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); 11783349b57fSJosef Bacik WARN_ON(test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, 11793349b57fSJosef Bacik &block_group->runtime_flags) && 11806a921de5SNaohiro Aota block_group->space_info->active_total_bytes 11816a921de5SNaohiro Aota < block_group->length); 1182e3e0520bSJosef Bacik } 1183b3470b5dSDavid Sterba block_group->space_info->total_bytes -= block_group->length; 11843349b57fSJosef Bacik if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) 11856a921de5SNaohiro Aota block_group->space_info->active_total_bytes -= block_group->length; 1186169e0da9SNaohiro Aota block_group->space_info->bytes_readonly -= 1187169e0da9SNaohiro Aota (block_group->length - block_group->zone_unusable); 1188169e0da9SNaohiro Aota block_group->space_info->bytes_zone_unusable -= 1189169e0da9SNaohiro Aota block_group->zone_unusable; 1190b3470b5dSDavid Sterba block_group->space_info->disk_total -= block_group->length * factor; 1191e3e0520bSJosef Bacik 1192e3e0520bSJosef Bacik spin_unlock(&block_group->space_info->lock); 1193e3e0520bSJosef Bacik 1194ffcb9d44SFilipe Manana /* 1195ffcb9d44SFilipe Manana * Remove the free space for the block group from the free space tree 1196ffcb9d44SFilipe Manana * and the block group's item from the extent tree before marking the 1197ffcb9d44SFilipe Manana * block group as removed. This is to prevent races with tasks that 1198ffcb9d44SFilipe Manana * freeze and unfreeze a block group, this task and another task 1199ffcb9d44SFilipe Manana * allocating a new block group - the unfreeze task ends up removing 1200ffcb9d44SFilipe Manana * the block group's extent map before the task calling this function 1201ffcb9d44SFilipe Manana * deletes the block group item from the extent tree, allowing for 1202ffcb9d44SFilipe Manana * another task to attempt to create another block group with the same 1203ffcb9d44SFilipe Manana * item key (and failing with -EEXIST and a transaction abort). 1204ffcb9d44SFilipe Manana */ 1205ffcb9d44SFilipe Manana ret = remove_block_group_free_space(trans, block_group); 1206ffcb9d44SFilipe Manana if (ret) 1207ffcb9d44SFilipe Manana goto out; 1208ffcb9d44SFilipe Manana 1209ffcb9d44SFilipe Manana ret = remove_block_group_item(trans, path, block_group); 1210ffcb9d44SFilipe Manana if (ret < 0) 1211ffcb9d44SFilipe Manana goto out; 1212ffcb9d44SFilipe Manana 1213e3e0520bSJosef Bacik spin_lock(&block_group->lock); 12143349b57fSJosef Bacik set_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags); 12153349b57fSJosef Bacik 1216e3e0520bSJosef Bacik /* 12176b7304afSFilipe Manana * At this point trimming or scrub can't start on this block group, 12186b7304afSFilipe Manana * because we removed the block group from the rbtree 12196b7304afSFilipe Manana * fs_info->block_group_cache_tree so no one can't find it anymore and 12206b7304afSFilipe Manana * even if someone already got this block group before we removed it 12216b7304afSFilipe Manana * from the rbtree, they have already incremented block_group->frozen - 12226b7304afSFilipe Manana * if they didn't, for the trimming case they won't find any free space 12236b7304afSFilipe Manana * entries because we already removed them all when we called 12246b7304afSFilipe Manana * btrfs_remove_free_space_cache(). 1225e3e0520bSJosef Bacik * 1226e3e0520bSJosef Bacik * And we must not remove the extent map from the fs_info->mapping_tree 1227e3e0520bSJosef Bacik * to prevent the same logical address range and physical device space 12286b7304afSFilipe Manana * ranges from being reused for a new block group. This is needed to 12296b7304afSFilipe Manana * avoid races with trimming and scrub. 12306b7304afSFilipe Manana * 12316b7304afSFilipe Manana * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is 1232e3e0520bSJosef Bacik * completely transactionless, so while it is trimming a range the 1233e3e0520bSJosef Bacik * currently running transaction might finish and a new one start, 1234e3e0520bSJosef Bacik * allowing for new block groups to be created that can reuse the same 1235e3e0520bSJosef Bacik * physical device locations unless we take this special care. 1236e3e0520bSJosef Bacik * 1237e3e0520bSJosef Bacik * There may also be an implicit trim operation if the file system 1238e3e0520bSJosef Bacik * is mounted with -odiscard. The same protections must remain 1239e3e0520bSJosef Bacik * in place until the extents have been discarded completely when 1240e3e0520bSJosef Bacik * the transaction commit has completed. 1241e3e0520bSJosef Bacik */ 12426b7304afSFilipe Manana remove_em = (atomic_read(&block_group->frozen) == 0); 1243e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1244e3e0520bSJosef Bacik 1245e3e0520bSJosef Bacik if (remove_em) { 1246e3e0520bSJosef Bacik struct extent_map_tree *em_tree; 1247e3e0520bSJosef Bacik 1248e3e0520bSJosef Bacik em_tree = &fs_info->mapping_tree; 1249e3e0520bSJosef Bacik write_lock(&em_tree->lock); 1250e3e0520bSJosef Bacik remove_extent_mapping(em_tree, em); 1251e3e0520bSJosef Bacik write_unlock(&em_tree->lock); 1252e3e0520bSJosef Bacik /* once for the tree */ 1253e3e0520bSJosef Bacik free_extent_map(em); 1254e3e0520bSJosef Bacik } 1255f6033c5eSXiyu Yang 12569fecd132SFilipe Manana out: 1257f6033c5eSXiyu Yang /* Once for the lookup reference */ 1258f6033c5eSXiyu Yang btrfs_put_block_group(block_group); 1259e3e0520bSJosef Bacik if (remove_rsv) 1260e3e0520bSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 1261e3e0520bSJosef Bacik btrfs_free_path(path); 1262e3e0520bSJosef Bacik return ret; 1263e3e0520bSJosef Bacik } 1264e3e0520bSJosef Bacik 1265e3e0520bSJosef Bacik struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( 1266e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info, const u64 chunk_offset) 1267e3e0520bSJosef Bacik { 1268dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 1269e3e0520bSJosef Bacik struct extent_map_tree *em_tree = &fs_info->mapping_tree; 1270e3e0520bSJosef Bacik struct extent_map *em; 1271e3e0520bSJosef Bacik struct map_lookup *map; 1272e3e0520bSJosef Bacik unsigned int num_items; 1273e3e0520bSJosef Bacik 1274e3e0520bSJosef Bacik read_lock(&em_tree->lock); 1275e3e0520bSJosef Bacik em = lookup_extent_mapping(em_tree, chunk_offset, 1); 1276e3e0520bSJosef Bacik read_unlock(&em_tree->lock); 1277e3e0520bSJosef Bacik ASSERT(em && em->start == chunk_offset); 1278e3e0520bSJosef Bacik 1279e3e0520bSJosef Bacik /* 1280e3e0520bSJosef Bacik * We need to reserve 3 + N units from the metadata space info in order 1281e3e0520bSJosef Bacik * to remove a block group (done at btrfs_remove_chunk() and at 1282e3e0520bSJosef Bacik * btrfs_remove_block_group()), which are used for: 1283e3e0520bSJosef Bacik * 1284e3e0520bSJosef Bacik * 1 unit for adding the free space inode's orphan (located in the tree 1285e3e0520bSJosef Bacik * of tree roots). 1286e3e0520bSJosef Bacik * 1 unit for deleting the block group item (located in the extent 1287e3e0520bSJosef Bacik * tree). 1288e3e0520bSJosef Bacik * 1 unit for deleting the free space item (located in tree of tree 1289e3e0520bSJosef Bacik * roots). 1290e3e0520bSJosef Bacik * N units for deleting N device extent items corresponding to each 1291e3e0520bSJosef Bacik * stripe (located in the device tree). 1292e3e0520bSJosef Bacik * 1293e3e0520bSJosef Bacik * In order to remove a block group we also need to reserve units in the 1294e3e0520bSJosef Bacik * system space info in order to update the chunk tree (update one or 1295e3e0520bSJosef Bacik * more device items and remove one chunk item), but this is done at 1296e3e0520bSJosef Bacik * btrfs_remove_chunk() through a call to check_system_chunk(). 1297e3e0520bSJosef Bacik */ 1298e3e0520bSJosef Bacik map = em->map_lookup; 1299e3e0520bSJosef Bacik num_items = 3 + map->num_stripes; 1300e3e0520bSJosef Bacik free_extent_map(em); 1301e3e0520bSJosef Bacik 1302dfe8aec4SJosef Bacik return btrfs_start_transaction_fallback_global_rsv(root, num_items); 1303e3e0520bSJosef Bacik } 1304e3e0520bSJosef Bacik 1305e3e0520bSJosef Bacik /* 130626ce2095SJosef Bacik * Mark block group @cache read-only, so later write won't happen to block 130726ce2095SJosef Bacik * group @cache. 130826ce2095SJosef Bacik * 130926ce2095SJosef Bacik * If @force is not set, this function will only mark the block group readonly 131026ce2095SJosef Bacik * if we have enough free space (1M) in other metadata/system block groups. 131126ce2095SJosef Bacik * If @force is not set, this function will mark the block group readonly 131226ce2095SJosef Bacik * without checking free space. 131326ce2095SJosef Bacik * 131426ce2095SJosef Bacik * NOTE: This function doesn't care if other block groups can contain all the 131526ce2095SJosef Bacik * data in this block group. That check should be done by relocation routine, 131626ce2095SJosef Bacik * not this function. 131726ce2095SJosef Bacik */ 131832da5386SDavid Sterba static int inc_block_group_ro(struct btrfs_block_group *cache, int force) 131926ce2095SJosef Bacik { 132026ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 132126ce2095SJosef Bacik u64 num_bytes; 132226ce2095SJosef Bacik int ret = -ENOSPC; 132326ce2095SJosef Bacik 132426ce2095SJosef Bacik spin_lock(&sinfo->lock); 132526ce2095SJosef Bacik spin_lock(&cache->lock); 132626ce2095SJosef Bacik 1327195a49eaSFilipe Manana if (cache->swap_extents) { 1328195a49eaSFilipe Manana ret = -ETXTBSY; 1329195a49eaSFilipe Manana goto out; 1330195a49eaSFilipe Manana } 1331195a49eaSFilipe Manana 133226ce2095SJosef Bacik if (cache->ro) { 133326ce2095SJosef Bacik cache->ro++; 133426ce2095SJosef Bacik ret = 0; 133526ce2095SJosef Bacik goto out; 133626ce2095SJosef Bacik } 133726ce2095SJosef Bacik 1338b3470b5dSDavid Sterba num_bytes = cache->length - cache->reserved - cache->pinned - 1339169e0da9SNaohiro Aota cache->bytes_super - cache->zone_unusable - cache->used; 134026ce2095SJosef Bacik 134126ce2095SJosef Bacik /* 1342a30a3d20SJosef Bacik * Data never overcommits, even in mixed mode, so do just the straight 1343a30a3d20SJosef Bacik * check of left over space in how much we have allocated. 1344a30a3d20SJosef Bacik */ 1345a30a3d20SJosef Bacik if (force) { 1346a30a3d20SJosef Bacik ret = 0; 1347a30a3d20SJosef Bacik } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) { 1348a30a3d20SJosef Bacik u64 sinfo_used = btrfs_space_info_used(sinfo, true); 1349a30a3d20SJosef Bacik 1350a30a3d20SJosef Bacik /* 135126ce2095SJosef Bacik * Here we make sure if we mark this bg RO, we still have enough 1352f8935566SJosef Bacik * free space as buffer. 135326ce2095SJosef Bacik */ 1354a30a3d20SJosef Bacik if (sinfo_used + num_bytes <= sinfo->total_bytes) 1355a30a3d20SJosef Bacik ret = 0; 1356a30a3d20SJosef Bacik } else { 1357a30a3d20SJosef Bacik /* 1358a30a3d20SJosef Bacik * We overcommit metadata, so we need to do the 1359a30a3d20SJosef Bacik * btrfs_can_overcommit check here, and we need to pass in 1360a30a3d20SJosef Bacik * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of 1361a30a3d20SJosef Bacik * leeway to allow us to mark this block group as read only. 1362a30a3d20SJosef Bacik */ 1363a30a3d20SJosef Bacik if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes, 1364a30a3d20SJosef Bacik BTRFS_RESERVE_NO_FLUSH)) 1365a30a3d20SJosef Bacik ret = 0; 1366a30a3d20SJosef Bacik } 1367a30a3d20SJosef Bacik 1368a30a3d20SJosef Bacik if (!ret) { 136926ce2095SJosef Bacik sinfo->bytes_readonly += num_bytes; 1370169e0da9SNaohiro Aota if (btrfs_is_zoned(cache->fs_info)) { 1371169e0da9SNaohiro Aota /* Migrate zone_unusable bytes to readonly */ 1372169e0da9SNaohiro Aota sinfo->bytes_readonly += cache->zone_unusable; 1373169e0da9SNaohiro Aota sinfo->bytes_zone_unusable -= cache->zone_unusable; 1374169e0da9SNaohiro Aota cache->zone_unusable = 0; 1375169e0da9SNaohiro Aota } 137626ce2095SJosef Bacik cache->ro++; 137726ce2095SJosef Bacik list_add_tail(&cache->ro_list, &sinfo->ro_bgs); 137826ce2095SJosef Bacik } 137926ce2095SJosef Bacik out: 138026ce2095SJosef Bacik spin_unlock(&cache->lock); 138126ce2095SJosef Bacik spin_unlock(&sinfo->lock); 138226ce2095SJosef Bacik if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { 138326ce2095SJosef Bacik btrfs_info(cache->fs_info, 1384b3470b5dSDavid Sterba "unable to make block group %llu ro", cache->start); 138526ce2095SJosef Bacik btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); 138626ce2095SJosef Bacik } 138726ce2095SJosef Bacik return ret; 138826ce2095SJosef Bacik } 138926ce2095SJosef Bacik 1390fe119a6eSNikolay Borisov static bool clean_pinned_extents(struct btrfs_trans_handle *trans, 1391fe119a6eSNikolay Borisov struct btrfs_block_group *bg) 139245bb5d6aSNikolay Borisov { 139345bb5d6aSNikolay Borisov struct btrfs_fs_info *fs_info = bg->fs_info; 1394fe119a6eSNikolay Borisov struct btrfs_transaction *prev_trans = NULL; 139545bb5d6aSNikolay Borisov const u64 start = bg->start; 139645bb5d6aSNikolay Borisov const u64 end = start + bg->length - 1; 139745bb5d6aSNikolay Borisov int ret; 139845bb5d6aSNikolay Borisov 1399fe119a6eSNikolay Borisov spin_lock(&fs_info->trans_lock); 1400fe119a6eSNikolay Borisov if (trans->transaction->list.prev != &fs_info->trans_list) { 1401fe119a6eSNikolay Borisov prev_trans = list_last_entry(&trans->transaction->list, 1402fe119a6eSNikolay Borisov struct btrfs_transaction, list); 1403fe119a6eSNikolay Borisov refcount_inc(&prev_trans->use_count); 1404fe119a6eSNikolay Borisov } 1405fe119a6eSNikolay Borisov spin_unlock(&fs_info->trans_lock); 1406fe119a6eSNikolay Borisov 140745bb5d6aSNikolay Borisov /* 140845bb5d6aSNikolay Borisov * Hold the unused_bg_unpin_mutex lock to avoid racing with 140945bb5d6aSNikolay Borisov * btrfs_finish_extent_commit(). If we are at transaction N, another 141045bb5d6aSNikolay Borisov * task might be running finish_extent_commit() for the previous 141145bb5d6aSNikolay Borisov * transaction N - 1, and have seen a range belonging to the block 1412fe119a6eSNikolay Borisov * group in pinned_extents before we were able to clear the whole block 1413fe119a6eSNikolay Borisov * group range from pinned_extents. This means that task can lookup for 1414fe119a6eSNikolay Borisov * the block group after we unpinned it from pinned_extents and removed 1415fe119a6eSNikolay Borisov * it, leading to a BUG_ON() at unpin_extent_range(). 141645bb5d6aSNikolay Borisov */ 141745bb5d6aSNikolay Borisov mutex_lock(&fs_info->unused_bg_unpin_mutex); 1418fe119a6eSNikolay Borisov if (prev_trans) { 1419fe119a6eSNikolay Borisov ret = clear_extent_bits(&prev_trans->pinned_extents, start, end, 142045bb5d6aSNikolay Borisov EXTENT_DIRTY); 142145bb5d6aSNikolay Borisov if (ret) 1422534cf531SFilipe Manana goto out; 1423fe119a6eSNikolay Borisov } 142445bb5d6aSNikolay Borisov 1425fe119a6eSNikolay Borisov ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end, 142645bb5d6aSNikolay Borisov EXTENT_DIRTY); 1427534cf531SFilipe Manana out: 142845bb5d6aSNikolay Borisov mutex_unlock(&fs_info->unused_bg_unpin_mutex); 14295150bf19SFilipe Manana if (prev_trans) 14305150bf19SFilipe Manana btrfs_put_transaction(prev_trans); 143145bb5d6aSNikolay Borisov 1432534cf531SFilipe Manana return ret == 0; 143345bb5d6aSNikolay Borisov } 143445bb5d6aSNikolay Borisov 143526ce2095SJosef Bacik /* 1436e3e0520bSJosef Bacik * Process the unused_bgs list and remove any that don't have any allocated 1437e3e0520bSJosef Bacik * space inside of them. 1438e3e0520bSJosef Bacik */ 1439e3e0520bSJosef Bacik void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) 1440e3e0520bSJosef Bacik { 144132da5386SDavid Sterba struct btrfs_block_group *block_group; 1442e3e0520bSJosef Bacik struct btrfs_space_info *space_info; 1443e3e0520bSJosef Bacik struct btrfs_trans_handle *trans; 14446e80d4f8SDennis Zhou const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC); 1445e3e0520bSJosef Bacik int ret = 0; 1446e3e0520bSJosef Bacik 1447e3e0520bSJosef Bacik if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 1448e3e0520bSJosef Bacik return; 1449e3e0520bSJosef Bacik 14502f12741fSJosef Bacik if (btrfs_fs_closing(fs_info)) 14512f12741fSJosef Bacik return; 14522f12741fSJosef Bacik 1453ddfd08cbSJosef Bacik /* 1454ddfd08cbSJosef Bacik * Long running balances can keep us blocked here for eternity, so 1455ddfd08cbSJosef Bacik * simply skip deletion if we're unable to get the mutex. 1456ddfd08cbSJosef Bacik */ 1457f3372065SJohannes Thumshirn if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) 1458ddfd08cbSJosef Bacik return; 1459ddfd08cbSJosef Bacik 1460e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1461e3e0520bSJosef Bacik while (!list_empty(&fs_info->unused_bgs)) { 1462e3e0520bSJosef Bacik int trimming; 1463e3e0520bSJosef Bacik 1464e3e0520bSJosef Bacik block_group = list_first_entry(&fs_info->unused_bgs, 146532da5386SDavid Sterba struct btrfs_block_group, 1466e3e0520bSJosef Bacik bg_list); 1467e3e0520bSJosef Bacik list_del_init(&block_group->bg_list); 1468e3e0520bSJosef Bacik 1469e3e0520bSJosef Bacik space_info = block_group->space_info; 1470e3e0520bSJosef Bacik 1471e3e0520bSJosef Bacik if (ret || btrfs_mixed_space_info(space_info)) { 1472e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1473e3e0520bSJosef Bacik continue; 1474e3e0520bSJosef Bacik } 1475e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1476e3e0520bSJosef Bacik 1477b0643e59SDennis Zhou btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 1478b0643e59SDennis Zhou 1479e3e0520bSJosef Bacik /* Don't want to race with allocators so take the groups_sem */ 1480e3e0520bSJosef Bacik down_write(&space_info->groups_sem); 14816e80d4f8SDennis Zhou 14826e80d4f8SDennis Zhou /* 14836e80d4f8SDennis Zhou * Async discard moves the final block group discard to be prior 14846e80d4f8SDennis Zhou * to the unused_bgs code path. Therefore, if it's not fully 14856e80d4f8SDennis Zhou * trimmed, punt it back to the async discard lists. 14866e80d4f8SDennis Zhou */ 14876e80d4f8SDennis Zhou if (btrfs_test_opt(fs_info, DISCARD_ASYNC) && 14886e80d4f8SDennis Zhou !btrfs_is_free_space_trimmed(block_group)) { 14896e80d4f8SDennis Zhou trace_btrfs_skip_unused_block_group(block_group); 14906e80d4f8SDennis Zhou up_write(&space_info->groups_sem); 14916e80d4f8SDennis Zhou /* Requeue if we failed because of async discard */ 14926e80d4f8SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 14936e80d4f8SDennis Zhou block_group); 14946e80d4f8SDennis Zhou goto next; 14956e80d4f8SDennis Zhou } 14966e80d4f8SDennis Zhou 1497e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1498e3e0520bSJosef Bacik if (block_group->reserved || block_group->pinned || 1499bf38be65SDavid Sterba block_group->used || block_group->ro || 1500e3e0520bSJosef Bacik list_is_singular(&block_group->list)) { 1501e3e0520bSJosef Bacik /* 1502e3e0520bSJosef Bacik * We want to bail if we made new allocations or have 1503e3e0520bSJosef Bacik * outstanding allocations in this block group. We do 1504e3e0520bSJosef Bacik * the ro check in case balance is currently acting on 1505e3e0520bSJosef Bacik * this block group. 1506e3e0520bSJosef Bacik */ 1507e3e0520bSJosef Bacik trace_btrfs_skip_unused_block_group(block_group); 1508e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1509e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1510e3e0520bSJosef Bacik goto next; 1511e3e0520bSJosef Bacik } 1512e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1513e3e0520bSJosef Bacik 1514e3e0520bSJosef Bacik /* We don't want to force the issue, only flip if it's ok. */ 1515e11c0406SJosef Bacik ret = inc_block_group_ro(block_group, 0); 1516e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1517e3e0520bSJosef Bacik if (ret < 0) { 1518e3e0520bSJosef Bacik ret = 0; 1519e3e0520bSJosef Bacik goto next; 1520e3e0520bSJosef Bacik } 1521e3e0520bSJosef Bacik 152274e91b12SNaohiro Aota ret = btrfs_zone_finish(block_group); 152374e91b12SNaohiro Aota if (ret < 0) { 152474e91b12SNaohiro Aota btrfs_dec_block_group_ro(block_group); 152574e91b12SNaohiro Aota if (ret == -EAGAIN) 152674e91b12SNaohiro Aota ret = 0; 152774e91b12SNaohiro Aota goto next; 152874e91b12SNaohiro Aota } 152974e91b12SNaohiro Aota 1530e3e0520bSJosef Bacik /* 1531e3e0520bSJosef Bacik * Want to do this before we do anything else so we can recover 1532e3e0520bSJosef Bacik * properly if we fail to join the transaction. 1533e3e0520bSJosef Bacik */ 1534e3e0520bSJosef Bacik trans = btrfs_start_trans_remove_block_group(fs_info, 1535b3470b5dSDavid Sterba block_group->start); 1536e3e0520bSJosef Bacik if (IS_ERR(trans)) { 1537e3e0520bSJosef Bacik btrfs_dec_block_group_ro(block_group); 1538e3e0520bSJosef Bacik ret = PTR_ERR(trans); 1539e3e0520bSJosef Bacik goto next; 1540e3e0520bSJosef Bacik } 1541e3e0520bSJosef Bacik 1542e3e0520bSJosef Bacik /* 1543e3e0520bSJosef Bacik * We could have pending pinned extents for this block group, 1544e3e0520bSJosef Bacik * just delete them, we don't care about them anymore. 1545e3e0520bSJosef Bacik */ 1546534cf531SFilipe Manana if (!clean_pinned_extents(trans, block_group)) { 1547534cf531SFilipe Manana btrfs_dec_block_group_ro(block_group); 1548e3e0520bSJosef Bacik goto end_trans; 1549534cf531SFilipe Manana } 1550e3e0520bSJosef Bacik 1551b0643e59SDennis Zhou /* 1552b0643e59SDennis Zhou * At this point, the block_group is read only and should fail 1553b0643e59SDennis Zhou * new allocations. However, btrfs_finish_extent_commit() can 1554b0643e59SDennis Zhou * cause this block_group to be placed back on the discard 1555b0643e59SDennis Zhou * lists because now the block_group isn't fully discarded. 1556b0643e59SDennis Zhou * Bail here and try again later after discarding everything. 