186736342SJosef Bacik // SPDX-License-Identifier: GPL-2.0 286736342SJosef Bacik 386736342SJosef Bacik #include "ctree.h" 486736342SJosef Bacik #include "delalloc-space.h" 586736342SJosef Bacik #include "block-rsv.h" 686736342SJosef Bacik #include "btrfs_inode.h" 786736342SJosef Bacik #include "space-info.h" 886736342SJosef Bacik #include "transaction.h" 986736342SJosef Bacik #include "qgroup.h" 1007730d87SJosef Bacik #include "block-group.h" 1186736342SJosef Bacik 126f4ad559SJosef Bacik /* 136f4ad559SJosef Bacik * HOW DOES THIS WORK 146f4ad559SJosef Bacik * 156f4ad559SJosef Bacik * There are two stages to data reservations, one for data and one for metadata 166f4ad559SJosef Bacik * to handle the new extents and checksums generated by writing data. 176f4ad559SJosef Bacik * 186f4ad559SJosef Bacik * 196f4ad559SJosef Bacik * DATA RESERVATION 206f4ad559SJosef Bacik * The general flow of the data reservation is as follows 216f4ad559SJosef Bacik * 226f4ad559SJosef Bacik * -> Reserve 236f4ad559SJosef Bacik * We call into btrfs_reserve_data_bytes() for the user request bytes that 246f4ad559SJosef Bacik * they wish to write. We make this reservation and add it to 256f4ad559SJosef Bacik * space_info->bytes_may_use. We set EXTENT_DELALLOC on the inode io_tree 266f4ad559SJosef Bacik * for the range and carry on if this is buffered, or follow up trying to 276f4ad559SJosef Bacik * make a real allocation if we are pre-allocating or doing O_DIRECT. 286f4ad559SJosef Bacik * 296f4ad559SJosef Bacik * -> Use 306f4ad559SJosef Bacik * At writepages()/prealloc/O_DIRECT time we will call into 316f4ad559SJosef Bacik * btrfs_reserve_extent() for some part or all of this range of bytes. We 326f4ad559SJosef Bacik * will make the allocation and subtract space_info->bytes_may_use by the 336f4ad559SJosef Bacik * original requested length and increase the space_info->bytes_reserved by 346f4ad559SJosef Bacik * the allocated length. This distinction is important because compression 356f4ad559SJosef Bacik * may allocate a smaller on disk extent than we previously reserved. 366f4ad559SJosef Bacik * 376f4ad559SJosef Bacik * -> Allocation 386f4ad559SJosef Bacik * finish_ordered_io() will insert the new file extent item for this range, 396f4ad559SJosef Bacik * and then add a delayed ref update for the extent tree. Once that delayed 406f4ad559SJosef Bacik * ref is written the extent size is subtracted from 416f4ad559SJosef Bacik * space_info->bytes_reserved and added to space_info->bytes_used. 426f4ad559SJosef Bacik * 436f4ad559SJosef Bacik * Error handling 446f4ad559SJosef Bacik * 456f4ad559SJosef Bacik * -> By the reservation maker 466f4ad559SJosef Bacik * This is the simplest case, we haven't completed our operation and we know 476f4ad559SJosef Bacik * how much we reserved, we can simply call 486f4ad559SJosef Bacik * btrfs_free_reserved_data_space*() and it will be removed from 496f4ad559SJosef Bacik * space_info->bytes_may_use. 506f4ad559SJosef Bacik * 516f4ad559SJosef Bacik * -> After the reservation has been made, but before cow_file_range() 526f4ad559SJosef Bacik * This is specifically for the delalloc case. You must clear 536f4ad559SJosef Bacik * EXTENT_DELALLOC with the EXTENT_CLEAR_DATA_RESV bit, and the range will 546f4ad559SJosef Bacik * be subtracted from space_info->bytes_may_use. 556f4ad559SJosef Bacik * 566f4ad559SJosef Bacik * METADATA RESERVATION 576f4ad559SJosef Bacik * The general metadata reservation lifetimes are discussed elsewhere, this 586f4ad559SJosef Bacik * will just focus on how it is used for delalloc space. 596f4ad559SJosef Bacik * 606f4ad559SJosef Bacik * We keep track of two things on a per inode bases 616f4ad559SJosef Bacik * 626f4ad559SJosef Bacik * ->outstanding_extents 636f4ad559SJosef Bacik * This is the number of file extent items we'll need to handle all of the 646f4ad559SJosef Bacik * outstanding DELALLOC space we have in this inode. We limit the maximum 656f4ad559SJosef Bacik * size of an extent, so a large contiguous dirty area may require more than 666f4ad559SJosef Bacik * one outstanding_extent, which is why count_max_extents() is used to 676f4ad559SJosef Bacik * determine how many outstanding_extents get added. 