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 543*c7eec3d9SBoris Burkov /* 544*c7eec3d9SBoris Burkov * Get an arbitrary extent item index / max_index through the block group 545*c7eec3d9SBoris Burkov * 546*c7eec3d9SBoris Burkov * @block_group the block group to sample from 547*c7eec3d9SBoris Burkov * @index: the integral step through the block group to grab from 548*c7eec3d9SBoris Burkov * @max_index: the granularity of the sampling 549*c7eec3d9SBoris Burkov * @key: return value parameter for the item we find 550*c7eec3d9SBoris Burkov * 551*c7eec3d9SBoris Burkov * Pre-conditions on indices: 552*c7eec3d9SBoris Burkov * 0 <= index <= max_index 553*c7eec3d9SBoris Burkov * 0 < max_index 554*c7eec3d9SBoris Burkov * 555*c7eec3d9SBoris Burkov * Returns: 0 on success, 1 if the search didn't yield a useful item, negative 556*c7eec3d9SBoris Burkov * error code on error. 557*c7eec3d9SBoris Burkov */ 558*c7eec3d9SBoris Burkov static int sample_block_group_extent_item(struct btrfs_caching_control *caching_ctl, 559*c7eec3d9SBoris Burkov struct btrfs_block_group *block_group, 560*c7eec3d9SBoris Burkov int index, int max_index, 561*c7eec3d9SBoris Burkov struct btrfs_key *key) 562*c7eec3d9SBoris Burkov { 563*c7eec3d9SBoris Burkov struct btrfs_fs_info *fs_info = block_group->fs_info; 564*c7eec3d9SBoris Burkov struct btrfs_root *extent_root; 565*c7eec3d9SBoris Burkov int ret = 0; 566*c7eec3d9SBoris Burkov u64 search_offset; 567*c7eec3d9SBoris Burkov u64 search_end = block_group->start + block_group->length; 568*c7eec3d9SBoris Burkov struct btrfs_path *path; 569*c7eec3d9SBoris Burkov 570*c7eec3d9SBoris Burkov ASSERT(index >= 0); 571*c7eec3d9SBoris Burkov ASSERT(index <= max_index); 572*c7eec3d9SBoris Burkov ASSERT(max_index > 0); 573*c7eec3d9SBoris Burkov lockdep_assert_held(&caching_ctl->mutex); 574*c7eec3d9SBoris Burkov lockdep_assert_held_read(&fs_info->commit_root_sem); 575*c7eec3d9SBoris Burkov 576*c7eec3d9SBoris Burkov path = btrfs_alloc_path(); 577*c7eec3d9SBoris Burkov if (!path) 578*c7eec3d9SBoris Burkov return -ENOMEM; 579*c7eec3d9SBoris Burkov 580*c7eec3d9SBoris Burkov extent_root = btrfs_extent_root(fs_info, max_t(u64, block_group->start, 581*c7eec3d9SBoris Burkov BTRFS_SUPER_INFO_OFFSET)); 582*c7eec3d9SBoris Burkov 583*c7eec3d9SBoris Burkov path->skip_locking = 1; 584*c7eec3d9SBoris Burkov path->search_commit_root = 1; 585*c7eec3d9SBoris Burkov path->reada = READA_FORWARD; 586*c7eec3d9SBoris Burkov 587*c7eec3d9SBoris Burkov search_offset = index * div_u64(block_group->length, max_index); 588*c7eec3d9SBoris Burkov key->objectid = block_group->start + search_offset; 589*c7eec3d9SBoris Burkov key->type = BTRFS_EXTENT_ITEM_KEY; 590*c7eec3d9SBoris Burkov key->offset = 0; 591*c7eec3d9SBoris Burkov 592*c7eec3d9SBoris Burkov while (1) { 593*c7eec3d9SBoris Burkov ret = btrfs_search_forward(extent_root, key, path, 0); 594*c7eec3d9SBoris Burkov if (ret != 0) 595*c7eec3d9SBoris Burkov goto out; 596*c7eec3d9SBoris Burkov /* Success; sampled an extent item in the block group */ 597*c7eec3d9SBoris Burkov if (key->type == BTRFS_EXTENT_ITEM_KEY && 598*c7eec3d9SBoris Burkov key->objectid >= block_group->start && 599*c7eec3d9SBoris Burkov key->objectid + key->offset <= search_end) 600*c7eec3d9SBoris Burkov goto out; 601*c7eec3d9SBoris Burkov 602*c7eec3d9SBoris Burkov /* We can't possibly find a valid extent item anymore */ 603*c7eec3d9SBoris Burkov if (key->objectid >= search_end) { 604*c7eec3d9SBoris Burkov ret = 1; 605*c7eec3d9SBoris Burkov break; 606*c7eec3d9SBoris Burkov } 607*c7eec3d9SBoris Burkov if (key->type < BTRFS_EXTENT_ITEM_KEY) 608*c7eec3d9SBoris Burkov key->type = BTRFS_EXTENT_ITEM_KEY; 609*c7eec3d9SBoris Burkov else 610*c7eec3d9SBoris Burkov key->objectid++; 611*c7eec3d9SBoris Burkov btrfs_release_path(path); 612*c7eec3d9SBoris Burkov up_read(&fs_info->commit_root_sem); 613*c7eec3d9SBoris Burkov mutex_unlock(&caching_ctl->mutex); 614*c7eec3d9SBoris Burkov cond_resched(); 615*c7eec3d9SBoris Burkov mutex_lock(&caching_ctl->mutex); 616*c7eec3d9SBoris Burkov down_read(&fs_info->commit_root_sem); 617*c7eec3d9SBoris Burkov } 618*c7eec3d9SBoris Burkov out: 619*c7eec3d9SBoris Burkov lockdep_assert_held(&caching_ctl->mutex); 620*c7eec3d9SBoris Burkov lockdep_assert_held_read(&fs_info->commit_root_sem); 621*c7eec3d9SBoris Burkov btrfs_free_path(path); 622*c7eec3d9SBoris Burkov return ret; 623*c7eec3d9SBoris Burkov } 624*c7eec3d9SBoris Burkov 625*c7eec3d9SBoris Burkov /* 626*c7eec3d9SBoris Burkov * Best effort attempt to compute a block group's size class while caching it. 627*c7eec3d9SBoris Burkov * 628*c7eec3d9SBoris Burkov * @block_group: the block group we are caching 629*c7eec3d9SBoris Burkov * 630*c7eec3d9SBoris Burkov * We cannot infer the size class while adding free space extents, because that 631*c7eec3d9SBoris Burkov * logic doesn't care about contiguous file extents (it doesn't differentiate 632*c7eec3d9SBoris Burkov * between a 100M extent and 100 contiguous 1M extents). So we need to read the 633*c7eec3d9SBoris Burkov * file extent items. Reading all of them is quite wasteful, because usually 634*c7eec3d9SBoris Burkov * only a handful are enough to give a good answer. Therefore, we just grab 5 of 635*c7eec3d9SBoris Burkov * them at even steps through the block group and pick the smallest size class 636*c7eec3d9SBoris Burkov * we see. Since size class is best effort, and not guaranteed in general, 637*c7eec3d9SBoris Burkov * inaccuracy is acceptable. 638*c7eec3d9SBoris Burkov * 639*c7eec3d9SBoris Burkov * To be more explicit about why this algorithm makes sense: 640*c7eec3d9SBoris Burkov * 641*c7eec3d9SBoris Burkov * If we are caching in a block group from disk, then there are three major cases 642*c7eec3d9SBoris Burkov * to consider: 643*c7eec3d9SBoris Burkov * 1. the block group is well behaved and all extents in it are the same size 644*c7eec3d9SBoris Burkov * class. 645*c7eec3d9SBoris Burkov * 2. the block group is mostly one size class with rare exceptions for last 646*c7eec3d9SBoris Burkov * ditch allocations 647*c7eec3d9SBoris Burkov * 3. the block group was populated before size classes and can have a totally 648*c7eec3d9SBoris Burkov * arbitrary mix of size classes. 649*c7eec3d9SBoris Burkov * 650*c7eec3d9SBoris Burkov * In case 1, looking at any extent in the block group will yield the correct 651*c7eec3d9SBoris Burkov * result. For the mixed cases, taking the minimum size class seems like a good 652*c7eec3d9SBoris Burkov * approximation, since gaps from frees will be usable to the size class. For 653*c7eec3d9SBoris Burkov * 2., a small handful of file extents is likely to yield the right answer. For 654*c7eec3d9SBoris Burkov * 3, we can either read every file extent, or admit that this is best effort 655*c7eec3d9SBoris Burkov * anyway and try to stay fast. 656*c7eec3d9SBoris Burkov * 657*c7eec3d9SBoris Burkov * Returns: 0 on success, negative error code on error. 658*c7eec3d9SBoris Burkov */ 659*c7eec3d9SBoris Burkov static int load_block_group_size_class(struct btrfs_caching_control *caching_ctl, 660*c7eec3d9SBoris Burkov struct btrfs_block_group *block_group) 661*c7eec3d9SBoris Burkov { 662*c7eec3d9SBoris Burkov struct btrfs_key key; 663*c7eec3d9SBoris Burkov int i; 664*c7eec3d9SBoris Burkov u64 min_size = block_group->length; 665*c7eec3d9SBoris Burkov enum btrfs_block_group_size_class size_class = BTRFS_BG_SZ_NONE; 666*c7eec3d9SBoris Burkov int ret; 667*c7eec3d9SBoris Burkov 668*c7eec3d9SBoris Burkov if (!btrfs_is_block_group_data_only(block_group)) 669*c7eec3d9SBoris Burkov return 0; 670*c7eec3d9SBoris Burkov 671*c7eec3d9SBoris Burkov for (i = 0; i < 5; ++i) { 672*c7eec3d9SBoris Burkov ret = sample_block_group_extent_item(caching_ctl, block_group, i, 5, &key); 673*c7eec3d9SBoris Burkov if (ret < 0) 674*c7eec3d9SBoris Burkov goto out; 675*c7eec3d9SBoris Burkov if (ret > 0) 676*c7eec3d9SBoris Burkov continue; 677*c7eec3d9SBoris Burkov min_size = min_t(u64, min_size, key.offset); 678*c7eec3d9SBoris Burkov size_class = btrfs_calc_block_group_size_class(min_size); 679*c7eec3d9SBoris Burkov } 680*c7eec3d9SBoris Burkov if (size_class != BTRFS_BG_SZ_NONE) { 681*c7eec3d9SBoris Burkov spin_lock(&block_group->lock); 682*c7eec3d9SBoris Burkov block_group->size_class = size_class; 683*c7eec3d9SBoris Burkov spin_unlock(&block_group->lock); 684*c7eec3d9SBoris Burkov } 685*c7eec3d9SBoris Burkov 686*c7eec3d9SBoris Burkov out: 687*c7eec3d9SBoris Burkov return ret; 688*c7eec3d9SBoris Burkov } 689*c7eec3d9SBoris Burkov 6909f21246dSJosef Bacik static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) 6919f21246dSJosef Bacik { 69232da5386SDavid Sterba struct btrfs_block_group *block_group = caching_ctl->block_group; 6939f21246dSJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 69429cbcf40SJosef Bacik struct btrfs_root *extent_root; 6959f21246dSJosef Bacik struct btrfs_path *path; 6969f21246dSJosef Bacik struct extent_buffer *leaf; 6979f21246dSJosef Bacik struct btrfs_key key; 6989f21246dSJosef Bacik u64 total_found = 0; 6999f21246dSJosef Bacik u64 last = 0; 7009f21246dSJosef Bacik u32 nritems; 7019f21246dSJosef Bacik int ret; 7029f21246dSJosef Bacik bool wakeup = true; 7039f21246dSJosef Bacik 7049f21246dSJosef Bacik path = btrfs_alloc_path(); 7059f21246dSJosef Bacik if (!path) 7069f21246dSJosef Bacik return -ENOMEM; 7079f21246dSJosef Bacik 708b3470b5dSDavid Sterba last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET); 70929cbcf40SJosef Bacik extent_root = btrfs_extent_root(fs_info, last); 7109f21246dSJosef Bacik 7119f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 7129f21246dSJosef Bacik /* 7139f21246dSJosef Bacik * If we're fragmenting we don't want to make anybody think we can 7149f21246dSJosef Bacik * allocate from this block group until we've had a chance to fragment 7159f21246dSJosef Bacik * the free space. 7169f21246dSJosef Bacik */ 7179f21246dSJosef Bacik if (btrfs_should_fragment_free_space(block_group)) 7189f21246dSJosef Bacik wakeup = false; 7199f21246dSJosef Bacik #endif 7209f21246dSJosef Bacik /* 7219f21246dSJosef Bacik * We don't want to deadlock with somebody trying to allocate a new 7229f21246dSJosef Bacik * extent for the extent root while also trying to search the extent 7239f21246dSJosef Bacik * root to add free space. So we skip locking and search the commit 7249f21246dSJosef Bacik * root, since its read-only 7259f21246dSJosef Bacik */ 7269f21246dSJosef Bacik path->skip_locking = 1; 7279f21246dSJosef Bacik path->search_commit_root = 1; 7289f21246dSJosef Bacik path->reada = READA_FORWARD; 7299f21246dSJosef Bacik 7309f21246dSJosef Bacik key.objectid = last; 7319f21246dSJosef Bacik key.offset = 0; 7329f21246dSJosef Bacik key.type = BTRFS_EXTENT_ITEM_KEY; 7339f21246dSJosef Bacik 7349f21246dSJosef Bacik next: 7359f21246dSJosef Bacik ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); 7369f21246dSJosef Bacik if (ret < 0) 7379f21246dSJosef Bacik goto out; 7389f21246dSJosef Bacik 7399f21246dSJosef Bacik leaf = path->nodes[0]; 7409f21246dSJosef Bacik nritems = btrfs_header_nritems(leaf); 7419f21246dSJosef Bacik 7429f21246dSJosef Bacik while (1) { 7439f21246dSJosef Bacik if (btrfs_fs_closing(fs_info) > 1) { 7449f21246dSJosef Bacik last = (u64)-1; 7459f21246dSJosef Bacik break; 7469f21246dSJosef Bacik } 7479f21246dSJosef Bacik 7489f21246dSJosef Bacik if (path->slots[0] < nritems) { 7499f21246dSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 7509f21246dSJosef Bacik } else { 7519f21246dSJosef Bacik ret = btrfs_find_next_key(extent_root, path, &key, 0, 0); 7529f21246dSJosef Bacik if (ret) 7539f21246dSJosef Bacik break; 7549f21246dSJosef Bacik 7559f21246dSJosef Bacik if (need_resched() || 7569f21246dSJosef Bacik rwsem_is_contended(&fs_info->commit_root_sem)) { 7579f21246dSJosef Bacik btrfs_release_path(path); 7589f21246dSJosef Bacik up_read(&fs_info->commit_root_sem); 7599f21246dSJosef Bacik mutex_unlock(&caching_ctl->mutex); 7609f21246dSJosef Bacik cond_resched(); 7619f21246dSJosef Bacik mutex_lock(&caching_ctl->mutex); 7629f21246dSJosef Bacik down_read(&fs_info->commit_root_sem); 7639f21246dSJosef Bacik goto next; 7649f21246dSJosef Bacik } 7659f21246dSJosef Bacik 7669f21246dSJosef Bacik ret = btrfs_next_leaf(extent_root, path); 7679f21246dSJosef Bacik if (ret < 0) 7689f21246dSJosef Bacik goto out; 7699f21246dSJosef Bacik if (ret) 7709f21246dSJosef Bacik break; 7719f21246dSJosef Bacik leaf = path->nodes[0]; 7729f21246dSJosef Bacik nritems = btrfs_header_nritems(leaf); 7739f21246dSJosef Bacik continue; 7749f21246dSJosef Bacik } 7759f21246dSJosef Bacik 7769f21246dSJosef Bacik if (key.objectid < last) { 7779f21246dSJosef Bacik key.objectid = last; 7789f21246dSJosef Bacik key.offset = 0; 7799f21246dSJosef Bacik key.type = BTRFS_EXTENT_ITEM_KEY; 7809f21246dSJosef Bacik btrfs_release_path(path); 7819f21246dSJosef Bacik goto next; 7829f21246dSJosef Bacik } 7839f21246dSJosef Bacik 784b3470b5dSDavid Sterba if (key.objectid < block_group->start) { 7859f21246dSJosef Bacik path->slots[0]++; 7869f21246dSJosef Bacik continue; 7879f21246dSJosef Bacik } 7889f21246dSJosef Bacik 789b3470b5dSDavid Sterba if (key.objectid >= block_group->start + block_group->length) 7909f21246dSJosef Bacik break; 7919f21246dSJosef Bacik 7929f21246dSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7939f21246dSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7949f21246dSJosef Bacik total_found += add_new_free_space(block_group, last, 7959f21246dSJosef Bacik key.objectid); 7969f21246dSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY) 7979f21246dSJosef Bacik last = key.objectid + 7989f21246dSJosef Bacik fs_info->nodesize; 7999f21246dSJosef Bacik else 8009f21246dSJosef Bacik last = key.objectid + key.offset; 8019f21246dSJosef Bacik 8029f21246dSJosef Bacik if (total_found > CACHING_CTL_WAKE_UP) { 8039f21246dSJosef Bacik total_found = 0; 8049f21246dSJosef Bacik if (wakeup) 8059f21246dSJosef Bacik wake_up(&caching_ctl->wait); 8069f21246dSJosef Bacik } 8079f21246dSJosef Bacik } 8089f21246dSJosef Bacik path->slots[0]++; 8099f21246dSJosef Bacik } 8109f21246dSJosef Bacik ret = 0; 8119f21246dSJosef Bacik 8129f21246dSJosef Bacik total_found += add_new_free_space(block_group, last, 813b3470b5dSDavid Sterba block_group->start + block_group->length); 8149f21246dSJosef Bacik 8159f21246dSJosef Bacik out: 8169f21246dSJosef Bacik btrfs_free_path(path); 8179f21246dSJosef Bacik return ret; 8189f21246dSJosef Bacik } 8199f21246dSJosef Bacik 8209f21246dSJosef Bacik static noinline void caching_thread(struct btrfs_work *work) 8219f21246dSJosef Bacik { 82232da5386SDavid Sterba struct btrfs_block_group *block_group; 8239f21246dSJosef Bacik struct btrfs_fs_info *fs_info; 8249f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 8259f21246dSJosef Bacik int ret; 8269f21246dSJosef Bacik 8279f21246dSJosef Bacik caching_ctl = container_of(work, struct btrfs_caching_control, work); 8289f21246dSJosef Bacik block_group = caching_ctl->block_group; 8299f21246dSJosef Bacik fs_info = block_group->fs_info; 8309f21246dSJosef Bacik 8319f21246dSJosef Bacik mutex_lock(&caching_ctl->mutex); 8329f21246dSJosef Bacik down_read(&fs_info->commit_root_sem); 8339f21246dSJosef Bacik 834*c7eec3d9SBoris Burkov load_block_group_size_class(caching_ctl, block_group); 835e747853cSJosef Bacik if (btrfs_test_opt(fs_info, SPACE_CACHE)) { 836e747853cSJosef Bacik ret = load_free_space_cache(block_group); 837e747853cSJosef Bacik if (ret == 1) { 838e747853cSJosef Bacik ret = 0; 839e747853cSJosef Bacik goto done; 840e747853cSJosef Bacik } 841e747853cSJosef Bacik 842e747853cSJosef Bacik /* 843e747853cSJosef Bacik * We failed to load the space cache, set ourselves to 844e747853cSJosef Bacik * CACHE_STARTED and carry on. 845e747853cSJosef Bacik */ 846e747853cSJosef Bacik spin_lock(&block_group->lock); 847e747853cSJosef Bacik block_group->cached = BTRFS_CACHE_STARTED; 848e747853cSJosef Bacik spin_unlock(&block_group->lock); 849e747853cSJosef Bacik wake_up(&caching_ctl->wait); 850e747853cSJosef Bacik } 851e747853cSJosef Bacik 8522f96e402SJosef Bacik /* 8532f96e402SJosef Bacik * If we are in the transaction that populated the free space tree we 8542f96e402SJosef Bacik * can't actually cache from the free space tree as our commit root and 8552f96e402SJosef Bacik * real root are the same, so we could change the contents of the blocks 8562f96e402SJosef Bacik * while caching. Instead do the slow caching in this case, and after 8572f96e402SJosef Bacik * the transaction has committed we will be safe. 8582f96e402SJosef Bacik */ 8592f96e402SJosef Bacik if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 8602f96e402SJosef Bacik !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags))) 8619f21246dSJosef Bacik ret = load_free_space_tree(caching_ctl); 8629f21246dSJosef Bacik else 8639f21246dSJosef Bacik ret = load_extent_tree_free(caching_ctl); 864e747853cSJosef Bacik done: 8659f21246dSJosef Bacik spin_lock(&block_group->lock); 8669f21246dSJosef Bacik block_group->caching_ctl = NULL; 8679f21246dSJosef Bacik block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; 8689f21246dSJosef Bacik spin_unlock(&block_group->lock); 8699f21246dSJosef Bacik 8709f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 8719f21246dSJosef Bacik if (btrfs_should_fragment_free_space(block_group)) { 8729f21246dSJosef Bacik u64 bytes_used; 8739f21246dSJosef Bacik 8749f21246dSJosef Bacik spin_lock(&block_group->space_info->lock); 8759f21246dSJosef Bacik spin_lock(&block_group->lock); 876b3470b5dSDavid Sterba bytes_used = block_group->length - block_group->used; 8779f21246dSJosef Bacik block_group->space_info->bytes_used += bytes_used >> 1; 8789f21246dSJosef Bacik spin_unlock(&block_group->lock); 8799f21246dSJosef Bacik spin_unlock(&block_group->space_info->lock); 880e11c0406SJosef Bacik fragment_free_space(block_group); 8819f21246dSJosef Bacik } 8829f21246dSJosef Bacik #endif 8839f21246dSJosef Bacik 8849f21246dSJosef Bacik up_read(&fs_info->commit_root_sem); 8859f21246dSJosef Bacik btrfs_free_excluded_extents(block_group); 8869f21246dSJosef Bacik mutex_unlock(&caching_ctl->mutex); 8879f21246dSJosef Bacik 8889f21246dSJosef Bacik wake_up(&caching_ctl->wait); 8899f21246dSJosef Bacik 8909f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 8919f21246dSJosef Bacik btrfs_put_block_group(block_group); 8929f21246dSJosef Bacik } 8939f21246dSJosef Bacik 894ced8ecf0SOmar Sandoval int btrfs_cache_block_group(struct btrfs_block_group *cache, bool wait) 8959f21246dSJosef Bacik { 8969f21246dSJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 897e747853cSJosef Bacik struct btrfs_caching_control *caching_ctl = NULL; 8989f21246dSJosef Bacik int ret = 0; 8999f21246dSJosef Bacik 9002eda5708SNaohiro Aota /* Allocator for zoned filesystems does not use the cache at all */ 9012eda5708SNaohiro Aota if (btrfs_is_zoned(fs_info)) 9022eda5708SNaohiro Aota return 0; 9032eda5708SNaohiro Aota 9049f21246dSJosef Bacik caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); 9059f21246dSJosef Bacik if (!caching_ctl) 9069f21246dSJosef Bacik return -ENOMEM; 9079f21246dSJosef Bacik 9089f21246dSJosef Bacik INIT_LIST_HEAD(&caching_ctl->list); 9099f21246dSJosef Bacik mutex_init(&caching_ctl->mutex); 9109f21246dSJosef Bacik init_waitqueue_head(&caching_ctl->wait); 9119f21246dSJosef Bacik caching_ctl->block_group = cache; 912e747853cSJosef Bacik refcount_set(&caching_ctl->count, 2); 913a0cac0ecSOmar Sandoval btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); 9149f21246dSJosef Bacik 9159f21246dSJosef Bacik spin_lock(&cache->lock); 9169f21246dSJosef Bacik if (cache->cached != BTRFS_CACHE_NO) { 9179f21246dSJosef Bacik kfree(caching_ctl); 918e747853cSJosef Bacik 919e747853cSJosef Bacik caching_ctl = cache->caching_ctl; 920e747853cSJosef Bacik if (caching_ctl) 921e747853cSJosef Bacik refcount_inc(&caching_ctl->count); 922e747853cSJosef Bacik spin_unlock(&cache->lock); 923e747853cSJosef Bacik goto out; 9249f21246dSJosef Bacik } 9259f21246dSJosef Bacik WARN_ON(cache->caching_ctl); 9269f21246dSJosef Bacik cache->caching_ctl = caching_ctl; 9279f21246dSJosef Bacik cache->cached = BTRFS_CACHE_STARTED; 9289f21246dSJosef Bacik spin_unlock(&cache->lock); 9299f21246dSJosef Bacik 93016b0c258SFilipe Manana write_lock(&fs_info->block_group_cache_lock); 9319f21246dSJosef Bacik refcount_inc(&caching_ctl->count); 9329f21246dSJosef Bacik list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); 93316b0c258SFilipe Manana write_unlock(&fs_info->block_group_cache_lock); 9349f21246dSJosef Bacik 9359f21246dSJosef Bacik btrfs_get_block_group(cache); 9369f21246dSJosef Bacik 9379f21246dSJosef Bacik btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); 938e747853cSJosef Bacik out: 939ced8ecf0SOmar Sandoval if (wait && caching_ctl) 940ced8ecf0SOmar Sandoval ret = btrfs_caching_ctl_wait_done(cache, caching_ctl); 941e747853cSJosef Bacik if (caching_ctl) 942e747853cSJosef Bacik btrfs_put_caching_control(caching_ctl); 9439f21246dSJosef Bacik 9449f21246dSJosef Bacik return ret; 9459f21246dSJosef Bacik } 946e3e0520bSJosef Bacik 947e3e0520bSJosef Bacik static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) 948e3e0520bSJosef Bacik { 949e3e0520bSJosef Bacik u64 extra_flags = chunk_to_extended(flags) & 950e3e0520bSJosef Bacik BTRFS_EXTENDED_PROFILE_MASK; 951e3e0520bSJosef Bacik 952e3e0520bSJosef Bacik write_seqlock(&fs_info->profiles_lock); 953e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 954e3e0520bSJosef Bacik fs_info->avail_data_alloc_bits &= ~extra_flags; 955e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_METADATA) 956e3e0520bSJosef Bacik fs_info->avail_metadata_alloc_bits &= ~extra_flags; 957e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 958e3e0520bSJosef Bacik fs_info->avail_system_alloc_bits &= ~extra_flags; 959e3e0520bSJosef Bacik write_sequnlock(&fs_info->profiles_lock); 960e3e0520bSJosef Bacik } 961e3e0520bSJosef Bacik 962e3e0520bSJosef Bacik /* 963e3e0520bSJosef Bacik * Clear incompat bits for the following feature(s): 964e3e0520bSJosef Bacik * 965e3e0520bSJosef Bacik * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group 966e3e0520bSJosef Bacik * in the whole filesystem 9679c907446SDavid Sterba * 9689c907446SDavid Sterba * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups 969e3e0520bSJosef Bacik */ 970e3e0520bSJosef Bacik static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) 971e3e0520bSJosef Bacik { 9729c907446SDavid Sterba bool found_raid56 = false; 9739c907446SDavid Sterba bool found_raid1c34 = false; 9749c907446SDavid Sterba 9759c907446SDavid Sterba if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) || 9769c907446SDavid Sterba (flags & BTRFS_BLOCK_GROUP_RAID1C3) || 9779c907446SDavid Sterba (flags & BTRFS_BLOCK_GROUP_RAID1C4)) { 978e3e0520bSJosef Bacik struct list_head *head = &fs_info->space_info; 979e3e0520bSJosef Bacik struct btrfs_space_info *sinfo; 980e3e0520bSJosef Bacik 981e3e0520bSJosef Bacik list_for_each_entry_rcu(sinfo, head, list) { 982e3e0520bSJosef Bacik down_read(&sinfo->groups_sem); 983e3e0520bSJosef Bacik if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) 9849c907446SDavid Sterba found_raid56 = true; 985e3e0520bSJosef Bacik if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) 9869c907446SDavid Sterba found_raid56 = true; 9879c907446SDavid Sterba if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3])) 9889c907446SDavid Sterba found_raid1c34 = true; 9899c907446SDavid Sterba if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4])) 9909c907446SDavid Sterba found_raid1c34 = true; 991e3e0520bSJosef Bacik up_read(&sinfo->groups_sem); 992e3e0520bSJosef Bacik } 993d8e6fd5cSFilipe Manana if (!found_raid56) 994e3e0520bSJosef Bacik btrfs_clear_fs_incompat(fs_info, RAID56); 995d8e6fd5cSFilipe Manana if (!found_raid1c34) 9969c907446SDavid Sterba btrfs_clear_fs_incompat(fs_info, RAID1C34); 997e3e0520bSJosef Bacik } 998e3e0520bSJosef Bacik } 999e3e0520bSJosef Bacik 10007357623aSQu Wenruo static int remove_block_group_item(struct btrfs_trans_handle *trans, 10017357623aSQu Wenruo struct btrfs_path *path, 10027357623aSQu Wenruo struct btrfs_block_group *block_group) 10037357623aSQu Wenruo { 10047357623aSQu Wenruo struct btrfs_fs_info *fs_info = trans->fs_info; 10057357623aSQu Wenruo struct btrfs_root *root; 10067357623aSQu Wenruo struct btrfs_key key; 10077357623aSQu Wenruo int ret; 10087357623aSQu Wenruo 1009dfe8aec4SJosef Bacik root = btrfs_block_group_root(fs_info); 10107357623aSQu Wenruo key.objectid = block_group->start; 10117357623aSQu Wenruo key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 10127357623aSQu Wenruo key.offset = block_group->length; 10137357623aSQu Wenruo 10147357623aSQu Wenruo ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 10157357623aSQu Wenruo if (ret > 0) 10167357623aSQu Wenruo ret = -ENOENT; 10177357623aSQu Wenruo if (ret < 0) 10187357623aSQu Wenruo return ret; 10197357623aSQu Wenruo 10207357623aSQu Wenruo ret = btrfs_del_item(trans, root, path); 10217357623aSQu Wenruo return ret; 10227357623aSQu Wenruo } 10237357623aSQu Wenruo 1024e3e0520bSJosef Bacik int btrfs_remove_block_group(struct btrfs_trans_handle *trans, 1025e3e0520bSJosef Bacik u64 group_start, struct extent_map *em) 1026e3e0520bSJosef Bacik { 1027e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 1028e3e0520bSJosef Bacik struct btrfs_path *path; 102932da5386SDavid Sterba struct btrfs_block_group *block_group; 1030e3e0520bSJosef Bacik struct btrfs_free_cluster *cluster; 1031e3e0520bSJosef Bacik struct inode *inode; 1032e3e0520bSJosef Bacik struct kobject *kobj = NULL; 1033e3e0520bSJosef Bacik int ret; 1034e3e0520bSJosef Bacik int index; 1035e3e0520bSJosef Bacik int factor; 1036e3e0520bSJosef Bacik struct btrfs_caching_control *caching_ctl = NULL; 1037e3e0520bSJosef Bacik bool remove_em; 1038e3e0520bSJosef Bacik bool remove_rsv = false; 1039e3e0520bSJosef Bacik 1040e3e0520bSJosef Bacik block_group = btrfs_lookup_block_group(fs_info, group_start); 1041e3e0520bSJosef Bacik BUG_ON(!block_group); 1042e3e0520bSJosef Bacik BUG_ON(!block_group->ro); 1043e3e0520bSJosef Bacik 1044e3e0520bSJosef Bacik trace_btrfs_remove_block_group(block_group); 1045e3e0520bSJosef Bacik /* 1046e3e0520bSJosef Bacik * Free the reserved super bytes from this block group before 1047e3e0520bSJosef Bacik * remove it. 1048e3e0520bSJosef Bacik */ 1049e3e0520bSJosef Bacik btrfs_free_excluded_extents(block_group); 1050b3470b5dSDavid Sterba btrfs_free_ref_tree_range(fs_info, block_group->start, 