik * See if restripe for this chunk_type is in progress, if so try to 81878d7b67SJosef Bacik * reduce to the target profile 82878d7b67SJosef Bacik */ 83878d7b67SJosef Bacik spin_lock(&fs_info->balance_lock); 84e11c0406SJosef Bacik target = get_restripe_target(fs_info, flags); 85878d7b67SJosef Bacik if (target) { 86878d7b67SJosef Bacik spin_unlock(&fs_info->balance_lock); 87878d7b67SJosef Bacik return extended_to_chunk(target); 88878d7b67SJosef Bacik } 89878d7b67SJosef Bacik spin_unlock(&fs_info->balance_lock); 90878d7b67SJosef Bacik 91878d7b67SJosef Bacik /* First, mask out the RAID levels which aren't possible */ 92878d7b67SJosef Bacik for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 93878d7b67SJosef Bacik if (num_devices >= btrfs_raid_array[raid_type].devs_min) 94878d7b67SJosef Bacik allowed |= btrfs_raid_array[raid_type].bg_flag; 95878d7b67SJosef Bacik } 96878d7b67SJosef Bacik allowed &= flags; 97878d7b67SJosef Bacik 98878d7b67SJosef Bacik if (allowed & BTRFS_BLOCK_GROUP_RAID6) 99878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID6; 100878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID5) 101878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID5; 102878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID10) 103878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID10; 104878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID1) 105878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID1; 106878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID0) 107878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID0; 108878d7b67SJosef Bacik 109878d7b67SJosef Bacik flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; 110878d7b67SJosef Bacik 111878d7b67SJosef Bacik return extended_to_chunk(flags | allowed); 112878d7b67SJosef Bacik } 113878d7b67SJosef Bacik 114ef0a82daSJohannes Thumshirn u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) 115878d7b67SJosef Bacik { 116878d7b67SJosef Bacik unsigned seq; 117878d7b67SJosef Bacik u64 flags; 118878d7b67SJosef Bacik 119878d7b67SJosef Bacik do { 120878d7b67SJosef Bacik flags = orig_flags; 121878d7b67SJosef Bacik seq = read_seqbegin(&fs_info->profiles_lock); 122878d7b67SJosef Bacik 123878d7b67SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 124878d7b67SJosef Bacik flags |= fs_info->avail_data_alloc_bits; 125878d7b67SJosef Bacik else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 126878d7b67SJosef Bacik flags |= fs_info->avail_system_alloc_bits; 127878d7b67SJosef Bacik else if (flags & BTRFS_BLOCK_GROUP_METADATA) 128878d7b67SJosef Bacik flags |= fs_info->avail_metadata_alloc_bits; 129878d7b67SJosef Bacik } while (read_seqretry(&fs_info->profiles_lock, seq)); 130878d7b67SJosef Bacik 131878d7b67SJosef Bacik return btrfs_reduce_alloc_profile(fs_info, flags); 132878d7b67SJosef Bacik } 133878d7b67SJosef Bacik 13432da5386SDavid Sterba void btrfs_get_block_group(struct btrfs_block_group *cache) 1353cad1284SJosef Bacik { 13648aaeebeSJosef Bacik refcount_inc(&cache->refs); 1373cad1284SJosef Bacik } 1383cad1284SJosef Bacik 13932da5386SDavid Sterba void btrfs_put_block_group(struct btrfs_block_group *cache) 1403cad1284SJosef Bacik { 14148aaeebeSJosef Bacik if (refcount_dec_and_test(&cache->refs)) { 1423cad1284SJosef Bacik WARN_ON(cache->pinned > 0); 14340cdc509SFilipe Manana /* 14440cdc509SFilipe Manana * If there was a failure to cleanup a log tree, very likely due 14540cdc509SFilipe Manana * to an IO failure on a writeback attempt of one or more of its 14640cdc509SFilipe Manana * extent buffers, we could not do proper (and cheap) unaccounting 14740cdc509SFilipe Manana * of their reserved space, so don't warn on reserved > 0 in that 14840cdc509SFilipe Manana * case. 14940cdc509SFilipe Manana */ 15040cdc509SFilipe Manana if (!(cache->flags & BTRFS_BLOCK_GROUP_METADATA) || 15140cdc509SFilipe Manana !BTRFS_FS_LOG_CLEANUP_ERROR(cache->fs_info)) 1523cad1284SJosef Bacik WARN_ON(cache->reserved > 0); 1533cad1284SJosef Bacik 1543cad1284SJosef Bacik /* 155b0643e59SDennis Zhou * A block_group shouldn't be on the discard_list anymore. 156b0643e59SDennis Zhou * Remove the block_group from the discard_list to prevent us 157b0643e59SDennis Zhou * from causing a panic due to NULL pointer dereference. 158b0643e59SDennis Zhou */ 159b0643e59SDennis Zhou if (WARN_ON(!list_empty(&cache->discard_list))) 160b0643e59SDennis Zhou btrfs_discard_cancel_work(&cache->fs_info->discard_ctl, 161b0643e59SDennis Zhou cache); 162b0643e59SDennis Zhou 163b0643e59SDennis Zhou /* 1643cad1284SJosef Bacik * If not empty, someone is still holding mutex of 1653cad1284SJosef Bacik * full_stripe_lock, which can only be released by caller. 1663cad1284SJosef Bacik * And it will definitely cause use-after-free when caller 1673cad1284SJosef Bacik * tries to release full stripe lock. 1683cad1284SJosef Bacik * 1693cad1284SJosef Bacik * No better way to resolve, but only to warn. 1703cad1284SJosef Bacik */ 1713cad1284SJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); 1723cad1284SJosef Bacik kfree(cache->free_space_ctl); 173dafc340dSNaohiro Aota kfree(cache->physical_map); 1743cad1284SJosef Bacik kfree(cache); 1753cad1284SJosef Bacik } 1763cad1284SJosef Bacik } 1773cad1284SJosef Bacik 1782e405ad8SJosef Bacik /* 1794358d963SJosef Bacik * This adds the block group to the fs_info rb tree for the block group cache 1804358d963SJosef Bacik */ 1814358d963SJosef Bacik static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, 18232da5386SDavid Sterba struct btrfs_block_group *block_group) 1834358d963SJosef Bacik { 1844358d963SJosef Bacik struct rb_node **p; 1854358d963SJosef Bacik struct rb_node *parent = NULL; 18632da5386SDavid Sterba struct btrfs_block_group *cache; 18708dddb29SFilipe Manana bool leftmost = true; 1884358d963SJosef Bacik 1899afc6649SQu Wenruo ASSERT(block_group->length != 0); 1909afc6649SQu Wenruo 19116b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 19208dddb29SFilipe Manana p = &info->block_group_cache_tree.rb_root.rb_node; 1934358d963SJosef Bacik 1944358d963SJosef Bacik while (*p) { 1954358d963SJosef Bacik parent = *p; 19632da5386SDavid Sterba cache = rb_entry(parent, struct btrfs_block_group, cache_node); 197b3470b5dSDavid Sterba if (block_group->start < cache->start) { 1984358d963SJosef Bacik p = &(*p)->rb_left; 199b3470b5dSDavid Sterba } else if (block_group->start > cache->start) { 2004358d963SJosef Bacik p = &(*p)->rb_right; 20108dddb29SFilipe Manana leftmost = false; 2024358d963SJosef Bacik } else { 20316b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 2044358d963SJosef Bacik return -EEXIST; 2054358d963SJosef Bacik } 2064358d963SJosef Bacik } 2074358d963SJosef Bacik 2084358d963SJosef Bacik rb_link_node(&block_group->cache_node, parent, p); 20908dddb29SFilipe Manana rb_insert_color_cached(&block_group->cache_node, 21008dddb29SFilipe Manana &info->block_group_cache_tree, leftmost); 2114358d963SJosef Bacik 21216b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 2134358d963SJosef Bacik 2144358d963SJosef Bacik return 0; 2154358d963SJosef Bacik } 2164358d963SJosef Bacik 2174358d963SJosef Bacik /* 2182e405ad8SJosef Bacik * This will return the block group at or after bytenr if contains is 0, else 2192e405ad8SJosef Bacik * it will return the block group that contains the bytenr 2202e405ad8SJosef Bacik */ 22132da5386SDavid Sterba static struct btrfs_block_group *block_group_cache_tree_search( 2222e405ad8SJosef Bacik struct btrfs_fs_info *info, u64 bytenr, int contains) 2232e405ad8SJosef Bacik { 22432da5386SDavid Sterba struct btrfs_block_group *cache, *ret = NULL; 2252e405ad8SJosef Bacik struct rb_node *n; 2262e405ad8SJosef Bacik u64 end, start; 2272e405ad8SJosef Bacik 22816b0c258SFilipe Manana read_lock(&info->block_group_cache_lock); 22908dddb29SFilipe Manana n = info->block_group_cache_tree.rb_root.rb_node; 2302e405ad8SJosef Bacik 2312e405ad8SJosef Bacik while (n) { 23232da5386SDavid Sterba cache = rb_entry(n, struct btrfs_block_group, cache_node); 233b3470b5dSDavid Sterba end = cache->start + cache->length - 1; 234b3470b5dSDavid Sterba start = cache->start; 2352e405ad8SJosef Bacik 2362e405ad8SJosef Bacik if (bytenr < start) { 237b3470b5dSDavid Sterba if (!contains && (!ret || start < ret->start)) 2382e405ad8SJosef Bacik ret = cache; 2392e405ad8SJosef Bacik n = n->rb_left; 2402e405ad8SJosef Bacik } else if (bytenr > start) { 2412e405ad8SJosef Bacik if (contains && bytenr <= end) { 2422e405ad8SJosef Bacik ret = cache; 2432e405ad8SJosef Bacik break; 2442e405ad8SJosef Bacik } 2452e405ad8SJosef Bacik n = n->rb_right; 2462e405ad8SJosef Bacik } else { 2472e405ad8SJosef Bacik ret = cache; 2482e405ad8SJosef Bacik break; 2492e405ad8SJosef Bacik } 2502e405ad8SJosef Bacik } 25108dddb29SFilipe Manana if (ret) 2522e405ad8SJosef Bacik btrfs_get_block_group(ret); 25316b0c258SFilipe Manana read_unlock(&info->block_group_cache_lock); 2542e405ad8SJosef Bacik 2552e405ad8SJosef Bacik return ret; 2562e405ad8SJosef Bacik } 2572e405ad8SJosef Bacik 2582e405ad8SJosef Bacik /* 2592e405ad8SJosef Bacik * Return the block group that starts at or after bytenr 2602e405ad8SJosef Bacik */ 26132da5386SDavid Sterba struct btrfs_block_group *btrfs_lookup_first_block_group( 2622e405ad8SJosef Bacik struct btrfs_fs_info *info, u64 bytenr) 2632e405ad8SJosef Bacik { 2642e405ad8SJosef Bacik return block_group_cache_tree_search(info, bytenr, 0); 2652e405ad8SJosef Bacik } 2662e405ad8SJosef Bacik 2672e405ad8SJosef Bacik /* 2682e405ad8SJosef Bacik * Return the block group that contains the given bytenr 2692e405ad8SJosef Bacik */ 27032da5386SDavid Sterba struct btrfs_block_group *btrfs_lookup_block_group( 2712e405ad8SJosef Bacik struct btrfs_fs_info *info, u64 bytenr) 2722e405ad8SJosef Bacik { 2732e405ad8SJosef Bacik return block_group_cache_tree_search(info, bytenr, 1); 2742e405ad8SJosef Bacik } 2752e405ad8SJosef Bacik 27632da5386SDavid Sterba struct btrfs_block_group *btrfs_next_block_group( 27732da5386SDavid Sterba struct btrfs_block_group *cache) 2782e405ad8SJosef Bacik { 2792e405ad8SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 2802e405ad8SJosef Bacik struct rb_node *node; 2812e405ad8SJosef Bacik 28216b0c258SFilipe Manana read_lock(&fs_info->block_group_cache_lock); 2832e405ad8SJosef Bacik 2842e405ad8SJosef Bacik /* If our block group was removed, we need a full search. */ 2852e405ad8SJosef Bacik if (RB_EMPTY_NODE(&cache->cache_node)) { 286b3470b5dSDavid Sterba const u64 next_bytenr = cache->start + cache->length; 2872e405ad8SJosef Bacik 28816b0c258SFilipe Manana read_unlock(&fs_info->block_group_cache_lock); 2892e405ad8SJosef Bacik btrfs_put_block_group(cache); 2908b01f931SFilipe Manana return btrfs_lookup_first_block_group(fs_info, next_bytenr); 2912e405ad8SJosef Bacik } 2922e405ad8SJosef Bacik node = rb_next(&cache->cache_node); 2932e405ad8SJosef Bacik btrfs_put_block_group(cache); 2942e405ad8SJosef Bacik if (node) { 29532da5386SDavid Sterba cache = rb_entry(node, struct btrfs_block_group, cache_node); 2962e405ad8SJosef Bacik btrfs_get_block_group(cache); 2972e405ad8SJosef Bacik } else 2982e405ad8SJosef Bacik cache = NULL; 29916b0c258SFilipe Manana read_unlock(&fs_info->block_group_cache_lock); 3002e405ad8SJosef Bacik return cache; 3012e405ad8SJosef Bacik } 3023eeb3226SJosef Bacik 30343dd529aSDavid Sterba /* 3042306e83eSFilipe Manana * Check if we can do a NOCOW write for a given extent. 3052306e83eSFilipe Manana * 3062306e83eSFilipe Manana * @fs_info: The filesystem information object. 3072306e83eSFilipe Manana * @bytenr: Logical start address of the extent. 3082306e83eSFilipe Manana * 3092306e83eSFilipe Manana * Check if we can do a NOCOW write for the given extent, and increments the 3102306e83eSFilipe Manana * number of NOCOW writers in the block group that contains the extent, as long 3112306e83eSFilipe Manana * as the block group exists and it's currently not in read-only mode. 3122306e83eSFilipe Manana * 3132306e83eSFilipe Manana * Returns: A non-NULL block group pointer if we can do a NOCOW write, the caller 3142306e83eSFilipe Manana * is responsible for calling btrfs_dec_nocow_writers() later. 3152306e83eSFilipe Manana * 3162306e83eSFilipe Manana * Or NULL if we can not do a NOCOW write 3172306e83eSFilipe Manana */ 3182306e83eSFilipe Manana struct btrfs_block_group *btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, 3192306e83eSFilipe Manana u64 bytenr) 3203eeb3226SJosef Bacik { 32132da5386SDavid Sterba struct btrfs_block_group *bg; 3222306e83eSFilipe Manana bool can_nocow = true; 3233eeb3226SJosef Bacik 3243eeb3226SJosef Bacik bg = btrfs_lookup_block_group(fs_info, bytenr); 3253eeb3226SJosef Bacik if (!bg) 3262306e83eSFilipe Manana return NULL; 3273eeb3226SJosef Bacik 3283eeb3226SJosef Bacik spin_lock(&bg->lock); 3293eeb3226SJosef Bacik if (bg->ro) 3302306e83eSFilipe Manana can_nocow = false; 3313eeb3226SJosef Bacik else 3323eeb3226SJosef Bacik atomic_inc(&bg->nocow_writers); 3333eeb3226SJosef Bacik spin_unlock(&bg->lock); 3343eeb3226SJosef Bacik 3352306e83eSFilipe Manana if (!can_nocow) { 3363eeb3226SJosef Bacik btrfs_put_block_group(bg); 3372306e83eSFilipe Manana return NULL; 3383eeb3226SJosef Bacik } 3393eeb3226SJosef Bacik 3402306e83eSFilipe Manana /* No put on block group, done by btrfs_dec_nocow_writers(). */ 3412306e83eSFilipe Manana return bg; 3422306e83eSFilipe Manana } 3433eeb3226SJosef Bacik 34443dd529aSDavid Sterba /* 3452306e83eSFilipe Manana * Decrement the number of NOCOW writers in a block group. 3462306e83eSFilipe Manana * 3472306e83eSFilipe Manana * This is meant to be called after a previous call to btrfs_inc_nocow_writers(), 3482306e83eSFilipe Manana * and on the block group returned by that call. Typically this is called after 3492306e83eSFilipe Manana * creating an ordered extent for a NOCOW write, to prevent races with scrub and 3502306e83eSFilipe Manana * relocation. 3512306e83eSFilipe Manana * 3522306e83eSFilipe Manana * After this call, the caller should not use the block group anymore. It it wants 3532306e83eSFilipe Manana * to use it, then it should get a reference on it before calling this function. 3542306e83eSFilipe Manana */ 3552306e83eSFilipe Manana void btrfs_dec_nocow_writers(struct btrfs_block_group *bg) 3562306e83eSFilipe Manana { 3573eeb3226SJosef Bacik if (atomic_dec_and_test(&bg->nocow_writers)) 3583eeb3226SJosef Bacik wake_up_var(&bg->nocow_writers); 3592306e83eSFilipe Manana 3602306e83eSFilipe Manana /* For the lookup done by a previous call to btrfs_inc_nocow_writers(). */ 3613eeb3226SJosef Bacik btrfs_put_block_group(bg); 3623eeb3226SJosef Bacik } 3633eeb3226SJosef Bacik 36432da5386SDavid Sterba void btrfs_wait_nocow_writers(struct btrfs_block_group *bg) 3653eeb3226SJosef Bacik { 3663eeb3226SJosef Bacik wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); 3673eeb3226SJosef Bacik } 3683eeb3226SJosef Bacik 3693eeb3226SJosef Bacik void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, 3703eeb3226SJosef Bacik const u64 start) 3713eeb3226SJosef Bacik { 37232da5386SDavid Sterba struct btrfs_block_group *bg; 3733eeb3226SJosef Bacik 3743eeb3226SJosef Bacik bg = btrfs_lookup_block_group(fs_info, start); 3753eeb3226SJosef Bacik ASSERT(bg); 3763eeb3226SJosef Bacik if (atomic_dec_and_test(&bg->reservations)) 3773eeb3226SJosef Bacik wake_up_var(&bg->reservations); 3783eeb3226SJosef Bacik btrfs_put_block_group(bg); 3793eeb3226SJosef Bacik } 3803eeb3226SJosef Bacik 38132da5386SDavid Sterba void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg) 3823eeb3226SJosef Bacik { 3833eeb3226SJosef Bacik struct btrfs_space_info *space_info = bg->space_info; 3843eeb3226SJosef Bacik 3853eeb3226SJosef Bacik ASSERT(bg->ro); 3863eeb3226SJosef Bacik 3873eeb3226SJosef Bacik if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) 3883eeb3226SJosef Bacik return; 3893eeb3226SJosef Bacik 3903eeb3226SJosef Bacik /* 3913eeb3226SJosef Bacik * Our block group is read only but before we set it to read only, 3923eeb3226SJosef Bacik * some task might have had allocated an extent from it already, but it 3933eeb3226SJosef Bacik * has not yet created a respective ordered extent (and added it to a 3943eeb3226SJosef Bacik * root's list of ordered extents). 3953eeb3226SJosef Bacik * Therefore wait for any task currently allocating extents, since the 3963eeb3226SJosef Bacik * block group's reservations counter is incremented while a read lock 3973eeb3226SJosef Bacik * on the groups' semaphore is held and decremented after releasing 3983eeb3226SJosef Bacik * the read access on that semaphore and creating the ordered extent. 3993eeb3226SJosef Bacik */ 4003eeb3226SJosef Bacik down_write(&space_info->groups_sem); 4013eeb3226SJosef Bacik up_write(&space_info->groups_sem); 4023eeb3226SJosef Bacik 4033eeb3226SJosef Bacik wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); 4043eeb3226SJosef Bacik } 4059f21246dSJosef Bacik 4069f21246dSJosef Bacik struct btrfs_caching_control *btrfs_get_caching_control( 40732da5386SDavid Sterba struct btrfs_block_group *cache) 4089f21246dSJosef Bacik { 4099f21246dSJosef Bacik struct btrfs_caching_control *ctl; 4109f21246dSJosef Bacik 4119f21246dSJosef Bacik spin_lock(&cache->lock); 4129f21246dSJosef Bacik if (!cache->caching_ctl) { 4139f21246dSJosef Bacik spin_unlock(&cache->lock); 4149f21246dSJosef Bacik return NULL; 4159f21246dSJosef Bacik } 4169f21246dSJosef Bacik 4179f21246dSJosef Bacik ctl = cache->caching_ctl; 4189f21246dSJosef Bacik refcount_inc(&ctl->count); 4199f21246dSJosef Bacik spin_unlock(&cache->lock); 4209f21246dSJosef Bacik return ctl; 4219f21246dSJosef Bacik } 4229f21246dSJosef Bacik 4239f21246dSJosef Bacik void btrfs_put_caching_control(struct btrfs_caching_control *ctl) 4249f21246dSJosef Bacik { 4259f21246dSJosef Bacik if (refcount_dec_and_test(&ctl->count)) 4269f21246dSJosef Bacik kfree(ctl); 4279f21246dSJosef Bacik } 4289f21246dSJosef Bacik 4299f21246dSJosef Bacik /* 4309f21246dSJosef Bacik * When we wait for progress in the block group caching, its because our 4319f21246dSJosef Bacik * allocation attempt failed at least once. So, we must sleep and let some 4329f21246dSJosef Bacik * progress happen before we try again. 4339f21246dSJosef Bacik * 4349f21246dSJosef Bacik * This function will sleep at least once waiting for new free space to show 4359f21246dSJosef Bacik * up, and then it will check the block group free space numbers for our min 4369f21246dSJosef Bacik * num_bytes. Another option is to have it go ahead and look in the rbtree for 4379f21246dSJosef Bacik * a free extent of a given size, but this is a good start. 4389f21246dSJosef Bacik * 4399f21246dSJosef Bacik * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using 4409f21246dSJosef Bacik * any of the information in this block group. 4419f21246dSJosef Bacik */ 44232da5386SDavid Sterba void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache, 4439f21246dSJosef Bacik u64 num_bytes) 4449f21246dSJosef Bacik { 4459f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 4469f21246dSJosef Bacik 4479f21246dSJosef Bacik caching_ctl = btrfs_get_caching_control(cache); 4489f21246dSJosef Bacik if (!caching_ctl) 4499f21246dSJosef Bacik return; 4509f21246dSJosef Bacik 45132da5386SDavid Sterba wait_event(caching_ctl->wait, btrfs_block_group_done(cache) || 4529f21246dSJosef Bacik (cache->free_space_ctl->free_space >= num_bytes)); 4539f21246dSJosef Bacik 4549f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 4559f21246dSJosef Bacik } 4569f21246dSJosef Bacik 457ced8ecf0SOmar Sandoval static int btrfs_caching_ctl_wait_done(struct btrfs_block_group *cache, 458ced8ecf0SOmar Sandoval struct btrfs_caching_control *caching_ctl) 459ced8ecf0SOmar Sandoval { 460ced8ecf0SOmar Sandoval wait_event(caching_ctl->wait, btrfs_block_group_done(cache)); 461ced8ecf0SOmar Sandoval return cache->cached == BTRFS_CACHE_ERROR ? -EIO : 0; 462ced8ecf0SOmar Sandoval } 463ced8ecf0SOmar Sandoval 464ced8ecf0SOmar Sandoval static int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache) 4659f21246dSJosef Bacik { 4669f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 467ced8ecf0SOmar Sandoval int ret; 4689f21246dSJosef Bacik 4699f21246dSJosef Bacik caching_ctl = btrfs_get_caching_control(cache); 4709f21246dSJosef Bacik if (!caching_ctl) 4719f21246dSJosef Bacik return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; 472ced8ecf0SOmar Sandoval ret = btrfs_caching_ctl_wait_done(cache, caching_ctl); 4739f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 4749f21246dSJosef Bacik return ret; 4759f21246dSJosef Bacik } 4769f21246dSJosef Bacik 4779f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 47832da5386SDavid Sterba static void fragment_free_space(struct btrfs_block_group *block_group) 4799f21246dSJosef Bacik { 4809f21246dSJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 481b3470b5dSDavid Sterba u64 start = block_group->start; 482b3470b5dSDavid Sterba u64 len = block_group->length; 4839f21246dSJosef Bacik u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? 4849f21246dSJosef Bacik fs_info->nodesize : fs_info->sectorsize; 4859f21246dSJosef Bacik u64 step = chunk << 1; 4869f21246dSJosef Bacik 4879f21246dSJosef Bacik while (len > chunk) { 4889f21246dSJosef Bacik btrfs_remove_free_space(block_group, start, chunk); 4899f21246dSJosef Bacik start += step; 4909f21246dSJosef Bacik if (len < step) 4919f21246dSJosef Bacik len = 0; 4929f21246dSJosef Bacik else 4939f21246dSJosef Bacik len -= step; 4949f21246dSJosef Bacik } 4959f21246dSJosef Bacik } 4969f21246dSJosef Bacik #endif 4979f21246dSJosef Bacik 4989f21246dSJosef Bacik /* 4999f21246dSJosef Bacik * This is only called by btrfs_cache_block_group, since we could have freed 5009f21246dSJosef Bacik * extents we need to check the pinned_extents for any extents that can't be 5019f21246dSJosef Bacik * used yet since their free space will be released as soon as the transaction 5029f21246dSJosef Bacik * commits. 5039f21246dSJosef Bacik */ 50432da5386SDavid Sterba u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end) 5059f21246dSJosef Bacik { 5069f21246dSJosef Bacik struct btrfs_fs_info *info = block_group->fs_info; 5079f21246dSJosef Bacik u64 extent_start, extent_end, size, total_added = 0; 5089f21246dSJosef Bacik int ret; 5099f21246dSJosef Bacik 5109f21246dSJosef Bacik while (start < end) { 511fe119a6eSNikolay Borisov ret = find_first_extent_bit(&info->excluded_extents, start, 5129f21246dSJosef Bacik &extent_start, &extent_end, 5139f21246dSJosef Bacik EXTENT_DIRTY | EXTENT_UPTODATE, 5149f21246dSJosef Bacik NULL); 5159f21246dSJosef Bacik if (ret) 5169f21246dSJosef Bacik break; 5179f21246dSJosef Bacik 5189f21246dSJosef Bacik if (extent_start <= start) { 5199f21246dSJosef Bacik start = extent_end + 1; 5209f21246dSJosef Bacik } else if (extent_start > start && extent_start < end) { 5219f21246dSJosef Bacik size = extent_start - start; 5229f21246dSJosef Bacik total_added += size; 523b0643e59SDennis Zhou ret = btrfs_add_free_space_async_trimmed(block_group, 524b0643e59SDennis Zhou start, size); 5259f21246dSJosef Bacik BUG_ON(ret); /* -ENOMEM or logic error */ 5269f21246dSJosef Bacik start = extent_end + 1; 5279f21246dSJosef Bacik } else { 5289f21246dSJosef Bacik break; 5299f21246dSJosef Bacik } 5309f21246dSJosef Bacik } 5319f21246dSJosef Bacik 5329f21246dSJosef Bacik if (start < end) { 5339f21246dSJosef Bacik size = end - start; 5349f21246dSJosef Bacik total_added += size; 535b0643e59SDennis Zhou ret = btrfs_add_free_space_async_trimmed(block_group, start, 536b0643e59SDennis Zhou size); 5379f21246dSJosef Bacik BUG_ON(ret); /* -ENOMEM or logic error */ 5389f21246dSJosef Bacik } 5399f21246dSJosef Bacik 5409f21246dSJosef Bacik return total_added; 5419f21246dSJosef Bacik } 5429f21246dSJosef Bacik 543c7eec3d9SBoris Burkov /* 544c7eec3d9SBoris Burkov * Get an arbitrary extent item index / max_index through the block group 545c7eec3d9SBoris Burkov * 546c7eec3d9SBoris Burkov * @block_group the block group to sample from 547c7eec3d9SBoris Burkov * @index: the integral step through the block group to grab from 548c7eec3d9SBoris Burkov * @max_index: the granularity of the sampling 549c7eec3d9SBoris Burkov * @key: return value parameter for the item we find 550c7eec3d9SBoris Burkov * 551c7eec3d9SBoris Burkov * Pre-conditions on indices: 552c7eec3d9SBoris Burkov * 0 <= index <= max_index 553c7eec3d9SBoris Burkov * 0 < max_index 554c7eec3d9SBoris Burkov * 555c7eec3d9SBoris Burkov * Returns: 0 on success, 1 if the search didn't yield a useful item, negative 556c7eec3d9SBoris Burkov * error code on error. 557c7eec3d9SBoris Burkov */ 558c7eec3d9SBoris Burkov static int sample_block_group_extent_item(struct btrfs_caching_control *caching_ctl, 559c7eec3d9SBoris Burkov struct btrfs_block_group *block_group, 560c7eec3d9SBoris Burkov int index, int max_index, 56112148367SBoris Burkov struct btrfs_key *found_key) 562c7eec3d9SBoris Burkov { 563c7eec3d9SBoris Burkov struct btrfs_fs_info *fs_info = block_group->fs_info; 564c7eec3d9SBoris Burkov struct btrfs_root *extent_root; 565c7eec3d9SBoris Burkov u64 search_offset; 566c7eec3d9SBoris Burkov u64 search_end = block_group->start + block_group->length; 567c7eec3d9SBoris Burkov struct btrfs_path *path; 56812148367SBoris Burkov struct btrfs_key search_key; 56912148367SBoris Burkov int ret = 0; 570c7eec3d9SBoris Burkov 571c7eec3d9SBoris Burkov ASSERT(index >= 0); 572c7eec3d9SBoris Burkov ASSERT(index <= max_index); 573c7eec3d9SBoris Burkov ASSERT(max_index > 0); 574c7eec3d9SBoris Burkov lockdep_assert_held(&caching_ctl->mutex); 575c7eec3d9SBoris Burkov lockdep_assert_held_read(&fs_info->commit_root_sem); 576c7eec3d9SBoris Burkov 577c7eec3d9SBoris Burkov path = btrfs_alloc_path(); 578c7eec3d9SBoris Burkov if (!path) 579c7eec3d9SBoris Burkov return -ENOMEM; 580c7eec3d9SBoris Burkov 581c7eec3d9SBoris Burkov extent_root = btrfs_extent_root(fs_info, max_t(u64, block_group->start, 582c7eec3d9SBoris Burkov BTRFS_SUPER_INFO_OFFSET)); 583c7eec3d9SBoris Burkov 584c7eec3d9SBoris Burkov path->skip_locking = 1; 585c7eec3d9SBoris Burkov path->search_commit_root = 1; 586c7eec3d9SBoris Burkov path->reada = READA_FORWARD; 587c7eec3d9SBoris Burkov 588c7eec3d9SBoris Burkov search_offset = index * div_u64(block_group->length, max_index); 58912148367SBoris Burkov search_key.objectid = block_group->start + search_offset; 59012148367SBoris Burkov search_key.type = BTRFS_EXTENT_ITEM_KEY; 59112148367SBoris Burkov search_key.offset = 0; 592c7eec3d9SBoris Burkov 59312148367SBoris Burkov btrfs_for_each_slot(extent_root, &search_key, found_key, path, ret) { 594c7eec3d9SBoris Burkov /* Success; sampled an extent item in the block group */ 59512148367SBoris Burkov if (found_key->type == BTRFS_EXTENT_ITEM_KEY && 59612148367SBoris Burkov found_key->objectid >= block_group->start && 59712148367SBoris Burkov found_key->objectid + found_key->offset <= search_end) 59812148367SBoris Burkov break; 599c7eec3d9SBoris Burkov 600c7eec3d9SBoris Burkov /* We can't possibly find a valid extent item anymore */ 60112148367SBoris Burkov if (found_key->objectid >= search_end) { 602c7eec3d9SBoris Burkov ret = 1; 603c7eec3d9SBoris Burkov break; 604c7eec3d9SBoris Burkov } 605c7eec3d9SBoris Burkov } 60612148367SBoris Burkov 607c7eec3d9SBoris Burkov lockdep_assert_held(&caching_ctl->mutex); 608c7eec3d9SBoris Burkov lockdep_assert_held_read(&fs_info->commit_root_sem); 609c7eec3d9SBoris Burkov btrfs_free_path(path); 610c7eec3d9SBoris Burkov return ret; 611c7eec3d9SBoris Burkov } 612c7eec3d9SBoris Burkov 613c7eec3d9SBoris Burkov /* 614c7eec3d9SBoris Burkov * Best effort attempt to compute a block group's size class while caching it. 615c7eec3d9SBoris Burkov * 616c7eec3d9SBoris Burkov * @block_group: the block group we are caching 617c7eec3d9SBoris Burkov * 618c7eec3d9SBoris Burkov * We cannot infer the size class while adding free space extents, because that 619c7eec3d9SBoris Burkov * logic doesn't care about contiguous file extents (it doesn't differentiate 620c7eec3d9SBoris Burkov * between a 100M extent and 100 contiguous 1M extents). So we need to read the 621c7eec3d9SBoris Burkov * file extent items. Reading all of them is quite wasteful, because usually 622c7eec3d9SBoris Burkov * only a handful are enough to give a good answer. Therefore, we just grab 5 of 623c7eec3d9SBoris Burkov * them at even steps through the block group and pick the smallest size class 624c7eec3d9SBoris Burkov * we see. Since size class is best effort, and not guaranteed in general, 625c7eec3d9SBoris Burkov * inaccuracy is acceptable. 626c7eec3d9SBoris Burkov * 627c7eec3d9SBoris Burkov * To be more explicit about why this algorithm makes sense: 628c7eec3d9SBoris Burkov * 629c7eec3d9SBoris Burkov * If we are caching in a block group from disk, then there are three major cases 630c7eec3d9SBoris Burkov * to consider: 631c7eec3d9SBoris Burkov * 1. the block group is well behaved and all extents in it are the same size 632c7eec3d9SBoris Burkov * class. 633c7eec3d9SBoris Burkov * 2. the block group is mostly one size class with rare exceptions for last 634c7eec3d9SBoris Burkov * ditch allocations 635c7eec3d9SBoris Burkov * 3. the block group was populated before size classes and can have a totally 636c7eec3d9SBoris Burkov * arbitrary mix of size classes. 