12e405ad8SJosef Bacik // SPDX-License-Identifier: GPL-2.0 22e405ad8SJosef Bacik 3784352feSDavid Sterba #include "misc.h" 42e405ad8SJosef Bacik #include "ctree.h" 52e405ad8SJosef Bacik #include "block-group.h" 63eeb3226SJosef Bacik #include "space-info.h" 79f21246dSJosef Bacik #include "disk-io.h" 89f21246dSJosef Bacik #include "free-space-cache.h" 99f21246dSJosef Bacik #include "free-space-tree.h" 10e3e0520bSJosef Bacik #include "disk-io.h" 11e3e0520bSJosef Bacik #include "volumes.h" 12e3e0520bSJosef Bacik #include "transaction.h" 13e3e0520bSJosef Bacik #include "ref-verify.h" 144358d963SJosef Bacik #include "sysfs.h" 154358d963SJosef Bacik #include "tree-log.h" 1677745c05SJosef Bacik #include "delalloc-space.h" 17b0643e59SDennis Zhou #include "discard.h" 1896a14336SNikolay Borisov #include "raid56.h" 192e405ad8SJosef Bacik 20878d7b67SJosef Bacik /* 21878d7b67SJosef Bacik * Return target flags in extended format or 0 if restripe for this chunk_type 22878d7b67SJosef Bacik * is not in progress 23878d7b67SJosef Bacik * 24878d7b67SJosef Bacik * Should be called with balance_lock held 25878d7b67SJosef Bacik */ 26e11c0406SJosef Bacik static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) 27878d7b67SJosef Bacik { 28878d7b67SJosef Bacik struct btrfs_balance_control *bctl = fs_info->balance_ctl; 29878d7b67SJosef Bacik u64 target = 0; 30878d7b67SJosef Bacik 31878d7b67SJosef Bacik if (!bctl) 32878d7b67SJosef Bacik return 0; 33878d7b67SJosef Bacik 34878d7b67SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA && 35878d7b67SJosef Bacik bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { 36878d7b67SJosef Bacik target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; 37878d7b67SJosef Bacik } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && 38878d7b67SJosef Bacik bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { 39878d7b67SJosef Bacik target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; 40878d7b67SJosef Bacik } else if (flags & BTRFS_BLOCK_GROUP_METADATA && 41878d7b67SJosef Bacik bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { 42878d7b67SJosef Bacik target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; 43878d7b67SJosef Bacik } 44878d7b67SJosef Bacik 45878d7b67SJosef Bacik return target; 46878d7b67SJosef Bacik } 47878d7b67SJosef Bacik 48878d7b67SJosef Bacik /* 49878d7b67SJosef Bacik * @flags: available profiles in extended format (see ctree.h) 50878d7b67SJosef Bacik * 51878d7b67SJosef Bacik * Return reduced profile in chunk format. If profile changing is in progress 52878d7b67SJosef Bacik * (either running or paused) picks the target profile (if it's already 53878d7b67SJosef Bacik * available), otherwise falls back to plain reducing. 54878d7b67SJosef Bacik */ 55878d7b67SJosef Bacik static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) 56878d7b67SJosef Bacik { 57878d7b67SJosef Bacik u64 num_devices = fs_info->fs_devices->rw_devices; 58878d7b67SJosef Bacik u64 target; 59878d7b67SJosef Bacik u64 raid_type; 60878d7b67SJosef Bacik u64 allowed = 0; 61878d7b67SJosef Bacik 62878d7b67SJosef Bacik /* 63878d7b67SJosef Bacik * See if restripe for this chunk_type is in progress, if so try to 64878d7b67SJosef Bacik * reduce to the target profile 65878d7b67SJosef Bacik */ 66878d7b67SJosef Bacik spin_lock(&fs_info->balance_lock); 67e11c0406SJosef Bacik target = get_restripe_target(fs_info, flags); 68878d7b67SJosef Bacik if (target) { 69878d7b67SJosef Bacik /* Pick target profile only if it's already available */ 70878d7b67SJosef Bacik if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { 71878d7b67SJosef Bacik spin_unlock(&fs_info->balance_lock); 72878d7b67SJosef Bacik return extended_to_chunk(target); 73878d7b67SJosef Bacik } 74878d7b67SJosef Bacik } 75878d7b67SJosef Bacik spin_unlock(&fs_info->balance_lock); 76878d7b67SJosef Bacik 77878d7b67SJosef Bacik /* First, mask out the RAID levels which aren't possible */ 78878d7b67SJosef Bacik for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 79878d7b67SJosef Bacik if (num_devices >= btrfs_raid_array[raid_type].devs_min) 80878d7b67SJosef Bacik allowed |= btrfs_raid_array[raid_type].bg_flag; 81878d7b67SJosef Bacik } 82878d7b67SJosef Bacik allowed &= flags; 83878d7b67SJosef Bacik 84878d7b67SJosef Bacik if (allowed & BTRFS_BLOCK_GROUP_RAID6) 85878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID6; 86878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID5) 87878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID5; 88878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID10) 89878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID10; 90878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID1) 91878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID1; 92878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID0) 93878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID0; 94878d7b67SJosef Bacik 95878d7b67SJosef Bacik flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; 96878d7b67SJosef Bacik 97878d7b67SJosef Bacik return extended_to_chunk(flags | allowed); 98878d7b67SJosef Bacik } 99878d7b67SJosef Bacik 100ef0a82daSJohannes Thumshirn u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) 101878d7b67SJosef Bacik { 102878d7b67SJosef Bacik unsigned seq; 103878d7b67SJosef Bacik u64 flags; 104878d7b67SJosef Bacik 105878d7b67SJosef Bacik do { 106878d7b67SJosef Bacik flags = orig_flags; 107878d7b67SJosef Bacik seq = read_seqbegin(&fs_info->profiles_lock); 108878d7b67SJosef Bacik 109878d7b67SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 110878d7b67SJosef Bacik flags |= fs_info->avail_data_alloc_bits; 111878d7b67SJosef Bacik else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 112878d7b67SJosef Bacik flags |= fs_info->avail_system_alloc_bits; 113878d7b67SJosef Bacik else if (flags & BTRFS_BLOCK_GROUP_METADATA) 114878d7b67SJosef Bacik flags |= fs_info->avail_metadata_alloc_bits; 115878d7b67SJosef Bacik } while (read_seqretry(&fs_info->profiles_lock, seq)); 116878d7b67SJosef Bacik 117878d7b67SJosef Bacik return btrfs_reduce_alloc_profile(fs_info, flags); 118878d7b67SJosef Bacik } 119878d7b67SJosef Bacik 12032da5386SDavid Sterba void btrfs_get_block_group(struct btrfs_block_group *cache) 1213cad1284SJosef Bacik { 1223cad1284SJosef Bacik atomic_inc(&cache->count); 1233cad1284SJosef Bacik } 1243cad1284SJosef Bacik 12532da5386SDavid Sterba void btrfs_put_block_group(struct btrfs_block_group *cache) 1263cad1284SJosef Bacik { 1273cad1284SJosef Bacik if (atomic_dec_and_test(&cache->count)) { 1283cad1284SJosef Bacik WARN_ON(cache->pinned > 0); 1293cad1284SJosef Bacik WARN_ON(cache->reserved > 0); 1303cad1284SJosef Bacik 1313cad1284SJosef Bacik /* 132b0643e59SDennis Zhou * A block_group shouldn't be on the discard_list anymore. 133b0643e59SDennis Zhou * Remove the block_group from the discard_list to prevent us 134b0643e59SDennis Zhou * from causing a panic due to NULL pointer dereference. 135b0643e59SDennis Zhou */ 136b0643e59SDennis Zhou if (WARN_ON(!list_empty(&cache->discard_list))) 137b0643e59SDennis Zhou btrfs_discard_cancel_work(&cache->fs_info->discard_ctl, 138b0643e59SDennis Zhou cache); 139b0643e59SDennis Zhou 140b0643e59SDennis Zhou /* 1413cad1284SJosef Bacik * If not empty, someone is still holding mutex of 1423cad1284SJosef Bacik * full_stripe_lock, which can only be released by caller. 1433cad1284SJosef Bacik * And it will definitely cause use-after-free when caller 1443cad1284SJosef Bacik * tries to release full stripe lock. 1453cad1284SJosef Bacik * 1463cad1284SJosef Bacik * No better way to resolve, but only to warn. 1473cad1284SJosef Bacik */ 1483cad1284SJosef Bacik WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); 1493cad1284SJosef Bacik kfree(cache->free_space_ctl); 1503cad1284SJosef Bacik kfree(cache); 1513cad1284SJosef Bacik } 1523cad1284SJosef Bacik } 1533cad1284SJosef Bacik 1542e405ad8SJosef Bacik /* 1554358d963SJosef Bacik * This adds the block group to the fs_info rb tree for the block group cache 1564358d963SJosef Bacik */ 1574358d963SJosef Bacik static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, 15832da5386SDavid Sterba struct btrfs_block_group *block_group) 1594358d963SJosef Bacik { 1604358d963SJosef Bacik struct rb_node **p; 1614358d963SJosef Bacik struct rb_node *parent = NULL; 16232da5386SDavid Sterba struct btrfs_block_group *cache; 1634358d963SJosef Bacik 1644358d963SJosef Bacik spin_lock(&info->block_group_cache_lock); 1654358d963SJosef Bacik p = &info->block_group_cache_tree.rb_node; 1664358d963SJosef Bacik 1674358d963SJosef Bacik while (*p) { 1684358d963SJosef Bacik parent = *p; 16932da5386SDavid Sterba cache = rb_entry(parent, struct btrfs_block_group, cache_node); 170b3470b5dSDavid Sterba if (block_group->start < cache->start) { 1714358d963SJosef Bacik p = &(*p)->rb_left; 172b3470b5dSDavid Sterba } else if (block_group->start > cache->start) { 1734358d963SJosef Bacik p = &(*p)->rb_right; 1744358d963SJosef Bacik } else { 1754358d963SJosef Bacik spin_unlock(&info->block_group_cache_lock); 1764358d963SJosef Bacik return -EEXIST; 1774358d963SJosef Bacik } 1784358d963SJosef Bacik } 1794358d963SJosef Bacik 1804358d963SJosef Bacik rb_link_node(&block_group->cache_node, parent, p); 1814358d963SJosef Bacik rb_insert_color(&block_group->cache_node, 1824358d963SJosef Bacik &info->block_group_cache_tree); 1834358d963SJosef Bacik 184b3470b5dSDavid Sterba if (info->first_logical_byte > block_group->start) 185b3470b5dSDavid Sterba info->first_logical_byte = block_group->start; 1864358d963SJosef Bacik 1874358d963SJosef Bacik spin_unlock(&info->block_group_cache_lock); 1884358d963SJosef Bacik 1894358d963SJosef Bacik return 0; 1904358d963SJosef Bacik } 1914358d963SJosef Bacik 1924358d963SJosef Bacik /* 1932e405ad8SJosef Bacik * This will return the block group at or after bytenr if contains is 0, else 1942e405ad8SJosef Bacik * it will return the block group that contains the bytenr 1952e405ad8SJosef Bacik */ 19632da5386SDavid Sterba static struct btrfs_block_group *block_group_cache_tree_search( 1972e405ad8SJosef Bacik struct btrfs_fs_info *info, u64 bytenr, int contains) 1982e405ad8SJosef Bacik { 19932da5386SDavid Sterba struct btrfs_block_group *cache, *ret = NULL; 2002e405ad8SJosef Bacik struct rb_node *n; 2012e405ad8SJosef Bacik u64 end, start; 2022e405ad8SJosef Bacik 2032e405ad8SJosef Bacik spin_lock(&info->block_group_cache_lock); 2042e405ad8SJosef Bacik n = info->block_group_cache_tree.rb_node; 2052e405ad8SJosef Bacik 2062e405ad8SJosef Bacik while (n) { 20732da5386SDavid Sterba cache = rb_entry(n, struct btrfs_block_group, cache_node); 208b3470b5dSDavid Sterba end = cache->start + cache->length - 1; 209b3470b5dSDavid Sterba start = cache->start; 2102e405ad8SJosef Bacik 2112e405ad8SJosef Bacik if (bytenr < start) { 212b3470b5dSDavid Sterba if (!contains && (!ret || start < ret->start)) 2132e405ad8SJosef Bacik ret = cache; 2142e405ad8SJosef Bacik n = n->rb_left; 2152e405ad8SJosef Bacik } else if (bytenr > start) { 2162e405ad8SJosef Bacik if (contains && bytenr <= end) { 2172e405ad8SJosef Bacik ret = cache; 2182e405ad8SJosef Bacik break; 2192e405ad8SJosef Bacik } 2202e405ad8SJosef Bacik n = n->rb_right; 2212e405ad8SJosef Bacik } else { 2222e405ad8SJosef Bacik ret = cache; 2232e405ad8SJosef Bacik break; 2242e405ad8SJosef Bacik } 2252e405ad8SJosef Bacik } 2262e405ad8SJosef Bacik if (ret) { 2272e405ad8SJosef Bacik btrfs_get_block_group(ret); 228b3470b5dSDavid Sterba if (bytenr == 0 && info->first_logical_byte > ret->start) 229b3470b5dSDavid Sterba info->first_logical_byte = ret->start; 2302e405ad8SJosef Bacik } 2312e405ad8SJosef Bacik spin_unlock(&info->block_group_cache_lock); 2322e405ad8SJosef Bacik 2332e405ad8SJosef Bacik return ret; 2342e405ad8SJosef Bacik } 2352e405ad8SJosef Bacik 2362e405ad8SJosef Bacik /* 2372e405ad8SJosef Bacik * Return the block group that starts at or after bytenr 2382e405ad8SJosef Bacik */ 23932da5386SDavid Sterba struct btrfs_block_group *btrfs_lookup_first_block_group( 2402e405ad8SJosef Bacik struct btrfs_fs_info *info, u64 bytenr) 2412e405ad8SJosef Bacik { 2422e405ad8SJosef Bacik return block_group_cache_tree_search(info, bytenr, 0); 2432e405ad8SJosef Bacik } 2442e405ad8SJosef Bacik 2452e405ad8SJosef Bacik /* 2462e405ad8SJosef Bacik * Return the block group that contains the given bytenr 2472e405ad8SJosef Bacik */ 24832da5386SDavid Sterba struct btrfs_block_group *btrfs_lookup_block_group( 2492e405ad8SJosef Bacik struct btrfs_fs_info *info, u64 bytenr) 2502e405ad8SJosef Bacik { 2512e405ad8SJosef Bacik return block_group_cache_tree_search(info, bytenr, 1); 2522e405ad8SJosef Bacik } 2532e405ad8SJosef Bacik 25432da5386SDavid Sterba struct btrfs_block_group *btrfs_next_block_group( 25532da5386SDavid Sterba struct btrfs_block_group *cache) 2562e405ad8SJosef Bacik { 2572e405ad8SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 2582e405ad8SJosef Bacik struct rb_node *node; 2592e405ad8SJosef Bacik 2602e405ad8SJosef Bacik spin_lock(&fs_info->block_group_cache_lock); 2612e405ad8SJosef Bacik 2622e405ad8SJosef Bacik /* If our block group was removed, we need a full search. */ 2632e405ad8SJosef Bacik if (RB_EMPTY_NODE(&cache->cache_node)) { 264b3470b5dSDavid Sterba const u64 next_bytenr = cache->start + cache->length; 2652e405ad8SJosef Bacik 2662e405ad8SJosef Bacik spin_unlock(&fs_info->block_group_cache_lock); 2672e405ad8SJosef Bacik btrfs_put_block_group(cache); 2682e405ad8SJosef Bacik cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; 2692e405ad8SJosef Bacik } 2702e405ad8SJosef Bacik node = rb_next(&cache->cache_node); 2712e405ad8SJosef Bacik btrfs_put_block_group(cache); 2722e405ad8SJosef Bacik if (node) { 27332da5386SDavid Sterba cache = rb_entry(node, struct btrfs_block_group, cache_node); 2742e405ad8SJosef Bacik btrfs_get_block_group(cache); 2752e405ad8SJosef Bacik } else 2762e405ad8SJosef Bacik cache = NULL; 2772e405ad8SJosef Bacik spin_unlock(&fs_info->block_group_cache_lock); 2782e405ad8SJosef Bacik return cache; 2792e405ad8SJosef Bacik } 2803eeb3226SJosef Bacik 2813eeb3226SJosef Bacik bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) 2823eeb3226SJosef Bacik { 28332da5386SDavid Sterba struct btrfs_block_group *bg; 2843eeb3226SJosef Bacik bool ret = true; 2853eeb3226SJosef Bacik 2863eeb3226SJosef Bacik bg = btrfs_lookup_block_group(fs_info, bytenr); 2873eeb3226SJosef Bacik if (!bg) 2883eeb3226SJosef Bacik return false; 2893eeb3226SJosef Bacik 2903eeb3226SJosef Bacik spin_lock(&bg->lock); 2913eeb3226SJosef Bacik if (bg->ro) 2923eeb3226SJosef Bacik ret = false; 2933eeb3226SJosef Bacik else 2943eeb3226SJosef Bacik atomic_inc(&bg->nocow_writers); 2953eeb3226SJosef Bacik spin_unlock(&bg->lock); 2963eeb3226SJosef Bacik 2973eeb3226SJosef Bacik /* No put on block group, done by btrfs_dec_nocow_writers */ 2983eeb3226SJosef Bacik if (!ret) 2993eeb3226SJosef Bacik btrfs_put_block_group(bg); 3003eeb3226SJosef Bacik 3013eeb3226SJosef Bacik return ret; 3023eeb3226SJosef Bacik } 3033eeb3226SJosef Bacik 3043eeb3226SJosef Bacik void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) 3053eeb3226SJosef Bacik { 30632da5386SDavid Sterba struct btrfs_block_group *bg; 3073eeb3226SJosef Bacik 3083eeb3226SJosef Bacik bg = btrfs_lookup_block_group(fs_info, bytenr); 3093eeb3226SJosef Bacik ASSERT(bg); 3103eeb3226SJosef Bacik if (atomic_dec_and_test(&bg->nocow_writers)) 3113eeb3226SJosef Bacik wake_up_var(&bg->nocow_writers); 3123eeb3226SJosef Bacik /* 3133eeb3226SJosef Bacik * Once for our lookup and once for the lookup done by a previous call 3143eeb3226SJosef Bacik * to btrfs_inc_nocow_writers() 3153eeb3226SJosef Bacik */ 3163eeb3226SJosef Bacik btrfs_put_block_group(bg); 3173eeb3226SJosef Bacik btrfs_put_block_group(bg); 3183eeb3226SJosef Bacik } 3193eeb3226SJosef Bacik 32032da5386SDavid Sterba void btrfs_wait_nocow_writers(struct btrfs_block_group *bg) 3213eeb3226SJosef Bacik { 3223eeb3226SJosef Bacik wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); 3233eeb3226SJosef Bacik } 3243eeb3226SJosef Bacik 3253eeb3226SJosef Bacik void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, 3263eeb3226SJosef Bacik const u64 start) 3273eeb3226SJosef Bacik { 32832da5386SDavid Sterba struct btrfs_block_group *bg; 3293eeb3226SJosef Bacik 3303eeb3226SJosef Bacik bg = btrfs_lookup_block_group(fs_info, start); 3313eeb3226SJosef Bacik ASSERT(bg); 3323eeb3226SJosef Bacik if (atomic_dec_and_test(&bg->reservations)) 3333eeb3226SJosef Bacik wake_up_var(&bg->reservations); 3343eeb3226SJosef Bacik btrfs_put_block_group(bg); 3353eeb3226SJosef Bacik } 3363eeb3226SJosef Bacik 33732da5386SDavid Sterba void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg) 3383eeb3226SJosef Bacik { 3393eeb3226SJosef Bacik struct btrfs_space_info *space_info = bg->space_info; 3403eeb3226SJosef Bacik 3413eeb3226SJosef Bacik ASSERT(bg->ro); 3423eeb3226SJosef Bacik 3433eeb3226SJosef Bacik if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) 3443eeb3226SJosef Bacik return; 3453eeb3226SJosef Bacik 3463eeb3226SJosef Bacik /* 3473eeb3226SJosef Bacik * Our block group is read only but before we set it to read only, 3483eeb3226SJosef Bacik * some task might have had allocated an extent from it already, but it 3493eeb3226SJosef Bacik * has not yet created a respective ordered extent (and added it to a 3503eeb3226SJosef Bacik * root's list of ordered extents). 3513eeb3226SJosef Bacik * Therefore wait for any task currently allocating extents, since the 3523eeb3226SJosef Bacik * block group's reservations counter is incremented while a read lock 3533eeb3226SJosef Bacik * on the groups' semaphore is held and decremented after releasing 3543eeb3226SJosef Bacik * the read access on that semaphore and creating the ordered extent. 3553eeb3226SJosef Bacik */ 3563eeb3226SJosef Bacik down_write(&space_info->groups_sem); 3573eeb3226SJosef Bacik up_write(&space_info->groups_sem); 3583eeb3226SJosef Bacik 3593eeb3226SJosef Bacik wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); 3603eeb3226SJosef Bacik } 3619f21246dSJosef Bacik 3629f21246dSJosef Bacik struct btrfs_caching_control *btrfs_get_caching_control( 36332da5386SDavid Sterba struct btrfs_block_group *cache) 3649f21246dSJosef Bacik { 3659f21246dSJosef Bacik struct btrfs_caching_control *ctl; 3669f21246dSJosef Bacik 3679f21246dSJosef Bacik spin_lock(&cache->lock); 3689f21246dSJosef Bacik if (!cache->caching_ctl) { 3699f21246dSJosef Bacik spin_unlock(&cache->lock); 3709f21246dSJosef Bacik return NULL; 3719f21246dSJosef Bacik } 3729f21246dSJosef Bacik 3739f21246dSJosef Bacik ctl = cache->caching_ctl; 3749f21246dSJosef Bacik refcount_inc(&ctl->count); 3759f21246dSJosef Bacik spin_unlock(&cache->lock); 3769f21246dSJosef Bacik return ctl; 3779f21246dSJosef Bacik } 3789f21246dSJosef Bacik 3799f21246dSJosef Bacik void btrfs_put_caching_control(struct btrfs_caching_control *ctl) 3809f21246dSJosef Bacik { 3819f21246dSJosef Bacik if (refcount_dec_and_test(&ctl->count)) 3829f21246dSJosef Bacik kfree(ctl); 3839f21246dSJosef Bacik } 3849f21246dSJosef Bacik 3859f21246dSJosef Bacik /* 3869f21246dSJosef Bacik * When we wait for progress in the block group caching, its because our 3879f21246dSJosef Bacik * allocation attempt failed at least once. So, we must sleep and let some 3889f21246dSJosef Bacik * progress happen before we try again. 3899f21246dSJosef Bacik * 3909f21246dSJosef Bacik * This function will sleep at least once waiting for new free space to show 3919f21246dSJosef Bacik * up, and then it will check the block group free space numbers for our min 3929f21246dSJosef Bacik * num_bytes. Another option is to have it go ahead and look in the rbtree for 3939f21246dSJosef Bacik * a free extent of a given size, but this is a good start. 3949f21246dSJosef Bacik * 3959f21246dSJosef Bacik * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using 3969f21246dSJosef Bacik * any of the information in this block group. 3979f21246dSJosef Bacik */ 39832da5386SDavid Sterba void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache, 3999f21246dSJosef Bacik u64 num_bytes) 4009f21246dSJosef Bacik { 4019f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 4029f21246dSJosef Bacik 4039f21246dSJosef Bacik caching_ctl = btrfs_get_caching_control(cache); 4049f21246dSJosef Bacik if (!caching_ctl) 4059f21246dSJosef Bacik return; 4069f21246dSJosef Bacik 40732da5386SDavid Sterba wait_event(caching_ctl->wait, btrfs_block_group_done(cache) || 4089f21246dSJosef Bacik (cache->free_space_ctl->free_space >= num_bytes)); 4099f21246dSJosef Bacik 4109f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 4119f21246dSJosef Bacik } 4129f21246dSJosef Bacik 41332da5386SDavid Sterba int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache) 4149f21246dSJosef Bacik { 4159f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 4169f21246dSJosef Bacik int ret = 0; 4179f21246dSJosef Bacik 4189f21246dSJosef Bacik caching_ctl = btrfs_get_caching_control(cache); 4199f21246dSJosef Bacik if (!caching_ctl) 4209f21246dSJosef Bacik return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; 4219f21246dSJosef Bacik 42232da5386SDavid Sterba wait_event(caching_ctl->wait, btrfs_block_group_done(cache)); 4239f21246dSJosef Bacik if (cache->cached == BTRFS_CACHE_ERROR) 4249f21246dSJosef Bacik ret = -EIO; 4259f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 4269f21246dSJosef Bacik return ret; 4279f21246dSJosef Bacik } 4289f21246dSJosef Bacik 4299f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 43032da5386SDavid Sterba static void fragment_free_space(struct btrfs_block_group *block_group) 4319f21246dSJosef Bacik { 4329f21246dSJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 433b3470b5dSDavid Sterba u64 start = block_group->start; 434b3470b5dSDavid Sterba u64 len = block_group->length; 4359f21246dSJosef Bacik u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? 