12e405ad8SJosef Bacik // SPDX-License-Identifier: GPL-2.0 22e405ad8SJosef Bacik 352bb7a21SBoris Burkov #include <linux/sizes.h> 42ca0ec77SJohannes Thumshirn #include <linux/list_sort.h> 5784352feSDavid Sterba #include "misc.h" 62e405ad8SJosef Bacik #include "ctree.h" 72e405ad8SJosef Bacik #include "block-group.h" 83eeb3226SJosef Bacik #include "space-info.h" 99f21246dSJosef Bacik #include "disk-io.h" 109f21246dSJosef Bacik #include "free-space-cache.h" 119f21246dSJosef Bacik #include "free-space-tree.h" 12e3e0520bSJosef Bacik #include "volumes.h" 13e3e0520bSJosef Bacik #include "transaction.h" 14e3e0520bSJosef Bacik #include "ref-verify.h" 154358d963SJosef Bacik #include "sysfs.h" 164358d963SJosef Bacik #include "tree-log.h" 1777745c05SJosef Bacik #include "delalloc-space.h" 18b0643e59SDennis Zhou #include "discard.h" 1996a14336SNikolay Borisov #include "raid56.h" 2008e11a3dSNaohiro Aota #include "zoned.h" 21c7f13d42SJosef Bacik #include "fs.h" 2207e81dc9SJosef Bacik #include "accessors.h" 23a0231804SJosef Bacik #include "extent-tree.h" 242e405ad8SJosef Bacik 2506d61cb1SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 2606d61cb1SJosef Bacik int btrfs_should_fragment_free_space(struct btrfs_block_group *block_group) 2706d61cb1SJosef Bacik { 2806d61cb1SJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 2906d61cb1SJosef Bacik 3006d61cb1SJosef Bacik return (btrfs_test_opt(fs_info, FRAGMENT_METADATA) && 3106d61cb1SJosef Bacik block_group->flags & BTRFS_BLOCK_GROUP_METADATA) || 3206d61cb1SJosef Bacik (btrfs_test_opt(fs_info, FRAGMENT_DATA) && 3306d61cb1SJosef Bacik block_group->flags & BTRFS_BLOCK_GROUP_DATA); 3406d61cb1SJosef Bacik } 3506d61cb1SJosef Bacik #endif 3606d61cb1SJosef Bacik 37878d7b67SJosef Bacik /* 38878d7b67SJosef Bacik * Return target flags in extended format or 0 if restripe for this chunk_type 39878d7b67SJosef Bacik * is not in progress 40878d7b67SJosef Bacik * 41878d7b67SJosef Bacik * Should be called with balance_lock held 42878d7b67SJosef Bacik */ 43e11c0406SJosef Bacik static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) 44878d7b67SJosef Bacik { 45878d7b67SJosef Bacik struct btrfs_balance_control *bctl = fs_info->balance_ctl; 46878d7b67SJosef Bacik u64 target = 0; 47878d7b67SJosef Bacik 48878d7b67SJosef Bacik if (!bctl) 49878d7b67SJosef Bacik return 0; 50878d7b67SJosef Bacik 51878d7b67SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA && 52878d7b67SJosef Bacik bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { 53878d7b67SJosef Bacik target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; 54878d7b67SJosef Bacik } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && 55878d7b67SJosef Bacik bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { 56878d7b67SJosef Bacik target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; 57878d7b67SJosef Bacik } else if (flags & BTRFS_BLOCK_GROUP_METADATA && 58878d7b67SJosef Bacik bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { 59878d7b67SJosef Bacik target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; 60878d7b67SJosef Bacik } 61878d7b67SJosef Bacik 62878d7b67SJosef Bacik return target; 63878d7b67SJosef Bacik } 64878d7b67SJosef Bacik 65878d7b67SJosef Bacik /* 66878d7b67SJosef Bacik * @flags: available profiles in extended format (see ctree.h) 67878d7b67SJosef Bacik * 68878d7b67SJosef Bacik * Return reduced profile in chunk format. If profile changing is in progress 69878d7b67SJosef Bacik * (either running or paused) picks the target profile (if it's already 70878d7b67SJosef Bacik * available), otherwise falls back to plain reducing. 71878d7b67SJosef Bacik */ 72878d7b67SJosef Bacik static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) 73878d7b67SJosef Bacik { 74878d7b67SJosef Bacik u64 num_devices = fs_info->fs_devices->rw_devices; 75878d7b67SJosef Bacik u64 target; 76878d7b67SJosef Bacik u64 raid_type; 77878d7b67SJosef Bacik u64 allowed = 0; 78878d7b67SJosef Bacik 79878d7b67SJosef Bacik /* 80878d7b67SJosef Bacik * See if restripe for this chunk_type is in progress, if so try to 81878d7b67SJosef Bacik * reduce to the target profile 82878d7b67SJosef Bacik */ 83878d7b67SJosef Bacik spin_lock(&fs_info->balance_lock); 84e11c0406SJosef Bacik target = get_restripe_target(fs_info, flags); 85878d7b67SJosef Bacik if (target) { 86878d7b67SJosef Bacik spin_unlock(&fs_info->balance_lock); 87878d7b67SJosef Bacik return extended_to_chunk(target); 88878d7b67SJosef Bacik } 89878d7b67SJosef Bacik spin_unlock(&fs_info->balance_lock); 90878d7b67SJosef Bacik 91878d7b67SJosef Bacik /* First, mask out the RAID levels which aren't possible */ 92878d7b67SJosef Bacik for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 93878d7b67SJosef Bacik if (num_devices >= btrfs_raid_array[raid_type].devs_min) 94878d7b67SJosef Bacik allowed |= btrfs_raid_array[raid_type].bg_flag; 95878d7b67SJosef Bacik } 96878d7b67SJosef Bacik allowed &= flags; 97878d7b67SJosef Bacik 98160fe8f6SMatt Corallo /* Select the highest-redundancy RAID level. */ 99160fe8f6SMatt Corallo if (allowed & BTRFS_BLOCK_GROUP_RAID1C4) 100160fe8f6SMatt Corallo allowed = BTRFS_BLOCK_GROUP_RAID1C4; 101160fe8f6SMatt Corallo else if (allowed & BTRFS_BLOCK_GROUP_RAID6) 102878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID6; 103160fe8f6SMatt Corallo else if (allowed & BTRFS_BLOCK_GROUP_RAID1C3) 104160fe8f6SMatt Corallo allowed = BTRFS_BLOCK_GROUP_RAID1C3; 105878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID5) 106878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID5; 107878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID10) 108878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID10; 109878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID1) 110878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID1; 111160fe8f6SMatt Corallo else if (allowed & BTRFS_BLOCK_GROUP_DUP) 112160fe8f6SMatt Corallo allowed = BTRFS_BLOCK_GROUP_DUP; 113878d7b67SJosef Bacik else if (allowed & BTRFS_BLOCK_GROUP_RAID0) 114878d7b67SJosef Bacik allowed = BTRFS_BLOCK_GROUP_RAID0; 115878d7b67SJosef Bacik 116878d7b67SJosef Bacik flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; 117878d7b67SJosef Bacik 118878d7b67SJosef Bacik return extended_to_chunk(flags | allowed); 119878d7b67SJosef Bacik } 120878d7b67SJosef Bacik 121ef0a82daSJohannes Thumshirn u64 btrfs_get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) 122878d7b67SJosef Bacik { 123878d7b67SJosef Bacik unsigned seq; 124878d7b67SJosef Bacik u64 flags; 125878d7b67SJosef Bacik 126878d7b67SJosef Bacik do { 127878d7b67SJosef Bacik flags = orig_flags; 128878d7b67SJosef Bacik seq = read_seqbegin(&fs_info->profiles_lock); 129878d7b67SJosef Bacik 130878d7b67SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 131878d7b67SJosef Bacik flags |= fs_info->avail_data_alloc_bits; 132878d7b67SJosef Bacik else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 133878d7b67SJosef Bacik flags |= fs_info->avail_system_alloc_bits; 134878d7b67SJosef Bacik else if (flags & BTRFS_BLOCK_GROUP_METADATA) 135878d7b67SJosef Bacik flags |= fs_info->avail_metadata_alloc_bits; 136878d7b67SJosef Bacik } while (read_seqretry(&fs_info->profiles_lock, seq)); 137878d7b67SJosef Bacik 138878d7b67SJosef Bacik return btrfs_reduce_alloc_profile(fs_info, flags); 139878d7b67SJosef Bacik } 140878d7b67SJosef Bacik 14132da5386SDavid Sterba void btrfs_get_block_group(struct btrfs_block_group *cache) 1423cad1284SJosef Bacik { 14348aaeebeSJosef Bacik refcount_inc(&cache->refs); 1443cad1284SJosef Bacik } 1453cad1284SJosef Bacik 14632da5386SDavid Sterba void btrfs_put_block_group(struct btrfs_block_group *cache) 1473cad1284SJosef Bacik { 14848aaeebeSJosef Bacik if (refcount_dec_and_test(&cache->refs)) { 1493cad1284SJosef Bacik WARN_ON(cache->pinned > 0); 15040cdc509SFilipe Manana /* 15140cdc509SFilipe Manana * If there was a failure to cleanup a log tree, very likely due 15240cdc509SFilipe Manana * to an IO failure on a writeback attempt of one or more of its 15340cdc509SFilipe Manana * extent buffers, we could not do proper (and cheap) unaccounting 15440cdc509SFilipe Manana * of their reserved space, so don't warn on reserved > 0 in that 15540cdc509SFilipe Manana * case. 15640cdc509SFilipe Manana */ 15740cdc509SFilipe Manana if (!(cache->flags & BTRFS_BLOCK_GROUP_METADATA) || 15840cdc509SFilipe Manana !BTRFS_FS_LOG_CLEANUP_ERROR(cache->fs_info)) 1593cad1284SJosef Bacik WARN_ON(cache->reserved > 0); 1603cad1284SJosef Bacik 1613cad1284SJosef Bacik /* 162b0643e59SDennis Zhou * A block_group shouldn't be on the discard_list anymore. 163b0643e59SDennis Zhou * Remove the block_group from the discard_list to prevent us 164b0643e59SDennis Zhou * from causing a panic due to NULL pointer dereference. 165b0643e59SDennis Zhou */ 166b0643e59SDennis Zhou if (WARN_ON(!list_empty(&cache->discard_list))) 167b0643e59SDennis Zhou btrfs_discard_cancel_work(&cache->fs_info->discard_ctl, 168b0643e59SDennis Zhou cache); 169b0643e59SDennis Zhou 1703cad1284SJosef Bacik kfree(cache->free_space_ctl); 171dafc340dSNaohiro Aota kfree(cache->physical_map); 1723cad1284SJosef Bacik kfree(cache); 1733cad1284SJosef Bacik } 1743cad1284SJosef Bacik } 1753cad1284SJosef Bacik 1762e405ad8SJosef Bacik /* 1774358d963SJosef Bacik * This adds the block group to the fs_info rb tree for the block group cache 1784358d963SJosef Bacik */ 1794358d963SJosef Bacik static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, 18032da5386SDavid Sterba struct btrfs_block_group *block_group) 1814358d963SJosef Bacik { 1824358d963SJosef Bacik struct rb_node **p; 1834358d963SJosef Bacik struct rb_node *parent = NULL; 18432da5386SDavid Sterba struct btrfs_block_group *cache; 18508dddb29SFilipe Manana bool leftmost = true; 1864358d963SJosef Bacik 1879afc6649SQu Wenruo ASSERT(block_group->length != 0); 1889afc6649SQu Wenruo 18916b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 19008dddb29SFilipe Manana p = &info->block_group_cache_tree.rb_root.rb_node; 1914358d963SJosef Bacik 1924358d963SJosef Bacik while (*p) { 1934358d963SJosef Bacik parent = *p; 19432da5386SDavid Sterba cache = rb_entry(parent, struct btrfs_block_group, cache_node); 195b3470b5dSDavid Sterba if (block_group->start < cache->start) { 1964358d963SJosef Bacik p = &(*p)->rb_left; 197b3470b5dSDavid Sterba } else if (block_group->start > cache->start) { 1984358d963SJosef Bacik p = &(*p)->rb_right; 19908dddb29SFilipe Manana leftmost = false; 2004358d963SJosef Bacik } else { 20116b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 2024358d963SJosef Bacik return -EEXIST; 2034358d963SJosef Bacik } 2044358d963SJosef Bacik } 2054358d963SJosef Bacik 2064358d963SJosef Bacik rb_link_node(&block_group->cache_node, parent, p); 20708dddb29SFilipe Manana rb_insert_color_cached(&block_group->cache_node, 20808dddb29SFilipe Manana &info->block_group_cache_tree, leftmost); 2094358d963SJosef Bacik 21016b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 2114358d963SJosef Bacik 2124358d963SJosef Bacik return 0; 2134358d963SJosef Bacik } 2144358d963SJosef Bacik 2154358d963SJosef Bacik /* 2162e405ad8SJosef Bacik * This will return the block group at or after bytenr if contains is 0, else 2172e405ad8SJosef Bacik * it will return the block group that contains the bytenr 2182e405ad8SJosef Bacik */ 21932da5386SDavid Sterba static struct btrfs_block_group *block_group_cache_tree_search( 2202e405ad8SJosef Bacik struct btrfs_fs_info *info, u64 bytenr, int contains) 2212e405ad8SJosef Bacik { 22232da5386SDavid Sterba struct btrfs_block_group *cache, *ret = NULL; 2232e405ad8SJosef Bacik struct rb_node *n; 2242e405ad8SJosef Bacik u64 end, start; 2252e405ad8SJosef Bacik 22616b0c258SFilipe Manana read_lock(&info->block_group_cache_lock); 22708dddb29SFilipe Manana n = info->block_group_cache_tree.rb_root.rb_node; 2282e405ad8SJosef Bacik 2292e405ad8SJosef Bacik while (n) { 23032da5386SDavid Sterba cache = rb_entry(n, struct btrfs_block_group, cache_node); 231b3470b5dSDavid Sterba end = cache->start + cache->length - 1; 232b3470b5dSDavid Sterba start = cache->start; 2332e405ad8SJosef Bacik 2342e405ad8SJosef Bacik if (bytenr < start) { 235b3470b5dSDavid Sterba if (!contains && (!ret || start < ret->start)) 2362e405ad8SJosef Bacik ret = cache; 2372e405ad8SJosef Bacik n = n->rb_left; 2382e405ad8SJosef Bacik } else if (bytenr > start) { 2392e405ad8SJosef Bacik if (contains && bytenr <= end) { 2402e405ad8SJosef Bacik ret = cache; 2412e405ad8SJosef Bacik break; 2422e405ad8SJosef Bacik } 2432e405ad8SJosef Bacik n = n->rb_right; 2442e405ad8SJosef Bacik } else { 2452e405ad8SJosef Bacik ret = cache; 2462e405ad8SJosef Bacik break; 2472e405ad8SJosef Bacik } 2482e405ad8SJosef Bacik } 24908dddb29SFilipe Manana if (ret) 2502e405ad8SJosef Bacik btrfs_get_block_group(ret); 25116b0c258SFilipe Manana read_unlock(&info->block_group_cache_lock); 2522e405ad8SJosef Bacik 2532e405ad8SJosef Bacik return ret; 2542e405ad8SJosef Bacik } 2552e405ad8SJosef Bacik 2562e405ad8SJosef Bacik /* 2572e405ad8SJosef Bacik * Return the block group that starts at or after bytenr 2582e405ad8SJosef Bacik */ 25932da5386SDavid Sterba struct btrfs_block_group *btrfs_lookup_first_block_group( 2602e405ad8SJosef Bacik struct btrfs_fs_info *info, u64 bytenr) 2612e405ad8SJosef Bacik { 2622e405ad8SJosef Bacik return block_group_cache_tree_search(info, bytenr, 0); 2632e405ad8SJosef Bacik } 2642e405ad8SJosef Bacik 2652e405ad8SJosef Bacik /* 2662e405ad8SJosef Bacik * Return the block group that contains the given bytenr 2672e405ad8SJosef Bacik */ 26832da5386SDavid Sterba struct btrfs_block_group *btrfs_lookup_block_group( 2692e405ad8SJosef Bacik struct btrfs_fs_info *info, u64 bytenr) 2702e405ad8SJosef Bacik { 2712e405ad8SJosef Bacik return block_group_cache_tree_search(info, bytenr, 1); 2722e405ad8SJosef Bacik } 2732e405ad8SJosef Bacik 27432da5386SDavid Sterba struct btrfs_block_group *btrfs_next_block_group( 27532da5386SDavid Sterba struct btrfs_block_group *cache) 2762e405ad8SJosef Bacik { 2772e405ad8SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 2782e405ad8SJosef Bacik struct rb_node *node; 2792e405ad8SJosef Bacik 28016b0c258SFilipe Manana read_lock(&fs_info->block_group_cache_lock); 2812e405ad8SJosef Bacik 2822e405ad8SJosef Bacik /* If our block group was removed, we need a full search. */ 2832e405ad8SJosef Bacik if (RB_EMPTY_NODE(&cache->cache_node)) { 284b3470b5dSDavid Sterba const u64 next_bytenr = cache->start + cache->length; 2852e405ad8SJosef Bacik 28616b0c258SFilipe Manana read_unlock(&fs_info->block_group_cache_lock); 2872e405ad8SJosef Bacik btrfs_put_block_group(cache); 2888b01f931SFilipe Manana return btrfs_lookup_first_block_group(fs_info, next_bytenr); 2892e405ad8SJosef Bacik } 2902e405ad8SJosef Bacik node = rb_next(&cache->cache_node); 2912e405ad8SJosef Bacik btrfs_put_block_group(cache); 2922e405ad8SJosef Bacik if (node) { 29332da5386SDavid Sterba cache = rb_entry(node, struct btrfs_block_group, cache_node); 2942e405ad8SJosef Bacik btrfs_get_block_group(cache); 2952e405ad8SJosef Bacik } else 2962e405ad8SJosef Bacik cache = NULL; 29716b0c258SFilipe Manana read_unlock(&fs_info->block_group_cache_lock); 2982e405ad8SJosef Bacik return cache; 2992e405ad8SJosef Bacik } 3003eeb3226SJosef Bacik 30143dd529aSDavid Sterba /* 3022306e83eSFilipe Manana * Check if we can do a NOCOW write for a given extent. 3032306e83eSFilipe Manana * 3042306e83eSFilipe Manana * @fs_info: The filesystem information object. 3052306e83eSFilipe Manana * @bytenr: Logical start address of the extent. 3062306e83eSFilipe Manana * 3072306e83eSFilipe Manana * Check if we can do a NOCOW write for the given extent, and increments the 3082306e83eSFilipe Manana * number of NOCOW writers in the block group that contains the extent, as long 3092306e83eSFilipe Manana * as the block group exists and it's currently not in read-only mode. 3102306e83eSFilipe Manana * 3112306e83eSFilipe Manana * Returns: A non-NULL block group pointer if we can do a NOCOW write, the caller 3122306e83eSFilipe Manana * is responsible for calling btrfs_dec_nocow_writers() later. 3132306e83eSFilipe Manana * 3142306e83eSFilipe Manana * Or NULL if we can not do a NOCOW write 3152306e83eSFilipe Manana */ 3162306e83eSFilipe Manana struct btrfs_block_group *btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, 3172306e83eSFilipe Manana u64 bytenr) 3183eeb3226SJosef Bacik { 31932da5386SDavid Sterba struct btrfs_block_group *bg; 3202306e83eSFilipe Manana bool can_nocow = true; 3213eeb3226SJosef Bacik 3223eeb3226SJosef Bacik bg = btrfs_lookup_block_group(fs_info, bytenr); 3233eeb3226SJosef Bacik if (!bg) 3242306e83eSFilipe Manana return NULL; 3253eeb3226SJosef Bacik 3263eeb3226SJosef Bacik spin_lock(&bg->lock); 3273eeb3226SJosef Bacik if (bg->ro) 3282306e83eSFilipe Manana can_nocow = false; 3293eeb3226SJosef Bacik else 3303eeb3226SJosef Bacik atomic_inc(&bg->nocow_writers); 3313eeb3226SJosef Bacik spin_unlock(&bg->lock); 3323eeb3226SJosef Bacik 3332306e83eSFilipe Manana if (!can_nocow) { 3343eeb3226SJosef Bacik btrfs_put_block_group(bg); 3352306e83eSFilipe Manana return NULL; 3363eeb3226SJosef Bacik } 3373eeb3226SJosef Bacik 3382306e83eSFilipe Manana /* No put on block group, done by btrfs_dec_nocow_writers(). */ 3392306e83eSFilipe Manana return bg; 3402306e83eSFilipe Manana } 3413eeb3226SJosef Bacik 34243dd529aSDavid Sterba /* 3432306e83eSFilipe Manana * Decrement the number of NOCOW writers in a block group. 3442306e83eSFilipe Manana * 3452306e83eSFilipe Manana * This is meant to be called after a previous call to btrfs_inc_nocow_writers(), 3462306e83eSFilipe Manana * and on the block group returned by that call. Typically this is called after 3472306e83eSFilipe Manana * creating an ordered extent for a NOCOW write, to prevent races with scrub and 3482306e83eSFilipe Manana * relocation. 3492306e83eSFilipe Manana * 3502306e83eSFilipe Manana * After this call, the caller should not use the block group anymore. It it wants 3512306e83eSFilipe Manana * to use it, then it should get a reference on it before calling this function. 3522306e83eSFilipe Manana */ 3532306e83eSFilipe Manana void btrfs_dec_nocow_writers(struct btrfs_block_group *bg) 3542306e83eSFilipe Manana { 3553eeb3226SJosef Bacik if (atomic_dec_and_test(&bg->nocow_writers)) 3563eeb3226SJosef Bacik wake_up_var(&bg->nocow_writers); 3572306e83eSFilipe Manana 3582306e83eSFilipe Manana /* For the lookup done by a previous call to btrfs_inc_nocow_writers(). */ 3593eeb3226SJosef Bacik btrfs_put_block_group(bg); 3603eeb3226SJosef Bacik } 3613eeb3226SJosef Bacik 36232da5386SDavid Sterba void btrfs_wait_nocow_writers(struct btrfs_block_group *bg) 3633eeb3226SJosef Bacik { 3643eeb3226SJosef Bacik wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); 3653eeb3226SJosef Bacik } 3663eeb3226SJosef Bacik 3673eeb3226SJosef Bacik void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, 3683eeb3226SJosef Bacik const u64 start) 3693eeb3226SJosef Bacik { 37032da5386SDavid Sterba struct btrfs_block_group *bg; 3713eeb3226SJosef Bacik 3723eeb3226SJosef Bacik bg = btrfs_lookup_block_group(fs_info, start); 3733eeb3226SJosef Bacik ASSERT(bg); 3743eeb3226SJosef Bacik if (atomic_dec_and_test(&bg->reservations)) 3753eeb3226SJosef Bacik wake_up_var(&bg->reservations); 3763eeb3226SJosef Bacik btrfs_put_block_group(bg); 3773eeb3226SJosef Bacik } 3783eeb3226SJosef Bacik 37932da5386SDavid Sterba void btrfs_wait_block_group_reservations(struct btrfs_block_group *bg) 3803eeb3226SJosef Bacik { 3813eeb3226SJosef Bacik struct btrfs_space_info *space_info = bg->space_info; 3823eeb3226SJosef Bacik 3833eeb3226SJosef Bacik ASSERT(bg->ro); 3843eeb3226SJosef Bacik 3853eeb3226SJosef Bacik if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) 3863eeb3226SJosef Bacik return; 3873eeb3226SJosef Bacik 3883eeb3226SJosef Bacik /* 3893eeb3226SJosef Bacik * Our block group is read only but before we set it to read only, 3903eeb3226SJosef Bacik * some task might have had allocated an extent from it already, but it 3913eeb3226SJosef Bacik * has not yet created a respective ordered extent (and added it to a 3923eeb3226SJosef Bacik * root's list of ordered extents). 3933eeb3226SJosef Bacik * Therefore wait for any task currently allocating extents, since the 3943eeb3226SJosef Bacik * block group's reservations counter is incremented while a read lock 3953eeb3226SJosef Bacik * on the groups' semaphore is held and decremented after releasing 3963eeb3226SJosef Bacik * the read access on that semaphore and creating the ordered extent. 3973eeb3226SJosef Bacik */ 3983eeb3226SJosef Bacik down_write(&space_info->groups_sem); 3993eeb3226SJosef Bacik up_write(&space_info->groups_sem); 4003eeb3226SJosef Bacik 4013eeb3226SJosef Bacik wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); 4023eeb3226SJosef Bacik } 4039f21246dSJosef Bacik 4049f21246dSJosef Bacik struct btrfs_caching_control *btrfs_get_caching_control( 40532da5386SDavid Sterba struct btrfs_block_group *cache) 4069f21246dSJosef Bacik { 4079f21246dSJosef Bacik struct btrfs_caching_control *ctl; 4089f21246dSJosef Bacik 4099f21246dSJosef Bacik spin_lock(&cache->lock); 4109f21246dSJosef Bacik if (!cache->caching_ctl) { 4119f21246dSJosef Bacik spin_unlock(&cache->lock); 4129f21246dSJosef Bacik return NULL; 4139f21246dSJosef Bacik } 4149f21246dSJosef Bacik 4159f21246dSJosef Bacik ctl = cache->caching_ctl; 4169f21246dSJosef Bacik refcount_inc(&ctl->count); 4179f21246dSJosef Bacik spin_unlock(&cache->lock); 4189f21246dSJosef Bacik return ctl; 4199f21246dSJosef Bacik } 4209f21246dSJosef Bacik 4219f21246dSJosef Bacik void btrfs_put_caching_control(struct btrfs_caching_control *ctl) 4229f21246dSJosef Bacik { 4239f21246dSJosef Bacik if (refcount_dec_and_test(&ctl->count)) 4249f21246dSJosef Bacik kfree(ctl); 4259f21246dSJosef Bacik } 4269f21246dSJosef Bacik 4279f21246dSJosef Bacik /* 4289f21246dSJosef Bacik * When we wait for progress in the block group caching, its because our 4299f21246dSJosef Bacik * allocation attempt failed at least once. So, we must sleep and let some 4309f21246dSJosef Bacik * progress happen before we try again. 4319f21246dSJosef Bacik * 4329f21246dSJosef Bacik * This function will sleep at least once waiting for new free space to show 4339f21246dSJosef Bacik * up, and then it will check the block group free space numbers for our min 4349f21246dSJosef Bacik * num_bytes. Another option is to have it go ahead and look in the rbtree for 4359f21246dSJosef Bacik * a free extent of a given size, but this is a good start. 4369f21246dSJosef Bacik * 4379f21246dSJosef Bacik * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using 4389f21246dSJosef Bacik * any of the information in this block group. 4399f21246dSJosef Bacik */ 44032da5386SDavid Sterba void btrfs_wait_block_group_cache_progress(struct btrfs_block_group *cache, 4419f21246dSJosef Bacik u64 num_bytes) 4429f21246dSJosef Bacik { 4439f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 4449f21246dSJosef Bacik 4459f21246dSJosef Bacik caching_ctl = btrfs_get_caching_control(cache); 4469f21246dSJosef Bacik if (!caching_ctl) 4479f21246dSJosef Bacik return; 4489f21246dSJosef Bacik 44932da5386SDavid Sterba wait_event(caching_ctl->wait, btrfs_block_group_done(cache) || 4509f21246dSJosef Bacik (cache->free_space_ctl->free_space >= num_bytes)); 4519f21246dSJosef Bacik 4529f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 4539f21246dSJosef Bacik } 4549f21246dSJosef Bacik 455ced8ecf0SOmar Sandoval static int btrfs_caching_ctl_wait_done(struct btrfs_block_group *cache, 456ced8ecf0SOmar Sandoval struct btrfs_caching_control *caching_ctl) 457ced8ecf0SOmar Sandoval { 458ced8ecf0SOmar Sandoval wait_event(caching_ctl->wait, btrfs_block_group_done(cache)); 459ced8ecf0SOmar Sandoval return cache->cached == BTRFS_CACHE_ERROR ? -EIO : 0; 460ced8ecf0SOmar Sandoval } 461ced8ecf0SOmar Sandoval 462ced8ecf0SOmar Sandoval static int btrfs_wait_block_group_cache_done(struct btrfs_block_group *cache) 4639f21246dSJosef Bacik { 4649f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 465ced8ecf0SOmar Sandoval int ret; 4669f21246dSJosef Bacik 4679f21246dSJosef Bacik caching_ctl = btrfs_get_caching_control(cache); 4689f21246dSJosef Bacik if (!caching_ctl) 4699f21246dSJosef Bacik return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; 470ced8ecf0SOmar Sandoval ret = btrfs_caching_ctl_wait_done(cache, caching_ctl); 4719f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 4729f21246dSJosef Bacik return ret; 4739f21246dSJosef Bacik } 4749f21246dSJosef Bacik 4759f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 47632da5386SDavid Sterba static void fragment_free_space(struct btrfs_block_group *block_group) 4779f21246dSJosef Bacik { 4789f21246dSJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 479b3470b5dSDavid Sterba u64 start = block_group->start; 480b3470b5dSDavid Sterba u64 len = block_group->length; 4819f21246dSJosef Bacik u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? 4829f21246dSJosef Bacik fs_info->nodesize : fs_info->sectorsize; 4839f21246dSJosef Bacik u64 step = chunk << 1; 4849f21246dSJosef Bacik 4859f21246dSJosef Bacik while (len > chunk) { 4869f21246dSJosef Bacik btrfs_remove_free_space(block_group, start, chunk); 4879f21246dSJosef Bacik start += step; 4889f21246dSJosef Bacik if (len < step) 4899f21246dSJosef Bacik len = 0; 4909f21246dSJosef Bacik else 4919f21246dSJosef Bacik len -= step; 4929f21246dSJosef Bacik } 4939f21246dSJosef Bacik } 4949f21246dSJosef Bacik #endif 4959f21246dSJosef Bacik 4969f21246dSJosef Bacik /* 4979f21246dSJosef Bacik * This is only called by btrfs_cache_block_group, since we could have freed 4989f21246dSJosef Bacik * extents we need to check the pinned_extents for any extents that can't be 4999f21246dSJosef Bacik * used yet since their free space will be released as soon as the transaction 5009f21246dSJosef Bacik * commits. 5019f21246dSJosef Bacik */ 50232da5386SDavid Sterba u64 add_new_free_space(struct btrfs_block_group *block_group, u64 start, u64 end) 5039f21246dSJosef Bacik { 5049f21246dSJosef Bacik struct btrfs_fs_info *info = block_group->fs_info; 5059f21246dSJosef Bacik u64 extent_start, extent_end, size, total_added = 0; 5069f21246dSJosef Bacik int ret; 5079f21246dSJosef Bacik 5089f21246dSJosef Bacik while (start < end) { 509fe119a6eSNikolay Borisov ret = find_first_extent_bit(&info->excluded_extents, start, 5109f21246dSJosef Bacik &extent_start, &extent_end, 5119f21246dSJosef Bacik EXTENT_DIRTY | EXTENT_UPTODATE, 5129f21246dSJosef Bacik NULL); 5139f21246dSJosef Bacik if (ret) 5149f21246dSJosef Bacik break; 5159f21246dSJosef Bacik 5169f21246dSJosef Bacik if (extent_start <= start) { 5179f21246dSJosef Bacik start = extent_end + 1; 5189f21246dSJosef Bacik } else if (extent_start > start && extent_start < end) { 5199f21246dSJosef Bacik size = extent_start - start; 5209f21246dSJosef Bacik total_added += size; 521b0643e59SDennis Zhou ret = btrfs_add_free_space_async_trimmed(block_group, 522b0643e59SDennis Zhou start, size); 5239f21246dSJosef Bacik BUG_ON(ret); /* -ENOMEM or logic error */ 5249f21246dSJosef Bacik start = extent_end + 1; 5259f21246dSJosef Bacik } else { 5269f21246dSJosef Bacik break; 5279f21246dSJosef Bacik } 5289f21246dSJosef Bacik } 5299f21246dSJosef Bacik 5309f21246dSJosef Bacik if (start < end) { 5319f21246dSJosef Bacik size = end - start; 5329f21246dSJosef Bacik total_added += size; 533b0643e59SDennis Zhou ret = btrfs_add_free_space_async_trimmed(block_group, start, 534b0643e59SDennis Zhou size); 5359f21246dSJosef Bacik BUG_ON(ret); /* -ENOMEM or logic error */ 5369f21246dSJosef Bacik } 5379f21246dSJosef Bacik 5389f21246dSJosef Bacik return total_added; 5399f21246dSJosef Bacik } 5409f21246dSJosef Bacik 541c7eec3d9SBoris Burkov /* 542c7eec3d9SBoris Burkov * Get an arbitrary extent item index / max_index through the block group 543c7eec3d9SBoris Burkov * 544c7eec3d9SBoris Burkov * @block_group the block group to sample from 545c7eec3d9SBoris Burkov * @index: the integral step through the block group to grab from 546c7eec3d9SBoris Burkov * @max_index: the granularity of the sampling 547c7eec3d9SBoris Burkov * @key: return value parameter for the item we find 548c7eec3d9SBoris Burkov * 549c7eec3d9SBoris Burkov * Pre-conditions on indices: 550c7eec3d9SBoris Burkov * 0 <= index <= max_index 551c7eec3d9SBoris Burkov * 0 < max_index 552c7eec3d9SBoris Burkov * 553c7eec3d9SBoris Burkov * Returns: 0 on success, 1 if the search didn't yield a useful item, negative 554c7eec3d9SBoris Burkov * error code on error. 