637c7eec3d9SBoris Burkov * 638c7eec3d9SBoris Burkov * In case 1, looking at any extent in the block group will yield the correct 639c7eec3d9SBoris Burkov * result. For the mixed cases, taking the minimum size class seems like a good 640c7eec3d9SBoris Burkov * approximation, since gaps from frees will be usable to the size class. For 641c7eec3d9SBoris Burkov * 2., a small handful of file extents is likely to yield the right answer. For 642c7eec3d9SBoris Burkov * 3, we can either read every file extent, or admit that this is best effort 643c7eec3d9SBoris Burkov * anyway and try to stay fast. 644c7eec3d9SBoris Burkov * 645c7eec3d9SBoris Burkov * Returns: 0 on success, negative error code on error. 646c7eec3d9SBoris Burkov */ 647c7eec3d9SBoris Burkov static int load_block_group_size_class(struct btrfs_caching_control *caching_ctl, 648c7eec3d9SBoris Burkov struct btrfs_block_group *block_group) 649c7eec3d9SBoris Burkov { 65012148367SBoris Burkov struct btrfs_fs_info *fs_info = block_group->fs_info; 651c7eec3d9SBoris Burkov struct btrfs_key key; 652c7eec3d9SBoris Burkov int i; 653c7eec3d9SBoris Burkov u64 min_size = block_group->length; 654c7eec3d9SBoris Burkov enum btrfs_block_group_size_class size_class = BTRFS_BG_SZ_NONE; 655c7eec3d9SBoris Burkov int ret; 656c7eec3d9SBoris Burkov 657cb0922f2SBoris Burkov if (!btrfs_block_group_should_use_size_class(block_group)) 658c7eec3d9SBoris Burkov return 0; 659c7eec3d9SBoris Burkov 66012148367SBoris Burkov lockdep_assert_held(&caching_ctl->mutex); 66112148367SBoris Burkov lockdep_assert_held_read(&fs_info->commit_root_sem); 662c7eec3d9SBoris Burkov for (i = 0; i < 5; ++i) { 663c7eec3d9SBoris Burkov ret = sample_block_group_extent_item(caching_ctl, block_group, i, 5, &key); 664c7eec3d9SBoris Burkov if (ret < 0) 665c7eec3d9SBoris Burkov goto out; 666c7eec3d9SBoris Burkov if (ret > 0) 667c7eec3d9SBoris Burkov continue; 668c7eec3d9SBoris Burkov min_size = min_t(u64, min_size, key.offset); 669c7eec3d9SBoris Burkov size_class = btrfs_calc_block_group_size_class(min_size); 670c7eec3d9SBoris Burkov } 671c7eec3d9SBoris Burkov if (size_class != BTRFS_BG_SZ_NONE) { 672c7eec3d9SBoris Burkov spin_lock(&block_group->lock); 673c7eec3d9SBoris Burkov block_group->size_class = size_class; 674c7eec3d9SBoris Burkov spin_unlock(&block_group->lock); 675c7eec3d9SBoris Burkov } 676c7eec3d9SBoris Burkov out: 677c7eec3d9SBoris Burkov return ret; 678c7eec3d9SBoris Burkov } 679c7eec3d9SBoris Burkov 6809f21246dSJosef Bacik static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) 6819f21246dSJosef Bacik { 68232da5386SDavid Sterba struct btrfs_block_group *block_group = caching_ctl->block_group; 6839f21246dSJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 68429cbcf40SJosef Bacik struct btrfs_root *extent_root; 6859f21246dSJosef Bacik struct btrfs_path *path; 6869f21246dSJosef Bacik struct extent_buffer *leaf; 6879f21246dSJosef Bacik struct btrfs_key key; 6889f21246dSJosef Bacik u64 total_found = 0; 6899f21246dSJosef Bacik u64 last = 0; 6909f21246dSJosef Bacik u32 nritems; 6919f21246dSJosef Bacik int ret; 6929f21246dSJosef Bacik bool wakeup = true; 6939f21246dSJosef Bacik 6949f21246dSJosef Bacik path = btrfs_alloc_path(); 6959f21246dSJosef Bacik if (!path) 6969f21246dSJosef Bacik return -ENOMEM; 6979f21246dSJosef Bacik 698b3470b5dSDavid Sterba last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET); 69929cbcf40SJosef Bacik extent_root = btrfs_extent_root(fs_info, last); 7009f21246dSJosef Bacik 7019f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 7029f21246dSJosef Bacik /* 7039f21246dSJosef Bacik * If we're fragmenting we don't want to make anybody think we can 7049f21246dSJosef Bacik * allocate from this block group until we've had a chance to fragment 7059f21246dSJosef Bacik * the free space. 7069f21246dSJosef Bacik */ 7079f21246dSJosef Bacik if (btrfs_should_fragment_free_space(block_group)) 7089f21246dSJosef Bacik wakeup = false; 7099f21246dSJosef Bacik #endif 7109f21246dSJosef Bacik /* 7119f21246dSJosef Bacik * We don't want to deadlock with somebody trying to allocate a new 7129f21246dSJosef Bacik * extent for the extent root while also trying to search the extent 7139f21246dSJosef Bacik * root to add free space. So we skip locking and search the commit 7149f21246dSJosef Bacik * root, since its read-only 7159f21246dSJosef Bacik */ 7169f21246dSJosef Bacik path->skip_locking = 1; 7179f21246dSJosef Bacik path->search_commit_root = 1; 7189f21246dSJosef Bacik path->reada = READA_FORWARD; 7199f21246dSJosef Bacik 7209f21246dSJosef Bacik key.objectid = last; 7219f21246dSJosef Bacik key.offset = 0; 7229f21246dSJosef Bacik key.type = BTRFS_EXTENT_ITEM_KEY; 7239f21246dSJosef Bacik 7249f21246dSJosef Bacik next: 7259f21246dSJosef Bacik ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); 7269f21246dSJosef Bacik if (ret < 0) 7279f21246dSJosef Bacik goto out; 7289f21246dSJosef Bacik 7299f21246dSJosef Bacik leaf = path->nodes[0]; 7309f21246dSJosef Bacik nritems = btrfs_header_nritems(leaf); 7319f21246dSJosef Bacik 7329f21246dSJosef Bacik while (1) { 7339f21246dSJosef Bacik if (btrfs_fs_closing(fs_info) > 1) { 7349f21246dSJosef Bacik last = (u64)-1; 7359f21246dSJosef Bacik break; 7369f21246dSJosef Bacik } 7379f21246dSJosef Bacik 7389f21246dSJosef Bacik if (path->slots[0] < nritems) { 7399f21246dSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 7409f21246dSJosef Bacik } else { 7419f21246dSJosef Bacik ret = btrfs_find_next_key(extent_root, path, &key, 0, 0); 7429f21246dSJosef Bacik if (ret) 7439f21246dSJosef Bacik break; 7449f21246dSJosef Bacik 7459f21246dSJosef Bacik if (need_resched() || 7469f21246dSJosef Bacik rwsem_is_contended(&fs_info->commit_root_sem)) { 7479f21246dSJosef Bacik btrfs_release_path(path); 7489f21246dSJosef Bacik up_read(&fs_info->commit_root_sem); 7499f21246dSJosef Bacik mutex_unlock(&caching_ctl->mutex); 7509f21246dSJosef Bacik cond_resched(); 7519f21246dSJosef Bacik mutex_lock(&caching_ctl->mutex); 7529f21246dSJosef Bacik down_read(&fs_info->commit_root_sem); 7539f21246dSJosef Bacik goto next; 7549f21246dSJosef Bacik } 7559f21246dSJosef Bacik 7569f21246dSJosef Bacik ret = btrfs_next_leaf(extent_root, path); 7579f21246dSJosef Bacik if (ret < 0) 7589f21246dSJosef Bacik goto out; 7599f21246dSJosef Bacik if (ret) 7609f21246dSJosef Bacik break; 7619f21246dSJosef Bacik leaf = path->nodes[0]; 7629f21246dSJosef Bacik nritems = btrfs_header_nritems(leaf); 7639f21246dSJosef Bacik continue; 7649f21246dSJosef Bacik } 7659f21246dSJosef Bacik 7669f21246dSJosef Bacik if (key.objectid < last) { 7679f21246dSJosef Bacik key.objectid = last; 7689f21246dSJosef Bacik key.offset = 0; 7699f21246dSJosef Bacik key.type = BTRFS_EXTENT_ITEM_KEY; 7709f21246dSJosef Bacik btrfs_release_path(path); 7719f21246dSJosef Bacik goto next; 7729f21246dSJosef Bacik } 7739f21246dSJosef Bacik 774b3470b5dSDavid Sterba if (key.objectid < block_group->start) { 7759f21246dSJosef Bacik path->slots[0]++; 7769f21246dSJosef Bacik continue; 7779f21246dSJosef Bacik } 7789f21246dSJosef Bacik 779b3470b5dSDavid Sterba if (key.objectid >= block_group->start + block_group->length) 7809f21246dSJosef Bacik break; 7819f21246dSJosef Bacik 7829f21246dSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7839f21246dSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7849f21246dSJosef Bacik total_found += add_new_free_space(block_group, last, 7859f21246dSJosef Bacik key.objectid); 7869f21246dSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY) 7879f21246dSJosef Bacik last = key.objectid + 7889f21246dSJosef Bacik fs_info->nodesize; 7899f21246dSJosef Bacik else 7909f21246dSJosef Bacik last = key.objectid + key.offset; 7919f21246dSJosef Bacik 7929f21246dSJosef Bacik if (total_found > CACHING_CTL_WAKE_UP) { 7939f21246dSJosef Bacik total_found = 0; 7949f21246dSJosef Bacik if (wakeup) 7959f21246dSJosef Bacik wake_up(&caching_ctl->wait); 7969f21246dSJosef Bacik } 7979f21246dSJosef Bacik } 7989f21246dSJosef Bacik path->slots[0]++; 7999f21246dSJosef Bacik } 8009f21246dSJosef Bacik ret = 0; 8019f21246dSJosef Bacik 8029f21246dSJosef Bacik total_found += add_new_free_space(block_group, last, 803b3470b5dSDavid Sterba block_group->start + block_group->length); 8049f21246dSJosef Bacik 8059f21246dSJosef Bacik out: 8069f21246dSJosef Bacik btrfs_free_path(path); 8079f21246dSJosef Bacik return ret; 8089f21246dSJosef Bacik } 8099f21246dSJosef Bacik 8109f21246dSJosef Bacik static noinline void caching_thread(struct btrfs_work *work) 8119f21246dSJosef Bacik { 81232da5386SDavid Sterba struct btrfs_block_group *block_group; 8139f21246dSJosef Bacik struct btrfs_fs_info *fs_info; 8149f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 8159f21246dSJosef Bacik int ret; 8169f21246dSJosef Bacik 8179f21246dSJosef Bacik caching_ctl = container_of(work, struct btrfs_caching_control, work); 8189f21246dSJosef Bacik block_group = caching_ctl->block_group; 8199f21246dSJosef Bacik fs_info = block_group->fs_info; 8209f21246dSJosef Bacik 8219f21246dSJosef Bacik mutex_lock(&caching_ctl->mutex); 8229f21246dSJosef Bacik down_read(&fs_info->commit_root_sem); 8239f21246dSJosef Bacik 824c7eec3d9SBoris Burkov load_block_group_size_class(caching_ctl, block_group); 825e747853cSJosef Bacik if (btrfs_test_opt(fs_info, SPACE_CACHE)) { 826e747853cSJosef Bacik ret = load_free_space_cache(block_group); 827e747853cSJosef Bacik if (ret == 1) { 828e747853cSJosef Bacik ret = 0; 829e747853cSJosef Bacik goto done; 830e747853cSJosef Bacik } 831e747853cSJosef Bacik 832e747853cSJosef Bacik /* 833e747853cSJosef Bacik * We failed to load the space cache, set ourselves to 834e747853cSJosef Bacik * CACHE_STARTED and carry on. 835e747853cSJosef Bacik */ 836e747853cSJosef Bacik spin_lock(&block_group->lock); 837e747853cSJosef Bacik block_group->cached = BTRFS_CACHE_STARTED; 838e747853cSJosef Bacik spin_unlock(&block_group->lock); 839e747853cSJosef Bacik wake_up(&caching_ctl->wait); 840e747853cSJosef Bacik } 841e747853cSJosef Bacik 8422f96e402SJosef Bacik /* 8432f96e402SJosef Bacik * If we are in the transaction that populated the free space tree we 8442f96e402SJosef Bacik * can't actually cache from the free space tree as our commit root and 8452f96e402SJosef Bacik * real root are the same, so we could change the contents of the blocks 8462f96e402SJosef Bacik * while caching. Instead do the slow caching in this case, and after 8472f96e402SJosef Bacik * the transaction has committed we will be safe. 8482f96e402SJosef Bacik */ 8492f96e402SJosef Bacik if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 8502f96e402SJosef Bacik !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags))) 8519f21246dSJosef Bacik ret = load_free_space_tree(caching_ctl); 8529f21246dSJosef Bacik else 8539f21246dSJosef Bacik ret = load_extent_tree_free(caching_ctl); 854e747853cSJosef Bacik done: 8559f21246dSJosef Bacik spin_lock(&block_group->lock); 8569f21246dSJosef Bacik block_group->caching_ctl = NULL; 8579f21246dSJosef Bacik block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; 8589f21246dSJosef Bacik spin_unlock(&block_group->lock); 8599f21246dSJosef Bacik 8609f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 8619f21246dSJosef Bacik if (btrfs_should_fragment_free_space(block_group)) { 8629f21246dSJosef Bacik u64 bytes_used; 8639f21246dSJosef Bacik 8649f21246dSJosef Bacik spin_lock(&block_group->space_info->lock); 8659f21246dSJosef Bacik spin_lock(&block_group->lock); 866b3470b5dSDavid Sterba bytes_used = block_group->length - block_group->used; 8679f21246dSJosef Bacik block_group->space_info->bytes_used += bytes_used >> 1; 8689f21246dSJosef Bacik spin_unlock(&block_group->lock); 8699f21246dSJosef Bacik spin_unlock(&block_group->space_info->lock); 870e11c0406SJosef Bacik fragment_free_space(block_group); 8719f21246dSJosef Bacik } 8729f21246dSJosef Bacik #endif 8739f21246dSJosef Bacik 8749f21246dSJosef Bacik up_read(&fs_info->commit_root_sem); 8759f21246dSJosef Bacik btrfs_free_excluded_extents(block_group); 8769f21246dSJosef Bacik mutex_unlock(&caching_ctl->mutex); 8779f21246dSJosef Bacik 8789f21246dSJosef Bacik wake_up(&caching_ctl->wait); 8799f21246dSJosef Bacik 8809f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 8819f21246dSJosef Bacik btrfs_put_block_group(block_group); 8829f21246dSJosef Bacik } 8839f21246dSJosef Bacik 884ced8ecf0SOmar Sandoval int btrfs_cache_block_group(struct btrfs_block_group *cache, bool wait) 8859f21246dSJosef Bacik { 8869f21246dSJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 887e747853cSJosef Bacik struct btrfs_caching_control *caching_ctl = NULL; 8889f21246dSJosef Bacik int ret = 0; 8899f21246dSJosef Bacik 8902eda5708SNaohiro Aota /* Allocator for zoned filesystems does not use the cache at all */ 8912eda5708SNaohiro Aota if (btrfs_is_zoned(fs_info)) 8922eda5708SNaohiro Aota return 0; 8932eda5708SNaohiro Aota 8949f21246dSJosef Bacik caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); 8959f21246dSJosef Bacik if (!caching_ctl) 8969f21246dSJosef Bacik return -ENOMEM; 8979f21246dSJosef Bacik 8989f21246dSJosef Bacik INIT_LIST_HEAD(&caching_ctl->list); 8999f21246dSJosef Bacik mutex_init(&caching_ctl->mutex); 9009f21246dSJosef Bacik init_waitqueue_head(&caching_ctl->wait); 9019f21246dSJosef Bacik caching_ctl->block_group = cache; 902e747853cSJosef Bacik refcount_set(&caching_ctl->count, 2); 903a0cac0ecSOmar Sandoval btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); 9049f21246dSJosef Bacik 9059f21246dSJosef Bacik spin_lock(&cache->lock); 9069f21246dSJosef Bacik if (cache->cached != BTRFS_CACHE_NO) { 9079f21246dSJosef Bacik kfree(caching_ctl); 908e747853cSJosef Bacik 909e747853cSJosef Bacik caching_ctl = cache->caching_ctl; 910e747853cSJosef Bacik if (caching_ctl) 911e747853cSJosef Bacik refcount_inc(&caching_ctl->count); 912e747853cSJosef Bacik spin_unlock(&cache->lock); 913e747853cSJosef Bacik goto out; 9149f21246dSJosef Bacik } 9159f21246dSJosef Bacik WARN_ON(cache->caching_ctl); 9169f21246dSJosef Bacik cache->caching_ctl = caching_ctl; 9179f21246dSJosef Bacik cache->cached = BTRFS_CACHE_STARTED; 9189f21246dSJosef Bacik spin_unlock(&cache->lock); 9199f21246dSJosef Bacik 92016b0c258SFilipe Manana write_lock(&fs_info->block_group_cache_lock); 9219f21246dSJosef Bacik refcount_inc(&caching_ctl->count); 9229f21246dSJosef Bacik list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); 92316b0c258SFilipe Manana write_unlock(&fs_info->block_group_cache_lock); 9249f21246dSJosef Bacik 9259f21246dSJosef Bacik btrfs_get_block_group(cache); 9269f21246dSJosef Bacik 9279f21246dSJosef Bacik btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); 928e747853cSJosef Bacik out: 929ced8ecf0SOmar Sandoval if (wait && caching_ctl) 930ced8ecf0SOmar Sandoval ret = btrfs_caching_ctl_wait_done(cache, caching_ctl); 931e747853cSJosef Bacik if (caching_ctl) 932e747853cSJosef Bacik btrfs_put_caching_control(caching_ctl); 9339f21246dSJosef Bacik 9349f21246dSJosef Bacik return ret; 9359f21246dSJosef Bacik } 936e3e0520bSJosef Bacik 937e3e0520bSJosef Bacik static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) 938e3e0520bSJosef Bacik { 939e3e0520bSJosef Bacik u64 extra_flags = chunk_to_extended(flags) & 940e3e0520bSJosef Bacik BTRFS_EXTENDED_PROFILE_MASK; 941e3e0520bSJosef Bacik 942e3e0520bSJosef Bacik write_seqlock(&fs_info->profiles_lock); 943e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 944e3e0520bSJosef Bacik fs_info->avail_data_alloc_bits &= ~extra_flags; 945e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_METADATA) 946e3e0520bSJosef Bacik fs_info->avail_metadata_alloc_bits &= ~extra_flags; 947e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 948e3e0520bSJosef Bacik fs_info->avail_system_alloc_bits &= ~extra_flags; 949e3e0520bSJosef Bacik write_sequnlock(&fs_info->profiles_lock); 950e3e0520bSJosef Bacik } 951e3e0520bSJosef Bacik 952e3e0520bSJosef Bacik /* 953e3e0520bSJosef Bacik * Clear incompat bits for the following feature(s): 954e3e0520bSJosef Bacik * 955e3e0520bSJosef Bacik * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group 956e3e0520bSJosef Bacik * in the whole filesystem 9579c907446SDavid Sterba * 9589c907446SDavid Sterba * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups 959e3e0520bSJosef Bacik */ 960e3e0520bSJosef Bacik static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) 961e3e0520bSJosef Bacik { 9629c907446SDavid Sterba bool found_raid56 = false; 9639c907446SDavid Sterba bool found_raid1c34 = false; 9649c907446SDavid Sterba 9659c907446SDavid Sterba if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) || 9669c907446SDavid Sterba (flags & BTRFS_BLOCK_GROUP_RAID1C3) || 9679c907446SDavid Sterba (flags & BTRFS_BLOCK_GROUP_RAID1C4)) { 968e3e0520bSJosef Bacik struct list_head *head = &fs_info->space_info; 969e3e0520bSJosef Bacik struct btrfs_space_info *sinfo; 970e3e0520bSJosef Bacik 971e3e0520bSJosef Bacik list_for_each_entry_rcu(sinfo, head, list) { 972e3e0520bSJosef Bacik down_read(&sinfo->groups_sem); 973e3e0520bSJosef Bacik if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) 9749c907446SDavid Sterba found_raid56 = true; 975e3e0520bSJosef Bacik if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) 9769c907446SDavid Sterba found_raid56 = true; 9779c907446SDavid Sterba if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3])) 9789c907446SDavid Sterba found_raid1c34 = true; 9799c907446SDavid Sterba if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4])) 9809c907446SDavid Sterba found_raid1c34 = true; 981e3e0520bSJosef Bacik up_read(&sinfo->groups_sem); 982e3e0520bSJosef Bacik } 983d8e6fd5cSFilipe Manana if (!found_raid56) 984e3e0520bSJosef Bacik btrfs_clear_fs_incompat(fs_info, RAID56); 985d8e6fd5cSFilipe Manana if (!found_raid1c34) 9869c907446SDavid Sterba btrfs_clear_fs_incompat(fs_info, RAID1C34); 987e3e0520bSJosef Bacik } 988e3e0520bSJosef Bacik } 989e3e0520bSJosef Bacik 9907357623aSQu Wenruo static int remove_block_group_item(struct btrfs_trans_handle *trans, 9917357623aSQu Wenruo struct btrfs_path *path, 9927357623aSQu Wenruo struct btrfs_block_group *block_group) 9937357623aSQu Wenruo { 9947357623aSQu Wenruo struct btrfs_fs_info *fs_info = trans->fs_info; 9957357623aSQu Wenruo struct btrfs_root *root; 9967357623aSQu Wenruo struct btrfs_key key; 9977357623aSQu Wenruo int ret; 9987357623aSQu Wenruo 999dfe8aec4SJosef Bacik root = btrfs_block_group_root(fs_info); 10007357623aSQu Wenruo key.objectid = block_group->start; 10017357623aSQu Wenruo key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 10027357623aSQu Wenruo key.offset = block_group->length; 10037357623aSQu Wenruo 10047357623aSQu Wenruo ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 10057357623aSQu Wenruo if (ret > 0) 10067357623aSQu Wenruo ret = -ENOENT; 10077357623aSQu Wenruo if (ret < 0) 10087357623aSQu Wenruo return ret; 10097357623aSQu Wenruo 10107357623aSQu Wenruo ret = btrfs_del_item(trans, root, path); 10117357623aSQu Wenruo return ret; 10127357623aSQu Wenruo } 10137357623aSQu Wenruo 1014e3e0520bSJosef Bacik int btrfs_remove_block_group(struct btrfs_trans_handle *trans, 1015e3e0520bSJosef Bacik u64 group_start, struct extent_map *em) 1016e3e0520bSJosef Bacik { 1017e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 1018e3e0520bSJosef Bacik struct btrfs_path *path; 101932da5386SDavid Sterba struct btrfs_block_group *block_group; 1020e3e0520bSJosef Bacik struct btrfs_free_cluster *cluster; 1021e3e0520bSJosef Bacik struct inode *inode; 1022e3e0520bSJosef Bacik struct kobject *kobj = NULL; 1023e3e0520bSJosef Bacik int ret; 1024e3e0520bSJosef Bacik int index; 1025e3e0520bSJosef Bacik int factor; 1026e3e0520bSJosef Bacik struct btrfs_caching_control *caching_ctl = NULL; 1027e3e0520bSJosef Bacik bool remove_em; 1028e3e0520bSJosef Bacik bool remove_rsv = false; 1029e3e0520bSJosef Bacik 1030e3e0520bSJosef Bacik block_group = btrfs_lookup_block_group(fs_info, group_start); 1031e3e0520bSJosef Bacik BUG_ON(!block_group); 1032e3e0520bSJosef Bacik BUG_ON(!block_group->ro); 1033e3e0520bSJosef Bacik 1034e3e0520bSJosef Bacik trace_btrfs_remove_block_group(block_group); 1035e3e0520bSJosef Bacik /* 1036e3e0520bSJosef Bacik * Free the reserved super bytes from this block group before 1037e3e0520bSJosef Bacik * remove it. 1038e3e0520bSJosef Bacik */ 1039e3e0520bSJosef Bacik btrfs_free_excluded_extents(block_group); 1040b3470b5dSDavid Sterba btrfs_free_ref_tree_range(fs_info, block_group->start, 1041b3470b5dSDavid Sterba block_group->length); 1042e3e0520bSJosef Bacik 1043e3e0520bSJosef Bacik index = btrfs_bg_flags_to_raid_index(block_group->flags); 1044e3e0520bSJosef Bacik factor = btrfs_bg_type_to_factor(block_group->flags); 1045e3e0520bSJosef Bacik 1046e3e0520bSJosef Bacik /* make sure this block group isn't part of an allocation cluster */ 1047e3e0520bSJosef Bacik cluster = &fs_info->data_alloc_cluster; 1048e3e0520bSJosef Bacik spin_lock(&cluster->refill_lock); 1049e3e0520bSJosef Bacik btrfs_return_cluster_to_free_space(block_group, cluster); 1050e3e0520bSJosef Bacik spin_unlock(&cluster->refill_lock); 1051e3e0520bSJosef Bacik 1052e3e0520bSJosef Bacik /* 1053e3e0520bSJosef Bacik * make sure this block group isn't part of a metadata 1054e3e0520bSJosef Bacik * allocation cluster 1055e3e0520bSJosef Bacik */ 1056e3e0520bSJosef Bacik cluster = &fs_info->meta_alloc_cluster; 1057e3e0520bSJosef Bacik spin_lock(&cluster->refill_lock); 1058e3e0520bSJosef Bacik btrfs_return_cluster_to_free_space(block_group, cluster); 1059e3e0520bSJosef Bacik spin_unlock(&cluster->refill_lock); 1060e3e0520bSJosef Bacik 106140ab3be1SNaohiro Aota btrfs_clear_treelog_bg(block_group); 1062c2707a25SJohannes Thumshirn btrfs_clear_data_reloc_bg(block_group); 106340ab3be1SNaohiro Aota 1064e3e0520bSJosef Bacik path = btrfs_alloc_path(); 1065e3e0520bSJosef Bacik if (!path) { 1066e3e0520bSJosef Bacik ret = -ENOMEM; 10679fecd132SFilipe Manana goto out; 1068e3e0520bSJosef Bacik } 1069e3e0520bSJosef Bacik 1070e3e0520bSJosef Bacik /* 1071e3e0520bSJosef Bacik * get the inode first so any iput calls done for the io_list 1072e3e0520bSJosef Bacik * aren't the final iput (no unlinks allowed now) 1073e3e0520bSJosef Bacik */ 1074e3e0520bSJosef Bacik inode = lookup_free_space_inode(block_group, path); 1075e3e0520bSJosef Bacik 1076e3e0520bSJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 1077e3e0520bSJosef Bacik /* 1078e3e0520bSJosef Bacik * Make sure our free space cache IO is done before removing the 1079e3e0520bSJosef Bacik * free space inode 1080e3e0520bSJosef Bacik */ 1081e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1082e3e0520bSJosef Bacik if (!list_empty(&block_group->io_list)) { 1083e3e0520bSJosef Bacik list_del_init(&block_group->io_list); 1084e3e0520bSJosef Bacik 1085e3e0520bSJosef Bacik WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); 1086e3e0520bSJosef Bacik 1087e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1088e3e0520bSJosef Bacik btrfs_wait_cache_io(trans, block_group, path); 1089e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1090e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1091e3e0520bSJosef Bacik } 1092e3e0520bSJosef Bacik 1093e3e0520bSJosef Bacik if (!list_empty(&block_group->dirty_list)) { 1094e3e0520bSJosef Bacik list_del_init(&block_group->dirty_list); 1095e3e0520bSJosef Bacik remove_rsv = true; 1096e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1097e3e0520bSJosef Bacik } 1098e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1099e3e0520bSJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 1100e3e0520bSJosef Bacik 110136b216c8SBoris Burkov ret = btrfs_remove_free_space_inode(trans, inode, block_group); 1102e3e0520bSJosef Bacik if (ret) 11039fecd132SFilipe Manana goto out; 1104e3e0520bSJosef Bacik 110516b0c258SFilipe Manana write_lock(&fs_info->block_group_cache_lock); 110608dddb29SFilipe Manana rb_erase_cached(&block_group->cache_node, 1107e3e0520bSJosef Bacik &fs_info->block_group_cache_tree); 1108e3e0520bSJosef Bacik RB_CLEAR_NODE(&block_group->cache_node); 1109e3e0520bSJosef Bacik 11109fecd132SFilipe Manana /* Once for the block groups rbtree */ 11119fecd132SFilipe Manana btrfs_put_block_group(block_group); 11129fecd132SFilipe Manana 111316b0c258SFilipe Manana write_unlock(&fs_info->block_group_cache_lock); 1114e3e0520bSJosef Bacik 1115e3e0520bSJosef Bacik down_write(&block_group->space_info->groups_sem); 1116e3e0520bSJosef Bacik /* 1117e3e0520bSJosef Bacik * we must use list_del_init so people can check to see if they 1118e3e0520bSJosef Bacik * are still on the list after taking the semaphore 1119e3e0520bSJosef Bacik */ 1120e3e0520bSJosef Bacik list_del_init(&block_group->list); 1121e3e0520bSJosef Bacik if (list_empty(&block_group->space_info->block_groups[index])) { 1122e3e0520bSJosef Bacik kobj = block_group->space_info->block_group_kobjs[index]; 1123e3e0520bSJosef Bacik block_group->space_info->block_group_kobjs[index] = NULL; 1124e3e0520bSJosef Bacik clear_avail_alloc_bits(fs_info, block_group->flags); 1125e3e0520bSJosef Bacik } 1126e3e0520bSJosef Bacik up_write(&block_group->space_info->groups_sem); 1127e3e0520bSJosef Bacik clear_incompat_bg_bits(fs_info, block_group->flags); 1128e3e0520bSJosef Bacik if (kobj) { 1129e3e0520bSJosef Bacik kobject_del(kobj); 1130e3e0520bSJosef Bacik kobject_put(kobj); 1131e3e0520bSJosef Bacik } 1132e3e0520bSJosef Bacik 1133e3e0520bSJosef Bacik if (block_group->cached == BTRFS_CACHE_STARTED) 1134e3e0520bSJosef Bacik btrfs_wait_block_group_cache_done(block_group); 11357b9c293bSJosef Bacik 113616b0c258SFilipe Manana write_lock(&fs_info->block_group_cache_lock); 11377b9c293bSJosef Bacik caching_ctl = btrfs_get_caching_control(block_group); 1138e3e0520bSJosef Bacik if (!caching_ctl) { 1139e3e0520bSJosef Bacik struct btrfs_caching_control *ctl; 1140e3e0520bSJosef Bacik 11417b9c293bSJosef Bacik list_for_each_entry(ctl, &fs_info->caching_block_groups, list) { 1142e3e0520bSJosef Bacik if (ctl->block_group == block_group) { 1143e3e0520bSJosef Bacik caching_ctl = ctl; 1144e3e0520bSJosef Bacik refcount_inc(&caching_ctl->count); 1145e3e0520bSJosef Bacik break; 1146e3e0520bSJosef Bacik } 1147e3e0520bSJosef Bacik } 11487b9c293bSJosef Bacik } 1149e3e0520bSJosef Bacik if (caching_ctl) 1150e3e0520bSJosef Bacik list_del_init(&caching_ctl->list); 115116b0c258SFilipe Manana write_unlock(&fs_info->block_group_cache_lock); 11527b9c293bSJosef Bacik 1153e3e0520bSJosef Bacik if (caching_ctl) { 1154e3e0520bSJosef Bacik /* Once for the caching bgs list and once for us. */ 1155e3e0520bSJosef Bacik btrfs_put_caching_control(caching_ctl); 1156e3e0520bSJosef Bacik btrfs_put_caching_control(caching_ctl); 1157e3e0520bSJosef Bacik } 1158e3e0520bSJosef Bacik 1159e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1160e3e0520bSJosef Bacik WARN_ON(!list_empty(&block_group->dirty_list)); 1161e3e0520bSJosef Bacik WARN_ON(!list_empty(&block_group->io_list)); 1162e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1163e3e0520bSJosef Bacik 1164e3e0520bSJosef Bacik btrfs_remove_free_space_cache(block_group); 1165e3e0520bSJosef Bacik 1166e3e0520bSJosef Bacik spin_lock(&block_group->space_info->lock); 1167e3e0520bSJosef Bacik list_del_init(&block_group->ro_list); 1168e3e0520bSJosef Bacik 1169e3e0520bSJosef Bacik if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { 1170e3e0520bSJosef Bacik WARN_ON(block_group->space_info->total_bytes 1171b3470b5dSDavid Sterba < block_group->length); 1172e3e0520bSJosef Bacik WARN_ON(block_group->space_info->bytes_readonly 1173169e0da9SNaohiro Aota < block_group->length - block_group->zone_unusable); 1174169e0da9SNaohiro Aota WARN_ON(block_group->space_info->bytes_zone_unusable 1175169e0da9SNaohiro Aota < block_group->zone_unusable); 1176e3e0520bSJosef Bacik WARN_ON(block_group->space_info->disk_total 1177b3470b5dSDavid Sterba < block_group->length * factor); 1178e3e0520bSJosef Bacik } 1179b3470b5dSDavid Sterba block_group->space_info->total_bytes -= block_group->length; 1180169e0da9SNaohiro Aota block_group->space_info->bytes_readonly -= 1181169e0da9SNaohiro Aota (block_group->length - block_group->zone_unusable); 1182169e0da9SNaohiro Aota block_group->space_info->bytes_zone_unusable -= 1183169e0da9SNaohiro Aota block_group->zone_unusable; 1184b3470b5dSDavid Sterba block_group->space_info->disk_total -= block_group->length * factor; 1185e3e0520bSJosef Bacik 1186e3e0520bSJosef Bacik spin_unlock(&block_group->space_info->lock); 1187e3e0520bSJosef Bacik 1188ffcb9d44SFilipe Manana /* 1189ffcb9d44SFilipe Manana * Remove the free space for the block group from the free space tree 1190ffcb9d44SFilipe Manana * and the block group's item from the extent tree before marking the 1191ffcb9d44SFilipe Manana * block group as removed. This is to prevent races with tasks that 1192ffcb9d44SFilipe Manana * freeze and unfreeze a block group, this task and another task 1193ffcb9d44SFilipe Manana * allocating a new block group - the unfreeze task ends up removing 1194ffcb9d44SFilipe Manana * the block group's extent map before the task calling this function 1195ffcb9d44SFilipe Manana * deletes the block group item from the extent tree, allowing for 1196ffcb9d44SFilipe Manana * another task to attempt to create another block group with the same 1197ffcb9d44SFilipe Manana * item key (and failing with -EEXIST and a transaction abort). 