1557b0643e59SDennis Zhou */ 1558b0643e59SDennis Zhou spin_lock(&fs_info->discard_ctl.lock); 1559b0643e59SDennis Zhou if (!list_empty(&block_group->discard_list)) { 1560b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1561b0643e59SDennis Zhou btrfs_dec_block_group_ro(block_group); 1562b0643e59SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 1563b0643e59SDennis Zhou block_group); 1564b0643e59SDennis Zhou goto end_trans; 1565b0643e59SDennis Zhou } 1566b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1567b0643e59SDennis Zhou 1568e3e0520bSJosef Bacik /* Reset pinned so btrfs_put_block_group doesn't complain */ 1569e3e0520bSJosef Bacik spin_lock(&space_info->lock); 1570e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1571e3e0520bSJosef Bacik 1572e3e0520bSJosef Bacik btrfs_space_info_update_bytes_pinned(fs_info, space_info, 1573e3e0520bSJosef Bacik -block_group->pinned); 1574e3e0520bSJosef Bacik space_info->bytes_readonly += block_group->pinned; 1575e3e0520bSJosef Bacik block_group->pinned = 0; 1576e3e0520bSJosef Bacik 1577e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1578e3e0520bSJosef Bacik spin_unlock(&space_info->lock); 1579e3e0520bSJosef Bacik 15806e80d4f8SDennis Zhou /* 15816e80d4f8SDennis Zhou * The normal path here is an unused block group is passed here, 15826e80d4f8SDennis Zhou * then trimming is handled in the transaction commit path. 15836e80d4f8SDennis Zhou * Async discard interposes before this to do the trimming 15846e80d4f8SDennis Zhou * before coming down the unused block group path as trimming 15856e80d4f8SDennis Zhou * will no longer be done later in the transaction commit path. 15866e80d4f8SDennis Zhou */ 15876e80d4f8SDennis Zhou if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC)) 15886e80d4f8SDennis Zhou goto flip_async; 15896e80d4f8SDennis Zhou 1590dcba6e48SNaohiro Aota /* 1591dcba6e48SNaohiro Aota * DISCARD can flip during remount. On zoned filesystems, we 1592dcba6e48SNaohiro Aota * need to reset sequential-required zones. 1593dcba6e48SNaohiro Aota */ 1594dcba6e48SNaohiro Aota trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) || 1595dcba6e48SNaohiro Aota btrfs_is_zoned(fs_info); 1596e3e0520bSJosef Bacik 1597e3e0520bSJosef Bacik /* Implicit trim during transaction commit. */ 1598e3e0520bSJosef Bacik if (trimming) 15996b7304afSFilipe Manana btrfs_freeze_block_group(block_group); 1600e3e0520bSJosef Bacik 1601e3e0520bSJosef Bacik /* 1602e3e0520bSJosef Bacik * Btrfs_remove_chunk will abort the transaction if things go 1603e3e0520bSJosef Bacik * horribly wrong. 1604e3e0520bSJosef Bacik */ 1605b3470b5dSDavid Sterba ret = btrfs_remove_chunk(trans, block_group->start); 1606e3e0520bSJosef Bacik 1607e3e0520bSJosef Bacik if (ret) { 1608e3e0520bSJosef Bacik if (trimming) 16096b7304afSFilipe Manana btrfs_unfreeze_block_group(block_group); 1610e3e0520bSJosef Bacik goto end_trans; 1611e3e0520bSJosef Bacik } 1612e3e0520bSJosef Bacik 1613e3e0520bSJosef Bacik /* 1614e3e0520bSJosef Bacik * If we're not mounted with -odiscard, we can just forget 1615e3e0520bSJosef Bacik * about this block group. Otherwise we'll need to wait 1616e3e0520bSJosef Bacik * until transaction commit to do the actual discard. 1617e3e0520bSJosef Bacik */ 1618e3e0520bSJosef Bacik if (trimming) { 1619e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1620e3e0520bSJosef Bacik /* 1621e3e0520bSJosef Bacik * A concurrent scrub might have added us to the list 1622e3e0520bSJosef Bacik * fs_info->unused_bgs, so use a list_move operation 1623e3e0520bSJosef Bacik * to add the block group to the deleted_bgs list. 1624e3e0520bSJosef Bacik */ 1625e3e0520bSJosef Bacik list_move(&block_group->bg_list, 1626e3e0520bSJosef Bacik &trans->transaction->deleted_bgs); 1627e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1628e3e0520bSJosef Bacik btrfs_get_block_group(block_group); 1629e3e0520bSJosef Bacik } 1630e3e0520bSJosef Bacik end_trans: 1631e3e0520bSJosef Bacik btrfs_end_transaction(trans); 1632e3e0520bSJosef Bacik next: 1633e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1634e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1635e3e0520bSJosef Bacik } 1636e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1637f3372065SJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 16386e80d4f8SDennis Zhou return; 16396e80d4f8SDennis Zhou 16406e80d4f8SDennis Zhou flip_async: 16416e80d4f8SDennis Zhou btrfs_end_transaction(trans); 1642f3372065SJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 16436e80d4f8SDennis Zhou btrfs_put_block_group(block_group); 16446e80d4f8SDennis Zhou btrfs_discard_punt_unused_bgs_list(fs_info); 1645e3e0520bSJosef Bacik } 1646e3e0520bSJosef Bacik 164732da5386SDavid Sterba void btrfs_mark_bg_unused(struct btrfs_block_group *bg) 1648e3e0520bSJosef Bacik { 1649e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info = bg->fs_info; 1650e3e0520bSJosef Bacik 1651e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1652e3e0520bSJosef Bacik if (list_empty(&bg->bg_list)) { 1653e3e0520bSJosef Bacik btrfs_get_block_group(bg); 1654e3e0520bSJosef Bacik trace_btrfs_add_unused_block_group(bg); 1655e3e0520bSJosef Bacik list_add_tail(&bg->bg_list, &fs_info->unused_bgs); 1656e3e0520bSJosef Bacik } 1657e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1658e3e0520bSJosef Bacik } 16594358d963SJosef Bacik 16602ca0ec77SJohannes Thumshirn /* 16612ca0ec77SJohannes Thumshirn * We want block groups with a low number of used bytes to be in the beginning 16622ca0ec77SJohannes Thumshirn * of the list, so they will get reclaimed first. 16632ca0ec77SJohannes Thumshirn */ 16642ca0ec77SJohannes Thumshirn static int reclaim_bgs_cmp(void *unused, const struct list_head *a, 16652ca0ec77SJohannes Thumshirn const struct list_head *b) 16662ca0ec77SJohannes Thumshirn { 16672ca0ec77SJohannes Thumshirn const struct btrfs_block_group *bg1, *bg2; 16682ca0ec77SJohannes Thumshirn 16692ca0ec77SJohannes Thumshirn bg1 = list_entry(a, struct btrfs_block_group, bg_list); 16702ca0ec77SJohannes Thumshirn bg2 = list_entry(b, struct btrfs_block_group, bg_list); 16712ca0ec77SJohannes Thumshirn 16722ca0ec77SJohannes Thumshirn return bg1->used > bg2->used; 16732ca0ec77SJohannes Thumshirn } 16742ca0ec77SJohannes Thumshirn 16753687fcb0SJohannes Thumshirn static inline bool btrfs_should_reclaim(struct btrfs_fs_info *fs_info) 16763687fcb0SJohannes Thumshirn { 16773687fcb0SJohannes Thumshirn if (btrfs_is_zoned(fs_info)) 16783687fcb0SJohannes Thumshirn return btrfs_zoned_should_reclaim(fs_info); 16793687fcb0SJohannes Thumshirn return true; 16803687fcb0SJohannes Thumshirn } 16813687fcb0SJohannes Thumshirn 168281531225SBoris Burkov static bool should_reclaim_block_group(struct btrfs_block_group *bg, u64 bytes_freed) 168381531225SBoris Burkov { 168481531225SBoris Burkov const struct btrfs_space_info *space_info = bg->space_info; 168581531225SBoris Burkov const int reclaim_thresh = READ_ONCE(space_info->bg_reclaim_threshold); 168681531225SBoris Burkov const u64 new_val = bg->used; 168781531225SBoris Burkov const u64 old_val = new_val + bytes_freed; 168881531225SBoris Burkov u64 thresh; 168981531225SBoris Burkov 169081531225SBoris Burkov if (reclaim_thresh == 0) 169181531225SBoris Burkov return false; 169281531225SBoris Burkov 1693428c8e03SDavid Sterba thresh = mult_perc(bg->length, reclaim_thresh); 169481531225SBoris Burkov 169581531225SBoris Burkov /* 169681531225SBoris Burkov * If we were below the threshold before don't reclaim, we are likely a 169781531225SBoris Burkov * brand new block group and we don't want to relocate new block groups. 169881531225SBoris Burkov */ 169981531225SBoris Burkov if (old_val < thresh) 170081531225SBoris Burkov return false; 170181531225SBoris Burkov if (new_val >= thresh) 170281531225SBoris Burkov return false; 170381531225SBoris Burkov return true; 170481531225SBoris Burkov } 170581531225SBoris Burkov 170618bb8bbfSJohannes Thumshirn void btrfs_reclaim_bgs_work(struct work_struct *work) 170718bb8bbfSJohannes Thumshirn { 170818bb8bbfSJohannes Thumshirn struct btrfs_fs_info *fs_info = 170918bb8bbfSJohannes Thumshirn container_of(work, struct btrfs_fs_info, reclaim_bgs_work); 171018bb8bbfSJohannes Thumshirn struct btrfs_block_group *bg; 171118bb8bbfSJohannes Thumshirn struct btrfs_space_info *space_info; 171218bb8bbfSJohannes Thumshirn 171318bb8bbfSJohannes Thumshirn if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 171418bb8bbfSJohannes Thumshirn return; 171518bb8bbfSJohannes Thumshirn 17162f12741fSJosef Bacik if (btrfs_fs_closing(fs_info)) 17172f12741fSJosef Bacik return; 17182f12741fSJosef Bacik 17193687fcb0SJohannes Thumshirn if (!btrfs_should_reclaim(fs_info)) 17203687fcb0SJohannes Thumshirn return; 17213687fcb0SJohannes Thumshirn 1722ca5e4ea0SNaohiro Aota sb_start_write(fs_info->sb); 1723ca5e4ea0SNaohiro Aota 1724ca5e4ea0SNaohiro Aota if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 1725ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 172618bb8bbfSJohannes Thumshirn return; 1727ca5e4ea0SNaohiro Aota } 172818bb8bbfSJohannes Thumshirn 17299cc0b837SJohannes Thumshirn /* 17309cc0b837SJohannes Thumshirn * Long running balances can keep us blocked here for eternity, so 17319cc0b837SJohannes Thumshirn * simply skip reclaim if we're unable to get the mutex. 17329cc0b837SJohannes Thumshirn */ 17339cc0b837SJohannes Thumshirn if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) { 17349cc0b837SJohannes Thumshirn btrfs_exclop_finish(fs_info); 1735ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 17369cc0b837SJohannes Thumshirn return; 17379cc0b837SJohannes Thumshirn } 17389cc0b837SJohannes Thumshirn 173918bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 17402ca0ec77SJohannes Thumshirn /* 17412ca0ec77SJohannes Thumshirn * Sort happens under lock because we can't simply splice it and sort. 17422ca0ec77SJohannes Thumshirn * The block groups might still be in use and reachable via bg_list, 17432ca0ec77SJohannes Thumshirn * and their presence in the reclaim_bgs list must be preserved. 17442ca0ec77SJohannes Thumshirn */ 17452ca0ec77SJohannes Thumshirn list_sort(NULL, &fs_info->reclaim_bgs, reclaim_bgs_cmp); 174618bb8bbfSJohannes Thumshirn while (!list_empty(&fs_info->reclaim_bgs)) { 17475f93e776SJohannes Thumshirn u64 zone_unusable; 17481cea5cf0SFilipe Manana int ret = 0; 17491cea5cf0SFilipe Manana 175018bb8bbfSJohannes Thumshirn bg = list_first_entry(&fs_info->reclaim_bgs, 175118bb8bbfSJohannes Thumshirn struct btrfs_block_group, 175218bb8bbfSJohannes Thumshirn bg_list); 175318bb8bbfSJohannes Thumshirn list_del_init(&bg->bg_list); 175418bb8bbfSJohannes Thumshirn 175518bb8bbfSJohannes Thumshirn space_info = bg->space_info; 175618bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 175718bb8bbfSJohannes Thumshirn 175818bb8bbfSJohannes Thumshirn /* Don't race with allocators so take the groups_sem */ 175918bb8bbfSJohannes Thumshirn down_write(&space_info->groups_sem); 176018bb8bbfSJohannes Thumshirn 176118bb8bbfSJohannes Thumshirn spin_lock(&bg->lock); 176218bb8bbfSJohannes Thumshirn if (bg->reserved || bg->pinned || bg->ro) { 176318bb8bbfSJohannes Thumshirn /* 176418bb8bbfSJohannes Thumshirn * We want to bail if we made new allocations or have 176518bb8bbfSJohannes Thumshirn * outstanding allocations in this block group. We do 176618bb8bbfSJohannes Thumshirn * the ro check in case balance is currently acting on 176718bb8bbfSJohannes Thumshirn * this block group. 176818bb8bbfSJohannes Thumshirn */ 176918bb8bbfSJohannes Thumshirn spin_unlock(&bg->lock); 177018bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 177118bb8bbfSJohannes Thumshirn goto next; 177218bb8bbfSJohannes Thumshirn } 1773cc4804bfSBoris Burkov if (bg->used == 0) { 1774cc4804bfSBoris Burkov /* 1775cc4804bfSBoris Burkov * It is possible that we trigger relocation on a block 1776cc4804bfSBoris Burkov * group as its extents are deleted and it first goes 1777cc4804bfSBoris Burkov * below the threshold, then shortly after goes empty. 1778cc4804bfSBoris Burkov * 1779cc4804bfSBoris Burkov * In this case, relocating it does delete it, but has 1780cc4804bfSBoris Burkov * some overhead in relocation specific metadata, looking 1781cc4804bfSBoris Burkov * for the non-existent extents and running some extra 1782cc4804bfSBoris Burkov * transactions, which we can avoid by using one of the 1783cc4804bfSBoris Burkov * other mechanisms for dealing with empty block groups. 1784cc4804bfSBoris Burkov */ 1785cc4804bfSBoris Burkov if (!btrfs_test_opt(fs_info, DISCARD_ASYNC)) 1786cc4804bfSBoris Burkov btrfs_mark_bg_unused(bg); 1787cc4804bfSBoris Burkov spin_unlock(&bg->lock); 1788cc4804bfSBoris Burkov up_write(&space_info->groups_sem); 1789cc4804bfSBoris Burkov goto next; 179081531225SBoris Burkov 179181531225SBoris Burkov } 179281531225SBoris Burkov /* 179381531225SBoris Burkov * The block group might no longer meet the reclaim condition by 179481531225SBoris Burkov * the time we get around to reclaiming it, so to avoid 179581531225SBoris Burkov * reclaiming overly full block_groups, skip reclaiming them. 179681531225SBoris Burkov * 179781531225SBoris Burkov * Since the decision making process also depends on the amount 179881531225SBoris Burkov * being freed, pass in a fake giant value to skip that extra 179981531225SBoris Burkov * check, which is more meaningful when adding to the list in 180081531225SBoris Burkov * the first place. 180181531225SBoris Burkov */ 180281531225SBoris Burkov if (!should_reclaim_block_group(bg, bg->length)) { 180381531225SBoris Burkov spin_unlock(&bg->lock); 180481531225SBoris Burkov up_write(&space_info->groups_sem); 180581531225SBoris Burkov goto next; 1806cc4804bfSBoris Burkov } 180718bb8bbfSJohannes Thumshirn spin_unlock(&bg->lock); 180818bb8bbfSJohannes Thumshirn 180918bb8bbfSJohannes Thumshirn /* Get out fast, in case we're unmounting the filesystem */ 181018bb8bbfSJohannes Thumshirn if (btrfs_fs_closing(fs_info)) { 181118bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 181218bb8bbfSJohannes Thumshirn goto next; 181318bb8bbfSJohannes Thumshirn } 181418bb8bbfSJohannes Thumshirn 18155f93e776SJohannes Thumshirn /* 18165f93e776SJohannes Thumshirn * Cache the zone_unusable value before turning the block group 18175f93e776SJohannes Thumshirn * to read only. As soon as the blog group is read only it's 18185f93e776SJohannes Thumshirn * zone_unusable value gets moved to the block group's read-only 18195f93e776SJohannes Thumshirn * bytes and isn't available for calculations anymore. 18205f93e776SJohannes Thumshirn */ 18215f93e776SJohannes Thumshirn zone_unusable = bg->zone_unusable; 182218bb8bbfSJohannes Thumshirn ret = inc_block_group_ro(bg, 0); 182318bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 182418bb8bbfSJohannes Thumshirn if (ret < 0) 182518bb8bbfSJohannes Thumshirn goto next; 182618bb8bbfSJohannes Thumshirn 18275f93e776SJohannes Thumshirn btrfs_info(fs_info, 18285f93e776SJohannes Thumshirn "reclaiming chunk %llu with %llu%% used %llu%% unusable", 182995cd356cSJohannes Thumshirn bg->start, 183095cd356cSJohannes Thumshirn div64_u64(bg->used * 100, bg->length), 18315f93e776SJohannes Thumshirn div64_u64(zone_unusable * 100, bg->length)); 183218bb8bbfSJohannes Thumshirn trace_btrfs_reclaim_block_group(bg); 183318bb8bbfSJohannes Thumshirn ret = btrfs_relocate_chunk(fs_info, bg->start); 183474944c87SJosef Bacik if (ret) { 183574944c87SJosef Bacik btrfs_dec_block_group_ro(bg); 183618bb8bbfSJohannes Thumshirn btrfs_err(fs_info, "error relocating chunk %llu", 183718bb8bbfSJohannes Thumshirn bg->start); 183874944c87SJosef Bacik } 183918bb8bbfSJohannes Thumshirn 184018bb8bbfSJohannes Thumshirn next: 18411cea5cf0SFilipe Manana btrfs_put_block_group(bg); 1842d96b3424SFilipe Manana spin_lock(&fs_info->unused_bgs_lock); 184318bb8bbfSJohannes Thumshirn } 184418bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 184518bb8bbfSJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 184618bb8bbfSJohannes Thumshirn btrfs_exclop_finish(fs_info); 1847ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 184818bb8bbfSJohannes Thumshirn } 184918bb8bbfSJohannes Thumshirn 185018bb8bbfSJohannes Thumshirn void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info) 185118bb8bbfSJohannes Thumshirn { 185218bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 185318bb8bbfSJohannes Thumshirn if (!list_empty(&fs_info->reclaim_bgs)) 185418bb8bbfSJohannes Thumshirn queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work); 185518bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 185618bb8bbfSJohannes Thumshirn } 185718bb8bbfSJohannes Thumshirn 185818bb8bbfSJohannes Thumshirn void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg) 185918bb8bbfSJohannes Thumshirn { 186018bb8bbfSJohannes Thumshirn struct btrfs_fs_info *fs_info = bg->fs_info; 186118bb8bbfSJohannes Thumshirn 186218bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 186318bb8bbfSJohannes Thumshirn if (list_empty(&bg->bg_list)) { 186418bb8bbfSJohannes Thumshirn btrfs_get_block_group(bg); 186518bb8bbfSJohannes Thumshirn trace_btrfs_add_reclaim_block_group(bg); 186618bb8bbfSJohannes Thumshirn list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs); 186718bb8bbfSJohannes Thumshirn } 186818bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 186918bb8bbfSJohannes Thumshirn } 187018bb8bbfSJohannes Thumshirn 1871e3ba67a1SJohannes Thumshirn static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key, 1872e3ba67a1SJohannes Thumshirn struct btrfs_path *path) 1873e3ba67a1SJohannes Thumshirn { 1874e3ba67a1SJohannes Thumshirn struct extent_map_tree *em_tree; 1875e3ba67a1SJohannes Thumshirn struct extent_map *em; 1876e3ba67a1SJohannes Thumshirn struct btrfs_block_group_item bg; 1877e3ba67a1SJohannes Thumshirn struct extent_buffer *leaf; 1878e3ba67a1SJohannes Thumshirn int slot; 1879e3ba67a1SJohannes Thumshirn u64 flags; 1880e3ba67a1SJohannes Thumshirn int ret = 0; 1881e3ba67a1SJohannes Thumshirn 1882e3ba67a1SJohannes Thumshirn slot = path->slots[0]; 1883e3ba67a1SJohannes Thumshirn leaf = path->nodes[0]; 1884e3ba67a1SJohannes Thumshirn 1885e3ba67a1SJohannes Thumshirn em_tree = &fs_info->mapping_tree; 1886e3ba67a1SJohannes Thumshirn read_lock(&em_tree->lock); 1887e3ba67a1SJohannes Thumshirn em = lookup_extent_mapping(em_tree, key->objectid, key->offset); 1888e3ba67a1SJohannes Thumshirn read_unlock(&em_tree->lock); 1889e3ba67a1SJohannes Thumshirn if (!em) { 1890e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1891e3ba67a1SJohannes Thumshirn "logical %llu len %llu found bg but no related chunk", 1892e3ba67a1SJohannes Thumshirn key->objectid, key->offset); 1893e3ba67a1SJohannes Thumshirn return -ENOENT; 1894e3ba67a1SJohannes Thumshirn } 1895e3ba67a1SJohannes Thumshirn 1896e3ba67a1SJohannes Thumshirn if (em->start != key->objectid || em->len != key->offset) { 1897e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1898e3ba67a1SJohannes Thumshirn "block group %llu len %llu mismatch with chunk %llu len %llu", 1899e3ba67a1SJohannes Thumshirn key->objectid, key->offset, em->start, em->len); 1900e3ba67a1SJohannes Thumshirn ret = -EUCLEAN; 1901e3ba67a1SJohannes Thumshirn goto out_free_em; 1902e3ba67a1SJohannes Thumshirn } 1903e3ba67a1SJohannes Thumshirn 1904e3ba67a1SJohannes Thumshirn read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot), 1905e3ba67a1SJohannes Thumshirn sizeof(bg)); 1906e3ba67a1SJohannes Thumshirn flags = btrfs_stack_block_group_flags(&bg) & 1907e3ba67a1SJohannes Thumshirn BTRFS_BLOCK_GROUP_TYPE_MASK; 1908e3ba67a1SJohannes Thumshirn 1909e3ba67a1SJohannes Thumshirn if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 1910e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1911e3ba67a1SJohannes Thumshirn "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", 1912e3ba67a1SJohannes Thumshirn key->objectid, key->offset, flags, 1913e3ba67a1SJohannes Thumshirn (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type)); 1914e3ba67a1SJohannes Thumshirn ret = -EUCLEAN; 1915e3ba67a1SJohannes Thumshirn } 1916e3ba67a1SJohannes Thumshirn 1917e3ba67a1SJohannes Thumshirn out_free_em: 1918e3ba67a1SJohannes Thumshirn free_extent_map(em); 1919e3ba67a1SJohannes Thumshirn return ret; 1920e3ba67a1SJohannes Thumshirn } 1921e3ba67a1SJohannes Thumshirn 19224358d963SJosef Bacik static int find_first_block_group(struct btrfs_fs_info *fs_info, 19234358d963SJosef Bacik struct btrfs_path *path, 19244358d963SJosef Bacik struct btrfs_key *key) 19254358d963SJosef Bacik { 1926dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 1927e3ba67a1SJohannes Thumshirn int ret; 19284358d963SJosef Bacik struct btrfs_key found_key; 19294358d963SJosef Bacik 193036dfbbe2SGabriel Niebler btrfs_for_each_slot(root, key, &found_key, path, ret) { 19314358d963SJosef Bacik if (found_key.objectid >= key->objectid && 19324358d963SJosef Bacik found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { 193336dfbbe2SGabriel Niebler return read_bg_from_eb(fs_info, &found_key, path); 1934e3ba67a1SJohannes Thumshirn } 19354358d963SJosef Bacik } 19364358d963SJosef Bacik return ret; 19374358d963SJosef Bacik } 19384358d963SJosef Bacik 19394358d963SJosef Bacik static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) 19404358d963SJosef Bacik { 19414358d963SJosef Bacik u64 extra_flags = chunk_to_extended(flags) & 19424358d963SJosef Bacik BTRFS_EXTENDED_PROFILE_MASK; 19434358d963SJosef Bacik 19444358d963SJosef Bacik write_seqlock(&fs_info->profiles_lock); 19454358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 19464358d963SJosef Bacik fs_info->avail_data_alloc_bits |= extra_flags; 19474358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_METADATA) 19484358d963SJosef Bacik fs_info->avail_metadata_alloc_bits |= extra_flags; 19494358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 19504358d963SJosef Bacik fs_info->avail_system_alloc_bits |= extra_flags; 19514358d963SJosef Bacik write_sequnlock(&fs_info->profiles_lock); 19524358d963SJosef Bacik } 19534358d963SJosef Bacik 195443dd529aSDavid Sterba /* 195543dd529aSDavid Sterba * Map a physical disk address to a list of logical addresses. 19569ee9b979SNikolay Borisov * 19579ee9b979SNikolay Borisov * @fs_info: the filesystem 195896a14336SNikolay Borisov * @chunk_start: logical address of block group 195996a14336SNikolay Borisov * @physical: physical address to map to logical addresses 196096a14336SNikolay Borisov * @logical: return array of logical addresses which map to @physical 196196a14336SNikolay Borisov * @naddrs: length of @logical 196296a14336SNikolay Borisov * @stripe_len: size of IO stripe for the given block group 196396a14336SNikolay Borisov * 196496a14336SNikolay Borisov * Maps a particular @physical disk address to a list of @logical addresses. 196596a14336SNikolay Borisov * Used primarily to exclude those portions of a block group that contain super 196696a14336SNikolay Borisov * block copies. 