1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 36cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 679154b1bSChris Mason #include <linux/fs.h> 75a0e3ad6STejun Heo #include <linux/slab.h> 834088780SChris Mason #include <linux/sched.h> 9d3c2fdcfSChris Mason #include <linux/writeback.h> 105f39d397SChris Mason #include <linux/pagemap.h> 115f2cc086SChris Mason #include <linux/blkdev.h> 128ea05e3aSAlexander Block #include <linux/uuid.h> 13602cbe91SDavid Sterba #include "misc.h" 1479154b1bSChris Mason #include "ctree.h" 1579154b1bSChris Mason #include "disk-io.h" 1679154b1bSChris Mason #include "transaction.h" 17925baeddSChris Mason #include "locking.h" 18e02119d5SChris Mason #include "tree-log.h" 19733f4fbbSStefan Behrens #include "volumes.h" 208dabb742SStefan Behrens #include "dev-replace.h" 21fcebe456SJosef Bacik #include "qgroup.h" 22aac0023cSJosef Bacik #include "block-group.h" 239c343784SJosef Bacik #include "space-info.h" 24d3575156SNaohiro Aota #include "zoned.h" 2579154b1bSChris Mason 260f7d52f4SChris Mason #define BTRFS_ROOT_TRANS_TAG 0 270f7d52f4SChris Mason 2861c047b5SQu Wenruo /* 2961c047b5SQu Wenruo * Transaction states and transitions 3061c047b5SQu Wenruo * 3161c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 3261c047b5SQu Wenruo * | 3361c047b5SQu Wenruo * | To next stage: 3461c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 3561c047b5SQu Wenruo * V 3661c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 3761c047b5SQu Wenruo * | 3861c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 3961c047b5SQu Wenruo * | start_transaction() variants. 4061c047b5SQu Wenruo * | 4161c047b5SQu Wenruo * | To next stage: 4261c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 4361c047b5SQu Wenruo * | transaction N 4461c047b5SQu Wenruo * V 4561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 4661c047b5SQu Wenruo * | 4761c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 4861c047b5SQu Wenruo * | is one 4961c047b5SQu Wenruo * | 5061c047b5SQu Wenruo * | Then one of the following happes: 5161c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 5261c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 5361c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 5461c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 5561c047b5SQu Wenruo * | 5661c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 5761c047b5SQu Wenruo * | to this running transaction. 5861c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 5961c047b5SQu Wenruo * | transaction N+1. 6061c047b5SQu Wenruo * | 6161c047b5SQu Wenruo * | To next stage: 6261c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 6361c047b5SQu Wenruo * | haven been released. 6461c047b5SQu Wenruo * V 6561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 6661c047b5SQu Wenruo * | 6761c047b5SQu Wenruo * | The heavy lifting transaction work is started. 6861c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 6961c047b5SQu Wenruo * | snapshots, running qgroups. 7061c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 7161c047b5SQu Wenruo * | trees. 7261c047b5SQu Wenruo * | 7361c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 7461c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 7561c047b5SQu Wenruo * | 7661c047b5SQu Wenruo * | To next stage: 7761c047b5SQu Wenruo * | Until all supporting trees are updated. 7861c047b5SQu Wenruo * V 7961c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 8061c047b5SQu Wenruo * | Transaction N+1 8161c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 8261c047b5SQu Wenruo * | need to write them back to disk and update | 8361c047b5SQu Wenruo * | super blocks. | 8461c047b5SQu Wenruo * | | 8561c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 8661c047b5SQu Wenruo * | start. | 8761c047b5SQu Wenruo * | All new start_transaction() calls will be | 8861c047b5SQu Wenruo * | attached to transid N+1. | 8961c047b5SQu Wenruo * | | 9061c047b5SQu Wenruo * | To next stage: | 9161c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 9261c047b5SQu Wenruo * | written to block devices | 9361c047b5SQu Wenruo * V | 9461c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 9561c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 9661c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 9761c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 9861c047b5SQu Wenruo */ 99e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1004a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 101bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 102bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1034a9d8bdeSMiao Xie __TRANS_ATTACH | 104a6d155d2SFilipe Manana __TRANS_JOIN | 105a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 106bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1074a9d8bdeSMiao Xie __TRANS_ATTACH | 1084a9d8bdeSMiao Xie __TRANS_JOIN | 109a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 110a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 111d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 112d0c2f4faSFilipe Manana __TRANS_ATTACH | 113d0c2f4faSFilipe Manana __TRANS_JOIN | 114d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 115d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 116bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1174a9d8bdeSMiao Xie __TRANS_ATTACH | 1184a9d8bdeSMiao Xie __TRANS_JOIN | 119a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 120a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1214a9d8bdeSMiao Xie }; 1224a9d8bdeSMiao Xie 123724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 12479154b1bSChris Mason { 1259b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1269b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 127a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1285c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1295c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 13081f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 13181f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1321262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 133ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 134ab8d0fc4SJeff Mahoney "pending csums is %llu", 1351262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1367785a663SFilipe Manana /* 1377785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1387785a663SFilipe Manana * because the transaction was aborted and a commit did not 1397785a663SFilipe Manana * happen (things failed before writing the new superblock 1407785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1417785a663SFilipe Manana * discard the physical locations of the block groups. 1427785a663SFilipe Manana */ 1437785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 14432da5386SDavid Sterba struct btrfs_block_group *cache; 1457785a663SFilipe Manana 1467785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 14732da5386SDavid Sterba struct btrfs_block_group, 1487785a663SFilipe Manana bg_list); 1497785a663SFilipe Manana list_del_init(&cache->bg_list); 1506b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1517785a663SFilipe Manana btrfs_put_block_group(cache); 1527785a663SFilipe Manana } 153bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1544b5faeacSDavid Sterba kfree(transaction); 15579154b1bSChris Mason } 15678fae27eSChris Mason } 15779154b1bSChris Mason 158889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 159817d52f8SJosef Bacik { 160889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 16116916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1629e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 16327d56e62SJosef Bacik struct btrfs_caching_control *caching_ctl, *next; 1649e351cc8SJosef Bacik 165dfba78dcSFilipe Manana /* 166dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 167dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 168dfba78dcSFilipe Manana */ 169dfba78dcSFilipe Manana ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); 170dfba78dcSFilipe Manana 1719e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 172*d96b3424SFilipe Manana 173*d96b3424SFilipe Manana if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 174*d96b3424SFilipe Manana fs_info->last_reloc_trans = trans->transid; 175*d96b3424SFilipe Manana 176889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1779e351cc8SJosef Bacik dirty_list) { 1789e351cc8SJosef Bacik list_del_init(&root->dirty_list); 179817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 180817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 18141e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 182370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1839e351cc8SJosef Bacik } 1842b9dbef2SJosef Bacik 1852b9dbef2SJosef Bacik /* We can free old roots now. */ 186889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 187889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 188889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1892b9dbef2SJosef Bacik struct btrfs_root, root_list); 1902b9dbef2SJosef Bacik list_del_init(&root->root_list); 191889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 192889bfa39SJosef Bacik btrfs_free_log(trans, root); 1932b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 194889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 1952b9dbef2SJosef Bacik } 196889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 19727d56e62SJosef Bacik 19827d56e62SJosef Bacik /* 19927d56e62SJosef Bacik * We have to update the last_byte_to_unpin under the commit_root_sem, 20027d56e62SJosef Bacik * at the same time we swap out the commit roots. 20127d56e62SJosef Bacik * 20227d56e62SJosef Bacik * This is because we must have a real view of the last spot the caching 20327d56e62SJosef Bacik * kthreads were while caching. Consider the following views of the 20427d56e62SJosef Bacik * extent tree for a block group 20527d56e62SJosef Bacik * 20627d56e62SJosef Bacik * commit root 20727d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20827d56e62SJosef Bacik * |\\\\| |\\\\|\\\\| |\\\\|\\\\| 20927d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 21027d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 21127d56e62SJosef Bacik * 21227d56e62SJosef Bacik * new commit root 21327d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 21427d56e62SJosef Bacik * | | | |\\\\| | |\\\\| 21527d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 21627d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 21727d56e62SJosef Bacik * 21827d56e62SJosef Bacik * If the cache_ctl->progress was at 3, then we are only allowed to 21927d56e62SJosef Bacik * unpin [0,1) and [2,3], because the caching thread has already 22027d56e62SJosef Bacik * processed those extents. We are not allowed to unpin [5,6), because 22127d56e62SJosef Bacik * the caching thread will re-start it's search from 3, and thus find 22227d56e62SJosef Bacik * the hole from [4,6) to add to the free space cache. 22327d56e62SJosef Bacik */ 224bbb86a37SJosef Bacik spin_lock(&fs_info->block_group_cache_lock); 22527d56e62SJosef Bacik list_for_each_entry_safe(caching_ctl, next, 22627d56e62SJosef Bacik &fs_info->caching_block_groups, list) { 22727d56e62SJosef Bacik struct btrfs_block_group *cache = caching_ctl->block_group; 22827d56e62SJosef Bacik 22927d56e62SJosef Bacik if (btrfs_block_group_done(cache)) { 23027d56e62SJosef Bacik cache->last_byte_to_unpin = (u64)-1; 23127d56e62SJosef Bacik list_del_init(&caching_ctl->list); 23227d56e62SJosef Bacik btrfs_put_caching_control(caching_ctl); 23327d56e62SJosef Bacik } else { 23427d56e62SJosef Bacik cache->last_byte_to_unpin = caching_ctl->progress; 23527d56e62SJosef Bacik } 23627d56e62SJosef Bacik } 237bbb86a37SJosef Bacik spin_unlock(&fs_info->block_group_cache_lock); 2389e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 239817d52f8SJosef Bacik } 240817d52f8SJosef Bacik 2410860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2420860adfdSMiao Xie unsigned int type) 2430860adfdSMiao Xie { 2440860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2450860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2460860adfdSMiao Xie } 2470860adfdSMiao Xie 2480860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2490860adfdSMiao Xie unsigned int type) 2500860adfdSMiao Xie { 2510860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2520860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2530860adfdSMiao Xie } 2540860adfdSMiao Xie 2550860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2560860adfdSMiao Xie unsigned int type) 2570860adfdSMiao Xie { 2580860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2590860adfdSMiao Xie } 2600860adfdSMiao Xie 2610860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2620860adfdSMiao Xie { 2630860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 264178260b2SMiao Xie } 265178260b2SMiao Xie 266d352ac68SChris Mason /* 26779bd3712SFilipe Manana * To be called after doing the chunk btree updates right after allocating a new 26879bd3712SFilipe Manana * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a 26979bd3712SFilipe Manana * chunk after all chunk btree updates and after finishing the second phase of 27079bd3712SFilipe Manana * chunk allocation (btrfs_create_pending_block_groups()) in case some block 27179bd3712SFilipe Manana * group had its chunk item insertion delayed to the second phase. 