555c7eec3d9SBoris Burkov */ 556c7eec3d9SBoris Burkov static int sample_block_group_extent_item(struct btrfs_caching_control *caching_ctl, 557c7eec3d9SBoris Burkov struct btrfs_block_group *block_group, 558c7eec3d9SBoris Burkov int index, int max_index, 55912148367SBoris Burkov struct btrfs_key *found_key) 560c7eec3d9SBoris Burkov { 561c7eec3d9SBoris Burkov struct btrfs_fs_info *fs_info = block_group->fs_info; 562c7eec3d9SBoris Burkov struct btrfs_root *extent_root; 563c7eec3d9SBoris Burkov u64 search_offset; 564c7eec3d9SBoris Burkov u64 search_end = block_group->start + block_group->length; 565c7eec3d9SBoris Burkov struct btrfs_path *path; 56612148367SBoris Burkov struct btrfs_key search_key; 56712148367SBoris Burkov int ret = 0; 568c7eec3d9SBoris Burkov 569c7eec3d9SBoris Burkov ASSERT(index >= 0); 570c7eec3d9SBoris Burkov ASSERT(index <= max_index); 571c7eec3d9SBoris Burkov ASSERT(max_index > 0); 572c7eec3d9SBoris Burkov lockdep_assert_held(&caching_ctl->mutex); 573c7eec3d9SBoris Burkov lockdep_assert_held_read(&fs_info->commit_root_sem); 574c7eec3d9SBoris Burkov 575c7eec3d9SBoris Burkov path = btrfs_alloc_path(); 576c7eec3d9SBoris Burkov if (!path) 577c7eec3d9SBoris Burkov return -ENOMEM; 578c7eec3d9SBoris Burkov 579c7eec3d9SBoris Burkov extent_root = btrfs_extent_root(fs_info, max_t(u64, block_group->start, 580c7eec3d9SBoris Burkov BTRFS_SUPER_INFO_OFFSET)); 581c7eec3d9SBoris Burkov 582c7eec3d9SBoris Burkov path->skip_locking = 1; 583c7eec3d9SBoris Burkov path->search_commit_root = 1; 584c7eec3d9SBoris Burkov path->reada = READA_FORWARD; 585c7eec3d9SBoris Burkov 586c7eec3d9SBoris Burkov search_offset = index * div_u64(block_group->length, max_index); 58712148367SBoris Burkov search_key.objectid = block_group->start + search_offset; 58812148367SBoris Burkov search_key.type = BTRFS_EXTENT_ITEM_KEY; 58912148367SBoris Burkov search_key.offset = 0; 590c7eec3d9SBoris Burkov 59112148367SBoris Burkov btrfs_for_each_slot(extent_root, &search_key, found_key, path, ret) { 592c7eec3d9SBoris Burkov /* Success; sampled an extent item in the block group */ 59312148367SBoris Burkov if (found_key->type == BTRFS_EXTENT_ITEM_KEY && 59412148367SBoris Burkov found_key->objectid >= block_group->start && 59512148367SBoris Burkov found_key->objectid + found_key->offset <= search_end) 59612148367SBoris Burkov break; 597c7eec3d9SBoris Burkov 598c7eec3d9SBoris Burkov /* We can't possibly find a valid extent item anymore */ 59912148367SBoris Burkov if (found_key->objectid >= search_end) { 600c7eec3d9SBoris Burkov ret = 1; 601c7eec3d9SBoris Burkov break; 602c7eec3d9SBoris Burkov } 603c7eec3d9SBoris Burkov } 60412148367SBoris Burkov 605c7eec3d9SBoris Burkov lockdep_assert_held(&caching_ctl->mutex); 606c7eec3d9SBoris Burkov lockdep_assert_held_read(&fs_info->commit_root_sem); 607c7eec3d9SBoris Burkov btrfs_free_path(path); 608c7eec3d9SBoris Burkov return ret; 609c7eec3d9SBoris Burkov } 610c7eec3d9SBoris Burkov 611c7eec3d9SBoris Burkov /* 612c7eec3d9SBoris Burkov * Best effort attempt to compute a block group's size class while caching it. 613c7eec3d9SBoris Burkov * 614c7eec3d9SBoris Burkov * @block_group: the block group we are caching 615c7eec3d9SBoris Burkov * 616c7eec3d9SBoris Burkov * We cannot infer the size class while adding free space extents, because that 617c7eec3d9SBoris Burkov * logic doesn't care about contiguous file extents (it doesn't differentiate 618c7eec3d9SBoris Burkov * between a 100M extent and 100 contiguous 1M extents). So we need to read the 619c7eec3d9SBoris Burkov * file extent items. Reading all of them is quite wasteful, because usually 620c7eec3d9SBoris Burkov * only a handful are enough to give a good answer. Therefore, we just grab 5 of 621c7eec3d9SBoris Burkov * them at even steps through the block group and pick the smallest size class 622c7eec3d9SBoris Burkov * we see. Since size class is best effort, and not guaranteed in general, 623c7eec3d9SBoris Burkov * inaccuracy is acceptable. 624c7eec3d9SBoris Burkov * 625c7eec3d9SBoris Burkov * To be more explicit about why this algorithm makes sense: 626c7eec3d9SBoris Burkov * 627c7eec3d9SBoris Burkov * If we are caching in a block group from disk, then there are three major cases 628c7eec3d9SBoris Burkov * to consider: 629c7eec3d9SBoris Burkov * 1. the block group is well behaved and all extents in it are the same size 630c7eec3d9SBoris Burkov * class. 631c7eec3d9SBoris Burkov * 2. the block group is mostly one size class with rare exceptions for last 632c7eec3d9SBoris Burkov * ditch allocations 633c7eec3d9SBoris Burkov * 3. the block group was populated before size classes and can have a totally 634c7eec3d9SBoris Burkov * arbitrary mix of size classes. 635c7eec3d9SBoris Burkov * 636c7eec3d9SBoris Burkov * In case 1, looking at any extent in the block group will yield the correct 637c7eec3d9SBoris Burkov * result. For the mixed cases, taking the minimum size class seems like a good 638c7eec3d9SBoris Burkov * approximation, since gaps from frees will be usable to the size class. For 639c7eec3d9SBoris Burkov * 2., a small handful of file extents is likely to yield the right answer. For 640c7eec3d9SBoris Burkov * 3, we can either read every file extent, or admit that this is best effort 641c7eec3d9SBoris Burkov * anyway and try to stay fast. 642c7eec3d9SBoris Burkov * 643c7eec3d9SBoris Burkov * Returns: 0 on success, negative error code on error. 644c7eec3d9SBoris Burkov */ 645c7eec3d9SBoris Burkov static int load_block_group_size_class(struct btrfs_caching_control *caching_ctl, 646c7eec3d9SBoris Burkov struct btrfs_block_group *block_group) 647c7eec3d9SBoris Burkov { 64812148367SBoris Burkov struct btrfs_fs_info *fs_info = block_group->fs_info; 649c7eec3d9SBoris Burkov struct btrfs_key key; 650c7eec3d9SBoris Burkov int i; 651c7eec3d9SBoris Burkov u64 min_size = block_group->length; 652c7eec3d9SBoris Burkov enum btrfs_block_group_size_class size_class = BTRFS_BG_SZ_NONE; 653c7eec3d9SBoris Burkov int ret; 654c7eec3d9SBoris Burkov 655cb0922f2SBoris Burkov if (!btrfs_block_group_should_use_size_class(block_group)) 656c7eec3d9SBoris Burkov return 0; 657c7eec3d9SBoris Burkov 65812148367SBoris Burkov lockdep_assert_held(&caching_ctl->mutex); 65912148367SBoris Burkov lockdep_assert_held_read(&fs_info->commit_root_sem); 660c7eec3d9SBoris Burkov for (i = 0; i < 5; ++i) { 661c7eec3d9SBoris Burkov ret = sample_block_group_extent_item(caching_ctl, block_group, i, 5, &key); 662c7eec3d9SBoris Burkov if (ret < 0) 663c7eec3d9SBoris Burkov goto out; 664c7eec3d9SBoris Burkov if (ret > 0) 665c7eec3d9SBoris Burkov continue; 666c7eec3d9SBoris Burkov min_size = min_t(u64, min_size, key.offset); 667c7eec3d9SBoris Burkov size_class = btrfs_calc_block_group_size_class(min_size); 668c7eec3d9SBoris Burkov } 669c7eec3d9SBoris Burkov if (size_class != BTRFS_BG_SZ_NONE) { 670c7eec3d9SBoris Burkov spin_lock(&block_group->lock); 671c7eec3d9SBoris Burkov block_group->size_class = size_class; 672c7eec3d9SBoris Burkov spin_unlock(&block_group->lock); 673c7eec3d9SBoris Burkov } 674c7eec3d9SBoris Burkov out: 675c7eec3d9SBoris Burkov return ret; 676c7eec3d9SBoris Burkov } 677c7eec3d9SBoris Burkov 6789f21246dSJosef Bacik static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) 6799f21246dSJosef Bacik { 68032da5386SDavid Sterba struct btrfs_block_group *block_group = caching_ctl->block_group; 6819f21246dSJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 68229cbcf40SJosef Bacik struct btrfs_root *extent_root; 6839f21246dSJosef Bacik struct btrfs_path *path; 6849f21246dSJosef Bacik struct extent_buffer *leaf; 6859f21246dSJosef Bacik struct btrfs_key key; 6869f21246dSJosef Bacik u64 total_found = 0; 6879f21246dSJosef Bacik u64 last = 0; 6889f21246dSJosef Bacik u32 nritems; 6899f21246dSJosef Bacik int ret; 6909f21246dSJosef Bacik bool wakeup = true; 6919f21246dSJosef Bacik 6929f21246dSJosef Bacik path = btrfs_alloc_path(); 6939f21246dSJosef Bacik if (!path) 6949f21246dSJosef Bacik return -ENOMEM; 6959f21246dSJosef Bacik 696b3470b5dSDavid Sterba last = max_t(u64, block_group->start, BTRFS_SUPER_INFO_OFFSET); 69729cbcf40SJosef Bacik extent_root = btrfs_extent_root(fs_info, last); 6989f21246dSJosef Bacik 6999f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 7009f21246dSJosef Bacik /* 7019f21246dSJosef Bacik * If we're fragmenting we don't want to make anybody think we can 7029f21246dSJosef Bacik * allocate from this block group until we've had a chance to fragment 7039f21246dSJosef Bacik * the free space. 7049f21246dSJosef Bacik */ 7059f21246dSJosef Bacik if (btrfs_should_fragment_free_space(block_group)) 7069f21246dSJosef Bacik wakeup = false; 7079f21246dSJosef Bacik #endif 7089f21246dSJosef Bacik /* 7099f21246dSJosef Bacik * We don't want to deadlock with somebody trying to allocate a new 7109f21246dSJosef Bacik * extent for the extent root while also trying to search the extent 7119f21246dSJosef Bacik * root to add free space. So we skip locking and search the commit 7129f21246dSJosef Bacik * root, since its read-only 7139f21246dSJosef Bacik */ 7149f21246dSJosef Bacik path->skip_locking = 1; 7159f21246dSJosef Bacik path->search_commit_root = 1; 7169f21246dSJosef Bacik path->reada = READA_FORWARD; 7179f21246dSJosef Bacik 7189f21246dSJosef Bacik key.objectid = last; 7199f21246dSJosef Bacik key.offset = 0; 7209f21246dSJosef Bacik key.type = BTRFS_EXTENT_ITEM_KEY; 7219f21246dSJosef Bacik 7229f21246dSJosef Bacik next: 7239f21246dSJosef Bacik ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); 7249f21246dSJosef Bacik if (ret < 0) 7259f21246dSJosef Bacik goto out; 7269f21246dSJosef Bacik 7279f21246dSJosef Bacik leaf = path->nodes[0]; 7289f21246dSJosef Bacik nritems = btrfs_header_nritems(leaf); 7299f21246dSJosef Bacik 7309f21246dSJosef Bacik while (1) { 7319f21246dSJosef Bacik if (btrfs_fs_closing(fs_info) > 1) { 7329f21246dSJosef Bacik last = (u64)-1; 7339f21246dSJosef Bacik break; 7349f21246dSJosef Bacik } 7359f21246dSJosef Bacik 7369f21246dSJosef Bacik if (path->slots[0] < nritems) { 7379f21246dSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 7389f21246dSJosef Bacik } else { 7399f21246dSJosef Bacik ret = btrfs_find_next_key(extent_root, path, &key, 0, 0); 7409f21246dSJosef Bacik if (ret) 7419f21246dSJosef Bacik break; 7429f21246dSJosef Bacik 7439f21246dSJosef Bacik if (need_resched() || 7449f21246dSJosef Bacik rwsem_is_contended(&fs_info->commit_root_sem)) { 7459f21246dSJosef Bacik btrfs_release_path(path); 7469f21246dSJosef Bacik up_read(&fs_info->commit_root_sem); 7479f21246dSJosef Bacik mutex_unlock(&caching_ctl->mutex); 7489f21246dSJosef Bacik cond_resched(); 7499f21246dSJosef Bacik mutex_lock(&caching_ctl->mutex); 7509f21246dSJosef Bacik down_read(&fs_info->commit_root_sem); 7519f21246dSJosef Bacik goto next; 7529f21246dSJosef Bacik } 7539f21246dSJosef Bacik 7549f21246dSJosef Bacik ret = btrfs_next_leaf(extent_root, path); 7559f21246dSJosef Bacik if (ret < 0) 7569f21246dSJosef Bacik goto out; 7579f21246dSJosef Bacik if (ret) 7589f21246dSJosef Bacik break; 7599f21246dSJosef Bacik leaf = path->nodes[0]; 7609f21246dSJosef Bacik nritems = btrfs_header_nritems(leaf); 7619f21246dSJosef Bacik continue; 7629f21246dSJosef Bacik } 7639f21246dSJosef Bacik 7649f21246dSJosef Bacik if (key.objectid < last) { 7659f21246dSJosef Bacik key.objectid = last; 7669f21246dSJosef Bacik key.offset = 0; 7679f21246dSJosef Bacik key.type = BTRFS_EXTENT_ITEM_KEY; 7689f21246dSJosef Bacik btrfs_release_path(path); 7699f21246dSJosef Bacik goto next; 7709f21246dSJosef Bacik } 7719f21246dSJosef Bacik 772b3470b5dSDavid Sterba if (key.objectid < block_group->start) { 7739f21246dSJosef Bacik path->slots[0]++; 7749f21246dSJosef Bacik continue; 7759f21246dSJosef Bacik } 7769f21246dSJosef Bacik 777b3470b5dSDavid Sterba if (key.objectid >= block_group->start + block_group->length) 7789f21246dSJosef Bacik break; 7799f21246dSJosef Bacik 7809f21246dSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7819f21246dSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7829f21246dSJosef Bacik total_found += add_new_free_space(block_group, last, 7839f21246dSJosef Bacik key.objectid); 7849f21246dSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY) 7859f21246dSJosef Bacik last = key.objectid + 7869f21246dSJosef Bacik fs_info->nodesize; 7879f21246dSJosef Bacik else 7889f21246dSJosef Bacik last = key.objectid + key.offset; 7899f21246dSJosef Bacik 7909f21246dSJosef Bacik if (total_found > CACHING_CTL_WAKE_UP) { 7919f21246dSJosef Bacik total_found = 0; 7929f21246dSJosef Bacik if (wakeup) 7939f21246dSJosef Bacik wake_up(&caching_ctl->wait); 7949f21246dSJosef Bacik } 7959f21246dSJosef Bacik } 7969f21246dSJosef Bacik path->slots[0]++; 7979f21246dSJosef Bacik } 7989f21246dSJosef Bacik ret = 0; 7999f21246dSJosef Bacik 8009f21246dSJosef Bacik total_found += add_new_free_space(block_group, last, 801b3470b5dSDavid Sterba block_group->start + block_group->length); 8029f21246dSJosef Bacik 8039f21246dSJosef Bacik out: 8049f21246dSJosef Bacik btrfs_free_path(path); 8059f21246dSJosef Bacik return ret; 8069f21246dSJosef Bacik } 8079f21246dSJosef Bacik 8089f21246dSJosef Bacik static noinline void caching_thread(struct btrfs_work *work) 8099f21246dSJosef Bacik { 81032da5386SDavid Sterba struct btrfs_block_group *block_group; 8119f21246dSJosef Bacik struct btrfs_fs_info *fs_info; 8129f21246dSJosef Bacik struct btrfs_caching_control *caching_ctl; 8139f21246dSJosef Bacik int ret; 8149f21246dSJosef Bacik 8159f21246dSJosef Bacik caching_ctl = container_of(work, struct btrfs_caching_control, work); 8169f21246dSJosef Bacik block_group = caching_ctl->block_group; 8179f21246dSJosef Bacik fs_info = block_group->fs_info; 8189f21246dSJosef Bacik 8199f21246dSJosef Bacik mutex_lock(&caching_ctl->mutex); 8209f21246dSJosef Bacik down_read(&fs_info->commit_root_sem); 8219f21246dSJosef Bacik 822c7eec3d9SBoris Burkov load_block_group_size_class(caching_ctl, block_group); 823e747853cSJosef Bacik if (btrfs_test_opt(fs_info, SPACE_CACHE)) { 824e747853cSJosef Bacik ret = load_free_space_cache(block_group); 825e747853cSJosef Bacik if (ret == 1) { 826e747853cSJosef Bacik ret = 0; 827e747853cSJosef Bacik goto done; 828e747853cSJosef Bacik } 829e747853cSJosef Bacik 830e747853cSJosef Bacik /* 831e747853cSJosef Bacik * We failed to load the space cache, set ourselves to 832e747853cSJosef Bacik * CACHE_STARTED and carry on. 833e747853cSJosef Bacik */ 834e747853cSJosef Bacik spin_lock(&block_group->lock); 835e747853cSJosef Bacik block_group->cached = BTRFS_CACHE_STARTED; 836e747853cSJosef Bacik spin_unlock(&block_group->lock); 837e747853cSJosef Bacik wake_up(&caching_ctl->wait); 838e747853cSJosef Bacik } 839e747853cSJosef Bacik 8402f96e402SJosef Bacik /* 8412f96e402SJosef Bacik * If we are in the transaction that populated the free space tree we 8422f96e402SJosef Bacik * can't actually cache from the free space tree as our commit root and 8432f96e402SJosef Bacik * real root are the same, so we could change the contents of the blocks 8442f96e402SJosef Bacik * while caching. Instead do the slow caching in this case, and after 8452f96e402SJosef Bacik * the transaction has committed we will be safe. 8462f96e402SJosef Bacik */ 8472f96e402SJosef Bacik if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 8482f96e402SJosef Bacik !(test_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags))) 8499f21246dSJosef Bacik ret = load_free_space_tree(caching_ctl); 8509f21246dSJosef Bacik else 8519f21246dSJosef Bacik ret = load_extent_tree_free(caching_ctl); 852e747853cSJosef Bacik done: 8539f21246dSJosef Bacik spin_lock(&block_group->lock); 8549f21246dSJosef Bacik block_group->caching_ctl = NULL; 8559f21246dSJosef Bacik block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; 8569f21246dSJosef Bacik spin_unlock(&block_group->lock); 8579f21246dSJosef Bacik 8589f21246dSJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 8599f21246dSJosef Bacik if (btrfs_should_fragment_free_space(block_group)) { 8609f21246dSJosef Bacik u64 bytes_used; 8619f21246dSJosef Bacik 8629f21246dSJosef Bacik spin_lock(&block_group->space_info->lock); 8639f21246dSJosef Bacik spin_lock(&block_group->lock); 864b3470b5dSDavid Sterba bytes_used = block_group->length - block_group->used; 8659f21246dSJosef Bacik block_group->space_info->bytes_used += bytes_used >> 1; 8669f21246dSJosef Bacik spin_unlock(&block_group->lock); 8679f21246dSJosef Bacik spin_unlock(&block_group->space_info->lock); 868e11c0406SJosef Bacik fragment_free_space(block_group); 8699f21246dSJosef Bacik } 8709f21246dSJosef Bacik #endif 8719f21246dSJosef Bacik 8729f21246dSJosef Bacik up_read(&fs_info->commit_root_sem); 8739f21246dSJosef Bacik btrfs_free_excluded_extents(block_group); 8749f21246dSJosef Bacik mutex_unlock(&caching_ctl->mutex); 8759f21246dSJosef Bacik 8769f21246dSJosef Bacik wake_up(&caching_ctl->wait); 8779f21246dSJosef Bacik 8789f21246dSJosef Bacik btrfs_put_caching_control(caching_ctl); 8799f21246dSJosef Bacik btrfs_put_block_group(block_group); 8809f21246dSJosef Bacik } 8819f21246dSJosef Bacik 882ced8ecf0SOmar Sandoval int btrfs_cache_block_group(struct btrfs_block_group *cache, bool wait) 8839f21246dSJosef Bacik { 8849f21246dSJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 885e747853cSJosef Bacik struct btrfs_caching_control *caching_ctl = NULL; 8869f21246dSJosef Bacik int ret = 0; 8879f21246dSJosef Bacik 8882eda5708SNaohiro Aota /* Allocator for zoned filesystems does not use the cache at all */ 8892eda5708SNaohiro Aota if (btrfs_is_zoned(fs_info)) 8902eda5708SNaohiro Aota return 0; 8912eda5708SNaohiro Aota 8929f21246dSJosef Bacik caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); 8939f21246dSJosef Bacik if (!caching_ctl) 8949f21246dSJosef Bacik return -ENOMEM; 8959f21246dSJosef Bacik 8969f21246dSJosef Bacik INIT_LIST_HEAD(&caching_ctl->list); 8979f21246dSJosef Bacik mutex_init(&caching_ctl->mutex); 8989f21246dSJosef Bacik init_waitqueue_head(&caching_ctl->wait); 8999f21246dSJosef Bacik caching_ctl->block_group = cache; 900e747853cSJosef Bacik refcount_set(&caching_ctl->count, 2); 901a0cac0ecSOmar Sandoval btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); 9029f21246dSJosef Bacik 9039f21246dSJosef Bacik spin_lock(&cache->lock); 9049f21246dSJosef Bacik if (cache->cached != BTRFS_CACHE_NO) { 9059f21246dSJosef Bacik kfree(caching_ctl); 906e747853cSJosef Bacik 907e747853cSJosef Bacik caching_ctl = cache->caching_ctl; 908e747853cSJosef Bacik if (caching_ctl) 909e747853cSJosef Bacik refcount_inc(&caching_ctl->count); 910e747853cSJosef Bacik spin_unlock(&cache->lock); 911e747853cSJosef Bacik goto out; 9129f21246dSJosef Bacik } 9139f21246dSJosef Bacik WARN_ON(cache->caching_ctl); 9149f21246dSJosef Bacik cache->caching_ctl = caching_ctl; 9159f21246dSJosef Bacik cache->cached = BTRFS_CACHE_STARTED; 9169f21246dSJosef Bacik spin_unlock(&cache->lock); 9179f21246dSJosef Bacik 91816b0c258SFilipe Manana write_lock(&fs_info->block_group_cache_lock); 9199f21246dSJosef Bacik refcount_inc(&caching_ctl->count); 9209f21246dSJosef Bacik list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); 92116b0c258SFilipe Manana write_unlock(&fs_info->block_group_cache_lock); 9229f21246dSJosef Bacik 9239f21246dSJosef Bacik btrfs_get_block_group(cache); 9249f21246dSJosef Bacik 9259f21246dSJosef Bacik btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); 926e747853cSJosef Bacik out: 927ced8ecf0SOmar Sandoval if (wait && caching_ctl) 928ced8ecf0SOmar Sandoval ret = btrfs_caching_ctl_wait_done(cache, caching_ctl); 929e747853cSJosef Bacik if (caching_ctl) 930e747853cSJosef Bacik btrfs_put_caching_control(caching_ctl); 9319f21246dSJosef Bacik 9329f21246dSJosef Bacik return ret; 9339f21246dSJosef Bacik } 934e3e0520bSJosef Bacik 935e3e0520bSJosef Bacik static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) 936e3e0520bSJosef Bacik { 937e3e0520bSJosef Bacik u64 extra_flags = chunk_to_extended(flags) & 938e3e0520bSJosef Bacik BTRFS_EXTENDED_PROFILE_MASK; 939e3e0520bSJosef Bacik 940e3e0520bSJosef Bacik write_seqlock(&fs_info->profiles_lock); 941e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 942e3e0520bSJosef Bacik fs_info->avail_data_alloc_bits &= ~extra_flags; 943e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_METADATA) 944e3e0520bSJosef Bacik fs_info->avail_metadata_alloc_bits &= ~extra_flags; 945e3e0520bSJosef Bacik if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 946e3e0520bSJosef Bacik fs_info->avail_system_alloc_bits &= ~extra_flags; 947e3e0520bSJosef Bacik write_sequnlock(&fs_info->profiles_lock); 948e3e0520bSJosef Bacik } 949e3e0520bSJosef Bacik 950e3e0520bSJosef Bacik /* 951e3e0520bSJosef Bacik * Clear incompat bits for the following feature(s): 952e3e0520bSJosef Bacik * 953e3e0520bSJosef Bacik * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group 954e3e0520bSJosef Bacik * in the whole filesystem 9559c907446SDavid Sterba * 9569c907446SDavid Sterba * - RAID1C34 - same as above for RAID1C3 and RAID1C4 block groups 957e3e0520bSJosef Bacik */ 958e3e0520bSJosef Bacik static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) 959e3e0520bSJosef Bacik { 9609c907446SDavid Sterba bool found_raid56 = false; 9619c907446SDavid Sterba bool found_raid1c34 = false; 9629c907446SDavid Sterba 9639c907446SDavid Sterba if ((flags & BTRFS_BLOCK_GROUP_RAID56_MASK) || 9649c907446SDavid Sterba (flags & BTRFS_BLOCK_GROUP_RAID1C3) || 9659c907446SDavid Sterba (flags & BTRFS_BLOCK_GROUP_RAID1C4)) { 966e3e0520bSJosef Bacik struct list_head *head = &fs_info->space_info; 967e3e0520bSJosef Bacik struct btrfs_space_info *sinfo; 968e3e0520bSJosef Bacik 969e3e0520bSJosef Bacik list_for_each_entry_rcu(sinfo, head, list) { 970e3e0520bSJosef Bacik down_read(&sinfo->groups_sem); 971e3e0520bSJosef Bacik if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) 9729c907446SDavid Sterba found_raid56 = true; 973e3e0520bSJosef Bacik if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) 9749c907446SDavid Sterba found_raid56 = true; 9759c907446SDavid Sterba if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C3])) 9769c907446SDavid Sterba found_raid1c34 = true; 9779c907446SDavid Sterba if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID1C4])) 9789c907446SDavid Sterba found_raid1c34 = true; 979e3e0520bSJosef Bacik up_read(&sinfo->groups_sem); 980e3e0520bSJosef Bacik } 981d8e6fd5cSFilipe Manana if (!found_raid56) 982e3e0520bSJosef Bacik btrfs_clear_fs_incompat(fs_info, RAID56); 983d8e6fd5cSFilipe Manana if (!found_raid1c34) 9849c907446SDavid Sterba btrfs_clear_fs_incompat(fs_info, RAID1C34); 985e3e0520bSJosef Bacik } 986e3e0520bSJosef Bacik } 987e3e0520bSJosef Bacik 9887357623aSQu Wenruo static int remove_block_group_item(struct btrfs_trans_handle *trans, 9897357623aSQu Wenruo struct btrfs_path *path, 9907357623aSQu Wenruo struct btrfs_block_group *block_group) 9917357623aSQu Wenruo { 9927357623aSQu Wenruo struct btrfs_fs_info *fs_info = trans->fs_info; 9937357623aSQu Wenruo struct btrfs_root *root; 9947357623aSQu Wenruo struct btrfs_key key; 9957357623aSQu Wenruo int ret; 9967357623aSQu Wenruo 997dfe8aec4SJosef Bacik root = btrfs_block_group_root(fs_info); 9987357623aSQu Wenruo key.objectid = block_group->start; 9997357623aSQu Wenruo key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 10007357623aSQu Wenruo key.offset = block_group->length; 10017357623aSQu Wenruo 10027357623aSQu Wenruo ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 10037357623aSQu Wenruo if (ret > 0) 10047357623aSQu Wenruo ret = -ENOENT; 10057357623aSQu Wenruo if (ret < 0) 10067357623aSQu Wenruo return ret; 10077357623aSQu Wenruo 10087357623aSQu Wenruo ret = btrfs_del_item(trans, root, path); 10097357623aSQu Wenruo return ret; 10107357623aSQu Wenruo } 10117357623aSQu Wenruo 1012e3e0520bSJosef Bacik int btrfs_remove_block_group(struct btrfs_trans_handle *trans, 1013e3e0520bSJosef Bacik u64 group_start, struct extent_map *em) 1014e3e0520bSJosef Bacik { 1015e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 1016e3e0520bSJosef Bacik struct btrfs_path *path; 101732da5386SDavid Sterba struct btrfs_block_group *block_group; 1018e3e0520bSJosef Bacik struct btrfs_free_cluster *cluster; 1019e3e0520bSJosef Bacik struct inode *inode; 1020e3e0520bSJosef Bacik struct kobject *kobj = NULL; 1021e3e0520bSJosef Bacik int ret; 1022e3e0520bSJosef Bacik int index; 1023e3e0520bSJosef Bacik int factor; 1024e3e0520bSJosef Bacik struct btrfs_caching_control *caching_ctl = NULL; 1025e3e0520bSJosef Bacik bool remove_em; 1026e3e0520bSJosef Bacik bool remove_rsv = false; 1027e3e0520bSJosef Bacik 1028e3e0520bSJosef Bacik block_group = btrfs_lookup_block_group(fs_info, group_start); 1029e3e0520bSJosef Bacik BUG_ON(!block_group); 1030e3e0520bSJosef Bacik BUG_ON(!block_group->ro); 1031e3e0520bSJosef Bacik 1032e3e0520bSJosef Bacik trace_btrfs_remove_block_group(block_group); 1033e3e0520bSJosef Bacik /* 1034e3e0520bSJosef Bacik * Free the reserved super bytes from this block group before 1035e3e0520bSJosef Bacik * remove it. 1036e3e0520bSJosef Bacik */ 1037e3e0520bSJosef Bacik btrfs_free_excluded_extents(block_group); 1038b3470b5dSDavid Sterba btrfs_free_ref_tree_range(fs_info, block_group->start, 1039b3470b5dSDavid Sterba block_group->length); 1040e3e0520bSJosef Bacik 1041e3e0520bSJosef Bacik index = btrfs_bg_flags_to_raid_index(block_group->flags); 1042e3e0520bSJosef Bacik factor = btrfs_bg_type_to_factor(block_group->flags); 1043e3e0520bSJosef Bacik 1044e3e0520bSJosef Bacik /* make sure this block group isn't part of an allocation cluster */ 1045e3e0520bSJosef Bacik cluster = &fs_info->data_alloc_cluster; 1046e3e0520bSJosef Bacik spin_lock(&cluster->refill_lock); 1047e3e0520bSJosef Bacik btrfs_return_cluster_to_free_space(block_group, cluster); 1048e3e0520bSJosef Bacik spin_unlock(&cluster->refill_lock); 1049e3e0520bSJosef Bacik 1050e3e0520bSJosef Bacik /* 1051e3e0520bSJosef Bacik * make sure this block group isn't part of a metadata 1052e3e0520bSJosef Bacik * allocation cluster 1053e3e0520bSJosef Bacik */ 1054e3e0520bSJosef Bacik cluster = &fs_info->meta_alloc_cluster; 1055e3e0520bSJosef Bacik spin_lock(&cluster->refill_lock); 1056e3e0520bSJosef Bacik btrfs_return_cluster_to_free_space(block_group, cluster); 1057e3e0520bSJosef Bacik spin_unlock(&cluster->refill_lock); 1058e3e0520bSJosef Bacik 105940ab3be1SNaohiro Aota btrfs_clear_treelog_bg(block_group); 1060c2707a25SJohannes Thumshirn btrfs_clear_data_reloc_bg(block_group); 106140ab3be1SNaohiro Aota 1062e3e0520bSJosef Bacik path = btrfs_alloc_path(); 1063e3e0520bSJosef Bacik if (!path) { 1064e3e0520bSJosef Bacik ret = -ENOMEM; 10659fecd132SFilipe Manana goto out; 1066e3e0520bSJosef Bacik } 1067e3e0520bSJosef Bacik 1068e3e0520bSJosef Bacik /* 1069e3e0520bSJosef Bacik * get the inode first so any iput calls done for the io_list 1070e3e0520bSJosef Bacik * aren't the final iput (no unlinks allowed now) 1071e3e0520bSJosef Bacik */ 1072e3e0520bSJosef Bacik inode = lookup_free_space_inode(block_group, path); 1073e3e0520bSJosef Bacik 1074e3e0520bSJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 1075e3e0520bSJosef Bacik /* 1076e3e0520bSJosef Bacik * Make sure our free space cache IO is done before removing the 1077e3e0520bSJosef Bacik * free space inode 1078e3e0520bSJosef Bacik */ 1079e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1080e3e0520bSJosef Bacik if (!list_empty(&block_group->io_list)) { 1081e3e0520bSJosef Bacik list_del_init(&block_group->io_list); 1082e3e0520bSJosef Bacik 1083e3e0520bSJosef Bacik WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); 1084e3e0520bSJosef Bacik 1085e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1086e3e0520bSJosef Bacik btrfs_wait_cache_io(trans, block_group, path); 1087e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1088e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1089e3e0520bSJosef Bacik } 1090e3e0520bSJosef Bacik 1091e3e0520bSJosef Bacik if (!list_empty(&block_group->dirty_list)) { 1092e3e0520bSJosef Bacik list_del_init(&block_group->dirty_list); 1093e3e0520bSJosef Bacik remove_rsv = true; 1094e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1095e3e0520bSJosef Bacik } 1096e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1097e3e0520bSJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 1098e3e0520bSJosef Bacik 109936b216c8SBoris Burkov ret = btrfs_remove_free_space_inode(trans, inode, block_group); 1100e3e0520bSJosef Bacik if (ret) 11019fecd132SFilipe Manana goto out; 1102e3e0520bSJosef Bacik 110316b0c258SFilipe Manana write_lock(&fs_info->block_group_cache_lock); 110408dddb29SFilipe Manana rb_erase_cached(&block_group->cache_node, 1105e3e0520bSJosef Bacik &fs_info->block_group_cache_tree); 1106e3e0520bSJosef Bacik RB_CLEAR_NODE(&block_group->cache_node); 1107e3e0520bSJosef Bacik 11089fecd132SFilipe Manana /* Once for the block groups rbtree */ 11099fecd132SFilipe Manana btrfs_put_block_group(block_group); 11109fecd132SFilipe Manana 111116b0c258SFilipe Manana write_unlock(&fs_info->block_group_cache_lock); 1112e3e0520bSJosef Bacik 1113e3e0520bSJosef Bacik down_write(&block_group->space_info->groups_sem); 1114e3e0520bSJosef Bacik /* 1115e3e0520bSJosef Bacik * we must use list_del_init so people can check to see if they 1116e3e0520bSJosef Bacik * are still on the list after taking the semaphore 1117e3e0520bSJosef Bacik */ 1118e3e0520bSJosef Bacik list_del_init(&block_group->list); 1119e3e0520bSJosef Bacik if (list_empty(&block_group->space_info->block_groups[index])) { 1120e3e0520bSJosef Bacik kobj = block_group->space_info->block_group_kobjs[index]; 1121e3e0520bSJosef Bacik block_group->space_info->block_group_kobjs[index] = NULL; 1122e3e0520bSJosef Bacik clear_avail_alloc_bits(fs_info, block_group->flags); 1123e3e0520bSJosef Bacik } 1124e3e0520bSJosef Bacik up_write(&block_group->space_info->groups_sem); 1125e3e0520bSJosef Bacik clear_incompat_bg_bits(fs_info, block_group->flags); 1126e3e0520bSJosef Bacik if (kobj) { 1127e3e0520bSJosef Bacik kobject_del(kobj); 1128e3e0520bSJosef Bacik kobject_put(kobj); 1129e3e0520bSJosef Bacik } 1130e3e0520bSJosef Bacik 1131e3e0520bSJosef Bacik if (block_group->cached == BTRFS_CACHE_STARTED) 1132e3e0520bSJosef Bacik btrfs_wait_block_group_cache_done(block_group); 11337b9c293bSJosef Bacik 113416b0c258SFilipe Manana write_lock(&fs_info->block_group_cache_lock); 11357b9c293bSJosef Bacik caching_ctl = btrfs_get_caching_control(block_group); 1136e3e0520bSJosef Bacik if (!caching_ctl) { 1137e3e0520bSJosef Bacik struct btrfs_caching_control *ctl; 1138e3e0520bSJosef Bacik 11397b9c293bSJosef Bacik list_for_each_entry(ctl, &fs_info->caching_block_groups, list) { 1140e3e0520bSJosef Bacik if (ctl->block_group == block_group) { 1141e3e0520bSJosef Bacik caching_ctl = ctl; 1142e3e0520bSJosef Bacik refcount_inc(&caching_ctl->count); 1143e3e0520bSJosef Bacik break; 1144e3e0520bSJosef Bacik } 1145e3e0520bSJosef Bacik } 11467b9c293bSJosef Bacik } 1147e3e0520bSJosef Bacik if (caching_ctl) 1148e3e0520bSJosef Bacik list_del_init(&caching_ctl->list); 114916b0c258SFilipe Manana write_unlock(&fs_info->block_group_cache_lock); 11507b9c293bSJosef Bacik 1151e3e0520bSJosef Bacik if (caching_ctl) { 1152e3e0520bSJosef Bacik /* Once for the caching bgs list and once for us. */ 1153e3e0520bSJosef Bacik btrfs_put_caching_control(caching_ctl); 1154e3e0520bSJosef Bacik btrfs_put_caching_control(caching_ctl); 1155e3e0520bSJosef Bacik } 1156e3e0520bSJosef Bacik 1157e3e0520bSJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 1158e3e0520bSJosef Bacik WARN_ON(!list_empty(&block_group->dirty_list)); 1159e3e0520bSJosef Bacik WARN_ON(!list_empty(&block_group->io_list)); 1160e3e0520bSJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 1161e3e0520bSJosef Bacik 1162e3e0520bSJosef Bacik btrfs_remove_free_space_cache(block_group); 1163e3e0520bSJosef Bacik 1164e3e0520bSJosef Bacik spin_lock(&block_group->space_info->lock); 1165e3e0520bSJosef Bacik list_del_init(&block_group->ro_list); 1166e3e0520bSJosef Bacik 1167e3e0520bSJosef Bacik if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { 1168e3e0520bSJosef Bacik WARN_ON(block_group->space_info->total_bytes 1169b3470b5dSDavid Sterba < block_group->length); 1170e3e0520bSJosef Bacik WARN_ON(block_group->space_info->bytes_readonly 1171169e0da9SNaohiro Aota < block_group->length - block_group->zone_unusable); 1172169e0da9SNaohiro Aota WARN_ON(block_group->space_info->bytes_zone_unusable 1173169e0da9SNaohiro Aota < block_group->zone_unusable); 1174e3e0520bSJosef Bacik WARN_ON(block_group->space_info->disk_total 1175b3470b5dSDavid Sterba < block_group->length * factor); 1176e3e0520bSJosef Bacik } 1177b3470b5dSDavid Sterba block_group->space_info->total_bytes -= block_group->length; 1178169e0da9SNaohiro Aota block_group->space_info->bytes_readonly -= 1179169e0da9SNaohiro Aota (block_group->length - block_group->zone_unusable); 1180169e0da9SNaohiro Aota block_group->space_info->bytes_zone_unusable -= 1181169e0da9SNaohiro Aota block_group->zone_unusable; 1182b3470b5dSDavid Sterba block_group->space_info->disk_total -= block_group->length * factor; 1183e3e0520bSJosef Bacik 1184e3e0520bSJosef Bacik spin_unlock(&block_group->space_info->lock); 1185e3e0520bSJosef Bacik 1186ffcb9d44SFilipe Manana /* 1187ffcb9d44SFilipe Manana * Remove the free space for the block group from the free space tree 1188ffcb9d44SFilipe Manana * and the block group's item from the extent tree before marking the 1189ffcb9d44SFilipe Manana * block group as removed. This is to prevent races with tasks that 1190ffcb9d44SFilipe Manana * freeze and unfreeze a block group, this task and another task 1191ffcb9d44SFilipe Manana * allocating a new block group - the unfreeze task ends up removing 1192ffcb9d44SFilipe Manana * the block group's extent map before the task calling this function 1193ffcb9d44SFilipe Manana * deletes the block group item from the extent tree, allowing for 1194ffcb9d44SFilipe Manana * another task to attempt to create another block group with the same 1195ffcb9d44SFilipe Manana * item key (and failing with -EEXIST and a transaction abort). 