196796a14336SNikolay Borisov */ 196896a14336SNikolay Borisov int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, 19691eb82ef8SChristoph Hellwig u64 physical, u64 **logical, int *naddrs, int *stripe_len) 197096a14336SNikolay Borisov { 197196a14336SNikolay Borisov struct extent_map *em; 197296a14336SNikolay Borisov struct map_lookup *map; 197396a14336SNikolay Borisov u64 *buf; 197496a14336SNikolay Borisov u64 bytenr; 19751776ad17SNikolay Borisov u64 data_stripe_length; 19761776ad17SNikolay Borisov u64 io_stripe_size; 19771776ad17SNikolay Borisov int i, nr = 0; 19781776ad17SNikolay Borisov int ret = 0; 197996a14336SNikolay Borisov 198096a14336SNikolay Borisov em = btrfs_get_chunk_map(fs_info, chunk_start, 1); 198196a14336SNikolay Borisov if (IS_ERR(em)) 198296a14336SNikolay Borisov return -EIO; 198396a14336SNikolay Borisov 198496a14336SNikolay Borisov map = em->map_lookup; 19859e22b925SNikolay Borisov data_stripe_length = em->orig_block_len; 19861776ad17SNikolay Borisov io_stripe_size = map->stripe_len; 1987138082f3SNaohiro Aota chunk_start = em->start; 198896a14336SNikolay Borisov 19899e22b925SNikolay Borisov /* For RAID5/6 adjust to a full IO stripe length */ 19909e22b925SNikolay Borisov if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 19911776ad17SNikolay Borisov io_stripe_size = map->stripe_len * nr_data_stripes(map); 199296a14336SNikolay Borisov 199396a14336SNikolay Borisov buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); 19941776ad17SNikolay Borisov if (!buf) { 19951776ad17SNikolay Borisov ret = -ENOMEM; 19961776ad17SNikolay Borisov goto out; 19971776ad17SNikolay Borisov } 199896a14336SNikolay Borisov 199996a14336SNikolay Borisov for (i = 0; i < map->num_stripes; i++) { 20001776ad17SNikolay Borisov bool already_inserted = false; 20011776ad17SNikolay Borisov u64 stripe_nr; 2002138082f3SNaohiro Aota u64 offset; 20031776ad17SNikolay Borisov int j; 20041776ad17SNikolay Borisov 20051776ad17SNikolay Borisov if (!in_range(physical, map->stripes[i].physical, 20061776ad17SNikolay Borisov data_stripe_length)) 200796a14336SNikolay Borisov continue; 200896a14336SNikolay Borisov 200996a14336SNikolay Borisov stripe_nr = physical - map->stripes[i].physical; 2010138082f3SNaohiro Aota stripe_nr = div64_u64_rem(stripe_nr, map->stripe_len, &offset); 201196a14336SNikolay Borisov 2012ac067734SDavid Sterba if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 2013ac067734SDavid Sterba BTRFS_BLOCK_GROUP_RAID10)) { 201496a14336SNikolay Borisov stripe_nr = stripe_nr * map->num_stripes + i; 201596a14336SNikolay Borisov stripe_nr = div_u64(stripe_nr, map->sub_stripes); 201696a14336SNikolay Borisov } 201796a14336SNikolay Borisov /* 201896a14336SNikolay Borisov * The remaining case would be for RAID56, multiply by 201996a14336SNikolay Borisov * nr_data_stripes(). Alternatively, just use rmap_len below 202096a14336SNikolay Borisov * instead of map->stripe_len 202196a14336SNikolay Borisov */ 202296a14336SNikolay Borisov 2023138082f3SNaohiro Aota bytenr = chunk_start + stripe_nr * io_stripe_size + offset; 20241776ad17SNikolay Borisov 20251776ad17SNikolay Borisov /* Ensure we don't add duplicate addresses */ 202696a14336SNikolay Borisov for (j = 0; j < nr; j++) { 20271776ad17SNikolay Borisov if (buf[j] == bytenr) { 20281776ad17SNikolay Borisov already_inserted = true; 202996a14336SNikolay Borisov break; 203096a14336SNikolay Borisov } 203196a14336SNikolay Borisov } 20321776ad17SNikolay Borisov 20331776ad17SNikolay Borisov if (!already_inserted) 20341776ad17SNikolay Borisov buf[nr++] = bytenr; 203596a14336SNikolay Borisov } 203696a14336SNikolay Borisov 203796a14336SNikolay Borisov *logical = buf; 203896a14336SNikolay Borisov *naddrs = nr; 20391776ad17SNikolay Borisov *stripe_len = io_stripe_size; 20401776ad17SNikolay Borisov out: 204196a14336SNikolay Borisov free_extent_map(em); 20421776ad17SNikolay Borisov return ret; 204396a14336SNikolay Borisov } 204496a14336SNikolay Borisov 204532da5386SDavid Sterba static int exclude_super_stripes(struct btrfs_block_group *cache) 20464358d963SJosef Bacik { 20474358d963SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 204812659251SNaohiro Aota const bool zoned = btrfs_is_zoned(fs_info); 20494358d963SJosef Bacik u64 bytenr; 20504358d963SJosef Bacik u64 *logical; 20514358d963SJosef Bacik int stripe_len; 20524358d963SJosef Bacik int i, nr, ret; 20534358d963SJosef Bacik 2054b3470b5dSDavid Sterba if (cache->start < BTRFS_SUPER_INFO_OFFSET) { 2055b3470b5dSDavid Sterba stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start; 20564358d963SJosef Bacik cache->bytes_super += stripe_len; 2057b3470b5dSDavid Sterba ret = btrfs_add_excluded_extent(fs_info, cache->start, 20584358d963SJosef Bacik stripe_len); 20594358d963SJosef Bacik if (ret) 20604358d963SJosef Bacik return ret; 20614358d963SJosef Bacik } 20624358d963SJosef Bacik 20634358d963SJosef Bacik for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 20644358d963SJosef Bacik bytenr = btrfs_sb_offset(i); 20651eb82ef8SChristoph Hellwig ret = btrfs_rmap_block(fs_info, cache->start, 20664358d963SJosef Bacik bytenr, &logical, &nr, &stripe_len); 20674358d963SJosef Bacik if (ret) 20684358d963SJosef Bacik return ret; 20694358d963SJosef Bacik 207012659251SNaohiro Aota /* Shouldn't have super stripes in sequential zones */ 207112659251SNaohiro Aota if (zoned && nr) { 207212659251SNaohiro Aota btrfs_err(fs_info, 207312659251SNaohiro Aota "zoned: block group %llu must not contain super block", 207412659251SNaohiro Aota cache->start); 207512659251SNaohiro Aota return -EUCLEAN; 207612659251SNaohiro Aota } 207712659251SNaohiro Aota 20784358d963SJosef Bacik while (nr--) { 207996f9b0f2SNikolay Borisov u64 len = min_t(u64, stripe_len, 208096f9b0f2SNikolay Borisov cache->start + cache->length - logical[nr]); 20814358d963SJosef Bacik 20824358d963SJosef Bacik cache->bytes_super += len; 208396f9b0f2SNikolay Borisov ret = btrfs_add_excluded_extent(fs_info, logical[nr], 208496f9b0f2SNikolay Borisov len); 20854358d963SJosef Bacik if (ret) { 20864358d963SJosef Bacik kfree(logical); 20874358d963SJosef Bacik return ret; 20884358d963SJosef Bacik } 20894358d963SJosef Bacik } 20904358d963SJosef Bacik 20914358d963SJosef Bacik kfree(logical); 20924358d963SJosef Bacik } 20934358d963SJosef Bacik return 0; 20944358d963SJosef Bacik } 20954358d963SJosef Bacik 209632da5386SDavid Sterba static struct btrfs_block_group *btrfs_create_block_group_cache( 20979afc6649SQu Wenruo struct btrfs_fs_info *fs_info, u64 start) 20984358d963SJosef Bacik { 209932da5386SDavid Sterba struct btrfs_block_group *cache; 21004358d963SJosef Bacik 21014358d963SJosef Bacik cache = kzalloc(sizeof(*cache), GFP_NOFS); 21024358d963SJosef Bacik if (!cache) 21034358d963SJosef Bacik return NULL; 21044358d963SJosef Bacik 21054358d963SJosef Bacik cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), 21064358d963SJosef Bacik GFP_NOFS); 21074358d963SJosef Bacik if (!cache->free_space_ctl) { 21084358d963SJosef Bacik kfree(cache); 21094358d963SJosef Bacik return NULL; 21104358d963SJosef Bacik } 21114358d963SJosef Bacik 2112b3470b5dSDavid Sterba cache->start = start; 21134358d963SJosef Bacik 21144358d963SJosef Bacik cache->fs_info = fs_info; 21154358d963SJosef Bacik cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); 21164358d963SJosef Bacik 21176e80d4f8SDennis Zhou cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED; 21186e80d4f8SDennis Zhou 211948aaeebeSJosef Bacik refcount_set(&cache->refs, 1); 21204358d963SJosef Bacik spin_lock_init(&cache->lock); 21214358d963SJosef Bacik init_rwsem(&cache->data_rwsem); 21224358d963SJosef Bacik INIT_LIST_HEAD(&cache->list); 21234358d963SJosef Bacik INIT_LIST_HEAD(&cache->cluster_list); 21244358d963SJosef Bacik INIT_LIST_HEAD(&cache->bg_list); 21254358d963SJosef Bacik INIT_LIST_HEAD(&cache->ro_list); 2126b0643e59SDennis Zhou INIT_LIST_HEAD(&cache->discard_list); 21274358d963SJosef Bacik INIT_LIST_HEAD(&cache->dirty_list); 21284358d963SJosef Bacik INIT_LIST_HEAD(&cache->io_list); 2129afba2bc0SNaohiro Aota INIT_LIST_HEAD(&cache->active_bg_list); 2130cd79909bSJosef Bacik btrfs_init_free_space_ctl(cache, cache->free_space_ctl); 21316b7304afSFilipe Manana atomic_set(&cache->frozen, 0); 21324358d963SJosef Bacik mutex_init(&cache->free_space_lock); 2133c29abab4SJosef Bacik cache->full_stripe_locks_root.root = RB_ROOT; 2134c29abab4SJosef Bacik mutex_init(&cache->full_stripe_locks_root.lock); 21354358d963SJosef Bacik 21364358d963SJosef Bacik return cache; 21374358d963SJosef Bacik } 21384358d963SJosef Bacik 21394358d963SJosef Bacik /* 21404358d963SJosef Bacik * Iterate all chunks and verify that each of them has the corresponding block 21414358d963SJosef Bacik * group 21424358d963SJosef Bacik */ 21434358d963SJosef Bacik static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) 21444358d963SJosef Bacik { 21454358d963SJosef Bacik struct extent_map_tree *map_tree = &fs_info->mapping_tree; 21464358d963SJosef Bacik struct extent_map *em; 214732da5386SDavid Sterba struct btrfs_block_group *bg; 21484358d963SJosef Bacik u64 start = 0; 21494358d963SJosef Bacik int ret = 0; 21504358d963SJosef Bacik 21514358d963SJosef Bacik while (1) { 21524358d963SJosef Bacik read_lock(&map_tree->lock); 21534358d963SJosef Bacik /* 21544358d963SJosef Bacik * lookup_extent_mapping will return the first extent map 21554358d963SJosef Bacik * intersecting the range, so setting @len to 1 is enough to 21564358d963SJosef Bacik * get the first chunk. 21574358d963SJosef Bacik */ 21584358d963SJosef Bacik em = lookup_extent_mapping(map_tree, start, 1); 21594358d963SJosef Bacik read_unlock(&map_tree->lock); 21604358d963SJosef Bacik if (!em) 21614358d963SJosef Bacik break; 21624358d963SJosef Bacik 21634358d963SJosef Bacik bg = btrfs_lookup_block_group(fs_info, em->start); 21644358d963SJosef Bacik if (!bg) { 21654358d963SJosef Bacik btrfs_err(fs_info, 21664358d963SJosef Bacik "chunk start=%llu len=%llu doesn't have corresponding block group", 21674358d963SJosef Bacik em->start, em->len); 21684358d963SJosef Bacik ret = -EUCLEAN; 21694358d963SJosef Bacik free_extent_map(em); 21704358d963SJosef Bacik break; 21714358d963SJosef Bacik } 2172b3470b5dSDavid Sterba if (bg->start != em->start || bg->length != em->len || 21734358d963SJosef Bacik (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != 21744358d963SJosef Bacik (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 21754358d963SJosef Bacik btrfs_err(fs_info, 21764358d963SJosef Bacik "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", 21774358d963SJosef Bacik em->start, em->len, 21784358d963SJosef Bacik em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, 2179b3470b5dSDavid Sterba bg->start, bg->length, 21804358d963SJosef Bacik bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); 21814358d963SJosef Bacik ret = -EUCLEAN; 21824358d963SJosef Bacik free_extent_map(em); 21834358d963SJosef Bacik btrfs_put_block_group(bg); 21844358d963SJosef Bacik break; 21854358d963SJosef Bacik } 21864358d963SJosef Bacik start = em->start + em->len; 21874358d963SJosef Bacik free_extent_map(em); 21884358d963SJosef Bacik btrfs_put_block_group(bg); 21894358d963SJosef Bacik } 21904358d963SJosef Bacik return ret; 21914358d963SJosef Bacik } 21924358d963SJosef Bacik 2193ffb9e0f0SQu Wenruo static int read_one_block_group(struct btrfs_fs_info *info, 21944afd2fe8SJohannes Thumshirn struct btrfs_block_group_item *bgi, 2195d49a2ddbSQu Wenruo const struct btrfs_key *key, 2196ffb9e0f0SQu Wenruo int need_clear) 2197ffb9e0f0SQu Wenruo { 219832da5386SDavid Sterba struct btrfs_block_group *cache; 2199ffb9e0f0SQu Wenruo const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS); 2200ffb9e0f0SQu Wenruo int ret; 2201ffb9e0f0SQu Wenruo 2202d49a2ddbSQu Wenruo ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY); 2203ffb9e0f0SQu Wenruo 22049afc6649SQu Wenruo cache = btrfs_create_block_group_cache(info, key->objectid); 2205ffb9e0f0SQu Wenruo if (!cache) 2206ffb9e0f0SQu Wenruo return -ENOMEM; 2207ffb9e0f0SQu Wenruo 22084afd2fe8SJohannes Thumshirn cache->length = key->offset; 22094afd2fe8SJohannes Thumshirn cache->used = btrfs_stack_block_group_used(bgi); 22107248e0ceSQu Wenruo cache->commit_used = cache->used; 22114afd2fe8SJohannes Thumshirn cache->flags = btrfs_stack_block_group_flags(bgi); 2212f7238e50SJosef Bacik cache->global_root_id = btrfs_stack_block_group_chunk_objectid(bgi); 22139afc6649SQu Wenruo 2214e3e39c72SMarcos Paulo de Souza set_free_space_tree_thresholds(cache); 2215e3e39c72SMarcos Paulo de Souza 2216ffb9e0f0SQu Wenruo if (need_clear) { 2217ffb9e0f0SQu Wenruo /* 2218ffb9e0f0SQu Wenruo * When we mount with old space cache, we need to 2219ffb9e0f0SQu Wenruo * set BTRFS_DC_CLEAR and set dirty flag. 2220ffb9e0f0SQu Wenruo * 2221ffb9e0f0SQu Wenruo * a) Setting 'BTRFS_DC_CLEAR' makes sure that we 2222ffb9e0f0SQu Wenruo * truncate the old free space cache inode and 2223ffb9e0f0SQu Wenruo * setup a new one. 2224ffb9e0f0SQu Wenruo * b) Setting 'dirty flag' makes sure that we flush 2225ffb9e0f0SQu Wenruo * the new space cache info onto disk. 2226ffb9e0f0SQu Wenruo */ 2227ffb9e0f0SQu Wenruo if (btrfs_test_opt(info, SPACE_CACHE)) 2228ffb9e0f0SQu Wenruo cache->disk_cache_state = BTRFS_DC_CLEAR; 2229ffb9e0f0SQu Wenruo } 2230ffb9e0f0SQu Wenruo if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && 2231ffb9e0f0SQu Wenruo (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { 2232ffb9e0f0SQu Wenruo btrfs_err(info, 2233ffb9e0f0SQu Wenruo "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", 2234ffb9e0f0SQu Wenruo cache->start); 2235ffb9e0f0SQu Wenruo ret = -EINVAL; 2236ffb9e0f0SQu Wenruo goto error; 2237ffb9e0f0SQu Wenruo } 2238ffb9e0f0SQu Wenruo 2239a94794d5SNaohiro Aota ret = btrfs_load_block_group_zone_info(cache, false); 224008e11a3dSNaohiro Aota if (ret) { 224108e11a3dSNaohiro Aota btrfs_err(info, "zoned: failed to load zone info of bg %llu", 224208e11a3dSNaohiro Aota cache->start); 224308e11a3dSNaohiro Aota goto error; 224408e11a3dSNaohiro Aota } 224508e11a3dSNaohiro Aota 2246ffb9e0f0SQu Wenruo /* 2247ffb9e0f0SQu Wenruo * We need to exclude the super stripes now so that the space info has 2248ffb9e0f0SQu Wenruo * super bytes accounted for, otherwise we'll think we have more space 2249ffb9e0f0SQu Wenruo * than we actually do. 2250ffb9e0f0SQu Wenruo */ 2251ffb9e0f0SQu Wenruo ret = exclude_super_stripes(cache); 2252ffb9e0f0SQu Wenruo if (ret) { 2253ffb9e0f0SQu Wenruo /* We may have excluded something, so call this just in case. */ 2254ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2255ffb9e0f0SQu Wenruo goto error; 2256ffb9e0f0SQu Wenruo } 2257ffb9e0f0SQu Wenruo 2258ffb9e0f0SQu Wenruo /* 2259169e0da9SNaohiro Aota * For zoned filesystem, space after the allocation offset is the only 2260169e0da9SNaohiro Aota * free space for a block group. So, we don't need any caching work. 2261169e0da9SNaohiro Aota * btrfs_calc_zone_unusable() will set the amount of free space and 2262169e0da9SNaohiro Aota * zone_unusable space. 2263169e0da9SNaohiro Aota * 2264169e0da9SNaohiro Aota * For regular filesystem, check for two cases, either we are full, and 2265169e0da9SNaohiro Aota * therefore don't need to bother with the caching work since we won't 2266169e0da9SNaohiro Aota * find any space, or we are empty, and we can just add all the space 2267169e0da9SNaohiro Aota * in and be done with it. This saves us _a_lot_ of time, particularly 2268169e0da9SNaohiro Aota * in the full case. 2269ffb9e0f0SQu Wenruo */ 2270169e0da9SNaohiro Aota if (btrfs_is_zoned(info)) { 2271169e0da9SNaohiro Aota btrfs_calc_zone_unusable(cache); 2272c46c4247SNaohiro Aota /* Should not have any excluded extents. Just in case, though. */ 2273c46c4247SNaohiro Aota btrfs_free_excluded_extents(cache); 2274169e0da9SNaohiro Aota } else if (cache->length == cache->used) { 2275ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 2276ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2277ffb9e0f0SQu Wenruo } else if (cache->used == 0) { 2278ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 22799afc6649SQu Wenruo add_new_free_space(cache, cache->start, 22809afc6649SQu Wenruo cache->start + cache->length); 2281ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2282ffb9e0f0SQu Wenruo } 2283ffb9e0f0SQu Wenruo 2284ffb9e0f0SQu Wenruo ret = btrfs_add_block_group_cache(info, cache); 2285ffb9e0f0SQu Wenruo if (ret) { 2286ffb9e0f0SQu Wenruo btrfs_remove_free_space_cache(cache); 2287ffb9e0f0SQu Wenruo goto error; 2288ffb9e0f0SQu Wenruo } 2289ffb9e0f0SQu Wenruo trace_btrfs_add_block_group(info, cache, 0); 2290723de71dSJosef Bacik btrfs_add_bg_to_space_info(info, cache); 2291ffb9e0f0SQu Wenruo 2292ffb9e0f0SQu Wenruo set_avail_alloc_bits(info, cache->flags); 2293a09f23c3SAnand Jain if (btrfs_chunk_writeable(info, cache->start)) { 2294a09f23c3SAnand Jain if (cache->used == 0) { 2295ffb9e0f0SQu Wenruo ASSERT(list_empty(&cache->bg_list)); 22966e80d4f8SDennis Zhou if (btrfs_test_opt(info, DISCARD_ASYNC)) 22976e80d4f8SDennis Zhou btrfs_discard_queue_work(&info->discard_ctl, cache); 22986e80d4f8SDennis Zhou else 2299ffb9e0f0SQu Wenruo btrfs_mark_bg_unused(cache); 2300ffb9e0f0SQu Wenruo } 2301a09f23c3SAnand Jain } else { 2302a09f23c3SAnand Jain inc_block_group_ro(cache, 1); 2303a09f23c3SAnand Jain } 2304a09f23c3SAnand Jain 2305ffb9e0f0SQu Wenruo return 0; 2306ffb9e0f0SQu Wenruo error: 2307ffb9e0f0SQu Wenruo btrfs_put_block_group(cache); 2308ffb9e0f0SQu Wenruo return ret; 2309ffb9e0f0SQu Wenruo } 2310ffb9e0f0SQu Wenruo 231142437a63SJosef Bacik static int fill_dummy_bgs(struct btrfs_fs_info *fs_info) 231242437a63SJosef Bacik { 231342437a63SJosef Bacik struct extent_map_tree *em_tree = &fs_info->mapping_tree; 231442437a63SJosef Bacik struct rb_node *node; 231542437a63SJosef Bacik int ret = 0; 231642437a63SJosef Bacik 231742437a63SJosef Bacik for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 231842437a63SJosef Bacik struct extent_map *em; 231942437a63SJosef Bacik struct map_lookup *map; 232042437a63SJosef Bacik struct btrfs_block_group *bg; 232142437a63SJosef Bacik 232242437a63SJosef Bacik em = rb_entry(node, struct extent_map, rb_node); 232342437a63SJosef Bacik map = em->map_lookup; 232442437a63SJosef Bacik bg = btrfs_create_block_group_cache(fs_info, em->start); 232542437a63SJosef Bacik if (!bg) { 232642437a63SJosef Bacik ret = -ENOMEM; 232742437a63SJosef Bacik break; 232842437a63SJosef Bacik } 232942437a63SJosef Bacik 233042437a63SJosef Bacik /* Fill dummy cache as FULL */ 233142437a63SJosef Bacik bg->length = em->len; 233242437a63SJosef Bacik bg->flags = map->type; 233342437a63SJosef Bacik bg->cached = BTRFS_CACHE_FINISHED; 233442437a63SJosef Bacik bg->used = em->len; 233542437a63SJosef Bacik bg->flags = map->type; 233642437a63SJosef Bacik ret = btrfs_add_block_group_cache(fs_info, bg); 23372b29726cSQu Wenruo /* 23382b29726cSQu Wenruo * We may have some valid block group cache added already, in 23392b29726cSQu Wenruo * that case we skip to the next one. 23402b29726cSQu Wenruo */ 23412b29726cSQu Wenruo if (ret == -EEXIST) { 23422b29726cSQu Wenruo ret = 0; 23432b29726cSQu Wenruo btrfs_put_block_group(bg); 23442b29726cSQu Wenruo continue; 23452b29726cSQu Wenruo } 23462b29726cSQu Wenruo 234742437a63SJosef Bacik if (ret) { 234842437a63SJosef Bacik btrfs_remove_free_space_cache(bg); 234942437a63SJosef Bacik btrfs_put_block_group(bg); 235042437a63SJosef Bacik break; 235142437a63SJosef Bacik } 23522b29726cSQu Wenruo 2353723de71dSJosef Bacik btrfs_add_bg_to_space_info(fs_info, bg); 235442437a63SJosef Bacik 235542437a63SJosef Bacik set_avail_alloc_bits(fs_info, bg->flags); 235642437a63SJosef Bacik } 235742437a63SJosef Bacik if (!ret) 235842437a63SJosef Bacik btrfs_init_global_block_rsv(fs_info); 235942437a63SJosef Bacik return ret; 236042437a63SJosef Bacik } 236142437a63SJosef Bacik 23624358d963SJosef Bacik int btrfs_read_block_groups(struct btrfs_fs_info *info) 23634358d963SJosef Bacik { 2364dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(info); 23654358d963SJosef Bacik struct btrfs_path *path; 23664358d963SJosef Bacik int ret; 236732da5386SDavid Sterba struct btrfs_block_group *cache; 23684358d963SJosef Bacik struct btrfs_space_info *space_info; 23694358d963SJosef Bacik struct btrfs_key key; 23704358d963SJosef Bacik int need_clear = 0; 23714358d963SJosef Bacik u64 cache_gen; 23724358d963SJosef Bacik 237381d5d614SQu Wenruo /* 237481d5d614SQu Wenruo * Either no extent root (with ibadroots rescue option) or we have 237581d5d614SQu Wenruo * unsupported RO options. The fs can never be mounted read-write, so no 237681d5d614SQu Wenruo * need to waste time searching block group items. 237781d5d614SQu Wenruo * 237881d5d614SQu Wenruo * This also allows new extent tree related changes to be RO compat, 237981d5d614SQu Wenruo * no need for a full incompat flag. 238081d5d614SQu Wenruo */ 238181d5d614SQu Wenruo if (!root || (btrfs_super_compat_ro_flags(info->super_copy) & 238281d5d614SQu Wenruo ~BTRFS_FEATURE_COMPAT_RO_SUPP)) 238342437a63SJosef Bacik return fill_dummy_bgs(info); 238442437a63SJosef Bacik 23854358d963SJosef Bacik key.objectid = 0; 23864358d963SJosef Bacik key.offset = 0; 23874358d963SJosef Bacik key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 23884358d963SJosef Bacik path = btrfs_alloc_path(); 23894358d963SJosef Bacik if (!path) 23904358d963SJosef Bacik return -ENOMEM; 23914358d963SJosef Bacik 23924358d963SJosef Bacik cache_gen = btrfs_super_cache_generation(info->super_copy); 23934358d963SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 23944358d963SJosef Bacik btrfs_super_generation(info->super_copy) != cache_gen) 23954358d963SJosef Bacik need_clear = 1; 23964358d963SJosef Bacik if (btrfs_test_opt(info, CLEAR_CACHE)) 23974358d963SJosef Bacik need_clear = 1; 23984358d963SJosef Bacik 23994358d963SJosef Bacik while (1) { 24004afd2fe8SJohannes Thumshirn struct btrfs_block_group_item bgi; 24014afd2fe8SJohannes Thumshirn struct extent_buffer *leaf; 24024afd2fe8SJohannes Thumshirn int slot; 24034afd2fe8SJohannes Thumshirn 24044358d963SJosef Bacik ret = find_first_block_group(info, path, &key); 24054358d963SJosef Bacik if (ret > 0) 24064358d963SJosef Bacik break; 24074358d963SJosef Bacik if (ret != 0) 24084358d963SJosef Bacik goto error; 24094358d963SJosef Bacik 24104afd2fe8SJohannes Thumshirn leaf = path->nodes[0]; 24114afd2fe8SJohannes Thumshirn slot = path->slots[0]; 24124afd2fe8SJohannes Thumshirn 24134afd2fe8SJohannes Thumshirn read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), 24144afd2fe8SJohannes Thumshirn sizeof(bgi)); 24154afd2fe8SJohannes Thumshirn 24164afd2fe8SJohannes Thumshirn btrfs_item_key_to_cpu(leaf, &key, slot); 24174afd2fe8SJohannes Thumshirn btrfs_release_path(path); 24184afd2fe8SJohannes Thumshirn ret = read_one_block_group(info, &bgi, &key, need_clear); 2419ffb9e0f0SQu Wenruo if (ret < 0) 24204358d963SJosef Bacik goto error; 2421ffb9e0f0SQu Wenruo key.objectid += key.offset; 2422ffb9e0f0SQu Wenruo key.offset = 0; 24234358d963SJosef Bacik } 24247837fa88SJosef Bacik btrfs_release_path(path); 24254358d963SJosef Bacik 242672804905SJosef Bacik list_for_each_entry(space_info, &info->space_info, list) { 242749ea112dSJosef Bacik int i; 242849ea112dSJosef Bacik 242949ea112dSJosef Bacik for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 243049ea112dSJosef Bacik if (list_empty(&space_info->block_groups[i])) 243149ea112dSJosef Bacik continue; 243249ea112dSJosef Bacik cache = list_first_entry(&space_info->block_groups[i], 243349ea112dSJosef Bacik struct btrfs_block_group, 243449ea112dSJosef Bacik list); 243549ea112dSJosef Bacik btrfs_sysfs_add_block_group_type(cache); 243649ea112dSJosef Bacik } 243749ea112dSJosef Bacik 24384358d963SJosef Bacik if (!