272fb6dea26SJosef Bacik */ 273fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 274fb6dea26SJosef Bacik { 275fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 276fb6dea26SJosef Bacik 277fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 278fb6dea26SJosef Bacik return; 279fb6dea26SJosef Bacik 280fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 28163f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 282fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 283fb6dea26SJosef Bacik } 284fb6dea26SJosef Bacik 285fb6dea26SJosef Bacik /* 286d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 287d352ac68SChris Mason */ 2882ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2892ff7e61eSJeff Mahoney unsigned int type) 29079154b1bSChris Mason { 29179154b1bSChris Mason struct btrfs_transaction *cur_trans; 292a4abeea4SJosef Bacik 29319ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 294d43317dcSChris Mason loop: 29549b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 29684961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 29719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 29849b25e05SJeff Mahoney return -EROFS; 29949b25e05SJeff Mahoney } 30049b25e05SJeff Mahoney 30119ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 302a4abeea4SJosef Bacik if (cur_trans) { 303bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 30419ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 30549b25e05SJeff Mahoney return cur_trans->aborted; 306871383beSDavid Sterba } 3074a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 308178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 309178260b2SMiao Xie return -EBUSY; 310178260b2SMiao Xie } 3119b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 312a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 3130860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 31419ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 315a4abeea4SJosef Bacik return 0; 316a4abeea4SJosef Bacik } 31719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 318a4abeea4SJosef Bacik 319354aa0fbSMiao Xie /* 320354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 321354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 322354aa0fbSMiao Xie */ 323354aa0fbSMiao Xie if (type == TRANS_ATTACH) 324354aa0fbSMiao Xie return -ENOENT; 325354aa0fbSMiao Xie 3264a9d8bdeSMiao Xie /* 3274a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3284a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3294a9d8bdeSMiao Xie */ 3304a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3314a9d8bdeSMiao Xie 3324b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 333db5b493aSTsutomu Itoh if (!cur_trans) 334db5b493aSTsutomu Itoh return -ENOMEM; 335d43317dcSChris Mason 33619ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 33719ae4e81SJan Schmidt if (fs_info->running_transaction) { 338d43317dcSChris Mason /* 339d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3404a9d8bdeSMiao Xie * to redo the checks above 341d43317dcSChris Mason */ 3424b5faeacSDavid Sterba kfree(cur_trans); 343d43317dcSChris Mason goto loop; 34484961539SJosef Bacik } else if (BTRFS_FS_ERROR(fs_info)) { 345e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3464b5faeacSDavid Sterba kfree(cur_trans); 3477b8b92afSJosef Bacik return -EROFS; 348a4abeea4SJosef Bacik } 349d43317dcSChris Mason 350ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 35148778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 35248778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 35313c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3540860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 35579154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 35679154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3574a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 358a4abeea4SJosef Bacik /* 359a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 360a4abeea4SJosef Bacik * commit the transaction. 361a4abeea4SJosef Bacik */ 3629b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3633204d33cSJosef Bacik cur_trans->flags = 0; 364afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 36556bec294SChris Mason 366a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 367a099d0fdSAlexandru Moise 3685c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3693368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 370d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 37120b297d6SJan Schmidt 37220b297d6SJan Schmidt /* 37320b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 37420b297d6SJan Schmidt * the log must never go across transaction boundaries. 37520b297d6SJan Schmidt */ 37620b297d6SJan Schmidt smp_mb(); 37731b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3785d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 37931b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3805d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 381fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 38220b297d6SJan Schmidt 38356bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 38456bec294SChris Mason 3853063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 386bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3879e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 388ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3891bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3902b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3911bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 392ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 393e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3942b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 395d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 396d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 39719ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 398c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 39943eb5f29SQu Wenruo IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); 400fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 401fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 40219ae4e81SJan Schmidt fs_info->generation++; 40319ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 40419ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 40549b25e05SJeff Mahoney cur_trans->aborted = 0; 40619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 40715ee9bc7SJosef Bacik 40879154b1bSChris Mason return 0; 40979154b1bSChris Mason } 41079154b1bSChris Mason 411d352ac68SChris Mason /* 41292a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 41392a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 41492a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 41592a7cc42SQu Wenruo * transaction commits. 416d352ac68SChris Mason */ 4177585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 4186426c7adSQu Wenruo struct btrfs_root *root, 4196426c7adSQu Wenruo int force) 4206702ed49SChris Mason { 4210b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 42203a7e111SJosef Bacik int ret = 0; 4230b246afaSJeff Mahoney 42492a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4256426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4264d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4275d4f98a2SYan Zheng 4287585717fSChris Mason /* 42927cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4307585717fSChris Mason * we have the reloc mutex held now, so there 4317585717fSChris Mason * is only one writer in this function 4327585717fSChris Mason */ 43327cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4347585717fSChris Mason 43527cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4367585717fSChris Mason * they find our root->last_trans update 4377585717fSChris Mason */ 4387585717fSChris Mason smp_wmb(); 4397585717fSChris Mason 4400b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4416426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 4420b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 443a4abeea4SJosef Bacik return 0; 444a4abeea4SJosef Bacik } 4450b246afaSJeff Mahoney radix_tree_tag_set(&fs_info->fs_roots_radix, 4466702ed49SChris Mason (unsigned long)root->root_key.objectid, 4476702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 4480b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4497585717fSChris Mason root->last_trans = trans->transid; 4507585717fSChris Mason 4517585717fSChris Mason /* this is pretty tricky. We don't want to 4527585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4537585717fSChris Mason * unless we're really doing the first setup for this root in 4547585717fSChris Mason * this transaction. 4557585717fSChris Mason * 4567585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4577585717fSChris Mason * if we want to take the expensive mutex. 4587585717fSChris Mason * 4597585717fSChris Mason * But, we have to set root->last_trans before we 4607585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4617585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 46227cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4637585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4647585717fSChris Mason * 4657585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4667585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4677585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4687585717fSChris Mason * done before we pop in the zero below 4697585717fSChris Mason */ 47003a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 471c7548af6SChris Mason smp_mb__before_atomic(); 47227cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4736702ed49SChris Mason } 47403a7e111SJosef Bacik return ret; 4756702ed49SChris Mason } 4765d4f98a2SYan Zheng 4777585717fSChris Mason 4782b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4792b9dbef2SJosef Bacik struct btrfs_root *root) 4802b9dbef2SJosef Bacik { 4810b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4822b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4832b9dbef2SJosef Bacik 4842b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4852b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4862b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4872b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4882b9dbef2SJosef Bacik 4892b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 4900b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4910b246afaSJeff Mahoney radix_tree_tag_clear(&fs_info->fs_roots_radix, 4922b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4932b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 4940b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4952b9dbef2SJosef Bacik } 4962b9dbef2SJosef Bacik 4977585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4987585717fSChris Mason struct btrfs_root *root) 4997585717fSChris Mason { 5000b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5011409e6ccSJosef Bacik int ret; 5020b246afaSJeff Mahoney 50392a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 5047585717fSChris Mason return 0; 5057585717fSChris Mason 5067585717fSChris Mason /* 50727cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 5087585717fSChris Mason * and barriers 5097585717fSChris Mason */ 5107585717fSChris Mason smp_rmb(); 5117585717fSChris Mason if (root->last_trans == trans->transid && 51227cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 5137585717fSChris Mason return 0; 5147585717fSChris Mason 5150b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 5161409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 5170b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 5187585717fSChris Mason 5191409e6ccSJosef Bacik return ret; 5207585717fSChris Mason } 5217585717fSChris Mason 5224a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5234a9d8bdeSMiao Xie { 5243296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 525501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 526bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5274a9d8bdeSMiao Xie } 5284a9d8bdeSMiao Xie 529d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 530d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 531d352ac68SChris Mason * transaction might not be fully on disk. 532d352ac68SChris Mason */ 5332ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 53479154b1bSChris Mason { 535f9295749SChris Mason struct btrfs_transaction *cur_trans; 53679154b1bSChris Mason 5370b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5380b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5394a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5409b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5410b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 54272d63ed6SLi Zefan 5430b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 544501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 545bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 546724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 547a4abeea4SJosef Bacik } else { 5480b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 549f9295749SChris Mason } 55037d1aeeeSChris Mason } 55137d1aeeeSChris Mason 5522ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 55337d1aeeeSChris Mason { 5540b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 555a4abeea4SJosef Bacik return 0; 556a4abeea4SJosef Bacik 55792e2f7e3SNikolay Borisov if (type == TRANS_START) 558a4abeea4SJosef Bacik return 1; 559a4abeea4SJosef Bacik 560a22285a6SYan, Zheng return 0; 561a22285a6SYan, Zheng } 562a22285a6SYan, Zheng 56320dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 56420dd2cbfSMiao Xie { 5650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5660b246afaSJeff Mahoney 5670b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 56892a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 56920dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 57020dd2cbfSMiao Xie root->reloc_root) 57120dd2cbfSMiao Xie return false; 57220dd2cbfSMiao Xie 57320dd2cbfSMiao Xie return true; 57420dd2cbfSMiao Xie } 57520dd2cbfSMiao Xie 57608e007d2SMiao Xie static struct btrfs_trans_handle * 5775aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 578003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 579003d7c59SJeff Mahoney bool enforce_qgroups) 580a22285a6SYan, Zheng { 5810b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 582ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 583a22285a6SYan, Zheng struct btrfs_trans_handle *h; 584a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 585b5009945SJosef Bacik u64 num_bytes = 0; 586c5567237SArne Jansen u64 qgroup_reserved = 0; 58720dd2cbfSMiao Xie bool reloc_reserved = false; 5889c343784SJosef Bacik bool do_chunk_alloc = false; 58920dd2cbfSMiao Xie int ret; 590acce952bSliubo 