1051b3470b5dSDavid Sterba block_group->length); 1052e3e0520bSJosef Bacik 1053e3e0520bSJosef Bacik index = btrfs_bg_flags_to_raid_index(block_group->flags); 1054e3e0520bSJosef Bacik factor = btrfs_bg_type_to_factor(block_group->flags); 1055e3e0520bSJosef Bacik 1056e3e0520bSJosef Bacik /* make sure this block group isn't part of an allocation cluster */ 1057e3e0520bSJosef Bacik cluster = &fs_info->data_alloc_cluster; 1058e3e0520bSJosef Bacik spin_lock(&cluster->refill_lock); 1059e3e0520bSJosef Bacik btrfs_return_cluster_to_free_space(block_group, cluster); 1060e3e0520bSJosef Bacik spin_unlock(&cluster->refill_lock); 1061e3e0520bSJosef Bacik 1062e3e0520bSJosef Bacik /* 1063e3e0520bSJosef Bacik * make sure this block group isn't part of a metadata 1064e3e0520bSJosef Bacik * allocation cluster 1065e3e0520bSJosef Bacik */ 1066e3e0520bSJosef Bacik cluster = &fs_info->meta_alloc_cluster; 1067e3e0520bSJosef Bacik spin_lock(&cluster->refill_lock); 1068e3e0520bSJosef Bacik btrfs_return_cluster_to_free_space(block_group, cluster); 1069e3e0520bSJosef Bacik spin_unlock(&cluster->refill_lock); 1070e3e0520bSJosef Bacik 107140ab3be1SNaohiro Aota btrfs_clear_treelog_bg(block_group); 1072c2707a25SJohannes Thumshirn btrfs_clear_data_reloc_bg(block_group); 107340ab3be1SNaohiro Aota 1074e3e0520bSJosef Bacik path = btrfs_alloc_path(); 1075e3e0520bSJosef Bacik if (!path) { 1076e3e0520bSJosef Bacik ret = -ENOMEM; 10779fecd132SFilipe Manana goto out; 1078e3e0520bSJosef Bacik } 1079e3e0520bSJosef Bacik 1080e3e0520bSJosef Bacik /* 1081e3e0520bSJosef Bacik * get the inode first so any iput calls done for the io_list 1082e3e0520bSJosef Bacik * aren't the final iput (no unlinks allowed now) 1083e3e0520bSJosef Bacik */ 1084e3e0520bSJosef Bacik inode = lookup_free_space_inode(block_group, path); 1085e3e0520bSJosef Bacik 1086e3e0520bSJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 1087e3e0520bSJosef Bacik /* 1088e3e0520bSJosef Bacik * Make sure our free space cache IO is done before removing the 1089e3e0520bSJosef Bacik * free space inode 1090e3e0520bSJosef Bacik */ 1091e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1092e3e0520bSJosef Bacik if (!list_empty(&block_group->io_list)) { 1093e3e0520bSJosef Bacik list_del_init(&block_group->io_list); 1094e3e0520bSJosef Bacik 1095e3e0520bSJosef Bacik WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); 1096e3e0520bSJosef Bacik 1097e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1098e3e0520bSJosef Bacik btrfs_wait_cache_io(trans, block_group, path); 1099e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1100e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1101e3e0520bSJosef Bacik } 1102e3e0520bSJosef Bacik 1103e3e0520bSJosef Bacik if (!list_empty(&block_group->dirty_list)) { 1104e3e0520bSJosef Bacik list_del_init(&block_group->dirty_list); 1105e3e0520bSJosef Bacik remove_rsv = true; 1106e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1107e3e0520bSJosef Bacik } 1108e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1109e3e0520bSJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 1110e3e0520bSJosef Bacik 111136b216c8SBoris Burkov ret = btrfs_remove_free_space_inode(trans, inode, block_group); 1112e3e0520bSJosef Bacik if (ret) 11139fecd132SFilipe Manana goto out; 1114e3e0520bSJosef Bacik 111516b0c258SFilipe Manana write_lock(&fs_info->block_group_cache_lock); 111608dddb29SFilipe Manana rb_erase_cached(&block_group->cache_node, 1117e3e0520bSJosef Bacik &fs_info->block_group_cache_tree); 1118e3e0520bSJosef Bacik RB_CLEAR_NODE(&block_group->cache_node); 1119e3e0520bSJosef Bacik 11209fecd132SFilipe Manana /* Once for the block groups rbtree */ 11219fecd132SFilipe Manana btrfs_put_block_group(block_group); 11229fecd132SFilipe Manana 112316b0c258SFilipe Manana write_unlock(&fs_info->block_group_cache_lock); 1124e3e0520bSJosef Bacik 1125e3e0520bSJosef Bacik down_write(&block_group->space_info->groups_sem); 1126e3e0520bSJosef Bacik /* 1127e3e0520bSJosef Bacik * we must use list_del_init so people can check to see if they 1128e3e0520bSJosef Bacik * are still on the list after taking the semaphore 1129e3e0520bSJosef Bacik */ 1130e3e0520bSJosef Bacik list_del_init(&block_group->list); 1131e3e0520bSJosef Bacik if (list_empty(&block_group->space_info->block_groups[index])) { 1132e3e0520bSJosef Bacik kobj = block_group->space_info->block_group_kobjs[index]; 1133e3e0520bSJosef Bacik block_group->space_info->block_group_kobjs[index] = NULL; 1134e3e0520bSJosef Bacik clear_avail_alloc_bits(fs_info, block_group->flags); 1135e3e0520bSJosef Bacik } 1136e3e0520bSJosef Bacik up_write(&block_group->space_info->groups_sem); 1137e3e0520bSJosef Bacik clear_incompat_bg_bits(fs_info, block_group->flags); 1138e3e0520bSJosef Bacik if (kobj) { 1139e3e0520bSJosef Bacik kobject_del(kobj); 1140e3e0520bSJosef Bacik kobject_put(kobj); 1141e3e0520bSJosef Bacik } 1142e3e0520bSJosef Bacik 1143e3e0520bSJosef Bacik if (block_group->cached == BTRFS_CACHE_STARTED) 1144e3e0520bSJosef Bacik btrfs_wait_block_group_cache_done(block_group); 11457b9c293bSJosef Bacik 114616b0c258SFilipe Manana write_lock(&fs_info->block_group_cache_lock); 11477b9c293bSJosef Bacik caching_ctl = btrfs_get_caching_control(block_group); 1148e3e0520bSJosef Bacik if (!caching_ctl) { 1149e3e0520bSJosef Bacik struct btrfs_caching_control *ctl; 1150e3e0520bSJosef Bacik 11517b9c293bSJosef Bacik list_for_each_entry(ctl, &fs_info->caching_block_groups, list) { 1152e3e0520bSJosef Bacik if (ctl->block_group == block_group) { 1153e3e0520bSJosef Bacik caching_ctl = ctl; 1154e3e0520bSJosef Bacik refcount_inc(&caching_ctl->count); 1155e3e0520bSJosef Bacik break; 1156e3e0520bSJosef Bacik } 1157e3e0520bSJosef Bacik } 11587b9c293bSJosef Bacik } 1159e3e0520bSJosef Bacik if (caching_ctl) 1160e3e0520bSJosef Bacik list_del_init(&caching_ctl->list); 116116b0c258SFilipe Manana write_unlock(&fs_info->block_group_cache_lock); 11627b9c293bSJosef Bacik 1163e3e0520bSJosef Bacik if (caching_ctl) { 1164e3e0520bSJosef Bacik /* Once for the caching bgs list and once for us. */ 1165e3e0520bSJosef Bacik btrfs_put_caching_control(caching_ctl); 1166e3e0520bSJosef Bacik btrfs_put_caching_control(caching_ctl); 1167e3e0520bSJosef Bacik } 1168e3e0520bSJosef Bacik 1169e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1170e3e0520bSJosef Bacik WARN_ON(!list_empty(&block_group->dirty_list)); 1171e3e0520bSJosef Bacik WARN_ON(!list_empty(&block_group->io_list)); 1172e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1173e3e0520bSJosef Bacik 1174e3e0520bSJosef Bacik btrfs_remove_free_space_cache(block_group); 1175e3e0520bSJosef Bacik 1176e3e0520bSJosef Bacik spin_lock(&block_group->space_info->lock); 1177e3e0520bSJosef Bacik list_del_init(&block_group->ro_list); 1178e3e0520bSJosef Bacik 1179e3e0520bSJosef Bacik if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { 1180e3e0520bSJosef Bacik WARN_ON(block_group->space_info->total_bytes 1181b3470b5dSDavid Sterba < block_group->length); 1182e3e0520bSJosef Bacik WARN_ON(block_group->space_info->bytes_readonly 1183169e0da9SNaohiro Aota < block_group->length - block_group->zone_unusable); 1184169e0da9SNaohiro Aota WARN_ON(block_group->space_info->bytes_zone_unusable 1185169e0da9SNaohiro Aota < block_group->zone_unusable); 1186e3e0520bSJosef Bacik WARN_ON(block_group->space_info->disk_total 1187b3470b5dSDavid Sterba < block_group->length * factor); 11883349b57fSJosef Bacik WARN_ON(test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, 11893349b57fSJosef Bacik &block_group->runtime_flags) && 11906a921de5SNaohiro Aota block_group->space_info->active_total_bytes 11916a921de5SNaohiro Aota < block_group->length); 1192e3e0520bSJosef Bacik } 1193b3470b5dSDavid Sterba block_group->space_info->total_bytes -= block_group->length; 11943349b57fSJosef Bacik if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) 11956a921de5SNaohiro Aota block_group->space_info->active_total_bytes -= block_group->length; 1196169e0da9SNaohiro Aota block_group->space_info->bytes_readonly -= 1197169e0da9SNaohiro Aota (block_group->length - block_group->zone_unusable); 1198169e0da9SNaohiro Aota block_group->space_info->bytes_zone_unusable -= 1199169e0da9SNaohiro Aota block_group->zone_unusable; 1200b3470b5dSDavid Sterba block_group->space_info->disk_total -= block_group->length * factor; 1201e3e0520bSJosef Bacik 1202e3e0520bSJosef Bacik spin_unlock(&block_group->space_info->lock); 1203e3e0520bSJosef Bacik 1204ffcb9d44SFilipe Manana /* 1205ffcb9d44SFilipe Manana * Remove the free space for the block group from the free space tree 1206ffcb9d44SFilipe Manana * and the block group's item from the extent tree before marking the 1207ffcb9d44SFilipe Manana * block group as removed. This is to prevent races with tasks that 1208ffcb9d44SFilipe Manana * freeze and unfreeze a block group, this task and another task 1209ffcb9d44SFilipe Manana * allocating a new block group - the unfreeze task ends up removing 1210ffcb9d44SFilipe Manana * the block group's extent map before the task calling this function 1211ffcb9d44SFilipe Manana * deletes the block group item from the extent tree, allowing for 1212ffcb9d44SFilipe Manana * another task to attempt to create another block group with the same 1213ffcb9d44SFilipe Manana * item key (and failing with -EEXIST and a transaction abort). 1214ffcb9d44SFilipe Manana */ 1215ffcb9d44SFilipe Manana ret = remove_block_group_free_space(trans, block_group); 1216ffcb9d44SFilipe Manana if (ret) 1217ffcb9d44SFilipe Manana goto out; 1218ffcb9d44SFilipe Manana 1219ffcb9d44SFilipe Manana ret = remove_block_group_item(trans, path, block_group); 1220ffcb9d44SFilipe Manana if (ret < 0) 1221ffcb9d44SFilipe Manana goto out; 1222ffcb9d44SFilipe Manana 1223e3e0520bSJosef Bacik spin_lock(&block_group->lock); 12243349b57fSJosef Bacik set_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags); 12253349b57fSJosef Bacik 1226e3e0520bSJosef Bacik /* 12276b7304afSFilipe Manana * At this point trimming or scrub can't start on this block group, 12286b7304afSFilipe Manana * because we removed the block group from the rbtree 12296b7304afSFilipe Manana * fs_info->block_group_cache_tree so no one can't find it anymore and 12306b7304afSFilipe Manana * even if someone already got this block group before we removed it 12316b7304afSFilipe Manana * from the rbtree, they have already incremented block_group->frozen - 12326b7304afSFilipe Manana * if they didn't, for the trimming case they won't find any free space 12336b7304afSFilipe Manana * entries because we already removed them all when we called 12346b7304afSFilipe Manana * btrfs_remove_free_space_cache(). 1235e3e0520bSJosef Bacik * 1236e3e0520bSJosef Bacik * And we must not remove the extent map from the fs_info->mapping_tree 1237e3e0520bSJosef Bacik * to prevent the same logical address range and physical device space 12386b7304afSFilipe Manana * ranges from being reused for a new block group. This is needed to 12396b7304afSFilipe Manana * avoid races with trimming and scrub. 12406b7304afSFilipe Manana * 12416b7304afSFilipe Manana * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is 1242e3e0520bSJosef Bacik * completely transactionless, so while it is trimming a range the 1243e3e0520bSJosef Bacik * currently running transaction might finish and a new one start, 1244e3e0520bSJosef Bacik * allowing for new block groups to be created that can reuse the same 1245e3e0520bSJosef Bacik * physical device locations unless we take this special care. 1246e3e0520bSJosef Bacik * 1247e3e0520bSJosef Bacik * There may also be an implicit trim operation if the file system 1248e3e0520bSJosef Bacik * is mounted with -odiscard. The same protections must remain 1249e3e0520bSJosef Bacik * in place until the extents have been discarded completely when 1250e3e0520bSJosef Bacik * the transaction commit has completed. 1251e3e0520bSJosef Bacik */ 12526b7304afSFilipe Manana remove_em = (atomic_read(&block_group->frozen) == 0); 1253e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1254e3e0520bSJosef Bacik 1255e3e0520bSJosef Bacik if (remove_em) { 1256e3e0520bSJosef Bacik struct extent_map_tree *em_tree; 1257e3e0520bSJosef Bacik 1258e3e0520bSJosef Bacik em_tree = &fs_info->mapping_tree; 1259e3e0520bSJosef Bacik write_lock(&em_tree->lock); 1260e3e0520bSJosef Bacik remove_extent_mapping(em_tree, em); 1261e3e0520bSJosef Bacik write_unlock(&em_tree->lock); 1262e3e0520bSJosef Bacik /* once for the tree */ 1263e3e0520bSJosef Bacik free_extent_map(em); 1264e3e0520bSJosef Bacik } 1265f6033c5eSXiyu Yang 12669fecd132SFilipe Manana out: 1267f6033c5eSXiyu Yang /* Once for the lookup reference */ 1268f6033c5eSXiyu Yang btrfs_put_block_group(block_group); 1269e3e0520bSJosef Bacik if (remove_rsv) 1270e3e0520bSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 1271e3e0520bSJosef Bacik btrfs_free_path(path); 1272e3e0520bSJosef Bacik return ret; 1273e3e0520bSJosef Bacik } 1274e3e0520bSJosef Bacik 1275e3e0520bSJosef Bacik struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( 1276e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info, const u64 chunk_offset) 1277e3e0520bSJosef Bacik { 1278dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 1279e3e0520bSJosef Bacik struct extent_map_tree *em_tree = &fs_info->mapping_tree; 1280e3e0520bSJosef Bacik struct extent_map *em; 1281e3e0520bSJosef Bacik struct map_lookup *map; 1282e3e0520bSJosef Bacik unsigned int num_items; 1283e3e0520bSJosef Bacik 1284e3e0520bSJosef Bacik read_lock(&em_tree->lock); 1285e3e0520bSJosef Bacik em = lookup_extent_mapping(em_tree, chunk_offset, 1); 1286e3e0520bSJosef Bacik read_unlock(&em_tree->lock); 1287e3e0520bSJosef Bacik ASSERT(em && em->start == chunk_offset); 1288e3e0520bSJosef Bacik 1289e3e0520bSJosef Bacik /* 1290e3e0520bSJosef Bacik * We need to reserve 3 + N units from the metadata space info in order 1291e3e0520bSJosef Bacik * to remove a block group (done at btrfs_remove_chunk() and at 1292e3e0520bSJosef Bacik * btrfs_remove_block_group()), which are used for: 1293e3e0520bSJosef Bacik * 1294e3e0520bSJosef Bacik * 1 unit for adding the free space inode's orphan (located in the tree 1295e3e0520bSJosef Bacik * of tree roots). 1296e3e0520bSJosef Bacik * 1 unit for deleting the block group item (located in the extent 1297e3e0520bSJosef Bacik * tree). 1298e3e0520bSJosef Bacik * 1 unit for deleting the free space item (located in tree of tree 1299e3e0520bSJosef Bacik * roots). 1300e3e0520bSJosef Bacik * N units for deleting N device extent items corresponding to each 1301e3e0520bSJosef Bacik * stripe (located in the device tree). 1302e3e0520bSJosef Bacik * 1303e3e0520bSJosef Bacik * In order to remove a block group we also need to reserve units in the 1304e3e0520bSJosef Bacik * system space info in order to update the chunk tree (update one or 1305e3e0520bSJosef Bacik * more device items and remove one chunk item), but this is done at 1306e3e0520bSJosef Bacik * btrfs_remove_chunk() through a call to check_system_chunk(). 1307e3e0520bSJosef Bacik */ 1308e3e0520bSJosef Bacik map = em->map_lookup; 1309e3e0520bSJosef Bacik num_items = 3 + map->num_stripes; 1310e3e0520bSJosef Bacik free_extent_map(em); 1311e3e0520bSJosef Bacik 1312dfe8aec4SJosef Bacik return btrfs_start_transaction_fallback_global_rsv(root, num_items); 1313e3e0520bSJosef Bacik } 1314e3e0520bSJosef Bacik 1315e3e0520bSJosef Bacik /* 131626ce2095SJosef Bacik * Mark block group @cache read-only, so later write won't happen to block 131726ce2095SJosef Bacik * group @cache. 131826ce2095SJosef Bacik * 131926ce2095SJosef Bacik * If @force is not set, this function will only mark the block group readonly 132026ce2095SJosef Bacik * if we have enough free space (1M) in other metadata/system block groups. 132126ce2095SJosef Bacik * If @force is not set, this function will mark the block group readonly 132226ce2095SJosef Bacik * without checking free space. 132326ce2095SJosef Bacik * 132426ce2095SJosef Bacik * NOTE: This function doesn't care if other block groups can contain all the 132526ce2095SJosef Bacik * data in this block group. That check should be done by relocation routine, 132626ce2095SJosef Bacik * not this function. 132726ce2095SJosef Bacik */ 132832da5386SDavid Sterba static int inc_block_group_ro(struct btrfs_block_group *cache, int force) 132926ce2095SJosef Bacik { 133026ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 133126ce2095SJosef Bacik u64 num_bytes; 133226ce2095SJosef Bacik int ret = -ENOSPC; 133326ce2095SJosef Bacik 133426ce2095SJosef Bacik spin_lock(&sinfo->lock); 133526ce2095SJosef Bacik spin_lock(&cache->lock); 133626ce2095SJosef Bacik 1337195a49eaSFilipe Manana if (cache->swap_extents) { 1338195a49eaSFilipe Manana ret = -ETXTBSY; 1339195a49eaSFilipe Manana goto out; 1340195a49eaSFilipe Manana } 1341195a49eaSFilipe Manana 134226ce2095SJosef Bacik if (cache->ro) { 134326ce2095SJosef Bacik cache->ro++; 134426ce2095SJosef Bacik ret = 0; 134526ce2095SJosef Bacik goto out; 134626ce2095SJosef Bacik } 134726ce2095SJosef Bacik 1348b3470b5dSDavid Sterba num_bytes = cache->length - cache->reserved - cache->pinned - 1349169e0da9SNaohiro Aota cache->bytes_super - cache->zone_unusable - cache->used; 135026ce2095SJosef Bacik 135126ce2095SJosef Bacik /* 1352a30a3d20SJosef Bacik * Data never overcommits, even in mixed mode, so do just the straight 1353a30a3d20SJosef Bacik * check of left over space in how much we have allocated. 1354a30a3d20SJosef Bacik */ 1355a30a3d20SJosef Bacik if (force) { 1356a30a3d20SJosef Bacik ret = 0; 1357a30a3d20SJosef Bacik } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) { 1358a30a3d20SJosef Bacik u64 sinfo_used = btrfs_space_info_used(sinfo, true); 1359a30a3d20SJosef Bacik 1360a30a3d20SJosef Bacik /* 136126ce2095SJosef Bacik * Here we make sure if we mark this bg RO, we still have enough 1362f8935566SJosef Bacik * free space as buffer. 136326ce2095SJosef Bacik */ 1364a30a3d20SJosef Bacik if (sinfo_used + num_bytes <= sinfo->total_bytes) 1365a30a3d20SJosef Bacik ret = 0; 1366a30a3d20SJosef Bacik } else { 1367a30a3d20SJosef Bacik /* 1368a30a3d20SJosef Bacik * We overcommit metadata, so we need to do the 1369a30a3d20SJosef Bacik * btrfs_can_overcommit check here, and we need to pass in 1370a30a3d20SJosef Bacik * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of 1371a30a3d20SJosef Bacik * leeway to allow us to mark this block group as read only. 1372a30a3d20SJosef Bacik */ 1373a30a3d20SJosef Bacik if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes, 1374a30a3d20SJosef Bacik BTRFS_RESERVE_NO_FLUSH)) 1375a30a3d20SJosef Bacik ret = 0; 1376a30a3d20SJosef Bacik } 1377a30a3d20SJosef Bacik 1378a30a3d20SJosef Bacik if (!ret) { 137926ce2095SJosef Bacik sinfo->bytes_readonly += num_bytes; 1380169e0da9SNaohiro Aota if (btrfs_is_zoned(cache->fs_info)) { 1381169e0da9SNaohiro Aota /* Migrate zone_unusable bytes to readonly */ 1382169e0da9SNaohiro Aota sinfo->bytes_readonly += cache->zone_unusable; 1383169e0da9SNaohiro Aota sinfo->bytes_zone_unusable -= cache->zone_unusable; 1384169e0da9SNaohiro Aota cache->zone_unusable = 0; 1385169e0da9SNaohiro Aota } 138626ce2095SJosef Bacik cache->ro++; 138726ce2095SJosef Bacik list_add_tail(&cache->ro_list, &sinfo->ro_bgs); 138826ce2095SJosef Bacik } 138926ce2095SJosef Bacik out: 139026ce2095SJosef Bacik spin_unlock(&cache->lock); 139126ce2095SJosef Bacik spin_unlock(&sinfo->lock); 139226ce2095SJosef Bacik if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { 139326ce2095SJosef Bacik btrfs_info(cache->fs_info, 1394b3470b5dSDavid Sterba "unable to make block group %llu ro", cache->start); 139526ce2095SJosef Bacik btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); 139626ce2095SJosef Bacik } 139726ce2095SJosef Bacik return ret; 139826ce2095SJosef Bacik } 139926ce2095SJosef Bacik 1400fe119a6eSNikolay Borisov static bool clean_pinned_extents(struct btrfs_trans_handle *trans, 1401fe119a6eSNikolay Borisov struct btrfs_block_group *bg) 140245bb5d6aSNikolay Borisov { 140345bb5d6aSNikolay Borisov struct btrfs_fs_info *fs_info = bg->fs_info; 1404fe119a6eSNikolay Borisov struct btrfs_transaction *prev_trans = NULL; 140545bb5d6aSNikolay Borisov const u64 start = bg->start; 140645bb5d6aSNikolay Borisov const u64 end = start + bg->length - 1; 140745bb5d6aSNikolay Borisov int ret; 140845bb5d6aSNikolay Borisov 1409fe119a6eSNikolay Borisov spin_lock(&fs_info->trans_lock); 1410fe119a6eSNikolay Borisov if (trans->transaction->list.prev != &fs_info->trans_list) { 1411fe119a6eSNikolay Borisov prev_trans = list_last_entry(&trans->transaction->list, 1412fe119a6eSNikolay Borisov struct btrfs_transaction, list); 1413fe119a6eSNikolay Borisov refcount_inc(&prev_trans->use_count); 1414fe119a6eSNikolay Borisov } 1415fe119a6eSNikolay Borisov spin_unlock(&fs_info->trans_lock); 1416fe119a6eSNikolay Borisov 141745bb5d6aSNikolay Borisov /* 141845bb5d6aSNikolay Borisov * Hold the unused_bg_unpin_mutex lock to avoid racing with 141945bb5d6aSNikolay Borisov * btrfs_finish_extent_commit(). If we are at transaction N, another 142045bb5d6aSNikolay Borisov * task might be running finish_extent_commit() for the previous 142145bb5d6aSNikolay Borisov * transaction N - 1, and have seen a range belonging to the block 1422fe119a6eSNikolay Borisov * group in pinned_extents before we were able to clear the whole block 1423fe119a6eSNikolay Borisov * group range from pinned_extents. This means that task can lookup for 1424fe119a6eSNikolay Borisov * the block group after we unpinned it from pinned_extents and removed 1425fe119a6eSNikolay Borisov * it, leading to a BUG_ON() at unpin_extent_range(). 142645bb5d6aSNikolay Borisov */ 142745bb5d6aSNikolay Borisov mutex_lock(&fs_info->unused_bg_unpin_mutex); 1428fe119a6eSNikolay Borisov if (prev_trans) { 1429fe119a6eSNikolay Borisov ret = clear_extent_bits(&prev_trans->pinned_extents, start, end, 143045bb5d6aSNikolay Borisov EXTENT_DIRTY); 143145bb5d6aSNikolay Borisov if (ret) 1432534cf531SFilipe Manana goto out; 1433fe119a6eSNikolay Borisov } 143445bb5d6aSNikolay Borisov 1435fe119a6eSNikolay Borisov ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end, 143645bb5d6aSNikolay Borisov EXTENT_DIRTY); 1437534cf531SFilipe Manana out: 143845bb5d6aSNikolay Borisov mutex_unlock(&fs_info->unused_bg_unpin_mutex); 14395150bf19SFilipe Manana if (prev_trans) 14405150bf19SFilipe Manana btrfs_put_transaction(prev_trans); 144145bb5d6aSNikolay Borisov 1442534cf531SFilipe Manana return ret == 0; 144345bb5d6aSNikolay Borisov } 144445bb5d6aSNikolay Borisov 144526ce2095SJosef Bacik /* 1446e3e0520bSJosef Bacik * Process the unused_bgs list and remove any that don't have any allocated 1447e3e0520bSJosef Bacik * space inside of them. 1448e3e0520bSJosef Bacik */ 1449e3e0520bSJosef Bacik void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) 1450e3e0520bSJosef Bacik { 145132da5386SDavid Sterba struct btrfs_block_group *block_group; 1452e3e0520bSJosef Bacik struct btrfs_space_info *space_info; 1453e3e0520bSJosef Bacik struct btrfs_trans_handle *trans; 14546e80d4f8SDennis Zhou const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC); 1455e3e0520bSJosef Bacik int ret = 0; 1456e3e0520bSJosef Bacik 1457e3e0520bSJosef Bacik if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 1458e3e0520bSJosef Bacik return; 1459e3e0520bSJosef Bacik 14602f12741fSJosef Bacik if (btrfs_fs_closing(fs_info)) 14612f12741fSJosef Bacik return; 14622f12741fSJosef Bacik 1463ddfd08cbSJosef Bacik /* 1464ddfd08cbSJosef Bacik * Long running balances can keep us blocked here for eternity, so 1465ddfd08cbSJosef Bacik * simply skip deletion if we're unable to get the mutex. 1466ddfd08cbSJosef Bacik */ 1467f3372065SJohannes Thumshirn if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) 1468ddfd08cbSJosef Bacik return; 1469ddfd08cbSJosef Bacik 1470e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1471e3e0520bSJosef Bacik while (!list_empty(&fs_info->unused_bgs)) { 1472e3e0520bSJosef Bacik int trimming; 1473e3e0520bSJosef Bacik 1474e3e0520bSJosef Bacik block_group = list_first_entry(&fs_info->unused_bgs, 147532da5386SDavid Sterba struct btrfs_block_group, 1476e3e0520bSJosef Bacik bg_list); 1477e3e0520bSJosef Bacik list_del_init(&block_group->bg_list); 1478e3e0520bSJosef Bacik 1479e3e0520bSJosef Bacik space_info = block_group->space_info; 1480e3e0520bSJosef Bacik 1481e3e0520bSJosef Bacik if (ret || btrfs_mixed_space_info(space_info)) { 1482e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1483e3e0520bSJosef Bacik continue; 1484e3e0520bSJosef Bacik } 1485e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1486e3e0520bSJosef Bacik 1487b0643e59SDennis Zhou btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 1488b0643e59SDennis Zhou 1489e3e0520bSJosef Bacik /* Don't want to race with allocators so take the groups_sem */ 1490e3e0520bSJosef Bacik down_write(&space_info->groups_sem); 14916e80d4f8SDennis Zhou 14926e80d4f8SDennis Zhou /* 14936e80d4f8SDennis Zhou * Async discard moves the final block group discard to be prior 14946e80d4f8SDennis Zhou * to the unused_bgs code path. Therefore, if it's not fully 14956e80d4f8SDennis Zhou * trimmed, punt it back to the async discard lists. 14966e80d4f8SDennis Zhou */ 14976e80d4f8SDennis Zhou if (btrfs_test_opt(fs_info, DISCARD_ASYNC) && 14986e80d4f8SDennis Zhou !btrfs_is_free_space_trimmed(block_group)) { 14996e80d4f8SDennis Zhou trace_btrfs_skip_unused_block_group(block_group); 15006e80d4f8SDennis Zhou up_write(&space_info->groups_sem); 15016e80d4f8SDennis Zhou /* Requeue if we failed because of async discard */ 15026e80d4f8SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 15036e80d4f8SDennis Zhou block_group); 15046e80d4f8SDennis Zhou goto next; 15056e80d4f8SDennis Zhou } 15066e80d4f8SDennis Zhou 1507e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1508e3e0520bSJosef Bacik if (block_group->reserved || block_group->pinned || 1509bf38be65SDavid Sterba block_group->used || block_group->ro || 1510e3e0520bSJosef Bacik list_is_singular(&block_group->list)) { 1511e3e0520bSJosef Bacik /* 1512e3e0520bSJosef Bacik * We want to bail if we made new allocations or have 1513e3e0520bSJosef Bacik * outstanding allocations in this block group. We do 1514e3e0520bSJosef Bacik * the ro check in case balance is currently acting on 1515e3e0520bSJosef Bacik * this block group. 