1196ffcb9d44SFilipe Manana */ 1197ffcb9d44SFilipe Manana ret = remove_block_group_free_space(trans, block_group); 1198ffcb9d44SFilipe Manana if (ret) 1199ffcb9d44SFilipe Manana goto out; 1200ffcb9d44SFilipe Manana 1201ffcb9d44SFilipe Manana ret = remove_block_group_item(trans, path, block_group); 1202ffcb9d44SFilipe Manana if (ret < 0) 1203ffcb9d44SFilipe Manana goto out; 1204ffcb9d44SFilipe Manana 1205e3e0520bSJosef Bacik spin_lock(&block_group->lock); 12063349b57fSJosef Bacik set_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags); 12073349b57fSJosef Bacik 1208e3e0520bSJosef Bacik /* 12096b7304afSFilipe Manana * At this point trimming or scrub can't start on this block group, 12106b7304afSFilipe Manana * because we removed the block group from the rbtree 12116b7304afSFilipe Manana * fs_info->block_group_cache_tree so no one can't find it anymore and 12126b7304afSFilipe Manana * even if someone already got this block group before we removed it 12136b7304afSFilipe Manana * from the rbtree, they have already incremented block_group->frozen - 12146b7304afSFilipe Manana * if they didn't, for the trimming case they won't find any free space 12156b7304afSFilipe Manana * entries because we already removed them all when we called 12166b7304afSFilipe Manana * btrfs_remove_free_space_cache(). 1217e3e0520bSJosef Bacik * 1218e3e0520bSJosef Bacik * And we must not remove the extent map from the fs_info->mapping_tree 1219e3e0520bSJosef Bacik * to prevent the same logical address range and physical device space 12206b7304afSFilipe Manana * ranges from being reused for a new block group. This is needed to 12216b7304afSFilipe Manana * avoid races with trimming and scrub. 12226b7304afSFilipe Manana * 12236b7304afSFilipe Manana * An fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is 1224e3e0520bSJosef Bacik * completely transactionless, so while it is trimming a range the 1225e3e0520bSJosef Bacik * currently running transaction might finish and a new one start, 1226e3e0520bSJosef Bacik * allowing for new block groups to be created that can reuse the same 1227e3e0520bSJosef Bacik * physical device locations unless we take this special care. 1228e3e0520bSJosef Bacik * 1229e3e0520bSJosef Bacik * There may also be an implicit trim operation if the file system 1230e3e0520bSJosef Bacik * is mounted with -odiscard. The same protections must remain 1231e3e0520bSJosef Bacik * in place until the extents have been discarded completely when 1232e3e0520bSJosef Bacik * the transaction commit has completed. 1233e3e0520bSJosef Bacik */ 12346b7304afSFilipe Manana remove_em = (atomic_read(&block_group->frozen) == 0); 1235e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1236e3e0520bSJosef Bacik 1237e3e0520bSJosef Bacik if (remove_em) { 1238e3e0520bSJosef Bacik struct extent_map_tree *em_tree; 1239e3e0520bSJosef Bacik 1240e3e0520bSJosef Bacik em_tree = &fs_info->mapping_tree; 1241e3e0520bSJosef Bacik write_lock(&em_tree->lock); 1242e3e0520bSJosef Bacik remove_extent_mapping(em_tree, em); 1243e3e0520bSJosef Bacik write_unlock(&em_tree->lock); 1244e3e0520bSJosef Bacik /* once for the tree */ 1245e3e0520bSJosef Bacik free_extent_map(em); 1246e3e0520bSJosef Bacik } 1247f6033c5eSXiyu Yang 12489fecd132SFilipe Manana out: 1249f6033c5eSXiyu Yang /* Once for the lookup reference */ 1250f6033c5eSXiyu Yang btrfs_put_block_group(block_group); 1251e3e0520bSJosef Bacik if (remove_rsv) 1252e3e0520bSJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 1253e3e0520bSJosef Bacik btrfs_free_path(path); 1254e3e0520bSJosef Bacik return ret; 1255e3e0520bSJosef Bacik } 1256e3e0520bSJosef Bacik 1257e3e0520bSJosef Bacik struct btrfs_trans_handle *btrfs_start_trans_remove_block_group( 1258e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info, const u64 chunk_offset) 1259e3e0520bSJosef Bacik { 1260dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 1261e3e0520bSJosef Bacik struct extent_map_tree *em_tree = &fs_info->mapping_tree; 1262e3e0520bSJosef Bacik struct extent_map *em; 1263e3e0520bSJosef Bacik struct map_lookup *map; 1264e3e0520bSJosef Bacik unsigned int num_items; 1265e3e0520bSJosef Bacik 1266e3e0520bSJosef Bacik read_lock(&em_tree->lock); 1267e3e0520bSJosef Bacik em = lookup_extent_mapping(em_tree, chunk_offset, 1); 1268e3e0520bSJosef Bacik read_unlock(&em_tree->lock); 1269e3e0520bSJosef Bacik ASSERT(em && em->start == chunk_offset); 1270e3e0520bSJosef Bacik 1271e3e0520bSJosef Bacik /* 1272e3e0520bSJosef Bacik * We need to reserve 3 + N units from the metadata space info in order 1273e3e0520bSJosef Bacik * to remove a block group (done at btrfs_remove_chunk() and at 1274e3e0520bSJosef Bacik * btrfs_remove_block_group()), which are used for: 1275e3e0520bSJosef Bacik * 1276e3e0520bSJosef Bacik * 1 unit for adding the free space inode's orphan (located in the tree 1277e3e0520bSJosef Bacik * of tree roots). 1278e3e0520bSJosef Bacik * 1 unit for deleting the block group item (located in the extent 1279e3e0520bSJosef Bacik * tree). 1280e3e0520bSJosef Bacik * 1 unit for deleting the free space item (located in tree of tree 1281e3e0520bSJosef Bacik * roots). 1282e3e0520bSJosef Bacik * N units for deleting N device extent items corresponding to each 1283e3e0520bSJosef Bacik * stripe (located in the device tree). 1284e3e0520bSJosef Bacik * 1285e3e0520bSJosef Bacik * In order to remove a block group we also need to reserve units in the 1286e3e0520bSJosef Bacik * system space info in order to update the chunk tree (update one or 1287e3e0520bSJosef Bacik * more device items and remove one chunk item), but this is done at 1288e3e0520bSJosef Bacik * btrfs_remove_chunk() through a call to check_system_chunk(). 1289e3e0520bSJosef Bacik */ 1290e3e0520bSJosef Bacik map = em->map_lookup; 1291e3e0520bSJosef Bacik num_items = 3 + map->num_stripes; 1292e3e0520bSJosef Bacik free_extent_map(em); 1293e3e0520bSJosef Bacik 1294dfe8aec4SJosef Bacik return btrfs_start_transaction_fallback_global_rsv(root, num_items); 1295e3e0520bSJosef Bacik } 1296e3e0520bSJosef Bacik 1297e3e0520bSJosef Bacik /* 129826ce2095SJosef Bacik * Mark block group @cache read-only, so later write won't happen to block 129926ce2095SJosef Bacik * group @cache. 130026ce2095SJosef Bacik * 130126ce2095SJosef Bacik * If @force is not set, this function will only mark the block group readonly 130226ce2095SJosef Bacik * if we have enough free space (1M) in other metadata/system block groups. 130326ce2095SJosef Bacik * If @force is not set, this function will mark the block group readonly 130426ce2095SJosef Bacik * without checking free space. 130526ce2095SJosef Bacik * 130626ce2095SJosef Bacik * NOTE: This function doesn't care if other block groups can contain all the 130726ce2095SJosef Bacik * data in this block group. That check should be done by relocation routine, 130826ce2095SJosef Bacik * not this function. 130926ce2095SJosef Bacik */ 131032da5386SDavid Sterba static int inc_block_group_ro(struct btrfs_block_group *cache, int force) 131126ce2095SJosef Bacik { 131226ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 131326ce2095SJosef Bacik u64 num_bytes; 131426ce2095SJosef Bacik int ret = -ENOSPC; 131526ce2095SJosef Bacik 131626ce2095SJosef Bacik spin_lock(&sinfo->lock); 131726ce2095SJosef Bacik spin_lock(&cache->lock); 131826ce2095SJosef Bacik 1319195a49eaSFilipe Manana if (cache->swap_extents) { 1320195a49eaSFilipe Manana ret = -ETXTBSY; 1321195a49eaSFilipe Manana goto out; 1322195a49eaSFilipe Manana } 1323195a49eaSFilipe Manana 132426ce2095SJosef Bacik if (cache->ro) { 132526ce2095SJosef Bacik cache->ro++; 132626ce2095SJosef Bacik ret = 0; 132726ce2095SJosef Bacik goto out; 132826ce2095SJosef Bacik } 132926ce2095SJosef Bacik 1330b3470b5dSDavid Sterba num_bytes = cache->length - cache->reserved - cache->pinned - 1331169e0da9SNaohiro Aota cache->bytes_super - cache->zone_unusable - cache->used; 133226ce2095SJosef Bacik 133326ce2095SJosef Bacik /* 1334a30a3d20SJosef Bacik * Data never overcommits, even in mixed mode, so do just the straight 1335a30a3d20SJosef Bacik * check of left over space in how much we have allocated. 1336a30a3d20SJosef Bacik */ 1337a30a3d20SJosef Bacik if (force) { 1338a30a3d20SJosef Bacik ret = 0; 1339a30a3d20SJosef Bacik } else if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) { 1340a30a3d20SJosef Bacik u64 sinfo_used = btrfs_space_info_used(sinfo, true); 1341a30a3d20SJosef Bacik 1342a30a3d20SJosef Bacik /* 134326ce2095SJosef Bacik * Here we make sure if we mark this bg RO, we still have enough 1344f8935566SJosef Bacik * free space as buffer. 134526ce2095SJosef Bacik */ 1346a30a3d20SJosef Bacik if (sinfo_used + num_bytes <= sinfo->total_bytes) 1347a30a3d20SJosef Bacik ret = 0; 1348a30a3d20SJosef Bacik } else { 1349a30a3d20SJosef Bacik /* 1350a30a3d20SJosef Bacik * We overcommit metadata, so we need to do the 1351a30a3d20SJosef Bacik * btrfs_can_overcommit check here, and we need to pass in 1352a30a3d20SJosef Bacik * BTRFS_RESERVE_NO_FLUSH to give ourselves the most amount of 1353a30a3d20SJosef Bacik * leeway to allow us to mark this block group as read only. 1354a30a3d20SJosef Bacik */ 1355a30a3d20SJosef Bacik if (btrfs_can_overcommit(cache->fs_info, sinfo, num_bytes, 1356a30a3d20SJosef Bacik BTRFS_RESERVE_NO_FLUSH)) 1357a30a3d20SJosef Bacik ret = 0; 1358a30a3d20SJosef Bacik } 1359a30a3d20SJosef Bacik 1360a30a3d20SJosef Bacik if (!ret) { 136126ce2095SJosef Bacik sinfo->bytes_readonly += num_bytes; 1362169e0da9SNaohiro Aota if (btrfs_is_zoned(cache->fs_info)) { 1363169e0da9SNaohiro Aota /* Migrate zone_unusable bytes to readonly */ 1364169e0da9SNaohiro Aota sinfo->bytes_readonly += cache->zone_unusable; 1365169e0da9SNaohiro Aota sinfo->bytes_zone_unusable -= cache->zone_unusable; 1366169e0da9SNaohiro Aota cache->zone_unusable = 0; 1367169e0da9SNaohiro Aota } 136826ce2095SJosef Bacik cache->ro++; 136926ce2095SJosef Bacik list_add_tail(&cache->ro_list, &sinfo->ro_bgs); 137026ce2095SJosef Bacik } 137126ce2095SJosef Bacik out: 137226ce2095SJosef Bacik spin_unlock(&cache->lock); 137326ce2095SJosef Bacik spin_unlock(&sinfo->lock); 137426ce2095SJosef Bacik if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { 137526ce2095SJosef Bacik btrfs_info(cache->fs_info, 1376b3470b5dSDavid Sterba "unable to make block group %llu ro", cache->start); 137726ce2095SJosef Bacik btrfs_dump_space_info(cache->fs_info, cache->space_info, 0, 0); 137826ce2095SJosef Bacik } 137926ce2095SJosef Bacik return ret; 138026ce2095SJosef Bacik } 138126ce2095SJosef Bacik 1382fe119a6eSNikolay Borisov static bool clean_pinned_extents(struct btrfs_trans_handle *trans, 1383fe119a6eSNikolay Borisov struct btrfs_block_group *bg) 138445bb5d6aSNikolay Borisov { 138545bb5d6aSNikolay Borisov struct btrfs_fs_info *fs_info = bg->fs_info; 1386fe119a6eSNikolay Borisov struct btrfs_transaction *prev_trans = NULL; 138745bb5d6aSNikolay Borisov const u64 start = bg->start; 138845bb5d6aSNikolay Borisov const u64 end = start + bg->length - 1; 138945bb5d6aSNikolay Borisov int ret; 139045bb5d6aSNikolay Borisov 1391fe119a6eSNikolay Borisov spin_lock(&fs_info->trans_lock); 1392fe119a6eSNikolay Borisov if (trans->transaction->list.prev != &fs_info->trans_list) { 1393fe119a6eSNikolay Borisov prev_trans = list_last_entry(&trans->transaction->list, 1394fe119a6eSNikolay Borisov struct btrfs_transaction, list); 1395fe119a6eSNikolay Borisov refcount_inc(&prev_trans->use_count); 1396fe119a6eSNikolay Borisov } 1397fe119a6eSNikolay Borisov spin_unlock(&fs_info->trans_lock); 1398fe119a6eSNikolay Borisov 139945bb5d6aSNikolay Borisov /* 140045bb5d6aSNikolay Borisov * Hold the unused_bg_unpin_mutex lock to avoid racing with 140145bb5d6aSNikolay Borisov * btrfs_finish_extent_commit(). If we are at transaction N, another 140245bb5d6aSNikolay Borisov * task might be running finish_extent_commit() for the previous 140345bb5d6aSNikolay Borisov * transaction N - 1, and have seen a range belonging to the block 1404fe119a6eSNikolay Borisov * group in pinned_extents before we were able to clear the whole block 1405fe119a6eSNikolay Borisov * group range from pinned_extents. This means that task can lookup for 1406fe119a6eSNikolay Borisov * the block group after we unpinned it from pinned_extents and removed 1407fe119a6eSNikolay Borisov * it, leading to a BUG_ON() at unpin_extent_range(). 140845bb5d6aSNikolay Borisov */ 140945bb5d6aSNikolay Borisov mutex_lock(&fs_info->unused_bg_unpin_mutex); 1410fe119a6eSNikolay Borisov if (prev_trans) { 1411fe119a6eSNikolay Borisov ret = clear_extent_bits(&prev_trans->pinned_extents, start, end, 141245bb5d6aSNikolay Borisov EXTENT_DIRTY); 141345bb5d6aSNikolay Borisov if (ret) 1414534cf531SFilipe Manana goto out; 1415fe119a6eSNikolay Borisov } 141645bb5d6aSNikolay Borisov 1417fe119a6eSNikolay Borisov ret = clear_extent_bits(&trans->transaction->pinned_extents, start, end, 141845bb5d6aSNikolay Borisov EXTENT_DIRTY); 1419534cf531SFilipe Manana out: 142045bb5d6aSNikolay Borisov mutex_unlock(&fs_info->unused_bg_unpin_mutex); 14215150bf19SFilipe Manana if (prev_trans) 14225150bf19SFilipe Manana btrfs_put_transaction(prev_trans); 142345bb5d6aSNikolay Borisov 1424534cf531SFilipe Manana return ret == 0; 142545bb5d6aSNikolay Borisov } 142645bb5d6aSNikolay Borisov 142726ce2095SJosef Bacik /* 1428e3e0520bSJosef Bacik * Process the unused_bgs list and remove any that don't have any allocated 1429e3e0520bSJosef Bacik * space inside of them. 1430e3e0520bSJosef Bacik */ 1431e3e0520bSJosef Bacik void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) 1432e3e0520bSJosef Bacik { 143332da5386SDavid Sterba struct btrfs_block_group *block_group; 1434e3e0520bSJosef Bacik struct btrfs_space_info *space_info; 1435e3e0520bSJosef Bacik struct btrfs_trans_handle *trans; 14366e80d4f8SDennis Zhou const bool async_trim_enabled = btrfs_test_opt(fs_info, DISCARD_ASYNC); 1437e3e0520bSJosef Bacik int ret = 0; 1438e3e0520bSJosef Bacik 1439e3e0520bSJosef Bacik if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 1440e3e0520bSJosef Bacik return; 1441e3e0520bSJosef Bacik 14422f12741fSJosef Bacik if (btrfs_fs_closing(fs_info)) 14432f12741fSJosef Bacik return; 14442f12741fSJosef Bacik 1445ddfd08cbSJosef Bacik /* 1446ddfd08cbSJosef Bacik * Long running balances can keep us blocked here for eternity, so 1447ddfd08cbSJosef Bacik * simply skip deletion if we're unable to get the mutex. 1448ddfd08cbSJosef Bacik */ 1449f3372065SJohannes Thumshirn if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) 1450ddfd08cbSJosef Bacik return; 1451ddfd08cbSJosef Bacik 1452e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1453e3e0520bSJosef Bacik while (!list_empty(&fs_info->unused_bgs)) { 1454e3e0520bSJosef Bacik int trimming; 1455e3e0520bSJosef Bacik 1456e3e0520bSJosef Bacik block_group = list_first_entry(&fs_info->unused_bgs, 145732da5386SDavid Sterba struct btrfs_block_group, 1458e3e0520bSJosef Bacik bg_list); 1459e3e0520bSJosef Bacik list_del_init(&block_group->bg_list); 1460e3e0520bSJosef Bacik 1461e3e0520bSJosef Bacik space_info = block_group->space_info; 1462e3e0520bSJosef Bacik 1463e3e0520bSJosef Bacik if (ret || btrfs_mixed_space_info(space_info)) { 1464e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1465e3e0520bSJosef Bacik continue; 1466e3e0520bSJosef Bacik } 1467e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1468e3e0520bSJosef Bacik 1469b0643e59SDennis Zhou btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 1470b0643e59SDennis Zhou 1471e3e0520bSJosef Bacik /* Don't want to race with allocators so take the groups_sem */ 1472e3e0520bSJosef Bacik down_write(&space_info->groups_sem); 14736e80d4f8SDennis Zhou 14746e80d4f8SDennis Zhou /* 14756e80d4f8SDennis Zhou * Async discard moves the final block group discard to be prior 14766e80d4f8SDennis Zhou * to the unused_bgs code path. Therefore, if it's not fully 14776e80d4f8SDennis Zhou * trimmed, punt it back to the async discard lists. 14786e80d4f8SDennis Zhou */ 14796e80d4f8SDennis Zhou if (btrfs_test_opt(fs_info, DISCARD_ASYNC) && 14806e80d4f8SDennis Zhou !btrfs_is_free_space_trimmed(block_group)) { 14816e80d4f8SDennis Zhou trace_btrfs_skip_unused_block_group(block_group); 14826e80d4f8SDennis Zhou up_write(&space_info->groups_sem); 14836e80d4f8SDennis Zhou /* Requeue if we failed because of async discard */ 14846e80d4f8SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 14856e80d4f8SDennis Zhou block_group); 14866e80d4f8SDennis Zhou goto next; 14876e80d4f8SDennis Zhou } 14886e80d4f8SDennis Zhou 1489e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1490e3e0520bSJosef Bacik if (block_group->reserved || block_group->pinned || 1491bf38be65SDavid Sterba block_group->used || block_group->ro || 1492e3e0520bSJosef Bacik list_is_singular(&block_group->list)) { 1493e3e0520bSJosef Bacik /* 1494e3e0520bSJosef Bacik * We want to bail if we made new allocations or have 1495e3e0520bSJosef Bacik * outstanding allocations in this block group. We do 1496e3e0520bSJosef Bacik * the ro check in case balance is currently acting on 1497e3e0520bSJosef Bacik * this block group. 1498e3e0520bSJosef Bacik */ 1499e3e0520bSJosef Bacik trace_btrfs_skip_unused_block_group(block_group); 1500e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1501e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1502e3e0520bSJosef Bacik goto next; 1503e3e0520bSJosef Bacik } 1504e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1505e3e0520bSJosef Bacik 1506e3e0520bSJosef Bacik /* We don't want to force the issue, only flip if it's ok. */ 1507e11c0406SJosef Bacik ret = inc_block_group_ro(block_group, 0); 1508e3e0520bSJosef Bacik up_write(&space_info->groups_sem); 1509e3e0520bSJosef Bacik if (ret < 0) { 1510e3e0520bSJosef Bacik ret = 0; 1511e3e0520bSJosef Bacik goto next; 1512e3e0520bSJosef Bacik } 1513e3e0520bSJosef Bacik 151474e91b12SNaohiro Aota ret = btrfs_zone_finish(block_group); 151574e91b12SNaohiro Aota if (ret < 0) { 151674e91b12SNaohiro Aota btrfs_dec_block_group_ro(block_group); 151774e91b12SNaohiro Aota if (ret == -EAGAIN) 151874e91b12SNaohiro Aota ret = 0; 151974e91b12SNaohiro Aota goto next; 152074e91b12SNaohiro Aota } 152174e91b12SNaohiro Aota 1522e3e0520bSJosef Bacik /* 1523e3e0520bSJosef Bacik * Want to do this before we do anything else so we can recover 1524e3e0520bSJosef Bacik * properly if we fail to join the transaction. 1525e3e0520bSJosef Bacik */ 1526e3e0520bSJosef Bacik trans = btrfs_start_trans_remove_block_group(fs_info, 1527b3470b5dSDavid Sterba block_group->start); 1528e3e0520bSJosef Bacik if (IS_ERR(trans)) { 1529e3e0520bSJosef Bacik btrfs_dec_block_group_ro(block_group); 1530e3e0520bSJosef Bacik ret = PTR_ERR(trans); 1531e3e0520bSJosef Bacik goto next; 1532e3e0520bSJosef Bacik } 1533e3e0520bSJosef Bacik 1534e3e0520bSJosef Bacik /* 1535e3e0520bSJosef Bacik * We could have pending pinned extents for this block group, 1536e3e0520bSJosef Bacik * just delete them, we don't care about them anymore. 1537e3e0520bSJosef Bacik */ 1538534cf531SFilipe Manana if (!clean_pinned_extents(trans, block_group)) { 1539534cf531SFilipe Manana btrfs_dec_block_group_ro(block_group); 1540e3e0520bSJosef Bacik goto end_trans; 1541534cf531SFilipe Manana } 1542e3e0520bSJosef Bacik 1543b0643e59SDennis Zhou /* 1544b0643e59SDennis Zhou * At this point, the block_group is read only and should fail 1545b0643e59SDennis Zhou * new allocations. However, btrfs_finish_extent_commit() can 1546b0643e59SDennis Zhou * cause this block_group to be placed back on the discard 1547b0643e59SDennis Zhou * lists because now the block_group isn't fully discarded. 1548b0643e59SDennis Zhou * Bail here and try again later after discarding everything. 1549b0643e59SDennis Zhou */ 1550b0643e59SDennis Zhou spin_lock(&fs_info->discard_ctl.lock); 1551b0643e59SDennis Zhou if (!list_empty(&block_group->discard_list)) { 1552b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1553b0643e59SDennis Zhou btrfs_dec_block_group_ro(block_group); 1554b0643e59SDennis Zhou btrfs_discard_queue_work(&fs_info->discard_ctl, 1555b0643e59SDennis Zhou block_group); 1556b0643e59SDennis Zhou goto end_trans; 1557b0643e59SDennis Zhou } 1558b0643e59SDennis Zhou spin_unlock(&fs_info->discard_ctl.lock); 1559b0643e59SDennis Zhou 1560e3e0520bSJosef Bacik /* Reset pinned so btrfs_put_block_group doesn't complain */ 1561e3e0520bSJosef Bacik spin_lock(&space_info->lock); 1562e3e0520bSJosef Bacik spin_lock(&block_group->lock); 1563e3e0520bSJosef Bacik 1564e3e0520bSJosef Bacik btrfs_space_info_update_bytes_pinned(fs_info, space_info, 1565e3e0520bSJosef Bacik -block_group->pinned); 1566e3e0520bSJosef Bacik space_info->bytes_readonly += block_group->pinned; 1567e3e0520bSJosef Bacik block_group->pinned = 0; 1568e3e0520bSJosef Bacik 1569e3e0520bSJosef Bacik spin_unlock(&block_group->lock); 1570e3e0520bSJosef Bacik spin_unlock(&space_info->lock); 1571e3e0520bSJosef Bacik 15726e80d4f8SDennis Zhou /* 15736e80d4f8SDennis Zhou * The normal path here is an unused block group is passed here, 15746e80d4f8SDennis Zhou * then trimming is handled in the transaction commit path. 15756e80d4f8SDennis Zhou * Async discard interposes before this to do the trimming 15766e80d4f8SDennis Zhou * before coming down the unused block group path as trimming 15776e80d4f8SDennis Zhou * will no longer be done later in the transaction commit path. 15786e80d4f8SDennis Zhou */ 15796e80d4f8SDennis Zhou if (!async_trim_enabled && btrfs_test_opt(fs_info, DISCARD_ASYNC)) 15806e80d4f8SDennis Zhou goto flip_async; 15816e80d4f8SDennis Zhou 1582dcba6e48SNaohiro Aota /* 1583dcba6e48SNaohiro Aota * DISCARD can flip during remount. On zoned filesystems, we 1584dcba6e48SNaohiro Aota * need to reset sequential-required zones. 1585dcba6e48SNaohiro Aota */ 1586dcba6e48SNaohiro Aota trimming = btrfs_test_opt(fs_info, DISCARD_SYNC) || 1587dcba6e48SNaohiro Aota btrfs_is_zoned(fs_info); 1588e3e0520bSJosef Bacik 1589e3e0520bSJosef Bacik /* Implicit trim during transaction commit. */ 1590e3e0520bSJosef Bacik if (trimming) 15916b7304afSFilipe Manana btrfs_freeze_block_group(block_group); 1592e3e0520bSJosef Bacik 1593e3e0520bSJosef Bacik /* 1594e3e0520bSJosef Bacik * Btrfs_remove_chunk will abort the transaction if things go 1595e3e0520bSJosef Bacik * horribly wrong. 1596e3e0520bSJosef Bacik */ 1597b3470b5dSDavid Sterba ret = btrfs_remove_chunk(trans, block_group->start); 1598e3e0520bSJosef Bacik 1599e3e0520bSJosef Bacik if (ret) { 1600e3e0520bSJosef Bacik if (trimming) 16016b7304afSFilipe Manana btrfs_unfreeze_block_group(block_group); 1602e3e0520bSJosef Bacik goto end_trans; 1603e3e0520bSJosef Bacik } 1604e3e0520bSJosef Bacik 1605e3e0520bSJosef Bacik /* 1606e3e0520bSJosef Bacik * If we're not mounted with -odiscard, we can just forget 1607e3e0520bSJosef Bacik * about this block group. Otherwise we'll need to wait 1608e3e0520bSJosef Bacik * until transaction commit to do the actual discard. 1609e3e0520bSJosef Bacik */ 1610e3e0520bSJosef Bacik if (trimming) { 1611e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1612e3e0520bSJosef Bacik /* 1613e3e0520bSJosef Bacik * A concurrent scrub might have added us to the list 1614e3e0520bSJosef Bacik * fs_info->unused_bgs, so use a list_move operation 1615e3e0520bSJosef Bacik * to add the block group to the deleted_bgs list. 1616e3e0520bSJosef Bacik */ 1617e3e0520bSJosef Bacik list_move(&block_group->bg_list, 1618e3e0520bSJosef Bacik &trans->transaction->deleted_bgs); 1619e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1620e3e0520bSJosef Bacik btrfs_get_block_group(block_group); 1621e3e0520bSJosef Bacik } 1622e3e0520bSJosef Bacik end_trans: 1623e3e0520bSJosef Bacik btrfs_end_transaction(trans); 1624e3e0520bSJosef Bacik next: 1625e3e0520bSJosef Bacik btrfs_put_block_group(block_group); 1626e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1627e3e0520bSJosef Bacik } 1628e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1629f3372065SJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 16306e80d4f8SDennis Zhou return; 16316e80d4f8SDennis Zhou 16326e80d4f8SDennis Zhou flip_async: 16336e80d4f8SDennis Zhou btrfs_end_transaction(trans); 1634f3372065SJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 16356e80d4f8SDennis Zhou btrfs_put_block_group(block_group); 16366e80d4f8SDennis Zhou btrfs_discard_punt_unused_bgs_list(fs_info); 1637e3e0520bSJosef Bacik } 1638e3e0520bSJosef Bacik 163932da5386SDavid Sterba void btrfs_mark_bg_unused(struct btrfs_block_group *bg) 1640e3e0520bSJosef Bacik { 1641e3e0520bSJosef Bacik struct btrfs_fs_info *fs_info = bg->fs_info; 1642e3e0520bSJosef Bacik 1643e3e0520bSJosef Bacik spin_lock(&fs_info->unused_bgs_lock); 1644e3e0520bSJosef Bacik if (list_empty(&bg->bg_list)) { 1645e3e0520bSJosef Bacik btrfs_get_block_group(bg); 1646*0657b20cSFilipe Manana trace_btrfs_add_unused_block_group(bg); 1647e3e0520bSJosef Bacik list_add_tail(&bg->bg_list, &fs_info->unused_bgs); 1648*0657b20cSFilipe Manana } else if (!test_bit(BLOCK_GROUP_FLAG_NEW, &bg->runtime_flags)) { 1649a9f18971SNaohiro Aota /* Pull out the block group from the reclaim_bgs list. */ 1650*0657b20cSFilipe Manana trace_btrfs_add_unused_block_group(bg); 1651a9f18971SNaohiro Aota list_move_tail(&bg->bg_list, &fs_info->unused_bgs); 1652e3e0520bSJosef Bacik } 1653e3e0520bSJosef Bacik spin_unlock(&fs_info->unused_bgs_lock); 1654e3e0520bSJosef Bacik } 16554358d963SJosef Bacik 16562ca0ec77SJohannes Thumshirn /* 16572ca0ec77SJohannes Thumshirn * We want block groups with a low number of used bytes to be in the beginning 16582ca0ec77SJohannes Thumshirn * of the list, so they will get reclaimed first. 16592ca0ec77SJohannes Thumshirn */ 16602ca0ec77SJohannes Thumshirn static int reclaim_bgs_cmp(void *unused, const struct list_head *a, 16612ca0ec77SJohannes Thumshirn const struct list_head *b) 16622ca0ec77SJohannes Thumshirn { 16632ca0ec77SJohannes Thumshirn const struct btrfs_block_group *bg1, *bg2; 16642ca0ec77SJohannes Thumshirn 16652ca0ec77SJohannes Thumshirn bg1 = list_entry(a, struct btrfs_block_group, bg_list); 16662ca0ec77SJohannes Thumshirn bg2 = list_entry(b, struct btrfs_block_group, bg_list); 16672ca0ec77SJohannes Thumshirn 16682ca0ec77SJohannes Thumshirn return bg1->used > bg2->used; 16692ca0ec77SJohannes Thumshirn } 16702ca0ec77SJohannes Thumshirn 16713687fcb0SJohannes Thumshirn static inline bool btrfs_should_reclaim(struct btrfs_fs_info *fs_info) 16723687fcb0SJohannes Thumshirn { 16733687fcb0SJohannes Thumshirn if (btrfs_is_zoned(fs_info)) 16743687fcb0SJohannes Thumshirn return btrfs_zoned_should_reclaim(fs_info); 16753687fcb0SJohannes Thumshirn return true; 16763687fcb0SJohannes Thumshirn } 16773687fcb0SJohannes Thumshirn 167881531225SBoris Burkov static bool should_reclaim_block_group(struct btrfs_block_group *bg, u64 bytes_freed) 167981531225SBoris Burkov { 168081531225SBoris Burkov const struct btrfs_space_info *space_info = bg->space_info; 168181531225SBoris Burkov const int reclaim_thresh = READ_ONCE(space_info->bg_reclaim_threshold); 168281531225SBoris Burkov const u64 new_val = bg->used; 168381531225SBoris Burkov const u64 old_val = new_val + bytes_freed; 168481531225SBoris Burkov u64 thresh; 168581531225SBoris Burkov 168681531225SBoris Burkov if (reclaim_thresh == 0) 168781531225SBoris Burkov return false; 168881531225SBoris Burkov 1689428c8e03SDavid Sterba thresh = mult_perc(bg->length, reclaim_thresh); 169081531225SBoris Burkov 169181531225SBoris Burkov /* 169281531225SBoris Burkov * If we were below the threshold before don't reclaim, we are likely a 169381531225SBoris Burkov * brand new block group and we don't want to relocate new block groups. 