59184961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) 592acce952bSliubo return ERR_PTR(-EROFS); 5932a1eb461SJosef Bacik 59446c4e71eSFilipe Manana if (current->journal_info) { 5950860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5962a1eb461SJosef Bacik h = current->journal_info; 597b50fff81SDavid Sterba refcount_inc(&h->use_count); 598b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5992a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 6002a1eb461SJosef Bacik h->block_rsv = NULL; 6012a1eb461SJosef Bacik goto got_it; 6022a1eb461SJosef Bacik } 603b5009945SJosef Bacik 604b5009945SJosef Bacik /* 605b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 606b5009945SJosef Bacik * the appropriate flushing if need be. 607b5009945SJosef Bacik */ 608003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 609ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 610ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 611ba2c4d4eSJosef Bacik 6120b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 613733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 614003d7c59SJeff Mahoney enforce_qgroups); 615c5567237SArne Jansen if (ret) 616c5567237SArne Jansen return ERR_PTR(ret); 617c5567237SArne Jansen 618ba2c4d4eSJosef Bacik /* 619ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 620ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 621ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 622ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 623ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 624ba2c4d4eSJosef Bacik */ 6252bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6267f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 6277f9fe614SJosef Bacik delayed_refs_rsv->full == 0) { 628ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 629ba2c4d4eSJosef Bacik num_bytes <<= 1; 630ba2c4d4eSJosef Bacik } 631ba2c4d4eSJosef Bacik 63220dd2cbfSMiao Xie /* 63320dd2cbfSMiao Xie * Do the reservation for the relocation root creation 63420dd2cbfSMiao Xie */ 635ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6360b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 63720dd2cbfSMiao Xie reloc_reserved = true; 63820dd2cbfSMiao Xie } 63920dd2cbfSMiao Xie 6409270501cSJosef Bacik ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 641ba2c4d4eSJosef Bacik if (ret) 642ba2c4d4eSJosef Bacik goto reserve_fail; 643ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 644ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 645ba2c4d4eSJosef Bacik delayed_refs_bytes); 646ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 647ba2c4d4eSJosef Bacik } 6489c343784SJosef Bacik 6499c343784SJosef Bacik if (rsv->space_info->force_alloc) 6509c343784SJosef Bacik do_chunk_alloc = true; 651ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 652ba2c4d4eSJosef Bacik !delayed_refs_rsv->full) { 653ba2c4d4eSJosef Bacik /* 654ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 655ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 656ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 657ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 658ba2c4d4eSJosef Bacik * here. 659ba2c4d4eSJosef Bacik */ 660ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 661b5009945SJosef Bacik if (ret) 662843fcf35SMiao Xie goto reserve_fail; 663b5009945SJosef Bacik } 664a22285a6SYan, Zheng again: 665f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 666843fcf35SMiao Xie if (!h) { 667843fcf35SMiao Xie ret = -ENOMEM; 668843fcf35SMiao Xie goto alloc_fail; 669843fcf35SMiao Xie } 670a22285a6SYan, Zheng 67198114659SJosef Bacik /* 67298114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 67398114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 67498114659SJosef Bacik * because we're already holding a ref. We need this because we could 67598114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 67698114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 677354aa0fbSMiao Xie * 678354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 679354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 68098114659SJosef Bacik */ 6810860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6820b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 683b2b5ef5cSJan Kara 6842ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6852ff7e61eSJeff Mahoney wait_current_trans(fs_info); 686a22285a6SYan, Zheng 687a4abeea4SJosef Bacik do { 6882ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 689178260b2SMiao Xie if (ret == -EBUSY) { 6902ff7e61eSJeff Mahoney wait_current_trans(fs_info); 691a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 692a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 693178260b2SMiao Xie ret = -ENOENT; 694178260b2SMiao Xie } 695a4abeea4SJosef Bacik } while (ret == -EBUSY); 696a4abeea4SJosef Bacik 697a43f7f82SLiu Bo if (ret < 0) 698843fcf35SMiao Xie goto join_fail; 6990f7d52f4SChris Mason 7000b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 701a22285a6SYan, Zheng 702a22285a6SYan, Zheng h->transid = cur_trans->transid; 703a22285a6SYan, Zheng h->transaction = cur_trans; 704b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 70564b63580SJeff Mahoney h->fs_info = root->fs_info; 7067174109cSQu Wenruo 707a698d075SMiao Xie h->type = type; 708ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 709b7ec40d7SChris Mason 710a22285a6SYan, Zheng smp_mb(); 7113296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 7122ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 713abdd2e80SFilipe Manana current->journal_info = h; 7143a45bb20SJeff Mahoney btrfs_commit_transaction(h); 715a22285a6SYan, Zheng goto again; 716a22285a6SYan, Zheng } 7179ed74f2dSJosef Bacik 718b5009945SJosef Bacik if (num_bytes) { 7190b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7202bcc0328SLiu Bo h->transid, num_bytes, 1); 7210b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 722b5009945SJosef Bacik h->bytes_reserved = num_bytes; 72320dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 724a22285a6SYan, Zheng } 725a22285a6SYan, Zheng 7262a1eb461SJosef Bacik got_it: 727bcf3a3e7SNikolay Borisov if (!current->journal_info) 728a22285a6SYan, Zheng current->journal_info = h; 729fcc99734SQu Wenruo 730fcc99734SQu Wenruo /* 7319c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7329c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7339c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7349c343784SJosef Bacik * value here. 7359c343784SJosef Bacik */ 7369c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7379c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7389c343784SJosef Bacik 7399c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7409c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7419c343784SJosef Bacik } 7429c343784SJosef Bacik 7439c343784SJosef Bacik /* 744fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 745fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 746fcc99734SQu Wenruo * transaction. 747fcc99734SQu Wenruo * 748fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 749fcc99734SQu Wenruo * or we can deadlock. 750fcc99734SQu Wenruo */ 75168075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 75268075ea8SJosef Bacik if (ret) { 75368075ea8SJosef Bacik /* 75468075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 75568075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 75668075ea8SJosef Bacik * not just freed. 75768075ea8SJosef Bacik */ 75868075ea8SJosef Bacik btrfs_end_transaction(h); 75968075ea8SJosef Bacik return ERR_PTR(ret); 76068075ea8SJosef Bacik } 761fcc99734SQu Wenruo 76279154b1bSChris Mason return h; 763843fcf35SMiao Xie 764843fcf35SMiao Xie join_fail: 7650860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7660b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 767843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 768843fcf35SMiao Xie alloc_fail: 769843fcf35SMiao Xie if (num_bytes) 7702ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 77163f018beSNikolay Borisov num_bytes, NULL); 772843fcf35SMiao Xie reserve_fail: 773733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 774843fcf35SMiao Xie return ERR_PTR(ret); 77579154b1bSChris Mason } 77679154b1bSChris Mason 777f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7785aed1dd8SAlexandru Moise unsigned int num_items) 779f9295749SChris Mason { 78008e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 781003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 782f9295749SChris Mason } 783003d7c59SJeff Mahoney 7848eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7858eab77ffSFilipe Manana struct btrfs_root *root, 7867f9fe614SJosef Bacik unsigned int num_items) 7878eab77ffSFilipe Manana { 7887f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7897f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7908eab77ffSFilipe Manana } 7918407aa46SMiao Xie 7927a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 793f9295749SChris Mason { 794003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 795003d7c59SJeff Mahoney true); 796f9295749SChris Mason } 797f9295749SChris Mason 7988d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7990af3d00bSJosef Bacik { 800575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 801003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8020af3d00bSJosef Bacik } 8030af3d00bSJosef Bacik 804d4edf39bSMiao Xie /* 805a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 806a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 807a6d155d2SFilipe Manana */ 808a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 809a6d155d2SFilipe Manana { 810a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 811a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 812a6d155d2SFilipe Manana } 813a6d155d2SFilipe Manana 814a6d155d2SFilipe Manana /* 815d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 816d4edf39bSMiao Xie * 817d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 818d4edf39bSMiao Xie * don't want to start a new one. 819d4edf39bSMiao Xie * 820d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 821d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 822d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 823d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 824d4edf39bSMiao Xie * invoke 825d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 826d4edf39bSMiao Xie */ 827354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 82860376ce4SJosef Bacik { 829575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 830003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 83160376ce4SJosef Bacik } 83260376ce4SJosef Bacik 833d4edf39bSMiao Xie /* 83490b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 835d4edf39bSMiao Xie * 83652042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 837d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 838d4edf39bSMiao Xie * complete. 839d4edf39bSMiao Xie */ 840d4edf39bSMiao Xie struct btrfs_trans_handle * 841d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 842d4edf39bSMiao Xie { 843d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 844d4edf39bSMiao Xie 845575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 846003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8478d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8482ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 849d4edf39bSMiao Xie 850d4edf39bSMiao Xie return trans; 851d4edf39bSMiao Xie } 852d4edf39bSMiao Xie 853d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 854d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 855d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 85689ce8a63SChris Mason { 857d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 85889ce8a63SChris Mason } 85989ce8a63SChris Mason 8602ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 86146204592SSage Weil { 86246204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8638cd2807fSMiao Xie int ret = 0; 86446204592SSage Weil 86546204592SSage Weil if (transid) { 8660b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 867a4abeea4SJosef Bacik goto out; 86846204592SSage Weil 86946204592SSage Weil /* find specified transaction */ 8700b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8710b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 87246204592SSage Weil if (t->transid == transid) { 87346204592SSage Weil cur_trans = t; 8749b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8758cd2807fSMiao Xie ret = 0; 87646204592SSage Weil break; 87746204592SSage Weil } 8788cd2807fSMiao Xie if (t->transid > transid) { 8798cd2807fSMiao Xie ret = 0; 88046204592SSage Weil break; 88146204592SSage Weil } 8828cd2807fSMiao Xie } 8830b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 88442383020SSage Weil 88542383020SSage Weil /* 88642383020SSage Weil * The specified transaction doesn't exist, or we 88742383020SSage Weil * raced with btrfs_commit_transaction 88842383020SSage Weil */ 88942383020SSage Weil if (!cur_trans) { 8900b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 89142383020SSage Weil ret = -EINVAL; 8928cd2807fSMiao Xie goto out; 89342383020SSage Weil } 89446204592SSage Weil } else { 89546204592SSage Weil /* find newest transaction that is committing | committed */ 8960b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8970b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 89846204592SSage Weil list) { 8994a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 9004a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 9013473f3c0SJosef Bacik break; 90246204592SSage Weil cur_trans = t; 9039b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 90446204592SSage Weil break; 90546204592SSage Weil } 90646204592SSage Weil } 9070b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 90846204592SSage Weil if (!cur_trans) 909a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 91046204592SSage Weil } 91146204592SSage Weil 912d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 913724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 914a4abeea4SJosef Bacik out: 91546204592SSage Weil return ret; 91646204592SSage