1516e3e0520bSJosef Bacik */ 1517e3e0520bSJosef Bacik trace_btrfs_skip_unused_block_group(block_group); 1518e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1519e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1520e3e0520bSJosef Bacik goto next; 1521e3e0520bSJosef Bacik } 1522e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1523e3e0520bSJosef Bacik 1524e3e0520bSJosef Bacik /* We don't want to force the issue, only flip if it's ok. */ 1525e11c0406SJosef Bacik ret = inc_block_group_ro(block_group, 0); 1526e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1527e3e0520bSJosef Bacik if (ret < 0) { 1528e3e0520bSJosef Bacik ret = 0; 1529e3e0520bSJosef Bacik goto next; 1530e3e0520bSJosef Bacik } 1531e3e0520bSJosef Bacik 153274e91b12SNaohiro Aota ret = btrfs_zone_finish(block_group); 153374e91b12SNaohiro Aota if (ret < 0) { 153474e91b12SNaohiro Aota btrfs_dec_block_group_ro(block_group); 153574e91b12SNaohiro Aota if (ret == -EAGAIN) 153674e91b12SNaohiro Aota ret = 0; 153774e91b12SNaohiro Aota goto next; 153874e91b12SNaohiro Aota } 153974e91b12SNaohiro Aota 1540e3e0520bSJosef Bacik /* 1541e3e0520bSJosef Bacik * Want to do this before we do anything else so we can recover 1542e3e0520bSJosef Bacik * properly if we fail to join the transaction. 1543e3e0520bSJosef Bacik */ 1544e3e0520bSJosef Bacik trans = btrfs_start_trans_remove_block_group(fs_info, 1545b3470b5dSDavid Sterba block_group->start); 1546e3e0520bSJosef Bacik if (IS_ERR(trans)) { 1547e3e0520bSJosef Bacik btrfs_dec_block_group_ro(block_group); 1548e3e0520bSJosef Bacik ret = PTR_ERR(trans); 1549e3e0520bSJosef Bacik goto next; 1550e3e0520bSJosef Bacik } 1551e3e0520bSJosef Bacik 1552e3e0520bSJosef Bacik /* 1553e3e0520bSJosef Bacik * We could have pending pinned extents for this block group, 1554e3e0520bSJosef Bacik * just delete them, we don't care about them anymore. 1555e3e0520bSJosef Bacik */ 1556534cf531SFilipe Manana if (!clean_pinned_extents(trans, block_group)) { 1557534cf531SFilipe Manana btrfs_dec_block_group_ro(block_group); 1558e3e0520bSJosef Bacik goto end_trans; 1559534cf531SFilipe Manana } 1560e3e0520bSJosef Bacik 1561b0643e59SDennis Zhou /* 1562b0643e59SDennis Zhou * At this point, the block_group is read only and should fail 1563b0643e59SDennis Zhou * new allocations. However, btrfs_finish_extent_commit() can 1564b0643e59SDennis Zhou * cause this block_group to be placed back on the discard 1565b0643e59SDennis Zhou * lists because now the block_group isn't fully discarded. 1566b0643e59SDennis Zhou * Bail here and try again later after discarding everything. 1567b0643e59SDennis Zhou */ 1568b0643e59SDennis Zhou spin_lock(&fs_info->discard_ctl.lock); 1569b0643e59SDennis Zhou if (!list_empty(&block_group->discard_list)) { 1570b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1571b0643e59SDennis Zhou btrfs_dec_block_group_ro(block_group); 1572b0643e59SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 1573b0643e59SDennis Zhou block_group); 1574b0643e59SDennis Zhou goto end_trans; 1575b0643e59SDennis Zhou } 1576b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1577b0643e59SDennis Zhou 1578e3e0520bSJosef Bacik /* Reset pinned so btrfs_put_block_group doesn't complain */ 1579e3e0520bSJosef Bacik spin_lock(&space_info->lock); 1580e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1581e3e0520bSJosef Bacik 1582e3e0520bSJosef Bacik btrfs_space_info_update_bytes_pinned(fs_info, space_info, 1583e3e0520bSJosef Bacik -block_group->pinned); 1584e3e0520bSJosef Bacik space_info->bytes_readonly += block_group->pinned; 1585e3e0520bSJosef Bacik block_group->pinned = 0; 1586e3e0520bSJosef Bacik 1587e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1588e3e0520bSJosef Bacik spin_unlock(&space_info->lock); 1589e3e0520bSJosef Bacik 15906e80d4f8SDennis Zhou /* 15916e80d4f8SDennis Zhou * The normal path here is an unused block group is passed here, 15926e80d4f8SDennis Zhou * then trimming is handled in the transaction commit path. 15936e80d4f8SDennis Zhou * Async discard interposes before this to do the trimming 15946e80d4f8SDennis Zhou * before coming down the unused block group path as trimming 15956e80d4f8SDennis Zhou * will no longer be done later in the transaction commit path. 15966e80d4f8SDennis Zhou */ 15976e80d4f8SDennis Zhou if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC)) 15986e80d4f8SDennis Zhou goto flip_async; 15996e80d4f8SDennis Zhou 1600dcba6e48SNaohiro Aota /* 1601dcba6e48SNaohiro Aota * DISCARD can flip during remount. On zoned filesystems, we 1602dcba6e48SNaohiro Aota * need to reset sequential-required zones. 1603dcba6e48SNaohiro Aota */ 1604dcba6e48SNaohiro Aota trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) || 1605dcba6e48SNaohiro Aota btrfs_is_zoned(fs_info); 1606e3e0520bSJosef Bacik 1607e3e0520bSJosef Bacik /* Implicit trim during transaction commit. */ 1608e3e0520bSJosef Bacik if (trimming) 16096b7304afSFilipe Manana btrfs_freeze_block_group(block_group); 1610e3e0520bSJosef Bacik 1611e3e0520bSJosef Bacik /* 1612e3e0520bSJosef Bacik * Btrfs_remove_chunk will abort the transaction if things go 1613e3e0520bSJosef Bacik * horribly wrong. 1614e3e0520bSJosef Bacik */ 1615b3470b5dSDavid Sterba ret = btrfs_remove_chunk(trans, block_group->start); 1616e3e0520bSJosef Bacik 1617e3e0520bSJosef Bacik if (ret) { 1618e3e0520bSJosef Bacik if (trimming) 16196b7304afSFilipe Manana btrfs_unfreeze_block_group(block_group); 1620e3e0520bSJosef Bacik goto end_trans; 1621e3e0520bSJosef Bacik } 1622e3e0520bSJosef Bacik 1623e3e0520bSJosef Bacik /* 1624e3e0520bSJosef Bacik * If we're not mounted with -odiscard, we can just forget 1625e3e0520bSJosef Bacik * about this block group. Otherwise we'll need to wait 1626e3e0520bSJosef Bacik * until transaction commit to do the actual discard. 1627e3e0520bSJosef Bacik */ 1628e3e0520bSJosef Bacik if (trimming) { 1629e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1630e3e0520bSJosef Bacik /* 1631e3e0520bSJosef Bacik * A concurrent scrub might have added us to the list 1632e3e0520bSJosef Bacik * fs_info->unused_bgs, so use a list_move operation 1633e3e0520bSJosef Bacik * to add the block group to the deleted_bgs list. 1634e3e0520bSJosef Bacik */ 1635e3e0520bSJosef Bacik list_move(&block_group->bg_list, 1636e3e0520bSJosef Bacik &trans->transaction->deleted_bgs); 1637e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1638e3e0520bSJosef Bacik btrfs_get_block_group(block_group); 1639e3e0520bSJosef Bacik } 1640e3e0520bSJosef Bacik end_trans: 1641e3e0520bSJosef Bacik btrfs_end_transaction(trans); 1642e3e0520bSJosef Bacik next: 1643e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1644e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1645e3e0520bSJosef Bacik } 1646e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1647f3372065SJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 16486e80d4f8SDennis Zhou return; 16496e80d4f8SDennis Zhou 16506e80d4f8SDennis Zhou flip_async: 16516e80d4f8SDennis Zhou btrfs_end_transaction(trans); 1652f3372065SJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 16536e80d4f8SDennis Zhou btrfs_put_block_group(block_group); 16546e80d4f8SDennis Zhou btrfs_discard_punt_unused_bgs_list(fs_info); 1655e3e0520bSJosef Bacik } 1656e3e0520bSJosef Bacik 165732da5386SDavid Sterba void btrfs_mark_bg_unused(struct btrfs_block_group *bg) 1658e3e0520bSJosef Bacik { 1659e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info = bg->fs_info; 1660e3e0520bSJosef Bacik 1661e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1662e3e0520bSJosef Bacik if (list_empty(&bg->bg_list)) { 1663e3e0520bSJosef Bacik btrfs_get_block_group(bg); 1664e3e0520bSJosef Bacik trace_btrfs_add_unused_block_group(bg); 1665e3e0520bSJosef Bacik list_add_tail(&bg->bg_list, &fs_info->unused_bgs); 1666e3e0520bSJosef Bacik } 1667e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1668e3e0520bSJosef Bacik } 16694358d963SJosef Bacik 16702ca0ec77SJohannes Thumshirn /* 16712ca0ec77SJohannes Thumshirn * We want block groups with a low number of used bytes to be in the beginning 16722ca0ec77SJohannes Thumshirn * of the list, so they will get reclaimed first. 16732ca0ec77SJohannes Thumshirn */ 16742ca0ec77SJohannes Thumshirn static int reclaim_bgs_cmp(void *unused, const struct list_head *a, 16752ca0ec77SJohannes Thumshirn const struct list_head *b) 16762ca0ec77SJohannes Thumshirn { 16772ca0ec77SJohannes Thumshirn const struct btrfs_block_group *bg1, *bg2; 16782ca0ec77SJohannes Thumshirn 16792ca0ec77SJohannes Thumshirn bg1 = list_entry(a, struct btrfs_block_group, bg_list); 16802ca0ec77SJohannes Thumshirn bg2 = list_entry(b, struct btrfs_block_group, bg_list); 16812ca0ec77SJohannes Thumshirn 16822ca0ec77SJohannes Thumshirn return bg1->used > bg2->used; 16832ca0ec77SJohannes Thumshirn } 16842ca0ec77SJohannes Thumshirn 16853687fcb0SJohannes Thumshirn static inline bool btrfs_should_reclaim(struct btrfs_fs_info *fs_info) 16863687fcb0SJohannes Thumshirn { 16873687fcb0SJohannes Thumshirn if (btrfs_is_zoned(fs_info)) 16883687fcb0SJohannes Thumshirn return btrfs_zoned_should_reclaim(fs_info); 16893687fcb0SJohannes Thumshirn return true; 16903687fcb0SJohannes Thumshirn } 16913687fcb0SJohannes Thumshirn 169281531225SBoris Burkov static bool should_reclaim_block_group(struct btrfs_block_group *bg, u64 bytes_freed) 169381531225SBoris Burkov { 169481531225SBoris Burkov const struct btrfs_space_info *space_info = bg->space_info; 169581531225SBoris Burkov const int reclaim_thresh = READ_ONCE(space_info->bg_reclaim_threshold); 169681531225SBoris Burkov const u64 new_val = bg->used; 169781531225SBoris Burkov const u64 old_val = new_val + bytes_freed; 169881531225SBoris Burkov u64 thresh; 169981531225SBoris Burkov 170081531225SBoris Burkov if (reclaim_thresh == 0) 170181531225SBoris Burkov return false; 170281531225SBoris Burkov 1703428c8e03SDavid Sterba thresh = mult_perc(bg->length, reclaim_thresh); 170481531225SBoris Burkov 170581531225SBoris Burkov /* 170681531225SBoris Burkov * If we were below the threshold before don't reclaim, we are likely a 170781531225SBoris Burkov * brand new block group and we don't want to relocate new block groups. 170881531225SBoris Burkov */ 170981531225SBoris Burkov if (old_val < thresh) 171081531225SBoris Burkov return false; 171181531225SBoris Burkov if (new_val >= thresh) 171281531225SBoris Burkov return false; 171381531225SBoris Burkov return true; 171481531225SBoris Burkov } 171581531225SBoris Burkov 171618bb8bbfSJohannes Thumshirn void btrfs_reclaim_bgs_work(struct work_struct *work) 171718bb8bbfSJohannes Thumshirn { 171818bb8bbfSJohannes Thumshirn struct btrfs_fs_info *fs_info = 171918bb8bbfSJohannes Thumshirn container_of(work, struct btrfs_fs_info, reclaim_bgs_work); 172018bb8bbfSJohannes Thumshirn struct btrfs_block_group *bg; 172118bb8bbfSJohannes Thumshirn struct btrfs_space_info *space_info; 172218bb8bbfSJohannes Thumshirn 172318bb8bbfSJohannes Thumshirn if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 172418bb8bbfSJohannes Thumshirn return; 172518bb8bbfSJohannes Thumshirn 17262f12741fSJosef Bacik if (btrfs_fs_closing(fs_info)) 17272f12741fSJosef Bacik return; 17282f12741fSJosef Bacik 17293687fcb0SJohannes Thumshirn if (!btrfs_should_reclaim(fs_info)) 17303687fcb0SJohannes Thumshirn return; 17313687fcb0SJohannes Thumshirn 1732ca5e4ea0SNaohiro Aota sb_start_write(fs_info->sb); 1733ca5e4ea0SNaohiro Aota 1734ca5e4ea0SNaohiro Aota if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 1735ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 173618bb8bbfSJohannes Thumshirn return; 1737ca5e4ea0SNaohiro Aota } 173818bb8bbfSJohannes Thumshirn 17399cc0b837SJohannes Thumshirn /* 17409cc0b837SJohannes Thumshirn * Long running balances can keep us blocked here for eternity, so 17419cc0b837SJohannes Thumshirn * simply skip reclaim if we're unable to get the mutex. 17429cc0b837SJohannes Thumshirn */ 17439cc0b837SJohannes Thumshirn if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) { 17449cc0b837SJohannes Thumshirn btrfs_exclop_finish(fs_info); 1745ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 17469cc0b837SJohannes Thumshirn return; 17479cc0b837SJohannes Thumshirn } 17489cc0b837SJohannes Thumshirn 174918bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 17502ca0ec77SJohannes Thumshirn /* 17512ca0ec77SJohannes Thumshirn * Sort happens under lock because we can't simply splice it and sort. 17522ca0ec77SJohannes Thumshirn * The block groups might still be in use and reachable via bg_list, 17532ca0ec77SJohannes Thumshirn * and their presence in the reclaim_bgs list must be preserved. 17542ca0ec77SJohannes Thumshirn */ 17552ca0ec77SJohannes Thumshirn list_sort(NULL, &fs_info->reclaim_bgs, reclaim_bgs_cmp); 175618bb8bbfSJohannes Thumshirn while (!list_empty(&fs_info->reclaim_bgs)) { 17575f93e776SJohannes Thumshirn u64 zone_unusable; 17581cea5cf0SFilipe Manana int ret = 0; 17591cea5cf0SFilipe Manana 176018bb8bbfSJohannes Thumshirn bg = list_first_entry(&fs_info->reclaim_bgs, 176118bb8bbfSJohannes Thumshirn struct btrfs_block_group, 176218bb8bbfSJohannes Thumshirn bg_list); 176318bb8bbfSJohannes Thumshirn list_del_init(&bg->bg_list); 176418bb8bbfSJohannes Thumshirn 176518bb8bbfSJohannes Thumshirn space_info = bg->space_info; 176618bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 176718bb8bbfSJohannes Thumshirn 176818bb8bbfSJohannes Thumshirn /* Don't race with allocators so take the groups_sem */ 176918bb8bbfSJohannes Thumshirn down_write(&space_info->groups_sem); 177018bb8bbfSJohannes Thumshirn 177118bb8bbfSJohannes Thumshirn spin_lock(&bg->lock); 177218bb8bbfSJohannes Thumshirn if (bg->reserved || bg->pinned || bg->ro) { 177318bb8bbfSJohannes Thumshirn /* 177418bb8bbfSJohannes Thumshirn * We want to bail if we made new allocations or have 177518bb8bbfSJohannes Thumshirn * outstanding allocations in this block group. We do 177618bb8bbfSJohannes Thumshirn * the ro check in case balance is currently acting on 177718bb8bbfSJohannes Thumshirn * this block group. 177818bb8bbfSJohannes Thumshirn */ 177918bb8bbfSJohannes Thumshirn spin_unlock(&bg->lock); 178018bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 178118bb8bbfSJohannes Thumshirn goto next; 178218bb8bbfSJohannes Thumshirn } 1783cc4804bfSBoris Burkov if (bg->used == 0) { 1784cc4804bfSBoris Burkov /* 1785cc4804bfSBoris Burkov * It is possible that we trigger relocation on a block 1786cc4804bfSBoris Burkov * group as its extents are deleted and it first goes 1787cc4804bfSBoris Burkov * below the threshold, then shortly after goes empty. 1788cc4804bfSBoris Burkov * 1789cc4804bfSBoris Burkov * In this case, relocating it does delete it, but has 1790cc4804bfSBoris Burkov * some overhead in relocation specific metadata, looking 1791cc4804bfSBoris Burkov * for the non-existent extents and running some extra 1792cc4804bfSBoris Burkov * transactions, which we can avoid by using one of the 1793cc4804bfSBoris Burkov * other mechanisms for dealing with empty block groups. 1794cc4804bfSBoris Burkov */ 1795cc4804bfSBoris Burkov if (!btrfs_test_opt(fs_info, DISCARD_ASYNC)) 1796cc4804bfSBoris Burkov btrfs_mark_bg_unused(bg); 1797cc4804bfSBoris Burkov spin_unlock(&bg->lock); 1798cc4804bfSBoris Burkov up_write(&space_info->groups_sem); 1799cc4804bfSBoris Burkov goto next; 180081531225SBoris Burkov 180181531225SBoris Burkov } 180281531225SBoris Burkov /* 180381531225SBoris Burkov * The block group might no longer meet the reclaim condition by 180481531225SBoris Burkov * the time we get around to reclaiming it, so to avoid 180581531225SBoris Burkov * reclaiming overly full block_groups, skip reclaiming them. 180681531225SBoris Burkov * 180781531225SBoris Burkov * Since the decision making process also depends on the amount 180881531225SBoris Burkov * being freed, pass in a fake giant value to skip that extra 180981531225SBoris Burkov * check, which is more meaningful when adding to the list in 181081531225SBoris Burkov * the first place. 181181531225SBoris Burkov */ 181281531225SBoris Burkov if (!should_reclaim_block_group(bg, bg->length)) { 181381531225SBoris Burkov spin_unlock(&bg->lock); 181481531225SBoris Burkov up_write(&space_info->groups_sem); 181581531225SBoris Burkov goto next; 1816cc4804bfSBoris Burkov } 181718bb8bbfSJohannes Thumshirn spin_unlock(&bg->lock); 181818bb8bbfSJohannes Thumshirn 181918bb8bbfSJohannes Thumshirn /* Get out fast, in case we're unmounting the filesystem */ 182018bb8bbfSJohannes Thumshirn if (btrfs_fs_closing(fs_info)) { 182118bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 182218bb8bbfSJohannes Thumshirn goto next; 182318bb8bbfSJohannes Thumshirn } 182418bb8bbfSJohannes Thumshirn 18255f93e776SJohannes Thumshirn /* 18265f93e776SJohannes Thumshirn * Cache the zone_unusable value before turning the block group 18275f93e776SJohannes Thumshirn * to read only. As soon as the blog group is read only it's 18285f93e776SJohannes Thumshirn * zone_unusable value gets moved to the block group's read-only 18295f93e776SJohannes Thumshirn * bytes and isn't available for calculations anymore. 18305f93e776SJohannes Thumshirn */ 18315f93e776SJohannes Thumshirn zone_unusable = bg->zone_unusable; 183218bb8bbfSJohannes Thumshirn ret = inc_block_group_ro(bg, 0); 183318bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 183418bb8bbfSJohannes Thumshirn if (ret < 0) 183518bb8bbfSJohannes Thumshirn goto next; 183618bb8bbfSJohannes Thumshirn 18375f93e776SJohannes Thumshirn btrfs_info(fs_info, 18385f93e776SJohannes Thumshirn "reclaiming chunk %llu with %llu%% used %llu%% unusable", 18395f93e776SJohannes Thumshirn bg->start, div_u64(bg->used * 100, bg->length), 18405f93e776SJohannes Thumshirn div64_u64(zone_unusable * 100, bg->length)); 184118bb8bbfSJohannes Thumshirn trace_btrfs_reclaim_block_group(bg); 184218bb8bbfSJohannes Thumshirn ret = btrfs_relocate_chunk(fs_info, bg->start); 184374944c87SJosef Bacik if (ret) { 184474944c87SJosef Bacik btrfs_dec_block_group_ro(bg); 184518bb8bbfSJohannes Thumshirn btrfs_err(fs_info, "error relocating chunk %llu", 184618bb8bbfSJohannes Thumshirn bg->start); 184774944c87SJosef Bacik } 184818bb8bbfSJohannes Thumshirn 184918bb8bbfSJohannes Thumshirn next: 18501cea5cf0SFilipe Manana btrfs_put_block_group(bg); 1851d96b3424SFilipe Manana spin_lock(&fs_info->unused_bgs_lock); 185218bb8bbfSJohannes Thumshirn } 185318bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 185418bb8bbfSJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 185518bb8bbfSJohannes Thumshirn btrfs_exclop_finish(fs_info); 1856ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 185718bb8bbfSJohannes Thumshirn } 185818bb8bbfSJohannes Thumshirn 185918bb8bbfSJohannes Thumshirn void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info) 186018bb8bbfSJohannes Thumshirn { 186118bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 186218bb8bbfSJohannes Thumshirn if (!list_empty(&fs_info->reclaim_bgs)) 186318bb8bbfSJohannes Thumshirn queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work); 186418bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 186518bb8bbfSJohannes Thumshirn } 186618bb8bbfSJohannes Thumshirn 186718bb8bbfSJohannes Thumshirn void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg) 186818bb8bbfSJohannes Thumshirn { 186918bb8bbfSJohannes Thumshirn struct btrfs_fs_info *fs_info = bg->fs_info; 187018bb8bbfSJohannes Thumshirn 187118bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 187218bb8bbfSJohannes Thumshirn if (list_empty(&bg->bg_list)) { 187318bb8bbfSJohannes Thumshirn btrfs_get_block_group(bg); 187418bb8bbfSJohannes Thumshirn trace_btrfs_add_reclaim_block_group(bg); 187518bb8bbfSJohannes Thumshirn list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs); 187618bb8bbfSJohannes Thumshirn } 187718bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 187818bb8bbfSJohannes Thumshirn } 187918bb8bbfSJohannes Thumshirn 1880e3ba67a1SJohannes Thumshirn static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key, 1881e3ba67a1SJohannes Thumshirn struct btrfs_path *path) 1882e3ba67a1SJohannes Thumshirn { 1883e3ba67a1SJohannes Thumshirn struct extent_map_tree *em_tree; 1884e3ba67a1SJohannes Thumshirn struct extent_map *em; 1885e3ba67a1SJohannes Thumshirn struct btrfs_block_group_item bg; 1886e3ba67a1SJohannes Thumshirn struct extent_buffer *leaf; 1887e3ba67a1SJohannes Thumshirn int slot; 1888e3ba67a1SJohannes Thumshirn u64 flags; 1889e3ba67a1SJohannes Thumshirn int ret = 0; 1890e3ba67a1SJohannes Thumshirn 1891e3ba67a1SJohannes Thumshirn slot = path->slots[0]; 1892e3ba67a1SJohannes Thumshirn leaf = path->nodes[0]; 1893e3ba67a1SJohannes Thumshirn 1894e3ba67a1SJohannes Thumshirn em_tree = &fs_info->mapping_tree; 1895e3ba67a1SJohannes Thumshirn read_lock(&em_tree->lock); 1896e3ba67a1SJohannes Thumshirn em = lookup_extent_mapping(em_tree, key->objectid, key->offset); 1897e3ba67a1SJohannes Thumshirn read_unlock(&em_tree->lock); 1898e3ba67a1SJohannes Thumshirn if (!em) { 1899e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1900e3ba67a1SJohannes Thumshirn "logical %llu len %llu found bg but no related chunk", 1901e3ba67a1SJohannes Thumshirn key->objectid, key->offset); 1902e3ba67a1SJohannes Thumshirn return -ENOENT; 1903e3ba67a1SJohannes Thumshirn } 1904e3ba67a1SJohannes Thumshirn 1905e3ba67a1SJohannes Thumshirn if (em->start != key->objectid || em->len != key->offset) { 1906e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1907e3ba67a1SJohannes Thumshirn "block group %llu len %llu mismatch with chunk %llu len %llu", 1908e3ba67a1SJohannes Thumshirn key->objectid, key->offset, em->start, em->len); 1909e3ba67a1SJohannes Thumshirn ret = -EUCLEAN; 1910e3ba67a1SJohannes Thumshirn goto out_free_em; 1911e3ba67a1SJohannes Thumshirn } 1912e3ba67a1SJohannes Thumshirn 1913e3ba67a1SJohannes Thumshirn read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot), 1914e3ba67a1SJohannes Thumshirn sizeof(bg)); 1915e3ba67a1SJohannes Thumshirn flags = btrfs_stack_block_group_flags(&bg) & 1916e3ba67a1SJohannes Thumshirn BTRFS_BLOCK_GROUP_TYPE_MASK; 1917e3ba67a1SJohannes Thumshirn 1918e3ba67a1SJohannes Thumshirn if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 1919e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1920e3ba67a1SJohannes Thumshirn "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", 1921e3ba67a1SJohannes Thumshirn key->objectid, key->offset, flags, 1922e3ba67a1SJohannes Thumshirn (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type)); 1923e3ba67a1SJohannes Thumshirn ret = -EUCLEAN; 1924e3ba67a1SJohannes Thumshirn } 1925e3ba67a1SJohannes Thumshirn 1926e3ba67a1SJohannes Thumshirn out_free_em: 1927e3ba67a1SJohannes Thumshirn free_extent_map(em); 1928e3ba67a1SJohannes Thumshirn return ret; 1929e3ba67a1SJohannes Thumshirn } 1930e3ba67a1SJohannes Thumshirn 19314358d963SJosef Bacik static int find_first_block_group(struct btrfs_fs_info *fs_info, 19324358d963SJosef Bacik struct btrfs_path *path, 19334358d963SJosef Bacik struct btrfs_key *key) 19344358d963SJosef Bacik { 1935dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 1936e3ba67a1SJohannes Thumshirn int ret; 19374358d963SJosef Bacik struct btrfs_key found_key; 19384358d963SJosef Bacik 193936dfbbe2SGabriel Niebler btrfs_for_each_slot(root, key, &found_key, path, ret) { 19404358d963SJosef Bacik if (found_key.objectid >= key->objectid && 19414358d963SJosef Bacik found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { 194236dfbbe2SGabriel Niebler return read_bg_from_eb(fs_info, &found_key, path); 1943e3ba67a1SJohannes Thumshirn } 19444358d963SJosef Bacik } 19454358d963SJosef Bacik return ret; 19464358d963SJosef Bacik } 19474358d963SJosef Bacik 19484358d963SJosef Bacik static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) 19494358d963SJosef Bacik { 19504358d963SJosef Bacik u64 extra_flags = chunk_to_extended(flags) & 19514358d963SJosef Bacik BTRFS_EXTENDED_PROFILE_MASK; 19524358d963SJosef Bacik 19534358d963SJosef Bacik write_seqlock(&fs_info->profiles_lock); 19544358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 19554358d963SJosef Bacik fs_info->avail_data_alloc_bits |= extra_flags; 19564358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_METADATA) 19574358d963SJosef Bacik fs_info->avail_metadata_alloc_bits |= extra_flags; 19584358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 19594358d963SJosef Bacik fs_info->avail_system_alloc_bits |= extra_flags; 19604358d963SJosef Bacik write_sequnlock(&fs_info->profiles_lock); 19614358d963SJosef Bacik } 19624358d963SJosef Bacik 196343dd529aSDavid Sterba /* 196443dd529aSDavid Sterba * Map a physical disk address to a list of logical addresses. 19659ee9b979SNikolay Borisov * 19669ee9b979SNikolay Borisov * @fs_info: the filesystem 196796a14336SNikolay Borisov * @chunk_start: logical address of block group 1968138082f3SNaohiro Aota * @bdev: physical device to resolve, can be NULL to indicate any device 196996a14336SNikolay Borisov * @physical: physical address to map to logical addresses 197096a14336SNikolay Borisov * @logical: return array of logical addresses which map to @physical 197196a14336SNikolay Borisov * @naddrs: length of @logical 197296a14336SNikolay Borisov * @stripe_len: size of IO stripe for the given block group 197396a14336SNikolay Borisov * 197496a14336SNikolay Borisov * Maps a particular @physical disk address to a list of @logical addresses. 197596a14336SNikolay Borisov * Used primarily to exclude those portions of a block group that contain super 197696a14336SNikolay Borisov * block copies. 