4369f21246dSJosef Bacik fs_info->nodesize : fs_info->sectorsize; 4379f21246dSJosef Bacik u64 step = chunk << 1; 4389f21246dSJosef Bacik 4399f21246dSJosef Bacik while (len > chunk) { 4409f21246dSJosef Bacik btrfs_remove_free_space(block_group, start, chunk); 4419f21246dSJosef Bacik start += step; 4429f21246dSJosef Bacik if (len < step) 4439f21246dSJosef Bacik len = 0; 4449f21246dSJosef Bacik else 4459f21246dSJosef Bacik len -= step; 4469f21246dSJosef Bacik } 4479f21246dSJosef Bacik } 4489f21246dSJosef Bacik #endif 4499f21246dSJosef Bacik 4509f21246dSJosef Bacik /* 4519f21246dSJosef Bacik * This is only called by btrfs_cache_block_group, since we could have freed 4529f21246dSJosef Bacik * extents we need to check the pinned_extents for any extents that can't be 4539f21246dSJosef Bacik * used yet since their free space will be released as soon as the transaction 4549f21246dSJosef Bacik * commits. 4559f21246dSJosef Bacik */ 45632da5386SDavid Sterba u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end) 4579f21246dSJosef Bacik { 4589f21246dSJosef Bacik struct btrfs_fs_info *info = block_group->fs_info; 4599f21246dSJosef Bacik u64 extent_start, extent_end, size, total_added = 0; 4609f21246dSJosef Bacik int ret; 4619f21246dSJosef Bacik 4629f21246dSJosef Bacik while (start < end) { 463fe119a6eSNikolay Borisov ret = find_first_extent_bit(&info->excluded_extents, start, 4649f21246dSJosef Bacik &extent_start, &extent_end, 4659f21246dSJosef Bacik EXTENT_DIRTY | EXTENT_UPTODATE, 4669f21246dSJosef Bacik NULL); 4679f21246dSJosef Bacik if (ret) 4689f21246dSJosef Bacik break; 4699f21246dSJosef Bacik 4709f21246dSJosef Bacik if (extent_start <= start) { 4719f21246dSJosef Bacik start = extent_end + 1; 4729f21246dSJosef Bacik } else if (extent_start > start && extent_start < end) { 4739f21246dSJosef Bacik size = extent_start - start; 4749f21246dSJosef Bacik total_added += size; 475b0643e59SDennis Zhou ret = btrfs_add_free_space_async_trimmed(block_group, 476b0643e59SDennis Zhou start, size); 4779f21246dSJosef Bacik BUG_ON(ret); /* -ENOMEM or logic error */ 4789f21246dSJosef Bacik start = extent_end + 1; 4799f21246dSJosef Bacik } else { 4809f21246dSJosef Bacik break; 4819f21246dSJosef Bacik } 4829f21246dSJosef Bacik } 4839f21246dSJosef Bacik 4849f21246dSJosef Bacik if (start < end) { 4859f21246dSJosef Bacik size = end - start; 4869f21246dSJosef Bacik total_added += size; 487b0643e59SDennis Zhou ret = btrfs_add_free_space_async_trimmed(block_group, start, 488b0643e59SDennis Zhou size); 4899f21246dSJosef Bacik BUG_ON(ret); /* -ENOMEM or logic error */ 4909f21246dSJosef Bacik } 4919f21246dSJosef Bacik 4929f21246dSJosef Bacik return total_added; 4939f21246dSJosef Bacik } 4949f21246dSJosef Bacik 4959f21246dSJosef Bacik static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) 4969f21246dSJosef Bacik { 49732da5386SDavid Sterba struct btrfs_block_group *block_group = caching_ctl->block_group; 4989f21246dSJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 4999f21246dSJosef Bacik struct btrfs_root *extent_root = fs_info->extent_root; 5009f21246dSJosef Bacik struct btrfs_path *path; 5019f21246dSJosef Bacik struct extent_buffer *leaf; 5029f21246dSJosef Bacik struct btrfs_key key; 5039f21246dSJosef Bacik u64 total_found = 0; 5049f21246dSJosef Bacik u64 last = 0; 5059f21246dSJosef Bacik u32 nritems; 5069f21246dSJosef Bacik int ret; 5079f21246dSJosef Bacik bool wakeup = true; 5089f21246dSJosef Bacik 5099f21246dSJosef Bacik path = btrfs_alloc_path(); 5109f21246dSJosef Bacik if (!path) 5119f21246dSJosef Bacik return -ENOMEM; 5129f21246dSJosef Bacik 513b3470b5dSDavid Sterba last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET); 5149f21246dSJosef Bacik 5159f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 5169f21246dSJosef Bacik /* 5179f21246dSJosef Bacik * If we're fragmenting we don't want to make anybody think we can 5189f21246dSJosef Bacik * allocate from this block group until we've had a chance to fragment 5199f21246dSJosef Bacik * the free space. 5209f21246dSJosef Bacik */ 5219f21246dSJosef Bacik if (btrfs_should_fragment_free_space(block_group)) 5229f21246dSJosef Bacik wakeup = false; 5239f21246dSJosef Bacik #endif 5249f21246dSJosef Bacik /* 5259f21246dSJosef Bacik * We don't want to deadlock with somebody trying to allocate a new 5269f21246dSJosef Bacik * extent for the extent root while also trying to search the extent 5279f21246dSJosef Bacik * root to add free space. So we skip locking and search the commit 5289f21246dSJosef Bacik * root, since its read-only 5299f21246dSJosef Bacik */ 5309f21246dSJosef Bacik path->skip_locking = 1; 5319f21246dSJosef Bacik path->search_commit_root = 1; 5329f21246dSJosef Bacik path->reada = READA_FORWARD; 5339f21246dSJosef Bacik 5349f21246dSJosef Bacik key.objectid = last; 5359f21246dSJosef Bacik key.offset = 0; 5369f21246dSJosef Bacik key.type = BTRFS_EXTENT_ITEM_KEY; 5379f21246dSJosef Bacik 5389f21246dSJosef Bacik next: 5399f21246dSJosef Bacik ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); 5409f21246dSJosef Bacik if (ret < 0) 5419f21246dSJosef Bacik goto out; 5429f21246dSJosef Bacik 5439f21246dSJosef Bacik leaf = path->nodes[0]; 5449f21246dSJosef Bacik nritems = btrfs_header_nritems(leaf); 5459f21246dSJosef Bacik 5469f21246dSJosef Bacik while (1) { 5479f21246dSJosef Bacik if (btrfs_fs_closing(fs_info) > 1) { 5489f21246dSJosef Bacik last = (u64)-1; 5499f21246dSJosef Bacik break; 5509f21246dSJosef Bacik } 5519f21246dSJosef Bacik 5529f21246dSJosef Bacik if (path->slots[0] < nritems) { 5539f21246dSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 5549f21246dSJosef Bacik } else { 5559f21246dSJosef Bacik ret = btrfs_find_next_key(extent_root, path, &key, 0, 0); 5569f21246dSJosef Bacik if (ret) 5579f21246dSJosef Bacik break; 5589f21246dSJosef Bacik 5599f21246dSJosef Bacik if (need_resched() || 5609f21246dSJosef Bacik rwsem_is_contended(&fs_info->commit_root_sem)) { 5619f21246dSJosef Bacik if (wakeup) 5629f21246dSJosef Bacik caching_ctl->progress = last; 5639f21246dSJosef Bacik btrfs_release_path(path); 5649f21246dSJosef Bacik up_read(&fs_info->commit_root_sem); 5659f21246dSJosef Bacik mutex_unlock(&caching_ctl->mutex); 5669f21246dSJosef Bacik cond_resched(); 5679f21246dSJosef Bacik mutex_lock(&caching_ctl->mutex); 5689f21246dSJosef Bacik down_read(&fs_info->commit_root_sem); 5699f21246dSJosef Bacik goto next; 5709f21246dSJosef Bacik } 5719f21246dSJosef Bacik 5729f21246dSJosef Bacik ret = btrfs_next_leaf(extent_root, path); 5739f21246dSJosef Bacik if (ret < 0) 5749f21246dSJosef Bacik goto out; 5759f21246dSJosef Bacik if (ret) 5769f21246dSJosef Bacik break; 5779f21246dSJosef Bacik leaf = path->nodes[0]; 5789f21246dSJosef Bacik nritems = btrfs_header_nritems(leaf); 5799f21246dSJosef Bacik continue; 5809f21246dSJosef Bacik } 5819f21246dSJosef Bacik 5829f21246dSJosef Bacik if (key.objectid < last) { 5839f21246dSJosef Bacik key.objectid = last; 5849f21246dSJosef Bacik key.offset = 0; 5859f21246dSJosef Bacik key.type = BTRFS_EXTENT_ITEM_KEY; 5869f21246dSJosef Bacik 5879f21246dSJosef Bacik if (wakeup) 5889f21246dSJosef Bacik caching_ctl->progress = last; 5899f21246dSJosef Bacik btrfs_release_path(path); 5909f21246dSJosef Bacik goto next; 5919f21246dSJosef Bacik } 5929f21246dSJosef Bacik 593b3470b5dSDavid Sterba if (key.objectid < block_group->start) { 5949f21246dSJosef Bacik path->slots[0]++; 5959f21246dSJosef Bacik continue; 5969f21246dSJosef Bacik } 5979f21246dSJosef Bacik 598b3470b5dSDavid Sterba if (key.objectid >= block_group->start + block_group->length) 5999f21246dSJosef Bacik break; 6009f21246dSJosef Bacik 6019f21246dSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 6029f21246dSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 6039f21246dSJosef Bacik total_found += add_new_free_space(block_group, last, 6049f21246dSJosef Bacik key.objectid); 6059f21246dSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY) 6069f21246dSJosef Bacik last = key.objectid + 6079f21246dSJosef Bacik fs_info->nodesize; 6089f21246dSJosef Bacik else 6099f21246dSJosef Bacik last = key.objectid + key.offset; 6109f21246dSJosef Bacik 6119f21246dSJosef Bacik if (total_found > CACHING_CTL_WAKE_UP) { 6129f21246dSJosef Bacik total_found = 0; 6139f21246dSJosef Bacik if (wakeup) 6149f21246dSJosef Bacik wake_up(&caching_ctl->wait); 6159f21246dSJosef Bacik } 6169f21246dSJosef Bacik } 6179f21246dSJosef Bacik path->slots[0]++; 6189f21246dSJosef Bacik } 6199f21246dSJosef Bacik ret = 0; 6209f21246dSJosef Bacik 6219f21246dSJosef Bacik total_found += add_new_free_space(block_group, last, 622b3470b5dSDavid Sterba block_group->start + block_group->length); 6239f21246dSJosef Bacik caching_ctl->progress = (u64)-1; 6249f21246dSJosef Bacik 6259f21246dSJosef Bacik out: 6269f21246dSJosef Bacik btrfs_free_path(path); 6279f21246dSJosef Bacik return ret; 6289f21246dSJosef Bacik } 6299f21246dSJosef Bacik 6309f21246dSJosef Bacik static noinline void caching_thread(struct btrfs_work *work) 6319f21246dSJosef Bacik { 63232da5386SDavid Sterba struct btrfs_block_group *block_group; 6339f21246dSJosef Bacik struct btrfs_fs_info *fs_info; 6349f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 6359f21246dSJosef Bacik int ret; 6369f21246dSJosef Bacik 6379f21246dSJosef Bacik caching_ctl = container_of(work, struct btrfs_caching_control, work); 6389f21246dSJosef Bacik block_group = caching_ctl->block_group; 6399f21246dSJosef Bacik fs_info = block_group->fs_info; 6409f21246dSJosef Bacik 6419f21246dSJosef Bacik mutex_lock(&caching_ctl->mutex); 6429f21246dSJosef Bacik down_read(&fs_info->commit_root_sem); 6439f21246dSJosef Bacik 6449f21246dSJosef Bacik if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) 6459f21246dSJosef Bacik ret = load_free_space_tree(caching_ctl); 6469f21246dSJosef Bacik else 6479f21246dSJosef Bacik ret = load_extent_tree_free(caching_ctl); 6489f21246dSJosef Bacik 6499f21246dSJosef Bacik spin_lock(&block_group->lock); 6509f21246dSJosef Bacik block_group->caching_ctl = NULL; 6519f21246dSJosef Bacik block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; 6529f21246dSJosef Bacik spin_unlock(&block_group->lock); 6539f21246dSJosef Bacik 6549f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 6559f21246dSJosef Bacik if (btrfs_should_fragment_free_space(block_group)) { 6569f21246dSJosef Bacik u64 bytes_used; 6579f21246dSJosef Bacik 6589f21246dSJosef Bacik spin_lock(&block_group->space_info->lock); 6599f21246dSJosef Bacik spin_lock(&block_group->lock); 660b3470b5dSDavid Sterba bytes_used = block_group->length - block_group->used; 6619f21246dSJosef Bacik block_group->space_info->bytes_used += bytes_used >> 1; 6629f21246dSJosef Bacik spin_unlock(&block_group->lock); 6639f21246dSJosef Bacik spin_unlock(&block_group->space_info->lock); 664e11c0406SJosef Bacik fragment_free_space(block_group); 6659f21246dSJosef Bacik } 6669f21246dSJosef Bacik #endif 6679f21246dSJosef Bacik 6689f21246dSJosef Bacik caching_ctl->progress = (u64)-1; 6699f21246dSJosef Bacik 6709f21246dSJosef Bacik up_read(&fs_info->commit_root_sem); 6719f21246dSJosef Bacik btrfs_free_excluded_extents(block_group); 6729f21246dSJosef Bacik mutex_unlock(&caching_ctl->mutex); 6739f21246dSJosef Bacik 6749f21246dSJosef Bacik wake_up(&caching_ctl->wait); 6759f21246dSJosef Bacik 6769f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 6779f21246dSJosef Bacik btrfs_put_block_group(block_group); 6789f21246dSJosef Bacik } 6799f21246dSJosef Bacik 68032da5386SDavid Sterba int btrfs_cache_block_group(struct btrfs_block_group *cache, int load_cache_only) 6819f21246dSJosef Bacik { 6829f21246dSJosef Bacik DEFINE_WAIT(wait); 6839f21246dSJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 6849f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 6859f21246dSJosef Bacik int ret = 0; 6869f21246dSJosef Bacik 6879f21246dSJosef Bacik caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); 6889f21246dSJosef Bacik if (!caching_ctl) 6899f21246dSJosef Bacik return -ENOMEM; 6909f21246dSJosef Bacik 6919f21246dSJosef Bacik INIT_LIST_HEAD(&caching_ctl->list); 6929f21246dSJosef Bacik mutex_init(&caching_ctl->mutex); 6939f21246dSJosef Bacik init_waitqueue_head(&caching_ctl->wait); 6949f21246dSJosef Bacik caching_ctl->block_group = cache; 695b3470b5dSDavid Sterba caching_ctl->progress = cache->start; 6969f21246dSJosef Bacik refcount_set(&caching_ctl->count, 1); 697a0cac0ecSOmar Sandoval btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); 6989f21246dSJosef Bacik 6999f21246dSJosef Bacik spin_lock(&cache->lock); 7009f21246dSJosef Bacik /* 7019f21246dSJosef Bacik * This should be a rare occasion, but this could happen I think in the 7029f21246dSJosef Bacik * case where one thread starts to load the space cache info, and then 7039f21246dSJosef Bacik * some other thread starts a transaction commit which tries to do an 7049f21246dSJosef Bacik * allocation while the other thread is still loading the space cache 7059f21246dSJosef Bacik * info. The previous loop should have kept us from choosing this block 7069f21246dSJosef Bacik * group, but if we've moved to the state where we will wait on caching 7079f21246dSJosef Bacik * block groups we need to first check if we're doing a fast load here, 7089f21246dSJosef Bacik * so we can wait for it to finish, otherwise we could end up allocating 7099f21246dSJosef Bacik * from a block group who's cache gets evicted for one reason or 7109f21246dSJosef Bacik * another. 7119f21246dSJosef Bacik */ 7129f21246dSJosef Bacik while (cache->cached == BTRFS_CACHE_FAST) { 7139f21246dSJosef Bacik struct btrfs_caching_control *ctl; 7149f21246dSJosef Bacik 7159f21246dSJosef Bacik ctl = cache->caching_ctl; 7169f21246dSJosef Bacik refcount_inc(&ctl->count); 7179f21246dSJosef Bacik prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); 7189f21246dSJosef Bacik spin_unlock(&cache->lock); 7199f21246dSJosef Bacik 7209f21246dSJosef Bacik schedule(); 7219f21246dSJosef Bacik 7229f21246dSJosef Bacik finish_wait(&ctl->wait, &wait); 7239f21246dSJosef Bacik btrfs_put_caching_control(ctl); 7249f21246dSJosef Bacik spin_lock(&cache->lock); 7259f21246dSJosef Bacik } 7269f21246dSJosef Bacik 7279f21246dSJosef Bacik if (cache->cached != BTRFS_CACHE_NO) { 7289f21246dSJosef Bacik spin_unlock(&cache->lock); 7299f21246dSJosef Bacik kfree(caching_ctl); 7309f21246dSJosef Bacik return 0; 7319f21246dSJosef Bacik } 7329f21246dSJosef Bacik WARN_ON(cache->caching_ctl); 7339f21246dSJosef Bacik cache->caching_ctl = caching_ctl; 7349f21246dSJosef Bacik cache->cached = BTRFS_CACHE_FAST; 7359f21246dSJosef Bacik spin_unlock(&cache->lock); 7369f21246dSJosef Bacik 7379f21246dSJosef Bacik if (btrfs_test_opt(fs_info, SPACE_CACHE)) { 7389f21246dSJosef Bacik mutex_lock(&caching_ctl->mutex); 7399f21246dSJosef Bacik ret = load_free_space_cache(cache); 7409f21246dSJosef Bacik 7419f21246dSJosef Bacik spin_lock(&cache->lock); 7429f21246dSJosef Bacik if (ret == 1) { 7439f21246dSJosef Bacik cache->caching_ctl = NULL; 7449f21246dSJosef Bacik cache->cached = BTRFS_CACHE_FINISHED; 7459f21246dSJosef Bacik cache->last_byte_to_unpin = (u64)-1; 7469f21246dSJosef Bacik caching_ctl->progress = (u64)-1; 7479f21246dSJosef Bacik } else { 7489f21246dSJosef Bacik if (load_cache_only) { 7499f21246dSJosef Bacik cache->caching_ctl = NULL; 7509f21246dSJosef Bacik cache->cached = BTRFS_CACHE_NO; 7519f21246dSJosef Bacik } else { 7529f21246dSJosef Bacik cache->cached = BTRFS_CACHE_STARTED; 7539f21246dSJosef Bacik cache->has_caching_ctl = 1; 7549f21246dSJosef Bacik } 7559f21246dSJosef Bacik } 7569f21246dSJosef Bacik spin_unlock(&cache->lock); 7579f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 7589f21246dSJosef Bacik if (ret == 1 && 7599f21246dSJosef Bacik btrfs_should_fragment_free_space(cache)) { 7609f21246dSJosef Bacik u64 bytes_used; 7619f21246dSJosef Bacik 7629f21246dSJosef Bacik spin_lock(&cache->space_info->lock); 7639f21246dSJosef Bacik spin_lock(&cache->lock); 764b3470b5dSDavid Sterba bytes_used = cache->length - cache->used; 7659f21246dSJosef Bacik cache->space_info->bytes_used += bytes_used >> 1; 7669f21246dSJosef Bacik spin_unlock(&cache->lock); 7679f21246dSJosef Bacik spin_unlock(&cache->space_info->lock); 768e11c0406SJosef Bacik fragment_free_space(cache); 7699f21246dSJosef Bacik } 7709f21246dSJosef Bacik #endif 7719f21246dSJosef Bacik mutex_unlock(&caching_ctl->mutex); 7729f21246dSJosef Bacik 7739f21246dSJosef Bacik wake_up(&caching_ctl->wait); 7749f21246dSJosef Bacik if (ret == 1) { 7759f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 7769f21246dSJosef Bacik btrfs_free_excluded_extents(cache); 7779f21246dSJosef Bacik return 0; 7789f21246dSJosef Bacik } 7799f21246dSJosef Bacik } else { 7809f21246dSJosef Bacik /* 7819f21246dSJosef Bacik * We're either using the free space tree or no caching at all. 7829f21246dSJosef Bacik * Set cached to the appropriate value and wakeup any waiters. 7839f21246dSJosef Bacik */ 7849f21246dSJosef Bacik spin_lock(&cache->lock); 7859f21246dSJosef Bacik if (load_cache_only) { 7869f21246dSJosef Bacik cache->caching_ctl = NULL; 7879f21246dSJosef Bacik cache->cached = BTRFS_CACHE_NO; 7889f21246dSJosef Bacik } else { 7899f21246dSJosef Bacik cache->cached = BTRFS_CACHE_STARTED; 7909f21246dSJosef Bacik cache->has_caching_ctl = 1; 7919f21246dSJosef Bacik } 7929f21246dSJosef Bacik spin_unlock(&cache->lock); 7939f21246dSJosef Bacik wake_up(&caching_ctl->wait); 7949f21246dSJosef Bacik } 7959f21246dSJosef Bacik 7969f21246dSJosef Bacik if (load_cache_only) { 7979f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 7989f21246dSJosef Bacik return 0; 7999f21246dSJosef Bacik } 8009f21246dSJosef Bacik 8019f21246dSJosef Bacik down_write(&fs_info->commit_root_sem); 8029f21246dSJosef Bacik refcount_inc(&caching_ctl->count); 8039f21246dSJosef Bacik list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); 8049f21246dSJosef Bacik up_write(&fs_info->commit_root_sem); 8059f21246dSJosef Bacik 8069f21246dSJosef Bacik btrfs_get_block_group(cache); 8079f21246dSJosef Bacik 8089f21246dSJosef Bacik btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); 8099f21246dSJosef Bacik 8109f21246dSJosef Bacik return ret; 8119f21246dSJosef Bacik } 812e3e0520bSJosef Bacik 813e3e0520bSJosef Bacik static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) 814e3e0520bSJosef Bacik { 815e3e0520bSJosef Bacik u64 extra_flags = chunk_to_extended(flags) & 816e3e0520bSJosef Bacik BTRFS_EXTENDED_PROFILE_MASK; 817e3e0520bSJosef Bacik 818e3e0520bSJosef Bacik write_seqlock(&fs_info->profiles_lock); 819e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 820e3e0520bSJosef Bacik fs_info->avail_data_alloc_bits &= ~extra_flags; 821e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_METADATA) 822e3e0520bSJosef Bacik fs_info->avail_metadata_alloc_bits &= ~extra_flags; 823e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 824e3e0520bSJosef Bacik fs_info->avail_system_alloc_bits &= ~extra_flags; 825e3e0520bSJosef Bacik write_sequnlock(&fs_info->profiles_lock); 826e3e0520bSJosef Bacik } 827e3e0520bSJosef Bacik 828e3e0520bSJosef Bacik /* 829e3e0520bSJosef Bacik * Clear incompat bits for the following feature(s): 830e3e0520bSJosef Bacik * 831e3e0520bSJosef Bacik * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group 832e3e0520bSJosef Bacik * in the whole filesystem 8339c907446SDavid Sterba * 8349c907446SDavid Sterba * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups 835e3e0520bSJosef Bacik */ 836e3e0520bSJosef Bacik static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) 837e3e0520bSJosef Bacik { 8389c907446SDavid Sterba bool found_raid56 = false; 8399c907446SDavid Sterba bool found_raid1c34 = false; 8409c907446SDavid Sterba 8419c907446SDavid Sterba if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) || 8429c907446SDavid Sterba (flags & BTRFS_BLOCK_GROUP_RAID1C3) || 8439c907446SDavid Sterba (flags & BTRFS_BLOCK_GROUP_RAID1C4)) { 844e3e0520bSJosef Bacik struct list_head *head = &fs_info->space_info; 845e3e0520bSJosef Bacik struct btrfs_space_info *sinfo; 846e3e0520bSJosef Bacik 847e3e0520bSJosef Bacik list_for_each_entry_rcu(sinfo, head, list) { 848e3e0520bSJosef Bacik down_read(&sinfo->groups_sem); 849e3e0520bSJosef Bacik if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) 8509c907446SDavid Sterba found_raid56 = true; 851e3e0520bSJosef Bacik if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) 8529c907446SDavid Sterba found_raid56 = true; 8539c907446SDavid Sterba if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3])) 8549c907446SDavid Sterba found_raid1c34 = true; 8559c907446SDavid Sterba if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4])) 8569c907446SDavid Sterba found_raid1c34 = true; 857e3e0520bSJosef Bacik up_read(&sinfo->groups_sem); 858e3e0520bSJosef Bacik } 859d8e6fd5cSFilipe Manana if (!found_raid56) 860e3e0520bSJosef Bacik btrfs_clear_fs_incompat(fs_info, RAID56); 861d8e6fd5cSFilipe Manana if (!found_raid1c34) 8629c907446SDavid Sterba btrfs_clear_fs_incompat(fs_info, RAID1C34); 863e3e0520bSJosef Bacik } 864e3e0520bSJosef Bacik } 865e3e0520bSJosef Bacik 866e3e0520bSJosef Bacik int btrfs_remove_block_group(struct btrfs_trans_handle *trans, 867e3e0520bSJosef Bacik u64 group_start, struct extent_map *em) 868e3e0520bSJosef Bacik { 869e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 870e3e0520bSJosef Bacik struct btrfs_root *root = fs_info->extent_root; 871e3e0520bSJosef Bacik struct btrfs_path *path; 87232da5386SDavid Sterba struct btrfs_block_group *block_group; 873e3e0520bSJosef Bacik struct btrfs_free_cluster *cluster; 874e3e0520bSJosef Bacik struct btrfs_root *tree_root = fs_info->tree_root; 875e3e0520bSJosef Bacik struct btrfs_key key; 