(btrfs_get_alloc_profile(info, space_info->flags) & 24394358d963SJosef Bacik (BTRFS_BLOCK_GROUP_RAID10 | 24404358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID1_MASK | 24414358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID56_MASK | 24424358d963SJosef Bacik BTRFS_BLOCK_GROUP_DUP))) 24434358d963SJosef Bacik continue; 24444358d963SJosef Bacik /* 24454358d963SJosef Bacik * Avoid allocating from un-mirrored block group if there are 24464358d963SJosef Bacik * mirrored block groups. 24474358d963SJosef Bacik */ 24484358d963SJosef Bacik list_for_each_entry(cache, 24494358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_RAID0], 24504358d963SJosef Bacik list) 2451e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 24524358d963SJosef Bacik list_for_each_entry(cache, 24534358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_SINGLE], 24544358d963SJosef Bacik list) 2455e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 24564358d963SJosef Bacik } 24574358d963SJosef Bacik 24584358d963SJosef Bacik btrfs_init_global_block_rsv(info); 24594358d963SJosef Bacik ret = check_chunk_block_group_mappings(info); 24604358d963SJosef Bacik error: 24614358d963SJosef Bacik btrfs_free_path(path); 24622b29726cSQu Wenruo /* 24632b29726cSQu Wenruo * We've hit some error while reading the extent tree, and have 24642b29726cSQu Wenruo * rescue=ibadroots mount option. 24652b29726cSQu Wenruo * Try to fill the tree using dummy block groups so that the user can 24662b29726cSQu Wenruo * continue to mount and grab their data. 24672b29726cSQu Wenruo */ 24682b29726cSQu Wenruo if (ret && btrfs_test_opt(info, IGNOREBADROOTS)) 24692b29726cSQu Wenruo ret = fill_dummy_bgs(info); 24704358d963SJosef Bacik return ret; 24714358d963SJosef Bacik } 24724358d963SJosef Bacik 247379bd3712SFilipe Manana /* 247479bd3712SFilipe Manana * This function, insert_block_group_item(), belongs to the phase 2 of chunk 247579bd3712SFilipe Manana * allocation. 247679bd3712SFilipe Manana * 247779bd3712SFilipe Manana * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 247879bd3712SFilipe Manana * phases. 247979bd3712SFilipe Manana */ 248097f4728aSQu Wenruo static int insert_block_group_item(struct btrfs_trans_handle *trans, 248197f4728aSQu Wenruo struct btrfs_block_group *block_group) 248297f4728aSQu Wenruo { 248397f4728aSQu Wenruo struct btrfs_fs_info *fs_info = trans->fs_info; 248497f4728aSQu Wenruo struct btrfs_block_group_item bgi; 2485dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 248697f4728aSQu Wenruo struct btrfs_key key; 2487*675dfe12SFilipe Manana u64 old_commit_used; 2488*675dfe12SFilipe Manana int ret; 248997f4728aSQu Wenruo 249097f4728aSQu Wenruo spin_lock(&block_group->lock); 249197f4728aSQu Wenruo btrfs_set_stack_block_group_used(&bgi, block_group->used); 249297f4728aSQu Wenruo btrfs_set_stack_block_group_chunk_objectid(&bgi, 2493f7238e50SJosef Bacik block_group->global_root_id); 249497f4728aSQu Wenruo btrfs_set_stack_block_group_flags(&bgi, block_group->flags); 2495*675dfe12SFilipe Manana old_commit_used = block_group->commit_used; 2496*675dfe12SFilipe Manana block_group->commit_used = block_group->used; 249797f4728aSQu Wenruo key.objectid = block_group->start; 249897f4728aSQu Wenruo key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 249997f4728aSQu Wenruo key.offset = block_group->length; 250097f4728aSQu Wenruo spin_unlock(&block_group->lock); 250197f4728aSQu Wenruo 2502*675dfe12SFilipe Manana ret = btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi)); 2503*675dfe12SFilipe Manana if (ret < 0) { 2504*675dfe12SFilipe Manana spin_lock(&block_group->lock); 2505*675dfe12SFilipe Manana block_group->commit_used = old_commit_used; 2506*675dfe12SFilipe Manana spin_unlock(&block_group->lock); 2507*675dfe12SFilipe Manana } 2508*675dfe12SFilipe Manana 2509*675dfe12SFilipe Manana return ret; 251097f4728aSQu Wenruo } 251197f4728aSQu Wenruo 25122eadb9e7SNikolay Borisov static int insert_dev_extent(struct btrfs_trans_handle *trans, 25132eadb9e7SNikolay Borisov struct btrfs_device *device, u64 chunk_offset, 25142eadb9e7SNikolay Borisov u64 start, u64 num_bytes) 25152eadb9e7SNikolay Borisov { 25162eadb9e7SNikolay Borisov struct btrfs_fs_info *fs_info = device->fs_info; 25172eadb9e7SNikolay Borisov struct btrfs_root *root = fs_info->dev_root; 25182eadb9e7SNikolay Borisov struct btrfs_path *path; 25192eadb9e7SNikolay Borisov struct btrfs_dev_extent *extent; 25202eadb9e7SNikolay Borisov struct extent_buffer *leaf; 25212eadb9e7SNikolay Borisov struct btrfs_key key; 25222eadb9e7SNikolay Borisov int ret; 25232eadb9e7SNikolay Borisov 25242eadb9e7SNikolay Borisov WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); 25252eadb9e7SNikolay Borisov WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 25262eadb9e7SNikolay Borisov path = btrfs_alloc_path(); 25272eadb9e7SNikolay Borisov if (!path) 25282eadb9e7SNikolay Borisov return -ENOMEM; 25292eadb9e7SNikolay Borisov 25302eadb9e7SNikolay Borisov key.objectid = device->devid; 25312eadb9e7SNikolay Borisov key.type = BTRFS_DEV_EXTENT_KEY; 25322eadb9e7SNikolay Borisov key.offset = start; 25332eadb9e7SNikolay Borisov ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*extent)); 25342eadb9e7SNikolay Borisov if (ret) 25352eadb9e7SNikolay Borisov goto out; 25362eadb9e7SNikolay Borisov 25372eadb9e7SNikolay Borisov leaf = path->nodes[0]; 25382eadb9e7SNikolay Borisov extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); 25392eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID); 25402eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_objectid(leaf, extent, 25412eadb9e7SNikolay Borisov BTRFS_FIRST_CHUNK_TREE_OBJECTID); 25422eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); 25432eadb9e7SNikolay Borisov 25442eadb9e7SNikolay Borisov btrfs_set_dev_extent_length(leaf, extent, num_bytes); 25452eadb9e7SNikolay Borisov btrfs_mark_buffer_dirty(leaf); 25462eadb9e7SNikolay Borisov out: 25472eadb9e7SNikolay Borisov btrfs_free_path(path); 25482eadb9e7SNikolay Borisov return ret; 25492eadb9e7SNikolay Borisov } 25502eadb9e7SNikolay Borisov 25512eadb9e7SNikolay Borisov /* 25522eadb9e7SNikolay Borisov * This function belongs to phase 2. 25532eadb9e7SNikolay Borisov * 25542eadb9e7SNikolay Borisov * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 25552eadb9e7SNikolay Borisov * phases. 25562eadb9e7SNikolay Borisov */ 25572eadb9e7SNikolay Borisov static int insert_dev_extents(struct btrfs_trans_handle *trans, 25582eadb9e7SNikolay Borisov u64 chunk_offset, u64 chunk_size) 25592eadb9e7SNikolay Borisov { 25602eadb9e7SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 25612eadb9e7SNikolay Borisov struct btrfs_device *device; 25622eadb9e7SNikolay Borisov struct extent_map *em; 25632eadb9e7SNikolay Borisov struct map_lookup *map; 25642eadb9e7SNikolay Borisov u64 dev_offset; 25652eadb9e7SNikolay Borisov u64 stripe_size; 25662eadb9e7SNikolay Borisov int i; 25672eadb9e7SNikolay Borisov int ret = 0; 25682eadb9e7SNikolay Borisov 25692eadb9e7SNikolay Borisov em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); 25702eadb9e7SNikolay Borisov if (IS_ERR(em)) 25712eadb9e7SNikolay Borisov return PTR_ERR(em); 25722eadb9e7SNikolay Borisov 25732eadb9e7SNikolay Borisov map = em->map_lookup; 25742eadb9e7SNikolay Borisov stripe_size = em->orig_block_len; 25752eadb9e7SNikolay Borisov 25762eadb9e7SNikolay Borisov /* 25772eadb9e7SNikolay Borisov * Take the device list mutex to prevent races with the final phase of 25782eadb9e7SNikolay Borisov * a device replace operation that replaces the device object associated 25792eadb9e7SNikolay Borisov * with the map's stripes, because the device object's id can change 25802eadb9e7SNikolay Borisov * at any time during that final phase of the device replace operation 25812eadb9e7SNikolay Borisov * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 25822eadb9e7SNikolay Borisov * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 25832eadb9e7SNikolay Borisov * resulting in persisting a device extent item with such ID. 25842eadb9e7SNikolay Borisov */ 25852eadb9e7SNikolay Borisov mutex_lock(&fs_info->fs_devices->device_list_mutex); 25862eadb9e7SNikolay Borisov for (i = 0; i < map->num_stripes; i++) { 25872eadb9e7SNikolay Borisov device = map->stripes[i].dev; 25882eadb9e7SNikolay Borisov dev_offset = map->stripes[i].physical; 25892eadb9e7SNikolay Borisov 25902eadb9e7SNikolay Borisov ret = insert_dev_extent(trans, device, chunk_offset, dev_offset, 25912eadb9e7SNikolay Borisov stripe_size); 25922eadb9e7SNikolay Borisov if (ret) 25932eadb9e7SNikolay Borisov break; 25942eadb9e7SNikolay Borisov } 25952eadb9e7SNikolay Borisov mutex_unlock(&fs_info->fs_devices->device_list_mutex); 25962eadb9e7SNikolay Borisov 25972eadb9e7SNikolay Borisov free_extent_map(em); 25982eadb9e7SNikolay Borisov return ret; 25992eadb9e7SNikolay Borisov } 26002eadb9e7SNikolay Borisov 260179bd3712SFilipe Manana /* 260279bd3712SFilipe Manana * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of 260379bd3712SFilipe Manana * chunk allocation. 260479bd3712SFilipe Manana * 260579bd3712SFilipe Manana * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 260679bd3712SFilipe Manana * phases. 260779bd3712SFilipe Manana */ 26084358d963SJosef Bacik void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) 26094358d963SJosef Bacik { 26104358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 261132da5386SDavid Sterba struct btrfs_block_group *block_group; 26124358d963SJosef Bacik int ret = 0; 26134358d963SJosef Bacik 26144358d963SJosef Bacik while (!list_empty(&trans->new_bgs)) { 261549ea112dSJosef Bacik int index; 261649ea112dSJosef Bacik 26174358d963SJosef Bacik block_group = list_first_entry(&trans->new_bgs, 261832da5386SDavid Sterba struct btrfs_block_group, 26194358d963SJosef Bacik bg_list); 26204358d963SJosef Bacik if (ret) 26214358d963SJosef Bacik goto next; 26224358d963SJosef Bacik 262349ea112dSJosef Bacik index = btrfs_bg_flags_to_raid_index(block_group->flags); 262449ea112dSJosef Bacik 262597f4728aSQu Wenruo ret = insert_block_group_item(trans, block_group); 26264358d963SJosef Bacik if (ret) 26274358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 26283349b57fSJosef Bacik if (!test_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, 26293349b57fSJosef Bacik &block_group->runtime_flags)) { 263079bd3712SFilipe Manana mutex_lock(&fs_info->chunk_mutex); 263179bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, block_group); 263279bd3712SFilipe Manana mutex_unlock(&fs_info->chunk_mutex); 263379bd3712SFilipe Manana if (ret) 263479bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 263579bd3712SFilipe Manana } 26362eadb9e7SNikolay Borisov ret = insert_dev_extents(trans, block_group->start, 263797f4728aSQu Wenruo block_group->length); 26384358d963SJosef Bacik if (ret) 26394358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 26404358d963SJosef Bacik add_block_group_free_space(trans, block_group); 264149ea112dSJosef Bacik 264249ea112dSJosef Bacik /* 264349ea112dSJosef Bacik * If we restriped during balance, we may have added a new raid 264449ea112dSJosef Bacik * type, so now add the sysfs entries when it is safe to do so. 264549ea112dSJosef Bacik * We don't have to worry about locking here as it's handled in 264649ea112dSJosef Bacik * btrfs_sysfs_add_block_group_type. 264749ea112dSJosef Bacik */ 264849ea112dSJosef Bacik if (block_group->space_info->block_group_kobjs[index] == NULL) 264949ea112dSJosef Bacik btrfs_sysfs_add_block_group_type(block_group); 265049ea112dSJosef Bacik 26514358d963SJosef Bacik /* Already aborted the transaction if it failed. */ 26524358d963SJosef Bacik next: 26534358d963SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 26544358d963SJosef Bacik list_del_init(&block_group->bg_list); 26554358d963SJosef Bacik } 26564358d963SJosef Bacik btrfs_trans_release_chunk_metadata(trans); 26574358d963SJosef Bacik } 26584358d963SJosef Bacik 2659f7238e50SJosef Bacik /* 2660f7238e50SJosef Bacik * For extent tree v2 we use the block_group_item->chunk_offset to point at our 2661f7238e50SJosef Bacik * global root id. For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID. 2662f7238e50SJosef Bacik */ 2663f7238e50SJosef Bacik static u64 calculate_global_root_id(struct btrfs_fs_info *fs_info, u64 offset) 2664f7238e50SJosef Bacik { 2665f7238e50SJosef Bacik u64 div = SZ_1G; 2666f7238e50SJosef Bacik u64 index; 2667f7238e50SJosef Bacik 2668f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 2669f7238e50SJosef Bacik return BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2670f7238e50SJosef Bacik 2671f7238e50SJosef Bacik /* If we have a smaller fs index based on 128MiB. */ 2672f7238e50SJosef Bacik if (btrfs_super_total_bytes(fs_info->super_copy) <= (SZ_1G * 10ULL)) 2673f7238e50SJosef Bacik div = SZ_128M; 2674f7238e50SJosef Bacik 2675f7238e50SJosef Bacik offset = div64_u64(offset, div); 2676f7238e50SJosef Bacik div64_u64_rem(offset, fs_info->nr_global_roots, &index); 2677f7238e50SJosef Bacik return index; 2678f7238e50SJosef Bacik } 2679f7238e50SJosef Bacik 268079bd3712SFilipe Manana struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans, 268179bd3712SFilipe Manana u64 bytes_used, u64 type, 268279bd3712SFilipe Manana u64 chunk_offset, u64 size) 26834358d963SJosef Bacik { 26844358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 268532da5386SDavid Sterba struct btrfs_block_group *cache; 26864358d963SJosef Bacik int ret; 26874358d963SJosef Bacik 26884358d963SJosef Bacik btrfs_set_log_full_commit(trans); 26894358d963SJosef Bacik 26909afc6649SQu Wenruo cache = btrfs_create_block_group_cache(fs_info, chunk_offset); 26914358d963SJosef Bacik if (!cache) 269279bd3712SFilipe Manana return ERR_PTR(-ENOMEM); 26934358d963SJosef Bacik 26949afc6649SQu Wenruo cache->length = size; 2695e3e39c72SMarcos Paulo de Souza set_free_space_tree_thresholds(cache); 2696bf38be65SDavid Sterba cache->used = bytes_used; 26974358d963SJosef Bacik cache->flags = type; 26984358d963SJosef Bacik cache->cached = BTRFS_CACHE_FINISHED; 2699f7238e50SJosef Bacik cache->global_root_id = calculate_global_root_id(fs_info, cache->start); 2700f7238e50SJosef Bacik 2701997e3e2eSBoris Burkov if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) 27020d7764ffSDavid Sterba set_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &cache->runtime_flags); 270308e11a3dSNaohiro Aota 2704a94794d5SNaohiro Aota ret = btrfs_load_block_group_zone_info(cache, true); 270508e11a3dSNaohiro Aota if (ret) { 270608e11a3dSNaohiro Aota btrfs_put_block_group(cache); 270779bd3712SFilipe Manana return ERR_PTR(ret); 270808e11a3dSNaohiro Aota } 270908e11a3dSNaohiro Aota 27104358d963SJosef Bacik ret = exclude_super_stripes(cache); 27114358d963SJosef Bacik if (ret) { 27124358d963SJosef Bacik /* We may have excluded something, so call this just in case */ 27134358d963SJosef Bacik btrfs_free_excluded_extents(cache); 27144358d963SJosef Bacik btrfs_put_block_group(cache); 271579bd3712SFilipe Manana return ERR_PTR(ret); 27164358d963SJosef Bacik } 27174358d963SJosef Bacik 27184358d963SJosef Bacik add_new_free_space(cache, chunk_offset, chunk_offset + size); 27194358d963SJosef Bacik 27204358d963SJosef Bacik btrfs_free_excluded_extents(cache); 27214358d963SJosef Bacik 27224358d963SJosef Bacik /* 27234358d963SJosef Bacik * Ensure the corresponding space_info object is created and 27244358d963SJosef Bacik * assigned to our block group. We want our bg to be added to the rbtree 27254358d963SJosef Bacik * with its ->space_info set. 27264358d963SJosef Bacik */ 27274358d963SJosef Bacik cache->space_info = btrfs_find_space_info(fs_info, cache->flags); 27284358d963SJosef Bacik ASSERT(cache->space_info); 27294358d963SJosef Bacik 27304358d963SJosef Bacik ret = btrfs_add_block_group_cache(fs_info, cache); 27314358d963SJosef Bacik if (ret) { 27324358d963SJosef Bacik btrfs_remove_free_space_cache(cache); 27334358d963SJosef Bacik btrfs_put_block_group(cache); 273479bd3712SFilipe Manana return ERR_PTR(ret); 27354358d963SJosef Bacik } 27364358d963SJosef Bacik 27374358d963SJosef Bacik /* 27384358d963SJosef Bacik * Now that our block group has its ->space_info set and is inserted in 27394358d963SJosef Bacik * the rbtree, update the space info's counters. 27404358d963SJosef Bacik */ 27414358d963SJosef Bacik trace_btrfs_add_block_group(fs_info, cache, 1); 2742723de71dSJosef Bacik btrfs_add_bg_to_space_info(fs_info, cache); 27434358d963SJosef Bacik btrfs_update_global_block_rsv(fs_info); 27444358d963SJosef Bacik 27459d4b0a12SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 27469d4b0a12SJosef Bacik if (btrfs_should_fragment_free_space(cache)) { 27479d4b0a12SJosef Bacik u64 new_bytes_used = size - bytes_used; 27489d4b0a12SJosef Bacik 27499d4b0a12SJosef Bacik cache->space_info->bytes_used += new_bytes_used >> 1; 27509d4b0a12SJosef Bacik fragment_free_space(cache); 27519d4b0a12SJosef Bacik } 27529d4b0a12SJosef Bacik #endif 27534358d963SJosef Bacik 27544358d963SJosef Bacik list_add_tail(&cache->bg_list, &trans->new_bgs); 27554358d963SJosef Bacik trans->delayed_ref_updates++; 27564358d963SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 27574358d963SJosef Bacik 27584358d963SJosef Bacik set_avail_alloc_bits(fs_info, type); 275979bd3712SFilipe Manana return cache; 27604358d963SJosef Bacik } 276126ce2095SJosef Bacik 2762b12de528SQu Wenruo /* 2763b12de528SQu Wenruo * Mark one block group RO, can be called several times for the same block 2764b12de528SQu Wenruo * group. 2765b12de528SQu Wenruo * 2766b12de528SQu Wenruo * @cache: the destination block group 2767b12de528SQu Wenruo * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to 2768b12de528SQu Wenruo * ensure we still have some free space after marking this 2769b12de528SQu Wenruo * block group RO. 2770b12de528SQu Wenruo */ 2771b12de528SQu Wenruo int btrfs_inc_block_group_ro(struct btrfs_block_group *cache, 2772b12de528SQu Wenruo bool do_chunk_alloc) 277326ce2095SJosef Bacik { 277426ce2095SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 277526ce2095SJosef Bacik struct btrfs_trans_handle *trans; 2776dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 277726ce2095SJosef Bacik u64 alloc_flags; 277826ce2095SJosef Bacik int ret; 2779b6e9f16cSNikolay Borisov bool dirty_bg_running; 278026ce2095SJosef Bacik 27812d192fc4SQu Wenruo /* 27822d192fc4SQu Wenruo * This can only happen when we are doing read-only scrub on read-only 27832d192fc4SQu Wenruo * mount. 27842d192fc4SQu Wenruo * In that case we should not start a new transaction on read-only fs. 27852d192fc4SQu Wenruo * Thus here we skip all chunk allocations. 27862d192fc4SQu Wenruo */ 27872d192fc4SQu Wenruo if (sb_rdonly(fs_info->sb)) { 27882d192fc4SQu Wenruo mutex_lock(&fs_info->ro_block_group_mutex); 27892d192fc4SQu Wenruo ret = inc_block_group_ro(cache, 0); 27902d192fc4SQu Wenruo mutex_unlock(&fs_info->ro_block_group_mutex); 27912d192fc4SQu Wenruo return ret; 27922d192fc4SQu Wenruo } 27932d192fc4SQu Wenruo 2794b6e9f16cSNikolay Borisov do { 2795dfe8aec4SJosef Bacik trans = btrfs_join_transaction(root); 279626ce2095SJosef Bacik if (IS_ERR(trans)) 279726ce2095SJosef Bacik return PTR_ERR(trans); 279826ce2095SJosef Bacik 2799b6e9f16cSNikolay Borisov dirty_bg_running = false; 2800b6e9f16cSNikolay Borisov 280126ce2095SJosef Bacik /* 2802b6e9f16cSNikolay Borisov * We're not allowed to set block groups readonly after the dirty 2803b6e9f16cSNikolay Borisov * block group cache has started writing. If it already started, 2804b6e9f16cSNikolay Borisov * back off and let this transaction commit. 280526ce2095SJosef Bacik */ 280626ce2095SJosef Bacik mutex_lock(&fs_info->ro_block_group_mutex); 280726ce2095SJosef Bacik if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { 280826ce2095SJosef Bacik u64 transid = trans->transid; 280926ce2095SJosef Bacik 281026ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 281126ce2095SJosef Bacik btrfs_end_transaction(trans); 281226ce2095SJosef Bacik 281326ce2095SJosef Bacik ret = btrfs_wait_for_commit(fs_info, transid); 281426ce2095SJosef Bacik if (ret) 281526ce2095SJosef Bacik return ret; 2816b6e9f16cSNikolay Borisov dirty_bg_running = true; 281726ce2095SJosef Bacik } 2818b6e9f16cSNikolay Borisov } while (dirty_bg_running); 281926ce2095SJosef Bacik 2820b12de528SQu Wenruo if (do_chunk_alloc) { 282126ce2095SJosef Bacik /* 2822b12de528SQu Wenruo * If we are changing raid levels, try to allocate a 2823b12de528SQu Wenruo * corresponding block group with the new raid level. 282426ce2095SJosef Bacik */ 2825349e120eSJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); 282626ce2095SJosef Bacik if (alloc_flags != cache->flags) { 2827b12de528SQu Wenruo ret = btrfs_chunk_alloc(trans, alloc_flags, 2828b12de528SQu Wenruo CHUNK_ALLOC_FORCE); 282926ce2095SJosef Bacik /* 283026ce2095SJosef Bacik * ENOSPC is allowed here, we may have enough space 2831b12de528SQu Wenruo * already allocated at the new raid level to carry on 283226ce2095SJosef Bacik */ 283326ce2095SJosef Bacik if (ret == -ENOSPC) 283426ce2095SJosef Bacik ret = 0; 283526ce2095SJosef Bacik if (ret < 0) 283626ce2095SJosef Bacik goto out; 283726ce2095SJosef Bacik } 2838b12de528SQu Wenruo } 283926ce2095SJosef Bacik 2840a7a63accSJosef Bacik ret = inc_block_group_ro(cache, 0); 2841195a49eaSFilipe Manana if (!do_chunk_alloc || ret == -ETXTBSY) 2842b12de528SQu Wenruo goto unlock_out; 284326ce2095SJosef Bacik if (!ret) 284426ce2095SJosef Bacik goto out; 284526ce2095SJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); 284626ce2095SJosef Bacik ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 284726ce2095SJosef Bacik if (ret < 0) 284826ce2095SJosef Bacik goto out; 2849b6a98021SNaohiro Aota /* 2850b6a98021SNaohiro Aota * We have allocated a new chunk. We also need to activate that chunk to 2851b6a98021SNaohiro Aota * grant metadata tickets for zoned filesystem. 