197796a14336SNikolay Borisov */ 197896a14336SNikolay Borisov int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, 1979138082f3SNaohiro Aota struct block_device *bdev, u64 physical, u64 **logical, 1980138082f3SNaohiro Aota int *naddrs, int *stripe_len) 198196a14336SNikolay Borisov { 198296a14336SNikolay Borisov struct extent_map *em; 198396a14336SNikolay Borisov struct map_lookup *map; 198496a14336SNikolay Borisov u64 *buf; 198596a14336SNikolay Borisov u64 bytenr; 19861776ad17SNikolay Borisov u64 data_stripe_length; 19871776ad17SNikolay Borisov u64 io_stripe_size; 19881776ad17SNikolay Borisov int i, nr = 0; 19891776ad17SNikolay Borisov int ret = 0; 199096a14336SNikolay Borisov 199196a14336SNikolay Borisov em = btrfs_get_chunk_map(fs_info, chunk_start, 1); 199296a14336SNikolay Borisov if (IS_ERR(em)) 199396a14336SNikolay Borisov return -EIO; 199496a14336SNikolay Borisov 199596a14336SNikolay Borisov map = em->map_lookup; 19969e22b925SNikolay Borisov data_stripe_length = em->orig_block_len; 19971776ad17SNikolay Borisov io_stripe_size = map->stripe_len; 1998138082f3SNaohiro Aota chunk_start = em->start; 199996a14336SNikolay Borisov 20009e22b925SNikolay Borisov /* For RAID5/6 adjust to a full IO stripe length */ 20019e22b925SNikolay Borisov if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 20021776ad17SNikolay Borisov io_stripe_size = map->stripe_len * nr_data_stripes(map); 200396a14336SNikolay Borisov 200496a14336SNikolay Borisov buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); 20051776ad17SNikolay Borisov if (!buf) { 20061776ad17SNikolay Borisov ret = -ENOMEM; 20071776ad17SNikolay Borisov goto out; 20081776ad17SNikolay Borisov } 200996a14336SNikolay Borisov 201096a14336SNikolay Borisov for (i = 0; i < map->num_stripes; i++) { 20111776ad17SNikolay Borisov bool already_inserted = false; 20121776ad17SNikolay Borisov u64 stripe_nr; 2013138082f3SNaohiro Aota u64 offset; 20141776ad17SNikolay Borisov int j; 20151776ad17SNikolay Borisov 20161776ad17SNikolay Borisov if (!in_range(physical, map->stripes[i].physical, 20171776ad17SNikolay Borisov data_stripe_length)) 201896a14336SNikolay Borisov continue; 201996a14336SNikolay Borisov 2020138082f3SNaohiro Aota if (bdev && map->stripes[i].dev->bdev != bdev) 2021138082f3SNaohiro Aota continue; 2022138082f3SNaohiro Aota 202396a14336SNikolay Borisov stripe_nr = physical - map->stripes[i].physical; 2024138082f3SNaohiro Aota stripe_nr = div64_u64_rem(stripe_nr, map->stripe_len, &offset); 202596a14336SNikolay Borisov 2026ac067734SDavid Sterba if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 2027ac067734SDavid Sterba BTRFS_BLOCK_GROUP_RAID10)) { 202896a14336SNikolay Borisov stripe_nr = stripe_nr * map->num_stripes + i; 202996a14336SNikolay Borisov stripe_nr = div_u64(stripe_nr, map->sub_stripes); 203096a14336SNikolay Borisov } 203196a14336SNikolay Borisov /* 203296a14336SNikolay Borisov * The remaining case would be for RAID56, multiply by 203396a14336SNikolay Borisov * nr_data_stripes(). Alternatively, just use rmap_len below 203496a14336SNikolay Borisov * instead of map->stripe_len 203596a14336SNikolay Borisov */ 203696a14336SNikolay Borisov 2037138082f3SNaohiro Aota bytenr = chunk_start + stripe_nr * io_stripe_size + offset; 20381776ad17SNikolay Borisov 20391776ad17SNikolay Borisov /* Ensure we don't add duplicate addresses */ 204096a14336SNikolay Borisov for (j = 0; j < nr; j++) { 20411776ad17SNikolay Borisov if (buf[j] == bytenr) { 20421776ad17SNikolay Borisov already_inserted = true; 204396a14336SNikolay Borisov break; 204496a14336SNikolay Borisov } 204596a14336SNikolay Borisov } 20461776ad17SNikolay Borisov 20471776ad17SNikolay Borisov if (!already_inserted) 20481776ad17SNikolay Borisov buf[nr++] = bytenr; 204996a14336SNikolay Borisov } 205096a14336SNikolay Borisov 205196a14336SNikolay Borisov *logical = buf; 205296a14336SNikolay Borisov *naddrs = nr; 20531776ad17SNikolay Borisov *stripe_len = io_stripe_size; 20541776ad17SNikolay Borisov out: 205596a14336SNikolay Borisov free_extent_map(em); 20561776ad17SNikolay Borisov return ret; 205796a14336SNikolay Borisov } 205896a14336SNikolay Borisov 205932da5386SDavid Sterba static int exclude_super_stripes(struct btrfs_block_group *cache) 20604358d963SJosef Bacik { 20614358d963SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 206212659251SNaohiro Aota const bool zoned = btrfs_is_zoned(fs_info); 20634358d963SJosef Bacik u64 bytenr; 20644358d963SJosef Bacik u64 *logical; 20654358d963SJosef Bacik int stripe_len; 20664358d963SJosef Bacik int i, nr, ret; 20674358d963SJosef Bacik 2068b3470b5dSDavid Sterba if (cache->start < BTRFS_SUPER_INFO_OFFSET) { 2069b3470b5dSDavid Sterba stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start; 20704358d963SJosef Bacik cache->bytes_super += stripe_len; 2071b3470b5dSDavid Sterba ret = btrfs_add_excluded_extent(fs_info, cache->start, 20724358d963SJosef Bacik stripe_len); 20734358d963SJosef Bacik if (ret) 20744358d963SJosef Bacik return ret; 20754358d963SJosef Bacik } 20764358d963SJosef Bacik 20774358d963SJosef Bacik for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 20784358d963SJosef Bacik bytenr = btrfs_sb_offset(i); 2079138082f3SNaohiro Aota ret = btrfs_rmap_block(fs_info, cache->start, NULL, 20804358d963SJosef Bacik bytenr, &logical, &nr, &stripe_len); 20814358d963SJosef Bacik if (ret) 20824358d963SJosef Bacik return ret; 20834358d963SJosef Bacik 208412659251SNaohiro Aota /* Shouldn't have super stripes in sequential zones */ 208512659251SNaohiro Aota if (zoned && nr) { 208612659251SNaohiro Aota btrfs_err(fs_info, 208712659251SNaohiro Aota "zoned: block group %llu must not contain super block", 208812659251SNaohiro Aota cache->start); 208912659251SNaohiro Aota return -EUCLEAN; 209012659251SNaohiro Aota } 209112659251SNaohiro Aota 20924358d963SJosef Bacik while (nr--) { 209396f9b0f2SNikolay Borisov u64 len = min_t(u64, stripe_len, 209496f9b0f2SNikolay Borisov cache->start + cache->length - logical[nr]); 20954358d963SJosef Bacik 20964358d963SJosef Bacik cache->bytes_super += len; 209796f9b0f2SNikolay Borisov ret = btrfs_add_excluded_extent(fs_info, logical[nr], 209896f9b0f2SNikolay Borisov len); 20994358d963SJosef Bacik if (ret) { 21004358d963SJosef Bacik kfree(logical); 21014358d963SJosef Bacik return ret; 21024358d963SJosef Bacik } 21034358d963SJosef Bacik } 21044358d963SJosef Bacik 21054358d963SJosef Bacik kfree(logical); 21064358d963SJosef Bacik } 21074358d963SJosef Bacik return 0; 21084358d963SJosef Bacik } 21094358d963SJosef Bacik 211032da5386SDavid Sterba static struct btrfs_block_group *btrfs_create_block_group_cache( 21119afc6649SQu Wenruo struct btrfs_fs_info *fs_info, u64 start) 21124358d963SJosef Bacik { 211332da5386SDavid Sterba struct btrfs_block_group *cache; 21144358d963SJosef Bacik 21154358d963SJosef Bacik cache = kzalloc(sizeof(*cache), GFP_NOFS); 21164358d963SJosef Bacik if (!cache) 21174358d963SJosef Bacik return NULL; 21184358d963SJosef Bacik 21194358d963SJosef Bacik cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), 21204358d963SJosef Bacik GFP_NOFS); 21214358d963SJosef Bacik if (!cache->free_space_ctl) { 21224358d963SJosef Bacik kfree(cache); 21234358d963SJosef Bacik return NULL; 21244358d963SJosef Bacik } 21254358d963SJosef Bacik 2126b3470b5dSDavid Sterba cache->start = start; 21274358d963SJosef Bacik 21284358d963SJosef Bacik cache->fs_info = fs_info; 21294358d963SJosef Bacik cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); 21304358d963SJosef Bacik 21316e80d4f8SDennis Zhou cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED; 21326e80d4f8SDennis Zhou 213348aaeebeSJosef Bacik refcount_set(&cache->refs, 1); 21344358d963SJosef Bacik spin_lock_init(&cache->lock); 21354358d963SJosef Bacik init_rwsem(&cache->data_rwsem); 21364358d963SJosef Bacik INIT_LIST_HEAD(&cache->list); 21374358d963SJosef Bacik INIT_LIST_HEAD(&cache->cluster_list); 21384358d963SJosef Bacik INIT_LIST_HEAD(&cache->bg_list); 21394358d963SJosef Bacik INIT_LIST_HEAD(&cache->ro_list); 2140b0643e59SDennis Zhou INIT_LIST_HEAD(&cache->discard_list); 21414358d963SJosef Bacik INIT_LIST_HEAD(&cache->dirty_list); 21424358d963SJosef Bacik INIT_LIST_HEAD(&cache->io_list); 2143afba2bc0SNaohiro Aota INIT_LIST_HEAD(&cache->active_bg_list); 2144cd79909bSJosef Bacik btrfs_init_free_space_ctl(cache, cache->free_space_ctl); 21456b7304afSFilipe Manana atomic_set(&cache->frozen, 0); 21464358d963SJosef Bacik mutex_init(&cache->free_space_lock); 2147c29abab4SJosef Bacik cache->full_stripe_locks_root.root = RB_ROOT; 2148c29abab4SJosef Bacik mutex_init(&cache->full_stripe_locks_root.lock); 21494358d963SJosef Bacik 21504358d963SJosef Bacik return cache; 21514358d963SJosef Bacik } 21524358d963SJosef Bacik 21534358d963SJosef Bacik /* 21544358d963SJosef Bacik * Iterate all chunks and verify that each of them has the corresponding block 21554358d963SJosef Bacik * group 21564358d963SJosef Bacik */ 21574358d963SJosef Bacik static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) 21584358d963SJosef Bacik { 21594358d963SJosef Bacik struct extent_map_tree *map_tree = &fs_info->mapping_tree; 21604358d963SJosef Bacik struct extent_map *em; 216132da5386SDavid Sterba struct btrfs_block_group *bg; 21624358d963SJosef Bacik u64 start = 0; 21634358d963SJosef Bacik int ret = 0; 21644358d963SJosef Bacik 21654358d963SJosef Bacik while (1) { 21664358d963SJosef Bacik read_lock(&map_tree->lock); 21674358d963SJosef Bacik /* 21684358d963SJosef Bacik * lookup_extent_mapping will return the first extent map 21694358d963SJosef Bacik * intersecting the range, so setting @len to 1 is enough to 21704358d963SJosef Bacik * get the first chunk. 21714358d963SJosef Bacik */ 21724358d963SJosef Bacik em = lookup_extent_mapping(map_tree, start, 1); 21734358d963SJosef Bacik read_unlock(&map_tree->lock); 21744358d963SJosef Bacik if (!em) 21754358d963SJosef Bacik break; 21764358d963SJosef Bacik 21774358d963SJosef Bacik bg = btrfs_lookup_block_group(fs_info, em->start); 21784358d963SJosef Bacik if (!bg) { 21794358d963SJosef Bacik btrfs_err(fs_info, 21804358d963SJosef Bacik "chunk start=%llu len=%llu doesn't have corresponding block group", 21814358d963SJosef Bacik em->start, em->len); 21824358d963SJosef Bacik ret = -EUCLEAN; 21834358d963SJosef Bacik free_extent_map(em); 21844358d963SJosef Bacik break; 21854358d963SJosef Bacik } 2186b3470b5dSDavid Sterba if (bg->start != em->start || bg->length != em->len || 21874358d963SJosef Bacik (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != 21884358d963SJosef Bacik (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 21894358d963SJosef Bacik btrfs_err(fs_info, 21904358d963SJosef Bacik "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", 21914358d963SJosef Bacik em->start, em->len, 21924358d963SJosef Bacik em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, 2193b3470b5dSDavid Sterba bg->start, bg->length, 21944358d963SJosef Bacik bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); 21954358d963SJosef Bacik ret = -EUCLEAN; 21964358d963SJosef Bacik free_extent_map(em); 21974358d963SJosef Bacik btrfs_put_block_group(bg); 21984358d963SJosef Bacik break; 21994358d963SJosef Bacik } 22004358d963SJosef Bacik start = em->start + em->len; 22014358d963SJosef Bacik free_extent_map(em); 22024358d963SJosef Bacik btrfs_put_block_group(bg); 22034358d963SJosef Bacik } 22044358d963SJosef Bacik return ret; 22054358d963SJosef Bacik } 22064358d963SJosef Bacik 2207ffb9e0f0SQu Wenruo static int read_one_block_group(struct btrfs_fs_info *info, 22084afd2fe8SJohannes Thumshirn struct btrfs_block_group_item *bgi, 2209d49a2ddbSQu Wenruo const struct btrfs_key *key, 2210ffb9e0f0SQu Wenruo int need_clear) 2211ffb9e0f0SQu Wenruo { 221232da5386SDavid Sterba struct btrfs_block_group *cache; 2213ffb9e0f0SQu Wenruo const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS); 2214ffb9e0f0SQu Wenruo int ret; 2215ffb9e0f0SQu Wenruo 2216d49a2ddbSQu Wenruo ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY); 2217ffb9e0f0SQu Wenruo 22189afc6649SQu Wenruo cache = btrfs_create_block_group_cache(info, key->objectid); 2219ffb9e0f0SQu Wenruo if (!cache) 2220ffb9e0f0SQu Wenruo return -ENOMEM; 2221ffb9e0f0SQu Wenruo 22224afd2fe8SJohannes Thumshirn cache->length = key->offset; 22234afd2fe8SJohannes Thumshirn cache->used = btrfs_stack_block_group_used(bgi); 22247248e0ceSQu Wenruo cache->commit_used = cache->used; 22254afd2fe8SJohannes Thumshirn cache->flags = btrfs_stack_block_group_flags(bgi); 2226f7238e50SJosef Bacik cache->global_root_id = btrfs_stack_block_group_chunk_objectid(bgi); 22279afc6649SQu Wenruo 2228e3e39c72SMarcos Paulo de Souza set_free_space_tree_thresholds(cache); 2229e3e39c72SMarcos Paulo de Souza 2230ffb9e0f0SQu Wenruo if (need_clear) { 2231ffb9e0f0SQu Wenruo /* 2232ffb9e0f0SQu Wenruo * When we mount with old space cache, we need to 2233ffb9e0f0SQu Wenruo * set BTRFS_DC_CLEAR and set dirty flag. 2234ffb9e0f0SQu Wenruo * 2235ffb9e0f0SQu Wenruo * a) Setting 'BTRFS_DC_CLEAR' makes sure that we 2236ffb9e0f0SQu Wenruo * truncate the old free space cache inode and 2237ffb9e0f0SQu Wenruo * setup a new one. 2238ffb9e0f0SQu Wenruo * b) Setting 'dirty flag' makes sure that we flush 2239ffb9e0f0SQu Wenruo * the new space cache info onto disk. 2240ffb9e0f0SQu Wenruo */ 2241ffb9e0f0SQu Wenruo if (btrfs_test_opt(info, SPACE_CACHE)) 2242ffb9e0f0SQu Wenruo cache->disk_cache_state = BTRFS_DC_CLEAR; 2243ffb9e0f0SQu Wenruo } 2244ffb9e0f0SQu Wenruo if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && 2245ffb9e0f0SQu Wenruo (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { 2246ffb9e0f0SQu Wenruo btrfs_err(info, 2247ffb9e0f0SQu Wenruo "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", 2248ffb9e0f0SQu Wenruo cache->start); 2249ffb9e0f0SQu Wenruo ret = -EINVAL; 2250ffb9e0f0SQu Wenruo goto error; 2251ffb9e0f0SQu Wenruo } 2252ffb9e0f0SQu Wenruo 2253a94794d5SNaohiro Aota ret = btrfs_load_block_group_zone_info(cache, false); 225408e11a3dSNaohiro Aota if (ret) { 225508e11a3dSNaohiro Aota btrfs_err(info, "zoned: failed to load zone info of bg %llu", 225608e11a3dSNaohiro Aota cache->start); 225708e11a3dSNaohiro Aota goto error; 225808e11a3dSNaohiro Aota } 225908e11a3dSNaohiro Aota 2260ffb9e0f0SQu Wenruo /* 2261ffb9e0f0SQu Wenruo * We need to exclude the super stripes now so that the space info has 2262ffb9e0f0SQu Wenruo * super bytes accounted for, otherwise we'll think we have more space 2263ffb9e0f0SQu Wenruo * than we actually do. 2264ffb9e0f0SQu Wenruo */ 2265ffb9e0f0SQu Wenruo ret = exclude_super_stripes(cache); 2266ffb9e0f0SQu Wenruo if (ret) { 2267ffb9e0f0SQu Wenruo /* We may have excluded something, so call this just in case. */ 2268ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2269ffb9e0f0SQu Wenruo goto error; 2270ffb9e0f0SQu Wenruo } 2271ffb9e0f0SQu Wenruo 2272ffb9e0f0SQu Wenruo /* 2273169e0da9SNaohiro Aota * For zoned filesystem, space after the allocation offset is the only 2274169e0da9SNaohiro Aota * free space for a block group. So, we don't need any caching work. 2275169e0da9SNaohiro Aota * btrfs_calc_zone_unusable() will set the amount of free space and 2276169e0da9SNaohiro Aota * zone_unusable space. 2277169e0da9SNaohiro Aota * 2278169e0da9SNaohiro Aota * For regular filesystem, check for two cases, either we are full, and 2279169e0da9SNaohiro Aota * therefore don't need to bother with the caching work since we won't 2280169e0da9SNaohiro Aota * find any space, or we are empty, and we can just add all the space 2281169e0da9SNaohiro Aota * in and be done with it. This saves us _a_lot_ of time, particularly 2282169e0da9SNaohiro Aota * in the full case. 2283ffb9e0f0SQu Wenruo */ 2284169e0da9SNaohiro Aota if (btrfs_is_zoned(info)) { 2285169e0da9SNaohiro Aota btrfs_calc_zone_unusable(cache); 2286c46c4247SNaohiro Aota /* Should not have any excluded extents. Just in case, though. */ 2287c46c4247SNaohiro Aota btrfs_free_excluded_extents(cache); 2288169e0da9SNaohiro Aota } else if (cache->length == cache->used) { 2289ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 2290ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2291ffb9e0f0SQu Wenruo } else if (cache->used == 0) { 2292ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 22939afc6649SQu Wenruo add_new_free_space(cache, cache->start, 22949afc6649SQu Wenruo cache->start + cache->length); 2295ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2296ffb9e0f0SQu Wenruo } 2297ffb9e0f0SQu Wenruo 2298ffb9e0f0SQu Wenruo ret = btrfs_add_block_group_cache(info, cache); 2299ffb9e0f0SQu Wenruo if (ret) { 2300ffb9e0f0SQu Wenruo btrfs_remove_free_space_cache(cache); 2301ffb9e0f0SQu Wenruo goto error; 2302ffb9e0f0SQu Wenruo } 2303ffb9e0f0SQu Wenruo trace_btrfs_add_block_group(info, cache, 0); 2304723de71dSJosef Bacik btrfs_add_bg_to_space_info(info, cache); 2305ffb9e0f0SQu Wenruo 2306ffb9e0f0SQu Wenruo set_avail_alloc_bits(info, cache->flags); 2307a09f23c3SAnand Jain if (btrfs_chunk_writeable(info, cache->start)) { 2308a09f23c3SAnand Jain if (cache->used == 0) { 2309ffb9e0f0SQu Wenruo ASSERT(list_empty(&cache->bg_list)); 23106e80d4f8SDennis Zhou if (btrfs_test_opt(info, DISCARD_ASYNC)) 23116e80d4f8SDennis Zhou btrfs_discard_queue_work(&info->discard_ctl, cache); 23126e80d4f8SDennis Zhou else 2313ffb9e0f0SQu Wenruo btrfs_mark_bg_unused(cache); 2314ffb9e0f0SQu Wenruo } 2315a09f23c3SAnand Jain } else { 2316a09f23c3SAnand Jain inc_block_group_ro(cache, 1); 2317a09f23c3SAnand Jain } 2318a09f23c3SAnand Jain 2319ffb9e0f0SQu Wenruo return 0; 2320ffb9e0f0SQu Wenruo error: 2321ffb9e0f0SQu Wenruo btrfs_put_block_group(cache); 2322ffb9e0f0SQu Wenruo return ret; 2323ffb9e0f0SQu Wenruo } 2324ffb9e0f0SQu Wenruo 232542437a63SJosef Bacik static int fill_dummy_bgs(struct btrfs_fs_info *fs_info) 232642437a63SJosef Bacik { 232742437a63SJosef Bacik struct extent_map_tree *em_tree = &fs_info->mapping_tree; 232842437a63SJosef Bacik struct rb_node *node; 232942437a63SJosef Bacik int ret = 0; 233042437a63SJosef Bacik 233142437a63SJosef Bacik for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 233242437a63SJosef Bacik struct extent_map *em; 233342437a63SJosef Bacik struct map_lookup *map; 233442437a63SJosef Bacik struct btrfs_block_group *bg; 233542437a63SJosef Bacik 233642437a63SJosef Bacik em = rb_entry(node, struct extent_map, rb_node); 233742437a63SJosef Bacik map = em->map_lookup; 233842437a63SJosef Bacik bg = btrfs_create_block_group_cache(fs_info, em->start); 233942437a63SJosef Bacik if (!bg) { 234042437a63SJosef Bacik ret = -ENOMEM; 234142437a63SJosef Bacik break; 234242437a63SJosef Bacik } 234342437a63SJosef Bacik 234442437a63SJosef Bacik /* Fill dummy cache as FULL */ 234542437a63SJosef Bacik bg->length = em->len; 234642437a63SJosef Bacik bg->flags = map->type; 234742437a63SJosef Bacik bg->cached = BTRFS_CACHE_FINISHED; 234842437a63SJosef Bacik bg->used = em->len; 234942437a63SJosef Bacik bg->flags = map->type; 235042437a63SJosef Bacik ret = btrfs_add_block_group_cache(fs_info, bg); 23512b29726cSQu Wenruo /* 23522b29726cSQu Wenruo * We may have some valid block group cache added already, in 23532b29726cSQu Wenruo * that case we skip to the next one. 23542b29726cSQu Wenruo */ 23552b29726cSQu Wenruo if (ret == -EEXIST) { 23562b29726cSQu Wenruo ret = 0; 23572b29726cSQu Wenruo btrfs_put_block_group(bg); 23582b29726cSQu Wenruo continue; 23592b29726cSQu Wenruo } 23602b29726cSQu Wenruo 236142437a63SJosef Bacik if (ret) { 236242437a63SJosef Bacik btrfs_remove_free_space_cache(bg); 236342437a63SJosef Bacik btrfs_put_block_group(bg); 236442437a63SJosef Bacik break; 236542437a63SJosef Bacik } 23662b29726cSQu Wenruo 2367723de71dSJosef Bacik btrfs_add_bg_to_space_info(fs_info, bg); 236842437a63SJosef Bacik 236942437a63SJosef Bacik set_avail_alloc_bits(fs_info, bg->flags); 237042437a63SJosef Bacik } 237142437a63SJosef Bacik if (!ret) 237242437a63SJosef Bacik btrfs_init_global_block_rsv(fs_info); 237342437a63SJosef Bacik return ret; 237442437a63SJosef Bacik } 237542437a63SJosef Bacik 23764358d963SJosef Bacik int btrfs_read_block_groups(struct btrfs_fs_info *info) 23774358d963SJosef Bacik { 2378dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(info); 23794358d963SJosef Bacik struct btrfs_path *path; 23804358d963SJosef Bacik int ret; 238132da5386SDavid Sterba struct btrfs_block_group *cache; 23824358d963SJosef Bacik struct btrfs_space_info *space_info; 23834358d963SJosef Bacik struct btrfs_key key; 23844358d963SJosef Bacik int need_clear = 0; 23854358d963SJosef Bacik u64 cache_gen; 23864358d963SJosef Bacik 238781d5d614SQu Wenruo /* 238881d5d614SQu Wenruo * Either no extent root (with ibadroots rescue option) or we have 238981d5d614SQu Wenruo * unsupported RO options. The fs can never be mounted read-write, so no 239081d5d614SQu Wenruo * need to waste time searching block group items. 239181d5d614SQu Wenruo * 239281d5d614SQu Wenruo * This also allows new extent tree related changes to be RO compat, 239381d5d614SQu Wenruo * no need for a full incompat flag. 239481d5d614SQu Wenruo */ 239581d5d614SQu Wenruo if (!root || (btrfs_super_compat_ro_flags(info->super_copy) & 239681d5d614SQu Wenruo ~BTRFS_FEATURE_COMPAT_RO_SUPP)) 239742437a63SJosef Bacik return fill_dummy_bgs(info); 239842437a63SJosef Bacik 23994358d963SJosef Bacik key.objectid = 0; 24004358d963SJosef Bacik key.offset = 0; 24014358d963SJosef Bacik key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 24024358d963SJosef Bacik path = btrfs_alloc_path(); 24034358d963SJosef Bacik if (!path) 24044358d963SJosef Bacik return -ENOMEM; 24054358d963SJosef Bacik 24064358d963SJosef Bacik cache_gen = btrfs_super_cache_generation(info->super_copy); 24074358d963SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 24084358d963SJosef Bacik btrfs_super_generation(info->super_copy) != cache_gen) 24094358d963SJosef Bacik need_clear = 1; 24104358d963SJosef Bacik if (btrfs_test_opt(info, CLEAR_CACHE)) 24114358d963SJosef Bacik need_clear = 1; 24124358d963SJosef Bacik 24134358d963SJosef Bacik while (1) { 24144afd2fe8SJohannes Thumshirn struct btrfs_block_group_item bgi; 24154afd2fe8SJohannes Thumshirn struct extent_buffer *leaf; 24164afd2fe8SJohannes Thumshirn int slot; 24174afd2fe8SJohannes Thumshirn 24184358d963SJosef Bacik ret = find_first_block_group(info, path, &key); 24194358d963SJosef Bacik if (ret > 0) 24204358d963SJosef Bacik break; 24214358d963SJosef Bacik if (ret != 0) 24224358d963SJosef Bacik goto error; 24234358d963SJosef Bacik 24244afd2fe8SJohannes Thumshirn leaf = path->nodes[0]; 24254afd2fe8SJohannes Thumshirn slot = path->slots[0]; 24264afd2fe8SJohannes Thumshirn 24274afd2fe8SJohannes Thumshirn read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), 24284afd2fe8SJohannes Thumshirn sizeof(bgi)); 24294afd2fe8SJohannes Thumshirn 24304afd2fe8SJohannes Thumshirn btrfs_item_key_to_cpu(leaf, &key, slot); 24314afd2fe8SJohannes Thumshirn btrfs_release_path(path); 24324afd2fe8SJohannes Thumshirn ret = read_one_block_group(info, &bgi, &key, need_clear); 2433ffb9e0f0SQu Wenruo if (ret < 0) 24344358d963SJosef Bacik goto error; 2435ffb9e0f0SQu Wenruo key.objectid += key.offset; 2436ffb9e0f0SQu Wenruo key.offset = 0; 24374358d963SJosef Bacik } 24387837fa88SJosef Bacik btrfs_release_path(path); 24394358d963SJosef Bacik 244072804905SJosef Bacik list_for_each_entry(space_info, &info->space_info, list) { 244149ea112dSJosef Bacik int i; 244249ea112dSJosef Bacik 244349ea112dSJosef Bacik for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 244449ea112dSJosef Bacik if (list_empty(&space_info->block_groups[i])) 244549ea112dSJosef Bacik continue; 244649ea112dSJosef Bacik cache = list_first_entry(&space_info->block_groups[i], 244749ea112dSJosef Bacik struct btrfs_block_group, 244849ea112dSJosef Bacik list); 244949ea112dSJosef Bacik btrfs_sysfs_add_block_group_type(cache); 245049ea112dSJosef Bacik } 245149ea112dSJosef Bacik 24524358d963SJosef Bacik if (!(btrfs_get_alloc_profile(info, space_info->flags) & 24534358d963SJosef Bacik (BTRFS_BLOCK_GROUP_RAID10 | 24544358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID1_MASK | 24554358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID56_MASK | 24564358d963SJosef Bacik BTRFS_BLOCK_GROUP_DUP))) 24574358d963SJosef Bacik continue; 24584358d963SJosef Bacik /* 24594358d963SJosef Bacik * Avoid allocating from un-mirrored block group if there are 24604358d963SJosef Bacik * mirrored block groups. 24614358d963SJosef Bacik */ 24624358d963SJosef Bacik list_for_each_entry(cache, 24634358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_RAID0], 24644358d963SJosef Bacik list) 2465e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 24664358d963SJosef Bacik list_for_each_entry(cache, 24674358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_SINGLE], 24684358d963SJosef Bacik list) 2469e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 24704358d963SJosef Bacik } 24714358d963SJosef Bacik 24724358d963SJosef Bacik btrfs_init_global_block_rsv(info); 24734358d963SJosef Bacik ret = check_chunk_block_group_mappings(info); 24744358d963SJosef Bacik error: 24754358d963SJosef Bacik btrfs_free_path(path); 24762b29726cSQu Wenruo /* 24772b29726cSQu Wenruo * We've hit some error while reading the extent tree, and have 24782b29726cSQu Wenruo * rescue=ibadroots mount option. 24792b29726cSQu Wenruo * Try to fill the tree using dummy block groups so that the user can 24802b29726cSQu Wenruo * continue to mount and grab their data. 24812b29726cSQu Wenruo */ 24822b29726cSQu Wenruo if (ret && btrfs_test_opt(info, IGNOREBADROOTS)) 24832b29726cSQu Wenruo ret = fill_dummy_bgs(info); 24844358d963SJosef Bacik return ret; 24854358d963SJosef Bacik } 24864358d963SJosef Bacik 248779bd3712SFilipe Manana /* 248879bd3712SFilipe Manana * This function, insert_block_group_item(), belongs to the phase 2 of chunk 248979bd3712SFilipe Manana * allocation. 249079bd3712SFilipe Manana * 249179bd3712SFilipe Manana * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 249279bd3712SFilipe Manana * phases. 