876e3e0520bSJosef Bacik struct inode *inode; 877e3e0520bSJosef Bacik struct kobject *kobj = NULL; 878e3e0520bSJosef Bacik int ret; 879e3e0520bSJosef Bacik int index; 880e3e0520bSJosef Bacik int factor; 881e3e0520bSJosef Bacik struct btrfs_caching_control *caching_ctl = NULL; 882e3e0520bSJosef Bacik bool remove_em; 883e3e0520bSJosef Bacik bool remove_rsv = false; 884e3e0520bSJosef Bacik 885e3e0520bSJosef Bacik block_group = btrfs_lookup_block_group(fs_info, group_start); 886e3e0520bSJosef Bacik BUG_ON(!block_group); 887e3e0520bSJosef Bacik BUG_ON(!block_group->ro); 888e3e0520bSJosef Bacik 889e3e0520bSJosef Bacik trace_btrfs_remove_block_group(block_group); 890e3e0520bSJosef Bacik /* 891e3e0520bSJosef Bacik * Free the reserved super bytes from this block group before 892e3e0520bSJosef Bacik * remove it. 893e3e0520bSJosef Bacik */ 894e3e0520bSJosef Bacik btrfs_free_excluded_extents(block_group); 895b3470b5dSDavid Sterba btrfs_free_ref_tree_range(fs_info, block_group->start, 896b3470b5dSDavid Sterba block_group->length); 897e3e0520bSJosef Bacik 898e3e0520bSJosef Bacik index = btrfs_bg_flags_to_raid_index(block_group->flags); 899e3e0520bSJosef Bacik factor = btrfs_bg_type_to_factor(block_group->flags); 900e3e0520bSJosef Bacik 901e3e0520bSJosef Bacik /* make sure this block group isn't part of an allocation cluster */ 902e3e0520bSJosef Bacik cluster = &fs_info->data_alloc_cluster; 903e3e0520bSJosef Bacik spin_lock(&cluster->refill_lock); 904e3e0520bSJosef Bacik btrfs_return_cluster_to_free_space(block_group, cluster); 905e3e0520bSJosef Bacik spin_unlock(&cluster->refill_lock); 906e3e0520bSJosef Bacik 907e3e0520bSJosef Bacik /* 908e3e0520bSJosef Bacik * make sure this block group isn't part of a metadata 909e3e0520bSJosef Bacik * allocation cluster 910e3e0520bSJosef Bacik */ 911e3e0520bSJosef Bacik cluster = &fs_info->meta_alloc_cluster; 912e3e0520bSJosef Bacik spin_lock(&cluster->refill_lock); 913e3e0520bSJosef Bacik btrfs_return_cluster_to_free_space(block_group, cluster); 914e3e0520bSJosef Bacik spin_unlock(&cluster->refill_lock); 915e3e0520bSJosef Bacik 916e3e0520bSJosef Bacik path = btrfs_alloc_path(); 917e3e0520bSJosef Bacik if (!path) { 918e3e0520bSJosef Bacik ret = -ENOMEM; 919e3e0520bSJosef Bacik goto out; 920e3e0520bSJosef Bacik } 921e3e0520bSJosef Bacik 922e3e0520bSJosef Bacik /* 923e3e0520bSJosef Bacik * get the inode first so any iput calls done for the io_list 924e3e0520bSJosef Bacik * aren't the final iput (no unlinks allowed now) 925e3e0520bSJosef Bacik */ 926e3e0520bSJosef Bacik inode = lookup_free_space_inode(block_group, path); 927e3e0520bSJosef Bacik 928e3e0520bSJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 929e3e0520bSJosef Bacik /* 930e3e0520bSJosef Bacik * Make sure our free space cache IO is done before removing the 931e3e0520bSJosef Bacik * free space inode 932e3e0520bSJosef Bacik */ 933e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 934e3e0520bSJosef Bacik if (!list_empty(&block_group->io_list)) { 935e3e0520bSJosef Bacik list_del_init(&block_group->io_list); 936e3e0520bSJosef Bacik 937e3e0520bSJosef Bacik WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); 938e3e0520bSJosef Bacik 939e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 940e3e0520bSJosef Bacik btrfs_wait_cache_io(trans, block_group, path); 941e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 942e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 943e3e0520bSJosef Bacik } 944e3e0520bSJosef Bacik 945e3e0520bSJosef Bacik if (!list_empty(&block_group->dirty_list)) { 946e3e0520bSJosef Bacik list_del_init(&block_group->dirty_list); 947e3e0520bSJosef Bacik remove_rsv = true; 948e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 949e3e0520bSJosef Bacik } 950e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 951e3e0520bSJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 952e3e0520bSJosef Bacik 953e3e0520bSJosef Bacik if (!IS_ERR(inode)) { 954e3e0520bSJosef Bacik ret = btrfs_orphan_add(trans, BTRFS_I(inode)); 955e3e0520bSJosef Bacik if (ret) { 956e3e0520bSJosef Bacik btrfs_add_delayed_iput(inode); 957e3e0520bSJosef Bacik goto out; 958e3e0520bSJosef Bacik } 959e3e0520bSJosef Bacik clear_nlink(inode); 960e3e0520bSJosef Bacik /* One for the block groups ref */ 961e3e0520bSJosef Bacik spin_lock(&block_group->lock); 962e3e0520bSJosef Bacik if (block_group->iref) { 963e3e0520bSJosef Bacik block_group->iref = 0; 964e3e0520bSJosef Bacik block_group->inode = NULL; 965e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 966e3e0520bSJosef Bacik iput(inode); 967e3e0520bSJosef Bacik } else { 968e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 969e3e0520bSJosef Bacik } 970e3e0520bSJosef Bacik /* One for our lookup ref */ 971e3e0520bSJosef Bacik btrfs_add_delayed_iput(inode); 972e3e0520bSJosef Bacik } 973e3e0520bSJosef Bacik 974e3e0520bSJosef Bacik key.objectid = BTRFS_FREE_SPACE_OBJECTID; 975e3e0520bSJosef Bacik key.type = 0; 976b3470b5dSDavid Sterba key.offset = block_group->start; 977e3e0520bSJosef Bacik 978e3e0520bSJosef Bacik ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); 979e3e0520bSJosef Bacik if (ret < 0) 980e3e0520bSJosef Bacik goto out; 981e3e0520bSJosef Bacik if (ret > 0) 982e3e0520bSJosef Bacik btrfs_release_path(path); 983e3e0520bSJosef Bacik if (ret == 0) { 984e3e0520bSJosef Bacik ret = btrfs_del_item(trans, tree_root, path); 985e3e0520bSJosef Bacik if (ret) 986e3e0520bSJosef Bacik goto out; 987e3e0520bSJosef Bacik btrfs_release_path(path); 988e3e0520bSJosef Bacik } 989e3e0520bSJosef Bacik 990e3e0520bSJosef Bacik spin_lock(&fs_info->block_group_cache_lock); 991e3e0520bSJosef Bacik rb_erase(&block_group->cache_node, 992e3e0520bSJosef Bacik &fs_info->block_group_cache_tree); 993e3e0520bSJosef Bacik RB_CLEAR_NODE(&block_group->cache_node); 994e3e0520bSJosef Bacik 995b3470b5dSDavid Sterba if (fs_info->first_logical_byte == block_group->start) 996e3e0520bSJosef Bacik fs_info->first_logical_byte = (u64)-1; 997e3e0520bSJosef Bacik spin_unlock(&fs_info->block_group_cache_lock); 998e3e0520bSJosef Bacik 999e3e0520bSJosef Bacik down_write(&block_group->space_info->groups_sem); 1000e3e0520bSJosef Bacik /* 1001e3e0520bSJosef Bacik * we must use list_del_init so people can check to see if they 1002e3e0520bSJosef Bacik * are still on the list after taking the semaphore 1003e3e0520bSJosef Bacik */ 1004e3e0520bSJosef Bacik list_del_init(&block_group->list); 1005e3e0520bSJosef Bacik if (list_empty(&block_group->space_info->block_groups[index])) { 1006e3e0520bSJosef Bacik kobj = block_group->space_info->block_group_kobjs[index]; 1007e3e0520bSJosef Bacik block_group->space_info->block_group_kobjs[index] = NULL; 1008e3e0520bSJosef Bacik clear_avail_alloc_bits(fs_info, block_group->flags); 1009e3e0520bSJosef Bacik } 1010e3e0520bSJosef Bacik up_write(&block_group->space_info->groups_sem); 1011e3e0520bSJosef Bacik clear_incompat_bg_bits(fs_info, block_group->flags); 1012e3e0520bSJosef Bacik if (kobj) { 1013e3e0520bSJosef Bacik kobject_del(kobj); 1014e3e0520bSJosef Bacik kobject_put(kobj); 1015e3e0520bSJosef Bacik } 1016e3e0520bSJosef Bacik 1017e3e0520bSJosef Bacik if (block_group->has_caching_ctl) 1018e3e0520bSJosef Bacik caching_ctl = btrfs_get_caching_control(block_group); 1019e3e0520bSJosef Bacik if (block_group->cached == BTRFS_CACHE_STARTED) 1020e3e0520bSJosef Bacik btrfs_wait_block_group_cache_done(block_group); 1021e3e0520bSJosef Bacik if (block_group->has_caching_ctl) { 1022e3e0520bSJosef Bacik down_write(&fs_info->commit_root_sem); 1023e3e0520bSJosef Bacik if (!caching_ctl) { 1024e3e0520bSJosef Bacik struct btrfs_caching_control *ctl; 1025e3e0520bSJosef Bacik 1026e3e0520bSJosef Bacik list_for_each_entry(ctl, 1027e3e0520bSJosef Bacik &fs_info->caching_block_groups, list) 1028e3e0520bSJosef Bacik if (ctl->block_group == block_group) { 1029e3e0520bSJosef Bacik caching_ctl = ctl; 1030e3e0520bSJosef Bacik refcount_inc(&caching_ctl->count); 1031e3e0520bSJosef Bacik break; 1032e3e0520bSJosef Bacik } 1033e3e0520bSJosef Bacik } 1034e3e0520bSJosef Bacik if (caching_ctl) 1035e3e0520bSJosef Bacik list_del_init(&caching_ctl->list); 1036e3e0520bSJosef Bacik up_write(&fs_info->commit_root_sem); 1037e3e0520bSJosef Bacik if (caching_ctl) { 1038e3e0520bSJosef Bacik /* Once for the caching bgs list and once for us. */ 1039e3e0520bSJosef Bacik btrfs_put_caching_control(caching_ctl); 1040e3e0520bSJosef Bacik btrfs_put_caching_control(caching_ctl); 1041e3e0520bSJosef Bacik } 1042e3e0520bSJosef Bacik } 1043e3e0520bSJosef Bacik 1044e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1045e3e0520bSJosef Bacik WARN_ON(!list_empty(&block_group->dirty_list)); 1046e3e0520bSJosef Bacik WARN_ON(!list_empty(&block_group->io_list)); 1047e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1048e3e0520bSJosef Bacik 1049e3e0520bSJosef Bacik btrfs_remove_free_space_cache(block_group); 1050e3e0520bSJosef Bacik 1051e3e0520bSJosef Bacik spin_lock(&block_group->space_info->lock); 1052e3e0520bSJosef Bacik list_del_init(&block_group->ro_list); 1053e3e0520bSJosef Bacik 1054e3e0520bSJosef Bacik if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { 1055e3e0520bSJosef Bacik WARN_ON(block_group->space_info->total_bytes 1056b3470b5dSDavid Sterba < block_group->length); 1057e3e0520bSJosef Bacik WARN_ON(block_group->space_info->bytes_readonly 1058b3470b5dSDavid Sterba < block_group->length); 1059e3e0520bSJosef Bacik WARN_ON(block_group->space_info->disk_total 1060b3470b5dSDavid Sterba < block_group->length * factor); 1061e3e0520bSJosef Bacik } 1062b3470b5dSDavid Sterba block_group->space_info->total_bytes -= block_group->length; 1063b3470b5dSDavid Sterba block_group->space_info->bytes_readonly -= block_group->length; 1064b3470b5dSDavid Sterba block_group->space_info->disk_total -= block_group->length * factor; 1065e3e0520bSJosef Bacik 1066e3e0520bSJosef Bacik spin_unlock(&block_group->space_info->lock); 1067e3e0520bSJosef Bacik 1068b3470b5dSDavid Sterba key.objectid = block_group->start; 1069b3470b5dSDavid Sterba key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 1070b3470b5dSDavid Sterba key.offset = block_group->length; 1071e3e0520bSJosef Bacik 1072e3e0520bSJosef Bacik mutex_lock(&fs_info->chunk_mutex); 1073e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1074e3e0520bSJosef Bacik block_group->removed = 1; 1075e3e0520bSJosef Bacik /* 1076e3e0520bSJosef Bacik * At this point trimming can't start on this block group, because we 1077e3e0520bSJosef Bacik * removed the block group from the tree fs_info->block_group_cache_tree 1078e3e0520bSJosef Bacik * so no one can't find it anymore and even if someone already got this 1079e3e0520bSJosef Bacik * block group before we removed it from the rbtree, they have already 1080e3e0520bSJosef Bacik * incremented block_group->trimming - if they didn't, they won't find 1081e3e0520bSJosef Bacik * any free space entries because we already removed them all when we 1082e3e0520bSJosef Bacik * called btrfs_remove_free_space_cache(). 1083e3e0520bSJosef Bacik * 1084e3e0520bSJosef Bacik * And we must not remove the extent map from the fs_info->mapping_tree 1085e3e0520bSJosef Bacik * to prevent the same logical address range and physical device space 1086e3e0520bSJosef Bacik * ranges from being reused for a new block group. This is because our 1087e3e0520bSJosef Bacik * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is 1088e3e0520bSJosef Bacik * completely transactionless, so while it is trimming a range the 1089e3e0520bSJosef Bacik * currently running transaction might finish and a new one start, 1090e3e0520bSJosef Bacik * allowing for new block groups to be created that can reuse the same 1091e3e0520bSJosef Bacik * physical device locations unless we take this special care. 1092e3e0520bSJosef Bacik * 1093e3e0520bSJosef Bacik * There may also be an implicit trim operation if the file system 1094e3e0520bSJosef Bacik * is mounted with -odiscard. The same protections must remain 1095e3e0520bSJosef Bacik * in place until the extents have been discarded completely when 1096e3e0520bSJosef Bacik * the transaction commit has completed. 1097e3e0520bSJosef Bacik */ 1098e3e0520bSJosef Bacik remove_em = (atomic_read(&block_group->trimming) == 0); 1099e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1100e3e0520bSJosef Bacik 1101e3e0520bSJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 1102e3e0520bSJosef Bacik 1103e3e0520bSJosef Bacik ret = remove_block_group_free_space(trans, block_group); 1104e3e0520bSJosef Bacik if (ret) 1105e3e0520bSJosef Bacik goto out; 1106e3e0520bSJosef Bacik 1107e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1108e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1109e3e0520bSJosef Bacik 1110e3e0520bSJosef Bacik ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1111e3e0520bSJosef Bacik if (ret > 0) 1112e3e0520bSJosef Bacik ret = -EIO; 1113e3e0520bSJosef Bacik if (ret < 0) 1114e3e0520bSJosef Bacik goto out; 1115e3e0520bSJosef Bacik 1116e3e0520bSJosef Bacik ret = btrfs_del_item(trans, root, path); 1117e3e0520bSJosef Bacik if (ret) 1118e3e0520bSJosef Bacik goto out; 1119e3e0520bSJosef Bacik 1120e3e0520bSJosef Bacik if (remove_em) { 1121e3e0520bSJosef Bacik struct extent_map_tree *em_tree; 1122e3e0520bSJosef Bacik 1123e3e0520bSJosef Bacik em_tree = &fs_info->mapping_tree; 1124e3e0520bSJosef Bacik write_lock(&em_tree->lock); 1125e3e0520bSJosef Bacik remove_extent_mapping(em_tree, em); 1126e3e0520bSJosef Bacik write_unlock(&em_tree->lock); 1127e3e0520bSJosef Bacik /* once for the tree */ 1128e3e0520bSJosef Bacik free_extent_map(em); 1129e3e0520bSJosef Bacik } 1130e3e0520bSJosef Bacik out: 1131e3e0520bSJosef Bacik if (remove_rsv) 1132e3e0520bSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 1133e3e0520bSJosef Bacik btrfs_free_path(path); 1134e3e0520bSJosef Bacik return ret; 1135e3e0520bSJosef Bacik } 1136e3e0520bSJosef Bacik 1137e3e0520bSJosef Bacik struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( 1138e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info, const u64 chunk_offset) 1139e3e0520bSJosef Bacik { 1140e3e0520bSJosef Bacik struct extent_map_tree *em_tree = &fs_info->mapping_tree; 1141e3e0520bSJosef Bacik struct extent_map *em; 1142e3e0520bSJosef Bacik struct map_lookup *map; 1143e3e0520bSJosef Bacik unsigned int num_items; 1144e3e0520bSJosef Bacik 1145e3e0520bSJosef Bacik read_lock(&em_tree->lock); 1146e3e0520bSJosef Bacik em = lookup_extent_mapping(em_tree, chunk_offset, 1); 1147e3e0520bSJosef Bacik read_unlock(&em_tree->lock); 1148e3e0520bSJosef Bacik ASSERT(em && em->start == chunk_offset); 1149e3e0520bSJosef Bacik 1150e3e0520bSJosef Bacik /* 1151e3e0520bSJosef Bacik * We need to reserve 3 + N units from the metadata space info in order 1152e3e0520bSJosef Bacik * to remove a block group (done at btrfs_remove_chunk() and at 1153e3e0520bSJosef Bacik * btrfs_remove_block_group()), which are used for: 1154e3e0520bSJosef Bacik * 1155e3e0520bSJosef Bacik * 1 unit for adding the free space inode's orphan (located in the tree 1156e3e0520bSJosef Bacik * of tree roots). 1157e3e0520bSJosef Bacik * 1 unit for deleting the block group item (located in the extent 1158e3e0520bSJosef Bacik * tree). 1159e3e0520bSJosef Bacik * 1 unit for deleting the free space item (located in tree of tree 1160e3e0520bSJosef Bacik * roots). 1161e3e0520bSJosef Bacik * N units for deleting N device extent items corresponding to each 1162e3e0520bSJosef Bacik * stripe (located in the device tree). 1163e3e0520bSJosef Bacik * 1164e3e0520bSJosef Bacik * In order to remove a block group we also need to reserve units in the 1165e3e0520bSJosef Bacik * system space info in order to update the chunk tree (update one or 1166e3e0520bSJosef Bacik * more device items and remove one chunk item), but this is done at 1167e3e0520bSJosef Bacik * btrfs_remove_chunk() through a call to check_system_chunk(). 1168e3e0520bSJosef Bacik */ 1169e3e0520bSJosef Bacik map = em->map_lookup; 1170e3e0520bSJosef Bacik num_items = 3 + map->num_stripes; 1171e3e0520bSJosef Bacik free_extent_map(em); 1172e3e0520bSJosef Bacik 1173e3e0520bSJosef Bacik return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, 1174e3e0520bSJosef Bacik num_items, 1); 1175e3e0520bSJosef Bacik } 1176e3e0520bSJosef Bacik 1177e3e0520bSJosef Bacik /* 117826ce2095SJosef Bacik * Mark block group @cache read-only, so later write won't happen to block 117926ce2095SJosef Bacik * group @cache. 118026ce2095SJosef Bacik * 118126ce2095SJosef Bacik * If @force is not set, this function will only mark the block group readonly 118226ce2095SJosef Bacik * if we have enough free space (1M) in other metadata/system block groups. 118326ce2095SJosef Bacik * If @force is not set, this function will mark the block group readonly 118426ce2095SJosef Bacik * without checking free space. 118526ce2095SJosef Bacik * 118626ce2095SJosef Bacik * NOTE: This function doesn't care if other block groups can contain all the 118726ce2095SJosef Bacik * data in this block group. That check should be done by relocation routine, 118826ce2095SJosef Bacik * not this function. 118926ce2095SJosef Bacik */ 119032da5386SDavid Sterba static int inc_block_group_ro(struct btrfs_block_group *cache, int force) 119126ce2095SJosef Bacik { 119226ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 119326ce2095SJosef Bacik u64 num_bytes; 119426ce2095SJosef Bacik int ret = -ENOSPC; 119526ce2095SJosef Bacik 119626ce2095SJosef Bacik spin_lock(&sinfo->lock); 119726ce2095SJosef Bacik spin_lock(&cache->lock); 119826ce2095SJosef Bacik 119926ce2095SJosef Bacik if (cache->ro) { 120026ce2095SJosef Bacik cache->ro++; 120126ce2095SJosef Bacik ret = 0; 120226ce2095SJosef Bacik goto out; 120326ce2095SJosef Bacik } 120426ce2095SJosef Bacik 1205b3470b5dSDavid Sterba num_bytes = cache->length - cache->reserved - cache->pinned - 1206bf38be65SDavid Sterba cache->bytes_super - cache->used; 120726ce2095SJosef Bacik 120826ce2095SJosef Bacik /* 1209a30a3d20SJosef Bacik * Data never overcommits, even in mixed mode, so do just the straight 1210a30a3d20SJosef Bacik * check of left over space in how much we have allocated. 1211a30a3d20SJosef Bacik */ 1212a30a3d20SJosef Bacik if (force) { 1213a30a3d20SJosef Bacik ret = 0; 1214a30a3d20SJosef Bacik } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) { 1215a30a3d20SJosef Bacik u64 sinfo_used = btrfs_space_info_used(sinfo, true); 1216a30a3d20SJosef Bacik 1217a30a3d20SJosef Bacik /* 121826ce2095SJosef Bacik * Here we make sure if we mark this bg RO, we still have enough 1219f8935566SJosef Bacik * free space as buffer. 122026ce2095SJosef Bacik */ 1221a30a3d20SJosef Bacik if (sinfo_used + num_bytes <= sinfo->total_bytes) 1222a30a3d20SJosef Bacik ret = 0; 1223a30a3d20SJosef Bacik } else { 1224a30a3d20SJosef Bacik /* 1225a30a3d20SJosef Bacik * We overcommit metadata, so we need to do the 1226a30a3d20SJosef Bacik * btrfs_can_overcommit check here, and we need to pass in 1227a30a3d20SJosef Bacik * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of 1228a30a3d20SJosef Bacik * leeway to allow us to mark this block group as read only. 1229a30a3d20SJosef Bacik */ 1230a30a3d20SJosef Bacik if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes, 1231a30a3d20SJosef Bacik BTRFS_RESERVE_NO_FLUSH)) 1232a30a3d20SJosef Bacik ret = 0; 1233a30a3d20SJosef Bacik } 1234a30a3d20SJosef Bacik 1235a30a3d20SJosef Bacik if (!ret) { 123626ce2095SJosef Bacik sinfo->bytes_readonly += num_bytes; 123726ce2095SJosef Bacik cache->ro++; 123826ce2095SJosef Bacik list_add_tail(&cache->ro_list, &sinfo->ro_bgs); 123926ce2095SJosef Bacik } 124026ce2095SJosef Bacik out: 124126ce2095SJosef Bacik spin_unlock(&cache->lock); 124226ce2095SJosef Bacik spin_unlock(&sinfo->lock); 124326ce2095SJosef Bacik if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { 124426ce2095SJosef Bacik btrfs_info(cache->fs_info, 1245b3470b5dSDavid Sterba "unable to make block group %llu ro", cache->start); 124626ce2095SJosef Bacik btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); 124726ce2095SJosef Bacik } 124826ce2095SJosef Bacik return ret; 124926ce2095SJosef Bacik } 125026ce2095SJosef Bacik 1251fe119a6eSNikolay Borisov static bool clean_pinned_extents(struct btrfs_trans_handle *trans, 1252fe119a6eSNikolay Borisov struct btrfs_block_group *bg) 125345bb5d6aSNikolay Borisov { 125445bb5d6aSNikolay Borisov struct btrfs_fs_info *fs_info = bg->fs_info; 1255fe119a6eSNikolay Borisov struct btrfs_transaction *prev_trans = NULL; 125645bb5d6aSNikolay Borisov const u64 start = bg->start; 125745bb5d6aSNikolay Borisov const u64 end = start + bg->length - 1; 125845bb5d6aSNikolay Borisov int ret; 125945bb5d6aSNikolay Borisov 1260fe119a6eSNikolay Borisov spin_lock(&fs_info->trans_lock); 1261fe119a6eSNikolay Borisov if (trans->transaction->list.prev != &fs_info->trans_list) { 1262fe119a6eSNikolay Borisov prev_trans = list_last_entry(&trans->transaction->list, 1263fe119a6eSNikolay Borisov struct btrfs_transaction, list); 1264fe119a6eSNikolay Borisov refcount_inc(&prev_trans->use_count); 1265fe119a6eSNikolay Borisov } 1266fe119a6eSNikolay Borisov spin_unlock(&fs_info->trans_lock); 1267fe119a6eSNikolay Borisov 126845bb5d6aSNikolay Borisov /* 126945bb5d6aSNikolay Borisov * Hold the unused_bg_unpin_mutex lock to avoid racing with 127045bb5d6aSNikolay Borisov * btrfs_finish_extent_commit(). If we are at transaction N, another 127145bb5d6aSNikolay Borisov * task might be running finish_extent_commit() for the previous 127245bb5d6aSNikolay Borisov * transaction N - 1, and have seen a range belonging to the block 1273fe119a6eSNikolay Borisov * group in pinned_extents before we were able to clear the whole block 1274fe119a6eSNikolay Borisov * group range from pinned_extents. This means that task can lookup for 1275fe119a6eSNikolay Borisov * the block group after we unpinned it from pinned_extents and removed 1276fe119a6eSNikolay Borisov * it, leading to a BUG_ON() at unpin_extent_range(). 