686f4ad559SJosef Bacik * 696f4ad559SJosef Bacik * ->csum_bytes 706f4ad559SJosef Bacik * This is essentially how many dirty bytes we have for this inode, so we 716f4ad559SJosef Bacik * can calculate the number of checksum items we would have to add in order 726f4ad559SJosef Bacik * to checksum our outstanding data. 736f4ad559SJosef Bacik * 746f4ad559SJosef Bacik * We keep a per-inode block_rsv in order to make it easier to keep track of 756f4ad559SJosef Bacik * our reservation. We use btrfs_calculate_inode_block_rsv_size() to 766f4ad559SJosef Bacik * calculate the current theoretical maximum reservation we would need for the 776f4ad559SJosef Bacik * metadata for this inode. We call this and then adjust our reservation as 786f4ad559SJosef Bacik * necessary, either by attempting to reserve more space, or freeing up excess 796f4ad559SJosef Bacik * space. 806f4ad559SJosef Bacik * 816f4ad559SJosef Bacik * OUTSTANDING_EXTENTS HANDLING 826f4ad559SJosef Bacik * 836f4ad559SJosef Bacik * ->outstanding_extents is used for keeping track of how many extents we will 846f4ad559SJosef Bacik * need to use for this inode, and it will fluctuate depending on where you are 856f4ad559SJosef Bacik * in the life cycle of the dirty data. Consider the following normal case for 866f4ad559SJosef Bacik * a completely clean inode, with a num_bytes < our maximum allowed extent size 876f4ad559SJosef Bacik * 886f4ad559SJosef Bacik * -> reserve 896f4ad559SJosef Bacik * ->outstanding_extents += 1 (current value is 1) 906f4ad559SJosef Bacik * 916f4ad559SJosef Bacik * -> set_delalloc 926f4ad559SJosef Bacik * ->outstanding_extents += 1 (currrent value is 2) 936f4ad559SJosef Bacik * 946f4ad559SJosef Bacik * -> btrfs_delalloc_release_extents() 956f4ad559SJosef Bacik * ->outstanding_extents -= 1 (current value is 1) 966f4ad559SJosef Bacik * 976f4ad559SJosef Bacik * We must call this once we are done, as we hold our reservation for the 986f4ad559SJosef Bacik * duration of our operation, and then assume set_delalloc will update the 996f4ad559SJosef Bacik * counter appropriately. 1006f4ad559SJosef Bacik * 1016f4ad559SJosef Bacik * -> add ordered extent 1026f4ad559SJosef Bacik * ->outstanding_extents += 1 (current value is 2) 1036f4ad559SJosef Bacik * 1046f4ad559SJosef Bacik * -> btrfs_clear_delalloc_extent 1056f4ad559SJosef Bacik * ->outstanding_extents -= 1 (current value is 1) 1066f4ad559SJosef Bacik * 1076f4ad559SJosef Bacik * -> finish_ordered_io/btrfs_remove_ordered_extent 1086f4ad559SJosef Bacik * ->outstanding_extents -= 1 (current value is 0) 1096f4ad559SJosef Bacik * 1106f4ad559SJosef Bacik * Each stage is responsible for their own accounting of the extent, thus 1116f4ad559SJosef Bacik * making error handling and cleanup easier. 1126f4ad559SJosef Bacik */ 1136f4ad559SJosef Bacik 11486736342SJosef Bacik int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) 11586736342SJosef Bacik { 11686736342SJosef Bacik struct btrfs_root *root = inode->root; 11786736342SJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 11886736342SJosef Bacik struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; 11986736342SJosef Bacik u64 used; 12086736342SJosef Bacik int ret = 0; 12186736342SJosef Bacik int need_commit = 2; 12286736342SJosef Bacik int have_pinned_space; 12386736342SJosef Bacik 12486736342SJosef Bacik /* Make sure bytes are sectorsize aligned */ 12586736342SJosef Bacik bytes = ALIGN(bytes, fs_info->sectorsize); 12686736342SJosef Bacik 12786736342SJosef Bacik if (btrfs_is_free_space_inode(inode)) { 12886736342SJosef Bacik need_commit = 0; 12986736342SJosef Bacik ASSERT(current->journal_info); 13086736342SJosef Bacik } 13186736342SJosef Bacik 13286736342SJosef Bacik again: 13386736342SJosef Bacik /* Make sure we have enough space to handle the data first */ 13486736342SJosef Bacik spin_lock(&data_sinfo->lock); 13586736342SJosef Bacik used = btrfs_space_info_used(data_sinfo, true); 13686736342SJosef Bacik 13786736342SJosef Bacik if (used + bytes > data_sinfo->total_bytes) { 13886736342SJosef Bacik struct btrfs_trans_handle *trans; 13986736342SJosef Bacik 14086736342SJosef Bacik /* 14186736342SJosef Bacik * If we don't have enough free bytes in this space then we need 14286736342SJosef Bacik * to alloc a new chunk. 