1198ffcb9d44SFilipe Manana */ 1199ffcb9d44SFilipe Manana ret = remove_block_group_free_space(trans, block_group); 1200ffcb9d44SFilipe Manana if (ret) 1201ffcb9d44SFilipe Manana goto out; 1202ffcb9d44SFilipe Manana 1203ffcb9d44SFilipe Manana ret = remove_block_group_item(trans, path, block_group); 1204ffcb9d44SFilipe Manana if (ret < 0) 1205ffcb9d44SFilipe Manana goto out; 1206ffcb9d44SFilipe Manana 1207e3e0520bSJosef Bacik spin_lock(&block_group->lock); 12083349b57fSJosef Bacik set_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags); 12093349b57fSJosef Bacik 1210e3e0520bSJosef Bacik /* 12116b7304afSFilipe Manana * At this point trimming or scrub can't start on this block group, 12126b7304afSFilipe Manana * because we removed the block group from the rbtree 12136b7304afSFilipe Manana * fs_info->block_group_cache_tree so no one can't find it anymore and 12146b7304afSFilipe Manana * even if someone already got this block group before we removed it 12156b7304afSFilipe Manana * from the rbtree, they have already incremented block_group->frozen - 12166b7304afSFilipe Manana * if they didn't, for the trimming case they won't find any free space 12176b7304afSFilipe Manana * entries because we already removed them all when we called 12186b7304afSFilipe Manana * btrfs_remove_free_space_cache(). 1219e3e0520bSJosef Bacik * 1220e3e0520bSJosef Bacik * And we must not remove the extent map from the fs_info->mapping_tree 1221e3e0520bSJosef Bacik * to prevent the same logical address range and physical device space 12226b7304afSFilipe Manana * ranges from being reused for a new block group. This is needed to 12236b7304afSFilipe Manana * avoid races with trimming and scrub. 12246b7304afSFilipe Manana * 12256b7304afSFilipe Manana * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is 1226e3e0520bSJosef Bacik * completely transactionless, so while it is trimming a range the 1227e3e0520bSJosef Bacik * currently running transaction might finish and a new one start, 1228e3e0520bSJosef Bacik * allowing for new block groups to be created that can reuse the same 1229e3e0520bSJosef Bacik * physical device locations unless we take this special care. 1230e3e0520bSJosef Bacik * 1231e3e0520bSJosef Bacik * There may also be an implicit trim operation if the file system 1232e3e0520bSJosef Bacik * is mounted with -odiscard. The same protections must remain 1233e3e0520bSJosef Bacik * in place until the extents have been discarded completely when 1234e3e0520bSJosef Bacik * the transaction commit has completed. 1235e3e0520bSJosef Bacik */ 12366b7304afSFilipe Manana remove_em = (atomic_read(&block_group->frozen) == 0); 1237e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1238e3e0520bSJosef Bacik 1239e3e0520bSJosef Bacik if (remove_em) { 1240e3e0520bSJosef Bacik struct extent_map_tree *em_tree; 1241e3e0520bSJosef Bacik 1242e3e0520bSJosef Bacik em_tree = &fs_info->mapping_tree; 1243e3e0520bSJosef Bacik write_lock(&em_tree->lock); 1244e3e0520bSJosef Bacik remove_extent_mapping(em_tree, em); 1245e3e0520bSJosef Bacik write_unlock(&em_tree->lock); 1246e3e0520bSJosef Bacik /* once for the tree */ 1247e3e0520bSJosef Bacik free_extent_map(em); 1248e3e0520bSJosef Bacik } 1249f6033c5eSXiyu Yang 12509fecd132SFilipe Manana out: 1251f6033c5eSXiyu Yang /* Once for the lookup reference */ 1252f6033c5eSXiyu Yang btrfs_put_block_group(block_group); 1253e3e0520bSJosef Bacik if (remove_rsv) 1254e3e0520bSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 1255e3e0520bSJosef Bacik btrfs_free_path(path); 1256e3e0520bSJosef Bacik return ret; 1257e3e0520bSJosef Bacik } 1258e3e0520bSJosef Bacik 1259e3e0520bSJosef Bacik struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( 1260e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info, const u64 chunk_offset) 1261e3e0520bSJosef Bacik { 1262dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 1263e3e0520bSJosef Bacik struct extent_map_tree *em_tree = &fs_info->mapping_tree; 1264e3e0520bSJosef Bacik struct extent_map *em; 1265e3e0520bSJosef Bacik struct map_lookup *map; 1266e3e0520bSJosef Bacik unsigned int num_items; 1267e3e0520bSJosef Bacik 1268e3e0520bSJosef Bacik read_lock(&em_tree->lock); 1269e3e0520bSJosef Bacik em = lookup_extent_mapping(em_tree, chunk_offset, 1); 1270e3e0520bSJosef Bacik read_unlock(&em_tree->lock); 1271e3e0520bSJosef Bacik ASSERT(em && em->start == chunk_offset); 1272e3e0520bSJosef Bacik 1273e3e0520bSJosef Bacik /* 1274e3e0520bSJosef Bacik * We need to reserve 3 + N units from the metadata space info in order 1275e3e0520bSJosef Bacik * to remove a block group (done at btrfs_remove_chunk() and at 1276e3e0520bSJosef Bacik * btrfs_remove_block_group()), which are used for: 1277e3e0520bSJosef Bacik * 1278e3e0520bSJosef Bacik * 1 unit for adding the free space inode's orphan (located in the tree 1279e3e0520bSJosef Bacik * of tree roots). 1280e3e0520bSJosef Bacik * 1 unit for deleting the block group item (located in the extent 1281e3e0520bSJosef Bacik * tree). 1282e3e0520bSJosef Bacik * 1 unit for deleting the free space item (located in tree of tree 1283e3e0520bSJosef Bacik * roots). 1284e3e0520bSJosef Bacik * N units for deleting N device extent items corresponding to each 1285e3e0520bSJosef Bacik * stripe (located in the device tree). 1286e3e0520bSJosef Bacik * 1287e3e0520bSJosef Bacik * In order to remove a block group we also need to reserve units in the 1288e3e0520bSJosef Bacik * system space info in order to update the chunk tree (update one or 1289e3e0520bSJosef Bacik * more device items and remove one chunk item), but this is done at 1290e3e0520bSJosef Bacik * btrfs_remove_chunk() through a call to check_system_chunk(). 1291e3e0520bSJosef Bacik */ 1292e3e0520bSJosef Bacik map = em->map_lookup; 1293e3e0520bSJosef Bacik num_items = 3 + map->num_stripes; 1294e3e0520bSJosef Bacik free_extent_map(em); 1295e3e0520bSJosef Bacik 1296dfe8aec4SJosef Bacik return btrfs_start_transaction_fallback_global_rsv(root, num_items); 1297e3e0520bSJosef Bacik } 1298e3e0520bSJosef Bacik 1299e3e0520bSJosef Bacik /* 130026ce2095SJosef Bacik * Mark block group @cache read-only, so later write won't happen to block 130126ce2095SJosef Bacik * group @cache. 130226ce2095SJosef Bacik * 130326ce2095SJosef Bacik * If @force is not set, this function will only mark the block group readonly 130426ce2095SJosef Bacik * if we have enough free space (1M) in other metadata/system block groups. 130526ce2095SJosef Bacik * If @force is not set, this function will mark the block group readonly 130626ce2095SJosef Bacik * without checking free space. 130726ce2095SJosef Bacik * 130826ce2095SJosef Bacik * NOTE: This function doesn't care if other block groups can contain all the 130926ce2095SJosef Bacik * data in this block group. That check should be done by relocation routine, 131026ce2095SJosef Bacik * not this function. 131126ce2095SJosef Bacik */ 131232da5386SDavid Sterba static int inc_block_group_ro(struct btrfs_block_group *cache, int force) 131326ce2095SJosef Bacik { 131426ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 131526ce2095SJosef Bacik u64 num_bytes; 131626ce2095SJosef Bacik int ret = -ENOSPC; 131726ce2095SJosef Bacik 131826ce2095SJosef Bacik spin_lock(&sinfo->lock); 131926ce2095SJosef Bacik spin_lock(&cache->lock); 132026ce2095SJosef Bacik 1321195a49eaSFilipe Manana if (cache->swap_extents) { 1322195a49eaSFilipe Manana ret = -ETXTBSY; 1323195a49eaSFilipe Manana goto out; 1324195a49eaSFilipe Manana } 1325195a49eaSFilipe Manana 132626ce2095SJosef Bacik if (cache->ro) { 132726ce2095SJosef Bacik cache->ro++; 132826ce2095SJosef Bacik ret = 0; 132926ce2095SJosef Bacik goto out; 133026ce2095SJosef Bacik } 133126ce2095SJosef Bacik 1332b3470b5dSDavid Sterba num_bytes = cache->length - cache->reserved - cache->pinned - 1333169e0da9SNaohiro Aota cache->bytes_super - cache->zone_unusable - cache->used; 133426ce2095SJosef Bacik 133526ce2095SJosef Bacik /* 1336a30a3d20SJosef Bacik * Data never overcommits, even in mixed mode, so do just the straight 1337a30a3d20SJosef Bacik * check of left over space in how much we have allocated. 1338a30a3d20SJosef Bacik */ 1339a30a3d20SJosef Bacik if (force) { 1340a30a3d20SJosef Bacik ret = 0; 1341a30a3d20SJosef Bacik } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) { 1342a30a3d20SJosef Bacik u64 sinfo_used = btrfs_space_info_used(sinfo, true); 1343a30a3d20SJosef Bacik 1344a30a3d20SJosef Bacik /* 134526ce2095SJosef Bacik * Here we make sure if we mark this bg RO, we still have enough 1346f8935566SJosef Bacik * free space as buffer. 134726ce2095SJosef Bacik */ 1348a30a3d20SJosef Bacik if (sinfo_used + num_bytes <= sinfo->total_bytes) 1349a30a3d20SJosef Bacik ret = 0; 1350a30a3d20SJosef Bacik } else { 1351a30a3d20SJosef Bacik /* 1352a30a3d20SJosef Bacik * We overcommit metadata, so we need to do the 1353a30a3d20SJosef Bacik * btrfs_can_overcommit check here, and we need to pass in 1354a30a3d20SJosef Bacik * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of 1355a30a3d20SJosef Bacik * leeway to allow us to mark this block group as read only. 1356a30a3d20SJosef Bacik */ 1357a30a3d20SJosef Bacik if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes, 1358a30a3d20SJosef Bacik BTRFS_RESERVE_NO_FLUSH)) 1359a30a3d20SJosef Bacik ret = 0; 1360a30a3d20SJosef Bacik } 1361a30a3d20SJosef Bacik 1362a30a3d20SJosef Bacik if (!ret) { 136326ce2095SJosef Bacik sinfo->bytes_readonly += num_bytes; 1364169e0da9SNaohiro Aota if (btrfs_is_zoned(cache->fs_info)) { 1365169e0da9SNaohiro Aota /* Migrate zone_unusable bytes to readonly */ 1366169e0da9SNaohiro Aota sinfo->bytes_readonly += cache->zone_unusable; 1367169e0da9SNaohiro Aota sinfo->bytes_zone_unusable -= cache->zone_unusable; 1368169e0da9SNaohiro Aota cache->zone_unusable = 0; 1369169e0da9SNaohiro Aota } 137026ce2095SJosef Bacik cache->ro++; 137126ce2095SJosef Bacik list_add_tail(&cache->ro_list, &sinfo->ro_bgs); 137226ce2095SJosef Bacik } 137326ce2095SJosef Bacik out: 137426ce2095SJosef Bacik spin_unlock(&cache->lock); 137526ce2095SJosef Bacik spin_unlock(&sinfo->lock); 137626ce2095SJosef Bacik if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { 137726ce2095SJosef Bacik btrfs_info(cache->fs_info, 1378b3470b5dSDavid Sterba "unable to make block group %llu ro", cache->start); 137926ce2095SJosef Bacik btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); 138026ce2095SJosef Bacik } 138126ce2095SJosef Bacik return ret; 138226ce2095SJosef Bacik } 138326ce2095SJosef Bacik 1384fe119a6eSNikolay Borisov static bool clean_pinned_extents(struct btrfs_trans_handle *trans, 1385fe119a6eSNikolay Borisov struct btrfs_block_group *bg) 138645bb5d6aSNikolay Borisov { 138745bb5d6aSNikolay Borisov struct btrfs_fs_info *fs_info = bg->fs_info; 1388fe119a6eSNikolay Borisov struct btrfs_transaction *prev_trans = NULL; 138945bb5d6aSNikolay Borisov const u64 start = bg->start; 139045bb5d6aSNikolay Borisov const u64 end = start + bg->length - 1; 139145bb5d6aSNikolay Borisov int ret; 139245bb5d6aSNikolay Borisov 1393fe119a6eSNikolay Borisov spin_lock(&fs_info->trans_lock); 1394fe119a6eSNikolay Borisov if (trans->transaction->list.prev != &fs_info->trans_list) { 1395fe119a6eSNikolay Borisov prev_trans = list_last_entry(&trans->transaction->list, 1396fe119a6eSNikolay Borisov struct btrfs_transaction, list); 1397fe119a6eSNikolay Borisov refcount_inc(&prev_trans->use_count); 1398fe119a6eSNikolay Borisov } 1399fe119a6eSNikolay Borisov spin_unlock(&fs_info->trans_lock); 1400fe119a6eSNikolay Borisov 140145bb5d6aSNikolay Borisov /* 140245bb5d6aSNikolay Borisov * Hold the unused_bg_unpin_mutex lock to avoid racing with 140345bb5d6aSNikolay Borisov * btrfs_finish_extent_commit(). If we are at transaction N, another 140445bb5d6aSNikolay Borisov * task might be running finish_extent_commit() for the previous 140545bb5d6aSNikolay Borisov * transaction N - 1, and have seen a range belonging to the block 1406fe119a6eSNikolay Borisov * group in pinned_extents before we were able to clear the whole block 1407fe119a6eSNikolay Borisov * group range from pinned_extents. This means that task can lookup for 1408fe119a6eSNikolay Borisov * the block group after we unpinned it from pinned_extents and removed 1409fe119a6eSNikolay Borisov * it, leading to a BUG_ON() at unpin_extent_range(). 141045bb5d6aSNikolay Borisov */ 141145bb5d6aSNikolay Borisov mutex_lock(&fs_info->unused_bg_unpin_mutex); 1412fe119a6eSNikolay Borisov if (prev_trans) { 1413fe119a6eSNikolay Borisov ret = clear_extent_bits(&prev_trans->pinned_extents, start, end, 141445bb5d6aSNikolay Borisov EXTENT_DIRTY); 141545bb5d6aSNikolay Borisov if (ret) 1416534cf531SFilipe Manana goto out; 1417fe119a6eSNikolay Borisov } 141845bb5d6aSNikolay Borisov 1419fe119a6eSNikolay Borisov ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end, 142045bb5d6aSNikolay Borisov EXTENT_DIRTY); 1421534cf531SFilipe Manana out: 142245bb5d6aSNikolay Borisov mutex_unlock(&fs_info->unused_bg_unpin_mutex); 14235150bf19SFilipe Manana if (prev_trans) 14245150bf19SFilipe Manana btrfs_put_transaction(prev_trans); 142545bb5d6aSNikolay Borisov 1426534cf531SFilipe Manana return ret == 0; 142745bb5d6aSNikolay Borisov } 142845bb5d6aSNikolay Borisov 142926ce2095SJosef Bacik /* 1430e3e0520bSJosef Bacik * Process the unused_bgs list and remove any that don't have any allocated 1431e3e0520bSJosef Bacik * space inside of them. 1432e3e0520bSJosef Bacik */ 1433e3e0520bSJosef Bacik void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) 1434e3e0520bSJosef Bacik { 143532da5386SDavid Sterba struct btrfs_block_group *block_group; 1436e3e0520bSJosef Bacik struct btrfs_space_info *space_info; 1437e3e0520bSJosef Bacik struct btrfs_trans_handle *trans; 14386e80d4f8SDennis Zhou const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC); 1439e3e0520bSJosef Bacik int ret = 0; 1440e3e0520bSJosef Bacik 1441e3e0520bSJosef Bacik if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 1442e3e0520bSJosef Bacik return; 1443e3e0520bSJosef Bacik 14442f12741fSJosef Bacik if (btrfs_fs_closing(fs_info)) 14452f12741fSJosef Bacik return; 14462f12741fSJosef Bacik 1447ddfd08cbSJosef Bacik /* 1448ddfd08cbSJosef Bacik * Long running balances can keep us blocked here for eternity, so 1449ddfd08cbSJosef Bacik * simply skip deletion if we're unable to get the mutex. 1450ddfd08cbSJosef Bacik */ 1451f3372065SJohannes Thumshirn if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) 1452ddfd08cbSJosef Bacik return; 1453ddfd08cbSJosef Bacik 1454e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1455e3e0520bSJosef Bacik while (!list_empty(&fs_info->unused_bgs)) { 1456e3e0520bSJosef Bacik int trimming; 1457e3e0520bSJosef Bacik 1458e3e0520bSJosef Bacik block_group = list_first_entry(&fs_info->unused_bgs, 145932da5386SDavid Sterba struct btrfs_block_group, 1460e3e0520bSJosef Bacik bg_list); 1461e3e0520bSJosef Bacik list_del_init(&block_group->bg_list); 1462e3e0520bSJosef Bacik 1463e3e0520bSJosef Bacik space_info = block_group->space_info; 1464e3e0520bSJosef Bacik 1465e3e0520bSJosef Bacik if (ret || btrfs_mixed_space_info(space_info)) { 1466e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1467e3e0520bSJosef Bacik continue; 1468e3e0520bSJosef Bacik } 1469e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1470e3e0520bSJosef Bacik 1471b0643e59SDennis Zhou btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 1472b0643e59SDennis Zhou 1473e3e0520bSJosef Bacik /* Don't want to race with allocators so take the groups_sem */ 1474e3e0520bSJosef Bacik down_write(&space_info->groups_sem); 14756e80d4f8SDennis Zhou 14766e80d4f8SDennis Zhou /* 14776e80d4f8SDennis Zhou * Async discard moves the final block group discard to be prior 14786e80d4f8SDennis Zhou * to the unused_bgs code path. Therefore, if it's not fully 14796e80d4f8SDennis Zhou * trimmed, punt it back to the async discard lists. 14806e80d4f8SDennis Zhou */ 14816e80d4f8SDennis Zhou if (btrfs_test_opt(fs_info, DISCARD_ASYNC) && 14826e80d4f8SDennis Zhou !btrfs_is_free_space_trimmed(block_group)) { 14836e80d4f8SDennis Zhou trace_btrfs_skip_unused_block_group(block_group); 14846e80d4f8SDennis Zhou up_write(&space_info->groups_sem); 14856e80d4f8SDennis Zhou /* Requeue if we failed because of async discard */ 14866e80d4f8SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 14876e80d4f8SDennis Zhou block_group); 14886e80d4f8SDennis Zhou goto next; 14896e80d4f8SDennis Zhou } 14906e80d4f8SDennis Zhou 1491e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1492e3e0520bSJosef Bacik if (block_group->reserved || block_group->pinned || 1493bf38be65SDavid Sterba block_group->used || block_group->ro || 1494e3e0520bSJosef Bacik list_is_singular(&block_group->list)) { 1495e3e0520bSJosef Bacik /* 1496e3e0520bSJosef Bacik * We want to bail if we made new allocations or have 1497e3e0520bSJosef Bacik * outstanding allocations in this block group. We do 1498e3e0520bSJosef Bacik * the ro check in case balance is currently acting on 1499e3e0520bSJosef Bacik * this block group. 1500e3e0520bSJosef Bacik */ 1501e3e0520bSJosef Bacik trace_btrfs_skip_unused_block_group(block_group); 1502e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1503e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1504e3e0520bSJosef Bacik goto next; 1505e3e0520bSJosef Bacik } 1506e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1507e3e0520bSJosef Bacik 1508e3e0520bSJosef Bacik /* We don't want to force the issue, only flip if it's ok. */ 1509e11c0406SJosef Bacik ret = inc_block_group_ro(block_group, 0); 1510e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1511e3e0520bSJosef Bacik if (ret < 0) { 1512e3e0520bSJosef Bacik ret = 0; 1513e3e0520bSJosef Bacik goto next; 1514e3e0520bSJosef Bacik } 1515e3e0520bSJosef Bacik 151674e91b12SNaohiro Aota ret = btrfs_zone_finish(block_group); 151774e91b12SNaohiro Aota if (ret < 0) { 151874e91b12SNaohiro Aota btrfs_dec_block_group_ro(block_group); 151974e91b12SNaohiro Aota if (ret == -EAGAIN) 152074e91b12SNaohiro Aota ret = 0; 152174e91b12SNaohiro Aota goto next; 152274e91b12SNaohiro Aota } 152374e91b12SNaohiro Aota 1524e3e0520bSJosef Bacik /* 1525e3e0520bSJosef Bacik * Want to do this before we do anything else so we can recover 1526e3e0520bSJosef Bacik * properly if we fail to join the transaction. 1527e3e0520bSJosef Bacik */ 1528e3e0520bSJosef Bacik trans = btrfs_start_trans_remove_block_group(fs_info, 1529b3470b5dSDavid Sterba block_group->start); 1530e3e0520bSJosef Bacik if (IS_ERR(trans)) { 1531e3e0520bSJosef Bacik btrfs_dec_block_group_ro(block_group); 1532e3e0520bSJosef Bacik ret = PTR_ERR(trans); 1533e3e0520bSJosef Bacik goto next; 1534e3e0520bSJosef Bacik } 1535e3e0520bSJosef Bacik 1536e3e0520bSJosef Bacik /* 1537e3e0520bSJosef Bacik * We could have pending pinned extents for this block group, 1538e3e0520bSJosef Bacik * just delete them, we don't care about them anymore. 1539e3e0520bSJosef Bacik */ 1540534cf531SFilipe Manana if (!clean_pinned_extents(trans, block_group)) { 1541534cf531SFilipe Manana btrfs_dec_block_group_ro(block_group); 1542e3e0520bSJosef Bacik goto end_trans; 1543534cf531SFilipe Manana } 1544e3e0520bSJosef Bacik 1545b0643e59SDennis Zhou /* 1546b0643e59SDennis Zhou * At this point, the block_group is read only and should fail 1547b0643e59SDennis Zhou * new allocations. However, btrfs_finish_extent_commit() can 1548b0643e59SDennis Zhou * cause this block_group to be placed back on the discard 1549b0643e59SDennis Zhou * lists because now the block_group isn't fully discarded. 1550b0643e59SDennis Zhou * Bail here and try again later after discarding everything. 1551b0643e59SDennis Zhou */ 1552b0643e59SDennis Zhou spin_lock(&fs_info->discard_ctl.lock); 1553b0643e59SDennis Zhou if (!list_empty(&block_group->discard_list)) { 1554b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1555b0643e59SDennis Zhou btrfs_dec_block_group_ro(block_group); 1556b0643e59SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 1557b0643e59SDennis Zhou block_group); 1558b0643e59SDennis Zhou goto end_trans; 1559b0643e59SDennis Zhou } 1560b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1561b0643e59SDennis Zhou 1562e3e0520bSJosef Bacik /* Reset pinned so btrfs_put_block_group doesn't complain */ 1563e3e0520bSJosef Bacik spin_lock(&space_info->lock); 1564e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1565e3e0520bSJosef Bacik 1566e3e0520bSJosef Bacik btrfs_space_info_update_bytes_pinned(fs_info, space_info, 1567e3e0520bSJosef Bacik -block_group->pinned); 1568e3e0520bSJosef Bacik space_info->bytes_readonly += block_group->pinned; 1569e3e0520bSJosef Bacik block_group->pinned = 0; 1570e3e0520bSJosef Bacik 1571e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1572e3e0520bSJosef Bacik spin_unlock(&space_info->lock); 1573e3e0520bSJosef Bacik 15746e80d4f8SDennis Zhou /* 15756e80d4f8SDennis Zhou * The normal path here is an unused block group is passed here, 15766e80d4f8SDennis Zhou * then trimming is handled in the transaction commit path. 15776e80d4f8SDennis Zhou * Async discard interposes before this to do the trimming 15786e80d4f8SDennis Zhou * before coming down the unused block group path as trimming 15796e80d4f8SDennis Zhou * will no longer be done later in the transaction commit path. 15806e80d4f8SDennis Zhou */ 15816e80d4f8SDennis Zhou if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC)) 15826e80d4f8SDennis Zhou goto flip_async; 15836e80d4f8SDennis Zhou 1584dcba6e48SNaohiro Aota /* 1585dcba6e48SNaohiro Aota * DISCARD can flip during remount. On zoned filesystems, we 1586dcba6e48SNaohiro Aota * need to reset sequential-required zones. 1587dcba6e48SNaohiro Aota */ 1588dcba6e48SNaohiro Aota trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) || 1589dcba6e48SNaohiro Aota btrfs_is_zoned(fs_info); 1590e3e0520bSJosef Bacik 1591e3e0520bSJosef Bacik /* Implicit trim during transaction commit. */ 1592e3e0520bSJosef Bacik if (trimming) 15936b7304afSFilipe Manana btrfs_freeze_block_group(block_group); 1594e3e0520bSJosef Bacik 1595e3e0520bSJosef Bacik /* 1596e3e0520bSJosef Bacik * Btrfs_remove_chunk will abort the transaction if things go 1597e3e0520bSJosef Bacik * horribly wrong. 1598e3e0520bSJosef Bacik */ 1599b3470b5dSDavid Sterba ret = btrfs_remove_chunk(trans, block_group->start); 1600e3e0520bSJosef Bacik 1601e3e0520bSJosef Bacik if (ret) { 1602e3e0520bSJosef Bacik if (trimming) 16036b7304afSFilipe Manana btrfs_unfreeze_block_group(block_group); 1604e3e0520bSJosef Bacik goto end_trans; 1605e3e0520bSJosef Bacik } 1606e3e0520bSJosef Bacik 1607e3e0520bSJosef Bacik /* 1608e3e0520bSJosef Bacik * If we're not mounted with -odiscard, we can just forget 1609e3e0520bSJosef Bacik * about this block group. Otherwise we'll need to wait 1610e3e0520bSJosef Bacik * until transaction commit to do the actual discard. 1611e3e0520bSJosef Bacik */ 1612e3e0520bSJosef Bacik if (trimming) { 1613e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1614e3e0520bSJosef Bacik /* 1615e3e0520bSJosef Bacik * A concurrent scrub might have added us to the list 1616e3e0520bSJosef Bacik * fs_info->unused_bgs, so use a list_move operation 1617e3e0520bSJosef Bacik * to add the block group to the deleted_bgs list. 1618e3e0520bSJosef Bacik */ 1619e3e0520bSJosef Bacik list_move(&block_group->bg_list, 1620e3e0520bSJosef Bacik &trans->transaction->deleted_bgs); 1621e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1622e3e0520bSJosef Bacik btrfs_get_block_group(block_group); 1623e3e0520bSJosef Bacik } 1624e3e0520bSJosef Bacik end_trans: 1625e3e0520bSJosef Bacik btrfs_end_transaction(trans); 1626e3e0520bSJosef Bacik next: 1627e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1628e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1629e3e0520bSJosef Bacik } 1630e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1631f3372065SJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 16326e80d4f8SDennis Zhou return; 16336e80d4f8SDennis Zhou 16346e80d4f8SDennis Zhou flip_async: 16356e80d4f8SDennis Zhou btrfs_end_transaction(trans); 1636f3372065SJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 16376e80d4f8SDennis Zhou btrfs_put_block_group(block_group); 16386e80d4f8SDennis Zhou btrfs_discard_punt_unused_bgs_list(fs_info); 1639e3e0520bSJosef Bacik } 1640e3e0520bSJosef Bacik 164132da5386SDavid Sterba void btrfs_mark_bg_unused(struct btrfs_block_group *bg) 1642e3e0520bSJosef Bacik { 1643e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info = bg->fs_info; 1644e3e0520bSJosef Bacik 1645e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1646e3e0520bSJosef Bacik if (list_empty(&bg->bg_list)) { 1647e3e0520bSJosef Bacik btrfs_get_block_group(bg); 1648e3e0520bSJosef Bacik trace_btrfs_add_unused_block_group(bg); 1649e3e0520bSJosef Bacik list_add_tail(&bg->bg_list, &fs_info->unused_bgs); 1650e3e0520bSJosef Bacik } 1651e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1652e3e0520bSJosef Bacik } 16534358d963SJosef Bacik 16542ca0ec77SJohannes Thumshirn /* 16552ca0ec77SJohannes Thumshirn * We want block groups with a low number of used bytes to be in the beginning 16562ca0ec77SJohannes Thumshirn * of the list, so they will get reclaimed first. 16572ca0ec77SJohannes Thumshirn */ 16582ca0ec77SJohannes Thumshirn static int reclaim_bgs_cmp(void *unused, const struct list_head *a, 16592ca0ec77SJohannes Thumshirn const struct list_head *b) 16602ca0ec77SJohannes Thumshirn { 16612ca0ec77SJohannes Thumshirn const struct btrfs_block_group *bg1, *bg2; 16622ca0ec77SJohannes Thumshirn 16632ca0ec77SJohannes Thumshirn bg1 = list_entry(a, struct btrfs_block_group, bg_list); 16642ca0ec77SJohannes Thumshirn bg2 = list_entry(b, struct btrfs_block_group, bg_list); 16652ca0ec77SJohannes Thumshirn 16662ca0ec77SJohannes Thumshirn return bg1->used > bg2->used; 16672ca0ec77SJohannes Thumshirn } 16682ca0ec77SJohannes Thumshirn 16693687fcb0SJohannes Thumshirn static inline bool btrfs_should_reclaim(struct btrfs_fs_info *fs_info) 16703687fcb0SJohannes Thumshirn { 16713687fcb0SJohannes Thumshirn if (btrfs_is_zoned(fs_info)) 16723687fcb0SJohannes Thumshirn return btrfs_zoned_should_reclaim(fs_info); 16733687fcb0SJohannes Thumshirn return true; 16743687fcb0SJohannes Thumshirn } 16753687fcb0SJohannes Thumshirn 167681531225SBoris Burkov static bool should_reclaim_block_group(struct btrfs_block_group *bg, u64 bytes_freed) 167781531225SBoris Burkov { 167881531225SBoris Burkov const struct btrfs_space_info *space_info = bg->space_info; 167981531225SBoris Burkov const int reclaim_thresh = READ_ONCE(space_info->bg_reclaim_threshold); 168081531225SBoris Burkov const u64 new_val = bg->used; 168181531225SBoris Burkov const u64 old_val = new_val + bytes_freed; 168281531225SBoris Burkov u64 thresh; 168381531225SBoris Burkov 168481531225SBoris Burkov if (reclaim_thresh == 0) 168581531225SBoris Burkov return false; 168681531225SBoris Burkov 1687428c8e03SDavid Sterba thresh = mult_perc(bg->length, reclaim_thresh); 168881531225SBoris Burkov 168981531225SBoris Burkov /* 169081531225SBoris Burkov * If we were below the threshold before don't reclaim, we are likely a 169181531225SBoris Burkov * brand new block group and we don't want to relocate new block groups. 169281531225SBoris Burkov */ 169381531225SBoris Burkov if (old_val < thresh) 169481531225SBoris Burkov return false; 169581531225SBoris Burkov if (new_val >= thresh) 169681531225SBoris Burkov return false; 169781531225SBoris Burkov return true; 169881531225SBoris Burkov } 169981531225SBoris Burkov 170018bb8bbfSJohannes Thumshirn void btrfs_reclaim_bgs_work(struct work_struct *work) 170118bb8bbfSJohannes Thumshirn { 170218bb8bbfSJohannes Thumshirn struct btrfs_fs_info *fs_info = 170318bb8bbfSJohannes Thumshirn container_of(work, struct btrfs_fs_info, reclaim_bgs_work); 170418bb8bbfSJohannes Thumshirn struct btrfs_block_group *bg; 170518bb8bbfSJohannes Thumshirn struct btrfs_space_info *space_info; 170618bb8bbfSJohannes Thumshirn 170718bb8bbfSJohannes Thumshirn if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 170818bb8bbfSJohannes Thumshirn return; 170918bb8bbfSJohannes Thumshirn 17102f12741fSJosef Bacik if (btrfs_fs_closing(fs_info)) 17112f12741fSJosef Bacik return; 17122f12741fSJosef Bacik 17133687fcb0SJohannes Thumshirn if (!btrfs_should_reclaim(fs_info)) 17143687fcb0SJohannes Thumshirn return; 17153687fcb0SJohannes Thumshirn 1716ca5e4ea0SNaohiro Aota sb_start_write(fs_info->sb); 1717ca5e4ea0SNaohiro Aota 1718ca5e4ea0SNaohiro Aota if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 1719ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 172018bb8bbfSJohannes Thumshirn return; 1721ca5e4ea0SNaohiro Aota } 172218bb8bbfSJohannes Thumshirn 17239cc0b837SJohannes Thumshirn /* 17249cc0b837SJohannes Thumshirn * Long running balances can keep us blocked here for eternity, so 17259cc0b837SJohannes Thumshirn * simply skip reclaim if we're unable to get the mutex. 17269cc0b837SJohannes Thumshirn */ 17279cc0b837SJohannes Thumshirn if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) { 17289cc0b837SJohannes Thumshirn btrfs_exclop_finish(fs_info); 1729ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 17309cc0b837SJohannes Thumshirn return; 17319cc0b837SJohannes Thumshirn } 17329cc0b837SJohannes Thumshirn 173318bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 17342ca0ec77SJohannes Thumshirn /* 17352ca0ec77SJohannes Thumshirn * Sort happens under lock because we can't simply splice it and sort. 17362ca0ec77SJohannes Thumshirn * The block groups might still be in use and reachable via bg_list, 17372ca0ec77SJohannes Thumshirn * and their presence in the reclaim_bgs list must be preserved. 