169481531225SBoris Burkov */ 169581531225SBoris Burkov if (old_val < thresh) 169681531225SBoris Burkov return false; 169781531225SBoris Burkov if (new_val >= thresh) 169881531225SBoris Burkov return false; 169981531225SBoris Burkov return true; 170081531225SBoris Burkov } 170181531225SBoris Burkov 170218bb8bbfSJohannes Thumshirn void btrfs_reclaim_bgs_work(struct work_struct *work) 170318bb8bbfSJohannes Thumshirn { 170418bb8bbfSJohannes Thumshirn struct btrfs_fs_info *fs_info = 170518bb8bbfSJohannes Thumshirn container_of(work, struct btrfs_fs_info, reclaim_bgs_work); 170618bb8bbfSJohannes Thumshirn struct btrfs_block_group *bg; 170718bb8bbfSJohannes Thumshirn struct btrfs_space_info *space_info; 170818bb8bbfSJohannes Thumshirn 170918bb8bbfSJohannes Thumshirn if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 171018bb8bbfSJohannes Thumshirn return; 171118bb8bbfSJohannes Thumshirn 17122f12741fSJosef Bacik if (btrfs_fs_closing(fs_info)) 17132f12741fSJosef Bacik return; 17142f12741fSJosef Bacik 17153687fcb0SJohannes Thumshirn if (!btrfs_should_reclaim(fs_info)) 17163687fcb0SJohannes Thumshirn return; 17173687fcb0SJohannes Thumshirn 1718ca5e4ea0SNaohiro Aota sb_start_write(fs_info->sb); 1719ca5e4ea0SNaohiro Aota 1720ca5e4ea0SNaohiro Aota if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 1721ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 172218bb8bbfSJohannes Thumshirn return; 1723ca5e4ea0SNaohiro Aota } 172418bb8bbfSJohannes Thumshirn 17259cc0b837SJohannes Thumshirn /* 17269cc0b837SJohannes Thumshirn * Long running balances can keep us blocked here for eternity, so 17279cc0b837SJohannes Thumshirn * simply skip reclaim if we're unable to get the mutex. 17289cc0b837SJohannes Thumshirn */ 17299cc0b837SJohannes Thumshirn if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) { 17309cc0b837SJohannes Thumshirn btrfs_exclop_finish(fs_info); 1731ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 17329cc0b837SJohannes Thumshirn return; 17339cc0b837SJohannes Thumshirn } 17349cc0b837SJohannes Thumshirn 173518bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 17362ca0ec77SJohannes Thumshirn /* 17372ca0ec77SJohannes Thumshirn * Sort happens under lock because we can't simply splice it and sort. 17382ca0ec77SJohannes Thumshirn * The block groups might still be in use and reachable via bg_list, 17392ca0ec77SJohannes Thumshirn * and their presence in the reclaim_bgs list must be preserved. 17402ca0ec77SJohannes Thumshirn */ 17412ca0ec77SJohannes Thumshirn list_sort(NULL, &fs_info->reclaim_bgs, reclaim_bgs_cmp); 174218bb8bbfSJohannes Thumshirn while (!list_empty(&fs_info->reclaim_bgs)) { 17435f93e776SJohannes Thumshirn u64 zone_unusable; 17441cea5cf0SFilipe Manana int ret = 0; 17451cea5cf0SFilipe Manana 174618bb8bbfSJohannes Thumshirn bg = list_first_entry(&fs_info->reclaim_bgs, 174718bb8bbfSJohannes Thumshirn struct btrfs_block_group, 174818bb8bbfSJohannes Thumshirn bg_list); 174918bb8bbfSJohannes Thumshirn list_del_init(&bg->bg_list); 175018bb8bbfSJohannes Thumshirn 175118bb8bbfSJohannes Thumshirn space_info = bg->space_info; 175218bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 175318bb8bbfSJohannes Thumshirn 175418bb8bbfSJohannes Thumshirn /* Don't race with allocators so take the groups_sem */ 175518bb8bbfSJohannes Thumshirn down_write(&space_info->groups_sem); 175618bb8bbfSJohannes Thumshirn 175718bb8bbfSJohannes Thumshirn spin_lock(&bg->lock); 175818bb8bbfSJohannes Thumshirn if (bg->reserved || bg->pinned || bg->ro) { 175918bb8bbfSJohannes Thumshirn /* 176018bb8bbfSJohannes Thumshirn * We want to bail if we made new allocations or have 176118bb8bbfSJohannes Thumshirn * outstanding allocations in this block group. We do 176218bb8bbfSJohannes Thumshirn * the ro check in case balance is currently acting on 176318bb8bbfSJohannes Thumshirn * this block group. 176418bb8bbfSJohannes Thumshirn */ 176518bb8bbfSJohannes Thumshirn spin_unlock(&bg->lock); 176618bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 176718bb8bbfSJohannes Thumshirn goto next; 176818bb8bbfSJohannes Thumshirn } 1769cc4804bfSBoris Burkov if (bg->used == 0) { 1770cc4804bfSBoris Burkov /* 1771cc4804bfSBoris Burkov * It is possible that we trigger relocation on a block 1772cc4804bfSBoris Burkov * group as its extents are deleted and it first goes 1773cc4804bfSBoris Burkov * below the threshold, then shortly after goes empty. 1774cc4804bfSBoris Burkov * 1775cc4804bfSBoris Burkov * In this case, relocating it does delete it, but has 1776cc4804bfSBoris Burkov * some overhead in relocation specific metadata, looking 1777cc4804bfSBoris Burkov * for the non-existent extents and running some extra 1778cc4804bfSBoris Burkov * transactions, which we can avoid by using one of the 1779cc4804bfSBoris Burkov * other mechanisms for dealing with empty block groups. 1780cc4804bfSBoris Burkov */ 1781cc4804bfSBoris Burkov if (!btrfs_test_opt(fs_info, DISCARD_ASYNC)) 1782cc4804bfSBoris Burkov btrfs_mark_bg_unused(bg); 1783cc4804bfSBoris Burkov spin_unlock(&bg->lock); 1784cc4804bfSBoris Burkov up_write(&space_info->groups_sem); 1785cc4804bfSBoris Burkov goto next; 178681531225SBoris Burkov 178781531225SBoris Burkov } 178881531225SBoris Burkov /* 178981531225SBoris Burkov * The block group might no longer meet the reclaim condition by 179081531225SBoris Burkov * the time we get around to reclaiming it, so to avoid 179181531225SBoris Burkov * reclaiming overly full block_groups, skip reclaiming them. 179281531225SBoris Burkov * 179381531225SBoris Burkov * Since the decision making process also depends on the amount 179481531225SBoris Burkov * being freed, pass in a fake giant value to skip that extra 179581531225SBoris Burkov * check, which is more meaningful when adding to the list in 179681531225SBoris Burkov * the first place. 179781531225SBoris Burkov */ 179881531225SBoris Burkov if (!should_reclaim_block_group(bg, bg->length)) { 179981531225SBoris Burkov spin_unlock(&bg->lock); 180081531225SBoris Burkov up_write(&space_info->groups_sem); 180181531225SBoris Burkov goto next; 1802cc4804bfSBoris Burkov } 180318bb8bbfSJohannes Thumshirn spin_unlock(&bg->lock); 180418bb8bbfSJohannes Thumshirn 180593463ff7SNaohiro Aota /* 180693463ff7SNaohiro Aota * Get out fast, in case we're read-only or unmounting the 180793463ff7SNaohiro Aota * filesystem. It is OK to drop block groups from the list even 180893463ff7SNaohiro Aota * for the read-only case. As we did sb_start_write(), 180993463ff7SNaohiro Aota * "mount -o remount,ro" won't happen and read-only filesystem 181093463ff7SNaohiro Aota * means it is forced read-only due to a fatal error. So, it 181193463ff7SNaohiro Aota * never gets back to read-write to let us reclaim again. 181293463ff7SNaohiro Aota */ 181393463ff7SNaohiro Aota if (btrfs_need_cleaner_sleep(fs_info)) { 181418bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 181518bb8bbfSJohannes Thumshirn goto next; 181618bb8bbfSJohannes Thumshirn } 181718bb8bbfSJohannes Thumshirn 18185f93e776SJohannes Thumshirn /* 18195f93e776SJohannes Thumshirn * Cache the zone_unusable value before turning the block group 18205f93e776SJohannes Thumshirn * to read only. As soon as the blog group is read only it's 18215f93e776SJohannes Thumshirn * zone_unusable value gets moved to the block group's read-only 18225f93e776SJohannes Thumshirn * bytes and isn't available for calculations anymore. 18235f93e776SJohannes Thumshirn */ 18245f93e776SJohannes Thumshirn zone_unusable = bg->zone_unusable; 182518bb8bbfSJohannes Thumshirn ret = inc_block_group_ro(bg, 0); 182618bb8bbfSJohannes Thumshirn up_write(&space_info->groups_sem); 182718bb8bbfSJohannes Thumshirn if (ret < 0) 182818bb8bbfSJohannes Thumshirn goto next; 182918bb8bbfSJohannes Thumshirn 18305f93e776SJohannes Thumshirn btrfs_info(fs_info, 18315f93e776SJohannes Thumshirn "reclaiming chunk %llu with %llu%% used %llu%% unusable", 183295cd356cSJohannes Thumshirn bg->start, 183395cd356cSJohannes Thumshirn div64_u64(bg->used * 100, bg->length), 18345f93e776SJohannes Thumshirn div64_u64(zone_unusable * 100, bg->length)); 183518bb8bbfSJohannes Thumshirn trace_btrfs_reclaim_block_group(bg); 183618bb8bbfSJohannes Thumshirn ret = btrfs_relocate_chunk(fs_info, bg->start); 183774944c87SJosef Bacik if (ret) { 183874944c87SJosef Bacik btrfs_dec_block_group_ro(bg); 183918bb8bbfSJohannes Thumshirn btrfs_err(fs_info, "error relocating chunk %llu", 184018bb8bbfSJohannes Thumshirn bg->start); 184174944c87SJosef Bacik } 184218bb8bbfSJohannes Thumshirn 184318bb8bbfSJohannes Thumshirn next: 18447e271809SNaohiro Aota if (ret) 18457e271809SNaohiro Aota btrfs_mark_bg_to_reclaim(bg); 18461cea5cf0SFilipe Manana btrfs_put_block_group(bg); 18473ed01616SNaohiro Aota 18483ed01616SNaohiro Aota mutex_unlock(&fs_info->reclaim_bgs_lock); 18493ed01616SNaohiro Aota /* 18503ed01616SNaohiro Aota * Reclaiming all the block groups in the list can take really 18513ed01616SNaohiro Aota * long. Prioritize cleaning up unused block groups. 18523ed01616SNaohiro Aota */ 18533ed01616SNaohiro Aota btrfs_delete_unused_bgs(fs_info); 18543ed01616SNaohiro Aota /* 18553ed01616SNaohiro Aota * If we are interrupted by a balance, we can just bail out. The 18563ed01616SNaohiro Aota * cleaner thread restart again if necessary. 18573ed01616SNaohiro Aota */ 18583ed01616SNaohiro Aota if (!mutex_trylock(&fs_info->reclaim_bgs_lock)) 18593ed01616SNaohiro Aota goto end; 1860d96b3424SFilipe Manana spin_lock(&fs_info->unused_bgs_lock); 186118bb8bbfSJohannes Thumshirn } 186218bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 186318bb8bbfSJohannes Thumshirn mutex_unlock(&fs_info->reclaim_bgs_lock); 18643ed01616SNaohiro Aota end: 186518bb8bbfSJohannes Thumshirn btrfs_exclop_finish(fs_info); 1866ca5e4ea0SNaohiro Aota sb_end_write(fs_info->sb); 186718bb8bbfSJohannes Thumshirn } 186818bb8bbfSJohannes Thumshirn 186918bb8bbfSJohannes Thumshirn void btrfs_reclaim_bgs(struct btrfs_fs_info *fs_info) 187018bb8bbfSJohannes Thumshirn { 187118bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 187218bb8bbfSJohannes Thumshirn if (!list_empty(&fs_info->reclaim_bgs)) 187318bb8bbfSJohannes Thumshirn queue_work(system_unbound_wq, &fs_info->reclaim_bgs_work); 187418bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 187518bb8bbfSJohannes Thumshirn } 187618bb8bbfSJohannes Thumshirn 187718bb8bbfSJohannes Thumshirn void btrfs_mark_bg_to_reclaim(struct btrfs_block_group *bg) 187818bb8bbfSJohannes Thumshirn { 187918bb8bbfSJohannes Thumshirn struct btrfs_fs_info *fs_info = bg->fs_info; 188018bb8bbfSJohannes Thumshirn 188118bb8bbfSJohannes Thumshirn spin_lock(&fs_info->unused_bgs_lock); 188218bb8bbfSJohannes Thumshirn if (list_empty(&bg->bg_list)) { 188318bb8bbfSJohannes Thumshirn btrfs_get_block_group(bg); 188418bb8bbfSJohannes Thumshirn trace_btrfs_add_reclaim_block_group(bg); 188518bb8bbfSJohannes Thumshirn list_add_tail(&bg->bg_list, &fs_info->reclaim_bgs); 188618bb8bbfSJohannes Thumshirn } 188718bb8bbfSJohannes Thumshirn spin_unlock(&fs_info->unused_bgs_lock); 188818bb8bbfSJohannes Thumshirn } 188918bb8bbfSJohannes Thumshirn 1890e3ba67a1SJohannes Thumshirn static int read_bg_from_eb(struct btrfs_fs_info *fs_info, struct btrfs_key *key, 1891e3ba67a1SJohannes Thumshirn struct btrfs_path *path) 1892e3ba67a1SJohannes Thumshirn { 1893e3ba67a1SJohannes Thumshirn struct extent_map_tree *em_tree; 1894e3ba67a1SJohannes Thumshirn struct extent_map *em; 1895e3ba67a1SJohannes Thumshirn struct btrfs_block_group_item bg; 1896e3ba67a1SJohannes Thumshirn struct extent_buffer *leaf; 1897e3ba67a1SJohannes Thumshirn int slot; 1898e3ba67a1SJohannes Thumshirn u64 flags; 1899e3ba67a1SJohannes Thumshirn int ret = 0; 1900e3ba67a1SJohannes Thumshirn 1901e3ba67a1SJohannes Thumshirn slot = path->slots[0]; 1902e3ba67a1SJohannes Thumshirn leaf = path->nodes[0]; 1903e3ba67a1SJohannes Thumshirn 1904e3ba67a1SJohannes Thumshirn em_tree = &fs_info->mapping_tree; 1905e3ba67a1SJohannes Thumshirn read_lock(&em_tree->lock); 1906e3ba67a1SJohannes Thumshirn em = lookup_extent_mapping(em_tree, key->objectid, key->offset); 1907e3ba67a1SJohannes Thumshirn read_unlock(&em_tree->lock); 1908e3ba67a1SJohannes Thumshirn if (!em) { 1909e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1910e3ba67a1SJohannes Thumshirn "logical %llu len %llu found bg but no related chunk", 1911e3ba67a1SJohannes Thumshirn key->objectid, key->offset); 1912e3ba67a1SJohannes Thumshirn return -ENOENT; 1913e3ba67a1SJohannes Thumshirn } 1914e3ba67a1SJohannes Thumshirn 1915e3ba67a1SJohannes Thumshirn if (em->start != key->objectid || em->len != key->offset) { 1916e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1917e3ba67a1SJohannes Thumshirn "block group %llu len %llu mismatch with chunk %llu len %llu", 1918e3ba67a1SJohannes Thumshirn key->objectid, key->offset, em->start, em->len); 1919e3ba67a1SJohannes Thumshirn ret = -EUCLEAN; 1920e3ba67a1SJohannes Thumshirn goto out_free_em; 1921e3ba67a1SJohannes Thumshirn } 1922e3ba67a1SJohannes Thumshirn 1923e3ba67a1SJohannes Thumshirn read_extent_buffer(leaf, &bg, btrfs_item_ptr_offset(leaf, slot), 1924e3ba67a1SJohannes Thumshirn sizeof(bg)); 1925e3ba67a1SJohannes Thumshirn flags = btrfs_stack_block_group_flags(&bg) & 1926e3ba67a1SJohannes Thumshirn BTRFS_BLOCK_GROUP_TYPE_MASK; 1927e3ba67a1SJohannes Thumshirn 1928e3ba67a1SJohannes Thumshirn if (flags != (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 1929e3ba67a1SJohannes Thumshirn btrfs_err(fs_info, 1930e3ba67a1SJohannes Thumshirn "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", 1931e3ba67a1SJohannes Thumshirn key->objectid, key->offset, flags, 1932e3ba67a1SJohannes Thumshirn (BTRFS_BLOCK_GROUP_TYPE_MASK & em->map_lookup->type)); 1933e3ba67a1SJohannes Thumshirn ret = -EUCLEAN; 1934e3ba67a1SJohannes Thumshirn } 1935e3ba67a1SJohannes Thumshirn 1936e3ba67a1SJohannes Thumshirn out_free_em: 1937e3ba67a1SJohannes Thumshirn free_extent_map(em); 1938e3ba67a1SJohannes Thumshirn return ret; 1939e3ba67a1SJohannes Thumshirn } 1940e3ba67a1SJohannes Thumshirn 19414358d963SJosef Bacik static int find_first_block_group(struct btrfs_fs_info *fs_info, 19424358d963SJosef Bacik struct btrfs_path *path, 19434358d963SJosef Bacik struct btrfs_key *key) 19444358d963SJosef Bacik { 1945dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 1946e3ba67a1SJohannes Thumshirn int ret; 19474358d963SJosef Bacik struct btrfs_key found_key; 19484358d963SJosef Bacik 194936dfbbe2SGabriel Niebler btrfs_for_each_slot(root, key, &found_key, path, ret) { 19504358d963SJosef Bacik if (found_key.objectid >= key->objectid && 19514358d963SJosef Bacik found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { 195236dfbbe2SGabriel Niebler return read_bg_from_eb(fs_info, &found_key, path); 1953e3ba67a1SJohannes Thumshirn } 19544358d963SJosef Bacik } 19554358d963SJosef Bacik return ret; 19564358d963SJosef Bacik } 19574358d963SJosef Bacik 19584358d963SJosef Bacik static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) 19594358d963SJosef Bacik { 19604358d963SJosef Bacik u64 extra_flags = chunk_to_extended(flags) & 19614358d963SJosef Bacik BTRFS_EXTENDED_PROFILE_MASK; 19624358d963SJosef Bacik 19634358d963SJosef Bacik write_seqlock(&fs_info->profiles_lock); 19644358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA) 19654358d963SJosef Bacik fs_info->avail_data_alloc_bits |= extra_flags; 19664358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_METADATA) 19674358d963SJosef Bacik fs_info->avail_metadata_alloc_bits |= extra_flags; 19684358d963SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 19694358d963SJosef Bacik fs_info->avail_system_alloc_bits |= extra_flags; 19704358d963SJosef Bacik write_sequnlock(&fs_info->profiles_lock); 19714358d963SJosef Bacik } 19724358d963SJosef Bacik 197343dd529aSDavid Sterba /* 197443dd529aSDavid Sterba * Map a physical disk address to a list of logical addresses. 19759ee9b979SNikolay Borisov * 19769ee9b979SNikolay Borisov * @fs_info: the filesystem 197796a14336SNikolay Borisov * @chunk_start: logical address of block group 197896a14336SNikolay Borisov * @physical: physical address to map to logical addresses 197996a14336SNikolay Borisov * @logical: return array of logical addresses which map to @physical 198096a14336SNikolay Borisov * @naddrs: length of @logical 198196a14336SNikolay Borisov * @stripe_len: size of IO stripe for the given block group 198296a14336SNikolay Borisov * 198396a14336SNikolay Borisov * Maps a particular @physical disk address to a list of @logical addresses. 198496a14336SNikolay Borisov * Used primarily to exclude those portions of a block group that contain super 198596a14336SNikolay Borisov * block copies. 198696a14336SNikolay Borisov */ 198796a14336SNikolay Borisov int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, 19881eb82ef8SChristoph Hellwig u64 physical, u64 **logical, int *naddrs, int *stripe_len) 198996a14336SNikolay Borisov { 199096a14336SNikolay Borisov struct extent_map *em; 199196a14336SNikolay Borisov struct map_lookup *map; 199296a14336SNikolay Borisov u64 *buf; 199396a14336SNikolay Borisov u64 bytenr; 19941776ad17SNikolay Borisov u64 data_stripe_length; 19951776ad17SNikolay Borisov u64 io_stripe_size; 19961776ad17SNikolay Borisov int i, nr = 0; 19971776ad17SNikolay Borisov int ret = 0; 199896a14336SNikolay Borisov 199996a14336SNikolay Borisov em = btrfs_get_chunk_map(fs_info, chunk_start, 1); 200096a14336SNikolay Borisov if (IS_ERR(em)) 200196a14336SNikolay Borisov return -EIO; 200296a14336SNikolay Borisov 200396a14336SNikolay Borisov map = em->map_lookup; 20049e22b925SNikolay Borisov data_stripe_length = em->orig_block_len; 2005a97699d1SQu Wenruo io_stripe_size = BTRFS_STRIPE_LEN; 2006138082f3SNaohiro Aota chunk_start = em->start; 200796a14336SNikolay Borisov 20089e22b925SNikolay Borisov /* For RAID5/6 adjust to a full IO stripe length */ 20099e22b925SNikolay Borisov if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 2010a97699d1SQu Wenruo io_stripe_size = nr_data_stripes(map) << BTRFS_STRIPE_LEN_SHIFT; 201196a14336SNikolay Borisov 201296a14336SNikolay Borisov buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); 20131776ad17SNikolay Borisov if (!buf) { 20141776ad17SNikolay Borisov ret = -ENOMEM; 20151776ad17SNikolay Borisov goto out; 20161776ad17SNikolay Borisov } 201796a14336SNikolay Borisov 201896a14336SNikolay Borisov for (i = 0; i < map->num_stripes; i++) { 20191776ad17SNikolay Borisov bool already_inserted = false; 20206ded22c1SQu Wenruo u32 stripe_nr; 20216ded22c1SQu Wenruo u32 offset; 20221776ad17SNikolay Borisov int j; 20231776ad17SNikolay Borisov 20241776ad17SNikolay Borisov if (!in_range(physical, map->stripes[i].physical, 20251776ad17SNikolay Borisov data_stripe_length)) 202696a14336SNikolay Borisov continue; 202796a14336SNikolay Borisov 2028a97699d1SQu Wenruo stripe_nr = (physical - map->stripes[i].physical) >> 2029a97699d1SQu Wenruo BTRFS_STRIPE_LEN_SHIFT; 2030a97699d1SQu Wenruo offset = (physical - map->stripes[i].physical) & 2031a97699d1SQu Wenruo BTRFS_STRIPE_LEN_MASK; 203296a14336SNikolay Borisov 2033ac067734SDavid Sterba if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 20346ded22c1SQu Wenruo BTRFS_BLOCK_GROUP_RAID10)) 20356ded22c1SQu Wenruo stripe_nr = div_u64(stripe_nr * map->num_stripes + i, 20366ded22c1SQu Wenruo map->sub_stripes); 203796a14336SNikolay Borisov /* 203896a14336SNikolay Borisov * The remaining case would be for RAID56, multiply by 203996a14336SNikolay Borisov * nr_data_stripes(). Alternatively, just use rmap_len below 204096a14336SNikolay Borisov * instead of map->stripe_len 204196a14336SNikolay Borisov */ 2042138082f3SNaohiro Aota bytenr = chunk_start + stripe_nr * io_stripe_size + offset; 20431776ad17SNikolay Borisov 20441776ad17SNikolay Borisov /* Ensure we don't add duplicate addresses */ 204596a14336SNikolay Borisov for (j = 0; j < nr; j++) { 20461776ad17SNikolay Borisov if (buf[j] == bytenr) { 20471776ad17SNikolay Borisov already_inserted = true; 204896a14336SNikolay Borisov break; 204996a14336SNikolay Borisov } 205096a14336SNikolay Borisov } 20511776ad17SNikolay Borisov 20521776ad17SNikolay Borisov if (!already_inserted) 20531776ad17SNikolay Borisov buf[nr++] = bytenr; 205496a14336SNikolay Borisov } 205596a14336SNikolay Borisov 205696a14336SNikolay Borisov *logical = buf; 205796a14336SNikolay Borisov *naddrs = nr; 20581776ad17SNikolay Borisov *stripe_len = io_stripe_size; 20591776ad17SNikolay Borisov out: 206096a14336SNikolay Borisov free_extent_map(em); 20611776ad17SNikolay Borisov return ret; 206296a14336SNikolay Borisov } 206396a14336SNikolay Borisov 206432da5386SDavid Sterba static int exclude_super_stripes(struct btrfs_block_group *cache) 20654358d963SJosef Bacik { 20664358d963SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 206712659251SNaohiro Aota const bool zoned = btrfs_is_zoned(fs_info); 20684358d963SJosef Bacik u64 bytenr; 20694358d963SJosef Bacik u64 *logical; 20704358d963SJosef Bacik int stripe_len; 20714358d963SJosef Bacik int i, nr, ret; 20724358d963SJosef Bacik 2073b3470b5dSDavid Sterba if (cache->start < BTRFS_SUPER_INFO_OFFSET) { 2074b3470b5dSDavid Sterba stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->start; 20754358d963SJosef Bacik cache->bytes_super += stripe_len; 2076b3470b5dSDavid Sterba ret = btrfs_add_excluded_extent(fs_info, cache->start, 20774358d963SJosef Bacik stripe_len); 20784358d963SJosef Bacik if (ret) 20794358d963SJosef Bacik return ret; 20804358d963SJosef Bacik } 20814358d963SJosef Bacik 20824358d963SJosef Bacik for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 20834358d963SJosef Bacik bytenr = btrfs_sb_offset(i); 20841eb82ef8SChristoph Hellwig ret = btrfs_rmap_block(fs_info, cache->start, 20854358d963SJosef Bacik bytenr, &logical, &nr, &stripe_len); 20864358d963SJosef Bacik if (ret) 20874358d963SJosef Bacik return ret; 20884358d963SJosef Bacik 208912659251SNaohiro Aota /* Shouldn't have super stripes in sequential zones */ 209012659251SNaohiro Aota if (zoned && nr) { 209112659251SNaohiro Aota btrfs_err(fs_info, 209212659251SNaohiro Aota "zoned: block group %llu must not contain super block", 209312659251SNaohiro Aota cache->start); 209412659251SNaohiro Aota return -EUCLEAN; 209512659251SNaohiro Aota } 209612659251SNaohiro Aota 20974358d963SJosef Bacik while (nr--) { 209896f9b0f2SNikolay Borisov u64 len = min_t(u64, stripe_len, 209996f9b0f2SNikolay Borisov cache->start + cache->length - logical[nr]); 21004358d963SJosef Bacik 21014358d963SJosef Bacik cache->bytes_super += len; 210296f9b0f2SNikolay Borisov ret = btrfs_add_excluded_extent(fs_info, logical[nr], 210396f9b0f2SNikolay Borisov len); 21044358d963SJosef Bacik if (ret) { 21054358d963SJosef Bacik kfree(logical); 21064358d963SJosef Bacik return ret; 21074358d963SJosef Bacik } 21084358d963SJosef Bacik } 21094358d963SJosef Bacik 21104358d963SJosef Bacik kfree(logical); 21114358d963SJosef Bacik } 21124358d963SJosef Bacik return 0; 21134358d963SJosef Bacik } 21144358d963SJosef Bacik 211532da5386SDavid Sterba static struct btrfs_block_group *btrfs_create_block_group_cache( 21169afc6649SQu Wenruo struct btrfs_fs_info *fs_info, u64 start) 21174358d963SJosef Bacik { 211832da5386SDavid Sterba struct btrfs_block_group *cache; 21194358d963SJosef Bacik 21204358d963SJosef Bacik cache = kzalloc(sizeof(*cache), GFP_NOFS); 21214358d963SJosef Bacik if (!cache) 21224358d963SJosef Bacik return NULL; 21234358d963SJosef Bacik 21244358d963SJosef Bacik cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), 21254358d963SJosef Bacik GFP_NOFS); 21264358d963SJosef Bacik if (!cache->free_space_ctl) { 21274358d963SJosef Bacik kfree(cache); 21284358d963SJosef Bacik return NULL; 21294358d963SJosef Bacik } 21304358d963SJosef Bacik 2131b3470b5dSDavid Sterba cache->start = start; 21324358d963SJosef Bacik 21334358d963SJosef Bacik cache->fs_info = fs_info; 21344358d963SJosef Bacik cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); 21354358d963SJosef Bacik 21366e80d4f8SDennis Zhou cache->discard_index = BTRFS_DISCARD_INDEX_UNUSED; 21376e80d4f8SDennis Zhou 213848aaeebeSJosef Bacik refcount_set(&cache->refs, 1); 21394358d963SJosef Bacik spin_lock_init(&cache->lock); 21404358d963SJosef Bacik init_rwsem(&cache->data_rwsem); 21414358d963SJosef Bacik INIT_LIST_HEAD(&cache->list); 21424358d963SJosef Bacik INIT_LIST_HEAD(&cache->cluster_list); 21434358d963SJosef Bacik INIT_LIST_HEAD(&cache->bg_list); 21444358d963SJosef Bacik INIT_LIST_HEAD(&cache->ro_list); 2145b0643e59SDennis Zhou INIT_LIST_HEAD(&cache->discard_list); 21464358d963SJosef Bacik INIT_LIST_HEAD(&cache->dirty_list); 21474358d963SJosef Bacik INIT_LIST_HEAD(&cache->io_list); 2148afba2bc0SNaohiro Aota INIT_LIST_HEAD(&cache->active_bg_list); 2149cd79909bSJosef Bacik btrfs_init_free_space_ctl(cache, cache->free_space_ctl); 21506b7304afSFilipe Manana atomic_set(&cache->frozen, 0); 21514358d963SJosef Bacik mutex_init(&cache->free_space_lock); 21524358d963SJosef Bacik 21534358d963SJosef Bacik return cache; 21544358d963SJosef Bacik } 21554358d963SJosef Bacik 21564358d963SJosef Bacik /* 21574358d963SJosef Bacik * Iterate all chunks and verify that each of them has the corresponding block 21584358d963SJosef Bacik * group 21594358d963SJosef Bacik */ 21604358d963SJosef Bacik static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) 21614358d963SJosef Bacik { 21624358d963SJosef Bacik struct extent_map_tree *map_tree = &fs_info->mapping_tree; 21634358d963SJosef Bacik struct extent_map *em; 216432da5386SDavid Sterba struct btrfs_block_group *bg; 21654358d963SJosef Bacik u64 start = 0; 21664358d963SJosef Bacik int ret = 0; 21674358d963SJosef Bacik 21684358d963SJosef Bacik while (1) { 21694358d963SJosef Bacik read_lock(&map_tree->lock); 21704358d963SJosef Bacik /* 21714358d963SJosef Bacik * lookup_extent_mapping will return the first extent map 21724358d963SJosef Bacik * intersecting the range, so setting @len to 1 is enough to 21734358d963SJosef Bacik * get the first chunk. 21744358d963SJosef Bacik */ 21754358d963SJosef Bacik em = lookup_extent_mapping(map_tree, start, 1); 21764358d963SJosef Bacik read_unlock(&map_tree->lock); 21774358d963SJosef Bacik if (!em) 21784358d963SJosef Bacik break; 21794358d963SJosef Bacik 21804358d963SJosef Bacik bg = btrfs_lookup_block_group(fs_info, em->start); 21814358d963SJosef Bacik if (!bg) { 21824358d963SJosef Bacik btrfs_err(fs_info, 21834358d963SJosef Bacik "chunk start=%llu len=%llu doesn't have corresponding block group", 21844358d963SJosef Bacik em->start, em->len); 21854358d963SJosef Bacik ret = -EUCLEAN; 21864358d963SJosef Bacik free_extent_map(em); 21874358d963SJosef Bacik break; 21884358d963SJosef Bacik } 2189b3470b5dSDavid Sterba if (bg->start != em->start || bg->length != em->len || 21904358d963SJosef Bacik (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != 21914358d963SJosef Bacik (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 21924358d963SJosef Bacik btrfs_err(fs_info, 21934358d963SJosef Bacik "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", 21944358d963SJosef Bacik em->start, em->len, 21954358d963SJosef Bacik em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, 2196b3470b5dSDavid Sterba bg->start, bg->length, 21974358d963SJosef Bacik bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); 21984358d963SJosef Bacik ret = -EUCLEAN; 21994358d963SJosef Bacik free_extent_map(em); 22004358d963SJosef Bacik btrfs_put_block_group(bg); 22014358d963SJosef Bacik break; 22024358d963SJosef Bacik } 22034358d963SJosef Bacik start = em->start + em->len; 22044358d963SJosef Bacik free_extent_map(em); 22054358d963SJosef Bacik btrfs_put_block_group(bg); 22064358d963SJosef Bacik } 22074358d963SJosef Bacik return ret; 22084358d963SJosef Bacik } 22094358d963SJosef Bacik 2210ffb9e0f0SQu Wenruo static int read_one_block_group(struct btrfs_fs_info *info, 22114afd2fe8SJohannes Thumshirn struct btrfs_block_group_item *bgi, 2212d49a2ddbSQu Wenruo const struct btrfs_key *key, 2213ffb9e0f0SQu Wenruo int need_clear) 2214ffb9e0f0SQu Wenruo { 221532da5386SDavid Sterba struct btrfs_block_group *cache; 2216ffb9e0f0SQu Wenruo const bool mixed = btrfs_fs_incompat(info, MIXED_GROUPS); 2217ffb9e0f0SQu Wenruo int ret; 2218ffb9e0f0SQu Wenruo 2219d49a2ddbSQu Wenruo ASSERT(key->type == BTRFS_BLOCK_GROUP_ITEM_KEY); 2220ffb9e0f0SQu Wenruo 22219afc6649SQu Wenruo cache = btrfs_create_block_group_cache(info, key->objectid); 2222ffb9e0f0SQu Wenruo if (!cache) 2223ffb9e0f0SQu Wenruo return -ENOMEM; 2224ffb9e0f0SQu Wenruo 22254afd2fe8SJohannes Thumshirn cache->length = key->offset; 22264afd2fe8SJohannes Thumshirn cache->used = btrfs_stack_block_group_used(bgi); 22277248e0ceSQu Wenruo cache->commit_used = cache->used; 22284afd2fe8SJohannes Thumshirn cache->flags = btrfs_stack_block_group_flags(bgi); 2229f7238e50SJosef Bacik cache->global_root_id = btrfs_stack_block_group_chunk_objectid(bgi); 22309afc6649SQu Wenruo 2231e3e39c72SMarcos Paulo de Souza set_free_space_tree_thresholds(cache); 2232e3e39c72SMarcos Paulo de Souza 2233ffb9e0f0SQu Wenruo if (need_clear) { 2234ffb9e0f0SQu Wenruo /* 2235ffb9e0f0SQu Wenruo * When we mount with old space cache, we need to 2236ffb9e0f0SQu Wenruo * set BTRFS_DC_CLEAR and set dirty flag. 2237ffb9e0f0SQu Wenruo * 2238ffb9e0f0SQu Wenruo * a) Setting 'BTRFS_DC_CLEAR' makes sure that we 2239ffb9e0f0SQu Wenruo * truncate the old free space cache inode and 2240ffb9e0f0SQu Wenruo * setup a new one. 2241ffb9e0f0SQu Wenruo * b) Setting 'dirty flag' makes sure that we flush 2242ffb9e0f0SQu Wenruo * the new space cache info onto disk. 2243ffb9e0f0SQu Wenruo */ 2244ffb9e0f0SQu Wenruo if (btrfs_test_opt(info, SPACE_CACHE)) 2245ffb9e0f0SQu Wenruo cache->disk_cache_state = BTRFS_DC_CLEAR; 2246ffb9e0f0SQu Wenruo } 2247ffb9e0f0SQu Wenruo if (!mixed && ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && 2248ffb9e0f0SQu Wenruo (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { 2249ffb9e0f0SQu Wenruo btrfs_err(info, 2250ffb9e0f0SQu Wenruo "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", 2251ffb9e0f0SQu Wenruo cache->start); 2252ffb9e0f0SQu Wenruo ret = -EINVAL; 2253ffb9e0f0SQu Wenruo goto error; 2254ffb9e0f0SQu Wenruo } 2255ffb9e0f0SQu Wenruo 2256a94794d5SNaohiro Aota ret = btrfs_load_block_group_zone_info(cache, false); 225708e11a3dSNaohiro Aota if (ret) { 225808e11a3dSNaohiro Aota btrfs_err(info, "zoned: failed to load zone info of bg %llu", 225908e11a3dSNaohiro Aota cache->start); 226008e11a3dSNaohiro Aota goto error; 226108e11a3dSNaohiro Aota } 226208e11a3dSNaohiro Aota 2263ffb9e0f0SQu Wenruo /* 2264ffb9e0f0SQu Wenruo * We need to exclude the super stripes now so that the space info has 2265ffb9e0f0SQu Wenruo * super bytes accounted for, otherwise we'll think we have more space 2266ffb9e0f0SQu Wenruo * than we actually do. 2267ffb9e0f0SQu Wenruo */ 2268ffb9e0f0SQu Wenruo ret = exclude_super_stripes(cache); 2269ffb9e0f0SQu Wenruo if (ret) { 2270ffb9e0f0SQu Wenruo /* We may have excluded something, so call this just in case. */ 2271ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2272ffb9e0f0SQu Wenruo goto error; 2273ffb9e0f0SQu Wenruo } 2274ffb9e0f0SQu Wenruo 2275ffb9e0f0SQu Wenruo /* 2276169e0da9SNaohiro Aota * For zoned filesystem, space after the allocation offset is the only 2277169e0da9SNaohiro Aota * free space for a block group. So, we don't need any caching work. 2278169e0da9SNaohiro Aota * btrfs_calc_zone_unusable() will set the amount of free space and 2279169e0da9SNaohiro Aota * zone_unusable space. 