14386736342SJosef Bacik */ 14486736342SJosef Bacik if (!data_sinfo->full) { 14586736342SJosef Bacik u64 alloc_target; 14686736342SJosef Bacik 14786736342SJosef Bacik data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; 14886736342SJosef Bacik spin_unlock(&data_sinfo->lock); 14986736342SJosef Bacik 15086736342SJosef Bacik alloc_target = btrfs_data_alloc_profile(fs_info); 15186736342SJosef Bacik /* 15286736342SJosef Bacik * It is ugly that we don't call nolock join 15386736342SJosef Bacik * transaction for the free space inode case here. 15486736342SJosef Bacik * But it is safe because we only do the data space 15586736342SJosef Bacik * reservation for the free space cache in the 15686736342SJosef Bacik * transaction context, the common join transaction 15786736342SJosef Bacik * just increase the counter of the current transaction 15886736342SJosef Bacik * handler, doesn't try to acquire the trans_lock of 15986736342SJosef Bacik * the fs. 16086736342SJosef Bacik */ 16186736342SJosef Bacik trans = btrfs_join_transaction(root); 16286736342SJosef Bacik if (IS_ERR(trans)) 16386736342SJosef Bacik return PTR_ERR(trans); 16486736342SJosef Bacik 16586736342SJosef Bacik ret = btrfs_chunk_alloc(trans, alloc_target, 16686736342SJosef Bacik CHUNK_ALLOC_NO_FORCE); 16786736342SJosef Bacik btrfs_end_transaction(trans); 16886736342SJosef Bacik if (ret < 0) { 16986736342SJosef Bacik if (ret != -ENOSPC) 17086736342SJosef Bacik return ret; 17186736342SJosef Bacik else { 17286736342SJosef Bacik have_pinned_space = 1; 17386736342SJosef Bacik goto commit_trans; 17486736342SJosef Bacik } 17586736342SJosef Bacik } 17686736342SJosef Bacik 17786736342SJosef Bacik goto again; 17886736342SJosef Bacik } 17986736342SJosef Bacik 18086736342SJosef Bacik /* 18186736342SJosef Bacik * If we don't have enough pinned space to deal with this 18286736342SJosef Bacik * allocation, and no removed chunk in current transaction, 18386736342SJosef Bacik * don't bother committing the transaction. 18486736342SJosef Bacik */ 18586736342SJosef Bacik have_pinned_space = __percpu_counter_compare( 18686736342SJosef Bacik &data_sinfo->total_bytes_pinned, 18786736342SJosef Bacik used + bytes - data_sinfo->total_bytes, 18886736342SJosef Bacik BTRFS_TOTAL_BYTES_PINNED_BATCH); 18986736342SJosef Bacik spin_unlock(&data_sinfo->lock); 19086736342SJosef Bacik 19186736342SJosef Bacik /* Commit the current transaction and try again */ 19286736342SJosef Bacik commit_trans: 19386736342SJosef Bacik if (need_commit) { 19486736342SJosef Bacik need_commit--; 19586736342SJosef Bacik 19686736342SJosef Bacik if (need_commit > 0) { 19786736342SJosef Bacik btrfs_start_delalloc_roots(fs_info, -1); 19886736342SJosef Bacik btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, 19986736342SJosef Bacik (u64)-1); 20086736342SJosef Bacik } 20186736342SJosef Bacik 20286736342SJosef Bacik trans = btrfs_join_transaction(root); 20386736342SJosef Bacik if (IS_ERR(trans)) 20486736342SJosef Bacik return PTR_ERR(trans); 20586736342SJosef Bacik if (have_pinned_space >= 0 || 20686736342SJosef Bacik test_bit(BTRFS_TRANS_HAVE_FREE_BGS, 20786736342SJosef Bacik &trans->transaction->flags) || 20886736342SJosef Bacik need_commit > 0) { 20986736342SJosef Bacik ret = btrfs_commit_transaction(trans); 21086736342SJosef Bacik if (ret) 21186736342SJosef Bacik return ret; 21286736342SJosef Bacik /* 21386736342SJosef Bacik * The cleaner kthread might still be doing iput 21486736342SJosef Bacik * operations. Wait for it to finish so that 21586736342SJosef Bacik * more space is released. We don't need to 21686736342SJosef Bacik * explicitly run the delayed iputs here because 21786736342SJosef Bacik * the commit_transaction would have woken up 21886736342SJosef Bacik * the cleaner. 