Weil } 91746204592SSage Weil 9182ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 91937d1aeeeSChris Mason { 9202ff7e61eSJeff Mahoney wait_current_trans(fs_info); 92137d1aeeeSChris Mason } 92237d1aeeeSChris Mason 9238a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9248929ecfaSYan, Zheng { 9252ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9260b246afaSJeff Mahoney 92764403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9288a8f4deaSNikolay Borisov return true; 92936ba022aSJosef Bacik 9302ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9318929ecfaSYan, Zheng } 9328929ecfaSYan, Zheng 933a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9348929ecfaSYan, Zheng { 9358929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9368929ecfaSYan, Zheng 9373296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 938e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 939a2633b6aSNikolay Borisov return true; 9408929ecfaSYan, Zheng 9412ff7e61eSJeff Mahoney return should_end_transaction(trans); 9428929ecfaSYan, Zheng } 9438929ecfaSYan, Zheng 944dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 945dc60c525SNikolay Borisov 9460e34693fSNikolay Borisov { 947dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 948dc60c525SNikolay Borisov 9490e34693fSNikolay Borisov if (!trans->block_rsv) { 9500e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9510e34693fSNikolay Borisov return; 9520e34693fSNikolay Borisov } 9530e34693fSNikolay Borisov 9540e34693fSNikolay Borisov if (!trans->bytes_reserved) 9550e34693fSNikolay Borisov return; 9560e34693fSNikolay Borisov 9570e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9580e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9590e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9600e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 96163f018beSNikolay Borisov trans->bytes_reserved, NULL); 9620e34693fSNikolay Borisov trans->bytes_reserved = 0; 9630e34693fSNikolay Borisov } 9640e34693fSNikolay Borisov 96589ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9663a45bb20SJeff Mahoney int throttle) 96779154b1bSChris Mason { 9683a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9698929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9704edc2ca3SDave Jones int err = 0; 971d6e4a428SChris Mason 972b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 973b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9742a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9752a1eb461SJosef Bacik return 0; 9762a1eb461SJosef Bacik } 9772a1eb461SJosef Bacik 978dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9794c13d758SJosef Bacik trans->block_rsv = NULL; 980c5567237SArne Jansen 9816c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 982ea658badSJosef Bacik 9834fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 9844fbcdf66SFilipe Manana 9850860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 9860b246afaSJeff Mahoney sb_end_intwrite(info->sb); 9876df7881aSJosef Bacik 9888929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 98913c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 99013c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 9910860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 99289ce8a63SChris Mason 993093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 994724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 9959ed74f2dSJosef Bacik 9969ed74f2dSJosef Bacik if (current->journal_info == trans) 9979ed74f2dSJosef Bacik current->journal_info = NULL; 998ab78c84dSChris Mason 99924bbcf04SYan, Zheng if (throttle) 10002ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 100124bbcf04SYan, Zheng 100284961539SJosef Bacik if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 10034e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 1004fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 1005fbabd4a3SJosef Bacik err = trans->aborted; 1006fbabd4a3SJosef Bacik else 1007fbabd4a3SJosef Bacik err = -EROFS; 10084e121c06SJosef Bacik } 100949b25e05SJeff Mahoney 10104edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10114edc2ca3SDave Jones return err; 101279154b1bSChris Mason } 101379154b1bSChris Mason 10143a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 101589ce8a63SChris Mason { 10163a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 101789ce8a63SChris Mason } 101889ce8a63SChris Mason 10193a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 102089ce8a63SChris Mason { 10213a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 102216cdcec7SMiao Xie } 102316cdcec7SMiao Xie 1024d352ac68SChris Mason /* 1025d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1026d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1027690587d1SChris Mason * those extents are sent to disk but does not wait on them 1028d352ac68SChris Mason */ 10292ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10308cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 103179154b1bSChris Mason { 1032777e6bd7SChris Mason int err = 0; 10337c4452b9SChris Mason int werr = 0; 10340b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1035e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1036777e6bd7SChris Mason u64 start = 0; 10375f39d397SChris Mason u64 end; 10387c4452b9SChris Mason 10396300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10401728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1041e6138876SJosef Bacik mark, &cached_state)) { 1042663dfbb0SFilipe Manana bool wait_writeback = false; 1043663dfbb0SFilipe Manana 1044663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1045663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1046210aa277SDavid Sterba mark, &cached_state); 1047663dfbb0SFilipe Manana /* 1048663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1049663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1050663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1051663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1052bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1053bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1054bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1055bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1056bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1057663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 105841e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1059663dfbb0SFilipe Manana */ 1060663dfbb0SFilipe Manana if (err == -ENOMEM) { 1061663dfbb0SFilipe Manana err = 0; 1062663dfbb0SFilipe Manana wait_writeback = true; 1063663dfbb0SFilipe Manana } 1064663dfbb0SFilipe Manana if (!err) 10651728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10667c4452b9SChris Mason if (err) 10677c4452b9SChris Mason werr = err; 1068663dfbb0SFilipe Manana else if (wait_writeback) 1069663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1070e38e2ed7SFilipe Manana free_extent_state(cached_state); 1071663dfbb0SFilipe Manana cached_state = NULL; 10721728366eSJosef Bacik cond_resched(); 10731728366eSJosef Bacik start = end + 1; 10747c4452b9SChris Mason } 10756300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1076690587d1SChris Mason return werr; 1077690587d1SChris Mason } 1078690587d1SChris Mason 1079690587d1SChris Mason /* 1080690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1081690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1082690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1083690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1084690587d1SChris Mason */ 1085bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1086bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1087690587d1SChris Mason { 1088690587d1SChris Mason int err = 0; 1089690587d1SChris Mason int werr = 0; 10900b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1091e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1092690587d1SChris Mason u64 start = 0; 1093690587d1SChris Mason u64 end; 1094690587d1SChris Mason 10951728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1096e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1097663dfbb0SFilipe Manana /* 1098663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1099663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1100663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1101663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1102663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 110341e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1104663dfbb0SFilipe Manana */ 1105663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1106ae0f1625SDavid Sterba EXTENT_NEED_WAIT, 0, 0, &cached_state); 1107663dfbb0SFilipe Manana if (err == -ENOMEM) 1108663dfbb0SFilipe Manana err = 0; 1109663dfbb0SFilipe Manana if (!err) 11101728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1111777e6bd7SChris Mason if (err) 1112777e6bd7SChris Mason werr = err; 1113e38e2ed7SFilipe Manana free_extent_state(cached_state); 1114e38e2ed7SFilipe Manana cached_state = NULL; 1115777e6bd7SChris Mason cond_resched(); 11161728366eSJosef Bacik start = end + 1; 1117777e6bd7SChris Mason } 11187c4452b9SChris Mason if (err) 11197c4452b9SChris Mason werr = err; 1120bf89d38fSJeff Mahoney return werr; 1121bf89d38fSJeff Mahoney } 1122656f30dbSFilipe Manana 1123b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1124bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1125bf89d38fSJeff Mahoney { 1126bf89d38fSJeff Mahoney bool errors = false; 1127bf89d38fSJeff Mahoney int err; 1128bf89d38fSJeff Mahoney 1129bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1130bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1131bf89d38fSJeff Mahoney errors = true; 1132bf89d38fSJeff Mahoney 1133bf89d38fSJeff Mahoney if (errors && !err) 1134bf89d38fSJeff Mahoney err = -EIO; 1135bf89d38fSJeff Mahoney return err; 1136bf89d38fSJeff Mahoney } 1137bf89d38fSJeff Mahoney 1138bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1139bf89d38fSJeff Mahoney { 1140bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1141bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1142bf89d38fSJeff Mahoney bool errors = false; 1143bf89d38fSJeff Mahoney int err; 1144bf89d38fSJeff Mahoney 1145bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1146bf89d38fSJeff Mahoney 1147bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1148656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11490b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1150656f30dbSFilipe Manana errors = true; 1151656f30dbSFilipe Manana 1152656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11530b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1154656f30dbSFilipe Manana errors = true; 1155656f30dbSFilipe Manana 1156bf89d38fSJeff Mahoney if (errors && !err) 1157bf89d38fSJeff Mahoney err = -EIO; 1158bf89d38fSJeff Mahoney return err; 115979154b1bSChris Mason } 116079154b1bSChris Mason 1161690587d1SChris Mason /* 1162c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1163c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1164c9b577c0SNikolay Borisov * log commit. 1165c9b577c0SNikolay Borisov * 1166c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1167690587d1SChris Mason */ 116870458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1169d0c803c4SChris Mason { 1170663dfbb0SFilipe Manana int ret; 1171c9b577c0SNikolay Borisov int ret2; 1172c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 117370458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1174c9b577c0SNikolay Borisov struct blk_plug plug; 1175663dfbb0SFilipe Manana 1176c9b577c0SNikolay Borisov blk_start_plug(&plug); 1177c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1178c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1179c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1180c9b577c0SNikolay Borisov 118141e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1182663dfbb0SFilipe Manana 1183c9b577c0SNikolay Borisov if (ret) 1184663dfbb0SFilipe Manana return ret; 1185c9b577c0SNikolay Borisov else if (ret2) 1186c9b577c0SNikolay Borisov return ret2; 1187c9b577c0SNikolay Borisov else 1188c9b577c0SNikolay Borisov return 0; 1189d0c803c4SChris Mason } 1190d0c803c4SChris Mason 1191d352ac68SChris Mason /* 1192d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1193d352ac68SChris Mason * 1194d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1195d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1196d352ac68SChris Mason * allocation tree. 1197d352ac68SChris Mason * 1198d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1199d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1200d352ac68SChris Mason */ 12010b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 120279154b1bSChris Mason struct btrfs_root *root) 120379154b1bSChris Mason { 120479154b1bSChris Mason int ret; 12050b86a832SChris Mason u64 old_root_bytenr; 120686b9f2ecSYan, Zheng u64 old_root_used; 12070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12080b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 120979154b1bSChris Mason 121086b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 121156bec294SChris Mason 121279154b1bSChris Mason while (1) { 12130b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 121486b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1215ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 121679154b1bSChris Mason break; 121787ef2bb4SChris Mason 12185d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 121979154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12200b86a832SChris Mason &root->root_key, 12210b86a832SChris Mason &root->root_item); 122249b25e05SJeff Mahoney if (ret) 122349b25e05SJeff Mahoney return ret; 122456bec294SChris Mason 122586b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1226e7070be1SJosef Bacik } 1227276e680dSYan Zheng 12280b86a832SChris Mason return 0; 12290b86a832SChris Mason } 12300b86a832SChris Mason 1231d352ac68SChris Mason /* 1232d352ac68SChris Mason * update all the cowonly tree roots on disk 123349b25e05SJeff Mahoney * 123449b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 