2852b6a98021SNaohiro Aota */ 2853b6a98021SNaohiro Aota ret = btrfs_zoned_activate_one_bg(fs_info, cache->space_info, true); 2854b6a98021SNaohiro Aota if (ret < 0) 2855b6a98021SNaohiro Aota goto out; 2856b6a98021SNaohiro Aota 2857e11c0406SJosef Bacik ret = inc_block_group_ro(cache, 0); 2858195a49eaSFilipe Manana if (ret == -ETXTBSY) 2859195a49eaSFilipe Manana goto unlock_out; 286026ce2095SJosef Bacik out: 286126ce2095SJosef Bacik if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { 2862349e120eSJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); 286326ce2095SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 286426ce2095SJosef Bacik check_system_chunk(trans, alloc_flags); 286526ce2095SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 286626ce2095SJosef Bacik } 2867b12de528SQu Wenruo unlock_out: 286826ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 286926ce2095SJosef Bacik 287026ce2095SJosef Bacik btrfs_end_transaction(trans); 287126ce2095SJosef Bacik return ret; 287226ce2095SJosef Bacik } 287326ce2095SJosef Bacik 287432da5386SDavid Sterba void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) 287526ce2095SJosef Bacik { 287626ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 287726ce2095SJosef Bacik u64 num_bytes; 287826ce2095SJosef Bacik 287926ce2095SJosef Bacik BUG_ON(!cache->ro); 288026ce2095SJosef Bacik 288126ce2095SJosef Bacik spin_lock(&sinfo->lock); 288226ce2095SJosef Bacik spin_lock(&cache->lock); 288326ce2095SJosef Bacik if (!--cache->ro) { 2884169e0da9SNaohiro Aota if (btrfs_is_zoned(cache->fs_info)) { 2885169e0da9SNaohiro Aota /* Migrate zone_unusable bytes back */ 288698173255SNaohiro Aota cache->zone_unusable = 288798173255SNaohiro Aota (cache->alloc_offset - cache->used) + 288898173255SNaohiro Aota (cache->length - cache->zone_capacity); 2889169e0da9SNaohiro Aota sinfo->bytes_zone_unusable += cache->zone_unusable; 2890169e0da9SNaohiro Aota sinfo->bytes_readonly -= cache->zone_unusable; 2891169e0da9SNaohiro Aota } 2892f9f28e5bSNaohiro Aota num_bytes = cache->length - cache->reserved - 2893f9f28e5bSNaohiro Aota cache->pinned - cache->bytes_super - 2894f9f28e5bSNaohiro Aota cache->zone_unusable - cache->used; 2895f9f28e5bSNaohiro Aota sinfo->bytes_readonly -= num_bytes; 289626ce2095SJosef Bacik list_del_init(&cache->ro_list); 289726ce2095SJosef Bacik } 289826ce2095SJosef Bacik spin_unlock(&cache->lock); 289926ce2095SJosef Bacik spin_unlock(&sinfo->lock); 290026ce2095SJosef Bacik } 290177745c05SJosef Bacik 29023be4d8efSQu Wenruo static int update_block_group_item(struct btrfs_trans_handle *trans, 290377745c05SJosef Bacik struct btrfs_path *path, 290432da5386SDavid Sterba struct btrfs_block_group *cache) 290577745c05SJosef Bacik { 290677745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 290777745c05SJosef Bacik int ret; 2908dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 290977745c05SJosef Bacik unsigned long bi; 291077745c05SJosef Bacik struct extent_buffer *leaf; 2911bf38be65SDavid Sterba struct btrfs_block_group_item bgi; 2912b3470b5dSDavid Sterba struct btrfs_key key; 29137248e0ceSQu Wenruo u64 old_commit_used; 29147248e0ceSQu Wenruo u64 used; 29157248e0ceSQu Wenruo 29167248e0ceSQu Wenruo /* 29177248e0ceSQu Wenruo * Block group items update can be triggered out of commit transaction 29187248e0ceSQu Wenruo * critical section, thus we need a consistent view of used bytes. 29197248e0ceSQu Wenruo * We cannot use cache->used directly outside of the spin lock, as it 29207248e0ceSQu Wenruo * may be changed. 29217248e0ceSQu Wenruo */ 29227248e0ceSQu Wenruo spin_lock(&cache->lock); 29237248e0ceSQu Wenruo old_commit_used = cache->commit_used; 29247248e0ceSQu Wenruo used = cache->used; 29257248e0ceSQu Wenruo /* No change in used bytes, can safely skip it. */ 29267248e0ceSQu Wenruo if (cache->commit_used == used) { 29277248e0ceSQu Wenruo spin_unlock(&cache->lock); 29287248e0ceSQu Wenruo return 0; 29297248e0ceSQu Wenruo } 29307248e0ceSQu Wenruo cache->commit_used = used; 29317248e0ceSQu Wenruo spin_unlock(&cache->lock); 293277745c05SJosef Bacik 2933b3470b5dSDavid Sterba key.objectid = cache->start; 2934b3470b5dSDavid Sterba key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 2935b3470b5dSDavid Sterba key.offset = cache->length; 2936b3470b5dSDavid Sterba 29373be4d8efSQu Wenruo ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 293877745c05SJosef Bacik if (ret) { 293977745c05SJosef Bacik if (ret > 0) 294077745c05SJosef Bacik ret = -ENOENT; 294177745c05SJosef Bacik goto fail; 294277745c05SJosef Bacik } 294377745c05SJosef Bacik 294477745c05SJosef Bacik leaf = path->nodes[0]; 294577745c05SJosef Bacik bi = btrfs_item_ptr_offset(leaf, path->slots[0]); 29467248e0ceSQu Wenruo btrfs_set_stack_block_group_used(&bgi, used); 2947de0dc456SDavid Sterba btrfs_set_stack_block_group_chunk_objectid(&bgi, 2948f7238e50SJosef Bacik cache->global_root_id); 2949de0dc456SDavid Sterba btrfs_set_stack_block_group_flags(&bgi, cache->flags); 2950bf38be65SDavid Sterba write_extent_buffer(leaf, &bgi, bi, sizeof(bgi)); 295177745c05SJosef Bacik btrfs_mark_buffer_dirty(leaf); 295277745c05SJosef Bacik fail: 295377745c05SJosef Bacik btrfs_release_path(path); 29547248e0ceSQu Wenruo /* We didn't update the block group item, need to revert @commit_used. */ 29557248e0ceSQu Wenruo if (ret < 0) { 29567248e0ceSQu Wenruo spin_lock(&cache->lock); 29577248e0ceSQu Wenruo cache->commit_used = old_commit_used; 29587248e0ceSQu Wenruo spin_unlock(&cache->lock); 29597248e0ceSQu Wenruo } 296077745c05SJosef Bacik return ret; 296177745c05SJosef Bacik 296277745c05SJosef Bacik } 296377745c05SJosef Bacik 296432da5386SDavid Sterba static int cache_save_setup(struct btrfs_block_group *block_group, 296577745c05SJosef Bacik struct btrfs_trans_handle *trans, 296677745c05SJosef Bacik struct btrfs_path *path) 296777745c05SJosef Bacik { 296877745c05SJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 296977745c05SJosef Bacik struct btrfs_root *root = fs_info->tree_root; 297077745c05SJosef Bacik struct inode *inode = NULL; 297177745c05SJosef Bacik struct extent_changeset *data_reserved = NULL; 297277745c05SJosef Bacik u64 alloc_hint = 0; 297377745c05SJosef Bacik int dcs = BTRFS_DC_ERROR; 29740044ae11SQu Wenruo u64 cache_size = 0; 297577745c05SJosef Bacik int retries = 0; 297677745c05SJosef Bacik int ret = 0; 297777745c05SJosef Bacik 2978af456a2cSBoris Burkov if (!btrfs_test_opt(fs_info, SPACE_CACHE)) 2979af456a2cSBoris Burkov return 0; 2980af456a2cSBoris Burkov 298177745c05SJosef Bacik /* 298277745c05SJosef Bacik * If this block group is smaller than 100 megs don't bother caching the 298377745c05SJosef Bacik * block group. 298477745c05SJosef Bacik */ 2985b3470b5dSDavid Sterba if (block_group->length < (100 * SZ_1M)) { 298677745c05SJosef Bacik spin_lock(&block_group->lock); 298777745c05SJosef Bacik block_group->disk_cache_state = BTRFS_DC_WRITTEN; 298877745c05SJosef Bacik spin_unlock(&block_group->lock); 298977745c05SJosef Bacik return 0; 299077745c05SJosef Bacik } 299177745c05SJosef Bacik 2992bf31f87fSDavid Sterba if (TRANS_ABORTED(trans)) 299377745c05SJosef Bacik return 0; 299477745c05SJosef Bacik again: 299577745c05SJosef Bacik inode = lookup_free_space_inode(block_group, path); 299677745c05SJosef Bacik if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { 299777745c05SJosef Bacik ret = PTR_ERR(inode); 299877745c05SJosef Bacik btrfs_release_path(path); 299977745c05SJosef Bacik goto out; 300077745c05SJosef Bacik } 300177745c05SJosef Bacik 300277745c05SJosef Bacik if (IS_ERR(inode)) { 300377745c05SJosef Bacik BUG_ON(retries); 300477745c05SJosef Bacik retries++; 300577745c05SJosef Bacik 300677745c05SJosef Bacik if (block_group->ro) 300777745c05SJosef Bacik goto out_free; 300877745c05SJosef Bacik 300977745c05SJosef Bacik ret = create_free_space_inode(trans, block_group, path); 301077745c05SJosef Bacik if (ret) 301177745c05SJosef Bacik goto out_free; 301277745c05SJosef Bacik goto again; 301377745c05SJosef Bacik } 301477745c05SJosef Bacik 301577745c05SJosef Bacik /* 301677745c05SJosef Bacik * We want to set the generation to 0, that way if anything goes wrong 301777745c05SJosef Bacik * from here on out we know not to trust this cache when we load up next 301877745c05SJosef Bacik * time. 301977745c05SJosef Bacik */ 302077745c05SJosef Bacik BTRFS_I(inode)->generation = 0; 30219a56fcd1SNikolay Borisov ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); 302277745c05SJosef Bacik if (ret) { 302377745c05SJosef Bacik /* 302477745c05SJosef Bacik * So theoretically we could recover from this, simply set the 302577745c05SJosef Bacik * super cache generation to 0 so we know to invalidate the 302677745c05SJosef Bacik * cache, but then we'd have to keep track of the block groups 302777745c05SJosef Bacik * that fail this way so we know we _have_ to reset this cache 302877745c05SJosef Bacik * before the next commit or risk reading stale cache. So to 302977745c05SJosef Bacik * limit our exposure to horrible edge cases lets just abort the 303077745c05SJosef Bacik * transaction, this only happens in really bad situations 303177745c05SJosef Bacik * anyway. 303277745c05SJosef Bacik */ 303377745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 303477745c05SJosef Bacik goto out_put; 303577745c05SJosef Bacik } 303677745c05SJosef Bacik WARN_ON(ret); 303777745c05SJosef Bacik 303877745c05SJosef Bacik /* We've already setup this transaction, go ahead and exit */ 303977745c05SJosef Bacik if (block_group->cache_generation == trans->transid && 304077745c05SJosef Bacik i_size_read(inode)) { 304177745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 304277745c05SJosef Bacik goto out_put; 304377745c05SJosef Bacik } 304477745c05SJosef Bacik 304577745c05SJosef Bacik if (i_size_read(inode) > 0) { 304677745c05SJosef Bacik ret = btrfs_check_trunc_cache_free_space(fs_info, 304777745c05SJosef Bacik &fs_info->global_block_rsv); 304877745c05SJosef Bacik if (ret) 304977745c05SJosef Bacik goto out_put; 305077745c05SJosef Bacik 305177745c05SJosef Bacik ret = btrfs_truncate_free_space_cache(trans, NULL, inode); 305277745c05SJosef Bacik if (ret) 305377745c05SJosef Bacik goto out_put; 305477745c05SJosef Bacik } 305577745c05SJosef Bacik 305677745c05SJosef Bacik spin_lock(&block_group->lock); 305777745c05SJosef Bacik if (block_group->cached != BTRFS_CACHE_FINISHED || 305877745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) { 305977745c05SJosef Bacik /* 306077745c05SJosef Bacik * don't bother trying to write stuff out _if_ 306177745c05SJosef Bacik * a) we're not cached, 306277745c05SJosef Bacik * b) we're with nospace_cache mount option, 306377745c05SJosef Bacik * c) we're with v2 space_cache (FREE_SPACE_TREE). 306477745c05SJosef Bacik */ 306577745c05SJosef Bacik dcs = BTRFS_DC_WRITTEN; 306677745c05SJosef Bacik spin_unlock(&block_group->lock); 306777745c05SJosef Bacik goto out_put; 306877745c05SJosef Bacik } 306977745c05SJosef Bacik spin_unlock(&block_group->lock); 307077745c05SJosef Bacik 307177745c05SJosef Bacik /* 307277745c05SJosef Bacik * We hit an ENOSPC when setting up the cache in this transaction, just 307377745c05SJosef Bacik * skip doing the setup, we've already cleared the cache so we're safe. 307477745c05SJosef Bacik */ 307577745c05SJosef Bacik if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { 307677745c05SJosef Bacik ret = -ENOSPC; 307777745c05SJosef Bacik goto out_put; 307877745c05SJosef Bacik } 307977745c05SJosef Bacik 308077745c05SJosef Bacik /* 308177745c05SJosef Bacik * Try to preallocate enough space based on how big the block group is. 308277745c05SJosef Bacik * Keep in mind this has to include any pinned space which could end up 308377745c05SJosef Bacik * taking up quite a bit since it's not folded into the other space 308477745c05SJosef Bacik * cache. 308577745c05SJosef Bacik */ 30860044ae11SQu Wenruo cache_size = div_u64(block_group->length, SZ_256M); 30870044ae11SQu Wenruo if (!cache_size) 30880044ae11SQu Wenruo cache_size = 1; 308977745c05SJosef Bacik 30900044ae11SQu Wenruo cache_size *= 16; 30910044ae11SQu Wenruo cache_size *= fs_info->sectorsize; 309277745c05SJosef Bacik 309336ea6f3eSNikolay Borisov ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0, 30941daedb1dSJosef Bacik cache_size, false); 309577745c05SJosef Bacik if (ret) 309677745c05SJosef Bacik goto out_put; 309777745c05SJosef Bacik 30980044ae11SQu Wenruo ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size, 30990044ae11SQu Wenruo cache_size, cache_size, 310077745c05SJosef Bacik &alloc_hint); 310177745c05SJosef Bacik /* 310277745c05SJosef Bacik * Our cache requires contiguous chunks so that we don't modify a bunch 310377745c05SJosef Bacik * of metadata or split extents when writing the cache out, which means 310477745c05SJosef Bacik * we can enospc if we are heavily fragmented in addition to just normal 310577745c05SJosef Bacik * out of space conditions. So if we hit this just skip setting up any 310677745c05SJosef Bacik * other block groups for this transaction, maybe we'll unpin enough 310777745c05SJosef Bacik * space the next time around. 310877745c05SJosef Bacik */ 310977745c05SJosef Bacik if (!ret) 311077745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 311177745c05SJosef Bacik else if (ret == -ENOSPC) 311277745c05SJosef Bacik set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); 311377745c05SJosef Bacik 311477745c05SJosef Bacik out_put: 311577745c05SJosef Bacik iput(inode); 311677745c05SJosef Bacik out_free: 311777745c05SJosef Bacik btrfs_release_path(path); 311877745c05SJosef Bacik out: 311977745c05SJosef Bacik spin_lock(&block_group->lock); 312077745c05SJosef Bacik if (!ret && dcs == BTRFS_DC_SETUP) 312177745c05SJosef Bacik block_group->cache_generation = trans->transid; 312277745c05SJosef Bacik block_group->disk_cache_state = dcs; 312377745c05SJosef Bacik spin_unlock(&block_group->lock); 312477745c05SJosef Bacik 312577745c05SJosef Bacik extent_changeset_free(data_reserved); 312677745c05SJosef Bacik return ret; 312777745c05SJosef Bacik } 312877745c05SJosef Bacik 312977745c05SJosef Bacik int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) 313077745c05SJosef Bacik { 313177745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 313232da5386SDavid Sterba struct btrfs_block_group *cache, *tmp; 313377745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 313477745c05SJosef Bacik struct btrfs_path *path; 313577745c05SJosef Bacik 313677745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs) || 313777745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) 313877745c05SJosef Bacik return 0; 313977745c05SJosef Bacik 314077745c05SJosef Bacik path = btrfs_alloc_path(); 314177745c05SJosef Bacik if (!path) 314277745c05SJosef Bacik return -ENOMEM; 314377745c05SJosef Bacik 314477745c05SJosef Bacik /* Could add new block groups, use _safe just in case */ 314577745c05SJosef Bacik list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, 314677745c05SJosef Bacik dirty_list) { 314777745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_CLEAR) 314877745c05SJosef Bacik cache_save_setup(cache, trans, path); 314977745c05SJosef Bacik } 315077745c05SJosef Bacik 315177745c05SJosef Bacik btrfs_free_path(path); 315277745c05SJosef Bacik return 0; 315377745c05SJosef Bacik } 315477745c05SJosef Bacik 315577745c05SJosef Bacik /* 315677745c05SJosef Bacik * Transaction commit does final block group cache writeback during a critical 315777745c05SJosef Bacik * section where nothing is allowed to change the FS. This is required in 315877745c05SJosef Bacik * order for the cache to actually match the block group, but can introduce a 315977745c05SJosef Bacik * lot of latency into the commit. 316077745c05SJosef Bacik * 316177745c05SJosef Bacik * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. 316277745c05SJosef Bacik * There's a chance we'll have to redo some of it if the block group changes 316377745c05SJosef Bacik * again during the commit, but it greatly reduces the commit latency by 316477745c05SJosef Bacik * getting rid of the easy block groups while we're still allowing others to 316577745c05SJosef Bacik * join the commit. 316677745c05SJosef Bacik */ 316777745c05SJosef Bacik int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) 316877745c05SJosef Bacik { 316977745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 317032da5386SDavid Sterba struct btrfs_block_group *cache; 317177745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 317277745c05SJosef Bacik int ret = 0; 317377745c05SJosef Bacik int should_put; 317477745c05SJosef Bacik struct btrfs_path *path = NULL; 317577745c05SJosef Bacik LIST_HEAD(dirty); 317677745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 317777745c05SJosef Bacik int loops = 0; 317877745c05SJosef Bacik 317977745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 318077745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs)) { 318177745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 318277745c05SJosef Bacik return 0; 318377745c05SJosef Bacik } 318477745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 318577745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 318677745c05SJosef Bacik 318777745c05SJosef Bacik again: 318877745c05SJosef Bacik /* Make sure all the block groups on our dirty list actually exist */ 318977745c05SJosef Bacik btrfs_create_pending_block_groups(trans); 319077745c05SJosef Bacik 319177745c05SJosef Bacik if (!path) { 319277745c05SJosef Bacik path = btrfs_alloc_path(); 3193938fcbfbSJosef Bacik if (!path) { 3194938fcbfbSJosef Bacik ret = -ENOMEM; 3195938fcbfbSJosef Bacik goto out; 3196938fcbfbSJosef Bacik } 319777745c05SJosef Bacik } 319877745c05SJosef Bacik 319977745c05SJosef Bacik /* 320077745c05SJosef Bacik * cache_write_mutex is here only to save us from balance or automatic 320177745c05SJosef Bacik * removal of empty block groups deleting this block group while we are 320277745c05SJosef Bacik * writing out the cache 320377745c05SJosef Bacik */ 320477745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 320577745c05SJosef Bacik while (!list_empty(&dirty)) { 320677745c05SJosef Bacik bool drop_reserve = true; 320777745c05SJosef Bacik 320832da5386SDavid Sterba cache = list_first_entry(&dirty, struct btrfs_block_group, 320977745c05SJosef Bacik dirty_list); 321077745c05SJosef Bacik /* 321177745c05SJosef Bacik * This can happen if something re-dirties a block group that 321277745c05SJosef Bacik * is already under IO. Just wait for it to finish and then do 321377745c05SJosef Bacik * it all again 321477745c05SJosef Bacik */ 321577745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 321677745c05SJosef Bacik list_del_init(&cache->io_list); 321777745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 321877745c05SJosef Bacik btrfs_put_block_group(cache); 321977745c05SJosef Bacik } 322077745c05SJosef Bacik 322177745c05SJosef Bacik 322277745c05SJosef Bacik /* 322377745c05SJosef Bacik * btrfs_wait_cache_io uses the cache->dirty_list to decide if 322477745c05SJosef Bacik * it should update the cache_state. Don't delete until after 322577745c05SJosef Bacik * we wait. 322677745c05SJosef Bacik * 322777745c05SJosef Bacik * Since we're not running in the commit critical section 322877745c05SJosef Bacik * we need the dirty_bgs_lock to protect from update_block_group 322977745c05SJosef Bacik */ 323077745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 323177745c05SJosef Bacik list_del_init(&cache->dirty_list); 323277745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 323377745c05SJosef Bacik 323477745c05SJosef Bacik should_put = 1; 323577745c05SJosef Bacik 323677745c05SJosef Bacik cache_save_setup(cache, trans, path); 323777745c05SJosef Bacik 323877745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_SETUP) { 323977745c05SJosef Bacik cache->io_ctl.inode = NULL; 324077745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 324177745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 324277745c05SJosef Bacik should_put = 0; 324377745c05SJosef Bacik 324477745c05SJosef Bacik /* 324577745c05SJosef Bacik * The cache_write_mutex is protecting the 324677745c05SJosef Bacik * io_list, also refer to the definition of 324777745c05SJosef Bacik * btrfs_transaction::io_bgs for more details 324877745c05SJosef Bacik */ 324977745c05SJosef Bacik list_add_tail(&cache->io_list, io); 325077745c05SJosef Bacik } else { 325177745c05SJosef Bacik /* 325277745c05SJosef Bacik * If we failed to write the cache, the 325377745c05SJosef Bacik * generation will be bad and life goes on 325477745c05SJosef Bacik */ 325577745c05SJosef Bacik ret = 0; 325677745c05SJosef Bacik } 325777745c05SJosef Bacik } 325877745c05SJosef Bacik if (!ret) { 32593be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 326077745c05SJosef Bacik /* 326177745c05SJosef Bacik * Our block group might still be attached to the list 326277745c05SJosef Bacik * of new block groups in the transaction handle of some 326377745c05SJosef Bacik * other task (struct btrfs_trans_handle->new_bgs). This 326477745c05SJosef Bacik * means its block group item isn't yet in the extent 326577745c05SJosef Bacik * tree. If this happens ignore the error, as we will 326677745c05SJosef Bacik * try again later in the critical section of the 326777745c05SJosef Bacik * transaction commit. 