21986736342SJosef Bacik */ 22086736342SJosef Bacik ret = btrfs_wait_on_delayed_iputs(fs_info); 22186736342SJosef Bacik if (ret) 22286736342SJosef Bacik return ret; 22386736342SJosef Bacik goto again; 22486736342SJosef Bacik } else { 22586736342SJosef Bacik btrfs_end_transaction(trans); 22686736342SJosef Bacik } 22786736342SJosef Bacik } 22886736342SJosef Bacik 22986736342SJosef Bacik trace_btrfs_space_reservation(fs_info, 23086736342SJosef Bacik "space_info:enospc", 23186736342SJosef Bacik data_sinfo->flags, bytes, 1); 23286736342SJosef Bacik return -ENOSPC; 23386736342SJosef Bacik } 23486736342SJosef Bacik btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, bytes); 23586736342SJosef Bacik spin_unlock(&data_sinfo->lock); 23686736342SJosef Bacik 23786736342SJosef Bacik return 0; 23886736342SJosef Bacik } 23986736342SJosef Bacik 24036ea6f3eSNikolay Borisov int btrfs_check_data_free_space(struct btrfs_inode *inode, 24186736342SJosef Bacik struct extent_changeset **reserved, u64 start, u64 len) 24286736342SJosef Bacik { 24336ea6f3eSNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 24486736342SJosef Bacik int ret; 24586736342SJosef Bacik 24686736342SJosef Bacik /* align the range */ 24786736342SJosef Bacik len = round_up(start + len, fs_info->sectorsize) - 24886736342SJosef Bacik round_down(start, fs_info->sectorsize); 24986736342SJosef Bacik start = round_down(start, fs_info->sectorsize); 25086736342SJosef Bacik 25136ea6f3eSNikolay Borisov ret = btrfs_alloc_data_chunk_ondemand(inode, len); 25286736342SJosef Bacik if (ret < 0) 25386736342SJosef Bacik return ret; 25486736342SJosef Bacik 25586736342SJosef Bacik /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ 25636ea6f3eSNikolay Borisov ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); 25786736342SJosef Bacik if (ret < 0) 2589db5d510SNikolay Borisov btrfs_free_reserved_data_space_noquota(fs_info, len); 25986736342SJosef Bacik else 26086736342SJosef Bacik ret = 0; 26186736342SJosef Bacik return ret; 26286736342SJosef Bacik } 26386736342SJosef Bacik 26486736342SJosef Bacik /* 26586736342SJosef Bacik * Called if we need to clear a data reservation for this inode 26686736342SJosef Bacik * Normally in a error case. 26786736342SJosef Bacik * 26886736342SJosef Bacik * This one will *NOT* use accurate qgroup reserved space API, just for case 26986736342SJosef Bacik * which we can't sleep and is sure it won't affect qgroup reserved space. 27086736342SJosef Bacik * Like clear_bit_hook(). 27186736342SJosef Bacik */ 2729db5d510SNikolay Borisov void btrfs_free_reserved_data_space_noquota(struct btrfs_fs_info *fs_info, 27386736342SJosef Bacik u64 len) 27486736342SJosef Bacik { 27586736342SJosef Bacik struct btrfs_space_info *data_sinfo; 27686736342SJosef Bacik 27746d4dac8SFilipe Manana ASSERT(IS_ALIGNED(len, fs_info->sectorsize)); 27886736342SJosef Bacik 27986736342SJosef Bacik data_sinfo = fs_info->data_sinfo; 28086736342SJosef Bacik spin_lock(&data_sinfo->lock); 28186736342SJosef Bacik btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, -len); 28286736342SJosef Bacik spin_unlock(&data_sinfo->lock); 28386736342SJosef Bacik } 28486736342SJosef Bacik 28586736342SJosef Bacik /* 28686736342SJosef Bacik * Called if we need to clear a data reservation for this inode 28786736342SJosef Bacik * Normally in a error case. 28886736342SJosef Bacik * 28986736342SJosef Bacik * This one will handle the per-inode data rsv map for accurate reserved 29086736342SJosef Bacik * space framework. 29186736342SJosef Bacik */ 29225ce28caSNikolay Borisov void btrfs_free_reserved_data_space(struct btrfs_inode *inode, 29386736342SJosef Bacik struct extent_changeset *reserved, u64 start, u64 len) 29486736342SJosef Bacik { 29525ce28caSNikolay Borisov struct btrfs_fs_info *fs_info = inode->root->fs_info; 29686736342SJosef Bacik 29786736342SJosef Bacik /* Make sure the range is aligned to sectorsize */ 29825ce28caSNikolay Borisov len = round_up(start + len, fs_info->sectorsize) - 29925ce28caSNikolay Borisov round_down(start, fs_info->sectorsize); 30025ce28caSNikolay Borisov start = round_down(start, fs_info->sectorsize); 30186736342SJosef Bacik 30225ce28caSNikolay Borisov btrfs_free_reserved_data_space_noquota(fs_info, len); 30325ce28caSNikolay Borisov btrfs_qgroup_free_data(inode, reserved, start, len); 30486736342SJosef Bacik } 30586736342SJosef Bacik 30686736342SJosef Bacik /** 30786736342SJosef Bacik * btrfs_inode_rsv_release - release any excessive reservation. 