2280169e0da9SNaohiro Aota * 2281169e0da9SNaohiro Aota * For regular filesystem, check for two cases, either we are full, and 2282169e0da9SNaohiro Aota * therefore don't need to bother with the caching work since we won't 2283169e0da9SNaohiro Aota * find any space, or we are empty, and we can just add all the space 2284169e0da9SNaohiro Aota * in and be done with it. This saves us _a_lot_ of time, particularly 2285169e0da9SNaohiro Aota * in the full case. 2286ffb9e0f0SQu Wenruo */ 2287169e0da9SNaohiro Aota if (btrfs_is_zoned(info)) { 2288169e0da9SNaohiro Aota btrfs_calc_zone_unusable(cache); 2289c46c4247SNaohiro Aota /* Should not have any excluded extents. Just in case, though. */ 2290c46c4247SNaohiro Aota btrfs_free_excluded_extents(cache); 2291169e0da9SNaohiro Aota } else if (cache->length == cache->used) { 2292ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 2293ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2294ffb9e0f0SQu Wenruo } else if (cache->used == 0) { 2295ffb9e0f0SQu Wenruo cache->cached = BTRFS_CACHE_FINISHED; 22969afc6649SQu Wenruo add_new_free_space(cache, cache->start, 22979afc6649SQu Wenruo cache->start + cache->length); 2298ffb9e0f0SQu Wenruo btrfs_free_excluded_extents(cache); 2299ffb9e0f0SQu Wenruo } 2300ffb9e0f0SQu Wenruo 2301ffb9e0f0SQu Wenruo ret = btrfs_add_block_group_cache(info, cache); 2302ffb9e0f0SQu Wenruo if (ret) { 2303ffb9e0f0SQu Wenruo btrfs_remove_free_space_cache(cache); 2304ffb9e0f0SQu Wenruo goto error; 2305ffb9e0f0SQu Wenruo } 2306ffb9e0f0SQu Wenruo trace_btrfs_add_block_group(info, cache, 0); 2307723de71dSJosef Bacik btrfs_add_bg_to_space_info(info, cache); 2308ffb9e0f0SQu Wenruo 2309ffb9e0f0SQu Wenruo set_avail_alloc_bits(info, cache->flags); 2310a09f23c3SAnand Jain if (btrfs_chunk_writeable(info, cache->start)) { 2311a09f23c3SAnand Jain if (cache->used == 0) { 2312ffb9e0f0SQu Wenruo ASSERT(list_empty(&cache->bg_list)); 23136e80d4f8SDennis Zhou if (btrfs_test_opt(info, DISCARD_ASYNC)) 23146e80d4f8SDennis Zhou btrfs_discard_queue_work(&info->discard_ctl, cache); 23156e80d4f8SDennis Zhou else 2316ffb9e0f0SQu Wenruo btrfs_mark_bg_unused(cache); 2317ffb9e0f0SQu Wenruo } 2318a09f23c3SAnand Jain } else { 2319a09f23c3SAnand Jain inc_block_group_ro(cache, 1); 2320a09f23c3SAnand Jain } 2321a09f23c3SAnand Jain 2322ffb9e0f0SQu Wenruo return 0; 2323ffb9e0f0SQu Wenruo error: 2324ffb9e0f0SQu Wenruo btrfs_put_block_group(cache); 2325ffb9e0f0SQu Wenruo return ret; 2326ffb9e0f0SQu Wenruo } 2327ffb9e0f0SQu Wenruo 232842437a63SJosef Bacik static int fill_dummy_bgs(struct btrfs_fs_info *fs_info) 232942437a63SJosef Bacik { 233042437a63SJosef Bacik struct extent_map_tree *em_tree = &fs_info->mapping_tree; 233142437a63SJosef Bacik struct rb_node *node; 233242437a63SJosef Bacik int ret = 0; 233342437a63SJosef Bacik 233442437a63SJosef Bacik for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 233542437a63SJosef Bacik struct extent_map *em; 233642437a63SJosef Bacik struct map_lookup *map; 233742437a63SJosef Bacik struct btrfs_block_group *bg; 233842437a63SJosef Bacik 233942437a63SJosef Bacik em = rb_entry(node, struct extent_map, rb_node); 234042437a63SJosef Bacik map = em->map_lookup; 234142437a63SJosef Bacik bg = btrfs_create_block_group_cache(fs_info, em->start); 234242437a63SJosef Bacik if (!bg) { 234342437a63SJosef Bacik ret = -ENOMEM; 234442437a63SJosef Bacik break; 234542437a63SJosef Bacik } 234642437a63SJosef Bacik 234742437a63SJosef Bacik /* Fill dummy cache as FULL */ 234842437a63SJosef Bacik bg->length = em->len; 234942437a63SJosef Bacik bg->flags = map->type; 235042437a63SJosef Bacik bg->cached = BTRFS_CACHE_FINISHED; 235142437a63SJosef Bacik bg->used = em->len; 235242437a63SJosef Bacik bg->flags = map->type; 235342437a63SJosef Bacik ret = btrfs_add_block_group_cache(fs_info, bg); 23542b29726cSQu Wenruo /* 23552b29726cSQu Wenruo * We may have some valid block group cache added already, in 23562b29726cSQu Wenruo * that case we skip to the next one. 23572b29726cSQu Wenruo */ 23582b29726cSQu Wenruo if (ret == -EEXIST) { 23592b29726cSQu Wenruo ret = 0; 23602b29726cSQu Wenruo btrfs_put_block_group(bg); 23612b29726cSQu Wenruo continue; 23622b29726cSQu Wenruo } 23632b29726cSQu Wenruo 236442437a63SJosef Bacik if (ret) { 236542437a63SJosef Bacik btrfs_remove_free_space_cache(bg); 236642437a63SJosef Bacik btrfs_put_block_group(bg); 236742437a63SJosef Bacik break; 236842437a63SJosef Bacik } 23692b29726cSQu Wenruo 2370723de71dSJosef Bacik btrfs_add_bg_to_space_info(fs_info, bg); 237142437a63SJosef Bacik 237242437a63SJosef Bacik set_avail_alloc_bits(fs_info, bg->flags); 237342437a63SJosef Bacik } 237442437a63SJosef Bacik if (!ret) 237542437a63SJosef Bacik btrfs_init_global_block_rsv(fs_info); 237642437a63SJosef Bacik return ret; 237742437a63SJosef Bacik } 237842437a63SJosef Bacik 23794358d963SJosef Bacik int btrfs_read_block_groups(struct btrfs_fs_info *info) 23804358d963SJosef Bacik { 2381dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(info); 23824358d963SJosef Bacik struct btrfs_path *path; 23834358d963SJosef Bacik int ret; 238432da5386SDavid Sterba struct btrfs_block_group *cache; 23854358d963SJosef Bacik struct btrfs_space_info *space_info; 23864358d963SJosef Bacik struct btrfs_key key; 23874358d963SJosef Bacik int need_clear = 0; 23884358d963SJosef Bacik u64 cache_gen; 23894358d963SJosef Bacik 239081d5d614SQu Wenruo /* 239181d5d614SQu Wenruo * Either no extent root (with ibadroots rescue option) or we have 239281d5d614SQu Wenruo * unsupported RO options. The fs can never be mounted read-write, so no 239381d5d614SQu Wenruo * need to waste time searching block group items. 239481d5d614SQu Wenruo * 239581d5d614SQu Wenruo * This also allows new extent tree related changes to be RO compat, 239681d5d614SQu Wenruo * no need for a full incompat flag. 239781d5d614SQu Wenruo */ 239881d5d614SQu Wenruo if (!root || (btrfs_super_compat_ro_flags(info->super_copy) & 239981d5d614SQu Wenruo ~BTRFS_FEATURE_COMPAT_RO_SUPP)) 240042437a63SJosef Bacik return fill_dummy_bgs(info); 240142437a63SJosef Bacik 24024358d963SJosef Bacik key.objectid = 0; 24034358d963SJosef Bacik key.offset = 0; 24044358d963SJosef Bacik key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 24054358d963SJosef Bacik path = btrfs_alloc_path(); 24064358d963SJosef Bacik if (!path) 24074358d963SJosef Bacik return -ENOMEM; 24084358d963SJosef Bacik 24094358d963SJosef Bacik cache_gen = btrfs_super_cache_generation(info->super_copy); 24104358d963SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 24114358d963SJosef Bacik btrfs_super_generation(info->super_copy) != cache_gen) 24124358d963SJosef Bacik need_clear = 1; 24134358d963SJosef Bacik if (btrfs_test_opt(info, CLEAR_CACHE)) 24144358d963SJosef Bacik need_clear = 1; 24154358d963SJosef Bacik 24164358d963SJosef Bacik while (1) { 24174afd2fe8SJohannes Thumshirn struct btrfs_block_group_item bgi; 24184afd2fe8SJohannes Thumshirn struct extent_buffer *leaf; 24194afd2fe8SJohannes Thumshirn int slot; 24204afd2fe8SJohannes Thumshirn 24214358d963SJosef Bacik ret = find_first_block_group(info, path, &key); 24224358d963SJosef Bacik if (ret > 0) 24234358d963SJosef Bacik break; 24244358d963SJosef Bacik if (ret != 0) 24254358d963SJosef Bacik goto error; 24264358d963SJosef Bacik 24274afd2fe8SJohannes Thumshirn leaf = path->nodes[0]; 24284afd2fe8SJohannes Thumshirn slot = path->slots[0]; 24294afd2fe8SJohannes Thumshirn 24304afd2fe8SJohannes Thumshirn read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), 24314afd2fe8SJohannes Thumshirn sizeof(bgi)); 24324afd2fe8SJohannes Thumshirn 24334afd2fe8SJohannes Thumshirn btrfs_item_key_to_cpu(leaf, &key, slot); 24344afd2fe8SJohannes Thumshirn btrfs_release_path(path); 24354afd2fe8SJohannes Thumshirn ret = read_one_block_group(info, &bgi, &key, need_clear); 2436ffb9e0f0SQu Wenruo if (ret < 0) 24374358d963SJosef Bacik goto error; 2438ffb9e0f0SQu Wenruo key.objectid += key.offset; 2439ffb9e0f0SQu Wenruo key.offset = 0; 24404358d963SJosef Bacik } 24417837fa88SJosef Bacik btrfs_release_path(path); 24424358d963SJosef Bacik 244372804905SJosef Bacik list_for_each_entry(space_info, &info->space_info, list) { 244449ea112dSJosef Bacik int i; 244549ea112dSJosef Bacik 244649ea112dSJosef Bacik for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 244749ea112dSJosef Bacik if (list_empty(&space_info->block_groups[i])) 244849ea112dSJosef Bacik continue; 244949ea112dSJosef Bacik cache = list_first_entry(&space_info->block_groups[i], 245049ea112dSJosef Bacik struct btrfs_block_group, 245149ea112dSJosef Bacik list); 245249ea112dSJosef Bacik btrfs_sysfs_add_block_group_type(cache); 245349ea112dSJosef Bacik } 245449ea112dSJosef Bacik 24554358d963SJosef Bacik if (!(btrfs_get_alloc_profile(info, space_info->flags) & 24564358d963SJosef Bacik (BTRFS_BLOCK_GROUP_RAID10 | 24574358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID1_MASK | 24584358d963SJosef Bacik BTRFS_BLOCK_GROUP_RAID56_MASK | 24594358d963SJosef Bacik BTRFS_BLOCK_GROUP_DUP))) 24604358d963SJosef Bacik continue; 24614358d963SJosef Bacik /* 24624358d963SJosef Bacik * Avoid allocating from un-mirrored block group if there are 24634358d963SJosef Bacik * mirrored block groups. 24644358d963SJosef Bacik */ 24654358d963SJosef Bacik list_for_each_entry(cache, 24664358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_RAID0], 24674358d963SJosef Bacik list) 2468e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 24694358d963SJosef Bacik list_for_each_entry(cache, 24704358d963SJosef Bacik &space_info->block_groups[BTRFS_RAID_SINGLE], 24714358d963SJosef Bacik list) 2472e11c0406SJosef Bacik inc_block_group_ro(cache, 1); 24734358d963SJosef Bacik } 24744358d963SJosef Bacik 24754358d963SJosef Bacik btrfs_init_global_block_rsv(info); 24764358d963SJosef Bacik ret = check_chunk_block_group_mappings(info); 24774358d963SJosef Bacik error: 24784358d963SJosef Bacik btrfs_free_path(path); 24792b29726cSQu Wenruo /* 24802b29726cSQu Wenruo * We've hit some error while reading the extent tree, and have 24812b29726cSQu Wenruo * rescue=ibadroots mount option. 24822b29726cSQu Wenruo * Try to fill the tree using dummy block groups so that the user can 24832b29726cSQu Wenruo * continue to mount and grab their data. 24842b29726cSQu Wenruo */ 24852b29726cSQu Wenruo if (ret && btrfs_test_opt(info, IGNOREBADROOTS)) 24862b29726cSQu Wenruo ret = fill_dummy_bgs(info); 24874358d963SJosef Bacik return ret; 24884358d963SJosef Bacik } 24894358d963SJosef Bacik 249079bd3712SFilipe Manana /* 249179bd3712SFilipe Manana * This function, insert_block_group_item(), belongs to the phase 2 of chunk 249279bd3712SFilipe Manana * allocation. 249379bd3712SFilipe Manana * 249479bd3712SFilipe Manana * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 249579bd3712SFilipe Manana * phases. 249679bd3712SFilipe Manana */ 249797f4728aSQu Wenruo static int insert_block_group_item(struct btrfs_trans_handle *trans, 249897f4728aSQu Wenruo struct btrfs_block_group *block_group) 249997f4728aSQu Wenruo { 250097f4728aSQu Wenruo struct btrfs_fs_info *fs_info = trans->fs_info; 250197f4728aSQu Wenruo struct btrfs_block_group_item bgi; 2502dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 250397f4728aSQu Wenruo struct btrfs_key key; 2504675dfe12SFilipe Manana u64 old_commit_used; 2505675dfe12SFilipe Manana int ret; 250697f4728aSQu Wenruo 250797f4728aSQu Wenruo spin_lock(&block_group->lock); 250897f4728aSQu Wenruo btrfs_set_stack_block_group_used(&bgi, block_group->used); 250997f4728aSQu Wenruo btrfs_set_stack_block_group_chunk_objectid(&bgi, 2510f7238e50SJosef Bacik block_group->global_root_id); 251197f4728aSQu Wenruo btrfs_set_stack_block_group_flags(&bgi, block_group->flags); 2512675dfe12SFilipe Manana old_commit_used = block_group->commit_used; 2513675dfe12SFilipe Manana block_group->commit_used = block_group->used; 251497f4728aSQu Wenruo key.objectid = block_group->start; 251597f4728aSQu Wenruo key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 251697f4728aSQu Wenruo key.offset = block_group->length; 251797f4728aSQu Wenruo spin_unlock(&block_group->lock); 251897f4728aSQu Wenruo 2519675dfe12SFilipe Manana ret = btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi)); 2520675dfe12SFilipe Manana if (ret < 0) { 2521675dfe12SFilipe Manana spin_lock(&block_group->lock); 2522675dfe12SFilipe Manana block_group->commit_used = old_commit_used; 2523675dfe12SFilipe Manana spin_unlock(&block_group->lock); 2524675dfe12SFilipe Manana } 2525675dfe12SFilipe Manana 2526675dfe12SFilipe Manana return ret; 252797f4728aSQu Wenruo } 252897f4728aSQu Wenruo 25292eadb9e7SNikolay Borisov static int insert_dev_extent(struct btrfs_trans_handle *trans, 25302eadb9e7SNikolay Borisov struct btrfs_device *device, u64 chunk_offset, 25312eadb9e7SNikolay Borisov u64 start, u64 num_bytes) 25322eadb9e7SNikolay Borisov { 25332eadb9e7SNikolay Borisov struct btrfs_fs_info *fs_info = device->fs_info; 25342eadb9e7SNikolay Borisov struct btrfs_root *root = fs_info->dev_root; 25352eadb9e7SNikolay Borisov struct btrfs_path *path; 25362eadb9e7SNikolay Borisov struct btrfs_dev_extent *extent; 25372eadb9e7SNikolay Borisov struct extent_buffer *leaf; 25382eadb9e7SNikolay Borisov struct btrfs_key key; 25392eadb9e7SNikolay Borisov int ret; 25402eadb9e7SNikolay Borisov 25412eadb9e7SNikolay Borisov WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); 25422eadb9e7SNikolay Borisov WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 25432eadb9e7SNikolay Borisov path = btrfs_alloc_path(); 25442eadb9e7SNikolay Borisov if (!path) 25452eadb9e7SNikolay Borisov return -ENOMEM; 25462eadb9e7SNikolay Borisov 25472eadb9e7SNikolay Borisov key.objectid = device->devid; 25482eadb9e7SNikolay Borisov key.type = BTRFS_DEV_EXTENT_KEY; 25492eadb9e7SNikolay Borisov key.offset = start; 25502eadb9e7SNikolay Borisov ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*extent)); 25512eadb9e7SNikolay Borisov if (ret) 25522eadb9e7SNikolay Borisov goto out; 25532eadb9e7SNikolay Borisov 25542eadb9e7SNikolay Borisov leaf = path->nodes[0]; 25552eadb9e7SNikolay Borisov extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); 25562eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_tree(leaf, extent, BTRFS_CHUNK_TREE_OBJECTID); 25572eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_objectid(leaf, extent, 25582eadb9e7SNikolay Borisov BTRFS_FIRST_CHUNK_TREE_OBJECTID); 25592eadb9e7SNikolay Borisov btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); 25602eadb9e7SNikolay Borisov 25612eadb9e7SNikolay Borisov btrfs_set_dev_extent_length(leaf, extent, num_bytes); 25622eadb9e7SNikolay Borisov btrfs_mark_buffer_dirty(leaf); 25632eadb9e7SNikolay Borisov out: 25642eadb9e7SNikolay Borisov btrfs_free_path(path); 25652eadb9e7SNikolay Borisov return ret; 25662eadb9e7SNikolay Borisov } 25672eadb9e7SNikolay Borisov 25682eadb9e7SNikolay Borisov /* 25692eadb9e7SNikolay Borisov * This function belongs to phase 2. 25702eadb9e7SNikolay Borisov * 25712eadb9e7SNikolay Borisov * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 25722eadb9e7SNikolay Borisov * phases. 25732eadb9e7SNikolay Borisov */ 25742eadb9e7SNikolay Borisov static int insert_dev_extents(struct btrfs_trans_handle *trans, 25752eadb9e7SNikolay Borisov u64 chunk_offset, u64 chunk_size) 25762eadb9e7SNikolay Borisov { 25772eadb9e7SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 25782eadb9e7SNikolay Borisov struct btrfs_device *device; 25792eadb9e7SNikolay Borisov struct extent_map *em; 25802eadb9e7SNikolay Borisov struct map_lookup *map; 25812eadb9e7SNikolay Borisov u64 dev_offset; 25822eadb9e7SNikolay Borisov u64 stripe_size; 25832eadb9e7SNikolay Borisov int i; 25842eadb9e7SNikolay Borisov int ret = 0; 25852eadb9e7SNikolay Borisov 25862eadb9e7SNikolay Borisov em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); 25872eadb9e7SNikolay Borisov if (IS_ERR(em)) 25882eadb9e7SNikolay Borisov return PTR_ERR(em); 25892eadb9e7SNikolay Borisov 25902eadb9e7SNikolay Borisov map = em->map_lookup; 25912eadb9e7SNikolay Borisov stripe_size = em->orig_block_len; 25922eadb9e7SNikolay Borisov 25932eadb9e7SNikolay Borisov /* 25942eadb9e7SNikolay Borisov * Take the device list mutex to prevent races with the final phase of 25952eadb9e7SNikolay Borisov * a device replace operation that replaces the device object associated 25962eadb9e7SNikolay Borisov * with the map's stripes, because the device object's id can change 25972eadb9e7SNikolay Borisov * at any time during that final phase of the device replace operation 25982eadb9e7SNikolay Borisov * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 25992eadb9e7SNikolay Borisov * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 26002eadb9e7SNikolay Borisov * resulting in persisting a device extent item with such ID. 26012eadb9e7SNikolay Borisov */ 26022eadb9e7SNikolay Borisov mutex_lock(&fs_info->fs_devices->device_list_mutex); 26032eadb9e7SNikolay Borisov for (i = 0; i < map->num_stripes; i++) { 26042eadb9e7SNikolay Borisov device = map->stripes[i].dev; 26052eadb9e7SNikolay Borisov dev_offset = map->stripes[i].physical; 26062eadb9e7SNikolay Borisov 26072eadb9e7SNikolay Borisov ret = insert_dev_extent(trans, device, chunk_offset, dev_offset, 26082eadb9e7SNikolay Borisov stripe_size); 26092eadb9e7SNikolay Borisov if (ret) 26102eadb9e7SNikolay Borisov break; 26112eadb9e7SNikolay Borisov } 26122eadb9e7SNikolay Borisov mutex_unlock(&fs_info->fs_devices->device_list_mutex); 26132eadb9e7SNikolay Borisov 26142eadb9e7SNikolay Borisov free_extent_map(em); 26152eadb9e7SNikolay Borisov return ret; 26162eadb9e7SNikolay Borisov } 26172eadb9e7SNikolay Borisov 261879bd3712SFilipe Manana /* 261979bd3712SFilipe Manana * This function, btrfs_create_pending_block_groups(), belongs to the phase 2 of 262079bd3712SFilipe Manana * chunk allocation. 262179bd3712SFilipe Manana * 262279bd3712SFilipe Manana * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 262379bd3712SFilipe Manana * phases. 262479bd3712SFilipe Manana */ 26254358d963SJosef Bacik void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) 26264358d963SJosef Bacik { 26274358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 262832da5386SDavid Sterba struct btrfs_block_group *block_group; 26294358d963SJosef Bacik int ret = 0; 26304358d963SJosef Bacik 26314358d963SJosef Bacik while (!list_empty(&trans->new_bgs)) { 263249ea112dSJosef Bacik int index; 263349ea112dSJosef Bacik 26344358d963SJosef Bacik block_group = list_first_entry(&trans->new_bgs, 263532da5386SDavid Sterba struct btrfs_block_group, 26364358d963SJosef Bacik bg_list); 26374358d963SJosef Bacik if (ret) 26384358d963SJosef Bacik goto next; 26394358d963SJosef Bacik 264049ea112dSJosef Bacik index = btrfs_bg_flags_to_raid_index(block_group->flags); 264149ea112dSJosef Bacik 264297f4728aSQu Wenruo ret = insert_block_group_item(trans, block_group); 26434358d963SJosef Bacik if (ret) 26444358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 26453349b57fSJosef Bacik if (!test_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, 26463349b57fSJosef Bacik &block_group->runtime_flags)) { 264779bd3712SFilipe Manana mutex_lock(&fs_info->chunk_mutex); 264879bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, block_group); 264979bd3712SFilipe Manana mutex_unlock(&fs_info->chunk_mutex); 265079bd3712SFilipe Manana if (ret) 265179bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 265279bd3712SFilipe Manana } 26532eadb9e7SNikolay Borisov ret = insert_dev_extents(trans, block_group->start, 265497f4728aSQu Wenruo block_group->length); 26554358d963SJosef Bacik if (ret) 26564358d963SJosef Bacik btrfs_abort_transaction(trans, ret); 26574358d963SJosef Bacik add_block_group_free_space(trans, block_group); 265849ea112dSJosef Bacik 265949ea112dSJosef Bacik /* 266049ea112dSJosef Bacik * If we restriped during balance, we may have added a new raid 266149ea112dSJosef Bacik * type, so now add the sysfs entries when it is safe to do so. 266249ea112dSJosef Bacik * We don't have to worry about locking here as it's handled in 266349ea112dSJosef Bacik * btrfs_sysfs_add_block_group_type. 266449ea112dSJosef Bacik */ 266549ea112dSJosef Bacik if (block_group->space_info->block_group_kobjs[index] == NULL) 266649ea112dSJosef Bacik btrfs_sysfs_add_block_group_type(block_group); 266749ea112dSJosef Bacik 26684358d963SJosef Bacik /* Already aborted the transaction if it failed. */ 26694358d963SJosef Bacik next: 26704358d963SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 26714358d963SJosef Bacik list_del_init(&block_group->bg_list); 2672*0657b20cSFilipe Manana clear_bit(BLOCK_GROUP_FLAG_NEW, &block_group->runtime_flags); 26734358d963SJosef Bacik } 26744358d963SJosef Bacik btrfs_trans_release_chunk_metadata(trans); 26754358d963SJosef Bacik } 26764358d963SJosef Bacik 2677f7238e50SJosef Bacik /* 2678f7238e50SJosef Bacik * For extent tree v2 we use the block_group_item->chunk_offset to point at our 2679f7238e50SJosef Bacik * global root id. For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID. 2680f7238e50SJosef Bacik */ 2681f7238e50SJosef Bacik static u64 calculate_global_root_id(struct btrfs_fs_info *fs_info, u64 offset) 2682f7238e50SJosef Bacik { 2683f7238e50SJosef Bacik u64 div = SZ_1G; 2684f7238e50SJosef Bacik u64 index; 2685f7238e50SJosef Bacik 2686f7238e50SJosef Bacik if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 2687f7238e50SJosef Bacik return BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2688f7238e50SJosef Bacik 2689f7238e50SJosef Bacik /* If we have a smaller fs index based on 128MiB. */ 2690f7238e50SJosef Bacik if (btrfs_super_total_bytes(fs_info->super_copy) <= (SZ_1G * 10ULL)) 2691f7238e50SJosef Bacik div = SZ_128M; 2692f7238e50SJosef Bacik 2693f7238e50SJosef Bacik offset = div64_u64(offset, div); 2694f7238e50SJosef Bacik div64_u64_rem(offset, fs_info->nr_global_roots, &index); 2695f7238e50SJosef Bacik return index; 2696f7238e50SJosef Bacik } 2697f7238e50SJosef Bacik 269879bd3712SFilipe Manana struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *trans, 26995758d1bdSFilipe Manana u64 type, 270079bd3712SFilipe Manana u64 chunk_offset, u64 size) 27014358d963SJosef Bacik { 27024358d963SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 270332da5386SDavid Sterba struct btrfs_block_group *cache; 27044358d963SJosef Bacik int ret; 27054358d963SJosef Bacik 27064358d963SJosef Bacik btrfs_set_log_full_commit(trans); 27074358d963SJosef Bacik 27089afc6649SQu Wenruo cache = btrfs_create_block_group_cache(fs_info, chunk_offset); 27094358d963SJosef Bacik if (!cache) 271079bd3712SFilipe Manana return ERR_PTR(-ENOMEM); 27114358d963SJosef Bacik 2712*0657b20cSFilipe Manana /* 2713*0657b20cSFilipe Manana * Mark it as new before adding it to the rbtree of block groups or any 2714*0657b20cSFilipe Manana * list, so that no other task finds it and calls btrfs_mark_bg_unused() 2715*0657b20cSFilipe Manana * before the new flag is set. 2716*0657b20cSFilipe Manana */ 2717*0657b20cSFilipe Manana set_bit(BLOCK_GROUP_FLAG_NEW, &cache->runtime_flags); 2718*0657b20cSFilipe Manana 27199afc6649SQu Wenruo cache->length = size; 2720e3e39c72SMarcos Paulo de Souza set_free_space_tree_thresholds(cache); 27214358d963SJosef Bacik cache->flags = type; 27224358d963SJosef Bacik cache->cached = BTRFS_CACHE_FINISHED; 2723f7238e50SJosef Bacik cache->global_root_id = calculate_global_root_id(fs_info, cache->start); 2724f7238e50SJosef Bacik 2725997e3e2eSBoris Burkov if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) 27260d7764ffSDavid Sterba set_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &cache->runtime_flags); 272708e11a3dSNaohiro Aota 2728a94794d5SNaohiro Aota ret = btrfs_load_block_group_zone_info(cache, true); 272908e11a3dSNaohiro Aota if (ret) { 273008e11a3dSNaohiro Aota btrfs_put_block_group(cache); 273179bd3712SFilipe Manana return ERR_PTR(ret); 273208e11a3dSNaohiro Aota } 273308e11a3dSNaohiro Aota 27344358d963SJosef Bacik ret = exclude_super_stripes(cache); 27354358d963SJosef Bacik if (ret) { 27364358d963SJosef Bacik /* We may have excluded something, so call this just in case */ 27374358d963SJosef Bacik btrfs_free_excluded_extents(cache); 27384358d963SJosef Bacik btrfs_put_block_group(cache); 273979bd3712SFilipe Manana return ERR_PTR(ret); 27404358d963SJosef Bacik } 27414358d963SJosef Bacik 27424358d963SJosef Bacik add_new_free_space(cache, chunk_offset, chunk_offset + size); 27434358d963SJosef Bacik 27444358d963SJosef Bacik btrfs_free_excluded_extents(cache); 27454358d963SJosef Bacik 27464358d963SJosef Bacik /* 27474358d963SJosef Bacik * Ensure the corresponding space_info object is created and 27484358d963SJosef Bacik * assigned to our block group. We want our bg to be added to the rbtree 27494358d963SJosef Bacik * with its ->space_info set. 27504358d963SJosef Bacik */ 27514358d963SJosef Bacik cache->space_info = btrfs_find_space_info(fs_info, cache->flags); 27524358d963SJosef Bacik ASSERT(cache->space_info); 27534358d963SJosef Bacik 27544358d963SJosef Bacik ret = btrfs_add_block_group_cache(fs_info, cache); 27554358d963SJosef Bacik if (ret) { 27564358d963SJosef Bacik btrfs_remove_free_space_cache(cache); 27574358d963SJosef Bacik btrfs_put_block_group(cache); 275879bd3712SFilipe Manana return ERR_PTR(ret); 27594358d963SJosef Bacik } 27604358d963SJosef Bacik 27614358d963SJosef Bacik /* 27624358d963SJosef Bacik * Now that our block group has its ->space_info set and is inserted in 27634358d963SJosef Bacik * the rbtree, update the space info's counters. 27644358d963SJosef Bacik */ 27654358d963SJosef Bacik trace_btrfs_add_block_group(fs_info, cache, 1); 2766723de71dSJosef Bacik btrfs_add_bg_to_space_info(fs_info, cache); 27674358d963SJosef Bacik btrfs_update_global_block_rsv(fs_info); 27684358d963SJosef Bacik 27699d4b0a12SJosef Bacik #ifdef CONFIG_BTRFS_DEBUG 27709d4b0a12SJosef Bacik if (btrfs_should_fragment_free_space(cache)) { 27715758d1bdSFilipe Manana cache->space_info->bytes_used += size >> 1; 27729d4b0a12SJosef Bacik fragment_free_space(cache); 27739d4b0a12SJosef Bacik } 27749d4b0a12SJosef Bacik #endif 27754358d963SJosef Bacik 27764358d963SJosef Bacik list_add_tail(&cache->bg_list, &trans->new_bgs); 27774358d963SJosef Bacik trans->delayed_ref_updates++; 27784358d963SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 27794358d963SJosef Bacik 27804358d963SJosef Bacik set_avail_alloc_bits(fs_info, type); 278179bd3712SFilipe Manana return cache; 27824358d963SJosef Bacik } 278326ce2095SJosef Bacik 2784b12de528SQu Wenruo /* 2785b12de528SQu Wenruo * Mark one block group RO, can be called several times for the same block 2786b12de528SQu Wenruo * group. 2787b12de528SQu Wenruo * 2788b12de528SQu Wenruo * @cache: the destination block group 2789b12de528SQu Wenruo * @do_chunk_alloc: whether need to do chunk pre-allocation, this is to 2790b12de528SQu Wenruo * ensure we still have some free space after marking this 2791b12de528SQu Wenruo * block group RO. 2792b12de528SQu Wenruo */ 2793b12de528SQu Wenruo int btrfs_inc_block_group_ro(struct btrfs_block_group *cache, 2794b12de528SQu Wenruo bool do_chunk_alloc) 279526ce2095SJosef Bacik { 279626ce2095SJosef Bacik struct btrfs_fs_info *fs_info = cache->fs_info; 279726ce2095SJosef Bacik struct btrfs_trans_handle *trans; 2798dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 279926ce2095SJosef Bacik u64 alloc_flags; 280026ce2095SJosef Bacik int ret; 2801b6e9f16cSNikolay Borisov bool dirty_bg_running; 280226ce2095SJosef Bacik 28032d192fc4SQu Wenruo /* 28042d192fc4SQu Wenruo * This can only happen when we are doing read-only scrub on read-only 28052d192fc4SQu Wenruo * mount. 28062d192fc4SQu Wenruo * In that case we should not start a new transaction on read-only fs. 28072d192fc4SQu Wenruo * Thus here we skip all chunk allocations. 28082d192fc4SQu Wenruo */ 28092d192fc4SQu Wenruo if (sb_rdonly(fs_info->sb)) { 28102d192fc4SQu Wenruo mutex_lock(&fs_info->ro_block_group_mutex); 28112d192fc4SQu Wenruo ret = inc_block_group_ro(cache, 0); 28122d192fc4SQu Wenruo mutex_unlock(&fs_info->ro_block_group_mutex); 28132d192fc4SQu Wenruo return ret; 28142d192fc4SQu Wenruo } 28152d192fc4SQu Wenruo 2816b6e9f16cSNikolay Borisov do { 2817dfe8aec4SJosef Bacik trans = btrfs_join_transaction(root); 281826ce2095SJosef Bacik if (IS_ERR(trans)) 281926ce2095SJosef Bacik return PTR_ERR(trans); 282026ce2095SJosef Bacik 2821b6e9f16cSNikolay Borisov dirty_bg_running = false; 2822b6e9f16cSNikolay Borisov 282326ce2095SJosef Bacik /* 2824b6e9f16cSNikolay Borisov * We're not allowed to set block groups readonly after the dirty 2825b6e9f16cSNikolay Borisov * block group cache has started writing. If it already started, 2826b6e9f16cSNikolay Borisov * back off and let this transaction commit. 282726ce2095SJosef Bacik */ 282826ce2095SJosef Bacik mutex_lock(&fs_info->ro_block_group_mutex); 282926ce2095SJosef Bacik if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { 283026ce2095SJosef Bacik u64 transid = trans->transid; 283126ce2095SJosef Bacik 283226ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 283326ce2095SJosef Bacik btrfs_end_transaction(trans); 283426ce2095SJosef Bacik 283526ce2095SJosef Bacik ret = btrfs_wait_for_commit(fs_info, transid); 283626ce2095SJosef Bacik if (ret) 283726ce2095SJosef Bacik return ret; 2838b6e9f16cSNikolay Borisov dirty_bg_running = true; 283926ce2095SJosef Bacik } 2840b6e9f16cSNikolay Borisov } while (dirty_bg_running); 284126ce2095SJosef Bacik 2842b12de528SQu Wenruo if (do_chunk_alloc) { 284326ce2095SJosef Bacik /* 2844b12de528SQu Wenruo * If we are changing raid levels, try to allocate a 2845b12de528SQu Wenruo * corresponding block group with the new raid level. 