326877745c05SJosef Bacik */ 326977745c05SJosef Bacik if (ret == -ENOENT) { 327077745c05SJosef Bacik ret = 0; 327177745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 327277745c05SJosef Bacik if (list_empty(&cache->dirty_list)) { 327377745c05SJosef Bacik list_add_tail(&cache->dirty_list, 327477745c05SJosef Bacik &cur_trans->dirty_bgs); 327577745c05SJosef Bacik btrfs_get_block_group(cache); 327677745c05SJosef Bacik drop_reserve = false; 327777745c05SJosef Bacik } 327877745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 327977745c05SJosef Bacik } else if (ret) { 328077745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 328177745c05SJosef Bacik } 328277745c05SJosef Bacik } 328377745c05SJosef Bacik 328477745c05SJosef Bacik /* If it's not on the io list, we need to put the block group */ 328577745c05SJosef Bacik if (should_put) 328677745c05SJosef Bacik btrfs_put_block_group(cache); 328777745c05SJosef Bacik if (drop_reserve) 328877745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 328977745c05SJosef Bacik /* 329077745c05SJosef Bacik * Avoid blocking other tasks for too long. It might even save 329177745c05SJosef Bacik * us from writing caches for block groups that are going to be 329277745c05SJosef Bacik * removed. 329377745c05SJosef Bacik */ 329477745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 3295938fcbfbSJosef Bacik if (ret) 3296938fcbfbSJosef Bacik goto out; 329777745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 329877745c05SJosef Bacik } 329977745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 330077745c05SJosef Bacik 330177745c05SJosef Bacik /* 330277745c05SJosef Bacik * Go through delayed refs for all the stuff we've just kicked off 330377745c05SJosef Bacik * and then loop back (just once) 330477745c05SJosef Bacik */ 330534d1eb0eSJosef Bacik if (!ret) 330677745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 0); 330777745c05SJosef Bacik if (!ret && loops == 0) { 330877745c05SJosef Bacik loops++; 330977745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 331077745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 331177745c05SJosef Bacik /* 331277745c05SJosef Bacik * dirty_bgs_lock protects us from concurrent block group 331377745c05SJosef Bacik * deletes too (not just cache_write_mutex). 331477745c05SJosef Bacik */ 331577745c05SJosef Bacik if (!list_empty(&dirty)) { 331677745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 331777745c05SJosef Bacik goto again; 331877745c05SJosef Bacik } 331977745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 3320938fcbfbSJosef Bacik } 3321938fcbfbSJosef Bacik out: 3322938fcbfbSJosef Bacik if (ret < 0) { 3323938fcbfbSJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 3324938fcbfbSJosef Bacik list_splice_init(&dirty, &cur_trans->dirty_bgs); 3325938fcbfbSJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 332677745c05SJosef Bacik btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 332777745c05SJosef Bacik } 332877745c05SJosef Bacik 332977745c05SJosef Bacik btrfs_free_path(path); 333077745c05SJosef Bacik return ret; 333177745c05SJosef Bacik } 333277745c05SJosef Bacik 333377745c05SJosef Bacik int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) 333477745c05SJosef Bacik { 333577745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 333632da5386SDavid Sterba struct btrfs_block_group *cache; 333777745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 333877745c05SJosef Bacik int ret = 0; 333977745c05SJosef Bacik int should_put; 334077745c05SJosef Bacik struct btrfs_path *path; 334177745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 334277745c05SJosef Bacik 334377745c05SJosef Bacik path = btrfs_alloc_path(); 334477745c05SJosef Bacik if (!path) 334577745c05SJosef Bacik return -ENOMEM; 334677745c05SJosef Bacik 334777745c05SJosef Bacik /* 334877745c05SJosef Bacik * Even though we are in the critical section of the transaction commit, 334977745c05SJosef Bacik * we can still have concurrent tasks adding elements to this 335077745c05SJosef Bacik * transaction's list of dirty block groups. These tasks correspond to 335177745c05SJosef Bacik * endio free space workers started when writeback finishes for a 335277745c05SJosef Bacik * space cache, which run inode.c:btrfs_finish_ordered_io(), and can 335377745c05SJosef Bacik * allocate new block groups as a result of COWing nodes of the root 335477745c05SJosef Bacik * tree when updating the free space inode. The writeback for the space 335577745c05SJosef Bacik * caches is triggered by an earlier call to 335677745c05SJosef Bacik * btrfs_start_dirty_block_groups() and iterations of the following 335777745c05SJosef Bacik * loop. 335877745c05SJosef Bacik * Also we want to do the cache_save_setup first and then run the 335977745c05SJosef Bacik * delayed refs to make sure we have the best chance at doing this all 336077745c05SJosef Bacik * in one shot. 336177745c05SJosef Bacik */ 336277745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 336377745c05SJosef Bacik while (!list_empty(&cur_trans->dirty_bgs)) { 336477745c05SJosef Bacik cache = list_first_entry(&cur_trans->dirty_bgs, 336532da5386SDavid Sterba struct btrfs_block_group, 336677745c05SJosef Bacik dirty_list); 336777745c05SJosef Bacik 336877745c05SJosef Bacik /* 336977745c05SJosef Bacik * This can happen if cache_save_setup re-dirties a block group 337077745c05SJosef Bacik * that is already under IO. Just wait for it to finish and 337177745c05SJosef Bacik * then do it all again 337277745c05SJosef Bacik */ 337377745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 337477745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 337577745c05SJosef Bacik list_del_init(&cache->io_list); 337677745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 337777745c05SJosef Bacik btrfs_put_block_group(cache); 337877745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 337977745c05SJosef Bacik } 338077745c05SJosef Bacik 338177745c05SJosef Bacik /* 338277745c05SJosef Bacik * Don't remove from the dirty list until after we've waited on 338377745c05SJosef Bacik * any pending IO 338477745c05SJosef Bacik */ 338577745c05SJosef Bacik list_del_init(&cache->dirty_list); 338677745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 338777745c05SJosef Bacik should_put = 1; 338877745c05SJosef Bacik 338977745c05SJosef Bacik cache_save_setup(cache, trans, path); 339077745c05SJosef Bacik 339177745c05SJosef Bacik if (!ret) 339277745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 339377745c05SJosef Bacik (unsigned long) -1); 339477745c05SJosef Bacik 339577745c05SJosef Bacik if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { 339677745c05SJosef Bacik cache->io_ctl.inode = NULL; 339777745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 339877745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 339977745c05SJosef Bacik should_put = 0; 340077745c05SJosef Bacik list_add_tail(&cache->io_list, io); 340177745c05SJosef Bacik } else { 340277745c05SJosef Bacik /* 340377745c05SJosef Bacik * If we failed to write the cache, the 340477745c05SJosef Bacik * generation will be bad and life goes on 340577745c05SJosef Bacik */ 340677745c05SJosef Bacik ret = 0; 340777745c05SJosef Bacik } 340877745c05SJosef Bacik } 340977745c05SJosef Bacik if (!ret) { 34103be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 341177745c05SJosef Bacik /* 341277745c05SJosef Bacik * One of the free space endio workers might have 341377745c05SJosef Bacik * created a new block group while updating a free space 341477745c05SJosef Bacik * cache's inode (at inode.c:btrfs_finish_ordered_io()) 341577745c05SJosef Bacik * and hasn't released its transaction handle yet, in 341677745c05SJosef Bacik * which case the new block group is still attached to 341777745c05SJosef Bacik * its transaction handle and its creation has not 341877745c05SJosef Bacik * finished yet (no block group item in the extent tree 341977745c05SJosef Bacik * yet, etc). If this is the case, wait for all free 342077745c05SJosef Bacik * space endio workers to finish and retry. This is a 3421260db43cSRandy Dunlap * very rare case so no need for a more efficient and 342277745c05SJosef Bacik * complex approach. 342377745c05SJosef Bacik */ 342477745c05SJosef Bacik if (ret == -ENOENT) { 342577745c05SJosef Bacik wait_event(cur_trans->writer_wait, 342677745c05SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 34273be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 342877745c05SJosef Bacik } 342977745c05SJosef Bacik if (ret) 343077745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 343177745c05SJosef Bacik } 343277745c05SJosef Bacik 343377745c05SJosef Bacik /* If its not on the io list, we need to put the block group */ 343477745c05SJosef Bacik if (should_put) 343577745c05SJosef Bacik btrfs_put_block_group(cache); 343677745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 343777745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 343877745c05SJosef Bacik } 343977745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 344077745c05SJosef Bacik 344177745c05SJosef Bacik /* 344277745c05SJosef Bacik * Refer to the definition of io_bgs member for details why it's safe 344377745c05SJosef Bacik * to use it without any locking 344477745c05SJosef Bacik */ 344577745c05SJosef Bacik while (!list_empty(io)) { 344632da5386SDavid Sterba cache = list_first_entry(io, struct btrfs_block_group, 344777745c05SJosef Bacik io_list); 344877745c05SJosef Bacik list_del_init(&cache->io_list); 344977745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 345077745c05SJosef Bacik btrfs_put_block_group(cache); 345177745c05SJosef Bacik } 345277745c05SJosef Bacik 345377745c05SJosef Bacik btrfs_free_path(path); 345477745c05SJosef Bacik return ret; 345577745c05SJosef Bacik } 3456606d1bf1SJosef Bacik 3457606d1bf1SJosef Bacik int btrfs_update_block_group(struct btrfs_trans_handle *trans, 345811b66fa6SAnand Jain u64 bytenr, u64 num_bytes, bool alloc) 3459606d1bf1SJosef Bacik { 3460606d1bf1SJosef Bacik struct btrfs_fs_info *info = trans->fs_info; 346132da5386SDavid Sterba struct btrfs_block_group *cache = NULL; 3462606d1bf1SJosef Bacik u64 total = num_bytes; 3463606d1bf1SJosef Bacik u64 old_val; 3464606d1bf1SJosef Bacik u64 byte_in_group; 3465606d1bf1SJosef Bacik int factor; 3466606d1bf1SJosef Bacik int ret = 0; 3467606d1bf1SJosef Bacik 3468606d1bf1SJosef Bacik /* Block accounting for super block */ 3469606d1bf1SJosef Bacik spin_lock(&info->delalloc_root_lock); 3470606d1bf1SJosef Bacik old_val = btrfs_super_bytes_used(info->super_copy); 3471606d1bf1SJosef Bacik if (alloc) 3472606d1bf1SJosef Bacik old_val += num_bytes; 3473606d1bf1SJosef Bacik else 3474606d1bf1SJosef Bacik old_val -= num_bytes; 3475606d1bf1SJosef Bacik btrfs_set_super_bytes_used(info->super_copy, old_val); 3476606d1bf1SJosef Bacik spin_unlock(&info->delalloc_root_lock); 3477606d1bf1SJosef Bacik 3478606d1bf1SJosef Bacik while (total) { 3479efbf35a1SJosef Bacik bool reclaim = false; 3480ac2f1e63SJosef Bacik 3481606d1bf1SJosef Bacik cache = btrfs_lookup_block_group(info, bytenr); 3482606d1bf1SJosef Bacik if (!cache) { 3483606d1bf1SJosef Bacik ret = -ENOENT; 3484606d1bf1SJosef Bacik break; 3485606d1bf1SJosef Bacik } 3486606d1bf1SJosef Bacik factor = btrfs_bg_type_to_factor(cache->flags); 3487606d1bf1SJosef Bacik 3488606d1bf1SJosef Bacik /* 3489606d1bf1SJosef Bacik * If this block group has free space cache written out, we 3490606d1bf1SJosef Bacik * need to make sure to load it if we are removing space. This 3491606d1bf1SJosef Bacik * is because we need the unpinning stage to actually add the 3492606d1bf1SJosef Bacik * space back to the block group, otherwise we will leak space. 3493606d1bf1SJosef Bacik */ 349432da5386SDavid Sterba if (!alloc && !btrfs_block_group_done(cache)) 3495ced8ecf0SOmar Sandoval btrfs_cache_block_group(cache, true); 3496606d1bf1SJosef Bacik 3497b3470b5dSDavid Sterba byte_in_group = bytenr - cache->start; 3498b3470b5dSDavid Sterba WARN_ON(byte_in_group > cache->length); 3499606d1bf1SJosef Bacik 3500606d1bf1SJosef Bacik spin_lock(&cache->space_info->lock); 3501606d1bf1SJosef Bacik spin_lock(&cache->lock); 3502606d1bf1SJosef Bacik 3503606d1bf1SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 3504606d1bf1SJosef Bacik cache->disk_cache_state < BTRFS_DC_CLEAR) 3505606d1bf1SJosef Bacik cache->disk_cache_state = BTRFS_DC_CLEAR; 3506606d1bf1SJosef Bacik 3507bf38be65SDavid Sterba old_val = cache->used; 3508b3470b5dSDavid Sterba num_bytes = min(total, cache->length - byte_in_group); 3509606d1bf1SJosef Bacik if (alloc) { 3510606d1bf1SJosef Bacik old_val += num_bytes; 3511bf38be65SDavid Sterba cache->used = old_val; 3512606d1bf1SJosef Bacik cache->reserved -= num_bytes; 3513606d1bf1SJosef Bacik cache->space_info->bytes_reserved -= num_bytes; 3514606d1bf1SJosef Bacik cache->space_info->bytes_used += num_bytes; 3515606d1bf1SJosef Bacik cache->space_info->disk_used += num_bytes * factor; 3516606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3517606d1bf1SJosef Bacik spin_unlock(&cache->space_info->lock); 3518606d1bf1SJosef Bacik } else { 3519606d1bf1SJosef Bacik old_val -= num_bytes; 3520bf38be65SDavid Sterba cache->used = old_val; 3521606d1bf1SJosef Bacik cache->pinned += num_bytes; 3522606d1bf1SJosef Bacik btrfs_space_info_update_bytes_pinned(info, 3523606d1bf1SJosef Bacik cache->space_info, num_bytes); 3524606d1bf1SJosef Bacik cache->space_info->bytes_used -= num_bytes; 3525606d1bf1SJosef Bacik cache->space_info->disk_used -= num_bytes * factor; 3526ac2f1e63SJosef Bacik 3527ac2f1e63SJosef Bacik reclaim = should_reclaim_block_group(cache, num_bytes); 352852bb7a21SBoris Burkov 3529606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3530606d1bf1SJosef Bacik spin_unlock(&cache->space_info->lock); 3531606d1bf1SJosef Bacik 3532fe119a6eSNikolay Borisov set_extent_dirty(&trans->transaction->pinned_extents, 3533606d1bf1SJosef Bacik bytenr, bytenr + num_bytes - 1, 3534606d1bf1SJosef Bacik GFP_NOFS | __GFP_NOFAIL); 3535606d1bf1SJosef Bacik } 3536606d1bf1SJosef Bacik 3537606d1bf1SJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 3538606d1bf1SJosef Bacik if (list_empty(&cache->dirty_list)) { 3539606d1bf1SJosef Bacik list_add_tail(&cache->dirty_list, 3540606d1bf1SJosef Bacik &trans->transaction->dirty_bgs); 3541606d1bf1SJosef Bacik trans->delayed_ref_updates++; 3542606d1bf1SJosef Bacik btrfs_get_block_group(cache); 3543606d1bf1SJosef Bacik } 3544606d1bf1SJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 3545606d1bf1SJosef Bacik 3546606d1bf1SJosef Bacik /* 3547606d1bf1SJosef Bacik * No longer have used bytes in this block group, queue it for 3548606d1bf1SJosef Bacik * deletion. We do this after adding the block group to the 3549606d1bf1SJosef Bacik * dirty list to avoid races between cleaner kthread and space 3550606d1bf1SJosef Bacik * cache writeout. 3551606d1bf1SJosef Bacik */ 35526e80d4f8SDennis Zhou if (!alloc && old_val == 0) { 35536e80d4f8SDennis Zhou if (!btrfs_test_opt(info, DISCARD_ASYNC)) 3554606d1bf1SJosef Bacik btrfs_mark_bg_unused(cache); 3555ac2f1e63SJosef Bacik } else if (!alloc && reclaim) { 3556ac2f1e63SJosef Bacik btrfs_mark_bg_to_reclaim(cache); 35576e80d4f8SDennis Zhou } 3558606d1bf1SJosef Bacik 3559606d1bf1SJosef Bacik btrfs_put_block_group(cache); 3560606d1bf1SJosef Bacik total -= num_bytes; 3561606d1bf1SJosef Bacik bytenr += num_bytes; 3562606d1bf1SJosef Bacik } 3563606d1bf1SJosef Bacik 3564606d1bf1SJosef Bacik /* Modified block groups are accounted for in the delayed_refs_rsv. */ 3565606d1bf1SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 3566606d1bf1SJosef Bacik return ret; 3567606d1bf1SJosef Bacik } 3568606d1bf1SJosef Bacik 356943dd529aSDavid Sterba /* 357043dd529aSDavid Sterba * Update the block_group and space info counters. 357143dd529aSDavid Sterba * 3572606d1bf1SJosef Bacik * @cache: The cache we are manipulating 3573606d1bf1SJosef Bacik * @ram_bytes: The number of bytes of file content, and will be same to 3574606d1bf1SJosef Bacik * @num_bytes except for the compress path. 3575606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 3576606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 3577606d1bf1SJosef Bacik * 3578606d1bf1SJosef Bacik * This is called by the allocator when it reserves space. If this is a 3579606d1bf1SJosef Bacik * reservation and the block group has become read only we cannot make the 3580606d1bf1SJosef Bacik * reservation and return -EAGAIN, otherwise this function always succeeds. 3581606d1bf1SJosef Bacik */ 358232da5386SDavid Sterba int btrfs_add_reserved_bytes(struct btrfs_block_group *cache, 358352bb7a21SBoris Burkov u64 ram_bytes, u64 num_bytes, int delalloc, 358452bb7a21SBoris Burkov bool force_wrong_size_class) 3585606d1bf1SJosef Bacik { 3586606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 358752bb7a21SBoris Burkov enum btrfs_block_group_size_class size_class; 3588606d1bf1SJosef Bacik int ret = 0; 3589606d1bf1SJosef Bacik 3590606d1bf1SJosef Bacik spin_lock(&space_info->lock); 3591606d1bf1SJosef Bacik spin_lock(&cache->lock); 3592606d1bf1SJosef Bacik if (cache->ro) { 3593606d1bf1SJosef Bacik ret = -EAGAIN; 359452bb7a21SBoris Burkov goto out; 359552bb7a21SBoris Burkov } 359652bb7a21SBoris Burkov 3597cb0922f2SBoris Burkov if (btrfs_block_group_should_use_size_class(cache)) { 359852bb7a21SBoris Burkov size_class = btrfs_calc_block_group_size_class(num_bytes); 359952bb7a21SBoris Burkov ret = btrfs_use_block_group_size_class(cache, size_class, force_wrong_size_class); 360052bb7a21SBoris Burkov if (ret) 360152bb7a21SBoris Burkov goto out; 360252bb7a21SBoris Burkov } 3603606d1bf1SJosef Bacik cache->reserved += num_bytes; 3604606d1bf1SJosef Bacik space_info->bytes_reserved += num_bytes; 3605a43c3835SJosef Bacik trace_btrfs_space_reservation(cache->fs_info, "space_info", 3606a43c3835SJosef Bacik space_info->flags, num_bytes, 1); 3607606d1bf1SJosef Bacik btrfs_space_info_update_bytes_may_use(cache->fs_info, 3608606d1bf1SJosef Bacik space_info, -ram_bytes); 3609606d1bf1SJosef Bacik if (delalloc) 3610606d1bf1SJosef Bacik cache->delalloc_bytes += num_bytes; 361199ffb43eSJosef Bacik 361299ffb43eSJosef Bacik /* 361352bb7a21SBoris Burkov * Compression can use less space than we reserved, so wake tickets if 361452bb7a21SBoris Burkov * that happens. 361599ffb43eSJosef Bacik */ 361699ffb43eSJosef Bacik if (num_bytes < ram_bytes) 361799ffb43eSJosef Bacik btrfs_try_granting_tickets(cache->fs_info, space_info); 361852bb7a21SBoris Burkov out: 3619606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3620606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 3621606d1bf1SJosef Bacik return ret; 3622606d1bf1SJosef Bacik } 3623606d1bf1SJosef Bacik 362443dd529aSDavid Sterba /* 362543dd529aSDavid Sterba * Update the block_group and space info counters. 362643dd529aSDavid Sterba * 3627606d1bf1SJosef Bacik * @cache: The cache we are manipulating 3628606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 3629606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 3630606d1bf1SJosef Bacik * 3631606d1bf1SJosef Bacik * This is called by somebody who is freeing space that was never actually used 3632606d1bf1SJosef Bacik * on disk. For example if you reserve some space for a new leaf in transaction 3633606d1bf1SJosef Bacik * A and before transaction A commits you free that leaf, you call this with 3634606d1bf1SJosef Bacik * reserve set to 0 in order to clear the reservation. 3635606d1bf1SJosef Bacik */ 363632da5386SDavid Sterba void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, 3637606d1bf1SJosef Bacik u64 num_bytes, int delalloc) 3638606d1bf1SJosef Bacik { 3639606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 3640606d1bf1SJosef Bacik 3641606d1bf1SJosef Bacik spin_lock(&space_info->lock); 3642606d1bf1SJosef Bacik spin_lock(&cache->lock); 3643606d1bf1SJosef Bacik if (cache->ro) 3644606d1bf1SJosef Bacik space_info->bytes_readonly += num_bytes; 3645606d1bf1SJosef Bacik cache->reserved -= num_bytes; 3646606d1bf1SJosef Bacik space_info->bytes_reserved -= num_bytes; 3647606d1bf1SJosef Bacik space_info->max_extent_size = 0; 3648606d1bf1SJosef Bacik 3649606d1bf1SJosef Bacik if (delalloc) 3650606d1bf1SJosef Bacik cache->delalloc_bytes -= num_bytes; 3651606d1bf1SJosef Bacik spin_unlock(&cache->lock); 36523308234aSJosef Bacik 36533308234aSJosef Bacik btrfs_try_granting_tickets(cache->fs_info, space_info); 3654606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 3655606d1bf1SJosef Bacik } 365607730d87SJosef Bacik 365707730d87SJosef Bacik static void force_metadata_allocation(struct btrfs_fs_info *info) 365807730d87SJosef Bacik { 365907730d87SJosef Bacik struct list_head *head = &info->space_info; 366007730d87SJosef Bacik struct btrfs_space_info *found; 366107730d87SJosef Bacik 366272804905SJosef Bacik list_for_each_entry(found, head, list) { 366307730d87SJosef Bacik if (found->flags & BTRFS_BLOCK_GROUP_METADATA) 366407730d87SJosef Bacik found->force_alloc = CHUNK_ALLOC_FORCE; 366507730d87SJosef Bacik } 366607730d87SJosef Bacik } 366707730d87SJosef Bacik 366807730d87SJosef Bacik static int should_alloc_chunk(struct btrfs_fs_info *fs_info, 366907730d87SJosef Bacik struct btrfs_space_info *sinfo, int force) 367007730d87SJosef Bacik { 367107730d87SJosef Bacik u64 bytes_used = btrfs_space_info_used(sinfo, false); 367207730d87SJosef Bacik u64 thresh; 367307730d87SJosef Bacik 367407730d87SJosef Bacik if (force == CHUNK_ALLOC_FORCE) 367507730d87SJosef Bacik return 1; 367607730d87SJosef Bacik 367707730d87SJosef Bacik /* 367807730d87SJosef Bacik * in limited mode, we want to have some free space up to 367907730d87SJosef Bacik * about 1% of the FS size. 368007730d87SJosef Bacik */ 368107730d87SJosef Bacik if (force == CHUNK_ALLOC_LIMITED) { 368207730d87SJosef Bacik thresh = btrfs_super_total_bytes(fs_info->super_copy); 3683428c8e03SDavid Sterba thresh = max_t(u64, SZ_64M, mult_perc(thresh, 1)); 368407730d87SJosef Bacik 368507730d87SJosef Bacik if (sinfo->total_bytes - bytes_used < thresh) 368607730d87SJosef Bacik return 1; 368707730d87SJosef Bacik } 368807730d87SJosef Bacik 3689428c8e03SDavid Sterba if (bytes_used + SZ_2M < mult_perc(sinfo->total_bytes, 80)) 369007730d87SJosef Bacik return 0; 369107730d87SJosef Bacik return 1; 369207730d87SJosef Bacik } 369307730d87SJosef Bacik 369407730d87SJosef Bacik int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) 369507730d87SJosef Bacik { 369607730d87SJosef Bacik u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); 369707730d87SJosef Bacik 369807730d87SJosef Bacik return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 369907730d87SJosef Bacik } 370007730d87SJosef Bacik 3701820c363bSNaohiro Aota static struct btrfs_block_group *do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags) 370279bd3712SFilipe Manana { 370379bd3712SFilipe Manana struct btrfs_block_group *bg; 370479bd3712SFilipe Manana int ret; 370579bd3712SFilipe Manana 370607730d87SJosef Bacik /* 370779bd3712SFilipe Manana * Check if we have enough space in the system space info because we 370879bd3712SFilipe Manana * will need to update device items in the chunk btree and insert a new 370979bd3712SFilipe Manana * chunk item in the chunk btree as well. This will allocate a new 371079bd3712SFilipe Manana * system block group if needed. 