123549b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 123649b25e05SJeff Mahoney * to clean up the delayed refs. 1237d352ac68SChris Mason */ 12389386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12390b86a832SChris Mason { 12409386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1241ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12421bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12430b86a832SChris Mason struct list_head *next; 124484234f3aSYan Zheng struct extent_buffer *eb; 124556bec294SChris Mason int ret; 124684234f3aSYan Zheng 1247dfba78dcSFilipe Manana /* 1248dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1249dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1250dfba78dcSFilipe Manana */ 1251dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1252dfba78dcSFilipe Manana 125384234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 125449b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12559631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 125684234f3aSYan Zheng btrfs_tree_unlock(eb); 125784234f3aSYan Zheng free_extent_buffer(eb); 12580b86a832SChris Mason 125949b25e05SJeff Mahoney if (ret) 126049b25e05SJeff Mahoney return ret; 126149b25e05SJeff Mahoney 1262196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1263c16ce190SJosef Bacik if (ret) 1264c16ce190SJosef Bacik return ret; 12652b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1266c16ce190SJosef Bacik if (ret) 1267c16ce190SJosef Bacik return ret; 1268280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1269c16ce190SJosef Bacik if (ret) 1270c16ce190SJosef Bacik return ret; 1271546adb0dSJan Schmidt 1272bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1273dcdf7f6dSJosef Bacik if (ret) 1274dcdf7f6dSJosef Bacik return ret; 1275dcdf7f6dSJosef Bacik 1276ea526d18SJosef Bacik again: 12770b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12782ff7e61eSJeff Mahoney struct btrfs_root *root; 12790b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 12800b86a832SChris Mason list_del_init(next); 12810b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1282e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 128387ef2bb4SChris Mason 12849e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 12859e351cc8SJosef Bacik &trans->transaction->switch_commits); 128649b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 128749b25e05SJeff Mahoney if (ret) 128849b25e05SJeff Mahoney return ret; 1289488bc2a2SJosef Bacik } 1290488bc2a2SJosef Bacik 1291488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1292c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1293ea526d18SJosef Bacik if (ret) 1294ea526d18SJosef Bacik return ret; 1295276e680dSYan Zheng 12961bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 12975742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1298ea526d18SJosef Bacik if (ret) 1299ea526d18SJosef Bacik return ret; 1300488bc2a2SJosef Bacik 1301488bc2a2SJosef Bacik /* 1302488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1303488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1304488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1305488bc2a2SJosef Bacik * everything gets run. 1306488bc2a2SJosef Bacik */ 1307c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1308ea526d18SJosef Bacik if (ret) 1309ea526d18SJosef Bacik return ret; 1310ea526d18SJosef Bacik } 1311ea526d18SJosef Bacik 1312ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1313ea526d18SJosef Bacik goto again; 1314ea526d18SJosef Bacik 13159f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13169f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13179f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13188dabb742SStefan Behrens 131979154b1bSChris Mason return 0; 132079154b1bSChris Mason } 132179154b1bSChris Mason 1322d352ac68SChris Mason /* 1323d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1324d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1325d352ac68SChris Mason * be deleted 1326d352ac68SChris Mason */ 1327cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13285eda7b5eSChris Mason { 13290b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13300b246afaSJeff Mahoney 13310b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1332dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1333dc9492c1SJosef Bacik btrfs_grab_root(root); 13340b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1335dc9492c1SJosef Bacik } 13360b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13375eda7b5eSChris Mason } 13385eda7b5eSChris Mason 1339d352ac68SChris Mason /* 1340dfba78dcSFilipe Manana * Update each subvolume root and its relocation root, if it exists, in the tree 1341dfba78dcSFilipe Manana * of tree roots. Also free log roots if they exist. 1342d352ac68SChris Mason */ 13437e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13440f7d52f4SChris Mason { 13457e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 13460f7d52f4SChris Mason struct btrfs_root *gang[8]; 13470f7d52f4SChris Mason int i; 13480f7d52f4SChris Mason int ret; 134954aa1f4dSChris Mason 1350dfba78dcSFilipe Manana /* 1351dfba78dcSFilipe Manana * At this point no one can be using this transaction to modify any tree 1352dfba78dcSFilipe Manana * and no one can start another transaction to modify any tree either. 1353dfba78dcSFilipe Manana */ 1354dfba78dcSFilipe Manana ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1355dfba78dcSFilipe Manana 1356a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 13570f7d52f4SChris Mason while (1) { 13585d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 13595d4f98a2SYan Zheng (void **)gang, 0, 13600f7d52f4SChris Mason ARRAY_SIZE(gang), 13610f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 13620f7d52f4SChris Mason if (ret == 0) 13630f7d52f4SChris Mason break; 13640f7d52f4SChris Mason for (i = 0; i < ret; i++) { 13655b4aacefSJeff Mahoney struct btrfs_root *root = gang[i]; 13664f4317c1SJosef Bacik int ret2; 13674f4317c1SJosef Bacik 1368dfba78dcSFilipe Manana /* 1369dfba78dcSFilipe Manana * At this point we can neither have tasks logging inodes 1370dfba78dcSFilipe Manana * from a root nor trying to commit a log tree. 1371dfba78dcSFilipe Manana */ 1372dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_writers) == 0); 1373dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[0]) == 0); 1374dfba78dcSFilipe Manana ASSERT(atomic_read(&root->log_commit[1]) == 0); 1375dfba78dcSFilipe Manana 13765d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 13772619ba1fSChris Mason (unsigned long)root->root_key.objectid, 13780f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1379a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 138031153d81SYan Zheng 1381e02119d5SChris Mason btrfs_free_log(trans, root); 13822dd8298eSJosef Bacik ret2 = btrfs_update_reloc_root(trans, root); 13832dd8298eSJosef Bacik if (ret2) 13842dd8298eSJosef Bacik return ret2; 1385e02119d5SChris Mason 1386f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 138727cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1388c7548af6SChris Mason smp_mb__after_atomic(); 1389f1ebcc74SLiu Bo 1390978d910dSYan Zheng if (root->commit_root != root->node) { 13919e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13929e351cc8SJosef Bacik &trans->transaction->switch_commits); 1393978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 1394978d910dSYan Zheng root->node); 1395978d910dSYan Zheng } 139631153d81SYan Zheng 13974f4317c1SJosef Bacik ret2 = btrfs_update_root(trans, fs_info->tree_root, 13980f7d52f4SChris Mason &root->root_key, 13990f7d52f4SChris Mason &root->root_item); 14004f4317c1SJosef Bacik if (ret2) 14014f4317c1SJosef Bacik return ret2; 1402a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1403733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 14040f7d52f4SChris Mason } 14059f3a7427SChris Mason } 1406a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 14074f4317c1SJosef Bacik return 0; 14080f7d52f4SChris Mason } 14090f7d52f4SChris Mason 1410d352ac68SChris Mason /* 1411de78b51aSEric Sandeen * defrag a given btree. 1412de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1413d352ac68SChris Mason */ 1414de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1415e9d0b13bSChris Mason { 1416e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1417e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 14188929ecfaSYan, Zheng int ret; 1419e9d0b13bSChris Mason 142027cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1421e9d0b13bSChris Mason return 0; 14228929ecfaSYan, Zheng 14236b80053dSChris Mason while (1) { 14248929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14256819703fSDavid Sterba if (IS_ERR(trans)) { 14266819703fSDavid Sterba ret = PTR_ERR(trans); 14276819703fSDavid Sterba break; 14286819703fSDavid Sterba } 14298929ecfaSYan, Zheng 1430de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14318929ecfaSYan, Zheng 14323a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14332ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1434e9d0b13bSChris Mason cond_resched(); 1435e9d0b13bSChris Mason 1436ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1437e9d0b13bSChris Mason break; 1438210549ebSDavid Sterba 1439ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1440ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1441210549ebSDavid Sterba ret = -EAGAIN; 1442210549ebSDavid Sterba break; 1443210549ebSDavid Sterba } 1444e9d0b13bSChris Mason } 144527cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14468929ecfaSYan, Zheng return ret; 1447e9d0b13bSChris Mason } 1448e9d0b13bSChris Mason 1449d352ac68SChris Mason /* 14506426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 14516426c7adSQu Wenruo * 14526426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 14536426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 14546426c7adSQu Wenruo * qgroup works. 14556426c7adSQu Wenruo */ 14566426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 14576426c7adSQu Wenruo struct btrfs_root *src, 14586426c7adSQu Wenruo struct btrfs_root *parent, 14596426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 14606426c7adSQu Wenruo u64 dst_objectid) 14616426c7adSQu Wenruo { 14626426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 14636426c7adSQu Wenruo int ret; 14646426c7adSQu Wenruo 14656426c7adSQu Wenruo /* 14666426c7adSQu Wenruo * Save some performance in the case that qgroups are not 14676426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 14686426c7adSQu Wenruo * kick in anyway. 14696426c7adSQu Wenruo */ 14709ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 14716426c7adSQu Wenruo return 0; 14726426c7adSQu Wenruo 14736426c7adSQu Wenruo /* 147452042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 14754d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 14764d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 14774d31778aSQu Wenruo * item. 14784d31778aSQu Wenruo */ 14791c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 14801c442d22SJosef Bacik if (ret) 14811c442d22SJosef Bacik return ret; 14824d31778aSQu Wenruo 14834d31778aSQu Wenruo /* 14842a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 14852a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 14862a4d84c1SJosef Bacik * 14872a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 14882a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 14892a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 14902a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 14912a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 14922a4d84c1SJosef Bacik * qgroup counters could end up wrong. 14932a4d84c1SJosef Bacik */ 14942a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 14952a4d84c1SJosef Bacik if (ret) { 14962a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 149744365827SNaohiro Aota return ret; 14982a4d84c1SJosef Bacik } 14992a4d84c1SJosef Bacik 15007e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 15016426c7adSQu Wenruo if (ret) 15026426c7adSQu Wenruo goto out; 1503460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 15046426c7adSQu Wenruo if (ret < 0) 15056426c7adSQu Wenruo goto out; 15066426c7adSQu Wenruo 15076426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1508a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 15096426c7adSQu Wenruo inherit); 15106426c7adSQu Wenruo if (ret < 0) 15116426c7adSQu Wenruo goto out; 15126426c7adSQu Wenruo 15136426c7adSQu Wenruo /* 15146426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 15156426c7adSQu Wenruo * 1) commit all subvolume and extent tree 15166426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 15176426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15186426c7adSQu Wenruo * 2) write all btree blocks onto disk 15196426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15206426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15216426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15226426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15236426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15246426c7adSQu Wenruo */ 15259386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15266426c7adSQu Wenruo if (ret) 15276426c7adSQu Wenruo goto out; 1528889bfa39SJosef Bacik switch_commit_roots(trans); 152970458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15306426c7adSQu Wenruo if (ret) 1531f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15326426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15336426c7adSQu Wenruo 15346426c7adSQu Wenruo out: 15356426c7adSQu Wenruo /* 15366426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15376426c7adSQu Wenruo * last_trans == cur_transid. 15386426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15396426c7adSQu Wenruo * insert_dir_item() 15406426c7adSQu Wenruo */ 15416426c7adSQu Wenruo if (!ret) 15421c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15436426c7adSQu Wenruo return ret; 15446426c7adSQu Wenruo } 15456426c7adSQu Wenruo 15466426c7adSQu Wenruo /* 1547d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1548aec8030aSMiao Xie * transaction commit. This does the actual creation. 