127745bb5d6aSNikolay Borisov */ 127845bb5d6aSNikolay Borisov mutex_lock(&fs_info->unused_bg_unpin_mutex); 1279fe119a6eSNikolay Borisov if (prev_trans) { 1280fe119a6eSNikolay Borisov ret = clear_extent_bits(&prev_trans->pinned_extents, start, end, 128145bb5d6aSNikolay Borisov EXTENT_DIRTY); 128245bb5d6aSNikolay Borisov if (ret) 128345bb5d6aSNikolay Borisov goto err; 1284fe119a6eSNikolay Borisov } 128545bb5d6aSNikolay Borisov 1286fe119a6eSNikolay Borisov ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end, 128745bb5d6aSNikolay Borisov EXTENT_DIRTY); 128845bb5d6aSNikolay Borisov if (ret) 128945bb5d6aSNikolay Borisov goto err; 129045bb5d6aSNikolay Borisov mutex_unlock(&fs_info->unused_bg_unpin_mutex); 129145bb5d6aSNikolay Borisov 129245bb5d6aSNikolay Borisov return true; 129345bb5d6aSNikolay Borisov 129445bb5d6aSNikolay Borisov err: 129545bb5d6aSNikolay Borisov mutex_unlock(&fs_info->unused_bg_unpin_mutex); 129645bb5d6aSNikolay Borisov btrfs_dec_block_group_ro(bg); 129745bb5d6aSNikolay Borisov return false; 129845bb5d6aSNikolay Borisov } 129945bb5d6aSNikolay Borisov 130026ce2095SJosef Bacik /* 1301e3e0520bSJosef Bacik * Process the unused_bgs list and remove any that don't have any allocated 1302e3e0520bSJosef Bacik * space inside of them. 1303e3e0520bSJosef Bacik */ 1304e3e0520bSJosef Bacik void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) 1305e3e0520bSJosef Bacik { 130632da5386SDavid Sterba struct btrfs_block_group *block_group; 1307e3e0520bSJosef Bacik struct btrfs_space_info *space_info; 1308e3e0520bSJosef Bacik struct btrfs_trans_handle *trans; 13096e80d4f8SDennis Zhou const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC); 1310e3e0520bSJosef Bacik int ret = 0; 1311e3e0520bSJosef Bacik 1312e3e0520bSJosef Bacik if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 1313e3e0520bSJosef Bacik return; 1314e3e0520bSJosef Bacik 1315e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1316e3e0520bSJosef Bacik while (!list_empty(&fs_info->unused_bgs)) { 1317e3e0520bSJosef Bacik int trimming; 1318e3e0520bSJosef Bacik 1319e3e0520bSJosef Bacik block_group = list_first_entry(&fs_info->unused_bgs, 132032da5386SDavid Sterba struct btrfs_block_group, 1321e3e0520bSJosef Bacik bg_list); 1322e3e0520bSJosef Bacik list_del_init(&block_group->bg_list); 1323e3e0520bSJosef Bacik 1324e3e0520bSJosef Bacik space_info = block_group->space_info; 1325e3e0520bSJosef Bacik 1326e3e0520bSJosef Bacik if (ret || btrfs_mixed_space_info(space_info)) { 1327e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1328e3e0520bSJosef Bacik continue; 1329e3e0520bSJosef Bacik } 1330e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1331e3e0520bSJosef Bacik 1332b0643e59SDennis Zhou btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 1333b0643e59SDennis Zhou 1334e3e0520bSJosef Bacik mutex_lock(&fs_info->delete_unused_bgs_mutex); 1335e3e0520bSJosef Bacik 1336e3e0520bSJosef Bacik /* Don't want to race with allocators so take the groups_sem */ 1337e3e0520bSJosef Bacik down_write(&space_info->groups_sem); 13386e80d4f8SDennis Zhou 13396e80d4f8SDennis Zhou /* 13406e80d4f8SDennis Zhou * Async discard moves the final block group discard to be prior 13416e80d4f8SDennis Zhou * to the unused_bgs code path. Therefore, if it's not fully 13426e80d4f8SDennis Zhou * trimmed, punt it back to the async discard lists. 13436e80d4f8SDennis Zhou */ 13446e80d4f8SDennis Zhou if (btrfs_test_opt(fs_info, DISCARD_ASYNC) && 13456e80d4f8SDennis Zhou !btrfs_is_free_space_trimmed(block_group)) { 13466e80d4f8SDennis Zhou trace_btrfs_skip_unused_block_group(block_group); 13476e80d4f8SDennis Zhou up_write(&space_info->groups_sem); 13486e80d4f8SDennis Zhou /* Requeue if we failed because of async discard */ 13496e80d4f8SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 13506e80d4f8SDennis Zhou block_group); 13516e80d4f8SDennis Zhou goto next; 13526e80d4f8SDennis Zhou } 13536e80d4f8SDennis Zhou 1354e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1355e3e0520bSJosef Bacik if (block_group->reserved || block_group->pinned || 1356bf38be65SDavid Sterba block_group->used || block_group->ro || 1357e3e0520bSJosef Bacik list_is_singular(&block_group->list)) { 1358e3e0520bSJosef Bacik /* 1359e3e0520bSJosef Bacik * We want to bail if we made new allocations or have 1360e3e0520bSJosef Bacik * outstanding allocations in this block group. We do 1361e3e0520bSJosef Bacik * the ro check in case balance is currently acting on 1362e3e0520bSJosef Bacik * this block group. 1363e3e0520bSJosef Bacik */ 1364e3e0520bSJosef Bacik trace_btrfs_skip_unused_block_group(block_group); 1365e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1366e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1367e3e0520bSJosef Bacik goto next; 1368e3e0520bSJosef Bacik } 1369e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1370e3e0520bSJosef Bacik 1371e3e0520bSJosef Bacik /* We don't want to force the issue, only flip if it's ok. */ 1372e11c0406SJosef Bacik ret = inc_block_group_ro(block_group, 0); 1373e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1374e3e0520bSJosef Bacik if (ret < 0) { 1375e3e0520bSJosef Bacik ret = 0; 1376e3e0520bSJosef Bacik goto next; 1377e3e0520bSJosef Bacik } 1378e3e0520bSJosef Bacik 1379e3e0520bSJosef Bacik /* 1380e3e0520bSJosef Bacik * Want to do this before we do anything else so we can recover 1381e3e0520bSJosef Bacik * properly if we fail to join the transaction. 1382e3e0520bSJosef Bacik */ 1383e3e0520bSJosef Bacik trans = btrfs_start_trans_remove_block_group(fs_info, 1384b3470b5dSDavid Sterba block_group->start); 1385e3e0520bSJosef Bacik if (IS_ERR(trans)) { 1386e3e0520bSJosef Bacik btrfs_dec_block_group_ro(block_group); 1387e3e0520bSJosef Bacik ret = PTR_ERR(trans); 1388e3e0520bSJosef Bacik goto next; 1389e3e0520bSJosef Bacik } 1390e3e0520bSJosef Bacik 1391e3e0520bSJosef Bacik /* 1392e3e0520bSJosef Bacik * We could have pending pinned extents for this block group, 1393e3e0520bSJosef Bacik * just delete them, we don't care about them anymore. 1394e3e0520bSJosef Bacik */ 1395fe119a6eSNikolay Borisov if (!clean_pinned_extents(trans, block_group)) 1396e3e0520bSJosef Bacik goto end_trans; 1397e3e0520bSJosef Bacik 1398b0643e59SDennis Zhou /* 1399b0643e59SDennis Zhou * At this point, the block_group is read only and should fail 1400b0643e59SDennis Zhou * new allocations. However, btrfs_finish_extent_commit() can 1401b0643e59SDennis Zhou * cause this block_group to be placed back on the discard 1402b0643e59SDennis Zhou * lists because now the block_group isn't fully discarded. 1403b0643e59SDennis Zhou * Bail here and try again later after discarding everything. 1404b0643e59SDennis Zhou */ 1405b0643e59SDennis Zhou spin_lock(&fs_info->discard_ctl.lock); 1406b0643e59SDennis Zhou if (!list_empty(&block_group->discard_list)) { 1407b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1408b0643e59SDennis Zhou btrfs_dec_block_group_ro(block_group); 1409b0643e59SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 1410b0643e59SDennis Zhou block_group); 1411b0643e59SDennis Zhou goto end_trans; 1412b0643e59SDennis Zhou } 1413b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1414b0643e59SDennis Zhou 1415e3e0520bSJosef Bacik /* Reset pinned so btrfs_put_block_group doesn't complain */ 1416e3e0520bSJosef Bacik spin_lock(&space_info->lock); 1417e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1418e3e0520bSJosef Bacik 1419e3e0520bSJosef Bacik btrfs_space_info_update_bytes_pinned(fs_info, space_info, 1420e3e0520bSJosef Bacik -block_group->pinned); 1421e3e0520bSJosef Bacik space_info->bytes_readonly += block_group->pinned; 1422e3e0520bSJosef Bacik percpu_counter_add_batch(&space_info->total_bytes_pinned, 1423e3e0520bSJosef Bacik -block_group->pinned, 1424e3e0520bSJosef Bacik BTRFS_TOTAL_BYTES_PINNED_BATCH); 1425e3e0520bSJosef Bacik block_group->pinned = 0; 1426e3e0520bSJosef Bacik 1427e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1428e3e0520bSJosef Bacik spin_unlock(&space_info->lock); 1429e3e0520bSJosef Bacik 14306e80d4f8SDennis Zhou /* 14316e80d4f8SDennis Zhou * The normal path here is an unused block group is passed here, 14326e80d4f8SDennis Zhou * then trimming is handled in the transaction commit path. 14336e80d4f8SDennis Zhou * Async discard interposes before this to do the trimming 14346e80d4f8SDennis Zhou * before coming down the unused block group path as trimming 14356e80d4f8SDennis Zhou * will no longer be done later in the transaction commit path. 14366e80d4f8SDennis Zhou */ 14376e80d4f8SDennis Zhou if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC)) 14386e80d4f8SDennis Zhou goto flip_async; 14396e80d4f8SDennis Zhou 1440e3e0520bSJosef Bacik /* DISCARD can flip during remount */ 144146b27f50SDennis Zhou trimming = btrfs_test_opt(fs_info, DISCARD_SYNC); 1442e3e0520bSJosef Bacik 1443e3e0520bSJosef Bacik /* Implicit trim during transaction commit. */ 1444e3e0520bSJosef Bacik if (trimming) 1445e3e0520bSJosef Bacik btrfs_get_block_group_trimming(block_group); 1446e3e0520bSJosef Bacik 1447e3e0520bSJosef Bacik /* 1448e3e0520bSJosef Bacik * Btrfs_remove_chunk will abort the transaction if things go 1449e3e0520bSJosef Bacik * horribly wrong. 1450e3e0520bSJosef Bacik */ 1451b3470b5dSDavid Sterba ret = btrfs_remove_chunk(trans, block_group->start); 1452e3e0520bSJosef Bacik 1453e3e0520bSJosef Bacik if (ret) { 1454e3e0520bSJosef Bacik if (trimming) 1455e3e0520bSJosef Bacik btrfs_put_block_group_trimming(block_group); 1456e3e0520bSJosef Bacik goto end_trans; 1457e3e0520bSJosef Bacik } 1458e3e0520bSJosef Bacik 1459e3e0520bSJosef Bacik /* 1460e3e0520bSJosef Bacik * If we're not mounted with -odiscard, we can just forget 1461e3e0520bSJosef Bacik * about this block group. Otherwise we'll need to wait 1462e3e0520bSJosef Bacik * until transaction commit to do the actual discard. 1463e3e0520bSJosef Bacik */ 1464e3e0520bSJosef Bacik if (trimming) { 1465e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1466e3e0520bSJosef Bacik /* 1467e3e0520bSJosef Bacik * A concurrent scrub might have added us to the list 1468e3e0520bSJosef Bacik * fs_info->unused_bgs, so use a list_move operation 1469e3e0520bSJosef Bacik * to add the block group to the deleted_bgs list. 1470e3e0520bSJosef Bacik */ 1471e3e0520bSJosef Bacik list_move(&block_group->bg_list, 1472e3e0520bSJosef Bacik &trans->transaction->deleted_bgs); 1473e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1474e3e0520bSJosef Bacik btrfs_get_block_group(block_group); 1475e3e0520bSJosef Bacik } 1476e3e0520bSJosef Bacik end_trans: 1477e3e0520bSJosef Bacik btrfs_end_transaction(trans); 1478e3e0520bSJosef Bacik next: 1479e3e0520bSJosef Bacik mutex_unlock(&fs_info->delete_unused_bgs_mutex); 1480e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1481e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1482e3e0520bSJosef Bacik } 1483e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 14846e80d4f8SDennis Zhou return; 14856e80d4f8SDennis Zhou 14866e80d4f8SDennis Zhou flip_async: 14876e80d4f8SDennis Zhou btrfs_end_transaction(trans); 14886e80d4f8SDennis Zhou mutex_unlock(&fs_info->delete_unused_bgs_mutex); 14896e80d4f8SDennis Zhou btrfs_put_block_group(block_group); 14906e80d4f8SDennis Zhou btrfs_discard_punt_unused_bgs_list(fs_info); 1491e3e0520bSJosef Bacik } 1492e3e0520bSJosef Bacik 149332da5386SDavid Sterba void btrfs_mark_bg_unused(struct btrfs_block_group *bg) 1494e3e0520bSJosef Bacik { 1495e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info = bg->fs_info; 1496e3e0520bSJosef Bacik 1497e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1498e3e0520bSJosef Bacik if (list_empty(&bg->bg_list)) { 1499e3e0520bSJosef Bacik btrfs_get_block_group(bg); 1500e3e0520bSJosef Bacik trace_btrfs_add_unused_block_group(bg); 1501e3e0520bSJosef Bacik list_add_tail(&bg->bg_list, &fs_info->unused_bgs); 1502e3e0520bSJosef Bacik } 1503e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1504e3e0520bSJosef Bacik } 15054358d963SJosef Bacik 15064358d963SJosef Bacik static int find_first_block_group(struct btrfs_fs_info *fs_info, 15074358d963SJosef Bacik struct btrfs_path *path, 15084358d963SJosef Bacik struct btrfs_key *key) 15094358d963SJosef Bacik { 15104358d963SJosef Bacik struct btrfs_root *root = fs_info->extent_root; 15114358d963SJosef Bacik int ret = 0; 15124358d963SJosef Bacik struct btrfs_key found_key; 15134358d963SJosef Bacik struct extent_buffer *leaf; 15144358d963SJosef Bacik struct btrfs_block_group_item bg; 15154358d963SJosef Bacik u64 flags; 15164358d963SJosef Bacik int slot; 15174358d963SJosef Bacik 15184358d963SJosef Bacik ret = btrfs_search_slot(NULL, root, key, path, 0, 0); 15194358d963SJosef Bacik if (ret < 0) 15204358d963SJosef Bacik goto out; 15214358d963SJosef Bacik 15224358d963SJosef Bacik while (1) { 15234358d963SJosef Bacik slot = path->slots[0]; 15244358d963SJosef Bacik leaf = path->nodes[0]; 15254358d963SJosef Bacik if (slot >= btrfs_header_nritems(leaf)) { 15264358d963SJosef Bacik ret = btrfs_next_leaf(root, path); 15274358d963SJosef Bacik if (ret == 0) 15284358d963SJosef Bacik continue; 15294358d963SJosef Bacik if (ret < 0) 15304358d963SJosef Bacik goto out; 15314358d963SJosef Bacik break; 15324358d963SJosef Bacik } 15334358d963SJosef Bacik btrfs_item_key_to_cpu(leaf, &found_key, slot); 15344358d963SJosef Bacik 15354358d963SJosef Bacik if (found_key.objectid >= key->objectid && 15364358d963SJosef Bacik found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { 15374358d963SJosef Bacik struct extent_map_tree *em_tree; 15384358d963SJosef Bacik struct extent_map *em; 15394358d963SJosef Bacik 15404358d963SJosef Bacik em_tree = &root->fs_info->mapping_tree; 15414358d963SJosef Bacik read_lock(&em_tree->lock); 15424358d963SJosef Bacik em = lookup_extent_mapping(em_tree, found_key.objectid, 15434358d963SJosef Bacik found_key.offset); 15444358d963SJosef Bacik read_unlock(&em_tree->lock); 15454358d963SJosef Bacik if (!em) { 15464358d963SJosef Bacik btrfs_err(fs_info, 15474358d963SJosef Bacik "logical %llu len %llu found bg but no related chunk", 15484358d963SJosef Bacik found_key.objectid, found_key.offset); 15494358d963SJosef Bacik ret = -ENOENT; 15504358d963SJosef Bacik } else if (em->start != found_key.objectid || 15514358d963SJosef Bacik em->len != found_key.offset) { 15524358d963SJosef Bacik btrfs_err(fs_info, 15534358d963SJosef Bacik "block group %llu len %llu mismatch with chunk %llu len %llu", 15544358d963SJosef Bacik found_key.objectid, found_key.offset, 15554358d963SJosef Bacik em->start, em->len); 15564358d963SJosef Bacik ret = -EUCLEAN; 15574358d963SJosef Bacik } else { 15584358d963SJosef Bacik read_extent_buffer(leaf, &bg, 15594358d963SJosef Bacik btrfs_item_ptr_offset(leaf, slot), 15604358d963SJosef Bacik sizeof(bg)); 1561de0dc456SDavid Sterba flags = btrfs_stack_block_group_flags(&bg) & 15624358d963SJosef Bacik BTRFS_BLOCK_GROUP_TYPE_MASK; 15634358d963SJosef Bacik 15644358d963SJosef Bacik if (flags != (em->map_lookup->type & 15654358d963SJosef Bacik BTRFS_BLOCK_GROUP_TYPE_MASK)) { 15664358d963SJosef Bacik btrfs_err(fs_info, 15674358d963SJosef Bacik "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", 15684358d963SJosef Bacik found_key.objectid, 15694358d963SJosef Bacik found_key.offset, flags, 15704358d963SJosef Bacik (BTRFS_BLOCK_GROUP_TYPE_MASK & 15714358d963SJosef Bacik em->map_lookup->type)); 15724358d963SJosef Bacik ret = -EUCLEAN; 15734358d963SJosef Bacik } else { 15744358d963SJosef Bacik ret = 0; 15754358d963SJosef Bacik } 15764358d963SJosef Bacik } 15774358d963SJosef Bacik free_extent_map(em); 15784358d963SJosef Bacik goto out; 15794358d963SJosef Bacik } 15804358d963SJosef Bacik path->slots[0]++; 15814358d963SJosef Bacik } 15824358d963SJosef Bacik out: 15834358d963SJosef Bacik return ret; 15844358d963SJosef Bacik } 15854358d963SJosef Bacik 15864358d963SJosef Bacik static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) 15874358d963SJosef Bacik { 15884358d963SJosef Bacik u64 extra_flags = chunk_to_extended(flags) & 15894358d963SJosef Bacik BTRFS_EXTENDED_PROFILE_MASK; 15904358d963SJosef Bacik 15914358d963SJosef Bacik write_seqlock(&fs_info->profiles_lock); 15924358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 15934358d963SJosef Bacik fs_info->avail_data_alloc_bits |= extra_flags; 15944358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_METADATA) 15954358d963SJosef Bacik fs_info->avail_metadata_alloc_bits |= extra_flags; 15964358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 15974358d963SJosef Bacik fs_info->avail_system_alloc_bits |= extra_flags; 15984358d963SJosef Bacik write_sequnlock(&fs_info->profiles_lock); 15994358d963SJosef Bacik } 16004358d963SJosef Bacik 160196a14336SNikolay Borisov /** 160296a14336SNikolay Borisov * btrfs_rmap_block - Map a physical disk address to a list of logical addresses 160396a14336SNikolay Borisov * @chunk_start: logical address of block group 160496a14336SNikolay Borisov * @physical: physical address to map to logical addresses 160596a14336SNikolay Borisov * @logical: return array of logical addresses which map to @physical 160696a14336SNikolay Borisov * @naddrs: length of @logical 160796a14336SNikolay Borisov * @stripe_len: size of IO stripe for the given block group 160896a14336SNikolay Borisov * 160996a14336SNikolay Borisov * Maps a particular @physical disk address to a list of @logical addresses. 161096a14336SNikolay Borisov * Used primarily to exclude those portions of a block group that contain super 161196a14336SNikolay Borisov * block copies. 161296a14336SNikolay Borisov */ 161396a14336SNikolay Borisov EXPORT_FOR_TESTS 161496a14336SNikolay Borisov int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, 161596a14336SNikolay Borisov u64 physical, u64 **logical, int *naddrs, int *stripe_len) 161696a14336SNikolay Borisov { 161796a14336SNikolay Borisov struct extent_map *em; 161896a14336SNikolay Borisov struct map_lookup *map; 161996a14336SNikolay Borisov u64 *buf; 162096a14336SNikolay Borisov u64 bytenr; 16211776ad17SNikolay Borisov u64 data_stripe_length; 16221776ad17SNikolay Borisov u64 io_stripe_size; 16231776ad17SNikolay Borisov int i, nr = 0; 16241776ad17SNikolay Borisov int ret = 0; 162596a14336SNikolay Borisov 162696a14336SNikolay Borisov em = btrfs_get_chunk_map(fs_info, chunk_start, 1); 162796a14336SNikolay Borisov if (IS_ERR(em)) 162896a14336SNikolay Borisov return -EIO; 162996a14336SNikolay Borisov 163096a14336SNikolay Borisov map = em->map_lookup; 16311776ad17SNikolay Borisov data_stripe_length = em->len; 16321776ad17SNikolay Borisov io_stripe_size = map->stripe_len; 163396a14336SNikolay Borisov 163496a14336SNikolay Borisov if (map->type & BTRFS_BLOCK_GROUP_RAID10) 16351776ad17SNikolay Borisov data_stripe_length = div_u64(data_stripe_length, 16361776ad17SNikolay Borisov map->num_stripes / map->sub_stripes); 163796a14336SNikolay Borisov else if (map->type & BTRFS_BLOCK_GROUP_RAID0) 16381776ad17SNikolay Borisov data_stripe_length = div_u64(data_stripe_length, map->num_stripes); 163996a14336SNikolay Borisov else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 16401776ad17SNikolay Borisov data_stripe_length = div_u64(data_stripe_length, 16411776ad17SNikolay Borisov nr_data_stripes(map)); 16421776ad17SNikolay Borisov io_stripe_size = map->stripe_len * nr_data_stripes(map); 164396a14336SNikolay Borisov } 164496a14336SNikolay Borisov 164596a14336SNikolay Borisov buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); 16461776ad17SNikolay Borisov if (!buf) { 16471776ad17SNikolay Borisov ret = -ENOMEM; 16481776ad17SNikolay Borisov goto out; 16491776ad17SNikolay Borisov } 165096a14336SNikolay Borisov 165196a14336SNikolay Borisov for (i = 0; i < map->num_stripes; i++) { 16521776ad17SNikolay Borisov bool already_inserted = false; 16531776ad17SNikolay Borisov u64 stripe_nr; 16541776ad17SNikolay Borisov int j; 16551776ad17SNikolay Borisov 16561776ad17SNikolay Borisov if (!in_range(physical, map->stripes[i].physical, 16571776ad17SNikolay Borisov data_stripe_length)) 165896a14336SNikolay Borisov continue; 165996a14336SNikolay Borisov 166096a14336SNikolay Borisov stripe_nr = physical - map->stripes[i].physical; 166196a14336SNikolay Borisov stripe_nr = div64_u64(stripe_nr, map->stripe_len); 166296a14336SNikolay Borisov 166396a14336SNikolay Borisov if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 166496a14336SNikolay Borisov stripe_nr = stripe_nr * map->num_stripes + i; 166596a14336SNikolay Borisov stripe_nr = div_u64(stripe_nr, map->sub_stripes); 166696a14336SNikolay Borisov } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 166796a14336SNikolay Borisov stripe_nr = stripe_nr * map->num_stripes + i; 166896a14336SNikolay Borisov } 166996a14336SNikolay Borisov /* 167096a14336SNikolay Borisov * The remaining case would be for RAID56, multiply by 167196a14336SNikolay Borisov * nr_data_stripes(). Alternatively, just use rmap_len below 167296a14336SNikolay Borisov * instead of map->stripe_len 167396a14336SNikolay Borisov */ 167496a14336SNikolay Borisov 16751776ad17SNikolay Borisov bytenr = chunk_start + stripe_nr * io_stripe_size; 16761776ad17SNikolay Borisov 16771776ad17SNikolay Borisov /* Ensure we don't add duplicate addresses */ 167896a14336SNikolay Borisov for (j = 0; j < nr; j++) { 16791776ad17SNikolay Borisov if (buf[j] == bytenr) { 16801776ad17SNikolay Borisov already_inserted = true; 168196a14336SNikolay Borisov break; 168296a14336SNikolay Borisov } 168396a14336SNikolay Borisov } 16841776ad17SNikolay Borisov 16851776ad17SNikolay Borisov if (!already_inserted) 16861776ad17SNikolay Borisov buf[nr++] = bytenr; 168796a14336SNikolay