17382ca0ec77SJohannes Thumshirn */ 17392ca0ec77SJohannes Thumshirn list_sort(NULL, &fs_info->reclaim_bgs, reclaim_bgs_cmp); 174018bb8bbfSJohannes Thumshirn while (!list_empty(&fs_info->reclaim_bgs)) { 17415f93e776SJohannes Thumshirn u64 zone_unusable; 17421cea5cf0SFilipe Manana int ret = 0; 17431cea5cf0SFilipe Manana 174418bb8bbfSJohannes Thumshirn bg = list_first_entry(&fs_info->reclaim_bgs, 174518bb8bbfSJohannes Thumshirn struct btrfs_block_group, 174618bb8bbfSJohannes Thumshirn bg_list); 174718bb8bbfSJohannes Thumshirn list_del_init(&bg->bg_list); 174818bb8bbfSJohannes Thumshirn 174918bb8bbfSJohannes Thumshirn space_info = bg->space_info; 175018bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 175118bb8bbfSJohannes Thumshirn 175218bb8bbfSJohannes Thumshirn /* Don't race with allocators so take the groups_sem */ 175318bb8bbfSJohannes Thumshirn down_write(&space_info->groups_sem); 175418bb8bbfSJohannes Thumshirn 175518bb8bbfSJohannes Thumshirn spin_lock(&bg->lock); 175618bb8bbfSJohannes Thumshirn if (bg->reserved || bg->pinned || bg->ro) { 175718bb8bbfSJohannes Thumshirn /* 175818bb8bbfSJohannes Thumshirn * We want to bail if we made new allocations or have 175918bb8bbfSJohannes Thumshirn * outstanding allocations in this block group. We do 176018bb8bbfSJohannes Thumshirn * the ro check in case balance is currently acting on 176118bb8bbfSJohannes Thumshirn * this block group. 176218bb8bbfSJohannes Thumshirn */ 176318bb8bbfSJohannes Thumshirn spin_unlock(&bg->lock); 176418bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 176518bb8bbfSJohannes Thumshirn goto next; 176618bb8bbfSJohannes Thumshirn } 1767cc4804bfSBoris Burkov if (bg->used == 0) { 1768cc4804bfSBoris Burkov /* 1769cc4804bfSBoris Burkov * It is possible that we trigger relocation on a block 1770cc4804bfSBoris Burkov * group as its extents are deleted and it first goes 1771cc4804bfSBoris Burkov * below the threshold, then shortly after goes empty. 1772cc4804bfSBoris Burkov * 1773cc4804bfSBoris Burkov * In this case, relocating it does delete it, but has 1774cc4804bfSBoris Burkov * some overhead in relocation specific metadata, looking 1775cc4804bfSBoris Burkov * for the non-existent extents and running some extra 1776cc4804bfSBoris Burkov * transactions, which we can avoid by using one of the 1777cc4804bfSBoris Burkov * other mechanisms for dealing with empty block groups. 1778cc4804bfSBoris Burkov */ 1779cc4804bfSBoris Burkov if (!btrfs_test_opt(fs_info, DISCARD_ASYNC)) 1780cc4804bfSBoris Burkov btrfs_mark_bg_unused(bg); 1781cc4804bfSBoris Burkov spin_unlock(&bg->lock); 1782cc4804bfSBoris Burkov up_write(&space_info->groups_sem); 1783cc4804bfSBoris Burkov goto next; 178481531225SBoris Burkov 178581531225SBoris Burkov } 178681531225SBoris Burkov /* 178781531225SBoris Burkov * The block group might no longer meet the reclaim condition by 178881531225SBoris Burkov * the time we get around to reclaiming it, so to avoid 178981531225SBoris Burkov * reclaiming overly full block_groups, skip reclaiming them. 179081531225SBoris Burkov * 179181531225SBoris Burkov * Since the decision making process also depends on the amount 179281531225SBoris Burkov * being freed, pass in a fake giant value to skip that extra 179381531225SBoris Burkov * check, which is more meaningful when adding to the list in 179481531225SBoris Burkov * the first place. 179581531225SBoris Burkov */ 179681531225SBoris Burkov if (!should_reclaim_block_group(bg, bg->length)) { 179781531225SBoris Burkov spin_unlock(&bg->lock); 179881531225SBoris Burkov up_write(&space_info->groups_sem); 179981531225SBoris Burkov goto next; 1800cc4804bfSBoris Burkov } 180118bb8bbfSJohannes Thumshirn spin_unlock(&bg->lock); 180218bb8bbfSJohannes Thumshirn 180318bb8bbfSJohannes Thumshirn /* Get out fast, in case we're unmounting the filesystem */ 180418bb8bbfSJohannes Thumshirn if (btrfs_fs_closing(fs_info)) { 180518bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 180618bb8bbfSJohannes Thumshirn goto next; 180718bb8bbfSJohannes Thumshirn } 180818bb8bbfSJohannes Thumshirn 18095f93e776SJohannes Thumshirn /* 18105f93e776SJohannes Thumshirn * Cache the zone_unusable value before turning the block group 18115f93e776SJohannes Thumshirn * to read only. As soon as the blog group is read only it's 18125f93e776SJohannes Thumshirn * zone_unusable value gets moved to the block group's read-only 18135f93e776SJohannes Thumshirn * bytes and isn't available for calculations anymore. 18145f93e776SJohannes Thumshirn */ 18155f93e776SJohannes Thumshirn zone_unusable = bg->zone_unusable; 181618bb8bbfSJohannes Thumshirn ret = inc_block_group_ro(bg, 0); 181718bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 181818bb8bbfSJohannes Thumshirn if (ret < 0) 181918bb8bbfSJohannes Thumshirn goto next; 182018bb8bbfSJohannes Thumshirn 18215f93e776SJohannes Thumshirn btrfs_info(fs_info, 18225f93e776SJohannes Thumshirn "reclaiming chunk %llu with %llu%% used %llu%% unusable", 182395cd356cSJohannes Thumshirn bg->start, 182495cd356cSJohannes Thumshirn div64_u64(bg->used * 100, bg->length), 18255f93e776SJohannes Thumshirn div64_u64(zone_unusable * 100, bg->length)); 182618bb8bbfSJohannes Thumshirn trace_btrfs_reclaim_block_group(bg); 182718bb8bbfSJohannes Thumshirn ret = btrfs_relocate_chunk(fs_info, bg->start); 182874944c87SJosef Bacik if (ret) { 182974944c87SJosef Bacik btrfs_dec_block_group_ro(bg); 183018bb8bbfSJohannes Thumshirn btrfs_err(fs_info, "error relocating chunk %llu", 183118bb8bbfSJohannes Thumshirn bg->start); 183274944c87SJosef Bacik } 183318bb8bbfSJohannes Thumshirn 183418bb8bbfSJohannes Thumshirn next: 18351cea5cf0SFilipe Manana btrfs_put_block_group(bg); 1836d96b3424SFilipe Manana spin_lock(&fs_info->unused_bgs_lock); 183718bb8bbfSJohannes Thumshirn } 183818bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 183918bb8bbfSJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 184018bb8bbfSJohannes Thumshirn btrfs_exclop_finish(fs_info); 1841ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 184218bb8bbfSJohannes Thumshirn } 184318bb8bbfSJohannes Thumshirn 184418bb8bbfSJohannes Thumshirn void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info) 184518bb8bbfSJohannes Thumshirn { 184618bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 184718bb8bbfSJohannes Thumshirn if (!list_empty(&fs_info->reclaim_bgs)) 184818bb8bbfSJohannes Thumshirn queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work); 184918bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 185018bb8bbfSJohannes Thumshirn } 185118bb8bbfSJohannes Thumshirn 185218bb8bbfSJohannes Thumshirn void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg) 185318bb8bbfSJohannes Thumshirn { 185418bb8bbfSJohannes Thumshirn struct btrfs_fs_info *fs_info = bg->fs_info; 185518bb8bbfSJohannes Thumshirn 185618bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 185718bb8bbfSJohannes Thumshirn if (list_empty(&bg->bg_list)) { 185818bb8bbfSJohannes Thumshirn btrfs_get_block_group(bg); 185918bb8bbfSJohannes Thumshirn trace_btrfs_add_reclaim_block_group(bg); 186018bb8bbfSJohannes Thumshirn list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs); 186118bb8bbfSJohannes Thumshirn } 186218bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 186318bb8bbfSJohannes Thumshirn } 186418bb8bbfSJohannes Thumshirn 1865e3ba67a1SJohannes Thumshirn static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key, 1866e3ba67a1SJohannes Thumshirn struct btrfs_path *path) 1867e3ba67a1SJohannes Thumshirn { 1868e3ba67a1SJohannes Thumshirn struct extent_map_tree *em_tree; 1869e3ba67a1SJohannes Thumshirn struct extent_map *em; 1870e3ba67a1SJohannes Thumshirn struct btrfs_block_group_item bg; 1871e3ba67a1SJohannes Thumshirn struct extent_buffer *leaf; 1872e3ba67a1SJohannes Thumshirn int slot; 1873e3ba67a1SJohannes Thumshirn u64 flags; 1874e3ba67a1SJohannes Thumshirn int ret = 0; 1875e3ba67a1SJohannes Thumshirn 1876e3ba67a1SJohannes Thumshirn slot = path->slots[0]; 1877e3ba67a1SJohannes Thumshirn leaf = path->nodes[0]; 1878e3ba67a1SJohannes Thumshirn 1879e3ba67a1SJohannes Thumshirn em_tree = &fs_info->mapping_tree; 1880e3ba67a1SJohannes Thumshirn read_lock(&em_tree->lock); 1881e3ba67a1SJohannes Thumshirn em = lookup_extent_mapping(em_tree, key->objectid, key->offset); 1882e3ba67a1SJohannes Thumshirn read_unlock(&em_tree->lock); 1883e3ba67a1SJohannes Thumshirn if (!em) { 1884e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1885e3ba67a1SJohannes Thumshirn "logical %llu len %llu found bg but no related chunk", 1886e3ba67a1SJohannes Thumshirn key->objectid, key->offset); 1887e3ba67a1SJohannes Thumshirn return -ENOENT; 1888e3ba67a1SJohannes Thumshirn } 1889e3ba67a1SJohannes Thumshirn 1890e3ba67a1SJohannes Thumshirn if (em->start != key->objectid || em->len != key->offset) { 1891e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1892e3ba67a1SJohannes Thumshirn "block group %llu len %llu mismatch with chunk %llu len %llu", 1893e3ba67a1SJohannes Thumshirn key->objectid, key->offset, em->start, em->len); 1894e3ba67a1SJohannes Thumshirn ret = -EUCLEAN; 1895e3ba67a1SJohannes Thumshirn goto out_free_em; 1896e3ba67a1SJohannes Thumshirn } 1897e3ba67a1SJohannes Thumshirn 1898e3ba67a1SJohannes Thumshirn read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot), 1899e3ba67a1SJohannes Thumshirn sizeof(bg)); 1900e3ba67a1SJohannes Thumshirn flags = btrfs_stack_block_group_flags(&bg) & 1901e3ba67a1SJohannes Thumshirn BTRFS_BLOCK_GROUP_TYPE_MASK; 1902e3ba67a1SJohannes Thumshirn 1903e3ba67a1SJohannes Thumshirn if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 1904e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1905e3ba67a1SJohannes Thumshirn "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", 1906e3ba67a1SJohannes Thumshirn key->objectid, key->offset, flags, 1907e3ba67a1SJohannes Thumshirn (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type)); 1908e3ba67a1SJohannes Thumshirn ret = -EUCLEAN; 1909e3ba67a1SJohannes Thumshirn } 1910e3ba67a1SJohannes Thumshirn 1911e3ba67a1SJohannes Thumshirn out_free_em: 1912e3ba67a1SJohannes Thumshirn free_extent_map(em); 1913e3ba67a1SJohannes Thumshirn return ret; 1914e3ba67a1SJohannes Thumshirn } 1915e3ba67a1SJohannes Thumshirn 19164358d963SJosef Bacik static int find_first_block_group(struct btrfs_fs_info *fs_info, 19174358d963SJosef Bacik struct btrfs_path *path, 19184358d963SJosef Bacik struct btrfs_key *key) 19194358d963SJosef Bacik { 1920dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 1921e3ba67a1SJohannes Thumshirn int ret; 19224358d963SJosef Bacik struct btrfs_key found_key; 19234358d963SJosef Bacik 192436dfbbe2SGabriel Niebler btrfs_for_each_slot(root, key, &found_key, path, ret) { 19254358d963SJosef Bacik if (found_key.objectid >= key->objectid && 19264358d963SJosef Bacik found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { 192736dfbbe2SGabriel Niebler return read_bg_from_eb(fs_info, &found_key, path); 1928e3ba67a1SJohannes Thumshirn } 19294358d963SJosef Bacik } 19304358d963SJosef Bacik return ret; 19314358d963SJosef Bacik } 19324358d963SJosef Bacik 19334358d963SJosef Bacik static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) 19344358d963SJosef Bacik { 19354358d963SJosef Bacik u64 extra_flags = chunk_to_extended(flags) & 19364358d963SJosef Bacik BTRFS_EXTENDED_PROFILE_MASK; 19374358d963SJosef Bacik 19384358d963SJosef Bacik write_seqlock(&fs_info->profiles_lock); 19394358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 19404358d963SJosef Bacik fs_info->avail_data_alloc_bits |= extra_flags; 19414358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_METADATA) 19424358d963SJosef Bacik fs_info->avail_metadata_alloc_bits |= extra_flags; 19434358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 19444358d963SJosef Bacik fs_info->avail_system_alloc_bits |= extra_flags; 19454358d963SJosef Bacik write_sequnlock(&fs_info->profiles_lock); 19464358d963SJosef Bacik } 19474358d963SJosef Bacik 194843dd529aSDavid Sterba /* 194943dd529aSDavid Sterba * Map a physical disk address to a list of logical addresses. 19509ee9b979SNikolay Borisov * 19519ee9b979SNikolay Borisov * @fs_info: the filesystem 195296a14336SNikolay Borisov * @chunk_start: logical address of block group 195396a14336SNikolay Borisov * @physical: physical address to map to logical addresses 195496a14336SNikolay Borisov * @logical: return array of logical addresses which map to @physical 195596a14336SNikolay Borisov * @naddrs: length of @logical 195696a14336SNikolay Borisov * @stripe_len: size of IO stripe for the given block group 195796a14336SNikolay Borisov * 195896a14336SNikolay Borisov * Maps a particular @physical disk address to a list of @logical addresses. 195996a14336SNikolay Borisov * Used primarily to exclude those portions of a block group that contain super 196096a14336SNikolay Borisov * block copies. 196196a14336SNikolay Borisov */ 196296a14336SNikolay Borisov int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, 19631eb82ef8SChristoph Hellwig u64 physical, u64 **logical, int *naddrs, int *stripe_len) 196496a14336SNikolay Borisov { 196596a14336SNikolay Borisov struct extent_map *em; 196696a14336SNikolay Borisov struct map_lookup *map; 196796a14336SNikolay Borisov u64 *buf; 196896a14336SNikolay Borisov u64 bytenr; 19691776ad17SNikolay Borisov u64 data_stripe_length; 19701776ad17SNikolay Borisov u64 io_stripe_size; 19711776ad17SNikolay Borisov int i, nr = 0; 19721776ad17SNikolay Borisov int ret = 0; 197396a14336SNikolay Borisov 197496a14336SNikolay Borisov em = btrfs_get_chunk_map(fs_info, chunk_start, 1); 197596a14336SNikolay Borisov if (IS_ERR(em)) 197696a14336SNikolay Borisov return -EIO; 197796a14336SNikolay Borisov 197896a14336SNikolay Borisov map = em->map_lookup; 19799e22b925SNikolay Borisov data_stripe_length = em->orig_block_len; 1980a97699d1SQu Wenruo io_stripe_size = BTRFS_STRIPE_LEN; 1981138082f3SNaohiro Aota chunk_start = em->start; 198296a14336SNikolay Borisov 19839e22b925SNikolay Borisov /* For RAID5/6 adjust to a full IO stripe length */ 19849e22b925SNikolay Borisov if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 1985a97699d1SQu Wenruo io_stripe_size = nr_data_stripes(map) << BTRFS_STRIPE_LEN_SHIFT; 198696a14336SNikolay Borisov 198796a14336SNikolay Borisov buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); 19881776ad17SNikolay Borisov if (!buf) { 19891776ad17SNikolay Borisov ret = -ENOMEM; 19901776ad17SNikolay Borisov goto out; 19911776ad17SNikolay Borisov } 199296a14336SNikolay Borisov 199396a14336SNikolay Borisov for (i = 0; i < map->num_stripes; i++) { 19941776ad17SNikolay Borisov bool already_inserted = false; 19956ded22c1SQu Wenruo u32 stripe_nr; 19966ded22c1SQu Wenruo u32 offset; 19971776ad17SNikolay Borisov int j; 19981776ad17SNikolay Borisov 19991776ad17SNikolay Borisov if (!in_range(physical, map->stripes[i].physical, 20001776ad17SNikolay Borisov data_stripe_length)) 200196a14336SNikolay Borisov continue; 200296a14336SNikolay Borisov 2003a97699d1SQu Wenruo stripe_nr = (physical - map->stripes[i].physical) >> 2004a97699d1SQu Wenruo BTRFS_STRIPE_LEN_SHIFT; 2005a97699d1SQu Wenruo offset = (physical - map->stripes[i].physical) & 2006a97699d1SQu Wenruo BTRFS_STRIPE_LEN_MASK; 200796a14336SNikolay Borisov 2008ac067734SDavid Sterba if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 20096ded22c1SQu Wenruo BTRFS_BLOCK_GROUP_RAID10)) 20106ded22c1SQu Wenruo stripe_nr = div_u64(stripe_nr * map->num_stripes + i, 20116ded22c1SQu Wenruo map->sub_stripes); 201296a14336SNikolay Borisov /* 201396a14336SNikolay Borisov * The remaining case would be for RAID56, multiply by 201496a14336SNikolay Borisov * nr_data_stripes(). Alternatively, just use rmap_len below 201596a14336SNikolay Borisov * instead of map->stripe_len 201696a14336SNikolay Borisov */ 2017138082f3SNaohiro Aota bytenr = chunk_start + stripe_nr * io_stripe_size + offset; 20181776ad17SNikolay Borisov 20191776ad17SNikolay Borisov /* Ensure we don't add duplicate addresses */ 202096a14336SNikolay Borisov for (j = 0; j < nr; j++) { 20211776ad17SNikolay Borisov if (buf[j] == bytenr) { 20221776ad17SNikolay Borisov already_inserted = true; 202396a14336SNikolay Borisov break; 202496a14336SNikolay Borisov } 202596a14336SNikolay Borisov } 20261776ad17SNikolay Borisov 20271776ad17SNikolay Borisov if (!already_inserted) 20281776ad17SNikolay Borisov buf[nr++] = bytenr; 202996a14336SNikolay Borisov } 203096a14336SNikolay Borisov 203196a14336SNikolay Borisov *logical = buf; 203296a14336SNikolay Borisov *naddrs = nr; 20331776ad17SNikolay Borisov *stripe_len = io_stripe_size; 20341776ad17SNikolay Borisov out: 203596a14336SNikolay Borisov free_extent_map(em); 20361776ad17SNikolay Borisov return ret; 203796a14336SNikolay Borisov } 203896a14336SNikolay Borisov 203932da5386SDavid Sterba static int exclude_super_stripes(struct btrfs_block_group *cache) 20404358d963SJosef Bacik { 20414358d963SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 204212659251SNaohiro Aota const bool zoned = btrfs_is_zoned(fs_info); 20434358d963SJosef Bacik u64 bytenr; 20444358d963SJosef Bacik u64 *logical; 20454358d963SJosef Bacik int stripe_len; 20464358d963SJosef Bacik int i, nr, ret; 20474358d963SJosef Bacik 2048b3470b5dSDavid Sterba if (cache->start < BTRFS_SUPER_INFO_OFFSET) { 2049b3470b5dSDavid Sterba stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start; 20504358d963SJosef Bacik cache->bytes_super += stripe_len; 2051b3470b5dSDavid Sterba ret = btrfs_add_excluded_extent(fs_info, cache->start, 20524358d963SJosef Bacik stripe_len); 20534358d963SJosef Bacik if (ret) 20544358d963SJosef Bacik return ret; 20554358d963SJosef Bacik } 20564358d963SJosef Bacik 20574358d963SJosef Bacik for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 20584358d963SJosef Bacik bytenr = btrfs_sb_offset(i); 20591eb82ef8SChristoph Hellwig ret = btrfs_rmap_block(fs_info, cache->start, 20604358d963SJosef Bacik bytenr, &logical, &nr, &stripe_len); 20614358d963SJosef Bacik if (ret) 20624358d963SJosef Bacik return ret; 20634358d963SJosef Bacik 206412659251SNaohiro Aota /* Shouldn't have super stripes in sequential zones */ 206512659251SNaohiro Aota if (zoned && nr) { 206612659251SNaohiro Aota btrfs_err(fs_info, 206712659251SNaohiro Aota "zoned: block group %llu must not contain super block", 206812659251SNaohiro Aota cache->start); 206912659251SNaohiro Aota return -EUCLEAN; 207012659251SNaohiro Aota } 207112659251SNaohiro Aota 20724358d963SJosef Bacik while (nr--) { 207396f9b0f2SNikolay Borisov u64 len = min_t(u64, stripe_len, 207496f9b0f2SNikolay Borisov cache->start + cache->length - logical[nr]); 20754358d963SJosef Bacik 20764358d963SJosef Bacik cache->bytes_super += len; 207796f9b0f2SNikolay Borisov ret = btrfs_add_excluded_extent(fs_info, logical[nr], 207896f9b0f2SNikolay Borisov len); 20794358d963SJosef Bacik if (ret) { 20804358d963SJosef Bacik kfree(logical); 20814358d963SJosef Bacik return ret; 20824358d963SJosef Bacik } 20834358d963SJosef Bacik } 20844358d963SJosef Bacik 20854358d963SJosef Bacik kfree(logical); 20864358d963SJosef Bacik } 20874358d963SJosef Bacik return 0; 20884358d963SJosef Bacik } 20894358d963SJosef Bacik 209032da5386SDavid Sterba static struct btrfs_block_group *btrfs_create_block_group_cache( 20919afc6649SQu Wenruo struct btrfs_fs_info *fs_info, u64 start) 20924358d963SJosef Bacik { 209332da5386SDavid Sterba struct btrfs_block_group *cache; 20944358d963SJosef Bacik 20954358d963SJosef Bacik cache = kzalloc(sizeof(*cache), GFP_NOFS); 20964358d963SJosef Bacik if (!cache) 20974358d963SJosef Bacik return NULL; 20984358d963SJosef Bacik 20994358d963SJosef Bacik cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), 21004358d963SJosef Bacik GFP_NOFS); 21014358d963SJosef Bacik if (!cache->free_space_ctl) { 21024358d963SJosef Bacik kfree(cache); 21034358d963SJosef Bacik return NULL; 21044358d963SJosef Bacik } 21054358d963SJosef Bacik 2106b3470b5dSDavid Sterba cache->start = start; 21074358d963SJosef Bacik 21084358d963SJosef Bacik cache->fs_info = fs_info; 21094358d963SJosef Bacik cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); 21104358d963SJosef Bacik 21116e80d4f8SDennis Zhou cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED; 21126e80d4f8SDennis Zhou 211348aaeebeSJosef Bacik refcount_set(&cache->refs, 1); 21144358d963SJosef Bacik spin_lock_init(&cache->lock); 21154358d963SJosef Bacik init_rwsem(&cache->data_rwsem); 21164358d963SJosef Bacik INIT_LIST_HEAD(&cache->list); 21174358d963SJosef Bacik INIT_LIST_HEAD(&cache->cluster_list); 21184358d963SJosef Bacik INIT_LIST_HEAD(&cache->bg_list); 21194358d963SJosef Bacik INIT_LIST_HEAD(&cache->ro_list); 2120b0643e59SDennis Zhou INIT_LIST_HEAD(&cache->discard_list); 21214358d963SJosef Bacik INIT_LIST_HEAD(&cache->dirty_list); 21224358d963SJosef Bacik INIT_LIST_HEAD(&cache->io_list); 2123afba2bc0SNaohiro Aota INIT_LIST_HEAD(&cache->active_bg_list); 2124cd79909bSJosef Bacik btrfs_init_free_space_ctl(cache, cache->free_space_ctl); 21256b7304afSFilipe Manana atomic_set(&cache->frozen, 0); 21264358d963SJosef Bacik mutex_init(&cache->free_space_lock); 2127c29abab4SJosef Bacik cache->full_stripe_locks_root.root = RB_ROOT; 2128c29abab4SJosef Bacik mutex_init(&cache->full_stripe_locks_root.lock); 21294358d963SJosef Bacik 21304358d963SJosef Bacik return cache; 21314358d963SJosef Bacik } 21324358d963SJosef Bacik 21334358d963SJosef Bacik /* 21344358d963SJosef Bacik * Iterate all chunks and verify that each of them has the corresponding block 21354358d963SJosef Bacik * group 21364358d963SJosef Bacik */ 21374358d963SJosef Bacik static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) 21384358d963SJosef Bacik { 21394358d963SJosef Bacik struct extent_map_tree *map_tree = &fs_info->mapping_tree; 21404358d963SJosef Bacik struct extent_map *em; 214132da5386SDavid Sterba struct btrfs_block_group *bg; 21424358d963SJosef Bacik u64 start = 0; 21434358d963SJosef Bacik int ret = 0; 21444358d963SJosef Bacik 21454358d963SJosef Bacik while (1) { 21464358d963SJosef Bacik read_lock(&map_tree->lock); 21474358d963SJosef Bacik /* 21484358d963SJosef Bacik * lookup_extent_mapping will return the first extent map 21494358d963SJosef Bacik * intersecting the range, so setting @len to 1 is enough to 21504358d963SJosef Bacik * get the first chunk. 21514358d963SJosef Bacik */ 21524358d963SJosef Bacik em = lookup_extent_mapping(map_tree, start, 1); 21534358d963SJosef Bacik read_unlock(&map_tree->lock); 21544358d963SJosef Bacik if (!em) 21554358d963SJosef Bacik break; 21564358d963SJosef Bacik 21574358d963SJosef Bacik bg = btrfs_lookup_block_group(fs_info, em->start); 21584358d963SJosef Bacik if (!bg) { 21594358d963SJosef Bacik btrfs_err(fs_info, 21604358d963SJosef Bacik "chunk start=%llu len=%llu doesn't have corresponding block group", 21614358d963SJosef Bacik em->start, em->len); 21624358d963SJosef Bacik ret = -EUCLEAN; 21634358d963SJosef Bacik free_extent_map(em); 21644358d963SJosef Bacik break; 21654358d963SJosef Bacik } 2166b3470b5dSDavid Sterba if (bg->start != em->start || bg->length != em->len || 21674358d963SJosef Bacik (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != 21684358d963SJosef Bacik (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 21694358d963SJosef Bacik btrfs_err(fs_info, 21704358d963SJosef Bacik "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", 21714358d963SJosef Bacik em->start, em->len, 21724358d963SJosef Bacik em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, 2173b3470b5dSDavid Sterba bg->start, bg->length, 21744358d963SJosef Bacik bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); 21754358d963SJosef Bacik ret = -EUCLEAN; 21764358d963SJosef Bacik free_extent_map(em); 21774358d963SJosef Bacik btrfs_put_block_group(bg); 21784358d963SJosef Bacik break; 21794358d963SJosef Bacik } 21804358d963SJosef Bacik start = em->start + em->len; 21814358d963SJosef Bacik free_extent_map(em); 21824358d963SJosef Bacik btrfs_put_block_group(bg); 21834358d963SJosef Bacik } 21844358d963SJosef Bacik return ret; 21854358d963SJosef Bacik } 21864358d963SJosef Bacik 2187ffb9e0f0SQu Wenruo static int read_one_block_group(struct btrfs_fs_info *info, 21884afd2fe8SJohannes Thumshirn struct btrfs_block_group_item *bgi, 2189d49a2ddbSQu Wenruo const struct btrfs_key *key, 2190ffb9e0f0SQu Wenruo int need_clear) 2191ffb9e0f0SQu Wenruo { 219232da5386SDavid Sterba struct btrfs_block_group *cache; 2193ffb9e0f0SQu Wenruo const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS); 2194ffb9e0f0SQu Wenruo int ret; 2195ffb9e0f0SQu Wenruo 2196d49a2ddbSQu Wenruo ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY); 2197ffb9e0f0SQu Wenruo 21989afc6649SQu Wenruo cache = btrfs_create_block_group_cache(info, key->objectid); 2199ffb9e0f0SQu Wenruo if (!cache) 2200ffb9e0f0SQu Wenruo return -ENOMEM; 2201ffb9e0f0SQu Wenruo 22024afd2fe8SJohannes Thumshirn cache->length = key->offset; 22034afd2fe8SJohannes Thumshirn cache->used = btrfs_stack_block_group_used(bgi); 22047248e0ceSQu Wenruo cache->commit_used = cache->used; 22054afd2fe8SJohannes Thumshirn cache->flags = btrfs_stack_block_group_flags(bgi); 2206f7238e50SJosef Bacik cache->global_root_id = btrfs_stack_block_group_chunk_objectid(bgi); 22079afc6649SQu Wenruo 2208e3e39c72SMarcos Paulo de Souza set_free_space_tree_thresholds(cache); 2209e3e39c72SMarcos Paulo de Souza 2210ffb9e0f0SQu Wenruo if (need_clear) { 2211ffb9e0f0SQu Wenruo /* 2212ffb9e0f0SQu Wenruo * When we mount with old space cache, we need to 2213ffb9e0f0SQu Wenruo * set BTRFS_DC_CLEAR and set dirty flag. 2214ffb9e0f0SQu Wenruo * 2215ffb9e0f0SQu Wenruo * a) Setting 'BTRFS_DC_CLEAR' makes sure that we 2216ffb9e0f0SQu Wenruo * truncate the old free space cache inode and 2217ffb9e0f0SQu Wenruo * setup a new one. 2218ffb9e0f0SQu Wenruo * b) Setting 'dirty flag' makes sure that we flush 2219ffb9e0f0SQu Wenruo * the new space cache info onto disk. 2220ffb9e0f0SQu Wenruo */ 2221ffb9e0f0SQu Wenruo if (btrfs_test_opt(info, SPACE_CACHE)) 2222ffb9e0f0SQu Wenruo cache->disk_cache_state = BTRFS_DC_CLEAR; 2223ffb9e0f0SQu Wenruo } 2224ffb9e0f0SQu Wenruo if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && 2225ffb9e0f0SQu Wenruo (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { 2226ffb9e0f0SQu Wenruo btrfs_err(info, 2227ffb9e0f0SQu Wenruo "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", 2228ffb9e0f0SQu Wenruo cache->start); 2229ffb9e0f0SQu Wenruo ret = -EINVAL; 2230ffb9e0f0SQu Wenruo goto error; 2231ffb9e0f0SQu Wenruo } 2232ffb9e0f0SQu Wenruo 2233a94794d5SNaohiro Aota ret = btrfs_load_block_group_zone_info(cache, false); 223408e11a3dSNaohiro Aota if (ret) { 223508e11a3dSNaohiro Aota btrfs_err(info, "zoned: failed to load zone info of bg %llu", 223608e11a3dSNaohiro Aota cache->start); 223708e11a3dSNaohiro Aota goto error; 223808e11a3dSNaohiro Aota } 223908e11a3dSNaohiro Aota 2240ffb9e0f0SQu Wenruo /* 2241ffb9e0f0SQu Wenruo * We need to exclude the super stripes now so that the space info has 2242ffb9e0f0SQu Wenruo * super bytes accounted for, otherwise we'll think we have more space 2243ffb9e0f0SQu Wenruo * than we actually do. 2244ffb9e0f0SQu Wenruo */ 2245ffb9e0f0SQu Wenruo ret = exclude_super_stripes(cache); 2246ffb9e0f0SQu Wenruo if (ret) { 2247ffb9e0f0SQu Wenruo /* We may have excluded something, so call this just in case. */ 2248ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2249ffb9e0f0SQu Wenruo goto error; 2250ffb9e0f0SQu Wenruo } 2251ffb9e0f0SQu Wenruo 2252ffb9e0f0SQu Wenruo /* 2253169e0da9SNaohiro Aota * For zoned filesystem, space after the allocation offset is the only 2254169e0da9SNaohiro Aota * free space for a block group. So, we don't need any caching work. 2255169e0da9SNaohiro Aota * btrfs_calc_zone_unusable() will set the amount of free space and 2256169e0da9SNaohiro Aota * zone_unusable space. 