30886736342SJosef Bacik * @inode - the inode we need to release from. 30986736342SJosef Bacik * @qgroup_free - free or convert qgroup meta. 31086736342SJosef Bacik * Unlike normal operation, qgroup meta reservation needs to know if we are 31186736342SJosef Bacik * freeing qgroup reservation or just converting it into per-trans. Normally 31286736342SJosef Bacik * @qgroup_free is true for error handling, and false for normal release. 31386736342SJosef Bacik * 31486736342SJosef Bacik * This is the same as btrfs_block_rsv_release, except that it handles the 31586736342SJosef Bacik * tracepoint for the reservation. 31686736342SJosef Bacik */ 31786736342SJosef Bacik static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) 31886736342SJosef Bacik { 31986736342SJosef Bacik struct btrfs_fs_info *fs_info = inode->root->fs_info; 32086736342SJosef Bacik struct btrfs_block_rsv *block_rsv = &inode->block_rsv; 32186736342SJosef Bacik u64 released = 0; 32286736342SJosef Bacik u64 qgroup_to_release = 0; 32386736342SJosef Bacik 32486736342SJosef Bacik /* 32586736342SJosef Bacik * Since we statically set the block_rsv->size we just want to say we 32686736342SJosef Bacik * are releasing 0 bytes, and then we'll just get the reservation over 32786736342SJosef Bacik * the size free'd. 32886736342SJosef Bacik */ 32963f018beSNikolay Borisov released = btrfs_block_rsv_release(fs_info, block_rsv, 0, 33086736342SJosef Bacik &qgroup_to_release); 33186736342SJosef Bacik if (released > 0) 33286736342SJosef Bacik trace_btrfs_space_reservation(fs_info, "delalloc", 33386736342SJosef Bacik btrfs_ino(inode), released, 0); 33486736342SJosef Bacik if (qgroup_free) 33586736342SJosef Bacik btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); 33686736342SJosef Bacik else 33786736342SJosef Bacik btrfs_qgroup_convert_reserved_meta(inode->root, 33886736342SJosef Bacik qgroup_to_release); 33986736342SJosef Bacik } 34086736342SJosef Bacik 34186736342SJosef Bacik static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, 34286736342SJosef Bacik struct btrfs_inode *inode) 34386736342SJosef Bacik { 34486736342SJosef Bacik struct btrfs_block_rsv *block_rsv = &inode->block_rsv; 34586736342SJosef Bacik u64 reserve_size = 0; 34686736342SJosef Bacik u64 qgroup_rsv_size = 0; 34786736342SJosef Bacik u64 csum_leaves; 34886736342SJosef Bacik unsigned outstanding_extents; 34986736342SJosef Bacik 35086736342SJosef Bacik lockdep_assert_held(&inode->lock); 35186736342SJosef Bacik outstanding_extents = inode->outstanding_extents; 352bcacf5f3SJosef Bacik 353bcacf5f3SJosef Bacik /* 354bcacf5f3SJosef Bacik * Insert size for the number of outstanding extents, 1 normal size for 355bcacf5f3SJosef Bacik * updating the inode. 356bcacf5f3SJosef Bacik */ 357bcacf5f3SJosef Bacik if (outstanding_extents) { 3582bd36e7bSJosef Bacik reserve_size = btrfs_calc_insert_metadata_size(fs_info, 359bcacf5f3SJosef Bacik outstanding_extents); 360bcacf5f3SJosef Bacik reserve_size += btrfs_calc_metadata_size(fs_info, 1); 361bcacf5f3SJosef Bacik } 36286736342SJosef Bacik csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, 36386736342SJosef Bacik inode->csum_bytes); 3642bd36e7bSJosef Bacik reserve_size += btrfs_calc_insert_metadata_size(fs_info, 36586736342SJosef Bacik csum_leaves); 36686736342SJosef Bacik /* 36786736342SJosef Bacik * For qgroup rsv, the calculation is very simple: 36886736342SJosef Bacik * account one nodesize for each outstanding extent 36986736342SJosef Bacik * 37086736342SJosef Bacik * This is overestimating in most cases. 