249379bd3712SFilipe Manana */ 249497f4728aSQu Wenruo static int insert_block_group_item(struct btrfs_trans_handle *trans, 249597f4728aSQu Wenruo struct btrfs_block_group *block_group) 249697f4728aSQu Wenruo { 249797f4728aSQu Wenruo struct btrfs_fs_info *fs_info = trans->fs_info; 249897f4728aSQu Wenruo struct btrfs_block_group_item bgi; 2499dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 250097f4728aSQu Wenruo struct btrfs_key key; 250197f4728aSQu Wenruo 250297f4728aSQu Wenruo spin_lock(&block_group->lock); 250397f4728aSQu Wenruo btrfs_set_stack_block_group_used(&bgi, block_group->used); 250497f4728aSQu Wenruo btrfs_set_stack_block_group_chunk_objectid(&bgi, 2505f7238e50SJosef Bacik block_group->global_root_id); 250697f4728aSQu Wenruo btrfs_set_stack_block_group_flags(&bgi, block_group->flags); 250797f4728aSQu Wenruo key.objectid = block_group->start; 250897f4728aSQu Wenruo key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 250997f4728aSQu Wenruo key.offset = block_group->length; 251097f4728aSQu Wenruo spin_unlock(&block_group->lock); 251197f4728aSQu Wenruo 251297f4728aSQu Wenruo return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi)); 251397f4728aSQu Wenruo } 251497f4728aSQu Wenruo 25152eadb9e7SNikolay Borisov static int insert_dev_extent(struct btrfs_trans_handle *trans, 25162eadb9e7SNikolay Borisov struct btrfs_device *device, u64 chunk_offset, 25172eadb9e7SNikolay Borisov u64 start, u64 num_bytes) 25182eadb9e7SNikolay Borisov { 25192eadb9e7SNikolay Borisov struct btrfs_fs_info *fs_info = device->fs_info; 25202eadb9e7SNikolay Borisov struct btrfs_root *root = fs_info->dev_root; 25212eadb9e7SNikolay Borisov struct btrfs_path *path; 25222eadb9e7SNikolay Borisov struct btrfs_dev_extent *extent; 25232eadb9e7SNikolay Borisov struct extent_buffer *leaf; 25242eadb9e7SNikolay Borisov struct btrfs_key key; 25252eadb9e7SNikolay Borisov int ret; 25262eadb9e7SNikolay Borisov 25272eadb9e7SNikolay Borisov WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); 25282eadb9e7SNikolay Borisov WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 25292eadb9e7SNikolay Borisov path = btrfs_alloc_path(); 25302eadb9e7SNikolay Borisov if (!path) 25312eadb9e7SNikolay Borisov return -ENOMEM; 25322eadb9e7SNikolay Borisov 25332eadb9e7SNikolay Borisov key.objectid = device->devid; 25342eadb9e7SNikolay Borisov key.type = BTRFS_DEV_EXTENT_KEY; 25352eadb9e7SNikolay Borisov key.offset = start; 25362eadb9e7SNikolay Borisov ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*extent)); 25372eadb9e7SNikolay Borisov if (ret) 25382eadb9e7SNikolay Borisov goto out; 25392eadb9e7SNikolay Borisov 25402eadb9e7SNikolay Borisov leaf = path->nodes[0]; 25412eadb9e7SNikolay Borisov extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); 25422eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID); 25432eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_objectid(leaf, extent, 25442eadb9e7SNikolay Borisov BTRFS_FIRST_CHUNK_TREE_OBJECTID); 25452eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); 25462eadb9e7SNikolay Borisov 25472eadb9e7SNikolay Borisov btrfs_set_dev_extent_length(leaf, extent, num_bytes); 25482eadb9e7SNikolay Borisov btrfs_mark_buffer_dirty(leaf); 25492eadb9e7SNikolay Borisov out: 25502eadb9e7SNikolay Borisov btrfs_free_path(path); 25512eadb9e7SNikolay Borisov return ret; 25522eadb9e7SNikolay Borisov } 25532eadb9e7SNikolay Borisov 25542eadb9e7SNikolay Borisov /* 25552eadb9e7SNikolay Borisov * This function belongs to phase 2. 25562eadb9e7SNikolay Borisov * 25572eadb9e7SNikolay Borisov * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 25582eadb9e7SNikolay Borisov * phases. 25592eadb9e7SNikolay Borisov */ 25602eadb9e7SNikolay Borisov static int insert_dev_extents(struct btrfs_trans_handle *trans, 25612eadb9e7SNikolay Borisov u64 chunk_offset, u64 chunk_size) 25622eadb9e7SNikolay Borisov { 25632eadb9e7SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 25642eadb9e7SNikolay Borisov struct btrfs_device *device; 25652eadb9e7SNikolay Borisov struct extent_map *em; 25662eadb9e7SNikolay Borisov struct map_lookup *map; 25672eadb9e7SNikolay Borisov u64 dev_offset; 25682eadb9e7SNikolay Borisov u64 stripe_size; 25692eadb9e7SNikolay Borisov int i; 25702eadb9e7SNikolay Borisov int ret = 0; 25712eadb9e7SNikolay Borisov 25722eadb9e7SNikolay Borisov em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); 25732eadb9e7SNikolay Borisov if (IS_ERR(em)) 25742eadb9e7SNikolay Borisov return PTR_ERR(em); 25752eadb9e7SNikolay Borisov 25762eadb9e7SNikolay Borisov map = em->map_lookup; 25772eadb9e7SNikolay Borisov stripe_size = em->orig_block_len; 25782eadb9e7SNikolay Borisov 25792eadb9e7SNikolay Borisov /* 25802eadb9e7SNikolay Borisov * Take the device list mutex to prevent races with the final phase of 25812eadb9e7SNikolay Borisov * a device replace operation that replaces the device object associated 25822eadb9e7SNikolay Borisov * with the map's stripes, because the device object's id can change 25832eadb9e7SNikolay Borisov * at any time during that final phase of the device replace operation 25842eadb9e7SNikolay Borisov * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 25852eadb9e7SNikolay Borisov * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 25862eadb9e7SNikolay Borisov * resulting in persisting a device extent item with such ID. 25872eadb9e7SNikolay Borisov */ 25882eadb9e7SNikolay Borisov mutex_lock(&fs_info->fs_devices->device_list_mutex); 25892eadb9e7SNikolay Borisov for (i = 0; i < map->num_stripes; i++) { 25902eadb9e7SNikolay Borisov device = map->stripes[i].dev; 25912eadb9e7SNikolay Borisov dev_offset = map->stripes[i].physical; 25922eadb9e7SNikolay Borisov 25932eadb9e7SNikolay Borisov ret = insert_dev_extent(trans, device, chunk_offset, dev_offset, 25942eadb9e7SNikolay Borisov stripe_size); 25952eadb9e7SNikolay Borisov if (ret) 25962eadb9e7SNikolay Borisov break; 25972eadb9e7SNikolay Borisov } 25982eadb9e7SNikolay Borisov mutex_unlock(&fs_info->fs_devices->device_list_mutex); 25992eadb9e7SNikolay Borisov 26002eadb9e7SNikolay Borisov free_extent_map(em); 26012eadb9e7SNikolay Borisov return ret; 26022eadb9e7SNikolay Borisov } 26032eadb9e7SNikolay Borisov 260479bd3712SFilipe Manana /* 260579bd3712SFilipe Manana * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of 260679bd3712SFilipe Manana * chunk allocation. 260779bd3712SFilipe Manana * 260879bd3712SFilipe Manana * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 260979bd3712SFilipe Manana * phases. 261079bd3712SFilipe Manana */ 26114358d963SJosef Bacik void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) 26124358d963SJosef Bacik { 26134358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 261432da5386SDavid Sterba struct btrfs_block_group *block_group; 26154358d963SJosef Bacik int ret = 0; 26164358d963SJosef Bacik 26174358d963SJosef Bacik while (!list_empty(&trans->new_bgs)) { 261849ea112dSJosef Bacik int index; 261949ea112dSJosef Bacik 26204358d963SJosef Bacik block_group = list_first_entry(&trans->new_bgs, 262132da5386SDavid Sterba struct btrfs_block_group, 26224358d963SJosef Bacik bg_list); 26234358d963SJosef Bacik if (ret) 26244358d963SJosef Bacik goto next; 26254358d963SJosef Bacik 262649ea112dSJosef Bacik index = btrfs_bg_flags_to_raid_index(block_group->flags); 262749ea112dSJosef Bacik 262897f4728aSQu Wenruo ret = insert_block_group_item(trans, block_group); 26294358d963SJosef Bacik if (ret) 26304358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 26313349b57fSJosef Bacik if (!test_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, 26323349b57fSJosef Bacik &block_group->runtime_flags)) { 263379bd3712SFilipe Manana mutex_lock(&fs_info->chunk_mutex); 263479bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, block_group); 263579bd3712SFilipe Manana mutex_unlock(&fs_info->chunk_mutex); 263679bd3712SFilipe Manana if (ret) 263779bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 263879bd3712SFilipe Manana } 26392eadb9e7SNikolay Borisov ret = insert_dev_extents(trans, block_group->start, 264097f4728aSQu Wenruo block_group->length); 26414358d963SJosef Bacik if (ret) 26424358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 26434358d963SJosef Bacik add_block_group_free_space(trans, block_group); 264449ea112dSJosef Bacik 264549ea112dSJosef Bacik /* 264649ea112dSJosef Bacik * If we restriped during balance, we may have added a new raid 264749ea112dSJosef Bacik * type, so now add the sysfs entries when it is safe to do so. 264849ea112dSJosef Bacik * We don't have to worry about locking here as it's handled in 264949ea112dSJosef Bacik * btrfs_sysfs_add_block_group_type. 265049ea112dSJosef Bacik */ 265149ea112dSJosef Bacik if (block_group->space_info->block_group_kobjs[index] == NULL) 265249ea112dSJosef Bacik btrfs_sysfs_add_block_group_type(block_group); 265349ea112dSJosef Bacik 26544358d963SJosef Bacik /* Already aborted the transaction if it failed. */ 26554358d963SJosef Bacik next: 26564358d963SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 26574358d963SJosef Bacik list_del_init(&block_group->bg_list); 26584358d963SJosef Bacik } 26594358d963SJosef Bacik btrfs_trans_release_chunk_metadata(trans); 26604358d963SJosef Bacik } 26614358d963SJosef Bacik 2662f7238e50SJosef Bacik /* 2663f7238e50SJosef Bacik * For extent tree v2 we use the block_group_item->chunk_offset to point at our 2664f7238e50SJosef Bacik * global root id. For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID. 2665f7238e50SJosef Bacik */ 2666f7238e50SJosef Bacik static u64 calculate_global_root_id(struct btrfs_fs_info *fs_info, u64 offset) 2667f7238e50SJosef Bacik { 2668f7238e50SJosef Bacik u64 div = SZ_1G; 2669f7238e50SJosef Bacik u64 index; 2670f7238e50SJosef Bacik 2671f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 2672f7238e50SJosef Bacik return BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2673f7238e50SJosef Bacik 2674f7238e50SJosef Bacik /* If we have a smaller fs index based on 128MiB. */ 2675f7238e50SJosef Bacik if (btrfs_super_total_bytes(fs_info->super_copy) <= (SZ_1G * 10ULL)) 2676f7238e50SJosef Bacik div = SZ_128M; 2677f7238e50SJosef Bacik 2678f7238e50SJosef Bacik offset = div64_u64(offset, div); 2679f7238e50SJosef Bacik div64_u64_rem(offset, fs_info->nr_global_roots, &index); 2680f7238e50SJosef Bacik return index; 2681f7238e50SJosef Bacik } 2682f7238e50SJosef Bacik 268379bd3712SFilipe Manana struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans, 268479bd3712SFilipe Manana u64 bytes_used, u64 type, 268579bd3712SFilipe Manana u64 chunk_offset, u64 size) 26864358d963SJosef Bacik { 26874358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 268832da5386SDavid Sterba struct btrfs_block_group *cache; 26894358d963SJosef Bacik int ret; 26904358d963SJosef Bacik 26914358d963SJosef Bacik btrfs_set_log_full_commit(trans); 26924358d963SJosef Bacik 26939afc6649SQu Wenruo cache = btrfs_create_block_group_cache(fs_info, chunk_offset); 26944358d963SJosef Bacik if (!cache) 269579bd3712SFilipe Manana return ERR_PTR(-ENOMEM); 26964358d963SJosef Bacik 26979afc6649SQu Wenruo cache->length = size; 2698e3e39c72SMarcos Paulo de Souza set_free_space_tree_thresholds(cache); 2699bf38be65SDavid Sterba cache->used = bytes_used; 27004358d963SJosef Bacik cache->flags = type; 27014358d963SJosef Bacik cache->cached = BTRFS_CACHE_FINISHED; 2702f7238e50SJosef Bacik cache->global_root_id = calculate_global_root_id(fs_info, cache->start); 2703f7238e50SJosef Bacik 2704997e3e2eSBoris Burkov if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) 27050d7764ffSDavid Sterba set_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &cache->runtime_flags); 270608e11a3dSNaohiro Aota 2707a94794d5SNaohiro Aota ret = btrfs_load_block_group_zone_info(cache, true); 270808e11a3dSNaohiro Aota if (ret) { 270908e11a3dSNaohiro Aota btrfs_put_block_group(cache); 271079bd3712SFilipe Manana return ERR_PTR(ret); 271108e11a3dSNaohiro Aota } 271208e11a3dSNaohiro Aota 27134358d963SJosef Bacik ret = exclude_super_stripes(cache); 27144358d963SJosef Bacik if (ret) { 27154358d963SJosef Bacik /* We may have excluded something, so call this just in case */ 27164358d963SJosef Bacik btrfs_free_excluded_extents(cache); 27174358d963SJosef Bacik btrfs_put_block_group(cache); 271879bd3712SFilipe Manana return ERR_PTR(ret); 27194358d963SJosef Bacik } 27204358d963SJosef Bacik 27214358d963SJosef Bacik add_new_free_space(cache, chunk_offset, chunk_offset + size); 27224358d963SJosef Bacik 27234358d963SJosef Bacik btrfs_free_excluded_extents(cache); 27244358d963SJosef Bacik 27254358d963SJosef Bacik /* 27264358d963SJosef Bacik * Ensure the corresponding space_info object is created and 27274358d963SJosef Bacik * assigned to our block group. We want our bg to be added to the rbtree 27284358d963SJosef Bacik * with its ->space_info set. 27294358d963SJosef Bacik */ 27304358d963SJosef Bacik cache->space_info = btrfs_find_space_info(fs_info, cache->flags); 27314358d963SJosef Bacik ASSERT(cache->space_info); 27324358d963SJosef Bacik 27334358d963SJosef Bacik ret = btrfs_add_block_group_cache(fs_info, cache); 27344358d963SJosef Bacik if (ret) { 27354358d963SJosef Bacik btrfs_remove_free_space_cache(cache); 27364358d963SJosef Bacik btrfs_put_block_group(cache); 273779bd3712SFilipe Manana return ERR_PTR(ret); 27384358d963SJosef Bacik } 27394358d963SJosef Bacik 27404358d963SJosef Bacik /* 27414358d963SJosef Bacik * Now that our block group has its ->space_info set and is inserted in 27424358d963SJosef Bacik * the rbtree, update the space info's counters. 27434358d963SJosef Bacik */ 27444358d963SJosef Bacik trace_btrfs_add_block_group(fs_info, cache, 1); 2745723de71dSJosef Bacik btrfs_add_bg_to_space_info(fs_info, cache); 27464358d963SJosef Bacik btrfs_update_global_block_rsv(fs_info); 27474358d963SJosef Bacik 27489d4b0a12SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 27499d4b0a12SJosef Bacik if (btrfs_should_fragment_free_space(cache)) { 27509d4b0a12SJosef Bacik u64 new_bytes_used = size - bytes_used; 27519d4b0a12SJosef Bacik 27529d4b0a12SJosef Bacik cache->space_info->bytes_used += new_bytes_used >> 1; 27539d4b0a12SJosef Bacik fragment_free_space(cache); 27549d4b0a12SJosef Bacik } 27559d4b0a12SJosef Bacik #endif 27564358d963SJosef Bacik 27574358d963SJosef Bacik list_add_tail(&cache->bg_list, &trans->new_bgs); 27584358d963SJosef Bacik trans->delayed_ref_updates++; 27594358d963SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 27604358d963SJosef Bacik 27614358d963SJosef Bacik set_avail_alloc_bits(fs_info, type); 276279bd3712SFilipe Manana return cache; 27634358d963SJosef Bacik } 276426ce2095SJosef Bacik 2765b12de528SQu Wenruo /* 2766b12de528SQu Wenruo * Mark one block group RO, can be called several times for the same block 2767b12de528SQu Wenruo * group. 2768b12de528SQu Wenruo * 2769b12de528SQu Wenruo * @cache: the destination block group 2770b12de528SQu Wenruo * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to 2771b12de528SQu Wenruo * ensure we still have some free space after marking this 2772b12de528SQu Wenruo * block group RO. 2773b12de528SQu Wenruo */ 2774b12de528SQu Wenruo int btrfs_inc_block_group_ro(struct btrfs_block_group *cache, 2775b12de528SQu Wenruo bool do_chunk_alloc) 277626ce2095SJosef Bacik { 277726ce2095SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 277826ce2095SJosef Bacik struct btrfs_trans_handle *trans; 2779dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 278026ce2095SJosef Bacik u64 alloc_flags; 278126ce2095SJosef Bacik int ret; 2782b6e9f16cSNikolay Borisov bool dirty_bg_running; 278326ce2095SJosef Bacik 27842d192fc4SQu Wenruo /* 27852d192fc4SQu Wenruo * This can only happen when we are doing read-only scrub on read-only 27862d192fc4SQu Wenruo * mount. 27872d192fc4SQu Wenruo * In that case we should not start a new transaction on read-only fs. 27882d192fc4SQu Wenruo * Thus here we skip all chunk allocations. 27892d192fc4SQu Wenruo */ 27902d192fc4SQu Wenruo if (sb_rdonly(fs_info->sb)) { 27912d192fc4SQu Wenruo mutex_lock(&fs_info->ro_block_group_mutex); 27922d192fc4SQu Wenruo ret = inc_block_group_ro(cache, 0); 27932d192fc4SQu Wenruo mutex_unlock(&fs_info->ro_block_group_mutex); 27942d192fc4SQu Wenruo return ret; 27952d192fc4SQu Wenruo } 27962d192fc4SQu Wenruo 2797b6e9f16cSNikolay Borisov do { 2798dfe8aec4SJosef Bacik trans = btrfs_join_transaction(root); 279926ce2095SJosef Bacik if (IS_ERR(trans)) 280026ce2095SJosef Bacik return PTR_ERR(trans); 280126ce2095SJosef Bacik 2802b6e9f16cSNikolay Borisov dirty_bg_running = false; 2803b6e9f16cSNikolay Borisov 280426ce2095SJosef Bacik /* 2805b6e9f16cSNikolay Borisov * We're not allowed to set block groups readonly after the dirty 2806b6e9f16cSNikolay Borisov * block group cache has started writing. If it already started, 2807b6e9f16cSNikolay Borisov * back off and let this transaction commit. 280826ce2095SJosef Bacik */ 280926ce2095SJosef Bacik mutex_lock(&fs_info->ro_block_group_mutex); 281026ce2095SJosef Bacik if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { 281126ce2095SJosef Bacik u64 transid = trans->transid; 281226ce2095SJosef Bacik 281326ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 281426ce2095SJosef Bacik btrfs_end_transaction(trans); 281526ce2095SJosef Bacik 281626ce2095SJosef Bacik ret = btrfs_wait_for_commit(fs_info, transid); 281726ce2095SJosef Bacik if (ret) 281826ce2095SJosef Bacik return ret; 2819b6e9f16cSNikolay Borisov dirty_bg_running = true; 282026ce2095SJosef Bacik } 2821b6e9f16cSNikolay Borisov } while (dirty_bg_running); 282226ce2095SJosef Bacik 2823b12de528SQu Wenruo if (do_chunk_alloc) { 282426ce2095SJosef Bacik /* 2825b12de528SQu Wenruo * If we are changing raid levels, try to allocate a 2826b12de528SQu Wenruo * corresponding block group with the new raid level. 282726ce2095SJosef Bacik */ 2828349e120eSJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); 282926ce2095SJosef Bacik if (alloc_flags != cache->flags) { 2830b12de528SQu Wenruo ret = btrfs_chunk_alloc(trans, alloc_flags, 2831b12de528SQu Wenruo CHUNK_ALLOC_FORCE); 283226ce2095SJosef Bacik /* 283326ce2095SJosef Bacik * ENOSPC is allowed here, we may have enough space 2834b12de528SQu Wenruo * already allocated at the new raid level to carry on 283526ce2095SJosef Bacik */ 283626ce2095SJosef Bacik if (ret == -ENOSPC) 283726ce2095SJosef Bacik ret = 0; 283826ce2095SJosef Bacik if (ret < 0) 283926ce2095SJosef Bacik goto out; 284026ce2095SJosef Bacik } 2841b12de528SQu Wenruo } 284226ce2095SJosef Bacik 2843a7a63accSJosef Bacik ret = inc_block_group_ro(cache, 0); 2844195a49eaSFilipe Manana if (!do_chunk_alloc || ret == -ETXTBSY) 2845b12de528SQu Wenruo goto unlock_out; 284626ce2095SJosef Bacik if (!ret) 284726ce2095SJosef Bacik goto out; 284826ce2095SJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); 284926ce2095SJosef Bacik ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 285026ce2095SJosef Bacik if (ret < 0) 285126ce2095SJosef Bacik goto out; 2852b6a98021SNaohiro Aota /* 2853b6a98021SNaohiro Aota * We have allocated a new chunk. We also need to activate that chunk to 2854b6a98021SNaohiro Aota * grant metadata tickets for zoned filesystem. 2855b6a98021SNaohiro Aota */ 2856b6a98021SNaohiro Aota ret = btrfs_zoned_activate_one_bg(fs_info, cache->space_info, true); 2857b6a98021SNaohiro Aota if (ret < 0) 2858b6a98021SNaohiro Aota goto out; 2859b6a98021SNaohiro Aota 2860e11c0406SJosef Bacik ret = inc_block_group_ro(cache, 0); 2861195a49eaSFilipe Manana if (ret == -ETXTBSY) 2862195a49eaSFilipe Manana goto unlock_out; 286326ce2095SJosef Bacik out: 286426ce2095SJosef Bacik if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { 2865349e120eSJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); 286626ce2095SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 286726ce2095SJosef Bacik check_system_chunk(trans, alloc_flags); 286826ce2095SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 286926ce2095SJosef Bacik } 2870b12de528SQu Wenruo unlock_out: 287126ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 287226ce2095SJosef Bacik 287326ce2095SJosef Bacik btrfs_end_transaction(trans); 287426ce2095SJosef Bacik return ret; 287526ce2095SJosef Bacik } 287626ce2095SJosef Bacik 287732da5386SDavid Sterba void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) 287826ce2095SJosef Bacik { 287926ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 288026ce2095SJosef Bacik u64 num_bytes; 288126ce2095SJosef Bacik 288226ce2095SJosef Bacik BUG_ON(!cache->ro); 288326ce2095SJosef Bacik 288426ce2095SJosef Bacik spin_lock(&sinfo->lock); 288526ce2095SJosef Bacik spin_lock(&cache->lock); 288626ce2095SJosef Bacik if (!--cache->ro) { 2887169e0da9SNaohiro Aota if (btrfs_is_zoned(cache->fs_info)) { 2888169e0da9SNaohiro Aota /* Migrate zone_unusable bytes back */ 288998173255SNaohiro Aota cache->zone_unusable = 289098173255SNaohiro Aota (cache->alloc_offset - cache->used) + 289198173255SNaohiro Aota (cache->length - cache->zone_capacity); 2892169e0da9SNaohiro Aota sinfo->bytes_zone_unusable += cache->zone_unusable; 2893169e0da9SNaohiro Aota sinfo->bytes_readonly -= cache->zone_unusable; 2894169e0da9SNaohiro Aota } 2895f9f28e5bSNaohiro Aota num_bytes = cache->length - cache->reserved - 2896f9f28e5bSNaohiro Aota cache->pinned - cache->bytes_super - 2897f9f28e5bSNaohiro Aota cache->zone_unusable - cache->used; 2898f9f28e5bSNaohiro Aota sinfo->bytes_readonly -= num_bytes; 289926ce2095SJosef Bacik list_del_init(&cache->ro_list); 290026ce2095SJosef Bacik } 290126ce2095SJosef Bacik spin_unlock(&cache->lock); 290226ce2095SJosef Bacik spin_unlock(&sinfo->lock); 290326ce2095SJosef Bacik } 290477745c05SJosef Bacik 29053be4d8efSQu Wenruo static int update_block_group_item(struct btrfs_trans_handle *trans, 290677745c05SJosef Bacik struct btrfs_path *path, 290732da5386SDavid Sterba struct btrfs_block_group *cache) 290877745c05SJosef Bacik { 290977745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 291077745c05SJosef Bacik int ret; 2911dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 291277745c05SJosef Bacik unsigned long bi; 291377745c05SJosef Bacik struct extent_buffer *leaf; 2914bf38be65SDavid Sterba struct btrfs_block_group_item bgi; 2915b3470b5dSDavid Sterba struct btrfs_key key; 29167248e0ceSQu Wenruo u64 old_commit_used; 29177248e0ceSQu Wenruo u64 used; 29187248e0ceSQu Wenruo 29197248e0ceSQu Wenruo /* 29207248e0ceSQu Wenruo * Block group items update can be triggered out of commit transaction 29217248e0ceSQu Wenruo * critical section, thus we need a consistent view of used bytes. 29227248e0ceSQu Wenruo * We cannot use cache->used directly outside of the spin lock, as it 29237248e0ceSQu Wenruo * may be changed. 29247248e0ceSQu Wenruo */ 29257248e0ceSQu Wenruo spin_lock(&cache->lock); 29267248e0ceSQu Wenruo old_commit_used = cache->commit_used; 29277248e0ceSQu Wenruo used = cache->used; 29287248e0ceSQu Wenruo /* No change in used bytes, can safely skip it. */ 29297248e0ceSQu Wenruo if (cache->commit_used == used) { 29307248e0ceSQu Wenruo spin_unlock(&cache->lock); 29317248e0ceSQu Wenruo return 0; 29327248e0ceSQu Wenruo } 29337248e0ceSQu Wenruo cache->commit_used = used; 29347248e0ceSQu Wenruo spin_unlock(&cache->lock); 293577745c05SJosef Bacik 2936b3470b5dSDavid Sterba key.objectid = cache->start; 2937b3470b5dSDavid Sterba key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 2938b3470b5dSDavid Sterba key.offset = cache->length; 2939b3470b5dSDavid Sterba 29403be4d8efSQu Wenruo ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 294177745c05SJosef Bacik if (ret) { 294277745c05SJosef Bacik if (ret > 0) 294377745c05SJosef Bacik ret = -ENOENT; 294477745c05SJosef Bacik goto fail; 294577745c05SJosef Bacik } 294677745c05SJosef Bacik 294777745c05SJosef Bacik leaf = path->nodes[0]; 294877745c05SJosef Bacik bi = btrfs_item_ptr_offset(leaf, path->slots[0]); 29497248e0ceSQu Wenruo btrfs_set_stack_block_group_used(&bgi, used); 2950de0dc456SDavid Sterba btrfs_set_stack_block_group_chunk_objectid(&bgi, 2951f7238e50SJosef Bacik cache->global_root_id); 2952de0dc456SDavid Sterba btrfs_set_stack_block_group_flags(&bgi, cache->flags); 2953bf38be65SDavid Sterba write_extent_buffer(leaf, &bgi, bi, sizeof(bgi)); 295477745c05SJosef Bacik btrfs_mark_buffer_dirty(leaf); 295577745c05SJosef Bacik fail: 295677745c05SJosef Bacik btrfs_release_path(path); 29577248e0ceSQu Wenruo /* We didn't update the block group item, need to revert @commit_used. */ 29587248e0ceSQu Wenruo if (ret < 0) { 29597248e0ceSQu Wenruo spin_lock(&cache->lock); 29607248e0ceSQu Wenruo cache->commit_used = old_commit_used; 29617248e0ceSQu Wenruo spin_unlock(&cache->lock); 29627248e0ceSQu Wenruo } 296377745c05SJosef Bacik return ret; 296477745c05SJosef Bacik 296577745c05SJosef Bacik } 296677745c05SJosef Bacik 296732da5386SDavid Sterba static int cache_save_setup(struct btrfs_block_group *block_group, 296877745c05SJosef Bacik struct btrfs_trans_handle *trans, 296977745c05SJosef Bacik struct btrfs_path *path) 297077745c05SJosef Bacik { 297177745c05SJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 297277745c05SJosef Bacik struct btrfs_root *root = fs_info->tree_root; 297377745c05SJosef Bacik struct inode *inode = NULL; 297477745c05SJosef Bacik struct extent_changeset *data_reserved = NULL; 297577745c05SJosef Bacik u64 alloc_hint = 0; 297677745c05SJosef Bacik int dcs = BTRFS_DC_ERROR; 29770044ae11SQu Wenruo u64 cache_size = 0; 297877745c05SJosef Bacik int retries = 0; 297977745c05SJosef Bacik int ret = 0; 298077745c05SJosef Bacik 2981af456a2cSBoris Burkov if (!btrfs_test_opt(fs_info, SPACE_CACHE)) 2982af456a2cSBoris Burkov return 0; 2983af456a2cSBoris Burkov 298477745c05SJosef Bacik /* 298577745c05SJosef Bacik * If this block group is smaller than 100 megs don't bother caching the 298677745c05SJosef Bacik * block group. 298777745c05SJosef Bacik */ 2988b3470b5dSDavid Sterba if (block_group->length < (100 * SZ_1M)) { 298977745c05SJosef Bacik spin_lock(&block_group->lock); 299077745c05SJosef Bacik block_group->disk_cache_state = BTRFS_DC_WRITTEN; 299177745c05SJosef Bacik spin_unlock(&block_group->lock); 299277745c05SJosef Bacik return 0; 299377745c05SJosef Bacik } 299477745c05SJosef Bacik 2995bf31f87fSDavid Sterba if (TRANS_ABORTED(trans)) 299677745c05SJosef Bacik return 0; 299777745c05SJosef Bacik again: 299877745c05SJosef Bacik inode = lookup_free_space_inode(block_group, path); 299977745c05SJosef Bacik if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { 300077745c05SJosef Bacik ret = PTR_ERR(inode); 300177745c05SJosef Bacik btrfs_release_path(path); 300277745c05SJosef Bacik goto out; 300377745c05SJosef Bacik } 300477745c05SJosef Bacik 300577745c05SJosef Bacik if (IS_ERR(inode)) { 300677745c05SJosef Bacik BUG_ON(retries); 300777745c05SJosef Bacik retries++; 300877745c05SJosef Bacik 300977745c05SJosef Bacik if (block_group->ro) 301077745c05SJosef Bacik goto out_free; 301177745c05SJosef Bacik 301277745c05SJosef Bacik ret = create_free_space_inode(trans, block_group, path); 301377745c05SJosef Bacik if (ret) 301477745c05SJosef Bacik goto out_free; 301577745c05SJosef Bacik goto again; 301677745c05SJosef Bacik } 301777745c05SJosef Bacik 301877745c05SJosef Bacik /* 301977745c05SJosef Bacik * We want to set the generation to 0, that way if anything goes wrong 302077745c05SJosef Bacik * from here on out we know not to trust this cache when we load up next 302177745c05SJosef Bacik * time. 302277745c05SJosef Bacik */ 302377745c05SJosef Bacik BTRFS_I(inode)->generation = 0; 30249a56fcd1SNikolay Borisov ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); 302577745c05SJosef Bacik if (ret) { 302677745c05SJosef Bacik /* 302777745c05SJosef Bacik * So theoretically we could recover from this, simply set the 302877745c05SJosef Bacik * super cache generation to 0 so we know to invalidate the 302977745c05SJosef Bacik * cache, but then we'd have to keep track of the block groups 303077745c05SJosef Bacik * that fail this way so we know we _have_ to reset this cache 303177745c05SJosef Bacik * before the next commit or risk reading stale cache. So to 303277745c05SJosef Bacik * limit our exposure to horrible edge cases lets just abort the 303377745c05SJosef Bacik * transaction, this only happens in really bad situations 303477745c05SJosef Bacik * anyway. 