Borisov } 168896a14336SNikolay Borisov 168996a14336SNikolay Borisov *logical = buf; 169096a14336SNikolay Borisov *naddrs = nr; 16911776ad17SNikolay Borisov *stripe_len = io_stripe_size; 16921776ad17SNikolay Borisov out: 169396a14336SNikolay Borisov free_extent_map(em); 16941776ad17SNikolay Borisov return ret; 169596a14336SNikolay Borisov } 169696a14336SNikolay Borisov 169732da5386SDavid Sterba static int exclude_super_stripes(struct btrfs_block_group *cache) 16984358d963SJosef Bacik { 16994358d963SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 17004358d963SJosef Bacik u64 bytenr; 17014358d963SJosef Bacik u64 *logical; 17024358d963SJosef Bacik int stripe_len; 17034358d963SJosef Bacik int i, nr, ret; 17044358d963SJosef Bacik 1705b3470b5dSDavid Sterba if (cache->start < BTRFS_SUPER_INFO_OFFSET) { 1706b3470b5dSDavid Sterba stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start; 17074358d963SJosef Bacik cache->bytes_super += stripe_len; 1708b3470b5dSDavid Sterba ret = btrfs_add_excluded_extent(fs_info, cache->start, 17094358d963SJosef Bacik stripe_len); 17104358d963SJosef Bacik if (ret) 17114358d963SJosef Bacik return ret; 17124358d963SJosef Bacik } 17134358d963SJosef Bacik 17144358d963SJosef Bacik for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 17154358d963SJosef Bacik bytenr = btrfs_sb_offset(i); 1716b3470b5dSDavid Sterba ret = btrfs_rmap_block(fs_info, cache->start, 17174358d963SJosef Bacik bytenr, &logical, &nr, &stripe_len); 17184358d963SJosef Bacik if (ret) 17194358d963SJosef Bacik return ret; 17204358d963SJosef Bacik 17214358d963SJosef Bacik while (nr--) { 17224358d963SJosef Bacik u64 start, len; 17234358d963SJosef Bacik 1724b3470b5dSDavid Sterba if (logical[nr] > cache->start + cache->length) 17254358d963SJosef Bacik continue; 17264358d963SJosef Bacik 1727b3470b5dSDavid Sterba if (logical[nr] + stripe_len <= cache->start) 17284358d963SJosef Bacik continue; 17294358d963SJosef Bacik 17304358d963SJosef Bacik start = logical[nr]; 1731b3470b5dSDavid Sterba if (start < cache->start) { 1732b3470b5dSDavid Sterba start = cache->start; 17334358d963SJosef Bacik len = (logical[nr] + stripe_len) - start; 17344358d963SJosef Bacik } else { 17354358d963SJosef Bacik len = min_t(u64, stripe_len, 1736b3470b5dSDavid Sterba cache->start + cache->length - start); 17374358d963SJosef Bacik } 17384358d963SJosef Bacik 17394358d963SJosef Bacik cache->bytes_super += len; 17404358d963SJosef Bacik ret = btrfs_add_excluded_extent(fs_info, start, len); 17414358d963SJosef Bacik if (ret) { 17424358d963SJosef Bacik kfree(logical); 17434358d963SJosef Bacik return ret; 17444358d963SJosef Bacik } 17454358d963SJosef Bacik } 17464358d963SJosef Bacik 17474358d963SJosef Bacik kfree(logical); 17484358d963SJosef Bacik } 17494358d963SJosef Bacik return 0; 17504358d963SJosef Bacik } 17514358d963SJosef Bacik 175232da5386SDavid Sterba static void link_block_group(struct btrfs_block_group *cache) 17534358d963SJosef Bacik { 17544358d963SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 17554358d963SJosef Bacik int index = btrfs_bg_flags_to_raid_index(cache->flags); 17564358d963SJosef Bacik bool first = false; 17574358d963SJosef Bacik 17584358d963SJosef Bacik down_write(&space_info->groups_sem); 17594358d963SJosef Bacik if (list_empty(&space_info->block_groups[index])) 17604358d963SJosef Bacik first = true; 17614358d963SJosef Bacik list_add_tail(&cache->list, &space_info->block_groups[index]); 17624358d963SJosef Bacik up_write(&space_info->groups_sem); 17634358d963SJosef Bacik 17644358d963SJosef Bacik if (first) 17654358d963SJosef Bacik btrfs_sysfs_add_block_group_type(cache); 17664358d963SJosef Bacik } 17674358d963SJosef Bacik 176832da5386SDavid Sterba static struct btrfs_block_group *btrfs_create_block_group_cache( 17694358d963SJosef Bacik struct btrfs_fs_info *fs_info, u64 start, u64 size) 17704358d963SJosef Bacik { 177132da5386SDavid Sterba struct btrfs_block_group *cache; 17724358d963SJosef Bacik 17734358d963SJosef Bacik cache = kzalloc(sizeof(*cache), GFP_NOFS); 17744358d963SJosef Bacik if (!cache) 17754358d963SJosef Bacik return NULL; 17764358d963SJosef Bacik 17774358d963SJosef Bacik cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), 17784358d963SJosef Bacik GFP_NOFS); 17794358d963SJosef Bacik if (!cache->free_space_ctl) { 17804358d963SJosef Bacik kfree(cache); 17814358d963SJosef Bacik return NULL; 17824358d963SJosef Bacik } 17834358d963SJosef Bacik 1784b3470b5dSDavid Sterba cache->start = start; 1785b3470b5dSDavid Sterba cache->length = size; 17864358d963SJosef Bacik 17874358d963SJosef Bacik cache->fs_info = fs_info; 17884358d963SJosef Bacik cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); 17894358d963SJosef Bacik set_free_space_tree_thresholds(cache); 17904358d963SJosef Bacik 17916e80d4f8SDennis Zhou cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED; 17926e80d4f8SDennis Zhou 17934358d963SJosef Bacik atomic_set(&cache->count, 1); 17944358d963SJosef Bacik spin_lock_init(&cache->lock); 17954358d963SJosef Bacik init_rwsem(&cache->data_rwsem); 17964358d963SJosef Bacik INIT_LIST_HEAD(&cache->list); 17974358d963SJosef Bacik INIT_LIST_HEAD(&cache->cluster_list); 17984358d963SJosef Bacik INIT_LIST_HEAD(&cache->bg_list); 17994358d963SJosef Bacik INIT_LIST_HEAD(&cache->ro_list); 1800b0643e59SDennis Zhou INIT_LIST_HEAD(&cache->discard_list); 18014358d963SJosef Bacik INIT_LIST_HEAD(&cache->dirty_list); 18024358d963SJosef Bacik INIT_LIST_HEAD(&cache->io_list); 18034358d963SJosef Bacik btrfs_init_free_space_ctl(cache); 18044358d963SJosef Bacik atomic_set(&cache->trimming, 0); 18054358d963SJosef Bacik mutex_init(&cache->free_space_lock); 18064358d963SJosef Bacik btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); 18074358d963SJosef Bacik 18084358d963SJosef Bacik return cache; 18094358d963SJosef Bacik } 18104358d963SJosef Bacik 18114358d963SJosef Bacik /* 18124358d963SJosef Bacik * Iterate all chunks and verify that each of them has the corresponding block 18134358d963SJosef Bacik * group 18144358d963SJosef Bacik */ 18154358d963SJosef Bacik static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) 18164358d963SJosef Bacik { 18174358d963SJosef Bacik struct extent_map_tree *map_tree = &fs_info->mapping_tree; 18184358d963SJosef Bacik struct extent_map *em; 181932da5386SDavid Sterba struct btrfs_block_group *bg; 18204358d963SJosef Bacik u64 start = 0; 18214358d963SJosef Bacik int ret = 0; 18224358d963SJosef Bacik 18234358d963SJosef Bacik while (1) { 18244358d963SJosef Bacik read_lock(&map_tree->lock); 18254358d963SJosef Bacik /* 18264358d963SJosef Bacik * lookup_extent_mapping will return the first extent map 18274358d963SJosef Bacik * intersecting the range, so setting @len to 1 is enough to 18284358d963SJosef Bacik * get the first chunk. 18294358d963SJosef Bacik */ 18304358d963SJosef Bacik em = lookup_extent_mapping(map_tree, start, 1); 18314358d963SJosef Bacik read_unlock(&map_tree->lock); 18324358d963SJosef Bacik if (!em) 18334358d963SJosef Bacik break; 18344358d963SJosef Bacik 18354358d963SJosef Bacik bg = btrfs_lookup_block_group(fs_info, em->start); 18364358d963SJosef Bacik if (!bg) { 18374358d963SJosef Bacik btrfs_err(fs_info, 18384358d963SJosef Bacik "chunk start=%llu len=%llu doesn't have corresponding block group", 18394358d963SJosef Bacik em->start, em->len); 18404358d963SJosef Bacik ret = -EUCLEAN; 18414358d963SJosef Bacik free_extent_map(em); 18424358d963SJosef Bacik break; 18434358d963SJosef Bacik } 1844b3470b5dSDavid Sterba if (bg->start != em->start || bg->length != em->len || 18454358d963SJosef Bacik (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != 18464358d963SJosef Bacik (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 18474358d963SJosef Bacik btrfs_err(fs_info, 18484358d963SJosef Bacik "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", 18494358d963SJosef Bacik em->start, em->len, 18504358d963SJosef Bacik em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, 1851b3470b5dSDavid Sterba bg->start, bg->length, 18524358d963SJosef Bacik bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); 18534358d963SJosef Bacik ret = -EUCLEAN; 18544358d963SJosef Bacik free_extent_map(em); 18554358d963SJosef Bacik btrfs_put_block_group(bg); 18564358d963SJosef Bacik break; 18574358d963SJosef Bacik } 18584358d963SJosef Bacik start = em->start + em->len; 18594358d963SJosef Bacik free_extent_map(em); 18604358d963SJosef Bacik btrfs_put_block_group(bg); 18614358d963SJosef Bacik } 18624358d963SJosef Bacik return ret; 18634358d963SJosef Bacik } 18644358d963SJosef Bacik 1865ffb9e0f0SQu Wenruo static int read_one_block_group(struct btrfs_fs_info *info, 1866ffb9e0f0SQu Wenruo struct btrfs_path *path, 1867d49a2ddbSQu Wenruo const struct btrfs_key *key, 1868ffb9e0f0SQu Wenruo int need_clear) 1869ffb9e0f0SQu Wenruo { 1870ffb9e0f0SQu Wenruo struct extent_buffer *leaf = path->nodes[0]; 187132da5386SDavid Sterba struct btrfs_block_group *cache; 1872ffb9e0f0SQu Wenruo struct btrfs_space_info *space_info; 1873ffb9e0f0SQu Wenruo struct btrfs_block_group_item bgi; 1874ffb9e0f0SQu Wenruo const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS); 1875ffb9e0f0SQu Wenruo int slot = path->slots[0]; 1876ffb9e0f0SQu Wenruo int ret; 1877ffb9e0f0SQu Wenruo 1878d49a2ddbSQu Wenruo ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY); 1879ffb9e0f0SQu Wenruo 1880d49a2ddbSQu Wenruo cache = btrfs_create_block_group_cache(info, key->objectid, key->offset); 1881ffb9e0f0SQu Wenruo if (!cache) 1882ffb9e0f0SQu Wenruo return -ENOMEM; 1883ffb9e0f0SQu Wenruo 1884ffb9e0f0SQu Wenruo if (need_clear) { 1885ffb9e0f0SQu Wenruo /* 1886ffb9e0f0SQu Wenruo * When we mount with old space cache, we need to 1887ffb9e0f0SQu Wenruo * set BTRFS_DC_CLEAR and set dirty flag. 1888ffb9e0f0SQu Wenruo * 1889ffb9e0f0SQu Wenruo * a) Setting 'BTRFS_DC_CLEAR' makes sure that we 1890ffb9e0f0SQu Wenruo * truncate the old free space cache inode and 1891ffb9e0f0SQu Wenruo * setup a new one. 1892ffb9e0f0SQu Wenruo * b) Setting 'dirty flag' makes sure that we flush 1893ffb9e0f0SQu Wenruo * the new space cache info onto disk. 1894ffb9e0f0SQu Wenruo */ 1895ffb9e0f0SQu Wenruo if (btrfs_test_opt(info, SPACE_CACHE)) 1896ffb9e0f0SQu Wenruo cache->disk_cache_state = BTRFS_DC_CLEAR; 1897ffb9e0f0SQu Wenruo } 1898ffb9e0f0SQu Wenruo read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), 1899ffb9e0f0SQu Wenruo sizeof(bgi)); 1900ffb9e0f0SQu Wenruo cache->used = btrfs_stack_block_group_used(&bgi); 1901ffb9e0f0SQu Wenruo cache->flags = btrfs_stack_block_group_flags(&bgi); 1902ffb9e0f0SQu Wenruo if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && 1903ffb9e0f0SQu Wenruo (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { 1904ffb9e0f0SQu Wenruo btrfs_err(info, 1905ffb9e0f0SQu Wenruo "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", 1906ffb9e0f0SQu Wenruo cache->start); 1907ffb9e0f0SQu Wenruo ret = -EINVAL; 1908ffb9e0f0SQu Wenruo goto error; 1909ffb9e0f0SQu Wenruo } 1910ffb9e0f0SQu Wenruo 1911ffb9e0f0SQu Wenruo /* 1912ffb9e0f0SQu Wenruo * We need to exclude the super stripes now so that the space info has 1913ffb9e0f0SQu Wenruo * super bytes accounted for, otherwise we'll think we have more space 1914ffb9e0f0SQu Wenruo * than we actually do. 1915ffb9e0f0SQu Wenruo */ 1916ffb9e0f0SQu Wenruo ret = exclude_super_stripes(cache); 1917ffb9e0f0SQu Wenruo if (ret) { 1918ffb9e0f0SQu Wenruo /* We may have excluded something, so call this just in case. */ 1919ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 1920ffb9e0f0SQu Wenruo goto error; 1921ffb9e0f0SQu Wenruo } 1922ffb9e0f0SQu Wenruo 1923ffb9e0f0SQu Wenruo /* 1924ffb9e0f0SQu Wenruo * Check for two cases, either we are full, and therefore don't need 1925ffb9e0f0SQu Wenruo * to bother with the caching work since we won't find any space, or we 1926ffb9e0f0SQu Wenruo * are empty, and we can just add all the space in and be done with it. 1927ffb9e0f0SQu Wenruo * This saves us _a_lot_ of time, particularly in the full case. 1928ffb9e0f0SQu Wenruo */ 1929d49a2ddbSQu Wenruo if (key->offset == cache->used) { 1930ffb9e0f0SQu Wenruo cache->last_byte_to_unpin = (u64)-1; 1931ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 1932ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 1933ffb9e0f0SQu Wenruo } else if (cache->used == 0) { 1934ffb9e0f0SQu Wenruo cache->last_byte_to_unpin = (u64)-1; 1935ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 1936d49a2ddbSQu Wenruo add_new_free_space(cache, key->objectid, 1937d49a2ddbSQu Wenruo key->objectid + key->offset); 1938ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 1939ffb9e0f0SQu Wenruo } 1940ffb9e0f0SQu Wenruo 1941ffb9e0f0SQu Wenruo ret = btrfs_add_block_group_cache(info, cache); 1942ffb9e0f0SQu Wenruo if (ret) { 1943ffb9e0f0SQu Wenruo btrfs_remove_free_space_cache(cache); 1944ffb9e0f0SQu Wenruo goto error; 1945ffb9e0f0SQu Wenruo } 1946ffb9e0f0SQu Wenruo trace_btrfs_add_block_group(info, cache, 0); 1947d49a2ddbSQu Wenruo btrfs_update_space_info(info, cache->flags, key->offset, 1948ffb9e0f0SQu Wenruo cache->used, cache->bytes_super, &space_info); 1949ffb9e0f0SQu Wenruo 1950ffb9e0f0SQu Wenruo cache->space_info = space_info; 1951ffb9e0f0SQu Wenruo 1952ffb9e0f0SQu Wenruo link_block_group(cache); 1953ffb9e0f0SQu Wenruo 1954ffb9e0f0SQu Wenruo set_avail_alloc_bits(info, cache->flags); 1955ffb9e0f0SQu Wenruo if (btrfs_chunk_readonly(info, cache->start)) { 1956ffb9e0f0SQu Wenruo inc_block_group_ro(cache, 1); 1957ffb9e0f0SQu Wenruo } else if (cache->used == 0) { 1958ffb9e0f0SQu Wenruo ASSERT(list_empty(&cache->bg_list)); 19596e80d4f8SDennis Zhou if (btrfs_test_opt(info, DISCARD_ASYNC)) 19606e80d4f8SDennis Zhou btrfs_discard_queue_work(&info->discard_ctl, cache); 19616e80d4f8SDennis Zhou else 1962ffb9e0f0SQu Wenruo btrfs_mark_bg_unused(cache); 1963ffb9e0f0SQu Wenruo } 1964ffb9e0f0SQu Wenruo return 0; 1965ffb9e0f0SQu Wenruo error: 1966ffb9e0f0SQu Wenruo btrfs_put_block_group(cache); 1967ffb9e0f0SQu Wenruo return ret; 1968ffb9e0f0SQu Wenruo } 1969ffb9e0f0SQu Wenruo 19704358d963SJosef Bacik int btrfs_read_block_groups(struct btrfs_fs_info *info) 19714358d963SJosef Bacik { 19724358d963SJosef Bacik struct btrfs_path *path; 19734358d963SJosef Bacik int ret; 197432da5386SDavid Sterba struct btrfs_block_group *cache; 19754358d963SJosef Bacik struct btrfs_space_info *space_info; 19764358d963SJosef Bacik struct btrfs_key key; 19774358d963SJosef Bacik int need_clear = 0; 19784358d963SJosef Bacik u64 cache_gen; 19794358d963SJosef Bacik 19804358d963SJosef Bacik key.objectid = 0; 19814358d963SJosef Bacik key.offset = 0; 19824358d963SJosef Bacik key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 19834358d963SJosef Bacik path = btrfs_alloc_path(); 19844358d963SJosef Bacik if (!path) 19854358d963SJosef Bacik return -ENOMEM; 19864358d963SJosef Bacik path->reada = READA_FORWARD; 19874358d963SJosef Bacik 19884358d963SJosef Bacik cache_gen = btrfs_super_cache_generation(info->super_copy); 19894358d963SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 19904358d963SJosef Bacik btrfs_super_generation(info->super_copy) != cache_gen) 19914358d963SJosef Bacik need_clear = 1; 19924358d963SJosef Bacik if (btrfs_test_opt(info, CLEAR_CACHE)) 19934358d963SJosef Bacik need_clear = 1; 19944358d963SJosef Bacik 19954358d963SJosef Bacik while (1) { 19964358d963SJosef Bacik ret = find_first_block_group(info, path, &key); 19974358d963SJosef Bacik if (ret > 0) 19984358d963SJosef Bacik break; 19994358d963SJosef Bacik if (ret != 0) 20004358d963SJosef Bacik goto error; 20014358d963SJosef Bacik 2002ffb9e0f0SQu Wenruo btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2003d49a2ddbSQu Wenruo ret = read_one_block_group(info, path, &key, need_clear); 2004ffb9e0f0SQu Wenruo if (ret < 0) 20054358d963SJosef Bacik goto error; 2006ffb9e0f0SQu Wenruo key.objectid += key.offset; 2007ffb9e0f0SQu Wenruo key.offset = 0; 20084358d963SJosef Bacik btrfs_release_path(path); 20094358d963SJosef Bacik } 20104358d963SJosef Bacik 201129566c9cSMadhuparna Bhowmik rcu_read_lock(); 20124358d963SJosef Bacik list_for_each_entry_rcu(space_info, &info->space_info, list) { 20134358d963SJosef Bacik if (!(btrfs_get_alloc_profile(info, space_info->flags) & 20144358d963SJosef Bacik (BTRFS_BLOCK_GROUP_RAID10 | 20154358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID1_MASK | 20164358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID56_MASK | 20174358d963SJosef Bacik BTRFS_BLOCK_GROUP_DUP))) 20184358d963SJosef Bacik continue; 20194358d963SJosef Bacik /* 20204358d963SJosef Bacik * Avoid allocating from un-mirrored block group if there are 20214358d963SJosef Bacik * mirrored block groups. 20224358d963SJosef Bacik */ 20234358d963SJosef Bacik list_for_each_entry(cache, 20244358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_RAID0], 20254358d963SJosef Bacik list) 2026e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 20274358d963SJosef Bacik list_for_each_entry(cache, 20284358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_SINGLE], 20294358d963SJosef Bacik list) 2030e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 20314358d963SJosef Bacik } 203229566c9cSMadhuparna Bhowmik rcu_read_unlock(); 20334358d963SJosef Bacik 20344358d963SJosef Bacik btrfs_init_global_block_rsv(info); 20354358d963SJosef Bacik ret = check_chunk_block_group_mappings(info); 20364358d963SJosef Bacik error: 20374358d963SJosef Bacik btrfs_free_path(path); 20384358d963SJosef Bacik return ret; 20394358d963SJosef Bacik } 20404358d963SJosef Bacik 20414358d963SJosef Bacik void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) 20424358d963SJosef Bacik { 20434358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 204432da5386SDavid Sterba struct btrfs_block_group *block_group; 20454358d963SJosef Bacik struct btrfs_root *extent_root = fs_info->extent_root; 20464358d963SJosef Bacik struct btrfs_block_group_item item; 20474358d963SJosef Bacik struct btrfs_key key; 20484358d963SJosef Bacik int ret = 0; 20494358d963SJosef Bacik 20504358d963SJosef Bacik if (!trans->can_flush_pending_bgs) 20514358d963SJosef Bacik return; 20524358d963SJosef Bacik 20534358d963SJosef Bacik while (!list_empty(&trans->new_bgs)) { 20544358d963SJosef Bacik block_group = list_first_entry(&trans->new_bgs, 205532da5386SDavid Sterba struct btrfs_block_group, 20564358d963SJosef Bacik bg_list); 20574358d963SJosef Bacik if (ret) 20584358d963SJosef Bacik goto next; 20594358d963SJosef Bacik 20604358d963SJosef Bacik spin_lock(&block_group->lock); 2061de0dc456SDavid Sterba btrfs_set_stack_block_group_used(&item, block_group->used); 2062de0dc456SDavid Sterba btrfs_set_stack_block_group_chunk_objectid(&item, 20633d976388SDavid Sterba BTRFS_FIRST_CHUNK_TREE_OBJECTID); 2064de0dc456SDavid Sterba btrfs_set_stack_block_group_flags(&item, block_group->flags); 2065b3470b5dSDavid Sterba key.objectid = block_group->start; 2066b3470b5dSDavid Sterba key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 2067b3470b5dSDavid Sterba key.offset = block_group->length; 20684358d963SJosef Bacik spin_unlock(&block_group->lock); 20694358d963SJosef Bacik 20704358d963SJosef Bacik ret = btrfs_insert_item(trans, extent_root, &key, &item, 20714358d963SJosef Bacik sizeof(item)); 20724358d963SJosef Bacik if (ret) 20734358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 20744358d963SJosef Bacik ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset); 20754358d963SJosef Bacik if (ret) 20764358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 20774358d963SJosef Bacik add_block_group_free_space(trans, block_group); 20784358d963SJosef Bacik /* Already aborted the transaction if it failed. */ 20794358d963SJosef Bacik next: 20804358d963SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 20814358d963SJosef Bacik list_del_init(&block_group->bg_list); 20824358d963SJosef Bacik } 20834358d963SJosef Bacik btrfs_trans_release_chunk_metadata(trans); 20844358d963SJosef Bacik } 20854358d963SJosef Bacik 20864358d963SJosef Bacik int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, 20874358d963SJosef Bacik u64 type, u64 chunk_offset, u64 size) 20884358d963SJosef Bacik { 20894358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 209032da5386SDavid Sterba struct btrfs_block_group *cache; 20914358d963SJosef Bacik int ret; 20924358d963SJosef Bacik 20934358d963SJosef Bacik btrfs_set_log_full_commit(trans); 20944358d963SJosef Bacik 20954358d963SJosef Bacik cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); 20964358d963SJosef Bacik if (!cache) 20974358d963SJosef Bacik return -ENOMEM; 20984358d963SJosef Bacik 2099bf38be65SDavid Sterba cache->used = bytes_used; 21004358d963SJosef Bacik cache->flags = type; 21014358d963SJosef Bacik cache->last_byte_to_unpin = (u64)-1; 21024358d963SJosef Bacik cache->cached = BTRFS_CACHE_FINISHED; 21034358d963SJosef Bacik cache->needs_free_space = 1; 21044358d963SJosef Bacik ret = exclude_super_stripes(cache); 21054358d963SJosef Bacik if (ret) { 21064358d963SJosef Bacik /* We may have excluded something, so call this just in case */ 21074358d963SJosef Bacik btrfs_free_excluded_extents(cache); 21084358d963SJosef Bacik btrfs_put_block_group(cache); 21094358d963SJosef Bacik return ret; 21104358d963SJosef Bacik } 21114358d963SJosef Bacik 21124358d963SJosef Bacik add_new_free_space(cache, chunk_offset, chunk_offset + size); 21134358d963SJosef Bacik 21144358d963SJosef Bacik btrfs_free_excluded_extents(cache); 21154358d963SJosef Bacik 21164358d963SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 21174358d963SJosef Bacik if (btrfs_should_fragment_free_space(cache)) { 21184358d963SJosef Bacik u64 new_bytes_used = size - bytes_used; 21194358d963SJosef Bacik 21204358d963SJosef Bacik bytes_used += new_bytes_used >> 1; 2121e11c0406SJosef Bacik fragment_free_space(cache); 21224358d963SJosef Bacik } 21234358d963SJosef Bacik #endif 21244358d963SJosef Bacik /* 21254358d963SJosef Bacik * Ensure the corresponding space_info object is created and 21264358d963SJosef Bacik * assigned to our block group. We want our bg to be added to the rbtree 21274358d963SJosef Bacik * with its ->space_info set. 21284358d963SJosef Bacik */ 21294358d963SJosef Bacik cache->space_info = btrfs_find_space_info(fs_info, cache->flags); 21304358d963SJosef Bacik ASSERT(cache->space_info); 21314358d963SJosef Bacik 21324358d963SJosef Bacik ret = btrfs_add_block_group_cache(fs_info, cache); 21334358d963SJosef Bacik if (ret) { 21344358d963SJosef Bacik btrfs_remove_free_space_cache(cache); 21354358d963SJosef Bacik btrfs_put_block_group(cache); 21364358d963SJosef Bacik return ret; 21374358d963SJosef Bacik } 21384358d963SJosef Bacik 21394358d963SJosef Bacik /* 21404358d963SJosef Bacik * Now that our block group has its ->space_info set and is inserted in 21414358d963SJosef Bacik * the rbtree, update the space info's counters. 