1549aec8030aSMiao Xie * 1550aec8030aSMiao Xie * Note: 1551aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1552aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1553aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1554d352ac68SChris Mason */ 155580b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 15563063d29fSChris Mason struct btrfs_pending_snapshot *pending) 15573063d29fSChris Mason { 155808d50ca3SNikolay Borisov 155908d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 15603063d29fSChris Mason struct btrfs_key key; 156180b6794dSChris Mason struct btrfs_root_item *new_root_item; 15623063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 15633063d29fSChris Mason struct btrfs_root *root = pending->root; 15646bdb72deSSage Weil struct btrfs_root *parent_root; 156598c9942aSLiu Bo struct btrfs_block_rsv *rsv; 15666bdb72deSSage Weil struct inode *parent_inode; 156742874b3dSMiao Xie struct btrfs_path *path; 156842874b3dSMiao Xie struct btrfs_dir_item *dir_item; 1569a22285a6SYan, Zheng struct dentry *dentry; 15703063d29fSChris Mason struct extent_buffer *tmp; 1571925baeddSChris Mason struct extent_buffer *old; 157295582b00SDeepa Dinamani struct timespec64 cur_time; 1573aec8030aSMiao Xie int ret = 0; 1574d68fc57bSYan, Zheng u64 to_reserve = 0; 15756bdb72deSSage Weil u64 index = 0; 1576a22285a6SYan, Zheng u64 objectid; 1577b83cc969SLi Zefan u64 root_flags; 15783063d29fSChris Mason 15798546b570SDavid Sterba ASSERT(pending->path); 15808546b570SDavid Sterba path = pending->path; 158142874b3dSMiao Xie 1582b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1583b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1584a22285a6SYan, Zheng 1585543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1586aec8030aSMiao Xie if (pending->error) 15876fa9700eSMiao Xie goto no_free_objectid; 15883063d29fSChris Mason 1589d6726335SQu Wenruo /* 1590d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1591d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1592d6726335SQu Wenruo */ 1593d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1594d6726335SQu Wenruo 1595147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1596d68fc57bSYan, Zheng 1597d68fc57bSYan, Zheng if (to_reserve > 0) { 15989270501cSJosef Bacik pending->error = btrfs_block_rsv_add(fs_info, 1599aec8030aSMiao Xie &pending->block_rsv, 160008e007d2SMiao Xie to_reserve, 160108e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1602aec8030aSMiao Xie if (pending->error) 1603d6726335SQu Wenruo goto clear_skip_qgroup; 1604d68fc57bSYan, Zheng } 1605d68fc57bSYan, Zheng 16063063d29fSChris Mason key.objectid = objectid; 1607a22285a6SYan, Zheng key.offset = (u64)-1; 1608a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 16093063d29fSChris Mason 16106fa9700eSMiao Xie rsv = trans->block_rsv; 1611a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 16122382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 16130b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 161488d3a5aaSJosef Bacik trans->transid, 161588d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1616a22285a6SYan, Zheng dentry = pending->dentry; 1617e9662f70SMiao Xie parent_inode = pending->dir; 1618a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1619f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1620f0118cb6SJosef Bacik if (ret) 1621f0118cb6SJosef Bacik goto fail; 1622c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 162304b285f3SDeepa Dinamani 16246bdb72deSSage Weil /* 16256bdb72deSSage Weil * insert the directory item 16266bdb72deSSage Weil */ 1627877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 162849b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 162942874b3dSMiao Xie 163042874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 163142874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16324a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 163342874b3dSMiao Xie dentry->d_name.name, 163442874b3dSMiao Xie dentry->d_name.len, 0); 163542874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1636fe66a05aSChris Mason pending->error = -EEXIST; 1637aec8030aSMiao Xie goto dir_item_existed; 163842874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 163942874b3dSMiao Xie ret = PTR_ERR(dir_item); 164066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16418732d44fSMiao Xie goto fail; 164279787eaaSJeff Mahoney } 164342874b3dSMiao Xie btrfs_release_path(path); 16446bdb72deSSage Weil 1645e999376fSChris Mason /* 1646e999376fSChris Mason * pull in the delayed directory update 1647e999376fSChris Mason * and the delayed inode item 1648e999376fSChris Mason * otherwise we corrupt the FS during 1649e999376fSChris Mason * snapshot 1650e999376fSChris Mason */ 1651e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 16528732d44fSMiao Xie if (ret) { /* Transaction aborted */ 165366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16548732d44fSMiao Xie goto fail; 16558732d44fSMiao Xie } 1656e999376fSChris Mason 1657f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1658f0118cb6SJosef Bacik if (ret) { 1659f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1660f0118cb6SJosef Bacik goto fail; 1661f0118cb6SJosef Bacik } 16626bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 16636bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 166408fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 16656bdb72deSSage Weil 1666b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1667b83cc969SLi Zefan if (pending->readonly) 1668b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1669b83cc969SLi Zefan else 1670b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1671b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1672b83cc969SLi Zefan 16738ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 16748ea05e3aSAlexander Block trans->transid); 1675807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 16768ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 16778ea05e3aSAlexander Block BTRFS_UUID_SIZE); 167870023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 167970023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 168070023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 16818ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 16828ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 16838ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 16848ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 168570023da2SStefan Behrens } 16863cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 16873cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 168870023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 16898ea05e3aSAlexander Block 1690925baeddSChris Mason old = btrfs_lock_root_node(root); 16919631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 16929631e4ccSJosef Bacik BTRFS_NESTING_COW); 169379787eaaSJeff Mahoney if (ret) { 169479787eaaSJeff Mahoney btrfs_tree_unlock(old); 169579787eaaSJeff Mahoney free_extent_buffer(old); 169666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16978732d44fSMiao Xie goto fail; 169879787eaaSJeff Mahoney } 169949b25e05SJeff Mahoney 170049b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 170179787eaaSJeff Mahoney /* clean up in any case */ 1702925baeddSChris Mason btrfs_tree_unlock(old); 1703925baeddSChris Mason free_extent_buffer(old); 17048732d44fSMiao Xie if (ret) { 170566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17068732d44fSMiao Xie goto fail; 17078732d44fSMiao Xie } 1708f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 170927cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1710f1ebcc74SLiu Bo smp_wmb(); 1711f1ebcc74SLiu Bo 17125d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1713a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1714a22285a6SYan, Zheng key.offset = trans->transid; 1715a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1716925baeddSChris Mason btrfs_tree_unlock(tmp); 17173063d29fSChris Mason free_extent_buffer(tmp); 17188732d44fSMiao Xie if (ret) { 171966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17208732d44fSMiao Xie goto fail; 17218732d44fSMiao Xie } 17220660b5afSChris Mason 1723a22285a6SYan, Zheng /* 1724a22285a6SYan, Zheng * insert root back/forward references 1725a22285a6SYan, Zheng */ 17266025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1727a22285a6SYan, Zheng parent_root->root_key.objectid, 17284a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 1729a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 17308732d44fSMiao Xie if (ret) { 173166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17328732d44fSMiao Xie goto fail; 17338732d44fSMiao Xie } 1734a22285a6SYan, Zheng 1735a22285a6SYan, Zheng key.offset = (u64)-1; 17362dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 173779787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 173879787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 17392d892ccdSFilipe Manana pending->snap = NULL; 174066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17418732d44fSMiao Xie goto fail; 174279787eaaSJeff Mahoney } 1743d68fc57bSYan, Zheng 174449b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 17458732d44fSMiao Xie if (ret) { 174666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17478732d44fSMiao Xie goto fail; 17488732d44fSMiao Xie } 1749361048f5SMiao Xie 17506426c7adSQu Wenruo /* 17516426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 17526426c7adSQu Wenruo * snapshot hack to do fast snapshot. 17536426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 17546426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 17556426c7adSQu Wenruo */ 17566426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 17576426c7adSQu Wenruo pending->inherit, objectid); 17586426c7adSQu Wenruo if (ret < 0) 17596426c7adSQu Wenruo goto fail; 17606426c7adSQu Wenruo 1761684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, dentry->d_name.name, 1762684572dfSLu Fengqi dentry->d_name.len, BTRFS_I(parent_inode), 1763684572dfSLu Fengqi &key, BTRFS_FT_DIR, index); 176442874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 17659c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 17668732d44fSMiao Xie if (ret) { 176766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17688732d44fSMiao Xie goto fail; 17698732d44fSMiao Xie } 177042874b3dSMiao Xie 17716ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 177242874b3dSMiao Xie dentry->d_name.len * 2); 177304b285f3SDeepa Dinamani parent_inode->i_mtime = parent_inode->i_ctime = 1774c2050a45SDeepa Dinamani current_time(parent_inode); 1775729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1776dd5f9615SStefan Behrens if (ret) { 177766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1778dd5f9615SStefan Behrens goto fail; 1779dd5f9615SStefan Behrens } 1780807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1781807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1782cdb345a8SLu Fengqi objectid); 1783dd5f9615SStefan Behrens if (ret) { 178466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1785dd5f9615SStefan Behrens goto fail; 1786dd5f9615SStefan Behrens } 1787dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1788cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1789dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1790dd5f9615SStefan Behrens objectid); 1791dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 179266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1793dd5f9615SStefan Behrens goto fail; 1794dd5f9615SStefan Behrens } 1795dd5f9615SStefan Behrens } 1796d6726335SQu Wenruo 17973063d29fSChris Mason fail: 1798aec8030aSMiao Xie pending->error = ret; 1799aec8030aSMiao Xie dir_item_existed: 180098c9942aSLiu Bo trans->block_rsv = rsv; 18012382c5ccSLiu Bo trans->bytes_reserved = 0; 1802d6726335SQu Wenruo clear_skip_qgroup: 1803d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 18046fa9700eSMiao Xie no_free_objectid: 18056fa9700eSMiao Xie kfree(new_root_item); 1806b0c0ea63SDavid Sterba pending->root_item = NULL; 180742874b3dSMiao Xie btrfs_free_path(path); 18088546b570SDavid Sterba pending->path = NULL; 18098546b570SDavid Sterba 181049b25e05SJeff Mahoney return ret; 18113063d29fSChris Mason } 18123063d29fSChris Mason 1813d352ac68SChris Mason /* 1814d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1815d352ac68SChris Mason */ 181608d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 18173063d29fSChris Mason { 1818aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18193063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1820aec8030aSMiao Xie int ret = 0; 18213de4586cSChris Mason 1822aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1823aec8030aSMiao Xie list_del(&pending->list); 182408d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1825aec8030aSMiao Xie if (ret) 1826aec8030aSMiao Xie break; 1827aec8030aSMiao Xie } 1828aec8030aSMiao Xie return ret; 18293de4586cSChris Mason } 18303de4586cSChris Mason 18312ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 18325d4f98a2SYan Zheng { 18335d4f98a2SYan Zheng struct btrfs_root_item *root_item; 18345d4f98a2SYan Zheng struct btrfs_super_block *super; 18355d4f98a2SYan Zheng 18360b246afaSJeff Mahoney super = fs_info->super_copy; 18375d4f98a2SYan Zheng 18380b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1839093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1840093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1841093e037cSDavid Sterba super->chunk_root_level = root_item->level; 18425d4f98a2SYan Zheng 18430b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1844093e037cSDavid Sterba super->root = root_item->bytenr; 1845093e037cSDavid Sterba super->generation = root_item->generation; 1846093e037cSDavid Sterba super->root_level = root_item->level; 18470b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1848093e037cSDavid Sterba super->cache_generation = root_item->generation; 184994846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 