37186736342SJosef Bacik */ 37286736342SJosef Bacik qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize; 37386736342SJosef Bacik 37486736342SJosef Bacik spin_lock(&block_rsv->lock); 37586736342SJosef Bacik block_rsv->size = reserve_size; 37686736342SJosef Bacik block_rsv->qgroup_rsv_size = qgroup_rsv_size; 37786736342SJosef Bacik spin_unlock(&block_rsv->lock); 37886736342SJosef Bacik } 37986736342SJosef Bacik 38086736342SJosef Bacik static void calc_inode_reservations(struct btrfs_fs_info *fs_info, 38186736342SJosef Bacik u64 num_bytes, u64 *meta_reserve, 38286736342SJosef Bacik u64 *qgroup_reserve) 38386736342SJosef Bacik { 38486736342SJosef Bacik u64 nr_extents = count_max_extents(num_bytes); 38586736342SJosef Bacik u64 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, num_bytes); 386bcacf5f3SJosef Bacik u64 inode_update = btrfs_calc_metadata_size(fs_info, 1); 38786736342SJosef Bacik 3882bd36e7bSJosef Bacik *meta_reserve = btrfs_calc_insert_metadata_size(fs_info, 389bcacf5f3SJosef Bacik nr_extents + csum_leaves); 390bcacf5f3SJosef Bacik 391bcacf5f3SJosef Bacik /* 392bcacf5f3SJosef Bacik * finish_ordered_io has to update the inode, so add the space required 393bcacf5f3SJosef Bacik * for an inode update. 394bcacf5f3SJosef Bacik */ 395bcacf5f3SJosef Bacik *meta_reserve += inode_update; 39686736342SJosef Bacik *qgroup_reserve = nr_extents * fs_info->nodesize; 39786736342SJosef Bacik } 39886736342SJosef Bacik 39986736342SJosef Bacik int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) 40086736342SJosef Bacik { 40186736342SJosef Bacik struct btrfs_root *root = inode->root; 40286736342SJosef Bacik struct btrfs_fs_info *fs_info = root->fs_info; 40386736342SJosef Bacik struct btrfs_block_rsv *block_rsv = &inode->block_rsv; 40486736342SJosef Bacik u64 meta_reserve, qgroup_reserve; 40586736342SJosef Bacik unsigned nr_extents; 40686736342SJosef Bacik enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; 40786736342SJosef Bacik int ret = 0; 40886736342SJosef Bacik 40986736342SJosef Bacik /* 41086736342SJosef Bacik * If we are a free space inode we need to not flush since we will be in 41186736342SJosef Bacik * the middle of a transaction commit. We also don't need the delalloc 41286736342SJosef Bacik * mutex since we won't race with anybody. We need this mostly to make 41386736342SJosef Bacik * lockdep shut its filthy mouth. 41486736342SJosef Bacik * 41586736342SJosef Bacik * If we have a transaction open (can happen if we call truncate_block 41686736342SJosef Bacik * from truncate), then we need FLUSH_LIMIT so we don't deadlock. 41786736342SJosef Bacik */ 41886736342SJosef Bacik if (btrfs_is_free_space_inode(inode)) { 41986736342SJosef Bacik flush = BTRFS_RESERVE_NO_FLUSH; 42086736342SJosef Bacik } else { 42186736342SJosef Bacik if (current->journal_info) 42286736342SJosef Bacik flush = BTRFS_RESERVE_FLUSH_LIMIT; 42386736342SJosef Bacik 42486736342SJosef Bacik if (btrfs_transaction_in_commit(fs_info)) 42586736342SJosef Bacik schedule_timeout(1); 42686736342SJosef Bacik } 42786736342SJosef Bacik 42886736342SJosef Bacik num_bytes = ALIGN(num_bytes, fs_info->sectorsize); 42986736342SJosef Bacik 43086736342SJosef Bacik /* 43186736342SJosef Bacik * We always want to do it this way, every other way is wrong and ends 43286736342SJosef Bacik * in tears. Pre-reserving the amount we are going to add will always 43386736342SJosef Bacik * be the right way, because otherwise if we have enough parallelism we 43486736342SJosef Bacik * could end up with thousands of inodes all holding little bits of 43586736342SJosef Bacik * reservations they were able to make previously and the only way to 43686736342SJosef Bacik * reclaim that space is to ENOSPC out the operations and clear 43786736342SJosef Bacik * everything out and try again, which is bad. This way we just 43886736342SJosef Bacik * over-reserve slightly, and clean up the mess when we are done. 43986736342SJosef Bacik */ 44086736342SJosef Bacik calc_inode_reservations(fs_info, num_bytes, &meta_reserve, 44186736342SJosef Bacik &qgroup_reserve); 44286736342SJosef Bacik ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserve, true); 44386736342SJosef Bacik if (ret) 44416ad3be1SFilipe Manana return ret; 44586736342SJosef Bacik ret = btrfs_reserve_metadata_bytes(root, block_rsv, meta_reserve, flush); 44616ad3be1SFilipe Manana if (ret) { 44716ad3be1SFilipe Manana btrfs_qgroup_free_meta_prealloc(root, qgroup_reserve); 44816ad3be1SFilipe Manana return ret; 44916ad3be1SFilipe Manana } 45086736342SJosef Bacik 45186736342SJosef Bacik /* 45286736342SJosef Bacik * Now we need to update our outstanding extents and csum bytes _first_ 45386736342SJosef Bacik * and then add the reservation to the block_rsv. This keeps us from 45486736342SJosef Bacik * racing with an ordered completion or some such that would think it 45586736342SJosef Bacik * needs to free the reservation we just made. 45686736342SJosef Bacik */ 45786736342SJosef Bacik spin_lock(&inode->lock); 45886736342SJosef Bacik nr_extents = count_max_extents(num_bytes); 45986736342SJosef Bacik btrfs_mod_outstanding_extents(inode, nr_extents); 46086736342SJosef Bacik inode->csum_bytes += num_bytes; 46186736342SJosef Bacik btrfs_calculate_inode_block_rsv_size(fs_info, inode); 46286736342SJosef Bacik spin_unlock(&inode->lock); 46386736342SJosef Bacik 46486736342SJosef Bacik /* Now we can safely add our space to our block rsv */ 46586736342SJosef Bacik btrfs_block_rsv_add_bytes(block_rsv, meta_reserve, false); 46686736342SJosef Bacik trace_btrfs_space_reservation(root->fs_info, "delalloc", 46786736342SJosef Bacik btrfs_ino(inode), meta_reserve, 1); 46886736342SJosef Bacik 46986736342SJosef Bacik spin_lock(&block_rsv->lock); 47086736342SJosef Bacik block_rsv->qgroup_rsv_reserved += qgroup_reserve; 47186736342SJosef Bacik spin_unlock(&block_rsv->lock); 47286736342SJosef Bacik 47386736342SJosef Bacik return 0; 47486736342SJosef Bacik } 47586736342SJosef Bacik 47686736342SJosef Bacik /** 47786736342SJosef Bacik * btrfs_delalloc_release_metadata - release a metadata reservation for an inode 47886736342SJosef Bacik * @inode: the inode to release the reservation for. 47986736342SJosef Bacik * @num_bytes: the number of bytes we are releasing. 48086736342SJosef Bacik * @qgroup_free: free qgroup reservation or convert it to per-trans reservation 48186736342SJosef Bacik * 48286736342SJosef Bacik * This will release the metadata reservation for an inode. This can be called 48386736342SJosef Bacik * once we complete IO for a given set of bytes to release their metadata 48486736342SJosef Bacik * reservations, or on error for the same reason. 48586736342SJosef Bacik */ 48686736342SJosef Bacik void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, 48786736342SJosef Bacik bool qgroup_free) 48886736342SJosef Bacik { 48986736342SJosef Bacik struct btrfs_fs_info *fs_info = inode->root->fs_info; 49086736342SJosef Bacik 49186736342SJosef Bacik num_bytes = ALIGN(num_bytes, fs_info->sectorsize); 49286736342SJosef Bacik spin_lock(&inode->lock); 49386736342SJosef Bacik inode->csum_bytes -= num_bytes; 49486736342SJosef Bacik btrfs_calculate_inode_block_rsv_size(fs_info, inode); 49586736342SJosef Bacik spin_unlock(&inode->lock); 49686736342SJosef Bacik 49786736342SJosef Bacik if (btrfs_is_testing(fs_info)) 49886736342SJosef Bacik return; 49986736342SJosef Bacik 50086736342SJosef Bacik btrfs_inode_rsv_release(inode, qgroup_free); 50186736342SJosef Bacik } 50286736342SJosef Bacik 50386736342SJosef Bacik /** 50486736342SJosef Bacik * btrfs_delalloc_release_extents - release our outstanding_extents 50586736342SJosef Bacik * @inode: the inode to balance the reservation for. 50686736342SJosef Bacik * @num_bytes: the number of bytes we originally reserved with 50786736342SJosef Bacik * 50886736342SJosef Bacik * When we reserve space we increase outstanding_extents for the extents we may 50986736342SJosef Bacik * add. Once we've set the range as delalloc or created our ordered extents we 51086736342SJosef Bacik * have outstanding_extents to track the real usage, so we use this to free our 51186736342SJosef Bacik * temporarily tracked outstanding_extents. This _must_ be used in conjunction 51286736342SJosef Bacik * with btrfs_delalloc_reserve_metadata. 