371179bd3712SFilipe Manana */ 371279bd3712SFilipe Manana check_system_chunk(trans, flags); 371379bd3712SFilipe Manana 3714f6f39f7aSNikolay Borisov bg = btrfs_create_chunk(trans, flags); 371579bd3712SFilipe Manana if (IS_ERR(bg)) { 371679bd3712SFilipe Manana ret = PTR_ERR(bg); 371779bd3712SFilipe Manana goto out; 371879bd3712SFilipe Manana } 371979bd3712SFilipe Manana 372079bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); 372179bd3712SFilipe Manana /* 372279bd3712SFilipe Manana * Normally we are not expected to fail with -ENOSPC here, since we have 372379bd3712SFilipe Manana * previously reserved space in the system space_info and allocated one 3724ecd84d54SFilipe Manana * new system chunk if necessary. However there are three exceptions: 372579bd3712SFilipe Manana * 372679bd3712SFilipe Manana * 1) We may have enough free space in the system space_info but all the 372779bd3712SFilipe Manana * existing system block groups have a profile which can not be used 372879bd3712SFilipe Manana * for extent allocation. 372979bd3712SFilipe Manana * 373079bd3712SFilipe Manana * This happens when mounting in degraded mode. For example we have a 373179bd3712SFilipe Manana * RAID1 filesystem with 2 devices, lose one device and mount the fs 373279bd3712SFilipe Manana * using the other device in degraded mode. If we then allocate a chunk, 373379bd3712SFilipe Manana * we may have enough free space in the existing system space_info, but 373479bd3712SFilipe Manana * none of the block groups can be used for extent allocation since they 373579bd3712SFilipe Manana * have a RAID1 profile, and because we are in degraded mode with a 373679bd3712SFilipe Manana * single device, we are forced to allocate a new system chunk with a 373779bd3712SFilipe Manana * SINGLE profile. Making check_system_chunk() iterate over all system 373879bd3712SFilipe Manana * block groups and check if they have a usable profile and enough space 373979bd3712SFilipe Manana * can be slow on very large filesystems, so we tolerate the -ENOSPC and 374079bd3712SFilipe Manana * try again after forcing allocation of a new system chunk. Like this 374179bd3712SFilipe Manana * we avoid paying the cost of that search in normal circumstances, when 374279bd3712SFilipe Manana * we were not mounted in degraded mode; 374379bd3712SFilipe Manana * 374479bd3712SFilipe Manana * 2) We had enough free space info the system space_info, and one suitable 374579bd3712SFilipe Manana * block group to allocate from when we called check_system_chunk() 374679bd3712SFilipe Manana * above. However right after we called it, the only system block group 374779bd3712SFilipe Manana * with enough free space got turned into RO mode by a running scrub, 374879bd3712SFilipe Manana * and in this case we have to allocate a new one and retry. We only 374979bd3712SFilipe Manana * need do this allocate and retry once, since we have a transaction 3750ecd84d54SFilipe Manana * handle and scrub uses the commit root to search for block groups; 3751ecd84d54SFilipe Manana * 3752ecd84d54SFilipe Manana * 3) We had one system block group with enough free space when we called 3753ecd84d54SFilipe Manana * check_system_chunk(), but after that, right before we tried to 3754ecd84d54SFilipe Manana * allocate the last extent buffer we needed, a discard operation came 3755ecd84d54SFilipe Manana * in and it temporarily removed the last free space entry from the 3756ecd84d54SFilipe Manana * block group (discard removes a free space entry, discards it, and 3757ecd84d54SFilipe Manana * then adds back the entry to the block group cache). 375879bd3712SFilipe Manana */ 375979bd3712SFilipe Manana if (ret == -ENOSPC) { 376079bd3712SFilipe Manana const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info); 376179bd3712SFilipe Manana struct btrfs_block_group *sys_bg; 376279bd3712SFilipe Manana 3763f6f39f7aSNikolay Borisov sys_bg = btrfs_create_chunk(trans, sys_flags); 376479bd3712SFilipe Manana if (IS_ERR(sys_bg)) { 376579bd3712SFilipe Manana ret = PTR_ERR(sys_bg); 376679bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 376779bd3712SFilipe Manana goto out; 376879bd3712SFilipe Manana } 376979bd3712SFilipe Manana 377079bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 377179bd3712SFilipe Manana if (ret) { 377279bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 377379bd3712SFilipe Manana goto out; 377479bd3712SFilipe Manana } 377579bd3712SFilipe Manana 377679bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); 377779bd3712SFilipe Manana if (ret) { 377879bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 377979bd3712SFilipe Manana goto out; 378079bd3712SFilipe Manana } 378179bd3712SFilipe Manana } else if (ret) { 378279bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 378379bd3712SFilipe Manana goto out; 378479bd3712SFilipe Manana } 378579bd3712SFilipe Manana out: 378679bd3712SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 378779bd3712SFilipe Manana 3788820c363bSNaohiro Aota if (ret) 3789820c363bSNaohiro Aota return ERR_PTR(ret); 3790820c363bSNaohiro Aota 3791820c363bSNaohiro Aota btrfs_get_block_group(bg); 3792820c363bSNaohiro Aota return bg; 379379bd3712SFilipe Manana } 379479bd3712SFilipe Manana 379579bd3712SFilipe Manana /* 379679bd3712SFilipe Manana * Chunk allocation is done in 2 phases: 379779bd3712SFilipe Manana * 379879bd3712SFilipe Manana * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for 379979bd3712SFilipe Manana * the chunk, the chunk mapping, create its block group and add the items 380079bd3712SFilipe Manana * that belong in the chunk btree to it - more specifically, we need to 380179bd3712SFilipe Manana * update device items in the chunk btree and add a new chunk item to it. 380279bd3712SFilipe Manana * 380379bd3712SFilipe Manana * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block 380479bd3712SFilipe Manana * group item to the extent btree and the device extent items to the devices 380579bd3712SFilipe Manana * btree. 380679bd3712SFilipe Manana * 380779bd3712SFilipe Manana * This is done to prevent deadlocks. For example when COWing a node from the 380879bd3712SFilipe Manana * extent btree we are holding a write lock on the node's parent and if we 380979bd3712SFilipe Manana * trigger chunk allocation and attempted to insert the new block group item 381079bd3712SFilipe Manana * in the extent btree right way, we could deadlock because the path for the 381179bd3712SFilipe Manana * insertion can include that parent node. At first glance it seems impossible 381279bd3712SFilipe Manana * to trigger chunk allocation after starting a transaction since tasks should 381379bd3712SFilipe Manana * reserve enough transaction units (metadata space), however while that is true 381479bd3712SFilipe Manana * most of the time, chunk allocation may still be triggered for several reasons: 381579bd3712SFilipe Manana * 381679bd3712SFilipe Manana * 1) When reserving metadata, we check if there is enough free space in the 381779bd3712SFilipe Manana * metadata space_info and therefore don't trigger allocation of a new chunk. 381879bd3712SFilipe Manana * However later when the task actually tries to COW an extent buffer from 381979bd3712SFilipe Manana * the extent btree or from the device btree for example, it is forced to 382079bd3712SFilipe Manana * allocate a new block group (chunk) because the only one that had enough 382179bd3712SFilipe Manana * free space was just turned to RO mode by a running scrub for example (or 382279bd3712SFilipe Manana * device replace, block group reclaim thread, etc), so we can not use it 382379bd3712SFilipe Manana * for allocating an extent and end up being forced to allocate a new one; 382479bd3712SFilipe Manana * 382579bd3712SFilipe Manana * 2) Because we only check that the metadata space_info has enough free bytes, 382679bd3712SFilipe Manana * we end up not allocating a new metadata chunk in that case. However if 382779bd3712SFilipe Manana * the filesystem was mounted in degraded mode, none of the existing block 382879bd3712SFilipe Manana * groups might be suitable for extent allocation due to their incompatible 382979bd3712SFilipe Manana * profile (for e.g. mounting a 2 devices filesystem, where all block groups 383079bd3712SFilipe Manana * use a RAID1 profile, in degraded mode using a single device). In this case 383179bd3712SFilipe Manana * when the task attempts to COW some extent buffer of the extent btree for 383279bd3712SFilipe Manana * example, it will trigger allocation of a new metadata block group with a 383379bd3712SFilipe Manana * suitable profile (SINGLE profile in the example of the degraded mount of 383479bd3712SFilipe Manana * the RAID1 filesystem); 383579bd3712SFilipe Manana * 383679bd3712SFilipe Manana * 3) The task has reserved enough transaction units / metadata space, but when 383779bd3712SFilipe Manana * it attempts to COW an extent buffer from the extent or device btree for 383879bd3712SFilipe Manana * example, it does not find any free extent in any metadata block group, 383979bd3712SFilipe Manana * therefore forced to try to allocate a new metadata block group. 384079bd3712SFilipe Manana * This is because some other task allocated all available extents in the 384179bd3712SFilipe Manana * meanwhile - this typically happens with tasks that don't reserve space 384279bd3712SFilipe Manana * properly, either intentionally or as a bug. One example where this is 384379bd3712SFilipe Manana * done intentionally is fsync, as it does not reserve any transaction units 384479bd3712SFilipe Manana * and ends up allocating a variable number of metadata extents for log 3845ecd84d54SFilipe Manana * tree extent buffers; 3846ecd84d54SFilipe Manana * 3847ecd84d54SFilipe Manana * 4) The task has reserved enough transaction units / metadata space, but right 3848ecd84d54SFilipe Manana * before it tries to allocate the last extent buffer it needs, a discard 3849ecd84d54SFilipe Manana * operation comes in and, temporarily, removes the last free space entry from 3850ecd84d54SFilipe Manana * the only metadata block group that had free space (discard starts by 3851ecd84d54SFilipe Manana * removing a free space entry from a block group, then does the discard 3852ecd84d54SFilipe Manana * operation and, once it's done, it adds back the free space entry to the 3853ecd84d54SFilipe Manana * block group). 385479bd3712SFilipe Manana * 385579bd3712SFilipe Manana * We also need this 2 phases setup when adding a device to a filesystem with 385679bd3712SFilipe Manana * a seed device - we must create new metadata and system chunks without adding 385779bd3712SFilipe Manana * any of the block group items to the chunk, extent and device btrees. If we 385879bd3712SFilipe Manana * did not do it this way, we would get ENOSPC when attempting to update those 385979bd3712SFilipe Manana * btrees, since all the chunks from the seed device are read-only. 386079bd3712SFilipe Manana * 386179bd3712SFilipe Manana * Phase 1 does the updates and insertions to the chunk btree because if we had 386279bd3712SFilipe Manana * it done in phase 2 and have a thundering herd of tasks allocating chunks in 386379bd3712SFilipe Manana * parallel, we risk having too many system chunks allocated by many tasks if 386479bd3712SFilipe Manana * many tasks reach phase 1 without the previous ones completing phase 2. In the 386579bd3712SFilipe Manana * extreme case this leads to exhaustion of the system chunk array in the 386679bd3712SFilipe Manana * superblock. This is easier to trigger if using a btree node/leaf size of 64K 386779bd3712SFilipe Manana * and with RAID filesystems (so we have more device items in the chunk btree). 386879bd3712SFilipe Manana * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of 386979bd3712SFilipe Manana * the system chunk array due to concurrent allocations") provides more details. 387079bd3712SFilipe Manana * 38712bb2e00eSFilipe Manana * Allocation of system chunks does not happen through this function. A task that 38722bb2e00eSFilipe Manana * needs to update the chunk btree (the only btree that uses system chunks), must 38732bb2e00eSFilipe Manana * preallocate chunk space by calling either check_system_chunk() or 38742bb2e00eSFilipe Manana * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or 38752bb2e00eSFilipe Manana * metadata chunk or when removing a chunk, while the later is used before doing 38762bb2e00eSFilipe Manana * a modification to the chunk btree - use cases for the later are adding, 38772bb2e00eSFilipe Manana * removing and resizing a device as well as relocation of a system chunk. 38782bb2e00eSFilipe Manana * See the comment below for more details. 387979bd3712SFilipe Manana * 388079bd3712SFilipe Manana * The reservation of system space, done through check_system_chunk(), as well 388179bd3712SFilipe Manana * as all the updates and insertions into the chunk btree must be done while 388279bd3712SFilipe Manana * holding fs_info->chunk_mutex. This is important to guarantee that while COWing 388379bd3712SFilipe Manana * an extent buffer from the chunks btree we never trigger allocation of a new 388479bd3712SFilipe Manana * system chunk, which would result in a deadlock (trying to lock twice an 388579bd3712SFilipe Manana * extent buffer of the chunk btree, first time before triggering the chunk 388679bd3712SFilipe Manana * allocation and the second time during chunk allocation while attempting to 388779bd3712SFilipe Manana * update the chunks btree). The system chunk array is also updated while holding 388879bd3712SFilipe Manana * that mutex. The same logic applies to removing chunks - we must reserve system 388979bd3712SFilipe Manana * space, update the chunk btree and the system chunk array in the superblock 389079bd3712SFilipe Manana * while holding fs_info->chunk_mutex. 389179bd3712SFilipe Manana * 389279bd3712SFilipe Manana * This function, btrfs_chunk_alloc(), belongs to phase 1. 389379bd3712SFilipe Manana * 389479bd3712SFilipe Manana * If @force is CHUNK_ALLOC_FORCE: 389507730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 389607730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 389779bd3712SFilipe Manana * If @force is NOT CHUNK_ALLOC_FORCE: 389807730d87SJosef Bacik * - return 0 if it doesn't need to allocate a new chunk, 389907730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 390007730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 390107730d87SJosef Bacik */ 390207730d87SJosef Bacik int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, 390307730d87SJosef Bacik enum btrfs_chunk_alloc_enum force) 390407730d87SJosef Bacik { 390507730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 390607730d87SJosef Bacik struct btrfs_space_info *space_info; 3907820c363bSNaohiro Aota struct btrfs_block_group *ret_bg; 390807730d87SJosef Bacik bool wait_for_alloc = false; 390907730d87SJosef Bacik bool should_alloc = false; 3910760e69c4SNaohiro Aota bool from_extent_allocation = false; 391107730d87SJosef Bacik int ret = 0; 391207730d87SJosef Bacik 3913760e69c4SNaohiro Aota if (force == CHUNK_ALLOC_FORCE_FOR_EXTENT) { 3914760e69c4SNaohiro Aota from_extent_allocation = true; 3915760e69c4SNaohiro Aota force = CHUNK_ALLOC_FORCE; 3916760e69c4SNaohiro Aota } 3917760e69c4SNaohiro Aota 391807730d87SJosef Bacik /* Don't re-enter if we're already allocating a chunk */ 391907730d87SJosef Bacik if (trans->allocating_chunk) 392007730d87SJosef Bacik return -ENOSPC; 392179bd3712SFilipe Manana /* 39222bb2e00eSFilipe Manana * Allocation of system chunks can not happen through this path, as we 39232bb2e00eSFilipe Manana * could end up in a deadlock if we are allocating a data or metadata 39242bb2e00eSFilipe Manana * chunk and there is another task modifying the chunk btree. 39252bb2e00eSFilipe Manana * 39262bb2e00eSFilipe Manana * This is because while we are holding the chunk mutex, we will attempt 39272bb2e00eSFilipe Manana * to add the new chunk item to the chunk btree or update an existing 39282bb2e00eSFilipe Manana * device item in the chunk btree, while the other task that is modifying 39292bb2e00eSFilipe Manana * the chunk btree is attempting to COW an extent buffer while holding a 39302bb2e00eSFilipe Manana * lock on it and on its parent - if the COW operation triggers a system 39312bb2e00eSFilipe Manana * chunk allocation, then we can deadlock because we are holding the 39322bb2e00eSFilipe Manana * chunk mutex and we may need to access that extent buffer or its parent 39332bb2e00eSFilipe Manana * in order to add the chunk item or update a device item. 39342bb2e00eSFilipe Manana * 39352bb2e00eSFilipe Manana * Tasks that want to modify the chunk tree should reserve system space 39362bb2e00eSFilipe Manana * before updating the chunk btree, by calling either 39372bb2e00eSFilipe Manana * btrfs_reserve_chunk_metadata() or check_system_chunk(). 39382bb2e00eSFilipe Manana * It's possible that after a task reserves the space, it still ends up 39392bb2e00eSFilipe Manana * here - this happens in the cases described above at do_chunk_alloc(). 39402bb2e00eSFilipe Manana * The task will have to either retry or fail. 394179bd3712SFilipe Manana */ 39422bb2e00eSFilipe Manana if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 394379bd3712SFilipe Manana return -ENOSPC; 394407730d87SJosef Bacik 394507730d87SJosef Bacik space_info = btrfs_find_space_info(fs_info, flags); 394607730d87SJosef Bacik ASSERT(space_info); 394707730d87SJosef Bacik 394807730d87SJosef Bacik do { 394907730d87SJosef Bacik spin_lock(&space_info->lock); 395007730d87SJosef Bacik if (force < space_info->force_alloc) 395107730d87SJosef Bacik force = space_info->force_alloc; 395207730d87SJosef Bacik should_alloc = should_alloc_chunk(fs_info, space_info, force); 395307730d87SJosef Bacik if (space_info->full) { 395407730d87SJosef Bacik /* No more free physical space */ 395507730d87SJosef Bacik if (should_alloc) 395607730d87SJosef Bacik ret = -ENOSPC; 395707730d87SJosef Bacik else 395807730d87SJosef Bacik ret = 0; 395907730d87SJosef Bacik spin_unlock(&space_info->lock); 396007730d87SJosef Bacik return ret; 396107730d87SJosef Bacik } else if (!should_alloc) { 396207730d87SJosef Bacik spin_unlock(&space_info->lock); 396307730d87SJosef Bacik return 0; 396407730d87SJosef Bacik } else if (space_info->chunk_alloc) { 396507730d87SJosef Bacik /* 396607730d87SJosef Bacik * Someone is already allocating, so we need to block 396707730d87SJosef Bacik * until this someone is finished and then loop to 396807730d87SJosef Bacik * recheck if we should continue with our allocation 396907730d87SJosef Bacik * attempt. 397007730d87SJosef Bacik */ 397107730d87SJosef Bacik wait_for_alloc = true; 39721314ca78SJosef Bacik force = CHUNK_ALLOC_NO_FORCE; 397307730d87SJosef Bacik spin_unlock(&space_info->lock); 397407730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 397507730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 397607730d87SJosef Bacik } else { 397707730d87SJosef Bacik /* Proceed with allocation */ 397807730d87SJosef Bacik space_info->chunk_alloc = 1; 397907730d87SJosef Bacik wait_for_alloc = false; 398007730d87SJosef Bacik spin_unlock(&space_info->lock); 398107730d87SJosef Bacik } 398207730d87SJosef Bacik 398307730d87SJosef Bacik cond_resched(); 398407730d87SJosef Bacik } while (wait_for_alloc); 398507730d87SJosef Bacik 398607730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 398707730d87SJosef Bacik trans->allocating_chunk = true; 398807730d87SJosef Bacik 398907730d87SJosef Bacik /* 399007730d87SJosef Bacik * If we have mixed data/metadata chunks we want to make sure we keep 399107730d87SJosef Bacik * allocating mixed chunks instead of individual chunks. 399207730d87SJosef Bacik */ 399307730d87SJosef Bacik if (btrfs_mixed_space_info(space_info)) 399407730d87SJosef Bacik flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); 399507730d87SJosef Bacik 399607730d87SJosef Bacik /* 399707730d87SJosef Bacik * if we're doing a data chunk, go ahead and make sure that 399807730d87SJosef Bacik * we keep a reasonable number of metadata chunks allocated in the 399907730d87SJosef Bacik * FS as well. 400007730d87SJosef Bacik */ 400107730d87SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { 400207730d87SJosef Bacik fs_info->data_chunk_allocations++; 400307730d87SJosef Bacik if (!(fs_info->data_chunk_allocations % 400407730d87SJosef Bacik fs_info->metadata_ratio)) 400507730d87SJosef Bacik force_metadata_allocation(fs_info); 400607730d87SJosef Bacik } 400707730d87SJosef Bacik 4008820c363bSNaohiro Aota ret_bg = do_chunk_alloc(trans, flags); 400907730d87SJosef Bacik trans->allocating_chunk = false; 401007730d87SJosef Bacik 4011760e69c4SNaohiro Aota if (IS_ERR(ret_bg)) { 4012820c363bSNaohiro Aota ret = PTR_ERR(ret_bg); 4013760e69c4SNaohiro Aota } else if (from_extent_allocation) { 4014760e69c4SNaohiro Aota /* 4015760e69c4SNaohiro Aota * New block group is likely to be used soon. Try to activate 4016760e69c4SNaohiro Aota * it now. Failure is OK for now. 4017760e69c4SNaohiro Aota */ 4018760e69c4SNaohiro Aota btrfs_zone_activate(ret_bg); 4019760e69c4SNaohiro Aota } 4020760e69c4SNaohiro Aota 4021760e69c4SNaohiro Aota if (!ret) 4022820c363bSNaohiro Aota btrfs_put_block_group(ret_bg); 4023820c363bSNaohiro Aota 402407730d87SJosef Bacik spin_lock(&space_info->lock); 402507730d87SJosef Bacik if (ret < 0) { 402607730d87SJosef Bacik if (ret == -ENOSPC) 402707730d87SJosef Bacik space_info->full = 1; 402807730d87SJosef Bacik else 402907730d87SJosef Bacik goto out; 403007730d87SJosef Bacik } else { 403107730d87SJosef Bacik ret = 1; 403207730d87SJosef Bacik space_info->max_extent_size = 0; 403307730d87SJosef Bacik } 403407730d87SJosef Bacik 403507730d87SJosef Bacik space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; 403607730d87SJosef Bacik out: 403707730d87SJosef Bacik space_info->chunk_alloc = 0; 403807730d87SJosef Bacik spin_unlock(&space_info->lock); 403907730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 404007730d87SJosef Bacik 404107730d87SJosef Bacik return ret; 404207730d87SJosef Bacik } 404307730d87SJosef Bacik 404407730d87SJosef Bacik static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) 404507730d87SJosef Bacik { 404607730d87SJosef Bacik u64 num_dev; 404707730d87SJosef Bacik 404807730d87SJosef Bacik num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; 404907730d87SJosef Bacik if (!num_dev) 405007730d87SJosef Bacik num_dev = fs_info->fs_devices->rw_devices; 405107730d87SJosef Bacik 405207730d87SJosef Bacik return num_dev; 405307730d87SJosef Bacik } 405407730d87SJosef Bacik 40552bb2e00eSFilipe Manana static void reserve_chunk_space(struct btrfs_trans_handle *trans, 40562bb2e00eSFilipe Manana u64 bytes, 40572bb2e00eSFilipe Manana u64 type) 405807730d87SJosef Bacik { 405907730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 406007730d87SJosef Bacik struct btrfs_space_info *info; 406107730d87SJosef Bacik u64 left; 406207730d87SJosef Bacik int ret = 0; 406307730d87SJosef Bacik 406407730d87SJosef Bacik /* 406507730d87SJosef Bacik * Needed because we can end up allocating a system chunk and for an 406607730d87SJosef Bacik * atomic and race free space reservation in the chunk block reserve. 