185094846229SBoris Burkov super->cache_generation = 0; 18510b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1852093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 18535d4f98a2SYan Zheng } 18545d4f98a2SYan Zheng 1855f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1856f36f3042SChris Mason { 18574a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1858f36f3042SChris Mason int ret = 0; 18594a9d8bdeSMiao Xie 1860a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18614a9d8bdeSMiao Xie trans = info->running_transaction; 18624a9d8bdeSMiao Xie if (trans) 18634a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1864a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1865f36f3042SChris Mason return ret; 1866f36f3042SChris Mason } 1867f36f3042SChris Mason 18688929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 18698929ecfaSYan, Zheng { 18704a9d8bdeSMiao Xie struct btrfs_transaction *trans; 18718929ecfaSYan, Zheng int ret = 0; 18724a9d8bdeSMiao Xie 1873a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18744a9d8bdeSMiao Xie trans = info->running_transaction; 18754a9d8bdeSMiao Xie if (trans) 18764a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1877a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 18788929ecfaSYan, Zheng return ret; 18798929ecfaSYan, Zheng } 18808929ecfaSYan, Zheng 1881fdfbf020SJosef Bacik void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1882bb9c12c9SSage Weil { 18833a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1884bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1885bb9c12c9SSage Weil 1886fdfbf020SJosef Bacik /* Kick the transaction kthread. */ 1887fdfbf020SJosef Bacik set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1888fdfbf020SJosef Bacik wake_up_process(fs_info->transaction_kthread); 1889bb9c12c9SSage Weil 1890bb9c12c9SSage Weil /* take transaction reference */ 1891bb9c12c9SSage Weil cur_trans = trans->transaction; 18929b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1893bb9c12c9SSage Weil 18943a45bb20SJeff Mahoney btrfs_end_transaction(trans); 18956fc4e354SSage Weil 18966fc4e354SSage Weil /* 1897ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1898ae5d29d4SDavid Sterba * subsequent transaction joins 1899ae5d29d4SDavid Sterba */ 1900ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1901ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1902ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 1903724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1904bb9c12c9SSage Weil } 1905bb9c12c9SSage Weil 190697cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 190749b25e05SJeff Mahoney { 190897cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 190949b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 191049b25e05SJeff Mahoney 1911b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 191249b25e05SJeff Mahoney 191366642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19147b8b92afSJosef Bacik 19150b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 191666b6135bSLiu Bo 191725d8c284SMiao Xie /* 191825d8c284SMiao Xie * If the transaction is removed from the list, it means this 191925d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 192025d8c284SMiao Xie * to call the cleanup function. 192125d8c284SMiao Xie */ 192225d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 192366b6135bSLiu Bo 19240b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19254a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19260b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1927f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 1928f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 1929f094ac32SLiu Bo 19300b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1931d7096fc3SJosef Bacik } 1932061dde82SFilipe Manana 1933061dde82SFilipe Manana /* 1934061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 1935061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 1936061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 1937061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 1938061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 1939061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 1940061dde82SFilipe Manana */ 1941061dde82SFilipe Manana list_del_init(&cur_trans->list); 1942061dde82SFilipe Manana 19430b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 194449b25e05SJeff Mahoney 19452ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 194649b25e05SJeff Mahoney 19470b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 19480b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 19490b246afaSJeff Mahoney fs_info->running_transaction = NULL; 19500b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 19514a9d8bdeSMiao Xie 1952e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 19530b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 1954724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1955724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 195649b25e05SJeff Mahoney 19572e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 195849b25e05SJeff Mahoney 195949b25e05SJeff Mahoney if (current->journal_info == trans) 196049b25e05SJeff Mahoney current->journal_info = NULL; 19610b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 196249b25e05SJeff Mahoney 196349b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 196449b25e05SJeff Mahoney } 196549b25e05SJeff Mahoney 1966c7cc64a9SDavid Sterba /* 1967c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 1968c7cc64a9SDavid Sterba * transaction and remove them from the list 1969c7cc64a9SDavid Sterba */ 1970c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 1971c7cc64a9SDavid Sterba { 1972c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 197332da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 1974c7cc64a9SDavid Sterba 1975c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 1976c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 1977c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 1978c7cc64a9SDavid Sterba } 1979c7cc64a9SDavid Sterba } 1980c7cc64a9SDavid Sterba 198188090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 198282436617SMiao Xie { 1983ce8ea7ccSJosef Bacik /* 1984ce8ea7ccSJosef Bacik * We use writeback_inodes_sb here because if we used 1985ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 1986ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 1987ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 1988ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 1989ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 1990ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 1991ce8ea7ccSJosef Bacik */ 199288090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 1993ce8ea7ccSJosef Bacik writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 199482436617SMiao Xie return 0; 199582436617SMiao Xie } 199682436617SMiao Xie 199788090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 199882436617SMiao Xie { 199988090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20006374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 200182436617SMiao Xie } 200282436617SMiao Xie 20033a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 200479154b1bSChris Mason { 20053a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 200649b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 20078fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 200825287e0aSMiao Xie int ret; 200979154b1bSChris Mason 201035b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 201135b814f3SNikolay Borisov 20128d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2013bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 201425287e0aSMiao Xie ret = cur_trans->aborted; 20153a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2016e4a2bcacSJosef Bacik return ret; 201725287e0aSMiao Xie } 201849b25e05SJeff Mahoney 2019f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2020f45c752bSJosef Bacik trans->block_rsv = NULL; 2021f45c752bSJosef Bacik 2022e19eb11fSJosef Bacik /* 2023e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2024e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2025e19eb11fSJosef Bacik */ 2026e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2027e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2028e19eb11fSJosef Bacik /* 2029e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2030e19eb11fSJosef Bacik * Any running threads may add more while we are here. 203156bec294SChris Mason */ 2032c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 2033e4a2bcacSJosef Bacik if (ret) { 20343a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2035e4a2bcacSJosef Bacik return ret; 2036e4a2bcacSJosef Bacik } 2037e19eb11fSJosef Bacik } 203856bec294SChris Mason 20396c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2040ea658badSJosef Bacik 20413204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 20421bbc621eSChris Mason int run_it = 0; 20431bbc621eSChris Mason 20441bbc621eSChris Mason /* this mutex is also taken before trying to set 20451bbc621eSChris Mason * block groups readonly. We need to make sure 20461bbc621eSChris Mason * that nobody has set a block group readonly 20471bbc621eSChris Mason * after a extents from that block group have been 20481bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 20491bbc621eSChris Mason * will wait for the transaction to commit if it 20503204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 20511bbc621eSChris Mason * 20523204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 20533204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 20541bbc621eSChris Mason * hurt to have more than one go through, but there's no 20551bbc621eSChris Mason * real advantage to it either. 20561bbc621eSChris Mason */ 20570b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 20583204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 20593204d33cSJosef Bacik &cur_trans->flags)) 20601bbc621eSChris Mason run_it = 1; 20610b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 20621bbc621eSChris Mason 2063f9cacae3SNikolay Borisov if (run_it) { 206421217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 20651bbc621eSChris Mason if (ret) { 20663a45bb20SJeff Mahoney btrfs_end_transaction(trans); 20671bbc621eSChris Mason return ret; 20681bbc621eSChris Mason } 2069f9cacae3SNikolay Borisov } 2070f9cacae3SNikolay Borisov } 20711bbc621eSChris Mason 20720b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 20734a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2074d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2075d0c2f4faSFilipe Manana 20760b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 20779b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2078ccd467d6SChris Mason 2079d0c2f4faSFilipe Manana if (trans->in_fsync) 2080d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2081d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2082d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 208315ee9bc7SJosef Bacik 2084bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2085b4924a0fSLiu Bo ret = cur_trans->aborted; 2086b4924a0fSLiu Bo 2087724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 208815ee9bc7SJosef Bacik 208949b25e05SJeff Mahoney return ret; 209079154b1bSChris Mason } 20914313b399SChris Mason 20924a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 20930b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 2094bb9c12c9SSage Weil 20950b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2096d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2097d0c2f4faSFilipe Manana 2098d0c2f4faSFilipe Manana if (trans->in_fsync) 2099d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2100d0c2f4faSFilipe Manana 2101ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2102ccd467d6SChris Mason struct btrfs_transaction, list); 2103d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 21049b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 21050b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2106ccd467d6SChris Mason 2107d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2108d0c2f4faSFilipe Manana 2109bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2110ccd467d6SChris Mason 2111724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 21121f9b8c8fSFilipe Manana if (ret) 21131f9b8c8fSFilipe Manana goto cleanup_transaction; 2114a4abeea4SJosef Bacik } else { 21150b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2116ccd467d6SChris Mason } 2117a4abeea4SJosef Bacik } else { 21180b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2119cb2d3dadSFilipe Manana /* 2120cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2121cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2122cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2123cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2124cb2d3dadSFilipe Manana */ 212584961539SJosef Bacik if (BTRFS_FS_ERROR(fs_info)) { 2126cb2d3dadSFilipe Manana ret = -EROFS; 2127cb2d3dadSFilipe Manana goto cleanup_transaction; 2128cb2d3dadSFilipe Manana } 2129ccd467d6SChris Mason } 213015ee9bc7SJosef Bacik 21310860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 21320860adfdSMiao Xie 213388090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 213482436617SMiao Xie if (ret) 213582436617SMiao Xie goto cleanup_transaction; 213682436617SMiao Xie 2137e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 213849b25e05SJeff Mahoney if (ret) 213949b25e05SJeff Mahoney goto cleanup_transaction; 214016cdcec7SMiao Xie 2141581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2142581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2143ed3b3d31SChris Mason 2144581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2145e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2146ca469637SMiao Xie if (ret) 2147ca469637SMiao Xie goto cleanup_transaction; 2148ca469637SMiao Xie 214988090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2150cb7ab021SWang Shilong 215148778179SFilipe Manana /* 215248778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 215348778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 215448778179SFilipe Manana * extents complete we lose logged data after a power failure. 