2257169e0da9SNaohiro Aota * 2258169e0da9SNaohiro Aota * For regular filesystem, check for two cases, either we are full, and 2259169e0da9SNaohiro Aota * therefore don't need to bother with the caching work since we won't 2260169e0da9SNaohiro Aota * find any space, or we are empty, and we can just add all the space 2261169e0da9SNaohiro Aota * in and be done with it. This saves us _a_lot_ of time, particularly 2262169e0da9SNaohiro Aota * in the full case. 2263ffb9e0f0SQu Wenruo */ 2264169e0da9SNaohiro Aota if (btrfs_is_zoned(info)) { 2265169e0da9SNaohiro Aota btrfs_calc_zone_unusable(cache); 2266c46c4247SNaohiro Aota /* Should not have any excluded extents. Just in case, though. */ 2267c46c4247SNaohiro Aota btrfs_free_excluded_extents(cache); 2268169e0da9SNaohiro Aota } else if (cache->length == cache->used) { 2269ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 2270ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2271ffb9e0f0SQu Wenruo } else if (cache->used == 0) { 2272ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 22739afc6649SQu Wenruo add_new_free_space(cache, cache->start, 22749afc6649SQu Wenruo cache->start + cache->length); 2275ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2276ffb9e0f0SQu Wenruo } 2277ffb9e0f0SQu Wenruo 2278ffb9e0f0SQu Wenruo ret = btrfs_add_block_group_cache(info, cache); 2279ffb9e0f0SQu Wenruo if (ret) { 2280ffb9e0f0SQu Wenruo btrfs_remove_free_space_cache(cache); 2281ffb9e0f0SQu Wenruo goto error; 2282ffb9e0f0SQu Wenruo } 2283ffb9e0f0SQu Wenruo trace_btrfs_add_block_group(info, cache, 0); 2284723de71dSJosef Bacik btrfs_add_bg_to_space_info(info, cache); 2285ffb9e0f0SQu Wenruo 2286ffb9e0f0SQu Wenruo set_avail_alloc_bits(info, cache->flags); 2287a09f23c3SAnand Jain if (btrfs_chunk_writeable(info, cache->start)) { 2288a09f23c3SAnand Jain if (cache->used == 0) { 2289ffb9e0f0SQu Wenruo ASSERT(list_empty(&cache->bg_list)); 22906e80d4f8SDennis Zhou if (btrfs_test_opt(info, DISCARD_ASYNC)) 22916e80d4f8SDennis Zhou btrfs_discard_queue_work(&info->discard_ctl, cache); 22926e80d4f8SDennis Zhou else 2293ffb9e0f0SQu Wenruo btrfs_mark_bg_unused(cache); 2294ffb9e0f0SQu Wenruo } 2295a09f23c3SAnand Jain } else { 2296a09f23c3SAnand Jain inc_block_group_ro(cache, 1); 2297a09f23c3SAnand Jain } 2298a09f23c3SAnand Jain 2299ffb9e0f0SQu Wenruo return 0; 2300ffb9e0f0SQu Wenruo error: 2301ffb9e0f0SQu Wenruo btrfs_put_block_group(cache); 2302ffb9e0f0SQu Wenruo return ret; 2303ffb9e0f0SQu Wenruo } 2304ffb9e0f0SQu Wenruo 230542437a63SJosef Bacik static int fill_dummy_bgs(struct btrfs_fs_info *fs_info) 230642437a63SJosef Bacik { 230742437a63SJosef Bacik struct extent_map_tree *em_tree = &fs_info->mapping_tree; 230842437a63SJosef Bacik struct rb_node *node; 230942437a63SJosef Bacik int ret = 0; 231042437a63SJosef Bacik 231142437a63SJosef Bacik for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 231242437a63SJosef Bacik struct extent_map *em; 231342437a63SJosef Bacik struct map_lookup *map; 231442437a63SJosef Bacik struct btrfs_block_group *bg; 231542437a63SJosef Bacik 231642437a63SJosef Bacik em = rb_entry(node, struct extent_map, rb_node); 231742437a63SJosef Bacik map = em->map_lookup; 231842437a63SJosef Bacik bg = btrfs_create_block_group_cache(fs_info, em->start); 231942437a63SJosef Bacik if (!bg) { 232042437a63SJosef Bacik ret = -ENOMEM; 232142437a63SJosef Bacik break; 232242437a63SJosef Bacik } 232342437a63SJosef Bacik 232442437a63SJosef Bacik /* Fill dummy cache as FULL */ 232542437a63SJosef Bacik bg->length = em->len; 232642437a63SJosef Bacik bg->flags = map->type; 232742437a63SJosef Bacik bg->cached = BTRFS_CACHE_FINISHED; 232842437a63SJosef Bacik bg->used = em->len; 232942437a63SJosef Bacik bg->flags = map->type; 233042437a63SJosef Bacik ret = btrfs_add_block_group_cache(fs_info, bg); 23312b29726cSQu Wenruo /* 23322b29726cSQu Wenruo * We may have some valid block group cache added already, in 23332b29726cSQu Wenruo * that case we skip to the next one. 23342b29726cSQu Wenruo */ 23352b29726cSQu Wenruo if (ret == -EEXIST) { 23362b29726cSQu Wenruo ret = 0; 23372b29726cSQu Wenruo btrfs_put_block_group(bg); 23382b29726cSQu Wenruo continue; 23392b29726cSQu Wenruo } 23402b29726cSQu Wenruo 234142437a63SJosef Bacik if (ret) { 234242437a63SJosef Bacik btrfs_remove_free_space_cache(bg); 234342437a63SJosef Bacik btrfs_put_block_group(bg); 234442437a63SJosef Bacik break; 234542437a63SJosef Bacik } 23462b29726cSQu Wenruo 2347723de71dSJosef Bacik btrfs_add_bg_to_space_info(fs_info, bg); 234842437a63SJosef Bacik 234942437a63SJosef Bacik set_avail_alloc_bits(fs_info, bg->flags); 235042437a63SJosef Bacik } 235142437a63SJosef Bacik if (!ret) 235242437a63SJosef Bacik btrfs_init_global_block_rsv(fs_info); 235342437a63SJosef Bacik return ret; 235442437a63SJosef Bacik } 235542437a63SJosef Bacik 23564358d963SJosef Bacik int btrfs_read_block_groups(struct btrfs_fs_info *info) 23574358d963SJosef Bacik { 2358dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(info); 23594358d963SJosef Bacik struct btrfs_path *path; 23604358d963SJosef Bacik int ret; 236132da5386SDavid Sterba struct btrfs_block_group *cache; 23624358d963SJosef Bacik struct btrfs_space_info *space_info; 23634358d963SJosef Bacik struct btrfs_key key; 23644358d963SJosef Bacik int need_clear = 0; 23654358d963SJosef Bacik u64 cache_gen; 23664358d963SJosef Bacik 236781d5d614SQu Wenruo /* 236881d5d614SQu Wenruo * Either no extent root (with ibadroots rescue option) or we have 236981d5d614SQu Wenruo * unsupported RO options. The fs can never be mounted read-write, so no 237081d5d614SQu Wenruo * need to waste time searching block group items. 237181d5d614SQu Wenruo * 237281d5d614SQu Wenruo * This also allows new extent tree related changes to be RO compat, 237381d5d614SQu Wenruo * no need for a full incompat flag. 237481d5d614SQu Wenruo */ 237581d5d614SQu Wenruo if (!root || (btrfs_super_compat_ro_flags(info->super_copy) & 237681d5d614SQu Wenruo ~BTRFS_FEATURE_COMPAT_RO_SUPP)) 237742437a63SJosef Bacik return fill_dummy_bgs(info); 237842437a63SJosef Bacik 23794358d963SJosef Bacik key.objectid = 0; 23804358d963SJosef Bacik key.offset = 0; 23814358d963SJosef Bacik key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 23824358d963SJosef Bacik path = btrfs_alloc_path(); 23834358d963SJosef Bacik if (!path) 23844358d963SJosef Bacik return -ENOMEM; 23854358d963SJosef Bacik 23864358d963SJosef Bacik cache_gen = btrfs_super_cache_generation(info->super_copy); 23874358d963SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 23884358d963SJosef Bacik btrfs_super_generation(info->super_copy) != cache_gen) 23894358d963SJosef Bacik need_clear = 1; 23904358d963SJosef Bacik if (btrfs_test_opt(info, CLEAR_CACHE)) 23914358d963SJosef Bacik need_clear = 1; 23924358d963SJosef Bacik 23934358d963SJosef Bacik while (1) { 23944afd2fe8SJohannes Thumshirn struct btrfs_block_group_item bgi; 23954afd2fe8SJohannes Thumshirn struct extent_buffer *leaf; 23964afd2fe8SJohannes Thumshirn int slot; 23974afd2fe8SJohannes Thumshirn 23984358d963SJosef Bacik ret = find_first_block_group(info, path, &key); 23994358d963SJosef Bacik if (ret > 0) 24004358d963SJosef Bacik break; 24014358d963SJosef Bacik if (ret != 0) 24024358d963SJosef Bacik goto error; 24034358d963SJosef Bacik 24044afd2fe8SJohannes Thumshirn leaf = path->nodes[0]; 24054afd2fe8SJohannes Thumshirn slot = path->slots[0]; 24064afd2fe8SJohannes Thumshirn 24074afd2fe8SJohannes Thumshirn read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), 24084afd2fe8SJohannes Thumshirn sizeof(bgi)); 24094afd2fe8SJohannes Thumshirn 24104afd2fe8SJohannes Thumshirn btrfs_item_key_to_cpu(leaf, &key, slot); 24114afd2fe8SJohannes Thumshirn btrfs_release_path(path); 24124afd2fe8SJohannes Thumshirn ret = read_one_block_group(info, &bgi, &key, need_clear); 2413ffb9e0f0SQu Wenruo if (ret < 0) 24144358d963SJosef Bacik goto error; 2415ffb9e0f0SQu Wenruo key.objectid += key.offset; 2416ffb9e0f0SQu Wenruo key.offset = 0; 24174358d963SJosef Bacik } 24187837fa88SJosef Bacik btrfs_release_path(path); 24194358d963SJosef Bacik 242072804905SJosef Bacik list_for_each_entry(space_info, &info->space_info, list) { 242149ea112dSJosef Bacik int i; 242249ea112dSJosef Bacik 242349ea112dSJosef Bacik for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 242449ea112dSJosef Bacik if (list_empty(&space_info->block_groups[i])) 242549ea112dSJosef Bacik continue; 242649ea112dSJosef Bacik cache = list_first_entry(&space_info->block_groups[i], 242749ea112dSJosef Bacik struct btrfs_block_group, 242849ea112dSJosef Bacik list); 242949ea112dSJosef Bacik btrfs_sysfs_add_block_group_type(cache); 243049ea112dSJosef Bacik } 243149ea112dSJosef Bacik 24324358d963SJosef Bacik if (!(btrfs_get_alloc_profile(info, space_info->flags) & 24334358d963SJosef Bacik (BTRFS_BLOCK_GROUP_RAID10 | 24344358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID1_MASK | 24354358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID56_MASK | 24364358d963SJosef Bacik BTRFS_BLOCK_GROUP_DUP))) 24374358d963SJosef Bacik continue; 24384358d963SJosef Bacik /* 24394358d963SJosef Bacik * Avoid allocating from un-mirrored block group if there are 24404358d963SJosef Bacik * mirrored block groups. 24414358d963SJosef Bacik */ 24424358d963SJosef Bacik list_for_each_entry(cache, 24434358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_RAID0], 24444358d963SJosef Bacik list) 2445e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 24464358d963SJosef Bacik list_for_each_entry(cache, 24474358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_SINGLE], 24484358d963SJosef Bacik list) 2449e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 24504358d963SJosef Bacik } 24514358d963SJosef Bacik 24524358d963SJosef Bacik btrfs_init_global_block_rsv(info); 24534358d963SJosef Bacik ret = check_chunk_block_group_mappings(info); 24544358d963SJosef Bacik error: 24554358d963SJosef Bacik btrfs_free_path(path); 24562b29726cSQu Wenruo /* 24572b29726cSQu Wenruo * We've hit some error while reading the extent tree, and have 24582b29726cSQu Wenruo * rescue=ibadroots mount option. 24592b29726cSQu Wenruo * Try to fill the tree using dummy block groups so that the user can 24602b29726cSQu Wenruo * continue to mount and grab their data. 24612b29726cSQu Wenruo */ 24622b29726cSQu Wenruo if (ret && btrfs_test_opt(info, IGNOREBADROOTS)) 24632b29726cSQu Wenruo ret = fill_dummy_bgs(info); 24644358d963SJosef Bacik return ret; 24654358d963SJosef Bacik } 24664358d963SJosef Bacik 246779bd3712SFilipe Manana /* 246879bd3712SFilipe Manana * This function, insert_block_group_item(), belongs to the phase 2 of chunk 246979bd3712SFilipe Manana * allocation. 247079bd3712SFilipe Manana * 247179bd3712SFilipe Manana * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 247279bd3712SFilipe Manana * phases. 247379bd3712SFilipe Manana */ 247497f4728aSQu Wenruo static int insert_block_group_item(struct btrfs_trans_handle *trans, 247597f4728aSQu Wenruo struct btrfs_block_group *block_group) 247697f4728aSQu Wenruo { 247797f4728aSQu Wenruo struct btrfs_fs_info *fs_info = trans->fs_info; 247897f4728aSQu Wenruo struct btrfs_block_group_item bgi; 2479dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 248097f4728aSQu Wenruo struct btrfs_key key; 2481675dfe12SFilipe Manana u64 old_commit_used; 2482675dfe12SFilipe Manana int ret; 248397f4728aSQu Wenruo 248497f4728aSQu Wenruo spin_lock(&block_group->lock); 248597f4728aSQu Wenruo btrfs_set_stack_block_group_used(&bgi, block_group->used); 248697f4728aSQu Wenruo btrfs_set_stack_block_group_chunk_objectid(&bgi, 2487f7238e50SJosef Bacik block_group->global_root_id); 248897f4728aSQu Wenruo btrfs_set_stack_block_group_flags(&bgi, block_group->flags); 2489675dfe12SFilipe Manana old_commit_used = block_group->commit_used; 2490675dfe12SFilipe Manana block_group->commit_used = block_group->used; 249197f4728aSQu Wenruo key.objectid = block_group->start; 249297f4728aSQu Wenruo key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 249397f4728aSQu Wenruo key.offset = block_group->length; 249497f4728aSQu Wenruo spin_unlock(&block_group->lock); 249597f4728aSQu Wenruo 2496675dfe12SFilipe Manana ret = btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi)); 2497675dfe12SFilipe Manana if (ret < 0) { 2498675dfe12SFilipe Manana spin_lock(&block_group->lock); 2499675dfe12SFilipe Manana block_group->commit_used = old_commit_used; 2500675dfe12SFilipe Manana spin_unlock(&block_group->lock); 2501675dfe12SFilipe Manana } 2502675dfe12SFilipe Manana 2503675dfe12SFilipe Manana return ret; 250497f4728aSQu Wenruo } 250597f4728aSQu Wenruo 25062eadb9e7SNikolay Borisov static int insert_dev_extent(struct btrfs_trans_handle *trans, 25072eadb9e7SNikolay Borisov struct btrfs_device *device, u64 chunk_offset, 25082eadb9e7SNikolay Borisov u64 start, u64 num_bytes) 25092eadb9e7SNikolay Borisov { 25102eadb9e7SNikolay Borisov struct btrfs_fs_info *fs_info = device->fs_info; 25112eadb9e7SNikolay Borisov struct btrfs_root *root = fs_info->dev_root; 25122eadb9e7SNikolay Borisov struct btrfs_path *path; 25132eadb9e7SNikolay Borisov struct btrfs_dev_extent *extent; 25142eadb9e7SNikolay Borisov struct extent_buffer *leaf; 25152eadb9e7SNikolay Borisov struct btrfs_key key; 25162eadb9e7SNikolay Borisov int ret; 25172eadb9e7SNikolay Borisov 25182eadb9e7SNikolay Borisov WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); 25192eadb9e7SNikolay Borisov WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 25202eadb9e7SNikolay Borisov path = btrfs_alloc_path(); 25212eadb9e7SNikolay Borisov if (!path) 25222eadb9e7SNikolay Borisov return -ENOMEM; 25232eadb9e7SNikolay Borisov 25242eadb9e7SNikolay Borisov key.objectid = device->devid; 25252eadb9e7SNikolay Borisov key.type = BTRFS_DEV_EXTENT_KEY; 25262eadb9e7SNikolay Borisov key.offset = start; 25272eadb9e7SNikolay Borisov ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*extent)); 25282eadb9e7SNikolay Borisov if (ret) 25292eadb9e7SNikolay Borisov goto out; 25302eadb9e7SNikolay Borisov 25312eadb9e7SNikolay Borisov leaf = path->nodes[0]; 25322eadb9e7SNikolay Borisov extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); 25332eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID); 25342eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_objectid(leaf, extent, 25352eadb9e7SNikolay Borisov BTRFS_FIRST_CHUNK_TREE_OBJECTID); 25362eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); 25372eadb9e7SNikolay Borisov 25382eadb9e7SNikolay Borisov btrfs_set_dev_extent_length(leaf, extent, num_bytes); 25392eadb9e7SNikolay Borisov btrfs_mark_buffer_dirty(leaf); 25402eadb9e7SNikolay Borisov out: 25412eadb9e7SNikolay Borisov btrfs_free_path(path); 25422eadb9e7SNikolay Borisov return ret; 25432eadb9e7SNikolay Borisov } 25442eadb9e7SNikolay Borisov 25452eadb9e7SNikolay Borisov /* 25462eadb9e7SNikolay Borisov * This function belongs to phase 2. 25472eadb9e7SNikolay Borisov * 25482eadb9e7SNikolay Borisov * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 25492eadb9e7SNikolay Borisov * phases. 25502eadb9e7SNikolay Borisov */ 25512eadb9e7SNikolay Borisov static int insert_dev_extents(struct btrfs_trans_handle *trans, 25522eadb9e7SNikolay Borisov u64 chunk_offset, u64 chunk_size) 25532eadb9e7SNikolay Borisov { 25542eadb9e7SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 25552eadb9e7SNikolay Borisov struct btrfs_device *device; 25562eadb9e7SNikolay Borisov struct extent_map *em; 25572eadb9e7SNikolay Borisov struct map_lookup *map; 25582eadb9e7SNikolay Borisov u64 dev_offset; 25592eadb9e7SNikolay Borisov u64 stripe_size; 25602eadb9e7SNikolay Borisov int i; 25612eadb9e7SNikolay Borisov int ret = 0; 25622eadb9e7SNikolay Borisov 25632eadb9e7SNikolay Borisov em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); 25642eadb9e7SNikolay Borisov if (IS_ERR(em)) 25652eadb9e7SNikolay Borisov return PTR_ERR(em); 25662eadb9e7SNikolay Borisov 25672eadb9e7SNikolay Borisov map = em->map_lookup; 25682eadb9e7SNikolay Borisov stripe_size = em->orig_block_len; 25692eadb9e7SNikolay Borisov 25702eadb9e7SNikolay Borisov /* 25712eadb9e7SNikolay Borisov * Take the device list mutex to prevent races with the final phase of 25722eadb9e7SNikolay Borisov * a device replace operation that replaces the device object associated 25732eadb9e7SNikolay Borisov * with the map's stripes, because the device object's id can change 25742eadb9e7SNikolay Borisov * at any time during that final phase of the device replace operation 25752eadb9e7SNikolay Borisov * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 25762eadb9e7SNikolay Borisov * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 25772eadb9e7SNikolay Borisov * resulting in persisting a device extent item with such ID. 25782eadb9e7SNikolay Borisov */ 25792eadb9e7SNikolay Borisov mutex_lock(&fs_info->fs_devices->device_list_mutex); 25802eadb9e7SNikolay Borisov for (i = 0; i < map->num_stripes; i++) { 25812eadb9e7SNikolay Borisov device = map->stripes[i].dev; 25822eadb9e7SNikolay Borisov dev_offset = map->stripes[i].physical; 25832eadb9e7SNikolay Borisov 25842eadb9e7SNikolay Borisov ret = insert_dev_extent(trans, device, chunk_offset, dev_offset, 25852eadb9e7SNikolay Borisov stripe_size); 25862eadb9e7SNikolay Borisov if (ret) 25872eadb9e7SNikolay Borisov break; 25882eadb9e7SNikolay Borisov } 25892eadb9e7SNikolay Borisov mutex_unlock(&fs_info->fs_devices->device_list_mutex); 25902eadb9e7SNikolay Borisov 25912eadb9e7SNikolay Borisov free_extent_map(em); 25922eadb9e7SNikolay Borisov return ret; 25932eadb9e7SNikolay Borisov } 25942eadb9e7SNikolay Borisov 259579bd3712SFilipe Manana /* 259679bd3712SFilipe Manana * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of 259779bd3712SFilipe Manana * chunk allocation. 259879bd3712SFilipe Manana * 259979bd3712SFilipe Manana * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 260079bd3712SFilipe Manana * phases. 260179bd3712SFilipe Manana */ 26024358d963SJosef Bacik void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) 26034358d963SJosef Bacik { 26044358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 260532da5386SDavid Sterba struct btrfs_block_group *block_group; 26064358d963SJosef Bacik int ret = 0; 26074358d963SJosef Bacik 26084358d963SJosef Bacik while (!list_empty(&trans->new_bgs)) { 260949ea112dSJosef Bacik int index; 261049ea112dSJosef Bacik 26114358d963SJosef Bacik block_group = list_first_entry(&trans->new_bgs, 261232da5386SDavid Sterba struct btrfs_block_group, 26134358d963SJosef Bacik bg_list); 26144358d963SJosef Bacik if (ret) 26154358d963SJosef Bacik goto next; 26164358d963SJosef Bacik 261749ea112dSJosef Bacik index = btrfs_bg_flags_to_raid_index(block_group->flags); 261849ea112dSJosef Bacik 261997f4728aSQu Wenruo ret = insert_block_group_item(trans, block_group); 26204358d963SJosef Bacik if (ret) 26214358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 26223349b57fSJosef Bacik if (!test_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, 26233349b57fSJosef Bacik &block_group->runtime_flags)) { 262479bd3712SFilipe Manana mutex_lock(&fs_info->chunk_mutex); 262579bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, block_group); 262679bd3712SFilipe Manana mutex_unlock(&fs_info->chunk_mutex); 262779bd3712SFilipe Manana if (ret) 262879bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 262979bd3712SFilipe Manana } 26302eadb9e7SNikolay Borisov ret = insert_dev_extents(trans, block_group->start, 263197f4728aSQu Wenruo block_group->length); 26324358d963SJosef Bacik if (ret) 26334358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 26344358d963SJosef Bacik add_block_group_free_space(trans, block_group); 263549ea112dSJosef Bacik 263649ea112dSJosef Bacik /* 263749ea112dSJosef Bacik * If we restriped during balance, we may have added a new raid 263849ea112dSJosef Bacik * type, so now add the sysfs entries when it is safe to do so. 263949ea112dSJosef Bacik * We don't have to worry about locking here as it's handled in 264049ea112dSJosef Bacik * btrfs_sysfs_add_block_group_type. 264149ea112dSJosef Bacik */ 264249ea112dSJosef Bacik if (block_group->space_info->block_group_kobjs[index] == NULL) 264349ea112dSJosef Bacik btrfs_sysfs_add_block_group_type(block_group); 264449ea112dSJosef Bacik 26454358d963SJosef Bacik /* Already aborted the transaction if it failed. */ 26464358d963SJosef Bacik next: 26474358d963SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 26484358d963SJosef Bacik list_del_init(&block_group->bg_list); 26494358d963SJosef Bacik } 26504358d963SJosef Bacik btrfs_trans_release_chunk_metadata(trans); 26514358d963SJosef Bacik } 26524358d963SJosef Bacik 2653f7238e50SJosef Bacik /* 2654f7238e50SJosef Bacik * For extent tree v2 we use the block_group_item->chunk_offset to point at our 2655f7238e50SJosef Bacik * global root id. For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID. 2656f7238e50SJosef Bacik */ 2657f7238e50SJosef Bacik static u64 calculate_global_root_id(struct btrfs_fs_info *fs_info, u64 offset) 2658f7238e50SJosef Bacik { 2659f7238e50SJosef Bacik u64 div = SZ_1G; 2660f7238e50SJosef Bacik u64 index; 2661f7238e50SJosef Bacik 2662f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 2663f7238e50SJosef Bacik return BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2664f7238e50SJosef Bacik 2665f7238e50SJosef Bacik /* If we have a smaller fs index based on 128MiB. */ 2666f7238e50SJosef Bacik if (btrfs_super_total_bytes(fs_info->super_copy) <= (SZ_1G * 10ULL)) 2667f7238e50SJosef Bacik div = SZ_128M; 2668f7238e50SJosef Bacik 2669f7238e50SJosef Bacik offset = div64_u64(offset, div); 2670f7238e50SJosef Bacik div64_u64_rem(offset, fs_info->nr_global_roots, &index); 2671f7238e50SJosef Bacik return index; 2672f7238e50SJosef Bacik } 2673f7238e50SJosef Bacik 267479bd3712SFilipe Manana struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans, 2675*5758d1bdSFilipe Manana u64 type, 267679bd3712SFilipe Manana u64 chunk_offset, u64 size) 26774358d963SJosef Bacik { 26784358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 267932da5386SDavid Sterba struct btrfs_block_group *cache; 26804358d963SJosef Bacik int ret; 26814358d963SJosef Bacik 26824358d963SJosef Bacik btrfs_set_log_full_commit(trans); 26834358d963SJosef Bacik 26849afc6649SQu Wenruo cache = btrfs_create_block_group_cache(fs_info, chunk_offset); 26854358d963SJosef Bacik if (!cache) 268679bd3712SFilipe Manana return ERR_PTR(-ENOMEM); 26874358d963SJosef Bacik 26889afc6649SQu Wenruo cache->length = size; 2689e3e39c72SMarcos Paulo de Souza set_free_space_tree_thresholds(cache); 26904358d963SJosef Bacik cache->flags = type; 26914358d963SJosef Bacik cache->cached = BTRFS_CACHE_FINISHED; 2692f7238e50SJosef Bacik cache->global_root_id = calculate_global_root_id(fs_info, cache->start); 2693f7238e50SJosef Bacik 2694997e3e2eSBoris Burkov if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) 26950d7764ffSDavid Sterba set_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &cache->runtime_flags); 269608e11a3dSNaohiro Aota 2697a94794d5SNaohiro Aota ret = btrfs_load_block_group_zone_info(cache, true); 269808e11a3dSNaohiro Aota if (ret) { 269908e11a3dSNaohiro Aota btrfs_put_block_group(cache); 270079bd3712SFilipe Manana return ERR_PTR(ret); 270108e11a3dSNaohiro Aota } 270208e11a3dSNaohiro Aota 27034358d963SJosef Bacik ret = exclude_super_stripes(cache); 27044358d963SJosef Bacik if (ret) { 27054358d963SJosef Bacik /* We may have excluded something, so call this just in case */ 27064358d963SJosef Bacik btrfs_free_excluded_extents(cache); 27074358d963SJosef Bacik btrfs_put_block_group(cache); 270879bd3712SFilipe Manana return ERR_PTR(ret); 27094358d963SJosef Bacik } 27104358d963SJosef Bacik 27114358d963SJosef Bacik add_new_free_space(cache, chunk_offset, chunk_offset + size); 27124358d963SJosef Bacik 27134358d963SJosef Bacik btrfs_free_excluded_extents(cache); 27144358d963SJosef Bacik 27154358d963SJosef Bacik /* 27164358d963SJosef Bacik * Ensure the corresponding space_info object is created and 27174358d963SJosef Bacik * assigned to our block group. We want our bg to be added to the rbtree 27184358d963SJosef Bacik * with its ->space_info set. 27194358d963SJosef Bacik */ 27204358d963SJosef Bacik cache->space_info = btrfs_find_space_info(fs_info, cache->flags); 27214358d963SJosef Bacik ASSERT(cache->space_info); 27224358d963SJosef Bacik 27234358d963SJosef Bacik ret = btrfs_add_block_group_cache(fs_info, cache); 27244358d963SJosef Bacik if (ret) { 27254358d963SJosef Bacik btrfs_remove_free_space_cache(cache); 27264358d963SJosef Bacik btrfs_put_block_group(cache); 272779bd3712SFilipe Manana return ERR_PTR(ret); 27284358d963SJosef Bacik } 27294358d963SJosef Bacik 27304358d963SJosef Bacik /* 27314358d963SJosef Bacik * Now that our block group has its ->space_info set and is inserted in 27324358d963SJosef Bacik * the rbtree, update the space info's counters. 27334358d963SJosef Bacik */ 27344358d963SJosef Bacik trace_btrfs_add_block_group(fs_info, cache, 1); 2735723de71dSJosef Bacik btrfs_add_bg_to_space_info(fs_info, cache); 27364358d963SJosef Bacik btrfs_update_global_block_rsv(fs_info); 27374358d963SJosef Bacik 27389d4b0a12SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 27399d4b0a12SJosef Bacik if (btrfs_should_fragment_free_space(cache)) { 2740*5758d1bdSFilipe Manana cache->space_info->bytes_used += size >> 1; 27419d4b0a12SJosef Bacik fragment_free_space(cache); 27429d4b0a12SJosef Bacik } 27439d4b0a12SJosef Bacik #endif 27444358d963SJosef Bacik 27454358d963SJosef Bacik list_add_tail(&cache->bg_list, &trans->new_bgs); 27464358d963SJosef Bacik trans->delayed_ref_updates++; 27474358d963SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 27484358d963SJosef Bacik 27494358d963SJosef Bacik set_avail_alloc_bits(fs_info, type); 275079bd3712SFilipe Manana return cache; 27514358d963SJosef Bacik } 275226ce2095SJosef Bacik 2753b12de528SQu Wenruo /* 2754b12de528SQu Wenruo * Mark one block group RO, can be called several times for the same block 2755b12de528SQu Wenruo * group. 2756b12de528SQu Wenruo * 2757b12de528SQu Wenruo * @cache: the destination block group 2758b12de528SQu Wenruo * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to 2759b12de528SQu Wenruo * ensure we still have some free space after marking this 2760b12de528SQu Wenruo * block group RO. 2761b12de528SQu Wenruo */ 2762b12de528SQu Wenruo int btrfs_inc_block_group_ro(struct btrfs_block_group *cache, 2763b12de528SQu Wenruo bool do_chunk_alloc) 276426ce2095SJosef Bacik { 276526ce2095SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 276626ce2095SJosef Bacik struct btrfs_trans_handle *trans; 2767dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 276826ce2095SJosef Bacik u64 alloc_flags; 276926ce2095SJosef Bacik int ret; 2770b6e9f16cSNikolay Borisov bool dirty_bg_running; 277126ce2095SJosef Bacik 27722d192fc4SQu Wenruo /* 27732d192fc4SQu Wenruo * This can only happen when we are doing read-only scrub on read-only 27742d192fc4SQu Wenruo * mount. 27752d192fc4SQu Wenruo * In that case we should not start a new transaction on read-only fs. 27762d192fc4SQu Wenruo * Thus here we skip all chunk allocations. 27772d192fc4SQu Wenruo */ 27782d192fc4SQu Wenruo if (sb_rdonly(fs_info->sb)) { 27792d192fc4SQu Wenruo mutex_lock(&fs_info->ro_block_group_mutex); 27802d192fc4SQu Wenruo ret = inc_block_group_ro(cache, 0); 27812d192fc4SQu Wenruo mutex_unlock(&fs_info->ro_block_group_mutex); 27822d192fc4SQu Wenruo return ret; 27832d192fc4SQu Wenruo } 27842d192fc4SQu Wenruo 2785b6e9f16cSNikolay Borisov do { 2786dfe8aec4SJosef Bacik trans = btrfs_join_transaction(root); 278726ce2095SJosef Bacik if (IS_ERR(trans)) 278826ce2095SJosef Bacik return PTR_ERR(trans); 278926ce2095SJosef Bacik 2790b6e9f16cSNikolay Borisov dirty_bg_running = false; 2791b6e9f16cSNikolay Borisov 279226ce2095SJosef Bacik /* 2793b6e9f16cSNikolay Borisov * We're not allowed to set block groups readonly after the dirty 2794b6e9f16cSNikolay Borisov * block group cache has started writing. If it already started, 2795b6e9f16cSNikolay Borisov * back off and let this transaction commit. 279626ce2095SJosef Bacik */ 279726ce2095SJosef Bacik mutex_lock(&fs_info->ro_block_group_mutex); 279826ce2095SJosef Bacik if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { 279926ce2095SJosef Bacik u64 transid = trans->transid; 280026ce2095SJosef Bacik 280126ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 280226ce2095SJosef Bacik btrfs_end_transaction(trans); 280326ce2095SJosef Bacik 280426ce2095SJosef Bacik ret = btrfs_wait_for_commit(fs_info, transid); 280526ce2095SJosef Bacik if (ret) 280626ce2095SJosef Bacik return ret; 2807b6e9f16cSNikolay Borisov dirty_bg_running = true; 280826ce2095SJosef Bacik } 2809b6e9f16cSNikolay Borisov } while (dirty_bg_running); 281026ce2095SJosef Bacik 2811b12de528SQu Wenruo if (do_chunk_alloc) { 281226ce2095SJosef Bacik /* 2813b12de528SQu Wenruo * If we are changing raid levels, try to allocate a 2814b12de528SQu Wenruo * corresponding block group with the new raid level. 281526ce2095SJosef Bacik */ 2816349e120eSJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); 281726ce2095SJosef Bacik if (alloc_flags != cache->flags) { 2818b12de528SQu Wenruo ret = btrfs_chunk_alloc(trans, alloc_flags, 2819b12de528SQu Wenruo CHUNK_ALLOC_FORCE); 282026ce2095SJosef Bacik /* 282126ce2095SJosef Bacik * ENOSPC is allowed here, we may have enough space 2822b12de528SQu Wenruo * already allocated at the new raid level to carry on 282326ce2095SJosef Bacik */ 282426ce2095SJosef Bacik if (ret == -ENOSPC) 282526ce2095SJosef Bacik ret = 0; 282626ce2095SJosef Bacik if (ret < 0) 282726ce2095SJosef Bacik goto out; 282826ce2095SJosef Bacik } 2829b12de528SQu Wenruo } 283026ce2095SJosef Bacik 2831a7a63accSJosef Bacik ret = inc_block_group_ro(cache, 0); 2832195a49eaSFilipe Manana if (!do_chunk_alloc || ret == -ETXTBSY) 2833b12de528SQu Wenruo goto unlock_out; 283426ce2095SJosef Bacik if (!ret) 283526ce2095SJosef Bacik goto out; 283626ce2095SJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); 283726ce2095SJosef Bacik ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 283826ce2095SJosef Bacik if (ret < 0) 283926ce2095SJosef Bacik goto out; 2840b6a98021SNaohiro Aota /* 2841b6a98021SNaohiro Aota * We have allocated a new chunk. We also need to activate that chunk to 2842b6a98021SNaohiro Aota * grant metadata tickets for zoned filesystem. 