284626ce2095SJosef Bacik */ 2847349e120eSJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); 284826ce2095SJosef Bacik if (alloc_flags != cache->flags) { 2849b12de528SQu Wenruo ret = btrfs_chunk_alloc(trans, alloc_flags, 2850b12de528SQu Wenruo CHUNK_ALLOC_FORCE); 285126ce2095SJosef Bacik /* 285226ce2095SJosef Bacik * ENOSPC is allowed here, we may have enough space 2853b12de528SQu Wenruo * already allocated at the new raid level to carry on 285426ce2095SJosef Bacik */ 285526ce2095SJosef Bacik if (ret == -ENOSPC) 285626ce2095SJosef Bacik ret = 0; 285726ce2095SJosef Bacik if (ret < 0) 285826ce2095SJosef Bacik goto out; 285926ce2095SJosef Bacik } 2860b12de528SQu Wenruo } 286126ce2095SJosef Bacik 2862a7a63accSJosef Bacik ret = inc_block_group_ro(cache, 0); 286326ce2095SJosef Bacik if (!ret) 286426ce2095SJosef Bacik goto out; 28657561551eSQu Wenruo if (ret == -ETXTBSY) 28667561551eSQu Wenruo goto unlock_out; 28677561551eSQu Wenruo 28687561551eSQu Wenruo /* 28697561551eSQu Wenruo * Skip chunk alloction if the bg is SYSTEM, this is to avoid system 28707561551eSQu Wenruo * chunk allocation storm to exhaust the system chunk array. Otherwise 28717561551eSQu Wenruo * we still want to try our best to mark the block group read-only. 28727561551eSQu Wenruo */ 28737561551eSQu Wenruo if (!do_chunk_alloc && ret == -ENOSPC && 28747561551eSQu Wenruo (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM)) 28757561551eSQu Wenruo goto unlock_out; 28767561551eSQu Wenruo 287726ce2095SJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->space_info->flags); 287826ce2095SJosef Bacik ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 287926ce2095SJosef Bacik if (ret < 0) 288026ce2095SJosef Bacik goto out; 2881b6a98021SNaohiro Aota /* 2882b6a98021SNaohiro Aota * We have allocated a new chunk. We also need to activate that chunk to 2883b6a98021SNaohiro Aota * grant metadata tickets for zoned filesystem. 2884b6a98021SNaohiro Aota */ 2885b6a98021SNaohiro Aota ret = btrfs_zoned_activate_one_bg(fs_info, cache->space_info, true); 2886b6a98021SNaohiro Aota if (ret < 0) 2887b6a98021SNaohiro Aota goto out; 2888b6a98021SNaohiro Aota 2889e11c0406SJosef Bacik ret = inc_block_group_ro(cache, 0); 2890195a49eaSFilipe Manana if (ret == -ETXTBSY) 2891195a49eaSFilipe Manana goto unlock_out; 289226ce2095SJosef Bacik out: 289326ce2095SJosef Bacik if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { 2894349e120eSJosef Bacik alloc_flags = btrfs_get_alloc_profile(fs_info, cache->flags); 289526ce2095SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 289626ce2095SJosef Bacik check_system_chunk(trans, alloc_flags); 289726ce2095SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 289826ce2095SJosef Bacik } 2899b12de528SQu Wenruo unlock_out: 290026ce2095SJosef Bacik mutex_unlock(&fs_info->ro_block_group_mutex); 290126ce2095SJosef Bacik 290226ce2095SJosef Bacik btrfs_end_transaction(trans); 290326ce2095SJosef Bacik return ret; 290426ce2095SJosef Bacik } 290526ce2095SJosef Bacik 290632da5386SDavid Sterba void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) 290726ce2095SJosef Bacik { 290826ce2095SJosef Bacik struct btrfs_space_info *sinfo = cache->space_info; 290926ce2095SJosef Bacik u64 num_bytes; 291026ce2095SJosef Bacik 291126ce2095SJosef Bacik BUG_ON(!cache->ro); 291226ce2095SJosef Bacik 291326ce2095SJosef Bacik spin_lock(&sinfo->lock); 291426ce2095SJosef Bacik spin_lock(&cache->lock); 291526ce2095SJosef Bacik if (!--cache->ro) { 2916169e0da9SNaohiro Aota if (btrfs_is_zoned(cache->fs_info)) { 2917169e0da9SNaohiro Aota /* Migrate zone_unusable bytes back */ 291898173255SNaohiro Aota cache->zone_unusable = 291998173255SNaohiro Aota (cache->alloc_offset - cache->used) + 292098173255SNaohiro Aota (cache->length - cache->zone_capacity); 2921169e0da9SNaohiro Aota sinfo->bytes_zone_unusable += cache->zone_unusable; 2922169e0da9SNaohiro Aota sinfo->bytes_readonly -= cache->zone_unusable; 2923169e0da9SNaohiro Aota } 2924f9f28e5bSNaohiro Aota num_bytes = cache->length - cache->reserved - 2925f9f28e5bSNaohiro Aota cache->pinned - cache->bytes_super - 2926f9f28e5bSNaohiro Aota cache->zone_unusable - cache->used; 2927f9f28e5bSNaohiro Aota sinfo->bytes_readonly -= num_bytes; 292826ce2095SJosef Bacik list_del_init(&cache->ro_list); 292926ce2095SJosef Bacik } 293026ce2095SJosef Bacik spin_unlock(&cache->lock); 293126ce2095SJosef Bacik spin_unlock(&sinfo->lock); 293226ce2095SJosef Bacik } 293377745c05SJosef Bacik 29343be4d8efSQu Wenruo static int update_block_group_item(struct btrfs_trans_handle *trans, 293577745c05SJosef Bacik struct btrfs_path *path, 293632da5386SDavid Sterba struct btrfs_block_group *cache) 293777745c05SJosef Bacik { 293877745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 293977745c05SJosef Bacik int ret; 2940dfe8aec4SJosef Bacik struct btrfs_root *root = btrfs_block_group_root(fs_info); 294177745c05SJosef Bacik unsigned long bi; 294277745c05SJosef Bacik struct extent_buffer *leaf; 2943bf38be65SDavid Sterba struct btrfs_block_group_item bgi; 2944b3470b5dSDavid Sterba struct btrfs_key key; 29457248e0ceSQu Wenruo u64 old_commit_used; 29467248e0ceSQu Wenruo u64 used; 29477248e0ceSQu Wenruo 29487248e0ceSQu Wenruo /* 29497248e0ceSQu Wenruo * Block group items update can be triggered out of commit transaction 29507248e0ceSQu Wenruo * critical section, thus we need a consistent view of used bytes. 29517248e0ceSQu Wenruo * We cannot use cache->used directly outside of the spin lock, as it 29527248e0ceSQu Wenruo * may be changed. 29537248e0ceSQu Wenruo */ 29547248e0ceSQu Wenruo spin_lock(&cache->lock); 29557248e0ceSQu Wenruo old_commit_used = cache->commit_used; 29567248e0ceSQu Wenruo used = cache->used; 29577248e0ceSQu Wenruo /* No change in used bytes, can safely skip it. */ 29587248e0ceSQu Wenruo if (cache->commit_used == used) { 29597248e0ceSQu Wenruo spin_unlock(&cache->lock); 29607248e0ceSQu Wenruo return 0; 29617248e0ceSQu Wenruo } 29627248e0ceSQu Wenruo cache->commit_used = used; 29637248e0ceSQu Wenruo spin_unlock(&cache->lock); 296477745c05SJosef Bacik 2965b3470b5dSDavid Sterba key.objectid = cache->start; 2966b3470b5dSDavid Sterba key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; 2967b3470b5dSDavid Sterba key.offset = cache->length; 2968b3470b5dSDavid Sterba 29693be4d8efSQu Wenruo ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 297077745c05SJosef Bacik if (ret) { 297177745c05SJosef Bacik if (ret > 0) 297277745c05SJosef Bacik ret = -ENOENT; 297377745c05SJosef Bacik goto fail; 297477745c05SJosef Bacik } 297577745c05SJosef Bacik 297677745c05SJosef Bacik leaf = path->nodes[0]; 297777745c05SJosef Bacik bi = btrfs_item_ptr_offset(leaf, path->slots[0]); 29787248e0ceSQu Wenruo btrfs_set_stack_block_group_used(&bgi, used); 2979de0dc456SDavid Sterba btrfs_set_stack_block_group_chunk_objectid(&bgi, 2980f7238e50SJosef Bacik cache->global_root_id); 2981de0dc456SDavid Sterba btrfs_set_stack_block_group_flags(&bgi, cache->flags); 2982bf38be65SDavid Sterba write_extent_buffer(leaf, &bgi, bi, sizeof(bgi)); 298377745c05SJosef Bacik btrfs_mark_buffer_dirty(leaf); 298477745c05SJosef Bacik fail: 298577745c05SJosef Bacik btrfs_release_path(path); 29867248e0ceSQu Wenruo /* We didn't update the block group item, need to revert @commit_used. */ 29877248e0ceSQu Wenruo if (ret < 0) { 29887248e0ceSQu Wenruo spin_lock(&cache->lock); 29897248e0ceSQu Wenruo cache->commit_used = old_commit_used; 29907248e0ceSQu Wenruo spin_unlock(&cache->lock); 29917248e0ceSQu Wenruo } 299277745c05SJosef Bacik return ret; 299377745c05SJosef Bacik 299477745c05SJosef Bacik } 299577745c05SJosef Bacik 299632da5386SDavid Sterba static int cache_save_setup(struct btrfs_block_group *block_group, 299777745c05SJosef Bacik struct btrfs_trans_handle *trans, 299877745c05SJosef Bacik struct btrfs_path *path) 299977745c05SJosef Bacik { 300077745c05SJosef Bacik struct btrfs_fs_info *fs_info = block_group->fs_info; 300177745c05SJosef Bacik struct btrfs_root *root = fs_info->tree_root; 300277745c05SJosef Bacik struct inode *inode = NULL; 300377745c05SJosef Bacik struct extent_changeset *data_reserved = NULL; 300477745c05SJosef Bacik u64 alloc_hint = 0; 300577745c05SJosef Bacik int dcs = BTRFS_DC_ERROR; 30060044ae11SQu Wenruo u64 cache_size = 0; 300777745c05SJosef Bacik int retries = 0; 300877745c05SJosef Bacik int ret = 0; 300977745c05SJosef Bacik 3010af456a2cSBoris Burkov if (!btrfs_test_opt(fs_info, SPACE_CACHE)) 3011af456a2cSBoris Burkov return 0; 3012af456a2cSBoris Burkov 301377745c05SJosef Bacik /* 301477745c05SJosef Bacik * If this block group is smaller than 100 megs don't bother caching the 301577745c05SJosef Bacik * block group. 301677745c05SJosef Bacik */ 3017b3470b5dSDavid Sterba if (block_group->length < (100 * SZ_1M)) { 301877745c05SJosef Bacik spin_lock(&block_group->lock); 301977745c05SJosef Bacik block_group->disk_cache_state = BTRFS_DC_WRITTEN; 302077745c05SJosef Bacik spin_unlock(&block_group->lock); 302177745c05SJosef Bacik return 0; 302277745c05SJosef Bacik } 302377745c05SJosef Bacik 3024bf31f87fSDavid Sterba if (TRANS_ABORTED(trans)) 302577745c05SJosef Bacik return 0; 302677745c05SJosef Bacik again: 302777745c05SJosef Bacik inode = lookup_free_space_inode(block_group, path); 302877745c05SJosef Bacik if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { 302977745c05SJosef Bacik ret = PTR_ERR(inode); 303077745c05SJosef Bacik btrfs_release_path(path); 303177745c05SJosef Bacik goto out; 303277745c05SJosef Bacik } 303377745c05SJosef Bacik 303477745c05SJosef Bacik if (IS_ERR(inode)) { 303577745c05SJosef Bacik BUG_ON(retries); 303677745c05SJosef Bacik retries++; 303777745c05SJosef Bacik 303877745c05SJosef Bacik if (block_group->ro) 303977745c05SJosef Bacik goto out_free; 304077745c05SJosef Bacik 304177745c05SJosef Bacik ret = create_free_space_inode(trans, block_group, path); 304277745c05SJosef Bacik if (ret) 304377745c05SJosef Bacik goto out_free; 304477745c05SJosef Bacik goto again; 304577745c05SJosef Bacik } 304677745c05SJosef Bacik 304777745c05SJosef Bacik /* 304877745c05SJosef Bacik * We want to set the generation to 0, that way if anything goes wrong 304977745c05SJosef Bacik * from here on out we know not to trust this cache when we load up next 305077745c05SJosef Bacik * time. 305177745c05SJosef Bacik */ 305277745c05SJosef Bacik BTRFS_I(inode)->generation = 0; 30539a56fcd1SNikolay Borisov ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); 305477745c05SJosef Bacik if (ret) { 305577745c05SJosef Bacik /* 305677745c05SJosef Bacik * So theoretically we could recover from this, simply set the 305777745c05SJosef Bacik * super cache generation to 0 so we know to invalidate the 305877745c05SJosef Bacik * cache, but then we'd have to keep track of the block groups 305977745c05SJosef Bacik * that fail this way so we know we _have_ to reset this cache 306077745c05SJosef Bacik * before the next commit or risk reading stale cache. So to 306177745c05SJosef Bacik * limit our exposure to horrible edge cases lets just abort the 306277745c05SJosef Bacik * transaction, this only happens in really bad situations 306377745c05SJosef Bacik * anyway. 306477745c05SJosef Bacik */ 306577745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 306677745c05SJosef Bacik goto out_put; 306777745c05SJosef Bacik } 306877745c05SJosef Bacik WARN_ON(ret); 306977745c05SJosef Bacik 307077745c05SJosef Bacik /* We've already setup this transaction, go ahead and exit */ 307177745c05SJosef Bacik if (block_group->cache_generation == trans->transid && 307277745c05SJosef Bacik i_size_read(inode)) { 307377745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 307477745c05SJosef Bacik goto out_put; 307577745c05SJosef Bacik } 307677745c05SJosef Bacik 307777745c05SJosef Bacik if (i_size_read(inode) > 0) { 307877745c05SJosef Bacik ret = btrfs_check_trunc_cache_free_space(fs_info, 307977745c05SJosef Bacik &fs_info->global_block_rsv); 308077745c05SJosef Bacik if (ret) 308177745c05SJosef Bacik goto out_put; 308277745c05SJosef Bacik 308377745c05SJosef Bacik ret = btrfs_truncate_free_space_cache(trans, NULL, inode); 308477745c05SJosef Bacik if (ret) 308577745c05SJosef Bacik goto out_put; 308677745c05SJosef Bacik } 308777745c05SJosef Bacik 308877745c05SJosef Bacik spin_lock(&block_group->lock); 308977745c05SJosef Bacik if (block_group->cached != BTRFS_CACHE_FINISHED || 309077745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) { 309177745c05SJosef Bacik /* 309277745c05SJosef Bacik * don't bother trying to write stuff out _if_ 309377745c05SJosef Bacik * a) we're not cached, 309477745c05SJosef Bacik * b) we're with nospace_cache mount option, 309577745c05SJosef Bacik * c) we're with v2 space_cache (FREE_SPACE_TREE). 309677745c05SJosef Bacik */ 309777745c05SJosef Bacik dcs = BTRFS_DC_WRITTEN; 309877745c05SJosef Bacik spin_unlock(&block_group->lock); 309977745c05SJosef Bacik goto out_put; 310077745c05SJosef Bacik } 310177745c05SJosef Bacik spin_unlock(&block_group->lock); 310277745c05SJosef Bacik 310377745c05SJosef Bacik /* 310477745c05SJosef Bacik * We hit an ENOSPC when setting up the cache in this transaction, just 310577745c05SJosef Bacik * skip doing the setup, we've already cleared the cache so we're safe. 310677745c05SJosef Bacik */ 310777745c05SJosef Bacik if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { 310877745c05SJosef Bacik ret = -ENOSPC; 310977745c05SJosef Bacik goto out_put; 311077745c05SJosef Bacik } 311177745c05SJosef Bacik 311277745c05SJosef Bacik /* 311377745c05SJosef Bacik * Try to preallocate enough space based on how big the block group is. 311477745c05SJosef Bacik * Keep in mind this has to include any pinned space which could end up 311577745c05SJosef Bacik * taking up quite a bit since it's not folded into the other space 311677745c05SJosef Bacik * cache. 311777745c05SJosef Bacik */ 31180044ae11SQu Wenruo cache_size = div_u64(block_group->length, SZ_256M); 31190044ae11SQu Wenruo if (!cache_size) 31200044ae11SQu Wenruo cache_size = 1; 312177745c05SJosef Bacik 31220044ae11SQu Wenruo cache_size *= 16; 31230044ae11SQu Wenruo cache_size *= fs_info->sectorsize; 312477745c05SJosef Bacik 312536ea6f3eSNikolay Borisov ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved, 0, 31261daedb1dSJosef Bacik cache_size, false); 312777745c05SJosef Bacik if (ret) 312877745c05SJosef Bacik goto out_put; 312977745c05SJosef Bacik 31300044ae11SQu Wenruo ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, cache_size, 31310044ae11SQu Wenruo cache_size, cache_size, 313277745c05SJosef Bacik &alloc_hint); 313377745c05SJosef Bacik /* 313477745c05SJosef Bacik * Our cache requires contiguous chunks so that we don't modify a bunch 313577745c05SJosef Bacik * of metadata or split extents when writing the cache out, which means 313677745c05SJosef Bacik * we can enospc if we are heavily fragmented in addition to just normal 313777745c05SJosef Bacik * out of space conditions. So if we hit this just skip setting up any 313877745c05SJosef Bacik * other block groups for this transaction, maybe we'll unpin enough 313977745c05SJosef Bacik * space the next time around. 314077745c05SJosef Bacik */ 314177745c05SJosef Bacik if (!ret) 314277745c05SJosef Bacik dcs = BTRFS_DC_SETUP; 314377745c05SJosef Bacik else if (ret == -ENOSPC) 314477745c05SJosef Bacik set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); 314577745c05SJosef Bacik 314677745c05SJosef Bacik out_put: 314777745c05SJosef Bacik iput(inode); 314877745c05SJosef Bacik out_free: 314977745c05SJosef Bacik btrfs_release_path(path); 315077745c05SJosef Bacik out: 315177745c05SJosef Bacik spin_lock(&block_group->lock); 315277745c05SJosef Bacik if (!ret && dcs == BTRFS_DC_SETUP) 315377745c05SJosef Bacik block_group->cache_generation = trans->transid; 315477745c05SJosef Bacik block_group->disk_cache_state = dcs; 315577745c05SJosef Bacik spin_unlock(&block_group->lock); 315677745c05SJosef Bacik 315777745c05SJosef Bacik extent_changeset_free(data_reserved); 315877745c05SJosef Bacik return ret; 315977745c05SJosef Bacik } 316077745c05SJosef Bacik 316177745c05SJosef Bacik int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) 316277745c05SJosef Bacik { 316377745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 316432da5386SDavid Sterba struct btrfs_block_group *cache, *tmp; 316577745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 316677745c05SJosef Bacik struct btrfs_path *path; 316777745c05SJosef Bacik 316877745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs) || 316977745c05SJosef Bacik !btrfs_test_opt(fs_info, SPACE_CACHE)) 317077745c05SJosef Bacik return 0; 317177745c05SJosef Bacik 317277745c05SJosef Bacik path = btrfs_alloc_path(); 317377745c05SJosef Bacik if (!path) 317477745c05SJosef Bacik return -ENOMEM; 317577745c05SJosef Bacik 317677745c05SJosef Bacik /* Could add new block groups, use _safe just in case */ 317777745c05SJosef Bacik list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, 317877745c05SJosef Bacik dirty_list) { 317977745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_CLEAR) 318077745c05SJosef Bacik cache_save_setup(cache, trans, path); 318177745c05SJosef Bacik } 318277745c05SJosef Bacik 318377745c05SJosef Bacik btrfs_free_path(path); 318477745c05SJosef Bacik return 0; 318577745c05SJosef Bacik } 318677745c05SJosef Bacik 318777745c05SJosef Bacik /* 318877745c05SJosef Bacik * Transaction commit does final block group cache writeback during a critical 318977745c05SJosef Bacik * section where nothing is allowed to change the FS. This is required in 319077745c05SJosef Bacik * order for the cache to actually match the block group, but can introduce a 319177745c05SJosef Bacik * lot of latency into the commit. 319277745c05SJosef Bacik * 319377745c05SJosef Bacik * So, btrfs_start_dirty_block_groups is here to kick off block group cache IO. 319477745c05SJosef Bacik * There's a chance we'll have to redo some of it if the block group changes 319577745c05SJosef Bacik * again during the commit, but it greatly reduces the commit latency by 319677745c05SJosef Bacik * getting rid of the easy block groups while we're still allowing others to 319777745c05SJosef Bacik * join the commit. 319877745c05SJosef Bacik */ 319977745c05SJosef Bacik int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) 320077745c05SJosef Bacik { 320177745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 320232da5386SDavid Sterba struct btrfs_block_group *cache; 320377745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 320477745c05SJosef Bacik int ret = 0; 320577745c05SJosef Bacik int should_put; 320677745c05SJosef Bacik struct btrfs_path *path = NULL; 320777745c05SJosef Bacik LIST_HEAD(dirty); 320877745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 320977745c05SJosef Bacik int loops = 0; 321077745c05SJosef Bacik 321177745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 321277745c05SJosef Bacik if (list_empty(&cur_trans->dirty_bgs)) { 321377745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 321477745c05SJosef Bacik return 0; 321577745c05SJosef Bacik } 321677745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 321777745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 321877745c05SJosef Bacik 321977745c05SJosef Bacik again: 322077745c05SJosef Bacik /* Make sure all the block groups on our dirty list actually exist */ 322177745c05SJosef Bacik btrfs_create_pending_block_groups(trans); 322277745c05SJosef Bacik 322377745c05SJosef Bacik if (!path) { 322477745c05SJosef Bacik path = btrfs_alloc_path(); 3225938fcbfbSJosef Bacik if (!path) { 3226938fcbfbSJosef Bacik ret = -ENOMEM; 3227938fcbfbSJosef Bacik goto out; 3228938fcbfbSJosef Bacik } 322977745c05SJosef Bacik } 323077745c05SJosef Bacik 323177745c05SJosef Bacik /* 323277745c05SJosef Bacik * cache_write_mutex is here only to save us from balance or automatic 323377745c05SJosef Bacik * removal of empty block groups deleting this block group while we are 323477745c05SJosef Bacik * writing out the cache 323577745c05SJosef Bacik */ 323677745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 323777745c05SJosef Bacik while (!list_empty(&dirty)) { 323877745c05SJosef Bacik bool drop_reserve = true; 323977745c05SJosef Bacik 324032da5386SDavid Sterba cache = list_first_entry(&dirty, struct btrfs_block_group, 324177745c05SJosef Bacik dirty_list); 324277745c05SJosef Bacik /* 324377745c05SJosef Bacik * This can happen if something re-dirties a block group that 324477745c05SJosef Bacik * is already under IO. Just wait for it to finish and then do 324577745c05SJosef Bacik * it all again 324677745c05SJosef Bacik */ 324777745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 324877745c05SJosef Bacik list_del_init(&cache->io_list); 324977745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 325077745c05SJosef Bacik btrfs_put_block_group(cache); 325177745c05SJosef Bacik } 325277745c05SJosef Bacik 325377745c05SJosef Bacik 325477745c05SJosef Bacik /* 325577745c05SJosef Bacik * btrfs_wait_cache_io uses the cache->dirty_list to decide if 325677745c05SJosef Bacik * it should update the cache_state. Don't delete until after 325777745c05SJosef Bacik * we wait. 325877745c05SJosef Bacik * 325977745c05SJosef Bacik * Since we're not running in the commit critical section 326077745c05SJosef Bacik * we need the dirty_bgs_lock to protect from update_block_group 326177745c05SJosef Bacik */ 326277745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 326377745c05SJosef Bacik list_del_init(&cache->dirty_list); 326477745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 326577745c05SJosef Bacik 326677745c05SJosef Bacik should_put = 1; 326777745c05SJosef Bacik 326877745c05SJosef Bacik cache_save_setup(cache, trans, path); 326977745c05SJosef Bacik 327077745c05SJosef Bacik if (cache->disk_cache_state == BTRFS_DC_SETUP) { 327177745c05SJosef Bacik cache->io_ctl.inode = NULL; 327277745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 327377745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 327477745c05SJosef Bacik should_put = 0; 327577745c05SJosef Bacik 327677745c05SJosef Bacik /* 327777745c05SJosef Bacik * The cache_write_mutex is protecting the 327877745c05SJosef Bacik * io_list, also refer to the definition of 327977745c05SJosef Bacik * btrfs_transaction::io_bgs for more details 328077745c05SJosef Bacik */ 328177745c05SJosef Bacik list_add_tail(&cache->io_list, io); 328277745c05SJosef Bacik } else { 328377745c05SJosef Bacik /* 328477745c05SJosef Bacik * If we failed to write the cache, the 328577745c05SJosef Bacik * generation will be bad and life goes on 328677745c05SJosef Bacik */ 328777745c05SJosef Bacik ret = 0; 328877745c05SJosef Bacik } 328977745c05SJosef Bacik } 329077745c05SJosef Bacik if (!ret) { 32913be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 329277745c05SJosef Bacik /* 329377745c05SJosef Bacik * Our block group might still be attached to the list 329477745c05SJosef Bacik * of new block groups in the transaction handle of some 329577745c05SJosef Bacik * other task (struct btrfs_trans_handle->new_bgs). This 329677745c05SJosef Bacik * means its block group item isn't yet in the extent 329777745c05SJosef Bacik * tree. If this happens ignore the error, as we will 329877745c05SJosef Bacik * try again later in the critical section of the 329977745c05SJosef Bacik * transaction commit. 330077745c05SJosef Bacik */ 330177745c05SJosef Bacik if (ret == -ENOENT) { 330277745c05SJosef Bacik ret = 0; 330377745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 330477745c05SJosef Bacik if (list_empty(&cache->dirty_list)) { 330577745c05SJosef Bacik list_add_tail(&cache->dirty_list, 330677745c05SJosef Bacik &cur_trans->dirty_bgs); 330777745c05SJosef Bacik btrfs_get_block_group(cache); 330877745c05SJosef Bacik drop_reserve = false; 330977745c05SJosef Bacik } 331077745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 331177745c05SJosef Bacik } else if (ret) { 331277745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 331377745c05SJosef Bacik } 331477745c05SJosef Bacik } 331577745c05SJosef Bacik 331677745c05SJosef Bacik /* If it's not on the io list, we need to put the block group */ 331777745c05SJosef Bacik if (should_put) 331877745c05SJosef Bacik btrfs_put_block_group(cache); 331977745c05SJosef Bacik if (drop_reserve) 332077745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 332177745c05SJosef Bacik /* 332277745c05SJosef Bacik * Avoid blocking other tasks for too long. It might even save 332377745c05SJosef Bacik * us from writing caches for block groups that are going to be 332477745c05SJosef Bacik * removed. 332577745c05SJosef Bacik */ 332677745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 3327938fcbfbSJosef Bacik if (ret) 3328938fcbfbSJosef Bacik goto out; 332977745c05SJosef Bacik mutex_lock(&trans->transaction->cache_write_mutex); 333077745c05SJosef Bacik } 333177745c05SJosef Bacik mutex_unlock(&trans->transaction->cache_write_mutex); 333277745c05SJosef Bacik 333377745c05SJosef Bacik /* 333477745c05SJosef Bacik * Go through delayed refs for all the stuff we've just kicked off 333577745c05SJosef Bacik * and then loop back (just once) 333677745c05SJosef Bacik */ 333734d1eb0eSJosef Bacik if (!ret) 333877745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 0); 333977745c05SJosef Bacik if (!ret && loops == 0) { 334077745c05SJosef Bacik loops++; 334177745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 334277745c05SJosef Bacik list_splice_init(&cur_trans->dirty_bgs, &dirty); 334377745c05SJosef Bacik /* 334477745c05SJosef Bacik * dirty_bgs_lock protects us from concurrent block group 334577745c05SJosef Bacik * deletes too (not just cache_write_mutex). 334677745c05SJosef Bacik */ 334777745c05SJosef Bacik if (!list_empty(&dirty)) { 334877745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 334977745c05SJosef Bacik goto again; 335077745c05SJosef Bacik } 335177745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 3352938fcbfbSJosef Bacik } 3353938fcbfbSJosef Bacik out: 3354938fcbfbSJosef Bacik if (ret < 0) { 3355938fcbfbSJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 3356938fcbfbSJosef Bacik list_splice_init(&dirty, &cur_trans->dirty_bgs); 3357938fcbfbSJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 335877745c05SJosef Bacik btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 335977745c05SJosef Bacik } 336077745c05SJosef Bacik 336177745c05SJosef Bacik btrfs_free_path(path); 336277745c05SJosef Bacik return ret; 336377745c05SJosef Bacik } 336477745c05SJosef Bacik 336577745c05SJosef Bacik int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) 336677745c05SJosef Bacik { 336777745c05SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 336832da5386SDavid Sterba struct btrfs_block_group *cache; 336977745c05SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 337077745c05SJosef Bacik int ret = 0; 337177745c05SJosef Bacik int should_put; 337277745c05SJosef Bacik struct btrfs_path *path; 337377745c05SJosef Bacik struct list_head *io = &cur_trans->io_bgs; 337477745c05SJosef Bacik 337577745c05SJosef Bacik path = btrfs_alloc_path(); 337677745c05SJosef Bacik if (!path) 337777745c05SJosef Bacik return -ENOMEM; 337877745c05SJosef Bacik 337977745c05SJosef Bacik /* 338077745c05SJosef Bacik * Even though we are in the critical section of the transaction commit, 338177745c05SJosef Bacik * we can still have concurrent tasks adding elements to this 338277745c05SJosef Bacik * transaction's list of dirty block groups. These tasks correspond to 338377745c05SJosef Bacik * endio free space workers started when writeback finishes for a 338477745c05SJosef Bacik * space cache, which run inode.c:btrfs_finish_ordered_io(), and can 338577745c05SJosef Bacik * allocate new block groups as a result of COWing nodes of the root 338677745c05SJosef Bacik * tree when updating the free space inode. The writeback for the space 338777745c05SJosef Bacik * caches is triggered by an earlier call to 338877745c05SJosef Bacik * btrfs_start_dirty_block_groups() and iterations of the following 338977745c05SJosef Bacik * loop. 339077745c05SJosef Bacik * Also we want to do the cache_save_setup first and then run the 339177745c05SJosef Bacik * delayed refs to make sure we have the best chance at doing this all 339277745c05SJosef Bacik * in one shot. 339377745c05SJosef Bacik */ 339477745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 339577745c05SJosef Bacik while (!list_empty(&cur_trans->dirty_bgs)) { 339677745c05SJosef Bacik cache = list_first_entry(&cur_trans->dirty_bgs, 339732da5386SDavid Sterba struct btrfs_block_group, 339877745c05SJosef Bacik dirty_list); 339977745c05SJosef Bacik 340077745c05SJosef Bacik /* 340177745c05SJosef Bacik * This can happen if cache_save_setup re-dirties a block group 340277745c05SJosef Bacik * that is already under IO. Just wait for it to finish and 340377745c05SJosef Bacik * then do it all again 340477745c05SJosef Bacik */ 340577745c05SJosef Bacik if (!list_empty(&cache->io_list)) { 340677745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 340777745c05SJosef Bacik list_del_init(&cache->io_list); 340877745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 340977745c05SJosef Bacik btrfs_put_block_group(cache); 341077745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 341177745c05SJosef Bacik } 341277745c05SJosef Bacik 341377745c05SJosef Bacik /* 341477745c05SJosef Bacik * Don't remove from the dirty list until after we've waited on 341577745c05SJosef Bacik * any pending IO 341677745c05SJosef Bacik */ 341777745c05SJosef Bacik list_del_init(&cache->dirty_list); 341877745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 341977745c05SJosef Bacik should_put = 1; 342077745c05SJosef Bacik 342177745c05SJosef Bacik cache_save_setup(cache, trans, path); 342277745c05SJosef Bacik 342377745c05SJosef Bacik if (!ret) 342477745c05SJosef Bacik ret = btrfs_run_delayed_refs(trans, 342577745c05SJosef Bacik (unsigned long) -1); 342677745c05SJosef Bacik 342777745c05SJosef Bacik if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { 342877745c05SJosef Bacik cache->io_ctl.inode = NULL; 342977745c05SJosef Bacik ret = btrfs_write_out_cache(trans, cache, path); 343077745c05SJosef Bacik if (ret == 0 && cache->io_ctl.inode) { 343177745c05SJosef Bacik should_put = 0; 343277745c05SJosef Bacik list_add_tail(&cache->io_list, io); 343377745c05SJosef Bacik } else { 343477745c05SJosef Bacik /* 343577745c05SJosef Bacik * If we failed to write the cache, the 343677745c05SJosef Bacik * generation will be bad and life goes on 343777745c05SJosef Bacik */ 343877745c05SJosef Bacik ret = 0; 343977745c05SJosef Bacik } 344077745c05SJosef Bacik } 344177745c05SJosef Bacik if (!ret) { 34423be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 344377745c05SJosef Bacik /* 344477745c05SJosef Bacik * One of the free space endio workers might have 344577745c05SJosef Bacik * created a new block group while updating a free space 344677745c05SJosef Bacik * cache's inode (at inode.c:btrfs_finish_ordered_io()) 344777745c05SJosef Bacik * and hasn't released its transaction handle yet, in 344877745c05SJosef Bacik * which case the new block group is still attached to 344977745c05SJosef Bacik * its transaction handle and its creation has not 345077745c05SJosef Bacik * finished yet (no block group item in the extent tree 345177745c05SJosef Bacik * yet, etc). If this is the case, wait for all free 345277745c05SJosef Bacik * space endio workers to finish and retry. This is a 3453260db43cSRandy Dunlap * very rare case so no need for a more efficient and 345477745c05SJosef Bacik * complex approach. 