215548778179SFilipe Manana */ 215648778179SFilipe Manana wait_event(cur_trans->pending_wait, 215748778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 215848778179SFilipe Manana 21592ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2160ed0ca140SJosef Bacik /* 2161ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2162ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 21634a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2164ed0ca140SJosef Bacik */ 21650b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21664a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 21670b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2168ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2169ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 217015ee9bc7SJosef Bacik 2171fdfbf020SJosef Bacik /* 2172fdfbf020SJosef Bacik * We've started the commit, clear the flag in case we were triggered to 2173fdfbf020SJosef Bacik * do an async commit but somebody else started before the transaction 2174fdfbf020SJosef Bacik * kthread could do the work. 2175fdfbf020SJosef Bacik */ 2176fdfbf020SJosef Bacik clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2177fdfbf020SJosef Bacik 2178bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 21792cba30f1SMiao Xie ret = cur_trans->aborted; 21806cf7f77eSWang Shilong goto scrub_continue; 21812cba30f1SMiao Xie } 21827585717fSChris Mason /* 21837585717fSChris Mason * the reloc mutex makes sure that we stop 21847585717fSChris Mason * the balancing code from coming in and moving 21857585717fSChris Mason * extents around in the middle of the commit 21867585717fSChris Mason */ 21870b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 21887585717fSChris Mason 218942874b3dSMiao Xie /* 219042874b3dSMiao Xie * We needn't worry about the delayed items because we will 219142874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 219242874b3dSMiao Xie * core function of the snapshot creation. 219342874b3dSMiao Xie */ 219408d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 219556e9f6eaSDavid Sterba if (ret) 219656e9f6eaSDavid Sterba goto unlock_reloc; 21973063d29fSChris Mason 219842874b3dSMiao Xie /* 219942874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 220042874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 220142874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 220242874b3dSMiao Xie * them. 220342874b3dSMiao Xie * 220442874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 220542874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 220642874b3dSMiao Xie * the nodes and leaves. 220742874b3dSMiao Xie */ 2208e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 220956e9f6eaSDavid Sterba if (ret) 221056e9f6eaSDavid Sterba goto unlock_reloc; 221116cdcec7SMiao Xie 2212c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 221356e9f6eaSDavid Sterba if (ret) 221456e9f6eaSDavid Sterba goto unlock_reloc; 221556bec294SChris Mason 2216e999376fSChris Mason /* 2217e999376fSChris Mason * make sure none of the code above managed to slip in a 2218e999376fSChris Mason * delayed item 2219e999376fSChris Mason */ 2220ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2221e999376fSChris Mason 22222c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2223dc17ff8fSChris Mason 22247e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 222556e9f6eaSDavid Sterba if (ret) 2226dfba78dcSFilipe Manana goto unlock_reloc; 222754aa1f4dSChris Mason 22283818aea2SQu Wenruo /* 22297e1876acSDavid Sterba * Since the transaction is done, we can apply the pending changes 22307e1876acSDavid Sterba * before the next transaction. 22313818aea2SQu Wenruo */ 22320b246afaSJeff Mahoney btrfs_apply_pending_changes(fs_info); 22333818aea2SQu Wenruo 22345d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2235e02119d5SChris Mason * safe to free the root of tree log roots 2236e02119d5SChris Mason */ 22370b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2238e02119d5SChris Mason 22390ed4792aSQu Wenruo /* 22400ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 22410ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 22420ed4792aSQu Wenruo */ 2243460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 224456e9f6eaSDavid Sterba if (ret < 0) 2245dfba78dcSFilipe Manana goto unlock_reloc; 22460ed4792aSQu Wenruo 22479386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 224856e9f6eaSDavid Sterba if (ret) 2249dfba78dcSFilipe Manana goto unlock_reloc; 225054aa1f4dSChris Mason 22512cba30f1SMiao Xie /* 22522cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 22532cba30f1SMiao Xie * update ->aborted, check it. 22542cba30f1SMiao Xie */ 2255bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 22562cba30f1SMiao Xie ret = cur_trans->aborted; 2257dfba78dcSFilipe Manana goto unlock_reloc; 22582cba30f1SMiao Xie } 22592cba30f1SMiao Xie 22600b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 22615f39d397SChris Mason 22620b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 22630b246afaSJeff Mahoney fs_info->tree_root->node); 22640b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 22659e351cc8SJosef Bacik &cur_trans->switch_commits); 22665d4f98a2SYan Zheng 22670b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 22680b246afaSJeff Mahoney fs_info->chunk_root->node); 22690b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 22709e351cc8SJosef Bacik &cur_trans->switch_commits); 22719e351cc8SJosef Bacik 2272889bfa39SJosef Bacik switch_commit_roots(trans); 22735d4f98a2SYan Zheng 2274ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 22751bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 22762ff7e61eSJeff Mahoney update_super_roots(fs_info); 2277e02119d5SChris Mason 22780b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 22790b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 22800b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 22810b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2282ccd467d6SChris Mason 2283bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2284935e5cc9SMiao Xie 22850b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 22860b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2287656f30dbSFilipe Manana 22884fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 22894fbcdf66SFilipe Manana 2290dfba78dcSFilipe Manana /* 2291dfba78dcSFilipe Manana * Before changing the transaction state to TRANS_STATE_UNBLOCKED and 2292dfba78dcSFilipe Manana * setting fs_info->running_transaction to NULL, lock tree_log_mutex to 2293dfba78dcSFilipe Manana * make sure that before we commit our superblock, no other task can 2294dfba78dcSFilipe Manana * start a new transaction and commit a log tree before we commit our 2295dfba78dcSFilipe Manana * superblock. Anyone trying to commit a log tree locks this mutex before 2296dfba78dcSFilipe Manana * writing its superblock. 2297dfba78dcSFilipe Manana */ 2298dfba78dcSFilipe Manana mutex_lock(&fs_info->tree_log_mutex); 2299dfba78dcSFilipe Manana 23000b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 23014a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 23020b246afaSJeff Mahoney fs_info->running_transaction = NULL; 23030b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 23040b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2305b7ec40d7SChris Mason 23060b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 2307e6dcd2dcSChris Mason 230870458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 230949b25e05SJeff Mahoney if (ret) { 23100b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 231108748810SDavid Sterba "Error while writing out transaction"); 23120b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 23136cf7f77eSWang Shilong goto scrub_continue; 231449b25e05SJeff Mahoney } 231549b25e05SJeff Mahoney 2316d3575156SNaohiro Aota /* 2317d3575156SNaohiro Aota * At this point, we should have written all the tree blocks allocated 2318d3575156SNaohiro Aota * in this transaction. So it's now safe to free the redirtyied extent 2319d3575156SNaohiro Aota * buffers. 2320d3575156SNaohiro Aota */ 2321d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 2322d3575156SNaohiro Aota 2323eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2324e02119d5SChris Mason /* 2325e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2326e02119d5SChris Mason * to go about their business 2327e02119d5SChris Mason */ 23280b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2329c1f32b7cSAnand Jain if (ret) 2330c1f32b7cSAnand Jain goto scrub_continue; 2331e02119d5SChris Mason 2332d0c2f4faSFilipe Manana /* 2333d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2334d0c2f4faSFilipe Manana * which can change it. 2335d0c2f4faSFilipe Manana */ 2336d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2337d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 2338d0c2f4faSFilipe Manana 23395ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 23404313b399SChris Mason 23413204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 23420b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 234313212b54SZhao Lei 23440b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 23454a9d8bdeSMiao Xie /* 23464a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 23474a9d8bdeSMiao Xie * which can change it. 23484a9d8bdeSMiao Xie */ 23494a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 23502c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 23513de4586cSChris Mason 23520b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 235313c5a93eSJosef Bacik list_del_init(&cur_trans->list); 23540b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2355a4abeea4SJosef Bacik 2356724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2357724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 235858176a96SJosef Bacik 23590860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 23600b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2361b2b5ef5cSJan Kara 23622e4e97abSJosef Bacik trace_btrfs_transaction_commit(fs_info); 23631abe9b8aSliubo 23642ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2365a2de733cSArne Jansen 23669ed74f2dSJosef Bacik if (current->journal_info == trans) 23679ed74f2dSJosef Bacik current->journal_info = NULL; 23689ed74f2dSJosef Bacik 23692c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 237024bbcf04SYan, Zheng 237179154b1bSChris Mason return ret; 237249b25e05SJeff Mahoney 237356e9f6eaSDavid Sterba unlock_reloc: 237456e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 23756cf7f77eSWang Shilong scrub_continue: 23762ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 237749b25e05SJeff Mahoney cleanup_transaction: 2378dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2379c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 23804fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 23810e721106SJosef Bacik trans->block_rsv = NULL; 23820b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 238349b25e05SJeff Mahoney if (current->journal_info == trans) 238449b25e05SJeff Mahoney current->journal_info = NULL; 238597cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 238649b25e05SJeff Mahoney 238749b25e05SJeff Mahoney return ret; 238879154b1bSChris Mason } 238979154b1bSChris Mason 2390d352ac68SChris Mason /* 23919d1a2a3aSDavid Sterba * return < 0 if error 23929d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 23939d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 23949d1a2a3aSDavid Sterba * 23959d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 23969d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 23979d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 23989d1a2a3aSDavid Sterba * few seconds later. 2399d352ac68SChris Mason */ 24009d1a2a3aSDavid Sterba int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) 2401e9d0b13bSChris Mason { 24029d1a2a3aSDavid Sterba int ret; 24035d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 2404e9d0b13bSChris Mason 2405a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 24069d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 24079d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 24089d1a2a3aSDavid Sterba return 0; 24099d1a2a3aSDavid Sterba } 24109d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 24119d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2412cfad392bSJosef Bacik list_del_init(&root->root_list); 2413a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 24145d4f98a2SYan Zheng 24154fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 241676dda93cSYan, Zheng 241716cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 241816cdcec7SMiao Xie 241976dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 242076dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 24210078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 242276dda93cSYan, Zheng else 24230078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 242432471dc2SDavid Sterba 2425dc9492c1SJosef Bacik btrfs_put_root(root); 24266596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2427e9d0b13bSChris Mason } 2428572d9ab7SDavid Sterba 2429572d9ab7SDavid Sterba void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) 2430572d9ab7SDavid Sterba { 2431572d9ab7SDavid Sterba unsigned long prev; 2432572d9ab7SDavid Sterba unsigned long bit; 2433572d9ab7SDavid Sterba 24346c9fe14fSQu Wenruo prev = xchg(&fs_info->pending_changes, 0); 2435572d9ab7SDavid Sterba if (!prev) 2436572d9ab7SDavid Sterba return; 2437572d9ab7SDavid Sterba 2438d51033d0SDavid Sterba bit = 1 << BTRFS_PENDING_COMMIT; 2439d51033d0SDavid Sterba if (prev & bit) 2440d51033d0SDavid Sterba btrfs_debug(fs_info, "pending commit done"); 2441d51033d0SDavid Sterba prev &= ~bit; 2442d51033d0SDavid Sterba 2443572d9ab7SDavid Sterba if (prev) 2444572d9ab7SDavid Sterba btrfs_warn(fs_info, 2445572d9ab7SDavid Sterba "unknown pending changes left 0x%lx, ignoring", prev); 2446572d9ab7SDavid Sterba } 2447