303577745c05SJosef Bacik */ 303677745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 303777745c05SJosef Bacik goto out_put; 303877745c05SJosef Bacik } 303977745c05SJosef Bacik WARN_ON(ret); 304077745c05SJosef Bacik 304177745c05SJosef Bacik /* We've already setup this transaction, go ahead and exit */ 304277745c05SJosef Bacik if (block_group->cache_generation == trans->transid && 304377745c05SJosef Bacik i_size_read(inode)) { 304477745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 304577745c05SJosef Bacik goto out_put; 304677745c05SJosef Bacik } 304777745c05SJosef Bacik 304877745c05SJosef Bacik if (i_size_read(inode) > 0) { 304977745c05SJosef Bacik ret = btrfs_check_trunc_cache_free_space(fs_info, 305077745c05SJosef Bacik &fs_info->global_block_rsv); 305177745c05SJosef Bacik if (ret) 305277745c05SJosef Bacik goto out_put; 305377745c05SJosef Bacik 305477745c05SJosef Bacik ret = btrfs_truncate_free_space_cache(trans, NULL, inode); 305577745c05SJosef Bacik if (ret) 305677745c05SJosef Bacik goto out_put; 305777745c05SJosef Bacik } 305877745c05SJosef Bacik 305977745c05SJosef Bacik spin_lock(&block_group->lock); 306077745c05SJosef Bacik if (block_group->cached != BTRFS_CACHE_FINISHED || 306177745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) { 306277745c05SJosef Bacik /* 306377745c05SJosef Bacik * don't bother trying to write stuff out _if_ 306477745c05SJosef Bacik * a) we're not cached, 306577745c05SJosef Bacik * b) we're with nospace_cache mount option, 306677745c05SJosef Bacik * c) we're with v2 space_cache (FREE_SPACE_TREE). 306777745c05SJosef Bacik */ 306877745c05SJosef Bacik dcs = BTRFS_DC_WRITTEN; 306977745c05SJosef Bacik spin_unlock(&block_group->lock); 307077745c05SJosef Bacik goto out_put; 307177745c05SJosef Bacik } 307277745c05SJosef Bacik spin_unlock(&block_group->lock); 307377745c05SJosef Bacik 307477745c05SJosef Bacik /* 307577745c05SJosef Bacik * We hit an ENOSPC when setting up the cache in this transaction, just 307677745c05SJosef Bacik * skip doing the setup, we've already cleared the cache so we're safe. 307777745c05SJosef Bacik */ 307877745c05SJosef Bacik if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { 307977745c05SJosef Bacik ret = -ENOSPC; 308077745c05SJosef Bacik goto out_put; 308177745c05SJosef Bacik } 308277745c05SJosef Bacik 308377745c05SJosef Bacik /* 308477745c05SJosef Bacik * Try to preallocate enough space based on how big the block group is. 308577745c05SJosef Bacik * Keep in mind this has to include any pinned space which could end up 308677745c05SJosef Bacik * taking up quite a bit since it's not folded into the other space 308777745c05SJosef Bacik * cache. 308877745c05SJosef Bacik */ 30890044ae11SQu Wenruo cache_size = div_u64(block_group->length, SZ_256M); 30900044ae11SQu Wenruo if (!cache_size) 30910044ae11SQu Wenruo cache_size = 1; 309277745c05SJosef Bacik 30930044ae11SQu Wenruo cache_size *= 16; 30940044ae11SQu Wenruo cache_size *= fs_info->sectorsize; 309577745c05SJosef Bacik 309636ea6f3eSNikolay Borisov ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0, 30971daedb1dSJosef Bacik cache_size, false); 309877745c05SJosef Bacik if (ret) 309977745c05SJosef Bacik goto out_put; 310077745c05SJosef Bacik 31010044ae11SQu Wenruo ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size, 31020044ae11SQu Wenruo cache_size, cache_size, 310377745c05SJosef Bacik &alloc_hint); 310477745c05SJosef Bacik /* 310577745c05SJosef Bacik * Our cache requires contiguous chunks so that we don't modify a bunch 310677745c05SJosef Bacik * of metadata or split extents when writing the cache out, which means 310777745c05SJosef Bacik * we can enospc if we are heavily fragmented in addition to just normal 310877745c05SJosef Bacik * out of space conditions. So if we hit this just skip setting up any 310977745c05SJosef Bacik * other block groups for this transaction, maybe we'll unpin enough 311077745c05SJosef Bacik * space the next time around. 311177745c05SJosef Bacik */ 311277745c05SJosef Bacik if (!ret) 311377745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 311477745c05SJosef Bacik else if (ret == -ENOSPC) 311577745c05SJosef Bacik set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); 311677745c05SJosef Bacik 311777745c05SJosef Bacik out_put: 311877745c05SJosef Bacik iput(inode); 311977745c05SJosef Bacik out_free: 312077745c05SJosef Bacik btrfs_release_path(path); 312177745c05SJosef Bacik out: 312277745c05SJosef Bacik spin_lock(&block_group->lock); 312377745c05SJosef Bacik if (!ret && dcs == BTRFS_DC_SETUP) 312477745c05SJosef Bacik block_group->cache_generation = trans->transid; 312577745c05SJosef Bacik block_group->disk_cache_state = dcs; 312677745c05SJosef Bacik spin_unlock(&block_group->lock); 312777745c05SJosef Bacik 312877745c05SJosef Bacik extent_changeset_free(data_reserved); 312977745c05SJosef Bacik return ret; 313077745c05SJosef Bacik } 313177745c05SJosef Bacik 313277745c05SJosef Bacik int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) 313377745c05SJosef Bacik { 313477745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 313532da5386SDavid Sterba struct btrfs_block_group *cache, *tmp; 313677745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 313777745c05SJosef Bacik struct btrfs_path *path; 313877745c05SJosef Bacik 313977745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs) || 314077745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) 314177745c05SJosef Bacik return 0; 314277745c05SJosef Bacik 314377745c05SJosef Bacik path = btrfs_alloc_path(); 314477745c05SJosef Bacik if (!path) 314577745c05SJosef Bacik return -ENOMEM; 314677745c05SJosef Bacik 314777745c05SJosef Bacik /* Could add new block groups, use _safe just in case */ 314877745c05SJosef Bacik list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, 314977745c05SJosef Bacik dirty_list) { 315077745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_CLEAR) 315177745c05SJosef Bacik cache_save_setup(cache, trans, path); 315277745c05SJosef Bacik } 315377745c05SJosef Bacik 315477745c05SJosef Bacik btrfs_free_path(path); 315577745c05SJosef Bacik return 0; 315677745c05SJosef Bacik } 315777745c05SJosef Bacik 315877745c05SJosef Bacik /* 315977745c05SJosef Bacik * Transaction commit does final block group cache writeback during a critical 316077745c05SJosef Bacik * section where nothing is allowed to change the FS. This is required in 316177745c05SJosef Bacik * order for the cache to actually match the block group, but can introduce a 316277745c05SJosef Bacik * lot of latency into the commit. 316377745c05SJosef Bacik * 316477745c05SJosef Bacik * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. 316577745c05SJosef Bacik * There's a chance we'll have to redo some of it if the block group changes 316677745c05SJosef Bacik * again during the commit, but it greatly reduces the commit latency by 316777745c05SJosef Bacik * getting rid of the easy block groups while we're still allowing others to 316877745c05SJosef Bacik * join the commit. 316977745c05SJosef Bacik */ 317077745c05SJosef Bacik int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) 317177745c05SJosef Bacik { 317277745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 317332da5386SDavid Sterba struct btrfs_block_group *cache; 317477745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 317577745c05SJosef Bacik int ret = 0; 317677745c05SJosef Bacik int should_put; 317777745c05SJosef Bacik struct btrfs_path *path = NULL; 317877745c05SJosef Bacik LIST_HEAD(dirty); 317977745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 318077745c05SJosef Bacik int loops = 0; 318177745c05SJosef Bacik 318277745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 318377745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs)) { 318477745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 318577745c05SJosef Bacik return 0; 318677745c05SJosef Bacik } 318777745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 318877745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 318977745c05SJosef Bacik 319077745c05SJosef Bacik again: 319177745c05SJosef Bacik /* Make sure all the block groups on our dirty list actually exist */ 319277745c05SJosef Bacik btrfs_create_pending_block_groups(trans); 319377745c05SJosef Bacik 319477745c05SJosef Bacik if (!path) { 319577745c05SJosef Bacik path = btrfs_alloc_path(); 3196938fcbfbSJosef Bacik if (!path) { 3197938fcbfbSJosef Bacik ret = -ENOMEM; 3198938fcbfbSJosef Bacik goto out; 3199938fcbfbSJosef Bacik } 320077745c05SJosef Bacik } 320177745c05SJosef Bacik 320277745c05SJosef Bacik /* 320377745c05SJosef Bacik * cache_write_mutex is here only to save us from balance or automatic 320477745c05SJosef Bacik * removal of empty block groups deleting this block group while we are 320577745c05SJosef Bacik * writing out the cache 320677745c05SJosef Bacik */ 320777745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 320877745c05SJosef Bacik while (!list_empty(&dirty)) { 320977745c05SJosef Bacik bool drop_reserve = true; 321077745c05SJosef Bacik 321132da5386SDavid Sterba cache = list_first_entry(&dirty, struct btrfs_block_group, 321277745c05SJosef Bacik dirty_list); 321377745c05SJosef Bacik /* 321477745c05SJosef Bacik * This can happen if something re-dirties a block group that 321577745c05SJosef Bacik * is already under IO. Just wait for it to finish and then do 321677745c05SJosef Bacik * it all again 321777745c05SJosef Bacik */ 321877745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 321977745c05SJosef Bacik list_del_init(&cache->io_list); 322077745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 322177745c05SJosef Bacik btrfs_put_block_group(cache); 322277745c05SJosef Bacik } 322377745c05SJosef Bacik 322477745c05SJosef Bacik 322577745c05SJosef Bacik /* 322677745c05SJosef Bacik * btrfs_wait_cache_io uses the cache->dirty_list to decide if 322777745c05SJosef Bacik * it should update the cache_state. Don't delete until after 322877745c05SJosef Bacik * we wait. 322977745c05SJosef Bacik * 323077745c05SJosef Bacik * Since we're not running in the commit critical section 323177745c05SJosef Bacik * we need the dirty_bgs_lock to protect from update_block_group 323277745c05SJosef Bacik */ 323377745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 323477745c05SJosef Bacik list_del_init(&cache->dirty_list); 323577745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 323677745c05SJosef Bacik 323777745c05SJosef Bacik should_put = 1; 323877745c05SJosef Bacik 323977745c05SJosef Bacik cache_save_setup(cache, trans, path); 324077745c05SJosef Bacik 324177745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_SETUP) { 324277745c05SJosef Bacik cache->io_ctl.inode = NULL; 324377745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 324477745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 324577745c05SJosef Bacik should_put = 0; 324677745c05SJosef Bacik 324777745c05SJosef Bacik /* 324877745c05SJosef Bacik * The cache_write_mutex is protecting the 324977745c05SJosef Bacik * io_list, also refer to the definition of 325077745c05SJosef Bacik * btrfs_transaction::io_bgs for more details 325177745c05SJosef Bacik */ 325277745c05SJosef Bacik list_add_tail(&cache->io_list, io); 325377745c05SJosef Bacik } else { 325477745c05SJosef Bacik /* 325577745c05SJosef Bacik * If we failed to write the cache, the 325677745c05SJosef Bacik * generation will be bad and life goes on 325777745c05SJosef Bacik */ 325877745c05SJosef Bacik ret = 0; 325977745c05SJosef Bacik } 326077745c05SJosef Bacik } 326177745c05SJosef Bacik if (!ret) { 32623be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 326377745c05SJosef Bacik /* 326477745c05SJosef Bacik * Our block group might still be attached to the list 326577745c05SJosef Bacik * of new block groups in the transaction handle of some 326677745c05SJosef Bacik * other task (struct btrfs_trans_handle->new_bgs). This 326777745c05SJosef Bacik * means its block group item isn't yet in the extent 326877745c05SJosef Bacik * tree. If this happens ignore the error, as we will 326977745c05SJosef Bacik * try again later in the critical section of the 327077745c05SJosef Bacik * transaction commit. 327177745c05SJosef Bacik */ 327277745c05SJosef Bacik if (ret == -ENOENT) { 327377745c05SJosef Bacik ret = 0; 327477745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 327577745c05SJosef Bacik if (list_empty(&cache->dirty_list)) { 327677745c05SJosef Bacik list_add_tail(&cache->dirty_list, 327777745c05SJosef Bacik &cur_trans->dirty_bgs); 327877745c05SJosef Bacik btrfs_get_block_group(cache); 327977745c05SJosef Bacik drop_reserve = false; 328077745c05SJosef Bacik } 328177745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 328277745c05SJosef Bacik } else if (ret) { 328377745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 328477745c05SJosef Bacik } 328577745c05SJosef Bacik } 328677745c05SJosef Bacik 328777745c05SJosef Bacik /* If it's not on the io list, we need to put the block group */ 328877745c05SJosef Bacik if (should_put) 328977745c05SJosef Bacik btrfs_put_block_group(cache); 329077745c05SJosef Bacik if (drop_reserve) 329177745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 329277745c05SJosef Bacik /* 329377745c05SJosef Bacik * Avoid blocking other tasks for too long. It might even save 329477745c05SJosef Bacik * us from writing caches for block groups that are going to be 329577745c05SJosef Bacik * removed. 329677745c05SJosef Bacik */ 329777745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 3298938fcbfbSJosef Bacik if (ret) 3299938fcbfbSJosef Bacik goto out; 330077745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 330177745c05SJosef Bacik } 330277745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 330377745c05SJosef Bacik 330477745c05SJosef Bacik /* 330577745c05SJosef Bacik * Go through delayed refs for all the stuff we've just kicked off 330677745c05SJosef Bacik * and then loop back (just once) 330777745c05SJosef Bacik */ 330834d1eb0eSJosef Bacik if (!ret) 330977745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 0); 331077745c05SJosef Bacik if (!ret && loops == 0) { 331177745c05SJosef Bacik loops++; 331277745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 331377745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 331477745c05SJosef Bacik /* 331577745c05SJosef Bacik * dirty_bgs_lock protects us from concurrent block group 331677745c05SJosef Bacik * deletes too (not just cache_write_mutex). 331777745c05SJosef Bacik */ 331877745c05SJosef Bacik if (!list_empty(&dirty)) { 331977745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 332077745c05SJosef Bacik goto again; 332177745c05SJosef Bacik } 332277745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 3323938fcbfbSJosef Bacik } 3324938fcbfbSJosef Bacik out: 3325938fcbfbSJosef Bacik if (ret < 0) { 3326938fcbfbSJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 3327938fcbfbSJosef Bacik list_splice_init(&dirty, &cur_trans->dirty_bgs); 3328938fcbfbSJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 332977745c05SJosef Bacik btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 333077745c05SJosef Bacik } 333177745c05SJosef Bacik 333277745c05SJosef Bacik btrfs_free_path(path); 333377745c05SJosef Bacik return ret; 333477745c05SJosef Bacik } 333577745c05SJosef Bacik 333677745c05SJosef Bacik int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) 333777745c05SJosef Bacik { 333877745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 333932da5386SDavid Sterba struct btrfs_block_group *cache; 334077745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 334177745c05SJosef Bacik int ret = 0; 334277745c05SJosef Bacik int should_put; 334377745c05SJosef Bacik struct btrfs_path *path; 334477745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 334577745c05SJosef Bacik 334677745c05SJosef Bacik path = btrfs_alloc_path(); 334777745c05SJosef Bacik if (!path) 334877745c05SJosef Bacik return -ENOMEM; 334977745c05SJosef Bacik 335077745c05SJosef Bacik /* 335177745c05SJosef Bacik * Even though we are in the critical section of the transaction commit, 335277745c05SJosef Bacik * we can still have concurrent tasks adding elements to this 335377745c05SJosef Bacik * transaction's list of dirty block groups. These tasks correspond to 335477745c05SJosef Bacik * endio free space workers started when writeback finishes for a 335577745c05SJosef Bacik * space cache, which run inode.c:btrfs_finish_ordered_io(), and can 335677745c05SJosef Bacik * allocate new block groups as a result of COWing nodes of the root 335777745c05SJosef Bacik * tree when updating the free space inode. The writeback for the space 335877745c05SJosef Bacik * caches is triggered by an earlier call to 335977745c05SJosef Bacik * btrfs_start_dirty_block_groups() and iterations of the following 336077745c05SJosef Bacik * loop. 336177745c05SJosef Bacik * Also we want to do the cache_save_setup first and then run the 336277745c05SJosef Bacik * delayed refs to make sure we have the best chance at doing this all 336377745c05SJosef Bacik * in one shot. 336477745c05SJosef Bacik */ 336577745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 336677745c05SJosef Bacik while (!list_empty(&cur_trans->dirty_bgs)) { 336777745c05SJosef Bacik cache = list_first_entry(&cur_trans->dirty_bgs, 336832da5386SDavid Sterba struct btrfs_block_group, 336977745c05SJosef Bacik dirty_list); 337077745c05SJosef Bacik 337177745c05SJosef Bacik /* 337277745c05SJosef Bacik * This can happen if cache_save_setup re-dirties a block group 337377745c05SJosef Bacik * that is already under IO. Just wait for it to finish and 337477745c05SJosef Bacik * then do it all again 337577745c05SJosef Bacik */ 337677745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 337777745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 337877745c05SJosef Bacik list_del_init(&cache->io_list); 337977745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 338077745c05SJosef Bacik btrfs_put_block_group(cache); 338177745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 338277745c05SJosef Bacik } 338377745c05SJosef Bacik 338477745c05SJosef Bacik /* 338577745c05SJosef Bacik * Don't remove from the dirty list until after we've waited on 338677745c05SJosef Bacik * any pending IO 338777745c05SJosef Bacik */ 338877745c05SJosef Bacik list_del_init(&cache->dirty_list); 338977745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 339077745c05SJosef Bacik should_put = 1; 339177745c05SJosef Bacik 339277745c05SJosef Bacik cache_save_setup(cache, trans, path); 339377745c05SJosef Bacik 339477745c05SJosef Bacik if (!ret) 339577745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 339677745c05SJosef Bacik (unsigned long) -1); 339777745c05SJosef Bacik 339877745c05SJosef Bacik if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { 339977745c05SJosef Bacik cache->io_ctl.inode = NULL; 340077745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 340177745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 340277745c05SJosef Bacik should_put = 0; 340377745c05SJosef Bacik list_add_tail(&cache->io_list, io); 340477745c05SJosef Bacik } else { 340577745c05SJosef Bacik /* 340677745c05SJosef Bacik * If we failed to write the cache, the 340777745c05SJosef Bacik * generation will be bad and life goes on 340877745c05SJosef Bacik */ 340977745c05SJosef Bacik ret = 0; 341077745c05SJosef Bacik } 341177745c05SJosef Bacik } 341277745c05SJosef Bacik if (!ret) { 34133be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 341477745c05SJosef Bacik /* 341577745c05SJosef Bacik * One of the free space endio workers might have 341677745c05SJosef Bacik * created a new block group while updating a free space 341777745c05SJosef Bacik * cache's inode (at inode.c:btrfs_finish_ordered_io()) 341877745c05SJosef Bacik * and hasn't released its transaction handle yet, in 341977745c05SJosef Bacik * which case the new block group is still attached to 342077745c05SJosef Bacik * its transaction handle and its creation has not 342177745c05SJosef Bacik * finished yet (no block group item in the extent tree 342277745c05SJosef Bacik * yet, etc). If this is the case, wait for all free 342377745c05SJosef Bacik * space endio workers to finish and retry. This is a 3424260db43cSRandy Dunlap * very rare case so no need for a more efficient and 342577745c05SJosef Bacik * complex approach. 342677745c05SJosef Bacik */ 342777745c05SJosef Bacik if (ret == -ENOENT) { 342877745c05SJosef Bacik wait_event(cur_trans->writer_wait, 342977745c05SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 34303be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 343177745c05SJosef Bacik } 343277745c05SJosef Bacik if (ret) 343377745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 343477745c05SJosef Bacik } 343577745c05SJosef Bacik 343677745c05SJosef Bacik /* If its not on the io list, we need to put the block group */ 343777745c05SJosef Bacik if (should_put) 343877745c05SJosef Bacik btrfs_put_block_group(cache); 343977745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 344077745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 344177745c05SJosef Bacik } 344277745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 344377745c05SJosef Bacik 344477745c05SJosef Bacik /* 344577745c05SJosef Bacik * Refer to the definition of io_bgs member for details why it's safe 344677745c05SJosef Bacik * to use it without any locking 344777745c05SJosef Bacik */ 344877745c05SJosef Bacik while (!list_empty(io)) { 344932da5386SDavid Sterba cache = list_first_entry(io, struct btrfs_block_group, 345077745c05SJosef Bacik io_list); 345177745c05SJosef Bacik list_del_init(&cache->io_list); 345277745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 345377745c05SJosef Bacik btrfs_put_block_group(cache); 345477745c05SJosef Bacik } 345577745c05SJosef Bacik 345677745c05SJosef Bacik btrfs_free_path(path); 345777745c05SJosef Bacik return ret; 345877745c05SJosef Bacik } 3459606d1bf1SJosef Bacik 3460606d1bf1SJosef Bacik int btrfs_update_block_group(struct btrfs_trans_handle *trans, 346111b66fa6SAnand Jain u64 bytenr, u64 num_bytes, bool alloc) 3462606d1bf1SJosef Bacik { 3463606d1bf1SJosef Bacik struct btrfs_fs_info *info = trans->fs_info; 346432da5386SDavid Sterba struct btrfs_block_group *cache = NULL; 3465606d1bf1SJosef Bacik u64 total = num_bytes; 3466606d1bf1SJosef Bacik u64 old_val; 3467606d1bf1SJosef Bacik u64 byte_in_group; 3468606d1bf1SJosef Bacik int factor; 3469606d1bf1SJosef Bacik int ret = 0; 3470606d1bf1SJosef Bacik 3471606d1bf1SJosef Bacik /* Block accounting for super block */ 3472606d1bf1SJosef Bacik spin_lock(&info->delalloc_root_lock); 3473606d1bf1SJosef Bacik old_val = btrfs_super_bytes_used(info->super_copy); 3474606d1bf1SJosef Bacik if (alloc) 3475606d1bf1SJosef Bacik old_val += num_bytes; 3476606d1bf1SJosef Bacik else 3477606d1bf1SJosef Bacik old_val -= num_bytes; 3478606d1bf1SJosef Bacik btrfs_set_super_bytes_used(info->super_copy, old_val); 3479606d1bf1SJosef Bacik spin_unlock(&info->delalloc_root_lock); 3480606d1bf1SJosef Bacik 3481606d1bf1SJosef Bacik while (total) { 3482efbf35a1SJosef Bacik bool reclaim = false; 3483ac2f1e63SJosef Bacik 3484606d1bf1SJosef Bacik cache = btrfs_lookup_block_group(info, bytenr); 3485606d1bf1SJosef Bacik if (!cache) { 3486606d1bf1SJosef Bacik ret = -ENOENT; 3487606d1bf1SJosef Bacik break; 3488606d1bf1SJosef Bacik } 3489606d1bf1SJosef Bacik factor = btrfs_bg_type_to_factor(cache->flags); 3490606d1bf1SJosef Bacik 3491606d1bf1SJosef Bacik /* 3492606d1bf1SJosef Bacik * If this block group has free space cache written out, we 3493606d1bf1SJosef Bacik * need to make sure to load it if we are removing space. This 3494606d1bf1SJosef Bacik * is because we need the unpinning stage to actually add the 3495606d1bf1SJosef Bacik * space back to the block group, otherwise we will leak space. 3496606d1bf1SJosef Bacik */ 349732da5386SDavid Sterba if (!alloc && !btrfs_block_group_done(cache)) 3498ced8ecf0SOmar Sandoval btrfs_cache_block_group(cache, true); 3499606d1bf1SJosef Bacik 3500b3470b5dSDavid Sterba byte_in_group = bytenr - cache->start; 3501b3470b5dSDavid Sterba WARN_ON(byte_in_group > cache->length); 3502606d1bf1SJosef Bacik 3503606d1bf1SJosef Bacik spin_lock(&cache->space_info->lock); 3504606d1bf1SJosef Bacik spin_lock(&cache->lock); 3505606d1bf1SJosef Bacik 3506606d1bf1SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 3507606d1bf1SJosef Bacik cache->disk_cache_state < BTRFS_DC_CLEAR) 3508606d1bf1SJosef Bacik cache->disk_cache_state = BTRFS_DC_CLEAR; 3509606d1bf1SJosef Bacik 3510bf38be65SDavid Sterba old_val = cache->used; 3511b3470b5dSDavid Sterba num_bytes = min(total, cache->length - byte_in_group); 3512606d1bf1SJosef Bacik if (alloc) { 3513606d1bf1SJosef Bacik old_val += num_bytes; 3514bf38be65SDavid Sterba cache->used = old_val; 3515606d1bf1SJosef Bacik cache->reserved -= num_bytes; 3516606d1bf1SJosef Bacik cache->space_info->bytes_reserved -= num_bytes; 3517606d1bf1SJosef Bacik cache->space_info->bytes_used += num_bytes; 3518606d1bf1SJosef Bacik cache->space_info->disk_used += num_bytes * factor; 3519606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3520606d1bf1SJosef Bacik spin_unlock(&cache->space_info->lock); 3521606d1bf1SJosef Bacik } else { 3522606d1bf1SJosef Bacik old_val -= num_bytes; 3523bf38be65SDavid Sterba cache->used = old_val; 3524606d1bf1SJosef Bacik cache->pinned += num_bytes; 3525606d1bf1SJosef Bacik btrfs_space_info_update_bytes_pinned(info, 3526606d1bf1SJosef Bacik cache->space_info, num_bytes); 3527606d1bf1SJosef Bacik cache->space_info->bytes_used -= num_bytes; 3528606d1bf1SJosef Bacik cache->space_info->disk_used -= num_bytes * factor; 3529ac2f1e63SJosef Bacik 3530ac2f1e63SJosef Bacik reclaim = should_reclaim_block_group(cache, num_bytes); 353152bb7a21SBoris Burkov 3532606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3533606d1bf1SJosef Bacik spin_unlock(&cache->space_info->lock); 3534606d1bf1SJosef Bacik 3535fe119a6eSNikolay Borisov set_extent_dirty(&trans->transaction->pinned_extents, 3536606d1bf1SJosef Bacik bytenr, bytenr + num_bytes - 1, 3537606d1bf1SJosef Bacik GFP_NOFS | __GFP_NOFAIL); 3538606d1bf1SJosef Bacik } 3539606d1bf1SJosef Bacik 3540606d1bf1SJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 3541606d1bf1SJosef Bacik if (list_empty(&cache->dirty_list)) { 3542606d1bf1SJosef Bacik list_add_tail(&cache->dirty_list, 3543606d1bf1SJosef Bacik &trans->transaction->dirty_bgs); 3544606d1bf1SJosef Bacik trans->delayed_ref_updates++; 3545606d1bf1SJosef Bacik btrfs_get_block_group(cache); 3546606d1bf1SJosef Bacik } 3547606d1bf1SJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 3548606d1bf1SJosef Bacik 3549606d1bf1SJosef Bacik /* 3550606d1bf1SJosef Bacik * No longer have used bytes in this block group, queue it for 3551606d1bf1SJosef Bacik * deletion. We do this after adding the block group to the 3552606d1bf1SJosef Bacik * dirty list to avoid races between cleaner kthread and space 3553606d1bf1SJosef Bacik * cache writeout. 