51386736342SJosef Bacik */ 5148702ba93SQu Wenruo void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes) 51586736342SJosef Bacik { 51686736342SJosef Bacik struct btrfs_fs_info *fs_info = inode->root->fs_info; 51786736342SJosef Bacik unsigned num_extents; 51886736342SJosef Bacik 51986736342SJosef Bacik spin_lock(&inode->lock); 52086736342SJosef Bacik num_extents = count_max_extents(num_bytes); 52186736342SJosef Bacik btrfs_mod_outstanding_extents(inode, -num_extents); 52286736342SJosef Bacik btrfs_calculate_inode_block_rsv_size(fs_info, inode); 52386736342SJosef Bacik spin_unlock(&inode->lock); 52486736342SJosef Bacik 52586736342SJosef Bacik if (btrfs_is_testing(fs_info)) 52686736342SJosef Bacik return; 52786736342SJosef Bacik 5288702ba93SQu Wenruo btrfs_inode_rsv_release(inode, true); 52986736342SJosef Bacik } 53086736342SJosef Bacik 53186736342SJosef Bacik /** 53286736342SJosef Bacik * btrfs_delalloc_reserve_space - reserve data and metadata space for 53386736342SJosef Bacik * delalloc 53486736342SJosef Bacik * @inode: inode we're writing to 53586736342SJosef Bacik * @start: start range we are writing to 53686736342SJosef Bacik * @len: how long the range we are writing to 53786736342SJosef Bacik * @reserved: mandatory parameter, record actually reserved qgroup ranges of 53886736342SJosef Bacik * current reservation. 53986736342SJosef Bacik * 54086736342SJosef Bacik * This will do the following things 54186736342SJosef Bacik * 54286736342SJosef Bacik * - reserve space in data space info for num bytes 54386736342SJosef Bacik * and reserve precious corresponding qgroup space 54486736342SJosef Bacik * (Done in check_data_free_space) 54586736342SJosef Bacik * 54686736342SJosef Bacik * - reserve space for metadata space, based on the number of outstanding 54786736342SJosef Bacik * extents and how much csums will be needed 54886736342SJosef Bacik * also reserve metadata space in a per root over-reserve method. 54986736342SJosef Bacik * - add to the inodes->delalloc_bytes 55086736342SJosef Bacik * - add it to the fs_info's delalloc inodes list. 55186736342SJosef Bacik * (Above 3 all done in delalloc_reserve_metadata) 55286736342SJosef Bacik * 55386736342SJosef Bacik * Return 0 for success 55486736342SJosef Bacik * Return <0 for error(-ENOSPC or -EQUOT) 55586736342SJosef Bacik */ 55686736342SJosef Bacik int btrfs_delalloc_reserve_space(struct inode *inode, 55786736342SJosef Bacik struct extent_changeset **reserved, u64 start, u64 len) 55886736342SJosef Bacik { 55986736342SJosef Bacik int ret; 56086736342SJosef Bacik 56136ea6f3eSNikolay Borisov ret = btrfs_check_data_free_space(BTRFS_I(inode), reserved, start, len); 56286736342SJosef Bacik if (ret < 0) 56386736342SJosef Bacik return ret; 56486736342SJosef Bacik ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); 56586736342SJosef Bacik if (ret < 0) 56625ce28caSNikolay Borisov btrfs_free_reserved_data_space(BTRFS_I(inode), *reserved, start, len); 56786736342SJosef Bacik return ret; 56886736342SJosef Bacik } 56986736342SJosef Bacik 57086736342SJosef Bacik /** 57186736342SJosef Bacik * btrfs_delalloc_release_space - release data and metadata space for delalloc 57286736342SJosef Bacik * @inode: inode we're releasing space for 57386736342SJosef Bacik * @start: start position of the space already reserved 57486736342SJosef Bacik * @len: the len of the space already reserved 57586736342SJosef Bacik * @release_bytes: the len of the space we consumed or didn't use 57686736342SJosef Bacik * 57786736342SJosef Bacik * This function will release the metadata space that was not used and will 57886736342SJosef Bacik * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes 57986736342SJosef Bacik * list if there are no delalloc bytes left. 58086736342SJosef Bacik * Also it will handle the qgroup reserved space. 58186736342SJosef Bacik */ 58286d52921SNikolay Borisov void btrfs_delalloc_release_space(struct btrfs_inode *inode, 58386736342SJosef Bacik struct extent_changeset *reserved, 58486736342SJosef Bacik u64 start, u64 len, bool qgroup_free) 58586736342SJosef Bacik { 58686d52921SNikolay Borisov btrfs_delalloc_release_metadata(inode, len, qgroup_free); 58786d52921SNikolay Borisov btrfs_free_reserved_data_space(inode, reserved, start, len); 58886736342SJosef Bacik } 589