345577745c05SJosef Bacik */ 345677745c05SJosef Bacik if (ret == -ENOENT) { 345777745c05SJosef Bacik wait_event(cur_trans->writer_wait, 345877745c05SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 34593be4d8efSQu Wenruo ret = update_block_group_item(trans, path, cache); 346077745c05SJosef Bacik } 346177745c05SJosef Bacik if (ret) 346277745c05SJosef Bacik btrfs_abort_transaction(trans, ret); 346377745c05SJosef Bacik } 346477745c05SJosef Bacik 346577745c05SJosef Bacik /* If its not on the io list, we need to put the block group */ 346677745c05SJosef Bacik if (should_put) 346777745c05SJosef Bacik btrfs_put_block_group(cache); 346877745c05SJosef Bacik btrfs_delayed_refs_rsv_release(fs_info, 1); 346977745c05SJosef Bacik spin_lock(&cur_trans->dirty_bgs_lock); 347077745c05SJosef Bacik } 347177745c05SJosef Bacik spin_unlock(&cur_trans->dirty_bgs_lock); 347277745c05SJosef Bacik 347377745c05SJosef Bacik /* 347477745c05SJosef Bacik * Refer to the definition of io_bgs member for details why it's safe 347577745c05SJosef Bacik * to use it without any locking 347677745c05SJosef Bacik */ 347777745c05SJosef Bacik while (!list_empty(io)) { 347832da5386SDavid Sterba cache = list_first_entry(io, struct btrfs_block_group, 347977745c05SJosef Bacik io_list); 348077745c05SJosef Bacik list_del_init(&cache->io_list); 348177745c05SJosef Bacik btrfs_wait_cache_io(trans, cache, path); 348277745c05SJosef Bacik btrfs_put_block_group(cache); 348377745c05SJosef Bacik } 348477745c05SJosef Bacik 348577745c05SJosef Bacik btrfs_free_path(path); 348677745c05SJosef Bacik return ret; 348777745c05SJosef Bacik } 3488606d1bf1SJosef Bacik 3489606d1bf1SJosef Bacik int btrfs_update_block_group(struct btrfs_trans_handle *trans, 349011b66fa6SAnand Jain u64 bytenr, u64 num_bytes, bool alloc) 3491606d1bf1SJosef Bacik { 3492606d1bf1SJosef Bacik struct btrfs_fs_info *info = trans->fs_info; 349332da5386SDavid Sterba struct btrfs_block_group *cache = NULL; 3494606d1bf1SJosef Bacik u64 total = num_bytes; 3495606d1bf1SJosef Bacik u64 old_val; 3496606d1bf1SJosef Bacik u64 byte_in_group; 3497606d1bf1SJosef Bacik int factor; 3498606d1bf1SJosef Bacik int ret = 0; 3499606d1bf1SJosef Bacik 3500606d1bf1SJosef Bacik /* Block accounting for super block */ 3501606d1bf1SJosef Bacik spin_lock(&info->delalloc_root_lock); 3502606d1bf1SJosef Bacik old_val = btrfs_super_bytes_used(info->super_copy); 3503606d1bf1SJosef Bacik if (alloc) 3504606d1bf1SJosef Bacik old_val += num_bytes; 3505606d1bf1SJosef Bacik else 3506606d1bf1SJosef Bacik old_val -= num_bytes; 3507606d1bf1SJosef Bacik btrfs_set_super_bytes_used(info->super_copy, old_val); 3508606d1bf1SJosef Bacik spin_unlock(&info->delalloc_root_lock); 3509606d1bf1SJosef Bacik 3510606d1bf1SJosef Bacik while (total) { 3511df384da5SJosef Bacik struct btrfs_space_info *space_info; 3512efbf35a1SJosef Bacik bool reclaim = false; 3513ac2f1e63SJosef Bacik 3514606d1bf1SJosef Bacik cache = btrfs_lookup_block_group(info, bytenr); 3515606d1bf1SJosef Bacik if (!cache) { 3516606d1bf1SJosef Bacik ret = -ENOENT; 3517606d1bf1SJosef Bacik break; 3518606d1bf1SJosef Bacik } 3519df384da5SJosef Bacik space_info = cache->space_info; 3520606d1bf1SJosef Bacik factor = btrfs_bg_type_to_factor(cache->flags); 3521606d1bf1SJosef Bacik 3522606d1bf1SJosef Bacik /* 3523606d1bf1SJosef Bacik * If this block group has free space cache written out, we 3524606d1bf1SJosef Bacik * need to make sure to load it if we are removing space. This 3525606d1bf1SJosef Bacik * is because we need the unpinning stage to actually add the 3526606d1bf1SJosef Bacik * space back to the block group, otherwise we will leak space. 3527606d1bf1SJosef Bacik */ 352832da5386SDavid Sterba if (!alloc && !btrfs_block_group_done(cache)) 3529ced8ecf0SOmar Sandoval btrfs_cache_block_group(cache, true); 3530606d1bf1SJosef Bacik 3531b3470b5dSDavid Sterba byte_in_group = bytenr - cache->start; 3532b3470b5dSDavid Sterba WARN_ON(byte_in_group > cache->length); 3533606d1bf1SJosef Bacik 3534df384da5SJosef Bacik spin_lock(&space_info->lock); 3535606d1bf1SJosef Bacik spin_lock(&cache->lock); 3536606d1bf1SJosef Bacik 3537606d1bf1SJosef Bacik if (btrfs_test_opt(info, SPACE_CACHE) && 3538606d1bf1SJosef Bacik cache->disk_cache_state < BTRFS_DC_CLEAR) 3539606d1bf1SJosef Bacik cache->disk_cache_state = BTRFS_DC_CLEAR; 3540606d1bf1SJosef Bacik 3541bf38be65SDavid Sterba old_val = cache->used; 3542b3470b5dSDavid Sterba num_bytes = min(total, cache->length - byte_in_group); 3543606d1bf1SJosef Bacik if (alloc) { 3544606d1bf1SJosef Bacik old_val += num_bytes; 3545bf38be65SDavid Sterba cache->used = old_val; 3546606d1bf1SJosef Bacik cache->reserved -= num_bytes; 3547df384da5SJosef Bacik space_info->bytes_reserved -= num_bytes; 3548df384da5SJosef Bacik space_info->bytes_used += num_bytes; 3549df384da5SJosef Bacik space_info->disk_used += num_bytes * factor; 3550606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3551df384da5SJosef Bacik spin_unlock(&space_info->lock); 3552606d1bf1SJosef Bacik } else { 3553606d1bf1SJosef Bacik old_val -= num_bytes; 3554bf38be65SDavid Sterba cache->used = old_val; 3555606d1bf1SJosef Bacik cache->pinned += num_bytes; 3556df384da5SJosef Bacik btrfs_space_info_update_bytes_pinned(info, space_info, 3557df384da5SJosef Bacik num_bytes); 3558df384da5SJosef Bacik space_info->bytes_used -= num_bytes; 3559df384da5SJosef Bacik space_info->disk_used -= num_bytes * factor; 3560ac2f1e63SJosef Bacik 3561ac2f1e63SJosef Bacik reclaim = should_reclaim_block_group(cache, num_bytes); 356252bb7a21SBoris Burkov 3563606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3564df384da5SJosef Bacik spin_unlock(&space_info->lock); 3565606d1bf1SJosef Bacik 3566fe1a598cSDavid Sterba set_extent_bit(&trans->transaction->pinned_extents, 3567606d1bf1SJosef Bacik bytenr, bytenr + num_bytes - 1, 35681d126800SDavid Sterba EXTENT_DIRTY, NULL); 3569606d1bf1SJosef Bacik } 3570606d1bf1SJosef Bacik 3571606d1bf1SJosef Bacik spin_lock(&trans->transaction->dirty_bgs_lock); 3572606d1bf1SJosef Bacik if (list_empty(&cache->dirty_list)) { 3573606d1bf1SJosef Bacik list_add_tail(&cache->dirty_list, 3574606d1bf1SJosef Bacik &trans->transaction->dirty_bgs); 3575606d1bf1SJosef Bacik trans->delayed_ref_updates++; 3576606d1bf1SJosef Bacik btrfs_get_block_group(cache); 3577606d1bf1SJosef Bacik } 3578606d1bf1SJosef Bacik spin_unlock(&trans->transaction->dirty_bgs_lock); 3579606d1bf1SJosef Bacik 3580606d1bf1SJosef Bacik /* 3581606d1bf1SJosef Bacik * No longer have used bytes in this block group, queue it for 3582606d1bf1SJosef Bacik * deletion. We do this after adding the block group to the 3583606d1bf1SJosef Bacik * dirty list to avoid races between cleaner kthread and space 3584606d1bf1SJosef Bacik * cache writeout. 3585606d1bf1SJosef Bacik */ 35866e80d4f8SDennis Zhou if (!alloc && old_val == 0) { 35876e80d4f8SDennis Zhou if (!btrfs_test_opt(info, DISCARD_ASYNC)) 3588606d1bf1SJosef Bacik btrfs_mark_bg_unused(cache); 3589ac2f1e63SJosef Bacik } else if (!alloc && reclaim) { 3590ac2f1e63SJosef Bacik btrfs_mark_bg_to_reclaim(cache); 35916e80d4f8SDennis Zhou } 3592606d1bf1SJosef Bacik 3593606d1bf1SJosef Bacik btrfs_put_block_group(cache); 3594606d1bf1SJosef Bacik total -= num_bytes; 3595606d1bf1SJosef Bacik bytenr += num_bytes; 3596606d1bf1SJosef Bacik } 3597606d1bf1SJosef Bacik 3598606d1bf1SJosef Bacik /* Modified block groups are accounted for in the delayed_refs_rsv. */ 3599606d1bf1SJosef Bacik btrfs_update_delayed_refs_rsv(trans); 3600606d1bf1SJosef Bacik return ret; 3601606d1bf1SJosef Bacik } 3602606d1bf1SJosef Bacik 360343dd529aSDavid Sterba /* 360443dd529aSDavid Sterba * Update the block_group and space info counters. 360543dd529aSDavid Sterba * 3606606d1bf1SJosef Bacik * @cache: The cache we are manipulating 3607606d1bf1SJosef Bacik * @ram_bytes: The number of bytes of file content, and will be same to 3608606d1bf1SJosef Bacik * @num_bytes except for the compress path. 3609606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 3610606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 3611606d1bf1SJosef Bacik * 3612606d1bf1SJosef Bacik * This is called by the allocator when it reserves space. If this is a 3613606d1bf1SJosef Bacik * reservation and the block group has become read only we cannot make the 3614606d1bf1SJosef Bacik * reservation and return -EAGAIN, otherwise this function always succeeds. 3615606d1bf1SJosef Bacik */ 361632da5386SDavid Sterba int btrfs_add_reserved_bytes(struct btrfs_block_group *cache, 361752bb7a21SBoris Burkov u64 ram_bytes, u64 num_bytes, int delalloc, 361852bb7a21SBoris Burkov bool force_wrong_size_class) 3619606d1bf1SJosef Bacik { 3620606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 362152bb7a21SBoris Burkov enum btrfs_block_group_size_class size_class; 3622606d1bf1SJosef Bacik int ret = 0; 3623606d1bf1SJosef Bacik 3624606d1bf1SJosef Bacik spin_lock(&space_info->lock); 3625606d1bf1SJosef Bacik spin_lock(&cache->lock); 3626606d1bf1SJosef Bacik if (cache->ro) { 3627606d1bf1SJosef Bacik ret = -EAGAIN; 362852bb7a21SBoris Burkov goto out; 362952bb7a21SBoris Burkov } 363052bb7a21SBoris Burkov 3631cb0922f2SBoris Burkov if (btrfs_block_group_should_use_size_class(cache)) { 363252bb7a21SBoris Burkov size_class = btrfs_calc_block_group_size_class(num_bytes); 363352bb7a21SBoris Burkov ret = btrfs_use_block_group_size_class(cache, size_class, force_wrong_size_class); 363452bb7a21SBoris Burkov if (ret) 363552bb7a21SBoris Burkov goto out; 363652bb7a21SBoris Burkov } 3637606d1bf1SJosef Bacik cache->reserved += num_bytes; 3638606d1bf1SJosef Bacik space_info->bytes_reserved += num_bytes; 3639a43c3835SJosef Bacik trace_btrfs_space_reservation(cache->fs_info, "space_info", 3640a43c3835SJosef Bacik space_info->flags, num_bytes, 1); 3641606d1bf1SJosef Bacik btrfs_space_info_update_bytes_may_use(cache->fs_info, 3642606d1bf1SJosef Bacik space_info, -ram_bytes); 3643606d1bf1SJosef Bacik if (delalloc) 3644606d1bf1SJosef Bacik cache->delalloc_bytes += num_bytes; 364599ffb43eSJosef Bacik 364699ffb43eSJosef Bacik /* 364752bb7a21SBoris Burkov * Compression can use less space than we reserved, so wake tickets if 364852bb7a21SBoris Burkov * that happens. 364999ffb43eSJosef Bacik */ 365099ffb43eSJosef Bacik if (num_bytes < ram_bytes) 365199ffb43eSJosef Bacik btrfs_try_granting_tickets(cache->fs_info, space_info); 365252bb7a21SBoris Burkov out: 3653606d1bf1SJosef Bacik spin_unlock(&cache->lock); 3654606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 3655606d1bf1SJosef Bacik return ret; 3656606d1bf1SJosef Bacik } 3657606d1bf1SJosef Bacik 365843dd529aSDavid Sterba /* 365943dd529aSDavid Sterba * Update the block_group and space info counters. 366043dd529aSDavid Sterba * 3661606d1bf1SJosef Bacik * @cache: The cache we are manipulating 3662606d1bf1SJosef Bacik * @num_bytes: The number of bytes in question 3663606d1bf1SJosef Bacik * @delalloc: The blocks are allocated for the delalloc write 3664606d1bf1SJosef Bacik * 3665606d1bf1SJosef Bacik * This is called by somebody who is freeing space that was never actually used 3666606d1bf1SJosef Bacik * on disk. For example if you reserve some space for a new leaf in transaction 3667606d1bf1SJosef Bacik * A and before transaction A commits you free that leaf, you call this with 3668606d1bf1SJosef Bacik * reserve set to 0 in order to clear the reservation. 3669606d1bf1SJosef Bacik */ 367032da5386SDavid Sterba void btrfs_free_reserved_bytes(struct btrfs_block_group *cache, 3671606d1bf1SJosef Bacik u64 num_bytes, int delalloc) 3672606d1bf1SJosef Bacik { 3673606d1bf1SJosef Bacik struct btrfs_space_info *space_info = cache->space_info; 3674606d1bf1SJosef Bacik 3675606d1bf1SJosef Bacik spin_lock(&space_info->lock); 3676606d1bf1SJosef Bacik spin_lock(&cache->lock); 3677606d1bf1SJosef Bacik if (cache->ro) 3678606d1bf1SJosef Bacik space_info->bytes_readonly += num_bytes; 3679606d1bf1SJosef Bacik cache->reserved -= num_bytes; 3680606d1bf1SJosef Bacik space_info->bytes_reserved -= num_bytes; 3681606d1bf1SJosef Bacik space_info->max_extent_size = 0; 3682606d1bf1SJosef Bacik 3683606d1bf1SJosef Bacik if (delalloc) 3684606d1bf1SJosef Bacik cache->delalloc_bytes -= num_bytes; 3685606d1bf1SJosef Bacik spin_unlock(&cache->lock); 36863308234aSJosef Bacik 36873308234aSJosef Bacik btrfs_try_granting_tickets(cache->fs_info, space_info); 3688606d1bf1SJosef Bacik spin_unlock(&space_info->lock); 3689606d1bf1SJosef Bacik } 369007730d87SJosef Bacik 369107730d87SJosef Bacik static void force_metadata_allocation(struct btrfs_fs_info *info) 369207730d87SJosef Bacik { 369307730d87SJosef Bacik struct list_head *head = &info->space_info; 369407730d87SJosef Bacik struct btrfs_space_info *found; 369507730d87SJosef Bacik 369672804905SJosef Bacik list_for_each_entry(found, head, list) { 369707730d87SJosef Bacik if (found->flags & BTRFS_BLOCK_GROUP_METADATA) 369807730d87SJosef Bacik found->force_alloc = CHUNK_ALLOC_FORCE; 369907730d87SJosef Bacik } 370007730d87SJosef Bacik } 370107730d87SJosef Bacik 370207730d87SJosef Bacik static int should_alloc_chunk(struct btrfs_fs_info *fs_info, 370307730d87SJosef Bacik struct btrfs_space_info *sinfo, int force) 370407730d87SJosef Bacik { 370507730d87SJosef Bacik u64 bytes_used = btrfs_space_info_used(sinfo, false); 370607730d87SJosef Bacik u64 thresh; 370707730d87SJosef Bacik 370807730d87SJosef Bacik if (force == CHUNK_ALLOC_FORCE) 370907730d87SJosef Bacik return 1; 371007730d87SJosef Bacik 371107730d87SJosef Bacik /* 371207730d87SJosef Bacik * in limited mode, we want to have some free space up to 371307730d87SJosef Bacik * about 1% of the FS size. 371407730d87SJosef Bacik */ 371507730d87SJosef Bacik if (force == CHUNK_ALLOC_LIMITED) { 371607730d87SJosef Bacik thresh = btrfs_super_total_bytes(fs_info->super_copy); 3717428c8e03SDavid Sterba thresh = max_t(u64, SZ_64M, mult_perc(thresh, 1)); 371807730d87SJosef Bacik 371907730d87SJosef Bacik if (sinfo->total_bytes - bytes_used < thresh) 372007730d87SJosef Bacik return 1; 372107730d87SJosef Bacik } 372207730d87SJosef Bacik 3723428c8e03SDavid Sterba if (bytes_used + SZ_2M < mult_perc(sinfo->total_bytes, 80)) 372407730d87SJosef Bacik return 0; 372507730d87SJosef Bacik return 1; 372607730d87SJosef Bacik } 372707730d87SJosef Bacik 372807730d87SJosef Bacik int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) 372907730d87SJosef Bacik { 373007730d87SJosef Bacik u64 alloc_flags = btrfs_get_alloc_profile(trans->fs_info, type); 373107730d87SJosef Bacik 373207730d87SJosef Bacik return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); 373307730d87SJosef Bacik } 373407730d87SJosef Bacik 3735820c363bSNaohiro Aota static struct btrfs_block_group *do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags) 373679bd3712SFilipe Manana { 373779bd3712SFilipe Manana struct btrfs_block_group *bg; 373879bd3712SFilipe Manana int ret; 373979bd3712SFilipe Manana 374007730d87SJosef Bacik /* 374179bd3712SFilipe Manana * Check if we have enough space in the system space info because we 374279bd3712SFilipe Manana * will need to update device items in the chunk btree and insert a new 374379bd3712SFilipe Manana * chunk item in the chunk btree as well. This will allocate a new 374479bd3712SFilipe Manana * system block group if needed. 374579bd3712SFilipe Manana */ 374679bd3712SFilipe Manana check_system_chunk(trans, flags); 374779bd3712SFilipe Manana 3748f6f39f7aSNikolay Borisov bg = btrfs_create_chunk(trans, flags); 374979bd3712SFilipe Manana if (IS_ERR(bg)) { 375079bd3712SFilipe Manana ret = PTR_ERR(bg); 375179bd3712SFilipe Manana goto out; 375279bd3712SFilipe Manana } 375379bd3712SFilipe Manana 375479bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); 375579bd3712SFilipe Manana /* 375679bd3712SFilipe Manana * Normally we are not expected to fail with -ENOSPC here, since we have 375779bd3712SFilipe Manana * previously reserved space in the system space_info and allocated one 3758ecd84d54SFilipe Manana * new system chunk if necessary. However there are three exceptions: 375979bd3712SFilipe Manana * 376079bd3712SFilipe Manana * 1) We may have enough free space in the system space_info but all the 376179bd3712SFilipe Manana * existing system block groups have a profile which can not be used 376279bd3712SFilipe Manana * for extent allocation. 376379bd3712SFilipe Manana * 376479bd3712SFilipe Manana * This happens when mounting in degraded mode. For example we have a 376579bd3712SFilipe Manana * RAID1 filesystem with 2 devices, lose one device and mount the fs 376679bd3712SFilipe Manana * using the other device in degraded mode. If we then allocate a chunk, 376779bd3712SFilipe Manana * we may have enough free space in the existing system space_info, but 376879bd3712SFilipe Manana * none of the block groups can be used for extent allocation since they 376979bd3712SFilipe Manana * have a RAID1 profile, and because we are in degraded mode with a 377079bd3712SFilipe Manana * single device, we are forced to allocate a new system chunk with a 377179bd3712SFilipe Manana * SINGLE profile. Making check_system_chunk() iterate over all system 377279bd3712SFilipe Manana * block groups and check if they have a usable profile and enough space 377379bd3712SFilipe Manana * can be slow on very large filesystems, so we tolerate the -ENOSPC and 377479bd3712SFilipe Manana * try again after forcing allocation of a new system chunk. Like this 377579bd3712SFilipe Manana * we avoid paying the cost of that search in normal circumstances, when 377679bd3712SFilipe Manana * we were not mounted in degraded mode; 377779bd3712SFilipe Manana * 377879bd3712SFilipe Manana * 2) We had enough free space info the system space_info, and one suitable 377979bd3712SFilipe Manana * block group to allocate from when we called check_system_chunk() 378079bd3712SFilipe Manana * above. However right after we called it, the only system block group 378179bd3712SFilipe Manana * with enough free space got turned into RO mode by a running scrub, 378279bd3712SFilipe Manana * and in this case we have to allocate a new one and retry. We only 378379bd3712SFilipe Manana * need do this allocate and retry once, since we have a transaction 3784ecd84d54SFilipe Manana * handle and scrub uses the commit root to search for block groups; 3785ecd84d54SFilipe Manana * 3786ecd84d54SFilipe Manana * 3) We had one system block group with enough free space when we called 3787ecd84d54SFilipe Manana * check_system_chunk(), but after that, right before we tried to 3788ecd84d54SFilipe Manana * allocate the last extent buffer we needed, a discard operation came 3789ecd84d54SFilipe Manana * in and it temporarily removed the last free space entry from the 3790ecd84d54SFilipe Manana * block group (discard removes a free space entry, discards it, and 3791ecd84d54SFilipe Manana * then adds back the entry to the block group cache). 379279bd3712SFilipe Manana */ 379379bd3712SFilipe Manana if (ret == -ENOSPC) { 379479bd3712SFilipe Manana const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info); 379579bd3712SFilipe Manana struct btrfs_block_group *sys_bg; 379679bd3712SFilipe Manana 3797f6f39f7aSNikolay Borisov sys_bg = btrfs_create_chunk(trans, sys_flags); 379879bd3712SFilipe Manana if (IS_ERR(sys_bg)) { 379979bd3712SFilipe Manana ret = PTR_ERR(sys_bg); 380079bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 380179bd3712SFilipe Manana goto out; 380279bd3712SFilipe Manana } 380379bd3712SFilipe Manana 380479bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 380579bd3712SFilipe Manana if (ret) { 380679bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 380779bd3712SFilipe Manana goto out; 380879bd3712SFilipe Manana } 380979bd3712SFilipe Manana 381079bd3712SFilipe Manana ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); 381179bd3712SFilipe Manana if (ret) { 381279bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 381379bd3712SFilipe Manana goto out; 381479bd3712SFilipe Manana } 381579bd3712SFilipe Manana } else if (ret) { 381679bd3712SFilipe Manana btrfs_abort_transaction(trans, ret); 381779bd3712SFilipe Manana goto out; 381879bd3712SFilipe Manana } 381979bd3712SFilipe Manana out: 382079bd3712SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 382179bd3712SFilipe Manana 3822820c363bSNaohiro Aota if (ret) 3823820c363bSNaohiro Aota return ERR_PTR(ret); 3824820c363bSNaohiro Aota 3825820c363bSNaohiro Aota btrfs_get_block_group(bg); 3826820c363bSNaohiro Aota return bg; 382779bd3712SFilipe Manana } 382879bd3712SFilipe Manana 382979bd3712SFilipe Manana /* 383079bd3712SFilipe Manana * Chunk allocation is done in 2 phases: 383179bd3712SFilipe Manana * 383279bd3712SFilipe Manana * 1) Phase 1 - through btrfs_chunk_alloc() we allocate device extents for 383379bd3712SFilipe Manana * the chunk, the chunk mapping, create its block group and add the items 383479bd3712SFilipe Manana * that belong in the chunk btree to it - more specifically, we need to 383579bd3712SFilipe Manana * update device items in the chunk btree and add a new chunk item to it. 383679bd3712SFilipe Manana * 383779bd3712SFilipe Manana * 2) Phase 2 - through btrfs_create_pending_block_groups(), we add the block 383879bd3712SFilipe Manana * group item to the extent btree and the device extent items to the devices 383979bd3712SFilipe Manana * btree. 384079bd3712SFilipe Manana * 384179bd3712SFilipe Manana * This is done to prevent deadlocks. For example when COWing a node from the 384279bd3712SFilipe Manana * extent btree we are holding a write lock on the node's parent and if we 384379bd3712SFilipe Manana * trigger chunk allocation and attempted to insert the new block group item 384479bd3712SFilipe Manana * in the extent btree right way, we could deadlock because the path for the 384579bd3712SFilipe Manana * insertion can include that parent node. At first glance it seems impossible 384679bd3712SFilipe Manana * to trigger chunk allocation after starting a transaction since tasks should 384779bd3712SFilipe Manana * reserve enough transaction units (metadata space), however while that is true 384879bd3712SFilipe Manana * most of the time, chunk allocation may still be triggered for several reasons: 384979bd3712SFilipe Manana * 385079bd3712SFilipe Manana * 1) When reserving metadata, we check if there is enough free space in the 385179bd3712SFilipe Manana * metadata space_info and therefore don't trigger allocation of a new chunk. 385279bd3712SFilipe Manana * However later when the task actually tries to COW an extent buffer from 385379bd3712SFilipe Manana * the extent btree or from the device btree for example, it is forced to 385479bd3712SFilipe Manana * allocate a new block group (chunk) because the only one that had enough 385579bd3712SFilipe Manana * free space was just turned to RO mode by a running scrub for example (or 385679bd3712SFilipe Manana * device replace, block group reclaim thread, etc), so we can not use it 385779bd3712SFilipe Manana * for allocating an extent and end up being forced to allocate a new one; 385879bd3712SFilipe Manana * 385979bd3712SFilipe Manana * 2) Because we only check that the metadata space_info has enough free bytes, 386079bd3712SFilipe Manana * we end up not allocating a new metadata chunk in that case. However if 386179bd3712SFilipe Manana * the filesystem was mounted in degraded mode, none of the existing block 386279bd3712SFilipe Manana * groups might be suitable for extent allocation due to their incompatible 386379bd3712SFilipe Manana * profile (for e.g. mounting a 2 devices filesystem, where all block groups 386479bd3712SFilipe Manana * use a RAID1 profile, in degraded mode using a single device). In this case 386579bd3712SFilipe Manana * when the task attempts to COW some extent buffer of the extent btree for 386679bd3712SFilipe Manana * example, it will trigger allocation of a new metadata block group with a 386779bd3712SFilipe Manana * suitable profile (SINGLE profile in the example of the degraded mount of 386879bd3712SFilipe Manana * the RAID1 filesystem); 386979bd3712SFilipe Manana * 387079bd3712SFilipe Manana * 3) The task has reserved enough transaction units / metadata space, but when 387179bd3712SFilipe Manana * it attempts to COW an extent buffer from the extent or device btree for 387279bd3712SFilipe Manana * example, it does not find any free extent in any metadata block group, 387379bd3712SFilipe Manana * therefore forced to try to allocate a new metadata block group. 387479bd3712SFilipe Manana * This is because some other task allocated all available extents in the 387579bd3712SFilipe Manana * meanwhile - this typically happens with tasks that don't reserve space 387679bd3712SFilipe Manana * properly, either intentionally or as a bug. One example where this is 387779bd3712SFilipe Manana * done intentionally is fsync, as it does not reserve any transaction units 387879bd3712SFilipe Manana * and ends up allocating a variable number of metadata extents for log 3879ecd84d54SFilipe Manana * tree extent buffers; 3880ecd84d54SFilipe Manana * 3881ecd84d54SFilipe Manana * 4) The task has reserved enough transaction units / metadata space, but right 3882ecd84d54SFilipe Manana * before it tries to allocate the last extent buffer it needs, a discard 3883ecd84d54SFilipe Manana * operation comes in and, temporarily, removes the last free space entry from 3884ecd84d54SFilipe Manana * the only metadata block group that had free space (discard starts by 3885ecd84d54SFilipe Manana * removing a free space entry from a block group, then does the discard 3886ecd84d54SFilipe Manana * operation and, once it's done, it adds back the free space entry to the 3887ecd84d54SFilipe Manana * block group). 388879bd3712SFilipe Manana * 388979bd3712SFilipe Manana * We also need this 2 phases setup when adding a device to a filesystem with 389079bd3712SFilipe Manana * a seed device - we must create new metadata and system chunks without adding 389179bd3712SFilipe Manana * any of the block group items to the chunk, extent and device btrees. If we 389279bd3712SFilipe Manana * did not do it this way, we would get ENOSPC when attempting to update those 389379bd3712SFilipe Manana * btrees, since all the chunks from the seed device are read-only. 389479bd3712SFilipe Manana * 389579bd3712SFilipe Manana * Phase 1 does the updates and insertions to the chunk btree because if we had 389679bd3712SFilipe Manana * it done in phase 2 and have a thundering herd of tasks allocating chunks in 389779bd3712SFilipe Manana * parallel, we risk having too many system chunks allocated by many tasks if 389879bd3712SFilipe Manana * many tasks reach phase 1 without the previous ones completing phase 2. In the 389979bd3712SFilipe Manana * extreme case this leads to exhaustion of the system chunk array in the 390079bd3712SFilipe Manana * superblock. This is easier to trigger if using a btree node/leaf size of 64K 390179bd3712SFilipe Manana * and with RAID filesystems (so we have more device items in the chunk btree). 390279bd3712SFilipe Manana * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of 390379bd3712SFilipe Manana * the system chunk array due to concurrent allocations") provides more details. 390479bd3712SFilipe Manana * 39052bb2e00eSFilipe Manana * Allocation of system chunks does not happen through this function. A task that 39062bb2e00eSFilipe Manana * needs to update the chunk btree (the only btree that uses system chunks), must 39072bb2e00eSFilipe Manana * preallocate chunk space by calling either check_system_chunk() or 39082bb2e00eSFilipe Manana * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or 39092bb2e00eSFilipe Manana * metadata chunk or when removing a chunk, while the later is used before doing 39102bb2e00eSFilipe Manana * a modification to the chunk btree - use cases for the later are adding, 39112bb2e00eSFilipe Manana * removing and resizing a device as well as relocation of a system chunk. 39122bb2e00eSFilipe Manana * See the comment below for more details. 391379bd3712SFilipe Manana * 391479bd3712SFilipe Manana * The reservation of system space, done through check_system_chunk(), as well 391579bd3712SFilipe Manana * as all the updates and insertions into the chunk btree must be done while 391679bd3712SFilipe Manana * holding fs_info->chunk_mutex. This is important to guarantee that while COWing 391779bd3712SFilipe Manana * an extent buffer from the chunks btree we never trigger allocation of a new 391879bd3712SFilipe Manana * system chunk, which would result in a deadlock (trying to lock twice an 391979bd3712SFilipe Manana * extent buffer of the chunk btree, first time before triggering the chunk 392079bd3712SFilipe Manana * allocation and the second time during chunk allocation while attempting to 392179bd3712SFilipe Manana * update the chunks btree). The system chunk array is also updated while holding 392279bd3712SFilipe Manana * that mutex. The same logic applies to removing chunks - we must reserve system 392379bd3712SFilipe Manana * space, update the chunk btree and the system chunk array in the superblock 392479bd3712SFilipe Manana * while holding fs_info->chunk_mutex. 392579bd3712SFilipe Manana * 392679bd3712SFilipe Manana * This function, btrfs_chunk_alloc(), belongs to phase 1. 