3554606d1bf1SJosef Bacik */ 35556e80d4f8SDennis Zhou if (!alloc && old_val == 0) { 35566e80d4f8SDennis Zhou if (!btrfs_test_opt(info, DISCARD_ASYNC)) 3557606d1bf1SJosef Bacik btrfs_mark_bg_unused(cache); 3558ac2f1e63SJosef Bacik } else if (!alloc && reclaim) { 3559ac2f1e63SJosef Bacik btrfs_mark_bg_to_reclaim(cache); 35606e80d4f8SDennis Zhou } 3561606d1bf1SJosef Bacik 3562606d1bf1SJosef Bacik btrfs_put_block_group(cache); 3563606d1bf1SJosef Bacik total -= num_bytes; 3564606d1bf1SJosef Bacik bytenr += num_bytes; 3565606d1bf1SJosef Bacik } 3566606d1bf1SJosef Bacik 3567606d1bf1SJosef Bacik /* Modified block groups are accounted for in the delayed_refs_rsv. */ 3568606d1bf1SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 3569606d1bf1SJosef Bacik return ret; 3570606d1bf1SJosef Bacik } 3571606d1bf1SJosef Bacik 357243dd529aSDavid Sterba /* 357343dd529aSDavid Sterba * Update the block_group and space info counters. 357443dd529aSDavid Sterba * 3575606d1bf1SJosef Bacik * @cache: The cache we are manipulating 3576606d1bf1SJosef Bacik * @ram_bytes: The number of bytes of file content, and will be same to 3577606d1bf1SJosef Bacik * @num_bytes except for the compress path. 3578606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 3579606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 3580606d1bf1SJosef Bacik * 3581606d1bf1SJosef Bacik * This is called by the allocator when it reserves space. If this is a 3582606d1bf1SJosef Bacik * reservation and the block group has become read only we cannot make the 3583606d1bf1SJosef Bacik * reservation and return -EAGAIN, otherwise this function always succeeds. 3584606d1bf1SJosef Bacik */ 358532da5386SDavid Sterba int btrfs_add_reserved_bytes(struct btrfs_block_group *cache, 358652bb7a21SBoris Burkov u64 ram_bytes, u64 num_bytes, int delalloc, 358752bb7a21SBoris Burkov bool force_wrong_size_class) 3588606d1bf1SJosef Bacik { 3589606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 359052bb7a21SBoris Burkov enum btrfs_block_group_size_class size_class; 3591606d1bf1SJosef Bacik int ret = 0; 3592606d1bf1SJosef Bacik 3593606d1bf1SJosef Bacik spin_lock(&space_info->lock); 3594606d1bf1SJosef Bacik spin_lock(&cache->lock); 3595606d1bf1SJosef Bacik if (cache->ro) { 3596606d1bf1SJosef Bacik ret = -EAGAIN; 359752bb7a21SBoris Burkov goto out; 359852bb7a21SBoris Burkov } 359952bb7a21SBoris Burkov 360052bb7a21SBoris Burkov if (btrfs_is_block_group_data_only(cache)) { 360152bb7a21SBoris Burkov size_class = btrfs_calc_block_group_size_class(num_bytes); 360252bb7a21SBoris Burkov ret = btrfs_use_block_group_size_class(cache, size_class, force_wrong_size_class); 360352bb7a21SBoris Burkov if (ret) 360452bb7a21SBoris Burkov goto out; 360552bb7a21SBoris Burkov } 3606606d1bf1SJosef Bacik cache->reserved += num_bytes; 3607606d1bf1SJosef Bacik space_info->bytes_reserved += num_bytes; 3608a43c3835SJosef Bacik trace_btrfs_space_reservation(cache->fs_info, "space_info", 3609a43c3835SJosef Bacik space_info->flags, num_bytes, 1); 3610606d1bf1SJosef Bacik btrfs_space_info_update_bytes_may_use(cache->fs_info, 3611606d1bf1SJosef Bacik space_info, -ram_bytes); 3612606d1bf1SJosef Bacik if (delalloc) 3613606d1bf1SJosef Bacik cache->delalloc_bytes += num_bytes; 361499ffb43eSJosef Bacik 361599ffb43eSJosef Bacik /* 361652bb7a21SBoris Burkov * Compression can use less space than we reserved, so wake tickets if 361752bb7a21SBoris Burkov * that happens. 361899ffb43eSJosef Bacik */ 361999ffb43eSJosef Bacik if (num_bytes < ram_bytes) 362099ffb43eSJosef Bacik btrfs_try_granting_tickets(cache->fs_info, space_info); 362152bb7a21SBoris Burkov out: 3622606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3623606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 3624606d1bf1SJosef Bacik return ret; 3625606d1bf1SJosef Bacik } 3626606d1bf1SJosef Bacik 362743dd529aSDavid Sterba /* 362843dd529aSDavid Sterba * Update the block_group and space info counters. 362943dd529aSDavid Sterba * 3630606d1bf1SJosef Bacik * @cache: The cache we are manipulating 3631606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 3632606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 3633606d1bf1SJosef Bacik * 3634606d1bf1SJosef Bacik * This is called by somebody who is freeing space that was never actually used 3635606d1bf1SJosef Bacik * on disk. For example if you reserve some space for a new leaf in transaction 3636606d1bf1SJosef Bacik * A and before transaction A commits you free that leaf, you call this with 3637606d1bf1SJosef Bacik * reserve set to 0 in order to clear the reservation. 3638606d1bf1SJosef Bacik */ 363932da5386SDavid Sterba void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, 3640606d1bf1SJosef Bacik u64 num_bytes, int delalloc) 3641606d1bf1SJosef Bacik { 3642606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 3643606d1bf1SJosef Bacik 3644606d1bf1SJosef Bacik spin_lock(&space_info->lock); 3645606d1bf1SJosef Bacik spin_lock(&cache->lock); 3646606d1bf1SJosef Bacik if (cache->ro) 3647606d1bf1SJosef Bacik space_info->bytes_readonly += num_bytes; 3648606d1bf1SJosef Bacik cache->reserved -= num_bytes; 3649606d1bf1SJosef Bacik space_info->bytes_reserved -= num_bytes; 3650606d1bf1SJosef Bacik space_info->max_extent_size = 0; 3651606d1bf1SJosef Bacik 3652606d1bf1SJosef Bacik if (delalloc) 3653606d1bf1SJosef Bacik cache->delalloc_bytes -= num_bytes; 3654606d1bf1SJosef Bacik spin_unlock(&cache->lock); 36553308234aSJosef Bacik 36563308234aSJosef Bacik btrfs_try_granting_tickets(cache->fs_info, space_info); 3657606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 3658606d1bf1SJosef Bacik } 365907730d87SJosef Bacik 366007730d87SJosef Bacik static void force_metadata_allocation(struct btrfs_fs_info *info) 366107730d87SJosef Bacik { 366207730d87SJosef Bacik struct list_head *head = &info->space_info; 366307730d87SJosef Bacik struct btrfs_space_info *found; 366407730d87SJosef Bacik 366572804905SJosef Bacik list_for_each_entry(found, head, list) { 366607730d87SJosef Bacik if (found->flags & BTRFS_BLOCK_GROUP_METADATA) 366707730d87SJosef Bacik found->force_alloc = CHUNK_ALLOC_FORCE; 366807730d87SJosef Bacik } 366907730d87SJosef Bacik } 367007730d87SJosef Bacik 367107730d87SJosef Bacik static int should_alloc_chunk(struct btrfs_fs_info *fs_info, 367207730d87SJosef Bacik struct btrfs_space_info *sinfo, int force) 367307730d87SJosef Bacik { 367407730d87SJosef Bacik u64 bytes_used = btrfs_space_info_used(sinfo, false); 367507730d87SJosef Bacik u64 thresh; 367607730d87SJosef Bacik 367707730d87SJosef Bacik if (force == CHUNK_ALLOC_FORCE) 367807730d87SJosef Bacik return 1; 367907730d87SJosef Bacik 368007730d87SJosef Bacik /* 368107730d87SJosef Bacik * in limited mode, we want to have some free space up to 368207730d87SJosef Bacik * about 1% of the FS size. 368307730d87SJosef Bacik */ 368407730d87SJosef Bacik if (force == CHUNK_ALLOC_LIMITED) { 368507730d87SJosef Bacik thresh = btrfs_super_total_bytes(fs_info->super_copy); 3686428c8e03SDavid Sterba thresh = max_t(u64, SZ_64M, mult_perc(thresh, 1)); 368707730d87SJosef Bacik 368807730d87SJosef Bacik if (sinfo->total_bytes - bytes_used < thresh) 368907730d87SJosef Bacik return 1; 369007730d87SJosef Bacik } 369107730d87SJosef Bacik 3692428c8e03SDavid Sterba if (bytes_used + SZ_2M < mult_perc(sinfo->total_bytes, 80)) 369307730d87SJosef Bacik return 0; 369407730d87SJosef Bacik return 1; 369507730d87SJosef Bacik } 369607730d87SJosef Bacik 369707730d87SJosef Bacik int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) 369807730d87SJosef Bacik { 369907730d87SJosef Bacik u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); 370007730d87SJosef Bacik 370107730d87SJosef Bacik return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 370207730d87SJosef Bacik } 370307730d87SJosef Bacik 3704820c363bSNaohiro Aota static struct btrfs_block_group *do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags) 370579bd3712SFilipe Manana { 370679bd3712SFilipe Manana struct btrfs_block_group *bg; 370779bd3712SFilipe Manana int ret; 370879bd3712SFilipe Manana 370907730d87SJosef Bacik /* 371079bd3712SFilipe Manana * Check if we have enough space in the system space info because we 371179bd3712SFilipe Manana * will need to update device items in the chunk btree and insert a new 371279bd3712SFilipe Manana * chunk item in the chunk btree as well. This will allocate a new 371379bd3712SFilipe Manana * system block group if needed. 371479bd3712SFilipe Manana */ 371579bd3712SFilipe Manana check_system_chunk(trans, flags); 371679bd3712SFilipe Manana 3717f6f39f7aSNikolay Borisov bg = btrfs_create_chunk(trans, flags); 371879bd3712SFilipe Manana if (IS_ERR(bg)) { 371979bd3712SFilipe Manana ret = PTR_ERR(bg); 372079bd3712SFilipe Manana goto out; 372179bd3712SFilipe Manana } 372279bd3712SFilipe Manana 372379bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); 372479bd3712SFilipe Manana /* 372579bd3712SFilipe Manana * Normally we are not expected to fail with -ENOSPC here, since we have 372679bd3712SFilipe Manana * previously reserved space in the system space_info and allocated one 3727ecd84d54SFilipe Manana * new system chunk if necessary. However there are three exceptions: 372879bd3712SFilipe Manana * 372979bd3712SFilipe Manana * 1) We may have enough free space in the system space_info but all the 373079bd3712SFilipe Manana * existing system block groups have a profile which can not be used 373179bd3712SFilipe Manana * for extent allocation. 373279bd3712SFilipe Manana * 373379bd3712SFilipe Manana * This happens when mounting in degraded mode. For example we have a 373479bd3712SFilipe Manana * RAID1 filesystem with 2 devices, lose one device and mount the fs 373579bd3712SFilipe Manana * using the other device in degraded mode. If we then allocate a chunk, 373679bd3712SFilipe Manana * we may have enough free space in the existing system space_info, but 373779bd3712SFilipe Manana * none of the block groups can be used for extent allocation since they 373879bd3712SFilipe Manana * have a RAID1 profile, and because we are in degraded mode with a 373979bd3712SFilipe Manana * single device, we are forced to allocate a new system chunk with a 374079bd3712SFilipe Manana * SINGLE profile. Making check_system_chunk() iterate over all system 374179bd3712SFilipe Manana * block groups and check if they have a usable profile and enough space 374279bd3712SFilipe Manana * can be slow on very large filesystems, so we tolerate the -ENOSPC and 374379bd3712SFilipe Manana * try again after forcing allocation of a new system chunk. Like this 374479bd3712SFilipe Manana * we avoid paying the cost of that search in normal circumstances, when 374579bd3712SFilipe Manana * we were not mounted in degraded mode; 374679bd3712SFilipe Manana * 374779bd3712SFilipe Manana * 2) We had enough free space info the system space_info, and one suitable 374879bd3712SFilipe Manana * block group to allocate from when we called check_system_chunk() 374979bd3712SFilipe Manana * above. However right after we called it, the only system block group 375079bd3712SFilipe Manana * with enough free space got turned into RO mode by a running scrub, 375179bd3712SFilipe Manana * and in this case we have to allocate a new one and retry. We only 375279bd3712SFilipe Manana * need do this allocate and retry once, since we have a transaction 3753ecd84d54SFilipe Manana * handle and scrub uses the commit root to search for block groups; 3754ecd84d54SFilipe Manana * 3755ecd84d54SFilipe Manana * 3) We had one system block group with enough free space when we called 3756ecd84d54SFilipe Manana * check_system_chunk(), but after that, right before we tried to 3757ecd84d54SFilipe Manana * allocate the last extent buffer we needed, a discard operation came 3758ecd84d54SFilipe Manana * in and it temporarily removed the last free space entry from the 3759ecd84d54SFilipe Manana * block group (discard removes a free space entry, discards it, and 3760ecd84d54SFilipe Manana * then adds back the entry to the block group cache). 376179bd3712SFilipe Manana */ 376279bd3712SFilipe Manana if (ret == -ENOSPC) { 376379bd3712SFilipe Manana const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info); 376479bd3712SFilipe Manana struct btrfs_block_group *sys_bg; 376579bd3712SFilipe Manana 3766f6f39f7aSNikolay Borisov sys_bg = btrfs_create_chunk(trans, sys_flags); 376779bd3712SFilipe Manana if (IS_ERR(sys_bg)) { 376879bd3712SFilipe Manana ret = PTR_ERR(sys_bg); 376979bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 377079bd3712SFilipe Manana goto out; 377179bd3712SFilipe Manana } 377279bd3712SFilipe Manana 377379bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 377479bd3712SFilipe Manana if (ret) { 377579bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 377679bd3712SFilipe Manana goto out; 377779bd3712SFilipe Manana } 377879bd3712SFilipe Manana 377979bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); 378079bd3712SFilipe Manana if (ret) { 378179bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 378279bd3712SFilipe Manana goto out; 378379bd3712SFilipe Manana } 378479bd3712SFilipe Manana } else if (ret) { 378579bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 378679bd3712SFilipe Manana goto out; 378779bd3712SFilipe Manana } 378879bd3712SFilipe Manana out: 378979bd3712SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 379079bd3712SFilipe Manana 3791820c363bSNaohiro Aota if (ret) 3792820c363bSNaohiro Aota return ERR_PTR(ret); 3793820c363bSNaohiro Aota 3794820c363bSNaohiro Aota btrfs_get_block_group(bg); 3795820c363bSNaohiro Aota return bg; 379679bd3712SFilipe Manana } 379779bd3712SFilipe Manana 379879bd3712SFilipe Manana /* 379979bd3712SFilipe Manana * Chunk allocation is done in 2 phases: 380079bd3712SFilipe Manana * 380179bd3712SFilipe Manana * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for 380279bd3712SFilipe Manana * the chunk, the chunk mapping, create its block group and add the items 380379bd3712SFilipe Manana * that belong in the chunk btree to it - more specifically, we need to 380479bd3712SFilipe Manana * update device items in the chunk btree and add a new chunk item to it. 380579bd3712SFilipe Manana * 380679bd3712SFilipe Manana * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block 380779bd3712SFilipe Manana * group item to the extent btree and the device extent items to the devices 380879bd3712SFilipe Manana * btree. 380979bd3712SFilipe Manana * 381079bd3712SFilipe Manana * This is done to prevent deadlocks. For example when COWing a node from the 381179bd3712SFilipe Manana * extent btree we are holding a write lock on the node's parent and if we 381279bd3712SFilipe Manana * trigger chunk allocation and attempted to insert the new block group item 381379bd3712SFilipe Manana * in the extent btree right way, we could deadlock because the path for the 381479bd3712SFilipe Manana * insertion can include that parent node. At first glance it seems impossible 381579bd3712SFilipe Manana * to trigger chunk allocation after starting a transaction since tasks should 381679bd3712SFilipe Manana * reserve enough transaction units (metadata space), however while that is true 381779bd3712SFilipe Manana * most of the time, chunk allocation may still be triggered for several reasons: 381879bd3712SFilipe Manana * 381979bd3712SFilipe Manana * 1) When reserving metadata, we check if there is enough free space in the 382079bd3712SFilipe Manana * metadata space_info and therefore don't trigger allocation of a new chunk. 382179bd3712SFilipe Manana * However later when the task actually tries to COW an extent buffer from 382279bd3712SFilipe Manana * the extent btree or from the device btree for example, it is forced to 382379bd3712SFilipe Manana * allocate a new block group (chunk) because the only one that had enough 382479bd3712SFilipe Manana * free space was just turned to RO mode by a running scrub for example (or 382579bd3712SFilipe Manana * device replace, block group reclaim thread, etc), so we can not use it 382679bd3712SFilipe Manana * for allocating an extent and end up being forced to allocate a new one; 382779bd3712SFilipe Manana * 382879bd3712SFilipe Manana * 2) Because we only check that the metadata space_info has enough free bytes, 382979bd3712SFilipe Manana * we end up not allocating a new metadata chunk in that case. However if 383079bd3712SFilipe Manana * the filesystem was mounted in degraded mode, none of the existing block 383179bd3712SFilipe Manana * groups might be suitable for extent allocation due to their incompatible 383279bd3712SFilipe Manana * profile (for e.g. mounting a 2 devices filesystem, where all block groups 383379bd3712SFilipe Manana * use a RAID1 profile, in degraded mode using a single device). In this case 383479bd3712SFilipe Manana * when the task attempts to COW some extent buffer of the extent btree for 383579bd3712SFilipe Manana * example, it will trigger allocation of a new metadata block group with a 383679bd3712SFilipe Manana * suitable profile (SINGLE profile in the example of the degraded mount of 383779bd3712SFilipe Manana * the RAID1 filesystem); 383879bd3712SFilipe Manana * 383979bd3712SFilipe Manana * 3) The task has reserved enough transaction units / metadata space, but when 384079bd3712SFilipe Manana * it attempts to COW an extent buffer from the extent or device btree for 384179bd3712SFilipe Manana * example, it does not find any free extent in any metadata block group, 384279bd3712SFilipe Manana * therefore forced to try to allocate a new metadata block group. 384379bd3712SFilipe Manana * This is because some other task allocated all available extents in the 384479bd3712SFilipe Manana * meanwhile - this typically happens with tasks that don't reserve space 384579bd3712SFilipe Manana * properly, either intentionally or as a bug. One example where this is 384679bd3712SFilipe Manana * done intentionally is fsync, as it does not reserve any transaction units 384779bd3712SFilipe Manana * and ends up allocating a variable number of metadata extents for log 3848ecd84d54SFilipe Manana * tree extent buffers; 3849ecd84d54SFilipe Manana * 3850ecd84d54SFilipe Manana * 4) The task has reserved enough transaction units / metadata space, but right 3851ecd84d54SFilipe Manana * before it tries to allocate the last extent buffer it needs, a discard 3852ecd84d54SFilipe Manana * operation comes in and, temporarily, removes the last free space entry from 3853ecd84d54SFilipe Manana * the only metadata block group that had free space (discard starts by 3854ecd84d54SFilipe Manana * removing a free space entry from a block group, then does the discard 3855ecd84d54SFilipe Manana * operation and, once it's done, it adds back the free space entry to the 3856ecd84d54SFilipe Manana * block group). 385779bd3712SFilipe Manana * 385879bd3712SFilipe Manana * We also need this 2 phases setup when adding a device to a filesystem with 385979bd3712SFilipe Manana * a seed device - we must create new metadata and system chunks without adding 386079bd3712SFilipe Manana * any of the block group items to the chunk, extent and device btrees. If we 386179bd3712SFilipe Manana * did not do it this way, we would get ENOSPC when attempting to update those 386279bd3712SFilipe Manana * btrees, since all the chunks from the seed device are read-only. 386379bd3712SFilipe Manana * 386479bd3712SFilipe Manana * Phase 1 does the updates and insertions to the chunk btree because if we had 386579bd3712SFilipe Manana * it done in phase 2 and have a thundering herd of tasks allocating chunks in 386679bd3712SFilipe Manana * parallel, we risk having too many system chunks allocated by many tasks if 386779bd3712SFilipe Manana * many tasks reach phase 1 without the previous ones completing phase 2. In the 386879bd3712SFilipe Manana * extreme case this leads to exhaustion of the system chunk array in the 386979bd3712SFilipe Manana * superblock. This is easier to trigger if using a btree node/leaf size of 64K 387079bd3712SFilipe Manana * and with RAID filesystems (so we have more device items in the chunk btree). 387179bd3712SFilipe Manana * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of 387279bd3712SFilipe Manana * the system chunk array due to concurrent allocations") provides more details. 387379bd3712SFilipe Manana * 38742bb2e00eSFilipe Manana * Allocation of system chunks does not happen through this function. A task that 38752bb2e00eSFilipe Manana * needs to update the chunk btree (the only btree that uses system chunks), must 38762bb2e00eSFilipe Manana * preallocate chunk space by calling either check_system_chunk() or 38772bb2e00eSFilipe Manana * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or 38782bb2e00eSFilipe Manana * metadata chunk or when removing a chunk, while the later is used before doing 38792bb2e00eSFilipe Manana * a modification to the chunk btree - use cases for the later are adding, 38802bb2e00eSFilipe Manana * removing and resizing a device as well as relocation of a system chunk. 38812bb2e00eSFilipe Manana * See the comment below for more details. 388279bd3712SFilipe Manana * 388379bd3712SFilipe Manana * The reservation of system space, done through check_system_chunk(), as well 388479bd3712SFilipe Manana * as all the updates and insertions into the chunk btree must be done while 388579bd3712SFilipe Manana * holding fs_info->chunk_mutex. This is important to guarantee that while COWing 388679bd3712SFilipe Manana * an extent buffer from the chunks btree we never trigger allocation of a new 388779bd3712SFilipe Manana * system chunk, which would result in a deadlock (trying to lock twice an 388879bd3712SFilipe Manana * extent buffer of the chunk btree, first time before triggering the chunk 388979bd3712SFilipe Manana * allocation and the second time during chunk allocation while attempting to 389079bd3712SFilipe Manana * update the chunks btree). The system chunk array is also updated while holding 389179bd3712SFilipe Manana * that mutex. The same logic applies to removing chunks - we must reserve system 389279bd3712SFilipe Manana * space, update the chunk btree and the system chunk array in the superblock 389379bd3712SFilipe Manana * while holding fs_info->chunk_mutex. 389479bd3712SFilipe Manana * 389579bd3712SFilipe Manana * This function, btrfs_chunk_alloc(), belongs to phase 1. 389679bd3712SFilipe Manana * 389779bd3712SFilipe Manana * If @force is CHUNK_ALLOC_FORCE: 389807730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 389907730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 390079bd3712SFilipe Manana * If @force is NOT CHUNK_ALLOC_FORCE: 390107730d87SJosef Bacik * - return 0 if it doesn't need to allocate a new chunk, 390207730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 390307730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 390407730d87SJosef Bacik */ 390507730d87SJosef Bacik int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, 390607730d87SJosef Bacik enum btrfs_chunk_alloc_enum force) 390707730d87SJosef Bacik { 390807730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 390907730d87SJosef Bacik struct btrfs_space_info *space_info; 3910820c363bSNaohiro Aota struct btrfs_block_group *ret_bg; 391107730d87SJosef Bacik bool wait_for_alloc = false; 391207730d87SJosef Bacik bool should_alloc = false; 3913760e69c4SNaohiro Aota bool from_extent_allocation = false; 391407730d87SJosef Bacik int ret = 0; 391507730d87SJosef Bacik 3916760e69c4SNaohiro Aota if (force == CHUNK_ALLOC_FORCE_FOR_EXTENT) { 3917760e69c4SNaohiro Aota from_extent_allocation = true; 3918760e69c4SNaohiro Aota force = CHUNK_ALLOC_FORCE; 3919760e69c4SNaohiro Aota } 3920760e69c4SNaohiro Aota 392107730d87SJosef Bacik /* Don't re-enter if we're already allocating a chunk */ 392207730d87SJosef Bacik if (trans->allocating_chunk) 392307730d87SJosef Bacik return -ENOSPC; 392479bd3712SFilipe Manana /* 39252bb2e00eSFilipe Manana * Allocation of system chunks can not happen through this path, as we 39262bb2e00eSFilipe Manana * could end up in a deadlock if we are allocating a data or metadata 39272bb2e00eSFilipe Manana * chunk and there is another task modifying the chunk btree. 39282bb2e00eSFilipe Manana * 39292bb2e00eSFilipe Manana * This is because while we are holding the chunk mutex, we will attempt 39302bb2e00eSFilipe Manana * to add the new chunk item to the chunk btree or update an existing 39312bb2e00eSFilipe Manana * device item in the chunk btree, while the other task that is modifying 39322bb2e00eSFilipe Manana * the chunk btree is attempting to COW an extent buffer while holding a 39332bb2e00eSFilipe Manana * lock on it and on its parent - if the COW operation triggers a system 39342bb2e00eSFilipe Manana * chunk allocation, then we can deadlock because we are holding the 39352bb2e00eSFilipe Manana * chunk mutex and we may need to access that extent buffer or its parent 39362bb2e00eSFilipe Manana * in order to add the chunk item or update a device item. 39372bb2e00eSFilipe Manana * 39382bb2e00eSFilipe Manana * Tasks that want to modify the chunk tree should reserve system space 39392bb2e00eSFilipe Manana * before updating the chunk btree, by calling either 39402bb2e00eSFilipe Manana * btrfs_reserve_chunk_metadata() or check_system_chunk(). 39412bb2e00eSFilipe Manana * It's possible that after a task reserves the space, it still ends up 39422bb2e00eSFilipe Manana * here - this happens in the cases described above at do_chunk_alloc(). 