406707730d87SJosef Bacik */ 406807730d87SJosef Bacik lockdep_assert_held(&fs_info->chunk_mutex); 406907730d87SJosef Bacik 407007730d87SJosef Bacik info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); 407107730d87SJosef Bacik spin_lock(&info->lock); 407207730d87SJosef Bacik left = info->total_bytes - btrfs_space_info_used(info, true); 407307730d87SJosef Bacik spin_unlock(&info->lock); 407407730d87SJosef Bacik 40752bb2e00eSFilipe Manana if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { 407607730d87SJosef Bacik btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", 40772bb2e00eSFilipe Manana left, bytes, type); 407807730d87SJosef Bacik btrfs_dump_space_info(fs_info, info, 0, 0); 407907730d87SJosef Bacik } 408007730d87SJosef Bacik 40812bb2e00eSFilipe Manana if (left < bytes) { 408207730d87SJosef Bacik u64 flags = btrfs_system_alloc_profile(fs_info); 408379bd3712SFilipe Manana struct btrfs_block_group *bg; 408407730d87SJosef Bacik 408507730d87SJosef Bacik /* 408607730d87SJosef Bacik * Ignore failure to create system chunk. We might end up not 408707730d87SJosef Bacik * needing it, as we might not need to COW all nodes/leafs from 408807730d87SJosef Bacik * the paths we visit in the chunk tree (they were already COWed 408907730d87SJosef Bacik * or created in the current transaction for example). 409007730d87SJosef Bacik */ 4091f6f39f7aSNikolay Borisov bg = btrfs_create_chunk(trans, flags); 409279bd3712SFilipe Manana if (IS_ERR(bg)) { 409379bd3712SFilipe Manana ret = PTR_ERR(bg); 40942bb2e00eSFilipe Manana } else { 409579bd3712SFilipe Manana /* 4096b6a98021SNaohiro Aota * We have a new chunk. We also need to activate it for 4097b6a98021SNaohiro Aota * zoned filesystem. 4098b6a98021SNaohiro Aota */ 4099b6a98021SNaohiro Aota ret = btrfs_zoned_activate_one_bg(fs_info, info, true); 4100b6a98021SNaohiro Aota if (ret < 0) 4101b6a98021SNaohiro Aota return; 4102b6a98021SNaohiro Aota 4103b6a98021SNaohiro Aota /* 410479bd3712SFilipe Manana * If we fail to add the chunk item here, we end up 410579bd3712SFilipe Manana * trying again at phase 2 of chunk allocation, at 410679bd3712SFilipe Manana * btrfs_create_pending_block_groups(). So ignore 41072bb2e00eSFilipe Manana * any error here. An ENOSPC here could happen, due to 41082bb2e00eSFilipe Manana * the cases described at do_chunk_alloc() - the system 41092bb2e00eSFilipe Manana * block group we just created was just turned into RO 41102bb2e00eSFilipe Manana * mode by a scrub for example, or a running discard 41112bb2e00eSFilipe Manana * temporarily removed its free space entries, etc. 411279bd3712SFilipe Manana */ 411379bd3712SFilipe Manana btrfs_chunk_alloc_add_chunk_item(trans, bg); 411479bd3712SFilipe Manana } 411507730d87SJosef Bacik } 411607730d87SJosef Bacik 411707730d87SJosef Bacik if (!ret) { 41189270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, 411907730d87SJosef Bacik &fs_info->chunk_block_rsv, 41202bb2e00eSFilipe Manana bytes, BTRFS_RESERVE_NO_FLUSH); 41211cb3db1cSFilipe Manana if (!ret) 41222bb2e00eSFilipe Manana trans->chunk_bytes_reserved += bytes; 412307730d87SJosef Bacik } 412407730d87SJosef Bacik } 412507730d87SJosef Bacik 41262bb2e00eSFilipe Manana /* 41272bb2e00eSFilipe Manana * Reserve space in the system space for allocating or removing a chunk. 41282bb2e00eSFilipe Manana * The caller must be holding fs_info->chunk_mutex. 41292bb2e00eSFilipe Manana */ 41302bb2e00eSFilipe Manana void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) 41312bb2e00eSFilipe Manana { 41322bb2e00eSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 41332bb2e00eSFilipe Manana const u64 num_devs = get_profile_num_devs(fs_info, type); 41342bb2e00eSFilipe Manana u64 bytes; 41352bb2e00eSFilipe Manana 41362bb2e00eSFilipe Manana /* num_devs device items to update and 1 chunk item to add or remove. */ 41372bb2e00eSFilipe Manana bytes = btrfs_calc_metadata_size(fs_info, num_devs) + 41382bb2e00eSFilipe Manana btrfs_calc_insert_metadata_size(fs_info, 1); 41392bb2e00eSFilipe Manana 41402bb2e00eSFilipe Manana reserve_chunk_space(trans, bytes, type); 41412bb2e00eSFilipe Manana } 41422bb2e00eSFilipe Manana 41432bb2e00eSFilipe Manana /* 41442bb2e00eSFilipe Manana * Reserve space in the system space, if needed, for doing a modification to the 41452bb2e00eSFilipe Manana * chunk btree. 41462bb2e00eSFilipe Manana * 41472bb2e00eSFilipe Manana * @trans: A transaction handle. 41482bb2e00eSFilipe Manana * @is_item_insertion: Indicate if the modification is for inserting a new item 41492bb2e00eSFilipe Manana * in the chunk btree or if it's for the deletion or update 41502bb2e00eSFilipe Manana * of an existing item. 41512bb2e00eSFilipe Manana * 41522bb2e00eSFilipe Manana * This is used in a context where we need to update the chunk btree outside 41532bb2e00eSFilipe Manana * block group allocation and removal, to avoid a deadlock with a concurrent 41542bb2e00eSFilipe Manana * task that is allocating a metadata or data block group and therefore needs to 41552bb2e00eSFilipe Manana * update the chunk btree while holding the chunk mutex. After the update to the 41562bb2e00eSFilipe Manana * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called. 41572bb2e00eSFilipe Manana * 41582bb2e00eSFilipe Manana */ 41592bb2e00eSFilipe Manana void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans, 41602bb2e00eSFilipe Manana bool is_item_insertion) 41612bb2e00eSFilipe Manana { 41622bb2e00eSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 41632bb2e00eSFilipe Manana u64 bytes; 41642bb2e00eSFilipe Manana 41652bb2e00eSFilipe Manana if (is_item_insertion) 41662bb2e00eSFilipe Manana bytes = btrfs_calc_insert_metadata_size(fs_info, 1); 41672bb2e00eSFilipe Manana else 41682bb2e00eSFilipe Manana bytes = btrfs_calc_metadata_size(fs_info, 1); 41692bb2e00eSFilipe Manana 41702bb2e00eSFilipe Manana mutex_lock(&fs_info->chunk_mutex); 41712bb2e00eSFilipe Manana reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM); 41722bb2e00eSFilipe Manana mutex_unlock(&fs_info->chunk_mutex); 41732bb2e00eSFilipe Manana } 41742bb2e00eSFilipe Manana 41753e43c279SJosef Bacik void btrfs_put_block_group_cache(struct btrfs_fs_info *info) 41763e43c279SJosef Bacik { 417732da5386SDavid Sterba struct btrfs_block_group *block_group; 41783e43c279SJosef Bacik 417950c31eaaSJosef Bacik block_group = btrfs_lookup_first_block_group(info, 0); 41803e43c279SJosef Bacik while (block_group) { 41813e43c279SJosef Bacik btrfs_wait_block_group_cache_done(block_group); 41823e43c279SJosef Bacik spin_lock(&block_group->lock); 418350c31eaaSJosef Bacik if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, 418450c31eaaSJosef Bacik &block_group->runtime_flags)) { 418550c31eaaSJosef Bacik struct inode *inode = block_group->inode; 41863e43c279SJosef Bacik 41873e43c279SJosef Bacik block_group->inode = NULL; 41883e43c279SJosef Bacik spin_unlock(&block_group->lock); 418950c31eaaSJosef Bacik 41903e43c279SJosef Bacik ASSERT(block_group->io_ctl.inode == NULL); 41913e43c279SJosef Bacik iput(inode); 419250c31eaaSJosef Bacik } else { 419350c31eaaSJosef Bacik spin_unlock(&block_group->lock); 419450c31eaaSJosef Bacik } 419550c31eaaSJosef Bacik block_group = btrfs_next_block_group(block_group); 41963e43c279SJosef Bacik } 41973e43c279SJosef Bacik } 41983e43c279SJosef Bacik 41993e43c279SJosef Bacik /* 42003e43c279SJosef Bacik * Must be called only after stopping all workers, since we could have block 42013e43c279SJosef Bacik * group caching kthreads running, and therefore they could race with us if we 42023e43c279SJosef Bacik * freed the block groups before stopping them. 42033e43c279SJosef Bacik */ 42043e43c279SJosef Bacik int btrfs_free_block_groups(struct btrfs_fs_info *info) 42053e43c279SJosef Bacik { 420632da5386SDavid Sterba struct btrfs_block_group *block_group; 42073e43c279SJosef Bacik struct btrfs_space_info *space_info; 42083e43c279SJosef Bacik struct btrfs_caching_control *caching_ctl; 42093e43c279SJosef Bacik struct rb_node *n; 42103e43c279SJosef Bacik 421116b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 42123e43c279SJosef Bacik while (!list_empty(&info->caching_block_groups)) { 42133e43c279SJosef Bacik caching_ctl = list_entry(info->caching_block_groups.next, 42143e43c279SJosef Bacik struct btrfs_caching_control, list); 42153e43c279SJosef Bacik list_del(&caching_ctl->list); 42163e43c279SJosef Bacik btrfs_put_caching_control(caching_ctl); 42173e43c279SJosef Bacik } 421816b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42193e43c279SJosef Bacik 42203e43c279SJosef Bacik spin_lock(&info->unused_bgs_lock); 42213e43c279SJosef Bacik while (!list_empty(&info->unused_bgs)) { 42223e43c279SJosef Bacik block_group = list_first_entry(&info->unused_bgs, 422332da5386SDavid Sterba struct btrfs_block_group, 42243e43c279SJosef Bacik bg_list); 42253e43c279SJosef Bacik list_del_init(&block_group->bg_list); 42263e43c279SJosef Bacik btrfs_put_block_group(block_group); 42273e43c279SJosef Bacik } 42283e43c279SJosef Bacik 422918bb8bbfSJohannes Thumshirn while (!list_empty(&info->reclaim_bgs)) { 423018bb8bbfSJohannes Thumshirn block_group = list_first_entry(&info->reclaim_bgs, 423118bb8bbfSJohannes Thumshirn struct btrfs_block_group, 423218bb8bbfSJohannes Thumshirn bg_list); 423318bb8bbfSJohannes Thumshirn list_del_init(&block_group->bg_list); 423418bb8bbfSJohannes Thumshirn btrfs_put_block_group(block_group); 423518bb8bbfSJohannes Thumshirn } 423618bb8bbfSJohannes Thumshirn spin_unlock(&info->unused_bgs_lock); 423718bb8bbfSJohannes Thumshirn 4238afba2bc0SNaohiro Aota spin_lock(&info->zone_active_bgs_lock); 4239afba2bc0SNaohiro Aota while (!list_empty(&info->zone_active_bgs)) { 4240afba2bc0SNaohiro Aota block_group = list_first_entry(&info->zone_active_bgs, 4241afba2bc0SNaohiro Aota struct btrfs_block_group, 4242afba2bc0SNaohiro Aota active_bg_list); 4243afba2bc0SNaohiro Aota list_del_init(&block_group->active_bg_list); 4244afba2bc0SNaohiro Aota btrfs_put_block_group(block_group); 4245afba2bc0SNaohiro Aota } 4246afba2bc0SNaohiro Aota spin_unlock(&info->zone_active_bgs_lock); 4247afba2bc0SNaohiro Aota 424816b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 424908dddb29SFilipe Manana while ((n = rb_last(&info->block_group_cache_tree.rb_root)) != NULL) { 425032da5386SDavid Sterba block_group = rb_entry(n, struct btrfs_block_group, 42513e43c279SJosef Bacik cache_node); 425208dddb29SFilipe Manana rb_erase_cached(&block_group->cache_node, 42533e43c279SJosef Bacik &info->block_group_cache_tree); 42543e43c279SJosef Bacik RB_CLEAR_NODE(&block_group->cache_node); 425516b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42563e43c279SJosef Bacik 42573e43c279SJosef Bacik down_write(&block_group->space_info->groups_sem); 42583e43c279SJosef Bacik list_del(&block_group->list); 42593e43c279SJosef Bacik up_write(&block_group->space_info->groups_sem); 42603e43c279SJosef Bacik 42613e43c279SJosef Bacik /* 42623e43c279SJosef Bacik * We haven't cached this block group, which means we could 42633e43c279SJosef Bacik * possibly have excluded extents on this block group. 42643e43c279SJosef Bacik */ 42653e43c279SJosef Bacik if (block_group->cached == BTRFS_CACHE_NO || 42663e43c279SJosef Bacik block_group->cached == BTRFS_CACHE_ERROR) 42673e43c279SJosef Bacik btrfs_free_excluded_extents(block_group); 42683e43c279SJosef Bacik 42693e43c279SJosef Bacik btrfs_remove_free_space_cache(block_group); 42703e43c279SJosef Bacik ASSERT(block_group->cached != BTRFS_CACHE_STARTED); 42713e43c279SJosef Bacik ASSERT(list_empty(&block_group->dirty_list)); 42723e43c279SJosef Bacik ASSERT(list_empty(&block_group->io_list)); 42733e43c279SJosef Bacik ASSERT(list_empty(&block_group->bg_list)); 427448aaeebeSJosef Bacik ASSERT(refcount_read(&block_group->refs) == 1); 4275195a49eaSFilipe Manana ASSERT(block_group->swap_extents == 0); 42763e43c279SJosef Bacik btrfs_put_block_group(block_group); 42773e43c279SJosef Bacik 427816b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 42793e43c279SJosef Bacik } 428016b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42813e43c279SJosef Bacik 42823e43c279SJosef Bacik btrfs_release_global_block_rsv(info); 42833e43c279SJosef Bacik 42843e43c279SJosef Bacik while (!list_empty(&info->space_info)) { 42853e43c279SJosef Bacik space_info = list_entry(info->space_info.next, 42863e43c279SJosef Bacik struct btrfs_space_info, 42873e43c279SJosef Bacik list); 42883e43c279SJosef Bacik 42893e43c279SJosef Bacik /* 42903e43c279SJosef Bacik * Do not hide this behind enospc_debug, this is actually 42913e43c279SJosef Bacik * important and indicates a real bug if this happens. 42923e43c279SJosef Bacik */ 42933e43c279SJosef Bacik if (WARN_ON(space_info->bytes_pinned > 0 || 42943e43c279SJosef Bacik space_info->bytes_may_use > 0)) 42953e43c279SJosef Bacik btrfs_dump_space_info(info, space_info, 0, 0); 429640cdc509SFilipe Manana 429740cdc509SFilipe Manana /* 429840cdc509SFilipe Manana * If there was a failure to cleanup a log tree, very likely due 429940cdc509SFilipe Manana * to an IO failure on a writeback attempt of one or more of its 430040cdc509SFilipe Manana * extent buffers, we could not do proper (and cheap) unaccounting 430140cdc509SFilipe Manana * of their reserved space, so don't warn on bytes_reserved > 0 in 430240cdc509SFilipe Manana * that case. 430340cdc509SFilipe Manana */ 430440cdc509SFilipe Manana if (!(space_info->flags & BTRFS_BLOCK_GROUP_METADATA) || 430540cdc509SFilipe Manana !BTRFS_FS_LOG_CLEANUP_ERROR(info)) { 430640cdc509SFilipe Manana if (WARN_ON(space_info->bytes_reserved > 0)) 430740cdc509SFilipe Manana btrfs_dump_space_info(info, space_info, 0, 0); 430840cdc509SFilipe Manana } 430940cdc509SFilipe Manana 4310d611add4SFilipe Manana WARN_ON(space_info->reclaim_size > 0); 43113e43c279SJosef Bacik list_del(&space_info->list); 43123e43c279SJosef Bacik btrfs_sysfs_remove_space_info(space_info); 43133e43c279SJosef Bacik } 43143e43c279SJosef Bacik return 0; 43153e43c279SJosef Bacik } 4316684b752bSFilipe Manana 4317684b752bSFilipe Manana void btrfs_freeze_block_group(struct btrfs_block_group *cache) 4318684b752bSFilipe Manana { 4319684b752bSFilipe Manana atomic_inc(&cache->frozen); 4320684b752bSFilipe Manana } 4321684b752bSFilipe Manana 4322684b752bSFilipe Manana void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group) 4323684b752bSFilipe Manana { 4324684b752bSFilipe Manana struct btrfs_fs_info *fs_info = block_group->fs_info; 4325684b752bSFilipe Manana struct extent_map_tree *em_tree; 4326684b752bSFilipe Manana struct extent_map *em; 4327684b752bSFilipe Manana bool cleanup; 4328684b752bSFilipe Manana 4329684b752bSFilipe Manana spin_lock(&block_group->lock); 4330684b752bSFilipe Manana cleanup = (atomic_dec_and_test(&block_group->frozen) && 43313349b57fSJosef Bacik test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)); 4332684b752bSFilipe Manana spin_unlock(&block_group->lock); 4333684b752bSFilipe Manana 4334684b752bSFilipe Manana if (cleanup) { 4335684b752bSFilipe Manana em_tree = &fs_info->mapping_tree; 4336684b752bSFilipe Manana write_lock(&em_tree->lock); 4337684b752bSFilipe Manana em = lookup_extent_mapping(em_tree, block_group->start, 4338684b752bSFilipe Manana 1); 4339684b752bSFilipe Manana BUG_ON(!em); /* logic error, can't happen */ 4340684b752bSFilipe Manana remove_extent_mapping(em_tree, em); 4341684b752bSFilipe Manana write_unlock(&em_tree->lock); 4342684b752bSFilipe Manana 4343684b752bSFilipe Manana /* once for us and once for the tree */ 4344684b752bSFilipe Manana free_extent_map(em); 4345684b752bSFilipe Manana free_extent_map(em); 4346684b752bSFilipe Manana 4347684b752bSFilipe Manana /* 4348684b752bSFilipe Manana * We may have left one free space entry and other possible 4349684b752bSFilipe Manana * tasks trimming this block group have left 1 entry each one. 4350684b752bSFilipe Manana * Free them if any. 4351684b752bSFilipe Manana */ 4352fc80f7acSJosef Bacik btrfs_remove_free_space_cache(block_group); 4353684b752bSFilipe Manana } 4354684b752bSFilipe Manana } 4355195a49eaSFilipe Manana 4356195a49eaSFilipe Manana bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg) 4357195a49eaSFilipe Manana { 4358195a49eaSFilipe Manana bool ret = true; 4359195a49eaSFilipe Manana 4360195a49eaSFilipe Manana spin_lock(&bg->lock); 4361195a49eaSFilipe Manana if (bg->ro) 4362195a49eaSFilipe Manana ret = false; 4363195a49eaSFilipe Manana else 4364195a49eaSFilipe Manana bg->swap_extents++; 4365195a49eaSFilipe Manana spin_unlock(&bg->lock); 4366195a49eaSFilipe Manana 4367195a49eaSFilipe Manana return ret; 4368195a49eaSFilipe Manana } 4369195a49eaSFilipe Manana 4370195a49eaSFilipe Manana void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount) 4371195a49eaSFilipe Manana { 4372195a49eaSFilipe Manana spin_lock(&bg->lock); 4373195a49eaSFilipe Manana ASSERT(!bg->ro); 4374195a49eaSFilipe Manana ASSERT(bg->swap_extents >= amount); 4375195a49eaSFilipe Manana bg->swap_extents -= amount; 4376195a49eaSFilipe Manana spin_unlock(&bg->lock); 4377195a49eaSFilipe Manana } 437852bb7a21SBoris Burkov 437952bb7a21SBoris Burkov enum btrfs_block_group_size_class btrfs_calc_block_group_size_class(u64 size) 438052bb7a21SBoris Burkov { 438152bb7a21SBoris Burkov if (size <= SZ_128K) 438252bb7a21SBoris Burkov return BTRFS_BG_SZ_SMALL; 438352bb7a21SBoris Burkov if (size <= SZ_8M) 438452bb7a21SBoris Burkov return BTRFS_BG_SZ_MEDIUM; 438552bb7a21SBoris Burkov return BTRFS_BG_SZ_LARGE; 438652bb7a21SBoris Burkov } 438752bb7a21SBoris Burkov 438852bb7a21SBoris Burkov /* 438952bb7a21SBoris Burkov * Handle a block group allocating an extent in a size class 439052bb7a21SBoris Burkov * 439152bb7a21SBoris Burkov * @bg: The block group we allocated in. 439252bb7a21SBoris Burkov * @size_class: The size class of the allocation. 439352bb7a21SBoris Burkov * @force_wrong_size_class: Whether we are desperate enough to allow 439452bb7a21SBoris Burkov * mismatched size classes. 439552bb7a21SBoris Burkov * 439652bb7a21SBoris Burkov * Returns: 0 if the size class was valid for this block_group, -EAGAIN in the 439752bb7a21SBoris Burkov * case of a race that leads to the wrong size class without 439852bb7a21SBoris Burkov * force_wrong_size_class set. 439952bb7a21SBoris Burkov * 440052bb7a21SBoris Burkov * find_free_extent will skip block groups with a mismatched size class until 440152bb7a21SBoris Burkov * it really needs to avoid ENOSPC. In that case it will set 440252bb7a21SBoris Burkov * force_wrong_size_class. However, if a block group is newly allocated and 440352bb7a21SBoris Burkov * doesn't yet have a size class, then it is possible for two allocations of 440452bb7a21SBoris Burkov * different sizes to race and both try to use it. The loser is caught here and 440552bb7a21SBoris Burkov * has to retry. 440652bb7a21SBoris Burkov */ 440752bb7a21SBoris Burkov int btrfs_use_block_group_size_class(struct btrfs_block_group *bg, 440852bb7a21SBoris Burkov enum btrfs_block_group_size_class size_class, 440952bb7a21SBoris Burkov bool force_wrong_size_class) 441052bb7a21SBoris Burkov { 441152bb7a21SBoris Burkov ASSERT(size_class != BTRFS_BG_SZ_NONE); 441252bb7a21SBoris Burkov 441352bb7a21SBoris Burkov /* The new allocation is in the right size class, do nothing */ 441452bb7a21SBoris Burkov if (bg->size_class == size_class) 441552bb7a21SBoris Burkov return 0; 441652bb7a21SBoris Burkov /* 441752bb7a21SBoris Burkov * The new allocation is in a mismatched size class. 441852bb7a21SBoris Burkov * This means one of two things: 441952bb7a21SBoris Burkov * 442052bb7a21SBoris Burkov * 1. Two tasks in find_free_extent for different size_classes raced 442152bb7a21SBoris Burkov * and hit the same empty block_group. Make the loser try again. 442252bb7a21SBoris Burkov * 2. A call to find_free_extent got desperate enough to set 442352bb7a21SBoris Burkov * 'force_wrong_slab'. Don't change the size_class, but allow the 442452bb7a21SBoris Burkov * allocation. 442552bb7a21SBoris Burkov */ 442652bb7a21SBoris Burkov if (bg->size_class != BTRFS_BG_SZ_NONE) { 442752bb7a21SBoris Burkov if (force_wrong_size_class) 442852bb7a21SBoris Burkov return 0; 442952bb7a21SBoris Burkov return -EAGAIN; 443052bb7a21SBoris Burkov } 443152bb7a21SBoris Burkov /* 443252bb7a21SBoris Burkov * The happy new block group case: the new allocation is the first 443352bb7a21SBoris Burkov * one in the block_group so we set size_class. 443452bb7a21SBoris Burkov */ 443552bb7a21SBoris Burkov bg->size_class = size_class; 443652bb7a21SBoris Burkov 443752bb7a21SBoris Burkov return 0; 443852bb7a21SBoris Burkov } 4439cb0922f2SBoris Burkov 4440cb0922f2SBoris Burkov bool btrfs_block_group_should_use_size_class(struct btrfs_block_group *bg) 4441cb0922f2SBoris Burkov { 4442cb0922f2SBoris Burkov if (btrfs_is_zoned(bg->fs_info)) 4443cb0922f2SBoris Burkov return false; 4444cb0922f2SBoris Burkov if (!btrfs_is_block_group_data_only(bg)) 4445cb0922f2SBoris Burkov return false; 4446cb0922f2SBoris Burkov return true; 4447cb0922f2SBoris Burkov } 4448