21424358d963SJosef Bacik */ 21434358d963SJosef Bacik trace_btrfs_add_block_group(fs_info, cache, 1); 21444358d963SJosef Bacik btrfs_update_space_info(fs_info, cache->flags, size, bytes_used, 21454358d963SJosef Bacik cache->bytes_super, &cache->space_info); 21464358d963SJosef Bacik btrfs_update_global_block_rsv(fs_info); 21474358d963SJosef Bacik 21484358d963SJosef Bacik link_block_group(cache); 21494358d963SJosef Bacik 21504358d963SJosef Bacik list_add_tail(&cache->bg_list, &trans->new_bgs); 21514358d963SJosef Bacik trans->delayed_ref_updates++; 21524358d963SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 21534358d963SJosef Bacik 21544358d963SJosef Bacik set_avail_alloc_bits(fs_info, type); 21554358d963SJosef Bacik return 0; 21564358d963SJosef Bacik } 215726ce2095SJosef Bacik 215826ce2095SJosef Bacik static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) 215926ce2095SJosef Bacik { 216026ce2095SJosef Bacik u64 num_devices; 216126ce2095SJosef Bacik u64 stripped; 216226ce2095SJosef Bacik 216326ce2095SJosef Bacik /* 216426ce2095SJosef Bacik * if restripe for this chunk_type is on pick target profile and 216526ce2095SJosef Bacik * return, otherwise do the usual balance 216626ce2095SJosef Bacik */ 2167e11c0406SJosef Bacik stripped = get_restripe_target(fs_info, flags); 216826ce2095SJosef Bacik if (stripped) 216926ce2095SJosef Bacik return extended_to_chunk(stripped); 217026ce2095SJosef Bacik 217126ce2095SJosef Bacik num_devices = fs_info->fs_devices->rw_devices; 217226ce2095SJosef Bacik 217326ce2095SJosef Bacik stripped = BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID56_MASK | 217426ce2095SJosef Bacik BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10; 217526ce2095SJosef Bacik 217626ce2095SJosef Bacik if (num_devices == 1) { 217726ce2095SJosef Bacik stripped |= BTRFS_BLOCK_GROUP_DUP; 217826ce2095SJosef Bacik stripped = flags & ~stripped; 217926ce2095SJosef Bacik 218026ce2095SJosef Bacik /* turn raid0 into single device chunks */ 218126ce2095SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_RAID0) 218226ce2095SJosef Bacik return stripped; 218326ce2095SJosef Bacik 218426ce2095SJosef Bacik /* turn mirroring into duplication */ 218526ce2095SJosef Bacik if (flags & (BTRFS_BLOCK_GROUP_RAID1_MASK | 218626ce2095SJosef Bacik BTRFS_BLOCK_GROUP_RAID10)) 218726ce2095SJosef Bacik return stripped | BTRFS_BLOCK_GROUP_DUP; 218826ce2095SJosef Bacik } else { 218926ce2095SJosef Bacik /* they already had raid on here, just return */ 219026ce2095SJosef Bacik if (flags & stripped) 219126ce2095SJosef Bacik return flags; 219226ce2095SJosef Bacik 219326ce2095SJosef Bacik stripped |= BTRFS_BLOCK_GROUP_DUP; 219426ce2095SJosef Bacik stripped = flags & ~stripped; 219526ce2095SJosef Bacik 219626ce2095SJosef Bacik /* switch duplicated blocks with raid1 */ 219726ce2095SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DUP) 219826ce2095SJosef Bacik return stripped | BTRFS_BLOCK_GROUP_RAID1; 219926ce2095SJosef Bacik 220026ce2095SJosef Bacik /* this is drive concat, leave it alone */ 220126ce2095SJosef Bacik } 220226ce2095SJosef Bacik 220326ce2095SJosef Bacik return flags; 220426ce2095SJosef Bacik } 220526ce2095SJosef Bacik 2206b12de528SQu Wenruo /* 2207b12de528SQu Wenruo * Mark one block group RO, can be called several times for the same block 2208b12de528SQu Wenruo * group. 2209b12de528SQu Wenruo * 2210b12de528SQu Wenruo * @cache: the destination block group 2211b12de528SQu Wenruo * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to 2212b12de528SQu Wenruo * ensure we still have some free space after marking this 2213b12de528SQu Wenruo * block group RO. 2214b12de528SQu Wenruo */ 2215b12de528SQu Wenruo int btrfs_inc_block_group_ro(struct btrfs_block_group *cache, 2216b12de528SQu Wenruo bool do_chunk_alloc) 221726ce2095SJosef Bacik { 221826ce2095SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 221926ce2095SJosef Bacik struct btrfs_trans_handle *trans; 222026ce2095SJosef Bacik u64 alloc_flags; 222126ce2095SJosef Bacik int ret; 222226ce2095SJosef Bacik 222326ce2095SJosef Bacik again: 222426ce2095SJosef Bacik trans = btrfs_join_transaction(fs_info->extent_root); 222526ce2095SJosef Bacik if (IS_ERR(trans)) 222626ce2095SJosef Bacik return PTR_ERR(trans); 222726ce2095SJosef Bacik 222826ce2095SJosef Bacik /* 222926ce2095SJosef Bacik * we're not allowed to set block groups readonly after the dirty 223026ce2095SJosef Bacik * block groups cache has started writing. If it already started, 223126ce2095SJosef Bacik * back off and let this transaction commit 223226ce2095SJosef Bacik */ 223326ce2095SJosef Bacik mutex_lock(&fs_info->ro_block_group_mutex); 223426ce2095SJosef Bacik if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { 223526ce2095SJosef Bacik u64 transid = trans->transid; 223626ce2095SJosef Bacik 223726ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 223826ce2095SJosef Bacik btrfs_end_transaction(trans); 223926ce2095SJosef Bacik 224026ce2095SJosef Bacik ret = btrfs_wait_for_commit(fs_info, transid); 224126ce2095SJosef Bacik if (ret) 224226ce2095SJosef Bacik return ret; 224326ce2095SJosef Bacik goto again; 224426ce2095SJosef Bacik } 224526ce2095SJosef Bacik 2246b12de528SQu Wenruo if (do_chunk_alloc) { 224726ce2095SJosef Bacik /* 2248b12de528SQu Wenruo * If we are changing raid levels, try to allocate a 2249b12de528SQu Wenruo * corresponding block group with the new raid level. 225026ce2095SJosef Bacik */ 225126ce2095SJosef Bacik alloc_flags = update_block_group_flags(fs_info, cache->flags); 225226ce2095SJosef Bacik if (alloc_flags != cache->flags) { 2253b12de528SQu Wenruo ret = btrfs_chunk_alloc(trans, alloc_flags, 2254b12de528SQu Wenruo CHUNK_ALLOC_FORCE); 225526ce2095SJosef Bacik /* 225626ce2095SJosef Bacik * ENOSPC is allowed here, we may have enough space 2257b12de528SQu Wenruo * already allocated at the new raid level to carry on 225826ce2095SJosef Bacik */ 225926ce2095SJosef Bacik if (ret == -ENOSPC) 226026ce2095SJosef Bacik ret = 0; 226126ce2095SJosef Bacik if (ret < 0) 226226ce2095SJosef Bacik goto out; 226326ce2095SJosef Bacik } 2264b12de528SQu Wenruo } 226526ce2095SJosef Bacik 2266a7a63accSJosef Bacik ret = inc_block_group_ro(cache, 0); 2267b12de528SQu Wenruo if (!do_chunk_alloc) 2268b12de528SQu Wenruo goto unlock_out; 226926ce2095SJosef Bacik if (!ret) 227026ce2095SJosef Bacik goto out; 227126ce2095SJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); 227226ce2095SJosef Bacik ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 227326ce2095SJosef Bacik if (ret < 0) 227426ce2095SJosef Bacik goto out; 2275e11c0406SJosef Bacik ret = inc_block_group_ro(cache, 0); 227626ce2095SJosef Bacik out: 227726ce2095SJosef Bacik if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { 227826ce2095SJosef Bacik alloc_flags = update_block_group_flags(fs_info, cache->flags); 227926ce2095SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 228026ce2095SJosef Bacik check_system_chunk(trans, alloc_flags); 228126ce2095SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 228226ce2095SJosef Bacik } 2283b12de528SQu Wenruo unlock_out: 228426ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 228526ce2095SJosef Bacik 228626ce2095SJosef Bacik btrfs_end_transaction(trans); 228726ce2095SJosef Bacik return ret; 228826ce2095SJosef Bacik } 228926ce2095SJosef Bacik 229032da5386SDavid Sterba void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) 229126ce2095SJosef Bacik { 229226ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 229326ce2095SJosef Bacik u64 num_bytes; 229426ce2095SJosef Bacik 229526ce2095SJosef Bacik BUG_ON(!cache->ro); 229626ce2095SJosef Bacik 229726ce2095SJosef Bacik spin_lock(&sinfo->lock); 229826ce2095SJosef Bacik spin_lock(&cache->lock); 229926ce2095SJosef Bacik if (!--cache->ro) { 2300b3470b5dSDavid Sterba num_bytes = cache->length - cache->reserved - 2301bf38be65SDavid Sterba cache->pinned - cache->bytes_super - cache->used; 230226ce2095SJosef Bacik sinfo->bytes_readonly -= num_bytes; 230326ce2095SJosef Bacik list_del_init(&cache->ro_list); 230426ce2095SJosef Bacik } 230526ce2095SJosef Bacik spin_unlock(&cache->lock); 230626ce2095SJosef Bacik spin_unlock(&sinfo->lock); 230726ce2095SJosef Bacik } 230877745c05SJosef Bacik 230977745c05SJosef Bacik static int write_one_cache_group(struct btrfs_trans_handle *trans, 231077745c05SJosef Bacik struct btrfs_path *path, 231132da5386SDavid Sterba struct btrfs_block_group *cache) 231277745c05SJosef Bacik { 231377745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 231477745c05SJosef Bacik int ret; 231577745c05SJosef Bacik struct btrfs_root *extent_root = fs_info->extent_root; 231677745c05SJosef Bacik unsigned long bi; 231777745c05SJosef Bacik struct extent_buffer *leaf; 2318bf38be65SDavid Sterba struct btrfs_block_group_item bgi; 2319b3470b5dSDavid Sterba struct btrfs_key key; 232077745c05SJosef Bacik 2321b3470b5dSDavid Sterba key.objectid = cache->start; 2322b3470b5dSDavid Sterba key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 2323b3470b5dSDavid Sterba key.offset = cache->length; 2324b3470b5dSDavid Sterba 2325b3470b5dSDavid Sterba ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1); 232677745c05SJosef Bacik if (ret) { 232777745c05SJosef Bacik if (ret > 0) 232877745c05SJosef Bacik ret = -ENOENT; 232977745c05SJosef Bacik goto fail; 233077745c05SJosef Bacik } 233177745c05SJosef Bacik 233277745c05SJosef Bacik leaf = path->nodes[0]; 233377745c05SJosef Bacik bi = btrfs_item_ptr_offset(leaf, path->slots[0]); 2334de0dc456SDavid Sterba btrfs_set_stack_block_group_used(&bgi, cache->used); 2335de0dc456SDavid Sterba btrfs_set_stack_block_group_chunk_objectid(&bgi, 23363d976388SDavid Sterba BTRFS_FIRST_CHUNK_TREE_OBJECTID); 2337de0dc456SDavid Sterba btrfs_set_stack_block_group_flags(&bgi, cache->flags); 2338bf38be65SDavid Sterba write_extent_buffer(leaf, &bgi, bi, sizeof(bgi)); 233977745c05SJosef Bacik btrfs_mark_buffer_dirty(leaf); 234077745c05SJosef Bacik fail: 234177745c05SJosef Bacik btrfs_release_path(path); 234277745c05SJosef Bacik return ret; 234377745c05SJosef Bacik 234477745c05SJosef Bacik } 234577745c05SJosef Bacik 234632da5386SDavid Sterba static int cache_save_setup(struct btrfs_block_group *block_group, 234777745c05SJosef Bacik struct btrfs_trans_handle *trans, 234877745c05SJosef Bacik struct btrfs_path *path) 234977745c05SJosef Bacik { 235077745c05SJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 235177745c05SJosef Bacik struct btrfs_root *root = fs_info->tree_root; 235277745c05SJosef Bacik struct inode *inode = NULL; 235377745c05SJosef Bacik struct extent_changeset *data_reserved = NULL; 235477745c05SJosef Bacik u64 alloc_hint = 0; 235577745c05SJosef Bacik int dcs = BTRFS_DC_ERROR; 235677745c05SJosef Bacik u64 num_pages = 0; 235777745c05SJosef Bacik int retries = 0; 235877745c05SJosef Bacik int ret = 0; 235977745c05SJosef Bacik 236077745c05SJosef Bacik /* 236177745c05SJosef Bacik * If this block group is smaller than 100 megs don't bother caching the 236277745c05SJosef Bacik * block group. 236377745c05SJosef Bacik */ 2364b3470b5dSDavid Sterba if (block_group->length < (100 * SZ_1M)) { 236577745c05SJosef Bacik spin_lock(&block_group->lock); 236677745c05SJosef Bacik block_group->disk_cache_state = BTRFS_DC_WRITTEN; 236777745c05SJosef Bacik spin_unlock(&block_group->lock); 236877745c05SJosef Bacik return 0; 236977745c05SJosef Bacik } 237077745c05SJosef Bacik 2371bf31f87fSDavid Sterba if (TRANS_ABORTED(trans)) 237277745c05SJosef Bacik return 0; 237377745c05SJosef Bacik again: 237477745c05SJosef Bacik inode = lookup_free_space_inode(block_group, path); 237577745c05SJosef Bacik if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { 237677745c05SJosef Bacik ret = PTR_ERR(inode); 237777745c05SJosef Bacik btrfs_release_path(path); 237877745c05SJosef Bacik goto out; 237977745c05SJosef Bacik } 238077745c05SJosef Bacik 238177745c05SJosef Bacik if (IS_ERR(inode)) { 238277745c05SJosef Bacik BUG_ON(retries); 238377745c05SJosef Bacik retries++; 238477745c05SJosef Bacik 238577745c05SJosef Bacik if (block_group->ro) 238677745c05SJosef Bacik goto out_free; 238777745c05SJosef Bacik 238877745c05SJosef Bacik ret = create_free_space_inode(trans, block_group, path); 238977745c05SJosef Bacik if (ret) 239077745c05SJosef Bacik goto out_free; 239177745c05SJosef Bacik goto again; 239277745c05SJosef Bacik } 239377745c05SJosef Bacik 239477745c05SJosef Bacik /* 239577745c05SJosef Bacik * We want to set the generation to 0, that way if anything goes wrong 239677745c05SJosef Bacik * from here on out we know not to trust this cache when we load up next 239777745c05SJosef Bacik * time. 239877745c05SJosef Bacik */ 239977745c05SJosef Bacik BTRFS_I(inode)->generation = 0; 240077745c05SJosef Bacik ret = btrfs_update_inode(trans, root, inode); 240177745c05SJosef Bacik if (ret) { 240277745c05SJosef Bacik /* 240377745c05SJosef Bacik * So theoretically we could recover from this, simply set the 240477745c05SJosef Bacik * super cache generation to 0 so we know to invalidate the 240577745c05SJosef Bacik * cache, but then we'd have to keep track of the block groups 240677745c05SJosef Bacik * that fail this way so we know we _have_ to reset this cache 240777745c05SJosef Bacik * before the next commit or risk reading stale cache. So to 240877745c05SJosef Bacik * limit our exposure to horrible edge cases lets just abort the 240977745c05SJosef Bacik * transaction, this only happens in really bad situations 241077745c05SJosef Bacik * anyway. 241177745c05SJosef Bacik */ 241277745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 241377745c05SJosef Bacik goto out_put; 241477745c05SJosef Bacik } 241577745c05SJosef Bacik WARN_ON(ret); 241677745c05SJosef Bacik 241777745c05SJosef Bacik /* We've already setup this transaction, go ahead and exit */ 241877745c05SJosef Bacik if (block_group->cache_generation == trans->transid && 241977745c05SJosef Bacik i_size_read(inode)) { 242077745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 242177745c05SJosef Bacik goto out_put; 242277745c05SJosef Bacik } 242377745c05SJosef Bacik 242477745c05SJosef Bacik if (i_size_read(inode) > 0) { 242577745c05SJosef Bacik ret = btrfs_check_trunc_cache_free_space(fs_info, 242677745c05SJosef Bacik &fs_info->global_block_rsv); 242777745c05SJosef Bacik if (ret) 242877745c05SJosef Bacik goto out_put; 242977745c05SJosef Bacik 243077745c05SJosef Bacik ret = btrfs_truncate_free_space_cache(trans, NULL, inode); 243177745c05SJosef Bacik if (ret) 243277745c05SJosef Bacik goto out_put; 243377745c05SJosef Bacik } 243477745c05SJosef Bacik 243577745c05SJosef Bacik spin_lock(&block_group->lock); 243677745c05SJosef Bacik if (block_group->cached != BTRFS_CACHE_FINISHED || 243777745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) { 243877745c05SJosef Bacik /* 243977745c05SJosef Bacik * don't bother trying to write stuff out _if_ 244077745c05SJosef Bacik * a) we're not cached, 244177745c05SJosef Bacik * b) we're with nospace_cache mount option, 244277745c05SJosef Bacik * c) we're with v2 space_cache (FREE_SPACE_TREE). 244377745c05SJosef Bacik */ 244477745c05SJosef Bacik dcs = BTRFS_DC_WRITTEN; 244577745c05SJosef Bacik spin_unlock(&block_group->lock); 244677745c05SJosef Bacik goto out_put; 244777745c05SJosef Bacik } 244877745c05SJosef Bacik spin_unlock(&block_group->lock); 244977745c05SJosef Bacik 245077745c05SJosef Bacik /* 245177745c05SJosef Bacik * We hit an ENOSPC when setting up the cache in this transaction, just 245277745c05SJosef Bacik * skip doing the setup, we've already cleared the cache so we're safe. 245377745c05SJosef Bacik */ 245477745c05SJosef Bacik if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { 245577745c05SJosef Bacik ret = -ENOSPC; 245677745c05SJosef Bacik goto out_put; 245777745c05SJosef Bacik } 245877745c05SJosef Bacik 245977745c05SJosef Bacik /* 246077745c05SJosef Bacik * Try to preallocate enough space based on how big the block group is. 246177745c05SJosef Bacik * Keep in mind this has to include any pinned space which could end up 246277745c05SJosef Bacik * taking up quite a bit since it's not folded into the other space 246377745c05SJosef Bacik * cache. 246477745c05SJosef Bacik */ 2465b3470b5dSDavid Sterba num_pages = div_u64(block_group->length, SZ_256M); 246677745c05SJosef Bacik if (!num_pages) 246777745c05SJosef Bacik num_pages = 1; 246877745c05SJosef Bacik 246977745c05SJosef Bacik num_pages *= 16; 247077745c05SJosef Bacik num_pages *= PAGE_SIZE; 247177745c05SJosef Bacik 247277745c05SJosef Bacik ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); 247377745c05SJosef Bacik if (ret) 247477745c05SJosef Bacik goto out_put; 247577745c05SJosef Bacik 247677745c05SJosef Bacik ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, 247777745c05SJosef Bacik num_pages, num_pages, 247877745c05SJosef Bacik &alloc_hint); 247977745c05SJosef Bacik /* 248077745c05SJosef Bacik * Our cache requires contiguous chunks so that we don't modify a bunch 248177745c05SJosef Bacik * of metadata or split extents when writing the cache out, which means 248277745c05SJosef Bacik * we can enospc if we are heavily fragmented in addition to just normal 248377745c05SJosef Bacik * out of space conditions. So if we hit this just skip setting up any 248477745c05SJosef Bacik * other block groups for this transaction, maybe we'll unpin enough 248577745c05SJosef Bacik * space the next time around. 248677745c05SJosef Bacik */ 248777745c05SJosef Bacik if (!ret) 248877745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 248977745c05SJosef Bacik else if (ret == -ENOSPC) 249077745c05SJosef Bacik set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); 249177745c05SJosef Bacik 249277745c05SJosef Bacik out_put: 249377745c05SJosef Bacik iput(inode); 249477745c05SJosef Bacik out_free: 249577745c05SJosef Bacik btrfs_release_path(path); 249677745c05SJosef Bacik out: 249777745c05SJosef Bacik spin_lock(&block_group->lock); 249877745c05SJosef Bacik if (!ret && dcs == BTRFS_DC_SETUP) 249977745c05SJosef Bacik block_group->cache_generation = trans->transid; 250077745c05SJosef Bacik block_group->disk_cache_state = dcs; 250177745c05SJosef Bacik spin_unlock(&block_group->lock); 250277745c05SJosef Bacik 250377745c05SJosef Bacik extent_changeset_free(data_reserved); 250477745c05SJosef Bacik return ret; 250577745c05SJosef Bacik } 250677745c05SJosef Bacik 250777745c05SJosef Bacik int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) 250877745c05SJosef Bacik { 250977745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 251032da5386SDavid Sterba struct btrfs_block_group *cache, *tmp; 251177745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 251277745c05SJosef Bacik struct btrfs_path *path; 251377745c05SJosef Bacik 251477745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs) || 251577745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) 251677745c05SJosef Bacik return 0; 251777745c05SJosef Bacik 251877745c05SJosef Bacik path = btrfs_alloc_path(); 251977745c05SJosef Bacik if (!path) 252077745c05SJosef Bacik return -ENOMEM; 252177745c05SJosef Bacik 252277745c05SJosef Bacik /* Could add new block groups, use _safe just in case */ 252377745c05SJosef Bacik list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, 252477745c05SJosef Bacik dirty_list) { 252577745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_CLEAR) 252677745c05SJosef Bacik cache_save_setup(cache, trans, path); 252777745c05SJosef Bacik } 252877745c05SJosef Bacik 252977745c05SJosef Bacik btrfs_free_path(path); 253077745c05SJosef Bacik return 0; 253177745c05SJosef Bacik } 253277745c05SJosef Bacik 253377745c05SJosef Bacik /* 253477745c05SJosef Bacik * Transaction commit does final block group cache writeback during a critical 253577745c05SJosef Bacik * section where nothing is allowed to change the FS. This is required in 253677745c05SJosef Bacik * order for the cache to actually match the block group, but can introduce a 253777745c05SJosef Bacik * lot of latency into the commit. 253877745c05SJosef Bacik * 253977745c05SJosef Bacik * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. 254077745c05SJosef Bacik * There's a chance we'll have to redo some of it if the block group changes 254177745c05SJosef Bacik * again during the commit, but it greatly reduces the commit latency by 254277745c05SJosef Bacik * getting rid of the easy block groups while we're still allowing others to 254377745c05SJosef Bacik * join the commit. 