39432bb2e00eSFilipe Manana * The task will have to either retry or fail. 394479bd3712SFilipe Manana */ 39452bb2e00eSFilipe Manana if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 394679bd3712SFilipe Manana return -ENOSPC; 394707730d87SJosef Bacik 394807730d87SJosef Bacik space_info = btrfs_find_space_info(fs_info, flags); 394907730d87SJosef Bacik ASSERT(space_info); 395007730d87SJosef Bacik 395107730d87SJosef Bacik do { 395207730d87SJosef Bacik spin_lock(&space_info->lock); 395307730d87SJosef Bacik if (force < space_info->force_alloc) 395407730d87SJosef Bacik force = space_info->force_alloc; 395507730d87SJosef Bacik should_alloc = should_alloc_chunk(fs_info, space_info, force); 395607730d87SJosef Bacik if (space_info->full) { 395707730d87SJosef Bacik /* No more free physical space */ 395807730d87SJosef Bacik if (should_alloc) 395907730d87SJosef Bacik ret = -ENOSPC; 396007730d87SJosef Bacik else 396107730d87SJosef Bacik ret = 0; 396207730d87SJosef Bacik spin_unlock(&space_info->lock); 396307730d87SJosef Bacik return ret; 396407730d87SJosef Bacik } else if (!should_alloc) { 396507730d87SJosef Bacik spin_unlock(&space_info->lock); 396607730d87SJosef Bacik return 0; 396707730d87SJosef Bacik } else if (space_info->chunk_alloc) { 396807730d87SJosef Bacik /* 396907730d87SJosef Bacik * Someone is already allocating, so we need to block 397007730d87SJosef Bacik * until this someone is finished and then loop to 397107730d87SJosef Bacik * recheck if we should continue with our allocation 397207730d87SJosef Bacik * attempt. 397307730d87SJosef Bacik */ 397407730d87SJosef Bacik wait_for_alloc = true; 39751314ca78SJosef Bacik force = CHUNK_ALLOC_NO_FORCE; 397607730d87SJosef Bacik spin_unlock(&space_info->lock); 397707730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 397807730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 397907730d87SJosef Bacik } else { 398007730d87SJosef Bacik /* Proceed with allocation */ 398107730d87SJosef Bacik space_info->chunk_alloc = 1; 398207730d87SJosef Bacik wait_for_alloc = false; 398307730d87SJosef Bacik spin_unlock(&space_info->lock); 398407730d87SJosef Bacik } 398507730d87SJosef Bacik 398607730d87SJosef Bacik cond_resched(); 398707730d87SJosef Bacik } while (wait_for_alloc); 398807730d87SJosef Bacik 398907730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 399007730d87SJosef Bacik trans->allocating_chunk = true; 399107730d87SJosef Bacik 399207730d87SJosef Bacik /* 399307730d87SJosef Bacik * If we have mixed data/metadata chunks we want to make sure we keep 399407730d87SJosef Bacik * allocating mixed chunks instead of individual chunks. 399507730d87SJosef Bacik */ 399607730d87SJosef Bacik if (btrfs_mixed_space_info(space_info)) 399707730d87SJosef Bacik flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); 399807730d87SJosef Bacik 399907730d87SJosef Bacik /* 400007730d87SJosef Bacik * if we're doing a data chunk, go ahead and make sure that 400107730d87SJosef Bacik * we keep a reasonable number of metadata chunks allocated in the 400207730d87SJosef Bacik * FS as well. 400307730d87SJosef Bacik */ 400407730d87SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { 400507730d87SJosef Bacik fs_info->data_chunk_allocations++; 400607730d87SJosef Bacik if (!(fs_info->data_chunk_allocations % 400707730d87SJosef Bacik fs_info->metadata_ratio)) 400807730d87SJosef Bacik force_metadata_allocation(fs_info); 400907730d87SJosef Bacik } 401007730d87SJosef Bacik 4011820c363bSNaohiro Aota ret_bg = do_chunk_alloc(trans, flags); 401207730d87SJosef Bacik trans->allocating_chunk = false; 401307730d87SJosef Bacik 4014760e69c4SNaohiro Aota if (IS_ERR(ret_bg)) { 4015820c363bSNaohiro Aota ret = PTR_ERR(ret_bg); 4016760e69c4SNaohiro Aota } else if (from_extent_allocation) { 4017760e69c4SNaohiro Aota /* 4018760e69c4SNaohiro Aota * New block group is likely to be used soon. Try to activate 4019760e69c4SNaohiro Aota * it now. Failure is OK for now. 4020760e69c4SNaohiro Aota */ 4021760e69c4SNaohiro Aota btrfs_zone_activate(ret_bg); 4022760e69c4SNaohiro Aota } 4023760e69c4SNaohiro Aota 4024760e69c4SNaohiro Aota if (!ret) 4025820c363bSNaohiro Aota btrfs_put_block_group(ret_bg); 4026820c363bSNaohiro Aota 402707730d87SJosef Bacik spin_lock(&space_info->lock); 402807730d87SJosef Bacik if (ret < 0) { 402907730d87SJosef Bacik if (ret == -ENOSPC) 403007730d87SJosef Bacik space_info->full = 1; 403107730d87SJosef Bacik else 403207730d87SJosef Bacik goto out; 403307730d87SJosef Bacik } else { 403407730d87SJosef Bacik ret = 1; 403507730d87SJosef Bacik space_info->max_extent_size = 0; 403607730d87SJosef Bacik } 403707730d87SJosef Bacik 403807730d87SJosef Bacik space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; 403907730d87SJosef Bacik out: 404007730d87SJosef Bacik space_info->chunk_alloc = 0; 404107730d87SJosef Bacik spin_unlock(&space_info->lock); 404207730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 404307730d87SJosef Bacik 404407730d87SJosef Bacik return ret; 404507730d87SJosef Bacik } 404607730d87SJosef Bacik 404707730d87SJosef Bacik static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) 404807730d87SJosef Bacik { 404907730d87SJosef Bacik u64 num_dev; 405007730d87SJosef Bacik 405107730d87SJosef Bacik num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; 405207730d87SJosef Bacik if (!num_dev) 405307730d87SJosef Bacik num_dev = fs_info->fs_devices->rw_devices; 405407730d87SJosef Bacik 405507730d87SJosef Bacik return num_dev; 405607730d87SJosef Bacik } 405707730d87SJosef Bacik 40582bb2e00eSFilipe Manana static void reserve_chunk_space(struct btrfs_trans_handle *trans, 40592bb2e00eSFilipe Manana u64 bytes, 40602bb2e00eSFilipe Manana u64 type) 406107730d87SJosef Bacik { 406207730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 406307730d87SJosef Bacik struct btrfs_space_info *info; 406407730d87SJosef Bacik u64 left; 406507730d87SJosef Bacik int ret = 0; 406607730d87SJosef Bacik 406707730d87SJosef Bacik /* 406807730d87SJosef Bacik * Needed because we can end up allocating a system chunk and for an 406907730d87SJosef Bacik * atomic and race free space reservation in the chunk block reserve. 407007730d87SJosef Bacik */ 407107730d87SJosef Bacik lockdep_assert_held(&fs_info->chunk_mutex); 407207730d87SJosef Bacik 407307730d87SJosef Bacik info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); 407407730d87SJosef Bacik spin_lock(&info->lock); 407507730d87SJosef Bacik left = info->total_bytes - btrfs_space_info_used(info, true); 407607730d87SJosef Bacik spin_unlock(&info->lock); 407707730d87SJosef Bacik 40782bb2e00eSFilipe Manana if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { 407907730d87SJosef Bacik btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", 40802bb2e00eSFilipe Manana left, bytes, type); 408107730d87SJosef Bacik btrfs_dump_space_info(fs_info, info, 0, 0); 408207730d87SJosef Bacik } 408307730d87SJosef Bacik 40842bb2e00eSFilipe Manana if (left < bytes) { 408507730d87SJosef Bacik u64 flags = btrfs_system_alloc_profile(fs_info); 408679bd3712SFilipe Manana struct btrfs_block_group *bg; 408707730d87SJosef Bacik 408807730d87SJosef Bacik /* 408907730d87SJosef Bacik * Ignore failure to create system chunk. We might end up not 409007730d87SJosef Bacik * needing it, as we might not need to COW all nodes/leafs from 409107730d87SJosef Bacik * the paths we visit in the chunk tree (they were already COWed 409207730d87SJosef Bacik * or created in the current transaction for example). 409307730d87SJosef Bacik */ 4094f6f39f7aSNikolay Borisov bg = btrfs_create_chunk(trans, flags); 409579bd3712SFilipe Manana if (IS_ERR(bg)) { 409679bd3712SFilipe Manana ret = PTR_ERR(bg); 40972bb2e00eSFilipe Manana } else { 409879bd3712SFilipe Manana /* 4099b6a98021SNaohiro Aota * We have a new chunk. We also need to activate it for 4100b6a98021SNaohiro Aota * zoned filesystem. 4101b6a98021SNaohiro Aota */ 4102b6a98021SNaohiro Aota ret = btrfs_zoned_activate_one_bg(fs_info, info, true); 4103b6a98021SNaohiro Aota if (ret < 0) 4104b6a98021SNaohiro Aota return; 4105b6a98021SNaohiro Aota 4106b6a98021SNaohiro Aota /* 410779bd3712SFilipe Manana * If we fail to add the chunk item here, we end up 410879bd3712SFilipe Manana * trying again at phase 2 of chunk allocation, at 410979bd3712SFilipe Manana * btrfs_create_pending_block_groups(). So ignore 41102bb2e00eSFilipe Manana * any error here. An ENOSPC here could happen, due to 41112bb2e00eSFilipe Manana * the cases described at do_chunk_alloc() - the system 41122bb2e00eSFilipe Manana * block group we just created was just turned into RO 41132bb2e00eSFilipe Manana * mode by a scrub for example, or a running discard 41142bb2e00eSFilipe Manana * temporarily removed its free space entries, etc. 411579bd3712SFilipe Manana */ 411679bd3712SFilipe Manana btrfs_chunk_alloc_add_chunk_item(trans, bg); 411779bd3712SFilipe Manana } 411807730d87SJosef Bacik } 411907730d87SJosef Bacik 412007730d87SJosef Bacik if (!ret) { 41219270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, 412207730d87SJosef Bacik &fs_info->chunk_block_rsv, 41232bb2e00eSFilipe Manana bytes, BTRFS_RESERVE_NO_FLUSH); 41241cb3db1cSFilipe Manana if (!ret) 41252bb2e00eSFilipe Manana trans->chunk_bytes_reserved += bytes; 412607730d87SJosef Bacik } 412707730d87SJosef Bacik } 412807730d87SJosef Bacik 41292bb2e00eSFilipe Manana /* 41302bb2e00eSFilipe Manana * Reserve space in the system space for allocating or removing a chunk. 41312bb2e00eSFilipe Manana * The caller must be holding fs_info->chunk_mutex. 41322bb2e00eSFilipe Manana */ 41332bb2e00eSFilipe Manana void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) 41342bb2e00eSFilipe Manana { 41352bb2e00eSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 41362bb2e00eSFilipe Manana const u64 num_devs = get_profile_num_devs(fs_info, type); 41372bb2e00eSFilipe Manana u64 bytes; 41382bb2e00eSFilipe Manana 41392bb2e00eSFilipe Manana /* num_devs device items to update and 1 chunk item to add or remove. */ 41402bb2e00eSFilipe Manana bytes = btrfs_calc_metadata_size(fs_info, num_devs) + 41412bb2e00eSFilipe Manana btrfs_calc_insert_metadata_size(fs_info, 1); 41422bb2e00eSFilipe Manana 41432bb2e00eSFilipe Manana reserve_chunk_space(trans, bytes, type); 41442bb2e00eSFilipe Manana } 41452bb2e00eSFilipe Manana 41462bb2e00eSFilipe Manana /* 41472bb2e00eSFilipe Manana * Reserve space in the system space, if needed, for doing a modification to the 41482bb2e00eSFilipe Manana * chunk btree. 41492bb2e00eSFilipe Manana * 41502bb2e00eSFilipe Manana * @trans: A transaction handle. 41512bb2e00eSFilipe Manana * @is_item_insertion: Indicate if the modification is for inserting a new item 41522bb2e00eSFilipe Manana * in the chunk btree or if it's for the deletion or update 41532bb2e00eSFilipe Manana * of an existing item. 41542bb2e00eSFilipe Manana * 41552bb2e00eSFilipe Manana * This is used in a context where we need to update the chunk btree outside 41562bb2e00eSFilipe Manana * block group allocation and removal, to avoid a deadlock with a concurrent 41572bb2e00eSFilipe Manana * task that is allocating a metadata or data block group and therefore needs to 41582bb2e00eSFilipe Manana * update the chunk btree while holding the chunk mutex. After the update to the 41592bb2e00eSFilipe Manana * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called. 41602bb2e00eSFilipe Manana * 41612bb2e00eSFilipe Manana */ 41622bb2e00eSFilipe Manana void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans, 41632bb2e00eSFilipe Manana bool is_item_insertion) 41642bb2e00eSFilipe Manana { 41652bb2e00eSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 41662bb2e00eSFilipe Manana u64 bytes; 41672bb2e00eSFilipe Manana 41682bb2e00eSFilipe Manana if (is_item_insertion) 41692bb2e00eSFilipe Manana bytes = btrfs_calc_insert_metadata_size(fs_info, 1); 41702bb2e00eSFilipe Manana else 41712bb2e00eSFilipe Manana bytes = btrfs_calc_metadata_size(fs_info, 1); 41722bb2e00eSFilipe Manana 41732bb2e00eSFilipe Manana mutex_lock(&fs_info->chunk_mutex); 41742bb2e00eSFilipe Manana reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM); 41752bb2e00eSFilipe Manana mutex_unlock(&fs_info->chunk_mutex); 41762bb2e00eSFilipe Manana } 41772bb2e00eSFilipe Manana 41783e43c279SJosef Bacik void btrfs_put_block_group_cache(struct btrfs_fs_info *info) 41793e43c279SJosef Bacik { 418032da5386SDavid Sterba struct btrfs_block_group *block_group; 41813e43c279SJosef Bacik 418250c31eaaSJosef Bacik block_group = btrfs_lookup_first_block_group(info, 0); 41833e43c279SJosef Bacik while (block_group) { 41843e43c279SJosef Bacik btrfs_wait_block_group_cache_done(block_group); 41853e43c279SJosef Bacik spin_lock(&block_group->lock); 418650c31eaaSJosef Bacik if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, 418750c31eaaSJosef Bacik &block_group->runtime_flags)) { 418850c31eaaSJosef Bacik struct inode *inode = block_group->inode; 41893e43c279SJosef Bacik 41903e43c279SJosef Bacik block_group->inode = NULL; 41913e43c279SJosef Bacik spin_unlock(&block_group->lock); 419250c31eaaSJosef Bacik 41933e43c279SJosef Bacik ASSERT(block_group->io_ctl.inode == NULL); 41943e43c279SJosef Bacik iput(inode); 419550c31eaaSJosef Bacik } else { 419650c31eaaSJosef Bacik spin_unlock(&block_group->lock); 419750c31eaaSJosef Bacik } 419850c31eaaSJosef Bacik block_group = btrfs_next_block_group(block_group); 41993e43c279SJosef Bacik } 42003e43c279SJosef Bacik } 42013e43c279SJosef Bacik 42023e43c279SJosef Bacik /* 42033e43c279SJosef Bacik * Must be called only after stopping all workers, since we could have block 42043e43c279SJosef Bacik * group caching kthreads running, and therefore they could race with us if we 42053e43c279SJosef Bacik * freed the block groups before stopping them. 42063e43c279SJosef Bacik */ 42073e43c279SJosef Bacik int btrfs_free_block_groups(struct btrfs_fs_info *info) 42083e43c279SJosef Bacik { 420932da5386SDavid Sterba struct btrfs_block_group *block_group; 42103e43c279SJosef Bacik struct btrfs_space_info *space_info; 42113e43c279SJosef Bacik struct btrfs_caching_control *caching_ctl; 42123e43c279SJosef Bacik struct rb_node *n; 42133e43c279SJosef Bacik 421416b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 42153e43c279SJosef Bacik while (!list_empty(&info->caching_block_groups)) { 42163e43c279SJosef Bacik caching_ctl = list_entry(info->caching_block_groups.next, 42173e43c279SJosef Bacik struct btrfs_caching_control, list); 42183e43c279SJosef Bacik list_del(&caching_ctl->list); 42193e43c279SJosef Bacik btrfs_put_caching_control(caching_ctl); 42203e43c279SJosef Bacik } 422116b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42223e43c279SJosef Bacik 42233e43c279SJosef Bacik spin_lock(&info->unused_bgs_lock); 42243e43c279SJosef Bacik while (!list_empty(&info->unused_bgs)) { 42253e43c279SJosef Bacik block_group = list_first_entry(&info->unused_bgs, 422632da5386SDavid Sterba struct btrfs_block_group, 42273e43c279SJosef Bacik bg_list); 42283e43c279SJosef Bacik list_del_init(&block_group->bg_list); 42293e43c279SJosef Bacik btrfs_put_block_group(block_group); 42303e43c279SJosef Bacik } 42313e43c279SJosef Bacik 423218bb8bbfSJohannes Thumshirn while (!list_empty(&info->reclaim_bgs)) { 423318bb8bbfSJohannes Thumshirn block_group = list_first_entry(&info->reclaim_bgs, 423418bb8bbfSJohannes Thumshirn struct btrfs_block_group, 423518bb8bbfSJohannes Thumshirn bg_list); 423618bb8bbfSJohannes Thumshirn list_del_init(&block_group->bg_list); 423718bb8bbfSJohannes Thumshirn btrfs_put_block_group(block_group); 423818bb8bbfSJohannes Thumshirn } 423918bb8bbfSJohannes Thumshirn spin_unlock(&info->unused_bgs_lock); 424018bb8bbfSJohannes Thumshirn 4241afba2bc0SNaohiro Aota spin_lock(&info->zone_active_bgs_lock); 4242afba2bc0SNaohiro Aota while (!list_empty(&info->zone_active_bgs)) { 4243afba2bc0SNaohiro Aota block_group = list_first_entry(&info->zone_active_bgs, 4244afba2bc0SNaohiro Aota struct btrfs_block_group, 4245afba2bc0SNaohiro Aota active_bg_list); 4246afba2bc0SNaohiro Aota list_del_init(&block_group->active_bg_list); 4247afba2bc0SNaohiro Aota btrfs_put_block_group(block_group); 4248afba2bc0SNaohiro Aota } 4249afba2bc0SNaohiro Aota spin_unlock(&info->zone_active_bgs_lock); 4250afba2bc0SNaohiro Aota 425116b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 425208dddb29SFilipe Manana while ((n = rb_last(&info->block_group_cache_tree.rb_root)) != NULL) { 425332da5386SDavid Sterba block_group = rb_entry(n, struct btrfs_block_group, 42543e43c279SJosef Bacik cache_node); 425508dddb29SFilipe Manana rb_erase_cached(&block_group->cache_node, 42563e43c279SJosef Bacik &info->block_group_cache_tree); 42573e43c279SJosef Bacik RB_CLEAR_NODE(&block_group->cache_node); 425816b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42593e43c279SJosef Bacik 42603e43c279SJosef Bacik down_write(&block_group->space_info->groups_sem); 42613e43c279SJosef Bacik list_del(&block_group->list); 42623e43c279SJosef Bacik up_write(&block_group->space_info->groups_sem); 42633e43c279SJosef Bacik 42643e43c279SJosef Bacik /* 42653e43c279SJosef Bacik * We haven't cached this block group, which means we could 42663e43c279SJosef Bacik * possibly have excluded extents on this block group. 42673e43c279SJosef Bacik */ 42683e43c279SJosef Bacik if (block_group->cached == BTRFS_CACHE_NO || 42693e43c279SJosef Bacik block_group->cached == BTRFS_CACHE_ERROR) 42703e43c279SJosef Bacik btrfs_free_excluded_extents(block_group); 42713e43c279SJosef Bacik 42723e43c279SJosef Bacik btrfs_remove_free_space_cache(block_group); 42733e43c279SJosef Bacik ASSERT(block_group->cached != BTRFS_CACHE_STARTED); 42743e43c279SJosef Bacik ASSERT(list_empty(&block_group->dirty_list)); 42753e43c279SJosef Bacik ASSERT(list_empty(&block_group->io_list)); 42763e43c279SJosef Bacik ASSERT(list_empty(&block_group->bg_list)); 427748aaeebeSJosef Bacik ASSERT(refcount_read(&block_group->refs) == 1); 4278195a49eaSFilipe Manana ASSERT(block_group->swap_extents == 0); 42793e43c279SJosef Bacik btrfs_put_block_group(block_group); 42803e43c279SJosef Bacik 428116b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 42823e43c279SJosef Bacik } 428316b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42843e43c279SJosef Bacik 42853e43c279SJosef Bacik btrfs_release_global_block_rsv(info); 42863e43c279SJosef Bacik 42873e43c279SJosef Bacik while (!list_empty(&info->space_info)) { 42883e43c279SJosef Bacik space_info = list_entry(info->space_info.next, 42893e43c279SJosef Bacik struct btrfs_space_info, 42903e43c279SJosef Bacik list); 42913e43c279SJosef Bacik 42923e43c279SJosef Bacik /* 42933e43c279SJosef Bacik * Do not hide this behind enospc_debug, this is actually 42943e43c279SJosef Bacik * important and indicates a real bug if this happens. 42953e43c279SJosef Bacik */ 42963e43c279SJosef Bacik if (WARN_ON(space_info->bytes_pinned > 0 || 42973e43c279SJosef Bacik space_info->bytes_may_use > 0)) 42983e43c279SJosef Bacik btrfs_dump_space_info(info, space_info, 0, 0); 429940cdc509SFilipe Manana 430040cdc509SFilipe Manana /* 430140cdc509SFilipe Manana * If there was a failure to cleanup a log tree, very likely due 430240cdc509SFilipe Manana * to an IO failure on a writeback attempt of one or more of its 430340cdc509SFilipe Manana * extent buffers, we could not do proper (and cheap) unaccounting 430440cdc509SFilipe Manana * of their reserved space, so don't warn on bytes_reserved > 0 in 430540cdc509SFilipe Manana * that case. 430640cdc509SFilipe Manana */ 430740cdc509SFilipe Manana if (!(space_info->flags & BTRFS_BLOCK_GROUP_METADATA) || 430840cdc509SFilipe Manana !BTRFS_FS_LOG_CLEANUP_ERROR(info)) { 430940cdc509SFilipe Manana if (WARN_ON(space_info->bytes_reserved > 0)) 431040cdc509SFilipe Manana btrfs_dump_space_info(info, space_info, 0, 0); 431140cdc509SFilipe Manana } 431240cdc509SFilipe Manana 4313d611add4SFilipe Manana WARN_ON(space_info->reclaim_size > 0); 43143e43c279SJosef Bacik list_del(&space_info->list); 43153e43c279SJosef Bacik btrfs_sysfs_remove_space_info(space_info); 43163e43c279SJosef Bacik } 43173e43c279SJosef Bacik return 0; 43183e43c279SJosef Bacik } 4319684b752bSFilipe Manana 4320684b752bSFilipe Manana void btrfs_freeze_block_group(struct btrfs_block_group *cache) 4321684b752bSFilipe Manana { 4322684b752bSFilipe Manana atomic_inc(&cache->frozen); 4323684b752bSFilipe Manana } 4324684b752bSFilipe Manana 4325684b752bSFilipe Manana void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group) 4326684b752bSFilipe Manana { 4327684b752bSFilipe Manana struct btrfs_fs_info *fs_info = block_group->fs_info; 4328684b752bSFilipe Manana struct extent_map_tree *em_tree; 4329684b752bSFilipe Manana struct extent_map *em; 4330684b752bSFilipe Manana bool cleanup; 4331684b752bSFilipe Manana 4332684b752bSFilipe Manana spin_lock(&block_group->lock); 4333684b752bSFilipe Manana cleanup = (atomic_dec_and_test(&block_group->frozen) && 43343349b57fSJosef Bacik test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)); 4335684b752bSFilipe Manana spin_unlock(&block_group->lock); 4336684b752bSFilipe Manana 4337684b752bSFilipe Manana if (cleanup) { 4338684b752bSFilipe Manana em_tree = &fs_info->mapping_tree; 4339684b752bSFilipe Manana write_lock(&em_tree->lock); 4340684b752bSFilipe Manana em = lookup_extent_mapping(em_tree, block_group->start, 4341684b752bSFilipe Manana 1); 4342684b752bSFilipe Manana BUG_ON(!em); /* logic error, can't happen */ 4343684b752bSFilipe Manana remove_extent_mapping(em_tree, em); 4344684b752bSFilipe Manana write_unlock(&em_tree->lock); 4345684b752bSFilipe Manana 4346684b752bSFilipe Manana /* once for us and once for the tree */ 4347684b752bSFilipe Manana free_extent_map(em); 4348684b752bSFilipe Manana free_extent_map(em); 4349684b752bSFilipe Manana 4350684b752bSFilipe Manana /* 4351684b752bSFilipe Manana * We may have left one free space entry and other possible 4352684b752bSFilipe Manana * tasks trimming this block group have left 1 entry each one. 4353684b752bSFilipe Manana * Free them if any. 4354684b752bSFilipe Manana */ 4355fc80f7acSJosef Bacik btrfs_remove_free_space_cache(block_group); 4356684b752bSFilipe Manana } 4357684b752bSFilipe Manana } 4358195a49eaSFilipe Manana 4359195a49eaSFilipe Manana bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg) 4360195a49eaSFilipe Manana { 4361195a49eaSFilipe Manana bool ret = true; 4362195a49eaSFilipe Manana 4363195a49eaSFilipe Manana spin_lock(&bg->lock); 4364195a49eaSFilipe Manana if (bg->ro) 4365195a49eaSFilipe Manana ret = false; 4366195a49eaSFilipe Manana else 4367195a49eaSFilipe Manana bg->swap_extents++; 4368195a49eaSFilipe Manana spin_unlock(&bg->lock); 4369195a49eaSFilipe Manana 4370195a49eaSFilipe Manana return ret; 4371195a49eaSFilipe Manana } 4372195a49eaSFilipe Manana 4373195a49eaSFilipe Manana void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount) 4374195a49eaSFilipe Manana { 4375195a49eaSFilipe Manana spin_lock(&bg->lock); 4376195a49eaSFilipe Manana ASSERT(!bg->ro); 4377195a49eaSFilipe Manana ASSERT(bg->swap_extents >= amount); 4378195a49eaSFilipe Manana bg->swap_extents -= amount; 4379195a49eaSFilipe Manana spin_unlock(&bg->lock); 4380195a49eaSFilipe Manana } 438152bb7a21SBoris Burkov 438252bb7a21SBoris Burkov enum btrfs_block_group_size_class btrfs_calc_block_group_size_class(u64 size) 438352bb7a21SBoris Burkov { 438452bb7a21SBoris Burkov if (size <= SZ_128K) 438552bb7a21SBoris Burkov return BTRFS_BG_SZ_SMALL; 438652bb7a21SBoris Burkov if (size <= SZ_8M) 438752bb7a21SBoris Burkov return BTRFS_BG_SZ_MEDIUM; 438852bb7a21SBoris Burkov return BTRFS_BG_SZ_LARGE; 438952bb7a21SBoris Burkov } 439052bb7a21SBoris Burkov 439152bb7a21SBoris Burkov /* 439252bb7a21SBoris Burkov * Handle a block group allocating an extent in a size class 439352bb7a21SBoris Burkov * 439452bb7a21SBoris Burkov * @bg: The block group we allocated in. 439552bb7a21SBoris Burkov * @size_class: The size class of the allocation. 439652bb7a21SBoris Burkov * @force_wrong_size_class: Whether we are desperate enough to allow 439752bb7a21SBoris Burkov * mismatched size classes. 439852bb7a21SBoris Burkov * 439952bb7a21SBoris Burkov * Returns: 0 if the size class was valid for this block_group, -EAGAIN in the 440052bb7a21SBoris Burkov * case of a race that leads to the wrong size class without 440152bb7a21SBoris Burkov * force_wrong_size_class set. 440252bb7a21SBoris Burkov * 440352bb7a21SBoris Burkov * find_free_extent will skip block groups with a mismatched size class until 440452bb7a21SBoris Burkov * it really needs to avoid ENOSPC. In that case it will set 440552bb7a21SBoris Burkov * force_wrong_size_class. However, if a block group is newly allocated and 440652bb7a21SBoris Burkov * doesn't yet have a size class, then it is possible for two allocations of 440752bb7a21SBoris Burkov * different sizes to race and both try to use it. The loser is caught here and 440852bb7a21SBoris Burkov * has to retry. 440952bb7a21SBoris Burkov */ 441052bb7a21SBoris Burkov int btrfs_use_block_group_size_class(struct btrfs_block_group *bg, 441152bb7a21SBoris Burkov enum btrfs_block_group_size_class size_class, 441252bb7a21SBoris Burkov bool force_wrong_size_class) 441352bb7a21SBoris Burkov { 441452bb7a21SBoris Burkov ASSERT(size_class != BTRFS_BG_SZ_NONE); 441552bb7a21SBoris Burkov 441652bb7a21SBoris Burkov /* The new allocation is in the right size class, do nothing */ 441752bb7a21SBoris Burkov if (bg->size_class == size_class) 441852bb7a21SBoris Burkov return 0; 441952bb7a21SBoris Burkov /* 442052bb7a21SBoris Burkov * The new allocation is in a mismatched size class. 442152bb7a21SBoris Burkov * This means one of two things: 442252bb7a21SBoris Burkov * 442352bb7a21SBoris Burkov * 1. Two tasks in find_free_extent for different size_classes raced 442452bb7a21SBoris Burkov * and hit the same empty block_group. Make the loser try again. 442552bb7a21SBoris Burkov * 2. A call to find_free_extent got desperate enough to set 442652bb7a21SBoris Burkov * 'force_wrong_slab'. Don't change the size_class, but allow the 442752bb7a21SBoris Burkov * allocation. 442852bb7a21SBoris Burkov */ 442952bb7a21SBoris Burkov if (bg->size_class != BTRFS_BG_SZ_NONE) { 443052bb7a21SBoris Burkov if (force_wrong_size_class) 443152bb7a21SBoris Burkov return 0; 443252bb7a21SBoris Burkov return -EAGAIN; 443352bb7a21SBoris Burkov } 443452bb7a21SBoris Burkov /* 443552bb7a21SBoris Burkov * The happy new block group case: the new allocation is the first 443652bb7a21SBoris Burkov * one in the block_group so we set size_class. 443752bb7a21SBoris Burkov */ 443852bb7a21SBoris Burkov bg->size_class = size_class; 443952bb7a21SBoris Burkov 444052bb7a21SBoris Burkov return 0; 444152bb7a21SBoris Burkov } 4442