2843b6a98021SNaohiro Aota */ 2844b6a98021SNaohiro Aota ret = btrfs_zoned_activate_one_bg(fs_info, cache->space_info, true); 2845b6a98021SNaohiro Aota if (ret < 0) 2846b6a98021SNaohiro Aota goto out; 2847b6a98021SNaohiro Aota 2848e11c0406SJosef Bacik ret = inc_block_group_ro(cache, 0); 2849195a49eaSFilipe Manana if (ret == -ETXTBSY) 2850195a49eaSFilipe Manana goto unlock_out; 285126ce2095SJosef Bacik out: 285226ce2095SJosef Bacik if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { 2853349e120eSJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); 285426ce2095SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 285526ce2095SJosef Bacik check_system_chunk(trans, alloc_flags); 285626ce2095SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 285726ce2095SJosef Bacik } 2858b12de528SQu Wenruo unlock_out: 285926ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 286026ce2095SJosef Bacik 286126ce2095SJosef Bacik btrfs_end_transaction(trans); 286226ce2095SJosef Bacik return ret; 286326ce2095SJosef Bacik } 286426ce2095SJosef Bacik 286532da5386SDavid Sterba void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) 286626ce2095SJosef Bacik { 286726ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 286826ce2095SJosef Bacik u64 num_bytes; 286926ce2095SJosef Bacik 287026ce2095SJosef Bacik BUG_ON(!cache->ro); 287126ce2095SJosef Bacik 287226ce2095SJosef Bacik spin_lock(&sinfo->lock); 287326ce2095SJosef Bacik spin_lock(&cache->lock); 287426ce2095SJosef Bacik if (!--cache->ro) { 2875169e0da9SNaohiro Aota if (btrfs_is_zoned(cache->fs_info)) { 2876169e0da9SNaohiro Aota /* Migrate zone_unusable bytes back */ 287798173255SNaohiro Aota cache->zone_unusable = 287898173255SNaohiro Aota (cache->alloc_offset - cache->used) + 287998173255SNaohiro Aota (cache->length - cache->zone_capacity); 2880169e0da9SNaohiro Aota sinfo->bytes_zone_unusable += cache->zone_unusable; 2881169e0da9SNaohiro Aota sinfo->bytes_readonly -= cache->zone_unusable; 2882169e0da9SNaohiro Aota } 2883f9f28e5bSNaohiro Aota num_bytes = cache->length - cache->reserved - 2884f9f28e5bSNaohiro Aota cache->pinned - cache->bytes_super - 2885f9f28e5bSNaohiro Aota cache->zone_unusable - cache->used; 2886f9f28e5bSNaohiro Aota sinfo->bytes_readonly -= num_bytes; 288726ce2095SJosef Bacik list_del_init(&cache->ro_list); 288826ce2095SJosef Bacik } 288926ce2095SJosef Bacik spin_unlock(&cache->lock); 289026ce2095SJosef Bacik spin_unlock(&sinfo->lock); 289126ce2095SJosef Bacik } 289277745c05SJosef Bacik 28933be4d8efSQu Wenruo static int update_block_group_item(struct btrfs_trans_handle *trans, 289477745c05SJosef Bacik struct btrfs_path *path, 289532da5386SDavid Sterba struct btrfs_block_group *cache) 289677745c05SJosef Bacik { 289777745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 289877745c05SJosef Bacik int ret; 2899dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 290077745c05SJosef Bacik unsigned long bi; 290177745c05SJosef Bacik struct extent_buffer *leaf; 2902bf38be65SDavid Sterba struct btrfs_block_group_item bgi; 2903b3470b5dSDavid Sterba struct btrfs_key key; 29047248e0ceSQu Wenruo u64 old_commit_used; 29057248e0ceSQu Wenruo u64 used; 29067248e0ceSQu Wenruo 29077248e0ceSQu Wenruo /* 29087248e0ceSQu Wenruo * Block group items update can be triggered out of commit transaction 29097248e0ceSQu Wenruo * critical section, thus we need a consistent view of used bytes. 29107248e0ceSQu Wenruo * We cannot use cache->used directly outside of the spin lock, as it 29117248e0ceSQu Wenruo * may be changed. 29127248e0ceSQu Wenruo */ 29137248e0ceSQu Wenruo spin_lock(&cache->lock); 29147248e0ceSQu Wenruo old_commit_used = cache->commit_used; 29157248e0ceSQu Wenruo used = cache->used; 29167248e0ceSQu Wenruo /* No change in used bytes, can safely skip it. */ 29177248e0ceSQu Wenruo if (cache->commit_used == used) { 29187248e0ceSQu Wenruo spin_unlock(&cache->lock); 29197248e0ceSQu Wenruo return 0; 29207248e0ceSQu Wenruo } 29217248e0ceSQu Wenruo cache->commit_used = used; 29227248e0ceSQu Wenruo spin_unlock(&cache->lock); 292377745c05SJosef Bacik 2924b3470b5dSDavid Sterba key.objectid = cache->start; 2925b3470b5dSDavid Sterba key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 2926b3470b5dSDavid Sterba key.offset = cache->length; 2927b3470b5dSDavid Sterba 29283be4d8efSQu Wenruo ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 292977745c05SJosef Bacik if (ret) { 293077745c05SJosef Bacik if (ret > 0) 293177745c05SJosef Bacik ret = -ENOENT; 293277745c05SJosef Bacik goto fail; 293377745c05SJosef Bacik } 293477745c05SJosef Bacik 293577745c05SJosef Bacik leaf = path->nodes[0]; 293677745c05SJosef Bacik bi = btrfs_item_ptr_offset(leaf, path->slots[0]); 29377248e0ceSQu Wenruo btrfs_set_stack_block_group_used(&bgi, used); 2938de0dc456SDavid Sterba btrfs_set_stack_block_group_chunk_objectid(&bgi, 2939f7238e50SJosef Bacik cache->global_root_id); 2940de0dc456SDavid Sterba btrfs_set_stack_block_group_flags(&bgi, cache->flags); 2941bf38be65SDavid Sterba write_extent_buffer(leaf, &bgi, bi, sizeof(bgi)); 294277745c05SJosef Bacik btrfs_mark_buffer_dirty(leaf); 294377745c05SJosef Bacik fail: 294477745c05SJosef Bacik btrfs_release_path(path); 29457248e0ceSQu Wenruo /* We didn't update the block group item, need to revert @commit_used. */ 29467248e0ceSQu Wenruo if (ret < 0) { 29477248e0ceSQu Wenruo spin_lock(&cache->lock); 29487248e0ceSQu Wenruo cache->commit_used = old_commit_used; 29497248e0ceSQu Wenruo spin_unlock(&cache->lock); 29507248e0ceSQu Wenruo } 295177745c05SJosef Bacik return ret; 295277745c05SJosef Bacik 295377745c05SJosef Bacik } 295477745c05SJosef Bacik 295532da5386SDavid Sterba static int cache_save_setup(struct btrfs_block_group *block_group, 295677745c05SJosef Bacik struct btrfs_trans_handle *trans, 295777745c05SJosef Bacik struct btrfs_path *path) 295877745c05SJosef Bacik { 295977745c05SJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 296077745c05SJosef Bacik struct btrfs_root *root = fs_info->tree_root; 296177745c05SJosef Bacik struct inode *inode = NULL; 296277745c05SJosef Bacik struct extent_changeset *data_reserved = NULL; 296377745c05SJosef Bacik u64 alloc_hint = 0; 296477745c05SJosef Bacik int dcs = BTRFS_DC_ERROR; 29650044ae11SQu Wenruo u64 cache_size = 0; 296677745c05SJosef Bacik int retries = 0; 296777745c05SJosef Bacik int ret = 0; 296877745c05SJosef Bacik 2969af456a2cSBoris Burkov if (!btrfs_test_opt(fs_info, SPACE_CACHE)) 2970af456a2cSBoris Burkov return 0; 2971af456a2cSBoris Burkov 297277745c05SJosef Bacik /* 297377745c05SJosef Bacik * If this block group is smaller than 100 megs don't bother caching the 297477745c05SJosef Bacik * block group. 297577745c05SJosef Bacik */ 2976b3470b5dSDavid Sterba if (block_group->length < (100 * SZ_1M)) { 297777745c05SJosef Bacik spin_lock(&block_group->lock); 297877745c05SJosef Bacik block_group->disk_cache_state = BTRFS_DC_WRITTEN; 297977745c05SJosef Bacik spin_unlock(&block_group->lock); 298077745c05SJosef Bacik return 0; 298177745c05SJosef Bacik } 298277745c05SJosef Bacik 2983bf31f87fSDavid Sterba if (TRANS_ABORTED(trans)) 298477745c05SJosef Bacik return 0; 298577745c05SJosef Bacik again: 298677745c05SJosef Bacik inode = lookup_free_space_inode(block_group, path); 298777745c05SJosef Bacik if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { 298877745c05SJosef Bacik ret = PTR_ERR(inode); 298977745c05SJosef Bacik btrfs_release_path(path); 299077745c05SJosef Bacik goto out; 299177745c05SJosef Bacik } 299277745c05SJosef Bacik 299377745c05SJosef Bacik if (IS_ERR(inode)) { 299477745c05SJosef Bacik BUG_ON(retries); 299577745c05SJosef Bacik retries++; 299677745c05SJosef Bacik 299777745c05SJosef Bacik if (block_group->ro) 299877745c05SJosef Bacik goto out_free; 299977745c05SJosef Bacik 300077745c05SJosef Bacik ret = create_free_space_inode(trans, block_group, path); 300177745c05SJosef Bacik if (ret) 300277745c05SJosef Bacik goto out_free; 300377745c05SJosef Bacik goto again; 300477745c05SJosef Bacik } 300577745c05SJosef Bacik 300677745c05SJosef Bacik /* 300777745c05SJosef Bacik * We want to set the generation to 0, that way if anything goes wrong 300877745c05SJosef Bacik * from here on out we know not to trust this cache when we load up next 300977745c05SJosef Bacik * time. 301077745c05SJosef Bacik */ 301177745c05SJosef Bacik BTRFS_I(inode)->generation = 0; 30129a56fcd1SNikolay Borisov ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); 301377745c05SJosef Bacik if (ret) { 301477745c05SJosef Bacik /* 301577745c05SJosef Bacik * So theoretically we could recover from this, simply set the 301677745c05SJosef Bacik * super cache generation to 0 so we know to invalidate the 301777745c05SJosef Bacik * cache, but then we'd have to keep track of the block groups 301877745c05SJosef Bacik * that fail this way so we know we _have_ to reset this cache 301977745c05SJosef Bacik * before the next commit or risk reading stale cache. So to 302077745c05SJosef Bacik * limit our exposure to horrible edge cases lets just abort the 302177745c05SJosef Bacik * transaction, this only happens in really bad situations 302277745c05SJosef Bacik * anyway. 302377745c05SJosef Bacik */ 302477745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 302577745c05SJosef Bacik goto out_put; 302677745c05SJosef Bacik } 302777745c05SJosef Bacik WARN_ON(ret); 302877745c05SJosef Bacik 302977745c05SJosef Bacik /* We've already setup this transaction, go ahead and exit */ 303077745c05SJosef Bacik if (block_group->cache_generation == trans->transid && 303177745c05SJosef Bacik i_size_read(inode)) { 303277745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 303377745c05SJosef Bacik goto out_put; 303477745c05SJosef Bacik } 303577745c05SJosef Bacik 303677745c05SJosef Bacik if (i_size_read(inode) > 0) { 303777745c05SJosef Bacik ret = btrfs_check_trunc_cache_free_space(fs_info, 303877745c05SJosef Bacik &fs_info->global_block_rsv); 303977745c05SJosef Bacik if (ret) 304077745c05SJosef Bacik goto out_put; 304177745c05SJosef Bacik 304277745c05SJosef Bacik ret = btrfs_truncate_free_space_cache(trans, NULL, inode); 304377745c05SJosef Bacik if (ret) 304477745c05SJosef Bacik goto out_put; 304577745c05SJosef Bacik } 304677745c05SJosef Bacik 304777745c05SJosef Bacik spin_lock(&block_group->lock); 304877745c05SJosef Bacik if (block_group->cached != BTRFS_CACHE_FINISHED || 304977745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) { 305077745c05SJosef Bacik /* 305177745c05SJosef Bacik * don't bother trying to write stuff out _if_ 305277745c05SJosef Bacik * a) we're not cached, 305377745c05SJosef Bacik * b) we're with nospace_cache mount option, 305477745c05SJosef Bacik * c) we're with v2 space_cache (FREE_SPACE_TREE). 305577745c05SJosef Bacik */ 305677745c05SJosef Bacik dcs = BTRFS_DC_WRITTEN; 305777745c05SJosef Bacik spin_unlock(&block_group->lock); 305877745c05SJosef Bacik goto out_put; 305977745c05SJosef Bacik } 306077745c05SJosef Bacik spin_unlock(&block_group->lock); 306177745c05SJosef Bacik 306277745c05SJosef Bacik /* 306377745c05SJosef Bacik * We hit an ENOSPC when setting up the cache in this transaction, just 306477745c05SJosef Bacik * skip doing the setup, we've already cleared the cache so we're safe. 306577745c05SJosef Bacik */ 306677745c05SJosef Bacik if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { 306777745c05SJosef Bacik ret = -ENOSPC; 306877745c05SJosef Bacik goto out_put; 306977745c05SJosef Bacik } 307077745c05SJosef Bacik 307177745c05SJosef Bacik /* 307277745c05SJosef Bacik * Try to preallocate enough space based on how big the block group is. 307377745c05SJosef Bacik * Keep in mind this has to include any pinned space which could end up 307477745c05SJosef Bacik * taking up quite a bit since it's not folded into the other space 307577745c05SJosef Bacik * cache. 307677745c05SJosef Bacik */ 30770044ae11SQu Wenruo cache_size = div_u64(block_group->length, SZ_256M); 30780044ae11SQu Wenruo if (!cache_size) 30790044ae11SQu Wenruo cache_size = 1; 308077745c05SJosef Bacik 30810044ae11SQu Wenruo cache_size *= 16; 30820044ae11SQu Wenruo cache_size *= fs_info->sectorsize; 308377745c05SJosef Bacik 308436ea6f3eSNikolay Borisov ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0, 30851daedb1dSJosef Bacik cache_size, false); 308677745c05SJosef Bacik if (ret) 308777745c05SJosef Bacik goto out_put; 308877745c05SJosef Bacik 30890044ae11SQu Wenruo ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size, 30900044ae11SQu Wenruo cache_size, cache_size, 309177745c05SJosef Bacik &alloc_hint); 309277745c05SJosef Bacik /* 309377745c05SJosef Bacik * Our cache requires contiguous chunks so that we don't modify a bunch 309477745c05SJosef Bacik * of metadata or split extents when writing the cache out, which means 309577745c05SJosef Bacik * we can enospc if we are heavily fragmented in addition to just normal 309677745c05SJosef Bacik * out of space conditions. So if we hit this just skip setting up any 309777745c05SJosef Bacik * other block groups for this transaction, maybe we'll unpin enough 309877745c05SJosef Bacik * space the next time around. 309977745c05SJosef Bacik */ 310077745c05SJosef Bacik if (!ret) 310177745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 310277745c05SJosef Bacik else if (ret == -ENOSPC) 310377745c05SJosef Bacik set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); 310477745c05SJosef Bacik 310577745c05SJosef Bacik out_put: 310677745c05SJosef Bacik iput(inode); 310777745c05SJosef Bacik out_free: 310877745c05SJosef Bacik btrfs_release_path(path); 310977745c05SJosef Bacik out: 311077745c05SJosef Bacik spin_lock(&block_group->lock); 311177745c05SJosef Bacik if (!ret && dcs == BTRFS_DC_SETUP) 311277745c05SJosef Bacik block_group->cache_generation = trans->transid; 311377745c05SJosef Bacik block_group->disk_cache_state = dcs; 311477745c05SJosef Bacik spin_unlock(&block_group->lock); 311577745c05SJosef Bacik 311677745c05SJosef Bacik extent_changeset_free(data_reserved); 311777745c05SJosef Bacik return ret; 311877745c05SJosef Bacik } 311977745c05SJosef Bacik 312077745c05SJosef Bacik int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) 312177745c05SJosef Bacik { 312277745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 312332da5386SDavid Sterba struct btrfs_block_group *cache, *tmp; 312477745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 312577745c05SJosef Bacik struct btrfs_path *path; 312677745c05SJosef Bacik 312777745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs) || 312877745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) 312977745c05SJosef Bacik return 0; 313077745c05SJosef Bacik 313177745c05SJosef Bacik path = btrfs_alloc_path(); 313277745c05SJosef Bacik if (!path) 313377745c05SJosef Bacik return -ENOMEM; 313477745c05SJosef Bacik 313577745c05SJosef Bacik /* Could add new block groups, use _safe just in case */ 313677745c05SJosef Bacik list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, 313777745c05SJosef Bacik dirty_list) { 313877745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_CLEAR) 313977745c05SJosef Bacik cache_save_setup(cache, trans, path); 314077745c05SJosef Bacik } 314177745c05SJosef Bacik 314277745c05SJosef Bacik btrfs_free_path(path); 314377745c05SJosef Bacik return 0; 314477745c05SJosef Bacik } 314577745c05SJosef Bacik 314677745c05SJosef Bacik /* 314777745c05SJosef Bacik * Transaction commit does final block group cache writeback during a critical 314877745c05SJosef Bacik * section where nothing is allowed to change the FS. This is required in 314977745c05SJosef Bacik * order for the cache to actually match the block group, but can introduce a 315077745c05SJosef Bacik * lot of latency into the commit. 315177745c05SJosef Bacik * 315277745c05SJosef Bacik * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. 315377745c05SJosef Bacik * There's a chance we'll have to redo some of it if the block group changes 315477745c05SJosef Bacik * again during the commit, but it greatly reduces the commit latency by 315577745c05SJosef Bacik * getting rid of the easy block groups while we're still allowing others to 315677745c05SJosef Bacik * join the commit. 315777745c05SJosef Bacik */ 315877745c05SJosef Bacik int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) 315977745c05SJosef Bacik { 316077745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 316132da5386SDavid Sterba struct btrfs_block_group *cache; 316277745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 316377745c05SJosef Bacik int ret = 0; 316477745c05SJosef Bacik int should_put; 316577745c05SJosef Bacik struct btrfs_path *path = NULL; 316677745c05SJosef Bacik LIST_HEAD(dirty); 316777745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 316877745c05SJosef Bacik int loops = 0; 316977745c05SJosef Bacik 317077745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 317177745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs)) { 317277745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 317377745c05SJosef Bacik return 0; 317477745c05SJosef Bacik } 317577745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 317677745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 317777745c05SJosef Bacik 317877745c05SJosef Bacik again: 317977745c05SJosef Bacik /* Make sure all the block groups on our dirty list actually exist */ 318077745c05SJosef Bacik btrfs_create_pending_block_groups(trans); 318177745c05SJosef Bacik 318277745c05SJosef Bacik if (!path) { 318377745c05SJosef Bacik path = btrfs_alloc_path(); 3184938fcbfbSJosef Bacik if (!path) { 3185938fcbfbSJosef Bacik ret = -ENOMEM; 3186938fcbfbSJosef Bacik goto out; 3187938fcbfbSJosef Bacik } 318877745c05SJosef Bacik } 318977745c05SJosef Bacik 319077745c05SJosef Bacik /* 319177745c05SJosef Bacik * cache_write_mutex is here only to save us from balance or automatic 319277745c05SJosef Bacik * removal of empty block groups deleting this block group while we are 319377745c05SJosef Bacik * writing out the cache 319477745c05SJosef Bacik */ 319577745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 319677745c05SJosef Bacik while (!list_empty(&dirty)) { 319777745c05SJosef Bacik bool drop_reserve = true; 319877745c05SJosef Bacik 319932da5386SDavid Sterba cache = list_first_entry(&dirty, struct btrfs_block_group, 320077745c05SJosef Bacik dirty_list); 320177745c05SJosef Bacik /* 320277745c05SJosef Bacik * This can happen if something re-dirties a block group that 320377745c05SJosef Bacik * is already under IO. Just wait for it to finish and then do 320477745c05SJosef Bacik * it all again 320577745c05SJosef Bacik */ 320677745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 320777745c05SJosef Bacik list_del_init(&cache->io_list); 320877745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 320977745c05SJosef Bacik btrfs_put_block_group(cache); 321077745c05SJosef Bacik } 321177745c05SJosef Bacik 321277745c05SJosef Bacik 321377745c05SJosef Bacik /* 321477745c05SJosef Bacik * btrfs_wait_cache_io uses the cache->dirty_list to decide if 321577745c05SJosef Bacik * it should update the cache_state. Don't delete until after 321677745c05SJosef Bacik * we wait. 321777745c05SJosef Bacik * 321877745c05SJosef Bacik * Since we're not running in the commit critical section 321977745c05SJosef Bacik * we need the dirty_bgs_lock to protect from update_block_group 322077745c05SJosef Bacik */ 322177745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 322277745c05SJosef Bacik list_del_init(&cache->dirty_list); 322377745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 322477745c05SJosef Bacik 322577745c05SJosef Bacik should_put = 1; 322677745c05SJosef Bacik 322777745c05SJosef Bacik cache_save_setup(cache, trans, path); 322877745c05SJosef Bacik 322977745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_SETUP) { 323077745c05SJosef Bacik cache->io_ctl.inode = NULL; 323177745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 323277745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 323377745c05SJosef Bacik should_put = 0; 323477745c05SJosef Bacik 323577745c05SJosef Bacik /* 323677745c05SJosef Bacik * The cache_write_mutex is protecting the 323777745c05SJosef Bacik * io_list, also refer to the definition of 323877745c05SJosef Bacik * btrfs_transaction::io_bgs for more details 323977745c05SJosef Bacik */ 324077745c05SJosef Bacik list_add_tail(&cache->io_list, io); 324177745c05SJosef Bacik } else { 324277745c05SJosef Bacik /* 324377745c05SJosef Bacik * If we failed to write the cache, the 324477745c05SJosef Bacik * generation will be bad and life goes on 324577745c05SJosef Bacik */ 324677745c05SJosef Bacik ret = 0; 324777745c05SJosef Bacik } 324877745c05SJosef Bacik } 324977745c05SJosef Bacik if (!ret) { 32503be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 325177745c05SJosef Bacik /* 325277745c05SJosef Bacik * Our block group might still be attached to the list 325377745c05SJosef Bacik * of new block groups in the transaction handle of some 325477745c05SJosef Bacik * other task (struct btrfs_trans_handle->new_bgs). This 325577745c05SJosef Bacik * means its block group item isn't yet in the extent 325677745c05SJosef Bacik * tree. If this happens ignore the error, as we will 325777745c05SJosef Bacik * try again later in the critical section of the 325877745c05SJosef Bacik * transaction commit. 325977745c05SJosef Bacik */ 326077745c05SJosef Bacik if (ret == -ENOENT) { 326177745c05SJosef Bacik ret = 0; 326277745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 326377745c05SJosef Bacik if (list_empty(&cache->dirty_list)) { 326477745c05SJosef Bacik list_add_tail(&cache->dirty_list, 326577745c05SJosef Bacik &cur_trans->dirty_bgs); 326677745c05SJosef Bacik btrfs_get_block_group(cache); 326777745c05SJosef Bacik drop_reserve = false; 326877745c05SJosef Bacik } 326977745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 327077745c05SJosef Bacik } else if (ret) { 327177745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 327277745c05SJosef Bacik } 327377745c05SJosef Bacik } 327477745c05SJosef Bacik 327577745c05SJosef Bacik /* If it's not on the io list, we need to put the block group */ 327677745c05SJosef Bacik if (should_put) 327777745c05SJosef Bacik btrfs_put_block_group(cache); 327877745c05SJosef Bacik if (drop_reserve) 327977745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 328077745c05SJosef Bacik /* 328177745c05SJosef Bacik * Avoid blocking other tasks for too long. It might even save 328277745c05SJosef Bacik * us from writing caches for block groups that are going to be 328377745c05SJosef Bacik * removed. 328477745c05SJosef Bacik */ 328577745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 3286938fcbfbSJosef Bacik if (ret) 3287938fcbfbSJosef Bacik goto out; 328877745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 328977745c05SJosef Bacik } 329077745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 329177745c05SJosef Bacik 329277745c05SJosef Bacik /* 329377745c05SJosef Bacik * Go through delayed refs for all the stuff we've just kicked off 329477745c05SJosef Bacik * and then loop back (just once) 329577745c05SJosef Bacik */ 329634d1eb0eSJosef Bacik if (!ret) 329777745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 0); 329877745c05SJosef Bacik if (!ret && loops == 0) { 329977745c05SJosef Bacik loops++; 330077745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 330177745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 330277745c05SJosef Bacik /* 330377745c05SJosef Bacik * dirty_bgs_lock protects us from concurrent block group 330477745c05SJosef Bacik * deletes too (not just cache_write_mutex). 330577745c05SJosef Bacik */ 330677745c05SJosef Bacik if (!list_empty(&dirty)) { 330777745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 330877745c05SJosef Bacik goto again; 330977745c05SJosef Bacik } 331077745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 3311938fcbfbSJosef Bacik } 3312938fcbfbSJosef Bacik out: 3313938fcbfbSJosef Bacik if (ret < 0) { 3314938fcbfbSJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 3315938fcbfbSJosef Bacik list_splice_init(&dirty, &cur_trans->dirty_bgs); 3316938fcbfbSJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 331777745c05SJosef Bacik btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 331877745c05SJosef Bacik } 331977745c05SJosef Bacik 332077745c05SJosef Bacik btrfs_free_path(path); 332177745c05SJosef Bacik return ret; 332277745c05SJosef Bacik } 332377745c05SJosef Bacik 332477745c05SJosef Bacik int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) 332577745c05SJosef Bacik { 332677745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 332732da5386SDavid Sterba struct btrfs_block_group *cache; 332877745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 332977745c05SJosef Bacik int ret = 0; 333077745c05SJosef Bacik int should_put; 333177745c05SJosef Bacik struct btrfs_path *path; 333277745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 333377745c05SJosef Bacik 333477745c05SJosef Bacik path = btrfs_alloc_path(); 333577745c05SJosef Bacik if (!path) 333677745c05SJosef Bacik return -ENOMEM; 333777745c05SJosef Bacik 333877745c05SJosef Bacik /* 333977745c05SJosef Bacik * Even though we are in the critical section of the transaction commit, 334077745c05SJosef Bacik * we can still have concurrent tasks adding elements to this 334177745c05SJosef Bacik * transaction's list of dirty block groups. These tasks correspond to 334277745c05SJosef Bacik * endio free space workers started when writeback finishes for a 334377745c05SJosef Bacik * space cache, which run inode.c:btrfs_finish_ordered_io(), and can 334477745c05SJosef Bacik * allocate new block groups as a result of COWing nodes of the root 334577745c05SJosef Bacik * tree when updating the free space inode. The writeback for the space 334677745c05SJosef Bacik * caches is triggered by an earlier call to 334777745c05SJosef Bacik * btrfs_start_dirty_block_groups() and iterations of the following 334877745c05SJosef Bacik * loop. 334977745c05SJosef Bacik * Also we want to do the cache_save_setup first and then run the 335077745c05SJosef Bacik * delayed refs to make sure we have the best chance at doing this all 335177745c05SJosef Bacik * in one shot. 335277745c05SJosef Bacik */ 335377745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 335477745c05SJosef Bacik while (!list_empty(&cur_trans->dirty_bgs)) { 335577745c05SJosef Bacik cache = list_first_entry(&cur_trans->dirty_bgs, 335632da5386SDavid Sterba struct btrfs_block_group, 335777745c05SJosef Bacik dirty_list); 335877745c05SJosef Bacik 335977745c05SJosef Bacik /* 336077745c05SJosef Bacik * This can happen if cache_save_setup re-dirties a block group 336177745c05SJosef Bacik * that is already under IO. Just wait for it to finish and 336277745c05SJosef Bacik * then do it all again 336377745c05SJosef Bacik */ 336477745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 336577745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 336677745c05SJosef Bacik list_del_init(&cache->io_list); 336777745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 336877745c05SJosef Bacik btrfs_put_block_group(cache); 336977745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 337077745c05SJosef Bacik } 337177745c05SJosef Bacik 337277745c05SJosef Bacik /* 337377745c05SJosef Bacik * Don't remove from the dirty list until after we've waited on 337477745c05SJosef Bacik * any pending IO 337577745c05SJosef Bacik */ 337677745c05SJosef Bacik list_del_init(&cache->dirty_list); 337777745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 337877745c05SJosef Bacik should_put = 1; 337977745c05SJosef Bacik 338077745c05SJosef Bacik cache_save_setup(cache, trans, path); 338177745c05SJosef Bacik 338277745c05SJosef Bacik if (!ret) 338377745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 338477745c05SJosef Bacik (unsigned long) -1); 338577745c05SJosef Bacik 338677745c05SJosef Bacik if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { 338777745c05SJosef Bacik cache->io_ctl.inode = NULL; 338877745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 338977745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 339077745c05SJosef Bacik should_put = 0; 339177745c05SJosef Bacik list_add_tail(&cache->io_list, io); 339277745c05SJosef Bacik } else { 339377745c05SJosef Bacik /* 339477745c05SJosef Bacik * If we failed to write the cache, the 339577745c05SJosef Bacik * generation will be bad and life goes on 339677745c05SJosef Bacik */ 339777745c05SJosef Bacik ret = 0; 339877745c05SJosef Bacik } 339977745c05SJosef Bacik } 340077745c05SJosef Bacik if (!ret) { 34013be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 340277745c05SJosef Bacik /* 340377745c05SJosef Bacik * One of the free space endio workers might have 340477745c05SJosef Bacik * created a new block group while updating a free space 340577745c05SJosef Bacik * cache's inode (at inode.c:btrfs_finish_ordered_io()) 340677745c05SJosef Bacik * and hasn't released its transaction handle yet, in 340777745c05SJosef Bacik * which case the new block group is still attached to 340877745c05SJosef Bacik * its transaction handle and its creation has not 340977745c05SJosef Bacik * finished yet (no block group item in the extent tree 341077745c05SJosef Bacik * yet, etc). If this is the case, wait for all free 341177745c05SJosef Bacik * space endio workers to finish and retry. This is a 3412260db43cSRandy Dunlap * very rare case so no need for a more efficient and 341377745c05SJosef Bacik * complex approach. 