392779bd3712SFilipe Manana * 392879bd3712SFilipe Manana * If @force is CHUNK_ALLOC_FORCE: 392907730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 393007730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 393179bd3712SFilipe Manana * If @force is NOT CHUNK_ALLOC_FORCE: 393207730d87SJosef Bacik * - return 0 if it doesn't need to allocate a new chunk, 393307730d87SJosef Bacik * - return 1 if it successfully allocates a chunk, 393407730d87SJosef Bacik * - return errors including -ENOSPC otherwise. 393507730d87SJosef Bacik */ 393607730d87SJosef Bacik int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, 393707730d87SJosef Bacik enum btrfs_chunk_alloc_enum force) 393807730d87SJosef Bacik { 393907730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 394007730d87SJosef Bacik struct btrfs_space_info *space_info; 3941820c363bSNaohiro Aota struct btrfs_block_group *ret_bg; 394207730d87SJosef Bacik bool wait_for_alloc = false; 394307730d87SJosef Bacik bool should_alloc = false; 3944760e69c4SNaohiro Aota bool from_extent_allocation = false; 394507730d87SJosef Bacik int ret = 0; 394607730d87SJosef Bacik 3947760e69c4SNaohiro Aota if (force == CHUNK_ALLOC_FORCE_FOR_EXTENT) { 3948760e69c4SNaohiro Aota from_extent_allocation = true; 3949760e69c4SNaohiro Aota force = CHUNK_ALLOC_FORCE; 3950760e69c4SNaohiro Aota } 3951760e69c4SNaohiro Aota 395207730d87SJosef Bacik /* Don't re-enter if we're already allocating a chunk */ 395307730d87SJosef Bacik if (trans->allocating_chunk) 395407730d87SJosef Bacik return -ENOSPC; 395579bd3712SFilipe Manana /* 39562bb2e00eSFilipe Manana * Allocation of system chunks can not happen through this path, as we 39572bb2e00eSFilipe Manana * could end up in a deadlock if we are allocating a data or metadata 39582bb2e00eSFilipe Manana * chunk and there is another task modifying the chunk btree. 39592bb2e00eSFilipe Manana * 39602bb2e00eSFilipe Manana * This is because while we are holding the chunk mutex, we will attempt 39612bb2e00eSFilipe Manana * to add the new chunk item to the chunk btree or update an existing 39622bb2e00eSFilipe Manana * device item in the chunk btree, while the other task that is modifying 39632bb2e00eSFilipe Manana * the chunk btree is attempting to COW an extent buffer while holding a 39642bb2e00eSFilipe Manana * lock on it and on its parent - if the COW operation triggers a system 39652bb2e00eSFilipe Manana * chunk allocation, then we can deadlock because we are holding the 39662bb2e00eSFilipe Manana * chunk mutex and we may need to access that extent buffer or its parent 39672bb2e00eSFilipe Manana * in order to add the chunk item or update a device item. 39682bb2e00eSFilipe Manana * 39692bb2e00eSFilipe Manana * Tasks that want to modify the chunk tree should reserve system space 39702bb2e00eSFilipe Manana * before updating the chunk btree, by calling either 39712bb2e00eSFilipe Manana * btrfs_reserve_chunk_metadata() or check_system_chunk(). 39722bb2e00eSFilipe Manana * It's possible that after a task reserves the space, it still ends up 39732bb2e00eSFilipe Manana * here - this happens in the cases described above at do_chunk_alloc(). 39742bb2e00eSFilipe Manana * The task will have to either retry or fail. 397579bd3712SFilipe Manana */ 39762bb2e00eSFilipe Manana if (flags & BTRFS_BLOCK_GROUP_SYSTEM) 397779bd3712SFilipe Manana return -ENOSPC; 397807730d87SJosef Bacik 397907730d87SJosef Bacik space_info = btrfs_find_space_info(fs_info, flags); 398007730d87SJosef Bacik ASSERT(space_info); 398107730d87SJosef Bacik 398207730d87SJosef Bacik do { 398307730d87SJosef Bacik spin_lock(&space_info->lock); 398407730d87SJosef Bacik if (force < space_info->force_alloc) 398507730d87SJosef Bacik force = space_info->force_alloc; 398607730d87SJosef Bacik should_alloc = should_alloc_chunk(fs_info, space_info, force); 398707730d87SJosef Bacik if (space_info->full) { 398807730d87SJosef Bacik /* No more free physical space */ 398907730d87SJosef Bacik if (should_alloc) 399007730d87SJosef Bacik ret = -ENOSPC; 399107730d87SJosef Bacik else 399207730d87SJosef Bacik ret = 0; 399307730d87SJosef Bacik spin_unlock(&space_info->lock); 399407730d87SJosef Bacik return ret; 399507730d87SJosef Bacik } else if (!should_alloc) { 399607730d87SJosef Bacik spin_unlock(&space_info->lock); 399707730d87SJosef Bacik return 0; 399807730d87SJosef Bacik } else if (space_info->chunk_alloc) { 399907730d87SJosef Bacik /* 400007730d87SJosef Bacik * Someone is already allocating, so we need to block 400107730d87SJosef Bacik * until this someone is finished and then loop to 400207730d87SJosef Bacik * recheck if we should continue with our allocation 400307730d87SJosef Bacik * attempt. 400407730d87SJosef Bacik */ 400507730d87SJosef Bacik wait_for_alloc = true; 40061314ca78SJosef Bacik force = CHUNK_ALLOC_NO_FORCE; 400707730d87SJosef Bacik spin_unlock(&space_info->lock); 400807730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 400907730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 401007730d87SJosef Bacik } else { 401107730d87SJosef Bacik /* Proceed with allocation */ 401207730d87SJosef Bacik space_info->chunk_alloc = 1; 401307730d87SJosef Bacik wait_for_alloc = false; 401407730d87SJosef Bacik spin_unlock(&space_info->lock); 401507730d87SJosef Bacik } 401607730d87SJosef Bacik 401707730d87SJosef Bacik cond_resched(); 401807730d87SJosef Bacik } while (wait_for_alloc); 401907730d87SJosef Bacik 402007730d87SJosef Bacik mutex_lock(&fs_info->chunk_mutex); 402107730d87SJosef Bacik trans->allocating_chunk = true; 402207730d87SJosef Bacik 402307730d87SJosef Bacik /* 402407730d87SJosef Bacik * If we have mixed data/metadata chunks we want to make sure we keep 402507730d87SJosef Bacik * allocating mixed chunks instead of individual chunks. 402607730d87SJosef Bacik */ 402707730d87SJosef Bacik if (btrfs_mixed_space_info(space_info)) 402807730d87SJosef Bacik flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); 402907730d87SJosef Bacik 403007730d87SJosef Bacik /* 403107730d87SJosef Bacik * if we're doing a data chunk, go ahead and make sure that 403207730d87SJosef Bacik * we keep a reasonable number of metadata chunks allocated in the 403307730d87SJosef Bacik * FS as well. 403407730d87SJosef Bacik */ 403507730d87SJosef Bacik if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { 403607730d87SJosef Bacik fs_info->data_chunk_allocations++; 403707730d87SJosef Bacik if (!(fs_info->data_chunk_allocations % 403807730d87SJosef Bacik fs_info->metadata_ratio)) 403907730d87SJosef Bacik force_metadata_allocation(fs_info); 404007730d87SJosef Bacik } 404107730d87SJosef Bacik 4042820c363bSNaohiro Aota ret_bg = do_chunk_alloc(trans, flags); 404307730d87SJosef Bacik trans->allocating_chunk = false; 404407730d87SJosef Bacik 4045760e69c4SNaohiro Aota if (IS_ERR(ret_bg)) { 4046820c363bSNaohiro Aota ret = PTR_ERR(ret_bg); 4047760e69c4SNaohiro Aota } else if (from_extent_allocation) { 4048760e69c4SNaohiro Aota /* 4049760e69c4SNaohiro Aota * New block group is likely to be used soon. Try to activate 4050760e69c4SNaohiro Aota * it now. Failure is OK for now. 4051760e69c4SNaohiro Aota */ 4052760e69c4SNaohiro Aota btrfs_zone_activate(ret_bg); 4053760e69c4SNaohiro Aota } 4054760e69c4SNaohiro Aota 4055760e69c4SNaohiro Aota if (!ret) 4056820c363bSNaohiro Aota btrfs_put_block_group(ret_bg); 4057820c363bSNaohiro Aota 405807730d87SJosef Bacik spin_lock(&space_info->lock); 405907730d87SJosef Bacik if (ret < 0) { 406007730d87SJosef Bacik if (ret == -ENOSPC) 406107730d87SJosef Bacik space_info->full = 1; 406207730d87SJosef Bacik else 406307730d87SJosef Bacik goto out; 406407730d87SJosef Bacik } else { 406507730d87SJosef Bacik ret = 1; 406607730d87SJosef Bacik space_info->max_extent_size = 0; 406707730d87SJosef Bacik } 406807730d87SJosef Bacik 406907730d87SJosef Bacik space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; 407007730d87SJosef Bacik out: 407107730d87SJosef Bacik space_info->chunk_alloc = 0; 407207730d87SJosef Bacik spin_unlock(&space_info->lock); 407307730d87SJosef Bacik mutex_unlock(&fs_info->chunk_mutex); 407407730d87SJosef Bacik 407507730d87SJosef Bacik return ret; 407607730d87SJosef Bacik } 407707730d87SJosef Bacik 407807730d87SJosef Bacik static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) 407907730d87SJosef Bacik { 408007730d87SJosef Bacik u64 num_dev; 408107730d87SJosef Bacik 408207730d87SJosef Bacik num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; 408307730d87SJosef Bacik if (!num_dev) 408407730d87SJosef Bacik num_dev = fs_info->fs_devices->rw_devices; 408507730d87SJosef Bacik 408607730d87SJosef Bacik return num_dev; 408707730d87SJosef Bacik } 408807730d87SJosef Bacik 40892bb2e00eSFilipe Manana static void reserve_chunk_space(struct btrfs_trans_handle *trans, 40902bb2e00eSFilipe Manana u64 bytes, 40912bb2e00eSFilipe Manana u64 type) 409207730d87SJosef Bacik { 409307730d87SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 409407730d87SJosef Bacik struct btrfs_space_info *info; 409507730d87SJosef Bacik u64 left; 409607730d87SJosef Bacik int ret = 0; 409707730d87SJosef Bacik 409807730d87SJosef Bacik /* 409907730d87SJosef Bacik * Needed because we can end up allocating a system chunk and for an 410007730d87SJosef Bacik * atomic and race free space reservation in the chunk block reserve. 410107730d87SJosef Bacik */ 410207730d87SJosef Bacik lockdep_assert_held(&fs_info->chunk_mutex); 410307730d87SJosef Bacik 410407730d87SJosef Bacik info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); 410507730d87SJosef Bacik spin_lock(&info->lock); 410607730d87SJosef Bacik left = info->total_bytes - btrfs_space_info_used(info, true); 410707730d87SJosef Bacik spin_unlock(&info->lock); 410807730d87SJosef Bacik 41092bb2e00eSFilipe Manana if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { 411007730d87SJosef Bacik btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", 41112bb2e00eSFilipe Manana left, bytes, type); 411207730d87SJosef Bacik btrfs_dump_space_info(fs_info, info, 0, 0); 411307730d87SJosef Bacik } 411407730d87SJosef Bacik 41152bb2e00eSFilipe Manana if (left < bytes) { 411607730d87SJosef Bacik u64 flags = btrfs_system_alloc_profile(fs_info); 411779bd3712SFilipe Manana struct btrfs_block_group *bg; 411807730d87SJosef Bacik 411907730d87SJosef Bacik /* 412007730d87SJosef Bacik * Ignore failure to create system chunk. We might end up not 412107730d87SJosef Bacik * needing it, as we might not need to COW all nodes/leafs from 412207730d87SJosef Bacik * the paths we visit in the chunk tree (they were already COWed 412307730d87SJosef Bacik * or created in the current transaction for example). 412407730d87SJosef Bacik */ 4125f6f39f7aSNikolay Borisov bg = btrfs_create_chunk(trans, flags); 412679bd3712SFilipe Manana if (IS_ERR(bg)) { 412779bd3712SFilipe Manana ret = PTR_ERR(bg); 41282bb2e00eSFilipe Manana } else { 412979bd3712SFilipe Manana /* 4130b6a98021SNaohiro Aota * We have a new chunk. We also need to activate it for 4131b6a98021SNaohiro Aota * zoned filesystem. 4132b6a98021SNaohiro Aota */ 4133b6a98021SNaohiro Aota ret = btrfs_zoned_activate_one_bg(fs_info, info, true); 4134b6a98021SNaohiro Aota if (ret < 0) 4135b6a98021SNaohiro Aota return; 4136b6a98021SNaohiro Aota 4137b6a98021SNaohiro Aota /* 413879bd3712SFilipe Manana * If we fail to add the chunk item here, we end up 413979bd3712SFilipe Manana * trying again at phase 2 of chunk allocation, at 414079bd3712SFilipe Manana * btrfs_create_pending_block_groups(). So ignore 41412bb2e00eSFilipe Manana * any error here. An ENOSPC here could happen, due to 41422bb2e00eSFilipe Manana * the cases described at do_chunk_alloc() - the system 41432bb2e00eSFilipe Manana * block group we just created was just turned into RO 41442bb2e00eSFilipe Manana * mode by a scrub for example, or a running discard 41452bb2e00eSFilipe Manana * temporarily removed its free space entries, etc. 414679bd3712SFilipe Manana */ 414779bd3712SFilipe Manana btrfs_chunk_alloc_add_chunk_item(trans, bg); 414879bd3712SFilipe Manana } 414907730d87SJosef Bacik } 415007730d87SJosef Bacik 415107730d87SJosef Bacik if (!ret) { 41529270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, 415307730d87SJosef Bacik &fs_info->chunk_block_rsv, 41542bb2e00eSFilipe Manana bytes, BTRFS_RESERVE_NO_FLUSH); 41551cb3db1cSFilipe Manana if (!ret) 41562bb2e00eSFilipe Manana trans->chunk_bytes_reserved += bytes; 415707730d87SJosef Bacik } 415807730d87SJosef Bacik } 415907730d87SJosef Bacik 41602bb2e00eSFilipe Manana /* 41612bb2e00eSFilipe Manana * Reserve space in the system space for allocating or removing a chunk. 41622bb2e00eSFilipe Manana * The caller must be holding fs_info->chunk_mutex. 41632bb2e00eSFilipe Manana */ 41642bb2e00eSFilipe Manana void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) 41652bb2e00eSFilipe Manana { 41662bb2e00eSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 41672bb2e00eSFilipe Manana const u64 num_devs = get_profile_num_devs(fs_info, type); 41682bb2e00eSFilipe Manana u64 bytes; 41692bb2e00eSFilipe Manana 41702bb2e00eSFilipe Manana /* num_devs device items to update and 1 chunk item to add or remove. */ 41712bb2e00eSFilipe Manana bytes = btrfs_calc_metadata_size(fs_info, num_devs) + 41722bb2e00eSFilipe Manana btrfs_calc_insert_metadata_size(fs_info, 1); 41732bb2e00eSFilipe Manana 41742bb2e00eSFilipe Manana reserve_chunk_space(trans, bytes, type); 41752bb2e00eSFilipe Manana } 41762bb2e00eSFilipe Manana 41772bb2e00eSFilipe Manana /* 41782bb2e00eSFilipe Manana * Reserve space in the system space, if needed, for doing a modification to the 41792bb2e00eSFilipe Manana * chunk btree. 41802bb2e00eSFilipe Manana * 41812bb2e00eSFilipe Manana * @trans: A transaction handle. 41822bb2e00eSFilipe Manana * @is_item_insertion: Indicate if the modification is for inserting a new item 41832bb2e00eSFilipe Manana * in the chunk btree or if it's for the deletion or update 41842bb2e00eSFilipe Manana * of an existing item. 41852bb2e00eSFilipe Manana * 41862bb2e00eSFilipe Manana * This is used in a context where we need to update the chunk btree outside 41872bb2e00eSFilipe Manana * block group allocation and removal, to avoid a deadlock with a concurrent 41882bb2e00eSFilipe Manana * task that is allocating a metadata or data block group and therefore needs to 41892bb2e00eSFilipe Manana * update the chunk btree while holding the chunk mutex. After the update to the 41902bb2e00eSFilipe Manana * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called. 41912bb2e00eSFilipe Manana * 41922bb2e00eSFilipe Manana */ 41932bb2e00eSFilipe Manana void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans, 41942bb2e00eSFilipe Manana bool is_item_insertion) 41952bb2e00eSFilipe Manana { 41962bb2e00eSFilipe Manana struct btrfs_fs_info *fs_info = trans->fs_info; 41972bb2e00eSFilipe Manana u64 bytes; 41982bb2e00eSFilipe Manana 41992bb2e00eSFilipe Manana if (is_item_insertion) 42002bb2e00eSFilipe Manana bytes = btrfs_calc_insert_metadata_size(fs_info, 1); 42012bb2e00eSFilipe Manana else 42022bb2e00eSFilipe Manana bytes = btrfs_calc_metadata_size(fs_info, 1); 42032bb2e00eSFilipe Manana 42042bb2e00eSFilipe Manana mutex_lock(&fs_info->chunk_mutex); 42052bb2e00eSFilipe Manana reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM); 42062bb2e00eSFilipe Manana mutex_unlock(&fs_info->chunk_mutex); 42072bb2e00eSFilipe Manana } 42082bb2e00eSFilipe Manana 42093e43c279SJosef Bacik void btrfs_put_block_group_cache(struct btrfs_fs_info *info) 42103e43c279SJosef Bacik { 421132da5386SDavid Sterba struct btrfs_block_group *block_group; 42123e43c279SJosef Bacik 421350c31eaaSJosef Bacik block_group = btrfs_lookup_first_block_group(info, 0); 42143e43c279SJosef Bacik while (block_group) { 42153e43c279SJosef Bacik btrfs_wait_block_group_cache_done(block_group); 42163e43c279SJosef Bacik spin_lock(&block_group->lock); 421750c31eaaSJosef Bacik if (test_and_clear_bit(BLOCK_GROUP_FLAG_IREF, 421850c31eaaSJosef Bacik &block_group->runtime_flags)) { 421950c31eaaSJosef Bacik struct inode *inode = block_group->inode; 42203e43c279SJosef Bacik 42213e43c279SJosef Bacik block_group->inode = NULL; 42223e43c279SJosef Bacik spin_unlock(&block_group->lock); 422350c31eaaSJosef Bacik 42243e43c279SJosef Bacik ASSERT(block_group->io_ctl.inode == NULL); 42253e43c279SJosef Bacik iput(inode); 422650c31eaaSJosef Bacik } else { 422750c31eaaSJosef Bacik spin_unlock(&block_group->lock); 422850c31eaaSJosef Bacik } 422950c31eaaSJosef Bacik block_group = btrfs_next_block_group(block_group); 42303e43c279SJosef Bacik } 42313e43c279SJosef Bacik } 42323e43c279SJosef Bacik 42333e43c279SJosef Bacik /* 42343e43c279SJosef Bacik * Must be called only after stopping all workers, since we could have block 42353e43c279SJosef Bacik * group caching kthreads running, and therefore they could race with us if we 42363e43c279SJosef Bacik * freed the block groups before stopping them. 42373e43c279SJosef Bacik */ 42383e43c279SJosef Bacik int btrfs_free_block_groups(struct btrfs_fs_info *info) 42393e43c279SJosef Bacik { 424032da5386SDavid Sterba struct btrfs_block_group *block_group; 42413e43c279SJosef Bacik struct btrfs_space_info *space_info; 42423e43c279SJosef Bacik struct btrfs_caching_control *caching_ctl; 42433e43c279SJosef Bacik struct rb_node *n; 42443e43c279SJosef Bacik 424516b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 42463e43c279SJosef Bacik while (!list_empty(&info->caching_block_groups)) { 42473e43c279SJosef Bacik caching_ctl = list_entry(info->caching_block_groups.next, 42483e43c279SJosef Bacik struct btrfs_caching_control, list); 42493e43c279SJosef Bacik list_del(&caching_ctl->list); 42503e43c279SJosef Bacik btrfs_put_caching_control(caching_ctl); 42513e43c279SJosef Bacik } 425216b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42533e43c279SJosef Bacik 42543e43c279SJosef Bacik spin_lock(&info->unused_bgs_lock); 42553e43c279SJosef Bacik while (!list_empty(&info->unused_bgs)) { 42563e43c279SJosef Bacik block_group = list_first_entry(&info->unused_bgs, 425732da5386SDavid Sterba struct btrfs_block_group, 42583e43c279SJosef Bacik bg_list); 42593e43c279SJosef Bacik list_del_init(&block_group->bg_list); 42603e43c279SJosef Bacik btrfs_put_block_group(block_group); 42613e43c279SJosef Bacik } 42623e43c279SJosef Bacik 426318bb8bbfSJohannes Thumshirn while (!list_empty(&info->reclaim_bgs)) { 426418bb8bbfSJohannes Thumshirn block_group = list_first_entry(&info->reclaim_bgs, 426518bb8bbfSJohannes Thumshirn struct btrfs_block_group, 426618bb8bbfSJohannes Thumshirn bg_list); 426718bb8bbfSJohannes Thumshirn list_del_init(&block_group->bg_list); 426818bb8bbfSJohannes Thumshirn btrfs_put_block_group(block_group); 426918bb8bbfSJohannes Thumshirn } 427018bb8bbfSJohannes Thumshirn spin_unlock(&info->unused_bgs_lock); 427118bb8bbfSJohannes Thumshirn 4272afba2bc0SNaohiro Aota spin_lock(&info->zone_active_bgs_lock); 4273afba2bc0SNaohiro Aota while (!list_empty(&info->zone_active_bgs)) { 4274afba2bc0SNaohiro Aota block_group = list_first_entry(&info->zone_active_bgs, 4275afba2bc0SNaohiro Aota struct btrfs_block_group, 4276afba2bc0SNaohiro Aota active_bg_list); 4277afba2bc0SNaohiro Aota list_del_init(&block_group->active_bg_list); 4278afba2bc0SNaohiro Aota btrfs_put_block_group(block_group); 4279afba2bc0SNaohiro Aota } 4280afba2bc0SNaohiro Aota spin_unlock(&info->zone_active_bgs_lock); 4281afba2bc0SNaohiro Aota 428216b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 428308dddb29SFilipe Manana while ((n = rb_last(&info->block_group_cache_tree.rb_root)) != NULL) { 428432da5386SDavid Sterba block_group = rb_entry(n, struct btrfs_block_group, 42853e43c279SJosef Bacik cache_node); 428608dddb29SFilipe Manana rb_erase_cached(&block_group->cache_node, 42873e43c279SJosef Bacik &info->block_group_cache_tree); 42883e43c279SJosef Bacik RB_CLEAR_NODE(&block_group->cache_node); 428916b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 42903e43c279SJosef Bacik 42913e43c279SJosef Bacik down_write(&block_group->space_info->groups_sem); 42923e43c279SJosef Bacik list_del(&block_group->list); 42933e43c279SJosef Bacik up_write(&block_group->space_info->groups_sem); 42943e43c279SJosef Bacik 42953e43c279SJosef Bacik /* 42963e43c279SJosef Bacik * We haven't cached this block group, which means we could 42973e43c279SJosef Bacik * possibly have excluded extents on this block group. 42983e43c279SJosef Bacik */ 42993e43c279SJosef Bacik if (block_group->cached == BTRFS_CACHE_NO || 43003e43c279SJosef Bacik block_group->cached == BTRFS_CACHE_ERROR) 43013e43c279SJosef Bacik btrfs_free_excluded_extents(block_group); 43023e43c279SJosef Bacik 43033e43c279SJosef Bacik btrfs_remove_free_space_cache(block_group); 43043e43c279SJosef Bacik ASSERT(block_group->cached != BTRFS_CACHE_STARTED); 43053e43c279SJosef Bacik ASSERT(list_empty(&block_group->dirty_list)); 43063e43c279SJosef Bacik ASSERT(list_empty(&block_group->io_list)); 43073e43c279SJosef Bacik ASSERT(list_empty(&block_group->bg_list)); 430848aaeebeSJosef Bacik ASSERT(refcount_read(&block_group->refs) == 1); 4309195a49eaSFilipe Manana ASSERT(block_group->swap_extents == 0); 43103e43c279SJosef Bacik btrfs_put_block_group(block_group); 43113e43c279SJosef Bacik 431216b0c258SFilipe Manana write_lock(&info->block_group_cache_lock); 43133e43c279SJosef Bacik } 431416b0c258SFilipe Manana write_unlock(&info->block_group_cache_lock); 43153e43c279SJosef Bacik 43163e43c279SJosef Bacik btrfs_release_global_block_rsv(info); 43173e43c279SJosef Bacik 43183e43c279SJosef Bacik while (!list_empty(&info->space_info)) { 43193e43c279SJosef Bacik space_info = list_entry(info->space_info.next, 43203e43c279SJosef Bacik struct btrfs_space_info, 43213e43c279SJosef Bacik list); 43223e43c279SJosef Bacik 43233e43c279SJosef Bacik /* 43243e43c279SJosef Bacik * Do not hide this behind enospc_debug, this is actually 43253e43c279SJosef Bacik * important and indicates a real bug if this happens. 43263e43c279SJosef Bacik */ 43273e43c279SJosef Bacik if (WARN_ON(space_info->bytes_pinned > 0 || 43283e43c279SJosef Bacik space_info->bytes_may_use > 0)) 43293e43c279SJosef Bacik btrfs_dump_space_info(info, space_info, 0, 0); 433040cdc509SFilipe Manana 433140cdc509SFilipe Manana /* 433240cdc509SFilipe Manana * If there was a failure to cleanup a log tree, very likely due 433340cdc509SFilipe Manana * to an IO failure on a writeback attempt of one or more of its 433440cdc509SFilipe Manana * extent buffers, we could not do proper (and cheap) unaccounting 433540cdc509SFilipe Manana * of their reserved space, so don't warn on bytes_reserved > 0 in 433640cdc509SFilipe Manana * that case. 433740cdc509SFilipe Manana */ 433840cdc509SFilipe Manana if (!(space_info->flags & BTRFS_BLOCK_GROUP_METADATA) || 433940cdc509SFilipe Manana !BTRFS_FS_LOG_CLEANUP_ERROR(info)) { 434040cdc509SFilipe Manana if (WARN_ON(space_info->bytes_reserved > 0)) 434140cdc509SFilipe Manana btrfs_dump_space_info(info, space_info, 0, 0); 434240cdc509SFilipe Manana } 434340cdc509SFilipe Manana 4344d611add4SFilipe Manana WARN_ON(space_info->reclaim_size > 0); 43453e43c279SJosef Bacik list_del(&space_info->list); 43463e43c279SJosef Bacik btrfs_sysfs_remove_space_info(space_info); 43473e43c279SJosef Bacik } 43483e43c279SJosef Bacik return 0; 43493e43c279SJosef Bacik } 4350684b752bSFilipe Manana 4351684b752bSFilipe Manana void btrfs_freeze_block_group(struct btrfs_block_group *cache) 4352684b752bSFilipe Manana { 4353684b752bSFilipe Manana atomic_inc(&cache->frozen); 4354684b752bSFilipe Manana } 4355684b752bSFilipe Manana 4356684b752bSFilipe Manana void btrfs_unfreeze_block_group(struct btrfs_block_group *block_group) 4357684b752bSFilipe Manana { 4358684b752bSFilipe Manana struct btrfs_fs_info *fs_info = block_group->fs_info; 4359684b752bSFilipe Manana struct extent_map_tree *em_tree; 4360684b752bSFilipe Manana struct extent_map *em; 4361684b752bSFilipe Manana bool cleanup; 4362684b752bSFilipe Manana 4363684b752bSFilipe Manana spin_lock(&block_group->lock); 4364684b752bSFilipe Manana cleanup = (atomic_dec_and_test(&block_group->frozen) && 43653349b57fSJosef Bacik test_bit(BLOCK_GROUP_FLAG_REMOVED, &block_group->runtime_flags)); 4366684b752bSFilipe Manana spin_unlock(&block_group->lock); 4367684b752bSFilipe Manana 4368684b752bSFilipe Manana if (cleanup) { 4369684b752bSFilipe Manana em_tree = &fs_info->mapping_tree; 4370684b752bSFilipe Manana write_lock(&em_tree->lock); 4371684b752bSFilipe Manana em = lookup_extent_mapping(em_tree, block_group->start, 4372684b752bSFilipe Manana 1); 4373684b752bSFilipe Manana BUG_ON(!em); /* logic error, can't happen */ 4374684b752bSFilipe Manana remove_extent_mapping(em_tree, em); 4375684b752bSFilipe Manana write_unlock(&em_tree->lock); 4376684b752bSFilipe Manana 4377684b752bSFilipe Manana /* once for us and once for the tree */ 4378684b752bSFilipe Manana free_extent_map(em); 4379684b752bSFilipe Manana free_extent_map(em); 4380684b752bSFilipe Manana 4381684b752bSFilipe Manana /* 4382684b752bSFilipe Manana * We may have left one free space entry and other possible 4383684b752bSFilipe Manana * tasks trimming this block group have left 1 entry each one. 4384684b752bSFilipe Manana * Free them if any. 4385684b752bSFilipe Manana */ 4386fc80f7acSJosef Bacik btrfs_remove_free_space_cache(block_group); 4387684b752bSFilipe Manana } 4388684b752bSFilipe Manana } 4389195a49eaSFilipe Manana 4390195a49eaSFilipe Manana bool btrfs_inc_block_group_swap_extents(struct btrfs_block_group *bg) 4391195a49eaSFilipe Manana { 4392195a49eaSFilipe Manana bool ret = true; 4393195a49eaSFilipe Manana 4394195a49eaSFilipe Manana spin_lock(&bg->lock); 4395195a49eaSFilipe Manana if (bg->ro) 4396195a49eaSFilipe Manana ret = false; 4397195a49eaSFilipe Manana else 4398195a49eaSFilipe Manana bg->swap_extents++; 4399195a49eaSFilipe Manana spin_unlock(&bg->lock); 4400195a49eaSFilipe Manana 4401195a49eaSFilipe Manana return ret; 4402195a49eaSFilipe Manana } 4403195a49eaSFilipe Manana 4404195a49eaSFilipe Manana void btrfs_dec_block_group_swap_extents(struct btrfs_block_group *bg, int amount) 4405195a49eaSFilipe Manana { 4406195a49eaSFilipe Manana spin_lock(&bg->lock); 4407195a49eaSFilipe Manana ASSERT(!bg->ro); 4408195a49eaSFilipe Manana ASSERT(bg->swap_extents >= amount); 4409195a49eaSFilipe Manana bg->swap_extents -= amount; 4410195a49eaSFilipe Manana spin_unlock(&bg->lock); 4411195a49eaSFilipe Manana } 441252bb7a21SBoris Burkov 441352bb7a21SBoris Burkov enum btrfs_block_group_size_class btrfs_calc_block_group_size_class(u64 size) 441452bb7a21SBoris Burkov { 441552bb7a21SBoris Burkov if (size <= SZ_128K) 441652bb7a21SBoris Burkov return BTRFS_BG_SZ_SMALL; 441752bb7a21SBoris Burkov if (size <= SZ_8M) 441852bb7a21SBoris Burkov return BTRFS_BG_SZ_MEDIUM; 441952bb7a21SBoris Burkov return BTRFS_BG_SZ_LARGE; 442052bb7a21SBoris Burkov } 442152bb7a21SBoris Burkov 442252bb7a21SBoris Burkov /* 442352bb7a21SBoris Burkov * Handle a block group allocating an extent in a size class 442452bb7a21SBoris Burkov * 442552bb7a21SBoris Burkov * @bg: The block group we allocated in. 442652bb7a21SBoris Burkov * @size_class: The size class of the allocation. 442752bb7a21SBoris Burkov * @force_wrong_size_class: Whether we are desperate enough to allow 442852bb7a21SBoris Burkov * mismatched size classes. 442952bb7a21SBoris Burkov * 443052bb7a21SBoris Burkov * Returns: 0 if the size class was valid for this block_group, -EAGAIN in the 443152bb7a21SBoris Burkov * case of a race that leads to the wrong size class without 443252bb7a21SBoris Burkov * force_wrong_size_class set. 443352bb7a21SBoris Burkov * 443452bb7a21SBoris Burkov * find_free_extent will skip block groups with a mismatched size class until 443552bb7a21SBoris Burkov * it really needs to avoid ENOSPC. In that case it will set 443652bb7a21SBoris Burkov * force_wrong_size_class. However, if a block group is newly allocated and 443752bb7a21SBoris Burkov * doesn't yet have a size class, then it is possible for two allocations of 443852bb7a21SBoris Burkov * different sizes to race and both try to use it. The loser is caught here and 443952bb7a21SBoris Burkov * has to retry. 444052bb7a21SBoris Burkov */ 444152bb7a21SBoris Burkov int btrfs_use_block_group_size_class(struct btrfs_block_group *bg, 444252bb7a21SBoris Burkov enum btrfs_block_group_size_class size_class, 444352bb7a21SBoris Burkov bool force_wrong_size_class) 444452bb7a21SBoris Burkov { 444552bb7a21SBoris Burkov ASSERT(size_class != BTRFS_BG_SZ_NONE); 444652bb7a21SBoris Burkov 444752bb7a21SBoris Burkov /* The new allocation is in the right size class, do nothing */ 444852bb7a21SBoris Burkov if (bg->size_class == size_class) 444952bb7a21SBoris Burkov return 0; 445052bb7a21SBoris Burkov /* 445152bb7a21SBoris Burkov * The new allocation is in a mismatched size class. 445252bb7a21SBoris Burkov * This means one of two things: 445352bb7a21SBoris Burkov * 445452bb7a21SBoris Burkov * 1. Two tasks in find_free_extent for different size_classes raced 445552bb7a21SBoris Burkov * and hit the same empty block_group. Make the loser try again. 445652bb7a21SBoris Burkov * 2. A call to find_free_extent got desperate enough to set 445752bb7a21SBoris Burkov * 'force_wrong_slab'. Don't change the size_class, but allow the 445852bb7a21SBoris Burkov * allocation. 445952bb7a21SBoris Burkov */ 446052bb7a21SBoris Burkov if (bg->size_class != BTRFS_BG_SZ_NONE) { 446152bb7a21SBoris Burkov if (force_wrong_size_class) 446252bb7a21SBoris Burkov return 0; 446352bb7a21SBoris Burkov return -EAGAIN; 446452bb7a21SBoris Burkov } 446552bb7a21SBoris Burkov /* 446652bb7a21SBoris Burkov * The happy new block group case: the new allocation is the first 446752bb7a21SBoris Burkov * one in the block_group so we set size_class. 446852bb7a21SBoris Burkov */ 446952bb7a21SBoris Burkov bg->size_class = size_class; 447052bb7a21SBoris Burkov 447152bb7a21SBoris Burkov return 0; 447252bb7a21SBoris Burkov } 4473cb0922f2SBoris Burkov 4474cb0922f2SBoris Burkov bool btrfs_block_group_should_use_size_class(struct btrfs_block_group *bg) 4475cb0922f2SBoris Burkov { 4476cb0922f2SBoris Burkov if (btrfs_is_zoned(bg->fs_info)) 4477cb0922f2SBoris Burkov return false; 4478cb0922f2SBoris Burkov if (!btrfs_is_block_group_data_only(bg)) 4479cb0922f2SBoris Burkov return false; 4480cb0922f2SBoris Burkov return true; 4481cb0922f2SBoris Burkov } 4482