254477745c05SJosef Bacik */ 254577745c05SJosef Bacik int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) 254677745c05SJosef Bacik { 254777745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 254832da5386SDavid Sterba struct btrfs_block_group *cache; 254977745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 255077745c05SJosef Bacik int ret = 0; 255177745c05SJosef Bacik int should_put; 255277745c05SJosef Bacik struct btrfs_path *path = NULL; 255377745c05SJosef Bacik LIST_HEAD(dirty); 255477745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 255577745c05SJosef Bacik int num_started = 0; 255677745c05SJosef Bacik int loops = 0; 255777745c05SJosef Bacik 255877745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 255977745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs)) { 256077745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 256177745c05SJosef Bacik return 0; 256277745c05SJosef Bacik } 256377745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 256477745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 256577745c05SJosef Bacik 256677745c05SJosef Bacik again: 256777745c05SJosef Bacik /* Make sure all the block groups on our dirty list actually exist */ 256877745c05SJosef Bacik btrfs_create_pending_block_groups(trans); 256977745c05SJosef Bacik 257077745c05SJosef Bacik if (!path) { 257177745c05SJosef Bacik path = btrfs_alloc_path(); 257277745c05SJosef Bacik if (!path) 257377745c05SJosef Bacik return -ENOMEM; 257477745c05SJosef Bacik } 257577745c05SJosef Bacik 257677745c05SJosef Bacik /* 257777745c05SJosef Bacik * cache_write_mutex is here only to save us from balance or automatic 257877745c05SJosef Bacik * removal of empty block groups deleting this block group while we are 257977745c05SJosef Bacik * writing out the cache 258077745c05SJosef Bacik */ 258177745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 258277745c05SJosef Bacik while (!list_empty(&dirty)) { 258377745c05SJosef Bacik bool drop_reserve = true; 258477745c05SJosef Bacik 258532da5386SDavid Sterba cache = list_first_entry(&dirty, struct btrfs_block_group, 258677745c05SJosef Bacik dirty_list); 258777745c05SJosef Bacik /* 258877745c05SJosef Bacik * This can happen if something re-dirties a block group that 258977745c05SJosef Bacik * is already under IO. Just wait for it to finish and then do 259077745c05SJosef Bacik * it all again 259177745c05SJosef Bacik */ 259277745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 259377745c05SJosef Bacik list_del_init(&cache->io_list); 259477745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 259577745c05SJosef Bacik btrfs_put_block_group(cache); 259677745c05SJosef Bacik } 259777745c05SJosef Bacik 259877745c05SJosef Bacik 259977745c05SJosef Bacik /* 260077745c05SJosef Bacik * btrfs_wait_cache_io uses the cache->dirty_list to decide if 260177745c05SJosef Bacik * it should update the cache_state. Don't delete until after 260277745c05SJosef Bacik * we wait. 260377745c05SJosef Bacik * 260477745c05SJosef Bacik * Since we're not running in the commit critical section 260577745c05SJosef Bacik * we need the dirty_bgs_lock to protect from update_block_group 260677745c05SJosef Bacik */ 260777745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 260877745c05SJosef Bacik list_del_init(&cache->dirty_list); 260977745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 261077745c05SJosef Bacik 261177745c05SJosef Bacik should_put = 1; 261277745c05SJosef Bacik 261377745c05SJosef Bacik cache_save_setup(cache, trans, path); 261477745c05SJosef Bacik 261577745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_SETUP) { 261677745c05SJosef Bacik cache->io_ctl.inode = NULL; 261777745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 261877745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 261977745c05SJosef Bacik num_started++; 262077745c05SJosef Bacik should_put = 0; 262177745c05SJosef Bacik 262277745c05SJosef Bacik /* 262377745c05SJosef Bacik * The cache_write_mutex is protecting the 262477745c05SJosef Bacik * io_list, also refer to the definition of 262577745c05SJosef Bacik * btrfs_transaction::io_bgs for more details 262677745c05SJosef Bacik */ 262777745c05SJosef Bacik list_add_tail(&cache->io_list, io); 262877745c05SJosef Bacik } else { 262977745c05SJosef Bacik /* 263077745c05SJosef Bacik * If we failed to write the cache, the 263177745c05SJosef Bacik * generation will be bad and life goes on 263277745c05SJosef Bacik */ 263377745c05SJosef Bacik ret = 0; 263477745c05SJosef Bacik } 263577745c05SJosef Bacik } 263677745c05SJosef Bacik if (!ret) { 263777745c05SJosef Bacik ret = write_one_cache_group(trans, path, cache); 263877745c05SJosef Bacik /* 263977745c05SJosef Bacik * Our block group might still be attached to the list 264077745c05SJosef Bacik * of new block groups in the transaction handle of some 264177745c05SJosef Bacik * other task (struct btrfs_trans_handle->new_bgs). This 264277745c05SJosef Bacik * means its block group item isn't yet in the extent 264377745c05SJosef Bacik * tree. If this happens ignore the error, as we will 264477745c05SJosef Bacik * try again later in the critical section of the 264577745c05SJosef Bacik * transaction commit. 264677745c05SJosef Bacik */ 264777745c05SJosef Bacik if (ret == -ENOENT) { 264877745c05SJosef Bacik ret = 0; 264977745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 265077745c05SJosef Bacik if (list_empty(&cache->dirty_list)) { 265177745c05SJosef Bacik list_add_tail(&cache->dirty_list, 265277745c05SJosef Bacik &cur_trans->dirty_bgs); 265377745c05SJosef Bacik btrfs_get_block_group(cache); 265477745c05SJosef Bacik drop_reserve = false; 265577745c05SJosef Bacik } 265677745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 265777745c05SJosef Bacik } else if (ret) { 265877745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 265977745c05SJosef Bacik } 266077745c05SJosef Bacik } 266177745c05SJosef Bacik 266277745c05SJosef Bacik /* If it's not on the io list, we need to put the block group */ 266377745c05SJosef Bacik if (should_put) 266477745c05SJosef Bacik btrfs_put_block_group(cache); 266577745c05SJosef Bacik if (drop_reserve) 266677745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 266777745c05SJosef Bacik 266877745c05SJosef Bacik if (ret) 266977745c05SJosef Bacik break; 267077745c05SJosef Bacik 267177745c05SJosef Bacik /* 267277745c05SJosef Bacik * Avoid blocking other tasks for too long. It might even save 267377745c05SJosef Bacik * us from writing caches for block groups that are going to be 267477745c05SJosef Bacik * removed. 267577745c05SJosef Bacik */ 267677745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 267777745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 267877745c05SJosef Bacik } 267977745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 268077745c05SJosef Bacik 268177745c05SJosef Bacik /* 268277745c05SJosef Bacik * Go through delayed refs for all the stuff we've just kicked off 268377745c05SJosef Bacik * and then loop back (just once) 268477745c05SJosef Bacik */ 268577745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 0); 268677745c05SJosef Bacik if (!ret && loops == 0) { 268777745c05SJosef Bacik loops++; 268877745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 268977745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 269077745c05SJosef Bacik /* 269177745c05SJosef Bacik * dirty_bgs_lock protects us from concurrent block group 269277745c05SJosef Bacik * deletes too (not just cache_write_mutex). 269377745c05SJosef Bacik */ 269477745c05SJosef Bacik if (!list_empty(&dirty)) { 269577745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 269677745c05SJosef Bacik goto again; 269777745c05SJosef Bacik } 269877745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 269977745c05SJosef Bacik } else if (ret < 0) { 270077745c05SJosef Bacik btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 270177745c05SJosef Bacik } 270277745c05SJosef Bacik 270377745c05SJosef Bacik btrfs_free_path(path); 270477745c05SJosef Bacik return ret; 270577745c05SJosef Bacik } 270677745c05SJosef Bacik 270777745c05SJosef Bacik int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) 270877745c05SJosef Bacik { 270977745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 271032da5386SDavid Sterba struct btrfs_block_group *cache; 271177745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 271277745c05SJosef Bacik int ret = 0; 271377745c05SJosef Bacik int should_put; 271477745c05SJosef Bacik struct btrfs_path *path; 271577745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 271677745c05SJosef Bacik int num_started = 0; 271777745c05SJosef Bacik 271877745c05SJosef Bacik path = btrfs_alloc_path(); 271977745c05SJosef Bacik if (!path) 272077745c05SJosef Bacik return -ENOMEM; 272177745c05SJosef Bacik 272277745c05SJosef Bacik /* 272377745c05SJosef Bacik * Even though we are in the critical section of the transaction commit, 272477745c05SJosef Bacik * we can still have concurrent tasks adding elements to this 272577745c05SJosef Bacik * transaction's list of dirty block groups. These tasks correspond to 272677745c05SJosef Bacik * endio free space workers started when writeback finishes for a 272777745c05SJosef Bacik * space cache, which run inode.c:btrfs_finish_ordered_io(), and can 272877745c05SJosef Bacik * allocate new block groups as a result of COWing nodes of the root 272977745c05SJosef Bacik * tree when updating the free space inode. The writeback for the space 273077745c05SJosef Bacik * caches is triggered by an earlier call to 273177745c05SJosef Bacik * btrfs_start_dirty_block_groups() and iterations of the following 273277745c05SJosef Bacik * loop. 273377745c05SJosef Bacik * Also we want to do the cache_save_setup first and then run the 273477745c05SJosef Bacik * delayed refs to make sure we have the best chance at doing this all 273577745c05SJosef Bacik * in one shot. 273677745c05SJosef Bacik */ 273777745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 273877745c05SJosef Bacik while (!list_empty(&cur_trans->dirty_bgs)) { 273977745c05SJosef Bacik cache = list_first_entry(&cur_trans->dirty_bgs, 274032da5386SDavid Sterba struct btrfs_block_group, 274177745c05SJosef Bacik dirty_list); 274277745c05SJosef Bacik 274377745c05SJosef Bacik /* 274477745c05SJosef Bacik * This can happen if cache_save_setup re-dirties a block group 274577745c05SJosef Bacik * that is already under IO. Just wait for it to finish and 274677745c05SJosef Bacik * then do it all again 274777745c05SJosef Bacik */ 274877745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 274977745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 275077745c05SJosef Bacik list_del_init(&cache->io_list); 275177745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 275277745c05SJosef Bacik btrfs_put_block_group(cache); 275377745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 275477745c05SJosef Bacik } 275577745c05SJosef Bacik 275677745c05SJosef Bacik /* 275777745c05SJosef Bacik * Don't remove from the dirty list until after we've waited on 275877745c05SJosef Bacik * any pending IO 275977745c05SJosef Bacik */ 276077745c05SJosef Bacik list_del_init(&cache->dirty_list); 276177745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 276277745c05SJosef Bacik should_put = 1; 276377745c05SJosef Bacik 276477745c05SJosef Bacik cache_save_setup(cache, trans, path); 276577745c05SJosef Bacik 276677745c05SJosef Bacik if (!ret) 276777745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 276877745c05SJosef Bacik (unsigned long) -1); 276977745c05SJosef Bacik 277077745c05SJosef Bacik if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { 277177745c05SJosef Bacik cache->io_ctl.inode = NULL; 277277745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 277377745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 277477745c05SJosef Bacik num_started++; 277577745c05SJosef Bacik should_put = 0; 277677745c05SJosef Bacik list_add_tail(&cache->io_list, io); 277777745c05SJosef Bacik } else { 277877745c05SJosef Bacik /* 277977745c05SJosef Bacik * If we failed to write the cache, the 278077745c05SJosef Bacik * generation will be bad and life goes on 278177745c05SJosef Bacik */ 278277745c05SJosef Bacik ret = 0; 278377745c05SJosef Bacik } 278477745c05SJosef Bacik } 278577745c05SJosef Bacik if (!ret) { 278677745c05SJosef Bacik ret = write_one_cache_group(trans, path, cache); 278777745c05SJosef Bacik /* 278877745c05SJosef Bacik * One of the free space endio workers might have 278977745c05SJosef Bacik * created a new block group while updating a free space 279077745c05SJosef Bacik * cache's inode (at inode.c:btrfs_finish_ordered_io()) 279177745c05SJosef Bacik * and hasn't released its transaction handle yet, in 279277745c05SJosef Bacik * which case the new block group is still attached to 279377745c05SJosef Bacik * its transaction handle and its creation has not 279477745c05SJosef Bacik * finished yet (no block group item in the extent tree 279577745c05SJosef Bacik * yet, etc). If this is the case, wait for all free 279677745c05SJosef Bacik * space endio workers to finish and retry. This is a 279777745c05SJosef Bacik * a very rare case so no need for a more efficient and 279877745c05SJosef Bacik * complex approach. 279977745c05SJosef Bacik */ 280077745c05SJosef Bacik if (ret == -ENOENT) { 280177745c05SJosef Bacik wait_event(cur_trans->writer_wait, 280277745c05SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 280377745c05SJosef Bacik ret = write_one_cache_group(trans, path, cache); 280477745c05SJosef Bacik } 280577745c05SJosef Bacik if (ret) 280677745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 280777745c05SJosef Bacik } 280877745c05SJosef Bacik 280977745c05SJosef Bacik /* If its not on the io list, we need to put the block group */ 281077745c05SJosef Bacik if (should_put) 281177745c05SJosef Bacik btrfs_put_block_group(cache); 281277745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 281377745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 281477745c05SJosef Bacik } 281577745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 281677745c05SJosef Bacik 281777745c05SJosef Bacik /* 281877745c05SJosef Bacik * Refer to the definition of io_bgs member for details why it's safe 281977745c05SJosef Bacik * to use it without any locking 282077745c05SJosef Bacik */ 282177745c05SJosef Bacik while (!list_empty(io)) { 282232da5386SDavid Sterba cache = list_first_entry(io, struct btrfs_block_group, 282377745c05SJosef Bacik io_list); 282477745c05SJosef Bacik list_del_init(&cache->io_list); 282577745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 282677745c05SJosef Bacik btrfs_put_block_group(cache); 282777745c05SJosef Bacik } 282877745c05SJosef Bacik 282977745c05SJosef Bacik btrfs_free_path(path); 283077745c05SJosef Bacik return ret; 283177745c05SJosef Bacik } 2832606d1bf1SJosef Bacik 2833606d1bf1SJosef Bacik int btrfs_update_block_group(struct btrfs_trans_handle *trans, 2834606d1bf1SJosef Bacik u64 bytenr, u64 num_bytes, int alloc) 2835606d1bf1SJosef Bacik { 2836606d1bf1SJosef Bacik struct btrfs_fs_info *info = trans->fs_info; 283732da5386SDavid Sterba struct btrfs_block_group *cache = NULL; 2838606d1bf1SJosef Bacik u64 total = num_bytes; 2839606d1bf1SJosef Bacik u64 old_val; 2840606d1bf1SJosef Bacik u64 byte_in_group; 2841606d1bf1SJosef Bacik int factor; 2842606d1bf1SJosef Bacik int ret = 0; 2843606d1bf1SJosef Bacik 2844606d1bf1SJosef Bacik /* Block accounting for super block */ 2845606d1bf1SJosef Bacik spin_lock(&info->delalloc_root_lock); 2846606d1bf1SJosef Bacik old_val = btrfs_super_bytes_used(info->super_copy); 2847606d1bf1SJosef Bacik if (alloc) 2848606d1bf1SJosef Bacik old_val += num_bytes; 2849606d1bf1SJosef Bacik else 2850606d1bf1SJosef Bacik old_val -= num_bytes; 2851606d1bf1SJosef Bacik btrfs_set_super_bytes_used(info->super_copy, old_val); 2852606d1bf1SJosef Bacik spin_unlock(&info->delalloc_root_lock); 2853606d1bf1SJosef Bacik 2854606d1bf1SJosef Bacik while (total) { 2855606d1bf1SJosef Bacik cache = btrfs_lookup_block_group(info, bytenr); 2856606d1bf1SJosef Bacik if (!cache) { 2857606d1bf1SJosef Bacik ret = -ENOENT; 2858606d1bf1SJosef Bacik break; 2859606d1bf1SJosef Bacik } 2860606d1bf1SJosef Bacik factor = btrfs_bg_type_to_factor(cache->flags); 2861606d1bf1SJosef Bacik 2862606d1bf1SJosef Bacik /* 2863606d1bf1SJosef Bacik * If this block group has free space cache written out, we 2864606d1bf1SJosef Bacik * need to make sure to load it if we are removing space. This 2865606d1bf1SJosef Bacik * is because we need the unpinning stage to actually add the 2866606d1bf1SJosef Bacik * space back to the block group, otherwise we will leak space. 2867606d1bf1SJosef Bacik */ 286832da5386SDavid Sterba if (!alloc && !btrfs_block_group_done(cache)) 2869606d1bf1SJosef Bacik btrfs_cache_block_group(cache, 1); 2870606d1bf1SJosef Bacik 2871b3470b5dSDavid Sterba byte_in_group = bytenr - cache->start; 2872b3470b5dSDavid Sterba WARN_ON(byte_in_group > cache->length); 2873606d1bf1SJosef Bacik 2874606d1bf1SJosef Bacik spin_lock(&cache->space_info->lock); 2875606d1bf1SJosef Bacik spin_lock(&cache->lock); 2876606d1bf1SJosef Bacik 2877606d1bf1SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 2878606d1bf1SJosef Bacik cache->disk_cache_state < BTRFS_DC_CLEAR) 2879606d1bf1SJosef Bacik cache->disk_cache_state = BTRFS_DC_CLEAR; 2880606d1bf1SJosef Bacik 2881bf38be65SDavid Sterba old_val = cache->used; 2882b3470b5dSDavid Sterba num_bytes = min(total, cache->length - byte_in_group); 2883606d1bf1SJosef Bacik if (alloc) { 2884606d1bf1SJosef Bacik old_val += num_bytes; 2885bf38be65SDavid Sterba cache->used = old_val; 2886606d1bf1SJosef Bacik cache->reserved -= num_bytes; 2887606d1bf1SJosef Bacik cache->space_info->bytes_reserved -= num_bytes; 2888606d1bf1SJosef Bacik cache->space_info->bytes_used += num_bytes; 2889606d1bf1SJosef Bacik cache->space_info->disk_used += num_bytes * factor; 2890606d1bf1SJosef Bacik spin_unlock(&cache->lock); 2891606d1bf1SJosef Bacik spin_unlock(&cache->space_info->lock); 2892606d1bf1SJosef Bacik } else { 2893606d1bf1SJosef Bacik old_val -= num_bytes; 2894bf38be65SDavid Sterba cache->used = old_val; 2895606d1bf1SJosef Bacik cache->pinned += num_bytes; 2896606d1bf1SJosef Bacik btrfs_space_info_update_bytes_pinned(info, 2897606d1bf1SJosef Bacik cache->space_info, num_bytes); 2898606d1bf1SJosef Bacik cache->space_info->bytes_used -= num_bytes; 2899606d1bf1SJosef Bacik cache->space_info->disk_used -= num_bytes * factor; 2900606d1bf1SJosef Bacik spin_unlock(&cache->lock); 2901606d1bf1SJosef Bacik spin_unlock(&cache->space_info->lock); 2902606d1bf1SJosef Bacik 2903606d1bf1SJosef Bacik percpu_counter_add_batch( 2904606d1bf1SJosef Bacik &cache->space_info->total_bytes_pinned, 2905606d1bf1SJosef Bacik num_bytes, 2906606d1bf1SJosef Bacik BTRFS_TOTAL_BYTES_PINNED_BATCH); 2907fe119a6eSNikolay Borisov set_extent_dirty(&trans->transaction->pinned_extents, 2908606d1bf1SJosef Bacik bytenr, bytenr + num_bytes - 1, 2909606d1bf1SJosef Bacik GFP_NOFS | __GFP_NOFAIL); 2910606d1bf1SJosef Bacik } 2911606d1bf1SJosef Bacik 2912606d1bf1SJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 2913606d1bf1SJosef Bacik if (list_empty(&cache->dirty_list)) { 2914606d1bf1SJosef Bacik list_add_tail(&cache->dirty_list, 2915606d1bf1SJosef Bacik &trans->transaction->dirty_bgs); 2916606d1bf1SJosef Bacik trans->delayed_ref_updates++; 2917606d1bf1SJosef Bacik btrfs_get_block_group(cache); 2918606d1bf1SJosef Bacik } 2919606d1bf1SJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 2920606d1bf1SJosef Bacik 2921606d1bf1SJosef Bacik /* 2922606d1bf1SJosef Bacik * No longer have used bytes in this block group, queue it for 2923606d1bf1SJosef Bacik * deletion. We do this after adding the block group to the 2924606d1bf1SJosef Bacik * dirty list to avoid races between cleaner kthread and space 2925606d1bf1SJosef Bacik * cache writeout. 2926606d1bf1SJosef Bacik */ 29276e80d4f8SDennis Zhou if (!alloc && old_val == 0) { 29286e80d4f8SDennis Zhou if (!btrfs_test_opt(info, DISCARD_ASYNC)) 2929606d1bf1SJosef Bacik btrfs_mark_bg_unused(cache); 29306e80d4f8SDennis Zhou } 2931606d1bf1SJosef Bacik 2932606d1bf1SJosef Bacik btrfs_put_block_group(cache); 2933606d1bf1SJosef Bacik total -= num_bytes; 2934606d1bf1SJosef Bacik bytenr += num_bytes; 2935606d1bf1SJosef Bacik } 2936606d1bf1SJosef Bacik 2937606d1bf1SJosef Bacik /* Modified block groups are accounted for in the delayed_refs_rsv. */ 2938606d1bf1SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 2939606d1bf1SJosef Bacik return ret; 2940606d1bf1SJosef Bacik } 2941606d1bf1SJosef Bacik 2942606d1bf1SJosef Bacik /** 2943606d1bf1SJosef Bacik * btrfs_add_reserved_bytes - update the block_group and space info counters 2944606d1bf1SJosef Bacik * @cache: The cache we are manipulating 2945606d1bf1SJosef Bacik * @ram_bytes: The number of bytes of file content, and will be same to 2946606d1bf1SJosef Bacik * @num_bytes except for the compress path. 2947606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 2948606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 2949606d1bf1SJosef Bacik * 2950606d1bf1SJosef Bacik * This is called by the allocator when it reserves space. If this is a 2951606d1bf1SJosef Bacik * reservation and the block group has become read only we cannot make the 2952606d1bf1SJosef Bacik * reservation and return -EAGAIN, otherwise this function always succeeds. 