341477745c05SJosef Bacik */ 341577745c05SJosef Bacik if (ret == -ENOENT) { 341677745c05SJosef Bacik wait_event(cur_trans->writer_wait, 341777745c05SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 34183be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 341977745c05SJosef Bacik } 342077745c05SJosef Bacik if (ret) 342177745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 342277745c05SJosef Bacik } 342377745c05SJosef Bacik 342477745c05SJosef Bacik /* If its not on the io list, we need to put the block group */ 342577745c05SJosef Bacik if (should_put) 342677745c05SJosef Bacik btrfs_put_block_group(cache); 342777745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 342877745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 342977745c05SJosef Bacik } 343077745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 343177745c05SJosef Bacik 343277745c05SJosef Bacik /* 343377745c05SJosef Bacik * Refer to the definition of io_bgs member for details why it's safe 343477745c05SJosef Bacik * to use it without any locking 343577745c05SJosef Bacik */ 343677745c05SJosef Bacik while (!list_empty(io)) { 343732da5386SDavid Sterba cache = list_first_entry(io, struct btrfs_block_group, 343877745c05SJosef Bacik io_list); 343977745c05SJosef Bacik list_del_init(&cache->io_list); 344077745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 344177745c05SJosef Bacik btrfs_put_block_group(cache); 344277745c05SJosef Bacik } 344377745c05SJosef Bacik 344477745c05SJosef Bacik btrfs_free_path(path); 344577745c05SJosef Bacik return ret; 344677745c05SJosef Bacik } 3447606d1bf1SJosef Bacik 3448606d1bf1SJosef Bacik int btrfs_update_block_group(struct btrfs_trans_handle *trans, 344911b66fa6SAnand Jain u64 bytenr, u64 num_bytes, bool alloc) 3450606d1bf1SJosef Bacik { 3451606d1bf1SJosef Bacik struct btrfs_fs_info *info = trans->fs_info; 345232da5386SDavid Sterba struct btrfs_block_group *cache = NULL; 3453606d1bf1SJosef Bacik u64 total = num_bytes; 3454606d1bf1SJosef Bacik u64 old_val; 3455606d1bf1SJosef Bacik u64 byte_in_group; 3456606d1bf1SJosef Bacik int factor; 3457606d1bf1SJosef Bacik int ret = 0; 3458606d1bf1SJosef Bacik 3459606d1bf1SJosef Bacik /* Block accounting for super block */ 3460606d1bf1SJosef Bacik spin_lock(&info->delalloc_root_lock); 3461606d1bf1SJosef Bacik old_val = btrfs_super_bytes_used(info->super_copy); 3462606d1bf1SJosef Bacik if (alloc) 3463606d1bf1SJosef Bacik old_val += num_bytes; 3464606d1bf1SJosef Bacik else 3465606d1bf1SJosef Bacik old_val -= num_bytes; 3466606d1bf1SJosef Bacik btrfs_set_super_bytes_used(info->super_copy, old_val); 3467606d1bf1SJosef Bacik spin_unlock(&info->delalloc_root_lock); 3468606d1bf1SJosef Bacik 3469606d1bf1SJosef Bacik while (total) { 3470df384da5SJosef Bacik struct btrfs_space_info *space_info; 3471efbf35a1SJosef Bacik bool reclaim = false; 3472ac2f1e63SJosef Bacik 3473606d1bf1SJosef Bacik cache = btrfs_lookup_block_group(info, bytenr); 3474606d1bf1SJosef Bacik if (!cache) { 3475606d1bf1SJosef Bacik ret = -ENOENT; 3476606d1bf1SJosef Bacik break; 3477606d1bf1SJosef Bacik } 3478df384da5SJosef Bacik space_info = cache->space_info; 3479606d1bf1SJosef Bacik factor = btrfs_bg_type_to_factor(cache->flags); 3480606d1bf1SJosef Bacik 3481606d1bf1SJosef Bacik /* 3482606d1bf1SJosef Bacik * If this block group has free space cache written out, we 3483606d1bf1SJosef Bacik * need to make sure to load it if we are removing space. This 3484606d1bf1SJosef Bacik * is because we need the unpinning stage to actually add the 3485606d1bf1SJosef Bacik * space back to the block group, otherwise we will leak space. 3486606d1bf1SJosef Bacik */ 348732da5386SDavid Sterba if (!alloc && !btrfs_block_group_done(cache)) 3488ced8ecf0SOmar Sandoval btrfs_cache_block_group(cache, true); 3489606d1bf1SJosef Bacik 3490b3470b5dSDavid Sterba byte_in_group = bytenr - cache->start; 3491b3470b5dSDavid Sterba WARN_ON(byte_in_group > cache->length); 3492606d1bf1SJosef Bacik 3493df384da5SJosef Bacik spin_lock(&space_info->lock); 3494606d1bf1SJosef Bacik spin_lock(&cache->lock); 3495606d1bf1SJosef Bacik 3496606d1bf1SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 3497606d1bf1SJosef Bacik cache->disk_cache_state < BTRFS_DC_CLEAR) 3498606d1bf1SJosef Bacik cache->disk_cache_state = BTRFS_DC_CLEAR; 3499606d1bf1SJosef Bacik 3500bf38be65SDavid Sterba old_val = cache->used; 3501b3470b5dSDavid Sterba num_bytes = min(total, cache->length - byte_in_group); 3502606d1bf1SJosef Bacik if (alloc) { 3503606d1bf1SJosef Bacik old_val += num_bytes; 3504bf38be65SDavid Sterba cache->used = old_val; 3505606d1bf1SJosef Bacik cache->reserved -= num_bytes; 3506df384da5SJosef Bacik space_info->bytes_reserved -= num_bytes; 3507df384da5SJosef Bacik space_info->bytes_used += num_bytes; 3508df384da5SJosef Bacik space_info->disk_used += num_bytes * factor; 3509606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3510df384da5SJosef Bacik spin_unlock(&space_info->lock); 3511606d1bf1SJosef Bacik } else { 3512606d1bf1SJosef Bacik old_val -= num_bytes; 3513bf38be65SDavid Sterba cache->used = old_val; 3514606d1bf1SJosef Bacik cache->pinned += num_bytes; 3515df384da5SJosef Bacik btrfs_space_info_update_bytes_pinned(info, space_info, 3516df384da5SJosef Bacik num_bytes); 3517df384da5SJosef Bacik space_info->bytes_used -= num_bytes; 3518df384da5SJosef Bacik space_info->disk_used -= num_bytes * factor; 3519ac2f1e63SJosef Bacik 3520ac2f1e63SJosef Bacik reclaim = should_reclaim_block_group(cache, num_bytes); 352152bb7a21SBoris Burkov 3522606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3523df384da5SJosef Bacik spin_unlock(&space_info->lock); 3524606d1bf1SJosef Bacik 3525fe119a6eSNikolay Borisov set_extent_dirty(&trans->transaction->pinned_extents, 3526606d1bf1SJosef Bacik bytenr, bytenr + num_bytes - 1, 3527606d1bf1SJosef Bacik GFP_NOFS | __GFP_NOFAIL); 3528606d1bf1SJosef Bacik } 3529606d1bf1SJosef Bacik 3530606d1bf1SJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 3531606d1bf1SJosef Bacik if (list_empty(&cache->dirty_list)) { 3532606d1bf1SJosef Bacik list_add_tail(&cache->dirty_list, 3533606d1bf1SJosef Bacik &trans->transaction->dirty_bgs); 3534606d1bf1SJosef Bacik trans->delayed_ref_updates++; 3535606d1bf1SJosef Bacik btrfs_get_block_group(cache); 3536606d1bf1SJosef Bacik } 3537606d1bf1SJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 3538606d1bf1SJosef Bacik 3539606d1bf1SJosef Bacik /* 3540606d1bf1SJosef Bacik * No longer have used bytes in this block group, queue it for 3541606d1bf1SJosef Bacik * deletion. We do this after adding the block group to the 3542606d1bf1SJosef Bacik * dirty list to avoid races between cleaner kthread and space 3543606d1bf1SJosef Bacik * cache writeout. 3544606d1bf1SJosef Bacik */ 35456e80d4f8SDennis Zhou if (!alloc && old_val == 0) { 35466e80d4f8SDennis Zhou if (!btrfs_test_opt(info, DISCARD_ASYNC)) 3547606d1bf1SJosef Bacik btrfs_mark_bg_unused(cache); 3548ac2f1e63SJosef Bacik } else if (!alloc && reclaim) { 3549ac2f1e63SJosef Bacik btrfs_mark_bg_to_reclaim(cache); 35506e80d4f8SDennis Zhou } 3551606d1bf1SJosef Bacik 3552606d1bf1SJosef Bacik btrfs_put_block_group(cache); 3553606d1bf1SJosef Bacik total -= num_bytes; 3554606d1bf1SJosef Bacik bytenr += num_bytes; 3555606d1bf1SJosef Bacik } 3556606d1bf1SJosef Bacik 3557606d1bf1SJosef Bacik /* Modified block groups are accounted for in the delayed_refs_rsv. */ 3558606d1bf1SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 3559606d1bf1SJosef Bacik return ret; 3560606d1bf1SJosef Bacik } 3561606d1bf1SJosef Bacik 356243dd529aSDavid Sterba /* 356343dd529aSDavid Sterba * Update the block_group and space info counters. 356443dd529aSDavid Sterba * 3565606d1bf1SJosef Bacik * @cache: The cache we are manipulating 3566606d1bf1SJosef Bacik * @ram_bytes: The number of bytes of file content, and will be same to 3567606d1bf1SJosef Bacik * @num_bytes except for the compress path. 3568606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 3569606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 3570606d1bf1SJosef Bacik * 3571606d1bf1SJosef Bacik * This is called by the allocator when it reserves space. If this is a 3572606d1bf1SJosef Bacik * reservation and the block group has become read only we cannot make the 3573606d1bf1SJosef Bacik * reservation and return -EAGAIN, otherwise this function always succeeds. 3574606d1bf1SJosef Bacik */ 357532da5386SDavid Sterba int btrfs_add_reserved_bytes(struct btrfs_block_group *cache, 357652bb7a21SBoris Burkov u64 ram_bytes, u64 num_bytes, int delalloc, 357752bb7a21SBoris Burkov bool force_wrong_size_class) 3578606d1bf1SJosef Bacik { 3579606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 358052bb7a21SBoris Burkov enum btrfs_block_group_size_class size_class; 3581606d1bf1SJosef Bacik int ret = 0; 3582606d1bf1SJosef Bacik 3583606d1bf1SJosef Bacik spin_lock(&space_info->lock); 3584606d1bf1SJosef Bacik spin_lock(&cache->lock); 3585606d1bf1SJosef Bacik if (cache->ro) { 3586606d1bf1SJosef Bacik ret = -EAGAIN; 358752bb7a21SBoris Burkov goto out; 358852bb7a21SBoris Burkov } 358952bb7a21SBoris Burkov 3590cb0922f2SBoris Burkov if (btrfs_block_group_should_use_size_class(cache)) { 359152bb7a21SBoris Burkov size_class = btrfs_calc_block_group_size_class(num_bytes); 359252bb7a21SBoris Burkov ret = btrfs_use_block_group_size_class(cache, size_class, force_wrong_size_class); 359352bb7a21SBoris Burkov if (ret) 359452bb7a21SBoris Burkov goto out; 359552bb7a21SBoris Burkov } 3596606d1bf1SJosef Bacik cache->reserved += num_bytes; 3597606d1bf1SJosef Bacik space_info->bytes_reserved += num_bytes; 3598a43c3835SJosef Bacik trace_btrfs_space_reservation(cache->fs_info, "space_info", 3599a43c3835SJosef Bacik space_info->flags, num_bytes, 1); 3600606d1bf1SJosef Bacik btrfs_space_info_update_bytes_may_use(cache->fs_info, 3601606d1bf1SJosef Bacik space_info, -ram_bytes); 3602606d1bf1SJosef Bacik if (delalloc) 3603606d1bf1SJosef Bacik cache->delalloc_bytes += num_bytes; 360499ffb43eSJosef Bacik 360599ffb43eSJosef Bacik /* 360652bb7a21SBoris Burkov * Compression can use less space than we reserved, so wake tickets if 360752bb7a21SBoris Burkov * that happens. 360899ffb43eSJosef Bacik */ 360999ffb43eSJosef Bacik if (num_bytes < ram_bytes) 361099ffb43eSJosef Bacik btrfs_try_granting_tickets(cache->fs_info, space_info); 361152bb7a21SBoris Burkov out: 3612606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3613606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 3614606d1bf1SJosef Bacik return ret; 3615606d1bf1SJosef Bacik } 3616606d1bf1SJosef Bacik 361743dd529aSDavid Sterba /* 361843dd529aSDavid Sterba * Update the block_group and space info counters. 361943dd529aSDavid Sterba * 3620606d1bf1SJosef Bacik * @cache: The cache we are manipulating 3621606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 3622606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 3623606d1bf1SJosef Bacik * 3624606d1bf1SJosef Bacik * This is called by somebody who is freeing space that was never actually used 3625606d1bf1SJosef Bacik * on disk. For example if you reserve some space for a new leaf in transaction 3626606d1bf1SJosef Bacik * A and before transaction A commits you free that leaf, you call this with 3627606d1bf1SJosef Bacik * reserve set to 0 in order to clear the reservation. 3628606d1bf1SJosef Bacik */ 362932da5386SDavid Sterba void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, 3630606d1bf1SJosef Bacik u64 num_bytes, int delalloc) 3631606d1bf1SJosef Bacik { 3632606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 3633606d1bf1SJosef Bacik 3634606d1bf1SJosef Bacik spin_lock(&space_info->lock); 3635606d1bf1SJosef Bacik spin_lock(&cache->lock); 3636606d1bf1SJosef Bacik if (cache->ro) 3637606d1bf1SJosef Bacik space_info->bytes_readonly += num_bytes; 3638606d1bf1SJosef Bacik cache->reserved -= num_bytes; 3639606d1bf1SJosef Bacik space_info->bytes_reserved -= num_bytes; 3640606d1bf1SJosef Bacik space_info->max_extent_size = 0; 3641606d1bf1SJosef Bacik 3642606d1bf1SJosef Bacik if (delalloc) 3643606d1bf1SJosef Bacik cache->delalloc_bytes -= num_bytes; 3644606d1bf1SJosef Bacik spin_unlock(&cache->lock); 36453308234aSJosef Bacik 36463308234aSJosef Bacik btrfs_try_granting_tickets(cache->fs_info, space_info); 3647606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 3648606d1bf1SJosef Bacik } 364907730d87SJosef Bacik 365007730d87SJosef Bacik static void force_metadata_allocation(struct btrfs_fs_info *info) 365107730d87SJosef Bacik { 365207730d87SJosef Bacik struct list_head *head = &info->space_info; 365307730d87SJosef Bacik struct btrfs_space_info *found; 365407730d87SJosef Bacik 365572804905SJosef Bacik list_for_each_entry(found, head, list) { 365607730d87SJosef Bacik if (found->flags & BTRFS_BLOCK_GROUP_METADATA) 365707730d87SJosef Bacik found->force_alloc = CHUNK_ALLOC_FORCE; 365807730d87SJosef Bacik } 365907730d87SJosef Bacik } 366007730d87SJosef Bacik 366107730d87SJosef Bacik static int should_alloc_chunk(struct btrfs_fs_info *fs_info, 366207730d87SJosef Bacik struct btrfs_space_info *sinfo, int force) 366307730d87SJosef Bacik { 366407730d87SJosef Bacik u64 bytes_used = btrfs_space_info_used(sinfo, false); 366507730d87SJosef Bacik u64 thresh; 366607730d87SJosef Bacik 366707730d87SJosef Bacik if (force == CHUNK_ALLOC_FORCE) 366807730d87SJosef Bacik return 1; 366907730d87SJosef Bacik 367007730d87SJosef Bacik /* 367107730d87SJosef Bacik * in limited mode, we want to have some free space up to 367207730d87SJosef Bacik * about 1% of the FS size. 367307730d87SJosef Bacik */ 367407730d87SJosef Bacik if (force == CHUNK_ALLOC_LIMITED) { 367507730d87SJosef Bacik thresh = btrfs_super_total_bytes(fs_info->super_copy); 3676428c8e03SDavid Sterba thresh = max_t(u64, SZ_64M, mult_perc(thresh, 1)); 367707730d87SJosef Bacik 367807730d87SJosef Bacik if (sinfo->total_bytes - bytes_used < thresh) 367907730d87SJosef Bacik return 1; 368007730d87SJosef Bacik } 368107730d87SJosef Bacik 3682428c8e03SDavid Sterba if (bytes_used + SZ_2M < mult_perc(sinfo->total_bytes, 80)) 368307730d87SJosef Bacik return 0; 368407730d87SJosef Bacik return 1; 368507730d87SJosef Bacik } 368607730d87SJosef Bacik 368707730d87SJosef Bacik int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) 368807730d87SJosef Bacik { 368907730d87SJosef Bacik u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); 369007730d87SJosef Bacik 369107730d87SJosef Bacik return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 369207730d87SJosef Bacik } 369307730d87SJosef Bacik 3694820c363bSNaohiro Aota static struct btrfs_block_group *do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags) 369579bd3712SFilipe Manana { 369679bd3712SFilipe Manana struct btrfs_block_group *bg; 369779bd3712SFilipe Manana int ret; 369879bd3712SFilipe Manana 369907730d87SJosef Bacik /* 370079bd3712SFilipe Manana * Check if we have enough space in the system space info because we 370179bd3712SFilipe Manana * will need to update device items in the chunk btree and insert a new 370279bd3712SFilipe Manana * chunk item in the chunk btree as well. This will allocate a new 370379bd3712SFilipe Manana * system block group if needed. 370479bd3712SFilipe Manana */ 370579bd3712SFilipe Manana check_system_chunk(trans, flags); 370679bd3712SFilipe Manana 3707f6f39f7aSNikolay Borisov bg = btrfs_create_chunk(trans, flags); 370879bd3712SFilipe Manana if (IS_ERR(bg)) { 370979bd3712SFilipe Manana ret = PTR_ERR(bg); 371079bd3712SFilipe Manana goto out; 371179bd3712SFilipe Manana } 371279bd3712SFilipe Manana 371379bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); 371479bd3712SFilipe Manana /* 371579bd3712SFilipe Manana * Normally we are not expected to fail with -ENOSPC here, since we have 371679bd3712SFilipe Manana * previously reserved space in the system space_info and allocated one 3717ecd84d54SFilipe Manana * new system chunk if necessary. However there are three exceptions: 371879bd3712SFilipe Manana * 371979bd3712SFilipe Manana * 1) We may have enough free space in the system space_info but all the 372079bd3712SFilipe Manana * existing system block groups have a profile which can not be used 372179bd3712SFilipe Manana * for extent allocation. 372279bd3712SFilipe Manana * 372379bd3712SFilipe Manana * This happens when mounting in degraded mode. For example we have a 372479bd3712SFilipe Manana * RAID1 filesystem with 2 devices, lose one device and mount the fs 372579bd3712SFilipe Manana * using the other device in degraded mode. If we then allocate a chunk, 372679bd3712SFilipe Manana * we may have enough free space in the existing system space_info, but 372779bd3712SFilipe Manana * none of the block groups can be used for extent allocation since they 372879bd3712SFilipe Manana * have a RAID1 profile, and because we are in degraded mode with a 372979bd3712SFilipe Manana * single device, we are forced to allocate a new system chunk with a 373079bd3712SFilipe Manana * SINGLE profile. Making check_system_chunk() iterate over all system 373179bd3712SFilipe Manana * block groups and check if they have a usable profile and enough space 373279bd3712SFilipe Manana * can be slow on very large filesystems, so we tolerate the -ENOSPC and 373379bd3712SFilipe Manana * try again after forcing allocation of a new system chunk. Like this 373479bd3712SFilipe Manana * we avoid paying the cost of that search in normal circumstances, when 373579bd3712SFilipe Manana * we were not mounted in degraded mode; 373679bd3712SFilipe Manana * 373779bd3712SFilipe Manana * 2) We had enough free space info the system space_info, and one suitable 373879bd3712SFilipe Manana * block group to allocate from when we called check_system_chunk() 373979bd3712SFilipe Manana * above. However right after we called it, the only system block group 374079bd3712SFilipe Manana * with enough free space got turned into RO mode by a running scrub, 374179bd3712SFilipe Manana * and in this case we have to allocate a new one and retry. We only 374279bd3712SFilipe Manana * need do this allocate and retry once, since we have a transaction 3743ecd84d54SFilipe Manana * handle and scrub uses the commit root to search for block groups; 3744ecd84d54SFilipe Manana * 3745ecd84d54SFilipe Manana * 3) We had one system block group with enough free space when we called 3746ecd84d54SFilipe Manana * check_system_chunk(), but after that, right before we tried to 3747ecd84d54SFilipe Manana * allocate the last extent buffer we needed, a discard operation came 3748ecd84d54SFilipe Manana * in and it temporarily removed the last free space entry from the 3749ecd84d54SFilipe Manana * block group (discard removes a free space entry, discards it, and 3750ecd84d54SFilipe Manana * then adds back the entry to the block group cache). 375179bd3712SFilipe Manana */ 375279bd3712SFilipe Manana if (ret == -ENOSPC) { 375379bd3712SFilipe Manana const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info); 375479bd3712SFilipe Manana struct btrfs_block_group *sys_bg; 375579bd3712SFilipe Manana 3756f6f39f7aSNikolay Borisov sys_bg = btrfs_create_chunk(trans, sys_flags); 375779bd3712SFilipe Manana if (IS_ERR(sys_bg)) { 375879bd3712SFilipe Manana ret = PTR_ERR(sys_bg); 375979bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 376079bd3712SFilipe Manana goto out; 376179bd3712SFilipe Manana } 376279bd3712SFilipe Manana 376379bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 376479bd3712SFilipe Manana if (ret) { 376579bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 376679bd3712SFilipe Manana goto out; 376779bd3712SFilipe Manana } 376879bd3712SFilipe Manana 376979bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); 377079bd3712SFilipe Manana if (ret) { 377179bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 377279bd3712SFilipe Manana goto out; 377379bd3712SFilipe Manana } 377479bd3712SFilipe Manana } else if (ret) { 377579bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 377679bd3712SFilipe Manana goto out; 377779bd3712SFilipe Manana } 377879bd3712SFilipe Manana out: 377979bd3712SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 378079bd3712SFilipe Manana 3781820c363bSNaohiro Aota if (ret) 3782820c363bSNaohiro Aota return ERR_PTR(ret); 3783820c363bSNaohiro Aota 3784820c363bSNaohiro Aota btrfs_get_block_group(bg); 3785820c363bSNaohiro Aota return bg; 378679bd3712SFilipe Manana } 378779bd3712SFilipe Manana 378879bd3712SFilipe Manana /* 378979bd3712SFilipe Manana * Chunk allocation is done in 2 phases: 379079bd3712SFilipe Manana * 379179bd3712SFilipe Manana * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for 379279bd3712SFilipe Manana * the chunk, the chunk mapping, create its block group and add the items 379379bd3712SFilipe Manana * that belong in the chunk btree to it - more specifically, we need to 379479bd3712SFilipe Manana * update device items in the chunk btree and add a new chunk item to it. 379579bd3712SFilipe Manana * 379679bd3712SFilipe Manana * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block 379779bd3712SFilipe Manana * group item to the extent btree and the device extent items to the devices 379879bd3712SFilipe Manana * btree. 379979bd3712SFilipe Manana * 380079bd3712SFilipe Manana * This is done to prevent deadlocks. For example when COWing a node from the 380179bd3712SFilipe Manana * extent btree we are holding a write lock on the node's parent and if we 380279bd3712SFilipe Manana * trigger chunk allocation and attempted to insert the new block group item 380379bd3712SFilipe Manana * in the extent btree right way, we could deadlock because the path for the 380479bd3712SFilipe Manana * insertion can include that parent node. At first glance it seems impossible 380579bd3712SFilipe Manana * to trigger chunk allocation after starting a transaction since tasks should 380679bd3712SFilipe Manana * reserve enough transaction units (metadata space), however while that is true 380779bd3712SFilipe Manana * most of the time, chunk allocation may still be triggered for several reasons: 380879bd3712SFilipe Manana * 380979bd3712SFilipe Manana * 1) When reserving metadata, we check if there is enough free space in the 381079bd3712SFilipe Manana * metadata space_info and therefore don't trigger allocation of a new chunk. 381179bd3712SFilipe Manana * However later when the task actually tries to COW an extent buffer from 381279bd3712SFilipe Manana * the extent btree or from the device btree for example, it is forced to 381379bd3712SFilipe Manana * allocate a new block group (chunk) because the only one that had enough 381479bd3712SFilipe Manana * free space was just turned to RO mode by a running scrub for example (or 381579bd3712SFilipe Manana * device replace, block group reclaim thread, etc), so we can not use it 381679bd3712SFilipe Manana * for allocating an extent and end up being forced to allocate a new one; 381779bd3712SFilipe Manana * 381879bd3712SFilipe Manana * 2) Because we only check that the metadata space_info has enough free bytes, 381979bd3712SFilipe Manana * we end up not allocating a new metadata chunk in that case. However if 382079bd3712SFilipe Manana * the filesystem was mounted in degraded mode, none of the existing block 382179bd3712SFilipe Manana * groups might be suitable for extent allocation due to their incompatible 382279bd3712SFilipe Manana * profile (for e.g. mounting a 2 devices filesystem, where all block groups 382379bd3712SFilipe Manana * use a RAID1 profile, in degraded mode using a single device). In this case 382479bd3712SFilipe Manana * when the task attempts to COW some extent buffer of the extent btree for 382579bd3712SFilipe Manana * example, it will trigger allocation of a new metadata block group with a 382679bd3712SFilipe Manana * suitable profile (SINGLE profile in the example of the degraded mount of 382779bd3712SFilipe Manana * the RAID1 filesystem); 382879bd3712SFilipe Manana * 382979bd3712SFilipe Manana * 3) The task has reserved enough transaction units / metadata space, but when 383079bd3712SFilipe Manana * it attempts to COW an extent buffer from the extent or device btree for 383179bd3712SFilipe Manana * example, it does not find any free extent in any metadata block group, 383279bd3712SFilipe Manana * therefore forced to try to allocate a new metadata block group. 383379bd3712SFilipe Manana * This is because some other task allocated all available extents in the 383479bd3712SFilipe Manana * meanwhile - this typically happens with tasks that don't reserve space 383579bd3712SFilipe Manana * properly, either intentionally or as a bug. One example where this is 383679bd3712SFilipe Manana * done intentionally is fsync, as it does not reserve any transaction units 383779bd3712SFilipe Manana * and ends up allocating a variable number of metadata extents for log 3838ecd84d54SFilipe Manana * tree extent buffers; 3839ecd84d54SFilipe Manana * 3840ecd84d54SFilipe Manana * 4) The task has reserved enough transaction units / metadata space, but right 3841ecd84d54SFilipe Manana * before it tries to allocate the last extent buffer it needs, a discard 3842ecd84d54SFilipe Manana * operation comes in and, temporarily, removes the last free space entry from 3843ecd84d54SFilipe Manana * the only metadata block group that had free space (discard starts by 3844ecd84d54SFilipe Manana * removing a free space entry from a block group, then does the discard 3845ecd84d54SFilipe Manana * operation and, once it's done, it adds back the free space entry to the 3846ecd84d54SFilipe Manana * block group). 384779bd3712SFilipe Manana * 384879bd3712SFilipe Manana * We also need this 2 phases setup when adding a device to a filesystem with 384979bd3712SFilipe Manana * a seed device - we must create new metadata and system chunks without adding 385079bd3712SFilipe Manana * any of the block group items to the chunk, extent and device btrees. If we 385179bd3712SFilipe Manana * did not do it this way, we would get ENOSPC when attempting to update those 385279bd3712SFilipe Manana * btrees, since all the chunks from the seed device are read-only. 385379bd3712SFilipe Manana * 385479bd3712SFilipe Manana * Phase 1 does the updates and insertions to the chunk btree because if we had 385579bd3712SFilipe Manana * it done in phase 2 and have a thundering herd of tasks allocating chunks in 385679bd3712SFilipe Manana * parallel, we risk having too many system chunks allocated by many tasks if 385779bd3712SFilipe Manana * many tasks reach phase 1 without the previous ones completing phase 2. In the 385879bd3712SFilipe Manana * extreme case this leads to exhaustion of the system chunk array in the 385979bd3712SFilipe Manana * superblock. This is easier to trigger if using a btree node/leaf size of 64K 386079bd3712SFilipe Manana * and with RAID filesystems (so we have more device items in the chunk btree). 386179bd3712SFilipe Manana * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of 386279bd3712SFilipe Manana * the system chunk array due to concurrent allocations") provides more details. 386379bd3712SFilipe Manana * 38642bb2e00eSFilipe Manana * Allocation of system chunks does not happen through this function. A task that 38652bb2e00eSFilipe Manana * needs to update the chunk btree (the only btree that uses system chunks), must 38662bb2e00eSFilipe Manana * preallocate chunk space by calling either check_system_chunk() or 38672bb2e00eSFilipe Manana * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or 38682bb2e00eSFilipe Manana * metadata chunk or when removing a chunk, while the later is used before doing 38692bb2e00eSFilipe Manana * a modification to the chunk btree - use cases for the later are adding, 38702bb2e00eSFilipe Manana * removing and resizing a device as well as relocation of a system chunk. 38712bb2e00eSFilipe Manana * See the comment below for more details. 387279bd3712SFilipe Manana * 387379bd3712SFilipe Manana * The reservation of system space, done through check_system_chunk(), as well 387479bd3712SFilipe Manana * as all the updates and insertions into the chunk btree must be done while 387579bd3712SFilipe Manana * holding fs_info->chunk_mutex. This is important to guarantee that while COWing 387679bd3712SFilipe Manana * an extent buffer from the chunks btree we never trigger allocation of a new 387779bd3712SFilipe Manana * system chunk, which would result in a deadlock (trying to lock twice an 387879bd3712SFilipe Manana * extent buffer of the chunk btree, first time before triggering the chunk 387979bd3712SFilipe Manana * allocation and the second time during chunk allocation while attempting to 388079bd3712SFilipe Manana * update the chunks btree). The system chunk array is also updated while holding 388179bd3712SFilipe Manana * that mutex. The same logic applies to removing chunks - we must reserve system 388279bd3712SFilipe Manana * space, update the chunk btree and the system chunk array in the superblock 388379bd3712SFilipe Manana * while holding fs_info->chunk_mutex. 388479bd3712SFilipe Manana * 388579bd3712SFilipe Manana * This function, btrfs_chunk_alloc(), belongs to phase 1. 388679bd3712SFilipe Manana * 388779bd3712SFilipe Manana * If @force is CHUNK_ALLOC_FORCE: 388807730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 388907730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 389079bd3712SFilipe Manana * If @force is NOT CHUNK_ALLOC_FORCE: 389107730d87SJosef Bacik * - return 0 if it doesn't need to allocate a new chunk, 389207730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 389307730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 389407730d87SJosef Bacik */ 389507730d87SJosef Bacik int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, 389607730d87SJosef Bacik enum btrfs_chunk_alloc_enum force) 389707730d87SJosef Bacik { 389807730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 389907730d87SJosef Bacik struct btrfs_space_info *space_info; 3900820c363bSNaohiro Aota struct btrfs_block_group *ret_bg; 390107730d87SJosef Bacik bool wait_for_alloc = false; 390207730d87SJosef Bacik bool should_alloc = false; 3903760e69c4SNaohiro Aota bool from_extent_allocation = false; 390407730d87SJosef Bacik int ret = 0; 390507730d87SJosef Bacik 3906760e69c4SNaohiro Aota if (force == CHUNK_ALLOC_FORCE_FOR_EXTENT) { 3907760e69c4SNaohiro Aota from_extent_allocation = true; 3908760e69c4SNaohiro Aota force = CHUNK_ALLOC_FORCE; 3909760e69c4SNaohiro Aota } 3910760e69c4SNaohiro Aota 391107730d87SJosef Bacik /* Don't re-enter if we're already allocating a chunk */ 391207730d87SJosef Bacik if (trans->allocating_chunk) 391307730d87SJosef Bacik return -ENOSPC; 391479bd3712SFilipe Manana /* 39152bb2e00eSFilipe Manana * Allocation of system chunks can not happen through this path, as we 39162bb2e00eSFilipe Manana * could end up in a deadlock if we are allocating a data or metadata 39172bb2e00eSFilipe Manana * chunk and there is another task modifying the chunk btree. 39182bb2e00eSFilipe Manana * 39192bb2e00eSFilipe Manana * This is because while we are holding the chunk mutex, we will attempt 39202bb2e00eSFilipe Manana * to add the new chunk item to the chunk btree or update an existing 39212bb2e00eSFilipe Manana * device item in the chunk btree, while the other task that is modifying 39222bb2e00eSFilipe Manana * the chunk btree is attempting to COW an extent buffer while holding a 39232bb2e00eSFilipe Manana * lock on it and on its parent - if the COW operation triggers a system 39242bb2e00eSFilipe Manana * chunk allocation, then we can deadlock because we are holding the 39252bb2e00eSFilipe Manana * chunk mutex and we may need to access that extent buffer or its parent 39262bb2e00eSFilipe Manana * in order to add the chunk item or update a device item. 39272bb2e00eSFilipe Manana * 39282bb2e00eSFilipe Manana * Tasks that want to modify the chunk tree should reserve system space 39292bb2e00eSFilipe Manana * before updating the chunk btree, by calling either 39302bb2e00eSFilipe Manana * btrfs_reserve_chunk_metadata() or check_system_chunk(). 39312bb2e00eSFilipe Manana * It's possible that after a task reserves the space, it still ends up 39322bb2e00eSFilipe Manana * here - this happens in the cases described above at do_chunk_alloc(). 39332bb2e00eSFilipe Manana * The task will have to either retry or fail. 393479bd3712SFilipe Manana */ 39352bb2e00eSFilipe Manana if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 393679bd3712SFilipe Manana return -ENOSPC; 393707730d87SJosef Bacik 393807730d87SJosef Bacik space_info = btrfs_find_space_info(fs_info, flags); 393907730d87SJosef Bacik ASSERT(space_info); 394007730d87SJosef Bacik 394107730d87SJosef Bacik do { 394207730d87SJosef Bacik spin_lock(&space_info->lock); 394307730d87SJosef Bacik if (force < space_info->force_alloc) 394407730d87SJosef Bacik force = space_info->force_alloc; 394507730d87SJosef Bacik should_alloc = should_alloc_chunk(fs_info, space_info, force); 394607730d87SJosef Bacik if (space_info->full) { 394707730d87SJosef Bacik /* No more free physical space */ 394807730d87SJosef Bacik if (should_alloc) 394907730d87SJosef Bacik ret = -ENOSPC; 395007730d87SJosef Bacik else 395107730d87SJosef Bacik ret = 0; 395207730d87SJosef Bacik spin_unlock(&space_info->lock); 395307730d87SJosef Bacik return ret; 395407730d87SJosef Bacik } else if (!should_alloc) { 395507730d87SJosef Bacik spin_unlock(&space_info->lock); 395607730d87SJosef Bacik return 0; 395707730d87SJosef Bacik } else if (space_info->chunk_alloc) { 395807730d87SJosef Bacik /* 395907730d87SJosef Bacik * Someone is already allocating, so we need to block 396007730d87SJosef Bacik * until this someone is finished and then loop to 396107730d87SJosef Bacik * recheck if we should continue with our allocation 396207730d87SJosef Bacik * attempt. 396307730d87SJosef Bacik */ 396407730d87SJosef Bacik wait_for_alloc = true; 39651314ca78SJosef Bacik force = CHUNK_ALLOC_NO_FORCE; 396607730d87SJosef Bacik spin_unlock(&space_info->lock); 396707730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 396807730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 396907730d87SJosef Bacik } else { 397007730d87SJosef Bacik /* Proceed with allocation */ 397107730d87SJosef Bacik space_info->chunk_alloc = 1; 397207730d87SJosef Bacik wait_for_alloc = false; 397307730d87SJosef Bacik spin_unlock(&space_info->lock); 397407730d87SJosef Bacik } 397507730d87SJosef Bacik 397607730d87SJosef Bacik cond_resched(); 397707730d87SJosef Bacik } while (wait_for_alloc); 397807730d87SJosef Bacik 397907730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 398007730d87SJosef Bacik trans->allocating_chunk = true; 398107730d87SJosef Bacik 398207730d87SJosef Bacik /* 398307730d87SJosef Bacik * If we have mixed data/metadata chunks we want to make sure we keep 398407730d87SJosef Bacik * allocating mixed chunks instead of individual chunks. 398507730d87SJosef Bacik */ 398607730d87SJosef Bacik if (btrfs_mixed_space_info(space_info)) 398707730d87SJosef Bacik flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); 398807730d87SJosef Bacik 398907730d87SJosef Bacik /* 399007730d87SJosef Bacik * if we're doing a data chunk, go ahead and make sure that 399107730d87SJosef Bacik * we keep a reasonable number of metadata chunks allocated in the 399207730d87SJosef Bacik * FS as well. 399307730d87SJosef Bacik */ 399407730d87SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { 399507730d87SJosef Bacik fs_info->data_chunk_allocations++; 399607730d87SJosef Bacik if (!(fs_info->data_chunk_allocations % 399707730d87SJosef Bacik fs_info->metadata_ratio)) 399807730d87SJosef Bacik force_metadata_allocation(fs_info); 399907730d87SJosef Bacik } 400007730d87SJosef Bacik 4001820c363bSNaohiro Aota ret_bg = do_chunk_alloc(trans, flags); 400207730d87SJosef Bacik trans->allocating_chunk = false; 400307730d87SJosef Bacik 4004760e69c4SNaohiro Aota if (IS_ERR(ret_bg)) { 4005820c363bSNaohiro Aota ret = PTR_ERR(ret_bg); 4006760e69c4SNaohiro Aota } else if (from_extent_allocation) { 4007760e69c4SNaohiro Aota /* 4008760e69c4SNaohiro Aota * New block group is likely to be used soon. Try to activate 4009760e69c4SNaohiro Aota * it now. Failure is OK for now. 4010760e69c4SNaohiro Aota */ 4011760e69c4SNaohiro Aota btrfs_zone_activate(ret_bg); 4012760e69c4SNaohiro Aota } 4013760e69c4SNaohiro Aota 4014760e69c4SNaohiro Aota if (!ret) 4015820c363bSNaohiro Aota btrfs_put_block_group(ret_bg); 4016820c363bSNaohiro Aota 401707730d87SJosef Bacik spin_lock(&space_info->lock); 401807730d87SJosef Bacik if (ret < 0) { 401907730d87SJosef Bacik if (ret == -ENOSPC) 402007730d87SJosef Bacik space_info->full = 1; 402107730d87SJosef Bacik else 402207730d87SJosef Bacik goto out; 402307730d87SJosef Bacik } else { 402407730d87SJosef Bacik ret = 1; 402507730d87SJosef Bacik space_info->max_extent_size = 0; 402607730d87SJosef Bacik } 402707730d87SJosef Bacik 402807730d87SJosef Bacik space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; 402907730d87SJosef Bacik out: 403007730d87SJosef Bacik space_info->chunk_alloc = 0; 403107730d87SJosef Bacik spin_unlock(&space_info->lock); 403207730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 403307730d87SJosef Bacik 403407730d87SJosef Bacik return ret; 403507730d87SJosef Bacik } 403607730d87SJosef Bacik 403707730d87SJosef Bacik static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) 403807730d87SJosef Bacik { 403907730d87SJosef Bacik u64 num_dev; 404007730d87SJosef Bacik 404107730d87SJosef Bacik num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; 404207730d87SJosef Bacik if (!num_dev) 404307730d87SJosef Bacik num_dev = fs_info->fs_devices->rw_devices; 404407730d87SJosef Bacik 404507730d87SJosef Bacik return num_dev; 404607730d87SJosef Bacik } 404707730d87SJosef Bacik 40482bb2e00eSFilipe Manana static void reserve_chunk_space(struct btrfs_trans_handle *trans, 40492bb2e00eSFilipe Manana u64 bytes, 40502bb2e00eSFilipe Manana u64 type) 405107730d87SJosef Bacik { 405207730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 405307730d87SJosef Bacik struct btrfs_space_info *info; 405407730d87SJosef Bacik u64 left; 405507730d87SJosef Bacik int ret = 0; 405607730d87SJosef Bacik 405707730d87SJosef Bacik /* 405807730d87SJosef Bacik * Needed because we can end up allocating a system chunk and for an 405907730d87SJosef Bacik * atomic and race free space reservation in the chunk block reserve. 406007730d87SJosef Bacik */ 406107730d87SJosef Bacik lockdep_assert_held(&fs_info->chunk_mutex); 406207730d87SJosef Bacik 406307730d87SJosef Bacik info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); 406407730d87SJosef Bacik spin_lock(&info->lock); 406507730d87SJosef Bacik left = info->total_bytes - btrfs_space_info_used(info, true); 406607730d87SJosef Bacik spin_unlock(&info->lock); 406707730d87SJosef Bacik 40682bb2e00eSFilipe Manana if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { 406907730d87SJosef Bacik btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", 40702bb2e00eSFilipe Manana left, bytes, type); 407107730d87SJosef Bacik btrfs_dump_space_info(fs_info, info, 0, 0); 407207730d87SJosef Bacik } 407307730d87SJosef Bacik 40742bb2e00eSFilipe Manana if (left < bytes) { 407507730d87SJosef Bacik u64 flags = btrfs_system_alloc_profile(fs_info); 407679bd3712SFilipe Manana struct btrfs_block_group *bg; 407707730d87SJosef Bacik 407807730d87SJosef Bacik /* 407907730d87SJosef Bacik * Ignore failure to create system chunk. We might end up not 408007730d87SJosef Bacik * needing it, as we might not need to COW all nodes/leafs from 408107730d87SJosef Bacik * the paths we visit in the chunk tree (they were already COWed 408207730d87SJosef Bacik * or created in the current transaction for example). 408307730d87SJosef Bacik */ 4084f6f39f7aSNikolay Borisov bg = btrfs_create_chunk(trans, flags); 408579bd3712SFilipe Manana if (IS_ERR(bg)) { 408679bd3712SFilipe Manana ret = PTR_ERR(bg); 40872bb2e00eSFilipe Manana } else { 408879bd3712SFilipe Manana /* 4089b6a98021SNaohiro Aota * We have a new chunk. We also need to activate it for 4090b6a98021SNaohiro Aota * zoned filesystem. 4091b6a98021SNaohiro Aota */ 4092b6a98021SNaohiro Aota ret = btrfs_zoned_activate_one_bg(fs_info, info, true); 4093b6a98021SNaohiro Aota if (ret < 0) 4094b6a98021SNaohiro Aota return; 4095b6a98021SNaohiro Aota 4096b6a98021SNaohiro Aota /* 409779bd3712SFilipe Manana * If we fail to add the chunk item here, we end up 409879bd3712SFilipe Manana * trying again at phase 2 of chunk allocation, at 409979bd3712SFilipe Manana * btrfs_create_pending_block_groups(). So ignore 41002bb2e00eSFilipe Manana * any error here. An ENOSPC here could happen, due to 41012bb2e00eSFilipe Manana * the cases described at do_chunk_alloc() - the system 41022bb2e00eSFilipe Manana * block group we just created was just turned into RO 41032bb2e00eSFilipe Manana * mode by a scrub for example, or a running discard 41042bb2e00eSFilipe Manana * temporarily removed its free space entries, etc. 410579bd3712SFilipe Manana */ 410679bd3712SFilipe Manana btrfs_chunk_alloc_add_chunk_item(trans, bg); 410779bd3712SFilipe Manana } 410807730d87SJosef Bacik } 410907730d87SJosef Bacik 411007730d87SJosef Bacik if (!ret) { 41119270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, 411207730d87SJosef Bacik &fs_info->chunk_block_rsv, 41132bb2e00eSFilipe Manana bytes, BTRFS_RESERVE_NO_FLUSH); 41141cb3db1cSFilipe Manana if (!ret) 41152bb2e00eSFilipe Manana trans->chunk_bytes_reserved += bytes; 411607730d87SJosef Bacik } 411707730d87SJosef Bacik } 411807730d87SJosef Bacik 41192bb2e00eSFilipe Manana /* 41202bb2e00eSFilipe Manana * Reserve space in the system space for allocating or removing a chunk. 41212bb2e00eSFilipe Manana * The caller must be holding fs_info->chunk_mutex. 41222bb2e00eSFilipe Manana */ 41232bb2e00eSFilipe Manana void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) 41242bb2e00eSFilipe Manana { 41252bb2e00eSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 41262bb2e00eSFilipe Manana const u64 num_devs = get_profile_num_devs(fs_info, type); 41272bb2e00eSFilipe Manana u64 bytes; 41282bb2e00eSFilipe Manana 41292bb2e00eSFilipe Manana /* num_devs device items to update and 1 chunk item to add or remove. */ 41302bb2e00eSFilipe Manana bytes = btrfs_calc_metadata_size(fs_info, num_devs) + 41312bb2e00eSFilipe Manana btrfs_calc_insert_metadata_size(fs_info, 1); 41322bb2e00eSFilipe Manana 41332bb2e00eSFilipe Manana reserve_chunk_space(trans, bytes, type); 41342bb2e00eSFilipe Manana } 41352bb2e00eSFilipe Manana 41362bb2e00eSFilipe Manana /* 41372bb2e00eSFilipe Manana * Reserve space in the system space, if needed, for doing a modification to the 41382bb2e00eSFilipe Manana * chunk btree. 41392bb2e00eSFilipe Manana * 41402bb2e00eSFilipe Manana * @trans: A transaction handle. 41412bb2e00eSFilipe Manana * @is_item_insertion: Indicate if the modification is for inserting a new item 41422bb2e00eSFilipe Manana * in the chunk btree or if it's for the deletion or update 41432bb2e00eSFilipe Manana * of an existing item. 41442bb2e00eSFilipe Manana * 41452bb2e00eSFilipe Manana * This is used in a context where we need to update the chunk btree outside 41462bb2e00eSFilipe Manana * block group allocation and removal, to avoid a deadlock with a concurrent 41472bb2e00eSFilipe Manana * task that is allocating a metadata or data block group and therefore needs to 41482bb2e00eSFilipe Manana * update the chunk btree while holding the chunk mutex. After the update to the 41492bb2e00eSFilipe Manana * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called. 41502bb2e00eSFilipe Manana * 41512bb2e00eSFilipe Manana */ 41522bb2e00eSFilipe Manana void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans, 41532bb2e00eSFilipe Manana bool is_item_insertion) 41542bb2e00eSFilipe Manana { 41552bb2e00eSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 41562bb2e00eSFilipe Manana u64 bytes; 41572bb2e00eSFilipe Manana 41582bb2e00eSFilipe Manana if (is_item_insertion) 41592bb2e00eSFilipe Manana bytes = btrfs_calc_insert_metadata_size(fs_info, 1); 41602bb2e00eSFilipe Manana else 41612bb2e00eSFilipe Manana bytes = btrfs_calc_metadata_size(fs_info, 1); 41622bb2e00eSFilipe Manana 41632bb2e00eSFilipe Manana mutex_lock(&fs_info->chunk_mutex); 41642bb2e00eSFilipe Manana reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM); 41652bb2e00eSFilipe Manana mutex_unlock(&fs_info->chunk_mutex); 41662bb2e00eSFilipe Manana } 41672bb2e00eSFilipe Manana 41683e43c279SJosef Bacik void btrfs_put_block_group_cache(struct btrfs_fs_info *info) 41693e43c279SJosef Bacik { 417032da5386SDavid Sterba struct btrfs_block_group *block_group; 41713e43c279SJosef Bacik 417250c31eaaSJosef Bacik block_group = btrfs_lookup_first_block_group(info, 0); 41733e43c279SJosef Bacik while (block_group) { 41743e43c279SJosef Bacik btrfs_wait_block_group_cache_done(block_group); 41753e43c279SJosef Bacik spin_lock(&block_group->lock); 417650c31eaaSJosef Bacik if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, 417750c31eaaSJosef Bacik &block_group->runtime_flags)) { 417850c31eaaSJosef Bacik struct inode *inode = block_group->inode; 41793e43c279SJosef Bacik 41803e43c279SJosef Bacik block_group->inode = NULL; 41813e43c279SJosef Bacik spin_unlock(&block_group->lock); 418250c31eaaSJosef Bacik 41833e43c279SJosef Bacik ASSERT(block_group->io_ctl.inode == NULL); 41843e43c279SJosef Bacik iput(inode); 418550c31eaaSJosef Bacik } else { 418650c31eaaSJosef Bacik spin_unlock(&block_group->lock); 418750c31eaaSJosef Bacik } 418850c31eaaSJosef Bacik block_group = btrfs_next_block_group(block_group); 41893e43c279SJosef Bacik } 41903e43c279SJosef Bacik } 41913e43c279SJosef Bacik 41923e43c279SJosef Bacik /* 41933e43c279SJosef Bacik * Must be called only after stopping all workers, since we could have block 41943e43c279SJosef Bacik * group caching kthreads running, and therefore they could race with us if we 41953e43c279SJosef Bacik * freed the block groups before stopping them. 41963e43c279SJosef Bacik */ 41973e43c279SJosef Bacik int btrfs_free_block_groups(struct btrfs_fs_info *info) 41983e43c279SJosef Bacik { 419932da5386SDavid Sterba struct btrfs_block_group *block_group; 42003e43c279SJosef Bacik struct btrfs_space_info *space_info; 42013e43c279SJosef Bacik struct btrfs_caching_control *caching_ctl; 42023e43c279SJosef Bacik struct rb_node *n; 42033e43c279SJosef Bacik 420416b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 42053e43c279SJosef Bacik while (!list_empty(&info->caching_block_groups)) { 42063e43c279SJosef Bacik caching_ctl = list_entry(info->caching_block_groups.next, 42073e43c279SJosef Bacik struct btrfs_caching_control, list); 42083e43c279SJosef Bacik list_del(&caching_ctl->list); 42093e43c279SJosef Bacik btrfs_put_caching_control(caching_ctl); 42103e43c279SJosef Bacik } 421116b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42123e43c279SJosef Bacik 42133e43c279SJosef Bacik spin_lock(&info->unused_bgs_lock); 42143e43c279SJosef Bacik while (!list_empty(&info->unused_bgs)) { 42153e43c279SJosef Bacik block_group = list_first_entry(&info->unused_bgs, 421632da5386SDavid Sterba struct btrfs_block_group, 42173e43c279SJosef Bacik bg_list); 42183e43c279SJosef Bacik list_del_init(&block_group->bg_list); 42193e43c279SJosef Bacik btrfs_put_block_group(block_group); 42203e43c279SJosef Bacik } 42213e43c279SJosef Bacik 422218bb8bbfSJohannes Thumshirn while (!list_empty(&info->reclaim_bgs)) { 422318bb8bbfSJohannes Thumshirn block_group = list_first_entry(&info->reclaim_bgs, 422418bb8bbfSJohannes Thumshirn struct btrfs_block_group, 422518bb8bbfSJohannes Thumshirn bg_list); 422618bb8bbfSJohannes Thumshirn list_del_init(&block_group->bg_list); 422718bb8bbfSJohannes Thumshirn btrfs_put_block_group(block_group); 422818bb8bbfSJohannes Thumshirn } 422918bb8bbfSJohannes Thumshirn spin_unlock(&info->unused_bgs_lock); 423018bb8bbfSJohannes Thumshirn 4231afba2bc0SNaohiro Aota spin_lock(&info->zone_active_bgs_lock); 4232afba2bc0SNaohiro Aota while (!list_empty(&info->zone_active_bgs)) { 4233afba2bc0SNaohiro Aota block_group = list_first_entry(&info->zone_active_bgs, 4234afba2bc0SNaohiro Aota struct btrfs_block_group, 4235afba2bc0SNaohiro Aota active_bg_list); 4236afba2bc0SNaohiro Aota list_del_init(&block_group->active_bg_list); 4237afba2bc0SNaohiro Aota btrfs_put_block_group(block_group); 4238afba2bc0SNaohiro Aota } 4239afba2bc0SNaohiro Aota spin_unlock(&info->zone_active_bgs_lock); 4240afba2bc0SNaohiro Aota 424116b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 424208dddb29SFilipe Manana while ((n = rb_last(&info->block_group_cache_tree.rb_root)) != NULL) { 424332da5386SDavid Sterba block_group = rb_entry(n, struct btrfs_block_group, 42443e43c279SJosef Bacik cache_node); 424508dddb29SFilipe Manana rb_erase_cached(&block_group->cache_node, 42463e43c279SJosef Bacik &info->block_group_cache_tree); 42473e43c279SJosef Bacik RB_CLEAR_NODE(&block_group->cache_node); 424816b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42493e43c279SJosef Bacik 42503e43c279SJosef Bacik down_write(&block_group->space_info->groups_sem); 42513e43c279SJosef Bacik list_del(&block_group->list); 42523e43c279SJosef Bacik up_write(&block_group->space_info->groups_sem); 42533e43c279SJosef Bacik 42543e43c279SJosef Bacik /* 42553e43c279SJosef Bacik * We haven't cached this block group, which means we could 42563e43c279SJosef Bacik * possibly have excluded extents on this block group. 42573e43c279SJosef Bacik */ 42583e43c279SJosef Bacik if (block_group->cached == BTRFS_CACHE_NO || 42593e43c279SJosef Bacik block_group->cached == BTRFS_CACHE_ERROR) 42603e43c279SJosef Bacik btrfs_free_excluded_extents(block_group); 42613e43c279SJosef Bacik 42623e43c279SJosef Bacik btrfs_remove_free_space_cache(block_group); 42633e43c279SJosef Bacik ASSERT(block_group->cached != BTRFS_CACHE_STARTED); 42643e43c279SJosef Bacik ASSERT(list_empty(&block_group->dirty_list)); 42653e43c279SJosef Bacik ASSERT(list_empty(&block_group->io_list)); 42663e43c279SJosef Bacik ASSERT(list_empty(&block_group->bg_list)); 426748aaeebeSJosef Bacik ASSERT(refcount_read(&block_group->refs) == 1); 4268195a49eaSFilipe Manana ASSERT(block_group->swap_extents == 0); 42693e43c279SJosef Bacik btrfs_put_block_group(block_group); 42703e43c279SJosef Bacik 427116b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 42723e43c279SJosef Bacik } 427316b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42743e43c279SJosef Bacik 42753e43c279SJosef Bacik btrfs_release_global_block_rsv(info); 42763e43c279SJosef Bacik 42773e43c279SJosef Bacik while (!list_empty(&info->space_info)) { 42783e43c279SJosef Bacik space_info = list_entry(info->space_info.next, 42793e43c279SJosef Bacik struct btrfs_space_info, 42803e43c279SJosef Bacik list); 42813e43c279SJosef Bacik 42823e43c279SJosef Bacik /* 42833e43c279SJosef Bacik * Do not hide this behind enospc_debug, this is actually 42843e43c279SJosef Bacik * important and indicates a real bug if this happens. 42853e43c279SJosef Bacik */ 42863e43c279SJosef Bacik if (WARN_ON(space_info->bytes_pinned > 0 || 42873e43c279SJosef Bacik space_info->bytes_may_use > 0)) 42883e43c279SJosef Bacik btrfs_dump_space_info(info, space_info, 0, 0); 428940cdc509SFilipe Manana 429040cdc509SFilipe Manana /* 429140cdc509SFilipe Manana * If there was a failure to cleanup a log tree, very likely due 429240cdc509SFilipe Manana * to an IO failure on a writeback attempt of one or more of its 429340cdc509SFilipe Manana * extent buffers, we could not do proper (and cheap) unaccounting 429440cdc509SFilipe Manana * of their reserved space, so don't warn on bytes_reserved > 0 in 429540cdc509SFilipe Manana * that case. 429640cdc509SFilipe Manana */ 429740cdc509SFilipe Manana if (!(space_info->flags & BTRFS_BLOCK_GROUP_METADATA) || 429840cdc509SFilipe Manana !BTRFS_FS_LOG_CLEANUP_ERROR(info)) { 429940cdc509SFilipe Manana if (WARN_ON(space_info->bytes_reserved > 0)) 430040cdc509SFilipe Manana btrfs_dump_space_info(info, space_info, 0, 0); 430140cdc509SFilipe Manana } 430240cdc509SFilipe Manana 4303d611add4SFilipe Manana WARN_ON(space_info->reclaim_size > 0); 43043e43c279SJosef Bacik list_del(&space_info->list); 43053e43c279SJosef Bacik btrfs_sysfs_remove_space_info(space_info); 43063e43c279SJosef Bacik } 43073e43c279SJosef Bacik return 0; 43083e43c279SJosef Bacik } 4309684b752bSFilipe Manana 4310684b752bSFilipe Manana void btrfs_freeze_block_group(struct btrfs_block_group *cache) 4311684b752bSFilipe Manana { 4312684b752bSFilipe Manana atomic_inc(&cache->frozen); 4313684b752bSFilipe Manana } 4314684b752bSFilipe Manana 4315684b752bSFilipe Manana void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group) 4316684b752bSFilipe Manana { 4317684b752bSFilipe Manana struct btrfs_fs_info *fs_info = block_group->fs_info; 4318684b752bSFilipe Manana struct extent_map_tree *em_tree; 4319684b752bSFilipe Manana struct extent_map *em; 4320684b752bSFilipe Manana bool cleanup; 4321684b752bSFilipe Manana 4322684b752bSFilipe Manana spin_lock(&block_group->lock); 4323684b752bSFilipe Manana cleanup = (atomic_dec_and_test(&block_group->frozen) && 43243349b57fSJosef Bacik test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)); 4325684b752bSFilipe Manana spin_unlock(&block_group->lock); 4326684b752bSFilipe Manana 4327684b752bSFilipe Manana if (cleanup) { 4328684b752bSFilipe Manana em_tree = &fs_info->mapping_tree; 4329684b752bSFilipe Manana write_lock(&em_tree->lock); 4330684b752bSFilipe Manana em = lookup_extent_mapping(em_tree, block_group->start, 4331684b752bSFilipe Manana 1); 4332684b752bSFilipe Manana BUG_ON(!em); /* logic error, can't happen */ 4333684b752bSFilipe Manana remove_extent_mapping(em_tree, em); 4334684b752bSFilipe Manana write_unlock(&em_tree->lock); 4335684b752bSFilipe Manana 4336684b752bSFilipe Manana /* once for us and once for the tree */ 4337684b752bSFilipe Manana free_extent_map(em); 4338684b752bSFilipe Manana free_extent_map(em); 4339684b752bSFilipe Manana 4340684b752bSFilipe Manana /* 4341684b752bSFilipe Manana * We may have left one free space entry and other possible 4342684b752bSFilipe Manana * tasks trimming this block group have left 1 entry each one. 4343684b752bSFilipe Manana * Free them if any. 4344684b752bSFilipe Manana */ 4345fc80f7acSJosef Bacik btrfs_remove_free_space_cache(block_group); 4346684b752bSFilipe Manana } 4347684b752bSFilipe Manana } 4348195a49eaSFilipe Manana 4349195a49eaSFilipe Manana bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg) 4350195a49eaSFilipe Manana { 4351195a49eaSFilipe Manana bool ret = true; 4352195a49eaSFilipe Manana 4353195a49eaSFilipe Manana spin_lock(&bg->lock); 4354195a49eaSFilipe Manana if (bg->ro) 4355195a49eaSFilipe Manana ret = false; 4356195a49eaSFilipe Manana else 4357195a49eaSFilipe Manana bg->swap_extents++; 4358195a49eaSFilipe Manana spin_unlock(&bg->lock); 4359195a49eaSFilipe Manana 4360195a49eaSFilipe Manana return ret; 4361195a49eaSFilipe Manana } 4362195a49eaSFilipe Manana 4363195a49eaSFilipe Manana void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount) 4364195a49eaSFilipe Manana { 4365195a49eaSFilipe Manana spin_lock(&bg->lock); 4366195a49eaSFilipe Manana ASSERT(!bg->ro); 4367195a49eaSFilipe Manana ASSERT(bg->swap_extents >= amount); 4368195a49eaSFilipe Manana bg->swap_extents -= amount; 4369195a49eaSFilipe Manana spin_unlock(&bg->lock); 4370195a49eaSFilipe Manana } 437152bb7a21SBoris Burkov 437252bb7a21SBoris Burkov enum btrfs_block_group_size_class btrfs_calc_block_group_size_class(u64 size) 437352bb7a21SBoris Burkov { 437452bb7a21SBoris Burkov if (size <= SZ_128K) 437552bb7a21SBoris Burkov return BTRFS_BG_SZ_SMALL; 437652bb7a21SBoris Burkov if (size <= SZ_8M) 437752bb7a21SBoris Burkov return BTRFS_BG_SZ_MEDIUM; 437852bb7a21SBoris Burkov return BTRFS_BG_SZ_LARGE; 437952bb7a21SBoris Burkov } 438052bb7a21SBoris Burkov 438152bb7a21SBoris Burkov /* 438252bb7a21SBoris Burkov * Handle a block group allocating an extent in a size class 438352bb7a21SBoris Burkov * 438452bb7a21SBoris Burkov * @bg: The block group we allocated in. 438552bb7a21SBoris Burkov * @size_class: The size class of the allocation. 438652bb7a21SBoris Burkov * @force_wrong_size_class: Whether we are desperate enough to allow 438752bb7a21SBoris Burkov * mismatched size classes. 438852bb7a21SBoris Burkov * 438952bb7a21SBoris Burkov * Returns: 0 if the size class was valid for this block_group, -EAGAIN in the 439052bb7a21SBoris Burkov * case of a race that leads to the wrong size class without 439152bb7a21SBoris Burkov * force_wrong_size_class set. 439252bb7a21SBoris Burkov * 439352bb7a21SBoris Burkov * find_free_extent will skip block groups with a mismatched size class until 439452bb7a21SBoris Burkov * it really needs to avoid ENOSPC. In that case it will set 439552bb7a21SBoris Burkov * force_wrong_size_class. However, if a block group is newly allocated and 439652bb7a21SBoris Burkov * doesn't yet have a size class, then it is possible for two allocations of 439752bb7a21SBoris Burkov * different sizes to race and both try to use it. The loser is caught here and 439852bb7a21SBoris Burkov * has to retry. 439952bb7a21SBoris Burkov */ 440052bb7a21SBoris Burkov int btrfs_use_block_group_size_class(struct btrfs_block_group *bg, 440152bb7a21SBoris Burkov enum btrfs_block_group_size_class size_class, 440252bb7a21SBoris Burkov bool force_wrong_size_class) 440352bb7a21SBoris Burkov { 440452bb7a21SBoris Burkov ASSERT(size_class != BTRFS_BG_SZ_NONE); 440552bb7a21SBoris Burkov 440652bb7a21SBoris Burkov /* The new allocation is in the right size class, do nothing */ 440752bb7a21SBoris Burkov if (bg->size_class == size_class) 440852bb7a21SBoris Burkov return 0; 440952bb7a21SBoris Burkov /* 441052bb7a21SBoris Burkov * The new allocation is in a mismatched size class. 441152bb7a21SBoris Burkov * This means one of two things: 441252bb7a21SBoris Burkov * 441352bb7a21SBoris Burkov * 1. Two tasks in find_free_extent for different size_classes raced 441452bb7a21SBoris Burkov * and hit the same empty block_group. Make the loser try again. 441552bb7a21SBoris Burkov * 2. A call to find_free_extent got desperate enough to set 441652bb7a21SBoris Burkov * 'force_wrong_slab'. Don't change the size_class, but allow the 441752bb7a21SBoris Burkov * allocation. 441852bb7a21SBoris Burkov */ 441952bb7a21SBoris Burkov if (bg->size_class != BTRFS_BG_SZ_NONE) { 442052bb7a21SBoris Burkov if (force_wrong_size_class) 442152bb7a21SBoris Burkov return 0; 442252bb7a21SBoris Burkov return -EAGAIN; 442352bb7a21SBoris Burkov } 442452bb7a21SBoris Burkov /* 442552bb7a21SBoris Burkov * The happy new block group case: the new allocation is the first 442652bb7a21SBoris Burkov * one in the block_group so we set size_class. 442752bb7a21SBoris Burkov */ 442852bb7a21SBoris Burkov bg->size_class = size_class; 442952bb7a21SBoris Burkov 443052bb7a21SBoris Burkov return 0; 443152bb7a21SBoris Burkov } 4432cb0922f2SBoris Burkov 4433cb0922f2SBoris Burkov bool btrfs_block_group_should_use_size_class(struct btrfs_block_group *bg) 4434cb0922f2SBoris Burkov { 4435cb0922f2SBoris Burkov if (btrfs_is_zoned(bg->fs_info)) 4436cb0922f2SBoris Burkov return false; 4437cb0922f2SBoris Burkov if (!btrfs_is_block_group_data_only(bg)) 4438cb0922f2SBoris Burkov return false; 4439cb0922f2SBoris Burkov return true; 4440cb0922f2SBoris Burkov } 4441