2953606d1bf1SJosef Bacik */ 295432da5386SDavid Sterba int btrfs_add_reserved_bytes(struct btrfs_block_group *cache, 2955606d1bf1SJosef Bacik u64 ram_bytes, u64 num_bytes, int delalloc) 2956606d1bf1SJosef Bacik { 2957606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 2958606d1bf1SJosef Bacik int ret = 0; 2959606d1bf1SJosef Bacik 2960606d1bf1SJosef Bacik spin_lock(&space_info->lock); 2961606d1bf1SJosef Bacik spin_lock(&cache->lock); 2962606d1bf1SJosef Bacik if (cache->ro) { 2963606d1bf1SJosef Bacik ret = -EAGAIN; 2964606d1bf1SJosef Bacik } else { 2965606d1bf1SJosef Bacik cache->reserved += num_bytes; 2966606d1bf1SJosef Bacik space_info->bytes_reserved += num_bytes; 2967a43c3835SJosef Bacik trace_btrfs_space_reservation(cache->fs_info, "space_info", 2968a43c3835SJosef Bacik space_info->flags, num_bytes, 1); 2969606d1bf1SJosef Bacik btrfs_space_info_update_bytes_may_use(cache->fs_info, 2970606d1bf1SJosef Bacik space_info, -ram_bytes); 2971606d1bf1SJosef Bacik if (delalloc) 2972606d1bf1SJosef Bacik cache->delalloc_bytes += num_bytes; 2973606d1bf1SJosef Bacik } 2974606d1bf1SJosef Bacik spin_unlock(&cache->lock); 2975606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 2976606d1bf1SJosef Bacik return ret; 2977606d1bf1SJosef Bacik } 2978606d1bf1SJosef Bacik 2979606d1bf1SJosef Bacik /** 2980606d1bf1SJosef Bacik * btrfs_free_reserved_bytes - update the block_group and space info counters 2981606d1bf1SJosef Bacik * @cache: The cache we are manipulating 2982606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 2983606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 2984606d1bf1SJosef Bacik * 2985606d1bf1SJosef Bacik * This is called by somebody who is freeing space that was never actually used 2986606d1bf1SJosef Bacik * on disk. For example if you reserve some space for a new leaf in transaction 2987606d1bf1SJosef Bacik * A and before transaction A commits you free that leaf, you call this with 2988606d1bf1SJosef Bacik * reserve set to 0 in order to clear the reservation. 2989606d1bf1SJosef Bacik */ 299032da5386SDavid Sterba void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, 2991606d1bf1SJosef Bacik u64 num_bytes, int delalloc) 2992606d1bf1SJosef Bacik { 2993606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 2994606d1bf1SJosef Bacik 2995606d1bf1SJosef Bacik spin_lock(&space_info->lock); 2996606d1bf1SJosef Bacik spin_lock(&cache->lock); 2997606d1bf1SJosef Bacik if (cache->ro) 2998606d1bf1SJosef Bacik space_info->bytes_readonly += num_bytes; 2999606d1bf1SJosef Bacik cache->reserved -= num_bytes; 3000606d1bf1SJosef Bacik space_info->bytes_reserved -= num_bytes; 3001606d1bf1SJosef Bacik space_info->max_extent_size = 0; 3002606d1bf1SJosef Bacik 3003606d1bf1SJosef Bacik if (delalloc) 3004606d1bf1SJosef Bacik cache->delalloc_bytes -= num_bytes; 3005606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3006606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 3007606d1bf1SJosef Bacik } 300807730d87SJosef Bacik 300907730d87SJosef Bacik static void force_metadata_allocation(struct btrfs_fs_info *info) 301007730d87SJosef Bacik { 301107730d87SJosef Bacik struct list_head *head = &info->space_info; 301207730d87SJosef Bacik struct btrfs_space_info *found; 301307730d87SJosef Bacik 301407730d87SJosef Bacik rcu_read_lock(); 301507730d87SJosef Bacik list_for_each_entry_rcu(found, head, list) { 301607730d87SJosef Bacik if (found->flags & BTRFS_BLOCK_GROUP_METADATA) 301707730d87SJosef Bacik found->force_alloc = CHUNK_ALLOC_FORCE; 301807730d87SJosef Bacik } 301907730d87SJosef Bacik rcu_read_unlock(); 302007730d87SJosef Bacik } 302107730d87SJosef Bacik 302207730d87SJosef Bacik static int should_alloc_chunk(struct btrfs_fs_info *fs_info, 302307730d87SJosef Bacik struct btrfs_space_info *sinfo, int force) 302407730d87SJosef Bacik { 302507730d87SJosef Bacik u64 bytes_used = btrfs_space_info_used(sinfo, false); 302607730d87SJosef Bacik u64 thresh; 302707730d87SJosef Bacik 302807730d87SJosef Bacik if (force == CHUNK_ALLOC_FORCE) 302907730d87SJosef Bacik return 1; 303007730d87SJosef Bacik 303107730d87SJosef Bacik /* 303207730d87SJosef Bacik * in limited mode, we want to have some free space up to 303307730d87SJosef Bacik * about 1% of the FS size. 303407730d87SJosef Bacik */ 303507730d87SJosef Bacik if (force == CHUNK_ALLOC_LIMITED) { 303607730d87SJosef Bacik thresh = btrfs_super_total_bytes(fs_info->super_copy); 303707730d87SJosef Bacik thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); 303807730d87SJosef Bacik 303907730d87SJosef Bacik if (sinfo->total_bytes - bytes_used < thresh) 304007730d87SJosef Bacik return 1; 304107730d87SJosef Bacik } 304207730d87SJosef Bacik 304307730d87SJosef Bacik if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) 304407730d87SJosef Bacik return 0; 304507730d87SJosef Bacik return 1; 304607730d87SJosef Bacik } 304707730d87SJosef Bacik 304807730d87SJosef Bacik int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) 304907730d87SJosef Bacik { 305007730d87SJosef Bacik u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); 305107730d87SJosef Bacik 305207730d87SJosef Bacik return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 305307730d87SJosef Bacik } 305407730d87SJosef Bacik 305507730d87SJosef Bacik /* 305607730d87SJosef Bacik * If force is CHUNK_ALLOC_FORCE: 305707730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 305807730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 305907730d87SJosef Bacik * If force is NOT CHUNK_ALLOC_FORCE: 306007730d87SJosef Bacik * - return 0 if it doesn't need to allocate a new chunk, 306107730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 306207730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 306307730d87SJosef Bacik */ 306407730d87SJosef Bacik int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, 306507730d87SJosef Bacik enum btrfs_chunk_alloc_enum force) 306607730d87SJosef Bacik { 306707730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 306807730d87SJosef Bacik struct btrfs_space_info *space_info; 306907730d87SJosef Bacik bool wait_for_alloc = false; 307007730d87SJosef Bacik bool should_alloc = false; 307107730d87SJosef Bacik int ret = 0; 307207730d87SJosef Bacik 307307730d87SJosef Bacik /* Don't re-enter if we're already allocating a chunk */ 307407730d87SJosef Bacik if (trans->allocating_chunk) 307507730d87SJosef Bacik return -ENOSPC; 307607730d87SJosef Bacik 307707730d87SJosef Bacik space_info = btrfs_find_space_info(fs_info, flags); 307807730d87SJosef Bacik ASSERT(space_info); 307907730d87SJosef Bacik 308007730d87SJosef Bacik do { 308107730d87SJosef Bacik spin_lock(&space_info->lock); 308207730d87SJosef Bacik if (force < space_info->force_alloc) 308307730d87SJosef Bacik force = space_info->force_alloc; 308407730d87SJosef Bacik should_alloc = should_alloc_chunk(fs_info, space_info, force); 308507730d87SJosef Bacik if (space_info->full) { 308607730d87SJosef Bacik /* No more free physical space */ 308707730d87SJosef Bacik if (should_alloc) 308807730d87SJosef Bacik ret = -ENOSPC; 308907730d87SJosef Bacik else 309007730d87SJosef Bacik ret = 0; 309107730d87SJosef Bacik spin_unlock(&space_info->lock); 309207730d87SJosef Bacik return ret; 309307730d87SJosef Bacik } else if (!should_alloc) { 309407730d87SJosef Bacik spin_unlock(&space_info->lock); 309507730d87SJosef Bacik return 0; 309607730d87SJosef Bacik } else if (space_info->chunk_alloc) { 309707730d87SJosef Bacik /* 309807730d87SJosef Bacik * Someone is already allocating, so we need to block 309907730d87SJosef Bacik * until this someone is finished and then loop to 310007730d87SJosef Bacik * recheck if we should continue with our allocation 310107730d87SJosef Bacik * attempt. 310207730d87SJosef Bacik */ 310307730d87SJosef Bacik wait_for_alloc = true; 310407730d87SJosef Bacik spin_unlock(&space_info->lock); 310507730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 310607730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 310707730d87SJosef Bacik } else { 310807730d87SJosef Bacik /* Proceed with allocation */ 310907730d87SJosef Bacik space_info->chunk_alloc = 1; 311007730d87SJosef Bacik wait_for_alloc = false; 311107730d87SJosef Bacik spin_unlock(&space_info->lock); 311207730d87SJosef Bacik } 311307730d87SJosef Bacik 311407730d87SJosef Bacik cond_resched(); 311507730d87SJosef Bacik } while (wait_for_alloc); 311607730d87SJosef Bacik 311707730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 311807730d87SJosef Bacik trans->allocating_chunk = true; 311907730d87SJosef Bacik 312007730d87SJosef Bacik /* 312107730d87SJosef Bacik * If we have mixed data/metadata chunks we want to make sure we keep 312207730d87SJosef Bacik * allocating mixed chunks instead of individual chunks. 312307730d87SJosef Bacik */ 312407730d87SJosef Bacik if (btrfs_mixed_space_info(space_info)) 312507730d87SJosef Bacik flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); 312607730d87SJosef Bacik 312707730d87SJosef Bacik /* 312807730d87SJosef Bacik * if we're doing a data chunk, go ahead and make sure that 312907730d87SJosef Bacik * we keep a reasonable number of metadata chunks allocated in the 313007730d87SJosef Bacik * FS as well. 313107730d87SJosef Bacik */ 313207730d87SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { 313307730d87SJosef Bacik fs_info->data_chunk_allocations++; 313407730d87SJosef Bacik if (!(fs_info->data_chunk_allocations % 313507730d87SJosef Bacik fs_info->metadata_ratio)) 313607730d87SJosef Bacik force_metadata_allocation(fs_info); 313707730d87SJosef Bacik } 313807730d87SJosef Bacik 313907730d87SJosef Bacik /* 314007730d87SJosef Bacik * Check if we have enough space in SYSTEM chunk because we may need 314107730d87SJosef Bacik * to update devices. 314207730d87SJosef Bacik */ 314307730d87SJosef Bacik check_system_chunk(trans, flags); 314407730d87SJosef Bacik 314507730d87SJosef Bacik ret = btrfs_alloc_chunk(trans, flags); 314607730d87SJosef Bacik trans->allocating_chunk = false; 314707730d87SJosef Bacik 314807730d87SJosef Bacik spin_lock(&space_info->lock); 314907730d87SJosef Bacik if (ret < 0) { 315007730d87SJosef Bacik if (ret == -ENOSPC) 315107730d87SJosef Bacik space_info->full = 1; 315207730d87SJosef Bacik else 315307730d87SJosef Bacik goto out; 315407730d87SJosef Bacik } else { 315507730d87SJosef Bacik ret = 1; 315607730d87SJosef Bacik space_info->max_extent_size = 0; 315707730d87SJosef Bacik } 315807730d87SJosef Bacik 315907730d87SJosef Bacik space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; 316007730d87SJosef Bacik out: 316107730d87SJosef Bacik space_info->chunk_alloc = 0; 316207730d87SJosef Bacik spin_unlock(&space_info->lock); 316307730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 316407730d87SJosef Bacik /* 316507730d87SJosef Bacik * When we allocate a new chunk we reserve space in the chunk block 316607730d87SJosef Bacik * reserve to make sure we can COW nodes/leafs in the chunk tree or 316707730d87SJosef Bacik * add new nodes/leafs to it if we end up needing to do it when 316807730d87SJosef Bacik * inserting the chunk item and updating device items as part of the 316907730d87SJosef Bacik * second phase of chunk allocation, performed by 317007730d87SJosef Bacik * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a 317107730d87SJosef Bacik * large number of new block groups to create in our transaction 317207730d87SJosef Bacik * handle's new_bgs list to avoid exhausting the chunk block reserve 317307730d87SJosef Bacik * in extreme cases - like having a single transaction create many new 317407730d87SJosef Bacik * block groups when starting to write out the free space caches of all 317507730d87SJosef Bacik * the block groups that were made dirty during the lifetime of the 317607730d87SJosef Bacik * transaction. 317707730d87SJosef Bacik */ 317807730d87SJosef Bacik if (trans->chunk_bytes_reserved >= (u64)SZ_2M) 317907730d87SJosef Bacik btrfs_create_pending_block_groups(trans); 318007730d87SJosef Bacik 318107730d87SJosef Bacik return ret; 318207730d87SJosef Bacik } 318307730d87SJosef Bacik 318407730d87SJosef Bacik static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) 318507730d87SJosef Bacik { 318607730d87SJosef Bacik u64 num_dev; 318707730d87SJosef Bacik 318807730d87SJosef Bacik num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; 318907730d87SJosef Bacik if (!num_dev) 319007730d87SJosef Bacik num_dev = fs_info->fs_devices->rw_devices; 319107730d87SJosef Bacik 319207730d87SJosef Bacik return num_dev; 319307730d87SJosef Bacik } 319407730d87SJosef Bacik 319507730d87SJosef Bacik /* 3196a9143bd3SMarcos Paulo de Souza * Reserve space in the system space for allocating or removing a chunk 319707730d87SJosef Bacik */ 319807730d87SJosef Bacik void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) 319907730d87SJosef Bacik { 320007730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 320107730d87SJosef Bacik struct btrfs_space_info *info; 320207730d87SJosef Bacik u64 left; 320307730d87SJosef Bacik u64 thresh; 320407730d87SJosef Bacik int ret = 0; 320507730d87SJosef Bacik u64 num_devs; 320607730d87SJosef Bacik 320707730d87SJosef Bacik /* 320807730d87SJosef Bacik * Needed because we can end up allocating a system chunk and for an 320907730d87SJosef Bacik * atomic and race free space reservation in the chunk block reserve. 321007730d87SJosef Bacik */ 321107730d87SJosef Bacik lockdep_assert_held(&fs_info->chunk_mutex); 321207730d87SJosef Bacik 321307730d87SJosef Bacik info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); 321407730d87SJosef Bacik spin_lock(&info->lock); 321507730d87SJosef Bacik left = info->total_bytes - btrfs_space_info_used(info, true); 321607730d87SJosef Bacik spin_unlock(&info->lock); 321707730d87SJosef Bacik 321807730d87SJosef Bacik num_devs = get_profile_num_devs(fs_info, type); 321907730d87SJosef Bacik 322007730d87SJosef Bacik /* num_devs device items to update and 1 chunk item to add or remove */ 32212bd36e7bSJosef Bacik thresh = btrfs_calc_metadata_size(fs_info, num_devs) + 32222bd36e7bSJosef Bacik btrfs_calc_insert_metadata_size(fs_info, 1); 322307730d87SJosef Bacik 322407730d87SJosef Bacik if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { 322507730d87SJosef Bacik btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", 322607730d87SJosef Bacik left, thresh, type); 322707730d87SJosef Bacik btrfs_dump_space_info(fs_info, info, 0, 0); 322807730d87SJosef Bacik } 322907730d87SJosef Bacik 323007730d87SJosef Bacik if (left < thresh) { 323107730d87SJosef Bacik u64 flags = btrfs_system_alloc_profile(fs_info); 323207730d87SJosef Bacik 323307730d87SJosef Bacik /* 323407730d87SJosef Bacik * Ignore failure to create system chunk. We might end up not 323507730d87SJosef Bacik * needing it, as we might not need to COW all nodes/leafs from 323607730d87SJosef Bacik * the paths we visit in the chunk tree (they were already COWed 323707730d87SJosef Bacik * or created in the current transaction for example). 323807730d87SJosef Bacik */ 323907730d87SJosef Bacik ret = btrfs_alloc_chunk(trans, flags); 324007730d87SJosef Bacik } 324107730d87SJosef Bacik 324207730d87SJosef Bacik if (!ret) { 324307730d87SJosef Bacik ret = btrfs_block_rsv_add(fs_info->chunk_root, 324407730d87SJosef Bacik &fs_info->chunk_block_rsv, 324507730d87SJosef Bacik thresh, BTRFS_RESERVE_NO_FLUSH); 324607730d87SJosef Bacik if (!ret) 324707730d87SJosef Bacik trans->chunk_bytes_reserved += thresh; 324807730d87SJosef Bacik } 324907730d87SJosef Bacik } 325007730d87SJosef Bacik 32513e43c279SJosef Bacik void btrfs_put_block_group_cache(struct btrfs_fs_info *info) 32523e43c279SJosef Bacik { 325332da5386SDavid Sterba struct btrfs_block_group *block_group; 32543e43c279SJosef Bacik u64 last = 0; 32553e43c279SJosef Bacik 32563e43c279SJosef Bacik while (1) { 32573e43c279SJosef Bacik struct inode *inode; 32583e43c279SJosef Bacik 32593e43c279SJosef Bacik block_group = btrfs_lookup_first_block_group(info, last); 32603e43c279SJosef Bacik while (block_group) { 32613e43c279SJosef Bacik btrfs_wait_block_group_cache_done(block_group); 32623e43c279SJosef Bacik spin_lock(&block_group->lock); 32633e43c279SJosef Bacik if (block_group->iref) 32643e43c279SJosef Bacik break; 32653e43c279SJosef Bacik spin_unlock(&block_group->lock); 32663e43c279SJosef Bacik block_group = btrfs_next_block_group(block_group); 32673e43c279SJosef Bacik } 32683e43c279SJosef Bacik if (!block_group) { 32693e43c279SJosef Bacik if (last == 0) 32703e43c279SJosef Bacik break; 32713e43c279SJosef Bacik last = 0; 32723e43c279SJosef Bacik continue; 32733e43c279SJosef Bacik } 32743e43c279SJosef Bacik 32753e43c279SJosef Bacik inode = block_group->inode; 32763e43c279SJosef Bacik block_group->iref = 0; 32773e43c279SJosef Bacik block_group->inode = NULL; 32783e43c279SJosef Bacik spin_unlock(&block_group->lock); 32793e43c279SJosef Bacik ASSERT(block_group->io_ctl.inode == NULL); 32803e43c279SJosef Bacik iput(inode); 3281b3470b5dSDavid Sterba last = block_group->start + block_group->length; 32823e43c279SJosef Bacik btrfs_put_block_group(block_group); 32833e43c279SJosef Bacik } 32843e43c279SJosef Bacik } 32853e43c279SJosef Bacik 32863e43c279SJosef Bacik /* 32873e43c279SJosef Bacik * Must be called only after stopping all workers, since we could have block 32883e43c279SJosef Bacik * group caching kthreads running, and therefore they could race with us if we 32893e43c279SJosef Bacik * freed the block groups before stopping them. 32903e43c279SJosef Bacik */ 32913e43c279SJosef Bacik int btrfs_free_block_groups(struct btrfs_fs_info *info) 32923e43c279SJosef Bacik { 329332da5386SDavid Sterba struct btrfs_block_group *block_group; 32943e43c279SJosef Bacik struct btrfs_space_info *space_info; 32953e43c279SJosef Bacik struct btrfs_caching_control *caching_ctl; 32963e43c279SJosef Bacik struct rb_node *n; 32973e43c279SJosef Bacik 32983e43c279SJosef Bacik down_write(&info->commit_root_sem); 32993e43c279SJosef Bacik while (!list_empty(&info->caching_block_groups)) { 33003e43c279SJosef Bacik caching_ctl = list_entry(info->caching_block_groups.next, 33013e43c279SJosef Bacik struct btrfs_caching_control, list); 33023e43c279SJosef Bacik list_del(&caching_ctl->list); 33033e43c279SJosef Bacik btrfs_put_caching_control(caching_ctl); 33043e43c279SJosef Bacik } 33053e43c279SJosef Bacik up_write(&info->commit_root_sem); 33063e43c279SJosef Bacik 33073e43c279SJosef Bacik spin_lock(&info->unused_bgs_lock); 33083e43c279SJosef Bacik while (!list_empty(&info->unused_bgs)) { 33093e43c279SJosef Bacik block_group = list_first_entry(&info->unused_bgs, 331032da5386SDavid Sterba struct btrfs_block_group, 33113e43c279SJosef Bacik bg_list); 33123e43c279SJosef Bacik list_del_init(&block_group->bg_list); 33133e43c279SJosef Bacik btrfs_put_block_group(block_group); 33143e43c279SJosef Bacik } 33153e43c279SJosef Bacik spin_unlock(&info->unused_bgs_lock); 33163e43c279SJosef Bacik 33173e43c279SJosef Bacik spin_lock(&info->block_group_cache_lock); 33183e43c279SJosef Bacik while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { 331932da5386SDavid Sterba block_group = rb_entry(n, struct btrfs_block_group, 33203e43c279SJosef Bacik cache_node); 33213e43c279SJosef Bacik rb_erase(&block_group->cache_node, 33223e43c279SJosef Bacik &info->block_group_cache_tree); 33233e43c279SJosef Bacik RB_CLEAR_NODE(&block_group->cache_node); 33243e43c279SJosef Bacik spin_unlock(&info->block_group_cache_lock); 33253e43c279SJosef Bacik 33263e43c279SJosef Bacik down_write(&block_group->space_info->groups_sem); 33273e43c279SJosef Bacik list_del(&block_group->list); 33283e43c279SJosef Bacik up_write(&block_group->space_info->groups_sem); 33293e43c279SJosef Bacik 33303e43c279SJosef Bacik /* 33313e43c279SJosef Bacik * We haven't cached this block group, which means we could 33323e43c279SJosef Bacik * possibly have excluded extents on this block group. 33333e43c279SJosef Bacik */ 33343e43c279SJosef Bacik if (block_group->cached == BTRFS_CACHE_NO || 33353e43c279SJosef Bacik block_group->cached == BTRFS_CACHE_ERROR) 33363e43c279SJosef Bacik btrfs_free_excluded_extents(block_group); 33373e43c279SJosef Bacik 33383e43c279SJosef Bacik btrfs_remove_free_space_cache(block_group); 33393e43c279SJosef Bacik ASSERT(block_group->cached != BTRFS_CACHE_STARTED); 33403e43c279SJosef Bacik ASSERT(list_empty(&block_group->dirty_list)); 33413e43c279SJosef Bacik ASSERT(list_empty(&block_group->io_list)); 33423e43c279SJosef Bacik ASSERT(list_empty(&block_group->bg_list)); 33433e43c279SJosef Bacik ASSERT(atomic_read(&block_group->count) == 1); 33443e43c279SJosef Bacik btrfs_put_block_group(block_group); 33453e43c279SJosef Bacik 33463e43c279SJosef Bacik spin_lock(&info->block_group_cache_lock); 33473e43c279SJosef Bacik } 33483e43c279SJosef Bacik spin_unlock(&info->block_group_cache_lock); 33493e43c279SJosef Bacik 33503e43c279SJosef Bacik /* 33513e43c279SJosef Bacik * Now that all the block groups are freed, go through and free all the 33523e43c279SJosef Bacik * space_info structs. This is only called during the final stages of 33533e43c279SJosef Bacik * unmount, and so we know nobody is using them. We call 33543e43c279SJosef Bacik * synchronize_rcu() once before we start, just to be on the safe side. 33553e43c279SJosef Bacik */ 33563e43c279SJosef Bacik synchronize_rcu(); 33573e43c279SJosef Bacik 33583e43c279SJosef Bacik btrfs_release_global_block_rsv(info); 33593e43c279SJosef Bacik 33603e43c279SJosef Bacik while (!list_empty(&info->space_info)) { 33613e43c279SJosef Bacik space_info = list_entry(info->space_info.next, 33623e43c279SJosef Bacik struct btrfs_space_info, 33633e43c279SJosef Bacik list); 33643e43c279SJosef Bacik 33653e43c279SJosef Bacik /* 33663e43c279SJosef Bacik * Do not hide this behind enospc_debug, this is actually 33673e43c279SJosef Bacik * important and indicates a real bug if this happens. 33683e43c279SJosef Bacik */ 33693e43c279SJosef Bacik if (WARN_ON(space_info->bytes_pinned > 0 || 33703e43c279SJosef Bacik space_info->bytes_reserved > 0 || 33713e43c279SJosef Bacik space_info->bytes_may_use > 0)) 33723e43c279SJosef Bacik btrfs_dump_space_info(info, space_info, 0, 0); 3373*d611add4SFilipe Manana WARN_ON(space_info->reclaim_size > 0); 33743e43c279SJosef Bacik list_del(&space_info->list); 33753e43c279SJosef Bacik btrfs_sysfs_remove_space_info(space_info); 33763e43c279SJosef Bacik } 33773e43c279SJosef Bacik return 0; 33783e43c279SJosef Bacik } 3379